xref: /openbsd-src/sys/dev/pci/if_mcx.c (revision cba26e98faa2b48aa4705f205ed876af460243a2)
1 /*	$OpenBSD: if_mcx.c,v 1.93 2021/01/04 23:12:05 dlg 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 #include <sys/timetc.h>
37 
38 #include <machine/bus.h>
39 #include <machine/intr.h>
40 
41 #include <net/if.h>
42 #include <net/if_dl.h>
43 #include <net/if_media.h>
44 #include <net/toeplitz.h>
45 
46 #if NBPFILTER > 0
47 #include <net/bpf.h>
48 #endif
49 
50 #if NKSTAT > 0
51 #include <sys/kstat.h>
52 #endif
53 
54 #include <netinet/in.h>
55 #include <netinet/if_ether.h>
56 
57 #include <dev/pci/pcireg.h>
58 #include <dev/pci/pcivar.h>
59 #include <dev/pci/pcidevs.h>
60 
61 #define BUS_DMASYNC_PRERW	(BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE)
62 #define BUS_DMASYNC_POSTRW	(BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE)
63 
64 #define MCX_HCA_BAR	PCI_MAPREG_START /* BAR 0 */
65 
66 #define MCX_FW_VER			0x0000
67 #define  MCX_FW_VER_MAJOR(_v)			((_v) & 0xffff)
68 #define  MCX_FW_VER_MINOR(_v)			((_v) >> 16)
69 #define MCX_CMDIF_FW_SUBVER		0x0004
70 #define  MCX_FW_VER_SUBMINOR(_v)		((_v) & 0xffff)
71 #define  MCX_CMDIF(_v)				((_v) >> 16)
72 
73 #define MCX_ISSI			1 /* as per the PRM */
74 #define MCX_CMD_IF_SUPPORTED		5
75 
76 #define MCX_HARDMTU			9500
77 
78 #define MCX_PAGE_SHIFT			12
79 #define MCX_PAGE_SIZE			(1 << MCX_PAGE_SHIFT)
80 
81 /* queue sizes */
82 #define MCX_LOG_EQ_SIZE			7
83 #define MCX_LOG_CQ_SIZE			12
84 #define MCX_LOG_RQ_SIZE			10
85 #define MCX_LOG_SQ_SIZE			11
86 
87 #define MCX_MAX_QUEUES			1
88 
89 /* completion event moderation - about 10khz, or 90% of the cq */
90 #define MCX_CQ_MOD_PERIOD		50
91 #define MCX_CQ_MOD_COUNTER		\
92 	(((1 << (MCX_LOG_CQ_SIZE - 1)) * 9) / 10)
93 
94 #define MCX_LOG_SQ_ENTRY_SIZE		6
95 #define MCX_SQ_ENTRY_MAX_SLOTS		4
96 #define MCX_SQ_SEGS_PER_SLOT		\
97 	(sizeof(struct mcx_sq_entry) / sizeof(struct mcx_sq_entry_seg))
98 #define MCX_SQ_MAX_SEGMENTS		\
99 	1 + ((MCX_SQ_ENTRY_MAX_SLOTS-1) * MCX_SQ_SEGS_PER_SLOT)
100 
101 #define MCX_LOG_FLOW_TABLE_SIZE		5
102 #define MCX_NUM_STATIC_FLOWS		4 /* promisc, allmulti, ucast, bcast */
103 #define MCX_NUM_MCAST_FLOWS 		\
104 	((1 << MCX_LOG_FLOW_TABLE_SIZE) - MCX_NUM_STATIC_FLOWS)
105 
106 #define MCX_SQ_INLINE_SIZE		18
107 CTASSERT(ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN == MCX_SQ_INLINE_SIZE);
108 
109 /* doorbell offsets */
110 #define MCX_DOORBELL_AREA_SIZE		MCX_PAGE_SIZE
111 
112 #define MCX_CQ_DOORBELL_BASE		0
113 #define MCX_CQ_DOORBELL_STRIDE		64
114 
115 #define MCX_WQ_DOORBELL_BASE		MCX_PAGE_SIZE/2
116 #define MCX_WQ_DOORBELL_STRIDE		64
117 /* make sure the doorbells fit */
118 CTASSERT(MCX_MAX_QUEUES * MCX_CQ_DOORBELL_STRIDE < MCX_WQ_DOORBELL_BASE);
119 CTASSERT(MCX_MAX_QUEUES * MCX_WQ_DOORBELL_STRIDE <
120     MCX_DOORBELL_AREA_SIZE - MCX_WQ_DOORBELL_BASE);
121 
122 #define MCX_WQ_DOORBELL_MASK		0xffff
123 
124 /* uar registers */
125 #define MCX_UAR_CQ_DOORBELL		0x20
126 #define MCX_UAR_EQ_DOORBELL_ARM		0x40
127 #define MCX_UAR_EQ_DOORBELL		0x48
128 #define MCX_UAR_BF			0x800
129 
130 #define MCX_CMDQ_ADDR_HI		0x0010
131 #define MCX_CMDQ_ADDR_LO		0x0014
132 #define MCX_CMDQ_ADDR_NMASK		0xfff
133 #define MCX_CMDQ_LOG_SIZE(_v)		((_v) >> 4 & 0xf)
134 #define MCX_CMDQ_LOG_STRIDE(_v)		((_v) >> 0 & 0xf)
135 #define MCX_CMDQ_INTERFACE_MASK		(0x3 << 8)
136 #define MCX_CMDQ_INTERFACE_FULL_DRIVER	(0x0 << 8)
137 #define MCX_CMDQ_INTERFACE_DISABLED	(0x1 << 8)
138 
139 #define MCX_CMDQ_DOORBELL		0x0018
140 
141 #define MCX_STATE			0x01fc
142 #define MCX_STATE_MASK				(1 << 31)
143 #define MCX_STATE_INITIALIZING			(1 << 31)
144 #define MCX_STATE_READY				(0 << 31)
145 #define MCX_STATE_INTERFACE_MASK		(0x3 << 24)
146 #define MCX_STATE_INTERFACE_FULL_DRIVER		(0x0 << 24)
147 #define MCX_STATE_INTERFACE_DISABLED		(0x1 << 24)
148 
149 #define MCX_INTERNAL_TIMER		0x1000
150 #define MCX_INTERNAL_TIMER_H		0x1000
151 #define MCX_INTERNAL_TIMER_L		0x1004
152 
153 #define MCX_CLEAR_INT			0x100c
154 
155 #define MCX_REG_OP_WRITE		0
156 #define MCX_REG_OP_READ			1
157 
158 #define MCX_REG_PMLP			0x5002
159 #define MCX_REG_PMTU			0x5003
160 #define MCX_REG_PTYS			0x5004
161 #define MCX_REG_PAOS			0x5006
162 #define MCX_REG_PFCC			0x5007
163 #define MCX_REG_PPCNT			0x5008
164 #define MCX_REG_MTCAP			0x9009 /* mgmt temp capabilities */
165 #define MCX_REG_MTMP			0x900a /* mgmt temp */
166 #define MCX_REG_MCIA			0x9014
167 
168 #define MCX_ETHER_CAP_SGMII		0
169 #define MCX_ETHER_CAP_1000_KX		1
170 #define MCX_ETHER_CAP_10G_CX4		2
171 #define MCX_ETHER_CAP_10G_KX4		3
172 #define MCX_ETHER_CAP_10G_KR		4
173 #define MCX_ETHER_CAP_40G_CR4		6
174 #define MCX_ETHER_CAP_40G_KR4		7
175 #define MCX_ETHER_CAP_10G_CR		12
176 #define MCX_ETHER_CAP_10G_SR		13
177 #define MCX_ETHER_CAP_10G_LR		14
178 #define MCX_ETHER_CAP_40G_SR4		15
179 #define MCX_ETHER_CAP_40G_LR4		16
180 #define MCX_ETHER_CAP_50G_SR2		18
181 #define MCX_ETHER_CAP_100G_CR4		20
182 #define MCX_ETHER_CAP_100G_SR4		21
183 #define MCX_ETHER_CAP_100G_KR4		22
184 #define MCX_ETHER_CAP_25G_CR		27
185 #define MCX_ETHER_CAP_25G_KR		28
186 #define MCX_ETHER_CAP_25G_SR		29
187 #define MCX_ETHER_CAP_50G_CR2		30
188 #define MCX_ETHER_CAP_50G_KR2		31
189 
190 #define MCX_MAX_CQE			32
191 
192 #define MCX_CMD_QUERY_HCA_CAP		0x100
193 #define MCX_CMD_QUERY_ADAPTER		0x101
194 #define MCX_CMD_INIT_HCA		0x102
195 #define MCX_CMD_TEARDOWN_HCA		0x103
196 #define MCX_CMD_ENABLE_HCA		0x104
197 #define MCX_CMD_DISABLE_HCA		0x105
198 #define MCX_CMD_QUERY_PAGES		0x107
199 #define MCX_CMD_MANAGE_PAGES		0x108
200 #define MCX_CMD_SET_HCA_CAP		0x109
201 #define MCX_CMD_QUERY_ISSI		0x10a
202 #define MCX_CMD_SET_ISSI		0x10b
203 #define MCX_CMD_SET_DRIVER_VERSION	0x10d
204 #define MCX_CMD_QUERY_SPECIAL_CONTEXTS	0x203
205 #define MCX_CMD_CREATE_EQ		0x301
206 #define MCX_CMD_DESTROY_EQ		0x302
207 #define MCX_CMD_QUERY_EQ		0x303
208 #define MCX_CMD_CREATE_CQ		0x400
209 #define MCX_CMD_DESTROY_CQ		0x401
210 #define MCX_CMD_QUERY_CQ		0x402
211 #define MCX_CMD_QUERY_NIC_VPORT_CONTEXT	0x754
212 #define MCX_CMD_MODIFY_NIC_VPORT_CONTEXT \
213 					0x755
214 #define MCX_CMD_QUERY_VPORT_COUNTERS	0x770
215 #define MCX_CMD_ALLOC_PD		0x800
216 #define MCX_CMD_ALLOC_UAR		0x802
217 #define MCX_CMD_ACCESS_REG		0x805
218 #define MCX_CMD_ALLOC_TRANSPORT_DOMAIN	0x816
219 #define MCX_CMD_CREATE_TIR		0x900
220 #define MCX_CMD_DESTROY_TIR		0x902
221 #define MCX_CMD_CREATE_SQ		0x904
222 #define MCX_CMD_MODIFY_SQ		0x905
223 #define MCX_CMD_DESTROY_SQ		0x906
224 #define MCX_CMD_QUERY_SQ		0x907
225 #define MCX_CMD_CREATE_RQ		0x908
226 #define MCX_CMD_MODIFY_RQ		0x909
227 #define MCX_CMD_DESTROY_RQ		0x90a
228 #define MCX_CMD_QUERY_RQ		0x90b
229 #define MCX_CMD_CREATE_TIS		0x912
230 #define MCX_CMD_DESTROY_TIS		0x914
231 #define MCX_CMD_CREATE_RQT		0x916
232 #define MCX_CMD_DESTROY_RQT		0x918
233 #define MCX_CMD_SET_FLOW_TABLE_ROOT	0x92f
234 #define MCX_CMD_CREATE_FLOW_TABLE	0x930
235 #define MCX_CMD_DESTROY_FLOW_TABLE	0x931
236 #define MCX_CMD_QUERY_FLOW_TABLE	0x932
237 #define MCX_CMD_CREATE_FLOW_GROUP	0x933
238 #define MCX_CMD_DESTROY_FLOW_GROUP	0x934
239 #define MCX_CMD_QUERY_FLOW_GROUP	0x935
240 #define MCX_CMD_SET_FLOW_TABLE_ENTRY	0x936
241 #define MCX_CMD_QUERY_FLOW_TABLE_ENTRY	0x937
242 #define MCX_CMD_DELETE_FLOW_TABLE_ENTRY	0x938
243 #define MCX_CMD_ALLOC_FLOW_COUNTER	0x939
244 #define MCX_CMD_QUERY_FLOW_COUNTER	0x93b
245 
246 #define MCX_QUEUE_STATE_RST		0
247 #define MCX_QUEUE_STATE_RDY		1
248 #define MCX_QUEUE_STATE_ERR		3
249 
250 #define MCX_FLOW_TABLE_TYPE_RX		0
251 #define MCX_FLOW_TABLE_TYPE_TX		1
252 
253 #define MCX_CMDQ_INLINE_DATASIZE 16
254 
255 struct mcx_cmdq_entry {
256 	uint8_t			cq_type;
257 #define MCX_CMDQ_TYPE_PCIE		0x7
258 	uint8_t			cq_reserved0[3];
259 
260 	uint32_t		cq_input_length;
261 	uint64_t		cq_input_ptr;
262 	uint8_t			cq_input_data[MCX_CMDQ_INLINE_DATASIZE];
263 
264 	uint8_t			cq_output_data[MCX_CMDQ_INLINE_DATASIZE];
265 	uint64_t		cq_output_ptr;
266 	uint32_t		cq_output_length;
267 
268 	uint8_t			cq_token;
269 	uint8_t			cq_signature;
270 	uint8_t			cq_reserved1[1];
271 	uint8_t			cq_status;
272 #define MCX_CQ_STATUS_SHIFT		1
273 #define MCX_CQ_STATUS_MASK		(0x7f << MCX_CQ_STATUS_SHIFT)
274 #define MCX_CQ_STATUS_OK		(0x00 << MCX_CQ_STATUS_SHIFT)
275 #define MCX_CQ_STATUS_INT_ERR		(0x01 << MCX_CQ_STATUS_SHIFT)
276 #define MCX_CQ_STATUS_BAD_OPCODE	(0x02 << MCX_CQ_STATUS_SHIFT)
277 #define MCX_CQ_STATUS_BAD_PARAM		(0x03 << MCX_CQ_STATUS_SHIFT)
278 #define MCX_CQ_STATUS_BAD_SYS_STATE	(0x04 << MCX_CQ_STATUS_SHIFT)
279 #define MCX_CQ_STATUS_BAD_RESOURCE	(0x05 << MCX_CQ_STATUS_SHIFT)
280 #define MCX_CQ_STATUS_RESOURCE_BUSY	(0x06 << MCX_CQ_STATUS_SHIFT)
281 #define MCX_CQ_STATUS_EXCEED_LIM	(0x08 << MCX_CQ_STATUS_SHIFT)
282 #define MCX_CQ_STATUS_BAD_RES_STATE	(0x09 << MCX_CQ_STATUS_SHIFT)
283 #define MCX_CQ_STATUS_BAD_INDEX		(0x0a << MCX_CQ_STATUS_SHIFT)
284 #define MCX_CQ_STATUS_NO_RESOURCES	(0x0f << MCX_CQ_STATUS_SHIFT)
285 #define MCX_CQ_STATUS_BAD_INPUT_LEN	(0x50 << MCX_CQ_STATUS_SHIFT)
286 #define MCX_CQ_STATUS_BAD_OUTPUT_LEN	(0x51 << MCX_CQ_STATUS_SHIFT)
287 #define MCX_CQ_STATUS_BAD_RESOURCE_STATE \
288 					(0x10 << MCX_CQ_STATUS_SHIFT)
289 #define MCX_CQ_STATUS_BAD_SIZE		(0x40 << MCX_CQ_STATUS_SHIFT)
290 #define MCX_CQ_STATUS_OWN_MASK		0x1
291 #define MCX_CQ_STATUS_OWN_SW		0x0
292 #define MCX_CQ_STATUS_OWN_HW		0x1
293 } __packed __aligned(8);
294 
295 #define MCX_CMDQ_MAILBOX_DATASIZE	512
296 
297 struct mcx_cmdq_mailbox {
298 	uint8_t			mb_data[MCX_CMDQ_MAILBOX_DATASIZE];
299 	uint8_t			mb_reserved0[48];
300 	uint64_t		mb_next_ptr;
301 	uint32_t		mb_block_number;
302 	uint8_t			mb_reserved1[1];
303 	uint8_t			mb_token;
304 	uint8_t			mb_ctrl_signature;
305 	uint8_t			mb_signature;
306 } __packed __aligned(8);
307 
308 #define MCX_CMDQ_MAILBOX_ALIGN	(1 << 10)
309 #define MCX_CMDQ_MAILBOX_SIZE	roundup(sizeof(struct mcx_cmdq_mailbox), \
310 				    MCX_CMDQ_MAILBOX_ALIGN)
311 /*
312  * command mailbox structres
313  */
314 
315 struct mcx_cmd_enable_hca_in {
316 	uint16_t		cmd_opcode;
317 	uint8_t			cmd_reserved0[4];
318 	uint16_t		cmd_op_mod;
319 	uint8_t			cmd_reserved1[2];
320 	uint16_t		cmd_function_id;
321 	uint8_t			cmd_reserved2[4];
322 } __packed __aligned(4);
323 
324 struct mcx_cmd_enable_hca_out {
325 	uint8_t			cmd_status;
326 	uint8_t			cmd_reserved0[3];
327 	uint32_t		cmd_syndrome;
328 	uint8_t			cmd_reserved1[4];
329 } __packed __aligned(4);
330 
331 struct mcx_cmd_init_hca_in {
332 	uint16_t		cmd_opcode;
333 	uint8_t			cmd_reserved0[4];
334 	uint16_t		cmd_op_mod;
335 	uint8_t			cmd_reserved1[8];
336 } __packed __aligned(4);
337 
338 struct mcx_cmd_init_hca_out {
339 	uint8_t			cmd_status;
340 	uint8_t			cmd_reserved0[3];
341 	uint32_t		cmd_syndrome;
342 	uint8_t			cmd_reserved1[8];
343 } __packed __aligned(4);
344 
345 struct mcx_cmd_teardown_hca_in {
346 	uint16_t		cmd_opcode;
347 	uint8_t			cmd_reserved0[4];
348 	uint16_t		cmd_op_mod;
349 	uint8_t			cmd_reserved1[2];
350 #define MCX_CMD_TEARDOWN_HCA_GRACEFUL	0x0
351 #define MCX_CMD_TEARDOWN_HCA_PANIC	0x1
352 	uint16_t		cmd_profile;
353 	uint8_t			cmd_reserved2[4];
354 } __packed __aligned(4);
355 
356 struct mcx_cmd_teardown_hca_out {
357 	uint8_t			cmd_status;
358 	uint8_t			cmd_reserved0[3];
359 	uint32_t		cmd_syndrome;
360 	uint8_t			cmd_reserved1[8];
361 } __packed __aligned(4);
362 
363 struct mcx_cmd_access_reg_in {
364 	uint16_t		cmd_opcode;
365 	uint8_t			cmd_reserved0[4];
366 	uint16_t		cmd_op_mod;
367 	uint8_t			cmd_reserved1[2];
368 	uint16_t		cmd_register_id;
369 	uint32_t		cmd_argument;
370 } __packed __aligned(4);
371 
372 struct mcx_cmd_access_reg_out {
373 	uint8_t			cmd_status;
374 	uint8_t			cmd_reserved0[3];
375 	uint32_t		cmd_syndrome;
376 	uint8_t			cmd_reserved1[8];
377 } __packed __aligned(4);
378 
379 struct mcx_reg_pmtu {
380 	uint8_t			rp_reserved1;
381 	uint8_t			rp_local_port;
382 	uint8_t			rp_reserved2[2];
383 	uint16_t		rp_max_mtu;
384 	uint8_t			rp_reserved3[2];
385 	uint16_t		rp_admin_mtu;
386 	uint8_t			rp_reserved4[2];
387 	uint16_t		rp_oper_mtu;
388 	uint8_t			rp_reserved5[2];
389 } __packed __aligned(4);
390 
391 struct mcx_reg_ptys {
392 	uint8_t			rp_reserved1;
393 	uint8_t			rp_local_port;
394 	uint8_t			rp_reserved2;
395 	uint8_t			rp_proto_mask;
396 #define MCX_REG_PTYS_PROTO_MASK_ETH		(1 << 2)
397 	uint8_t			rp_reserved3[8];
398 	uint32_t		rp_eth_proto_cap;
399 	uint8_t			rp_reserved4[8];
400 	uint32_t		rp_eth_proto_admin;
401 	uint8_t			rp_reserved5[8];
402 	uint32_t		rp_eth_proto_oper;
403 	uint8_t			rp_reserved6[24];
404 } __packed __aligned(4);
405 
406 struct mcx_reg_paos {
407 	uint8_t			rp_reserved1;
408 	uint8_t			rp_local_port;
409 	uint8_t			rp_admin_status;
410 #define MCX_REG_PAOS_ADMIN_STATUS_UP		1
411 #define MCX_REG_PAOS_ADMIN_STATUS_DOWN		2
412 #define MCX_REG_PAOS_ADMIN_STATUS_UP_ONCE	3
413 #define MCX_REG_PAOS_ADMIN_STATUS_DISABLED	4
414 	uint8_t			rp_oper_status;
415 #define MCX_REG_PAOS_OPER_STATUS_UP		1
416 #define MCX_REG_PAOS_OPER_STATUS_DOWN		2
417 #define MCX_REG_PAOS_OPER_STATUS_FAILED		4
418 	uint8_t			rp_admin_state_update;
419 #define MCX_REG_PAOS_ADMIN_STATE_UPDATE_EN	(1 << 7)
420 	uint8_t			rp_reserved2[11];
421 } __packed __aligned(4);
422 
423 struct mcx_reg_pfcc {
424 	uint8_t			rp_reserved1;
425 	uint8_t			rp_local_port;
426 	uint8_t			rp_reserved2[3];
427 	uint8_t			rp_prio_mask_tx;
428 	uint8_t			rp_reserved3;
429 	uint8_t			rp_prio_mask_rx;
430 	uint8_t			rp_pptx_aptx;
431 	uint8_t			rp_pfctx;
432 	uint8_t			rp_fctx_dis;
433 	uint8_t			rp_reserved4;
434 	uint8_t			rp_pprx_aprx;
435 	uint8_t			rp_pfcrx;
436 	uint8_t			rp_reserved5[2];
437 	uint16_t		rp_dev_stall_min;
438 	uint16_t		rp_dev_stall_crit;
439 	uint8_t			rp_reserved6[12];
440 } __packed __aligned(4);
441 
442 #define MCX_PMLP_MODULE_NUM_MASK	0xff
443 struct mcx_reg_pmlp {
444 	uint8_t			rp_rxtx;
445 	uint8_t			rp_local_port;
446 	uint8_t			rp_reserved0;
447 	uint8_t			rp_width;
448 	uint32_t		rp_lane0_mapping;
449 	uint32_t		rp_lane1_mapping;
450 	uint32_t		rp_lane2_mapping;
451 	uint32_t		rp_lane3_mapping;
452 	uint8_t			rp_reserved1[44];
453 } __packed __aligned(4);
454 
455 struct mcx_reg_ppcnt {
456 	uint8_t			ppcnt_swid;
457 	uint8_t			ppcnt_local_port;
458 	uint8_t			ppcnt_pnat;
459 	uint8_t			ppcnt_grp;
460 #define MCX_REG_PPCNT_GRP_IEEE8023		0x00
461 #define MCX_REG_PPCNT_GRP_RFC2863		0x01
462 #define MCX_REG_PPCNT_GRP_RFC2819		0x02
463 #define MCX_REG_PPCNT_GRP_RFC3635		0x03
464 #define MCX_REG_PPCNT_GRP_PER_PRIO		0x10
465 #define MCX_REG_PPCNT_GRP_PER_TC		0x11
466 #define MCX_REG_PPCNT_GRP_PER_RX_BUFFER		0x11
467 
468 	uint8_t			ppcnt_clr;
469 	uint8_t			ppcnt_reserved1[2];
470 	uint8_t			ppcnt_prio_tc;
471 #define MCX_REG_PPCNT_CLR			(1 << 7)
472 
473 	uint8_t			ppcnt_counter_set[248];
474 } __packed __aligned(8);
475 CTASSERT(sizeof(struct mcx_reg_ppcnt) == 256);
476 CTASSERT((offsetof(struct mcx_reg_ppcnt, ppcnt_counter_set) %
477     sizeof(uint64_t)) == 0);
478 
479 enum mcx_ppcnt_ieee8023 {
480 	frames_transmitted_ok,
481 	frames_received_ok,
482 	frame_check_sequence_errors,
483 	alignment_errors,
484 	octets_transmitted_ok,
485 	octets_received_ok,
486 	multicast_frames_xmitted_ok,
487 	broadcast_frames_xmitted_ok,
488 	multicast_frames_received_ok,
489 	broadcast_frames_received_ok,
490 	in_range_length_errors,
491 	out_of_range_length_field,
492 	frame_too_long_errors,
493 	symbol_error_during_carrier,
494 	mac_control_frames_transmitted,
495 	mac_control_frames_received,
496 	unsupported_opcodes_received,
497 	pause_mac_ctrl_frames_received,
498 	pause_mac_ctrl_frames_transmitted,
499 
500 	mcx_ppcnt_ieee8023_count
501 };
502 CTASSERT(mcx_ppcnt_ieee8023_count * sizeof(uint64_t) == 0x98);
503 
504 enum mcx_ppcnt_rfc2863 {
505 	in_octets,
506 	in_ucast_pkts,
507 	in_discards,
508 	in_errors,
509 	in_unknown_protos,
510 	out_octets,
511 	out_ucast_pkts,
512 	out_discards,
513 	out_errors,
514 	in_multicast_pkts,
515 	in_broadcast_pkts,
516 	out_multicast_pkts,
517 	out_broadcast_pkts,
518 
519 	mcx_ppcnt_rfc2863_count
520 };
521 CTASSERT(mcx_ppcnt_rfc2863_count * sizeof(uint64_t) == 0x68);
522 
523 enum mcx_ppcnt_rfc2819 {
524 	drop_events,
525 	octets,
526 	pkts,
527 	broadcast_pkts,
528 	multicast_pkts,
529 	crc_align_errors,
530 	undersize_pkts,
531 	oversize_pkts,
532 	fragments,
533 	jabbers,
534 	collisions,
535 	pkts64octets,
536 	pkts65to127octets,
537 	pkts128to255octets,
538 	pkts256to511octets,
539 	pkts512to1023octets,
540 	pkts1024to1518octets,
541 	pkts1519to2047octets,
542 	pkts2048to4095octets,
543 	pkts4096to8191octets,
544 	pkts8192to10239octets,
545 
546 	mcx_ppcnt_rfc2819_count
547 };
548 CTASSERT((mcx_ppcnt_rfc2819_count * sizeof(uint64_t)) == 0xa8);
549 
550 enum mcx_ppcnt_rfc3635 {
551 	dot3stats_alignment_errors,
552 	dot3stats_fcs_errors,
553 	dot3stats_single_collision_frames,
554 	dot3stats_multiple_collision_frames,
555 	dot3stats_sqe_test_errors,
556 	dot3stats_deferred_transmissions,
557 	dot3stats_late_collisions,
558 	dot3stats_excessive_collisions,
559 	dot3stats_internal_mac_transmit_errors,
560 	dot3stats_carrier_sense_errors,
561 	dot3stats_frame_too_longs,
562 	dot3stats_internal_mac_receive_errors,
563 	dot3stats_symbol_errors,
564 	dot3control_in_unknown_opcodes,
565 	dot3in_pause_frames,
566 	dot3out_pause_frames,
567 
568 	mcx_ppcnt_rfc3635_count
569 };
570 CTASSERT((mcx_ppcnt_rfc3635_count * sizeof(uint64_t)) == 0x80);
571 
572 struct mcx_reg_mtcap {
573 	uint8_t			_reserved1[3];
574 	uint8_t			mtcap_sensor_count;
575 	uint8_t			_reserved2[4];
576 
577 	uint64_t		mtcap_sensor_map;
578 };
579 
580 struct mcx_reg_mtmp {
581 	uint8_t			_reserved1[2];
582 	uint16_t		mtmp_sensor_index;
583 
584 	uint8_t			_reserved2[2];
585 	uint16_t		mtmp_temperature;
586 
587 	uint16_t		mtmp_mte_mtr;
588 #define MCX_REG_MTMP_MTE		(1 << 15)
589 #define MCX_REG_MTMP_MTR		(1 << 14)
590 	uint16_t		mtmp_max_temperature;
591 
592 	uint16_t		mtmp_tee;
593 #define MCX_REG_MTMP_TEE_NOPE		(0 << 14)
594 #define MCX_REG_MTMP_TEE_GENERATE	(1 << 14)
595 #define MCX_REG_MTMP_TEE_GENERATE_ONE	(2 << 14)
596 	uint16_t		mtmp_temperature_threshold_hi;
597 
598 	uint8_t			_reserved3[2];
599 	uint16_t		mtmp_temperature_threshold_lo;
600 
601 	uint8_t			_reserved4[4];
602 
603 	uint8_t			mtmp_sensor_name[8];
604 };
605 CTASSERT(sizeof(struct mcx_reg_mtmp) == 0x20);
606 CTASSERT(offsetof(struct mcx_reg_mtmp, mtmp_sensor_name) == 0x18);
607 
608 #define MCX_MCIA_EEPROM_BYTES	32
609 struct mcx_reg_mcia {
610 	uint8_t			rm_l;
611 	uint8_t			rm_module;
612 	uint8_t			rm_reserved0;
613 	uint8_t			rm_status;
614 	uint8_t			rm_i2c_addr;
615 	uint8_t			rm_page_num;
616 	uint16_t		rm_dev_addr;
617 	uint16_t		rm_reserved1;
618 	uint16_t		rm_size;
619 	uint32_t		rm_reserved2;
620 	uint8_t			rm_data[48];
621 } __packed __aligned(4);
622 
623 struct mcx_cmd_query_issi_in {
624 	uint16_t		cmd_opcode;
625 	uint8_t			cmd_reserved0[4];
626 	uint16_t		cmd_op_mod;
627 	uint8_t			cmd_reserved1[8];
628 } __packed __aligned(4);
629 
630 struct mcx_cmd_query_issi_il_out {
631 	uint8_t			cmd_status;
632 	uint8_t			cmd_reserved0[3];
633 	uint32_t		cmd_syndrome;
634 	uint8_t			cmd_reserved1[2];
635 	uint16_t		cmd_current_issi;
636 	uint8_t			cmd_reserved2[4];
637 } __packed __aligned(4);
638 
639 CTASSERT(sizeof(struct mcx_cmd_query_issi_il_out) == MCX_CMDQ_INLINE_DATASIZE);
640 
641 struct mcx_cmd_query_issi_mb_out {
642 	uint8_t			cmd_reserved2[16];
643 	uint8_t			cmd_supported_issi[80]; /* very big endian */
644 } __packed __aligned(4);
645 
646 CTASSERT(sizeof(struct mcx_cmd_query_issi_mb_out) <= MCX_CMDQ_MAILBOX_DATASIZE);
647 
648 struct mcx_cmd_set_issi_in {
649 	uint16_t		cmd_opcode;
650 	uint8_t			cmd_reserved0[4];
651 	uint16_t		cmd_op_mod;
652 	uint8_t			cmd_reserved1[2];
653 	uint16_t		cmd_current_issi;
654 	uint8_t			cmd_reserved2[4];
655 } __packed __aligned(4);
656 
657 CTASSERT(sizeof(struct mcx_cmd_set_issi_in) <= MCX_CMDQ_INLINE_DATASIZE);
658 
659 struct mcx_cmd_set_issi_out {
660 	uint8_t			cmd_status;
661 	uint8_t			cmd_reserved0[3];
662 	uint32_t		cmd_syndrome;
663 	uint8_t			cmd_reserved1[8];
664 } __packed __aligned(4);
665 
666 CTASSERT(sizeof(struct mcx_cmd_set_issi_out) <= MCX_CMDQ_INLINE_DATASIZE);
667 
668 struct mcx_cmd_query_pages_in {
669 	uint16_t		cmd_opcode;
670 	uint8_t			cmd_reserved0[4];
671 	uint16_t		cmd_op_mod;
672 #define MCX_CMD_QUERY_PAGES_BOOT	0x01
673 #define MCX_CMD_QUERY_PAGES_INIT	0x02
674 #define MCX_CMD_QUERY_PAGES_REGULAR	0x03
675 	uint8_t			cmd_reserved1[8];
676 } __packed __aligned(4);
677 
678 struct mcx_cmd_query_pages_out {
679 	uint8_t			cmd_status;
680 	uint8_t			cmd_reserved0[3];
681 	uint32_t		cmd_syndrome;
682 	uint8_t			cmd_reserved1[2];
683 	uint16_t		cmd_func_id;
684 	int32_t			cmd_num_pages;
685 } __packed __aligned(4);
686 
687 struct mcx_cmd_manage_pages_in {
688 	uint16_t		cmd_opcode;
689 	uint8_t			cmd_reserved0[4];
690 	uint16_t		cmd_op_mod;
691 #define MCX_CMD_MANAGE_PAGES_ALLOC_FAIL \
692 					0x00
693 #define MCX_CMD_MANAGE_PAGES_ALLOC_SUCCESS \
694 					0x01
695 #define MCX_CMD_MANAGE_PAGES_HCA_RETURN_PAGES \
696 					0x02
697 	uint8_t			cmd_reserved1[2];
698 	uint16_t		cmd_func_id;
699 	uint32_t		cmd_input_num_entries;
700 } __packed __aligned(4);
701 
702 CTASSERT(sizeof(struct mcx_cmd_manage_pages_in) == MCX_CMDQ_INLINE_DATASIZE);
703 
704 struct mcx_cmd_manage_pages_out {
705 	uint8_t			cmd_status;
706 	uint8_t			cmd_reserved0[3];
707 	uint32_t		cmd_syndrome;
708 	uint32_t		cmd_output_num_entries;
709 	uint8_t			cmd_reserved1[4];
710 } __packed __aligned(4);
711 
712 CTASSERT(sizeof(struct mcx_cmd_manage_pages_out) == MCX_CMDQ_INLINE_DATASIZE);
713 
714 struct mcx_cmd_query_hca_cap_in {
715 	uint16_t		cmd_opcode;
716 	uint8_t			cmd_reserved0[4];
717 	uint16_t		cmd_op_mod;
718 #define MCX_CMD_QUERY_HCA_CAP_MAX	(0x0 << 0)
719 #define MCX_CMD_QUERY_HCA_CAP_CURRENT	(0x1 << 0)
720 #define MCX_CMD_QUERY_HCA_CAP_DEVICE	(0x0 << 1)
721 #define MCX_CMD_QUERY_HCA_CAP_OFFLOAD	(0x1 << 1)
722 #define MCX_CMD_QUERY_HCA_CAP_FLOW	(0x7 << 1)
723 	uint8_t			cmd_reserved1[8];
724 } __packed __aligned(4);
725 
726 struct mcx_cmd_query_hca_cap_out {
727 	uint8_t			cmd_status;
728 	uint8_t			cmd_reserved0[3];
729 	uint32_t		cmd_syndrome;
730 	uint8_t			cmd_reserved1[8];
731 } __packed __aligned(4);
732 
733 #define MCX_HCA_CAP_LEN			0x1000
734 #define MCX_HCA_CAP_NMAILBOXES		\
735 	(MCX_HCA_CAP_LEN / MCX_CMDQ_MAILBOX_DATASIZE)
736 
737 #if __GNUC_PREREQ__(4, 3)
738 #define __counter__		__COUNTER__
739 #else
740 #define __counter__		__LINE__
741 #endif
742 
743 #define __token(_tok, _num)	_tok##_num
744 #define _token(_tok, _num)	__token(_tok, _num)
745 #define __reserved__		_token(__reserved, __counter__)
746 
747 struct mcx_cap_device {
748 	uint8_t			reserved0[16];
749 
750 	uint8_t			log_max_srq_sz;
751 	uint8_t			log_max_qp_sz;
752 	uint8_t			__reserved__[1];
753 	uint8_t			log_max_qp; /* 5 bits */
754 #define MCX_CAP_DEVICE_LOG_MAX_QP	0x1f
755 
756 	uint8_t			__reserved__[1];
757 	uint8_t			log_max_srq; /* 5 bits */
758 #define MCX_CAP_DEVICE_LOG_MAX_SRQ	0x1f
759 	uint8_t			__reserved__[2];
760 
761 	uint8_t			__reserved__[1];
762 	uint8_t			log_max_cq_sz;
763 	uint8_t			__reserved__[1];
764 	uint8_t			log_max_cq; /* 5 bits */
765 #define MCX_CAP_DEVICE_LOG_MAX_CQ	0x1f
766 
767 	uint8_t			log_max_eq_sz;
768 	uint8_t			log_max_mkey; /* 6 bits */
769 #define MCX_CAP_DEVICE_LOG_MAX_MKEY	0x3f
770 	uint8_t			__reserved__[1];
771 	uint8_t			log_max_eq; /* 4 bits */
772 #define MCX_CAP_DEVICE_LOG_MAX_EQ	0x0f
773 
774 	uint8_t			max_indirection;
775 	uint8_t			log_max_mrw_sz; /* 7 bits */
776 #define MCX_CAP_DEVICE_LOG_MAX_MRW_SZ	0x7f
777 	uint8_t			teardown_log_max_msf_list_size;
778 #define MCX_CAP_DEVICE_FORCE_TEARDOWN	0x80
779 #define MCX_CAP_DEVICE_LOG_MAX_MSF_LIST_SIZE \
780 					0x3f
781 	uint8_t			log_max_klm_list_size; /* 6 bits */
782 #define MCX_CAP_DEVICE_LOG_MAX_KLM_LIST_SIZE \
783 					0x3f
784 
785 	uint8_t			__reserved__[1];
786 	uint8_t			log_max_ra_req_dc; /* 6 bits */
787 #define MCX_CAP_DEVICE_LOG_MAX_REQ_DC	0x3f
788 	uint8_t			__reserved__[1];
789 	uint8_t			log_max_ra_res_dc; /* 6 bits */
790 #define MCX_CAP_DEVICE_LOG_MAX_RA_RES_DC \
791 					0x3f
792 
793 	uint8_t			__reserved__[1];
794 	uint8_t			log_max_ra_req_qp; /* 6 bits */
795 #define MCX_CAP_DEVICE_LOG_MAX_RA_REQ_QP \
796 					0x3f
797 	uint8_t			__reserved__[1];
798 	uint8_t			log_max_ra_res_qp; /* 6 bits */
799 #define MCX_CAP_DEVICE_LOG_MAX_RA_RES_QP \
800 					0x3f
801 
802 	uint8_t			flags1;
803 #define MCX_CAP_DEVICE_END_PAD		0x80
804 #define MCX_CAP_DEVICE_CC_QUERY_ALLOWED	0x40
805 #define MCX_CAP_DEVICE_CC_MODIFY_ALLOWED \
806 					0x20
807 #define MCX_CAP_DEVICE_START_PAD	0x10
808 #define MCX_CAP_DEVICE_128BYTE_CACHELINE \
809 					0x08
810 	uint8_t			__reserved__[1];
811 	uint16_t		gid_table_size;
812 
813 	uint16_t		flags2;
814 #define MCX_CAP_DEVICE_OUT_OF_SEQ_CNT	0x8000
815 #define MCX_CAP_DEVICE_VPORT_COUNTERS	0x4000
816 #define MCX_CAP_DEVICE_RETRANSMISSION_Q_COUNTERS \
817 					0x2000
818 #define MCX_CAP_DEVICE_DEBUG		0x1000
819 #define MCX_CAP_DEVICE_MODIFY_RQ_COUNTERS_SET_ID \
820 					0x8000
821 #define MCX_CAP_DEVICE_RQ_DELAY_DROP	0x4000
822 #define MCX_CAP_DEVICe_MAX_QP_CNT_MASK	0x03ff
823 	uint16_t		pkey_table_size;
824 
825 	uint8_t			flags3;
826 #define MCX_CAP_DEVICE_VPORT_GROUP_MANAGER \
827 					0x80
828 #define MCX_CAP_DEVICE_VHCA_GROUP_MANAGER \
829 					0x40
830 #define MCX_CAP_DEVICE_IB_VIRTUAL	0x20
831 #define MCX_CAP_DEVICE_ETH_VIRTUAL	0x10
832 #define MCX_CAP_DEVICE_ETS		0x04
833 #define MCX_CAP_DEVICE_NIC_FLOW_TABLE	0x02
834 #define MCX_CAP_DEVICE_ESWITCH_FLOW_TABLE \
835 					0x01
836 	uint8_t			local_ca_ack_delay; /* 5 bits */
837 #define MCX_CAP_DEVICE_LOCAL_CA_ACK_DELAY \
838 					0x1f
839 	uint8_t			port_type;
840 #define MCX_CAP_DEVICE_PORT_MODULE_EVENT \
841 					0x80
842 #define MCX_CAP_DEVICE_PORT_TYPE	0x03
843 #define MCX_CAP_DEVICE_PORT_TYPE_ETH	0x01
844 	uint8_t			num_ports;
845 
846 	uint8_t			snapshot_log_max_msg;
847 #define MCX_CAP_DEVICE_SNAPSHOT		0x80
848 #define MCX_CAP_DEVICE_LOG_MAX_MSG	0x1f
849 	uint8_t			max_tc; /* 4 bits */
850 #define MCX_CAP_DEVICE_MAX_TC		0x0f
851 	uint8_t			flags4;
852 #define MCX_CAP_DEVICE_TEMP_WARN_EVENT	0x80
853 #define MCX_CAP_DEVICE_DCBX		0x40
854 #define MCX_CAP_DEVICE_ROL_S		0x02
855 #define MCX_CAP_DEVICE_ROL_G		0x01
856 	uint8_t			wol;
857 #define MCX_CAP_DEVICE_WOL_S		0x40
858 #define MCX_CAP_DEVICE_WOL_G		0x20
859 #define MCX_CAP_DEVICE_WOL_A		0x10
860 #define MCX_CAP_DEVICE_WOL_B		0x08
861 #define MCX_CAP_DEVICE_WOL_M		0x04
862 #define MCX_CAP_DEVICE_WOL_U		0x02
863 #define MCX_CAP_DEVICE_WOL_P		0x01
864 
865 	uint16_t		stat_rate_support;
866 	uint8_t			__reserved__[1];
867 	uint8_t			cqe_version; /* 4 bits */
868 #define MCX_CAP_DEVICE_CQE_VERSION	0x0f
869 
870 	uint32_t		flags5;
871 #define MCX_CAP_DEVICE_COMPACT_ADDRESS_VECTOR \
872 					0x80000000
873 #define MCX_CAP_DEVICE_STRIDING_RQ	0x40000000
874 #define MCX_CAP_DEVICE_IPOIP_ENHANCED_OFFLOADS \
875 					0x10000000
876 #define MCX_CAP_DEVICE_IPOIP_IPOIP_OFFLOADS \
877 					0x08000000
878 #define MCX_CAP_DEVICE_DC_CONNECT_CP	0x00040000
879 #define MCX_CAP_DEVICE_DC_CNAK_DRACE	0x00020000
880 #define MCX_CAP_DEVICE_DRAIN_SIGERR	0x00010000
881 #define MCX_CAP_DEVICE_DRAIN_SIGERR	0x00010000
882 #define MCX_CAP_DEVICE_CMDIF_CHECKSUM	0x0000c000
883 #define MCX_CAP_DEVICE_SIGERR_QCE	0x00002000
884 #define MCX_CAP_DEVICE_WQ_SIGNATURE	0x00000800
885 #define MCX_CAP_DEVICE_SCTR_DATA_CQE	0x00000400
886 #define MCX_CAP_DEVICE_SHO		0x00000100
887 #define MCX_CAP_DEVICE_TPH		0x00000080
888 #define MCX_CAP_DEVICE_RF		0x00000040
889 #define MCX_CAP_DEVICE_DCT		0x00000020
890 #define MCX_CAP_DEVICE_QOS		0x00000010
891 #define MCX_CAP_DEVICe_ETH_NET_OFFLOADS	0x00000008
892 #define MCX_CAP_DEVICE_ROCE		0x00000004
893 #define MCX_CAP_DEVICE_ATOMIC		0x00000002
894 
895 	uint32_t		flags6;
896 #define MCX_CAP_DEVICE_CQ_OI		0x80000000
897 #define MCX_CAP_DEVICE_CQ_RESIZE	0x40000000
898 #define MCX_CAP_DEVICE_CQ_MODERATION	0x20000000
899 #define MCX_CAP_DEVICE_CQ_PERIOD_MODE_MODIFY \
900 					0x10000000
901 #define MCX_CAP_DEVICE_CQ_INVALIDATE	0x08000000
902 #define MCX_CAP_DEVICE_RESERVED_AT_255	0x04000000
903 #define MCX_CAP_DEVICE_CQ_EQ_REMAP	0x02000000
904 #define MCX_CAP_DEVICE_PG		0x01000000
905 #define MCX_CAP_DEVICE_BLOCK_LB_MC	0x00800000
906 #define MCX_CAP_DEVICE_EXPONENTIAL_BACKOFF \
907 					0x00400000
908 #define MCX_CAP_DEVICE_SCQE_BREAK_MODERATION \
909 					0x00200000
910 #define MCX_CAP_DEVICE_CQ_PERIOD_START_FROM_CQE \
911 					0x00100000
912 #define MCX_CAP_DEVICE_CD		0x00080000
913 #define MCX_CAP_DEVICE_ATM		0x00040000
914 #define MCX_CAP_DEVICE_APM		0x00020000
915 #define MCX_CAP_DEVICE_IMAICL		0x00010000
916 #define MCX_CAP_DEVICE_QKV		0x00000200
917 #define MCX_CAP_DEVICE_PKV		0x00000100
918 #define MCX_CAP_DEVICE_SET_DETH_SQPN	0x00000080
919 #define MCX_CAP_DEVICE_XRC		0x00000008
920 #define MCX_CAP_DEVICE_UD		0x00000004
921 #define MCX_CAP_DEVICE_UC		0x00000002
922 #define MCX_CAP_DEVICE_RC		0x00000001
923 
924 	uint8_t			uar_flags;
925 #define MCX_CAP_DEVICE_UAR_4K		0x80
926 	uint8_t			uar_sz;	/* 6 bits */
927 #define MCX_CAP_DEVICE_UAR_SZ		0x3f
928 	uint8_t			__reserved__[1];
929 	uint8_t			log_pg_sz;
930 
931 	uint8_t			flags7;
932 #define MCX_CAP_DEVICE_BF		0x80
933 #define MCX_CAP_DEVICE_DRIVER_VERSION	0x40
934 #define MCX_CAP_DEVICE_PAD_TX_ETH_PACKET \
935 					0x20
936 	uint8_t			log_bf_reg_size; /* 5 bits */
937 #define MCX_CAP_DEVICE_LOG_BF_REG_SIZE	0x1f
938 	uint8_t			__reserved__[2];
939 
940 	uint16_t		num_of_diagnostic_counters;
941 	uint16_t		max_wqe_sz_sq;
942 
943 	uint8_t			__reserved__[2];
944 	uint16_t		max_wqe_sz_rq;
945 
946 	uint8_t			__reserved__[2];
947 	uint16_t		max_wqe_sz_sq_dc;
948 
949 	uint32_t		max_qp_mcg; /* 25 bits */
950 #define MCX_CAP_DEVICE_MAX_QP_MCG	0x1ffffff
951 
952 	uint8_t			__reserved__[3];
953 	uint8_t			log_max_mcq;
954 
955 	uint8_t			log_max_transport_domain; /* 5 bits */
956 #define MCX_CAP_DEVICE_LOG_MAX_TRANSORT_DOMAIN \
957 					0x1f
958 	uint8_t			log_max_pd; /* 5 bits */
959 #define MCX_CAP_DEVICE_LOG_MAX_PD	0x1f
960 	uint8_t			__reserved__[1];
961 	uint8_t			log_max_xrcd; /* 5 bits */
962 #define MCX_CAP_DEVICE_LOG_MAX_XRCD	0x1f
963 
964 	uint8_t			__reserved__[2];
965 	uint16_t		max_flow_counter;
966 
967 	uint8_t			log_max_rq; /* 5 bits */
968 #define MCX_CAP_DEVICE_LOG_MAX_RQ	0x1f
969 	uint8_t			log_max_sq; /* 5 bits */
