xref: /netbsd-src/sys/dev/pci/if_mcx.c (revision afab4e300d3a9fb07dd8c80daf53d0feb3345706)
1 /*	$NetBSD: if_mcx.c,v 1.25 2022/08/28 07:54:03 skrll Exp $ */
2 /*	$OpenBSD: if_mcx.c,v 1.101 2021/06/02 19:16:11 patrick Exp $ */
3 
4 /*
5  * Copyright (c) 2017 David Gwynne <dlg@openbsd.org>
6  * Copyright (c) 2019 Jonathan Matthew <jmatthew@openbsd.org>
7  *
8  * Permission to use, copy, modify, and distribute this software for any
9  * purpose with or without fee is hereby granted, provided that the above
10  * copyright notice and this permission notice appear in all copies.
11  *
12  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
13  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
14  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
15  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
16  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
17  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
18  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
19  */
20 
21 #ifdef _KERNEL_OPT
22 #include "opt_net_mpsafe.h"
23 #endif
24 
25 #include <sys/cdefs.h>
26 __KERNEL_RCSID(0, "$NetBSD: if_mcx.c,v 1.25 2022/08/28 07:54:03 skrll Exp $");
27 
28 #include <sys/param.h>
29 #include <sys/systm.h>
30 #include <sys/sockio.h>
31 #include <sys/mbuf.h>
32 #include <sys/kernel.h>
33 #include <sys/socket.h>
34 #include <sys/device.h>
35 #include <sys/pool.h>
36 #include <sys/queue.h>
37 #include <sys/callout.h>
38 #include <sys/workqueue.h>
39 #include <sys/atomic.h>
40 #include <sys/timetc.h>
41 #include <sys/kmem.h>
42 #include <sys/bus.h>
43 #include <sys/interrupt.h>
44 #include <sys/pcq.h>
45 #include <sys/cpu.h>
46 #include <sys/bitops.h>
47 
48 #include <machine/intr.h>
49 
50 #include <net/if.h>
51 #include <net/if_dl.h>
52 #include <net/if_ether.h>
53 #include <net/if_media.h>
54 #include <net/if_vlanvar.h>
55 #include <net/toeplitz.h>
56 
57 #include <net/bpf.h>
58 
59 #include <netinet/in.h>
60 
61 #include <dev/pci/pcireg.h>
62 #include <dev/pci/pcivar.h>
63 #include <dev/pci/pcidevs.h>
64 
65 /* TODO: Port kstat key/value stuff to evcnt/sysmon */
66 #define	NKSTAT		0
67 
68 /* XXX This driver is not yet MP-safe; don't claim to be! */
69 /* #ifdef NET_MPSAFE */
70 /* #define	MCX_MPSAFE	1 */
71 /* #define	CALLOUT_FLAGS	CALLOUT_MPSAFE */
72 /* #else */
73 #define	CALLOUT_FLAGS	0
74 /* #endif */
75 
76 #define	MCX_TXQ_NUM		2048
77 
78 #define BUS_DMASYNC_PRERW	(BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE)
79 #define BUS_DMASYNC_POSTRW	(BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE)
80 
81 #define MCX_HCA_BAR	PCI_MAPREG_START /* BAR 0 */
82 
83 #define MCX_FW_VER			0x0000
84 #define  MCX_FW_VER_MAJOR(_v)			((_v) & 0xffff)
85 #define  MCX_FW_VER_MINOR(_v)			((_v) >> 16)
86 #define MCX_CMDIF_FW_SUBVER		0x0004
87 #define  MCX_FW_VER_SUBMINOR(_v)		((_v) & 0xffff)
88 #define  MCX_CMDIF(_v)				((_v) >> 16)
89 
90 #define MCX_ISSI			1 /* as per the PRM */
91 #define MCX_CMD_IF_SUPPORTED		5
92 
93 #define MCX_HARDMTU			9500
94 
95 #define MCX_PAGE_SHIFT			12
96 #define MCX_PAGE_SIZE			(1 << MCX_PAGE_SHIFT)
97 
98 /* queue sizes */
99 #define MCX_LOG_EQ_SIZE			7
100 #define MCX_LOG_CQ_SIZE			12
101 #define MCX_LOG_RQ_SIZE			10
102 #define MCX_LOG_SQ_SIZE			11
103 
104 #define MCX_MAX_QUEUES			16
105 
106 /* completion event moderation - about 10khz, or 90% of the cq */
107 #define MCX_CQ_MOD_PERIOD		50
108 #define MCX_CQ_MOD_COUNTER		\
109 	(((1 << (MCX_LOG_CQ_SIZE - 1)) * 9) / 10)
110 
111 #define MCX_LOG_SQ_ENTRY_SIZE		6
112 #define MCX_SQ_ENTRY_MAX_SLOTS		4
113 #define MCX_SQ_SEGS_PER_SLOT		\
114 	(sizeof(struct mcx_sq_entry) / sizeof(struct mcx_sq_entry_seg))
115 #define MCX_SQ_MAX_SEGMENTS		\
116 	1 + ((MCX_SQ_ENTRY_MAX_SLOTS-1) * MCX_SQ_SEGS_PER_SLOT)
117 
118 #define MCX_LOG_FLOW_TABLE_SIZE		5
119 #define MCX_NUM_STATIC_FLOWS		4 /* promisc, allmulti, ucast, bcast */
120 #define MCX_NUM_MCAST_FLOWS 		\
121 	((1 << MCX_LOG_FLOW_TABLE_SIZE) - MCX_NUM_STATIC_FLOWS)
122 
123 #define MCX_SQ_INLINE_SIZE		18
124 CTASSERT(ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN == MCX_SQ_INLINE_SIZE);
125 
126 /* doorbell offsets */
127 #define MCX_DOORBELL_AREA_SIZE		MCX_PAGE_SIZE
128 
129 #define MCX_CQ_DOORBELL_BASE		0
130 #define MCX_CQ_DOORBELL_STRIDE		64
131 
132 #define MCX_WQ_DOORBELL_BASE		MCX_PAGE_SIZE/2
133 #define MCX_WQ_DOORBELL_STRIDE		64
134 /* make sure the doorbells fit */
135 CTASSERT(MCX_MAX_QUEUES * MCX_CQ_DOORBELL_STRIDE < MCX_WQ_DOORBELL_BASE);
136 CTASSERT(MCX_MAX_QUEUES * MCX_WQ_DOORBELL_STRIDE <
137     MCX_DOORBELL_AREA_SIZE - MCX_WQ_DOORBELL_BASE);
138 
139 #define MCX_WQ_DOORBELL_MASK		0xffff
140 
141 /* uar registers */
142 #define MCX_UAR_CQ_DOORBELL		0x20
143 #define MCX_UAR_EQ_DOORBELL_ARM		0x40
144 #define MCX_UAR_EQ_DOORBELL		0x48
145 #define MCX_UAR_BF			0x800
146 
147 #define MCX_CMDQ_ADDR_HI		0x0010
148 #define MCX_CMDQ_ADDR_LO		0x0014
149 #define MCX_CMDQ_ADDR_NMASK		0xfff
150 #define MCX_CMDQ_LOG_SIZE(_v)		((_v) >> 4 & 0xf)
151 #define MCX_CMDQ_LOG_STRIDE(_v)		((_v) >> 0 & 0xf)
152 #define MCX_CMDQ_INTERFACE_MASK		(0x3 << 8)
153 #define MCX_CMDQ_INTERFACE_FULL_DRIVER	(0x0 << 8)
154 #define MCX_CMDQ_INTERFACE_DISABLED	(0x1 << 8)
155 
156 #define MCX_CMDQ_DOORBELL		0x0018
157 
158 #define MCX_STATE			0x01fc
159 #define MCX_STATE_MASK				(1U << 31)
160 #define MCX_STATE_INITIALIZING			(1 << 31)
161 #define MCX_STATE_READY				(0 << 31)
162 #define MCX_STATE_INTERFACE_MASK		(0x3 << 24)
163 #define MCX_STATE_INTERFACE_FULL_DRIVER		(0x0 << 24)
164 #define MCX_STATE_INTERFACE_DISABLED		(0x1 << 24)
165 
166 #define MCX_INTERNAL_TIMER		0x1000
167 #define MCX_INTERNAL_TIMER_H		0x1000
168 #define MCX_INTERNAL_TIMER_L		0x1004
169 
170 #define MCX_CLEAR_INT			0x100c
171 
172 #define MCX_REG_OP_WRITE		0
173 #define MCX_REG_OP_READ			1
174 
175 #define MCX_REG_PMLP			0x5002
176 #define MCX_REG_PMTU			0x5003
177 #define MCX_REG_PTYS			0x5004
178 #define MCX_REG_PAOS			0x5006
179 #define MCX_REG_PFCC			0x5007
180 #define MCX_REG_PPCNT			0x5008
181 #define MCX_REG_MTCAP			0x9009 /* mgmt temp capabilities */
182 #define MCX_REG_MTMP			0x900a /* mgmt temp */
183 #define MCX_REG_MCIA			0x9014
184 #define MCX_REG_MCAM			0x907f
185 
186 #define MCX_ETHER_CAP_SGMII		0
187 #define MCX_ETHER_CAP_1000_KX		1
188 #define MCX_ETHER_CAP_10G_CX4		2
189 #define MCX_ETHER_CAP_10G_KX4		3
190 #define MCX_ETHER_CAP_10G_KR		4
191 #define MCX_ETHER_CAP_20G_KR2		5
192 #define MCX_ETHER_CAP_40G_CR4		6
193 #define MCX_ETHER_CAP_40G_KR4		7
194 #define MCX_ETHER_CAP_56G_R4		8
195 #define MCX_ETHER_CAP_10G_CR		12
196 #define MCX_ETHER_CAP_10G_SR		13
197 #define MCX_ETHER_CAP_10G_LR		14
198 #define MCX_ETHER_CAP_40G_SR4		15
199 #define MCX_ETHER_CAP_40G_LR4		16
200 #define MCX_ETHER_CAP_50G_SR2		18
201 #define MCX_ETHER_CAP_100G_CR4		20
202 #define MCX_ETHER_CAP_100G_SR4		21
203 #define MCX_ETHER_CAP_100G_KR4		22
204 #define MCX_ETHER_CAP_100G_LR4		23
205 #define MCX_ETHER_CAP_100_TX		24
206 #define MCX_ETHER_CAP_1000_T		25
207 #define MCX_ETHER_CAP_10G_T		26
208 #define MCX_ETHER_CAP_25G_CR		27
209 #define MCX_ETHER_CAP_25G_KR		28
210 #define MCX_ETHER_CAP_25G_SR		29
211 #define MCX_ETHER_CAP_50G_CR2		30
212 #define MCX_ETHER_CAP_50G_KR2		31
213 
214 #define MCX_MAX_CQE			32
215 
216 #define MCX_CMD_QUERY_HCA_CAP		0x100
217 #define MCX_CMD_QUERY_ADAPTER		0x101
218 #define MCX_CMD_INIT_HCA		0x102
219 #define MCX_CMD_TEARDOWN_HCA		0x103
220 #define MCX_CMD_ENABLE_HCA		0x104
221 #define MCX_CMD_DISABLE_HCA		0x105
222 #define MCX_CMD_QUERY_PAGES		0x107
223 #define MCX_CMD_MANAGE_PAGES		0x108
224 #define MCX_CMD_SET_HCA_CAP		0x109
225 #define MCX_CMD_QUERY_ISSI		0x10a
226 #define MCX_CMD_SET_ISSI		0x10b
227 #define MCX_CMD_SET_DRIVER_VERSION	0x10d
228 #define MCX_CMD_QUERY_SPECIAL_CONTEXTS	0x203
229 #define MCX_CMD_CREATE_EQ		0x301
230 #define MCX_CMD_DESTROY_EQ		0x302
231 #define MCX_CMD_QUERY_EQ		0x303
232 #define MCX_CMD_CREATE_CQ		0x400
233 #define MCX_CMD_DESTROY_CQ		0x401
234 #define MCX_CMD_QUERY_CQ		0x402
235 #define MCX_CMD_QUERY_NIC_VPORT_CONTEXT 0x754
236 #define MCX_CMD_MODIFY_NIC_VPORT_CONTEXT \
237 					0x755
238 #define MCX_CMD_QUERY_VPORT_COUNTERS 	0x770
239 #define MCX_CMD_ALLOC_PD		0x800
240 #define MCX_CMD_ALLOC_UAR		0x802
241 #define MCX_CMD_ACCESS_REG		0x805
242 #define MCX_CMD_ALLOC_TRANSPORT_DOMAIN	0x816
243 #define MCX_CMD_CREATE_TIR		0x900
244 #define MCX_CMD_DESTROY_TIR		0x902
245 #define MCX_CMD_CREATE_SQ		0x904
246 #define MCX_CMD_MODIFY_SQ		0x905
247 #define MCX_CMD_DESTROY_SQ		0x906
248 #define MCX_CMD_QUERY_SQ		0x907
249 #define MCX_CMD_CREATE_RQ		0x908
250 #define MCX_CMD_MODIFY_RQ		0x909
251 #define MCX_CMD_DESTROY_RQ		0x90a
252 #define MCX_CMD_QUERY_RQ		0x90b
253 #define MCX_CMD_CREATE_TIS		0x912
254 #define MCX_CMD_DESTROY_TIS		0x914
255 #define MCX_CMD_CREATE_RQT		0x916
256 #define MCX_CMD_DESTROY_RQT		0x918
257 #define MCX_CMD_SET_FLOW_TABLE_ROOT	0x92f
258 #define MCX_CMD_CREATE_FLOW_TABLE	0x930
259 #define MCX_CMD_DESTROY_FLOW_TABLE	0x931
260 #define MCX_CMD_QUERY_FLOW_TABLE	0x932
261 #define MCX_CMD_CREATE_FLOW_GROUP	0x933
262 #define MCX_CMD_DESTROY_FLOW_GROUP	0x934
263 #define MCX_CMD_QUERY_FLOW_GROUP	0x935
264 #define MCX_CMD_SET_FLOW_TABLE_ENTRY	0x936
265 #define MCX_CMD_QUERY_FLOW_TABLE_ENTRY	0x937
266 #define MCX_CMD_DELETE_FLOW_TABLE_ENTRY	0x938
267 #define MCX_CMD_ALLOC_FLOW_COUNTER	0x939
268 #define MCX_CMD_QUERY_FLOW_COUNTER	0x93b
269 
270 #define MCX_QUEUE_STATE_RST		0
271 #define MCX_QUEUE_STATE_RDY		1
272 #define MCX_QUEUE_STATE_ERR		3
273 
274 #define MCX_FLOW_TABLE_TYPE_RX		0
275 #define MCX_FLOW_TABLE_TYPE_TX		1
276 
277 #define MCX_CMDQ_INLINE_DATASIZE	16
278 
279 struct mcx_cmdq_entry {
280 	uint8_t			cq_type;
281 #define MCX_CMDQ_TYPE_PCIE		0x7
282 	uint8_t			cq_reserved0[3];
283 
284 	uint32_t		cq_input_length;
285 	uint64_t		cq_input_ptr;
286 	uint8_t			cq_input_data[MCX_CMDQ_INLINE_DATASIZE];
287 
288 	uint8_t			cq_output_data[MCX_CMDQ_INLINE_DATASIZE];
289 	uint64_t		cq_output_ptr;
290 	uint32_t		cq_output_length;
291 
292 	uint8_t			cq_token;
293 	uint8_t			cq_signature;
294 	uint8_t			cq_reserved1[1];
295 	uint8_t			cq_status;
296 #define MCX_CQ_STATUS_SHIFT		1
297 #define MCX_CQ_STATUS_MASK		(0x7f << MCX_CQ_STATUS_SHIFT)
298 #define MCX_CQ_STATUS_OK		(0x00 << MCX_CQ_STATUS_SHIFT)
299 #define MCX_CQ_STATUS_INT_ERR		(0x01 << MCX_CQ_STATUS_SHIFT)
300 #define MCX_CQ_STATUS_BAD_OPCODE	(0x02 << MCX_CQ_STATUS_SHIFT)
301 #define MCX_CQ_STATUS_BAD_PARAM		(0x03 << MCX_CQ_STATUS_SHIFT)
302 #define MCX_CQ_STATUS_BAD_SYS_STATE	(0x04 << MCX_CQ_STATUS_SHIFT)
303 #define MCX_CQ_STATUS_BAD_RESOURCE	(0x05 << MCX_CQ_STATUS_SHIFT)
304 #define MCX_CQ_STATUS_RESOURCE_BUSY	(0x06 << MCX_CQ_STATUS_SHIFT)
305 #define MCX_CQ_STATUS_EXCEED_LIM	(0x08 << MCX_CQ_STATUS_SHIFT)
306 #define MCX_CQ_STATUS_BAD_RES_STATE	(0x09 << MCX_CQ_STATUS_SHIFT)
307 #define MCX_CQ_STATUS_BAD_INDEX		(0x0a << MCX_CQ_STATUS_SHIFT)
308 #define MCX_CQ_STATUS_NO_RESOURCES	(0x0f << MCX_CQ_STATUS_SHIFT)
309 #define MCX_CQ_STATUS_BAD_INPUT_LEN	(0x50 << MCX_CQ_STATUS_SHIFT)
310 #define MCX_CQ_STATUS_BAD_OUTPUT_LEN	(0x51 << MCX_CQ_STATUS_SHIFT)
311 #define MCX_CQ_STATUS_BAD_RESOURCE_STATE \
312 					(0x10 << MCX_CQ_STATUS_SHIFT)
313 #define MCX_CQ_STATUS_BAD_SIZE		(0x40 << MCX_CQ_STATUS_SHIFT)
314 #define MCX_CQ_STATUS_OWN_MASK		0x1
315 #define MCX_CQ_STATUS_OWN_SW		0x0
316 #define MCX_CQ_STATUS_OWN_HW		0x1
317 } __packed __aligned(8);
318 
319 #define MCX_CMDQ_MAILBOX_DATASIZE	512
320 
321 struct mcx_cmdq_mailbox {
322 	uint8_t			mb_data[MCX_CMDQ_MAILBOX_DATASIZE];
323 	uint8_t			mb_reserved0[48];
324 	uint64_t		mb_next_ptr;
325 	uint32_t		mb_block_number;
326 	uint8_t			mb_reserved1[1];
327 	uint8_t			mb_token;
328 	uint8_t			mb_ctrl_signature;
329 	uint8_t			mb_signature;
330 } __packed __aligned(8);
331 
332 #define MCX_CMDQ_MAILBOX_ALIGN	(1 << 10)
333 #define MCX_CMDQ_MAILBOX_SIZE	roundup(sizeof(struct mcx_cmdq_mailbox), \
334 				    MCX_CMDQ_MAILBOX_ALIGN)
335 /*
336  * command mailbox structures
337  */
338 
339 struct mcx_cmd_enable_hca_in {
340 	uint16_t		cmd_opcode;
341 	uint8_t			cmd_reserved0[4];
342 	uint16_t		cmd_op_mod;
343 	uint8_t			cmd_reserved1[2];
344 	uint16_t		cmd_function_id;
345 	uint8_t			cmd_reserved2[4];
346 } __packed __aligned(4);
347 
348 struct mcx_cmd_enable_hca_out {
349 	uint8_t			cmd_status;
350 	uint8_t			cmd_reserved0[3];
351 	uint32_t		cmd_syndrome;
352 	uint8_t			cmd_reserved1[4];
353 } __packed __aligned(4);
354 
355 struct mcx_cmd_init_hca_in {
356 	uint16_t		cmd_opcode;
357 	uint8_t			cmd_reserved0[4];
358 	uint16_t		cmd_op_mod;
359 	uint8_t			cmd_reserved1[8];
360 } __packed __aligned(4);
361 
362 struct mcx_cmd_init_hca_out {
363 	uint8_t			cmd_status;
364 	uint8_t			cmd_reserved0[3];
365 	uint32_t		cmd_syndrome;
366 	uint8_t			cmd_reserved1[8];
367 } __packed __aligned(4);
368 
369 struct mcx_cmd_teardown_hca_in {
370 	uint16_t		cmd_opcode;
371 	uint8_t			cmd_reserved0[4];
372 	uint16_t		cmd_op_mod;
373 	uint8_t			cmd_reserved1[2];
374 #define MCX_CMD_TEARDOWN_HCA_GRACEFUL	0x0
375 #define MCX_CMD_TEARDOWN_HCA_PANIC	0x1
376 	uint16_t		cmd_profile;
377 	uint8_t			cmd_reserved2[4];
378 } __packed __aligned(4);
379 
380 struct mcx_cmd_teardown_hca_out {
381 	uint8_t			cmd_status;
382 	uint8_t			cmd_reserved0[3];
383 	uint32_t		cmd_syndrome;
384 	uint8_t			cmd_reserved1[8];
385 } __packed __aligned(4);
386 
387 struct mcx_cmd_access_reg_in {
388 	uint16_t		cmd_opcode;
389 	uint8_t			cmd_reserved0[4];
390 	uint16_t		cmd_op_mod;
391 	uint8_t			cmd_reserved1[2];
392 	uint16_t		cmd_register_id;
393 	uint32_t		cmd_argument;
394 } __packed __aligned(4);
395 
396 struct mcx_cmd_access_reg_out {
397 	uint8_t			cmd_status;
398 	uint8_t			cmd_reserved0[3];
399 	uint32_t		cmd_syndrome;
400 	uint8_t			cmd_reserved1[8];
401 } __packed __aligned(4);
402 
403 struct mcx_reg_pmtu {
404 	uint8_t			rp_reserved1;
405 	uint8_t			rp_local_port;
406 	uint8_t			rp_reserved2[2];
407 	uint16_t		rp_max_mtu;
408 	uint8_t			rp_reserved3[2];
409 	uint16_t		rp_admin_mtu;
410 	uint8_t			rp_reserved4[2];
411 	uint16_t		rp_oper_mtu;
412 	uint8_t			rp_reserved5[2];
413 } __packed __aligned(4);
414 
415 struct mcx_reg_ptys {
416 	uint8_t			rp_reserved1;
417 	uint8_t			rp_local_port;
418 	uint8_t			rp_reserved2;
419 	uint8_t			rp_proto_mask;
420 #define MCX_REG_PTYS_PROTO_MASK_ETH		(1 << 2)
421 	uint8_t			rp_reserved3[8];
422 	uint32_t		rp_eth_proto_cap;
423 	uint8_t			rp_reserved4[8];
424 	uint32_t		rp_eth_proto_admin;
425 	uint8_t			rp_reserved5[8];
426 	uint32_t		rp_eth_proto_oper;
427 	uint8_t			rp_reserved6[24];
428 } __packed __aligned(4);
429 
430 struct mcx_reg_paos {
431 	uint8_t			rp_reserved1;
432 	uint8_t			rp_local_port;
433 	uint8_t			rp_admin_status;
434 #define MCX_REG_PAOS_ADMIN_STATUS_UP		1
435 #define MCX_REG_PAOS_ADMIN_STATUS_DOWN		2
436 #define MCX_REG_PAOS_ADMIN_STATUS_UP_ONCE	3
437 #define MCX_REG_PAOS_ADMIN_STATUS_DISABLED	4
438 	uint8_t			rp_oper_status;
439 #define MCX_REG_PAOS_OPER_STATUS_UP		1
440 #define MCX_REG_PAOS_OPER_STATUS_DOWN		2
441 #define MCX_REG_PAOS_OPER_STATUS_FAILED		4
442 	uint8_t			rp_admin_state_update;
443 #define MCX_REG_PAOS_ADMIN_STATE_UPDATE_EN	(1 << 7)
444 	uint8_t			rp_reserved2[11];
445 } __packed __aligned(4);
446 
447 struct mcx_reg_pfcc {
448 	uint8_t			rp_reserved1;
449 	uint8_t			rp_local_port;
450 	uint8_t			rp_reserved2[3];
451 	uint8_t			rp_prio_mask_tx;
452 	uint8_t			rp_reserved3;
453 	uint8_t			rp_prio_mask_rx;
454 	uint8_t			rp_pptx_aptx;
455 	uint8_t			rp_pfctx;
456 	uint8_t			rp_fctx_dis;
457 	uint8_t			rp_reserved4;
458 	uint8_t			rp_pprx_aprx;
459 	uint8_t			rp_pfcrx;
460 	uint8_t			rp_reserved5[2];
461 	uint16_t		rp_dev_stall_min;
462 	uint16_t		rp_dev_stall_crit;
463 	uint8_t			rp_reserved6[12];
464 } __packed __aligned(4);
465 
466 #define MCX_PMLP_MODULE_NUM_MASK	0xff
467 struct mcx_reg_pmlp {
468 	uint8_t			rp_rxtx;
469 	uint8_t			rp_local_port;
470 	uint8_t			rp_reserved0;
471 	uint8_t			rp_width;
472 	uint32_t		rp_lane0_mapping;
473 	uint32_t		rp_lane1_mapping;
474 	uint32_t		rp_lane2_mapping;
475 	uint32_t		rp_lane3_mapping;
476 	uint8_t			rp_reserved1[44];
477 } __packed __aligned(4);
478 
479 struct mcx_reg_ppcnt {
480 	uint8_t			ppcnt_swid;
481 	uint8_t			ppcnt_local_port;
482 	uint8_t			ppcnt_pnat;
483 	uint8_t			ppcnt_grp;
484 #define MCX_REG_PPCNT_GRP_IEEE8023		0x00
485 #define MCX_REG_PPCNT_GRP_RFC2863		0x01
486 #define MCX_REG_PPCNT_GRP_RFC2819		0x02
487 #define MCX_REG_PPCNT_GRP_RFC3635		0x03
488 #define MCX_REG_PPCNT_GRP_PER_PRIO		0x10
489 #define MCX_REG_PPCNT_GRP_PER_TC		0x11
490 #define MCX_REG_PPCNT_GRP_PER_RX_BUFFER		0x11
491 
492 	uint8_t			ppcnt_clr;
493 	uint8_t			ppcnt_reserved1[2];
494 	uint8_t			ppcnt_prio_tc;
495 #define MCX_REG_PPCNT_CLR			(1 << 7)
496 
497 	uint8_t			ppcnt_counter_set[248];
498 } __packed __aligned(8);
499 CTASSERT(sizeof(struct mcx_reg_ppcnt) == 256);
500 CTASSERT((offsetof(struct mcx_reg_ppcnt, ppcnt_counter_set) %
501     sizeof(uint64_t)) == 0);
502 
503 enum mcx_ppcnt_ieee8023 {
504 	frames_transmitted_ok,
505 	frames_received_ok,
506 	frame_check_sequence_errors,
507 	alignment_errors,
508 	octets_transmitted_ok,
509 	octets_received_ok,
510 	multicast_frames_xmitted_ok,
511 	broadcast_frames_xmitted_ok,
512 	multicast_frames_received_ok,
513 	broadcast_frames_received_ok,
514 	in_range_length_errors,
515 	out_of_range_length_field,
516 	frame_too_long_errors,
517 	symbol_error_during_carrier,
518 	mac_control_frames_transmitted,
519 	mac_control_frames_received,
520 	unsupported_opcodes_received,
521 	pause_mac_ctrl_frames_received,
522 	pause_mac_ctrl_frames_transmitted,
523 
524 	mcx_ppcnt_ieee8023_count
525 };
526 CTASSERT(mcx_ppcnt_ieee8023_count * sizeof(uint64_t) == 0x98);
527 
528 enum mcx_ppcnt_rfc2863 {
529 	in_octets,
530 	in_ucast_pkts,
531 	in_discards,
532 	in_errors,
533 	in_unknown_protos,
534 	out_octets,
535 	out_ucast_pkts,
536 	out_discards,
537 	out_errors,
538 	in_multicast_pkts,
539 	in_broadcast_pkts,
540 	out_multicast_pkts,
541 	out_broadcast_pkts,
542 
543 	mcx_ppcnt_rfc2863_count
544 };
545 CTASSERT(mcx_ppcnt_rfc2863_count * sizeof(uint64_t) == 0x68);
546 
547 enum mcx_ppcnt_rfc2819 {
548 	drop_events,
549 	octets,
550 	pkts,
551 	broadcast_pkts,
552 	multicast_pkts,
553 	crc_align_errors,
554 	undersize_pkts,
555 	oversize_pkts,
556 	fragments,
557 	jabbers,
558 	collisions,
559 	pkts64octets,
560 	pkts65to127octets,
561 	pkts128to255octets,
562 	pkts256to511octets,
563 	pkts512to1023octets,
564 	pkts1024to1518octets,
565 	pkts1519to2047octets,
566 	pkts2048to4095octets,
567 	pkts4096to8191octets,
568 	pkts8192to10239octets,
569 
570 	mcx_ppcnt_rfc2819_count
571 };
572 CTASSERT((mcx_ppcnt_rfc2819_count * sizeof(uint64_t)) == 0xa8);
573 
574 enum mcx_ppcnt_rfc3635 {
575 	dot3stats_alignment_errors,
576 	dot3stats_fcs_errors,
577 	dot3stats_single_collision_frames,
578 	dot3stats_multiple_collision_frames,
579 	dot3stats_sqe_test_errors,
580 	dot3stats_deferred_transmissions,
