xref: /openbsd-src/sys/dev/pci/if_mcx.c (revision 505ee9ea3b177e2387d907a91ca7da069f3f14d8)
1 /*	$OpenBSD: if_mcx.c,v 1.70 2020/07/17 03:37:37 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			6 /* one page */
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_CREATE_CQ		0x400
207 #define MCX_CMD_DESTROY_CQ		0x401
208 #define MCX_CMD_QUERY_NIC_VPORT_CONTEXT	0x754
209 #define MCX_CMD_MODIFY_NIC_VPORT_CONTEXT \
210 					0x755
211 #define MCX_CMD_QUERY_VPORT_COUNTERS	0x770
212 #define MCX_CMD_ALLOC_PD		0x800
213 #define MCX_CMD_ALLOC_UAR		0x802
214 #define MCX_CMD_ACCESS_REG		0x805
215 #define MCX_CMD_ALLOC_TRANSPORT_DOMAIN	0x816
216 #define MCX_CMD_CREATE_TIR		0x900
217 #define MCX_CMD_DESTROY_TIR		0x902
218 #define MCX_CMD_CREATE_SQ		0x904
219 #define MCX_CMD_MODIFY_SQ		0x905
220 #define MCX_CMD_DESTROY_SQ		0x906
221 #define MCX_CMD_QUERY_SQ		0x907
222 #define MCX_CMD_CREATE_RQ		0x908
223 #define MCX_CMD_MODIFY_RQ		0x909
224 #define MCX_CMD_DESTROY_RQ		0x90a
225 #define MCX_CMD_QUERY_RQ		0x90b
226 #define MCX_CMD_CREATE_TIS		0x912
227 #define MCX_CMD_DESTROY_TIS		0x914
228 #define MCX_CMD_CREATE_RQT		0x916
229 #define MCX_CMD_DESTROY_RQT		0x918
230 #define MCX_CMD_SET_FLOW_TABLE_ROOT	0x92f
231 #define MCX_CMD_CREATE_FLOW_TABLE	0x930
232 #define MCX_CMD_DESTROY_FLOW_TABLE	0x931
233 #define MCX_CMD_QUERY_FLOW_TABLE	0x932
234 #define MCX_CMD_CREATE_FLOW_GROUP	0x933
235 #define MCX_CMD_DESTROY_FLOW_GROUP	0x934
236 #define MCX_CMD_QUERY_FLOW_GROUP	0x935
237 #define MCX_CMD_SET_FLOW_TABLE_ENTRY	0x936
238 #define MCX_CMD_QUERY_FLOW_TABLE_ENTRY	0x937
239 #define MCX_CMD_DELETE_FLOW_TABLE_ENTRY	0x938
240 #define MCX_CMD_ALLOC_FLOW_COUNTER	0x939
241 #define MCX_CMD_QUERY_FLOW_COUNTER	0x93b
242 
243 #define MCX_QUEUE_STATE_RST		0
244 #define MCX_QUEUE_STATE_RDY		1
245 #define MCX_QUEUE_STATE_ERR		3
246 
247 #define MCX_FLOW_TABLE_TYPE_RX		0
248 #define MCX_FLOW_TABLE_TYPE_TX		1
249 
250 #define MCX_CMDQ_INLINE_DATASIZE 16
251 
252 struct mcx_cmdq_entry {
253 	uint8_t			cq_type;
254 #define MCX_CMDQ_TYPE_PCIE		0x7
255 	uint8_t			cq_reserved0[3];
256 
257 	uint32_t		cq_input_length;
258 	uint64_t		cq_input_ptr;
259 	uint8_t			cq_input_data[MCX_CMDQ_INLINE_DATASIZE];
260 
261 	uint8_t			cq_output_data[MCX_CMDQ_INLINE_DATASIZE];
262 	uint64_t		cq_output_ptr;
263 	uint32_t		cq_output_length;
264 
265 	uint8_t			cq_token;
266 	uint8_t			cq_signature;
267 	uint8_t			cq_reserved1[1];
268 	uint8_t			cq_status;
269 #define MCX_CQ_STATUS_SHIFT		1
270 #define MCX_CQ_STATUS_MASK		(0x7f << MCX_CQ_STATUS_SHIFT)
271 #define MCX_CQ_STATUS_OK		(0x00 << MCX_CQ_STATUS_SHIFT)
272 #define MCX_CQ_STATUS_INT_ERR		(0x01 << MCX_CQ_STATUS_SHIFT)
273 #define MCX_CQ_STATUS_BAD_OPCODE	(0x02 << MCX_CQ_STATUS_SHIFT)
274 #define MCX_CQ_STATUS_BAD_PARAM		(0x03 << MCX_CQ_STATUS_SHIFT)
275 #define MCX_CQ_STATUS_BAD_SYS_STATE	(0x04 << MCX_CQ_STATUS_SHIFT)
276 #define MCX_CQ_STATUS_BAD_RESOURCE	(0x05 << MCX_CQ_STATUS_SHIFT)
277 #define MCX_CQ_STATUS_RESOURCE_BUSY	(0x06 << MCX_CQ_STATUS_SHIFT)
278 #define MCX_CQ_STATUS_EXCEED_LIM	(0x08 << MCX_CQ_STATUS_SHIFT)
279 #define MCX_CQ_STATUS_BAD_RES_STATE	(0x09 << MCX_CQ_STATUS_SHIFT)
280 #define MCX_CQ_STATUS_BAD_INDEX		(0x0a << MCX_CQ_STATUS_SHIFT)
281 #define MCX_CQ_STATUS_NO_RESOURCES	(0x0f << MCX_CQ_STATUS_SHIFT)
282 #define MCX_CQ_STATUS_BAD_INPUT_LEN	(0x50 << MCX_CQ_STATUS_SHIFT)
283 #define MCX_CQ_STATUS_BAD_OUTPUT_LEN	(0x51 << MCX_CQ_STATUS_SHIFT)
284 #define MCX_CQ_STATUS_BAD_RESOURCE_STATE \
285 					(0x10 << MCX_CQ_STATUS_SHIFT)
286 #define MCX_CQ_STATUS_BAD_SIZE		(0x40 << MCX_CQ_STATUS_SHIFT)
287 #define MCX_CQ_STATUS_OWN_MASK		0x1
288 #define MCX_CQ_STATUS_OWN_SW		0x0
289 #define MCX_CQ_STATUS_OWN_HW		0x1
290 } __packed __aligned(8);
291 
292 #define MCX_CMDQ_MAILBOX_DATASIZE	512
293 
294 struct mcx_cmdq_mailbox {
295 	uint8_t			mb_data[MCX_CMDQ_MAILBOX_DATASIZE];
296 	uint8_t			mb_reserved0[48];
297 	uint64_t		mb_next_ptr;
298 	uint32_t		mb_block_number;
299 	uint8_t			mb_reserved1[1];
300 	uint8_t			mb_token;
301 	uint8_t			mb_ctrl_signature;
302 	uint8_t			mb_signature;
303 } __packed __aligned(8);
304 
305 #define MCX_CMDQ_MAILBOX_ALIGN	(1 << 10)
306 #define MCX_CMDQ_MAILBOX_SIZE	roundup(sizeof(struct mcx_cmdq_mailbox), \
307 				    MCX_CMDQ_MAILBOX_ALIGN)
308 /*
309  * command mailbox structres
310  */
311 
312 struct mcx_cmd_enable_hca_in {
313 	uint16_t		cmd_opcode;
314 	uint8_t			cmd_reserved0[4];
315 	uint16_t		cmd_op_mod;
316 	uint8_t			cmd_reserved1[2];
317 	uint16_t		cmd_function_id;
318 	uint8_t			cmd_reserved2[4];
319 } __packed __aligned(4);
320 
321 struct mcx_cmd_enable_hca_out {
322 	uint8_t			cmd_status;
323 	uint8_t			cmd_reserved0[3];
324 	uint32_t		cmd_syndrome;
325 	uint8_t			cmd_reserved1[4];
326 } __packed __aligned(4);
327 
328 struct mcx_cmd_init_hca_in {
329 	uint16_t		cmd_opcode;
330 	uint8_t			cmd_reserved0[4];
331 	uint16_t		cmd_op_mod;
332 	uint8_t			cmd_reserved1[8];
333 } __packed __aligned(4);
334 
335 struct mcx_cmd_init_hca_out {
336 	uint8_t			cmd_status;
337 	uint8_t			cmd_reserved0[3];
338 	uint32_t		cmd_syndrome;
339 	uint8_t			cmd_reserved1[8];
340 } __packed __aligned(4);
341 
342 struct mcx_cmd_teardown_hca_in {
343 	uint16_t		cmd_opcode;
344 	uint8_t			cmd_reserved0[4];
345 	uint16_t		cmd_op_mod;
346 	uint8_t			cmd_reserved1[2];
347 #define MCX_CMD_TEARDOWN_HCA_GRACEFUL	0x0
348 #define MCX_CMD_TEARDOWN_HCA_PANIC	0x1
349 	uint16_t		cmd_profile;
350 	uint8_t			cmd_reserved2[4];
351 } __packed __aligned(4);
352 
353 struct mcx_cmd_teardown_hca_out {
354 	uint8_t			cmd_status;
355 	uint8_t			cmd_reserved0[3];
356 	uint32_t		cmd_syndrome;
357 	uint8_t			cmd_reserved1[8];
358 } __packed __aligned(4);
359 
360 struct mcx_cmd_access_reg_in {
361 	uint16_t		cmd_opcode;
362 	uint8_t			cmd_reserved0[4];
363 	uint16_t		cmd_op_mod;
364 	uint8_t			cmd_reserved1[2];
365 	uint16_t		cmd_register_id;
366 	uint32_t		cmd_argument;
367 } __packed __aligned(4);
368 
369 struct mcx_cmd_access_reg_out {
370 	uint8_t			cmd_status;
371 	uint8_t			cmd_reserved0[3];
372 	uint32_t		cmd_syndrome;
373 	uint8_t			cmd_reserved1[8];
374 } __packed __aligned(4);
375 
376 struct mcx_reg_pmtu {
377 	uint8_t			rp_reserved1;
378 	uint8_t			rp_local_port;
379 	uint8_t			rp_reserved2[2];
380 	uint16_t		rp_max_mtu;
381 	uint8_t			rp_reserved3[2];
382 	uint16_t		rp_admin_mtu;
383 	uint8_t			rp_reserved4[2];
384 	uint16_t		rp_oper_mtu;
385 	uint8_t			rp_reserved5[2];
386 } __packed __aligned(4);
387 
388 struct mcx_reg_ptys {
389 	uint8_t			rp_reserved1;
390 	uint8_t			rp_local_port;
391 	uint8_t			rp_reserved2;
392 	uint8_t			rp_proto_mask;
393 #define MCX_REG_PTYS_PROTO_MASK_ETH		(1 << 2)
394 	uint8_t			rp_reserved3[8];
395 	uint32_t		rp_eth_proto_cap;
396 	uint8_t			rp_reserved4[8];
397 	uint32_t		rp_eth_proto_admin;
398 	uint8_t			rp_reserved5[8];
399 	uint32_t		rp_eth_proto_oper;
400 	uint8_t			rp_reserved6[24];
401 } __packed __aligned(4);
402 
403 struct mcx_reg_paos {
404 	uint8_t			rp_reserved1;
405 	uint8_t			rp_local_port;
406 	uint8_t			rp_admin_status;
407 #define MCX_REG_PAOS_ADMIN_STATUS_UP		1
408 #define MCX_REG_PAOS_ADMIN_STATUS_DOWN		2
409 #define MCX_REG_PAOS_ADMIN_STATUS_UP_ONCE	3
410 #define MCX_REG_PAOS_ADMIN_STATUS_DISABLED	4
411 	uint8_t			rp_oper_status;
412 #define MCX_REG_PAOS_OPER_STATUS_UP		1
413 #define MCX_REG_PAOS_OPER_STATUS_DOWN		2
414 #define MCX_REG_PAOS_OPER_STATUS_FAILED		4
415 	uint8_t			rp_admin_state_update;
416 #define MCX_REG_PAOS_ADMIN_STATE_UPDATE_EN	(1 << 7)
417 	uint8_t			rp_reserved2[11];
418 } __packed __aligned(4);
419 
420 struct mcx_reg_pfcc {
421 	uint8_t			rp_reserved1;
422 	uint8_t			rp_local_port;
423 	uint8_t			rp_reserved2[3];
424 	uint8_t			rp_prio_mask_tx;
425 	uint8_t			rp_reserved3;
426 	uint8_t			rp_prio_mask_rx;
427 	uint8_t			rp_pptx_aptx;
428 	uint8_t			rp_pfctx;
429 	uint8_t			rp_fctx_dis;
430 	uint8_t			rp_reserved4;
431 	uint8_t			rp_pprx_aprx;
432 	uint8_t			rp_pfcrx;
433 	uint8_t			rp_reserved5[2];
434 	uint16_t		rp_dev_stall_min;
435 	uint16_t		rp_dev_stall_crit;
436 	uint8_t			rp_reserved6[12];
437 } __packed __aligned(4);
438 
439 #define MCX_PMLP_MODULE_NUM_MASK	0xff
440 struct mcx_reg_pmlp {
441 	uint8_t			rp_rxtx;
442 	uint8_t			rp_local_port;
443 	uint8_t			rp_reserved0;
444 	uint8_t			rp_width;
445 	uint32_t		rp_lane0_mapping;
446 	uint32_t		rp_lane1_mapping;
447 	uint32_t		rp_lane2_mapping;
448 	uint32_t		rp_lane3_mapping;
449 	uint8_t			rp_reserved1[44];
450 } __packed __aligned(4);
451 
452 struct mcx_reg_ppcnt {
453 	uint8_t			ppcnt_swid;
454 	uint8_t			ppcnt_local_port;
455 	uint8_t			ppcnt_pnat;
456 	uint8_t			ppcnt_grp;
457 #define MCX_REG_PPCNT_GRP_IEEE8023		0x00
458 #define MCX_REG_PPCNT_GRP_RFC2863		0x01
459 #define MCX_REG_PPCNT_GRP_RFC2819		0x02
460 #define MCX_REG_PPCNT_GRP_RFC3635		0x03
461 #define MCX_REG_PPCNT_GRP_PER_PRIO		0x10
462 #define MCX_REG_PPCNT_GRP_PER_TC		0x11
463 #define MCX_REG_PPCNT_GRP_PER_RX_BUFFER		0x11
464 
465 	uint8_t			ppcnt_clr;
466 	uint8_t			ppcnt_reserved1[2];
467 	uint8_t			ppcnt_prio_tc;
468 #define MCX_REG_PPCNT_CLR			(1 << 7)
469 
470 	uint8_t			ppcnt_counter_set[248];
471 } __packed __aligned(8);
472 CTASSERT(sizeof(struct mcx_reg_ppcnt) == 256);
473 CTASSERT((offsetof(struct mcx_reg_ppcnt, ppcnt_counter_set) %
474     sizeof(uint64_t)) == 0);
475 
476 enum mcx_ppcnt_ieee8023 {
477 	frames_transmitted_ok,
478 	frames_received_ok,
479 	frame_check_sequence_errors,
480 	alignment_errors,
481 	octets_transmitted_ok,
482 	octets_received_ok,
483 	multicast_frames_xmitted_ok,
484 	broadcast_frames_xmitted_ok,
485 	multicast_frames_received_ok,
486 	broadcast_frames_received_ok,
487 	in_range_length_errors,
488 	out_of_range_length_field,
489 	frame_too_long_errors,
490 	symbol_error_during_carrier,
491 	mac_control_frames_transmitted,
492 	mac_control_frames_received,
493 	unsupported_opcodes_received,
494 	pause_mac_ctrl_frames_received,
495 	pause_mac_ctrl_frames_transmitted,
496 
497 	mcx_ppcnt_ieee8023_count
498 };
499 CTASSERT(mcx_ppcnt_ieee8023_count * sizeof(uint64_t) == 0x98);
500 
501 enum mcx_ppcnt_rfc2863 {
502 	in_octets,
503 	in_ucast_pkts,
504 	in_discards,
505 	in_errors,
506 	in_unknown_protos,
507 	out_octets,
508 	out_ucast_pkts,
509 	out_discards,
510 	out_errors,
511 	in_multicast_pkts,
512 	in_broadcast_pkts,
513 	out_multicast_pkts,
514 	out_broadcast_pkts,
515 
516 	mcx_ppcnt_rfc2863_count
517 };
518 CTASSERT(mcx_ppcnt_rfc2863_count * sizeof(uint64_t) == 0x68);
519 
520 enum mcx_ppcnt_rfc2819 {
521 	drop_events,
522 	octets,
523 	pkts,
524 	broadcast_pkts,
525 	multicast_pkts,
526 	crc_align_errors,
527 	undersize_pkts,
528 	oversize_pkts,
529 	fragments,
530 	jabbers,
531 	collisions,
532 	pkts64octets,
533 	pkts65to127octets,
534 	pkts128to255octets,
535 	pkts256to511octets,
536 	pkts512to1023octets,
537 	pkts1024to1518octets,
538 	pkts1519to2047octets,
539 	pkts2048to4095octets,
540 	pkts4096to8191octets,
541 	pkts8192to10239octets,
542 
543 	mcx_ppcnt_rfc2819_count
544 };
545 CTASSERT((mcx_ppcnt_rfc2819_count * sizeof(uint64_t)) == 0xa8);
546 
547 enum mcx_ppcnt_rfc3635 {
548 	dot3stats_alignment_errors,
549 	dot3stats_fcs_errors,
550 	dot3stats_single_collision_frames,
551 	dot3stats_multiple_collision_frames,
552 	dot3stats_sqe_test_errors,
553 	dot3stats_deferred_transmissions,
554 	dot3stats_late_collisions,
555 	dot3stats_excessive_collisions,
556 	dot3stats_internal_mac_transmit_errors,
557 	dot3stats_carrier_sense_errors,
558 	dot3stats_frame_too_longs,
559 	dot3stats_internal_mac_receive_errors,
560 	dot3stats_symbol_errors,
561 	dot3control_in_unknown_opcodes,
562 	dot3in_pause_frames,
563 	dot3out_pause_frames,
564 
565 	mcx_ppcnt_rfc3635_count
566 };
567 CTASSERT((mcx_ppcnt_rfc3635_count * sizeof(uint64_t)) == 0x80);
568 
569 struct mcx_reg_mtcap {
570 	uint8_t			_reserved1[3];
571 	uint8_t			mtcap_sensor_count;
572 	uint8_t			_reserved2[4];
573 
574 	uint64_t		mtcap_sensor_map;
575 };
576 
577 struct mcx_reg_mtmp {
578 	uint8_t			_reserved1[2];
579 	uint16_t		mtmp_sensor_index;
580 
581 	uint8_t			_reserved2[2];
582 	uint16_t		mtmp_temperature;
583 
584 	uint16_t		mtmp_mte_mtr;
585 #define MCX_REG_MTMP_MTE		(1 << 15)
586 #define MCX_REG_MTMP_MTR		(1 << 14)
587 	uint16_t		mtmp_max_temperature;
588 
589 	uint16_t		mtmp_tee;
590 #define MCX_REG_MTMP_TEE_NOPE		(0 << 14)
591 #define MCX_REG_MTMP_TEE_GENERATE	(1 << 14)
592 #define MCX_REG_MTMP_TEE_GENERATE_ONE	(2 << 14)
593 	uint16_t		mtmp_temperature_threshold_hi;
594 
595 	uint8_t			_reserved3[2];
596 	uint16_t		mtmp_temperature_threshold_lo;
597 
598 	uint8_t			_reserved4[4];
599 
600 	uint8_t			mtmp_sensor_name[8];
601 };
602 CTASSERT(sizeof(struct mcx_reg_mtmp) == 0x20);
603 CTASSERT(offsetof(struct mcx_reg_mtmp, mtmp_sensor_name) == 0x18);
604 
605 #define MCX_MCIA_EEPROM_BYTES	32
606 struct mcx_reg_mcia {
607 	uint8_t			rm_l;
608 	uint8_t			rm_module;
609 	uint8_t			rm_reserved0;
610 	uint8_t			rm_status;
611 	uint8_t			rm_i2c_addr;
612 	uint8_t			rm_page_num;
613 	uint16_t		rm_dev_addr;
614 	uint16_t		rm_reserved1;
615 	uint16_t		rm_size;
616 	uint32_t		rm_reserved2;
617 	uint8_t			rm_data[48];
618 } __packed __aligned(4);
619 
620 struct mcx_cmd_query_issi_in {
621 	uint16_t		cmd_opcode;
622 	uint8_t			cmd_reserved0[4];
623 	uint16_t		cmd_op_mod;
624 	uint8_t			cmd_reserved1[8];
625 } __packed __aligned(4);
626 
627 struct mcx_cmd_query_issi_il_out {
628 	uint8_t			cmd_status;
629 	uint8_t			cmd_reserved0[3];
630 	uint32_t		cmd_syndrome;
631 	uint8_t			cmd_reserved1[2];
632 	uint16_t		cmd_current_issi;
633 	uint8_t			cmd_reserved2[4];
634 } __packed __aligned(4);
635 
636 CTASSERT(sizeof(struct mcx_cmd_query_issi_il_out) == MCX_CMDQ_INLINE_DATASIZE);
637 
638 struct mcx_cmd_query_issi_mb_out {
639 	uint8_t			cmd_reserved2[16];
640 	uint8_t			cmd_supported_issi[80]; /* very big endian */
641 } __packed __aligned(4);
642 
643 CTASSERT(sizeof(struct mcx_cmd_query_issi_mb_out) <= MCX_CMDQ_MAILBOX_DATASIZE);
644 
645 struct mcx_cmd_set_issi_in {
646 	uint16_t		cmd_opcode;
647 	uint8_t			cmd_reserved0[4];
648 	uint16_t		cmd_op_mod;
649 	uint8_t			cmd_reserved1[2];
650 	uint16_t		cmd_current_issi;
651 	uint8_t			cmd_reserved2[4];
652 } __packed __aligned(4);
653 
654 CTASSERT(sizeof(struct mcx_cmd_set_issi_in) <= MCX_CMDQ_INLINE_DATASIZE);
655 
656 struct mcx_cmd_set_issi_out {
657 	uint8_t			cmd_status;
658 	uint8_t			cmd_reserved0[3];
659 	uint32_t		cmd_syndrome;
660 	uint8_t			cmd_reserved1[8];
661 } __packed __aligned(4);
662 
663 CTASSERT(sizeof(struct mcx_cmd_set_issi_out) <= MCX_CMDQ_INLINE_DATASIZE);
664 
665 struct mcx_cmd_query_pages_in {
666 	uint16_t		cmd_opcode;
667 	uint8_t			cmd_reserved0[4];
668 	uint16_t		cmd_op_mod;
669 #define MCX_CMD_QUERY_PAGES_BOOT	0x01
670 #define MCX_CMD_QUERY_PAGES_INIT	0x02
671 #define MCX_CMD_QUERY_PAGES_REGULAR	0x03
672 	uint8_t			cmd_reserved1[8];
673 } __packed __aligned(4);
674 
675 struct mcx_cmd_query_pages_out {
676 	uint8_t			cmd_status;
677 	uint8_t			cmd_reserved0[3];
678 	uint32_t		cmd_syndrome;
679 	uint8_t			cmd_reserved1[2];
680 	uint16_t		cmd_func_id;
681 	int32_t			cmd_num_pages;
682 } __packed __aligned(4);
683 
684 struct mcx_cmd_manage_pages_in {
685 	uint16_t		cmd_opcode;
686 	uint8_t			cmd_reserved0[4];
687 	uint16_t		cmd_op_mod;
688 #define MCX_CMD_MANAGE_PAGES_ALLOC_FAIL \
689 					0x00
690 #define MCX_CMD_MANAGE_PAGES_ALLOC_SUCCESS \
691 					0x01
692 #define MCX_CMD_MANAGE_PAGES_HCA_RETURN_PAGES \
693 					0x02
694 	uint8_t			cmd_reserved1[2];
695 	uint16_t		cmd_func_id;
696 	uint32_t		cmd_input_num_entries;
697 } __packed __aligned(4);
698 
699 CTASSERT(sizeof(struct mcx_cmd_manage_pages_in) == MCX_CMDQ_INLINE_DATASIZE);
700 
701 struct mcx_cmd_manage_pages_out {
702 	uint8_t			cmd_status;
703 	uint8_t			cmd_reserved0[3];
704 	uint32_t		cmd_syndrome;
705 	uint32_t		cmd_output_num_entries;
706 	uint8_t			cmd_reserved1[4];
707 } __packed __aligned(4);
708 
709 CTASSERT(sizeof(struct mcx_cmd_manage_pages_out) == MCX_CMDQ_INLINE_DATASIZE);
710 
711 struct mcx_cmd_query_hca_cap_in {
712 	uint16_t		cmd_opcode;
713 	uint8_t			cmd_reserved0[4];
714 	uint16_t		cmd_op_mod;
715 #define MCX_CMD_QUERY_HCA_CAP_MAX	(0x0 << 0)
716 #define MCX_CMD_QUERY_HCA_CAP_CURRENT	(0x1 << 0)
717 #define MCX_CMD_QUERY_HCA_CAP_DEVICE	(0x0 << 1)
718 #define MCX_CMD_QUERY_HCA_CAP_OFFLOAD	(0x1 << 1)
719 #define MCX_CMD_QUERY_HCA_CAP_FLOW	(0x7 << 1)
720 	uint8_t			cmd_reserved1[8];
721 } __packed __aligned(4);
722 
723 struct mcx_cmd_query_hca_cap_out {
724 	uint8_t			cmd_status;
725 	uint8_t			cmd_reserved0[3];
726 	uint32_t		cmd_syndrome;
727 	uint8_t			cmd_reserved1[8];
728 } __packed __aligned(4);
729 
730 #define MCX_HCA_CAP_LEN			0x1000
731 #define MCX_HCA_CAP_NMAILBOXES		\
732 	(MCX_HCA_CAP_LEN / MCX_CMDQ_MAILBOX_DATASIZE)
733 
734 #if __GNUC_PREREQ__(4, 3)
735 #define __counter__		__COUNTER__
736 #else
737 #define __counter__		__LINE__
738 #endif
739 
740 #define __token(_tok, _num)	_tok##_num
741 #define _token(_tok, _num)	__token(_tok, _num)
742 #define __reserved__		_token(__reserved, __counter__)
743 
744 struct mcx_cap_device {
745 	uint8_t			reserved0[16];
746 
747 	uint8_t			log_max_srq_sz;
748 	uint8_t			log_max_qp_sz;
749 	uint8_t			__reserved__[1];
750 	uint8_t			log_max_qp; /* 5 bits */
751 #define MCX_CAP_DEVICE_LOG_MAX_QP	0x1f
752 
753 	uint8_t			__reserved__[1];
754 	uint8_t			log_max_srq; /* 5 bits */
755 #define MCX_CAP_DEVICE_LOG_MAX_SRQ	0x1f
756 	uint8_t			__reserved__[2];
757 
758 	uint8_t			__reserved__[1];
759 	uint8_t			log_max_cq_sz;
760 	uint8_t			__reserved__[1];
761 	uint8_t			log_max_cq; /* 5 bits */
762 #define MCX_CAP_DEVICE_LOG_MAX_CQ	0x1f
763 
764 	uint8_t			log_max_eq_sz;
765 	uint8_t			log_max_mkey; /* 6 bits */
766 #define MCX_CAP_DEVICE_LOG_MAX_MKEY	0x3f
767 	uint8_t			__reserved__[1];
768 	uint8_t			log_max_eq; /* 4 bits */
769 #define MCX_CAP_DEVICE_LOG_MAX_EQ	0x0f
770 
771 	uint8_t			max_indirection;
772 	uint8_t			log_max_mrw_sz; /* 7 bits */
773 #define MCX_CAP_DEVICE_LOG_MAX_MRW_SZ	0x7f
774 	uint8_t			teardown_log_max_msf_list_size;
775 #define MCX_CAP_DEVICE_FORCE_TEARDOWN	0x80
776 #define MCX_CAP_DEVICE_LOG_MAX_MSF_LIST_SIZE \
777 					0x3f
778 	uint8_t			log_max_klm_list_size; /* 6 bits */
779 #define MCX_CAP_DEVICE_LOG_MAX_KLM_LIST_SIZE \
780 					0x3f
781 
782 	uint8_t			__reserved__[1];
783 	uint8_t			log_max_ra_req_dc; /* 6 bits */
784 #define MCX_CAP_DEVICE_LOG_MAX_REQ_DC	0x3f
785 	uint8_t			__reserved__[1];
786 	uint8_t			log_max_ra_res_dc; /* 6 bits */
787 #define MCX_CAP_DEVICE_LOG_MAX_RA_RES_DC \
788 					0x3f
789 
790 	uint8_t			__reserved__[1];
791 	uint8_t			log_max_ra_req_qp; /* 6 bits */
792 #define MCX_CAP_DEVICE_LOG_MAX_RA_REQ_QP \
793 					0x3f
794 	uint8_t			__reserved__[1];
795 	uint8_t			log_max_ra_res_qp; /* 6 bits */
796 #define MCX_CAP_DEVICE_LOG_MAX_RA_RES_QP \
797 					0x3f
798 
799 	uint8_t			flags1;
800 #define MCX_CAP_DEVICE_END_PAD		0x80
801 #define MCX_CAP_DEVICE_CC_QUERY_ALLOWED	0x40
802 #define MCX_CAP_DEVICE_CC_MODIFY_ALLOWED \
803 					0x20
804 #define MCX_CAP_DEVICE_START_PAD	0x10
805 #define MCX_CAP_DEVICE_128BYTE_CACHELINE \
806 					0x08
807 	uint8_t			__reserved__[1];
808 	uint16_t		gid_table_size;
809 
810 	uint16_t		flags2;
811 #define MCX_CAP_DEVICE_OUT_OF_SEQ_CNT	0x8000
812 #define MCX_CAP_DEVICE_VPORT_COUNTERS	0x4000
813 #define MCX_CAP_DEVICE_RETRANSMISSION_Q_COUNTERS \
814 					0x2000
815 #define MCX_CAP_DEVICE_DEBUG		0x1000
816 #define MCX_CAP_DEVICE_MODIFY_RQ_COUNTERS_SET_ID \
817 					0x8000
818 #define MCX_CAP_DEVICE_RQ_DELAY_DROP	0x4000
819 #define MCX_CAP_DEVICe_MAX_QP_CNT_MASK	0x03ff
820 	uint16_t		pkey_table_size;
821 
822 	uint8_t			flags3;
823 #define MCX_CAP_DEVICE_VPORT_GROUP_MANAGER \
824 					0x80
825 #define MCX_CAP_DEVICE_VHCA_GROUP_MANAGER \
826 					0x40
827 #define MCX_CAP_DEVICE_IB_VIRTUAL	0x20
828 #define MCX_CAP_DEVICE_ETH_VIRTUAL	0x10
829 #define MCX_CAP_DEVICE_ETS		0x04
830 #define MCX_CAP_DEVICE_NIC_FLOW_TABLE	0x02
831 #define MCX_CAP_DEVICE_ESWITCH_FLOW_TABLE \
832 					0x01
833 	uint8_t			local_ca_ack_delay; /* 5 bits */
834 #define MCX_CAP_DEVICE_LOCAL_CA_ACK_DELAY \
835 					0x1f
836 	uint8_t			port_type;
837 #define MCX_CAP_DEVICE_PORT_MODULE_EVENT \
838 					0x80
839 #define MCX_CAP_DEVICE_PORT_TYPE	0x03
840 	uint8_t			num_ports;
841 
842 	uint8_t			snapshot_log_max_msg;
843 #define MCX_CAP_DEVICE_SNAPSHOT		0x80
844 #define MCX_CAP_DEVICE_LOG_MAX_MSG	0x1f
845 	uint8_t			max_tc; /* 4 bits */
846 #define MCX_CAP_DEVICE_MAX_TC		0x0f
847 	uint8_t			flags4;
848 #define MCX_CAP_DEVICE_TEMP_WARN_EVENT	0x80
849 #define MCX_CAP_DEVICE_DCBX		0x40
850 #define MCX_CAP_DEVICE_ROL_S		0x02
851 #define MCX_CAP_DEVICE_ROL_G		0x01
852 	uint8_t			wol;
853 #define MCX_CAP_DEVICE_WOL_S		0x40
854 #define MCX_CAP_DEVICE_WOL_G		0x20
855 #define MCX_CAP_DEVICE_WOL_A		0x10
856 #define MCX_CAP_DEVICE_WOL_B		0x08
857 #define MCX_CAP_DEVICE_WOL_M		0x04
858 #define MCX_CAP_DEVICE_WOL_U		0x02
859 #define MCX_CAP_DEVICE_WOL_P		0x01
860 
861 	uint16_t		stat_rate_support;
862 	uint8_t			__reserved__[1];
863 	uint8_t			cqe_version; /* 4 bits */
864 #define MCX_CAP_DEVICE_CQE_VERSION	0x0f
865 
866 	uint32_t		flags5;
867 #define MCX_CAP_DEVICE_COMPACT_ADDRESS_VECTOR \
868 					0x80000000
869 #define MCX_CAP_DEVICE_STRIDING_RQ	0x40000000
870 #define MCX_CAP_DEVICE_IPOIP_ENHANCED_OFFLOADS \
871 					0x10000000
872 #define MCX_CAP_DEVICE_IPOIP_IPOIP_OFFLOADS \
873 					0x08000000
874 #define MCX_CAP_DEVICE_DC_CONNECT_CP	0x00040000
875 #define MCX_CAP_DEVICE_DC_CNAK_DRACE	0x00020000
876 #define MCX_CAP_DEVICE_DRAIN_SIGERR	0x00010000
877 #define MCX_CAP_DEVICE_DRAIN_SIGERR	0x00010000
878 #define MCX_CAP_DEVICE_CMDIF_CHECKSUM	0x0000c000
879 #define MCX_CAP_DEVICE_SIGERR_QCE	0x00002000
880 #define MCX_CAP_DEVICE_WQ_SIGNATURE	0x00000800
881 #define MCX_CAP_DEVICE_SCTR_DATA_CQE	0x00000400
882 #define MCX_CAP_DEVICE_SHO		0x00000100
883 #define MCX_CAP_DEVICE_TPH		0x00000080
884 #define MCX_CAP_DEVICE_RF		0x00000040
885 #define MCX_CAP_DEVICE_DCT		0x00000020
886 #define MCX_CAP_DEVICE_QOS		0x00000010
887 #define MCX_CAP_DEVICe_ETH_NET_OFFLOADS	0x00000008
888 #define MCX_CAP_DEVICE_ROCE		0x00000004
889 #define MCX_CAP_DEVICE_ATOMIC		0x00000002
890 
891 	uint32_t		flags6;
892 #define MCX_CAP_DEVICE_CQ_OI		0x80000000
893 #define MCX_CAP_DEVICE_CQ_RESIZE	0x40000000
894 #define MCX_CAP_DEVICE_CQ_MODERATION	0x20000000
895 #define MCX_CAP_DEVICE_CQ_PERIOD_MODE_MODIFY \
896 					0x10000000
897 #define MCX_CAP_DEVICE_CQ_INVALIDATE	0x08000000
898 #define MCX_CAP_DEVICE_RESERVED_AT_255	0x04000000
899 #define MCX_CAP_DEVICE_CQ_EQ_REMAP	0x02000000
900 #define MCX_CAP_DEVICE_PG		0x01000000
901 #define MCX_CAP_DEVICE_BLOCK_LB_MC	0x00800000
902 #define MCX_CAP_DEVICE_EXPONENTIAL_BACKOFF \
903 					0x00400000
904 #define MCX_CAP_DEVICE_SCQE_BREAK_MODERATION \
905 					0x00200000
906 #define MCX_CAP_DEVICE_CQ_PERIOD_START_FROM_CQE \
907 					0x00100000
908 #define MCX_CAP_DEVICE_CD		0x00080000
909 #define MCX_CAP_DEVICE_ATM		0x00040000
910 #define MCX_CAP_DEVICE_APM		0x00020000
911 #define MCX_CAP_DEVICE_IMAICL		0x00010000
912 #define MCX_CAP_DEVICE_QKV		0x00000200
913 #define MCX_CAP_DEVICE_PKV		0x00000100
914 #define MCX_CAP_DEVICE_SET_DETH_SQPN	0x00000080
915 #define MCX_CAP_DEVICE_XRC		0x00000008
916 #define MCX_CAP_DEVICE_UD		0x00000004
917 #define MCX_CAP_DEVICE_UC		0x00000002
918 #define MCX_CAP_DEVICE_RC		0x00000001
919 
920 	uint8_t			uar_flags;
921 #define MCX_CAP_DEVICE_UAR_4K		0x80
922 	uint8_t			uar_sz;	/* 6 bits */
923 #define MCX_CAP_DEVICE_UAR_SZ		0x3f
924 	uint8_t			__reserved__[1];
925 	uint8_t			log_pg_sz;
926 
927 	uint8_t			flags7;
928 #define MCX_CAP_DEVICE_BF		0x80
929 #define MCX_CAP_DEVICE_DRIVER_VERSION	0x40
930 #define MCX_CAP_DEVICE_PAD_TX_ETH_PACKET \
931 					0x20
932 	uint8_t			log_bf_reg_size; /* 5 bits */
933 #define MCX_CAP_DEVICE_LOG_BF_REG_SIZE	0x1f
934 	uint8_t			__reserved__[2];
935 
936 	uint16_t		num_of_diagnostic_counters;
937 	uint16_t		max_wqe_sz_sq;
938 
939 	uint8_t			__reserved__[2];
940 	uint16_t		max_wqe_sz_rq;
941 
942 	uint8_t			__reserved__[2];
943 	uint16_t		max_wqe_sz_sq_dc;
944 
945 	uint32_t		max_qp_mcg; /* 25 bits */
946 #define MCX_CAP_DEVICE_MAX_QP_MCG	0x1ffffff
947 
948 	uint8_t			__reserved__[3];
949 	uint8_t			log_max_mcq;
950 
951 	uint8_t			log_max_transport_domain; /* 5 bits */
952 #define MCX_CAP_DEVICE_LOG_MAX_TRANSORT_DOMAIN \
953 					0x1f
954 	uint8_t			log_max_pd; /* 5 bits */
955 #define MCX_CAP_DEVICE_LOG_MAX_PD	0x1f
956 	uint8_t			__reserved__[1];
957 	uint8_t			log_max_xrcd; /* 5 bits */
958 #define MCX_CAP_DEVICE_LOG_MAX_XRCD	0x1f
959 
960 	uint8_t			__reserved__[2];
961 	uint16_t		max_flow_counter;
962 
963 	uint8_t			log_max_rq; /* 5 bits */
964 #define MCX_CAP_DEVICE_LOG_MAX_RQ	0x1f
965 	uint8_t			log_max_sq; /* 5 bits */
966 #define MCX_CAP_DEVICE_LOG_MAX_SQ	0x1f
967 	uint8_t			log_max_tir; /* 5 bits */
968 #define MCX_CAP_DEVICE_LOG_MAX_TIR	0x1f
969 	uint8_t			log_max_tis; /* 5 bits */
970 #define MCX_CAP_DEVICE_LOG_MAX_TIS	0x1f
971 
972 	uint8_t 		flags8;
973 #define MCX_CAP_DEVICE_BASIC_CYCLIC_RCV_WQE \
974 					0x80
975 #define MCX_CAP_DEVICE_LOG_MAX_RMP	0x1f
976 	uint8_t			log_max_rqt; /* 5 bits */
977 #define MCX_CAP_DEVICE_LOG_MAX_RQT	0x1f
978 	uint8_t			log_max_rqt_size; /* 5 bits */
979 #define MCX_CAP_DEVICE_LOG_MAX_RQT_SIZE	0x1f
980 	uint8_t			log_max_tis_per_sq; /* 5 bits */
981 #define MCX_CAP_DEVICE_LOG_MAX_TIS_PER_SQ \
982 					0x1f
983 } __packed __aligned(8);
984 
985 CTASSERT(offsetof(struct mcx_cap_device, max_indirection) == 0x20);
986 CTASSERT(offsetof(struct mcx_cap_device, flags1) == 0x2c);
987 CTASSERT(offsetof(struct mcx_cap_device, flags2) == 0x30);
988 CTASSERT(offsetof(struct mcx_cap_device, snapshot_log_max_msg) == 0x38);
989 CTASSERT(offsetof(struct mcx_cap_device, flags5) == 0x40);
990 CTASSERT(offsetof(struct mcx_cap_device, flags7) == 0x4c);
991 CTASSERT(sizeof(struct mcx_cap_device) <= MCX_CMDQ_MAILBOX_DATASIZE);
992 
993 struct mcx_cmd_set_driver_version_in {
994 	uint16_t		cmd_opcode;
995 	uint8_t			cmd_reserved0[4];
996 	uint16_t		cmd_op_mod;
997 	uint8_t			cmd_reserved1[8];
998 } __packed __aligned(4);
999 
1000 struct mcx_cmd_set_driver_version_out {
1001 	uint8_t			cmd_status;
1002 	uint8_t			cmd_reserved0[3];
1003 	uint32_t		cmd_syndrome;
1004 	uint8_t			cmd_reserved1[8];
1005 } __packed __aligned(4);
1006 
1007 struct mcx_cmd_set_driver_version {
1008 	uint8_t			cmd_driver_version[64];
1009 } __packed __aligned(8);
1010 
1011 struct mcx_cmd_modify_nic_vport_context_in {
1012 	uint16_t		cmd_opcode;
1013 	uint8_t			cmd_reserved0[4];
1014 	uint16_t		cmd_op_mod;
1015 	uint8_t			cmd_reserved1[4];
1016 	uint32_t		cmd_field_select;
1017 #define MCX_CMD_MODIFY_NIC_VPORT_CONTEXT_FIELD_ADDR	0x04
1018 #define MCX_CMD_MODIFY_NIC_VPORT_CONTEXT_FIELD_PROMISC	0x10
1019 #define MCX_CMD_MODIFY_NIC_VPORT_CONTEXT_FIELD_MTU	0x40
1020 } __packed __aligned(4);
1021 
1022 struct mcx_cmd_modify_nic_vport_context_out {
1023 	uint8_t			cmd_status;
1024 	uint8_t			cmd_reserved0[3];
1025 	uint32_t		cmd_syndrome;
1026 	uint8_t			cmd_reserved1[8];
1027 } __packed __aligned(4);
1028 
1029 struct mcx_cmd_query_nic_vport_context_in {
1030 	uint16_t		cmd_opcode;
1031 	uint8_t			cmd_reserved0[4];
1032 	uint16_t		cmd_op_mod;
1033 	uint8_t			cmd_reserved1[4];
1034 	uint8_t			cmd_allowed_list_type;
1035 	uint8_t			cmd_reserved2[3];
1036 } __packed __aligned(4);
1037 
1038 struct mcx_cmd_query_nic_vport_context_out {
1039 	uint8_t			cmd_status;
1040 	uint8_t			cmd_reserved0[3];
1041 	uint32_t		cmd_syndrome;
1042 	uint8_t			cmd_reserved1[8];
1043 } __packed __aligned(4);
1044 
1045 struct mcx_nic_vport_ctx {
1046 	uint32_t		vp_min_wqe_inline_mode;
1047 	uint8_t			vp_reserved0[32];
1048 	uint32_t		vp_mtu;
1049 	uint8_t			vp_reserved1[200];
1050 	uint16_t		vp_flags;
1051 #define MCX_NIC_VPORT_CTX_LIST_UC_MAC			(0)
1052 #define MCX_NIC_VPORT_CTX_LIST_MC_MAC			(1 << 24)
1053 #define MCX_NIC_VPORT_CTX_LIST_VLAN			(2 << 24)
1054 #define MCX_NIC_VPORT_CTX_PROMISC_ALL			(1 << 13)
1055 #define MCX_NIC_VPORT_CTX_PROMISC_MCAST			(1 << 14)
1056 #define MCX_NIC_VPORT_CTX_PROMISC_UCAST			(1 << 15)
1057 	uint16_t		vp_allowed_list_size;
1058 	uint64_t		vp_perm_addr;
1059 	uint8_t			vp_reserved2[4];
1060 	/* allowed list follows */
1061 } __packed __aligned(4);
1062 
1063 struct mcx_counter {
1064 	uint64_t		packets;
1065 	uint64_t		octets;
1066 } __packed __aligned(4);
1067 
1068 struct mcx_nic_vport_counters {
1069 	struct mcx_counter	rx_err;
1070 	struct mcx_counter	tx_err;
1071 	uint8_t			reserved0[64]; /* 0x30 */
1072 	struct mcx_counter	rx_bcast;
1073 	struct mcx_counter	tx_bcast;
1074 	struct mcx_counter	rx_ucast;
1075 	struct mcx_counter	tx_ucast;
1076 	struct mcx_counter	rx_mcast;
1077 	struct mcx_counter	tx_mcast;
1078 	uint8_t			reserved1[0x210 - 0xd0];
1079 } __packed __aligned(4);
1080 
1081 struct mcx_cmd_query_vport_counters_in {
1082 	uint16_t		cmd_opcode;
1083 	uint8_t			cmd_reserved0[4];
1084 	uint16_t		cmd_op_mod;
1085 	uint8_t			cmd_reserved1[8];
1086 } __packed __aligned(4);
1087 
1088 struct mcx_cmd_query_vport_counters_mb_in {
1089 	uint8_t			cmd_reserved0[8];
1090 	uint8_t			cmd_clear;
1091 	uint8_t			cmd_reserved1[7];
1092 } __packed __aligned(4);
1093 
1094 struct mcx_cmd_query_vport_counters_out {
1095 	uint8_t			cmd_status;
1096 	uint8_t			cmd_reserved0[3];
1097 	uint32_t		cmd_syndrome;
1098 	uint8_t			cmd_reserved1[8];
1099 } __packed __aligned(4);
1100 
1101 struct mcx_cmd_query_flow_counter_in {
1102 	uint16_t		cmd_opcode;
1103 	uint8_t			cmd_reserved0[4];
1104 	uint16_t		cmd_op_mod;
1105 	uint8_t			cmd_reserved1[8];
1106 } __packed __aligned(4);
1107 
1108 struct mcx_cmd_query_flow_counter_mb_in {
1109 	uint8_t			cmd_reserved0[8];
1110 	uint8_t			cmd_clear;
1111 	uint8_t			cmd_reserved1[5];
1112 	uint16_t		cmd_flow_counter_id;
1113 } __packed __aligned(4);
1114 
1115 struct mcx_cmd_query_flow_counter_out {
1116 	uint8_t			cmd_status;
1117 	uint8_t			cmd_reserved0[3];
1118 	uint32_t		cmd_syndrome;
1119 	uint8_t			cmd_reserved1[8];
1120 } __packed __aligned(4);
1121 
1122 struct mcx_cmd_alloc_uar_in {
1123 	uint16_t		cmd_opcode;
1124 	uint8_t			cmd_reserved0[4];
1125 	uint16_t		cmd_op_mod;
1126 	uint8_t			cmd_reserved1[8];
1127 } __packed __aligned(4);
1128 
1129 struct mcx_cmd_alloc_uar_out {
1130 	uint8_t			cmd_status;
1131 	uint8_t			cmd_reserved0[3];
1132 	uint32_t		cmd_syndrome;
1133 	uint32_t		cmd_uar;
1134 	uint8_t			cmd_reserved1[4];
1135 } __packed __aligned(4);
1136 
1137 struct mcx_cmd_query_special_ctx_in {
1138 	uint16_t		cmd_opcode;
1139 	uint8_t			cmd_reserved0[4];
1140 	uint16_t		cmd_op_mod;
1141 	uint8_t			cmd_reserved1[8];
1142 } __packed __aligned(4);
1143 
1144 struct mcx_cmd_query_special_ctx_out {
1145 	uint8_t			cmd_status;
1146 	uint8_t			cmd_reserved0[3];
1147 	uint32_t		cmd_syndrome;
1148 	uint8_t			cmd_reserved1[4];
1149 	uint32_t		cmd_resd_lkey;
1150 } __packed __aligned(4);
1151 
1152 struct mcx_eq_ctx {
1153 	uint32_t		eq_status;
1154 #define MCX_EQ_CTX_ST_SHIFT		8
1155 #define MCX_EQ_CTX_ST_MASK		(0xf << MCX_EQ_CTX_ST_SHIFT)
1156 #define MCX_EQ_CTX_ST_ARMED		(0x9 << MCX_EQ_CTX_ST_SHIFT)
1157 #define MCX_EQ_CTX_ST_FIRED		(0xa << MCX_EQ_CTX_ST_SHIFT)
1158 #define MCX_EQ_CTX_OI_SHIFT		17
1159 #define MCX_EQ_CTX_OI			(1 << MCX_EQ_CTX_OI_SHIFT)
1160 #define MCX_EQ_CTX_EC_SHIFT		18
1161 #define MCX_EQ_CTX_EC			(1 << MCX_EQ_CTX_EC_SHIFT)
1162 #define MCX_EQ_CTX_STATUS_SHIFT		28
1163 #define MCX_EQ_CTX_STATUS_MASK		(0xf << MCX_EQ_CTX_STATUS_SHIFT)
1164 #define MCX_EQ_CTX_STATUS_OK		(0x0 << MCX_EQ_CTX_STATUS_SHIFT)
1165 #define MCX_EQ_CTX_STATUS_EQ_WRITE_FAILURE \
1166 					(0xa << MCX_EQ_CTX_STATUS_SHIFT)
1167 	uint32_t		eq_reserved1;
1168 	uint32_t		eq_page_offset;
1169 #define MCX_EQ_CTX_PAGE_OFFSET_SHIFT	5