970 #define MCX_CAP_DEVICE_LOG_MAX_SQ	0x1f
971 	uint8_t			log_max_tir; /* 5 bits */
972 #define MCX_CAP_DEVICE_LOG_MAX_TIR	0x1f
973 	uint8_t			log_max_tis; /* 5 bits */
974 #define MCX_CAP_DEVICE_LOG_MAX_TIS	0x1f
975 
976 	uint8_t 		flags8;
977 #define MCX_CAP_DEVICE_BASIC_CYCLIC_RCV_WQE \
978 					0x80
979 #define MCX_CAP_DEVICE_LOG_MAX_RMP	0x1f
980 	uint8_t			log_max_rqt; /* 5 bits */
981 #define MCX_CAP_DEVICE_LOG_MAX_RQT	0x1f
982 	uint8_t			log_max_rqt_size; /* 5 bits */
983 #define MCX_CAP_DEVICE_LOG_MAX_RQT_SIZE	0x1f
984 	uint8_t			log_max_tis_per_sq; /* 5 bits */
985 #define MCX_CAP_DEVICE_LOG_MAX_TIS_PER_SQ \
986 					0x1f
987 
988 	uint8_t			flags9;
989 #define MXC_CAP_DEVICE_EXT_STRIDE_NUM_RANGES \
990 					0x80
991 #define MXC_CAP_DEVICE_LOG_MAX_STRIDE_SZ_RQ \
992 					0x1f
993 	uint8_t			log_min_stride_sz_rq; /* 5 bits */
994 #define MXC_CAP_DEVICE_LOG_MIN_STRIDE_SZ_RQ \
995 					0x1f
996 	uint8_t			log_max_stride_sz_sq; /* 5 bits */
997 #define MXC_CAP_DEVICE_LOG_MAX_STRIDE_SZ_SQ \
998 					0x1f
999 	uint8_t			log_min_stride_sz_sq; /* 5 bits */
1000 #define MXC_CAP_DEVICE_LOG_MIN_STRIDE_SZ_SQ \
1001 					0x1f
1002 
1003 	uint8_t			log_max_hairpin_queues;
1004 #define MXC_CAP_DEVICE_HAIRPIN		0x80
1005 #define MXC_CAP_DEVICE_LOG_MAX_HAIRPIN_QUEUES \
1006 					0x1f
1007 	uint8_t			log_min_hairpin_queues;
1008 #define MXC_CAP_DEVICE_LOG_MIN_HAIRPIN_QUEUES \
1009 					0x1f
1010 	uint8_t			log_max_hairpin_num_packets;
1011 #define MXC_CAP_DEVICE_LOG_MAX_HAIRPIN_NUM_PACKETS \
1012 					0x1f
1013 	uint8_t			log_max_mq_sz;
1014 #define MXC_CAP_DEVICE_LOG_MAX_WQ_SZ \
1015 					0x1f
1016 
1017 	uint8_t			log_min_hairpin_wq_data_sz;
1018 #define MXC_CAP_DEVICE_NIC_VPORT_CHANGE_EVENT \
1019 					0x80
1020 #define MXC_CAP_DEVICE_DISABLE_LOCAL_LB_UC \
1021 					0x40
1022 #define MXC_CAP_DEVICE_DISABLE_LOCAL_LB_MC \
1023 					0x20
1024 #define MCX_CAP_DEVICE_LOG_MIN_HAIRPIN_WQ_DATA_SZ \
1025 					0x1f
1026 	uint8_t			log_max_vlan_list;
1027 #define MXC_CAP_DEVICE_SYSTEM_IMAGE_GUID_MODIFIABLE \
1028 					0x80
1029 #define MXC_CAP_DEVICE_LOG_MAX_VLAN_LIST \
1030 					0x1f
1031 	uint8_t			log_max_current_mc_list;
1032 #define MXC_CAP_DEVICE_LOG_MAX_CURRENT_MC_LIST \
1033 					0x1f
1034 	uint8_t			log_max_current_uc_list;
1035 #define MXC_CAP_DEVICE_LOG_MAX_CURRENT_UC_LIST \
1036 					0x1f
1037 
1038 	uint8_t			__reserved__[4];
1039 
1040 	uint32_t		create_qp_start_hint; /* 24 bits */
1041 
1042 	uint8_t			log_max_uctx; /* 5 bits */
1043 #define MXC_CAP_DEVICE_LOG_MAX_UCTX	0x1f
1044 	uint8_t			log_max_umem; /* 5 bits */
1045 #define MXC_CAP_DEVICE_LOG_MAX_UMEM	0x1f
1046 	uint16_t		max_num_eqs;
1047 
1048 	uint8_t			log_max_l2_table; /* 5 bits */
1049 #define MXC_CAP_DEVICE_LOG_MAX_L2_TABLE	0x1f
1050 	uint8_t			__reserved__[1];
1051 	uint16_t		log_uar_page_sz;
1052 
1053 	uint8_t			__reserved__[8];
1054 
1055 	uint32_t		device_frequency_mhz;
1056 	uint32_t		device_frequency_khz;
1057 } __packed __aligned(8);
1058 
1059 CTASSERT(offsetof(struct mcx_cap_device, max_indirection) == 0x20);
1060 CTASSERT(offsetof(struct mcx_cap_device, flags1) == 0x2c);
1061 CTASSERT(offsetof(struct mcx_cap_device, flags2) == 0x30);
1062 CTASSERT(offsetof(struct mcx_cap_device, snapshot_log_max_msg) == 0x38);
1063 CTASSERT(offsetof(struct mcx_cap_device, flags5) == 0x40);
1064 CTASSERT(offsetof(struct mcx_cap_device, flags7) == 0x4c);
1065 CTASSERT(offsetof(struct mcx_cap_device, device_frequency_mhz) == 0x98);
1066 CTASSERT(offsetof(struct mcx_cap_device, device_frequency_khz) == 0x9c);
1067 CTASSERT(sizeof(struct mcx_cap_device) <= MCX_CMDQ_MAILBOX_DATASIZE);
1068 
1069 struct mcx_cmd_set_driver_version_in {
1070 	uint16_t		cmd_opcode;
1071 	uint8_t			cmd_reserved0[4];
1072 	uint16_t		cmd_op_mod;
1073 	uint8_t			cmd_reserved1[8];
1074 } __packed __aligned(4);
1075 
1076 struct mcx_cmd_set_driver_version_out {
1077 	uint8_t			cmd_status;
1078 	uint8_t			cmd_reserved0[3];
1079 	uint32_t		cmd_syndrome;
1080 	uint8_t			cmd_reserved1[8];
1081 } __packed __aligned(4);
1082 
1083 struct mcx_cmd_set_driver_version {
1084 	uint8_t			cmd_driver_version[64];
1085 } __packed __aligned(8);
1086 
1087 struct mcx_cmd_modify_nic_vport_context_in {
1088 	uint16_t		cmd_opcode;
1089 	uint8_t			cmd_reserved0[4];
1090 	uint16_t		cmd_op_mod;
1091 	uint8_t			cmd_reserved1[4];
1092 	uint32_t		cmd_field_select;
1093 #define MCX_CMD_MODIFY_NIC_VPORT_CONTEXT_FIELD_ADDR	0x04
1094 #define MCX_CMD_MODIFY_NIC_VPORT_CONTEXT_FIELD_PROMISC	0x10
1095 #define MCX_CMD_MODIFY_NIC_VPORT_CONTEXT_FIELD_MTU	0x40
1096 } __packed __aligned(4);
1097 
1098 struct mcx_cmd_modify_nic_vport_context_out {
1099 	uint8_t			cmd_status;
1100 	uint8_t			cmd_reserved0[3];
1101 	uint32_t		cmd_syndrome;
1102 	uint8_t			cmd_reserved1[8];
1103 } __packed __aligned(4);
1104 
1105 struct mcx_cmd_query_nic_vport_context_in {
1106 	uint16_t		cmd_opcode;
1107 	uint8_t			cmd_reserved0[4];
1108 	uint16_t		cmd_op_mod;
1109 	uint8_t			cmd_reserved1[4];
1110 	uint8_t			cmd_allowed_list_type;
1111 	uint8_t			cmd_reserved2[3];
1112 } __packed __aligned(4);
1113 
1114 struct mcx_cmd_query_nic_vport_context_out {
1115 	uint8_t			cmd_status;
1116 	uint8_t			cmd_reserved0[3];
1117 	uint32_t		cmd_syndrome;
1118 	uint8_t			cmd_reserved1[8];
1119 } __packed __aligned(4);
1120 
1121 struct mcx_nic_vport_ctx {
1122 	uint32_t		vp_min_wqe_inline_mode;
1123 	uint8_t			vp_reserved0[32];
1124 	uint32_t		vp_mtu;
1125 	uint8_t			vp_reserved1[200];
1126 	uint16_t		vp_flags;
1127 #define MCX_NIC_VPORT_CTX_LIST_UC_MAC			(0)
1128 #define MCX_NIC_VPORT_CTX_LIST_MC_MAC			(1 << 24)
1129 #define MCX_NIC_VPORT_CTX_LIST_VLAN			(2 << 24)
1130 #define MCX_NIC_VPORT_CTX_PROMISC_ALL			(1 << 13)
1131 #define MCX_NIC_VPORT_CTX_PROMISC_MCAST			(1 << 14)
1132 #define MCX_NIC_VPORT_CTX_PROMISC_UCAST			(1 << 15)
1133 	uint16_t		vp_allowed_list_size;
1134 	uint64_t		vp_perm_addr;
1135 	uint8_t			vp_reserved2[4];
1136 	/* allowed list follows */
1137 } __packed __aligned(4);
1138 
1139 struct mcx_counter {
1140 	uint64_t		packets;
1141 	uint64_t		octets;
1142 } __packed __aligned(4);
1143 
1144 struct mcx_nic_vport_counters {
1145 	struct mcx_counter	rx_err;
1146 	struct mcx_counter	tx_err;
1147 	uint8_t			reserved0[64]; /* 0x30 */
1148 	struct mcx_counter	rx_bcast;
1149 	struct mcx_counter	tx_bcast;
1150 	struct mcx_counter	rx_ucast;
1151 	struct mcx_counter	tx_ucast;
1152 	struct mcx_counter	rx_mcast;
1153 	struct mcx_counter	tx_mcast;
1154 	uint8_t			reserved1[0x210 - 0xd0];
1155 } __packed __aligned(4);
1156 
1157 struct mcx_cmd_query_vport_counters_in {
1158 	uint16_t		cmd_opcode;
1159 	uint8_t			cmd_reserved0[4];
1160 	uint16_t		cmd_op_mod;
1161 	uint8_t			cmd_reserved1[8];
1162 } __packed __aligned(4);
1163 
1164 struct mcx_cmd_query_vport_counters_mb_in {
1165 	uint8_t			cmd_reserved0[8];
1166 	uint8_t			cmd_clear;
1167 	uint8_t			cmd_reserved1[7];
1168 } __packed __aligned(4);
1169 
1170 struct mcx_cmd_query_vport_counters_out {
1171 	uint8_t			cmd_status;
1172 	uint8_t			cmd_reserved0[3];
1173 	uint32_t		cmd_syndrome;
1174 	uint8_t			cmd_reserved1[8];
1175 } __packed __aligned(4);
1176 
1177 struct mcx_cmd_query_flow_counter_in {
1178 	uint16_t		cmd_opcode;
1179 	uint8_t			cmd_reserved0[4];
1180 	uint16_t		cmd_op_mod;
1181 	uint8_t			cmd_reserved1[8];
1182 } __packed __aligned(4);
1183 
1184 struct mcx_cmd_query_flow_counter_mb_in {
1185 	uint8_t			cmd_reserved0[8];
1186 	uint8_t			cmd_clear;
1187 	uint8_t			cmd_reserved1[5];
1188 	uint16_t		cmd_flow_counter_id;
1189 } __packed __aligned(4);
1190 
1191 struct mcx_cmd_query_flow_counter_out {
1192 	uint8_t			cmd_status;
1193 	uint8_t			cmd_reserved0[3];
1194 	uint32_t		cmd_syndrome;
1195 	uint8_t			cmd_reserved1[8];
1196 } __packed __aligned(4);
1197 
1198 struct mcx_cmd_alloc_uar_in {
1199 	uint16_t		cmd_opcode;
1200 	uint8_t			cmd_reserved0[4];
1201 	uint16_t		cmd_op_mod;
1202 	uint8_t			cmd_reserved1[8];
1203 } __packed __aligned(4);
1204 
1205 struct mcx_cmd_alloc_uar_out {
1206 	uint8_t			cmd_status;
1207 	uint8_t			cmd_reserved0[3];
1208 	uint32_t		cmd_syndrome;
1209 	uint32_t		cmd_uar;
1210 	uint8_t			cmd_reserved1[4];
1211 } __packed __aligned(4);
1212 
1213 struct mcx_cmd_query_special_ctx_in {
1214 	uint16_t		cmd_opcode;
1215 	uint8_t			cmd_reserved0[4];
1216 	uint16_t		cmd_op_mod;
1217 	uint8_t			cmd_reserved1[8];
1218 } __packed __aligned(4);
1219 
1220 struct mcx_cmd_query_special_ctx_out {
1221 	uint8_t			cmd_status;
1222 	uint8_t			cmd_reserved0[3];
1223 	uint32_t		cmd_syndrome;
1224 	uint8_t			cmd_reserved1[4];
1225 	uint32_t		cmd_resd_lkey;
1226 } __packed __aligned(4);
1227 
1228 struct mcx_eq_ctx {
1229 	uint32_t		eq_status;
1230 #define MCX_EQ_CTX_STATE_SHIFT		8
1231 #define MCX_EQ_CTX_STATE_MASK		(0xf << MCX_EQ_CTX_STATE_SHIFT)
1232 #define MCX_EQ_CTX_STATE_ARMED		0x9
1233 #define MCX_EQ_CTX_STATE_FIRED		0xa
1234 #define MCX_EQ_CTX_OI_SHIFT		17
1235 #define MCX_EQ_CTX_OI			(1 << MCX_EQ_CTX_OI_SHIFT)
1236 #define MCX_EQ_CTX_EC_SHIFT		18
1237 #define MCX_EQ_CTX_EC			(1 << MCX_EQ_CTX_EC_SHIFT)
1238 #define MCX_EQ_CTX_STATUS_SHIFT		28
1239 #define MCX_EQ_CTX_STATUS_MASK		(0xf << MCX_EQ_CTX_STATUS_SHIFT)
1240 #define MCX_EQ_CTX_STATUS_OK		0x0
1241 #define MCX_EQ_CTX_STATUS_EQ_WRITE_FAILURE 0xa
1242 	uint32_t		eq_reserved1;
1243 	uint32_t		eq_page_offset;
1244 #define MCX_EQ_CTX_PAGE_OFFSET_SHIFT	5
1245 	uint32_t		eq_uar_size;
1246 #define MCX_EQ_CTX_UAR_PAGE_MASK	0xffffff
1247 #define MCX_EQ_CTX_LOG_EQ_SIZE_SHIFT	24
1248 	uint32_t		eq_reserved2;
1249 	uint8_t			eq_reserved3[3];
1250 	uint8_t			eq_intr;
1251 	uint32_t		eq_log_page_size;
1252 #define MCX_EQ_CTX_LOG_PAGE_SIZE_SHIFT	24
1253 	uint32_t		eq_reserved4[3];
1254 	uint32_t		eq_consumer_counter;
1255 	uint32_t		eq_producer_counter;
1256 #define MCX_EQ_CTX_COUNTER_MASK		0xffffff
1257 	uint32_t		eq_reserved5[4];
1258 } __packed __aligned(4);
1259 
1260 CTASSERT(sizeof(struct mcx_eq_ctx) == 64);
1261 
1262 struct mcx_cmd_create_eq_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_eq_mb_in {
1270 	struct mcx_eq_ctx	cmd_eq_ctx;
1271 	uint8_t			cmd_reserved0[8];
1272 	uint64_t		cmd_event_bitmask;
1273 #define MCX_EVENT_TYPE_COMPLETION	0x00
1274 #define MCX_EVENT_TYPE_CQ_ERROR		0x04
1275 #define MCX_EVENT_TYPE_INTERNAL_ERROR	0x08
1276 #define MCX_EVENT_TYPE_PORT_CHANGE	0x09
1277 #define MCX_EVENT_TYPE_CMD_COMPLETION	0x0a
1278 #define MCX_EVENT_TYPE_PAGE_REQUEST	0x0b
1279 #define MCX_EVENT_TYPE_LAST_WQE		0x13
1280 	uint8_t			cmd_reserved1[176];
1281 } __packed __aligned(4);
1282 
1283 struct mcx_cmd_create_eq_out {
1284 	uint8_t			cmd_status;
1285 	uint8_t			cmd_reserved0[3];
1286 	uint32_t		cmd_syndrome;
1287 	uint32_t		cmd_eqn;
1288 	uint8_t			cmd_reserved1[4];
1289 } __packed __aligned(4);
1290 
1291 struct mcx_cmd_query_eq_in {
1292 	uint16_t		cmd_opcode;
1293 	uint8_t			cmd_reserved0[4];
1294 	uint16_t		cmd_op_mod;
1295 	uint32_t		cmd_eqn;
1296 	uint8_t			cmd_reserved1[4];
1297 } __packed __aligned(4);
1298 
1299 struct mcx_cmd_query_eq_out {
1300 	uint8_t			cmd_status;
1301 	uint8_t			cmd_reserved0[3];
1302 	uint32_t		cmd_syndrome;
1303 	uint8_t			cmd_reserved1[8];
1304 } __packed __aligned(4);
1305 
1306 struct mcx_eq_entry {
1307 	uint8_t			eq_reserved1;
1308 	uint8_t			eq_event_type;
1309 	uint8_t			eq_reserved2;
1310 	uint8_t			eq_event_sub_type;
1311 
1312 	uint8_t			eq_reserved3[28];
1313 	uint32_t		eq_event_data[7];
1314 	uint8_t			eq_reserved4[2];
1315 	uint8_t			eq_signature;
1316 	uint8_t			eq_owner;
1317 #define MCX_EQ_ENTRY_OWNER_INIT			1
1318 } __packed __aligned(4);
1319 
1320 CTASSERT(sizeof(struct mcx_eq_entry) == 64);
1321 
1322 struct mcx_cmd_alloc_pd_in {
1323 	uint16_t		cmd_opcode;
1324 	uint8_t			cmd_reserved0[4];
1325 	uint16_t		cmd_op_mod;
1326 	uint8_t			cmd_reserved1[8];
1327 } __packed __aligned(4);
1328 
1329 struct mcx_cmd_alloc_pd_out {
1330 	uint8_t			cmd_status;
1331 	uint8_t			cmd_reserved0[3];
1332 	uint32_t		cmd_syndrome;
1333 	uint32_t		cmd_pd;
1334 	uint8_t			cmd_reserved1[4];
1335 } __packed __aligned(4);
1336 
1337 struct mcx_cmd_alloc_td_in {
1338 	uint16_t		cmd_opcode;
1339 	uint8_t			cmd_reserved0[4];
1340 	uint16_t		cmd_op_mod;
1341 	uint8_t			cmd_reserved1[8];
1342 } __packed __aligned(4);
1343 
1344 struct mcx_cmd_alloc_td_out {
1345 	uint8_t			cmd_status;
1346 	uint8_t			cmd_reserved0[3];
1347 	uint32_t		cmd_syndrome;
1348 	uint32_t		cmd_tdomain;
1349 	uint8_t			cmd_reserved1[4];
1350 } __packed __aligned(4);
1351 
1352 struct mcx_cmd_create_tir_in {
1353 	uint16_t		cmd_opcode;
1354 	uint8_t			cmd_reserved0[4];
1355 	uint16_t		cmd_op_mod;
1356 	uint8_t			cmd_reserved1[8];
1357 } __packed __aligned(4);
1358 
1359 struct mcx_cmd_create_tir_mb_in {
1360 	uint8_t			cmd_reserved0[20];
1361 	uint32_t		cmd_disp_type;
1362 #define MCX_TIR_CTX_DISP_TYPE_DIRECT	0
1363 #define MCX_TIR_CTX_DISP_TYPE_INDIRECT	1
1364 #define MCX_TIR_CTX_DISP_TYPE_SHIFT	28
1365 	uint8_t			cmd_reserved1[8];
1366 	uint32_t		cmd_lro;
1367 	uint8_t			cmd_reserved2[8];
1368 	uint32_t		cmd_inline_rqn;
1369 	uint32_t		cmd_indir_table;
1370 	uint32_t		cmd_tdomain;
1371 #define MCX_TIR_CTX_HASH_TOEPLITZ	2
1372 #define MCX_TIR_CTX_HASH_SHIFT		28
1373 	uint8_t			cmd_rx_hash_key[40];
1374 	uint32_t		cmd_rx_hash_sel_outer;
1375 #define MCX_TIR_CTX_HASH_SEL_SRC_IP	(1 << 0)
1376 #define MCX_TIR_CTX_HASH_SEL_DST_IP	(1 << 1)
1377 #define MCX_TIR_CTX_HASH_SEL_SPORT	(1 << 2)
1378 #define MCX_TIR_CTX_HASH_SEL_DPORT	(1 << 3)
1379 #define MCX_TIR_CTX_HASH_SEL_IPV4	(0 << 31)
1380 #define MCX_TIR_CTX_HASH_SEL_IPV6	(1 << 31)
1381 #define MCX_TIR_CTX_HASH_SEL_TCP	(0 << 30)
1382 #define MCX_TIR_CTX_HASH_SEL_UDP	(1 << 30)
1383 	uint32_t		cmd_rx_hash_sel_inner;
1384 	uint8_t			cmd_reserved3[152];
1385 } __packed __aligned(4);
1386 
1387 struct mcx_cmd_create_tir_out {
1388 	uint8_t			cmd_status;
1389 	uint8_t			cmd_reserved0[3];
1390 	uint32_t		cmd_syndrome;
1391 	uint32_t		cmd_tirn;
1392 	uint8_t			cmd_reserved1[4];
1393 } __packed __aligned(4);
1394 
1395 struct mcx_cmd_destroy_tir_in {
1396 	uint16_t		cmd_opcode;
1397 	uint8_t			cmd_reserved0[4];
1398 	uint16_t		cmd_op_mod;
1399 	uint32_t		cmd_tirn;
1400 	uint8_t			cmd_reserved1[4];
1401 } __packed __aligned(4);
1402 
1403 struct mcx_cmd_destroy_tir_out {
1404 	uint8_t			cmd_status;
1405 	uint8_t			cmd_reserved0[3];
1406 	uint32_t		cmd_syndrome;
1407 	uint8_t			cmd_reserved1[8];
1408 } __packed __aligned(4);
1409 
1410 struct mcx_cmd_create_tis_in {
1411 	uint16_t		cmd_opcode;
1412 	uint8_t			cmd_reserved0[4];
1413 	uint16_t		cmd_op_mod;
1414 	uint8_t			cmd_reserved1[8];
1415 } __packed __aligned(4);
1416 
1417 struct mcx_cmd_create_tis_mb_in {
1418 	uint8_t			cmd_reserved[16];
1419 	uint32_t		cmd_prio;
1420 	uint8_t			cmd_reserved1[32];
1421 	uint32_t		cmd_tdomain;
1422 	uint8_t			cmd_reserved2[120];
1423 } __packed __aligned(4);
1424 
1425 struct mcx_cmd_create_tis_out {
1426 	uint8_t			cmd_status;
1427 	uint8_t			cmd_reserved0[3];
1428 	uint32_t		cmd_syndrome;
1429 	uint32_t		cmd_tisn;
1430 	uint8_t			cmd_reserved1[4];
1431 } __packed __aligned(4);
1432 
1433 struct mcx_cmd_destroy_tis_in {
1434 	uint16_t		cmd_opcode;
1435 	uint8_t			cmd_reserved0[4];
1436 	uint16_t		cmd_op_mod;
1437 	uint32_t		cmd_tisn;
1438 	uint8_t			cmd_reserved1[4];
1439 } __packed __aligned(4);
1440 
1441 struct mcx_cmd_destroy_tis_out {
1442 	uint8_t			cmd_status;
1443 	uint8_t			cmd_reserved0[3];
1444 	uint32_t		cmd_syndrome;
1445 	uint8_t			cmd_reserved1[8];
1446 } __packed __aligned(4);
1447 
1448 struct mcx_cmd_create_rqt_in {
1449 	uint16_t		cmd_opcode;
1450 	uint8_t			cmd_reserved0[4];
1451 	uint16_t		cmd_op_mod;
1452 	uint8_t			cmd_reserved1[8];
1453 } __packed __aligned(4);
1454 
1455 struct mcx_rqt_ctx {
1456 	uint8_t			cmd_reserved0[20];
1457 	uint16_t		cmd_reserved1;
1458 	uint16_t		cmd_rqt_max_size;
1459 	uint16_t		cmd_reserved2;
1460 	uint16_t		cmd_rqt_actual_size;
1461 	uint8_t			cmd_reserved3[212];
1462 } __packed __aligned(4);
1463 
1464 struct mcx_cmd_create_rqt_mb_in {
1465 	uint8_t			cmd_reserved0[16];
1466 	struct mcx_rqt_ctx	cmd_rqt;
1467 } __packed __aligned(4);
1468 
1469 struct mcx_cmd_create_rqt_out {
1470 	uint8_t			cmd_status;
1471 	uint8_t			cmd_reserved0[3];
1472 	uint32_t		cmd_syndrome;
1473 	uint32_t		cmd_rqtn;
1474 	uint8_t			cmd_reserved1[4];
1475 } __packed __aligned(4);
1476 
1477 struct mcx_cmd_destroy_rqt_in {
1478 	uint16_t		cmd_opcode;
1479 	uint8_t			cmd_reserved0[4];
1480 	uint16_t		cmd_op_mod;
1481 	uint32_t		cmd_rqtn;
1482 	uint8_t			cmd_reserved1[4];
1483 } __packed __aligned(4);
1484 
1485 struct mcx_cmd_destroy_rqt_out {
1486 	uint8_t			cmd_status;
1487 	uint8_t			cmd_reserved0[3];
1488 	uint32_t		cmd_syndrome;
1489 	uint8_t			cmd_reserved1[8];
1490 } __packed __aligned(4);
1491 
1492 struct mcx_cq_ctx {
1493 	uint32_t		cq_status;
1494 #define MCX_CQ_CTX_STATUS_SHIFT		28
1495 #define MCX_CQ_CTX_STATUS_MASK		(0xf << MCX_CQ_CTX_STATUS_SHIFT)
1496 #define MCX_CQ_CTX_STATUS_OK		0x0
1497 #define MCX_CQ_CTX_STATUS_OVERFLOW	0x9
1498 #define MCX_CQ_CTX_STATUS_WRITE_FAIL	0xa
1499 #define MCX_CQ_CTX_STATE_SHIFT		8
1500 #define MCX_CQ_CTX_STATE_MASK		(0xf << MCX_CQ_CTX_STATE_SHIFT)
1501 #define MCX_CQ_CTX_STATE_SOLICITED	0x6
1502 #define MCX_CQ_CTX_STATE_ARMED		0x9
1503 #define MCX_CQ_CTX_STATE_FIRED		0xa
1504 	uint32_t		cq_reserved1;
1505 	uint32_t		cq_page_offset;
1506 	uint32_t		cq_uar_size;
1507 #define MCX_CQ_CTX_UAR_PAGE_MASK	0xffffff
1508 #define MCX_CQ_CTX_LOG_CQ_SIZE_SHIFT	24
1509 	uint32_t		cq_period_max_count;
1510 #define MCX_CQ_CTX_PERIOD_SHIFT		16
1511 	uint32_t		cq_eqn;
1512 	uint32_t		cq_log_page_size;
1513 #define MCX_CQ_CTX_LOG_PAGE_SIZE_SHIFT	24
1514 	uint32_t		cq_reserved2;
1515 	uint32_t		cq_last_notified;
1516 	uint32_t		cq_last_solicit;
1517 	uint32_t		cq_consumer_counter;
1518 	uint32_t		cq_producer_counter;
1519 	uint8_t			cq_reserved3[8];
1520 	uint64_t		cq_doorbell;
1521 } __packed __aligned(4);
1522 
1523 CTASSERT(sizeof(struct mcx_cq_ctx) == 64);
1524 
1525 struct mcx_cmd_create_cq_in {
1526 	uint16_t		cmd_opcode;
1527 	uint8_t			cmd_reserved0[4];
1528 	uint16_t		cmd_op_mod;
1529 	uint8_t			cmd_reserved1[8];
1530 } __packed __aligned(4);
1531 
1532 struct mcx_cmd_create_cq_mb_in {
1533 	struct mcx_cq_ctx	cmd_cq_ctx;
1534 	uint8_t			cmd_reserved1[192];
1535 } __packed __aligned(4);
1536 
1537 struct mcx_cmd_create_cq_out {
1538 	uint8_t			cmd_status;
1539 	uint8_t			cmd_reserved0[3];
1540 	uint32_t		cmd_syndrome;
1541 	uint32_t		cmd_cqn;
1542 	uint8_t			cmd_reserved1[4];
1543 } __packed __aligned(4);
1544 
1545 struct mcx_cmd_destroy_cq_in {
1546 	uint16_t		cmd_opcode;
1547 	uint8_t			cmd_reserved0[4];
1548 	uint16_t		cmd_op_mod;
1549 	uint32_t		cmd_cqn;
1550 	uint8_t			cmd_reserved1[4];
1551 } __packed __aligned(4);
1552 
1553 struct mcx_cmd_destroy_cq_out {
1554 	uint8_t			cmd_status;
1555 	uint8_t			cmd_reserved0[3];
1556 	uint32_t		cmd_syndrome;
1557 	uint8_t			cmd_reserved1[8];
1558 } __packed __aligned(4);
1559 
1560 struct mcx_cmd_query_cq_in {
1561 	uint16_t		cmd_opcode;
1562 	uint8_t			cmd_reserved0[4];
1563 	uint16_t		cmd_op_mod;
1564 	uint32_t		cmd_cqn;
1565 	uint8_t			cmd_reserved1[4];
1566 } __packed __aligned(4);
1567 
1568 struct mcx_cmd_query_cq_out {
1569 	uint8_t			cmd_status;
1570 	uint8_t			cmd_reserved0[3];
1571 	uint32_t		cmd_syndrome;
1572 	uint8_t			cmd_reserved1[8];
1573 } __packed __aligned(4);
1574 
1575 struct mcx_cq_entry {
1576 	uint32_t		__reserved__;
1577 	uint32_t		cq_lro;
1578 	uint32_t		cq_lro_ack_seq_num;
1579 	uint32_t		cq_rx_hash;
1580 	uint8_t			cq_rx_hash_type;
1581 	uint8_t			cq_ml_path;
1582 	uint16_t		__reserved__;
1583 	uint32_t		cq_checksum;
1584 	uint32_t		__reserved__;
1585 	uint32_t		cq_flags;
1586 #define MCX_CQ_ENTRY_FLAGS_L4_OK		(1 << 26)
1587 #define MCX_CQ_ENTRY_FLAGS_L3_OK		(1 << 25)
1588 #define MCX_CQ_ENTRY_FLAGS_L2_OK		(1 << 24)
1589 #define MCX_CQ_ENTRY_FLAGS_CV			(1 << 16)
1590 #define MCX_CQ_ENTRY_FLAGS_VLAN_MASK		(0xffff)
1591 
1592 	uint32_t		cq_lro_srqn;
1593 	uint32_t		__reserved__[2];
1594 	uint32_t		cq_byte_cnt;
1595 	uint64_t		cq_timestamp;
1596 	uint8_t			cq_rx_drops;
1597 	uint8_t			cq_flow_tag[3];
1598 	uint16_t		cq_wqe_count;
1599 	uint8_t			cq_signature;
1600 	uint8_t			cq_opcode_owner;
1601 #define MCX_CQ_ENTRY_FLAG_OWNER			(1 << 0)
1602 #define MCX_CQ_ENTRY_FLAG_SE			(1 << 1)
1603 #define MCX_CQ_ENTRY_FORMAT_SHIFT		2
1604 #define MCX_CQ_ENTRY_OPCODE_SHIFT		4
1605 
1606 #define MCX_CQ_ENTRY_FORMAT_NO_INLINE		0
1607 #define MCX_CQ_ENTRY_FORMAT_INLINE_32		1
1608 #define MCX_CQ_ENTRY_FORMAT_INLINE_64		2
1609 #define MCX_CQ_ENTRY_FORMAT_COMPRESSED		3
1610 
1611 #define MCX_CQ_ENTRY_OPCODE_REQ			0
1612 #define MCX_CQ_ENTRY_OPCODE_SEND		2
1613 #define MCX_CQ_ENTRY_OPCODE_REQ_ERR		13
1614 #define MCX_CQ_ENTRY_OPCODE_SEND_ERR		14
1615 #define MCX_CQ_ENTRY_OPCODE_INVALID		15
1616 
1617 } __packed __aligned(4);
1618 
1619 CTASSERT(sizeof(struct mcx_cq_entry) == 64);
1620 
1621 struct mcx_cq_doorbell {
1622 	uint32_t		 db_update_ci;
1623 	uint32_t		 db_arm_ci;
1624 #define MCX_CQ_DOORBELL_ARM_CMD_SN_SHIFT	28
1625 #define MCX_CQ_DOORBELL_ARM_CMD			(1 << 24)
1626 #define MCX_CQ_DOORBELL_ARM_CI_MASK		(0xffffff)
1627 } __packed __aligned(8);
1628 
1629 struct mcx_wq_ctx {
1630 	uint8_t			 wq_type;
1631 #define MCX_WQ_CTX_TYPE_CYCLIC			(1 << 4)
1632 #define MCX_WQ_CTX_TYPE_SIGNATURE		(1 << 3)
1633 	uint8_t			 wq_reserved0[5];
1634 	uint16_t		 wq_lwm;
1635 	uint32_t		 wq_pd;
1636 	uint32_t		 wq_uar_page;
1637 	uint64_t		 wq_doorbell;
1638 	uint32_t		 wq_hw_counter;
1639 	uint32_t		 wq_sw_counter;
1640 	uint16_t		 wq_log_stride;
1641 	uint8_t			 wq_log_page_sz;
1642 	uint8_t			 wq_log_size;
1643 	uint8_t			 wq_reserved1[156];
1644 } __packed __aligned(4);
1645 
1646 CTASSERT(sizeof(struct mcx_wq_ctx) == 0xC0);
1647 
1648 struct mcx_sq_ctx {
1649 	uint32_t		sq_flags;
1650 #define MCX_SQ_CTX_RLKEY			(1 << 31)
1651 #define MCX_SQ_CTX_FRE_SHIFT			(1 << 29)
1652 #define MCX_SQ_CTX_FLUSH_IN_ERROR		(1 << 28)
1653 #define MCX_SQ_CTX_MIN_WQE_INLINE_SHIFT		24
1654 #define MCX_SQ_CTX_STATE_SHIFT			20
1655 #define MCX_SQ_CTX_STATE_MASK			(0xf << 20)
1656 #define MCX_SQ_CTX_STATE_RST			0
1657 #define MCX_SQ_CTX_STATE_RDY			1
1658 #define MCX_SQ_CTX_STATE_ERR			3
1659 	uint32_t		sq_user_index;
1660 	uint32_t		sq_cqn;
1661 	uint32_t		sq_reserved1[5];
1662 	uint32_t		sq_tis_lst_sz;
1663 #define MCX_SQ_CTX_TIS_LST_SZ_SHIFT		16
1664 	uint32_t		sq_reserved2[2];
1665 	uint32_t		sq_tis_num;
1666 	struct mcx_wq_ctx	sq_wq;
1667 } __packed __aligned(4);
1668 
1669 struct mcx_sq_entry_seg {
1670 	uint32_t		sqs_byte_count;
1671 	uint32_t		sqs_lkey;
1672 	uint64_t		sqs_addr;
1673 } __packed __aligned(4);
1674 
1675 struct mcx_sq_entry {
1676 	/* control segment */
1677 	uint32_t		sqe_opcode_index;
1678 #define MCX_SQE_WQE_INDEX_SHIFT			8
1679 #define MCX_SQE_WQE_OPCODE_NOP			0x00
1680 #define MCX_SQE_WQE_OPCODE_SEND			0x0a
1681 	uint32_t		sqe_ds_sq_num;
1682 #define MCX_SQE_SQ_NUM_SHIFT			8
1683 	uint32_t		sqe_signature;
1684 #define MCX_SQE_SIGNATURE_SHIFT			24
1685 #define MCX_SQE_SOLICITED_EVENT			0x02
1686 #define MCX_SQE_CE_CQE_ON_ERR			0x00
1687 #define MCX_SQE_CE_CQE_FIRST_ERR		0x04
1688 #define MCX_SQE_CE_CQE_ALWAYS			0x08
1689 #define MCX_SQE_CE_CQE_SOLICIT			0x0C
1690 #define MCX_SQE_FM_NO_FENCE			0x00
1691 #define MCX_SQE_FM_SMALL_FENCE			0x40
1692 	uint32_t		sqe_mkey;
1693 
1694 	/* ethernet segment */
1695 	uint32_t		sqe_reserved1;
1696 	uint32_t		sqe_mss_csum;
1697 #define MCX_SQE_L4_CSUM				(1 << 31)
1698 #define MCX_SQE_L3_CSUM				(1 << 30)
1699 	uint32_t		sqe_reserved2;
1700 	uint16_t		sqe_inline_header_size;
1701 	uint16_t		sqe_inline_headers[9];
1702 
1703 	/* data segment */
1704 	struct mcx_sq_entry_seg sqe_segs[1];
1705 } __packed __aligned(64);
1706 
1707 CTASSERT(sizeof(struct mcx_sq_entry) == 64);
1708 
1709 struct mcx_cmd_create_sq_in {
1710 	uint16_t		cmd_opcode;
1711 	uint8_t			cmd_reserved0[4];
1712 	uint16_t		cmd_op_mod;
1713 	uint8_t			cmd_reserved1[8];
1714 } __packed __aligned(4);
1715 
1716 struct mcx_cmd_create_sq_out {
1717 	uint8_t			cmd_status;
1718 	uint8_t			cmd_reserved0[3];
1719 	uint32_t		cmd_syndrome;
1720 	uint32_t		cmd_sqn;
1721 	uint8_t			cmd_reserved1[4];
1722 } __packed __aligned(4);
1723 
1724 struct mcx_cmd_modify_sq_in {
1725 	uint16_t		cmd_opcode;
1726 	uint8_t			cmd_reserved0[4];
1727 	uint16_t		cmd_op_mod;
1728 	uint32_t		cmd_sq_state;
1729 	uint8_t			cmd_reserved1[4];
1730 } __packed __aligned(4);
1731 
1732 struct mcx_cmd_modify_sq_mb_in {
1733 	uint32_t		cmd_modify_hi;
1734 	uint32_t		cmd_modify_lo;
1735 	uint8_t			cmd_reserved0[8];
1736 	struct mcx_sq_ctx	cmd_sq_ctx;
1737 } __packed __aligned(4);
1738 
1739 struct mcx_cmd_modify_sq_out {
1740 	uint8_t			cmd_status;
1741 	uint8_t			cmd_reserved0[3];
1742 	uint32_t		cmd_syndrome;
1743 	uint8_t			cmd_reserved1[8];
1744 } __packed __aligned(4);
1745 
1746 struct mcx_cmd_destroy_sq_in {
1747 	uint16_t		cmd_opcode;
1748 	uint8_t			cmd_reserved0[4];
1749 	uint16_t		cmd_op_mod;
1750 	uint32_t		cmd_sqn;
1751 	uint8_t			cmd_reserved1[4];
1752 } __packed __aligned(4);
1753 
1754 struct mcx_cmd_destroy_sq_out {
1755 	uint8_t			cmd_status;
1756 	uint8_t			cmd_reserved0[3];
1757 	uint32_t		cmd_syndrome;
1758 	uint8_t			cmd_reserved1[8];
1759 } __packed __aligned(4);
1760 
1761 
1762 struct mcx_rq_ctx {
1763 	uint32_t		rq_flags;
1764 #define MCX_RQ_CTX_RLKEY			(1 << 31)
1765 #define MCX_RQ_CTX_VLAN_STRIP_DIS		(1 << 28)
1766 #define MCX_RQ_CTX_MEM_RQ_TYPE_SHIFT		24
1767 #define MCX_RQ_CTX_STATE_SHIFT			20
1768 #define MCX_RQ_CTX_STATE_MASK			(0xf << 20)
1769 #define MCX_RQ_CTX_STATE_RST			0
1770 #define MCX_RQ_CTX_STATE_RDY			1
1771 #define MCX_RQ_CTX_STATE_ERR			3
1772 #define MCX_RQ_CTX_FLUSH_IN_ERROR		(1 << 18)
1773 	uint32_t		rq_user_index;
1774 	uint32_t		rq_cqn;
1775 	uint32_t		rq_reserved1;
1776 	uint32_t		rq_rmpn;
1777 	uint32_t		rq_reserved2[7];
1778 	struct mcx_wq_ctx	rq_wq;
1779 } __packed __aligned(4);
1780 
1781 struct mcx_rq_entry {
1782 	uint32_t		rqe_byte_count;
1783 	uint32_t		rqe_lkey;
1784 	uint64_t		rqe_addr;
1785 } __packed __aligned(16);
1786 
1787 struct mcx_cmd_create_rq_in {
1788 	uint16_t		cmd_opcode;
1789 	uint8_t			cmd_reserved0[4];
1790 	uint16_t		cmd_op_mod;
1791 	uint8_t			cmd_reserved1[8];
1792 } __packed __aligned(4);
1793 
1794 struct mcx_cmd_create_rq_out {
1795 	uint8_t			cmd_status;
1796 	uint8_t			cmd_reserved0[3];
1797 	uint32_t		cmd_syndrome;
1798 	uint32_t		cmd_rqn;
1799 	uint8_t			cmd_reserved1[4];
1800 } __packed __aligned(4);
1801 
1802 struct mcx_cmd_modify_rq_in {
1803 	uint16_t		cmd_opcode;
1804 	uint8_t			cmd_reserved0[4];
1805 	uint16_t		cmd_op_mod;
1806 	uint32_t		cmd_rq_state;
1807 	uint8_t			cmd_reserved1[4];
1808 } __packed __aligned(4);
1809 
1810 struct mcx_cmd_modify_rq_mb_in {
1811 	uint32_t		cmd_modify_hi;
1812 	uint32_t		cmd_modify_lo;
1813 	uint8_t			cmd_reserved0[8];
1814 	struct mcx_rq_ctx	cmd_rq_ctx;
1815 } __packed __aligned(4);
1816 
1817 struct mcx_cmd_modify_rq_out {
1818 	uint8_t			cmd_status;
1819 	uint8_t			cmd_reserved0[3];
1820 	uint32_t		cmd_syndrome;
1821 	uint8_t			cmd_reserved1[8];
1822 } __packed __aligned(4);
1823 
1824 struct mcx_cmd_destroy_rq_in {
1825 	uint16_t		cmd_opcode;
1826 	uint8_t			cmd_reserved0[4];
1827 	uint16_t		cmd_op_mod;
1828 	uint32_t		cmd_rqn;
1829 	uint8_t			cmd_reserved1[4];
1830 } __packed __aligned(4);
1831 
1832 struct mcx_cmd_destroy_rq_out {
1833 	uint8_t			cmd_status;
1834 	uint8_t			cmd_reserved0[3];
1835 	uint32_t		cmd_syndrome;
1836 	uint8_t			cmd_reserved1[8];
1837 } __packed __aligned(4);
1838 
1839 struct mcx_cmd_create_flow_table_in {
1840 	uint16_t		cmd_opcode;
1841 	uint8_t			cmd_reserved0[4];
1842 	uint16_t		cmd_op_mod;
1843 	uint8_t			cmd_reserved1[8];
1844 } __packed __aligned(4);
1845 
1846 struct mcx_flow_table_ctx {
1847 	uint8_t			ft_miss_action;
1848 	uint8_t			ft_level;
1849 	uint8_t			ft_reserved0;
1850 	uint8_t			ft_log_size;
1851 	uint32_t		ft_table_miss_id;
1852 	uint8_t			ft_reserved1[28];
1853 } __packed __aligned(4);
1854 
1855 struct mcx_cmd_create_flow_table_mb_in {
1856 	uint8_t			cmd_table_type;
1857 	uint8_t			cmd_reserved0[7];
1858 	struct mcx_flow_table_ctx cmd_ctx;
1859 } __packed __aligned(4);
1860 
1861 struct mcx_cmd_create_flow_table_out {
1862 	uint8_t			cmd_status;
1863 	uint8_t			cmd_reserved0[3];
1864 	uint32_t		cmd_syndrome;
1865 	uint32_t		cmd_table_id;
1866 	uint8_t			cmd_reserved1[4];
1867 } __packed __aligned(4);
1868 
1869 struct mcx_cmd_destroy_flow_table_in {
1870 	uint16_t		cmd_opcode;
1871 	uint8_t			cmd_reserved0[4];
1872 	uint16_t		cmd_op_mod;
1873 	uint8_t			cmd_reserved1[8];
1874 } __packed __aligned(4);
1875 
1876 struct mcx_cmd_destroy_flow_table_mb_in {
1877 	uint8_t			cmd_table_type;
1878 	uint8_t			cmd_reserved0[3];
1879 	uint32_t		cmd_table_id;
1880 	uint8_t			cmd_reserved1[40];
1881 } __packed __aligned(4);
1882 
1883 struct mcx_cmd_destroy_flow_table_out {
1884 	uint8_t			cmd_status;
1885 	uint8_t			cmd_reserved0[3];
1886 	uint32_t		cmd_syndrome;
1887 	uint8_t			cmd_reserved1[8];
1888 } __packed __aligned(4);
1889 
1890 struct mcx_cmd_set_flow_table_root_in {
1891 	uint16_t		cmd_opcode;
1892 	uint8_t			cmd_reserved0[4];
1893 	uint16_t		cmd_op_mod;
1894 	uint8_t			cmd_reserved1[8];
1895 } __packed __aligned(4);
1896 
1897 struct mcx_cmd_set_flow_table_root_mb_in {
1898 	uint8_t			cmd_table_type;
1899 	uint8_t			cmd_reserved0[3];
1900 	uint32_t		cmd_table_id;
1901 	uint8_t			cmd_reserved1[56];
1902 } __packed __aligned(4);
1903 
1904 struct mcx_cmd_set_flow_table_root_out {
1905 	uint8_t			cmd_status;
1906 	uint8_t			cmd_reserved0[3];
1907 	uint32_t		cmd_syndrome;
1908 	uint8_t			cmd_reserved1[8];
1909 } __packed __aligned(4);
1910 
1911 struct mcx_flow_match {
1912 	/* outer headers */
1913 	uint8_t			mc_src_mac[6];
1914 	uint16_t		mc_ethertype;
1915 	uint8_t			mc_dest_mac[6];
1916 	uint16_t		mc_first_vlan;
1917 	uint8_t			mc_ip_proto;
1918 	uint8_t			mc_ip_dscp_ecn;
1919 	uint8_t			mc_vlan_flags;
1920 #define MCX_FLOW_MATCH_IP_FRAG	(1 << 5)
1921 	uint8_t			mc_tcp_flags;
1922 	uint16_t		mc_tcp_sport;
1923 	uint16_t		mc_tcp_dport;
1924 	uint32_t		mc_reserved0;
1925 	uint16_t		mc_udp_sport;
1926 	uint16_t		mc_udp_dport;
1927 	uint8_t			mc_src_ip[16];
1928 	uint8_t			mc_dest_ip[16];
1929 
1930 	/* misc parameters */
1931 	uint8_t			mc_reserved1[8];
1932 	uint16_t		mc_second_vlan;
1933 	uint8_t			mc_reserved2[2];
1934 	uint8_t			mc_second_vlan_flags;
1935 	uint8_t			mc_reserved3[15];
1936 	uint32_t		mc_outer_ipv6_flow_label;