581 	dot3stats_late_collisions,
582 	dot3stats_excessive_collisions,
583 	dot3stats_internal_mac_transmit_errors,
584 	dot3stats_carrier_sense_errors,
585 	dot3stats_frame_too_longs,
586 	dot3stats_internal_mac_receive_errors,
587 	dot3stats_symbol_errors,
588 	dot3control_in_unknown_opcodes,
589 	dot3in_pause_frames,
590 	dot3out_pause_frames,
591 
592 	mcx_ppcnt_rfc3635_count
593 };
594 CTASSERT((mcx_ppcnt_rfc3635_count * sizeof(uint64_t)) == 0x80);
595 
596 struct mcx_reg_mcam {
597 	uint8_t			_reserved1[1];
598 	uint8_t			mcam_feature_group;
599 	uint8_t			_reserved2[1];
600 	uint8_t			mcam_access_reg_group;
601 	uint8_t			_reserved3[4];
602 	uint8_t			mcam_access_reg_cap_mask[16];
603 	uint8_t			_reserved4[16];
604 	uint8_t			mcam_feature_cap_mask[16];
605 	uint8_t			_reserved5[16];
606 } __packed __aligned(4);
607 
608 #define MCX_BITFIELD_BIT(bf, b)	(bf[(sizeof bf - 1) - (b / 8)] & (b % 8))
609 
610 #define MCX_MCAM_FEATURE_CAP_SENSOR_MAP	6
611 
612 struct mcx_reg_mtcap {
613 	uint8_t			_reserved1[3];
614 	uint8_t			mtcap_sensor_count;
615 	uint8_t			_reserved2[4];
616 
617 	uint64_t		mtcap_sensor_map;
618 };
619 
620 struct mcx_reg_mtmp {
621 	uint8_t			_reserved1[2];
622 	uint16_t		mtmp_sensor_index;
623 
624 	uint8_t			_reserved2[2];
625 	uint16_t		mtmp_temperature;
626 
627 	uint16_t		mtmp_mte_mtr;
628 #define MCX_REG_MTMP_MTE		(1 << 15)
629 #define MCX_REG_MTMP_MTR		(1 << 14)
630 	uint16_t		mtmp_max_temperature;
631 
632 	uint16_t		mtmp_tee;
633 #define MCX_REG_MTMP_TEE_NOPE		(0 << 14)
634 #define MCX_REG_MTMP_TEE_GENERATE	(1 << 14)
635 #define MCX_REG_MTMP_TEE_GENERATE_ONE	(2 << 14)
636 	uint16_t		mtmp_temperature_threshold_hi;
637 
638 	uint8_t			_reserved3[2];
639 	uint16_t		mtmp_temperature_threshold_lo;
640 
641 	uint8_t			_reserved4[4];
642 
643 	uint8_t			mtmp_sensor_name[8];
644 };
645 CTASSERT(sizeof(struct mcx_reg_mtmp) == 0x20);
646 CTASSERT(offsetof(struct mcx_reg_mtmp, mtmp_sensor_name) == 0x18);
647 
648 #define MCX_MCIA_EEPROM_BYTES	32
649 struct mcx_reg_mcia {
650 	uint8_t			rm_l;
651 	uint8_t			rm_module;
652 	uint8_t			rm_reserved0;
653 	uint8_t			rm_status;
654 	uint8_t			rm_i2c_addr;
655 	uint8_t			rm_page_num;
656 	uint16_t		rm_dev_addr;
657 	uint16_t		rm_reserved1;
658 	uint16_t		rm_size;
659 	uint32_t		rm_reserved2;
660 	uint8_t			rm_data[48];
661 } __packed __aligned(4);
662 
663 struct mcx_cmd_query_issi_in {
664 	uint16_t		cmd_opcode;
665 	uint8_t			cmd_reserved0[4];
666 	uint16_t		cmd_op_mod;
667 	uint8_t			cmd_reserved1[8];
668 } __packed __aligned(4);
669 
670 struct mcx_cmd_query_issi_il_out {
671 	uint8_t			cmd_status;
672 	uint8_t			cmd_reserved0[3];
673 	uint32_t		cmd_syndrome;
674 	uint8_t			cmd_reserved1[2];
675 	uint16_t		cmd_current_issi;
676 	uint8_t			cmd_reserved2[4];
677 } __packed __aligned(4);
678 
679 CTASSERT(sizeof(struct mcx_cmd_query_issi_il_out) == MCX_CMDQ_INLINE_DATASIZE);
680 
681 struct mcx_cmd_query_issi_mb_out {
682 	uint8_t			cmd_reserved2[16];
683 	uint8_t			cmd_supported_issi[80]; /* very big endian */
684 } __packed __aligned(4);
685 
686 CTASSERT(sizeof(struct mcx_cmd_query_issi_mb_out) <= MCX_CMDQ_MAILBOX_DATASIZE);
687 
688 struct mcx_cmd_set_issi_in {
689 	uint16_t		cmd_opcode;
690 	uint8_t			cmd_reserved0[4];
691 	uint16_t		cmd_op_mod;
692 	uint8_t			cmd_reserved1[2];
693 	uint16_t		cmd_current_issi;
694 	uint8_t			cmd_reserved2[4];
695 } __packed __aligned(4);
696 
697 CTASSERT(sizeof(struct mcx_cmd_set_issi_in) <= MCX_CMDQ_INLINE_DATASIZE);
698 
699 struct mcx_cmd_set_issi_out {
700 	uint8_t			cmd_status;
701 	uint8_t			cmd_reserved0[3];
702 	uint32_t		cmd_syndrome;
703 	uint8_t			cmd_reserved1[8];
704 } __packed __aligned(4);
705 
706 CTASSERT(sizeof(struct mcx_cmd_set_issi_out) <= MCX_CMDQ_INLINE_DATASIZE);
707 
708 struct mcx_cmd_query_pages_in {
709 	uint16_t		cmd_opcode;
710 	uint8_t			cmd_reserved0[4];
711 	uint16_t		cmd_op_mod;
712 #define MCX_CMD_QUERY_PAGES_BOOT	0x01
713 #define MCX_CMD_QUERY_PAGES_INIT	0x02
714 #define MCX_CMD_QUERY_PAGES_REGULAR	0x03
715 	uint8_t			cmd_reserved1[8];
716 } __packed __aligned(4);
717 
718 struct mcx_cmd_query_pages_out {
719 	uint8_t			cmd_status;
720 	uint8_t			cmd_reserved0[3];
721 	uint32_t		cmd_syndrome;
722 	uint8_t			cmd_reserved1[2];
723 	uint16_t		cmd_func_id;
724 	int32_t			cmd_num_pages;
725 } __packed __aligned(4);
726 
727 struct mcx_cmd_manage_pages_in {
728 	uint16_t		cmd_opcode;
729 	uint8_t			cmd_reserved0[4];
730 	uint16_t		cmd_op_mod;
731 #define MCX_CMD_MANAGE_PAGES_ALLOC_FAIL \
732 					0x00
733 #define MCX_CMD_MANAGE_PAGES_ALLOC_SUCCESS \
734 					0x01
735 #define MCX_CMD_MANAGE_PAGES_HCA_RETURN_PAGES \
736 					0x02
737 	uint8_t			cmd_reserved1[2];
738 	uint16_t		cmd_func_id;
739 	uint32_t		cmd_input_num_entries;
740 } __packed __aligned(4);
741 
742 CTASSERT(sizeof(struct mcx_cmd_manage_pages_in) == MCX_CMDQ_INLINE_DATASIZE);
743 
744 struct mcx_cmd_manage_pages_out {
745 	uint8_t			cmd_status;
746 	uint8_t			cmd_reserved0[3];
747 	uint32_t		cmd_syndrome;
748 	uint32_t		cmd_output_num_entries;
749 	uint8_t			cmd_reserved1[4];
750 } __packed __aligned(4);
751 
752 CTASSERT(sizeof(struct mcx_cmd_manage_pages_out) == MCX_CMDQ_INLINE_DATASIZE);
753 
754 struct mcx_cmd_query_hca_cap_in {
755 	uint16_t		cmd_opcode;
756 	uint8_t			cmd_reserved0[4];
757 	uint16_t		cmd_op_mod;
758 #define MCX_CMD_QUERY_HCA_CAP_MAX	(0x0 << 0)
759 #define MCX_CMD_QUERY_HCA_CAP_CURRENT	(0x1 << 0)
760 #define MCX_CMD_QUERY_HCA_CAP_DEVICE	(0x0 << 1)
761 #define MCX_CMD_QUERY_HCA_CAP_OFFLOAD	(0x1 << 1)
762 #define MCX_CMD_QUERY_HCA_CAP_FLOW	(0x7 << 1)
763 	uint8_t			cmd_reserved1[8];
764 } __packed __aligned(4);
765 
766 struct mcx_cmd_query_hca_cap_out {
767 	uint8_t			cmd_status;
768 	uint8_t			cmd_reserved0[3];
769 	uint32_t		cmd_syndrome;
770 	uint8_t			cmd_reserved1[8];
771 } __packed __aligned(4);
772 
773 #define MCX_HCA_CAP_LEN			0x1000
774 #define MCX_HCA_CAP_NMAILBOXES		\
775 	(MCX_HCA_CAP_LEN / MCX_CMDQ_MAILBOX_DATASIZE)
776 
777 #if __GNUC_PREREQ__(4, 3)
778 #define __counter__		__COUNTER__
779 #else
780 #define __counter__		__LINE__
781 #endif
782 
783 #define __token(_tok, _num)	_tok##_num
784 #define _token(_tok, _num)	__token(_tok, _num)
785 #define __reserved__		_token(__reserved, __counter__)
786 
787 struct mcx_cap_device {
788 	uint8_t			reserved0[16];
789 
790 	uint8_t			log_max_srq_sz;
791 	uint8_t			log_max_qp_sz;
792 	uint8_t			__reserved__[1];
793 	uint8_t			log_max_qp; /* 5 bits */
794 #define MCX_CAP_DEVICE_LOG_MAX_QP	0x1f
795 
796 	uint8_t			__reserved__[1];
797 	uint8_t			log_max_srq; /* 5 bits */
798 #define MCX_CAP_DEVICE_LOG_MAX_SRQ	0x1f
799 	uint8_t			__reserved__[2];
800 
801 	uint8_t			__reserved__[1];
802 	uint8_t			log_max_cq_sz;
803 	uint8_t			__reserved__[1];
804 	uint8_t			log_max_cq; /* 5 bits */
805 #define MCX_CAP_DEVICE_LOG_MAX_CQ	0x1f
806 
807 	uint8_t			log_max_eq_sz;
808 	uint8_t			log_max_mkey; /* 6 bits */
809 #define MCX_CAP_DEVICE_LOG_MAX_MKEY	0x3f
810 	uint8_t			__reserved__[1];
811 	uint8_t			log_max_eq; /* 4 bits */
812 #define MCX_CAP_DEVICE_LOG_MAX_EQ	0x0f
813 
814 	uint8_t			max_indirection;
815 	uint8_t			log_max_mrw_sz; /* 7 bits */
816 #define MCX_CAP_DEVICE_LOG_MAX_MRW_SZ	0x7f
817 	uint8_t			teardown_log_max_msf_list_size;
818 #define MCX_CAP_DEVICE_FORCE_TEARDOWN	0x80
819 #define MCX_CAP_DEVICE_LOG_MAX_MSF_LIST_SIZE \
820 					0x3f
821 	uint8_t			log_max_klm_list_size; /* 6 bits */
822 #define MCX_CAP_DEVICE_LOG_MAX_KLM_LIST_SIZE \
823 					0x3f
824 
825 	uint8_t			__reserved__[1];
826 	uint8_t			log_max_ra_req_dc; /* 6 bits */
827 #define MCX_CAP_DEVICE_LOG_MAX_REQ_DC	0x3f
828 	uint8_t			__reserved__[1];
829 	uint8_t			log_max_ra_res_dc; /* 6 bits */
830 #define MCX_CAP_DEVICE_LOG_MAX_RA_RES_DC \
831 					0x3f
832 
833 	uint8_t			__reserved__[1];
834 	uint8_t			log_max_ra_req_qp; /* 6 bits */
835 #define MCX_CAP_DEVICE_LOG_MAX_RA_REQ_QP \
836 					0x3f
837 	uint8_t			__reserved__[1];
838 	uint8_t			log_max_ra_res_qp; /* 6 bits */
839 #define MCX_CAP_DEVICE_LOG_MAX_RA_RES_QP \
840 					0x3f
841 
842 	uint8_t			flags1;
843 #define MCX_CAP_DEVICE_END_PAD		0x80
844 #define MCX_CAP_DEVICE_CC_QUERY_ALLOWED	0x40
845 #define MCX_CAP_DEVICE_CC_MODIFY_ALLOWED \
846 					0x20
847 #define MCX_CAP_DEVICE_START_PAD	0x10
848 #define MCX_CAP_DEVICE_128BYTE_CACHELINE \
849 					0x08
850 	uint8_t			__reserved__[1];
851 	uint16_t		gid_table_size;
852 
853 	uint16_t		flags2;
854 #define MCX_CAP_DEVICE_OUT_OF_SEQ_CNT	0x8000
855 #define MCX_CAP_DEVICE_VPORT_COUNTERS	0x4000
856 #define MCX_CAP_DEVICE_RETRANSMISSION_Q_COUNTERS \
857 					0x2000
858 #define MCX_CAP_DEVICE_DEBUG		0x1000
859 #define MCX_CAP_DEVICE_MODIFY_RQ_COUNTERS_SET_ID \
860 					0x8000
861 #define MCX_CAP_DEVICE_RQ_DELAY_DROP	0x4000
862 #define MCX_CAP_DEVICe_MAX_QP_CNT_MASK	0x03ff
863 	uint16_t		pkey_table_size;
864 
865 	uint8_t			flags3;
866 #define MCX_CAP_DEVICE_VPORT_GROUP_MANAGER \
867 					0x80
868 #define MCX_CAP_DEVICE_VHCA_GROUP_MANAGER \
869 					0x40
870 #define MCX_CAP_DEVICE_IB_VIRTUAL	0x20
871 #define MCX_CAP_DEVICE_ETH_VIRTUAL	0x10
872 #define MCX_CAP_DEVICE_ETS		0x04
873 #define MCX_CAP_DEVICE_NIC_FLOW_TABLE	0x02
874 #define MCX_CAP_DEVICE_ESWITCH_FLOW_TABLE \
875 					0x01
876 	uint8_t			local_ca_ack_delay; /* 5 bits */
877 #define MCX_CAP_DEVICE_LOCAL_CA_ACK_DELAY \
878 					0x1f
879 #define MCX_CAP_DEVICE_MCAM_REG		0x40
880 	uint8_t			port_type;
881 #define MCX_CAP_DEVICE_PORT_MODULE_EVENT \
882 					0x80
883 #define MCX_CAP_DEVICE_PORT_TYPE	0x03
884 #define MCX_CAP_DEVICE_PORT_TYPE_ETH	0x01
885 	uint8_t			num_ports;
886 
887 	uint8_t			snapshot_log_max_msg;
888 #define MCX_CAP_DEVICE_SNAPSHOT		0x80
889 #define MCX_CAP_DEVICE_LOG_MAX_MSG	0x1f
890 	uint8_t			max_tc; /* 4 bits */
891 #define MCX_CAP_DEVICE_MAX_TC		0x0f
892 	uint8_t			flags4;
893 #define MCX_CAP_DEVICE_TEMP_WARN_EVENT	0x80
894 #define MCX_CAP_DEVICE_DCBX		0x40
895 #define MCX_CAP_DEVICE_ROL_S		0x02
896 #define MCX_CAP_DEVICE_ROL_G		0x01
897 	uint8_t			wol;
898 #define MCX_CAP_DEVICE_WOL_S		0x40
899 #define MCX_CAP_DEVICE_WOL_G		0x20
900 #define MCX_CAP_DEVICE_WOL_A		0x10
901 #define MCX_CAP_DEVICE_WOL_B		0x08
902 #define MCX_CAP_DEVICE_WOL_M		0x04
903 #define MCX_CAP_DEVICE_WOL_U		0x02
904 #define MCX_CAP_DEVICE_WOL_P		0x01
905 
906 	uint16_t		stat_rate_support;
907 	uint8_t			__reserved__[1];
908 	uint8_t			cqe_version; /* 4 bits */
909 #define MCX_CAP_DEVICE_CQE_VERSION	0x0f
910 
911 	uint32_t		flags5;
912 #define MCX_CAP_DEVICE_COMPACT_ADDRESS_VECTOR \
913 					0x80000000
914 #define MCX_CAP_DEVICE_STRIDING_RQ	0x40000000
915 #define MCX_CAP_DEVICE_IPOIP_ENHANCED_OFFLOADS \
916 					0x10000000
917 #define MCX_CAP_DEVICE_IPOIP_IPOIP_OFFLOADS \
918 					0x08000000
919 #define MCX_CAP_DEVICE_DC_CONNECT_CP	0x00040000
920 #define MCX_CAP_DEVICE_DC_CNAK_DRACE	0x00020000
921 #define MCX_CAP_DEVICE_DRAIN_SIGERR	0x00010000
922 #define MCX_CAP_DEVICE_DRAIN_SIGERR	0x00010000
923 #define MCX_CAP_DEVICE_CMDIF_CHECKSUM	0x0000c000
924 #define MCX_CAP_DEVICE_SIGERR_QCE	0x00002000
925 #define MCX_CAP_DEVICE_WQ_SIGNATURE	0x00000800
926 #define MCX_CAP_DEVICE_SCTR_DATA_CQE	0x00000400
927 #define MCX_CAP_DEVICE_SHO		0x00000100
928 #define MCX_CAP_DEVICE_TPH		0x00000080
929 #define MCX_CAP_DEVICE_RF		0x00000040
930 #define MCX_CAP_DEVICE_DCT		0x00000020
931 #define MCX_CAP_DEVICE_QOS		0x00000010
932 #define MCX_CAP_DEVICe_ETH_NET_OFFLOADS	0x00000008
933 #define MCX_CAP_DEVICE_ROCE		0x00000004
934 #define MCX_CAP_DEVICE_ATOMIC		0x00000002
935 
936 	uint32_t		flags6;
937 #define MCX_CAP_DEVICE_CQ_OI		0x80000000
938 #define MCX_CAP_DEVICE_CQ_RESIZE	0x40000000
939 #define MCX_CAP_DEVICE_CQ_MODERATION	0x20000000
940 #define MCX_CAP_DEVICE_CQ_PERIOD_MODE_MODIFY \
941 					0x10000000
942 #define MCX_CAP_DEVICE_CQ_INVALIDATE	0x08000000
943 #define MCX_CAP_DEVICE_RESERVED_AT_255	0x04000000
944 #define MCX_CAP_DEVICE_CQ_EQ_REMAP	0x02000000
945 #define MCX_CAP_DEVICE_PG		0x01000000
946 #define MCX_CAP_DEVICE_BLOCK_LB_MC	0x00800000
947 #define MCX_CAP_DEVICE_EXPONENTIAL_BACKOFF \
948 					0x00400000
949 #define MCX_CAP_DEVICE_SCQE_BREAK_MODERATION \
950 					0x00200000
951 #define MCX_CAP_DEVICE_CQ_PERIOD_START_FROM_CQE \
952 					0x00100000
953 #define MCX_CAP_DEVICE_CD		0x00080000
954 #define MCX_CAP_DEVICE_ATM		0x00040000
955 #define MCX_CAP_DEVICE_APM		0x00020000
956 #define MCX_CAP_DEVICE_IMAICL		0x00010000
957 #define MCX_CAP_DEVICE_QKV		0x00000200
958 #define MCX_CAP_DEVICE_PKV		0x00000100
959 #define MCX_CAP_DEVICE_SET_DETH_SQPN	0x00000080
960 #define MCX_CAP_DEVICE_XRC		0x00000008
961 #define MCX_CAP_DEVICE_UD		0x00000004
962 #define MCX_CAP_DEVICE_UC		0x00000002
963 #define MCX_CAP_DEVICE_RC		0x00000001
964 
965 	uint8_t			uar_flags;
966 #define MCX_CAP_DEVICE_UAR_4K		0x80
967 	uint8_t			uar_sz;	/* 6 bits */
968 #define MCX_CAP_DEVICE_UAR_SZ		0x3f
969 	uint8_t			__reserved__[1];
970 	uint8_t			log_pg_sz;
971 
972 	uint8_t			flags7;
973 #define MCX_CAP_DEVICE_BF		0x80
974 #define MCX_CAP_DEVICE_DRIVER_VERSION	0x40
975 #define MCX_CAP_DEVICE_PAD_TX_ETH_PACKET \
976 					0x20
977 	uint8_t			log_bf_reg_size; /* 5 bits */
978 #define MCX_CAP_DEVICE_LOG_BF_REG_SIZE	0x1f
979 	uint8_t			__reserved__[2];
980 
981 	uint16_t		num_of_diagnostic_counters;
982 	uint16_t		max_wqe_sz_sq;
983 
984 	uint8_t			__reserved__[2];
985 	uint16_t		max_wqe_sz_rq;
986 
987 	uint8_t			__reserved__[2];
988 	uint16_t		max_wqe_sz_sq_dc;
989 
990 	uint32_t		max_qp_mcg; /* 25 bits */
991 #define MCX_CAP_DEVICE_MAX_QP_MCG	0x1ffffff
992 
993 	uint8_t			__reserved__[3];
994 	uint8_t			log_max_mcq;
995 
996 	uint8_t			log_max_transport_domain; /* 5 bits */
997 #define MCX_CAP_DEVICE_LOG_MAX_TRANSORT_DOMAIN \
998 					0x1f
999 	uint8_t			log_max_pd; /* 5 bits */
1000 #define MCX_CAP_DEVICE_LOG_MAX_PD	0x1f
1001 	uint8_t			__reserved__[1];
1002 	uint8_t			log_max_xrcd; /* 5 bits */
1003 #define MCX_CAP_DEVICE_LOG_MAX_XRCD	0x1f
1004 
1005 	uint8_t			__reserved__[2];
1006 	uint16_t		max_flow_counter;
1007 
1008 	uint8_t			log_max_rq; /* 5 bits */
1009 #define MCX_CAP_DEVICE_LOG_MAX_RQ	0x1f
1010 	uint8_t			log_max_sq; /* 5 bits */
1011 #define MCX_CAP_DEVICE_LOG_MAX_SQ	0x1f
1012 	uint8_t			log_max_tir; /* 5 bits */
1013 #define MCX_CAP_DEVICE_LOG_MAX_TIR	0x1f
1014 	uint8_t			log_max_tis; /* 5 bits */
1015 #define MCX_CAP_DEVICE_LOG_MAX_TIS	0x1f
1016 
1017 	uint8_t 		flags8;
1018 #define MCX_CAP_DEVICE_BASIC_CYCLIC_RCV_WQE \
1019 					0x80
1020 #define MCX_CAP_DEVICE_LOG_MAX_RMP	0x1f
1021 	uint8_t			log_max_rqt; /* 5 bits */
1022 #define MCX_CAP_DEVICE_LOG_MAX_RQT	0x1f
1023 	uint8_t			log_max_rqt_size; /* 5 bits */
1024 #define MCX_CAP_DEVICE_LOG_MAX_RQT_SIZE	0x1f
1025 	uint8_t			log_max_tis_per_sq; /* 5 bits */
1026 #define MCX_CAP_DEVICE_LOG_MAX_TIS_PER_SQ \
1027 					0x1f
1028 
1029 	uint8_t			flags9;
1030 #define MXC_CAP_DEVICE_EXT_STRIDE_NUM_RANGES \
1031 					0x80
1032 #define MXC_CAP_DEVICE_LOG_MAX_STRIDE_SZ_RQ \
1033 					0x1f
1034 	uint8_t			log_min_stride_sz_rq; /* 5 bits */
1035 #define MXC_CAP_DEVICE_LOG_MIN_STRIDE_SZ_RQ \
1036 					0x1f
1037 	uint8_t			log_max_stride_sz_sq; /* 5 bits */
1038 #define MXC_CAP_DEVICE_LOG_MAX_STRIDE_SZ_SQ \
1039 					0x1f
1040 	uint8_t			log_min_stride_sz_sq; /* 5 bits */
1041 #define MXC_CAP_DEVICE_LOG_MIN_STRIDE_SZ_SQ \
1042 					0x1f
1043 
1044 	uint8_t			log_max_hairpin_queues;
1045 #define MXC_CAP_DEVICE_HAIRPIN		0x80
1046 #define MXC_CAP_DEVICE_LOG_MAX_HAIRPIN_QUEUES \
1047 					0x1f
1048 	uint8_t			log_min_hairpin_queues;
1049 #define MXC_CAP_DEVICE_LOG_MIN_HAIRPIN_QUEUES \
1050 					0x1f
1051 	uint8_t			log_max_hairpin_num_packets;
1052 #define MXC_CAP_DEVICE_LOG_MAX_HAIRPIN_NUM_PACKETS \
1053 					0x1f
1054 	uint8_t			log_max_mq_sz;
1055 #define MXC_CAP_DEVICE_LOG_MAX_WQ_SZ \
1056 					0x1f
1057 
1058 	uint8_t			log_min_hairpin_wq_data_sz;
1059 #define MXC_CAP_DEVICE_NIC_VPORT_CHANGE_EVENT \
1060 					0x80
1061 #define MXC_CAP_DEVICE_DISABLE_LOCAL_LB_UC \
1062 					0x40
1063 #define MXC_CAP_DEVICE_DISABLE_LOCAL_LB_MC \
1064 					0x20
1065 #define MCX_CAP_DEVICE_LOG_MIN_HAIRPIN_WQ_DATA_SZ \
1066 					0x1f
1067 	uint8_t			log_max_vlan_list;
1068 #define MXC_CAP_DEVICE_SYSTEM_IMAGE_GUID_MODIFIABLE \
1069 					0x80
1070 #define MXC_CAP_DEVICE_LOG_MAX_VLAN_LIST \
1071 					0x1f
1072 	uint8_t			log_max_current_mc_list;
1073 #define MXC_CAP_DEVICE_LOG_MAX_CURRENT_MC_LIST \
1074 					0x1f
1075 	uint8_t			log_max_current_uc_list;
1076 #define MXC_CAP_DEVICE_LOG_MAX_CURRENT_UC_LIST \
1077 					0x1f
1078 
1079 	uint8_t			__reserved__[4];
1080 
1081 	uint32_t		create_qp_start_hint; /* 24 bits */
1082 
1083 	uint8_t			log_max_uctx; /* 5 bits */
1084 #define MXC_CAP_DEVICE_LOG_MAX_UCTX	0x1f
1085 	uint8_t			log_max_umem; /* 5 bits */
1086 #define MXC_CAP_DEVICE_LOG_MAX_UMEM	0x1f
1087 	uint16_t		max_num_eqs;
1088 
1089 	uint8_t			log_max_l2_table; /* 5 bits */
1090 #define MXC_CAP_DEVICE_LOG_MAX_L2_TABLE	0x1f
1091 	uint8_t			__reserved__[1];
1092 	uint16_t		log_uar_page_sz;
1093 
1094 	uint8_t			__reserved__[8];
1095 
1096 	uint32_t		device_frequency_mhz;
1097 	uint32_t		device_frequency_khz;
1098 } __packed __aligned(8);
1099 
1100 CTASSERT(offsetof(struct mcx_cap_device, max_indirection) == 0x20);
1101 CTASSERT(offsetof(struct mcx_cap_device, flags1) == 0x2c);
1102 CTASSERT(offsetof(struct mcx_cap_device, flags2) == 0x30);
1103 CTASSERT(offsetof(struct mcx_cap_device, snapshot_log_max_msg) == 0x38);
1104 CTASSERT(offsetof(struct mcx_cap_device, flags5) == 0x40);
1105 CTASSERT(offsetof(struct mcx_cap_device, flags7) == 0x4c);
1106 CTASSERT(offsetof(struct mcx_cap_device, device_frequency_mhz) == 0x98);
1107 CTASSERT(offsetof(struct mcx_cap_device, device_frequency_khz) == 0x9c);
1108 CTASSERT(sizeof(struct mcx_cap_device) <= MCX_CMDQ_MAILBOX_DATASIZE);
1109 
1110 struct mcx_cmd_set_driver_version_in {
1111 	uint16_t		cmd_opcode;
1112 	uint8_t			cmd_reserved0[4];
1113 	uint16_t		cmd_op_mod;
1114 	uint8_t			cmd_reserved1[8];
1115 } __packed __aligned(4);
1116 
1117 struct mcx_cmd_set_driver_version_out {
1118 	uint8_t			cmd_status;
1119 	uint8_t			cmd_reserved0[3];
1120 	uint32_t		cmd_syndrome;
1121 	uint8_t			cmd_reserved1[8];
1122 } __packed __aligned(4);
1123 
1124 struct mcx_cmd_set_driver_version {
1125 	uint8_t			cmd_driver_version[64];
1126 } __packed __aligned(8);
1127 
1128 struct mcx_cmd_modify_nic_vport_context_in {
1129 	uint16_t		cmd_opcode;
1130 	uint8_t			cmd_reserved0[4];
1131 	uint16_t		cmd_op_mod;
1132 	uint8_t			cmd_reserved1[4];
1133 	uint32_t		cmd_field_select;
1134 #define MCX_CMD_MODIFY_NIC_VPORT_CONTEXT_FIELD_ADDR	0x04
1135 #define MCX_CMD_MODIFY_NIC_VPORT_CONTEXT_FIELD_PROMISC	0x10
1136 #define MCX_CMD_MODIFY_NIC_VPORT_CONTEXT_FIELD_MTU	0x40
1137 } __packed __aligned(4);
1138 
1139 struct mcx_cmd_modify_nic_vport_context_out {
1140 	uint8_t			cmd_status;
1141 	uint8_t			cmd_reserved0[3];
1142 	uint32_t		cmd_syndrome;
1143 	uint8_t			cmd_reserved1[8];
1144 } __packed __aligned(4);
1145 
1146 struct mcx_cmd_query_nic_vport_context_in {
1147 	uint16_t		cmd_opcode;
