xref: /freebsd/sys/dev/cxgbe/t4_main.c (revision 85732ac8)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2011 Chelsio Communications, Inc.
5  * All rights reserved.
6  * Written by: Navdeep Parhar <np@FreeBSD.org>
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  */
29 
30 #include <sys/cdefs.h>
31 __FBSDID("$FreeBSD$");
32 
33 #include "opt_ddb.h"
34 #include "opt_inet.h"
35 #include "opt_inet6.h"
36 #include "opt_ratelimit.h"
37 #include "opt_rss.h"
38 
39 #include <sys/param.h>
40 #include <sys/conf.h>
41 #include <sys/priv.h>
42 #include <sys/kernel.h>
43 #include <sys/bus.h>
44 #include <sys/module.h>
45 #include <sys/malloc.h>
46 #include <sys/queue.h>
47 #include <sys/taskqueue.h>
48 #include <sys/pciio.h>
49 #include <dev/pci/pcireg.h>
50 #include <dev/pci/pcivar.h>
51 #include <dev/pci/pci_private.h>
52 #include <sys/firmware.h>
53 #include <sys/sbuf.h>
54 #include <sys/smp.h>
55 #include <sys/socket.h>
56 #include <sys/sockio.h>
57 #include <sys/sysctl.h>
58 #include <net/ethernet.h>
59 #include <net/if.h>
60 #include <net/if_types.h>
61 #include <net/if_dl.h>
62 #include <net/if_vlan_var.h>
63 #ifdef RSS
64 #include <net/rss_config.h>
65 #endif
66 #include <netinet/in.h>
67 #include <netinet/ip.h>
68 #if defined(__i386__) || defined(__amd64__)
69 #include <machine/md_var.h>
70 #include <machine/cputypes.h>
71 #include <vm/vm.h>
72 #include <vm/pmap.h>
73 #endif
74 #include <crypto/rijndael/rijndael.h>
75 #ifdef DDB
76 #include <ddb/ddb.h>
77 #include <ddb/db_lex.h>
78 #endif
79 
80 #include "common/common.h"
81 #include "common/t4_msg.h"
82 #include "common/t4_regs.h"
83 #include "common/t4_regs_values.h"
84 #include "cudbg/cudbg.h"
85 #include "t4_clip.h"
86 #include "t4_ioctl.h"
87 #include "t4_l2t.h"
88 #include "t4_mp_ring.h"
89 #include "t4_if.h"
90 #include "t4_smt.h"
91 
92 /* T4 bus driver interface */
93 static int t4_probe(device_t);
94 static int t4_attach(device_t);
95 static int t4_detach(device_t);
96 static int t4_child_location_str(device_t, device_t, char *, size_t);
97 static int t4_ready(device_t);
98 static int t4_read_port_device(device_t, int, device_t *);
99 static device_method_t t4_methods[] = {
100 	DEVMETHOD(device_probe,		t4_probe),
101 	DEVMETHOD(device_attach,	t4_attach),
102 	DEVMETHOD(device_detach,	t4_detach),
103 
104 	DEVMETHOD(bus_child_location_str, t4_child_location_str),
105 
106 	DEVMETHOD(t4_is_main_ready,	t4_ready),
107 	DEVMETHOD(t4_read_port_device,	t4_read_port_device),
108 
109 	DEVMETHOD_END
110 };
111 static driver_t t4_driver = {
112 	"t4nex",
113 	t4_methods,
114 	sizeof(struct adapter)
115 };
116 
117 
118 /* T4 port (cxgbe) interface */
119 static int cxgbe_probe(device_t);
120 static int cxgbe_attach(device_t);
121 static int cxgbe_detach(device_t);
122 device_method_t cxgbe_methods[] = {
123 	DEVMETHOD(device_probe,		cxgbe_probe),
124 	DEVMETHOD(device_attach,	cxgbe_attach),
125 	DEVMETHOD(device_detach,	cxgbe_detach),
126 	{ 0, 0 }
127 };
128 static driver_t cxgbe_driver = {
129 	"cxgbe",
130 	cxgbe_methods,
131 	sizeof(struct port_info)
132 };
133 
134 /* T4 VI (vcxgbe) interface */
135 static int vcxgbe_probe(device_t);
136 static int vcxgbe_attach(device_t);
137 static int vcxgbe_detach(device_t);
138 static device_method_t vcxgbe_methods[] = {
139 	DEVMETHOD(device_probe,		vcxgbe_probe),
140 	DEVMETHOD(device_attach,	vcxgbe_attach),
141 	DEVMETHOD(device_detach,	vcxgbe_detach),
142 	{ 0, 0 }
143 };
144 static driver_t vcxgbe_driver = {
145 	"vcxgbe",
146 	vcxgbe_methods,
147 	sizeof(struct vi_info)
148 };
149 
150 static d_ioctl_t t4_ioctl;
151 
152 static struct cdevsw t4_cdevsw = {
153        .d_version = D_VERSION,
154        .d_ioctl = t4_ioctl,
155        .d_name = "t4nex",
156 };
157 
158 /* T5 bus driver interface */
159 static int t5_probe(device_t);
160 static device_method_t t5_methods[] = {
161 	DEVMETHOD(device_probe,		t5_probe),
162 	DEVMETHOD(device_attach,	t4_attach),
163 	DEVMETHOD(device_detach,	t4_detach),
164 
165 	DEVMETHOD(bus_child_location_str, t4_child_location_str),
166 
167 	DEVMETHOD(t4_is_main_ready,	t4_ready),
168 	DEVMETHOD(t4_read_port_device,	t4_read_port_device),
169 
170 	DEVMETHOD_END
171 };
172 static driver_t t5_driver = {
173 	"t5nex",
174 	t5_methods,
175 	sizeof(struct adapter)
176 };
177 
178 
179 /* T5 port (cxl) interface */
180 static driver_t cxl_driver = {
181 	"cxl",
182 	cxgbe_methods,
183 	sizeof(struct port_info)
184 };
185 
186 /* T5 VI (vcxl) interface */
187 static driver_t vcxl_driver = {
188 	"vcxl",
189 	vcxgbe_methods,
190 	sizeof(struct vi_info)
191 };
192 
193 /* T6 bus driver interface */
194 static int t6_probe(device_t);
195 static device_method_t t6_methods[] = {
196 	DEVMETHOD(device_probe,		t6_probe),
197 	DEVMETHOD(device_attach,	t4_attach),
198 	DEVMETHOD(device_detach,	t4_detach),
199 
200 	DEVMETHOD(bus_child_location_str, t4_child_location_str),
201 
202 	DEVMETHOD(t4_is_main_ready,	t4_ready),
203 	DEVMETHOD(t4_read_port_device,	t4_read_port_device),
204 
205 	DEVMETHOD_END
206 };
207 static driver_t t6_driver = {
208 	"t6nex",
209 	t6_methods,
210 	sizeof(struct adapter)
211 };
212 
213 
214 /* T6 port (cc) interface */
215 static driver_t cc_driver = {
216 	"cc",
217 	cxgbe_methods,
218 	sizeof(struct port_info)
219 };
220 
221 /* T6 VI (vcc) interface */
222 static driver_t vcc_driver = {
223 	"vcc",
224 	vcxgbe_methods,
225 	sizeof(struct vi_info)
226 };
227 
228 /* ifnet interface */
229 static void cxgbe_init(void *);
230 static int cxgbe_ioctl(struct ifnet *, unsigned long, caddr_t);
231 static int cxgbe_transmit(struct ifnet *, struct mbuf *);
232 static void cxgbe_qflush(struct ifnet *);
233 
234 MALLOC_DEFINE(M_CXGBE, "cxgbe", "Chelsio T4/T5 Ethernet driver and services");
235 
236 /*
237  * Correct lock order when you need to acquire multiple locks is t4_list_lock,
238  * then ADAPTER_LOCK, then t4_uld_list_lock.
239  */
240 static struct sx t4_list_lock;
241 SLIST_HEAD(, adapter) t4_list;
242 #ifdef TCP_OFFLOAD
243 static struct sx t4_uld_list_lock;
244 SLIST_HEAD(, uld_info) t4_uld_list;
245 #endif
246 
247 /*
248  * Tunables.  See tweak_tunables() too.
249  *
250  * Each tunable is set to a default value here if it's known at compile-time.
251  * Otherwise it is set to -n as an indication to tweak_tunables() that it should
252  * provide a reasonable default (upto n) when the driver is loaded.
253  *
254  * Tunables applicable to both T4 and T5 are under hw.cxgbe.  Those specific to
255  * T5 are under hw.cxl.
256  */
257 SYSCTL_NODE(_hw, OID_AUTO, cxgbe, CTLFLAG_RD, 0, "cxgbe(4) parameters");
258 SYSCTL_NODE(_hw, OID_AUTO, cxl, CTLFLAG_RD, 0, "cxgbe(4) T5+ parameters");
259 SYSCTL_NODE(_hw_cxgbe, OID_AUTO, toe, CTLFLAG_RD, 0, "cxgbe(4) TOE parameters");
260 
261 /*
262  * Number of queues for tx and rx, NIC and offload.
263  */
264 #define NTXQ 16
265 int t4_ntxq = -NTXQ;
266 SYSCTL_INT(_hw_cxgbe, OID_AUTO, ntxq, CTLFLAG_RDTUN, &t4_ntxq, 0,
267     "Number of TX queues per port");
268 TUNABLE_INT("hw.cxgbe.ntxq10g", &t4_ntxq);	/* Old name, undocumented */
269 
270 #define NRXQ 8
271 int t4_nrxq = -NRXQ;
272 SYSCTL_INT(_hw_cxgbe, OID_AUTO, nrxq, CTLFLAG_RDTUN, &t4_nrxq, 0,
273     "Number of RX queues per port");
274 TUNABLE_INT("hw.cxgbe.nrxq10g", &t4_nrxq);	/* Old name, undocumented */
275 
276 #define NTXQ_VI 1
277 static int t4_ntxq_vi = -NTXQ_VI;
278 SYSCTL_INT(_hw_cxgbe, OID_AUTO, ntxq_vi, CTLFLAG_RDTUN, &t4_ntxq_vi, 0,
279     "Number of TX queues per VI");
280 
281 #define NRXQ_VI 1
282 static int t4_nrxq_vi = -NRXQ_VI;
283 SYSCTL_INT(_hw_cxgbe, OID_AUTO, nrxq_vi, CTLFLAG_RDTUN, &t4_nrxq_vi, 0,
284     "Number of RX queues per VI");
285 
286 static int t4_rsrv_noflowq = 0;
287 SYSCTL_INT(_hw_cxgbe, OID_AUTO, rsrv_noflowq, CTLFLAG_RDTUN, &t4_rsrv_noflowq,
288     0, "Reserve TX queue 0 of each VI for non-flowid packets");
289 
290 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
291 #define NOFLDTXQ 8
292 static int t4_nofldtxq = -NOFLDTXQ;
293 SYSCTL_INT(_hw_cxgbe, OID_AUTO, nofldtxq, CTLFLAG_RDTUN, &t4_nofldtxq, 0,
294     "Number of offload TX queues per port");
295 
296 #define NOFLDRXQ 2
297 static int t4_nofldrxq = -NOFLDRXQ;
298 SYSCTL_INT(_hw_cxgbe, OID_AUTO, nofldrxq, CTLFLAG_RDTUN, &t4_nofldrxq, 0,
299     "Number of offload RX queues per port");
300 
301 #define NOFLDTXQ_VI 1
302 static int t4_nofldtxq_vi = -NOFLDTXQ_VI;
303 SYSCTL_INT(_hw_cxgbe, OID_AUTO, nofldtxq_vi, CTLFLAG_RDTUN, &t4_nofldtxq_vi, 0,
304     "Number of offload TX queues per VI");
305 
306 #define NOFLDRXQ_VI 1
307 static int t4_nofldrxq_vi = -NOFLDRXQ_VI;
308 SYSCTL_INT(_hw_cxgbe, OID_AUTO, nofldrxq_vi, CTLFLAG_RDTUN, &t4_nofldrxq_vi, 0,
309     "Number of offload RX queues per VI");
310 
311 #define TMR_IDX_OFLD 1
312 int t4_tmr_idx_ofld = TMR_IDX_OFLD;
313 SYSCTL_INT(_hw_cxgbe, OID_AUTO, holdoff_timer_idx_ofld, CTLFLAG_RDTUN,
314     &t4_tmr_idx_ofld, 0, "Holdoff timer index for offload queues");
315 
316 #define PKTC_IDX_OFLD (-1)
317 int t4_pktc_idx_ofld = PKTC_IDX_OFLD;
318 SYSCTL_INT(_hw_cxgbe, OID_AUTO, holdoff_pktc_idx_ofld, CTLFLAG_RDTUN,
319     &t4_pktc_idx_ofld, 0, "holdoff packet counter index for offload queues");
320 
321 /* 0 means chip/fw default, non-zero number is value in microseconds */
322 static u_long t4_toe_keepalive_idle = 0;
323 SYSCTL_ULONG(_hw_cxgbe_toe, OID_AUTO, keepalive_idle, CTLFLAG_RDTUN,
324     &t4_toe_keepalive_idle, 0, "TOE keepalive idle timer (us)");
325 
326 /* 0 means chip/fw default, non-zero number is value in microseconds */
327 static u_long t4_toe_keepalive_interval = 0;
328 SYSCTL_ULONG(_hw_cxgbe_toe, OID_AUTO, keepalive_interval, CTLFLAG_RDTUN,
329     &t4_toe_keepalive_interval, 0, "TOE keepalive interval timer (us)");
330 
331 /* 0 means chip/fw default, non-zero number is # of keepalives before abort */
332 static int t4_toe_keepalive_count = 0;
333 SYSCTL_INT(_hw_cxgbe_toe, OID_AUTO, keepalive_count, CTLFLAG_RDTUN,
334     &t4_toe_keepalive_count, 0, "Number of TOE keepalive probes before abort");
335 
336 /* 0 means chip/fw default, non-zero number is value in microseconds */
337 static u_long t4_toe_rexmt_min = 0;
338 SYSCTL_ULONG(_hw_cxgbe_toe, OID_AUTO, rexmt_min, CTLFLAG_RDTUN,
339     &t4_toe_rexmt_min, 0, "Minimum TOE retransmit interval (us)");
340 
341 /* 0 means chip/fw default, non-zero number is value in microseconds */
342 static u_long t4_toe_rexmt_max = 0;
343 SYSCTL_ULONG(_hw_cxgbe_toe, OID_AUTO, rexmt_max, CTLFLAG_RDTUN,
344     &t4_toe_rexmt_max, 0, "Maximum TOE retransmit interval (us)");
345 
346 /* 0 means chip/fw default, non-zero number is # of rexmt before abort */
347 static int t4_toe_rexmt_count = 0;
348 SYSCTL_INT(_hw_cxgbe_toe, OID_AUTO, rexmt_count, CTLFLAG_RDTUN,
349     &t4_toe_rexmt_count, 0, "Number of TOE retransmissions before abort");
350 
351 /* -1 means chip/fw default, other values are raw backoff values to use */
352 static int t4_toe_rexmt_backoff[16] = {
353 	-1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1
354 };
355 SYSCTL_NODE(_hw_cxgbe_toe, OID_AUTO, rexmt_backoff, CTLFLAG_RD, 0,
356     "cxgbe(4) TOE retransmit backoff values");
357 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 0, CTLFLAG_RDTUN,
358     &t4_toe_rexmt_backoff[0], 0, "");
359 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 1, CTLFLAG_RDTUN,
360     &t4_toe_rexmt_backoff[1], 0, "");
361 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 2, CTLFLAG_RDTUN,
362     &t4_toe_rexmt_backoff[2], 0, "");
363 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 3, CTLFLAG_RDTUN,
364     &t4_toe_rexmt_backoff[3], 0, "");
365 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 4, CTLFLAG_RDTUN,
366     &t4_toe_rexmt_backoff[4], 0, "");
367 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 5, CTLFLAG_RDTUN,
368     &t4_toe_rexmt_backoff[5], 0, "");
369 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 6, CTLFLAG_RDTUN,
370     &t4_toe_rexmt_backoff[6], 0, "");
371 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 7, CTLFLAG_RDTUN,
372     &t4_toe_rexmt_backoff[7], 0, "");
373 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 8, CTLFLAG_RDTUN,
374     &t4_toe_rexmt_backoff[8], 0, "");
375 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 9, CTLFLAG_RDTUN,
376     &t4_toe_rexmt_backoff[9], 0, "");
377 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 10, CTLFLAG_RDTUN,
378     &t4_toe_rexmt_backoff[10], 0, "");
379 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 11, CTLFLAG_RDTUN,
380     &t4_toe_rexmt_backoff[11], 0, "");
381 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 12, CTLFLAG_RDTUN,
382     &t4_toe_rexmt_backoff[12], 0, "");
383 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 13, CTLFLAG_RDTUN,
384     &t4_toe_rexmt_backoff[13], 0, "");
385 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 14, CTLFLAG_RDTUN,
386     &t4_toe_rexmt_backoff[14], 0, "");
387 SYSCTL_INT(_hw_cxgbe_toe_rexmt_backoff, OID_AUTO, 15, CTLFLAG_RDTUN,
388     &t4_toe_rexmt_backoff[15], 0, "");
389 #endif
390 
391 #ifdef DEV_NETMAP
392 #define NNMTXQ_VI 2
393 static int t4_nnmtxq_vi = -NNMTXQ_VI;
394 SYSCTL_INT(_hw_cxgbe, OID_AUTO, nnmtxq_vi, CTLFLAG_RDTUN, &t4_nnmtxq_vi, 0,
395     "Number of netmap TX queues per VI");
396 
397 #define NNMRXQ_VI 2
398 static int t4_nnmrxq_vi = -NNMRXQ_VI;
399 SYSCTL_INT(_hw_cxgbe, OID_AUTO, nnmrxq_vi, CTLFLAG_RDTUN, &t4_nnmrxq_vi, 0,
400     "Number of netmap RX queues per VI");
401 #endif
402 
403 /*
404  * Holdoff parameters for ports.
405  */
406 #define TMR_IDX 1
407 int t4_tmr_idx = TMR_IDX;
408 SYSCTL_INT(_hw_cxgbe, OID_AUTO, holdoff_timer_idx, CTLFLAG_RDTUN, &t4_tmr_idx,
409     0, "Holdoff timer index");
410 TUNABLE_INT("hw.cxgbe.holdoff_timer_idx_10G", &t4_tmr_idx);	/* Old name */
411 
412 #define PKTC_IDX (-1)
413 int t4_pktc_idx = PKTC_IDX;
414 SYSCTL_INT(_hw_cxgbe, OID_AUTO, holdoff_pktc_idx, CTLFLAG_RDTUN, &t4_pktc_idx,
415     0, "Holdoff packet counter index");
416 TUNABLE_INT("hw.cxgbe.holdoff_pktc_idx_10G", &t4_pktc_idx);	/* Old name */
417 
418 /*
419  * Size (# of entries) of each tx and rx queue.
420  */
421 unsigned int t4_qsize_txq = TX_EQ_QSIZE;
422 SYSCTL_INT(_hw_cxgbe, OID_AUTO, qsize_txq, CTLFLAG_RDTUN, &t4_qsize_txq, 0,
423     "Number of descriptors in each TX queue");
424 
425 unsigned int t4_qsize_rxq = RX_IQ_QSIZE;
426 SYSCTL_INT(_hw_cxgbe, OID_AUTO, qsize_rxq, CTLFLAG_RDTUN, &t4_qsize_rxq, 0,
427     "Number of descriptors in each RX queue");
428 
429 /*
430  * Interrupt types allowed (bits 0, 1, 2 = INTx, MSI, MSI-X respectively).
431  */
432 int t4_intr_types = INTR_MSIX | INTR_MSI | INTR_INTX;
433 SYSCTL_INT(_hw_cxgbe, OID_AUTO, interrupt_types, CTLFLAG_RDTUN, &t4_intr_types,
434     0, "Interrupt types allowed (bit 0 = INTx, 1 = MSI, 2 = MSI-X)");
435 
436 /*
437  * Configuration file.  All the _CF names here are special.
438  */
439 #define DEFAULT_CF	"default"
440 #define BUILTIN_CF	"built-in"
441 #define FLASH_CF	"flash"
442 #define UWIRE_CF	"uwire"
443 #define FPGA_CF		"fpga"
444 static char t4_cfg_file[32] = DEFAULT_CF;
445 SYSCTL_STRING(_hw_cxgbe, OID_AUTO, config_file, CTLFLAG_RDTUN, t4_cfg_file,
446     sizeof(t4_cfg_file), "Firmware configuration file");
447 
448 /*
449  * PAUSE settings (bit 0, 1, 2 = rx_pause, tx_pause, pause_autoneg respectively).
450  * rx_pause = 1 to heed incoming PAUSE frames, 0 to ignore them.
451  * tx_pause = 1 to emit PAUSE frames when the rx FIFO reaches its high water
452  *            mark or when signalled to do so, 0 to never emit PAUSE.
453  * pause_autoneg = 1 means PAUSE will be negotiated if possible and the
454  *                 negotiated settings will override rx_pause/tx_pause.
455  *                 Otherwise rx_pause/tx_pause are applied forcibly.
456  */
457 static int t4_pause_settings = PAUSE_RX | PAUSE_TX | PAUSE_AUTONEG;
458 SYSCTL_INT(_hw_cxgbe, OID_AUTO, pause_settings, CTLFLAG_RDTUN,
459     &t4_pause_settings, 0,
460     "PAUSE settings (bit 0 = rx_pause, 1 = tx_pause, 2 = pause_autoneg)");
461 
462 /*
463  * Forward Error Correction settings (bit 0, 1 = RS, BASER respectively).
464  * -1 to run with the firmware default.  Same as FEC_AUTO (bit 5)
465  *  0 to disable FEC.
466  */
467 static int t4_fec = -1;
468 SYSCTL_INT(_hw_cxgbe, OID_AUTO, fec, CTLFLAG_RDTUN, &t4_fec, 0,
469     "Forward Error Correction (bit 0 = RS, bit 1 = BASER_RS)");
470 
471 /*
472  * Link autonegotiation.
473  * -1 to run with the firmware default.
474  *  0 to disable.
475  *  1 to enable.
476  */
477 static int t4_autoneg = -1;
478 SYSCTL_INT(_hw_cxgbe, OID_AUTO, autoneg, CTLFLAG_RDTUN, &t4_autoneg, 0,
479     "Link autonegotiation");
480 
481 /*
482  * Firmware auto-install by driver during attach (0, 1, 2 = prohibited, allowed,
483  * encouraged respectively).
484  */
485 static unsigned int t4_fw_install = 1;
486 SYSCTL_INT(_hw_cxgbe, OID_AUTO, fw_install, CTLFLAG_RDTUN, &t4_fw_install, 0,
487     "Firmware auto-install (0 = prohibited, 1 = allowed, 2 = encouraged)");
488 
489 /*
490  * ASIC features that will be used.  Disable the ones you don't want so that the
491  * chip resources aren't wasted on features that will not be used.
492  */
493 static int t4_nbmcaps_allowed = 0;
494 SYSCTL_INT(_hw_cxgbe, OID_AUTO, nbmcaps_allowed, CTLFLAG_RDTUN,
495     &t4_nbmcaps_allowed, 0, "Default NBM capabilities");
496 
497 static int t4_linkcaps_allowed = 0;	/* No DCBX, PPP, etc. by default */
498 SYSCTL_INT(_hw_cxgbe, OID_AUTO, linkcaps_allowed, CTLFLAG_RDTUN,
499     &t4_linkcaps_allowed, 0, "Default link capabilities");
500 
501 static int t4_switchcaps_allowed = FW_CAPS_CONFIG_SWITCH_INGRESS |
502     FW_CAPS_CONFIG_SWITCH_EGRESS;
503 SYSCTL_INT(_hw_cxgbe, OID_AUTO, switchcaps_allowed, CTLFLAG_RDTUN,
504     &t4_switchcaps_allowed, 0, "Default switch capabilities");
505 
506 #ifdef RATELIMIT
507 static int t4_niccaps_allowed = FW_CAPS_CONFIG_NIC |
508 	FW_CAPS_CONFIG_NIC_HASHFILTER | FW_CAPS_CONFIG_NIC_ETHOFLD;
509 #else
510 static int t4_niccaps_allowed = FW_CAPS_CONFIG_NIC |
511 	FW_CAPS_CONFIG_NIC_HASHFILTER;
512 #endif
513 SYSCTL_INT(_hw_cxgbe, OID_AUTO, niccaps_allowed, CTLFLAG_RDTUN,
514     &t4_niccaps_allowed, 0, "Default NIC capabilities");
515 
516 static int t4_toecaps_allowed = -1;
517 SYSCTL_INT(_hw_cxgbe, OID_AUTO, toecaps_allowed, CTLFLAG_RDTUN,
518     &t4_toecaps_allowed, 0, "Default TCP offload capabilities");
519 
520 static int t4_rdmacaps_allowed = -1;
521 SYSCTL_INT(_hw_cxgbe, OID_AUTO, rdmacaps_allowed, CTLFLAG_RDTUN,
522     &t4_rdmacaps_allowed, 0, "Default RDMA capabilities");
523 
524 static int t4_cryptocaps_allowed = -1;
525 SYSCTL_INT(_hw_cxgbe, OID_AUTO, cryptocaps_allowed, CTLFLAG_RDTUN,
526     &t4_cryptocaps_allowed, 0, "Default crypto capabilities");
527 
528 static int t4_iscsicaps_allowed = -1;
529 SYSCTL_INT(_hw_cxgbe, OID_AUTO, iscsicaps_allowed, CTLFLAG_RDTUN,
530     &t4_iscsicaps_allowed, 0, "Default iSCSI capabilities");
531 
532 static int t4_fcoecaps_allowed = 0;
533 SYSCTL_INT(_hw_cxgbe, OID_AUTO, fcoecaps_allowed, CTLFLAG_RDTUN,
534     &t4_fcoecaps_allowed, 0, "Default FCoE capabilities");
535 
536 static int t5_write_combine = 0;
537 SYSCTL_INT(_hw_cxl, OID_AUTO, write_combine, CTLFLAG_RDTUN, &t5_write_combine,
538     0, "Use WC instead of UC for BAR2");
539 
540 static int t4_num_vis = 1;
541 SYSCTL_INT(_hw_cxgbe, OID_AUTO, num_vis, CTLFLAG_RDTUN, &t4_num_vis, 0,
542     "Number of VIs per port");
543 
544 /*
545  * PCIe Relaxed Ordering.
546  * -1: driver should figure out a good value.
547  * 0: disable RO.
548  * 1: enable RO.
549  * 2: leave RO alone.
550  */
551 static int pcie_relaxed_ordering = -1;
552 SYSCTL_INT(_hw_cxgbe, OID_AUTO, pcie_relaxed_ordering, CTLFLAG_RDTUN,
553     &pcie_relaxed_ordering, 0,
554     "PCIe Relaxed Ordering: 0 = disable, 1 = enable, 2 = leave alone");
555 
556 static int t4_panic_on_fatal_err = 0;
557 SYSCTL_INT(_hw_cxgbe, OID_AUTO, panic_on_fatal_err, CTLFLAG_RDTUN,
558     &t4_panic_on_fatal_err, 0, "panic on fatal firmware errors");
559 
560 #ifdef TCP_OFFLOAD
561 /*
562  * TOE tunables.
563  */
564 static int t4_cop_managed_offloading = 0;
565 SYSCTL_INT(_hw_cxgbe, OID_AUTO, cop_managed_offloading, CTLFLAG_RDTUN,
566     &t4_cop_managed_offloading, 0,
567     "COP (Connection Offload Policy) controls all TOE offload");
568 #endif
569 
570 /* Functions used by VIs to obtain unique MAC addresses for each VI. */
571 static int vi_mac_funcs[] = {
572 	FW_VI_FUNC_ETH,
573 	FW_VI_FUNC_OFLD,
574 	FW_VI_FUNC_IWARP,
575 	FW_VI_FUNC_OPENISCSI,
576 	FW_VI_FUNC_OPENFCOE,
577 	FW_VI_FUNC_FOISCSI,
578 	FW_VI_FUNC_FOFCOE,
579 };
580 
581 struct intrs_and_queues {
582 	uint16_t intr_type;	/* INTx, MSI, or MSI-X */
583 	uint16_t num_vis;	/* number of VIs for each port */
584 	uint16_t nirq;		/* Total # of vectors */
585 	uint16_t ntxq;		/* # of NIC txq's for each port */
586 	uint16_t nrxq;		/* # of NIC rxq's for each port */
587 	uint16_t nofldtxq;	/* # of TOE/ETHOFLD txq's for each port */
588 	uint16_t nofldrxq;	/* # of TOE rxq's for each port */
589 
590 	/* The vcxgbe/vcxl interfaces use these and not the ones above. */
591 	uint16_t ntxq_vi;	/* # of NIC txq's */
592 	uint16_t nrxq_vi;	/* # of NIC rxq's */
593 	uint16_t nofldtxq_vi;	/* # of TOE txq's */
594 	uint16_t nofldrxq_vi;	/* # of TOE rxq's */
595 	uint16_t nnmtxq_vi;	/* # of netmap txq's */
596 	uint16_t nnmrxq_vi;	/* # of netmap rxq's */
597 };
598 
599 static void setup_memwin(struct adapter *);
600 static void position_memwin(struct adapter *, int, uint32_t);
601 static int validate_mem_range(struct adapter *, uint32_t, uint32_t);
602 static int fwmtype_to_hwmtype(int);
603 static int validate_mt_off_len(struct adapter *, int, uint32_t, uint32_t,
604     uint32_t *);
605 static int fixup_devlog_params(struct adapter *);
606 static int cfg_itype_and_nqueues(struct adapter *, struct intrs_and_queues *);
607 static int contact_firmware(struct adapter *);
608 static int partition_resources(struct adapter *);
609 static int get_params__pre_init(struct adapter *);
610 static int get_params__post_init(struct adapter *);
611 static int set_params__post_init(struct adapter *);
612 static void t4_set_desc(struct adapter *);
613 static bool fixed_ifmedia(struct port_info *);
614 static void build_medialist(struct port_info *);
615 static void init_link_config(struct port_info *);
616 static int fixup_link_config(struct port_info *);
617 static int apply_link_config(struct port_info *);
618 static int cxgbe_init_synchronized(struct vi_info *);
619 static int cxgbe_uninit_synchronized(struct vi_info *);
620 static void quiesce_txq(struct adapter *, struct sge_txq *);
621 static void quiesce_wrq(struct adapter *, struct sge_wrq *);
622 static void quiesce_iq(struct adapter *, struct sge_iq *);
623 static void quiesce_fl(struct adapter *, struct sge_fl *);
624 static int t4_alloc_irq(struct adapter *, struct irq *, int rid,
625     driver_intr_t *, void *, char *);
626 static int t4_free_irq(struct adapter *, struct irq *);
627 static void get_regs(struct adapter *, struct t4_regdump *, uint8_t *);
628 static void vi_refresh_stats(struct adapter *, struct vi_info *);
629 static void cxgbe_refresh_stats(struct adapter *, struct port_info *);
630 static void cxgbe_tick(void *);
631 static void cxgbe_sysctls(struct port_info *);
632 static int sysctl_int_array(SYSCTL_HANDLER_ARGS);
633 static int sysctl_bitfield_8b(SYSCTL_HANDLER_ARGS);
634 static int sysctl_bitfield_16b(SYSCTL_HANDLER_ARGS);
635 static int sysctl_btphy(SYSCTL_HANDLER_ARGS);
636 static int sysctl_noflowq(SYSCTL_HANDLER_ARGS);
637 static int sysctl_holdoff_tmr_idx(SYSCTL_HANDLER_ARGS);
638 static int sysctl_holdoff_pktc_idx(SYSCTL_HANDLER_ARGS);
639 static int sysctl_qsize_rxq(SYSCTL_HANDLER_ARGS);
640 static int sysctl_qsize_txq(SYSCTL_HANDLER_ARGS);
641 static int sysctl_pause_settings(SYSCTL_HANDLER_ARGS);
642 static int sysctl_fec(SYSCTL_HANDLER_ARGS);
643 static int sysctl_autoneg(SYSCTL_HANDLER_ARGS);
644 static int sysctl_handle_t4_reg64(SYSCTL_HANDLER_ARGS);
645 static int sysctl_temperature(SYSCTL_HANDLER_ARGS);
646 static int sysctl_loadavg(SYSCTL_HANDLER_ARGS);
647 static int sysctl_cctrl(SYSCTL_HANDLER_ARGS);
648 static int sysctl_cim_ibq_obq(SYSCTL_HANDLER_ARGS);
649 static int sysctl_cim_la(SYSCTL_HANDLER_ARGS);
650 static int sysctl_cim_la_t6(SYSCTL_HANDLER_ARGS);
651 static int sysctl_cim_ma_la(SYSCTL_HANDLER_ARGS);
652 static int sysctl_cim_pif_la(SYSCTL_HANDLER_ARGS);
653 static int sysctl_cim_qcfg(SYSCTL_HANDLER_ARGS);
654 static int sysctl_cpl_stats(SYSCTL_HANDLER_ARGS);
655 static int sysctl_ddp_stats(SYSCTL_HANDLER_ARGS);
656 static int sysctl_devlog(SYSCTL_HANDLER_ARGS);
657 static int sysctl_fcoe_stats(SYSCTL_HANDLER_ARGS);
658 static int sysctl_hw_sched(SYSCTL_HANDLER_ARGS);
659 static int sysctl_lb_stats(SYSCTL_HANDLER_ARGS);
660 static int sysctl_linkdnrc(SYSCTL_HANDLER_ARGS);
661 static int sysctl_meminfo(SYSCTL_HANDLER_ARGS);
662 static int sysctl_mps_tcam(SYSCTL_HANDLER_ARGS);
663 static int sysctl_mps_tcam_t6(SYSCTL_HANDLER_ARGS);
664 static int sysctl_path_mtus(SYSCTL_HANDLER_ARGS);
665 static int sysctl_pm_stats(SYSCTL_HANDLER_ARGS);
666 static int sysctl_rdma_stats(SYSCTL_HANDLER_ARGS);
667 static int sysctl_tcp_stats(SYSCTL_HANDLER_ARGS);
668 static int sysctl_tids(SYSCTL_HANDLER_ARGS);
669 static int sysctl_tp_err_stats(SYSCTL_HANDLER_ARGS);
670 static int sysctl_tp_la_mask(SYSCTL_HANDLER_ARGS);
671 static int sysctl_tp_la(SYSCTL_HANDLER_ARGS);
672 static int sysctl_tx_rate(SYSCTL_HANDLER_ARGS);
673 static int sysctl_ulprx_la(SYSCTL_HANDLER_ARGS);
674 static int sysctl_wcwr_stats(SYSCTL_HANDLER_ARGS);
675 static int sysctl_cpus(SYSCTL_HANDLER_ARGS);
676 #ifdef TCP_OFFLOAD
677 static int sysctl_tls_rx_ports(SYSCTL_HANDLER_ARGS);
678 static int sysctl_tp_tick(SYSCTL_HANDLER_ARGS);
679 static int sysctl_tp_dack_timer(SYSCTL_HANDLER_ARGS);
680 static int sysctl_tp_timer(SYSCTL_HANDLER_ARGS);
681 static int sysctl_tp_shift_cnt(SYSCTL_HANDLER_ARGS);
682 static int sysctl_tp_backoff(SYSCTL_HANDLER_ARGS);
683 static int sysctl_holdoff_tmr_idx_ofld(SYSCTL_HANDLER_ARGS);
684 static int sysctl_holdoff_pktc_idx_ofld(SYSCTL_HANDLER_ARGS);
685 #endif
686 static int get_sge_context(struct adapter *, struct t4_sge_context *);
687 static int load_fw(struct adapter *, struct t4_data *);
688 static int load_cfg(struct adapter *, struct t4_data *);
689 static int load_boot(struct adapter *, struct t4_bootrom *);
690 static int load_bootcfg(struct adapter *, struct t4_data *);
691 static int cudbg_dump(struct adapter *, struct t4_cudbg_dump *);
692 static void free_offload_policy(struct t4_offload_policy *);
693 static int set_offload_policy(struct adapter *, struct t4_offload_policy *);
694 static int read_card_mem(struct adapter *, int, struct t4_mem_range *);
695 static int read_i2c(struct adapter *, struct t4_i2c_data *);
696 #ifdef TCP_OFFLOAD
697 static int toe_capability(struct vi_info *, int);
698 #endif
699 static int mod_event(module_t, int, void *);
700 static int notify_siblings(device_t, int);
701 
702 struct {
703 	uint16_t device;
704 	char *desc;
705 } t4_pciids[] = {
706 	{0xa000, "Chelsio Terminator 4 FPGA"},
707 	{0x4400, "Chelsio T440-dbg"},
708 	{0x4401, "Chelsio T420-CR"},
709 	{0x4402, "Chelsio T422-CR"},
710 	{0x4403, "Chelsio T440-CR"},
711 	{0x4404, "Chelsio T420-BCH"},
712 	{0x4405, "Chelsio T440-BCH"},
713 	{0x4406, "Chelsio T440-CH"},
714 	{0x4407, "Chelsio T420-SO"},
715 	{0x4408, "Chelsio T420-CX"},
716 	{0x4409, "Chelsio T420-BT"},
717 	{0x440a, "Chelsio T404-BT"},
718 	{0x440e, "Chelsio T440-LP-CR"},
719 }, t5_pciids[] = {
720 	{0xb000, "Chelsio Terminator 5 FPGA"},
721 	{0x5400, "Chelsio T580-dbg"},
722 	{0x5401,  "Chelsio T520-CR"},		/* 2 x 10G */
723 	{0x5402,  "Chelsio T522-CR"},		/* 2 x 10G, 2 X 1G */
724 	{0x5403,  "Chelsio T540-CR"},		/* 4 x 10G */
725 	{0x5407,  "Chelsio T520-SO"},		/* 2 x 10G, nomem */
726 	{0x5409,  "Chelsio T520-BT"},		/* 2 x 10GBaseT */
727 	{0x540a,  "Chelsio T504-BT"},		/* 4 x 1G */
728 	{0x540d,  "Chelsio T580-CR"},		/* 2 x 40G */
729 	{0x540e,  "Chelsio T540-LP-CR"},	/* 4 x 10G */
730 	{0x5410,  "Chelsio T580-LP-CR"},	/* 2 x 40G */
731 	{0x5411,  "Chelsio T520-LL-CR"},	/* 2 x 10G */
732 	{0x5412,  "Chelsio T560-CR"},		/* 1 x 40G, 2 x 10G */
733 	{0x5414,  "Chelsio T580-LP-SO-CR"},	/* 2 x 40G, nomem */
734 	{0x5415,  "Chelsio T502-BT"},		/* 2 x 1G */
735 	{0x5418,  "Chelsio T540-BT"},		/* 4 x 10GBaseT */
736 	{0x5419,  "Chelsio T540-LP-BT"},	/* 4 x 10GBaseT */
737 	{0x541a,  "Chelsio T540-SO-BT"},	/* 4 x 10GBaseT, nomem */
738 	{0x541b,  "Chelsio T540-SO-CR"},	/* 4 x 10G, nomem */
739 }, t6_pciids[] = {
740 	{0xc006, "Chelsio Terminator 6 FPGA"},	/* T6 PE10K6 FPGA (PF0) */
741 	{0x6400, "Chelsio T6-DBG-25"},		/* 2 x 10/25G, debug */
742 	{0x6401, "Chelsio T6225-CR"},		/* 2 x 10/25G */
743 	{0x6402, "Chelsio T6225-SO-CR"},	/* 2 x 10/25G, nomem */
744 	{0x6403, "Chelsio T6425-CR"},		/* 4 x 10/25G */
745 	{0x6404, "Chelsio T6425-SO-CR"},	/* 4 x 10/25G, nomem */
746 	{0x6405, "Chelsio T6225-OCP-SO"},	/* 2 x 10/25G, nomem */
747 	{0x6406, "Chelsio T62100-OCP-SO"},	/* 2 x 40/50/100G, nomem */
748 	{0x6407, "Chelsio T62100-LP-CR"},	/* 2 x 40/50/100G */
749 	{0x6408, "Chelsio T62100-SO-CR"},	/* 2 x 40/50/100G, nomem */
750 	{0x6409, "Chelsio T6210-BT"},		/* 2 x 10GBASE-T */
751 	{0x640d, "Chelsio T62100-CR"},		/* 2 x 40/50/100G */
752 	{0x6410, "Chelsio T6-DBG-100"},		/* 2 x 40/50/100G, debug */
753 	{0x6411, "Chelsio T6225-LL-CR"},	/* 2 x 10/25G */
754 	{0x6414, "Chelsio T61100-OCP-SO"},	/* 1 x 40/50/100G, nomem */
755 	{0x6415, "Chelsio T6201-BT"},		/* 2 x 1000BASE-T */
756 
757 	/* Custom */
758 	{0x6480, "Custom T6225-CR"},
759 	{0x6481, "Custom T62100-CR"},
760 	{0x6482, "Custom T6225-CR"},
761 	{0x6483, "Custom T62100-CR"},
762 	{0x6484, "Custom T64100-CR"},
763 	{0x6485, "Custom T6240-SO"},
764 	{0x6486, "Custom T6225-SO-CR"},
765 	{0x6487, "Custom T6225-CR"},
766 };
767 
768 #ifdef TCP_OFFLOAD
769 /*
770  * service_iq_fl() has an iq and needs the fl.  Offset of fl from the iq should
771  * be exactly the same for both rxq and ofld_rxq.
772  */
773 CTASSERT(offsetof(struct sge_ofld_rxq, iq) == offsetof(struct sge_rxq, iq));
774 CTASSERT(offsetof(struct sge_ofld_rxq, fl) == offsetof(struct sge_rxq, fl));
775 #endif
776 CTASSERT(sizeof(struct cluster_metadata) <= CL_METADATA_SIZE);
777 
778 static int
779 t4_probe(device_t dev)
780 {
781 	int i;
782 	uint16_t v = pci_get_vendor(dev);
783 	uint16_t d = pci_get_device(dev);
784 	uint8_t f = pci_get_function(dev);
785 
786 	if (v != PCI_VENDOR_ID_CHELSIO)
787 		return (ENXIO);
788 
789 	/* Attach only to PF0 of the FPGA */
790 	if (d == 0xa000 && f != 0)
791 		return (ENXIO);
792 
793 	for (i = 0; i < nitems(t4_pciids); i++) {
794 		if (d == t4_pciids[i].device) {
795 			device_set_desc(dev, t4_pciids[i].desc);
796 			return (BUS_PROBE_DEFAULT);
797 		}
798 	}
799 
800 	return (ENXIO);
801 }
802 
803 static int
804 t5_probe(device_t dev)
805 {
806 	int i;
807 	uint16_t v = pci_get_vendor(dev);
808 	uint16_t d = pci_get_device(dev);
809 	uint8_t f = pci_get_function(dev);
810 
811 	if (v != PCI_VENDOR_ID_CHELSIO)
812 		return (ENXIO);
813 
814 	/* Attach only to PF0 of the FPGA */
815 	if (d == 0xb000 && f != 0)
816 		return (ENXIO);
817 
818 	for (i = 0; i < nitems(t5_pciids); i++) {
819 		if (d == t5_pciids[i].device) {
820 			device_set_desc(dev, t5_pciids[i].desc);
821 			return (BUS_PROBE_DEFAULT);
822 		}
823 	}
824 
825 	return (ENXIO);
826 }
827 
828 static int
829 t6_probe(device_t dev)
830 {
831 	int i;
832 	uint16_t v = pci_get_vendor(dev);
833 	uint16_t d = pci_get_device(dev);
834 
835 	if (v != PCI_VENDOR_ID_CHELSIO)
836 		return (ENXIO);
837 
838 	for (i = 0; i < nitems(t6_pciids); i++) {
839 		if (d == t6_pciids[i].device) {
840 			device_set_desc(dev, t6_pciids[i].desc);
841 			return (BUS_PROBE_DEFAULT);
842 		}
843 	}
844 
845 	return (ENXIO);
846 }
847 
848 static void
849 t5_attribute_workaround(device_t dev)
850 {
851 	device_t root_port;
852 	uint32_t v;
853 
854 	/*
855 	 * The T5 chips do not properly echo the No Snoop and Relaxed
856 	 * Ordering attributes when replying to a TLP from a Root
857 	 * Port.  As a workaround, find the parent Root Port and
858 	 * disable No Snoop and Relaxed Ordering.  Note that this
859 	 * affects all devices under this root port.
860 	 */
861 	root_port = pci_find_pcie_root_port(dev);
862 	if (root_port == NULL) {
863 		device_printf(dev, "Unable to find parent root port\n");
864 		return;
865 	}
866 
867 	v = pcie_adjust_config(root_port, PCIER_DEVICE_CTL,
868 	    PCIEM_CTL_RELAXED_ORD_ENABLE | PCIEM_CTL_NOSNOOP_ENABLE, 0, 2);
869 	if ((v & (PCIEM_CTL_RELAXED_ORD_ENABLE | PCIEM_CTL_NOSNOOP_ENABLE)) !=
870 	    0)
871 		device_printf(dev, "Disabled No Snoop/Relaxed Ordering on %s\n",
872 		    device_get_nameunit(root_port));
873 }
874 
875 static const struct devnames devnames[] = {
876 	{
877 		.nexus_name = "t4nex",
878 		.ifnet_name = "cxgbe",
879 		.vi_ifnet_name = "vcxgbe",
880 		.pf03_drv_name = "t4iov",
881 		.vf_nexus_name = "t4vf",
882 		.vf_ifnet_name = "cxgbev"
883 	}, {
884 		.nexus_name = "t5nex",
885 		.ifnet_name = "cxl",
886 		.vi_ifnet_name = "vcxl",
887 		.pf03_drv_name = "t5iov",
888 		.vf_nexus_name = "t5vf",
889 		.vf_ifnet_name = "cxlv"
890 	}, {
891 		.nexus_name = "t6nex",
892 		.ifnet_name = "cc",
893 		.vi_ifnet_name = "vcc",
894 		.pf03_drv_name = "t6iov",
895 		.vf_nexus_name = "t6vf",
896 		.vf_ifnet_name = "ccv"
897 	}
898 };
899 
900 void
901 t4_init_devnames(struct adapter *sc)
902 {
903 	int id;
904 
905 	id = chip_id(sc);
906 	if (id >= CHELSIO_T4 && id - CHELSIO_T4 < nitems(devnames))
907 		sc->names = &devnames[id - CHELSIO_T4];
908 	else {
909 		device_printf(sc->dev, "chip id %d is not supported.\n", id);
910 		sc->names = NULL;
911 	}
912 }
913 
914 static int
915 t4_ifnet_unit(struct adapter *sc, struct port_info *pi)
916 {
917 	const char *parent, *name;
918 	long value;
919 	int line, unit;
920 
921 	line = 0;
922 	parent = device_get_nameunit(sc->dev);
923 	name = sc->names->ifnet_name;
924 	while (resource_find_dev(&line, name, &unit, "at", parent) == 0) {
925 		if (resource_long_value(name, unit, "port", &value) == 0 &&
926 		    value == pi->port_id)
927 			return (unit);
928 	}
929 	return (-1);
930 }
931 
932 static int
933 t4_attach(device_t dev)
934 {
935 	struct adapter *sc;
936 	int rc = 0, i, j, rqidx, tqidx, nports;
937 	struct make_dev_args mda;
938 	struct intrs_and_queues iaq;
939 	struct sge *s;
940 	uint32_t *buf;
941 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
942 	int ofld_tqidx;
943 #endif
944 #ifdef TCP_OFFLOAD
945 	int ofld_rqidx;
946 #endif
947 #ifdef DEV_NETMAP
948 	int nm_rqidx, nm_tqidx;
949 #endif
950 	int num_vis;
951 
952 	sc = device_get_softc(dev);
953 	sc->dev = dev;
954 	TUNABLE_INT_FETCH("hw.cxgbe.dflags", &sc->debug_flags);
955 
956 	if ((pci_get_device(dev) & 0xff00) == 0x5400)
957 		t5_attribute_workaround(dev);
958 	pci_enable_busmaster(dev);
959 	if (pci_find_cap(dev, PCIY_EXPRESS, &i) == 0) {
960 		uint32_t v;
961 
962 		pci_set_max_read_req(dev, 4096);
963 		v = pci_read_config(dev, i + PCIER_DEVICE_CTL, 2);
964 		sc->params.pci.mps = 128 << ((v & PCIEM_CTL_MAX_PAYLOAD) >> 5);
965 		if (pcie_relaxed_ordering == 0 &&
966 		    (v & PCIEM_CTL_RELAXED_ORD_ENABLE) != 0) {
967 			v &= ~PCIEM_CTL_RELAXED_ORD_ENABLE;
968 			pci_write_config(dev, i + PCIER_DEVICE_CTL, v, 2);
969 		} else if (pcie_relaxed_ordering == 1 &&
970 		    (v & PCIEM_CTL_RELAXED_ORD_ENABLE) == 0) {
971 			v |= PCIEM_CTL_RELAXED_ORD_ENABLE;
972 			pci_write_config(dev, i + PCIER_DEVICE_CTL, v, 2);
973 		}
974 	}
975 
976 	sc->sge_gts_reg = MYPF_REG(A_SGE_PF_GTS);
977 	sc->sge_kdoorbell_reg = MYPF_REG(A_SGE_PF_KDOORBELL);
978 	sc->traceq = -1;
979 	mtx_init(&sc->ifp_lock, sc->ifp_lockname, 0, MTX_DEF);
980 	snprintf(sc->ifp_lockname, sizeof(sc->ifp_lockname), "%s tracer",
981 	    device_get_nameunit(dev));
982 
983 	snprintf(sc->lockname, sizeof(sc->lockname), "%s",
984 	    device_get_nameunit(dev));
985 	mtx_init(&sc->sc_lock, sc->lockname, 0, MTX_DEF);
986 	t4_add_adapter(sc);
987 
988 	mtx_init(&sc->sfl_lock, "starving freelists", 0, MTX_DEF);
989 	TAILQ_INIT(&sc->sfl);
990 	callout_init_mtx(&sc->sfl_callout, &sc->sfl_lock, 0);
991 
992 	mtx_init(&sc->reg_lock, "indirect register access", 0, MTX_DEF);
993 
994 	sc->policy = NULL;
995 	rw_init(&sc->policy_lock, "connection offload policy");
996 
997 	rc = t4_map_bars_0_and_4(sc);
998 	if (rc != 0)
999 		goto done; /* error message displayed already */
1000 
1001 	memset(sc->chan_map, 0xff, sizeof(sc->chan_map));
1002 
1003 	/* Prepare the adapter for operation. */
1004 	buf = malloc(PAGE_SIZE, M_CXGBE, M_ZERO | M_WAITOK);
1005 	rc = -t4_prep_adapter(sc, buf);
1006 	free(buf, M_CXGBE);
1007 	if (rc != 0) {
1008 		device_printf(dev, "failed to prepare adapter: %d.\n", rc);
1009 		goto done;
1010 	}
1011 
1012 	/*
1013 	 * This is the real PF# to which we're attaching.  Works from within PCI
1014 	 * passthrough environments too, where pci_get_function() could return a
1015 	 * different PF# depending on the passthrough configuration.  We need to
1016 	 * use the real PF# in all our communication with the firmware.
1017 	 */
1018 	j = t4_read_reg(sc, A_PL_WHOAMI);
1019 	sc->pf = chip_id(sc) <= CHELSIO_T5 ? G_SOURCEPF(j) : G_T6_SOURCEPF(j);
1020 	sc->mbox = sc->pf;
1021 
1022 	t4_init_devnames(sc);
1023 	if (sc->names == NULL) {
1024 		rc = ENOTSUP;
1025 		goto done; /* error message displayed already */
1026 	}
1027 
1028 	/*
1029 	 * Do this really early, with the memory windows set up even before the
1030 	 * character device.  The userland tool's register i/o and mem read
1031 	 * will work even in "recovery mode".
1032 	 */
1033 	setup_memwin(sc);
1034 	if (t4_init_devlog_params(sc, 0) == 0)
1035 		fixup_devlog_params(sc);
1036 	make_dev_args_init(&mda);
1037 	mda.mda_devsw = &t4_cdevsw;
1038 	mda.mda_uid = UID_ROOT;
1039 	mda.mda_gid = GID_WHEEL;
1040 	mda.mda_mode = 0600;
1041 	mda.mda_si_drv1 = sc;
1042 	rc = make_dev_s(&mda, &sc->cdev, "%s", device_get_nameunit(dev));
1043 	if (rc != 0)
1044 		device_printf(dev, "failed to create nexus char device: %d.\n",
1045 		    rc);
1046 
1047 	/* Go no further if recovery mode has been requested. */
1048 	if (TUNABLE_INT_FETCH("hw.cxgbe.sos", &i) && i != 0) {
1049 		device_printf(dev, "recovery mode.\n");
1050 		goto done;
1051 	}
1052 
1053 #if defined(__i386__)
1054 	if ((cpu_feature & CPUID_CX8) == 0) {
1055 		device_printf(dev, "64 bit atomics not available.\n");
1056 		rc = ENOTSUP;
1057 		goto done;
1058 	}
1059 #endif
1060 
1061 	/* Contact the firmware and try to become the master driver. */
1062 	rc = contact_firmware(sc);
1063 	if (rc != 0)
1064 		goto done; /* error message displayed already */
1065 	MPASS(sc->flags & FW_OK);
1066 
1067 	rc = get_params__pre_init(sc);
1068 	if (rc != 0)
1069 		goto done; /* error message displayed already */
1070 
1071 	if (sc->flags & MASTER_PF) {
1072 		rc = partition_resources(sc);
1073 		if (rc != 0)
1074 			goto done; /* error message displayed already */
1075 	}
1076 
1077 	rc = get_params__post_init(sc);
1078 	if (rc != 0)
1079 		goto done; /* error message displayed already */
1080 
1081 	rc = set_params__post_init(sc);
1082 	if (rc != 0)
1083 		goto done; /* error message displayed already */
1084 
1085 	rc = t4_map_bar_2(sc);
1086 	if (rc != 0)
1087 		goto done; /* error message displayed already */
1088 
1089 	rc = t4_create_dma_tag(sc);
1090 	if (rc != 0)
1091 		goto done; /* error message displayed already */
1092 
1093 	/*
1094 	 * First pass over all the ports - allocate VIs and initialize some
1095 	 * basic parameters like mac address, port type, etc.
1096 	 */
1097 	for_each_port(sc, i) {
1098 		struct port_info *pi;
1099 
1100 		pi = malloc(sizeof(*pi), M_CXGBE, M_ZERO | M_WAITOK);
1101 		sc->port[i] = pi;
1102 
1103 		/* These must be set before t4_port_init */
1104 		pi->adapter = sc;
1105 		pi->port_id = i;
1106 		/*
1107 		 * XXX: vi[0] is special so we can't delay this allocation until
1108 		 * pi->nvi's final value is known.
1109 		 */
1110 		pi->vi = malloc(sizeof(struct vi_info) * t4_num_vis, M_CXGBE,
1111 		    M_ZERO | M_WAITOK);
1112 
1113 		/*
1114 		 * Allocate the "main" VI and initialize parameters
1115 		 * like mac addr.
1116 		 */
1117 		rc = -t4_port_init(sc, sc->mbox, sc->pf, 0, i);
1118 		if (rc != 0) {
1119 			device_printf(dev, "unable to initialize port %d: %d\n",
1120 			    i, rc);
1121 			free(pi->vi, M_CXGBE);
1122 			free(pi, M_CXGBE);
1123 			sc->port[i] = NULL;
1124 			goto done;
1125 		}
1126 
1127 		snprintf(pi->lockname, sizeof(pi->lockname), "%sp%d",
1128 		    device_get_nameunit(dev), i);
1129 		mtx_init(&pi->pi_lock, pi->lockname, 0, MTX_DEF);
1130 		sc->chan_map[pi->tx_chan] = i;
1131 
1132 		/* All VIs on this port share this media. */
1133 		ifmedia_init(&pi->media, IFM_IMASK, cxgbe_media_change,
1134 		    cxgbe_media_status);
1135 
1136 		PORT_LOCK(pi);
1137 		init_link_config(pi);
1138 		fixup_link_config(pi);
1139 		build_medialist(pi);
1140 		if (fixed_ifmedia(pi))
1141 			pi->flags |= FIXED_IFMEDIA;
1142 		PORT_UNLOCK(pi);
1143 
1144 		pi->dev = device_add_child(dev, sc->names->ifnet_name,
1145 		    t4_ifnet_unit(sc, pi));
1146 		if (pi->dev == NULL) {
1147 			device_printf(dev,
1148 			    "failed to add device for port %d.\n", i);
1149 			rc = ENXIO;
1150 			goto done;
1151 		}
1152 		pi->vi[0].dev = pi->dev;
1153 		device_set_softc(pi->dev, pi);
1154 	}
1155 
1156 	/*
1157 	 * Interrupt type, # of interrupts, # of rx/tx queues, etc.