1937 	uint8_t			mc_reserved4[32];
1938 
1939 	uint8_t			mc_reserved[384];
1940 } __packed __aligned(4);
1941 
1942 CTASSERT(sizeof(struct mcx_flow_match) == 512);
1943 
1944 struct mcx_cmd_create_flow_group_in {
1945 	uint16_t		cmd_opcode;
1946 	uint8_t			cmd_reserved0[4];
1947 	uint16_t		cmd_op_mod;
1948 	uint8_t			cmd_reserved1[8];
1949 } __packed __aligned(4);
1950 
1951 struct mcx_cmd_create_flow_group_mb_in {
1952 	uint8_t			cmd_table_type;
1953 	uint8_t			cmd_reserved0[3];
1954 	uint32_t		cmd_table_id;
1955 	uint8_t			cmd_reserved1[4];
1956 	uint32_t		cmd_start_flow_index;
1957 	uint8_t			cmd_reserved2[4];
1958 	uint32_t		cmd_end_flow_index;
1959 	uint8_t			cmd_reserved3[23];
1960 	uint8_t			cmd_match_criteria_enable;
1961 #define MCX_CREATE_FLOW_GROUP_CRIT_OUTER	(1 << 0)
1962 #define MCX_CREATE_FLOW_GROUP_CRIT_MISC		(1 << 1)
1963 #define MCX_CREATE_FLOW_GROUP_CRIT_INNER	(1 << 2)
1964 	struct mcx_flow_match	cmd_match_criteria;
1965 	uint8_t			cmd_reserved4[448];
1966 } __packed __aligned(4);
1967 
1968 struct mcx_cmd_create_flow_group_out {
1969 	uint8_t			cmd_status;
1970 	uint8_t			cmd_reserved0[3];
1971 	uint32_t		cmd_syndrome;
1972 	uint32_t		cmd_group_id;
1973 	uint8_t			cmd_reserved1[4];
1974 } __packed __aligned(4);
1975 
1976 struct mcx_flow_ctx {
1977 	uint8_t			fc_reserved0[4];
1978 	uint32_t		fc_group_id;
1979 	uint32_t		fc_flow_tag;
1980 	uint32_t		fc_action;
1981 #define MCX_FLOW_CONTEXT_ACTION_ALLOW		(1 << 0)
1982 #define MCX_FLOW_CONTEXT_ACTION_DROP		(1 << 1)
1983 #define MCX_FLOW_CONTEXT_ACTION_FORWARD		(1 << 2)
1984 #define MCX_FLOW_CONTEXT_ACTION_COUNT		(1 << 3)
1985 	uint32_t		fc_dest_list_size;
1986 	uint32_t		fc_counter_list_size;
1987 	uint8_t			fc_reserved1[40];
1988 	struct mcx_flow_match	fc_match_value;
1989 	uint8_t			fc_reserved2[192];
1990 } __packed __aligned(4);
1991 
1992 #define MCX_FLOW_CONTEXT_DEST_TYPE_TABLE	(1 << 24)
1993 #define MCX_FLOW_CONTEXT_DEST_TYPE_TIR		(2 << 24)
1994 
1995 struct mcx_cmd_destroy_flow_group_in {
1996 	uint16_t		cmd_opcode;
1997 	uint8_t			cmd_reserved0[4];
1998 	uint16_t		cmd_op_mod;
1999 	uint8_t			cmd_reserved1[8];
2000 } __packed __aligned(4);
2001 
2002 struct mcx_cmd_destroy_flow_group_mb_in {
2003 	uint8_t			cmd_table_type;
2004 	uint8_t			cmd_reserved0[3];
2005 	uint32_t		cmd_table_id;
2006 	uint32_t		cmd_group_id;
2007 	uint8_t			cmd_reserved1[36];
2008 } __packed __aligned(4);
2009 
2010 struct mcx_cmd_destroy_flow_group_out {
2011 	uint8_t			cmd_status;
2012 	uint8_t			cmd_reserved0[3];
2013 	uint32_t		cmd_syndrome;
2014 	uint8_t			cmd_reserved1[8];
2015 } __packed __aligned(4);
2016 
2017 struct mcx_cmd_set_flow_table_entry_in {
2018 	uint16_t		cmd_opcode;
2019 	uint8_t			cmd_reserved0[4];
2020 	uint16_t		cmd_op_mod;
2021 	uint8_t			cmd_reserved1[8];
2022 } __packed __aligned(4);
2023 
2024 struct mcx_cmd_set_flow_table_entry_mb_in {
2025 	uint8_t			cmd_table_type;
2026 	uint8_t			cmd_reserved0[3];
2027 	uint32_t		cmd_table_id;
2028 	uint32_t		cmd_modify_enable_mask;
2029 	uint8_t			cmd_reserved1[4];
2030 	uint32_t		cmd_flow_index;
2031 	uint8_t			cmd_reserved2[28];
2032 	struct mcx_flow_ctx	cmd_flow_ctx;
2033 } __packed __aligned(4);
2034 
2035 struct mcx_cmd_set_flow_table_entry_out {
2036 	uint8_t			cmd_status;
2037 	uint8_t			cmd_reserved0[3];
2038 	uint32_t		cmd_syndrome;
2039 	uint8_t			cmd_reserved1[8];
2040 } __packed __aligned(4);
2041 
2042 struct mcx_cmd_query_flow_table_entry_in {
2043 	uint16_t		cmd_opcode;
2044 	uint8_t			cmd_reserved0[4];
2045 	uint16_t		cmd_op_mod;
2046 	uint8_t			cmd_reserved1[8];
2047 } __packed __aligned(4);
2048 
2049 struct mcx_cmd_query_flow_table_entry_mb_in {
2050 	uint8_t			cmd_table_type;
2051 	uint8_t			cmd_reserved0[3];
2052 	uint32_t		cmd_table_id;
2053 	uint8_t			cmd_reserved1[8];
2054 	uint32_t		cmd_flow_index;
2055 	uint8_t			cmd_reserved2[28];
2056 } __packed __aligned(4);
2057 
2058 struct mcx_cmd_query_flow_table_entry_out {
2059 	uint8_t			cmd_status;
2060 	uint8_t			cmd_reserved0[3];
2061 	uint32_t		cmd_syndrome;
2062 	uint8_t			cmd_reserved1[8];
2063 } __packed __aligned(4);
2064 
2065 struct mcx_cmd_query_flow_table_entry_mb_out {
2066 	uint8_t			cmd_reserved0[48];
2067 	struct mcx_flow_ctx	cmd_flow_ctx;
2068 } __packed __aligned(4);
2069 
2070 struct mcx_cmd_delete_flow_table_entry_in {
2071 	uint16_t		cmd_opcode;
2072 	uint8_t			cmd_reserved0[4];
2073 	uint16_t		cmd_op_mod;
2074 	uint8_t			cmd_reserved1[8];
2075 } __packed __aligned(4);
2076 
2077 struct mcx_cmd_delete_flow_table_entry_mb_in {
2078 	uint8_t			cmd_table_type;
2079 	uint8_t			cmd_reserved0[3];
2080 	uint32_t		cmd_table_id;
2081 	uint8_t			cmd_reserved1[8];
2082 	uint32_t		cmd_flow_index;
2083 	uint8_t			cmd_reserved2[28];
2084 } __packed __aligned(4);
2085 
2086 struct mcx_cmd_delete_flow_table_entry_out {
2087 	uint8_t			cmd_status;
2088 	uint8_t			cmd_reserved0[3];
2089 	uint32_t		cmd_syndrome;
2090 	uint8_t			cmd_reserved1[8];
2091 } __packed __aligned(4);
2092 
2093 struct mcx_cmd_query_flow_group_in {
2094 	uint16_t		cmd_opcode;
2095 	uint8_t			cmd_reserved0[4];
2096 	uint16_t		cmd_op_mod;
2097 	uint8_t			cmd_reserved1[8];
2098 } __packed __aligned(4);
2099 
2100 struct mcx_cmd_query_flow_group_mb_in {
2101 	uint8_t			cmd_table_type;
2102 	uint8_t			cmd_reserved0[3];
2103 	uint32_t		cmd_table_id;
2104 	uint32_t		cmd_group_id;
2105 	uint8_t			cmd_reserved1[36];
2106 } __packed __aligned(4);
2107 
2108 struct mcx_cmd_query_flow_group_out {
2109 	uint8_t			cmd_status;
2110 	uint8_t			cmd_reserved0[3];
2111 	uint32_t		cmd_syndrome;
2112 	uint8_t			cmd_reserved1[8];
2113 } __packed __aligned(4);
2114 
2115 struct mcx_cmd_query_flow_group_mb_out {
2116 	uint8_t			cmd_reserved0[12];
2117 	uint32_t		cmd_start_flow_index;
2118 	uint8_t			cmd_reserved1[4];
2119 	uint32_t		cmd_end_flow_index;
2120 	uint8_t			cmd_reserved2[20];
2121 	uint32_t		cmd_match_criteria_enable;
2122 	uint8_t			cmd_match_criteria[512];
2123 	uint8_t			cmd_reserved4[448];
2124 } __packed __aligned(4);
2125 
2126 struct mcx_cmd_query_flow_table_in {
2127 	uint16_t		cmd_opcode;
2128 	uint8_t			cmd_reserved0[4];
2129 	uint16_t		cmd_op_mod;
2130 	uint8_t			cmd_reserved1[8];
2131 } __packed __aligned(4);
2132 
2133 struct mcx_cmd_query_flow_table_mb_in {
2134 	uint8_t			cmd_table_type;
2135 	uint8_t			cmd_reserved0[3];
2136 	uint32_t		cmd_table_id;
2137 	uint8_t			cmd_reserved1[40];
2138 } __packed __aligned(4);
2139 
2140 struct mcx_cmd_query_flow_table_out {
2141 	uint8_t			cmd_status;
2142 	uint8_t			cmd_reserved0[3];
2143 	uint32_t		cmd_syndrome;
2144 	uint8_t			cmd_reserved1[8];
2145 } __packed __aligned(4);
2146 
2147 struct mcx_cmd_query_flow_table_mb_out {
2148 	uint8_t			cmd_reserved0[4];
2149 	struct mcx_flow_table_ctx cmd_ctx;
2150 } __packed __aligned(4);
2151 
2152 struct mcx_cmd_alloc_flow_counter_in {
2153 	uint16_t		cmd_opcode;
2154 	uint8_t			cmd_reserved0[4];
2155 	uint16_t		cmd_op_mod;
2156 	uint8_t			cmd_reserved1[8];
2157 } __packed __aligned(4);
2158 
2159 struct mcx_cmd_query_rq_in {
2160 	uint16_t		cmd_opcode;
2161 	uint8_t			cmd_reserved0[4];
2162 	uint16_t		cmd_op_mod;
2163 	uint32_t		cmd_rqn;
2164 	uint8_t			cmd_reserved1[4];
2165 } __packed __aligned(4);
2166 
2167 struct mcx_cmd_query_rq_out {
2168 	uint8_t			cmd_status;
2169 	uint8_t			cmd_reserved0[3];
2170 	uint32_t		cmd_syndrome;
2171 	uint8_t			cmd_reserved1[8];
2172 } __packed __aligned(4);
2173 
2174 struct mcx_cmd_query_rq_mb_out {
2175 	uint8_t			cmd_reserved0[16];
2176 	struct mcx_rq_ctx	cmd_ctx;
2177 };
2178 
2179 struct mcx_cmd_query_sq_in {
2180 	uint16_t		cmd_opcode;
2181 	uint8_t			cmd_reserved0[4];
2182 	uint16_t		cmd_op_mod;
2183 	uint32_t		cmd_sqn;
2184 	uint8_t			cmd_reserved1[4];
2185 } __packed __aligned(4);
2186 
2187 struct mcx_cmd_query_sq_out {
2188 	uint8_t			cmd_status;
2189 	uint8_t			cmd_reserved0[3];
2190 	uint32_t		cmd_syndrome;
2191 	uint8_t			cmd_reserved1[8];
2192 } __packed __aligned(4);
2193 
2194 struct mcx_cmd_query_sq_mb_out {
2195 	uint8_t			cmd_reserved0[16];
2196 	struct mcx_sq_ctx	cmd_ctx;
2197 };
2198 
2199 struct mcx_cmd_alloc_flow_counter_out {
2200 	uint8_t			cmd_status;
2201 	uint8_t			cmd_reserved0[3];
2202 	uint32_t		cmd_syndrome;
2203 	uint8_t			cmd_reserved1[2];
2204 	uint16_t		cmd_flow_counter_id;
2205 	uint8_t			cmd_reserved2[4];
2206 } __packed __aligned(4);
2207 
2208 struct mcx_wq_doorbell {
2209 	uint32_t		 db_recv_counter;
2210 	uint32_t		 db_send_counter;
2211 } __packed __aligned(8);
2212 
2213 struct mcx_dmamem {
2214 	bus_dmamap_t		 mxm_map;
2215 	bus_dma_segment_t	 mxm_seg;
2216 	int			 mxm_nsegs;
2217 	size_t			 mxm_size;
2218 	caddr_t			 mxm_kva;
2219 };
2220 #define MCX_DMA_MAP(_mxm)	((_mxm)->mxm_map)
2221 #define MCX_DMA_DVA(_mxm)	((_mxm)->mxm_map->dm_segs[0].ds_addr)
2222 #define MCX_DMA_KVA(_mxm)	((void *)(_mxm)->mxm_kva)
2223 #define MCX_DMA_OFF(_mxm, _off)	((void *)((_mxm)->mxm_kva + (_off)))
2224 #define MCX_DMA_LEN(_mxm)	((_mxm)->mxm_size)
2225 
2226 struct mcx_hwmem {
2227 	bus_dmamap_t		 mhm_map;
2228 	bus_dma_segment_t	*mhm_segs;
2229 	unsigned int		 mhm_seg_count;
2230 	unsigned int		 mhm_npages;
2231 };
2232 
2233 struct mcx_slot {
2234 	bus_dmamap_t		 ms_map;
2235 	struct mbuf		*ms_m;
2236 };
2237 
2238 struct mcx_eq {
2239 	int			 eq_n;
2240 	uint32_t		 eq_cons;
2241 	struct mcx_dmamem	 eq_mem;
2242 };
2243 
2244 struct mcx_cq {
2245 	int			 cq_n;
2246 	struct mcx_dmamem	 cq_mem;
2247 	bus_addr_t		 cq_doorbell;
2248 	uint32_t		 cq_cons;
2249 	uint32_t		 cq_count;
2250 };
2251 
2252 struct mcx_calibration {
2253 	uint64_t		 c_timestamp;	/* previous mcx chip time */
2254 	uint64_t		 c_uptime;	/* previous kernel nanouptime */
2255 	uint64_t		 c_tbase;	/* mcx chip time */
2256 	uint64_t		 c_ubase;	/* kernel nanouptime */
2257 	uint64_t		 c_ratio;
2258 };
2259 
2260 #define MCX_CALIBRATE_FIRST    2
2261 #define MCX_CALIBRATE_NORMAL   32
2262 
2263 struct mcx_rx {
2264 	struct mcx_softc	*rx_softc;
2265 	struct ifiqueue		*rx_ifiq;
2266 
2267 	int			 rx_rqn;
2268 	struct mcx_dmamem	 rx_rq_mem;
2269 	struct mcx_slot		*rx_slots;
2270 	bus_addr_t		 rx_doorbell;
2271 
2272 	uint32_t		 rx_prod;
2273 	struct timeout		 rx_refill;
2274 	struct if_rxring	 rx_rxr;
2275 } __aligned(64);
2276 
2277 struct mcx_tx {
2278 	struct mcx_softc	*tx_softc;
2279 	struct ifqueue		*tx_ifq;
2280 
2281 	int			 tx_uar;
2282 	int			 tx_sqn;
2283 	struct mcx_dmamem	 tx_sq_mem;
2284 	struct mcx_slot		*tx_slots;
2285 	bus_addr_t		 tx_doorbell;
2286 	int			 tx_bf_offset;
2287 
2288 	uint32_t		 tx_cons;
2289 	uint32_t		 tx_prod;
2290 } __aligned(64);
2291 
2292 struct mcx_queues {
2293 	char			 q_name[16];
2294 	void			*q_ihc;
2295 	struct mcx_softc	*q_sc;
2296 	int			 q_uar;
2297 	int			 q_index;
2298 	struct mcx_rx		 q_rx;
2299 	struct mcx_tx		 q_tx;
2300 	struct mcx_cq		 q_cq;
2301 	struct mcx_eq		 q_eq;
2302 #if NKSTAT > 0
2303 	struct kstat		*q_kstat;
2304 #endif
2305 };
2306 
2307 struct mcx_flow_group {
2308 	int			 g_id;
2309 	int			 g_table;
2310 	int			 g_start;
2311 	int			 g_size;
2312 };
2313 
2314 #define MCX_FLOW_GROUP_PROMISC	 0
2315 #define MCX_FLOW_GROUP_ALLMULTI	 1
2316 #define MCX_FLOW_GROUP_MAC	 2
2317 #define MCX_FLOW_GROUP_RSS_L4	 3
2318 #define MCX_FLOW_GROUP_RSS_L3	 4
2319 #define MCX_FLOW_GROUP_RSS_NONE	 5
2320 #define MCX_NUM_FLOW_GROUPS	 6
2321 
2322 #define MCX_HASH_SEL_L3		MCX_TIR_CTX_HASH_SEL_SRC_IP | \
2323 				MCX_TIR_CTX_HASH_SEL_DST_IP
2324 #define MCX_HASH_SEL_L4		MCX_HASH_SEL_L3 | MCX_TIR_CTX_HASH_SEL_SPORT | \
2325 				MCX_TIR_CTX_HASH_SEL_DPORT
2326 
2327 #define MCX_RSS_HASH_SEL_V4_TCP MCX_HASH_SEL_L4 | MCX_TIR_CTX_HASH_SEL_TCP  |\
2328 				MCX_TIR_CTX_HASH_SEL_IPV4
2329 #define MCX_RSS_HASH_SEL_V6_TCP	MCX_HASH_SEL_L4 | MCX_TIR_CTX_HASH_SEL_TCP | \
2330 				MCX_TIR_CTX_HASH_SEL_IPV6
2331 #define MCX_RSS_HASH_SEL_V4_UDP	MCX_HASH_SEL_L4 | MCX_TIR_CTX_HASH_SEL_UDP | \
2332 				MCX_TIR_CTX_HASH_SEL_IPV4
2333 #define MCX_RSS_HASH_SEL_V6_UDP	MCX_HASH_SEL_L4 | MCX_TIR_CTX_HASH_SEL_UDP | \
2334 				MCX_TIR_CTX_HASH_SEL_IPV6
2335 #define MCX_RSS_HASH_SEL_V4	MCX_HASH_SEL_L3 | MCX_TIR_CTX_HASH_SEL_IPV4
2336 #define MCX_RSS_HASH_SEL_V6	MCX_HASH_SEL_L3 | MCX_TIR_CTX_HASH_SEL_IPV6
2337 
2338 /*
2339  * There are a few different pieces involved in configuring RSS.
2340  * A Receive Queue Table (RQT) is the indirection table that maps packets to
2341  * different rx queues based on a hash value.  We only create one, because
2342  * we want to scatter any traffic we can apply RSS to across all our rx
2343  * queues.  Anything else will only be delivered to the first rx queue,
2344  * which doesn't require an RQT.
2345  *
2346  * A Transport Interface Receive (TIR) delivers packets to either a single rx
2347  * queue or an RQT, and in the latter case, specifies the set of fields
2348  * hashed, the hash function, and the hash key.  We need one of these for each
2349  * type of RSS traffic - v4 TCP, v6 TCP, v4 UDP, v6 UDP, other v4, other v6,
2350  * and one for non-RSS traffic.
2351  *
2352  * Flow tables hold flow table entries in sequence.  The first entry that
2353  * matches a packet is applied, sending the packet to either another flow
2354  * table or a TIR.  We use one flow table to select packets based on
2355  * destination MAC address, and a second to apply RSS.  The entries in the
2356  * first table send matching packets to the second, and the entries in the
2357  * RSS table send packets to RSS TIRs if possible, or the non-RSS TIR.
2358  *
2359  * The flow table entry that delivers packets to an RSS TIR must include match
2360  * criteria that ensure packets delivered to the TIR include all the fields
2361  * that the TIR hashes on - so for a v4 TCP TIR, the flow table entry must
2362  * only accept v4 TCP packets.  Accordingly, we need flow table entries for
2363  * each TIR.
2364  *
2365  * All of this is a lot more flexible than we need, and we can describe most
2366  * of the stuff we need with a simple array.
2367  *
2368  * An RSS config creates a TIR with hashing enabled on a set of fields,
2369  * pointing to either the first rx queue or the RQT containing all the rx
2370  * queues, and a flow table entry that matches on an ether type and
2371  * optionally an ip proto, that delivers packets to the TIR.
2372  */
2373 static struct mcx_rss_rule {
2374 	int			hash_sel;
2375 	int			flow_group;
2376 	int			ethertype;
2377 	int			ip_proto;
2378 } mcx_rss_config[] = {
2379 	/* udp and tcp for v4/v6 */
2380 	{ MCX_RSS_HASH_SEL_V4_TCP, MCX_FLOW_GROUP_RSS_L4,
2381 	  ETHERTYPE_IP, IPPROTO_TCP },
2382 	{ MCX_RSS_HASH_SEL_V6_TCP, MCX_FLOW_GROUP_RSS_L4,
2383 	  ETHERTYPE_IPV6, IPPROTO_TCP },
2384 	{ MCX_RSS_HASH_SEL_V4_UDP, MCX_FLOW_GROUP_RSS_L4,
2385 	  ETHERTYPE_IP, IPPROTO_UDP },
2386 	{ MCX_RSS_HASH_SEL_V6_UDP, MCX_FLOW_GROUP_RSS_L4,
2387 	  ETHERTYPE_IPV6, IPPROTO_UDP },
2388 
2389 	/* other v4/v6 */
2390 	{ MCX_RSS_HASH_SEL_V4, MCX_FLOW_GROUP_RSS_L3,
2391 	  ETHERTYPE_IP, 0 },
2392 	{ MCX_RSS_HASH_SEL_V6, MCX_FLOW_GROUP_RSS_L3,
2393 	  ETHERTYPE_IPV6, 0 },
2394 
2395 	/* non v4/v6 */
2396 	{ 0, MCX_FLOW_GROUP_RSS_NONE, 0, 0 }
2397 };
2398 
2399 struct mcx_softc {
2400 	struct device		 sc_dev;
2401 	struct arpcom		 sc_ac;
2402 	struct ifmedia		 sc_media;
2403 	uint64_t		 sc_media_status;
2404 	uint64_t		 sc_media_active;
2405 
2406 	pci_chipset_tag_t	 sc_pc;
2407 	pci_intr_handle_t	 sc_ih;
2408 	void			*sc_ihc;
2409 	pcitag_t		 sc_tag;
2410 
2411 	bus_dma_tag_t		 sc_dmat;
2412 	bus_space_tag_t		 sc_memt;
2413 	bus_space_handle_t	 sc_memh;
2414 	bus_size_t		 sc_mems;
2415 
2416 	struct mcx_dmamem	 sc_cmdq_mem;
2417 	unsigned int		 sc_cmdq_mask;
2418 	unsigned int		 sc_cmdq_size;
2419 
2420 	unsigned int		 sc_cmdq_token;
2421 
2422 	struct mcx_hwmem	 sc_boot_pages;
2423 	struct mcx_hwmem	 sc_init_pages;
2424 	struct mcx_hwmem	 sc_regular_pages;
2425 
2426 	int			 sc_uar;
2427 	int			 sc_pd;
2428 	int			 sc_tdomain;
2429 	uint32_t		 sc_lkey;
2430 	int			 sc_tis;
2431 	int			 sc_tir[nitems(mcx_rss_config)];
2432 	int			 sc_rqt;
2433 
2434 	struct mcx_dmamem	 sc_doorbell_mem;
2435 
2436 	struct mcx_eq		 sc_admin_eq;
2437 	struct mcx_eq		 sc_queue_eq;
2438 
2439 	int			 sc_hardmtu;
2440 	int			 sc_rxbufsz;
2441 
2442 	int			 sc_bf_size;
2443 	int			 sc_max_rqt_size;
2444 
2445 	struct task		 sc_port_change;
2446 
2447 	int			 sc_mac_flow_table_id;
2448 	int			 sc_rss_flow_table_id;
2449 	struct mcx_flow_group	 sc_flow_group[MCX_NUM_FLOW_GROUPS];
2450 	int			 sc_promisc_flow_enabled;
2451 	int			 sc_allmulti_flow_enabled;
2452 	int			 sc_mcast_flow_base;
2453 	int			 sc_extra_mcast;
2454 	uint8_t			 sc_mcast_flows[MCX_NUM_MCAST_FLOWS][ETHER_ADDR_LEN];
2455 
2456 	struct mcx_calibration	 sc_calibration[2];
2457 	unsigned int		 sc_calibration_gen;
2458 	struct timeout		 sc_calibrate;
2459 	uint32_t		 sc_mhz;
2460 	uint32_t		 sc_khz;
2461 
2462 	struct mcx_queues	 sc_queues[MCX_MAX_QUEUES];
2463 	unsigned int		 sc_nqueues;
2464 
2465 #if NKSTAT > 0
2466 	struct kstat		*sc_kstat_ieee8023;
2467 	struct kstat		*sc_kstat_rfc2863;
2468 	struct kstat		*sc_kstat_rfc2819;
2469 	struct kstat		*sc_kstat_rfc3635;
2470 	unsigned int		 sc_kstat_mtmp_count;
2471 	struct kstat		**sc_kstat_mtmp;
2472 #endif
2473 
2474 	struct timecounter	 sc_timecounter;
2475 };
2476 #define DEVNAME(_sc) ((_sc)->sc_dev.dv_xname)
2477 
2478 static int	mcx_match(struct device *, void *, void *);
2479 static void	mcx_attach(struct device *, struct device *, void *);
2480 
2481 #if NKSTAT > 0
2482 static void	mcx_kstat_attach(struct mcx_softc *);
2483 #endif
2484 
2485 static void	mcx_timecounter_attach(struct mcx_softc *);
2486 
2487 static int	mcx_version(struct mcx_softc *);
2488 static int	mcx_init_wait(struct mcx_softc *);
2489 static int	mcx_enable_hca(struct mcx_softc *);
2490 static int	mcx_teardown_hca(struct mcx_softc *, uint16_t);
2491 static int	mcx_access_hca_reg(struct mcx_softc *, uint16_t, int, void *,
2492 		    int);
2493 static int	mcx_issi(struct mcx_softc *);
2494 static int	mcx_pages(struct mcx_softc *, struct mcx_hwmem *, uint16_t);
2495 static int	mcx_hca_max_caps(struct mcx_softc *);
2496 static int	mcx_hca_set_caps(struct mcx_softc *);
2497 static int	mcx_init_hca(struct mcx_softc *);
2498 static int	mcx_set_driver_version(struct mcx_softc *);
2499 static int	mcx_iff(struct mcx_softc *);
2500 static int	mcx_alloc_uar(struct mcx_softc *, int *);
2501 static int	mcx_alloc_pd(struct mcx_softc *);
2502 static int	mcx_alloc_tdomain(struct mcx_softc *);
2503 static int	mcx_create_eq(struct mcx_softc *, struct mcx_eq *, int,
2504 		    uint64_t, int);
2505 static int	mcx_query_nic_vport_context(struct mcx_softc *);
2506 static int	mcx_query_special_contexts(struct mcx_softc *);
2507 static int	mcx_set_port_mtu(struct mcx_softc *, int);
2508 static int	mcx_create_cq(struct mcx_softc *, struct mcx_cq *, int, int,
2509 		    int);
2510 static int	mcx_destroy_cq(struct mcx_softc *, struct mcx_cq *);
2511 static int	mcx_create_sq(struct mcx_softc *, struct mcx_tx *, int, int,
2512 		    int);
2513 static int	mcx_destroy_sq(struct mcx_softc *, struct mcx_tx *);
2514 static int	mcx_ready_sq(struct mcx_softc *, struct mcx_tx *);
2515 static int	mcx_create_rq(struct mcx_softc *, struct mcx_rx *, int, int);
2516 static int	mcx_destroy_rq(struct mcx_softc *, struct mcx_rx *);
2517 static int	mcx_ready_rq(struct mcx_softc *, struct mcx_rx *);
2518 static int	mcx_create_tir_direct(struct mcx_softc *, struct mcx_rx *,
2519 		    int *);
2520 static int	mcx_create_tir_indirect(struct mcx_softc *, int, uint32_t,
2521 		    int *);
2522 static int	mcx_destroy_tir(struct mcx_softc *, int);
2523 static int	mcx_create_tis(struct mcx_softc *, int *);
2524 static int	mcx_destroy_tis(struct mcx_softc *, int);
2525 static int	mcx_create_rqt(struct mcx_softc *, int, int *, int *);
2526 static int	mcx_destroy_rqt(struct mcx_softc *, int);
2527 static int	mcx_create_flow_table(struct mcx_softc *, int, int, int *);
2528 static int	mcx_set_flow_table_root(struct mcx_softc *, int);
2529 static int	mcx_destroy_flow_table(struct mcx_softc *, int);
2530 static int	mcx_create_flow_group(struct mcx_softc *, int, int, int,
2531 		    int, int, struct mcx_flow_match *);
2532 static int	mcx_destroy_flow_group(struct mcx_softc *, int);
2533 static int	mcx_set_flow_table_entry_mac(struct mcx_softc *, int, int,
2534 		    uint8_t *, uint32_t);
2535 static int	mcx_set_flow_table_entry_proto(struct mcx_softc *, int, int,
2536 		    int, int, uint32_t);
2537 static int	mcx_delete_flow_table_entry(struct mcx_softc *, int, int);
2538 
2539 #if NKSTAT > 0
2540 static int	mcx_query_rq(struct mcx_softc *, struct mcx_rx *, struct mcx_rq_ctx *);
2541 static int	mcx_query_sq(struct mcx_softc *, struct mcx_tx *, struct mcx_sq_ctx *);
2542 static int	mcx_query_cq(struct mcx_softc *, struct mcx_cq *, struct mcx_cq_ctx *);
2543 static int	mcx_query_eq(struct mcx_softc *, struct mcx_eq *, struct mcx_eq_ctx *);
2544 #endif
2545 
2546 #if 0
2547 static int	mcx_dump_flow_table(struct mcx_softc *, int);
2548 static int	mcx_dump_flow_table_entry(struct mcx_softc *, int, int);
2549 static int	mcx_dump_flow_group(struct mcx_softc *, int);
2550 #endif
2551 
2552 
2553 /*
2554 static void	mcx_cmdq_dump(const struct mcx_cmdq_entry *);
2555 static void	mcx_cmdq_mbox_dump(struct mcx_dmamem *, int);
2556 */
2557 static void	mcx_refill(void *);
2558 static int	mcx_process_rx(struct mcx_softc *, struct mcx_rx *,
2559 		    struct mcx_cq_entry *, struct mbuf_list *,
2560 		    const struct mcx_calibration *);
2561 static int	mcx_process_txeof(struct mcx_softc *, struct mcx_tx *,
2562 		    struct mcx_cq_entry *);
2563 static void	mcx_process_cq(struct mcx_softc *, struct mcx_queues *,
2564 		    struct mcx_cq *);
2565 
2566 static void	mcx_arm_cq(struct mcx_softc *, struct mcx_cq *, int);
2567 static void	mcx_arm_eq(struct mcx_softc *, struct mcx_eq *, int);
2568 static int	mcx_admin_intr(void *);
2569 static int	mcx_cq_intr(void *);
2570 
2571 static int	mcx_up(struct mcx_softc *);
2572 static void	mcx_down(struct mcx_softc *);
2573 static int	mcx_ioctl(struct ifnet *, u_long, caddr_t);
2574 static int	mcx_rxrinfo(struct mcx_softc *, struct if_rxrinfo *);
2575 static void	mcx_start(struct ifqueue *);
2576 static void	mcx_watchdog(struct ifnet *);
2577 static void	mcx_media_add_types(struct mcx_softc *);
2578 static void	mcx_media_status(struct ifnet *, struct ifmediareq *);
2579 static int	mcx_media_change(struct ifnet *);
2580 static int	mcx_get_sffpage(struct ifnet *, struct if_sffpage *);
2581 static void	mcx_port_change(void *);
2582 
2583 static void	mcx_calibrate_first(struct mcx_softc *);
2584 static void	mcx_calibrate(void *);
2585 
2586 static inline uint32_t
2587 		mcx_rd(struct mcx_softc *, bus_size_t);
2588 static inline void
2589 		mcx_wr(struct mcx_softc *, bus_size_t, uint32_t);
2590 static inline void
2591 		mcx_bar(struct mcx_softc *, bus_size_t, bus_size_t, int);
2592 
2593 static uint64_t	mcx_timer(struct mcx_softc *);
2594 
2595 static int	mcx_dmamem_alloc(struct mcx_softc *, struct mcx_dmamem *,
2596 		    bus_size_t, u_int align);
2597 static void	mcx_dmamem_zero(struct mcx_dmamem *);
2598 static void	mcx_dmamem_free(struct mcx_softc *, struct mcx_dmamem *);
2599 
2600 static int	mcx_hwmem_alloc(struct mcx_softc *, struct mcx_hwmem *,
2601 		    unsigned int);
2602 static void	mcx_hwmem_free(struct mcx_softc *, struct mcx_hwmem *);
2603 
2604 struct cfdriver mcx_cd = {
2605 	NULL,
2606 	"mcx",
2607 	DV_IFNET,
2608 };
2609 
2610 struct cfattach mcx_ca = {
2611 	sizeof(struct mcx_softc),
2612 	mcx_match,
2613 	mcx_attach,
2614 };
2615 
2616 static const struct pci_matchid mcx_devices[] = {
2617 	{ PCI_VENDOR_MELLANOX,	PCI_PRODUCT_MELLANOX_MT27700 },
2618 	{ PCI_VENDOR_MELLANOX,	PCI_PRODUCT_MELLANOX_MT27700VF },
2619 	{ PCI_VENDOR_MELLANOX,	PCI_PRODUCT_MELLANOX_MT27710 },
2620 	{ PCI_VENDOR_MELLANOX,	PCI_PRODUCT_MELLANOX_MT27710VF },
2621 	{ PCI_VENDOR_MELLANOX,	PCI_PRODUCT_MELLANOX_MT27800 },
2622 	{ PCI_VENDOR_MELLANOX,	PCI_PRODUCT_MELLANOX_MT27800VF },
2623 	{ PCI_VENDOR_MELLANOX,	PCI_PRODUCT_MELLANOX_MT28800 },
2624 	{ PCI_VENDOR_MELLANOX,	PCI_PRODUCT_MELLANOX_MT28800VF },
2625 	{ PCI_VENDOR_MELLANOX,	PCI_PRODUCT_MELLANOX_MT28908 },
2626 	{ PCI_VENDOR_MELLANOX,	PCI_PRODUCT_MELLANOX_MT2892  },
2627 };
2628 
2629 struct mcx_eth_proto_capability {
2630 	uint64_t	cap_media;
2631 	uint64_t	cap_baudrate;
2632 };
2633 
2634 static const struct mcx_eth_proto_capability mcx_eth_cap_map[] = {
2635 	[MCX_ETHER_CAP_SGMII]		= { IFM_1000_SGMII,	IF_Gbps(1) },
2636 	[MCX_ETHER_CAP_1000_KX]		= { IFM_1000_KX,	IF_Gbps(1) },
2637 	[MCX_ETHER_CAP_10G_CX4]		= { IFM_10G_CX4,	IF_Gbps(10) },
2638 	[MCX_ETHER_CAP_10G_KX4]		= { IFM_10G_KX4,	IF_Gbps(10) },
2639 	[MCX_ETHER_CAP_10G_KR]		= { IFM_10G_KR,		IF_Gbps(10) },
2640 	[MCX_ETHER_CAP_40G_CR4]		= { IFM_40G_CR4,	IF_Gbps(40) },
2641 	[MCX_ETHER_CAP_40G_KR4]		= { IFM_40G_KR4,	IF_Gbps(40) },
2642 	[MCX_ETHER_CAP_10G_CR]		= { IFM_10G_SFP_CU,	IF_Gbps(10) },
2643 	[MCX_ETHER_CAP_10G_SR]		= { IFM_10G_SR,		IF_Gbps(10) },
2644 	[MCX_ETHER_CAP_10G_LR]		= { IFM_10G_LR,		IF_Gbps(10) },
2645 	[MCX_ETHER_CAP_40G_SR4]		= { IFM_40G_SR4,	IF_Gbps(40) },
2646 	[MCX_ETHER_CAP_40G_LR4]		= { IFM_40G_LR4,	IF_Gbps(40) },
2647 	[MCX_ETHER_CAP_50G_SR2]		= { 0 /*IFM_50G_SR2*/,	IF_Gbps(50) },
2648 	[MCX_ETHER_CAP_100G_CR4]	= { IFM_100G_CR4,	IF_Gbps(100) },
2649 	[MCX_ETHER_CAP_100G_SR4]	= { IFM_100G_SR4,	IF_Gbps(100) },
2650 	[MCX_ETHER_CAP_100G_KR4]	= { IFM_100G_KR4,	IF_Gbps(100) },
2651 	[MCX_ETHER_CAP_25G_CR]		= { IFM_25G_CR,		IF_Gbps(25) },
2652 	[MCX_ETHER_CAP_25G_KR]		= { IFM_25G_KR,		IF_Gbps(25) },
2653 	[MCX_ETHER_CAP_25G_SR]		= { IFM_25G_SR,		IF_Gbps(25) },
2654 	[MCX_ETHER_CAP_50G_CR2]		= { IFM_50G_CR2,	IF_Gbps(50) },
2655 	[MCX_ETHER_CAP_50G_KR2]		= { IFM_50G_KR2,	IF_Gbps(50) },
2656 };
2657 
2658 static int
2659 mcx_get_id(uint32_t val)
2660 {
2661 	return betoh32(val) & 0x00ffffff;
2662 }
2663 
2664 static int
2665 mcx_match(struct device *parent, void *match, void *aux)
2666 {
2667 	return (pci_matchbyid(aux, mcx_devices, nitems(mcx_devices)));
2668 }
2669 
2670 void
2671 mcx_attach(struct device *parent, struct device *self, void *aux)
2672 {
2673 	struct mcx_softc *sc = (struct mcx_softc *)self;
2674 	struct ifnet *ifp = &sc->sc_ac.ac_if;
2675 	struct pci_attach_args *pa = aux;
2676 	pcireg_t memtype;
2677 	uint32_t r;
2678 	unsigned int cq_stride;
2679 	unsigned int cq_size;
2680 	const char *intrstr;
2681 	int i, msix;
2682 
2683 	sc->sc_pc = pa->pa_pc;
2684 	sc->sc_tag = pa->pa_tag;
2685 	sc->sc_dmat = pa->pa_dmat;
2686 
2687 	/* Map the PCI memory space */
2688 	memtype = pci_mapreg_type(sc->sc_pc, sc->sc_tag, MCX_HCA_BAR);
2689 	if (pci_mapreg_map(pa, MCX_HCA_BAR, memtype,
2690 	    BUS_SPACE_MAP_PREFETCHABLE, &sc->sc_memt, &sc->sc_memh,
2691 	    NULL, &sc->sc_mems, 0)) {
2692 		printf(": unable to map register memory\n");
2693 		return;
2694 	}
2695 
2696 	if (mcx_version(sc) != 0) {
2697 		/* error printed by mcx_version */
2698 		goto unmap;
2699 	}
2700 
2701 	r = mcx_rd(sc, MCX_CMDQ_ADDR_LO);
2702 	cq_stride = 1 << MCX_CMDQ_LOG_STRIDE(r); /* size of the entries */
2703 	cq_size = 1 << MCX_CMDQ_LOG_SIZE(r); /* number of entries */
2704 	if (cq_size > MCX_MAX_CQE) {
2705 		printf(", command queue size overflow %u\n", cq_size);
2706 		goto unmap;
2707 	}
2708 	if (cq_stride < sizeof(struct mcx_cmdq_entry)) {
2709 		printf(", command queue entry size underflow %u\n", cq_stride);
2710 		goto unmap;
2711 	}
2712 	if (cq_stride * cq_size > MCX_PAGE_SIZE) {
2713 		printf(", command queue page overflow\n");
2714 		goto unmap;
2715 	}
2716 
2717 	if (mcx_dmamem_alloc(sc, &sc->sc_doorbell_mem, MCX_DOORBELL_AREA_SIZE,
2718 	    MCX_PAGE_SIZE) != 0) {
2719 		printf(", unable to allocate doorbell memory\n");
2720 		goto unmap;
2721 	}
2722 
2723 	if (mcx_dmamem_alloc(sc, &sc->sc_cmdq_mem, MCX_PAGE_SIZE,
2724 	    MCX_PAGE_SIZE) != 0) {
2725 		printf(", unable to allocate command queue\n");
2726 		goto dbfree;
2727 	}
2728 
2729 	mcx_wr(sc, MCX_CMDQ_ADDR_HI, MCX_DMA_DVA(&sc->sc_cmdq_mem) >> 32);
2730 	mcx_bar(sc, MCX_CMDQ_ADDR_HI, sizeof(uint32_t),
2731 	    BUS_SPACE_BARRIER_WRITE);
2732 	mcx_wr(sc, MCX_CMDQ_ADDR_LO, MCX_DMA_DVA(&sc->sc_cmdq_mem));
2733 	mcx_bar(sc, MCX_CMDQ_ADDR_LO, sizeof(uint32_t),
2734 	    BUS_SPACE_BARRIER_WRITE);
2735 
2736 	if (mcx_init_wait(sc) != 0) {
2737 		printf(", timeout waiting for init\n");
2738 		goto cqfree;
2739 	}
2740 
2741 	sc->sc_cmdq_mask = cq_size - 1;
2742 	sc->sc_cmdq_size = cq_stride;
2743 
2744 	if (mcx_enable_hca(sc) != 0) {
2745 		/* error printed by mcx_enable_hca */
2746 		goto cqfree;
2747 	}
2748 
2749 	if (mcx_issi(sc) != 0) {
2750 		/* error printed by mcx_issi */
2751 		goto teardown;
2752 	}
2753 
2754 	if (mcx_pages(sc, &sc->sc_boot_pages,
2755 	    htobe16(MCX_CMD_QUERY_PAGES_BOOT)) != 0) {
2756 		/* error printed by mcx_pages */
2757 		goto teardown;
2758 	}
2759 
2760 	if (mcx_hca_max_caps(sc) != 0) {
2761 		/* error printed by mcx_hca_max_caps */
2762 		goto teardown;
2763 	}
2764 
2765 	if (mcx_hca_set_caps(sc) != 0) {
2766 		/* error printed by mcx_hca_set_caps */
2767 		goto teardown;
2768 	}
2769 
2770 	if (mcx_pages(sc, &sc->sc_init_pages,
2771 	    htobe16(MCX_CMD_QUERY_PAGES_INIT)) != 0) {
2772 		/* error printed by mcx_pages */
2773 		goto teardown;
2774 	}
2775 
2776 	if (mcx_init_hca(sc) != 0) {
2777 		/* error printed by mcx_init_hca */
2778 		goto teardown;
2779 	}
2780 
2781 	if (mcx_pages(sc, &sc->sc_regular_pages,
2782 	    htobe16(MCX_CMD_QUERY_PAGES_REGULAR)) != 0) {
2783 		/* error printed by mcx_pages */
2784 		goto teardown;
2785 	}
2786 
2787 	/* apparently not necessary? */
2788 	if (mcx_set_driver_version(sc) != 0) {
2789 		/* error printed by mcx_set_driver_version */
2790 		goto teardown;
2791 	}
2792 
2793 	if (mcx_iff(sc) != 0) {	/* modify nic vport context */
2794 		/* error printed by mcx_iff? */
2795 		goto teardown;
2796 	}
2797 
2798 	if (mcx_alloc_uar(sc, &sc->sc_uar) != 0) {
2799 		/* error printed by mcx_alloc_uar */
2800 		goto teardown;
2801 	}
2802 
2803 	if (mcx_alloc_pd(sc) != 0) {
2804 		/* error printed by mcx_alloc_pd */
2805 		goto teardown;
2806 	}
2807 
2808 	if (mcx_alloc_tdomain(sc) != 0) {
2809 		/* error printed by mcx_alloc_tdomain */
2810 		goto teardown;
2811 	}
2812 
2813 	/*
2814 	 * PRM makes no mention of msi interrupts, just legacy and msi-x.
2815 	 * mellanox support tells me legacy interrupts are not supported,
2816 	 * so we're stuck with just msi-x.
