xref: /freebsd/sys/dev/cxgbe/adapter.h (revision 06c3fb27)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
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 
31 #ifndef __T4_ADAPTER_H__
32 #define __T4_ADAPTER_H__
33 
34 #include <sys/kernel.h>
35 #include <sys/bus.h>
36 #include <sys/counter.h>
37 #include <sys/rman.h>
38 #include <sys/types.h>
39 #include <sys/lock.h>
40 #include <sys/malloc.h>
41 #include <sys/rwlock.h>
42 #include <sys/seqc.h>
43 #include <sys/sx.h>
44 #include <sys/vmem.h>
45 #include <vm/uma.h>
46 
47 #include <dev/pci/pcivar.h>
48 #include <dev/pci/pcireg.h>
49 #include <machine/bus.h>
50 #include <sys/socket.h>
51 #include <sys/sysctl.h>
52 #include <sys/taskqueue.h>
53 #include <net/ethernet.h>
54 #include <net/if.h>
55 #include <net/if_var.h>
56 #include <net/if_media.h>
57 #include <net/pfil.h>
58 #include <netinet/in.h>
59 #include <netinet/tcp_lro.h>
60 
61 #include "offload.h"
62 #include "t4_ioctl.h"
63 #include "common/t4_msg.h"
64 #include "firmware/t4fw_interface.h"
65 
66 #define KTR_CXGBE	KTR_SPARE3
67 MALLOC_DECLARE(M_CXGBE);
68 #define CXGBE_UNIMPLEMENTED(s) \
69     panic("%s (%s, line %d) not implemented yet.", s, __FILE__, __LINE__)
70 
71 /*
72  * Same as LIST_HEAD from queue.h.  This is to avoid conflict with LinuxKPI's
73  * LIST_HEAD when building iw_cxgbe.
74  */
75 #define	CXGBE_LIST_HEAD(name, type)					\
76 struct name {								\
77 	struct type *lh_first;	/* first element */			\
78 }
79 
80 #ifndef SYSCTL_ADD_UQUAD
81 #define SYSCTL_ADD_UQUAD SYSCTL_ADD_QUAD
82 #define sysctl_handle_64 sysctl_handle_quad
83 #define CTLTYPE_U64 CTLTYPE_QUAD
84 #endif
85 
86 SYSCTL_DECL(_hw_cxgbe);
87 
88 struct adapter;
89 typedef struct adapter adapter_t;
90 
91 enum {
92 	/*
93 	 * All ingress queues use this entry size.  Note that the firmware event
94 	 * queue and any iq expecting CPL_RX_PKT in the descriptor needs this to
95 	 * be at least 64.
96 	 */
97 	IQ_ESIZE = 64,
98 
99 	/* Default queue sizes for all kinds of ingress queues */
100 	FW_IQ_QSIZE = 256,
101 	RX_IQ_QSIZE = 1024,
102 
103 	/* All egress queues use this entry size */
104 	EQ_ESIZE = 64,
105 
106 	/* Default queue sizes for all kinds of egress queues */
107 	CTRL_EQ_QSIZE = 1024,
108 	TX_EQ_QSIZE = 1024,
109 
110 	SW_ZONE_SIZES = 4,	/* cluster, jumbop, jumbo9k, jumbo16k */
111 	CL_METADATA_SIZE = CACHE_LINE_SIZE,
112 
113 	SGE_MAX_WR_NDESC = SGE_MAX_WR_LEN / EQ_ESIZE, /* max WR size in desc */
114 	TX_SGL_SEGS = 39,
115 	TX_SGL_SEGS_TSO = 38,
116 	TX_SGL_SEGS_VM = 38,
117 	TX_SGL_SEGS_VM_TSO = 37,
118 	TX_SGL_SEGS_EO_TSO = 30,	/* XXX: lower for IPv6. */
119 	TX_SGL_SEGS_VXLAN_TSO = 37,
120 	TX_WR_FLITS = SGE_MAX_WR_LEN / 8
121 };
122 
123 enum {
124 	/* adapter intr_type */
125 	INTR_INTX	= (1 << 0),
126 	INTR_MSI	= (1 << 1),
127 	INTR_MSIX	= (1 << 2)
128 };
129 
130 enum {
131 	XGMAC_MTU	= (1 << 0),
132 	XGMAC_PROMISC	= (1 << 1),
133 	XGMAC_ALLMULTI	= (1 << 2),
134 	XGMAC_VLANEX	= (1 << 3),
135 	XGMAC_UCADDR	= (1 << 4),
136 	XGMAC_MCADDRS	= (1 << 5),
137 
138 	XGMAC_ALL	= 0xffff
139 };
140 
141 enum {
142 	/* flags understood by begin_synchronized_op */
143 	HOLD_LOCK	= (1 << 0),
144 	SLEEP_OK	= (1 << 1),
145 	INTR_OK		= (1 << 2),
146 
147 	/* flags understood by end_synchronized_op */
148 	LOCK_HELD	= HOLD_LOCK,
149 };
150 
151 enum {
152 	/* adapter flags.  synch_op or adapter_lock. */
153 	FULL_INIT_DONE	= (1 << 0),
154 	FW_OK		= (1 << 1),
155 	CHK_MBOX_ACCESS	= (1 << 2),
156 	MASTER_PF	= (1 << 3),
157 	BUF_PACKING_OK	= (1 << 6),
158 	IS_VF		= (1 << 7),
159 	KERN_TLS_ON	= (1 << 8),	/* HW is configured for KERN_TLS */
160 	CXGBE_BUSY	= (1 << 9),
161 
162 	/* adapter error_flags.  reg_lock for HW_OFF_LIMITS, atomics for the rest. */
163 	ADAP_STOPPED	= (1 << 0),	/* Adapter has been stopped. */
164 	ADAP_FATAL_ERR	= (1 << 1),	/* Encountered a fatal error. */
165 	HW_OFF_LIMITS	= (1 << 2),	/* off limits to all except reset_thread */
166 	ADAP_CIM_ERR	= (1 << 3),	/* Error was related to FW/CIM. */
167 
168 	/* port flags */
169 	HAS_TRACEQ	= (1 << 3),
170 	FIXED_IFMEDIA	= (1 << 4),	/* ifmedia list doesn't change. */
171 
172 	/* VI flags */
173 	DOOMED		= (1 << 0),
174 	VI_INIT_DONE	= (1 << 1),
175 	/* 1 << 2 is unused, was VI_SYSCTL_CTX */
176 	TX_USES_VM_WR	= (1 << 3),
177 	VI_SKIP_STATS	= (1 << 4),
178 
179 	/* adapter debug_flags */
180 	DF_DUMP_MBOX		= (1 << 0),	/* Log all mbox cmd/rpl. */
181 	DF_LOAD_FW_ANYTIME	= (1 << 1),	/* Allow LOAD_FW after init */
182 	DF_DISABLE_TCB_CACHE	= (1 << 2),	/* Disable TCB cache (T6+) */
183 	DF_DISABLE_CFG_RETRY	= (1 << 3),	/* Disable fallback config */
184 	DF_VERBOSE_SLOWINTR	= (1 << 4),	/* Chatty slow intr handler */
185 };
186 
187 #define IS_DOOMED(vi)	((vi)->flags & DOOMED)
188 #define SET_DOOMED(vi)	do {(vi)->flags |= DOOMED;} while (0)
189 #define IS_BUSY(sc)	((sc)->flags & CXGBE_BUSY)
190 #define SET_BUSY(sc)	do {(sc)->flags |= CXGBE_BUSY;} while (0)
191 #define CLR_BUSY(sc)	do {(sc)->flags &= ~CXGBE_BUSY;} while (0)
192 
193 struct vi_info {
194 	device_t dev;
195 	struct port_info *pi;
196 	struct adapter *adapter;
197 
198 	if_t ifp;
199 	struct pfil_head *pfil;
200 
201 	unsigned long flags;
202 	int if_flags;
203 
204 	uint16_t *rss, *nm_rss;
205 	uint16_t viid;		/* opaque VI identifier */
206 	uint16_t smt_idx;
207 	uint16_t vin;
208 	uint8_t vfvld;
209 	int16_t  xact_addr_filt;/* index of exact MAC address filter */
210 	uint16_t rss_size;	/* size of VI's RSS table slice */
211 	uint16_t rss_base;	/* start of VI's RSS table slice */
212 	int hashen;
213 
214 	int nintr;
215 	int first_intr;
216 
217 	/* These need to be int as they are used in sysctl */
218 	int ntxq;		/* # of tx queues */
219 	int first_txq;		/* index of first tx queue */
220 	int rsrv_noflowq;	/* Reserve queue 0 for non-flowid packets */
221 	int nrxq;		/* # of rx queues */
222 	int first_rxq;		/* index of first rx queue */
223 	int nofldtxq;		/* # of offload tx queues */
224 	int first_ofld_txq;	/* index of first offload tx queue */
225 	int nofldrxq;		/* # of offload rx queues */
226 	int first_ofld_rxq;	/* index of first offload rx queue */
227 	int nnmtxq;
228 	int first_nm_txq;
229 	int nnmrxq;
230 	int first_nm_rxq;
231 	int tmr_idx;
232 	int ofld_tmr_idx;
233 	int pktc_idx;
234 	int ofld_pktc_idx;
235 	int qsize_rxq;
236 	int qsize_txq;
237 
238 	struct timeval last_refreshed;
239 	struct fw_vi_stats_vf stats;
240 	struct mtx tick_mtx;