1158 	 */
1159 	nports = sc->params.nports;
1160 	rc = cfg_itype_and_nqueues(sc, &iaq);
1161 	if (rc != 0)
1162 		goto done; /* error message displayed already */
1163 
1164 	num_vis = iaq.num_vis;
1165 	sc->intr_type = iaq.intr_type;
1166 	sc->intr_count = iaq.nirq;
1167 
1168 	s = &sc->sge;
1169 	s->nrxq = nports * iaq.nrxq;
1170 	s->ntxq = nports * iaq.ntxq;
1171 	if (num_vis > 1) {
1172 		s->nrxq += nports * (num_vis - 1) * iaq.nrxq_vi;
1173 		s->ntxq += nports * (num_vis - 1) * iaq.ntxq_vi;
1174 	}
1175 	s->neq = s->ntxq + s->nrxq;	/* the free list in an rxq is an eq */
1176 	s->neq += nports;		/* ctrl queues: 1 per port */
1177 	s->niq = s->nrxq + 1;		/* 1 extra for firmware event queue */
1178 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1179 	if (is_offload(sc) || is_ethoffload(sc)) {
1180 		s->nofldtxq = nports * iaq.nofldtxq;
1181 		if (num_vis > 1)
1182 			s->nofldtxq += nports * (num_vis - 1) * iaq.nofldtxq_vi;
1183 		s->neq += s->nofldtxq;
1184 
1185 		s->ofld_txq = malloc(s->nofldtxq * sizeof(struct sge_wrq),
1186 		    M_CXGBE, M_ZERO | M_WAITOK);
1187 	}
1188 #endif
1189 #ifdef TCP_OFFLOAD
1190 	if (is_offload(sc)) {
1191 		s->nofldrxq = nports * iaq.nofldrxq;
1192 		if (num_vis > 1)
1193 			s->nofldrxq += nports * (num_vis - 1) * iaq.nofldrxq_vi;
1194 		s->neq += s->nofldrxq;	/* free list */
1195 		s->niq += s->nofldrxq;
1196 
1197 		s->ofld_rxq = malloc(s->nofldrxq * sizeof(struct sge_ofld_rxq),
1198 		    M_CXGBE, M_ZERO | M_WAITOK);
1199 	}
1200 #endif
1201 #ifdef DEV_NETMAP
1202 	if (num_vis > 1) {
1203 		s->nnmrxq = nports * (num_vis - 1) * iaq.nnmrxq_vi;
1204 		s->nnmtxq = nports * (num_vis - 1) * iaq.nnmtxq_vi;
1205 	}
1206 	s->neq += s->nnmtxq + s->nnmrxq;
1207 	s->niq += s->nnmrxq;
1208 
1209 	s->nm_rxq = malloc(s->nnmrxq * sizeof(struct sge_nm_rxq),
1210 	    M_CXGBE, M_ZERO | M_WAITOK);
1211 	s->nm_txq = malloc(s->nnmtxq * sizeof(struct sge_nm_txq),
1212 	    M_CXGBE, M_ZERO | M_WAITOK);
1213 #endif
1214 
1215 	s->ctrlq = malloc(nports * sizeof(struct sge_wrq), M_CXGBE,
1216 	    M_ZERO | M_WAITOK);
1217 	s->rxq = malloc(s->nrxq * sizeof(struct sge_rxq), M_CXGBE,
1218 	    M_ZERO | M_WAITOK);
1219 	s->txq = malloc(s->ntxq * sizeof(struct sge_txq), M_CXGBE,
1220 	    M_ZERO | M_WAITOK);
1221 	s->iqmap = malloc(s->niq * sizeof(struct sge_iq *), M_CXGBE,
1222 	    M_ZERO | M_WAITOK);
1223 	s->eqmap = malloc(s->neq * sizeof(struct sge_eq *), M_CXGBE,
1224 	    M_ZERO | M_WAITOK);
1225 
1226 	sc->irq = malloc(sc->intr_count * sizeof(struct irq), M_CXGBE,
1227 	    M_ZERO | M_WAITOK);
1228 
1229 	t4_init_l2t(sc, M_WAITOK);
1230 	t4_init_smt(sc, M_WAITOK);
1231 	t4_init_tx_sched(sc);
1232 #ifdef RATELIMIT
1233 	t4_init_etid_table(sc);
1234 #endif
1235 #ifdef INET6
1236 	t4_init_clip_table(sc);
1237 #endif
1238 	if (sc->vres.key.size != 0)
1239 		sc->key_map = vmem_create("T4TLS key map", sc->vres.key.start,
1240 		    sc->vres.key.size, 32, 0, M_FIRSTFIT | M_WAITOK);
1241 
1242 	/*
1243 	 * Second pass over the ports.  This time we know the number of rx and
1244 	 * tx queues that each port should get.
1245 	 */
1246 	rqidx = tqidx = 0;
1247 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1248 	ofld_tqidx = 0;
1249 #endif
1250 #ifdef TCP_OFFLOAD
1251 	ofld_rqidx = 0;
1252 #endif
1253 #ifdef DEV_NETMAP
1254 	nm_rqidx = nm_tqidx = 0;
1255 #endif
1256 	for_each_port(sc, i) {
1257 		struct port_info *pi = sc->port[i];
1258 		struct vi_info *vi;
1259 
1260 		if (pi == NULL)
1261 			continue;
1262 
1263 		pi->nvi = num_vis;
1264 		for_each_vi(pi, j, vi) {
1265 			vi->pi = pi;
1266 			vi->qsize_rxq = t4_qsize_rxq;
1267 			vi->qsize_txq = t4_qsize_txq;
1268 
1269 			vi->first_rxq = rqidx;
1270 			vi->first_txq = tqidx;
1271 			vi->tmr_idx = t4_tmr_idx;
1272 			vi->pktc_idx = t4_pktc_idx;
1273 			vi->nrxq = j == 0 ? iaq.nrxq : iaq.nrxq_vi;
1274 			vi->ntxq = j == 0 ? iaq.ntxq : iaq.ntxq_vi;
1275 
1276 			rqidx += vi->nrxq;
1277 			tqidx += vi->ntxq;
1278 
1279 			if (j == 0 && vi->ntxq > 1)
1280 				vi->rsrv_noflowq = t4_rsrv_noflowq ? 1 : 0;
1281 			else
1282 				vi->rsrv_noflowq = 0;
1283 
1284 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1285 			vi->first_ofld_txq = ofld_tqidx;
1286 			vi->nofldtxq = j == 0 ? iaq.nofldtxq : iaq.nofldtxq_vi;
1287 			ofld_tqidx += vi->nofldtxq;
1288 #endif
1289 #ifdef TCP_OFFLOAD
1290 			vi->ofld_tmr_idx = t4_tmr_idx_ofld;
1291 			vi->ofld_pktc_idx = t4_pktc_idx_ofld;
1292 			vi->first_ofld_rxq = ofld_rqidx;
1293 			vi->nofldrxq = j == 0 ? iaq.nofldrxq : iaq.nofldrxq_vi;
1294 
1295 			ofld_rqidx += vi->nofldrxq;
1296 #endif
1297 #ifdef DEV_NETMAP
1298 			if (j > 0) {
1299 				vi->first_nm_rxq = nm_rqidx;
1300 				vi->first_nm_txq = nm_tqidx;
1301 				vi->nnmrxq = iaq.nnmrxq_vi;
1302 				vi->nnmtxq = iaq.nnmtxq_vi;
1303 				nm_rqidx += vi->nnmrxq;
1304 				nm_tqidx += vi->nnmtxq;
1305 			}
1306 #endif
1307 		}
1308 	}
1309 
1310 	rc = t4_setup_intr_handlers(sc);
1311 	if (rc != 0) {
1312 		device_printf(dev,
1313 		    "failed to setup interrupt handlers: %d\n", rc);
1314 		goto done;
1315 	}
1316 
1317 	rc = bus_generic_probe(dev);
1318 	if (rc != 0) {
1319 		device_printf(dev, "failed to probe child drivers: %d\n", rc);
1320 		goto done;
1321 	}
1322 
1323 	/*
1324 	 * Ensure thread-safe mailbox access (in debug builds).
1325 	 *
1326 	 * So far this was the only thread accessing the mailbox but various
1327 	 * ifnets and sysctls are about to be created and their handlers/ioctls
1328 	 * will access the mailbox from different threads.
1329 	 */
1330 	sc->flags |= CHK_MBOX_ACCESS;
1331 
1332 	rc = bus_generic_attach(dev);
1333 	if (rc != 0) {
1334 		device_printf(dev,
1335 		    "failed to attach all child ports: %d\n", rc);
1336 		goto done;
1337 	}
1338 
1339 	device_printf(dev,
1340 	    "PCIe gen%d x%d, %d ports, %d %s interrupt%s, %d eq, %d iq\n",
1341 	    sc->params.pci.speed, sc->params.pci.width, sc->params.nports,
1342 	    sc->intr_count, sc->intr_type == INTR_MSIX ? "MSI-X" :
1343 	    (sc->intr_type == INTR_MSI ? "MSI" : "INTx"),
1344 	    sc->intr_count > 1 ? "s" : "", sc->sge.neq, sc->sge.niq);
1345 
1346 	t4_set_desc(sc);
1347 
1348 	notify_siblings(dev, 0);
1349 
1350 done:
1351 	if (rc != 0 && sc->cdev) {
1352 		/* cdev was created and so cxgbetool works; recover that way. */
1353 		device_printf(dev,
1354 		    "error during attach, adapter is now in recovery mode.\n");
1355 		rc = 0;
1356 	}
1357 
1358 	if (rc != 0)
1359 		t4_detach_common(dev);
1360 	else
1361 		t4_sysctls(sc);
1362 
1363 	return (rc);
1364 }
1365 
1366 static int
1367 t4_child_location_str(device_t bus, device_t dev, char *buf, size_t buflen)
1368 {
1369 	struct port_info *pi;
1370 
1371 	pi = device_get_softc(dev);
1372 	snprintf(buf, buflen, "port=%d", pi->port_id);
1373 	return (0);
1374 }
1375 
1376 static int
1377 t4_ready(device_t dev)
1378 {
1379 	struct adapter *sc;
1380 
1381 	sc = device_get_softc(dev);
1382 	if (sc->flags & FW_OK)
1383 		return (0);
1384 	return (ENXIO);
1385 }
1386 
1387 static int
1388 t4_read_port_device(device_t dev, int port, device_t *child)
1389 {
1390 	struct adapter *sc;
1391 	struct port_info *pi;
1392 
1393 	sc = device_get_softc(dev);
1394 	if (port < 0 || port >= MAX_NPORTS)
1395 		return (EINVAL);
1396 	pi = sc->port[port];
1397 	if (pi == NULL || pi->dev == NULL)
1398 		return (ENXIO);
1399 	*child = pi->dev;
1400 	return (0);
1401 }
1402 
1403 static int
1404 notify_siblings(device_t dev, int detaching)
1405 {
1406 	device_t sibling;
1407 	int error, i;
1408 
1409 	error = 0;
1410 	for (i = 0; i < PCI_FUNCMAX; i++) {
1411 		if (i == pci_get_function(dev))
1412 			continue;
1413 		sibling = pci_find_dbsf(pci_get_domain(dev), pci_get_bus(dev),
1414 		    pci_get_slot(dev), i);
1415 		if (sibling == NULL || !device_is_attached(sibling))
1416 			continue;
1417 		if (detaching)
1418 			error = T4_DETACH_CHILD(sibling);
1419 		else
1420 			(void)T4_ATTACH_CHILD(sibling);
1421 		if (error)
1422 			break;
1423 	}
1424 	return (error);
1425 }
1426 
1427 /*
1428  * Idempotent
1429  */
1430 static int
1431 t4_detach(device_t dev)
1432 {
1433 	struct adapter *sc;
1434 	int rc;
1435 
1436 	sc = device_get_softc(dev);
1437 
1438 	rc = notify_siblings(dev, 1);
1439 	if (rc) {
1440 		device_printf(dev,
1441 		    "failed to detach sibling devices: %d\n", rc);
1442 		return (rc);
1443 	}
1444 
1445 	return (t4_detach_common(dev));
1446 }
1447 
1448 int
1449 t4_detach_common(device_t dev)
1450 {
1451 	struct adapter *sc;
1452 	struct port_info *pi;
1453 	int i, rc;
1454 
1455 	sc = device_get_softc(dev);
1456 
1457 	if (sc->cdev) {
1458 		destroy_dev(sc->cdev);
1459 		sc->cdev = NULL;
1460 	}
1461 
1462 	sc->flags &= ~CHK_MBOX_ACCESS;
1463 	if (sc->flags & FULL_INIT_DONE) {
1464 		if (!(sc->flags & IS_VF))
1465 			t4_intr_disable(sc);
1466 	}
1467 
1468 	if (device_is_attached(dev)) {
1469 		rc = bus_generic_detach(dev);
1470 		if (rc) {
1471 			device_printf(dev,
1472 			    "failed to detach child devices: %d\n", rc);
1473 			return (rc);
1474 		}
1475 	}
1476 
1477 	for (i = 0; i < sc->intr_count; i++)
1478 		t4_free_irq(sc, &sc->irq[i]);
1479 
1480 	if ((sc->flags & (IS_VF | FW_OK)) == FW_OK)
1481 		t4_free_tx_sched(sc);
1482 
1483 	for (i = 0; i < MAX_NPORTS; i++) {
1484 		pi = sc->port[i];
1485 		if (pi) {
1486 			t4_free_vi(sc, sc->mbox, sc->pf, 0, pi->vi[0].viid);
1487 			if (pi->dev)
1488 				device_delete_child(dev, pi->dev);
1489 
1490 			mtx_destroy(&pi->pi_lock);
1491 			free(pi->vi, M_CXGBE);
1492 			free(pi, M_CXGBE);
1493 		}
1494 	}
1495 
1496 	device_delete_children(dev);
1497 
1498 	if (sc->flags & FULL_INIT_DONE)
1499 		adapter_full_uninit(sc);
1500 
1501 	if ((sc->flags & (IS_VF | FW_OK)) == FW_OK)
1502 		t4_fw_bye(sc, sc->mbox);
1503 
1504 	if (sc->intr_type == INTR_MSI || sc->intr_type == INTR_MSIX)
1505 		pci_release_msi(dev);
1506 
1507 	if (sc->regs_res)
1508 		bus_release_resource(dev, SYS_RES_MEMORY, sc->regs_rid,
1509 		    sc->regs_res);
1510 
1511 	if (sc->udbs_res)
1512 		bus_release_resource(dev, SYS_RES_MEMORY, sc->udbs_rid,
1513 		    sc->udbs_res);
1514 
1515 	if (sc->msix_res)
1516 		bus_release_resource(dev, SYS_RES_MEMORY, sc->msix_rid,
1517 		    sc->msix_res);
1518 
1519 	if (sc->l2t)
1520 		t4_free_l2t(sc->l2t);
1521 	if (sc->smt)
1522 		t4_free_smt(sc->smt);
1523 #ifdef RATELIMIT
1524 	t4_free_etid_table(sc);
1525 #endif
1526 	if (sc->key_map)
1527 		vmem_destroy(sc->key_map);
1528 #ifdef INET6
1529 	t4_destroy_clip_table(sc);
1530 #endif
1531 
1532 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1533 	free(sc->sge.ofld_txq, M_CXGBE);
1534 #endif
1535 #ifdef TCP_OFFLOAD
1536 	free(sc->sge.ofld_rxq, M_CXGBE);
1537 #endif
1538 #ifdef DEV_NETMAP
1539 	free(sc->sge.nm_rxq, M_CXGBE);
1540 	free(sc->sge.nm_txq, M_CXGBE);
1541 #endif
1542 	free(sc->irq, M_CXGBE);
1543 	free(sc->sge.rxq, M_CXGBE);
1544 	free(sc->sge.txq, M_CXGBE);
1545 	free(sc->sge.ctrlq, M_CXGBE);
1546 	free(sc->sge.iqmap, M_CXGBE);
1547 	free(sc->sge.eqmap, M_CXGBE);
1548 	free(sc->tids.ftid_tab, M_CXGBE);
1549 	free(sc->tids.hpftid_tab, M_CXGBE);
1550 	free_hftid_hash(&sc->tids);
1551 	free(sc->tids.atid_tab, M_CXGBE);
1552 	free(sc->tids.tid_tab, M_CXGBE);
1553 	free(sc->tt.tls_rx_ports, M_CXGBE);
1554 	t4_destroy_dma_tag(sc);
1555 	if (mtx_initialized(&sc->sc_lock)) {
1556 		sx_xlock(&t4_list_lock);
1557 		SLIST_REMOVE(&t4_list, sc, adapter, link);
1558 		sx_xunlock(&t4_list_lock);
1559 		mtx_destroy(&sc->sc_lock);
1560 	}
1561 
1562 	callout_drain(&sc->sfl_callout);
1563 	if (mtx_initialized(&sc->tids.ftid_lock)) {
1564 		mtx_destroy(&sc->tids.ftid_lock);
1565 		cv_destroy(&sc->tids.ftid_cv);
1566 	}
1567 	if (mtx_initialized(&sc->tids.atid_lock))
1568 		mtx_destroy(&sc->tids.atid_lock);
1569 	if (mtx_initialized(&sc->sfl_lock))
1570 		mtx_destroy(&sc->sfl_lock);
1571 	if (mtx_initialized(&sc->ifp_lock))
1572 		mtx_destroy(&sc->ifp_lock);
1573 	if (mtx_initialized(&sc->reg_lock))
1574 		mtx_destroy(&sc->reg_lock);
1575 
1576 	if (rw_initialized(&sc->policy_lock)) {
1577 		rw_destroy(&sc->policy_lock);
1578 #ifdef TCP_OFFLOAD
1579 		if (sc->policy != NULL)
1580 			free_offload_policy(sc->policy);
1581 #endif
1582 	}
1583 
1584 	for (i = 0; i < NUM_MEMWIN; i++) {
1585 		struct memwin *mw = &sc->memwin[i];
1586 
1587 		if (rw_initialized(&mw->mw_lock))
1588 			rw_destroy(&mw->mw_lock);
1589 	}
1590 
1591 	bzero(sc, sizeof(*sc));
1592 
1593 	return (0);
1594 }
1595 
1596 static int
1597 cxgbe_probe(device_t dev)
1598 {
1599 	char buf[128];
1600 	struct port_info *pi = device_get_softc(dev);
1601 
1602 	snprintf(buf, sizeof(buf), "port %d", pi->port_id);
1603 	device_set_desc_copy(dev, buf);
1604 
1605 	return (BUS_PROBE_DEFAULT);
1606 }
1607 
1608 #define T4_CAP (IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | \
1609     IFCAP_VLAN_HWCSUM | IFCAP_TSO | IFCAP_JUMBO_MTU | IFCAP_LRO | \
1610     IFCAP_VLAN_HWTSO | IFCAP_LINKSTATE | IFCAP_HWCSUM_IPV6 | IFCAP_HWSTATS | \
1611     IFCAP_HWRXTSTMP)
1612 #define T4_CAP_ENABLE (T4_CAP)
1613 
1614 static int
1615 cxgbe_vi_attach(device_t dev, struct vi_info *vi)
1616 {
1617 	struct ifnet *ifp;
1618 	struct sbuf *sb;
1619 
1620 	vi->xact_addr_filt = -1;
1621 	callout_init(&vi->tick, 1);
1622 
1623 	/* Allocate an ifnet and set it up */
1624 	ifp = if_alloc(IFT_ETHER);
1625 	if (ifp == NULL) {
1626 		device_printf(dev, "Cannot allocate ifnet\n");
1627 		return (ENOMEM);
1628 	}
1629 	vi->ifp = ifp;
1630 	ifp->if_softc = vi;
1631 
1632 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1633 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1634 
1635 	ifp->if_init = cxgbe_init;
1636 	ifp->if_ioctl = cxgbe_ioctl;
1637 	ifp->if_transmit = cxgbe_transmit;
1638 	ifp->if_qflush = cxgbe_qflush;
1639 	ifp->if_get_counter = cxgbe_get_counter;
1640 #ifdef RATELIMIT
1641 	ifp->if_snd_tag_alloc = cxgbe_snd_tag_alloc;
1642 	ifp->if_snd_tag_modify = cxgbe_snd_tag_modify;
1643 	ifp->if_snd_tag_query = cxgbe_snd_tag_query;
1644 	ifp->if_snd_tag_free = cxgbe_snd_tag_free;
1645 #endif
1646 
1647 	ifp->if_capabilities = T4_CAP;
1648 	ifp->if_capenable = T4_CAP_ENABLE;
1649 #ifdef TCP_OFFLOAD
1650 	if (vi->nofldrxq != 0)
1651 		ifp->if_capabilities |= IFCAP_TOE;
1652 #endif
1653 #ifdef RATELIMIT
1654 	if (is_ethoffload(vi->pi->adapter) && vi->nofldtxq != 0) {
1655 		ifp->if_capabilities |= IFCAP_TXRTLMT;
1656 		ifp->if_capenable |= IFCAP_TXRTLMT;
1657 	}
1658 #endif
1659 	ifp->if_hwassist = CSUM_TCP | CSUM_UDP | CSUM_IP | CSUM_TSO |
1660 	    CSUM_UDP_IPV6 | CSUM_TCP_IPV6;
1661 
1662 	ifp->if_hw_tsomax = IP_MAXPACKET;
1663 	ifp->if_hw_tsomaxsegcount = TX_SGL_SEGS_TSO;
1664 #ifdef RATELIMIT
1665 	if (is_ethoffload(vi->pi->adapter) && vi->nofldtxq != 0)
1666 		ifp->if_hw_tsomaxsegcount = TX_SGL_SEGS_EO_TSO;
1667 #endif
1668 	ifp->if_hw_tsomaxsegsize = 65536;
1669 
1670 	ether_ifattach(ifp, vi->hw_addr);
1671 #ifdef DEV_NETMAP
1672 	if (vi->nnmrxq != 0)
1673 		cxgbe_nm_attach(vi);
1674 #endif
1675 	sb = sbuf_new_auto();
1676 	sbuf_printf(sb, "%d txq, %d rxq (NIC)", vi->ntxq, vi->nrxq);
1677 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
1678 	switch (ifp->if_capabilities & (IFCAP_TOE | IFCAP_TXRTLMT)) {
1679 	case IFCAP_TOE:
1680 		sbuf_printf(sb, "; %d txq (TOE)", vi->nofldtxq);
1681 		break;
1682 	case IFCAP_TOE | IFCAP_TXRTLMT:
1683 		sbuf_printf(sb, "; %d txq (TOE/ETHOFLD)", vi->nofldtxq);
1684 		break;
1685 	case IFCAP_TXRTLMT:
1686 		sbuf_printf(sb, "; %d txq (ETHOFLD)", vi->nofldtxq);
1687 		break;
1688 	}
1689 #endif
1690 #ifdef TCP_OFFLOAD
1691 	if (ifp->if_capabilities & IFCAP_TOE)
1692 		sbuf_printf(sb, ", %d rxq (TOE)", vi->nofldrxq);
1693 #endif
1694 #ifdef DEV_NETMAP
1695 	if (ifp->if_capabilities & IFCAP_NETMAP)
1696 		sbuf_printf(sb, "; %d txq, %d rxq (netmap)",
1697 		    vi->nnmtxq, vi->nnmrxq);
1698 #endif
1699 	sbuf_finish(sb);
1700 	device_printf(dev, "%s\n", sbuf_data(sb));
1701 	sbuf_delete(sb);
1702 
1703 	vi_sysctls(vi);
1704 
1705 	return (0);
1706 }
1707 
1708 static int
1709 cxgbe_attach(device_t dev)
1710 {
1711 	struct port_info *pi = device_get_softc(dev);
1712 	struct adapter *sc = pi->adapter;
1713 	struct vi_info *vi;
1714 	int i, rc;
1715 
1716 	callout_init_mtx(&pi->tick, &pi->pi_lock, 0);
1717 
1718 	rc = cxgbe_vi_attach(dev, &pi->vi[0]);
1719 	if (rc)
1720 		return (rc);
1721 
1722 	for_each_vi(pi, i, vi) {
1723 		if (i == 0)
1724 			continue;
1725 		vi->dev = device_add_child(dev, sc->names->vi_ifnet_name, -1);
1726 		if (vi->dev == NULL) {
1727 			device_printf(dev, "failed to add VI %d\n", i);
1728 			continue;
1729 		}
1730 		device_set_softc(vi->dev, vi);
1731 	}
1732 
1733 	cxgbe_sysctls(pi);
1734 
1735 	bus_generic_attach(dev);
1736 
1737 	return (0);
1738 }
1739 
1740 static void
1741 cxgbe_vi_detach(struct vi_info *vi)
1742 {
1743 	struct ifnet *ifp = vi->ifp;
1744 
1745 	ether_ifdetach(ifp);
1746 
1747 	/* Let detach proceed even if these fail. */
1748 #ifdef DEV_NETMAP
1749 	if (ifp->if_capabilities & IFCAP_NETMAP)
1750 		cxgbe_nm_detach(vi);
1751 #endif
1752 	cxgbe_uninit_synchronized(vi);
1753 	callout_drain(&vi->tick);
1754 	vi_full_uninit(vi);
1755 
1756 	if_free(vi->ifp);
1757 	vi->ifp = NULL;
1758 }
1759 
1760 static int
1761 cxgbe_detach(device_t dev)
1762 {
1763 	struct port_info *pi = device_get_softc(dev);
1764 	struct adapter *sc = pi->adapter;
1765 	int rc;
1766 
1767 	/* Detach the extra VIs first. */
1768 	rc = bus_generic_detach(dev);
1769 	if (rc)
1770 		return (rc);
1771 	device_delete_children(dev);
1772 
1773 	doom_vi(sc, &pi->vi[0]);
1774 
1775 	if (pi->flags & HAS_TRACEQ) {
1776 		sc->traceq = -1;	/* cloner should not create ifnet */
1777 		t4_tracer_port_detach(sc);
1778 	}
1779 
1780 	cxgbe_vi_detach(&pi->vi[0]);
1781 	callout_drain(&pi->tick);
1782 	ifmedia_removeall(&pi->media);
1783 
1784 	end_synchronized_op(sc, 0);
1785 
1786 	return (0);
1787 }
1788 
1789 static void
1790 cxgbe_init(void *arg)
1791 {
1792 	struct vi_info *vi = arg;
1793 	struct adapter *sc = vi->pi->adapter;
1794 
1795 	if (begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4init") != 0)
1796 		return;
1797 	cxgbe_init_synchronized(vi);
1798 	end_synchronized_op(sc, 0);
1799 }
1800 
1801 static int
1802 cxgbe_ioctl(struct ifnet *ifp, unsigned long cmd, caddr_t data)
1803 {
1804 	int rc = 0, mtu, flags;
1805 	struct vi_info *vi = ifp->if_softc;
1806 	struct port_info *pi = vi->pi;
1807 	struct adapter *sc = pi->adapter;
1808 	struct ifreq *ifr = (struct ifreq *)data;
1809 	uint32_t mask;
1810 
1811 	switch (cmd) {
1812 	case SIOCSIFMTU:
1813 		mtu = ifr->ifr_mtu;
1814 		if (mtu < ETHERMIN || mtu > MAX_MTU)
1815 			return (EINVAL);
1816 
1817 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4mtu");
1818 		if (rc)
1819 			return (rc);
1820 		ifp->if_mtu = mtu;
1821 		if (vi->flags & VI_INIT_DONE) {
1822 			t4_update_fl_bufsize(ifp);
1823 			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1824 				rc = update_mac_settings(ifp, XGMAC_MTU);
1825 		}
1826 		end_synchronized_op(sc, 0);
1827 		break;
1828 
1829 	case SIOCSIFFLAGS:
1830 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4flg");
1831 		if (rc)
1832 			return (rc);
1833 
1834 		if (ifp->if_flags & IFF_UP) {
1835 			if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1836 				flags = vi->if_flags;
1837 				if ((ifp->if_flags ^ flags) &
1838 				    (IFF_PROMISC | IFF_ALLMULTI)) {
1839 					rc = update_mac_settings(ifp,
1840 					    XGMAC_PROMISC | XGMAC_ALLMULTI);
1841 				}
1842 			} else {
1843 				rc = cxgbe_init_synchronized(vi);
1844 			}
1845 			vi->if_flags = ifp->if_flags;
1846 		} else if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1847 			rc = cxgbe_uninit_synchronized(vi);
1848 		}
1849 		end_synchronized_op(sc, 0);
1850 		break;
1851 
1852 	case SIOCADDMULTI:
1853 	case SIOCDELMULTI:
1854 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4multi");
1855 		if (rc)
1856 			return (rc);
1857 		if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1858 			rc = update_mac_settings(ifp, XGMAC_MCADDRS);
1859 		end_synchronized_op(sc, 0);
1860 		break;
1861 
1862 	case SIOCSIFCAP:
1863 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4cap");
1864 		if (rc)
1865 			return (rc);
1866 
1867 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1868 		if (mask & IFCAP_TXCSUM) {
1869 			ifp->if_capenable ^= IFCAP_TXCSUM;
1870 			ifp->if_hwassist ^= (CSUM_TCP | CSUM_UDP | CSUM_IP);
1871 
1872 			if (IFCAP_TSO4 & ifp->if_capenable &&
1873 			    !(IFCAP_TXCSUM & ifp->if_capenable)) {
1874 				ifp->if_capenable &= ~IFCAP_TSO4;
1875 				if_printf(ifp,
1876 				    "tso4 disabled due to -txcsum.\n");
1877 			}
1878 		}
1879 		if (mask & IFCAP_TXCSUM_IPV6) {
1880 			ifp->if_capenable ^= IFCAP_TXCSUM_IPV6;
1881 			ifp->if_hwassist ^= (CSUM_UDP_IPV6 | CSUM_TCP_IPV6);
1882 
1883 			if (IFCAP_TSO6 & ifp->if_capenable &&
1884 			    !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
1885 				ifp->if_capenable &= ~IFCAP_TSO6;
1886 				if_printf(ifp,
1887 				    "tso6 disabled due to -txcsum6.\n");
1888 			}
1889 		}
1890 		if (mask & IFCAP_RXCSUM)
1891 			ifp->if_capenable ^= IFCAP_RXCSUM;
1892 		if (mask & IFCAP_RXCSUM_IPV6)
1893 			ifp->if_capenable ^= IFCAP_RXCSUM_IPV6;
1894 
1895 		/*
1896 		 * Note that we leave CSUM_TSO alone (it is always set).  The
1897 		 * kernel takes both IFCAP_TSOx and CSUM_TSO into account before
1898 		 * sending a TSO request our way, so it's sufficient to toggle
1899 		 * IFCAP_TSOx only.
1900 		 */
1901 		if (mask & IFCAP_TSO4) {
1902 			if (!(IFCAP_TSO4 & ifp->if_capenable) &&
1903 			    !(IFCAP_TXCSUM & ifp->if_capenable)) {
1904 				if_printf(ifp, "enable txcsum first.\n");
1905 				rc = EAGAIN;
1906 				goto fail;
1907 			}
1908 			ifp->if_capenable ^= IFCAP_TSO4;
1909 		}
1910 		if (mask & IFCAP_TSO6) {
1911 			if (!(IFCAP_TSO6 & ifp->if_capenable) &&
1912 			    !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) {
1913 				if_printf(ifp, "enable txcsum6 first.\n");
1914 				rc = EAGAIN;
1915 				goto fail;
1916 			}
1917 			ifp->if_capenable ^= IFCAP_TSO6;
1918 		}
1919 		if (mask & IFCAP_LRO) {
1920 #if defined(INET) || defined(INET6)
1921 			int i;
1922 			struct sge_rxq *rxq;
1923 
1924 			ifp->if_capenable ^= IFCAP_LRO;
1925 			for_each_rxq(vi, i, rxq) {
1926 				if (ifp->if_capenable & IFCAP_LRO)
1927 					rxq->iq.flags |= IQ_LRO_ENABLED;
1928 				else
1929 					rxq->iq.flags &= ~IQ_LRO_ENABLED;
1930 			}
1931 #endif
1932 		}
1933 #ifdef TCP_OFFLOAD
1934 		if (mask & IFCAP_TOE) {
1935 			int enable = (ifp->if_capenable ^ mask) & IFCAP_TOE;
1936 
1937 			rc = toe_capability(vi, enable);
1938 			if (rc != 0)
1939 				goto fail;
1940 
1941 			ifp->if_capenable ^= mask;
1942 		}
1943 #endif
1944 		if (mask & IFCAP_VLAN_HWTAGGING) {
1945 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
1946 			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1947 				rc = update_mac_settings(ifp, XGMAC_VLANEX);
1948 		}
1949 		if (mask & IFCAP_VLAN_MTU) {
1950 			ifp->if_capenable ^= IFCAP_VLAN_MTU;
1951 
1952 			/* Need to find out how to disable auto-mtu-inflation */
1953 		}
1954 		if (mask & IFCAP_VLAN_HWTSO)
1955 			ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
1956 		if (mask & IFCAP_VLAN_HWCSUM)
1957 			ifp->if_capenable ^= IFCAP_VLAN_HWCSUM;
1958 #ifdef RATELIMIT
1959 		if (mask & IFCAP_TXRTLMT)
1960 			ifp->if_capenable ^= IFCAP_TXRTLMT;
1961 #endif
1962 		if (mask & IFCAP_HWRXTSTMP) {
1963 			int i;
1964 			struct sge_rxq *rxq;
1965 
1966 			ifp->if_capenable ^= IFCAP_HWRXTSTMP;
1967 			for_each_rxq(vi, i, rxq) {
1968 				if (ifp->if_capenable & IFCAP_HWRXTSTMP)
1969 					rxq->iq.flags |= IQ_RX_TIMESTAMP;
1970 				else
1971 					rxq->iq.flags &= ~IQ_RX_TIMESTAMP;
1972 			}
1973 		}
1974 
1975 #ifdef VLAN_CAPABILITIES
1976 		VLAN_CAPABILITIES(ifp);
1977 #endif
1978 fail:
1979 		end_synchronized_op(sc, 0);
1980 		break;
1981 
1982 	case SIOCSIFMEDIA:
1983 	case SIOCGIFMEDIA:
1984 	case SIOCGIFXMEDIA:
1985 		ifmedia_ioctl(ifp, ifr, &pi->media, cmd);
1986 		break;
1987 
1988 	case SIOCGI2C: {
1989 		struct ifi2creq i2c;
1990 
1991 		rc = copyin(ifr_data_get_ptr(ifr), &i2c, sizeof(i2c));
1992 		if (rc != 0)
1993 			break;
1994 		if (i2c.dev_addr != 0xA0 && i2c.dev_addr != 0xA2) {
1995 			rc = EPERM;
1996 			break;
1997 		}
1998 		if (i2c.len > sizeof(i2c.data)) {
1999 			rc = EINVAL;
2000 			break;
2001 		}
2002 		rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4i2c");
2003 		if (rc)
2004 			return (rc);
2005 		rc = -t4_i2c_rd(sc, sc->mbox, pi->port_id, i2c.dev_addr,
2006 		    i2c.offset, i2c.len, &i2c.data[0]);
2007 		end_synchronized_op(sc, 0);
2008 		if (rc == 0)
2009 			rc = copyout(&i2c, ifr_data_get_ptr(ifr), sizeof(i2c));
2010 		break;
2011 	}
2012 
2013 	default:
2014 		rc = ether_ioctl(ifp, cmd, data);
2015 	}
2016 
2017 	return (rc);
2018 }
2019 
2020 static int
2021 cxgbe_transmit(struct ifnet *ifp, struct mbuf *m)
2022 {
2023 	struct vi_info *vi = ifp->if_softc;
2024 	struct port_info *pi = vi->pi;
2025 	struct adapter *sc = pi->adapter;
2026 	struct sge_txq *txq;
2027 	void *items[1];
2028 	int rc;
2029 
2030 	M_ASSERTPKTHDR(m);
2031 	MPASS(m->m_nextpkt == NULL);	/* not quite ready for this yet */
2032 
2033 	if (__predict_false(pi->link_cfg.link_ok == false)) {
2034 		m_freem(m);
2035 		return (ENETDOWN);
2036 	}
2037 
2038 	rc = parse_pkt(sc, &m);
2039 	if (__predict_false(rc != 0)) {
2040 		MPASS(m == NULL);			/* was freed already */
2041 		atomic_add_int(&pi->tx_parse_error, 1);	/* rare, atomic is ok */
2042 		return (rc);
2043 	}
2044 #ifdef RATELIMIT
2045 	if (m->m_pkthdr.snd_tag != NULL) {
2046 		/* EAGAIN tells the stack we are not the correct interface. */
2047 		if (__predict_false(ifp != m->m_pkthdr.snd_tag->ifp)) {
2048 			m_freem(m);
2049 			return (EAGAIN);
2050 		}
2051 
2052 		return (ethofld_transmit(ifp, m));
2053 	}
2054 #endif
2055 
2056 	/* Select a txq. */
2057 	txq = &sc->sge.txq[vi->first_txq];
2058 	if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
2059 		txq += ((m->m_pkthdr.flowid % (vi->ntxq - vi->rsrv_noflowq)) +
2060 		    vi->rsrv_noflowq);
2061 
2062 	items[0] = m;
2063 	rc = mp_ring_enqueue(txq->r, items, 1, 4096);
2064 	if (__predict_false(rc != 0))
2065 		m_freem(m);
2066 
2067 	return (rc);
2068 }
2069 
2070 static void
2071 cxgbe_qflush(struct ifnet *ifp)
2072 {
2073 	struct vi_info *vi = ifp->if_softc;
2074 	struct sge_txq *txq;
2075 	int i;
2076 
2077 	/* queues do not exist if !VI_INIT_DONE. */
2078 	if (vi->flags & VI_INIT_DONE) {
2079 		for_each_txq(vi, i, txq) {
2080 			TXQ_LOCK(txq);
2081 			txq->eq.flags |= EQ_QFLUSH;
2082 			TXQ_UNLOCK(txq);
2083 			while (!mp_ring_is_idle(txq->r)) {
2084 				mp_ring_check_drainage(txq->r, 0);
2085 				pause("qflush", 1);
2086 			}
2087 			TXQ_LOCK(txq);
2088 			txq->eq.flags &= ~EQ_QFLUSH;
2089 			TXQ_UNLOCK(txq);
2090 		}
2091 	}
2092 	if_qflush(ifp);
2093 }
2094 
2095 static uint64_t
2096 vi_get_counter(struct ifnet *ifp, ift_counter c)
2097 {
2098 	struct vi_info *vi = ifp->if_softc;
2099 	struct fw_vi_stats_vf *s = &vi->stats;
2100 
2101 	vi_refresh_stats(vi->pi->adapter, vi);
2102 
2103 	switch (c) {
2104 	case IFCOUNTER_IPACKETS:
2105 		return (s->rx_bcast_frames + s->rx_mcast_frames +
2106 		    s->rx_ucast_frames);
2107 	case IFCOUNTER_IERRORS:
2108 		return (s->rx_err_frames);
2109 	case IFCOUNTER_OPACKETS:
2110 		return (s->tx_bcast_frames + s->tx_mcast_frames +
2111 		    s->tx_ucast_frames + s->tx_offload_frames);
2112 	case IFCOUNTER_OERRORS:
2113 		return (s->tx_drop_frames);
2114 	case IFCOUNTER_IBYTES:
2115 		return (s->rx_bcast_bytes + s->rx_mcast_bytes +
2116 		    s->rx_ucast_bytes);
2117 	case IFCOUNTER_OBYTES:
2118 		return (s->tx_bcast_bytes + s->tx_mcast_bytes +
2119 		    s->tx_ucast_bytes + s->tx_offload_bytes);
2120 	case IFCOUNTER_IMCASTS:
2121 		return (s->rx_mcast_frames);
2122 	case IFCOUNTER_OMCASTS:
2123 		return (s->tx_mcast_frames);
2124 	case IFCOUNTER_OQDROPS: {
2125 		uint64_t drops;
2126 
2127 		drops = 0;
2128 		if (vi->flags & VI_INIT_DONE) {
2129 			int i;
2130 			struct sge_txq *txq;
2131 
2132 			for_each_txq(vi, i, txq)
2133 				drops += counter_u64_fetch(txq->r->drops);
2134 		}
2135 
2136 		return (drops);
2137 
2138 	}
2139 
2140 	default:
2141 		return (if_get_counter_default(ifp, c));
2142 	}
2143 }
2144 
2145 uint64_t
2146 cxgbe_get_counter(struct ifnet *ifp, ift_counter c)
2147 {
2148 	struct vi_info *vi = ifp->if_softc;
2149 	struct port_info *pi = vi->pi;
2150 	struct adapter *sc = pi->adapter;
2151 	struct port_stats *s = &pi->stats;
2152 
2153 	if (pi->nvi > 1 || sc->flags & IS_VF)
2154 		return (vi_get_counter(ifp, c));
2155 
2156 	cxgbe_refresh_stats(sc, pi);
2157 
2158 	switch (c) {
2159 	case IFCOUNTER_IPACKETS:
2160 		return (s->rx_frames);
2161 
2162 	case IFCOUNTER_IERRORS:
2163 		return (s->rx_jabber + s->rx_runt + s->rx_too_long +
2164 		    s->rx_fcs_err + s->rx_len_err);
2165 
2166 	case IFCOUNTER_OPACKETS:
2167 		return (s->tx_frames);
2168 
2169 	case IFCOUNTER_OERRORS:
2170 		return (s->tx_error_frames);
2171 
2172 	case IFCOUNTER_IBYTES:
2173 		return (s->rx_octets);
2174 
2175 	case IFCOUNTER_OBYTES:
2176 		return (s->tx_octets);
2177 
2178 	case IFCOUNTER_IMCASTS:
2179 		return (s->rx_mcast_frames);
2180 
2181 	case IFCOUNTER_OMCASTS:
2182 		return (s->tx_mcast_frames);
2183 
2184 	case IFCOUNTER_IQDROPS:
2185 		return (s->rx_ovflow0 + s->rx_ovflow1 + s->rx_ovflow2 +
2186 		    s->rx_ovflow3 + s->rx_trunc0 + s->rx_trunc1 + s->rx_trunc2 +
2187 		    s->rx_trunc3 + pi->tnl_cong_drops);
2188 
2189 	case IFCOUNTER_OQDROPS: {
2190 		uint64_t drops;
2191 
2192 		drops = s->tx_drop;
2193 		if (vi->flags & VI_INIT_DONE) {
2194 			int i;
2195 			struct sge_txq *txq;
2196 
2197 			for_each_txq(vi, i, txq)
2198 				drops += counter_u64_fetch(txq->r->drops);
2199 		}
2200 
2201 		return (drops);
2202 
2203 	}
2204 
2205 	default:
2206 		return (if_get_counter_default(ifp, c));
2207 	}
2208 }
2209 
2210 /*
2211  * The kernel picks a media from the list we had provided but we still validate
2212  * the requeste.
2213  */
2214 int
2215 cxgbe_media_change(struct ifnet *ifp)
2216 {
2217 	struct vi_info *vi = ifp->if_softc;
2218 	struct port_info *pi = vi->pi;
2219 	struct ifmedia *ifm = &pi->media;
2220 	struct link_config *lc = &pi->link_cfg;
2221 	struct adapter *sc = pi->adapter;
2222 	int rc;
2223 
2224 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4mec");
2225 	if (rc != 0)
2226 		return (rc);
2227 	PORT_LOCK(pi);
2228 	if (IFM_SUBTYPE(ifm->ifm_media) == IFM_AUTO) {
2229 		/* ifconfig .. media autoselect */
2230 		if (!(lc->supported & FW_PORT_CAP32_ANEG)) {
2231 			rc = ENOTSUP; /* AN not supported by transceiver */
2232 			goto done;
2233 		}
2234 		lc->requested_aneg = AUTONEG_ENABLE;
2235 		lc->requested_speed = 0;
2236 		lc->requested_fc |= PAUSE_AUTONEG;
2237 	} else {
2238 		lc->requested_aneg = AUTONEG_DISABLE;
2239 		lc->requested_speed =
2240 		    ifmedia_baudrate(ifm->ifm_media) / 1000000;
2241 		lc->requested_fc = 0;
2242 		if (IFM_OPTIONS(ifm->ifm_media) & IFM_ETH_RXPAUSE)
2243 			lc->requested_fc |= PAUSE_RX;
2244 		if (IFM_OPTIONS(ifm->ifm_media) & IFM_ETH_TXPAUSE)
2245 			lc->requested_fc |= PAUSE_TX;
2246 	}
2247 	if (pi->up_vis > 0) {
2248 		fixup_link_config(pi);
2249 		rc = apply_link_config(pi);
2250 	}
2251 done:
2252 	PORT_UNLOCK(pi);
2253 	end_synchronized_op(sc, 0);
2254 	return (rc);
2255 }
2256 
2257 /*
2258  * Base media word (without ETHER, pause, link active, etc.) for the port at the
2259  * given speed.
2260  */
2261 static int
2262 port_mword(struct port_info *pi, uint32_t speed)
2263 {
2264 
2265 	MPASS(speed & M_FW_PORT_CAP32_SPEED);
2266 	MPASS(powerof2(speed));
2267 
2268 	switch(pi->port_type) {
2269 	case FW_PORT_TYPE_BT_SGMII:
2270 	case FW_PORT_TYPE_BT_XFI:
2271 	case FW_PORT_TYPE_BT_XAUI:
2272 		/* BaseT */
2273 		switch (speed) {
2274 		case FW_PORT_CAP32_SPEED_100M:
2275 			return (IFM_100_T);
2276 		case FW_PORT_CAP32_SPEED_1G:
2277 			return (IFM_1000_T);
2278 		case FW_PORT_CAP32_SPEED_10G:
2279 			return (IFM_10G_T);
2280 		}
2281 		break;
2282 	case FW_PORT_TYPE_KX4:
2283 		if (speed == FW_PORT_CAP32_SPEED_10G)
2284 			return (IFM_10G_KX4);
2285 		break;
2286 	case FW_PORT_TYPE_CX4:
2287 		if (speed == FW_PORT_CAP32_SPEED_10G)
2288 			return (IFM_10G_CX4);
2289 		break;
2290 	case FW_PORT_TYPE_KX:
2291 		if (speed == FW_PORT_CAP32_SPEED_1G)
2292 			return (IFM_1000_KX);
2293 		break;
2294 	case FW_PORT_TYPE_KR:
2295 	case FW_PORT_TYPE_BP_AP:
2296 	case FW_PORT_TYPE_BP4_AP:
2297 	case FW_PORT_TYPE_BP40_BA:
2298 	case FW_PORT_TYPE_KR4_100G:
2299 	case FW_PORT_TYPE_KR_SFP28:
2300 	case FW_PORT_TYPE_KR_XLAUI:
2301 		switch (speed) {
2302 		case FW_PORT_CAP32_SPEED_1G:
2303 			return (IFM_1000_KX);
2304 		case FW_PORT_CAP32_SPEED_10G:
2305 			return (IFM_10G_KR);
2306 		case FW_PORT_CAP32_SPEED_25G:
2307 			return (IFM_25G_KR);
2308 		case FW_PORT_CAP32_SPEED_40G:
2309 			return (IFM_40G_KR4);
2310 		case FW_PORT_CAP32_SPEED_50G:
2311 			return (IFM_50G_KR2);
2312 		case FW_PORT_CAP32_SPEED_100G:
2313 			return (IFM_100G_KR4);
2314 		}
2315 		break;
2316 	case FW_PORT_TYPE_FIBER_XFI:
2317 	case FW_PORT_TYPE_FIBER_XAUI:
2318 	case FW_PORT_TYPE_SFP:
2319 	case FW_PORT_TYPE_QSFP_10G:
2320 	case FW_PORT_TYPE_QSA:
2321 	case FW_PORT_TYPE_QSFP:
2322 	case FW_PORT_TYPE_CR4_QSFP:
2323 	case FW_PORT_TYPE_CR_QSFP:
2324 	case FW_PORT_TYPE_CR2_QSFP:
2325 	case FW_PORT_TYPE_SFP28:
2326 		/* Pluggable transceiver */
2327 		switch (pi->mod_type) {
2328 		case FW_PORT_MOD_TYPE_LR:
2329 			switch (speed) {
2330 			case FW_PORT_CAP32_SPEED_1G:
2331 				return (IFM_1000_LX);
2332 			case FW_PORT_CAP32_SPEED_10G:
2333 				return (IFM_10G_LR);
2334 			case FW_PORT_CAP32_SPEED_25G:
2335 				return (IFM_25G_LR);
2336 			case FW_PORT_CAP32_SPEED_40G:
2337 				return (IFM_40G_LR4);
2338 			case FW_PORT_CAP32_SPEED_50G:
2339 				return (IFM_50G_LR2);
2340 			case FW_PORT_CAP32_SPEED_100G:
2341 				return (IFM_100G_LR4);
2342 			}
2343 			break;
2344 		case FW_PORT_MOD_TYPE_SR:
2345 			switch (speed) {
2346 			case FW_PORT_CAP32_SPEED_1G:
2347 				return (IFM_1000_SX);
2348 			case FW_PORT_CAP32_SPEED_10G:
2349 				return (IFM_10G_SR);
2350 			case FW_PORT_CAP32_SPEED_25G:
2351 				return (IFM_25G_SR);
2352 			case FW_PORT_CAP32_SPEED_40G:
2353 				return (IFM_40G_SR4);
2354 			case FW_PORT_CAP32_SPEED_50G:
2355 				return (IFM_50G_SR2);
2356 			case FW_PORT_CAP32_SPEED_100G:
2357 				return (IFM_100G_SR4);
2358 			}
2359 			break;
2360 		case FW_PORT_MOD_TYPE_ER:
2361 			if (speed == FW_PORT_CAP32_SPEED_10G)
2362 				return (IFM_10G_ER);
2363 			break;
2364 		case FW_PORT_MOD_TYPE_TWINAX_PASSIVE:
2365 		case FW_PORT_MOD_TYPE_TWINAX_ACTIVE:
2366 			switch (speed) {
2367 			case FW_PORT_CAP32_SPEED_1G:
2368 				return (IFM_1000_CX);
2369 			case FW_PORT_CAP32_SPEED_10G:
2370 				return (IFM_10G_TWINAX);
2371 			case FW_PORT_CAP32_SPEED_25G:
2372 				return (IFM_25G_CR);
2373 			case FW_PORT_CAP32_SPEED_40G:
2374 				return (IFM_40G_CR4);
2375 			case FW_PORT_CAP32_SPEED_50G:
2376 				return (IFM_50G_CR2);
2377 			case FW_PORT_CAP32_SPEED_100G:
2378 				return (IFM_100G_CR4);
2379 			}
2380 			break;
2381 		case FW_PORT_MOD_TYPE_LRM:
2382 			if (speed == FW_PORT_CAP32_SPEED_10G)
2383 				return (IFM_10G_LRM);
2384 			break;
2385 		case FW_PORT_MOD_TYPE_NA:
2386 			MPASS(0);	/* Not pluggable? */
2387 			/* fall throough */
2388 		case FW_PORT_MOD_TYPE_ERROR:
2389 		case FW_PORT_MOD_TYPE_UNKNOWN:
2390 		case FW_PORT_MOD_TYPE_NOTSUPPORTED:
2391 			break;
2392 		case FW_PORT_MOD_TYPE_NONE:
2393 			return (IFM_NONE);
2394 		}
2395 		break;
2396 	case FW_PORT_TYPE_NONE:
2397 		return (IFM_NONE);
2398 	}
2399 
2400 	return (IFM_UNKNOWN);
2401 }
2402 
2403 void
2404 cxgbe_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
2405 {
2406 	struct vi_info *vi = ifp->if_softc;
2407 	struct port_info *pi = vi->pi;
2408 	struct adapter *sc = pi->adapter;
2409 	struct link_config *lc = &pi->link_cfg;
2410 
2411 	if (begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4med") != 0)
2412 		return;
2413 	PORT_LOCK(pi);
2414 
2415 	if (pi->up_vis == 0) {
2416 		/*
2417 		 * If all the interfaces are administratively down the firmware
2418 		 * does not report transceiver changes.  Refresh port info here
2419 		 * so that ifconfig displays accurate ifmedia at all times.
2420 		 * This is the only reason we have a synchronized op in this
2421 		 * function.  Just PORT_LOCK would have been enough otherwise.
2422 		 */
2423 		t4_update_port_info(pi);
2424 		build_medialist(pi);
2425 	}
2426 
2427 	/* ifm_status */
2428 	ifmr->ifm_status = IFM_AVALID;
2429 	if (lc->link_ok == false)
2430 		goto done;
2431 	ifmr->ifm_status |= IFM_ACTIVE;
2432 
2433 	/* ifm_active */
2434 	ifmr->ifm_active = IFM_ETHER | IFM_FDX;
2435 	ifmr->ifm_active &= ~(IFM_ETH_TXPAUSE | IFM_ETH_RXPAUSE);
2436 	if (lc->fc & PAUSE_RX)
2437 		ifmr->ifm_active |= IFM_ETH_RXPAUSE;
2438 	if (lc->fc & PAUSE_TX)
2439 		ifmr->ifm_active |= IFM_ETH_TXPAUSE;
2440 	ifmr->ifm_active |= port_mword(pi, speed_to_fwcap(lc->speed));
2441 done:
2442 	PORT_UNLOCK(pi);
2443 	end_synchronized_op(sc, 0);
2444 }
2445 
2446 static int
2447 vcxgbe_probe(device_t dev)
2448 {
2449 	char buf[128];
2450 	struct vi_info *vi = device_get_softc(dev);
2451 
2452 	snprintf(buf, sizeof(buf), "port %d vi %td", vi->pi->port_id,
2453 	    vi - vi->pi->vi);
2454 	device_set_desc_copy(dev, buf);
2455 
2456 	return (BUS_PROBE_DEFAULT);
2457 }
2458 
2459 static int
2460 alloc_extra_vi(struct adapter *sc, struct port_info *pi, struct vi_info *vi)
2461 {
2462 	int func, index, rc;
2463 	uint32_t param, val;
2464 
2465 	ASSERT_SYNCHRONIZED_OP(sc);
2466 
2467 	index = vi - pi->vi;
2468 	MPASS(index > 0);	/* This function deals with _extra_ VIs only */
2469 	KASSERT(index < nitems(vi_mac_funcs),
2470 	    ("%s: VI %s doesn't have a MAC func", __func__,
2471 	    device_get_nameunit(vi->dev)));
2472 	func = vi_mac_funcs[index];
2473 	rc = t4_alloc_vi_func(sc, sc->mbox, pi->tx_chan, sc->pf, 0, 1,
2474 	    vi->hw_addr, &vi->rss_size, func, 0);
2475 	if (rc < 0) {
2476 		device_printf(vi->dev, "failed to allocate virtual interface %d"
2477 		    "for port %d: %d\n", index, pi->port_id, -rc);
2478 		return (-rc);
2479 	}
2480 	vi->viid = rc;
2481 	if (chip_id(sc) <= CHELSIO_T5)
2482 		vi->smt_idx = (rc & 0x7f) << 1;
2483 	else
2484 		vi->smt_idx = (rc & 0x7f);
2485 
2486 	if (vi->rss_size == 1) {
2487 		/*
2488 		 * This VI didn't get a slice of the RSS table.  Reduce the
2489 		 * number of VIs being created (hw.cxgbe.num_vis) or modify the
2490 		 * configuration file (nvi, rssnvi for this PF) if this is a
2491 		 * problem.
2492 		 */
2493 		device_printf(vi->dev, "RSS table not available.\n");
2494 		vi->rss_base = 0xffff;
2495 
2496 		return (0);
2497 	}
2498 
2499 	param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
2500 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_RSSINFO) |
2501 	    V_FW_PARAMS_PARAM_YZ(vi->viid);
2502 	rc = t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
2503 	if (rc)
2504 		vi->rss_base = 0xffff;
2505 	else {
2506 		MPASS((val >> 16) == vi->rss_size);
2507 		vi->rss_base = val & 0xffff;
2508 	}
2509 
2510 	return (0);
2511 }
2512 
2513 static int
2514 vcxgbe_attach(device_t dev)
2515 {
2516 	struct vi_info *vi;
2517 	struct port_info *pi;
2518 	struct adapter *sc;
2519 	int rc;
2520 
2521 	vi = device_get_softc(dev);
2522 	pi = vi->pi;
2523 	sc = pi->adapter;
2524 
2525 	rc = begin_synchronized_op(sc, vi, SLEEP_OK | INTR_OK, "t4via");
2526 	if (rc)
2527 		return (rc);
2528 	rc = alloc_extra_vi(sc, pi, vi);
2529 	end_synchronized_op(sc, 0);
2530 	if (rc)
2531 		return (rc);
2532 
2533 	rc = cxgbe_vi_attach(dev, vi);
2534 	if (rc) {
2535 		t4_free_vi(sc, sc->mbox, sc->pf, 0, vi->viid);
2536 		return (rc);
2537 	}
2538 	return (0);
2539 }
2540 
2541 static int
2542 vcxgbe_detach(device_t dev)
2543 {
2544 	struct vi_info *vi;
2545 	struct adapter *sc;
2546 
2547 	vi = device_get_softc(dev);
2548 	sc = vi->pi->adapter;
2549 
2550 	doom_vi(sc, vi);
2551 
2552 	cxgbe_vi_detach(vi);
2553 	t4_free_vi(sc, sc->mbox, sc->pf, 0, vi->viid);
2554 
2555 	end_synchronized_op(sc, 0);
2556 
2557 	return (0);
2558 }
2559 
2560 void
2561 t4_fatal_err(struct adapter *sc)
2562 {
2563 	t4_set_reg_field(sc, A_SGE_CONTROL, F_GLOBALENABLE, 0);
2564 	t4_intr_disable(sc);
2565 	log(LOG_EMERG, "%s: encountered fatal error, adapter stopped.\n",
2566 	    device_get_nameunit(sc->dev));
2567 	if (t4_panic_on_fatal_err)
2568 		panic("panic requested on fatal error");
2569 }
2570 
2571 void
2572 t4_add_adapter(struct adapter *sc)
2573 {
2574 	sx_xlock(&t4_list_lock);
2575 	SLIST_INSERT_HEAD(&t4_list, sc, link);
2576 	sx_xunlock(&t4_list_lock);
2577 }
2578 
2579 int
2580 t4_map_bars_0_and_4(struct adapter *sc)
2581 {
2582 	sc->regs_rid = PCIR_BAR(0);
2583 	sc->regs_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2584 	    &sc->regs_rid, RF_ACTIVE);
2585 	if (sc->regs_res == NULL) {
2586 		device_printf(sc->dev, "cannot map registers.\n");
2587 		return (ENXIO);
2588 	}
2589 	sc->bt = rman_get_bustag(sc->regs_res);
2590 	sc->bh = rman_get_bushandle(sc->regs_res);
2591 	sc->mmio_len = rman_get_size(sc->regs_res);
2592 	setbit(&sc->doorbells, DOORBELL_KDB);
2593 
2594 	sc->msix_rid = PCIR_BAR(4);
2595 	sc->msix_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2596 	    &sc->msix_rid, RF_ACTIVE);
2597 	if (sc->msix_res == NULL) {
2598 		device_printf(sc->dev, "cannot map MSI-X BAR.\n");
2599 		return (ENXIO);
2600 	}
2601 
2602 	return (0);
2603 }
2604 
2605 int
2606 t4_map_bar_2(struct adapter *sc)
2607 {
2608 
2609 	/*
2610 	 * T4: only iWARP driver uses the userspace doorbells.  There is no need
2611 	 * to map it if RDMA is disabled.