1170 	uint32_t		eq_uar_size;
1171 #define MCX_EQ_CTX_UAR_PAGE_MASK	0xffffff
1172 #define MCX_EQ_CTX_LOG_EQ_SIZE_SHIFT	24
1173 	uint32_t		eq_reserved2;
1174 	uint8_t			eq_reserved3[3];
1175 	uint8_t			eq_intr;
1176 	uint32_t		eq_log_page_size;
1177 #define MCX_EQ_CTX_LOG_PAGE_SIZE_SHIFT	24
1178 	uint32_t		eq_reserved4[3];
1179 	uint32_t		eq_consumer_counter;
1180 	uint32_t		eq_producer_counter;
1181 #define MCX_EQ_CTX_COUNTER_MASK		0xffffff
1182 	uint32_t		eq_reserved5[4];
1183 } __packed __aligned(4);
1184 
1185 CTASSERT(sizeof(struct mcx_eq_ctx) == 64);
1186 
1187 struct mcx_cmd_create_eq_in {
1188 	uint16_t		cmd_opcode;
1189 	uint8_t			cmd_reserved0[4];
1190 	uint16_t		cmd_op_mod;
1191 	uint8_t			cmd_reserved1[8];
1192 } __packed __aligned(4);
1193 
1194 struct mcx_cmd_create_eq_mb_in {
1195 	struct mcx_eq_ctx	cmd_eq_ctx;
1196 	uint8_t			cmd_reserved0[8];
1197 	uint64_t		cmd_event_bitmask;
1198 #define MCX_EVENT_TYPE_COMPLETION	0x00
1199 #define MCX_EVENT_TYPE_CQ_ERROR		0x04
1200 #define MCX_EVENT_TYPE_INTERNAL_ERROR	0x08
1201 #define MCX_EVENT_TYPE_PORT_CHANGE	0x09
1202 #define MCX_EVENT_TYPE_CMD_COMPLETION	0x0a
1203 #define MCX_EVENT_TYPE_PAGE_REQUEST	0x0b
1204 #define MCX_EVENT_TYPE_LAST_WQE		0x13
1205 	uint8_t			cmd_reserved1[176];
1206 } __packed __aligned(4);
1207 
1208 struct mcx_cmd_create_eq_out {
1209 	uint8_t			cmd_status;
1210 	uint8_t			cmd_reserved0[3];
1211 	uint32_t		cmd_syndrome;
1212 	uint32_t		cmd_eqn;
1213 	uint8_t			cmd_reserved1[4];
1214 } __packed __aligned(4);
1215 
1216 struct mcx_eq_entry {
1217 	uint8_t			eq_reserved1;
1218 	uint8_t			eq_event_type;
1219 	uint8_t			eq_reserved2;
1220 	uint8_t			eq_event_sub_type;
1221 
1222 	uint8_t			eq_reserved3[28];
1223 	uint32_t		eq_event_data[7];
1224 	uint8_t			eq_reserved4[2];
1225 	uint8_t			eq_signature;
1226 	uint8_t			eq_owner;
1227 #define MCX_EQ_ENTRY_OWNER_INIT			1
1228 } __packed __aligned(4);
1229 
1230 CTASSERT(sizeof(struct mcx_eq_entry) == 64);
1231 
1232 struct mcx_cmd_alloc_pd_in {
1233 	uint16_t		cmd_opcode;
1234 	uint8_t			cmd_reserved0[4];
1235 	uint16_t		cmd_op_mod;
1236 	uint8_t			cmd_reserved1[8];
1237 } __packed __aligned(4);
1238 
1239 struct mcx_cmd_alloc_pd_out {
1240 	uint8_t			cmd_status;
1241 	uint8_t			cmd_reserved0[3];
1242 	uint32_t		cmd_syndrome;
1243 	uint32_t		cmd_pd;
1244 	uint8_t			cmd_reserved1[4];
1245 } __packed __aligned(4);
1246 
1247 struct mcx_cmd_alloc_td_in {
1248 	uint16_t		cmd_opcode;
1249 	uint8_t			cmd_reserved0[4];
1250 	uint16_t		cmd_op_mod;
1251 	uint8_t			cmd_reserved1[8];
1252 } __packed __aligned(4);
1253 
1254 struct mcx_cmd_alloc_td_out {
1255 	uint8_t			cmd_status;
1256 	uint8_t			cmd_reserved0[3];
1257 	uint32_t		cmd_syndrome;
1258 	uint32_t		cmd_tdomain;
1259 	uint8_t			cmd_reserved1[4];
1260 } __packed __aligned(4);
1261 
1262 struct mcx_cmd_create_tir_in {
1263 	uint16_t		cmd_opcode;
1264 	uint8_t			cmd_reserved0[4];
1265 	uint16_t		cmd_op_mod;
1266 	uint8_t			cmd_reserved1[8];
1267 } __packed __aligned(4);
1268 
1269 struct mcx_cmd_create_tir_mb_in {
1270 	uint8_t			cmd_reserved0[20];
1271 	uint32_t		cmd_disp_type;
1272 #define MCX_TIR_CTX_DISP_TYPE_DIRECT	0
1273 #define MCX_TIR_CTX_DISP_TYPE_INDIRECT	1
1274 #define MCX_TIR_CTX_DISP_TYPE_SHIFT	28
1275 	uint8_t			cmd_reserved1[8];
1276 	uint32_t		cmd_lro;
1277 	uint8_t			cmd_reserved2[8];
1278 	uint32_t		cmd_inline_rqn;
1279 	uint32_t		cmd_indir_table;
1280 	uint32_t		cmd_tdomain;
1281 #define MCX_TIR_CTX_HASH_TOEPLITZ	2
1282 #define MCX_TIR_CTX_HASH_SHIFT		28
1283 	uint8_t			cmd_rx_hash_key[40];
1284 	uint32_t		cmd_rx_hash_sel_outer;
1285 #define MCX_TIR_CTX_HASH_SEL_SRC_IP	(1 << 0)
1286 #define MCX_TIR_CTX_HASH_SEL_DST_IP	(1 << 1)
1287 #define MCX_TIR_CTX_HASH_SEL_SPORT	(1 << 2)
1288 #define MCX_TIR_CTX_HASH_SEL_DPORT	(1 << 3)
1289 #define MCX_TIR_CTX_HASH_SEL_IPV4	(0 << 31)
1290 #define MCX_TIR_CTX_HASH_SEL_IPV6	(1 << 31)
1291 #define MCX_TIR_CTX_HASH_SEL_TCP	(0 << 30)
1292 #define MCX_TIR_CTX_HASH_SEL_UDP	(1 << 30)
1293 	uint32_t		cmd_rx_hash_sel_inner;
1294 	uint8_t			cmd_reserved3[152];
1295 } __packed __aligned(4);
1296 
1297 struct mcx_cmd_create_tir_out {
1298 	uint8_t			cmd_status;
1299 	uint8_t			cmd_reserved0[3];
1300 	uint32_t		cmd_syndrome;
1301 	uint32_t		cmd_tirn;
1302 	uint8_t			cmd_reserved1[4];
1303 } __packed __aligned(4);
1304 
1305 struct mcx_cmd_destroy_tir_in {
1306 	uint16_t		cmd_opcode;
1307 	uint8_t			cmd_reserved0[4];
1308 	uint16_t		cmd_op_mod;
1309 	uint32_t		cmd_tirn;
1310 	uint8_t			cmd_reserved1[4];
1311 } __packed __aligned(4);
1312 
1313 struct mcx_cmd_destroy_tir_out {
1314 	uint8_t			cmd_status;
1315 	uint8_t			cmd_reserved0[3];
1316 	uint32_t		cmd_syndrome;
1317 	uint8_t			cmd_reserved1[8];
1318 } __packed __aligned(4);
1319 
1320 struct mcx_cmd_create_tis_in {
1321 	uint16_t		cmd_opcode;
1322 	uint8_t			cmd_reserved0[4];
1323 	uint16_t		cmd_op_mod;
1324 	uint8_t			cmd_reserved1[8];
1325 } __packed __aligned(4);
1326 
1327 struct mcx_cmd_create_tis_mb_in {
1328 	uint8_t			cmd_reserved[16];
1329 	uint32_t		cmd_prio;
1330 	uint8_t			cmd_reserved1[32];
1331 	uint32_t		cmd_tdomain;
1332 	uint8_t			cmd_reserved2[120];
1333 } __packed __aligned(4);
1334 
1335 struct mcx_cmd_create_tis_out {
1336 	uint8_t			cmd_status;
1337 	uint8_t			cmd_reserved0[3];
1338 	uint32_t		cmd_syndrome;
1339 	uint32_t		cmd_tisn;
1340 	uint8_t			cmd_reserved1[4];
1341 } __packed __aligned(4);
1342 
1343 struct mcx_cmd_destroy_tis_in {
1344 	uint16_t		cmd_opcode;
1345 	uint8_t			cmd_reserved0[4];
1346 	uint16_t		cmd_op_mod;
1347 	uint32_t		cmd_tisn;
1348 	uint8_t			cmd_reserved1[4];
1349 } __packed __aligned(4);
1350 
1351 struct mcx_cmd_destroy_tis_out {
1352 	uint8_t			cmd_status;
1353 	uint8_t			cmd_reserved0[3];
1354 	uint32_t		cmd_syndrome;
1355 	uint8_t			cmd_reserved1[8];
1356 } __packed __aligned(4);
1357 
1358 struct mcx_cmd_create_rqt_in {
1359 	uint16_t		cmd_opcode;
1360 	uint8_t			cmd_reserved0[4];
1361 	uint16_t		cmd_op_mod;
1362 	uint8_t			cmd_reserved1[8];
1363 } __packed __aligned(4);
1364 
1365 struct mcx_rqt_ctx {
1366 	uint8_t			cmd_reserved0[20];
1367 	uint16_t		cmd_reserved1;
1368 	uint16_t		cmd_rqt_max_size;
1369 	uint16_t		cmd_reserved2;
1370 	uint16_t		cmd_rqt_actual_size;
1371 	uint8_t			cmd_reserved3[212];
1372 } __packed __aligned(4);
1373 
1374 struct mcx_cmd_create_rqt_mb_in {
1375 	uint8_t			cmd_reserved0[16];
1376 	struct mcx_rqt_ctx	cmd_rqt;
1377 } __packed __aligned(4);
1378 
1379 struct mcx_cmd_create_rqt_out {
1380 	uint8_t			cmd_status;
1381 	uint8_t			cmd_reserved0[3];
1382 	uint32_t		cmd_syndrome;
1383 	uint32_t		cmd_rqtn;
1384 	uint8_t			cmd_reserved1[4];
1385 } __packed __aligned(4);
1386 
1387 struct mcx_cmd_destroy_rqt_in {
1388 	uint16_t		cmd_opcode;
1389 	uint8_t			cmd_reserved0[4];
1390 	uint16_t		cmd_op_mod;
1391 	uint32_t		cmd_rqtn;
1392 	uint8_t			cmd_reserved1[4];
1393 } __packed __aligned(4);
1394 
1395 struct mcx_cmd_destroy_rqt_out {
1396 	uint8_t			cmd_status;
1397 	uint8_t			cmd_reserved0[3];
1398 	uint32_t		cmd_syndrome;
1399 	uint8_t			cmd_reserved1[8];
1400 } __packed __aligned(4);
1401 
1402 struct mcx_cq_ctx {
1403 	uint32_t		cq_status;
1404 	uint32_t		cq_reserved1;
1405 	uint32_t		cq_page_offset;
1406 	uint32_t		cq_uar_size;
1407 #define MCX_CQ_CTX_UAR_PAGE_MASK	0xffffff
1408 #define MCX_CQ_CTX_LOG_CQ_SIZE_SHIFT	24
1409 	uint32_t		cq_period_max_count;
1410 #define MCX_CQ_CTX_PERIOD_SHIFT		16
1411 	uint32_t		cq_eqn;
1412 	uint32_t		cq_log_page_size;
1413 #define MCX_CQ_CTX_LOG_PAGE_SIZE_SHIFT	24
1414 	uint32_t		cq_reserved2;
1415 	uint32_t		cq_last_notified;
1416 	uint32_t		cq_last_solicit;
1417 	uint32_t		cq_consumer_counter;
1418 	uint32_t		cq_producer_counter;
1419 	uint8_t			cq_reserved3[8];
1420 	uint64_t		cq_doorbell;
1421 } __packed __aligned(4);
1422 
1423 CTASSERT(sizeof(struct mcx_cq_ctx) == 64);
1424 
1425 struct mcx_cmd_create_cq_in {
1426 	uint16_t		cmd_opcode;
1427 	uint8_t			cmd_reserved0[4];
1428 	uint16_t		cmd_op_mod;
1429 	uint8_t			cmd_reserved1[8];
1430 } __packed __aligned(4);
1431 
1432 struct mcx_cmd_create_cq_mb_in {
1433 	struct mcx_cq_ctx	cmd_cq_ctx;
1434 	uint8_t			cmd_reserved1[192];
1435 } __packed __aligned(4);
1436 
1437 struct mcx_cmd_create_cq_out {
1438 	uint8_t			cmd_status;
1439 	uint8_t			cmd_reserved0[3];
1440 	uint32_t		cmd_syndrome;
1441 	uint32_t		cmd_cqn;
1442 	uint8_t			cmd_reserved1[4];
1443 } __packed __aligned(4);
1444 
1445 struct mcx_cmd_destroy_cq_in {
1446 	uint16_t		cmd_opcode;
1447 	uint8_t			cmd_reserved0[4];
1448 	uint16_t		cmd_op_mod;
1449 	uint32_t		cmd_cqn;
1450 	uint8_t			cmd_reserved1[4];
1451 } __packed __aligned(4);
1452 
1453 struct mcx_cmd_destroy_cq_out {
1454 	uint8_t			cmd_status;
1455 	uint8_t			cmd_reserved0[3];
1456 	uint32_t		cmd_syndrome;
1457 	uint8_t			cmd_reserved1[8];
1458 } __packed __aligned(4);
1459 
1460 struct mcx_cq_entry {
1461 	uint32_t		__reserved__;
1462 	uint32_t		cq_lro;
1463 	uint32_t		cq_lro_ack_seq_num;
1464 	uint32_t		cq_rx_hash;
1465 	uint8_t			cq_rx_hash_type;
1466 	uint8_t			cq_ml_path;
1467 	uint16_t		__reserved__;
1468 	uint32_t		cq_checksum;
1469 	uint32_t		__reserved__;
1470 	uint32_t		cq_flags;
1471 #define MCX_CQ_ENTRY_FLAGS_L4_OK		(1 << 26)
1472 #define MCX_CQ_ENTRY_FLAGS_L3_OK		(1 << 25)
1473 #define MCX_CQ_ENTRY_FLAGS_L2_OK		(1 << 24)
1474 #define MCX_CQ_ENTRY_FLAGS_CV			(1 << 16)
1475 #define MCX_CQ_ENTRY_FLAGS_VLAN_MASK		(0xffff)
1476 
1477 	uint32_t		cq_lro_srqn;
1478 	uint32_t		__reserved__[2];
1479 	uint32_t		cq_byte_cnt;
1480 	uint64_t		cq_timestamp;
1481 	uint8_t			cq_rx_drops;
1482 	uint8_t			cq_flow_tag[3];
1483 	uint16_t		cq_wqe_count;
1484 	uint8_t			cq_signature;
1485 	uint8_t			cq_opcode_owner;
1486 #define MCX_CQ_ENTRY_FLAG_OWNER			(1 << 0)
1487 #define MCX_CQ_ENTRY_FLAG_SE			(1 << 1)
1488 #define MCX_CQ_ENTRY_FORMAT_SHIFT		2
1489 #define MCX_CQ_ENTRY_OPCODE_SHIFT		4
1490 
1491 #define MCX_CQ_ENTRY_FORMAT_NO_INLINE		0
1492 #define MCX_CQ_ENTRY_FORMAT_INLINE_32		1
1493 #define MCX_CQ_ENTRY_FORMAT_INLINE_64		2
1494 #define MCX_CQ_ENTRY_FORMAT_COMPRESSED		3
1495 
1496 #define MCX_CQ_ENTRY_OPCODE_REQ			0
1497 #define MCX_CQ_ENTRY_OPCODE_SEND		2
1498 #define MCX_CQ_ENTRY_OPCODE_REQ_ERR		13
1499 #define MCX_CQ_ENTRY_OPCODE_SEND_ERR		14
1500 #define MCX_CQ_ENTRY_OPCODE_INVALID		15
1501 
1502 } __packed __aligned(4);
1503 
1504 CTASSERT(sizeof(struct mcx_cq_entry) == 64);
1505 
1506 struct mcx_cq_doorbell {
1507 	uint32_t		 db_update_ci;
1508 	uint32_t		 db_arm_ci;
1509 #define MCX_CQ_DOORBELL_ARM_CMD_SN_SHIFT	28
1510 #define MCX_CQ_DOORBELL_ARM_CMD			(1 << 24)
1511 #define MCX_CQ_DOORBELL_ARM_CI_MASK		(0xffffff)
1512 } __packed __aligned(8);
1513 
1514 struct mcx_wq_ctx {
1515 	uint8_t			 wq_type;
1516 #define MCX_WQ_CTX_TYPE_CYCLIC			(1 << 4)
1517 #define MCX_WQ_CTX_TYPE_SIGNATURE		(1 << 3)
1518 	uint8_t			 wq_reserved0[5];
1519 	uint16_t		 wq_lwm;
1520 	uint32_t		 wq_pd;
1521 	uint32_t		 wq_uar_page;
1522 	uint64_t		 wq_doorbell;
1523 	uint32_t		 wq_hw_counter;
1524 	uint32_t		 wq_sw_counter;
1525 	uint16_t		 wq_log_stride;
1526 	uint8_t			 wq_log_page_sz;
1527 	uint8_t			 wq_log_size;
1528 	uint8_t			 wq_reserved1[156];
1529 } __packed __aligned(4);
1530 
1531 CTASSERT(sizeof(struct mcx_wq_ctx) == 0xC0);
1532 
1533 struct mcx_sq_ctx {
1534 	uint32_t		sq_flags;
1535 #define MCX_SQ_CTX_RLKEY			(1 << 31)
1536 #define MCX_SQ_CTX_FRE_SHIFT			(1 << 29)
1537 #define MCX_SQ_CTX_FLUSH_IN_ERROR		(1 << 28)
1538 #define MCX_SQ_CTX_MIN_WQE_INLINE_SHIFT		24
1539 #define MCX_SQ_CTX_STATE_SHIFT			20
1540 	uint32_t		sq_user_index;
1541 	uint32_t		sq_cqn;
1542 	uint32_t		sq_reserved1[5];
1543 	uint32_t		sq_tis_lst_sz;
1544 #define MCX_SQ_CTX_TIS_LST_SZ_SHIFT		16
1545 	uint32_t		sq_reserved2[2];
1546 	uint32_t		sq_tis_num;
1547 	struct mcx_wq_ctx	sq_wq;
1548 } __packed __aligned(4);
1549 
1550 struct mcx_sq_entry_seg {
1551 	uint32_t		sqs_byte_count;
1552 	uint32_t		sqs_lkey;
1553 	uint64_t		sqs_addr;
1554 } __packed __aligned(4);
1555 
1556 struct mcx_sq_entry {
1557 	/* control segment */
1558 	uint32_t		sqe_opcode_index;
1559 #define MCX_SQE_WQE_INDEX_SHIFT			8
1560 #define MCX_SQE_WQE_OPCODE_NOP			0x00
1561 #define MCX_SQE_WQE_OPCODE_SEND			0x0a
1562 	uint32_t		sqe_ds_sq_num;
1563 #define MCX_SQE_SQ_NUM_SHIFT			8
1564 	uint32_t		sqe_signature;
1565 #define MCX_SQE_SIGNATURE_SHIFT			24
1566 #define MCX_SQE_SOLICITED_EVENT			0x02
1567 #define MCX_SQE_CE_CQE_ON_ERR			0x00
1568 #define MCX_SQE_CE_CQE_FIRST_ERR		0x04
1569 #define MCX_SQE_CE_CQE_ALWAYS			0x08
1570 #define MCX_SQE_CE_CQE_SOLICIT			0x0C
1571 #define MCX_SQE_FM_NO_FENCE			0x00
1572 #define MCX_SQE_FM_SMALL_FENCE			0x40
1573 	uint32_t		sqe_mkey;
1574 
1575 	/* ethernet segment */
1576 	uint32_t		sqe_reserved1;
1577 	uint32_t		sqe_mss_csum;
1578 #define MCX_SQE_L4_CSUM				(1 << 31)
1579 #define MCX_SQE_L3_CSUM				(1 << 30)
1580 	uint32_t		sqe_reserved2;
1581 	uint16_t		sqe_inline_header_size;
1582 	uint16_t		sqe_inline_headers[9];
1583 
1584 	/* data segment */
1585 	struct mcx_sq_entry_seg sqe_segs[1];
1586 } __packed __aligned(64);
1587 
1588 CTASSERT(sizeof(struct mcx_sq_entry) == 64);
1589 
1590 struct mcx_cmd_create_sq_in {
1591 	uint16_t		cmd_opcode;
1592 	uint8_t			cmd_reserved0[4];
1593 	uint16_t		cmd_op_mod;
1594 	uint8_t			cmd_reserved1[8];
1595 } __packed __aligned(4);
1596 
1597 struct mcx_cmd_create_sq_out {
1598 	uint8_t			cmd_status;
1599 	uint8_t			cmd_reserved0[3];
1600 	uint32_t		cmd_syndrome;
1601 	uint32_t		cmd_sqn;
1602 	uint8_t			cmd_reserved1[4];
1603 } __packed __aligned(4);
1604 
1605 struct mcx_cmd_modify_sq_in {
1606 	uint16_t		cmd_opcode;
1607 	uint8_t			cmd_reserved0[4];
1608 	uint16_t		cmd_op_mod;
1609 	uint32_t		cmd_sq_state;
1610 	uint8_t			cmd_reserved1[4];
1611 } __packed __aligned(4);
1612 
1613 struct mcx_cmd_modify_sq_mb_in {
1614 	uint32_t		cmd_modify_hi;
1615 	uint32_t		cmd_modify_lo;
1616 	uint8_t			cmd_reserved0[8];
1617 	struct mcx_sq_ctx	cmd_sq_ctx;
1618 } __packed __aligned(4);
1619 
1620 struct mcx_cmd_modify_sq_out {
1621 	uint8_t			cmd_status;
1622 	uint8_t			cmd_reserved0[3];
1623 	uint32_t		cmd_syndrome;
1624 	uint8_t			cmd_reserved1[8];
1625 } __packed __aligned(4);
1626 
1627 struct mcx_cmd_destroy_sq_in {
1628 	uint16_t		cmd_opcode;
1629 	uint8_t			cmd_reserved0[4];
1630 	uint16_t		cmd_op_mod;
1631 	uint32_t		cmd_sqn;
1632 	uint8_t			cmd_reserved1[4];
1633 } __packed __aligned(4);
1634 
1635 struct mcx_cmd_destroy_sq_out {
1636 	uint8_t			cmd_status;
1637 	uint8_t			cmd_reserved0[3];
1638 	uint32_t		cmd_syndrome;
1639 	uint8_t			cmd_reserved1[8];
1640 } __packed __aligned(4);
1641 
1642 
1643 struct mcx_rq_ctx {
1644 	uint32_t		rq_flags;
1645 #define MCX_RQ_CTX_RLKEY			(1 << 31)
1646 #define MCX_RQ_CTX_VLAN_STRIP_DIS		(1 << 28)
1647 #define MCX_RQ_CTX_MEM_RQ_TYPE_SHIFT		24
1648 #define MCX_RQ_CTX_STATE_SHIFT			20
1649 #define MCX_RQ_CTX_FLUSH_IN_ERROR		(1 << 18)
1650 	uint32_t		rq_user_index;
1651 	uint32_t		rq_cqn;
1652 	uint32_t		rq_reserved1;
1653 	uint32_t		rq_rmpn;
1654 	uint32_t		rq_reserved2[7];
1655 	struct mcx_wq_ctx	rq_wq;
1656 } __packed __aligned(4);
1657 
1658 struct mcx_rq_entry {
1659 	uint32_t		rqe_byte_count;
1660 	uint32_t		rqe_lkey;
1661 	uint64_t		rqe_addr;
1662 } __packed __aligned(16);
1663 
1664 struct mcx_cmd_create_rq_in {
1665 	uint16_t		cmd_opcode;
1666 	uint8_t			cmd_reserved0[4];
1667 	uint16_t		cmd_op_mod;
1668 	uint8_t			cmd_reserved1[8];
1669 } __packed __aligned(4);
1670 
1671 struct mcx_cmd_create_rq_out {
1672 	uint8_t			cmd_status;
1673 	uint8_t			cmd_reserved0[3];
1674 	uint32_t		cmd_syndrome;
1675 	uint32_t		cmd_rqn;
1676 	uint8_t			cmd_reserved1[4];
1677 } __packed __aligned(4);
1678 
1679 struct mcx_cmd_modify_rq_in {
1680 	uint16_t		cmd_opcode;
1681 	uint8_t			cmd_reserved0[4];
1682 	uint16_t		cmd_op_mod;
1683 	uint32_t		cmd_rq_state;
1684 	uint8_t			cmd_reserved1[4];
1685 } __packed __aligned(4);
1686 
1687 struct mcx_cmd_modify_rq_mb_in {
1688 	uint32_t		cmd_modify_hi;
1689 	uint32_t		cmd_modify_lo;
1690 	uint8_t			cmd_reserved0[8];
1691 	struct mcx_rq_ctx	cmd_rq_ctx;
1692 } __packed __aligned(4);
1693 
1694 struct mcx_cmd_modify_rq_out {
1695 	uint8_t			cmd_status;
1696 	uint8_t			cmd_reserved0[3];
1697 	uint32_t		cmd_syndrome;
1698 	uint8_t			cmd_reserved1[8];
1699 } __packed __aligned(4);
1700 
1701 struct mcx_cmd_destroy_rq_in {
1702 	uint16_t		cmd_opcode;
1703 	uint8_t			cmd_reserved0[4];
1704 	uint16_t		cmd_op_mod;
1705 	uint32_t		cmd_rqn;
1706 	uint8_t			cmd_reserved1[4];
1707 } __packed __aligned(4);
1708 
1709 struct mcx_cmd_destroy_rq_out {
1710 	uint8_t			cmd_status;
1711 	uint8_t			cmd_reserved0[3];
1712 	uint32_t		cmd_syndrome;
1713 	uint8_t			cmd_reserved1[8];
1714 } __packed __aligned(4);
1715 
1716 struct mcx_cmd_create_flow_table_in {
1717 	uint16_t		cmd_opcode;
1718 	uint8_t			cmd_reserved0[4];
1719 	uint16_t		cmd_op_mod;
1720 	uint8_t			cmd_reserved1[8];
1721 } __packed __aligned(4);
1722 
1723 struct mcx_flow_table_ctx {
1724 	uint8_t			ft_miss_action;
1725 	uint8_t			ft_level;
1726 	uint8_t			ft_reserved0;
1727 	uint8_t			ft_log_size;
1728 	uint32_t		ft_table_miss_id;
1729 	uint8_t			ft_reserved1[28];
1730 } __packed __aligned(4);
1731 
1732 struct mcx_cmd_create_flow_table_mb_in {
1733 	uint8_t			cmd_table_type;
1734 	uint8_t			cmd_reserved0[7];
1735 	struct mcx_flow_table_ctx cmd_ctx;
1736 } __packed __aligned(4);
1737 
1738 struct mcx_cmd_create_flow_table_out {
1739 	uint8_t			cmd_status;
1740 	uint8_t			cmd_reserved0[3];
1741 	uint32_t		cmd_syndrome;
1742 	uint32_t		cmd_table_id;
1743 	uint8_t			cmd_reserved1[4];
1744 } __packed __aligned(4);
1745 
1746 struct mcx_cmd_destroy_flow_table_in {
1747 	uint16_t		cmd_opcode;
1748 	uint8_t			cmd_reserved0[4];
1749 	uint16_t		cmd_op_mod;
1750 	uint8_t			cmd_reserved1[8];
1751 } __packed __aligned(4);
1752 
1753 struct mcx_cmd_destroy_flow_table_mb_in {
1754 	uint8_t			cmd_table_type;
1755 	uint8_t			cmd_reserved0[3];
1756 	uint32_t		cmd_table_id;
1757 	uint8_t			cmd_reserved1[40];
1758 } __packed __aligned(4);
1759 
1760 struct mcx_cmd_destroy_flow_table_out {
1761 	uint8_t			cmd_status;
1762 	uint8_t			cmd_reserved0[3];
1763 	uint32_t		cmd_syndrome;
1764 	uint8_t			cmd_reserved1[8];
1765 } __packed __aligned(4);
1766 
1767 struct mcx_cmd_set_flow_table_root_in {
1768 	uint16_t		cmd_opcode;
1769 	uint8_t			cmd_reserved0[4];
1770 	uint16_t		cmd_op_mod;
1771 	uint8_t			cmd_reserved1[8];
1772 } __packed __aligned(4);
1773 
1774 struct mcx_cmd_set_flow_table_root_mb_in {
1775 	uint8_t			cmd_table_type;
1776 	uint8_t			cmd_reserved0[3];
1777 	uint32_t		cmd_table_id;
1778 	uint8_t			cmd_reserved1[56];
1779 } __packed __aligned(4);
1780 
1781 struct mcx_cmd_set_flow_table_root_out {
1782 	uint8_t			cmd_status;
1783 	uint8_t			cmd_reserved0[3];
1784 	uint32_t		cmd_syndrome;
1785 	uint8_t			cmd_reserved1[8];
1786 } __packed __aligned(4);
1787 
1788 struct mcx_flow_match {
1789 	/* outer headers */
1790 	uint8_t			mc_src_mac[6];
1791 	uint16_t		mc_ethertype;
1792 	uint8_t			mc_dest_mac[6];
1793 	uint16_t		mc_first_vlan;
1794 	uint8_t			mc_ip_proto;
1795 	uint8_t			mc_ip_dscp_ecn;
1796 	uint8_t			mc_vlan_flags;
1797 #define MCX_FLOW_MATCH_IP_FRAG	(1 << 5)
1798 	uint8_t			mc_tcp_flags;
1799 	uint16_t		mc_tcp_sport;
1800 	uint16_t		mc_tcp_dport;
1801 	uint32_t		mc_reserved0;
1802 	uint16_t		mc_udp_sport;
1803 	uint16_t		mc_udp_dport;
1804 	uint8_t			mc_src_ip[16];
1805 	uint8_t			mc_dest_ip[16];
1806 
1807 	/* misc parameters */
1808 	uint8_t			mc_reserved1[8];
1809 	uint16_t		mc_second_vlan;
1810 	uint8_t			mc_reserved2[2];
1811 	uint8_t			mc_second_vlan_flags;
1812 	uint8_t			mc_reserved3[15];
1813 	uint32_t		mc_outer_ipv6_flow_label;
1814 	uint8_t			mc_reserved4[32];
1815 
1816 	uint8_t			mc_reserved[384];
1817 } __packed __aligned(4);
1818 
1819 CTASSERT(sizeof(struct mcx_flow_match) == 512);
1820 
1821 struct mcx_cmd_create_flow_group_in {
1822 	uint16_t		cmd_opcode;
1823 	uint8_t			cmd_reserved0[4];
1824 	uint16_t		cmd_op_mod;
1825 	uint8_t			cmd_reserved1[8];
1826 } __packed __aligned(4);
1827 
1828 struct mcx_cmd_create_flow_group_mb_in {
1829 	uint8_t			cmd_table_type;
1830 	uint8_t			cmd_reserved0[3];
1831 	uint32_t		cmd_table_id;
1832 	uint8_t			cmd_reserved1[4];
1833 	uint32_t		cmd_start_flow_index;
1834 	uint8_t			cmd_reserved2[4];
1835 	uint32_t		cmd_end_flow_index;
1836 	uint8_t			cmd_reserved3[23];
1837 	uint8_t			cmd_match_criteria_enable;
1838 #define MCX_CREATE_FLOW_GROUP_CRIT_OUTER	(1 << 0)
1839 #define MCX_CREATE_FLOW_GROUP_CRIT_MISC		(1 << 1)
1840 #define MCX_CREATE_FLOW_GROUP_CRIT_INNER	(1 << 2)
1841 	struct mcx_flow_match	cmd_match_criteria;
1842 	uint8_t			cmd_reserved4[448];
1843 } __packed __aligned(4);
1844 
1845 struct mcx_cmd_create_flow_group_out {
1846 	uint8_t			cmd_status;
1847 	uint8_t			cmd_reserved0[3];
1848 	uint32_t		cmd_syndrome;
1849 	uint32_t		cmd_group_id;
1850 	uint8_t			cmd_reserved1[4];
1851 } __packed __aligned(4);
1852 
1853 struct mcx_flow_ctx {
1854 	uint8_t			fc_reserved0[4];
1855 	uint32_t		fc_group_id;
1856 	uint32_t		fc_flow_tag;
1857 	uint32_t		fc_action;
1858 #define MCX_FLOW_CONTEXT_ACTION_ALLOW		(1 << 0)
1859 #define MCX_FLOW_CONTEXT_ACTION_DROP		(1 << 1)
1860 #define MCX_FLOW_CONTEXT_ACTION_FORWARD		(1 << 2)
1861 #define MCX_FLOW_CONTEXT_ACTION_COUNT		(1 << 3)
1862 	uint32_t		fc_dest_list_size;
1863 	uint32_t		fc_counter_list_size;
1864 	uint8_t			fc_reserved1[40];
1865 	struct mcx_flow_match	fc_match_value;
1866 	uint8_t			fc_reserved2[192];
1867 } __packed __aligned(4);
1868 
1869 #define MCX_FLOW_CONTEXT_DEST_TYPE_TABLE	(1 << 24)
1870 #define MCX_FLOW_CONTEXT_DEST_TYPE_TIR		(2 << 24)
1871 
1872 struct mcx_cmd_destroy_flow_group_in {
1873 	uint16_t		cmd_opcode;
1874 	uint8_t			cmd_reserved0[4];
1875 	uint16_t		cmd_op_mod;
1876 	uint8_t			cmd_reserved1[8];
1877 } __packed __aligned(4);
1878 
1879 struct mcx_cmd_destroy_flow_group_mb_in {
1880 	uint8_t			cmd_table_type;
1881 	uint8_t			cmd_reserved0[3];
1882 	uint32_t		cmd_table_id;
1883 	uint32_t		cmd_group_id;
1884 	uint8_t			cmd_reserved1[36];
1885 } __packed __aligned(4);
1886 
1887 struct mcx_cmd_destroy_flow_group_out {
1888 	uint8_t			cmd_status;
1889 	uint8_t			cmd_reserved0[3];
1890 	uint32_t		cmd_syndrome;
1891 	uint8_t			cmd_reserved1[8];
1892 } __packed __aligned(4);
1893 
1894 struct mcx_cmd_set_flow_table_entry_in {
1895 	uint16_t		cmd_opcode;
1896 	uint8_t			cmd_reserved0[4];
1897 	uint16_t		cmd_op_mod;
1898 	uint8_t			cmd_reserved1[8];
1899 } __packed __aligned(4);
1900 
1901 struct mcx_cmd_set_flow_table_entry_mb_in {
1902 	uint8_t			cmd_table_type;
1903 	uint8_t			cmd_reserved0[3];
1904 	uint32_t		cmd_table_id;
1905 	uint32_t		cmd_modify_enable_mask;
1906 	uint8_t			cmd_reserved1[4];
1907 	uint32_t		cmd_flow_index;
1908 	uint8_t			cmd_reserved2[28];
1909 	struct mcx_flow_ctx	cmd_flow_ctx;
1910 } __packed __aligned(4);
1911 
1912 struct mcx_cmd_set_flow_table_entry_out {
1913 	uint8_t			cmd_status;
1914 	uint8_t			cmd_reserved0[3];
1915 	uint32_t		cmd_syndrome;
1916 	uint8_t			cmd_reserved1[8];
1917 } __packed __aligned(4);
1918 
1919 struct mcx_cmd_query_flow_table_entry_in {
1920 	uint16_t		cmd_opcode;
1921 	uint8_t			cmd_reserved0[4];
1922 	uint16_t		cmd_op_mod;
1923 	uint8_t			cmd_reserved1[8];
1924 } __packed __aligned(4);
1925 
1926 struct mcx_cmd_query_flow_table_entry_mb_in {
1927 	uint8_t			cmd_table_type;
1928 	uint8_t			cmd_reserved0[3];
1929 	uint32_t		cmd_table_id;
1930 	uint8_t			cmd_reserved1[8];
1931 	uint32_t		cmd_flow_index;
1932 	uint8_t			cmd_reserved2[28];
1933 } __packed __aligned(4);
1934 
1935 struct mcx_cmd_query_flow_table_entry_out {
1936 	uint8_t			cmd_status;
1937 	uint8_t			cmd_reserved0[3];
1938 	uint32_t		cmd_syndrome;
1939 	uint8_t			cmd_reserved1[8];
1940 } __packed __aligned(4);
1941 
1942 struct mcx_cmd_query_flow_table_entry_mb_out {
1943 	uint8_t			cmd_reserved0[48];
1944 	struct mcx_flow_ctx	cmd_flow_ctx;
1945 } __packed __aligned(4);
1946 
1947 struct mcx_cmd_delete_flow_table_entry_in {
1948 	uint16_t		cmd_opcode;
1949 	uint8_t			cmd_reserved0[4];
1950 	uint16_t		cmd_op_mod;
1951 	uint8_t			cmd_reserved1[8];
1952 } __packed __aligned(4);
1953 
1954 struct mcx_cmd_delete_flow_table_entry_mb_in {
1955 	uint8_t			cmd_table_type;
1956 	uint8_t			cmd_reserved0[3];
1957 	uint32_t		cmd_table_id;
1958 	uint8_t			cmd_reserved1[8];
1959 	uint32_t		cmd_flow_index;
1960 	uint8_t			cmd_reserved2[28];
1961 } __packed __aligned(4);
1962 
1963 struct mcx_cmd_delete_flow_table_entry_out {
1964 	uint8_t			cmd_status;
1965 	uint8_t			cmd_reserved0[3];
1966 	uint32_t		cmd_syndrome;
1967 	uint8_t			cmd_reserved1[8];
1968 } __packed __aligned(4);
1969 
1970 struct mcx_cmd_query_flow_group_in {
1971 	uint16_t		cmd_opcode;
1972 	uint8_t			cmd_reserved0[4];
1973 	uint16_t		cmd_op_mod;
1974 	uint8_t			cmd_reserved1[8];
1975 } __packed __aligned(4);
1976 
1977 struct mcx_cmd_query_flow_group_mb_in {
1978 	uint8_t			cmd_table_type;
1979 	uint8_t			cmd_reserved0[3];
1980 	uint32_t		cmd_table_id;
1981 	uint32_t		cmd_group_id;
1982 	uint8_t			cmd_reserved1[36];
1983 } __packed __aligned(4);
1984 
1985 struct mcx_cmd_query_flow_group_out {
1986 	uint8_t			cmd_status;
1987 	uint8_t			cmd_reserved0[3];
1988 	uint32_t		cmd_syndrome;
1989 	uint8_t			cmd_reserved1[8];
1990 } __packed __aligned(4);
1991 
1992 struct mcx_cmd_query_flow_group_mb_out {
1993 	uint8_t			cmd_reserved0[12];
1994 	uint32_t		cmd_start_flow_index;
1995 	uint8_t			cmd_reserved1[4];
1996 	uint32_t		cmd_end_flow_index;
1997 	uint8_t			cmd_reserved2[20];
1998 	uint32_t		cmd_match_criteria_enable;
1999 	uint8_t			cmd_match_criteria[512];
2000 	uint8_t			cmd_reserved4[448];
2001 } __packed __aligned(4);
2002 
2003 struct mcx_cmd_query_flow_table_in {
2004 	uint16_t		cmd_opcode;
2005 	uint8_t			cmd_reserved0[4];
2006 	uint16_t		cmd_op_mod;
2007 	uint8_t			cmd_reserved1[8];
2008 } __packed __aligned(4);
2009 
2010 struct mcx_cmd_query_flow_table_mb_in {
2011 	uint8_t			cmd_table_type;
2012 	uint8_t			cmd_reserved0[3];
2013 	uint32_t		cmd_table_id;
2014 	uint8_t			cmd_reserved1[40];
2015 } __packed __aligned(4);
2016 
2017 struct mcx_cmd_query_flow_table_out {
2018 	uint8_t			cmd_status;
2019 	uint8_t			cmd_reserved0[3];
2020 	uint32_t		cmd_syndrome;
2021 	uint8_t			cmd_reserved1[8];
2022 } __packed __aligned(4);
2023 
2024 struct mcx_cmd_query_flow_table_mb_out {
2025 	uint8_t			cmd_reserved0[4];
2026 	struct mcx_flow_table_ctx cmd_ctx;
2027 } __packed __aligned(4);
2028 
2029 struct mcx_cmd_alloc_flow_counter_in {
2030 	uint16_t		cmd_opcode;
2031 	uint8_t			cmd_reserved0[4];
2032 	uint16_t		cmd_op_mod;
2033 	uint8_t			cmd_reserved1[8];
2034 } __packed __aligned(4);
2035 
2036 struct mcx_cmd_query_rq_in {
2037 	uint16_t		cmd_opcode;
2038 	uint8_t			cmd_reserved0[4];
2039 	uint16_t		cmd_op_mod;
2040 	uint32_t		cmd_rqn;
2041 	uint8_t			cmd_reserved1[4];
2042 } __packed __aligned(4);
2043 
2044 struct mcx_cmd_query_rq_out {
2045 	uint8_t			cmd_status;
2046 	uint8_t			cmd_reserved0[3];
2047 	uint32_t		cmd_syndrome;
2048 	uint8_t			cmd_reserved1[8];
2049 } __packed __aligned(4);
2050 
2051 struct mcx_cmd_query_rq_mb_out {
2052 	uint8_t			cmd_reserved0[16];
2053 	struct mcx_rq_ctx	cmd_ctx;
2054 };
2055 
2056 struct mcx_cmd_query_sq_in {
2057 	uint16_t		cmd_opcode;
2058 	uint8_t			cmd_reserved0[4];
2059 	uint16_t		cmd_op_mod;
2060 	uint32_t		cmd_sqn;
2061 	uint8_t			cmd_reserved1[4];
2062 } __packed __aligned(4);
2063 
2064 struct mcx_cmd_query_sq_out {
2065 	uint8_t			cmd_status;
2066 	uint8_t			cmd_reserved0[3];
2067 	uint32_t		cmd_syndrome;
2068 	uint8_t			cmd_reserved1[8];
2069 } __packed __aligned(4);
2070 
2071 struct mcx_cmd_query_sq_mb_out {
2072 	uint8_t			cmd_reserved0[16];
2073 	struct mcx_sq_ctx	cmd_ctx;
2074 };
2075 
2076 struct mcx_cmd_alloc_flow_counter_out {
2077 	uint8_t			cmd_status;
2078 	uint8_t			cmd_reserved0[3];
2079 	uint32_t		cmd_syndrome;
2080 	uint8_t			cmd_reserved1[2];
2081 	uint16_t		cmd_flow_counter_id;
2082 	uint8_t			cmd_reserved2[4];
2083 } __packed __aligned(4);
2084 
2085 struct mcx_wq_doorbell {
2086 	uint32_t		 db_recv_counter;
2087 	uint32_t		 db_send_counter;
2088 } __packed __aligned(8);
2089 
2090 struct mcx_dmamem {
2091 	bus_dmamap_t		 mxm_map;
2092 	bus_dma_segment_t	 mxm_seg;
2093 	int			 mxm_nsegs;
2094 	size_t			 mxm_size;
2095 	caddr_t			 mxm_kva;
2096 };
2097 #define MCX_DMA_MAP(_mxm)	((_mxm)->mxm_map)
2098 #define MCX_DMA_DVA(_mxm)	((_mxm)->mxm_map->dm_segs[0].ds_addr)
2099 #define MCX_DMA_KVA(_mxm)	((void *)(_mxm)->mxm_kva)
2100 #define MCX_DMA_LEN(_mxm)	((_mxm)->mxm_size)
2101 
2102 struct mcx_hwmem {
2103 	bus_dmamap_t		 mhm_map;
2104 	bus_dma_segment_t	*mhm_segs;
2105 	unsigned int		 mhm_seg_count;
2106 	unsigned int		 mhm_npages;
2107 };
2108 
2109 struct mcx_slot {
2110 	bus_dmamap_t		 ms_map;
2111 	struct mbuf		*ms_m;
2112 };
2113 
2114 struct mcx_eq {
2115 	int			 eq_n;
2116 	uint32_t		 eq_cons;
2117 	struct mcx_dmamem	 eq_mem;
2118 };
2119 
2120 struct mcx_cq {
2121 	int			 cq_n;
2122 	struct mcx_dmamem	 cq_mem;
2123 	uint32_t		*cq_doorbell;
2124 	uint32_t		 cq_cons;
2125 	uint32_t		 cq_count;
2126 };
2127 
2128 struct mcx_calibration {
2129 	uint64_t		 c_timestamp;	/* previous mcx chip time */
2130 	uint64_t		 c_uptime;	/* previous kernel nanouptime */
2131 	uint64_t		 c_tbase;	/* mcx chip time */
2132 	uint64_t		 c_ubase;	/* kernel nanouptime */
2133 	uint64_t		 c_tdiff;
2134 	uint64_t		 c_udiff;
2135 };
2136 
2137 #define MCX_CALIBRATE_FIRST    2
2138 #define MCX_CALIBRATE_NORMAL   30
2139 
2140 struct mcx_rx {
2141 	struct mcx_softc	*rx_softc;
2142 	struct ifiqueue		*rx_ifiq;
2143 
2144 	int			 rx_rqn;
2145 	struct mcx_dmamem	 rx_rq_mem;
2146 	struct mcx_slot		*rx_slots;
2147 	uint32_t		*rx_doorbell;
2148 
2149 	uint32_t		 rx_prod;
2150 	struct timeout		 rx_refill;
2151 	struct if_rxring	 rx_rxr;
2152 } __aligned(64);
2153 
2154 struct mcx_tx {
2155 	struct mcx_softc	*tx_softc;
2156 	struct ifqueue		*tx_ifq;
2157 
2158 	int			 tx_uar;
2159 	int			 tx_sqn;
2160 	struct mcx_dmamem	 tx_sq_mem;
2161 	struct mcx_slot		*tx_slots;
2162 	uint32_t		*tx_doorbell;
2163 	int			 tx_bf_offset;
2164 
2165 	uint32_t		 tx_cons;
2166 	uint32_t		 tx_prod;
2167 } __aligned(64);
2168 
2169 struct mcx_queues {
2170 	char			 q_name[16];
2171 	void			*q_ihc;
2172 	struct mcx_softc	*q_sc;
2173 	int			 q_uar;
2174 	int			 q_index;
2175 	struct mcx_rx		 q_rx;
2176 	struct mcx_tx		 q_tx;
2177 	struct mcx_cq		 q_cq;
2178 	struct mcx_eq		 q_eq;
2179 };
2180 
2181 struct mcx_flow_group {
2182 	int			 g_id;
2183 	int			 g_table;
2184 	int			 g_start;
2185 	int			 g_size;
2186 };
2187 
2188 #define MCX_FLOW_GROUP_PROMISC	 0
2189 #define MCX_FLOW_GROUP_ALLMULTI	 1
2190 #define MCX_FLOW_GROUP_MAC	 2
2191 #define MCX_FLOW_GROUP_RSS_L4	 3
2192 #define MCX_FLOW_GROUP_RSS_L3	 4
2193 #define MCX_FLOW_GROUP_RSS_NONE	 5
2194 #define MCX_NUM_FLOW_GROUPS	 6
2195 
2196 #define MCX_HASH_SEL_L3		MCX_TIR_CTX_HASH_SEL_SRC_IP | \
2197 				MCX_TIR_CTX_HASH_SEL_DST_IP
2198 #define MCX_HASH_SEL_L4		MCX_HASH_SEL_L3 | MCX_TIR_CTX_HASH_SEL_SPORT | \
2199 				MCX_TIR_CTX_HASH_SEL_DPORT
2200 
2201 #define MCX_RSS_HASH_SEL_V4_TCP MCX_HASH_SEL_L4 | MCX_TIR_CTX_HASH_SEL_TCP  |\
2202 				MCX_TIR_CTX_HASH_SEL_IPV4
2203 #define MCX_RSS_HASH_SEL_V6_TCP	MCX_HASH_SEL_L4 | MCX_TIR_CTX_HASH_SEL_TCP | \
2204 				MCX_TIR_CTX_HASH_SEL_IPV6
2205 #define MCX_RSS_HASH_SEL_V4_UDP	MCX_HASH_SEL_L4 | MCX_TIR_CTX_HASH_SEL_UDP | \
2206 				MCX_TIR_CTX_HASH_SEL_IPV4
2207 #define MCX_RSS_HASH_SEL_V6_UDP	MCX_HASH_SEL_L4 | MCX_TIR_CTX_HASH_SEL_UDP | \
2208 				MCX_TIR_CTX_HASH_SEL_IPV6
2209 #define MCX_RSS_HASH_SEL_V4	MCX_HASH_SEL_L3 | MCX_TIR_CTX_HASH_SEL_IPV4
2210 #define MCX_RSS_HASH_SEL_V6	MCX_HASH_SEL_L3 | MCX_TIR_CTX_HASH_SEL_IPV6
2211 
2212 /*
2213  * There are a few different pieces involved in configuring RSS.
