xref: /openbsd/sys/dev/pci/if_mcx.c (revision 905646f0)
1 /*	$OpenBSD: if_mcx.c,v 1.72 2020/08/21 04:47:52 jmatthew Exp $ */
2 
3 /*
4  * Copyright (c) 2017 David Gwynne <dlg@openbsd.org>
5  * Copyright (c) 2019 Jonathan Matthew <jmatthew@openbsd.org>
6  *
7  * Permission to use, copy, modify, and distribute this software for any
8  * purpose with or without fee is hereby granted, provided that the above
9  * copyright notice and this permission notice appear in all copies.
10  *
11  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
12  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
13  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
14  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
15  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
16  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
17  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
18  */
19 
20 #include "bpfilter.h"
21 #include "vlan.h"
22 #include "kstat.h"
23 
24 #include <sys/param.h>
25 #include <sys/systm.h>
26 #include <sys/sockio.h>
27 #include <sys/mbuf.h>
28 #include <sys/kernel.h>
29 #include <sys/socket.h>
30 #include <sys/device.h>
31 #include <sys/pool.h>
32 #include <sys/queue.h>
33 #include <sys/timeout.h>
34 #include <sys/task.h>
35 #include <sys/atomic.h>
36 
37 #include <machine/bus.h>
38 #include <machine/intr.h>
39 
40 #include <net/if.h>
41 #include <net/if_dl.h>
42 #include <net/if_media.h>
43 #include <net/toeplitz.h>
44 
45 #if NBPFILTER > 0
46 #include <net/bpf.h>
47 #endif
48 
49 #if NKSTAT > 0
50 #include <sys/kstat.h>
51 #endif
52 
53 #include <netinet/in.h>
54 #include <netinet/if_ether.h>
55 
56 #include <dev/pci/pcireg.h>
57 #include <dev/pci/pcivar.h>
58 #include <dev/pci/pcidevs.h>
59 
60 #define BUS_DMASYNC_PRERW	(BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE)
61 #define BUS_DMASYNC_POSTRW	(BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE)
62 
63 #define MCX_HCA_BAR	PCI_MAPREG_START /* BAR 0 */
64 
65 #define MCX_FW_VER			0x0000
66 #define  MCX_FW_VER_MAJOR(_v)			((_v) & 0xffff)
67 #define  MCX_FW_VER_MINOR(_v)			((_v) >> 16)
68 #define MCX_CMDIF_FW_SUBVER		0x0004
69 #define  MCX_FW_VER_SUBMINOR(_v)		((_v) & 0xffff)
70 #define  MCX_CMDIF(_v)				((_v) >> 16)
71 
72 #define MCX_ISSI			1 /* as per the PRM */
73 #define MCX_CMD_IF_SUPPORTED		5
74 
75 #define MCX_HARDMTU			9500
76 
77 #define MCX_PAGE_SHIFT			12
78 #define MCX_PAGE_SIZE			(1 << MCX_PAGE_SHIFT)
79 
80 /* queue sizes */
81 #define MCX_LOG_EQ_SIZE			7
82 #define MCX_LOG_CQ_SIZE			12
83 #define MCX_LOG_RQ_SIZE			10
84 #define MCX_LOG_SQ_SIZE			11
85 
86 #define MCX_MAX_QUEUES			1
87 
88 /* completion event moderation - about 10khz, or 90% of the cq */
89 #define MCX_CQ_MOD_PERIOD		50
90 #define MCX_CQ_MOD_COUNTER		\
91 	(((1 << (MCX_LOG_CQ_SIZE - 1)) * 9) / 10)
92 
93 #define MCX_LOG_SQ_ENTRY_SIZE		6
94 #define MCX_SQ_ENTRY_MAX_SLOTS		4
95 #define MCX_SQ_SEGS_PER_SLOT		\
96 	(sizeof(struct mcx_sq_entry) / sizeof(struct mcx_sq_entry_seg))
97 #define MCX_SQ_MAX_SEGMENTS		\
98 	1 + ((MCX_SQ_ENTRY_MAX_SLOTS-1) * MCX_SQ_SEGS_PER_SLOT)
99 
100 #define MCX_LOG_FLOW_TABLE_SIZE		5
101 #define MCX_NUM_STATIC_FLOWS		4 /* promisc, allmulti, ucast, bcast */
102 #define MCX_NUM_MCAST_FLOWS 		\
103 	((1 << MCX_LOG_FLOW_TABLE_SIZE) - MCX_NUM_STATIC_FLOWS)
104 
105 #define MCX_SQ_INLINE_SIZE		18
106 CTASSERT(ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN == MCX_SQ_INLINE_SIZE);
107 
108 /* doorbell offsets */
109 #define MCX_DOORBELL_AREA_SIZE		MCX_PAGE_SIZE
110 
111 #define MCX_CQ_DOORBELL_BASE		0
112 #define MCX_CQ_DOORBELL_STRIDE		64
113 
114 #define MCX_WQ_DOORBELL_BASE		MCX_PAGE_SIZE/2
115 #define MCX_WQ_DOORBELL_STRIDE		64
116 /* make sure the doorbells fit */
117 CTASSERT(MCX_MAX_QUEUES * MCX_CQ_DOORBELL_STRIDE < MCX_WQ_DOORBELL_BASE);
118 CTASSERT(MCX_MAX_QUEUES * MCX_WQ_DOORBELL_STRIDE <
119     MCX_DOORBELL_AREA_SIZE - MCX_WQ_DOORBELL_BASE);
120 
121 #define MCX_WQ_DOORBELL_MASK		0xffff
122 
123 /* uar registers */
124 #define MCX_UAR_CQ_DOORBELL		0x20
125 #define MCX_UAR_EQ_DOORBELL_ARM		0x40
126 #define MCX_UAR_EQ_DOORBELL		0x48
127 #define MCX_UAR_BF			0x800
128 
129 #define MCX_CMDQ_ADDR_HI		0x0010
130 #define MCX_CMDQ_ADDR_LO		0x0014
131 #define MCX_CMDQ_ADDR_NMASK		0xfff
132 #define MCX_CMDQ_LOG_SIZE(_v)		((_v) >> 4 & 0xf)
133 #define MCX_CMDQ_LOG_STRIDE(_v)		((_v) >> 0 & 0xf)
134 #define MCX_CMDQ_INTERFACE_MASK		(0x3 << 8)
135 #define MCX_CMDQ_INTERFACE_FULL_DRIVER	(0x0 << 8)
136 #define MCX_CMDQ_INTERFACE_DISABLED	(0x1 << 8)
137 
138 #define MCX_CMDQ_DOORBELL		0x0018
139 
140 #define MCX_STATE			0x01fc
141 #define MCX_STATE_MASK				(1 << 31)
142 #define MCX_STATE_INITIALIZING			(1 << 31)
143 #define MCX_STATE_READY				(0 << 31)
144 #define MCX_STATE_INTERFACE_MASK		(0x3 << 24)
145 #define MCX_STATE_INTERFACE_FULL_DRIVER		(0x0 << 24)
146 #define MCX_STATE_INTERFACE_DISABLED		(0x1 << 24)
147 
148 #define MCX_INTERNAL_TIMER		0x1000
149 #define MCX_INTERNAL_TIMER_H		0x1000
150 #define MCX_INTERNAL_TIMER_L		0x1004
151 
152 #define MCX_CLEAR_INT			0x100c
153 
154 #define MCX_REG_OP_WRITE		0
155 #define MCX_REG_OP_READ			1
156 
157 #define MCX_REG_PMLP			0x5002
158 #define MCX_REG_PMTU			0x5003
159 #define MCX_REG_PTYS			0x5004
160 #define MCX_REG_PAOS			0x5006
161 #define MCX_REG_PFCC			0x5007
162 #define MCX_REG_PPCNT			0x5008
163 #define MCX_REG_MTCAP			0x9009 /* mgmt temp capabilities */
164 #define MCX_REG_MTMP			0x900a /* mgmt temp */
165 #define MCX_REG_MCIA			0x9014
166 
167 #define MCX_ETHER_CAP_SGMII		0
168 #define MCX_ETHER_CAP_1000_KX		1
169 #define MCX_ETHER_CAP_10G_CX4		2
170 #define MCX_ETHER_CAP_10G_KX4		3
171 #define MCX_ETHER_CAP_10G_KR		4
172 #define MCX_ETHER_CAP_40G_CR4		6
173 #define MCX_ETHER_CAP_40G_KR4		7
174 #define MCX_ETHER_CAP_10G_CR		12
175 #define MCX_ETHER_CAP_10G_SR		13
176 #define MCX_ETHER_CAP_10G_LR		14
177 #define MCX_ETHER_CAP_40G_SR4		15
178 #define MCX_ETHER_CAP_40G_LR4		16
179 #define MCX_ETHER_CAP_50G_SR2		18
180 #define MCX_ETHER_CAP_100G_CR4		20
181 #define MCX_ETHER_CAP_100G_SR4		21
182 #define MCX_ETHER_CAP_100G_KR4		22
183 #define MCX_ETHER_CAP_25G_CR		27
184 #define MCX_ETHER_CAP_25G_KR		28
185 #define MCX_ETHER_CAP_25G_SR		29
186 #define MCX_ETHER_CAP_50G_CR2		30
187 #define MCX_ETHER_CAP_50G_KR2		31
188 
189 #define MCX_MAX_CQE			32
190 
191 #define MCX_CMD_QUERY_HCA_CAP		0x100
192 #define MCX_CMD_QUERY_ADAPTER		0x101
193 #define MCX_CMD_INIT_HCA		0x102
194 #define MCX_CMD_TEARDOWN_HCA		0x103
195 #define MCX_CMD_ENABLE_HCA		0x104
196 #define MCX_CMD_DISABLE_HCA		0x105
197 #define MCX_CMD_QUERY_PAGES		0x107
198 #define MCX_CMD_MANAGE_PAGES		0x108
199 #define MCX_CMD_SET_HCA_CAP		0x109
200 #define MCX_CMD_QUERY_ISSI		0x10a
201 #define MCX_CMD_SET_ISSI		0x10b
202 #define MCX_CMD_SET_DRIVER_VERSION	0x10d
203 #define MCX_CMD_QUERY_SPECIAL_CONTEXTS	0x203
204 #define MCX_CMD_CREATE_EQ		0x301
205 #define MCX_CMD_DESTROY_EQ		0x302
206 #define MCX_CMD_QUERY_EQ		0x303
207 #define MCX_CMD_CREATE_CQ		0x400
208 #define MCX_CMD_DESTROY_CQ		0x401
209 #define MCX_CMD_QUERY_CQ		0x402
210 #define MCX_CMD_QUERY_NIC_VPORT_CONTEXT	0x754
211 #define MCX_CMD_MODIFY_NIC_VPORT_CONTEXT \
212 					0x755
213 #define MCX_CMD_QUERY_VPORT_COUNTERS	0x770
214 #define MCX_CMD_ALLOC_PD		0x800
215 #define MCX_CMD_ALLOC_UAR		0x802
216 #define MCX_CMD_ACCESS_REG		0x805
217 #define MCX_CMD_ALLOC_TRANSPORT_DOMAIN	0x816
218 #define MCX_CMD_CREATE_TIR		0x900
219 #define MCX_CMD_DESTROY_TIR		0x902
220 #define MCX_CMD_CREATE_SQ		0x904
221 #define MCX_CMD_MODIFY_SQ		0x905
222 #define MCX_CMD_DESTROY_SQ		0x906
223 #define MCX_CMD_QUERY_SQ		0x907
224 #define MCX_CMD_CREATE_RQ		0x908
225 #define MCX_CMD_MODIFY_RQ		0x909
226 #define MCX_CMD_DESTROY_RQ		0x90a
227 #define MCX_CMD_QUERY_RQ		0x90b
228 #define MCX_CMD_CREATE_TIS		0x912
229 #define MCX_CMD_DESTROY_TIS		0x914
230 #define MCX_CMD_CREATE_RQT		0x916
231 #define MCX_CMD_DESTROY_RQT		0x918
232 #define MCX_CMD_SET_FLOW_TABLE_ROOT	0x92f
233 #define MCX_CMD_CREATE_FLOW_TABLE	0x930
234 #define MCX_CMD_DESTROY_FLOW_TABLE	0x931
235 #define MCX_CMD_QUERY_FLOW_TABLE	0x932
236 #define MCX_CMD_CREATE_FLOW_GROUP	0x933
237 #define MCX_CMD_DESTROY_FLOW_GROUP	0x934
238 #define MCX_CMD_QUERY_FLOW_GROUP	0x935
239 #define MCX_CMD_SET_FLOW_TABLE_ENTRY	0x936
240 #define MCX_CMD_QUERY_FLOW_TABLE_ENTRY	0x937
241 #define MCX_CMD_DELETE_FLOW_TABLE_ENTRY	0x938
242 #define MCX_CMD_ALLOC_FLOW_COUNTER	0x939
243 #define MCX_CMD_QUERY_FLOW_COUNTER	0x93b
244 
245 #define MCX_QUEUE_STATE_RST		0
246 #define MCX_QUEUE_STATE_RDY		1
247 #define MCX_QUEUE_STATE_ERR		3
248 
249 #define MCX_FLOW_TABLE_TYPE_RX		0
250 #define MCX_FLOW_TABLE_TYPE_TX		1
251 
252 #define MCX_CMDQ_INLINE_DATASIZE 16
253 
254 struct mcx_cmdq_entry {
255 	uint8_t			cq_type;
256 #define MCX_CMDQ_TYPE_PCIE		0x7
257 	uint8_t			cq_reserved0[3];
258 
259 	uint32_t		cq_input_length;
260 	uint64_t		cq_input_ptr;
261 	uint8_t			cq_input_data[MCX_CMDQ_INLINE_DATASIZE];
262 
263 	uint8_t			cq_output_data[MCX_CMDQ_INLINE_DATASIZE];
264 	uint64_t		cq_output_ptr;
265 	uint32_t		cq_output_length;
266 
267 	uint8_t			cq_token;
268 	uint8_t			cq_signature;
269 	uint8_t			cq_reserved1[1];
270 	uint8_t			cq_status;
271 #define MCX_CQ_STATUS_SHIFT		1
272 #define MCX_CQ_STATUS_MASK		(0x7f << MCX_CQ_STATUS_SHIFT)
273 #define MCX_CQ_STATUS_OK		(0x00 << MCX_CQ_STATUS_SHIFT)
274 #define MCX_CQ_STATUS_INT_ERR		(0x01 << MCX_CQ_STATUS_SHIFT)
275 #define MCX_CQ_STATUS_BAD_OPCODE	(0x02 << MCX_CQ_STATUS_SHIFT)
276 #define MCX_CQ_STATUS_BAD_PARAM		(0x03 << MCX_CQ_STATUS_SHIFT)
277 #define MCX_CQ_STATUS_BAD_SYS_STATE	(0x04 << MCX_CQ_STATUS_SHIFT)
278 #define MCX_CQ_STATUS_BAD_RESOURCE	(0x05 << MCX_CQ_STATUS_SHIFT)
279 #define MCX_CQ_STATUS_RESOURCE_BUSY	(0x06 << MCX_CQ_STATUS_SHIFT)
280 #define MCX_CQ_STATUS_EXCEED_LIM	(0x08 << MCX_CQ_STATUS_SHIFT)
281 #define MCX_CQ_STATUS_BAD_RES_STATE	(0x09 << MCX_CQ_STATUS_SHIFT)
282 #define MCX_CQ_STATUS_BAD_INDEX		(0x0a << MCX_CQ_STATUS_SHIFT)
283 #define MCX_CQ_STATUS_NO_RESOURCES	(0x0f << MCX_CQ_STATUS_SHIFT)
284 #define MCX_CQ_STATUS_BAD_INPUT_LEN	(0x50 << MCX_CQ_STATUS_SHIFT)
285 #define MCX_CQ_STATUS_BAD_OUTPUT_LEN	(0x51 << MCX_CQ_STATUS_SHIFT)
286 #define MCX_CQ_STATUS_BAD_RESOURCE_STATE \
287 					(0x10 << MCX_CQ_STATUS_SHIFT)
288 #define MCX_CQ_STATUS_BAD_SIZE		(0x40 << MCX_CQ_STATUS_SHIFT)
289 #define MCX_CQ_STATUS_OWN_MASK		0x1
290 #define MCX_CQ_STATUS_OWN_SW		0x0
291 #define MCX_CQ_STATUS_OWN_HW		0x1
292 } __packed __aligned(8);
293 
294 #define MCX_CMDQ_MAILBOX_DATASIZE	512
295 
296 struct mcx_cmdq_mailbox {
297 	uint8_t			mb_data[MCX_CMDQ_MAILBOX_DATASIZE];
298 	uint8_t			mb_reserved0[48];
299 	uint64_t		mb_next_ptr;
300 	uint32_t		mb_block_number;
301 	uint8_t			mb_reserved1[1];
302 	uint8_t			mb_token;
303 	uint8_t			mb_ctrl_signature;
304 	uint8_t			mb_signature;
305 } __packed __aligned(8);
306 
307 #define MCX_CMDQ_MAILBOX_ALIGN	(1 << 10)
308 #define MCX_CMDQ_MAILBOX_SIZE	roundup(sizeof(struct mcx_cmdq_mailbox), \
309 				    MCX_CMDQ_MAILBOX_ALIGN)
310 /*
311  * command mailbox structres
312  */
313 
314 struct mcx_cmd_enable_hca_in {
315 	uint16_t		cmd_opcode;
316 	uint8_t			cmd_reserved0[4];
317 	uint16_t		cmd_op_mod;
318 	uint8_t			cmd_reserved1[2];
319 	uint16_t		cmd_function_id;
320 	uint8_t			cmd_reserved2[4];
321 } __packed __aligned(4);
322 
323 struct mcx_cmd_enable_hca_out {
324 	uint8_t			cmd_status;
325 	uint8_t			cmd_reserved0[3];
326 	uint32_t		cmd_syndrome;
327 	uint8_t			cmd_reserved1[4];
328 } __packed __aligned(4);
329 
330 struct mcx_cmd_init_hca_in {
331 	uint16_t		cmd_opcode;
332 	uint8_t			cmd_reserved0[4];
333 	uint16_t		cmd_op_mod;
334 	uint8_t			cmd_reserved1[8];
335 } __packed __aligned(4);
336 
337 struct mcx_cmd_init_hca_out {
338 	uint8_t			cmd_status;
339 	uint8_t			cmd_reserved0[3];
340 	uint32_t		cmd_syndrome;
341 	uint8_t			cmd_reserved1[8];
342 } __packed __aligned(4);
343 
344 struct mcx_cmd_teardown_hca_in {
345 	uint16_t		cmd_opcode;
346 	uint8_t			cmd_reserved0[4];
347 	uint16_t		cmd_op_mod;
348 	uint8_t			cmd_reserved1[2];
349 #define MCX_CMD_TEARDOWN_HCA_GRACEFUL	0x0
350 #define MCX_CMD_TEARDOWN_HCA_PANIC	0x1
351 	uint16_t		cmd_profile;
352 	uint8_t			cmd_reserved2[4];
353 } __packed __aligned(4);
354 
355 struct mcx_cmd_teardown_hca_out {
356 	uint8_t			cmd_status;
357 	uint8_t			cmd_reserved0[3];
358 	uint32_t		cmd_syndrome;
359 	uint8_t			cmd_reserved1[8];
360 } __packed __aligned(4);
361 
362 struct mcx_cmd_access_reg_in {
363 	uint16_t		cmd_opcode;
364 	uint8_t			cmd_reserved0[4];
365 	uint16_t		cmd_op_mod;
366 	uint8_t			cmd_reserved1[2];
367 	uint16_t		cmd_register_id;
368 	uint32_t		cmd_argument;
369 } __packed __aligned(4);
370 
371 struct mcx_cmd_access_reg_out {
372 	uint8_t			cmd_status;
373 	uint8_t			cmd_reserved0[3];
374 	uint32_t		cmd_syndrome;
375 	uint8_t			cmd_reserved1[8];
376 } __packed __aligned(4);
377 
378 struct mcx_reg_pmtu {
379 	uint8_t			rp_reserved1;
380 	uint8_t			rp_local_port;
381 	uint8_t			rp_reserved2[2];
382 	uint16_t		rp_max_mtu;
383 	uint8_t			rp_reserved3[2];
384 	uint16_t		rp_admin_mtu;
385 	uint8_t			rp_reserved4[2];
386 	uint16_t		rp_oper_mtu;
387 	uint8_t			rp_reserved5[2];
388 } __packed __aligned(4);
389 
390 struct mcx_reg_ptys {
391 	uint8_t			rp_reserved1;
392 	uint8_t			rp_local_port;
393 	uint8_t			rp_reserved2;
394 	uint8_t			rp_proto_mask;
395 #define MCX_REG_PTYS_PROTO_MASK_ETH		(1 << 2)
396 	uint8_t			rp_reserved3[8];
397 	uint32_t		rp_eth_proto_cap;
398 	uint8_t			rp_reserved4[8];
399 	uint32_t		rp_eth_proto_admin;
400 	uint8_t			rp_reserved5[8];
401 	uint32_t		rp_eth_proto_oper;
402 	uint8_t			rp_reserved6[24];
403 } __packed __aligned(4);
404 
405 struct mcx_reg_paos {
406 	uint8_t			rp_reserved1;
407 	uint8_t			rp_local_port;
408 	uint8_t			rp_admin_status;
409 #define MCX_REG_PAOS_ADMIN_STATUS_UP		1
410 #define MCX_REG_PAOS_ADMIN_STATUS_DOWN		2
411 #define MCX_REG_PAOS_ADMIN_STATUS_UP_ONCE	3
412 #define MCX_REG_PAOS_ADMIN_STATUS_DISABLED	4
413 	uint8_t			rp_oper_status;
414 #define MCX_REG_PAOS_OPER_STATUS_UP		1
415 #define MCX_REG_PAOS_OPER_STATUS_DOWN		2
416 #define MCX_REG_PAOS_OPER_STATUS_FAILED		4
417 	uint8_t			rp_admin_state_update;
418 #define MCX_REG_PAOS_ADMIN_STATE_UPDATE_EN	(1 << 7)
419 	uint8_t			rp_reserved2[11];
420 } __packed __aligned(4);
421 
422 struct mcx_reg_pfcc {
423 	uint8_t			rp_reserved1;
424 	uint8_t			rp_local_port;
425 	uint8_t			rp_reserved2[3];
426 	uint8_t			rp_prio_mask_tx;
427 	uint8_t			rp_reserved3;
428 	uint8_t			rp_prio_mask_rx;
429 	uint8_t			rp_pptx_aptx;
430 	uint8_t			rp_pfctx;
431 	uint8_t			rp_fctx_dis;
432 	uint8_t			rp_reserved4;
433 	uint8_t			rp_pprx_aprx;
434 	uint8_t			rp_pfcrx;
435 	uint8_t			rp_reserved5[2];
436 	uint16_t		rp_dev_stall_min;
437 	uint16_t		rp_dev_stall_crit;
438 	uint8_t			rp_reserved6[12];
439 } __packed __aligned(4);
440 
441 #define MCX_PMLP_MODULE_NUM_MASK	0xff
442 struct mcx_reg_pmlp {
443 	uint8_t			rp_rxtx;
444 	uint8_t			rp_local_port;
445 	uint8_t			rp_reserved0;
446 	uint8_t			rp_width;
447 	uint32_t		rp_lane0_mapping;
448 	uint32_t		rp_lane1_mapping;
449 	uint32_t		rp_lane2_mapping;
450 	uint32_t		rp_lane3_mapping;
451 	uint8_t			rp_reserved1[44];
452 } __packed __aligned(4);
453 
454 struct mcx_reg_ppcnt {
455 	uint8_t			ppcnt_swid;
456 	uint8_t			ppcnt_local_port;
457 	uint8_t			ppcnt_pnat;
458 	uint8_t			ppcnt_grp;
459 #define MCX_REG_PPCNT_GRP_IEEE8023		0x00
460 #define MCX_REG_PPCNT_GRP_RFC2863		0x01
461 #define MCX_REG_PPCNT_GRP_RFC2819		0x02
462 #define MCX_REG_PPCNT_GRP_RFC3635		0x03
463 #define MCX_REG_PPCNT_GRP_PER_PRIO		0x10
464 #define MCX_REG_PPCNT_GRP_PER_TC		0x11
465 #define MCX_REG_PPCNT_GRP_PER_RX_BUFFER		0x11
466 
467 	uint8_t			ppcnt_clr;
468 	uint8_t			ppcnt_reserved1[2];
469 	uint8_t			ppcnt_prio_tc;
470 #define MCX_REG_PPCNT_CLR			(1 << 7)
471 
472 	uint8_t			ppcnt_counter_set[248];
473 } __packed __aligned(8);
474 CTASSERT(sizeof(struct mcx_reg_ppcnt) == 256);
475 CTASSERT((offsetof(struct mcx_reg_ppcnt, ppcnt_counter_set) %
476     sizeof(uint64_t)) == 0);
477 
478 enum mcx_ppcnt_ieee8023 {
479 	frames_transmitted_ok,
480 	frames_received_ok,
481 	frame_check_sequence_errors,
482 	alignment_errors,
483 	octets_transmitted_ok,
484 	octets_received_ok,
485 	multicast_frames_xmitted_ok,
486 	broadcast_frames_xmitted_ok,
487 	multicast_frames_received_ok,
488 	broadcast_frames_received_ok,
489 	in_range_length_errors,
490 	out_of_range_length_field,
491 	frame_too_long_errors,
492 	symbol_error_during_carrier,
493 	mac_control_frames_transmitted,
494 	mac_control_frames_received,
495 	unsupported_opcodes_received,
496 	pause_mac_ctrl_frames_received,
497 	pause_mac_ctrl_frames_transmitted,
498 
499 	mcx_ppcnt_ieee8023_count
500 };
501 CTASSERT(mcx_ppcnt_ieee8023_count * sizeof(uint64_t) == 0x98);
502 
503 enum mcx_ppcnt_rfc2863 {
504 	in_octets,
505 	in_ucast_pkts,
506 	in_discards,
507 	in_errors,
508 	in_unknown_protos,
509 	out_octets,
510 	out_ucast_pkts,
511 	out_discards,
512 	out_errors,
513 	in_multicast_pkts,
514 	in_broadcast_pkts,
515 	out_multicast_pkts,
516 	out_broadcast_pkts,
517 
518 	mcx_ppcnt_rfc2863_count
519 };
520 CTASSERT(mcx_ppcnt_rfc2863_count * sizeof(uint64_t) == 0x68);
521 
522 enum mcx_ppcnt_rfc2819 {
523 	drop_events,
524 	octets,
525 	pkts,
526 	broadcast_pkts,
527 	multicast_pkts,
528 	crc_align_errors,
529 	undersize_pkts,
530 	oversize_pkts,
531 	fragments,
532 	jabbers,
533 	collisions,
534 	pkts64octets,
535 	pkts65to127octets,
536 	pkts128to255octets,
537 	pkts256to511octets,
538 	pkts512to1023octets,
539 	pkts1024to1518octets,
540 	pkts1519to2047octets,
541 	pkts2048to4095octets,
542 	pkts4096to8191octets,
543 	pkts8192to10239octets,
544 
545 	mcx_ppcnt_rfc2819_count
546 };
547 CTASSERT((mcx_ppcnt_rfc2819_count * sizeof(uint64_t)) == 0xa8);
548 
549 enum mcx_ppcnt_rfc3635 {
550 	dot3stats_alignment_errors,
551 	dot3stats_fcs_errors,
552 	dot3stats_single_collision_frames,
553 	dot3stats_multiple_collision_frames,
554 	dot3stats_sqe_test_errors,
555 	dot3stats_deferred_transmissions,
556 	dot3stats_late_collisions,
557 	dot3stats_excessive_collisions,
558 	dot3stats_internal_mac_transmit_errors,
559 	dot3stats_carrier_sense_errors,
560 	dot3stats_frame_too_longs,
561 	dot3stats_internal_mac_receive_errors,
562 	dot3stats_symbol_errors,
563 	dot3control_in_unknown_opcodes,
564 	dot3in_pause_frames,
565 	dot3out_pause_frames,
566 
567 	mcx_ppcnt_rfc3635_count
568 };
569 CTASSERT((mcx_ppcnt_rfc3635_count * sizeof(uint64_t)) == 0x80);
570 
571 struct mcx_reg_mtcap {
572 	uint8_t			_reserved1[3];
573 	uint8_t			mtcap_sensor_count;
574 	uint8_t			_reserved2[4];
575 
576 	uint64_t		mtcap_sensor_map;
577 };
578 
579 struct mcx_reg_mtmp {
580 	uint8_t			_reserved1[2];
581 	uint16_t		mtmp_sensor_index;
582 
583 	uint8_t			_reserved2[2];
584 	uint16_t		mtmp_temperature;
585 
586 	uint16_t		mtmp_mte_mtr;
587 #define MCX_REG_MTMP_MTE		(1 << 15)
588 #define MCX_REG_MTMP_MTR		(1 << 14)
589 	uint16_t		mtmp_max_temperature;
590 
591 	uint16_t		mtmp_tee;
592 #define MCX_REG_MTMP_TEE_NOPE		(0 << 14)
593 #define MCX_REG_MTMP_TEE_GENERATE	(1 << 14)
594 #define MCX_REG_MTMP_TEE_GENERATE_ONE	(2 << 14)
595 	uint16_t		mtmp_temperature_threshold_hi;
596 
597 	uint8_t			_reserved3[2];
598 	uint16_t		mtmp_temperature_threshold_lo;
599 
600 	uint8_t			_reserved4[4];
601 
602 	uint8_t			mtmp_sensor_name[8];
603 };
604 CTASSERT(sizeof(struct mcx_reg_mtmp) == 0x20);
605 CTASSERT(offsetof(struct mcx_reg_mtmp, mtmp_sensor_name) == 0x18);
606 
607 #define MCX_MCIA_EEPROM_BYTES	32
608 struct mcx_reg_mcia {
609 	uint8_t			rm_l;
610 	uint8_t			rm_module;
611 	uint8_t			rm_reserved0;
612 	uint8_t			rm_status;
613 	uint8_t			rm_i2c_addr;
614 	uint8_t			rm_page_num;
615 	uint16_t		rm_dev_addr;
616 	uint16_t		rm_reserved1;
617 	uint16_t		rm_size;
618 	uint32_t		rm_reserved2;
619 	uint8_t			rm_data[48];
620 } __packed __aligned(4);
621 
622 struct mcx_cmd_query_issi_in {
623 	uint16_t		cmd_opcode;
624 	uint8_t			cmd_reserved0[4];
625 	uint16_t		cmd_op_mod;
626 	uint8_t			cmd_reserved1[8];
627 } __packed __aligned(4);
628 
629 struct mcx_cmd_query_issi_il_out {
630 	uint8_t			cmd_status;
631 	uint8_t			cmd_reserved0[3];
632 	uint32_t		cmd_syndrome;
633 	uint8_t			cmd_reserved1[2];
634 	uint16_t		cmd_current_issi;
635 	uint8_t			cmd_reserved2[4];
636 } __packed __aligned(4);
637 
638 CTASSERT(sizeof(struct mcx_cmd_query_issi_il_out) == MCX_CMDQ_INLINE_DATASIZE);
639 
640 struct mcx_cmd_query_issi_mb_out {
641 	uint8_t			cmd_reserved2[16];
642 	uint8_t			cmd_supported_issi[80]; /* very big endian */
643 } __packed __aligned(4);
644 
645 CTASSERT(sizeof(struct mcx_cmd_query_issi_mb_out) <= MCX_CMDQ_MAILBOX_DATASIZE);
646 
647 struct mcx_cmd_set_issi_in {
648 	uint16_t		cmd_opcode;
649 	uint8_t			cmd_reserved0[4];
650 	uint16_t		cmd_op_mod;
651 	uint8_t			cmd_reserved1[2];
652 	uint16_t		cmd_current_issi;
653 	uint8_t			cmd_reserved2[4];
654 } __packed __aligned(4);
655 
656 CTASSERT(sizeof(struct mcx_cmd_set_issi_in) <= MCX_CMDQ_INLINE_DATASIZE);
657 
658 struct mcx_cmd_set_issi_out {
659 	uint8_t			cmd_status;
660 	uint8_t			cmd_reserved0[3];
661 	uint32_t		cmd_syndrome;
662 	uint8_t			cmd_reserved1[8];
663 } __packed __aligned(4);
664 
665 CTASSERT(sizeof(struct mcx_cmd_set_issi_out) <= MCX_CMDQ_INLINE_DATASIZE);
666 
667 struct mcx_cmd_query_pages_in {
668 	uint16_t		cmd_opcode;
669 	uint8_t			cmd_reserved0[4];
670 	uint16_t		cmd_op_mod;
671 #define MCX_CMD_QUERY_PAGES_BOOT	0x01
672 #define MCX_CMD_QUERY_PAGES_INIT	0x02
673 #define MCX_CMD_QUERY_PAGES_REGULAR	0x03
674 	uint8_t			cmd_reserved1[8];
675 } __packed __aligned(4);
676 
677 struct mcx_cmd_query_pages_out {
678 	uint8_t			cmd_status;
679 	uint8_t			cmd_reserved0[3];
680 	uint32_t		cmd_syndrome;
681 	uint8_t			cmd_reserved1[2];
682 	uint16_t		cmd_func_id;
683 	int32_t			cmd_num_pages;
684 } __packed __aligned(4);
685 
686 struct mcx_cmd_manage_pages_in {
687 	uint16_t		cmd_opcode;
688 	uint8_t			cmd_reserved0[4];
689 	uint16_t		cmd_op_mod;
690 #define MCX_CMD_MANAGE_PAGES_ALLOC_FAIL \
691 					0x00
692 #define MCX_CMD_MANAGE_PAGES_ALLOC_SUCCESS \
693 					0x01
694 #define MCX_CMD_MANAGE_PAGES_HCA_RETURN_PAGES \
695 					0x02
696 	uint8_t			cmd_reserved1[2];
697 	uint16_t		cmd_func_id;
698 	uint32_t		cmd_input_num_entries;
699 } __packed __aligned(4);
700 
701 CTASSERT(sizeof(struct mcx_cmd_manage_pages_in) == MCX_CMDQ_INLINE_DATASIZE);
702 
703 struct mcx_cmd_manage_pages_out {
704 	uint8_t			cmd_status;
705 	uint8_t			cmd_reserved0[3];
706 	uint32_t		cmd_syndrome;
707 	uint32_t		cmd_output_num_entries;
708 	uint8_t			cmd_reserved1[4];
709 } __packed __aligned(4);
710 
711 CTASSERT(sizeof(struct mcx_cmd_manage_pages_out) == MCX_CMDQ_INLINE_DATASIZE);
712 
713 struct mcx_cmd_query_hca_cap_in {
714 	uint16_t		cmd_opcode;
715 	uint8_t			cmd_reserved0[4];
716 	uint16_t		cmd_op_mod;
717 #define MCX_CMD_QUERY_HCA_CAP_MAX	(0x0 << 0)
718 #define MCX_CMD_QUERY_HCA_CAP_CURRENT	(0x1 << 0)
719 #define MCX_CMD_QUERY_HCA_CAP_DEVICE	(0x0 << 1)
720 #define MCX_CMD_QUERY_HCA_CAP_OFFLOAD	(0x1 << 1)
721 #define MCX_CMD_QUERY_HCA_CAP_FLOW	(0x7 << 1)
722 	uint8_t			cmd_reserved1[8];
723 } __packed __aligned(4);
724 
725 struct mcx_cmd_query_hca_cap_out {
726 	uint8_t			cmd_status;
727 	uint8_t			cmd_reserved0[3];
728 	uint32_t		cmd_syndrome;
729 	uint8_t			cmd_reserved1[8];
730 } __packed __aligned(4);
731 
732 #define MCX_HCA_CAP_LEN			0x1000
733 #define MCX_HCA_CAP_NMAILBOXES		\
734 	(MCX_HCA_CAP_LEN / MCX_CMDQ_MAILBOX_DATASIZE)
735 
736 #if __GNUC_PREREQ__(4, 3)
737 #define __counter__		__COUNTER__
738 #else
739 #define __counter__		__LINE__
740 #endif
741 
742 #define __token(_tok, _num)	_tok##_num
743 #define _token(_tok, _num)	__token(_tok, _num)
744 #define __reserved__		_token(__reserved, __counter__)
745 
746 struct mcx_cap_device {
747 	uint8_t			reserved0[16];
748 
749 	uint8_t			log_max_srq_sz;
750 	uint8_t			log_max_qp_sz;
751 	uint8_t			__reserved__[1];
752 	uint8_t			log_max_qp; /* 5 bits */
753 #define MCX_CAP_DEVICE_LOG_MAX_QP	0x1f
754 
755 	uint8_t			__reserved__[1];
756 	uint8_t			log_max_srq; /* 5 bits */
757 #define MCX_CAP_DEVICE_LOG_MAX_SRQ	0x1f
758 	uint8_t			__reserved__[2];
759 
760 	uint8_t			__reserved__[1];
761 	uint8_t			log_max_cq_sz;
762 	uint8_t			__reserved__[1];
763 	uint8_t			log_max_cq; /* 5 bits */
764 #define MCX_CAP_DEVICE_LOG_MAX_CQ	0x1f
765 
766 	uint8_t			log_max_eq_sz;
767 	uint8_t			log_max_mkey; /* 6 bits */
768 #define MCX_CAP_DEVICE_LOG_MAX_MKEY	0x3f
769 	uint8_t			__reserved__[1];
770 	uint8_t			log_max_eq; /* 4 bits */
771 #define MCX_CAP_DEVICE_LOG_MAX_EQ	0x0f
772 
773 	uint8_t			max_indirection;
774 	uint8_t			log_max_mrw_sz; /* 7 bits */
775 #define MCX_CAP_DEVICE_LOG_MAX_MRW_SZ	0x7f
776 	uint8_t			teardown_log_max_msf_list_size;
777 #define MCX_CAP_DEVICE_FORCE_TEARDOWN	0x80
778 #define MCX_CAP_DEVICE_LOG_MAX_MSF_LIST_SIZE \
779 					0x3f
780 	uint8_t			log_max_klm_list_size; /* 6 bits */
781 #define MCX_CAP_DEVICE_LOG_MAX_KLM_LIST_SIZE \
782 					0x3f
783 
784 	uint8_t			__reserved__[1];
785 	uint8_t			log_max_ra_req_dc; /* 6 bits */
786 #define MCX_CAP_DEVICE_LOG_MAX_REQ_DC	0x3f
787 	uint8_t			__reserved__[1];
788 	uint8_t			log_max_ra_res_dc; /* 6 bits */
789 #define MCX_CAP_DEVICE_LOG_MAX_RA_RES_DC \
790 					0x3f
791 
792 	uint8_t			__reserved__[1];
793 	uint8_t			log_max_ra_req_qp; /* 6 bits */
794 #define MCX_CAP_DEVICE_LOG_MAX_RA_REQ_QP \
795 					0x3f
796 	uint8_t			__reserved__[1];
797 	uint8_t			log_max_ra_res_qp; /* 6 bits */
798 #define MCX_CAP_DEVICE_LOG_MAX_RA_RES_QP \
799 					0x3f
800 
801 	uint8_t			flags1;
802 #define MCX_CAP_DEVICE_END_PAD		0x80
803 #define MCX_CAP_DEVICE_CC_QUERY_ALLOWED	0x40
804 #define MCX_CAP_DEVICE_CC_MODIFY_ALLOWED \
805 					0x20
806 #define MCX_CAP_DEVICE_START_PAD	0x10
807 #define MCX_CAP_DEVICE_128BYTE_CACHELINE \
808 					0x08
809 	uint8_t			__reserved__[1];
810 	uint16_t		gid_table_size;
811 
812 	uint16_t		flags2;
813 #define MCX_CAP_DEVICE_OUT_OF_SEQ_CNT	0x8000
814 #define MCX_CAP_DEVICE_VPORT_COUNTERS	0x4000
815 #define MCX_CAP_DEVICE_RETRANSMISSION_Q_COUNTERS \
816 					0x2000
817 #define MCX_CAP_DEVICE_DEBUG		0x1000
818 #define MCX_CAP_DEVICE_MODIFY_RQ_COUNTERS_SET_ID \
819 					0x8000
820 #define MCX_CAP_DEVICE_RQ_DELAY_DROP	0x4000
821 #define MCX_CAP_DEVICe_MAX_QP_CNT_MASK	0x03ff
822 	uint16_t		pkey_table_size;
823 
824 	uint8_t			flags3;
825 #define MCX_CAP_DEVICE_VPORT_GROUP_MANAGER \
826 					0x80
827 #define MCX_CAP_DEVICE_VHCA_GROUP_MANAGER \
828 					0x40
829 #define MCX_CAP_DEVICE_IB_VIRTUAL	0x20
830 #define MCX_CAP_DEVICE_ETH_VIRTUAL	0x10
831 #define MCX_CAP_DEVICE_ETS		0x04
832 #define MCX_CAP_DEVICE_NIC_FLOW_TABLE	0x02
833 #define MCX_CAP_DEVICE_ESWITCH_FLOW_TABLE \
834 					0x01
835 	uint8_t			local_ca_ack_delay; /* 5 bits */
836 #define MCX_CAP_DEVICE_LOCAL_CA_ACK_DELAY \
837 					0x1f
838 	uint8_t			port_type;
839 #define MCX_CAP_DEVICE_PORT_MODULE_EVENT \
840 					0x80
841 #define MCX_CAP_DEVICE_PORT_TYPE	0x03
842 	uint8_t			num_ports;
843 
844 	uint8_t			snapshot_log_max_msg;
845 #define MCX_CAP_DEVICE_SNAPSHOT		0x80
846 #define MCX_CAP_DEVICE_LOG_MAX_MSG	0x1f
847 	uint8_t			max_tc; /* 4 bits */
848 #define MCX_CAP_DEVICE_MAX_TC		0x0f
849 	uint8_t			flags4;
850 #define MCX_CAP_DEVICE_TEMP_WARN_EVENT	0x80
851 #define MCX_CAP_DEVICE_DCBX		0x40
852 #define MCX_CAP_DEVICE_ROL_S		0x02
853 #define MCX_CAP_DEVICE_ROL_G		0x01
854 	uint8_t			wol;
855 #define MCX_CAP_DEVICE_WOL_S		0x40
856 #define MCX_CAP_DEVICE_WOL_G		0x20
857 #define MCX_CAP_DEVICE_WOL_A		0x10
858 #define MCX_CAP_DEVICE_WOL_B		0x08
859 #define MCX_CAP_DEVICE_WOL_M		0x04
860 #define MCX_CAP_DEVICE_WOL_U		0x02
861 #define MCX_CAP_DEVICE_WOL_P		0x01
862 
863 	uint16_t		stat_rate_support;
864 	uint8_t			__reserved__[1];
865 	uint8_t			cqe_version; /* 4 bits */
866 #define MCX_CAP_DEVICE_CQE_VERSION	0x0f
867 
868 	uint32_t		flags5;
869 #define MCX_CAP_DEVICE_COMPACT_ADDRESS_VECTOR \
870 					0x80000000
871 #define MCX_CAP_DEVICE_STRIDING_RQ	0x40000000
872 #define MCX_CAP_DEVICE_IPOIP_ENHANCED_OFFLOADS \
873 					0x10000000
874 #define MCX_CAP_DEVICE_IPOIP_IPOIP_OFFLOADS \
875 					0x08000000
876 #define MCX_CAP_DEVICE_DC_CONNECT_CP	0x00040000
877 #define MCX_CAP_DEVICE_DC_CNAK_DRACE	0x00020000
878 #define MCX_CAP_DEVICE_DRAIN_SIGERR	0x00010000
879 #define MCX_CAP_DEVICE_DRAIN_SIGERR	0x00010000
880 #define MCX_CAP_DEVICE_CMDIF_CHECKSUM	0x0000c000
881 #define MCX_CAP_DEVICE_SIGERR_QCE	0x00002000
882 #define MCX_CAP_DEVICE_WQ_SIGNATURE	0x00000800
883 #define MCX_CAP_DEVICE_SCTR_DATA_CQE	0x00000400
884 #define MCX_CAP_DEVICE_SHO		0x00000100
885 #define MCX_CAP_DEVICE_TPH		0x00000080
886 #define MCX_CAP_DEVICE_RF		0x00000040
887 #define MCX_CAP_DEVICE_DCT		0x00000020
888 #define MCX_CAP_DEVICE_QOS		0x00000010
889 #define MCX_CAP_DEVICe_ETH_NET_OFFLOADS	0x00000008
890 #define MCX_CAP_DEVICE_ROCE		0x00000004
891 #define MCX_CAP_DEVICE_ATOMIC		0x00000002
892 
893 	uint32_t		flags6;
894 #define MCX_CAP_DEVICE_CQ_OI		0x80000000
895 #define MCX_CAP_DEVICE_CQ_RESIZE	0x40000000
896 #define MCX_CAP_DEVICE_CQ_MODERATION	0x20000000
897 #define MCX_CAP_DEVICE_CQ_PERIOD_MODE_MODIFY \
898 					0x10000000
899 #define MCX_CAP_DEVICE_CQ_INVALIDATE	0x08000000
900 #define MCX_CAP_DEVICE_RESERVED_AT_255	0x04000000
901 #define MCX_CAP_DEVICE_CQ_EQ_REMAP	0x02000000
902 #define MCX_CAP_DEVICE_PG		0x01000000
903 #define MCX_CAP_DEVICE_BLOCK_LB_MC	0x00800000
904 #define MCX_CAP_DEVICE_EXPONENTIAL_BACKOFF \
905 					0x00400000
906 #define MCX_CAP_DEVICE_SCQE_BREAK_MODERATION \
907 					0x00200000
908 #define MCX_CAP_DEVICE_CQ_PERIOD_START_FROM_CQE \
909 					0x00100000
910 #define MCX_CAP_DEVICE_CD		0x00080000
911 #define MCX_CAP_DEVICE_ATM		0x00040000
912 #define MCX_CAP_DEVICE_APM		0x00020000
913 #define MCX_CAP_DEVICE_IMAICL		0x00010000
914 #define MCX_CAP_DEVICE_QKV		0x00000200
915 #define MCX_CAP_DEVICE_PKV		0x00000100
916 #define MCX_CAP_DEVICE_SET_DETH_SQPN	0x00000080