2817 	 */
2818 	if (pci_intr_map_msix(pa, 0, &sc->sc_ih) != 0) {
2819 		printf(": unable to map interrupt\n");
2820 		goto teardown;
2821 	}
2822 	intrstr = pci_intr_string(sc->sc_pc, sc->sc_ih);
2823 	sc->sc_ihc = pci_intr_establish(sc->sc_pc, sc->sc_ih,
2824 	    IPL_NET | IPL_MPSAFE, mcx_admin_intr, sc, DEVNAME(sc));
2825 	if (sc->sc_ihc == NULL) {
2826 		printf(": unable to establish interrupt");
2827 		if (intrstr != NULL)
2828 			printf(" at %s", intrstr);
2829 		printf("\n");
2830 		goto teardown;
2831 	}
2832 
2833 	if (mcx_create_eq(sc, &sc->sc_admin_eq, sc->sc_uar,
2834 	    (1ull << MCX_EVENT_TYPE_INTERNAL_ERROR) |
2835 	    (1ull << MCX_EVENT_TYPE_PORT_CHANGE) |
2836 	    (1ull << MCX_EVENT_TYPE_CMD_COMPLETION) |
2837 	    (1ull << MCX_EVENT_TYPE_PAGE_REQUEST), 0) != 0) {
2838 		/* error printed by mcx_create_eq */
2839 		goto teardown;
2840 	}
2841 
2842 	if (mcx_query_nic_vport_context(sc) != 0) {
2843 		/* error printed by mcx_query_nic_vport_context */
2844 		goto teardown;
2845 	}
2846 
2847 	if (mcx_query_special_contexts(sc) != 0) {
2848 		/* error printed by mcx_query_special_contexts */
2849 		goto teardown;
2850 	}
2851 
2852 	if (mcx_set_port_mtu(sc, MCX_HARDMTU) != 0) {
2853 		/* error printed by mcx_set_port_mtu */
2854 		goto teardown;
2855 	}
2856 
2857 	printf(", %s, address %s\n", intrstr,
2858 	    ether_sprintf(sc->sc_ac.ac_enaddr));
2859 
2860 	msix = pci_intr_msix_count(pa->pa_pc, pa->pa_tag);
2861 	sc->sc_nqueues = 1;
2862 
2863 	strlcpy(ifp->if_xname, DEVNAME(sc), IFNAMSIZ);
2864 	ifp->if_softc = sc;
2865 	ifp->if_flags = IFF_BROADCAST | IFF_MULTICAST | IFF_SIMPLEX;
2866 	ifp->if_xflags = IFXF_MPSAFE;
2867 	ifp->if_ioctl = mcx_ioctl;
2868 	ifp->if_qstart = mcx_start;
2869 	ifp->if_watchdog = mcx_watchdog;
2870 	ifp->if_hardmtu = sc->sc_hardmtu;
2871 	ifp->if_capabilities = IFCAP_VLAN_MTU | IFCAP_CSUM_IPv4 |
2872 	    IFCAP_CSUM_UDPv4 | IFCAP_CSUM_UDPv6 | IFCAP_CSUM_TCPv4 |
2873 	    IFCAP_CSUM_TCPv6;
2874 #if NVLAN > 0
2875 	ifp->if_capabilities |= IFCAP_VLAN_HWTAGGING;
2876 #endif
2877 	ifq_set_maxlen(&ifp->if_snd, 1024);
2878 
2879 	ifmedia_init(&sc->sc_media, IFM_IMASK, mcx_media_change,
2880 	    mcx_media_status);
2881 	mcx_media_add_types(sc);
2882 	ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL);
2883 	ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO);
2884 
2885 	if_attach(ifp);
2886 	ether_ifattach(ifp);
2887 
2888 	if_attach_iqueues(ifp, sc->sc_nqueues);
2889 	if_attach_queues(ifp, sc->sc_nqueues);
2890 	for (i = 0; i < sc->sc_nqueues; i++) {
2891 		struct ifiqueue *ifiq = ifp->if_iqs[i];
2892 		struct ifqueue *ifq = ifp->if_ifqs[i];
2893 		struct mcx_queues *q = &sc->sc_queues[i];
2894 		struct mcx_rx *rx = &q->q_rx;
2895 		struct mcx_tx *tx = &q->q_tx;
2896 		pci_intr_handle_t ih;
2897 		int vec;
2898 
2899 		vec = i + 1;
2900 		q->q_sc = sc;
2901 		q->q_index = i;
2902 
2903 		if (mcx_alloc_uar(sc, &q->q_uar) != 0) {
2904 			printf("%s: unable to alloc uar %d\n",
2905 			    DEVNAME(sc), i);
2906 			goto teardown;
2907 		}
2908 
2909 		if (mcx_create_eq(sc, &q->q_eq, q->q_uar, 0, vec) != 0) {
2910 			printf("%s: unable to create event queue %d\n",
2911 			    DEVNAME(sc), i);
2912 			goto teardown;
2913 		}
2914 
2915 		rx->rx_softc = sc;
2916 		rx->rx_ifiq = ifiq;
2917 		timeout_set(&rx->rx_refill, mcx_refill, rx);
2918 		ifiq->ifiq_softc = rx;
2919 
2920 		tx->tx_softc = sc;
2921 		tx->tx_ifq = ifq;
2922 		ifq->ifq_softc = tx;
2923 
2924 		if (pci_intr_map_msix(pa, vec, &ih) != 0) {
2925 			printf("%s: unable to map queue interrupt %d\n",
2926 			    DEVNAME(sc), i);
2927 			goto teardown;
2928 		}
2929 		snprintf(q->q_name, sizeof(q->q_name), "%s:%d",
2930 		    DEVNAME(sc), i);
2931 		q->q_ihc = pci_intr_establish(sc->sc_pc, ih,
2932 		    IPL_NET | IPL_MPSAFE, mcx_cq_intr, q, q->q_name);
2933 	}
2934 
2935 	timeout_set(&sc->sc_calibrate, mcx_calibrate, sc);
2936 
2937 	task_set(&sc->sc_port_change, mcx_port_change, sc);
2938 	mcx_port_change(sc);
2939 
2940 	sc->sc_mac_flow_table_id = -1;
2941 	sc->sc_rss_flow_table_id = -1;
2942 	sc->sc_rqt = -1;
2943 	for (i = 0; i < MCX_NUM_FLOW_GROUPS; i++) {
2944 		struct mcx_flow_group *mfg = &sc->sc_flow_group[i];
2945 		mfg->g_id = -1;
2946 		mfg->g_table = -1;
2947 		mfg->g_size = 0;
2948 		mfg->g_start = 0;
2949 	}
2950 	sc->sc_extra_mcast = 0;
2951 	memset(sc->sc_mcast_flows, 0, sizeof(sc->sc_mcast_flows));
2952 
2953 #if NKSTAT > 0
2954 	mcx_kstat_attach(sc);
2955 #endif
2956 	mcx_timecounter_attach(sc);
2957 	return;
2958 
2959 teardown:
2960 	mcx_teardown_hca(sc, htobe16(MCX_CMD_TEARDOWN_HCA_GRACEFUL));
2961 	/* error printed by mcx_teardown_hca, and we're already unwinding */
2962 cqfree:
2963 	mcx_wr(sc, MCX_CMDQ_ADDR_HI, MCX_DMA_DVA(&sc->sc_cmdq_mem) >> 32);
2964 	mcx_bar(sc, MCX_CMDQ_ADDR_HI, sizeof(uint64_t),
2965 	    BUS_SPACE_BARRIER_WRITE);
2966 	mcx_wr(sc, MCX_CMDQ_ADDR_LO, MCX_DMA_DVA(&sc->sc_cmdq_mem) |
2967 	    MCX_CMDQ_INTERFACE_DISABLED);
2968 	mcx_bar(sc, MCX_CMDQ_ADDR_LO, sizeof(uint64_t),
2969 	    BUS_SPACE_BARRIER_WRITE);
2970 
2971 	mcx_wr(sc, MCX_CMDQ_ADDR_HI, 0);
2972 	mcx_bar(sc, MCX_CMDQ_ADDR_HI, sizeof(uint64_t),
2973 	    BUS_SPACE_BARRIER_WRITE);
2974 	mcx_wr(sc, MCX_CMDQ_ADDR_LO, MCX_CMDQ_INTERFACE_DISABLED);
2975 
2976 	mcx_dmamem_free(sc, &sc->sc_cmdq_mem);
2977 dbfree:
2978 	mcx_dmamem_free(sc, &sc->sc_doorbell_mem);
2979 unmap:
2980 	bus_space_unmap(sc->sc_memt, sc->sc_memh, sc->sc_mems);
2981 	sc->sc_mems = 0;
2982 }
2983 
2984 static int
2985 mcx_version(struct mcx_softc *sc)
2986 {
2987 	uint32_t fw0, fw1;
2988 	uint16_t cmdif;
2989 
2990 	fw0 = mcx_rd(sc, MCX_FW_VER);
2991 	fw1 = mcx_rd(sc, MCX_CMDIF_FW_SUBVER);
2992 
2993 	printf(": FW %u.%u.%04u", MCX_FW_VER_MAJOR(fw0),
2994 	    MCX_FW_VER_MINOR(fw0), MCX_FW_VER_SUBMINOR(fw1));
2995 
2996 	cmdif = MCX_CMDIF(fw1);
2997 	if (cmdif != MCX_CMD_IF_SUPPORTED) {
2998 		printf(", unsupported command interface %u\n", cmdif);
2999 		return (-1);
3000 	}
3001 
3002 	return (0);
3003 }
3004 
3005 static int
3006 mcx_init_wait(struct mcx_softc *sc)
3007 {
3008 	unsigned int i;
3009 	uint32_t r;
3010 
3011 	for (i = 0; i < 2000; i++) {
3012 		r = mcx_rd(sc, MCX_STATE);
3013 		if ((r & MCX_STATE_MASK) == MCX_STATE_READY)
3014 			return (0);
3015 
3016 		delay(1000);
3017 		mcx_bar(sc, MCX_STATE, sizeof(uint32_t),
3018 		    BUS_SPACE_BARRIER_READ);
3019 	}
3020 
3021 	return (-1);
3022 }
3023 
3024 static uint8_t
3025 mcx_cmdq_poll(struct mcx_softc *sc, struct mcx_cmdq_entry *cqe,
3026     unsigned int msec)
3027 {
3028 	unsigned int i;
3029 
3030 	for (i = 0; i < msec; i++) {
3031 		bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_cmdq_mem),
3032 		    0, MCX_DMA_LEN(&sc->sc_cmdq_mem), BUS_DMASYNC_POSTRW);
3033 
3034 		if ((cqe->cq_status & MCX_CQ_STATUS_OWN_MASK) ==
3035 		    MCX_CQ_STATUS_OWN_SW)
3036 			return (0);
3037 
3038 		delay(1000);
3039 	}
3040 
3041 	return (ETIMEDOUT);
3042 }
3043 
3044 static uint32_t
3045 mcx_mix_u64(uint32_t xor, uint64_t u64)
3046 {
3047 	xor ^= u64 >> 32;
3048 	xor ^= u64;
3049 
3050 	return (xor);
3051 }
3052 
3053 static uint32_t
3054 mcx_mix_u32(uint32_t xor, uint32_t u32)
3055 {
3056 	xor ^= u32;
3057 
3058 	return (xor);
3059 }
3060 
3061 static uint32_t
3062 mcx_mix_u8(uint32_t xor, uint8_t u8)
3063 {
3064 	xor ^= u8;
3065 
3066 	return (xor);
3067 }
3068 
3069 static uint8_t
3070 mcx_mix_done(uint32_t xor)
3071 {
3072 	xor ^= xor >> 16;
3073 	xor ^= xor >> 8;
3074 
3075 	return (xor);
3076 }
3077 
3078 static uint8_t
3079 mcx_xor(const void *buf, size_t len)
3080 {
3081 	const uint32_t *dwords = buf;
3082 	uint32_t xor = 0xff;
3083 	size_t i;
3084 
3085 	len /= sizeof(*dwords);
3086 
3087 	for (i = 0; i < len; i++)
3088 		xor ^= dwords[i];
3089 
3090 	return (mcx_mix_done(xor));
3091 }
3092 
3093 static uint8_t
3094 mcx_cmdq_token(struct mcx_softc *sc)
3095 {
3096 	uint8_t token;
3097 
3098 	do {
3099 		token = ++sc->sc_cmdq_token;
3100 	} while (token == 0);
3101 
3102 	return (token);
3103 }
3104 
3105 static void
3106 mcx_cmdq_init(struct mcx_softc *sc, struct mcx_cmdq_entry *cqe,
3107     uint32_t ilen, uint32_t olen, uint8_t token)
3108 {
3109 	memset(cqe, 0, sc->sc_cmdq_size);
3110 
3111 	cqe->cq_type = MCX_CMDQ_TYPE_PCIE;
3112 	htobem32(&cqe->cq_input_length, ilen);
3113 	htobem32(&cqe->cq_output_length, olen);
3114 	cqe->cq_token = token;
3115 	cqe->cq_status = MCX_CQ_STATUS_OWN_HW;
3116 }
3117 
3118 static void
3119 mcx_cmdq_sign(struct mcx_cmdq_entry *cqe)
3120 {
3121 	cqe->cq_signature = ~mcx_xor(cqe, sizeof(*cqe));
3122 }
3123 
3124 static int
3125 mcx_cmdq_verify(const struct mcx_cmdq_entry *cqe)
3126 {
3127 	/* return (mcx_xor(cqe, sizeof(*cqe)) ? -1 :  0); */
3128 	return (0);
3129 }
3130 
3131 static void *
3132 mcx_cmdq_in(struct mcx_cmdq_entry *cqe)
3133 {
3134 	return (&cqe->cq_input_data);
3135 }
3136 
3137 static void *
3138 mcx_cmdq_out(struct mcx_cmdq_entry *cqe)
3139 {
3140 	return (&cqe->cq_output_data);
3141 }
3142 
3143 static void
3144 mcx_cmdq_post(struct mcx_softc *sc, struct mcx_cmdq_entry *cqe,
3145     unsigned int slot)
3146 {
3147 	mcx_cmdq_sign(cqe);
3148 
3149 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_cmdq_mem),
3150 	    0, MCX_DMA_LEN(&sc->sc_cmdq_mem), BUS_DMASYNC_PRERW);
3151 
3152 	mcx_wr(sc, MCX_CMDQ_DOORBELL, 1U << slot);
3153 	mcx_bar(sc, MCX_CMDQ_DOORBELL, sizeof(uint32_t),
3154 	    BUS_SPACE_BARRIER_WRITE);
3155 }
3156 
3157 static int
3158 mcx_enable_hca(struct mcx_softc *sc)
3159 {
3160 	struct mcx_cmdq_entry *cqe;
3161 	struct mcx_cmd_enable_hca_in *in;
3162 	struct mcx_cmd_enable_hca_out *out;
3163 	int error;
3164 	uint8_t status;
3165 
3166 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
3167 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
3168 
3169 	in = mcx_cmdq_in(cqe);
3170 	in->cmd_opcode = htobe16(MCX_CMD_ENABLE_HCA);
3171 	in->cmd_op_mod = htobe16(0);
3172 	in->cmd_function_id = htobe16(0);
3173 
3174 	mcx_cmdq_post(sc, cqe, 0);
3175 
3176 	error = mcx_cmdq_poll(sc, cqe, 1000);
3177 	if (error != 0) {
3178 		printf(", hca enable timeout\n");
3179 		return (-1);
3180 	}
3181 	if (mcx_cmdq_verify(cqe) != 0) {
3182 		printf(", hca enable command corrupt\n");
3183 		return (-1);
3184 	}
3185 
3186 	status = cqe->cq_output_data[0];
3187 	if (status != MCX_CQ_STATUS_OK) {
3188 		printf(", hca enable failed (%x)\n", status);
3189 		return (-1);
3190 	}
3191 
3192 	return (0);
3193 }
3194 
3195 static int
3196 mcx_teardown_hca(struct mcx_softc *sc, uint16_t profile)
3197 {
3198 	struct mcx_cmdq_entry *cqe;
3199 	struct mcx_cmd_teardown_hca_in *in;
3200 	struct mcx_cmd_teardown_hca_out *out;
3201 	int error;
3202 	uint8_t status;
3203 
3204 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
3205 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
3206 
3207 	in = mcx_cmdq_in(cqe);
3208 	in->cmd_opcode = htobe16(MCX_CMD_TEARDOWN_HCA);
3209 	in->cmd_op_mod = htobe16(0);
3210 	in->cmd_profile = profile;
3211 
3212 	mcx_cmdq_post(sc, cqe, 0);
3213 
3214 	error = mcx_cmdq_poll(sc, cqe, 1000);
3215 	if (error != 0) {
3216 		printf(", hca teardown timeout\n");
3217 		return (-1);
3218 	}
3219 	if (mcx_cmdq_verify(cqe) != 0) {
3220 		printf(", hca teardown command corrupt\n");
3221 		return (-1);
3222 	}
3223 
3224 	status = cqe->cq_output_data[0];
3225 	if (status != MCX_CQ_STATUS_OK) {
3226 		printf(", hca teardown failed (%x)\n", status);
3227 		return (-1);
3228 	}
3229 
3230 	return (0);
3231 }
3232 
3233 static int
3234 mcx_cmdq_mboxes_alloc(struct mcx_softc *sc, struct mcx_dmamem *mxm,
3235     unsigned int nmb, uint64_t *ptr, uint8_t token)
3236 {
3237 	caddr_t kva;
3238 	uint64_t dva;
3239 	int i;
3240 	int error;
3241 
3242 	error = mcx_dmamem_alloc(sc, mxm,
3243 	    nmb * MCX_CMDQ_MAILBOX_SIZE, MCX_CMDQ_MAILBOX_ALIGN);
3244 	if (error != 0)
3245 		return (error);
3246 
3247 	mcx_dmamem_zero(mxm);
3248 
3249 	dva = MCX_DMA_DVA(mxm);
3250 	kva = MCX_DMA_KVA(mxm);
3251 	for (i = 0; i < nmb; i++) {
3252 		struct mcx_cmdq_mailbox *mbox = (struct mcx_cmdq_mailbox *)kva;
3253 
3254 		/* patch the cqe or mbox pointing at this one */
3255 		htobem64(ptr, dva);
3256 
3257 		/* fill in this mbox */
3258 		htobem32(&mbox->mb_block_number, i);
3259 		mbox->mb_token = token;
3260 
3261 		/* move to the next one */
3262 		ptr = &mbox->mb_next_ptr;
3263 
3264 		dva += MCX_CMDQ_MAILBOX_SIZE;
3265 		kva += MCX_CMDQ_MAILBOX_SIZE;
3266 	}
3267 
3268 	return (0);
3269 }
3270 
3271 static uint32_t
3272 mcx_cmdq_mbox_ctrl_sig(const struct mcx_cmdq_mailbox *mb)
3273 {
3274 	uint32_t xor = 0xff;
3275 
3276 	/* only 3 fields get set, so mix them directly */
3277 	xor = mcx_mix_u64(xor, mb->mb_next_ptr);
3278 	xor = mcx_mix_u32(xor, mb->mb_block_number);
3279 	xor = mcx_mix_u8(xor, mb->mb_token);
3280 
3281 	return (mcx_mix_done(xor));
3282 }
3283 
3284 static void
3285 mcx_cmdq_mboxes_sign(struct mcx_dmamem *mxm, unsigned int nmb)
3286 {
3287 	caddr_t kva;
3288 	int i;
3289 
3290 	kva = MCX_DMA_KVA(mxm);
3291 
3292 	for (i = 0; i < nmb; i++) {
3293 		struct mcx_cmdq_mailbox *mb = (struct mcx_cmdq_mailbox *)kva;
3294 		uint8_t sig = mcx_cmdq_mbox_ctrl_sig(mb);
3295 		mb->mb_ctrl_signature = sig;
3296 		mb->mb_signature = sig ^
3297 		    mcx_xor(mb->mb_data, sizeof(mb->mb_data));
3298 
3299 		kva += MCX_CMDQ_MAILBOX_SIZE;
3300 	}
3301 }
3302 
3303 static void
3304 mcx_cmdq_mboxes_sync(struct mcx_softc *sc, struct mcx_dmamem *mxm, int ops)
3305 {
3306 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(mxm),
3307 	    0, MCX_DMA_LEN(mxm), ops);
3308 }
3309 
3310 static struct mcx_cmdq_mailbox *
3311 mcx_cq_mbox(struct mcx_dmamem *mxm, unsigned int i)
3312 {
3313 	caddr_t kva;
3314 
3315 	kva = MCX_DMA_KVA(mxm);
3316 	kva += i * MCX_CMDQ_MAILBOX_SIZE;
3317 
3318 	return ((struct mcx_cmdq_mailbox *)kva);
3319 }
3320 
3321 static inline void *
3322 mcx_cq_mbox_data(struct mcx_cmdq_mailbox *mb)
3323 {
3324 	return (&mb->mb_data);
3325 }
3326 
3327 static void
3328 mcx_cmdq_mboxes_copyin(struct mcx_dmamem *mxm, unsigned int nmb,
3329     void *b, size_t len)
3330 {
3331 	caddr_t buf = b;
3332 	struct mcx_cmdq_mailbox *mb;
3333 	int i;
3334 
3335 	mb = (struct mcx_cmdq_mailbox *)MCX_DMA_KVA(mxm);
3336 	for (i = 0; i < nmb; i++) {
3337 
3338 		memcpy(mb->mb_data, buf, min(sizeof(mb->mb_data), len));
3339 
3340 		if (sizeof(mb->mb_data) >= len)
3341 			break;
3342 
3343 		buf += sizeof(mb->mb_data);
3344 		len -= sizeof(mb->mb_data);
3345 		mb++;
3346 	}
3347 }
3348 
3349 static void
3350 mcx_cmdq_mboxes_pas(struct mcx_dmamem *mxm, int offset, int npages,
3351     struct mcx_dmamem *buf)
3352 {
3353 	uint64_t *pas;
3354 	int mbox, mbox_pages, i;
3355 
3356 	mbox = offset / MCX_CMDQ_MAILBOX_DATASIZE;
3357 	offset %= MCX_CMDQ_MAILBOX_DATASIZE;
3358 
3359 	pas = mcx_cq_mbox_data(mcx_cq_mbox(mxm, mbox));
3360 	pas += (offset / sizeof(*pas));
3361 	mbox_pages = (MCX_CMDQ_MAILBOX_DATASIZE - offset) / sizeof(*pas);
3362 	for (i = 0; i < npages; i++) {
3363 		if (i == mbox_pages) {
3364 			mbox++;
3365 			pas = mcx_cq_mbox_data(mcx_cq_mbox(mxm, mbox));
3366 			mbox_pages += MCX_CMDQ_MAILBOX_DATASIZE / sizeof(*pas);
3367 		}
3368 		*pas = htobe64(MCX_DMA_DVA(buf) + (i * MCX_PAGE_SIZE));
3369 		pas++;
3370 	}
3371 }
3372 
3373 static void
3374 mcx_cmdq_mboxes_copyout(struct mcx_dmamem *mxm, int nmb, void *b, size_t len)
3375 {
3376 	caddr_t buf = b;
3377 	struct mcx_cmdq_mailbox *mb;
3378 	int i;
3379 
3380 	mb = (struct mcx_cmdq_mailbox *)MCX_DMA_KVA(mxm);
3381 	for (i = 0; i < nmb; i++) {
3382 		memcpy(buf, mb->mb_data, min(sizeof(mb->mb_data), len));
3383 
3384 		if (sizeof(mb->mb_data) >= len)
3385 			break;
3386 
3387 		buf += sizeof(mb->mb_data);
3388 		len -= sizeof(mb->mb_data);
3389 		mb++;
3390 	}
3391 }
3392 
3393 static void
3394 mcx_cq_mboxes_free(struct mcx_softc *sc, struct mcx_dmamem *mxm)
3395 {
3396 	mcx_dmamem_free(sc, mxm);
3397 }
3398 
3399 #if 0
3400 static void
3401 mcx_cmdq_dump(const struct mcx_cmdq_entry *cqe)
3402 {
3403 	unsigned int i;
3404 
3405 	printf(" type %02x, ilen %u, iptr %016llx", cqe->cq_type,
3406 	    bemtoh32(&cqe->cq_input_length), bemtoh64(&cqe->cq_input_ptr));
3407 
3408 	printf(", idata ");
3409 	for (i = 0; i < sizeof(cqe->cq_input_data); i++)
3410 		printf("%02x", cqe->cq_input_data[i]);
3411 
3412 	printf(", odata ");
3413 	for (i = 0; i < sizeof(cqe->cq_output_data); i++)
3414 		printf("%02x", cqe->cq_output_data[i]);
3415 
3416 	printf(", optr %016llx, olen %u, token %02x, sig %02x, status %02x",
3417 	    bemtoh64(&cqe->cq_output_ptr), bemtoh32(&cqe->cq_output_length),
3418 	    cqe->cq_token, cqe->cq_signature, cqe->cq_status);
3419 }
3420 
3421 static void
3422 mcx_cmdq_mbox_dump(struct mcx_dmamem *mboxes, int num)
3423 {
3424 	int i, j;
3425 	uint8_t *d;
3426 
3427 	for (i = 0; i < num; i++) {
3428 		struct mcx_cmdq_mailbox *mbox;
3429 		mbox = mcx_cq_mbox(mboxes, i);
3430 
3431 		d = mcx_cq_mbox_data(mbox);
3432 		for (j = 0; j < MCX_CMDQ_MAILBOX_DATASIZE; j++) {
3433 			if (j != 0 && (j % 16 == 0))
3434 				printf("\n");
3435 			printf("%.2x ", d[j]);
3436 		}
3437 	}
3438 }
3439 #endif
3440 
3441 static int
3442 mcx_access_hca_reg(struct mcx_softc *sc, uint16_t reg, int op, void *data,
3443     int len)
3444 {
3445 	struct mcx_dmamem mxm;
3446 	struct mcx_cmdq_entry *cqe;
3447 	struct mcx_cmd_access_reg_in *in;
3448 	struct mcx_cmd_access_reg_out *out;
3449 	uint8_t token = mcx_cmdq_token(sc);
3450 	int error, nmb;
3451 
3452 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
3453 	mcx_cmdq_init(sc, cqe, sizeof(*in) + len, sizeof(*out) + len,
3454 	    token);
3455 
3456 	in = mcx_cmdq_in(cqe);
3457 	in->cmd_opcode = htobe16(MCX_CMD_ACCESS_REG);
3458 	in->cmd_op_mod = htobe16(op);
3459 	in->cmd_register_id = htobe16(reg);
3460 
3461 	nmb = howmany(len, MCX_CMDQ_MAILBOX_DATASIZE);
3462 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, nmb,
3463 	    &cqe->cq_output_ptr, token) != 0) {
3464 		printf(", unable to allocate access reg mailboxen\n");
3465 		return (-1);
3466 	}
3467 	cqe->cq_input_ptr = cqe->cq_output_ptr;
3468 	mcx_cmdq_mboxes_copyin(&mxm, nmb, data, len);
3469 	mcx_cmdq_mboxes_sign(&mxm, nmb);
3470 	mcx_cmdq_mboxes_sync(sc, &mxm, BUS_DMASYNC_PRERW);
3471 
3472 	mcx_cmdq_post(sc, cqe, 0);
3473 	error = mcx_cmdq_poll(sc, cqe, 1000);
3474 	mcx_cmdq_mboxes_sync(sc, &mxm, BUS_DMASYNC_POSTRW);
3475 
3476 	if (error != 0) {
3477 		printf("%s: access reg (%s %x) timeout\n", DEVNAME(sc),
3478 		    (op == MCX_REG_OP_WRITE ? "write" : "read"), reg);
3479 		goto free;
3480 	}
3481 	error = mcx_cmdq_verify(cqe);
3482 	if (error != 0) {
3483 		printf("%s: access reg (%s %x) reply corrupt\n",
3484 		    (op == MCX_REG_OP_WRITE ? "write" : "read"), DEVNAME(sc),
3485 		    reg);
3486 		goto free;
3487 	}
3488 
3489 	out = mcx_cmdq_out(cqe);
3490 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
3491 		printf("%s: access reg (%s %x) failed (%x, %.6x)\n",
3492 		    DEVNAME(sc), (op == MCX_REG_OP_WRITE ? "write" : "read"),
3493 		    reg, out->cmd_status, betoh32(out->cmd_syndrome));
3494 		error = -1;
3495 		goto free;
3496 	}
3497 
3498 	mcx_cmdq_mboxes_copyout(&mxm, nmb, data, len);
3499 free:
3500 	mcx_dmamem_free(sc, &mxm);
3501 
3502 	return (error);
3503 }
3504 
3505 static int
3506 mcx_set_issi(struct mcx_softc *sc, struct mcx_cmdq_entry *cqe,
3507     unsigned int slot)
3508 {
3509 	struct mcx_cmd_set_issi_in *in;
3510 	struct mcx_cmd_set_issi_out *out;
3511 	uint8_t status;
3512 
3513 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
3514 
3515 	in = mcx_cmdq_in(cqe);
3516 	in->cmd_opcode = htobe16(MCX_CMD_SET_ISSI);
3517 	in->cmd_op_mod = htobe16(0);
3518 	in->cmd_current_issi = htobe16(MCX_ISSI);
3519 
3520 	mcx_cmdq_post(sc, cqe, slot);
3521 	if (mcx_cmdq_poll(sc, cqe, 1000) != 0)
3522 		return (-1);
3523 	if (mcx_cmdq_verify(cqe) != 0)
3524 		return (-1);
3525 
3526 	status = cqe->cq_output_data[0];
3527 	if (status != MCX_CQ_STATUS_OK)
3528 		return (-1);
3529 
3530 	return (0);
3531 }
3532 
3533 static int
3534 mcx_issi(struct mcx_softc *sc)
3535 {
3536 	struct mcx_dmamem mxm;
3537 	struct mcx_cmdq_entry *cqe;
3538 	struct mcx_cmd_query_issi_in *in;
3539 	struct mcx_cmd_query_issi_il_out *out;
3540 	struct mcx_cmd_query_issi_mb_out *mb;
3541 	uint8_t token = mcx_cmdq_token(sc);
3542 	uint8_t status;
3543 	int error;
3544 
3545 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
3546 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out) + sizeof(*mb), token);
3547 
3548 	in = mcx_cmdq_in(cqe);
3549 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_ISSI);
3550 	in->cmd_op_mod = htobe16(0);
3551 
3552 	CTASSERT(sizeof(*mb) <= MCX_CMDQ_MAILBOX_DATASIZE);
3553 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
3554 	    &cqe->cq_output_ptr, token) != 0) {
3555 		printf(", unable to allocate query issi mailbox\n");
3556 		return (-1);
3557 	}
3558 	mcx_cmdq_mboxes_sign(&mxm, 1);
3559 
3560 	mcx_cmdq_post(sc, cqe, 0);
3561 	error = mcx_cmdq_poll(sc, cqe, 1000);
3562 	if (error != 0) {
3563 		printf(", query issi timeout\n");
3564 		goto free;
3565 	}
3566 	error = mcx_cmdq_verify(cqe);
3567 	if (error != 0) {
3568 		printf(", query issi reply corrupt\n");
3569 		goto free;
3570 	}
3571 
3572 	status = cqe->cq_output_data[0];
3573 	switch (status) {
3574 	case MCX_CQ_STATUS_OK:
3575 		break;
3576 	case MCX_CQ_STATUS_BAD_OPCODE:
3577 		/* use ISSI 0 */
3578 		goto free;
3579 	default:
3580 		printf(", query issi failed (%x)\n", status);
3581 		error = -1;
3582 		goto free;
3583 	}
3584 
3585 	out = mcx_cmdq_out(cqe);
3586 	if (out->cmd_current_issi == htobe16(MCX_ISSI)) {
3587 		/* use ISSI 1 */
3588 		goto free;
3589 	}
3590 
3591 	/* don't need to read cqe anymore, can be used for SET ISSI */
3592 
3593 	mb = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
3594 	CTASSERT(MCX_ISSI < NBBY);
3595 	 /* XXX math is hard */
3596 	if (!ISSET(mb->cmd_supported_issi[79], 1 << MCX_ISSI)) {
3597 		/* use ISSI 0 */
3598 		goto free;
3599 	}
3600 
3601 	if (mcx_set_issi(sc, cqe, 0) != 0) {
3602 		/* ignore the error, just use ISSI 0 */
3603 	} else {
3604 		/* use ISSI 1 */
3605 	}
3606 
3607 free:
3608 	mcx_cq_mboxes_free(sc, &mxm);
3609 	return (error);
3610 }
3611 
3612 static int
3613 mcx_query_pages(struct mcx_softc *sc, uint16_t type,
3614     int32_t *npages, uint16_t *func_id)
3615 {
3616 	struct mcx_cmdq_entry *cqe;
3617 	struct mcx_cmd_query_pages_in *in;
3618 	struct mcx_cmd_query_pages_out *out;
3619 
3620 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
3621 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
3622 
3623 	in = mcx_cmdq_in(cqe);
3624 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_PAGES);
3625 	in->cmd_op_mod = type;
3626 
3627 	mcx_cmdq_post(sc, cqe, 0);
3628 	if (mcx_cmdq_poll(sc, cqe, 1000) != 0) {
3629 		printf(", query pages timeout\n");
3630 		return (-1);
3631 	}
3632 	if (mcx_cmdq_verify(cqe) != 0) {
3633 		printf(", query pages reply corrupt\n");
3634 		return (-1);
3635 	}
3636 
3637 	out = mcx_cmdq_out(cqe);
3638 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
3639 		printf(", query pages failed (%x)\n", out->cmd_status);
3640 		return (-1);
3641 	}
3642 
3643 	*func_id = out->cmd_func_id;
3644 	*npages = bemtoh32(&out->cmd_num_pages);
3645 
3646 	return (0);
3647 }
3648 
3649 struct bus_dma_iter {
3650 	bus_dmamap_t		i_map;
3651 	bus_size_t		i_offset;
3652 	unsigned int		i_index;
3653 };
3654 
3655 static void
3656 bus_dma_iter_init(struct bus_dma_iter *i, bus_dmamap_t map)
3657 {
3658 	i->i_map = map;
3659 	i->i_offset = 0;
3660 	i->i_index = 0;
3661 }
3662 
3663 static bus_addr_t
3664 bus_dma_iter_addr(struct bus_dma_iter *i)
3665 {
3666 	return (i->i_map->dm_segs[i->i_index].ds_addr + i->i_offset);
3667 }
3668 
3669 static void
3670 bus_dma_iter_add(struct bus_dma_iter *i, bus_size_t size)
3671 {
3672 	bus_dma_segment_t *seg = i->i_map->dm_segs + i->i_index;
3673 	bus_size_t diff;
3674 
3675 	do {
3676 		diff = seg->ds_len - i->i_offset;
3677 		if (size < diff)
3678 			break;
3679 
3680 		size -= diff;
3681 
3682 		seg++;
3683 
3684 		i->i_offset = 0;
3685 		i->i_index++;
3686 	} while (size > 0);
3687 
3688 	i->i_offset += size;
3689 }
3690 
3691 static int
3692 mcx_add_pages(struct mcx_softc *sc, struct mcx_hwmem *mhm, uint16_t func_id)
3693 {
3694 	struct mcx_dmamem mxm;
3695 	struct mcx_cmdq_entry *cqe;
3696 	struct mcx_cmd_manage_pages_in *in;
3697 	struct mcx_cmd_manage_pages_out *out;
3698 	unsigned int paslen, nmb, i, j, npages;
3699 	struct bus_dma_iter iter;
3700 	uint64_t *pas;
3701 	uint8_t status;
3702 	uint8_t token = mcx_cmdq_token(sc);
3703 	int error;
3704 
3705 	npages = mhm->mhm_npages;
3706 
3707 	paslen = sizeof(*pas) * npages;
3708 	nmb = howmany(paslen, MCX_CMDQ_MAILBOX_DATASIZE);
3709 
3710 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
3711 	mcx_cmdq_init(sc, cqe, sizeof(*in) + paslen, sizeof(*out), token);
3712 
3713 	in = mcx_cmdq_in(cqe);
3714 	in->cmd_opcode = htobe16(MCX_CMD_MANAGE_PAGES);
3715 	in->cmd_op_mod = htobe16(MCX_CMD_MANAGE_PAGES_ALLOC_SUCCESS);
3716 	in->cmd_func_id = func_id;
3717 	htobem32(&in->cmd_input_num_entries, npages);
3718 
3719 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, nmb,
3720 	    &cqe->cq_input_ptr, token) != 0) {
3721 		printf(", unable to allocate manage pages mailboxen\n");
3722 		return (-1);
3723 	}
3724 
3725 	bus_dma_iter_init(&iter, mhm->mhm_map);
3726 	for (i = 0; i < nmb; i++) {
3727 		unsigned int lim;
3728 
3729 		pas = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, i));
3730 		lim = min(MCX_CMDQ_MAILBOX_DATASIZE / sizeof(*pas), npages);
3731 
3732 		for (j = 0; j < lim; j++) {
3733 			htobem64(&pas[j], bus_dma_iter_addr(&iter));
3734 			bus_dma_iter_add(&iter, MCX_PAGE_SIZE);
3735 		}
3736 
3737 		npages -= lim;
3738 	}
3739 
3740 	mcx_cmdq_mboxes_sign(&mxm, nmb);
3741 
3742 	mcx_cmdq_post(sc, cqe, 0);
3743 	error = mcx_cmdq_poll(sc, cqe, 1000);
3744 	if (error != 0) {
3745 		printf(", manage pages timeout\n");
3746 		goto free;
3747 	}
3748 	error = mcx_cmdq_verify(cqe);
3749 	if (error != 0) {
3750 		printf(", manage pages reply corrupt\n");
3751 		goto free;
3752 	}
3753 
3754 	status = cqe->cq_output_data[0];
3755 	if (status != MCX_CQ_STATUS_OK) {
3756 		printf(", manage pages failed (%x)\n", status);
3757 		error = -1;
3758 		goto free;
3759 	}
3760 
3761 free:
3762 	mcx_dmamem_free(sc, &mxm);
3763 
3764 	return (error);
3765 }
3766 
3767 static int
3768 mcx_pages(struct mcx_softc *sc, struct mcx_hwmem *mhm, uint16_t type)
3769 {
3770 	int32_t npages;
3771 	uint16_t func_id;
3772 
3773 	if (mcx_query_pages(sc, type, &npages, &func_id) != 0) {
3774 		/* error printed by mcx_query_pages */
3775 		return (-1);
3776 	}
3777 
3778 	if (npages < 1)
3779 		return (0);
3780 
3781 	if (mcx_hwmem_alloc(sc, mhm, npages) != 0) {
3782 		printf(", unable to allocate hwmem\n");
3783 		return (-1);
3784 	}
3785 
3786 	if (mcx_add_pages(sc, mhm, func_id) != 0) {
3787 		printf(", unable to add hwmem\n");
3788 		goto free;
3789 	}
3790 
3791 	return (0);
3792 
3793 free:
3794 	mcx_hwmem_free(sc, mhm);
3795 
3796 	return (-1);
3797 }
3798 
3799 static int
3800 mcx_hca_max_caps(struct mcx_softc *sc)
3801 {
3802 	struct mcx_dmamem mxm;
3803 	struct mcx_cmdq_entry *cqe;
3804 	struct mcx_cmd_query_hca_cap_in *in;
3805 	struct mcx_cmd_query_hca_cap_out *out;
3806 	struct mcx_cmdq_mailbox *mb;
3807 	struct mcx_cap_device *hca;
3808 	uint8_t status;
3809 	uint8_t token = mcx_cmdq_token(sc);
3810 	int error;
3811 
3812 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
3813 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out) + MCX_HCA_CAP_LEN,
3814 	    token);
3815 
3816 	in = mcx_cmdq_in(cqe);
3817 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_HCA_CAP);
3818 	in->cmd_op_mod = htobe16(MCX_CMD_QUERY_HCA_CAP_MAX |
3819 	    MCX_CMD_QUERY_HCA_CAP_DEVICE);
3820 
3821 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, MCX_HCA_CAP_NMAILBOXES,
3822 	    &cqe->cq_output_ptr, token) != 0) {
3823 		printf(", unable to allocate query hca caps mailboxen\n");
3824 		return (-1);
3825 	}
3826 	mcx_cmdq_mboxes_sign(&mxm, MCX_HCA_CAP_NMAILBOXES);
3827 	mcx_cmdq_mboxes_sync(sc, &mxm, BUS_DMASYNC_PRERW);
3828 
3829 	mcx_cmdq_post(sc, cqe, 0);
3830 	error = mcx_cmdq_poll(sc, cqe, 1000);
3831 	mcx_cmdq_mboxes_sync(sc, &mxm, BUS_DMASYNC_POSTRW);
3832 
3833 	if (error != 0) {
3834 		printf(", query hca caps timeout\n");
3835 		goto free;
3836 	}
3837 	error = mcx_cmdq_verify(cqe);
3838 	if (error != 0) {
3839 		printf(", query hca caps reply corrupt\n");
3840 		goto free;
3841 	}
3842 
3843 	status = cqe->cq_output_data[0];
3844 	if (status != MCX_CQ_STATUS_OK) {
3845 		printf(", query hca caps failed (%x)\n", status);
3846 		error = -1;
3847 		goto free;
3848 	}
3849 
3850 	mb = mcx_cq_mbox(&mxm, 0);
3851 	hca = mcx_cq_mbox_data(mb);
3852 
3853 	if ((hca->port_type & MCX_CAP_DEVICE_PORT_TYPE)
3854 	    != MCX_CAP_DEVICE_PORT_TYPE_ETH) {
3855 		printf(", not in ethernet mode\n");
3856 		error = -1;
3857 		goto free;
3858 	}
3859 	if (hca->log_pg_sz > PAGE_SHIFT) {
3860 		printf(", minimum system page shift %u is too large\n",
3861 		    hca->log_pg_sz);
3862 		error = -1;
3863 		goto free;
3864 	}
3865 	/*
3866 	 * blueflame register is split into two buffers, and we must alternate
3867 	 * between the two of them.
3868 	 */
3869 	sc->sc_bf_size = (1 << hca->log_bf_reg_size) / 2;
3870 	sc->sc_max_rqt_size = (1 << hca->log_max_rqt_size);
3871 
3872 	sc->sc_mhz = bemtoh32(&hca->device_frequency_mhz);
3873 	sc->sc_khz = bemtoh32(&hca->device_frequency_khz);
3874 
3875 free:
3876 	mcx_dmamem_free(sc, &mxm);
3877 
3878 	return (error);
3879 }
3880 
3881 static int
3882 mcx_hca_set_caps(struct mcx_softc *sc)
3883 {
3884 	struct mcx_dmamem mxm;
3885 	struct mcx_cmdq_entry *cqe;
3886 	struct mcx_cmd_query_hca_cap_in *in;
3887 	struct mcx_cmd_query_hca_cap_out *out;
3888 	struct mcx_cmdq_mailbox *mb;
3889 	struct mcx_cap_device *hca;
3890 	uint8_t status;
3891 	uint8_t token = mcx_cmdq_token(sc);
3892 	int error;
3893 
3894 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
3895 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out) + MCX_HCA_CAP_LEN,
3896 	    token);
3897 
3898 	in = mcx_cmdq_in(cqe);
3899 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_HCA_CAP);
3900 	in->cmd_op_mod = htobe16(MCX_CMD_QUERY_HCA_CAP_CURRENT |
3901 	    MCX_CMD_QUERY_HCA_CAP_DEVICE);
3902 
3903 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, MCX_HCA_CAP_NMAILBOXES,
3904 	    &cqe->cq_output_ptr, token) != 0) {
3905 		printf(", unable to allocate manage pages mailboxen\n");
3906 		return (-1);
3907 	}
3908 	mcx_cmdq_mboxes_sign(&mxm, MCX_HCA_CAP_NMAILBOXES);
3909 	mcx_cmdq_mboxes_sync(sc, &mxm, BUS_DMASYNC_PRERW);
3910 
3911 	mcx_cmdq_post(sc, cqe, 0);
3912 	error = mcx_cmdq_poll(sc, cqe, 1000);
3913 	mcx_cmdq_mboxes_sync(sc, &mxm, BUS_DMASYNC_POSTRW);
3914 
3915 	if (error != 0) {
3916 		printf(", query hca caps timeout\n");
3917 		goto free;
3918 	}
3919 	error = mcx_cmdq_verify(cqe);
3920 	if (error != 0) {
3921 		printf(", query hca caps reply corrupt\n");
3922 		goto free;
3923 	}
3924 
3925 	status = cqe->cq_output_data[0];
3926 	if (status != MCX_CQ_STATUS_OK) {
3927 		printf(", query hca caps failed (%x)\n", status);
3928 		error = -1;
3929 		goto free;
3930 	}
3931 
3932 	mb = mcx_cq_mbox(&mxm, 0);
3933 	hca = mcx_cq_mbox_data(mb);
3934 
3935 	hca->log_pg_sz = PAGE_SHIFT;
3936 
3937 free:
3938 	mcx_dmamem_free(sc, &mxm);
3939 
3940 	return (error);
3941 }
3942 
3943 
3944 static int
3945 mcx_init_hca(struct mcx_softc *sc)
3946 {
3947 	struct mcx_cmdq_entry *cqe;
3948 	struct mcx_cmd_init_hca_in *in;
3949 	struct mcx_cmd_init_hca_out *out;
3950 	int error;
3951 	uint8_t status;
3952 
3953 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
3954 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
3955 
3956 	in = mcx_cmdq_in(cqe);
3957 	in->cmd_opcode = htobe16(MCX_CMD_INIT_HCA);
3958 	in->cmd_op_mod = htobe16(0);
3959 
3960 	mcx_cmdq_post(sc, cqe, 0);
3961 
3962 	error = mcx_cmdq_poll(sc, cqe, 1000);
3963 	if (error != 0) {
3964 		printf(", hca init timeout\n");
3965 		return (-1);
3966 	}
3967 	if (mcx_cmdq_verify(cqe) != 0) {
3968 		printf(", hca init command corrupt\n");
3969 		return (-1);
3970 	}
3971 
3972 	status = cqe->cq_output_data[0];
3973 	if (status != MCX_CQ_STATUS_OK) {
3974 		printf(", hca init failed (%x)\n", status);
3975 		return (-1);
3976 	}
3977 
3978 	return (0);
3979 }
3980 
3981 static int
3982 mcx_set_driver_version(struct mcx_softc *sc)
3983 {
3984 	struct mcx_dmamem mxm;
3985 	struct mcx_cmdq_entry *cqe;
3986 	struct mcx_cmd_set_driver_version_in *in;
3987 	struct mcx_cmd_set_driver_version_out *out;
3988 	int error;
3989 	int token;
3990 	uint8_t status;
3991 
3992 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
3993 	token = mcx_cmdq_token(sc);
3994 	mcx_cmdq_init(sc, cqe, sizeof(*in) +
3995 	    sizeof(struct mcx_cmd_set_driver_version), sizeof(*out), token);
3996 
3997 	in = mcx_cmdq_in(cqe);
3998 	in->cmd_opcode = htobe16(MCX_CMD_SET_DRIVER_VERSION);
3999 	in->cmd_op_mod = htobe16(0);
4000 
4001 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
4002 	    &cqe->cq_input_ptr, token) != 0) {
4003 		printf(", unable to allocate set driver version mailboxen\n");
4004 		return (-1);
4005 	}
4006 	strlcpy(mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)),
4007 	    "OpenBSD,mcx,1.000.000000", MCX_CMDQ_MAILBOX_DATASIZE);
4008 
4009 	mcx_cmdq_mboxes_sign(&mxm, 1);
4010 	mcx_cmdq_post(sc, cqe, 0);
4011 
4012 	error = mcx_cmdq_poll(sc, cqe, 1000);
4013 	if (error != 0) {
4014 		printf(", set driver version timeout\n");
4015 		goto free;
4016 	}
4017 	if (mcx_cmdq_verify(cqe) != 0) {
4018 		printf(", set driver version command corrupt\n");
4019 		goto free;
4020 	}
4021 
4022 	status = cqe->cq_output_data[0];
4023 	if (status != MCX_CQ_STATUS_OK) {
4024 		printf(", set driver version failed (%x)\n", status);
4025 		error = -1;
4026 		goto free;
4027 	}
4028 
4029 free:
4030 	mcx_dmamem_free(sc, &mxm);
4031 
4032 	return (error);
4033 }
4034 
4035 static int
4036 mcx_iff(struct mcx_softc *sc)
4037 {
4038 	struct ifnet *ifp = &sc->sc_ac.ac_if;
4039 	struct mcx_dmamem mxm;
4040 	struct mcx_cmdq_entry *cqe;
4041 	struct mcx_cmd_modify_nic_vport_context_in *in;
4042 	struct mcx_cmd_modify_nic_vport_context_out *out;
4043 	struct mcx_nic_vport_ctx *ctx;
4044 	int error;
4045 	int token;
4046 	int insize;
4047 	uint32_t dest;
4048 
4049 	dest = MCX_FLOW_CONTEXT_DEST_TYPE_TABLE |
4050 	    sc->sc_rss_flow_table_id;
4051 
4052 	/* enable or disable the promisc flow */
4053 	if (ISSET(ifp->if_flags, IFF_PROMISC)) {
4054 		if (sc->sc_promisc_flow_enabled == 0) {
4055 			mcx_set_flow_table_entry_mac(sc,
4056 			    MCX_FLOW_GROUP_PROMISC, 0, NULL, dest);
4057 			sc->sc_promisc_flow_enabled = 1;
4058 		}
4059 	} else if (sc->sc_promisc_flow_enabled != 0) {
4060 		mcx_delete_flow_table_entry(sc, MCX_FLOW_GROUP_PROMISC, 0);
4061 		sc->sc_promisc_flow_enabled = 0;
4062 	}
4063 
4064 	/* enable or disable the all-multicast flow */
4065 	if (ISSET(ifp->if_flags, IFF_ALLMULTI)) {
4066 		if (sc->sc_allmulti_flow_enabled == 0) {
4067 			uint8_t mcast[ETHER_ADDR_LEN];
4068 
4069 			memset(mcast, 0, sizeof(mcast));
4070 			mcast[0] = 0x01;
4071 			mcx_set_flow_table_entry_mac(sc,
4072 			    MCX_FLOW_GROUP_ALLMULTI, 0, mcast, dest);
4073 			sc->sc_allmulti_flow_enabled = 1;
4074 		}
4075 	} else if (sc->sc_allmulti_flow_enabled != 0) {
4076 		mcx_delete_flow_table_entry(sc, MCX_FLOW_GROUP_ALLMULTI, 0);
4077 		sc->sc_allmulti_flow_enabled = 0;
4078 	}
4079 
4080 	insize = sizeof(struct mcx_nic_vport_ctx) + 240;
4081 
4082 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4083 	token = mcx_cmdq_token(sc);
4084 	mcx_cmdq_init(sc, cqe, sizeof(*in) + insize, sizeof(*out), token);
4085 
4086 	in = mcx_cmdq_in(cqe);
4087 	in->cmd_opcode = htobe16(MCX_CMD_MODIFY_NIC_VPORT_CONTEXT);
4088 	in->cmd_op_mod = htobe16(0);
4089 	in->cmd_field_select = htobe32(
4090 	    MCX_CMD_MODIFY_NIC_VPORT_CONTEXT_FIELD_PROMISC |
4091 	    MCX_CMD_MODIFY_NIC_VPORT_CONTEXT_FIELD_MTU);
4092 
4093 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1, &cqe->cq_input_ptr, token) != 0) {
4094 		printf(", unable to allocate modify "
4095 		    "nic vport context mailboxen\n");
4096 		return (-1);
4097 	}
4098 	ctx = (struct mcx_nic_vport_ctx *)
4099 	    (((char *)mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0))) + 240);
4100 	ctx->vp_mtu = htobe32(sc->sc_hardmtu);
4101 	/*
4102          * always leave promisc-all enabled on the vport since we
4103          * can't give it a vlan list, and we're already doing multicast
4104          * filtering in the flow table.