2612 	 */
2613 	if (is_t4(sc) && sc->rdmacaps == 0)
2614 		return (0);
2615 
2616 	sc->udbs_rid = PCIR_BAR(2);
2617 	sc->udbs_res = bus_alloc_resource_any(sc->dev, SYS_RES_MEMORY,
2618 	    &sc->udbs_rid, RF_ACTIVE);
2619 	if (sc->udbs_res == NULL) {
2620 		device_printf(sc->dev, "cannot map doorbell BAR.\n");
2621 		return (ENXIO);
2622 	}
2623 	sc->udbs_base = rman_get_virtual(sc->udbs_res);
2624 
2625 	if (chip_id(sc) >= CHELSIO_T5) {
2626 		setbit(&sc->doorbells, DOORBELL_UDB);
2627 #if defined(__i386__) || defined(__amd64__)
2628 		if (t5_write_combine) {
2629 			int rc, mode;
2630 
2631 			/*
2632 			 * Enable write combining on BAR2.  This is the
2633 			 * userspace doorbell BAR and is split into 128B
2634 			 * (UDBS_SEG_SIZE) doorbell regions, each associated
2635 			 * with an egress queue.  The first 64B has the doorbell
2636 			 * and the second 64B can be used to submit a tx work
2637 			 * request with an implicit doorbell.
2638 			 */
2639 
2640 			rc = pmap_change_attr((vm_offset_t)sc->udbs_base,
2641 			    rman_get_size(sc->udbs_res), PAT_WRITE_COMBINING);
2642 			if (rc == 0) {
2643 				clrbit(&sc->doorbells, DOORBELL_UDB);
2644 				setbit(&sc->doorbells, DOORBELL_WCWR);
2645 				setbit(&sc->doorbells, DOORBELL_UDBWC);
2646 			} else {
2647 				device_printf(sc->dev,
2648 				    "couldn't enable write combining: %d\n",
2649 				    rc);
2650 			}
2651 
2652 			mode = is_t5(sc) ? V_STATMODE(0) : V_T6_STATMODE(0);
2653 			t4_write_reg(sc, A_SGE_STAT_CFG,
2654 			    V_STATSOURCE_T5(7) | mode);
2655 		}
2656 #endif
2657 	}
2658 	sc->iwt.wc_en = isset(&sc->doorbells, DOORBELL_UDBWC) ? 1 : 0;
2659 
2660 	return (0);
2661 }
2662 
2663 struct memwin_init {
2664 	uint32_t base;
2665 	uint32_t aperture;
2666 };
2667 
2668 static const struct memwin_init t4_memwin[NUM_MEMWIN] = {
2669 	{ MEMWIN0_BASE, MEMWIN0_APERTURE },
2670 	{ MEMWIN1_BASE, MEMWIN1_APERTURE },
2671 	{ MEMWIN2_BASE_T4, MEMWIN2_APERTURE_T4 }
2672 };
2673 
2674 static const struct memwin_init t5_memwin[NUM_MEMWIN] = {
2675 	{ MEMWIN0_BASE, MEMWIN0_APERTURE },
2676 	{ MEMWIN1_BASE, MEMWIN1_APERTURE },
2677 	{ MEMWIN2_BASE_T5, MEMWIN2_APERTURE_T5 },
2678 };
2679 
2680 static void
2681 setup_memwin(struct adapter *sc)
2682 {
2683 	const struct memwin_init *mw_init;
2684 	struct memwin *mw;
2685 	int i;
2686 	uint32_t bar0;
2687 
2688 	if (is_t4(sc)) {
2689 		/*
2690 		 * Read low 32b of bar0 indirectly via the hardware backdoor
2691 		 * mechanism.  Works from within PCI passthrough environments
2692 		 * too, where rman_get_start() can return a different value.  We
2693 		 * need to program the T4 memory window decoders with the actual
2694 		 * addresses that will be coming across the PCIe link.
2695 		 */
2696 		bar0 = t4_hw_pci_read_cfg4(sc, PCIR_BAR(0));
2697 		bar0 &= (uint32_t) PCIM_BAR_MEM_BASE;
2698 
2699 		mw_init = &t4_memwin[0];
2700 	} else {
2701 		/* T5+ use the relative offset inside the PCIe BAR */
2702 		bar0 = 0;
2703 
2704 		mw_init = &t5_memwin[0];
2705 	}
2706 
2707 	for (i = 0, mw = &sc->memwin[0]; i < NUM_MEMWIN; i++, mw_init++, mw++) {
2708 		rw_init(&mw->mw_lock, "memory window access");
2709 		mw->mw_base = mw_init->base;
2710 		mw->mw_aperture = mw_init->aperture;
2711 		mw->mw_curpos = 0;
2712 		t4_write_reg(sc,
2713 		    PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, i),
2714 		    (mw->mw_base + bar0) | V_BIR(0) |
2715 		    V_WINDOW(ilog2(mw->mw_aperture) - 10));
2716 		rw_wlock(&mw->mw_lock);
2717 		position_memwin(sc, i, 0);
2718 		rw_wunlock(&mw->mw_lock);
2719 	}
2720 
2721 	/* flush */
2722 	t4_read_reg(sc, PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_BASE_WIN, 2));
2723 }
2724 
2725 /*
2726  * Positions the memory window at the given address in the card's address space.
2727  * There are some alignment requirements and the actual position may be at an
2728  * address prior to the requested address.  mw->mw_curpos always has the actual
2729  * position of the window.
2730  */
2731 static void
2732 position_memwin(struct adapter *sc, int idx, uint32_t addr)
2733 {
2734 	struct memwin *mw;
2735 	uint32_t pf;
2736 	uint32_t reg;
2737 
2738 	MPASS(idx >= 0 && idx < NUM_MEMWIN);
2739 	mw = &sc->memwin[idx];
2740 	rw_assert(&mw->mw_lock, RA_WLOCKED);
2741 
2742 	if (is_t4(sc)) {
2743 		pf = 0;
2744 		mw->mw_curpos = addr & ~0xf;	/* start must be 16B aligned */
2745 	} else {
2746 		pf = V_PFNUM(sc->pf);
2747 		mw->mw_curpos = addr & ~0x7f;	/* start must be 128B aligned */
2748 	}
2749 	reg = PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_OFFSET, idx);
2750 	t4_write_reg(sc, reg, mw->mw_curpos | pf);
2751 	t4_read_reg(sc, reg);	/* flush */
2752 }
2753 
2754 int
2755 rw_via_memwin(struct adapter *sc, int idx, uint32_t addr, uint32_t *val,
2756     int len, int rw)
2757 {
2758 	struct memwin *mw;
2759 	uint32_t mw_end, v;
2760 
2761 	MPASS(idx >= 0 && idx < NUM_MEMWIN);
2762 
2763 	/* Memory can only be accessed in naturally aligned 4 byte units */
2764 	if (addr & 3 || len & 3 || len <= 0)
2765 		return (EINVAL);
2766 
2767 	mw = &sc->memwin[idx];
2768 	while (len > 0) {
2769 		rw_rlock(&mw->mw_lock);
2770 		mw_end = mw->mw_curpos + mw->mw_aperture;
2771 		if (addr >= mw_end || addr < mw->mw_curpos) {
2772 			/* Will need to reposition the window */
2773 			if (!rw_try_upgrade(&mw->mw_lock)) {
2774 				rw_runlock(&mw->mw_lock);
2775 				rw_wlock(&mw->mw_lock);
2776 			}
2777 			rw_assert(&mw->mw_lock, RA_WLOCKED);
2778 			position_memwin(sc, idx, addr);
2779 			rw_downgrade(&mw->mw_lock);
2780 			mw_end = mw->mw_curpos + mw->mw_aperture;
2781 		}
2782 		rw_assert(&mw->mw_lock, RA_RLOCKED);
2783 		while (addr < mw_end && len > 0) {
2784 			if (rw == 0) {
2785 				v = t4_read_reg(sc, mw->mw_base + addr -
2786 				    mw->mw_curpos);
2787 				*val++ = le32toh(v);
2788 			} else {
2789 				v = *val++;
2790 				t4_write_reg(sc, mw->mw_base + addr -
2791 				    mw->mw_curpos, htole32(v));
2792 			}
2793 			addr += 4;
2794 			len -= 4;
2795 		}
2796 		rw_runlock(&mw->mw_lock);
2797 	}
2798 
2799 	return (0);
2800 }
2801 
2802 int
2803 alloc_atid_tab(struct tid_info *t, int flags)
2804 {
2805 	int i;
2806 
2807 	MPASS(t->natids > 0);
2808 	MPASS(t->atid_tab == NULL);
2809 
2810 	t->atid_tab = malloc(t->natids * sizeof(*t->atid_tab), M_CXGBE,
2811 	    M_ZERO | flags);
2812 	if (t->atid_tab == NULL)
2813 		return (ENOMEM);
2814 	mtx_init(&t->atid_lock, "atid lock", NULL, MTX_DEF);
2815 	t->afree = t->atid_tab;
2816 	t->atids_in_use = 0;
2817 	for (i = 1; i < t->natids; i++)
2818 		t->atid_tab[i - 1].next = &t->atid_tab[i];
2819 	t->atid_tab[t->natids - 1].next = NULL;
2820 
2821 	return (0);
2822 }
2823 
2824 void
2825 free_atid_tab(struct tid_info *t)
2826 {
2827 
2828 	KASSERT(t->atids_in_use == 0,
2829 	    ("%s: %d atids still in use.", __func__, t->atids_in_use));
2830 
2831 	if (mtx_initialized(&t->atid_lock))
2832 		mtx_destroy(&t->atid_lock);
2833 	free(t->atid_tab, M_CXGBE);
2834 	t->atid_tab = NULL;
2835 }
2836 
2837 int
2838 alloc_atid(struct adapter *sc, void *ctx)
2839 {
2840 	struct tid_info *t = &sc->tids;
2841 	int atid = -1;
2842 
2843 	mtx_lock(&t->atid_lock);
2844 	if (t->afree) {
2845 		union aopen_entry *p = t->afree;
2846 
2847 		atid = p - t->atid_tab;
2848 		MPASS(atid <= M_TID_TID);
2849 		t->afree = p->next;
2850 		p->data = ctx;
2851 		t->atids_in_use++;
2852 	}
2853 	mtx_unlock(&t->atid_lock);
2854 	return (atid);
2855 }
2856 
2857 void *
2858 lookup_atid(struct adapter *sc, int atid)
2859 {
2860 	struct tid_info *t = &sc->tids;
2861 
2862 	return (t->atid_tab[atid].data);
2863 }
2864 
2865 void
2866 free_atid(struct adapter *sc, int atid)
2867 {
2868 	struct tid_info *t = &sc->tids;
2869 	union aopen_entry *p = &t->atid_tab[atid];
2870 
2871 	mtx_lock(&t->atid_lock);
2872 	p->next = t->afree;
2873 	t->afree = p;
2874 	t->atids_in_use--;
2875 	mtx_unlock(&t->atid_lock);
2876 }
2877 
2878 static void
2879 queue_tid_release(struct adapter *sc, int tid)
2880 {
2881 
2882 	CXGBE_UNIMPLEMENTED("deferred tid release");
2883 }
2884 
2885 void
2886 release_tid(struct adapter *sc, int tid, struct sge_wrq *ctrlq)
2887 {
2888 	struct wrqe *wr;
2889 	struct cpl_tid_release *req;
2890 
2891 	wr = alloc_wrqe(sizeof(*req), ctrlq);
2892 	if (wr == NULL) {
2893 		queue_tid_release(sc, tid);	/* defer */
2894 		return;
2895 	}
2896 	req = wrtod(wr);
2897 
2898 	INIT_TP_WR_MIT_CPL(req, CPL_TID_RELEASE, tid);
2899 
2900 	t4_wrq_tx(sc, wr);
2901 }
2902 
2903 static int
2904 t4_range_cmp(const void *a, const void *b)
2905 {
2906 	return ((const struct t4_range *)a)->start -
2907 	       ((const struct t4_range *)b)->start;
2908 }
2909 
2910 /*
2911  * Verify that the memory range specified by the addr/len pair is valid within
2912  * the card's address space.
2913  */
2914 static int
2915 validate_mem_range(struct adapter *sc, uint32_t addr, uint32_t len)
2916 {
2917 	struct t4_range mem_ranges[4], *r, *next;
2918 	uint32_t em, addr_len;
2919 	int i, n, remaining;
2920 
2921 	/* Memory can only be accessed in naturally aligned 4 byte units */
2922 	if (addr & 3 || len & 3 || len == 0)
2923 		return (EINVAL);
2924 
2925 	/* Enabled memories */
2926 	em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
2927 
2928 	r = &mem_ranges[0];
2929 	n = 0;
2930 	bzero(r, sizeof(mem_ranges));
2931 	if (em & F_EDRAM0_ENABLE) {
2932 		addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
2933 		r->size = G_EDRAM0_SIZE(addr_len) << 20;
2934 		if (r->size > 0) {
2935 			r->start = G_EDRAM0_BASE(addr_len) << 20;
2936 			if (addr >= r->start &&
2937 			    addr + len <= r->start + r->size)
2938 				return (0);
2939 			r++;
2940 			n++;
2941 		}
2942 	}
2943 	if (em & F_EDRAM1_ENABLE) {
2944 		addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
2945 		r->size = G_EDRAM1_SIZE(addr_len) << 20;
2946 		if (r->size > 0) {
2947 			r->start = G_EDRAM1_BASE(addr_len) << 20;
2948 			if (addr >= r->start &&
2949 			    addr + len <= r->start + r->size)
2950 				return (0);
2951 			r++;
2952 			n++;
2953 		}
2954 	}
2955 	if (em & F_EXT_MEM_ENABLE) {
2956 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
2957 		r->size = G_EXT_MEM_SIZE(addr_len) << 20;
2958 		if (r->size > 0) {
2959 			r->start = G_EXT_MEM_BASE(addr_len) << 20;
2960 			if (addr >= r->start &&
2961 			    addr + len <= r->start + r->size)
2962 				return (0);
2963 			r++;
2964 			n++;
2965 		}
2966 	}
2967 	if (is_t5(sc) && em & F_EXT_MEM1_ENABLE) {
2968 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
2969 		r->size = G_EXT_MEM1_SIZE(addr_len) << 20;
2970 		if (r->size > 0) {
2971 			r->start = G_EXT_MEM1_BASE(addr_len) << 20;
2972 			if (addr >= r->start &&
2973 			    addr + len <= r->start + r->size)
2974 				return (0);
2975 			r++;
2976 			n++;
2977 		}
2978 	}
2979 	MPASS(n <= nitems(mem_ranges));
2980 
2981 	if (n > 1) {
2982 		/* Sort and merge the ranges. */
2983 		qsort(mem_ranges, n, sizeof(struct t4_range), t4_range_cmp);
2984 
2985 		/* Start from index 0 and examine the next n - 1 entries. */
2986 		r = &mem_ranges[0];
2987 		for (remaining = n - 1; remaining > 0; remaining--, r++) {
2988 
2989 			MPASS(r->size > 0);	/* r is a valid entry. */
2990 			next = r + 1;
2991 			MPASS(next->size > 0);	/* and so is the next one. */
2992 
2993 			while (r->start + r->size >= next->start) {
2994 				/* Merge the next one into the current entry. */
2995 				r->size = max(r->start + r->size,
2996 				    next->start + next->size) - r->start;
2997 				n--;	/* One fewer entry in total. */
2998 				if (--remaining == 0)
2999 					goto done;	/* short circuit */
3000 				next++;
3001 			}
3002 			if (next != r + 1) {
3003 				/*
3004 				 * Some entries were merged into r and next
3005 				 * points to the first valid entry that couldn't
3006 				 * be merged.
3007 				 */
3008 				MPASS(next->size > 0);	/* must be valid */
3009 				memcpy(r + 1, next, remaining * sizeof(*r));
3010 #ifdef INVARIANTS
3011 				/*
3012 				 * This so that the foo->size assertion in the
3013 				 * next iteration of the loop do the right
3014 				 * thing for entries that were pulled up and are
3015 				 * no longer valid.
3016 				 */
3017 				MPASS(n < nitems(mem_ranges));
3018 				bzero(&mem_ranges[n], (nitems(mem_ranges) - n) *
3019 				    sizeof(struct t4_range));
3020 #endif
3021 			}
3022 		}
3023 done:
3024 		/* Done merging the ranges. */
3025 		MPASS(n > 0);
3026 		r = &mem_ranges[0];
3027 		for (i = 0; i < n; i++, r++) {
3028 			if (addr >= r->start &&
3029 			    addr + len <= r->start + r->size)
3030 				return (0);
3031 		}
3032 	}
3033 
3034 	return (EFAULT);
3035 }
3036 
3037 static int
3038 fwmtype_to_hwmtype(int mtype)
3039 {
3040 
3041 	switch (mtype) {
3042 	case FW_MEMTYPE_EDC0:
3043 		return (MEM_EDC0);
3044 	case FW_MEMTYPE_EDC1:
3045 		return (MEM_EDC1);
3046 	case FW_MEMTYPE_EXTMEM:
3047 		return (MEM_MC0);
3048 	case FW_MEMTYPE_EXTMEM1:
3049 		return (MEM_MC1);
3050 	default:
3051 		panic("%s: cannot translate fw mtype %d.", __func__, mtype);
3052 	}
3053 }
3054 
3055 /*
3056  * Verify that the memory range specified by the memtype/offset/len pair is
3057  * valid and lies entirely within the memtype specified.  The global address of
3058  * the start of the range is returned in addr.
3059  */
3060 static int
3061 validate_mt_off_len(struct adapter *sc, int mtype, uint32_t off, uint32_t len,
3062     uint32_t *addr)
3063 {
3064 	uint32_t em, addr_len, maddr;
3065 
3066 	/* Memory can only be accessed in naturally aligned 4 byte units */
3067 	if (off & 3 || len & 3 || len == 0)
3068 		return (EINVAL);
3069 
3070 	em = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
3071 	switch (fwmtype_to_hwmtype(mtype)) {
3072 	case MEM_EDC0:
3073 		if (!(em & F_EDRAM0_ENABLE))
3074 			return (EINVAL);
3075 		addr_len = t4_read_reg(sc, A_MA_EDRAM0_BAR);
3076 		maddr = G_EDRAM0_BASE(addr_len) << 20;
3077 		break;
3078 	case MEM_EDC1:
3079 		if (!(em & F_EDRAM1_ENABLE))
3080 			return (EINVAL);
3081 		addr_len = t4_read_reg(sc, A_MA_EDRAM1_BAR);
3082 		maddr = G_EDRAM1_BASE(addr_len) << 20;
3083 		break;
3084 	case MEM_MC:
3085 		if (!(em & F_EXT_MEM_ENABLE))
3086 			return (EINVAL);
3087 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
3088 		maddr = G_EXT_MEM_BASE(addr_len) << 20;
3089 		break;
3090 	case MEM_MC1:
3091 		if (!is_t5(sc) || !(em & F_EXT_MEM1_ENABLE))
3092 			return (EINVAL);
3093 		addr_len = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
3094 		maddr = G_EXT_MEM1_BASE(addr_len) << 20;
3095 		break;
3096 	default:
3097 		return (EINVAL);
3098 	}
3099 
3100 	*addr = maddr + off;	/* global address */
3101 	return (validate_mem_range(sc, *addr, len));
3102 }
3103 
3104 static int
3105 fixup_devlog_params(struct adapter *sc)
3106 {
3107 	struct devlog_params *dparams = &sc->params.devlog;
3108 	int rc;
3109 
3110 	rc = validate_mt_off_len(sc, dparams->memtype, dparams->start,
3111 	    dparams->size, &dparams->addr);
3112 
3113 	return (rc);
3114 }
3115 
3116 static void
3117 update_nirq(struct intrs_and_queues *iaq, int nports)
3118 {
3119 	int extra = T4_EXTRA_INTR;
3120 
3121 	iaq->nirq = extra;
3122 	iaq->nirq += nports * (iaq->nrxq + iaq->nofldrxq);
3123 	iaq->nirq += nports * (iaq->num_vis - 1) *
3124 	    max(iaq->nrxq_vi, iaq->nnmrxq_vi);
3125 	iaq->nirq += nports * (iaq->num_vis - 1) * iaq->nofldrxq_vi;
3126 }
3127 
3128 /*
3129  * Adjust requirements to fit the number of interrupts available.
3130  */
3131 static void
3132 calculate_iaq(struct adapter *sc, struct intrs_and_queues *iaq, int itype,
3133     int navail)
3134 {
3135 	int old_nirq;
3136 	const int nports = sc->params.nports;
3137 
3138 	MPASS(nports > 0);
3139 	MPASS(navail > 0);
3140 
3141 	bzero(iaq, sizeof(*iaq));
3142 	iaq->intr_type = itype;
3143 	iaq->num_vis = t4_num_vis;
3144 	iaq->ntxq = t4_ntxq;
3145 	iaq->ntxq_vi = t4_ntxq_vi;
3146 	iaq->nrxq = t4_nrxq;
3147 	iaq->nrxq_vi = t4_nrxq_vi;
3148 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
3149 	if (is_offload(sc) || is_ethoffload(sc)) {
3150 		iaq->nofldtxq = t4_nofldtxq;
3151 		iaq->nofldtxq_vi = t4_nofldtxq_vi;
3152 	}
3153 #endif
3154 #ifdef TCP_OFFLOAD
3155 	if (is_offload(sc)) {
3156 		iaq->nofldrxq = t4_nofldrxq;
3157 		iaq->nofldrxq_vi = t4_nofldrxq_vi;
3158 	}
3159 #endif
3160 #ifdef DEV_NETMAP
3161 	iaq->nnmtxq_vi = t4_nnmtxq_vi;
3162 	iaq->nnmrxq_vi = t4_nnmrxq_vi;
3163 #endif
3164 
3165 	update_nirq(iaq, nports);
3166 	if (iaq->nirq <= navail &&
3167 	    (itype != INTR_MSI || powerof2(iaq->nirq))) {
3168 		/*
3169 		 * This is the normal case -- there are enough interrupts for
3170 		 * everything.
3171 		 */
3172 		goto done;
3173 	}
3174 
3175 	/*
3176 	 * If extra VIs have been configured try reducing their count and see if
3177 	 * that works.
3178 	 */
3179 	while (iaq->num_vis > 1) {
3180 		iaq->num_vis--;
3181 		update_nirq(iaq, nports);
3182 		if (iaq->nirq <= navail &&
3183 		    (itype != INTR_MSI || powerof2(iaq->nirq))) {
3184 			device_printf(sc->dev, "virtual interfaces per port "
3185 			    "reduced to %d from %d.  nrxq=%u, nofldrxq=%u, "
3186 			    "nrxq_vi=%u nofldrxq_vi=%u, nnmrxq_vi=%u.  "
3187 			    "itype %d, navail %u, nirq %d.\n",
3188 			    iaq->num_vis, t4_num_vis, iaq->nrxq, iaq->nofldrxq,
3189 			    iaq->nrxq_vi, iaq->nofldrxq_vi, iaq->nnmrxq_vi,
3190 			    itype, navail, iaq->nirq);
3191 			goto done;
3192 		}
3193 	}
3194 
3195 	/*
3196 	 * Extra VIs will not be created.  Log a message if they were requested.
3197 	 */
3198 	MPASS(iaq->num_vis == 1);
3199 	iaq->ntxq_vi = iaq->nrxq_vi = 0;
3200 	iaq->nofldtxq_vi = iaq->nofldrxq_vi = 0;
3201 	iaq->nnmtxq_vi = iaq->nnmrxq_vi = 0;
3202 	if (iaq->num_vis != t4_num_vis) {
3203 		device_printf(sc->dev, "extra virtual interfaces disabled.  "
3204 		    "nrxq=%u, nofldrxq=%u, nrxq_vi=%u nofldrxq_vi=%u, "
3205 		    "nnmrxq_vi=%u.  itype %d, navail %u, nirq %d.\n",
3206 		    iaq->nrxq, iaq->nofldrxq, iaq->nrxq_vi, iaq->nofldrxq_vi,
3207 		    iaq->nnmrxq_vi, itype, navail, iaq->nirq);
3208 	}
3209 
3210 	/*
3211 	 * Keep reducing the number of NIC rx queues to the next lower power of
3212 	 * 2 (for even RSS distribution) and halving the TOE rx queues and see
3213 	 * if that works.
3214 	 */
3215 	do {
3216 		if (iaq->nrxq > 1) {
3217 			do {
3218 				iaq->nrxq--;
3219 			} while (!powerof2(iaq->nrxq));
3220 		}
3221 		if (iaq->nofldrxq > 1)
3222 			iaq->nofldrxq >>= 1;
3223 
3224 		old_nirq = iaq->nirq;
3225 		update_nirq(iaq, nports);
3226 		if (iaq->nirq <= navail &&
3227 		    (itype != INTR_MSI || powerof2(iaq->nirq))) {
3228 			device_printf(sc->dev, "running with reduced number of "
3229 			    "rx queues because of shortage of interrupts.  "
3230 			    "nrxq=%u, nofldrxq=%u.  "
3231 			    "itype %d, navail %u, nirq %d.\n", iaq->nrxq,
3232 			    iaq->nofldrxq, itype, navail, iaq->nirq);
3233 			goto done;
3234 		}
3235 	} while (old_nirq != iaq->nirq);
3236 
3237 	/* One interrupt for everything.  Ugh. */
3238 	device_printf(sc->dev, "running with minimal number of queues.  "
3239 	    "itype %d, navail %u.\n", itype, navail);
3240 	iaq->nirq = 1;
3241 	MPASS(iaq->nrxq == 1);
3242 	iaq->ntxq = 1;
3243 	if (iaq->nofldrxq > 1)
3244 		iaq->nofldtxq = 1;
3245 done:
3246 	MPASS(iaq->num_vis > 0);
3247 	if (iaq->num_vis > 1) {
3248 		MPASS(iaq->nrxq_vi > 0);
3249 		MPASS(iaq->ntxq_vi > 0);
3250 	}
3251 	MPASS(iaq->nirq > 0);
3252 	MPASS(iaq->nrxq > 0);
3253 	MPASS(iaq->ntxq > 0);
3254 	if (itype == INTR_MSI) {
3255 		MPASS(powerof2(iaq->nirq));
3256 	}
3257 }
3258 
3259 static int
3260 cfg_itype_and_nqueues(struct adapter *sc, struct intrs_and_queues *iaq)
3261 {
3262 	int rc, itype, navail, nalloc;
3263 
3264 	for (itype = INTR_MSIX; itype; itype >>= 1) {
3265 
3266 		if ((itype & t4_intr_types) == 0)
3267 			continue;	/* not allowed */
3268 
3269 		if (itype == INTR_MSIX)
3270 			navail = pci_msix_count(sc->dev);
3271 		else if (itype == INTR_MSI)
3272 			navail = pci_msi_count(sc->dev);
3273 		else
3274 			navail = 1;
3275 restart:
3276 		if (navail == 0)
3277 			continue;
3278 
3279 		calculate_iaq(sc, iaq, itype, navail);
3280 		nalloc = iaq->nirq;
3281 		rc = 0;
3282 		if (itype == INTR_MSIX)
3283 			rc = pci_alloc_msix(sc->dev, &nalloc);
3284 		else if (itype == INTR_MSI)
3285 			rc = pci_alloc_msi(sc->dev, &nalloc);
3286 
3287 		if (rc == 0 && nalloc > 0) {
3288 			if (nalloc == iaq->nirq)
3289 				return (0);
3290 
3291 			/*
3292 			 * Didn't get the number requested.  Use whatever number
3293 			 * the kernel is willing to allocate.
3294 			 */
3295 			device_printf(sc->dev, "fewer vectors than requested, "
3296 			    "type=%d, req=%d, rcvd=%d; will downshift req.\n",
3297 			    itype, iaq->nirq, nalloc);
3298 			pci_release_msi(sc->dev);
3299 			navail = nalloc;
3300 			goto restart;
3301 		}
3302 
3303 		device_printf(sc->dev,
3304 		    "failed to allocate vectors:%d, type=%d, req=%d, rcvd=%d\n",
3305 		    itype, rc, iaq->nirq, nalloc);
3306 	}
3307 
3308 	device_printf(sc->dev,
3309 	    "failed to find a usable interrupt type.  "
3310 	    "allowed=%d, msi-x=%d, msi=%d, intx=1", t4_intr_types,
3311 	    pci_msix_count(sc->dev), pci_msi_count(sc->dev));
3312 
3313 	return (ENXIO);
3314 }
3315 
3316 #define FW_VERSION(chip) ( \
3317     V_FW_HDR_FW_VER_MAJOR(chip##FW_VERSION_MAJOR) | \
3318     V_FW_HDR_FW_VER_MINOR(chip##FW_VERSION_MINOR) | \
3319     V_FW_HDR_FW_VER_MICRO(chip##FW_VERSION_MICRO) | \
3320     V_FW_HDR_FW_VER_BUILD(chip##FW_VERSION_BUILD))
3321 #define FW_INTFVER(chip, intf) (chip##FW_HDR_INTFVER_##intf)
3322 
3323 struct fw_info {
3324 	uint8_t chip;
3325 	char *kld_name;
3326 	char *fw_mod_name;
3327 	struct fw_hdr fw_hdr;	/* XXX: waste of space, need a sparse struct */
3328 } fw_info[] = {
3329 	{
3330 		.chip = CHELSIO_T4,
3331 		.kld_name = "t4fw_cfg",
3332 		.fw_mod_name = "t4fw",
3333 		.fw_hdr = {
3334 			.chip = FW_HDR_CHIP_T4,
3335 			.fw_ver = htobe32(FW_VERSION(T4)),
3336 			.intfver_nic = FW_INTFVER(T4, NIC),
3337 			.intfver_vnic = FW_INTFVER(T4, VNIC),
3338 			.intfver_ofld = FW_INTFVER(T4, OFLD),
3339 			.intfver_ri = FW_INTFVER(T4, RI),
3340 			.intfver_iscsipdu = FW_INTFVER(T4, ISCSIPDU),
3341 			.intfver_iscsi = FW_INTFVER(T4, ISCSI),
3342 			.intfver_fcoepdu = FW_INTFVER(T4, FCOEPDU),
3343 			.intfver_fcoe = FW_INTFVER(T4, FCOE),
3344 		},
3345 	}, {
3346 		.chip = CHELSIO_T5,
3347 		.kld_name = "t5fw_cfg",
3348 		.fw_mod_name = "t5fw",
3349 		.fw_hdr = {
3350 			.chip = FW_HDR_CHIP_T5,
3351 			.fw_ver = htobe32(FW_VERSION(T5)),
3352 			.intfver_nic = FW_INTFVER(T5, NIC),
3353 			.intfver_vnic = FW_INTFVER(T5, VNIC),
3354 			.intfver_ofld = FW_INTFVER(T5, OFLD),
3355 			.intfver_ri = FW_INTFVER(T5, RI),
3356 			.intfver_iscsipdu = FW_INTFVER(T5, ISCSIPDU),
3357 			.intfver_iscsi = FW_INTFVER(T5, ISCSI),
3358 			.intfver_fcoepdu = FW_INTFVER(T5, FCOEPDU),
3359 			.intfver_fcoe = FW_INTFVER(T5, FCOE),
3360 		},
3361 	}, {
3362 		.chip = CHELSIO_T6,
3363 		.kld_name = "t6fw_cfg",
3364 		.fw_mod_name = "t6fw",
3365 		.fw_hdr = {
3366 			.chip = FW_HDR_CHIP_T6,
3367 			.fw_ver = htobe32(FW_VERSION(T6)),
3368 			.intfver_nic = FW_INTFVER(T6, NIC),
3369 			.intfver_vnic = FW_INTFVER(T6, VNIC),
3370 			.intfver_ofld = FW_INTFVER(T6, OFLD),
3371 			.intfver_ri = FW_INTFVER(T6, RI),
3372 			.intfver_iscsipdu = FW_INTFVER(T6, ISCSIPDU),
3373 			.intfver_iscsi = FW_INTFVER(T6, ISCSI),
3374 			.intfver_fcoepdu = FW_INTFVER(T6, FCOEPDU),
3375 			.intfver_fcoe = FW_INTFVER(T6, FCOE),
3376 		},
3377 	}
3378 };
3379 
3380 static struct fw_info *
3381 find_fw_info(int chip)
3382 {
3383 	int i;
3384 
3385 	for (i = 0; i < nitems(fw_info); i++) {
3386 		if (fw_info[i].chip == chip)
3387 			return (&fw_info[i]);
3388 	}
3389 	return (NULL);
3390 }
3391 
3392 /*
3393  * Is the given firmware API compatible with the one the driver was compiled
3394  * with?
3395  */
3396 static int
3397 fw_compatible(const struct fw_hdr *hdr1, const struct fw_hdr *hdr2)
3398 {
3399 
3400 	/* short circuit if it's the exact same firmware version */
3401 	if (hdr1->chip == hdr2->chip && hdr1->fw_ver == hdr2->fw_ver)
3402 		return (1);
3403 
3404 	/*
3405 	 * XXX: Is this too conservative?  Perhaps I should limit this to the
3406 	 * features that are supported in the driver.
3407 	 */
3408 #define SAME_INTF(x) (hdr1->intfver_##x == hdr2->intfver_##x)
3409 	if (hdr1->chip == hdr2->chip && SAME_INTF(nic) && SAME_INTF(vnic) &&
3410 	    SAME_INTF(ofld) && SAME_INTF(ri) && SAME_INTF(iscsipdu) &&
3411 	    SAME_INTF(iscsi) && SAME_INTF(fcoepdu) && SAME_INTF(fcoe))
3412 		return (1);
3413 #undef SAME_INTF
3414 
3415 	return (0);
3416 }
3417 
3418 static int
3419 load_fw_module(struct adapter *sc, const struct firmware **dcfg,
3420     const struct firmware **fw)
3421 {
3422 	struct fw_info *fw_info;
3423 
3424 	*dcfg = NULL;
3425 	if (fw != NULL)
3426 		*fw = NULL;
3427 
3428 	fw_info = find_fw_info(chip_id(sc));
3429 	if (fw_info == NULL) {
3430 		device_printf(sc->dev,
3431 		    "unable to look up firmware information for chip %d.\n",
3432 		    chip_id(sc));
3433 		return (EINVAL);
3434 	}
3435 
3436 	*dcfg = firmware_get(fw_info->kld_name);
3437 	if (*dcfg != NULL) {
3438 		if (fw != NULL)
3439 			*fw = firmware_get(fw_info->fw_mod_name);
3440 		return (0);
3441 	}
3442 
3443 	return (ENOENT);
3444 }
3445 
3446 static void
3447 unload_fw_module(struct adapter *sc, const struct firmware *dcfg,
3448     const struct firmware *fw)
3449 {
3450 
3451 	if (fw != NULL)
3452 		firmware_put(fw, FIRMWARE_UNLOAD);
3453 	if (dcfg != NULL)
3454 		firmware_put(dcfg, FIRMWARE_UNLOAD);
3455 }
3456 
3457 /*
3458  * Return values:
3459  * 0 means no firmware install attempted.
3460  * ERESTART means a firmware install was attempted and was successful.
3461  * +ve errno means a firmware install was attempted but failed.
3462  */
3463 static int
3464 install_kld_firmware(struct adapter *sc, struct fw_hdr *card_fw,
3465     const struct fw_hdr *drv_fw, const char *reason, int *already)
3466 {
3467 	const struct firmware *cfg, *fw;
3468 	const uint32_t c = be32toh(card_fw->fw_ver);
3469 	const uint32_t d = be32toh(drv_fw->fw_ver);
3470 	uint32_t k;
3471 	int rc;
3472 
3473 	if (reason != NULL)
3474 		goto install;
3475 
3476 	if ((sc->flags & FW_OK) == 0) {
3477 
3478 		if (c == 0xffffffff) {
3479 			reason = "missing";
3480 			goto install;
3481 		}
3482 
3483 		return (0);
3484 	}
3485 
3486 	if (!fw_compatible(card_fw, drv_fw)) {
3487 		reason = "incompatible or unusable";
3488 		goto install;
3489 	}
3490 
3491 	if (d > c) {
3492 		reason = "older than the version bundled with this driver";
3493 		goto install;
3494 	}
3495 
3496 	if (t4_fw_install == 2 && d != c) {
3497 		reason = "different than the version bundled with this driver";
3498 		goto install;
3499 	}
3500 
3501 	return (0);
3502 
3503 install:
3504 	if ((*already)++)
3505 		return (0);
3506 
3507 	if (t4_fw_install == 0) {
3508 		device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
3509 		    "but the driver is prohibited from installing a firmware "
3510 		    "on the card.\n",
3511 		    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
3512 		    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason);
3513 
3514 		return (0);
3515 	}
3516 
3517 	device_printf(sc->dev, "firmware on card (%u.%u.%u.%u) is %s, "
3518 	    "installing firmware %u.%u.%u.%u on card.\n",
3519 	    G_FW_HDR_FW_VER_MAJOR(c), G_FW_HDR_FW_VER_MINOR(c),
3520 	    G_FW_HDR_FW_VER_MICRO(c), G_FW_HDR_FW_VER_BUILD(c), reason,
3521 	    G_FW_HDR_FW_VER_MAJOR(d), G_FW_HDR_FW_VER_MINOR(d),
3522 	    G_FW_HDR_FW_VER_MICRO(d), G_FW_HDR_FW_VER_BUILD(d));
3523 
3524 	rc = load_fw_module(sc, &cfg, &fw);
3525 	if (rc != 0 || fw == NULL) {
3526 		device_printf(sc->dev,
3527 		    "failed to load firmware module: %d. cfg %p, fw %p\n", rc,
3528 		    cfg, fw);
3529 		rc = sc->flags & FW_OK ? 0 : ENOENT;
3530 		goto done;
3531 	}
3532 	k = be32toh(((const struct fw_hdr *)fw->data)->fw_ver);
3533 	if (k != d) {
3534 		device_printf(sc->dev,
3535 		    "firmware in KLD (%u.%u.%u.%u) is not what the driver was "
3536 		    "compiled with and will not be used.\n",
3537 		    G_FW_HDR_FW_VER_MAJOR(k), G_FW_HDR_FW_VER_MINOR(k),
3538 		    G_FW_HDR_FW_VER_MICRO(k), G_FW_HDR_FW_VER_BUILD(k));
3539 		rc = sc->flags & FW_OK ? 0 : EINVAL;
3540 		goto done;
3541 	}
3542 
3543 	rc = -t4_fw_upgrade(sc, sc->mbox, fw->data, fw->datasize, 0);
3544 	if (rc != 0) {
3545 		device_printf(sc->dev, "failed to install firmware: %d\n", rc);
3546 	} else {
3547 		/* Installed successfully, update the cached header too. */
3548 		rc = ERESTART;
3549 		memcpy(card_fw, fw->data, sizeof(*card_fw));
3550 	}
3551 done:
3552 	unload_fw_module(sc, cfg, fw);
3553 
3554 	return (rc);
3555 }
3556 
3557 /*
3558  * Establish contact with the firmware and attempt to become the master driver.
3559  *
3560  * A firmware will be installed to the card if needed (if the driver is allowed
3561  * to do so).
3562  */
3563 static int
3564 contact_firmware(struct adapter *sc)
3565 {
3566 	int rc, already = 0;
3567 	enum dev_state state;
3568 	struct fw_info *fw_info;
3569 	struct fw_hdr *card_fw;		/* fw on the card */
3570 	const struct fw_hdr *drv_fw;	/* fw bundled with the driver */
3571 
3572 	fw_info = find_fw_info(chip_id(sc));
3573 	if (fw_info == NULL) {
3574 		device_printf(sc->dev,
3575 		    "unable to look up firmware information for chip %d.\n",
3576 		    chip_id(sc));
3577 		return (EINVAL);
3578 	}
3579 	drv_fw = &fw_info->fw_hdr;
3580 
3581 	/* Read the header of the firmware on the card */
3582 	card_fw = malloc(sizeof(*card_fw), M_CXGBE, M_ZERO | M_WAITOK);
3583 restart:
3584 	rc = -t4_get_fw_hdr(sc, card_fw);
3585 	if (rc != 0) {
3586 		device_printf(sc->dev,
3587 		    "unable to read firmware header from card's flash: %d\n",
3588 		    rc);
3589 		goto done;
3590 	}
3591 
3592 	rc = install_kld_firmware(sc, card_fw, drv_fw, NULL, &already);
3593 	if (rc == ERESTART)
3594 		goto restart;
3595 	if (rc != 0)
3596 		goto done;
3597 
3598 	rc = t4_fw_hello(sc, sc->mbox, sc->mbox, MASTER_MAY, &state);
3599 	if (rc < 0 || state == DEV_STATE_ERR) {
3600 		rc = -rc;
3601 		device_printf(sc->dev,
3602 		    "failed to connect to the firmware: %d, %d.  "
3603 		    "PCIE_FW 0x%08x\n", rc, state, t4_read_reg(sc, A_PCIE_FW));
3604 #if 0
3605 		if (install_kld_firmware(sc, card_fw, drv_fw,
3606 		    "not responding properly to HELLO", &already) == ERESTART)
3607 			goto restart;
3608 #endif
3609 		goto done;
3610 	}
3611 	MPASS(be32toh(card_fw->flags) & FW_HDR_FLAGS_RESET_HALT);
3612 	sc->flags |= FW_OK;	/* The firmware responded to the FW_HELLO. */
3613 
3614 	if (rc == sc->pf) {
3615 		sc->flags |= MASTER_PF;
3616 		rc = install_kld_firmware(sc, card_fw, drv_fw, NULL, &already);
3617 		if (rc == ERESTART)
3618 			rc = 0;
3619 		else if (rc != 0)
3620 			goto done;
3621 	} else if (state == DEV_STATE_UNINIT) {
3622 		/*
3623 		 * We didn't get to be the master so we definitely won't be
3624 		 * configuring the chip.  It's a bug if someone else hasn't
3625 		 * configured it already.
3626 		 */
3627 		device_printf(sc->dev, "couldn't be master(%d), "
3628 		    "device not already initialized either(%d).  "
3629 		    "PCIE_FW 0x%08x\n", rc, state, t4_read_reg(sc, A_PCIE_FW));
3630 		rc = EPROTO;
3631 		goto done;
3632 	} else {
3633 		/*
3634 		 * Some other PF is the master and has configured the chip.
3635 		 * This is allowed but untested.
3636 		 */
3637 		device_printf(sc->dev, "PF%d is master, device state %d.  "
3638 		    "PCIE_FW 0x%08x\n", rc, state, t4_read_reg(sc, A_PCIE_FW));
3639 		snprintf(sc->cfg_file, sizeof(sc->cfg_file), "pf%d", rc);
3640 		sc->cfcsum = 0;
3641 		rc = 0;
3642 	}
3643 done:
3644 	if (rc != 0 && sc->flags & FW_OK) {
3645 		t4_fw_bye(sc, sc->mbox);
3646 		sc->flags &= ~FW_OK;
3647 	}
3648 	free(card_fw, M_CXGBE);
3649 	return (rc);
3650 }
3651 
3652 static int
3653 copy_cfg_file_to_card(struct adapter *sc, char *cfg_file,
3654     uint32_t mtype, uint32_t moff)
3655 {
3656 	struct fw_info *fw_info;
3657 	const struct firmware *dcfg, *rcfg = NULL;
3658 	const uint32_t *cfdata;
3659 	uint32_t cflen, addr;
3660 	int rc;
3661 
3662 	load_fw_module(sc, &dcfg, NULL);
3663 
3664 	/* Card specific interpretation of "default". */
3665 	if (strncmp(cfg_file, DEFAULT_CF, sizeof(t4_cfg_file)) == 0) {
3666 		if (pci_get_device(sc->dev) == 0x440a)
3667 			snprintf(cfg_file, sizeof(t4_cfg_file), UWIRE_CF);
3668 		if (is_fpga(sc))
3669 			snprintf(cfg_file, sizeof(t4_cfg_file), FPGA_CF);
3670 	}
3671 
3672 	if (strncmp(cfg_file, DEFAULT_CF, sizeof(t4_cfg_file)) == 0) {
3673 		if (dcfg == NULL) {
3674 			device_printf(sc->dev,
3675 			    "KLD with default config is not available.\n");
3676 			rc = ENOENT;
3677 			goto done;
3678 		}
3679 		cfdata = dcfg->data;
3680 		cflen = dcfg->datasize & ~3;
3681 	} else {
3682 		char s[32];
3683 
3684 		fw_info = find_fw_info(chip_id(sc));
3685 		if (fw_info == NULL) {
3686 			device_printf(sc->dev,
3687 			    "unable to look up firmware information for chip %d.\n",
3688 			    chip_id(sc));
3689 			rc = EINVAL;
3690 			goto done;
3691 		}
3692 		snprintf(s, sizeof(s), "%s_%s", fw_info->kld_name, cfg_file);
3693 
3694 		rcfg = firmware_get(s);
3695 		if (rcfg == NULL) {
3696 			device_printf(sc->dev,
3697 			    "unable to load module \"%s\" for configuration "
3698 			    "profile \"%s\".\n", s, cfg_file);
3699 			rc = ENOENT;
3700 			goto done;
3701 		}
3702 		cfdata = rcfg->data;
3703 		cflen = rcfg->datasize & ~3;
3704 	}
3705 
3706 	if (cflen > FLASH_CFG_MAX_SIZE) {
3707 		device_printf(sc->dev,
3708 		    "config file too long (%d, max allowed is %d).\n",
3709 		    cflen, FLASH_CFG_MAX_SIZE);
3710 		rc = EINVAL;
3711 		goto done;
3712 	}
3713 
3714 	rc = validate_mt_off_len(sc, mtype, moff, cflen, &addr);
3715 	if (rc != 0) {
3716 		device_printf(sc->dev,
3717 		    "%s: addr (%d/0x%x) or len %d is not valid: %d.\n",
3718 		    __func__, mtype, moff, cflen, rc);
3719 		rc = EINVAL;
3720 		goto done;
3721 	}
3722 	write_via_memwin(sc, 2, addr, cfdata, cflen);
3723 done:
3724 	if (rcfg != NULL)
3725 		firmware_put(rcfg, FIRMWARE_UNLOAD);
3726 	unload_fw_module(sc, dcfg, NULL);
3727 	return (rc);
3728 }
3729 
3730 struct caps_allowed {
3731 	uint16_t nbmcaps;
3732 	uint16_t linkcaps;
3733 	uint16_t switchcaps;
3734 	uint16_t niccaps;
3735 	uint16_t toecaps;
3736 	uint16_t rdmacaps;
3737 	uint16_t cryptocaps;
3738 	uint16_t iscsicaps;
3739 	uint16_t fcoecaps;
3740 };
3741 
3742 #define FW_PARAM_DEV(param) \
3743 	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) | \
3744 	 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_##param))
3745 #define FW_PARAM_PFVF(param) \
3746 	(V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_PFVF) | \
3747 	 V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_PFVF_##param))
3748 
3749 /*
3750  * Provide a configuration profile to the firmware and have it initialize the
3751  * chip accordingly.  This may involve uploading a configuration file to the
3752  * card.
3753  */
3754 static int
3755 apply_cfg_and_initialize(struct adapter *sc, char *cfg_file,
3756     const struct caps_allowed *caps_allowed)
3757 {
3758 	int rc;
3759 	struct fw_caps_config_cmd caps;
3760 	uint32_t mtype, moff, finicsum, cfcsum, param, val;
3761 
3762 	rc = -t4_fw_reset(sc, sc->mbox, F_PIORSTMODE | F_PIORST);
3763 	if (rc != 0) {
3764 		device_printf(sc->dev, "firmware reset failed: %d.\n", rc);
3765 		return (rc);
3766 	}
3767 
3768 	bzero(&caps, sizeof(caps));
3769 	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3770 	    F_FW_CMD_REQUEST | F_FW_CMD_READ);
3771 	if (strncmp(cfg_file, BUILTIN_CF, sizeof(t4_cfg_file)) == 0) {
3772 		mtype = 0;
3773 		moff = 0;
3774 		caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
3775 	} else if (strncmp(cfg_file, FLASH_CF, sizeof(t4_cfg_file)) == 0) {
3776 		mtype = FW_MEMTYPE_FLASH;
3777 		moff = t4_flash_cfg_addr(sc);
3778 		caps.cfvalid_to_len16 = htobe32(F_FW_CAPS_CONFIG_CMD_CFVALID |
3779 		    V_FW_CAPS_CONFIG_CMD_MEMTYPE_CF(mtype) |
3780 		    V_FW_CAPS_CONFIG_CMD_MEMADDR64K_CF(moff >> 16) |
3781 		    FW_LEN16(caps));
3782 	} else {
3783 		/*
3784 		 * Ask the firmware where it wants us to upload the config file.
3785 		 */
3786 		param = FW_PARAM_DEV(CF);
3787 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
3788 		if (rc != 0) {
3789 			/* No support for config file?  Shouldn't happen. */
3790 			device_printf(sc->dev,
3791 			    "failed to query config file location: %d.\n", rc);
3792 			goto done;
3793 		}
3794 		mtype = G_FW_PARAMS_PARAM_Y(val);
3795 		moff = G_FW_PARAMS_PARAM_Z(val) << 16;
3796 		caps.cfvalid_to_len16 = htobe32(F_FW_CAPS_CONFIG_CMD_CFVALID |
3797 		    V_FW_CAPS_CONFIG_CMD_MEMTYPE_CF(mtype) |
3798 		    V_FW_CAPS_CONFIG_CMD_MEMADDR64K_CF(moff >> 16) |
3799 		    FW_LEN16(caps));
3800 
3801 		rc = copy_cfg_file_to_card(sc, cfg_file, mtype, moff);
3802 		if (rc != 0) {
3803 			device_printf(sc->dev,
3804 			    "failed to upload config file to card: %d.\n", rc);
3805 			goto done;
3806 		}
3807 	}
3808 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
3809 	if (rc != 0) {
3810 		device_printf(sc->dev, "failed to pre-process config file: %d "
3811 		    "(mtype %d, moff 0x%x).\n", rc, mtype, moff);
3812 		goto done;
3813 	}
3814 
3815 	finicsum = be32toh(caps.finicsum);
3816 	cfcsum = be32toh(caps.cfcsum);	/* actual */
3817 	if (finicsum != cfcsum) {
3818 		device_printf(sc->dev,
3819 		    "WARNING: config file checksum mismatch: %08x %08x\n",
3820 		    finicsum, cfcsum);
3821 	}
3822 	sc->cfcsum = cfcsum;
3823 	snprintf(sc->cfg_file, sizeof(sc->cfg_file), "%s", cfg_file);
3824 
3825 	/*
3826 	 * Let the firmware know what features will (not) be used so it can tune
3827 	 * things accordingly.
3828 	 */
3829 #define LIMIT_CAPS(x) do { \
3830 	caps.x##caps &= htobe16(caps_allowed->x##caps); \
3831 } while (0)
3832 	LIMIT_CAPS(nbm);
3833 	LIMIT_CAPS(link);
3834 	LIMIT_CAPS(switch);
3835 	LIMIT_CAPS(nic);
3836 	LIMIT_CAPS(toe);
3837 	LIMIT_CAPS(rdma);
3838 	LIMIT_CAPS(crypto);
3839 	LIMIT_CAPS(iscsi);
3840 	LIMIT_CAPS(fcoe);
3841 #undef LIMIT_CAPS
3842 	if (caps.niccaps & htobe16(FW_CAPS_CONFIG_NIC_HASHFILTER)) {
3843 		/*
3844 		 * TOE and hashfilters are mutually exclusive.  It is a config
3845 		 * file or firmware bug if both are reported as available.  Try
3846 		 * to cope with the situation in non-debug builds by disabling
3847 		 * TOE.
3848 		 */
3849 		MPASS(caps.toecaps == 0);
3850 
3851 		caps.toecaps = 0;
3852 		caps.rdmacaps = 0;
3853 		caps.iscsicaps = 0;
3854 	}
3855 
3856 	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
3857 	    F_FW_CMD_REQUEST | F_FW_CMD_WRITE);
3858 	caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
3859 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), NULL);
3860 	if (rc != 0) {
3861 		device_printf(sc->dev,
3862 		    "failed to process config file: %d.\n", rc);
3863 		goto done;
3864 	}
3865 
3866 	t4_tweak_chip_settings(sc);
3867 
3868 	/* get basic stuff going */
3869 	rc = -t4_fw_initialize(sc, sc->mbox);
3870 	if (rc != 0) {
3871 		device_printf(sc->dev, "fw_initialize failed: %d.\n", rc);
3872 		goto done;
3873 	}
3874 done:
3875 	return (rc);
3876 }
3877 
3878 /*
3879  * Partition chip resources for use between various PFs, VFs, etc.
3880  */
3881 static int
3882 partition_resources(struct adapter *sc)
3883 {
3884 	char cfg_file[sizeof(t4_cfg_file)];
3885 	struct caps_allowed caps_allowed;
3886 	int rc;
3887 	bool fallback;
3888 
3889 	/* Only the master driver gets to configure the chip resources. */
3890 	MPASS(sc->flags & MASTER_PF);
3891 
3892 #define COPY_CAPS(x) do { \
3893 	caps_allowed.x##caps = t4_##x##caps_allowed; \
3894 } while (0)
3895 	bzero(&caps_allowed, sizeof(caps_allowed));
3896 	COPY_CAPS(nbm);
3897 	COPY_CAPS(link);
3898 	COPY_CAPS(switch);
3899 	COPY_CAPS(nic);
3900 	COPY_CAPS(toe);
3901 	COPY_CAPS(rdma);
3902 	COPY_CAPS(crypto);
3903 	COPY_CAPS(iscsi);
3904 	COPY_CAPS(fcoe);
3905 	fallback = sc->debug_flags & DF_DISABLE_CFG_RETRY ? false : true;
3906 	snprintf(cfg_file, sizeof(cfg_file), "%s", t4_cfg_file);
3907 retry:
3908 	rc = apply_cfg_and_initialize(sc, cfg_file, &caps_allowed);
3909 	if (rc != 0 && fallback) {
3910 		device_printf(sc->dev,
3911 		    "failed (%d) to configure card with \"%s\" profile, "
3912 		    "will fall back to a basic configuration and retry.\n",
3913 		    rc, cfg_file);
3914 		snprintf(cfg_file, sizeof(cfg_file), "%s", BUILTIN_CF);
3915 		bzero(&caps_allowed, sizeof(caps_allowed));
3916 		COPY_CAPS(nbm);
3917 		COPY_CAPS(link);
3918 		COPY_CAPS(switch);
3919 		COPY_CAPS(nic);
3920 		fallback = false;
3921 		goto retry;
3922 	}
3923 #undef COPY_CAPS
3924 	return (rc);
3925 }
3926 
3927 /*
3928  * Retrieve parameters that are needed (or nice to have) very early.
3929  */
3930 static int
3931 get_params__pre_init(struct adapter *sc)
3932 {
3933 	int rc;
3934 	uint32_t param[2], val[2];
3935 
3936 	t4_get_version_info(sc);
3937 
3938 	snprintf(sc->fw_version, sizeof(sc->fw_version), "%u.%u.%u.%u",
3939 	    G_FW_HDR_FW_VER_MAJOR(sc->params.fw_vers),
3940 	    G_FW_HDR_FW_VER_MINOR(sc->params.fw_vers),
3941 	    G_FW_HDR_FW_VER_MICRO(sc->params.fw_vers),
3942 	    G_FW_HDR_FW_VER_BUILD(sc->params.fw_vers));
3943 
3944 	snprintf(sc->bs_version, sizeof(sc->bs_version), "%u.%u.%u.%u",
3945 	    G_FW_HDR_FW_VER_MAJOR(sc->params.bs_vers),
3946 	    G_FW_HDR_FW_VER_MINOR(sc->params.bs_vers),
3947 	    G_FW_HDR_FW_VER_MICRO(sc->params.bs_vers),
3948 	    G_FW_HDR_FW_VER_BUILD(sc->params.bs_vers));
3949 
3950 	snprintf(sc->tp_version, sizeof(sc->tp_version), "%u.%u.%u.%u",
3951 	    G_FW_HDR_FW_VER_MAJOR(sc->params.tp_vers),
3952 	    G_FW_HDR_FW_VER_MINOR(sc->params.tp_vers),
3953 	    G_FW_HDR_FW_VER_MICRO(sc->params.tp_vers),
3954 	    G_FW_HDR_FW_VER_BUILD(sc->params.tp_vers));
3955 
3956 	snprintf(sc->er_version, sizeof(sc->er_version), "%u.%u.%u.%u",
3957 	    G_FW_HDR_FW_VER_MAJOR(sc->params.er_vers),
3958 	    G_FW_HDR_FW_VER_MINOR(sc->params.er_vers),
3959 	    G_FW_HDR_FW_VER_MICRO(sc->params.er_vers),
3960 	    G_FW_HDR_FW_VER_BUILD(sc->params.er_vers));
3961 
3962 	param[0] = FW_PARAM_DEV(PORTVEC);
3963 	param[1] = FW_PARAM_DEV(CCLK);
3964 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
3965 	if (rc != 0) {
3966 		device_printf(sc->dev,
3967 		    "failed to query parameters (pre_init): %d.\n", rc);
3968 		return (rc);
3969 	}
3970 
3971 	sc->params.portvec = val[0];
3972 	sc->params.nports = bitcount32(val[0]);
3973 	sc->params.vpd.cclk = val[1];
3974 
3975 	/* Read device log parameters. */
3976 	rc = -t4_init_devlog_params(sc, 1);
3977 	if (rc == 0)
3978 		fixup_devlog_params(sc);
3979 	else {
3980 		device_printf(sc->dev,
3981 		    "failed to get devlog parameters: %d.\n", rc);
3982 		rc = 0;	/* devlog isn't critical for device operation */
3983 	}
3984 
3985 	return (rc);
3986 }
3987 
3988 /*
3989  * Retrieve various parameters that are of interest to the driver.  The device
3990  * has been initialized by the firmware at this point.