917 #define MCX_CAP_DEVICE_XRC		0x00000008
918 #define MCX_CAP_DEVICE_UD		0x00000004
919 #define MCX_CAP_DEVICE_UC		0x00000002
920 #define MCX_CAP_DEVICE_RC		0x00000001
921 
922 	uint8_t			uar_flags;
923 #define MCX_CAP_DEVICE_UAR_4K		0x80
924 	uint8_t			uar_sz;	/* 6 bits */
925 #define MCX_CAP_DEVICE_UAR_SZ		0x3f
926 	uint8_t			__reserved__[1];
927 	uint8_t			log_pg_sz;
928 
929 	uint8_t			flags7;
930 #define MCX_CAP_DEVICE_BF		0x80
931 #define MCX_CAP_DEVICE_DRIVER_VERSION	0x40
932 #define MCX_CAP_DEVICE_PAD_TX_ETH_PACKET \
933 					0x20
934 	uint8_t			log_bf_reg_size; /* 5 bits */
935 #define MCX_CAP_DEVICE_LOG_BF_REG_SIZE	0x1f
936 	uint8_t			__reserved__[2];
937 
938 	uint16_t		num_of_diagnostic_counters;
939 	uint16_t		max_wqe_sz_sq;
940 
941 	uint8_t			__reserved__[2];
942 	uint16_t		max_wqe_sz_rq;
943 
944 	uint8_t			__reserved__[2];
945 	uint16_t		max_wqe_sz_sq_dc;
946 
947 	uint32_t		max_qp_mcg; /* 25 bits */
948 #define MCX_CAP_DEVICE_MAX_QP_MCG	0x1ffffff
949 
950 	uint8_t			__reserved__[3];
951 	uint8_t			log_max_mcq;
952 
953 	uint8_t			log_max_transport_domain; /* 5 bits */
954 #define MCX_CAP_DEVICE_LOG_MAX_TRANSORT_DOMAIN \
955 					0x1f
956 	uint8_t			log_max_pd; /* 5 bits */
957 #define MCX_CAP_DEVICE_LOG_MAX_PD	0x1f
958 	uint8_t			__reserved__[1];
959 	uint8_t			log_max_xrcd; /* 5 bits */
960 #define MCX_CAP_DEVICE_LOG_MAX_XRCD	0x1f
961 
962 	uint8_t			__reserved__[2];
963 	uint16_t		max_flow_counter;
964 
965 	uint8_t			log_max_rq; /* 5 bits */
966 #define MCX_CAP_DEVICE_LOG_MAX_RQ	0x1f
967 	uint8_t			log_max_sq; /* 5 bits */
968 #define MCX_CAP_DEVICE_LOG_MAX_SQ	0x1f
969 	uint8_t			log_max_tir; /* 5 bits */
970 #define MCX_CAP_DEVICE_LOG_MAX_TIR	0x1f
971 	uint8_t			log_max_tis; /* 5 bits */
972 #define MCX_CAP_DEVICE_LOG_MAX_TIS	0x1f
973 
974 	uint8_t 		flags8;
975 #define MCX_CAP_DEVICE_BASIC_CYCLIC_RCV_WQE \
976 					0x80
977 #define MCX_CAP_DEVICE_LOG_MAX_RMP	0x1f
978 	uint8_t			log_max_rqt; /* 5 bits */
979 #define MCX_CAP_DEVICE_LOG_MAX_RQT	0x1f
980 	uint8_t			log_max_rqt_size; /* 5 bits */
981 #define MCX_CAP_DEVICE_LOG_MAX_RQT_SIZE	0x1f
982 	uint8_t			log_max_tis_per_sq; /* 5 bits */
983 #define MCX_CAP_DEVICE_LOG_MAX_TIS_PER_SQ \
984 					0x1f
985 } __packed __aligned(8);
986 
987 CTASSERT(offsetof(struct mcx_cap_device, max_indirection) == 0x20);
988 CTASSERT(offsetof(struct mcx_cap_device, flags1) == 0x2c);
989 CTASSERT(offsetof(struct mcx_cap_device, flags2) == 0x30);
990 CTASSERT(offsetof(struct mcx_cap_device, snapshot_log_max_msg) == 0x38);
991 CTASSERT(offsetof(struct mcx_cap_device, flags5) == 0x40);
992 CTASSERT(offsetof(struct mcx_cap_device, flags7) == 0x4c);
993 CTASSERT(sizeof(struct mcx_cap_device) <= MCX_CMDQ_MAILBOX_DATASIZE);
994 
995 struct mcx_cmd_set_driver_version_in {
996 	uint16_t		cmd_opcode;
997 	uint8_t			cmd_reserved0[4];
998 	uint16_t		cmd_op_mod;
999 	uint8_t			cmd_reserved1[8];
1000 } __packed __aligned(4);
1001 
1002 struct mcx_cmd_set_driver_version_out {
1003 	uint8_t			cmd_status;
1004 	uint8_t			cmd_reserved0[3];
1005 	uint32_t		cmd_syndrome;
1006 	uint8_t			cmd_reserved1[8];
1007 } __packed __aligned(4);
1008 
1009 struct mcx_cmd_set_driver_version {
1010 	uint8_t			cmd_driver_version[64];
1011 } __packed __aligned(8);
1012 
1013 struct mcx_cmd_modify_nic_vport_context_in {
1014 	uint16_t		cmd_opcode;
1015 	uint8_t			cmd_reserved0[4];
1016 	uint16_t		cmd_op_mod;
1017 	uint8_t			cmd_reserved1[4];
1018 	uint32_t		cmd_field_select;
1019 #define MCX_CMD_MODIFY_NIC_VPORT_CONTEXT_FIELD_ADDR	0x04
1020 #define MCX_CMD_MODIFY_NIC_VPORT_CONTEXT_FIELD_PROMISC	0x10
1021 #define MCX_CMD_MODIFY_NIC_VPORT_CONTEXT_FIELD_MTU	0x40
1022 } __packed __aligned(4);
1023 
1024 struct mcx_cmd_modify_nic_vport_context_out {
1025 	uint8_t			cmd_status;
1026 	uint8_t			cmd_reserved0[3];
1027 	uint32_t		cmd_syndrome;
1028 	uint8_t			cmd_reserved1[8];
1029 } __packed __aligned(4);
1030 
1031 struct mcx_cmd_query_nic_vport_context_in {
1032 	uint16_t		cmd_opcode;
1033 	uint8_t			cmd_reserved0[4];
1034 	uint16_t		cmd_op_mod;
1035 	uint8_t			cmd_reserved1[4];
1036 	uint8_t			cmd_allowed_list_type;
1037 	uint8_t			cmd_reserved2[3];
1038 } __packed __aligned(4);
1039 
1040 struct mcx_cmd_query_nic_vport_context_out {
1041 	uint8_t			cmd_status;
1042 	uint8_t			cmd_reserved0[3];
1043 	uint32_t		cmd_syndrome;
1044 	uint8_t			cmd_reserved1[8];
1045 } __packed __aligned(4);
1046 
1047 struct mcx_nic_vport_ctx {
1048 	uint32_t		vp_min_wqe_inline_mode;
1049 	uint8_t			vp_reserved0[32];
1050 	uint32_t		vp_mtu;
1051 	uint8_t			vp_reserved1[200];
1052 	uint16_t		vp_flags;
1053 #define MCX_NIC_VPORT_CTX_LIST_UC_MAC			(0)
1054 #define MCX_NIC_VPORT_CTX_LIST_MC_MAC			(1 << 24)
1055 #define MCX_NIC_VPORT_CTX_LIST_VLAN			(2 << 24)
1056 #define MCX_NIC_VPORT_CTX_PROMISC_ALL			(1 << 13)
1057 #define MCX_NIC_VPORT_CTX_PROMISC_MCAST			(1 << 14)
1058 #define MCX_NIC_VPORT_CTX_PROMISC_UCAST			(1 << 15)
1059 	uint16_t		vp_allowed_list_size;
1060 	uint64_t		vp_perm_addr;
1061 	uint8_t			vp_reserved2[4];
1062 	/* allowed list follows */
1063 } __packed __aligned(4);
1064 
1065 struct mcx_counter {
1066 	uint64_t		packets;
1067 	uint64_t		octets;
1068 } __packed __aligned(4);
1069 
1070 struct mcx_nic_vport_counters {
1071 	struct mcx_counter	rx_err;
1072 	struct mcx_counter	tx_err;
1073 	uint8_t			reserved0[64]; /* 0x30 */
1074 	struct mcx_counter	rx_bcast;
1075 	struct mcx_counter	tx_bcast;
1076 	struct mcx_counter	rx_ucast;
1077 	struct mcx_counter	tx_ucast;
1078 	struct mcx_counter	rx_mcast;
1079 	struct mcx_counter	tx_mcast;
1080 	uint8_t			reserved1[0x210 - 0xd0];
1081 } __packed __aligned(4);
1082 
1083 struct mcx_cmd_query_vport_counters_in {
1084 	uint16_t		cmd_opcode;
1085 	uint8_t			cmd_reserved0[4];
1086 	uint16_t		cmd_op_mod;
1087 	uint8_t			cmd_reserved1[8];
1088 } __packed __aligned(4);
1089 
1090 struct mcx_cmd_query_vport_counters_mb_in {
1091 	uint8_t			cmd_reserved0[8];
1092 	uint8_t			cmd_clear;
1093 	uint8_t			cmd_reserved1[7];
1094 } __packed __aligned(4);
1095 
1096 struct mcx_cmd_query_vport_counters_out {
1097 	uint8_t			cmd_status;
1098 	uint8_t			cmd_reserved0[3];
1099 	uint32_t		cmd_syndrome;
1100 	uint8_t			cmd_reserved1[8];
1101 } __packed __aligned(4);
1102 
1103 struct mcx_cmd_query_flow_counter_in {
1104 	uint16_t		cmd_opcode;
1105 	uint8_t			cmd_reserved0[4];
1106 	uint16_t		cmd_op_mod;
1107 	uint8_t			cmd_reserved1[8];
1108 } __packed __aligned(4);
1109 
1110 struct mcx_cmd_query_flow_counter_mb_in {
1111 	uint8_t			cmd_reserved0[8];
1112 	uint8_t			cmd_clear;
1113 	uint8_t			cmd_reserved1[5];
1114 	uint16_t		cmd_flow_counter_id;
1115 } __packed __aligned(4);
1116 
1117 struct mcx_cmd_query_flow_counter_out {
1118 	uint8_t			cmd_status;
1119 	uint8_t			cmd_reserved0[3];
1120 	uint32_t		cmd_syndrome;
1121 	uint8_t			cmd_reserved1[8];
1122 } __packed __aligned(4);
1123 
1124 struct mcx_cmd_alloc_uar_in {
1125 	uint16_t		cmd_opcode;
1126 	uint8_t			cmd_reserved0[4];
1127 	uint16_t		cmd_op_mod;
1128 	uint8_t			cmd_reserved1[8];
1129 } __packed __aligned(4);
1130 
1131 struct mcx_cmd_alloc_uar_out {
1132 	uint8_t			cmd_status;
1133 	uint8_t			cmd_reserved0[3];
1134 	uint32_t		cmd_syndrome;
1135 	uint32_t		cmd_uar;
1136 	uint8_t			cmd_reserved1[4];
1137 } __packed __aligned(4);
1138 
1139 struct mcx_cmd_query_special_ctx_in {
1140 	uint16_t		cmd_opcode;
1141 	uint8_t			cmd_reserved0[4];
1142 	uint16_t		cmd_op_mod;
1143 	uint8_t			cmd_reserved1[8];
1144 } __packed __aligned(4);
1145 
1146 struct mcx_cmd_query_special_ctx_out {
1147 	uint8_t			cmd_status;
1148 	uint8_t			cmd_reserved0[3];
1149 	uint32_t		cmd_syndrome;
1150 	uint8_t			cmd_reserved1[4];
1151 	uint32_t		cmd_resd_lkey;
1152 } __packed __aligned(4);
1153 
1154 struct mcx_eq_ctx {
1155 	uint32_t		eq_status;
1156 #define MCX_EQ_CTX_STATE_SHIFT		8
1157 #define MCX_EQ_CTX_STATE_MASK		(0xf << MCX_EQ_CTX_STATE_SHIFT)
1158 #define MCX_EQ_CTX_STATE_ARMED		0x9
1159 #define MCX_EQ_CTX_STATE_FIRED		0xa
1160 #define MCX_EQ_CTX_OI_SHIFT		17
1161 #define MCX_EQ_CTX_OI			(1 << MCX_EQ_CTX_OI_SHIFT)
1162 #define MCX_EQ_CTX_EC_SHIFT		18
1163 #define MCX_EQ_CTX_EC			(1 << MCX_EQ_CTX_EC_SHIFT)
1164 #define MCX_EQ_CTX_STATUS_SHIFT		28
1165 #define MCX_EQ_CTX_STATUS_MASK		(0xf << MCX_EQ_CTX_STATUS_SHIFT)
1166 #define MCX_EQ_CTX_STATUS_OK		0x0
1167 #define MCX_EQ_CTX_STATUS_EQ_WRITE_FAILURE 0xa
1168 	uint32_t		eq_reserved1;
1169 	uint32_t		eq_page_offset;
1170 #define MCX_EQ_CTX_PAGE_OFFSET_SHIFT	5
1171 	uint32_t		eq_uar_size;
1172 #define MCX_EQ_CTX_UAR_PAGE_MASK	0xffffff
1173 #define MCX_EQ_CTX_LOG_EQ_SIZE_SHIFT	24
1174 	uint32_t		eq_reserved2;
1175 	uint8_t			eq_reserved3[3];
1176 	uint8_t			eq_intr;
1177 	uint32_t		eq_log_page_size;
1178 #define MCX_EQ_CTX_LOG_PAGE_SIZE_SHIFT	24
1179 	uint32_t		eq_reserved4[3];
1180 	uint32_t		eq_consumer_counter;
1181 	uint32_t		eq_producer_counter;
1182 #define MCX_EQ_CTX_COUNTER_MASK		0xffffff
1183 	uint32_t		eq_reserved5[4];
1184 } __packed __aligned(4);
1185 
1186 CTASSERT(sizeof(struct mcx_eq_ctx) == 64);
1187 
1188 struct mcx_cmd_create_eq_in {
1189 	uint16_t		cmd_opcode;
1190 	uint8_t			cmd_reserved0[4];
1191 	uint16_t		cmd_op_mod;
1192 	uint8_t			cmd_reserved1[8];
1193 } __packed __aligned(4);
1194 
1195 struct mcx_cmd_create_eq_mb_in {
1196 	struct mcx_eq_ctx	cmd_eq_ctx;
1197 	uint8_t			cmd_reserved0[8];
1198 	uint64_t		cmd_event_bitmask;
1199 #define MCX_EVENT_TYPE_COMPLETION	0x00
1200 #define MCX_EVENT_TYPE_CQ_ERROR		0x04
1201 #define MCX_EVENT_TYPE_INTERNAL_ERROR	0x08
1202 #define MCX_EVENT_TYPE_PORT_CHANGE	0x09
1203 #define MCX_EVENT_TYPE_CMD_COMPLETION	0x0a
1204 #define MCX_EVENT_TYPE_PAGE_REQUEST	0x0b
1205 #define MCX_EVENT_TYPE_LAST_WQE		0x13
1206 	uint8_t			cmd_reserved1[176];
1207 } __packed __aligned(4);
1208 
1209 struct mcx_cmd_create_eq_out {
1210 	uint8_t			cmd_status;
1211 	uint8_t			cmd_reserved0[3];
1212 	uint32_t		cmd_syndrome;
1213 	uint32_t		cmd_eqn;
1214 	uint8_t			cmd_reserved1[4];
1215 } __packed __aligned(4);
1216 
1217 struct mcx_cmd_query_eq_in {
1218 	uint16_t		cmd_opcode;
1219 	uint8_t			cmd_reserved0[4];
1220 	uint16_t		cmd_op_mod;
1221 	uint32_t		cmd_eqn;
1222 	uint8_t			cmd_reserved1[4];
1223 } __packed __aligned(4);
1224 
1225 struct mcx_cmd_query_eq_out {
1226 	uint8_t			cmd_status;
1227 	uint8_t			cmd_reserved0[3];
1228 	uint32_t		cmd_syndrome;
1229 	uint8_t			cmd_reserved1[8];
1230 } __packed __aligned(4);
1231 
1232 struct mcx_eq_entry {
1233 	uint8_t			eq_reserved1;
1234 	uint8_t			eq_event_type;
1235 	uint8_t			eq_reserved2;
1236 	uint8_t			eq_event_sub_type;
1237 
1238 	uint8_t			eq_reserved3[28];
1239 	uint32_t		eq_event_data[7];
1240 	uint8_t			eq_reserved4[2];
1241 	uint8_t			eq_signature;
1242 	uint8_t			eq_owner;
1243 #define MCX_EQ_ENTRY_OWNER_INIT			1
1244 } __packed __aligned(4);
1245 
1246 CTASSERT(sizeof(struct mcx_eq_entry) == 64);
1247 
1248 struct mcx_cmd_alloc_pd_in {
1249 	uint16_t		cmd_opcode;
1250 	uint8_t			cmd_reserved0[4];
1251 	uint16_t		cmd_op_mod;
1252 	uint8_t			cmd_reserved1[8];
1253 } __packed __aligned(4);
1254 
1255 struct mcx_cmd_alloc_pd_out {
1256 	uint8_t			cmd_status;
1257 	uint8_t			cmd_reserved0[3];
1258 	uint32_t		cmd_syndrome;
1259 	uint32_t		cmd_pd;
1260 	uint8_t			cmd_reserved1[4];
1261 } __packed __aligned(4);
1262 
1263 struct mcx_cmd_alloc_td_in {
1264 	uint16_t		cmd_opcode;
1265 	uint8_t			cmd_reserved0[4];
1266 	uint16_t		cmd_op_mod;
1267 	uint8_t			cmd_reserved1[8];
1268 } __packed __aligned(4);
1269 
1270 struct mcx_cmd_alloc_td_out {
1271 	uint8_t			cmd_status;
1272 	uint8_t			cmd_reserved0[3];
1273 	uint32_t		cmd_syndrome;
1274 	uint32_t		cmd_tdomain;
1275 	uint8_t			cmd_reserved1[4];
1276 } __packed __aligned(4);
1277 
1278 struct mcx_cmd_create_tir_in {
1279 	uint16_t		cmd_opcode;
1280 	uint8_t			cmd_reserved0[4];
1281 	uint16_t		cmd_op_mod;
1282 	uint8_t			cmd_reserved1[8];
1283 } __packed __aligned(4);
1284 
1285 struct mcx_cmd_create_tir_mb_in {
1286 	uint8_t			cmd_reserved0[20];
1287 	uint32_t		cmd_disp_type;
1288 #define MCX_TIR_CTX_DISP_TYPE_DIRECT	0
1289 #define MCX_TIR_CTX_DISP_TYPE_INDIRECT	1
1290 #define MCX_TIR_CTX_DISP_TYPE_SHIFT	28
1291 	uint8_t			cmd_reserved1[8];
1292 	uint32_t		cmd_lro;
1293 	uint8_t			cmd_reserved2[8];
1294 	uint32_t		cmd_inline_rqn;
1295 	uint32_t		cmd_indir_table;
1296 	uint32_t		cmd_tdomain;
1297 #define MCX_TIR_CTX_HASH_TOEPLITZ	2
1298 #define MCX_TIR_CTX_HASH_SHIFT		28
1299 	uint8_t			cmd_rx_hash_key[40];
1300 	uint32_t		cmd_rx_hash_sel_outer;
1301 #define MCX_TIR_CTX_HASH_SEL_SRC_IP	(1 << 0)
1302 #define MCX_TIR_CTX_HASH_SEL_DST_IP	(1 << 1)
1303 #define MCX_TIR_CTX_HASH_SEL_SPORT	(1 << 2)
1304 #define MCX_TIR_CTX_HASH_SEL_DPORT	(1 << 3)
1305 #define MCX_TIR_CTX_HASH_SEL_IPV4	(0 << 31)
1306 #define MCX_TIR_CTX_HASH_SEL_IPV6	(1 << 31)
1307 #define MCX_TIR_CTX_HASH_SEL_TCP	(0 << 30)
1308 #define MCX_TIR_CTX_HASH_SEL_UDP	(1 << 30)
1309 	uint32_t		cmd_rx_hash_sel_inner;
1310 	uint8_t			cmd_reserved3[152];
1311 } __packed __aligned(4);
1312 
1313 struct mcx_cmd_create_tir_out {
1314 	uint8_t			cmd_status;
1315 	uint8_t			cmd_reserved0[3];
1316 	uint32_t		cmd_syndrome;
1317 	uint32_t		cmd_tirn;
1318 	uint8_t			cmd_reserved1[4];
1319 } __packed __aligned(4);
1320 
1321 struct mcx_cmd_destroy_tir_in {
1322 	uint16_t		cmd_opcode;
1323 	uint8_t			cmd_reserved0[4];
1324 	uint16_t		cmd_op_mod;
1325 	uint32_t		cmd_tirn;
1326 	uint8_t			cmd_reserved1[4];
1327 } __packed __aligned(4);
1328 
1329 struct mcx_cmd_destroy_tir_out {
1330 	uint8_t			cmd_status;
1331 	uint8_t			cmd_reserved0[3];
1332 	uint32_t		cmd_syndrome;
1333 	uint8_t			cmd_reserved1[8];
1334 } __packed __aligned(4);
1335 
1336 struct mcx_cmd_create_tis_in {
1337 	uint16_t		cmd_opcode;
1338 	uint8_t			cmd_reserved0[4];
1339 	uint16_t		cmd_op_mod;
1340 	uint8_t			cmd_reserved1[8];
1341 } __packed __aligned(4);
1342 
1343 struct mcx_cmd_create_tis_mb_in {
1344 	uint8_t			cmd_reserved[16];
1345 	uint32_t		cmd_prio;
1346 	uint8_t			cmd_reserved1[32];
1347 	uint32_t		cmd_tdomain;
1348 	uint8_t			cmd_reserved2[120];
1349 } __packed __aligned(4);
1350 
1351 struct mcx_cmd_create_tis_out {
1352 	uint8_t			cmd_status;
1353 	uint8_t			cmd_reserved0[3];
1354 	uint32_t		cmd_syndrome;
1355 	uint32_t		cmd_tisn;
1356 	uint8_t			cmd_reserved1[4];
1357 } __packed __aligned(4);
1358 
1359 struct mcx_cmd_destroy_tis_in {
1360 	uint16_t		cmd_opcode;
1361 	uint8_t			cmd_reserved0[4];
1362 	uint16_t		cmd_op_mod;
1363 	uint32_t		cmd_tisn;
1364 	uint8_t			cmd_reserved1[4];
1365 } __packed __aligned(4);
1366 
1367 struct mcx_cmd_destroy_tis_out {
1368 	uint8_t			cmd_status;
1369 	uint8_t			cmd_reserved0[3];
1370 	uint32_t		cmd_syndrome;
1371 	uint8_t			cmd_reserved1[8];
1372 } __packed __aligned(4);
1373 
1374 struct mcx_cmd_create_rqt_in {
1375 	uint16_t		cmd_opcode;
1376 	uint8_t			cmd_reserved0[4];
1377 	uint16_t		cmd_op_mod;
1378 	uint8_t			cmd_reserved1[8];
1379 } __packed __aligned(4);
1380 
1381 struct mcx_rqt_ctx {
1382 	uint8_t			cmd_reserved0[20];
1383 	uint16_t		cmd_reserved1;
1384 	uint16_t		cmd_rqt_max_size;
1385 	uint16_t		cmd_reserved2;
1386 	uint16_t		cmd_rqt_actual_size;
1387 	uint8_t			cmd_reserved3[212];
1388 } __packed __aligned(4);
1389 
1390 struct mcx_cmd_create_rqt_mb_in {
1391 	uint8_t			cmd_reserved0[16];
1392 	struct mcx_rqt_ctx	cmd_rqt;
1393 } __packed __aligned(4);
1394 
1395 struct mcx_cmd_create_rqt_out {
1396 	uint8_t			cmd_status;
1397 	uint8_t			cmd_reserved0[3];
1398 	uint32_t		cmd_syndrome;
1399 	uint32_t		cmd_rqtn;
1400 	uint8_t			cmd_reserved1[4];
1401 } __packed __aligned(4);
1402 
1403 struct mcx_cmd_destroy_rqt_in {
1404 	uint16_t		cmd_opcode;
1405 	uint8_t			cmd_reserved0[4];
1406 	uint16_t		cmd_op_mod;
1407 	uint32_t		cmd_rqtn;
1408 	uint8_t			cmd_reserved1[4];
1409 } __packed __aligned(4);
1410 
1411 struct mcx_cmd_destroy_rqt_out {
1412 	uint8_t			cmd_status;
1413 	uint8_t			cmd_reserved0[3];
1414 	uint32_t		cmd_syndrome;
1415 	uint8_t			cmd_reserved1[8];
1416 } __packed __aligned(4);
1417 
1418 struct mcx_cq_ctx {
1419 	uint32_t		cq_status;
1420 #define MCX_CQ_CTX_STATUS_SHIFT		28
1421 #define MCX_CQ_CTX_STATUS_MASK		(0xf << MCX_CQ_CTX_STATUS_SHIFT)
1422 #define MCX_CQ_CTX_STATUS_OK		0x0
1423 #define MCX_CQ_CTX_STATUS_OVERFLOW	0x9
1424 #define MCX_CQ_CTX_STATUS_WRITE_FAIL	0xa
1425 #define MCX_CQ_CTX_STATE_SHIFT		8
1426 #define MCX_CQ_CTX_STATE_MASK		(0xf << MCX_CQ_CTX_STATE_SHIFT)
1427 #define MCX_CQ_CTX_STATE_SOLICITED	0x6
1428 #define MCX_CQ_CTX_STATE_ARMED		0x9
1429 #define MCX_CQ_CTX_STATE_FIRED		0xa
1430 	uint32_t		cq_reserved1;
1431 	uint32_t		cq_page_offset;
1432 	uint32_t		cq_uar_size;
1433 #define MCX_CQ_CTX_UAR_PAGE_MASK	0xffffff
1434 #define MCX_CQ_CTX_LOG_CQ_SIZE_SHIFT	24
1435 	uint32_t		cq_period_max_count;
1436 #define MCX_CQ_CTX_PERIOD_SHIFT		16
1437 	uint32_t		cq_eqn;
1438 	uint32_t		cq_log_page_size;
1439 #define MCX_CQ_CTX_LOG_PAGE_SIZE_SHIFT	24
1440 	uint32_t		cq_reserved2;
1441 	uint32_t		cq_last_notified;
1442 	uint32_t		cq_last_solicit;
1443 	uint32_t		cq_consumer_counter;
1444 	uint32_t		cq_producer_counter;
1445 	uint8_t			cq_reserved3[8];
1446 	uint64_t		cq_doorbell;
1447 } __packed __aligned(4);
1448 
1449 CTASSERT(sizeof(struct mcx_cq_ctx) == 64);
1450 
1451 struct mcx_cmd_create_cq_in {
1452 	uint16_t		cmd_opcode;
1453 	uint8_t			cmd_reserved0[4];
1454 	uint16_t		cmd_op_mod;
1455 	uint8_t			cmd_reserved1[8];
1456 } __packed __aligned(4);
1457 
1458 struct mcx_cmd_create_cq_mb_in {
1459 	struct mcx_cq_ctx	cmd_cq_ctx;
1460 	uint8_t			cmd_reserved1[192];
1461 } __packed __aligned(4);
1462 
1463 struct mcx_cmd_create_cq_out {
1464 	uint8_t			cmd_status;
1465 	uint8_t			cmd_reserved0[3];
1466 	uint32_t		cmd_syndrome;
1467 	uint32_t		cmd_cqn;
1468 	uint8_t			cmd_reserved1[4];
1469 } __packed __aligned(4);
1470 
1471 struct mcx_cmd_destroy_cq_in {
1472 	uint16_t		cmd_opcode;
1473 	uint8_t			cmd_reserved0[4];
1474 	uint16_t		cmd_op_mod;
1475 	uint32_t		cmd_cqn;
1476 	uint8_t			cmd_reserved1[4];
1477 } __packed __aligned(4);
1478 
1479 struct mcx_cmd_destroy_cq_out {
1480 	uint8_t			cmd_status;
1481 	uint8_t			cmd_reserved0[3];
1482 	uint32_t		cmd_syndrome;
1483 	uint8_t			cmd_reserved1[8];
1484 } __packed __aligned(4);
1485 
1486 struct mcx_cmd_query_cq_in {
1487 	uint16_t		cmd_opcode;
1488 	uint8_t			cmd_reserved0[4];
1489 	uint16_t		cmd_op_mod;
1490 	uint32_t		cmd_cqn;
1491 	uint8_t			cmd_reserved1[4];
1492 } __packed __aligned(4);
1493 
1494 struct mcx_cmd_query_cq_out {
1495 	uint8_t			cmd_status;
1496 	uint8_t			cmd_reserved0[3];
1497 	uint32_t		cmd_syndrome;
1498 	uint8_t			cmd_reserved1[8];
1499 } __packed __aligned(4);
1500 
1501 struct mcx_cq_entry {
1502 	uint32_t		__reserved__;
1503 	uint32_t		cq_lro;
1504 	uint32_t		cq_lro_ack_seq_num;
1505 	uint32_t		cq_rx_hash;
1506 	uint8_t			cq_rx_hash_type;
1507 	uint8_t			cq_ml_path;
1508 	uint16_t		__reserved__;
1509 	uint32_t		cq_checksum;
1510 	uint32_t		__reserved__;
1511 	uint32_t		cq_flags;
1512 #define MCX_CQ_ENTRY_FLAGS_L4_OK		(1 << 26)
1513 #define MCX_CQ_ENTRY_FLAGS_L3_OK		(1 << 25)
1514 #define MCX_CQ_ENTRY_FLAGS_L2_OK		(1 << 24)
1515 #define MCX_CQ_ENTRY_FLAGS_CV			(1 << 16)
1516 #define MCX_CQ_ENTRY_FLAGS_VLAN_MASK		(0xffff)
1517 
1518 	uint32_t		cq_lro_srqn;
1519 	uint32_t		__reserved__[2];
1520 	uint32_t		cq_byte_cnt;
1521 	uint64_t		cq_timestamp;
1522 	uint8_t			cq_rx_drops;
1523 	uint8_t			cq_flow_tag[3];
1524 	uint16_t		cq_wqe_count;
1525 	uint8_t			cq_signature;
1526 	uint8_t			cq_opcode_owner;
1527 #define MCX_CQ_ENTRY_FLAG_OWNER			(1 << 0)
1528 #define MCX_CQ_ENTRY_FLAG_SE			(1 << 1)
1529 #define MCX_CQ_ENTRY_FORMAT_SHIFT		2
1530 #define MCX_CQ_ENTRY_OPCODE_SHIFT		4
1531 
1532 #define MCX_CQ_ENTRY_FORMAT_NO_INLINE		0
1533 #define MCX_CQ_ENTRY_FORMAT_INLINE_32		1
1534 #define MCX_CQ_ENTRY_FORMAT_INLINE_64		2
1535 #define MCX_CQ_ENTRY_FORMAT_COMPRESSED		3
1536 
1537 #define MCX_CQ_ENTRY_OPCODE_REQ			0
1538 #define MCX_CQ_ENTRY_OPCODE_SEND		2
1539 #define MCX_CQ_ENTRY_OPCODE_REQ_ERR		13
1540 #define MCX_CQ_ENTRY_OPCODE_SEND_ERR		14
1541 #define MCX_CQ_ENTRY_OPCODE_INVALID		15
1542 
1543 } __packed __aligned(4);
1544 
1545 CTASSERT(sizeof(struct mcx_cq_entry) == 64);
1546 
1547 struct mcx_cq_doorbell {
1548 	uint32_t		 db_update_ci;
1549 	uint32_t		 db_arm_ci;
1550 #define MCX_CQ_DOORBELL_ARM_CMD_SN_SHIFT	28
1551 #define MCX_CQ_DOORBELL_ARM_CMD			(1 << 24)
1552 #define MCX_CQ_DOORBELL_ARM_CI_MASK		(0xffffff)
1553 } __packed __aligned(8);
1554 
1555 struct mcx_wq_ctx {
1556 	uint8_t			 wq_type;
1557 #define MCX_WQ_CTX_TYPE_CYCLIC			(1 << 4)
1558 #define MCX_WQ_CTX_TYPE_SIGNATURE		(1 << 3)
1559 	uint8_t			 wq_reserved0[5];
1560 	uint16_t		 wq_lwm;
1561 	uint32_t		 wq_pd;
1562 	uint32_t		 wq_uar_page;
1563 	uint64_t		 wq_doorbell;
1564 	uint32_t		 wq_hw_counter;
1565 	uint32_t		 wq_sw_counter;
1566 	uint16_t		 wq_log_stride;
1567 	uint8_t			 wq_log_page_sz;
1568 	uint8_t			 wq_log_size;
1569 	uint8_t			 wq_reserved1[156];
1570 } __packed __aligned(4);
1571 
1572 CTASSERT(sizeof(struct mcx_wq_ctx) == 0xC0);
1573 
1574 struct mcx_sq_ctx {
1575 	uint32_t		sq_flags;
1576 #define MCX_SQ_CTX_RLKEY			(1 << 31)
1577 #define MCX_SQ_CTX_FRE_SHIFT			(1 << 29)
1578 #define MCX_SQ_CTX_FLUSH_IN_ERROR		(1 << 28)
1579 #define MCX_SQ_CTX_MIN_WQE_INLINE_SHIFT		24
1580 #define MCX_SQ_CTX_STATE_SHIFT			20
1581 #define MCX_SQ_CTX_STATE_MASK			(0xf << 20)
1582 #define MCX_SQ_CTX_STATE_RST			0
1583 #define MCX_SQ_CTX_STATE_RDY			1
1584 #define MCX_SQ_CTX_STATE_ERR			3
1585 	uint32_t		sq_user_index;
1586 	uint32_t		sq_cqn;
1587 	uint32_t		sq_reserved1[5];
1588 	uint32_t		sq_tis_lst_sz;
1589 #define MCX_SQ_CTX_TIS_LST_SZ_SHIFT		16
1590 	uint32_t		sq_reserved2[2];
1591 	uint32_t		sq_tis_num;
1592 	struct mcx_wq_ctx	sq_wq;
1593 } __packed __aligned(4);
1594 
1595 struct mcx_sq_entry_seg {
1596 	uint32_t		sqs_byte_count;
1597 	uint32_t		sqs_lkey;
1598 	uint64_t		sqs_addr;
1599 } __packed __aligned(4);
1600 
1601 struct mcx_sq_entry {
1602 	/* control segment */
1603 	uint32_t		sqe_opcode_index;
1604 #define MCX_SQE_WQE_INDEX_SHIFT			8
1605 #define MCX_SQE_WQE_OPCODE_NOP			0x00
1606 #define MCX_SQE_WQE_OPCODE_SEND			0x0a
1607 	uint32_t		sqe_ds_sq_num;
1608 #define MCX_SQE_SQ_NUM_SHIFT			8
1609 	uint32_t		sqe_signature;
1610 #define MCX_SQE_SIGNATURE_SHIFT			24
1611 #define MCX_SQE_SOLICITED_EVENT			0x02
1612 #define MCX_SQE_CE_CQE_ON_ERR			0x00
1613 #define MCX_SQE_CE_CQE_FIRST_ERR		0x04
1614 #define MCX_SQE_CE_CQE_ALWAYS			0x08
1615 #define MCX_SQE_CE_CQE_SOLICIT			0x0C
1616 #define MCX_SQE_FM_NO_FENCE			0x00
1617 #define MCX_SQE_FM_SMALL_FENCE			0x40
1618 	uint32_t		sqe_mkey;
1619 
1620 	/* ethernet segment */
1621 	uint32_t		sqe_reserved1;
1622 	uint32_t		sqe_mss_csum;
1623 #define MCX_SQE_L4_CSUM				(1 << 31)
1624 #define MCX_SQE_L3_CSUM				(1 << 30)
1625 	uint32_t		sqe_reserved2;
1626 	uint16_t		sqe_inline_header_size;
1627 	uint16_t		sqe_inline_headers[9];
1628 
1629 	/* data segment */
1630 	struct mcx_sq_entry_seg sqe_segs[1];
1631 } __packed __aligned(64);
1632 
1633 CTASSERT(sizeof(struct mcx_sq_entry) == 64);
1634 
1635 struct mcx_cmd_create_sq_in {
1636 	uint16_t		cmd_opcode;
1637 	uint8_t			cmd_reserved0[4];
1638 	uint16_t		cmd_op_mod;
1639 	uint8_t			cmd_reserved1[8];
1640 } __packed __aligned(4);
1641 
1642 struct mcx_cmd_create_sq_out {
1643 	uint8_t			cmd_status;
1644 	uint8_t			cmd_reserved0[3];
1645 	uint32_t		cmd_syndrome;
1646 	uint32_t		cmd_sqn;
1647 	uint8_t			cmd_reserved1[4];
1648 } __packed __aligned(4);
1649 
1650 struct mcx_cmd_modify_sq_in {
1651 	uint16_t		cmd_opcode;
1652 	uint8_t			cmd_reserved0[4];
1653 	uint16_t		cmd_op_mod;
1654 	uint32_t		cmd_sq_state;
1655 	uint8_t			cmd_reserved1[4];
1656 } __packed __aligned(4);
1657 
1658 struct mcx_cmd_modify_sq_mb_in {
1659 	uint32_t		cmd_modify_hi;
1660 	uint32_t		cmd_modify_lo;
1661 	uint8_t			cmd_reserved0[8];
1662 	struct mcx_sq_ctx	cmd_sq_ctx;
1663 } __packed __aligned(4);
1664 
1665 struct mcx_cmd_modify_sq_out {
1666 	uint8_t			cmd_status;
1667 	uint8_t			cmd_reserved0[3];
1668 	uint32_t		cmd_syndrome;
1669 	uint8_t			cmd_reserved1[8];
1670 } __packed __aligned(4);
1671 
1672 struct mcx_cmd_destroy_sq_in {
1673 	uint16_t		cmd_opcode;
1674 	uint8_t			cmd_reserved0[4];
1675 	uint16_t		cmd_op_mod;
1676 	uint32_t		cmd_sqn;
1677 	uint8_t			cmd_reserved1[4];
1678 } __packed __aligned(4);
1679 
1680 struct mcx_cmd_destroy_sq_out {
1681 	uint8_t			cmd_status;
1682 	uint8_t			cmd_reserved0[3];
1683 	uint32_t		cmd_syndrome;
1684 	uint8_t			cmd_reserved1[8];
1685 } __packed __aligned(4);
1686 
1687 
1688 struct mcx_rq_ctx {
1689 	uint32_t		rq_flags;
1690 #define MCX_RQ_CTX_RLKEY			(1 << 31)
1691 #define MCX_RQ_CTX_VLAN_STRIP_DIS		(1 << 28)
1692 #define MCX_RQ_CTX_MEM_RQ_TYPE_SHIFT		24
1693 #define MCX_RQ_CTX_STATE_SHIFT			20
1694 #define MCX_RQ_CTX_STATE_MASK			(0xf << 20)
1695 #define MCX_RQ_CTX_STATE_RST			0
1696 #define MCX_RQ_CTX_STATE_RDY			1
1697 #define MCX_RQ_CTX_STATE_ERR			3
1698 #define MCX_RQ_CTX_FLUSH_IN_ERROR		(1 << 18)
1699 	uint32_t		rq_user_index;
1700 	uint32_t		rq_cqn;
1701 	uint32_t		rq_reserved1;
1702 	uint32_t		rq_rmpn;
1703 	uint32_t		rq_reserved2[7];
1704 	struct mcx_wq_ctx	rq_wq;
1705 } __packed __aligned(4);
1706 
1707 struct mcx_rq_entry {
1708 	uint32_t		rqe_byte_count;
1709 	uint32_t		rqe_lkey;
1710 	uint64_t		rqe_addr;
1711 } __packed __aligned(16);
1712 
1713 struct mcx_cmd_create_rq_in {
1714 	uint16_t		cmd_opcode;
1715 	uint8_t			cmd_reserved0[4];
1716 	uint16_t		cmd_op_mod;
1717 	uint8_t			cmd_reserved1[8];
1718 } __packed __aligned(4);
1719 
1720 struct mcx_cmd_create_rq_out {
1721 	uint8_t			cmd_status;
1722 	uint8_t			cmd_reserved0[3];
1723 	uint32_t		cmd_syndrome;
1724 	uint32_t		cmd_rqn;
1725 	uint8_t			cmd_reserved1[4];
1726 } __packed __aligned(4);
1727 
1728 struct mcx_cmd_modify_rq_in {
1729 	uint16_t		cmd_opcode;
1730 	uint8_t			cmd_reserved0[4];
1731 	uint16_t		cmd_op_mod;
1732 	uint32_t		cmd_rq_state;
1733 	uint8_t			cmd_reserved1[4];
1734 } __packed __aligned(4);
1735 
1736 struct mcx_cmd_modify_rq_mb_in {
1737 	uint32_t		cmd_modify_hi;
1738 	uint32_t		cmd_modify_lo;
1739 	uint8_t			cmd_reserved0[8];
1740 	struct mcx_rq_ctx	cmd_rq_ctx;
1741 } __packed __aligned(4);
1742 
1743 struct mcx_cmd_modify_rq_out {
1744 	uint8_t			cmd_status;
1745 	uint8_t			cmd_reserved0[3];
1746 	uint32_t		cmd_syndrome;
1747 	uint8_t			cmd_reserved1[8];
1748 } __packed __aligned(4);
1749 
1750 struct mcx_cmd_destroy_rq_in {
1751 	uint16_t		cmd_opcode;
1752 	uint8_t			cmd_reserved0[4];
1753 	uint16_t		cmd_op_mod;
1754 	uint32_t		cmd_rqn;
1755 	uint8_t			cmd_reserved1[4];
1756 } __packed __aligned(4);
1757 
1758 struct mcx_cmd_destroy_rq_out {
1759 	uint8_t			cmd_status;
1760 	uint8_t			cmd_reserved0[3];
1761 	uint32_t		cmd_syndrome;
1762 	uint8_t			cmd_reserved1[8];
1763 } __packed __aligned(4);
1764 
1765 struct mcx_cmd_create_flow_table_in {
1766 	uint16_t		cmd_opcode;
1767 	uint8_t			cmd_reserved0[4];
1768 	uint16_t		cmd_op_mod;
1769 	uint8_t			cmd_reserved1[8];
1770 } __packed __aligned(4);
1771 
1772 struct mcx_flow_table_ctx {
1773 	uint8_t			ft_miss_action;
1774 	uint8_t			ft_level;
1775 	uint8_t			ft_reserved0;
1776 	uint8_t			ft_log_size;
1777 	uint32_t		ft_table_miss_id;
1778 	uint8_t			ft_reserved1[28];
1779 } __packed __aligned(4);
1780 
1781 struct mcx_cmd_create_flow_table_mb_in {
1782 	uint8_t			cmd_table_type;
1783 	uint8_t			cmd_reserved0[7];
1784 	struct mcx_flow_table_ctx cmd_ctx;
1785 } __packed __aligned(4);
1786 
1787 struct mcx_cmd_create_flow_table_out {
1788 	uint8_t			cmd_status;
1789 	uint8_t			cmd_reserved0[3];
1790 	uint32_t		cmd_syndrome;
1791 	uint32_t		cmd_table_id;
1792 	uint8_t			cmd_reserved1[4];
1793 } __packed __aligned(4);
1794 
1795 struct mcx_cmd_destroy_flow_table_in {
1796 	uint16_t		cmd_opcode;
1797 	uint8_t			cmd_reserved0[4];
1798 	uint16_t		cmd_op_mod;
1799 	uint8_t			cmd_reserved1[8];
1800 } __packed __aligned(4);
1801 
1802 struct mcx_cmd_destroy_flow_table_mb_in {
1803 	uint8_t			cmd_table_type;
1804 	uint8_t			cmd_reserved0[3];
1805 	uint32_t		cmd_table_id;
1806 	uint8_t			cmd_reserved1[40];
1807 } __packed __aligned(4);
1808 
1809 struct mcx_cmd_destroy_flow_table_out {
1810 	uint8_t			cmd_status;
1811 	uint8_t			cmd_reserved0[3];
1812 	uint32_t		cmd_syndrome;
1813 	uint8_t			cmd_reserved1[8];
1814 } __packed __aligned(4);
1815 
1816 struct mcx_cmd_set_flow_table_root_in {
1817 	uint16_t		cmd_opcode;
1818 	uint8_t			cmd_reserved0[4];
1819 	uint16_t		cmd_op_mod;
1820 	uint8_t			cmd_reserved1[8];
1821 } __packed __aligned(4);
1822 
1823 struct mcx_cmd_set_flow_table_root_mb_in {
1824 	uint8_t			cmd_table_type;
1825 	uint8_t			cmd_reserved0[3];
1826 	uint32_t		cmd_table_id;
1827 	uint8_t			cmd_reserved1[56];
1828 } __packed __aligned(4);
1829 
1830 struct mcx_cmd_set_flow_table_root_out {
1831 	uint8_t			cmd_status;
1832 	uint8_t			cmd_reserved0[3];
1833 	uint32_t		cmd_syndrome;
1834 	uint8_t			cmd_reserved1[8];
1835 } __packed __aligned(4);
1836 
1837 struct mcx_flow_match {
1838 	/* outer headers */
1839 	uint8_t			mc_src_mac[6];
1840 	uint16_t		mc_ethertype;
1841 	uint8_t			mc_dest_mac[6];
1842 	uint16_t		mc_first_vlan;
1843 	uint8_t			mc_ip_proto;
1844 	uint8_t			mc_ip_dscp_ecn;
1845 	uint8_t			mc_vlan_flags;
1846 #define MCX_FLOW_MATCH_IP_FRAG	(1 << 5)
1847 	uint8_t			mc_tcp_flags;
1848 	uint16_t		mc_tcp_sport;
1849 	uint16_t		mc_tcp_dport;
1850 	uint32_t		mc_reserved0;
1851 	uint16_t		mc_udp_sport;
1852 	uint16_t		mc_udp_dport;
1853 	uint8_t			mc_src_ip[16];
1854 	uint8_t			mc_dest_ip[16];
1855 
1856 	/* misc parameters */
1857 	uint8_t			mc_reserved1[8];
1858 	uint16_t		mc_second_vlan;
1859 	uint8_t			mc_reserved2[2];
1860 	uint8_t			mc_second_vlan_flags;
1861 	uint8_t			mc_reserved3[15];
1862 	uint32_t		mc_outer_ipv6_flow_label;
1863 	uint8_t			mc_reserved4[32];
1864 
1865 	uint8_t			mc_reserved[384];
1866 } __packed __aligned(4);
1867 
1868 CTASSERT(sizeof(struct mcx_flow_match) == 512);
1869 
1870 struct mcx_cmd_create_flow_group_in {
1871 	uint16_t		cmd_opcode;
1872 	uint8_t			cmd_reserved0[4];