4105 	 */
4106 	ctx->vp_flags = htobe16(MCX_NIC_VPORT_CTX_PROMISC_ALL);
4107 
4108 	mcx_cmdq_mboxes_sign(&mxm, 1);
4109 	mcx_cmdq_post(sc, cqe, 0);
4110 
4111 	error = mcx_cmdq_poll(sc, cqe, 1000);
4112 	if (error != 0) {
4113 		printf(", modify nic vport context timeout\n");
4114 		goto free;
4115 	}
4116 	if (mcx_cmdq_verify(cqe) != 0) {
4117 		printf(", modify nic vport context command corrupt\n");
4118 		goto free;
4119 	}
4120 
4121 	out = mcx_cmdq_out(cqe);
4122 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4123 		printf(", modify nic vport context failed (%x, %x)\n",
4124 		    out->cmd_status, betoh32(out->cmd_syndrome));
4125 		error = -1;
4126 		goto free;
4127 	}
4128 
4129 free:
4130 	mcx_dmamem_free(sc, &mxm);
4131 
4132 	return (error);
4133 }
4134 
4135 static int
4136 mcx_alloc_uar(struct mcx_softc *sc, int *uar)
4137 {
4138 	struct mcx_cmdq_entry *cqe;
4139 	struct mcx_cmd_alloc_uar_in *in;
4140 	struct mcx_cmd_alloc_uar_out *out;
4141 	int error;
4142 
4143 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4144 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
4145 
4146 	in = mcx_cmdq_in(cqe);
4147 	in->cmd_opcode = htobe16(MCX_CMD_ALLOC_UAR);
4148 	in->cmd_op_mod = htobe16(0);
4149 
4150 	mcx_cmdq_post(sc, cqe, 0);
4151 
4152 	error = mcx_cmdq_poll(sc, cqe, 1000);
4153 	if (error != 0) {
4154 		printf(", alloc uar timeout\n");
4155 		return (-1);
4156 	}
4157 	if (mcx_cmdq_verify(cqe) != 0) {
4158 		printf(", alloc uar command corrupt\n");
4159 		return (-1);
4160 	}
4161 
4162 	out = mcx_cmdq_out(cqe);
4163 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4164 		printf(", alloc uar failed (%x)\n", out->cmd_status);
4165 		return (-1);
4166 	}
4167 
4168 	*uar = mcx_get_id(out->cmd_uar);
4169 	return (0);
4170 }
4171 
4172 static int
4173 mcx_create_eq(struct mcx_softc *sc, struct mcx_eq *eq, int uar,
4174     uint64_t events, int vector)
4175 {
4176 	struct mcx_cmdq_entry *cqe;
4177 	struct mcx_dmamem mxm;
4178 	struct mcx_cmd_create_eq_in *in;
4179 	struct mcx_cmd_create_eq_mb_in *mbin;
4180 	struct mcx_cmd_create_eq_out *out;
4181 	struct mcx_eq_entry *eqe;
4182 	int error;
4183 	uint64_t *pas;
4184 	int insize, npages, paslen, i, token;
4185 
4186 	eq->eq_cons = 0;
4187 
4188 	npages = howmany((1 << MCX_LOG_EQ_SIZE) * sizeof(struct mcx_eq_entry),
4189 	    MCX_PAGE_SIZE);
4190 	paslen = npages * sizeof(*pas);
4191 	insize = sizeof(struct mcx_cmd_create_eq_mb_in) + paslen;
4192 
4193 	if (mcx_dmamem_alloc(sc, &eq->eq_mem, npages * MCX_PAGE_SIZE,
4194 	    MCX_PAGE_SIZE) != 0) {
4195 		printf(", unable to allocate event queue memory\n");
4196 		return (-1);
4197 	}
4198 
4199 	eqe = (struct mcx_eq_entry *)MCX_DMA_KVA(&eq->eq_mem);
4200 	for (i = 0; i < (1 << MCX_LOG_EQ_SIZE); i++) {
4201 		eqe[i].eq_owner = MCX_EQ_ENTRY_OWNER_INIT;
4202 	}
4203 
4204 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4205 	token = mcx_cmdq_token(sc);
4206 	mcx_cmdq_init(sc, cqe, sizeof(*in) + insize, sizeof(*out), token);
4207 
4208 	in = mcx_cmdq_in(cqe);
4209 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_EQ);
4210 	in->cmd_op_mod = htobe16(0);
4211 
4212 	if (mcx_cmdq_mboxes_alloc(sc, &mxm,
4213 	    howmany(insize, MCX_CMDQ_MAILBOX_DATASIZE),
4214 	    &cqe->cq_input_ptr, token) != 0) {
4215 		printf(", unable to allocate create eq mailboxen\n");
4216 		goto free_eq;
4217 	}
4218 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
4219 	mbin->cmd_eq_ctx.eq_uar_size = htobe32(
4220 	    (MCX_LOG_EQ_SIZE << MCX_EQ_CTX_LOG_EQ_SIZE_SHIFT) | uar);
4221 	mbin->cmd_eq_ctx.eq_intr = vector;
4222 	mbin->cmd_event_bitmask = htobe64(events);
4223 
4224 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&eq->eq_mem),
4225 	    0, MCX_DMA_LEN(&eq->eq_mem), BUS_DMASYNC_PREREAD);
4226 
4227 	/* physical addresses follow the mailbox in data */
4228 	mcx_cmdq_mboxes_pas(&mxm, sizeof(*mbin), npages, &eq->eq_mem);
4229 	mcx_cmdq_mboxes_sign(&mxm, howmany(insize, MCX_CMDQ_MAILBOX_DATASIZE));
4230 	mcx_cmdq_post(sc, cqe, 0);
4231 
4232 	error = mcx_cmdq_poll(sc, cqe, 1000);
4233 	if (error != 0) {
4234 		printf(", create eq timeout\n");
4235 		goto free_mxm;
4236 	}
4237 	if (mcx_cmdq_verify(cqe) != 0) {
4238 		printf(", create eq command corrupt\n");
4239 		goto free_mxm;
4240 	}
4241 
4242 	out = mcx_cmdq_out(cqe);
4243 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4244 		printf(", create eq failed (%x, %x)\n", out->cmd_status,
4245 		    betoh32(out->cmd_syndrome));
4246 		goto free_mxm;
4247 	}
4248 
4249 	eq->eq_n = mcx_get_id(out->cmd_eqn);
4250 
4251 	mcx_dmamem_free(sc, &mxm);
4252 
4253 	mcx_arm_eq(sc, eq, uar);
4254 
4255 	return (0);
4256 
4257 free_mxm:
4258 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&eq->eq_mem),
4259 	    0, MCX_DMA_LEN(&eq->eq_mem), BUS_DMASYNC_POSTREAD);
4260 	mcx_dmamem_free(sc, &mxm);
4261 free_eq:
4262 	mcx_dmamem_free(sc, &eq->eq_mem);
4263 	return (-1);
4264 }
4265 
4266 static int
4267 mcx_alloc_pd(struct mcx_softc *sc)
4268 {
4269 	struct mcx_cmdq_entry *cqe;
4270 	struct mcx_cmd_alloc_pd_in *in;
4271 	struct mcx_cmd_alloc_pd_out *out;
4272 	int error;
4273 
4274 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4275 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
4276 
4277 	in = mcx_cmdq_in(cqe);
4278 	in->cmd_opcode = htobe16(MCX_CMD_ALLOC_PD);
4279 	in->cmd_op_mod = htobe16(0);
4280 
4281 	mcx_cmdq_post(sc, cqe, 0);
4282 
4283 	error = mcx_cmdq_poll(sc, cqe, 1000);
4284 	if (error != 0) {
4285 		printf(", alloc pd timeout\n");
4286 		return (-1);
4287 	}
4288 	if (mcx_cmdq_verify(cqe) != 0) {
4289 		printf(", alloc pd command corrupt\n");
4290 		return (-1);
4291 	}
4292 
4293 	out = mcx_cmdq_out(cqe);
4294 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4295 		printf(", alloc pd failed (%x)\n", out->cmd_status);
4296 		return (-1);
4297 	}
4298 
4299 	sc->sc_pd = mcx_get_id(out->cmd_pd);
4300 	return (0);
4301 }
4302 
4303 static int
4304 mcx_alloc_tdomain(struct mcx_softc *sc)
4305 {
4306 	struct mcx_cmdq_entry *cqe;
4307 	struct mcx_cmd_alloc_td_in *in;
4308 	struct mcx_cmd_alloc_td_out *out;
4309 	int error;
4310 
4311 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4312 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
4313 
4314 	in = mcx_cmdq_in(cqe);
4315 	in->cmd_opcode = htobe16(MCX_CMD_ALLOC_TRANSPORT_DOMAIN);
4316 	in->cmd_op_mod = htobe16(0);
4317 
4318 	mcx_cmdq_post(sc, cqe, 0);
4319 
4320 	error = mcx_cmdq_poll(sc, cqe, 1000);
4321 	if (error != 0) {
4322 		printf(", alloc transport domain timeout\n");
4323 		return (-1);
4324 	}
4325 	if (mcx_cmdq_verify(cqe) != 0) {
4326 		printf(", alloc transport domain command corrupt\n");
4327 		return (-1);
4328 	}
4329 
4330 	out = mcx_cmdq_out(cqe);
4331 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4332 		printf(", alloc transport domain failed (%x)\n",
4333 		    out->cmd_status);
4334 		return (-1);
4335 	}
4336 
4337 	sc->sc_tdomain = mcx_get_id(out->cmd_tdomain);
4338 	return (0);
4339 }
4340 
4341 static int
4342 mcx_query_nic_vport_context(struct mcx_softc *sc)
4343 {
4344 	struct mcx_dmamem mxm;
4345 	struct mcx_cmdq_entry *cqe;
4346 	struct mcx_cmd_query_nic_vport_context_in *in;
4347 	struct mcx_cmd_query_nic_vport_context_out *out;
4348 	struct mcx_nic_vport_ctx *ctx;
4349 	uint8_t *addr;
4350 	int error, token, i;
4351 
4352 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4353 	token = mcx_cmdq_token(sc);
4354 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out) + sizeof(*ctx), token);
4355 
4356 	in = mcx_cmdq_in(cqe);
4357 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_NIC_VPORT_CONTEXT);
4358 	in->cmd_op_mod = htobe16(0);
4359 	in->cmd_allowed_list_type = 0;
4360 
4361 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
4362 	    &cqe->cq_output_ptr, token) != 0) {
4363 		printf(", unable to allocate "
4364 		    "query nic vport context mailboxen\n");
4365 		return (-1);
4366 	}
4367 	mcx_cmdq_mboxes_sign(&mxm, 1);
4368 	mcx_cmdq_post(sc, cqe, 0);
4369 
4370 	error = mcx_cmdq_poll(sc, cqe, 1000);
4371 	if (error != 0) {
4372 		printf(", query nic vport context timeout\n");
4373 		goto free;
4374 	}
4375 	if (mcx_cmdq_verify(cqe) != 0) {
4376 		printf(", query nic vport context command corrupt\n");
4377 		goto free;
4378 	}
4379 
4380 	out = mcx_cmdq_out(cqe);
4381 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4382 		printf(", query nic vport context failed (%x, %x)\n",
4383 		    out->cmd_status, betoh32(out->cmd_syndrome));
4384 		error = -1;
4385 		goto free;
4386 	}
4387 
4388 	ctx = (struct mcx_nic_vport_ctx *)
4389 	    mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
4390 	addr = (uint8_t *)&ctx->vp_perm_addr;
4391 	for (i = 0; i < ETHER_ADDR_LEN; i++) {
4392 		sc->sc_ac.ac_enaddr[i] = addr[i + 2];
4393 	}
4394 free:
4395 	mcx_dmamem_free(sc, &mxm);
4396 
4397 	return (error);
4398 }
4399 
4400 static int
4401 mcx_query_special_contexts(struct mcx_softc *sc)
4402 {
4403 	struct mcx_cmdq_entry *cqe;
4404 	struct mcx_cmd_query_special_ctx_in *in;
4405 	struct mcx_cmd_query_special_ctx_out *out;
4406 	int error;
4407 
4408 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4409 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
4410 
4411 	in = mcx_cmdq_in(cqe);
4412 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_SPECIAL_CONTEXTS);
4413 	in->cmd_op_mod = htobe16(0);
4414 
4415 	mcx_cmdq_post(sc, cqe, 0);
4416 
4417 	error = mcx_cmdq_poll(sc, cqe, 1000);
4418 	if (error != 0) {
4419 		printf(", query special contexts timeout\n");
4420 		return (-1);
4421 	}
4422 	if (mcx_cmdq_verify(cqe) != 0) {
4423 		printf(", query special contexts command corrupt\n");
4424 		return (-1);
4425 	}
4426 
4427 	out = mcx_cmdq_out(cqe);
4428 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4429 		printf(", query special contexts failed (%x)\n",
4430 		    out->cmd_status);
4431 		return (-1);
4432 	}
4433 
4434 	sc->sc_lkey = betoh32(out->cmd_resd_lkey);
4435 	return (0);
4436 }
4437 
4438 static int
4439 mcx_set_port_mtu(struct mcx_softc *sc, int mtu)
4440 {
4441 	struct mcx_reg_pmtu pmtu;
4442 	int error;
4443 
4444 	/* read max mtu */
4445 	memset(&pmtu, 0, sizeof(pmtu));
4446 	pmtu.rp_local_port = 1;
4447 	error = mcx_access_hca_reg(sc, MCX_REG_PMTU, MCX_REG_OP_READ, &pmtu,
4448 	    sizeof(pmtu));
4449 	if (error != 0) {
4450 		printf(", unable to get port MTU\n");
4451 		return error;
4452 	}
4453 
4454 	mtu = min(mtu, betoh16(pmtu.rp_max_mtu));
4455 	pmtu.rp_admin_mtu = htobe16(mtu);
4456 	error = mcx_access_hca_reg(sc, MCX_REG_PMTU, MCX_REG_OP_WRITE, &pmtu,
4457 	    sizeof(pmtu));
4458 	if (error != 0) {
4459 		printf(", unable to set port MTU\n");
4460 		return error;
4461 	}
4462 
4463 	sc->sc_hardmtu = mtu;
4464 	sc->sc_rxbufsz = roundup(mtu + ETHER_ALIGN, sizeof(long));
4465 	return 0;
4466 }
4467 
4468 static int
4469 mcx_create_cq(struct mcx_softc *sc, struct mcx_cq *cq, int uar, int db, int eqn)
4470 {
4471 	struct mcx_cmdq_entry *cmde;
4472 	struct mcx_cq_entry *cqe;
4473 	struct mcx_dmamem mxm;
4474 	struct mcx_cmd_create_cq_in *in;
4475 	struct mcx_cmd_create_cq_mb_in *mbin;
4476 	struct mcx_cmd_create_cq_out *out;
4477 	int error;
4478 	uint64_t *pas;
4479 	int insize, npages, paslen, i, token;
4480 
4481 	cq->cq_doorbell = MCX_CQ_DOORBELL_BASE + (MCX_CQ_DOORBELL_STRIDE * db);
4482 
4483 	npages = howmany((1 << MCX_LOG_CQ_SIZE) * sizeof(struct mcx_cq_entry),
4484 	    MCX_PAGE_SIZE);
4485 	paslen = npages * sizeof(*pas);
4486 	insize = sizeof(struct mcx_cmd_create_cq_mb_in) + paslen;
4487 
4488 	if (mcx_dmamem_alloc(sc, &cq->cq_mem, npages * MCX_PAGE_SIZE,
4489 	    MCX_PAGE_SIZE) != 0) {
4490 		printf("%s: unable to allocate completion queue memory\n",
4491 		    DEVNAME(sc));
4492 		return (-1);
4493 	}
4494 	cqe = MCX_DMA_KVA(&cq->cq_mem);
4495 	for (i = 0; i < (1 << MCX_LOG_CQ_SIZE); i++) {
4496 		cqe[i].cq_opcode_owner = MCX_CQ_ENTRY_FLAG_OWNER;
4497 	}
4498 
4499 	cmde = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4500 	token = mcx_cmdq_token(sc);
4501 	mcx_cmdq_init(sc, cmde, sizeof(*in) + insize, sizeof(*out), token);
4502 
4503 	in = mcx_cmdq_in(cmde);
4504 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_CQ);
4505 	in->cmd_op_mod = htobe16(0);
4506 
4507 	if (mcx_cmdq_mboxes_alloc(sc, &mxm,
4508 	    howmany(insize, MCX_CMDQ_MAILBOX_DATASIZE),
4509 	    &cmde->cq_input_ptr, token) != 0) {
4510 		printf("%s: unable to allocate create cq mailboxen\n",
4511 		    DEVNAME(sc));
4512 		goto free_cq;
4513 	}
4514 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
4515 	mbin->cmd_cq_ctx.cq_uar_size = htobe32(
4516 	    (MCX_LOG_CQ_SIZE << MCX_CQ_CTX_LOG_CQ_SIZE_SHIFT) | uar);
4517 	mbin->cmd_cq_ctx.cq_eqn = htobe32(eqn);
4518 	mbin->cmd_cq_ctx.cq_period_max_count = htobe32(
4519 	    (MCX_CQ_MOD_PERIOD << MCX_CQ_CTX_PERIOD_SHIFT) |
4520 	    MCX_CQ_MOD_COUNTER);
4521 	mbin->cmd_cq_ctx.cq_doorbell = htobe64(
4522 	    MCX_DMA_DVA(&sc->sc_doorbell_mem) + cq->cq_doorbell);
4523 
4524 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&cq->cq_mem),
4525 	    0, MCX_DMA_LEN(&cq->cq_mem), BUS_DMASYNC_PREREAD);
4526 
4527 	/* physical addresses follow the mailbox in data */
4528 	mcx_cmdq_mboxes_pas(&mxm, sizeof(*mbin), npages, &cq->cq_mem);
4529 	mcx_cmdq_post(sc, cmde, 0);
4530 
4531 	error = mcx_cmdq_poll(sc, cmde, 1000);
4532 	if (error != 0) {
4533 		printf("%s: create cq timeout\n", DEVNAME(sc));
4534 		goto free_mxm;
4535 	}
4536 	if (mcx_cmdq_verify(cmde) != 0) {
4537 		printf("%s: create cq command corrupt\n", DEVNAME(sc));
4538 		goto free_mxm;
4539 	}
4540 
4541 	out = mcx_cmdq_out(cmde);
4542 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4543 		printf("%s: create cq failed (%x, %x)\n", DEVNAME(sc),
4544 		    out->cmd_status, betoh32(out->cmd_syndrome));
4545 		goto free_mxm;
4546 	}
4547 
4548 	cq->cq_n = mcx_get_id(out->cmd_cqn);
4549 	cq->cq_cons = 0;
4550 	cq->cq_count = 0;
4551 
4552 	mcx_dmamem_free(sc, &mxm);
4553 
4554 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_doorbell_mem),
4555 	    cq->cq_doorbell, sizeof(struct mcx_cq_doorbell),
4556 	    BUS_DMASYNC_PREWRITE);
4557 
4558 	mcx_arm_cq(sc, cq, uar);
4559 
4560 	return (0);
4561 
4562 free_mxm:
4563 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&cq->cq_mem),
4564 	    0, MCX_DMA_LEN(&cq->cq_mem), BUS_DMASYNC_POSTREAD);
4565 	mcx_dmamem_free(sc, &mxm);
4566 free_cq:
4567 	mcx_dmamem_free(sc, &cq->cq_mem);
4568 	return (-1);
4569 }
4570 
4571 static int
4572 mcx_destroy_cq(struct mcx_softc *sc, struct mcx_cq *cq)
4573 {
4574 	struct mcx_cmdq_entry *cqe;
4575 	struct mcx_cmd_destroy_cq_in *in;
4576 	struct mcx_cmd_destroy_cq_out *out;
4577 	int error;
4578 	int token;
4579 
4580 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4581 	token = mcx_cmdq_token(sc);
4582 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), token);
4583 
4584 	in = mcx_cmdq_in(cqe);
4585 	in->cmd_opcode = htobe16(MCX_CMD_DESTROY_CQ);
4586 	in->cmd_op_mod = htobe16(0);
4587 	in->cmd_cqn = htobe32(cq->cq_n);
4588 
4589 	mcx_cmdq_post(sc, cqe, 0);
4590 	error = mcx_cmdq_poll(sc, cqe, 1000);
4591 	if (error != 0) {
4592 		printf("%s: destroy cq timeout\n", DEVNAME(sc));
4593 		return error;
4594 	}
4595 	if (mcx_cmdq_verify(cqe) != 0) {
4596 		printf("%s: destroy cq command corrupt\n", DEVNAME(sc));
4597 		return error;
4598 	}
4599 
4600 	out = mcx_cmdq_out(cqe);
4601 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4602 		printf("%s: destroy cq failed (%x, %x)\n", DEVNAME(sc),
4603 		    out->cmd_status, betoh32(out->cmd_syndrome));
4604 		return -1;
4605 	}
4606 
4607 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_doorbell_mem),
4608 	    cq->cq_doorbell, sizeof(struct mcx_cq_doorbell),
4609 	    BUS_DMASYNC_POSTWRITE);
4610 
4611 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&cq->cq_mem),
4612 	    0, MCX_DMA_LEN(&cq->cq_mem), BUS_DMASYNC_POSTREAD);
4613 	mcx_dmamem_free(sc, &cq->cq_mem);
4614 
4615 	cq->cq_n = 0;
4616 	cq->cq_cons = 0;
4617 	cq->cq_count = 0;
4618 	return 0;
4619 }
4620 
4621 static int
4622 mcx_create_rq(struct mcx_softc *sc, struct mcx_rx *rx, int db, int cqn)
4623 {
4624 	struct mcx_cmdq_entry *cqe;
4625 	struct mcx_dmamem mxm;
4626 	struct mcx_cmd_create_rq_in *in;
4627 	struct mcx_cmd_create_rq_out *out;
4628 	struct mcx_rq_ctx *mbin;
4629 	int error;
4630 	uint64_t *pas;
4631 	uint32_t rq_flags;
4632 	int insize, npages, paslen, token;
4633 
4634 	rx->rx_doorbell = MCX_WQ_DOORBELL_BASE +
4635 	    (db * MCX_WQ_DOORBELL_STRIDE);
4636 
4637 	npages = howmany((1 << MCX_LOG_RQ_SIZE) * sizeof(struct mcx_rq_entry),
4638 	    MCX_PAGE_SIZE);
4639 	paslen = npages * sizeof(*pas);
4640 	insize = 0x10 + sizeof(struct mcx_rq_ctx) + paslen;
4641 
4642 	if (mcx_dmamem_alloc(sc, &rx->rx_rq_mem, npages * MCX_PAGE_SIZE,
4643 	    MCX_PAGE_SIZE) != 0) {
4644 		printf("%s: unable to allocate receive queue memory\n",
4645 		    DEVNAME(sc));
4646 		return (-1);
4647 	}
4648 
4649 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4650 	token = mcx_cmdq_token(sc);
4651 	mcx_cmdq_init(sc, cqe, sizeof(*in) + insize, sizeof(*out), token);
4652 
4653 	in = mcx_cmdq_in(cqe);
4654 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_RQ);
4655 	in->cmd_op_mod = htobe16(0);
4656 
4657 	if (mcx_cmdq_mboxes_alloc(sc, &mxm,
4658 	    howmany(insize, MCX_CMDQ_MAILBOX_DATASIZE),
4659 	    &cqe->cq_input_ptr, token) != 0) {
4660 		printf("%s: unable to allocate create rq mailboxen\n",
4661 		    DEVNAME(sc));
4662 		goto free_rq;
4663 	}
4664 	mbin = (struct mcx_rq_ctx *)
4665 	    (((char *)mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0))) + 0x10);
4666 	rq_flags = MCX_RQ_CTX_RLKEY;
4667 #if NVLAN == 0
4668 	rq_flags |= MCX_RQ_CTX_VLAN_STRIP_DIS;
4669 #endif
4670 	mbin->rq_flags = htobe32(rq_flags);
4671 	mbin->rq_cqn = htobe32(cqn);
4672 	mbin->rq_wq.wq_type = MCX_WQ_CTX_TYPE_CYCLIC;
4673 	mbin->rq_wq.wq_pd = htobe32(sc->sc_pd);
4674 	mbin->rq_wq.wq_doorbell = htobe64(MCX_DMA_DVA(&sc->sc_doorbell_mem) +
4675 	    rx->rx_doorbell);
4676 	mbin->rq_wq.wq_log_stride = htobe16(4);
4677 	mbin->rq_wq.wq_log_size = MCX_LOG_RQ_SIZE;
4678 
4679 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&rx->rx_rq_mem),
4680 	    0, MCX_DMA_LEN(&rx->rx_rq_mem), BUS_DMASYNC_PREWRITE);
4681 
4682 	/* physical addresses follow the mailbox in data */
4683 	mcx_cmdq_mboxes_pas(&mxm, sizeof(*mbin) + 0x10, npages, &rx->rx_rq_mem);
4684 	mcx_cmdq_post(sc, cqe, 0);
4685 
4686 	error = mcx_cmdq_poll(sc, cqe, 1000);
4687 	if (error != 0) {
4688 		printf("%s: create rq timeout\n", DEVNAME(sc));
4689 		goto free_mxm;
4690 	}
4691 	if (mcx_cmdq_verify(cqe) != 0) {
4692 		printf("%s: create rq command corrupt\n", DEVNAME(sc));
4693 		goto free_mxm;
4694 	}
4695 
4696 	out = mcx_cmdq_out(cqe);
4697 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4698 		printf("%s: create rq failed (%x, %x)\n", DEVNAME(sc),
4699 		    out->cmd_status, betoh32(out->cmd_syndrome));
4700 		goto free_mxm;
4701 	}
4702 
4703 	rx->rx_rqn = mcx_get_id(out->cmd_rqn);
4704 
4705 	mcx_dmamem_free(sc, &mxm);
4706 
4707 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_doorbell_mem),
4708 	    rx->rx_doorbell, sizeof(uint32_t), BUS_DMASYNC_PREWRITE);
4709 
4710 	return (0);
4711 
4712 free_mxm:
4713 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&rx->rx_rq_mem),
4714 	    0, MCX_DMA_LEN(&rx->rx_rq_mem), BUS_DMASYNC_POSTWRITE);
4715 	mcx_dmamem_free(sc, &mxm);
4716 free_rq:
4717 	mcx_dmamem_free(sc, &rx->rx_rq_mem);
4718 	return (-1);
4719 }
4720 
4721 static int
4722 mcx_ready_rq(struct mcx_softc *sc, struct mcx_rx *rx)
4723 {
4724 	struct mcx_cmdq_entry *cqe;
4725 	struct mcx_dmamem mxm;
4726 	struct mcx_cmd_modify_rq_in *in;
4727 	struct mcx_cmd_modify_rq_mb_in *mbin;
4728 	struct mcx_cmd_modify_rq_out *out;
4729 	int error;
4730 	int token;
4731 
4732 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4733 	token = mcx_cmdq_token(sc);
4734 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
4735 	    sizeof(*out), token);
4736 
4737 	in = mcx_cmdq_in(cqe);
4738 	in->cmd_opcode = htobe16(MCX_CMD_MODIFY_RQ);
4739 	in->cmd_op_mod = htobe16(0);
4740 	in->cmd_rq_state = htobe32((MCX_QUEUE_STATE_RST << 28) | rx->rx_rqn);
4741 
4742 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
4743 	    &cqe->cq_input_ptr, token) != 0) {
4744 		printf("%s: unable to allocate modify rq mailbox\n",
4745 		    DEVNAME(sc));
4746 		return (-1);
4747 	}
4748 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
4749 	mbin->cmd_rq_ctx.rq_flags = htobe32(
4750 	    MCX_QUEUE_STATE_RDY << MCX_RQ_CTX_STATE_SHIFT);
4751 
4752 	mcx_cmdq_mboxes_sign(&mxm, 1);
4753 	mcx_cmdq_post(sc, cqe, 0);
4754 	error = mcx_cmdq_poll(sc, cqe, 1000);
4755 	if (error != 0) {
4756 		printf("%s: modify rq timeout\n", DEVNAME(sc));
4757 		goto free;
4758 	}
4759 	if (mcx_cmdq_verify(cqe) != 0) {
4760 		printf("%s: modify rq command corrupt\n", DEVNAME(sc));
4761 		goto free;
4762 	}
4763 
4764 	out = mcx_cmdq_out(cqe);
4765 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4766 		printf("%s: modify rq failed (%x, %x)\n", DEVNAME(sc),
4767 		    out->cmd_status, betoh32(out->cmd_syndrome));
4768 		error = -1;
4769 		goto free;
4770 	}
4771 
4772 free:
4773 	mcx_dmamem_free(sc, &mxm);
4774 	return (error);
4775 }
4776 
4777 static int
4778 mcx_destroy_rq(struct mcx_softc *sc, struct mcx_rx *rx)
4779 {
4780 	struct mcx_cmdq_entry *cqe;
4781 	struct mcx_cmd_destroy_rq_in *in;
4782 	struct mcx_cmd_destroy_rq_out *out;
4783 	int error;
4784 	int token;
4785 
4786 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4787 	token = mcx_cmdq_token(sc);
4788 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), token);
4789 
4790 	in = mcx_cmdq_in(cqe);
4791 	in->cmd_opcode = htobe16(MCX_CMD_DESTROY_RQ);
4792 	in->cmd_op_mod = htobe16(0);
4793 	in->cmd_rqn = htobe32(rx->rx_rqn);
4794 
4795 	mcx_cmdq_post(sc, cqe, 0);
4796 	error = mcx_cmdq_poll(sc, cqe, 1000);
4797 	if (error != 0) {
4798 		printf("%s: destroy rq timeout\n", DEVNAME(sc));
4799 		return error;
4800 	}
4801 	if (mcx_cmdq_verify(cqe) != 0) {
4802 		printf("%s: destroy rq command corrupt\n", DEVNAME(sc));
4803 		return error;
4804 	}
4805 
4806 	out = mcx_cmdq_out(cqe);
4807 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4808 		printf("%s: destroy rq failed (%x, %x)\n", DEVNAME(sc),
4809 		    out->cmd_status, betoh32(out->cmd_syndrome));
4810 		return -1;
4811 	}
4812 
4813 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_doorbell_mem),
4814 	    rx->rx_doorbell, sizeof(uint32_t), BUS_DMASYNC_POSTWRITE);
4815 
4816 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&rx->rx_rq_mem),
4817 	    0, MCX_DMA_LEN(&rx->rx_rq_mem), BUS_DMASYNC_POSTWRITE);
4818 	mcx_dmamem_free(sc, &rx->rx_rq_mem);
4819 
4820 	rx->rx_rqn = 0;
4821 	return 0;
4822 }
4823 
4824 static int
4825 mcx_create_tir_direct(struct mcx_softc *sc, struct mcx_rx *rx, int *tirn)
4826 {
4827 	struct mcx_cmdq_entry *cqe;
4828 	struct mcx_dmamem mxm;
4829 	struct mcx_cmd_create_tir_in *in;
4830 	struct mcx_cmd_create_tir_mb_in *mbin;
4831 	struct mcx_cmd_create_tir_out *out;
4832 	int error;
4833 	int token;
4834 
4835 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4836 	token = mcx_cmdq_token(sc);
4837 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
4838 	    sizeof(*out), token);
4839 
4840 	in = mcx_cmdq_in(cqe);
4841 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_TIR);
4842 	in->cmd_op_mod = htobe16(0);
4843 
4844 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
4845 	    &cqe->cq_input_ptr, token) != 0) {
4846 		printf("%s: unable to allocate create tir mailbox\n",
4847 		    DEVNAME(sc));
4848 		return (-1);
4849 	}
4850 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
4851 	/* leave disp_type = 0, so packets get sent to the inline rqn */
4852 	mbin->cmd_inline_rqn = htobe32(rx->rx_rqn);
4853 	mbin->cmd_tdomain = htobe32(sc->sc_tdomain);
4854 
4855 	mcx_cmdq_post(sc, cqe, 0);
4856 	error = mcx_cmdq_poll(sc, cqe, 1000);
4857 	if (error != 0) {
4858 		printf("%s: create tir timeout\n", DEVNAME(sc));
4859 		goto free;
4860 	}
4861 	if (mcx_cmdq_verify(cqe) != 0) {
4862 		printf("%s: create tir command corrupt\n", DEVNAME(sc));
4863 		goto free;
4864 	}
4865 
4866 	out = mcx_cmdq_out(cqe);
4867 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4868 		printf("%s: create tir failed (%x, %x)\n", DEVNAME(sc),
4869 		    out->cmd_status, betoh32(out->cmd_syndrome));
4870 		error = -1;
4871 		goto free;
4872 	}
4873 
4874 	*tirn = mcx_get_id(out->cmd_tirn);
4875 free:
4876 	mcx_dmamem_free(sc, &mxm);
4877 	return (error);
4878 }
4879 
4880 static int
4881 mcx_create_tir_indirect(struct mcx_softc *sc, int rqtn, uint32_t hash_sel,
4882     int *tirn)
4883 {
4884 	struct mcx_cmdq_entry *cqe;
4885 	struct mcx_dmamem mxm;
4886 	struct mcx_cmd_create_tir_in *in;
4887 	struct mcx_cmd_create_tir_mb_in *mbin;
4888 	struct mcx_cmd_create_tir_out *out;
4889 	int error;
4890 	int token;
4891 
4892 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4893 	token = mcx_cmdq_token(sc);
4894 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
4895 	    sizeof(*out), token);
4896 
4897 	in = mcx_cmdq_in(cqe);
4898 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_TIR);
4899 	in->cmd_op_mod = htobe16(0);
4900 
4901 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
4902 	    &cqe->cq_input_ptr, token) != 0) {
4903 		printf("%s: unable to allocate create tir mailbox\n",
4904 		    DEVNAME(sc));
4905 		return (-1);
4906 	}
4907 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
4908 	mbin->cmd_disp_type = htobe32(MCX_TIR_CTX_DISP_TYPE_INDIRECT
4909 	    << MCX_TIR_CTX_DISP_TYPE_SHIFT);
4910 	mbin->cmd_indir_table = htobe32(rqtn);
4911 	mbin->cmd_tdomain = htobe32(sc->sc_tdomain |
4912 	    MCX_TIR_CTX_HASH_TOEPLITZ << MCX_TIR_CTX_HASH_SHIFT);
4913 	mbin->cmd_rx_hash_sel_outer = htobe32(hash_sel);
4914 	stoeplitz_to_key(&mbin->cmd_rx_hash_key,
4915 	    sizeof(mbin->cmd_rx_hash_key));
4916 
4917 	mcx_cmdq_post(sc, cqe, 0);
4918 	error = mcx_cmdq_poll(sc, cqe, 1000);
4919 	if (error != 0) {
4920 		printf("%s: create tir timeout\n", DEVNAME(sc));
4921 		goto free;
4922 	}
4923 	if (mcx_cmdq_verify(cqe) != 0) {
4924 		printf("%s: create tir command corrupt\n", DEVNAME(sc));
4925 		goto free;
4926 	}
4927 
4928 	out = mcx_cmdq_out(cqe);
4929 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4930 		printf("%s: create tir failed (%x, %x)\n", DEVNAME(sc),
4931 		    out->cmd_status, betoh32(out->cmd_syndrome));
4932 		error = -1;
4933 		goto free;
4934 	}
4935 
4936 	*tirn = mcx_get_id(out->cmd_tirn);
4937 free:
4938 	mcx_dmamem_free(sc, &mxm);
4939 	return (error);
4940 }
4941 
4942 static int
4943 mcx_destroy_tir(struct mcx_softc *sc, int tirn)
4944 {
4945 	struct mcx_cmdq_entry *cqe;
4946 	struct mcx_cmd_destroy_tir_in *in;
4947 	struct mcx_cmd_destroy_tir_out *out;
4948 	int error;
4949 	int token;
4950 
4951 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4952 	token = mcx_cmdq_token(sc);
4953 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), token);
4954 
4955 	in = mcx_cmdq_in(cqe);
4956 	in->cmd_opcode = htobe16(MCX_CMD_DESTROY_TIR);
4957 	in->cmd_op_mod = htobe16(0);
4958 	in->cmd_tirn = htobe32(tirn);
4959 
4960 	mcx_cmdq_post(sc, cqe, 0);
4961 	error = mcx_cmdq_poll(sc, cqe, 1000);
4962 	if (error != 0) {
4963 		printf("%s: destroy tir timeout\n", DEVNAME(sc));
4964 		return error;
4965 	}
4966 	if (mcx_cmdq_verify(cqe) != 0) {
4967 		printf("%s: destroy tir command corrupt\n", DEVNAME(sc));
4968 		return error;
4969 	}
4970 
4971 	out = mcx_cmdq_out(cqe);
4972 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4973 		printf("%s: destroy tir failed (%x, %x)\n", DEVNAME(sc),
4974 		    out->cmd_status, betoh32(out->cmd_syndrome));
4975 		return -1;
4976 	}
4977 
4978 	return (0);
4979 }
4980 
4981 static int
4982 mcx_create_sq(struct mcx_softc *sc, struct mcx_tx *tx, int uar, int db,
4983     int cqn)
4984 {
4985 	struct mcx_cmdq_entry *cqe;
4986 	struct mcx_dmamem mxm;
4987 	struct mcx_cmd_create_sq_in *in;
4988 	struct mcx_sq_ctx *mbin;
4989 	struct mcx_cmd_create_sq_out *out;
4990 	int error;
4991 	uint64_t *pas;
4992 	int insize, npages, paslen, token;
4993 
4994 	tx->tx_doorbell = MCX_WQ_DOORBELL_BASE +
4995 	    (db * MCX_WQ_DOORBELL_STRIDE) + 4;
4996 
4997 	npages = howmany((1 << MCX_LOG_SQ_SIZE) * sizeof(struct mcx_sq_entry),
4998 	    MCX_PAGE_SIZE);
4999 	paslen = npages * sizeof(*pas);
5000 	insize = sizeof(struct mcx_sq_ctx) + paslen;
5001 
5002 	if (mcx_dmamem_alloc(sc, &tx->tx_sq_mem, npages * MCX_PAGE_SIZE,
5003 	    MCX_PAGE_SIZE) != 0) {
5004 		printf("%s: unable to allocate send queue memory\n",
5005 		    DEVNAME(sc));
5006 		return (-1);
5007 	}
5008 
5009 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5010 	token = mcx_cmdq_token(sc);
5011 	mcx_cmdq_init(sc, cqe, sizeof(*in) + insize + paslen, sizeof(*out),
5012 	    token);
5013 
5014 	in = mcx_cmdq_in(cqe);
5015 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_SQ);
5016 	in->cmd_op_mod = htobe16(0);
5017 
5018 	if (mcx_cmdq_mboxes_alloc(sc, &mxm,
5019 	    howmany(insize, MCX_CMDQ_MAILBOX_DATASIZE),
5020 	    &cqe->cq_input_ptr, token) != 0) {
5021 		printf("%s: unable to allocate create sq mailboxen\n",
5022 		    DEVNAME(sc));
5023 		goto free_sq;
5024 	}
5025 	mbin = (struct mcx_sq_ctx *)
5026 	    (((char *)mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0))) + 0x10);
5027 	mbin->sq_flags = htobe32(MCX_SQ_CTX_RLKEY |
5028 	    (1 << MCX_SQ_CTX_MIN_WQE_INLINE_SHIFT));
5029 	mbin->sq_cqn = htobe32(cqn);
5030 	mbin->sq_tis_lst_sz = htobe32(1 << MCX_SQ_CTX_TIS_LST_SZ_SHIFT);
5031 	mbin->sq_tis_num = htobe32(sc->sc_tis);
5032 	mbin->sq_wq.wq_type = MCX_WQ_CTX_TYPE_CYCLIC;
5033 	mbin->sq_wq.wq_pd = htobe32(sc->sc_pd);
5034 	mbin->sq_wq.wq_uar_page = htobe32(uar);
5035 	mbin->sq_wq.wq_doorbell = htobe64(MCX_DMA_DVA(&sc->sc_doorbell_mem) +
5036 	    tx->tx_doorbell);
5037 	mbin->sq_wq.wq_log_stride = htobe16(MCX_LOG_SQ_ENTRY_SIZE);
5038 	mbin->sq_wq.wq_log_size = MCX_LOG_SQ_SIZE;
5039 
5040 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&tx->tx_sq_mem),
5041 	    0, MCX_DMA_LEN(&tx->tx_sq_mem), BUS_DMASYNC_PREWRITE);
5042 
5043 	/* physical addresses follow the mailbox in data */
5044 	mcx_cmdq_mboxes_pas(&mxm, sizeof(*mbin) + 0x10,
5045 	    npages, &tx->tx_sq_mem);
5046 	mcx_cmdq_post(sc, cqe, 0);
5047 
5048 	error = mcx_cmdq_poll(sc, cqe, 1000);
5049 	if (error != 0) {
5050 		printf("%s: create sq timeout\n", DEVNAME(sc));
5051 		goto free_mxm;
5052 	}
5053 	if (mcx_cmdq_verify(cqe) != 0) {
5054 		printf("%s: create sq command corrupt\n", DEVNAME(sc));
5055 		goto free_mxm;
5056 	}
5057 
5058 	out = mcx_cmdq_out(cqe);
5059 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5060 		printf("%s: create sq failed (%x, %x)\n", DEVNAME(sc),
5061 		    out->cmd_status, betoh32(out->cmd_syndrome));
5062 		goto free_mxm;
5063 	}
5064 
5065 	tx->tx_uar = uar;
5066 	tx->tx_sqn = mcx_get_id(out->cmd_sqn);
5067 
5068 	mcx_dmamem_free(sc, &mxm);
5069 
5070 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_doorbell_mem),
5071 	    tx->tx_doorbell, sizeof(uint32_t), BUS_DMASYNC_PREWRITE);
5072 
5073 	return (0);
5074 
5075 free_mxm:
5076 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&tx->tx_sq_mem),
5077 	    0, MCX_DMA_LEN(&tx->tx_sq_mem), BUS_DMASYNC_POSTWRITE);
5078 	mcx_dmamem_free(sc, &mxm);
5079 free_sq:
5080 	mcx_dmamem_free(sc, &tx->tx_sq_mem);
5081 	return (-1);
5082 }
5083 
5084 static int
5085 mcx_destroy_sq(struct mcx_softc *sc, struct mcx_tx *tx)
5086 {
5087 	struct mcx_cmdq_entry *cqe;
5088 	struct mcx_cmd_destroy_sq_in *in;
5089 	struct mcx_cmd_destroy_sq_out *out;
5090 	int error;
5091 	int token;
5092 
5093 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5094 	token = mcx_cmdq_token(sc);
5095 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), token);
5096 
5097 	in = mcx_cmdq_in(cqe);
5098 	in->cmd_opcode = htobe16(MCX_CMD_DESTROY_SQ);
5099 	in->cmd_op_mod = htobe16(0);
5100 	in->cmd_sqn = htobe32(tx->tx_sqn);
5101 
5102 	mcx_cmdq_post(sc, cqe, 0);
5103 	error = mcx_cmdq_poll(sc, cqe, 1000);
5104 	if (error != 0) {
5105 		printf("%s: destroy sq timeout\n", DEVNAME(sc));
5106 		return error;
5107 	}
5108 	if (mcx_cmdq_verify(cqe) != 0) {
5109 		printf("%s: destroy sq command corrupt\n", DEVNAME(sc));
5110 		return error;
5111 	}
5112 
5113 	out = mcx_cmdq_out(cqe);
5114 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5115 		printf("%s: destroy sq failed (%x, %x)\n", DEVNAME(sc),
5116 		    out->cmd_status, betoh32(out->cmd_syndrome));
5117 		return -1;
5118 	}
5119 
5120 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_doorbell_mem),
5121 	    tx->tx_doorbell, sizeof(uint32_t), BUS_DMASYNC_POSTWRITE);
5122 
5123 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&tx->tx_sq_mem),
5124 	    0, MCX_DMA_LEN(&tx->tx_sq_mem), BUS_DMASYNC_POSTWRITE);
5125 	mcx_dmamem_free(sc, &tx->tx_sq_mem);
5126 
5127 	tx->tx_sqn = 0;
5128 	return 0;
5129 }
5130 
5131 static int
5132 mcx_ready_sq(struct mcx_softc *sc, struct mcx_tx *tx)
5133 {
5134 	struct mcx_cmdq_entry *cqe;
5135 	struct mcx_dmamem mxm;
5136 	struct mcx_cmd_modify_sq_in *in;
5137 	struct mcx_cmd_modify_sq_mb_in *mbin;
5138 	struct mcx_cmd_modify_sq_out *out;
5139 	int error;
5140 	int token;
5141 
5142 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5143 	token = mcx_cmdq_token(sc);
5144 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
5145 	    sizeof(*out), token);
5146 
5147 	in = mcx_cmdq_in(cqe);
5148 	in->cmd_opcode = htobe16(MCX_CMD_MODIFY_SQ);
5149 	in->cmd_op_mod = htobe16(0);
5150 	in->cmd_sq_state = htobe32((MCX_QUEUE_STATE_RST << 28) | tx->tx_sqn);
5151 
5152 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
5153 	    &cqe->cq_input_ptr, token) != 0) {
5154 		printf("%s: unable to allocate modify sq mailbox\n",
5155 		    DEVNAME(sc));
5156 		return (-1);
5157 	}
5158 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5159 	mbin->cmd_sq_ctx.sq_flags = htobe32(
5160 	    MCX_QUEUE_STATE_RDY << MCX_SQ_CTX_STATE_SHIFT);
5161 
5162 	mcx_cmdq_mboxes_sign(&mxm, 1);
5163 	mcx_cmdq_post(sc, cqe, 0);
5164 	error = mcx_cmdq_poll(sc, cqe, 1000);
5165 	if (error != 0) {
5166 		printf("%s: modify sq timeout\n", DEVNAME(sc));
5167 		goto free;
5168 	}
5169 	if (mcx_cmdq_verify(cqe) != 0) {
5170 		printf("%s: modify sq command corrupt\n", DEVNAME(sc));
5171 		goto free;
5172 	}
5173 
5174 	out = mcx_cmdq_out(cqe);
5175 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5176 		printf("%s: modify sq failed (%x, %x)\n", DEVNAME(sc),
5177 		    out->cmd_status, betoh32(out->cmd_syndrome));
5178 		error = -1;
5179 		goto free;
5180 	}
5181 
5182 free:
5183 	mcx_dmamem_free(sc, &mxm);
5184 	return (error);
5185 }
5186 
5187 static int
5188 mcx_create_tis(struct mcx_softc *sc, int *tis)
5189 {
5190 	struct mcx_cmdq_entry *cqe;
5191 	struct mcx_dmamem mxm;
5192 	struct mcx_cmd_create_tis_in *in;
5193 	struct mcx_cmd_create_tis_mb_in *mbin;
5194 	struct mcx_cmd_create_tis_out *out;
5195 	int error;
5196 	int token;
5197 
5198 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5199 	token = mcx_cmdq_token(sc);
5200 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
5201 	    sizeof(*out), token);
5202 
5203 	in = mcx_cmdq_in(cqe);
5204 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_TIS);
5205 	in->cmd_op_mod = htobe16(0);
5206 
5207 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
5208 	    &cqe->cq_input_ptr, token) != 0) {
5209 		printf("%s: unable to allocate create tis mailbox\n",
5210 		    DEVNAME(sc));
5211 		return (-1);
5212 	}
5213 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5214 	mbin->cmd_tdomain = htobe32(sc->sc_tdomain);
5215 
5216 	mcx_cmdq_mboxes_sign(&mxm, 1);
5217 	mcx_cmdq_post(sc, cqe, 0);
5218 	error = mcx_cmdq_poll(sc, cqe, 1000);
5219 	if (error != 0) {
5220 		printf("%s: create tis timeout\n", DEVNAME(sc));
5221 		goto free;
5222 	}
5223 	if (mcx_cmdq_verify(cqe) != 0) {
5224 		printf("%s: create tis command corrupt\n", DEVNAME(sc));
5225 		goto free;
5226 	}
5227 
5228 	out = mcx_cmdq_out(cqe);
5229 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5230 		printf("%s: create tis failed (%x, %x)\n", DEVNAME(sc),
5231 		    out->cmd_status, betoh32(out->cmd_syndrome));
5232 		error = -1;
5233 		goto free;
5234 	}
5235 
5236 	*tis = mcx_get_id(out->cmd_tisn);
5237 free:
5238 	mcx_dmamem_free(sc, &mxm);
5239 	return (error);
5240 }
5241 
5242 static int
5243 mcx_destroy_tis(struct mcx_softc *sc, int tis)
5244 {
5245 	struct mcx_cmdq_entry *cqe;
5246 	struct mcx_cmd_destroy_tis_in *in;
5247 	struct mcx_cmd_destroy_tis_out *out;
5248 	int error;
5249 	int token;
5250 
5251 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5252 	token = mcx_cmdq_token(sc);
5253 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), token);
5254 
5255 	in = mcx_cmdq_in(cqe);
5256 	in->cmd_opcode = htobe16(MCX_CMD_DESTROY_TIS);
5257 	in->cmd_op_mod = htobe16(0);
5258 	in->cmd_tisn = htobe32(tis);
5259 
5260 	mcx_cmdq_post(sc, cqe, 0);
5261 	error = mcx_cmdq_poll(sc, cqe, 1000);
5262 	if (error != 0) {
5263 		printf("%s: destroy tis timeout\n", DEVNAME(sc));
5264 		return error;
5265 	}
5266 	if (mcx_cmdq_verify(cqe) != 0) {
5267 		printf("%s: destroy tis command corrupt\n", DEVNAME(sc));
5268 		return error;
5269 	}
5270 
5271 	out = mcx_cmdq_out(cqe);
5272 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5273 		printf("%s: destroy tis failed (%x, %x)\n", DEVNAME(sc),
5274 		    out->cmd_status, betoh32(out->cmd_syndrome));
5275 		return -1;
5276 	}
5277 
5278 	return 0;
5279 }
5280 
5281 static int
5282 mcx_create_rqt(struct mcx_softc *sc, int size, int *rqns, int *rqt)
5283 {
5284 	struct mcx_cmdq_entry *cqe;
5285 	struct mcx_dmamem mxm;
5286 	struct mcx_cmd_create_rqt_in *in;
5287 	struct mcx_cmd_create_rqt_mb_in *mbin;
5288 	struct mcx_cmd_create_rqt_out *out;
5289 	struct mcx_rqt_ctx *rqt_ctx;
5290 	int *rqtn;
5291 	int error;
5292 	int token;
5293 	int i;
5294 
5295 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5296 	token = mcx_cmdq_token(sc);
5297 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin) +
5298 	    (size * sizeof(int)), sizeof(*out), token);
5299 
5300 	in = mcx_cmdq_in(cqe);
5301 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_RQT);
5302 	in->cmd_op_mod = htobe16(0);
5303 
5304 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
5305 	    &cqe->cq_input_ptr, token) != 0) {
5306 		printf("%s: unable to allocate create rqt mailbox\n",
5307 		    DEVNAME(sc));
5308 		return (-1);
5309 	}
5310 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5311 	rqt_ctx = &mbin->cmd_rqt;
5312 	rqt_ctx->cmd_rqt_max_size = htobe16(sc->sc_max_rqt_size);
5313 	rqt_ctx->cmd_rqt_actual_size = htobe16(size);
5314 
5315 	/* rqt list follows the rqt context */
5316 	rqtn = (int *)(rqt_ctx + 1);
5317 	for (i = 0; i < size; i++) {
5318 		rqtn[i] = htobe32(rqns[i]);
5319 	}
5320 
5321 	mcx_cmdq_mboxes_sign(&mxm, 1);
5322 	mcx_cmdq_post(sc, cqe, 0);
5323 	error = mcx_cmdq_poll(sc, cqe, 1000);
5324 	if (error != 0) {
5325 		printf("%s: create rqt timeout\n", DEVNAME(sc));
5326 		goto free;
5327 	}
5328 	if (mcx_cmdq_verify(cqe) != 0) {
5329 		printf("%s: create rqt command corrupt\n", DEVNAME(sc));
5330 		goto free;
5331 	}
5332 
5333 	out = mcx_cmdq_out(cqe);
5334 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5335 		printf("%s: create rqt failed (%x, %x)\n", DEVNAME(sc),
5336 		    out->cmd_status, betoh32(out->cmd_syndrome));
5337 		error = -1;
5338 		goto free;
5339 	}
5340 
5341 	*rqt = mcx_get_id(out->cmd_rqtn);
5342 	return (0);
5343 free:
5344 	mcx_dmamem_free(sc, &mxm);
5345 	return (error);
5346 }
5347 
5348 static int
5349 mcx_destroy_rqt(struct mcx_softc *sc, int rqt)
5350 {
5351 	struct mcx_cmdq_entry *cqe;
5352 	struct mcx_cmd_destroy_rqt_in *in;
5353 	struct mcx_cmd_destroy_rqt_out *out;
5354 	int error;
5355 	int token;
5356 
5357 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5358 	token = mcx_cmdq_token(sc);
5359 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), token);
5360 
5361 	in = mcx_cmdq_in(cqe);
5362 	in->cmd_opcode = htobe16(MCX_CMD_DESTROY_RQT);
5363 	in->cmd_op_mod = htobe16(0);
5364 	in->cmd_rqtn = htobe32(rqt);
5365 
5366 	mcx_cmdq_post(sc, cqe, 0);
5367 	error = mcx_cmdq_poll(sc, cqe, 1000);
5368 	if (error != 0) {
5369 		printf("%s: destroy rqt timeout\n", DEVNAME(sc));
5370 		return error;
5371 	}
5372 	if (mcx_cmdq_verify(cqe) != 0) {
5373 		printf("%s: destroy rqt command corrupt\n", DEVNAME(sc));
5374 		return error;
5375 	}
5376 
5377 	out = mcx_cmdq_out(cqe);
5378 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5379 		printf("%s: destroy rqt failed (%x, %x)\n", DEVNAME(sc),
5380 		    out->cmd_status, betoh32(out->cmd_syndrome));
5381 		return -1;
5382 	}
5383 
5384 	return 0;
5385 }
5386 
5387 #if 0
5388 static int
5389 mcx_alloc_flow_counter(struct mcx_softc *sc, int i)
5390 {
5391 	struct mcx_cmdq_entry *cqe;
5392 	struct mcx_cmd_alloc_flow_counter_in *in;
5393 	struct mcx_cmd_alloc_flow_counter_out *out;
5394 	int error;
5395 
5396 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5397 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
5398 
5399 	in = mcx_cmdq_in(cqe);
5400 	in->cmd_opcode = htobe16(MCX_CMD_ALLOC_FLOW_COUNTER);
5401 	in->cmd_op_mod = htobe16(0);
5402 
5403 	mcx_cmdq_post(sc, cqe, 0);
5404 
5405 	error = mcx_cmdq_poll(sc, cqe, 1000);
5406 	if (error != 0) {
5407 		printf("%s: alloc flow counter timeout\n", DEVNAME(sc));
5408 		return (-1);
5409 	}
5410 	if (mcx_cmdq_verify(cqe) != 0) {
5411 		printf("%s: alloc flow counter command corrupt\n", DEVNAME(sc));
5412 		return (-1);
5413 	}
5414 
5415 	out = (struct mcx_cmd_alloc_flow_counter_out *)cqe->cq_output_data;
5416 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5417 		printf("%s: alloc flow counter failed (%x)\n", DEVNAME(sc),
5418 		    out->cmd_status);
5419 		return (-1);
5420 	}
5421 
5422 	sc->sc_flow_counter_id[i]  = betoh16(out->cmd_flow_counter_id);