1148 	uint8_t			cmd_reserved0[4];
1149 	uint16_t		cmd_op_mod;
1150 	uint8_t			cmd_reserved1[4];
1151 	uint8_t			cmd_allowed_list_type;
1152 	uint8_t			cmd_reserved2[3];
1153 } __packed __aligned(4);
1154 
1155 struct mcx_cmd_query_nic_vport_context_out {
1156 	uint8_t			cmd_status;
1157 	uint8_t			cmd_reserved0[3];
1158 	uint32_t		cmd_syndrome;
1159 	uint8_t			cmd_reserved1[8];
1160 } __packed __aligned(4);
1161 
1162 struct mcx_nic_vport_ctx {
1163 	uint32_t		vp_min_wqe_inline_mode;
1164 	uint8_t			vp_reserved0[32];
1165 	uint32_t		vp_mtu;
1166 	uint8_t			vp_reserved1[200];
1167 	uint16_t		vp_flags;
1168 #define MCX_NIC_VPORT_CTX_LIST_UC_MAC			(0)
1169 #define MCX_NIC_VPORT_CTX_LIST_MC_MAC			(1 << 24)
1170 #define MCX_NIC_VPORT_CTX_LIST_VLAN			(2 << 24)
1171 #define MCX_NIC_VPORT_CTX_PROMISC_ALL			(1 << 13)
1172 #define MCX_NIC_VPORT_CTX_PROMISC_MCAST			(1 << 14)
1173 #define MCX_NIC_VPORT_CTX_PROMISC_UCAST			(1 << 15)
1174 	uint16_t		vp_allowed_list_size;
1175 	uint64_t		vp_perm_addr;
1176 	uint8_t			vp_reserved2[4];
1177 	/* allowed list follows */
1178 } __packed __aligned(4);
1179 
1180 struct mcx_counter {
1181 	uint64_t		packets;
1182 	uint64_t		octets;
1183 } __packed __aligned(4);
1184 
1185 struct mcx_nic_vport_counters {
1186 	struct mcx_counter	rx_err;
1187 	struct mcx_counter	tx_err;
1188 	uint8_t			reserved0[64]; /* 0x30 */
1189 	struct mcx_counter	rx_bcast;
1190 	struct mcx_counter	tx_bcast;
1191 	struct mcx_counter	rx_ucast;
1192 	struct mcx_counter	tx_ucast;
1193 	struct mcx_counter	rx_mcast;
1194 	struct mcx_counter	tx_mcast;
1195 	uint8_t			reserved1[0x210 - 0xd0];
1196 } __packed __aligned(4);
1197 
1198 struct mcx_cmd_query_vport_counters_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_query_vport_counters_mb_in {
1206 	uint8_t			cmd_reserved0[8];
1207 	uint8_t			cmd_clear;
1208 	uint8_t			cmd_reserved1[7];
1209 } __packed __aligned(4);
1210 
1211 struct mcx_cmd_query_vport_counters_out {
1212 	uint8_t			cmd_status;
1213 	uint8_t			cmd_reserved0[3];
1214 	uint32_t		cmd_syndrome;
1215 	uint8_t			cmd_reserved1[8];
1216 } __packed __aligned(4);
1217 
1218 struct mcx_cmd_query_flow_counter_in {
1219 	uint16_t		cmd_opcode;
1220 	uint8_t			cmd_reserved0[4];
1221 	uint16_t		cmd_op_mod;
1222 	uint8_t			cmd_reserved1[8];
1223 } __packed __aligned(4);
1224 
1225 struct mcx_cmd_query_flow_counter_mb_in {
1226 	uint8_t			cmd_reserved0[8];
1227 	uint8_t			cmd_clear;
1228 	uint8_t			cmd_reserved1[5];
1229 	uint16_t		cmd_flow_counter_id;
1230 } __packed __aligned(4);
1231 
1232 struct mcx_cmd_query_flow_counter_out {
1233 	uint8_t			cmd_status;
1234 	uint8_t			cmd_reserved0[3];
1235 	uint32_t		cmd_syndrome;
1236 	uint8_t			cmd_reserved1[8];
1237 } __packed __aligned(4);
1238 
1239 struct mcx_cmd_alloc_uar_in {
1240 	uint16_t		cmd_opcode;
1241 	uint8_t			cmd_reserved0[4];
1242 	uint16_t		cmd_op_mod;
1243 	uint8_t			cmd_reserved1[8];
1244 } __packed __aligned(4);
1245 
1246 struct mcx_cmd_alloc_uar_out {
1247 	uint8_t			cmd_status;
1248 	uint8_t			cmd_reserved0[3];
1249 	uint32_t		cmd_syndrome;
1250 	uint32_t		cmd_uar;
1251 	uint8_t			cmd_reserved1[4];
1252 } __packed __aligned(4);
1253 
1254 struct mcx_cmd_query_special_ctx_in {
1255 	uint16_t		cmd_opcode;
1256 	uint8_t			cmd_reserved0[4];
1257 	uint16_t		cmd_op_mod;
1258 	uint8_t			cmd_reserved1[8];
1259 } __packed __aligned(4);
1260 
1261 struct mcx_cmd_query_special_ctx_out {
1262 	uint8_t			cmd_status;
1263 	uint8_t			cmd_reserved0[3];
1264 	uint32_t		cmd_syndrome;
1265 	uint8_t			cmd_reserved1[4];
1266 	uint32_t		cmd_resd_lkey;
1267 } __packed __aligned(4);
1268 
1269 struct mcx_eq_ctx {
1270 	uint32_t		eq_status;
1271 #define MCX_EQ_CTX_STATE_SHIFT		8
1272 #define MCX_EQ_CTX_STATE_MASK		(0xf << MCX_EQ_CTX_STATE_SHIFT)
1273 #define MCX_EQ_CTX_STATE_ARMED		0x9
1274 #define MCX_EQ_CTX_STATE_FIRED		0xa
1275 #define MCX_EQ_CTX_OI_SHIFT		17
1276 #define MCX_EQ_CTX_OI			(1 << MCX_EQ_CTX_OI_SHIFT)
1277 #define MCX_EQ_CTX_EC_SHIFT		18
1278 #define MCX_EQ_CTX_EC			(1 << MCX_EQ_CTX_EC_SHIFT)
1279 #define MCX_EQ_CTX_STATUS_SHIFT		28
1280 #define MCX_EQ_CTX_STATUS_MASK		(0xf << MCX_EQ_CTX_STATUS_SHIFT)
1281 #define MCX_EQ_CTX_STATUS_OK		0x0
1282 #define MCX_EQ_CTX_STATUS_EQ_WRITE_FAILURE 0xa
1283 	uint32_t		eq_reserved1;
1284 	uint32_t		eq_page_offset;
1285 #define MCX_EQ_CTX_PAGE_OFFSET_SHIFT	5
1286 	uint32_t		eq_uar_size;
1287 #define MCX_EQ_CTX_UAR_PAGE_MASK	0xffffff
1288 #define MCX_EQ_CTX_LOG_EQ_SIZE_SHIFT	24
1289 	uint32_t		eq_reserved2;
1290 	uint8_t			eq_reserved3[3];
1291 	uint8_t			eq_intr;
1292 	uint32_t		eq_log_page_size;
1293 #define MCX_EQ_CTX_LOG_PAGE_SIZE_SHIFT	24
1294 	uint32_t		eq_reserved4[3];
1295 	uint32_t		eq_consumer_counter;
1296 	uint32_t		eq_producer_counter;
1297 #define MCX_EQ_CTX_COUNTER_MASK		0xffffff
1298 	uint32_t		eq_reserved5[4];
1299 } __packed __aligned(4);
1300 
1301 CTASSERT(sizeof(struct mcx_eq_ctx) == 64);
1302 
1303 struct mcx_cmd_create_eq_in {
1304 	uint16_t		cmd_opcode;
1305 	uint8_t			cmd_reserved0[4];
1306 	uint16_t		cmd_op_mod;
1307 	uint8_t			cmd_reserved1[8];
1308 } __packed __aligned(4);
1309 
1310 struct mcx_cmd_create_eq_mb_in {
1311 	struct mcx_eq_ctx	cmd_eq_ctx;
1312 	uint8_t			cmd_reserved0[8];
1313 	uint64_t		cmd_event_bitmask;
1314 #define MCX_EVENT_TYPE_COMPLETION	0x00
1315 #define MCX_EVENT_TYPE_CQ_ERROR		0x04
1316 #define MCX_EVENT_TYPE_INTERNAL_ERROR	0x08
1317 #define MCX_EVENT_TYPE_PORT_CHANGE	0x09
1318 #define MCX_EVENT_TYPE_CMD_COMPLETION	0x0a
1319 #define MCX_EVENT_TYPE_PAGE_REQUEST	0x0b
1320 #define MCX_EVENT_TYPE_LAST_WQE		0x13
1321 	uint8_t			cmd_reserved1[176];
1322 } __packed __aligned(4);
1323 
1324 struct mcx_cmd_create_eq_out {
1325 	uint8_t			cmd_status;
1326 	uint8_t			cmd_reserved0[3];
1327 	uint32_t		cmd_syndrome;
1328 	uint32_t		cmd_eqn;
1329 	uint8_t			cmd_reserved1[4];
1330 } __packed __aligned(4);
1331 
1332 struct mcx_cmd_query_eq_in {
1333 	uint16_t		cmd_opcode;
1334 	uint8_t			cmd_reserved0[4];
1335 	uint16_t		cmd_op_mod;
1336 	uint32_t		cmd_eqn;
1337 	uint8_t			cmd_reserved1[4];
1338 } __packed __aligned(4);
1339 
1340 struct mcx_cmd_query_eq_out {
1341 	uint8_t			cmd_status;
1342 	uint8_t			cmd_reserved0[3];
1343 	uint32_t		cmd_syndrome;
1344 	uint8_t			cmd_reserved1[8];
1345 } __packed __aligned(4);
1346 
1347 struct mcx_eq_entry {
1348 	uint8_t			eq_reserved1;
1349 	uint8_t			eq_event_type;
1350 	uint8_t			eq_reserved2;
1351 	uint8_t			eq_event_sub_type;
1352 
1353 	uint8_t			eq_reserved3[28];
1354 	uint32_t		eq_event_data[7];
1355 	uint8_t			eq_reserved4[2];
1356 	uint8_t			eq_signature;
1357 	uint8_t			eq_owner;
1358 #define MCX_EQ_ENTRY_OWNER_INIT			1
1359 } __packed __aligned(4);
1360 
1361 CTASSERT(sizeof(struct mcx_eq_entry) == 64);
1362 
1363 struct mcx_cmd_alloc_pd_in {
1364 	uint16_t		cmd_opcode;
1365 	uint8_t			cmd_reserved0[4];
1366 	uint16_t		cmd_op_mod;
1367 	uint8_t			cmd_reserved1[8];
1368 } __packed __aligned(4);
1369 
1370 struct mcx_cmd_alloc_pd_out {
1371 	uint8_t			cmd_status;
1372 	uint8_t			cmd_reserved0[3];
1373 	uint32_t		cmd_syndrome;
1374 	uint32_t		cmd_pd;
1375 	uint8_t			cmd_reserved1[4];
1376 } __packed __aligned(4);
1377 
1378 struct mcx_cmd_alloc_td_in {
1379 	uint16_t		cmd_opcode;
1380 	uint8_t			cmd_reserved0[4];
1381 	uint16_t		cmd_op_mod;
1382 	uint8_t			cmd_reserved1[8];
1383 } __packed __aligned(4);
1384 
1385 struct mcx_cmd_alloc_td_out {
1386 	uint8_t			cmd_status;
1387 	uint8_t			cmd_reserved0[3];
1388 	uint32_t		cmd_syndrome;
1389 	uint32_t		cmd_tdomain;
1390 	uint8_t			cmd_reserved1[4];
1391 } __packed __aligned(4);
1392 
1393 struct mcx_cmd_create_tir_in {
1394 	uint16_t		cmd_opcode;
1395 	uint8_t			cmd_reserved0[4];
1396 	uint16_t		cmd_op_mod;
1397 	uint8_t			cmd_reserved1[8];
1398 } __packed __aligned(4);
1399 
1400 struct mcx_cmd_create_tir_mb_in {
1401 	uint8_t			cmd_reserved0[20];
1402 	uint32_t		cmd_disp_type;
1403 #define MCX_TIR_CTX_DISP_TYPE_DIRECT	0
1404 #define MCX_TIR_CTX_DISP_TYPE_INDIRECT	1
1405 #define MCX_TIR_CTX_DISP_TYPE_SHIFT	28
1406 	uint8_t			cmd_reserved1[8];
1407 	uint32_t		cmd_lro;
1408 	uint8_t			cmd_reserved2[8];
1409 	uint32_t		cmd_inline_rqn;
1410 	uint32_t		cmd_indir_table;
1411 	uint32_t		cmd_tdomain;
1412 #define MCX_TIR_CTX_HASH_TOEPLITZ	2
1413 #define MCX_TIR_CTX_HASH_SHIFT		28
1414 	uint8_t			cmd_rx_hash_key[40];
1415 	uint32_t		cmd_rx_hash_sel_outer;
1416 #define MCX_TIR_CTX_HASH_SEL_SRC_IP	(1 << 0)
1417 #define MCX_TIR_CTX_HASH_SEL_DST_IP	(1 << 1)
1418 #define MCX_TIR_CTX_HASH_SEL_SPORT	(1 << 2)
1419 #define MCX_TIR_CTX_HASH_SEL_DPORT	(1 << 3)
1420 #define MCX_TIR_CTX_HASH_SEL_IPV4	(0 << 31)
1421 #define MCX_TIR_CTX_HASH_SEL_IPV6	(1 << 31)
1422 #define MCX_TIR_CTX_HASH_SEL_TCP	(0 << 30)
1423 #define MCX_TIR_CTX_HASH_SEL_UDP	(1 << 30)
1424 	uint32_t		cmd_rx_hash_sel_inner;
1425 	uint8_t			cmd_reserved3[152];
1426 } __packed __aligned(4);
1427 
1428 struct mcx_cmd_create_tir_out {
1429 	uint8_t			cmd_status;
1430 	uint8_t			cmd_reserved0[3];
1431 	uint32_t		cmd_syndrome;
1432 	uint32_t		cmd_tirn;
1433 	uint8_t			cmd_reserved1[4];
1434 } __packed __aligned(4);
1435 
1436 struct mcx_cmd_destroy_tir_in {
1437 	uint16_t		cmd_opcode;
1438 	uint8_t			cmd_reserved0[4];
1439 	uint16_t		cmd_op_mod;
1440 	uint32_t		cmd_tirn;
1441 	uint8_t			cmd_reserved1[4];
1442 } __packed __aligned(4);
1443 
1444 struct mcx_cmd_destroy_tir_out {
1445 	uint8_t			cmd_status;
1446 	uint8_t			cmd_reserved0[3];
1447 	uint32_t		cmd_syndrome;
1448 	uint8_t			cmd_reserved1[8];
1449 } __packed __aligned(4);
1450 
1451 struct mcx_cmd_create_tis_in {
1452 	uint16_t		cmd_opcode;
1453 	uint8_t			cmd_reserved0[4];
1454 	uint16_t		cmd_op_mod;
1455 	uint8_t			cmd_reserved1[8];
1456 } __packed __aligned(4);
1457 
1458 struct mcx_cmd_create_tis_mb_in {
1459 	uint8_t			cmd_reserved[16];
1460 	uint32_t		cmd_prio;
1461 	uint8_t			cmd_reserved1[32];
1462 	uint32_t		cmd_tdomain;
1463 	uint8_t			cmd_reserved2[120];
1464 } __packed __aligned(4);
1465 
1466 struct mcx_cmd_create_tis_out {
1467 	uint8_t			cmd_status;
1468 	uint8_t			cmd_reserved0[3];
1469 	uint32_t		cmd_syndrome;
1470 	uint32_t		cmd_tisn;
1471 	uint8_t			cmd_reserved1[4];
1472 } __packed __aligned(4);
1473 
1474 struct mcx_cmd_destroy_tis_in {
1475 	uint16_t		cmd_opcode;
1476 	uint8_t			cmd_reserved0[4];
1477 	uint16_t		cmd_op_mod;
1478 	uint32_t		cmd_tisn;
1479 	uint8_t			cmd_reserved1[4];
1480 } __packed __aligned(4);
1481 
1482 struct mcx_cmd_destroy_tis_out {
1483 	uint8_t			cmd_status;
1484 	uint8_t			cmd_reserved0[3];
1485 	uint32_t		cmd_syndrome;
1486 	uint8_t			cmd_reserved1[8];
1487 } __packed __aligned(4);
1488 
1489 struct mcx_cmd_create_rqt_in {
1490 	uint16_t		cmd_opcode;
1491 	uint8_t			cmd_reserved0[4];
1492 	uint16_t		cmd_op_mod;
1493 	uint8_t			cmd_reserved1[8];
1494 } __packed __aligned(4);
1495 
1496 struct mcx_rqt_ctx {
1497 	uint8_t			cmd_reserved0[20];
1498 	uint16_t		cmd_reserved1;
1499 	uint16_t		cmd_rqt_max_size;
1500 	uint16_t		cmd_reserved2;
1501 	uint16_t		cmd_rqt_actual_size;
1502 	uint8_t			cmd_reserved3[212];
1503 } __packed __aligned(4);
1504 
1505 struct mcx_cmd_create_rqt_mb_in {
1506 	uint8_t			cmd_reserved0[16];
1507 	struct mcx_rqt_ctx	cmd_rqt;
1508 } __packed __aligned(4);
1509 
1510 struct mcx_cmd_create_rqt_out {
1511 	uint8_t			cmd_status;
1512 	uint8_t			cmd_reserved0[3];
1513 	uint32_t		cmd_syndrome;
1514 	uint32_t		cmd_rqtn;
1515 	uint8_t			cmd_reserved1[4];
1516 } __packed __aligned(4);
1517 
1518 struct mcx_cmd_destroy_rqt_in {
1519 	uint16_t		cmd_opcode;
1520 	uint8_t			cmd_reserved0[4];
1521 	uint16_t		cmd_op_mod;
1522 	uint32_t		cmd_rqtn;
1523 	uint8_t			cmd_reserved1[4];
1524 } __packed __aligned(4);
1525 
1526 struct mcx_cmd_destroy_rqt_out {
1527 	uint8_t			cmd_status;
1528 	uint8_t			cmd_reserved0[3];
1529 	uint32_t		cmd_syndrome;
1530 	uint8_t			cmd_reserved1[8];
1531 } __packed __aligned(4);
1532 
1533 struct mcx_cq_ctx {
1534 	uint32_t		cq_status;
1535 #define MCX_CQ_CTX_STATUS_SHIFT		28
1536 #define MCX_CQ_CTX_STATUS_MASK		(0xf << MCX_CQ_CTX_STATUS_SHIFT)
1537 #define MCX_CQ_CTX_STATUS_OK		0x0
1538 #define MCX_CQ_CTX_STATUS_OVERFLOW	0x9
1539 #define MCX_CQ_CTX_STATUS_WRITE_FAIL	0xa
1540 #define MCX_CQ_CTX_STATE_SHIFT		8
1541 #define MCX_CQ_CTX_STATE_MASK		(0xf << MCX_CQ_CTX_STATE_SHIFT)
1542 #define MCX_CQ_CTX_STATE_SOLICITED	0x6
1543 #define MCX_CQ_CTX_STATE_ARMED		0x9
1544 #define MCX_CQ_CTX_STATE_FIRED		0xa
1545 	uint32_t		cq_reserved1;
1546 	uint32_t		cq_page_offset;
1547 	uint32_t		cq_uar_size;
1548 #define MCX_CQ_CTX_UAR_PAGE_MASK	0xffffff
1549 #define MCX_CQ_CTX_LOG_CQ_SIZE_SHIFT	24
1550 	uint32_t		cq_period_max_count;
1551 #define MCX_CQ_CTX_PERIOD_SHIFT		16
1552 	uint32_t		cq_eqn;
1553 	uint32_t		cq_log_page_size;
1554 #define MCX_CQ_CTX_LOG_PAGE_SIZE_SHIFT	24
1555 	uint32_t		cq_reserved2;
1556 	uint32_t		cq_last_notified;
1557 	uint32_t		cq_last_solicit;
1558 	uint32_t		cq_consumer_counter;
1559 	uint32_t		cq_producer_counter;
1560 	uint8_t			cq_reserved3[8];
1561 	uint64_t		cq_doorbell;
1562 } __packed __aligned(4);
1563 
1564 CTASSERT(sizeof(struct mcx_cq_ctx) == 64);
1565 
1566 struct mcx_cmd_create_cq_in {
1567 	uint16_t		cmd_opcode;
1568 	uint8_t			cmd_reserved0[4];
1569 	uint16_t		cmd_op_mod;
1570 	uint8_t			cmd_reserved1[8];
1571 } __packed __aligned(4);
1572 
1573 struct mcx_cmd_create_cq_mb_in {
1574 	struct mcx_cq_ctx	cmd_cq_ctx;
1575 	uint8_t			cmd_reserved1[192];
1576 } __packed __aligned(4);
1577 
1578 struct mcx_cmd_create_cq_out {
1579 	uint8_t			cmd_status;
1580 	uint8_t			cmd_reserved0[3];
1581 	uint32_t		cmd_syndrome;
1582 	uint32_t		cmd_cqn;
1583 	uint8_t			cmd_reserved1[4];
1584 } __packed __aligned(4);
1585 
1586 struct mcx_cmd_destroy_cq_in {
1587 	uint16_t		cmd_opcode;
1588 	uint8_t			cmd_reserved0[4];
1589 	uint16_t		cmd_op_mod;
1590 	uint32_t		cmd_cqn;
1591 	uint8_t			cmd_reserved1[4];
1592 } __packed __aligned(4);
1593 
1594 struct mcx_cmd_destroy_cq_out {
1595 	uint8_t			cmd_status;
1596 	uint8_t			cmd_reserved0[3];
1597 	uint32_t		cmd_syndrome;
1598 	uint8_t			cmd_reserved1[8];
1599 } __packed __aligned(4);
1600 
1601 struct mcx_cmd_query_cq_in {
1602 	uint16_t		cmd_opcode;
1603 	uint8_t			cmd_reserved0[4];
1604 	uint16_t		cmd_op_mod;
1605 	uint32_t		cmd_cqn;
1606 	uint8_t			cmd_reserved1[4];
1607 } __packed __aligned(4);
1608 
1609 struct mcx_cmd_query_cq_out {
1610 	uint8_t			cmd_status;
1611 	uint8_t			cmd_reserved0[3];
1612 	uint32_t		cmd_syndrome;
1613 	uint8_t			cmd_reserved1[8];
1614 } __packed __aligned(4);
1615 
1616 struct mcx_cq_entry {
1617 	uint32_t		__reserved__;
1618 	uint32_t		cq_lro;
1619 	uint32_t		cq_lro_ack_seq_num;
1620 	uint32_t		cq_rx_hash;
1621 	uint8_t			cq_rx_hash_type;
1622 	uint8_t			cq_ml_path;
1623 	uint16_t		__reserved__;
1624 	uint32_t		cq_checksum;
1625 	uint32_t		__reserved__;
1626 	uint32_t		cq_flags;
1627 #define MCX_CQ_ENTRY_FLAGS_L4_OK		(1 << 26)
1628 #define MCX_CQ_ENTRY_FLAGS_L3_OK		(1 << 25)
1629 #define MCX_CQ_ENTRY_FLAGS_L2_OK		(1 << 24)
1630 #define MCX_CQ_ENTRY_FLAGS_CV			(1 << 16)
1631 #define MCX_CQ_ENTRY_FLAGS_VLAN_MASK		(0xffff)
1632 
1633 	uint32_t		cq_lro_srqn;
1634 	uint32_t		__reserved__[2];
1635 	uint32_t		cq_byte_cnt;
1636 	uint64_t		cq_timestamp;
1637 	uint8_t			cq_rx_drops;
1638 	uint8_t			cq_flow_tag[3];
1639 	uint16_t		cq_wqe_count;
1640 	uint8_t			cq_signature;
1641 	uint8_t			cq_opcode_owner;
1642 #define MCX_CQ_ENTRY_FLAG_OWNER			(1 << 0)
1643 #define MCX_CQ_ENTRY_FLAG_SE			(1 << 1)
1644 #define MCX_CQ_ENTRY_FORMAT_SHIFT		2
1645 #define MCX_CQ_ENTRY_OPCODE_SHIFT		4
1646 
1647 #define MCX_CQ_ENTRY_FORMAT_NO_INLINE		0
1648 #define MCX_CQ_ENTRY_FORMAT_INLINE_32		1
1649 #define MCX_CQ_ENTRY_FORMAT_INLINE_64		2
1650 #define MCX_CQ_ENTRY_FORMAT_COMPRESSED		3
1651 
1652 #define MCX_CQ_ENTRY_OPCODE_REQ			0
1653 #define MCX_CQ_ENTRY_OPCODE_SEND		2
1654 #define MCX_CQ_ENTRY_OPCODE_REQ_ERR		13
1655 #define MCX_CQ_ENTRY_OPCODE_SEND_ERR		14
1656 #define MCX_CQ_ENTRY_OPCODE_INVALID		15
1657 
1658 } __packed __aligned(4);
1659 
1660 CTASSERT(sizeof(struct mcx_cq_entry) == 64);
1661 
1662 struct mcx_cq_doorbell {
1663 	uint32_t		 db_update_ci;
1664 	uint32_t		 db_arm_ci;
1665 #define MCX_CQ_DOORBELL_ARM_CMD_SN_SHIFT	28
1666 #define MCX_CQ_DOORBELL_ARM_CMD			(1 << 24)
1667 #define MCX_CQ_DOORBELL_ARM_CI_MASK		(0xffffff)
1668 } __packed __aligned(8);
1669 
1670 struct mcx_wq_ctx {
1671 	uint8_t			 wq_type;
1672 #define MCX_WQ_CTX_TYPE_CYCLIC			(1 << 4)
1673 #define MCX_WQ_CTX_TYPE_SIGNATURE		(1 << 3)
1674 	uint8_t			 wq_reserved0[5];
1675 	uint16_t		 wq_lwm;
1676 	uint32_t		 wq_pd;
1677 	uint32_t		 wq_uar_page;
1678 	uint64_t		 wq_doorbell;
1679 	uint32_t		 wq_hw_counter;
1680 	uint32_t		 wq_sw_counter;
1681 	uint16_t		 wq_log_stride;
1682 	uint8_t			 wq_log_page_sz;
1683 	uint8_t			 wq_log_size;
1684 	uint8_t			 wq_reserved1[156];
1685 } __packed __aligned(4);
1686 
1687 CTASSERT(sizeof(struct mcx_wq_ctx) == 0xC0);
1688 
1689 struct mcx_sq_ctx {
1690 	uint32_t		sq_flags;
1691 #define MCX_SQ_CTX_RLKEY			(1U << 31)
1692 #define MCX_SQ_CTX_FRE_SHIFT			(1 << 29)
1693 #define MCX_SQ_CTX_FLUSH_IN_ERROR		(1 << 28)
1694 #define MCX_SQ_CTX_MIN_WQE_INLINE_SHIFT		24
1695 #define MCX_SQ_CTX_STATE_SHIFT			20
1696 #define MCX_SQ_CTX_STATE_MASK			(0xf << 20)
1697 #define MCX_SQ_CTX_STATE_RST			0
1698 #define MCX_SQ_CTX_STATE_RDY			1
1699 #define MCX_SQ_CTX_STATE_ERR			3
1700 	uint32_t		sq_user_index;
1701 	uint32_t		sq_cqn;
1702 	uint32_t		sq_reserved1[5];
1703 	uint32_t		sq_tis_lst_sz;
1704 #define MCX_SQ_CTX_TIS_LST_SZ_SHIFT		16
1705 	uint32_t		sq_reserved2[2];
1706 	uint32_t		sq_tis_num;
1707 	struct mcx_wq_ctx	sq_wq;
1708 } __packed __aligned(4);
1709 
1710 struct mcx_sq_entry_seg {
1711 	uint32_t		sqs_byte_count;
1712 	uint32_t		sqs_lkey;
1713 	uint64_t		sqs_addr;
1714 } __packed __aligned(4);
1715 
1716 struct mcx_sq_entry {
1717 	/* control segment */
1718 	uint32_t		sqe_opcode_index;
1719 #define MCX_SQE_WQE_INDEX_SHIFT			8
1720 #define MCX_SQE_WQE_OPCODE_NOP			0x00
1721 #define MCX_SQE_WQE_OPCODE_SEND			0x0a
1722 	uint32_t		sqe_ds_sq_num;
1723 #define MCX_SQE_SQ_NUM_SHIFT			8
1724 	uint32_t		sqe_signature;
1725 #define MCX_SQE_SIGNATURE_SHIFT			24
1726 #define MCX_SQE_SOLICITED_EVENT			0x02
1727 #define MCX_SQE_CE_CQE_ON_ERR			0x00
1728 #define MCX_SQE_CE_CQE_FIRST_ERR		0x04
1729 #define MCX_SQE_CE_CQE_ALWAYS			0x08
1730 #define MCX_SQE_CE_CQE_SOLICIT			0x0C
1731 #define MCX_SQE_FM_NO_FENCE			0x00
1732 #define MCX_SQE_FM_SMALL_FENCE			0x40
1733 	uint32_t		sqe_mkey;
1734 
1735 	/* ethernet segment */
1736 	uint32_t		sqe_reserved1;
1737 	uint32_t		sqe_mss_csum;
1738 #define MCX_SQE_L4_CSUM				(1 << 31)
1739 #define MCX_SQE_L3_CSUM				(1 << 30)
1740 	uint32_t		sqe_reserved2;
1741 	uint16_t		sqe_inline_header_size;
1742 	uint16_t		sqe_inline_headers[9];
1743 
1744 	/* data segment */
1745 	struct mcx_sq_entry_seg sqe_segs[1];
1746 } __packed __aligned(64);
1747 
1748 CTASSERT(sizeof(struct mcx_sq_entry) == 64);
1749 
1750 struct mcx_cmd_create_sq_in {
1751 	uint16_t		cmd_opcode;