241 	struct callout tick;
242 
243 	struct sysctl_ctx_list ctx;
244 	struct sysctl_oid *rxq_oid;
245 	struct sysctl_oid *txq_oid;
246 	struct sysctl_oid *nm_rxq_oid;
247 	struct sysctl_oid *nm_txq_oid;
248 	struct sysctl_oid *ofld_rxq_oid;
249 	struct sysctl_oid *ofld_txq_oid;
250 
251 	uint8_t hw_addr[ETHER_ADDR_LEN]; /* factory MAC address, won't change */
252 	u_int txq_rr;
253 	u_int rxq_rr;
254 };
255 
256 struct tx_ch_rl_params {
257 	enum fw_sched_params_rate ratemode;	/* %port (REL) or kbps (ABS) */
258 	uint32_t maxrate;
259 };
260 
261 /* CLRL state */
262 enum clrl_state {
263 	CS_UNINITIALIZED = 0,
264 	CS_PARAMS_SET,			/* sw parameters have been set. */
265 	CS_HW_UPDATE_REQUESTED,		/* async HW update requested. */
266 	CS_HW_UPDATE_IN_PROGRESS,	/* sync hw update in progress. */
267 	CS_HW_CONFIGURED		/* configured in the hardware. */
268 };
269 
270 /* CLRL flags */
271 enum {
272 	CF_USER		= (1 << 0),	/* was configured by driver ioctl. */
273 };
274 
275 struct tx_cl_rl_params {
276 	enum clrl_state state;
277 	int refcount;
278 	uint8_t flags;
279 	enum fw_sched_params_rate ratemode;	/* %port REL or ABS value */
280 	enum fw_sched_params_unit rateunit;	/* kbps or pps (when ABS) */
281 	enum fw_sched_params_mode mode;		/* aggr or per-flow */
282 	uint32_t maxrate;
283 	uint16_t pktsize;
284 	uint16_t burstsize;
285 };
286 
287 /* Tx scheduler parameters for a channel/port */
288 struct tx_sched_params {
289 	/* Channel Rate Limiter */
290 	struct tx_ch_rl_params ch_rl;
291 
292 	/* Class WRR */
293 	/* XXX */
294 
295 	/* Class Rate Limiter (including the default pktsize and burstsize). */
296 	int pktsize;
297 	int burstsize;
298 	struct tx_cl_rl_params cl_rl[];
299 };
300 
301 struct port_info {
302 	device_t dev;
303 	struct adapter *adapter;
304 
305 	struct vi_info *vi;
306 	int nvi;
307 	int up_vis;
308 	int uld_vis;
309 	bool vxlan_tcam_entry;
310 
311 	struct tx_sched_params *sched_params;
312 
313 	struct mtx pi_lock;
314 	char lockname[16];
315 	unsigned long flags;
316 
317 	uint8_t  lport;		/* associated offload logical port */
318 	int8_t   mdio_addr;
319 	uint8_t  port_type;
320 	uint8_t  mod_type;
321 	uint8_t  port_id;
322 	uint8_t  tx_chan;
323 	uint8_t  mps_bg_map;	/* rx MPS buffer group bitmap */
324 	uint8_t  rx_e_chan_map;	/* rx TP e-channel bitmap */
325 	uint8_t  rx_c_chan;	/* rx TP c-channel */
326 
327 	struct link_config link_cfg;
328 	struct ifmedia media;
329 
330 	struct port_stats stats;
331 	u_int tnl_cong_drops;
332 	u_int tx_parse_error;
333 	int fcs_reg;
334 	uint64_t fcs_base;
335 
336 	struct sysctl_ctx_list ctx;
337 };
338 
339 #define	IS_MAIN_VI(vi)		((vi) == &((vi)->pi->vi[0]))
340 
341 struct cluster_metadata {
342 	uma_zone_t zone;
343 	caddr_t cl;
344 	u_int refcount;
345 };
346 
347 struct fl_sdesc {
348 	caddr_t cl;
349 	uint16_t nmbuf;	/* # of driver originated mbufs with ref on cluster */
350 	int16_t moff;	/* offset of metadata from cl */
351 	uint8_t zidx;
352 };
353 
354 struct tx_desc {
355 	__be64 flit[8];
356 };
357 
358 struct tx_sdesc {
359 	struct mbuf *m;		/* m_nextpkt linked chain of frames */
360 	uint8_t desc_used;	/* # of hardware descriptors used by the WR */
361 };
362 
363 
364 #define IQ_PAD (IQ_ESIZE - sizeof(struct rsp_ctrl) - sizeof(struct rss_header))
365 struct iq_desc {
366 	struct rss_header rss;
367 	uint8_t cpl[IQ_PAD];
368 	struct rsp_ctrl rsp;
369 };
370 #undef IQ_PAD
371 CTASSERT(sizeof(struct iq_desc) == IQ_ESIZE);
372 
373 enum {
374 	/* iq type */
375 	IQ_OTHER	= FW_IQ_IQTYPE_OTHER,
376 	IQ_ETH		= FW_IQ_IQTYPE_NIC,
377 	IQ_OFLD		= FW_IQ_IQTYPE_OFLD,
378 
379 	/* iq flags */
380 	IQ_SW_ALLOCATED	= (1 << 0),	/* sw resources allocated */
381 	IQ_HAS_FL	= (1 << 1),	/* iq associated with a freelist */
382 	IQ_RX_TIMESTAMP	= (1 << 2),	/* provide the SGE rx timestamp */
383 	IQ_LRO_ENABLED	= (1 << 3),	/* iq is an eth rxq with LRO enabled */
384 	IQ_ADJ_CREDIT	= (1 << 4),	/* hw is off by 1 credit for this iq */
385 	IQ_HW_ALLOCATED	= (1 << 5),	/* fw/hw resources allocated */
386 
387 	/* iq state */
388 	IQS_DISABLED	= 0,
389 	IQS_BUSY	= 1,
390 	IQS_IDLE	= 2,
391 
392 	/* netmap related flags */
393 	NM_OFF	= 0,
394 	NM_ON	= 1,
395 	NM_BUSY	= 2,
396 };
397 
398 enum {
399 	CPL_COOKIE_RESERVED = 0,
400 	CPL_COOKIE_FILTER,
401 	CPL_COOKIE_DDP0,
402 	CPL_COOKIE_DDP1,
403 	CPL_COOKIE_TOM,
404 	CPL_COOKIE_HASHFILTER,
405 	CPL_COOKIE_ETHOFLD,
406 	CPL_COOKIE_KERN_TLS,
407 
408 	NUM_CPL_COOKIES = 8	/* Limited by M_COOKIE.  Do not increase. */
409 };
410 
411 struct sge_iq;
412 struct rss_header;
413 typedef int (*cpl_handler_t)(struct sge_iq *, const struct rss_header *,
414     struct mbuf *);
415 typedef int (*an_handler_t)(struct sge_iq *, const struct rsp_ctrl *);
416 typedef int (*fw_msg_handler_t)(struct adapter *, const __be64 *);
417 
418 /*
419  * Ingress Queue: T4 is producer, driver is consumer.
420  */
421 struct sge_iq {
422 	uint16_t flags;
423 	uint8_t qtype;
424 	volatile int state;
425 	struct adapter *adapter;
426 	struct iq_desc  *desc;	/* KVA of descriptor ring */
427 	int8_t   intr_pktc_idx;	/* packet count threshold index */
428 	uint8_t  gen;		/* generation bit */
429 	uint8_t  intr_params;	/* interrupt holdoff parameters */
430 	int8_t   cong_drop;	/* congestion drop settings for the queue */
431 	uint16_t qsize;		/* size (# of entries) of the queue */
432 	uint16_t sidx;		/* index of the entry with the status page */
433 	uint16_t cidx;		/* consumer index */
434 	uint16_t cntxt_id;	/* SGE context id for the iq */
435 	uint16_t abs_id;	/* absolute SGE id for the iq */
436 	int16_t intr_idx;	/* interrupt used by the queue */
437 
438 	STAILQ_ENTRY(sge_iq) link;
439 
440 	bus_dma_tag_t desc_tag;
441 	bus_dmamap_t desc_map;
442 	bus_addr_t ba;		/* bus address of descriptor ring */
443 };
444 
445 enum {
446 	/* eq type */
447 	EQ_CTRL		= 1,
448 	EQ_ETH		= 2,
449 	EQ_OFLD		= 3,
450 
451 	/* eq flags */
452 	EQ_SW_ALLOCATED	= (1 << 0),	/* sw resources allocated */
453 	EQ_HW_ALLOCATED	= (1 << 1),	/* hw/fw resources allocated */
454 	EQ_ENABLED	= (1 << 3),	/* open for business */
455 	EQ_QFLUSH	= (1 << 4),	/* if_qflush in progress */
456 };
457 
458 /* Listed in order of preference.  Update t4_sysctls too if you change these */
459 enum {DOORBELL_UDB, DOORBELL_WCWR, DOORBELL_UDBWC, DOORBELL_KDB};
460 
461 /*
462  * Egress Queue: driver is producer, T4 is consumer.