3991  */
3992 static int
3993 get_params__post_init(struct adapter *sc)
3994 {
3995 	int rc;
3996 	uint32_t param[7], val[7];
3997 	struct fw_caps_config_cmd caps;
3998 
3999 	param[0] = FW_PARAM_PFVF(IQFLINT_START);
4000 	param[1] = FW_PARAM_PFVF(EQ_START);
4001 	param[2] = FW_PARAM_PFVF(FILTER_START);
4002 	param[3] = FW_PARAM_PFVF(FILTER_END);
4003 	param[4] = FW_PARAM_PFVF(L2T_START);
4004 	param[5] = FW_PARAM_PFVF(L2T_END);
4005 	param[6] = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
4006 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_DIAG) |
4007 	    V_FW_PARAMS_PARAM_Y(FW_PARAM_DEV_DIAG_VDD);
4008 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 7, param, val);
4009 	if (rc != 0) {
4010 		device_printf(sc->dev,
4011 		    "failed to query parameters (post_init): %d.\n", rc);
4012 		return (rc);
4013 	}
4014 
4015 	sc->sge.iq_start = val[0];
4016 	sc->sge.eq_start = val[1];
4017 	if ((int)val[3] > (int)val[2]) {
4018 		sc->tids.ftid_base = val[2];
4019 		sc->tids.ftid_end = val[3];
4020 		sc->tids.nftids = val[3] - val[2] + 1;
4021 	}
4022 	sc->vres.l2t.start = val[4];
4023 	sc->vres.l2t.size = val[5] - val[4] + 1;
4024 	KASSERT(sc->vres.l2t.size <= L2T_SIZE,
4025 	    ("%s: L2 table size (%u) larger than expected (%u)",
4026 	    __func__, sc->vres.l2t.size, L2T_SIZE));
4027 	sc->params.core_vdd = val[6];
4028 
4029 	if (chip_id(sc) >= CHELSIO_T6) {
4030 
4031 #ifdef INVARIANTS
4032 		if (sc->params.fw_vers >=
4033 		    (V_FW_HDR_FW_VER_MAJOR(1) | V_FW_HDR_FW_VER_MINOR(20) |
4034 		    V_FW_HDR_FW_VER_MICRO(1) | V_FW_HDR_FW_VER_BUILD(0))) {
4035 			/*
4036 			 * Note that the code to enable the region should run
4037 			 * before t4_fw_initialize and not here.  This is just a
4038 			 * reminder to add said code.
4039 			 */
4040 			device_printf(sc->dev,
4041 			    "hpfilter region not enabled.\n");
4042 		}
4043 #endif
4044 
4045 		sc->tids.tid_base = t4_read_reg(sc,
4046 		    A_LE_DB_ACTIVE_TABLE_START_INDEX);
4047 
4048 		param[0] = FW_PARAM_PFVF(HPFILTER_START);
4049 		param[1] = FW_PARAM_PFVF(HPFILTER_END);
4050 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
4051 		if (rc != 0) {
4052 			device_printf(sc->dev,
4053 			   "failed to query hpfilter parameters: %d.\n", rc);
4054 			return (rc);
4055 		}
4056 		if ((int)val[1] > (int)val[0]) {
4057 			sc->tids.hpftid_base = val[0];
4058 			sc->tids.hpftid_end = val[1];
4059 			sc->tids.nhpftids = val[1] - val[0] + 1;
4060 
4061 			/*
4062 			 * These should go off if the layout changes and the
4063 			 * driver needs to catch up.
4064 			 */
4065 			MPASS(sc->tids.hpftid_base == 0);
4066 			MPASS(sc->tids.tid_base == sc->tids.nhpftids);
4067 		}
4068 	}
4069 
4070 	/*
4071 	 * MPSBGMAP is queried separately because only recent firmwares support
4072 	 * it as a parameter and we don't want the compound query above to fail
4073 	 * on older firmwares.
4074 	 */
4075 	param[0] = FW_PARAM_DEV(MPSBGMAP);
4076 	val[0] = 0;
4077 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, param, val);
4078 	if (rc == 0)
4079 		sc->params.mps_bg_map = val[0];
4080 	else
4081 		sc->params.mps_bg_map = 0;
4082 
4083 	/*
4084 	 * Determine whether the firmware supports the filter2 work request.
4085 	 * This is queried separately for the same reason as MPSBGMAP above.
4086 	 */
4087 	param[0] = FW_PARAM_DEV(FILTER2_WR);
4088 	val[0] = 0;
4089 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, param, val);
4090 	if (rc == 0)
4091 		sc->params.filter2_wr_support = val[0] != 0;
4092 	else
4093 		sc->params.filter2_wr_support = 0;
4094 
4095 	/* get capabilites */
4096 	bzero(&caps, sizeof(caps));
4097 	caps.op_to_write = htobe32(V_FW_CMD_OP(FW_CAPS_CONFIG_CMD) |
4098 	    F_FW_CMD_REQUEST | F_FW_CMD_READ);
4099 	caps.cfvalid_to_len16 = htobe32(FW_LEN16(caps));
4100 	rc = -t4_wr_mbox(sc, sc->mbox, &caps, sizeof(caps), &caps);
4101 	if (rc != 0) {
4102 		device_printf(sc->dev,
4103 		    "failed to get card capabilities: %d.\n", rc);
4104 		return (rc);
4105 	}
4106 
4107 #define READ_CAPS(x) do { \
4108 	sc->x = htobe16(caps.x); \
4109 } while (0)
4110 	READ_CAPS(nbmcaps);
4111 	READ_CAPS(linkcaps);
4112 	READ_CAPS(switchcaps);
4113 	READ_CAPS(niccaps);
4114 	READ_CAPS(toecaps);
4115 	READ_CAPS(rdmacaps);
4116 	READ_CAPS(cryptocaps);
4117 	READ_CAPS(iscsicaps);
4118 	READ_CAPS(fcoecaps);
4119 
4120 	if (sc->niccaps & FW_CAPS_CONFIG_NIC_HASHFILTER) {
4121 		MPASS(chip_id(sc) > CHELSIO_T4);
4122 		MPASS(sc->toecaps == 0);
4123 		sc->toecaps = 0;
4124 
4125 		param[0] = FW_PARAM_DEV(NTID);
4126 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, param, val);
4127 		if (rc != 0) {
4128 			device_printf(sc->dev,
4129 			    "failed to query HASHFILTER parameters: %d.\n", rc);
4130 			return (rc);
4131 		}
4132 		sc->tids.ntids = val[0];
4133 		if (sc->params.fw_vers <
4134 		    (V_FW_HDR_FW_VER_MAJOR(1) | V_FW_HDR_FW_VER_MINOR(20) |
4135 		    V_FW_HDR_FW_VER_MICRO(5) | V_FW_HDR_FW_VER_BUILD(0))) {
4136 			MPASS(sc->tids.ntids >= sc->tids.nhpftids);
4137 			sc->tids.ntids -= sc->tids.nhpftids;
4138 		}
4139 		sc->tids.natids = min(sc->tids.ntids / 2, MAX_ATIDS);
4140 		sc->params.hash_filter = 1;
4141 	}
4142 	if (sc->niccaps & FW_CAPS_CONFIG_NIC_ETHOFLD) {
4143 		param[0] = FW_PARAM_PFVF(ETHOFLD_START);
4144 		param[1] = FW_PARAM_PFVF(ETHOFLD_END);
4145 		param[2] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
4146 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 3, param, val);
4147 		if (rc != 0) {
4148 			device_printf(sc->dev,
4149 			    "failed to query NIC parameters: %d.\n", rc);
4150 			return (rc);
4151 		}
4152 		if ((int)val[1] > (int)val[0]) {
4153 			sc->tids.etid_base = val[0];
4154 			sc->tids.etid_end = val[1];
4155 			sc->tids.netids = val[1] - val[0] + 1;
4156 			sc->params.eo_wr_cred = val[2];
4157 			sc->params.ethoffload = 1;
4158 		}
4159 	}
4160 	if (sc->toecaps) {
4161 		/* query offload-related parameters */
4162 		param[0] = FW_PARAM_DEV(NTID);
4163 		param[1] = FW_PARAM_PFVF(SERVER_START);
4164 		param[2] = FW_PARAM_PFVF(SERVER_END);
4165 		param[3] = FW_PARAM_PFVF(TDDP_START);
4166 		param[4] = FW_PARAM_PFVF(TDDP_END);
4167 		param[5] = FW_PARAM_DEV(FLOWC_BUFFIFO_SZ);
4168 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
4169 		if (rc != 0) {
4170 			device_printf(sc->dev,
4171 			    "failed to query TOE parameters: %d.\n", rc);
4172 			return (rc);
4173 		}
4174 		sc->tids.ntids = val[0];
4175 		if (sc->params.fw_vers <
4176 		    (V_FW_HDR_FW_VER_MAJOR(1) | V_FW_HDR_FW_VER_MINOR(20) |
4177 		    V_FW_HDR_FW_VER_MICRO(5) | V_FW_HDR_FW_VER_BUILD(0))) {
4178 			MPASS(sc->tids.ntids >= sc->tids.nhpftids);
4179 			sc->tids.ntids -= sc->tids.nhpftids;
4180 		}
4181 		sc->tids.natids = min(sc->tids.ntids / 2, MAX_ATIDS);
4182 		if ((int)val[2] > (int)val[1]) {
4183 			sc->tids.stid_base = val[1];
4184 			sc->tids.nstids = val[2] - val[1] + 1;
4185 		}
4186 		sc->vres.ddp.start = val[3];
4187 		sc->vres.ddp.size = val[4] - val[3] + 1;
4188 		sc->params.ofldq_wr_cred = val[5];
4189 		sc->params.offload = 1;
4190 	} else {
4191 		/*
4192 		 * The firmware attempts memfree TOE configuration for -SO cards
4193 		 * and will report toecaps=0 if it runs out of resources (this
4194 		 * depends on the config file).  It may not report 0 for other
4195 		 * capabilities dependent on the TOE in this case.  Set them to
4196 		 * 0 here so that the driver doesn't bother tracking resources
4197 		 * that will never be used.
4198 		 */
4199 		sc->iscsicaps = 0;
4200 		sc->rdmacaps = 0;
4201 	}
4202 	if (sc->rdmacaps) {
4203 		param[0] = FW_PARAM_PFVF(STAG_START);
4204 		param[1] = FW_PARAM_PFVF(STAG_END);
4205 		param[2] = FW_PARAM_PFVF(RQ_START);
4206 		param[3] = FW_PARAM_PFVF(RQ_END);
4207 		param[4] = FW_PARAM_PFVF(PBL_START);
4208 		param[5] = FW_PARAM_PFVF(PBL_END);
4209 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
4210 		if (rc != 0) {
4211 			device_printf(sc->dev,
4212 			    "failed to query RDMA parameters(1): %d.\n", rc);
4213 			return (rc);
4214 		}
4215 		sc->vres.stag.start = val[0];
4216 		sc->vres.stag.size = val[1] - val[0] + 1;
4217 		sc->vres.rq.start = val[2];
4218 		sc->vres.rq.size = val[3] - val[2] + 1;
4219 		sc->vres.pbl.start = val[4];
4220 		sc->vres.pbl.size = val[5] - val[4] + 1;
4221 
4222 		param[0] = FW_PARAM_PFVF(SQRQ_START);
4223 		param[1] = FW_PARAM_PFVF(SQRQ_END);
4224 		param[2] = FW_PARAM_PFVF(CQ_START);
4225 		param[3] = FW_PARAM_PFVF(CQ_END);
4226 		param[4] = FW_PARAM_PFVF(OCQ_START);
4227 		param[5] = FW_PARAM_PFVF(OCQ_END);
4228 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 6, param, val);
4229 		if (rc != 0) {
4230 			device_printf(sc->dev,
4231 			    "failed to query RDMA parameters(2): %d.\n", rc);
4232 			return (rc);
4233 		}
4234 		sc->vres.qp.start = val[0];
4235 		sc->vres.qp.size = val[1] - val[0] + 1;
4236 		sc->vres.cq.start = val[2];
4237 		sc->vres.cq.size = val[3] - val[2] + 1;
4238 		sc->vres.ocq.start = val[4];
4239 		sc->vres.ocq.size = val[5] - val[4] + 1;
4240 
4241 		param[0] = FW_PARAM_PFVF(SRQ_START);
4242 		param[1] = FW_PARAM_PFVF(SRQ_END);
4243 		param[2] = FW_PARAM_DEV(MAXORDIRD_QP);
4244 		param[3] = FW_PARAM_DEV(MAXIRD_ADAPTER);
4245 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 4, param, val);
4246 		if (rc != 0) {
4247 			device_printf(sc->dev,
4248 			    "failed to query RDMA parameters(3): %d.\n", rc);
4249 			return (rc);
4250 		}
4251 		sc->vres.srq.start = val[0];
4252 		sc->vres.srq.size = val[1] - val[0] + 1;
4253 		sc->params.max_ordird_qp = val[2];
4254 		sc->params.max_ird_adapter = val[3];
4255 	}
4256 	if (sc->iscsicaps) {
4257 		param[0] = FW_PARAM_PFVF(ISCSI_START);
4258 		param[1] = FW_PARAM_PFVF(ISCSI_END);
4259 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
4260 		if (rc != 0) {
4261 			device_printf(sc->dev,
4262 			    "failed to query iSCSI parameters: %d.\n", rc);
4263 			return (rc);
4264 		}
4265 		sc->vres.iscsi.start = val[0];
4266 		sc->vres.iscsi.size = val[1] - val[0] + 1;
4267 	}
4268 	if (sc->cryptocaps & FW_CAPS_CONFIG_TLSKEYS) {
4269 		param[0] = FW_PARAM_PFVF(TLS_START);
4270 		param[1] = FW_PARAM_PFVF(TLS_END);
4271 		rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 2, param, val);
4272 		if (rc != 0) {
4273 			device_printf(sc->dev,
4274 			    "failed to query TLS parameters: %d.\n", rc);
4275 			return (rc);
4276 		}
4277 		sc->vres.key.start = val[0];
4278 		sc->vres.key.size = val[1] - val[0] + 1;
4279 	}
4280 
4281 	t4_init_sge_params(sc);
4282 
4283 	/*
4284 	 * We've got the params we wanted to query via the firmware.  Now grab
4285 	 * some others directly from the chip.
4286 	 */
4287 	rc = t4_read_chip_settings(sc);
4288 
4289 	return (rc);
4290 }
4291 
4292 static int
4293 set_params__post_init(struct adapter *sc)
4294 {
4295 	uint32_t param, val;
4296 #ifdef TCP_OFFLOAD
4297 	int i, v, shift;
4298 #endif
4299 
4300 	/* ask for encapsulated CPLs */
4301 	param = FW_PARAM_PFVF(CPLFW4MSG_ENCAP);
4302 	val = 1;
4303 	(void)t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
4304 
4305 	/* Enable 32b port caps if the firmware supports it. */
4306 	param = FW_PARAM_PFVF(PORT_CAPS32);
4307 	val = 1;
4308 	if (t4_set_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val) == 0)
4309 		sc->params.port_caps32 = 1;
4310 
4311 	/* Let filter + maskhash steer to a part of the VI's RSS region. */
4312 	val = 1 << (G_MASKSIZE(t4_read_reg(sc, A_TP_RSS_CONFIG_TNL)) - 1);
4313 	t4_set_reg_field(sc, A_TP_RSS_CONFIG_TNL, V_MASKFILTER(M_MASKFILTER),
4314 	    V_MASKFILTER(val - 1));
4315 
4316 #ifdef TCP_OFFLOAD
4317 	/*
4318 	 * Override the TOE timers with user provided tunables.  This is not the
4319 	 * recommended way to change the timers (the firmware config file is) so
4320 	 * these tunables are not documented.
4321 	 *
4322 	 * All the timer tunables are in microseconds.
4323 	 */
4324 	if (t4_toe_keepalive_idle != 0) {
4325 		v = us_to_tcp_ticks(sc, t4_toe_keepalive_idle);
4326 		v &= M_KEEPALIVEIDLE;
4327 		t4_set_reg_field(sc, A_TP_KEEP_IDLE,
4328 		    V_KEEPALIVEIDLE(M_KEEPALIVEIDLE), V_KEEPALIVEIDLE(v));
4329 	}
4330 	if (t4_toe_keepalive_interval != 0) {
4331 		v = us_to_tcp_ticks(sc, t4_toe_keepalive_interval);
4332 		v &= M_KEEPALIVEINTVL;
4333 		t4_set_reg_field(sc, A_TP_KEEP_INTVL,
4334 		    V_KEEPALIVEINTVL(M_KEEPALIVEINTVL), V_KEEPALIVEINTVL(v));
4335 	}
4336 	if (t4_toe_keepalive_count != 0) {
4337 		v = t4_toe_keepalive_count & M_KEEPALIVEMAXR2;
4338 		t4_set_reg_field(sc, A_TP_SHIFT_CNT,
4339 		    V_KEEPALIVEMAXR1(M_KEEPALIVEMAXR1) |
4340 		    V_KEEPALIVEMAXR2(M_KEEPALIVEMAXR2),
4341 		    V_KEEPALIVEMAXR1(1) | V_KEEPALIVEMAXR2(v));
4342 	}
4343 	if (t4_toe_rexmt_min != 0) {
4344 		v = us_to_tcp_ticks(sc, t4_toe_rexmt_min);
4345 		v &= M_RXTMIN;
4346 		t4_set_reg_field(sc, A_TP_RXT_MIN,
4347 		    V_RXTMIN(M_RXTMIN), V_RXTMIN(v));
4348 	}
4349 	if (t4_toe_rexmt_max != 0) {
4350 		v = us_to_tcp_ticks(sc, t4_toe_rexmt_max);
4351 		v &= M_RXTMAX;
4352 		t4_set_reg_field(sc, A_TP_RXT_MAX,
4353 		    V_RXTMAX(M_RXTMAX), V_RXTMAX(v));
4354 	}
4355 	if (t4_toe_rexmt_count != 0) {
4356 		v = t4_toe_rexmt_count & M_RXTSHIFTMAXR2;
4357 		t4_set_reg_field(sc, A_TP_SHIFT_CNT,
4358 		    V_RXTSHIFTMAXR1(M_RXTSHIFTMAXR1) |
4359 		    V_RXTSHIFTMAXR2(M_RXTSHIFTMAXR2),
4360 		    V_RXTSHIFTMAXR1(1) | V_RXTSHIFTMAXR2(v));
4361 	}
4362 	for (i = 0; i < nitems(t4_toe_rexmt_backoff); i++) {
4363 		if (t4_toe_rexmt_backoff[i] != -1) {
4364 			v = t4_toe_rexmt_backoff[i] & M_TIMERBACKOFFINDEX0;
4365 			shift = (i & 3) << 3;
4366 			t4_set_reg_field(sc, A_TP_TCP_BACKOFF_REG0 + (i & ~3),
4367 			    M_TIMERBACKOFFINDEX0 << shift, v << shift);
4368 		}
4369 	}
4370 #endif
4371 	return (0);
4372 }
4373 
4374 #undef FW_PARAM_PFVF
4375 #undef FW_PARAM_DEV
4376 
4377 static void
4378 t4_set_desc(struct adapter *sc)
4379 {
4380 	char buf[128];
4381 	struct adapter_params *p = &sc->params;
4382 
4383 	snprintf(buf, sizeof(buf), "Chelsio %s", p->vpd.id);
4384 
4385 	device_set_desc_copy(sc->dev, buf);
4386 }
4387 
4388 static inline void
4389 ifmedia_add4(struct ifmedia *ifm, int m)
4390 {
4391 
4392 	ifmedia_add(ifm, m, 0, NULL);
4393 	ifmedia_add(ifm, m | IFM_ETH_TXPAUSE, 0, NULL);
4394 	ifmedia_add(ifm, m | IFM_ETH_RXPAUSE, 0, NULL);
4395 	ifmedia_add(ifm, m | IFM_ETH_TXPAUSE | IFM_ETH_RXPAUSE, 0, NULL);
4396 }
4397 
4398 /*
4399  * This is the selected media, which is not quite the same as the active media.
4400  * The media line in ifconfig is "media: Ethernet selected (active)" if selected
4401  * and active are not the same, and "media: Ethernet selected" otherwise.
4402  */
4403 static void
4404 set_current_media(struct port_info *pi)
4405 {
4406 	struct link_config *lc;
4407 	struct ifmedia *ifm;
4408 	int mword;
4409 	u_int speed;
4410 
4411 	PORT_LOCK_ASSERT_OWNED(pi);
4412 
4413 	/* Leave current media alone if it's already set to IFM_NONE. */
4414 	ifm = &pi->media;
4415 	if (ifm->ifm_cur != NULL &&
4416 	    IFM_SUBTYPE(ifm->ifm_cur->ifm_media) == IFM_NONE)
4417 		return;
4418 
4419 	lc = &pi->link_cfg;
4420 	if (lc->requested_aneg != AUTONEG_DISABLE &&
4421 	    lc->supported & FW_PORT_CAP32_ANEG) {
4422 		ifmedia_set(ifm, IFM_ETHER | IFM_AUTO);
4423 		return;
4424 	}
4425 	mword = IFM_ETHER | IFM_FDX;
4426 	if (lc->requested_fc & PAUSE_TX)
4427 		mword |= IFM_ETH_TXPAUSE;
4428 	if (lc->requested_fc & PAUSE_RX)
4429 		mword |= IFM_ETH_RXPAUSE;
4430 	if (lc->requested_speed == 0)
4431 		speed = port_top_speed(pi) * 1000;	/* Gbps -> Mbps */
4432 	else
4433 		speed = lc->requested_speed;
4434 	mword |= port_mword(pi, speed_to_fwcap(speed));
4435 	ifmedia_set(ifm, mword);
4436 }
4437 
4438 /*
4439  * Returns true if the ifmedia list for the port cannot change.
4440  */
4441 static bool
4442 fixed_ifmedia(struct port_info *pi)
4443 {
4444 
4445 	return (pi->port_type == FW_PORT_TYPE_BT_SGMII ||
4446 	    pi->port_type == FW_PORT_TYPE_BT_XFI ||
4447 	    pi->port_type == FW_PORT_TYPE_BT_XAUI ||
4448 	    pi->port_type == FW_PORT_TYPE_KX4 ||
4449 	    pi->port_type == FW_PORT_TYPE_KX ||
4450 	    pi->port_type == FW_PORT_TYPE_KR ||
4451 	    pi->port_type == FW_PORT_TYPE_BP_AP ||
4452 	    pi->port_type == FW_PORT_TYPE_BP4_AP ||
4453 	    pi->port_type == FW_PORT_TYPE_BP40_BA ||
4454 	    pi->port_type == FW_PORT_TYPE_KR4_100G ||
4455 	    pi->port_type == FW_PORT_TYPE_KR_SFP28 ||
4456 	    pi->port_type == FW_PORT_TYPE_KR_XLAUI);
4457 }
4458 
4459 static void
4460 build_medialist(struct port_info *pi)
4461 {
4462 	uint32_t ss, speed;
4463 	int unknown, mword, bit;
4464 	struct link_config *lc;
4465 	struct ifmedia *ifm;
4466 
4467 	PORT_LOCK_ASSERT_OWNED(pi);
4468 
4469 	if (pi->flags & FIXED_IFMEDIA)
4470 		return;
4471 
4472 	/*
4473 	 * Rebuild the ifmedia list.
4474 	 */
4475 	ifm = &pi->media;
4476 	ifmedia_removeall(ifm);
4477 	lc = &pi->link_cfg;
4478 	ss = G_FW_PORT_CAP32_SPEED(lc->supported); /* Supported Speeds */
4479 	if (__predict_false(ss == 0)) {	/* not supposed to happen. */
4480 		MPASS(ss != 0);
4481 no_media:
4482 		MPASS(LIST_EMPTY(&ifm->ifm_list));
4483 		ifmedia_add(ifm, IFM_ETHER | IFM_NONE, 0, NULL);
4484 		ifmedia_set(ifm, IFM_ETHER | IFM_NONE);
4485 		return;
4486 	}
4487 
4488 	unknown = 0;
4489 	for (bit = S_FW_PORT_CAP32_SPEED; bit < fls(ss); bit++) {
4490 		speed = 1 << bit;
4491 		MPASS(speed & M_FW_PORT_CAP32_SPEED);
4492 		if (ss & speed) {
4493 			mword = port_mword(pi, speed);
4494 			if (mword == IFM_NONE) {
4495 				goto no_media;
4496 			} else if (mword == IFM_UNKNOWN)
4497 				unknown++;
4498 			else
4499 				ifmedia_add4(ifm, IFM_ETHER | IFM_FDX | mword);
4500 		}
4501 	}
4502 	if (unknown > 0) /* Add one unknown for all unknown media types. */
4503 		ifmedia_add4(ifm, IFM_ETHER | IFM_FDX | IFM_UNKNOWN);
4504 	if (lc->supported & FW_PORT_CAP32_ANEG)
4505 		ifmedia_add(ifm, IFM_ETHER | IFM_AUTO, 0, NULL);
4506 
4507 	set_current_media(pi);
4508 }
4509 
4510 /*
4511  * Initialize the requested fields in the link config based on driver tunables.
4512  */
4513 static void
4514 init_link_config(struct port_info *pi)
4515 {
4516 	struct link_config *lc = &pi->link_cfg;
4517 
4518 	PORT_LOCK_ASSERT_OWNED(pi);
4519 
4520 	lc->requested_speed = 0;
4521 
4522 	if (t4_autoneg == 0)
4523 		lc->requested_aneg = AUTONEG_DISABLE;
4524 	else if (t4_autoneg == 1)
4525 		lc->requested_aneg = AUTONEG_ENABLE;
4526 	else
4527 		lc->requested_aneg = AUTONEG_AUTO;
4528 
4529 	lc->requested_fc = t4_pause_settings & (PAUSE_TX | PAUSE_RX |
4530 	    PAUSE_AUTONEG);
4531 
4532 	if (t4_fec == -1 || t4_fec & FEC_AUTO)
4533 		lc->requested_fec = FEC_AUTO;
4534 	else {
4535 		lc->requested_fec = FEC_NONE;
4536 		if (t4_fec & FEC_RS)
4537 			lc->requested_fec |= FEC_RS;
4538 		if (t4_fec & FEC_BASER_RS)
4539 			lc->requested_fec |= FEC_BASER_RS;
4540 	}
4541 }
4542 
4543 /*
4544  * Makes sure that all requested settings comply with what's supported by the
4545  * port.  Returns the number of settings that were invalid and had to be fixed.
4546  */
4547 static int
4548 fixup_link_config(struct port_info *pi)
4549 {
4550 	int n = 0;
4551 	struct link_config *lc = &pi->link_cfg;
4552 	uint32_t fwspeed;
4553 
4554 	PORT_LOCK_ASSERT_OWNED(pi);
4555 
4556 	/* Speed (when not autonegotiating) */
4557 	if (lc->requested_speed != 0) {
4558 		fwspeed = speed_to_fwcap(lc->requested_speed);
4559 		if ((fwspeed & lc->supported) == 0) {
4560 			n++;
4561 			lc->requested_speed = 0;
4562 		}
4563 	}
4564 
4565 	/* Link autonegotiation */
4566 	MPASS(lc->requested_aneg == AUTONEG_ENABLE ||
4567 	    lc->requested_aneg == AUTONEG_DISABLE ||
4568 	    lc->requested_aneg == AUTONEG_AUTO);
4569 	if (lc->requested_aneg == AUTONEG_ENABLE &&
4570 	    !(lc->supported & FW_PORT_CAP32_ANEG)) {
4571 		n++;
4572 		lc->requested_aneg = AUTONEG_AUTO;
4573 	}
4574 
4575 	/* Flow control */
4576 	MPASS((lc->requested_fc & ~(PAUSE_TX | PAUSE_RX | PAUSE_AUTONEG)) == 0);
4577 	if (lc->requested_fc & PAUSE_TX &&
4578 	    !(lc->supported & FW_PORT_CAP32_FC_TX)) {
4579 		n++;
4580 		lc->requested_fc &= ~PAUSE_TX;
4581 	}
4582 	if (lc->requested_fc & PAUSE_RX &&
4583 	    !(lc->supported & FW_PORT_CAP32_FC_RX)) {
4584 		n++;
4585 		lc->requested_fc &= ~PAUSE_RX;
4586 	}
4587 	if (!(lc->requested_fc & PAUSE_AUTONEG) &&
4588 	    !(lc->supported & FW_PORT_CAP32_FORCE_PAUSE)) {
4589 		n++;
4590 		lc->requested_fc |= PAUSE_AUTONEG;
4591 	}
4592 
4593 	/* FEC */
4594 	if ((lc->requested_fec & FEC_RS &&
4595 	    !(lc->supported & FW_PORT_CAP32_FEC_RS)) ||
4596 	    (lc->requested_fec & FEC_BASER_RS &&
4597 	    !(lc->supported & FW_PORT_CAP32_FEC_BASER_RS))) {
4598 		n++;
4599 		lc->requested_fec = FEC_AUTO;
4600 	}
4601 
4602 	return (n);
4603 }
4604 
4605 /*
4606  * Apply the requested L1 settings, which are expected to be valid, to the
4607  * hardware.
4608  */
4609 static int
4610 apply_link_config(struct port_info *pi)
4611 {
4612 	struct adapter *sc = pi->adapter;
4613 	struct link_config *lc = &pi->link_cfg;
4614 	int rc;
4615 
4616 #ifdef INVARIANTS
4617 	ASSERT_SYNCHRONIZED_OP(sc);
4618 	PORT_LOCK_ASSERT_OWNED(pi);
4619 
4620 	if (lc->requested_aneg == AUTONEG_ENABLE)
4621 		MPASS(lc->supported & FW_PORT_CAP32_ANEG);
4622 	if (!(lc->requested_fc & PAUSE_AUTONEG))
4623 		MPASS(lc->supported & FW_PORT_CAP32_FORCE_PAUSE);
4624 	if (lc->requested_fc & PAUSE_TX)
4625 		MPASS(lc->supported & FW_PORT_CAP32_FC_TX);
4626 	if (lc->requested_fc & PAUSE_RX)
4627 		MPASS(lc->supported & FW_PORT_CAP32_FC_RX);
4628 	if (lc->requested_fec & FEC_RS)
4629 		MPASS(lc->supported & FW_PORT_CAP32_FEC_RS);
4630 	if (lc->requested_fec & FEC_BASER_RS)
4631 		MPASS(lc->supported & FW_PORT_CAP32_FEC_BASER_RS);
4632 #endif
4633 	rc = -t4_link_l1cfg(sc, sc->mbox, pi->tx_chan, lc);
4634 	if (rc != 0) {
4635 		/* Don't complain if the VF driver gets back an EPERM. */
4636 		if (!(sc->flags & IS_VF) || rc != FW_EPERM)
4637 			device_printf(pi->dev, "l1cfg failed: %d\n", rc);
4638 	} else {
4639 		/*
4640 		 * An L1_CFG will almost always result in a link-change event if
4641 		 * the link is up, and the driver will refresh the actual
4642 		 * fec/fc/etc. when the notification is processed.  If the link
4643 		 * is down then the actual settings are meaningless.
4644 		 *
4645 		 * This takes care of the case where a change in the L1 settings
4646 		 * may not result in a notification.
4647 		 */
4648 		if (lc->link_ok && !(lc->requested_fc & PAUSE_AUTONEG))
4649 			lc->fc = lc->requested_fc & (PAUSE_TX | PAUSE_RX);
4650 	}
4651 	return (rc);
4652 }
4653 
4654 #define FW_MAC_EXACT_CHUNK	7
4655 
4656 /*
4657  * Program the port's XGMAC based on parameters in ifnet.  The caller also
4658  * indicates which parameters should be programmed (the rest are left alone).
4659  */
4660 int
4661 update_mac_settings(struct ifnet *ifp, int flags)
4662 {
4663 	int rc = 0;
4664 	struct vi_info *vi = ifp->if_softc;
4665 	struct port_info *pi = vi->pi;
4666 	struct adapter *sc = pi->adapter;
4667 	int mtu = -1, promisc = -1, allmulti = -1, vlanex = -1;
4668 
4669 	ASSERT_SYNCHRONIZED_OP(sc);
4670 	KASSERT(flags, ("%s: not told what to update.", __func__));
4671 
4672 	if (flags & XGMAC_MTU)
4673 		mtu = ifp->if_mtu;
4674 
4675 	if (flags & XGMAC_PROMISC)
4676 		promisc = ifp->if_flags & IFF_PROMISC ? 1 : 0;
4677 
4678 	if (flags & XGMAC_ALLMULTI)
4679 		allmulti = ifp->if_flags & IFF_ALLMULTI ? 1 : 0;
4680 
4681 	if (flags & XGMAC_VLANEX)
4682 		vlanex = ifp->if_capenable & IFCAP_VLAN_HWTAGGING ? 1 : 0;
4683 
4684 	if (flags & (XGMAC_MTU|XGMAC_PROMISC|XGMAC_ALLMULTI|XGMAC_VLANEX)) {
4685 		rc = -t4_set_rxmode(sc, sc->mbox, vi->viid, mtu, promisc,
4686 		    allmulti, 1, vlanex, false);
4687 		if (rc) {
4688 			if_printf(ifp, "set_rxmode (%x) failed: %d\n", flags,
4689 			    rc);
4690 			return (rc);
4691 		}
4692 	}
4693 
4694 	if (flags & XGMAC_UCADDR) {
4695 		uint8_t ucaddr[ETHER_ADDR_LEN];
4696 
4697 		bcopy(IF_LLADDR(ifp), ucaddr, sizeof(ucaddr));
4698 		rc = t4_change_mac(sc, sc->mbox, vi->viid, vi->xact_addr_filt,
4699 		    ucaddr, true, true);
4700 		if (rc < 0) {
4701 			rc = -rc;
4702 			if_printf(ifp, "change_mac failed: %d\n", rc);
4703 			return (rc);
4704 		} else {
4705 			vi->xact_addr_filt = rc;
4706 			rc = 0;
4707 		}
4708 	}
4709 
4710 	if (flags & XGMAC_MCADDRS) {
4711 		const uint8_t *mcaddr[FW_MAC_EXACT_CHUNK];
4712 		int del = 1;
4713 		uint64_t hash = 0;
4714 		struct ifmultiaddr *ifma;
4715 		int i = 0, j;
4716 
4717 		if_maddr_rlock(ifp);
4718 		CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
4719 			if (ifma->ifma_addr->sa_family != AF_LINK)
4720 				continue;
4721 			mcaddr[i] =
4722 			    LLADDR((struct sockaddr_dl *)ifma->ifma_addr);
4723 			MPASS(ETHER_IS_MULTICAST(mcaddr[i]));
4724 			i++;
4725 
4726 			if (i == FW_MAC_EXACT_CHUNK) {
4727 				rc = t4_alloc_mac_filt(sc, sc->mbox, vi->viid,
4728 				    del, i, mcaddr, NULL, &hash, 0);
4729 				if (rc < 0) {
4730 					rc = -rc;
4731 					for (j = 0; j < i; j++) {
4732 						if_printf(ifp,
4733 						    "failed to add mc address"
4734 						    " %02x:%02x:%02x:"
4735 						    "%02x:%02x:%02x rc=%d\n",
4736 						    mcaddr[j][0], mcaddr[j][1],
4737 						    mcaddr[j][2], mcaddr[j][3],
4738 						    mcaddr[j][4], mcaddr[j][5],
4739 						    rc);
4740 					}
4741 					goto mcfail;
4742 				}
4743 				del = 0;
4744 				i = 0;
4745 			}
4746 		}
4747 		if (i > 0) {
4748 			rc = t4_alloc_mac_filt(sc, sc->mbox, vi->viid, del, i,
4749 			    mcaddr, NULL, &hash, 0);
4750 			if (rc < 0) {
4751 				rc = -rc;
4752 				for (j = 0; j < i; j++) {
4753 					if_printf(ifp,
4754 					    "failed to add mc address"
4755 					    " %02x:%02x:%02x:"
4756 					    "%02x:%02x:%02x rc=%d\n",
4757 					    mcaddr[j][0], mcaddr[j][1],
4758 					    mcaddr[j][2], mcaddr[j][3],
4759 					    mcaddr[j][4], mcaddr[j][5],
4760 					    rc);
4761 				}
4762 				goto mcfail;
4763 			}
4764 		}
4765 
4766 		rc = -t4_set_addr_hash(sc, sc->mbox, vi->viid, 0, hash, 0);
4767 		if (rc != 0)
4768 			if_printf(ifp, "failed to set mc address hash: %d", rc);
4769 mcfail:
4770 		if_maddr_runlock(ifp);
4771 	}
4772 
4773 	return (rc);
4774 }
4775 
4776 /*
4777  * {begin|end}_synchronized_op must be called from the same thread.
4778  */
4779 int
4780 begin_synchronized_op(struct adapter *sc, struct vi_info *vi, int flags,
4781     char *wmesg)
4782 {
4783 	int rc, pri;
4784 
4785 #ifdef WITNESS
4786 	/* the caller thinks it's ok to sleep, but is it really? */
4787 	if (flags & SLEEP_OK)
4788 		WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
4789 		    "begin_synchronized_op");
4790 #endif
4791 
4792 	if (INTR_OK)
4793 		pri = PCATCH;
4794 	else
4795 		pri = 0;
4796 
4797 	ADAPTER_LOCK(sc);
4798 	for (;;) {
4799 
4800 		if (vi && IS_DOOMED(vi)) {
4801 			rc = ENXIO;
4802 			goto done;
4803 		}
4804 
4805 		if (!IS_BUSY(sc)) {
4806 			rc = 0;
4807 			break;
4808 		}
4809 
4810 		if (!(flags & SLEEP_OK)) {
4811 			rc = EBUSY;
4812 			goto done;
4813 		}
4814 
4815 		if (mtx_sleep(&sc->flags, &sc->sc_lock, pri, wmesg, 0)) {
4816 			rc = EINTR;
4817 			goto done;
4818 		}
4819 	}
4820 
4821 	KASSERT(!IS_BUSY(sc), ("%s: controller busy.", __func__));
4822 	SET_BUSY(sc);
4823 #ifdef INVARIANTS
4824 	sc->last_op = wmesg;
4825 	sc->last_op_thr = curthread;
4826 	sc->last_op_flags = flags;
4827 #endif
4828 
4829 done:
4830 	if (!(flags & HOLD_LOCK) || rc)
4831 		ADAPTER_UNLOCK(sc);
4832 
4833 	return (rc);
4834 }
4835 
4836 /*
4837  * Tell if_ioctl and if_init that the VI is going away.  This is
4838  * special variant of begin_synchronized_op and must be paired with a
4839  * call to end_synchronized_op.
4840  */
4841 void
4842 doom_vi(struct adapter *sc, struct vi_info *vi)
4843 {
4844 
4845 	ADAPTER_LOCK(sc);
4846 	SET_DOOMED(vi);
4847 	wakeup(&sc->flags);
4848 	while (IS_BUSY(sc))
4849 		mtx_sleep(&sc->flags, &sc->sc_lock, 0, "t4detach", 0);
4850 	SET_BUSY(sc);
4851 #ifdef INVARIANTS
4852 	sc->last_op = "t4detach";
4853 	sc->last_op_thr = curthread;
4854 	sc->last_op_flags = 0;
4855 #endif
4856 	ADAPTER_UNLOCK(sc);
4857 }
4858 
4859 /*
4860  * {begin|end}_synchronized_op must be called from the same thread.
4861  */
4862 void
4863 end_synchronized_op(struct adapter *sc, int flags)
4864 {
4865 
4866 	if (flags & LOCK_HELD)
4867 		ADAPTER_LOCK_ASSERT_OWNED(sc);
4868 	else
4869 		ADAPTER_LOCK(sc);
4870 
4871 	KASSERT(IS_BUSY(sc), ("%s: controller not busy.", __func__));
4872 	CLR_BUSY(sc);
4873 	wakeup(&sc->flags);
4874 	ADAPTER_UNLOCK(sc);
4875 }
4876 
4877 static int
4878 cxgbe_init_synchronized(struct vi_info *vi)
4879 {
4880 	struct port_info *pi = vi->pi;
4881 	struct adapter *sc = pi->adapter;
4882 	struct ifnet *ifp = vi->ifp;
4883 	int rc = 0, i;
4884 	struct sge_txq *txq;
4885 
4886 	ASSERT_SYNCHRONIZED_OP(sc);
4887 
4888 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
4889 		return (0);	/* already running */
4890 
4891 	if (!(sc->flags & FULL_INIT_DONE) &&
4892 	    ((rc = adapter_full_init(sc)) != 0))
4893 		return (rc);	/* error message displayed already */
4894 
4895 	if (!(vi->flags & VI_INIT_DONE) &&
4896 	    ((rc = vi_full_init(vi)) != 0))
4897 		return (rc); /* error message displayed already */
4898 
4899 	rc = update_mac_settings(ifp, XGMAC_ALL);
4900 	if (rc)
4901 		goto done;	/* error message displayed already */
4902 
4903 	PORT_LOCK(pi);
4904 	if (pi->up_vis == 0) {
4905 		t4_update_port_info(pi);
4906 		fixup_link_config(pi);
4907 		build_medialist(pi);
4908 		apply_link_config(pi);
4909 	}
4910 
4911 	rc = -t4_enable_vi(sc, sc->mbox, vi->viid, true, true);
4912 	if (rc != 0) {
4913 		if_printf(ifp, "enable_vi failed: %d\n", rc);
4914 		PORT_UNLOCK(pi);
4915 		goto done;
4916 	}
4917 
4918 	/*
4919 	 * Can't fail from this point onwards.  Review cxgbe_uninit_synchronized
4920 	 * if this changes.
4921 	 */
4922 
4923 	for_each_txq(vi, i, txq) {
4924 		TXQ_LOCK(txq);
4925 		txq->eq.flags |= EQ_ENABLED;
4926 		TXQ_UNLOCK(txq);
4927 	}
4928 
4929 	/*
4930 	 * The first iq of the first port to come up is used for tracing.
4931 	 */
4932 	if (sc->traceq < 0 && IS_MAIN_VI(vi)) {
4933 		sc->traceq = sc->sge.rxq[vi->first_rxq].iq.abs_id;
4934 		t4_write_reg(sc, is_t4(sc) ?  A_MPS_TRC_RSS_CONTROL :
4935 		    A_MPS_T5_TRC_RSS_CONTROL, V_RSSCONTROL(pi->tx_chan) |
4936 		    V_QUEUENUMBER(sc->traceq));
4937 		pi->flags |= HAS_TRACEQ;
4938 	}
4939 
4940 	/* all ok */
4941 	pi->up_vis++;
4942 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
4943 
4944 	if (pi->nvi > 1 || sc->flags & IS_VF)
4945 		callout_reset(&vi->tick, hz, vi_tick, vi);
4946 	else
4947 		callout_reset(&pi->tick, hz, cxgbe_tick, pi);
4948 	PORT_UNLOCK(pi);
4949 done:
4950 	if (rc != 0)
4951 		cxgbe_uninit_synchronized(vi);
4952 
4953 	return (rc);
4954 }
4955 
4956 /*
4957  * Idempotent.
4958  */
4959 static int
4960 cxgbe_uninit_synchronized(struct vi_info *vi)
4961 {
4962 	struct port_info *pi = vi->pi;
4963 	struct adapter *sc = pi->adapter;
4964 	struct ifnet *ifp = vi->ifp;
4965 	int rc, i;
4966 	struct sge_txq *txq;
4967 
4968 	ASSERT_SYNCHRONIZED_OP(sc);
4969 
4970 	if (!(vi->flags & VI_INIT_DONE)) {
4971 		if (__predict_false(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
4972 			KASSERT(0, ("uninited VI is running"));
4973 			if_printf(ifp, "uninited VI with running ifnet.  "
4974 			    "vi->flags 0x%016lx, if_flags 0x%08x, "
4975 			    "if_drv_flags 0x%08x\n", vi->flags, ifp->if_flags,
4976 			    ifp->if_drv_flags);
4977 		}
4978 		return (0);
4979 	}
4980 
4981 	/*
4982 	 * Disable the VI so that all its data in either direction is discarded
4983 	 * by the MPS.  Leave everything else (the queues, interrupts, and 1Hz
4984 	 * tick) intact as the TP can deliver negative advice or data that it's
4985 	 * holding in its RAM (for an offloaded connection) even after the VI is
4986 	 * disabled.
4987 	 */
4988 	rc = -t4_enable_vi(sc, sc->mbox, vi->viid, false, false);
4989 	if (rc) {
4990 		if_printf(ifp, "disable_vi failed: %d\n", rc);
4991 		return (rc);
4992 	}
4993 
4994 	for_each_txq(vi, i, txq) {
4995 		TXQ_LOCK(txq);
4996 		txq->eq.flags &= ~EQ_ENABLED;
4997 		TXQ_UNLOCK(txq);
4998 	}
4999 
5000 	PORT_LOCK(pi);
5001 	if (pi->nvi > 1 || sc->flags & IS_VF)
5002 		callout_stop(&vi->tick);
5003 	else
5004 		callout_stop(&pi->tick);
5005 	if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
5006 		PORT_UNLOCK(pi);
5007 		return (0);
5008 	}
5009 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
5010 	pi->up_vis--;
5011 	if (pi->up_vis > 0) {
5012 		PORT_UNLOCK(pi);
5013 		return (0);
5014 	}
5015 
5016 	pi->link_cfg.link_ok = false;
5017 	pi->link_cfg.speed = 0;
5018 	pi->link_cfg.link_down_rc = 255;
5019 	t4_os_link_changed(pi);
5020 	PORT_UNLOCK(pi);
5021 
5022 	return (0);
5023 }
5024 
5025 /*
5026  * It is ok for this function to fail midway and return right away.  t4_detach
5027  * will walk the entire sc->irq list and clean up whatever is valid.
5028  */
5029 int
5030 t4_setup_intr_handlers(struct adapter *sc)
5031 {
5032 	int rc, rid, p, q, v;
5033 	char s[8];
5034 	struct irq *irq;
5035 	struct port_info *pi;
5036 	struct vi_info *vi;
5037 	struct sge *sge = &sc->sge;
5038 	struct sge_rxq *rxq;
5039 #ifdef TCP_OFFLOAD
5040 	struct sge_ofld_rxq *ofld_rxq;
5041 #endif
5042 #ifdef DEV_NETMAP
5043 	struct sge_nm_rxq *nm_rxq;
5044 #endif
5045 #ifdef RSS
5046 	int nbuckets = rss_getnumbuckets();
5047 #endif
5048 
5049 	/*
5050 	 * Setup interrupts.
5051 	 */
5052 	irq = &sc->irq[0];
5053 	rid = sc->intr_type == INTR_INTX ? 0 : 1;
5054 	if (forwarding_intr_to_fwq(sc))
5055 		return (t4_alloc_irq(sc, irq, rid, t4_intr_all, sc, "all"));
5056 
5057 	/* Multiple interrupts. */
5058 	if (sc->flags & IS_VF)
5059 		KASSERT(sc->intr_count >= T4VF_EXTRA_INTR + sc->params.nports,
5060 		    ("%s: too few intr.", __func__));
5061 	else
5062 		KASSERT(sc->intr_count >= T4_EXTRA_INTR + sc->params.nports,
5063 		    ("%s: too few intr.", __func__));
5064 
5065 	/* The first one is always error intr on PFs */
5066 	if (!(sc->flags & IS_VF)) {
5067 		rc = t4_alloc_irq(sc, irq, rid, t4_intr_err, sc, "err");
5068 		if (rc != 0)
5069 			return (rc);
5070 		irq++;
5071 		rid++;
5072 	}
5073 
5074 	/* The second one is always the firmware event queue (first on VFs) */
5075 	rc = t4_alloc_irq(sc, irq, rid, t4_intr_evt, &sge->fwq, "evt");
5076 	if (rc != 0)
5077 		return (rc);
5078 	irq++;
5079 	rid++;
5080 
5081 	for_each_port(sc, p) {
5082 		pi = sc->port[p];
5083 		for_each_vi(pi, v, vi) {
5084 			vi->first_intr = rid - 1;
5085 
5086 			if (vi->nnmrxq > 0) {
5087 				int n = max(vi->nrxq, vi->nnmrxq);
5088 
5089 				rxq = &sge->rxq[vi->first_rxq];
5090 #ifdef DEV_NETMAP
5091 				nm_rxq = &sge->nm_rxq[vi->first_nm_rxq];
5092 #endif
5093 				for (q = 0; q < n; q++) {
5094 					snprintf(s, sizeof(s), "%x%c%x", p,
5095 					    'a' + v, q);
5096 					if (q < vi->nrxq)
5097 						irq->rxq = rxq++;
5098 #ifdef DEV_NETMAP
5099 					if (q < vi->nnmrxq)
5100 						irq->nm_rxq = nm_rxq++;
5101 
5102 					if (irq->nm_rxq != NULL &&
5103 					    irq->rxq == NULL) {
5104 						/* Netmap rx only */
5105 						rc = t4_alloc_irq(sc, irq, rid,
5106 						    t4_nm_intr, irq->nm_rxq, s);
5107 					}
5108 					if (irq->nm_rxq != NULL &&
5109 					    irq->rxq != NULL) {
5110 						/* NIC and Netmap rx */
5111 						rc = t4_alloc_irq(sc, irq, rid,
5112 						    t4_vi_intr, irq, s);
5113 					}
5114 #endif
5115 					if (irq->rxq != NULL &&
5116 					    irq->nm_rxq == NULL) {
5117 						/* NIC rx only */
5118 						rc = t4_alloc_irq(sc, irq, rid,
5119 						    t4_intr, irq->rxq, s);
5120 					}
5121 					if (rc != 0)
5122 						return (rc);
5123 #ifdef RSS
5124 					if (q < vi->nrxq) {
5125 						bus_bind_intr(sc->dev, irq->res,
5126 						    rss_getcpu(q % nbuckets));
5127 					}
5128 #endif
5129 					irq++;
5130 					rid++;
5131 					vi->nintr++;
5132 				}
5133 			} else {
5134 				for_each_rxq(vi, q, rxq) {
5135 					snprintf(s, sizeof(s), "%x%c%x", p,
5136 					    'a' + v, q);
5137 					rc = t4_alloc_irq(sc, irq, rid,
5138 					    t4_intr, rxq, s);
5139 					if (rc != 0)
5140 						return (rc);
5141 #ifdef RSS
5142 					bus_bind_intr(sc->dev, irq->res,
5143 					    rss_getcpu(q % nbuckets));
5144 #endif
5145 					irq++;
5146 					rid++;
5147 					vi->nintr++;
5148 				}
5149 			}
5150 #ifdef TCP_OFFLOAD
5151 			for_each_ofld_rxq(vi, q, ofld_rxq) {
5152 				snprintf(s, sizeof(s), "%x%c%x", p, 'A' + v, q);
5153 				rc = t4_alloc_irq(sc, irq, rid, t4_intr,
5154 				    ofld_rxq, s);
5155 				if (rc != 0)
5156 					return (rc);
5157 				irq++;
5158 				rid++;
5159 				vi->nintr++;
5160 			}
5161 #endif
5162 		}
5163 	}
5164 	MPASS(irq == &sc->irq[sc->intr_count]);
5165 
5166 	return (0);
5167 }
5168 
5169 int
5170 adapter_full_init(struct adapter *sc)
5171 {
5172 	int rc, i;
5173 #ifdef RSS
5174 	uint32_t raw_rss_key[RSS_KEYSIZE / sizeof(uint32_t)];
5175 	uint32_t rss_key[RSS_KEYSIZE / sizeof(uint32_t)];
5176 #endif
5177 
5178 	ASSERT_SYNCHRONIZED_OP(sc);
5179 	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
5180 	KASSERT((sc->flags & FULL_INIT_DONE) == 0,
5181 	    ("%s: FULL_INIT_DONE already", __func__));
5182 
5183 	/*
5184 	 * queues that belong to the adapter (not any particular port).
5185 	 */
5186 	rc = t4_setup_adapter_queues(sc);
5187 	if (rc != 0)
5188 		goto done;
5189 
5190 	for (i = 0; i < nitems(sc->tq); i++) {
5191 		sc->tq[i] = taskqueue_create("t4 taskq", M_NOWAIT,
5192 		    taskqueue_thread_enqueue, &sc->tq[i]);
5193 		if (sc->tq[i] == NULL) {
5194 			device_printf(sc->dev,
5195 			    "failed to allocate task queue %d\n", i);
5196 			rc = ENOMEM;
5197 			goto done;
5198 		}
5199 		taskqueue_start_threads(&sc->tq[i], 1, PI_NET, "%s tq%d",
5200 		    device_get_nameunit(sc->dev), i);
5201 	}
5202 #ifdef RSS
5203 	MPASS(RSS_KEYSIZE == 40);
5204 	rss_getkey((void *)&raw_rss_key[0]);
5205 	for (i = 0; i < nitems(rss_key); i++) {
5206 		rss_key[i] = htobe32(raw_rss_key[nitems(rss_key) - 1 - i]);
5207 	}
5208 	t4_write_rss_key(sc, &rss_key[0], -1, 1);
5209 #endif
5210 
5211 	if (!(sc->flags & IS_VF))
5212 		t4_intr_enable(sc);
5213 	sc->flags |= FULL_INIT_DONE;
5214 done:
5215 	if (rc != 0)
5216 		adapter_full_uninit(sc);
5217 
5218 	return (rc);
5219 }
5220 
5221 int
5222 adapter_full_uninit(struct adapter *sc)
5223 {
5224 	int i;
5225 
5226 	ADAPTER_LOCK_ASSERT_NOTOWNED(sc);
5227 
5228 	t4_teardown_adapter_queues(sc);
5229 
5230 	for (i = 0; i < nitems(sc->tq) && sc->tq[i]; i++) {
5231 		taskqueue_free(sc->tq[i]);
5232 		sc->tq[i] = NULL;
5233 	}
5234 
5235 	sc->flags &= ~FULL_INIT_DONE;
5236 
5237 	return (0);
5238 }
5239 
5240 #ifdef RSS
5241 #define SUPPORTED_RSS_HASHTYPES (RSS_HASHTYPE_RSS_IPV4 | \
5242     RSS_HASHTYPE_RSS_TCP_IPV4 | RSS_HASHTYPE_RSS_IPV6 | \
5243     RSS_HASHTYPE_RSS_TCP_IPV6 | RSS_HASHTYPE_RSS_UDP_IPV4 | \
5244     RSS_HASHTYPE_RSS_UDP_IPV6)
5245 
5246 /* Translates kernel hash types to hardware. */
5247 static int
5248 hashconfig_to_hashen(int hashconfig)
5249 {
5250 	int hashen = 0;
5251 
5252 	if (hashconfig & RSS_HASHTYPE_RSS_IPV4)
5253 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN;
5254 	if (hashconfig & RSS_HASHTYPE_RSS_IPV6)
5255 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN;
5256 	if (hashconfig & RSS_HASHTYPE_RSS_UDP_IPV4) {
5257 		hashen |= F_FW_RSS_VI_CONFIG_CMD_UDPEN |
5258 		    F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
5259 	}
5260 	if (hashconfig & RSS_HASHTYPE_RSS_UDP_IPV6) {
5261 		hashen |= F_FW_RSS_VI_CONFIG_CMD_UDPEN |
5262 		    F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
5263 	}
5264 	if (hashconfig & RSS_HASHTYPE_RSS_TCP_IPV4)
5265 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN;
5266 	if (hashconfig & RSS_HASHTYPE_RSS_TCP_IPV6)
5267 		hashen |= F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN;
5268 
5269 	return (hashen);
5270 }
5271 
5272 /* Translates hardware hash types to kernel. */
5273 static int
5274 hashen_to_hashconfig(int hashen)
5275 {
5276 	int hashconfig = 0;
5277 
5278 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_UDPEN) {
5279 		/*
5280 		 * If UDP hashing was enabled it must have been enabled for
5281 		 * either IPv4 or IPv6 (inclusive or).  Enabling UDP without
5282 		 * enabling any 4-tuple hash is nonsense configuration.