1873 	uint16_t		cmd_op_mod;
1874 	uint8_t			cmd_reserved1[8];
1875 } __packed __aligned(4);
1876 
1877 struct mcx_cmd_create_flow_group_mb_in {
1878 	uint8_t			cmd_table_type;
1879 	uint8_t			cmd_reserved0[3];
1880 	uint32_t		cmd_table_id;
1881 	uint8_t			cmd_reserved1[4];
1882 	uint32_t		cmd_start_flow_index;
1883 	uint8_t			cmd_reserved2[4];
1884 	uint32_t		cmd_end_flow_index;
1885 	uint8_t			cmd_reserved3[23];
1886 	uint8_t			cmd_match_criteria_enable;
1887 #define MCX_CREATE_FLOW_GROUP_CRIT_OUTER	(1 << 0)
1888 #define MCX_CREATE_FLOW_GROUP_CRIT_MISC		(1 << 1)
1889 #define MCX_CREATE_FLOW_GROUP_CRIT_INNER	(1 << 2)
1890 	struct mcx_flow_match	cmd_match_criteria;
1891 	uint8_t			cmd_reserved4[448];
1892 } __packed __aligned(4);
1893 
1894 struct mcx_cmd_create_flow_group_out {
1895 	uint8_t			cmd_status;
1896 	uint8_t			cmd_reserved0[3];
1897 	uint32_t		cmd_syndrome;
1898 	uint32_t		cmd_group_id;
1899 	uint8_t			cmd_reserved1[4];
1900 } __packed __aligned(4);
1901 
1902 struct mcx_flow_ctx {
1903 	uint8_t			fc_reserved0[4];
1904 	uint32_t		fc_group_id;
1905 	uint32_t		fc_flow_tag;
1906 	uint32_t		fc_action;
1907 #define MCX_FLOW_CONTEXT_ACTION_ALLOW		(1 << 0)
1908 #define MCX_FLOW_CONTEXT_ACTION_DROP		(1 << 1)
1909 #define MCX_FLOW_CONTEXT_ACTION_FORWARD		(1 << 2)
1910 #define MCX_FLOW_CONTEXT_ACTION_COUNT		(1 << 3)
1911 	uint32_t		fc_dest_list_size;
1912 	uint32_t		fc_counter_list_size;
1913 	uint8_t			fc_reserved1[40];
1914 	struct mcx_flow_match	fc_match_value;
1915 	uint8_t			fc_reserved2[192];
1916 } __packed __aligned(4);
1917 
1918 #define MCX_FLOW_CONTEXT_DEST_TYPE_TABLE	(1 << 24)
1919 #define MCX_FLOW_CONTEXT_DEST_TYPE_TIR		(2 << 24)
1920 
1921 struct mcx_cmd_destroy_flow_group_in {
1922 	uint16_t		cmd_opcode;
1923 	uint8_t			cmd_reserved0[4];
1924 	uint16_t		cmd_op_mod;
1925 	uint8_t			cmd_reserved1[8];
1926 } __packed __aligned(4);
1927 
1928 struct mcx_cmd_destroy_flow_group_mb_in {
1929 	uint8_t			cmd_table_type;
1930 	uint8_t			cmd_reserved0[3];
1931 	uint32_t		cmd_table_id;
1932 	uint32_t		cmd_group_id;
1933 	uint8_t			cmd_reserved1[36];
1934 } __packed __aligned(4);
1935 
1936 struct mcx_cmd_destroy_flow_group_out {
1937 	uint8_t			cmd_status;
1938 	uint8_t			cmd_reserved0[3];
1939 	uint32_t		cmd_syndrome;
1940 	uint8_t			cmd_reserved1[8];
1941 } __packed __aligned(4);
1942 
1943 struct mcx_cmd_set_flow_table_entry_in {
1944 	uint16_t		cmd_opcode;
1945 	uint8_t			cmd_reserved0[4];
1946 	uint16_t		cmd_op_mod;
1947 	uint8_t			cmd_reserved1[8];
1948 } __packed __aligned(4);
1949 
1950 struct mcx_cmd_set_flow_table_entry_mb_in {
1951 	uint8_t			cmd_table_type;
1952 	uint8_t			cmd_reserved0[3];
1953 	uint32_t		cmd_table_id;
1954 	uint32_t		cmd_modify_enable_mask;
1955 	uint8_t			cmd_reserved1[4];
1956 	uint32_t		cmd_flow_index;
1957 	uint8_t			cmd_reserved2[28];
1958 	struct mcx_flow_ctx	cmd_flow_ctx;
1959 } __packed __aligned(4);
1960 
1961 struct mcx_cmd_set_flow_table_entry_out {
1962 	uint8_t			cmd_status;
1963 	uint8_t			cmd_reserved0[3];
1964 	uint32_t		cmd_syndrome;
1965 	uint8_t			cmd_reserved1[8];
1966 } __packed __aligned(4);
1967 
1968 struct mcx_cmd_query_flow_table_entry_in {
1969 	uint16_t		cmd_opcode;
1970 	uint8_t			cmd_reserved0[4];
1971 	uint16_t		cmd_op_mod;
1972 	uint8_t			cmd_reserved1[8];
1973 } __packed __aligned(4);
1974 
1975 struct mcx_cmd_query_flow_table_entry_mb_in {
1976 	uint8_t			cmd_table_type;
1977 	uint8_t			cmd_reserved0[3];
1978 	uint32_t		cmd_table_id;
1979 	uint8_t			cmd_reserved1[8];
1980 	uint32_t		cmd_flow_index;
1981 	uint8_t			cmd_reserved2[28];
1982 } __packed __aligned(4);
1983 
1984 struct mcx_cmd_query_flow_table_entry_out {
1985 	uint8_t			cmd_status;
1986 	uint8_t			cmd_reserved0[3];
1987 	uint32_t		cmd_syndrome;
1988 	uint8_t			cmd_reserved1[8];
1989 } __packed __aligned(4);
1990 
1991 struct mcx_cmd_query_flow_table_entry_mb_out {
1992 	uint8_t			cmd_reserved0[48];
1993 	struct mcx_flow_ctx	cmd_flow_ctx;
1994 } __packed __aligned(4);
1995 
1996 struct mcx_cmd_delete_flow_table_entry_in {
1997 	uint16_t		cmd_opcode;
1998 	uint8_t			cmd_reserved0[4];
1999 	uint16_t		cmd_op_mod;
2000 	uint8_t			cmd_reserved1[8];
2001 } __packed __aligned(4);
2002 
2003 struct mcx_cmd_delete_flow_table_entry_mb_in {
2004 	uint8_t			cmd_table_type;
2005 	uint8_t			cmd_reserved0[3];
2006 	uint32_t		cmd_table_id;
2007 	uint8_t			cmd_reserved1[8];
2008 	uint32_t		cmd_flow_index;
2009 	uint8_t			cmd_reserved2[28];
2010 } __packed __aligned(4);
2011 
2012 struct mcx_cmd_delete_flow_table_entry_out {
2013 	uint8_t			cmd_status;
2014 	uint8_t			cmd_reserved0[3];
2015 	uint32_t		cmd_syndrome;
2016 	uint8_t			cmd_reserved1[8];
2017 } __packed __aligned(4);
2018 
2019 struct mcx_cmd_query_flow_group_in {
2020 	uint16_t		cmd_opcode;
2021 	uint8_t			cmd_reserved0[4];
2022 	uint16_t		cmd_op_mod;
2023 	uint8_t			cmd_reserved1[8];
2024 } __packed __aligned(4);
2025 
2026 struct mcx_cmd_query_flow_group_mb_in {
2027 	uint8_t			cmd_table_type;
2028 	uint8_t			cmd_reserved0[3];
2029 	uint32_t		cmd_table_id;
2030 	uint32_t		cmd_group_id;
2031 	uint8_t			cmd_reserved1[36];
2032 } __packed __aligned(4);
2033 
2034 struct mcx_cmd_query_flow_group_out {
2035 	uint8_t			cmd_status;
2036 	uint8_t			cmd_reserved0[3];
2037 	uint32_t		cmd_syndrome;
2038 	uint8_t			cmd_reserved1[8];
2039 } __packed __aligned(4);
2040 
2041 struct mcx_cmd_query_flow_group_mb_out {
2042 	uint8_t			cmd_reserved0[12];
2043 	uint32_t		cmd_start_flow_index;
2044 	uint8_t			cmd_reserved1[4];
2045 	uint32_t		cmd_end_flow_index;
2046 	uint8_t			cmd_reserved2[20];
2047 	uint32_t		cmd_match_criteria_enable;
2048 	uint8_t			cmd_match_criteria[512];
2049 	uint8_t			cmd_reserved4[448];
2050 } __packed __aligned(4);
2051 
2052 struct mcx_cmd_query_flow_table_in {
2053 	uint16_t		cmd_opcode;
2054 	uint8_t			cmd_reserved0[4];
2055 	uint16_t		cmd_op_mod;
2056 	uint8_t			cmd_reserved1[8];
2057 } __packed __aligned(4);
2058 
2059 struct mcx_cmd_query_flow_table_mb_in {
2060 	uint8_t			cmd_table_type;
2061 	uint8_t			cmd_reserved0[3];
2062 	uint32_t		cmd_table_id;
2063 	uint8_t			cmd_reserved1[40];
2064 } __packed __aligned(4);
2065 
2066 struct mcx_cmd_query_flow_table_out {
2067 	uint8_t			cmd_status;
2068 	uint8_t			cmd_reserved0[3];
2069 	uint32_t		cmd_syndrome;
2070 	uint8_t			cmd_reserved1[8];
2071 } __packed __aligned(4);
2072 
2073 struct mcx_cmd_query_flow_table_mb_out {
2074 	uint8_t			cmd_reserved0[4];
2075 	struct mcx_flow_table_ctx cmd_ctx;
2076 } __packed __aligned(4);
2077 
2078 struct mcx_cmd_alloc_flow_counter_in {
2079 	uint16_t		cmd_opcode;
2080 	uint8_t			cmd_reserved0[4];
2081 	uint16_t		cmd_op_mod;
2082 	uint8_t			cmd_reserved1[8];
2083 } __packed __aligned(4);
2084 
2085 struct mcx_cmd_query_rq_in {
2086 	uint16_t		cmd_opcode;
2087 	uint8_t			cmd_reserved0[4];
2088 	uint16_t		cmd_op_mod;
2089 	uint32_t		cmd_rqn;
2090 	uint8_t			cmd_reserved1[4];
2091 } __packed __aligned(4);
2092 
2093 struct mcx_cmd_query_rq_out {
2094 	uint8_t			cmd_status;
2095 	uint8_t			cmd_reserved0[3];
2096 	uint32_t		cmd_syndrome;
2097 	uint8_t			cmd_reserved1[8];
2098 } __packed __aligned(4);
2099 
2100 struct mcx_cmd_query_rq_mb_out {
2101 	uint8_t			cmd_reserved0[16];
2102 	struct mcx_rq_ctx	cmd_ctx;
2103 };
2104 
2105 struct mcx_cmd_query_sq_in {
2106 	uint16_t		cmd_opcode;
2107 	uint8_t			cmd_reserved0[4];
2108 	uint16_t		cmd_op_mod;
2109 	uint32_t		cmd_sqn;
2110 	uint8_t			cmd_reserved1[4];
2111 } __packed __aligned(4);
2112 
2113 struct mcx_cmd_query_sq_out {
2114 	uint8_t			cmd_status;
2115 	uint8_t			cmd_reserved0[3];
2116 	uint32_t		cmd_syndrome;
2117 	uint8_t			cmd_reserved1[8];
2118 } __packed __aligned(4);
2119 
2120 struct mcx_cmd_query_sq_mb_out {
2121 	uint8_t			cmd_reserved0[16];
2122 	struct mcx_sq_ctx	cmd_ctx;
2123 };
2124 
2125 struct mcx_cmd_alloc_flow_counter_out {
2126 	uint8_t			cmd_status;
2127 	uint8_t			cmd_reserved0[3];
2128 	uint32_t		cmd_syndrome;
2129 	uint8_t			cmd_reserved1[2];
2130 	uint16_t		cmd_flow_counter_id;
2131 	uint8_t			cmd_reserved2[4];
2132 } __packed __aligned(4);
2133 
2134 struct mcx_wq_doorbell {
2135 	uint32_t		 db_recv_counter;
2136 	uint32_t		 db_send_counter;
2137 } __packed __aligned(8);
2138 
2139 struct mcx_dmamem {
2140 	bus_dmamap_t		 mxm_map;
2141 	bus_dma_segment_t	 mxm_seg;
2142 	int			 mxm_nsegs;
2143 	size_t			 mxm_size;
2144 	caddr_t			 mxm_kva;
2145 };
2146 #define MCX_DMA_MAP(_mxm)	((_mxm)->mxm_map)
2147 #define MCX_DMA_DVA(_mxm)	((_mxm)->mxm_map->dm_segs[0].ds_addr)
2148 #define MCX_DMA_KVA(_mxm)	((void *)(_mxm)->mxm_kva)
2149 #define MCX_DMA_LEN(_mxm)	((_mxm)->mxm_size)
2150 
2151 struct mcx_hwmem {
2152 	bus_dmamap_t		 mhm_map;
2153 	bus_dma_segment_t	*mhm_segs;
2154 	unsigned int		 mhm_seg_count;
2155 	unsigned int		 mhm_npages;
2156 };
2157 
2158 struct mcx_slot {
2159 	bus_dmamap_t		 ms_map;
2160 	struct mbuf		*ms_m;
2161 };
2162 
2163 struct mcx_eq {
2164 	int			 eq_n;
2165 	uint32_t		 eq_cons;
2166 	struct mcx_dmamem	 eq_mem;
2167 };
2168 
2169 struct mcx_cq {
2170 	int			 cq_n;
2171 	struct mcx_dmamem	 cq_mem;
2172 	uint32_t		*cq_doorbell;
2173 	uint32_t		 cq_cons;
2174 	uint32_t		 cq_count;
2175 };
2176 
2177 struct mcx_calibration {
2178 	uint64_t		 c_timestamp;	/* previous mcx chip time */
2179 	uint64_t		 c_uptime;	/* previous kernel nanouptime */
2180 	uint64_t		 c_tbase;	/* mcx chip time */
2181 	uint64_t		 c_ubase;	/* kernel nanouptime */
2182 	uint64_t		 c_tdiff;
2183 	uint64_t		 c_udiff;
2184 };
2185 
2186 #define MCX_CALIBRATE_FIRST    2
2187 #define MCX_CALIBRATE_NORMAL   30
2188 
2189 struct mcx_rx {
2190 	struct mcx_softc	*rx_softc;
2191 	struct ifiqueue		*rx_ifiq;
2192 
2193 	int			 rx_rqn;
2194 	struct mcx_dmamem	 rx_rq_mem;
2195 	struct mcx_slot		*rx_slots;
2196 	uint32_t		*rx_doorbell;
2197 
2198 	uint32_t		 rx_prod;
2199 	struct timeout		 rx_refill;
2200 	struct if_rxring	 rx_rxr;
2201 } __aligned(64);
2202 
2203 struct mcx_tx {
2204 	struct mcx_softc	*tx_softc;
2205 	struct ifqueue		*tx_ifq;
2206 
2207 	int			 tx_uar;
2208 	int			 tx_sqn;
2209 	struct mcx_dmamem	 tx_sq_mem;
2210 	struct mcx_slot		*tx_slots;
2211 	uint32_t		*tx_doorbell;
2212 	int			 tx_bf_offset;
2213 
2214 	uint32_t		 tx_cons;
2215 	uint32_t		 tx_prod;
2216 } __aligned(64);
2217 
2218 struct mcx_queues {
2219 	char			 q_name[16];
2220 	void			*q_ihc;
2221 	struct mcx_softc	*q_sc;
2222 	int			 q_uar;
2223 	int			 q_index;
2224 	struct mcx_rx		 q_rx;
2225 	struct mcx_tx		 q_tx;
2226 	struct mcx_cq		 q_cq;
2227 	struct mcx_eq		 q_eq;
2228 #if NKSTAT > 0
2229 	struct kstat		*q_kstat;
2230 #endif
2231 };
2232 
2233 struct mcx_flow_group {
2234 	int			 g_id;
2235 	int			 g_table;
2236 	int			 g_start;
2237 	int			 g_size;
2238 };
2239 
2240 #define MCX_FLOW_GROUP_PROMISC	 0
2241 #define MCX_FLOW_GROUP_ALLMULTI	 1
2242 #define MCX_FLOW_GROUP_MAC	 2
2243 #define MCX_FLOW_GROUP_RSS_L4	 3
2244 #define MCX_FLOW_GROUP_RSS_L3	 4
2245 #define MCX_FLOW_GROUP_RSS_NONE	 5
2246 #define MCX_NUM_FLOW_GROUPS	 6
2247 
2248 #define MCX_HASH_SEL_L3		MCX_TIR_CTX_HASH_SEL_SRC_IP | \
2249 				MCX_TIR_CTX_HASH_SEL_DST_IP
2250 #define MCX_HASH_SEL_L4		MCX_HASH_SEL_L3 | MCX_TIR_CTX_HASH_SEL_SPORT | \
2251 				MCX_TIR_CTX_HASH_SEL_DPORT
2252 
2253 #define MCX_RSS_HASH_SEL_V4_TCP MCX_HASH_SEL_L4 | MCX_TIR_CTX_HASH_SEL_TCP  |\
2254 				MCX_TIR_CTX_HASH_SEL_IPV4
2255 #define MCX_RSS_HASH_SEL_V6_TCP	MCX_HASH_SEL_L4 | MCX_TIR_CTX_HASH_SEL_TCP | \
2256 				MCX_TIR_CTX_HASH_SEL_IPV6
2257 #define MCX_RSS_HASH_SEL_V4_UDP	MCX_HASH_SEL_L4 | MCX_TIR_CTX_HASH_SEL_UDP | \
2258 				MCX_TIR_CTX_HASH_SEL_IPV4
2259 #define MCX_RSS_HASH_SEL_V6_UDP	MCX_HASH_SEL_L4 | MCX_TIR_CTX_HASH_SEL_UDP | \
2260 				MCX_TIR_CTX_HASH_SEL_IPV6
2261 #define MCX_RSS_HASH_SEL_V4	MCX_HASH_SEL_L3 | MCX_TIR_CTX_HASH_SEL_IPV4
2262 #define MCX_RSS_HASH_SEL_V6	MCX_HASH_SEL_L3 | MCX_TIR_CTX_HASH_SEL_IPV6
2263 
2264 /*
2265  * There are a few different pieces involved in configuring RSS.
2266  * A Receive Queue Table (RQT) is the indirection table that maps packets to
2267  * different rx queues based on a hash value.  We only create one, because
2268  * we want to scatter any traffic we can apply RSS to across all our rx
2269  * queues.  Anything else will only be delivered to the first rx queue,
2270  * which doesn't require an RQT.
2271  *
2272  * A Transport Interface Receive (TIR) delivers packets to either a single rx
2273  * queue or an RQT, and in the latter case, specifies the set of fields
2274  * hashed, the hash function, and the hash key.  We need one of these for each
2275  * type of RSS traffic - v4 TCP, v6 TCP, v4 UDP, v6 UDP, other v4, other v6,
2276  * and one for non-RSS traffic.
2277  *
2278  * Flow tables hold flow table entries in sequence.  The first entry that
2279  * matches a packet is applied, sending the packet to either another flow
2280  * table or a TIR.  We use one flow table to select packets based on
2281  * destination MAC address, and a second to apply RSS.  The entries in the
2282  * first table send matching packets to the second, and the entries in the
2283  * RSS table send packets to RSS TIRs if possible, or the non-RSS TIR.
2284  *
2285  * The flow table entry that delivers packets to an RSS TIR must include match
2286  * criteria that ensure packets delivered to the TIR include all the fields
2287  * that the TIR hashes on - so for a v4 TCP TIR, the flow table entry must
2288  * only accept v4 TCP packets.  Accordingly, we need flow table entries for
2289  * each TIR.
2290  *
2291  * All of this is a lot more flexible than we need, and we can describe most
2292  * of the stuff we need with a simple array.
2293  *
2294  * An RSS config creates a TIR with hashing enabled on a set of fields,
2295  * pointing to either the first rx queue or the RQT containing all the rx
2296  * queues, and a flow table entry that matches on an ether type and
2297  * optionally an ip proto, that delivers packets to the TIR.
2298  */
2299 static struct mcx_rss_rule {
2300 	int			hash_sel;
2301 	int			flow_group;
2302 	int			ethertype;
2303 	int			ip_proto;
2304 } mcx_rss_config[] = {
2305 	/* udp and tcp for v4/v6 */
2306 	{ MCX_RSS_HASH_SEL_V4_TCP, MCX_FLOW_GROUP_RSS_L4,
2307 	  ETHERTYPE_IP, IPPROTO_TCP },
2308 	{ MCX_RSS_HASH_SEL_V6_TCP, MCX_FLOW_GROUP_RSS_L4,
2309 	  ETHERTYPE_IPV6, IPPROTO_TCP },
2310 	{ MCX_RSS_HASH_SEL_V4_UDP, MCX_FLOW_GROUP_RSS_L4,
2311 	  ETHERTYPE_IP, IPPROTO_UDP },
2312 	{ MCX_RSS_HASH_SEL_V6_UDP, MCX_FLOW_GROUP_RSS_L4,
2313 	  ETHERTYPE_IPV6, IPPROTO_UDP },
2314 
2315 	/* other v4/v6 */
2316 	{ MCX_RSS_HASH_SEL_V4, MCX_FLOW_GROUP_RSS_L3,
2317 	  ETHERTYPE_IP, 0 },
2318 	{ MCX_RSS_HASH_SEL_V6, MCX_FLOW_GROUP_RSS_L3,
2319 	  ETHERTYPE_IPV6, 0 },
2320 
2321 	/* non v4/v6 */
2322 	{ 0, MCX_FLOW_GROUP_RSS_NONE, 0, 0 }
2323 };
2324 
2325 struct mcx_softc {
2326 	struct device		 sc_dev;
2327 	struct arpcom		 sc_ac;
2328 	struct ifmedia		 sc_media;
2329 	uint64_t		 sc_media_status;
2330 	uint64_t		 sc_media_active;
2331 
2332 	pci_chipset_tag_t	 sc_pc;
2333 	pci_intr_handle_t	 sc_ih;
2334 	void			*sc_ihc;
2335 	pcitag_t		 sc_tag;
2336 
2337 	bus_dma_tag_t		 sc_dmat;
2338 	bus_space_tag_t		 sc_memt;
2339 	bus_space_handle_t	 sc_memh;
2340 	bus_size_t		 sc_mems;
2341 
2342 	struct mcx_dmamem	 sc_cmdq_mem;
2343 	unsigned int		 sc_cmdq_mask;
2344 	unsigned int		 sc_cmdq_size;
2345 
2346 	unsigned int		 sc_cmdq_token;
2347 
2348 	struct mcx_hwmem	 sc_boot_pages;
2349 	struct mcx_hwmem	 sc_init_pages;
2350 	struct mcx_hwmem	 sc_regular_pages;
2351 
2352 	int			 sc_uar;
2353 	int			 sc_pd;
2354 	int			 sc_tdomain;
2355 	uint32_t		 sc_lkey;
2356 	int			 sc_tis;
2357 	int			 sc_tir[nitems(mcx_rss_config)];
2358 	int			 sc_rqt;
2359 
2360 	struct mcx_dmamem	 sc_doorbell_mem;
2361 
2362 	struct mcx_eq		 sc_admin_eq;
2363 	struct mcx_eq		 sc_queue_eq;
2364 
2365 	int			 sc_hardmtu;
2366 	int			 sc_rxbufsz;
2367 
2368 	int			 sc_bf_size;
2369 	int			 sc_max_rqt_size;
2370 
2371 	struct task		 sc_port_change;
2372 
2373 	int			 sc_mac_flow_table_id;
2374 	int			 sc_rss_flow_table_id;
2375 	struct mcx_flow_group	 sc_flow_group[MCX_NUM_FLOW_GROUPS];
2376 	int			 sc_promisc_flow_enabled;
2377 	int			 sc_allmulti_flow_enabled;
2378 	int			 sc_mcast_flow_base;
2379 	int			 sc_extra_mcast;
2380 	uint8_t			 sc_mcast_flows[MCX_NUM_MCAST_FLOWS][ETHER_ADDR_LEN];
2381 
2382 	struct mcx_calibration	 sc_calibration[2];
2383 	unsigned int		 sc_calibration_gen;
2384 	struct timeout		 sc_calibrate;
2385 
2386 	struct mcx_queues	 sc_queues[MCX_MAX_QUEUES];
2387 	unsigned int		 sc_nqueues;
2388 
2389 #if NKSTAT > 0
2390 	struct kstat		*sc_kstat_ieee8023;
2391 	struct kstat		*sc_kstat_rfc2863;
2392 	struct kstat		*sc_kstat_rfc2819;
2393 	struct kstat		*sc_kstat_rfc3635;
2394 	unsigned int		 sc_kstat_mtmp_count;
2395 	struct kstat		**sc_kstat_mtmp;
2396 #endif
2397 };
2398 #define DEVNAME(_sc) ((_sc)->sc_dev.dv_xname)
2399 
2400 static int	mcx_match(struct device *, void *, void *);
2401 static void	mcx_attach(struct device *, struct device *, void *);
2402 
2403 #if NKSTAT > 0
2404 static void	mcx_kstat_attach(struct mcx_softc *);
2405 #endif
2406 
2407 static int	mcx_version(struct mcx_softc *);
2408 static int	mcx_init_wait(struct mcx_softc *);
2409 static int	mcx_enable_hca(struct mcx_softc *);
2410 static int	mcx_teardown_hca(struct mcx_softc *, uint16_t);
2411 static int	mcx_access_hca_reg(struct mcx_softc *, uint16_t, int, void *,
2412 		    int);
2413 static int	mcx_issi(struct mcx_softc *);
2414 static int	mcx_pages(struct mcx_softc *, struct mcx_hwmem *, uint16_t);
2415 static int	mcx_hca_max_caps(struct mcx_softc *);
2416 static int	mcx_hca_set_caps(struct mcx_softc *);
2417 static int	mcx_init_hca(struct mcx_softc *);
2418 static int	mcx_set_driver_version(struct mcx_softc *);
2419 static int	mcx_iff(struct mcx_softc *);
2420 static int	mcx_alloc_uar(struct mcx_softc *, int *);
2421 static int	mcx_alloc_pd(struct mcx_softc *);
2422 static int	mcx_alloc_tdomain(struct mcx_softc *);
2423 static int	mcx_create_eq(struct mcx_softc *, struct mcx_eq *, int,
2424 		    uint64_t, int);
2425 static int	mcx_query_nic_vport_context(struct mcx_softc *);
2426 static int	mcx_query_special_contexts(struct mcx_softc *);
2427 static int	mcx_set_port_mtu(struct mcx_softc *, int);
2428 static int	mcx_create_cq(struct mcx_softc *, struct mcx_cq *, int, int,
2429 		    int);
2430 static int	mcx_destroy_cq(struct mcx_softc *, struct mcx_cq *);
2431 static int	mcx_create_sq(struct mcx_softc *, struct mcx_tx *, int, int,
2432 		    int);
2433 static int	mcx_destroy_sq(struct mcx_softc *, struct mcx_tx *);
2434 static int	mcx_ready_sq(struct mcx_softc *, struct mcx_tx *);
2435 static int	mcx_create_rq(struct mcx_softc *, struct mcx_rx *, int, int);
2436 static int	mcx_destroy_rq(struct mcx_softc *, struct mcx_rx *);
2437 static int	mcx_ready_rq(struct mcx_softc *, struct mcx_rx *);
2438 static int	mcx_create_tir_direct(struct mcx_softc *, struct mcx_rx *,
2439 		    int *);
2440 static int	mcx_create_tir_indirect(struct mcx_softc *, int, uint32_t,
2441 		    int *);
2442 static int	mcx_destroy_tir(struct mcx_softc *, int);
2443 static int	mcx_create_tis(struct mcx_softc *, int *);
2444 static int	mcx_destroy_tis(struct mcx_softc *, int);
2445 static int	mcx_create_rqt(struct mcx_softc *, int, int *, int *);
2446 static int	mcx_destroy_rqt(struct mcx_softc *, int);
2447 static int	mcx_create_flow_table(struct mcx_softc *, int, int, int *);
2448 static int	mcx_set_flow_table_root(struct mcx_softc *, int);
2449 static int	mcx_destroy_flow_table(struct mcx_softc *, int);
2450 static int	mcx_create_flow_group(struct mcx_softc *, int, int, int,
2451 		    int, int, struct mcx_flow_match *);
2452 static int	mcx_destroy_flow_group(struct mcx_softc *, int);
2453 static int	mcx_set_flow_table_entry_mac(struct mcx_softc *, int, int,
2454 		    uint8_t *, uint32_t);
2455 static int	mcx_set_flow_table_entry_proto(struct mcx_softc *, int, int,
2456 		    int, int, uint32_t);
2457 static int	mcx_delete_flow_table_entry(struct mcx_softc *, int, int);
2458 
2459 #if NKSTAT > 0
2460 static int	mcx_query_rq(struct mcx_softc *, struct mcx_rx *, struct mcx_rq_ctx *);
2461 static int	mcx_query_sq(struct mcx_softc *, struct mcx_tx *, struct mcx_sq_ctx *);
2462 static int	mcx_query_cq(struct mcx_softc *, struct mcx_cq *, struct mcx_cq_ctx *);
2463 static int	mcx_query_eq(struct mcx_softc *, struct mcx_eq *, struct mcx_eq_ctx *);
2464 #endif
2465 
2466 #if 0
2467 static int	mcx_dump_flow_table(struct mcx_softc *, int);
2468 static int	mcx_dump_flow_table_entry(struct mcx_softc *, int, int);
2469 static int	mcx_dump_flow_group(struct mcx_softc *, int);
2470 #endif
2471 
2472 
2473 /*
2474 static void	mcx_cmdq_dump(const struct mcx_cmdq_entry *);
2475 static void	mcx_cmdq_mbox_dump(struct mcx_dmamem *, int);
2476 */
2477 static void	mcx_refill(void *);
2478 static int	mcx_process_rx(struct mcx_softc *, struct mcx_rx *,
2479 		    struct mcx_cq_entry *, struct mbuf_list *,
2480 		    const struct mcx_calibration *);
2481 static void	mcx_process_txeof(struct mcx_softc *, struct mcx_tx *,
2482 		    struct mcx_cq_entry *, int *);
2483 static void	mcx_process_cq(struct mcx_softc *, struct mcx_queues *,
2484 		    struct mcx_cq *);
2485 
2486 static void	mcx_arm_cq(struct mcx_softc *, struct mcx_cq *, int);
2487 static void	mcx_arm_eq(struct mcx_softc *, struct mcx_eq *, int);
2488 static int	mcx_admin_intr(void *);
2489 static int	mcx_cq_intr(void *);
2490 
2491 static int	mcx_up(struct mcx_softc *);
2492 static void	mcx_down(struct mcx_softc *);
2493 static int	mcx_ioctl(struct ifnet *, u_long, caddr_t);
2494 static int	mcx_rxrinfo(struct mcx_softc *, struct if_rxrinfo *);
2495 static void	mcx_start(struct ifqueue *);
2496 static void	mcx_watchdog(struct ifnet *);
2497 static void	mcx_media_add_types(struct mcx_softc *);
2498 static void	mcx_media_status(struct ifnet *, struct ifmediareq *);
2499 static int	mcx_media_change(struct ifnet *);
2500 static int	mcx_get_sffpage(struct ifnet *, struct if_sffpage *);
2501 static void	mcx_port_change(void *);
2502 
2503 static void	mcx_calibrate_first(struct mcx_softc *);
2504 static void	mcx_calibrate(void *);
2505 
2506 static inline uint32_t
2507 		mcx_rd(struct mcx_softc *, bus_size_t);
2508 static inline void
2509 		mcx_wr(struct mcx_softc *, bus_size_t, uint32_t);
2510 static inline void
2511 		mcx_bar(struct mcx_softc *, bus_size_t, bus_size_t, int);
2512 
2513 static uint64_t	mcx_timer(struct mcx_softc *);
2514 
2515 static int	mcx_dmamem_alloc(struct mcx_softc *, struct mcx_dmamem *,
2516 		    bus_size_t, u_int align);
2517 static void	mcx_dmamem_zero(struct mcx_dmamem *);
2518 static void	mcx_dmamem_free(struct mcx_softc *, struct mcx_dmamem *);
2519 
2520 static int	mcx_hwmem_alloc(struct mcx_softc *, struct mcx_hwmem *,
2521 		    unsigned int);
2522 static void	mcx_hwmem_free(struct mcx_softc *, struct mcx_hwmem *);
2523 
2524 struct cfdriver mcx_cd = {
2525 	NULL,
2526 	"mcx",
2527 	DV_IFNET,
2528 };
2529 
2530 struct cfattach mcx_ca = {
2531 	sizeof(struct mcx_softc),
2532 	mcx_match,
2533 	mcx_attach,
2534 };
2535 
2536 static const struct pci_matchid mcx_devices[] = {
2537 	{ PCI_VENDOR_MELLANOX,	PCI_PRODUCT_MELLANOX_MT27700 },
2538 	{ PCI_VENDOR_MELLANOX,	PCI_PRODUCT_MELLANOX_MT27700VF },
2539 	{ PCI_VENDOR_MELLANOX,	PCI_PRODUCT_MELLANOX_MT27710 },
2540 	{ PCI_VENDOR_MELLANOX,	PCI_PRODUCT_MELLANOX_MT27710VF },
2541 	{ PCI_VENDOR_MELLANOX,	PCI_PRODUCT_MELLANOX_MT27800 },
2542 	{ PCI_VENDOR_MELLANOX,	PCI_PRODUCT_MELLANOX_MT27800VF },
2543 	{ PCI_VENDOR_MELLANOX,	PCI_PRODUCT_MELLANOX_MT28800 },
2544 	{ PCI_VENDOR_MELLANOX,	PCI_PRODUCT_MELLANOX_MT28800VF },
2545 	{ PCI_VENDOR_MELLANOX,	PCI_PRODUCT_MELLANOX_MT2892  },
2546 };
2547 
2548 struct mcx_eth_proto_capability {
2549 	uint64_t	cap_media;
2550 	uint64_t	cap_baudrate;
2551 };
2552 
2553 static const struct mcx_eth_proto_capability mcx_eth_cap_map[] = {
2554 	[MCX_ETHER_CAP_SGMII]		= { IFM_1000_SGMII,	IF_Gbps(1) },
2555 	[MCX_ETHER_CAP_1000_KX]		= { IFM_1000_KX,	IF_Gbps(1) },
2556 	[MCX_ETHER_CAP_10G_CX4]		= { IFM_10G_CX4,	IF_Gbps(10) },
2557 	[MCX_ETHER_CAP_10G_KX4]		= { IFM_10G_KX4,	IF_Gbps(10) },
2558 	[MCX_ETHER_CAP_10G_KR]		= { IFM_10G_KR,		IF_Gbps(10) },
2559 	[MCX_ETHER_CAP_40G_CR4]		= { IFM_40G_CR4,	IF_Gbps(40) },
2560 	[MCX_ETHER_CAP_40G_KR4]		= { IFM_40G_KR4,	IF_Gbps(40) },
2561 	[MCX_ETHER_CAP_10G_CR]		= { IFM_10G_SFP_CU,	IF_Gbps(10) },
2562 	[MCX_ETHER_CAP_10G_SR]		= { IFM_10G_SR,		IF_Gbps(10) },
2563 	[MCX_ETHER_CAP_10G_LR]		= { IFM_10G_LR,		IF_Gbps(10) },
2564 	[MCX_ETHER_CAP_40G_SR4]		= { IFM_40G_SR4,	IF_Gbps(40) },
2565 	[MCX_ETHER_CAP_40G_LR4]		= { IFM_40G_LR4,	IF_Gbps(40) },
2566 	[MCX_ETHER_CAP_50G_SR2]		= { 0 /*IFM_50G_SR2*/,	IF_Gbps(50) },
2567 	[MCX_ETHER_CAP_100G_CR4]	= { IFM_100G_CR4,	IF_Gbps(100) },
2568 	[MCX_ETHER_CAP_100G_SR4]	= { IFM_100G_SR4,	IF_Gbps(100) },
2569 	[MCX_ETHER_CAP_100G_KR4]	= { IFM_100G_KR4,	IF_Gbps(100) },
2570 	[MCX_ETHER_CAP_25G_CR]		= { IFM_25G_CR,		IF_Gbps(25) },
2571 	[MCX_ETHER_CAP_25G_KR]		= { IFM_25G_KR,		IF_Gbps(25) },
2572 	[MCX_ETHER_CAP_25G_SR]		= { IFM_25G_SR,		IF_Gbps(25) },
2573 	[MCX_ETHER_CAP_50G_CR2]		= { IFM_50G_CR2,	IF_Gbps(50) },
2574 	[MCX_ETHER_CAP_50G_KR2]		= { IFM_50G_KR2,	IF_Gbps(50) },
2575 };
2576 
2577 static int
2578 mcx_get_id(uint32_t val)
2579 {
2580 	return betoh32(val) & 0x00ffffff;
2581 }
2582 
2583 static int
2584 mcx_match(struct device *parent, void *match, void *aux)
2585 {
2586 	return (pci_matchbyid(aux, mcx_devices, nitems(mcx_devices)));
2587 }
2588 
2589 void
2590 mcx_attach(struct device *parent, struct device *self, void *aux)
2591 {
2592 	struct mcx_softc *sc = (struct mcx_softc *)self;
2593 	struct ifnet *ifp = &sc->sc_ac.ac_if;
2594 	struct pci_attach_args *pa = aux;
2595 	pcireg_t memtype;
2596 	uint32_t r;
2597 	unsigned int cq_stride;
2598 	unsigned int cq_size;
2599 	const char *intrstr;
2600 	int i, msix;
2601 
2602 	sc->sc_pc = pa->pa_pc;
2603 	sc->sc_tag = pa->pa_tag;
2604 	sc->sc_dmat = pa->pa_dmat;
2605 
2606 	/* Map the PCI memory space */
2607 	memtype = pci_mapreg_type(sc->sc_pc, sc->sc_tag, MCX_HCA_BAR);
2608 	if (pci_mapreg_map(pa, MCX_HCA_BAR, memtype,
2609 	    BUS_SPACE_MAP_PREFETCHABLE, &sc->sc_memt, &sc->sc_memh,
2610 	    NULL, &sc->sc_mems, 0)) {
2611 		printf(": unable to map register memory\n");
2612 		return;
2613 	}
2614 
2615 	if (mcx_version(sc) != 0) {
2616 		/* error printed by mcx_version */
2617 		goto unmap;
2618 	}
2619 
2620 	r = mcx_rd(sc, MCX_CMDQ_ADDR_LO);
2621 	cq_stride = 1 << MCX_CMDQ_LOG_STRIDE(r); /* size of the entries */
2622 	cq_size = 1 << MCX_CMDQ_LOG_SIZE(r); /* number of entries */
2623 	if (cq_size > MCX_MAX_CQE) {
2624 		printf(", command queue size overflow %u\n", cq_size);
2625 		goto unmap;
2626 	}
2627 	if (cq_stride < sizeof(struct mcx_cmdq_entry)) {
2628 		printf(", command queue entry size underflow %u\n", cq_stride);
2629 		goto unmap;
2630 	}
2631 	if (cq_stride * cq_size > MCX_PAGE_SIZE) {
2632 		printf(", command queue page overflow\n");
2633 		goto unmap;
2634 	}
2635 
2636 	if (mcx_dmamem_alloc(sc, &sc->sc_doorbell_mem, MCX_DOORBELL_AREA_SIZE,
2637 	    MCX_PAGE_SIZE) != 0) {
2638 		printf(", unable to allocate doorbell memory\n");
2639 		goto unmap;
2640 	}
2641 
2642 	if (mcx_dmamem_alloc(sc, &sc->sc_cmdq_mem, MCX_PAGE_SIZE,
2643 	    MCX_PAGE_SIZE) != 0) {
2644 		printf(", unable to allocate command queue\n");
2645 		goto dbfree;
2646 	}
2647 
2648 	mcx_wr(sc, MCX_CMDQ_ADDR_HI, MCX_DMA_DVA(&sc->sc_cmdq_mem) >> 32);
2649 	mcx_bar(sc, MCX_CMDQ_ADDR_HI, sizeof(uint32_t),
2650 	    BUS_SPACE_BARRIER_WRITE);
2651 	mcx_wr(sc, MCX_CMDQ_ADDR_LO, MCX_DMA_DVA(&sc->sc_cmdq_mem));
2652 	mcx_bar(sc, MCX_CMDQ_ADDR_LO, sizeof(uint32_t),
2653 	    BUS_SPACE_BARRIER_WRITE);
2654 
2655 	if (mcx_init_wait(sc) != 0) {
2656 		printf(", timeout waiting for init\n");
2657 		goto cqfree;
2658 	}
2659 
2660 	sc->sc_cmdq_mask = cq_size - 1;
2661 	sc->sc_cmdq_size = cq_stride;
2662 
2663 	if (mcx_enable_hca(sc) != 0) {
2664 		/* error printed by mcx_enable_hca */
2665 		goto cqfree;
2666 	}
2667 
2668 	if (mcx_issi(sc) != 0) {
2669 		/* error printed by mcx_issi */
2670 		goto teardown;
2671 	}
2672 
2673 	if (mcx_pages(sc, &sc->sc_boot_pages,
2674 	    htobe16(MCX_CMD_QUERY_PAGES_BOOT)) != 0) {
2675 		/* error printed by mcx_pages */
2676 		goto teardown;
2677 	}
2678 
2679 	if (mcx_hca_max_caps(sc) != 0) {
2680 		/* error printed by mcx_hca_max_caps */
2681 		goto teardown;
2682 	}
2683 
2684 	if (mcx_hca_set_caps(sc) != 0) {
2685 		/* error printed by mcx_hca_set_caps */
2686 		goto teardown;
2687 	}
2688 
2689 	if (mcx_pages(sc, &sc->sc_init_pages,
2690 	    htobe16(MCX_CMD_QUERY_PAGES_INIT)) != 0) {
2691 		/* error printed by mcx_pages */
2692 		goto teardown;
2693 	}
2694 
2695 	if (mcx_init_hca(sc) != 0) {
2696 		/* error printed by mcx_init_hca */
2697 		goto teardown;
2698 	}
2699 
2700 	if (mcx_pages(sc, &sc->sc_regular_pages,
2701 	    htobe16(MCX_CMD_QUERY_PAGES_REGULAR)) != 0) {
2702 		/* error printed by mcx_pages */
2703 		goto teardown;
2704 	}
2705 
2706 	/* apparently not necessary? */
2707 	if (mcx_set_driver_version(sc) != 0) {
2708 		/* error printed by mcx_set_driver_version */
2709 		goto teardown;
2710 	}
2711 
2712 	if (mcx_iff(sc) != 0) {	/* modify nic vport context */
2713 		/* error printed by mcx_iff? */
2714 		goto teardown;
2715 	}
2716 
2717 	if (mcx_alloc_uar(sc, &sc->sc_uar) != 0) {
2718 		/* error printed by mcx_alloc_uar */
2719 		goto teardown;
2720 	}
2721 
2722 	if (mcx_alloc_pd(sc) != 0) {
2723 		/* error printed by mcx_alloc_pd */
2724 		goto teardown;
2725 	}
2726 
2727 	if (mcx_alloc_tdomain(sc) != 0) {
2728 		/* error printed by mcx_alloc_tdomain */
2729 		goto teardown;
2730 	}
2731 
2732 	/*
2733 	 * PRM makes no mention of msi interrupts, just legacy and msi-x.
2734 	 * mellanox support tells me legacy interrupts are not supported,
2735 	 * so we're stuck with just msi-x.