463  *
464  * Note: A free list is an egress queue (driver produces the buffers and T4
465  * consumes them) but it's special enough to have its own struct (see sge_fl).
466  */
467 struct sge_eq {
468 	unsigned int flags;	/* MUST be first */
469 	unsigned int cntxt_id;	/* SGE context id for the eq */
470 	unsigned int abs_id;	/* absolute SGE id for the eq */
471 	uint8_t type;		/* EQ_CTRL/EQ_ETH/EQ_OFLD */
472 	uint8_t doorbells;
473 	uint8_t tx_chan;	/* tx channel used by the eq */
474 	struct mtx eq_lock;
475 
476 	struct tx_desc *desc;	/* KVA of descriptor ring */
477 	volatile uint32_t *udb;	/* KVA of doorbell (lies within BAR2) */
478 	u_int udb_qid;		/* relative qid within the doorbell page */
479 	uint16_t sidx;		/* index of the entry with the status page */
480 	uint16_t cidx;		/* consumer idx (desc idx) */
481 	uint16_t pidx;		/* producer idx (desc idx) */
482 	uint16_t equeqidx;	/* EQUEQ last requested at this pidx */
483 	uint16_t dbidx;		/* pidx of the most recent doorbell */
484 	uint16_t iqid;		/* cached iq->cntxt_id (see iq below) */
485 	volatile u_int equiq;	/* EQUIQ outstanding */
486 	struct sge_iq *iq;	/* iq that receives egr_update for the eq */
487 
488 	bus_dma_tag_t desc_tag;
489 	bus_dmamap_t desc_map;
490 	bus_addr_t ba;		/* bus address of descriptor ring */
491 	char lockname[16];
492 };
493 
494 struct rx_buf_info {
495 	uma_zone_t zone;	/* zone that this cluster comes from */
496 	uint16_t size1;		/* same as size of cluster: 2K/4K/9K/16K.
497 				 * hwsize[hwidx1] = size1.  No spare. */
498 	uint16_t size2;		/* hwsize[hwidx2] = size2.
499 				 * spare in cluster = size1 - size2. */
500 	int8_t hwidx1;		/* SGE bufsize idx for size1 */
501 	int8_t hwidx2;		/* SGE bufsize idx for size2 */
502 	uint8_t type;		/* EXT_xxx type of the cluster */
503 };
504 
505 enum {
506 	NUM_MEMWIN = 3,
507 
508 	MEMWIN0_APERTURE = 2048,
509 	MEMWIN0_BASE     = 0x1b800,
510 
511 	MEMWIN1_APERTURE = 32768,
512 	MEMWIN1_BASE     = 0x28000,
513 
514 	MEMWIN2_APERTURE_T4 = 65536,
515 	MEMWIN2_BASE_T4     = 0x30000,
516 
517 	MEMWIN2_APERTURE_T5 = 128 * 1024,
518 	MEMWIN2_BASE_T5     = 0x60000,
519 };
520 
521 struct memwin {
522 	struct rwlock mw_lock __aligned(CACHE_LINE_SIZE);
523 	uint32_t mw_base;	/* constant after setup_memwin */
524 	uint32_t mw_aperture;	/* ditto */
525 	uint32_t mw_curpos;	/* protected by mw_lock */
526 };
527 
528 enum {
529 	FL_STARVING	= (1 << 0), /* on the adapter's list of starving fl's */
530 	FL_DOOMED	= (1 << 1), /* about to be destroyed */
531 	FL_BUF_PACKING	= (1 << 2), /* buffer packing enabled */
532 	FL_BUF_RESUME	= (1 << 3), /* resume from the middle of the frame */
533 };
534 
535 #define FL_RUNNING_LOW(fl) \
536     (IDXDIFF(fl->dbidx * 8, fl->cidx, fl->sidx * 8) <= fl->lowat)
537 #define FL_NOT_RUNNING_LOW(fl) \
538     (IDXDIFF(fl->dbidx * 8, fl->cidx, fl->sidx * 8) >= 2 * fl->lowat)
539 
540 struct sge_fl {
541 	struct mtx fl_lock;
542 	__be64 *desc;		/* KVA of descriptor ring, ptr to addresses */
543 	struct fl_sdesc *sdesc;	/* KVA of software descriptor ring */
544 	uint16_t zidx;		/* refill zone idx */
545 	uint16_t safe_zidx;
546 	uint16_t lowat;		/* # of buffers <= this means fl needs help */
547 	int flags;
548 	uint16_t buf_boundary;
549 
550 	/* The 16b idx all deal with hw descriptors */
551 	uint16_t dbidx;		/* hw pidx after last doorbell */
552 	uint16_t sidx;		/* index of status page */
553 	volatile uint16_t hw_cidx;
554 
555 	/* The 32b idx are all buffer idx, not hardware descriptor idx */
556 	uint32_t cidx;		/* consumer index */
557 	uint32_t pidx;		/* producer index */
558 
559 	uint32_t dbval;
560 	u_int rx_offset;	/* offset in fl buf (when buffer packing) */
561 	volatile uint32_t *udb;
562 
563 	uint64_t cl_allocated;	/* # of clusters allocated */
564 	uint64_t cl_recycled;	/* # of clusters recycled */
565 	uint64_t cl_fast_recycled; /* # of clusters recycled (fast) */
566 
567 	/* These 3 are valid when FL_BUF_RESUME is set, stale otherwise. */
568 	struct mbuf *m0;
569 	struct mbuf **pnext;
570 	u_int remaining;
571 
572 	uint16_t qsize;		/* # of hw descriptors (status page included) */
573 	uint16_t cntxt_id;	/* SGE context id for the freelist */
574 	TAILQ_ENTRY(sge_fl) link; /* All starving freelists */
575 	bus_dma_tag_t desc_tag;
576 	bus_dmamap_t desc_map;
577 	char lockname[16];
578 	bus_addr_t ba;		/* bus address of descriptor ring */
579 };
580 
581 struct mp_ring;
582 
583 struct txpkts {
584 	uint8_t wr_type;	/* type 0 or type 1 */
585 	uint8_t npkt;		/* # of packets in this work request */
586 	uint8_t len16;		/* # of 16B pieces used by this work request */
587 	uint8_t score;
588 	uint8_t max_npkt;	/* maximum number of packets allowed */
589 	uint16_t plen;		/* total payload (sum of all packets) */
590 
591 	/* straight from fw_eth_tx_pkts_vm_wr. */
592 	__u8   ethmacdst[6];
593 	__u8   ethmacsrc[6];
594 	__be16 ethtype;
595 	__be16 vlantci;
596 
597 	struct mbuf *mb[15];
598 };
599 
600 /* txq: SGE egress queue + what's needed for Ethernet NIC */
601 struct sge_txq {
602 	struct sge_eq eq;	/* MUST be first */
603 
604 	if_t ifp;		/* the interface this txq belongs to */
605 	struct mp_ring *r;	/* tx software ring */
606 	struct tx_sdesc *sdesc;	/* KVA of software descriptor ring */
607 	struct sglist *gl;
608 	__be32 cpl_ctrl0;	/* for convenience */
609 	int tc_idx;		/* traffic class */
610 	uint64_t last_tx;	/* cycle count when eth_tx was last called */
611 	struct txpkts txp;
612 
613 	struct task tx_reclaim_task;
614 	/* stats for common events first */
615 
616 	uint64_t txcsum;	/* # of times hardware assisted with checksum */