5423 	printf("flow counter id %d = %d\n", i, sc->sc_flow_counter_id[i]);
5424 
5425 	return (0);
5426 }
5427 #endif
5428 
5429 static int
5430 mcx_create_flow_table(struct mcx_softc *sc, int log_size, int level,
5431     int *flow_table_id)
5432 {
5433 	struct mcx_cmdq_entry *cqe;
5434 	struct mcx_dmamem mxm;
5435 	struct mcx_cmd_create_flow_table_in *in;
5436 	struct mcx_cmd_create_flow_table_mb_in *mbin;
5437 	struct mcx_cmd_create_flow_table_out *out;
5438 	int error;
5439 	int token;
5440 
5441 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5442 	token = mcx_cmdq_token(sc);
5443 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
5444 	    sizeof(*out), token);
5445 
5446 	in = mcx_cmdq_in(cqe);
5447 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_FLOW_TABLE);
5448 	in->cmd_op_mod = htobe16(0);
5449 
5450 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
5451 	    &cqe->cq_input_ptr, token) != 0) {
5452 		printf("%s: unable to allocate create flow table mailbox\n",
5453 		    DEVNAME(sc));
5454 		return (-1);
5455 	}
5456 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5457 	mbin->cmd_table_type = MCX_FLOW_TABLE_TYPE_RX;
5458 	mbin->cmd_ctx.ft_log_size = log_size;
5459 	mbin->cmd_ctx.ft_level = level;
5460 
5461 	mcx_cmdq_mboxes_sign(&mxm, 1);
5462 	mcx_cmdq_post(sc, cqe, 0);
5463 	error = mcx_cmdq_poll(sc, cqe, 1000);
5464 	if (error != 0) {
5465 		printf("%s: create flow table timeout\n", DEVNAME(sc));
5466 		goto free;
5467 	}
5468 	if (mcx_cmdq_verify(cqe) != 0) {
5469 		printf("%s: create flow table command corrupt\n", DEVNAME(sc));
5470 		goto free;
5471 	}
5472 
5473 	out = mcx_cmdq_out(cqe);
5474 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5475 		printf("%s: create flow table failed (%x, %x)\n", DEVNAME(sc),
5476 		    out->cmd_status, betoh32(out->cmd_syndrome));
5477 		error = -1;
5478 		goto free;
5479 	}
5480 
5481 	*flow_table_id = mcx_get_id(out->cmd_table_id);
5482 free:
5483 	mcx_dmamem_free(sc, &mxm);
5484 	return (error);
5485 }
5486 
5487 static int
5488 mcx_set_flow_table_root(struct mcx_softc *sc, int flow_table_id)
5489 {
5490 	struct mcx_cmdq_entry *cqe;
5491 	struct mcx_dmamem mxm;
5492 	struct mcx_cmd_set_flow_table_root_in *in;
5493 	struct mcx_cmd_set_flow_table_root_mb_in *mbin;
5494 	struct mcx_cmd_set_flow_table_root_out *out;
5495 	int error;
5496 	int token;
5497 
5498 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5499 	token = mcx_cmdq_token(sc);
5500 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
5501 	    sizeof(*out), token);
5502 
5503 	in = mcx_cmdq_in(cqe);
5504 	in->cmd_opcode = htobe16(MCX_CMD_SET_FLOW_TABLE_ROOT);
5505 	in->cmd_op_mod = htobe16(0);
5506 
5507 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
5508 	    &cqe->cq_input_ptr, token) != 0) {
5509 		printf("%s: unable to allocate set flow table root mailbox\n",
5510 		    DEVNAME(sc));
5511 		return (-1);
5512 	}
5513 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5514 	mbin->cmd_table_type = MCX_FLOW_TABLE_TYPE_RX;
5515 	mbin->cmd_table_id = htobe32(flow_table_id);
5516 
5517 	mcx_cmdq_mboxes_sign(&mxm, 1);
5518 	mcx_cmdq_post(sc, cqe, 0);
5519 	error = mcx_cmdq_poll(sc, cqe, 1000);
5520 	if (error != 0) {
5521 		printf("%s: set flow table root timeout\n", DEVNAME(sc));
5522 		goto free;
5523 	}
5524 	if (mcx_cmdq_verify(cqe) != 0) {
5525 		printf("%s: set flow table root command corrupt\n",
5526 		    DEVNAME(sc));
5527 		goto free;
5528 	}
5529 
5530 	out = mcx_cmdq_out(cqe);
5531 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5532 		printf("%s: set flow table root failed (%x, %x)\n",
5533 		    DEVNAME(sc), out->cmd_status, betoh32(out->cmd_syndrome));
5534 		error = -1;
5535 		goto free;
5536 	}
5537 
5538 free:
5539 	mcx_dmamem_free(sc, &mxm);
5540 	return (error);
5541 }
5542 
5543 static int
5544 mcx_destroy_flow_table(struct mcx_softc *sc, int flow_table_id)
5545 {
5546 	struct mcx_cmdq_entry *cqe;
5547 	struct mcx_dmamem mxm;
5548 	struct mcx_cmd_destroy_flow_table_in *in;
5549 	struct mcx_cmd_destroy_flow_table_mb_in *mb;
5550 	struct mcx_cmd_destroy_flow_table_out *out;
5551 	int error;
5552 	int token;
5553 
5554 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5555 	token = mcx_cmdq_token(sc);
5556 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mb), sizeof(*out), token);
5557 
5558 	in = mcx_cmdq_in(cqe);
5559 	in->cmd_opcode = htobe16(MCX_CMD_DESTROY_FLOW_TABLE);
5560 	in->cmd_op_mod = htobe16(0);
5561 
5562 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
5563 	    &cqe->cq_input_ptr, token) != 0) {
5564 		printf("%s: unable to allocate destroy flow table mailbox\n",
5565 		    DEVNAME(sc));
5566 		return (-1);
5567 	}
5568 	mb = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5569 	mb->cmd_table_type = MCX_FLOW_TABLE_TYPE_RX;
5570 	mb->cmd_table_id = htobe32(flow_table_id);
5571 
5572 	mcx_cmdq_mboxes_sign(&mxm, 1);
5573 	mcx_cmdq_post(sc, cqe, 0);
5574 	error = mcx_cmdq_poll(sc, cqe, 1000);
5575 	if (error != 0) {
5576 		printf("%s: destroy flow table timeout\n", DEVNAME(sc));
5577 		goto free;
5578 	}
5579 	if (mcx_cmdq_verify(cqe) != 0) {
5580 		printf("%s: destroy flow table command corrupt\n",
5581 		    DEVNAME(sc));
5582 		goto free;
5583 	}
5584 
5585 	out = mcx_cmdq_out(cqe);
5586 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5587 		printf("%s: destroy flow table failed (%x, %x)\n", DEVNAME(sc),
5588 		    out->cmd_status, betoh32(out->cmd_syndrome));
5589 		error = -1;
5590 		goto free;
5591 	}
5592 
5593 free:
5594 	mcx_dmamem_free(sc, &mxm);
5595 	return (error);
5596 }
5597 
5598 
5599 static int
5600 mcx_create_flow_group(struct mcx_softc *sc, int flow_table_id, int group,
5601     int start, int size, int match_enable, struct mcx_flow_match *match)
5602 {
5603 	struct mcx_cmdq_entry *cqe;
5604 	struct mcx_dmamem mxm;
5605 	struct mcx_cmd_create_flow_group_in *in;
5606 	struct mcx_cmd_create_flow_group_mb_in *mbin;
5607 	struct mcx_cmd_create_flow_group_out *out;
5608 	struct mcx_flow_group *mfg;
5609 	int error;
5610 	int token;
5611 
5612 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5613 	token = mcx_cmdq_token(sc);
5614 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin), sizeof(*out),
5615 	    token);
5616 
5617 	in = mcx_cmdq_in(cqe);
5618 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_FLOW_GROUP);
5619 	in->cmd_op_mod = htobe16(0);
5620 
5621 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2, &cqe->cq_input_ptr, token)
5622 	    != 0) {
5623 		printf("%s: unable to allocate create flow group mailbox\n",
5624 		    DEVNAME(sc));
5625 		return (-1);
5626 	}
5627 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5628 	mbin->cmd_table_type = MCX_FLOW_TABLE_TYPE_RX;
5629 	mbin->cmd_table_id = htobe32(flow_table_id);
5630 	mbin->cmd_start_flow_index = htobe32(start);
5631 	mbin->cmd_end_flow_index = htobe32(start + (size - 1));
5632 
5633 	mbin->cmd_match_criteria_enable = match_enable;
5634 	memcpy(&mbin->cmd_match_criteria, match, sizeof(*match));
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: create flow group timeout\n", DEVNAME(sc));
5641 		goto free;
5642 	}
5643 	if (mcx_cmdq_verify(cqe) != 0) {
5644 		printf("%s: create flow group command corrupt\n", DEVNAME(sc));
5645 		goto free;
5646 	}
5647 
5648 	out = mcx_cmdq_out(cqe);
5649 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5650 		printf("%s: create flow group failed (%x, %x)\n", DEVNAME(sc),
5651 		    out->cmd_status, betoh32(out->cmd_syndrome));
5652 		error = -1;
5653 		goto free;
5654 	}
5655 
5656 	mfg = &sc->sc_flow_group[group];
5657 	mfg->g_id = mcx_get_id(out->cmd_group_id);
5658 	mfg->g_table = flow_table_id;
5659 	mfg->g_start = start;
5660 	mfg->g_size = size;
5661 
5662 free:
5663 	mcx_dmamem_free(sc, &mxm);
5664 	return (error);
5665 }
5666 
5667 static int
5668 mcx_destroy_flow_group(struct mcx_softc *sc, int group)
5669 {
5670 	struct mcx_cmdq_entry *cqe;
5671 	struct mcx_dmamem mxm;
5672 	struct mcx_cmd_destroy_flow_group_in *in;
5673 	struct mcx_cmd_destroy_flow_group_mb_in *mb;
5674 	struct mcx_cmd_destroy_flow_group_out *out;
5675 	struct mcx_flow_group *mfg;
5676 	int error;
5677 	int token;
5678 
5679 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5680 	token = mcx_cmdq_token(sc);
5681 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mb), sizeof(*out), token);
5682 
5683 	in = mcx_cmdq_in(cqe);
5684 	in->cmd_opcode = htobe16(MCX_CMD_DESTROY_FLOW_GROUP);
5685 	in->cmd_op_mod = htobe16(0);
5686 
5687 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2,
5688 	    &cqe->cq_input_ptr, token) != 0) {
5689 		printf("%s: unable to allocate destroy flow group mailbox\n",
5690 		    DEVNAME(sc));
5691 		return (-1);
5692 	}
5693 	mb = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5694 	mb->cmd_table_type = MCX_FLOW_TABLE_TYPE_RX;
5695 	mfg = &sc->sc_flow_group[group];
5696 	mb->cmd_table_id = htobe32(mfg->g_table);
5697 	mb->cmd_group_id = htobe32(mfg->g_id);
5698 
5699 	mcx_cmdq_mboxes_sign(&mxm, 2);
5700 	mcx_cmdq_post(sc, cqe, 0);
5701 	error = mcx_cmdq_poll(sc, cqe, 1000);
5702 	if (error != 0) {
5703 		printf("%s: destroy flow group timeout\n", DEVNAME(sc));
5704 		goto free;
5705 	}
5706 	if (mcx_cmdq_verify(cqe) != 0) {
5707 		printf("%s: destroy flow group command corrupt\n", DEVNAME(sc));
5708 		goto free;
5709 	}
5710 
5711 	out = mcx_cmdq_out(cqe);
5712 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5713 		printf("%s: destroy flow group failed (%x, %x)\n", DEVNAME(sc),
5714 		    out->cmd_status, betoh32(out->cmd_syndrome));
5715 		error = -1;
5716 		goto free;
5717 	}
5718 
5719 	mfg->g_id = -1;
5720 	mfg->g_table = -1;
5721 	mfg->g_size = 0;
5722 	mfg->g_start = 0;
5723 free:
5724 	mcx_dmamem_free(sc, &mxm);
5725 	return (error);
5726 }
5727 
5728 static int
5729 mcx_set_flow_table_entry_mac(struct mcx_softc *sc, int group, int index,
5730     uint8_t *macaddr, uint32_t dest)
5731 {
5732 	struct mcx_cmdq_entry *cqe;
5733 	struct mcx_dmamem mxm;
5734 	struct mcx_cmd_set_flow_table_entry_in *in;
5735 	struct mcx_cmd_set_flow_table_entry_mb_in *mbin;
5736 	struct mcx_cmd_set_flow_table_entry_out *out;
5737 	struct mcx_flow_group *mfg;
5738 	uint32_t *pdest;
5739 	int error;
5740 	int token;
5741 
5742 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5743 	token = mcx_cmdq_token(sc);
5744 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin) + sizeof(*pdest),
5745 	    sizeof(*out), token);
5746 
5747 	in = mcx_cmdq_in(cqe);
5748 	in->cmd_opcode = htobe16(MCX_CMD_SET_FLOW_TABLE_ENTRY);
5749 	in->cmd_op_mod = htobe16(0);
5750 
5751 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2, &cqe->cq_input_ptr, token)
5752 	    != 0) {
5753 		printf("%s: unable to allocate set flow table entry mailbox\n",
5754 		    DEVNAME(sc));
5755 		return (-1);
5756 	}
5757 
5758 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5759 	mbin->cmd_table_type = MCX_FLOW_TABLE_TYPE_RX;
5760 
5761 	mfg = &sc->sc_flow_group[group];
5762 	mbin->cmd_table_id = htobe32(mfg->g_table);
5763 	mbin->cmd_flow_index = htobe32(mfg->g_start + index);
5764 	mbin->cmd_flow_ctx.fc_group_id = htobe32(mfg->g_id);
5765 
5766 	/* flow context ends at offset 0x330, 0x130 into the second mbox */
5767 	pdest = (uint32_t *)
5768 	    (((char *)mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 1))) + 0x130);
5769 	mbin->cmd_flow_ctx.fc_action = htobe32(MCX_FLOW_CONTEXT_ACTION_FORWARD);
5770 	mbin->cmd_flow_ctx.fc_dest_list_size = htobe32(1);
5771 	*pdest = htobe32(dest);
5772 
5773 	/* the only thing we match on at the moment is the dest mac address */
5774 	if (macaddr != NULL) {
5775 		memcpy(mbin->cmd_flow_ctx.fc_match_value.mc_dest_mac, macaddr,
5776 		    ETHER_ADDR_LEN);
5777 	}
5778 
5779 	mcx_cmdq_mboxes_sign(&mxm, 2);
5780 	mcx_cmdq_post(sc, cqe, 0);
5781 	error = mcx_cmdq_poll(sc, cqe, 1000);
5782 	if (error != 0) {
5783 		printf("%s: set flow table entry timeout\n", DEVNAME(sc));
5784 		goto free;
5785 	}
5786 	if (mcx_cmdq_verify(cqe) != 0) {
5787 		printf("%s: set flow table entry command corrupt\n",
5788 		    DEVNAME(sc));
5789 		goto free;
5790 	}
5791 
5792 	out = mcx_cmdq_out(cqe);
5793 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5794 		printf("%s: set flow table entry failed (%x, %x)\n",
5795 		    DEVNAME(sc), out->cmd_status, betoh32(out->cmd_syndrome));
5796 		error = -1;
5797 		goto free;
5798 	}
5799 
5800 free:
5801 	mcx_dmamem_free(sc, &mxm);
5802 	return (error);
5803 }
5804 
5805 static int
5806 mcx_set_flow_table_entry_proto(struct mcx_softc *sc, int group, int index,
5807     int ethertype, int ip_proto, uint32_t dest)
5808 {
5809 	struct mcx_cmdq_entry *cqe;
5810 	struct mcx_dmamem mxm;
5811 	struct mcx_cmd_set_flow_table_entry_in *in;
5812 	struct mcx_cmd_set_flow_table_entry_mb_in *mbin;
5813 	struct mcx_cmd_set_flow_table_entry_out *out;
5814 	struct mcx_flow_group *mfg;
5815 	uint32_t *pdest;
5816 	int error;
5817 	int token;
5818 
5819 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5820 	token = mcx_cmdq_token(sc);
5821 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin) + sizeof(*pdest),
5822 	    sizeof(*out), token);
5823 
5824 	in = mcx_cmdq_in(cqe);
5825 	in->cmd_opcode = htobe16(MCX_CMD_SET_FLOW_TABLE_ENTRY);
5826 	in->cmd_op_mod = htobe16(0);
5827 
5828 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2, &cqe->cq_input_ptr, token)
5829 	    != 0) {
5830 		printf("%s: unable to allocate set flow table entry mailbox\n",
5831 		    DEVNAME(sc));
5832 		return (-1);
5833 	}
5834 
5835 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5836 	mbin->cmd_table_type = MCX_FLOW_TABLE_TYPE_RX;
5837 
5838 	mfg = &sc->sc_flow_group[group];
5839 	mbin->cmd_table_id = htobe32(mfg->g_table);
5840 	mbin->cmd_flow_index = htobe32(mfg->g_start + index);
5841 	mbin->cmd_flow_ctx.fc_group_id = htobe32(mfg->g_id);
5842 
5843 	/* flow context ends at offset 0x330, 0x130 into the second mbox */
5844 	pdest = (uint32_t *)
5845 	    (((char *)mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 1))) + 0x130);
5846 	mbin->cmd_flow_ctx.fc_action = htobe32(MCX_FLOW_CONTEXT_ACTION_FORWARD);
5847 	mbin->cmd_flow_ctx.fc_dest_list_size = htobe32(1);
5848 	*pdest = htobe32(dest);
5849 
5850 	mbin->cmd_flow_ctx.fc_match_value.mc_ethertype = htobe16(ethertype);
5851 	mbin->cmd_flow_ctx.fc_match_value.mc_ip_proto = ip_proto;
5852 
5853 	mcx_cmdq_mboxes_sign(&mxm, 2);
5854 	mcx_cmdq_post(sc, cqe, 0);
5855 	error = mcx_cmdq_poll(sc, cqe, 1000);
5856 	if (error != 0) {
5857 		printf("%s: set flow table entry timeout\n", DEVNAME(sc));
5858 		goto free;
5859 	}
5860 	if (mcx_cmdq_verify(cqe) != 0) {
5861 		printf("%s: set flow table entry command corrupt\n",
5862 		    DEVNAME(sc));
5863 		goto free;
5864 	}
5865 
5866 	out = mcx_cmdq_out(cqe);
5867 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5868 		printf("%s: set flow table entry failed (%x, %x)\n",
5869 		    DEVNAME(sc), out->cmd_status, betoh32(out->cmd_syndrome));
5870 		error = -1;
5871 		goto free;
5872 	}
5873 
5874 free:
5875 	mcx_dmamem_free(sc, &mxm);
5876 	return (error);
5877 }
5878 
5879 static int
5880 mcx_delete_flow_table_entry(struct mcx_softc *sc, int group, int index)
5881 {
5882 	struct mcx_cmdq_entry *cqe;
5883 	struct mcx_dmamem mxm;
5884 	struct mcx_cmd_delete_flow_table_entry_in *in;
5885 	struct mcx_cmd_delete_flow_table_entry_mb_in *mbin;
5886 	struct mcx_cmd_delete_flow_table_entry_out *out;
5887 	struct mcx_flow_group *mfg;
5888 	int error;
5889 	int token;
5890 
5891 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5892 	token = mcx_cmdq_token(sc);
5893 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin), sizeof(*out),
5894 	    token);
5895 
5896 	in = mcx_cmdq_in(cqe);
5897 	in->cmd_opcode = htobe16(MCX_CMD_DELETE_FLOW_TABLE_ENTRY);
5898 	in->cmd_op_mod = htobe16(0);
5899 
5900 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2,
5901 	    &cqe->cq_input_ptr, token) != 0) {
5902 		printf("%s: unable to allocate "
5903 		    "delete flow table entry mailbox\n", DEVNAME(sc));
5904 		return (-1);
5905 	}
5906 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5907 	mbin->cmd_table_type = MCX_FLOW_TABLE_TYPE_RX;
5908 
5909 	mfg = &sc->sc_flow_group[group];
5910 	mbin->cmd_table_id = htobe32(mfg->g_table);
5911 	mbin->cmd_flow_index = htobe32(mfg->g_start + index);
5912 
5913 	mcx_cmdq_mboxes_sign(&mxm, 2);
5914 	mcx_cmdq_post(sc, cqe, 0);
5915 	error = mcx_cmdq_poll(sc, cqe, 1000);
5916 	if (error != 0) {
5917 		printf("%s: delete flow table entry timeout\n", DEVNAME(sc));
5918 		goto free;
5919 	}
5920 	if (mcx_cmdq_verify(cqe) != 0) {
5921 		printf("%s: delete flow table entry command corrupt\n",
5922 		    DEVNAME(sc));
5923 		goto free;
5924 	}
5925 
5926 	out = mcx_cmdq_out(cqe);
5927 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5928 		printf("%s: delete flow table entry %d:%d failed (%x, %x)\n",
5929 		    DEVNAME(sc), group, index, out->cmd_status,
5930 		    betoh32(out->cmd_syndrome));
5931 		error = -1;
5932 		goto free;
5933 	}
5934 
5935 free:
5936 	mcx_dmamem_free(sc, &mxm);
5937 	return (error);
5938 }
5939 
5940 #if 0
5941 int
5942 mcx_dump_flow_table(struct mcx_softc *sc, int flow_table_id)
5943 {
5944 	struct mcx_dmamem mxm;
5945 	struct mcx_cmdq_entry *cqe;
5946 	struct mcx_cmd_query_flow_table_in *in;
5947 	struct mcx_cmd_query_flow_table_mb_in *mbin;
5948 	struct mcx_cmd_query_flow_table_out *out;
5949 	struct mcx_cmd_query_flow_table_mb_out *mbout;
5950 	uint8_t token = mcx_cmdq_token(sc);
5951 	int error;
5952 	int i;
5953 	uint8_t *dump;
5954 
5955 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5956 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
5957 	    sizeof(*out) + sizeof(*mbout) + 16, token);
5958 
5959 	in = mcx_cmdq_in(cqe);
5960 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_FLOW_TABLE);
5961 	in->cmd_op_mod = htobe16(0);
5962 
5963 	CTASSERT(sizeof(*mbin) <= MCX_CMDQ_MAILBOX_DATASIZE);
5964 	CTASSERT(sizeof(*mbout) <= MCX_CMDQ_MAILBOX_DATASIZE);
5965 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2,
5966 	    &cqe->cq_output_ptr, token) != 0) {
5967 		printf(", unable to allocate query flow table mailboxes\n");
5968 		return (-1);
5969 	}
5970 	cqe->cq_input_ptr = cqe->cq_output_ptr;
5971 
5972 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5973 	mbin->cmd_table_type = 0;
5974 	mbin->cmd_table_id = htobe32(flow_table_id);
5975 
5976 	mcx_cmdq_mboxes_sign(&mxm, 1);
5977 
5978 	mcx_cmdq_post(sc, cqe, 0);
5979 	error = mcx_cmdq_poll(sc, cqe, 1000);
5980 	if (error != 0) {
5981 		printf("%s: query flow table timeout\n", DEVNAME(sc));
5982 		goto free;
5983 	}
5984 	error = mcx_cmdq_verify(cqe);
5985 	if (error != 0) {
5986 		printf("%s: query flow table reply corrupt\n", DEVNAME(sc));
5987 		goto free;
5988 	}
5989 
5990 	out = mcx_cmdq_out(cqe);
5991 	switch (out->cmd_status) {
5992 	case MCX_CQ_STATUS_OK:
5993 		break;
5994 	default:
5995 		printf("%s: query flow table failed (%x/%x)\n", DEVNAME(sc),
5996 		    out->cmd_status, betoh32(out->cmd_syndrome));
5997 		error = -1;
5998 		goto free;
5999 	}
6000 
6001         mbout = (struct mcx_cmd_query_flow_table_mb_out *)
6002 	    (mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)));
6003 	dump = (uint8_t *)mbout + 8;
6004 	for (i = 0; i < sizeof(struct mcx_flow_table_ctx); i++) {
6005 		printf("%.2x ", dump[i]);
6006 		if (i % 16 == 15)
6007 			printf("\n");
6008 	}
6009 free:
6010 	mcx_cq_mboxes_free(sc, &mxm);
6011 	return (error);
6012 }
6013 int
6014 mcx_dump_flow_table_entry(struct mcx_softc *sc, int flow_table_id, int index)
6015 {
6016 	struct mcx_dmamem mxm;
6017 	struct mcx_cmdq_entry *cqe;
6018 	struct mcx_cmd_query_flow_table_entry_in *in;
6019 	struct mcx_cmd_query_flow_table_entry_mb_in *mbin;
6020 	struct mcx_cmd_query_flow_table_entry_out *out;
6021 	struct mcx_cmd_query_flow_table_entry_mb_out *mbout;
6022 	uint8_t token = mcx_cmdq_token(sc);
6023 	int error;
6024 	int i;
6025 	uint8_t *dump;
6026 
6027 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
6028 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
6029 	    sizeof(*out) + sizeof(*mbout) + 16, token);
6030 
6031 	in = mcx_cmdq_in(cqe);
6032 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_FLOW_TABLE_ENTRY);
6033 	in->cmd_op_mod = htobe16(0);
6034 
6035 	CTASSERT(sizeof(*mbin) <= MCX_CMDQ_MAILBOX_DATASIZE);
6036 	CTASSERT(sizeof(*mbout) <= MCX_CMDQ_MAILBOX_DATASIZE*2);
6037 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2,
6038 	    &cqe->cq_output_ptr, token) != 0) {
6039 		printf(", unable to allocate "
6040 		    "query flow table entry mailboxes\n");
6041 		return (-1);
6042 	}
6043 	cqe->cq_input_ptr = cqe->cq_output_ptr;
6044 
6045 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
6046 	mbin->cmd_table_type = 0;
6047 	mbin->cmd_table_id = htobe32(flow_table_id);
6048 	mbin->cmd_flow_index = htobe32(index);
6049 
6050 	mcx_cmdq_mboxes_sign(&mxm, 1);
6051 
6052 	mcx_cmdq_post(sc, cqe, 0);
6053 	error = mcx_cmdq_poll(sc, cqe, 1000);
6054 	if (error != 0) {
6055 		printf("%s: query flow table entry timeout\n", DEVNAME(sc));
6056 		goto free;
6057 	}
6058 	error = mcx_cmdq_verify(cqe);
6059 	if (error != 0) {
6060 		printf("%s: query flow table entry reply corrupt\n",
6061 		    DEVNAME(sc));
6062 		goto free;
6063 	}
6064 
6065 	out = mcx_cmdq_out(cqe);
6066 	switch (out->cmd_status) {
6067 	case MCX_CQ_STATUS_OK:
6068 		break;
6069 	default:
6070 		printf("%s: query flow table entry failed (%x/%x)\n",
6071 		    DEVNAME(sc), out->cmd_status, betoh32(out->cmd_syndrome));
6072 		error = -1;
6073 		goto free;
6074 	}
6075 
6076         mbout = (struct mcx_cmd_query_flow_table_entry_mb_out *)
6077 	    (mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)));
6078 	dump = (uint8_t *)mbout;
6079 	for (i = 0; i < MCX_CMDQ_MAILBOX_DATASIZE; i++) {
6080 		printf("%.2x ", dump[i]);
6081 		if (i % 16 == 15)
6082 			printf("\n");
6083 	}
6084 
6085 free:
6086 	mcx_cq_mboxes_free(sc, &mxm);
6087 	return (error);
6088 }
6089 
6090 int
6091 mcx_dump_flow_group(struct mcx_softc *sc, int flow_table_id)
6092 {
6093 	struct mcx_dmamem mxm;
6094 	struct mcx_cmdq_entry *cqe;
6095 	struct mcx_cmd_query_flow_group_in *in;
6096 	struct mcx_cmd_query_flow_group_mb_in *mbin;
6097 	struct mcx_cmd_query_flow_group_out *out;
6098 	struct mcx_cmd_query_flow_group_mb_out *mbout;
6099 	uint8_t token = mcx_cmdq_token(sc);
6100 	int error;
6101 	int i;
6102 	uint8_t *dump;
6103 
6104 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
6105 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
6106 	    sizeof(*out) + sizeof(*mbout) + 16, token);
6107 
6108 	in = mcx_cmdq_in(cqe);
6109 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_FLOW_GROUP);
6110 	in->cmd_op_mod = htobe16(0);
6111 
6112 	CTASSERT(sizeof(*mbin) <= MCX_CMDQ_MAILBOX_DATASIZE);
6113 	CTASSERT(sizeof(*mbout) <= MCX_CMDQ_MAILBOX_DATASIZE*2);
6114 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2,
6115 	    &cqe->cq_output_ptr, token) != 0) {
6116 		printf(", unable to allocate query flow group mailboxes\n");
6117 		return (-1);
6118 	}
6119 	cqe->cq_input_ptr = cqe->cq_output_ptr;
6120 
6121 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
6122 	mbin->cmd_table_type = 0;
6123 	mbin->cmd_table_id = htobe32(flow_table_id);
6124 	mbin->cmd_group_id = htobe32(sc->sc_flow_group_id);
6125 
6126 	mcx_cmdq_mboxes_sign(&mxm, 1);
6127 
6128 	mcx_cmdq_post(sc, cqe, 0);
6129 	error = mcx_cmdq_poll(sc, cqe, 1000);
6130 	if (error != 0) {
6131 		printf("%s: query flow group timeout\n", DEVNAME(sc));
6132 		goto free;
6133 	}
6134 	error = mcx_cmdq_verify(cqe);
6135 	if (error != 0) {
6136 		printf("%s: query flow group reply corrupt\n", DEVNAME(sc));
6137 		goto free;
6138 	}
6139 
6140 	out = mcx_cmdq_out(cqe);
6141 	switch (out->cmd_status) {
6142 	case MCX_CQ_STATUS_OK:
6143 		break;
6144 	default:
6145 		printf("%s: query flow group failed (%x/%x)\n", DEVNAME(sc),
6146 		    out->cmd_status, betoh32(out->cmd_syndrome));
6147 		error = -1;
6148 		goto free;
6149 	}
6150 
6151         mbout = (struct mcx_cmd_query_flow_group_mb_out *)
6152 	    (mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)));
6153 	dump = (uint8_t *)mbout;
6154 	for (i = 0; i < MCX_CMDQ_MAILBOX_DATASIZE; i++) {
6155 		printf("%.2x ", dump[i]);
6156 		if (i % 16 == 15)
6157 			printf("\n");
6158 	}
6159 	dump = (uint8_t *)(mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 1)));
6160 	for (i = 0; i < MCX_CMDQ_MAILBOX_DATASIZE; i++) {
6161 		printf("%.2x ", dump[i]);
6162 		if (i % 16 == 15)
6163 			printf("\n");
6164 	}
6165 
6166 free:
6167 	mcx_cq_mboxes_free(sc, &mxm);
6168 	return (error);
6169 }
6170 
6171 static int
6172 mcx_dump_counters(struct mcx_softc *sc)
6173 {
6174 	struct mcx_dmamem mxm;
6175 	struct mcx_cmdq_entry *cqe;
6176 	struct mcx_cmd_query_vport_counters_in *in;
6177 	struct mcx_cmd_query_vport_counters_mb_in *mbin;
6178 	struct mcx_cmd_query_vport_counters_out *out;
6179 	struct mcx_nic_vport_counters *counters;
6180 	int error, token;
6181 
6182 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
6183 	token = mcx_cmdq_token(sc);
6184 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
6185 	    sizeof(*out) + sizeof(*counters), token);
6186 
6187 	in = mcx_cmdq_in(cqe);
6188 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_VPORT_COUNTERS);
6189 	in->cmd_op_mod = htobe16(0);
6190 
6191 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
6192 	    &cqe->cq_output_ptr, token) != 0) {
6193 		printf(", unable to allocate "
6194 		    "query nic vport counters mailboxen\n");
6195 		return (-1);
6196 	}
6197 	cqe->cq_input_ptr = cqe->cq_output_ptr;
6198 
6199 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
6200 	mbin->cmd_clear = 0x80;
6201 
6202 	mcx_cmdq_mboxes_sign(&mxm, 1);
6203 	mcx_cmdq_post(sc, cqe, 0);
6204 
6205 	error = mcx_cmdq_poll(sc, cqe, 1000);
6206 	if (error != 0) {
6207 		printf("%s: query nic vport counters timeout\n", DEVNAME(sc));
6208 		goto free;
6209 	}
6210 	if (mcx_cmdq_verify(cqe) != 0) {
6211 		printf("%s: query nic vport counters command corrupt\n",
6212 		    DEVNAME(sc));
6213 		goto free;
6214 	}
6215 
6216 	out = mcx_cmdq_out(cqe);
6217 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
6218 		printf("%s: query nic vport counters failed (%x, %x)\n",
6219 		    DEVNAME(sc), out->cmd_status, betoh32(out->cmd_syndrome));
6220 		error = -1;
6221 		goto free;
6222 	}
6223 
6224 	counters = (struct mcx_nic_vport_counters *)
6225 	    (mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)));
6226 	if (counters->rx_bcast.packets + counters->tx_bcast.packets +
6227 	    counters->rx_ucast.packets + counters->tx_ucast.packets +
6228 	    counters->rx_err.packets + counters->tx_err.packets)
6229 		printf("%s: err %llx/%llx uc %llx/%llx bc %llx/%llx\n",
6230 		    DEVNAME(sc),
6231 		    betoh64(counters->tx_err.packets),
6232 		    betoh64(counters->rx_err.packets),
6233 		    betoh64(counters->tx_ucast.packets),
6234 		    betoh64(counters->rx_ucast.packets),
6235 		    betoh64(counters->tx_bcast.packets),
6236 		    betoh64(counters->rx_bcast.packets));
6237 free:
6238 	mcx_dmamem_free(sc, &mxm);
6239 
6240 	return (error);
6241 }
6242 
6243 static int
6244 mcx_dump_flow_counter(struct mcx_softc *sc, int index, const char *what)
6245 {
6246 	struct mcx_dmamem mxm;
6247 	struct mcx_cmdq_entry *cqe;
6248 	struct mcx_cmd_query_flow_counter_in *in;
6249 	struct mcx_cmd_query_flow_counter_mb_in *mbin;
6250 	struct mcx_cmd_query_flow_counter_out *out;
6251 	struct mcx_counter *counters;
6252 	int error, token;
6253 
6254 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
6255 	token = mcx_cmdq_token(sc);
6256 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin), sizeof(*out) +
6257 	    sizeof(*counters), token);
6258 
6259 	in = mcx_cmdq_in(cqe);
6260 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_FLOW_COUNTER);
6261 	in->cmd_op_mod = htobe16(0);
6262 
6263 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
6264 	    &cqe->cq_output_ptr, token) != 0) {
6265 		printf(", unable to allocate query flow counter mailboxen\n");
6266 		return (-1);
6267 	}
6268 	cqe->cq_input_ptr = cqe->cq_output_ptr;
6269 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
6270 	mbin->cmd_flow_counter_id = htobe16(sc->sc_flow_counter_id[index]);
6271 	mbin->cmd_clear = 0x80;
6272 
6273 	mcx_cmdq_mboxes_sign(&mxm, 1);
6274 	mcx_cmdq_post(sc, cqe, 0);
6275 
6276 	error = mcx_cmdq_poll(sc, cqe, 1000);
6277 	if (error != 0) {
6278 		printf("%s: query flow counter timeout\n", DEVNAME(sc));
6279 		goto free;
6280 	}
6281 	if (mcx_cmdq_verify(cqe) != 0) {
6282 		printf("%s: query flow counter command corrupt\n", DEVNAME(sc));
6283 		goto free;
6284 	}
6285 
6286 	out = mcx_cmdq_out(cqe);
6287 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
6288 		printf("%s: query flow counter failed (%x, %x)\n", DEVNAME(sc),
6289 		    out->cmd_status, betoh32(out->cmd_syndrome));
6290 		error = -1;
6291 		goto free;
6292 	}
6293 
6294 	counters = (struct mcx_counter *)
6295 	    (mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)));
6296 	if (counters->packets)
6297 		printf("%s: %s inflow %llx\n", DEVNAME(sc), what,
6298 		    betoh64(counters->packets));
6299 free:
6300 	mcx_dmamem_free(sc, &mxm);
6301 
6302 	return (error);
6303 }
6304 
6305 #endif
6306 
6307 #if NKSTAT > 0
6308 
6309 int
6310 mcx_query_rq(struct mcx_softc *sc, struct mcx_rx *rx, struct mcx_rq_ctx *rq_ctx)
6311 {
6312 	struct mcx_dmamem mxm;
6313 	struct mcx_cmdq_entry *cqe;
6314 	struct mcx_cmd_query_rq_in *in;
6315 	struct mcx_cmd_query_rq_out *out;
6316 	struct mcx_cmd_query_rq_mb_out *mbout;
6317 	uint8_t token = mcx_cmdq_token(sc);
6318 	int error;
6319 
6320 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
6321 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out) + sizeof(*mbout) + 16,
6322 	    token);
6323 
6324 	in = mcx_cmdq_in(cqe);
6325 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_RQ);
6326 	in->cmd_op_mod = htobe16(0);
6327 	in->cmd_rqn = htobe32(rx->rx_rqn);
6328 
6329 	CTASSERT(sizeof(*mbout) <= MCX_CMDQ_MAILBOX_DATASIZE*2);
6330 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2,
6331 	    &cqe->cq_output_ptr, token) != 0) {
6332 		printf("%s: unable to allocate query rq mailboxes\n", DEVNAME(sc));
6333 		return (-1);
6334 	}
6335 
6336 	mcx_cmdq_mboxes_sign(&mxm, 1);
6337 
6338 	mcx_cmdq_post(sc, cqe, 0);
6339 	error = mcx_cmdq_poll(sc, cqe, 1000);
6340 	if (error != 0) {
6341 		printf("%s: query rq timeout\n", DEVNAME(sc));
6342 		goto free;
6343 	}
6344 	error = mcx_cmdq_verify(cqe);
6345 	if (error != 0) {
6346 		printf("%s: query rq reply corrupt\n", DEVNAME(sc));
6347 		goto free;
6348 	}
6349 
6350 	out = mcx_cmdq_out(cqe);
6351 	switch (out->cmd_status) {
6352 	case MCX_CQ_STATUS_OK:
6353 		break;
6354 	default:
6355 		printf("%s: query rq failed (%x/%x)\n", DEVNAME(sc),
6356 		    out->cmd_status, betoh32(out->cmd_syndrome));
6357 		error = -1;
6358 		goto free;
6359 	}
6360 
6361         mbout = (struct mcx_cmd_query_rq_mb_out *)
6362 	    (mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)));
6363 	memcpy(rq_ctx, &mbout->cmd_ctx, sizeof(*rq_ctx));
6364 
6365 free:
6366 	mcx_cq_mboxes_free(sc, &mxm);
6367 	return (error);
6368 }
6369 
6370 int
6371 mcx_query_sq(struct mcx_softc *sc, struct mcx_tx *tx, struct mcx_sq_ctx *sq_ctx)
6372 {
6373 	struct mcx_dmamem mxm;
6374 	struct mcx_cmdq_entry *cqe;
6375 	struct mcx_cmd_query_sq_in *in;
6376 	struct mcx_cmd_query_sq_out *out;
6377 	struct mcx_cmd_query_sq_mb_out *mbout;
6378 	uint8_t token = mcx_cmdq_token(sc);
6379 	int error;
6380 
6381 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
6382 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out) + sizeof(*mbout) + 16,
6383 	    token);
6384 
6385 	in = mcx_cmdq_in(cqe);
6386 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_SQ);
6387 	in->cmd_op_mod = htobe16(0);
6388 	in->cmd_sqn = htobe32(tx->tx_sqn);
6389 
6390 	CTASSERT(sizeof(*mbout) <= MCX_CMDQ_MAILBOX_DATASIZE*2);
6391 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2,
6392 	    &cqe->cq_output_ptr, token) != 0) {
6393 		printf("%s: unable to allocate query sq mailboxes\n", DEVNAME(sc));
6394 		return (-1);
6395 	}
6396 
6397 	mcx_cmdq_mboxes_sign(&mxm, 1);
6398 
6399 	mcx_cmdq_post(sc, cqe, 0);
6400 	error = mcx_cmdq_poll(sc, cqe, 1000);
6401 	if (error != 0) {
6402 		printf("%s: query sq timeout\n", DEVNAME(sc));
6403 		goto free;
6404 	}
6405 	error = mcx_cmdq_verify(cqe);
6406 	if (error != 0) {
6407 		printf("%s: query sq reply corrupt\n", DEVNAME(sc));
6408 		goto free;
6409 	}
6410 
6411 	out = mcx_cmdq_out(cqe);
6412 	switch (out->cmd_status) {
6413 	case MCX_CQ_STATUS_OK:
6414 		break;
6415 	default:
6416 		printf("%s: query sq failed (%x/%x)\n", DEVNAME(sc),
6417 		    out->cmd_status, betoh32(out->cmd_syndrome));
6418 		error = -1;
6419 		goto free;
6420 	}
6421 
6422         mbout = (struct mcx_cmd_query_sq_mb_out *)
6423 	    (mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)));
6424 	memcpy(sq_ctx, &mbout->cmd_ctx, sizeof(*sq_ctx));
6425 
6426 free:
6427 	mcx_cq_mboxes_free(sc, &mxm);
6428 	return (error);
6429 }
6430 
6431 int
6432 mcx_query_cq(struct mcx_softc *sc, struct mcx_cq *cq, struct mcx_cq_ctx *cq_ctx)
6433 {
6434 	struct mcx_dmamem mxm;
6435 	struct mcx_cmdq_entry *cqe;
6436 	struct mcx_cmd_query_cq_in *in;
6437 	struct mcx_cmd_query_cq_out *out;
6438 	struct mcx_cq_ctx *ctx;
6439 	uint8_t token = mcx_cmdq_token(sc);
6440 	int error;
6441 
6442 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
6443 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out) + sizeof(*ctx) + 16,
6444 	    token);
6445 
6446 	in = mcx_cmdq_in(cqe);
6447 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_CQ);
6448 	in->cmd_op_mod = htobe16(0);
6449 	in->cmd_cqn = htobe32(cq->cq_n);
6450 
6451 	CTASSERT(sizeof(*ctx) <= MCX_CMDQ_MAILBOX_DATASIZE*2);
6452 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2,
6453 	    &cqe->cq_output_ptr, token) != 0) {
6454 		printf("%s: unable to allocate query cq mailboxes\n", DEVNAME(sc));
6455 		return (-1);
6456 	}
6457 
6458 	mcx_cmdq_mboxes_sign(&mxm, 1);
6459 
6460 	mcx_cmdq_post(sc, cqe, 0);
6461 	error = mcx_cmdq_poll(sc, cqe, 1000);
6462 	if (error != 0) {
6463 		printf("%s: query cq timeout\n", DEVNAME(sc));
6464 		goto free;
6465 	}
6466 	error = mcx_cmdq_verify(cqe);
6467 	if (error != 0) {
6468 		printf("%s: query cq reply corrupt\n", DEVNAME(sc));
6469 		goto free;
6470 	}
6471 
6472 	out = mcx_cmdq_out(cqe);
6473 	switch (out->cmd_status) {
6474 	case MCX_CQ_STATUS_OK:
6475 		break;
6476 	default:
6477 		printf("%s: query cq failed (%x/%x)\n", DEVNAME(sc),
6478 		    out->cmd_status, betoh32(out->cmd_syndrome));
6479 		error = -1;
6480 		goto free;
6481 	}
6482 
6483         ctx = (struct mcx_cq_ctx *)(mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)));
6484 	memcpy(cq_ctx, ctx, sizeof(*cq_ctx));
6485 free:
6486 	mcx_cq_mboxes_free(sc, &mxm);
6487 	return (error);
6488 }
6489 
6490 int
6491 mcx_query_eq(struct mcx_softc *sc, struct mcx_eq *eq, struct mcx_eq_ctx *eq_ctx)
6492 {
6493 	struct mcx_dmamem mxm;
6494 	struct mcx_cmdq_entry *cqe;
6495 	struct mcx_cmd_query_eq_in *in;
6496 	struct mcx_cmd_query_eq_out *out;
6497 	struct mcx_eq_ctx *ctx;
6498 	uint8_t token = mcx_cmdq_token(sc);
6499 	int error;
6500 
6501 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
6502 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out) + sizeof(*ctx) + 16,
6503 	    token);
6504 
6505 	in = mcx_cmdq_in(cqe);
6506 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_EQ);
6507 	in->cmd_op_mod = htobe16(0);
6508 	in->cmd_eqn = htobe32(eq->eq_n);
6509 
6510 	CTASSERT(sizeof(*ctx) <= MCX_CMDQ_MAILBOX_DATASIZE*2);
6511 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2,
6512 	    &cqe->cq_output_ptr, token) != 0) {
6513 		printf("%s: unable to allocate query eq mailboxes\n", DEVNAME(sc));
6514 		return (-1);
6515 	}
6516 
6517 	mcx_cmdq_mboxes_sign(&mxm, 1);
6518 
6519 	mcx_cmdq_post(sc, cqe, 0);
6520 	error = mcx_cmdq_poll(sc, cqe, 1000);
6521 	if (error != 0) {
6522 		printf("%s: query eq timeout\n", DEVNAME(sc));
6523 		goto free;
6524 	}
6525 	error = mcx_cmdq_verify(cqe);
6526 	if (error != 0) {
6527 		printf("%s: query eq reply corrupt\n", DEVNAME(sc));
6528 		goto free;
6529 	}
6530 
6531 	out = mcx_cmdq_out(cqe);
6532 	switch (out->cmd_status) {
6533 	case MCX_CQ_STATUS_OK:
6534 		break;
6535 	default:
6536 		printf("%s: query eq failed (%x/%x)\n", DEVNAME(sc),
6537 		    out->cmd_status, betoh32(out->cmd_syndrome));
6538 		error = -1;
6539 		goto free;
6540 	}
6541 
6542         ctx = (struct mcx_eq_ctx *)(mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)));
6543 	memcpy(eq_ctx, ctx, sizeof(*eq_ctx));
6544 free:
6545 	mcx_cq_mboxes_free(sc, &mxm);
6546 	return (error);
6547 }
6548 
6549 #endif /* NKSTAT > 0 */
6550 
6551 static inline unsigned int
6552 mcx_rx_fill_slots(struct mcx_softc *sc, struct mcx_rx *rx, uint nslots)
6553 {
6554 	struct mcx_rq_entry *ring, *rqe;
6555 	struct mcx_slot *ms;
6556 	struct mbuf *m;
6557 	uint slot, p, fills;
6558 
6559 	ring = MCX_DMA_KVA(&rx->rx_rq_mem);
6560 	p = rx->rx_prod;
6561 
6562 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&rx->rx_rq_mem),
6563 	    0, MCX_DMA_LEN(&rx->rx_rq_mem), BUS_DMASYNC_POSTWRITE);
6564 
6565 	for (fills = 0; fills < nslots; fills++) {
6566 		slot = p % (1 << MCX_LOG_RQ_SIZE);
6567 
6568 		ms = &rx->rx_slots[slot];
6569 		rqe = &ring[slot];
6570 
6571 		m = MCLGETL(NULL, M_DONTWAIT, sc->sc_rxbufsz);
6572 		if (m == NULL)
6573 			break;
6574 
6575 		m->m_data += (m->m_ext.ext_size - sc->sc_rxbufsz);
6576 		m->m_data += ETHER_ALIGN;
6577 		m->m_len = m->m_pkthdr.len = sc->sc_hardmtu;
6578 
6579 		if (bus_dmamap_load_mbuf(sc->sc_dmat, ms->ms_map, m,
6580 		    BUS_DMA_NOWAIT) != 0) {
6581 			m_freem(m);
6582 			break;
6583 		}
6584 		ms->ms_m = m;
6585 
6586 		htobem32(&rqe->rqe_byte_count, ms->ms_map->dm_segs[0].ds_len);
6587 		htobem64(&rqe->rqe_addr, ms->ms_map->dm_segs[0].ds_addr);
6588 		htobem32(&rqe->rqe_lkey, sc->sc_lkey);
6589 
6590 		p++;
6591 	}
6592 
6593 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&rx->rx_rq_mem),
6594 	    0, MCX_DMA_LEN(&rx->rx_rq_mem), BUS_DMASYNC_PREWRITE);
6595 
6596 	rx->rx_prod = p;
6597 
6598 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_doorbell_mem),
6599 	    rx->rx_doorbell, sizeof(uint32_t), BUS_DMASYNC_POSTWRITE);
6600 	htobem32(MCX_DMA_OFF(&sc->sc_doorbell_mem, rx->rx_doorbell),
6601 	    p & MCX_WQ_DOORBELL_MASK);
6602 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_doorbell_mem),
6603 	    rx->rx_doorbell, sizeof(uint32_t), BUS_DMASYNC_PREWRITE);
6604 
6605 	return (nslots - fills);
6606 }
6607 
6608 int
6609 mcx_rx_fill(struct mcx_softc *sc, struct mcx_rx *rx)
6610 {
6611 	u_int slots;
6612 
6613 	slots = if_rxr_get(&rx->rx_rxr, (1 << MCX_LOG_RQ_SIZE));
6614 	if (slots == 0)
6615 		return (1);
6616 
6617 	slots = mcx_rx_fill_slots(sc, rx, slots);
6618 	if_rxr_put(&rx->rx_rxr, slots);
6619 	return (0);
6620 }
6621 
6622 void
6623 mcx_refill(void *xrx)
6624 {
6625 	struct mcx_rx *rx = xrx;
6626 	struct mcx_softc *sc = rx->rx_softc;
6627 
6628 	mcx_rx_fill(sc, rx);
6629 
6630 	if (if_rxr_inuse(&rx->rx_rxr) == 0)
6631 		timeout_add(&rx->rx_refill, 1);
6632 }
6633 
6634 static int
6635 mcx_process_txeof(struct mcx_softc *sc, struct mcx_tx *tx,
6636     struct mcx_cq_entry *cqe)
6637 {
6638 	struct mcx_slot *ms;
6639 	bus_dmamap_t map;
6640 	int slot, slots;
6641 
6642 	slot = betoh16(cqe->cq_wqe_count) % (1 << MCX_LOG_SQ_SIZE);
6643 
6644 	ms = &tx->tx_slots[slot];
6645 	map = ms->ms_map;
6646 	bus_dmamap_sync(sc->sc_dmat, map, 0, map->dm_mapsize,
6647 	    BUS_DMASYNC_POSTWRITE);
6648 
6649 	slots = 1;
6650 	if (map->dm_nsegs > 1)
6651 		slots += (map->dm_nsegs+2) / MCX_SQ_SEGS_PER_SLOT;
6652 
6653 	bus_dmamap_unload(sc->sc_dmat, map);
6654 	m_freem(ms->ms_m);
6655 	ms->ms_m = NULL;
6656 
6657 	return (slots);
6658 }
6659 
6660 static uint64_t
6661 mcx_uptime(void)
6662 {
6663 	struct timespec ts;
6664 
6665 	nanouptime(&ts);
6666 
6667 	return ((uint64_t)ts.tv_sec * 1000000000 + (uint64_t)ts.tv_nsec);
6668 }
6669 
6670 static void
6671 mcx_calibrate_first(struct mcx_softc *sc)
6672 {
6673 	struct mcx_calibration *c = &sc->sc_calibration[0];
6674 	int s;
6675 
6676 	sc->sc_calibration_gen = 0;
6677 
6678 	s = splhigh(); /* crit_enter? */
6679 	c->c_ubase = mcx_uptime();
6680 	c->c_tbase = mcx_timer(sc);
6681 	splx(s);
6682 	c->c_ratio = 0;
6683 
6684 #ifdef notyet
6685 	timeout_add_sec(&sc->sc_calibrate, MCX_CALIBRATE_FIRST);
6686 #endif
6687 }
6688 
6689 #define MCX_TIMESTAMP_SHIFT 24
6690 
6691 static void
6692 mcx_calibrate(void *arg)
6693 {
6694 	struct mcx_softc *sc = arg;
6695 	struct mcx_calibration *nc, *pc;
6696 	uint64_t udiff, tdiff;
6697 	unsigned int gen;
6698 	int s;
6699 
6700 	if (!ISSET(sc->sc_ac.ac_if.if_flags, IFF_RUNNING))
6701 		return;
6702 
6703 	timeout_add_sec(&sc->sc_calibrate, MCX_CALIBRATE_NORMAL);
6704 
6705 	gen = sc->sc_calibration_gen;
6706 	pc = &sc->sc_calibration[gen % nitems(sc->sc_calibration)];
6707 	gen++;
6708 	nc = &sc->sc_calibration[gen % nitems(sc->sc_calibration)];
6709 
6710 	nc->c_uptime = pc->c_ubase;
6711 	nc->c_timestamp = pc->c_tbase;
6712 
6713 	s = splhigh(); /* crit_enter? */
6714 	nc->c_ubase = mcx_uptime();
6715 	nc->c_tbase = mcx_timer(sc);
6716 	splx(s);
6717 
6718 	udiff = nc->c_ubase - nc->c_uptime;
6719 	tdiff = nc->c_tbase - nc->c_timestamp;
6720 
6721 	/*
6722 	 * udiff is the wall clock time between calibration ticks,
6723 	 * which should be 32 seconds or 32 billion nanoseconds. if
6724 	 * we squint, 1 billion nanoseconds is kind of like a 32 bit
6725 	 * number, so 32 billion should still have a lot of high bits
6726 	 * spare. we use this space by shifting the nanoseconds up
6727 	 * 24 bits so we have a nice big number to divide by the
6728 	 * number of mcx timer ticks.