2736 	 */
2737 	if (pci_intr_map_msix(pa, 0, &sc->sc_ih) != 0) {
2738 		printf(": unable to map interrupt\n");
2739 		goto teardown;
2740 	}
2741 	intrstr = pci_intr_string(sc->sc_pc, sc->sc_ih);
2742 	sc->sc_ihc = pci_intr_establish(sc->sc_pc, sc->sc_ih,
2743 	    IPL_NET | IPL_MPSAFE, mcx_admin_intr, sc, DEVNAME(sc));
2744 	if (sc->sc_ihc == NULL) {
2745 		printf(": unable to establish interrupt");
2746 		if (intrstr != NULL)
2747 			printf(" at %s", intrstr);
2748 		printf("\n");
2749 		goto teardown;
2750 	}
2751 
2752 	if (mcx_create_eq(sc, &sc->sc_admin_eq, sc->sc_uar,
2753 	    (1ull << MCX_EVENT_TYPE_INTERNAL_ERROR) |
2754 	    (1ull << MCX_EVENT_TYPE_PORT_CHANGE) |
2755 	    (1ull << MCX_EVENT_TYPE_CMD_COMPLETION) |
2756 	    (1ull << MCX_EVENT_TYPE_PAGE_REQUEST), 0) != 0) {
2757 		/* error printed by mcx_create_eq */
2758 		goto teardown;
2759 	}
2760 
2761 	if (mcx_query_nic_vport_context(sc) != 0) {
2762 		/* error printed by mcx_query_nic_vport_context */
2763 		goto teardown;
2764 	}
2765 
2766 	if (mcx_query_special_contexts(sc) != 0) {
2767 		/* error printed by mcx_query_special_contexts */
2768 		goto teardown;
2769 	}
2770 
2771 	if (mcx_set_port_mtu(sc, MCX_HARDMTU) != 0) {
2772 		/* error printed by mcx_set_port_mtu */
2773 		goto teardown;
2774 	}
2775 
2776 	printf(", %s, address %s\n", intrstr,
2777 	    ether_sprintf(sc->sc_ac.ac_enaddr));
2778 
2779 	msix = pci_intr_msix_count(pa->pa_pc, pa->pa_tag);
2780 	sc->sc_nqueues = 1;
2781 
2782 	strlcpy(ifp->if_xname, DEVNAME(sc), IFNAMSIZ);
2783 	ifp->if_softc = sc;
2784 	ifp->if_flags = IFF_BROADCAST | IFF_MULTICAST | IFF_SIMPLEX;
2785 	ifp->if_xflags = IFXF_MPSAFE;
2786 	ifp->if_ioctl = mcx_ioctl;
2787 	ifp->if_qstart = mcx_start;
2788 	ifp->if_watchdog = mcx_watchdog;
2789 	ifp->if_hardmtu = sc->sc_hardmtu;
2790 	ifp->if_capabilities = IFCAP_VLAN_MTU | IFCAP_CSUM_IPv4 |
2791 	    IFCAP_CSUM_UDPv4 | IFCAP_CSUM_UDPv6 | IFCAP_CSUM_TCPv4 |
2792 	    IFCAP_CSUM_TCPv6;
2793 #if NVLAN > 0
2794 	ifp->if_capabilities |= IFCAP_VLAN_HWTAGGING;
2795 #endif
2796 	ifq_set_maxlen(&ifp->if_snd, 1024);
2797 
2798 	ifmedia_init(&sc->sc_media, IFM_IMASK, mcx_media_change,
2799 	    mcx_media_status);
2800 	mcx_media_add_types(sc);
2801 	ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL);
2802 	ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO);
2803 
2804 	if_attach(ifp);
2805 	ether_ifattach(ifp);
2806 
2807 	if_attach_iqueues(ifp, sc->sc_nqueues);
2808 	if_attach_queues(ifp, sc->sc_nqueues);
2809 	for (i = 0; i < sc->sc_nqueues; i++) {
2810 		struct ifiqueue *ifiq = ifp->if_iqs[i];
2811 		struct ifqueue *ifq = ifp->if_ifqs[i];
2812 		struct mcx_queues *q = &sc->sc_queues[i];
2813 		struct mcx_rx *rx = &q->q_rx;
2814 		struct mcx_tx *tx = &q->q_tx;
2815 		pci_intr_handle_t ih;
2816 		int vec;
2817 
2818 		vec = i + 1;
2819 		q->q_sc = sc;
2820 		q->q_index = i;
2821 
2822 		if (mcx_alloc_uar(sc, &q->q_uar) != 0) {
2823 			printf("%s: unable to alloc uar %d\n",
2824 			    DEVNAME(sc), i);
2825 			goto teardown;
2826 		}
2827 
2828 		if (mcx_create_eq(sc, &q->q_eq, q->q_uar, 0, vec) != 0) {
2829 			printf("%s: unable to create event queue %d\n",
2830 			    DEVNAME(sc), i);
2831 			goto teardown;
2832 		}
2833 
2834 		rx->rx_softc = sc;
2835 		rx->rx_ifiq = ifiq;
2836 		timeout_set(&rx->rx_refill, mcx_refill, rx);
2837 		ifiq->ifiq_softc = rx;
2838 
2839 		tx->tx_softc = sc;
2840 		tx->tx_ifq = ifq;
2841 		ifq->ifq_softc = tx;
2842 
2843 		if (pci_intr_map_msix(pa, i + 1, &ih) != 0) {
2844 			printf("%s: unable to map queue interrupt %d\n",
2845 			    DEVNAME(sc), i);
2846 			goto teardown;
2847 		}
2848 		snprintf(q->q_name, sizeof(q->q_name), "%s:%d",
2849 		    DEVNAME(sc), i);
2850 		q->q_ihc = pci_intr_establish(sc->sc_pc, ih,
2851 		    IPL_NET | IPL_MPSAFE, mcx_cq_intr, q, q->q_name);
2852 	}
2853 
2854 	timeout_set(&sc->sc_calibrate, mcx_calibrate, sc);
2855 
2856 	task_set(&sc->sc_port_change, mcx_port_change, sc);
2857 	mcx_port_change(sc);
2858 
2859 	sc->sc_mac_flow_table_id = -1;
2860 	sc->sc_rss_flow_table_id = -1;
2861 	sc->sc_rqt = -1;
2862 	for (i = 0; i < MCX_NUM_FLOW_GROUPS; i++) {
2863 		struct mcx_flow_group *mfg = &sc->sc_flow_group[i];
2864 		mfg->g_id = -1;
2865 		mfg->g_table = -1;
2866 		mfg->g_size = 0;
2867 		mfg->g_start = 0;
2868 	}
2869 	sc->sc_extra_mcast = 0;
2870 	memset(sc->sc_mcast_flows, 0, sizeof(sc->sc_mcast_flows));
2871 
2872 #if NKSTAT > 0
2873 	mcx_kstat_attach(sc);
2874 #endif
2875 	return;
2876 
2877 teardown:
2878 	mcx_teardown_hca(sc, htobe16(MCX_CMD_TEARDOWN_HCA_GRACEFUL));
2879 	/* error printed by mcx_teardown_hca, and we're already unwinding */
2880 cqfree:
2881 	mcx_wr(sc, MCX_CMDQ_ADDR_HI, MCX_DMA_DVA(&sc->sc_cmdq_mem) >> 32);
2882 	mcx_bar(sc, MCX_CMDQ_ADDR_HI, sizeof(uint64_t),
2883 	    BUS_SPACE_BARRIER_WRITE);
2884 	mcx_wr(sc, MCX_CMDQ_ADDR_LO, MCX_DMA_DVA(&sc->sc_cmdq_mem) |
2885 	    MCX_CMDQ_INTERFACE_DISABLED);
2886 	mcx_bar(sc, MCX_CMDQ_ADDR_LO, sizeof(uint64_t),
2887 	    BUS_SPACE_BARRIER_WRITE);
2888 
2889 	mcx_wr(sc, MCX_CMDQ_ADDR_HI, 0);
2890 	mcx_bar(sc, MCX_CMDQ_ADDR_HI, sizeof(uint64_t),
2891 	    BUS_SPACE_BARRIER_WRITE);
2892 	mcx_wr(sc, MCX_CMDQ_ADDR_LO, MCX_CMDQ_INTERFACE_DISABLED);
2893 
2894 	mcx_dmamem_free(sc, &sc->sc_cmdq_mem);
2895 dbfree:
2896 	mcx_dmamem_free(sc, &sc->sc_doorbell_mem);
2897 unmap:
2898 	bus_space_unmap(sc->sc_memt, sc->sc_memh, sc->sc_mems);
2899 	sc->sc_mems = 0;
2900 }
2901 
2902 static int
2903 mcx_version(struct mcx_softc *sc)
2904 {
2905 	uint32_t fw0, fw1;
2906 	uint16_t cmdif;
2907 
2908 	fw0 = mcx_rd(sc, MCX_FW_VER);
2909 	fw1 = mcx_rd(sc, MCX_CMDIF_FW_SUBVER);
2910 
2911 	printf(": FW %u.%u.%04u", MCX_FW_VER_MAJOR(fw0),
2912 	    MCX_FW_VER_MINOR(fw0), MCX_FW_VER_SUBMINOR(fw1));
2913 
2914 	cmdif = MCX_CMDIF(fw1);
2915 	if (cmdif != MCX_CMD_IF_SUPPORTED) {
2916 		printf(", unsupported command interface %u\n", cmdif);
2917 		return (-1);
2918 	}
2919 
2920 	return (0);
2921 }
2922 
2923 static int
2924 mcx_init_wait(struct mcx_softc *sc)
2925 {
2926 	unsigned int i;
2927 	uint32_t r;
2928 
2929 	for (i = 0; i < 2000; i++) {
2930 		r = mcx_rd(sc, MCX_STATE);
2931 		if ((r & MCX_STATE_MASK) == MCX_STATE_READY)
2932 			return (0);
2933 
2934 		delay(1000);
2935 		mcx_bar(sc, MCX_STATE, sizeof(uint32_t),
2936 		    BUS_SPACE_BARRIER_READ);
2937 	}
2938 
2939 	return (-1);
2940 }
2941 
2942 static uint8_t
2943 mcx_cmdq_poll(struct mcx_softc *sc, struct mcx_cmdq_entry *cqe,
2944     unsigned int msec)
2945 {
2946 	unsigned int i;
2947 
2948 	for (i = 0; i < msec; i++) {
2949 		bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_cmdq_mem),
2950 		    0, MCX_DMA_LEN(&sc->sc_cmdq_mem), BUS_DMASYNC_POSTRW);
2951 
2952 		if ((cqe->cq_status & MCX_CQ_STATUS_OWN_MASK) ==
2953 		    MCX_CQ_STATUS_OWN_SW)
2954 			return (0);
2955 
2956 		delay(1000);
2957 	}
2958 
2959 	return (ETIMEDOUT);
2960 }
2961 
2962 static uint32_t
2963 mcx_mix_u64(uint32_t xor, uint64_t u64)
2964 {
2965 	xor ^= u64 >> 32;
2966 	xor ^= u64;
2967 
2968 	return (xor);
2969 }
2970 
2971 static uint32_t
2972 mcx_mix_u32(uint32_t xor, uint32_t u32)
2973 {
2974 	xor ^= u32;
2975 
2976 	return (xor);
2977 }
2978 
2979 static uint32_t
2980 mcx_mix_u8(uint32_t xor, uint8_t u8)
2981 {
2982 	xor ^= u8;
2983 
2984 	return (xor);
2985 }
2986 
2987 static uint8_t
2988 mcx_mix_done(uint32_t xor)
2989 {
2990 	xor ^= xor >> 16;
2991 	xor ^= xor >> 8;
2992 
2993 	return (xor);
2994 }
2995 
2996 static uint8_t
2997 mcx_xor(const void *buf, size_t len)
2998 {
2999 	const uint32_t *dwords = buf;
3000 	uint32_t xor = 0xff;
3001 	size_t i;
3002 
3003 	len /= sizeof(*dwords);
3004 
3005 	for (i = 0; i < len; i++)
3006 		xor ^= dwords[i];
3007 
3008 	return (mcx_mix_done(xor));
3009 }
3010 
3011 static uint8_t
3012 mcx_cmdq_token(struct mcx_softc *sc)
3013 {
3014 	uint8_t token;
3015 
3016 	do {
3017 		token = ++sc->sc_cmdq_token;
3018 	} while (token == 0);
3019 
3020 	return (token);
3021 }
3022 
3023 static void
3024 mcx_cmdq_init(struct mcx_softc *sc, struct mcx_cmdq_entry *cqe,
3025     uint32_t ilen, uint32_t olen, uint8_t token)
3026 {
3027 	memset(cqe, 0, sc->sc_cmdq_size);
3028 
3029 	cqe->cq_type = MCX_CMDQ_TYPE_PCIE;
3030 	htobem32(&cqe->cq_input_length, ilen);
3031 	htobem32(&cqe->cq_output_length, olen);
3032 	cqe->cq_token = token;
3033 	cqe->cq_status = MCX_CQ_STATUS_OWN_HW;
3034 }
3035 
3036 static void
3037 mcx_cmdq_sign(struct mcx_cmdq_entry *cqe)
3038 {
3039 	cqe->cq_signature = ~mcx_xor(cqe, sizeof(*cqe));
3040 }
3041 
3042 static int
3043 mcx_cmdq_verify(const struct mcx_cmdq_entry *cqe)
3044 {
3045 	/* return (mcx_xor(cqe, sizeof(*cqe)) ? -1 :  0); */
3046 	return (0);
3047 }
3048 
3049 static void *
3050 mcx_cmdq_in(struct mcx_cmdq_entry *cqe)
3051 {
3052 	return (&cqe->cq_input_data);
3053 }
3054 
3055 static void *
3056 mcx_cmdq_out(struct mcx_cmdq_entry *cqe)
3057 {
3058 	return (&cqe->cq_output_data);
3059 }
3060 
3061 static void
3062 mcx_cmdq_post(struct mcx_softc *sc, struct mcx_cmdq_entry *cqe,
3063     unsigned int slot)
3064 {
3065 	mcx_cmdq_sign(cqe);
3066 
3067 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_cmdq_mem),
3068 	    0, MCX_DMA_LEN(&sc->sc_cmdq_mem), BUS_DMASYNC_PRERW);
3069 
3070 	mcx_wr(sc, MCX_CMDQ_DOORBELL, 1U << slot);
3071 }
3072 
3073 static int
3074 mcx_enable_hca(struct mcx_softc *sc)
3075 {
3076 	struct mcx_cmdq_entry *cqe;
3077 	struct mcx_cmd_enable_hca_in *in;
3078 	struct mcx_cmd_enable_hca_out *out;
3079 	int error;
3080 	uint8_t status;
3081 
3082 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
3083 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
3084 
3085 	in = mcx_cmdq_in(cqe);
3086 	in->cmd_opcode = htobe16(MCX_CMD_ENABLE_HCA);
3087 	in->cmd_op_mod = htobe16(0);
3088 	in->cmd_function_id = htobe16(0);
3089 
3090 	mcx_cmdq_post(sc, cqe, 0);
3091 
3092 	error = mcx_cmdq_poll(sc, cqe, 1000);
3093 	if (error != 0) {
3094 		printf(", hca enable timeout\n");
3095 		return (-1);
3096 	}
3097 	if (mcx_cmdq_verify(cqe) != 0) {
3098 		printf(", hca enable command corrupt\n");
3099 		return (-1);
3100 	}
3101 
3102 	status = cqe->cq_output_data[0];
3103 	if (status != MCX_CQ_STATUS_OK) {
3104 		printf(", hca enable failed (%x)\n", status);
3105 		return (-1);
3106 	}
3107 
3108 	return (0);
3109 }
3110 
3111 static int
3112 mcx_teardown_hca(struct mcx_softc *sc, uint16_t profile)
3113 {
3114 	struct mcx_cmdq_entry *cqe;
3115 	struct mcx_cmd_teardown_hca_in *in;
3116 	struct mcx_cmd_teardown_hca_out *out;
3117 	int error;
3118 	uint8_t status;
3119 
3120 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
3121 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
3122 
3123 	in = mcx_cmdq_in(cqe);
3124 	in->cmd_opcode = htobe16(MCX_CMD_TEARDOWN_HCA);
3125 	in->cmd_op_mod = htobe16(0);
3126 	in->cmd_profile = profile;
3127 
3128 	mcx_cmdq_post(sc, cqe, 0);
3129 
3130 	error = mcx_cmdq_poll(sc, cqe, 1000);
3131 	if (error != 0) {
3132 		printf(", hca teardown timeout\n");
3133 		return (-1);
3134 	}
3135 	if (mcx_cmdq_verify(cqe) != 0) {
3136 		printf(", hca teardown command corrupt\n");
3137 		return (-1);
3138 	}
3139 
3140 	status = cqe->cq_output_data[0];
3141 	if (status != MCX_CQ_STATUS_OK) {
3142 		printf(", hca teardown failed (%x)\n", status);
3143 		return (-1);
3144 	}
3145 
3146 	return (0);
3147 }
3148 
3149 static int
3150 mcx_cmdq_mboxes_alloc(struct mcx_softc *sc, struct mcx_dmamem *mxm,
3151     unsigned int nmb, uint64_t *ptr, uint8_t token)
3152 {
3153 	caddr_t kva;
3154 	uint64_t dva;
3155 	int i;
3156 	int error;
3157 
3158 	error = mcx_dmamem_alloc(sc, mxm,
3159 	    nmb * MCX_CMDQ_MAILBOX_SIZE, MCX_CMDQ_MAILBOX_ALIGN);
3160 	if (error != 0)
3161 		return (error);
3162 
3163 	mcx_dmamem_zero(mxm);
3164 
3165 	dva = MCX_DMA_DVA(mxm);
3166 	kva = MCX_DMA_KVA(mxm);
3167 	for (i = 0; i < nmb; i++) {
3168 		struct mcx_cmdq_mailbox *mbox = (struct mcx_cmdq_mailbox *)kva;
3169 
3170 		/* patch the cqe or mbox pointing at this one */
3171 		htobem64(ptr, dva);
3172 
3173 		/* fill in this mbox */
3174 		htobem32(&mbox->mb_block_number, i);
3175 		mbox->mb_token = token;
3176 
3177 		/* move to the next one */
3178 		ptr = &mbox->mb_next_ptr;
3179 
3180 		dva += MCX_CMDQ_MAILBOX_SIZE;
3181 		kva += MCX_CMDQ_MAILBOX_SIZE;
3182 	}
3183 
3184 	return (0);
3185 }
3186 
3187 static uint32_t
3188 mcx_cmdq_mbox_ctrl_sig(const struct mcx_cmdq_mailbox *mb)
3189 {
3190 	uint32_t xor = 0xff;
3191 
3192 	/* only 3 fields get set, so mix them directly */
3193 	xor = mcx_mix_u64(xor, mb->mb_next_ptr);
3194 	xor = mcx_mix_u32(xor, mb->mb_block_number);
3195 	xor = mcx_mix_u8(xor, mb->mb_token);
3196 
3197 	return (mcx_mix_done(xor));
3198 }
3199 
3200 static void
3201 mcx_cmdq_mboxes_sign(struct mcx_dmamem *mxm, unsigned int nmb)
3202 {
3203 	caddr_t kva;
3204 	int i;
3205 
3206 	kva = MCX_DMA_KVA(mxm);
3207 
3208 	for (i = 0; i < nmb; i++) {
3209 		struct mcx_cmdq_mailbox *mb = (struct mcx_cmdq_mailbox *)kva;
3210 		uint8_t sig = mcx_cmdq_mbox_ctrl_sig(mb);
3211 		mb->mb_ctrl_signature = sig;
3212 		mb->mb_signature = sig ^
3213 		    mcx_xor(mb->mb_data, sizeof(mb->mb_data));
3214 
3215 		kva += MCX_CMDQ_MAILBOX_SIZE;
3216 	}
3217 }
3218 
3219 static void
3220 mcx_cmdq_mboxes_sync(struct mcx_softc *sc, struct mcx_dmamem *mxm, int ops)
3221 {
3222 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(mxm),
3223 	    0, MCX_DMA_LEN(mxm), ops);
3224 }
3225 
3226 static struct mcx_cmdq_mailbox *
3227 mcx_cq_mbox(struct mcx_dmamem *mxm, unsigned int i)
3228 {
3229 	caddr_t kva;
3230 
3231 	kva = MCX_DMA_KVA(mxm);
3232 	kva += i * MCX_CMDQ_MAILBOX_SIZE;
3233 
3234 	return ((struct mcx_cmdq_mailbox *)kva);
3235 }
3236 
3237 static inline void *
3238 mcx_cq_mbox_data(struct mcx_cmdq_mailbox *mb)
3239 {
3240 	return (&mb->mb_data);
3241 }
3242 
3243 static void
3244 mcx_cmdq_mboxes_copyin(struct mcx_dmamem *mxm, unsigned int nmb,
3245     void *b, size_t len)
3246 {
3247 	caddr_t buf = b;
3248 	struct mcx_cmdq_mailbox *mb;
3249 	int i;
3250 
3251 	mb = (struct mcx_cmdq_mailbox *)MCX_DMA_KVA(mxm);
3252 	for (i = 0; i < nmb; i++) {
3253 
3254 		memcpy(mb->mb_data, buf, min(sizeof(mb->mb_data), len));
3255 
3256 		if (sizeof(mb->mb_data) >= len)
3257 			break;
3258 
3259 		buf += sizeof(mb->mb_data);
3260 		len -= sizeof(mb->mb_data);
3261 		mb++;
3262 	}
3263 }
3264 
3265 static void
3266 mcx_cmdq_mboxes_pas(struct mcx_dmamem *mxm, int offset, int npages,
3267     struct mcx_dmamem *buf)
3268 {
3269 	uint64_t *pas;
3270 	int mbox, mbox_pages, i;
3271 
3272 	mbox = offset / MCX_CMDQ_MAILBOX_DATASIZE;
3273 	offset %= MCX_CMDQ_MAILBOX_DATASIZE;
3274 
3275 	pas = mcx_cq_mbox_data(mcx_cq_mbox(mxm, mbox));
3276 	pas += (offset / sizeof(*pas));
3277 	mbox_pages = (MCX_CMDQ_MAILBOX_DATASIZE - offset) / sizeof(*pas);
3278 	for (i = 0; i < npages; i++) {
3279 		if (i == mbox_pages) {
3280 			mbox++;
3281 			pas = mcx_cq_mbox_data(mcx_cq_mbox(mxm, mbox));
3282 			mbox_pages += MCX_CMDQ_MAILBOX_DATASIZE / sizeof(*pas);
3283 		}
3284 		*pas = htobe64(MCX_DMA_DVA(buf) + (i * MCX_PAGE_SIZE));
3285 		pas++;
3286 	}
3287 }
3288 
3289 static void
3290 mcx_cmdq_mboxes_copyout(struct mcx_dmamem *mxm, int nmb, void *b, size_t len)
3291 {
3292 	caddr_t buf = b;
3293 	struct mcx_cmdq_mailbox *mb;
3294 	int i;
3295 
3296 	mb = (struct mcx_cmdq_mailbox *)MCX_DMA_KVA(mxm);
3297 	for (i = 0; i < nmb; i++) {
3298 		memcpy(buf, mb->mb_data, min(sizeof(mb->mb_data), len));
3299 
3300 		if (sizeof(mb->mb_data) >= len)
3301 			break;
3302 
3303 		buf += sizeof(mb->mb_data);
3304 		len -= sizeof(mb->mb_data);
3305 		mb++;
3306 	}
3307 }
3308 
3309 static void
3310 mcx_cq_mboxes_free(struct mcx_softc *sc, struct mcx_dmamem *mxm)
3311 {
3312 	mcx_dmamem_free(sc, mxm);
3313 }
3314 
3315 #if 0
3316 static void
3317 mcx_cmdq_dump(const struct mcx_cmdq_entry *cqe)
3318 {
3319 	unsigned int i;
3320 
3321 	printf(" type %02x, ilen %u, iptr %016llx", cqe->cq_type,
3322 	    bemtoh32(&cqe->cq_input_length), bemtoh64(&cqe->cq_input_ptr));
3323 
3324 	printf(", idata ");
3325 	for (i = 0; i < sizeof(cqe->cq_input_data); i++)
3326 		printf("%02x", cqe->cq_input_data[i]);
3327 
3328 	printf(", odata ");
3329 	for (i = 0; i < sizeof(cqe->cq_output_data); i++)
3330 		printf("%02x", cqe->cq_output_data[i]);
3331 
3332 	printf(", optr %016llx, olen %u, token %02x, sig %02x, status %02x",
3333 	    bemtoh64(&cqe->cq_output_ptr), bemtoh32(&cqe->cq_output_length),
3334 	    cqe->cq_token, cqe->cq_signature, cqe->cq_status);
3335 }
3336 
3337 static void
3338 mcx_cmdq_mbox_dump(struct mcx_dmamem *mboxes, int num)
3339 {
3340 	int i, j;
3341 	uint8_t *d;
3342 
3343 	for (i = 0; i < num; i++) {
3344 		struct mcx_cmdq_mailbox *mbox;
3345 		mbox = mcx_cq_mbox(mboxes, i);
3346 
3347 		d = mcx_cq_mbox_data(mbox);
3348 		for (j = 0; j < MCX_CMDQ_MAILBOX_DATASIZE; j++) {
3349 			if (j != 0 && (j % 16 == 0))
3350 				printf("\n");
3351 			printf("%.2x ", d[j]);
3352 		}
3353 	}
3354 }
3355 #endif
3356 
3357 static int
3358 mcx_access_hca_reg(struct mcx_softc *sc, uint16_t reg, int op, void *data,
3359     int len)
3360 {
3361 	struct mcx_dmamem mxm;
3362 	struct mcx_cmdq_entry *cqe;
3363 	struct mcx_cmd_access_reg_in *in;
3364 	struct mcx_cmd_access_reg_out *out;
3365 	uint8_t token = mcx_cmdq_token(sc);
3366 	int error, nmb;
3367 
3368 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
3369 	mcx_cmdq_init(sc, cqe, sizeof(*in) + len, sizeof(*out) + len,
3370 	    token);
3371 
3372 	in = mcx_cmdq_in(cqe);
3373 	in->cmd_opcode = htobe16(MCX_CMD_ACCESS_REG);
3374 	in->cmd_op_mod = htobe16(op);
3375 	in->cmd_register_id = htobe16(reg);
3376 
3377 	nmb = howmany(len, MCX_CMDQ_MAILBOX_DATASIZE);
3378 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, nmb,
3379 	    &cqe->cq_output_ptr, token) != 0) {
3380 		printf(", unable to allocate access reg mailboxen\n");
3381 		return (-1);
3382 	}
3383 	cqe->cq_input_ptr = cqe->cq_output_ptr;
3384 	mcx_cmdq_mboxes_copyin(&mxm, nmb, data, len);
3385 	mcx_cmdq_mboxes_sign(&mxm, nmb);
3386 	mcx_cmdq_mboxes_sync(sc, &mxm, BUS_DMASYNC_PRERW);
3387 
3388 	mcx_cmdq_post(sc, cqe, 0);
3389 	error = mcx_cmdq_poll(sc, cqe, 1000);
3390 	mcx_cmdq_mboxes_sync(sc, &mxm, BUS_DMASYNC_POSTRW);
3391 
3392 	if (error != 0) {
3393 		printf("%s: access reg (%s %x) timeout\n", DEVNAME(sc),
3394 		    (op == MCX_REG_OP_WRITE ? "write" : "read"), reg);
3395 		goto free;
3396 	}
3397 	error = mcx_cmdq_verify(cqe);
3398 	if (error != 0) {
3399 		printf("%s: access reg (%s %x) reply corrupt\n",
3400 		    (op == MCX_REG_OP_WRITE ? "write" : "read"), DEVNAME(sc),
3401 		    reg);
3402 		goto free;
3403 	}
3404 
3405 	out = mcx_cmdq_out(cqe);
3406 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
3407 		printf("%s: access reg (%s %x) failed (%x, %.6x)\n",
3408 		    DEVNAME(sc), (op == MCX_REG_OP_WRITE ? "write" : "read"),
3409 		    reg, out->cmd_status, betoh32(out->cmd_syndrome));
3410 		error = -1;
3411 		goto free;
3412 	}
3413 
3414 	mcx_cmdq_mboxes_copyout(&mxm, nmb, data, len);
3415 free:
3416 	mcx_dmamem_free(sc, &mxm);
3417 
3418 	return (error);
3419 }
3420 
3421 static int
3422 mcx_set_issi(struct mcx_softc *sc, struct mcx_cmdq_entry *cqe,
3423     unsigned int slot)
3424 {
3425 	struct mcx_cmd_set_issi_in *in;
3426 	struct mcx_cmd_set_issi_out *out;
3427 	uint8_t status;
3428 
3429 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
3430 
3431 	in = mcx_cmdq_in(cqe);
3432 	in->cmd_opcode = htobe16(MCX_CMD_SET_ISSI);
3433 	in->cmd_op_mod = htobe16(0);
3434 	in->cmd_current_issi = htobe16(MCX_ISSI);
3435 
3436 	mcx_cmdq_post(sc, cqe, slot);
3437 	if (mcx_cmdq_poll(sc, cqe, 1000) != 0)
3438 		return (-1);
3439 	if (mcx_cmdq_verify(cqe) != 0)
3440 		return (-1);
3441 
3442 	status = cqe->cq_output_data[0];
3443 	if (status != MCX_CQ_STATUS_OK)
3444 		return (-1);
3445 
3446 	return (0);
3447 }
3448 
3449 static int
3450 mcx_issi(struct mcx_softc *sc)
3451 {
3452 	struct mcx_dmamem mxm;
3453 	struct mcx_cmdq_entry *cqe;
3454 	struct mcx_cmd_query_issi_in *in;
3455 	struct mcx_cmd_query_issi_il_out *out;
3456 	struct mcx_cmd_query_issi_mb_out *mb;
3457 	uint8_t token = mcx_cmdq_token(sc);
3458 	uint8_t status;
3459 	int error;
3460 
3461 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
3462 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out) + sizeof(*mb), token);
3463 
3464 	in = mcx_cmdq_in(cqe);
3465 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_ISSI);
3466 	in->cmd_op_mod = htobe16(0);
3467 
3468 	CTASSERT(sizeof(*mb) <= MCX_CMDQ_MAILBOX_DATASIZE);
3469 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
3470 	    &cqe->cq_output_ptr, token) != 0) {
3471 		printf(", unable to allocate query issi mailbox\n");
3472 		return (-1);
3473 	}
3474 	mcx_cmdq_mboxes_sign(&mxm, 1);
3475 
3476 	mcx_cmdq_post(sc, cqe, 0);
3477 	error = mcx_cmdq_poll(sc, cqe, 1000);
3478 	if (error != 0) {
3479 		printf(", query issi timeout\n");
3480 		goto free;
3481 	}
3482 	error = mcx_cmdq_verify(cqe);
3483 	if (error != 0) {
3484 		printf(", query issi reply corrupt\n");
3485 		goto free;
3486 	}
3487 
3488 	status = cqe->cq_output_data[0];
3489 	switch (status) {
3490 	case MCX_CQ_STATUS_OK:
3491 		break;
3492 	case MCX_CQ_STATUS_BAD_OPCODE:
3493 		/* use ISSI 0 */
3494 		goto free;
3495 	default:
3496 		printf(", query issi failed (%x)\n", status);
3497 		error = -1;
3498 		goto free;
3499 	}
3500 
3501 	out = mcx_cmdq_out(cqe);
3502 	if (out->cmd_current_issi == htobe16(MCX_ISSI)) {
3503 		/* use ISSI 1 */
3504 		goto free;
3505 	}
3506 
3507 	/* don't need to read cqe anymore, can be used for SET ISSI */
3508 
3509 	mb = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
3510 	CTASSERT(MCX_ISSI < NBBY);
3511 	 /* XXX math is hard */
3512 	if (!ISSET(mb->cmd_supported_issi[79], 1 << MCX_ISSI)) {
3513 		/* use ISSI 0 */
3514 		goto free;
3515 	}
3516 
3517 	if (mcx_set_issi(sc, cqe, 0) != 0) {
3518 		/* ignore the error, just use ISSI 0 */
3519 	} else {
3520 		/* use ISSI 1 */
3521 	}
3522 
3523 free:
3524 	mcx_cq_mboxes_free(sc, &mxm);
3525 	return (error);
3526 }
3527 
3528 static int
3529 mcx_query_pages(struct mcx_softc *sc, uint16_t type,
3530     int32_t *npages, uint16_t *func_id)
3531 {
3532 	struct mcx_cmdq_entry *cqe;
3533 	struct mcx_cmd_query_pages_in *in;
3534 	struct mcx_cmd_query_pages_out *out;
3535 
3536 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
3537 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
3538 
3539 	in = mcx_cmdq_in(cqe);
3540 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_PAGES);
3541 	in->cmd_op_mod = type;
3542 
3543 	mcx_cmdq_post(sc, cqe, 0);
3544 	if (mcx_cmdq_poll(sc, cqe, 1000) != 0) {
3545 		printf(", query pages timeout\n");
3546 		return (-1);
3547 	}
3548 	if (mcx_cmdq_verify(cqe) != 0) {
3549 		printf(", query pages reply corrupt\n");
3550 		return (-1);
3551 	}
3552 
3553 	out = mcx_cmdq_out(cqe);
3554 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
3555 		printf(", query pages failed (%x)\n", out->cmd_status);
3556 		return (-1);
3557 	}
3558 
3559 	*func_id = out->cmd_func_id;
3560 	*npages = bemtoh32(&out->cmd_num_pages);
3561 
3562 	return (0);
3563 }
3564 
3565 struct bus_dma_iter {
3566 	bus_dmamap_t		i_map;
3567 	bus_size_t		i_offset;
3568 	unsigned int		i_index;
3569 };
3570 
3571 static void
3572 bus_dma_iter_init(struct bus_dma_iter *i, bus_dmamap_t map)
3573 {
3574 	i->i_map = map;
3575 	i->i_offset = 0;
3576 	i->i_index = 0;
3577 }
3578 
3579 static bus_addr_t
3580 bus_dma_iter_addr(struct bus_dma_iter *i)
3581 {
3582 	return (i->i_map->dm_segs[i->i_index].ds_addr + i->i_offset);
3583 }
3584 
3585 static void
3586 bus_dma_iter_add(struct bus_dma_iter *i, bus_size_t size)
3587 {
3588 	bus_dma_segment_t *seg = i->i_map->dm_segs + i->i_index;
3589 	bus_size_t diff;
3590 
3591 	do {
3592 		diff = seg->ds_len - i->i_offset;
3593 		if (size < diff)
3594 			break;
3595 
3596 		size -= diff;
3597 
3598 		seg++;
3599 
3600 		i->i_offset = 0;
3601 		i->i_index++;
3602 	} while (size > 0);
3603 
3604 	i->i_offset += size;
3605 }
3606 
3607 static int
3608 mcx_add_pages(struct mcx_softc *sc, struct mcx_hwmem *mhm, uint16_t func_id)
3609 {
3610 	struct mcx_dmamem mxm;
3611 	struct mcx_cmdq_entry *cqe;
3612 	struct mcx_cmd_manage_pages_in *in;
3613 	struct mcx_cmd_manage_pages_out *out;
3614 	unsigned int paslen, nmb, i, j, npages;
3615 	struct bus_dma_iter iter;
3616 	uint64_t *pas;
3617 	uint8_t status;
3618 	uint8_t token = mcx_cmdq_token(sc);
3619 	int error;
3620 
3621 	npages = mhm->mhm_npages;
3622 
3623 	paslen = sizeof(*pas) * npages;
3624 	nmb = howmany(paslen, MCX_CMDQ_MAILBOX_DATASIZE);
3625 
3626 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
3627 	mcx_cmdq_init(sc, cqe, sizeof(*in) + paslen, sizeof(*out), token);
3628 
3629 	in = mcx_cmdq_in(cqe);
3630 	in->cmd_opcode = htobe16(MCX_CMD_MANAGE_PAGES);
3631 	in->cmd_op_mod = htobe16(MCX_CMD_MANAGE_PAGES_ALLOC_SUCCESS);
3632 	in->cmd_func_id = func_id;
3633 	htobem32(&in->cmd_input_num_entries, npages);
3634 
3635 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, nmb,
3636 	    &cqe->cq_input_ptr, token) != 0) {
3637 		printf(", unable to allocate manage pages mailboxen\n");
3638 		return (-1);
3639 	}
3640 
3641 	bus_dma_iter_init(&iter, mhm->mhm_map);
3642 	for (i = 0; i < nmb; i++) {
3643 		unsigned int lim;
3644 
3645 		pas = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, i));
3646 		lim = min(MCX_CMDQ_MAILBOX_DATASIZE / sizeof(*pas), npages);
3647 
3648 		for (j = 0; j < lim; j++) {
3649 			htobem64(&pas[j], bus_dma_iter_addr(&iter));
3650 			bus_dma_iter_add(&iter, MCX_PAGE_SIZE);
3651 		}
3652 
3653 		npages -= lim;
3654 	}
3655 
3656 	mcx_cmdq_mboxes_sign(&mxm, nmb);
3657 
3658 	mcx_cmdq_post(sc, cqe, 0);
3659 	error = mcx_cmdq_poll(sc, cqe, 1000);
3660 	if (error != 0) {
3661 		printf(", manage pages timeout\n");
3662 		goto free;
3663 	}
3664 	error = mcx_cmdq_verify(cqe);
3665 	if (error != 0) {
3666 		printf(", manage pages reply corrupt\n");
3667 		goto free;
3668 	}
3669 
3670 	status = cqe->cq_output_data[0];
3671 	if (status != MCX_CQ_STATUS_OK) {
3672 		printf(", manage pages failed (%x)\n", status);
3673 		error = -1;
3674 		goto free;
3675 	}
3676 
3677 free:
3678 	mcx_dmamem_free(sc, &mxm);
3679 
3680 	return (error);
3681 }
3682 
3683 static int
3684 mcx_pages(struct mcx_softc *sc, struct mcx_hwmem *mhm, uint16_t type)
3685 {
3686 	int32_t npages;
3687 	uint16_t func_id;
3688 
3689 	if (mcx_query_pages(sc, type, &npages, &func_id) != 0) {
3690 		/* error printed by mcx_query_pages */
3691 		return (-1);
3692 	}
3693 
3694 	if (npages < 1)
3695 		return (0);
3696 
3697 	if (mcx_hwmem_alloc(sc, mhm, npages) != 0) {
3698 		printf(", unable to allocate hwmem\n");
3699 		return (-1);
3700 	}
3701 
3702 	if (mcx_add_pages(sc, mhm, func_id) != 0) {
3703 		printf(", unable to add hwmem\n");
3704 		goto free;
3705 	}
3706 
3707 	return (0);
3708 
3709 free:
3710 	mcx_hwmem_free(sc, mhm);
3711 
3712 	return (-1);
3713 }
3714 
3715 static int
3716 mcx_hca_max_caps(struct mcx_softc *sc)
3717 {
3718 	struct mcx_dmamem mxm;
3719 	struct mcx_cmdq_entry *cqe;
3720 	struct mcx_cmd_query_hca_cap_in *in;
3721 	struct mcx_cmd_query_hca_cap_out *out;
3722 	struct mcx_cmdq_mailbox *mb;
3723 	struct mcx_cap_device *hca;
3724 	uint8_t status;
3725 	uint8_t token = mcx_cmdq_token(sc);
3726 	int error;
3727 
3728 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
3729 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out) + MCX_HCA_CAP_LEN,
3730 	    token);
3731 
3732 	in = mcx_cmdq_in(cqe);
3733 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_HCA_CAP);
3734 	in->cmd_op_mod = htobe16(MCX_CMD_QUERY_HCA_CAP_MAX |
3735 	    MCX_CMD_QUERY_HCA_CAP_DEVICE);
3736 
3737 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, MCX_HCA_CAP_NMAILBOXES,
3738 	    &cqe->cq_output_ptr, token) != 0) {
3739 		printf(", unable to allocate query hca caps mailboxen\n");
3740 		return (-1);
3741 	}
3742 	mcx_cmdq_mboxes_sign(&mxm, MCX_HCA_CAP_NMAILBOXES);
3743 	mcx_cmdq_mboxes_sync(sc, &mxm, BUS_DMASYNC_PRERW);
3744 
3745 	mcx_cmdq_post(sc, cqe, 0);
3746 	error = mcx_cmdq_poll(sc, cqe, 1000);
3747 	mcx_cmdq_mboxes_sync(sc, &mxm, BUS_DMASYNC_POSTRW);
3748 
3749 	if (error != 0) {
3750 		printf(", query hca caps timeout\n");
3751 		goto free;
3752 	}
3753 	error = mcx_cmdq_verify(cqe);
3754 	if (error != 0) {
3755 		printf(", query hca caps reply corrupt\n");
3756 		goto free;
3757 	}
3758 
3759 	status = cqe->cq_output_data[0];
3760 	if (status != MCX_CQ_STATUS_OK) {
3761 		printf(", query hca caps failed (%x)\n", status);
3762 		error = -1;
3763 		goto free;
3764 	}
3765 
3766 	mb = mcx_cq_mbox(&mxm, 0);
3767 	hca = mcx_cq_mbox_data(mb);
3768 
3769 	if (hca->log_pg_sz > PAGE_SHIFT) {
3770 		printf(", minimum system page shift %u is too large\n",
3771 		    hca->log_pg_sz);
3772 		error = -1;
3773 		goto free;
3774 	}
3775 	/*
3776 	 * blueflame register is split into two buffers, and we must alternate
3777 	 * between the two of them.
3778 	 */
3779 	sc->sc_bf_size = (1 << hca->log_bf_reg_size) / 2;
3780 	sc->sc_max_rqt_size = (1 << hca->log_max_rqt_size);
3781 
3782 free:
3783 	mcx_dmamem_free(sc, &mxm);
3784 
3785 	return (error);
3786 }
3787 
3788 static int
3789 mcx_hca_set_caps(struct mcx_softc *sc)
3790 {
3791 	struct mcx_dmamem mxm;
3792 	struct mcx_cmdq_entry *cqe;
3793 	struct mcx_cmd_query_hca_cap_in *in;
3794 	struct mcx_cmd_query_hca_cap_out *out;
3795 	struct mcx_cmdq_mailbox *mb;
3796 	struct mcx_cap_device *hca;
3797 	uint8_t status;
3798 	uint8_t token = mcx_cmdq_token(sc);
3799 	int error;
3800 
3801 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
3802 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out) + MCX_HCA_CAP_LEN,
3803 	    token);
3804 
3805 	in = mcx_cmdq_in(cqe);
3806 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_HCA_CAP);
3807 	in->cmd_op_mod = htobe16(MCX_CMD_QUERY_HCA_CAP_CURRENT |
3808 	    MCX_CMD_QUERY_HCA_CAP_DEVICE);
3809 
3810 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, MCX_HCA_CAP_NMAILBOXES,
3811 	    &cqe->cq_output_ptr, token) != 0) {
3812 		printf(", unable to allocate manage pages mailboxen\n");
3813 		return (-1);
3814 	}
3815 	mcx_cmdq_mboxes_sign(&mxm, MCX_HCA_CAP_NMAILBOXES);
3816 	mcx_cmdq_mboxes_sync(sc, &mxm, BUS_DMASYNC_PRERW);
3817 
3818 	mcx_cmdq_post(sc, cqe, 0);
3819 	error = mcx_cmdq_poll(sc, cqe, 1000);
3820 	mcx_cmdq_mboxes_sync(sc, &mxm, BUS_DMASYNC_POSTRW);
3821 
3822 	if (error != 0) {
3823 		printf(", query hca caps timeout\n");
3824 		goto free;
3825 	}
3826 	error = mcx_cmdq_verify(cqe);
3827 	if (error != 0) {
3828 		printf(", query hca caps reply corrupt\n");
3829 		goto free;
3830 	}
3831 
3832 	status = cqe->cq_output_data[0];
3833 	if (status != MCX_CQ_STATUS_OK) {
3834 		printf(", query hca caps failed (%x)\n", status);
3835 		error = -1;
3836 		goto free;
3837 	}
3838 
3839 	mb = mcx_cq_mbox(&mxm, 0);
3840 	hca = mcx_cq_mbox_data(mb);
3841 
3842 	hca->log_pg_sz = PAGE_SHIFT;
3843 
3844 free:
3845 	mcx_dmamem_free(sc, &mxm);
3846 
3847 	return (error);
3848 }
3849 
3850 
3851 static int
3852 mcx_init_hca(struct mcx_softc *sc)
3853 {
3854 	struct mcx_cmdq_entry *cqe;
3855 	struct mcx_cmd_init_hca_in *in;
3856 	struct mcx_cmd_init_hca_out *out;
3857 	int error;
3858 	uint8_t status;
3859 
3860 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
3861 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
3862 
3863 	in = mcx_cmdq_in(cqe);
3864 	in->cmd_opcode = htobe16(MCX_CMD_INIT_HCA);
3865 	in->cmd_op_mod = htobe16(0);
3866 
3867 	mcx_cmdq_post(sc, cqe, 0);
3868 
3869 	error = mcx_cmdq_poll(sc, cqe, 1000);
3870 	if (error != 0) {
3871 		printf(", hca init timeout\n");
3872 		return (-1);
3873 	}
3874 	if (mcx_cmdq_verify(cqe) != 0) {
3875 		printf(", hca init command corrupt\n");
3876 		return (-1);
3877 	}
3878 
3879 	status = cqe->cq_output_data[0];
3880 	if (status != MCX_CQ_STATUS_OK) {
3881 		printf(", hca init failed (%x)\n", status);
3882 		return (-1);
3883 	}
3884 
3885 	return (0);
3886 }
3887 
3888 static int
3889 mcx_set_driver_version(struct mcx_softc *sc)
3890 {
3891 	struct mcx_dmamem mxm;
3892 	struct mcx_cmdq_entry *cqe;
3893 	struct mcx_cmd_set_driver_version_in *in;
3894 	struct mcx_cmd_set_driver_version_out *out;
3895 	int error;
3896 	int token;
3897 	uint8_t status;
3898 
3899 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
3900 	token = mcx_cmdq_token(sc);
3901 	mcx_cmdq_init(sc, cqe, sizeof(*in) +
3902 	    sizeof(struct mcx_cmd_set_driver_version), sizeof(*out), token);
3903 
3904 	in = mcx_cmdq_in(cqe);
3905 	in->cmd_opcode = htobe16(MCX_CMD_SET_DRIVER_VERSION);
3906 	in->cmd_op_mod = htobe16(0);
3907 
3908 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
3909 	    &cqe->cq_input_ptr, token) != 0) {
3910 		printf(", unable to allocate set driver version mailboxen\n");
3911 		return (-1);
3912 	}
3913 	strlcpy(mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)),
3914 	    "OpenBSD,mcx,1.000.000000", MCX_CMDQ_MAILBOX_DATASIZE);
3915 
3916 	mcx_cmdq_mboxes_sign(&mxm, 1);
3917 	mcx_cmdq_post(sc, cqe, 0);
3918 
3919 	error = mcx_cmdq_poll(sc, cqe, 1000);
3920 	if (error != 0) {
3921 		printf(", set driver version timeout\n");
3922 		goto free;
3923 	}
3924 	if (mcx_cmdq_verify(cqe) != 0) {
3925 		printf(", set driver version command corrupt\n");
3926 		goto free;
3927 	}
3928 
3929 	status = cqe->cq_output_data[0];
3930 	if (status != MCX_CQ_STATUS_OK) {
3931 		printf(", set driver version failed (%x)\n", status);
3932 		error = -1;
3933 		goto free;
3934 	}
3935 
3936 free:
3937 	mcx_dmamem_free(sc, &mxm);
3938 
3939 	return (error);
3940 }
3941 
3942 static int
3943 mcx_iff(struct mcx_softc *sc)
3944 {
3945 	struct ifnet *ifp = &sc->sc_ac.ac_if;
3946 	struct mcx_dmamem mxm;
3947 	struct mcx_cmdq_entry *cqe;
3948 	struct mcx_cmd_modify_nic_vport_context_in *in;
3949 	struct mcx_cmd_modify_nic_vport_context_out *out;
3950 	struct mcx_nic_vport_ctx *ctx;
3951 	int error;
3952 	int token;
3953 	int insize;
3954 	uint32_t dest;
3955 
3956 	dest = MCX_FLOW_CONTEXT_DEST_TYPE_TABLE |
3957 	    sc->sc_rss_flow_table_id;
3958 
3959 	/* enable or disable the promisc flow */
3960 	if (ISSET(ifp->if_flags, IFF_PROMISC)) {
3961 		if (sc->sc_promisc_flow_enabled == 0) {
3962 			mcx_set_flow_table_entry_mac(sc,
3963 			    MCX_FLOW_GROUP_PROMISC, 0, NULL, dest);
3964 			sc->sc_promisc_flow_enabled = 1;
3965 		}
3966 	} else if (sc->sc_promisc_flow_enabled != 0) {
3967 		mcx_delete_flow_table_entry(sc, MCX_FLOW_GROUP_PROMISC, 0);
3968 		sc->sc_promisc_flow_enabled = 0;
3969 	}
3970 
3971 	/* enable or disable the all-multicast flow */
3972 	if (ISSET(ifp->if_flags, IFF_ALLMULTI)) {
3973 		if (sc->sc_allmulti_flow_enabled == 0) {
3974 			uint8_t mcast[ETHER_ADDR_LEN];
3975 
3976 			memset(mcast, 0, sizeof(mcast));
3977 			mcast[0] = 0x01;
3978 			mcx_set_flow_table_entry_mac(sc,
3979 			    MCX_FLOW_GROUP_ALLMULTI, 0, mcast, dest);
3980 			sc->sc_allmulti_flow_enabled = 1;
3981 		}
3982 	} else if (sc->sc_allmulti_flow_enabled != 0) {
3983 		mcx_delete_flow_table_entry(sc, MCX_FLOW_GROUP_ALLMULTI, 0);
3984 		sc->sc_allmulti_flow_enabled = 0;
3985 	}
3986 
3987 	insize = sizeof(struct mcx_nic_vport_ctx) + 240;
3988 
3989 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
3990 	token = mcx_cmdq_token(sc);
3991 	mcx_cmdq_init(sc, cqe, sizeof(*in) + insize, sizeof(*out), token);
3992 
3993 	in = mcx_cmdq_in(cqe);
3994 	in->cmd_opcode = htobe16(MCX_CMD_MODIFY_NIC_VPORT_CONTEXT);
3995 	in->cmd_op_mod = htobe16(0);
3996 	in->cmd_field_select = htobe32(
3997 	    MCX_CMD_MODIFY_NIC_VPORT_CONTEXT_FIELD_PROMISC |
3998 	    MCX_CMD_MODIFY_NIC_VPORT_CONTEXT_FIELD_MTU);
3999 
4000 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1, &cqe->cq_input_ptr, token) != 0) {
4001 		printf(", unable to allocate modify "
4002 		    "nic vport context mailboxen\n");
4003 		return (-1);
4004 	}
4005 	ctx = (struct mcx_nic_vport_ctx *)
4006 	    (((char *)mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0))) + 240);
4007 	ctx->vp_mtu = htobe32(sc->sc_hardmtu);
4008 	/*
4009          * always leave promisc-all enabled on the vport since we
4010          * can't give it a vlan list, and we're already doing multicast
4011          * filtering in the flow table.