617 	uint64_t tso_wrs;	/* # of TSO work requests */
618 	uint64_t vlan_insertion;/* # of times VLAN tag was inserted */
619 	uint64_t imm_wrs;	/* # of work requests with immediate data */
620 	uint64_t sgl_wrs;	/* # of work requests with direct SGL */
621 	uint64_t txpkt_wrs;	/* # of txpkt work requests (not coalesced) */
622 	uint64_t txpkts0_wrs;	/* # of type0 coalesced tx work requests */
623 	uint64_t txpkts1_wrs;	/* # of type1 coalesced tx work requests */
624 	uint64_t txpkts0_pkts;	/* # of frames in type0 coalesced tx WRs */
625 	uint64_t txpkts1_pkts;	/* # of frames in type1 coalesced tx WRs */
626 	uint64_t txpkts_flush;	/* # of times txp had to be sent by tx_update */
627 	uint64_t raw_wrs;	/* # of raw work requests (alloc_wr_mbuf) */
628 	uint64_t vxlan_tso_wrs;	/* # of VXLAN TSO work requests */
629 	uint64_t vxlan_txcsum;
630 
631 	uint64_t kern_tls_records;
632 	uint64_t kern_tls_short;
633 	uint64_t kern_tls_partial;
634 	uint64_t kern_tls_full;
635 	uint64_t kern_tls_octets;
636 	uint64_t kern_tls_waste;
637 	uint64_t kern_tls_options;
638 	uint64_t kern_tls_header;
639 	uint64_t kern_tls_fin;
640 	uint64_t kern_tls_fin_short;
641 	uint64_t kern_tls_cbc;
642 	uint64_t kern_tls_gcm;
643 
644 	/* stats for not-that-common events */
645 
646 	/* Optional scratch space for constructing work requests. */
647 	uint8_t ss[SGE_MAX_WR_LEN] __aligned(16);
648 } __aligned(CACHE_LINE_SIZE);
649 
650 /* rxq: SGE ingress queue + SGE free list + miscellaneous items */
651 struct sge_rxq {
652 	struct sge_iq iq;	/* MUST be first */
653 	struct sge_fl fl;	/* MUST follow iq */
654 
655 	if_t ifp;		/* the interface this rxq belongs to */
656 	struct lro_ctrl lro;	/* LRO state */
657 
658 	/* stats for common events first */
659 
660 	uint64_t rxcsum;	/* # of times hardware assisted with checksum */
661 	uint64_t vlan_extraction;/* # of times VLAN tag was extracted */
662 	uint64_t vxlan_rxcsum;
663 
664 	/* stats for not-that-common events */
665 
666 } __aligned(CACHE_LINE_SIZE);
667 
668 static inline struct sge_rxq *
669 iq_to_rxq(struct sge_iq *iq)
670 {
671 
672 	return (__containerof(iq, struct sge_rxq, iq));
673 }
674 
675 /* ofld_rxq: SGE ingress queue + SGE free list + miscellaneous items */
676 struct sge_ofld_rxq {
677 	struct sge_iq iq;	/* MUST be first */
678 	struct sge_fl fl;	/* MUST follow iq */
679 	counter_u64_t rx_iscsi_ddp_setup_ok;
680 	counter_u64_t rx_iscsi_ddp_setup_error;
681 	uint64_t rx_iscsi_ddp_pdus;
682 	uint64_t rx_iscsi_ddp_octets;
683 	uint64_t rx_iscsi_fl_pdus;
684 	uint64_t rx_iscsi_fl_octets;
685 	uint64_t rx_iscsi_padding_errors;
686 	uint64_t rx_iscsi_header_digest_errors;
687 	uint64_t rx_iscsi_data_digest_errors;
688 	u_long	rx_toe_tls_records;
689 	u_long	rx_toe_tls_octets;
690 } __aligned(CACHE_LINE_SIZE);
691 
692 static inline struct sge_ofld_rxq *
693 iq_to_ofld_rxq(struct sge_iq *iq)
694 {
695 
696 	return (__containerof(iq, struct sge_ofld_rxq, iq));
697 }
698 
699 struct wrqe {
700 	STAILQ_ENTRY(wrqe) link;
701 	struct sge_wrq *wrq;
702 	int wr_len;
703 	char wr[] __aligned(16);
704 };
705 
706 struct wrq_cookie {
707 	TAILQ_ENTRY(wrq_cookie) link;
708 	int ndesc;
709 	int pidx;
710 };
711 
712 /*
713  * wrq: SGE egress queue that is given prebuilt work requests.  Control queues
714  * are of this type.
715  */
716 struct sge_wrq {
717 	struct sge_eq eq;	/* MUST be first */
718 
719 	struct adapter *adapter;
720 	struct task wrq_tx_task;
721 
722 	/* Tx desc reserved but WR not "committed" yet. */
723 	TAILQ_HEAD(wrq_incomplete_wrs , wrq_cookie) incomplete_wrs;
724 
725 	/* List of WRs ready to go out as soon as descriptors are available. */
726 	STAILQ_HEAD(, wrqe) wr_list;
727 	u_int nwr_pending;
728 	u_int ndesc_needed;
729 
730 	/* stats for common events first */
731 
732 	uint64_t tx_wrs_direct;	/* # of WRs written directly to desc ring. */
733 	uint64_t tx_wrs_ss;	/* # of WRs copied from scratch space. */
734 	uint64_t tx_wrs_copied;	/* # of WRs queued and copied to desc ring. */
735 
736 	/* stats for not-that-common events */
737 
738 	/*
739 	 * Scratch space for work requests that wrap around after reaching the
740 	 * status page, and some information about the last WR that used it.
741 	 */
742 	uint16_t ss_pidx;
743 	uint16_t ss_len;
744 	uint8_t ss[SGE_MAX_WR_LEN];
745 
746 } __aligned(CACHE_LINE_SIZE);
747 
748 /* ofld_txq: SGE egress queue + miscellaneous items */
749 struct sge_ofld_txq {
750 	struct sge_wrq wrq;
751 	counter_u64_t tx_iscsi_pdus;
752 	counter_u64_t tx_iscsi_octets;
753 	counter_u64_t tx_iscsi_iso_wrs;
754 	counter_u64_t tx_toe_tls_records;
755 	counter_u64_t tx_toe_tls_octets;
756 } __aligned(CACHE_LINE_SIZE);
757 
758 #define INVALID_NM_RXQ_CNTXT_ID ((uint16_t)(-1))
759 struct sge_nm_rxq {
760 	/* Items used by the driver rx ithread are in this cacheline. */
761 	volatile int nm_state __aligned(CACHE_LINE_SIZE);	/* NM_OFF, NM_ON, or NM_BUSY */
762 	u_int nid;		/* netmap ring # for this queue */
763 	struct vi_info *vi;
764 
765 	struct iq_desc *iq_desc;
766 	uint16_t iq_abs_id;
767 	uint16_t iq_cntxt_id;
768 	uint16_t iq_cidx;
769 	uint16_t iq_sidx;
770 	uint8_t iq_gen;
771 	uint32_t fl_sidx;
772 
773 	/* Items used by netmap rxsync are in this cacheline. */
774 	__be64  *fl_desc __aligned(CACHE_LINE_SIZE);
775 	uint16_t fl_cntxt_id;
776 	uint32_t fl_pidx;
777 	uint32_t fl_sidx2;	/* copy of fl_sidx */
778 	uint32_t fl_db_val;
779 	u_int fl_db_saved;
780 	u_int fl_db_threshold;	/* in descriptors */
781 	u_int fl_hwidx:4;
782 
783 	/*
784 	 * fl_cidx is used by both the ithread and rxsync, the rest are not used
785 	 * in the rx fast path.