5283 		 */
5284 		MPASS(hashen & (F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
5285 		    F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN));
5286 
5287 		if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
5288 			hashconfig |= RSS_HASHTYPE_RSS_UDP_IPV4;
5289 		if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
5290 			hashconfig |= RSS_HASHTYPE_RSS_UDP_IPV6;
5291 	}
5292 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN)
5293 		hashconfig |= RSS_HASHTYPE_RSS_TCP_IPV4;
5294 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN)
5295 		hashconfig |= RSS_HASHTYPE_RSS_TCP_IPV6;
5296 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN)
5297 		hashconfig |= RSS_HASHTYPE_RSS_IPV4;
5298 	if (hashen & F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN)
5299 		hashconfig |= RSS_HASHTYPE_RSS_IPV6;
5300 
5301 	return (hashconfig);
5302 }
5303 #endif
5304 
5305 int
5306 vi_full_init(struct vi_info *vi)
5307 {
5308 	struct adapter *sc = vi->pi->adapter;
5309 	struct ifnet *ifp = vi->ifp;
5310 	uint16_t *rss;
5311 	struct sge_rxq *rxq;
5312 	int rc, i, j;
5313 #ifdef RSS
5314 	int nbuckets = rss_getnumbuckets();
5315 	int hashconfig = rss_gethashconfig();
5316 	int extra;
5317 #endif
5318 
5319 	ASSERT_SYNCHRONIZED_OP(sc);
5320 	KASSERT((vi->flags & VI_INIT_DONE) == 0,
5321 	    ("%s: VI_INIT_DONE already", __func__));
5322 
5323 	sysctl_ctx_init(&vi->ctx);
5324 	vi->flags |= VI_SYSCTL_CTX;
5325 
5326 	/*
5327 	 * Allocate tx/rx/fl queues for this VI.
5328 	 */
5329 	rc = t4_setup_vi_queues(vi);
5330 	if (rc != 0)
5331 		goto done;	/* error message displayed already */
5332 
5333 	/*
5334 	 * Setup RSS for this VI.  Save a copy of the RSS table for later use.
5335 	 */
5336 	if (vi->nrxq > vi->rss_size) {
5337 		if_printf(ifp, "nrxq (%d) > hw RSS table size (%d); "
5338 		    "some queues will never receive traffic.\n", vi->nrxq,
5339 		    vi->rss_size);
5340 	} else if (vi->rss_size % vi->nrxq) {
5341 		if_printf(ifp, "nrxq (%d), hw RSS table size (%d); "
5342 		    "expect uneven traffic distribution.\n", vi->nrxq,
5343 		    vi->rss_size);
5344 	}
5345 #ifdef RSS
5346 	if (vi->nrxq != nbuckets) {
5347 		if_printf(ifp, "nrxq (%d) != kernel RSS buckets (%d);"
5348 		    "performance will be impacted.\n", vi->nrxq, nbuckets);
5349 	}
5350 #endif
5351 	rss = malloc(vi->rss_size * sizeof (*rss), M_CXGBE, M_ZERO | M_WAITOK);
5352 	for (i = 0; i < vi->rss_size;) {
5353 #ifdef RSS
5354 		j = rss_get_indirection_to_bucket(i);
5355 		j %= vi->nrxq;
5356 		rxq = &sc->sge.rxq[vi->first_rxq + j];
5357 		rss[i++] = rxq->iq.abs_id;
5358 #else
5359 		for_each_rxq(vi, j, rxq) {
5360 			rss[i++] = rxq->iq.abs_id;
5361 			if (i == vi->rss_size)
5362 				break;
5363 		}
5364 #endif
5365 	}
5366 
5367 	rc = -t4_config_rss_range(sc, sc->mbox, vi->viid, 0, vi->rss_size, rss,
5368 	    vi->rss_size);
5369 	if (rc != 0) {
5370 		free(rss, M_CXGBE);
5371 		if_printf(ifp, "rss_config failed: %d\n", rc);
5372 		goto done;
5373 	}
5374 
5375 #ifdef RSS
5376 	vi->hashen = hashconfig_to_hashen(hashconfig);
5377 
5378 	/*
5379 	 * We may have had to enable some hashes even though the global config
5380 	 * wants them disabled.  This is a potential problem that must be
5381 	 * reported to the user.
5382 	 */
5383 	extra = hashen_to_hashconfig(vi->hashen) ^ hashconfig;
5384 
5385 	/*
5386 	 * If we consider only the supported hash types, then the enabled hashes
5387 	 * are a superset of the requested hashes.  In other words, there cannot
5388 	 * be any supported hash that was requested but not enabled, but there
5389 	 * can be hashes that were not requested but had to be enabled.
5390 	 */
5391 	extra &= SUPPORTED_RSS_HASHTYPES;
5392 	MPASS((extra & hashconfig) == 0);
5393 
5394 	if (extra) {
5395 		if_printf(ifp,
5396 		    "global RSS config (0x%x) cannot be accommodated.\n",
5397 		    hashconfig);
5398 	}
5399 	if (extra & RSS_HASHTYPE_RSS_IPV4)
5400 		if_printf(ifp, "IPv4 2-tuple hashing forced on.\n");
5401 	if (extra & RSS_HASHTYPE_RSS_TCP_IPV4)
5402 		if_printf(ifp, "TCP/IPv4 4-tuple hashing forced on.\n");
5403 	if (extra & RSS_HASHTYPE_RSS_IPV6)
5404 		if_printf(ifp, "IPv6 2-tuple hashing forced on.\n");
5405 	if (extra & RSS_HASHTYPE_RSS_TCP_IPV6)
5406 		if_printf(ifp, "TCP/IPv6 4-tuple hashing forced on.\n");
5407 	if (extra & RSS_HASHTYPE_RSS_UDP_IPV4)
5408 		if_printf(ifp, "UDP/IPv4 4-tuple hashing forced on.\n");
5409 	if (extra & RSS_HASHTYPE_RSS_UDP_IPV6)
5410 		if_printf(ifp, "UDP/IPv6 4-tuple hashing forced on.\n");
5411 #else
5412 	vi->hashen = F_FW_RSS_VI_CONFIG_CMD_IP6FOURTUPEN |
5413 	    F_FW_RSS_VI_CONFIG_CMD_IP6TWOTUPEN |
5414 	    F_FW_RSS_VI_CONFIG_CMD_IP4FOURTUPEN |
5415 	    F_FW_RSS_VI_CONFIG_CMD_IP4TWOTUPEN | F_FW_RSS_VI_CONFIG_CMD_UDPEN;
5416 #endif
5417 	rc = -t4_config_vi_rss(sc, sc->mbox, vi->viid, vi->hashen, rss[0], 0, 0);
5418 	if (rc != 0) {
5419 		free(rss, M_CXGBE);
5420 		if_printf(ifp, "rss hash/defaultq config failed: %d\n", rc);
5421 		goto done;
5422 	}
5423 
5424 	vi->rss = rss;
5425 	vi->flags |= VI_INIT_DONE;
5426 done:
5427 	if (rc != 0)
5428 		vi_full_uninit(vi);
5429 
5430 	return (rc);
5431 }
5432 
5433 /*
5434  * Idempotent.
5435  */
5436 int
5437 vi_full_uninit(struct vi_info *vi)
5438 {
5439 	struct port_info *pi = vi->pi;
5440 	struct adapter *sc = pi->adapter;
5441 	int i;
5442 	struct sge_rxq *rxq;
5443 	struct sge_txq *txq;
5444 #ifdef TCP_OFFLOAD
5445 	struct sge_ofld_rxq *ofld_rxq;
5446 	struct sge_wrq *ofld_txq;
5447 #endif
5448 
5449 	if (vi->flags & VI_INIT_DONE) {
5450 
5451 		/* Need to quiesce queues.  */
5452 
5453 		/* XXX: Only for the first VI? */
5454 		if (IS_MAIN_VI(vi) && !(sc->flags & IS_VF))
5455 			quiesce_wrq(sc, &sc->sge.ctrlq[pi->port_id]);
5456 
5457 		for_each_txq(vi, i, txq) {
5458 			quiesce_txq(sc, txq);
5459 		}
5460 
5461 #ifdef TCP_OFFLOAD
5462 		for_each_ofld_txq(vi, i, ofld_txq) {
5463 			quiesce_wrq(sc, ofld_txq);
5464 		}
5465 #endif
5466 
5467 		for_each_rxq(vi, i, rxq) {
5468 			quiesce_iq(sc, &rxq->iq);
5469 			quiesce_fl(sc, &rxq->fl);
5470 		}
5471 
5472 #ifdef TCP_OFFLOAD
5473 		for_each_ofld_rxq(vi, i, ofld_rxq) {
5474 			quiesce_iq(sc, &ofld_rxq->iq);
5475 			quiesce_fl(sc, &ofld_rxq->fl);
5476 		}
5477 #endif
5478 		free(vi->rss, M_CXGBE);
5479 		free(vi->nm_rss, M_CXGBE);
5480 	}
5481 
5482 	t4_teardown_vi_queues(vi);
5483 	vi->flags &= ~VI_INIT_DONE;
5484 
5485 	return (0);
5486 }
5487 
5488 static void
5489 quiesce_txq(struct adapter *sc, struct sge_txq *txq)
5490 {
5491 	struct sge_eq *eq = &txq->eq;
5492 	struct sge_qstat *spg = (void *)&eq->desc[eq->sidx];
5493 
5494 	(void) sc;	/* unused */
5495 
5496 #ifdef INVARIANTS
5497 	TXQ_LOCK(txq);
5498 	MPASS((eq->flags & EQ_ENABLED) == 0);
5499 	TXQ_UNLOCK(txq);
5500 #endif
5501 
5502 	/* Wait for the mp_ring to empty. */
5503 	while (!mp_ring_is_idle(txq->r)) {
5504 		mp_ring_check_drainage(txq->r, 0);
5505 		pause("rquiesce", 1);
5506 	}
5507 
5508 	/* Then wait for the hardware to finish. */
5509 	while (spg->cidx != htobe16(eq->pidx))
5510 		pause("equiesce", 1);
5511 
5512 	/* Finally, wait for the driver to reclaim all descriptors. */
5513 	while (eq->cidx != eq->pidx)
5514 		pause("dquiesce", 1);
5515 }
5516 
5517 static void
5518 quiesce_wrq(struct adapter *sc, struct sge_wrq *wrq)
5519 {
5520 
5521 	/* XXXTX */
5522 }
5523 
5524 static void
5525 quiesce_iq(struct adapter *sc, struct sge_iq *iq)
5526 {
5527 	(void) sc;	/* unused */
5528 
5529 	/* Synchronize with the interrupt handler */
5530 	while (!atomic_cmpset_int(&iq->state, IQS_IDLE, IQS_DISABLED))
5531 		pause("iqfree", 1);
5532 }
5533 
5534 static void
5535 quiesce_fl(struct adapter *sc, struct sge_fl *fl)
5536 {
5537 	mtx_lock(&sc->sfl_lock);
5538 	FL_LOCK(fl);
5539 	fl->flags |= FL_DOOMED;
5540 	FL_UNLOCK(fl);
5541 	callout_stop(&sc->sfl_callout);
5542 	mtx_unlock(&sc->sfl_lock);
5543 
5544 	KASSERT((fl->flags & FL_STARVING) == 0,
5545 	    ("%s: still starving", __func__));
5546 }
5547 
5548 static int
5549 t4_alloc_irq(struct adapter *sc, struct irq *irq, int rid,
5550     driver_intr_t *handler, void *arg, char *name)
5551 {
5552 	int rc;
5553 
5554 	irq->rid = rid;
5555 	irq->res = bus_alloc_resource_any(sc->dev, SYS_RES_IRQ, &irq->rid,
5556 	    RF_SHAREABLE | RF_ACTIVE);
5557 	if (irq->res == NULL) {
5558 		device_printf(sc->dev,
5559 		    "failed to allocate IRQ for rid %d, name %s.\n", rid, name);
5560 		return (ENOMEM);
5561 	}
5562 
5563 	rc = bus_setup_intr(sc->dev, irq->res, INTR_MPSAFE | INTR_TYPE_NET,
5564 	    NULL, handler, arg, &irq->tag);
5565 	if (rc != 0) {
5566 		device_printf(sc->dev,
5567 		    "failed to setup interrupt for rid %d, name %s: %d\n",
5568 		    rid, name, rc);
5569 	} else if (name)
5570 		bus_describe_intr(sc->dev, irq->res, irq->tag, "%s", name);
5571 
5572 	return (rc);
5573 }
5574 
5575 static int
5576 t4_free_irq(struct adapter *sc, struct irq *irq)
5577 {
5578 	if (irq->tag)
5579 		bus_teardown_intr(sc->dev, irq->res, irq->tag);
5580 	if (irq->res)
5581 		bus_release_resource(sc->dev, SYS_RES_IRQ, irq->rid, irq->res);
5582 
5583 	bzero(irq, sizeof(*irq));
5584 
5585 	return (0);
5586 }
5587 
5588 static void
5589 get_regs(struct adapter *sc, struct t4_regdump *regs, uint8_t *buf)
5590 {
5591 
5592 	regs->version = chip_id(sc) | chip_rev(sc) << 10;
5593 	t4_get_regs(sc, buf, regs->len);
5594 }
5595 
5596 #define	A_PL_INDIR_CMD	0x1f8
5597 
5598 #define	S_PL_AUTOINC	31
5599 #define	M_PL_AUTOINC	0x1U
5600 #define	V_PL_AUTOINC(x)	((x) << S_PL_AUTOINC)
5601 #define	G_PL_AUTOINC(x)	(((x) >> S_PL_AUTOINC) & M_PL_AUTOINC)
5602 
5603 #define	S_PL_VFID	20
5604 #define	M_PL_VFID	0xffU
5605 #define	V_PL_VFID(x)	((x) << S_PL_VFID)
5606 #define	G_PL_VFID(x)	(((x) >> S_PL_VFID) & M_PL_VFID)
5607 
5608 #define	S_PL_ADDR	0
5609 #define	M_PL_ADDR	0xfffffU
5610 #define	V_PL_ADDR(x)	((x) << S_PL_ADDR)
5611 #define	G_PL_ADDR(x)	(((x) >> S_PL_ADDR) & M_PL_ADDR)
5612 
5613 #define	A_PL_INDIR_DATA	0x1fc
5614 
5615 static uint64_t
5616 read_vf_stat(struct adapter *sc, unsigned int viid, int reg)
5617 {
5618 	u32 stats[2];
5619 
5620 	mtx_assert(&sc->reg_lock, MA_OWNED);
5621 	if (sc->flags & IS_VF) {
5622 		stats[0] = t4_read_reg(sc, VF_MPS_REG(reg));
5623 		stats[1] = t4_read_reg(sc, VF_MPS_REG(reg + 4));
5624 	} else {
5625 		t4_write_reg(sc, A_PL_INDIR_CMD, V_PL_AUTOINC(1) |
5626 		    V_PL_VFID(G_FW_VIID_VIN(viid)) |
5627 		    V_PL_ADDR(VF_MPS_REG(reg)));
5628 		stats[0] = t4_read_reg(sc, A_PL_INDIR_DATA);
5629 		stats[1] = t4_read_reg(sc, A_PL_INDIR_DATA);
5630 	}
5631 	return (((uint64_t)stats[1]) << 32 | stats[0]);
5632 }
5633 
5634 static void
5635 t4_get_vi_stats(struct adapter *sc, unsigned int viid,
5636     struct fw_vi_stats_vf *stats)
5637 {
5638 
5639 #define GET_STAT(name) \
5640 	read_vf_stat(sc, viid, A_MPS_VF_STAT_##name##_L)
5641 
5642 	stats->tx_bcast_bytes    = GET_STAT(TX_VF_BCAST_BYTES);
5643 	stats->tx_bcast_frames   = GET_STAT(TX_VF_BCAST_FRAMES);
5644 	stats->tx_mcast_bytes    = GET_STAT(TX_VF_MCAST_BYTES);
5645 	stats->tx_mcast_frames   = GET_STAT(TX_VF_MCAST_FRAMES);
5646 	stats->tx_ucast_bytes    = GET_STAT(TX_VF_UCAST_BYTES);
5647 	stats->tx_ucast_frames   = GET_STAT(TX_VF_UCAST_FRAMES);
5648 	stats->tx_drop_frames    = GET_STAT(TX_VF_DROP_FRAMES);
5649 	stats->tx_offload_bytes  = GET_STAT(TX_VF_OFFLOAD_BYTES);
5650 	stats->tx_offload_frames = GET_STAT(TX_VF_OFFLOAD_FRAMES);
5651 	stats->rx_bcast_bytes    = GET_STAT(RX_VF_BCAST_BYTES);
5652 	stats->rx_bcast_frames   = GET_STAT(RX_VF_BCAST_FRAMES);
5653 	stats->rx_mcast_bytes    = GET_STAT(RX_VF_MCAST_BYTES);
5654 	stats->rx_mcast_frames   = GET_STAT(RX_VF_MCAST_FRAMES);
5655 	stats->rx_ucast_bytes    = GET_STAT(RX_VF_UCAST_BYTES);
5656 	stats->rx_ucast_frames   = GET_STAT(RX_VF_UCAST_FRAMES);
5657 	stats->rx_err_frames     = GET_STAT(RX_VF_ERR_FRAMES);
5658 
5659 #undef GET_STAT
5660 }
5661 
5662 static void
5663 t4_clr_vi_stats(struct adapter *sc, unsigned int viid)
5664 {
5665 	int reg;
5666 
5667 	t4_write_reg(sc, A_PL_INDIR_CMD, V_PL_AUTOINC(1) |
5668 	    V_PL_VFID(G_FW_VIID_VIN(viid)) |
5669 	    V_PL_ADDR(VF_MPS_REG(A_MPS_VF_STAT_TX_VF_BCAST_BYTES_L)));
5670 	for (reg = A_MPS_VF_STAT_TX_VF_BCAST_BYTES_L;
5671 	     reg <= A_MPS_VF_STAT_RX_VF_ERR_FRAMES_H; reg += 4)
5672 		t4_write_reg(sc, A_PL_INDIR_DATA, 0);
5673 }
5674 
5675 static void
5676 vi_refresh_stats(struct adapter *sc, struct vi_info *vi)
5677 {
5678 	struct timeval tv;
5679 	const struct timeval interval = {0, 250000};	/* 250ms */
5680 
5681 	if (!(vi->flags & VI_INIT_DONE))
5682 		return;
5683 
5684 	getmicrotime(&tv);
5685 	timevalsub(&tv, &interval);
5686 	if (timevalcmp(&tv, &vi->last_refreshed, <))
5687 		return;
5688 
5689 	mtx_lock(&sc->reg_lock);
5690 	t4_get_vi_stats(sc, vi->viid, &vi->stats);
5691 	getmicrotime(&vi->last_refreshed);
5692 	mtx_unlock(&sc->reg_lock);
5693 }
5694 
5695 static void
5696 cxgbe_refresh_stats(struct adapter *sc, struct port_info *pi)
5697 {
5698 	u_int i, v, tnl_cong_drops, bg_map;
5699 	struct timeval tv;
5700 	const struct timeval interval = {0, 250000};	/* 250ms */
5701 
5702 	getmicrotime(&tv);
5703 	timevalsub(&tv, &interval);
5704 	if (timevalcmp(&tv, &pi->last_refreshed, <))
5705 		return;
5706 
5707 	tnl_cong_drops = 0;
5708 	t4_get_port_stats(sc, pi->tx_chan, &pi->stats);
5709 	bg_map = pi->mps_bg_map;
5710 	while (bg_map) {
5711 		i = ffs(bg_map) - 1;
5712 		mtx_lock(&sc->reg_lock);
5713 		t4_read_indirect(sc, A_TP_MIB_INDEX, A_TP_MIB_DATA, &v, 1,
5714 		    A_TP_MIB_TNL_CNG_DROP_0 + i);
5715 		mtx_unlock(&sc->reg_lock);
5716 		tnl_cong_drops += v;
5717 		bg_map &= ~(1 << i);
5718 	}
5719 	pi->tnl_cong_drops = tnl_cong_drops;
5720 	getmicrotime(&pi->last_refreshed);
5721 }
5722 
5723 static void
5724 cxgbe_tick(void *arg)
5725 {
5726 	struct port_info *pi = arg;
5727 	struct adapter *sc = pi->adapter;
5728 
5729 	PORT_LOCK_ASSERT_OWNED(pi);
5730 	cxgbe_refresh_stats(sc, pi);
5731 
5732 	callout_schedule(&pi->tick, hz);
5733 }
5734 
5735 void
5736 vi_tick(void *arg)
5737 {
5738 	struct vi_info *vi = arg;
5739 	struct adapter *sc = vi->pi->adapter;
5740 
5741 	vi_refresh_stats(sc, vi);
5742 
5743 	callout_schedule(&vi->tick, hz);
5744 }
5745 
5746 /*
5747  * Should match fw_caps_config_<foo> enums in t4fw_interface.h
5748  */
5749 static char *caps_decoder[] = {
5750 	"\20\001IPMI\002NCSI",				/* 0: NBM */
5751 	"\20\001PPP\002QFC\003DCBX",			/* 1: link */
5752 	"\20\001INGRESS\002EGRESS",			/* 2: switch */
5753 	"\20\001NIC\002VM\003IDS\004UM\005UM_ISGL"	/* 3: NIC */
5754 	    "\006HASHFILTER\007ETHOFLD",
5755 	"\20\001TOE",					/* 4: TOE */
5756 	"\20\001RDDP\002RDMAC",				/* 5: RDMA */
5757 	"\20\001INITIATOR_PDU\002TARGET_PDU"		/* 6: iSCSI */
5758 	    "\003INITIATOR_CNXOFLD\004TARGET_CNXOFLD"
5759 	    "\005INITIATOR_SSNOFLD\006TARGET_SSNOFLD"
5760 	    "\007T10DIF"
5761 	    "\010INITIATOR_CMDOFLD\011TARGET_CMDOFLD",
5762 	"\20\001LOOKASIDE\002TLSKEYS",			/* 7: Crypto */
5763 	"\20\001INITIATOR\002TARGET\003CTRL_OFLD"	/* 8: FCoE */
5764 		    "\004PO_INITIATOR\005PO_TARGET",
5765 };
5766 
5767 void
5768 t4_sysctls(struct adapter *sc)
5769 {
5770 	struct sysctl_ctx_list *ctx;
5771 	struct sysctl_oid *oid;
5772 	struct sysctl_oid_list *children, *c0;
5773 	static char *doorbells = {"\20\1UDB\2WCWR\3UDBWC\4KDB"};
5774 
5775 	ctx = device_get_sysctl_ctx(sc->dev);
5776 
5777 	/*
5778 	 * dev.t4nex.X.
5779 	 */
5780 	oid = device_get_sysctl_tree(sc->dev);
5781 	c0 = children = SYSCTL_CHILDREN(oid);
5782 
5783 	sc->sc_do_rxcopy = 1;
5784 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "do_rx_copy", CTLFLAG_RW,
5785 	    &sc->sc_do_rxcopy, 1, "Do RX copy of small frames");
5786 
5787 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nports", CTLFLAG_RD, NULL,
5788 	    sc->params.nports, "# of ports");
5789 
5790 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "doorbells",
5791 	    CTLTYPE_STRING | CTLFLAG_RD, doorbells, (uintptr_t)&sc->doorbells,
5792 	    sysctl_bitfield_8b, "A", "available doorbells");
5793 
5794 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "core_clock", CTLFLAG_RD, NULL,
5795 	    sc->params.vpd.cclk, "core clock frequency (in KHz)");
5796 
5797 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_timers",
5798 	    CTLTYPE_STRING | CTLFLAG_RD, sc->params.sge.timer_val,
5799 	    sizeof(sc->params.sge.timer_val), sysctl_int_array, "A",
5800 	    "interrupt holdoff timer values (us)");
5801 
5802 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pkt_counts",
5803 	    CTLTYPE_STRING | CTLFLAG_RD, sc->params.sge.counter_val,
5804 	    sizeof(sc->params.sge.counter_val), sysctl_int_array, "A",
5805 	    "interrupt holdoff packet counter values");
5806 
5807 	t4_sge_sysctls(sc, ctx, children);
5808 
5809 	sc->lro_timeout = 100;
5810 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "lro_timeout", CTLFLAG_RW,
5811 	    &sc->lro_timeout, 0, "lro inactive-flush timeout (in us)");
5812 
5813 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "dflags", CTLFLAG_RW,
5814 	    &sc->debug_flags, 0, "flags to enable runtime debugging");
5815 
5816 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "tp_version",
5817 	    CTLFLAG_RD, sc->tp_version, 0, "TP microcode version");
5818 
5819 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "firmware_version",
5820 	    CTLFLAG_RD, sc->fw_version, 0, "firmware version");
5821 
5822 	if (sc->flags & IS_VF)
5823 		return;
5824 
5825 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "hw_revision", CTLFLAG_RD,
5826 	    NULL, chip_rev(sc), "chip hardware revision");
5827 
5828 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "sn",
5829 	    CTLFLAG_RD, sc->params.vpd.sn, 0, "serial number");
5830 
5831 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "pn",
5832 	    CTLFLAG_RD, sc->params.vpd.pn, 0, "part number");
5833 
5834 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "ec",
5835 	    CTLFLAG_RD, sc->params.vpd.ec, 0, "engineering change");
5836 
5837 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "md_version",
5838 	    CTLFLAG_RD, sc->params.vpd.md, 0, "manufacturing diags version");
5839 
5840 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "na",
5841 	    CTLFLAG_RD, sc->params.vpd.na, 0, "network address");
5842 
5843 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "er_version", CTLFLAG_RD,
5844 	    sc->er_version, 0, "expansion ROM version");
5845 
5846 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "bs_version", CTLFLAG_RD,
5847 	    sc->bs_version, 0, "bootstrap firmware version");
5848 
5849 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "scfg_version", CTLFLAG_RD,
5850 	    NULL, sc->params.scfg_vers, "serial config version");
5851 
5852 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "vpd_version", CTLFLAG_RD,
5853 	    NULL, sc->params.vpd_vers, "VPD version");
5854 
5855 	SYSCTL_ADD_STRING(ctx, children, OID_AUTO, "cf",
5856 	    CTLFLAG_RD, sc->cfg_file, 0, "configuration file");
5857 
5858 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "cfcsum", CTLFLAG_RD, NULL,
5859 	    sc->cfcsum, "config file checksum");
5860 
5861 #define SYSCTL_CAP(name, n, text) \
5862 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, #name, \
5863 	    CTLTYPE_STRING | CTLFLAG_RD, caps_decoder[n], (uintptr_t)&sc->name, \
5864 	    sysctl_bitfield_16b, "A", "available " text " capabilities")
5865 
5866 	SYSCTL_CAP(nbmcaps, 0, "NBM");
5867 	SYSCTL_CAP(linkcaps, 1, "link");
5868 	SYSCTL_CAP(switchcaps, 2, "switch");
5869 	SYSCTL_CAP(niccaps, 3, "NIC");
5870 	SYSCTL_CAP(toecaps, 4, "TCP offload");
5871 	SYSCTL_CAP(rdmacaps, 5, "RDMA");
5872 	SYSCTL_CAP(iscsicaps, 6, "iSCSI");
5873 	SYSCTL_CAP(cryptocaps, 7, "crypto");
5874 	SYSCTL_CAP(fcoecaps, 8, "FCoE");
5875 #undef SYSCTL_CAP
5876 
5877 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nfilters", CTLFLAG_RD,
5878 	    NULL, sc->tids.nftids, "number of filters");
5879 
5880 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "temperature", CTLTYPE_INT |
5881 	    CTLFLAG_RD, sc, 0, sysctl_temperature, "I",
5882 	    "chip temperature (in Celsius)");
5883 
5884 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "loadavg", CTLTYPE_STRING |
5885 	    CTLFLAG_RD, sc, 0, sysctl_loadavg, "A",
5886 	    "microprocessor load averages (debug firmwares only)");
5887 
5888 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "core_vdd", CTLFLAG_RD,
5889 	    &sc->params.core_vdd, 0, "core Vdd (in mV)");
5890 
5891 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "local_cpus",
5892 	    CTLTYPE_STRING | CTLFLAG_RD, sc, LOCAL_CPUS,
5893 	    sysctl_cpus, "A", "local CPUs");
5894 
5895 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "intr_cpus",
5896 	    CTLTYPE_STRING | CTLFLAG_RD, sc, INTR_CPUS,
5897 	    sysctl_cpus, "A", "preferred CPUs for interrupts");
5898 
5899 	/*
5900 	 * dev.t4nex.X.misc.  Marked CTLFLAG_SKIP to avoid information overload.
5901 	 */
5902 	oid = SYSCTL_ADD_NODE(ctx, c0, OID_AUTO, "misc",
5903 	    CTLFLAG_RD | CTLFLAG_SKIP, NULL,
5904 	    "logs and miscellaneous information");
5905 	children = SYSCTL_CHILDREN(oid);
5906 
5907 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cctrl",
5908 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5909 	    sysctl_cctrl, "A", "congestion control");
5910 
5911 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_tp0",
5912 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5913 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 0 (TP0)");
5914 
5915 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_tp1",
5916 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 1,
5917 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 1 (TP1)");
5918 
5919 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_ulp",
5920 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 2,
5921 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 2 (ULP)");
5922 
5923 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_sge0",
5924 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 3,
5925 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 3 (SGE0)");
5926 
5927 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_sge1",
5928 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 4,
5929 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 4 (SGE1)");
5930 
5931 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ibq_ncsi",
5932 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 5,
5933 	    sysctl_cim_ibq_obq, "A", "CIM IBQ 5 (NCSI)");
5934 
5935 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_la",
5936 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5937 	    chip_id(sc) <= CHELSIO_T5 ? sysctl_cim_la : sysctl_cim_la_t6,
5938 	    "A", "CIM logic analyzer");
5939 
5940 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_ma_la",
5941 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5942 	    sysctl_cim_ma_la, "A", "CIM MA logic analyzer");
5943 
5944 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp0",
5945 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0 + CIM_NUM_IBQ,
5946 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 0 (ULP0)");
5947 
5948 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp1",
5949 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 1 + CIM_NUM_IBQ,
5950 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 1 (ULP1)");
5951 
5952 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp2",
5953 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 2 + CIM_NUM_IBQ,
5954 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 2 (ULP2)");
5955 
5956 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ulp3",
5957 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 3 + CIM_NUM_IBQ,
5958 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 3 (ULP3)");
5959 
5960 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge",
5961 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 4 + CIM_NUM_IBQ,
5962 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 4 (SGE)");
5963 
5964 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_ncsi",
5965 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 5 + CIM_NUM_IBQ,
5966 	    sysctl_cim_ibq_obq, "A", "CIM OBQ 5 (NCSI)");
5967 
5968 	if (chip_id(sc) > CHELSIO_T4) {
5969 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge0_rx",
5970 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 6 + CIM_NUM_IBQ,
5971 		    sysctl_cim_ibq_obq, "A", "CIM OBQ 6 (SGE0-RX)");
5972 
5973 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_obq_sge1_rx",
5974 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 7 + CIM_NUM_IBQ,
5975 		    sysctl_cim_ibq_obq, "A", "CIM OBQ 7 (SGE1-RX)");
5976 	}
5977 
5978 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_pif_la",
5979 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5980 	    sysctl_cim_pif_la, "A", "CIM PIF logic analyzer");
5981 
5982 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cim_qcfg",
5983 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5984 	    sysctl_cim_qcfg, "A", "CIM queue configuration");
5985 
5986 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "cpl_stats",
5987 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5988 	    sysctl_cpl_stats, "A", "CPL statistics");
5989 
5990 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "ddp_stats",
5991 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5992 	    sysctl_ddp_stats, "A", "non-TCP DDP statistics");
5993 
5994 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "devlog",
5995 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
5996 	    sysctl_devlog, "A", "firmware's device log");
5997 
5998 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fcoe_stats",
5999 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6000 	    sysctl_fcoe_stats, "A", "FCoE statistics");
6001 
6002 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "hw_sched",
6003 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6004 	    sysctl_hw_sched, "A", "hardware scheduler ");
6005 
6006 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "l2t",
6007 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6008 	    sysctl_l2t, "A", "hardware L2 table");
6009 
6010 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "smt",
6011 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6012 	    sysctl_smt, "A", "hardware source MAC table");
6013 
6014 #ifdef INET6
6015 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "clip",
6016 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6017 	    sysctl_clip, "A", "active CLIP table entries");
6018 #endif
6019 
6020 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "lb_stats",
6021 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6022 	    sysctl_lb_stats, "A", "loopback statistics");
6023 
6024 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "meminfo",
6025 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6026 	    sysctl_meminfo, "A", "memory regions");
6027 
6028 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "mps_tcam",
6029 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6030 	    chip_id(sc) <= CHELSIO_T5 ? sysctl_mps_tcam : sysctl_mps_tcam_t6,
6031 	    "A", "MPS TCAM entries");
6032 
6033 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "path_mtus",
6034 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6035 	    sysctl_path_mtus, "A", "path MTUs");
6036 
6037 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "pm_stats",
6038 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6039 	    sysctl_pm_stats, "A", "PM statistics");
6040 
6041 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rdma_stats",
6042 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6043 	    sysctl_rdma_stats, "A", "RDMA statistics");
6044 
6045 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tcp_stats",
6046 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6047 	    sysctl_tcp_stats, "A", "TCP statistics");
6048 
6049 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tids",
6050 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6051 	    sysctl_tids, "A", "TID information");
6052 
6053 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_err_stats",
6054 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6055 	    sysctl_tp_err_stats, "A", "TP error statistics");
6056 
6057 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_la_mask",
6058 	    CTLTYPE_INT | CTLFLAG_RW, sc, 0, sysctl_tp_la_mask, "I",
6059 	    "TP logic analyzer event capture mask");
6060 
6061 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tp_la",
6062 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6063 	    sysctl_tp_la, "A", "TP logic analyzer");
6064 
6065 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tx_rate",
6066 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6067 	    sysctl_tx_rate, "A", "Tx rate");
6068 
6069 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "ulprx_la",
6070 	    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6071 	    sysctl_ulprx_la, "A", "ULPRX logic analyzer");
6072 
6073 	if (chip_id(sc) >= CHELSIO_T5) {
6074 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "wcwr_stats",
6075 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 0,
6076 		    sysctl_wcwr_stats, "A", "write combined work requests");
6077 	}
6078 
6079 #ifdef TCP_OFFLOAD
6080 	if (is_offload(sc)) {
6081 		int i;
6082 		char s[4];
6083 
6084 		/*
6085 		 * dev.t4nex.X.toe.
6086 		 */
6087 		oid = SYSCTL_ADD_NODE(ctx, c0, OID_AUTO, "toe", CTLFLAG_RD,
6088 		    NULL, "TOE parameters");
6089 		children = SYSCTL_CHILDREN(oid);
6090 
6091 		sc->tt.cong_algorithm = -1;
6092 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "cong_algorithm",
6093 		    CTLFLAG_RW, &sc->tt.cong_algorithm, 0, "congestion control "
6094 		    "(-1 = default, 0 = reno, 1 = tahoe, 2 = newreno, "
6095 		    "3 = highspeed)");
6096 
6097 		sc->tt.sndbuf = 256 * 1024;
6098 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "sndbuf", CTLFLAG_RW,
6099 		    &sc->tt.sndbuf, 0, "max hardware send buffer size");
6100 
6101 		sc->tt.ddp = 0;
6102 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ddp", CTLFLAG_RW,
6103 		    &sc->tt.ddp, 0, "DDP allowed");
6104 
6105 		sc->tt.rx_coalesce = 1;
6106 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "rx_coalesce",
6107 		    CTLFLAG_RW, &sc->tt.rx_coalesce, 0, "receive coalescing");
6108 
6109 		sc->tt.tls = 0;
6110 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tls", CTLFLAG_RW,
6111 		    &sc->tt.tls, 0, "Inline TLS allowed");
6112 
6113 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "tls_rx_ports",
6114 		    CTLTYPE_INT | CTLFLAG_RW, sc, 0, sysctl_tls_rx_ports,
6115 		    "I", "TCP ports that use inline TLS+TOE RX");
6116 
6117 		sc->tt.tx_align = 1;
6118 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tx_align",
6119 		    CTLFLAG_RW, &sc->tt.tx_align, 0, "chop and align payload");
6120 
6121 		sc->tt.tx_zcopy = 0;
6122 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "tx_zcopy",
6123 		    CTLFLAG_RW, &sc->tt.tx_zcopy, 0,
6124 		    "Enable zero-copy aio_write(2)");
6125 
6126 		sc->tt.cop_managed_offloading = !!t4_cop_managed_offloading;
6127 		SYSCTL_ADD_INT(ctx, children, OID_AUTO,
6128 		    "cop_managed_offloading", CTLFLAG_RW,
6129 		    &sc->tt.cop_managed_offloading, 0,
6130 		    "COP (Connection Offload Policy) controls all TOE offload");
6131 
6132 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "timer_tick",
6133 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 0, sysctl_tp_tick, "A",
6134 		    "TP timer tick (us)");
6135 
6136 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "timestamp_tick",
6137 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 1, sysctl_tp_tick, "A",
6138 		    "TCP timestamp tick (us)");
6139 
6140 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "dack_tick",
6141 		    CTLTYPE_STRING | CTLFLAG_RD, sc, 2, sysctl_tp_tick, "A",
6142 		    "DACK tick (us)");
6143 
6144 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "dack_timer",
6145 		    CTLTYPE_UINT | CTLFLAG_RD, sc, 0, sysctl_tp_dack_timer,
6146 		    "IU", "DACK timer (us)");
6147 
6148 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_min",
6149 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_RXT_MIN,
6150 		    sysctl_tp_timer, "LU", "Minimum retransmit interval (us)");
6151 
6152 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_max",
6153 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_RXT_MAX,
6154 		    sysctl_tp_timer, "LU", "Maximum retransmit interval (us)");
6155 
6156 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "persist_min",
6157 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_PERS_MIN,
6158 		    sysctl_tp_timer, "LU", "Persist timer min (us)");
6159 
6160 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "persist_max",
6161 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_PERS_MAX,
6162 		    sysctl_tp_timer, "LU", "Persist timer max (us)");
6163 
6164 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_idle",
6165 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_KEEP_IDLE,
6166 		    sysctl_tp_timer, "LU", "Keepalive idle timer (us)");
6167 
6168 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_interval",
6169 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_KEEP_INTVL,
6170 		    sysctl_tp_timer, "LU", "Keepalive interval timer (us)");
6171 
6172 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "initial_srtt",
6173 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_INIT_SRTT,
6174 		    sysctl_tp_timer, "LU", "Initial SRTT (us)");
6175 
6176 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "finwait2_timer",
6177 		    CTLTYPE_ULONG | CTLFLAG_RD, sc, A_TP_FINWAIT2_TIMER,
6178 		    sysctl_tp_timer, "LU", "FINWAIT2 timer (us)");
6179 
6180 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "syn_rexmt_count",
6181 		    CTLTYPE_UINT | CTLFLAG_RD, sc, S_SYNSHIFTMAX,
6182 		    sysctl_tp_shift_cnt, "IU",
6183 		    "Number of SYN retransmissions before abort");
6184 
6185 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rexmt_count",
6186 		    CTLTYPE_UINT | CTLFLAG_RD, sc, S_RXTSHIFTMAXR2,
6187 		    sysctl_tp_shift_cnt, "IU",
6188 		    "Number of retransmissions before abort");
6189 
6190 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "keepalive_count",
6191 		    CTLTYPE_UINT | CTLFLAG_RD, sc, S_KEEPALIVEMAXR2,
6192 		    sysctl_tp_shift_cnt, "IU",
6193 		    "Number of keepalive probes before abort");
6194 
6195 		oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "rexmt_backoff",
6196 		    CTLFLAG_RD, NULL, "TOE retransmit backoffs");
6197 		children = SYSCTL_CHILDREN(oid);
6198 		for (i = 0; i < 16; i++) {
6199 			snprintf(s, sizeof(s), "%u", i);
6200 			SYSCTL_ADD_PROC(ctx, children, OID_AUTO, s,
6201 			    CTLTYPE_UINT | CTLFLAG_RD, sc, i, sysctl_tp_backoff,
6202 			    "IU", "TOE retransmit backoff");
6203 		}
6204 	}
6205 #endif
6206 }
6207 
6208 void
6209 vi_sysctls(struct vi_info *vi)
6210 {
6211 	struct sysctl_ctx_list *ctx;
6212 	struct sysctl_oid *oid;
6213 	struct sysctl_oid_list *children;
6214 
6215 	ctx = device_get_sysctl_ctx(vi->dev);
6216 
6217 	/*
6218 	 * dev.v?(cxgbe|cxl).X.
6219 	 */
6220 	oid = device_get_sysctl_tree(vi->dev);
6221 	children = SYSCTL_CHILDREN(oid);
6222 
6223 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "viid", CTLFLAG_RD, NULL,
6224 	    vi->viid, "VI identifer");
6225 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nrxq", CTLFLAG_RD,
6226 	    &vi->nrxq, 0, "# of rx queues");
6227 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "ntxq", CTLFLAG_RD,
6228 	    &vi->ntxq, 0, "# of tx queues");
6229 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_rxq", CTLFLAG_RD,
6230 	    &vi->first_rxq, 0, "index of first rx queue");
6231 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_txq", CTLFLAG_RD,
6232 	    &vi->first_txq, 0, "index of first tx queue");
6233 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "rss_base", CTLFLAG_RD, NULL,
6234 	    vi->rss_base, "start of RSS indirection table");
6235 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "rss_size", CTLFLAG_RD, NULL,
6236 	    vi->rss_size, "size of RSS indirection table");
6237 
6238 	if (IS_MAIN_VI(vi)) {
6239 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "rsrv_noflowq",
6240 		    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_noflowq, "IU",
6241 		    "Reserve queue 0 for non-flowid packets");
6242 	}
6243 
6244 #ifdef TCP_OFFLOAD
6245 	if (vi->nofldrxq != 0) {
6246 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nofldrxq", CTLFLAG_RD,
6247 		    &vi->nofldrxq, 0,
6248 		    "# of rx queues for offloaded TCP connections");
6249 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nofldtxq", CTLFLAG_RD,
6250 		    &vi->nofldtxq, 0,
6251 		    "# of tx queues for offloaded TCP connections");
6252 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_ofld_rxq",
6253 		    CTLFLAG_RD, &vi->first_ofld_rxq, 0,
6254 		    "index of first TOE rx queue");
6255 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_ofld_txq",
6256 		    CTLFLAG_RD, &vi->first_ofld_txq, 0,
6257 		    "index of first TOE tx queue");
6258 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_tmr_idx_ofld",
6259 		    CTLTYPE_INT | CTLFLAG_RW, vi, 0,
6260 		    sysctl_holdoff_tmr_idx_ofld, "I",
6261 		    "holdoff timer index for TOE queues");
6262 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pktc_idx_ofld",
6263 		    CTLTYPE_INT | CTLFLAG_RW, vi, 0,
6264 		    sysctl_holdoff_pktc_idx_ofld, "I",
6265 		    "holdoff packet counter index for TOE queues");
6266 	}
6267 #endif
6268 #ifdef DEV_NETMAP
6269 	if (vi->nnmrxq != 0) {
6270 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nnmrxq", CTLFLAG_RD,
6271 		    &vi->nnmrxq, 0, "# of netmap rx queues");
6272 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "nnmtxq", CTLFLAG_RD,
6273 		    &vi->nnmtxq, 0, "# of netmap tx queues");
6274 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_nm_rxq",
6275 		    CTLFLAG_RD, &vi->first_nm_rxq, 0,
6276 		    "index of first netmap rx queue");
6277 		SYSCTL_ADD_INT(ctx, children, OID_AUTO, "first_nm_txq",
6278 		    CTLFLAG_RD, &vi->first_nm_txq, 0,
6279 		    "index of first netmap tx queue");
6280 	}
6281 #endif
6282 
6283 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_tmr_idx",
6284 	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_holdoff_tmr_idx, "I",
6285 	    "holdoff timer index");
6286 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "holdoff_pktc_idx",
6287 	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_holdoff_pktc_idx, "I",
6288 	    "holdoff packet counter index");
6289 
6290 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "qsize_rxq",
6291 	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_qsize_rxq, "I",
6292 	    "rx queue size");
6293 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "qsize_txq",
6294 	    CTLTYPE_INT | CTLFLAG_RW, vi, 0, sysctl_qsize_txq, "I",
6295 	    "tx queue size");
6296 }
6297 
6298 static void
6299 cxgbe_sysctls(struct port_info *pi)
6300 {
6301 	struct sysctl_ctx_list *ctx;
6302 	struct sysctl_oid *oid;
6303 	struct sysctl_oid_list *children, *children2;
6304 	struct adapter *sc = pi->adapter;
6305 	int i;
6306 	char name[16];
6307 	static char *tc_flags = {"\20\1USER\2SYNC\3ASYNC\4ERR"};
6308 
6309 	ctx = device_get_sysctl_ctx(pi->dev);
6310 
6311 	/*
6312 	 * dev.cxgbe.X.
6313 	 */
6314 	oid = device_get_sysctl_tree(pi->dev);
6315 	children = SYSCTL_CHILDREN(oid);
6316 
6317 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "linkdnrc", CTLTYPE_STRING |
6318 	   CTLFLAG_RD, pi, 0, sysctl_linkdnrc, "A", "reason why link is down");
6319 	if (pi->port_type == FW_PORT_TYPE_BT_XAUI) {
6320 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "temperature",
6321 		    CTLTYPE_INT | CTLFLAG_RD, pi, 0, sysctl_btphy, "I",
6322 		    "PHY temperature (in Celsius)");
6323 		SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fw_version",
6324 		    CTLTYPE_INT | CTLFLAG_RD, pi, 1, sysctl_btphy, "I",
6325 		    "PHY firmware version");
6326 	}
6327 
6328 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "pause_settings",
6329 	    CTLTYPE_STRING | CTLFLAG_RW, pi, 0, sysctl_pause_settings, "A",
6330     "PAUSE settings (bit 0 = rx_pause, 1 = tx_pause, 2 = pause_autoneg)");
6331 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "fec",
6332 	    CTLTYPE_STRING | CTLFLAG_RW, pi, 0, sysctl_fec, "A",
6333 	    "Forward Error Correction (bit 0 = RS, bit 1 = BASER_RS)");
6334 	SYSCTL_ADD_PROC(ctx, children, OID_AUTO, "autoneg",
6335 	    CTLTYPE_INT | CTLFLAG_RW, pi, 0, sysctl_autoneg, "I",
6336 	    "autonegotiation (-1 = not supported)");
6337 
6338 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "max_speed", CTLFLAG_RD, NULL,
6339 	    port_top_speed(pi), "max speed (in Gbps)");
6340 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "mps_bg_map", CTLFLAG_RD, NULL,
6341 	    pi->mps_bg_map, "MPS buffer group map");
6342 	SYSCTL_ADD_INT(ctx, children, OID_AUTO, "rx_e_chan_map", CTLFLAG_RD,
6343 	    NULL, pi->rx_e_chan_map, "TP rx e-channel map");
6344 
6345 	if (sc->flags & IS_VF)
6346 		return;
6347 
6348 	/*
6349 	 * dev.(cxgbe|cxl).X.tc.
6350 	 */
6351 	oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "tc", CTLFLAG_RD, NULL,
6352 	    "Tx scheduler traffic classes (cl_rl)");
6353 	children2 = SYSCTL_CHILDREN(oid);
6354 	SYSCTL_ADD_UINT(ctx, children2, OID_AUTO, "pktsize",
6355 	    CTLFLAG_RW, &pi->sched_params->pktsize, 0,
6356 	    "pktsize for per-flow cl-rl (0 means up to the driver )");
6357 	SYSCTL_ADD_UINT(ctx, children2, OID_AUTO, "burstsize",
6358 	    CTLFLAG_RW, &pi->sched_params->burstsize, 0,
6359 	    "burstsize for per-flow cl-rl (0 means up to the driver)");
6360 	for (i = 0; i < sc->chip_params->nsched_cls; i++) {
6361 		struct tx_cl_rl_params *tc = &pi->sched_params->cl_rl[i];
6362 
6363 		snprintf(name, sizeof(name), "%d", i);
6364 		children2 = SYSCTL_CHILDREN(SYSCTL_ADD_NODE(ctx,
6365 		    SYSCTL_CHILDREN(oid), OID_AUTO, name, CTLFLAG_RD, NULL,
6366 		    "traffic class"));
6367 		SYSCTL_ADD_PROC(ctx, children2, OID_AUTO, "flags",
6368 		    CTLTYPE_STRING | CTLFLAG_RD, tc_flags, (uintptr_t)&tc->flags,
6369 		    sysctl_bitfield_8b, "A", "flags");
6370 		SYSCTL_ADD_UINT(ctx, children2, OID_AUTO, "refcount",
6371 		    CTLFLAG_RD, &tc->refcount, 0, "references to this class");
6372 		SYSCTL_ADD_PROC(ctx, children2, OID_AUTO, "params",
6373 		    CTLTYPE_STRING | CTLFLAG_RD, sc, (pi->port_id << 16) | i,
6374 		    sysctl_tc_params, "A", "traffic class parameters");
6375 	}
6376 
6377 	/*
6378 	 * dev.cxgbe.X.stats.