1752 	uint8_t			cmd_reserved0[4];
1753 	uint16_t		cmd_op_mod;
1754 	uint8_t			cmd_reserved1[8];
1755 } __packed __aligned(4);
1756 
1757 struct mcx_cmd_create_sq_out {
1758 	uint8_t			cmd_status;
1759 	uint8_t			cmd_reserved0[3];
1760 	uint32_t		cmd_syndrome;
1761 	uint32_t		cmd_sqn;
1762 	uint8_t			cmd_reserved1[4];
1763 } __packed __aligned(4);
1764 
1765 struct mcx_cmd_modify_sq_in {
1766 	uint16_t		cmd_opcode;
1767 	uint8_t			cmd_reserved0[4];
1768 	uint16_t		cmd_op_mod;
1769 	uint32_t		cmd_sq_state;
1770 	uint8_t			cmd_reserved1[4];
1771 } __packed __aligned(4);
1772 
1773 struct mcx_cmd_modify_sq_mb_in {
1774 	uint32_t		cmd_modify_hi;
1775 	uint32_t		cmd_modify_lo;
1776 	uint8_t			cmd_reserved0[8];
1777 	struct mcx_sq_ctx	cmd_sq_ctx;
1778 } __packed __aligned(4);
1779 
1780 struct mcx_cmd_modify_sq_out {
1781 	uint8_t			cmd_status;
1782 	uint8_t			cmd_reserved0[3];
1783 	uint32_t		cmd_syndrome;
1784 	uint8_t			cmd_reserved1[8];
1785 } __packed __aligned(4);
1786 
1787 struct mcx_cmd_destroy_sq_in {
1788 	uint16_t		cmd_opcode;
1789 	uint8_t			cmd_reserved0[4];
1790 	uint16_t		cmd_op_mod;
1791 	uint32_t		cmd_sqn;
1792 	uint8_t			cmd_reserved1[4];
1793 } __packed __aligned(4);
1794 
1795 struct mcx_cmd_destroy_sq_out {
1796 	uint8_t			cmd_status;
1797 	uint8_t			cmd_reserved0[3];
1798 	uint32_t		cmd_syndrome;
1799 	uint8_t			cmd_reserved1[8];
1800 } __packed __aligned(4);
1801 
1802 
1803 struct mcx_rq_ctx {
1804 	uint32_t		rq_flags;
1805 #define MCX_RQ_CTX_RLKEY			(1U << 31)
1806 #define MCX_RQ_CTX_VLAN_STRIP_DIS		(1 << 28)
1807 #define MCX_RQ_CTX_MEM_RQ_TYPE_SHIFT		24
1808 #define MCX_RQ_CTX_STATE_SHIFT			20
1809 #define MCX_RQ_CTX_STATE_MASK			(0xf << 20)
1810 #define MCX_RQ_CTX_STATE_RST			0
1811 #define MCX_RQ_CTX_STATE_RDY			1
1812 #define MCX_RQ_CTX_STATE_ERR			3
1813 #define MCX_RQ_CTX_FLUSH_IN_ERROR		(1 << 18)
1814 	uint32_t		rq_user_index;
1815 	uint32_t		rq_cqn;
1816 	uint32_t		rq_reserved1;
1817 	uint32_t		rq_rmpn;
1818 	uint32_t		rq_reserved2[7];
1819 	struct mcx_wq_ctx	rq_wq;
1820 } __packed __aligned(4);
1821 
1822 struct mcx_rq_entry {
1823 	uint32_t		rqe_byte_count;
1824 	uint32_t		rqe_lkey;
1825 	uint64_t		rqe_addr;
1826 } __packed __aligned(16);
1827 
1828 struct mcx_cmd_create_rq_in {
1829 	uint16_t		cmd_opcode;
1830 	uint8_t			cmd_reserved0[4];
1831 	uint16_t		cmd_op_mod;
1832 	uint8_t			cmd_reserved1[8];
1833 } __packed __aligned(4);
1834 
1835 struct mcx_cmd_create_rq_out {
1836 	uint8_t			cmd_status;
1837 	uint8_t			cmd_reserved0[3];
1838 	uint32_t		cmd_syndrome;
1839 	uint32_t		cmd_rqn;
1840 	uint8_t			cmd_reserved1[4];
1841 } __packed __aligned(4);
1842 
1843 struct mcx_cmd_modify_rq_in {
1844 	uint16_t		cmd_opcode;
1845 	uint8_t			cmd_reserved0[4];
1846 	uint16_t		cmd_op_mod;
1847 	uint32_t		cmd_rq_state;
1848 	uint8_t			cmd_reserved1[4];
1849 } __packed __aligned(4);
1850 
1851 struct mcx_cmd_modify_rq_mb_in {
1852 	uint32_t		cmd_modify_hi;
1853 	uint32_t		cmd_modify_lo;
1854 	uint8_t			cmd_reserved0[8];
1855 	struct mcx_rq_ctx	cmd_rq_ctx;
1856 } __packed __aligned(4);
1857 
1858 struct mcx_cmd_modify_rq_out {
1859 	uint8_t			cmd_status;
1860 	uint8_t			cmd_reserved0[3];
1861 	uint32_t		cmd_syndrome;
1862 	uint8_t			cmd_reserved1[8];
1863 } __packed __aligned(4);
1864 
1865 struct mcx_cmd_destroy_rq_in {
1866 	uint16_t		cmd_opcode;
1867 	uint8_t			cmd_reserved0[4];
1868 	uint16_t		cmd_op_mod;
1869 	uint32_t		cmd_rqn;
1870 	uint8_t			cmd_reserved1[4];
1871 } __packed __aligned(4);
1872 
1873 struct mcx_cmd_destroy_rq_out {
1874 	uint8_t			cmd_status;
1875 	uint8_t			cmd_reserved0[3];
1876 	uint32_t		cmd_syndrome;
1877 	uint8_t			cmd_reserved1[8];
1878 } __packed __aligned(4);
1879 
1880 struct mcx_cmd_create_flow_table_in {
1881 	uint16_t		cmd_opcode;
1882 	uint8_t			cmd_reserved0[4];
1883 	uint16_t		cmd_op_mod;
1884 	uint8_t			cmd_reserved1[8];
1885 } __packed __aligned(4);
1886 
1887 struct mcx_flow_table_ctx {
1888 	uint8_t			ft_miss_action;
1889 	uint8_t			ft_level;
1890 	uint8_t			ft_reserved0;
1891 	uint8_t			ft_log_size;
1892 	uint32_t		ft_table_miss_id;
1893 	uint8_t			ft_reserved1[28];
1894 } __packed __aligned(4);
1895 
1896 struct mcx_cmd_create_flow_table_mb_in {
1897 	uint8_t			cmd_table_type;
1898 	uint8_t			cmd_reserved0[7];
1899 	struct mcx_flow_table_ctx cmd_ctx;
1900 } __packed __aligned(4);
1901 
1902 struct mcx_cmd_create_flow_table_out {
1903 	uint8_t			cmd_status;
1904 	uint8_t			cmd_reserved0[3];
1905 	uint32_t		cmd_syndrome;
1906 	uint32_t		cmd_table_id;
1907 	uint8_t			cmd_reserved1[4];
1908 } __packed __aligned(4);
1909 
1910 struct mcx_cmd_destroy_flow_table_in {
1911 	uint16_t		cmd_opcode;
1912 	uint8_t			cmd_reserved0[4];
1913 	uint16_t		cmd_op_mod;
1914 	uint8_t			cmd_reserved1[8];
1915 } __packed __aligned(4);
1916 
1917 struct mcx_cmd_destroy_flow_table_mb_in {
1918 	uint8_t			cmd_table_type;
1919 	uint8_t			cmd_reserved0[3];
1920 	uint32_t		cmd_table_id;
1921 	uint8_t			cmd_reserved1[40];
1922 } __packed __aligned(4);
1923 
1924 struct mcx_cmd_destroy_flow_table_out {
1925 	uint8_t			cmd_status;
1926 	uint8_t			cmd_reserved0[3];
1927 	uint32_t		cmd_syndrome;
1928 	uint8_t			cmd_reserved1[8];
1929 } __packed __aligned(4);
1930 
1931 struct mcx_cmd_set_flow_table_root_in {
1932 	uint16_t		cmd_opcode;
1933 	uint8_t			cmd_reserved0[4];
1934 	uint16_t		cmd_op_mod;
1935 	uint8_t			cmd_reserved1[8];
1936 } __packed __aligned(4);
1937 
1938 struct mcx_cmd_set_flow_table_root_mb_in {
1939 	uint8_t			cmd_table_type;
1940 	uint8_t			cmd_reserved0[3];
1941 	uint32_t		cmd_table_id;
1942 	uint8_t			cmd_reserved1[56];
1943 } __packed __aligned(4);
1944 
1945 struct mcx_cmd_set_flow_table_root_out {
1946 	uint8_t			cmd_status;
1947 	uint8_t			cmd_reserved0[3];
1948 	uint32_t		cmd_syndrome;
1949 	uint8_t			cmd_reserved1[8];
1950 } __packed __aligned(4);
1951 
1952 struct mcx_flow_match {
1953 	/* outer headers */
1954 	uint8_t			mc_src_mac[6];
1955 	uint16_t		mc_ethertype;
1956 	uint8_t			mc_dest_mac[6];
1957 	uint16_t		mc_first_vlan;
1958 	uint8_t			mc_ip_proto;
1959 	uint8_t			mc_ip_dscp_ecn;
1960 	uint8_t			mc_vlan_flags;
1961 #define MCX_FLOW_MATCH_IP_FRAG	(1 << 5)
1962 	uint8_t			mc_tcp_flags;
1963 	uint16_t		mc_tcp_sport;
1964 	uint16_t		mc_tcp_dport;
1965 	uint32_t		mc_reserved0;
1966 	uint16_t		mc_udp_sport;
1967 	uint16_t		mc_udp_dport;
1968 	uint8_t			mc_src_ip[16];
1969 	uint8_t			mc_dest_ip[16];
1970 
1971 	/* misc parameters */
1972 	uint8_t			mc_reserved1[8];
1973 	uint16_t		mc_second_vlan;
1974 	uint8_t			mc_reserved2[2];
1975 	uint8_t			mc_second_vlan_flags;
1976 	uint8_t			mc_reserved3[15];
1977 	uint32_t		mc_outer_ipv6_flow_label;
1978 	uint8_t			mc_reserved4[32];
1979 
1980 	uint8_t			mc_reserved[384];
1981 } __packed __aligned(4);
1982 
1983 CTASSERT(sizeof(struct mcx_flow_match) == 512);
1984 
1985 struct mcx_cmd_create_flow_group_in {
1986 	uint16_t		cmd_opcode;
1987 	uint8_t			cmd_reserved0[4];
1988 	uint16_t		cmd_op_mod;
1989 	uint8_t			cmd_reserved1[8];
1990 } __packed __aligned(4);
1991 
1992 struct mcx_cmd_create_flow_group_mb_in {
1993 	uint8_t			cmd_table_type;
1994 	uint8_t			cmd_reserved0[3];
1995 	uint32_t		cmd_table_id;
1996 	uint8_t			cmd_reserved1[4];
1997 	uint32_t		cmd_start_flow_index;
1998 	uint8_t			cmd_reserved2[4];
1999 	uint32_t		cmd_end_flow_index;
2000 	uint8_t			cmd_reserved3[23];
2001 	uint8_t			cmd_match_criteria_enable;
2002 #define MCX_CREATE_FLOW_GROUP_CRIT_OUTER	(1 << 0)
2003 #define MCX_CREATE_FLOW_GROUP_CRIT_MISC		(1 << 1)
2004 #define MCX_CREATE_FLOW_GROUP_CRIT_INNER	(1 << 2)
2005 	struct mcx_flow_match	cmd_match_criteria;
2006 	uint8_t			cmd_reserved4[448];
2007 } __packed __aligned(4);
2008 
2009 struct mcx_cmd_create_flow_group_out {
2010 	uint8_t			cmd_status;
2011 	uint8_t			cmd_reserved0[3];
2012 	uint32_t		cmd_syndrome;
2013 	uint32_t		cmd_group_id;
2014 	uint8_t			cmd_reserved1[4];
2015 } __packed __aligned(4);
2016 
2017 struct mcx_flow_ctx {
2018 	uint8_t			fc_reserved0[4];
2019 	uint32_t		fc_group_id;
2020 	uint32_t		fc_flow_tag;
2021 	uint32_t		fc_action;
2022 #define MCX_FLOW_CONTEXT_ACTION_ALLOW		(1 << 0)
2023 #define MCX_FLOW_CONTEXT_ACTION_DROP		(1 << 1)
2024 #define MCX_FLOW_CONTEXT_ACTION_FORWARD		(1 << 2)
2025 #define MCX_FLOW_CONTEXT_ACTION_COUNT		(1 << 3)
2026 	uint32_t		fc_dest_list_size;
2027 	uint32_t		fc_counter_list_size;
2028 	uint8_t			fc_reserved1[40];
2029 	struct mcx_flow_match	fc_match_value;
2030 	uint8_t			fc_reserved2[192];
2031 } __packed __aligned(4);
2032 
2033 #define MCX_FLOW_CONTEXT_DEST_TYPE_TABLE	(1 << 24)
2034 #define MCX_FLOW_CONTEXT_DEST_TYPE_TIR		(2 << 24)
2035 
2036 struct mcx_cmd_destroy_flow_group_in {
2037 	uint16_t		cmd_opcode;
2038 	uint8_t			cmd_reserved0[4];
2039 	uint16_t		cmd_op_mod;
2040 	uint8_t			cmd_reserved1[8];
2041 } __packed __aligned(4);
2042 
2043 struct mcx_cmd_destroy_flow_group_mb_in {
2044 	uint8_t			cmd_table_type;
2045 	uint8_t			cmd_reserved0[3];
2046 	uint32_t		cmd_table_id;
2047 	uint32_t		cmd_group_id;
2048 	uint8_t			cmd_reserved1[36];
2049 } __packed __aligned(4);
2050 
2051 struct mcx_cmd_destroy_flow_group_out {
2052 	uint8_t			cmd_status;
2053 	uint8_t			cmd_reserved0[3];
2054 	uint32_t		cmd_syndrome;
2055 	uint8_t			cmd_reserved1[8];
2056 } __packed __aligned(4);
2057 
2058 struct mcx_cmd_set_flow_table_entry_in {
2059 	uint16_t		cmd_opcode;
2060 	uint8_t			cmd_reserved0[4];
2061 	uint16_t		cmd_op_mod;
2062 	uint8_t			cmd_reserved1[8];
2063 } __packed __aligned(4);
2064 
2065 struct mcx_cmd_set_flow_table_entry_mb_in {
2066 	uint8_t			cmd_table_type;
2067 	uint8_t			cmd_reserved0[3];
2068 	uint32_t		cmd_table_id;
2069 	uint32_t		cmd_modify_enable_mask;
2070 	uint8_t			cmd_reserved1[4];
2071 	uint32_t		cmd_flow_index;
2072 	uint8_t			cmd_reserved2[28];
2073 	struct mcx_flow_ctx	cmd_flow_ctx;
2074 } __packed __aligned(4);
2075 
2076 struct mcx_cmd_set_flow_table_entry_out {
2077 	uint8_t			cmd_status;
2078 	uint8_t			cmd_reserved0[3];
2079 	uint32_t		cmd_syndrome;
2080 	uint8_t			cmd_reserved1[8];
2081 } __packed __aligned(4);
2082 
2083 struct mcx_cmd_query_flow_table_entry_in {
2084 	uint16_t		cmd_opcode;
2085 	uint8_t			cmd_reserved0[4];
2086 	uint16_t		cmd_op_mod;
2087 	uint8_t			cmd_reserved1[8];
2088 } __packed __aligned(4);
2089 
2090 struct mcx_cmd_query_flow_table_entry_mb_in {
2091 	uint8_t			cmd_table_type;
2092 	uint8_t			cmd_reserved0[3];
2093 	uint32_t		cmd_table_id;
2094 	uint8_t			cmd_reserved1[8];
2095 	uint32_t		cmd_flow_index;
2096 	uint8_t			cmd_reserved2[28];
2097 } __packed __aligned(4);
2098 
2099 struct mcx_cmd_query_flow_table_entry_out {
2100 	uint8_t			cmd_status;
2101 	uint8_t			cmd_reserved0[3];
2102 	uint32_t		cmd_syndrome;
2103 	uint8_t			cmd_reserved1[8];
2104 } __packed __aligned(4);
2105 
2106 struct mcx_cmd_query_flow_table_entry_mb_out {
2107 	uint8_t			cmd_reserved0[48];
2108 	struct mcx_flow_ctx	cmd_flow_ctx;
2109 } __packed __aligned(4);
2110 
2111 struct mcx_cmd_delete_flow_table_entry_in {
2112 	uint16_t		cmd_opcode;
2113 	uint8_t			cmd_reserved0[4];
2114 	uint16_t		cmd_op_mod;
2115 	uint8_t			cmd_reserved1[8];
2116 } __packed __aligned(4);
2117 
2118 struct mcx_cmd_delete_flow_table_entry_mb_in {
2119 	uint8_t			cmd_table_type;
2120 	uint8_t			cmd_reserved0[3];
2121 	uint32_t		cmd_table_id;
2122 	uint8_t			cmd_reserved1[8];
2123 	uint32_t		cmd_flow_index;
2124 	uint8_t			cmd_reserved2[28];
2125 } __packed __aligned(4);
2126 
2127 struct mcx_cmd_delete_flow_table_entry_out {
2128 	uint8_t			cmd_status;
2129 	uint8_t			cmd_reserved0[3];
2130 	uint32_t		cmd_syndrome;
2131 	uint8_t			cmd_reserved1[8];
2132 } __packed __aligned(4);
2133 
2134 struct mcx_cmd_query_flow_group_in {
2135 	uint16_t		cmd_opcode;
2136 	uint8_t			cmd_reserved0[4];
2137 	uint16_t		cmd_op_mod;
2138 	uint8_t			cmd_reserved1[8];
2139 } __packed __aligned(4);
2140 
2141 struct mcx_cmd_query_flow_group_mb_in {
2142 	uint8_t			cmd_table_type;
2143 	uint8_t			cmd_reserved0[3];
2144 	uint32_t		cmd_table_id;
2145 	uint32_t		cmd_group_id;
2146 	uint8_t			cmd_reserved1[36];
2147 } __packed __aligned(4);
2148 
2149 struct mcx_cmd_query_flow_group_out {
2150 	uint8_t			cmd_status;
2151 	uint8_t			cmd_reserved0[3];
2152 	uint32_t		cmd_syndrome;
2153 	uint8_t			cmd_reserved1[8];
2154 } __packed __aligned(4);
2155 
2156 struct mcx_cmd_query_flow_group_mb_out {
2157 	uint8_t			cmd_reserved0[12];
2158 	uint32_t		cmd_start_flow_index;
2159 	uint8_t			cmd_reserved1[4];
2160 	uint32_t		cmd_end_flow_index;
2161 	uint8_t			cmd_reserved2[20];
2162 	uint32_t		cmd_match_criteria_enable;
2163 	uint8_t			cmd_match_criteria[512];
2164 	uint8_t			cmd_reserved4[448];
2165 } __packed __aligned(4);
2166 
2167 struct mcx_cmd_query_flow_table_in {
2168 	uint16_t		cmd_opcode;
2169 	uint8_t			cmd_reserved0[4];
2170 	uint16_t		cmd_op_mod;
2171 	uint8_t			cmd_reserved1[8];
2172 } __packed __aligned(4);
2173 
2174 struct mcx_cmd_query_flow_table_mb_in {
2175 	uint8_t			cmd_table_type;
2176 	uint8_t			cmd_reserved0[3];
2177 	uint32_t		cmd_table_id;
2178 	uint8_t			cmd_reserved1[40];
2179 } __packed __aligned(4);
2180 
2181 struct mcx_cmd_query_flow_table_out {
2182 	uint8_t			cmd_status;
2183 	uint8_t			cmd_reserved0[3];
2184 	uint32_t		cmd_syndrome;
2185 	uint8_t			cmd_reserved1[8];
2186 } __packed __aligned(4);
2187 
2188 struct mcx_cmd_query_flow_table_mb_out {
2189 	uint8_t			cmd_reserved0[4];
2190 	struct mcx_flow_table_ctx cmd_ctx;
2191 } __packed __aligned(4);
2192 
2193 struct mcx_cmd_alloc_flow_counter_in {
2194 	uint16_t		cmd_opcode;
2195 	uint8_t			cmd_reserved0[4];
2196 	uint16_t		cmd_op_mod;
2197 	uint8_t			cmd_reserved1[8];
2198 } __packed __aligned(4);
2199 
2200 struct mcx_cmd_query_rq_in {
2201 	uint16_t		cmd_opcode;
2202 	uint8_t			cmd_reserved0[4];
2203 	uint16_t		cmd_op_mod;
2204 	uint32_t		cmd_rqn;
2205 	uint8_t			cmd_reserved1[4];
2206 } __packed __aligned(4);
2207 
2208 struct mcx_cmd_query_rq_out {
2209 	uint8_t			cmd_status;
2210 	uint8_t			cmd_reserved0[3];
2211 	uint32_t		cmd_syndrome;
2212 	uint8_t			cmd_reserved1[8];
2213 } __packed __aligned(4);
2214 
2215 struct mcx_cmd_query_rq_mb_out {
2216 	uint8_t			cmd_reserved0[16];
2217 	struct mcx_rq_ctx	cmd_ctx;
2218 };
2219 
2220 struct mcx_cmd_query_sq_in {
2221 	uint16_t		cmd_opcode;
2222 	uint8_t			cmd_reserved0[4];
2223 	uint16_t		cmd_op_mod;
2224 	uint32_t		cmd_sqn;
2225 	uint8_t			cmd_reserved1[4];
2226 } __packed __aligned(4);
2227 
2228 struct mcx_cmd_query_sq_out {
2229 	uint8_t			cmd_status;
2230 	uint8_t			cmd_reserved0[3];
2231 	uint32_t		cmd_syndrome;
2232 	uint8_t			cmd_reserved1[8];
2233 } __packed __aligned(4);
2234 
2235 struct mcx_cmd_query_sq_mb_out {
2236 	uint8_t			cmd_reserved0[16];
2237 	struct mcx_sq_ctx	cmd_ctx;
2238 };
2239 
2240 struct mcx_cmd_alloc_flow_counter_out {
2241 	uint8_t			cmd_status;
2242 	uint8_t			cmd_reserved0[3];
2243 	uint32_t		cmd_syndrome;
2244 	uint8_t			cmd_reserved1[2];
2245 	uint16_t		cmd_flow_counter_id;
2246 	uint8_t			cmd_reserved2[4];
2247 } __packed __aligned(4);
2248 
2249 struct mcx_wq_doorbell {
2250 	uint32_t		 db_recv_counter;
2251 	uint32_t		 db_send_counter;
2252 } __packed __aligned(8);
2253 
2254 struct mcx_dmamem {
2255 	bus_dmamap_t		 mxm_map;
2256 	bus_dma_segment_t	 mxm_seg;
2257 	int			 mxm_nsegs;
2258 	size_t			 mxm_size;
2259 	void			*mxm_kva;
2260 };
2261 #define MCX_DMA_MAP(_mxm)	((_mxm)->mxm_map)
2262 #define MCX_DMA_DVA(_mxm)	((_mxm)->mxm_map->dm_segs[0].ds_addr)
2263 #define MCX_DMA_KVA(_mxm)	((void *)(_mxm)->mxm_kva)
2264 #define MCX_DMA_OFF(_mxm, _off)	((void *)((char *)(_mxm)->mxm_kva + (_off)))
2265 #define MCX_DMA_LEN(_mxm)	((_mxm)->mxm_size)
2266 
2267 struct mcx_hwmem {
2268 	bus_dmamap_t		 mhm_map;
2269 	bus_dma_segment_t	*mhm_segs;
2270 	unsigned int		 mhm_seg_count;
2271 	unsigned int		 mhm_npages;
2272 };
2273 
2274 struct mcx_slot {
2275 	bus_dmamap_t		 ms_map;
2276 	struct mbuf		*ms_m;
2277 };
2278 
2279 struct mcx_eq {
2280 	int			 eq_n;
2281 	uint32_t		 eq_cons;
2282 	struct mcx_dmamem	 eq_mem;
2283 };
2284 
2285 struct mcx_cq {
2286 	int			 cq_n;
2287 	struct mcx_dmamem	 cq_mem;
2288 	bus_addr_t		 cq_doorbell;
2289 	uint32_t		 cq_cons;
2290 	uint32_t		 cq_count;
2291 };
2292 
2293 struct mcx_calibration {
2294 	uint64_t		 c_timestamp;	/* previous mcx chip time */
2295 	uint64_t		 c_uptime;	/* previous kernel nanouptime */
2296 	uint64_t		 c_tbase;	/* mcx chip time */
2297 	uint64_t		 c_ubase;	/* kernel nanouptime */
2298 	uint64_t		 c_ratio;
2299 };
2300 
2301 #define MCX_CALIBRATE_FIRST    2
2302 #define MCX_CALIBRATE_NORMAL   32
2303 
2304 struct mcx_rxring {
2305 	u_int			 rxr_total;
2306 	u_int			 rxr_inuse;
2307 };
2308 
2309 MBUFQ_HEAD(mcx_mbufq);
2310 
2311 struct mcx_rx {
2312 	struct mcx_softc	*rx_softc;
2313 
2314 	int			 rx_rqn;
2315 	struct mcx_dmamem	 rx_rq_mem;
2316 	struct mcx_slot		*rx_slots;
2317 	bus_addr_t		 rx_doorbell;
2318 
2319 	uint32_t		 rx_prod;
2320 	callout_t		 rx_refill;
2321 	struct mcx_rxring	 rx_rxr;
2322 } __aligned(64);
2323 
2324 struct mcx_tx {
2325 	struct mcx_softc	*tx_softc;
2326 	kmutex_t		 tx_lock;
2327 	pcq_t			*tx_pcq;
2328 	void			*tx_softint;
2329 
2330 	int			 tx_uar;
2331 	int			 tx_sqn;
2332 	struct mcx_dmamem	 tx_sq_mem;
2333 	struct mcx_slot		*tx_slots;
2334 	bus_addr_t		 tx_doorbell;
2335 	int			 tx_bf_offset;
2336 
2337 	uint32_t		 tx_cons;
2338 	uint32_t		 tx_prod;
2339 } __aligned(64);
2340 
2341 struct mcx_queues {
2342 	void			*q_ihc;
2343 	struct mcx_softc	*q_sc;
2344 	int			 q_uar;
2345 	int			 q_index;
2346 	struct mcx_rx		 q_rx;
2347 	struct mcx_tx		 q_tx;
2348 	struct mcx_cq		 q_cq;
2349 	struct mcx_eq		 q_eq;
2350 #if NKSTAT > 0
2351 	struct kstat		*q_kstat;
2352 #endif
2353 };
2354 
2355 struct mcx_flow_group {
2356 	int			 g_id;
2357 	int			 g_table;
2358 	int			 g_start;
2359 	int			 g_size;
2360 };
2361 
2362 #define MCX_FLOW_GROUP_PROMISC		0
2363 #define MCX_FLOW_GROUP_ALLMULTI		1
2364 #define MCX_FLOW_GROUP_MAC		2
2365 #define MCX_FLOW_GROUP_RSS_L4		3
2366 #define MCX_FLOW_GROUP_RSS_L3		4
2367 #define MCX_FLOW_GROUP_RSS_NONE		5
2368 #define	MCX_NUM_FLOW_GROUPS		6
2369 
2370 #define MCX_HASH_SEL_L3		MCX_TIR_CTX_HASH_SEL_SRC_IP | \
2371 				MCX_TIR_CTX_HASH_SEL_DST_IP
2372 #define MCX_HASH_SEL_L4		MCX_HASH_SEL_L3 | MCX_TIR_CTX_HASH_SEL_SPORT | \
2373 				MCX_TIR_CTX_HASH_SEL_DPORT
2374 
2375 #define MCX_RSS_HASH_SEL_V4_TCP MCX_HASH_SEL_L4 | MCX_TIR_CTX_HASH_SEL_TCP  |\
2376 				MCX_TIR_CTX_HASH_SEL_IPV4
2377 #define MCX_RSS_HASH_SEL_V6_TCP	MCX_HASH_SEL_L4 | MCX_TIR_CTX_HASH_SEL_TCP | \
2378 				MCX_TIR_CTX_HASH_SEL_IPV6
2379 #define MCX_RSS_HASH_SEL_V4_UDP	MCX_HASH_SEL_L4 | MCX_TIR_CTX_HASH_SEL_UDP | \
2380 				MCX_TIR_CTX_HASH_SEL_IPV4
2381 #define MCX_RSS_HASH_SEL_V6_UDP	MCX_HASH_SEL_L4 | MCX_TIR_CTX_HASH_SEL_UDP | \
2382 				MCX_TIR_CTX_HASH_SEL_IPV6
2383 #define MCX_RSS_HASH_SEL_V4	MCX_HASH_SEL_L3 | MCX_TIR_CTX_HASH_SEL_IPV4
2384 #define MCX_RSS_HASH_SEL_V6	MCX_HASH_SEL_L3 | MCX_TIR_CTX_HASH_SEL_IPV6
2385 
2386 /*
2387  * There are a few different pieces involved in configuring RSS.