786 	 */
787 	uint32_t fl_cidx __aligned(CACHE_LINE_SIZE);
788 
789 	bus_dma_tag_t iq_desc_tag;
790 	bus_dmamap_t iq_desc_map;
791 	bus_addr_t iq_ba;
792 	int intr_idx;
793 
794 	bus_dma_tag_t fl_desc_tag;
795 	bus_dmamap_t fl_desc_map;
796 	bus_addr_t fl_ba;
797 };
798 
799 #define INVALID_NM_TXQ_CNTXT_ID ((u_int)(-1))
800 struct sge_nm_txq {
801 	struct tx_desc *desc;
802 	uint16_t cidx;
803 	uint16_t pidx;
804 	uint16_t sidx;
805 	uint16_t equiqidx;	/* EQUIQ last requested at this pidx */
806 	uint16_t equeqidx;	/* EQUEQ last requested at this pidx */
807 	uint16_t dbidx;		/* pidx of the most recent doorbell */
808 	uint8_t doorbells;
809 	volatile uint32_t *udb;
810 	u_int udb_qid;
811 	u_int cntxt_id;
812 	__be32 cpl_ctrl0;	/* for convenience */
813 	__be32 op_pkd;		/* ditto */
814 	u_int nid;		/* netmap ring # for this queue */
815 
816 	/* infrequently used items after this */
817 
818 	bus_dma_tag_t desc_tag;
819 	bus_dmamap_t desc_map;
820 	bus_addr_t ba;
821 	int iqidx;
822 } __aligned(CACHE_LINE_SIZE);
823 
824 struct sge {
825 	int nrxq;	/* total # of Ethernet rx queues */
826 	int ntxq;	/* total # of Ethernet tx queues */
827 	int nofldrxq;	/* total # of TOE rx queues */
828 	int nofldtxq;	/* total # of TOE tx queues */
829 	int nnmrxq;	/* total # of netmap rx queues */
830 	int nnmtxq;	/* total # of netmap tx queues */
831 	int niq;	/* total # of ingress queues */
832 	int neq;	/* total # of egress queues */
833 
834 	struct sge_iq fwq;	/* Firmware event queue */
835 	struct sge_wrq *ctrlq;	/* Control queues */
836 	struct sge_txq *txq;	/* NIC tx queues */
837 	struct sge_rxq *rxq;	/* NIC rx queues */
838 	struct sge_ofld_txq *ofld_txq;	/* TOE tx queues */
839 	struct sge_ofld_rxq *ofld_rxq;	/* TOE rx queues */
840 	struct sge_nm_txq *nm_txq;	/* netmap tx queues */
841 	struct sge_nm_rxq *nm_rxq;	/* netmap rx queues */
842 
843 	uint16_t iq_start;	/* first cntxt_id */
844 	uint16_t iq_base;	/* first abs_id */
845 	int eq_start;		/* first cntxt_id */
846 	int eq_base;		/* first abs_id */
847 	int iqmap_sz;
848 	int eqmap_sz;
849 	struct sge_iq **iqmap;	/* iq->cntxt_id to iq mapping */
850 	struct sge_eq **eqmap;	/* eq->cntxt_id to eq mapping */
851 
852 	int8_t safe_zidx;
853 	struct rx_buf_info rx_buf_info[SW_ZONE_SIZES];
854 };
855 
856 struct devnames {
857 	const char *nexus_name;
858 	const char *ifnet_name;
859 	const char *vi_ifnet_name;
860 	const char *pf03_drv_name;
861 	const char *vf_nexus_name;
862 	const char *vf_ifnet_name;
863 };
864 
865 struct clip_entry;
866 
867 #define CNT_CAL_INFO 3
868 struct clock_sync {
869 	uint64_t hw_cur;
870 	uint64_t hw_prev;
871 	sbintime_t sbt_cur;
872 	sbintime_t sbt_prev;
873 	seqc_t gen;
874 };
875 
876 struct adapter {
877 	SLIST_ENTRY(adapter) link;
878 	device_t dev;
879 	struct cdev *cdev;
880 	const struct devnames *names;
881 
882 	/* PCIe register resources */
883 	int regs_rid;
884 	struct resource *regs_res;
885 	int msix_rid;
886 	struct resource *msix_res;
887 	bus_space_handle_t bh;
888 	bus_space_tag_t bt;
889 	bus_size_t mmio_len;
890 	int udbs_rid;
891 	struct resource *udbs_res;
892 	volatile uint8_t *udbs_base;
893 
894 	unsigned int pf;
895 	unsigned int mbox;
896 	unsigned int vpd_busy;
897 	unsigned int vpd_flag;
898 
899 	/* Interrupt information */
900 	int intr_type;
901 	int intr_count;
902 	struct irq {
903 		struct resource *res;
904 		int rid;
905 		void *tag;
906 		struct sge_rxq *rxq;
907 		struct sge_nm_rxq *nm_rxq;
908 	} __aligned(CACHE_LINE_SIZE) *irq;
909 	int sge_gts_reg;
910 	int sge_kdoorbell_reg;
911 
912 	bus_dma_tag_t dmat;	/* Parent DMA tag */
913 
914 	struct sge sge;
915 	int lro_timeout;
916 	int sc_do_rxcopy;
917 
918 	int vxlan_port;
919 	u_int vxlan_refcount;
920 	int rawf_base;
921 	int nrawf;
922 
923 	struct taskqueue *tq[MAX_NCHAN];	/* General purpose taskqueues */
924 	struct port_info *port[MAX_NPORTS];
925 	uint8_t chan_map[MAX_NCHAN];		/* channel -> port */
926 
927 	CXGBE_LIST_HEAD(, clip_entry) *clip_table;
928 	TAILQ_HEAD(, clip_entry) clip_pending;	/* these need hw update. */
929 	u_long clip_mask;
930 	int clip_gen;
931 	struct timeout_task clip_task;
932 
933 	void *tom_softc;	/* (struct tom_data *) */
934 	struct tom_tunables tt;
935 	struct t4_offload_policy *policy;
936 	struct rwlock policy_lock;
937 
938 	void *iwarp_softc;	/* (struct c4iw_dev *) */
939 	struct iw_tunables iwt;
940 	void *iscsi_ulp_softc;	/* (struct cxgbei_data *) */
941 	struct l2t_data *l2t;	/* L2 table */
942 	struct smt_data *smt;	/* Source MAC Table */
943 	struct tid_info tids;
944 	vmem_t *key_map;
945 	struct tls_tunables tlst;
946 
947 	uint8_t doorbells;
948 	int offload_map;	/* port_id's with IFCAP_TOE enabled */
949 	int bt_map;		/* tx_chan's with BASE-T */
950 	int active_ulds;	/* ULDs activated on this adapter */
951 	int flags;
952 	int debug_flags;
953 	int error_flags;	/* Used by error handler and live reset. */
954 
955 	char ifp_lockname[16];
956 	struct mtx ifp_lock;
957 	if_t ifp;		/* tracer ifp */
958 	struct ifmedia media;
959 	int traceq;		/* iq used by all tracers, -1 if none */
960 	int tracer_valid;	/* bitmap of valid tracers */
961 	int tracer_enabled;	/* bitmap of enabled tracers */
962 
963 	char fw_version[16];
964 	char tp_version[16];
965 	char er_version[16];
966 	char bs_version[16];
967 	char cfg_file[32];
968 	u_int cfcsum;
969 	struct adapter_params params;
970 	const struct chip_params *chip_params;
971 	struct t4_virt_res vres;
972 
973 	uint16_t nbmcaps;
974 	uint16_t linkcaps;
975 	uint16_t switchcaps;
976 	uint16_t niccaps;
977 	uint16_t toecaps;
978 	uint16_t rdmacaps;
979 	uint16_t cryptocaps;
980 	uint16_t iscsicaps;
981 	uint16_t fcoecaps;
982 
983 	struct sysctl_ctx_list ctx;
984 	struct sysctl_oid *ctrlq_oid;
985 	struct sysctl_oid *fwq_oid;
986 
987 	struct mtx sc_lock;
988 	char lockname[16];
989 
990 	/* Starving free lists */
991 	struct mtx sfl_lock;	/* same cache-line as sc_lock? but that's ok */
992 	TAILQ_HEAD(, sge_fl) sfl;
993 	struct callout sfl_callout;
994 	struct callout cal_callout;
995 	struct clock_sync cal_info[CNT_CAL_INFO];
996 	int cal_current;
997 	int cal_count;
998 	uint32_t cal_gen;
999 
1000 	/*
1001 	 * Driver code that can run when the adapter is suspended must use this
1002 	 * lock or a synchronized_op and check for HW_OFF_LIMITS before
1003 	 * accessing hardware.
1004 	 *
1005 	 * XXX: could be changed to rwlock.  wlock in suspend/resume and for
1006 	 * indirect register access, rlock everywhere else.