2214  * A Receive Queue Table (RQT) is the indirection table that maps packets to
2215  * different rx queues based on a hash value.  We only create one, because
2216  * we want to scatter any traffic we can apply RSS to across all our rx
2217  * queues.  Anything else will only be delivered to the first rx queue,
2218  * which doesn't require an RQT.
2219  *
2220  * A Transport Interface Receive (TIR) delivers packets to either a single rx
2221  * queue or an RQT, and in the latter case, specifies the set of fields
2222  * hashed, the hash function, and the hash key.  We need one of these for each
2223  * type of RSS traffic - v4 TCP, v6 TCP, v4 UDP, v6 UDP, other v4, other v6,
2224  * and one for non-RSS traffic.
2225  *
2226  * Flow tables hold flow table entries in sequence.  The first entry that
2227  * matches a packet is applied, sending the packet to either another flow
2228  * table or a TIR.  We use one flow table to select packets based on
2229  * destination MAC address, and a second to apply RSS.  The entries in the
2230  * first table send matching packets to the second, and the entries in the
2231  * RSS table send packets to RSS TIRs if possible, or the non-RSS TIR.
2232  *
2233  * The flow table entry that delivers packets to an RSS TIR must include match
2234  * criteria that ensure packets delivered to the TIR include all the fields
2235  * that the TIR hashes on - so for a v4 TCP TIR, the flow table entry must
2236  * only accept v4 TCP packets.  Accordingly, we need flow table entries for
2237  * each TIR.
2238  *
2239  * All of this is a lot more flexible than we need, and we can describe most
2240  * of the stuff we need with a simple array.
2241  *
2242  * An RSS config creates a TIR with hashing enabled on a set of fields,
2243  * pointing to either the first rx queue or the RQT containing all the rx
2244  * queues, and a flow table entry that matches on an ether type and
2245  * optionally an ip proto, that delivers packets to the TIR.
2246  */
2247 static struct mcx_rss_rule {
2248 	int			hash_sel;
2249 	int			flow_group;
2250 	int			ethertype;
2251 	int			ip_proto;
2252 } mcx_rss_config[] = {
2253 	/* udp and tcp for v4/v6 */
2254 	{ MCX_RSS_HASH_SEL_V4_TCP, MCX_FLOW_GROUP_RSS_L4,
2255 	  ETHERTYPE_IP, IPPROTO_TCP },
2256 	{ MCX_RSS_HASH_SEL_V6_TCP, MCX_FLOW_GROUP_RSS_L4,
2257 	  ETHERTYPE_IPV6, IPPROTO_TCP },
2258 	{ MCX_RSS_HASH_SEL_V4_UDP, MCX_FLOW_GROUP_RSS_L4,
2259 	  ETHERTYPE_IP, IPPROTO_UDP },
2260 	{ MCX_RSS_HASH_SEL_V6_UDP, MCX_FLOW_GROUP_RSS_L4,
2261 	  ETHERTYPE_IPV6, IPPROTO_UDP },
2262 
2263 	/* other v4/v6 */
2264 	{ MCX_RSS_HASH_SEL_V4, MCX_FLOW_GROUP_RSS_L3,
2265 	  ETHERTYPE_IP, 0 },
2266 	{ MCX_RSS_HASH_SEL_V6, MCX_FLOW_GROUP_RSS_L3,
2267 	  ETHERTYPE_IPV6, 0 },
2268 
2269 	/* non v4/v6 */
2270 	{ 0, MCX_FLOW_GROUP_RSS_NONE, 0, 0 }
2271 };
2272 
2273 struct mcx_softc {
2274 	struct device		 sc_dev;
2275 	struct arpcom		 sc_ac;
2276 	struct ifmedia		 sc_media;
2277 	uint64_t		 sc_media_status;
2278 	uint64_t		 sc_media_active;
2279 
2280 	pci_chipset_tag_t	 sc_pc;
2281 	pci_intr_handle_t	 sc_ih;
2282 	void			*sc_ihc;
2283 	pcitag_t		 sc_tag;
2284 
2285 	bus_dma_tag_t		 sc_dmat;
2286 	bus_space_tag_t		 sc_memt;
2287 	bus_space_handle_t	 sc_memh;
2288 	bus_size_t		 sc_mems;
2289 
2290 	struct mcx_dmamem	 sc_cmdq_mem;
2291 	unsigned int		 sc_cmdq_mask;
2292 	unsigned int		 sc_cmdq_size;
2293 
2294 	unsigned int		 sc_cmdq_token;
2295 
2296 	struct mcx_hwmem	 sc_boot_pages;
2297 	struct mcx_hwmem	 sc_init_pages;
2298 	struct mcx_hwmem	 sc_regular_pages;
2299 
2300 	int			 sc_uar;
2301 	int			 sc_pd;
2302 	int			 sc_tdomain;
2303 	uint32_t		 sc_lkey;
2304 	int			 sc_tis;
2305 	int			 sc_tir[nitems(mcx_rss_config)];
2306 	int			 sc_rqt;
2307 
2308 	struct mcx_dmamem	 sc_doorbell_mem;
2309 
2310 	struct mcx_eq		 sc_admin_eq;
2311 	struct mcx_eq		 sc_queue_eq;
2312 
2313 	int			 sc_hardmtu;
2314 	int			 sc_rxbufsz;
2315 
2316 	int			 sc_bf_size;
2317 	int			 sc_max_rqt_size;
2318 
2319 	struct task		 sc_port_change;
2320 
2321 	int			 sc_mac_flow_table_id;
2322 	int			 sc_rss_flow_table_id;
2323 	struct mcx_flow_group	 sc_flow_group[MCX_NUM_FLOW_GROUPS];
2324 	int			 sc_promisc_flow_enabled;
2325 	int			 sc_allmulti_flow_enabled;
2326 	int			 sc_mcast_flow_base;
2327 	int			 sc_extra_mcast;
2328 	uint8_t			 sc_mcast_flows[MCX_NUM_MCAST_FLOWS][ETHER_ADDR_LEN];
2329 
2330 	struct mcx_calibration	 sc_calibration[2];
2331 	unsigned int		 sc_calibration_gen;
2332 	struct timeout		 sc_calibrate;
2333 
2334 	struct mcx_queues	 sc_queues[MCX_MAX_QUEUES];
2335 	unsigned int		 sc_nqueues;
2336 
2337 #if NKSTAT > 0
2338 	struct kstat		*sc_kstat_ieee8023;
2339 	struct kstat		*sc_kstat_rfc2863;
2340 	struct kstat		*sc_kstat_rfc2819;
2341 	struct kstat		*sc_kstat_rfc3635;
2342 	unsigned int		 sc_kstat_mtmp_count;
2343 	struct kstat		**sc_kstat_mtmp;
2344 #endif
2345 };
2346 #define DEVNAME(_sc) ((_sc)->sc_dev.dv_xname)
2347 
2348 static int	mcx_match(struct device *, void *, void *);
2349 static void	mcx_attach(struct device *, struct device *, void *);
2350 
2351 #if NKSTAT > 0
2352 static void	mcx_kstat_attach(struct mcx_softc *);
2353 #endif
2354 
2355 static int	mcx_version(struct mcx_softc *);
2356 static int	mcx_init_wait(struct mcx_softc *);
2357 static int	mcx_enable_hca(struct mcx_softc *);
2358 static int	mcx_teardown_hca(struct mcx_softc *, uint16_t);
2359 static int	mcx_access_hca_reg(struct mcx_softc *, uint16_t, int, void *,
2360 		    int);
2361 static int	mcx_issi(struct mcx_softc *);
2362 static int	mcx_pages(struct mcx_softc *, struct mcx_hwmem *, uint16_t);
2363 static int	mcx_hca_max_caps(struct mcx_softc *);
2364 static int	mcx_hca_set_caps(struct mcx_softc *);
2365 static int	mcx_init_hca(struct mcx_softc *);
2366 static int	mcx_set_driver_version(struct mcx_softc *);
2367 static int	mcx_iff(struct mcx_softc *);
2368 static int	mcx_alloc_uar(struct mcx_softc *, int *);
2369 static int	mcx_alloc_pd(struct mcx_softc *);
2370 static int	mcx_alloc_tdomain(struct mcx_softc *);
2371 static int	mcx_create_eq(struct mcx_softc *, struct mcx_eq *, int,
2372 		    uint64_t, int);
2373 static int	mcx_query_nic_vport_context(struct mcx_softc *);
2374 static int	mcx_query_special_contexts(struct mcx_softc *);
2375 static int	mcx_set_port_mtu(struct mcx_softc *, int);
2376 static int	mcx_create_cq(struct mcx_softc *, struct mcx_cq *, int, int,
2377 		    int);
2378 static int	mcx_destroy_cq(struct mcx_softc *, struct mcx_cq *);
2379 static int	mcx_create_sq(struct mcx_softc *, struct mcx_tx *, int, int,
2380 		    int);
2381 static int	mcx_destroy_sq(struct mcx_softc *, struct mcx_tx *);
2382 static int	mcx_ready_sq(struct mcx_softc *, struct mcx_tx *);
2383 static int	mcx_create_rq(struct mcx_softc *, struct mcx_rx *, int, int);
2384 static int	mcx_destroy_rq(struct mcx_softc *, struct mcx_rx *);
2385 static int	mcx_ready_rq(struct mcx_softc *, struct mcx_rx *);
2386 static int	mcx_create_tir_direct(struct mcx_softc *, struct mcx_rx *,
2387 		    int *);
2388 static int	mcx_create_tir_indirect(struct mcx_softc *, int, uint32_t,
2389 		    int *);
2390 static int	mcx_destroy_tir(struct mcx_softc *, int);
2391 static int	mcx_create_tis(struct mcx_softc *, int *);
2392 static int	mcx_destroy_tis(struct mcx_softc *, int);
2393 static int	mcx_create_rqt(struct mcx_softc *, int, int *, int *);
2394 static int	mcx_destroy_rqt(struct mcx_softc *, int);
2395 static int	mcx_create_flow_table(struct mcx_softc *, int, int, int *);
2396 static int	mcx_set_flow_table_root(struct mcx_softc *, int);
2397 static int	mcx_destroy_flow_table(struct mcx_softc *, int);
2398 static int	mcx_create_flow_group(struct mcx_softc *, int, int, int,
2399 		    int, int, struct mcx_flow_match *);
2400 static int	mcx_destroy_flow_group(struct mcx_softc *, int);
2401 static int	mcx_set_flow_table_entry_mac(struct mcx_softc *, int, int,
2402 		    uint8_t *, uint32_t);
2403 static int	mcx_set_flow_table_entry_proto(struct mcx_softc *, int, int,
2404 		    int, int, uint32_t);
2405 static int	mcx_delete_flow_table_entry(struct mcx_softc *, int, int);
2406 
2407 #if 0
2408 static int	mcx_dump_flow_table(struct mcx_softc *, int);
2409 static int	mcx_dump_flow_table_entry(struct mcx_softc *, int, int);
2410 static int	mcx_dump_flow_group(struct mcx_softc *, int);
2411 static int	mcx_dump_rq(struct mcx_softc *);
2412 static int	mcx_dump_sq(struct mcx_softc *);
2413 #endif
2414 
2415 
2416 /*
2417 static void	mcx_cmdq_dump(const struct mcx_cmdq_entry *);
2418 static void	mcx_cmdq_mbox_dump(struct mcx_dmamem *, int);
2419 */
2420 static void	mcx_refill(void *);
2421 static int	mcx_process_rx(struct mcx_softc *, struct mcx_rx *,
2422 		    struct mcx_cq_entry *, struct mbuf_list *,
2423 		    const struct mcx_calibration *);
2424 static void	mcx_process_txeof(struct mcx_softc *, struct mcx_tx *,
2425 		    struct mcx_cq_entry *, int *);
2426 static void	mcx_process_cq(struct mcx_softc *, struct mcx_queues *,
2427 		    struct mcx_cq *);
2428 
2429 static void	mcx_arm_cq(struct mcx_softc *, struct mcx_cq *, int);
2430 static void	mcx_arm_eq(struct mcx_softc *, struct mcx_eq *, int);
2431 static int	mcx_admin_intr(void *);
2432 static int	mcx_cq_intr(void *);
2433 
2434 static int	mcx_up(struct mcx_softc *);
2435 static void	mcx_down(struct mcx_softc *);
2436 static int	mcx_ioctl(struct ifnet *, u_long, caddr_t);
2437 static int	mcx_rxrinfo(struct mcx_softc *, struct if_rxrinfo *);
2438 static void	mcx_start(struct ifqueue *);
2439 static void	mcx_watchdog(struct ifnet *);
2440 static void	mcx_media_add_types(struct mcx_softc *);
2441 static void	mcx_media_status(struct ifnet *, struct ifmediareq *);
2442 static int	mcx_media_change(struct ifnet *);
2443 static int	mcx_get_sffpage(struct ifnet *, struct if_sffpage *);
2444 static void	mcx_port_change(void *);
2445 
2446 static void	mcx_calibrate_first(struct mcx_softc *);
2447 static void	mcx_calibrate(void *);
2448 
2449 static inline uint32_t
2450 		mcx_rd(struct mcx_softc *, bus_size_t);
2451 static inline void
2452 		mcx_wr(struct mcx_softc *, bus_size_t, uint32_t);
2453 static inline void
2454 		mcx_bar(struct mcx_softc *, bus_size_t, bus_size_t, int);
2455 
2456 static uint64_t	mcx_timer(struct mcx_softc *);
2457 
2458 static int	mcx_dmamem_alloc(struct mcx_softc *, struct mcx_dmamem *,
2459 		    bus_size_t, u_int align);
2460 static void	mcx_dmamem_zero(struct mcx_dmamem *);
2461 static void	mcx_dmamem_free(struct mcx_softc *, struct mcx_dmamem *);
2462 
2463 static int	mcx_hwmem_alloc(struct mcx_softc *, struct mcx_hwmem *,
2464 		    unsigned int);
2465 static void	mcx_hwmem_free(struct mcx_softc *, struct mcx_hwmem *);
2466 
2467 struct cfdriver mcx_cd = {
2468 	NULL,
2469 	"mcx",
2470 	DV_IFNET,
2471 };
2472 
2473 struct cfattach mcx_ca = {
2474 	sizeof(struct mcx_softc),
2475 	mcx_match,
2476 	mcx_attach,
2477 };
2478 
2479 static const struct pci_matchid mcx_devices[] = {
2480 	{ PCI_VENDOR_MELLANOX,	PCI_PRODUCT_MELLANOX_MT27700 },
2481 	{ PCI_VENDOR_MELLANOX,	PCI_PRODUCT_MELLANOX_MT27700VF },
2482 	{ PCI_VENDOR_MELLANOX,	PCI_PRODUCT_MELLANOX_MT27710 },
2483 	{ PCI_VENDOR_MELLANOX,	PCI_PRODUCT_MELLANOX_MT27710VF },
2484 	{ PCI_VENDOR_MELLANOX,	PCI_PRODUCT_MELLANOX_MT27800 },
2485 	{ PCI_VENDOR_MELLANOX,	PCI_PRODUCT_MELLANOX_MT27800VF },
2486 	{ PCI_VENDOR_MELLANOX,	PCI_PRODUCT_MELLANOX_MT28800 },
2487 	{ PCI_VENDOR_MELLANOX,	PCI_PRODUCT_MELLANOX_MT28800VF },
2488 	{ PCI_VENDOR_MELLANOX,	PCI_PRODUCT_MELLANOX_MT2892  },
2489 };
2490 
2491 struct mcx_eth_proto_capability {
2492 	uint64_t	cap_media;
2493 	uint64_t	cap_baudrate;
2494 };
2495 
2496 static const struct mcx_eth_proto_capability mcx_eth_cap_map[] = {
2497 	[MCX_ETHER_CAP_SGMII]		= { IFM_1000_SGMII,	IF_Gbps(1) },
2498 	[MCX_ETHER_CAP_1000_KX]		= { IFM_1000_KX,	IF_Gbps(1) },
2499 	[MCX_ETHER_CAP_10G_CX4]		= { IFM_10G_CX4,	IF_Gbps(10) },
2500 	[MCX_ETHER_CAP_10G_KX4]		= { IFM_10G_KX4,	IF_Gbps(10) },
2501 	[MCX_ETHER_CAP_10G_KR]		= { IFM_10G_KR,		IF_Gbps(10) },
2502 	[MCX_ETHER_CAP_40G_CR4]		= { IFM_40G_CR4,	IF_Gbps(40) },
2503 	[MCX_ETHER_CAP_40G_KR4]		= { IFM_40G_KR4,	IF_Gbps(40) },
2504 	[MCX_ETHER_CAP_10G_CR]		= { IFM_10G_SFP_CU,	IF_Gbps(10) },
2505 	[MCX_ETHER_CAP_10G_SR]		= { IFM_10G_SR,		IF_Gbps(10) },
2506 	[MCX_ETHER_CAP_10G_LR]		= { IFM_10G_LR,		IF_Gbps(10) },
2507 	[MCX_ETHER_CAP_40G_SR4]		= { IFM_40G_SR4,	IF_Gbps(40) },
2508 	[MCX_ETHER_CAP_40G_LR4]		= { IFM_40G_LR4,	IF_Gbps(40) },
2509 	[MCX_ETHER_CAP_50G_SR2]		= { 0 /*IFM_50G_SR2*/,	IF_Gbps(50) },
2510 	[MCX_ETHER_CAP_100G_CR4]	= { IFM_100G_CR4,	IF_Gbps(100) },
2511 	[MCX_ETHER_CAP_100G_SR4]	= { IFM_100G_SR4,	IF_Gbps(100) },
2512 	[MCX_ETHER_CAP_100G_KR4]	= { IFM_100G_KR4,	IF_Gbps(100) },
2513 	[MCX_ETHER_CAP_25G_CR]		= { IFM_25G_CR,		IF_Gbps(25) },
2514 	[MCX_ETHER_CAP_25G_KR]		= { IFM_25G_KR,		IF_Gbps(25) },
2515 	[MCX_ETHER_CAP_25G_SR]		= { IFM_25G_SR,		IF_Gbps(25) },
2516 	[MCX_ETHER_CAP_50G_CR2]		= { IFM_50G_CR2,	IF_Gbps(50) },
2517 	[MCX_ETHER_CAP_50G_KR2]		= { IFM_50G_KR2,	IF_Gbps(50) },
2518 };
2519 
2520 static int
2521 mcx_get_id(uint32_t val)
2522 {
2523 	return betoh32(val) & 0x00ffffff;
2524 }
2525 
2526 static int
2527 mcx_match(struct device *parent, void *match, void *aux)
2528 {
2529 	return (pci_matchbyid(aux, mcx_devices, nitems(mcx_devices)));
2530 }
2531 
2532 void
2533 mcx_attach(struct device *parent, struct device *self, void *aux)
2534 {
2535 	struct mcx_softc *sc = (struct mcx_softc *)self;
2536 	struct ifnet *ifp = &sc->sc_ac.ac_if;
2537 	struct pci_attach_args *pa = aux;
2538 	pcireg_t memtype;
2539 	uint32_t r;
2540 	unsigned int cq_stride;
2541 	unsigned int cq_size;
2542 	const char *intrstr;
2543 	int i, msix;
2544 
2545 	sc->sc_pc = pa->pa_pc;
2546 	sc->sc_tag = pa->pa_tag;
2547 	sc->sc_dmat = pa->pa_dmat;
2548 
2549 	/* Map the PCI memory space */
2550 	memtype = pci_mapreg_type(sc->sc_pc, sc->sc_tag, MCX_HCA_BAR);
2551 	if (pci_mapreg_map(pa, MCX_HCA_BAR, memtype,
2552 	    BUS_SPACE_MAP_PREFETCHABLE, &sc->sc_memt, &sc->sc_memh,
2553 	    NULL, &sc->sc_mems, 0)) {
2554 		printf(": unable to map register memory\n");
2555 		return;
2556 	}
2557 
2558 	if (mcx_version(sc) != 0) {
2559 		/* error printed by mcx_version */
2560 		goto unmap;
2561 	}
2562 
2563 	r = mcx_rd(sc, MCX_CMDQ_ADDR_LO);
2564 	cq_stride = 1 << MCX_CMDQ_LOG_STRIDE(r); /* size of the entries */
2565 	cq_size = 1 << MCX_CMDQ_LOG_SIZE(r); /* number of entries */
2566 	if (cq_size > MCX_MAX_CQE) {
2567 		printf(", command queue size overflow %u\n", cq_size);
2568 		goto unmap;
2569 	}
2570 	if (cq_stride < sizeof(struct mcx_cmdq_entry)) {
2571 		printf(", command queue entry size underflow %u\n", cq_stride);
2572 		goto unmap;
2573 	}
2574 	if (cq_stride * cq_size > MCX_PAGE_SIZE) {
2575 		printf(", command queue page overflow\n");
2576 		goto unmap;
2577 	}
2578 
2579 	if (mcx_dmamem_alloc(sc, &sc->sc_doorbell_mem, MCX_DOORBELL_AREA_SIZE,
2580 	    MCX_PAGE_SIZE) != 0) {
2581 		printf(", unable to allocate doorbell memory\n");
2582 		goto unmap;
2583 	}
2584 
2585 	if (mcx_dmamem_alloc(sc, &sc->sc_cmdq_mem, MCX_PAGE_SIZE,
2586 	    MCX_PAGE_SIZE) != 0) {
2587 		printf(", unable to allocate command queue\n");
2588 		goto dbfree;
2589 	}
2590 
2591 	mcx_wr(sc, MCX_CMDQ_ADDR_HI, MCX_DMA_DVA(&sc->sc_cmdq_mem) >> 32);
2592 	mcx_bar(sc, MCX_CMDQ_ADDR_HI, sizeof(uint32_t),
2593 	    BUS_SPACE_BARRIER_WRITE);
2594 	mcx_wr(sc, MCX_CMDQ_ADDR_LO, MCX_DMA_DVA(&sc->sc_cmdq_mem));
2595 	mcx_bar(sc, MCX_CMDQ_ADDR_LO, sizeof(uint32_t),
2596 	    BUS_SPACE_BARRIER_WRITE);
2597 
2598 	if (mcx_init_wait(sc) != 0) {
2599 		printf(", timeout waiting for init\n");
2600 		goto cqfree;
2601 	}
2602 
2603 	sc->sc_cmdq_mask = cq_size - 1;
2604 	sc->sc_cmdq_size = cq_stride;
2605 
2606 	if (mcx_enable_hca(sc) != 0) {
2607 		/* error printed by mcx_enable_hca */
2608 		goto cqfree;
2609 	}
2610 
2611 	if (mcx_issi(sc) != 0) {
2612 		/* error printed by mcx_issi */
2613 		goto teardown;
2614 	}
2615 
2616 	if (mcx_pages(sc, &sc->sc_boot_pages,
2617 	    htobe16(MCX_CMD_QUERY_PAGES_BOOT)) != 0) {
2618 		/* error printed by mcx_pages */
2619 		goto teardown;
2620 	}
2621 
2622 	if (mcx_hca_max_caps(sc) != 0) {
2623 		/* error printed by mcx_hca_max_caps */
2624 		goto teardown;
2625 	}
2626 
2627 	if (mcx_hca_set_caps(sc) != 0) {
2628 		/* error printed by mcx_hca_set_caps */
2629 		goto teardown;
2630 	}
2631 
2632 	if (mcx_pages(sc, &sc->sc_init_pages,
2633 	    htobe16(MCX_CMD_QUERY_PAGES_INIT)) != 0) {
2634 		/* error printed by mcx_pages */
2635 		goto teardown;
2636 	}
2637 
2638 	if (mcx_init_hca(sc) != 0) {
2639 		/* error printed by mcx_init_hca */
2640 		goto teardown;
2641 	}
2642 
2643 	if (mcx_pages(sc, &sc->sc_regular_pages,
2644 	    htobe16(MCX_CMD_QUERY_PAGES_REGULAR)) != 0) {
2645 		/* error printed by mcx_pages */
2646 		goto teardown;
2647 	}
2648 
2649 	/* apparently not necessary? */
2650 	if (mcx_set_driver_version(sc) != 0) {
2651 		/* error printed by mcx_set_driver_version */
2652 		goto teardown;
2653 	}
2654 
2655 	if (mcx_iff(sc) != 0) {	/* modify nic vport context */
2656 		/* error printed by mcx_iff? */
2657 		goto teardown;
2658 	}
2659 
2660 	if (mcx_alloc_uar(sc, &sc->sc_uar) != 0) {
2661 		/* error printed by mcx_alloc_uar */
2662 		goto teardown;
2663 	}
2664 
2665 	if (mcx_alloc_pd(sc) != 0) {
2666 		/* error printed by mcx_alloc_pd */
2667 		goto teardown;
2668 	}
2669 
2670 	if (mcx_alloc_tdomain(sc) != 0) {
2671 		/* error printed by mcx_alloc_tdomain */
2672 		goto teardown;
2673 	}
2674 
2675 	/*
2676 	 * PRM makes no mention of msi interrupts, just legacy and msi-x.
2677 	 * mellanox support tells me legacy interrupts are not supported,
2678 	 * so we're stuck with just msi-x.