6379 	 */
6380 	oid = SYSCTL_ADD_NODE(ctx, children, OID_AUTO, "stats", CTLFLAG_RD,
6381 	    NULL, "port statistics");
6382 	children = SYSCTL_CHILDREN(oid);
6383 	SYSCTL_ADD_UINT(ctx, children, OID_AUTO, "tx_parse_error", CTLFLAG_RD,
6384 	    &pi->tx_parse_error, 0,
6385 	    "# of tx packets with invalid length or # of segments");
6386 
6387 #define SYSCTL_ADD_T4_REG64(pi, name, desc, reg) \
6388 	SYSCTL_ADD_OID(ctx, children, OID_AUTO, name, \
6389 	    CTLTYPE_U64 | CTLFLAG_RD, sc, reg, \
6390 	    sysctl_handle_t4_reg64, "QU", desc)
6391 
6392 	SYSCTL_ADD_T4_REG64(pi, "tx_octets", "# of octets in good frames",
6393 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_BYTES_L));
6394 	SYSCTL_ADD_T4_REG64(pi, "tx_frames", "total # of good frames",
6395 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_FRAMES_L));
6396 	SYSCTL_ADD_T4_REG64(pi, "tx_bcast_frames", "# of broadcast frames",
6397 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_BCAST_L));
6398 	SYSCTL_ADD_T4_REG64(pi, "tx_mcast_frames", "# of multicast frames",
6399 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_MCAST_L));
6400 	SYSCTL_ADD_T4_REG64(pi, "tx_ucast_frames", "# of unicast frames",
6401 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_UCAST_L));
6402 	SYSCTL_ADD_T4_REG64(pi, "tx_error_frames", "# of error frames",
6403 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_ERROR_L));
6404 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_64",
6405 	    "# of tx frames in this range",
6406 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_64B_L));
6407 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_65_127",
6408 	    "# of tx frames in this range",
6409 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_65B_127B_L));
6410 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_128_255",
6411 	    "# of tx frames in this range",
6412 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_128B_255B_L));
6413 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_256_511",
6414 	    "# of tx frames in this range",
6415 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_256B_511B_L));
6416 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_512_1023",
6417 	    "# of tx frames in this range",
6418 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_512B_1023B_L));
6419 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_1024_1518",
6420 	    "# of tx frames in this range",
6421 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_1024B_1518B_L));
6422 	SYSCTL_ADD_T4_REG64(pi, "tx_frames_1519_max",
6423 	    "# of tx frames in this range",
6424 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_1519B_MAX_L));
6425 	SYSCTL_ADD_T4_REG64(pi, "tx_drop", "# of dropped tx frames",
6426 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_DROP_L));
6427 	SYSCTL_ADD_T4_REG64(pi, "tx_pause", "# of pause frames transmitted",
6428 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PAUSE_L));
6429 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp0", "# of PPP prio 0 frames transmitted",
6430 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP0_L));
6431 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp1", "# of PPP prio 1 frames transmitted",
6432 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP1_L));
6433 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp2", "# of PPP prio 2 frames transmitted",
6434 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP2_L));
6435 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp3", "# of PPP prio 3 frames transmitted",
6436 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP3_L));
6437 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp4", "# of PPP prio 4 frames transmitted",
6438 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP4_L));
6439 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp5", "# of PPP prio 5 frames transmitted",
6440 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP5_L));
6441 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp6", "# of PPP prio 6 frames transmitted",
6442 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP6_L));
6443 	SYSCTL_ADD_T4_REG64(pi, "tx_ppp7", "# of PPP prio 7 frames transmitted",
6444 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_TX_PORT_PPP7_L));
6445 
6446 	SYSCTL_ADD_T4_REG64(pi, "rx_octets", "# of octets in good frames",
6447 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_BYTES_L));
6448 	SYSCTL_ADD_T4_REG64(pi, "rx_frames", "total # of good frames",
6449 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_FRAMES_L));
6450 	SYSCTL_ADD_T4_REG64(pi, "rx_bcast_frames", "# of broadcast frames",
6451 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_BCAST_L));
6452 	SYSCTL_ADD_T4_REG64(pi, "rx_mcast_frames", "# of multicast frames",
6453 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MCAST_L));
6454 	SYSCTL_ADD_T4_REG64(pi, "rx_ucast_frames", "# of unicast frames",
6455 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_UCAST_L));
6456 	SYSCTL_ADD_T4_REG64(pi, "rx_too_long", "# of frames exceeding MTU",
6457 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MTU_ERROR_L));
6458 	SYSCTL_ADD_T4_REG64(pi, "rx_jabber", "# of jabber frames",
6459 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_MTU_CRC_ERROR_L));
6460 	SYSCTL_ADD_T4_REG64(pi, "rx_fcs_err",
6461 	    "# of frames received with bad FCS",
6462 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_CRC_ERROR_L));
6463 	SYSCTL_ADD_T4_REG64(pi, "rx_len_err",
6464 	    "# of frames received with length error",
6465 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_LEN_ERROR_L));
6466 	SYSCTL_ADD_T4_REG64(pi, "rx_symbol_err", "symbol errors",
6467 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_SYM_ERROR_L));
6468 	SYSCTL_ADD_T4_REG64(pi, "rx_runt", "# of short frames received",
6469 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_LESS_64B_L));
6470 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_64",
6471 	    "# of rx frames in this range",
6472 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_64B_L));
6473 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_65_127",
6474 	    "# of rx frames in this range",
6475 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_65B_127B_L));
6476 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_128_255",
6477 	    "# of rx frames in this range",
6478 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_128B_255B_L));
6479 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_256_511",
6480 	    "# of rx frames in this range",
6481 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_256B_511B_L));
6482 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_512_1023",
6483 	    "# of rx frames in this range",
6484 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_512B_1023B_L));
6485 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_1024_1518",
6486 	    "# of rx frames in this range",
6487 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_1024B_1518B_L));
6488 	SYSCTL_ADD_T4_REG64(pi, "rx_frames_1519_max",
6489 	    "# of rx frames in this range",
6490 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_1519B_MAX_L));
6491 	SYSCTL_ADD_T4_REG64(pi, "rx_pause", "# of pause frames received",
6492 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PAUSE_L));
6493 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp0", "# of PPP prio 0 frames received",
6494 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP0_L));
6495 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp1", "# of PPP prio 1 frames received",
6496 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP1_L));
6497 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp2", "# of PPP prio 2 frames received",
6498 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP2_L));
6499 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp3", "# of PPP prio 3 frames received",
6500 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP3_L));
6501 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp4", "# of PPP prio 4 frames received",
6502 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP4_L));
6503 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp5", "# of PPP prio 5 frames received",
6504 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP5_L));
6505 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp6", "# of PPP prio 6 frames received",
6506 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP6_L));
6507 	SYSCTL_ADD_T4_REG64(pi, "rx_ppp7", "# of PPP prio 7 frames received",
6508 	    PORT_REG(pi->tx_chan, A_MPS_PORT_STAT_RX_PORT_PPP7_L));
6509 
6510 #undef SYSCTL_ADD_T4_REG64
6511 
6512 #define SYSCTL_ADD_T4_PORTSTAT(name, desc) \
6513 	SYSCTL_ADD_UQUAD(ctx, children, OID_AUTO, #name, CTLFLAG_RD, \
6514 	    &pi->stats.name, desc)
6515 
6516 	/* We get these from port_stats and they may be stale by up to 1s */
6517 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow0,
6518 	    "# drops due to buffer-group 0 overflows");
6519 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow1,
6520 	    "# drops due to buffer-group 1 overflows");
6521 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow2,
6522 	    "# drops due to buffer-group 2 overflows");
6523 	SYSCTL_ADD_T4_PORTSTAT(rx_ovflow3,
6524 	    "# drops due to buffer-group 3 overflows");
6525 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc0,
6526 	    "# of buffer-group 0 truncated packets");
6527 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc1,
6528 	    "# of buffer-group 1 truncated packets");
6529 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc2,
6530 	    "# of buffer-group 2 truncated packets");
6531 	SYSCTL_ADD_T4_PORTSTAT(rx_trunc3,
6532 	    "# of buffer-group 3 truncated packets");
6533 
6534 #undef SYSCTL_ADD_T4_PORTSTAT
6535 
6536 	SYSCTL_ADD_ULONG(ctx, children, OID_AUTO, "tx_tls_records",
6537 	    CTLFLAG_RD, &pi->tx_tls_records,
6538 	    "# of TLS records transmitted");
6539 	SYSCTL_ADD_ULONG(ctx, children, OID_AUTO, "tx_tls_octets",
6540 	    CTLFLAG_RD, &pi->tx_tls_octets,
6541 	    "# of payload octets in transmitted TLS records");
6542 	SYSCTL_ADD_ULONG(ctx, children, OID_AUTO, "rx_tls_records",
6543 	    CTLFLAG_RD, &pi->rx_tls_records,
6544 	    "# of TLS records received");
6545 	SYSCTL_ADD_ULONG(ctx, children, OID_AUTO, "rx_tls_octets",
6546 	    CTLFLAG_RD, &pi->rx_tls_octets,
6547 	    "# of payload octets in received TLS records");
6548 }
6549 
6550 static int
6551 sysctl_int_array(SYSCTL_HANDLER_ARGS)
6552 {
6553 	int rc, *i, space = 0;
6554 	struct sbuf sb;
6555 
6556 	sbuf_new_for_sysctl(&sb, NULL, 64, req);
6557 	for (i = arg1; arg2; arg2 -= sizeof(int), i++) {
6558 		if (space)
6559 			sbuf_printf(&sb, " ");
6560 		sbuf_printf(&sb, "%d", *i);
6561 		space = 1;
6562 	}
6563 	rc = sbuf_finish(&sb);
6564 	sbuf_delete(&sb);
6565 	return (rc);
6566 }
6567 
6568 static int
6569 sysctl_bitfield_8b(SYSCTL_HANDLER_ARGS)
6570 {
6571 	int rc;
6572 	struct sbuf *sb;
6573 
6574 	rc = sysctl_wire_old_buffer(req, 0);
6575 	if (rc != 0)
6576 		return(rc);
6577 
6578 	sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
6579 	if (sb == NULL)
6580 		return (ENOMEM);
6581 
6582 	sbuf_printf(sb, "%b", *(uint8_t *)(uintptr_t)arg2, (char *)arg1);
6583 	rc = sbuf_finish(sb);
6584 	sbuf_delete(sb);
6585 
6586 	return (rc);
6587 }
6588 
6589 static int
6590 sysctl_bitfield_16b(SYSCTL_HANDLER_ARGS)
6591 {
6592 	int rc;
6593 	struct sbuf *sb;
6594 
6595 	rc = sysctl_wire_old_buffer(req, 0);
6596 	if (rc != 0)
6597 		return(rc);
6598 
6599 	sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
6600 	if (sb == NULL)
6601 		return (ENOMEM);
6602 
6603 	sbuf_printf(sb, "%b", *(uint16_t *)(uintptr_t)arg2, (char *)arg1);
6604 	rc = sbuf_finish(sb);
6605 	sbuf_delete(sb);
6606 
6607 	return (rc);
6608 }
6609 
6610 static int
6611 sysctl_btphy(SYSCTL_HANDLER_ARGS)
6612 {
6613 	struct port_info *pi = arg1;
6614 	int op = arg2;
6615 	struct adapter *sc = pi->adapter;
6616 	u_int v;
6617 	int rc;
6618 
6619 	rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK, "t4btt");
6620 	if (rc)
6621 		return (rc);
6622 	/* XXX: magic numbers */
6623 	rc = -t4_mdio_rd(sc, sc->mbox, pi->mdio_addr, 0x1e, op ? 0x20 : 0xc820,
6624 	    &v);
6625 	end_synchronized_op(sc, 0);
6626 	if (rc)
6627 		return (rc);
6628 	if (op == 0)
6629 		v /= 256;
6630 
6631 	rc = sysctl_handle_int(oidp, &v, 0, req);
6632 	return (rc);
6633 }
6634 
6635 static int
6636 sysctl_noflowq(SYSCTL_HANDLER_ARGS)
6637 {
6638 	struct vi_info *vi = arg1;
6639 	int rc, val;
6640 
6641 	val = vi->rsrv_noflowq;
6642 	rc = sysctl_handle_int(oidp, &val, 0, req);
6643 	if (rc != 0 || req->newptr == NULL)
6644 		return (rc);
6645 
6646 	if ((val >= 1) && (vi->ntxq > 1))
6647 		vi->rsrv_noflowq = 1;
6648 	else
6649 		vi->rsrv_noflowq = 0;
6650 
6651 	return (rc);
6652 }
6653 
6654 static int
6655 sysctl_holdoff_tmr_idx(SYSCTL_HANDLER_ARGS)
6656 {
6657 	struct vi_info *vi = arg1;
6658 	struct adapter *sc = vi->pi->adapter;
6659 	int idx, rc, i;
6660 	struct sge_rxq *rxq;
6661 	uint8_t v;
6662 
6663 	idx = vi->tmr_idx;
6664 
6665 	rc = sysctl_handle_int(oidp, &idx, 0, req);
6666 	if (rc != 0 || req->newptr == NULL)
6667 		return (rc);
6668 
6669 	if (idx < 0 || idx >= SGE_NTIMERS)
6670 		return (EINVAL);
6671 
6672 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
6673 	    "t4tmr");
6674 	if (rc)
6675 		return (rc);
6676 
6677 	v = V_QINTR_TIMER_IDX(idx) | V_QINTR_CNT_EN(vi->pktc_idx != -1);
6678 	for_each_rxq(vi, i, rxq) {
6679 #ifdef atomic_store_rel_8
6680 		atomic_store_rel_8(&rxq->iq.intr_params, v);
6681 #else
6682 		rxq->iq.intr_params = v;
6683 #endif
6684 	}
6685 	vi->tmr_idx = idx;
6686 
6687 	end_synchronized_op(sc, LOCK_HELD);
6688 	return (0);
6689 }
6690 
6691 static int
6692 sysctl_holdoff_pktc_idx(SYSCTL_HANDLER_ARGS)
6693 {
6694 	struct vi_info *vi = arg1;
6695 	struct adapter *sc = vi->pi->adapter;
6696 	int idx, rc;
6697 
6698 	idx = vi->pktc_idx;
6699 
6700 	rc = sysctl_handle_int(oidp, &idx, 0, req);
6701 	if (rc != 0 || req->newptr == NULL)
6702 		return (rc);
6703 
6704 	if (idx < -1 || idx >= SGE_NCOUNTERS)
6705 		return (EINVAL);
6706 
6707 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
6708 	    "t4pktc");
6709 	if (rc)
6710 		return (rc);
6711 
6712 	if (vi->flags & VI_INIT_DONE)
6713 		rc = EBUSY; /* cannot be changed once the queues are created */
6714 	else
6715 		vi->pktc_idx = idx;
6716 
6717 	end_synchronized_op(sc, LOCK_HELD);
6718 	return (rc);
6719 }
6720 
6721 static int
6722 sysctl_qsize_rxq(SYSCTL_HANDLER_ARGS)
6723 {
6724 	struct vi_info *vi = arg1;
6725 	struct adapter *sc = vi->pi->adapter;
6726 	int qsize, rc;
6727 
6728 	qsize = vi->qsize_rxq;
6729 
6730 	rc = sysctl_handle_int(oidp, &qsize, 0, req);
6731 	if (rc != 0 || req->newptr == NULL)
6732 		return (rc);
6733 
6734 	if (qsize < 128 || (qsize & 7))
6735 		return (EINVAL);
6736 
6737 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
6738 	    "t4rxqs");
6739 	if (rc)
6740 		return (rc);
6741 
6742 	if (vi->flags & VI_INIT_DONE)
6743 		rc = EBUSY; /* cannot be changed once the queues are created */
6744 	else
6745 		vi->qsize_rxq = qsize;
6746 
6747 	end_synchronized_op(sc, LOCK_HELD);
6748 	return (rc);
6749 }
6750 
6751 static int
6752 sysctl_qsize_txq(SYSCTL_HANDLER_ARGS)
6753 {
6754 	struct vi_info *vi = arg1;
6755 	struct adapter *sc = vi->pi->adapter;
6756 	int qsize, rc;
6757 
6758 	qsize = vi->qsize_txq;
6759 
6760 	rc = sysctl_handle_int(oidp, &qsize, 0, req);
6761 	if (rc != 0 || req->newptr == NULL)
6762 		return (rc);
6763 
6764 	if (qsize < 128 || qsize > 65536)
6765 		return (EINVAL);
6766 
6767 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
6768 	    "t4txqs");
6769 	if (rc)
6770 		return (rc);
6771 
6772 	if (vi->flags & VI_INIT_DONE)
6773 		rc = EBUSY; /* cannot be changed once the queues are created */
6774 	else
6775 		vi->qsize_txq = qsize;
6776 
6777 	end_synchronized_op(sc, LOCK_HELD);
6778 	return (rc);
6779 }
6780 
6781 static int
6782 sysctl_pause_settings(SYSCTL_HANDLER_ARGS)
6783 {
6784 	struct port_info *pi = arg1;
6785 	struct adapter *sc = pi->adapter;
6786 	struct link_config *lc = &pi->link_cfg;
6787 	int rc;
6788 
6789 	if (req->newptr == NULL) {
6790 		struct sbuf *sb;
6791 		static char *bits = "\20\1RX\2TX\3AUTO";
6792 
6793 		rc = sysctl_wire_old_buffer(req, 0);
6794 		if (rc != 0)
6795 			return(rc);
6796 
6797 		sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
6798 		if (sb == NULL)
6799 			return (ENOMEM);
6800 
6801 		if (lc->link_ok) {
6802 			sbuf_printf(sb, "%b", (lc->fc & (PAUSE_TX | PAUSE_RX)) |
6803 			    (lc->requested_fc & PAUSE_AUTONEG), bits);
6804 		} else {
6805 			sbuf_printf(sb, "%b", lc->requested_fc & (PAUSE_TX |
6806 			    PAUSE_RX | PAUSE_AUTONEG), bits);
6807 		}
6808 		rc = sbuf_finish(sb);
6809 		sbuf_delete(sb);
6810 	} else {
6811 		char s[2];
6812 		int n;
6813 
6814 		s[0] = '0' + (lc->requested_fc & (PAUSE_TX | PAUSE_RX |
6815 		    PAUSE_AUTONEG));
6816 		s[1] = 0;
6817 
6818 		rc = sysctl_handle_string(oidp, s, sizeof(s), req);
6819 		if (rc != 0)
6820 			return(rc);
6821 
6822 		if (s[1] != 0)
6823 			return (EINVAL);
6824 		if (s[0] < '0' || s[0] > '9')
6825 			return (EINVAL);	/* not a number */
6826 		n = s[0] - '0';
6827 		if (n & ~(PAUSE_TX | PAUSE_RX | PAUSE_AUTONEG))
6828 			return (EINVAL);	/* some other bit is set too */
6829 
6830 		rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
6831 		    "t4PAUSE");
6832 		if (rc)
6833 			return (rc);
6834 		PORT_LOCK(pi);
6835 		lc->requested_fc = n;
6836 		fixup_link_config(pi);
6837 		if (pi->up_vis > 0)
6838 			rc = apply_link_config(pi);
6839 		set_current_media(pi);
6840 		PORT_UNLOCK(pi);
6841 		end_synchronized_op(sc, 0);
6842 	}
6843 
6844 	return (rc);
6845 }
6846 
6847 static int
6848 sysctl_fec(SYSCTL_HANDLER_ARGS)
6849 {
6850 	struct port_info *pi = arg1;
6851 	struct adapter *sc = pi->adapter;
6852 	struct link_config *lc = &pi->link_cfg;
6853 	int rc;
6854 	int8_t old;
6855 
6856 	if (req->newptr == NULL) {
6857 		struct sbuf *sb;
6858 		static char *bits = "\20\1RS\2BASE-R\3RSVD1\4RSVD2\5RSVD3\6AUTO";
6859 
6860 		rc = sysctl_wire_old_buffer(req, 0);
6861 		if (rc != 0)
6862 			return(rc);
6863 
6864 		sb = sbuf_new_for_sysctl(NULL, NULL, 128, req);
6865 		if (sb == NULL)
6866 			return (ENOMEM);
6867 
6868 		/*
6869 		 * Display the requested_fec when the link is down -- the actual
6870 		 * FEC makes sense only when the link is up.
6871 		 */
6872 		if (lc->link_ok) {
6873 			sbuf_printf(sb, "%b", (lc->fec & M_FW_PORT_CAP32_FEC) |
6874 			    (lc->requested_fec & FEC_AUTO), bits);
6875 		} else {
6876 			sbuf_printf(sb, "%b", lc->requested_fec, bits);
6877 		}
6878 		rc = sbuf_finish(sb);
6879 		sbuf_delete(sb);
6880 	} else {
6881 		char s[3];
6882 		int n;
6883 
6884 		snprintf(s, sizeof(s), "%d",
6885 		    lc->requested_fec == FEC_AUTO ? -1 :
6886 		    lc->requested_fec & M_FW_PORT_CAP32_FEC);
6887 
6888 		rc = sysctl_handle_string(oidp, s, sizeof(s), req);
6889 		if (rc != 0)
6890 			return(rc);
6891 
6892 		n = strtol(&s[0], NULL, 0);
6893 		if (n < 0 || n & FEC_AUTO)
6894 			n = FEC_AUTO;
6895 		else {
6896 			if (n & ~M_FW_PORT_CAP32_FEC)
6897 				return (EINVAL);/* some other bit is set too */
6898 			if (!powerof2(n))
6899 				return (EINVAL);/* one bit can be set at most */
6900 		}
6901 
6902 		rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
6903 		    "t4fec");
6904 		if (rc)
6905 			return (rc);
6906 		PORT_LOCK(pi);
6907 		old = lc->requested_fec;
6908 		if (n == FEC_AUTO)
6909 			lc->requested_fec = FEC_AUTO;
6910 		else if (n == 0)
6911 			lc->requested_fec = FEC_NONE;
6912 		else {
6913 			if ((lc->supported | V_FW_PORT_CAP32_FEC(n)) !=
6914 			    lc->supported) {
6915 				rc = ENOTSUP;
6916 				goto done;
6917 			}
6918 			lc->requested_fec = n;
6919 		}
6920 		fixup_link_config(pi);
6921 		if (pi->up_vis > 0) {
6922 			rc = apply_link_config(pi);
6923 			if (rc != 0) {
6924 				lc->requested_fec = old;
6925 				if (rc == FW_EPROTO)
6926 					rc = ENOTSUP;
6927 			}
6928 		}
6929 done:
6930 		PORT_UNLOCK(pi);
6931 		end_synchronized_op(sc, 0);
6932 	}
6933 
6934 	return (rc);
6935 }
6936 
6937 static int
6938 sysctl_autoneg(SYSCTL_HANDLER_ARGS)
6939 {
6940 	struct port_info *pi = arg1;
6941 	struct adapter *sc = pi->adapter;
6942 	struct link_config *lc = &pi->link_cfg;
6943 	int rc, val;
6944 
6945 	if (lc->supported & FW_PORT_CAP32_ANEG)
6946 		val = lc->requested_aneg == AUTONEG_DISABLE ? 0 : 1;
6947 	else
6948 		val = -1;
6949 	rc = sysctl_handle_int(oidp, &val, 0, req);
6950 	if (rc != 0 || req->newptr == NULL)
6951 		return (rc);
6952 	if (val == 0)
6953 		val = AUTONEG_DISABLE;
6954 	else if (val == 1)
6955 		val = AUTONEG_ENABLE;
6956 	else
6957 		val = AUTONEG_AUTO;
6958 
6959 	rc = begin_synchronized_op(sc, &pi->vi[0], SLEEP_OK | INTR_OK,
6960 	    "t4aneg");
6961 	if (rc)
6962 		return (rc);
6963 	PORT_LOCK(pi);
6964 	if (val == AUTONEG_ENABLE && !(lc->supported & FW_PORT_CAP32_ANEG)) {
6965 		rc = ENOTSUP;
6966 		goto done;
6967 	}
6968 	lc->requested_aneg = val;
6969 	fixup_link_config(pi);
6970 	if (pi->up_vis > 0)
6971 		rc = apply_link_config(pi);
6972 	set_current_media(pi);
6973 done:
6974 	PORT_UNLOCK(pi);
6975 	end_synchronized_op(sc, 0);
6976 	return (rc);
6977 }
6978 
6979 static int
6980 sysctl_handle_t4_reg64(SYSCTL_HANDLER_ARGS)
6981 {
6982 	struct adapter *sc = arg1;
6983 	int reg = arg2;
6984 	uint64_t val;
6985 
6986 	val = t4_read_reg64(sc, reg);
6987 
6988 	return (sysctl_handle_64(oidp, &val, 0, req));
6989 }
6990 
6991 static int
6992 sysctl_temperature(SYSCTL_HANDLER_ARGS)
6993 {
6994 	struct adapter *sc = arg1;
6995 	int rc, t;
6996 	uint32_t param, val;
6997 
6998 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4temp");
6999 	if (rc)
7000 		return (rc);
7001 	param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
7002 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_DIAG) |
7003 	    V_FW_PARAMS_PARAM_Y(FW_PARAM_DEV_DIAG_TMP);
7004 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
7005 	end_synchronized_op(sc, 0);
7006 	if (rc)
7007 		return (rc);
7008 
7009 	/* unknown is returned as 0 but we display -1 in that case */
7010 	t = val == 0 ? -1 : val;
7011 
7012 	rc = sysctl_handle_int(oidp, &t, 0, req);
7013 	return (rc);
7014 }
7015 
7016 static int
7017 sysctl_loadavg(SYSCTL_HANDLER_ARGS)
7018 {
7019 	struct adapter *sc = arg1;
7020 	struct sbuf *sb;
7021 	int rc;
7022 	uint32_t param, val;
7023 
7024 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4lavg");
7025 	if (rc)
7026 		return (rc);
7027 	param = V_FW_PARAMS_MNEM(FW_PARAMS_MNEM_DEV) |
7028 	    V_FW_PARAMS_PARAM_X(FW_PARAMS_PARAM_DEV_LOAD);
7029 	rc = -t4_query_params(sc, sc->mbox, sc->pf, 0, 1, &param, &val);
7030 	end_synchronized_op(sc, 0);
7031 	if (rc)
7032 		return (rc);
7033 
7034 	rc = sysctl_wire_old_buffer(req, 0);
7035 	if (rc != 0)
7036 		return (rc);
7037 
7038 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7039 	if (sb == NULL)
7040 		return (ENOMEM);
7041 
7042 	if (val == 0xffffffff) {
7043 		/* Only debug and custom firmwares report load averages. */
7044 		sbuf_printf(sb, "not available");
7045 	} else {
7046 		sbuf_printf(sb, "%d %d %d", val & 0xff, (val >> 8) & 0xff,
7047 		    (val >> 16) & 0xff);
7048 	}
7049 	rc = sbuf_finish(sb);
7050 	sbuf_delete(sb);
7051 
7052 	return (rc);
7053 }
7054 
7055 static int
7056 sysctl_cctrl(SYSCTL_HANDLER_ARGS)
7057 {
7058 	struct adapter *sc = arg1;
7059 	struct sbuf *sb;
7060 	int rc, i;
7061 	uint16_t incr[NMTUS][NCCTRL_WIN];
7062 	static const char *dec_fac[] = {
7063 		"0.5", "0.5625", "0.625", "0.6875", "0.75", "0.8125", "0.875",
7064 		"0.9375"
7065 	};
7066 
7067 	rc = sysctl_wire_old_buffer(req, 0);
7068 	if (rc != 0)
7069 		return (rc);
7070 
7071 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7072 	if (sb == NULL)
7073 		return (ENOMEM);
7074 
7075 	t4_read_cong_tbl(sc, incr);
7076 
7077 	for (i = 0; i < NCCTRL_WIN; ++i) {
7078 		sbuf_printf(sb, "%2d: %4u %4u %4u %4u %4u %4u %4u %4u\n", i,
7079 		    incr[0][i], incr[1][i], incr[2][i], incr[3][i], incr[4][i],
7080 		    incr[5][i], incr[6][i], incr[7][i]);
7081 		sbuf_printf(sb, "%8u %4u %4u %4u %4u %4u %4u %4u %5u %s\n",
7082 		    incr[8][i], incr[9][i], incr[10][i], incr[11][i],
7083 		    incr[12][i], incr[13][i], incr[14][i], incr[15][i],
7084 		    sc->params.a_wnd[i], dec_fac[sc->params.b_wnd[i]]);
7085 	}
7086 
7087 	rc = sbuf_finish(sb);
7088 	sbuf_delete(sb);
7089 
7090 	return (rc);
7091 }
7092 
7093 static const char *qname[CIM_NUM_IBQ + CIM_NUM_OBQ_T5] = {
7094 	"TP0", "TP1", "ULP", "SGE0", "SGE1", "NC-SI",	/* ibq's */
7095 	"ULP0", "ULP1", "ULP2", "ULP3", "SGE", "NC-SI",	/* obq's */
7096 	"SGE0-RX", "SGE1-RX"	/* additional obq's (T5 onwards) */
7097 };
7098 
7099 static int
7100 sysctl_cim_ibq_obq(SYSCTL_HANDLER_ARGS)
7101 {
7102 	struct adapter *sc = arg1;
7103 	struct sbuf *sb;
7104 	int rc, i, n, qid = arg2;
7105 	uint32_t *buf, *p;
7106 	char *qtype;
7107 	u_int cim_num_obq = sc->chip_params->cim_num_obq;
7108 
7109 	KASSERT(qid >= 0 && qid < CIM_NUM_IBQ + cim_num_obq,
7110 	    ("%s: bad qid %d\n", __func__, qid));
7111 
7112 	if (qid < CIM_NUM_IBQ) {
7113 		/* inbound queue */
7114 		qtype = "IBQ";
7115 		n = 4 * CIM_IBQ_SIZE;
7116 		buf = malloc(n * sizeof(uint32_t), M_CXGBE, M_ZERO | M_WAITOK);
7117 		rc = t4_read_cim_ibq(sc, qid, buf, n);
7118 	} else {
7119 		/* outbound queue */
7120 		qtype = "OBQ";
7121 		qid -= CIM_NUM_IBQ;
7122 		n = 4 * cim_num_obq * CIM_OBQ_SIZE;
7123 		buf = malloc(n * sizeof(uint32_t), M_CXGBE, M_ZERO | M_WAITOK);
7124 		rc = t4_read_cim_obq(sc, qid, buf, n);
7125 	}
7126 
7127 	if (rc < 0) {
7128 		rc = -rc;
7129 		goto done;
7130 	}
7131 	n = rc * sizeof(uint32_t);	/* rc has # of words actually read */
7132 
7133 	rc = sysctl_wire_old_buffer(req, 0);
7134 	if (rc != 0)
7135 		goto done;
7136 
7137 	sb = sbuf_new_for_sysctl(NULL, NULL, PAGE_SIZE, req);
7138 	if (sb == NULL) {
7139 		rc = ENOMEM;
7140 		goto done;
7141 	}
7142 
7143 	sbuf_printf(sb, "%s%d %s", qtype , qid, qname[arg2]);
7144 	for (i = 0, p = buf; i < n; i += 16, p += 4)
7145 		sbuf_printf(sb, "\n%#06x: %08x %08x %08x %08x", i, p[0], p[1],
7146 		    p[2], p[3]);
7147 
7148 	rc = sbuf_finish(sb);
7149 	sbuf_delete(sb);
7150 done:
7151 	free(buf, M_CXGBE);
7152 	return (rc);
7153 }
7154 
7155 static int
7156 sysctl_cim_la(SYSCTL_HANDLER_ARGS)
7157 {
7158 	struct adapter *sc = arg1;
7159 	u_int cfg;
7160 	struct sbuf *sb;
7161 	uint32_t *buf, *p;
7162 	int rc;
7163 
7164 	MPASS(chip_id(sc) <= CHELSIO_T5);
7165 
7166 	rc = -t4_cim_read(sc, A_UP_UP_DBG_LA_CFG, 1, &cfg);
7167 	if (rc != 0)
7168 		return (rc);
7169 
7170 	rc = sysctl_wire_old_buffer(req, 0);
7171 	if (rc != 0)
7172 		return (rc);
7173 
7174 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7175 	if (sb == NULL)
7176 		return (ENOMEM);
7177 
7178 	buf = malloc(sc->params.cim_la_size * sizeof(uint32_t), M_CXGBE,
7179 	    M_ZERO | M_WAITOK);
7180 
7181 	rc = -t4_cim_read_la(sc, buf, NULL);
7182 	if (rc != 0)
7183 		goto done;
7184 
7185 	sbuf_printf(sb, "Status   Data      PC%s",
7186 	    cfg & F_UPDBGLACAPTPCONLY ? "" :
7187 	    "     LS0Stat  LS0Addr             LS0Data");
7188 
7189 	for (p = buf; p <= &buf[sc->params.cim_la_size - 8]; p += 8) {
7190 		if (cfg & F_UPDBGLACAPTPCONLY) {
7191 			sbuf_printf(sb, "\n  %02x   %08x %08x", p[5] & 0xff,
7192 			    p[6], p[7]);
7193 			sbuf_printf(sb, "\n  %02x   %02x%06x %02x%06x",
7194 			    (p[3] >> 8) & 0xff, p[3] & 0xff, p[4] >> 8,
7195 			    p[4] & 0xff, p[5] >> 8);
7196 			sbuf_printf(sb, "\n  %02x   %x%07x %x%07x",
7197 			    (p[0] >> 4) & 0xff, p[0] & 0xf, p[1] >> 4,
7198 			    p[1] & 0xf, p[2] >> 4);
7199 		} else {
7200 			sbuf_printf(sb,
7201 			    "\n  %02x   %x%07x %x%07x %08x %08x "
7202 			    "%08x%08x%08x%08x",
7203 			    (p[0] >> 4) & 0xff, p[0] & 0xf, p[1] >> 4,
7204 			    p[1] & 0xf, p[2] >> 4, p[2] & 0xf, p[3], p[4], p[5],
7205 			    p[6], p[7]);
7206 		}
7207 	}
7208 
7209 	rc = sbuf_finish(sb);
7210 	sbuf_delete(sb);
7211 done:
7212 	free(buf, M_CXGBE);
7213 	return (rc);
7214 }
7215 
7216 static int
7217 sysctl_cim_la_t6(SYSCTL_HANDLER_ARGS)
7218 {
7219 	struct adapter *sc = arg1;
7220 	u_int cfg;
7221 	struct sbuf *sb;
7222 	uint32_t *buf, *p;
7223 	int rc;
7224 
7225 	MPASS(chip_id(sc) > CHELSIO_T5);
7226 
7227 	rc = -t4_cim_read(sc, A_UP_UP_DBG_LA_CFG, 1, &cfg);
7228 	if (rc != 0)
7229 		return (rc);
7230 
7231 	rc = sysctl_wire_old_buffer(req, 0);
7232 	if (rc != 0)
7233 		return (rc);
7234 
7235 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7236 	if (sb == NULL)
7237 		return (ENOMEM);
7238 
7239 	buf = malloc(sc->params.cim_la_size * sizeof(uint32_t), M_CXGBE,
7240 	    M_ZERO | M_WAITOK);
7241 
7242 	rc = -t4_cim_read_la(sc, buf, NULL);
7243 	if (rc != 0)
7244 		goto done;
7245 
7246 	sbuf_printf(sb, "Status   Inst    Data      PC%s",
7247 	    cfg & F_UPDBGLACAPTPCONLY ? "" :
7248 	    "     LS0Stat  LS0Addr  LS0Data  LS1Stat  LS1Addr  LS1Data");
7249 
7250 	for (p = buf; p <= &buf[sc->params.cim_la_size - 10]; p += 10) {
7251 		if (cfg & F_UPDBGLACAPTPCONLY) {
7252 			sbuf_printf(sb, "\n  %02x   %08x %08x %08x",
7253 			    p[3] & 0xff, p[2], p[1], p[0]);
7254 			sbuf_printf(sb, "\n  %02x   %02x%06x %02x%06x %02x%06x",
7255 			    (p[6] >> 8) & 0xff, p[6] & 0xff, p[5] >> 8,
7256 			    p[5] & 0xff, p[4] >> 8, p[4] & 0xff, p[3] >> 8);
7257 			sbuf_printf(sb, "\n  %02x   %04x%04x %04x%04x %04x%04x",
7258 			    (p[9] >> 16) & 0xff, p[9] & 0xffff, p[8] >> 16,
7259 			    p[8] & 0xffff, p[7] >> 16, p[7] & 0xffff,
7260 			    p[6] >> 16);
7261 		} else {
7262 			sbuf_printf(sb, "\n  %02x   %04x%04x %04x%04x %04x%04x "
7263 			    "%08x %08x %08x %08x %08x %08x",
7264 			    (p[9] >> 16) & 0xff,
7265 			    p[9] & 0xffff, p[8] >> 16,
7266 			    p[8] & 0xffff, p[7] >> 16,
7267 			    p[7] & 0xffff, p[6] >> 16,
7268 			    p[2], p[1], p[0], p[5], p[4], p[3]);
7269 		}
7270 	}
7271 
7272 	rc = sbuf_finish(sb);
7273 	sbuf_delete(sb);
7274 done:
7275 	free(buf, M_CXGBE);
7276 	return (rc);
7277 }
7278 
7279 static int
7280 sysctl_cim_ma_la(SYSCTL_HANDLER_ARGS)
7281 {
7282 	struct adapter *sc = arg1;
7283 	u_int i;
7284 	struct sbuf *sb;
7285 	uint32_t *buf, *p;
7286 	int rc;
7287 
7288 	rc = sysctl_wire_old_buffer(req, 0);
7289 	if (rc != 0)
7290 		return (rc);
7291 
7292 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7293 	if (sb == NULL)
7294 		return (ENOMEM);
7295 
7296 	buf = malloc(2 * CIM_MALA_SIZE * 5 * sizeof(uint32_t), M_CXGBE,
7297 	    M_ZERO | M_WAITOK);
7298 
7299 	t4_cim_read_ma_la(sc, buf, buf + 5 * CIM_MALA_SIZE);
7300 	p = buf;
7301 
7302 	for (i = 0; i < CIM_MALA_SIZE; i++, p += 5) {
7303 		sbuf_printf(sb, "\n%02x%08x%08x%08x%08x", p[4], p[3], p[2],
7304 		    p[1], p[0]);
7305 	}
7306 
7307 	sbuf_printf(sb, "\n\nCnt ID Tag UE       Data       RDY VLD");
7308 	for (i = 0; i < CIM_MALA_SIZE; i++, p += 5) {
7309 		sbuf_printf(sb, "\n%3u %2u  %x   %u %08x%08x  %u   %u",
7310 		    (p[2] >> 10) & 0xff, (p[2] >> 7) & 7,
7311 		    (p[2] >> 3) & 0xf, (p[2] >> 2) & 1,
7312 		    (p[1] >> 2) | ((p[2] & 3) << 30),
7313 		    (p[0] >> 2) | ((p[1] & 3) << 30), (p[0] >> 1) & 1,
7314 		    p[0] & 1);
7315 	}
7316 
7317 	rc = sbuf_finish(sb);
7318 	sbuf_delete(sb);
7319 	free(buf, M_CXGBE);
7320 	return (rc);
7321 }
7322 
7323 static int
7324 sysctl_cim_pif_la(SYSCTL_HANDLER_ARGS)
7325 {
7326 	struct adapter *sc = arg1;
7327 	u_int i;
7328 	struct sbuf *sb;
7329 	uint32_t *buf, *p;
7330 	int rc;
7331 
7332 	rc = sysctl_wire_old_buffer(req, 0);
7333 	if (rc != 0)
7334 		return (rc);
7335 
7336 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7337 	if (sb == NULL)
7338 		return (ENOMEM);
7339 
7340 	buf = malloc(2 * CIM_PIFLA_SIZE * 6 * sizeof(uint32_t), M_CXGBE,
7341 	    M_ZERO | M_WAITOK);
7342 
7343 	t4_cim_read_pif_la(sc, buf, buf + 6 * CIM_PIFLA_SIZE, NULL, NULL);
7344 	p = buf;
7345 
7346 	sbuf_printf(sb, "Cntl ID DataBE   Addr                 Data");
7347 	for (i = 0; i < CIM_PIFLA_SIZE; i++, p += 6) {
7348 		sbuf_printf(sb, "\n %02x  %02x  %04x  %08x %08x%08x%08x%08x",
7349 		    (p[5] >> 22) & 0xff, (p[5] >> 16) & 0x3f, p[5] & 0xffff,
7350 		    p[4], p[3], p[2], p[1], p[0]);
7351 	}
7352 
7353 	sbuf_printf(sb, "\n\nCntl ID               Data");
7354 	for (i = 0; i < CIM_PIFLA_SIZE; i++, p += 6) {
7355 		sbuf_printf(sb, "\n %02x  %02x %08x%08x%08x%08x",
7356 		    (p[4] >> 6) & 0xff, p[4] & 0x3f, p[3], p[2], p[1], p[0]);
7357 	}
7358 
7359 	rc = sbuf_finish(sb);
7360 	sbuf_delete(sb);
7361 	free(buf, M_CXGBE);
7362 	return (rc);
7363 }
7364 
7365 static int
7366 sysctl_cim_qcfg(SYSCTL_HANDLER_ARGS)
7367 {
7368 	struct adapter *sc = arg1;
7369 	struct sbuf *sb;
7370 	int rc, i;
7371 	uint16_t base[CIM_NUM_IBQ + CIM_NUM_OBQ_T5];
7372 	uint16_t size[CIM_NUM_IBQ + CIM_NUM_OBQ_T5];
7373 	uint16_t thres[CIM_NUM_IBQ];
7374 	uint32_t obq_wr[2 * CIM_NUM_OBQ_T5], *wr = obq_wr;
7375 	uint32_t stat[4 * (CIM_NUM_IBQ + CIM_NUM_OBQ_T5)], *p = stat;
7376 	u_int cim_num_obq, ibq_rdaddr, obq_rdaddr, nq;
7377 
7378 	cim_num_obq = sc->chip_params->cim_num_obq;
7379 	if (is_t4(sc)) {
7380 		ibq_rdaddr = A_UP_IBQ_0_RDADDR;
7381 		obq_rdaddr = A_UP_OBQ_0_REALADDR;
7382 	} else {
7383 		ibq_rdaddr = A_UP_IBQ_0_SHADOW_RDADDR;
7384 		obq_rdaddr = A_UP_OBQ_0_SHADOW_REALADDR;
7385 	}
7386 	nq = CIM_NUM_IBQ + cim_num_obq;
7387 
7388 	rc = -t4_cim_read(sc, ibq_rdaddr, 4 * nq, stat);
7389 	if (rc == 0)
7390 		rc = -t4_cim_read(sc, obq_rdaddr, 2 * cim_num_obq, obq_wr);
7391 	if (rc != 0)
7392 		return (rc);
7393 
7394 	t4_read_cimq_cfg(sc, base, size, thres);
7395 
7396 	rc = sysctl_wire_old_buffer(req, 0);
7397 	if (rc != 0)
7398 		return (rc);
7399 
7400 	sb = sbuf_new_for_sysctl(NULL, NULL, PAGE_SIZE, req);
7401 	if (sb == NULL)
7402 		return (ENOMEM);
7403 
7404 	sbuf_printf(sb,
7405 	    "  Queue  Base  Size Thres  RdPtr WrPtr  SOP  EOP Avail");
7406 
7407 	for (i = 0; i < CIM_NUM_IBQ; i++, p += 4)
7408 		sbuf_printf(sb, "\n%7s %5x %5u %5u %6x  %4x %4u %4u %5u",
7409 		    qname[i], base[i], size[i], thres[i], G_IBQRDADDR(p[0]),
7410 		    G_IBQWRADDR(p[1]), G_QUESOPCNT(p[3]), G_QUEEOPCNT(p[3]),
7411 		    G_QUEREMFLITS(p[2]) * 16);
7412 	for ( ; i < nq; i++, p += 4, wr += 2)
7413 		sbuf_printf(sb, "\n%7s %5x %5u %12x  %4x %4u %4u %5u", qname[i],
7414 		    base[i], size[i], G_QUERDADDR(p[0]) & 0x3fff,
7415 		    wr[0] - base[i], G_QUESOPCNT(p[3]), G_QUEEOPCNT(p[3]),
7416 		    G_QUEREMFLITS(p[2]) * 16);
7417 
7418 	rc = sbuf_finish(sb);
7419 	sbuf_delete(sb);
7420 
7421 	return (rc);
7422 }
7423 
7424 static int
7425 sysctl_cpl_stats(SYSCTL_HANDLER_ARGS)
7426 {
7427 	struct adapter *sc = arg1;
7428 	struct sbuf *sb;
7429 	int rc;
7430 	struct tp_cpl_stats stats;
7431 
7432 	rc = sysctl_wire_old_buffer(req, 0);
7433 	if (rc != 0)
7434 		return (rc);
7435 
7436 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7437 	if (sb == NULL)
7438 		return (ENOMEM);
7439 
7440 	mtx_lock(&sc->reg_lock);
7441 	t4_tp_get_cpl_stats(sc, &stats, 0);
7442 	mtx_unlock(&sc->reg_lock);
7443 
7444 	if (sc->chip_params->nchan > 2) {
7445 		sbuf_printf(sb, "                 channel 0  channel 1"
7446 		    "  channel 2  channel 3");
7447 		sbuf_printf(sb, "\nCPL requests:   %10u %10u %10u %10u",
7448 		    stats.req[0], stats.req[1], stats.req[2], stats.req[3]);
7449 		sbuf_printf(sb, "\nCPL responses:   %10u %10u %10u %10u",
7450 		    stats.rsp[0], stats.rsp[1], stats.rsp[2], stats.rsp[3]);
7451 	} else {
7452 		sbuf_printf(sb, "                 channel 0  channel 1");
7453 		sbuf_printf(sb, "\nCPL requests:   %10u %10u",
7454 		    stats.req[0], stats.req[1]);
7455 		sbuf_printf(sb, "\nCPL responses:   %10u %10u",
7456 		    stats.rsp[0], stats.rsp[1]);
7457 	}
7458 
7459 	rc = sbuf_finish(sb);
7460 	sbuf_delete(sb);
7461 
7462 	return (rc);
7463 }
7464 
7465 static int
7466 sysctl_ddp_stats(SYSCTL_HANDLER_ARGS)
7467 {
7468 	struct adapter *sc = arg1;
7469 	struct sbuf *sb;
7470 	int rc;
7471 	struct tp_usm_stats stats;
7472 
7473 	rc = sysctl_wire_old_buffer(req, 0);
7474 	if (rc != 0)
7475 		return(rc);
7476 
7477 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7478 	if (sb == NULL)
7479 		return (ENOMEM);
7480 
7481 	t4_get_usm_stats(sc, &stats, 1);
7482 
7483 	sbuf_printf(sb, "Frames: %u\n", stats.frames);
7484 	sbuf_printf(sb, "Octets: %ju\n", stats.octets);
7485 	sbuf_printf(sb, "Drops:  %u", stats.drops);
7486 
7487 	rc = sbuf_finish(sb);
7488 	sbuf_delete(sb);
7489 
7490 	return (rc);
7491 }
7492 
7493 static const char * const devlog_level_strings[] = {
7494 	[FW_DEVLOG_LEVEL_EMERG]		= "EMERG",
7495 	[FW_DEVLOG_LEVEL_CRIT]		= "CRIT",
7496 	[FW_DEVLOG_LEVEL_ERR]		= "ERR",
7497 	[FW_DEVLOG_LEVEL_NOTICE]	= "NOTICE",
7498 	[FW_DEVLOG_LEVEL_INFO]		= "INFO",
7499 	[FW_DEVLOG_LEVEL_DEBUG]		= "DEBUG"
7500 };
7501 
7502 static const char * const devlog_facility_strings[] = {
7503 	[FW_DEVLOG_FACILITY_CORE]	= "CORE",
7504 	[FW_DEVLOG_FACILITY_CF]		= "CF",
7505 	[FW_DEVLOG_FACILITY_SCHED]	= "SCHED",
7506 	[FW_DEVLOG_FACILITY_TIMER]	= "TIMER",
7507 	[FW_DEVLOG_FACILITY_RES]	= "RES",
7508 	[FW_DEVLOG_FACILITY_HW]		= "HW",
7509 	[FW_DEVLOG_FACILITY_FLR]	= "FLR",
7510 	[FW_DEVLOG_FACILITY_DMAQ]	= "DMAQ",
7511 	[FW_DEVLOG_FACILITY_PHY]	= "PHY",
7512 	[FW_DEVLOG_FACILITY_MAC]	= "MAC",
7513 	[FW_DEVLOG_FACILITY_PORT]	= "PORT",
7514 	[FW_DEVLOG_FACILITY_VI]		= "VI",
7515 	[FW_DEVLOG_FACILITY_FILTER]	= "FILTER",
7516 	[FW_DEVLOG_FACILITY_ACL]	= "ACL",
7517 	[FW_DEVLOG_FACILITY_TM]		= "TM",
7518 	[FW_DEVLOG_FACILITY_QFC]	= "QFC",
7519 	[FW_DEVLOG_FACILITY_DCB]	= "DCB",
7520 	[FW_DEVLOG_FACILITY_ETH]	= "ETH",
7521 	[FW_DEVLOG_FACILITY_OFLD]	= "OFLD",
7522 	[FW_DEVLOG_FACILITY_RI]		= "RI",
7523 	[FW_DEVLOG_FACILITY_ISCSI]	= "ISCSI",
7524 	[FW_DEVLOG_FACILITY_FCOE]	= "FCOE",
7525 	[FW_DEVLOG_FACILITY_FOISCSI]	= "FOISCSI",
7526 	[FW_DEVLOG_FACILITY_FOFCOE]	= "FOFCOE",
7527 	[FW_DEVLOG_FACILITY_CHNET]	= "CHNET",
7528 };
7529 
7530 static int
7531 sysctl_devlog(SYSCTL_HANDLER_ARGS)
7532 {
7533 	struct adapter *sc = arg1;
7534 	struct devlog_params *dparams = &sc->params.devlog;
7535 	struct fw_devlog_e *buf, *e;
7536 	int i, j, rc, nentries, first = 0;
7537 	struct sbuf *sb;
7538 	uint64_t ftstamp = UINT64_MAX;
7539 
7540 	if (dparams->addr == 0)
7541 		return (ENXIO);
7542 
7543 	buf = malloc(dparams->size, M_CXGBE, M_NOWAIT);
7544 	if (buf == NULL)
7545 		return (ENOMEM);
7546 
7547 	rc = read_via_memwin(sc, 1, dparams->addr, (void *)buf, dparams->size);
7548 	if (rc != 0)
7549 		goto done;
7550 
7551 	nentries = dparams->size / sizeof(struct fw_devlog_e);
7552 	for (i = 0; i < nentries; i++) {
7553 		e = &buf[i];
7554 
7555 		if (e->timestamp == 0)
7556 			break;	/* end */
7557 
7558 		e->timestamp = be64toh(e->timestamp);
7559 		e->seqno = be32toh(e->seqno);
7560 		for (j = 0; j < 8; j++)
7561 			e->params[j] = be32toh(e->params[j]);
7562 
7563 		if (e->timestamp < ftstamp) {
7564 			ftstamp = e->timestamp;
7565 			first = i;
7566 		}
7567 	}
7568 
7569 	if (buf[first].timestamp == 0)
7570 		goto done;	/* nothing in the log */
7571 
7572 	rc = sysctl_wire_old_buffer(req, 0);
7573 	if (rc != 0)
7574 		goto done;
7575 
7576 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7577 	if (sb == NULL) {
7578 		rc = ENOMEM;
7579 		goto done;
7580 	}
7581 	sbuf_printf(sb, "%10s  %15s  %8s  %8s  %s\n",
7582 	    "Seq#", "Tstamp", "Level", "Facility", "Message");
7583 
7584 	i = first;
7585 	do {
7586 		e = &buf[i];
7587 		if (e->timestamp == 0)
7588 			break;	/* end */
7589 
7590 		sbuf_printf(sb, "%10d  %15ju  %8s  %8s  ",
7591 		    e->seqno, e->timestamp,
7592 		    (e->level < nitems(devlog_level_strings) ?
7593 			devlog_level_strings[e->level] : "UNKNOWN"),
7594 		    (e->facility < nitems(devlog_facility_strings) ?