4012 	 */
4013 	ctx->vp_flags = htobe16(MCX_NIC_VPORT_CTX_PROMISC_ALL);
4014 
4015 	mcx_cmdq_mboxes_sign(&mxm, 1);
4016 	mcx_cmdq_post(sc, cqe, 0);
4017 
4018 	error = mcx_cmdq_poll(sc, cqe, 1000);
4019 	if (error != 0) {
4020 		printf(", modify nic vport context timeout\n");
4021 		goto free;
4022 	}
4023 	if (mcx_cmdq_verify(cqe) != 0) {
4024 		printf(", modify nic vport context command corrupt\n");
4025 		goto free;
4026 	}
4027 
4028 	out = mcx_cmdq_out(cqe);
4029 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4030 		printf(", modify nic vport context failed (%x, %x)\n",
4031 		    out->cmd_status, betoh32(out->cmd_syndrome));
4032 		error = -1;
4033 		goto free;
4034 	}
4035 
4036 free:
4037 	mcx_dmamem_free(sc, &mxm);
4038 
4039 	return (error);
4040 }
4041 
4042 static int
4043 mcx_alloc_uar(struct mcx_softc *sc, int *uar)
4044 {
4045 	struct mcx_cmdq_entry *cqe;
4046 	struct mcx_cmd_alloc_uar_in *in;
4047 	struct mcx_cmd_alloc_uar_out *out;
4048 	int error;
4049 
4050 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4051 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
4052 
4053 	in = mcx_cmdq_in(cqe);
4054 	in->cmd_opcode = htobe16(MCX_CMD_ALLOC_UAR);
4055 	in->cmd_op_mod = htobe16(0);
4056 
4057 	mcx_cmdq_post(sc, cqe, 0);
4058 
4059 	error = mcx_cmdq_poll(sc, cqe, 1000);
4060 	if (error != 0) {
4061 		printf(", alloc uar timeout\n");
4062 		return (-1);
4063 	}
4064 	if (mcx_cmdq_verify(cqe) != 0) {
4065 		printf(", alloc uar command corrupt\n");
4066 		return (-1);
4067 	}
4068 
4069 	out = mcx_cmdq_out(cqe);
4070 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4071 		printf(", alloc uar failed (%x)\n", out->cmd_status);
4072 		return (-1);
4073 	}
4074 
4075 	*uar = mcx_get_id(out->cmd_uar);
4076 	return (0);
4077 }
4078 
4079 static int
4080 mcx_create_eq(struct mcx_softc *sc, struct mcx_eq *eq, int uar,
4081     uint64_t events, int vector)
4082 {
4083 	struct mcx_cmdq_entry *cqe;
4084 	struct mcx_dmamem mxm;
4085 	struct mcx_cmd_create_eq_in *in;
4086 	struct mcx_cmd_create_eq_mb_in *mbin;
4087 	struct mcx_cmd_create_eq_out *out;
4088 	struct mcx_eq_entry *eqe;
4089 	int error;
4090 	uint64_t *pas;
4091 	int insize, npages, paslen, i, token;
4092 
4093 	eq->eq_cons = 0;
4094 
4095 	npages = howmany((1 << MCX_LOG_EQ_SIZE) * sizeof(struct mcx_eq_entry),
4096 	    MCX_PAGE_SIZE);
4097 	paslen = npages * sizeof(*pas);
4098 	insize = sizeof(struct mcx_cmd_create_eq_mb_in) + paslen;
4099 
4100 	if (mcx_dmamem_alloc(sc, &eq->eq_mem, npages * MCX_PAGE_SIZE,
4101 	    MCX_PAGE_SIZE) != 0) {
4102 		printf(", unable to allocate event queue memory\n");
4103 		return (-1);
4104 	}
4105 
4106 	eqe = (struct mcx_eq_entry *)MCX_DMA_KVA(&eq->eq_mem);
4107 	for (i = 0; i < (1 << MCX_LOG_EQ_SIZE); i++) {
4108 		eqe[i].eq_owner = MCX_EQ_ENTRY_OWNER_INIT;
4109 	}
4110 
4111 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4112 	token = mcx_cmdq_token(sc);
4113 	mcx_cmdq_init(sc, cqe, sizeof(*in) + insize, sizeof(*out), token);
4114 
4115 	in = mcx_cmdq_in(cqe);
4116 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_EQ);
4117 	in->cmd_op_mod = htobe16(0);
4118 
4119 	if (mcx_cmdq_mboxes_alloc(sc, &mxm,
4120 	    howmany(insize, MCX_CMDQ_MAILBOX_DATASIZE),
4121 	    &cqe->cq_input_ptr, token) != 0) {
4122 		printf(", unable to allocate create eq mailboxen\n");
4123 		return (-1);
4124 	}
4125 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
4126 	mbin->cmd_eq_ctx.eq_uar_size = htobe32(
4127 	    (MCX_LOG_EQ_SIZE << MCX_EQ_CTX_LOG_EQ_SIZE_SHIFT) | uar);
4128 	mbin->cmd_eq_ctx.eq_intr = vector;
4129 	mbin->cmd_event_bitmask = htobe64(events);
4130 
4131 	/* physical addresses follow the mailbox in data */
4132 	mcx_cmdq_mboxes_pas(&mxm, sizeof(*mbin), npages, &eq->eq_mem);
4133 	mcx_cmdq_mboxes_sign(&mxm, howmany(insize, MCX_CMDQ_MAILBOX_DATASIZE));
4134 	mcx_cmdq_post(sc, cqe, 0);
4135 
4136 	error = mcx_cmdq_poll(sc, cqe, 1000);
4137 	if (error != 0) {
4138 		printf(", create eq timeout\n");
4139 		goto free;
4140 	}
4141 	if (mcx_cmdq_verify(cqe) != 0) {
4142 		printf(", create eq command corrupt\n");
4143 		goto free;
4144 	}
4145 
4146 	out = mcx_cmdq_out(cqe);
4147 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4148 		printf(", create eq failed (%x, %x)\n", out->cmd_status,
4149 		    betoh32(out->cmd_syndrome));
4150 		error = -1;
4151 		goto free;
4152 	}
4153 
4154 	eq->eq_n = mcx_get_id(out->cmd_eqn);
4155 	mcx_arm_eq(sc, eq, uar);
4156 free:
4157 	mcx_dmamem_free(sc, &mxm);
4158 	return (error);
4159 }
4160 
4161 static int
4162 mcx_alloc_pd(struct mcx_softc *sc)
4163 {
4164 	struct mcx_cmdq_entry *cqe;
4165 	struct mcx_cmd_alloc_pd_in *in;
4166 	struct mcx_cmd_alloc_pd_out *out;
4167 	int error;
4168 
4169 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4170 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
4171 
4172 	in = mcx_cmdq_in(cqe);
4173 	in->cmd_opcode = htobe16(MCX_CMD_ALLOC_PD);
4174 	in->cmd_op_mod = htobe16(0);
4175 
4176 	mcx_cmdq_post(sc, cqe, 0);
4177 
4178 	error = mcx_cmdq_poll(sc, cqe, 1000);
4179 	if (error != 0) {
4180 		printf(", alloc pd timeout\n");
4181 		return (-1);
4182 	}
4183 	if (mcx_cmdq_verify(cqe) != 0) {
4184 		printf(", alloc pd command corrupt\n");
4185 		return (-1);
4186 	}
4187 
4188 	out = mcx_cmdq_out(cqe);
4189 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4190 		printf(", alloc pd failed (%x)\n", out->cmd_status);
4191 		return (-1);
4192 	}
4193 
4194 	sc->sc_pd = mcx_get_id(out->cmd_pd);
4195 	return (0);
4196 }
4197 
4198 static int
4199 mcx_alloc_tdomain(struct mcx_softc *sc)
4200 {
4201 	struct mcx_cmdq_entry *cqe;
4202 	struct mcx_cmd_alloc_td_in *in;
4203 	struct mcx_cmd_alloc_td_out *out;
4204 	int error;
4205 
4206 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4207 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
4208 
4209 	in = mcx_cmdq_in(cqe);
4210 	in->cmd_opcode = htobe16(MCX_CMD_ALLOC_TRANSPORT_DOMAIN);
4211 	in->cmd_op_mod = htobe16(0);
4212 
4213 	mcx_cmdq_post(sc, cqe, 0);
4214 
4215 	error = mcx_cmdq_poll(sc, cqe, 1000);
4216 	if (error != 0) {
4217 		printf(", alloc transport domain timeout\n");
4218 		return (-1);
4219 	}
4220 	if (mcx_cmdq_verify(cqe) != 0) {
4221 		printf(", alloc transport domain command corrupt\n");
4222 		return (-1);
4223 	}
4224 
4225 	out = mcx_cmdq_out(cqe);
4226 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4227 		printf(", alloc transport domain failed (%x)\n",
4228 		    out->cmd_status);
4229 		return (-1);
4230 	}
4231 
4232 	sc->sc_tdomain = mcx_get_id(out->cmd_tdomain);
4233 	return (0);
4234 }
4235 
4236 static int
4237 mcx_query_nic_vport_context(struct mcx_softc *sc)
4238 {
4239 	struct mcx_dmamem mxm;
4240 	struct mcx_cmdq_entry *cqe;
4241 	struct mcx_cmd_query_nic_vport_context_in *in;
4242 	struct mcx_cmd_query_nic_vport_context_out *out;
4243 	struct mcx_nic_vport_ctx *ctx;
4244 	uint8_t *addr;
4245 	int error, token, i;
4246 
4247 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4248 	token = mcx_cmdq_token(sc);
4249 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out) + sizeof(*ctx), token);
4250 
4251 	in = mcx_cmdq_in(cqe);
4252 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_NIC_VPORT_CONTEXT);
4253 	in->cmd_op_mod = htobe16(0);
4254 	in->cmd_allowed_list_type = 0;
4255 
4256 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
4257 	    &cqe->cq_output_ptr, token) != 0) {
4258 		printf(", unable to allocate "
4259 		    "query nic vport context mailboxen\n");
4260 		return (-1);
4261 	}
4262 	mcx_cmdq_mboxes_sign(&mxm, 1);
4263 	mcx_cmdq_post(sc, cqe, 0);
4264 
4265 	error = mcx_cmdq_poll(sc, cqe, 1000);
4266 	if (error != 0) {
4267 		printf(", query nic vport context timeout\n");
4268 		goto free;
4269 	}
4270 	if (mcx_cmdq_verify(cqe) != 0) {
4271 		printf(", query nic vport context command corrupt\n");
4272 		goto free;
4273 	}
4274 
4275 	out = mcx_cmdq_out(cqe);
4276 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4277 		printf(", query nic vport context failed (%x, %x)\n",
4278 		    out->cmd_status, betoh32(out->cmd_syndrome));
4279 		error = -1;
4280 		goto free;
4281 	}
4282 
4283 	ctx = (struct mcx_nic_vport_ctx *)
4284 	    mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
4285 	addr = (uint8_t *)&ctx->vp_perm_addr;
4286 	for (i = 0; i < ETHER_ADDR_LEN; i++) {
4287 		sc->sc_ac.ac_enaddr[i] = addr[i + 2];
4288 	}
4289 free:
4290 	mcx_dmamem_free(sc, &mxm);
4291 
4292 	return (error);
4293 }
4294 
4295 static int
4296 mcx_query_special_contexts(struct mcx_softc *sc)
4297 {
4298 	struct mcx_cmdq_entry *cqe;
4299 	struct mcx_cmd_query_special_ctx_in *in;
4300 	struct mcx_cmd_query_special_ctx_out *out;
4301 	int error;
4302 
4303 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4304 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
4305 
4306 	in = mcx_cmdq_in(cqe);
4307 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_SPECIAL_CONTEXTS);
4308 	in->cmd_op_mod = htobe16(0);
4309 
4310 	mcx_cmdq_post(sc, cqe, 0);
4311 
4312 	error = mcx_cmdq_poll(sc, cqe, 1000);
4313 	if (error != 0) {
4314 		printf(", query special contexts timeout\n");
4315 		return (-1);
4316 	}
4317 	if (mcx_cmdq_verify(cqe) != 0) {
4318 		printf(", query special contexts command corrupt\n");
4319 		return (-1);
4320 	}
4321 
4322 	out = mcx_cmdq_out(cqe);
4323 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4324 		printf(", query special contexts failed (%x)\n",
4325 		    out->cmd_status);
4326 		return (-1);
4327 	}
4328 
4329 	sc->sc_lkey = betoh32(out->cmd_resd_lkey);
4330 	return (0);
4331 }
4332 
4333 static int
4334 mcx_set_port_mtu(struct mcx_softc *sc, int mtu)
4335 {
4336 	struct mcx_reg_pmtu pmtu;
4337 	int error;
4338 
4339 	/* read max mtu */
4340 	memset(&pmtu, 0, sizeof(pmtu));
4341 	pmtu.rp_local_port = 1;
4342 	error = mcx_access_hca_reg(sc, MCX_REG_PMTU, MCX_REG_OP_READ, &pmtu,
4343 	    sizeof(pmtu));
4344 	if (error != 0) {
4345 		printf(", unable to get port MTU\n");
4346 		return error;
4347 	}
4348 
4349 	mtu = min(mtu, betoh16(pmtu.rp_max_mtu));
4350 	pmtu.rp_admin_mtu = htobe16(mtu);
4351 	error = mcx_access_hca_reg(sc, MCX_REG_PMTU, MCX_REG_OP_WRITE, &pmtu,
4352 	    sizeof(pmtu));
4353 	if (error != 0) {
4354 		printf(", unable to set port MTU\n");
4355 		return error;
4356 	}
4357 
4358 	sc->sc_hardmtu = mtu;
4359 	sc->sc_rxbufsz = roundup(mtu + ETHER_ALIGN, sizeof(long));
4360 	return 0;
4361 }
4362 
4363 static int
4364 mcx_create_cq(struct mcx_softc *sc, struct mcx_cq *cq, int uar, int db, int eqn)
4365 {
4366 	struct mcx_cmdq_entry *cmde;
4367 	struct mcx_cq_entry *cqe;
4368 	struct mcx_dmamem mxm;
4369 	struct mcx_cmd_create_cq_in *in;
4370 	struct mcx_cmd_create_cq_mb_in *mbin;
4371 	struct mcx_cmd_create_cq_out *out;
4372 	int error;
4373 	uint64_t *pas;
4374 	int insize, npages, paslen, i, token;
4375 
4376 	npages = howmany((1 << MCX_LOG_CQ_SIZE) * sizeof(struct mcx_cq_entry),
4377 	    MCX_PAGE_SIZE);
4378 	paslen = npages * sizeof(*pas);
4379 	insize = sizeof(struct mcx_cmd_create_cq_mb_in) + paslen;
4380 
4381 	if (mcx_dmamem_alloc(sc, &cq->cq_mem, npages * MCX_PAGE_SIZE,
4382 	    MCX_PAGE_SIZE) != 0) {
4383 		printf("%s: unable to allocate completion queue memory\n",
4384 		    DEVNAME(sc));
4385 		return (-1);
4386 	}
4387 	cqe = MCX_DMA_KVA(&cq->cq_mem);
4388 	for (i = 0; i < (1 << MCX_LOG_CQ_SIZE); i++) {
4389 		cqe[i].cq_opcode_owner = MCX_CQ_ENTRY_FLAG_OWNER;
4390 	}
4391 
4392 	cmde = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4393 	token = mcx_cmdq_token(sc);
4394 	mcx_cmdq_init(sc, cmde, sizeof(*in) + insize, sizeof(*out), token);
4395 
4396 	in = mcx_cmdq_in(cmde);
4397 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_CQ);
4398 	in->cmd_op_mod = htobe16(0);
4399 
4400 	if (mcx_cmdq_mboxes_alloc(sc, &mxm,
4401 	    howmany(insize, MCX_CMDQ_MAILBOX_DATASIZE),
4402 	    &cmde->cq_input_ptr, token) != 0) {
4403 		printf("%s: unable to allocate create cq mailboxen\n",
4404 		    DEVNAME(sc));
4405 		error = -1;
4406 		goto free;
4407 	}
4408 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
4409 	mbin->cmd_cq_ctx.cq_uar_size = htobe32(
4410 	    (MCX_LOG_CQ_SIZE << MCX_CQ_CTX_LOG_CQ_SIZE_SHIFT) | uar);
4411 	mbin->cmd_cq_ctx.cq_eqn = htobe32(eqn);
4412 	mbin->cmd_cq_ctx.cq_period_max_count = htobe32(
4413 	    (MCX_CQ_MOD_PERIOD << MCX_CQ_CTX_PERIOD_SHIFT) |
4414 	    MCX_CQ_MOD_COUNTER);
4415 	mbin->cmd_cq_ctx.cq_doorbell = htobe64(
4416 	    MCX_DMA_DVA(&sc->sc_doorbell_mem) +
4417 	    MCX_CQ_DOORBELL_BASE + (MCX_CQ_DOORBELL_STRIDE * db));
4418 
4419 	/* physical addresses follow the mailbox in data */
4420 	mcx_cmdq_mboxes_pas(&mxm, sizeof(*mbin), npages, &cq->cq_mem);
4421 	mcx_cmdq_post(sc, cmde, 0);
4422 
4423 	error = mcx_cmdq_poll(sc, cmde, 1000);
4424 	if (error != 0) {
4425 		printf("%s: create cq timeout\n", DEVNAME(sc));
4426 		goto free;
4427 	}
4428 	if (mcx_cmdq_verify(cmde) != 0) {
4429 		printf("%s: create cq command corrupt\n", DEVNAME(sc));
4430 		goto free;
4431 	}
4432 
4433 	out = mcx_cmdq_out(cmde);
4434 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4435 		printf("%s: create cq failed (%x, %x)\n", DEVNAME(sc),
4436 		    out->cmd_status, betoh32(out->cmd_syndrome));
4437 		error = -1;
4438 		goto free;
4439 	}
4440 
4441 	cq->cq_n = mcx_get_id(out->cmd_cqn);
4442 	cq->cq_cons = 0;
4443 	cq->cq_count = 0;
4444 	cq->cq_doorbell = MCX_DMA_KVA(&sc->sc_doorbell_mem) +
4445 	    MCX_CQ_DOORBELL_BASE + (MCX_CQ_DOORBELL_STRIDE * db);
4446 	mcx_arm_cq(sc, cq, uar);
4447 
4448 free:
4449 	mcx_dmamem_free(sc, &mxm);
4450 	return (error);
4451 }
4452 
4453 static int
4454 mcx_destroy_cq(struct mcx_softc *sc, struct mcx_cq *cq)
4455 {
4456 	struct mcx_cmdq_entry *cqe;
4457 	struct mcx_cmd_destroy_cq_in *in;
4458 	struct mcx_cmd_destroy_cq_out *out;
4459 	int error;
4460 	int token;
4461 
4462 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4463 	token = mcx_cmdq_token(sc);
4464 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), token);
4465 
4466 	in = mcx_cmdq_in(cqe);
4467 	in->cmd_opcode = htobe16(MCX_CMD_DESTROY_CQ);
4468 	in->cmd_op_mod = htobe16(0);
4469 	in->cmd_cqn = htobe32(cq->cq_n);
4470 
4471 	mcx_cmdq_post(sc, cqe, 0);
4472 	error = mcx_cmdq_poll(sc, cqe, 1000);
4473 	if (error != 0) {
4474 		printf("%s: destroy cq timeout\n", DEVNAME(sc));
4475 		return error;
4476 	}
4477 	if (mcx_cmdq_verify(cqe) != 0) {
4478 		printf("%s: destroy cq command corrupt\n", DEVNAME(sc));
4479 		return error;
4480 	}
4481 
4482 	out = mcx_cmdq_out(cqe);
4483 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4484 		printf("%s: destroy cq failed (%x, %x)\n", DEVNAME(sc),
4485 		    out->cmd_status, betoh32(out->cmd_syndrome));
4486 		return -1;
4487 	}
4488 
4489 	cq->cq_n = 0;
4490 	mcx_dmamem_free(sc, &cq->cq_mem);
4491 	cq->cq_cons = 0;
4492 	cq->cq_count = 0;
4493 	return 0;
4494 }
4495 
4496 static int
4497 mcx_create_rq(struct mcx_softc *sc, struct mcx_rx *rx, int db, int cqn)
4498 {
4499 	struct mcx_cmdq_entry *cqe;
4500 	struct mcx_dmamem mxm;
4501 	struct mcx_cmd_create_rq_in *in;
4502 	struct mcx_cmd_create_rq_out *out;
4503 	struct mcx_rq_ctx *mbin;
4504 	int error;
4505 	uint64_t *pas;
4506 	uint32_t rq_flags;
4507 	uint8_t *doorbell;
4508 	int insize, npages, paslen, token;
4509 
4510 	npages = howmany((1 << MCX_LOG_RQ_SIZE) * sizeof(struct mcx_rq_entry),
4511 	    MCX_PAGE_SIZE);
4512 	paslen = npages * sizeof(*pas);
4513 	insize = 0x10 + sizeof(struct mcx_rq_ctx) + paslen;
4514 
4515 	if (mcx_dmamem_alloc(sc, &rx->rx_rq_mem, npages * MCX_PAGE_SIZE,
4516 	    MCX_PAGE_SIZE) != 0) {
4517 		printf("%s: unable to allocate receive queue memory\n",
4518 		    DEVNAME(sc));
4519 		return (-1);
4520 	}
4521 
4522 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4523 	token = mcx_cmdq_token(sc);
4524 	mcx_cmdq_init(sc, cqe, sizeof(*in) + insize, sizeof(*out), token);
4525 
4526 	in = mcx_cmdq_in(cqe);
4527 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_RQ);
4528 	in->cmd_op_mod = htobe16(0);
4529 
4530 	if (mcx_cmdq_mboxes_alloc(sc, &mxm,
4531 	    howmany(insize, MCX_CMDQ_MAILBOX_DATASIZE),
4532 	    &cqe->cq_input_ptr, token) != 0) {
4533 		printf("%s: unable to allocate create rq mailboxen\n",
4534 		    DEVNAME(sc));
4535 		error = -1;
4536 		goto free;
4537 	}
4538 	mbin = (struct mcx_rq_ctx *)
4539 	    (((char *)mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0))) + 0x10);
4540 	rq_flags = MCX_RQ_CTX_RLKEY;
4541 #if NVLAN == 0
4542 	rq_flags |= MCX_RQ_CTX_VLAN_STRIP_DIS;
4543 #endif
4544 	mbin->rq_flags = htobe32(rq_flags);
4545 	mbin->rq_cqn = htobe32(cqn);
4546 	mbin->rq_wq.wq_type = MCX_WQ_CTX_TYPE_CYCLIC;
4547 	mbin->rq_wq.wq_pd = htobe32(sc->sc_pd);
4548 	mbin->rq_wq.wq_doorbell = htobe64(MCX_DMA_DVA(&sc->sc_doorbell_mem) +
4549 	    MCX_WQ_DOORBELL_BASE + (db * MCX_WQ_DOORBELL_STRIDE));
4550 	mbin->rq_wq.wq_log_stride = htobe16(4);
4551 	mbin->rq_wq.wq_log_size = MCX_LOG_RQ_SIZE;
4552 
4553 	/* physical addresses follow the mailbox in data */
4554 	mcx_cmdq_mboxes_pas(&mxm, sizeof(*mbin) + 0x10, npages, &rx->rx_rq_mem);
4555 	mcx_cmdq_post(sc, cqe, 0);
4556 
4557 	error = mcx_cmdq_poll(sc, cqe, 1000);
4558 	if (error != 0) {
4559 		printf("%s: create rq timeout\n", DEVNAME(sc));
4560 		goto free;
4561 	}
4562 	if (mcx_cmdq_verify(cqe) != 0) {
4563 		printf("%s: create rq command corrupt\n", DEVNAME(sc));
4564 		goto free;
4565 	}
4566 
4567 	out = mcx_cmdq_out(cqe);
4568 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4569 		printf("%s: create rq failed (%x, %x)\n", DEVNAME(sc),
4570 		    out->cmd_status, betoh32(out->cmd_syndrome));
4571 		error = -1;
4572 		goto free;
4573 	}
4574 
4575 	rx->rx_rqn = mcx_get_id(out->cmd_rqn);
4576 
4577 	doorbell = MCX_DMA_KVA(&sc->sc_doorbell_mem);
4578 	rx->rx_doorbell = (uint32_t *)(doorbell + MCX_WQ_DOORBELL_BASE +
4579 	    (db * MCX_WQ_DOORBELL_STRIDE));
4580 
4581 free:
4582 	mcx_dmamem_free(sc, &mxm);
4583 	return (error);
4584 }
4585 
4586 static int
4587 mcx_ready_rq(struct mcx_softc *sc, struct mcx_rx *rx)
4588 {
4589 	struct mcx_cmdq_entry *cqe;
4590 	struct mcx_dmamem mxm;
4591 	struct mcx_cmd_modify_rq_in *in;
4592 	struct mcx_cmd_modify_rq_mb_in *mbin;
4593 	struct mcx_cmd_modify_rq_out *out;
4594 	int error;
4595 	int token;
4596 
4597 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4598 	token = mcx_cmdq_token(sc);
4599 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
4600 	    sizeof(*out), token);
4601 
4602 	in = mcx_cmdq_in(cqe);
4603 	in->cmd_opcode = htobe16(MCX_CMD_MODIFY_RQ);
4604 	in->cmd_op_mod = htobe16(0);
4605 	in->cmd_rq_state = htobe32((MCX_QUEUE_STATE_RST << 28) | rx->rx_rqn);
4606 
4607 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
4608 	    &cqe->cq_input_ptr, token) != 0) {
4609 		printf("%s: unable to allocate modify rq mailbox\n",
4610 		    DEVNAME(sc));
4611 		return (-1);
4612 	}
4613 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
4614 	mbin->cmd_rq_ctx.rq_flags = htobe32(
4615 	    MCX_QUEUE_STATE_RDY << MCX_RQ_CTX_STATE_SHIFT);
4616 
4617 	mcx_cmdq_mboxes_sign(&mxm, 1);
4618 	mcx_cmdq_post(sc, cqe, 0);
4619 	error = mcx_cmdq_poll(sc, cqe, 1000);
4620 	if (error != 0) {
4621 		printf("%s: modify rq timeout\n", DEVNAME(sc));
4622 		goto free;
4623 	}
4624 	if (mcx_cmdq_verify(cqe) != 0) {
4625 		printf("%s: modify rq command corrupt\n", DEVNAME(sc));
4626 		goto free;
4627 	}
4628 
4629 	out = mcx_cmdq_out(cqe);
4630 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4631 		printf("%s: modify rq failed (%x, %x)\n", DEVNAME(sc),
4632 		    out->cmd_status, betoh32(out->cmd_syndrome));
4633 		error = -1;
4634 		goto free;
4635 	}
4636 
4637 free:
4638 	mcx_dmamem_free(sc, &mxm);
4639 	return (error);
4640 }
4641 
4642 static int
4643 mcx_destroy_rq(struct mcx_softc *sc, struct mcx_rx *rx)
4644 {
4645 	struct mcx_cmdq_entry *cqe;
4646 	struct mcx_cmd_destroy_rq_in *in;
4647 	struct mcx_cmd_destroy_rq_out *out;
4648 	int error;
4649 	int token;
4650 
4651 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4652 	token = mcx_cmdq_token(sc);
4653 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), token);
4654 
4655 	in = mcx_cmdq_in(cqe);
4656 	in->cmd_opcode = htobe16(MCX_CMD_DESTROY_RQ);
4657 	in->cmd_op_mod = htobe16(0);
4658 	in->cmd_rqn = htobe32(rx->rx_rqn);
4659 
4660 	mcx_cmdq_post(sc, cqe, 0);
4661 	error = mcx_cmdq_poll(sc, cqe, 1000);
4662 	if (error != 0) {
4663 		printf("%s: destroy rq timeout\n", DEVNAME(sc));
4664 		return error;
4665 	}
4666 	if (mcx_cmdq_verify(cqe) != 0) {
4667 		printf("%s: destroy rq command corrupt\n", DEVNAME(sc));
4668 		return error;
4669 	}
4670 
4671 	out = mcx_cmdq_out(cqe);
4672 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4673 		printf("%s: destroy rq failed (%x, %x)\n", DEVNAME(sc),
4674 		    out->cmd_status, betoh32(out->cmd_syndrome));
4675 		return -1;
4676 	}
4677 
4678 	rx->rx_rqn = 0;
4679 	return 0;
4680 }
4681 
4682 static int
4683 mcx_create_tir_direct(struct mcx_softc *sc, struct mcx_rx *rx, int *tirn)
4684 {
4685 	struct mcx_cmdq_entry *cqe;
4686 	struct mcx_dmamem mxm;
4687 	struct mcx_cmd_create_tir_in *in;
4688 	struct mcx_cmd_create_tir_mb_in *mbin;
4689 	struct mcx_cmd_create_tir_out *out;
4690 	int error;
4691 	int token;
4692 
4693 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4694 	token = mcx_cmdq_token(sc);
4695 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
4696 	    sizeof(*out), token);
4697 
4698 	in = mcx_cmdq_in(cqe);
4699 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_TIR);
4700 	in->cmd_op_mod = htobe16(0);
4701 
4702 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
4703 	    &cqe->cq_input_ptr, token) != 0) {
4704 		printf("%s: unable to allocate create tir mailbox\n",
4705 		    DEVNAME(sc));
4706 		return (-1);
4707 	}
4708 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
4709 	/* leave disp_type = 0, so packets get sent to the inline rqn */
4710 	mbin->cmd_inline_rqn = htobe32(rx->rx_rqn);
4711 	mbin->cmd_tdomain = htobe32(sc->sc_tdomain);
4712 
4713 	mcx_cmdq_post(sc, cqe, 0);
4714 	error = mcx_cmdq_poll(sc, cqe, 1000);
4715 	if (error != 0) {
4716 		printf("%s: create tir timeout\n", DEVNAME(sc));
4717 		goto free;
4718 	}
4719 	if (mcx_cmdq_verify(cqe) != 0) {
4720 		printf("%s: create tir command corrupt\n", DEVNAME(sc));
4721 		goto free;
4722 	}
4723 
4724 	out = mcx_cmdq_out(cqe);
4725 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4726 		printf("%s: create tir failed (%x, %x)\n", DEVNAME(sc),
4727 		    out->cmd_status, betoh32(out->cmd_syndrome));
4728 		error = -1;
4729 		goto free;
4730 	}
4731 
4732 	*tirn = mcx_get_id(out->cmd_tirn);
4733 free:
4734 	mcx_dmamem_free(sc, &mxm);
4735 	return (error);
4736 }
4737 
4738 static int
4739 mcx_create_tir_indirect(struct mcx_softc *sc, int rqtn, uint32_t hash_sel,
4740     int *tirn)
4741 {
4742 	struct mcx_cmdq_entry *cqe;
4743 	struct mcx_dmamem mxm;
4744 	struct mcx_cmd_create_tir_in *in;
4745 	struct mcx_cmd_create_tir_mb_in *mbin;
4746 	struct mcx_cmd_create_tir_out *out;
4747 	int error;
4748 	int token;
4749 
4750 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4751 	token = mcx_cmdq_token(sc);
4752 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
4753 	    sizeof(*out), token);
4754 
4755 	in = mcx_cmdq_in(cqe);
4756 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_TIR);
4757 	in->cmd_op_mod = htobe16(0);
4758 
4759 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
4760 	    &cqe->cq_input_ptr, token) != 0) {
4761 		printf("%s: unable to allocate create tir mailbox\n",
4762 		    DEVNAME(sc));
4763 		return (-1);
4764 	}
4765 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
4766 	mbin->cmd_disp_type = htobe32(MCX_TIR_CTX_DISP_TYPE_INDIRECT
4767 	    << MCX_TIR_CTX_DISP_TYPE_SHIFT);
4768 	mbin->cmd_indir_table = htobe32(rqtn);
4769 	mbin->cmd_tdomain = htobe32(sc->sc_tdomain |
4770 	    MCX_TIR_CTX_HASH_TOEPLITZ << MCX_TIR_CTX_HASH_SHIFT);
4771 	mbin->cmd_rx_hash_sel_outer = htobe32(hash_sel);
4772 	stoeplitz_to_key(&mbin->cmd_rx_hash_key,
4773 	    sizeof(mbin->cmd_rx_hash_key));
4774 
4775 	mcx_cmdq_post(sc, cqe, 0);
4776 	error = mcx_cmdq_poll(sc, cqe, 1000);
4777 	if (error != 0) {
4778 		printf("%s: create tir timeout\n", DEVNAME(sc));
4779 		goto free;
4780 	}
4781 	if (mcx_cmdq_verify(cqe) != 0) {
4782 		printf("%s: create tir command corrupt\n", DEVNAME(sc));
4783 		goto free;
4784 	}
4785 
4786 	out = mcx_cmdq_out(cqe);
4787 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4788 		printf("%s: create tir failed (%x, %x)\n", DEVNAME(sc),
4789 		    out->cmd_status, betoh32(out->cmd_syndrome));
4790 		error = -1;
4791 		goto free;
4792 	}
4793 
4794 	*tirn = mcx_get_id(out->cmd_tirn);
4795 free:
4796 	mcx_dmamem_free(sc, &mxm);
4797 	return (error);
4798 }
4799 
4800 static int
4801 mcx_destroy_tir(struct mcx_softc *sc, int tirn)
4802 {
4803 	struct mcx_cmdq_entry *cqe;
4804 	struct mcx_cmd_destroy_tir_in *in;
4805 	struct mcx_cmd_destroy_tir_out *out;
4806 	int error;
4807 	int token;
4808 
4809 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4810 	token = mcx_cmdq_token(sc);
4811 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), token);
4812 
4813 	in = mcx_cmdq_in(cqe);
4814 	in->cmd_opcode = htobe16(MCX_CMD_DESTROY_TIR);
4815 	in->cmd_op_mod = htobe16(0);
4816 	in->cmd_tirn = htobe32(tirn);
4817 
4818 	mcx_cmdq_post(sc, cqe, 0);
4819 	error = mcx_cmdq_poll(sc, cqe, 1000);
4820 	if (error != 0) {
4821 		printf("%s: destroy tir timeout\n", DEVNAME(sc));
4822 		return error;
4823 	}
4824 	if (mcx_cmdq_verify(cqe) != 0) {
4825 		printf("%s: destroy tir command corrupt\n", DEVNAME(sc));
4826 		return error;
4827 	}
4828 
4829 	out = mcx_cmdq_out(cqe);
4830 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4831 		printf("%s: destroy tir failed (%x, %x)\n", DEVNAME(sc),
4832 		    out->cmd_status, betoh32(out->cmd_syndrome));
4833 		return -1;
4834 	}
4835 
4836 	return (0);
4837 }
4838 
4839 static int
4840 mcx_create_sq(struct mcx_softc *sc, struct mcx_tx *tx, int uar, int db,
4841     int cqn)
4842 {
4843 	struct mcx_cmdq_entry *cqe;
4844 	struct mcx_dmamem mxm;
4845 	struct mcx_cmd_create_sq_in *in;
4846 	struct mcx_sq_ctx *mbin;
4847 	struct mcx_cmd_create_sq_out *out;
4848 	int error;
4849 	uint64_t *pas;
4850 	uint8_t *doorbell;
4851 	int insize, npages, paslen, token;
4852 
4853 	npages = howmany((1 << MCX_LOG_SQ_SIZE) * sizeof(struct mcx_sq_entry),
4854 	    MCX_PAGE_SIZE);
4855 	paslen = npages * sizeof(*pas);
4856 	insize = sizeof(struct mcx_sq_ctx) + paslen;
4857 
4858 	if (mcx_dmamem_alloc(sc, &tx->tx_sq_mem, npages * MCX_PAGE_SIZE,
4859 	    MCX_PAGE_SIZE) != 0) {
4860 		printf("%s: unable to allocate send queue memory\n",
4861 		    DEVNAME(sc));
4862 		return (-1);
4863 	}
4864 
4865 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4866 	token = mcx_cmdq_token(sc);
4867 	mcx_cmdq_init(sc, cqe, sizeof(*in) + insize + paslen, sizeof(*out),
4868 	    token);
4869 
4870 	in = mcx_cmdq_in(cqe);
4871 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_SQ);
4872 	in->cmd_op_mod = htobe16(0);
4873 
4874 	if (mcx_cmdq_mboxes_alloc(sc, &mxm,
4875 	    howmany(insize, MCX_CMDQ_MAILBOX_DATASIZE),
4876 	    &cqe->cq_input_ptr, token) != 0) {
4877 		printf("%s: unable to allocate create sq mailboxen\n",
4878 		    DEVNAME(sc));
4879 		error = -1;
4880 		goto free;
4881 	}
4882 	mbin = (struct mcx_sq_ctx *)
4883 	    (((char *)mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0))) + 0x10);
4884 	mbin->sq_flags = htobe32(MCX_SQ_CTX_RLKEY |
4885 	    (1 << MCX_SQ_CTX_MIN_WQE_INLINE_SHIFT));
4886 	mbin->sq_cqn = htobe32(cqn);
4887 	mbin->sq_tis_lst_sz = htobe32(1 << MCX_SQ_CTX_TIS_LST_SZ_SHIFT);
4888 	mbin->sq_tis_num = htobe32(sc->sc_tis);
4889 	mbin->sq_wq.wq_type = MCX_WQ_CTX_TYPE_CYCLIC;
4890 	mbin->sq_wq.wq_pd = htobe32(sc->sc_pd);
4891 	mbin->sq_wq.wq_uar_page = htobe32(uar);
4892 	mbin->sq_wq.wq_doorbell = htobe64(MCX_DMA_DVA(&sc->sc_doorbell_mem) +
4893 	    MCX_WQ_DOORBELL_BASE + (db * MCX_WQ_DOORBELL_STRIDE));
4894 	mbin->sq_wq.wq_log_stride = htobe16(MCX_LOG_SQ_ENTRY_SIZE);
4895 	mbin->sq_wq.wq_log_size = MCX_LOG_SQ_SIZE;
4896 
4897 	/* physical addresses follow the mailbox in data */
4898 	mcx_cmdq_mboxes_pas(&mxm, sizeof(*mbin) + 0x10,
4899 	    npages, &tx->tx_sq_mem);
4900 	mcx_cmdq_post(sc, cqe, 0);
4901 
4902 	error = mcx_cmdq_poll(sc, cqe, 1000);
4903 	if (error != 0) {
4904 		printf("%s: create sq timeout\n", DEVNAME(sc));
4905 		goto free;
4906 	}
4907 	if (mcx_cmdq_verify(cqe) != 0) {
4908 		printf("%s: create sq command corrupt\n", DEVNAME(sc));
4909 		goto free;
4910 	}
4911 
4912 	out = mcx_cmdq_out(cqe);
4913 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4914 		printf("%s: create sq failed (%x, %x)\n", DEVNAME(sc),
4915 		    out->cmd_status, betoh32(out->cmd_syndrome));
4916 		error = -1;
4917 		goto free;
4918 	}
4919 
4920 	tx->tx_uar = uar;
4921 	tx->tx_sqn = mcx_get_id(out->cmd_sqn);
4922 
4923 	doorbell = MCX_DMA_KVA(&sc->sc_doorbell_mem);
4924 	tx->tx_doorbell = (uint32_t *)(doorbell + MCX_WQ_DOORBELL_BASE +
4925 	    (db * MCX_WQ_DOORBELL_STRIDE) + 4);
4926 free:
4927 	mcx_dmamem_free(sc, &mxm);
4928 	return (error);
4929 }
4930 
4931 static int
4932 mcx_destroy_sq(struct mcx_softc *sc, struct mcx_tx *tx)
4933 {
4934 	struct mcx_cmdq_entry *cqe;
4935 	struct mcx_cmd_destroy_sq_in *in;
4936 	struct mcx_cmd_destroy_sq_out *out;
4937 	int error;
4938 	int token;
4939 
4940 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4941 	token = mcx_cmdq_token(sc);
4942 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), token);
4943 
4944 	in = mcx_cmdq_in(cqe);
4945 	in->cmd_opcode = htobe16(MCX_CMD_DESTROY_SQ);
4946 	in->cmd_op_mod = htobe16(0);
4947 	in->cmd_sqn = htobe32(tx->tx_sqn);
4948 
4949 	mcx_cmdq_post(sc, cqe, 0);
4950 	error = mcx_cmdq_poll(sc, cqe, 1000);
4951 	if (error != 0) {
4952 		printf("%s: destroy sq timeout\n", DEVNAME(sc));
4953 		return error;
4954 	}
4955 	if (mcx_cmdq_verify(cqe) != 0) {
4956 		printf("%s: destroy sq command corrupt\n", DEVNAME(sc));
4957 		return error;
4958 	}
4959 
4960 	out = mcx_cmdq_out(cqe);
4961 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4962 		printf("%s: destroy sq failed (%x, %x)\n", DEVNAME(sc),
4963 		    out->cmd_status, betoh32(out->cmd_syndrome));
4964 		return -1;
4965 	}
4966 
4967 	tx->tx_sqn = 0;
4968 	return 0;
4969 }
4970 
4971 static int
4972 mcx_ready_sq(struct mcx_softc *sc, struct mcx_tx *tx)
4973 {
4974 	struct mcx_cmdq_entry *cqe;
4975 	struct mcx_dmamem mxm;
4976 	struct mcx_cmd_modify_sq_in *in;
4977 	struct mcx_cmd_modify_sq_mb_in *mbin;
4978 	struct mcx_cmd_modify_sq_out *out;
4979 	int error;
4980 	int token;
4981 
4982 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4983 	token = mcx_cmdq_token(sc);
4984 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
4985 	    sizeof(*out), token);
4986 
4987 	in = mcx_cmdq_in(cqe);
4988 	in->cmd_opcode = htobe16(MCX_CMD_MODIFY_SQ);
4989 	in->cmd_op_mod = htobe16(0);
4990 	in->cmd_sq_state = htobe32((MCX_QUEUE_STATE_RST << 28) | tx->tx_sqn);
4991 
4992 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
4993 	    &cqe->cq_input_ptr, token) != 0) {
4994 		printf("%s: unable to allocate modify sq mailbox\n",
4995 		    DEVNAME(sc));
4996 		return (-1);
4997 	}
4998 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
4999 	mbin->cmd_sq_ctx.sq_flags = htobe32(
5000 	    MCX_QUEUE_STATE_RDY << MCX_SQ_CTX_STATE_SHIFT);
5001 
5002 	mcx_cmdq_mboxes_sign(&mxm, 1);
5003 	mcx_cmdq_post(sc, cqe, 0);
5004 	error = mcx_cmdq_poll(sc, cqe, 1000);
5005 	if (error != 0) {
5006 		printf("%s: modify sq timeout\n", DEVNAME(sc));
5007 		goto free;
5008 	}
5009 	if (mcx_cmdq_verify(cqe) != 0) {
5010 		printf("%s: modify sq command corrupt\n", DEVNAME(sc));
5011 		goto free;
5012 	}
5013 
5014 	out = mcx_cmdq_out(cqe);
5015 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5016 		printf("%s: modify sq failed (%x, %x)\n", DEVNAME(sc),
5017 		    out->cmd_status, betoh32(out->cmd_syndrome));
5018 		error = -1;
5019 		goto free;
5020 	}
5021 
5022 free:
5023 	mcx_dmamem_free(sc, &mxm);
5024 	return (error);
5025 }
5026 
5027 static int
5028 mcx_create_tis(struct mcx_softc *sc, int *tis)
5029 {
5030 	struct mcx_cmdq_entry *cqe;
5031 	struct mcx_dmamem mxm;
5032 	struct mcx_cmd_create_tis_in *in;
5033 	struct mcx_cmd_create_tis_mb_in *mbin;
5034 	struct mcx_cmd_create_tis_out *out;
5035 	int error;
5036 	int token;
5037 
5038 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5039 	token = mcx_cmdq_token(sc);
5040 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
5041 	    sizeof(*out), token);
5042 
5043 	in = mcx_cmdq_in(cqe);
5044 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_TIS);
5045 	in->cmd_op_mod = htobe16(0);
5046 
5047 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
5048 	    &cqe->cq_input_ptr, token) != 0) {
5049 		printf("%s: unable to allocate create tis mailbox\n",
5050 		    DEVNAME(sc));
5051 		return (-1);
5052 	}
5053 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5054 	mbin->cmd_tdomain = htobe32(sc->sc_tdomain);
5055 
5056 	mcx_cmdq_mboxes_sign(&mxm, 1);
5057 	mcx_cmdq_post(sc, cqe, 0);
5058 	error = mcx_cmdq_poll(sc, cqe, 1000);
5059 	if (error != 0) {
5060 		printf("%s: create tis timeout\n", DEVNAME(sc));
5061 		goto free;
5062 	}
5063 	if (mcx_cmdq_verify(cqe) != 0) {
5064 		printf("%s: create tis command corrupt\n", DEVNAME(sc));
5065 		goto free;
5066 	}
5067 
5068 	out = mcx_cmdq_out(cqe);
5069 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5070 		printf("%s: create tis failed (%x, %x)\n", DEVNAME(sc),
5071 		    out->cmd_status, betoh32(out->cmd_syndrome));
5072 		error = -1;
5073 		goto free;
5074 	}
5075 
5076 	*tis = mcx_get_id(out->cmd_tisn);
5077 free:
5078 	mcx_dmamem_free(sc, &mxm);
5079 	return (error);
5080 }
5081 
5082 static int
5083 mcx_destroy_tis(struct mcx_softc *sc, int tis)
5084 {
5085 	struct mcx_cmdq_entry *cqe;
5086 	struct mcx_cmd_destroy_tis_in *in;
5087 	struct mcx_cmd_destroy_tis_out *out;
5088 	int error;
5089 	int token;
5090 
5091 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5092 	token = mcx_cmdq_token(sc);
5093 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), token);
5094 
5095 	in = mcx_cmdq_in(cqe);
5096 	in->cmd_opcode = htobe16(MCX_CMD_DESTROY_TIS);
5097 	in->cmd_op_mod = htobe16(0);
5098 	in->cmd_tisn = htobe32(tis);
5099 
5100 	mcx_cmdq_post(sc, cqe, 0);
5101 	error = mcx_cmdq_poll(sc, cqe, 1000);
5102 	if (error != 0) {
5103 		printf("%s: destroy tis timeout\n", DEVNAME(sc));
5104 		return error;
5105 	}
5106 	if (mcx_cmdq_verify(cqe) != 0) {