2679 	 */
2680 	if (pci_intr_map_msix(pa, 0, &sc->sc_ih) != 0) {
2681 		printf(": unable to map interrupt\n");
2682 		goto teardown;
2683 	}
2684 	intrstr = pci_intr_string(sc->sc_pc, sc->sc_ih);
2685 	sc->sc_ihc = pci_intr_establish(sc->sc_pc, sc->sc_ih,
2686 	    IPL_NET | IPL_MPSAFE, mcx_admin_intr, sc, DEVNAME(sc));
2687 	if (sc->sc_ihc == NULL) {
2688 		printf(": unable to establish interrupt");
2689 		if (intrstr != NULL)
2690 			printf(" at %s", intrstr);
2691 		printf("\n");
2692 		goto teardown;
2693 	}
2694 
2695 	if (mcx_create_eq(sc, &sc->sc_admin_eq, sc->sc_uar,
2696 	    (1ull << MCX_EVENT_TYPE_INTERNAL_ERROR) |
2697 	    (1ull << MCX_EVENT_TYPE_PORT_CHANGE) |
2698 	    (1ull << MCX_EVENT_TYPE_CMD_COMPLETION) |
2699 	    (1ull << MCX_EVENT_TYPE_PAGE_REQUEST), 0) != 0) {
2700 		/* error printed by mcx_create_eq */
2701 		goto teardown;
2702 	}
2703 
2704 	if (mcx_query_nic_vport_context(sc) != 0) {
2705 		/* error printed by mcx_query_nic_vport_context */
2706 		goto teardown;
2707 	}
2708 
2709 	if (mcx_query_special_contexts(sc) != 0) {
2710 		/* error printed by mcx_query_special_contexts */
2711 		goto teardown;
2712 	}
2713 
2714 	if (mcx_set_port_mtu(sc, MCX_HARDMTU) != 0) {
2715 		/* error printed by mcx_set_port_mtu */
2716 		goto teardown;
2717 	}
2718 
2719 	printf(", %s, address %s\n", intrstr,
2720 	    ether_sprintf(sc->sc_ac.ac_enaddr));
2721 
2722 	msix = pci_intr_msix_count(pa->pa_pc, pa->pa_tag);
2723 	sc->sc_nqueues = 1;
2724 
2725 	strlcpy(ifp->if_xname, DEVNAME(sc), IFNAMSIZ);
2726 	ifp->if_softc = sc;
2727 	ifp->if_flags = IFF_BROADCAST | IFF_MULTICAST | IFF_SIMPLEX;
2728 	ifp->if_xflags = IFXF_MPSAFE;
2729 	ifp->if_ioctl = mcx_ioctl;
2730 	ifp->if_qstart = mcx_start;
2731 	ifp->if_watchdog = mcx_watchdog;
2732 	ifp->if_hardmtu = sc->sc_hardmtu;
2733 	ifp->if_capabilities = IFCAP_VLAN_MTU | IFCAP_CSUM_IPv4 |
2734 	    IFCAP_CSUM_UDPv4 | IFCAP_CSUM_UDPv6 | IFCAP_CSUM_TCPv4 |
2735 	    IFCAP_CSUM_TCPv6;
2736 #if NVLAN > 0
2737 	ifp->if_capabilities |= IFCAP_VLAN_HWTAGGING;
2738 #endif
2739 	ifq_set_maxlen(&ifp->if_snd, 1024);
2740 
2741 	ifmedia_init(&sc->sc_media, IFM_IMASK, mcx_media_change,
2742 	    mcx_media_status);
2743 	mcx_media_add_types(sc);
2744 	ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL);
2745 	ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO);
2746 
2747 	if_attach(ifp);
2748 	ether_ifattach(ifp);
2749 
2750 	if_attach_iqueues(ifp, sc->sc_nqueues);
2751 	if_attach_queues(ifp, sc->sc_nqueues);
2752 	for (i = 0; i < sc->sc_nqueues; i++) {
2753 		struct ifiqueue *ifiq = ifp->if_iqs[i];
2754 		struct ifqueue *ifq = ifp->if_ifqs[i];
2755 		struct mcx_queues *q = &sc->sc_queues[i];
2756 		struct mcx_rx *rx = &q->q_rx;
2757 		struct mcx_tx *tx = &q->q_tx;
2758 		pci_intr_handle_t ih;
2759 		int vec;
2760 
2761 		vec = i + 1;
2762 		q->q_sc = sc;
2763 		q->q_index = i;
2764 
2765 		if (mcx_alloc_uar(sc, &q->q_uar) != 0) {
2766 			printf("%s: unable to alloc uar %d\n",
2767 			    DEVNAME(sc), i);
2768 			goto teardown;
2769 		}
2770 
2771 		if (mcx_create_eq(sc, &q->q_eq, q->q_uar, 0, vec) != 0) {
2772 			printf("%s: unable to create event queue %d\n",
2773 			    DEVNAME(sc), i);
2774 			goto teardown;
2775 		}
2776 
2777 		rx->rx_softc = sc;
2778 		rx->rx_ifiq = ifiq;
2779 		timeout_set(&rx->rx_refill, mcx_refill, rx);
2780 		ifiq->ifiq_softc = rx;
2781 
2782 		tx->tx_softc = sc;
2783 		tx->tx_ifq = ifq;
2784 		ifq->ifq_softc = tx;
2785 
2786 		if (pci_intr_map_msix(pa, i + 1, &ih) != 0) {
2787 			printf("%s: unable to map queue interrupt %d\n",
2788 			    DEVNAME(sc), i);
2789 			goto teardown;
2790 		}
2791 		snprintf(q->q_name, sizeof(q->q_name), "%s:%d",
2792 		    DEVNAME(sc), i);
2793 		q->q_ihc = pci_intr_establish(sc->sc_pc, ih,
2794 		    IPL_NET | IPL_MPSAFE, mcx_cq_intr, q, q->q_name);
2795 	}
2796 
2797 	timeout_set(&sc->sc_calibrate, mcx_calibrate, sc);
2798 
2799 	task_set(&sc->sc_port_change, mcx_port_change, sc);
2800 	mcx_port_change(sc);
2801 
2802 	sc->sc_mac_flow_table_id = -1;
2803 	sc->sc_rss_flow_table_id = -1;
2804 	sc->sc_rqt = -1;
2805 	for (i = 0; i < MCX_NUM_FLOW_GROUPS; i++) {
2806 		struct mcx_flow_group *mfg = &sc->sc_flow_group[i];
2807 		mfg->g_id = -1;
2808 		mfg->g_table = -1;
2809 		mfg->g_size = 0;
2810 		mfg->g_start = 0;
2811 	}
2812 	sc->sc_extra_mcast = 0;
2813 	memset(sc->sc_mcast_flows, 0, sizeof(sc->sc_mcast_flows));
2814 
2815 #if NKSTAT > 0
2816 	mcx_kstat_attach(sc);
2817 #endif
2818 	return;
2819 
2820 teardown:
2821 	mcx_teardown_hca(sc, htobe16(MCX_CMD_TEARDOWN_HCA_GRACEFUL));
2822 	/* error printed by mcx_teardown_hca, and we're already unwinding */
2823 cqfree:
2824 	mcx_wr(sc, MCX_CMDQ_ADDR_HI, MCX_DMA_DVA(&sc->sc_cmdq_mem) >> 32);
2825 	mcx_bar(sc, MCX_CMDQ_ADDR_HI, sizeof(uint64_t),
2826 	    BUS_SPACE_BARRIER_WRITE);
2827 	mcx_wr(sc, MCX_CMDQ_ADDR_LO, MCX_DMA_DVA(&sc->sc_cmdq_mem) |
2828 	    MCX_CMDQ_INTERFACE_DISABLED);
2829 	mcx_bar(sc, MCX_CMDQ_ADDR_LO, sizeof(uint64_t),
2830 	    BUS_SPACE_BARRIER_WRITE);
2831 
2832 	mcx_wr(sc, MCX_CMDQ_ADDR_HI, 0);
2833 	mcx_bar(sc, MCX_CMDQ_ADDR_HI, sizeof(uint64_t),
2834 	    BUS_SPACE_BARRIER_WRITE);
2835 	mcx_wr(sc, MCX_CMDQ_ADDR_LO, MCX_CMDQ_INTERFACE_DISABLED);
2836 
2837 	mcx_dmamem_free(sc, &sc->sc_cmdq_mem);
2838 dbfree:
2839 	mcx_dmamem_free(sc, &sc->sc_doorbell_mem);
2840 unmap:
2841 	bus_space_unmap(sc->sc_memt, sc->sc_memh, sc->sc_mems);
2842 	sc->sc_mems = 0;
2843 }
2844 
2845 static int
2846 mcx_version(struct mcx_softc *sc)
2847 {
2848 	uint32_t fw0, fw1;
2849 	uint16_t cmdif;
2850 
2851 	fw0 = mcx_rd(sc, MCX_FW_VER);
2852 	fw1 = mcx_rd(sc, MCX_CMDIF_FW_SUBVER);
2853 
2854 	printf(": FW %u.%u.%04u", MCX_FW_VER_MAJOR(fw0),
2855 	    MCX_FW_VER_MINOR(fw0), MCX_FW_VER_SUBMINOR(fw1));
2856 
2857 	cmdif = MCX_CMDIF(fw1);
2858 	if (cmdif != MCX_CMD_IF_SUPPORTED) {
2859 		printf(", unsupported command interface %u\n", cmdif);
2860 		return (-1);
2861 	}
2862 
2863 	return (0);
2864 }
2865 
2866 static int
2867 mcx_init_wait(struct mcx_softc *sc)
2868 {
2869 	unsigned int i;
2870 	uint32_t r;
2871 
2872 	for (i = 0; i < 2000; i++) {
2873 		r = mcx_rd(sc, MCX_STATE);
2874 		if ((r & MCX_STATE_MASK) == MCX_STATE_READY)
2875 			return (0);
2876 
2877 		delay(1000);
2878 		mcx_bar(sc, MCX_STATE, sizeof(uint32_t),
2879 		    BUS_SPACE_BARRIER_READ);
2880 	}
2881 
2882 	return (-1);
2883 }
2884 
2885 static uint8_t
2886 mcx_cmdq_poll(struct mcx_softc *sc, struct mcx_cmdq_entry *cqe,
2887     unsigned int msec)
2888 {
2889 	unsigned int i;
2890 
2891 	for (i = 0; i < msec; i++) {
2892 		bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_cmdq_mem),
2893 		    0, MCX_DMA_LEN(&sc->sc_cmdq_mem), BUS_DMASYNC_POSTRW);
2894 
2895 		if ((cqe->cq_status & MCX_CQ_STATUS_OWN_MASK) ==
2896 		    MCX_CQ_STATUS_OWN_SW)
2897 			return (0);
2898 
2899 		delay(1000);
2900 	}
2901 
2902 	return (ETIMEDOUT);
2903 }
2904 
2905 static uint32_t
2906 mcx_mix_u64(uint32_t xor, uint64_t u64)
2907 {
2908 	xor ^= u64 >> 32;
2909 	xor ^= u64;
2910 
2911 	return (xor);
2912 }
2913 
2914 static uint32_t
2915 mcx_mix_u32(uint32_t xor, uint32_t u32)
2916 {
2917 	xor ^= u32;
2918 
2919 	return (xor);
2920 }
2921 
2922 static uint32_t
2923 mcx_mix_u8(uint32_t xor, uint8_t u8)
2924 {
2925 	xor ^= u8;
2926 
2927 	return (xor);
2928 }
2929 
2930 static uint8_t
2931 mcx_mix_done(uint32_t xor)
2932 {
2933 	xor ^= xor >> 16;
2934 	xor ^= xor >> 8;
2935 
2936 	return (xor);
2937 }
2938 
2939 static uint8_t
2940 mcx_xor(const void *buf, size_t len)
2941 {
2942 	const uint32_t *dwords = buf;
2943 	uint32_t xor = 0xff;
2944 	size_t i;
2945 
2946 	len /= sizeof(*dwords);
2947 
2948 	for (i = 0; i < len; i++)
2949 		xor ^= dwords[i];
2950 
2951 	return (mcx_mix_done(xor));
2952 }
2953 
2954 static uint8_t
2955 mcx_cmdq_token(struct mcx_softc *sc)
2956 {
2957 	uint8_t token;
2958 
2959 	do {
2960 		token = ++sc->sc_cmdq_token;
2961 	} while (token == 0);
2962 
2963 	return (token);
2964 }
2965 
2966 static void
2967 mcx_cmdq_init(struct mcx_softc *sc, struct mcx_cmdq_entry *cqe,
2968     uint32_t ilen, uint32_t olen, uint8_t token)
2969 {
2970 	memset(cqe, 0, sc->sc_cmdq_size);
2971 
2972 	cqe->cq_type = MCX_CMDQ_TYPE_PCIE;
2973 	htobem32(&cqe->cq_input_length, ilen);
2974 	htobem32(&cqe->cq_output_length, olen);
2975 	cqe->cq_token = token;
2976 	cqe->cq_status = MCX_CQ_STATUS_OWN_HW;
2977 }
2978 
2979 static void
2980 mcx_cmdq_sign(struct mcx_cmdq_entry *cqe)
2981 {
2982 	cqe->cq_signature = ~mcx_xor(cqe, sizeof(*cqe));
2983 }
2984 
2985 static int
2986 mcx_cmdq_verify(const struct mcx_cmdq_entry *cqe)
2987 {
2988 	/* return (mcx_xor(cqe, sizeof(*cqe)) ? -1 :  0); */
2989 	return (0);
2990 }
2991 
2992 static void *
2993 mcx_cmdq_in(struct mcx_cmdq_entry *cqe)
2994 {
2995 	return (&cqe->cq_input_data);
2996 }
2997 
2998 static void *
2999 mcx_cmdq_out(struct mcx_cmdq_entry *cqe)
3000 {
3001 	return (&cqe->cq_output_data);
3002 }
3003 
3004 static void
3005 mcx_cmdq_post(struct mcx_softc *sc, struct mcx_cmdq_entry *cqe,
3006     unsigned int slot)
3007 {
3008 	mcx_cmdq_sign(cqe);
3009 
3010 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_cmdq_mem),
3011 	    0, MCX_DMA_LEN(&sc->sc_cmdq_mem), BUS_DMASYNC_PRERW);
3012 
3013 	mcx_wr(sc, MCX_CMDQ_DOORBELL, 1U << slot);
3014 }
3015 
3016 static int
3017 mcx_enable_hca(struct mcx_softc *sc)
3018 {
3019 	struct mcx_cmdq_entry *cqe;
3020 	struct mcx_cmd_enable_hca_in *in;
3021 	struct mcx_cmd_enable_hca_out *out;
3022 	int error;
3023 	uint8_t status;
3024 
3025 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
3026 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
3027 
3028 	in = mcx_cmdq_in(cqe);
3029 	in->cmd_opcode = htobe16(MCX_CMD_ENABLE_HCA);
3030 	in->cmd_op_mod = htobe16(0);
3031 	in->cmd_function_id = htobe16(0);
3032 
3033 	mcx_cmdq_post(sc, cqe, 0);
3034 
3035 	error = mcx_cmdq_poll(sc, cqe, 1000);
3036 	if (error != 0) {
3037 		printf(", hca enable timeout\n");
3038 		return (-1);
3039 	}
3040 	if (mcx_cmdq_verify(cqe) != 0) {
3041 		printf(", hca enable command corrupt\n");
3042 		return (-1);
3043 	}
3044 
3045 	status = cqe->cq_output_data[0];
3046 	if (status != MCX_CQ_STATUS_OK) {
3047 		printf(", hca enable failed (%x)\n", status);
3048 		return (-1);
3049 	}
3050 
3051 	return (0);
3052 }
3053 
3054 static int
3055 mcx_teardown_hca(struct mcx_softc *sc, uint16_t profile)
3056 {
3057 	struct mcx_cmdq_entry *cqe;
3058 	struct mcx_cmd_teardown_hca_in *in;
3059 	struct mcx_cmd_teardown_hca_out *out;
3060 	int error;
3061 	uint8_t status;
3062 
3063 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
3064 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
3065 
3066 	in = mcx_cmdq_in(cqe);
3067 	in->cmd_opcode = htobe16(MCX_CMD_TEARDOWN_HCA);
3068 	in->cmd_op_mod = htobe16(0);
3069 	in->cmd_profile = profile;
3070 
3071 	mcx_cmdq_post(sc, cqe, 0);
3072 
3073 	error = mcx_cmdq_poll(sc, cqe, 1000);
3074 	if (error != 0) {
3075 		printf(", hca teardown timeout\n");
3076 		return (-1);
3077 	}
3078 	if (mcx_cmdq_verify(cqe) != 0) {
3079 		printf(", hca teardown command corrupt\n");
3080 		return (-1);
3081 	}
3082 
3083 	status = cqe->cq_output_data[0];
3084 	if (status != MCX_CQ_STATUS_OK) {
3085 		printf(", hca teardown failed (%x)\n", status);
3086 		return (-1);
3087 	}
3088 
3089 	return (0);
3090 }
3091 
3092 static int
3093 mcx_cmdq_mboxes_alloc(struct mcx_softc *sc, struct mcx_dmamem *mxm,
3094     unsigned int nmb, uint64_t *ptr, uint8_t token)
3095 {
3096 	caddr_t kva;
3097 	uint64_t dva;
3098 	int i;
3099 	int error;
3100 
3101 	error = mcx_dmamem_alloc(sc, mxm,
3102 	    nmb * MCX_CMDQ_MAILBOX_SIZE, MCX_CMDQ_MAILBOX_ALIGN);
3103 	if (error != 0)
3104 		return (error);
3105 
3106 	mcx_dmamem_zero(mxm);
3107 
3108 	dva = MCX_DMA_DVA(mxm);
3109 	kva = MCX_DMA_KVA(mxm);
3110 	for (i = 0; i < nmb; i++) {
3111 		struct mcx_cmdq_mailbox *mbox = (struct mcx_cmdq_mailbox *)kva;
3112 
3113 		/* patch the cqe or mbox pointing at this one */
3114 		htobem64(ptr, dva);
3115 
3116 		/* fill in this mbox */
3117 		htobem32(&mbox->mb_block_number, i);
3118 		mbox->mb_token = token;
3119 
3120 		/* move to the next one */
3121 		ptr = &mbox->mb_next_ptr;
3122 
3123 		dva += MCX_CMDQ_MAILBOX_SIZE;
3124 		kva += MCX_CMDQ_MAILBOX_SIZE;
3125 	}
3126 
3127 	return (0);
3128 }
3129 
3130 static uint32_t
3131 mcx_cmdq_mbox_ctrl_sig(const struct mcx_cmdq_mailbox *mb)
3132 {
3133 	uint32_t xor = 0xff;
3134 
3135 	/* only 3 fields get set, so mix them directly */
3136 	xor = mcx_mix_u64(xor, mb->mb_next_ptr);
3137 	xor = mcx_mix_u32(xor, mb->mb_block_number);
3138 	xor = mcx_mix_u8(xor, mb->mb_token);
3139 
3140 	return (mcx_mix_done(xor));
3141 }
3142 
3143 static void
3144 mcx_cmdq_mboxes_sign(struct mcx_dmamem *mxm, unsigned int nmb)
3145 {
3146 	caddr_t kva;
3147 	int i;
3148 
3149 	kva = MCX_DMA_KVA(mxm);
3150 
3151 	for (i = 0; i < nmb; i++) {
3152 		struct mcx_cmdq_mailbox *mb = (struct mcx_cmdq_mailbox *)kva;
3153 		uint8_t sig = mcx_cmdq_mbox_ctrl_sig(mb);
3154 		mb->mb_ctrl_signature = sig;
3155 		mb->mb_signature = sig ^
3156 		    mcx_xor(mb->mb_data, sizeof(mb->mb_data));
3157 
3158 		kva += MCX_CMDQ_MAILBOX_SIZE;
3159 	}
3160 }
3161 
3162 static void
3163 mcx_cmdq_mboxes_sync(struct mcx_softc *sc, struct mcx_dmamem *mxm, int ops)
3164 {
3165 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(mxm),
3166 	    0, MCX_DMA_LEN(mxm), ops);
3167 }
3168 
3169 static struct mcx_cmdq_mailbox *
3170 mcx_cq_mbox(struct mcx_dmamem *mxm, unsigned int i)
3171 {
3172 	caddr_t kva;
3173 
3174 	kva = MCX_DMA_KVA(mxm);
3175 	kva += i * MCX_CMDQ_MAILBOX_SIZE;
3176 
3177 	return ((struct mcx_cmdq_mailbox *)kva);
3178 }
3179 
3180 static inline void *
3181 mcx_cq_mbox_data(struct mcx_cmdq_mailbox *mb)
3182 {
3183 	return (&mb->mb_data);
3184 }
3185 
3186 static void
3187 mcx_cmdq_mboxes_copyin(struct mcx_dmamem *mxm, unsigned int nmb,
3188     void *b, size_t len)
3189 {
3190 	caddr_t buf = b;
3191 	struct mcx_cmdq_mailbox *mb;
3192 	int i;
3193 
3194 	mb = (struct mcx_cmdq_mailbox *)MCX_DMA_KVA(mxm);
3195 	for (i = 0; i < nmb; i++) {
3196 
3197 		memcpy(mb->mb_data, buf, min(sizeof(mb->mb_data), len));
3198 
3199 		if (sizeof(mb->mb_data) >= len)
3200 			break;
3201 
3202 		buf += sizeof(mb->mb_data);
3203 		len -= sizeof(mb->mb_data);
3204 		mb++;
3205 	}
3206 }
3207 
3208 static void
3209 mcx_cmdq_mboxes_pas(struct mcx_dmamem *mxm, int offset, int npages,
3210     struct mcx_dmamem *buf)
3211 {
3212 	uint64_t *pas;
3213 	int mbox, mbox_pages, i;
3214 
3215 	mbox = offset / MCX_CMDQ_MAILBOX_DATASIZE;
3216 	offset %= MCX_CMDQ_MAILBOX_DATASIZE;
3217 
3218 	pas = mcx_cq_mbox_data(mcx_cq_mbox(mxm, mbox));
3219 	pas += (offset / sizeof(*pas));
3220 	mbox_pages = (MCX_CMDQ_MAILBOX_DATASIZE - offset) / sizeof(*pas);
3221 	for (i = 0; i < npages; i++) {
3222 		if (i == mbox_pages) {
3223 			mbox++;
3224 			pas = mcx_cq_mbox_data(mcx_cq_mbox(mxm, mbox));
3225 			mbox_pages += MCX_CMDQ_MAILBOX_DATASIZE / sizeof(*pas);
3226 		}
3227 		*pas = htobe64(MCX_DMA_DVA(buf) + (i * MCX_PAGE_SIZE));
3228 		pas++;
3229 	}
3230 }
3231 
3232 static void
3233 mcx_cmdq_mboxes_copyout(struct mcx_dmamem *mxm, int nmb, void *b, size_t len)
3234 {
3235 	caddr_t buf = b;
3236 	struct mcx_cmdq_mailbox *mb;
3237 	int i;
3238 
3239 	mb = (struct mcx_cmdq_mailbox *)MCX_DMA_KVA(mxm);
3240 	for (i = 0; i < nmb; i++) {
3241 		memcpy(buf, mb->mb_data, min(sizeof(mb->mb_data), len));
3242 
3243 		if (sizeof(mb->mb_data) >= len)
3244 			break;
3245 
3246 		buf += sizeof(mb->mb_data);
3247 		len -= sizeof(mb->mb_data);
3248 		mb++;
3249 	}
3250 }
3251 
3252 static void
3253 mcx_cq_mboxes_free(struct mcx_softc *sc, struct mcx_dmamem *mxm)
3254 {
3255 	mcx_dmamem_free(sc, mxm);
3256 }
3257 
3258 #if 0
3259 static void
3260 mcx_cmdq_dump(const struct mcx_cmdq_entry *cqe)
3261 {
3262 	unsigned int i;
3263 
3264 	printf(" type %02x, ilen %u, iptr %016llx", cqe->cq_type,
3265 	    bemtoh32(&cqe->cq_input_length), bemtoh64(&cqe->cq_input_ptr));
3266 
3267 	printf(", idata ");
3268 	for (i = 0; i < sizeof(cqe->cq_input_data); i++)
3269 		printf("%02x", cqe->cq_input_data[i]);
3270 
3271 	printf(", odata ");
3272 	for (i = 0; i < sizeof(cqe->cq_output_data); i++)
3273 		printf("%02x", cqe->cq_output_data[i]);
3274 
3275 	printf(", optr %016llx, olen %u, token %02x, sig %02x, status %02x",
3276 	    bemtoh64(&cqe->cq_output_ptr), bemtoh32(&cqe->cq_output_length),
3277 	    cqe->cq_token, cqe->cq_signature, cqe->cq_status);
3278 }
3279 
3280 static void
3281 mcx_cmdq_mbox_dump(struct mcx_dmamem *mboxes, int num)
3282 {
3283 	int i, j;
3284 	uint8_t *d;
3285 
3286 	for (i = 0; i < num; i++) {
3287 		struct mcx_cmdq_mailbox *mbox;
3288 		mbox = mcx_cq_mbox(mboxes, i);
3289 
3290 		d = mcx_cq_mbox_data(mbox);
3291 		for (j = 0; j < MCX_CMDQ_MAILBOX_DATASIZE; j++) {
3292 			if (j != 0 && (j % 16 == 0))
3293 				printf("\n");
3294 			printf("%.2x ", d[j]);
3295 		}
3296 	}
3297 }
3298 #endif
3299 
3300 static int
3301 mcx_access_hca_reg(struct mcx_softc *sc, uint16_t reg, int op, void *data,
3302     int len)
3303 {
3304 	struct mcx_dmamem mxm;
3305 	struct mcx_cmdq_entry *cqe;
3306 	struct mcx_cmd_access_reg_in *in;
3307 	struct mcx_cmd_access_reg_out *out;
3308 	uint8_t token = mcx_cmdq_token(sc);
3309 	int error, nmb;
3310 
3311 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
3312 	mcx_cmdq_init(sc, cqe, sizeof(*in) + len, sizeof(*out) + len,
3313 	    token);
3314 
3315 	in = mcx_cmdq_in(cqe);
3316 	in->cmd_opcode = htobe16(MCX_CMD_ACCESS_REG);
3317 	in->cmd_op_mod = htobe16(op);
3318 	in->cmd_register_id = htobe16(reg);
3319 
3320 	nmb = howmany(len, MCX_CMDQ_MAILBOX_DATASIZE);
3321 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, nmb,
3322 	    &cqe->cq_output_ptr, token) != 0) {
3323 		printf(", unable to allocate access reg mailboxen\n");
3324 		return (-1);
3325 	}
3326 	cqe->cq_input_ptr = cqe->cq_output_ptr;
3327 	mcx_cmdq_mboxes_copyin(&mxm, nmb, data, len);
3328 	mcx_cmdq_mboxes_sign(&mxm, nmb);
3329 	mcx_cmdq_mboxes_sync(sc, &mxm, BUS_DMASYNC_PRERW);
3330 
3331 	mcx_cmdq_post(sc, cqe, 0);
3332 	error = mcx_cmdq_poll(sc, cqe, 1000);
3333 	mcx_cmdq_mboxes_sync(sc, &mxm, BUS_DMASYNC_POSTRW);
3334 
3335 	if (error != 0) {
3336 		printf("%s: access reg (%s %x) timeout\n", DEVNAME(sc),
3337 		    (op == MCX_REG_OP_WRITE ? "write" : "read"), reg);
3338 		goto free;
3339 	}
3340 	error = mcx_cmdq_verify(cqe);
3341 	if (error != 0) {
3342 		printf("%s: access reg (%s %x) reply corrupt\n",
3343 		    (op == MCX_REG_OP_WRITE ? "write" : "read"), DEVNAME(sc),
3344 		    reg);
3345 		goto free;
3346 	}
3347 
3348 	out = mcx_cmdq_out(cqe);
3349 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
3350 		printf("%s: access reg (%s %x) failed (%x, %.6x)\n",
3351 		    DEVNAME(sc), (op == MCX_REG_OP_WRITE ? "write" : "read"),
3352 		    reg, out->cmd_status, betoh32(out->cmd_syndrome));
3353 		error = -1;
3354 		goto free;
3355 	}
3356 
3357 	mcx_cmdq_mboxes_copyout(&mxm, nmb, data, len);
3358 free:
3359 	mcx_dmamem_free(sc, &mxm);
3360 
3361 	return (error);
3362 }
3363 
3364 static int
3365 mcx_set_issi(struct mcx_softc *sc, struct mcx_cmdq_entry *cqe,
3366     unsigned int slot)
3367 {
3368 	struct mcx_cmd_set_issi_in *in;
3369 	struct mcx_cmd_set_issi_out *out;
3370 	uint8_t status;
3371 
3372 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
3373 
3374 	in = mcx_cmdq_in(cqe);
3375 	in->cmd_opcode = htobe16(MCX_CMD_SET_ISSI);
3376 	in->cmd_op_mod = htobe16(0);
3377 	in->cmd_current_issi = htobe16(MCX_ISSI);
3378 
3379 	mcx_cmdq_post(sc, cqe, slot);
3380 	if (mcx_cmdq_poll(sc, cqe, 1000) != 0)
3381 		return (-1);
3382 	if (mcx_cmdq_verify(cqe) != 0)
3383 		return (-1);
3384 
3385 	status = cqe->cq_output_data[0];
3386 	if (status != MCX_CQ_STATUS_OK)
3387 		return (-1);
3388 
3389 	return (0);
3390 }
3391 
3392 static int
3393 mcx_issi(struct mcx_softc *sc)
3394 {
3395 	struct mcx_dmamem mxm;
3396 	struct mcx_cmdq_entry *cqe;
3397 	struct mcx_cmd_query_issi_in *in;
3398 	struct mcx_cmd_query_issi_il_out *out;
3399 	struct mcx_cmd_query_issi_mb_out *mb;
3400 	uint8_t token = mcx_cmdq_token(sc);
3401 	uint8_t status;
3402 	int error;
3403 
3404 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
3405 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out) + sizeof(*mb), token);
3406 
3407 	in = mcx_cmdq_in(cqe);
3408 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_ISSI);
3409 	in->cmd_op_mod = htobe16(0);
3410 
3411 	CTASSERT(sizeof(*mb) <= MCX_CMDQ_MAILBOX_DATASIZE);
3412 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
3413 	    &cqe->cq_output_ptr, token) != 0) {
3414 		printf(", unable to allocate query issi mailbox\n");
3415 		return (-1);
3416 	}
3417 	mcx_cmdq_mboxes_sign(&mxm, 1);
3418 
3419 	mcx_cmdq_post(sc, cqe, 0);
3420 	error = mcx_cmdq_poll(sc, cqe, 1000);
3421 	if (error != 0) {
3422 		printf(", query issi timeout\n");
3423 		goto free;
3424 	}
3425 	error = mcx_cmdq_verify(cqe);
3426 	if (error != 0) {
3427 		printf(", query issi reply corrupt\n");
3428 		goto free;
3429 	}
3430 
3431 	status = cqe->cq_output_data[0];
3432 	switch (status) {
3433 	case MCX_CQ_STATUS_OK:
3434 		break;
3435 	case MCX_CQ_STATUS_BAD_OPCODE:
3436 		/* use ISSI 0 */
3437 		goto free;
3438 	default:
3439 		printf(", query issi failed (%x)\n", status);
3440 		error = -1;
3441 		goto free;
3442 	}
3443 
3444 	out = mcx_cmdq_out(cqe);
3445 	if (out->cmd_current_issi == htobe16(MCX_ISSI)) {
3446 		/* use ISSI 1 */
3447 		goto free;
3448 	}
3449 
3450 	/* don't need to read cqe anymore, can be used for SET ISSI */
3451 
3452 	mb = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
3453 	CTASSERT(MCX_ISSI < NBBY);
3454 	 /* XXX math is hard */
3455 	if (!ISSET(mb->cmd_supported_issi[79], 1 << MCX_ISSI)) {
3456 		/* use ISSI 0 */
3457 		goto free;
3458 	}
3459 
3460 	if (mcx_set_issi(sc, cqe, 0) != 0) {
3461 		/* ignore the error, just use ISSI 0 */
3462 	} else {
3463 		/* use ISSI 1 */
3464 	}
3465 
3466 free:
3467 	mcx_cq_mboxes_free(sc, &mxm);
3468 	return (error);
3469 }
3470 
3471 static int
3472 mcx_query_pages(struct mcx_softc *sc, uint16_t type,
3473     int32_t *npages, uint16_t *func_id)
3474 {
3475 	struct mcx_cmdq_entry *cqe;
3476 	struct mcx_cmd_query_pages_in *in;
3477 	struct mcx_cmd_query_pages_out *out;
3478 
3479 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
3480 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
3481 
3482 	in = mcx_cmdq_in(cqe);
3483 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_PAGES);
3484 	in->cmd_op_mod = type;
3485 
3486 	mcx_cmdq_post(sc, cqe, 0);
3487 	if (mcx_cmdq_poll(sc, cqe, 1000) != 0) {
3488 		printf(", query pages timeout\n");
3489 		return (-1);
3490 	}
3491 	if (mcx_cmdq_verify(cqe) != 0) {
3492 		printf(", query pages reply corrupt\n");
3493 		return (-1);
3494 	}
3495 
3496 	out = mcx_cmdq_out(cqe);
3497 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
3498 		printf(", query pages failed (%x)\n", out->cmd_status);
3499 		return (-1);
3500 	}
3501 
3502 	*func_id = out->cmd_func_id;
3503 	*npages = bemtoh32(&out->cmd_num_pages);
3504 
3505 	return (0);
3506 }
3507 
3508 struct bus_dma_iter {
3509 	bus_dmamap_t		i_map;
3510 	bus_size_t		i_offset;
3511 	unsigned int		i_index;
3512 };
3513 
3514 static void
3515 bus_dma_iter_init(struct bus_dma_iter *i, bus_dmamap_t map)
3516 {
3517 	i->i_map = map;
3518 	i->i_offset = 0;
3519 	i->i_index = 0;
3520 }
3521 
3522 static bus_addr_t
3523 bus_dma_iter_addr(struct bus_dma_iter *i)
3524 {
3525 	return (i->i_map->dm_segs[i->i_index].ds_addr + i->i_offset);
3526 }
3527 
3528 static void
3529 bus_dma_iter_add(struct bus_dma_iter *i, bus_size_t size)
3530 {
3531 	bus_dma_segment_t *seg = i->i_map->dm_segs + i->i_index;
3532 	bus_size_t diff;
3533 
3534 	do {
3535 		diff = seg->ds_len - i->i_offset;
3536 		if (size < diff)
3537 			break;
3538 
3539 		size -= diff;
3540 
3541 		seg++;
3542 
3543 		i->i_offset = 0;
3544 		i->i_index++;
3545 	} while (size > 0);
3546 
3547 	i->i_offset += size;
3548 }
3549 
3550 static int
3551 mcx_add_pages(struct mcx_softc *sc, struct mcx_hwmem *mhm, uint16_t func_id)
3552 {
3553 	struct mcx_dmamem mxm;
3554 	struct mcx_cmdq_entry *cqe;
3555 	struct mcx_cmd_manage_pages_in *in;
3556 	struct mcx_cmd_manage_pages_out *out;
3557 	unsigned int paslen, nmb, i, j, npages;
3558 	struct bus_dma_iter iter;
3559 	uint64_t *pas;
3560 	uint8_t status;
3561 	uint8_t token = mcx_cmdq_token(sc);
3562 	int error;
3563 
3564 	npages = mhm->mhm_npages;
3565 
3566 	paslen = sizeof(*pas) * npages;
3567 	nmb = howmany(paslen, MCX_CMDQ_MAILBOX_DATASIZE);
3568 
3569 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
3570 	mcx_cmdq_init(sc, cqe, sizeof(*in) + paslen, sizeof(*out), token);
3571 
3572 	in = mcx_cmdq_in(cqe);
3573 	in->cmd_opcode = htobe16(MCX_CMD_MANAGE_PAGES);
3574 	in->cmd_op_mod = htobe16(MCX_CMD_MANAGE_PAGES_ALLOC_SUCCESS);
3575 	in->cmd_func_id = func_id;
3576 	htobem32(&in->cmd_input_num_entries, npages);
3577 
3578 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, nmb,
3579 	    &cqe->cq_input_ptr, token) != 0) {
3580 		printf(", unable to allocate manage pages mailboxen\n");
3581 		return (-1);
3582 	}
3583 
3584 	bus_dma_iter_init(&iter, mhm->mhm_map);
3585 	for (i = 0; i < nmb; i++) {
3586 		unsigned int lim;
3587 
3588 		pas = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, i));
3589 		lim = min(MCX_CMDQ_MAILBOX_DATASIZE / sizeof(*pas), npages);
3590 
3591 		for (j = 0; j < lim; j++) {
3592 			htobem64(&pas[j], bus_dma_iter_addr(&iter));
3593 			bus_dma_iter_add(&iter, MCX_PAGE_SIZE);
3594 		}
3595 
3596 		npages -= lim;
3597 	}
3598 
3599 	mcx_cmdq_mboxes_sign(&mxm, nmb);
3600 
3601 	mcx_cmdq_post(sc, cqe, 0);
3602 	error = mcx_cmdq_poll(sc, cqe, 1000);
3603 	if (error != 0) {
3604 		printf(", manage pages timeout\n");
3605 		goto free;
3606 	}
3607 	error = mcx_cmdq_verify(cqe);
3608 	if (error != 0) {
3609 		printf(", manage pages reply corrupt\n");
3610 		goto free;
3611 	}
3612 
3613 	status = cqe->cq_output_data[0];
3614 	if (status != MCX_CQ_STATUS_OK) {
3615 		printf(", manage pages failed (%x)\n", status);
3616 		error = -1;
3617 		goto free;
3618 	}
3619 
3620 free:
3621 	mcx_dmamem_free(sc, &mxm);
3622 
3623 	return (error);
3624 }
3625 
3626 static int
3627 mcx_pages(struct mcx_softc *sc, struct mcx_hwmem *mhm, uint16_t type)
3628 {
3629 	int32_t npages;
3630 	uint16_t func_id;
3631 
3632 	if (mcx_query_pages(sc, type, &npages, &func_id) != 0) {
3633 		/* error printed by mcx_query_pages */
3634 		return (-1);
3635 	}
3636 
3637 	if (npages < 1)
3638 		return (0);
3639 
3640 	if (mcx_hwmem_alloc(sc, mhm, npages) != 0) {
3641 		printf(", unable to allocate hwmem\n");
3642 		return (-1);
3643 	}
3644 
3645 	if (mcx_add_pages(sc, mhm, func_id) != 0) {
3646 		printf(", unable to add hwmem\n");
3647 		goto free;
3648 	}
3649 
3650 	return (0);
3651 
3652 free:
3653 	mcx_hwmem_free(sc, mhm);
3654 
3655 	return (-1);
3656 }
3657 
3658 static int
3659 mcx_hca_max_caps(struct mcx_softc *sc)
3660 {
3661 	struct mcx_dmamem mxm;
3662 	struct mcx_cmdq_entry *cqe;
3663 	struct mcx_cmd_query_hca_cap_in *in;
3664 	struct mcx_cmd_query_hca_cap_out *out;
3665 	struct mcx_cmdq_mailbox *mb;
3666 	struct mcx_cap_device *hca;
3667 	uint8_t status;
3668 	uint8_t token = mcx_cmdq_token(sc);
3669 	int error;
3670 
3671 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
3672 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out) + MCX_HCA_CAP_LEN,
3673 	    token);
3674 
3675 	in = mcx_cmdq_in(cqe);
3676 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_HCA_CAP);
3677 	in->cmd_op_mod = htobe16(MCX_CMD_QUERY_HCA_CAP_MAX |
3678 	    MCX_CMD_QUERY_HCA_CAP_DEVICE);
3679 
3680 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, MCX_HCA_CAP_NMAILBOXES,
3681 	    &cqe->cq_output_ptr, token) != 0) {
3682 		printf(", unable to allocate query hca caps mailboxen\n");
3683 		return (-1);
3684 	}
3685 	mcx_cmdq_mboxes_sign(&mxm, MCX_HCA_CAP_NMAILBOXES);
3686 	mcx_cmdq_mboxes_sync(sc, &mxm, BUS_DMASYNC_PRERW);
3687 
3688 	mcx_cmdq_post(sc, cqe, 0);
3689 	error = mcx_cmdq_poll(sc, cqe, 1000);
3690 	mcx_cmdq_mboxes_sync(sc, &mxm, BUS_DMASYNC_POSTRW);
3691 
3692 	if (error != 0) {
3693 		printf(", query hca caps timeout\n");
3694 		goto free;
3695 	}
3696 	error = mcx_cmdq_verify(cqe);
3697 	if (error != 0) {
3698 		printf(", query hca caps reply corrupt\n");
3699 		goto free;
3700 	}
3701 
3702 	status = cqe->cq_output_data[0];
3703 	if (status != MCX_CQ_STATUS_OK) {
3704 		printf(", query hca caps failed (%x)\n", status);
3705 		error = -1;
3706 		goto free;
3707 	}
3708 
3709 	mb = mcx_cq_mbox(&mxm, 0);
3710 	hca = mcx_cq_mbox_data(mb);
3711 
3712 	if (hca->log_pg_sz > PAGE_SHIFT) {
3713 		printf(", minimum system page shift %u is too large\n",
3714 		    hca->log_pg_sz);
3715 		error = -1;
3716 		goto free;
3717 	}
3718 	/*
3719 	 * blueflame register is split into two buffers, and we must alternate
3720 	 * between the two of them.
3721 	 */
3722 	sc->sc_bf_size = (1 << hca->log_bf_reg_size) / 2;
3723 	sc->sc_max_rqt_size = (1 << hca->log_max_rqt_size);
3724 
3725 free:
3726 	mcx_dmamem_free(sc, &mxm);
3727 
3728 	return (error);
3729 }
3730 
3731 static int
3732 mcx_hca_set_caps(struct mcx_softc *sc)
3733 {
3734 	struct mcx_dmamem mxm;
3735 	struct mcx_cmdq_entry *cqe;
3736 	struct mcx_cmd_query_hca_cap_in *in;
3737 	struct mcx_cmd_query_hca_cap_out *out;
3738 	struct mcx_cmdq_mailbox *mb;
3739 	struct mcx_cap_device *hca;
3740 	uint8_t status;
3741 	uint8_t token = mcx_cmdq_token(sc);
3742 	int error;
3743 
3744 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
3745 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out) + MCX_HCA_CAP_LEN,
3746 	    token);
3747 
3748 	in = mcx_cmdq_in(cqe);
3749 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_HCA_CAP);
3750 	in->cmd_op_mod = htobe16(MCX_CMD_QUERY_HCA_CAP_CURRENT |
3751 	    MCX_CMD_QUERY_HCA_CAP_DEVICE);
3752 
3753 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, MCX_HCA_CAP_NMAILBOXES,
3754 	    &cqe->cq_output_ptr, token) != 0) {
3755 		printf(", unable to allocate manage pages mailboxen\n");
3756 		return (-1);
3757 	}
3758 	mcx_cmdq_mboxes_sign(&mxm, MCX_HCA_CAP_NMAILBOXES);
3759 	mcx_cmdq_mboxes_sync(sc, &mxm, BUS_DMASYNC_PRERW);
3760 
3761 	mcx_cmdq_post(sc, cqe, 0);
3762 	error = mcx_cmdq_poll(sc, cqe, 1000);
3763 	mcx_cmdq_mboxes_sync(sc, &mxm, BUS_DMASYNC_POSTRW);
3764 
3765 	if (error != 0) {
3766 		printf(", query hca caps timeout\n");
3767 		goto free;
3768 	}
3769 	error = mcx_cmdq_verify(cqe);
3770 	if (error != 0) {
3771 		printf(", query hca caps reply corrupt\n");
3772 		goto free;
3773 	}
3774 
3775 	status = cqe->cq_output_data[0];
3776 	if (status != MCX_CQ_STATUS_OK) {
3777 		printf(", query hca caps failed (%x)\n", status);
3778 		error = -1;
3779 		goto free;
3780 	}
3781 
3782 	mb = mcx_cq_mbox(&mxm, 0);
3783 	hca = mcx_cq_mbox_data(mb);
3784 
3785 	hca->log_pg_sz = PAGE_SHIFT;
3786 
3787 free:
3788 	mcx_dmamem_free(sc, &mxm);
3789 
3790 	return (error);
3791 }
3792 
3793 
3794 static int
3795 mcx_init_hca(struct mcx_softc *sc)
3796 {
3797 	struct mcx_cmdq_entry *cqe;
3798 	struct mcx_cmd_init_hca_in *in;
3799 	struct mcx_cmd_init_hca_out *out;
3800 	int error;
3801 	uint8_t status;
3802 
3803 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
3804 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
3805 
3806 	in = mcx_cmdq_in(cqe);
3807 	in->cmd_opcode = htobe16(MCX_CMD_INIT_HCA);
3808 	in->cmd_op_mod = htobe16(0);
3809 
3810 	mcx_cmdq_post(sc, cqe, 0);
3811 
3812 	error = mcx_cmdq_poll(sc, cqe, 1000);
3813 	if (error != 0) {
3814 		printf(", hca init timeout\n");
3815 		return (-1);
3816 	}
3817 	if (mcx_cmdq_verify(cqe) != 0) {
3818 		printf(", hca init command corrupt\n");
3819 		return (-1);
3820 	}
3821 
3822 	status = cqe->cq_output_data[0];
3823 	if (status != MCX_CQ_STATUS_OK) {
3824 		printf(", hca init failed (%x)\n", status);
3825 		return (-1);
3826 	}
3827 
3828 	return (0);
3829 }
3830 
3831 static int
3832 mcx_set_driver_version(struct mcx_softc *sc)
3833 {
3834 	struct mcx_dmamem mxm;
3835 	struct mcx_cmdq_entry *cqe;
3836 	struct mcx_cmd_set_driver_version_in *in;
3837 	struct mcx_cmd_set_driver_version_out *out;
3838 	int error;
3839 	int token;
3840 	uint8_t status;
3841 
3842 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
3843 	token = mcx_cmdq_token(sc);
3844 	mcx_cmdq_init(sc, cqe, sizeof(*in) +
3845 	    sizeof(struct mcx_cmd_set_driver_version), sizeof(*out), token);
3846 
3847 	in = mcx_cmdq_in(cqe);
3848 	in->cmd_opcode = htobe16(MCX_CMD_SET_DRIVER_VERSION);
3849 	in->cmd_op_mod = htobe16(0);
3850 
3851 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
3852 	    &cqe->cq_input_ptr, token) != 0) {
3853 		printf(", unable to allocate set driver version mailboxen\n");
3854 		return (-1);
3855 	}
3856 	strlcpy(mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)),
3857 	    "OpenBSD,mcx,1.000.000000", MCX_CMDQ_MAILBOX_DATASIZE);
3858 
3859 	mcx_cmdq_mboxes_sign(&mxm, 1);
3860 	mcx_cmdq_post(sc, cqe, 0);
3861 
3862 	error = mcx_cmdq_poll(sc, cqe, 1000);
3863 	if (error != 0) {
3864 		printf(", set driver version timeout\n");
3865 		goto free;
3866 	}
3867 	if (mcx_cmdq_verify(cqe) != 0) {
3868 		printf(", set driver version command corrupt\n");
3869 		goto free;
3870 	}
3871 
3872 	status = cqe->cq_output_data[0];
3873 	if (status != MCX_CQ_STATUS_OK) {
3874 		printf(", set driver version failed (%x)\n", status);
3875 		error = -1;
3876 		goto free;
3877 	}
3878 
3879 free:
3880 	mcx_dmamem_free(sc, &mxm);
3881 
3882 	return (error);
3883 }
3884 
3885 static int
3886 mcx_iff(struct mcx_softc *sc)
3887 {
3888 	struct ifnet *ifp = &sc->sc_ac.ac_if;
3889 	struct mcx_dmamem mxm;
3890 	struct mcx_cmdq_entry *cqe;
3891 	struct mcx_cmd_modify_nic_vport_context_in *in;
3892 	struct mcx_cmd_modify_nic_vport_context_out *out;
3893 	struct mcx_nic_vport_ctx *ctx;
3894 	int error;
3895 	int token;
3896 	int insize;
3897 	uint32_t dest;
3898 
3899 	dest = MCX_FLOW_CONTEXT_DEST_TYPE_TABLE |
3900 	    sc->sc_rss_flow_table_id;
3901 
3902 	/* enable or disable the promisc flow */
3903 	if (ISSET(ifp->if_flags, IFF_PROMISC)) {
3904 		if (sc->sc_promisc_flow_enabled == 0) {
3905 			mcx_set_flow_table_entry_mac(sc,
3906 			    MCX_FLOW_GROUP_PROMISC, 0, NULL, dest);
3907 			sc->sc_promisc_flow_enabled = 1;
3908 		}
3909 	} else if (sc->sc_promisc_flow_enabled != 0) {
3910 		mcx_delete_flow_table_entry(sc, MCX_FLOW_GROUP_PROMISC, 0);
3911 		sc->sc_promisc_flow_enabled = 0;
3912 	}
3913 
3914 	/* enable or disable the all-multicast flow */
3915 	if (ISSET(ifp->if_flags, IFF_ALLMULTI)) {
3916 		if (sc->sc_allmulti_flow_enabled == 0) {
3917 			uint8_t mcast[ETHER_ADDR_LEN];
3918 
3919 			memset(mcast, 0, sizeof(mcast));
3920 			mcast[0] = 0x01;
3921 			mcx_set_flow_table_entry_mac(sc,
3922 			    MCX_FLOW_GROUP_ALLMULTI, 0, mcast, dest);
3923 			sc->sc_allmulti_flow_enabled = 1;
3924 		}
3925 	} else if (sc->sc_allmulti_flow_enabled != 0) {
3926 		mcx_delete_flow_table_entry(sc, MCX_FLOW_GROUP_ALLMULTI, 0);
3927 		sc->sc_allmulti_flow_enabled = 0;
3928 	}
3929 
3930 	insize = sizeof(struct mcx_nic_vport_ctx) + 240;
3931 
3932 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
3933 	token = mcx_cmdq_token(sc);
3934 	mcx_cmdq_init(sc, cqe, sizeof(*in) + insize, sizeof(*out), token);
3935 
3936 	in = mcx_cmdq_in(cqe);
3937 	in->cmd_opcode = htobe16(MCX_CMD_MODIFY_NIC_VPORT_CONTEXT);
3938 	in->cmd_op_mod = htobe16(0);
3939 	in->cmd_field_select = htobe32(
3940 	    MCX_CMD_MODIFY_NIC_VPORT_CONTEXT_FIELD_PROMISC |
3941 	    MCX_CMD_MODIFY_NIC_VPORT_CONTEXT_FIELD_MTU);
3942 
3943 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1, &cqe->cq_input_ptr, token) != 0) {
3944 		printf(", unable to allocate modify "
3945 		    "nic vport context mailboxen\n");
3946 		return (-1);
3947 	}
3948 	ctx = (struct mcx_nic_vport_ctx *)
3949 	    (((char *)mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0))) + 240);
3950 	ctx->vp_mtu = htobe32(sc->sc_hardmtu);
3951 	/*
3952          * always leave promisc-all enabled on the vport since we
3953          * can't give it a vlan list, and we're already doing multicast
3954          * filtering in the flow table.