2388  * A Receive Queue Table (RQT) is the indirection table that maps packets to
2389  * different rx queues based on a hash value.  We only create one, because
2390  * we want to scatter any traffic we can apply RSS to across all our rx
2391  * queues.  Anything else will only be delivered to the first rx queue,
2392  * which doesn't require an RQT.
2393  *
2394  * A Transport Interface Receive (TIR) delivers packets to either a single rx
2395  * queue or an RQT, and in the latter case, specifies the set of fields
2396  * hashed, the hash function, and the hash key.  We need one of these for each
2397  * type of RSS traffic - v4 TCP, v6 TCP, v4 UDP, v6 UDP, other v4, other v6,
2398  * and one for non-RSS traffic.
2399  *
2400  * Flow tables hold flow table entries in sequence.  The first entry that
2401  * matches a packet is applied, sending the packet to either another flow
2402  * table or a TIR.  We use one flow table to select packets based on
2403  * destination MAC address, and a second to apply RSS.  The entries in the
2404  * first table send matching packets to the second, and the entries in the
2405  * RSS table send packets to RSS TIRs if possible, or the non-RSS TIR.
2406  *
2407  * The flow table entry that delivers packets to an RSS TIR must include match
2408  * criteria that ensure packets delivered to the TIR include all the fields
2409  * that the TIR hashes on - so for a v4 TCP TIR, the flow table entry must
2410  * only accept v4 TCP packets.  Accordingly, we need flow table entries for
2411  * each TIR.
2412  *
2413  * All of this is a lot more flexible than we need, and we can describe most
2414  * of the stuff we need with a simple array.
2415  *
2416  * An RSS config creates a TIR with hashing enabled on a set of fields,
2417  * pointing to either the first rx queue or the RQT containing all the rx
2418  * queues, and a flow table entry that matches on an ether type and
2419  * optionally an ip proto, that delivers packets to the TIR.
2420  */
2421 static struct mcx_rss_rule {
2422 	int			hash_sel;
2423 	int			flow_group;
2424 	int			ethertype;
2425 	int			ip_proto;
2426 } mcx_rss_config[] = {
2427 	/* udp and tcp for v4/v6 */
2428 	{ MCX_RSS_HASH_SEL_V4_TCP, MCX_FLOW_GROUP_RSS_L4,
2429 	  ETHERTYPE_IP, IPPROTO_TCP },
2430 	{ MCX_RSS_HASH_SEL_V6_TCP, MCX_FLOW_GROUP_RSS_L4,
2431 	  ETHERTYPE_IPV6, IPPROTO_TCP },
2432 	{ MCX_RSS_HASH_SEL_V4_UDP, MCX_FLOW_GROUP_RSS_L4,
2433 	  ETHERTYPE_IP, IPPROTO_UDP },
2434 	{ MCX_RSS_HASH_SEL_V6_UDP, MCX_FLOW_GROUP_RSS_L4,
2435 	  ETHERTYPE_IPV6, IPPROTO_UDP },
2436 
2437 	/* other v4/v6 */
2438 	{ MCX_RSS_HASH_SEL_V4, MCX_FLOW_GROUP_RSS_L3,
2439 	  ETHERTYPE_IP, 0 },
2440 	{ MCX_RSS_HASH_SEL_V6, MCX_FLOW_GROUP_RSS_L3,
2441 	  ETHERTYPE_IPV6, 0 },
2442 
2443 	/* non v4/v6 */
2444 	{ 0, MCX_FLOW_GROUP_RSS_NONE, 0, 0 }
2445 };
2446 
2447 struct mcx_softc {
2448 	device_t		 sc_dev;
2449 	struct ethercom		 sc_ec;
2450 	struct ifmedia		 sc_media;
2451 	uint64_t		 sc_media_status;
2452 	uint64_t		 sc_media_active;
2453 	kmutex_t		 sc_media_mutex;
2454 
2455 	pci_chipset_tag_t	 sc_pc;
2456 	pci_intr_handle_t	*sc_intrs;
2457 	void			*sc_ihc;
2458 	pcitag_t		 sc_tag;
2459 
2460 	bus_dma_tag_t		 sc_dmat;
2461 	bus_space_tag_t		 sc_memt;
2462 	bus_space_handle_t	 sc_memh;
2463 	bus_size_t		 sc_mems;
2464 
2465 	struct mcx_dmamem	 sc_cmdq_mem;
2466 	unsigned int		 sc_cmdq_mask;
2467 	unsigned int		 sc_cmdq_size;
2468 
2469 	unsigned int		 sc_cmdq_token;
2470 
2471 	struct mcx_hwmem	 sc_boot_pages;
2472 	struct mcx_hwmem	 sc_init_pages;
2473 	struct mcx_hwmem	 sc_regular_pages;
2474 
2475 	int			 sc_uar;
2476 	int			 sc_pd;
2477 	int			 sc_tdomain;
2478 	uint32_t		 sc_lkey;
2479 	int			 sc_tis;
2480 	int			 sc_tir[__arraycount(mcx_rss_config)];
2481 	int			 sc_rqt;
2482 
2483 	struct mcx_dmamem	 sc_doorbell_mem;
2484 
2485 	struct mcx_eq		 sc_admin_eq;
2486 	struct mcx_eq		 sc_queue_eq;
2487 
2488 	int			 sc_hardmtu;
2489 	int			 sc_rxbufsz;
2490 
2491 	int			 sc_bf_size;
2492 	int			 sc_max_rqt_size;
2493 
2494 	struct workqueue	*sc_workq;
2495 	struct work		 sc_port_change;
2496 
2497 	int			 sc_mac_flow_table_id;
2498 	int			 sc_rss_flow_table_id;
2499 	struct mcx_flow_group	 sc_flow_group[MCX_NUM_FLOW_GROUPS];
2500 	int			 sc_promisc_flow_enabled;
2501 	int			 sc_allmulti_flow_enabled;
2502 	int			 sc_mcast_flow_base;
2503 	int			 sc_extra_mcast;
2504 	uint8_t			 sc_mcast_flows[MCX_NUM_MCAST_FLOWS][ETHER_ADDR_LEN];
2505 
2506 	struct mcx_calibration	 sc_calibration[2];
2507 	unsigned int		 sc_calibration_gen;
2508 	callout_t		 sc_calibrate;
2509 	uint32_t		 sc_mhz;
2510 	uint32_t		 sc_khz;
2511 
2512 	struct mcx_queues	*sc_queues;
2513 	unsigned int		 sc_nqueues;
2514 
2515 	int			 sc_mcam_reg;
2516 
2517 #if NKSTAT > 0
2518 	struct kstat		*sc_kstat_ieee8023;
2519 	struct kstat		*sc_kstat_rfc2863;
2520 	struct kstat		*sc_kstat_rfc2819;
2521 	struct kstat		*sc_kstat_rfc3635;
2522 	unsigned int		 sc_kstat_mtmp_count;
2523 	struct kstat		**sc_kstat_mtmp;
2524 #endif
2525 
2526 	struct timecounter	 sc_timecounter;
2527 };
2528 #define DEVNAME(_sc) device_xname((_sc)->sc_dev)
2529 
2530 static int	mcx_match(device_t, cfdata_t, void *);
2531 static void	mcx_attach(device_t, device_t, void *);
2532 
2533 static void *	mcx_establish_intr(struct mcx_softc *, int, kcpuset_t *,
2534 		    int (*)(void *), void *, const char *);
2535 
2536 static void	mcx_rxr_init(struct mcx_rxring *, u_int, u_int);
2537 static u_int	mcx_rxr_get(struct mcx_rxring *, u_int);
2538 static void	mcx_rxr_put(struct mcx_rxring *, u_int);
2539 static u_int	mcx_rxr_inuse(struct mcx_rxring *);
2540 
2541 #if NKSTAT > 0
2542 static void	mcx_kstat_attach(struct mcx_softc *);
2543 #endif
2544 
2545 static void	mcx_timecounter_attach(struct mcx_softc *);
2546 
2547 static int	mcx_version(struct mcx_softc *);
2548 static int	mcx_init_wait(struct mcx_softc *);
2549 static int	mcx_enable_hca(struct mcx_softc *);
2550 static int	mcx_teardown_hca(struct mcx_softc *, uint16_t);
2551 static int	mcx_access_hca_reg(struct mcx_softc *, uint16_t, int, void *,
2552 		    int);
2553 static int	mcx_issi(struct mcx_softc *);
2554 static int	mcx_pages(struct mcx_softc *, struct mcx_hwmem *, uint16_t);
2555 static int	mcx_hca_max_caps(struct mcx_softc *);
2556 static int	mcx_hca_set_caps(struct mcx_softc *);
2557 static int	mcx_init_hca(struct mcx_softc *);
2558 static int	mcx_set_driver_version(struct mcx_softc *);
2559 static int	mcx_iff(struct mcx_softc *);
2560 static int	mcx_alloc_uar(struct mcx_softc *, int *);
2561 static int	mcx_alloc_pd(struct mcx_softc *);
2562 static int	mcx_alloc_tdomain(struct mcx_softc *);
2563 static int	mcx_create_eq(struct mcx_softc *, struct mcx_eq *, int,
2564 		    uint64_t, int);
2565 static int	mcx_query_nic_vport_context(struct mcx_softc *, uint8_t *);
2566 static int	mcx_query_special_contexts(struct mcx_softc *);
2567 static int	mcx_set_port_mtu(struct mcx_softc *, int);
2568 static int	mcx_create_cq(struct mcx_softc *, struct mcx_cq *, int, int,
2569 		    int);
2570 static int	mcx_destroy_cq(struct mcx_softc *, struct mcx_cq *);
2571 static int	mcx_create_sq(struct mcx_softc *, struct mcx_tx *, int, int,
2572 		    int);
2573 static int	mcx_destroy_sq(struct mcx_softc *, struct mcx_tx *);
2574 static int	mcx_ready_sq(struct mcx_softc *, struct mcx_tx *);
2575 static int	mcx_create_rq(struct mcx_softc *, struct mcx_rx *, int, int);
2576 static int	mcx_destroy_rq(struct mcx_softc *, struct mcx_rx *);
2577 static int	mcx_ready_rq(struct mcx_softc *, struct mcx_rx *);
2578 static int	mcx_create_tir_direct(struct mcx_softc *, struct mcx_rx *,
2579 		    int *);
2580 static int	mcx_create_tir_indirect(struct mcx_softc *, int, uint32_t,
2581 		    int *);
2582 static int	mcx_destroy_tir(struct mcx_softc *, int);
2583 static int	mcx_create_tis(struct mcx_softc *, int *);
2584 static int	mcx_destroy_tis(struct mcx_softc *, int);
2585 static int	mcx_create_rqt(struct mcx_softc *, int, int *, int *);
2586 static int	mcx_destroy_rqt(struct mcx_softc *, int);
2587 static int	mcx_create_flow_table(struct mcx_softc *, int, int, int *);
2588 static int	mcx_set_flow_table_root(struct mcx_softc *, int);
2589 static int	mcx_destroy_flow_table(struct mcx_softc *, int);
2590 static int	mcx_create_flow_group(struct mcx_softc *, int, int, int,
2591 		    int, int, struct mcx_flow_match *);
2592 static int	mcx_destroy_flow_group(struct mcx_softc *, int);
2593 static int	mcx_set_flow_table_entry_mac(struct mcx_softc *, int, int,
2594 		    const uint8_t *, uint32_t);
2595 static int	mcx_set_flow_table_entry_proto(struct mcx_softc *, int, int,
2596 		    int, int, uint32_t);
2597 static int	mcx_delete_flow_table_entry(struct mcx_softc *, int, int);
2598 
2599 #if NKSTAT > 0
2600 static int	mcx_query_rq(struct mcx_softc *, struct mcx_rx *, struct mcx_rq_ctx *);
2601 static int	mcx_query_sq(struct mcx_softc *, struct mcx_tx *, struct mcx_sq_ctx *);
2602 static int	mcx_query_cq(struct mcx_softc *, struct mcx_cq *, struct mcx_cq_ctx *);
2603 static int	mcx_query_eq(struct mcx_softc *, struct mcx_eq *, struct mcx_eq_ctx *);
2604 #endif
2605 
2606 #if 0
2607 static int	mcx_dump_flow_table(struct mcx_softc *, int);
2608 static int	mcx_dump_flow_table_entry(struct mcx_softc *, int, int);
2609 static int	mcx_dump_flow_group(struct mcx_softc *, int);
2610 #endif
2611 
2612 
2613 /*
2614 static void	mcx_cmdq_dump(const struct mcx_cmdq_entry *);
2615 static void	mcx_cmdq_mbox_dump(struct mcx_dmamem *, int);
2616 */
2617 static void	mcx_refill(void *);
2618 static int	mcx_process_rx(struct mcx_softc *, struct mcx_rx *,
2619 		    struct mcx_cq_entry *, struct mcx_mbufq *,
2620 		    const struct mcx_calibration *);
2621 static int	mcx_process_txeof(struct mcx_softc *, struct mcx_tx *,
2622 		    struct mcx_cq_entry *);
2623 static void	mcx_process_cq(struct mcx_softc *, struct mcx_queues *,
2624 		    struct mcx_cq *);
2625 
2626 static void	mcx_arm_cq(struct mcx_softc *, struct mcx_cq *, int);
2627 static void	mcx_arm_eq(struct mcx_softc *, struct mcx_eq *, int);
2628 static int	mcx_admin_intr(void *);
2629 static int	mcx_cq_intr(void *);
2630 
2631 static int	mcx_init(struct ifnet *);
2632 static void	mcx_stop(struct ifnet *, int);
2633 static int	mcx_ioctl(struct ifnet *, u_long, void *);
2634 static void	mcx_start(struct ifnet *);
2635 static int	mcx_transmit(struct ifnet *, struct mbuf *);
2636 static void	mcx_deferred_transmit(void *);
2637 static void	mcx_media_add_types(struct mcx_softc *);
2638 static void	mcx_media_status(struct ifnet *, struct ifmediareq *);
2639 static int	mcx_media_change(struct ifnet *);
2640 #if 0
2641 static int	mcx_get_sffpage(struct ifnet *, struct if_sffpage *);
2642 #endif
2643 static void	mcx_port_change(struct work *, void *);
2644 
2645 static void	mcx_calibrate_first(struct mcx_softc *);
2646 static void	mcx_calibrate(void *);
2647 
2648 static inline uint32_t
2649 		mcx_rd(struct mcx_softc *, bus_size_t);
2650 static inline void
2651 		mcx_wr(struct mcx_softc *, bus_size_t, uint32_t);
2652 static inline void
2653 		mcx_bar(struct mcx_softc *, bus_size_t, bus_size_t, int);
2654 
2655 static uint64_t	mcx_timer(struct mcx_softc *);
2656 
2657 static int	mcx_dmamem_alloc(struct mcx_softc *, struct mcx_dmamem *,
2658 		    bus_size_t, u_int align);
2659 static void	mcx_dmamem_zero(struct mcx_dmamem *);
2660 static void	mcx_dmamem_free(struct mcx_softc *, struct mcx_dmamem *);
2661 
2662 static int	mcx_hwmem_alloc(struct mcx_softc *, struct mcx_hwmem *,
2663 		    unsigned int);
2664 static void	mcx_hwmem_free(struct mcx_softc *, struct mcx_hwmem *);
2665 
2666 CFATTACH_DECL_NEW(mcx, sizeof(struct mcx_softc), mcx_match, mcx_attach, NULL, NULL);
2667 
2668 static const struct {
2669 	pci_vendor_id_t		vendor;
2670 	pci_product_id_t	product;
2671 } mcx_devices[] = {
2672 	{ PCI_VENDOR_MELLANOX,	PCI_PRODUCT_MELLANOX_MT27700 },
2673 	{ PCI_VENDOR_MELLANOX,	PCI_PRODUCT_MELLANOX_MT27700VF },
2674 	{ PCI_VENDOR_MELLANOX,	PCI_PRODUCT_MELLANOX_MT27710 },
2675 	{ PCI_VENDOR_MELLANOX,	PCI_PRODUCT_MELLANOX_MT27710VF },
2676 	{ PCI_VENDOR_MELLANOX,	PCI_PRODUCT_MELLANOX_MT27800 },
2677 	{ PCI_VENDOR_MELLANOX,	PCI_PRODUCT_MELLANOX_MT27800VF },
2678 	{ PCI_VENDOR_MELLANOX,	PCI_PRODUCT_MELLANOX_MT28800 },
2679 	{ PCI_VENDOR_MELLANOX,	PCI_PRODUCT_MELLANOX_MT28800VF },
2680 	{ PCI_VENDOR_MELLANOX,	PCI_PRODUCT_MELLANOX_MT28908 },
2681 	{ PCI_VENDOR_MELLANOX,	PCI_PRODUCT_MELLANOX_MT2892  },
2682 };
2683 
2684 struct mcx_eth_proto_capability {
2685 	uint64_t	cap_media;
2686 	uint64_t	cap_baudrate;
2687 };
2688 
2689 static const struct mcx_eth_proto_capability mcx_eth_cap_map[] = {
2690 	[MCX_ETHER_CAP_SGMII]		= { IFM_1000_SGMII,	IF_Gbps(1) },
2691 	[MCX_ETHER_CAP_1000_KX]		= { IFM_1000_KX,	IF_Gbps(1) },
2692 	[MCX_ETHER_CAP_10G_CX4]		= { IFM_10G_CX4,	IF_Gbps(10) },
2693 	[MCX_ETHER_CAP_10G_KX4]		= { IFM_10G_KX4,	IF_Gbps(10) },
2694 	[MCX_ETHER_CAP_10G_KR]		= { IFM_10G_KR,		IF_Gbps(10) },
2695 	[MCX_ETHER_CAP_20G_KR2]		= { IFM_20G_KR2,	IF_Gbps(20) },
2696 	[MCX_ETHER_CAP_40G_CR4]		= { IFM_40G_CR4,	IF_Gbps(40) },
2697 	[MCX_ETHER_CAP_40G_KR4]		= { IFM_40G_KR4,	IF_Gbps(40) },
2698 	[MCX_ETHER_CAP_56G_R4]		= { IFM_56G_R4,		IF_Gbps(56) },
2699 	[MCX_ETHER_CAP_10G_CR]		= { IFM_10G_CR1,	IF_Gbps(10) },
2700 	[MCX_ETHER_CAP_10G_SR]		= { IFM_10G_SR,		IF_Gbps(10) },
2701 	[MCX_ETHER_CAP_10G_LR]		= { IFM_10G_LR,		IF_Gbps(10) },
2702 	[MCX_ETHER_CAP_40G_SR4]		= { IFM_40G_SR4,	IF_Gbps(40) },
2703 	[MCX_ETHER_CAP_40G_LR4]		= { IFM_40G_LR4,	IF_Gbps(40) },
2704 	[MCX_ETHER_CAP_50G_SR2]		= { IFM_50G_SR2,	IF_Gbps(50) },
2705 	[MCX_ETHER_CAP_100G_CR4]	= { IFM_100G_CR4,	IF_Gbps(100) },
2706 	[MCX_ETHER_CAP_100G_SR4]	= { IFM_100G_SR4,	IF_Gbps(100) },
2707 	[MCX_ETHER_CAP_100G_KR4]	= { IFM_100G_KR4,	IF_Gbps(100) },
2708 	[MCX_ETHER_CAP_100G_LR4]	= { IFM_100G_LR4,	IF_Gbps(100) },
2709 	[MCX_ETHER_CAP_100_TX]		= { IFM_100_TX,		IF_Mbps(100) },
2710 	[MCX_ETHER_CAP_1000_T]		= { IFM_1000_T,		IF_Gbps(1) },
2711 	[MCX_ETHER_CAP_10G_T]		= { IFM_10G_T,		IF_Gbps(10) },
2712 	[MCX_ETHER_CAP_25G_CR]		= { IFM_25G_CR,		IF_Gbps(25) },
2713 	[MCX_ETHER_CAP_25G_KR]		= { IFM_25G_KR,		IF_Gbps(25) },
2714 	[MCX_ETHER_CAP_25G_SR]		= { IFM_25G_SR,		IF_Gbps(25) },
2715 	[MCX_ETHER_CAP_50G_CR2]		= { IFM_50G_CR2,	IF_Gbps(50) },
2716 	[MCX_ETHER_CAP_50G_KR2]		= { IFM_50G_KR2,	IF_Gbps(50) },
2717 };
2718 
2719 static int
2720 mcx_get_id(uint32_t val)
2721 {
2722 	return be32toh(val) & 0x00ffffff;
2723 }
2724 
2725 static int
2726 mcx_match(device_t parent, cfdata_t cf, void *aux)
2727 {
2728 	struct pci_attach_args *pa = aux;
2729 	int n;
2730 
2731 	for (n = 0; n < __arraycount(mcx_devices); n++) {
2732 		if (PCI_VENDOR(pa->pa_id) == mcx_devices[n].vendor &&
2733 		    PCI_PRODUCT(pa->pa_id) == mcx_devices[n].product)
2734 			return 1;
2735 	}
2736 
2737 	return 0;
2738 }
2739 
2740 void
2741 mcx_attach(device_t parent, device_t self, void *aux)
2742 {
2743 	struct mcx_softc *sc = device_private(self);
2744 	struct ifnet *ifp = &sc->sc_ec.ec_if;
2745 	struct pci_attach_args *pa = aux;
2746 	struct ifcapreq ifcr;
2747 	uint8_t enaddr[ETHER_ADDR_LEN];
2748 	int counts[PCI_INTR_TYPE_SIZE];
2749 	char intrxname[32];
2750 	pcireg_t memtype;
2751 	uint32_t r;
2752 	unsigned int cq_stride;
2753 	unsigned int cq_size;
2754 	int i, msix;
2755 	kcpuset_t *affinity;
2756 
2757 	sc->sc_dev = self;
2758 	sc->sc_pc = pa->pa_pc;
2759 	sc->sc_tag = pa->pa_tag;
2760 	if (pci_dma64_available(pa))
2761 		sc->sc_dmat = pa->pa_dmat64;
2762 	else
2763 		sc->sc_dmat = pa->pa_dmat;
2764 
2765 	/* Map the PCI memory space */
2766 	memtype = pci_mapreg_type(sc->sc_pc, sc->sc_tag, MCX_HCA_BAR);
2767 	if (pci_mapreg_map(pa, MCX_HCA_BAR, memtype,
2768 #ifdef __NetBSD__
2769 	    0,
2770 #else
2771 	    BUS_SPACE_MAP_PREFETCHABLE,
2772 #endif
2773 	    &sc->sc_memt, &sc->sc_memh,
2774 	    NULL, &sc->sc_mems)) {
2775 		aprint_error(": unable to map register memory\n");
2776 		return;
2777 	}
2778 
2779 	pci_aprint_devinfo(pa, "Ethernet controller");
2780 
2781 	mutex_init(&sc->sc_media_mutex, MUTEX_DEFAULT, IPL_SOFTNET);
2782 
2783 	if (mcx_version(sc) != 0) {
2784 		/* error printed by mcx_version */
2785 		goto unmap;
2786 	}
2787 
2788 	r = mcx_rd(sc, MCX_CMDQ_ADDR_LO);
2789 	cq_stride = 1 << MCX_CMDQ_LOG_STRIDE(r); /* size of the entries */
2790 	cq_size = 1 << MCX_CMDQ_LOG_SIZE(r); /* number of entries */
2791 	if (cq_size > MCX_MAX_CQE) {
2792 		aprint_error_dev(self,
2793 		    "command queue size overflow %u\n", cq_size);
2794 		goto unmap;
2795 	}
2796 	if (cq_stride < sizeof(struct mcx_cmdq_entry)) {
2797 		aprint_error_dev(self,
2798 		    "command queue entry size underflow %u\n", cq_stride);
2799 		goto unmap;
2800 	}
2801 	if (cq_stride * cq_size > MCX_PAGE_SIZE) {
2802 		aprint_error_dev(self, "command queue page overflow\n");
2803 		goto unmap;
2804 	}
2805 
2806 	if (mcx_dmamem_alloc(sc, &sc->sc_doorbell_mem, MCX_DOORBELL_AREA_SIZE,
2807 	    MCX_PAGE_SIZE) != 0) {
2808 		aprint_error_dev(self, "unable to allocate doorbell memory\n");
2809 		goto unmap;
2810 	}
2811 
2812 	if (mcx_dmamem_alloc(sc, &sc->sc_cmdq_mem, MCX_PAGE_SIZE,
2813 	    MCX_PAGE_SIZE) != 0) {
2814 		aprint_error_dev(self, "unable to allocate command queue\n");
2815 		goto dbfree;
2816 	}
2817 
2818 	mcx_wr(sc, MCX_CMDQ_ADDR_HI, MCX_DMA_DVA(&sc->sc_cmdq_mem) >> 32);
2819 	mcx_bar(sc, MCX_CMDQ_ADDR_HI, sizeof(uint32_t),
2820 	    BUS_SPACE_BARRIER_WRITE);
2821 	mcx_wr(sc, MCX_CMDQ_ADDR_LO, MCX_DMA_DVA(&sc->sc_cmdq_mem));
2822 	mcx_bar(sc, MCX_CMDQ_ADDR_LO, sizeof(uint32_t),
2823 	    BUS_SPACE_BARRIER_WRITE);
2824 
2825 	if (mcx_init_wait(sc) != 0) {
2826 		aprint_error_dev(self, "timeout waiting for init\n");
2827 		goto cqfree;
2828 	}
2829 
2830 	sc->sc_cmdq_mask = cq_size - 1;
2831 	sc->sc_cmdq_size = cq_stride;
2832 
2833 	if (mcx_enable_hca(sc) != 0) {
2834 		/* error printed by mcx_enable_hca */
2835 		goto cqfree;
2836 	}
2837 
2838 	if (mcx_issi(sc) != 0) {
2839 		/* error printed by mcx_issi */
2840 		goto teardown;
2841 	}
2842 
2843 	if (mcx_pages(sc, &sc->sc_boot_pages,
2844 	    htobe16(MCX_CMD_QUERY_PAGES_BOOT)) != 0) {
2845 		/* error printed by mcx_pages */
2846 		goto teardown;
2847 	}
2848 
2849 	if (mcx_hca_max_caps(sc) != 0) {
2850 		/* error printed by mcx_hca_max_caps */
2851 		goto teardown;
2852 	}
2853 
2854 	if (mcx_hca_set_caps(sc) != 0) {
2855 		/* error printed by mcx_hca_set_caps */
2856 		goto teardown;
2857 	}
2858 
2859 	if (mcx_pages(sc, &sc->sc_init_pages,
2860 	    htobe16(MCX_CMD_QUERY_PAGES_INIT)) != 0) {
2861 		/* error printed by mcx_pages */
2862 		goto teardown;
2863 	}
2864 
2865 	if (mcx_init_hca(sc) != 0) {
2866 		/* error printed by mcx_init_hca */
2867 		goto teardown;
2868 	}
2869 
2870 	if (mcx_pages(sc, &sc->sc_regular_pages,
2871 	    htobe16(MCX_CMD_QUERY_PAGES_REGULAR)) != 0) {
2872 		/* error printed by mcx_pages */
2873 		goto teardown;
2874 	}
2875 
2876 	/* apparently not necessary? */
2877 	if (mcx_set_driver_version(sc) != 0) {
2878 		/* error printed by mcx_set_driver_version */
2879 		goto teardown;
2880 	}
2881 
2882 	if (mcx_iff(sc) != 0) {	/* modify nic vport context */
2883 		/* error printed by mcx_iff? */
2884 		goto teardown;
2885 	}
2886 
2887 	if (mcx_alloc_uar(sc, &sc->sc_uar) != 0) {
2888 		/* error printed by mcx_alloc_uar */
2889 		goto teardown;
2890 	}
2891 
2892 	if (mcx_alloc_pd(sc) != 0) {
2893 		/* error printed by mcx_alloc_pd */
2894 		goto teardown;
2895 	}
2896 
2897 	if (mcx_alloc_tdomain(sc) != 0) {
2898 		/* error printed by mcx_alloc_tdomain */
2899 		goto teardown;
2900 	}
2901 
2902 	/*
2903 	 * PRM makes no mention of msi interrupts, just legacy and msi-x.