5107 		printf("%s: destroy tis command corrupt\n", DEVNAME(sc));
5108 		return error;
5109 	}
5110 
5111 	out = mcx_cmdq_out(cqe);
5112 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5113 		printf("%s: destroy tis failed (%x, %x)\n", DEVNAME(sc),
5114 		    out->cmd_status, betoh32(out->cmd_syndrome));
5115 		return -1;
5116 	}
5117 
5118 	return 0;
5119 }
5120 
5121 static int
5122 mcx_create_rqt(struct mcx_softc *sc, int size, int *rqns, int *rqt)
5123 {
5124 	struct mcx_cmdq_entry *cqe;
5125 	struct mcx_dmamem mxm;
5126 	struct mcx_cmd_create_rqt_in *in;
5127 	struct mcx_cmd_create_rqt_mb_in *mbin;
5128 	struct mcx_cmd_create_rqt_out *out;
5129 	struct mcx_rqt_ctx *rqt_ctx;
5130 	int *rqtn;
5131 	int error;
5132 	int token;
5133 	int i;
5134 
5135 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5136 	token = mcx_cmdq_token(sc);
5137 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin) +
5138 	    (size * sizeof(int)), sizeof(*out), token);
5139 
5140 	in = mcx_cmdq_in(cqe);
5141 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_RQT);
5142 	in->cmd_op_mod = htobe16(0);
5143 
5144 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
5145 	    &cqe->cq_input_ptr, token) != 0) {
5146 		printf("%s: unable to allocate create rqt mailbox\n",
5147 		    DEVNAME(sc));
5148 		return (-1);
5149 	}
5150 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5151 	rqt_ctx = &mbin->cmd_rqt;
5152 	rqt_ctx->cmd_rqt_max_size = htobe16(sc->sc_max_rqt_size);
5153 	rqt_ctx->cmd_rqt_actual_size = htobe16(size);
5154 
5155 	/* rqt list follows the rqt context */
5156 	rqtn = (int *)(rqt_ctx + 1);
5157 	for (i = 0; i < size; i++) {
5158 		rqtn[i] = htobe32(rqns[i]);
5159 	}
5160 
5161 	mcx_cmdq_mboxes_sign(&mxm, 1);
5162 	mcx_cmdq_post(sc, cqe, 0);
5163 	error = mcx_cmdq_poll(sc, cqe, 1000);
5164 	if (error != 0) {
5165 		printf("%s: create rqt timeout\n", DEVNAME(sc));
5166 		goto free;
5167 	}
5168 	if (mcx_cmdq_verify(cqe) != 0) {
5169 		printf("%s: create rqt command corrupt\n", DEVNAME(sc));
5170 		goto free;
5171 	}
5172 
5173 	out = mcx_cmdq_out(cqe);
5174 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5175 		printf("%s: create rqt failed (%x, %x)\n", DEVNAME(sc),
5176 		    out->cmd_status, betoh32(out->cmd_syndrome));
5177 		error = -1;
5178 		goto free;
5179 	}
5180 
5181 	*rqt = mcx_get_id(out->cmd_rqtn);
5182 	return (0);
5183 free:
5184 	mcx_dmamem_free(sc, &mxm);
5185 	return (error);
5186 }
5187 
5188 static int
5189 mcx_destroy_rqt(struct mcx_softc *sc, int rqt)
5190 {
5191 	struct mcx_cmdq_entry *cqe;
5192 	struct mcx_cmd_destroy_rqt_in *in;
5193 	struct mcx_cmd_destroy_rqt_out *out;
5194 	int error;
5195 	int token;
5196 
5197 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5198 	token = mcx_cmdq_token(sc);
5199 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), token);
5200 
5201 	in = mcx_cmdq_in(cqe);
5202 	in->cmd_opcode = htobe16(MCX_CMD_DESTROY_RQT);
5203 	in->cmd_op_mod = htobe16(0);
5204 	in->cmd_rqtn = htobe32(rqt);
5205 
5206 	mcx_cmdq_post(sc, cqe, 0);
5207 	error = mcx_cmdq_poll(sc, cqe, 1000);
5208 	if (error != 0) {
5209 		printf("%s: destroy rqt timeout\n", DEVNAME(sc));
5210 		return error;
5211 	}
5212 	if (mcx_cmdq_verify(cqe) != 0) {
5213 		printf("%s: destroy rqt command corrupt\n", DEVNAME(sc));
5214 		return error;
5215 	}
5216 
5217 	out = mcx_cmdq_out(cqe);
5218 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5219 		printf("%s: destroy rqt failed (%x, %x)\n", DEVNAME(sc),
5220 		    out->cmd_status, betoh32(out->cmd_syndrome));
5221 		return -1;
5222 	}
5223 
5224 	return 0;
5225 }
5226 
5227 #if 0
5228 static int
5229 mcx_alloc_flow_counter(struct mcx_softc *sc, int i)
5230 {
5231 	struct mcx_cmdq_entry *cqe;
5232 	struct mcx_cmd_alloc_flow_counter_in *in;
5233 	struct mcx_cmd_alloc_flow_counter_out *out;
5234 	int error;
5235 
5236 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5237 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
5238 
5239 	in = mcx_cmdq_in(cqe);
5240 	in->cmd_opcode = htobe16(MCX_CMD_ALLOC_FLOW_COUNTER);
5241 	in->cmd_op_mod = htobe16(0);
5242 
5243 	mcx_cmdq_post(sc, cqe, 0);
5244 
5245 	error = mcx_cmdq_poll(sc, cqe, 1000);
5246 	if (error != 0) {
5247 		printf("%s: alloc flow counter timeout\n", DEVNAME(sc));
5248 		return (-1);
5249 	}
5250 	if (mcx_cmdq_verify(cqe) != 0) {
5251 		printf("%s: alloc flow counter command corrupt\n", DEVNAME(sc));
5252 		return (-1);
5253 	}
5254 
5255 	out = (struct mcx_cmd_alloc_flow_counter_out *)cqe->cq_output_data;
5256 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5257 		printf("%s: alloc flow counter failed (%x)\n", DEVNAME(sc),
5258 		    out->cmd_status);
5259 		return (-1);
5260 	}
5261 
5262 	sc->sc_flow_counter_id[i]  = betoh16(out->cmd_flow_counter_id);
5263 	printf("flow counter id %d = %d\n", i, sc->sc_flow_counter_id[i]);
5264 
5265 	return (0);
5266 }
5267 #endif
5268 
5269 static int
5270 mcx_create_flow_table(struct mcx_softc *sc, int log_size, int level,
5271     int *flow_table_id)
5272 {
5273 	struct mcx_cmdq_entry *cqe;
5274 	struct mcx_dmamem mxm;
5275 	struct mcx_cmd_create_flow_table_in *in;
5276 	struct mcx_cmd_create_flow_table_mb_in *mbin;
5277 	struct mcx_cmd_create_flow_table_out *out;
5278 	int error;
5279 	int token;
5280 
5281 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5282 	token = mcx_cmdq_token(sc);
5283 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
5284 	    sizeof(*out), token);
5285 
5286 	in = mcx_cmdq_in(cqe);
5287 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_FLOW_TABLE);
5288 	in->cmd_op_mod = htobe16(0);
5289 
5290 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
5291 	    &cqe->cq_input_ptr, token) != 0) {
5292 		printf("%s: unable to allocate create flow table mailbox\n",
5293 		    DEVNAME(sc));
5294 		return (-1);
5295 	}
5296 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5297 	mbin->cmd_table_type = MCX_FLOW_TABLE_TYPE_RX;
5298 	mbin->cmd_ctx.ft_log_size = log_size;
5299 	mbin->cmd_ctx.ft_level = level;
5300 
5301 	mcx_cmdq_mboxes_sign(&mxm, 1);
5302 	mcx_cmdq_post(sc, cqe, 0);
5303 	error = mcx_cmdq_poll(sc, cqe, 1000);
5304 	if (error != 0) {
5305 		printf("%s: create flow table timeout\n", DEVNAME(sc));
5306 		goto free;
5307 	}
5308 	if (mcx_cmdq_verify(cqe) != 0) {
5309 		printf("%s: create flow table command corrupt\n", DEVNAME(sc));
5310 		goto free;
5311 	}
5312 
5313 	out = mcx_cmdq_out(cqe);
5314 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5315 		printf("%s: create flow table failed (%x, %x)\n", DEVNAME(sc),
5316 		    out->cmd_status, betoh32(out->cmd_syndrome));
5317 		error = -1;
5318 		goto free;
5319 	}
5320 
5321 	*flow_table_id = mcx_get_id(out->cmd_table_id);
5322 free:
5323 	mcx_dmamem_free(sc, &mxm);
5324 	return (error);
5325 }
5326 
5327 static int
5328 mcx_set_flow_table_root(struct mcx_softc *sc, int flow_table_id)
5329 {
5330 	struct mcx_cmdq_entry *cqe;
5331 	struct mcx_dmamem mxm;
5332 	struct mcx_cmd_set_flow_table_root_in *in;
5333 	struct mcx_cmd_set_flow_table_root_mb_in *mbin;
5334 	struct mcx_cmd_set_flow_table_root_out *out;
5335 	int error;
5336 	int token;
5337 
5338 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5339 	token = mcx_cmdq_token(sc);
5340 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
5341 	    sizeof(*out), token);
5342 
5343 	in = mcx_cmdq_in(cqe);
5344 	in->cmd_opcode = htobe16(MCX_CMD_SET_FLOW_TABLE_ROOT);
5345 	in->cmd_op_mod = htobe16(0);
5346 
5347 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
5348 	    &cqe->cq_input_ptr, token) != 0) {
5349 		printf("%s: unable to allocate set flow table root mailbox\n",
5350 		    DEVNAME(sc));
5351 		return (-1);
5352 	}
5353 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5354 	mbin->cmd_table_type = MCX_FLOW_TABLE_TYPE_RX;
5355 	mbin->cmd_table_id = htobe32(flow_table_id);
5356 
5357 	mcx_cmdq_mboxes_sign(&mxm, 1);
5358 	mcx_cmdq_post(sc, cqe, 0);
5359 	error = mcx_cmdq_poll(sc, cqe, 1000);
5360 	if (error != 0) {
5361 		printf("%s: set flow table root timeout\n", DEVNAME(sc));
5362 		goto free;
5363 	}
5364 	if (mcx_cmdq_verify(cqe) != 0) {
5365 		printf("%s: set flow table root command corrupt\n",
5366 		    DEVNAME(sc));
5367 		goto free;
5368 	}
5369 
5370 	out = mcx_cmdq_out(cqe);
5371 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5372 		printf("%s: set flow table root failed (%x, %x)\n",
5373 		    DEVNAME(sc), out->cmd_status, betoh32(out->cmd_syndrome));
5374 		error = -1;
5375 		goto free;
5376 	}
5377 
5378 free:
5379 	mcx_dmamem_free(sc, &mxm);
5380 	return (error);
5381 }
5382 
5383 static int
5384 mcx_destroy_flow_table(struct mcx_softc *sc, int flow_table_id)
5385 {
5386 	struct mcx_cmdq_entry *cqe;
5387 	struct mcx_dmamem mxm;
5388 	struct mcx_cmd_destroy_flow_table_in *in;
5389 	struct mcx_cmd_destroy_flow_table_mb_in *mb;
5390 	struct mcx_cmd_destroy_flow_table_out *out;
5391 	int error;
5392 	int token;
5393 
5394 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5395 	token = mcx_cmdq_token(sc);
5396 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mb), sizeof(*out), token);
5397 
5398 	in = mcx_cmdq_in(cqe);
5399 	in->cmd_opcode = htobe16(MCX_CMD_DESTROY_FLOW_TABLE);
5400 	in->cmd_op_mod = htobe16(0);
5401 
5402 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
5403 	    &cqe->cq_input_ptr, token) != 0) {
5404 		printf("%s: unable to allocate destroy flow table mailbox\n",
5405 		    DEVNAME(sc));
5406 		return (-1);
5407 	}
5408 	mb = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5409 	mb->cmd_table_type = MCX_FLOW_TABLE_TYPE_RX;
5410 	mb->cmd_table_id = htobe32(flow_table_id);
5411 
5412 	mcx_cmdq_mboxes_sign(&mxm, 1);
5413 	mcx_cmdq_post(sc, cqe, 0);
5414 	error = mcx_cmdq_poll(sc, cqe, 1000);
5415 	if (error != 0) {
5416 		printf("%s: destroy flow table timeout\n", DEVNAME(sc));
5417 		goto free;
5418 	}
5419 	if (mcx_cmdq_verify(cqe) != 0) {
5420 		printf("%s: destroy flow table command corrupt\n",
5421 		    DEVNAME(sc));
5422 		goto free;
5423 	}
5424 
5425 	out = mcx_cmdq_out(cqe);
5426 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5427 		printf("%s: destroy flow table failed (%x, %x)\n", DEVNAME(sc),
5428 		    out->cmd_status, betoh32(out->cmd_syndrome));
5429 		error = -1;
5430 		goto free;
5431 	}
5432 
5433 free:
5434 	mcx_dmamem_free(sc, &mxm);
5435 	return (error);
5436 }
5437 
5438 
5439 static int
5440 mcx_create_flow_group(struct mcx_softc *sc, int flow_table_id, int group,
5441     int start, int size, int match_enable, struct mcx_flow_match *match)
5442 {
5443 	struct mcx_cmdq_entry *cqe;
5444 	struct mcx_dmamem mxm;
5445 	struct mcx_cmd_create_flow_group_in *in;
5446 	struct mcx_cmd_create_flow_group_mb_in *mbin;
5447 	struct mcx_cmd_create_flow_group_out *out;
5448 	struct mcx_flow_group *mfg;
5449 	int error;
5450 	int token;
5451 
5452 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5453 	token = mcx_cmdq_token(sc);
5454 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin), sizeof(*out),
5455 	    token);
5456 
5457 	in = mcx_cmdq_in(cqe);
5458 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_FLOW_GROUP);
5459 	in->cmd_op_mod = htobe16(0);
5460 
5461 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2, &cqe->cq_input_ptr, token)
5462 	    != 0) {
5463 		printf("%s: unable to allocate create flow group mailbox\n",
5464 		    DEVNAME(sc));
5465 		return (-1);
5466 	}
5467 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5468 	mbin->cmd_table_type = MCX_FLOW_TABLE_TYPE_RX;
5469 	mbin->cmd_table_id = htobe32(flow_table_id);
5470 	mbin->cmd_start_flow_index = htobe32(start);
5471 	mbin->cmd_end_flow_index = htobe32(start + (size - 1));
5472 
5473 	mbin->cmd_match_criteria_enable = match_enable;
5474 	memcpy(&mbin->cmd_match_criteria, match, sizeof(*match));
5475 
5476 	mcx_cmdq_mboxes_sign(&mxm, 2);
5477 	mcx_cmdq_post(sc, cqe, 0);
5478 	error = mcx_cmdq_poll(sc, cqe, 1000);
5479 	if (error != 0) {
5480 		printf("%s: create flow group timeout\n", DEVNAME(sc));
5481 		goto free;
5482 	}
5483 	if (mcx_cmdq_verify(cqe) != 0) {
5484 		printf("%s: create flow group command corrupt\n", DEVNAME(sc));
5485 		goto free;
5486 	}
5487 
5488 	out = mcx_cmdq_out(cqe);
5489 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5490 		printf("%s: create flow group failed (%x, %x)\n", DEVNAME(sc),
5491 		    out->cmd_status, betoh32(out->cmd_syndrome));
5492 		error = -1;
5493 		goto free;
5494 	}
5495 
5496 	mfg = &sc->sc_flow_group[group];
5497 	mfg->g_id = mcx_get_id(out->cmd_group_id);
5498 	mfg->g_table = flow_table_id;
5499 	mfg->g_start = start;
5500 	mfg->g_size = size;
5501 
5502 free:
5503 	mcx_dmamem_free(sc, &mxm);
5504 	return (error);
5505 }
5506 
5507 static int
5508 mcx_destroy_flow_group(struct mcx_softc *sc, int group)
5509 {
5510 	struct mcx_cmdq_entry *cqe;
5511 	struct mcx_dmamem mxm;
5512 	struct mcx_cmd_destroy_flow_group_in *in;
5513 	struct mcx_cmd_destroy_flow_group_mb_in *mb;
5514 	struct mcx_cmd_destroy_flow_group_out *out;
5515 	struct mcx_flow_group *mfg;
5516 	int error;
5517 	int token;
5518 
5519 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5520 	token = mcx_cmdq_token(sc);
5521 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mb), sizeof(*out), token);
5522 
5523 	in = mcx_cmdq_in(cqe);
5524 	in->cmd_opcode = htobe16(MCX_CMD_DESTROY_FLOW_GROUP);
5525 	in->cmd_op_mod = htobe16(0);
5526 
5527 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2,
5528 	    &cqe->cq_input_ptr, token) != 0) {
5529 		printf("%s: unable to allocate destroy flow group mailbox\n",
5530 		    DEVNAME(sc));
5531 		return (-1);
5532 	}
5533 	mb = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5534 	mb->cmd_table_type = MCX_FLOW_TABLE_TYPE_RX;
5535 	mfg = &sc->sc_flow_group[group];
5536 	mb->cmd_table_id = htobe32(mfg->g_table);
5537 	mb->cmd_group_id = htobe32(mfg->g_id);
5538 
5539 	mcx_cmdq_mboxes_sign(&mxm, 2);
5540 	mcx_cmdq_post(sc, cqe, 0);
5541 	error = mcx_cmdq_poll(sc, cqe, 1000);
5542 	if (error != 0) {
5543 		printf("%s: destroy flow group timeout\n", DEVNAME(sc));
5544 		goto free;
5545 	}
5546 	if (mcx_cmdq_verify(cqe) != 0) {
5547 		printf("%s: destroy flow group command corrupt\n", DEVNAME(sc));
5548 		goto free;
5549 	}
5550 
5551 	out = mcx_cmdq_out(cqe);
5552 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5553 		printf("%s: destroy flow group failed (%x, %x)\n", DEVNAME(sc),
5554 		    out->cmd_status, betoh32(out->cmd_syndrome));
5555 		error = -1;
5556 		goto free;
5557 	}
5558 
5559 	mfg->g_id = -1;
5560 	mfg->g_table = -1;
5561 	mfg->g_size = 0;
5562 	mfg->g_start = 0;
5563 free:
5564 	mcx_dmamem_free(sc, &mxm);
5565 	return (error);
5566 }
5567 
5568 static int
5569 mcx_set_flow_table_entry_mac(struct mcx_softc *sc, int group, int index,
5570     uint8_t *macaddr, uint32_t dest)
5571 {
5572 	struct mcx_cmdq_entry *cqe;
5573 	struct mcx_dmamem mxm;
5574 	struct mcx_cmd_set_flow_table_entry_in *in;
5575 	struct mcx_cmd_set_flow_table_entry_mb_in *mbin;
5576 	struct mcx_cmd_set_flow_table_entry_out *out;
5577 	struct mcx_flow_group *mfg;
5578 	uint32_t *pdest;
5579 	int error;
5580 	int token;
5581 
5582 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5583 	token = mcx_cmdq_token(sc);
5584 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin) + sizeof(*pdest),
5585 	    sizeof(*out), token);
5586 
5587 	in = mcx_cmdq_in(cqe);
5588 	in->cmd_opcode = htobe16(MCX_CMD_SET_FLOW_TABLE_ENTRY);
5589 	in->cmd_op_mod = htobe16(0);
5590 
5591 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2, &cqe->cq_input_ptr, token)
5592 	    != 0) {
5593 		printf("%s: unable to allocate set flow table entry mailbox\n",
5594 		    DEVNAME(sc));
5595 		return (-1);
5596 	}
5597 
5598 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5599 	mbin->cmd_table_type = MCX_FLOW_TABLE_TYPE_RX;
5600 
5601 	mfg = &sc->sc_flow_group[group];
5602 	mbin->cmd_table_id = htobe32(mfg->g_table);
5603 	mbin->cmd_flow_index = htobe32(mfg->g_start + index);
5604 	mbin->cmd_flow_ctx.fc_group_id = htobe32(mfg->g_id);
5605 
5606 	/* flow context ends at offset 0x330, 0x130 into the second mbox */
5607 	pdest = (uint32_t *)
5608 	    (((char *)mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 1))) + 0x130);
5609 	mbin->cmd_flow_ctx.fc_action = htobe32(MCX_FLOW_CONTEXT_ACTION_FORWARD);
5610 	mbin->cmd_flow_ctx.fc_dest_list_size = htobe32(1);
5611 	*pdest = htobe32(dest);
5612 
5613 	/* the only thing we match on at the moment is the dest mac address */
5614 	if (macaddr != NULL) {
5615 		memcpy(mbin->cmd_flow_ctx.fc_match_value.mc_dest_mac, macaddr,
5616 		    ETHER_ADDR_LEN);
5617 	}
5618 
5619 	mcx_cmdq_mboxes_sign(&mxm, 2);
5620 	mcx_cmdq_post(sc, cqe, 0);
5621 	error = mcx_cmdq_poll(sc, cqe, 1000);
5622 	if (error != 0) {
5623 		printf("%s: set flow table entry timeout\n", DEVNAME(sc));
5624 		goto free;
5625 	}
5626 	if (mcx_cmdq_verify(cqe) != 0) {
5627 		printf("%s: set flow table entry command corrupt\n",
5628 		    DEVNAME(sc));
5629 		goto free;
5630 	}
5631 
5632 	out = mcx_cmdq_out(cqe);
5633 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5634 		printf("%s: set flow table entry failed (%x, %x)\n",
5635 		    DEVNAME(sc), out->cmd_status, betoh32(out->cmd_syndrome));
5636 		error = -1;
5637 		goto free;
5638 	}
5639 
5640 free:
5641 	mcx_dmamem_free(sc, &mxm);
5642 	return (error);
5643 }
5644 
5645 static int
5646 mcx_set_flow_table_entry_proto(struct mcx_softc *sc, int group, int index,
5647     int ethertype, int ip_proto, uint32_t dest)
5648 {
5649 	struct mcx_cmdq_entry *cqe;
5650 	struct mcx_dmamem mxm;
5651 	struct mcx_cmd_set_flow_table_entry_in *in;
5652 	struct mcx_cmd_set_flow_table_entry_mb_in *mbin;
5653 	struct mcx_cmd_set_flow_table_entry_out *out;
5654 	struct mcx_flow_group *mfg;
5655 	uint32_t *pdest;
5656 	int error;
5657 	int token;
5658 
5659 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5660 	token = mcx_cmdq_token(sc);
5661 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin) + sizeof(*pdest),
5662 	    sizeof(*out), token);
5663 
5664 	in = mcx_cmdq_in(cqe);
5665 	in->cmd_opcode = htobe16(MCX_CMD_SET_FLOW_TABLE_ENTRY);
5666 	in->cmd_op_mod = htobe16(0);
5667 
5668 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2, &cqe->cq_input_ptr, token)
5669 	    != 0) {
5670 		printf("%s: unable to allocate set flow table entry mailbox\n",
5671 		    DEVNAME(sc));
5672 		return (-1);
5673 	}
5674 
5675 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5676 	mbin->cmd_table_type = MCX_FLOW_TABLE_TYPE_RX;
5677 
5678 	mfg = &sc->sc_flow_group[group];
5679 	mbin->cmd_table_id = htobe32(mfg->g_table);
5680 	mbin->cmd_flow_index = htobe32(mfg->g_start + index);
5681 	mbin->cmd_flow_ctx.fc_group_id = htobe32(mfg->g_id);
5682 
5683 	/* flow context ends at offset 0x330, 0x130 into the second mbox */
5684 	pdest = (uint32_t *)
5685 	    (((char *)mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 1))) + 0x130);
5686 	mbin->cmd_flow_ctx.fc_action = htobe32(MCX_FLOW_CONTEXT_ACTION_FORWARD);
5687 	mbin->cmd_flow_ctx.fc_dest_list_size = htobe32(1);
5688 	*pdest = htobe32(dest);
5689 
5690 	mbin->cmd_flow_ctx.fc_match_value.mc_ethertype = htobe16(ethertype);
5691 	mbin->cmd_flow_ctx.fc_match_value.mc_ip_proto = ip_proto;
5692 
5693 	mcx_cmdq_mboxes_sign(&mxm, 2);
5694 	mcx_cmdq_post(sc, cqe, 0);
5695 	error = mcx_cmdq_poll(sc, cqe, 1000);
5696 	if (error != 0) {
5697 		printf("%s: set flow table entry timeout\n", DEVNAME(sc));
5698 		goto free;
5699 	}
5700 	if (mcx_cmdq_verify(cqe) != 0) {
5701 		printf("%s: set flow table entry command corrupt\n",
5702 		    DEVNAME(sc));
5703 		goto free;
5704 	}
5705 
5706 	out = mcx_cmdq_out(cqe);
5707 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5708 		printf("%s: set flow table entry failed (%x, %x)\n",
5709 		    DEVNAME(sc), out->cmd_status, betoh32(out->cmd_syndrome));
5710 		error = -1;
5711 		goto free;
5712 	}
5713 
5714 free:
5715 	mcx_dmamem_free(sc, &mxm);
5716 	return (error);
5717 }
5718 
5719 static int
5720 mcx_delete_flow_table_entry(struct mcx_softc *sc, int group, int index)
5721 {
5722 	struct mcx_cmdq_entry *cqe;
5723 	struct mcx_dmamem mxm;
5724 	struct mcx_cmd_delete_flow_table_entry_in *in;
5725 	struct mcx_cmd_delete_flow_table_entry_mb_in *mbin;
5726 	struct mcx_cmd_delete_flow_table_entry_out *out;
5727 	struct mcx_flow_group *mfg;
5728 	int error;
5729 	int token;
5730 
5731 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5732 	token = mcx_cmdq_token(sc);
5733 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin), sizeof(*out),
5734 	    token);
5735 
5736 	in = mcx_cmdq_in(cqe);
5737 	in->cmd_opcode = htobe16(MCX_CMD_DELETE_FLOW_TABLE_ENTRY);
5738 	in->cmd_op_mod = htobe16(0);
5739 
5740 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2,
5741 	    &cqe->cq_input_ptr, token) != 0) {
5742 		printf("%s: unable to allocate "
5743 		    "delete flow table entry mailbox\n", DEVNAME(sc));
5744 		return (-1);
5745 	}
5746 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5747 	mbin->cmd_table_type = MCX_FLOW_TABLE_TYPE_RX;
5748 
5749 	mfg = &sc->sc_flow_group[group];
5750 	mbin->cmd_table_id = htobe32(mfg->g_table);
5751 	mbin->cmd_flow_index = htobe32(mfg->g_start + index);
5752 
5753 	mcx_cmdq_mboxes_sign(&mxm, 2);
5754 	mcx_cmdq_post(sc, cqe, 0);
5755 	error = mcx_cmdq_poll(sc, cqe, 1000);
5756 	if (error != 0) {
5757 		printf("%s: delete flow table entry timeout\n", DEVNAME(sc));
5758 		goto free;
5759 	}
5760 	if (mcx_cmdq_verify(cqe) != 0) {
5761 		printf("%s: delete flow table entry command corrupt\n",
5762 		    DEVNAME(sc));
5763 		goto free;
5764 	}
5765 
5766 	out = mcx_cmdq_out(cqe);
5767 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5768 		printf("%s: delete flow table entry %d:%d failed (%x, %x)\n",
5769 		    DEVNAME(sc), group, index, out->cmd_status,
5770 		    betoh32(out->cmd_syndrome));
5771 		error = -1;
5772 		goto free;
5773 	}
5774 
5775 free:
5776 	mcx_dmamem_free(sc, &mxm);
5777 	return (error);
5778 }
5779 
5780 #if 0
5781 int
5782 mcx_dump_flow_table(struct mcx_softc *sc, int flow_table_id)
5783 {
5784 	struct mcx_dmamem mxm;
5785 	struct mcx_cmdq_entry *cqe;
5786 	struct mcx_cmd_query_flow_table_in *in;
5787 	struct mcx_cmd_query_flow_table_mb_in *mbin;
5788 	struct mcx_cmd_query_flow_table_out *out;
5789 	struct mcx_cmd_query_flow_table_mb_out *mbout;
5790 	uint8_t token = mcx_cmdq_token(sc);
5791 	int error;
5792 	int i;
5793 	uint8_t *dump;
5794 
5795 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5796 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
5797 	    sizeof(*out) + sizeof(*mbout) + 16, token);
5798 
5799 	in = mcx_cmdq_in(cqe);
5800 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_FLOW_TABLE);
5801 	in->cmd_op_mod = htobe16(0);
5802 
5803 	CTASSERT(sizeof(*mbin) <= MCX_CMDQ_MAILBOX_DATASIZE);
5804 	CTASSERT(sizeof(*mbout) <= MCX_CMDQ_MAILBOX_DATASIZE);
5805 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2,
5806 	    &cqe->cq_output_ptr, token) != 0) {
5807 		printf(", unable to allocate query flow table mailboxes\n");
5808 		return (-1);
5809 	}
5810 	cqe->cq_input_ptr = cqe->cq_output_ptr;
5811 
5812 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5813 	mbin->cmd_table_type = 0;
5814 	mbin->cmd_table_id = htobe32(flow_table_id);
5815 
5816 	mcx_cmdq_mboxes_sign(&mxm, 1);
5817 
5818 	mcx_cmdq_post(sc, cqe, 0);
5819 	error = mcx_cmdq_poll(sc, cqe, 1000);
5820 	if (error != 0) {
5821 		printf("%s: query flow table timeout\n", DEVNAME(sc));
5822 		goto free;
5823 	}
5824 	error = mcx_cmdq_verify(cqe);
5825 	if (error != 0) {
5826 		printf("%s: query flow table reply corrupt\n", DEVNAME(sc));
5827 		goto free;
5828 	}
5829 
5830 	out = mcx_cmdq_out(cqe);
5831 	switch (out->cmd_status) {
5832 	case MCX_CQ_STATUS_OK:
5833 		break;
5834 	default:
5835 		printf("%s: query flow table failed (%x/%x)\n", DEVNAME(sc),
5836 		    out->cmd_status, betoh32(out->cmd_syndrome));
5837 		error = -1;
5838 		goto free;
5839 	}
5840 
5841         mbout = (struct mcx_cmd_query_flow_table_mb_out *)
5842 	    (mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)));
5843 	dump = (uint8_t *)mbout + 8;
5844 	for (i = 0; i < sizeof(struct mcx_flow_table_ctx); i++) {
5845 		printf("%.2x ", dump[i]);
5846 		if (i % 16 == 15)
5847 			printf("\n");
5848 	}
5849 free:
5850 	mcx_cq_mboxes_free(sc, &mxm);
5851 	return (error);
5852 }
5853 int
5854 mcx_dump_flow_table_entry(struct mcx_softc *sc, int flow_table_id, int index)
5855 {
5856 	struct mcx_dmamem mxm;
5857 	struct mcx_cmdq_entry *cqe;
5858 	struct mcx_cmd_query_flow_table_entry_in *in;
5859 	struct mcx_cmd_query_flow_table_entry_mb_in *mbin;
5860 	struct mcx_cmd_query_flow_table_entry_out *out;
5861 	struct mcx_cmd_query_flow_table_entry_mb_out *mbout;
5862 	uint8_t token = mcx_cmdq_token(sc);
5863 	int error;
5864 	int i;
5865 	uint8_t *dump;
5866 
5867 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5868 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
5869 	    sizeof(*out) + sizeof(*mbout) + 16, token);
5870 
5871 	in = mcx_cmdq_in(cqe);
5872 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_FLOW_TABLE_ENTRY);
5873 	in->cmd_op_mod = htobe16(0);
5874 
5875 	CTASSERT(sizeof(*mbin) <= MCX_CMDQ_MAILBOX_DATASIZE);
5876 	CTASSERT(sizeof(*mbout) <= MCX_CMDQ_MAILBOX_DATASIZE*2);
5877 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2,
5878 	    &cqe->cq_output_ptr, token) != 0) {
5879 		printf(", unable to allocate "
5880 		    "query flow table entry mailboxes\n");
5881 		return (-1);
5882 	}
5883 	cqe->cq_input_ptr = cqe->cq_output_ptr;
5884 
5885 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5886 	mbin->cmd_table_type = 0;
5887 	mbin->cmd_table_id = htobe32(flow_table_id);
5888 	mbin->cmd_flow_index = htobe32(index);
5889 
5890 	mcx_cmdq_mboxes_sign(&mxm, 1);
5891 
5892 	mcx_cmdq_post(sc, cqe, 0);
5893 	error = mcx_cmdq_poll(sc, cqe, 1000);
5894 	if (error != 0) {
5895 		printf("%s: query flow table entry timeout\n", DEVNAME(sc));
5896 		goto free;
5897 	}
5898 	error = mcx_cmdq_verify(cqe);
5899 	if (error != 0) {
5900 		printf("%s: query flow table entry reply corrupt\n",
5901 		    DEVNAME(sc));
5902 		goto free;
5903 	}
5904 
5905 	out = mcx_cmdq_out(cqe);
5906 	switch (out->cmd_status) {
5907 	case MCX_CQ_STATUS_OK:
5908 		break;
5909 	default:
5910 		printf("%s: query flow table entry failed (%x/%x)\n",
5911 		    DEVNAME(sc), out->cmd_status, betoh32(out->cmd_syndrome));
5912 		error = -1;
5913 		goto free;
5914 	}
5915 
5916         mbout = (struct mcx_cmd_query_flow_table_entry_mb_out *)
5917 	    (mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)));
5918 	dump = (uint8_t *)mbout;
5919 	for (i = 0; i < MCX_CMDQ_MAILBOX_DATASIZE; i++) {
5920 		printf("%.2x ", dump[i]);
5921 		if (i % 16 == 15)
5922 			printf("\n");
5923 	}
5924 
5925 free:
5926 	mcx_cq_mboxes_free(sc, &mxm);
5927 	return (error);
5928 }
5929 
5930 int
5931 mcx_dump_flow_group(struct mcx_softc *sc, int flow_table_id)
5932 {
5933 	struct mcx_dmamem mxm;
5934 	struct mcx_cmdq_entry *cqe;
5935 	struct mcx_cmd_query_flow_group_in *in;
5936 	struct mcx_cmd_query_flow_group_mb_in *mbin;
5937 	struct mcx_cmd_query_flow_group_out *out;
5938 	struct mcx_cmd_query_flow_group_mb_out *mbout;
5939 	uint8_t token = mcx_cmdq_token(sc);
5940 	int error;
5941 	int i;
5942 	uint8_t *dump;
5943 
5944 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5945 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
5946 	    sizeof(*out) + sizeof(*mbout) + 16, token);
5947 
5948 	in = mcx_cmdq_in(cqe);
5949 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_FLOW_GROUP);
5950 	in->cmd_op_mod = htobe16(0);
5951 
5952 	CTASSERT(sizeof(*mbin) <= MCX_CMDQ_MAILBOX_DATASIZE);
5953 	CTASSERT(sizeof(*mbout) <= MCX_CMDQ_MAILBOX_DATASIZE*2);
5954 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2,
5955 	    &cqe->cq_output_ptr, token) != 0) {
5956 		printf(", unable to allocate query flow group mailboxes\n");
5957 		return (-1);
5958 	}
5959 	cqe->cq_input_ptr = cqe->cq_output_ptr;
5960 
5961 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5962 	mbin->cmd_table_type = 0;
5963 	mbin->cmd_table_id = htobe32(flow_table_id);
5964 	mbin->cmd_group_id = htobe32(sc->sc_flow_group_id);
5965 
5966 	mcx_cmdq_mboxes_sign(&mxm, 1);
5967 
5968 	mcx_cmdq_post(sc, cqe, 0);
5969 	error = mcx_cmdq_poll(sc, cqe, 1000);
5970 	if (error != 0) {
5971 		printf("%s: query flow group timeout\n", DEVNAME(sc));
5972 		goto free;
5973 	}
5974 	error = mcx_cmdq_verify(cqe);
5975 	if (error != 0) {
5976 		printf("%s: query flow group reply corrupt\n", DEVNAME(sc));
5977 		goto free;
5978 	}
5979 
5980 	out = mcx_cmdq_out(cqe);
5981 	switch (out->cmd_status) {
5982 	case MCX_CQ_STATUS_OK:
5983 		break;
5984 	default:
5985 		printf("%s: query flow group failed (%x/%x)\n", DEVNAME(sc),
5986 		    out->cmd_status, betoh32(out->cmd_syndrome));
5987 		error = -1;
5988 		goto free;
5989 	}
5990 
5991         mbout = (struct mcx_cmd_query_flow_group_mb_out *)
5992 	    (mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)));
5993 	dump = (uint8_t *)mbout;
5994 	for (i = 0; i < MCX_CMDQ_MAILBOX_DATASIZE; i++) {
5995 		printf("%.2x ", dump[i]);
5996 		if (i % 16 == 15)
5997 			printf("\n");
5998 	}
5999 	dump = (uint8_t *)(mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 1)));
6000 	for (i = 0; i < MCX_CMDQ_MAILBOX_DATASIZE; i++) {
6001 		printf("%.2x ", dump[i]);
6002 		if (i % 16 == 15)
6003 			printf("\n");
6004 	}
6005 
6006 free:
6007 	mcx_cq_mboxes_free(sc, &mxm);
6008 	return (error);
6009 }
6010 
6011 static int
6012 mcx_dump_counters(struct mcx_softc *sc)
6013 {
6014 	struct mcx_dmamem mxm;
6015 	struct mcx_cmdq_entry *cqe;
6016 	struct mcx_cmd_query_vport_counters_in *in;
6017 	struct mcx_cmd_query_vport_counters_mb_in *mbin;
6018 	struct mcx_cmd_query_vport_counters_out *out;
6019 	struct mcx_nic_vport_counters *counters;
6020 	int error, token;
6021 
6022 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
6023 	token = mcx_cmdq_token(sc);
6024 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
6025 	    sizeof(*out) + sizeof(*counters), token);
6026 
6027 	in = mcx_cmdq_in(cqe);
6028 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_VPORT_COUNTERS);
6029 	in->cmd_op_mod = htobe16(0);
6030 
6031 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
6032 	    &cqe->cq_output_ptr, token) != 0) {
6033 		printf(", unable to allocate "
6034 		    "query nic vport counters mailboxen\n");
6035 		return (-1);
6036 	}
6037 	cqe->cq_input_ptr = cqe->cq_output_ptr;
6038 
6039 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
6040 	mbin->cmd_clear = 0x80;
6041 
6042 	mcx_cmdq_mboxes_sign(&mxm, 1);
6043 	mcx_cmdq_post(sc, cqe, 0);
6044 
6045 	error = mcx_cmdq_poll(sc, cqe, 1000);
6046 	if (error != 0) {
6047 		printf("%s: query nic vport counters timeout\n", DEVNAME(sc));
6048 		goto free;
6049 	}
6050 	if (mcx_cmdq_verify(cqe) != 0) {
6051 		printf("%s: query nic vport counters command corrupt\n",
6052 		    DEVNAME(sc));
6053 		goto free;
6054 	}
6055 
6056 	out = mcx_cmdq_out(cqe);
6057 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
6058 		printf("%s: query nic vport counters failed (%x, %x)\n",
6059 		    DEVNAME(sc), out->cmd_status, betoh32(out->cmd_syndrome));
6060 		error = -1;
6061 		goto free;
6062 	}
6063 
6064 	counters = (struct mcx_nic_vport_counters *)
6065 	    (mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)));
6066 	if (counters->rx_bcast.packets + counters->tx_bcast.packets +
6067 	    counters->rx_ucast.packets + counters->tx_ucast.packets +
6068 	    counters->rx_err.packets + counters->tx_err.packets)
6069 		printf("%s: err %llx/%llx uc %llx/%llx bc %llx/%llx\n",
6070 		    DEVNAME(sc),
6071 		    betoh64(counters->tx_err.packets),
6072 		    betoh64(counters->rx_err.packets),
6073 		    betoh64(counters->tx_ucast.packets),
6074 		    betoh64(counters->rx_ucast.packets),
6075 		    betoh64(counters->tx_bcast.packets),
6076 		    betoh64(counters->rx_bcast.packets));
6077 free:
6078 	mcx_dmamem_free(sc, &mxm);
6079 
6080 	return (error);
6081 }
6082 
6083 static int
6084 mcx_dump_flow_counter(struct mcx_softc *sc, int index, const char *what)
6085 {
6086 	struct mcx_dmamem mxm;
6087 	struct mcx_cmdq_entry *cqe;
6088 	struct mcx_cmd_query_flow_counter_in *in;
6089 	struct mcx_cmd_query_flow_counter_mb_in *mbin;
6090 	struct mcx_cmd_query_flow_counter_out *out;
6091 	struct mcx_counter *counters;
6092 	int error, token;
6093 
6094 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
6095 	token = mcx_cmdq_token(sc);
6096 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin), sizeof(*out) +
6097 	    sizeof(*counters), token);
6098 
6099 	in = mcx_cmdq_in(cqe);
6100 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_FLOW_COUNTER);
6101 	in->cmd_op_mod = htobe16(0);
6102 
6103 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
6104 	    &cqe->cq_output_ptr, token) != 0) {
6105 		printf(", unable to allocate query flow counter mailboxen\n");
6106 		return (-1);
6107 	}
6108 	cqe->cq_input_ptr = cqe->cq_output_ptr;
6109 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
6110 	mbin->cmd_flow_counter_id = htobe16(sc->sc_flow_counter_id[index]);
6111 	mbin->cmd_clear = 0x80;
6112 
6113 	mcx_cmdq_mboxes_sign(&mxm, 1);
6114 	mcx_cmdq_post(sc, cqe, 0);
6115 
6116 	error = mcx_cmdq_poll(sc, cqe, 1000);
6117 	if (error != 0) {
6118 		printf("%s: query flow counter timeout\n", DEVNAME(sc));
6119 		goto free;
6120 	}
6121 	if (mcx_cmdq_verify(cqe) != 0) {
6122 		printf("%s: query flow counter command corrupt\n", DEVNAME(sc));
6123 		goto free;
6124 	}
6125 
6126 	out = mcx_cmdq_out(cqe);
6127 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
6128 		printf("%s: query flow counter failed (%x, %x)\n", DEVNAME(sc),
6129 		    out->cmd_status, betoh32(out->cmd_syndrome));
6130 		error = -1;
6131 		goto free;
6132 	}
6133 
6134 	counters = (struct mcx_counter *)
6135 	    (mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)));
6136 	if (counters->packets)
6137 		printf("%s: %s inflow %llx\n", DEVNAME(sc), what,
6138 		    betoh64(counters->packets));
6139 free:
6140 	mcx_dmamem_free(sc, &mxm);
6141 
6142 	return (error);
6143 }
6144 
6145 #endif
6146 
6147 #if NKSTAT > 0
6148 
6149 int
6150 mcx_query_rq(struct mcx_softc *sc, struct mcx_rx *rx, struct mcx_rq_ctx *rq_ctx)
6151 {
6152 	struct mcx_dmamem mxm;
6153 	struct mcx_cmdq_entry *cqe;
6154 	struct mcx_cmd_query_rq_in *in;
6155 	struct mcx_cmd_query_rq_out *out;
6156 	struct mcx_cmd_query_rq_mb_out *mbout;
6157 	uint8_t token = mcx_cmdq_token(sc);