6729 	 */
6730 	nc->c_ratio = (udiff << MCX_TIMESTAMP_SHIFT) / tdiff;
6731 
6732 	membar_producer();
6733 	sc->sc_calibration_gen = gen;
6734 }
6735 
6736 static int
6737 mcx_process_rx(struct mcx_softc *sc, struct mcx_rx *rx,
6738     struct mcx_cq_entry *cqe, struct mbuf_list *ml,
6739     const struct mcx_calibration *c)
6740 {
6741 	struct mcx_slot *ms;
6742 	struct mbuf *m;
6743 	uint32_t flags;
6744 	int slot;
6745 
6746 	slot = betoh16(cqe->cq_wqe_count) % (1 << MCX_LOG_RQ_SIZE);
6747 
6748 	ms = &rx->rx_slots[slot];
6749 	bus_dmamap_sync(sc->sc_dmat, ms->ms_map, 0, ms->ms_map->dm_mapsize,
6750 	    BUS_DMASYNC_POSTREAD);
6751 	bus_dmamap_unload(sc->sc_dmat, ms->ms_map);
6752 
6753 	m = ms->ms_m;
6754 	ms->ms_m = NULL;
6755 
6756 	m->m_pkthdr.len = m->m_len = bemtoh32(&cqe->cq_byte_cnt);
6757 
6758 	if (cqe->cq_rx_hash_type) {
6759 		m->m_pkthdr.ph_flowid = betoh32(cqe->cq_rx_hash);
6760 		m->m_pkthdr.csum_flags |= M_FLOWID;
6761 	}
6762 
6763 	flags = bemtoh32(&cqe->cq_flags);
6764 	if (flags & MCX_CQ_ENTRY_FLAGS_L3_OK)
6765 		m->m_pkthdr.csum_flags = M_IPV4_CSUM_IN_OK;
6766 	if (flags & MCX_CQ_ENTRY_FLAGS_L4_OK)
6767 		m->m_pkthdr.csum_flags |= M_TCP_CSUM_IN_OK |
6768 		    M_UDP_CSUM_IN_OK;
6769 #if NVLAN > 0
6770 	if (flags & MCX_CQ_ENTRY_FLAGS_CV) {
6771 		m->m_pkthdr.ether_vtag = (flags &
6772 		    MCX_CQ_ENTRY_FLAGS_VLAN_MASK);
6773 		m->m_flags |= M_VLANTAG;
6774 	}
6775 #endif
6776 
6777 #ifdef notyet
6778 	if (ISSET(sc->sc_ac.ac_if.if_flags, IFF_LINK0) && c->c_ratio) {
6779 		uint64_t t = bemtoh64(&cqe->cq_timestamp);
6780 		t -= c->c_timestamp;
6781 		t *= c->c_ratio;
6782 		t >>= MCX_TIMESTAMP_SHIFT;
6783 		t += c->c_uptime;
6784 
6785 		m->m_pkthdr.ph_timestamp = t;
6786 		SET(m->m_pkthdr.csum_flags, M_TIMESTAMP);
6787 	}
6788 #endif
6789 
6790 	ml_enqueue(ml, m);
6791 
6792 	return (1);
6793 }
6794 
6795 static struct mcx_cq_entry *
6796 mcx_next_cq_entry(struct mcx_softc *sc, struct mcx_cq *cq)
6797 {
6798 	struct mcx_cq_entry *cqe;
6799 	int next;
6800 
6801 	cqe = (struct mcx_cq_entry *)MCX_DMA_KVA(&cq->cq_mem);
6802 	next = cq->cq_cons % (1 << MCX_LOG_CQ_SIZE);
6803 
6804 	if ((cqe[next].cq_opcode_owner & MCX_CQ_ENTRY_FLAG_OWNER) ==
6805 	    ((cq->cq_cons >> MCX_LOG_CQ_SIZE) & 1)) {
6806 		return (&cqe[next]);
6807 	}
6808 
6809 	return (NULL);
6810 }
6811 
6812 static void
6813 mcx_arm_cq(struct mcx_softc *sc, struct mcx_cq *cq, int uar)
6814 {
6815 	struct mcx_cq_doorbell *db;
6816 	bus_size_t offset;
6817 	uint32_t val;
6818 	uint64_t uval;
6819 
6820 	val = ((cq->cq_count) & 3) << MCX_CQ_DOORBELL_ARM_CMD_SN_SHIFT;
6821 	val |= (cq->cq_cons & MCX_CQ_DOORBELL_ARM_CI_MASK);
6822 
6823 	db = MCX_DMA_OFF(&sc->sc_doorbell_mem, cq->cq_doorbell);
6824 
6825 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_doorbell_mem),
6826 	    cq->cq_doorbell, sizeof(*db), BUS_DMASYNC_POSTWRITE);
6827 
6828 	htobem32(&db->db_update_ci, cq->cq_cons & MCX_CQ_DOORBELL_ARM_CI_MASK);
6829 	htobem32(&db->db_arm_ci, val);
6830 
6831 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_doorbell_mem),
6832 	    cq->cq_doorbell, sizeof(*db), BUS_DMASYNC_PREWRITE);
6833 
6834 	offset = (MCX_PAGE_SIZE * uar) + MCX_UAR_CQ_DOORBELL;
6835 
6836 	uval = (uint64_t)val << 32;
6837 	uval |= cq->cq_n;
6838 
6839 	bus_space_write_raw_8(sc->sc_memt, sc->sc_memh, offset, htobe64(uval));
6840 	mcx_bar(sc, offset, sizeof(uval), BUS_SPACE_BARRIER_WRITE);
6841 }
6842 
6843 void
6844 mcx_process_cq(struct mcx_softc *sc, struct mcx_queues *q, struct mcx_cq *cq)
6845 {
6846 	struct mcx_rx *rx = &q->q_rx;
6847 	struct mcx_tx *tx = &q->q_tx;
6848 	const struct mcx_calibration *c;
6849 	unsigned int gen;
6850 	struct mcx_cq_entry *cqe;
6851 	uint8_t *cqp;
6852 	struct mbuf_list ml = MBUF_LIST_INITIALIZER();
6853 	int rxfree, txfree;
6854 
6855 	gen = sc->sc_calibration_gen;
6856 	membar_consumer();
6857 	c = &sc->sc_calibration[gen % nitems(sc->sc_calibration)];
6858 
6859 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&cq->cq_mem),
6860 	    0, MCX_DMA_LEN(&cq->cq_mem), BUS_DMASYNC_POSTREAD);
6861 
6862 	rxfree = 0;
6863 	txfree = 0;
6864 	while ((cqe = mcx_next_cq_entry(sc, cq))) {
6865 		uint8_t opcode;
6866 		opcode = (cqe->cq_opcode_owner >> MCX_CQ_ENTRY_OPCODE_SHIFT);
6867 		switch (opcode) {
6868 		case MCX_CQ_ENTRY_OPCODE_REQ:
6869 			txfree += mcx_process_txeof(sc, tx, cqe);
6870 			break;
6871 		case MCX_CQ_ENTRY_OPCODE_SEND:
6872 			rxfree += mcx_process_rx(sc, rx, cqe, &ml, c);
6873 			break;
6874 		case MCX_CQ_ENTRY_OPCODE_REQ_ERR:
6875 		case MCX_CQ_ENTRY_OPCODE_SEND_ERR:
6876 			cqp = (uint8_t *)cqe;
6877 			/* printf("%s: cq completion error: %x\n",
6878 			    DEVNAME(sc), cqp[0x37]); */
6879 			break;
6880 
6881 		default:
6882 			/* printf("%s: cq completion opcode %x??\n",
6883 			    DEVNAME(sc), opcode); */
6884 			break;
6885 		}
6886 
6887 		cq->cq_cons++;
6888 	}
6889 
6890 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&cq->cq_mem),
6891 	    0, MCX_DMA_LEN(&cq->cq_mem), BUS_DMASYNC_PREREAD);
6892 
6893 	cq->cq_count++;
6894 	mcx_arm_cq(sc, cq, q->q_uar);
6895 
6896 	if (rxfree > 0) {
6897 		if_rxr_put(&rx->rx_rxr, rxfree);
6898 		if (ifiq_input(rx->rx_ifiq, &ml))
6899 			if_rxr_livelocked(&rx->rx_rxr);
6900 
6901 		mcx_rx_fill(sc, rx);
6902 		if (if_rxr_inuse(&rx->rx_rxr) == 0)
6903 			timeout_add(&rx->rx_refill, 1);
6904 	}
6905 	if (txfree > 0) {
6906 		tx->tx_cons += txfree;
6907 		if (ifq_is_oactive(tx->tx_ifq))
6908 			ifq_restart(tx->tx_ifq);
6909 	}
6910 }
6911 
6912 
6913 static void
6914 mcx_arm_eq(struct mcx_softc *sc, struct mcx_eq *eq, int uar)
6915 {
6916 	bus_size_t offset;
6917 	uint32_t val;
6918 
6919 	offset = (MCX_PAGE_SIZE * uar) + MCX_UAR_EQ_DOORBELL_ARM;
6920 	val = (eq->eq_n << 24) | (eq->eq_cons & 0xffffff);
6921 
6922 	mcx_wr(sc, offset, val);
6923 	mcx_bar(sc, offset, sizeof(val), BUS_SPACE_BARRIER_WRITE);
6924 }
6925 
6926 static struct mcx_eq_entry *
6927 mcx_next_eq_entry(struct mcx_softc *sc, struct mcx_eq *eq)
6928 {
6929 	struct mcx_eq_entry *eqe;
6930 	int next;
6931 
6932 	eqe = (struct mcx_eq_entry *)MCX_DMA_KVA(&eq->eq_mem);
6933 	next = eq->eq_cons % (1 << MCX_LOG_EQ_SIZE);
6934 	if ((eqe[next].eq_owner & 1) ==
6935 	    ((eq->eq_cons >> MCX_LOG_EQ_SIZE) & 1)) {
6936 		eq->eq_cons++;
6937 		return (&eqe[next]);
6938 	}
6939 	return (NULL);
6940 }
6941 
6942 int
6943 mcx_admin_intr(void *xsc)
6944 {
6945 	struct mcx_softc *sc = (struct mcx_softc *)xsc;
6946 	struct mcx_eq *eq = &sc->sc_admin_eq;
6947 	struct mcx_eq_entry *eqe;
6948 
6949 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&eq->eq_mem),
6950 	    0, MCX_DMA_LEN(&eq->eq_mem), BUS_DMASYNC_POSTREAD);
6951 
6952 	while ((eqe = mcx_next_eq_entry(sc, eq)) != NULL) {
6953 		switch (eqe->eq_event_type) {
6954 		case MCX_EVENT_TYPE_LAST_WQE:
6955 			/* printf("%s: last wqe reached?\n", DEVNAME(sc)); */
6956 			break;
6957 
6958 		case MCX_EVENT_TYPE_CQ_ERROR:
6959 			/* printf("%s: cq error\n", DEVNAME(sc)); */
6960 			break;
6961 
6962 		case MCX_EVENT_TYPE_CMD_COMPLETION:
6963 			/* wakeup probably */
6964 			break;
6965 
6966 		case MCX_EVENT_TYPE_PORT_CHANGE:
6967 			task_add(systq, &sc->sc_port_change);
6968 			break;
6969 
6970 		default:
6971 			/* printf("%s: something happened\n", DEVNAME(sc)); */
6972 			break;
6973 		}
6974 	}
6975 
6976 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&eq->eq_mem),
6977 	    0, MCX_DMA_LEN(&eq->eq_mem), BUS_DMASYNC_PREREAD);
6978 
6979 	mcx_arm_eq(sc, eq, sc->sc_uar);
6980 
6981 	return (1);
6982 }
6983 
6984 int
6985 mcx_cq_intr(void *xq)
6986 {
6987 	struct mcx_queues *q = (struct mcx_queues *)xq;
6988 	struct mcx_softc *sc = q->q_sc;
6989 	struct mcx_eq *eq = &q->q_eq;
6990 	struct mcx_eq_entry *eqe;
6991 	int cqn;
6992 
6993 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&eq->eq_mem),
6994 	    0, MCX_DMA_LEN(&eq->eq_mem), BUS_DMASYNC_POSTREAD);
6995 
6996 	while ((eqe = mcx_next_eq_entry(sc, eq)) != NULL) {
6997 		switch (eqe->eq_event_type) {
6998 		case MCX_EVENT_TYPE_COMPLETION:
6999 			cqn = betoh32(eqe->eq_event_data[6]);
7000 			if (cqn == q->q_cq.cq_n)
7001 				mcx_process_cq(sc, q, &q->q_cq);
7002 			break;
7003 		}
7004 	}
7005 
7006 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&eq->eq_mem),
7007 	    0, MCX_DMA_LEN(&eq->eq_mem), BUS_DMASYNC_PREREAD);
7008 
7009 	mcx_arm_eq(sc, eq, q->q_uar);
7010 
7011 	return (1);
7012 }
7013 
7014 static void
7015 mcx_free_slots(struct mcx_softc *sc, struct mcx_slot *slots, int allocated,
7016     int total)
7017 {
7018 	struct mcx_slot *ms;
7019 
7020 	int i = allocated;
7021 	while (i-- > 0) {
7022 		ms = &slots[i];
7023 		bus_dmamap_destroy(sc->sc_dmat, ms->ms_map);
7024 		if (ms->ms_m != NULL)
7025 			m_freem(ms->ms_m);
7026 	}
7027 	free(slots, M_DEVBUF, total * sizeof(*ms));
7028 }
7029 
7030 static int
7031 mcx_queue_up(struct mcx_softc *sc, struct mcx_queues *q)
7032 {
7033 	struct mcx_rx *rx;
7034 	struct mcx_tx *tx;
7035 	struct mcx_slot *ms;
7036 	int i;
7037 
7038 	rx = &q->q_rx;
7039 	rx->rx_slots = mallocarray(sizeof(*ms), (1 << MCX_LOG_RQ_SIZE),
7040 	    M_DEVBUF, M_WAITOK | M_ZERO);
7041 	if (rx->rx_slots == NULL) {
7042 		printf("%s: failed to allocate rx slots\n", DEVNAME(sc));
7043 		return ENOMEM;
7044 	}
7045 
7046 	for (i = 0; i < (1 << MCX_LOG_RQ_SIZE); i++) {
7047 		ms = &rx->rx_slots[i];
7048 		if (bus_dmamap_create(sc->sc_dmat, sc->sc_hardmtu, 1,
7049 		    sc->sc_hardmtu, 0,
7050 		    BUS_DMA_WAITOK | BUS_DMA_ALLOCNOW | BUS_DMA_64BIT,
7051 		    &ms->ms_map) != 0) {
7052 			printf("%s: failed to allocate rx dma maps\n",
7053 			    DEVNAME(sc));
7054 			goto destroy_rx_slots;
7055 		}
7056 	}
7057 
7058 	tx = &q->q_tx;
7059 	tx->tx_slots = mallocarray(sizeof(*ms), (1 << MCX_LOG_SQ_SIZE),
7060 	    M_DEVBUF, M_WAITOK | M_ZERO);
7061 	if (tx->tx_slots == NULL) {
7062 		printf("%s: failed to allocate tx slots\n", DEVNAME(sc));
7063 		goto destroy_rx_slots;
7064 	}
7065 
7066 	for (i = 0; i < (1 << MCX_LOG_SQ_SIZE); i++) {
7067 		ms = &tx->tx_slots[i];
7068 		if (bus_dmamap_create(sc->sc_dmat, sc->sc_hardmtu,
7069 		    MCX_SQ_MAX_SEGMENTS, sc->sc_hardmtu, 0,
7070 		    BUS_DMA_WAITOK | BUS_DMA_ALLOCNOW | BUS_DMA_64BIT,
7071 		    &ms->ms_map) != 0) {
7072 			printf("%s: failed to allocate tx dma maps\n",
7073 			    DEVNAME(sc));
7074 			goto destroy_tx_slots;
7075 		}
7076 	}
7077 
7078 	if (mcx_create_cq(sc, &q->q_cq, q->q_uar, q->q_index,
7079 	    q->q_eq.eq_n) != 0)
7080 		goto destroy_tx_slots;
7081 
7082 	if (mcx_create_sq(sc, tx, q->q_uar, q->q_index, q->q_cq.cq_n)
7083 	    != 0)
7084 		goto destroy_cq;
7085 
7086 	if (mcx_create_rq(sc, rx, q->q_index, q->q_cq.cq_n) != 0)
7087 		goto destroy_sq;
7088 
7089 	return 0;
7090 
7091 destroy_sq:
7092 	mcx_destroy_sq(sc, tx);
7093 destroy_cq:
7094 	mcx_destroy_cq(sc, &q->q_cq);
7095 destroy_tx_slots:
7096 	mcx_free_slots(sc, tx->tx_slots, i, (1 << MCX_LOG_SQ_SIZE));
7097 	tx->tx_slots = NULL;
7098 
7099 	i = (1 << MCX_LOG_RQ_SIZE);
7100 destroy_rx_slots:
7101 	mcx_free_slots(sc, rx->rx_slots, i, (1 << MCX_LOG_RQ_SIZE));
7102 	rx->rx_slots = NULL;
7103 	return ENOMEM;
7104 }
7105 
7106 static int
7107 mcx_rss_group_entry_count(struct mcx_softc *sc, int group)
7108 {
7109 	int i;
7110 	int count;
7111 
7112 	count = 0;
7113 	for (i = 0; i < nitems(mcx_rss_config); i++) {
7114 		if (mcx_rss_config[i].flow_group == group)
7115 			count++;
7116 	}
7117 
7118 	return count;
7119 }
7120 
7121 static int
7122 mcx_up(struct mcx_softc *sc)
7123 {
7124 	struct ifnet *ifp = &sc->sc_ac.ac_if;
7125 	struct mcx_rx *rx;
7126 	struct mcx_tx *tx;
7127 	int i, start, count, flow_group, flow_index;
7128 	struct mcx_flow_match match_crit;
7129 	struct mcx_rss_rule *rss;
7130 	uint32_t dest;
7131 	int rqns[MCX_MAX_QUEUES];
7132 
7133 	if (mcx_create_tis(sc, &sc->sc_tis) != 0)
7134 		goto down;
7135 
7136 	for (i = 0; i < sc->sc_nqueues; i++) {
7137 		if (mcx_queue_up(sc, &sc->sc_queues[i]) != 0) {
7138 			goto down;
7139 		}
7140 	}
7141 
7142 	/* RSS flow table and flow groups */
7143 	if (mcx_create_flow_table(sc, MCX_LOG_FLOW_TABLE_SIZE, 1,
7144 	    &sc->sc_rss_flow_table_id) != 0)
7145 		goto down;
7146 
7147 	dest = MCX_FLOW_CONTEXT_DEST_TYPE_TABLE |
7148 	    sc->sc_rss_flow_table_id;
7149 
7150 	/* L4 RSS flow group (v4/v6 tcp/udp, no fragments) */
7151 	memset(&match_crit, 0, sizeof(match_crit));
7152 	match_crit.mc_ethertype = 0xffff;
7153 	match_crit.mc_ip_proto = 0xff;
7154 	match_crit.mc_vlan_flags = MCX_FLOW_MATCH_IP_FRAG;
7155 	start = 0;
7156 	count = mcx_rss_group_entry_count(sc, MCX_FLOW_GROUP_RSS_L4);
7157 	if (count != 0) {
7158 		if (mcx_create_flow_group(sc, sc->sc_rss_flow_table_id,
7159 		    MCX_FLOW_GROUP_RSS_L4, start, count,
7160 		    MCX_CREATE_FLOW_GROUP_CRIT_OUTER, &match_crit) != 0)
7161 			goto down;
7162 		start += count;
7163 	}
7164 
7165 	/* L3 RSS flow group (v4/v6, including fragments) */
7166 	memset(&match_crit, 0, sizeof(match_crit));
7167 	match_crit.mc_ethertype = 0xffff;
7168 	count = mcx_rss_group_entry_count(sc, MCX_FLOW_GROUP_RSS_L3);
7169 	if (mcx_create_flow_group(sc, sc->sc_rss_flow_table_id,
7170 	    MCX_FLOW_GROUP_RSS_L3, start, count,
7171 	    MCX_CREATE_FLOW_GROUP_CRIT_OUTER, &match_crit) != 0)
7172 		goto down;
7173 	start += count;
7174 
7175 	/* non-RSS flow group */
7176 	count = mcx_rss_group_entry_count(sc, MCX_FLOW_GROUP_RSS_NONE);
7177 	memset(&match_crit, 0, sizeof(match_crit));
7178 	if (mcx_create_flow_group(sc, sc->sc_rss_flow_table_id,
7179 	    MCX_FLOW_GROUP_RSS_NONE, start, count, 0, &match_crit) != 0)
7180 		goto down;
7181 
7182 	/* Root flow table, matching packets based on mac address */
7183 	if (mcx_create_flow_table(sc, MCX_LOG_FLOW_TABLE_SIZE, 0,
7184 	    &sc->sc_mac_flow_table_id) != 0)
7185 		goto down;
7186 
7187 	/* promisc flow group */
7188 	start = 0;
7189 	memset(&match_crit, 0, sizeof(match_crit));
7190 	if (mcx_create_flow_group(sc, sc->sc_mac_flow_table_id,
7191 	    MCX_FLOW_GROUP_PROMISC, start, 1, 0, &match_crit) != 0)
7192 		goto down;
7193 	sc->sc_promisc_flow_enabled = 0;
7194 	start++;
7195 
7196 	/* all multicast flow group */
7197 	match_crit.mc_dest_mac[0] = 0x01;
7198 	if (mcx_create_flow_group(sc, sc->sc_mac_flow_table_id,
7199 	    MCX_FLOW_GROUP_ALLMULTI, start, 1,
7200 	    MCX_CREATE_FLOW_GROUP_CRIT_OUTER, &match_crit) != 0)
7201 		goto down;
7202 	sc->sc_allmulti_flow_enabled = 0;
7203 	start++;
7204 
7205 	/* mac address matching flow group */
7206 	memset(&match_crit.mc_dest_mac, 0xff, sizeof(match_crit.mc_dest_mac));
7207 	if (mcx_create_flow_group(sc, sc->sc_mac_flow_table_id,
7208 	    MCX_FLOW_GROUP_MAC, start, (1 << MCX_LOG_FLOW_TABLE_SIZE) - start,
7209 	    MCX_CREATE_FLOW_GROUP_CRIT_OUTER, &match_crit) != 0)
7210 		goto down;
7211 
7212 	/* flow table entries for unicast and broadcast */
7213 	start = 0;
7214 	if (mcx_set_flow_table_entry_mac(sc, MCX_FLOW_GROUP_MAC, start,
7215 	    sc->sc_ac.ac_enaddr, dest) != 0)
7216 		goto down;
7217 	start++;
7218 
7219 	if (mcx_set_flow_table_entry_mac(sc, MCX_FLOW_GROUP_MAC, start,
7220 	    etherbroadcastaddr, dest) != 0)
7221 		goto down;
7222 	start++;
7223 
7224 	/* multicast entries go after that */
7225 	sc->sc_mcast_flow_base = start;
7226 
7227 	/* re-add any existing multicast flows */
7228 	for (i = 0; i < MCX_NUM_MCAST_FLOWS; i++) {
7229 		if (sc->sc_mcast_flows[i][0] != 0) {
7230 			mcx_set_flow_table_entry_mac(sc, MCX_FLOW_GROUP_MAC,
7231 			    sc->sc_mcast_flow_base + i,
7232 			    sc->sc_mcast_flows[i], dest);
7233 		}
7234 	}
7235 
7236 	if (mcx_set_flow_table_root(sc, sc->sc_mac_flow_table_id) != 0)
7237 		goto down;
7238 
7239 	/*
7240 	 * the RQT can be any size as long as it's a power of two.
7241 	 * since we also restrict the number of queues to a power of two,
7242 	 * we can just put each rx queue in once.
7243 	 */
7244 	for (i = 0; i < sc->sc_nqueues; i++)
7245 		rqns[i] = sc->sc_queues[i].q_rx.rx_rqn;
7246 
7247 	if (mcx_create_rqt(sc, sc->sc_nqueues, rqns, &sc->sc_rqt) != 0)
7248 		goto down;
7249 
7250 	start = 0;
7251 	flow_index = 0;
7252 	flow_group = -1;
7253 	for (i = 0; i < nitems(mcx_rss_config); i++) {
7254 		rss = &mcx_rss_config[i];
7255 		if (rss->flow_group != flow_group) {
7256 			flow_group = rss->flow_group;
7257 			flow_index = 0;
7258 		}
7259 
7260 		if (rss->hash_sel == 0) {
7261 			if (mcx_create_tir_direct(sc, &sc->sc_queues[0].q_rx,
7262 			    &sc->sc_tir[i]) != 0)
7263 				goto down;
7264 		} else {
7265 			if (mcx_create_tir_indirect(sc, sc->sc_rqt,
7266 			    rss->hash_sel, &sc->sc_tir[i]) != 0)
7267 				goto down;
7268 		}
7269 
7270 		if (mcx_set_flow_table_entry_proto(sc, flow_group,
7271 		    flow_index, rss->ethertype, rss->ip_proto,
7272 		    MCX_FLOW_CONTEXT_DEST_TYPE_TIR | sc->sc_tir[i]) != 0)
7273 			goto down;
7274 		flow_index++;
7275 	}
7276 
7277 	for (i = 0; i < sc->sc_nqueues; i++) {
7278 		struct mcx_queues *q = &sc->sc_queues[i];
7279 		rx = &q->q_rx;
7280 		tx = &q->q_tx;
7281 
7282 		/* start the queues */
7283 		if (mcx_ready_sq(sc, tx) != 0)
7284 			goto down;
7285 
7286 		if (mcx_ready_rq(sc, rx) != 0)
7287 			goto down;
7288 
7289 		if_rxr_init(&rx->rx_rxr, 1, (1 << MCX_LOG_RQ_SIZE));
7290 		rx->rx_prod = 0;
7291 		mcx_rx_fill(sc, rx);
7292 
7293 		tx->tx_cons = 0;
7294 		tx->tx_prod = 0;
7295 		ifq_clr_oactive(tx->tx_ifq);
7296 	}
7297 
7298 	mcx_calibrate_first(sc);
7299 
7300 	SET(ifp->if_flags, IFF_RUNNING);
7301 
7302 	return ENETRESET;
7303 down:
7304 	mcx_down(sc);
7305 	return ENOMEM;
7306 }
7307 
7308 static void
7309 mcx_down(struct mcx_softc *sc)
7310 {
7311 	struct ifnet *ifp = &sc->sc_ac.ac_if;
7312 	struct mcx_rss_rule *rss;
7313 	int group, i, flow_group, flow_index;
7314 
7315 	CLR(ifp->if_flags, IFF_RUNNING);
7316 
7317 	/*
7318 	 * delete flow table entries first, so no packets can arrive
7319 	 * after the barriers
7320 	 */
7321 	if (sc->sc_promisc_flow_enabled)
7322 		mcx_delete_flow_table_entry(sc, MCX_FLOW_GROUP_PROMISC, 0);
7323 	if (sc->sc_allmulti_flow_enabled)
7324 		mcx_delete_flow_table_entry(sc, MCX_FLOW_GROUP_ALLMULTI, 0);
7325 	mcx_delete_flow_table_entry(sc, MCX_FLOW_GROUP_MAC, 0);
7326 	mcx_delete_flow_table_entry(sc, MCX_FLOW_GROUP_MAC, 1);
7327 	for (i = 0; i < MCX_NUM_MCAST_FLOWS; i++) {
7328 		if (sc->sc_mcast_flows[i][0] != 0) {
7329 			mcx_delete_flow_table_entry(sc, MCX_FLOW_GROUP_MAC,
7330 			    sc->sc_mcast_flow_base + i);
7331 		}
7332 	}
7333 
7334 	flow_group = -1;
7335 	flow_index = 0;
7336 	for (i = 0; i < nitems(mcx_rss_config); i++) {
7337 		rss = &mcx_rss_config[i];
7338 		if (rss->flow_group != flow_group) {
7339 			flow_group = rss->flow_group;
7340 			flow_index = 0;
7341 		}
7342 
7343 		mcx_delete_flow_table_entry(sc, flow_group, flow_index);
7344 
7345 		mcx_destroy_tir(sc, sc->sc_tir[i]);
7346 		sc->sc_tir[i] = 0;
7347 
7348 		flow_index++;
7349 	}
7350 	intr_barrier(sc->sc_ihc);
7351 	for (i = 0; i < sc->sc_nqueues; i++) {
7352 		struct ifqueue *ifq = sc->sc_queues[i].q_tx.tx_ifq;
7353 		ifq_barrier(ifq);
7354 
7355 		timeout_del_barrier(&sc->sc_queues[i].q_rx.rx_refill);
7356 
7357 		intr_barrier(sc->sc_queues[i].q_ihc);
7358 	}
7359 
7360 	timeout_del_barrier(&sc->sc_calibrate);
7361 
7362 	for (group = 0; group < MCX_NUM_FLOW_GROUPS; group++) {
7363 		if (sc->sc_flow_group[group].g_id != -1)
7364 			mcx_destroy_flow_group(sc, group);
7365 	}
7366 
7367 	if (sc->sc_mac_flow_table_id != -1) {
7368 		mcx_destroy_flow_table(sc, sc->sc_mac_flow_table_id);
7369 		sc->sc_mac_flow_table_id = -1;
7370 	}
7371 	if (sc->sc_rss_flow_table_id != -1) {
7372 		mcx_destroy_flow_table(sc, sc->sc_rss_flow_table_id);
7373 		sc->sc_rss_flow_table_id = -1;
7374 	}
7375 	if (sc->sc_rqt != -1) {
7376 		mcx_destroy_rqt(sc, sc->sc_rqt);
7377 		sc->sc_rqt = -1;
7378 	}
7379 
7380 	for (i = 0; i < sc->sc_nqueues; i++) {
7381 		struct mcx_queues *q = &sc->sc_queues[i];
7382 		struct mcx_rx *rx = &q->q_rx;
7383 		struct mcx_tx *tx = &q->q_tx;
7384 		struct mcx_cq *cq = &q->q_cq;
7385 
7386 		if (rx->rx_rqn != 0)
7387 			mcx_destroy_rq(sc, rx);
7388 
7389 		if (tx->tx_sqn != 0)
7390 			mcx_destroy_sq(sc, tx);
7391 
7392 		if (tx->tx_slots != NULL) {
7393 			mcx_free_slots(sc, tx->tx_slots,
7394 			    (1 << MCX_LOG_SQ_SIZE), (1 << MCX_LOG_SQ_SIZE));
7395 			tx->tx_slots = NULL;
7396 		}
7397 		if (rx->rx_slots != NULL) {
7398 			mcx_free_slots(sc, rx->rx_slots,
7399 			    (1 << MCX_LOG_RQ_SIZE), (1 << MCX_LOG_RQ_SIZE));
7400 			rx->rx_slots = NULL;
7401 		}
7402 
7403 		if (cq->cq_n != 0)
7404 			mcx_destroy_cq(sc, cq);
7405 	}
7406 	if (sc->sc_tis != 0) {
7407 		mcx_destroy_tis(sc, sc->sc_tis);
7408 		sc->sc_tis = 0;
7409 	}
7410 }
7411 
7412 static int
7413 mcx_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
7414 {
7415 	struct mcx_softc *sc = (struct mcx_softc *)ifp->if_softc;
7416 	struct ifreq *ifr = (struct ifreq *)data;
7417 	uint8_t addrhi[ETHER_ADDR_LEN], addrlo[ETHER_ADDR_LEN];
7418 	int s, i, error = 0;
7419 	uint32_t dest;
7420 
7421 	s = splnet();
7422 	switch (cmd) {
7423 	case SIOCSIFADDR:
7424 		ifp->if_flags |= IFF_UP;
7425 		/* FALLTHROUGH */
7426 
7427 	case SIOCSIFFLAGS:
7428 		if (ISSET(ifp->if_flags, IFF_UP)) {
7429 			if (ISSET(ifp->if_flags, IFF_RUNNING))
7430 				error = ENETRESET;
7431 			else
7432 				error = mcx_up(sc);
7433 		} else {
7434 			if (ISSET(ifp->if_flags, IFF_RUNNING))
7435 				mcx_down(sc);
7436 		}
7437 		break;
7438 
7439 	case SIOCGIFMEDIA:
7440 	case SIOCSIFMEDIA:
7441 		error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
7442 		break;
7443 
7444 	case SIOCGIFSFFPAGE:
7445 		error = mcx_get_sffpage(ifp, (struct if_sffpage *)data);
7446 		break;
7447 
7448 	case SIOCGIFRXR:
7449 		error = mcx_rxrinfo(sc, (struct if_rxrinfo *)ifr->ifr_data);
7450 		break;
7451 
7452 	case SIOCADDMULTI:
7453 		if (ether_addmulti(ifr, &sc->sc_ac) == ENETRESET) {
7454 			error = ether_multiaddr(&ifr->ifr_addr, addrlo, addrhi);
7455 			if (error != 0)
7456 				return (error);
7457 
7458 			dest = MCX_FLOW_CONTEXT_DEST_TYPE_TABLE |
7459 			    sc->sc_rss_flow_table_id;
7460 
7461 			for (i = 0; i < MCX_NUM_MCAST_FLOWS; i++) {
7462 				if (sc->sc_mcast_flows[i][0] == 0) {
7463 					memcpy(sc->sc_mcast_flows[i], addrlo,
7464 					    ETHER_ADDR_LEN);
7465 					if (ISSET(ifp->if_flags, IFF_RUNNING)) {
7466 						mcx_set_flow_table_entry_mac(sc,
7467 						    MCX_FLOW_GROUP_MAC,
7468 						    sc->sc_mcast_flow_base + i,
7469 						    sc->sc_mcast_flows[i], dest);
7470 					}
7471 					break;
7472 				}
7473 			}
7474 
7475 			if (!ISSET(ifp->if_flags, IFF_ALLMULTI)) {
7476 				if (i == MCX_NUM_MCAST_FLOWS) {
7477 					SET(ifp->if_flags, IFF_ALLMULTI);
7478 					sc->sc_extra_mcast++;
7479 					error = ENETRESET;
7480 				}
7481 
7482 				if (sc->sc_ac.ac_multirangecnt > 0) {
7483 					SET(ifp->if_flags, IFF_ALLMULTI);
7484 					error = ENETRESET;
7485 				}
7486 			}
7487 		}
7488 		break;
7489 
7490 	case SIOCDELMULTI:
7491 		if (ether_delmulti(ifr, &sc->sc_ac) == ENETRESET) {
7492 			error = ether_multiaddr(&ifr->ifr_addr, addrlo, addrhi);
7493 			if (error != 0)
7494 				return (error);
7495 
7496 			for (i = 0; i < MCX_NUM_MCAST_FLOWS; i++) {
7497 				if (memcmp(sc->sc_mcast_flows[i], addrlo,
7498 				    ETHER_ADDR_LEN) == 0) {
7499 					if (ISSET(ifp->if_flags, IFF_RUNNING)) {
7500 						mcx_delete_flow_table_entry(sc,
7501 						    MCX_FLOW_GROUP_MAC,
7502 						    sc->sc_mcast_flow_base + i);
7503 					}
7504 					sc->sc_mcast_flows[i][0] = 0;
7505 					break;
7506 				}
7507 			}
7508 
7509 			if (i == MCX_NUM_MCAST_FLOWS)
7510 				sc->sc_extra_mcast--;
7511 
7512 			if (ISSET(ifp->if_flags, IFF_ALLMULTI) &&
7513 			    (sc->sc_extra_mcast == 0) &&
7514 			    (sc->sc_ac.ac_multirangecnt == 0)) {
7515 				CLR(ifp->if_flags, IFF_ALLMULTI);
7516 				error = ENETRESET;
7517 			}
7518 		}
7519 		break;
7520 
7521 	default:
7522 		error = ether_ioctl(ifp, &sc->sc_ac, cmd, data);
7523 	}
7524 
7525 	if (error == ENETRESET) {
7526 		if ((ifp->if_flags & (IFF_UP | IFF_RUNNING)) ==
7527 		    (IFF_UP | IFF_RUNNING))
7528 			mcx_iff(sc);
7529 		error = 0;
7530 	}
7531 	splx(s);
7532 
7533 	return (error);
7534 }
7535 
7536 static int
7537 mcx_get_sffpage(struct ifnet *ifp, struct if_sffpage *sff)
7538 {
7539 	struct mcx_softc *sc = (struct mcx_softc *)ifp->if_softc;
7540 	struct mcx_reg_mcia mcia;
7541 	struct mcx_reg_pmlp pmlp;
7542 	int offset, error;
7543 
7544 	/* get module number */
7545 	memset(&pmlp, 0, sizeof(pmlp));
7546 	pmlp.rp_local_port = 1;
7547 	error = mcx_access_hca_reg(sc, MCX_REG_PMLP, MCX_REG_OP_READ, &pmlp,
7548 	    sizeof(pmlp));
7549 	if (error != 0) {
7550 		printf("%s: unable to get eeprom module number\n",
7551 		    DEVNAME(sc));
7552 		return error;
7553 	}
7554 
7555 	for (offset = 0; offset < 256; offset += MCX_MCIA_EEPROM_BYTES) {
7556 		memset(&mcia, 0, sizeof(mcia));
7557 		mcia.rm_l = 0;
7558 		mcia.rm_module = betoh32(pmlp.rp_lane0_mapping) &
7559 		    MCX_PMLP_MODULE_NUM_MASK;
7560 		mcia.rm_i2c_addr = sff->sff_addr / 2;	/* apparently */
7561 		mcia.rm_page_num = sff->sff_page;
7562 		mcia.rm_dev_addr = htobe16(offset);
7563 		mcia.rm_size = htobe16(MCX_MCIA_EEPROM_BYTES);
7564 
7565 		error = mcx_access_hca_reg(sc, MCX_REG_MCIA, MCX_REG_OP_READ,
7566 		    &mcia, sizeof(mcia));
7567 		if (error != 0) {
7568 			printf("%s: unable to read eeprom at %x\n",
7569 			    DEVNAME(sc), offset);
7570 			return error;
7571 		}
7572 
7573 		memcpy(sff->sff_data + offset, mcia.rm_data,
7574 		    MCX_MCIA_EEPROM_BYTES);
7575 	}
7576 
7577 	return 0;
7578 }
7579 
7580 static int
7581 mcx_rxrinfo(struct mcx_softc *sc, struct if_rxrinfo *ifri)
7582 {
7583 	struct if_rxring_info *ifrs;
7584 	unsigned int i;
7585 	int error;
7586 
7587 	ifrs = mallocarray(sc->sc_nqueues, sizeof(*ifrs), M_TEMP,
7588 	    M_WAITOK|M_ZERO|M_CANFAIL);
7589 	if (ifrs == NULL)
7590 		return (ENOMEM);
7591 
7592 	for (i = 0; i < sc->sc_nqueues; i++) {
7593 		struct mcx_rx *rx = &sc->sc_queues[i].q_rx;
7594 		struct if_rxring_info *ifr = &ifrs[i];
7595 
7596 		snprintf(ifr->ifr_name, sizeof(ifr->ifr_name), "%u", i);
7597 		ifr->ifr_size = sc->sc_hardmtu;
7598 		ifr->ifr_info = rx->rx_rxr;
7599 	}
7600 
7601 	error = if_rxr_info_ioctl(ifri, i, ifrs);
7602 	free(ifrs, M_TEMP, i * sizeof(*ifrs));
7603 
7604 	return (error);
7605 }
7606 
7607 int
7608 mcx_load_mbuf(struct mcx_softc *sc, struct mcx_slot *ms, struct mbuf *m)
7609 {
7610 	switch (bus_dmamap_load_mbuf(sc->sc_dmat, ms->ms_map, m,
7611 	    BUS_DMA_STREAMING | BUS_DMA_NOWAIT)) {
7612 	case 0:
7613 		break;
7614 
7615 	case EFBIG:
7616 		if (m_defrag(m, M_DONTWAIT) == 0 &&
7617 		    bus_dmamap_load_mbuf(sc->sc_dmat, ms->ms_map, m,
7618 		    BUS_DMA_STREAMING | BUS_DMA_NOWAIT) == 0)
7619 			break;
7620 
7621 	default:
7622 		return (1);
7623 	}
7624 
7625 	ms->ms_m = m;
7626 	return (0);
7627 }
7628 
7629 static void
7630 mcx_start(struct ifqueue *ifq)
7631 {
7632 	struct mcx_tx *tx = ifq->ifq_softc;
7633 	struct ifnet *ifp = ifq->ifq_if;
7634 	struct mcx_softc *sc = ifp->if_softc;
7635 	struct mcx_sq_entry *sq, *sqe;
7636 	struct mcx_sq_entry_seg *sqs;
7637 	struct mcx_slot *ms;
7638 	bus_dmamap_t map;
7639 	struct mbuf *m;
7640 	u_int idx, free, used;
7641 	uint64_t *bf;
7642 	uint32_t csum;
7643 	size_t bf_base;
7644 	int i, seg, nseg;
7645 
7646 	bf_base = (tx->tx_uar * MCX_PAGE_SIZE) + MCX_UAR_BF;
7647 
7648 	idx = tx->tx_prod % (1 << MCX_LOG_SQ_SIZE);
7649 	free = (tx->tx_cons + (1 << MCX_LOG_SQ_SIZE)) - tx->tx_prod;
7650 
7651 	used = 0;
7652 	bf = NULL;
7653 
7654 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&tx->tx_sq_mem),
7655 	    0, MCX_DMA_LEN(&tx->tx_sq_mem), BUS_DMASYNC_POSTWRITE);
7656 
7657 	sq = (struct mcx_sq_entry *)MCX_DMA_KVA(&tx->tx_sq_mem);
7658 
7659 	for (;;) {
7660 		if (used + MCX_SQ_ENTRY_MAX_SLOTS >= free) {
7661 			ifq_set_oactive(ifq);
7662 			break;
7663 		}
7664 
7665 		m = ifq_dequeue(ifq);
7666 		if (m == NULL) {
7667 			break;
7668 		}
7669 
7670 		sqe = sq + idx;
7671 		ms = &tx->tx_slots[idx];
7672 		memset(sqe, 0, sizeof(*sqe));
7673 
7674 		/* ctrl segment */
7675 		sqe->sqe_opcode_index = htobe32(MCX_SQE_WQE_OPCODE_SEND |
7676 		    ((tx->tx_prod & 0xffff) << MCX_SQE_WQE_INDEX_SHIFT));
7677 		/* always generate a completion event */
7678 		sqe->sqe_signature = htobe32(MCX_SQE_CE_CQE_ALWAYS);
7679 
7680 		/* eth segment */
7681 		csum = 0;
7682 		if (m->m_pkthdr.csum_flags & M_IPV4_CSUM_OUT)
7683 			csum |= MCX_SQE_L3_CSUM;
7684 		if (m->m_pkthdr.csum_flags & (M_TCP_CSUM_OUT | M_UDP_CSUM_OUT))
7685 			csum |= MCX_SQE_L4_CSUM;
7686 		sqe->sqe_mss_csum = htobe32(csum);
7687 		sqe->sqe_inline_header_size = htobe16(MCX_SQ_INLINE_SIZE);
7688 #if NVLAN > 0
7689 		if (m->m_flags & M_VLANTAG) {
7690 			struct ether_vlan_header *evh;
7691 			evh = (struct ether_vlan_header *)
7692 			    &sqe->sqe_inline_headers;
7693 
7694 			/* slightly cheaper vlan_inject() */
7695 			m_copydata(m, 0, ETHER_HDR_LEN, (caddr_t)evh);
7696 			evh->evl_proto = evh->evl_encap_proto;
7697 			evh->evl_encap_proto = htons(ETHERTYPE_VLAN);
7698 			evh->evl_tag = htons(m->m_pkthdr.ether_vtag);
7699 
7700 			m_adj(m, ETHER_HDR_LEN);
7701 		} else
7702 #endif
7703 		{
7704 			m_copydata(m, 0, MCX_SQ_INLINE_SIZE,
7705 			    (caddr_t)sqe->sqe_inline_headers);
7706 			m_adj(m, MCX_SQ_INLINE_SIZE);
7707 		}
7708 
7709 		if (mcx_load_mbuf(sc, ms, m) != 0) {
7710 			m_freem(m);
7711 			ifp->if_oerrors++;
7712 			continue;
7713 		}
7714 		bf = (uint64_t *)sqe;
7715 
7716 #if NBPFILTER > 0
7717 		if (ifp->if_bpf)
7718 			bpf_mtap_hdr(ifp->if_bpf,
7719 			    (caddr_t)sqe->sqe_inline_headers,
7720 			    MCX_SQ_INLINE_SIZE, m, BPF_DIRECTION_OUT);
7721 #endif
7722 		map = ms->ms_map;
7723 		bus_dmamap_sync(sc->sc_dmat, map, 0, map->dm_mapsize,
7724 		    BUS_DMASYNC_PREWRITE);
7725 
7726 		sqe->sqe_ds_sq_num =
7727 		    htobe32((tx->tx_sqn << MCX_SQE_SQ_NUM_SHIFT) |
7728 		    (map->dm_nsegs + 3));
7729 
7730 		/* data segment - first wqe has one segment */
7731 		sqs = sqe->sqe_segs;
7732 		seg = 0;
7733 		nseg = 1;
7734 		for (i = 0; i < map->dm_nsegs; i++) {
7735 			if (seg == nseg) {
7736 				/* next slot */
7737 				idx++;
7738 				if (idx == (1 << MCX_LOG_SQ_SIZE))
7739 					idx = 0;
7740 				tx->tx_prod++;
7741 				used++;
7742 
7743 				sqs = (struct mcx_sq_entry_seg *)(sq + idx);
7744 				seg = 0;
7745 				nseg = MCX_SQ_SEGS_PER_SLOT;
7746 			}
7747 			sqs[seg].sqs_byte_count =
7748 			    htobe32(map->dm_segs[i].ds_len);
7749 			sqs[seg].sqs_lkey = htobe32(sc->sc_lkey);