3955 	 */
3956 	ctx->vp_flags = htobe16(MCX_NIC_VPORT_CTX_PROMISC_ALL);
3957 
3958 	mcx_cmdq_mboxes_sign(&mxm, 1);
3959 	mcx_cmdq_post(sc, cqe, 0);
3960 
3961 	error = mcx_cmdq_poll(sc, cqe, 1000);
3962 	if (error != 0) {
3963 		printf(", modify nic vport context timeout\n");
3964 		goto free;
3965 	}
3966 	if (mcx_cmdq_verify(cqe) != 0) {
3967 		printf(", modify nic vport context command corrupt\n");
3968 		goto free;
3969 	}
3970 
3971 	out = mcx_cmdq_out(cqe);
3972 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
3973 		printf(", modify nic vport context failed (%x, %x)\n",
3974 		    out->cmd_status, betoh32(out->cmd_syndrome));
3975 		error = -1;
3976 		goto free;
3977 	}
3978 
3979 free:
3980 	mcx_dmamem_free(sc, &mxm);
3981 
3982 	return (error);
3983 }
3984 
3985 static int
3986 mcx_alloc_uar(struct mcx_softc *sc, int *uar)
3987 {
3988 	struct mcx_cmdq_entry *cqe;
3989 	struct mcx_cmd_alloc_uar_in *in;
3990 	struct mcx_cmd_alloc_uar_out *out;
3991 	int error;
3992 
3993 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
3994 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
3995 
3996 	in = mcx_cmdq_in(cqe);
3997 	in->cmd_opcode = htobe16(MCX_CMD_ALLOC_UAR);
3998 	in->cmd_op_mod = htobe16(0);
3999 
4000 	mcx_cmdq_post(sc, cqe, 0);
4001 
4002 	error = mcx_cmdq_poll(sc, cqe, 1000);
4003 	if (error != 0) {
4004 		printf(", alloc uar timeout\n");
4005 		return (-1);
4006 	}
4007 	if (mcx_cmdq_verify(cqe) != 0) {
4008 		printf(", alloc uar command corrupt\n");
4009 		return (-1);
4010 	}
4011 
4012 	out = mcx_cmdq_out(cqe);
4013 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4014 		printf(", alloc uar failed (%x)\n", out->cmd_status);
4015 		return (-1);
4016 	}
4017 
4018 	*uar = mcx_get_id(out->cmd_uar);
4019 	return (0);
4020 }
4021 
4022 static int
4023 mcx_create_eq(struct mcx_softc *sc, struct mcx_eq *eq, int uar,
4024     uint64_t events, int vector)
4025 {
4026 	struct mcx_cmdq_entry *cqe;
4027 	struct mcx_dmamem mxm;
4028 	struct mcx_cmd_create_eq_in *in;
4029 	struct mcx_cmd_create_eq_mb_in *mbin;
4030 	struct mcx_cmd_create_eq_out *out;
4031 	struct mcx_eq_entry *eqe;
4032 	int error;
4033 	uint64_t *pas;
4034 	int insize, npages, paslen, i, token;
4035 
4036 	eq->eq_cons = 0;
4037 
4038 	npages = howmany((1 << MCX_LOG_EQ_SIZE) * sizeof(struct mcx_eq_entry),
4039 	    MCX_PAGE_SIZE);
4040 	paslen = npages * sizeof(*pas);
4041 	insize = sizeof(struct mcx_cmd_create_eq_mb_in) + paslen;
4042 
4043 	if (mcx_dmamem_alloc(sc, &eq->eq_mem, npages * MCX_PAGE_SIZE,
4044 	    MCX_PAGE_SIZE) != 0) {
4045 		printf(", unable to allocate event queue memory\n");
4046 		return (-1);
4047 	}
4048 
4049 	eqe = (struct mcx_eq_entry *)MCX_DMA_KVA(&eq->eq_mem);
4050 	for (i = 0; i < (1 << MCX_LOG_EQ_SIZE); i++) {
4051 		eqe[i].eq_owner = MCX_EQ_ENTRY_OWNER_INIT;
4052 	}
4053 
4054 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4055 	token = mcx_cmdq_token(sc);
4056 	mcx_cmdq_init(sc, cqe, sizeof(*in) + insize, sizeof(*out), token);
4057 
4058 	in = mcx_cmdq_in(cqe);
4059 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_EQ);
4060 	in->cmd_op_mod = htobe16(0);
4061 
4062 	if (mcx_cmdq_mboxes_alloc(sc, &mxm,
4063 	    howmany(insize, MCX_CMDQ_MAILBOX_DATASIZE),
4064 	    &cqe->cq_input_ptr, token) != 0) {
4065 		printf(", unable to allocate create eq mailboxen\n");
4066 		return (-1);
4067 	}
4068 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
4069 	mbin->cmd_eq_ctx.eq_uar_size = htobe32(
4070 	    (MCX_LOG_EQ_SIZE << MCX_EQ_CTX_LOG_EQ_SIZE_SHIFT) | uar);
4071 	mbin->cmd_eq_ctx.eq_intr = vector;
4072 	mbin->cmd_event_bitmask = htobe64(events);
4073 
4074 	/* physical addresses follow the mailbox in data */
4075 	mcx_cmdq_mboxes_pas(&mxm, sizeof(*mbin), npages, &eq->eq_mem);
4076 	mcx_cmdq_mboxes_sign(&mxm, howmany(insize, MCX_CMDQ_MAILBOX_DATASIZE));
4077 	mcx_cmdq_post(sc, cqe, 0);
4078 
4079 	error = mcx_cmdq_poll(sc, cqe, 1000);
4080 	if (error != 0) {
4081 		printf(", create eq timeout\n");
4082 		goto free;
4083 	}
4084 	if (mcx_cmdq_verify(cqe) != 0) {
4085 		printf(", create eq command corrupt\n");
4086 		goto free;
4087 	}
4088 
4089 	out = mcx_cmdq_out(cqe);
4090 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4091 		printf(", create eq failed (%x, %x)\n", out->cmd_status,
4092 		    betoh32(out->cmd_syndrome));
4093 		error = -1;
4094 		goto free;
4095 	}
4096 
4097 	eq->eq_n = mcx_get_id(out->cmd_eqn);
4098 	mcx_arm_eq(sc, eq, uar);
4099 free:
4100 	mcx_dmamem_free(sc, &mxm);
4101 	return (error);
4102 }
4103 
4104 static int
4105 mcx_alloc_pd(struct mcx_softc *sc)
4106 {
4107 	struct mcx_cmdq_entry *cqe;
4108 	struct mcx_cmd_alloc_pd_in *in;
4109 	struct mcx_cmd_alloc_pd_out *out;
4110 	int error;
4111 
4112 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4113 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
4114 
4115 	in = mcx_cmdq_in(cqe);
4116 	in->cmd_opcode = htobe16(MCX_CMD_ALLOC_PD);
4117 	in->cmd_op_mod = htobe16(0);
4118 
4119 	mcx_cmdq_post(sc, cqe, 0);
4120 
4121 	error = mcx_cmdq_poll(sc, cqe, 1000);
4122 	if (error != 0) {
4123 		printf(", alloc pd timeout\n");
4124 		return (-1);
4125 	}
4126 	if (mcx_cmdq_verify(cqe) != 0) {
4127 		printf(", alloc pd command corrupt\n");
4128 		return (-1);
4129 	}
4130 
4131 	out = mcx_cmdq_out(cqe);
4132 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4133 		printf(", alloc pd failed (%x)\n", out->cmd_status);
4134 		return (-1);
4135 	}
4136 
4137 	sc->sc_pd = mcx_get_id(out->cmd_pd);
4138 	return (0);
4139 }
4140 
4141 static int
4142 mcx_alloc_tdomain(struct mcx_softc *sc)
4143 {
4144 	struct mcx_cmdq_entry *cqe;
4145 	struct mcx_cmd_alloc_td_in *in;
4146 	struct mcx_cmd_alloc_td_out *out;
4147 	int error;
4148 
4149 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4150 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
4151 
4152 	in = mcx_cmdq_in(cqe);
4153 	in->cmd_opcode = htobe16(MCX_CMD_ALLOC_TRANSPORT_DOMAIN);
4154 	in->cmd_op_mod = htobe16(0);
4155 
4156 	mcx_cmdq_post(sc, cqe, 0);
4157 
4158 	error = mcx_cmdq_poll(sc, cqe, 1000);
4159 	if (error != 0) {
4160 		printf(", alloc transport domain timeout\n");
4161 		return (-1);
4162 	}
4163 	if (mcx_cmdq_verify(cqe) != 0) {
4164 		printf(", alloc transport domain command corrupt\n");
4165 		return (-1);
4166 	}
4167 
4168 	out = mcx_cmdq_out(cqe);
4169 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4170 		printf(", alloc transport domain failed (%x)\n",
4171 		    out->cmd_status);
4172 		return (-1);
4173 	}
4174 
4175 	sc->sc_tdomain = mcx_get_id(out->cmd_tdomain);
4176 	return (0);
4177 }
4178 
4179 static int
4180 mcx_query_nic_vport_context(struct mcx_softc *sc)
4181 {
4182 	struct mcx_dmamem mxm;
4183 	struct mcx_cmdq_entry *cqe;
4184 	struct mcx_cmd_query_nic_vport_context_in *in;
4185 	struct mcx_cmd_query_nic_vport_context_out *out;
4186 	struct mcx_nic_vport_ctx *ctx;
4187 	uint8_t *addr;
4188 	int error, token, i;
4189 
4190 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4191 	token = mcx_cmdq_token(sc);
4192 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out) + sizeof(*ctx), token);
4193 
4194 	in = mcx_cmdq_in(cqe);
4195 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_NIC_VPORT_CONTEXT);
4196 	in->cmd_op_mod = htobe16(0);
4197 	in->cmd_allowed_list_type = 0;
4198 
4199 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
4200 	    &cqe->cq_output_ptr, token) != 0) {
4201 		printf(", unable to allocate "
4202 		    "query nic vport context mailboxen\n");
4203 		return (-1);
4204 	}
4205 	mcx_cmdq_mboxes_sign(&mxm, 1);
4206 	mcx_cmdq_post(sc, cqe, 0);
4207 
4208 	error = mcx_cmdq_poll(sc, cqe, 1000);
4209 	if (error != 0) {
4210 		printf(", query nic vport context timeout\n");
4211 		goto free;
4212 	}
4213 	if (mcx_cmdq_verify(cqe) != 0) {
4214 		printf(", query nic vport context command corrupt\n");
4215 		goto free;
4216 	}
4217 
4218 	out = mcx_cmdq_out(cqe);
4219 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4220 		printf(", query nic vport context failed (%x, %x)\n",
4221 		    out->cmd_status, betoh32(out->cmd_syndrome));
4222 		error = -1;
4223 		goto free;
4224 	}
4225 
4226 	ctx = (struct mcx_nic_vport_ctx *)
4227 	    mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
4228 	addr = (uint8_t *)&ctx->vp_perm_addr;
4229 	for (i = 0; i < ETHER_ADDR_LEN; i++) {
4230 		sc->sc_ac.ac_enaddr[i] = addr[i + 2];
4231 	}
4232 free:
4233 	mcx_dmamem_free(sc, &mxm);
4234 
4235 	return (error);
4236 }
4237 
4238 static int
4239 mcx_query_special_contexts(struct mcx_softc *sc)
4240 {
4241 	struct mcx_cmdq_entry *cqe;
4242 	struct mcx_cmd_query_special_ctx_in *in;
4243 	struct mcx_cmd_query_special_ctx_out *out;
4244 	int error;
4245 
4246 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4247 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
4248 
4249 	in = mcx_cmdq_in(cqe);
4250 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_SPECIAL_CONTEXTS);
4251 	in->cmd_op_mod = htobe16(0);
4252 
4253 	mcx_cmdq_post(sc, cqe, 0);
4254 
4255 	error = mcx_cmdq_poll(sc, cqe, 1000);
4256 	if (error != 0) {
4257 		printf(", query special contexts timeout\n");
4258 		return (-1);
4259 	}
4260 	if (mcx_cmdq_verify(cqe) != 0) {
4261 		printf(", query special contexts command corrupt\n");
4262 		return (-1);
4263 	}
4264 
4265 	out = mcx_cmdq_out(cqe);
4266 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4267 		printf(", query special contexts failed (%x)\n",
4268 		    out->cmd_status);
4269 		return (-1);
4270 	}
4271 
4272 	sc->sc_lkey = betoh32(out->cmd_resd_lkey);
4273 	return (0);
4274 }
4275 
4276 static int
4277 mcx_set_port_mtu(struct mcx_softc *sc, int mtu)
4278 {
4279 	struct mcx_reg_pmtu pmtu;
4280 	int error;
4281 
4282 	/* read max mtu */
4283 	memset(&pmtu, 0, sizeof(pmtu));
4284 	pmtu.rp_local_port = 1;
4285 	error = mcx_access_hca_reg(sc, MCX_REG_PMTU, MCX_REG_OP_READ, &pmtu,
4286 	    sizeof(pmtu));
4287 	if (error != 0) {
4288 		printf(", unable to get port MTU\n");
4289 		return error;
4290 	}
4291 
4292 	mtu = min(mtu, betoh16(pmtu.rp_max_mtu));
4293 	pmtu.rp_admin_mtu = htobe16(mtu);
4294 	error = mcx_access_hca_reg(sc, MCX_REG_PMTU, MCX_REG_OP_WRITE, &pmtu,
4295 	    sizeof(pmtu));
4296 	if (error != 0) {
4297 		printf(", unable to set port MTU\n");
4298 		return error;
4299 	}
4300 
4301 	sc->sc_hardmtu = mtu;
4302 	sc->sc_rxbufsz = roundup(mtu + ETHER_ALIGN, sizeof(long));
4303 	return 0;
4304 }
4305 
4306 static int
4307 mcx_create_cq(struct mcx_softc *sc, struct mcx_cq *cq, int uar, int db, int eqn)
4308 {
4309 	struct mcx_cmdq_entry *cmde;
4310 	struct mcx_cq_entry *cqe;
4311 	struct mcx_dmamem mxm;
4312 	struct mcx_cmd_create_cq_in *in;
4313 	struct mcx_cmd_create_cq_mb_in *mbin;
4314 	struct mcx_cmd_create_cq_out *out;
4315 	int error;
4316 	uint64_t *pas;
4317 	int insize, npages, paslen, i, token;
4318 
4319 	npages = howmany((1 << MCX_LOG_CQ_SIZE) * sizeof(struct mcx_cq_entry),
4320 	    MCX_PAGE_SIZE);
4321 	paslen = npages * sizeof(*pas);
4322 	insize = sizeof(struct mcx_cmd_create_cq_mb_in) + paslen;
4323 
4324 	if (mcx_dmamem_alloc(sc, &cq->cq_mem, npages * MCX_PAGE_SIZE,
4325 	    MCX_PAGE_SIZE) != 0) {
4326 		printf("%s: unable to allocate completion queue memory\n",
4327 		    DEVNAME(sc));
4328 		return (-1);
4329 	}
4330 	cqe = MCX_DMA_KVA(&cq->cq_mem);
4331 	for (i = 0; i < (1 << MCX_LOG_CQ_SIZE); i++) {
4332 		cqe[i].cq_opcode_owner = MCX_CQ_ENTRY_FLAG_OWNER;
4333 	}
4334 
4335 	cmde = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4336 	token = mcx_cmdq_token(sc);
4337 	mcx_cmdq_init(sc, cmde, sizeof(*in) + insize, sizeof(*out), token);
4338 
4339 	in = mcx_cmdq_in(cmde);
4340 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_CQ);
4341 	in->cmd_op_mod = htobe16(0);
4342 
4343 	if (mcx_cmdq_mboxes_alloc(sc, &mxm,
4344 	    howmany(insize, MCX_CMDQ_MAILBOX_DATASIZE),
4345 	    &cmde->cq_input_ptr, token) != 0) {
4346 		printf("%s: unable to allocate create cq mailboxen\n",
4347 		    DEVNAME(sc));
4348 		error = -1;
4349 		goto free;
4350 	}
4351 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
4352 	mbin->cmd_cq_ctx.cq_uar_size = htobe32(
4353 	    (MCX_LOG_CQ_SIZE << MCX_CQ_CTX_LOG_CQ_SIZE_SHIFT) | uar);
4354 	mbin->cmd_cq_ctx.cq_eqn = htobe32(eqn);
4355 	mbin->cmd_cq_ctx.cq_period_max_count = htobe32(
4356 	    (MCX_CQ_MOD_PERIOD << MCX_CQ_CTX_PERIOD_SHIFT) |
4357 	    MCX_CQ_MOD_COUNTER);
4358 	mbin->cmd_cq_ctx.cq_doorbell = htobe64(
4359 	    MCX_DMA_DVA(&sc->sc_doorbell_mem) +
4360 	    MCX_CQ_DOORBELL_BASE + (MCX_CQ_DOORBELL_STRIDE * db));
4361 
4362 	/* physical addresses follow the mailbox in data */
4363 	mcx_cmdq_mboxes_pas(&mxm, sizeof(*mbin), npages, &cq->cq_mem);
4364 	mcx_cmdq_post(sc, cmde, 0);
4365 
4366 	error = mcx_cmdq_poll(sc, cmde, 1000);
4367 	if (error != 0) {
4368 		printf("%s: create cq timeout\n", DEVNAME(sc));
4369 		goto free;
4370 	}
4371 	if (mcx_cmdq_verify(cmde) != 0) {
4372 		printf("%s: create cq command corrupt\n", DEVNAME(sc));
4373 		goto free;
4374 	}
4375 
4376 	out = mcx_cmdq_out(cmde);
4377 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4378 		printf("%s: create cq failed (%x, %x)\n", DEVNAME(sc),
4379 		    out->cmd_status, betoh32(out->cmd_syndrome));
4380 		error = -1;
4381 		goto free;
4382 	}
4383 
4384 	cq->cq_n = mcx_get_id(out->cmd_cqn);
4385 	cq->cq_cons = 0;
4386 	cq->cq_count = 0;
4387 	cq->cq_doorbell = MCX_DMA_KVA(&sc->sc_doorbell_mem) +
4388 	    MCX_CQ_DOORBELL_BASE + (MCX_CQ_DOORBELL_STRIDE * db);
4389 	mcx_arm_cq(sc, cq, uar);
4390 
4391 free:
4392 	mcx_dmamem_free(sc, &mxm);
4393 	return (error);
4394 }
4395 
4396 static int
4397 mcx_destroy_cq(struct mcx_softc *sc, struct mcx_cq *cq)
4398 {
4399 	struct mcx_cmdq_entry *cqe;
4400 	struct mcx_cmd_destroy_cq_in *in;
4401 	struct mcx_cmd_destroy_cq_out *out;
4402 	int error;
4403 	int token;
4404 
4405 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4406 	token = mcx_cmdq_token(sc);
4407 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), token);
4408 
4409 	in = mcx_cmdq_in(cqe);
4410 	in->cmd_opcode = htobe16(MCX_CMD_DESTROY_CQ);
4411 	in->cmd_op_mod = htobe16(0);
4412 	in->cmd_cqn = htobe32(cq->cq_n);
4413 
4414 	mcx_cmdq_post(sc, cqe, 0);
4415 	error = mcx_cmdq_poll(sc, cqe, 1000);
4416 	if (error != 0) {
4417 		printf("%s: destroy cq timeout\n", DEVNAME(sc));
4418 		return error;
4419 	}
4420 	if (mcx_cmdq_verify(cqe) != 0) {
4421 		printf("%s: destroy cq command corrupt\n", DEVNAME(sc));
4422 		return error;
4423 	}
4424 
4425 	out = mcx_cmdq_out(cqe);
4426 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4427 		printf("%s: destroy cq failed (%x, %x)\n", DEVNAME(sc),
4428 		    out->cmd_status, betoh32(out->cmd_syndrome));
4429 		return -1;
4430 	}
4431 
4432 	cq->cq_n = 0;
4433 	mcx_dmamem_free(sc, &cq->cq_mem);
4434 	cq->cq_cons = 0;
4435 	cq->cq_count = 0;
4436 	return 0;
4437 }
4438 
4439 static int
4440 mcx_create_rq(struct mcx_softc *sc, struct mcx_rx *rx, int db, int cqn)
4441 {
4442 	struct mcx_cmdq_entry *cqe;
4443 	struct mcx_dmamem mxm;
4444 	struct mcx_cmd_create_rq_in *in;
4445 	struct mcx_cmd_create_rq_out *out;
4446 	struct mcx_rq_ctx *mbin;
4447 	int error;
4448 	uint64_t *pas;
4449 	uint32_t rq_flags;
4450 	uint8_t *doorbell;
4451 	int insize, npages, paslen, token;
4452 
4453 	npages = howmany((1 << MCX_LOG_RQ_SIZE) * sizeof(struct mcx_rq_entry),
4454 	    MCX_PAGE_SIZE);
4455 	paslen = npages * sizeof(*pas);
4456 	insize = 0x10 + sizeof(struct mcx_rq_ctx) + paslen;
4457 
4458 	if (mcx_dmamem_alloc(sc, &rx->rx_rq_mem, npages * MCX_PAGE_SIZE,
4459 	    MCX_PAGE_SIZE) != 0) {
4460 		printf("%s: unable to allocate receive queue memory\n",
4461 		    DEVNAME(sc));
4462 		return (-1);
4463 	}
4464 
4465 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4466 	token = mcx_cmdq_token(sc);
4467 	mcx_cmdq_init(sc, cqe, sizeof(*in) + insize, sizeof(*out), token);
4468 
4469 	in = mcx_cmdq_in(cqe);
4470 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_RQ);
4471 	in->cmd_op_mod = htobe16(0);
4472 
4473 	if (mcx_cmdq_mboxes_alloc(sc, &mxm,
4474 	    howmany(insize, MCX_CMDQ_MAILBOX_DATASIZE),
4475 	    &cqe->cq_input_ptr, token) != 0) {
4476 		printf("%s: unable to allocate create rq mailboxen\n",
4477 		    DEVNAME(sc));
4478 		error = -1;
4479 		goto free;
4480 	}
4481 	mbin = (struct mcx_rq_ctx *)
4482 	    (((char *)mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0))) + 0x10);
4483 	rq_flags = MCX_RQ_CTX_RLKEY;
4484 #if NVLAN == 0
4485 	rq_flags |= MCX_RQ_CTX_VLAN_STRIP_DIS;
4486 #endif
4487 	mbin->rq_flags = htobe32(rq_flags);
4488 	mbin->rq_cqn = htobe32(cqn);
4489 	mbin->rq_wq.wq_type = MCX_WQ_CTX_TYPE_CYCLIC;
4490 	mbin->rq_wq.wq_pd = htobe32(sc->sc_pd);
4491 	mbin->rq_wq.wq_doorbell = htobe64(MCX_DMA_DVA(&sc->sc_doorbell_mem) +
4492 	    MCX_WQ_DOORBELL_BASE + (db * MCX_WQ_DOORBELL_STRIDE));
4493 	mbin->rq_wq.wq_log_stride = htobe16(4);
4494 	mbin->rq_wq.wq_log_size = MCX_LOG_RQ_SIZE;
4495 
4496 	/* physical addresses follow the mailbox in data */
4497 	mcx_cmdq_mboxes_pas(&mxm, sizeof(*mbin) + 0x10, npages, &rx->rx_rq_mem);
4498 	mcx_cmdq_post(sc, cqe, 0);
4499 
4500 	error = mcx_cmdq_poll(sc, cqe, 1000);
4501 	if (error != 0) {
4502 		printf("%s: create rq timeout\n", DEVNAME(sc));
4503 		goto free;
4504 	}
4505 	if (mcx_cmdq_verify(cqe) != 0) {
4506 		printf("%s: create rq command corrupt\n", DEVNAME(sc));
4507 		goto free;
4508 	}
4509 
4510 	out = mcx_cmdq_out(cqe);
4511 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4512 		printf("%s: create rq failed (%x, %x)\n", DEVNAME(sc),
4513 		    out->cmd_status, betoh32(out->cmd_syndrome));
4514 		error = -1;
4515 		goto free;
4516 	}
4517 
4518 	rx->rx_rqn = mcx_get_id(out->cmd_rqn);
4519 
4520 	doorbell = MCX_DMA_KVA(&sc->sc_doorbell_mem);
4521 	rx->rx_doorbell = (uint32_t *)(doorbell + MCX_WQ_DOORBELL_BASE +
4522 	    (db * MCX_WQ_DOORBELL_STRIDE));
4523 
4524 free:
4525 	mcx_dmamem_free(sc, &mxm);
4526 	return (error);
4527 }
4528 
4529 static int
4530 mcx_ready_rq(struct mcx_softc *sc, struct mcx_rx *rx)
4531 {
4532 	struct mcx_cmdq_entry *cqe;
4533 	struct mcx_dmamem mxm;
4534 	struct mcx_cmd_modify_rq_in *in;
4535 	struct mcx_cmd_modify_rq_mb_in *mbin;
4536 	struct mcx_cmd_modify_rq_out *out;
4537 	int error;
4538 	int token;
4539 
4540 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4541 	token = mcx_cmdq_token(sc);
4542 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
4543 	    sizeof(*out), token);
4544 
4545 	in = mcx_cmdq_in(cqe);
4546 	in->cmd_opcode = htobe16(MCX_CMD_MODIFY_RQ);
4547 	in->cmd_op_mod = htobe16(0);
4548 	in->cmd_rq_state = htobe32((MCX_QUEUE_STATE_RST << 28) | rx->rx_rqn);
4549 
4550 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
4551 	    &cqe->cq_input_ptr, token) != 0) {
4552 		printf("%s: unable to allocate modify rq mailbox\n",
4553 		    DEVNAME(sc));
4554 		return (-1);
4555 	}
4556 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
4557 	mbin->cmd_rq_ctx.rq_flags = htobe32(
4558 	    MCX_QUEUE_STATE_RDY << MCX_RQ_CTX_STATE_SHIFT);
4559 
4560 	mcx_cmdq_mboxes_sign(&mxm, 1);
4561 	mcx_cmdq_post(sc, cqe, 0);
4562 	error = mcx_cmdq_poll(sc, cqe, 1000);
4563 	if (error != 0) {
4564 		printf("%s: modify rq timeout\n", DEVNAME(sc));
4565 		goto free;
4566 	}
4567 	if (mcx_cmdq_verify(cqe) != 0) {
4568 		printf("%s: modify rq command corrupt\n", DEVNAME(sc));
4569 		goto free;
4570 	}
4571 
4572 	out = mcx_cmdq_out(cqe);
4573 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4574 		printf("%s: modify rq failed (%x, %x)\n", DEVNAME(sc),
4575 		    out->cmd_status, betoh32(out->cmd_syndrome));
4576 		error = -1;
4577 		goto free;
4578 	}
4579 
4580 free:
4581 	mcx_dmamem_free(sc, &mxm);
4582 	return (error);
4583 }
4584 
4585 static int
4586 mcx_destroy_rq(struct mcx_softc *sc, struct mcx_rx *rx)
4587 {
4588 	struct mcx_cmdq_entry *cqe;
4589 	struct mcx_cmd_destroy_rq_in *in;
4590 	struct mcx_cmd_destroy_rq_out *out;
4591 	int error;
4592 	int token;
4593 
4594 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4595 	token = mcx_cmdq_token(sc);
4596 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), token);
4597 
4598 	in = mcx_cmdq_in(cqe);
4599 	in->cmd_opcode = htobe16(MCX_CMD_DESTROY_RQ);
4600 	in->cmd_op_mod = htobe16(0);
4601 	in->cmd_rqn = htobe32(rx->rx_rqn);
4602 
4603 	mcx_cmdq_post(sc, cqe, 0);
4604 	error = mcx_cmdq_poll(sc, cqe, 1000);
4605 	if (error != 0) {
4606 		printf("%s: destroy rq timeout\n", DEVNAME(sc));
4607 		return error;
4608 	}
4609 	if (mcx_cmdq_verify(cqe) != 0) {
4610 		printf("%s: destroy rq command corrupt\n", DEVNAME(sc));
4611 		return error;
4612 	}
4613 
4614 	out = mcx_cmdq_out(cqe);
4615 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4616 		printf("%s: destroy rq failed (%x, %x)\n", DEVNAME(sc),
4617 		    out->cmd_status, betoh32(out->cmd_syndrome));
4618 		return -1;
4619 	}
4620 
4621 	rx->rx_rqn = 0;
4622 	return 0;
4623 }
4624 
4625 static int
4626 mcx_create_tir_direct(struct mcx_softc *sc, struct mcx_rx *rx, int *tirn)
4627 {
4628 	struct mcx_cmdq_entry *cqe;
4629 	struct mcx_dmamem mxm;
4630 	struct mcx_cmd_create_tir_in *in;
4631 	struct mcx_cmd_create_tir_mb_in *mbin;
4632 	struct mcx_cmd_create_tir_out *out;
4633 	int error;
4634 	int token;
4635 
4636 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4637 	token = mcx_cmdq_token(sc);
4638 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
4639 	    sizeof(*out), token);
4640 
4641 	in = mcx_cmdq_in(cqe);
4642 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_TIR);
4643 	in->cmd_op_mod = htobe16(0);
4644 
4645 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
4646 	    &cqe->cq_input_ptr, token) != 0) {
4647 		printf("%s: unable to allocate create tir mailbox\n",
4648 		    DEVNAME(sc));
4649 		return (-1);
4650 	}
4651 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
4652 	/* leave disp_type = 0, so packets get sent to the inline rqn */
4653 	mbin->cmd_inline_rqn = htobe32(rx->rx_rqn);
4654 	mbin->cmd_tdomain = htobe32(sc->sc_tdomain);
4655 
4656 	mcx_cmdq_post(sc, cqe, 0);
4657 	error = mcx_cmdq_poll(sc, cqe, 1000);
4658 	if (error != 0) {
4659 		printf("%s: create tir timeout\n", DEVNAME(sc));
4660 		goto free;
4661 	}
4662 	if (mcx_cmdq_verify(cqe) != 0) {
4663 		printf("%s: create tir command corrupt\n", DEVNAME(sc));
4664 		goto free;
4665 	}
4666 
4667 	out = mcx_cmdq_out(cqe);
4668 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4669 		printf("%s: create tir failed (%x, %x)\n", DEVNAME(sc),
4670 		    out->cmd_status, betoh32(out->cmd_syndrome));
4671 		error = -1;
4672 		goto free;
4673 	}
4674 
4675 	*tirn = mcx_get_id(out->cmd_tirn);
4676 free:
4677 	mcx_dmamem_free(sc, &mxm);
4678 	return (error);
4679 }
4680 
4681 static int
4682 mcx_create_tir_indirect(struct mcx_softc *sc, int rqtn, uint32_t hash_sel,
4683     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 	mbin->cmd_disp_type = htobe32(MCX_TIR_CTX_DISP_TYPE_INDIRECT
4710 	    << MCX_TIR_CTX_DISP_TYPE_SHIFT);
4711 	mbin->cmd_indir_table = htobe32(rqtn);
4712 	mbin->cmd_tdomain = htobe32(sc->sc_tdomain |
4713 	    MCX_TIR_CTX_HASH_TOEPLITZ << MCX_TIR_CTX_HASH_SHIFT);
4714 	mbin->cmd_rx_hash_sel_outer = htobe32(hash_sel);
4715 	stoeplitz_to_key(&mbin->cmd_rx_hash_key,
4716 	    sizeof(mbin->cmd_rx_hash_key));
4717 
4718 	mcx_cmdq_post(sc, cqe, 0);
4719 	error = mcx_cmdq_poll(sc, cqe, 1000);
4720 	if (error != 0) {
4721 		printf("%s: create tir timeout\n", DEVNAME(sc));
4722 		goto free;
4723 	}
4724 	if (mcx_cmdq_verify(cqe) != 0) {
4725 		printf("%s: create tir command corrupt\n", DEVNAME(sc));
4726 		goto free;
4727 	}
4728 
4729 	out = mcx_cmdq_out(cqe);
4730 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4731 		printf("%s: create tir failed (%x, %x)\n", DEVNAME(sc),
4732 		    out->cmd_status, betoh32(out->cmd_syndrome));
4733 		error = -1;
4734 		goto free;
4735 	}
4736 
4737 	*tirn = mcx_get_id(out->cmd_tirn);
4738 free:
4739 	mcx_dmamem_free(sc, &mxm);
4740 	return (error);
4741 }
4742 
4743 static int
4744 mcx_destroy_tir(struct mcx_softc *sc, int tirn)
4745 {
4746 	struct mcx_cmdq_entry *cqe;
4747 	struct mcx_cmd_destroy_tir_in *in;
4748 	struct mcx_cmd_destroy_tir_out *out;
4749 	int error;
4750 	int token;
4751 
4752 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4753 	token = mcx_cmdq_token(sc);
4754 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), token);
4755 
4756 	in = mcx_cmdq_in(cqe);
4757 	in->cmd_opcode = htobe16(MCX_CMD_DESTROY_TIR);
4758 	in->cmd_op_mod = htobe16(0);
4759 	in->cmd_tirn = htobe32(tirn);
4760 
4761 	mcx_cmdq_post(sc, cqe, 0);
4762 	error = mcx_cmdq_poll(sc, cqe, 1000);
4763 	if (error != 0) {
4764 		printf("%s: destroy tir timeout\n", DEVNAME(sc));
4765 		return error;
4766 	}
4767 	if (mcx_cmdq_verify(cqe) != 0) {
4768 		printf("%s: destroy tir command corrupt\n", DEVNAME(sc));
4769 		return error;
4770 	}
4771 
4772 	out = mcx_cmdq_out(cqe);
4773 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4774 		printf("%s: destroy tir failed (%x, %x)\n", DEVNAME(sc),
4775 		    out->cmd_status, betoh32(out->cmd_syndrome));
4776 		return -1;
4777 	}
4778 
4779 	return (0);
4780 }
4781 
4782 static int
4783 mcx_create_sq(struct mcx_softc *sc, struct mcx_tx *tx, int uar, int db,
4784     int cqn)
4785 {
4786 	struct mcx_cmdq_entry *cqe;
4787 	struct mcx_dmamem mxm;
4788 	struct mcx_cmd_create_sq_in *in;
4789 	struct mcx_sq_ctx *mbin;
4790 	struct mcx_cmd_create_sq_out *out;
4791 	int error;
4792 	uint64_t *pas;
4793 	uint8_t *doorbell;
4794 	int insize, npages, paslen, token;
4795 
4796 	npages = howmany((1 << MCX_LOG_SQ_SIZE) * sizeof(struct mcx_sq_entry),
4797 	    MCX_PAGE_SIZE);
4798 	paslen = npages * sizeof(*pas);
4799 	insize = sizeof(struct mcx_sq_ctx) + paslen;
4800 
4801 	if (mcx_dmamem_alloc(sc, &tx->tx_sq_mem, npages * MCX_PAGE_SIZE,
4802 	    MCX_PAGE_SIZE) != 0) {
4803 		printf("%s: unable to allocate send queue memory\n",
4804 		    DEVNAME(sc));
4805 		return (-1);
4806 	}
4807 
4808 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4809 	token = mcx_cmdq_token(sc);
4810 	mcx_cmdq_init(sc, cqe, sizeof(*in) + insize + paslen, sizeof(*out),
4811 	    token);
4812 
4813 	in = mcx_cmdq_in(cqe);
4814 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_SQ);
4815 	in->cmd_op_mod = htobe16(0);
4816 
4817 	if (mcx_cmdq_mboxes_alloc(sc, &mxm,
4818 	    howmany(insize, MCX_CMDQ_MAILBOX_DATASIZE),
4819 	    &cqe->cq_input_ptr, token) != 0) {
4820 		printf("%s: unable to allocate create sq mailboxen\n",
4821 		    DEVNAME(sc));
4822 		error = -1;
4823 		goto free;
4824 	}
4825 	mbin = (struct mcx_sq_ctx *)
4826 	    (((char *)mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0))) + 0x10);
4827 	mbin->sq_flags = htobe32(MCX_SQ_CTX_RLKEY |
4828 	    (1 << MCX_SQ_CTX_MIN_WQE_INLINE_SHIFT));
4829 	mbin->sq_cqn = htobe32(cqn);
4830 	mbin->sq_tis_lst_sz = htobe32(1 << MCX_SQ_CTX_TIS_LST_SZ_SHIFT);
4831 	mbin->sq_tis_num = htobe32(sc->sc_tis);
4832 	mbin->sq_wq.wq_type = MCX_WQ_CTX_TYPE_CYCLIC;
4833 	mbin->sq_wq.wq_pd = htobe32(sc->sc_pd);
4834 	mbin->sq_wq.wq_uar_page = htobe32(uar);
4835 	mbin->sq_wq.wq_doorbell = htobe64(MCX_DMA_DVA(&sc->sc_doorbell_mem) +
4836 	    MCX_WQ_DOORBELL_BASE + (db * MCX_WQ_DOORBELL_STRIDE));
4837 	mbin->sq_wq.wq_log_stride = htobe16(MCX_LOG_SQ_ENTRY_SIZE);
4838 	mbin->sq_wq.wq_log_size = MCX_LOG_SQ_SIZE;
4839 
4840 	/* physical addresses follow the mailbox in data */
4841 	mcx_cmdq_mboxes_pas(&mxm, sizeof(*mbin) + 0x10,
4842 	    npages, &tx->tx_sq_mem);
4843 	mcx_cmdq_post(sc, cqe, 0);
4844 
4845 	error = mcx_cmdq_poll(sc, cqe, 1000);
4846 	if (error != 0) {
4847 		printf("%s: create sq timeout\n", DEVNAME(sc));
4848 		goto free;
4849 	}
4850 	if (mcx_cmdq_verify(cqe) != 0) {
4851 		printf("%s: create sq command corrupt\n", DEVNAME(sc));
4852 		goto free;
4853 	}
4854 
4855 	out = mcx_cmdq_out(cqe);
4856 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4857 		printf("%s: create sq failed (%x, %x)\n", DEVNAME(sc),
4858 		    out->cmd_status, betoh32(out->cmd_syndrome));
4859 		error = -1;
4860 		goto free;
4861 	}
4862 
4863 	tx->tx_uar = uar;
4864 	tx->tx_sqn = mcx_get_id(out->cmd_sqn);
4865 
4866 	doorbell = MCX_DMA_KVA(&sc->sc_doorbell_mem);
4867 	tx->tx_doorbell = (uint32_t *)(doorbell + MCX_WQ_DOORBELL_BASE +
4868 	    (db * MCX_WQ_DOORBELL_STRIDE) + 4);
4869 free:
4870 	mcx_dmamem_free(sc, &mxm);
4871 	return (error);
4872 }
4873 
4874 static int
4875 mcx_destroy_sq(struct mcx_softc *sc, struct mcx_tx *tx)
4876 {
4877 	struct mcx_cmdq_entry *cqe;
4878 	struct mcx_cmd_destroy_sq_in *in;
4879 	struct mcx_cmd_destroy_sq_out *out;
4880 	int error;
4881 	int token;
4882 
4883 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4884 	token = mcx_cmdq_token(sc);
4885 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), token);
4886 
4887 	in = mcx_cmdq_in(cqe);
4888 	in->cmd_opcode = htobe16(MCX_CMD_DESTROY_SQ);
4889 	in->cmd_op_mod = htobe16(0);
4890 	in->cmd_sqn = htobe32(tx->tx_sqn);
4891 
4892 	mcx_cmdq_post(sc, cqe, 0);
4893 	error = mcx_cmdq_poll(sc, cqe, 1000);
4894 	if (error != 0) {
4895 		printf("%s: destroy sq timeout\n", DEVNAME(sc));
4896 		return error;
4897 	}
4898 	if (mcx_cmdq_verify(cqe) != 0) {
4899 		printf("%s: destroy sq command corrupt\n", DEVNAME(sc));
4900 		return error;
4901 	}
4902 
4903 	out = mcx_cmdq_out(cqe);
4904 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4905 		printf("%s: destroy sq failed (%x, %x)\n", DEVNAME(sc),
4906 		    out->cmd_status, betoh32(out->cmd_syndrome));
4907 		return -1;
4908 	}
4909 
4910 	tx->tx_sqn = 0;
4911 	return 0;
4912 }
4913 
4914 static int
4915 mcx_ready_sq(struct mcx_softc *sc, struct mcx_tx *tx)
4916 {
4917 	struct mcx_cmdq_entry *cqe;
4918 	struct mcx_dmamem mxm;
4919 	struct mcx_cmd_modify_sq_in *in;
4920 	struct mcx_cmd_modify_sq_mb_in *mbin;
4921 	struct mcx_cmd_modify_sq_out *out;
4922 	int error;
4923 	int token;
4924 
4925 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4926 	token = mcx_cmdq_token(sc);
4927 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
4928 	    sizeof(*out), token);
4929 
4930 	in = mcx_cmdq_in(cqe);
4931 	in->cmd_opcode = htobe16(MCX_CMD_MODIFY_SQ);
4932 	in->cmd_op_mod = htobe16(0);
4933 	in->cmd_sq_state = htobe32((MCX_QUEUE_STATE_RST << 28) | tx->tx_sqn);
4934 
4935 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
4936 	    &cqe->cq_input_ptr, token) != 0) {
4937 		printf("%s: unable to allocate modify sq mailbox\n",
4938 		    DEVNAME(sc));
4939 		return (-1);
4940 	}
4941 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
4942 	mbin->cmd_sq_ctx.sq_flags = htobe32(
4943 	    MCX_QUEUE_STATE_RDY << MCX_SQ_CTX_STATE_SHIFT);
4944 
4945 	mcx_cmdq_mboxes_sign(&mxm, 1);
4946 	mcx_cmdq_post(sc, cqe, 0);
4947 	error = mcx_cmdq_poll(sc, cqe, 1000);
4948 	if (error != 0) {
4949 		printf("%s: modify sq timeout\n", DEVNAME(sc));
4950 		goto free;
4951 	}
4952 	if (mcx_cmdq_verify(cqe) != 0) {
4953 		printf("%s: modify sq command corrupt\n", DEVNAME(sc));
4954 		goto free;
4955 	}
4956 
4957 	out = mcx_cmdq_out(cqe);
4958 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4959 		printf("%s: modify sq failed (%x, %x)\n", DEVNAME(sc),
4960 		    out->cmd_status, betoh32(out->cmd_syndrome));
4961 		error = -1;
4962 		goto free;
4963 	}
4964 
4965 free:
4966 	mcx_dmamem_free(sc, &mxm);
4967 	return (error);
4968 }
4969 
4970 static int
4971 mcx_create_tis(struct mcx_softc *sc, int *tis)
4972 {
4973 	struct mcx_cmdq_entry *cqe;
4974 	struct mcx_dmamem mxm;
4975 	struct mcx_cmd_create_tis_in *in;
4976 	struct mcx_cmd_create_tis_mb_in *mbin;
4977 	struct mcx_cmd_create_tis_out *out;
4978 	int error;
4979 	int token;
4980 
4981 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4982 	token = mcx_cmdq_token(sc);
4983 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
4984 	    sizeof(*out), token);
4985 
4986 	in = mcx_cmdq_in(cqe);