7595 			devlog_facility_strings[e->facility] : "UNKNOWN"));
7596 		sbuf_printf(sb, e->fmt, e->params[0], e->params[1],
7597 		    e->params[2], e->params[3], e->params[4],
7598 		    e->params[5], e->params[6], e->params[7]);
7599 
7600 		if (++i == nentries)
7601 			i = 0;
7602 	} while (i != first);
7603 
7604 	rc = sbuf_finish(sb);
7605 	sbuf_delete(sb);
7606 done:
7607 	free(buf, M_CXGBE);
7608 	return (rc);
7609 }
7610 
7611 static int
7612 sysctl_fcoe_stats(SYSCTL_HANDLER_ARGS)
7613 {
7614 	struct adapter *sc = arg1;
7615 	struct sbuf *sb;
7616 	int rc;
7617 	struct tp_fcoe_stats stats[MAX_NCHAN];
7618 	int i, nchan = sc->chip_params->nchan;
7619 
7620 	rc = sysctl_wire_old_buffer(req, 0);
7621 	if (rc != 0)
7622 		return (rc);
7623 
7624 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7625 	if (sb == NULL)
7626 		return (ENOMEM);
7627 
7628 	for (i = 0; i < nchan; i++)
7629 		t4_get_fcoe_stats(sc, i, &stats[i], 1);
7630 
7631 	if (nchan > 2) {
7632 		sbuf_printf(sb, "                   channel 0        channel 1"
7633 		    "        channel 2        channel 3");
7634 		sbuf_printf(sb, "\noctetsDDP:  %16ju %16ju %16ju %16ju",
7635 		    stats[0].octets_ddp, stats[1].octets_ddp,
7636 		    stats[2].octets_ddp, stats[3].octets_ddp);
7637 		sbuf_printf(sb, "\nframesDDP:  %16u %16u %16u %16u",
7638 		    stats[0].frames_ddp, stats[1].frames_ddp,
7639 		    stats[2].frames_ddp, stats[3].frames_ddp);
7640 		sbuf_printf(sb, "\nframesDrop: %16u %16u %16u %16u",
7641 		    stats[0].frames_drop, stats[1].frames_drop,
7642 		    stats[2].frames_drop, stats[3].frames_drop);
7643 	} else {
7644 		sbuf_printf(sb, "                   channel 0        channel 1");
7645 		sbuf_printf(sb, "\noctetsDDP:  %16ju %16ju",
7646 		    stats[0].octets_ddp, stats[1].octets_ddp);
7647 		sbuf_printf(sb, "\nframesDDP:  %16u %16u",
7648 		    stats[0].frames_ddp, stats[1].frames_ddp);
7649 		sbuf_printf(sb, "\nframesDrop: %16u %16u",
7650 		    stats[0].frames_drop, stats[1].frames_drop);
7651 	}
7652 
7653 	rc = sbuf_finish(sb);
7654 	sbuf_delete(sb);
7655 
7656 	return (rc);
7657 }
7658 
7659 static int
7660 sysctl_hw_sched(SYSCTL_HANDLER_ARGS)
7661 {
7662 	struct adapter *sc = arg1;
7663 	struct sbuf *sb;
7664 	int rc, i;
7665 	unsigned int map, kbps, ipg, mode;
7666 	unsigned int pace_tab[NTX_SCHED];
7667 
7668 	rc = sysctl_wire_old_buffer(req, 0);
7669 	if (rc != 0)
7670 		return (rc);
7671 
7672 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
7673 	if (sb == NULL)
7674 		return (ENOMEM);
7675 
7676 	map = t4_read_reg(sc, A_TP_TX_MOD_QUEUE_REQ_MAP);
7677 	mode = G_TIMERMODE(t4_read_reg(sc, A_TP_MOD_CONFIG));
7678 	t4_read_pace_tbl(sc, pace_tab);
7679 
7680 	sbuf_printf(sb, "Scheduler  Mode   Channel  Rate (Kbps)   "
7681 	    "Class IPG (0.1 ns)   Flow IPG (us)");
7682 
7683 	for (i = 0; i < NTX_SCHED; ++i, map >>= 2) {
7684 		t4_get_tx_sched(sc, i, &kbps, &ipg, 1);
7685 		sbuf_printf(sb, "\n    %u      %-5s     %u     ", i,
7686 		    (mode & (1 << i)) ? "flow" : "class", map & 3);
7687 		if (kbps)
7688 			sbuf_printf(sb, "%9u     ", kbps);
7689 		else
7690 			sbuf_printf(sb, " disabled     ");
7691 
7692 		if (ipg)
7693 			sbuf_printf(sb, "%13u        ", ipg);
7694 		else
7695 			sbuf_printf(sb, "     disabled        ");
7696 
7697 		if (pace_tab[i])
7698 			sbuf_printf(sb, "%10u", pace_tab[i]);
7699 		else
7700 			sbuf_printf(sb, "  disabled");
7701 	}
7702 
7703 	rc = sbuf_finish(sb);
7704 	sbuf_delete(sb);
7705 
7706 	return (rc);
7707 }
7708 
7709 static int
7710 sysctl_lb_stats(SYSCTL_HANDLER_ARGS)
7711 {
7712 	struct adapter *sc = arg1;
7713 	struct sbuf *sb;
7714 	int rc, i, j;
7715 	uint64_t *p0, *p1;
7716 	struct lb_port_stats s[2];
7717 	static const char *stat_name[] = {
7718 		"OctetsOK:", "FramesOK:", "BcastFrames:", "McastFrames:",
7719 		"UcastFrames:", "ErrorFrames:", "Frames64:", "Frames65To127:",
7720 		"Frames128To255:", "Frames256To511:", "Frames512To1023:",
7721 		"Frames1024To1518:", "Frames1519ToMax:", "FramesDropped:",
7722 		"BG0FramesDropped:", "BG1FramesDropped:", "BG2FramesDropped:",
7723 		"BG3FramesDropped:", "BG0FramesTrunc:", "BG1FramesTrunc:",
7724 		"BG2FramesTrunc:", "BG3FramesTrunc:"
7725 	};
7726 
7727 	rc = sysctl_wire_old_buffer(req, 0);
7728 	if (rc != 0)
7729 		return (rc);
7730 
7731 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7732 	if (sb == NULL)
7733 		return (ENOMEM);
7734 
7735 	memset(s, 0, sizeof(s));
7736 
7737 	for (i = 0; i < sc->chip_params->nchan; i += 2) {
7738 		t4_get_lb_stats(sc, i, &s[0]);
7739 		t4_get_lb_stats(sc, i + 1, &s[1]);
7740 
7741 		p0 = &s[0].octets;
7742 		p1 = &s[1].octets;
7743 		sbuf_printf(sb, "%s                       Loopback %u"
7744 		    "           Loopback %u", i == 0 ? "" : "\n", i, i + 1);
7745 
7746 		for (j = 0; j < nitems(stat_name); j++)
7747 			sbuf_printf(sb, "\n%-17s %20ju %20ju", stat_name[j],
7748 				   *p0++, *p1++);
7749 	}
7750 
7751 	rc = sbuf_finish(sb);
7752 	sbuf_delete(sb);
7753 
7754 	return (rc);
7755 }
7756 
7757 static int
7758 sysctl_linkdnrc(SYSCTL_HANDLER_ARGS)
7759 {
7760 	int rc = 0;
7761 	struct port_info *pi = arg1;
7762 	struct link_config *lc = &pi->link_cfg;
7763 	struct sbuf *sb;
7764 
7765 	rc = sysctl_wire_old_buffer(req, 0);
7766 	if (rc != 0)
7767 		return(rc);
7768 	sb = sbuf_new_for_sysctl(NULL, NULL, 64, req);
7769 	if (sb == NULL)
7770 		return (ENOMEM);
7771 
7772 	if (lc->link_ok || lc->link_down_rc == 255)
7773 		sbuf_printf(sb, "n/a");
7774 	else
7775 		sbuf_printf(sb, "%s", t4_link_down_rc_str(lc->link_down_rc));
7776 
7777 	rc = sbuf_finish(sb);
7778 	sbuf_delete(sb);
7779 
7780 	return (rc);
7781 }
7782 
7783 struct mem_desc {
7784 	unsigned int base;
7785 	unsigned int limit;
7786 	unsigned int idx;
7787 };
7788 
7789 static int
7790 mem_desc_cmp(const void *a, const void *b)
7791 {
7792 	return ((const struct mem_desc *)a)->base -
7793 	       ((const struct mem_desc *)b)->base;
7794 }
7795 
7796 static void
7797 mem_region_show(struct sbuf *sb, const char *name, unsigned int from,
7798     unsigned int to)
7799 {
7800 	unsigned int size;
7801 
7802 	if (from == to)
7803 		return;
7804 
7805 	size = to - from + 1;
7806 	if (size == 0)
7807 		return;
7808 
7809 	/* XXX: need humanize_number(3) in libkern for a more readable 'size' */
7810 	sbuf_printf(sb, "%-15s %#x-%#x [%u]\n", name, from, to, size);
7811 }
7812 
7813 static int
7814 sysctl_meminfo(SYSCTL_HANDLER_ARGS)
7815 {
7816 	struct adapter *sc = arg1;
7817 	struct sbuf *sb;
7818 	int rc, i, n;
7819 	uint32_t lo, hi, used, alloc;
7820 	static const char *memory[] = {"EDC0:", "EDC1:", "MC:", "MC0:", "MC1:"};
7821 	static const char *region[] = {
7822 		"DBQ contexts:", "IMSG contexts:", "FLM cache:", "TCBs:",
7823 		"Pstructs:", "Timers:", "Rx FL:", "Tx FL:", "Pstruct FL:",
7824 		"Tx payload:", "Rx payload:", "LE hash:", "iSCSI region:",
7825 		"TDDP region:", "TPT region:", "STAG region:", "RQ region:",
7826 		"RQUDP region:", "PBL region:", "TXPBL region:",
7827 		"DBVFIFO region:", "ULPRX state:", "ULPTX state:",
7828 		"On-chip queues:", "TLS keys:",
7829 	};
7830 	struct mem_desc avail[4];
7831 	struct mem_desc mem[nitems(region) + 3];	/* up to 3 holes */
7832 	struct mem_desc *md = mem;
7833 
7834 	rc = sysctl_wire_old_buffer(req, 0);
7835 	if (rc != 0)
7836 		return (rc);
7837 
7838 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
7839 	if (sb == NULL)
7840 		return (ENOMEM);
7841 
7842 	for (i = 0; i < nitems(mem); i++) {
7843 		mem[i].limit = 0;
7844 		mem[i].idx = i;
7845 	}
7846 
7847 	/* Find and sort the populated memory ranges */
7848 	i = 0;
7849 	lo = t4_read_reg(sc, A_MA_TARGET_MEM_ENABLE);
7850 	if (lo & F_EDRAM0_ENABLE) {
7851 		hi = t4_read_reg(sc, A_MA_EDRAM0_BAR);
7852 		avail[i].base = G_EDRAM0_BASE(hi) << 20;
7853 		avail[i].limit = avail[i].base + (G_EDRAM0_SIZE(hi) << 20);
7854 		avail[i].idx = 0;
7855 		i++;
7856 	}
7857 	if (lo & F_EDRAM1_ENABLE) {
7858 		hi = t4_read_reg(sc, A_MA_EDRAM1_BAR);
7859 		avail[i].base = G_EDRAM1_BASE(hi) << 20;
7860 		avail[i].limit = avail[i].base + (G_EDRAM1_SIZE(hi) << 20);
7861 		avail[i].idx = 1;
7862 		i++;
7863 	}
7864 	if (lo & F_EXT_MEM_ENABLE) {
7865 		hi = t4_read_reg(sc, A_MA_EXT_MEMORY_BAR);
7866 		avail[i].base = G_EXT_MEM_BASE(hi) << 20;
7867 		avail[i].limit = avail[i].base +
7868 		    (G_EXT_MEM_SIZE(hi) << 20);
7869 		avail[i].idx = is_t5(sc) ? 3 : 2;	/* Call it MC0 for T5 */
7870 		i++;
7871 	}
7872 	if (is_t5(sc) && lo & F_EXT_MEM1_ENABLE) {
7873 		hi = t4_read_reg(sc, A_MA_EXT_MEMORY1_BAR);
7874 		avail[i].base = G_EXT_MEM1_BASE(hi) << 20;
7875 		avail[i].limit = avail[i].base +
7876 		    (G_EXT_MEM1_SIZE(hi) << 20);
7877 		avail[i].idx = 4;
7878 		i++;
7879 	}
7880 	if (!i)                                    /* no memory available */
7881 		return 0;
7882 	qsort(avail, i, sizeof(struct mem_desc), mem_desc_cmp);
7883 
7884 	(md++)->base = t4_read_reg(sc, A_SGE_DBQ_CTXT_BADDR);
7885 	(md++)->base = t4_read_reg(sc, A_SGE_IMSG_CTXT_BADDR);
7886 	(md++)->base = t4_read_reg(sc, A_SGE_FLM_CACHE_BADDR);
7887 	(md++)->base = t4_read_reg(sc, A_TP_CMM_TCB_BASE);
7888 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_BASE);
7889 	(md++)->base = t4_read_reg(sc, A_TP_CMM_TIMER_BASE);
7890 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_RX_FLST_BASE);
7891 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_TX_FLST_BASE);
7892 	(md++)->base = t4_read_reg(sc, A_TP_CMM_MM_PS_FLST_BASE);
7893 
7894 	/* the next few have explicit upper bounds */
7895 	md->base = t4_read_reg(sc, A_TP_PMM_TX_BASE);
7896 	md->limit = md->base - 1 +
7897 		    t4_read_reg(sc, A_TP_PMM_TX_PAGE_SIZE) *
7898 		    G_PMTXMAXPAGE(t4_read_reg(sc, A_TP_PMM_TX_MAX_PAGE));
7899 	md++;
7900 
7901 	md->base = t4_read_reg(sc, A_TP_PMM_RX_BASE);
7902 	md->limit = md->base - 1 +
7903 		    t4_read_reg(sc, A_TP_PMM_RX_PAGE_SIZE) *
7904 		    G_PMRXMAXPAGE(t4_read_reg(sc, A_TP_PMM_RX_MAX_PAGE));
7905 	md++;
7906 
7907 	if (t4_read_reg(sc, A_LE_DB_CONFIG) & F_HASHEN) {
7908 		if (chip_id(sc) <= CHELSIO_T5)
7909 			md->base = t4_read_reg(sc, A_LE_DB_HASH_TID_BASE);
7910 		else
7911 			md->base = t4_read_reg(sc, A_LE_DB_HASH_TBL_BASE_ADDR);
7912 		md->limit = 0;
7913 	} else {
7914 		md->base = 0;
7915 		md->idx = nitems(region);  /* hide it */
7916 	}
7917 	md++;
7918 
7919 #define ulp_region(reg) \
7920 	md->base = t4_read_reg(sc, A_ULP_ ## reg ## _LLIMIT);\
7921 	(md++)->limit = t4_read_reg(sc, A_ULP_ ## reg ## _ULIMIT)
7922 
7923 	ulp_region(RX_ISCSI);
7924 	ulp_region(RX_TDDP);
7925 	ulp_region(TX_TPT);
7926 	ulp_region(RX_STAG);
7927 	ulp_region(RX_RQ);
7928 	ulp_region(RX_RQUDP);
7929 	ulp_region(RX_PBL);
7930 	ulp_region(TX_PBL);
7931 #undef ulp_region
7932 
7933 	md->base = 0;
7934 	md->idx = nitems(region);
7935 	if (!is_t4(sc)) {
7936 		uint32_t size = 0;
7937 		uint32_t sge_ctrl = t4_read_reg(sc, A_SGE_CONTROL2);
7938 		uint32_t fifo_size = t4_read_reg(sc, A_SGE_DBVFIFO_SIZE);
7939 
7940 		if (is_t5(sc)) {
7941 			if (sge_ctrl & F_VFIFO_ENABLE)
7942 				size = G_DBVFIFO_SIZE(fifo_size);
7943 		} else
7944 			size = G_T6_DBVFIFO_SIZE(fifo_size);
7945 
7946 		if (size) {
7947 			md->base = G_BASEADDR(t4_read_reg(sc,
7948 			    A_SGE_DBVFIFO_BADDR));
7949 			md->limit = md->base + (size << 2) - 1;
7950 		}
7951 	}
7952 	md++;
7953 
7954 	md->base = t4_read_reg(sc, A_ULP_RX_CTX_BASE);
7955 	md->limit = 0;
7956 	md++;
7957 	md->base = t4_read_reg(sc, A_ULP_TX_ERR_TABLE_BASE);
7958 	md->limit = 0;
7959 	md++;
7960 
7961 	md->base = sc->vres.ocq.start;
7962 	if (sc->vres.ocq.size)
7963 		md->limit = md->base + sc->vres.ocq.size - 1;
7964 	else
7965 		md->idx = nitems(region);  /* hide it */
7966 	md++;
7967 
7968 	md->base = sc->vres.key.start;
7969 	if (sc->vres.key.size)
7970 		md->limit = md->base + sc->vres.key.size - 1;
7971 	else
7972 		md->idx = nitems(region);  /* hide it */
7973 	md++;
7974 
7975 	/* add any address-space holes, there can be up to 3 */
7976 	for (n = 0; n < i - 1; n++)
7977 		if (avail[n].limit < avail[n + 1].base)
7978 			(md++)->base = avail[n].limit;
7979 	if (avail[n].limit)
7980 		(md++)->base = avail[n].limit;
7981 
7982 	n = md - mem;
7983 	qsort(mem, n, sizeof(struct mem_desc), mem_desc_cmp);
7984 
7985 	for (lo = 0; lo < i; lo++)
7986 		mem_region_show(sb, memory[avail[lo].idx], avail[lo].base,
7987 				avail[lo].limit - 1);
7988 
7989 	sbuf_printf(sb, "\n");
7990 	for (i = 0; i < n; i++) {
7991 		if (mem[i].idx >= nitems(region))
7992 			continue;                        /* skip holes */
7993 		if (!mem[i].limit)
7994 			mem[i].limit = i < n - 1 ? mem[i + 1].base - 1 : ~0;
7995 		mem_region_show(sb, region[mem[i].idx], mem[i].base,
7996 				mem[i].limit);
7997 	}
7998 
7999 	sbuf_printf(sb, "\n");
8000 	lo = t4_read_reg(sc, A_CIM_SDRAM_BASE_ADDR);
8001 	hi = t4_read_reg(sc, A_CIM_SDRAM_ADDR_SIZE) + lo - 1;
8002 	mem_region_show(sb, "uP RAM:", lo, hi);
8003 
8004 	lo = t4_read_reg(sc, A_CIM_EXTMEM2_BASE_ADDR);
8005 	hi = t4_read_reg(sc, A_CIM_EXTMEM2_ADDR_SIZE) + lo - 1;
8006 	mem_region_show(sb, "uP Extmem2:", lo, hi);
8007 
8008 	lo = t4_read_reg(sc, A_TP_PMM_RX_MAX_PAGE);
8009 	sbuf_printf(sb, "\n%u Rx pages of size %uKiB for %u channels\n",
8010 		   G_PMRXMAXPAGE(lo),
8011 		   t4_read_reg(sc, A_TP_PMM_RX_PAGE_SIZE) >> 10,
8012 		   (lo & F_PMRXNUMCHN) ? 2 : 1);
8013 
8014 	lo = t4_read_reg(sc, A_TP_PMM_TX_MAX_PAGE);
8015 	hi = t4_read_reg(sc, A_TP_PMM_TX_PAGE_SIZE);
8016 	sbuf_printf(sb, "%u Tx pages of size %u%ciB for %u channels\n",
8017 		   G_PMTXMAXPAGE(lo),
8018 		   hi >= (1 << 20) ? (hi >> 20) : (hi >> 10),
8019 		   hi >= (1 << 20) ? 'M' : 'K', 1 << G_PMTXNUMCHN(lo));
8020 	sbuf_printf(sb, "%u p-structs\n",
8021 		   t4_read_reg(sc, A_TP_CMM_MM_MAX_PSTRUCT));
8022 
8023 	for (i = 0; i < 4; i++) {
8024 		if (chip_id(sc) > CHELSIO_T5)
8025 			lo = t4_read_reg(sc, A_MPS_RX_MAC_BG_PG_CNT0 + i * 4);
8026 		else
8027 			lo = t4_read_reg(sc, A_MPS_RX_PG_RSV0 + i * 4);
8028 		if (is_t5(sc)) {
8029 			used = G_T5_USED(lo);
8030 			alloc = G_T5_ALLOC(lo);
8031 		} else {
8032 			used = G_USED(lo);
8033 			alloc = G_ALLOC(lo);
8034 		}
8035 		/* For T6 these are MAC buffer groups */
8036 		sbuf_printf(sb, "\nPort %d using %u pages out of %u allocated",
8037 		    i, used, alloc);
8038 	}
8039 	for (i = 0; i < sc->chip_params->nchan; i++) {
8040 		if (chip_id(sc) > CHELSIO_T5)
8041 			lo = t4_read_reg(sc, A_MPS_RX_LPBK_BG_PG_CNT0 + i * 4);
8042 		else
8043 			lo = t4_read_reg(sc, A_MPS_RX_PG_RSV4 + i * 4);
8044 		if (is_t5(sc)) {
8045 			used = G_T5_USED(lo);
8046 			alloc = G_T5_ALLOC(lo);
8047 		} else {
8048 			used = G_USED(lo);
8049 			alloc = G_ALLOC(lo);
8050 		}
8051 		/* For T6 these are MAC buffer groups */
8052 		sbuf_printf(sb,
8053 		    "\nLoopback %d using %u pages out of %u allocated",
8054 		    i, used, alloc);
8055 	}
8056 
8057 	rc = sbuf_finish(sb);
8058 	sbuf_delete(sb);
8059 
8060 	return (rc);
8061 }
8062 
8063 static inline void
8064 tcamxy2valmask(uint64_t x, uint64_t y, uint8_t *addr, uint64_t *mask)
8065 {
8066 	*mask = x | y;
8067 	y = htobe64(y);
8068 	memcpy(addr, (char *)&y + 2, ETHER_ADDR_LEN);
8069 }
8070 
8071 static int
8072 sysctl_mps_tcam(SYSCTL_HANDLER_ARGS)
8073 {
8074 	struct adapter *sc = arg1;
8075 	struct sbuf *sb;
8076 	int rc, i;
8077 
8078 	MPASS(chip_id(sc) <= CHELSIO_T5);
8079 
8080 	rc = sysctl_wire_old_buffer(req, 0);
8081 	if (rc != 0)
8082 		return (rc);
8083 
8084 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8085 	if (sb == NULL)
8086 		return (ENOMEM);
8087 
8088 	sbuf_printf(sb,
8089 	    "Idx  Ethernet address     Mask     Vld Ports PF"
8090 	    "  VF              Replication             P0 P1 P2 P3  ML");
8091 	for (i = 0; i < sc->chip_params->mps_tcam_size; i++) {
8092 		uint64_t tcamx, tcamy, mask;
8093 		uint32_t cls_lo, cls_hi;
8094 		uint8_t addr[ETHER_ADDR_LEN];
8095 
8096 		tcamy = t4_read_reg64(sc, MPS_CLS_TCAM_Y_L(i));
8097 		tcamx = t4_read_reg64(sc, MPS_CLS_TCAM_X_L(i));
8098 		if (tcamx & tcamy)
8099 			continue;
8100 		tcamxy2valmask(tcamx, tcamy, addr, &mask);
8101 		cls_lo = t4_read_reg(sc, MPS_CLS_SRAM_L(i));
8102 		cls_hi = t4_read_reg(sc, MPS_CLS_SRAM_H(i));
8103 		sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x %012jx"
8104 			   "  %c   %#x%4u%4d", i, addr[0], addr[1], addr[2],
8105 			   addr[3], addr[4], addr[5], (uintmax_t)mask,
8106 			   (cls_lo & F_SRAM_VLD) ? 'Y' : 'N',
8107 			   G_PORTMAP(cls_hi), G_PF(cls_lo),
8108 			   (cls_lo & F_VF_VALID) ? G_VF(cls_lo) : -1);
8109 
8110 		if (cls_lo & F_REPLICATE) {
8111 			struct fw_ldst_cmd ldst_cmd;
8112 
8113 			memset(&ldst_cmd, 0, sizeof(ldst_cmd));
8114 			ldst_cmd.op_to_addrspace =
8115 			    htobe32(V_FW_CMD_OP(FW_LDST_CMD) |
8116 				F_FW_CMD_REQUEST | F_FW_CMD_READ |
8117 				V_FW_LDST_CMD_ADDRSPACE(FW_LDST_ADDRSPC_MPS));
8118 			ldst_cmd.cycles_to_len16 = htobe32(FW_LEN16(ldst_cmd));
8119 			ldst_cmd.u.mps.rplc.fid_idx =
8120 			    htobe16(V_FW_LDST_CMD_FID(FW_LDST_MPS_RPLC) |
8121 				V_FW_LDST_CMD_IDX(i));
8122 
8123 			rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK,
8124 			    "t4mps");
8125 			if (rc)
8126 				break;
8127 			rc = -t4_wr_mbox(sc, sc->mbox, &ldst_cmd,
8128 			    sizeof(ldst_cmd), &ldst_cmd);
8129 			end_synchronized_op(sc, 0);
8130 
8131 			if (rc != 0) {
8132 				sbuf_printf(sb, "%36d", rc);
8133 				rc = 0;
8134 			} else {
8135 				sbuf_printf(sb, " %08x %08x %08x %08x",
8136 				    be32toh(ldst_cmd.u.mps.rplc.rplc127_96),
8137 				    be32toh(ldst_cmd.u.mps.rplc.rplc95_64),
8138 				    be32toh(ldst_cmd.u.mps.rplc.rplc63_32),
8139 				    be32toh(ldst_cmd.u.mps.rplc.rplc31_0));
8140 			}
8141 		} else
8142 			sbuf_printf(sb, "%36s", "");
8143 
8144 		sbuf_printf(sb, "%4u%3u%3u%3u %#3x", G_SRAM_PRIO0(cls_lo),
8145 		    G_SRAM_PRIO1(cls_lo), G_SRAM_PRIO2(cls_lo),
8146 		    G_SRAM_PRIO3(cls_lo), (cls_lo >> S_MULTILISTEN0) & 0xf);
8147 	}
8148 
8149 	if (rc)
8150 		(void) sbuf_finish(sb);
8151 	else
8152 		rc = sbuf_finish(sb);
8153 	sbuf_delete(sb);
8154 
8155 	return (rc);
8156 }
8157 
8158 static int
8159 sysctl_mps_tcam_t6(SYSCTL_HANDLER_ARGS)
8160 {
8161 	struct adapter *sc = arg1;
8162 	struct sbuf *sb;
8163 	int rc, i;
8164 
8165 	MPASS(chip_id(sc) > CHELSIO_T5);
8166 
8167 	rc = sysctl_wire_old_buffer(req, 0);
8168 	if (rc != 0)
8169 		return (rc);
8170 
8171 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8172 	if (sb == NULL)
8173 		return (ENOMEM);
8174 
8175 	sbuf_printf(sb, "Idx  Ethernet address     Mask       VNI   Mask"
8176 	    "   IVLAN Vld DIP_Hit   Lookup  Port Vld Ports PF  VF"
8177 	    "                           Replication"
8178 	    "                                    P0 P1 P2 P3  ML\n");
8179 
8180 	for (i = 0; i < sc->chip_params->mps_tcam_size; i++) {
8181 		uint8_t dip_hit, vlan_vld, lookup_type, port_num;
8182 		uint16_t ivlan;
8183 		uint64_t tcamx, tcamy, val, mask;
8184 		uint32_t cls_lo, cls_hi, ctl, data2, vnix, vniy;
8185 		uint8_t addr[ETHER_ADDR_LEN];
8186 
8187 		ctl = V_CTLREQID(1) | V_CTLCMDTYPE(0) | V_CTLXYBITSEL(0);
8188 		if (i < 256)
8189 			ctl |= V_CTLTCAMINDEX(i) | V_CTLTCAMSEL(0);
8190 		else
8191 			ctl |= V_CTLTCAMINDEX(i - 256) | V_CTLTCAMSEL(1);
8192 		t4_write_reg(sc, A_MPS_CLS_TCAM_DATA2_CTL, ctl);
8193 		val = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA1_REQ_ID1);
8194 		tcamy = G_DMACH(val) << 32;
8195 		tcamy |= t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA0_REQ_ID1);
8196 		data2 = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA2_REQ_ID1);
8197 		lookup_type = G_DATALKPTYPE(data2);
8198 		port_num = G_DATAPORTNUM(data2);
8199 		if (lookup_type && lookup_type != M_DATALKPTYPE) {
8200 			/* Inner header VNI */
8201 			vniy = ((data2 & F_DATAVIDH2) << 23) |
8202 				       (G_DATAVIDH1(data2) << 16) | G_VIDL(val);
8203 			dip_hit = data2 & F_DATADIPHIT;
8204 			vlan_vld = 0;
8205 		} else {
8206 			vniy = 0;
8207 			dip_hit = 0;
8208 			vlan_vld = data2 & F_DATAVIDH2;
8209 			ivlan = G_VIDL(val);
8210 		}
8211 
8212 		ctl |= V_CTLXYBITSEL(1);
8213 		t4_write_reg(sc, A_MPS_CLS_TCAM_DATA2_CTL, ctl);
8214 		val = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA1_REQ_ID1);
8215 		tcamx = G_DMACH(val) << 32;
8216 		tcamx |= t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA0_REQ_ID1);
8217 		data2 = t4_read_reg(sc, A_MPS_CLS_TCAM_RDATA2_REQ_ID1);
8218 		if (lookup_type && lookup_type != M_DATALKPTYPE) {
8219 			/* Inner header VNI mask */
8220 			vnix = ((data2 & F_DATAVIDH2) << 23) |
8221 			       (G_DATAVIDH1(data2) << 16) | G_VIDL(val);
8222 		} else
8223 			vnix = 0;
8224 
8225 		if (tcamx & tcamy)
8226 			continue;
8227 		tcamxy2valmask(tcamx, tcamy, addr, &mask);
8228 
8229 		cls_lo = t4_read_reg(sc, MPS_CLS_SRAM_L(i));
8230 		cls_hi = t4_read_reg(sc, MPS_CLS_SRAM_H(i));
8231 
8232 		if (lookup_type && lookup_type != M_DATALKPTYPE) {
8233 			sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x "
8234 			    "%012jx %06x %06x    -    -   %3c"
8235 			    "      'I'  %4x   %3c   %#x%4u%4d", i, addr[0],
8236 			    addr[1], addr[2], addr[3], addr[4], addr[5],
8237 			    (uintmax_t)mask, vniy, vnix, dip_hit ? 'Y' : 'N',
8238 			    port_num, cls_lo & F_T6_SRAM_VLD ? 'Y' : 'N',
8239 			    G_PORTMAP(cls_hi), G_T6_PF(cls_lo),
8240 			    cls_lo & F_T6_VF_VALID ? G_T6_VF(cls_lo) : -1);
8241 		} else {
8242 			sbuf_printf(sb, "\n%3u %02x:%02x:%02x:%02x:%02x:%02x "
8243 			    "%012jx    -       -   ", i, addr[0], addr[1],
8244 			    addr[2], addr[3], addr[4], addr[5],
8245 			    (uintmax_t)mask);
8246 
8247 			if (vlan_vld)
8248 				sbuf_printf(sb, "%4u   Y     ", ivlan);
8249 			else
8250 				sbuf_printf(sb, "  -    N     ");
8251 
8252 			sbuf_printf(sb, "-      %3c  %4x   %3c   %#x%4u%4d",
8253 			    lookup_type ? 'I' : 'O', port_num,
8254 			    cls_lo & F_T6_SRAM_VLD ? 'Y' : 'N',
8255 			    G_PORTMAP(cls_hi), G_T6_PF(cls_lo),
8256 			    cls_lo & F_T6_VF_VALID ? G_T6_VF(cls_lo) : -1);
8257 		}
8258 
8259 
8260 		if (cls_lo & F_T6_REPLICATE) {
8261 			struct fw_ldst_cmd ldst_cmd;
8262 
8263 			memset(&ldst_cmd, 0, sizeof(ldst_cmd));
8264 			ldst_cmd.op_to_addrspace =
8265 			    htobe32(V_FW_CMD_OP(FW_LDST_CMD) |
8266 				F_FW_CMD_REQUEST | F_FW_CMD_READ |
8267 				V_FW_LDST_CMD_ADDRSPACE(FW_LDST_ADDRSPC_MPS));
8268 			ldst_cmd.cycles_to_len16 = htobe32(FW_LEN16(ldst_cmd));
8269 			ldst_cmd.u.mps.rplc.fid_idx =
8270 			    htobe16(V_FW_LDST_CMD_FID(FW_LDST_MPS_RPLC) |
8271 				V_FW_LDST_CMD_IDX(i));
8272 
8273 			rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK,
8274 			    "t6mps");
8275 			if (rc)
8276 				break;
8277 			rc = -t4_wr_mbox(sc, sc->mbox, &ldst_cmd,
8278 			    sizeof(ldst_cmd), &ldst_cmd);
8279 			end_synchronized_op(sc, 0);
8280 
8281 			if (rc != 0) {
8282 				sbuf_printf(sb, "%72d", rc);
8283 				rc = 0;
8284 			} else {
8285 				sbuf_printf(sb, " %08x %08x %08x %08x"
8286 				    " %08x %08x %08x %08x",
8287 				    be32toh(ldst_cmd.u.mps.rplc.rplc255_224),
8288 				    be32toh(ldst_cmd.u.mps.rplc.rplc223_192),
8289 				    be32toh(ldst_cmd.u.mps.rplc.rplc191_160),
8290 				    be32toh(ldst_cmd.u.mps.rplc.rplc159_128),
8291 				    be32toh(ldst_cmd.u.mps.rplc.rplc127_96),
8292 				    be32toh(ldst_cmd.u.mps.rplc.rplc95_64),
8293 				    be32toh(ldst_cmd.u.mps.rplc.rplc63_32),
8294 				    be32toh(ldst_cmd.u.mps.rplc.rplc31_0));
8295 			}
8296 		} else
8297 			sbuf_printf(sb, "%72s", "");
8298 
8299 		sbuf_printf(sb, "%4u%3u%3u%3u %#x",
8300 		    G_T6_SRAM_PRIO0(cls_lo), G_T6_SRAM_PRIO1(cls_lo),
8301 		    G_T6_SRAM_PRIO2(cls_lo), G_T6_SRAM_PRIO3(cls_lo),
8302 		    (cls_lo >> S_T6_MULTILISTEN0) & 0xf);
8303 	}
8304 
8305 	if (rc)
8306 		(void) sbuf_finish(sb);
8307 	else
8308 		rc = sbuf_finish(sb);
8309 	sbuf_delete(sb);
8310 
8311 	return (rc);
8312 }
8313 
8314 static int
8315 sysctl_path_mtus(SYSCTL_HANDLER_ARGS)
8316 {
8317 	struct adapter *sc = arg1;
8318 	struct sbuf *sb;
8319 	int rc;
8320 	uint16_t mtus[NMTUS];
8321 
8322 	rc = sysctl_wire_old_buffer(req, 0);
8323 	if (rc != 0)
8324 		return (rc);
8325 
8326 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8327 	if (sb == NULL)
8328 		return (ENOMEM);
8329 
8330 	t4_read_mtu_tbl(sc, mtus, NULL);
8331 
8332 	sbuf_printf(sb, "%u %u %u %u %u %u %u %u %u %u %u %u %u %u %u %u",
8333 	    mtus[0], mtus[1], mtus[2], mtus[3], mtus[4], mtus[5], mtus[6],
8334 	    mtus[7], mtus[8], mtus[9], mtus[10], mtus[11], mtus[12], mtus[13],
8335 	    mtus[14], mtus[15]);
8336 
8337 	rc = sbuf_finish(sb);
8338 	sbuf_delete(sb);
8339 
8340 	return (rc);
8341 }
8342 
8343 static int
8344 sysctl_pm_stats(SYSCTL_HANDLER_ARGS)
8345 {
8346 	struct adapter *sc = arg1;
8347 	struct sbuf *sb;
8348 	int rc, i;
8349 	uint32_t tx_cnt[MAX_PM_NSTATS], rx_cnt[MAX_PM_NSTATS];
8350 	uint64_t tx_cyc[MAX_PM_NSTATS], rx_cyc[MAX_PM_NSTATS];
8351 	static const char *tx_stats[MAX_PM_NSTATS] = {
8352 		"Read:", "Write bypass:", "Write mem:", "Bypass + mem:",
8353 		"Tx FIFO wait", NULL, "Tx latency"
8354 	};
8355 	static const char *rx_stats[MAX_PM_NSTATS] = {
8356 		"Read:", "Write bypass:", "Write mem:", "Flush:",
8357 		"Rx FIFO wait", NULL, "Rx latency"
8358 	};
8359 
8360 	rc = sysctl_wire_old_buffer(req, 0);
8361 	if (rc != 0)
8362 		return (rc);
8363 
8364 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8365 	if (sb == NULL)
8366 		return (ENOMEM);
8367 
8368 	t4_pmtx_get_stats(sc, tx_cnt, tx_cyc);
8369 	t4_pmrx_get_stats(sc, rx_cnt, rx_cyc);
8370 
8371 	sbuf_printf(sb, "                Tx pcmds             Tx bytes");
8372 	for (i = 0; i < 4; i++) {
8373 		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
8374 		    tx_cyc[i]);
8375 	}
8376 
8377 	sbuf_printf(sb, "\n                Rx pcmds             Rx bytes");
8378 	for (i = 0; i < 4; i++) {
8379 		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
8380 		    rx_cyc[i]);
8381 	}
8382 
8383 	if (chip_id(sc) > CHELSIO_T5) {
8384 		sbuf_printf(sb,
8385 		    "\n              Total wait      Total occupancy");
8386 		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
8387 		    tx_cyc[i]);
8388 		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
8389 		    rx_cyc[i]);
8390 
8391 		i += 2;
8392 		MPASS(i < nitems(tx_stats));
8393 
8394 		sbuf_printf(sb,
8395 		    "\n                   Reads           Total wait");
8396 		sbuf_printf(sb, "\n%-13s %10u %20ju", tx_stats[i], tx_cnt[i],
8397 		    tx_cyc[i]);
8398 		sbuf_printf(sb, "\n%-13s %10u %20ju", rx_stats[i], rx_cnt[i],
8399 		    rx_cyc[i]);
8400 	}
8401 
8402 	rc = sbuf_finish(sb);
8403 	sbuf_delete(sb);
8404 
8405 	return (rc);
8406 }
8407 
8408 static int
8409 sysctl_rdma_stats(SYSCTL_HANDLER_ARGS)
8410 {
8411 	struct adapter *sc = arg1;
8412 	struct sbuf *sb;
8413 	int rc;
8414 	struct tp_rdma_stats stats;
8415 
8416 	rc = sysctl_wire_old_buffer(req, 0);
8417 	if (rc != 0)
8418 		return (rc);
8419 
8420 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8421 	if (sb == NULL)
8422 		return (ENOMEM);
8423 
8424 	mtx_lock(&sc->reg_lock);
8425 	t4_tp_get_rdma_stats(sc, &stats, 0);
8426 	mtx_unlock(&sc->reg_lock);
8427 
8428 	sbuf_printf(sb, "NoRQEModDefferals: %u\n", stats.rqe_dfr_mod);
8429 	sbuf_printf(sb, "NoRQEPktDefferals: %u", stats.rqe_dfr_pkt);
8430 
8431 	rc = sbuf_finish(sb);
8432 	sbuf_delete(sb);
8433 
8434 	return (rc);
8435 }
8436 
8437 static int
8438 sysctl_tcp_stats(SYSCTL_HANDLER_ARGS)
8439 {
8440 	struct adapter *sc = arg1;
8441 	struct sbuf *sb;
8442 	int rc;
8443 	struct tp_tcp_stats v4, v6;
8444 
8445 	rc = sysctl_wire_old_buffer(req, 0);
8446 	if (rc != 0)
8447 		return (rc);
8448 
8449 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8450 	if (sb == NULL)
8451 		return (ENOMEM);
8452 
8453 	mtx_lock(&sc->reg_lock);
8454 	t4_tp_get_tcp_stats(sc, &v4, &v6, 0);
8455 	mtx_unlock(&sc->reg_lock);
8456 
8457 	sbuf_printf(sb,
8458 	    "                                IP                 IPv6\n");
8459 	sbuf_printf(sb, "OutRsts:      %20u %20u\n",
8460 	    v4.tcp_out_rsts, v6.tcp_out_rsts);
8461 	sbuf_printf(sb, "InSegs:       %20ju %20ju\n",
8462 	    v4.tcp_in_segs, v6.tcp_in_segs);
8463 	sbuf_printf(sb, "OutSegs:      %20ju %20ju\n",
8464 	    v4.tcp_out_segs, v6.tcp_out_segs);
8465 	sbuf_printf(sb, "RetransSegs:  %20ju %20ju",
8466 	    v4.tcp_retrans_segs, v6.tcp_retrans_segs);
8467 
8468 	rc = sbuf_finish(sb);
8469 	sbuf_delete(sb);
8470 
8471 	return (rc);
8472 }
8473 
8474 static int
8475 sysctl_tids(SYSCTL_HANDLER_ARGS)
8476 {
8477 	struct adapter *sc = arg1;
8478 	struct sbuf *sb;
8479 	int rc;
8480 	struct tid_info *t = &sc->tids;
8481 
8482 	rc = sysctl_wire_old_buffer(req, 0);
8483 	if (rc != 0)
8484 		return (rc);
8485 
8486 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8487 	if (sb == NULL)
8488 		return (ENOMEM);
8489 
8490 	if (t->natids) {
8491 		sbuf_printf(sb, "ATID range: 0-%u, in use: %u\n", t->natids - 1,
8492 		    t->atids_in_use);
8493 	}
8494 
8495 	if (t->nhpftids) {
8496 		sbuf_printf(sb, "HPFTID range: %u-%u, in use: %u\n",
8497 		    t->hpftid_base, t->hpftid_end, t->hpftids_in_use);
8498 	}
8499 
8500 	if (t->ntids) {
8501 		sbuf_printf(sb, "TID range: ");
8502 		if (t4_read_reg(sc, A_LE_DB_CONFIG) & F_HASHEN) {
8503 			uint32_t b, hb;
8504 
8505 			if (chip_id(sc) <= CHELSIO_T5) {
8506 				b = t4_read_reg(sc, A_LE_DB_SERVER_INDEX) / 4;
8507 				hb = t4_read_reg(sc, A_LE_DB_TID_HASHBASE) / 4;
8508 			} else {
8509 				b = t4_read_reg(sc, A_LE_DB_SRVR_START_INDEX);
8510 				hb = t4_read_reg(sc, A_T6_LE_DB_HASH_TID_BASE);
8511 			}
8512 
8513 			if (b)
8514 				sbuf_printf(sb, "%u-%u, ", t->tid_base, b - 1);
8515 			sbuf_printf(sb, "%u-%u", hb, t->ntids - 1);
8516 		} else
8517 			sbuf_printf(sb, "%u-%u", t->tid_base, t->ntids - 1);
8518 		sbuf_printf(sb, ", in use: %u\n",
8519 		    atomic_load_acq_int(&t->tids_in_use));
8520 	}
8521 
8522 	if (t->nstids) {
8523 		sbuf_printf(sb, "STID range: %u-%u, in use: %u\n", t->stid_base,
8524 		    t->stid_base + t->nstids - 1, t->stids_in_use);
8525 	}
8526 
8527 	if (t->nftids) {
8528 		sbuf_printf(sb, "FTID range: %u-%u, in use: %u\n", t->ftid_base,
8529 		    t->ftid_end, t->ftids_in_use);
8530 	}
8531 
8532 	if (t->netids) {
8533 		sbuf_printf(sb, "ETID range: %u-%u, in use: %u\n", t->etid_base,
8534 		    t->etid_base + t->netids - 1, t->etids_in_use);
8535 	}
8536 
8537 	sbuf_printf(sb, "HW TID usage: %u IP users, %u IPv6 users",
8538 	    t4_read_reg(sc, A_LE_DB_ACT_CNT_IPV4),
8539 	    t4_read_reg(sc, A_LE_DB_ACT_CNT_IPV6));
8540 
8541 	rc = sbuf_finish(sb);
8542 	sbuf_delete(sb);
8543 
8544 	return (rc);
8545 }
8546 
8547 static int
8548 sysctl_tp_err_stats(SYSCTL_HANDLER_ARGS)
8549 {
8550 	struct adapter *sc = arg1;
8551 	struct sbuf *sb;
8552 	int rc;
8553 	struct tp_err_stats stats;
8554 
8555 	rc = sysctl_wire_old_buffer(req, 0);
8556 	if (rc != 0)
8557 		return (rc);
8558 
8559 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8560 	if (sb == NULL)
8561 		return (ENOMEM);
8562 
8563 	mtx_lock(&sc->reg_lock);
8564 	t4_tp_get_err_stats(sc, &stats, 0);
8565 	mtx_unlock(&sc->reg_lock);
8566 
8567 	if (sc->chip_params->nchan > 2) {
8568 		sbuf_printf(sb, "                 channel 0  channel 1"
8569 		    "  channel 2  channel 3\n");
8570 		sbuf_printf(sb, "macInErrs:      %10u %10u %10u %10u\n",
8571 		    stats.mac_in_errs[0], stats.mac_in_errs[1],
8572 		    stats.mac_in_errs[2], stats.mac_in_errs[3]);
8573 		sbuf_printf(sb, "hdrInErrs:      %10u %10u %10u %10u\n",
8574 		    stats.hdr_in_errs[0], stats.hdr_in_errs[1],
8575 		    stats.hdr_in_errs[2], stats.hdr_in_errs[3]);
8576 		sbuf_printf(sb, "tcpInErrs:      %10u %10u %10u %10u\n",
8577 		    stats.tcp_in_errs[0], stats.tcp_in_errs[1],
8578 		    stats.tcp_in_errs[2], stats.tcp_in_errs[3]);
8579 		sbuf_printf(sb, "tcp6InErrs:     %10u %10u %10u %10u\n",
8580 		    stats.tcp6_in_errs[0], stats.tcp6_in_errs[1],
8581 		    stats.tcp6_in_errs[2], stats.tcp6_in_errs[3]);
8582 		sbuf_printf(sb, "tnlCongDrops:   %10u %10u %10u %10u\n",
8583 		    stats.tnl_cong_drops[0], stats.tnl_cong_drops[1],
8584 		    stats.tnl_cong_drops[2], stats.tnl_cong_drops[3]);
8585 		sbuf_printf(sb, "tnlTxDrops:     %10u %10u %10u %10u\n",
8586 		    stats.tnl_tx_drops[0], stats.tnl_tx_drops[1],
8587 		    stats.tnl_tx_drops[2], stats.tnl_tx_drops[3]);
8588 		sbuf_printf(sb, "ofldVlanDrops:  %10u %10u %10u %10u\n",
8589 		    stats.ofld_vlan_drops[0], stats.ofld_vlan_drops[1],
8590 		    stats.ofld_vlan_drops[2], stats.ofld_vlan_drops[3]);
8591 		sbuf_printf(sb, "ofldChanDrops:  %10u %10u %10u %10u\n\n",
8592 		    stats.ofld_chan_drops[0], stats.ofld_chan_drops[1],
8593 		    stats.ofld_chan_drops[2], stats.ofld_chan_drops[3]);
8594 	} else {
8595 		sbuf_printf(sb, "                 channel 0  channel 1\n");
8596 		sbuf_printf(sb, "macInErrs:      %10u %10u\n",
8597 		    stats.mac_in_errs[0], stats.mac_in_errs[1]);
8598 		sbuf_printf(sb, "hdrInErrs:      %10u %10u\n",
8599 		    stats.hdr_in_errs[0], stats.hdr_in_errs[1]);
8600 		sbuf_printf(sb, "tcpInErrs:      %10u %10u\n",
8601 		    stats.tcp_in_errs[0], stats.tcp_in_errs[1]);
8602 		sbuf_printf(sb, "tcp6InErrs:     %10u %10u\n",
8603 		    stats.tcp6_in_errs[0], stats.tcp6_in_errs[1]);
8604 		sbuf_printf(sb, "tnlCongDrops:   %10u %10u\n",
8605 		    stats.tnl_cong_drops[0], stats.tnl_cong_drops[1]);
8606 		sbuf_printf(sb, "tnlTxDrops:     %10u %10u\n",
8607 		    stats.tnl_tx_drops[0], stats.tnl_tx_drops[1]);
8608 		sbuf_printf(sb, "ofldVlanDrops:  %10u %10u\n",
8609 		    stats.ofld_vlan_drops[0], stats.ofld_vlan_drops[1]);
8610 		sbuf_printf(sb, "ofldChanDrops:  %10u %10u\n\n",
8611 		    stats.ofld_chan_drops[0], stats.ofld_chan_drops[1]);
8612 	}
8613 
8614 	sbuf_printf(sb, "ofldNoNeigh:    %u\nofldCongDefer:  %u",
8615 	    stats.ofld_no_neigh, stats.ofld_cong_defer);
8616 
8617 	rc = sbuf_finish(sb);
8618 	sbuf_delete(sb);
8619 
8620 	return (rc);
8621 }
8622 
8623 static int
8624 sysctl_tp_la_mask(SYSCTL_HANDLER_ARGS)
8625 {
8626 	struct adapter *sc = arg1;
8627 	struct tp_params *tpp = &sc->params.tp;
8628 	u_int mask;
8629 	int rc;
8630 
8631 	mask = tpp->la_mask >> 16;
8632 	rc = sysctl_handle_int(oidp, &mask, 0, req);
8633 	if (rc != 0 || req->newptr == NULL)
8634 		return (rc);
8635 	if (mask > 0xffff)
8636 		return (EINVAL);
8637 	tpp->la_mask = mask << 16;
8638 	t4_set_reg_field(sc, A_TP_DBG_LA_CONFIG, 0xffff0000U, tpp->la_mask);
8639 
8640 	return (0);
8641 }
8642 
8643 struct field_desc {
8644 	const char *name;
8645 	u_int start;
8646 	u_int width;
8647 };
8648 
8649 static void
8650 field_desc_show(struct sbuf *sb, uint64_t v, const struct field_desc *f)
8651 {
8652 	char buf[32];
8653 	int line_size = 0;
8654 
8655 	while (f->name) {
8656 		uint64_t mask = (1ULL << f->width) - 1;
8657 		int len = snprintf(buf, sizeof(buf), "%s: %ju", f->name,
8658 		    ((uintmax_t)v >> f->start) & mask);
8659 
8660 		if (line_size + len >= 79) {
8661 			line_size = 8;
8662 			sbuf_printf(sb, "\n        ");
8663 		}
8664 		sbuf_printf(sb, "%s ", buf);
8665 		line_size += len + 1;
8666 		f++;
8667 	}
8668 	sbuf_printf(sb, "\n");
8669 }
8670 
8671 static const struct field_desc tp_la0[] = {
8672 	{ "RcfOpCodeOut", 60, 4 },
8673 	{ "State", 56, 4 },
8674 	{ "WcfState", 52, 4 },
8675 	{ "RcfOpcSrcOut", 50, 2 },
8676 	{ "CRxError", 49, 1 },
8677 	{ "ERxError", 48, 1 },
8678 	{ "SanityFailed", 47, 1 },
8679 	{ "SpuriousMsg", 46, 1 },
8680 	{ "FlushInputMsg", 45, 1 },
8681 	{ "FlushInputCpl", 44, 1 },
8682 	{ "RssUpBit", 43, 1 },
8683 	{ "RssFilterHit", 42, 1 },
8684 	{ "Tid", 32, 10 },
8685 	{ "InitTcb", 31, 1 },
8686 	{ "LineNumber", 24, 7 },
8687 	{ "Emsg", 23, 1 },
8688 	{ "EdataOut", 22, 1 },
8689 	{ "Cmsg", 21, 1 },
8690 	{ "CdataOut", 20, 1 },
8691 	{ "EreadPdu", 19, 1 },
8692 	{ "CreadPdu", 18, 1 },
8693 	{ "TunnelPkt", 17, 1 },
8694 	{ "RcfPeerFin", 16, 1 },
8695 	{ "RcfReasonOut", 12, 4 },
8696 	{ "TxCchannel", 10, 2 },
8697 	{ "RcfTxChannel", 8, 2 },
8698 	{ "RxEchannel", 6, 2 },
8699 	{ "RcfRxChannel", 5, 1 },
8700 	{ "RcfDataOutSrdy", 4, 1 },
8701 	{ "RxDvld", 3, 1 },
8702 	{ "RxOoDvld", 2, 1 },
8703 	{ "RxCongestion", 1, 1 },
8704 	{ "TxCongestion", 0, 1 },
8705 	{ NULL }
8706 };
8707 
8708 static const struct field_desc tp_la1[] = {
8709 	{ "CplCmdIn", 56, 8 },
8710 	{ "CplCmdOut", 48, 8 },
8711 	{ "ESynOut", 47, 1 },
8712 	{ "EAckOut", 46, 1 },
8713 	{ "EFinOut", 45, 1 },
8714 	{ "ERstOut", 44, 1 },
8715 	{ "SynIn", 43, 1 },
8716 	{ "AckIn", 42, 1 },
8717 	{ "FinIn", 41, 1 },
8718 	{ "RstIn", 40, 1 },
8719 	{ "DataIn", 39, 1 },
8720 	{ "DataInVld", 38, 1 },
8721 	{ "PadIn", 37, 1 },
8722 	{ "RxBufEmpty", 36, 1 },
8723 	{ "RxDdp", 35, 1 },
8724 	{ "RxFbCongestion", 34, 1 },
8725 	{ "TxFbCongestion", 33, 1 },
8726 	{ "TxPktSumSrdy", 32, 1 },
8727 	{ "RcfUlpType", 28, 4 },
8728 	{ "Eread", 27, 1 },
8729 	{ "Ebypass", 26, 1 },
8730 	{ "Esave", 25, 1 },
8731 	{ "Static0", 24, 1 },
8732 	{ "Cread", 23, 1 },
8733 	{ "Cbypass", 22, 1 },
8734 	{ "Csave", 21, 1 },
8735 	{ "CPktOut", 20, 1 },
8736 	{ "RxPagePoolFull", 18, 2 },
8737 	{ "RxLpbkPkt", 17, 1 },
8738 	{ "TxLpbkPkt", 16, 1 },
8739 	{ "RxVfValid", 15, 1 },
8740 	{ "SynLearned", 14, 1 },
8741 	{ "SetDelEntry", 13, 1 },
8742 	{ "SetInvEntry", 12, 1 },
8743 	{ "CpcmdDvld", 11, 1 },
8744 	{ "CpcmdSave", 10, 1 },
8745 	{ "RxPstructsFull", 8, 2 },
8746 	{ "EpcmdDvld", 7, 1 },
8747 	{ "EpcmdFlush", 6, 1 },
8748 	{ "EpcmdTrimPrefix", 5, 1 },
8749 	{ "EpcmdTrimPostfix", 4, 1 },
8750 	{ "ERssIp4Pkt", 3, 1 },
8751 	{ "ERssIp6Pkt", 2, 1 },
8752 	{ "ERssTcpUdpPkt", 1, 1 },
8753 	{ "ERssFceFipPkt", 0, 1 },
8754 	{ NULL }
8755 };
8756 
8757 static const struct field_desc tp_la2[] = {
8758 	{ "CplCmdIn", 56, 8 },
8759 	{ "MpsVfVld", 55, 1 },
8760 	{ "MpsPf", 52, 3 },
8761 	{ "MpsVf", 44, 8 },
8762 	{ "SynIn", 43, 1 },
8763 	{ "AckIn", 42, 1 },
8764 	{ "FinIn", 41, 1 },
8765 	{ "RstIn", 40, 1 },
8766 	{ "DataIn", 39, 1 },
8767 	{ "DataInVld", 38, 1 },
8768 	{ "PadIn", 37, 1 },
8769 	{ "RxBufEmpty", 36, 1 },
8770 	{ "RxDdp", 35, 1 },
8771 	{ "RxFbCongestion", 34, 1 },
8772 	{ "TxFbCongestion", 33, 1 },
8773 	{ "TxPktSumSrdy", 32, 1 },
8774 	{ "RcfUlpType", 28, 4 },
8775 	{ "Eread", 27, 1 },
8776 	{ "Ebypass", 26, 1 },
8777 	{ "Esave", 25, 1 },
8778 	{ "Static0", 24, 1 },
8779 	{ "Cread", 23, 1 },
8780 	{ "Cbypass", 22, 1 },
8781 	{ "Csave", 21, 1 },
8782 	{ "CPktOut", 20, 1 },
8783 	{ "RxPagePoolFull", 18, 2 },
8784 	{ "RxLpbkPkt", 17, 1 },
8785 	{ "TxLpbkPkt", 16, 1 },
8786 	{ "RxVfValid", 15, 1 },
8787 	{ "SynLearned", 14, 1 },
8788 	{ "SetDelEntry", 13, 1 },
8789 	{ "SetInvEntry", 12, 1 },
8790 	{ "CpcmdDvld", 11, 1 },
8791 	{ "CpcmdSave", 10, 1 },
8792 	{ "RxPstructsFull", 8, 2 },
8793 	{ "EpcmdDvld", 7, 1 },
8794 	{ "EpcmdFlush", 6, 1 },
8795 	{ "EpcmdTrimPrefix", 5, 1 },
8796 	{ "EpcmdTrimPostfix", 4, 1 },
8797 	{ "ERssIp4Pkt", 3, 1 },
8798 	{ "ERssIp6Pkt", 2, 1 },
8799 	{ "ERssTcpUdpPkt", 1, 1 },
8800 	{ "ERssFceFipPkt", 0, 1 },
8801 	{ NULL }
8802 };
8803 
8804 static void
8805 tp_la_show(struct sbuf *sb, uint64_t *p, int idx)
8806 {
8807 
8808 	field_desc_show(sb, *p, tp_la0);
8809 }
8810 
8811 static void
8812 tp_la_show2(struct sbuf *sb, uint64_t *p, int idx)
8813 {
8814 
8815 	if (idx)
8816 		sbuf_printf(sb, "\n");
8817 	field_desc_show(sb, p[0], tp_la0);
8818 	if (idx < (TPLA_SIZE / 2 - 1) || p[1] != ~0ULL)
8819 		field_desc_show(sb, p[1], tp_la0);
8820 }
8821 
8822 static void
8823 tp_la_show3(struct sbuf *sb, uint64_t *p, int idx)
8824 {
8825 
8826 	if (idx)
8827 		sbuf_printf(sb, "\n");
8828 	field_desc_show(sb, p[0], tp_la0);
8829 	if (idx < (TPLA_SIZE / 2 - 1) || p[1] != ~0ULL)
8830 		field_desc_show(sb, p[1], (p[0] & (1 << 17)) ? tp_la2 : tp_la1);
8831 }
8832 
8833 static int
8834 sysctl_tp_la(SYSCTL_HANDLER_ARGS)
8835 {
8836 	struct adapter *sc = arg1;
8837 	struct sbuf *sb;
8838 	uint64_t *buf, *p;
8839 	int rc;
8840 	u_int i, inc;
8841 	void (*show_func)(struct sbuf *, uint64_t *, int);
8842 
8843 	rc = sysctl_wire_old_buffer(req, 0);
8844 	if (rc != 0)
8845 		return (rc);
8846 
8847 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8848 	if (sb == NULL)
8849 		return (ENOMEM);
8850 
8851 	buf = malloc(TPLA_SIZE * sizeof(uint64_t), M_CXGBE, M_ZERO | M_WAITOK);
8852 
8853 	t4_tp_read_la(sc, buf, NULL);
8854 	p = buf;
8855 
8856 	switch (G_DBGLAMODE(t4_read_reg(sc, A_TP_DBG_LA_CONFIG))) {
8857 	case 2:
8858 		inc = 2;
8859 		show_func = tp_la_show2;
8860 		break;
8861 	case 3:
8862 		inc = 2;
8863 		show_func = tp_la_show3;
8864 		break;
8865 	default:
8866 		inc = 1;
8867 		show_func = tp_la_show;
8868 	}
8869 
8870 	for (i = 0; i < TPLA_SIZE / inc; i++, p += inc)
8871 		(*show_func)(sb, p, i);
8872 
8873 	rc = sbuf_finish(sb);
8874 	sbuf_delete(sb);
8875 	free(buf, M_CXGBE);
8876 	return (rc);
8877 }
8878 
8879 static int
8880 sysctl_tx_rate(SYSCTL_HANDLER_ARGS)
8881 {
8882 	struct adapter *sc = arg1;
8883 	struct sbuf *sb;
8884 	int rc;
8885 	u64 nrate[MAX_NCHAN], orate[MAX_NCHAN];
8886 
8887 	rc = sysctl_wire_old_buffer(req, 0);
8888 	if (rc != 0)
8889 		return (rc);
8890 
8891 	sb = sbuf_new_for_sysctl(NULL, NULL, 256, req);
8892 	if (sb == NULL)
8893 		return (ENOMEM);
8894 
8895 	t4_get_chan_txrate(sc, nrate, orate);
8896 
8897 	if (sc->chip_params->nchan > 2) {
8898 		sbuf_printf(sb, "              channel 0   channel 1"
8899 		    "   channel 2   channel 3\n");
8900 		sbuf_printf(sb, "NIC B/s:     %10ju  %10ju  %10ju  %10ju\n",
8901 		    nrate[0], nrate[1], nrate[2], nrate[3]);
8902 		sbuf_printf(sb, "Offload B/s: %10ju  %10ju  %10ju  %10ju",
8903 		    orate[0], orate[1], orate[2], orate[3]);
8904 	} else {
8905 		sbuf_printf(sb, "              channel 0   channel 1\n");
8906 		sbuf_printf(sb, "NIC B/s:     %10ju  %10ju\n",
8907 		    nrate[0], nrate[1]);
8908 		sbuf_printf(sb, "Offload B/s: %10ju  %10ju",
8909 		    orate[0], orate[1]);
8910 	}
8911 
8912 	rc = sbuf_finish(sb);
8913 	sbuf_delete(sb);
8914 
8915 	return (rc);
8916 }
8917 
8918 static int
8919 sysctl_ulprx_la(SYSCTL_HANDLER_ARGS)
8920 {
8921 	struct adapter *sc = arg1;
8922 	struct sbuf *sb;
8923 	uint32_t *buf, *p;
8924 	int rc, i;
8925 
8926 	rc = sysctl_wire_old_buffer(req, 0);
8927 	if (rc != 0)
8928 		return (rc);
8929 
8930 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8931 	if (sb == NULL)
8932 		return (ENOMEM);
8933 
8934 	buf = malloc(ULPRX_LA_SIZE * 8 * sizeof(uint32_t), M_CXGBE,
8935 	    M_ZERO | M_WAITOK);
8936 
8937 	t4_ulprx_read_la(sc, buf);
8938 	p = buf;
8939 
8940 	sbuf_printf(sb, "      Pcmd        Type   Message"
8941 	    "                Data");
8942 	for (i = 0; i < ULPRX_LA_SIZE; i++, p += 8) {
8943 		sbuf_printf(sb, "\n%08x%08x  %4x  %08x  %08x%08x%08x%08x",
8944 		    p[1], p[0], p[2], p[3], p[7], p[6], p[5], p[4]);
8945 	}
8946 
8947 	rc = sbuf_finish(sb);
8948 	sbuf_delete(sb);
8949 	free(buf, M_CXGBE);
8950 	return (rc);
8951 }
8952 
8953 static int
8954 sysctl_wcwr_stats(SYSCTL_HANDLER_ARGS)
8955 {
8956 	struct adapter *sc = arg1;
8957 	struct sbuf *sb;
8958 	int rc, v;
8959 
8960 	MPASS(chip_id(sc) >= CHELSIO_T5);
8961 
8962 	rc = sysctl_wire_old_buffer(req, 0);
8963 	if (rc != 0)
8964 		return (rc);
8965 
8966 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
8967 	if (sb == NULL)
8968 		return (ENOMEM);
8969 
8970 	v = t4_read_reg(sc, A_SGE_STAT_CFG);
8971 	if (G_STATSOURCE_T5(v) == 7) {
8972 		int mode;
8973 
8974 		mode = is_t5(sc) ? G_STATMODE(v) : G_T6_STATMODE(v);
8975 		if (mode == 0) {
8976 			sbuf_printf(sb, "total %d, incomplete %d",
8977 			    t4_read_reg(sc, A_SGE_STAT_TOTAL),
8978 			    t4_read_reg(sc, A_SGE_STAT_MATCH));
8979 		} else if (mode == 1) {
8980 			sbuf_printf(sb, "total %d, data overflow %d",
8981 			    t4_read_reg(sc, A_SGE_STAT_TOTAL),
8982 			    t4_read_reg(sc, A_SGE_STAT_MATCH));
8983 		} else {
8984 			sbuf_printf(sb, "unknown mode %d", mode);
8985 		}
8986 	}
8987 	rc = sbuf_finish(sb);
8988 	sbuf_delete(sb);
8989 
8990 	return (rc);
8991 }
8992 
8993 static int
8994 sysctl_cpus(SYSCTL_HANDLER_ARGS)
8995 {
8996 	struct adapter *sc = arg1;
8997 	enum cpu_sets op = arg2;
8998 	cpuset_t cpuset;
8999 	struct sbuf *sb;
9000 	int i, rc;
9001 
9002 	MPASS(op == LOCAL_CPUS || op == INTR_CPUS);
9003 
9004 	CPU_ZERO(&cpuset);
9005 	rc = bus_get_cpus(sc->dev, op, sizeof(cpuset), &cpuset);
9006 	if (rc != 0)
9007 		return (rc);
9008 
9009 	rc = sysctl_wire_old_buffer(req, 0);
9010 	if (rc != 0)
9011 		return (rc);
9012 
9013 	sb = sbuf_new_for_sysctl(NULL, NULL, 4096, req);
9014 	if (sb == NULL)
9015 		return (ENOMEM);
9016 
9017 	CPU_FOREACH(i)
9018 		sbuf_printf(sb, "%d ", i);
9019 	rc = sbuf_finish(sb);
9020 	sbuf_delete(sb);
9021 
9022 	return (rc);
9023 }
9024 
9025 #ifdef TCP_OFFLOAD
9026 static int
9027 sysctl_tls_rx_ports(SYSCTL_HANDLER_ARGS)
9028 {
9029 	struct adapter *sc = arg1;
9030 	int *old_ports, *new_ports;
9031 	int i, new_count, rc;
9032 
9033 	if (req->newptr == NULL && req->oldptr == NULL)
9034 		return (SYSCTL_OUT(req, NULL, imax(sc->tt.num_tls_rx_ports, 1) *
9035 		    sizeof(sc->tt.tls_rx_ports[0])));
9036 
9037 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4tlsrx");
9038 	if (rc)
9039 		return (rc);
9040 
9041 	if (sc->tt.num_tls_rx_ports == 0) {
9042 		i = -1;
9043 		rc = SYSCTL_OUT(req, &i, sizeof(i));
9044 	} else
9045 		rc = SYSCTL_OUT(req, sc->tt.tls_rx_ports,
9046 		    sc->tt.num_tls_rx_ports * sizeof(sc->tt.tls_rx_ports[0]));
9047 	if (rc == 0 && req->newptr != NULL) {
9048 		new_count = req->newlen / sizeof(new_ports[0]);
9049 		new_ports = malloc(new_count * sizeof(new_ports[0]), M_CXGBE,
9050 		    M_WAITOK);
9051 		rc = SYSCTL_IN(req, new_ports, new_count *
9052 		    sizeof(new_ports[0]));
9053 		if (rc)
9054 			goto err;
9055 
9056 		/* Allow setting to a single '-1' to clear the list. */
9057 		if (new_count == 1 && new_ports[0] == -1) {
9058 			ADAPTER_LOCK(sc);
9059 			old_ports = sc->tt.tls_rx_ports;
9060 			sc->tt.tls_rx_ports = NULL;
9061 			sc->tt.num_tls_rx_ports = 0;
9062 			ADAPTER_UNLOCK(sc);
9063 			free(old_ports, M_CXGBE);
9064 		} else {
9065 			for (i = 0; i < new_count; i++) {
9066 				if (new_ports[i] < 1 ||
9067 				    new_ports[i] > IPPORT_MAX) {
9068 					rc = EINVAL;
9069 					goto err;
9070 				}
9071 			}
9072 
9073 			ADAPTER_LOCK(sc);
9074 			old_ports = sc->tt.tls_rx_ports;
9075 			sc->tt.tls_rx_ports = new_ports;
9076 			sc->tt.num_tls_rx_ports = new_count;
9077 			ADAPTER_UNLOCK(sc);
9078 			free(old_ports, M_CXGBE);
9079 			new_ports = NULL;
9080 		}
9081 	err:
9082 		free(new_ports, M_CXGBE);
9083 	}
9084 	end_synchronized_op(sc, 0);
9085 	return (rc);
9086 }
9087 
9088 static void
9089 unit_conv(char *buf, size_t len, u_int val, u_int factor)
9090 {
9091 	u_int rem = val % factor;
9092 
9093 	if (rem == 0)
9094 		snprintf(buf, len, "%u", val / factor);
9095 	else {
9096 		while (rem % 10 == 0)
9097 			rem /= 10;
9098 		snprintf(buf, len, "%u.%u", val / factor, rem);
9099 	}
9100 }
9101 
9102 static int
9103 sysctl_tp_tick(SYSCTL_HANDLER_ARGS)
9104 {
9105 	struct adapter *sc = arg1;
9106 	char buf[16];
9107 	u_int res, re;
9108 	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
9109 
9110 	res = t4_read_reg(sc, A_TP_TIMER_RESOLUTION);
9111 	switch (arg2) {
9112 	case 0:
9113 		/* timer_tick */
9114 		re = G_TIMERRESOLUTION(res);
9115 		break;
9116 	case 1:
9117 		/* TCP timestamp tick */
9118 		re = G_TIMESTAMPRESOLUTION(res);
9119 		break;
9120 	case 2:
9121 		/* DACK tick */
9122 		re = G_DELAYEDACKRESOLUTION(res);
9123 		break;
9124 	default:
9125 		return (EDOOFUS);
9126 	}
9127 
9128 	unit_conv(buf, sizeof(buf), (cclk_ps << re), 1000000);
9129 
9130 	return (sysctl_handle_string(oidp, buf, sizeof(buf), req));
9131 }
9132 
9133 static int
9134 sysctl_tp_dack_timer(SYSCTL_HANDLER_ARGS)
9135 {
9136 	struct adapter *sc = arg1;
9137 	u_int res, dack_re, v;
9138 	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
9139 
9140 	res = t4_read_reg(sc, A_TP_TIMER_RESOLUTION);
9141 	dack_re = G_DELAYEDACKRESOLUTION(res);
9142 	v = ((cclk_ps << dack_re) / 1000000) * t4_read_reg(sc, A_TP_DACK_TIMER);
9143 
9144 	return (sysctl_handle_int(oidp, &v, 0, req));
9145 }
9146 
9147 static int
9148 sysctl_tp_timer(SYSCTL_HANDLER_ARGS)
9149 {
9150 	struct adapter *sc = arg1;
9151 	int reg = arg2;
9152 	u_int tre;
9153 	u_long tp_tick_us, v;
9154 	u_int cclk_ps = 1000000000 / sc->params.vpd.cclk;
9155 
9156 	MPASS(reg == A_TP_RXT_MIN || reg == A_TP_RXT_MAX ||
9157 	    reg == A_TP_PERS_MIN  || reg == A_TP_PERS_MAX ||
9158 	    reg == A_TP_KEEP_IDLE || reg == A_TP_KEEP_INTVL ||
9159 	    reg == A_TP_INIT_SRTT || reg == A_TP_FINWAIT2_TIMER);
9160 
9161 	tre = G_TIMERRESOLUTION(t4_read_reg(sc, A_TP_TIMER_RESOLUTION));
9162 	tp_tick_us = (cclk_ps << tre) / 1000000;
9163 
9164 	if (reg == A_TP_INIT_SRTT)
9165 		v = tp_tick_us * G_INITSRTT(t4_read_reg(sc, reg));
9166 	else
9167 		v = tp_tick_us * t4_read_reg(sc, reg);
9168 
9169 	return (sysctl_handle_long(oidp, &v, 0, req));
9170 }
9171 
9172 /*
9173  * All fields in TP_SHIFT_CNT are 4b and the starting location of the field is
9174  * passed to this function.