7750 			sqs[seg].sqs_addr = htobe64(map->dm_segs[i].ds_addr);
7751 			seg++;
7752 		}
7753 
7754 		idx++;
7755 		if (idx == (1 << MCX_LOG_SQ_SIZE))
7756 			idx = 0;
7757 		tx->tx_prod++;
7758 		used++;
7759 	}
7760 
7761 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&tx->tx_sq_mem),
7762 	    0, MCX_DMA_LEN(&tx->tx_sq_mem), BUS_DMASYNC_PREWRITE);
7763 
7764 	if (used) {
7765 		bus_size_t blueflame;
7766 
7767 		bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_doorbell_mem),
7768 		    tx->tx_doorbell, sizeof(uint32_t), BUS_DMASYNC_POSTWRITE);
7769 		htobem32(MCX_DMA_OFF(&sc->sc_doorbell_mem, tx->tx_doorbell),
7770 		    tx->tx_prod & MCX_WQ_DOORBELL_MASK);
7771 		bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_doorbell_mem),
7772 		    tx->tx_doorbell, sizeof(uint32_t), BUS_DMASYNC_PREWRITE);
7773 
7774 		/*
7775 		 * write the first 64 bits of the last sqe we produced
7776 		 * to the blue flame buffer
7777 		 */
7778 
7779 		blueflame = bf_base + tx->tx_bf_offset;
7780 		bus_space_write_raw_8(sc->sc_memt, sc->sc_memh,
7781 		    blueflame, *bf);
7782 		mcx_bar(sc, blueflame, sizeof(*bf), BUS_SPACE_BARRIER_WRITE);
7783 
7784 		/* next write goes to the other buffer */
7785 		tx->tx_bf_offset ^= sc->sc_bf_size;
7786 	}
7787 }
7788 
7789 static void
7790 mcx_watchdog(struct ifnet *ifp)
7791 {
7792 }
7793 
7794 static void
7795 mcx_media_add_types(struct mcx_softc *sc)
7796 {
7797 	struct mcx_reg_ptys ptys;
7798 	int i;
7799 	uint32_t proto_cap;
7800 
7801 	memset(&ptys, 0, sizeof(ptys));
7802 	ptys.rp_local_port = 1;
7803 	ptys.rp_proto_mask = MCX_REG_PTYS_PROTO_MASK_ETH;
7804 	if (mcx_access_hca_reg(sc, MCX_REG_PTYS, MCX_REG_OP_READ, &ptys,
7805 	    sizeof(ptys)) != 0) {
7806 		printf("%s: unable to read port type/speed\n", DEVNAME(sc));
7807 		return;
7808 	}
7809 
7810 	proto_cap = betoh32(ptys.rp_eth_proto_cap);
7811 	for (i = 0; i < nitems(mcx_eth_cap_map); i++) {
7812 		const struct mcx_eth_proto_capability *cap;
7813 		if (!ISSET(proto_cap, 1 << i))
7814 			continue;
7815 
7816 		cap = &mcx_eth_cap_map[i];
7817 		if (cap->cap_media == 0)
7818 			continue;
7819 
7820 		ifmedia_add(&sc->sc_media, IFM_ETHER | cap->cap_media, 0, NULL);
7821 	}
7822 }
7823 
7824 static void
7825 mcx_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
7826 {
7827 	struct mcx_softc *sc = (struct mcx_softc *)ifp->if_softc;
7828 	struct mcx_reg_ptys ptys;
7829 	int i;
7830 	uint32_t proto_oper;
7831 	uint64_t media_oper;
7832 
7833 	memset(&ptys, 0, sizeof(ptys));
7834 	ptys.rp_local_port = 1;
7835 	ptys.rp_proto_mask = MCX_REG_PTYS_PROTO_MASK_ETH;
7836 
7837 	if (mcx_access_hca_reg(sc, MCX_REG_PTYS, MCX_REG_OP_READ, &ptys,
7838 	    sizeof(ptys)) != 0) {
7839 		printf("%s: unable to read port type/speed\n", DEVNAME(sc));
7840 		return;
7841 	}
7842 
7843 	proto_oper = betoh32(ptys.rp_eth_proto_oper);
7844 
7845 	media_oper = 0;
7846 
7847 	for (i = 0; i < nitems(mcx_eth_cap_map); i++) {
7848 		const struct mcx_eth_proto_capability *cap;
7849 		if (!ISSET(proto_oper, 1 << i))
7850 			continue;
7851 
7852 		cap = &mcx_eth_cap_map[i];
7853 
7854 		if (cap->cap_media != 0)
7855 			media_oper = cap->cap_media;
7856 	}
7857 
7858 	ifmr->ifm_status = IFM_AVALID;
7859 	if (proto_oper != 0) {
7860 		ifmr->ifm_status |= IFM_ACTIVE;
7861 		ifmr->ifm_active = IFM_ETHER | IFM_AUTO | media_oper;
7862 		/* txpause, rxpause, duplex? */
7863 	}
7864 }
7865 
7866 static int
7867 mcx_media_change(struct ifnet *ifp)
7868 {
7869 	struct mcx_softc *sc = (struct mcx_softc *)ifp->if_softc;
7870 	struct mcx_reg_ptys ptys;
7871 	struct mcx_reg_paos paos;
7872 	uint32_t media;
7873 	int i, error;
7874 
7875 	if (IFM_TYPE(sc->sc_media.ifm_media) != IFM_ETHER)
7876 		return EINVAL;
7877 
7878 	error = 0;
7879 
7880 	if (IFM_SUBTYPE(sc->sc_media.ifm_media) == IFM_AUTO) {
7881 		/* read ptys to get supported media */
7882 		memset(&ptys, 0, sizeof(ptys));
7883 		ptys.rp_local_port = 1;
7884 		ptys.rp_proto_mask = MCX_REG_PTYS_PROTO_MASK_ETH;
7885 		if (mcx_access_hca_reg(sc, MCX_REG_PTYS, MCX_REG_OP_READ,
7886 		    &ptys, sizeof(ptys)) != 0) {
7887 			printf("%s: unable to read port type/speed\n",
7888 			    DEVNAME(sc));
7889 			return EIO;
7890 		}
7891 
7892 		media = betoh32(ptys.rp_eth_proto_cap);
7893 	} else {
7894 		/* map media type */
7895 		media = 0;
7896 		for (i = 0; i < nitems(mcx_eth_cap_map); i++) {
7897 			const struct  mcx_eth_proto_capability *cap;
7898 
7899 			cap = &mcx_eth_cap_map[i];
7900 			if (cap->cap_media ==
7901 			    IFM_SUBTYPE(sc->sc_media.ifm_media)) {
7902 				media = (1 << i);
7903 				break;
7904 			}
7905 		}
7906 	}
7907 
7908 	/* disable the port */
7909 	memset(&paos, 0, sizeof(paos));
7910 	paos.rp_local_port = 1;
7911 	paos.rp_admin_status = MCX_REG_PAOS_ADMIN_STATUS_DOWN;
7912 	paos.rp_admin_state_update = MCX_REG_PAOS_ADMIN_STATE_UPDATE_EN;
7913 	if (mcx_access_hca_reg(sc, MCX_REG_PAOS, MCX_REG_OP_WRITE, &paos,
7914 	    sizeof(paos)) != 0) {
7915 		printf("%s: unable to set port state to down\n", DEVNAME(sc));
7916 		return EIO;
7917 	}
7918 
7919 	memset(&ptys, 0, sizeof(ptys));
7920 	ptys.rp_local_port = 1;
7921 	ptys.rp_proto_mask = MCX_REG_PTYS_PROTO_MASK_ETH;
7922 	ptys.rp_eth_proto_admin = htobe32(media);
7923 	if (mcx_access_hca_reg(sc, MCX_REG_PTYS, MCX_REG_OP_WRITE, &ptys,
7924 	    sizeof(ptys)) != 0) {
7925 		printf("%s: unable to set port media type/speed\n",
7926 		    DEVNAME(sc));
7927 		error = EIO;
7928 	}
7929 
7930 	/* re-enable the port to start negotiation */
7931 	memset(&paos, 0, sizeof(paos));
7932 	paos.rp_local_port = 1;
7933 	paos.rp_admin_status = MCX_REG_PAOS_ADMIN_STATUS_UP;
7934 	paos.rp_admin_state_update = MCX_REG_PAOS_ADMIN_STATE_UPDATE_EN;
7935 	if (mcx_access_hca_reg(sc, MCX_REG_PAOS, MCX_REG_OP_WRITE, &paos,
7936 	    sizeof(paos)) != 0) {
7937 		printf("%s: unable to set port state to up\n", DEVNAME(sc));
7938 		error = EIO;
7939 	}
7940 
7941 	return error;
7942 }
7943 
7944 static void
7945 mcx_port_change(void *xsc)
7946 {
7947 	struct mcx_softc *sc = xsc;
7948 	struct ifnet *ifp = &sc->sc_ac.ac_if;
7949 	struct mcx_reg_ptys ptys = {
7950 		.rp_local_port = 1,
7951 		.rp_proto_mask = MCX_REG_PTYS_PROTO_MASK_ETH,
7952 	};
7953 	int link_state = LINK_STATE_DOWN;
7954 
7955 	if (mcx_access_hca_reg(sc, MCX_REG_PTYS, MCX_REG_OP_READ, &ptys,
7956 	    sizeof(ptys)) == 0) {
7957 		uint32_t proto_oper = betoh32(ptys.rp_eth_proto_oper);
7958 		uint64_t baudrate = 0;
7959 		unsigned int i;
7960 
7961 		if (proto_oper != 0)
7962 			link_state = LINK_STATE_FULL_DUPLEX;
7963 
7964 		for (i = 0; i < nitems(mcx_eth_cap_map); i++) {
7965 			const struct mcx_eth_proto_capability *cap;
7966 			if (!ISSET(proto_oper, 1 << i))
7967 				continue;
7968 
7969 			cap = &mcx_eth_cap_map[i];
7970 			if (cap->cap_baudrate == 0)
7971 				continue;
7972 
7973 			baudrate = cap->cap_baudrate;
7974 			break;
7975 		}
7976 
7977 		ifp->if_baudrate = baudrate;
7978 	}
7979 
7980 	if (link_state != ifp->if_link_state) {
7981 		ifp->if_link_state = link_state;
7982 		if_link_state_change(ifp);
7983 	}
7984 }
7985 
7986 static inline uint32_t
7987 mcx_rd(struct mcx_softc *sc, bus_size_t r)
7988 {
7989 	uint32_t word;
7990 
7991 	word = bus_space_read_raw_4(sc->sc_memt, sc->sc_memh, r);
7992 
7993 	return (betoh32(word));
7994 }
7995 
7996 static inline void
7997 mcx_wr(struct mcx_softc *sc, bus_size_t r, uint32_t v)
7998 {
7999 	bus_space_write_raw_4(sc->sc_memt, sc->sc_memh, r, htobe32(v));
8000 }
8001 
8002 static inline void
8003 mcx_bar(struct mcx_softc *sc, bus_size_t r, bus_size_t l, int f)
8004 {
8005 	bus_space_barrier(sc->sc_memt, sc->sc_memh, r, l, f);
8006 }
8007 
8008 static uint64_t
8009 mcx_timer(struct mcx_softc *sc)
8010 {
8011 	uint32_t hi, lo, ni;
8012 
8013 	hi = mcx_rd(sc, MCX_INTERNAL_TIMER_H);
8014 	for (;;) {
8015 		lo = mcx_rd(sc, MCX_INTERNAL_TIMER_L);
8016 		mcx_bar(sc, MCX_INTERNAL_TIMER_L, 8, BUS_SPACE_BARRIER_READ);
8017 		ni = mcx_rd(sc, MCX_INTERNAL_TIMER_H);
8018 
8019 		if (ni == hi)
8020 			break;
8021 
8022 		hi = ni;
8023 	}
8024 
8025 	return (((uint64_t)hi << 32) | (uint64_t)lo);
8026 }
8027 
8028 static int
8029 mcx_dmamem_alloc(struct mcx_softc *sc, struct mcx_dmamem *mxm,
8030     bus_size_t size, u_int align)
8031 {
8032 	mxm->mxm_size = size;
8033 
8034 	if (bus_dmamap_create(sc->sc_dmat, mxm->mxm_size, 1,
8035 	    mxm->mxm_size, 0,
8036 	    BUS_DMA_WAITOK | BUS_DMA_ALLOCNOW | BUS_DMA_64BIT,
8037 	    &mxm->mxm_map) != 0)
8038 		return (1);
8039 	if (bus_dmamem_alloc(sc->sc_dmat, mxm->mxm_size,
8040 	    align, 0, &mxm->mxm_seg, 1, &mxm->mxm_nsegs,
8041 	    BUS_DMA_WAITOK | BUS_DMA_ZERO) != 0)
8042 		goto destroy;
8043 	if (bus_dmamem_map(sc->sc_dmat, &mxm->mxm_seg, mxm->mxm_nsegs,
8044 	    mxm->mxm_size, &mxm->mxm_kva, BUS_DMA_WAITOK) != 0)
8045 		goto free;
8046 	if (bus_dmamap_load(sc->sc_dmat, mxm->mxm_map, mxm->mxm_kva,
8047 	    mxm->mxm_size, NULL, BUS_DMA_WAITOK) != 0)
8048 		goto unmap;
8049 
8050 	return (0);
8051 unmap:
8052 	bus_dmamem_unmap(sc->sc_dmat, mxm->mxm_kva, mxm->mxm_size);
8053 free:
8054 	bus_dmamem_free(sc->sc_dmat, &mxm->mxm_seg, 1);
8055 destroy:
8056 	bus_dmamap_destroy(sc->sc_dmat, mxm->mxm_map);
8057 	return (1);
8058 }
8059 
8060 static void
8061 mcx_dmamem_zero(struct mcx_dmamem *mxm)
8062 {
8063 	memset(MCX_DMA_KVA(mxm), 0, MCX_DMA_LEN(mxm));
8064 }
8065 
8066 static void
8067 mcx_dmamem_free(struct mcx_softc *sc, struct mcx_dmamem *mxm)
8068 {
8069 	bus_dmamap_unload(sc->sc_dmat, mxm->mxm_map);
8070 	bus_dmamem_unmap(sc->sc_dmat, mxm->mxm_kva, mxm->mxm_size);
8071 	bus_dmamem_free(sc->sc_dmat, &mxm->mxm_seg, 1);
8072 	bus_dmamap_destroy(sc->sc_dmat, mxm->mxm_map);
8073 }
8074 
8075 static int
8076 mcx_hwmem_alloc(struct mcx_softc *sc, struct mcx_hwmem *mhm, unsigned int pages)
8077 {
8078 	bus_dma_segment_t *segs;
8079 	bus_size_t len = pages * MCX_PAGE_SIZE;
8080 	size_t seglen;
8081 
8082 	segs = mallocarray(sizeof(*segs), pages, M_DEVBUF, M_WAITOK|M_CANFAIL);
8083 	if (segs == NULL)
8084 		return (-1);
8085 
8086 	seglen = sizeof(*segs) * pages;
8087 
8088 	if (bus_dmamem_alloc(sc->sc_dmat, len, MCX_PAGE_SIZE, 0,
8089 	    segs, pages, &mhm->mhm_seg_count, BUS_DMA_NOWAIT) != 0)
8090 		goto free_segs;
8091 
8092 	if (mhm->mhm_seg_count < pages) {
8093 		size_t nseglen;
8094 
8095 		mhm->mhm_segs = mallocarray(sizeof(*mhm->mhm_segs),
8096 		    mhm->mhm_seg_count, M_DEVBUF, M_WAITOK|M_CANFAIL);
8097 		if (mhm->mhm_segs == NULL)
8098 			goto free_dmamem;
8099 
8100 		nseglen = sizeof(*mhm->mhm_segs) * mhm->mhm_seg_count;
8101 
8102 		memcpy(mhm->mhm_segs, segs, nseglen);
8103 
8104 		free(segs, M_DEVBUF, seglen);
8105 
8106 		segs = mhm->mhm_segs;
8107 		seglen = nseglen;
8108 	} else
8109 		mhm->mhm_segs = segs;
8110 
8111 	if (bus_dmamap_create(sc->sc_dmat, len, pages, MCX_PAGE_SIZE,
8112 	    MCX_PAGE_SIZE, BUS_DMA_NOWAIT|BUS_DMA_ALLOCNOW /*|BUS_DMA_64BIT*/,
8113 	    &mhm->mhm_map) != 0)
8114 		goto free_dmamem;
8115 
8116 	if (bus_dmamap_load_raw(sc->sc_dmat, mhm->mhm_map,
8117 	    mhm->mhm_segs, mhm->mhm_seg_count, len, BUS_DMA_NOWAIT) != 0)
8118 		goto destroy;
8119 
8120 	bus_dmamap_sync(sc->sc_dmat, mhm->mhm_map,
8121 	    0, mhm->mhm_map->dm_mapsize, BUS_DMASYNC_PRERW);
8122 
8123 	mhm->mhm_npages = pages;
8124 
8125 	return (0);
8126 
8127 destroy:
8128 	bus_dmamap_destroy(sc->sc_dmat, mhm->mhm_map);
8129 free_dmamem:
8130 	bus_dmamem_free(sc->sc_dmat, mhm->mhm_segs, mhm->mhm_seg_count);
8131 free_segs:
8132 	free(segs, M_DEVBUF, seglen);
8133 	mhm->mhm_segs = NULL;
8134 
8135 	return (-1);
8136 }
8137 
8138 static void
8139 mcx_hwmem_free(struct mcx_softc *sc, struct mcx_hwmem *mhm)
8140 {
8141 	if (mhm->mhm_npages == 0)
8142 		return;
8143 
8144 	bus_dmamap_sync(sc->sc_dmat, mhm->mhm_map,
8145 	    0, mhm->mhm_map->dm_mapsize, BUS_DMASYNC_POSTRW);
8146 
8147 	bus_dmamap_unload(sc->sc_dmat, mhm->mhm_map);
8148 	bus_dmamap_destroy(sc->sc_dmat, mhm->mhm_map);
8149 	bus_dmamem_free(sc->sc_dmat, mhm->mhm_segs, mhm->mhm_seg_count);
8150 	free(mhm->mhm_segs, M_DEVBUF,
8151 	    sizeof(*mhm->mhm_segs) * mhm->mhm_seg_count);
8152 
8153 	mhm->mhm_npages = 0;
8154 }
8155 
8156 #if NKSTAT > 0
8157 struct mcx_ppcnt {
8158 	char			 name[KSTAT_KV_NAMELEN];
8159 	enum kstat_kv_unit	 unit;
8160 };
8161 
8162 static const struct mcx_ppcnt mcx_ppcnt_ieee8023_tpl[] = {
8163 	{ "Good Tx",		KSTAT_KV_U_PACKETS, },
8164 	{ "Good Rx",		KSTAT_KV_U_PACKETS, },
8165 	{ "FCS errs",		KSTAT_KV_U_PACKETS, },
8166 	{ "Alignment Errs",	KSTAT_KV_U_PACKETS, },
8167 	{ "Good Tx",		KSTAT_KV_U_BYTES, },
8168 	{ "Good Rx",		KSTAT_KV_U_BYTES, },
8169 	{ "Multicast Tx",	KSTAT_KV_U_PACKETS, },
8170 	{ "Broadcast Tx",	KSTAT_KV_U_PACKETS, },
8171 	{ "Multicast Rx",	KSTAT_KV_U_PACKETS, },
8172 	{ "Broadcast Rx",	KSTAT_KV_U_PACKETS, },
8173 	{ "In Range Len",	KSTAT_KV_U_PACKETS, },
8174 	{ "Out Of Range Len",	KSTAT_KV_U_PACKETS, },
8175 	{ "Frame Too Long",	KSTAT_KV_U_PACKETS, },
8176 	{ "Symbol Errs",	KSTAT_KV_U_PACKETS, },
8177 	{ "MAC Ctrl Tx",	KSTAT_KV_U_PACKETS, },
8178 	{ "MAC Ctrl Rx",	KSTAT_KV_U_PACKETS, },
8179 	{ "MAC Ctrl Unsup",	KSTAT_KV_U_PACKETS, },
8180 	{ "Pause Rx",		KSTAT_KV_U_PACKETS, },
8181 	{ "Pause Tx",		KSTAT_KV_U_PACKETS, },
8182 };
8183 CTASSERT(nitems(mcx_ppcnt_ieee8023_tpl) == mcx_ppcnt_ieee8023_count);
8184 
8185 static const struct mcx_ppcnt mcx_ppcnt_rfc2863_tpl[] = {
8186 	{ "Rx Bytes",		KSTAT_KV_U_BYTES, },
8187 	{ "Rx Unicast",		KSTAT_KV_U_PACKETS, },
8188 	{ "Rx Discards",	KSTAT_KV_U_PACKETS, },
8189 	{ "Rx Errors",		KSTAT_KV_U_PACKETS, },
8190 	{ "Rx Unknown Proto",	KSTAT_KV_U_PACKETS, },
8191 	{ "Tx Bytes",		KSTAT_KV_U_BYTES, },
8192 	{ "Tx Unicast",		KSTAT_KV_U_PACKETS, },
8193 	{ "Tx Discards",	KSTAT_KV_U_PACKETS, },
8194 	{ "Tx Errors",		KSTAT_KV_U_PACKETS, },
8195 	{ "Rx Multicast",	KSTAT_KV_U_PACKETS, },
8196 	{ "Rx Broadcast",	KSTAT_KV_U_PACKETS, },
8197 	{ "Tx Multicast",	KSTAT_KV_U_PACKETS, },
8198 	{ "Tx Broadcast",	KSTAT_KV_U_PACKETS, },
8199 };
8200 CTASSERT(nitems(mcx_ppcnt_rfc2863_tpl) == mcx_ppcnt_rfc2863_count);
8201 
8202 static const struct mcx_ppcnt mcx_ppcnt_rfc2819_tpl[] = {
8203 	{ "Drop Events",	KSTAT_KV_U_PACKETS, },
8204 	{ "Octets",		KSTAT_KV_U_BYTES, },
8205 	{ "Packets",		KSTAT_KV_U_PACKETS, },
8206 	{ "Broadcasts",		KSTAT_KV_U_PACKETS, },
8207 	{ "Multicasts",		KSTAT_KV_U_PACKETS, },
8208 	{ "CRC Align Errs",	KSTAT_KV_U_PACKETS, },
8209 	{ "Undersize",		KSTAT_KV_U_PACKETS, },
8210 	{ "Oversize",		KSTAT_KV_U_PACKETS, },
8211 	{ "Fragments",		KSTAT_KV_U_PACKETS, },
8212 	{ "Jabbers",		KSTAT_KV_U_PACKETS, },
8213 	{ "Collisions",		KSTAT_KV_U_NONE, },
8214 	{ "64B",		KSTAT_KV_U_PACKETS, },
8215 	{ "65-127B",		KSTAT_KV_U_PACKETS, },
8216 	{ "128-255B",		KSTAT_KV_U_PACKETS, },
8217 	{ "256-511B",		KSTAT_KV_U_PACKETS, },
8218 	{ "512-1023B",		KSTAT_KV_U_PACKETS, },
8219 	{ "1024-1518B",		KSTAT_KV_U_PACKETS, },
8220 	{ "1519-2047B",		KSTAT_KV_U_PACKETS, },
8221 	{ "2048-4095B",		KSTAT_KV_U_PACKETS, },
8222 	{ "4096-8191B",		KSTAT_KV_U_PACKETS, },
8223 	{ "8192-10239B",	KSTAT_KV_U_PACKETS, },
8224 };
8225 CTASSERT(nitems(mcx_ppcnt_rfc2819_tpl) == mcx_ppcnt_rfc2819_count);
8226 
8227 static const struct mcx_ppcnt mcx_ppcnt_rfc3635_tpl[] = {
8228 	{ "Alignment Errs",	KSTAT_KV_U_PACKETS, },
8229 	{ "FCS Errs",		KSTAT_KV_U_PACKETS, },
8230 	{ "Single Colls",	KSTAT_KV_U_PACKETS, },
8231 	{ "Multiple Colls",	KSTAT_KV_U_PACKETS, },
8232 	{ "SQE Test Errs",	KSTAT_KV_U_NONE, },
8233 	{ "Deferred Tx",	KSTAT_KV_U_PACKETS, },
8234 	{ "Late Colls",		KSTAT_KV_U_NONE, },
8235 	{ "Exess Colls",	KSTAT_KV_U_NONE, },
8236 	{ "Int MAC Tx Errs",	KSTAT_KV_U_PACKETS, },
8237 	{ "CSM Sense Errs",	KSTAT_KV_U_NONE, },
8238 	{ "Too Long",		KSTAT_KV_U_PACKETS, },
8239 	{ "Int MAC Rx Errs",	KSTAT_KV_U_PACKETS, },
8240 	{ "Symbol Errs",	KSTAT_KV_U_NONE, },
8241 	{ "Unknown Control",	KSTAT_KV_U_PACKETS, },
8242 	{ "Pause Rx",		KSTAT_KV_U_PACKETS, },
8243 	{ "Pause Tx",		KSTAT_KV_U_PACKETS, },
8244 };
8245 CTASSERT(nitems(mcx_ppcnt_rfc3635_tpl) == mcx_ppcnt_rfc3635_count);
8246 
8247 struct mcx_kstat_ppcnt {
8248 	const char		*ksp_name;
8249 	const struct mcx_ppcnt	*ksp_tpl;
8250 	unsigned int		 ksp_n;
8251 	uint8_t			 ksp_grp;
8252 };
8253 
8254 static const struct mcx_kstat_ppcnt mcx_kstat_ppcnt_ieee8023 = {
8255 	.ksp_name =		"ieee802.3",
8256 	.ksp_tpl =		mcx_ppcnt_ieee8023_tpl,
8257 	.ksp_n =		nitems(mcx_ppcnt_ieee8023_tpl),
8258 	.ksp_grp =		MCX_REG_PPCNT_GRP_IEEE8023,
8259 };
8260 
8261 static const struct mcx_kstat_ppcnt mcx_kstat_ppcnt_rfc2863 = {
8262 	.ksp_name =		"rfc2863",
8263 	.ksp_tpl =		mcx_ppcnt_rfc2863_tpl,
8264 	.ksp_n =		nitems(mcx_ppcnt_rfc2863_tpl),
8265 	.ksp_grp =		MCX_REG_PPCNT_GRP_RFC2863,
8266 };
8267 
8268 static const struct mcx_kstat_ppcnt mcx_kstat_ppcnt_rfc2819 = {
8269 	.ksp_name =		"rfc2819",
8270 	.ksp_tpl =		mcx_ppcnt_rfc2819_tpl,
8271 	.ksp_n =		nitems(mcx_ppcnt_rfc2819_tpl),
8272 	.ksp_grp =		MCX_REG_PPCNT_GRP_RFC2819,
8273 };
8274 
8275 static const struct mcx_kstat_ppcnt mcx_kstat_ppcnt_rfc3635 = {
8276 	.ksp_name =		"rfc3635",
8277 	.ksp_tpl =		mcx_ppcnt_rfc3635_tpl,
8278 	.ksp_n =		nitems(mcx_ppcnt_rfc3635_tpl),
8279 	.ksp_grp =		MCX_REG_PPCNT_GRP_RFC3635,
8280 };
8281 
8282 static int	mcx_kstat_ppcnt_read(struct kstat *);
8283 
8284 static void	mcx_kstat_attach_tmps(struct mcx_softc *sc);
8285 static void	mcx_kstat_attach_queues(struct mcx_softc *sc);
8286 
8287 static struct kstat *
8288 mcx_kstat_attach_ppcnt(struct mcx_softc *sc,
8289     const struct mcx_kstat_ppcnt *ksp)
8290 {
8291 	struct kstat *ks;
8292 	struct kstat_kv *kvs;
8293 	unsigned int i;
8294 
8295 	ks = kstat_create(DEVNAME(sc), 0, ksp->ksp_name, 0, KSTAT_T_KV, 0);
8296 	if (ks == NULL)
8297 		return (NULL);
8298 
8299 	kvs = mallocarray(ksp->ksp_n, sizeof(*kvs),
8300 	    M_DEVBUF, M_WAITOK);
8301 
8302 	for (i = 0; i < ksp->ksp_n; i++) {
8303 		const struct mcx_ppcnt *tpl = &ksp->ksp_tpl[i];
8304 
8305 		kstat_kv_unit_init(&kvs[i], tpl->name,
8306 		    KSTAT_KV_T_COUNTER64, tpl->unit);
8307 	}
8308 
8309 	ks->ks_softc = sc;
8310 	ks->ks_ptr = (void *)ksp;
8311 	ks->ks_data = kvs;
8312 	ks->ks_datalen = ksp->ksp_n * sizeof(*kvs);
8313 	ks->ks_read = mcx_kstat_ppcnt_read;
8314 
8315 	kstat_install(ks);
8316 
8317 	return (ks);
8318 }
8319 
8320 static void
8321 mcx_kstat_attach(struct mcx_softc *sc)
8322 {
8323 	sc->sc_kstat_ieee8023 = mcx_kstat_attach_ppcnt(sc,
8324 	    &mcx_kstat_ppcnt_ieee8023);
8325 	sc->sc_kstat_rfc2863 = mcx_kstat_attach_ppcnt(sc,
8326 	    &mcx_kstat_ppcnt_rfc2863);
8327 	sc->sc_kstat_rfc2819 = mcx_kstat_attach_ppcnt(sc,
8328 	    &mcx_kstat_ppcnt_rfc2819);
8329 	sc->sc_kstat_rfc3635 = mcx_kstat_attach_ppcnt(sc,
8330 	    &mcx_kstat_ppcnt_rfc3635);
8331 
8332 	mcx_kstat_attach_tmps(sc);
8333 	mcx_kstat_attach_queues(sc);
8334 }
8335 
8336 static int
8337 mcx_kstat_ppcnt_read(struct kstat *ks)
8338 {
8339 	struct mcx_softc *sc = ks->ks_softc;
8340 	struct mcx_kstat_ppcnt *ksp = ks->ks_ptr;
8341 	struct mcx_reg_ppcnt ppcnt = {
8342 		.ppcnt_grp = ksp->ksp_grp,
8343 		.ppcnt_local_port = 1,
8344 	};
8345 	struct kstat_kv *kvs = ks->ks_data;
8346 	uint64_t *vs = (uint64_t *)&ppcnt.ppcnt_counter_set;
8347 	unsigned int i;
8348 	int rv;
8349 
8350 	KERNEL_LOCK(); /* XXX */
8351 	rv = mcx_access_hca_reg(sc, MCX_REG_PPCNT, MCX_REG_OP_READ,
8352 	    &ppcnt, sizeof(ppcnt));
8353 	KERNEL_UNLOCK();
8354 	if (rv != 0)
8355 		return (EIO);
8356 
8357 	nanouptime(&ks->ks_updated);
8358 
8359 	for (i = 0; i < ksp->ksp_n; i++)
8360 		kstat_kv_u64(&kvs[i]) = bemtoh64(&vs[i]);
8361 
8362 	return (0);
8363 }
8364 
8365 struct mcx_kstat_mtmp {
8366 	struct kstat_kv		ktmp_name;
8367 	struct kstat_kv		ktmp_temperature;
8368 	struct kstat_kv		ktmp_threshold_lo;
8369 	struct kstat_kv		ktmp_threshold_hi;
8370 };
8371 
8372 static const struct mcx_kstat_mtmp mcx_kstat_mtmp_tpl = {
8373 	KSTAT_KV_INITIALIZER("name",		KSTAT_KV_T_ISTR),
8374 	KSTAT_KV_INITIALIZER("temperature",	KSTAT_KV_T_TEMP),
8375 	KSTAT_KV_INITIALIZER("lo threshold",	KSTAT_KV_T_TEMP),
8376 	KSTAT_KV_INITIALIZER("hi threshold",	KSTAT_KV_T_TEMP),
8377 };
8378 
8379 static const struct timeval mcx_kstat_mtmp_rate = { 1, 0 };
8380 
8381 static int mcx_kstat_mtmp_read(struct kstat *);
8382 
8383 static void
8384 mcx_kstat_attach_tmps(struct mcx_softc *sc)
8385 {
8386 	struct kstat *ks;
8387 	struct mcx_reg_mtcap mtcap;
8388 	struct mcx_kstat_mtmp *ktmp;
8389 	uint64_t map;
8390 	unsigned int i, n;
8391 
8392 	memset(&mtcap, 0, sizeof(mtcap));
8393 
8394 	if (mcx_access_hca_reg(sc, MCX_REG_MTCAP, MCX_REG_OP_READ,
8395 	    &mtcap, sizeof(mtcap)) != 0) {
8396 		/* unable to find temperature sensors */
8397 		return;
8398 	}
8399 
8400 	sc->sc_kstat_mtmp_count = mtcap.mtcap_sensor_count;
8401 	sc->sc_kstat_mtmp = mallocarray(sc->sc_kstat_mtmp_count,
8402 	    sizeof(*sc->sc_kstat_mtmp), M_DEVBUF, M_WAITOK);
8403 
8404 	n = 0;
8405 	map = bemtoh64(&mtcap.mtcap_sensor_map);
8406 	for (i = 0; i < sizeof(map) * NBBY; i++) {
8407 		if (!ISSET(map, (1ULL << i)))
8408 			continue;
8409 
8410 		ks = kstat_create(DEVNAME(sc), 0, "temperature", i,
8411 		    KSTAT_T_KV, 0);
8412 		if (ks == NULL) {
8413 			/* unable to attach temperature sensor %u, i */
8414 			continue;
8415 		}
8416 
8417 		ktmp = malloc(sizeof(*ktmp), M_DEVBUF, M_WAITOK|M_ZERO);
8418 		*ktmp = mcx_kstat_mtmp_tpl;
8419 
8420 		ks->ks_data = ktmp;
8421 		ks->ks_datalen = sizeof(*ktmp);
8422 		TIMEVAL_TO_TIMESPEC(&mcx_kstat_mtmp_rate, &ks->ks_interval);
8423 		ks->ks_read = mcx_kstat_mtmp_read;
8424 
8425 		ks->ks_softc = sc;
8426 		kstat_install(ks);
8427 
8428 		sc->sc_kstat_mtmp[n++] = ks;
8429 		if (n >= sc->sc_kstat_mtmp_count)
8430 			break;
8431 	}
8432 }
8433 
8434 static uint64_t
8435 mcx_tmp_to_uK(uint16_t *t)
8436 {
8437 	int64_t mt = (int16_t)bemtoh16(t); /* 0.125 C units */
8438 	mt *= 1000000 / 8; /* convert to uC */
8439 	mt += 273150000; /* convert to uK */
8440 
8441 	return (mt);
8442 }
8443 
8444 static int
8445 mcx_kstat_mtmp_read(struct kstat *ks)
8446 {
8447 	struct mcx_softc *sc = ks->ks_softc;
8448 	struct mcx_kstat_mtmp *ktmp = ks->ks_data;
8449 	struct mcx_reg_mtmp mtmp;
8450 	int rv;
8451 	struct timeval updated;
8452 
8453 	TIMESPEC_TO_TIMEVAL(&updated, &ks->ks_updated);
8454 
8455 	if (!ratecheck(&updated, &mcx_kstat_mtmp_rate))
8456 		return (0);
8457 
8458 	memset(&mtmp, 0, sizeof(mtmp));
8459 	htobem16(&mtmp.mtmp_sensor_index, ks->ks_unit);
8460 
8461 	KERNEL_LOCK(); /* XXX */
8462 	rv = mcx_access_hca_reg(sc, MCX_REG_MTMP, MCX_REG_OP_READ,
8463 	    &mtmp, sizeof(mtmp));
8464 	KERNEL_UNLOCK();
8465 	if (rv != 0)
8466 		return (EIO);
8467 
8468 	memset(kstat_kv_istr(&ktmp->ktmp_name), 0,
8469 	    sizeof(kstat_kv_istr(&ktmp->ktmp_name)));
8470 	memcpy(kstat_kv_istr(&ktmp->ktmp_name),
8471 	    mtmp.mtmp_sensor_name, sizeof(mtmp.mtmp_sensor_name));
8472 	kstat_kv_temp(&ktmp->ktmp_temperature) =
8473 	    mcx_tmp_to_uK(&mtmp.mtmp_temperature);
8474 	kstat_kv_temp(&ktmp->ktmp_threshold_lo) =
8475 	    mcx_tmp_to_uK(&mtmp.mtmp_temperature_threshold_lo);
8476 	kstat_kv_temp(&ktmp->ktmp_threshold_hi) =
8477 	    mcx_tmp_to_uK(&mtmp.mtmp_temperature_threshold_hi);
8478 
8479 	TIMEVAL_TO_TIMESPEC(&updated, &ks->ks_updated);
8480 
8481 	return (0);
8482 }
8483 
8484 struct mcx_queuestat {
8485 	char			 name[KSTAT_KV_NAMELEN];
8486 	enum kstat_kv_type	 type;
8487 };
8488 
8489 static const struct mcx_queuestat mcx_queue_kstat_tpl[] = {
8490 	{ "RQ SW prod",		KSTAT_KV_T_COUNTER64 },
8491 	{ "RQ HW prod",		KSTAT_KV_T_COUNTER64 },
8492 	{ "RQ HW cons",		KSTAT_KV_T_COUNTER64 },
8493 	{ "RQ HW state",	KSTAT_KV_T_ISTR },
8494 
8495 	{ "SQ SW prod",		KSTAT_KV_T_COUNTER64 },
8496 	{ "SQ SW cons",		KSTAT_KV_T_COUNTER64 },
8497 	{ "SQ HW prod",		KSTAT_KV_T_COUNTER64 },
8498 	{ "SQ HW cons",		KSTAT_KV_T_COUNTER64 },
8499 	{ "SQ HW state",	KSTAT_KV_T_ISTR },
8500 
8501 	{ "CQ SW cons",		KSTAT_KV_T_COUNTER64 },
8502 	{ "CQ HW prod",		KSTAT_KV_T_COUNTER64 },
8503 	{ "CQ HW cons",		KSTAT_KV_T_COUNTER64 },
8504 	{ "CQ HW notify",	KSTAT_KV_T_COUNTER64 },
8505 	{ "CQ HW solicit",	KSTAT_KV_T_COUNTER64 },
8506 	{ "CQ HW status",	KSTAT_KV_T_ISTR },
8507 	{ "CQ HW state",	KSTAT_KV_T_ISTR },
8508 
8509 	{ "EQ SW cons",		KSTAT_KV_T_COUNTER64 },
8510 	{ "EQ HW prod",		KSTAT_KV_T_COUNTER64 },
8511 	{ "EQ HW cons",		KSTAT_KV_T_COUNTER64 },
8512 	{ "EQ HW status",	KSTAT_KV_T_ISTR },
8513 	{ "EQ HW state",	KSTAT_KV_T_ISTR },
8514 };
8515 
8516 static int	mcx_kstat_queue_read(struct kstat *);
8517 
8518 static void
8519 mcx_kstat_attach_queues(struct mcx_softc *sc)
8520 {
8521 	struct kstat *ks;
8522 	struct kstat_kv *kvs;
8523 	int q, i;
8524 
8525 	for (q = 0; q < sc->sc_nqueues; q++) {
8526 		ks = kstat_create(DEVNAME(sc), 0, "mcx-queues", q,
8527 		    KSTAT_T_KV, 0);
8528 		if (ks == NULL) {
8529 			/* unable to attach queue stats %u, q */
8530 			continue;
8531 		}
8532 
8533 		kvs = mallocarray(nitems(mcx_queue_kstat_tpl),
8534 		    sizeof(*kvs), M_DEVBUF, M_WAITOK);
8535 
8536 		for (i = 0; i < nitems(mcx_queue_kstat_tpl); i++) {
8537 			const struct mcx_queuestat *tpl =
8538 			    &mcx_queue_kstat_tpl[i];
8539 
8540 			kstat_kv_init(&kvs[i], tpl->name, tpl->type);
8541 		}
8542 
8543 		ks->ks_softc = &sc->sc_queues[q];
8544 		ks->ks_data = kvs;
8545 		ks->ks_datalen = nitems(mcx_queue_kstat_tpl) * sizeof(*kvs);
8546 		ks->ks_read = mcx_kstat_queue_read;
8547 
8548 		sc->sc_queues[q].q_kstat = ks;
8549 		kstat_install(ks);
8550 	}
8551 }
8552 
8553 static int
8554 mcx_kstat_queue_read(struct kstat *ks)
8555 {
8556 	struct mcx_queues *q = ks->ks_softc;
8557 	struct mcx_softc *sc = q->q_sc;
8558 	struct kstat_kv *kvs = ks->ks_data;
8559 	union {
8560 		struct mcx_rq_ctx rq;
8561 		struct mcx_sq_ctx sq;
8562 		struct mcx_cq_ctx cq;
8563 		struct mcx_eq_ctx eq;
8564 	} u;
8565 	const char *text;
8566 	int error = 0;
8567 
8568 	KERNEL_LOCK();
8569 
8570 	if (mcx_query_rq(sc, &q->q_rx, &u.rq) != 0) {
8571 		error = EIO;
8572 		goto out;
8573 	}
8574 
8575 	kstat_kv_u64(kvs++) = q->q_rx.rx_prod;
8576 	kstat_kv_u64(kvs++) = bemtoh32(&u.rq.rq_wq.wq_sw_counter);
8577 	kstat_kv_u64(kvs++) = bemtoh32(&u.rq.rq_wq.wq_hw_counter);
8578 	switch ((bemtoh32(&u.rq.rq_flags) & MCX_RQ_CTX_STATE_MASK) >>
8579 	    MCX_RQ_CTX_STATE_SHIFT) {
8580 	case MCX_RQ_CTX_STATE_RST:
8581 		text = "RST";
8582 		break;
8583 	case MCX_RQ_CTX_STATE_RDY:
8584 		text = "RDY";
8585 		break;
8586 	case MCX_RQ_CTX_STATE_ERR:
8587 		text = "ERR";
8588 		break;
8589 	default:
8590 		text = "unknown";
8591 		break;
8592 	}
8593 	strlcpy(kstat_kv_istr(kvs), text, sizeof(kstat_kv_istr(kvs)));
8594 	kvs++;
8595 
8596 	if (mcx_query_sq(sc, &q->q_tx, &u.sq) != 0) {
8597 		error = EIO;
8598 		goto out;
8599 	}
8600 
8601 	kstat_kv_u64(kvs++) = q->q_tx.tx_prod;
8602 	kstat_kv_u64(kvs++) = q->q_tx.tx_cons;
8603 	kstat_kv_u64(kvs++) = bemtoh32(&u.sq.sq_wq.wq_sw_counter);
8604 	kstat_kv_u64(kvs++) = bemtoh32(&u.sq.sq_wq.wq_hw_counter);
8605 	switch ((bemtoh32(&u.sq.sq_flags) & MCX_SQ_CTX_STATE_MASK) >>
8606 	    MCX_SQ_CTX_STATE_SHIFT) {
8607 	case MCX_SQ_CTX_STATE_RST:
8608 		text = "RST";
8609 		break;
8610 	case MCX_SQ_CTX_STATE_RDY:
8611 		text = "RDY";
8612 		break;
8613 	case MCX_SQ_CTX_STATE_ERR:
8614 		text = "ERR";
8615 		break;
8616 	default:
8617 		text = "unknown";
8618 		break;
8619 	}
8620 	strlcpy(kstat_kv_istr(kvs), text, sizeof(kstat_kv_istr(kvs)));
8621 	kvs++;
8622 
8623 	if (mcx_query_cq(sc, &q->q_cq, &u.cq) != 0) {
8624 		error = EIO;
8625 		goto out;
8626 	}
8627 
8628 	kstat_kv_u64(kvs++) = q->q_cq.cq_cons;
8629 	kstat_kv_u64(kvs++) = bemtoh32(&u.cq.cq_producer_counter);
8630 	kstat_kv_u64(kvs++) = bemtoh32(&u.cq.cq_consumer_counter);
8631 	kstat_kv_u64(kvs++) = bemtoh32(&u.cq.cq_last_notified);
8632 	kstat_kv_u64(kvs++) = bemtoh32(&u.cq.cq_last_solicit);
8633 
8634 	switch ((bemtoh32(&u.cq.cq_status) & MCX_CQ_CTX_STATUS_MASK) >>
8635 	    MCX_CQ_CTX_STATUS_SHIFT) {
8636 	case MCX_CQ_CTX_STATUS_OK:
8637 		text = "OK";
8638 		break;
8639 	case MCX_CQ_CTX_STATUS_OVERFLOW:
8640 		text = "overflow";
8641 		break;
8642 	case MCX_CQ_CTX_STATUS_WRITE_FAIL:
8643 		text = "write fail";
8644 		break;
8645 	default:
8646 		text = "unknown";
8647 		break;
8648 	}
8649 	strlcpy(kstat_kv_istr(kvs), text, sizeof(kstat_kv_istr(kvs)));
8650 	kvs++;
8651 
8652 	switch ((bemtoh32(&u.cq.cq_status) & MCX_CQ_CTX_STATE_MASK) >>
8653 	    MCX_CQ_CTX_STATE_SHIFT) {
8654 	case MCX_CQ_CTX_STATE_SOLICITED:
8655 		text = "solicited";
8656 		break;
8657 	case MCX_CQ_CTX_STATE_ARMED:
8658 		text = "armed";
8659 		break;
8660 	case MCX_CQ_CTX_STATE_FIRED:
8661 		text = "fired";
8662 		break;
8663 	default:
8664 		text = "unknown";
8665 		break;
8666 	}
8667 	strlcpy(kstat_kv_istr(kvs), text, sizeof(kstat_kv_istr(kvs)));
8668 	kvs++;
8669 
8670 	if (mcx_query_eq(sc, &q->q_eq, &u.eq) != 0) {
8671 		error = EIO;
8672 		goto out;
8673 	}
8674 
8675 	kstat_kv_u64(kvs++) = q->q_eq.eq_cons;
8676 	kstat_kv_u64(kvs++) = bemtoh32(&u.eq.eq_producer_counter);
8677 	kstat_kv_u64(kvs++) = bemtoh32(&u.eq.eq_consumer_counter);
8678 
8679 	switch ((bemtoh32(&u.eq.eq_status) & MCX_EQ_CTX_STATUS_MASK) >>
8680 	    MCX_EQ_CTX_STATUS_SHIFT) {
8681 	case MCX_EQ_CTX_STATUS_EQ_WRITE_FAILURE:
8682 		text = "write fail";
8683 		break;
8684 	case MCX_EQ_CTX_STATUS_OK:
8685 		text = "OK";
8686 		break;
8687 	default:
8688 		text = "unknown";
8689 		break;
8690 	}
8691 	strlcpy(kstat_kv_istr(kvs), text, sizeof(kstat_kv_istr(kvs)));
8692 	kvs++;
8693 
8694 	switch ((bemtoh32(&u.eq.eq_status) & MCX_EQ_CTX_STATE_MASK) >>
8695 	    MCX_EQ_CTX_STATE_SHIFT) {
8696 	case MCX_EQ_CTX_STATE_ARMED:
8697 		text = "armed";
8698 		break;
8699 	case MCX_EQ_CTX_STATE_FIRED:
8700 		text = "fired";
8701 		break;
8702 	default:
8703 		text = "unknown";
8704 		break;
8705 	}
8706 	strlcpy(kstat_kv_istr(kvs), text, sizeof(kstat_kv_istr(kvs)));
8707 	kvs++;
8708 
8709 	nanouptime(&ks->ks_updated);
8710 out:
8711 	KERNEL_UNLOCK();
8712 	return (error);
8713 }
8714 
8715 #endif /* NKSTAT > 0 */
8716 
8717 static unsigned int
8718 mcx_timecounter_read(struct timecounter *tc)
8719 {
8720 	struct mcx_softc *sc = tc->tc_priv;
8721 
8722 	return (mcx_rd(sc, MCX_INTERNAL_TIMER_L));
8723 }
8724 
8725 static void
8726 mcx_timecounter_attach(struct mcx_softc *sc)
8727 {
8728 	struct timecounter *tc = &sc->sc_timecounter;
8729 
8730 	tc->tc_get_timecount = mcx_timecounter_read;
8731 	tc->tc_counter_mask = ~0U;
8732 	tc->tc_frequency = sc->sc_khz * 1000;
8733 	tc->tc_name = sc->sc_dev.dv_xname;
8734 	tc->tc_quality = -100;
8735 	tc->tc_priv = sc;
8736 
8737 	tc_init(tc);
8738 }
8739