1007 	 */
1008 	struct mtx reg_lock;
1009 
1010 	struct memwin memwin[NUM_MEMWIN];	/* memory windows */
1011 
1012 	struct mtx tc_lock;
1013 	struct task tc_task;
1014 
1015 	struct task fatal_error_task;
1016 	struct task reset_task;
1017 	const void *reset_thread;
1018 	int num_resets;
1019 	int incarnation;
1020 
1021 	const char *last_op;
1022 	const void *last_op_thr;
1023 	int last_op_flags;
1024 
1025 	int swintr;
1026 	int sensor_resets;
1027 
1028 	struct callout ktls_tick;
1029 };
1030 
1031 #define ADAPTER_LOCK(sc)		mtx_lock(&(sc)->sc_lock)
1032 #define ADAPTER_UNLOCK(sc)		mtx_unlock(&(sc)->sc_lock)
1033 #define ADAPTER_LOCK_ASSERT_OWNED(sc)	mtx_assert(&(sc)->sc_lock, MA_OWNED)
1034 #define ADAPTER_LOCK_ASSERT_NOTOWNED(sc) mtx_assert(&(sc)->sc_lock, MA_NOTOWNED)
1035 
1036 #define ASSERT_SYNCHRONIZED_OP(sc)	\
1037     KASSERT(IS_BUSY(sc) && \
1038 	(mtx_owned(&(sc)->sc_lock) || sc->last_op_thr == curthread), \
1039 	("%s: operation not synchronized.", __func__))
1040 
1041 #define PORT_LOCK(pi)			mtx_lock(&(pi)->pi_lock)
1042 #define PORT_UNLOCK(pi)			mtx_unlock(&(pi)->pi_lock)
1043 #define PORT_LOCK_ASSERT_OWNED(pi)	mtx_assert(&(pi)->pi_lock, MA_OWNED)
1044 #define PORT_LOCK_ASSERT_NOTOWNED(pi)	mtx_assert(&(pi)->pi_lock, MA_NOTOWNED)
1045 
1046 #define FL_LOCK(fl)			mtx_lock(&(fl)->fl_lock)
1047 #define FL_TRYLOCK(fl)			mtx_trylock(&(fl)->fl_lock)
1048 #define FL_UNLOCK(fl)			mtx_unlock(&(fl)->fl_lock)
1049 #define FL_LOCK_ASSERT_OWNED(fl)	mtx_assert(&(fl)->fl_lock, MA_OWNED)
1050 #define FL_LOCK_ASSERT_NOTOWNED(fl)	mtx_assert(&(fl)->fl_lock, MA_NOTOWNED)
1051 
1052 #define RXQ_FL_LOCK(rxq)		FL_LOCK(&(rxq)->fl)
1053 #define RXQ_FL_UNLOCK(rxq)		FL_UNLOCK(&(rxq)->fl)
1054 #define RXQ_FL_LOCK_ASSERT_OWNED(rxq)	FL_LOCK_ASSERT_OWNED(&(rxq)->fl)
1055 #define RXQ_FL_LOCK_ASSERT_NOTOWNED(rxq) FL_LOCK_ASSERT_NOTOWNED(&(rxq)->fl)
1056 
1057 #define EQ_LOCK(eq)			mtx_lock(&(eq)->eq_lock)
1058 #define EQ_TRYLOCK(eq)			mtx_trylock(&(eq)->eq_lock)
1059 #define EQ_UNLOCK(eq)			mtx_unlock(&(eq)->eq_lock)
1060 #define EQ_LOCK_ASSERT_OWNED(eq)	mtx_assert(&(eq)->eq_lock, MA_OWNED)
1061 #define EQ_LOCK_ASSERT_NOTOWNED(eq)	mtx_assert(&(eq)->eq_lock, MA_NOTOWNED)
1062 
1063 #define TXQ_LOCK(txq)			EQ_LOCK(&(txq)->eq)
1064 #define TXQ_TRYLOCK(txq)		EQ_TRYLOCK(&(txq)->eq)
1065 #define TXQ_UNLOCK(txq)			EQ_UNLOCK(&(txq)->eq)
1066 #define TXQ_LOCK_ASSERT_OWNED(txq)	EQ_LOCK_ASSERT_OWNED(&(txq)->eq)
1067 #define TXQ_LOCK_ASSERT_NOTOWNED(txq)	EQ_LOCK_ASSERT_NOTOWNED(&(txq)->eq)
1068 
1069 #define for_each_txq(vi, iter, q) \
1070 	for (q = &vi->adapter->sge.txq[vi->first_txq], iter = 0; \
1071 	    iter < vi->ntxq; ++iter, ++q)
1072 #define for_each_rxq(vi, iter, q) \
1073 	for (q = &vi->adapter->sge.rxq[vi->first_rxq], iter = 0; \
1074 	    iter < vi->nrxq; ++iter, ++q)
1075 #define for_each_ofld_txq(vi, iter, q) \
1076 	for (q = &vi->adapter->sge.ofld_txq[vi->first_ofld_txq], iter = 0; \
1077 	    iter < vi->nofldtxq; ++iter, ++q)
1078 #define for_each_ofld_rxq(vi, iter, q) \
1079 	for (q = &vi->adapter->sge.ofld_rxq[vi->first_ofld_rxq], iter = 0; \
1080 	    iter < vi->nofldrxq; ++iter, ++q)
1081 #define for_each_nm_txq(vi, iter, q) \
1082 	for (q = &vi->adapter->sge.nm_txq[vi->first_nm_txq], iter = 0; \
1083 	    iter < vi->nnmtxq; ++iter, ++q)
1084 #define for_each_nm_rxq(vi, iter, q) \
1085 	for (q = &vi->adapter->sge.nm_rxq[vi->first_nm_rxq], iter = 0; \
1086 	    iter < vi->nnmrxq; ++iter, ++q)
1087 #define for_each_vi(_pi, _iter, _vi) \
1088 	for ((_vi) = (_pi)->vi, (_iter) = 0; (_iter) < (_pi)->nvi; \
1089 	     ++(_iter), ++(_vi))
1090 
1091 #define IDXINCR(idx, incr, wrap) do { \
1092 	idx = wrap - idx > incr ? idx + incr : incr - (wrap - idx); \
1093 } while (0)
1094 #define IDXDIFF(head, tail, wrap) \
1095 	((head) >= (tail) ? (head) - (tail) : (wrap) - (tail) + (head))
1096 
1097 /* One for errors, one for firmware events */
1098 #define T4_EXTRA_INTR 2
1099 
1100 /* One for firmware events */
1101 #define T4VF_EXTRA_INTR 1
1102 
1103 static inline int
1104 forwarding_intr_to_fwq(struct adapter *sc)
1105 {
1106 
1107 	return (sc->intr_count == 1);
1108 }
1109 
1110 /* Works reliably inside a sync_op or with reg_lock held. */
1111 static inline bool
1112 hw_off_limits(struct adapter *sc)
1113 {
1114 	int off_limits = atomic_load_int(&sc->error_flags) & HW_OFF_LIMITS;
1115 
1116 	return (__predict_false(off_limits != 0));
1117 }
1118 
1119 static inline int
1120 mbuf_nsegs(struct mbuf *m)
1121 {
1122 	M_ASSERTPKTHDR(m);
1123 	KASSERT(m->m_pkthdr.inner_l5hlen > 0,
1124 	    ("%s: mbuf %p missing information on # of segments.", __func__, m));
1125 
1126 	return (m->m_pkthdr.inner_l5hlen);
1127 }
1128 
1129 static inline void
1130 set_mbuf_nsegs(struct mbuf *m, uint8_t nsegs)
1131 {
1132 	M_ASSERTPKTHDR(m);
1133 	m->m_pkthdr.inner_l5hlen = nsegs;
1134 }
1135 
1136 /* Internal mbuf flags stored in PH_loc.eight[1]. */
1137 #define	MC_NOMAP		0x01
1138 #define	MC_RAW_WR		0x02
1139 #define	MC_TLS			0x04
1140 
1141 static inline int
1142 mbuf_cflags(struct mbuf *m)
1143 {
1144 	M_ASSERTPKTHDR(m);
1145 	return (m->m_pkthdr.PH_loc.eight[4]);
1146 }
1147 
1148 static inline void
1149 set_mbuf_cflags(struct mbuf *m, uint8_t flags)
1150 {
1151 	M_ASSERTPKTHDR(m);
1152 	m->m_pkthdr.PH_loc.eight[4] = flags;
1153 }
1154 
1155 static inline int
1156 mbuf_len16(struct mbuf *m)
1157 {
1158 	int n;
1159 
1160 	M_ASSERTPKTHDR(m);
1161 	n = m->m_pkthdr.PH_loc.eight[0];
1162 	if (!(mbuf_cflags(m) & MC_TLS))
1163 		MPASS(n > 0 && n <= SGE_MAX_WR_LEN / 16);
1164 
1165 	return (n);
1166 }
1167 
1168 static inline void
1169 set_mbuf_len16(struct mbuf *m, uint8_t len16)
1170 {
1171 	M_ASSERTPKTHDR(m);
1172 	if (!(mbuf_cflags(m) & MC_TLS))
1173 		MPASS(len16 > 0 && len16 <= SGE_MAX_WR_LEN / 16);
1174 	m->m_pkthdr.PH_loc.eight[0] = len16;
1175 }
1176 
1177 static inline uint32_t
1178 t4_read_reg(struct adapter *sc, uint32_t reg)
1179 {
1180 	if (hw_off_limits(sc))
1181 		MPASS(curthread == sc->reset_thread);
1182 	return bus_space_read_4(sc->bt, sc->bh, reg);
1183 }
1184 
1185 static inline void
1186 t4_write_reg(struct adapter *sc, uint32_t reg, uint32_t val)
1187 {
1188 	if (hw_off_limits(sc))
1189 		MPASS(curthread == sc->reset_thread);
1190 	bus_space_write_4(sc->bt, sc->bh, reg, val);
1191 }
1192 
1193 static inline uint64_t
1194 t4_read_reg64(struct adapter *sc, uint32_t reg)
1195 {
1196 	if (hw_off_limits(sc))
1197 		MPASS(curthread == sc->reset_thread);
1198 #ifdef __LP64__
1199 	return bus_space_read_8(sc->bt, sc->bh, reg);
1200 #else
1201 	return (uint64_t)bus_space_read_4(sc->bt, sc->bh, reg) +