9175  */
9176 static int
9177 sysctl_tp_shift_cnt(SYSCTL_HANDLER_ARGS)
9178 {
9179 	struct adapter *sc = arg1;
9180 	int idx = arg2;
9181 	u_int v;
9182 
9183 	MPASS(idx >= 0 && idx <= 24);
9184 
9185 	v = (t4_read_reg(sc, A_TP_SHIFT_CNT) >> idx) & 0xf;
9186 
9187 	return (sysctl_handle_int(oidp, &v, 0, req));
9188 }
9189 
9190 static int
9191 sysctl_tp_backoff(SYSCTL_HANDLER_ARGS)
9192 {
9193 	struct adapter *sc = arg1;
9194 	int idx = arg2;
9195 	u_int shift, v, r;
9196 
9197 	MPASS(idx >= 0 && idx < 16);
9198 
9199 	r = A_TP_TCP_BACKOFF_REG0 + (idx & ~3);
9200 	shift = (idx & 3) << 3;
9201 	v = (t4_read_reg(sc, r) >> shift) & M_TIMERBACKOFFINDEX0;
9202 
9203 	return (sysctl_handle_int(oidp, &v, 0, req));
9204 }
9205 
9206 static int
9207 sysctl_holdoff_tmr_idx_ofld(SYSCTL_HANDLER_ARGS)
9208 {
9209 	struct vi_info *vi = arg1;
9210 	struct adapter *sc = vi->pi->adapter;
9211 	int idx, rc, i;
9212 	struct sge_ofld_rxq *ofld_rxq;
9213 	uint8_t v;
9214 
9215 	idx = vi->ofld_tmr_idx;
9216 
9217 	rc = sysctl_handle_int(oidp, &idx, 0, req);
9218 	if (rc != 0 || req->newptr == NULL)
9219 		return (rc);
9220 
9221 	if (idx < 0 || idx >= SGE_NTIMERS)
9222 		return (EINVAL);
9223 
9224 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
9225 	    "t4otmr");
9226 	if (rc)
9227 		return (rc);
9228 
9229 	v = V_QINTR_TIMER_IDX(idx) | V_QINTR_CNT_EN(vi->ofld_pktc_idx != -1);
9230 	for_each_ofld_rxq(vi, i, ofld_rxq) {
9231 #ifdef atomic_store_rel_8
9232 		atomic_store_rel_8(&ofld_rxq->iq.intr_params, v);
9233 #else
9234 		ofld_rxq->iq.intr_params = v;
9235 #endif
9236 	}
9237 	vi->ofld_tmr_idx = idx;
9238 
9239 	end_synchronized_op(sc, LOCK_HELD);
9240 	return (0);
9241 }
9242 
9243 static int
9244 sysctl_holdoff_pktc_idx_ofld(SYSCTL_HANDLER_ARGS)
9245 {
9246 	struct vi_info *vi = arg1;
9247 	struct adapter *sc = vi->pi->adapter;
9248 	int idx, rc;
9249 
9250 	idx = vi->ofld_pktc_idx;
9251 
9252 	rc = sysctl_handle_int(oidp, &idx, 0, req);
9253 	if (rc != 0 || req->newptr == NULL)
9254 		return (rc);
9255 
9256 	if (idx < -1 || idx >= SGE_NCOUNTERS)
9257 		return (EINVAL);
9258 
9259 	rc = begin_synchronized_op(sc, vi, HOLD_LOCK | SLEEP_OK | INTR_OK,
9260 	    "t4opktc");
9261 	if (rc)
9262 		return (rc);
9263 
9264 	if (vi->flags & VI_INIT_DONE)
9265 		rc = EBUSY; /* cannot be changed once the queues are created */
9266 	else
9267 		vi->ofld_pktc_idx = idx;
9268 
9269 	end_synchronized_op(sc, LOCK_HELD);
9270 	return (rc);
9271 }
9272 #endif
9273 
9274 static int
9275 get_sge_context(struct adapter *sc, struct t4_sge_context *cntxt)
9276 {
9277 	int rc;
9278 
9279 	if (cntxt->cid > M_CTXTQID)
9280 		return (EINVAL);
9281 
9282 	if (cntxt->mem_id != CTXT_EGRESS && cntxt->mem_id != CTXT_INGRESS &&
9283 	    cntxt->mem_id != CTXT_FLM && cntxt->mem_id != CTXT_CNM)
9284 		return (EINVAL);
9285 
9286 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ctxt");
9287 	if (rc)
9288 		return (rc);
9289 
9290 	if (sc->flags & FW_OK) {
9291 		rc = -t4_sge_ctxt_rd(sc, sc->mbox, cntxt->cid, cntxt->mem_id,
9292 		    &cntxt->data[0]);
9293 		if (rc == 0)
9294 			goto done;
9295 	}
9296 
9297 	/*
9298 	 * Read via firmware failed or wasn't even attempted.  Read directly via
9299 	 * the backdoor.
9300 	 */
9301 	rc = -t4_sge_ctxt_rd_bd(sc, cntxt->cid, cntxt->mem_id, &cntxt->data[0]);
9302 done:
9303 	end_synchronized_op(sc, 0);
9304 	return (rc);
9305 }
9306 
9307 static int
9308 load_fw(struct adapter *sc, struct t4_data *fw)
9309 {
9310 	int rc;
9311 	uint8_t *fw_data;
9312 
9313 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldfw");
9314 	if (rc)
9315 		return (rc);
9316 
9317 	/*
9318 	 * The firmware, with the sole exception of the memory parity error
9319 	 * handler, runs from memory and not flash.  It is almost always safe to
9320 	 * install a new firmware on a running system.  Just set bit 1 in
9321 	 * hw.cxgbe.dflags or dev.<nexus>.<n>.dflags first.
9322 	 */
9323 	if (sc->flags & FULL_INIT_DONE &&
9324 	    (sc->debug_flags & DF_LOAD_FW_ANYTIME) == 0) {
9325 		rc = EBUSY;
9326 		goto done;
9327 	}
9328 
9329 	fw_data = malloc(fw->len, M_CXGBE, M_WAITOK);
9330 	if (fw_data == NULL) {
9331 		rc = ENOMEM;
9332 		goto done;
9333 	}
9334 
9335 	rc = copyin(fw->data, fw_data, fw->len);
9336 	if (rc == 0)
9337 		rc = -t4_load_fw(sc, fw_data, fw->len);
9338 
9339 	free(fw_data, M_CXGBE);
9340 done:
9341 	end_synchronized_op(sc, 0);
9342 	return (rc);
9343 }
9344 
9345 static int
9346 load_cfg(struct adapter *sc, struct t4_data *cfg)
9347 {
9348 	int rc;
9349 	uint8_t *cfg_data = NULL;
9350 
9351 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldcf");
9352 	if (rc)
9353 		return (rc);
9354 
9355 	if (cfg->len == 0) {
9356 		/* clear */
9357 		rc = -t4_load_cfg(sc, NULL, 0);
9358 		goto done;
9359 	}
9360 
9361 	cfg_data = malloc(cfg->len, M_CXGBE, M_WAITOK);
9362 	if (cfg_data == NULL) {
9363 		rc = ENOMEM;
9364 		goto done;
9365 	}
9366 
9367 	rc = copyin(cfg->data, cfg_data, cfg->len);
9368 	if (rc == 0)
9369 		rc = -t4_load_cfg(sc, cfg_data, cfg->len);
9370 
9371 	free(cfg_data, M_CXGBE);
9372 done:
9373 	end_synchronized_op(sc, 0);
9374 	return (rc);
9375 }
9376 
9377 static int
9378 load_boot(struct adapter *sc, struct t4_bootrom *br)
9379 {
9380 	int rc;
9381 	uint8_t *br_data = NULL;
9382 	u_int offset;
9383 
9384 	if (br->len > 1024 * 1024)
9385 		return (EFBIG);
9386 
9387 	if (br->pf_offset == 0) {
9388 		/* pfidx */
9389 		if (br->pfidx_addr > 7)
9390 			return (EINVAL);
9391 		offset = G_OFFSET(t4_read_reg(sc, PF_REG(br->pfidx_addr,
9392 		    A_PCIE_PF_EXPROM_OFST)));
9393 	} else if (br->pf_offset == 1) {
9394 		/* offset */
9395 		offset = G_OFFSET(br->pfidx_addr);
9396 	} else {
9397 		return (EINVAL);
9398 	}
9399 
9400 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldbr");
9401 	if (rc)
9402 		return (rc);
9403 
9404 	if (br->len == 0) {
9405 		/* clear */
9406 		rc = -t4_load_boot(sc, NULL, offset, 0);
9407 		goto done;
9408 	}
9409 
9410 	br_data = malloc(br->len, M_CXGBE, M_WAITOK);
9411 	if (br_data == NULL) {
9412 		rc = ENOMEM;
9413 		goto done;
9414 	}
9415 
9416 	rc = copyin(br->data, br_data, br->len);
9417 	if (rc == 0)
9418 		rc = -t4_load_boot(sc, br_data, offset, br->len);
9419 
9420 	free(br_data, M_CXGBE);
9421 done:
9422 	end_synchronized_op(sc, 0);
9423 	return (rc);
9424 }
9425 
9426 static int
9427 load_bootcfg(struct adapter *sc, struct t4_data *bc)
9428 {
9429 	int rc;
9430 	uint8_t *bc_data = NULL;
9431 
9432 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4ldcf");
9433 	if (rc)
9434 		return (rc);
9435 
9436 	if (bc->len == 0) {
9437 		/* clear */
9438 		rc = -t4_load_bootcfg(sc, NULL, 0);
9439 		goto done;
9440 	}
9441 
9442 	bc_data = malloc(bc->len, M_CXGBE, M_WAITOK);
9443 	if (bc_data == NULL) {
9444 		rc = ENOMEM;
9445 		goto done;
9446 	}
9447 
9448 	rc = copyin(bc->data, bc_data, bc->len);
9449 	if (rc == 0)
9450 		rc = -t4_load_bootcfg(sc, bc_data, bc->len);
9451 
9452 	free(bc_data, M_CXGBE);
9453 done:
9454 	end_synchronized_op(sc, 0);
9455 	return (rc);
9456 }
9457 
9458 static int
9459 cudbg_dump(struct adapter *sc, struct t4_cudbg_dump *dump)
9460 {
9461 	int rc;
9462 	struct cudbg_init *cudbg;
9463 	void *handle, *buf;
9464 
9465 	/* buf is large, don't block if no memory is available */
9466 	buf = malloc(dump->len, M_CXGBE, M_NOWAIT | M_ZERO);
9467 	if (buf == NULL)
9468 		return (ENOMEM);
9469 
9470 	handle = cudbg_alloc_handle();
9471 	if (handle == NULL) {
9472 		rc = ENOMEM;
9473 		goto done;
9474 	}
9475 
9476 	cudbg = cudbg_get_init(handle);
9477 	cudbg->adap = sc;
9478 	cudbg->print = (cudbg_print_cb)printf;
9479 
9480 #ifndef notyet
9481 	device_printf(sc->dev, "%s: wr_flash %u, len %u, data %p.\n",
9482 	    __func__, dump->wr_flash, dump->len, dump->data);
9483 #endif
9484 
9485 	if (dump->wr_flash)
9486 		cudbg->use_flash = 1;
9487 	MPASS(sizeof(cudbg->dbg_bitmap) == sizeof(dump->bitmap));
9488 	memcpy(cudbg->dbg_bitmap, dump->bitmap, sizeof(cudbg->dbg_bitmap));
9489 
9490 	rc = cudbg_collect(handle, buf, &dump->len);
9491 	if (rc != 0)
9492 		goto done;
9493 
9494 	rc = copyout(buf, dump->data, dump->len);
9495 done:
9496 	cudbg_free_handle(handle);
9497 	free(buf, M_CXGBE);
9498 	return (rc);
9499 }
9500 
9501 static void
9502 free_offload_policy(struct t4_offload_policy *op)
9503 {
9504 	struct offload_rule *r;
9505 	int i;
9506 
9507 	if (op == NULL)
9508 		return;
9509 
9510 	r = &op->rule[0];
9511 	for (i = 0; i < op->nrules; i++, r++) {
9512 		free(r->bpf_prog.bf_insns, M_CXGBE);
9513 	}
9514 	free(op->rule, M_CXGBE);
9515 	free(op, M_CXGBE);
9516 }
9517 
9518 static int
9519 set_offload_policy(struct adapter *sc, struct t4_offload_policy *uop)
9520 {
9521 	int i, rc, len;
9522 	struct t4_offload_policy *op, *old;
9523 	struct bpf_program *bf;
9524 	const struct offload_settings *s;
9525 	struct offload_rule *r;
9526 	void *u;
9527 
9528 	if (!is_offload(sc))
9529 		return (ENODEV);
9530 
9531 	if (uop->nrules == 0) {
9532 		/* Delete installed policies. */
9533 		op = NULL;
9534 		goto set_policy;
9535 	} if (uop->nrules > 256) { /* arbitrary */
9536 		return (E2BIG);
9537 	}
9538 
9539 	/* Copy userspace offload policy to kernel */
9540 	op = malloc(sizeof(*op), M_CXGBE, M_ZERO | M_WAITOK);
9541 	op->nrules = uop->nrules;
9542 	len = op->nrules * sizeof(struct offload_rule);
9543 	op->rule = malloc(len, M_CXGBE, M_ZERO | M_WAITOK);
9544 	rc = copyin(uop->rule, op->rule, len);
9545 	if (rc) {
9546 		free(op->rule, M_CXGBE);
9547 		free(op, M_CXGBE);
9548 		return (rc);
9549 	}
9550 
9551 	r = &op->rule[0];
9552 	for (i = 0; i < op->nrules; i++, r++) {
9553 
9554 		/* Validate open_type */
9555 		if (r->open_type != OPEN_TYPE_LISTEN &&
9556 		    r->open_type != OPEN_TYPE_ACTIVE &&
9557 		    r->open_type != OPEN_TYPE_PASSIVE &&
9558 		    r->open_type != OPEN_TYPE_DONTCARE) {
9559 error:
9560 			/*
9561 			 * Rules 0 to i have malloc'd filters that need to be
9562 			 * freed.  Rules i+1 to nrules have userspace pointers
9563 			 * and should be left alone.
9564 			 */
9565 			op->nrules = i;
9566 			free_offload_policy(op);
9567 			return (rc);
9568 		}
9569 
9570 		/* Validate settings */
9571 		s = &r->settings;
9572 		if ((s->offload != 0 && s->offload != 1) ||
9573 		    s->cong_algo < -1 || s->cong_algo > CONG_ALG_HIGHSPEED ||
9574 		    s->sched_class < -1 ||
9575 		    s->sched_class >= sc->chip_params->nsched_cls) {
9576 			rc = EINVAL;
9577 			goto error;
9578 		}
9579 
9580 		bf = &r->bpf_prog;
9581 		u = bf->bf_insns;	/* userspace ptr */
9582 		bf->bf_insns = NULL;
9583 		if (bf->bf_len == 0) {
9584 			/* legal, matches everything */
9585 			continue;
9586 		}
9587 		len = bf->bf_len * sizeof(*bf->bf_insns);
9588 		bf->bf_insns = malloc(len, M_CXGBE, M_ZERO | M_WAITOK);
9589 		rc = copyin(u, bf->bf_insns, len);
9590 		if (rc != 0)
9591 			goto error;
9592 
9593 		if (!bpf_validate(bf->bf_insns, bf->bf_len)) {
9594 			rc = EINVAL;
9595 			goto error;
9596 		}
9597 	}
9598 set_policy:
9599 	rw_wlock(&sc->policy_lock);
9600 	old = sc->policy;
9601 	sc->policy = op;
9602 	rw_wunlock(&sc->policy_lock);
9603 	free_offload_policy(old);
9604 
9605 	return (0);
9606 }
9607 
9608 #define MAX_READ_BUF_SIZE (128 * 1024)
9609 static int
9610 read_card_mem(struct adapter *sc, int win, struct t4_mem_range *mr)
9611 {
9612 	uint32_t addr, remaining, n;
9613 	uint32_t *buf;
9614 	int rc;
9615 	uint8_t *dst;
9616 
9617 	rc = validate_mem_range(sc, mr->addr, mr->len);
9618 	if (rc != 0)
9619 		return (rc);
9620 
9621 	buf = malloc(min(mr->len, MAX_READ_BUF_SIZE), M_CXGBE, M_WAITOK);
9622 	addr = mr->addr;
9623 	remaining = mr->len;
9624 	dst = (void *)mr->data;
9625 
9626 	while (remaining) {
9627 		n = min(remaining, MAX_READ_BUF_SIZE);
9628 		read_via_memwin(sc, 2, addr, buf, n);
9629 
9630 		rc = copyout(buf, dst, n);
9631 		if (rc != 0)
9632 			break;
9633 
9634 		dst += n;
9635 		remaining -= n;
9636 		addr += n;
9637 	}
9638 
9639 	free(buf, M_CXGBE);
9640 	return (rc);
9641 }
9642 #undef MAX_READ_BUF_SIZE
9643 
9644 static int
9645 read_i2c(struct adapter *sc, struct t4_i2c_data *i2cd)
9646 {
9647 	int rc;
9648 
9649 	if (i2cd->len == 0 || i2cd->port_id >= sc->params.nports)
9650 		return (EINVAL);
9651 
9652 	if (i2cd->len > sizeof(i2cd->data))
9653 		return (EFBIG);
9654 
9655 	rc = begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4i2crd");
9656 	if (rc)
9657 		return (rc);
9658 	rc = -t4_i2c_rd(sc, sc->mbox, i2cd->port_id, i2cd->dev_addr,
9659 	    i2cd->offset, i2cd->len, &i2cd->data[0]);
9660 	end_synchronized_op(sc, 0);
9661 
9662 	return (rc);
9663 }
9664 
9665 int
9666 t4_os_find_pci_capability(struct adapter *sc, int cap)
9667 {
9668 	int i;
9669 
9670 	return (pci_find_cap(sc->dev, cap, &i) == 0 ? i : 0);
9671 }
9672 
9673 int
9674 t4_os_pci_save_state(struct adapter *sc)
9675 {
9676 	device_t dev;
9677 	struct pci_devinfo *dinfo;
9678 
9679 	dev = sc->dev;
9680 	dinfo = device_get_ivars(dev);
9681 
9682 	pci_cfg_save(dev, dinfo, 0);
9683 	return (0);
9684 }
9685 
9686 int
9687 t4_os_pci_restore_state(struct adapter *sc)
9688 {
9689 	device_t dev;
9690 	struct pci_devinfo *dinfo;
9691 
9692 	dev = sc->dev;
9693 	dinfo = device_get_ivars(dev);
9694 
9695 	pci_cfg_restore(dev, dinfo);
9696 	return (0);
9697 }
9698 
9699 void
9700 t4_os_portmod_changed(struct port_info *pi)
9701 {
9702 	struct adapter *sc = pi->adapter;
9703 	struct vi_info *vi;
9704 	struct ifnet *ifp;
9705 	static const char *mod_str[] = {
9706 		NULL, "LR", "SR", "ER", "TWINAX", "active TWINAX", "LRM"
9707 	};
9708 
9709 	KASSERT((pi->flags & FIXED_IFMEDIA) == 0,
9710 	    ("%s: port_type %u", __func__, pi->port_type));
9711 
9712 	vi = &pi->vi[0];
9713 	if (begin_synchronized_op(sc, vi, HOLD_LOCK, "t4mod") == 0) {
9714 		PORT_LOCK(pi);
9715 		build_medialist(pi);
9716 		if (pi->mod_type != FW_PORT_MOD_TYPE_NONE) {
9717 			fixup_link_config(pi);
9718 			apply_link_config(pi);
9719 		}
9720 		PORT_UNLOCK(pi);
9721 		end_synchronized_op(sc, LOCK_HELD);
9722 	}
9723 
9724 	ifp = vi->ifp;
9725 	if (pi->mod_type == FW_PORT_MOD_TYPE_NONE)
9726 		if_printf(ifp, "transceiver unplugged.\n");
9727 	else if (pi->mod_type == FW_PORT_MOD_TYPE_UNKNOWN)
9728 		if_printf(ifp, "unknown transceiver inserted.\n");
9729 	else if (pi->mod_type == FW_PORT_MOD_TYPE_NOTSUPPORTED)
9730 		if_printf(ifp, "unsupported transceiver inserted.\n");
9731 	else if (pi->mod_type > 0 && pi->mod_type < nitems(mod_str)) {
9732 		if_printf(ifp, "%dGbps %s transceiver inserted.\n",
9733 		    port_top_speed(pi), mod_str[pi->mod_type]);
9734 	} else {
9735 		if_printf(ifp, "transceiver (type %d) inserted.\n",
9736 		    pi->mod_type);
9737 	}
9738 }
9739 
9740 void
9741 t4_os_link_changed(struct port_info *pi)
9742 {
9743 	struct vi_info *vi;
9744 	struct ifnet *ifp;
9745 	struct link_config *lc;
9746 	int v;
9747 
9748 	PORT_LOCK_ASSERT_OWNED(pi);
9749 
9750 	for_each_vi(pi, v, vi) {
9751 		ifp = vi->ifp;
9752 		if (ifp == NULL)
9753 			continue;
9754 
9755 		lc = &pi->link_cfg;
9756 		if (lc->link_ok) {
9757 			ifp->if_baudrate = IF_Mbps(lc->speed);
9758 			if_link_state_change(ifp, LINK_STATE_UP);
9759 		} else {
9760 			if_link_state_change(ifp, LINK_STATE_DOWN);
9761 		}
9762 	}
9763 }
9764 
9765 void
9766 t4_iterate(void (*func)(struct adapter *, void *), void *arg)
9767 {
9768 	struct adapter *sc;
9769 
9770 	sx_slock(&t4_list_lock);
9771 	SLIST_FOREACH(sc, &t4_list, link) {
9772 		/*
9773 		 * func should not make any assumptions about what state sc is
9774 		 * in - the only guarantee is that sc->sc_lock is a valid lock.
9775 		 */
9776 		func(sc, arg);
9777 	}
9778 	sx_sunlock(&t4_list_lock);
9779 }
9780 
9781 static int
9782 t4_ioctl(struct cdev *dev, unsigned long cmd, caddr_t data, int fflag,
9783     struct thread *td)
9784 {
9785 	int rc;
9786 	struct adapter *sc = dev->si_drv1;
9787 
9788 	rc = priv_check(td, PRIV_DRIVER);
9789 	if (rc != 0)
9790 		return (rc);
9791 
9792 	switch (cmd) {
9793 	case CHELSIO_T4_GETREG: {
9794 		struct t4_reg *edata = (struct t4_reg *)data;
9795 
9796 		if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
9797 			return (EFAULT);
9798 
9799 		if (edata->size == 4)
9800 			edata->val = t4_read_reg(sc, edata->addr);
9801 		else if (edata->size == 8)
9802 			edata->val = t4_read_reg64(sc, edata->addr);
9803 		else
9804 			return (EINVAL);
9805 
9806 		break;
9807 	}
9808 	case CHELSIO_T4_SETREG: {
9809 		struct t4_reg *edata = (struct t4_reg *)data;
9810 
9811 		if ((edata->addr & 0x3) != 0 || edata->addr >= sc->mmio_len)
9812 			return (EFAULT);
9813 
9814 		if (edata->size == 4) {
9815 			if (edata->val & 0xffffffff00000000)
9816 				return (EINVAL);
9817 			t4_write_reg(sc, edata->addr, (uint32_t) edata->val);
9818 		} else if (edata->size == 8)
9819 			t4_write_reg64(sc, edata->addr, edata->val);
9820 		else
9821 			return (EINVAL);
9822 		break;
9823 	}
9824 	case CHELSIO_T4_REGDUMP: {
9825 		struct t4_regdump *regs = (struct t4_regdump *)data;
9826 		int reglen = t4_get_regs_len(sc);
9827 		uint8_t *buf;
9828 
9829 		if (regs->len < reglen) {
9830 			regs->len = reglen; /* hint to the caller */
9831 			return (ENOBUFS);
9832 		}
9833 
9834 		regs->len = reglen;
9835 		buf = malloc(reglen, M_CXGBE, M_WAITOK | M_ZERO);
9836 		get_regs(sc, regs, buf);
9837 		rc = copyout(buf, regs->data, reglen);
9838 		free(buf, M_CXGBE);
9839 		break;
9840 	}
9841 	case CHELSIO_T4_GET_FILTER_MODE:
9842 		rc = get_filter_mode(sc, (uint32_t *)data);
9843 		break;
9844 	case CHELSIO_T4_SET_FILTER_MODE:
9845 		rc = set_filter_mode(sc, *(uint32_t *)data);
9846 		break;
9847 	case CHELSIO_T4_GET_FILTER:
9848 		rc = get_filter(sc, (struct t4_filter *)data);
9849 		break;
9850 	case CHELSIO_T4_SET_FILTER:
9851 		rc = set_filter(sc, (struct t4_filter *)data);
9852 		break;
9853 	case CHELSIO_T4_DEL_FILTER:
9854 		rc = del_filter(sc, (struct t4_filter *)data);
9855 		break;
9856 	case CHELSIO_T4_GET_SGE_CONTEXT:
9857 		rc = get_sge_context(sc, (struct t4_sge_context *)data);
9858 		break;
9859 	case CHELSIO_T4_LOAD_FW:
9860 		rc = load_fw(sc, (struct t4_data *)data);
9861 		break;
9862 	case CHELSIO_T4_GET_MEM:
9863 		rc = read_card_mem(sc, 2, (struct t4_mem_range *)data);
9864 		break;
9865 	case CHELSIO_T4_GET_I2C:
9866 		rc = read_i2c(sc, (struct t4_i2c_data *)data);
9867 		break;
9868 	case CHELSIO_T4_CLEAR_STATS: {
9869 		int i, v, bg_map;
9870 		u_int port_id = *(uint32_t *)data;
9871 		struct port_info *pi;
9872 		struct vi_info *vi;
9873 
9874 		if (port_id >= sc->params.nports)
9875 			return (EINVAL);
9876 		pi = sc->port[port_id];
9877 		if (pi == NULL)
9878 			return (EIO);
9879 
9880 		/* MAC stats */
9881 		t4_clr_port_stats(sc, pi->tx_chan);
9882 		pi->tx_parse_error = 0;
9883 		pi->tnl_cong_drops = 0;
9884 		mtx_lock(&sc->reg_lock);
9885 		for_each_vi(pi, v, vi) {
9886 			if (vi->flags & VI_INIT_DONE)
9887 				t4_clr_vi_stats(sc, vi->viid);
9888 		}
9889 		bg_map = pi->mps_bg_map;
9890 		v = 0;	/* reuse */
9891 		while (bg_map) {
9892 			i = ffs(bg_map) - 1;
9893 			t4_write_indirect(sc, A_TP_MIB_INDEX, A_TP_MIB_DATA, &v,
9894 			    1, A_TP_MIB_TNL_CNG_DROP_0 + i);
9895 			bg_map &= ~(1 << i);
9896 		}
9897 		mtx_unlock(&sc->reg_lock);
9898 
9899 		/*
9900 		 * Since this command accepts a port, clear stats for
9901 		 * all VIs on this port.
9902 		 */
9903 		for_each_vi(pi, v, vi) {
9904 			if (vi->flags & VI_INIT_DONE) {
9905 				struct sge_rxq *rxq;
9906 				struct sge_txq *txq;
9907 				struct sge_wrq *wrq;
9908 
9909 				for_each_rxq(vi, i, rxq) {
9910 #if defined(INET) || defined(INET6)
9911 					rxq->lro.lro_queued = 0;
9912 					rxq->lro.lro_flushed = 0;
9913 #endif
9914 					rxq->rxcsum = 0;
9915 					rxq->vlan_extraction = 0;
9916 				}
9917 
9918 				for_each_txq(vi, i, txq) {
9919 					txq->txcsum = 0;
9920 					txq->tso_wrs = 0;
9921 					txq->vlan_insertion = 0;
9922 					txq->imm_wrs = 0;
9923 					txq->sgl_wrs = 0;
9924 					txq->txpkt_wrs = 0;
9925 					txq->txpkts0_wrs = 0;
9926 					txq->txpkts1_wrs = 0;
9927 					txq->txpkts0_pkts = 0;
9928 					txq->txpkts1_pkts = 0;
9929 					txq->raw_wrs = 0;
9930 					mp_ring_reset_stats(txq->r);
9931 				}
9932 
9933 #ifdef TCP_OFFLOAD
9934 				/* nothing to clear for each ofld_rxq */
9935 
9936 				for_each_ofld_txq(vi, i, wrq) {
9937 					wrq->tx_wrs_direct = 0;
9938 					wrq->tx_wrs_copied = 0;
9939 				}
9940 #endif
9941 
9942 				if (IS_MAIN_VI(vi)) {
9943 					wrq = &sc->sge.ctrlq[pi->port_id];
9944 					wrq->tx_wrs_direct = 0;
9945 					wrq->tx_wrs_copied = 0;
9946 				}
9947 			}
9948 		}
9949 		break;
9950 	}
9951 	case CHELSIO_T4_SCHED_CLASS:
9952 		rc = t4_set_sched_class(sc, (struct t4_sched_params *)data);
9953 		break;
9954 	case CHELSIO_T4_SCHED_QUEUE:
9955 		rc = t4_set_sched_queue(sc, (struct t4_sched_queue *)data);
9956 		break;
9957 	case CHELSIO_T4_GET_TRACER:
9958 		rc = t4_get_tracer(sc, (struct t4_tracer *)data);
9959 		break;
9960 	case CHELSIO_T4_SET_TRACER:
9961 		rc = t4_set_tracer(sc, (struct t4_tracer *)data);
9962 		break;
9963 	case CHELSIO_T4_LOAD_CFG:
9964 		rc = load_cfg(sc, (struct t4_data *)data);
9965 		break;
9966 	case CHELSIO_T4_LOAD_BOOT:
9967 		rc = load_boot(sc, (struct t4_bootrom *)data);
9968 		break;
9969 	case CHELSIO_T4_LOAD_BOOTCFG:
9970 		rc = load_bootcfg(sc, (struct t4_data *)data);
9971 		break;
9972 	case CHELSIO_T4_CUDBG_DUMP:
9973 		rc = cudbg_dump(sc, (struct t4_cudbg_dump *)data);
9974 		break;
9975 	case CHELSIO_T4_SET_OFLD_POLICY:
9976 		rc = set_offload_policy(sc, (struct t4_offload_policy *)data);
9977 		break;
9978 	default:
9979 		rc = ENOTTY;
9980 	}
9981 
9982 	return (rc);
9983 }
9984 
9985 void
9986 t4_db_full(struct adapter *sc)
9987 {
9988 
9989 	CXGBE_UNIMPLEMENTED(__func__);
9990 }
9991 
9992 void
9993 t4_db_dropped(struct adapter *sc)
9994 {
9995 
9996 	CXGBE_UNIMPLEMENTED(__func__);
9997 }
9998 
9999 #ifdef TCP_OFFLOAD
10000 static int
10001 toe_capability(struct vi_info *vi, int enable)
10002 {
10003 	int rc;
10004 	struct port_info *pi = vi->pi;
10005 	struct adapter *sc = pi->adapter;
10006 
10007 	ASSERT_SYNCHRONIZED_OP(sc);
10008 
10009 	if (!is_offload(sc))
10010 		return (ENODEV);
10011 
10012 	if (enable) {
10013 		if ((vi->ifp->if_capenable & IFCAP_TOE) != 0) {
10014 			/* TOE is already enabled. */
10015 			return (0);
10016 		}
10017 
10018 		/*
10019 		 * We need the port's queues around so that we're able to send
10020 		 * and receive CPLs to/from the TOE even if the ifnet for this
10021 		 * port has never been UP'd administratively.
10022 		 */
10023 		if (!(vi->flags & VI_INIT_DONE)) {
10024 			rc = vi_full_init(vi);
10025 			if (rc)
10026 				return (rc);
10027 		}
10028 		if (!(pi->vi[0].flags & VI_INIT_DONE)) {
10029 			rc = vi_full_init(&pi->vi[0]);
10030 			if (rc)
10031 				return (rc);
10032 		}
10033 
10034 		if (isset(&sc->offload_map, pi->port_id)) {
10035 			/* TOE is enabled on another VI of this port. */
10036 			pi->uld_vis++;
10037 			return (0);
10038 		}
10039 
10040 		if (!uld_active(sc, ULD_TOM)) {
10041 			rc = t4_activate_uld(sc, ULD_TOM);
10042 			if (rc == EAGAIN) {
10043 				log(LOG_WARNING,
10044 				    "You must kldload t4_tom.ko before trying "
10045 				    "to enable TOE on a cxgbe interface.\n");
10046 			}
10047 			if (rc != 0)
10048 				return (rc);
10049 			KASSERT(sc->tom_softc != NULL,
10050 			    ("%s: TOM activated but softc NULL", __func__));
10051 			KASSERT(uld_active(sc, ULD_TOM),
10052 			    ("%s: TOM activated but flag not set", __func__));
10053 		}
10054 
10055 		/* Activate iWARP and iSCSI too, if the modules are loaded. */
10056 		if (!uld_active(sc, ULD_IWARP))
10057 			(void) t4_activate_uld(sc, ULD_IWARP);
10058 		if (!uld_active(sc, ULD_ISCSI))
10059 			(void) t4_activate_uld(sc, ULD_ISCSI);
10060 
10061 		pi->uld_vis++;
10062 		setbit(&sc->offload_map, pi->port_id);
10063 	} else {
10064 		pi->uld_vis--;
10065 
10066 		if (!isset(&sc->offload_map, pi->port_id) || pi->uld_vis > 0)
10067 			return (0);
10068 
10069 		KASSERT(uld_active(sc, ULD_TOM),
10070 		    ("%s: TOM never initialized?", __func__));
10071 		clrbit(&sc->offload_map, pi->port_id);
10072 	}
10073 
10074 	return (0);
10075 }
10076 
10077 /*
10078  * Add an upper layer driver to the global list.
10079  */
10080 int
10081 t4_register_uld(struct uld_info *ui)
10082 {
10083 	int rc = 0;
10084 	struct uld_info *u;
10085 
10086 	sx_xlock(&t4_uld_list_lock);
10087 	SLIST_FOREACH(u, &t4_uld_list, link) {
10088 	    if (u->uld_id == ui->uld_id) {
10089 		    rc = EEXIST;
10090 		    goto done;
10091 	    }
10092 	}
10093 
10094 	SLIST_INSERT_HEAD(&t4_uld_list, ui, link);
10095 	ui->refcount = 0;
10096 done:
10097 	sx_xunlock(&t4_uld_list_lock);
10098 	return (rc);
10099 }
10100 
10101 int
10102 t4_unregister_uld(struct uld_info *ui)
10103 {
10104 	int rc = EINVAL;
10105 	struct uld_info *u;
10106 
10107 	sx_xlock(&t4_uld_list_lock);
10108 
10109 	SLIST_FOREACH(u, &t4_uld_list, link) {
10110 	    if (u == ui) {
10111 		    if (ui->refcount > 0) {
10112 			    rc = EBUSY;
10113 			    goto done;
10114 		    }
10115 
10116 		    SLIST_REMOVE(&t4_uld_list, ui, uld_info, link);
10117 		    rc = 0;
10118 		    goto done;
10119 	    }
10120 	}
10121 done:
10122 	sx_xunlock(&t4_uld_list_lock);
10123 	return (rc);
10124 }
10125 
10126 int
10127 t4_activate_uld(struct adapter *sc, int id)
10128 {
10129 	int rc;
10130 	struct uld_info *ui;
10131 
10132 	ASSERT_SYNCHRONIZED_OP(sc);
10133 
10134 	if (id < 0 || id > ULD_MAX)
10135 		return (EINVAL);
10136 	rc = EAGAIN;	/* kldoad the module with this ULD and try again. */
10137 
10138 	sx_slock(&t4_uld_list_lock);
10139 
10140 	SLIST_FOREACH(ui, &t4_uld_list, link) {
10141 		if (ui->uld_id == id) {
10142 			if (!(sc->flags & FULL_INIT_DONE)) {
10143 				rc = adapter_full_init(sc);
10144 				if (rc != 0)
10145 					break;
10146 			}
10147 
10148 			rc = ui->activate(sc);
10149 			if (rc == 0) {
10150 				setbit(&sc->active_ulds, id);
10151 				ui->refcount++;
10152 			}
10153 			break;
10154 		}
10155 	}
10156 
10157 	sx_sunlock(&t4_uld_list_lock);
10158 
10159 	return (rc);
10160 }
10161 
10162 int
10163 t4_deactivate_uld(struct adapter *sc, int id)
10164 {
10165 	int rc;
10166 	struct uld_info *ui;
10167 
10168 	ASSERT_SYNCHRONIZED_OP(sc);
10169 
10170 	if (id < 0 || id > ULD_MAX)
10171 		return (EINVAL);
10172 	rc = ENXIO;
10173 
10174 	sx_slock(&t4_uld_list_lock);
10175 
10176 	SLIST_FOREACH(ui, &t4_uld_list, link) {
10177 		if (ui->uld_id == id) {
10178 			rc = ui->deactivate(sc);
10179 			if (rc == 0) {
10180 				clrbit(&sc->active_ulds, id);
10181 				ui->refcount--;
10182 			}
10183 			break;
10184 		}
10185 	}
10186 
10187 	sx_sunlock(&t4_uld_list_lock);
10188 
10189 	return (rc);
10190 }
10191 
10192 int
10193 uld_active(struct adapter *sc, int uld_id)
10194 {
10195 
10196 	MPASS(uld_id >= 0 && uld_id <= ULD_MAX);
10197 
10198 	return (isset(&sc->active_ulds, uld_id));
10199 }
10200 #endif
10201 
10202 /*
10203  * t  = ptr to tunable.
10204  * nc = number of CPUs.
10205  * c  = compiled in default for that tunable.
10206  */
10207 static void
10208 calculate_nqueues(int *t, int nc, const int c)
10209 {
10210 	int nq;
10211 
10212 	if (*t > 0)
10213 		return;
10214 	nq = *t < 0 ? -*t : c;
10215 	*t = min(nc, nq);
10216 }
10217 
10218 /*
10219  * Come up with reasonable defaults for some of the tunables, provided they're
10220  * not set by the user (in which case we'll use the values as is).
10221  */
10222 static void
10223 tweak_tunables(void)
10224 {
10225 	int nc = mp_ncpus;	/* our snapshot of the number of CPUs */
10226 
10227 	if (t4_ntxq < 1) {
10228 #ifdef RSS
10229 		t4_ntxq = rss_getnumbuckets();
10230 #else
10231 		calculate_nqueues(&t4_ntxq, nc, NTXQ);
10232 #endif
10233 	}
10234 
10235 	calculate_nqueues(&t4_ntxq_vi, nc, NTXQ_VI);
10236 
10237 	if (t4_nrxq < 1) {
10238 #ifdef RSS
10239 		t4_nrxq = rss_getnumbuckets();
10240 #else
10241 		calculate_nqueues(&t4_nrxq, nc, NRXQ);
10242 #endif
10243 	}
10244 
10245 	calculate_nqueues(&t4_nrxq_vi, nc, NRXQ_VI);
10246 
10247 #if defined(TCP_OFFLOAD) || defined(RATELIMIT)
10248 	calculate_nqueues(&t4_nofldtxq, nc, NOFLDTXQ);
10249 	calculate_nqueues(&t4_nofldtxq_vi, nc, NOFLDTXQ_VI);
10250 #endif
10251 #ifdef TCP_OFFLOAD
10252 	calculate_nqueues(&t4_nofldrxq, nc, NOFLDRXQ);
10253 	calculate_nqueues(&t4_nofldrxq_vi, nc, NOFLDRXQ_VI);
10254 
10255 	if (t4_toecaps_allowed == -1)
10256 		t4_toecaps_allowed = FW_CAPS_CONFIG_TOE;
10257 
10258 	if (t4_rdmacaps_allowed == -1) {
10259 		t4_rdmacaps_allowed = FW_CAPS_CONFIG_RDMA_RDDP |
10260 		    FW_CAPS_CONFIG_RDMA_RDMAC;
10261 	}
10262 
10263 	if (t4_iscsicaps_allowed == -1) {
10264 		t4_iscsicaps_allowed = FW_CAPS_CONFIG_ISCSI_INITIATOR_PDU |
10265 		    FW_CAPS_CONFIG_ISCSI_TARGET_PDU |
10266 		    FW_CAPS_CONFIG_ISCSI_T10DIF;
10267 	}
10268 
10269 	if (t4_tmr_idx_ofld < 0 || t4_tmr_idx_ofld >= SGE_NTIMERS)
10270 		t4_tmr_idx_ofld = TMR_IDX_OFLD;
10271 
10272 	if (t4_pktc_idx_ofld < -1 || t4_pktc_idx_ofld >= SGE_NCOUNTERS)
10273 		t4_pktc_idx_ofld = PKTC_IDX_OFLD;
10274 #else
10275 	if (t4_toecaps_allowed == -1)
10276 		t4_toecaps_allowed = 0;
10277 
10278 	if (t4_rdmacaps_allowed == -1)
10279 		t4_rdmacaps_allowed = 0;
10280 
10281 	if (t4_iscsicaps_allowed == -1)
10282 		t4_iscsicaps_allowed = 0;
10283 #endif
10284 
10285 #ifdef DEV_NETMAP
10286 	calculate_nqueues(&t4_nnmtxq_vi, nc, NNMTXQ_VI);
10287 	calculate_nqueues(&t4_nnmrxq_vi, nc, NNMRXQ_VI);
10288 #endif
10289 
10290 	if (t4_tmr_idx < 0 || t4_tmr_idx >= SGE_NTIMERS)
10291 		t4_tmr_idx = TMR_IDX;
10292 
10293 	if (t4_pktc_idx < -1 || t4_pktc_idx >= SGE_NCOUNTERS)
10294 		t4_pktc_idx = PKTC_IDX;
10295 
10296 	if (t4_qsize_txq < 128)
10297 		t4_qsize_txq = 128;
10298 
10299 	if (t4_qsize_rxq < 128)
10300 		t4_qsize_rxq = 128;
10301 	while (t4_qsize_rxq & 7)
10302 		t4_qsize_rxq++;
10303 
10304 	t4_intr_types &= INTR_MSIX | INTR_MSI | INTR_INTX;
10305 
10306 	/*
10307 	 * Number of VIs to create per-port.  The first VI is the "main" regular
10308 	 * VI for the port.  The rest are additional virtual interfaces on the
10309 	 * same physical port.  Note that the main VI does not have native
10310 	 * netmap support but the extra VIs do.
10311 	 *
10312 	 * Limit the number of VIs per port to the number of available
10313 	 * MAC addresses per port.
10314 	 */
10315 	if (t4_num_vis < 1)
10316 		t4_num_vis = 1;
10317 	if (t4_num_vis > nitems(vi_mac_funcs)) {
10318 		t4_num_vis = nitems(vi_mac_funcs);
10319 		printf("cxgbe: number of VIs limited to %d\n", t4_num_vis);
10320 	}
10321 
10322 	if (pcie_relaxed_ordering < 0 || pcie_relaxed_ordering > 2) {
10323 		pcie_relaxed_ordering = 1;
10324 #if defined(__i386__) || defined(__amd64__)
10325 		if (cpu_vendor_id == CPU_VENDOR_INTEL)
10326 			pcie_relaxed_ordering = 0;
10327 #endif
10328 	}
10329 }
10330 
10331 #ifdef DDB
10332 static void
10333 t4_dump_tcb(struct adapter *sc, int tid)
10334 {
10335 	uint32_t base, i, j, off, pf, reg, save, tcb_addr, win_pos;
10336 
10337 	reg = PCIE_MEM_ACCESS_REG(A_PCIE_MEM_ACCESS_OFFSET, 2);
10338 	save = t4_read_reg(sc, reg);
10339 	base = sc->memwin[2].mw_base;
10340 
10341 	/* Dump TCB for the tid */
10342 	tcb_addr = t4_read_reg(sc, A_TP_CMM_TCB_BASE);
10343 	tcb_addr += tid * TCB_SIZE;
10344 
10345 	if (is_t4(sc)) {
10346 		pf = 0;
10347 		win_pos = tcb_addr & ~0xf;	/* start must be 16B aligned */
10348 	} else {
10349 		pf = V_PFNUM(sc->pf);
10350 		win_pos = tcb_addr & ~0x7f;	/* start must be 128B aligned */
10351 	}
10352 	t4_write_reg(sc, reg, win_pos | pf);
10353 	t4_read_reg(sc, reg);
10354 
10355 	off = tcb_addr - win_pos;
10356 	for (i = 0; i < 4; i++) {
10357 		uint32_t buf[8];
10358 		for (j = 0; j < 8; j++, off += 4)
10359 			buf[j] = htonl(t4_read_reg(sc, base + off));
10360 
10361 		db_printf("%08x %08x %08x %08x %08x %08x %08x %08x\n",
10362 		    buf[0], buf[1], buf[2], buf[3], buf[4], buf[5], buf[6],
10363 		    buf[7]);
10364 	}
10365 
10366 	t4_write_reg(sc, reg, save);
10367 	t4_read_reg(sc, reg);
10368 }
10369 
10370 static void
10371 t4_dump_devlog(struct adapter *sc)
10372 {
10373 	struct devlog_params *dparams = &sc->params.devlog;
10374 	struct fw_devlog_e e;
10375 	int i, first, j, m, nentries, rc;
10376 	uint64_t ftstamp = UINT64_MAX;
10377 
10378 	if (dparams->start == 0) {
10379 		db_printf("devlog params not valid\n");
10380 		return;
10381 	}
10382 
10383 	nentries = dparams->size / sizeof(struct fw_devlog_e);
10384 	m = fwmtype_to_hwmtype(dparams->memtype);
10385 
10386 	/* Find the first entry. */
10387 	first = -1;
10388 	for (i = 0; i < nentries && !db_pager_quit; i++) {
10389 		rc = -t4_mem_read(sc, m, dparams->start + i * sizeof(e),
10390 		    sizeof(e), (void *)&e);
10391 		if (rc != 0)
10392 			break;
10393 
10394 		if (e.timestamp == 0)
10395 			break;
10396 
10397 		e.timestamp = be64toh(e.timestamp);
10398 		if (e.timestamp < ftstamp) {
10399 			ftstamp = e.timestamp;
10400 			first = i;
10401 		}
10402 	}
10403 
10404 	if (first == -1)
10405 		return;
10406 
10407 	i = first;
10408 	do {
10409 		rc = -t4_mem_read(sc, m, dparams->start + i * sizeof(e),
10410 		    sizeof(e), (void *)&e);
10411 		if (rc != 0)
10412 			return;
10413 
10414 		if (e.timestamp == 0)
10415 			return;
10416 
10417 		e.timestamp = be64toh(e.timestamp);
10418 		e.seqno = be32toh(e.seqno);
10419 		for (j = 0; j < 8; j++)
10420 			e.params[j] = be32toh(e.params[j]);
10421 
10422 		db_printf("%10d  %15ju  %8s  %8s  ",
10423 		    e.seqno, e.timestamp,
10424 		    (e.level < nitems(devlog_level_strings) ?
10425 			devlog_level_strings[e.level] : "UNKNOWN"),
10426 		    (e.facility < nitems(devlog_facility_strings) ?
10427 			devlog_facility_strings[e.facility] : "UNKNOWN"));
10428 		db_printf(e.fmt, e.params[0], e.params[1], e.params[2],
10429 		    e.params[3], e.params[4], e.params[5], e.params[6],
10430 		    e.params[7]);
10431 
10432 		if (++i == nentries)
10433 			i = 0;
10434 	} while (i != first && !db_pager_quit);
10435 }
10436 
10437 static struct command_table db_t4_table = LIST_HEAD_INITIALIZER(db_t4_table);
10438 _DB_SET(_show, t4, NULL, db_show_table, 0, &db_t4_table);
10439 
10440 DB_FUNC(devlog, db_show_devlog, db_t4_table, CS_OWN, NULL)
10441 {
10442 	device_t dev;
10443 	int t;
10444 	bool valid;
10445 
10446 	valid = false;
10447 	t = db_read_token();
10448 	if (t == tIDENT) {
10449 		dev = device_lookup_by_name(db_tok_string);
10450 		valid = true;
10451 	}
10452 	db_skip_to_eol();
10453 	if (!valid) {
10454 		db_printf("usage: show t4 devlog <nexus>\n");
10455 		return;
10456 	}
10457 
10458 	if (dev == NULL) {
10459 		db_printf("device not found\n");
10460 		return;
10461 	}
10462 
10463 	t4_dump_devlog(device_get_softc(dev));
10464 }
10465 
10466 DB_FUNC(tcb, db_show_t4tcb, db_t4_table, CS_OWN, NULL)
10467 {
10468 	device_t dev;
10469 	int radix, tid, t;
10470 	bool valid;
10471 
10472 	valid = false;
10473 	radix = db_radix;
10474 	db_radix = 10;
10475 	t = db_read_token();
10476 	if (t == tIDENT) {
10477 		dev = device_lookup_by_name(db_tok_string);
10478 		t = db_read_token();
10479 		if (t == tNUMBER) {
10480 			tid = db_tok_number;
10481 			valid = true;
10482 		}
10483 	}
10484 	db_radix = radix;
10485 	db_skip_to_eol();
10486 	if (!valid) {
10487 		db_printf("usage: show t4 tcb <nexus> <tid>\n");
10488 		return;
10489 	}
10490 
10491 	if (dev == NULL) {
10492 		db_printf("device not found\n");
10493 		return;
10494 	}
10495 	if (tid < 0) {
10496 		db_printf("invalid tid\n");
10497 		return;
10498 	}
10499 
10500 	t4_dump_tcb(device_get_softc(dev), tid);
10501 }
10502 #endif
10503 
10504 /*
10505  * Borrowed from cesa_prep_aes_key().
10506  *
10507  * NB: The crypto engine wants the words in the decryption key in reverse
10508  * order.
10509  */
10510 void
10511 t4_aes_getdeckey(void *dec_key, const void *enc_key, unsigned int kbits)
10512 {
10513 	uint32_t ek[4 * (RIJNDAEL_MAXNR + 1)];
10514 	uint32_t *dkey;
10515 	int i;
10516 
10517 	rijndaelKeySetupEnc(ek, enc_key, kbits);
10518 	dkey = dec_key;
10519 	dkey += (kbits / 8) / 4;
10520 
10521 	switch (kbits) {
10522 	case 128:
10523 		for (i = 0; i < 4; i++)
10524 			*--dkey = htobe32(ek[4 * 10 + i]);
10525 		break;
10526 	case 192:
10527 		for (i = 0; i < 2; i++)
10528 			*--dkey = htobe32(ek[4 * 11 + 2 + i]);
10529 		for (i = 0; i < 4; i++)
10530 			*--dkey = htobe32(ek[4 * 12 + i]);
10531 		break;
10532 	case 256:
10533 		for (i = 0; i < 4; i++)
10534 			*--dkey = htobe32(ek[4 * 13 + i]);
10535 		for (i = 0; i < 4; i++)
10536 			*--dkey = htobe32(ek[4 * 14 + i]);
10537 		break;
10538 	}
10539 	MPASS(dkey == dec_key);
10540 }
10541 
10542 static struct sx mlu;	/* mod load unload */
10543 SX_SYSINIT(cxgbe_mlu, &mlu, "cxgbe mod load/unload");
10544 
10545 static int
10546 mod_event(module_t mod, int cmd, void *arg)
10547 {
10548 	int rc = 0;
10549 	static int loaded = 0;
10550 
10551 	switch (cmd) {
10552 	case MOD_LOAD:
10553 		sx_xlock(&mlu);
10554 		if (loaded++ == 0) {
10555 			t4_sge_modload();
10556 			t4_register_shared_cpl_handler(CPL_SET_TCB_RPL,
10557 			    t4_filter_rpl, CPL_COOKIE_FILTER);
10558 			t4_register_shared_cpl_handler(CPL_L2T_WRITE_RPL,
10559 			    do_l2t_write_rpl, CPL_COOKIE_FILTER);
10560 			t4_register_shared_cpl_handler(CPL_ACT_OPEN_RPL,
10561 			    t4_hashfilter_ao_rpl, CPL_COOKIE_HASHFILTER);
10562 			t4_register_shared_cpl_handler(CPL_SET_TCB_RPL,
10563 			    t4_hashfilter_tcb_rpl, CPL_COOKIE_HASHFILTER);
10564 			t4_register_shared_cpl_handler(CPL_ABORT_RPL_RSS,
10565 			    t4_del_hashfilter_rpl, CPL_COOKIE_HASHFILTER);
10566 			t4_register_cpl_handler(CPL_TRACE_PKT, t4_trace_pkt);
10567 			t4_register_cpl_handler(CPL_T5_TRACE_PKT, t5_trace_pkt);
10568 			t4_register_cpl_handler(CPL_SMT_WRITE_RPL,
10569 			    do_smt_write_rpl);
10570 			sx_init(&t4_list_lock, "T4/T5 adapters");
10571 			SLIST_INIT(&t4_list);
10572 #ifdef TCP_OFFLOAD
10573 			sx_init(&t4_uld_list_lock, "T4/T5 ULDs");
10574 			SLIST_INIT(&t4_uld_list);
10575 #endif
10576 #ifdef INET6
10577 			t4_clip_modload();
10578 #endif
10579 			t4_tracer_modload();
10580 			tweak_tunables();
10581 		}
10582 		sx_xunlock(&mlu);
10583 		break;
10584 
10585 	case MOD_UNLOAD:
10586 		sx_xlock(&mlu);
10587 		if (--loaded == 0) {
10588 			int tries;
10589 
10590 			sx_slock(&t4_list_lock);
10591 			if (!SLIST_EMPTY(&t4_list)) {
10592 				rc = EBUSY;
10593 				sx_sunlock(&t4_list_lock);
10594 				goto done_unload;
10595 			}
10596 #ifdef TCP_OFFLOAD
10597 			sx_slock(&t4_uld_list_lock);
10598 			if (!SLIST_EMPTY(&t4_uld_list)) {
10599 				rc = EBUSY;
10600 				sx_sunlock(&t4_uld_list_lock);
10601 				sx_sunlock(&t4_list_lock);
10602 				goto done_unload;
10603 			}
10604 #endif
10605 			tries = 0;
10606 			while (tries++ < 5 && t4_sge_extfree_refs() != 0) {
10607 				uprintf("%ju clusters with custom free routine "
10608 				    "still is use.\n", t4_sge_extfree_refs());
10609 				pause("t4unload", 2 * hz);
10610 			}
10611 #ifdef TCP_OFFLOAD
10612 			sx_sunlock(&t4_uld_list_lock);
10613 #endif
10614 			sx_sunlock(&t4_list_lock);
10615 
10616 			if (t4_sge_extfree_refs() == 0) {
10617 				t4_tracer_modunload();
10618 #ifdef INET6
10619 				t4_clip_modunload();
10620 #endif
10621 #ifdef TCP_OFFLOAD
10622 				sx_destroy(&t4_uld_list_lock);
10623 #endif
10624 				sx_destroy(&t4_list_lock);
10625 				t4_sge_modunload();
10626 				loaded = 0;
10627 			} else {
10628 				rc = EBUSY;
10629 				loaded++;	/* undo earlier decrement */
10630 			}
10631 		}
10632 done_unload:
10633 		sx_xunlock(&mlu);
10634 		break;
10635 	}
10636 
10637 	return (rc);
10638 }
10639 
10640 static devclass_t t4_devclass, t5_devclass, t6_devclass;
10641 static devclass_t cxgbe_devclass, cxl_devclass, cc_devclass;
10642 static devclass_t vcxgbe_devclass, vcxl_devclass, vcc_devclass;
10643 
10644 DRIVER_MODULE(t4nex, pci, t4_driver, t4_devclass, mod_event, 0);
10645 MODULE_VERSION(t4nex, 1);
10646 MODULE_DEPEND(t4nex, firmware, 1, 1, 1);
10647 #ifdef DEV_NETMAP
10648 MODULE_DEPEND(t4nex, netmap, 1, 1, 1);
10649 #endif /* DEV_NETMAP */
10650 
10651 DRIVER_MODULE(t5nex, pci, t5_driver, t5_devclass, mod_event, 0);
10652 MODULE_VERSION(t5nex, 1);
10653 MODULE_DEPEND(t5nex, firmware, 1, 1, 1);
10654 #ifdef DEV_NETMAP
10655 MODULE_DEPEND(t5nex, netmap, 1, 1, 1);
10656 #endif /* DEV_NETMAP */
10657 
10658 DRIVER_MODULE(t6nex, pci, t6_driver, t6_devclass, mod_event, 0);
10659 MODULE_VERSION(t6nex, 1);
10660 MODULE_DEPEND(t6nex, firmware, 1, 1, 1);
10661 #ifdef DEV_NETMAP
10662 MODULE_DEPEND(t6nex, netmap, 1, 1, 1);
10663 #endif /* DEV_NETMAP */
10664 
10665 DRIVER_MODULE(cxgbe, t4nex, cxgbe_driver, cxgbe_devclass, 0, 0);
10666 MODULE_VERSION(cxgbe, 1);
10667 
10668 DRIVER_MODULE(cxl, t5nex, cxl_driver, cxl_devclass, 0, 0);
10669 MODULE_VERSION(cxl, 1);
10670 
10671 DRIVER_MODULE(cc, t6nex, cc_driver, cc_devclass, 0, 0);
10672 MODULE_VERSION(cc, 1);
10673 
10674 DRIVER_MODULE(vcxgbe, cxgbe, vcxgbe_driver, vcxgbe_devclass, 0, 0);
10675 MODULE_VERSION(vcxgbe, 1);
10676 
10677 DRIVER_MODULE(vcxl, cxl, vcxl_driver, vcxl_devclass, 0, 0);
10678 MODULE_VERSION(vcxl, 1);
10679 
10680 DRIVER_MODULE(vcc, cc, vcc_driver, vcc_devclass, 0, 0);
10681 MODULE_VERSION(vcc, 1);
10682