2904 	 * mellanox support tells me legacy interrupts are not supported,
2905 	 * so we're stuck with just msi-x.
2906 	 */
2907 	counts[PCI_INTR_TYPE_MSIX] = -1;
2908 	counts[PCI_INTR_TYPE_MSI] = 0;
2909 	counts[PCI_INTR_TYPE_INTX] = 0;
2910 	if (pci_intr_alloc(pa, &sc->sc_intrs, counts, PCI_INTR_TYPE_MSIX) != 0) {
2911 		aprint_error_dev(self, "unable to allocate interrupt\n");
2912 		goto teardown;
2913 	}
2914 	if (counts[PCI_INTR_TYPE_MSIX] < 2) {
2915 		aprint_error_dev(self, "not enough MSI-X vectors\n");
2916 		goto teardown;
2917 	}
2918 	KASSERT(pci_intr_type(sc->sc_pc, sc->sc_intrs[0]) == PCI_INTR_TYPE_MSIX);
2919 	snprintf(intrxname, sizeof(intrxname), "%s adminq", DEVNAME(sc));
2920 	sc->sc_ihc = mcx_establish_intr(sc, 0, NULL, mcx_admin_intr, sc,
2921 	    intrxname);
2922 	if (sc->sc_ihc == NULL) {
2923 		aprint_error_dev(self, "couldn't establish adminq interrupt\n");
2924 		goto teardown;
2925 	}
2926 
2927 	if (mcx_create_eq(sc, &sc->sc_admin_eq, sc->sc_uar,
2928 	    (1ull << MCX_EVENT_TYPE_INTERNAL_ERROR) |
2929 	    (1ull << MCX_EVENT_TYPE_PORT_CHANGE) |
2930 	    (1ull << MCX_EVENT_TYPE_CMD_COMPLETION) |
2931 	    (1ull << MCX_EVENT_TYPE_PAGE_REQUEST), 0) != 0) {
2932 		/* error printed by mcx_create_eq */
2933 		goto teardown;
2934 	}
2935 
2936 	if (mcx_query_nic_vport_context(sc, enaddr) != 0) {
2937 		/* error printed by mcx_query_nic_vport_context */
2938 		goto teardown;
2939 	}
2940 
2941 	if (mcx_query_special_contexts(sc) != 0) {
2942 		/* error printed by mcx_query_special_contexts */
2943 		goto teardown;
2944 	}
2945 
2946 	if (mcx_set_port_mtu(sc, MCX_HARDMTU) != 0) {
2947 		/* error printed by mcx_set_port_mtu */
2948 		goto teardown;
2949 	}
2950 
2951 	aprint_normal_dev(sc->sc_dev, "Ethernet address %s\n",
2952 	    ether_sprintf(enaddr));
2953 
2954 	msix = counts[PCI_INTR_TYPE_MSIX];
2955 	msix--; /* admin ops took one */
2956 
2957 	sc->sc_nqueues = uimin(MCX_MAX_QUEUES, msix);
2958 	sc->sc_nqueues = uimin(sc->sc_nqueues, ncpu);
2959 	/* Round down to a power of two.  */
2960 	sc->sc_nqueues = 1U << ilog2(sc->sc_nqueues);
2961 	sc->sc_queues = kmem_zalloc(sc->sc_nqueues * sizeof(*sc->sc_queues),
2962 	    KM_SLEEP);
2963 
2964 	strlcpy(ifp->if_xname, DEVNAME(sc), IFNAMSIZ);
2965 	ifp->if_softc = sc;
2966 	ifp->if_flags = IFF_BROADCAST | IFF_MULTICAST | IFF_SIMPLEX;
2967 #ifdef MCX_MPSAFE
2968 	ifp->if_extflags = IFEF_MPSAFE;
2969 #endif
2970 	ifp->if_init = mcx_init;
2971 	ifp->if_stop = mcx_stop;
2972 	ifp->if_ioctl = mcx_ioctl;
2973 	ifp->if_start = mcx_start;
2974 	if (sc->sc_nqueues > 1) {
2975 		ifp->if_transmit = mcx_transmit;
2976 	}
2977 	ifp->if_mtu = sc->sc_hardmtu;
2978 	ifp->if_capabilities = IFCAP_CSUM_IPv4_Rx | IFCAP_CSUM_IPv4_Tx |
2979 	    IFCAP_CSUM_UDPv4_Rx | IFCAP_CSUM_UDPv4_Tx |
2980 	    IFCAP_CSUM_UDPv6_Rx | IFCAP_CSUM_UDPv6_Tx |
2981 	    IFCAP_CSUM_TCPv4_Rx | IFCAP_CSUM_TCPv4_Tx |
2982 	    IFCAP_CSUM_TCPv6_Rx | IFCAP_CSUM_TCPv6_Tx;
2983 	IFQ_SET_MAXLEN(&ifp->if_snd, 1024);
2984 	IFQ_SET_READY(&ifp->if_snd);
2985 
2986 	sc->sc_ec.ec_capabilities = ETHERCAP_JUMBO_MTU |
2987 	    ETHERCAP_VLAN_MTU | ETHERCAP_VLAN_HWTAGGING;
2988 	sc->sc_ec.ec_capenable |= ETHERCAP_VLAN_HWTAGGING;
2989 
2990 	sc->sc_ec.ec_ifmedia = &sc->sc_media;
2991 	ifmedia_init_with_lock(&sc->sc_media, IFM_IMASK, mcx_media_change,
2992 	    mcx_media_status, &sc->sc_media_mutex);
2993 	mcx_media_add_types(sc);
2994 	ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL);
2995 	ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO);
2996 
2997 	if_attach(ifp);
2998 
2999 	/* Enable hardware offload by default */
3000 	memset(&ifcr, 0, sizeof(ifcr));
3001 	ifcr.ifcr_capenable = ifp->if_capabilities;
3002 	ifioctl_common(ifp, SIOCSIFCAP, &ifcr);
3003 
3004 	if_deferred_start_init(ifp, NULL);
3005 
3006 	ether_ifattach(ifp, enaddr);
3007 
3008 	kcpuset_create(&affinity, false);
3009 	kcpuset_set(affinity, 0);
3010 
3011 	for (i = 0; i < sc->sc_nqueues; i++) {
3012 		struct mcx_queues *q = &sc->sc_queues[i];
3013 		struct mcx_rx *rx = &q->q_rx;
3014 		struct mcx_tx *tx = &q->q_tx;
3015 		int vec;
3016 
3017 		vec = i + 1;
3018 		q->q_sc = sc;
3019 		q->q_index = i;
3020 
3021 		if (mcx_alloc_uar(sc, &q->q_uar) != 0) {
3022 			aprint_error_dev(self, "unable to alloc uar %d\n", i);
3023 			goto teardown;
3024 		}
3025 
3026 		if (mcx_create_eq(sc, &q->q_eq, q->q_uar, 0, vec) != 0) {
3027 			aprint_error_dev(self,
3028 			    "unable to create event queue %d\n", i);
3029 			goto teardown;
3030 		}
3031 
3032 		rx->rx_softc = sc;
3033 		callout_init(&rx->rx_refill, CALLOUT_FLAGS);
3034 		callout_setfunc(&rx->rx_refill, mcx_refill, rx);
3035 
3036 		tx->tx_softc = sc;
3037 		mutex_init(&tx->tx_lock, MUTEX_DEFAULT, IPL_NET);
3038 		tx->tx_pcq = pcq_create(MCX_TXQ_NUM, KM_SLEEP);
3039 		tx->tx_softint = softint_establish(SOFTINT_NET|SOFTINT_MPSAFE,
3040 		    mcx_deferred_transmit, tx);
3041 
3042 		snprintf(intrxname, sizeof(intrxname), "%s queue %d",
3043 		    DEVNAME(sc), i);
3044 		q->q_ihc = mcx_establish_intr(sc, vec, affinity, mcx_cq_intr,
3045 		    q, intrxname);
3046 	}
3047 
3048 	callout_init(&sc->sc_calibrate, CALLOUT_FLAGS);
3049 	callout_setfunc(&sc->sc_calibrate, mcx_calibrate, sc);
3050 
3051 	if (workqueue_create(&sc->sc_workq, "mcxportchg", mcx_port_change, sc,
3052 	    PRI_NONE, IPL_NET, 0) != 0) {
3053 		aprint_error_dev(self, "couldn't create port change workq\n");
3054 		goto teardown;
3055 	}
3056 
3057 	mcx_port_change(&sc->sc_port_change, sc);
3058 
3059 	sc->sc_mac_flow_table_id = -1;
3060 	sc->sc_rss_flow_table_id = -1;
3061 	sc->sc_rqt = -1;
3062 	for (i = 0; i < MCX_NUM_FLOW_GROUPS; i++) {
3063 		struct mcx_flow_group *mfg = &sc->sc_flow_group[i];
3064 		mfg->g_id = -1;
3065 		mfg->g_table = -1;
3066 		mfg->g_size = 0;
3067 		mfg->g_start = 0;
3068 	}
3069 	sc->sc_extra_mcast = 0;
3070 	memset(sc->sc_mcast_flows, 0, sizeof(sc->sc_mcast_flows));
3071 
3072 #if NKSTAT > 0
3073 	mcx_kstat_attach(sc);
3074 #endif
3075 	mcx_timecounter_attach(sc);
3076 	return;
3077 
3078 teardown:
3079 	mcx_teardown_hca(sc, htobe16(MCX_CMD_TEARDOWN_HCA_GRACEFUL));
3080 	/* error printed by mcx_teardown_hca, and we're already unwinding */
3081 cqfree:
3082 	mcx_wr(sc, MCX_CMDQ_ADDR_HI, MCX_DMA_DVA(&sc->sc_cmdq_mem) >> 32);
3083 	mcx_bar(sc, MCX_CMDQ_ADDR_HI, sizeof(uint64_t),
3084 	    BUS_SPACE_BARRIER_WRITE);
3085 	mcx_wr(sc, MCX_CMDQ_ADDR_LO, MCX_DMA_DVA(&sc->sc_cmdq_mem) |
3086 	    MCX_CMDQ_INTERFACE_DISABLED);
3087 	mcx_bar(sc, MCX_CMDQ_ADDR_LO, sizeof(uint64_t),
3088 	    BUS_SPACE_BARRIER_WRITE);
3089 
3090 	mcx_wr(sc, MCX_CMDQ_ADDR_HI, 0);
3091 	mcx_bar(sc, MCX_CMDQ_ADDR_HI, sizeof(uint64_t),
3092 	    BUS_SPACE_BARRIER_WRITE);
3093 	mcx_wr(sc, MCX_CMDQ_ADDR_LO, MCX_CMDQ_INTERFACE_DISABLED);
3094 
3095 	mcx_dmamem_free(sc, &sc->sc_cmdq_mem);
3096 dbfree:
3097 	mcx_dmamem_free(sc, &sc->sc_doorbell_mem);
3098 unmap:
3099 	bus_space_unmap(sc->sc_memt, sc->sc_memh, sc->sc_mems);
3100 	sc->sc_mems = 0;
3101 }
3102 
3103 static void *
3104 mcx_establish_intr(struct mcx_softc *sc, int index, kcpuset_t *affinity,
3105     int (*func)(void *), void *arg, const char *xname)
3106 {
3107 	char intrbuf[PCI_INTRSTR_LEN];
3108 	const char *intrstr;
3109 	void *ih;
3110 
3111 	pci_intr_setattr(sc->sc_pc, &sc->sc_intrs[index], PCI_INTR_MPSAFE,
3112 	    true);
3113 
3114 	intrstr = pci_intr_string(sc->sc_pc, sc->sc_intrs[index], intrbuf,
3115 	    sizeof(intrbuf));
3116 	ih = pci_intr_establish_xname(sc->sc_pc, sc->sc_intrs[index], IPL_NET,
3117 	    func, arg, xname);
3118 	if (ih == NULL) {
3119 		aprint_error_dev(sc->sc_dev,
3120 		    "unable to establish interrupt%s%s\n",
3121 		    intrstr ? " at " : "",
3122 		    intrstr ? intrstr : "");
3123 		return NULL;
3124 	}
3125 
3126 	if (affinity != NULL && index > 0) {
3127 		/* Round-robin affinity */
3128 		kcpuset_zero(affinity);
3129 		kcpuset_set(affinity, (index - 1) % ncpu);
3130 		interrupt_distribute(ih, affinity, NULL);
3131 	}
3132 
3133 	return ih;
3134 }
3135 
3136 static void
3137 mcx_rxr_init(struct mcx_rxring *rxr, u_int lwm __unused, u_int hwm)
3138 {
3139 	rxr->rxr_total = hwm;
3140 	rxr->rxr_inuse = 0;
3141 }
3142 
3143 static u_int
3144 mcx_rxr_get(struct mcx_rxring *rxr, u_int max)
3145 {
3146 	const u_int taken = MIN(max, rxr->rxr_total - rxr->rxr_inuse);
3147 
3148 	rxr->rxr_inuse += taken;
3149 
3150 	return taken;
3151 }
3152 
3153 static void
3154 mcx_rxr_put(struct mcx_rxring *rxr, u_int n)
3155 {
3156 	rxr->rxr_inuse -= n;
3157 }
3158 
3159 static u_int
3160 mcx_rxr_inuse(struct mcx_rxring *rxr)
3161 {
3162 	return rxr->rxr_inuse;
3163 }
3164 
3165 static int
3166 mcx_version(struct mcx_softc *sc)
3167 {
3168 	uint32_t fw0, fw1;
3169 	uint16_t cmdif;
3170 
3171 	fw0 = mcx_rd(sc, MCX_FW_VER);
3172 	fw1 = mcx_rd(sc, MCX_CMDIF_FW_SUBVER);
3173 
3174 	aprint_normal_dev(sc->sc_dev, "FW %u.%u.%04u\n", MCX_FW_VER_MAJOR(fw0),
3175 	    MCX_FW_VER_MINOR(fw0), MCX_FW_VER_SUBMINOR(fw1));
3176 
3177 	cmdif = MCX_CMDIF(fw1);
3178 	if (cmdif != MCX_CMD_IF_SUPPORTED) {
3179 		aprint_error_dev(sc->sc_dev,
3180 		    "unsupported command interface %u\n", cmdif);
3181 		return (-1);
3182 	}
3183 
3184 	return (0);
3185 }
3186 
3187 static int
3188 mcx_init_wait(struct mcx_softc *sc)
3189 {
3190 	unsigned int i;
3191 	uint32_t r;
3192 
3193 	for (i = 0; i < 2000; i++) {
3194 		r = mcx_rd(sc, MCX_STATE);
3195 		if ((r & MCX_STATE_MASK) == MCX_STATE_READY)
3196 			return (0);
3197 
3198 		delay(1000);
3199 		mcx_bar(sc, MCX_STATE, sizeof(uint32_t),
3200 		    BUS_SPACE_BARRIER_READ);
3201 	}
3202 
3203 	return (-1);
3204 }
3205 
3206 static uint8_t
3207 mcx_cmdq_poll(struct mcx_softc *sc, struct mcx_cmdq_entry *cqe,
3208     unsigned int msec)
3209 {
3210 	unsigned int i;
3211 
3212 	for (i = 0; i < msec; i++) {
3213 		bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_cmdq_mem),
3214 		    0, MCX_DMA_LEN(&sc->sc_cmdq_mem), BUS_DMASYNC_POSTRW);
3215 
3216 		if ((cqe->cq_status & MCX_CQ_STATUS_OWN_MASK) ==
3217 		    MCX_CQ_STATUS_OWN_SW)
3218 			return (0);
3219 
3220 		delay(1000);
3221 	}
3222 
3223 	return (ETIMEDOUT);
3224 }
3225 
3226 static uint32_t
3227 mcx_mix_u64(uint32_t xor, uint64_t u64)
3228 {
3229 	xor ^= u64 >> 32;
3230 	xor ^= u64;
3231 
3232 	return (xor);
3233 }
3234 
3235 static uint32_t
3236 mcx_mix_u32(uint32_t xor, uint32_t u32)
3237 {
3238 	xor ^= u32;
3239 
3240 	return (xor);
3241 }
3242 
3243 static uint32_t
3244 mcx_mix_u8(uint32_t xor, uint8_t u8)
3245 {
3246 	xor ^= u8;
3247 
3248 	return (xor);
3249 }
3250 
3251 static uint8_t
3252 mcx_mix_done(uint32_t xor)
3253 {
3254 	xor ^= xor >> 16;
3255 	xor ^= xor >> 8;
3256 
3257 	return (xor);
3258 }
3259 
3260 static uint8_t
3261 mcx_xor(const void *buf, size_t len)
3262 {
3263 	const uint32_t *dwords = buf;
3264 	uint32_t xor = 0xff;
3265 	size_t i;
3266 
3267 	len /= sizeof(*dwords);
3268 
3269 	for (i = 0; i < len; i++)
3270 		xor ^= dwords[i];
3271 
3272 	return (mcx_mix_done(xor));
3273 }
3274 
3275 static uint8_t
3276 mcx_cmdq_token(struct mcx_softc *sc)
3277 {
3278 	uint8_t token;
3279 
3280 	do {
3281 		token = ++sc->sc_cmdq_token;
3282 	} while (token == 0);
3283 
3284 	return (token);
3285 }
3286 
3287 static void
3288 mcx_cmdq_init(struct mcx_softc *sc, struct mcx_cmdq_entry *cqe,
3289     uint32_t ilen, uint32_t olen, uint8_t token)
3290 {
3291 	memset(cqe, 0, sc->sc_cmdq_size);
3292 
3293 	cqe->cq_type = MCX_CMDQ_TYPE_PCIE;
3294 	be32enc(&cqe->cq_input_length, ilen);
3295 	be32enc(&cqe->cq_output_length, olen);
3296 	cqe->cq_token = token;
3297 	cqe->cq_status = MCX_CQ_STATUS_OWN_HW;
3298 }
3299 
3300 static void
3301 mcx_cmdq_sign(struct mcx_cmdq_entry *cqe)
3302 {
3303 	cqe->cq_signature = ~mcx_xor(cqe, sizeof(*cqe));
3304 }
3305 
3306 static int
3307 mcx_cmdq_verify(const struct mcx_cmdq_entry *cqe)
3308 {
3309 	/* return (mcx_xor(cqe, sizeof(*cqe)) ? -1 :  0); */
3310 	return (0);
3311 }
3312 
3313 static void *
3314 mcx_cmdq_in(struct mcx_cmdq_entry *cqe)
3315 {
3316 	return (&cqe->cq_input_data);
3317 }
3318 
3319 static void *
3320 mcx_cmdq_out(struct mcx_cmdq_entry *cqe)
3321 {
3322 	return (&cqe->cq_output_data);
3323 }
3324 
3325 static void
3326 mcx_cmdq_post(struct mcx_softc *sc, struct mcx_cmdq_entry *cqe,
3327     unsigned int slot)
3328 {
3329 	mcx_cmdq_sign(cqe);
3330 
3331 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_cmdq_mem),
3332 	    0, MCX_DMA_LEN(&sc->sc_cmdq_mem), BUS_DMASYNC_PRERW);
3333 
3334 	mcx_wr(sc, MCX_CMDQ_DOORBELL, 1U << slot);
3335 	mcx_bar(sc, MCX_CMDQ_DOORBELL, sizeof(uint32_t),
3336 	    BUS_SPACE_BARRIER_WRITE);
3337 }
3338 
3339 static int
3340 mcx_enable_hca(struct mcx_softc *sc)
3341 {
3342 	struct mcx_cmdq_entry *cqe;
3343 	struct mcx_cmd_enable_hca_in *in;
3344 	struct mcx_cmd_enable_hca_out *out;
3345 	int error;
3346 	uint8_t status;
3347 
3348 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
3349 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
3350 
3351 	in = mcx_cmdq_in(cqe);
3352 	in->cmd_opcode = htobe16(MCX_CMD_ENABLE_HCA);
3353 	in->cmd_op_mod = htobe16(0);
3354 	in->cmd_function_id = htobe16(0);
3355 
3356 	mcx_cmdq_post(sc, cqe, 0);
3357 
3358 	error = mcx_cmdq_poll(sc, cqe, 1000);
3359 	if (error != 0) {
3360 		printf(", hca enable timeout\n");
3361 		return (-1);
3362 	}
3363 	if (mcx_cmdq_verify(cqe) != 0) {
3364 		printf(", hca enable command corrupt\n");
3365 		return (-1);
3366 	}
3367 
3368 	status = cqe->cq_output_data[0];
3369 	if (status != MCX_CQ_STATUS_OK) {
3370 		printf(", hca enable failed (%x)\n", status);
3371 		return (-1);
3372 	}
3373 
3374 	return (0);
3375 }
3376 
3377 static int
3378 mcx_teardown_hca(struct mcx_softc *sc, uint16_t profile)
3379 {
3380 	struct mcx_cmdq_entry *cqe;
3381 	struct mcx_cmd_teardown_hca_in *in;
3382 	struct mcx_cmd_teardown_hca_out *out;
3383 	int error;
3384 	uint8_t status;
3385 
3386 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
3387 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
3388 
3389 	in = mcx_cmdq_in(cqe);
3390 	in->cmd_opcode = htobe16(MCX_CMD_TEARDOWN_HCA);
3391 	in->cmd_op_mod = htobe16(0);
3392 	in->cmd_profile = profile;
3393 
3394 	mcx_cmdq_post(sc, cqe, 0);
3395 
3396 	error = mcx_cmdq_poll(sc, cqe, 1000);
3397 	if (error != 0) {
3398 		printf(", hca teardown timeout\n");
3399 		return (-1);
3400 	}
3401 	if (mcx_cmdq_verify(cqe) != 0) {
3402 		printf(", hca teardown command corrupt\n");
3403 		return (-1);
3404 	}
3405 
3406 	status = cqe->cq_output_data[0];
3407 	if (status != MCX_CQ_STATUS_OK) {
3408 		printf(", hca teardown failed (%x)\n", status);
3409 		return (-1);
3410 	}
3411 
3412 	return (0);
3413 }
3414 
3415 static int
3416 mcx_cmdq_mboxes_alloc(struct mcx_softc *sc, struct mcx_dmamem *mxm,
3417     unsigned int nmb, uint64_t *ptr, uint8_t token)
3418 {
3419 	uint8_t *kva;
3420 	uint64_t dva;
3421 	int i;
3422 	int error;
3423 
3424 	error = mcx_dmamem_alloc(sc, mxm,
3425 	    nmb * MCX_CMDQ_MAILBOX_SIZE, MCX_CMDQ_MAILBOX_ALIGN);
3426 	if (error != 0)
3427 		return (error);
3428 
3429 	mcx_dmamem_zero(mxm);
3430 
3431 	dva = MCX_DMA_DVA(mxm);
3432 	kva = MCX_DMA_KVA(mxm);
3433 	for (i = 0; i < nmb; i++) {
3434 		struct mcx_cmdq_mailbox *mbox = (struct mcx_cmdq_mailbox *)kva;
3435 
3436 		/* patch the cqe or mbox pointing at this one */
3437 		be64enc(ptr, dva);
3438 
3439 		/* fill in this mbox */
3440 		be32enc(&mbox->mb_block_number, i);
3441 		mbox->mb_token = token;
3442 
3443 		/* move to the next one */
3444 		ptr = &mbox->mb_next_ptr;
3445 
3446 		dva += MCX_CMDQ_MAILBOX_SIZE;
3447 		kva += MCX_CMDQ_MAILBOX_SIZE;
3448 	}
3449 
3450 	return (0);
3451 }
3452 
3453 static uint32_t
3454 mcx_cmdq_mbox_ctrl_sig(const struct mcx_cmdq_mailbox *mb)
3455 {
3456 	uint32_t xor = 0xff;
3457 
3458 	/* only 3 fields get set, so mix them directly */
3459 	xor = mcx_mix_u64(xor, mb->mb_next_ptr);
3460 	xor = mcx_mix_u32(xor, mb->mb_block_number);
3461 	xor = mcx_mix_u8(xor, mb->mb_token);
3462 
3463 	return (mcx_mix_done(xor));
3464 }
3465 
3466 static void
3467 mcx_cmdq_mboxes_sign(struct mcx_dmamem *mxm, unsigned int nmb)
3468 {
3469 	uint8_t *kva;
3470 	int i;
3471 
3472 	kva = MCX_DMA_KVA(mxm);
3473 
3474 	for (i = 0; i < nmb; i++) {
3475 		struct mcx_cmdq_mailbox *mb = (struct mcx_cmdq_mailbox *)kva;