6158 	int error;
6159 
6160 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
6161 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out) + sizeof(*mbout) + 16,
6162 	    token);
6163 
6164 	in = mcx_cmdq_in(cqe);
6165 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_RQ);
6166 	in->cmd_op_mod = htobe16(0);
6167 	in->cmd_rqn = htobe32(rx->rx_rqn);
6168 
6169 	CTASSERT(sizeof(*mbout) <= MCX_CMDQ_MAILBOX_DATASIZE*2);
6170 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2,
6171 	    &cqe->cq_output_ptr, token) != 0) {
6172 		printf("%s: unable to allocate query rq mailboxes\n", DEVNAME(sc));
6173 		return (-1);
6174 	}
6175 
6176 	mcx_cmdq_mboxes_sign(&mxm, 1);
6177 
6178 	mcx_cmdq_post(sc, cqe, 0);
6179 	error = mcx_cmdq_poll(sc, cqe, 1000);
6180 	if (error != 0) {
6181 		printf("%s: query rq timeout\n", DEVNAME(sc));
6182 		goto free;
6183 	}
6184 	error = mcx_cmdq_verify(cqe);
6185 	if (error != 0) {
6186 		printf("%s: query rq reply corrupt\n", DEVNAME(sc));
6187 		goto free;
6188 	}
6189 
6190 	out = mcx_cmdq_out(cqe);
6191 	switch (out->cmd_status) {
6192 	case MCX_CQ_STATUS_OK:
6193 		break;
6194 	default:
6195 		printf("%s: query rq failed (%x/%x)\n", DEVNAME(sc),
6196 		    out->cmd_status, betoh32(out->cmd_syndrome));
6197 		error = -1;
6198 		goto free;
6199 	}
6200 
6201         mbout = (struct mcx_cmd_query_rq_mb_out *)
6202 	    (mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)));
6203 	memcpy(rq_ctx, &mbout->cmd_ctx, sizeof(*rq_ctx));
6204 
6205 free:
6206 	mcx_cq_mboxes_free(sc, &mxm);
6207 	return (error);
6208 }
6209 
6210 int
6211 mcx_query_sq(struct mcx_softc *sc, struct mcx_tx *tx, struct mcx_sq_ctx *sq_ctx)
6212 {
6213 	struct mcx_dmamem mxm;
6214 	struct mcx_cmdq_entry *cqe;
6215 	struct mcx_cmd_query_sq_in *in;
6216 	struct mcx_cmd_query_sq_out *out;
6217 	struct mcx_cmd_query_sq_mb_out *mbout;
6218 	uint8_t token = mcx_cmdq_token(sc);
6219 	int error;
6220 
6221 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
6222 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out) + sizeof(*mbout) + 16,
6223 	    token);
6224 
6225 	in = mcx_cmdq_in(cqe);
6226 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_SQ);
6227 	in->cmd_op_mod = htobe16(0);
6228 	in->cmd_sqn = htobe32(tx->tx_sqn);
6229 
6230 	CTASSERT(sizeof(*mbout) <= MCX_CMDQ_MAILBOX_DATASIZE*2);
6231 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2,
6232 	    &cqe->cq_output_ptr, token) != 0) {
6233 		printf("%s: unable to allocate query sq mailboxes\n", DEVNAME(sc));
6234 		return (-1);
6235 	}
6236 
6237 	mcx_cmdq_mboxes_sign(&mxm, 1);
6238 
6239 	mcx_cmdq_post(sc, cqe, 0);
6240 	error = mcx_cmdq_poll(sc, cqe, 1000);
6241 	if (error != 0) {
6242 		printf("%s: query sq timeout\n", DEVNAME(sc));
6243 		goto free;
6244 	}
6245 	error = mcx_cmdq_verify(cqe);
6246 	if (error != 0) {
6247 		printf("%s: query sq reply corrupt\n", DEVNAME(sc));
6248 		goto free;
6249 	}
6250 
6251 	out = mcx_cmdq_out(cqe);
6252 	switch (out->cmd_status) {
6253 	case MCX_CQ_STATUS_OK:
6254 		break;
6255 	default:
6256 		printf("%s: query sq failed (%x/%x)\n", DEVNAME(sc),
6257 		    out->cmd_status, betoh32(out->cmd_syndrome));
6258 		error = -1;
6259 		goto free;
6260 	}
6261 
6262         mbout = (struct mcx_cmd_query_sq_mb_out *)
6263 	    (mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)));
6264 	memcpy(sq_ctx, &mbout->cmd_ctx, sizeof(*sq_ctx));
6265 
6266 free:
6267 	mcx_cq_mboxes_free(sc, &mxm);
6268 	return (error);
6269 }
6270 
6271 int
6272 mcx_query_cq(struct mcx_softc *sc, struct mcx_cq *cq, struct mcx_cq_ctx *cq_ctx)
6273 {
6274 	struct mcx_dmamem mxm;
6275 	struct mcx_cmdq_entry *cqe;
6276 	struct mcx_cmd_query_cq_in *in;
6277 	struct mcx_cmd_query_cq_out *out;
6278 	struct mcx_cq_ctx *ctx;
6279 	uint8_t token = mcx_cmdq_token(sc);
6280 	int error;
6281 
6282 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
6283 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out) + sizeof(*ctx) + 16,
6284 	    token);
6285 
6286 	in = mcx_cmdq_in(cqe);
6287 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_CQ);
6288 	in->cmd_op_mod = htobe16(0);
6289 	in->cmd_cqn = htobe32(cq->cq_n);
6290 
6291 	CTASSERT(sizeof(*ctx) <= MCX_CMDQ_MAILBOX_DATASIZE*2);
6292 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2,
6293 	    &cqe->cq_output_ptr, token) != 0) {
6294 		printf("%s: unable to allocate query cq mailboxes\n", DEVNAME(sc));
6295 		return (-1);
6296 	}
6297 
6298 	mcx_cmdq_mboxes_sign(&mxm, 1);
6299 
6300 	mcx_cmdq_post(sc, cqe, 0);
6301 	error = mcx_cmdq_poll(sc, cqe, 1000);
6302 	if (error != 0) {
6303 		printf("%s: query cq timeout\n", DEVNAME(sc));
6304 		goto free;
6305 	}
6306 	error = mcx_cmdq_verify(cqe);
6307 	if (error != 0) {
6308 		printf("%s: query cq reply corrupt\n", DEVNAME(sc));
6309 		goto free;
6310 	}
6311 
6312 	out = mcx_cmdq_out(cqe);
6313 	switch (out->cmd_status) {
6314 	case MCX_CQ_STATUS_OK:
6315 		break;
6316 	default:
6317 		printf("%s: query cq failed (%x/%x)\n", DEVNAME(sc),
6318 		    out->cmd_status, betoh32(out->cmd_syndrome));
6319 		error = -1;
6320 		goto free;
6321 	}
6322 
6323         ctx = (struct mcx_cq_ctx *)(mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)));
6324 	memcpy(cq_ctx, ctx, sizeof(*cq_ctx));
6325 free:
6326 	mcx_cq_mboxes_free(sc, &mxm);
6327 	return (error);
6328 }
6329 
6330 int
6331 mcx_query_eq(struct mcx_softc *sc, struct mcx_eq *eq, struct mcx_eq_ctx *eq_ctx)
6332 {
6333 	struct mcx_dmamem mxm;
6334 	struct mcx_cmdq_entry *cqe;
6335 	struct mcx_cmd_query_eq_in *in;
6336 	struct mcx_cmd_query_eq_out *out;
6337 	struct mcx_eq_ctx *ctx;
6338 	uint8_t token = mcx_cmdq_token(sc);
6339 	int error;
6340 
6341 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
6342 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out) + sizeof(*ctx) + 16,
6343 	    token);
6344 
6345 	in = mcx_cmdq_in(cqe);
6346 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_EQ);
6347 	in->cmd_op_mod = htobe16(0);
6348 	in->cmd_eqn = htobe32(eq->eq_n);
6349 
6350 	CTASSERT(sizeof(*ctx) <= MCX_CMDQ_MAILBOX_DATASIZE*2);
6351 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2,
6352 	    &cqe->cq_output_ptr, token) != 0) {
6353 		printf("%s: unable to allocate query eq mailboxes\n", DEVNAME(sc));
6354 		return (-1);
6355 	}
6356 
6357 	mcx_cmdq_mboxes_sign(&mxm, 1);
6358 
6359 	mcx_cmdq_post(sc, cqe, 0);
6360 	error = mcx_cmdq_poll(sc, cqe, 1000);
6361 	if (error != 0) {
6362 		printf("%s: query eq timeout\n", DEVNAME(sc));
6363 		goto free;
6364 	}
6365 	error = mcx_cmdq_verify(cqe);
6366 	if (error != 0) {
6367 		printf("%s: query eq reply corrupt\n", DEVNAME(sc));
6368 		goto free;
6369 	}
6370 
6371 	out = mcx_cmdq_out(cqe);
6372 	switch (out->cmd_status) {
6373 	case MCX_CQ_STATUS_OK:
6374 		break;
6375 	default:
6376 		printf("%s: query eq failed (%x/%x)\n", DEVNAME(sc),
6377 		    out->cmd_status, betoh32(out->cmd_syndrome));
6378 		error = -1;
6379 		goto free;
6380 	}
6381 
6382         ctx = (struct mcx_eq_ctx *)(mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)));
6383 	memcpy(eq_ctx, ctx, sizeof(*eq_ctx));
6384 free:
6385 	mcx_cq_mboxes_free(sc, &mxm);
6386 	return (error);
6387 }
6388 
6389 #endif /* NKSTAT > 0 */
6390 
6391 
6392 int
6393 mcx_rx_fill_slots(struct mcx_softc *sc, struct mcx_rx *rx,
6394     void *ring, struct mcx_slot *slots, uint *prod, uint nslots)
6395 {
6396 	struct mcx_rq_entry *rqe;
6397 	struct mcx_slot *ms;
6398 	struct mbuf *m;
6399 	uint slot, p, fills;
6400 
6401 	p = *prod;
6402 	slot = (p % (1 << MCX_LOG_RQ_SIZE));
6403 	rqe = ring;
6404 	for (fills = 0; fills < nslots; fills++) {
6405 		ms = &slots[slot];
6406 		m = MCLGETI(NULL, M_DONTWAIT, NULL, sc->sc_rxbufsz);
6407 		if (m == NULL)
6408 			break;
6409 
6410 		m->m_data += (m->m_ext.ext_size - sc->sc_rxbufsz);
6411 		m->m_data += ETHER_ALIGN;
6412 		m->m_len = m->m_pkthdr.len = sc->sc_hardmtu;
6413 		if (bus_dmamap_load_mbuf(sc->sc_dmat, ms->ms_map, m,
6414 		    BUS_DMA_NOWAIT) != 0) {
6415 			m_freem(m);
6416 			break;
6417 		}
6418 		ms->ms_m = m;
6419 
6420 		rqe[slot].rqe_byte_count =
6421 		    htobe32(ms->ms_map->dm_segs[0].ds_len);
6422 		rqe[slot].rqe_addr = htobe64(ms->ms_map->dm_segs[0].ds_addr);
6423 		rqe[slot].rqe_lkey = htobe32(sc->sc_lkey);
6424 
6425 		p++;
6426 		slot++;
6427 		if (slot == (1 << MCX_LOG_RQ_SIZE))
6428 			slot = 0;
6429 	}
6430 
6431 	if (fills != 0) {
6432 		*rx->rx_doorbell = htobe32(p & MCX_WQ_DOORBELL_MASK);
6433 		/* barrier? */
6434 	}
6435 
6436 	*prod = p;
6437 
6438 	return (nslots - fills);
6439 }
6440 
6441 int
6442 mcx_rx_fill(struct mcx_softc *sc, struct mcx_rx *rx)
6443 {
6444 	u_int slots;
6445 
6446 	slots = if_rxr_get(&rx->rx_rxr, (1 << MCX_LOG_RQ_SIZE));
6447 	if (slots == 0)
6448 		return (1);
6449 
6450 	slots = mcx_rx_fill_slots(sc, rx, MCX_DMA_KVA(&rx->rx_rq_mem),
6451 	    rx->rx_slots, &rx->rx_prod, slots);
6452 	if_rxr_put(&rx->rx_rxr, slots);
6453 	return (0);
6454 }
6455 
6456 void
6457 mcx_refill(void *xrx)
6458 {
6459 	struct mcx_rx *rx = xrx;
6460 	struct mcx_softc *sc = rx->rx_softc;
6461 
6462 	mcx_rx_fill(sc, rx);
6463 
6464 	if (if_rxr_inuse(&rx->rx_rxr) == 0)
6465 		timeout_add(&rx->rx_refill, 1);
6466 }
6467 
6468 void
6469 mcx_process_txeof(struct mcx_softc *sc, struct mcx_tx *tx,
6470     struct mcx_cq_entry *cqe, int *txfree)
6471 {
6472 	struct mcx_slot *ms;
6473 	bus_dmamap_t map;
6474 	int slot, slots;
6475 
6476 	slot = betoh16(cqe->cq_wqe_count) % (1 << MCX_LOG_SQ_SIZE);
6477 
6478 	ms = &tx->tx_slots[slot];
6479 	map = ms->ms_map;
6480 	bus_dmamap_sync(sc->sc_dmat, map, 0, map->dm_mapsize,
6481 	    BUS_DMASYNC_POSTWRITE);
6482 
6483 	slots = 1;
6484 	if (map->dm_nsegs > 1)
6485 		slots += (map->dm_nsegs+2) / MCX_SQ_SEGS_PER_SLOT;
6486 
6487 	(*txfree) += slots;
6488 	bus_dmamap_unload(sc->sc_dmat, map);
6489 	m_freem(ms->ms_m);
6490 	ms->ms_m = NULL;
6491 }
6492 
6493 static uint64_t
6494 mcx_uptime(void)
6495 {
6496 	struct timespec ts;
6497 
6498 	nanouptime(&ts);
6499 
6500 	return ((uint64_t)ts.tv_sec * 1000000000 + (uint64_t)ts.tv_nsec);
6501 }
6502 
6503 static void
6504 mcx_calibrate_first(struct mcx_softc *sc)
6505 {
6506 	struct mcx_calibration *c = &sc->sc_calibration[0];
6507 
6508 	sc->sc_calibration_gen = 0;
6509 
6510 	c->c_ubase = mcx_uptime();
6511 	c->c_tbase = mcx_timer(sc);
6512 	c->c_tdiff = 0;
6513 
6514 	timeout_add_sec(&sc->sc_calibrate, MCX_CALIBRATE_FIRST);
6515 }
6516 
6517 #define MCX_TIMESTAMP_SHIFT 10
6518 
6519 static void
6520 mcx_calibrate(void *arg)
6521 {
6522 	struct mcx_softc *sc = arg;
6523 	struct mcx_calibration *nc, *pc;
6524 	unsigned int gen;
6525 
6526 	if (!ISSET(sc->sc_ac.ac_if.if_flags, IFF_RUNNING))
6527 		return;
6528 
6529 	timeout_add_sec(&sc->sc_calibrate, MCX_CALIBRATE_NORMAL);
6530 
6531 	gen = sc->sc_calibration_gen;
6532 	pc = &sc->sc_calibration[gen % nitems(sc->sc_calibration)];
6533 	gen++;
6534 	nc = &sc->sc_calibration[gen % nitems(sc->sc_calibration)];
6535 
6536 	nc->c_uptime = pc->c_ubase;
6537 	nc->c_timestamp = pc->c_tbase;
6538 
6539 	nc->c_ubase = mcx_uptime();
6540 	nc->c_tbase = mcx_timer(sc);
6541 
6542 	nc->c_udiff = (nc->c_ubase - nc->c_uptime) >> MCX_TIMESTAMP_SHIFT;
6543 	nc->c_tdiff = (nc->c_tbase - nc->c_timestamp) >> MCX_TIMESTAMP_SHIFT;
6544 
6545 	membar_producer();
6546 	sc->sc_calibration_gen = gen;
6547 }
6548 
6549 static int
6550 mcx_process_rx(struct mcx_softc *sc, struct mcx_rx *rx,
6551     struct mcx_cq_entry *cqe, struct mbuf_list *ml,
6552     const struct mcx_calibration *c)
6553 {
6554 	struct mcx_slot *ms;
6555 	struct mbuf *m;
6556 	uint32_t flags;
6557 	int slot;
6558 
6559 	slot = betoh16(cqe->cq_wqe_count) % (1 << MCX_LOG_RQ_SIZE);
6560 
6561 	ms = &rx->rx_slots[slot];
6562 	bus_dmamap_sync(sc->sc_dmat, ms->ms_map, 0, ms->ms_map->dm_mapsize,
6563 	    BUS_DMASYNC_POSTREAD);
6564 	bus_dmamap_unload(sc->sc_dmat, ms->ms_map);
6565 
6566 	m = ms->ms_m;
6567 	ms->ms_m = NULL;
6568 
6569 	m->m_pkthdr.len = m->m_len = bemtoh32(&cqe->cq_byte_cnt);
6570 
6571 	if (cqe->cq_rx_hash_type) {
6572 		m->m_pkthdr.ph_flowid = betoh32(cqe->cq_rx_hash);
6573 		m->m_pkthdr.csum_flags |= M_FLOWID;
6574 	}
6575 
6576 	flags = bemtoh32(&cqe->cq_flags);
6577 	if (flags & MCX_CQ_ENTRY_FLAGS_L3_OK)
6578 		m->m_pkthdr.csum_flags = M_IPV4_CSUM_IN_OK;
6579 	if (flags & MCX_CQ_ENTRY_FLAGS_L4_OK)
6580 		m->m_pkthdr.csum_flags |= M_TCP_CSUM_IN_OK |
6581 		    M_UDP_CSUM_IN_OK;
6582 #if NVLAN > 0
6583 	if (flags & MCX_CQ_ENTRY_FLAGS_CV) {
6584 		m->m_pkthdr.ether_vtag = (flags &
6585 		    MCX_CQ_ENTRY_FLAGS_VLAN_MASK);
6586 		m->m_flags |= M_VLANTAG;
6587 	}
6588 #endif
6589 
6590 	if (c->c_tdiff) {
6591 		uint64_t t = bemtoh64(&cqe->cq_timestamp) - c->c_timestamp;
6592 		t *= c->c_udiff;
6593 		t /= c->c_tdiff;
6594 
6595 		m->m_pkthdr.ph_timestamp = c->c_uptime + t;
6596 		SET(m->m_pkthdr.csum_flags, M_TIMESTAMP);
6597 	}
6598 
6599 	ml_enqueue(ml, m);
6600 
6601 	return (1);
6602 }
6603 
6604 static struct mcx_cq_entry *
6605 mcx_next_cq_entry(struct mcx_softc *sc, struct mcx_cq *cq)
6606 {
6607 	struct mcx_cq_entry *cqe;
6608 	int next;
6609 
6610 	cqe = (struct mcx_cq_entry *)MCX_DMA_KVA(&cq->cq_mem);
6611 	next = cq->cq_cons % (1 << MCX_LOG_CQ_SIZE);
6612 
6613 	if ((cqe[next].cq_opcode_owner & MCX_CQ_ENTRY_FLAG_OWNER) ==
6614 	    ((cq->cq_cons >> MCX_LOG_CQ_SIZE) & 1)) {
6615 		return (&cqe[next]);
6616 	}
6617 
6618 	return (NULL);
6619 }
6620 
6621 static void
6622 mcx_arm_cq(struct mcx_softc *sc, struct mcx_cq *cq, int uar)
6623 {
6624 	bus_size_t offset;
6625 	uint32_t val;
6626 	uint64_t uval;
6627 
6628 	offset = (MCX_PAGE_SIZE * uar);
6629 	val = ((cq->cq_count) & 3) << MCX_CQ_DOORBELL_ARM_CMD_SN_SHIFT;
6630 	val |= (cq->cq_cons & MCX_CQ_DOORBELL_ARM_CI_MASK);
6631 
6632 	cq->cq_doorbell[0] = htobe32(cq->cq_cons & MCX_CQ_DOORBELL_ARM_CI_MASK);
6633 	cq->cq_doorbell[1] = htobe32(val);
6634 
6635 	uval = val;
6636 	uval <<= 32;
6637 	uval |= cq->cq_n;
6638 	bus_space_write_raw_8(sc->sc_memt, sc->sc_memh,
6639 	    offset + MCX_UAR_CQ_DOORBELL, htobe64(uval));
6640 	mcx_bar(sc, offset + MCX_UAR_CQ_DOORBELL, sizeof(uint64_t),
6641 	    BUS_SPACE_BARRIER_WRITE);
6642 }
6643 
6644 void
6645 mcx_process_cq(struct mcx_softc *sc, struct mcx_queues *q, struct mcx_cq *cq)
6646 {
6647 	struct mcx_rx *rx = &q->q_rx;
6648 	struct mcx_tx *tx = &q->q_tx;
6649 	const struct mcx_calibration *c;
6650 	unsigned int gen;
6651 	struct mcx_cq_entry *cqe;
6652 	uint8_t *cqp;
6653 	struct mbuf_list ml = MBUF_LIST_INITIALIZER();
6654 	int rxfree, txfree;
6655 
6656 	gen = sc->sc_calibration_gen;
6657 	membar_consumer();
6658 	c = &sc->sc_calibration[gen % nitems(sc->sc_calibration)];
6659 
6660 	rxfree = 0;
6661 	txfree = 0;
6662 	while ((cqe = mcx_next_cq_entry(sc, cq))) {
6663 		uint8_t opcode;
6664 		opcode = (cqe->cq_opcode_owner >> MCX_CQ_ENTRY_OPCODE_SHIFT);
6665 		switch (opcode) {
6666 		case MCX_CQ_ENTRY_OPCODE_REQ:
6667 			mcx_process_txeof(sc, tx, cqe, &txfree);
6668 			break;
6669 		case MCX_CQ_ENTRY_OPCODE_SEND:
6670 			rxfree += mcx_process_rx(sc, rx, cqe, &ml, c);
6671 			break;
6672 		case MCX_CQ_ENTRY_OPCODE_REQ_ERR:
6673 		case MCX_CQ_ENTRY_OPCODE_SEND_ERR:
6674 			cqp = (uint8_t *)cqe;
6675 			/* printf("%s: cq completion error: %x\n",
6676 			    DEVNAME(sc), cqp[0x37]); */
6677 			break;
6678 
6679 		default:
6680 			/* printf("%s: cq completion opcode %x??\n",
6681 			    DEVNAME(sc), opcode); */
6682 			break;
6683 		}
6684 
6685 		cq->cq_cons++;
6686 	}
6687 
6688 	cq->cq_count++;
6689 	mcx_arm_cq(sc, cq, q->q_uar);
6690 
6691 	if (rxfree > 0) {
6692 		if_rxr_put(&rx->rx_rxr, rxfree);
6693 		if (ifiq_input(rx->rx_ifiq, &ml))
6694 			if_rxr_livelocked(&rx->rx_rxr);
6695 
6696 		mcx_rx_fill(sc, rx);
6697 		if (if_rxr_inuse(&rx->rx_rxr) == 0)
6698 			timeout_add(&rx->rx_refill, 1);
6699 	}
6700 	if (txfree > 0) {
6701 		tx->tx_cons += txfree;
6702 		if (ifq_is_oactive(tx->tx_ifq))
6703 			ifq_restart(tx->tx_ifq);
6704 	}
6705 }
6706 
6707 
6708 static void
6709 mcx_arm_eq(struct mcx_softc *sc, struct mcx_eq *eq, int uar)
6710 {
6711 	bus_size_t offset;
6712 	uint32_t val;
6713 
6714 	offset = (MCX_PAGE_SIZE * uar) + MCX_UAR_EQ_DOORBELL_ARM;
6715 	val = (eq->eq_n << 24) | (eq->eq_cons & 0xffffff);
6716 
6717 	mcx_wr(sc, offset, val);
6718 	/* barrier? */
6719 }
6720 
6721 static struct mcx_eq_entry *
6722 mcx_next_eq_entry(struct mcx_softc *sc, struct mcx_eq *eq)
6723 {
6724 	struct mcx_eq_entry *eqe;
6725 	int next;
6726 
6727 	eqe = (struct mcx_eq_entry *)MCX_DMA_KVA(&eq->eq_mem);
6728 	next = eq->eq_cons % (1 << MCX_LOG_EQ_SIZE);
6729 	if ((eqe[next].eq_owner & 1) ==
6730 	    ((eq->eq_cons >> MCX_LOG_EQ_SIZE) & 1)) {
6731 		eq->eq_cons++;
6732 		return (&eqe[next]);
6733 	}
6734 	return (NULL);
6735 }
6736 
6737 int
6738 mcx_admin_intr(void *xsc)
6739 {
6740 	struct mcx_softc *sc = (struct mcx_softc *)xsc;
6741 	struct mcx_eq_entry *eqe;
6742 
6743 	while ((eqe = mcx_next_eq_entry(sc, &sc->sc_admin_eq))) {
6744 		switch (eqe->eq_event_type) {
6745 		case MCX_EVENT_TYPE_LAST_WQE:
6746 			/* printf("%s: last wqe reached?\n", DEVNAME(sc)); */
6747 			break;
6748 
6749 		case MCX_EVENT_TYPE_CQ_ERROR:
6750 			/* printf("%s: cq error\n", DEVNAME(sc)); */
6751 			break;
6752 
6753 		case MCX_EVENT_TYPE_CMD_COMPLETION:
6754 			/* wakeup probably */
6755 			break;
6756 
6757 		case MCX_EVENT_TYPE_PORT_CHANGE:
6758 			task_add(systq, &sc->sc_port_change);
6759 			break;
6760 
6761 		default:
6762 			/* printf("%s: something happened\n", DEVNAME(sc)); */
6763 			break;
6764 		}
6765 	}
6766 	mcx_arm_eq(sc, &sc->sc_admin_eq, sc->sc_uar);
6767 	return (1);
6768 }
6769 
6770 int
6771 mcx_cq_intr(void *xq)
6772 {
6773 	struct mcx_queues *q = (struct mcx_queues *)xq;
6774 	struct mcx_softc *sc = q->q_sc;
6775 	struct mcx_eq_entry *eqe;
6776 	int cqn;
6777 
6778 	while ((eqe = mcx_next_eq_entry(sc, &q->q_eq))) {
6779 		switch (eqe->eq_event_type) {
6780 		case MCX_EVENT_TYPE_COMPLETION:
6781 			cqn = betoh32(eqe->eq_event_data[6]);
6782 			if (cqn == q->q_cq.cq_n)
6783 				mcx_process_cq(sc, q, &q->q_cq);
6784 			break;
6785 		}
6786 	}
6787 
6788 	mcx_arm_eq(sc, &q->q_eq, q->q_uar);
6789 	return (1);
6790 }
6791 
6792 
6793 static void
6794 mcx_free_slots(struct mcx_softc *sc, struct mcx_slot *slots, int allocated,
6795     int total)
6796 {
6797 	struct mcx_slot *ms;
6798 
6799 	int i = allocated;
6800 	while (i-- > 0) {
6801 		ms = &slots[i];
6802 		bus_dmamap_destroy(sc->sc_dmat, ms->ms_map);
6803 		if (ms->ms_m != NULL)
6804 			m_freem(ms->ms_m);
6805 	}
6806 	free(slots, M_DEVBUF, total * sizeof(*ms));
6807 }
6808 
6809 static int
6810 mcx_queue_up(struct mcx_softc *sc, struct mcx_queues *q)
6811 {
6812 	struct mcx_rx *rx;
6813 	struct mcx_tx *tx;
6814 	struct mcx_slot *ms;
6815 	int i;
6816 
6817 	rx = &q->q_rx;
6818 	rx->rx_slots = mallocarray(sizeof(*ms), (1 << MCX_LOG_RQ_SIZE),
6819 	    M_DEVBUF, M_WAITOK | M_ZERO);
6820 	if (rx->rx_slots == NULL) {
6821 		printf("%s: failed to allocate rx slots\n", DEVNAME(sc));
6822 		return ENOMEM;
6823 	}
6824 
6825 	for (i = 0; i < (1 << MCX_LOG_RQ_SIZE); i++) {
6826 		ms = &rx->rx_slots[i];
6827 		if (bus_dmamap_create(sc->sc_dmat, sc->sc_hardmtu, 1,
6828 		    sc->sc_hardmtu, 0,
6829 		    BUS_DMA_WAITOK | BUS_DMA_ALLOCNOW | BUS_DMA_64BIT,
6830 		    &ms->ms_map) != 0) {
6831 			printf("%s: failed to allocate rx dma maps\n",
6832 			    DEVNAME(sc));
6833 			goto destroy_rx_slots;
6834 		}
6835 	}
6836 
6837 	tx = &q->q_tx;
6838 	tx->tx_slots = mallocarray(sizeof(*ms), (1 << MCX_LOG_SQ_SIZE),
6839 	    M_DEVBUF, M_WAITOK | M_ZERO);
6840 	if (tx->tx_slots == NULL) {
6841 		printf("%s: failed to allocate tx slots\n", DEVNAME(sc));
6842 		goto destroy_rx_slots;
6843 	}
6844 
6845 	for (i = 0; i < (1 << MCX_LOG_SQ_SIZE); i++) {
6846 		ms = &tx->tx_slots[i];
6847 		if (bus_dmamap_create(sc->sc_dmat, sc->sc_hardmtu,
6848 		    MCX_SQ_MAX_SEGMENTS, sc->sc_hardmtu, 0,
6849 		    BUS_DMA_WAITOK | BUS_DMA_ALLOCNOW | BUS_DMA_64BIT,
6850 		    &ms->ms_map) != 0) {
6851 			printf("%s: failed to allocate tx dma maps\n",
6852 			    DEVNAME(sc));
6853 			goto destroy_tx_slots;
6854 		}
6855 	}
6856 
6857 	if (mcx_create_cq(sc, &q->q_cq, q->q_uar, q->q_index,
6858 	    q->q_eq.eq_n) != 0)
6859 		return ENOMEM;
6860 
6861 	if (mcx_create_sq(sc, tx, q->q_uar, q->q_index, q->q_cq.cq_n)
6862 	    != 0)
6863 		return ENOMEM;
6864 
6865 	if (mcx_create_rq(sc, rx, q->q_index, q->q_cq.cq_n) != 0)
6866 		return ENOMEM;
6867 
6868 	return 0;
6869 destroy_tx_slots:
6870 	mcx_free_slots(sc, tx->tx_slots, i, (1 << MCX_LOG_SQ_SIZE));
6871 	tx->tx_slots = NULL;
6872 
6873 	i = (1 << MCX_LOG_RQ_SIZE);
6874 destroy_rx_slots:
6875 	mcx_free_slots(sc, rx->rx_slots, i, (1 << MCX_LOG_RQ_SIZE));
6876 	rx->rx_slots = NULL;
6877 	return ENOMEM;
6878 }
6879 
6880 static int
6881 mcx_rss_group_entry_count(struct mcx_softc *sc, int group)
6882 {
6883 	int i;
6884 	int count;
6885 
6886 	count = 0;
6887 	for (i = 0; i < nitems(mcx_rss_config); i++) {
6888 		if (mcx_rss_config[i].flow_group == group)
6889 			count++;
6890 	}
6891 
6892 	return count;
6893 }
6894 
6895 static int
6896 mcx_up(struct mcx_softc *sc)
6897 {
6898 	struct ifnet *ifp = &sc->sc_ac.ac_if;
6899 	struct mcx_rx *rx;
6900 	struct mcx_tx *tx;
6901 	int i, start, count, flow_group, flow_index;
6902 	struct mcx_flow_match match_crit;
6903 	struct mcx_rss_rule *rss;
6904 	uint32_t dest;
6905 	int rqns[MCX_MAX_QUEUES];
6906 
6907 	if (mcx_create_tis(sc, &sc->sc_tis) != 0)
6908 		goto down;
6909 
6910 	for (i = 0; i < sc->sc_nqueues; i++) {
6911 		if (mcx_queue_up(sc, &sc->sc_queues[i]) != 0) {
6912 			goto down;
6913 		}
6914 	}
6915 
6916 	/* RSS flow table and flow groups */
6917 	if (mcx_create_flow_table(sc, MCX_LOG_FLOW_TABLE_SIZE, 1,
6918 	    &sc->sc_rss_flow_table_id) != 0)
6919 		goto down;
6920 
6921 	dest = MCX_FLOW_CONTEXT_DEST_TYPE_TABLE |
6922 	    sc->sc_rss_flow_table_id;
6923 
6924 	/* L4 RSS flow group (v4/v6 tcp/udp, no fragments) */
6925 	memset(&match_crit, 0, sizeof(match_crit));
6926 	match_crit.mc_ethertype = 0xffff;
6927 	match_crit.mc_ip_proto = 0xff;
6928 	match_crit.mc_vlan_flags = MCX_FLOW_MATCH_IP_FRAG;
6929 	start = 0;
6930 	count = mcx_rss_group_entry_count(sc, MCX_FLOW_GROUP_RSS_L4);
6931 	if (count != 0) {
6932 		if (mcx_create_flow_group(sc, sc->sc_rss_flow_table_id,
6933 		    MCX_FLOW_GROUP_RSS_L4, start, count,
6934 		    MCX_CREATE_FLOW_GROUP_CRIT_OUTER, &match_crit) != 0)
6935 			goto down;
6936 		start += count;
6937 	}
6938 
6939 	/* L3 RSS flow group (v4/v6, including fragments) */
6940 	memset(&match_crit, 0, sizeof(match_crit));
6941 	match_crit.mc_ethertype = 0xffff;
6942 	count = mcx_rss_group_entry_count(sc, MCX_FLOW_GROUP_RSS_L3);
6943 	if (mcx_create_flow_group(sc, sc->sc_rss_flow_table_id,
6944 	    MCX_FLOW_GROUP_RSS_L3, start, count,
6945 	    MCX_CREATE_FLOW_GROUP_CRIT_OUTER, &match_crit) != 0)
6946 		goto down;
6947 	start += count;
6948 
6949 	/* non-RSS flow group */
6950 	count = mcx_rss_group_entry_count(sc, MCX_FLOW_GROUP_RSS_NONE);
6951 	memset(&match_crit, 0, sizeof(match_crit));
6952 	if (mcx_create_flow_group(sc, sc->sc_rss_flow_table_id,
6953 	    MCX_FLOW_GROUP_RSS_NONE, start, count, 0, &match_crit) != 0)
6954 		goto down;
6955 
6956 	/* Root flow table, matching packets based on mac address */
6957 	if (mcx_create_flow_table(sc, MCX_LOG_FLOW_TABLE_SIZE, 0,
6958 	    &sc->sc_mac_flow_table_id) != 0)
6959 		goto down;
6960 
6961 	/* promisc flow group */
6962 	start = 0;
6963 	memset(&match_crit, 0, sizeof(match_crit));
6964 	if (mcx_create_flow_group(sc, sc->sc_mac_flow_table_id,
6965 	    MCX_FLOW_GROUP_PROMISC, start, 1, 0, &match_crit) != 0)
6966 		goto down;
6967 	sc->sc_promisc_flow_enabled = 0;
6968 	start++;
6969 
6970 	/* all multicast flow group */
6971 	match_crit.mc_dest_mac[0] = 0x01;
6972 	if (mcx_create_flow_group(sc, sc->sc_mac_flow_table_id,
6973 	    MCX_FLOW_GROUP_ALLMULTI, start, 1,
6974 	    MCX_CREATE_FLOW_GROUP_CRIT_OUTER, &match_crit) != 0)
6975 		goto down;
6976 	sc->sc_allmulti_flow_enabled = 0;
6977 	start++;
6978 
6979 	/* mac address matching flow group */
6980 	memset(&match_crit.mc_dest_mac, 0xff, sizeof(match_crit.mc_dest_mac));
6981 	if (mcx_create_flow_group(sc, sc->sc_mac_flow_table_id,
6982 	    MCX_FLOW_GROUP_MAC, start, (1 << MCX_LOG_FLOW_TABLE_SIZE) - start,
6983 	    MCX_CREATE_FLOW_GROUP_CRIT_OUTER, &match_crit) != 0)
6984 		goto down;
6985 
6986 	/* flow table entries for unicast and broadcast */
6987 	start = 0;
6988 	if (mcx_set_flow_table_entry_mac(sc, MCX_FLOW_GROUP_MAC, start,
6989 	    sc->sc_ac.ac_enaddr, dest) != 0)
6990 		goto down;
6991 	start++;
6992 
6993 	if (mcx_set_flow_table_entry_mac(sc, MCX_FLOW_GROUP_MAC, start,
6994 	    etherbroadcastaddr, dest) != 0)
6995 		goto down;
6996 	start++;
6997 
6998 	/* multicast entries go after that */
6999 	sc->sc_mcast_flow_base = start;
7000 
7001 	/* re-add any existing multicast flows */
7002 	for (i = 0; i < MCX_NUM_MCAST_FLOWS; i++) {
7003 		if (sc->sc_mcast_flows[i][0] != 0) {
7004 			mcx_set_flow_table_entry_mac(sc, MCX_FLOW_GROUP_MAC,
7005 			    sc->sc_mcast_flow_base + i,
7006 			    sc->sc_mcast_flows[i], dest);
7007 		}
7008 	}
7009 
7010 	if (mcx_set_flow_table_root(sc, sc->sc_mac_flow_table_id) != 0)
7011 		goto down;
7012 
7013 	/*
7014 	 * the RQT can be any size as long as it's a power of two.
7015 	 * since we also restrict the number of queues to a power of two,
7016 	 * we can just put each rx queue in once.
7017 	 */
7018 	for (i = 0; i < sc->sc_nqueues; i++)
7019 		rqns[i] = sc->sc_queues[i].q_rx.rx_rqn;
7020 
7021 	if (mcx_create_rqt(sc, sc->sc_nqueues, rqns, &sc->sc_rqt) != 0)
7022 		goto down;
7023 
7024 	start = 0;
7025 	flow_index = 0;
7026 	flow_group = -1;
7027 	for (i = 0; i < nitems(mcx_rss_config); i++) {
7028 		rss = &mcx_rss_config[i];
7029 		if (rss->flow_group != flow_group) {
7030 			flow_group = rss->flow_group;
7031 			flow_index = 0;
7032 		}
7033 
7034 		if (rss->hash_sel == 0) {
7035 			if (mcx_create_tir_direct(sc, &sc->sc_queues[0].q_rx,
7036 			    &sc->sc_tir[i]) != 0)
7037 				goto down;
7038 		} else {
7039 			if (mcx_create_tir_indirect(sc, sc->sc_rqt,
7040 			    rss->hash_sel, &sc->sc_tir[i]) != 0)
7041 				goto down;
7042 		}
7043 
7044 		if (mcx_set_flow_table_entry_proto(sc, flow_group,
7045 		    flow_index, rss->ethertype, rss->ip_proto,
7046 		    MCX_FLOW_CONTEXT_DEST_TYPE_TIR | sc->sc_tir[i]) != 0)
7047 			goto down;
7048 		flow_index++;
7049 	}
7050 
7051 	for (i = 0; i < sc->sc_nqueues; i++) {
7052 		struct mcx_queues *q = &sc->sc_queues[i];
7053 		rx = &q->q_rx;
7054 		tx = &q->q_tx;
7055 
7056 		/* start the queues */
7057 		if (mcx_ready_sq(sc, tx) != 0)
7058 			goto down;
7059 
7060 		if (mcx_ready_rq(sc, rx) != 0)
7061 			goto down;
7062 
7063 		if_rxr_init(&rx->rx_rxr, 1, (1 << MCX_LOG_RQ_SIZE));
7064 		rx->rx_prod = 0;
7065 		mcx_rx_fill(sc, rx);
7066 
7067 		tx->tx_cons = 0;
7068 		tx->tx_prod = 0;
7069 		ifq_clr_oactive(tx->tx_ifq);
7070 	}
7071 
7072 	mcx_calibrate_first(sc);
7073 
7074 	SET(ifp->if_flags, IFF_RUNNING);
7075 
7076 	return ENETRESET;
7077 down:
7078 	mcx_down(sc);
7079 	return ENOMEM;
7080 }
7081 
7082 static void
7083 mcx_down(struct mcx_softc *sc)
7084 {
7085 	struct ifnet *ifp = &sc->sc_ac.ac_if;
7086 	struct mcx_rss_rule *rss;
7087 	int group, i, flow_group, flow_index;
7088 
7089 	CLR(ifp->if_flags, IFF_RUNNING);
7090 
7091 	/*
7092 	 * delete flow table entries first, so no packets can arrive
7093 	 * after the barriers
7094 	 */
7095 	if (sc->sc_promisc_flow_enabled)
7096 		mcx_delete_flow_table_entry(sc, MCX_FLOW_GROUP_PROMISC, 0);
7097 	if (sc->sc_allmulti_flow_enabled)
7098 		mcx_delete_flow_table_entry(sc, MCX_FLOW_GROUP_ALLMULTI, 0);
7099 	mcx_delete_flow_table_entry(sc, MCX_FLOW_GROUP_MAC, 0);
7100 	mcx_delete_flow_table_entry(sc, MCX_FLOW_GROUP_MAC, 1);
7101 	for (i = 0; i < MCX_NUM_MCAST_FLOWS; i++) {
7102 		if (sc->sc_mcast_flows[i][0] != 0) {
7103 			mcx_delete_flow_table_entry(sc, MCX_FLOW_GROUP_MAC,
7104 			    sc->sc_mcast_flow_base + i);
7105 		}
7106 	}
7107 
7108 	flow_group = -1;
7109 	flow_index = 0;
7110 	for (i = 0; i < nitems(mcx_rss_config); i++) {
7111 		rss = &mcx_rss_config[i];
7112 		if (rss->flow_group != flow_group) {
7113 			flow_group = rss->flow_group;
7114 			flow_index = 0;
7115 		}
7116 
7117 		mcx_delete_flow_table_entry(sc, flow_group, flow_index);
7118 
7119 		mcx_destroy_tir(sc, sc->sc_tir[i]);
7120 		sc->sc_tir[i] = 0;
7121 
7122 		flow_index++;
7123 	}
7124 	intr_barrier(sc->sc_ihc);
7125 	for (i = 0; i < sc->sc_nqueues; i++) {
7126 		struct ifqueue *ifq = sc->sc_queues[i].q_tx.tx_ifq;
7127 		ifq_barrier(ifq);
7128 
7129 		timeout_del_barrier(&sc->sc_queues[i].q_rx.rx_refill);
7130 
7131 		intr_barrier(sc->sc_queues[i].q_ihc);
7132 	}
7133 
7134 	timeout_del_barrier(&sc->sc_calibrate);
7135 
7136 	for (group = 0; group < MCX_NUM_FLOW_GROUPS; group++) {
7137 		if (sc->sc_flow_group[group].g_id != -1)
7138 			mcx_destroy_flow_group(sc, group);
7139 	}
7140 
7141 	if (sc->sc_mac_flow_table_id != -1) {
7142 		mcx_destroy_flow_table(sc, sc->sc_mac_flow_table_id);
7143 		sc->sc_mac_flow_table_id = -1;
7144 	}
7145 	if (sc->sc_rss_flow_table_id != -1) {
7146 		mcx_destroy_flow_table(sc, sc->sc_rss_flow_table_id);
7147 		sc->sc_rss_flow_table_id = -1;
7148 	}
7149 	if (sc->sc_rqt != -1) {
7150 		mcx_destroy_rqt(sc, sc->sc_rqt);
7151 		sc->sc_rqt = -1;
7152 	}
7153 
7154 	for (i = 0; i < sc->sc_nqueues; i++) {
7155 		struct mcx_queues *q = &sc->sc_queues[i];
7156 		struct mcx_rx *rx = &q->q_rx;
7157 		struct mcx_tx *tx = &q->q_tx;
7158 		struct mcx_cq *cq = &q->q_cq;
7159 
7160 		if (rx->rx_rqn != 0)
7161 			mcx_destroy_rq(sc, rx);
7162 
7163 		if (tx->tx_sqn != 0)
7164 			mcx_destroy_sq(sc, tx);
7165 
7166 		if (tx->tx_slots != NULL) {
7167 			mcx_free_slots(sc, tx->tx_slots,
7168 			    (1 << MCX_LOG_SQ_SIZE), (1 << MCX_LOG_SQ_SIZE));
7169 			tx->tx_slots = NULL;
7170 		}
7171 		if (rx->rx_slots != NULL) {
7172 			mcx_free_slots(sc, rx->rx_slots,
7173 			    (1 << MCX_LOG_RQ_SIZE), (1 << MCX_LOG_RQ_SIZE));
7174 			rx->rx_slots = NULL;
7175 		}
7176 
7177 		if (cq->cq_n != 0)
7178 			mcx_destroy_cq(sc, cq);
7179 	}
7180 	if (sc->sc_tis != 0) {
7181 		mcx_destroy_tis(sc, sc->sc_tis);
7182 		sc->sc_tis = 0;
7183 	}
7184 }
7185 
7186 static int
7187 mcx_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
7188 {
7189 	struct mcx_softc *sc = (struct mcx_softc *)ifp->if_softc;
7190 	struct ifreq *ifr = (struct ifreq *)data;
7191 	uint8_t addrhi[ETHER_ADDR_LEN], addrlo[ETHER_ADDR_LEN];
7192 	int s, i, error = 0;
7193 	uint32_t dest;
7194 
7195 	s = splnet();
7196 	switch (cmd) {
7197 	case SIOCSIFADDR:
7198 		ifp->if_flags |= IFF_UP;
7199 		/* FALLTHROUGH */
7200 
7201 	case SIOCSIFFLAGS:
7202 		if (ISSET(ifp->if_flags, IFF_UP)) {
7203 			if (ISSET(ifp->if_flags, IFF_RUNNING))
7204 				error = ENETRESET;
7205 			else
7206 				error = mcx_up(sc);
7207 		} else {
7208 			if (ISSET(ifp->if_flags, IFF_RUNNING))
7209 				mcx_down(sc);
7210 		}
7211 		break;
7212 
7213 	case SIOCGIFMEDIA:
7214 	case SIOCSIFMEDIA:
7215 		error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
7216 		break;
7217 
7218 	case SIOCGIFSFFPAGE:
7219 		error = mcx_get_sffpage(ifp, (struct if_sffpage *)data);
7220 		break;
7221 
7222 	case SIOCGIFRXR:
7223 		error = mcx_rxrinfo(sc, (struct if_rxrinfo *)ifr->ifr_data);
7224 		break;
7225 
7226 	case SIOCADDMULTI:
7227 		if (ether_addmulti(ifr, &sc->sc_ac) == ENETRESET) {
7228 			error = ether_multiaddr(&ifr->ifr_addr, addrlo, addrhi);
7229 			if (error != 0)
7230 				return (error);
7231 
7232 			dest = MCX_FLOW_CONTEXT_DEST_TYPE_TABLE |
7233 			    sc->sc_rss_flow_table_id;
7234 
7235 			for (i = 0; i < MCX_NUM_MCAST_FLOWS; i++) {
7236 				if (sc->sc_mcast_flows[i][0] == 0) {
7237 					memcpy(sc->sc_mcast_flows[i], addrlo,
7238 					    ETHER_ADDR_LEN);
7239 					if (ISSET(ifp->if_flags, IFF_RUNNING)) {
7240 						mcx_set_flow_table_entry_mac(sc,
7241 						    MCX_FLOW_GROUP_MAC,
7242 						    sc->sc_mcast_flow_base + i,
7243 						    sc->sc_mcast_flows[i], dest);
7244 					}
7245 					break;
7246 				}
7247 			}
7248 
7249 			if (!ISSET(ifp->if_flags, IFF_ALLMULTI)) {
7250 				if (i == MCX_NUM_MCAST_FLOWS) {
7251 					SET(ifp->if_flags, IFF_ALLMULTI);
7252 					sc->sc_extra_mcast++;
7253 					error = ENETRESET;
7254 				}
7255 
7256 				if (sc->sc_ac.ac_multirangecnt > 0) {
7257 					SET(ifp->if_flags, IFF_ALLMULTI);
7258 					error = ENETRESET;
7259 				}
7260 			}
7261 		}
7262 		break;
7263 
7264 	case SIOCDELMULTI:
7265 		if (ether_delmulti(ifr, &sc->sc_ac) == ENETRESET) {
7266 			error = ether_multiaddr(&ifr->ifr_addr, addrlo, addrhi);
7267 			if (error != 0)
7268 				return (error);
7269 
7270 			for (i = 0; i < MCX_NUM_MCAST_FLOWS; i++) {
7271 				if (memcmp(sc->sc_mcast_flows[i], addrlo,
7272 				    ETHER_ADDR_LEN) == 0) {
7273 					if (ISSET(ifp->if_flags, IFF_RUNNING)) {
7274 						mcx_delete_flow_table_entry(sc,
7275 						    MCX_FLOW_GROUP_MAC,
7276 						    sc->sc_mcast_flow_base + i);
7277 					}
7278 					sc->sc_mcast_flows[i][0] = 0;
7279 					break;
7280 				}
7281 			}
7282 
7283 			if (i == MCX_NUM_MCAST_FLOWS)
7284 				sc->sc_extra_mcast--;
7285 
7286 			if (ISSET(ifp->if_flags, IFF_ALLMULTI) &&
7287 			    (sc->sc_extra_mcast == 0) &&
7288 			    (sc->sc_ac.ac_multirangecnt == 0)) {
7289 				CLR(ifp->if_flags, IFF_ALLMULTI);
7290 				error = ENETRESET;
7291 			}
7292 		}
7293 		break;
7294 
7295 	default:
7296 		error = ether_ioctl(ifp, &sc->sc_ac, cmd, data);
7297 	}
7298 
7299 	if (error == ENETRESET) {
7300 		if ((ifp->if_flags & (IFF_UP | IFF_RUNNING)) ==
7301 		    (IFF_UP | IFF_RUNNING))
7302 			mcx_iff(sc);
7303 		error = 0;
7304 	}
7305 	splx(s);
7306 
7307 	return (error);
7308 }
7309 
7310 static int
7311 mcx_get_sffpage(struct ifnet *ifp, struct if_sffpage *sff)
7312 {
7313 	struct mcx_softc *sc = (struct mcx_softc *)ifp->if_softc;
7314 	struct mcx_reg_mcia mcia;
7315 	struct mcx_reg_pmlp pmlp;
7316 	int offset, error;
7317 
7318 	/* get module number */
7319 	memset(&pmlp, 0, sizeof(pmlp));
7320 	pmlp.rp_local_port = 1;
7321 	error = mcx_access_hca_reg(sc, MCX_REG_PMLP, MCX_REG_OP_READ, &pmlp,
7322 	    sizeof(pmlp));
7323 	if (error != 0) {
7324 		printf("%s: unable to get eeprom module number\n",
7325 		    DEVNAME(sc));
7326 		return error;
7327 	}
7328 
7329 	for (offset = 0; offset < 256; offset += MCX_MCIA_EEPROM_BYTES) {
7330 		memset(&mcia, 0, sizeof(mcia));
7331 		mcia.rm_l = 0;
7332 		mcia.rm_module = betoh32(pmlp.rp_lane0_mapping) &
7333 		    MCX_PMLP_MODULE_NUM_MASK;
7334 		mcia.rm_i2c_addr = sff->sff_addr / 2;	/* apparently */