1202 	    ((uint64_t)bus_space_read_4(sc->bt, sc->bh, reg + 4) << 32);
1203 
1204 #endif
1205 }
1206 
1207 static inline void
1208 t4_write_reg64(struct adapter *sc, uint32_t reg, uint64_t val)
1209 {
1210 	if (hw_off_limits(sc))
1211 		MPASS(curthread == sc->reset_thread);
1212 #ifdef __LP64__
1213 	bus_space_write_8(sc->bt, sc->bh, reg, val);
1214 #else
1215 	bus_space_write_4(sc->bt, sc->bh, reg, val);
1216 	bus_space_write_4(sc->bt, sc->bh, reg + 4, val>> 32);
1217 #endif
1218 }
1219 
1220 static inline void
1221 t4_os_pci_read_cfg1(struct adapter *sc, int reg, uint8_t *val)
1222 {
1223 	if (hw_off_limits(sc))
1224 		MPASS(curthread == sc->reset_thread);
1225 	*val = pci_read_config(sc->dev, reg, 1);
1226 }
1227 
1228 static inline void
1229 t4_os_pci_write_cfg1(struct adapter *sc, int reg, uint8_t val)
1230 {
1231 	if (hw_off_limits(sc))
1232 		MPASS(curthread == sc->reset_thread);
1233 	pci_write_config(sc->dev, reg, val, 1);
1234 }
1235 
1236 static inline void
1237 t4_os_pci_read_cfg2(struct adapter *sc, int reg, uint16_t *val)
1238 {
1239 
1240 	if (hw_off_limits(sc))
1241 		MPASS(curthread == sc->reset_thread);
1242 	*val = pci_read_config(sc->dev, reg, 2);
1243 }
1244 
1245 static inline void
1246 t4_os_pci_write_cfg2(struct adapter *sc, int reg, uint16_t val)
1247 {
1248 	if (hw_off_limits(sc))
1249 		MPASS(curthread == sc->reset_thread);
1250 	pci_write_config(sc->dev, reg, val, 2);
1251 }
1252 
1253 static inline void
1254 t4_os_pci_read_cfg4(struct adapter *sc, int reg, uint32_t *val)
1255 {
1256 	if (hw_off_limits(sc))
1257 		MPASS(curthread == sc->reset_thread);
1258 	*val = pci_read_config(sc->dev, reg, 4);
1259 }
1260 
1261 static inline void
1262 t4_os_pci_write_cfg4(struct adapter *sc, int reg, uint32_t val)
1263 {
1264 	if (hw_off_limits(sc))
1265 		MPASS(curthread == sc->reset_thread);
1266 	pci_write_config(sc->dev, reg, val, 4);
1267 }
1268 
1269 static inline struct port_info *
1270 adap2pinfo(struct adapter *sc, int idx)
1271 {
1272 
1273 	return (sc->port[idx]);
1274 }
1275 
1276 static inline void
1277 t4_os_set_hw_addr(struct port_info *pi, uint8_t hw_addr[])
1278 {
1279 
1280 	bcopy(hw_addr, pi->vi[0].hw_addr, ETHER_ADDR_LEN);
1281 }
1282 
1283 static inline int
1284 tx_resume_threshold(struct sge_eq *eq)
1285 {
1286 
1287 	/* not quite the same as qsize / 4, but this will do. */
1288 	return (eq->sidx / 4);
1289 }
1290 
1291 static inline int
1292 t4_use_ldst(struct adapter *sc)
1293 {
1294 
1295 #ifdef notyet
1296 	return (sc->flags & FW_OK || !sc->use_bd);
1297 #else
1298 	return (0);
1299 #endif
1300 }
1301 
1302 static inline void
1303 CH_DUMP_MBOX(struct adapter *sc, int mbox, const int reg,
1304     const char *msg, const __be64 *const p, const bool err)
1305 {
1306 
1307 	if (!(sc->debug_flags & DF_DUMP_MBOX) && !err)
1308 		return;
1309 	if (p != NULL) {
1310 		log(err ? LOG_ERR : LOG_DEBUG,
1311 		    "%s: mbox %u %s %016llx %016llx %016llx %016llx "
1312 		    "%016llx %016llx %016llx %016llx\n",
1313 		    device_get_nameunit(sc->dev), mbox, msg,
1314 		    (long long)be64_to_cpu(p[0]), (long long)be64_to_cpu(p[1]),
1315 		    (long long)be64_to_cpu(p[2]), (long long)be64_to_cpu(p[3]),
1316 		    (long long)be64_to_cpu(p[4]), (long long)be64_to_cpu(p[5]),
1317 		    (long long)be64_to_cpu(p[6]), (long long)be64_to_cpu(p[7]));
1318 	} else {
1319 		log(err ? LOG_ERR : LOG_DEBUG,
1320 		    "%s: mbox %u %s %016llx %016llx %016llx %016llx "
1321 		    "%016llx %016llx %016llx %016llx\n",
1322 		    device_get_nameunit(sc->dev), mbox, msg,
1323 		    (long long)t4_read_reg64(sc, reg),
1324 		    (long long)t4_read_reg64(sc, reg + 8),
1325 		    (long long)t4_read_reg64(sc, reg + 16),
1326 		    (long long)t4_read_reg64(sc, reg + 24),
1327 		    (long long)t4_read_reg64(sc, reg + 32),
1328 		    (long long)t4_read_reg64(sc, reg + 40),
1329 		    (long long)t4_read_reg64(sc, reg + 48),
1330 		    (long long)t4_read_reg64(sc, reg + 56));
1331 	}
1332 }
1333 
1334 /* t4_main.c */
1335 extern int t4_ntxq;
1336 extern int t4_nrxq;
1337 extern int t4_intr_types;
1338 extern int t4_tmr_idx;
1339 extern int t4_pktc_idx;
1340 extern unsigned int t4_qsize_rxq;
1341 extern unsigned int t4_qsize_txq;
1342 extern device_method_t cxgbe_methods[];
1343 
1344 int t4_os_find_pci_capability(struct adapter *, int);
1345 int t4_os_pci_save_state(struct adapter *);
1346 int t4_os_pci_restore_state(struct adapter *);
1347 void t4_os_portmod_changed(struct port_info *);
1348 void t4_os_link_changed(struct port_info *);
1349 void t4_iterate(void (*)(struct adapter *, void *), void *);
1350 void t4_init_devnames(struct adapter *);
1351 void t4_add_adapter(struct adapter *);
1352 int t4_detach_common(device_t);
1353 int t4_map_bars_0_and_4(struct adapter *);
1354 int t4_map_bar_2(struct adapter *);
1355 int t4_setup_intr_handlers(struct adapter *);
1356 void t4_sysctls(struct adapter *);
1357 int begin_synchronized_op(struct adapter *, struct vi_info *, int, char *);
1358 void doom_vi(struct adapter *, struct vi_info *);
1359 void end_synchronized_op(struct adapter *, int);
1360 int update_mac_settings(if_t, int);
1361 int adapter_init(struct adapter *);
1362 int vi_init(struct vi_info *);
1363 void vi_sysctls(struct vi_info *);
1364 int rw_via_memwin(struct adapter *, int, uint32_t, uint32_t *, int, int);
1365 int alloc_atid(struct adapter *, void *);
1366 void *lookup_atid(struct adapter *, int);
1367 void free_atid(struct adapter *, int);
1368 void release_tid(struct adapter *, int, struct sge_wrq *);
1369 int cxgbe_media_change(if_t);
1370 void cxgbe_media_status(if_t, struct ifmediareq *);
1371 void t4_os_cim_err(struct adapter *);
1372 
1373 #ifdef KERN_TLS
1374 /* t6_kern_tls.c */
1375 int t6_tls_tag_alloc(if_t, union if_snd_tag_alloc_params *,
1376     struct m_snd_tag **);
1377 void t6_ktls_modload(void);
1378 void t6_ktls_modunload(void);
1379 int t6_ktls_try(if_t, struct socket *, struct ktls_session *);
1380 int t6_ktls_parse_pkt(struct mbuf *);
1381 int t6_ktls_write_wr(struct sge_txq *, void *, struct mbuf *, u_int);
1382 #endif
1383 
1384 /* t4_keyctx.c */
1385 struct auth_hash;
1386 union authctx;
1387 #ifdef KERN_TLS
1388 struct ktls_session;
1389 struct tls_key_req;
1390 struct tls_keyctx;
1391 #endif
1392 
1393 void t4_aes_getdeckey(void *, const void *, unsigned int);
1394 void t4_copy_partial_hash(int, union authctx *, void *);
1395 void t4_init_gmac_hash(const char *, int, char *);
1396 void t4_init_hmac_digest(const struct auth_hash *, u_int, const char *, int,
1397     char *);
1398 #ifdef KERN_TLS