4987 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_TIS);
4988 	in->cmd_op_mod = htobe16(0);
4989 
4990 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
4991 	    &cqe->cq_input_ptr, token) != 0) {
4992 		printf("%s: unable to allocate create tis mailbox\n",
4993 		    DEVNAME(sc));
4994 		return (-1);
4995 	}
4996 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
4997 	mbin->cmd_tdomain = htobe32(sc->sc_tdomain);
4998 
4999 	mcx_cmdq_mboxes_sign(&mxm, 1);
5000 	mcx_cmdq_post(sc, cqe, 0);
5001 	error = mcx_cmdq_poll(sc, cqe, 1000);
5002 	if (error != 0) {
5003 		printf("%s: create tis timeout\n", DEVNAME(sc));
5004 		goto free;
5005 	}
5006 	if (mcx_cmdq_verify(cqe) != 0) {
5007 		printf("%s: create tis command corrupt\n", DEVNAME(sc));
5008 		goto free;
5009 	}
5010 
5011 	out = mcx_cmdq_out(cqe);
5012 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5013 		printf("%s: create tis failed (%x, %x)\n", DEVNAME(sc),
5014 		    out->cmd_status, betoh32(out->cmd_syndrome));
5015 		error = -1;
5016 		goto free;
5017 	}
5018 
5019 	*tis = mcx_get_id(out->cmd_tisn);
5020 free:
5021 	mcx_dmamem_free(sc, &mxm);
5022 	return (error);
5023 }
5024 
5025 static int
5026 mcx_destroy_tis(struct mcx_softc *sc, int tis)
5027 {
5028 	struct mcx_cmdq_entry *cqe;
5029 	struct mcx_cmd_destroy_tis_in *in;
5030 	struct mcx_cmd_destroy_tis_out *out;
5031 	int error;
5032 	int token;
5033 
5034 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5035 	token = mcx_cmdq_token(sc);
5036 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), token);
5037 
5038 	in = mcx_cmdq_in(cqe);
5039 	in->cmd_opcode = htobe16(MCX_CMD_DESTROY_TIS);
5040 	in->cmd_op_mod = htobe16(0);
5041 	in->cmd_tisn = htobe32(tis);
5042 
5043 	mcx_cmdq_post(sc, cqe, 0);
5044 	error = mcx_cmdq_poll(sc, cqe, 1000);
5045 	if (error != 0) {
5046 		printf("%s: destroy tis timeout\n", DEVNAME(sc));
5047 		return error;
5048 	}
5049 	if (mcx_cmdq_verify(cqe) != 0) {
5050 		printf("%s: destroy tis command corrupt\n", DEVNAME(sc));
5051 		return error;
5052 	}
5053 
5054 	out = mcx_cmdq_out(cqe);
5055 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5056 		printf("%s: destroy tis failed (%x, %x)\n", DEVNAME(sc),
5057 		    out->cmd_status, betoh32(out->cmd_syndrome));
5058 		return -1;
5059 	}
5060 
5061 	return 0;
5062 }
5063 
5064 static int
5065 mcx_create_rqt(struct mcx_softc *sc, int size, int *rqns, int *rqt)
5066 {
5067 	struct mcx_cmdq_entry *cqe;
5068 	struct mcx_dmamem mxm;
5069 	struct mcx_cmd_create_rqt_in *in;
5070 	struct mcx_cmd_create_rqt_mb_in *mbin;
5071 	struct mcx_cmd_create_rqt_out *out;
5072 	struct mcx_rqt_ctx *rqt_ctx;
5073 	int *rqtn;
5074 	int error;
5075 	int token;
5076 	int i;
5077 
5078 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5079 	token = mcx_cmdq_token(sc);
5080 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin) +
5081 	    (size * sizeof(int)), sizeof(*out), token);
5082 
5083 	in = mcx_cmdq_in(cqe);
5084 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_RQT);
5085 	in->cmd_op_mod = htobe16(0);
5086 
5087 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
5088 	    &cqe->cq_input_ptr, token) != 0) {
5089 		printf("%s: unable to allocate create rqt mailbox\n",
5090 		    DEVNAME(sc));
5091 		return (-1);
5092 	}
5093 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5094 	rqt_ctx = &mbin->cmd_rqt;
5095 	rqt_ctx->cmd_rqt_max_size = htobe16(sc->sc_max_rqt_size);
5096 	rqt_ctx->cmd_rqt_actual_size = htobe16(size);
5097 
5098 	/* rqt list follows the rqt context */
5099 	rqtn = (int *)(rqt_ctx + 1);
5100 	for (i = 0; i < size; i++) {
5101 		rqtn[i] = htobe32(rqns[i]);
5102 	}
5103 
5104 	mcx_cmdq_mboxes_sign(&mxm, 1);
5105 	mcx_cmdq_post(sc, cqe, 0);
5106 	error = mcx_cmdq_poll(sc, cqe, 1000);
5107 	if (error != 0) {
5108 		printf("%s: create rqt timeout\n", DEVNAME(sc));
5109 		goto free;
5110 	}
5111 	if (mcx_cmdq_verify(cqe) != 0) {
5112 		printf("%s: create rqt command corrupt\n", DEVNAME(sc));
5113 		goto free;
5114 	}
5115 
5116 	out = mcx_cmdq_out(cqe);
5117 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5118 		printf("%s: create rqt failed (%x, %x)\n", DEVNAME(sc),
5119 		    out->cmd_status, betoh32(out->cmd_syndrome));
5120 		error = -1;
5121 		goto free;
5122 	}
5123 
5124 	*rqt = mcx_get_id(out->cmd_rqtn);
5125 	return (0);
5126 free:
5127 	mcx_dmamem_free(sc, &mxm);
5128 	return (error);
5129 }
5130 
5131 static int
5132 mcx_destroy_rqt(struct mcx_softc *sc, int rqt)
5133 {
5134 	struct mcx_cmdq_entry *cqe;
5135 	struct mcx_cmd_destroy_rqt_in *in;
5136 	struct mcx_cmd_destroy_rqt_out *out;
5137 	int error;
5138 	int token;
5139 
5140 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5141 	token = mcx_cmdq_token(sc);
5142 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), token);
5143 
5144 	in = mcx_cmdq_in(cqe);
5145 	in->cmd_opcode = htobe16(MCX_CMD_DESTROY_RQT);
5146 	in->cmd_op_mod = htobe16(0);
5147 	in->cmd_rqtn = htobe32(rqt);
5148 
5149 	mcx_cmdq_post(sc, cqe, 0);
5150 	error = mcx_cmdq_poll(sc, cqe, 1000);
5151 	if (error != 0) {
5152 		printf("%s: destroy rqt timeout\n", DEVNAME(sc));
5153 		return error;
5154 	}
5155 	if (mcx_cmdq_verify(cqe) != 0) {
5156 		printf("%s: destroy rqt command corrupt\n", DEVNAME(sc));
5157 		return error;
5158 	}
5159 
5160 	out = mcx_cmdq_out(cqe);
5161 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5162 		printf("%s: destroy rqt failed (%x, %x)\n", DEVNAME(sc),
5163 		    out->cmd_status, betoh32(out->cmd_syndrome));
5164 		return -1;
5165 	}
5166 
5167 	return 0;
5168 }
5169 
5170 #if 0
5171 static int
5172 mcx_alloc_flow_counter(struct mcx_softc *sc, int i)
5173 {
5174 	struct mcx_cmdq_entry *cqe;
5175 	struct mcx_cmd_alloc_flow_counter_in *in;
5176 	struct mcx_cmd_alloc_flow_counter_out *out;
5177 	int error;
5178 
5179 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5180 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
5181 
5182 	in = mcx_cmdq_in(cqe);
5183 	in->cmd_opcode = htobe16(MCX_CMD_ALLOC_FLOW_COUNTER);
5184 	in->cmd_op_mod = htobe16(0);
5185 
5186 	mcx_cmdq_post(sc, cqe, 0);
5187 
5188 	error = mcx_cmdq_poll(sc, cqe, 1000);
5189 	if (error != 0) {
5190 		printf("%s: alloc flow counter timeout\n", DEVNAME(sc));
5191 		return (-1);
5192 	}
5193 	if (mcx_cmdq_verify(cqe) != 0) {
5194 		printf("%s: alloc flow counter command corrupt\n", DEVNAME(sc));
5195 		return (-1);
5196 	}
5197 
5198 	out = (struct mcx_cmd_alloc_flow_counter_out *)cqe->cq_output_data;
5199 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5200 		printf("%s: alloc flow counter failed (%x)\n", DEVNAME(sc),
5201 		    out->cmd_status);
5202 		return (-1);
5203 	}
5204 
5205 	sc->sc_flow_counter_id[i]  = betoh16(out->cmd_flow_counter_id);
5206 	printf("flow counter id %d = %d\n", i, sc->sc_flow_counter_id[i]);
5207 
5208 	return (0);
5209 }
5210 #endif
5211 
5212 static int
5213 mcx_create_flow_table(struct mcx_softc *sc, int log_size, int level,
5214     int *flow_table_id)
5215 {
5216 	struct mcx_cmdq_entry *cqe;
5217 	struct mcx_dmamem mxm;
5218 	struct mcx_cmd_create_flow_table_in *in;
5219 	struct mcx_cmd_create_flow_table_mb_in *mbin;
5220 	struct mcx_cmd_create_flow_table_out *out;
5221 	int error;
5222 	int token;
5223 
5224 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5225 	token = mcx_cmdq_token(sc);
5226 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
5227 	    sizeof(*out), token);
5228 
5229 	in = mcx_cmdq_in(cqe);
5230 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_FLOW_TABLE);
5231 	in->cmd_op_mod = htobe16(0);
5232 
5233 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
5234 	    &cqe->cq_input_ptr, token) != 0) {
5235 		printf("%s: unable to allocate create flow table mailbox\n",
5236 		    DEVNAME(sc));
5237 		return (-1);
5238 	}
5239 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5240 	mbin->cmd_table_type = MCX_FLOW_TABLE_TYPE_RX;
5241 	mbin->cmd_ctx.ft_log_size = log_size;
5242 	mbin->cmd_ctx.ft_level = level;
5243 
5244 	mcx_cmdq_mboxes_sign(&mxm, 1);
5245 	mcx_cmdq_post(sc, cqe, 0);
5246 	error = mcx_cmdq_poll(sc, cqe, 1000);
5247 	if (error != 0) {
5248 		printf("%s: create flow table timeout\n", DEVNAME(sc));
5249 		goto free;
5250 	}
5251 	if (mcx_cmdq_verify(cqe) != 0) {
5252 		printf("%s: create flow table command corrupt\n", DEVNAME(sc));
5253 		goto free;
5254 	}
5255 
5256 	out = mcx_cmdq_out(cqe);
5257 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5258 		printf("%s: create flow table failed (%x, %x)\n", DEVNAME(sc),
5259 		    out->cmd_status, betoh32(out->cmd_syndrome));
5260 		error = -1;
5261 		goto free;
5262 	}
5263 
5264 	*flow_table_id = mcx_get_id(out->cmd_table_id);
5265 free:
5266 	mcx_dmamem_free(sc, &mxm);
5267 	return (error);
5268 }
5269 
5270 static int
5271 mcx_set_flow_table_root(struct mcx_softc *sc, int flow_table_id)
5272 {
5273 	struct mcx_cmdq_entry *cqe;
5274 	struct mcx_dmamem mxm;
5275 	struct mcx_cmd_set_flow_table_root_in *in;
5276 	struct mcx_cmd_set_flow_table_root_mb_in *mbin;
5277 	struct mcx_cmd_set_flow_table_root_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_SET_FLOW_TABLE_ROOT);
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 set flow table root 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_table_id = htobe32(flow_table_id);
5299 
5300 	mcx_cmdq_mboxes_sign(&mxm, 1);
5301 	mcx_cmdq_post(sc, cqe, 0);
5302 	error = mcx_cmdq_poll(sc, cqe, 1000);
5303 	if (error != 0) {
5304 		printf("%s: set flow table root timeout\n", DEVNAME(sc));
5305 		goto free;
5306 	}
5307 	if (mcx_cmdq_verify(cqe) != 0) {
5308 		printf("%s: set flow table root command corrupt\n",
5309 		    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: set flow table root failed (%x, %x)\n",
5316 		    DEVNAME(sc), out->cmd_status, betoh32(out->cmd_syndrome));
5317 		error = -1;
5318 		goto free;
5319 	}
5320 
5321 free:
5322 	mcx_dmamem_free(sc, &mxm);
5323 	return (error);
5324 }
5325 
5326 static int
5327 mcx_destroy_flow_table(struct mcx_softc *sc, int flow_table_id)
5328 {
5329 	struct mcx_cmdq_entry *cqe;
5330 	struct mcx_dmamem mxm;
5331 	struct mcx_cmd_destroy_flow_table_in *in;
5332 	struct mcx_cmd_destroy_flow_table_mb_in *mb;
5333 	struct mcx_cmd_destroy_flow_table_out *out;
5334 	int error;
5335 	int token;
5336 
5337 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5338 	token = mcx_cmdq_token(sc);
5339 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mb), sizeof(*out), token);
5340 
5341 	in = mcx_cmdq_in(cqe);
5342 	in->cmd_opcode = htobe16(MCX_CMD_DESTROY_FLOW_TABLE);
5343 	in->cmd_op_mod = htobe16(0);
5344 
5345 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
5346 	    &cqe->cq_input_ptr, token) != 0) {
5347 		printf("%s: unable to allocate destroy flow table mailbox\n",
5348 		    DEVNAME(sc));
5349 		return (-1);
5350 	}
5351 	mb = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5352 	mb->cmd_table_type = MCX_FLOW_TABLE_TYPE_RX;
5353 	mb->cmd_table_id = htobe32(flow_table_id);
5354 
5355 	mcx_cmdq_mboxes_sign(&mxm, 1);
5356 	mcx_cmdq_post(sc, cqe, 0);
5357 	error = mcx_cmdq_poll(sc, cqe, 1000);
5358 	if (error != 0) {
5359 		printf("%s: destroy flow table timeout\n", DEVNAME(sc));
5360 		goto free;
5361 	}
5362 	if (mcx_cmdq_verify(cqe) != 0) {
5363 		printf("%s: destroy flow table command corrupt\n",
5364 		    DEVNAME(sc));
5365 		goto free;
5366 	}
5367 
5368 	out = mcx_cmdq_out(cqe);
5369 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5370 		printf("%s: destroy flow table failed (%x, %x)\n", DEVNAME(sc),
5371 		    out->cmd_status, betoh32(out->cmd_syndrome));
5372 		error = -1;
5373 		goto free;
5374 	}
5375 
5376 free:
5377 	mcx_dmamem_free(sc, &mxm);
5378 	return (error);
5379 }
5380 
5381 
5382 static int
5383 mcx_create_flow_group(struct mcx_softc *sc, int flow_table_id, int group,
5384     int start, int size, int match_enable, struct mcx_flow_match *match)
5385 {
5386 	struct mcx_cmdq_entry *cqe;
5387 	struct mcx_dmamem mxm;
5388 	struct mcx_cmd_create_flow_group_in *in;
5389 	struct mcx_cmd_create_flow_group_mb_in *mbin;
5390 	struct mcx_cmd_create_flow_group_out *out;
5391 	struct mcx_flow_group *mfg;
5392 	int error;
5393 	int token;
5394 
5395 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5396 	token = mcx_cmdq_token(sc);
5397 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin), sizeof(*out),
5398 	    token);
5399 
5400 	in = mcx_cmdq_in(cqe);
5401 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_FLOW_GROUP);
5402 	in->cmd_op_mod = htobe16(0);
5403 
5404 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2, &cqe->cq_input_ptr, token)
5405 	    != 0) {
5406 		printf("%s: unable to allocate create flow group mailbox\n",
5407 		    DEVNAME(sc));
5408 		return (-1);
5409 	}
5410 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5411 	mbin->cmd_table_type = MCX_FLOW_TABLE_TYPE_RX;
5412 	mbin->cmd_table_id = htobe32(flow_table_id);
5413 	mbin->cmd_start_flow_index = htobe32(start);
5414 	mbin->cmd_end_flow_index = htobe32(start + (size - 1));
5415 
5416 	mbin->cmd_match_criteria_enable = match_enable;
5417 	memcpy(&mbin->cmd_match_criteria, match, sizeof(*match));
5418 
5419 	mcx_cmdq_mboxes_sign(&mxm, 2);
5420 	mcx_cmdq_post(sc, cqe, 0);
5421 	error = mcx_cmdq_poll(sc, cqe, 1000);
5422 	if (error != 0) {
5423 		printf("%s: create flow group timeout\n", DEVNAME(sc));
5424 		goto free;
5425 	}
5426 	if (mcx_cmdq_verify(cqe) != 0) {
5427 		printf("%s: create flow group command corrupt\n", DEVNAME(sc));
5428 		goto free;
5429 	}
5430 
5431 	out = mcx_cmdq_out(cqe);
5432 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5433 		printf("%s: create flow group failed (%x, %x)\n", DEVNAME(sc),
5434 		    out->cmd_status, betoh32(out->cmd_syndrome));
5435 		error = -1;
5436 		goto free;
5437 	}
5438 
5439 	mfg = &sc->sc_flow_group[group];
5440 	mfg->g_id = mcx_get_id(out->cmd_group_id);
5441 	mfg->g_table = flow_table_id;
5442 	mfg->g_start = start;
5443 	mfg->g_size = size;
5444 
5445 free:
5446 	mcx_dmamem_free(sc, &mxm);
5447 	return (error);
5448 }
5449 
5450 static int
5451 mcx_destroy_flow_group(struct mcx_softc *sc, int group)
5452 {
5453 	struct mcx_cmdq_entry *cqe;
5454 	struct mcx_dmamem mxm;
5455 	struct mcx_cmd_destroy_flow_group_in *in;
5456 	struct mcx_cmd_destroy_flow_group_mb_in *mb;
5457 	struct mcx_cmd_destroy_flow_group_out *out;
5458 	struct mcx_flow_group *mfg;
5459 	int error;
5460 	int token;
5461 
5462 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5463 	token = mcx_cmdq_token(sc);
5464 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mb), sizeof(*out), token);
5465 
5466 	in = mcx_cmdq_in(cqe);
5467 	in->cmd_opcode = htobe16(MCX_CMD_DESTROY_FLOW_GROUP);
5468 	in->cmd_op_mod = htobe16(0);
5469 
5470 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2,
5471 	    &cqe->cq_input_ptr, token) != 0) {
5472 		printf("%s: unable to allocate destroy flow group mailbox\n",
5473 		    DEVNAME(sc));
5474 		return (-1);
5475 	}
5476 	mb = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5477 	mb->cmd_table_type = MCX_FLOW_TABLE_TYPE_RX;
5478 	mfg = &sc->sc_flow_group[group];
5479 	mb->cmd_table_id = htobe32(mfg->g_table);
5480 	mb->cmd_group_id = htobe32(mfg->g_id);
5481 
5482 	mcx_cmdq_mboxes_sign(&mxm, 2);
5483 	mcx_cmdq_post(sc, cqe, 0);
5484 	error = mcx_cmdq_poll(sc, cqe, 1000);
5485 	if (error != 0) {
5486 		printf("%s: destroy flow group timeout\n", DEVNAME(sc));
5487 		goto free;
5488 	}
5489 	if (mcx_cmdq_verify(cqe) != 0) {
5490 		printf("%s: destroy flow group command corrupt\n", DEVNAME(sc));
5491 		goto free;
5492 	}
5493 
5494 	out = mcx_cmdq_out(cqe);
5495 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5496 		printf("%s: destroy flow group failed (%x, %x)\n", DEVNAME(sc),
5497 		    out->cmd_status, betoh32(out->cmd_syndrome));
5498 		error = -1;
5499 		goto free;
5500 	}
5501 
5502 	mfg->g_id = -1;
5503 	mfg->g_table = -1;
5504 	mfg->g_size = 0;
5505 	mfg->g_start = 0;
5506 free:
5507 	mcx_dmamem_free(sc, &mxm);
5508 	return (error);
5509 }
5510 
5511 static int
5512 mcx_set_flow_table_entry_mac(struct mcx_softc *sc, int group, int index,
5513     uint8_t *macaddr, uint32_t dest)
5514 {
5515 	struct mcx_cmdq_entry *cqe;
5516 	struct mcx_dmamem mxm;
5517 	struct mcx_cmd_set_flow_table_entry_in *in;
5518 	struct mcx_cmd_set_flow_table_entry_mb_in *mbin;
5519 	struct mcx_cmd_set_flow_table_entry_out *out;
5520 	struct mcx_flow_group *mfg;
5521 	uint32_t *pdest;
5522 	int error;
5523 	int token;
5524 
5525 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5526 	token = mcx_cmdq_token(sc);
5527 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin) + sizeof(*pdest),
5528 	    sizeof(*out), token);
5529 
5530 	in = mcx_cmdq_in(cqe);
5531 	in->cmd_opcode = htobe16(MCX_CMD_SET_FLOW_TABLE_ENTRY);
5532 	in->cmd_op_mod = htobe16(0);
5533 
5534 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2, &cqe->cq_input_ptr, token)
5535 	    != 0) {
5536 		printf("%s: unable to allocate set flow table entry mailbox\n",
5537 		    DEVNAME(sc));
5538 		return (-1);
5539 	}
5540 
5541 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5542 	mbin->cmd_table_type = MCX_FLOW_TABLE_TYPE_RX;
5543 
5544 	mfg = &sc->sc_flow_group[group];
5545 	mbin->cmd_table_id = htobe32(mfg->g_table);
5546 	mbin->cmd_flow_index = htobe32(mfg->g_start + index);
5547 	mbin->cmd_flow_ctx.fc_group_id = htobe32(mfg->g_id);
5548 
5549 	/* flow context ends at offset 0x330, 0x130 into the second mbox */
5550 	pdest = (uint32_t *)
5551 	    (((char *)mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 1))) + 0x130);
5552 	mbin->cmd_flow_ctx.fc_action = htobe32(MCX_FLOW_CONTEXT_ACTION_FORWARD);
5553 	mbin->cmd_flow_ctx.fc_dest_list_size = htobe32(1);
5554 	*pdest = htobe32(dest);
5555 
5556 	/* the only thing we match on at the moment is the dest mac address */
5557 	if (macaddr != NULL) {
5558 		memcpy(mbin->cmd_flow_ctx.fc_match_value.mc_dest_mac, macaddr,
5559 		    ETHER_ADDR_LEN);
5560 	}
5561 
5562 	mcx_cmdq_mboxes_sign(&mxm, 2);
5563 	mcx_cmdq_post(sc, cqe, 0);
5564 	error = mcx_cmdq_poll(sc, cqe, 1000);
5565 	if (error != 0) {
5566 		printf("%s: set flow table entry timeout\n", DEVNAME(sc));
5567 		goto free;
5568 	}
5569 	if (mcx_cmdq_verify(cqe) != 0) {
5570 		printf("%s: set flow table entry command corrupt\n",
5571 		    DEVNAME(sc));
5572 		goto free;
5573 	}
5574 
5575 	out = mcx_cmdq_out(cqe);
5576 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5577 		printf("%s: set flow table entry failed (%x, %x)\n",
5578 		    DEVNAME(sc), out->cmd_status, betoh32(out->cmd_syndrome));
5579 		error = -1;
5580 		goto free;
5581 	}
5582 
5583 free:
5584 	mcx_dmamem_free(sc, &mxm);
5585 	return (error);
5586 }
5587 
5588 static int
5589 mcx_set_flow_table_entry_proto(struct mcx_softc *sc, int group, int index,
5590     int ethertype, int ip_proto, uint32_t dest)
5591 {
5592 	struct mcx_cmdq_entry *cqe;
5593 	struct mcx_dmamem mxm;
5594 	struct mcx_cmd_set_flow_table_entry_in *in;
5595 	struct mcx_cmd_set_flow_table_entry_mb_in *mbin;
5596 	struct mcx_cmd_set_flow_table_entry_out *out;
5597 	struct mcx_flow_group *mfg;
5598 	uint32_t *pdest;
5599 	int error;
5600 	int token;
5601 
5602 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5603 	token = mcx_cmdq_token(sc);
5604 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin) + sizeof(*pdest),
5605 	    sizeof(*out), token);
5606 
5607 	in = mcx_cmdq_in(cqe);
5608 	in->cmd_opcode = htobe16(MCX_CMD_SET_FLOW_TABLE_ENTRY);
5609 	in->cmd_op_mod = htobe16(0);
5610 
5611 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2, &cqe->cq_input_ptr, token)
5612 	    != 0) {
5613 		printf("%s: unable to allocate set flow table entry mailbox\n",
5614 		    DEVNAME(sc));
5615 		return (-1);
5616 	}
5617 
5618 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5619 	mbin->cmd_table_type = MCX_FLOW_TABLE_TYPE_RX;
5620 
5621 	mfg = &sc->sc_flow_group[group];
5622 	mbin->cmd_table_id = htobe32(mfg->g_table);
5623 	mbin->cmd_flow_index = htobe32(mfg->g_start + index);
5624 	mbin->cmd_flow_ctx.fc_group_id = htobe32(mfg->g_id);
5625 
5626 	/* flow context ends at offset 0x330, 0x130 into the second mbox */
5627 	pdest = (uint32_t *)
5628 	    (((char *)mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 1))) + 0x130);
5629 	mbin->cmd_flow_ctx.fc_action = htobe32(MCX_FLOW_CONTEXT_ACTION_FORWARD);
5630 	mbin->cmd_flow_ctx.fc_dest_list_size = htobe32(1);
5631 	*pdest = htobe32(dest);
5632 
5633 	mbin->cmd_flow_ctx.fc_match_value.mc_ethertype = htobe16(ethertype);
5634 	mbin->cmd_flow_ctx.fc_match_value.mc_ip_proto = ip_proto;
5635 
5636 	mcx_cmdq_mboxes_sign(&mxm, 2);
5637 	mcx_cmdq_post(sc, cqe, 0);
5638 	error = mcx_cmdq_poll(sc, cqe, 1000);
5639 	if (error != 0) {
5640 		printf("%s: set flow table entry timeout\n", DEVNAME(sc));
5641 		goto free;
5642 	}
5643 	if (mcx_cmdq_verify(cqe) != 0) {
5644 		printf("%s: set flow table entry command corrupt\n",
5645 		    DEVNAME(sc));
5646 		goto free;
5647 	}
5648 
5649 	out = mcx_cmdq_out(cqe);
5650 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5651 		printf("%s: set flow table entry failed (%x, %x)\n",
5652 		    DEVNAME(sc), out->cmd_status, betoh32(out->cmd_syndrome));
5653 		error = -1;
5654 		goto free;
5655 	}
5656 
5657 free:
5658 	mcx_dmamem_free(sc, &mxm);
5659 	return (error);
5660 }
5661 
5662 static int
5663 mcx_delete_flow_table_entry(struct mcx_softc *sc, int group, int index)
5664 {
5665 	struct mcx_cmdq_entry *cqe;
5666 	struct mcx_dmamem mxm;
5667 	struct mcx_cmd_delete_flow_table_entry_in *in;
5668 	struct mcx_cmd_delete_flow_table_entry_mb_in *mbin;
5669 	struct mcx_cmd_delete_flow_table_entry_out *out;
5670 	struct mcx_flow_group *mfg;
5671 	int error;
5672 	int token;
5673 
5674 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5675 	token = mcx_cmdq_token(sc);
5676 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin), sizeof(*out),
5677 	    token);
5678 
5679 	in = mcx_cmdq_in(cqe);
5680 	in->cmd_opcode = htobe16(MCX_CMD_DELETE_FLOW_TABLE_ENTRY);
5681 	in->cmd_op_mod = htobe16(0);
5682 
5683 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2,
5684 	    &cqe->cq_input_ptr, token) != 0) {
5685 		printf("%s: unable to allocate "
5686 		    "delete flow table entry mailbox\n", DEVNAME(sc));
5687 		return (-1);
5688 	}
5689 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5690 	mbin->cmd_table_type = MCX_FLOW_TABLE_TYPE_RX;
5691 
5692 	mfg = &sc->sc_flow_group[group];
5693 	mbin->cmd_table_id = htobe32(mfg->g_table);
5694 	mbin->cmd_flow_index = htobe32(mfg->g_start + index);
5695 
5696 	mcx_cmdq_mboxes_sign(&mxm, 2);
5697 	mcx_cmdq_post(sc, cqe, 0);
5698 	error = mcx_cmdq_poll(sc, cqe, 1000);
5699 	if (error != 0) {
5700 		printf("%s: delete flow table entry timeout\n", DEVNAME(sc));
5701 		goto free;
5702 	}
5703 	if (mcx_cmdq_verify(cqe) != 0) {
5704 		printf("%s: delete flow table entry command corrupt\n",
5705 		    DEVNAME(sc));
5706 		goto free;
5707 	}
5708 
5709 	out = mcx_cmdq_out(cqe);
5710 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5711 		printf("%s: delete flow table entry %d:%d failed (%x, %x)\n",
5712 		    DEVNAME(sc), group, index, out->cmd_status,
5713 		    betoh32(out->cmd_syndrome));
5714 		error = -1;
5715 		goto free;
5716 	}
5717 
5718 free:
5719 	mcx_dmamem_free(sc, &mxm);
5720 	return (error);
5721 }
5722 
5723 #if 0
5724 int
5725 mcx_dump_flow_table(struct mcx_softc *sc, int flow_table_id)
5726 {
5727 	struct mcx_dmamem mxm;
5728 	struct mcx_cmdq_entry *cqe;
5729 	struct mcx_cmd_query_flow_table_in *in;
5730 	struct mcx_cmd_query_flow_table_mb_in *mbin;
5731 	struct mcx_cmd_query_flow_table_out *out;
5732 	struct mcx_cmd_query_flow_table_mb_out *mbout;
5733 	uint8_t token = mcx_cmdq_token(sc);
5734 	int error;
5735 	int i;
5736 	uint8_t *dump;
5737 
5738 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5739 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
5740 	    sizeof(*out) + sizeof(*mbout) + 16, token);
5741 
5742 	in = mcx_cmdq_in(cqe);
5743 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_FLOW_TABLE);
5744 	in->cmd_op_mod = htobe16(0);
5745 
5746 	CTASSERT(sizeof(*mbin) <= MCX_CMDQ_MAILBOX_DATASIZE);
5747 	CTASSERT(sizeof(*mbout) <= MCX_CMDQ_MAILBOX_DATASIZE);
5748 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2,
5749 	    &cqe->cq_output_ptr, token) != 0) {
5750 		printf(", unable to allocate query flow table mailboxes\n");
5751 		return (-1);
5752 	}
5753 	cqe->cq_input_ptr = cqe->cq_output_ptr;
5754 
5755 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5756 	mbin->cmd_table_type = 0;
5757 	mbin->cmd_table_id = htobe32(flow_table_id);
5758 
5759 	mcx_cmdq_mboxes_sign(&mxm, 1);
5760 
5761 	mcx_cmdq_post(sc, cqe, 0);
5762 	error = mcx_cmdq_poll(sc, cqe, 1000);
5763 	if (error != 0) {
5764 		printf("%s: query flow table timeout\n", DEVNAME(sc));
5765 		goto free;
5766 	}
5767 	error = mcx_cmdq_verify(cqe);
5768 	if (error != 0) {
5769 		printf("%s: query flow table reply corrupt\n", DEVNAME(sc));
5770 		goto free;
5771 	}
5772 
5773 	out = mcx_cmdq_out(cqe);
5774 	switch (out->cmd_status) {
5775 	case MCX_CQ_STATUS_OK:
5776 		break;
5777 	default:
5778 		printf("%s: query flow table failed (%x/%x)\n", DEVNAME(sc),
5779 		    out->cmd_status, betoh32(out->cmd_syndrome));
5780 		error = -1;
5781 		goto free;
5782 	}
5783 
5784         mbout = (struct mcx_cmd_query_flow_table_mb_out *)
5785 	    (mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)));
5786 	dump = (uint8_t *)mbout + 8;
5787 	for (i = 0; i < sizeof(struct mcx_flow_table_ctx); i++) {
5788 		printf("%.2x ", dump[i]);
5789 		if (i % 16 == 15)
5790 			printf("\n");
5791 	}
5792 free:
5793 	mcx_cq_mboxes_free(sc, &mxm);
5794 	return (error);
5795 }
5796 int
5797 mcx_dump_flow_table_entry(struct mcx_softc *sc, int flow_table_id, int index)
5798 {
5799 	struct mcx_dmamem mxm;
5800 	struct mcx_cmdq_entry *cqe;
5801 	struct mcx_cmd_query_flow_table_entry_in *in;
5802 	struct mcx_cmd_query_flow_table_entry_mb_in *mbin;
5803 	struct mcx_cmd_query_flow_table_entry_out *out;
5804 	struct mcx_cmd_query_flow_table_entry_mb_out *mbout;
5805 	uint8_t token = mcx_cmdq_token(sc);
5806 	int error;
5807 	int i;
5808 	uint8_t *dump;
5809 
5810 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5811 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
5812 	    sizeof(*out) + sizeof(*mbout) + 16, token);
5813 
5814 	in = mcx_cmdq_in(cqe);
5815 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_FLOW_TABLE_ENTRY);
5816 	in->cmd_op_mod = htobe16(0);
5817 
5818 	CTASSERT(sizeof(*mbin) <= MCX_CMDQ_MAILBOX_DATASIZE);
5819 	CTASSERT(sizeof(*mbout) <= MCX_CMDQ_MAILBOX_DATASIZE*2);
5820 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2,
5821 	    &cqe->cq_output_ptr, token) != 0) {
5822 		printf(", unable to allocate "
5823 		    "query flow table entry mailboxes\n");
5824 		return (-1);
5825 	}
5826 	cqe->cq_input_ptr = cqe->cq_output_ptr;
5827 
5828 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5829 	mbin->cmd_table_type = 0;
5830 	mbin->cmd_table_id = htobe32(flow_table_id);
5831 	mbin->cmd_flow_index = htobe32(index);
5832 
5833 	mcx_cmdq_mboxes_sign(&mxm, 1);
5834 
5835 	mcx_cmdq_post(sc, cqe, 0);
5836 	error = mcx_cmdq_poll(sc, cqe, 1000);
5837 	if (error != 0) {
5838 		printf("%s: query flow table entry timeout\n", DEVNAME(sc));
5839 		goto free;
5840 	}
5841 	error = mcx_cmdq_verify(cqe);
5842 	if (error != 0) {
5843 		printf("%s: query flow table entry reply corrupt\n",
5844 		    DEVNAME(sc));
5845 		goto free;
5846 	}
5847 
5848 	out = mcx_cmdq_out(cqe);
5849 	switch (out->cmd_status) {
5850 	case MCX_CQ_STATUS_OK:
5851 		break;
5852 	default:
5853 		printf("%s: query flow table entry failed (%x/%x)\n",
5854 		    DEVNAME(sc), out->cmd_status, betoh32(out->cmd_syndrome));
5855 		error = -1;
5856 		goto free;
5857 	}
5858 
5859         mbout = (struct mcx_cmd_query_flow_table_entry_mb_out *)
5860 	    (mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)));
5861 	dump = (uint8_t *)mbout;
5862 	for (i = 0; i < MCX_CMDQ_MAILBOX_DATASIZE; i++) {
5863 		printf("%.2x ", dump[i]);
5864 		if (i % 16 == 15)
5865 			printf("\n");
5866 	}
5867 
5868 free:
5869 	mcx_cq_mboxes_free(sc, &mxm);
5870 	return (error);
5871 }
5872 
5873 int
5874 mcx_dump_flow_group(struct mcx_softc *sc, int flow_table_id)
5875 {
5876 	struct mcx_dmamem mxm;
5877 	struct mcx_cmdq_entry *cqe;
5878 	struct mcx_cmd_query_flow_group_in *in;
5879 	struct mcx_cmd_query_flow_group_mb_in *mbin;
5880 	struct mcx_cmd_query_flow_group_out *out;
5881 	struct mcx_cmd_query_flow_group_mb_out *mbout;
5882 	uint8_t token = mcx_cmdq_token(sc);
5883 	int error;
5884 	int i;
5885 	uint8_t *dump;
5886 
5887 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5888 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
5889 	    sizeof(*out) + sizeof(*mbout) + 16, token);
5890 
5891 	in = mcx_cmdq_in(cqe);
5892 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_FLOW_GROUP);
5893 	in->cmd_op_mod = htobe16(0);
5894 
5895 	CTASSERT(sizeof(*mbin) <= MCX_CMDQ_MAILBOX_DATASIZE);
5896 	CTASSERT(sizeof(*mbout) <= MCX_CMDQ_MAILBOX_DATASIZE*2);
5897 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2,
5898 	    &cqe->cq_output_ptr, token) != 0) {
5899 		printf(", unable to allocate query flow group mailboxes\n");
5900 		return (-1);
5901 	}
5902 	cqe->cq_input_ptr = cqe->cq_output_ptr;
5903 
5904 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5905 	mbin->cmd_table_type = 0;
5906 	mbin->cmd_table_id = htobe32(flow_table_id);
5907 	mbin->cmd_group_id = htobe32(sc->sc_flow_group_id);
5908 
5909 	mcx_cmdq_mboxes_sign(&mxm, 1);
5910 
5911 	mcx_cmdq_post(sc, cqe, 0);
5912 	error = mcx_cmdq_poll(sc, cqe, 1000);
5913 	if (error != 0) {
5914 		printf("%s: query flow group timeout\n", DEVNAME(sc));
5915 		goto free;
5916 	}
5917 	error = mcx_cmdq_verify(cqe);
5918 	if (error != 0) {
5919 		printf("%s: query flow group reply corrupt\n", DEVNAME(sc));
5920 		goto free;
5921 	}
5922 
5923 	out = mcx_cmdq_out(cqe);
5924 	switch (out->cmd_status) {
5925 	case MCX_CQ_STATUS_OK:
5926 		break;
5927 	default:
5928 		printf("%s: query flow group failed (%x/%x)\n", DEVNAME(sc),
5929 		    out->cmd_status, betoh32(out->cmd_syndrome));
5930 		error = -1;
5931 		goto free;
5932 	}
5933 
5934         mbout = (struct mcx_cmd_query_flow_group_mb_out *)
5935 	    (mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)));
5936 	dump = (uint8_t *)mbout;
5937 	for (i = 0; i < MCX_CMDQ_MAILBOX_DATASIZE; i++) {
5938 		printf("%.2x ", dump[i]);
5939 		if (i % 16 == 15)
5940 			printf("\n");
5941 	}
5942 	dump = (uint8_t *)(mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 1)));
5943 	for (i = 0; i < MCX_CMDQ_MAILBOX_DATASIZE; i++) {
5944 		printf("%.2x ", dump[i]);
5945 		if (i % 16 == 15)
5946 			printf("\n");
5947 	}
5948 
5949 free:
5950 	mcx_cq_mboxes_free(sc, &mxm);
5951 	return (error);
5952 }
5953 
5954 int
5955 mcx_dump_rq(struct mcx_softc *sc)
5956 {
5957 	struct mcx_dmamem mxm;
5958 	struct mcx_cmdq_entry *cqe;
5959 	struct mcx_cmd_query_rq_in *in;
5960 	struct mcx_cmd_query_rq_out *out;
5961 	struct mcx_cmd_query_rq_mb_out *mbout;
5962 	uint8_t token = mcx_cmdq_token(sc);
5963 	int error;
5964 
5965 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5966 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out) + sizeof(*mbout) + 16,
5967 	    token);
5968 
5969 	in = mcx_cmdq_in(cqe);
5970 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_RQ);
5971 	in->cmd_op_mod = htobe16(0);
5972 	in->cmd_rqn = htobe32(sc->sc_rqn);
5973 
5974 	CTASSERT(sizeof(*mbout) <= MCX_CMDQ_MAILBOX_DATASIZE*2);
5975 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2,
5976 	    &cqe->cq_output_ptr, token) != 0) {
5977 		printf(", unable to allocate query flow group mailboxes\n");
5978 		return (-1);
5979 	}
5980 
5981 	mcx_cmdq_mboxes_sign(&mxm, 1);
5982 