3476 		uint8_t sig = mcx_cmdq_mbox_ctrl_sig(mb);
3477 		mb->mb_ctrl_signature = sig;
3478 		mb->mb_signature = sig ^
3479 		    mcx_xor(mb->mb_data, sizeof(mb->mb_data));
3480 
3481 		kva += MCX_CMDQ_MAILBOX_SIZE;
3482 	}
3483 }
3484 
3485 static void
3486 mcx_cmdq_mboxes_sync(struct mcx_softc *sc, struct mcx_dmamem *mxm, int ops)
3487 {
3488 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(mxm),
3489 	    0, MCX_DMA_LEN(mxm), ops);
3490 }
3491 
3492 static struct mcx_cmdq_mailbox *
3493 mcx_cq_mbox(struct mcx_dmamem *mxm, unsigned int i)
3494 {
3495 	uint8_t *kva;
3496 
3497 	kva = MCX_DMA_KVA(mxm);
3498 	kva += i * MCX_CMDQ_MAILBOX_SIZE;
3499 
3500 	return ((struct mcx_cmdq_mailbox *)kva);
3501 }
3502 
3503 static inline void *
3504 mcx_cq_mbox_data(struct mcx_cmdq_mailbox *mb)
3505 {
3506 	return (&mb->mb_data);
3507 }
3508 
3509 static void
3510 mcx_cmdq_mboxes_copyin(struct mcx_dmamem *mxm, unsigned int nmb,
3511     void *b, size_t len)
3512 {
3513 	uint8_t *buf = b;
3514 	struct mcx_cmdq_mailbox *mb;
3515 	int i;
3516 
3517 	mb = (struct mcx_cmdq_mailbox *)MCX_DMA_KVA(mxm);
3518 	for (i = 0; i < nmb; i++) {
3519 
3520 		memcpy(mb->mb_data, buf, uimin(sizeof(mb->mb_data), len));
3521 
3522 		if (sizeof(mb->mb_data) >= len)
3523 			break;
3524 
3525 		buf += sizeof(mb->mb_data);
3526 		len -= sizeof(mb->mb_data);
3527 		mb++;
3528 	}
3529 }
3530 
3531 static void
3532 mcx_cmdq_mboxes_pas(struct mcx_dmamem *mxm, int offset, int npages,
3533     struct mcx_dmamem *buf)
3534 {
3535 	uint64_t *pas;
3536 	int mbox, mbox_pages, i;
3537 
3538 	mbox = offset / MCX_CMDQ_MAILBOX_DATASIZE;
3539 	offset %= MCX_CMDQ_MAILBOX_DATASIZE;
3540 
3541 	pas = mcx_cq_mbox_data(mcx_cq_mbox(mxm, mbox));
3542 	pas += (offset / sizeof(*pas));
3543 	mbox_pages = (MCX_CMDQ_MAILBOX_DATASIZE - offset) / sizeof(*pas);
3544 	for (i = 0; i < npages; i++) {
3545 		if (i == mbox_pages) {
3546 			mbox++;
3547 			pas = mcx_cq_mbox_data(mcx_cq_mbox(mxm, mbox));
3548 			mbox_pages += MCX_CMDQ_MAILBOX_DATASIZE / sizeof(*pas);
3549 		}
3550 		*pas = htobe64(MCX_DMA_DVA(buf) + (i * MCX_PAGE_SIZE));
3551 		pas++;
3552 	}
3553 }
3554 
3555 static void
3556 mcx_cmdq_mboxes_copyout(struct mcx_dmamem *mxm, int nmb, void *b, size_t len)
3557 {
3558 	uint8_t *buf = b;
3559 	struct mcx_cmdq_mailbox *mb;
3560 	int i;
3561 
3562 	mb = (struct mcx_cmdq_mailbox *)MCX_DMA_KVA(mxm);
3563 	for (i = 0; i < nmb; i++) {
3564 		memcpy(buf, mb->mb_data, uimin(sizeof(mb->mb_data), len));
3565 
3566 		if (sizeof(mb->mb_data) >= len)
3567 			break;
3568 
3569 		buf += sizeof(mb->mb_data);
3570 		len -= sizeof(mb->mb_data);
3571 		mb++;
3572 	}
3573 }
3574 
3575 static void
3576 mcx_cq_mboxes_free(struct mcx_softc *sc, struct mcx_dmamem *mxm)
3577 {
3578 	mcx_dmamem_free(sc, mxm);
3579 }
3580 
3581 #if 0
3582 static void
3583 mcx_cmdq_dump(const struct mcx_cmdq_entry *cqe)
3584 {
3585 	unsigned int i;
3586 
3587 	printf(" type %02x, ilen %u, iptr %016llx", cqe->cq_type,
3588 	    be32dec(&cqe->cq_input_length), be64dec(&cqe->cq_input_ptr));
3589 
3590 	printf(", idata ");
3591 	for (i = 0; i < sizeof(cqe->cq_input_data); i++)
3592 		printf("%02x", cqe->cq_input_data[i]);
3593 
3594 	printf(", odata ");
3595 	for (i = 0; i < sizeof(cqe->cq_output_data); i++)
3596 		printf("%02x", cqe->cq_output_data[i]);
3597 
3598 	printf(", optr %016llx, olen %u, token %02x, sig %02x, status %02x",
3599 	    be64dec(&cqe->cq_output_ptr), be32dec(&cqe->cq_output_length),
3600 	    cqe->cq_token, cqe->cq_signature, cqe->cq_status);
3601 }
3602 
3603 static void
3604 mcx_cmdq_mbox_dump(struct mcx_dmamem *mboxes, int num)
3605 {
3606 	int i, j;
3607 	uint8_t *d;
3608 
3609 	for (i = 0; i < num; i++) {
3610 		struct mcx_cmdq_mailbox *mbox;
3611 		mbox = mcx_cq_mbox(mboxes, i);
3612 
3613 		d = mcx_cq_mbox_data(mbox);
3614 		for (j = 0; j < MCX_CMDQ_MAILBOX_DATASIZE; j++) {
3615 			if (j != 0 && (j % 16 == 0))
3616 				printf("\n");
3617 			printf("%.2x ", d[j]);
3618 		}
3619 	}
3620 }
3621 #endif
3622 
3623 static int
3624 mcx_access_hca_reg(struct mcx_softc *sc, uint16_t reg, int op, void *data,
3625     int len)
3626 {
3627 	struct mcx_dmamem mxm;
3628 	struct mcx_cmdq_entry *cqe;
3629 	struct mcx_cmd_access_reg_in *in;
3630 	struct mcx_cmd_access_reg_out *out;
3631 	uint8_t token = mcx_cmdq_token(sc);
3632 	int error, nmb;
3633 
3634 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
3635 	mcx_cmdq_init(sc, cqe, sizeof(*in) + len, sizeof(*out) + len,
3636 	    token);
3637 
3638 	in = mcx_cmdq_in(cqe);
3639 	in->cmd_opcode = htobe16(MCX_CMD_ACCESS_REG);
3640 	in->cmd_op_mod = htobe16(op);
3641 	in->cmd_register_id = htobe16(reg);
3642 
3643 	nmb = howmany(len, MCX_CMDQ_MAILBOX_DATASIZE);
3644 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, nmb,
3645 	    &cqe->cq_output_ptr, token) != 0) {
3646 		printf(", unable to allocate access reg mailboxen\n");
3647 		return (-1);
3648 	}
3649 	cqe->cq_input_ptr = cqe->cq_output_ptr;
3650 	mcx_cmdq_mboxes_copyin(&mxm, nmb, data, len);
3651 	mcx_cmdq_mboxes_sign(&mxm, nmb);
3652 	mcx_cmdq_mboxes_sync(sc, &mxm, BUS_DMASYNC_PRERW);
3653 
3654 	mcx_cmdq_post(sc, cqe, 0);
3655 	error = mcx_cmdq_poll(sc, cqe, 1000);
3656 	mcx_cmdq_mboxes_sync(sc, &mxm, BUS_DMASYNC_POSTRW);
3657 
3658 	if (error != 0) {
3659 		printf("%s: access reg (%s %x) timeout\n", DEVNAME(sc),
3660 		    (op == MCX_REG_OP_WRITE ? "write" : "read"), reg);
3661 		goto free;
3662 	}
3663 	error = mcx_cmdq_verify(cqe);
3664 	if (error != 0) {
3665 		printf("%s: access reg (%s %x) reply corrupt\n",
3666 		    (op == MCX_REG_OP_WRITE ? "write" : "read"), DEVNAME(sc),
3667 		    reg);
3668 		goto free;
3669 	}
3670 
3671 	out = mcx_cmdq_out(cqe);
3672 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
3673 		printf("%s: access reg (%s %x) failed (%x, %.6x)\n",
3674 		    DEVNAME(sc), (op == MCX_REG_OP_WRITE ? "write" : "read"),
3675 		    reg, out->cmd_status, be32toh(out->cmd_syndrome));
3676 		error = -1;
3677 		goto free;
3678 	}
3679 
3680 	mcx_cmdq_mboxes_copyout(&mxm, nmb, data, len);
3681 free:
3682 	mcx_dmamem_free(sc, &mxm);
3683 
3684 	return (error);
3685 }
3686 
3687 static int
3688 mcx_set_issi(struct mcx_softc *sc, struct mcx_cmdq_entry *cqe,
3689     unsigned int slot)
3690 {
3691 	struct mcx_cmd_set_issi_in *in;
3692 	struct mcx_cmd_set_issi_out *out;
3693 	uint8_t status;
3694 
3695 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
3696 
3697 	in = mcx_cmdq_in(cqe);
3698 	in->cmd_opcode = htobe16(MCX_CMD_SET_ISSI);
3699 	in->cmd_op_mod = htobe16(0);
3700 	in->cmd_current_issi = htobe16(MCX_ISSI);
3701 
3702 	mcx_cmdq_post(sc, cqe, slot);
3703 	if (mcx_cmdq_poll(sc, cqe, 1000) != 0)
3704 		return (-1);
3705 	if (mcx_cmdq_verify(cqe) != 0)
3706 		return (-1);
3707 
3708 	status = cqe->cq_output_data[0];
3709 	if (status != MCX_CQ_STATUS_OK)
3710 		return (-1);
3711 
3712 	return (0);
3713 }
3714 
3715 static int
3716 mcx_issi(struct mcx_softc *sc)
3717 {
3718 	struct mcx_dmamem mxm;
3719 	struct mcx_cmdq_entry *cqe;
3720 	struct mcx_cmd_query_issi_in *in;
3721 	struct mcx_cmd_query_issi_il_out *out;
3722 	struct mcx_cmd_query_issi_mb_out *mb;
3723 	uint8_t token = mcx_cmdq_token(sc);
3724 	uint8_t status;
3725 	int error;
3726 
3727 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
3728 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out) + sizeof(*mb), token);
3729 
3730 	in = mcx_cmdq_in(cqe);
3731 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_ISSI);
3732 	in->cmd_op_mod = htobe16(0);
3733 
3734 	CTASSERT(sizeof(*mb) <= MCX_CMDQ_MAILBOX_DATASIZE);
3735 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
3736 	    &cqe->cq_output_ptr, token) != 0) {
3737 		printf(", unable to allocate query issi mailbox\n");
3738 		return (-1);
3739 	}
3740 	mcx_cmdq_mboxes_sign(&mxm, 1);
3741 
3742 	mcx_cmdq_post(sc, cqe, 0);
3743 	error = mcx_cmdq_poll(sc, cqe, 1000);
3744 	if (error != 0) {
3745 		printf(", query issi timeout\n");
3746 		goto free;
3747 	}
3748 	error = mcx_cmdq_verify(cqe);
3749 	if (error != 0) {
3750 		printf(", query issi reply corrupt\n");
3751 		goto free;
3752 	}
3753 
3754 	status = cqe->cq_output_data[0];
3755 	switch (status) {
3756 	case MCX_CQ_STATUS_OK:
3757 		break;
3758 	case MCX_CQ_STATUS_BAD_OPCODE:
3759 		/* use ISSI 0 */
3760 		goto free;
3761 	default:
3762 		printf(", query issi failed (%x)\n", status);
3763 		error = -1;
3764 		goto free;
3765 	}
3766 
3767 	out = mcx_cmdq_out(cqe);
3768 	if (out->cmd_current_issi == htobe16(MCX_ISSI)) {
3769 		/* use ISSI 1 */
3770 		goto free;
3771 	}
3772 
3773 	/* don't need to read cqe anymore, can be used for SET ISSI */
3774 
3775 	mb = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
3776 	CTASSERT(MCX_ISSI < NBBY);
3777 	 /* XXX math is hard */
3778 	if (!ISSET(mb->cmd_supported_issi[79], 1 << MCX_ISSI)) {
3779 		/* use ISSI 0 */
3780 		goto free;
3781 	}
3782 
3783 	if (mcx_set_issi(sc, cqe, 0) != 0) {
3784 		/* ignore the error, just use ISSI 0 */
3785 	} else {
3786 		/* use ISSI 1 */
3787 	}
3788 
3789 free:
3790 	mcx_cq_mboxes_free(sc, &mxm);
3791 	return (error);
3792 }
3793 
3794 static int
3795 mcx_query_pages(struct mcx_softc *sc, uint16_t type,
3796     int32_t *npages, uint16_t *func_id)
3797 {
3798 	struct mcx_cmdq_entry *cqe;
3799 	struct mcx_cmd_query_pages_in *in;
3800 	struct mcx_cmd_query_pages_out *out;
3801 
3802 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
3803 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
3804 
3805 	in = mcx_cmdq_in(cqe);
3806 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_PAGES);
3807 	in->cmd_op_mod = type;
3808 
3809 	mcx_cmdq_post(sc, cqe, 0);
3810 	if (mcx_cmdq_poll(sc, cqe, 1000) != 0) {
3811 		printf(", query pages timeout\n");
3812 		return (-1);
3813 	}
3814 	if (mcx_cmdq_verify(cqe) != 0) {
3815 		printf(", query pages reply corrupt\n");
3816 		return (-1);
3817 	}
3818 
3819 	out = mcx_cmdq_out(cqe);
3820 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
3821 		printf(", query pages failed (%x)\n", out->cmd_status);
3822 		return (-1);
3823 	}
3824 
3825 	*func_id = out->cmd_func_id;
3826 	*npages = be32dec(&out->cmd_num_pages);
3827 
3828 	return (0);
3829 }
3830 
3831 struct bus_dma_iter {
3832 	bus_dmamap_t		i_map;
3833 	bus_size_t		i_offset;
3834 	unsigned int		i_index;
3835 };
3836 
3837 static void
3838 bus_dma_iter_init(struct bus_dma_iter *i, bus_dmamap_t map)
3839 {
3840 	i->i_map = map;
3841 	i->i_offset = 0;
3842 	i->i_index = 0;
3843 }
3844 
3845 static bus_addr_t
3846 bus_dma_iter_addr(struct bus_dma_iter *i)
3847 {
3848 	return (i->i_map->dm_segs[i->i_index].ds_addr + i->i_offset);
3849 }
3850 
3851 static void
3852 bus_dma_iter_add(struct bus_dma_iter *i, bus_size_t size)
3853 {
3854 	bus_dma_segment_t *seg = i->i_map->dm_segs + i->i_index;
3855 	bus_size_t diff;
3856 
3857 	do {
3858 		diff = seg->ds_len - i->i_offset;
3859 		if (size < diff)
3860 			break;
3861 
3862 		size -= diff;
3863 
3864 		seg++;
3865 
3866 		i->i_offset = 0;
3867 		i->i_index++;
3868 	} while (size > 0);
3869 
3870 	i->i_offset += size;
3871 }
3872 
3873 static int
3874 mcx_add_pages(struct mcx_softc *sc, struct mcx_hwmem *mhm, uint16_t func_id)
3875 {
3876 	struct mcx_dmamem mxm;
3877 	struct mcx_cmdq_entry *cqe;
3878 	struct mcx_cmd_manage_pages_in *in;
3879 	struct mcx_cmd_manage_pages_out *out;
3880 	unsigned int paslen, nmb, i, j, npages;
3881 	struct bus_dma_iter iter;
3882 	uint64_t *pas;
3883 	uint8_t status;
3884 	uint8_t token = mcx_cmdq_token(sc);
3885 	int error;
3886 
3887 	npages = mhm->mhm_npages;
3888 
3889 	paslen = sizeof(*pas) * npages;
3890 	nmb = howmany(paslen, MCX_CMDQ_MAILBOX_DATASIZE);
3891 
3892 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
3893 	mcx_cmdq_init(sc, cqe, sizeof(*in) + paslen, sizeof(*out), token);
3894 
3895 	in = mcx_cmdq_in(cqe);
3896 	in->cmd_opcode = htobe16(MCX_CMD_MANAGE_PAGES);
3897 	in->cmd_op_mod = htobe16(MCX_CMD_MANAGE_PAGES_ALLOC_SUCCESS);
3898 	in->cmd_func_id = func_id;
3899 	be32enc(&in->cmd_input_num_entries, npages);
3900 
3901 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, nmb,
3902 	    &cqe->cq_input_ptr, token) != 0) {
3903 		printf(", unable to allocate manage pages mailboxen\n");
3904 		return (-1);
3905 	}
3906 
3907 	bus_dma_iter_init(&iter, mhm->mhm_map);
3908 	for (i = 0; i < nmb; i++) {
3909 		unsigned int lim;
3910 
3911 		pas = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, i));
3912 		lim = uimin(MCX_CMDQ_MAILBOX_DATASIZE / sizeof(*pas), npages);
3913 
3914 		for (j = 0; j < lim; j++) {
3915 			be64enc(&pas[j], bus_dma_iter_addr(&iter));
3916 			bus_dma_iter_add(&iter, MCX_PAGE_SIZE);
3917 		}
3918 
3919 		npages -= lim;
3920 	}
3921 
3922 	mcx_cmdq_mboxes_sign(&mxm, nmb);
3923 
3924 	mcx_cmdq_post(sc, cqe, 0);
3925 	error = mcx_cmdq_poll(sc, cqe, 1000);
3926 	if (error != 0) {
3927 		printf(", manage pages timeout\n");
3928 		goto free;
3929 	}
3930 	error = mcx_cmdq_verify(cqe);
3931 	if (error != 0) {
3932 		printf(", manage pages reply corrupt\n");
3933 		goto free;
3934 	}
3935 
3936 	status = cqe->cq_output_data[0];
3937 	if (status != MCX_CQ_STATUS_OK) {
3938 		printf(", manage pages failed (%x)\n", status);
3939 		error = -1;
3940 		goto free;
3941 	}
3942 
3943 free:
3944 	mcx_dmamem_free(sc, &mxm);
3945 
3946 	return (error);
3947 }
3948 
3949 static int
3950 mcx_pages(struct mcx_softc *sc, struct mcx_hwmem *mhm, uint16_t type)
3951 {
3952 	int32_t npages;
3953 	uint16_t func_id;
3954 
3955 	if (mcx_query_pages(sc, type, &npages, &func_id) != 0) {
3956 		/* error printed by mcx_query_pages */
3957 		return (-1);
3958 	}
3959 
3960 	if (npages < 1)
3961 		return (0);
3962 
3963 	if (mcx_hwmem_alloc(sc, mhm, npages) != 0) {
3964 		printf(", unable to allocate hwmem\n");
3965 		return (-1);
3966 	}
3967 
3968 	if (mcx_add_pages(sc, mhm, func_id) != 0) {
3969 		printf(", unable to add hwmem\n");
3970 		goto free;
3971 	}
3972 
3973 	return (0);
3974 
3975 free:
3976 	mcx_hwmem_free(sc, mhm);
3977 
3978 	return (-1);
3979 }
3980 
3981 static int
3982 mcx_hca_max_caps(struct mcx_softc *sc)
3983 {
3984 	struct mcx_dmamem mxm;
3985 	struct mcx_cmdq_entry *cqe;
3986 	struct mcx_cmd_query_hca_cap_in *in;
3987 	struct mcx_cmd_query_hca_cap_out *out;
3988 	struct mcx_cmdq_mailbox *mb;
3989 	struct mcx_cap_device *hca;
3990 	uint8_t status;
3991 	uint8_t token = mcx_cmdq_token(sc);
3992 	int error;
3993 
3994 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
3995 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out) + MCX_HCA_CAP_LEN,
3996 	    token);
3997 
3998 	in = mcx_cmdq_in(cqe);
3999 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_HCA_CAP);
4000 	in->cmd_op_mod = htobe16(MCX_CMD_QUERY_HCA_CAP_MAX |
4001 	    MCX_CMD_QUERY_HCA_CAP_DEVICE);
4002 
4003 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, MCX_HCA_CAP_NMAILBOXES,
4004 	    &cqe->cq_output_ptr, token) != 0) {
4005 		printf(", unable to allocate query hca caps mailboxen\n");
4006 		return (-1);
4007 	}
4008 	mcx_cmdq_mboxes_sign(&mxm, MCX_HCA_CAP_NMAILBOXES);
4009 	mcx_cmdq_mboxes_sync(sc, &mxm, BUS_DMASYNC_PRERW);
4010 
4011 	mcx_cmdq_post(sc, cqe, 0);
4012 	error = mcx_cmdq_poll(sc, cqe, 1000);
4013 	mcx_cmdq_mboxes_sync(sc, &mxm, BUS_DMASYNC_POSTRW);
4014 
4015 	if (error != 0) {
4016 		printf(", query hca caps timeout\n");
4017 		goto free;
4018 	}
4019 	error = mcx_cmdq_verify(cqe);
4020 	if (error != 0) {
4021 		printf(", query hca caps reply corrupt\n");
4022 		goto free;
4023 	}
4024 
4025 	status = cqe->cq_output_data[0];
4026 	if (status != MCX_CQ_STATUS_OK) {
4027 		printf(", query hca caps failed (%x)\n", status);
4028 		error = -1;
4029 		goto free;
4030 	}
4031 
4032 	mb = mcx_cq_mbox(&mxm, 0);
4033 	hca = mcx_cq_mbox_data(mb);
4034 
4035 	if ((hca->port_type & MCX_CAP_DEVICE_PORT_TYPE)
4036 	    != MCX_CAP_DEVICE_PORT_TYPE_ETH) {
4037 		printf(", not in ethernet mode\n");
4038 		error = -1;
4039 		goto free;
4040 	}
4041 	if (hca->log_pg_sz > PAGE_SHIFT) {
4042 		printf(", minimum system page shift %u is too large\n",
4043 		    hca->log_pg_sz);
4044 		error = -1;
4045 		goto free;
4046 	}
4047 	/*
4048 	 * blueflame register is split into two buffers, and we must alternate
4049 	 * between the two of them.