1399 u_int t4_tls_key_info_size(const struct ktls_session *);
1400 int t4_tls_proto_ver(const struct ktls_session *);
1401 int t4_tls_cipher_mode(const struct ktls_session *);
1402 int t4_tls_auth_mode(const struct ktls_session *);
1403 int t4_tls_hmac_ctrl(const struct ktls_session *);
1404 void t4_tls_key_ctx(const struct ktls_session *, int, struct tls_keyctx *);
1405 int t4_alloc_tls_keyid(struct adapter *);
1406 void t4_free_tls_keyid(struct adapter *, int);
1407 void t4_write_tlskey_wr(const struct ktls_session *, int, int, int, int,
1408     struct tls_key_req *);
1409 #endif
1410 
1411 #ifdef DEV_NETMAP
1412 /* t4_netmap.c */
1413 struct sge_nm_rxq;
1414 void cxgbe_nm_attach(struct vi_info *);
1415 void cxgbe_nm_detach(struct vi_info *);
1416 void service_nm_rxq(struct sge_nm_rxq *);
1417 int alloc_nm_rxq(struct vi_info *, struct sge_nm_rxq *, int, int);
1418 int free_nm_rxq(struct vi_info *, struct sge_nm_rxq *);
1419 int alloc_nm_txq(struct vi_info *, struct sge_nm_txq *, int, int);
1420 int free_nm_txq(struct vi_info *, struct sge_nm_txq *);
1421 #endif
1422 
1423 /* t4_sge.c */
1424 void t4_sge_modload(void);
1425 void t4_sge_modunload(void);
1426 uint64_t t4_sge_extfree_refs(void);
1427 void t4_tweak_chip_settings(struct adapter *);
1428 int t4_verify_chip_settings(struct adapter *);
1429 void t4_init_rx_buf_info(struct adapter *);
1430 int t4_create_dma_tag(struct adapter *);
1431 void t4_sge_sysctls(struct adapter *, struct sysctl_ctx_list *,
1432     struct sysctl_oid_list *);
1433 int t4_destroy_dma_tag(struct adapter *);
1434 int alloc_ring(struct adapter *, size_t, bus_dma_tag_t *, bus_dmamap_t *,
1435     bus_addr_t *, void **);
1436 int free_ring(struct adapter *, bus_dma_tag_t, bus_dmamap_t, bus_addr_t,
1437     void *);
1438 void free_fl_buffers(struct adapter *, struct sge_fl *);
1439 int t4_setup_adapter_queues(struct adapter *);
1440 int t4_teardown_adapter_queues(struct adapter *);
1441 int t4_setup_vi_queues(struct vi_info *);
1442 int t4_teardown_vi_queues(struct vi_info *);
1443 void t4_intr_all(void *);
1444 void t4_intr(void *);
1445 #ifdef DEV_NETMAP
1446 void t4_nm_intr(void *);
1447 void t4_vi_intr(void *);
1448 #endif
1449 void t4_intr_err(void *);
1450 void t4_intr_evt(void *);
1451 void t4_wrq_tx_locked(struct adapter *, struct sge_wrq *, struct wrqe *);
1452 void t4_update_fl_bufsize(if_t);
1453 struct mbuf *alloc_wr_mbuf(int, int);
1454 int parse_pkt(struct mbuf **, bool);
1455 void *start_wrq_wr(struct sge_wrq *, int, struct wrq_cookie *);
1456 void commit_wrq_wr(struct sge_wrq *, void *, struct wrq_cookie *);
1457 int t4_sge_set_conm_context(struct adapter *, int, int, int);
1458 void t4_register_an_handler(an_handler_t);
1459 void t4_register_fw_msg_handler(int, fw_msg_handler_t);
1460 void t4_register_cpl_handler(int, cpl_handler_t);
1461 void t4_register_shared_cpl_handler(int, cpl_handler_t, int);
1462 #ifdef RATELIMIT
1463 void send_etid_flush_wr(struct cxgbe_rate_tag *);
1464 #endif
1465 
1466 /* t4_tracer.c */
1467 struct t4_tracer;
1468 void t4_tracer_modload(void);
1469 void t4_tracer_modunload(void);
1470 void t4_tracer_port_detach(struct adapter *);
1471 int t4_get_tracer(struct adapter *, struct t4_tracer *);
1472 int t4_set_tracer(struct adapter *, struct t4_tracer *);
1473 int t4_trace_pkt(struct sge_iq *, const struct rss_header *, struct mbuf *);
1474 int t5_trace_pkt(struct sge_iq *, const struct rss_header *, struct mbuf *);
1475 
1476 /* t4_sched.c */
1477 int t4_set_sched_class(struct adapter *, struct t4_sched_params *);
1478 int t4_set_sched_queue(struct adapter *, struct t4_sched_queue *);
1479 int t4_init_tx_sched(struct adapter *);
1480 int t4_free_tx_sched(struct adapter *);
1481 void t4_update_tx_sched(struct adapter *);
1482 int t4_reserve_cl_rl_kbps(struct adapter *, int, u_int, int *);
1483 void t4_release_cl_rl(struct adapter *, int, int);
1484 int sysctl_tc(SYSCTL_HANDLER_ARGS);
1485 int sysctl_tc_params(SYSCTL_HANDLER_ARGS);
1486 #ifdef RATELIMIT
1487 void t4_init_etid_table(struct adapter *);
1488 void t4_free_etid_table(struct adapter *);
1489 struct cxgbe_rate_tag *lookup_etid(struct adapter *, int);
1490 int cxgbe_rate_tag_alloc(if_t, union if_snd_tag_alloc_params *,
1491     struct m_snd_tag **);
1492 void cxgbe_rate_tag_free_locked(struct cxgbe_rate_tag *);
1493 void cxgbe_ratelimit_query(if_t, struct if_ratelimit_query_results *);
1494 #endif
1495 
1496 /* t4_filter.c */
1497 int get_filter_mode(struct adapter *, uint32_t *);
1498 int set_filter_mode(struct adapter *, uint32_t);
1499 int set_filter_mask(struct adapter *, uint32_t);
1500 int get_filter(struct adapter *, struct t4_filter *);
1501 int set_filter(struct adapter *, struct t4_filter *);
1502 int del_filter(struct adapter *, struct t4_filter *);
1503 int t4_filter_rpl(struct sge_iq *, const struct rss_header *, struct mbuf *);
1504 int t4_hashfilter_ao_rpl(struct sge_iq *, const struct rss_header *, struct mbuf *);
1505 int t4_hashfilter_tcb_rpl(struct sge_iq *, const struct rss_header *, struct mbuf *);
1506 int t4_del_hashfilter_rpl(struct sge_iq *, const struct rss_header *, struct mbuf *);
1507 void free_hftid_hash(struct tid_info *);
1508 
1509 static inline struct wrqe *
1510 alloc_wrqe(int wr_len, struct sge_wrq *wrq)
1511 {
1512 	int len = offsetof(struct wrqe, wr) + wr_len;
1513 	struct wrqe *wr;
1514 
1515 	wr = malloc(len, M_CXGBE, M_NOWAIT);
1516 	if (__predict_false(wr == NULL))
1517 		return (NULL);
1518 	wr->wr_len = wr_len;
1519 	wr->wrq = wrq;
1520 	return (wr);
1521 }
1522 
1523 static inline void *
1524 wrtod(struct wrqe *wr)
1525 {
1526 	return (&wr->wr[0]);
1527 }
1528 
1529 static inline void
1530 free_wrqe(struct wrqe *wr)
1531 {
1532 	free(wr, M_CXGBE);
1533 }
1534 
1535 static inline void
1536 t4_wrq_tx(struct adapter *sc, struct wrqe *wr)
1537 {
1538 	struct sge_wrq *wrq = wr->wrq;
1539 
1540 	TXQ_LOCK(wrq);
1541 	t4_wrq_tx_locked(sc, wrq, wr);
1542 	TXQ_UNLOCK(wrq);
1543 }
1544 
1545 static inline int
1546 read_via_memwin(struct adapter *sc, int idx, uint32_t addr, uint32_t *val,
1547     int len)
1548 {
1549 
1550 	return (rw_via_memwin(sc, idx, addr, val, len, 0));
1551 }
1552 
1553 static inline int
1554 write_via_memwin(struct adapter *sc, int idx, uint32_t addr,
1555     const uint32_t *val, int len)
1556 {
1557 
1558 	return (rw_via_memwin(sc, idx, addr, (void *)(uintptr_t)val, len, 1));
1559 }
1560 
1561 /* Number of len16 -> number of descriptors */
1562 static inline int
1563 tx_len16_to_desc(int len16)
1564 {
1565 
1566 	return (howmany(len16, EQ_ESIZE / 16));
1567 }
1568 #endif
1569