5983 	mcx_cmdq_post(sc, cqe, 0);
5984 	error = mcx_cmdq_poll(sc, cqe, 1000);
5985 	if (error != 0) {
5986 		printf("%s: query rq timeout\n", DEVNAME(sc));
5987 		goto free;
5988 	}
5989 	error = mcx_cmdq_verify(cqe);
5990 	if (error != 0) {
5991 		printf("%s: query rq reply corrupt\n", DEVNAME(sc));
5992 		goto free;
5993 	}
5994 
5995 	out = mcx_cmdq_out(cqe);
5996 	switch (out->cmd_status) {
5997 	case MCX_CQ_STATUS_OK:
5998 		break;
5999 	default:
6000 		printf("%s: query rq failed (%x/%x)\n", DEVNAME(sc),
6001 		    out->cmd_status, betoh32(out->cmd_syndrome));
6002 		error = -1;
6003 		goto free;
6004 	}
6005 
6006         mbout = (struct mcx_cmd_query_rq_mb_out *)
6007 	    (mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)));
6008 	printf("%s: rq: state %d, ui %d, cqn %d, s/s %d/%d/%d, hw %d, sw %d\n",
6009 	    DEVNAME(sc),
6010 	    (betoh32(mbout->cmd_ctx.rq_flags) >> MCX_RQ_CTX_STATE_SHIFT) & 0x0f,
6011 	    betoh32(mbout->cmd_ctx.rq_user_index),
6012 	    betoh32(mbout->cmd_ctx.rq_cqn),
6013 	    betoh16(mbout->cmd_ctx.rq_wq.wq_log_stride),
6014 	    mbout->cmd_ctx.rq_wq.wq_log_page_sz,
6015 	    mbout->cmd_ctx.rq_wq.wq_log_size,
6016 	    betoh32(mbout->cmd_ctx.rq_wq.wq_hw_counter),
6017 	    betoh32(mbout->cmd_ctx.rq_wq.wq_sw_counter));
6018 
6019 free:
6020 	mcx_cq_mboxes_free(sc, &mxm);
6021 	return (error);
6022 }
6023 
6024 int
6025 mcx_dump_sq(struct mcx_softc *sc)
6026 {
6027 	struct mcx_dmamem mxm;
6028 	struct mcx_cmdq_entry *cqe;
6029 	struct mcx_cmd_query_sq_in *in;
6030 	struct mcx_cmd_query_sq_out *out;
6031 	struct mcx_cmd_query_sq_mb_out *mbout;
6032 	uint8_t token = mcx_cmdq_token(sc);
6033 	int error;
6034 	int i;
6035 	uint8_t *dump;
6036 
6037 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
6038 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out) + sizeof(*mbout) + 16,
6039 	    token);
6040 
6041 	in = mcx_cmdq_in(cqe);
6042 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_SQ);
6043 	in->cmd_op_mod = htobe16(0);
6044 	in->cmd_sqn = htobe32(tx->tx_sqn);
6045 
6046 	CTASSERT(sizeof(*mbout) <= MCX_CMDQ_MAILBOX_DATASIZE*2);
6047 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2,
6048 	    &cqe->cq_output_ptr, token) != 0) {
6049 		printf(", unable to allocate query sq mailboxes\n");
6050 		return (-1);
6051 	}
6052 
6053 	mcx_cmdq_mboxes_sign(&mxm, 1);
6054 
6055 	mcx_cmdq_post(sc, cqe, 0);
6056 	error = mcx_cmdq_poll(sc, cqe, 1000);
6057 	if (error != 0) {
6058 		printf("%s: query sq timeout\n", DEVNAME(sc));
6059 		goto free;
6060 	}
6061 	error = mcx_cmdq_verify(cqe);
6062 	if (error != 0) {
6063 		printf("%s: query sq reply corrupt\n", DEVNAME(sc));
6064 		goto free;
6065 	}
6066 
6067 	out = mcx_cmdq_out(cqe);
6068 	switch (out->cmd_status) {
6069 	case MCX_CQ_STATUS_OK:
6070 		break;
6071 	default:
6072 		printf("%s: query sq failed (%x/%x)\n", DEVNAME(sc),
6073 		    out->cmd_status, betoh32(out->cmd_syndrome));
6074 		error = -1;
6075 		goto free;
6076 	}
6077 
6078         mbout = (struct mcx_cmd_query_sq_mb_out *)
6079 	    (mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)));
6080 /*
6081 	printf("%s: rq: state %d, ui %d, cqn %d, s/s %d/%d/%d, hw %d, sw %d\n",
6082 	    DEVNAME(sc),
6083 	    (betoh32(mbout->cmd_ctx.rq_flags) >> MCX_RQ_CTX_STATE_SHIFT) & 0x0f,
6084 	    betoh32(mbout->cmd_ctx.rq_user_index),
6085 	    betoh32(mbout->cmd_ctx.rq_cqn),
6086 	    betoh16(mbout->cmd_ctx.rq_wq.wq_log_stride),
6087 	    mbout->cmd_ctx.rq_wq.wq_log_page_sz,
6088 	    mbout->cmd_ctx.rq_wq.wq_log_size,
6089 	    betoh32(mbout->cmd_ctx.rq_wq.wq_hw_counter),
6090 	    betoh32(mbout->cmd_ctx.rq_wq.wq_sw_counter));
6091 */
6092 	dump = (uint8_t *)mbout;
6093 	for (i = 0; i < MCX_CMDQ_MAILBOX_DATASIZE; i++) {
6094 		printf("%.2x ", dump[i]);
6095 		if (i % 16 == 15)
6096 			printf("\n");
6097 	}
6098 
6099 free:
6100 	mcx_cq_mboxes_free(sc, &mxm);
6101 	return (error);
6102 }
6103 
6104 static int
6105 mcx_dump_counters(struct mcx_softc *sc)
6106 {
6107 	struct mcx_dmamem mxm;
6108 	struct mcx_cmdq_entry *cqe;
6109 	struct mcx_cmd_query_vport_counters_in *in;
6110 	struct mcx_cmd_query_vport_counters_mb_in *mbin;
6111 	struct mcx_cmd_query_vport_counters_out *out;
6112 	struct mcx_nic_vport_counters *counters;
6113 	int error, token;
6114 
6115 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
6116 	token = mcx_cmdq_token(sc);
6117 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
6118 	    sizeof(*out) + sizeof(*counters), token);
6119 
6120 	in = mcx_cmdq_in(cqe);
6121 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_VPORT_COUNTERS);
6122 	in->cmd_op_mod = htobe16(0);
6123 
6124 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
6125 	    &cqe->cq_output_ptr, token) != 0) {
6126 		printf(", unable to allocate "
6127 		    "query nic vport counters mailboxen\n");
6128 		return (-1);
6129 	}
6130 	cqe->cq_input_ptr = cqe->cq_output_ptr;
6131 
6132 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
6133 	mbin->cmd_clear = 0x80;
6134 
6135 	mcx_cmdq_mboxes_sign(&mxm, 1);
6136 	mcx_cmdq_post(sc, cqe, 0);
6137 
6138 	error = mcx_cmdq_poll(sc, cqe, 1000);
6139 	if (error != 0) {
6140 		printf("%s: query nic vport counters timeout\n", DEVNAME(sc));
6141 		goto free;
6142 	}
6143 	if (mcx_cmdq_verify(cqe) != 0) {
6144 		printf("%s: query nic vport counters command corrupt\n",
6145 		    DEVNAME(sc));
6146 		goto free;
6147 	}
6148 
6149 	out = mcx_cmdq_out(cqe);
6150 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
6151 		printf("%s: query nic vport counters failed (%x, %x)\n",
6152 		    DEVNAME(sc), out->cmd_status, betoh32(out->cmd_syndrome));
6153 		error = -1;
6154 		goto free;
6155 	}
6156 
6157 	counters = (struct mcx_nic_vport_counters *)
6158 	    (mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)));
6159 	if (counters->rx_bcast.packets + counters->tx_bcast.packets +
6160 	    counters->rx_ucast.packets + counters->tx_ucast.packets +
6161 	    counters->rx_err.packets + counters->tx_err.packets)
6162 		printf("%s: err %llx/%llx uc %llx/%llx bc %llx/%llx\n",
6163 		    DEVNAME(sc),
6164 		    betoh64(counters->tx_err.packets),
6165 		    betoh64(counters->rx_err.packets),
6166 		    betoh64(counters->tx_ucast.packets),
6167 		    betoh64(counters->rx_ucast.packets),
6168 		    betoh64(counters->tx_bcast.packets),
6169 		    betoh64(counters->rx_bcast.packets));
6170 free:
6171 	mcx_dmamem_free(sc, &mxm);
6172 
6173 	return (error);
6174 }
6175 
6176 static int
6177 mcx_dump_flow_counter(struct mcx_softc *sc, int index, const char *what)
6178 {
6179 	struct mcx_dmamem mxm;
6180 	struct mcx_cmdq_entry *cqe;
6181 	struct mcx_cmd_query_flow_counter_in *in;
6182 	struct mcx_cmd_query_flow_counter_mb_in *mbin;
6183 	struct mcx_cmd_query_flow_counter_out *out;
6184 	struct mcx_counter *counters;
6185 	int error, token;
6186 
6187 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
6188 	token = mcx_cmdq_token(sc);
6189 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin), sizeof(*out) +
6190 	    sizeof(*counters), token);
6191 
6192 	in = mcx_cmdq_in(cqe);
6193 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_FLOW_COUNTER);
6194 	in->cmd_op_mod = htobe16(0);
6195 
6196 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
6197 	    &cqe->cq_output_ptr, token) != 0) {
6198 		printf(", unable to allocate query flow counter mailboxen\n");
6199 		return (-1);
6200 	}
6201 	cqe->cq_input_ptr = cqe->cq_output_ptr;
6202 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
6203 	mbin->cmd_flow_counter_id = htobe16(sc->sc_flow_counter_id[index]);
6204 	mbin->cmd_clear = 0x80;
6205 
6206 	mcx_cmdq_mboxes_sign(&mxm, 1);
6207 	mcx_cmdq_post(sc, cqe, 0);
6208 
6209 	error = mcx_cmdq_poll(sc, cqe, 1000);
6210 	if (error != 0) {
6211 		printf("%s: query flow counter timeout\n", DEVNAME(sc));
6212 		goto free;
6213 	}
6214 	if (mcx_cmdq_verify(cqe) != 0) {
6215 		printf("%s: query flow counter command corrupt\n", DEVNAME(sc));
6216 		goto free;
6217 	}
6218 
6219 	out = mcx_cmdq_out(cqe);
6220 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
6221 		printf("%s: query flow counter failed (%x, %x)\n", DEVNAME(sc),
6222 		    out->cmd_status, betoh32(out->cmd_syndrome));
6223 		error = -1;
6224 		goto free;
6225 	}
6226 
6227 	counters = (struct mcx_counter *)
6228 	    (mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)));
6229 	if (counters->packets)
6230 		printf("%s: %s inflow %llx\n", DEVNAME(sc), what,
6231 		    betoh64(counters->packets));
6232 free:
6233 	mcx_dmamem_free(sc, &mxm);
6234 
6235 	return (error);
6236 }
6237 
6238 #endif
6239 
6240 int
6241 mcx_rx_fill_slots(struct mcx_softc *sc, struct mcx_rx *rx,
6242     void *ring, struct mcx_slot *slots, uint *prod, uint nslots)
6243 {
6244 	struct mcx_rq_entry *rqe;
6245 	struct mcx_slot *ms;
6246 	struct mbuf *m;
6247 	uint slot, p, fills;
6248 
6249 	p = *prod;
6250 	slot = (p % (1 << MCX_LOG_RQ_SIZE));
6251 	rqe = ring;
6252 	for (fills = 0; fills < nslots; fills++) {
6253 		ms = &slots[slot];
6254 		m = MCLGETI(NULL, M_DONTWAIT, NULL, sc->sc_rxbufsz);
6255 		if (m == NULL)
6256 			break;
6257 
6258 		m->m_data += (m->m_ext.ext_size - sc->sc_rxbufsz);
6259 		m->m_data += ETHER_ALIGN;
6260 		m->m_len = m->m_pkthdr.len = sc->sc_hardmtu;
6261 		if (bus_dmamap_load_mbuf(sc->sc_dmat, ms->ms_map, m,
6262 		    BUS_DMA_NOWAIT) != 0) {
6263 			m_freem(m);
6264 			break;
6265 		}
6266 		ms->ms_m = m;
6267 
6268 		rqe[slot].rqe_byte_count =
6269 		    htobe32(ms->ms_map->dm_segs[0].ds_len);
6270 		rqe[slot].rqe_addr = htobe64(ms->ms_map->dm_segs[0].ds_addr);
6271 		rqe[slot].rqe_lkey = htobe32(sc->sc_lkey);
6272 
6273 		p++;
6274 		slot++;
6275 		if (slot == (1 << MCX_LOG_RQ_SIZE))
6276 			slot = 0;
6277 	}
6278 
6279 	if (fills != 0) {
6280 		*rx->rx_doorbell = htobe32(p & MCX_WQ_DOORBELL_MASK);
6281 		/* barrier? */
6282 	}
6283 
6284 	*prod = p;
6285 
6286 	return (nslots - fills);
6287 }
6288 
6289 int
6290 mcx_rx_fill(struct mcx_softc *sc, struct mcx_rx *rx)
6291 {
6292 	u_int slots;
6293 
6294 	slots = if_rxr_get(&rx->rx_rxr, (1 << MCX_LOG_RQ_SIZE));
6295 	if (slots == 0)
6296 		return (1);
6297 
6298 	slots = mcx_rx_fill_slots(sc, rx, MCX_DMA_KVA(&rx->rx_rq_mem),
6299 	    rx->rx_slots, &rx->rx_prod, slots);
6300 	if_rxr_put(&rx->rx_rxr, slots);
6301 	return (0);
6302 }
6303 
6304 void
6305 mcx_refill(void *xrx)
6306 {
6307 	struct mcx_rx *rx = xrx;
6308 	struct mcx_softc *sc = rx->rx_softc;
6309 
6310 	mcx_rx_fill(sc, rx);
6311 
6312 	if (if_rxr_inuse(&rx->rx_rxr) == 0)
6313 		timeout_add(&rx->rx_refill, 1);
6314 }
6315 
6316 void
6317 mcx_process_txeof(struct mcx_softc *sc, struct mcx_tx *tx,
6318     struct mcx_cq_entry *cqe, int *txfree)
6319 {
6320 	struct mcx_slot *ms;
6321 	bus_dmamap_t map;
6322 	int slot, slots;
6323 
6324 	slot = betoh16(cqe->cq_wqe_count) % (1 << MCX_LOG_SQ_SIZE);
6325 
6326 	ms = &tx->tx_slots[slot];
6327 	map = ms->ms_map;
6328 	bus_dmamap_sync(sc->sc_dmat, map, 0, map->dm_mapsize,
6329 	    BUS_DMASYNC_POSTWRITE);
6330 
6331 	slots = 1;
6332 	if (map->dm_nsegs > 1)
6333 		slots += (map->dm_nsegs+2) / MCX_SQ_SEGS_PER_SLOT;
6334 
6335 	(*txfree) += slots;
6336 	bus_dmamap_unload(sc->sc_dmat, map);
6337 	m_freem(ms->ms_m);
6338 	ms->ms_m = NULL;
6339 }
6340 
6341 static uint64_t
6342 mcx_uptime(void)
6343 {
6344 	struct timespec ts;
6345 
6346 	nanouptime(&ts);
6347 
6348 	return ((uint64_t)ts.tv_sec * 1000000000 + (uint64_t)ts.tv_nsec);
6349 }
6350 
6351 static void
6352 mcx_calibrate_first(struct mcx_softc *sc)
6353 {
6354 	struct mcx_calibration *c = &sc->sc_calibration[0];
6355 
6356 	sc->sc_calibration_gen = 0;
6357 
6358 	c->c_ubase = mcx_uptime();
6359 	c->c_tbase = mcx_timer(sc);
6360 	c->c_tdiff = 0;
6361 
6362 	timeout_add_sec(&sc->sc_calibrate, MCX_CALIBRATE_FIRST);
6363 }
6364 
6365 #define MCX_TIMESTAMP_SHIFT 10
6366 
6367 static void
6368 mcx_calibrate(void *arg)
6369 {
6370 	struct mcx_softc *sc = arg;
6371 	struct mcx_calibration *nc, *pc;
6372 	unsigned int gen;
6373 
6374 	if (!ISSET(sc->sc_ac.ac_if.if_flags, IFF_RUNNING))
6375 		return;
6376 
6377 	timeout_add_sec(&sc->sc_calibrate, MCX_CALIBRATE_NORMAL);
6378 
6379 	gen = sc->sc_calibration_gen;
6380 	pc = &sc->sc_calibration[gen % nitems(sc->sc_calibration)];
6381 	gen++;
6382 	nc = &sc->sc_calibration[gen % nitems(sc->sc_calibration)];
6383 
6384 	nc->c_uptime = pc->c_ubase;
6385 	nc->c_timestamp = pc->c_tbase;
6386 
6387 	nc->c_ubase = mcx_uptime();
6388 	nc->c_tbase = mcx_timer(sc);
6389 
6390 	nc->c_udiff = (nc->c_ubase - nc->c_uptime) >> MCX_TIMESTAMP_SHIFT;
6391 	nc->c_tdiff = (nc->c_tbase - nc->c_timestamp) >> MCX_TIMESTAMP_SHIFT;
6392 
6393 	membar_producer();
6394 	sc->sc_calibration_gen = gen;
6395 }
6396 
6397 static int
6398 mcx_process_rx(struct mcx_softc *sc, struct mcx_rx *rx,
6399     struct mcx_cq_entry *cqe, struct mbuf_list *ml,
6400     const struct mcx_calibration *c)
6401 {
6402 	struct mcx_slot *ms;
6403 	struct mbuf *m;
6404 	uint32_t flags;
6405 	int slot;
6406 
6407 	slot = betoh16(cqe->cq_wqe_count) % (1 << MCX_LOG_RQ_SIZE);
6408 
6409 	ms = &rx->rx_slots[slot];
6410 	bus_dmamap_sync(sc->sc_dmat, ms->ms_map, 0, ms->ms_map->dm_mapsize,
6411 	    BUS_DMASYNC_POSTREAD);
6412 	bus_dmamap_unload(sc->sc_dmat, ms->ms_map);
6413 
6414 	m = ms->ms_m;
6415 	ms->ms_m = NULL;
6416 
6417 	m->m_pkthdr.len = m->m_len = bemtoh32(&cqe->cq_byte_cnt);
6418 
6419 	if (cqe->cq_rx_hash_type) {
6420 		m->m_pkthdr.ph_flowid = betoh32(cqe->cq_rx_hash);
6421 		m->m_pkthdr.csum_flags |= M_FLOWID;
6422 	}
6423 
6424 	flags = bemtoh32(&cqe->cq_flags);
6425 	if (flags & MCX_CQ_ENTRY_FLAGS_L3_OK)
6426 		m->m_pkthdr.csum_flags = M_IPV4_CSUM_IN_OK;
6427 	if (flags & MCX_CQ_ENTRY_FLAGS_L4_OK)
6428 		m->m_pkthdr.csum_flags |= M_TCP_CSUM_IN_OK |
6429 		    M_UDP_CSUM_IN_OK;
6430 #if NVLAN > 0
6431 	if (flags & MCX_CQ_ENTRY_FLAGS_CV) {
6432 		m->m_pkthdr.ether_vtag = (flags &
6433 		    MCX_CQ_ENTRY_FLAGS_VLAN_MASK);
6434 		m->m_flags |= M_VLANTAG;
6435 	}
6436 #endif
6437 
6438 	if (c->c_tdiff) {
6439 		uint64_t t = bemtoh64(&cqe->cq_timestamp) - c->c_timestamp;
6440 		t *= c->c_udiff;
6441 		t /= c->c_tdiff;
6442 
6443 		m->m_pkthdr.ph_timestamp = c->c_uptime + t;
6444 		SET(m->m_pkthdr.csum_flags, M_TIMESTAMP);
6445 	}
6446 
6447 	ml_enqueue(ml, m);
6448 
6449 	return (1);
6450 }
6451 
6452 static struct mcx_cq_entry *
6453 mcx_next_cq_entry(struct mcx_softc *sc, struct mcx_cq *cq)
6454 {
6455 	struct mcx_cq_entry *cqe;
6456 	int next;
6457 
6458 	cqe = (struct mcx_cq_entry *)MCX_DMA_KVA(&cq->cq_mem);
6459 	next = cq->cq_cons % (1 << MCX_LOG_CQ_SIZE);
6460 
6461 	if ((cqe[next].cq_opcode_owner & MCX_CQ_ENTRY_FLAG_OWNER) ==
6462 	    ((cq->cq_cons >> MCX_LOG_CQ_SIZE) & 1)) {
6463 		return (&cqe[next]);
6464 	}
6465 
6466 	return (NULL);
6467 }
6468 
6469 static void
6470 mcx_arm_cq(struct mcx_softc *sc, struct mcx_cq *cq, int uar)
6471 {
6472 	bus_size_t offset;
6473 	uint32_t val;
6474 	uint64_t uval;
6475 
6476 	offset = (MCX_PAGE_SIZE * uar);
6477 	val = ((cq->cq_count) & 3) << MCX_CQ_DOORBELL_ARM_CMD_SN_SHIFT;
6478 	val |= (cq->cq_cons & MCX_CQ_DOORBELL_ARM_CI_MASK);
6479 
6480 	cq->cq_doorbell[0] = htobe32(cq->cq_cons & MCX_CQ_DOORBELL_ARM_CI_MASK);
6481 	cq->cq_doorbell[1] = htobe32(val);
6482 
6483 	uval = val;
6484 	uval <<= 32;
6485 	uval |= cq->cq_n;
6486 	bus_space_write_raw_8(sc->sc_memt, sc->sc_memh,
6487 	    offset + MCX_UAR_CQ_DOORBELL, htobe64(uval));
6488 	mcx_bar(sc, offset + MCX_UAR_CQ_DOORBELL, sizeof(uint64_t),
6489 	    BUS_SPACE_BARRIER_WRITE);
6490 }
6491 
6492 void
6493 mcx_process_cq(struct mcx_softc *sc, struct mcx_queues *q, struct mcx_cq *cq)
6494 {
6495 	struct mcx_rx *rx = &q->q_rx;
6496 	struct mcx_tx *tx = &q->q_tx;
6497 	const struct mcx_calibration *c;
6498 	unsigned int gen;
6499 	struct mcx_cq_entry *cqe;
6500 	uint8_t *cqp;
6501 	struct mbuf_list ml = MBUF_LIST_INITIALIZER();
6502 	int rxfree, txfree;
6503 
6504 	gen = sc->sc_calibration_gen;
6505 	membar_consumer();
6506 	c = &sc->sc_calibration[gen % nitems(sc->sc_calibration)];
6507 
6508 	rxfree = 0;
6509 	txfree = 0;
6510 	while ((cqe = mcx_next_cq_entry(sc, cq))) {
6511 		uint8_t opcode;
6512 		opcode = (cqe->cq_opcode_owner >> MCX_CQ_ENTRY_OPCODE_SHIFT);
6513 		switch (opcode) {
6514 		case MCX_CQ_ENTRY_OPCODE_REQ:
6515 			mcx_process_txeof(sc, tx, cqe, &txfree);
6516 			break;
6517 		case MCX_CQ_ENTRY_OPCODE_SEND:
6518 			rxfree += mcx_process_rx(sc, rx, cqe, &ml, c);
6519 			break;
6520 		case MCX_CQ_ENTRY_OPCODE_REQ_ERR:
6521 		case MCX_CQ_ENTRY_OPCODE_SEND_ERR:
6522 			cqp = (uint8_t *)cqe;
6523 			/* printf("%s: cq completion error: %x\n",
6524 			    DEVNAME(sc), cqp[0x37]); */
6525 			break;
6526 
6527 		default:
6528 			/* printf("%s: cq completion opcode %x??\n",
6529 			    DEVNAME(sc), opcode); */
6530 			break;
6531 		}
6532 
6533 		cq->cq_cons++;
6534 	}
6535 
6536 	cq->cq_count++;
6537 	mcx_arm_cq(sc, cq, q->q_uar);
6538 
6539 	if (rxfree > 0) {
6540 		if_rxr_put(&rx->rx_rxr, rxfree);
6541 		if (ifiq_input(rx->rx_ifiq, &ml))
6542 			if_rxr_livelocked(&rx->rx_rxr);
6543 
6544 		mcx_rx_fill(sc, rx);
6545 		if (if_rxr_inuse(&rx->rx_rxr) == 0)
6546 			timeout_add(&rx->rx_refill, 1);
6547 	}
6548 	if (txfree > 0) {
6549 		tx->tx_cons += txfree;
6550 		if (ifq_is_oactive(tx->tx_ifq))
6551 			ifq_restart(tx->tx_ifq);
6552 	}
6553 }
6554 
6555 
6556 static void
6557 mcx_arm_eq(struct mcx_softc *sc, struct mcx_eq *eq, int uar)
6558 {
6559 	bus_size_t offset;
6560 	uint32_t val;
6561 
6562 	offset = (MCX_PAGE_SIZE * uar) + MCX_UAR_EQ_DOORBELL_ARM;
6563 	val = (eq->eq_n << 24) | (eq->eq_cons & 0xffffff);
6564 
6565 	mcx_wr(sc, offset, val);
6566 	/* barrier? */
6567 }
6568 
6569 static struct mcx_eq_entry *
6570 mcx_next_eq_entry(struct mcx_softc *sc, struct mcx_eq *eq)
6571 {
6572 	struct mcx_eq_entry *eqe;
6573 	int next;
6574 
6575 	eqe = (struct mcx_eq_entry *)MCX_DMA_KVA(&eq->eq_mem);
6576 	next = eq->eq_cons % (1 << MCX_LOG_EQ_SIZE);
6577 	if ((eqe[next].eq_owner & 1) ==
6578 	    ((eq->eq_cons >> MCX_LOG_EQ_SIZE) & 1)) {
6579 		eq->eq_cons++;
6580 		return (&eqe[next]);
6581 	}
6582 	return (NULL);
6583 }
6584 
6585 int
6586 mcx_admin_intr(void *xsc)
6587 {
6588 	struct mcx_softc *sc = (struct mcx_softc *)xsc;
6589 	struct mcx_eq_entry *eqe;
6590 
6591 	while ((eqe = mcx_next_eq_entry(sc, &sc->sc_admin_eq))) {
6592 		switch (eqe->eq_event_type) {
6593 		case MCX_EVENT_TYPE_LAST_WQE:
6594 			/* printf("%s: last wqe reached?\n", DEVNAME(sc)); */
6595 			break;
6596 
6597 		case MCX_EVENT_TYPE_CQ_ERROR:
6598 			/* printf("%s: cq error\n", DEVNAME(sc)); */
6599 			break;
6600 
6601 		case MCX_EVENT_TYPE_CMD_COMPLETION:
6602 			/* wakeup probably */
6603 			break;
6604 
6605 		case MCX_EVENT_TYPE_PORT_CHANGE:
6606 			task_add(systq, &sc->sc_port_change);
6607 			break;
6608 
6609 		default:
6610 			/* printf("%s: something happened\n", DEVNAME(sc)); */
6611 			break;
6612 		}
6613 	}
6614 	mcx_arm_eq(sc, &sc->sc_admin_eq, sc->sc_uar);
6615 	return (1);
6616 }
6617 
6618 int
6619 mcx_cq_intr(void *xq)
6620 {
6621 	struct mcx_queues *q = (struct mcx_queues *)xq;
6622 	struct mcx_softc *sc = q->q_sc;
6623 	struct mcx_eq_entry *eqe;
6624 	int cqn;
6625 
6626 	while ((eqe = mcx_next_eq_entry(sc, &q->q_eq))) {
6627 		switch (eqe->eq_event_type) {
6628 		case MCX_EVENT_TYPE_COMPLETION:
6629 			cqn = betoh32(eqe->eq_event_data[6]);
6630 			if (cqn == q->q_cq.cq_n)
6631 				mcx_process_cq(sc, q, &q->q_cq);
6632 			break;
6633 		}
6634 	}
6635 
6636 	mcx_arm_eq(sc, &q->q_eq, q->q_uar);
6637 	return (1);
6638 }
6639 
6640 
6641 static void
6642 mcx_free_slots(struct mcx_softc *sc, struct mcx_slot *slots, int allocated,
6643     int total)
6644 {
6645 	struct mcx_slot *ms;
6646 
6647 	int i = allocated;
6648 	while (i-- > 0) {
6649 		ms = &slots[i];
6650 		bus_dmamap_destroy(sc->sc_dmat, ms->ms_map);
6651 		if (ms->ms_m != NULL)
6652 			m_freem(ms->ms_m);
6653 	}
6654 	free(slots, M_DEVBUF, total * sizeof(*ms));
6655 }
6656 
6657 static int
6658 mcx_queue_up(struct mcx_softc *sc, struct mcx_queues *q)
6659 {
6660 	struct mcx_rx *rx;
6661 	struct mcx_tx *tx;
6662 	struct mcx_slot *ms;
6663 	int i;
6664 
6665 	rx = &q->q_rx;
6666 	rx->rx_slots = mallocarray(sizeof(*ms), (1 << MCX_LOG_RQ_SIZE),
6667 	    M_DEVBUF, M_WAITOK | M_ZERO);
6668 	if (rx->rx_slots == NULL) {
6669 		printf("%s: failed to allocate rx slots\n", DEVNAME(sc));
6670 		return ENOMEM;
6671 	}
6672 
6673 	for (i = 0; i < (1 << MCX_LOG_RQ_SIZE); i++) {
6674 		ms = &rx->rx_slots[i];
6675 		if (bus_dmamap_create(sc->sc_dmat, sc->sc_hardmtu, 1,
6676 		    sc->sc_hardmtu, 0,
6677 		    BUS_DMA_WAITOK | BUS_DMA_ALLOCNOW | BUS_DMA_64BIT,
6678 		    &ms->ms_map) != 0) {
6679 			printf("%s: failed to allocate rx dma maps\n",
6680 			    DEVNAME(sc));
6681 			goto destroy_rx_slots;
6682 		}
6683 	}
6684 
6685 	tx = &q->q_tx;
6686 	tx->tx_slots = mallocarray(sizeof(*ms), (1 << MCX_LOG_SQ_SIZE),
6687 	    M_DEVBUF, M_WAITOK | M_ZERO);
6688 	if (tx->tx_slots == NULL) {
6689 		printf("%s: failed to allocate tx slots\n", DEVNAME(sc));
6690 		goto destroy_rx_slots;
6691 	}
6692 
6693 	for (i = 0; i < (1 << MCX_LOG_SQ_SIZE); i++) {
6694 		ms = &tx->tx_slots[i];
6695 		if (bus_dmamap_create(sc->sc_dmat, sc->sc_hardmtu,
6696 		    MCX_SQ_MAX_SEGMENTS, sc->sc_hardmtu, 0,
6697 		    BUS_DMA_WAITOK | BUS_DMA_ALLOCNOW | BUS_DMA_64BIT,
6698 		    &ms->ms_map) != 0) {
6699 			printf("%s: failed to allocate tx dma maps\n",
6700 			    DEVNAME(sc));
6701 			goto destroy_tx_slots;
6702 		}
6703 	}
6704 
6705 	if (mcx_create_cq(sc, &q->q_cq, q->q_uar, q->q_index,
6706 	    q->q_eq.eq_n) != 0)
6707 		return ENOMEM;
6708 
6709 	if (mcx_create_sq(sc, tx, q->q_uar, q->q_index, q->q_cq.cq_n)
6710 	    != 0)
6711 		return ENOMEM;
6712 
6713 	if (mcx_create_rq(sc, rx, q->q_index, q->q_cq.cq_n) != 0)
6714 		return ENOMEM;
6715 
6716 	return 0;
6717 destroy_tx_slots:
6718 	mcx_free_slots(sc, tx->tx_slots, i, (1 << MCX_LOG_SQ_SIZE));
6719 	tx->tx_slots = NULL;
6720 
6721 	i = (1 << MCX_LOG_RQ_SIZE);
6722 destroy_rx_slots:
6723 	mcx_free_slots(sc, rx->rx_slots, i, (1 << MCX_LOG_RQ_SIZE));
6724 	rx->rx_slots = NULL;
6725 	return ENOMEM;
6726 }
6727 
6728 static int
6729 mcx_rss_group_entry_count(struct mcx_softc *sc, int group)
6730 {
6731 	int i;
6732 	int count;
6733 
6734 	count = 0;
6735 	for (i = 0; i < nitems(mcx_rss_config); i++) {
6736 		if (mcx_rss_config[i].flow_group == group)
6737 			count++;
6738 	}
6739 
6740 	return count;
6741 }
6742 
6743 static int
6744 mcx_up(struct mcx_softc *sc)
6745 {
6746 	struct ifnet *ifp = &sc->sc_ac.ac_if;
6747 	struct mcx_rx *rx;
6748 	struct mcx_tx *tx;
6749 	int i, start, count, flow_group, flow_index;
6750 	struct mcx_flow_match match_crit;
6751 	struct mcx_rss_rule *rss;
6752 	uint32_t dest;
6753 	int rqns[MCX_MAX_QUEUES];
6754 
6755 	if (mcx_create_tis(sc, &sc->sc_tis) != 0)
6756 		goto down;
6757 
6758 	for (i = 0; i < sc->sc_nqueues; i++) {
6759 		if (mcx_queue_up(sc, &sc->sc_queues[i]) != 0) {
6760 			goto down;
6761 		}
6762 	}
6763 
6764 	/* RSS flow table and flow groups */
6765 	if (mcx_create_flow_table(sc, MCX_LOG_FLOW_TABLE_SIZE, 1,
6766 	    &sc->sc_rss_flow_table_id) != 0)
6767 		goto down;
6768 
6769 	dest = MCX_FLOW_CONTEXT_DEST_TYPE_TABLE |
6770 	    sc->sc_rss_flow_table_id;
6771 
6772 	/* L4 RSS flow group (v4/v6 tcp/udp, no fragments) */
6773 	memset(&match_crit, 0, sizeof(match_crit));
6774 	match_crit.mc_ethertype = 0xffff;
6775 	match_crit.mc_ip_proto = 0xff;
6776 	match_crit.mc_vlan_flags = MCX_FLOW_MATCH_IP_FRAG;
6777 	start = 0;
6778 	count = mcx_rss_group_entry_count(sc, MCX_FLOW_GROUP_RSS_L4);
6779 	if (count != 0) {
6780 		if (mcx_create_flow_group(sc, sc->sc_rss_flow_table_id,
6781 		    MCX_FLOW_GROUP_RSS_L4, start, count,
6782 		    MCX_CREATE_FLOW_GROUP_CRIT_OUTER, &match_crit) != 0)
6783 			goto down;
6784 		start += count;
6785 	}
6786 
6787 	/* L3 RSS flow group (v4/v6, including fragments) */
6788 	memset(&match_crit, 0, sizeof(match_crit));
6789 	match_crit.mc_ethertype = 0xffff;
6790 	count = mcx_rss_group_entry_count(sc, MCX_FLOW_GROUP_RSS_L3);
6791 	if (mcx_create_flow_group(sc, sc->sc_rss_flow_table_id,
6792 	    MCX_FLOW_GROUP_RSS_L3, start, count,
6793 	    MCX_CREATE_FLOW_GROUP_CRIT_OUTER, &match_crit) != 0)
6794 		goto down;
6795 	start += count;
6796 
6797 	/* non-RSS flow group */
6798 	count = mcx_rss_group_entry_count(sc, MCX_FLOW_GROUP_RSS_NONE);
6799 	memset(&match_crit, 0, sizeof(match_crit));
6800 	if (mcx_create_flow_group(sc, sc->sc_rss_flow_table_id,
6801 	    MCX_FLOW_GROUP_RSS_NONE, start, count, 0, &match_crit) != 0)
6802 		goto down;
6803 
6804 	/* Root flow table, matching packets based on mac address */
6805 	if (mcx_create_flow_table(sc, MCX_LOG_FLOW_TABLE_SIZE, 0,
6806 	    &sc->sc_mac_flow_table_id) != 0)
6807 		goto down;
6808 
6809 	/* promisc flow group */
6810 	start = 0;
6811 	memset(&match_crit, 0, sizeof(match_crit));
6812 	if (mcx_create_flow_group(sc, sc->sc_mac_flow_table_id,
6813 	    MCX_FLOW_GROUP_PROMISC, start, 1, 0, &match_crit) != 0)
6814 		goto down;
6815 	sc->sc_promisc_flow_enabled = 0;
6816 	start++;
6817 
6818 	/* all multicast flow group */
6819 	match_crit.mc_dest_mac[0] = 0x01;
6820 	if (mcx_create_flow_group(sc, sc->sc_mac_flow_table_id,
6821 	    MCX_FLOW_GROUP_ALLMULTI, start, 1,
6822 	    MCX_CREATE_FLOW_GROUP_CRIT_OUTER, &match_crit) != 0)
6823 		goto down;
6824 	sc->sc_allmulti_flow_enabled = 0;
6825 	start++;
6826 
6827 	/* mac address matching flow group */
6828 	memset(&match_crit.mc_dest_mac, 0xff, sizeof(match_crit.mc_dest_mac));
6829 	if (mcx_create_flow_group(sc, sc->sc_mac_flow_table_id,
6830 	    MCX_FLOW_GROUP_MAC, start, (1 << MCX_LOG_FLOW_TABLE_SIZE) - start,
6831 	    MCX_CREATE_FLOW_GROUP_CRIT_OUTER, &match_crit) != 0)
6832 		goto down;
6833 
6834 	/* flow table entries for unicast and broadcast */
6835 	start = 0;
6836 	if (mcx_set_flow_table_entry_mac(sc, MCX_FLOW_GROUP_MAC, start,
6837 	    sc->sc_ac.ac_enaddr, dest) != 0)
6838 		goto down;
6839 	start++;
6840 
6841 	if (mcx_set_flow_table_entry_mac(sc, MCX_FLOW_GROUP_MAC, start,
6842 	    etherbroadcastaddr, dest) != 0)
6843 		goto down;
6844 	start++;
6845 
6846 	/* multicast entries go after that */
6847 	sc->sc_mcast_flow_base = start;
6848 
6849 	/* re-add any existing multicast flows */
6850 	for (i = 0; i < MCX_NUM_MCAST_FLOWS; i++) {
6851 		if (sc->sc_mcast_flows[i][0] != 0) {
6852 			mcx_set_flow_table_entry_mac(sc, MCX_FLOW_GROUP_MAC,
6853 			    sc->sc_mcast_flow_base + i,
6854 			    sc->sc_mcast_flows[i], dest);
6855 		}
6856 	}
6857 
6858 	if (mcx_set_flow_table_root(sc, sc->sc_mac_flow_table_id) != 0)
6859 		goto down;
6860 
6861 	/*
6862 	 * the RQT can be any size as long as it's a power of two.
6863 	 * since we also restrict the number of queues to a power of two,
6864 	 * we can just put each rx queue in once.
6865 	 */
6866 	for (i = 0; i < sc->sc_nqueues; i++)
6867 		rqns[i] = sc->sc_queues[i].q_rx.rx_rqn;
6868 
6869 	if (mcx_create_rqt(sc, sc->sc_nqueues, rqns, &sc->sc_rqt) != 0)
6870 		goto down;
6871 
6872 	start = 0;
6873 	flow_index = 0;
6874 	flow_group = -1;
6875 	for (i = 0; i < nitems(mcx_rss_config); i++) {
6876 		rss = &mcx_rss_config[i];
6877 		if (rss->flow_group != flow_group) {
6878 			flow_group = rss->flow_group;
6879 			flow_index = 0;
6880 		}
6881 
6882 		if (rss->hash_sel == 0) {
6883 			if (mcx_create_tir_direct(sc, &sc->sc_queues[0].q_rx,
6884 			    &sc->sc_tir[i]) != 0)
6885 				goto down;
6886 		} else {
6887 			if (mcx_create_tir_indirect(sc, sc->sc_rqt,
6888 			    rss->hash_sel, &sc->sc_tir[i]) != 0)
6889 				goto down;
6890 		}
6891 
6892 		if (mcx_set_flow_table_entry_proto(sc, flow_group,
6893 		    flow_index, rss->ethertype, rss->ip_proto,
6894 		    MCX_FLOW_CONTEXT_DEST_TYPE_TIR | sc->sc_tir[i]) != 0)
6895 			goto down;
6896 		flow_index++;
6897 	}
6898 
6899 	for (i = 0; i < sc->sc_nqueues; i++) {
6900 		struct mcx_queues *q = &sc->sc_queues[i];
6901 		rx = &q->q_rx;
6902 		tx = &q->q_tx;
6903 
6904 		/* start the queues */
6905 		if (mcx_ready_sq(sc, tx) != 0)
6906 			goto down;
6907 
6908 		if (mcx_ready_rq(sc, rx) != 0)
6909 			goto down;
6910 
6911 		if_rxr_init(&rx->rx_rxr, 1, (1 << MCX_LOG_RQ_SIZE));
6912 		rx->rx_prod = 0;
6913 		mcx_rx_fill(sc, rx);
6914 
6915 		tx->tx_cons = 0;
6916 		tx->tx_prod = 0;
6917 		ifq_clr_oactive(tx->tx_ifq);
6918 	}
6919 
6920 	mcx_calibrate_first(sc);
6921 
6922 	SET(ifp->if_flags, IFF_RUNNING);
6923 
6924 	return ENETRESET;
6925 down:
6926 	mcx_down(sc);
6927 	return ENOMEM;
6928 }
6929 
6930 static void
6931 mcx_down(struct mcx_softc *sc)
6932 {
6933 	struct ifnet *ifp = &sc->sc_ac.ac_if;
6934 	struct mcx_rss_rule *rss;
6935 	int group, i, flow_group, flow_index;
6936 
6937 	CLR(ifp->if_flags, IFF_RUNNING);
6938 
6939 	/*
6940 	 * delete flow table entries first, so no packets can arrive
6941 	 * after the barriers
6942 	 */
6943 	if (sc->sc_promisc_flow_enabled)
6944 		mcx_delete_flow_table_entry(sc, MCX_FLOW_GROUP_PROMISC, 0);
6945 	if (sc->sc_allmulti_flow_enabled)
6946 		mcx_delete_flow_table_entry(sc, MCX_FLOW_GROUP_ALLMULTI, 0);
6947 	mcx_delete_flow_table_entry(sc, MCX_FLOW_GROUP_MAC, 0);
6948 	mcx_delete_flow_table_entry(sc, MCX_FLOW_GROUP_MAC, 1);
6949 	for (i = 0; i < MCX_NUM_MCAST_FLOWS; i++) {
6950 		if (sc->sc_mcast_flows[i][0] != 0) {
6951 			mcx_delete_flow_table_entry(sc, MCX_FLOW_GROUP_MAC,
6952 			    sc->sc_mcast_flow_base + i);
6953 		}
6954 	}
6955 
6956 	flow_group = -1;
6957 	flow_index = 0;
6958 	for (i = 0; i < nitems(mcx_rss_config); i++) {
6959 		rss = &mcx_rss_config[i];
6960 		if (rss->flow_group != flow_group) {
6961 			flow_group = rss->flow_group;
6962 			flow_index = 0;
6963 		}
6964 
6965 		mcx_delete_flow_table_entry(sc, flow_group, flow_index);
6966 
6967 		mcx_destroy_tir(sc, sc->sc_tir[i]);
6968 		sc->sc_tir[i] = 0;
6969 
6970 		flow_index++;
6971 	}
6972 	intr_barrier(sc->sc_ihc);
6973 	for (i = 0; i < sc->sc_nqueues; i++) {
6974 		struct ifqueue *ifq = sc->sc_queues[i].q_tx.tx_ifq;
6975 		ifq_barrier(ifq);
6976 
6977 		timeout_del_barrier(&sc->sc_queues[i].q_rx.rx_refill);
6978 
6979 		intr_barrier(sc->sc_queues[i].q_ihc);
6980 	}
6981 
6982 	timeout_del_barrier(&sc->sc_calibrate);
6983 
6984 	for (group = 0; group < MCX_NUM_FLOW_GROUPS; group++) {
6985 		if (sc->sc_flow_group[group].g_id != -1)
6986 			mcx_destroy_flow_group(sc, group);
6987 	}
6988 
6989 	if (sc->sc_mac_flow_table_id != -1) {
6990 		mcx_destroy_flow_table(sc, sc->sc_mac_flow_table_id);
6991 		sc->sc_mac_flow_table_id = -1;
6992 	}
6993 	if (sc->sc_rss_flow_table_id != -1) {
6994 		mcx_destroy_flow_table(sc, sc->sc_rss_flow_table_id);
6995 		sc->sc_rss_flow_table_id = -1;
6996 	}
6997 	if (sc->sc_rqt != -1) {
6998 		mcx_destroy_rqt(sc, sc->sc_rqt);
6999 		sc->sc_rqt = -1;
7000 	}
7001 
7002 	for (i = 0; i < sc->sc_nqueues; i++) {
7003 		struct mcx_queues *q = &sc->sc_queues[i];
7004 		struct mcx_rx *rx = &q->q_rx;
7005 		struct mcx_tx *tx = &q->q_tx;
7006 		struct mcx_cq *cq = &q->q_cq;
7007 
7008 		if (rx->rx_rqn != 0)
7009 			mcx_destroy_rq(sc, rx);
7010 
7011 		if (tx->tx_sqn != 0)
7012 			mcx_destroy_sq(sc, tx);
7013 
7014 		if (tx->tx_slots != NULL) {
7015 			mcx_free_slots(sc, tx->tx_slots,
7016 			    (1 << MCX_LOG_SQ_SIZE), (1 << MCX_LOG_SQ_SIZE));
7017 			tx->tx_slots = NULL;
7018 		}
7019 		if (rx->rx_slots != NULL) {
7020 			mcx_free_slots(sc, rx->rx_slots,
7021 			    (1 << MCX_LOG_RQ_SIZE), (1 << MCX_LOG_RQ_SIZE));
7022 			rx->rx_slots = NULL;
7023 		}
7024 
7025 		if (cq->cq_n != 0)
7026 			mcx_destroy_cq(sc, cq);
7027 	}
7028 	if (sc->sc_tis != 0) {
7029 		mcx_destroy_tis(sc, sc->sc_tis);
7030 		sc->sc_tis = 0;
7031 	}
7032 }
7033 
7034 static int
7035 mcx_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
7036 {
7037 	struct mcx_softc *sc = (struct mcx_softc *)ifp->if_softc;
7038 	struct ifreq *ifr = (struct ifreq *)data;
7039 	uint8_t addrhi[ETHER_ADDR_LEN], addrlo[ETHER_ADDR_LEN];
7040 	int s, i, error = 0;