7335 		mcia.rm_page_num = sff->sff_page;
7336 		mcia.rm_dev_addr = htobe16(offset);
7337 		mcia.rm_size = htobe16(MCX_MCIA_EEPROM_BYTES);
7338 
7339 		error = mcx_access_hca_reg(sc, MCX_REG_MCIA, MCX_REG_OP_READ,
7340 		    &mcia, sizeof(mcia));
7341 		if (error != 0) {
7342 			printf("%s: unable to read eeprom at %x\n",
7343 			    DEVNAME(sc), offset);
7344 			return error;
7345 		}
7346 
7347 		memcpy(sff->sff_data + offset, mcia.rm_data,
7348 		    MCX_MCIA_EEPROM_BYTES);
7349 	}
7350 
7351 	return 0;
7352 }
7353 
7354 static int
7355 mcx_rxrinfo(struct mcx_softc *sc, struct if_rxrinfo *ifri)
7356 {
7357 	struct if_rxring_info *ifrs;
7358 	unsigned int i;
7359 	int error;
7360 
7361 	ifrs = mallocarray(sc->sc_nqueues, sizeof(*ifrs), M_TEMP,
7362 	    M_WAITOK|M_ZERO|M_CANFAIL);
7363 	if (ifrs == NULL)
7364 		return (ENOMEM);
7365 
7366 	for (i = 0; i < sc->sc_nqueues; i++) {
7367 		struct mcx_rx *rx = &sc->sc_queues[i].q_rx;
7368 		struct if_rxring_info *ifr = &ifrs[i];
7369 
7370 		snprintf(ifr->ifr_name, sizeof(ifr->ifr_name), "%u", i);
7371 		ifr->ifr_size = sc->sc_hardmtu;
7372 		ifr->ifr_info = rx->rx_rxr;
7373 	}
7374 
7375 	error = if_rxr_info_ioctl(ifri, i, ifrs);
7376 	free(ifrs, M_TEMP, i * sizeof(*ifrs));
7377 
7378 	return (error);
7379 }
7380 
7381 int
7382 mcx_load_mbuf(struct mcx_softc *sc, struct mcx_slot *ms, struct mbuf *m)
7383 {
7384 	switch (bus_dmamap_load_mbuf(sc->sc_dmat, ms->ms_map, m,
7385 	    BUS_DMA_STREAMING | BUS_DMA_NOWAIT)) {
7386 	case 0:
7387 		break;
7388 
7389 	case EFBIG:
7390 		if (m_defrag(m, M_DONTWAIT) == 0 &&
7391 		    bus_dmamap_load_mbuf(sc->sc_dmat, ms->ms_map, m,
7392 		    BUS_DMA_STREAMING | BUS_DMA_NOWAIT) == 0)
7393 			break;
7394 
7395 	default:
7396 		return (1);
7397 	}
7398 
7399 	ms->ms_m = m;
7400 	return (0);
7401 }
7402 
7403 static void
7404 mcx_start(struct ifqueue *ifq)
7405 {
7406 	struct mcx_tx *tx = ifq->ifq_softc;
7407 	struct ifnet *ifp = ifq->ifq_if;
7408 	struct mcx_softc *sc = ifp->if_softc;
7409 	struct mcx_sq_entry *sq, *sqe;
7410 	struct mcx_sq_entry_seg *sqs;
7411 	struct mcx_slot *ms;
7412 	bus_dmamap_t map;
7413 	struct mbuf *m;
7414 	u_int idx, free, used;
7415 	uint64_t *bf;
7416 	uint32_t csum;
7417 	size_t bf_base;
7418 	int i, seg, nseg;
7419 
7420 	bf_base = (tx->tx_uar * MCX_PAGE_SIZE) + MCX_UAR_BF;
7421 
7422 	idx = tx->tx_prod % (1 << MCX_LOG_SQ_SIZE);
7423 	free = (tx->tx_cons + (1 << MCX_LOG_SQ_SIZE)) - tx->tx_prod;
7424 
7425 	used = 0;
7426 	bf = NULL;
7427 	sq = (struct mcx_sq_entry *)MCX_DMA_KVA(&tx->tx_sq_mem);
7428 
7429 	for (;;) {
7430 		if (used + MCX_SQ_ENTRY_MAX_SLOTS >= free) {
7431 			ifq_set_oactive(ifq);
7432 			break;
7433 		}
7434 
7435 		m = ifq_dequeue(ifq);
7436 		if (m == NULL) {
7437 			break;
7438 		}
7439 
7440 		sqe = sq + idx;
7441 		ms = &tx->tx_slots[idx];
7442 		memset(sqe, 0, sizeof(*sqe));
7443 
7444 		/* ctrl segment */
7445 		sqe->sqe_opcode_index = htobe32(MCX_SQE_WQE_OPCODE_SEND |
7446 		    ((tx->tx_prod & 0xffff) << MCX_SQE_WQE_INDEX_SHIFT));
7447 		/* always generate a completion event */
7448 		sqe->sqe_signature = htobe32(MCX_SQE_CE_CQE_ALWAYS);
7449 
7450 		/* eth segment */
7451 		csum = 0;
7452 		if (m->m_pkthdr.csum_flags & M_IPV4_CSUM_OUT)
7453 			csum |= MCX_SQE_L3_CSUM;
7454 		if (m->m_pkthdr.csum_flags & (M_TCP_CSUM_OUT | M_UDP_CSUM_OUT))
7455 			csum |= MCX_SQE_L4_CSUM;
7456 		sqe->sqe_mss_csum = htobe32(csum);
7457 		sqe->sqe_inline_header_size = htobe16(MCX_SQ_INLINE_SIZE);
7458 #if NVLAN > 0
7459 		if (m->m_flags & M_VLANTAG) {
7460 			struct ether_vlan_header *evh;
7461 			evh = (struct ether_vlan_header *)
7462 			    &sqe->sqe_inline_headers;
7463 
7464 			/* slightly cheaper vlan_inject() */
7465 			m_copydata(m, 0, ETHER_HDR_LEN, (caddr_t)evh);
7466 			evh->evl_proto = evh->evl_encap_proto;
7467 			evh->evl_encap_proto = htons(ETHERTYPE_VLAN);
7468 			evh->evl_tag = htons(m->m_pkthdr.ether_vtag);
7469 
7470 			m_adj(m, ETHER_HDR_LEN);
7471 		} else
7472 #endif
7473 		{
7474 			m_copydata(m, 0, MCX_SQ_INLINE_SIZE,
7475 			    (caddr_t)sqe->sqe_inline_headers);
7476 			m_adj(m, MCX_SQ_INLINE_SIZE);
7477 		}
7478 
7479 		if (mcx_load_mbuf(sc, ms, m) != 0) {
7480 			m_freem(m);
7481 			ifp->if_oerrors++;
7482 			continue;
7483 		}
7484 		bf = (uint64_t *)sqe;
7485 
7486 #if NBPFILTER > 0
7487 		if (ifp->if_bpf)
7488 			bpf_mtap_hdr(ifp->if_bpf,
7489 			    (caddr_t)sqe->sqe_inline_headers,
7490 			    MCX_SQ_INLINE_SIZE, m, BPF_DIRECTION_OUT);
7491 #endif
7492 		map = ms->ms_map;
7493 		bus_dmamap_sync(sc->sc_dmat, map, 0, map->dm_mapsize,
7494 		    BUS_DMASYNC_PREWRITE);
7495 
7496 		sqe->sqe_ds_sq_num =
7497 		    htobe32((tx->tx_sqn << MCX_SQE_SQ_NUM_SHIFT) |
7498 		    (map->dm_nsegs + 3));
7499 
7500 		/* data segment - first wqe has one segment */
7501 		sqs = sqe->sqe_segs;
7502 		seg = 0;
7503 		nseg = 1;
7504 		for (i = 0; i < map->dm_nsegs; i++) {
7505 			if (seg == nseg) {
7506 				/* next slot */
7507 				idx++;
7508 				if (idx == (1 << MCX_LOG_SQ_SIZE))
7509 					idx = 0;
7510 				tx->tx_prod++;
7511 				used++;
7512 
7513 				sqs = (struct mcx_sq_entry_seg *)(sq + idx);
7514 				seg = 0;
7515 				nseg = MCX_SQ_SEGS_PER_SLOT;
7516 			}
7517 			sqs[seg].sqs_byte_count =
7518 			    htobe32(map->dm_segs[i].ds_len);
7519 			sqs[seg].sqs_lkey = htobe32(sc->sc_lkey);
7520 			sqs[seg].sqs_addr = htobe64(map->dm_segs[i].ds_addr);
7521 			seg++;
7522 		}
7523 
7524 		idx++;
7525 		if (idx == (1 << MCX_LOG_SQ_SIZE))
7526 			idx = 0;
7527 		tx->tx_prod++;
7528 		used++;
7529 	}
7530 
7531 	if (used) {
7532 		htobem32(tx->tx_doorbell, tx->tx_prod & MCX_WQ_DOORBELL_MASK);
7533 
7534 		membar_sync();
7535 
7536 		/*
7537 		 * write the first 64 bits of the last sqe we produced
7538 		 * to the blue flame buffer
7539 		 */
7540 		bus_space_write_raw_8(sc->sc_memt, sc->sc_memh,
7541 		    bf_base + tx->tx_bf_offset, *bf);
7542 		/* next write goes to the other buffer */
7543 		tx->tx_bf_offset ^= sc->sc_bf_size;
7544 
7545 		membar_sync();
7546 	}
7547 }
7548 
7549 static void
7550 mcx_watchdog(struct ifnet *ifp)
7551 {
7552 }
7553 
7554 static void
7555 mcx_media_add_types(struct mcx_softc *sc)
7556 {
7557 	struct mcx_reg_ptys ptys;
7558 	int i;
7559 	uint32_t proto_cap;
7560 
7561 	memset(&ptys, 0, sizeof(ptys));
7562 	ptys.rp_local_port = 1;
7563 	ptys.rp_proto_mask = MCX_REG_PTYS_PROTO_MASK_ETH;
7564 	if (mcx_access_hca_reg(sc, MCX_REG_PTYS, MCX_REG_OP_READ, &ptys,
7565 	    sizeof(ptys)) != 0) {
7566 		printf("%s: unable to read port type/speed\n", DEVNAME(sc));
7567 		return;
7568 	}
7569 
7570 	proto_cap = betoh32(ptys.rp_eth_proto_cap);
7571 	for (i = 0; i < nitems(mcx_eth_cap_map); i++) {
7572 		const struct mcx_eth_proto_capability *cap;
7573 		if (!ISSET(proto_cap, 1 << i))
7574 			continue;
7575 
7576 		cap = &mcx_eth_cap_map[i];
7577 		if (cap->cap_media == 0)
7578 			continue;
7579 
7580 		ifmedia_add(&sc->sc_media, IFM_ETHER | cap->cap_media, 0, NULL);
7581 	}
7582 }
7583 
7584 static void
7585 mcx_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
7586 {
7587 	struct mcx_softc *sc = (struct mcx_softc *)ifp->if_softc;
7588 	struct mcx_reg_ptys ptys;
7589 	int i;
7590 	uint32_t proto_oper;
7591 	uint64_t media_oper;
7592 
7593 	memset(&ptys, 0, sizeof(ptys));
7594 	ptys.rp_local_port = 1;
7595 	ptys.rp_proto_mask = MCX_REG_PTYS_PROTO_MASK_ETH;
7596 
7597 	if (mcx_access_hca_reg(sc, MCX_REG_PTYS, MCX_REG_OP_READ, &ptys,
7598 	    sizeof(ptys)) != 0) {
7599 		printf("%s: unable to read port type/speed\n", DEVNAME(sc));
7600 		return;
7601 	}
7602 
7603 	proto_oper = betoh32(ptys.rp_eth_proto_oper);
7604 
7605 	media_oper = 0;
7606 
7607 	for (i = 0; i < nitems(mcx_eth_cap_map); i++) {
7608 		const struct mcx_eth_proto_capability *cap;
7609 		if (!ISSET(proto_oper, 1 << i))
7610 			continue;
7611 
7612 		cap = &mcx_eth_cap_map[i];
7613 
7614 		if (cap->cap_media != 0)
7615 			media_oper = cap->cap_media;
7616 	}
7617 
7618 	ifmr->ifm_status = IFM_AVALID;
7619 	if (proto_oper != 0) {
7620 		ifmr->ifm_status |= IFM_ACTIVE;
7621 		ifmr->ifm_active = IFM_ETHER | IFM_AUTO | media_oper;
7622 		/* txpause, rxpause, duplex? */
7623 	}
7624 }
7625 
7626 static int
7627 mcx_media_change(struct ifnet *ifp)
7628 {
7629 	struct mcx_softc *sc = (struct mcx_softc *)ifp->if_softc;
7630 	struct mcx_reg_ptys ptys;
7631 	struct mcx_reg_paos paos;
7632 	uint32_t media;
7633 	int i, error;
7634 
7635 	if (IFM_TYPE(sc->sc_media.ifm_media) != IFM_ETHER)
7636 		return EINVAL;
7637 
7638 	error = 0;
7639 
7640 	if (IFM_SUBTYPE(sc->sc_media.ifm_media) == IFM_AUTO) {
7641 		/* read ptys to get supported media */
7642 		memset(&ptys, 0, sizeof(ptys));
7643 		ptys.rp_local_port = 1;
7644 		ptys.rp_proto_mask = MCX_REG_PTYS_PROTO_MASK_ETH;
7645 		if (mcx_access_hca_reg(sc, MCX_REG_PTYS, MCX_REG_OP_READ,
7646 		    &ptys, sizeof(ptys)) != 0) {
7647 			printf("%s: unable to read port type/speed\n",
7648 			    DEVNAME(sc));
7649 			return EIO;
7650 		}
7651 
7652 		media = betoh32(ptys.rp_eth_proto_cap);
7653 	} else {
7654 		/* map media type */
7655 		media = 0;
7656 		for (i = 0; i < nitems(mcx_eth_cap_map); i++) {
7657 			const struct  mcx_eth_proto_capability *cap;
7658 
7659 			cap = &mcx_eth_cap_map[i];
7660 			if (cap->cap_media ==
7661 			    IFM_SUBTYPE(sc->sc_media.ifm_media)) {
7662 				media = (1 << i);
7663 				break;
7664 			}
7665 		}
7666 	}
7667 
7668 	/* disable the port */
7669 	memset(&paos, 0, sizeof(paos));
7670 	paos.rp_local_port = 1;
7671 	paos.rp_admin_status = MCX_REG_PAOS_ADMIN_STATUS_DOWN;
7672 	paos.rp_admin_state_update = MCX_REG_PAOS_ADMIN_STATE_UPDATE_EN;
7673 	if (mcx_access_hca_reg(sc, MCX_REG_PAOS, MCX_REG_OP_WRITE, &paos,
7674 	    sizeof(paos)) != 0) {
7675 		printf("%s: unable to set port state to down\n", DEVNAME(sc));
7676 		return EIO;
7677 	}
7678 
7679 	memset(&ptys, 0, sizeof(ptys));
7680 	ptys.rp_local_port = 1;
7681 	ptys.rp_proto_mask = MCX_REG_PTYS_PROTO_MASK_ETH;
7682 	ptys.rp_eth_proto_admin = htobe32(media);
7683 	if (mcx_access_hca_reg(sc, MCX_REG_PTYS, MCX_REG_OP_WRITE, &ptys,
7684 	    sizeof(ptys)) != 0) {
7685 		printf("%s: unable to set port media type/speed\n",
7686 		    DEVNAME(sc));
7687 		error = EIO;
7688 	}
7689 
7690 	/* re-enable the port to start negotiation */
7691 	memset(&paos, 0, sizeof(paos));
7692 	paos.rp_local_port = 1;
7693 	paos.rp_admin_status = MCX_REG_PAOS_ADMIN_STATUS_UP;
7694 	paos.rp_admin_state_update = MCX_REG_PAOS_ADMIN_STATE_UPDATE_EN;
7695 	if (mcx_access_hca_reg(sc, MCX_REG_PAOS, MCX_REG_OP_WRITE, &paos,
7696 	    sizeof(paos)) != 0) {
7697 		printf("%s: unable to set port state to up\n", DEVNAME(sc));
7698 		error = EIO;
7699 	}
7700 
7701 	return error;
7702 }
7703 
7704 static void
7705 mcx_port_change(void *xsc)
7706 {
7707 	struct mcx_softc *sc = xsc;
7708 	struct ifnet *ifp = &sc->sc_ac.ac_if;
7709 	struct mcx_reg_ptys ptys = {
7710 		.rp_local_port = 1,
7711 		.rp_proto_mask = MCX_REG_PTYS_PROTO_MASK_ETH,
7712 	};
7713 	int link_state = LINK_STATE_DOWN;
7714 
7715 	if (mcx_access_hca_reg(sc, MCX_REG_PTYS, MCX_REG_OP_READ, &ptys,
7716 	    sizeof(ptys)) == 0) {
7717 		uint32_t proto_oper = betoh32(ptys.rp_eth_proto_oper);
7718 		uint64_t baudrate = 0;
7719 		unsigned int i;
7720 
7721 		if (proto_oper != 0)
7722 			link_state = LINK_STATE_FULL_DUPLEX;
7723 
7724 		for (i = 0; i < nitems(mcx_eth_cap_map); i++) {
7725 			const struct mcx_eth_proto_capability *cap;
7726 			if (!ISSET(proto_oper, 1 << i))
7727 				continue;
7728 
7729 			cap = &mcx_eth_cap_map[i];
7730 			if (cap->cap_baudrate == 0)
7731 				continue;
7732 
7733 			baudrate = cap->cap_baudrate;
7734 			break;
7735 		}
7736 
7737 		ifp->if_baudrate = baudrate;
7738 	}
7739 
7740 	if (link_state != ifp->if_link_state) {
7741 		ifp->if_link_state = link_state;
7742 		if_link_state_change(ifp);
7743 	}
7744 }
7745 
7746 static inline uint32_t
7747 mcx_rd(struct mcx_softc *sc, bus_size_t r)
7748 {
7749 	uint32_t word;
7750 
7751 	word = bus_space_read_raw_4(sc->sc_memt, sc->sc_memh, r);
7752 
7753 	return (betoh32(word));
7754 }
7755 
7756 static inline void
7757 mcx_wr(struct mcx_softc *sc, bus_size_t r, uint32_t v)
7758 {
7759 	bus_space_write_raw_4(sc->sc_memt, sc->sc_memh, r, htobe32(v));
7760 }
7761 
7762 static inline void
7763 mcx_bar(struct mcx_softc *sc, bus_size_t r, bus_size_t l, int f)
7764 {
7765 	bus_space_barrier(sc->sc_memt, sc->sc_memh, r, l, f);
7766 }
7767 
7768 static uint64_t
7769 mcx_timer(struct mcx_softc *sc)
7770 {
7771 	uint32_t hi, lo, ni;
7772 
7773 	hi = mcx_rd(sc, MCX_INTERNAL_TIMER_H);
7774 	for (;;) {
7775 		lo = mcx_rd(sc, MCX_INTERNAL_TIMER_L);
7776 		mcx_bar(sc, MCX_INTERNAL_TIMER_L, 8, BUS_SPACE_BARRIER_READ);
7777 		ni = mcx_rd(sc, MCX_INTERNAL_TIMER_H);
7778 
7779 		if (ni == hi)
7780 			break;
7781 
7782 		hi = ni;
7783 	}
7784 
7785 	return (((uint64_t)hi << 32) | (uint64_t)lo);
7786 }
7787 
7788 static int
7789 mcx_dmamem_alloc(struct mcx_softc *sc, struct mcx_dmamem *mxm,
7790     bus_size_t size, u_int align)
7791 {
7792 	mxm->mxm_size = size;
7793 
7794 	if (bus_dmamap_create(sc->sc_dmat, mxm->mxm_size, 1,
7795 	    mxm->mxm_size, 0,
7796 	    BUS_DMA_WAITOK | BUS_DMA_ALLOCNOW | BUS_DMA_64BIT,
7797 	    &mxm->mxm_map) != 0)
7798 		return (1);
7799 	if (bus_dmamem_alloc(sc->sc_dmat, mxm->mxm_size,
7800 	    align, 0, &mxm->mxm_seg, 1, &mxm->mxm_nsegs,
7801 	    BUS_DMA_WAITOK | BUS_DMA_ZERO) != 0)
7802 		goto destroy;
7803 	if (bus_dmamem_map(sc->sc_dmat, &mxm->mxm_seg, mxm->mxm_nsegs,
7804 	    mxm->mxm_size, &mxm->mxm_kva, BUS_DMA_WAITOK) != 0)
7805 		goto free;
7806 	if (bus_dmamap_load(sc->sc_dmat, mxm->mxm_map, mxm->mxm_kva,
7807 	    mxm->mxm_size, NULL, BUS_DMA_WAITOK) != 0)
7808 		goto unmap;
7809 
7810 	return (0);
7811 unmap:
7812 	bus_dmamem_unmap(sc->sc_dmat, mxm->mxm_kva, mxm->mxm_size);
7813 free:
7814 	bus_dmamem_free(sc->sc_dmat, &mxm->mxm_seg, 1);
7815 destroy:
7816 	bus_dmamap_destroy(sc->sc_dmat, mxm->mxm_map);
7817 	return (1);
7818 }
7819 
7820 static void
7821 mcx_dmamem_zero(struct mcx_dmamem *mxm)
7822 {
7823 	memset(MCX_DMA_KVA(mxm), 0, MCX_DMA_LEN(mxm));
7824 }
7825 
7826 static void
7827 mcx_dmamem_free(struct mcx_softc *sc, struct mcx_dmamem *mxm)
7828 {
7829 	bus_dmamap_unload(sc->sc_dmat, mxm->mxm_map);
7830 	bus_dmamem_unmap(sc->sc_dmat, mxm->mxm_kva, mxm->mxm_size);
7831 	bus_dmamem_free(sc->sc_dmat, &mxm->mxm_seg, 1);
7832 	bus_dmamap_destroy(sc->sc_dmat, mxm->mxm_map);
7833 }
7834 
7835 static int
7836 mcx_hwmem_alloc(struct mcx_softc *sc, struct mcx_hwmem *mhm, unsigned int pages)
7837 {
7838 	bus_dma_segment_t *segs;
7839 	bus_size_t len = pages * MCX_PAGE_SIZE;
7840 	size_t seglen;
7841 
7842 	segs = mallocarray(sizeof(*segs), pages, M_DEVBUF, M_WAITOK|M_CANFAIL);
7843 	if (segs == NULL)
7844 		return (-1);
7845 
7846 	seglen = sizeof(*segs) * pages;
7847 
7848 	if (bus_dmamem_alloc(sc->sc_dmat, len, MCX_PAGE_SIZE, 0,
7849 	    segs, pages, &mhm->mhm_seg_count, BUS_DMA_NOWAIT) != 0)
7850 		goto free_segs;
7851 
7852 	if (mhm->mhm_seg_count < pages) {
7853 		size_t nseglen;
7854 
7855 		mhm->mhm_segs = mallocarray(sizeof(*mhm->mhm_segs),
7856 		    mhm->mhm_seg_count, M_DEVBUF, M_WAITOK|M_CANFAIL);
7857 		if (mhm->mhm_segs == NULL)
7858 			goto free_dmamem;
7859 
7860 		nseglen = sizeof(*mhm->mhm_segs) * mhm->mhm_seg_count;
7861 
7862 		memcpy(mhm->mhm_segs, segs, nseglen);
7863 
7864 		free(segs, M_DEVBUF, seglen);
7865 
7866 		segs = mhm->mhm_segs;
7867 		seglen = nseglen;
7868 	} else
7869 		mhm->mhm_segs = segs;
7870 
7871 	if (bus_dmamap_create(sc->sc_dmat, len, pages, MCX_PAGE_SIZE,
7872 	    MCX_PAGE_SIZE, BUS_DMA_NOWAIT|BUS_DMA_ALLOCNOW /*|BUS_DMA_64BIT*/,
7873 	    &mhm->mhm_map) != 0)
7874 		goto free_dmamem;
7875 
7876 	if (bus_dmamap_load_raw(sc->sc_dmat, mhm->mhm_map,
7877 	    mhm->mhm_segs, mhm->mhm_seg_count, len, BUS_DMA_NOWAIT) != 0)
7878 		goto destroy;
7879 
7880 	bus_dmamap_sync(sc->sc_dmat, mhm->mhm_map,
7881 	    0, mhm->mhm_map->dm_mapsize, BUS_DMASYNC_PRERW);
7882 
7883 	mhm->mhm_npages = pages;
7884 
7885 	return (0);
7886 
7887 destroy:
7888 	bus_dmamap_destroy(sc->sc_dmat, mhm->mhm_map);
7889 free_dmamem:
7890 	bus_dmamem_free(sc->sc_dmat, mhm->mhm_segs, mhm->mhm_seg_count);
7891 free_segs:
7892 	free(segs, M_DEVBUF, seglen);
7893 	mhm->mhm_segs = NULL;
7894 
7895 	return (-1);
7896 }
7897 
7898 static void
7899 mcx_hwmem_free(struct mcx_softc *sc, struct mcx_hwmem *mhm)
7900 {
7901 	if (mhm->mhm_npages == 0)
7902 		return;
7903 
7904 	bus_dmamap_sync(sc->sc_dmat, mhm->mhm_map,
7905 	    0, mhm->mhm_map->dm_mapsize, BUS_DMASYNC_POSTRW);
7906 
7907 	bus_dmamap_unload(sc->sc_dmat, mhm->mhm_map);
7908 	bus_dmamap_destroy(sc->sc_dmat, mhm->mhm_map);
7909 	bus_dmamem_free(sc->sc_dmat, mhm->mhm_segs, mhm->mhm_seg_count);
7910 	free(mhm->mhm_segs, M_DEVBUF,
7911 	    sizeof(*mhm->mhm_segs) * mhm->mhm_seg_count);
7912 
7913 	mhm->mhm_npages = 0;
7914 }
7915 
7916 #if NKSTAT > 0
7917 struct mcx_ppcnt {
7918 	char			 name[KSTAT_KV_NAMELEN];
7919 	enum kstat_kv_unit	 unit;
7920 };
7921 
7922 static const struct mcx_ppcnt mcx_ppcnt_ieee8023_tpl[] = {
7923 	{ "Good Tx",		KSTAT_KV_U_PACKETS, },
7924 	{ "Good Rx",		KSTAT_KV_U_PACKETS, },
7925 	{ "FCS errs",		KSTAT_KV_U_PACKETS, },
7926 	{ "Alignment Errs",	KSTAT_KV_U_PACKETS, },
7927 	{ "Good Tx",		KSTAT_KV_U_BYTES, },
7928 	{ "Good Rx",		KSTAT_KV_U_BYTES, },
7929 	{ "Multicast Tx",	KSTAT_KV_U_PACKETS, },
7930 	{ "Broadcast Tx",	KSTAT_KV_U_PACKETS, },
7931 	{ "Multicast Rx",	KSTAT_KV_U_PACKETS, },
7932 	{ "Broadcast Rx",	KSTAT_KV_U_PACKETS, },
7933 	{ "In Range Len",	KSTAT_KV_U_PACKETS, },
7934 	{ "Out Of Range Len",	KSTAT_KV_U_PACKETS, },
7935 	{ "Frame Too Long",	KSTAT_KV_U_PACKETS, },
7936 	{ "Symbol Errs",	KSTAT_KV_U_PACKETS, },
7937 	{ "MAC Ctrl Tx",	KSTAT_KV_U_PACKETS, },
7938 	{ "MAC Ctrl Rx",	KSTAT_KV_U_PACKETS, },
7939 	{ "MAC Ctrl Unsup",	KSTAT_KV_U_PACKETS, },
7940 	{ "Pause Rx",		KSTAT_KV_U_PACKETS, },
7941 	{ "Pause Tx",		KSTAT_KV_U_PACKETS, },
7942 };
7943 CTASSERT(nitems(mcx_ppcnt_ieee8023_tpl) == mcx_ppcnt_ieee8023_count);
7944 
7945 static const struct mcx_ppcnt mcx_ppcnt_rfc2863_tpl[] = {
7946 	{ "Rx Bytes",		KSTAT_KV_U_BYTES, },
7947 	{ "Rx Unicast",		KSTAT_KV_U_PACKETS, },
7948 	{ "Rx Discards",	KSTAT_KV_U_PACKETS, },
7949 	{ "Rx Errors",		KSTAT_KV_U_PACKETS, },
7950 	{ "Rx Unknown Proto",	KSTAT_KV_U_PACKETS, },
7951 	{ "Tx Bytes",		KSTAT_KV_U_BYTES, },
7952 	{ "Tx Unicast",		KSTAT_KV_U_PACKETS, },
7953 	{ "Tx Discards",	KSTAT_KV_U_PACKETS, },
7954 	{ "Tx Errors",		KSTAT_KV_U_PACKETS, },
7955 	{ "Rx Multicast",	KSTAT_KV_U_PACKETS, },
7956 	{ "Rx Broadcast",	KSTAT_KV_U_PACKETS, },
7957 	{ "Tx Multicast",	KSTAT_KV_U_PACKETS, },
7958 	{ "Tx Broadcast",	KSTAT_KV_U_PACKETS, },
7959 };
7960 CTASSERT(nitems(mcx_ppcnt_rfc2863_tpl) == mcx_ppcnt_rfc2863_count);
7961 
7962 static const struct mcx_ppcnt mcx_ppcnt_rfc2819_tpl[] = {
7963 	{ "Drop Events",	KSTAT_KV_U_PACKETS, },
7964 	{ "Octets",		KSTAT_KV_U_BYTES, },
7965 	{ "Packets",		KSTAT_KV_U_PACKETS, },
7966 	{ "Broadcasts",		KSTAT_KV_U_PACKETS, },
7967 	{ "Multicasts",		KSTAT_KV_U_PACKETS, },
7968 	{ "CRC Align Errs",	KSTAT_KV_U_PACKETS, },
7969 	{ "Undersize",		KSTAT_KV_U_PACKETS, },
7970 	{ "Oversize",		KSTAT_KV_U_PACKETS, },
7971 	{ "Fragments",		KSTAT_KV_U_PACKETS, },
7972 	{ "Jabbers",		KSTAT_KV_U_PACKETS, },
7973 	{ "Collisions",		KSTAT_KV_U_NONE, },
7974 	{ "64B",		KSTAT_KV_U_PACKETS, },
7975 	{ "65-127B",		KSTAT_KV_U_PACKETS, },
7976 	{ "128-255B",		KSTAT_KV_U_PACKETS, },
7977 	{ "256-511B",		KSTAT_KV_U_PACKETS, },
7978 	{ "512-1023B",		KSTAT_KV_U_PACKETS, },
7979 	{ "1024-1518B",		KSTAT_KV_U_PACKETS, },
7980 	{ "1519-2047B",		KSTAT_KV_U_PACKETS, },
7981 	{ "2048-4095B",		KSTAT_KV_U_PACKETS, },
7982 	{ "4096-8191B",		KSTAT_KV_U_PACKETS, },
7983 	{ "8192-10239B",	KSTAT_KV_U_PACKETS, },
7984 };
7985 CTASSERT(nitems(mcx_ppcnt_rfc2819_tpl) == mcx_ppcnt_rfc2819_count);
7986 
7987 static const struct mcx_ppcnt mcx_ppcnt_rfc3635_tpl[] = {
7988 	{ "Alignment Errs",	KSTAT_KV_U_PACKETS, },
7989 	{ "FCS Errs",		KSTAT_KV_U_PACKETS, },
7990 	{ "Single Colls",	KSTAT_KV_U_PACKETS, },
7991 	{ "Multiple Colls",	KSTAT_KV_U_PACKETS, },
7992 	{ "SQE Test Errs",	KSTAT_KV_U_NONE, },
7993 	{ "Deferred Tx",	KSTAT_KV_U_PACKETS, },
7994 	{ "Late Colls",		KSTAT_KV_U_NONE, },
7995 	{ "Exess Colls",	KSTAT_KV_U_NONE, },
7996 	{ "Int MAC Tx Errs",	KSTAT_KV_U_PACKETS, },
7997 	{ "CSM Sense Errs",	KSTAT_KV_U_NONE, },
7998 	{ "Too Long",		KSTAT_KV_U_PACKETS, },
7999 	{ "Int MAC Rx Errs",	KSTAT_KV_U_PACKETS, },
8000 	{ "Symbol Errs",	KSTAT_KV_U_NONE, },
8001 	{ "Unknown Control",	KSTAT_KV_U_PACKETS, },
8002 	{ "Pause Rx",		KSTAT_KV_U_PACKETS, },
8003 	{ "Pause Tx",		KSTAT_KV_U_PACKETS, },
8004 };
8005 CTASSERT(nitems(mcx_ppcnt_rfc3635_tpl) == mcx_ppcnt_rfc3635_count);
8006 
8007 struct mcx_kstat_ppcnt {
8008 	const char		*ksp_name;
8009 	const struct mcx_ppcnt	*ksp_tpl;
8010 	unsigned int		 ksp_n;
8011 	uint8_t			 ksp_grp;
8012 };
8013 
8014 static const struct mcx_kstat_ppcnt mcx_kstat_ppcnt_ieee8023 = {
8015 	.ksp_name =		"ieee802.3",
8016 	.ksp_tpl =		mcx_ppcnt_ieee8023_tpl,
8017 	.ksp_n =		nitems(mcx_ppcnt_ieee8023_tpl),
8018 	.ksp_grp =		MCX_REG_PPCNT_GRP_IEEE8023,
8019 };
8020 
8021 static const struct mcx_kstat_ppcnt mcx_kstat_ppcnt_rfc2863 = {
8022 	.ksp_name =		"rfc2863",
8023 	.ksp_tpl =		mcx_ppcnt_rfc2863_tpl,
8024 	.ksp_n =		nitems(mcx_ppcnt_rfc2863_tpl),
8025 	.ksp_grp =		MCX_REG_PPCNT_GRP_RFC2863,
8026 };
8027 
8028 static const struct mcx_kstat_ppcnt mcx_kstat_ppcnt_rfc2819 = {
8029 	.ksp_name =		"rfc2819",
8030 	.ksp_tpl =		mcx_ppcnt_rfc2819_tpl,
8031 	.ksp_n =		nitems(mcx_ppcnt_rfc2819_tpl),
8032 	.ksp_grp =		MCX_REG_PPCNT_GRP_RFC2819,
8033 };
8034 
8035 static const struct mcx_kstat_ppcnt mcx_kstat_ppcnt_rfc3635 = {
8036 	.ksp_name =		"rfc3635",
8037 	.ksp_tpl =		mcx_ppcnt_rfc3635_tpl,
8038 	.ksp_n =		nitems(mcx_ppcnt_rfc3635_tpl),
8039 	.ksp_grp =		MCX_REG_PPCNT_GRP_RFC3635,
8040 };
8041 
8042 static int	mcx_kstat_ppcnt_read(struct kstat *);
8043 
8044 static void	mcx_kstat_attach_tmps(struct mcx_softc *sc);
8045 static void	mcx_kstat_attach_queues(struct mcx_softc *sc);
8046 
8047 static struct kstat *
8048 mcx_kstat_attach_ppcnt(struct mcx_softc *sc,
8049     const struct mcx_kstat_ppcnt *ksp)
8050 {
8051 	struct kstat *ks;
8052 	struct kstat_kv *kvs;
8053 	unsigned int i;
8054 
8055 	ks = kstat_create(DEVNAME(sc), 0, ksp->ksp_name, 0, KSTAT_T_KV, 0);
8056 	if (ks == NULL)
8057 		return (NULL);
8058 
8059 	kvs = mallocarray(ksp->ksp_n, sizeof(*kvs),
8060 	    M_DEVBUF, M_WAITOK);
8061 
8062 	for (i = 0; i < ksp->ksp_n; i++) {
8063 		const struct mcx_ppcnt *tpl = &ksp->ksp_tpl[i];
8064 
8065 		kstat_kv_unit_init(&kvs[i], tpl->name,
8066 		    KSTAT_KV_T_COUNTER64, tpl->unit);
8067 	}
8068 
8069 	ks->ks_softc = sc;
8070 	ks->ks_ptr = (void *)ksp;
8071 	ks->ks_data = kvs;
8072 	ks->ks_datalen = ksp->ksp_n * sizeof(*kvs);
8073 	ks->ks_read = mcx_kstat_ppcnt_read;
8074 
8075 	kstat_install(ks);
8076 
8077 	return (ks);
8078 }
8079 
8080 static void
8081 mcx_kstat_attach(struct mcx_softc *sc)
8082 {
8083 	sc->sc_kstat_ieee8023 = mcx_kstat_attach_ppcnt(sc,
8084 	    &mcx_kstat_ppcnt_ieee8023);
8085 	sc->sc_kstat_rfc2863 = mcx_kstat_attach_ppcnt(sc,
8086 	    &mcx_kstat_ppcnt_rfc2863);
8087 	sc->sc_kstat_rfc2819 = mcx_kstat_attach_ppcnt(sc,
8088 	    &mcx_kstat_ppcnt_rfc2819);
8089 	sc->sc_kstat_rfc3635 = mcx_kstat_attach_ppcnt(sc,
8090 	    &mcx_kstat_ppcnt_rfc3635);
8091 
8092 	mcx_kstat_attach_tmps(sc);
8093 	mcx_kstat_attach_queues(sc);
8094 }
8095 
8096 static int
8097 mcx_kstat_ppcnt_read(struct kstat *ks)
8098 {
8099 	struct mcx_softc *sc = ks->ks_softc;
8100 	struct mcx_kstat_ppcnt *ksp = ks->ks_ptr;
8101 	struct mcx_reg_ppcnt ppcnt = {
8102 		.ppcnt_grp = ksp->ksp_grp,
8103 		.ppcnt_local_port = 1,
8104 	};
8105 	struct kstat_kv *kvs = ks->ks_data;
8106 	uint64_t *vs = (uint64_t *)&ppcnt.ppcnt_counter_set;
8107 	unsigned int i;
8108 	int rv;
8109 
8110 	KERNEL_LOCK(); /* XXX */
8111 	rv = mcx_access_hca_reg(sc, MCX_REG_PPCNT, MCX_REG_OP_READ,
8112 	    &ppcnt, sizeof(ppcnt));
8113 	KERNEL_UNLOCK();
8114 	if (rv != 0)
8115 		return (EIO);
8116 
8117 	nanouptime(&ks->ks_updated);
8118 
8119 	for (i = 0; i < ksp->ksp_n; i++)
8120 		kstat_kv_u64(&kvs[i]) = bemtoh64(&vs[i]);
8121 
8122 	return (0);
8123 }
8124 
8125 struct mcx_kstat_mtmp {
8126 	struct kstat_kv		ktmp_name;
8127 	struct kstat_kv		ktmp_temperature;
8128 	struct kstat_kv		ktmp_threshold_lo;
8129 	struct kstat_kv		ktmp_threshold_hi;
8130 };
8131 
8132 static const struct mcx_kstat_mtmp mcx_kstat_mtmp_tpl = {
8133 	KSTAT_KV_INITIALIZER("name",		KSTAT_KV_T_ISTR),
8134 	KSTAT_KV_INITIALIZER("temperature",	KSTAT_KV_T_TEMP),
8135 	KSTAT_KV_INITIALIZER("lo threshold",	KSTAT_KV_T_TEMP),
8136 	KSTAT_KV_INITIALIZER("hi threshold",	KSTAT_KV_T_TEMP),
8137 };
8138 
8139 static const struct timeval mcx_kstat_mtmp_rate = { 1, 0 };
8140 
8141 static int mcx_kstat_mtmp_read(struct kstat *);
8142 
8143 static void
8144 mcx_kstat_attach_tmps(struct mcx_softc *sc)
8145 {
8146 	struct kstat *ks;
8147 	struct mcx_reg_mtcap mtcap;
8148 	struct mcx_kstat_mtmp *ktmp;
8149 	uint64_t map;
8150 	unsigned int i, n;
8151 
8152 	memset(&mtcap, 0, sizeof(mtcap));
8153 
8154 	if (mcx_access_hca_reg(sc, MCX_REG_MTCAP, MCX_REG_OP_READ,
8155 	    &mtcap, sizeof(mtcap)) != 0) {
8156 		/* unable to find temperature sensors */
8157 		return;
8158 	}
8159 
8160 	sc->sc_kstat_mtmp_count = mtcap.mtcap_sensor_count;
8161 	sc->sc_kstat_mtmp = mallocarray(sc->sc_kstat_mtmp_count,
8162 	    sizeof(*sc->sc_kstat_mtmp), M_DEVBUF, M_WAITOK);
8163 
8164 	n = 0;
8165 	map = bemtoh64(&mtcap.mtcap_sensor_map);
8166 	for (i = 0; i < sizeof(map) * NBBY; i++) {
8167 		if (!ISSET(map, (1ULL << i)))
8168 			continue;
8169 
8170 		ks = kstat_create(DEVNAME(sc), 0, "temperature", i,
8171 		    KSTAT_T_KV, 0);
8172 		if (ks == NULL) {
8173 			/* unable to attach temperature sensor %u, i */
8174 			continue;
8175 		}
8176 
8177 		ktmp = malloc(sizeof(*ktmp), M_DEVBUF, M_WAITOK|M_ZERO);
8178 		*ktmp = mcx_kstat_mtmp_tpl;
8179 
8180 		ks->ks_data = ktmp;
8181 		ks->ks_datalen = sizeof(*ktmp);
8182 		TIMEVAL_TO_TIMESPEC(&mcx_kstat_mtmp_rate, &ks->ks_interval);
8183 		ks->ks_read = mcx_kstat_mtmp_read;
8184 
8185 		ks->ks_softc = sc;
8186 		kstat_install(ks);
8187 
8188 		sc->sc_kstat_mtmp[n++] = ks;
8189 		if (n >= sc->sc_kstat_mtmp_count)
8190 			break;
8191 	}
8192 }
8193 
8194 static uint64_t
8195 mcx_tmp_to_uK(uint16_t *t)
8196 {
8197 	int64_t mt = (int16_t)bemtoh16(t); /* 0.125 C units */
8198 	mt *= 1000000 / 8; /* convert to uC */
8199 	mt += 273150000; /* convert to uK */
8200 
8201 	return (mt);
8202 }
8203 
8204 static int
8205 mcx_kstat_mtmp_read(struct kstat *ks)
8206 {
8207 	struct mcx_softc *sc = ks->ks_softc;
8208 	struct mcx_kstat_mtmp *ktmp = ks->ks_data;
8209 	struct mcx_reg_mtmp mtmp;
8210 	int rv;
8211 	struct timeval updated;
8212 
8213 	TIMESPEC_TO_TIMEVAL(&updated, &ks->ks_updated);
8214 
8215 	if (!ratecheck(&updated, &mcx_kstat_mtmp_rate))
8216 		return (0);
8217 
8218 	memset(&mtmp, 0, sizeof(mtmp));
8219 	htobem16(&mtmp.mtmp_sensor_index, ks->ks_unit);
8220 
8221 	KERNEL_LOCK(); /* XXX */
8222 	rv = mcx_access_hca_reg(sc, MCX_REG_MTMP, MCX_REG_OP_READ,
8223 	    &mtmp, sizeof(mtmp));
8224 	KERNEL_UNLOCK();
8225 	if (rv != 0)
8226 		return (EIO);
8227 
8228 	memset(kstat_kv_istr(&ktmp->ktmp_name), 0,
8229 	    sizeof(kstat_kv_istr(&ktmp->ktmp_name)));
8230 	memcpy(kstat_kv_istr(&ktmp->ktmp_name),
8231 	    mtmp.mtmp_sensor_name, sizeof(mtmp.mtmp_sensor_name));
8232 	kstat_kv_temp(&ktmp->ktmp_temperature) =
8233 	    mcx_tmp_to_uK(&mtmp.mtmp_temperature);
8234 	kstat_kv_temp(&ktmp->ktmp_threshold_lo) =
8235 	    mcx_tmp_to_uK(&mtmp.mtmp_temperature_threshold_lo);
8236 	kstat_kv_temp(&ktmp->ktmp_threshold_hi) =
8237 	    mcx_tmp_to_uK(&mtmp.mtmp_temperature_threshold_hi);
8238 
8239 	TIMEVAL_TO_TIMESPEC(&updated, &ks->ks_updated);
8240 
8241 	return (0);
8242 }
8243 
8244 struct mcx_queuestat {
8245 	char			 name[KSTAT_KV_NAMELEN];
8246 	enum kstat_kv_type	 type;
8247 };
8248 
8249 static const struct mcx_queuestat mcx_queue_kstat_tpl[] = {
8250 	{ "RQ SW prod",		KSTAT_KV_T_COUNTER64 },
8251 	{ "RQ HW prod",		KSTAT_KV_T_COUNTER64 },
8252 	{ "RQ HW cons",		KSTAT_KV_T_COUNTER64 },
8253 	{ "RQ HW state",	KSTAT_KV_T_ISTR },
8254 
8255 	{ "SQ SW prod",		KSTAT_KV_T_COUNTER64 },
8256 	{ "SQ SW cons",		KSTAT_KV_T_COUNTER64 },
8257 	{ "SQ HW prod",		KSTAT_KV_T_COUNTER64 },
8258 	{ "SQ HW cons",		KSTAT_KV_T_COUNTER64 },
8259 	{ "SQ HW state",	KSTAT_KV_T_ISTR },
8260 
8261 	{ "CQ SW cons",		KSTAT_KV_T_COUNTER64 },
8262 	{ "CQ HW prod",		KSTAT_KV_T_COUNTER64 },
8263 	{ "CQ HW cons",		KSTAT_KV_T_COUNTER64 },
8264 	{ "CQ HW notify",	KSTAT_KV_T_COUNTER64 },
8265 	{ "CQ HW solicit",	KSTAT_KV_T_COUNTER64 },
8266 	{ "CQ HW status",	KSTAT_KV_T_ISTR },
8267 	{ "CQ HW state",	KSTAT_KV_T_ISTR },
8268 
8269 	{ "EQ SW cons",		KSTAT_KV_T_COUNTER64 },
8270 	{ "EQ HW prod",		KSTAT_KV_T_COUNTER64 },
8271 	{ "EQ HW cons",		KSTAT_KV_T_COUNTER64 },
8272 	{ "EQ HW status",	KSTAT_KV_T_ISTR },
8273 	{ "EQ HW state",	KSTAT_KV_T_ISTR },
8274 };
8275 
8276 static int	mcx_kstat_queue_read(struct kstat *);
8277 
8278 static void
8279 mcx_kstat_attach_queues(struct mcx_softc *sc)
8280 {
8281 	struct kstat *ks;
8282 	struct kstat_kv *kvs;
8283 	int q, i;
8284 
8285 	for (q = 0; q < sc->sc_nqueues; q++) {
8286 		ks = kstat_create(DEVNAME(sc), 0, "mcx-queues", q,
8287 		    KSTAT_T_KV, 0);
8288 		if (ks == NULL) {
8289 			/* unable to attach queue stats %u, q */
8290 			continue;
8291 		}
8292 
8293 		kvs = mallocarray(nitems(mcx_queue_kstat_tpl),
8294 		    sizeof(*kvs), M_DEVBUF, M_WAITOK);
8295 
8296 		for (i = 0; i < nitems(mcx_queue_kstat_tpl); i++) {
8297 			const struct mcx_queuestat *tpl =
8298 			    &mcx_queue_kstat_tpl[i];
8299 
8300 			kstat_kv_init(&kvs[i], tpl->name, tpl->type);
8301 		}
8302 
8303 		ks->ks_softc = &sc->sc_queues[q];
8304 		ks->ks_data = kvs;
8305 		ks->ks_datalen = nitems(mcx_queue_kstat_tpl) * sizeof(*kvs);
8306 		ks->ks_read = mcx_kstat_queue_read;
8307 
8308 		sc->sc_queues[q].q_kstat = ks;
8309 		kstat_install(ks);
8310 	}
8311 }
8312 
8313 static int
8314 mcx_kstat_queue_read(struct kstat *ks)
8315 {
8316 	struct mcx_queues *q = ks->ks_softc;
8317 	struct mcx_softc *sc = q->q_sc;
8318 	struct kstat_kv *kvs = ks->ks_data;
8319 	union {
8320 		struct mcx_rq_ctx rq;
8321 		struct mcx_sq_ctx sq;
8322 		struct mcx_cq_ctx cq;
8323 		struct mcx_eq_ctx eq;
8324 	} u;
8325 	const char *text;
8326 	int error = 0;
8327 
8328 	KERNEL_LOCK();
8329 
8330 	if (mcx_query_rq(sc, &q->q_rx, &u.rq) != 0) {
8331 		error = EIO;
8332 		goto out;
8333 	}
8334 
8335 	kstat_kv_u64(kvs++) = q->q_rx.rx_prod;
8336 	kstat_kv_u64(kvs++) = bemtoh32(&u.rq.rq_wq.wq_sw_counter);
8337 	kstat_kv_u64(kvs++) = bemtoh32(&u.rq.rq_wq.wq_hw_counter);
8338 	switch ((bemtoh32(&u.rq.rq_flags) & MCX_RQ_CTX_STATE_MASK) >>
8339 	    MCX_RQ_CTX_STATE_SHIFT) {
8340 	case MCX_RQ_CTX_STATE_RST:
8341 		text = "RST";
8342 		break;
8343 	case MCX_RQ_CTX_STATE_RDY:
8344 		text = "RDY";
8345 		break;
8346 	case MCX_RQ_CTX_STATE_ERR:
8347 		text = "ERR";
8348 		break;
8349 	default:
8350 		text = "unknown";
8351 		break;
8352 	}
8353 	strlcpy(kstat_kv_istr(kvs), text, sizeof(kstat_kv_istr(kvs)));
8354 	kvs++;
8355 
8356 	if (mcx_query_sq(sc, &q->q_tx, &u.sq) != 0) {
8357 		error = EIO;
8358 		goto out;
8359 	}
8360 
8361 	kstat_kv_u64(kvs++) = q->q_tx.tx_prod;
8362 	kstat_kv_u64(kvs++) = q->q_tx.tx_cons;
8363 	kstat_kv_u64(kvs++) = bemtoh32(&u.sq.sq_wq.wq_sw_counter);
8364 	kstat_kv_u64(kvs++) = bemtoh32(&u.sq.sq_wq.wq_hw_counter);
8365 	switch ((bemtoh32(&u.sq.sq_flags) & MCX_SQ_CTX_STATE_MASK) >>
8366 	    MCX_SQ_CTX_STATE_SHIFT) {
8367 	case MCX_SQ_CTX_STATE_RST:
8368 		text = "RST";
8369 		break;
8370 	case MCX_SQ_CTX_STATE_RDY:
8371 		text = "RDY";
8372 		break;
8373 	case MCX_SQ_CTX_STATE_ERR:
8374 		text = "ERR";
8375 		break;
8376 	default:
8377 		text = "unknown";
8378 		break;
8379 	}
8380 	strlcpy(kstat_kv_istr(kvs), text, sizeof(kstat_kv_istr(kvs)));
8381 	kvs++;
8382 
8383 	if (mcx_query_cq(sc, &q->q_cq, &u.cq) != 0) {
8384 		error = EIO;
8385 		goto out;
8386 	}
8387 
8388 	kstat_kv_u64(kvs++) = q->q_cq.cq_cons;
8389 	kstat_kv_u64(kvs++) = bemtoh32(&u.cq.cq_producer_counter);
8390 	kstat_kv_u64(kvs++) = bemtoh32(&u.cq.cq_consumer_counter);
8391 	kstat_kv_u64(kvs++) = bemtoh32(&u.cq.cq_last_notified);
8392 	kstat_kv_u64(kvs++) = bemtoh32(&u.cq.cq_last_solicit);
8393 
8394 	switch ((bemtoh32(&u.cq.cq_status) & MCX_CQ_CTX_STATUS_MASK) >>
8395 	    MCX_CQ_CTX_STATUS_SHIFT) {
8396 	case MCX_CQ_CTX_STATUS_OK:
8397 		text = "OK";
8398 		break;
8399 	case MCX_CQ_CTX_STATUS_OVERFLOW:
8400 		text = "overflow";
8401 		break;
8402 	case MCX_CQ_CTX_STATUS_WRITE_FAIL:
8403 		text = "write fail";
8404 		break;
8405 	default:
8406 		text = "unknown";
8407 		break;
8408 	}
8409 	strlcpy(kstat_kv_istr(kvs), text, sizeof(kstat_kv_istr(kvs)));
8410 	kvs++;
8411 
8412 	switch ((bemtoh32(&u.cq.cq_status) & MCX_CQ_CTX_STATE_MASK) >>
8413 	    MCX_CQ_CTX_STATE_SHIFT) {
8414 	case MCX_CQ_CTX_STATE_SOLICITED:
8415 		text = "solicited";
8416 		break;
8417 	case MCX_CQ_CTX_STATE_ARMED:
8418 		text = "armed";
8419 		break;
8420 	case MCX_CQ_CTX_STATE_FIRED:
8421 		text = "fired";
8422 		break;
8423 	default:
8424 		text = "unknown";
8425 		break;
8426 	}
8427 	strlcpy(kstat_kv_istr(kvs), text, sizeof(kstat_kv_istr(kvs)));
8428 	kvs++;
8429 
8430 	if (mcx_query_eq(sc, &q->q_eq, &u.eq) != 0) {
8431 		error = EIO;
8432 		goto out;
8433 	}
8434 
8435 	kstat_kv_u64(kvs++) = q->q_eq.eq_cons;
8436 	kstat_kv_u64(kvs++) = bemtoh32(&u.eq.eq_producer_counter);
8437 	kstat_kv_u64(kvs++) = bemtoh32(&u.eq.eq_consumer_counter);
8438 
8439 	switch ((bemtoh32(&u.eq.eq_status) & MCX_EQ_CTX_STATUS_MASK) >>
8440 	    MCX_EQ_CTX_STATUS_SHIFT) {
8441 	case MCX_EQ_CTX_STATUS_EQ_WRITE_FAILURE:
8442 		text = "write fail";
8443 		break;
8444 	case MCX_EQ_CTX_STATUS_OK:
8445 		text = "OK";
8446 		break;
8447 	default:
8448 		text = "unknown";
8449 		break;
8450 	}
8451 	strlcpy(kstat_kv_istr(kvs), text, sizeof(kstat_kv_istr(kvs)));
8452 	kvs++;
8453 
8454 	switch ((bemtoh32(&u.eq.eq_status) & MCX_EQ_CTX_STATE_MASK) >>
8455 	    MCX_EQ_CTX_STATE_SHIFT) {
8456 	case MCX_EQ_CTX_STATE_ARMED:
8457 		text = "armed";
8458 		break;
8459 	case MCX_EQ_CTX_STATE_FIRED:
8460 		text = "fired";
8461 		break;
8462 	default:
8463 		text = "unknown";
8464 		break;
8465 	}
8466 	strlcpy(kstat_kv_istr(kvs), text, sizeof(kstat_kv_istr(kvs)));
8467 	kvs++;
8468 
8469 	nanouptime(&ks->ks_updated);
8470 out:
8471 	KERNEL_UNLOCK();
8472 	return (error);
8473 }
8474 
8475 #endif /* NKSTAT > 0 */
8476