7041 	uint32_t dest;
7042 
7043 	s = splnet();
7044 	switch (cmd) {
7045 	case SIOCSIFADDR:
7046 		ifp->if_flags |= IFF_UP;
7047 		/* FALLTHROUGH */
7048 
7049 	case SIOCSIFFLAGS:
7050 		if (ISSET(ifp->if_flags, IFF_UP)) {
7051 			if (ISSET(ifp->if_flags, IFF_RUNNING))
7052 				error = ENETRESET;
7053 			else
7054 				error = mcx_up(sc);
7055 		} else {
7056 			if (ISSET(ifp->if_flags, IFF_RUNNING))
7057 				mcx_down(sc);
7058 		}
7059 		break;
7060 
7061 	case SIOCGIFMEDIA:
7062 	case SIOCSIFMEDIA:
7063 		error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
7064 		break;
7065 
7066 	case SIOCGIFSFFPAGE:
7067 		error = mcx_get_sffpage(ifp, (struct if_sffpage *)data);
7068 		break;
7069 
7070 	case SIOCGIFRXR:
7071 		error = mcx_rxrinfo(sc, (struct if_rxrinfo *)ifr->ifr_data);
7072 		break;
7073 
7074 	case SIOCADDMULTI:
7075 		if (ether_addmulti(ifr, &sc->sc_ac) == ENETRESET) {
7076 			error = ether_multiaddr(&ifr->ifr_addr, addrlo, addrhi);
7077 			if (error != 0)
7078 				return (error);
7079 
7080 			dest = MCX_FLOW_CONTEXT_DEST_TYPE_TABLE |
7081 			    sc->sc_rss_flow_table_id;
7082 
7083 			for (i = 0; i < MCX_NUM_MCAST_FLOWS; i++) {
7084 				if (sc->sc_mcast_flows[i][0] == 0) {
7085 					memcpy(sc->sc_mcast_flows[i], addrlo,
7086 					    ETHER_ADDR_LEN);
7087 					if (ISSET(ifp->if_flags, IFF_RUNNING)) {
7088 						mcx_set_flow_table_entry_mac(sc,
7089 						    MCX_FLOW_GROUP_MAC,
7090 						    sc->sc_mcast_flow_base + i,
7091 						    sc->sc_mcast_flows[i], dest);
7092 					}
7093 					break;
7094 				}
7095 			}
7096 
7097 			if (!ISSET(ifp->if_flags, IFF_ALLMULTI)) {
7098 				if (i == MCX_NUM_MCAST_FLOWS) {
7099 					SET(ifp->if_flags, IFF_ALLMULTI);
7100 					sc->sc_extra_mcast++;
7101 					error = ENETRESET;
7102 				}
7103 
7104 				if (sc->sc_ac.ac_multirangecnt > 0) {
7105 					SET(ifp->if_flags, IFF_ALLMULTI);
7106 					error = ENETRESET;
7107 				}
7108 			}
7109 		}
7110 		break;
7111 
7112 	case SIOCDELMULTI:
7113 		if (ether_delmulti(ifr, &sc->sc_ac) == ENETRESET) {
7114 			error = ether_multiaddr(&ifr->ifr_addr, addrlo, addrhi);
7115 			if (error != 0)
7116 				return (error);
7117 
7118 			for (i = 0; i < MCX_NUM_MCAST_FLOWS; i++) {
7119 				if (memcmp(sc->sc_mcast_flows[i], addrlo,
7120 				    ETHER_ADDR_LEN) == 0) {
7121 					if (ISSET(ifp->if_flags, IFF_RUNNING)) {
7122 						mcx_delete_flow_table_entry(sc,
7123 						    MCX_FLOW_GROUP_MAC,
7124 						    sc->sc_mcast_flow_base + i);
7125 					}
7126 					sc->sc_mcast_flows[i][0] = 0;
7127 					break;
7128 				}
7129 			}
7130 
7131 			if (i == MCX_NUM_MCAST_FLOWS)
7132 				sc->sc_extra_mcast--;
7133 
7134 			if (ISSET(ifp->if_flags, IFF_ALLMULTI) &&
7135 			    (sc->sc_extra_mcast == 0) &&
7136 			    (sc->sc_ac.ac_multirangecnt == 0)) {
7137 				CLR(ifp->if_flags, IFF_ALLMULTI);
7138 				error = ENETRESET;
7139 			}
7140 		}
7141 		break;
7142 
7143 	default:
7144 		error = ether_ioctl(ifp, &sc->sc_ac, cmd, data);
7145 	}
7146 
7147 	if (error == ENETRESET) {
7148 		if ((ifp->if_flags & (IFF_UP | IFF_RUNNING)) ==
7149 		    (IFF_UP | IFF_RUNNING))
7150 			mcx_iff(sc);
7151 		error = 0;
7152 	}
7153 	splx(s);
7154 
7155 	return (error);
7156 }
7157 
7158 static int
7159 mcx_get_sffpage(struct ifnet *ifp, struct if_sffpage *sff)
7160 {
7161 	struct mcx_softc *sc = (struct mcx_softc *)ifp->if_softc;
7162 	struct mcx_reg_mcia mcia;
7163 	struct mcx_reg_pmlp pmlp;
7164 	int offset, error;
7165 
7166 	/* get module number */
7167 	memset(&pmlp, 0, sizeof(pmlp));
7168 	pmlp.rp_local_port = 1;
7169 	error = mcx_access_hca_reg(sc, MCX_REG_PMLP, MCX_REG_OP_READ, &pmlp,
7170 	    sizeof(pmlp));
7171 	if (error != 0) {
7172 		printf("%s: unable to get eeprom module number\n",
7173 		    DEVNAME(sc));
7174 		return error;
7175 	}
7176 
7177 	for (offset = 0; offset < 256; offset += MCX_MCIA_EEPROM_BYTES) {
7178 		memset(&mcia, 0, sizeof(mcia));
7179 		mcia.rm_l = 0;
7180 		mcia.rm_module = betoh32(pmlp.rp_lane0_mapping) &
7181 		    MCX_PMLP_MODULE_NUM_MASK;
7182 		mcia.rm_i2c_addr = sff->sff_addr / 2;	/* apparently */
7183 		mcia.rm_page_num = sff->sff_page;
7184 		mcia.rm_dev_addr = htobe16(offset);
7185 		mcia.rm_size = htobe16(MCX_MCIA_EEPROM_BYTES);
7186 
7187 		error = mcx_access_hca_reg(sc, MCX_REG_MCIA, MCX_REG_OP_READ,
7188 		    &mcia, sizeof(mcia));
7189 		if (error != 0) {
7190 			printf("%s: unable to read eeprom at %x\n",
7191 			    DEVNAME(sc), offset);
7192 			return error;
7193 		}
7194 
7195 		memcpy(sff->sff_data + offset, mcia.rm_data,
7196 		    MCX_MCIA_EEPROM_BYTES);
7197 	}
7198 
7199 	return 0;
7200 }
7201 
7202 static int
7203 mcx_rxrinfo(struct mcx_softc *sc, struct if_rxrinfo *ifri)
7204 {
7205 	struct if_rxring_info *ifrs;
7206 	unsigned int i;
7207 	int error;
7208 
7209 	ifrs = mallocarray(sc->sc_nqueues, sizeof(*ifrs), M_TEMP,
7210 	    M_WAITOK|M_ZERO|M_CANFAIL);
7211 	if (ifrs == NULL)
7212 		return (ENOMEM);
7213 
7214 	for (i = 0; i < sc->sc_nqueues; i++) {
7215 		struct mcx_rx *rx = &sc->sc_queues[i].q_rx;
7216 		struct if_rxring_info *ifr = &ifrs[i];
7217 
7218 		snprintf(ifr->ifr_name, sizeof(ifr->ifr_name), "%u", i);
7219 		ifr->ifr_size = sc->sc_hardmtu;
7220 		ifr->ifr_info = rx->rx_rxr;
7221 	}
7222 
7223 	error = if_rxr_info_ioctl(ifri, i, ifrs);
7224 	free(ifrs, M_TEMP, i * sizeof(*ifrs));
7225 
7226 	return (error);
7227 }
7228 
7229 int
7230 mcx_load_mbuf(struct mcx_softc *sc, struct mcx_slot *ms, struct mbuf *m)
7231 {
7232 	switch (bus_dmamap_load_mbuf(sc->sc_dmat, ms->ms_map, m,
7233 	    BUS_DMA_STREAMING | BUS_DMA_NOWAIT)) {
7234 	case 0:
7235 		break;
7236 
7237 	case EFBIG:
7238 		if (m_defrag(m, M_DONTWAIT) == 0 &&
7239 		    bus_dmamap_load_mbuf(sc->sc_dmat, ms->ms_map, m,
7240 		    BUS_DMA_STREAMING | BUS_DMA_NOWAIT) == 0)
7241 			break;
7242 
7243 	default:
7244 		return (1);
7245 	}
7246 
7247 	ms->ms_m = m;
7248 	return (0);
7249 }
7250 
7251 static void
7252 mcx_start(struct ifqueue *ifq)
7253 {
7254 	struct mcx_tx *tx = ifq->ifq_softc;
7255 	struct ifnet *ifp = ifq->ifq_if;
7256 	struct mcx_softc *sc = ifp->if_softc;
7257 	struct mcx_sq_entry *sq, *sqe;
7258 	struct mcx_sq_entry_seg *sqs;
7259 	struct mcx_slot *ms;
7260 	bus_dmamap_t map;
7261 	struct mbuf *m;
7262 	u_int idx, free, used;
7263 	uint64_t *bf;
7264 	uint32_t csum;
7265 	size_t bf_base;
7266 	int i, seg, nseg;
7267 
7268 	bf_base = (tx->tx_uar * MCX_PAGE_SIZE) + MCX_UAR_BF;
7269 
7270 	idx = tx->tx_prod % (1 << MCX_LOG_SQ_SIZE);
7271 	free = (tx->tx_cons + (1 << MCX_LOG_SQ_SIZE)) - tx->tx_prod;
7272 
7273 	used = 0;
7274 	bf = NULL;
7275 	sq = (struct mcx_sq_entry *)MCX_DMA_KVA(&tx->tx_sq_mem);
7276 
7277 	for (;;) {
7278 		if (used + MCX_SQ_ENTRY_MAX_SLOTS >= free) {
7279 			ifq_set_oactive(ifq);
7280 			break;
7281 		}
7282 
7283 		m = ifq_dequeue(ifq);
7284 		if (m == NULL) {
7285 			break;
7286 		}
7287 
7288 		sqe = sq + idx;
7289 		ms = &tx->tx_slots[idx];
7290 		memset(sqe, 0, sizeof(*sqe));
7291 
7292 		/* ctrl segment */
7293 		sqe->sqe_opcode_index = htobe32(MCX_SQE_WQE_OPCODE_SEND |
7294 		    ((tx->tx_prod & 0xffff) << MCX_SQE_WQE_INDEX_SHIFT));
7295 		/* always generate a completion event */
7296 		sqe->sqe_signature = htobe32(MCX_SQE_CE_CQE_ALWAYS);
7297 
7298 		/* eth segment */
7299 		csum = 0;
7300 		if (m->m_pkthdr.csum_flags & M_IPV4_CSUM_OUT)
7301 			csum |= MCX_SQE_L3_CSUM;
7302 		if (m->m_pkthdr.csum_flags & (M_TCP_CSUM_OUT | M_UDP_CSUM_OUT))
7303 			csum |= MCX_SQE_L4_CSUM;
7304 		sqe->sqe_mss_csum = htobe32(csum);
7305 		sqe->sqe_inline_header_size = htobe16(MCX_SQ_INLINE_SIZE);
7306 #if NVLAN > 0
7307 		if (m->m_flags & M_VLANTAG) {
7308 			struct ether_vlan_header *evh;
7309 			evh = (struct ether_vlan_header *)
7310 			    &sqe->sqe_inline_headers;
7311 
7312 			/* slightly cheaper vlan_inject() */
7313 			m_copydata(m, 0, ETHER_HDR_LEN, (caddr_t)evh);
7314 			evh->evl_proto = evh->evl_encap_proto;
7315 			evh->evl_encap_proto = htons(ETHERTYPE_VLAN);
7316 			evh->evl_tag = htons(m->m_pkthdr.ether_vtag);
7317 
7318 			m_adj(m, ETHER_HDR_LEN);
7319 		} else
7320 #endif
7321 		{
7322 			m_copydata(m, 0, MCX_SQ_INLINE_SIZE,
7323 			    (caddr_t)sqe->sqe_inline_headers);
7324 			m_adj(m, MCX_SQ_INLINE_SIZE);
7325 		}
7326 
7327 		if (mcx_load_mbuf(sc, ms, m) != 0) {
7328 			m_freem(m);
7329 			ifp->if_oerrors++;
7330 			continue;
7331 		}
7332 		bf = (uint64_t *)sqe;
7333 
7334 #if NBPFILTER > 0
7335 		if (ifp->if_bpf)
7336 			bpf_mtap_hdr(ifp->if_bpf,
7337 			    (caddr_t)sqe->sqe_inline_headers,
7338 			    MCX_SQ_INLINE_SIZE, m, BPF_DIRECTION_OUT);
7339 #endif
7340 		map = ms->ms_map;
7341 		bus_dmamap_sync(sc->sc_dmat, map, 0, map->dm_mapsize,
7342 		    BUS_DMASYNC_PREWRITE);
7343 
7344 		sqe->sqe_ds_sq_num =
7345 		    htobe32((tx->tx_sqn << MCX_SQE_SQ_NUM_SHIFT) |
7346 		    (map->dm_nsegs + 3));
7347 
7348 		/* data segment - first wqe has one segment */
7349 		sqs = sqe->sqe_segs;
7350 		seg = 0;
7351 		nseg = 1;
7352 		for (i = 0; i < map->dm_nsegs; i++) {
7353 			if (seg == nseg) {
7354 				/* next slot */
7355 				idx++;
7356 				if (idx == (1 << MCX_LOG_SQ_SIZE))
7357 					idx = 0;
7358 				tx->tx_prod++;
7359 				used++;
7360 
7361 				sqs = (struct mcx_sq_entry_seg *)(sq + idx);
7362 				seg = 0;
7363 				nseg = MCX_SQ_SEGS_PER_SLOT;
7364 			}
7365 			sqs[seg].sqs_byte_count =
7366 			    htobe32(map->dm_segs[i].ds_len);
7367 			sqs[seg].sqs_lkey = htobe32(sc->sc_lkey);
7368 			sqs[seg].sqs_addr = htobe64(map->dm_segs[i].ds_addr);
7369 			seg++;
7370 		}
7371 
7372 		idx++;
7373 		if (idx == (1 << MCX_LOG_SQ_SIZE))
7374 			idx = 0;
7375 		tx->tx_prod++;
7376 		used++;
7377 	}
7378 
7379 	if (used) {
7380 		htobem32(tx->tx_doorbell, tx->tx_prod & MCX_WQ_DOORBELL_MASK);
7381 
7382 		membar_sync();
7383 
7384 		/*
7385 		 * write the first 64 bits of the last sqe we produced
7386 		 * to the blue flame buffer
7387 		 */
7388 		bus_space_write_raw_8(sc->sc_memt, sc->sc_memh,
7389 		    bf_base + tx->tx_bf_offset, *bf);
7390 		/* next write goes to the other buffer */
7391 		tx->tx_bf_offset ^= sc->sc_bf_size;
7392 
7393 		membar_sync();
7394 	}
7395 }
7396 
7397 static void
7398 mcx_watchdog(struct ifnet *ifp)
7399 {
7400 }
7401 
7402 static void
7403 mcx_media_add_types(struct mcx_softc *sc)
7404 {
7405 	struct mcx_reg_ptys ptys;
7406 	int i;
7407 	uint32_t proto_cap;
7408 
7409 	memset(&ptys, 0, sizeof(ptys));
7410 	ptys.rp_local_port = 1;
7411 	ptys.rp_proto_mask = MCX_REG_PTYS_PROTO_MASK_ETH;
7412 	if (mcx_access_hca_reg(sc, MCX_REG_PTYS, MCX_REG_OP_READ, &ptys,
7413 	    sizeof(ptys)) != 0) {
7414 		printf("%s: unable to read port type/speed\n", DEVNAME(sc));
7415 		return;
7416 	}
7417 
7418 	proto_cap = betoh32(ptys.rp_eth_proto_cap);
7419 	for (i = 0; i < nitems(mcx_eth_cap_map); i++) {
7420 		const struct mcx_eth_proto_capability *cap;
7421 		if (!ISSET(proto_cap, 1 << i))
7422 			continue;
7423 
7424 		cap = &mcx_eth_cap_map[i];
7425 		if (cap->cap_media == 0)
7426 			continue;
7427 
7428 		ifmedia_add(&sc->sc_media, IFM_ETHER | cap->cap_media, 0, NULL);
7429 	}
7430 }
7431 
7432 static void
7433 mcx_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
7434 {
7435 	struct mcx_softc *sc = (struct mcx_softc *)ifp->if_softc;
7436 	struct mcx_reg_ptys ptys;
7437 	int i;
7438 	uint32_t proto_oper;
7439 	uint64_t media_oper;
7440 
7441 	memset(&ptys, 0, sizeof(ptys));
7442 	ptys.rp_local_port = 1;
7443 	ptys.rp_proto_mask = MCX_REG_PTYS_PROTO_MASK_ETH;
7444 
7445 	if (mcx_access_hca_reg(sc, MCX_REG_PTYS, MCX_REG_OP_READ, &ptys,
7446 	    sizeof(ptys)) != 0) {
7447 		printf("%s: unable to read port type/speed\n", DEVNAME(sc));
7448 		return;
7449 	}
7450 
7451 	proto_oper = betoh32(ptys.rp_eth_proto_oper);
7452 
7453 	media_oper = 0;
7454 
7455 	for (i = 0; i < nitems(mcx_eth_cap_map); i++) {
7456 		const struct mcx_eth_proto_capability *cap;
7457 		if (!ISSET(proto_oper, 1 << i))
7458 			continue;
7459 
7460 		cap = &mcx_eth_cap_map[i];
7461 
7462 		if (cap->cap_media != 0)
7463 			media_oper = cap->cap_media;
7464 	}
7465 
7466 	ifmr->ifm_status = IFM_AVALID;
7467 	if (proto_oper != 0) {
7468 		ifmr->ifm_status |= IFM_ACTIVE;
7469 		ifmr->ifm_active = IFM_ETHER | IFM_AUTO | media_oper;
7470 		/* txpause, rxpause, duplex? */
7471 	}
7472 }
7473 
7474 static int
7475 mcx_media_change(struct ifnet *ifp)
7476 {
7477 	struct mcx_softc *sc = (struct mcx_softc *)ifp->if_softc;
7478 	struct mcx_reg_ptys ptys;
7479 	struct mcx_reg_paos paos;
7480 	uint32_t media;
7481 	int i, error;
7482 
7483 	if (IFM_TYPE(sc->sc_media.ifm_media) != IFM_ETHER)
7484 		return EINVAL;
7485 
7486 	error = 0;
7487 
7488 	if (IFM_SUBTYPE(sc->sc_media.ifm_media) == IFM_AUTO) {
7489 		/* read ptys to get supported media */
7490 		memset(&ptys, 0, sizeof(ptys));
7491 		ptys.rp_local_port = 1;
7492 		ptys.rp_proto_mask = MCX_REG_PTYS_PROTO_MASK_ETH;
7493 		if (mcx_access_hca_reg(sc, MCX_REG_PTYS, MCX_REG_OP_READ,
7494 		    &ptys, sizeof(ptys)) != 0) {
7495 			printf("%s: unable to read port type/speed\n",
7496 			    DEVNAME(sc));
7497 			return EIO;
7498 		}
7499 
7500 		media = betoh32(ptys.rp_eth_proto_cap);
7501 	} else {
7502 		/* map media type */
7503 		media = 0;
7504 		for (i = 0; i < nitems(mcx_eth_cap_map); i++) {
7505 			const struct  mcx_eth_proto_capability *cap;
7506 
7507 			cap = &mcx_eth_cap_map[i];
7508 			if (cap->cap_media ==
7509 			    IFM_SUBTYPE(sc->sc_media.ifm_media)) {
7510 				media = (1 << i);
7511 				break;
7512 			}
7513 		}
7514 	}
7515 
7516 	/* disable the port */
7517 	memset(&paos, 0, sizeof(paos));
7518 	paos.rp_local_port = 1;
7519 	paos.rp_admin_status = MCX_REG_PAOS_ADMIN_STATUS_DOWN;
7520 	paos.rp_admin_state_update = MCX_REG_PAOS_ADMIN_STATE_UPDATE_EN;
7521 	if (mcx_access_hca_reg(sc, MCX_REG_PAOS, MCX_REG_OP_WRITE, &paos,
7522 	    sizeof(paos)) != 0) {
7523 		printf("%s: unable to set port state to down\n", DEVNAME(sc));
7524 		return EIO;
7525 	}
7526 
7527 	memset(&ptys, 0, sizeof(ptys));
7528 	ptys.rp_local_port = 1;
7529 	ptys.rp_proto_mask = MCX_REG_PTYS_PROTO_MASK_ETH;
7530 	ptys.rp_eth_proto_admin = htobe32(media);
7531 	if (mcx_access_hca_reg(sc, MCX_REG_PTYS, MCX_REG_OP_WRITE, &ptys,
7532 	    sizeof(ptys)) != 0) {
7533 		printf("%s: unable to set port media type/speed\n",
7534 		    DEVNAME(sc));
7535 		error = EIO;
7536 	}
7537 
7538 	/* re-enable the port to start negotiation */
7539 	memset(&paos, 0, sizeof(paos));
7540 	paos.rp_local_port = 1;
7541 	paos.rp_admin_status = MCX_REG_PAOS_ADMIN_STATUS_UP;
7542 	paos.rp_admin_state_update = MCX_REG_PAOS_ADMIN_STATE_UPDATE_EN;
7543 	if (mcx_access_hca_reg(sc, MCX_REG_PAOS, MCX_REG_OP_WRITE, &paos,
7544 	    sizeof(paos)) != 0) {
7545 		printf("%s: unable to set port state to up\n", DEVNAME(sc));
7546 		error = EIO;
7547 	}
7548 
7549 	return error;
7550 }
7551 
7552 static void
7553 mcx_port_change(void *xsc)
7554 {
7555 	struct mcx_softc *sc = xsc;
7556 	struct ifnet *ifp = &sc->sc_ac.ac_if;
7557 	struct mcx_reg_ptys ptys = {
7558 		.rp_local_port = 1,
7559 		.rp_proto_mask = MCX_REG_PTYS_PROTO_MASK_ETH,
7560 	};
7561 	int link_state = LINK_STATE_DOWN;
7562 
7563 	if (mcx_access_hca_reg(sc, MCX_REG_PTYS, MCX_REG_OP_READ, &ptys,
7564 	    sizeof(ptys)) == 0) {
7565 		uint32_t proto_oper = betoh32(ptys.rp_eth_proto_oper);
7566 		uint64_t baudrate = 0;
7567 		unsigned int i;
7568 
7569 		if (proto_oper != 0)
7570 			link_state = LINK_STATE_FULL_DUPLEX;
7571 
7572 		for (i = 0; i < nitems(mcx_eth_cap_map); i++) {
7573 			const struct mcx_eth_proto_capability *cap;
7574 			if (!ISSET(proto_oper, 1 << i))
7575 				continue;
7576 
7577 			cap = &mcx_eth_cap_map[i];
7578 			if (cap->cap_baudrate == 0)
7579 				continue;
7580 
7581 			baudrate = cap->cap_baudrate;
7582 			break;
7583 		}
7584 
7585 		ifp->if_baudrate = baudrate;
7586 	}
7587 
7588 	if (link_state != ifp->if_link_state) {
7589 		ifp->if_link_state = link_state;
7590 		if_link_state_change(ifp);
7591 	}
7592 }
7593 
7594 static inline uint32_t
7595 mcx_rd(struct mcx_softc *sc, bus_size_t r)
7596 {
7597 	uint32_t word;
7598 
7599 	word = bus_space_read_raw_4(sc->sc_memt, sc->sc_memh, r);
7600 
7601 	return (betoh32(word));
7602 }
7603 
7604 static inline void
7605 mcx_wr(struct mcx_softc *sc, bus_size_t r, uint32_t v)
7606 {
7607 	bus_space_write_raw_4(sc->sc_memt, sc->sc_memh, r, htobe32(v));
7608 }
7609 
7610 static inline void
7611 mcx_bar(struct mcx_softc *sc, bus_size_t r, bus_size_t l, int f)
7612 {
7613 	bus_space_barrier(sc->sc_memt, sc->sc_memh, r, l, f);
7614 }
7615 
7616 static uint64_t
7617 mcx_timer(struct mcx_softc *sc)
7618 {
7619 	uint32_t hi, lo, ni;
7620 
7621 	hi = mcx_rd(sc, MCX_INTERNAL_TIMER_H);
7622 	for (;;) {
7623 		lo = mcx_rd(sc, MCX_INTERNAL_TIMER_L);
7624 		mcx_bar(sc, MCX_INTERNAL_TIMER_L, 8, BUS_SPACE_BARRIER_READ);
7625 		ni = mcx_rd(sc, MCX_INTERNAL_TIMER_H);
7626 
7627 		if (ni == hi)
7628 			break;
7629 
7630 		hi = ni;
7631 	}
7632 
7633 	return (((uint64_t)hi << 32) | (uint64_t)lo);
7634 }
7635 
7636 static int
7637 mcx_dmamem_alloc(struct mcx_softc *sc, struct mcx_dmamem *mxm,
7638     bus_size_t size, u_int align)
7639 {
7640 	mxm->mxm_size = size;
7641 
7642 	if (bus_dmamap_create(sc->sc_dmat, mxm->mxm_size, 1,
7643 	    mxm->mxm_size, 0,
7644 	    BUS_DMA_WAITOK | BUS_DMA_ALLOCNOW | BUS_DMA_64BIT,
7645 	    &mxm->mxm_map) != 0)
7646 		return (1);
7647 	if (bus_dmamem_alloc(sc->sc_dmat, mxm->mxm_size,
7648 	    align, 0, &mxm->mxm_seg, 1, &mxm->mxm_nsegs,
7649 	    BUS_DMA_WAITOK | BUS_DMA_ZERO) != 0)
7650 		goto destroy;
7651 	if (bus_dmamem_map(sc->sc_dmat, &mxm->mxm_seg, mxm->mxm_nsegs,
7652 	    mxm->mxm_size, &mxm->mxm_kva, BUS_DMA_WAITOK) != 0)
7653 		goto free;
7654 	if (bus_dmamap_load(sc->sc_dmat, mxm->mxm_map, mxm->mxm_kva,
7655 	    mxm->mxm_size, NULL, BUS_DMA_WAITOK) != 0)
7656 		goto unmap;
7657 
7658 	return (0);
7659 unmap:
7660 	bus_dmamem_unmap(sc->sc_dmat, mxm->mxm_kva, mxm->mxm_size);
7661 free:
7662 	bus_dmamem_free(sc->sc_dmat, &mxm->mxm_seg, 1);
7663 destroy:
7664 	bus_dmamap_destroy(sc->sc_dmat, mxm->mxm_map);
7665 	return (1);
7666 }
7667 
7668 static void
7669 mcx_dmamem_zero(struct mcx_dmamem *mxm)
7670 {
7671 	memset(MCX_DMA_KVA(mxm), 0, MCX_DMA_LEN(mxm));
7672 }
7673 
7674 static void
7675 mcx_dmamem_free(struct mcx_softc *sc, struct mcx_dmamem *mxm)
7676 {
7677 	bus_dmamap_unload(sc->sc_dmat, mxm->mxm_map);
7678 	bus_dmamem_unmap(sc->sc_dmat, mxm->mxm_kva, mxm->mxm_size);
7679 	bus_dmamem_free(sc->sc_dmat, &mxm->mxm_seg, 1);
7680 	bus_dmamap_destroy(sc->sc_dmat, mxm->mxm_map);
7681 }
7682 
7683 static int
7684 mcx_hwmem_alloc(struct mcx_softc *sc, struct mcx_hwmem *mhm, unsigned int pages)
7685 {
7686 	bus_dma_segment_t *segs;
7687 	bus_size_t len = pages * MCX_PAGE_SIZE;
7688 	size_t seglen;
7689 
7690 	segs = mallocarray(sizeof(*segs), pages, M_DEVBUF, M_WAITOK|M_CANFAIL);
7691 	if (segs == NULL)
7692 		return (-1);
7693 
7694 	seglen = sizeof(*segs) * pages;
7695 
7696 	if (bus_dmamem_alloc(sc->sc_dmat, len, MCX_PAGE_SIZE, 0,
7697 	    segs, pages, &mhm->mhm_seg_count, BUS_DMA_NOWAIT) != 0)
7698 		goto free_segs;
7699 
7700 	if (mhm->mhm_seg_count < pages) {
7701 		size_t nseglen;
7702 
7703 		mhm->mhm_segs = mallocarray(sizeof(*mhm->mhm_segs),
7704 		    mhm->mhm_seg_count, M_DEVBUF, M_WAITOK|M_CANFAIL);
7705 		if (mhm->mhm_segs == NULL)
7706 			goto free_dmamem;
7707 
7708 		nseglen = sizeof(*mhm->mhm_segs) * mhm->mhm_seg_count;
7709 
7710 		memcpy(mhm->mhm_segs, segs, nseglen);
7711 
7712 		free(segs, M_DEVBUF, seglen);
7713 
7714 		segs = mhm->mhm_segs;
7715 		seglen = nseglen;
7716 	} else
7717 		mhm->mhm_segs = segs;
7718 
7719 	if (bus_dmamap_create(sc->sc_dmat, len, pages, MCX_PAGE_SIZE,
7720 	    MCX_PAGE_SIZE, BUS_DMA_NOWAIT|BUS_DMA_ALLOCNOW /*|BUS_DMA_64BIT*/,
7721 	    &mhm->mhm_map) != 0)
7722 		goto free_dmamem;
7723 
7724 	if (bus_dmamap_load_raw(sc->sc_dmat, mhm->mhm_map,
7725 	    mhm->mhm_segs, mhm->mhm_seg_count, len, BUS_DMA_NOWAIT) != 0)
7726 		goto destroy;
7727 
7728 	bus_dmamap_sync(sc->sc_dmat, mhm->mhm_map,
7729 	    0, mhm->mhm_map->dm_mapsize, BUS_DMASYNC_PRERW);
7730 
7731 	mhm->mhm_npages = pages;
7732 
7733 	return (0);
7734 
7735 destroy:
7736 	bus_dmamap_destroy(sc->sc_dmat, mhm->mhm_map);
7737 free_dmamem:
7738 	bus_dmamem_free(sc->sc_dmat, mhm->mhm_segs, mhm->mhm_seg_count);
7739 free_segs:
7740 	free(segs, M_DEVBUF, seglen);
7741 	mhm->mhm_segs = NULL;
7742 
7743 	return (-1);
7744 }
7745 
7746 static void
7747 mcx_hwmem_free(struct mcx_softc *sc, struct mcx_hwmem *mhm)
7748 {
7749 	if (mhm->mhm_npages == 0)
7750 		return;
7751 
7752 	bus_dmamap_sync(sc->sc_dmat, mhm->mhm_map,
7753 	    0, mhm->mhm_map->dm_mapsize, BUS_DMASYNC_POSTRW);
7754 
7755 	bus_dmamap_unload(sc->sc_dmat, mhm->mhm_map);
7756 	bus_dmamap_destroy(sc->sc_dmat, mhm->mhm_map);
7757 	bus_dmamem_free(sc->sc_dmat, mhm->mhm_segs, mhm->mhm_seg_count);
7758 	free(mhm->mhm_segs, M_DEVBUF,
7759 	    sizeof(*mhm->mhm_segs) * mhm->mhm_seg_count);
7760 
7761 	mhm->mhm_npages = 0;
7762 }
7763 
7764 #if NKSTAT > 0
7765 struct mcx_ppcnt {
7766 	char			 name[KSTAT_KV_NAMELEN];
7767 	enum kstat_kv_unit	 unit;
7768 };
7769 
7770 static const struct mcx_ppcnt mcx_ppcnt_ieee8023_tpl[] = {
7771 	{ "Good Tx",		KSTAT_KV_U_PACKETS, },
7772 	{ "Good Rx",		KSTAT_KV_U_PACKETS, },
7773 	{ "FCS errs",		KSTAT_KV_U_PACKETS, },
7774 	{ "Alignment Errs",	KSTAT_KV_U_PACKETS, },
7775 	{ "Good Tx",		KSTAT_KV_U_BYTES, },
7776 	{ "Good Rx",		KSTAT_KV_U_BYTES, },
7777 	{ "Multicast Tx",	KSTAT_KV_U_PACKETS, },
7778 	{ "Broadcast Tx",	KSTAT_KV_U_PACKETS, },
7779 	{ "Multicast Rx",	KSTAT_KV_U_PACKETS, },
7780 	{ "Broadcast Rx",	KSTAT_KV_U_PACKETS, },
7781 	{ "In Range Len",	KSTAT_KV_U_PACKETS, },
7782 	{ "Out Of Range Len",	KSTAT_KV_U_PACKETS, },
7783 	{ "Frame Too Long",	KSTAT_KV_U_PACKETS, },
7784 	{ "Symbol Errs",	KSTAT_KV_U_PACKETS, },
7785 	{ "MAC Ctrl Tx",	KSTAT_KV_U_PACKETS, },
7786 	{ "MAC Ctrl Rx",	KSTAT_KV_U_PACKETS, },
7787 	{ "MAC Ctrl Unsup",	KSTAT_KV_U_PACKETS, },
7788 	{ "Pause Rx",		KSTAT_KV_U_PACKETS, },
7789 	{ "Pause Tx",		KSTAT_KV_U_PACKETS, },
7790 };
7791 CTASSERT(nitems(mcx_ppcnt_ieee8023_tpl) == mcx_ppcnt_ieee8023_count);
7792 
7793 static const struct mcx_ppcnt mcx_ppcnt_rfc2863_tpl[] = {
7794 	{ "Rx Bytes",		KSTAT_KV_U_BYTES, },
7795 	{ "Rx Unicast",		KSTAT_KV_U_PACKETS, },
7796 	{ "Rx Discards",	KSTAT_KV_U_PACKETS, },
7797 	{ "Rx Errors",		KSTAT_KV_U_PACKETS, },
7798 	{ "Rx Unknown Proto",	KSTAT_KV_U_PACKETS, },
7799 	{ "Tx Bytes",		KSTAT_KV_U_BYTES, },
7800 	{ "Tx Unicast",		KSTAT_KV_U_PACKETS, },
7801 	{ "Tx Discards",	KSTAT_KV_U_PACKETS, },
7802 	{ "Tx Errors",		KSTAT_KV_U_PACKETS, },
7803 	{ "Rx Multicast",	KSTAT_KV_U_PACKETS, },
7804 	{ "Rx Broadcast",	KSTAT_KV_U_PACKETS, },
7805 	{ "Tx Multicast",	KSTAT_KV_U_PACKETS, },
7806 	{ "Tx Broadcast",	KSTAT_KV_U_PACKETS, },
7807 };
7808 CTASSERT(nitems(mcx_ppcnt_rfc2863_tpl) == mcx_ppcnt_rfc2863_count);
7809 
7810 static const struct mcx_ppcnt mcx_ppcnt_rfc2819_tpl[] = {
7811 	{ "Drop Events",	KSTAT_KV_U_PACKETS, },
7812 	{ "Octets",		KSTAT_KV_U_BYTES, },
7813 	{ "Packets",		KSTAT_KV_U_PACKETS, },
7814 	{ "Broadcasts",		KSTAT_KV_U_PACKETS, },
7815 	{ "Multicasts",		KSTAT_KV_U_PACKETS, },
7816 	{ "CRC Align Errs",	KSTAT_KV_U_PACKETS, },
7817 	{ "Undersize",		KSTAT_KV_U_PACKETS, },
7818 	{ "Oversize",		KSTAT_KV_U_PACKETS, },
7819 	{ "Fragments",		KSTAT_KV_U_PACKETS, },
7820 	{ "Jabbers",		KSTAT_KV_U_PACKETS, },
7821 	{ "Collisions",		KSTAT_KV_U_NONE, },
7822 	{ "64B",		KSTAT_KV_U_PACKETS, },
7823 	{ "65-127B",		KSTAT_KV_U_PACKETS, },
7824 	{ "128-255B",		KSTAT_KV_U_PACKETS, },
7825 	{ "256-511B",		KSTAT_KV_U_PACKETS, },
7826 	{ "512-1023B",		KSTAT_KV_U_PACKETS, },
7827 	{ "1024-1518B",		KSTAT_KV_U_PACKETS, },
7828 	{ "1519-2047B",		KSTAT_KV_U_PACKETS, },
7829 	{ "2048-4095B",		KSTAT_KV_U_PACKETS, },
7830 	{ "4096-8191B",		KSTAT_KV_U_PACKETS, },
7831 	{ "8192-10239B",	KSTAT_KV_U_PACKETS, },
7832 };
7833 CTASSERT(nitems(mcx_ppcnt_rfc2819_tpl) == mcx_ppcnt_rfc2819_count);
7834 
7835 static const struct mcx_ppcnt mcx_ppcnt_rfc3635_tpl[] = {
7836 	{ "Alignment Errs",	KSTAT_KV_U_PACKETS, },
7837 	{ "FCS Errs",		KSTAT_KV_U_PACKETS, },
7838 	{ "Single Colls",	KSTAT_KV_U_PACKETS, },
7839 	{ "Multiple Colls",	KSTAT_KV_U_PACKETS, },
7840 	{ "SQE Test Errs",	KSTAT_KV_U_NONE, },
7841 	{ "Deferred Tx",	KSTAT_KV_U_PACKETS, },
7842 	{ "Late Colls",		KSTAT_KV_U_NONE, },
7843 	{ "Exess Colls",	KSTAT_KV_U_NONE, },
7844 	{ "Int MAC Tx Errs",	KSTAT_KV_U_PACKETS, },
7845 	{ "CSM Sense Errs",	KSTAT_KV_U_NONE, },
7846 	{ "Too Long",		KSTAT_KV_U_PACKETS, },
7847 	{ "Int MAC Rx Errs",	KSTAT_KV_U_PACKETS, },
7848 	{ "Symbol Errs",	KSTAT_KV_U_NONE, },
7849 	{ "Unknown Control",	KSTAT_KV_U_PACKETS, },
7850 	{ "Pause Rx",		KSTAT_KV_U_PACKETS, },
7851 	{ "Pause Tx",		KSTAT_KV_U_PACKETS, },
7852 };
7853 CTASSERT(nitems(mcx_ppcnt_rfc3635_tpl) == mcx_ppcnt_rfc3635_count);
7854 
7855 struct mcx_kstat_ppcnt {
7856 	const char		*ksp_name;
7857 	const struct mcx_ppcnt	*ksp_tpl;
7858 	unsigned int		 ksp_n;
7859 	uint8_t			 ksp_grp;
7860 };
7861 
7862 static const struct mcx_kstat_ppcnt mcx_kstat_ppcnt_ieee8023 = {
7863 	.ksp_name =		"ieee802.3",
7864 	.ksp_tpl =		mcx_ppcnt_ieee8023_tpl,
7865 	.ksp_n =		nitems(mcx_ppcnt_ieee8023_tpl),
7866 	.ksp_grp =		MCX_REG_PPCNT_GRP_IEEE8023,
7867 };
7868 
7869 static const struct mcx_kstat_ppcnt mcx_kstat_ppcnt_rfc2863 = {
7870 	.ksp_name =		"rfc2863",
7871 	.ksp_tpl =		mcx_ppcnt_rfc2863_tpl,
7872 	.ksp_n =		nitems(mcx_ppcnt_rfc2863_tpl),
7873 	.ksp_grp =		MCX_REG_PPCNT_GRP_RFC2863,
7874 };
7875 
7876 static const struct mcx_kstat_ppcnt mcx_kstat_ppcnt_rfc2819 = {
7877 	.ksp_name =		"rfc2819",
7878 	.ksp_tpl =		mcx_ppcnt_rfc2819_tpl,
7879 	.ksp_n =		nitems(mcx_ppcnt_rfc2819_tpl),
7880 	.ksp_grp =		MCX_REG_PPCNT_GRP_RFC2819,
7881 };
7882 
7883 static const struct mcx_kstat_ppcnt mcx_kstat_ppcnt_rfc3635 = {
7884 	.ksp_name =		"rfc3635",
7885 	.ksp_tpl =		mcx_ppcnt_rfc3635_tpl,
7886 	.ksp_n =		nitems(mcx_ppcnt_rfc3635_tpl),
7887 	.ksp_grp =		MCX_REG_PPCNT_GRP_RFC3635,
7888 };
7889 
7890 static int	mcx_kstat_ppcnt_read(struct kstat *);
7891 
7892 static void	mcx_kstat_attach_tmps(struct mcx_softc *sc);
7893 
7894 static struct kstat *
7895 mcx_kstat_attach_ppcnt(struct mcx_softc *sc,
7896     const struct mcx_kstat_ppcnt *ksp)
7897 {
7898 	struct kstat *ks;
7899 	struct kstat_kv *kvs;
7900 	unsigned int i;
7901 
7902 	ks = kstat_create(DEVNAME(sc), 0, ksp->ksp_name, 0, KSTAT_T_KV, 0);
7903 	if (ks == NULL)
7904 		return (NULL);
7905 
7906 	kvs = mallocarray(ksp->ksp_n, sizeof(*kvs),
7907 	    M_DEVBUF, M_WAITOK);
7908 
7909 	for (i = 0; i < ksp->ksp_n; i++) {
7910 		const struct mcx_ppcnt *tpl = &ksp->ksp_tpl[i];
7911 
7912 		kstat_kv_unit_init(&kvs[i], tpl->name,
7913 		    KSTAT_KV_T_COUNTER64, tpl->unit);
7914 	}
7915 
7916 	ks->ks_softc = sc;
7917 	ks->ks_ptr = (void *)ksp;
7918 	ks->ks_data = kvs;
7919 	ks->ks_datalen = ksp->ksp_n * sizeof(*kvs);
7920 	ks->ks_read = mcx_kstat_ppcnt_read;
7921 
7922 	kstat_install(ks);
7923 
7924 	return (ks);
7925 }
7926 
7927 static void
7928 mcx_kstat_attach(struct mcx_softc *sc)
7929 {
7930 	sc->sc_kstat_ieee8023 = mcx_kstat_attach_ppcnt(sc,
7931 	    &mcx_kstat_ppcnt_ieee8023);
7932 	sc->sc_kstat_rfc2863 = mcx_kstat_attach_ppcnt(sc,
7933 	    &mcx_kstat_ppcnt_rfc2863);
7934 	sc->sc_kstat_rfc2819 = mcx_kstat_attach_ppcnt(sc,
7935 	    &mcx_kstat_ppcnt_rfc2819);
7936 	sc->sc_kstat_rfc3635 = mcx_kstat_attach_ppcnt(sc,
7937 	    &mcx_kstat_ppcnt_rfc3635);
7938 
7939 	mcx_kstat_attach_tmps(sc);
7940 }
7941 
7942 static int
7943 mcx_kstat_ppcnt_read(struct kstat *ks)
7944 {
7945 	struct mcx_softc *sc = ks->ks_softc;
7946 	struct mcx_kstat_ppcnt *ksp = ks->ks_ptr;
7947 	struct mcx_reg_ppcnt ppcnt = {
7948 		.ppcnt_grp = ksp->ksp_grp,
7949 		.ppcnt_local_port = 1,
7950 	};
7951 	struct kstat_kv *kvs = ks->ks_data;
7952 	uint64_t *vs = (uint64_t *)&ppcnt.ppcnt_counter_set;
7953 	unsigned int i;
7954 	int rv;
7955 
7956 	KERNEL_LOCK(); /* XXX */
7957 	rv = mcx_access_hca_reg(sc, MCX_REG_PPCNT, MCX_REG_OP_READ,
7958 	    &ppcnt, sizeof(ppcnt));
7959 	KERNEL_UNLOCK();
7960 	if (rv != 0)
7961 		return (EIO);
7962 
7963 	nanouptime(&ks->ks_updated);
7964 
7965 	for (i = 0; i < ksp->ksp_n; i++)
7966 		kstat_kv_u64(&kvs[i]) = bemtoh64(&vs[i]);
7967 
7968 	return (0);
7969 }
7970 
7971 struct mcx_kstat_mtmp {
7972 	struct kstat_kv		ktmp_name;
7973 	struct kstat_kv		ktmp_temperature;
7974 	struct kstat_kv		ktmp_threshold_lo;
7975 	struct kstat_kv		ktmp_threshold_hi;
7976 };
7977 
7978 static const struct mcx_kstat_mtmp mcx_kstat_mtmp_tpl = {
7979 	KSTAT_KV_INITIALIZER("name",		KSTAT_KV_T_ISTR),
7980 	KSTAT_KV_INITIALIZER("temperature",	KSTAT_KV_T_TEMP),
7981 	KSTAT_KV_INITIALIZER("lo threshold",	KSTAT_KV_T_TEMP),
7982 	KSTAT_KV_INITIALIZER("hi threshold",	KSTAT_KV_T_TEMP),
7983 };
7984 
7985 static const struct timeval mcx_kstat_mtmp_rate = { 1, 0 };
7986 
7987 static int mcx_kstat_mtmp_read(struct kstat *);
7988 
7989 static void
7990 mcx_kstat_attach_tmps(struct mcx_softc *sc)
7991 {
7992 	struct kstat *ks;
7993 	struct mcx_reg_mtcap mtcap;
7994 	struct mcx_kstat_mtmp *ktmp;
7995 	uint64_t map;
7996 	unsigned int i, n;
7997 
7998 	memset(&mtcap, 0, sizeof(mtcap));
7999 
8000 	if (mcx_access_hca_reg(sc, MCX_REG_MTCAP, MCX_REG_OP_READ,
8001 	    &mtcap, sizeof(mtcap)) != 0) {
8002 		/* unable to find temperature sensors */
8003 		return;
8004 	}
8005 
8006 	sc->sc_kstat_mtmp_count = mtcap.mtcap_sensor_count;
8007 	sc->sc_kstat_mtmp = mallocarray(sc->sc_kstat_mtmp_count,
8008 	    sizeof(*sc->sc_kstat_mtmp), M_DEVBUF, M_WAITOK);
8009 
8010 	n = 0;
8011 	map = bemtoh64(&mtcap.mtcap_sensor_map);
8012 	for (i = 0; i < sizeof(map) * NBBY; i++) {
8013 		if (!ISSET(map, (1ULL << i)))
8014 			continue;
8015 
8016 		ks = kstat_create(DEVNAME(sc), 0, "temperature", i,
8017 		    KSTAT_T_KV, 0);
8018 		if (ks == NULL) {
8019 			/* unable to attach temperature sensor %u, i */
8020 			continue;
8021 		}
8022 
8023 		ktmp = malloc(sizeof(*ktmp), M_DEVBUF, M_WAITOK|M_ZERO);
8024 		*ktmp = mcx_kstat_mtmp_tpl;
8025 
8026 		ks->ks_data = ktmp;
8027 		ks->ks_datalen = sizeof(*ktmp);
8028 		TIMEVAL_TO_TIMESPEC(&mcx_kstat_mtmp_rate, &ks->ks_interval);
8029 		ks->ks_read = mcx_kstat_mtmp_read;
8030 
8031 		ks->ks_softc = sc;
8032 		kstat_install(ks);
8033 
8034 		sc->sc_kstat_mtmp[n++] = ks;
8035 		if (n >= sc->sc_kstat_mtmp_count)
8036 			break;
8037 	}
8038 }
8039 
8040 static uint64_t
8041 mcx_tmp_to_uK(uint16_t *t)
8042 {
8043 	int64_t mt = (int16_t)bemtoh16(t); /* 0.125 C units */
8044 	mt *= 1000000 / 8; /* convert to uC */
8045 	mt += 273150000; /* convert to uK */
8046 
8047 	return (mt);
8048 }
8049 
8050 static int
8051 mcx_kstat_mtmp_read(struct kstat *ks)
8052 {
8053 	struct mcx_softc *sc = ks->ks_softc;
8054 	struct mcx_kstat_mtmp *ktmp = ks->ks_data;
8055 	struct mcx_reg_mtmp mtmp;
8056 	int rv;
8057 	struct timeval updated;
8058 
8059 	TIMESPEC_TO_TIMEVAL(&updated, &ks->ks_updated);
8060 
8061 	if (!ratecheck(&updated, &mcx_kstat_mtmp_rate))
8062 		return (0);
8063 
8064 	memset(&mtmp, 0, sizeof(mtmp));
8065 	htobem16(&mtmp.mtmp_sensor_index, ks->ks_unit);
8066 
8067 	KERNEL_LOCK(); /* XXX */
8068 	rv = mcx_access_hca_reg(sc, MCX_REG_MTMP, MCX_REG_OP_READ,
8069 	    &mtmp, sizeof(mtmp));
8070 	KERNEL_UNLOCK();
8071 	if (rv != 0)
8072 		return (EIO);
8073 
8074 	memset(kstat_kv_istr(&ktmp->ktmp_name), 0,
8075 	    sizeof(kstat_kv_istr(&ktmp->ktmp_name)));
8076 	memcpy(kstat_kv_istr(&ktmp->ktmp_name),
8077 	    mtmp.mtmp_sensor_name, sizeof(mtmp.mtmp_sensor_name));
8078 	kstat_kv_temp(&ktmp->ktmp_temperature) =
8079 	    mcx_tmp_to_uK(&mtmp.mtmp_temperature);
8080 	kstat_kv_temp(&ktmp->ktmp_threshold_lo) =
8081 	    mcx_tmp_to_uK(&mtmp.mtmp_temperature_threshold_lo);
8082 	kstat_kv_temp(&ktmp->ktmp_threshold_hi) =
8083 	    mcx_tmp_to_uK(&mtmp.mtmp_temperature_threshold_hi);
8084 
8085 	TIMEVAL_TO_TIMESPEC(&updated, &ks->ks_updated);
8086 
8087 	return (0);
8088 }
8089 #endif /* NKSTAT > 0 */
8090