4050 	 */
4051 	sc->sc_bf_size = (1 << hca->log_bf_reg_size) / 2;
4052 	sc->sc_max_rqt_size = (1 << hca->log_max_rqt_size);
4053 
4054 	if (hca->local_ca_ack_delay & MCX_CAP_DEVICE_MCAM_REG)
4055 		sc->sc_mcam_reg = 1;
4056 
4057 	sc->sc_mhz = be32dec(&hca->device_frequency_mhz);
4058 	sc->sc_khz = be32dec(&hca->device_frequency_khz);
4059 
4060 free:
4061 	mcx_dmamem_free(sc, &mxm);
4062 
4063 	return (error);
4064 }
4065 
4066 static int
4067 mcx_hca_set_caps(struct mcx_softc *sc)
4068 {
4069 	struct mcx_dmamem mxm;
4070 	struct mcx_cmdq_entry *cqe;
4071 	struct mcx_cmd_query_hca_cap_in *in;
4072 	struct mcx_cmd_query_hca_cap_out *out;
4073 	struct mcx_cmdq_mailbox *mb;
4074 	struct mcx_cap_device *hca;
4075 	uint8_t status;
4076 	uint8_t token = mcx_cmdq_token(sc);
4077 	int error;
4078 
4079 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4080 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out) + MCX_HCA_CAP_LEN,
4081 	    token);
4082 
4083 	in = mcx_cmdq_in(cqe);
4084 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_HCA_CAP);
4085 	in->cmd_op_mod = htobe16(MCX_CMD_QUERY_HCA_CAP_CURRENT |
4086 	    MCX_CMD_QUERY_HCA_CAP_DEVICE);
4087 
4088 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, MCX_HCA_CAP_NMAILBOXES,
4089 	    &cqe->cq_output_ptr, token) != 0) {
4090 		printf(", unable to allocate manage pages mailboxen\n");
4091 		return (-1);
4092 	}
4093 	mcx_cmdq_mboxes_sign(&mxm, MCX_HCA_CAP_NMAILBOXES);
4094 	mcx_cmdq_mboxes_sync(sc, &mxm, BUS_DMASYNC_PRERW);
4095 
4096 	mcx_cmdq_post(sc, cqe, 0);
4097 	error = mcx_cmdq_poll(sc, cqe, 1000);
4098 	mcx_cmdq_mboxes_sync(sc, &mxm, BUS_DMASYNC_POSTRW);
4099 
4100 	if (error != 0) {
4101 		printf(", query hca caps timeout\n");
4102 		goto free;
4103 	}
4104 	error = mcx_cmdq_verify(cqe);
4105 	if (error != 0) {
4106 		printf(", query hca caps reply corrupt\n");
4107 		goto free;
4108 	}
4109 
4110 	status = cqe->cq_output_data[0];
4111 	if (status != MCX_CQ_STATUS_OK) {
4112 		printf(", query hca caps failed (%x)\n", status);
4113 		error = -1;
4114 		goto free;
4115 	}
4116 
4117 	mb = mcx_cq_mbox(&mxm, 0);
4118 	hca = mcx_cq_mbox_data(mb);
4119 
4120 	hca->log_pg_sz = PAGE_SHIFT;
4121 
4122 free:
4123 	mcx_dmamem_free(sc, &mxm);
4124 
4125 	return (error);
4126 }
4127 
4128 
4129 static int
4130 mcx_init_hca(struct mcx_softc *sc)
4131 {
4132 	struct mcx_cmdq_entry *cqe;
4133 	struct mcx_cmd_init_hca_in *in;
4134 	struct mcx_cmd_init_hca_out *out;
4135 	int error;
4136 	uint8_t status;
4137 
4138 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4139 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
4140 
4141 	in = mcx_cmdq_in(cqe);
4142 	in->cmd_opcode = htobe16(MCX_CMD_INIT_HCA);
4143 	in->cmd_op_mod = htobe16(0);
4144 
4145 	mcx_cmdq_post(sc, cqe, 0);
4146 
4147 	error = mcx_cmdq_poll(sc, cqe, 1000);
4148 	if (error != 0) {
4149 		printf(", hca init timeout\n");
4150 		return (-1);
4151 	}
4152 	if (mcx_cmdq_verify(cqe) != 0) {
4153 		printf(", hca init command corrupt\n");
4154 		return (-1);
4155 	}
4156 
4157 	status = cqe->cq_output_data[0];
4158 	if (status != MCX_CQ_STATUS_OK) {
4159 		printf(", hca init failed (%x)\n", status);
4160 		return (-1);
4161 	}
4162 
4163 	return (0);
4164 }
4165 
4166 static int
4167 mcx_set_driver_version(struct mcx_softc *sc)
4168 {
4169 	struct mcx_dmamem mxm;
4170 	struct mcx_cmdq_entry *cqe;
4171 	struct mcx_cmd_set_driver_version_in *in;
4172 	struct mcx_cmd_set_driver_version_out *out;
4173 	int error;
4174 	int token;
4175 	uint8_t status;
4176 
4177 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4178 	token = mcx_cmdq_token(sc);
4179 	mcx_cmdq_init(sc, cqe, sizeof(*in) +
4180 	    sizeof(struct mcx_cmd_set_driver_version), sizeof(*out), token);
4181 
4182 	in = mcx_cmdq_in(cqe);
4183 	in->cmd_opcode = htobe16(MCX_CMD_SET_DRIVER_VERSION);
4184 	in->cmd_op_mod = htobe16(0);
4185 
4186 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
4187 	    &cqe->cq_input_ptr, token) != 0) {
4188 		printf(", unable to allocate set driver version mailboxen\n");
4189 		return (-1);
4190 	}
4191 	strlcpy(mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)),
4192 	    "OpenBSD,mcx,1.000.000000", MCX_CMDQ_MAILBOX_DATASIZE);
4193 
4194 	mcx_cmdq_mboxes_sign(&mxm, 1);
4195 	mcx_cmdq_post(sc, cqe, 0);
4196 
4197 	error = mcx_cmdq_poll(sc, cqe, 1000);
4198 	if (error != 0) {
4199 		printf(", set driver version timeout\n");
4200 		goto free;
4201 	}
4202 	if (mcx_cmdq_verify(cqe) != 0) {
4203 		printf(", set driver version command corrupt\n");
4204 		goto free;
4205 	}
4206 
4207 	status = cqe->cq_output_data[0];
4208 	if (status != MCX_CQ_STATUS_OK) {
4209 		printf(", set driver version failed (%x)\n", status);
4210 		error = -1;
4211 		goto free;
4212 	}
4213 
4214 free:
4215 	mcx_dmamem_free(sc, &mxm);
4216 
4217 	return (error);
4218 }
4219 
4220 static int
4221 mcx_iff(struct mcx_softc *sc)
4222 {
4223 	struct ifnet *ifp = &sc->sc_ec.ec_if;
4224 	struct mcx_dmamem mxm;
4225 	struct mcx_cmdq_entry *cqe;
4226 	struct mcx_cmd_modify_nic_vport_context_in *in;
4227 	struct mcx_cmd_modify_nic_vport_context_out *out;
4228 	struct mcx_nic_vport_ctx *ctx;
4229 	int error;
4230 	int token;
4231 	int insize;
4232 	uint32_t dest;
4233 
4234 	dest = MCX_FLOW_CONTEXT_DEST_TYPE_TABLE |
4235 	    sc->sc_rss_flow_table_id;
4236 
4237 	/* enable or disable the promisc flow */
4238 	if (ISSET(ifp->if_flags, IFF_PROMISC)) {
4239 		if (sc->sc_promisc_flow_enabled == 0) {
4240 			mcx_set_flow_table_entry_mac(sc,
4241 			    MCX_FLOW_GROUP_PROMISC, 0, NULL, dest);
4242 			sc->sc_promisc_flow_enabled = 1;
4243 		}
4244 	} else if (sc->sc_promisc_flow_enabled != 0) {
4245 		mcx_delete_flow_table_entry(sc, MCX_FLOW_GROUP_PROMISC, 0);
4246 		sc->sc_promisc_flow_enabled = 0;
4247 	}
4248 
4249 	/* enable or disable the all-multicast flow */
4250 	if (ISSET(ifp->if_flags, IFF_ALLMULTI)) {
4251 		if (sc->sc_allmulti_flow_enabled == 0) {
4252 			uint8_t mcast[ETHER_ADDR_LEN];
4253 
4254 			memset(mcast, 0, sizeof(mcast));
4255 			mcast[0] = 0x01;
4256 			mcx_set_flow_table_entry_mac(sc,
4257 			    MCX_FLOW_GROUP_ALLMULTI, 0, mcast, dest);
4258 			sc->sc_allmulti_flow_enabled = 1;
4259 		}
4260 	} else if (sc->sc_allmulti_flow_enabled != 0) {
4261 		mcx_delete_flow_table_entry(sc, MCX_FLOW_GROUP_ALLMULTI, 0);
4262 		sc->sc_allmulti_flow_enabled = 0;
4263 	}
4264 
4265 	insize = sizeof(struct mcx_nic_vport_ctx) + 240;
4266 
4267 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4268 	token = mcx_cmdq_token(sc);
4269 	mcx_cmdq_init(sc, cqe, sizeof(*in) + insize, sizeof(*out), token);
4270 
4271 	in = mcx_cmdq_in(cqe);
4272 	in->cmd_opcode = htobe16(MCX_CMD_MODIFY_NIC_VPORT_CONTEXT);
4273 	in->cmd_op_mod = htobe16(0);
4274 	in->cmd_field_select = htobe32(
4275 	    MCX_CMD_MODIFY_NIC_VPORT_CONTEXT_FIELD_PROMISC |
4276 	    MCX_CMD_MODIFY_NIC_VPORT_CONTEXT_FIELD_MTU);
4277 
4278 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1, &cqe->cq_input_ptr, token) != 0) {
4279 		printf(", unable to allocate modify "
4280 		    "nic vport context mailboxen\n");
4281 		return (-1);
4282 	}
4283 	ctx = (struct mcx_nic_vport_ctx *)
4284 	    (((char *)mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0))) + 240);
4285 	ctx->vp_mtu = htobe32(sc->sc_hardmtu);
4286 	/*
4287          * always leave promisc-all enabled on the vport since we
4288          * can't give it a vlan list, and we're already doing multicast
4289          * filtering in the flow table.
4290 	 */
4291 	ctx->vp_flags = htobe16(MCX_NIC_VPORT_CTX_PROMISC_ALL);
4292 
4293 	mcx_cmdq_mboxes_sign(&mxm, 1);
4294 	mcx_cmdq_post(sc, cqe, 0);
4295 
4296 	error = mcx_cmdq_poll(sc, cqe, 1000);
4297 	if (error != 0) {
4298 		printf(", modify nic vport context timeout\n");
4299 		goto free;
4300 	}
4301 	if (mcx_cmdq_verify(cqe) != 0) {
4302 		printf(", modify nic vport context command corrupt\n");
4303 		goto free;
4304 	}
4305 
4306 	out = mcx_cmdq_out(cqe);
4307 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4308 		printf(", modify nic vport context failed (%x, %x)\n",
4309 		    out->cmd_status, be32toh(out->cmd_syndrome));
4310 		error = -1;
4311 		goto free;
4312 	}
4313 
4314 free:
4315 	mcx_dmamem_free(sc, &mxm);
4316 
4317 	return (error);
4318 }
4319 
4320 static int
4321 mcx_alloc_uar(struct mcx_softc *sc, int *uar)
4322 {
4323 	struct mcx_cmdq_entry *cqe;
4324 	struct mcx_cmd_alloc_uar_in *in;
4325 	struct mcx_cmd_alloc_uar_out *out;
4326 	int error;
4327 
4328 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4329 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
4330 
4331 	in = mcx_cmdq_in(cqe);
4332 	in->cmd_opcode = htobe16(MCX_CMD_ALLOC_UAR);
4333 	in->cmd_op_mod = htobe16(0);
4334 
4335 	mcx_cmdq_post(sc, cqe, 0);
4336 
4337 	error = mcx_cmdq_poll(sc, cqe, 1000);
4338 	if (error != 0) {
4339 		printf(", alloc uar timeout\n");
4340 		return (-1);
4341 	}
4342 	if (mcx_cmdq_verify(cqe) != 0) {
4343 		printf(", alloc uar command corrupt\n");
4344 		return (-1);
4345 	}
4346 
4347 	out = mcx_cmdq_out(cqe);
4348 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4349 		printf(", alloc uar failed (%x)\n", out->cmd_status);
4350 		return (-1);
4351 	}
4352 
4353 	*uar = mcx_get_id(out->cmd_uar);
4354 	return (0);
4355 }
4356 
4357 static int
4358 mcx_create_eq(struct mcx_softc *sc, struct mcx_eq *eq, int uar,
4359     uint64_t events, int vector)
4360 {
4361 	struct mcx_cmdq_entry *cqe;
4362 	struct mcx_dmamem mxm;
4363 	struct mcx_cmd_create_eq_in *in;
4364 	struct mcx_cmd_create_eq_mb_in *mbin;
4365 	struct mcx_cmd_create_eq_out *out;
4366 	struct mcx_eq_entry *eqe;
4367 	int error;
4368 	uint64_t *pas;
4369 	int insize, npages, paslen, i, token;
4370 
4371 	eq->eq_cons = 0;
4372 
4373 	npages = howmany((1 << MCX_LOG_EQ_SIZE) * sizeof(struct mcx_eq_entry),
4374 	    MCX_PAGE_SIZE);
4375 	paslen = npages * sizeof(*pas);
4376 	insize = sizeof(struct mcx_cmd_create_eq_mb_in) + paslen;
4377 
4378 	if (mcx_dmamem_alloc(sc, &eq->eq_mem, npages * MCX_PAGE_SIZE,
4379 	    MCX_PAGE_SIZE) != 0) {
4380 		printf(", unable to allocate event queue memory\n");
4381 		return (-1);
4382 	}
4383 
4384 	eqe = (struct mcx_eq_entry *)MCX_DMA_KVA(&eq->eq_mem);
4385 	for (i = 0; i < (1 << MCX_LOG_EQ_SIZE); i++) {
4386 		eqe[i].eq_owner = MCX_EQ_ENTRY_OWNER_INIT;
4387 	}
4388 
4389 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4390 	token = mcx_cmdq_token(sc);
4391 	mcx_cmdq_init(sc, cqe, sizeof(*in) + insize, sizeof(*out), token);
4392 
4393 	in = mcx_cmdq_in(cqe);
4394 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_EQ);
4395 	in->cmd_op_mod = htobe16(0);
4396 
4397 	if (mcx_cmdq_mboxes_alloc(sc, &mxm,
4398 	    howmany(insize, MCX_CMDQ_MAILBOX_DATASIZE),
4399 	    &cqe->cq_input_ptr, token) != 0) {
4400 		printf(", unable to allocate create eq mailboxen\n");
4401 		goto free_eq;
4402 	}
4403 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
4404 	mbin->cmd_eq_ctx.eq_uar_size = htobe32(
4405 	    (MCX_LOG_EQ_SIZE << MCX_EQ_CTX_LOG_EQ_SIZE_SHIFT) | uar);
4406 	mbin->cmd_eq_ctx.eq_intr = vector;
4407 	mbin->cmd_event_bitmask = htobe64(events);
4408 
4409 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&eq->eq_mem),
4410 	    0, MCX_DMA_LEN(&eq->eq_mem), BUS_DMASYNC_PREREAD);
4411 
4412 	/* physical addresses follow the mailbox in data */
4413 	mcx_cmdq_mboxes_pas(&mxm, sizeof(*mbin), npages, &eq->eq_mem);
4414 	mcx_cmdq_mboxes_sign(&mxm, howmany(insize, MCX_CMDQ_MAILBOX_DATASIZE));
4415 	mcx_cmdq_post(sc, cqe, 0);
4416 
4417 	error = mcx_cmdq_poll(sc, cqe, 1000);
4418 	if (error != 0) {
4419 		printf(", create eq timeout\n");
4420 		goto free_mxm;
4421 	}
4422 	if (mcx_cmdq_verify(cqe) != 0) {
4423 		printf(", create eq command corrupt\n");
4424 		goto free_mxm;
4425 	}
4426 
4427 	out = mcx_cmdq_out(cqe);
4428 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4429 		printf(", create eq failed (%x, %x)\n", out->cmd_status,
4430 		    be32toh(out->cmd_syndrome));
4431 		goto free_mxm;
4432 	}
4433 
4434 	eq->eq_n = mcx_get_id(out->cmd_eqn);
4435 
4436 	mcx_dmamem_free(sc, &mxm);
4437 
4438 	mcx_arm_eq(sc, eq, uar);
4439 
4440 	return (0);
4441 
4442 free_mxm:
4443 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&eq->eq_mem),
4444 	    0, MCX_DMA_LEN(&eq->eq_mem), BUS_DMASYNC_POSTREAD);
4445 	mcx_dmamem_free(sc, &mxm);
4446 free_eq:
4447 	mcx_dmamem_free(sc, &eq->eq_mem);
4448 	return (-1);
4449 }
4450 
4451 static int
4452 mcx_alloc_pd(struct mcx_softc *sc)
4453 {
4454 	struct mcx_cmdq_entry *cqe;
4455 	struct mcx_cmd_alloc_pd_in *in;
4456 	struct mcx_cmd_alloc_pd_out *out;
4457 	int error;
4458 
4459 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4460 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
4461 
4462 	in = mcx_cmdq_in(cqe);
4463 	in->cmd_opcode = htobe16(MCX_CMD_ALLOC_PD);
4464 	in->cmd_op_mod = htobe16(0);
4465 
4466 	mcx_cmdq_post(sc, cqe, 0);
4467 
4468 	error = mcx_cmdq_poll(sc, cqe, 1000);
4469 	if (error != 0) {
4470 		printf(", alloc pd timeout\n");
4471 		return (-1);
4472 	}
4473 	if (mcx_cmdq_verify(cqe) != 0) {
4474 		printf(", alloc pd command corrupt\n");
4475 		return (-1);
4476 	}
4477 
4478 	out = mcx_cmdq_out(cqe);
4479 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4480 		printf(", alloc pd failed (%x)\n", out->cmd_status);
4481 		return (-1);
4482 	}
4483 
4484 	sc->sc_pd = mcx_get_id(out->cmd_pd);
4485 	return (0);
4486 }
4487 
4488 static int
4489 mcx_alloc_tdomain(struct mcx_softc *sc)
4490 {
4491 	struct mcx_cmdq_entry *cqe;
4492 	struct mcx_cmd_alloc_td_in *in;
4493 	struct mcx_cmd_alloc_td_out *out;
4494 	int error;
4495 
4496 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4497 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
4498 
4499 	in = mcx_cmdq_in(cqe);
4500 	in->cmd_opcode = htobe16(MCX_CMD_ALLOC_TRANSPORT_DOMAIN);
4501 	in->cmd_op_mod = htobe16(0);
4502 
4503 	mcx_cmdq_post(sc, cqe, 0);
4504 
4505 	error = mcx_cmdq_poll(sc, cqe, 1000);
4506 	if (error != 0) {
4507 		printf(", alloc transport domain timeout\n");
4508 		return (-1);
4509 	}
4510 	if (mcx_cmdq_verify(cqe) != 0) {
4511 		printf(", alloc transport domain command corrupt\n");
4512 		return (-1);
4513 	}
4514 
4515 	out = mcx_cmdq_out(cqe);
4516 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4517 		printf(", alloc transport domain failed (%x)\n",
4518 		    out->cmd_status);
4519 		return (-1);
4520 	}
4521 
4522 	sc->sc_tdomain = mcx_get_id(out->cmd_tdomain);
4523 	return (0);
4524 }
4525 
4526 static int
4527 mcx_query_nic_vport_context(struct mcx_softc *sc, uint8_t *enaddr)
4528 {
4529 	struct mcx_dmamem mxm;
4530 	struct mcx_cmdq_entry *cqe;
4531 	struct mcx_cmd_query_nic_vport_context_in *in;
4532 	struct mcx_cmd_query_nic_vport_context_out *out;
4533 	struct mcx_nic_vport_ctx *ctx;
4534 	uint8_t *addr;
4535 	int error, token, i;
4536 
4537 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4538 	token = mcx_cmdq_token(sc);
4539 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out) + sizeof(*ctx), token);
4540 
4541 	in = mcx_cmdq_in(cqe);
4542 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_NIC_VPORT_CONTEXT);
4543 	in->cmd_op_mod = htobe16(0);
4544 	in->cmd_allowed_list_type = 0;
4545 
4546 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
4547 	    &cqe->cq_output_ptr, token) != 0) {
4548 		printf(", unable to allocate "
4549 		    "query nic vport context mailboxen\n");
4550 		return (-1);
4551 	}
4552 	mcx_cmdq_mboxes_sign(&mxm, 1);
4553 	mcx_cmdq_post(sc, cqe, 0);
4554 
4555 	error = mcx_cmdq_poll(sc, cqe, 1000);
4556 	if (error != 0) {
4557 		printf(", query nic vport context timeout\n");
4558 		goto free;
4559 	}
4560 	if (mcx_cmdq_verify(cqe) != 0) {
4561 		printf(", query nic vport context command corrupt\n");
4562 		goto free;
4563 	}
4564 
4565 	out = mcx_cmdq_out(cqe);
4566 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4567 		printf(", query nic vport context failed (%x, %x)\n",
4568 		    out->cmd_status, be32toh(out->cmd_syndrome));
4569 		error = -1;
4570 		goto free;
4571 	}
4572 
4573 	ctx = (struct mcx_nic_vport_ctx *)
4574 	    mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
4575 	addr = (uint8_t *)&ctx->vp_perm_addr;
4576 	for (i = 0; i < ETHER_ADDR_LEN; i++) {
4577 		enaddr[i] = addr[i + 2];
4578 	}
4579 free:
4580 	mcx_dmamem_free(sc, &mxm);
4581 
4582 	return (error);
4583 }
4584 
4585 static int
4586 mcx_query_special_contexts(struct mcx_softc *sc)
4587 {
4588 	struct mcx_cmdq_entry *cqe;
4589 	struct mcx_cmd_query_special_ctx_in *in;
4590 	struct mcx_cmd_query_special_ctx_out *out;
4591 	int error;
4592 
4593 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4594 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
4595 
4596 	in = mcx_cmdq_in(cqe);
4597 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_SPECIAL_CONTEXTS);
4598 	in->cmd_op_mod = htobe16(0);
4599 
4600 	mcx_cmdq_post(sc, cqe, 0);
4601 
4602 	error = mcx_cmdq_poll(sc, cqe, 1000);
4603 	if (error != 0) {
4604 		printf(", query special contexts timeout\n");
4605 		return (-1);
4606 	}
4607 	if (mcx_cmdq_verify(cqe) != 0) {
4608 		printf(", query special contexts command corrupt\n");
4609 		return (-1);
4610 	}
4611 
4612 	out = mcx_cmdq_out(cqe);
4613 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4614 		printf(", query special contexts failed (%x)\n",
4615 		    out->cmd_status);
4616 		return (-1);
4617 	}
4618 
4619 	sc->sc_lkey = be32toh(out->cmd_resd_lkey);
4620 	return (0);
4621 }
4622 
4623 static int
4624 mcx_set_port_mtu(struct mcx_softc *sc, int mtu)
4625 {
4626 	struct mcx_reg_pmtu pmtu;
4627 	int error;
4628 
4629 	/* read max mtu */
4630 	memset(&pmtu, 0, sizeof(pmtu));
4631 	pmtu.rp_local_port = 1;
4632 	error = mcx_access_hca_reg(sc, MCX_REG_PMTU, MCX_REG_OP_READ, &pmtu,
4633 	    sizeof(pmtu));
4634 	if (error != 0) {
4635 		printf(", unable to get port MTU\n");
4636 		return error;
4637 	}
4638 
4639 	mtu = uimin(mtu, be16toh(pmtu.rp_max_mtu));
4640 	pmtu.rp_admin_mtu = htobe16(mtu);
4641 	error = mcx_access_hca_reg(sc, MCX_REG_PMTU, MCX_REG_OP_WRITE, &pmtu,
4642 	    sizeof(pmtu));
4643 	if (error != 0) {
4644 		printf(", unable to set port MTU\n");
4645 		return error;
4646 	}
4647 
4648 	sc->sc_hardmtu = mtu;
4649 	sc->sc_rxbufsz = roundup(mtu + ETHER_ALIGN, sizeof(long));
4650 	return 0;
4651 }
4652 
4653 static int
4654 mcx_create_cq(struct mcx_softc *sc, struct mcx_cq *cq, int uar, int db, int eqn)
4655 {
4656 	struct mcx_cmdq_entry *cmde;
4657 	struct mcx_cq_entry *cqe;
4658 	struct mcx_dmamem mxm;
4659 	struct mcx_cmd_create_cq_in *in;
4660 	struct mcx_cmd_create_cq_mb_in *mbin;
4661 	struct mcx_cmd_create_cq_out *out;
4662 	int error;
4663 	uint64_t *pas;
4664 	int insize, npages, paslen, i, token;
4665 
4666 	cq->cq_doorbell = MCX_CQ_DOORBELL_BASE + (MCX_CQ_DOORBELL_STRIDE * db);
4667 
4668 	npages = howmany((1 << MCX_LOG_CQ_SIZE) * sizeof(struct mcx_cq_entry),
4669 	    MCX_PAGE_SIZE);
4670 	paslen = npages * sizeof(*pas);
4671 	insize = sizeof(struct mcx_cmd_create_cq_mb_in) + paslen;
4672 
4673 	if (mcx_dmamem_alloc(sc, &cq->cq_mem, npages * MCX_PAGE_SIZE,
4674 	    MCX_PAGE_SIZE) != 0) {
4675 		printf("%s: unable to allocate completion queue memory\n",
4676 		    DEVNAME(sc));
4677 		return (-1);
4678 	}
4679 	cqe = MCX_DMA_KVA(&cq->cq_mem);
4680 	for (i = 0; i < (1 << MCX_LOG_CQ_SIZE); i++) {
4681 		cqe[i].cq_opcode_owner = MCX_CQ_ENTRY_FLAG_OWNER;
4682 	}
4683 
4684 	cmde = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4685 	token = mcx_cmdq_token(sc);
4686 	mcx_cmdq_init(sc, cmde, sizeof(*in) + insize, sizeof(*out), token);
4687 
4688 	in = mcx_cmdq_in(cmde);
4689 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_CQ);
4690 	in->cmd_op_mod = htobe16(0);
4691 
4692 	if (mcx_cmdq_mboxes_alloc(sc, &mxm,
4693 	    howmany(insize, MCX_CMDQ_MAILBOX_DATASIZE),
4694 	    &cmde->cq_input_ptr, token) != 0) {
4695 		printf("%s: unable to allocate create cq mailboxen\n",
4696 		    DEVNAME(sc));
4697 		goto free_cq;
4698 	}
4699 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
4700 	mbin->cmd_cq_ctx.cq_uar_size = htobe32(
4701 	    (MCX_LOG_CQ_SIZE << MCX_CQ_CTX_LOG_CQ_SIZE_SHIFT) | uar);
4702 	mbin->cmd_cq_ctx.cq_eqn = htobe32(eqn);
4703 	mbin->cmd_cq_ctx.cq_period_max_count = htobe32(
4704 	    (MCX_CQ_MOD_PERIOD << MCX_CQ_CTX_PERIOD_SHIFT) |
4705 	    MCX_CQ_MOD_COUNTER);
4706 	mbin->cmd_cq_ctx.cq_doorbell = htobe64(
4707 	    MCX_DMA_DVA(&sc->sc_doorbell_mem) + cq->cq_doorbell);
4708 
4709 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&cq->cq_mem),
4710 	    0, MCX_DMA_LEN(&cq->cq_mem), BUS_DMASYNC_PREREAD);
4711 
4712 	/* physical addresses follow the mailbox in data */
4713 	mcx_cmdq_mboxes_pas(&mxm, sizeof(*mbin), npages, &cq->cq_mem);
4714 	mcx_cmdq_post(sc, cmde, 0);
4715 
4716 	error = mcx_cmdq_poll(sc, cmde, 1000);
4717 	if (error != 0) {
4718 		printf("%s: create cq timeout\n", DEVNAME(sc));
4719 		goto free_mxm;
4720 	}
4721 	if (mcx_cmdq_verify(cmde) != 0) {
4722 		printf("%s: create cq command corrupt\n", DEVNAME(sc));
4723 		goto free_mxm;
4724 	}
4725 
4726 	out = mcx_cmdq_out(cmde);
4727 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4728 		printf("%s: create cq failed (%x, %x)\n", DEVNAME(sc),
4729 		    out->cmd_status, be32toh(out->cmd_syndrome));
4730 		goto free_mxm;
4731 	}
4732 
4733 	cq->cq_n = mcx_get_id(out->cmd_cqn);
4734 	cq->cq_cons = 0;
4735 	cq->cq_count = 0;
4736 
4737 	mcx_dmamem_free(sc, &mxm);
4738 
4739 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_doorbell_mem),
4740 	    cq->cq_doorbell, sizeof(struct mcx_cq_doorbell),
4741 	    BUS_DMASYNC_PREWRITE);
4742 
4743 	mcx_arm_cq(sc, cq, uar);
4744 
4745 	return (0);
4746 
4747 free_mxm:
4748 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&cq->cq_mem),
4749 	    0, MCX_DMA_LEN(&cq->cq_mem), BUS_DMASYNC_POSTREAD);
4750 	mcx_dmamem_free(sc, &mxm);
4751 free_cq:
4752 	mcx_dmamem_free(sc, &cq->cq_mem);
4753 	return (-1);
4754 }
4755 
4756 static int
4757 mcx_destroy_cq(struct mcx_softc *sc, struct mcx_cq *cq)
4758 {
4759 	struct mcx_cmdq_entry *cqe;
4760 	struct mcx_cmd_destroy_cq_in *in;
4761 	struct mcx_cmd_destroy_cq_out *out;
4762 	int error;
4763 	int token;
4764 
4765 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4766 	token = mcx_cmdq_token(sc);
4767 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), token);
4768 
4769 	in = mcx_cmdq_in(cqe);
4770 	in->cmd_opcode = htobe16(MCX_CMD_DESTROY_CQ);
4771 	in->cmd_op_mod = htobe16(0);
4772 	in->cmd_cqn = htobe32(cq->cq_n);
4773 
4774 	mcx_cmdq_post(sc, cqe, 0);
4775 	error = mcx_cmdq_poll(sc, cqe, 1000);
4776 	if (error != 0) {
4777 		printf("%s: destroy cq timeout\n", DEVNAME(sc));
4778 		return error;
4779 	}
4780 	if (mcx_cmdq_verify(cqe) != 0) {
4781 		printf("%s: destroy cq command corrupt\n", DEVNAME(sc));
4782 		return error;
4783 	}
4784 
4785 	out = mcx_cmdq_out(cqe);
4786 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4787 		printf("%s: destroy cq failed (%x, %x)\n", DEVNAME(sc),
4788 		    out->cmd_status, be32toh(out->cmd_syndrome));
4789 		return -1;
4790 	}
4791 
4792 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_doorbell_mem),
4793 	    cq->cq_doorbell, sizeof(struct mcx_cq_doorbell),
4794 	    BUS_DMASYNC_POSTWRITE);
4795 
4796 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&cq->cq_mem),
4797 	    0, MCX_DMA_LEN(&cq->cq_mem), BUS_DMASYNC_POSTREAD);
4798 	mcx_dmamem_free(sc, &cq->cq_mem);
4799 
4800 	cq->cq_n = 0;
4801 	cq->cq_cons = 0;
4802 	cq->cq_count = 0;
4803 	return 0;
4804 }
4805 
4806 static int
4807 mcx_create_rq(struct mcx_softc *sc, struct mcx_rx *rx, int db, int cqn)
4808 {
4809 	struct mcx_cmdq_entry *cqe;
4810 	struct mcx_dmamem mxm;
4811 	struct mcx_cmd_create_rq_in *in;
4812 	struct mcx_cmd_create_rq_out *out;
4813 	struct mcx_rq_ctx *mbin;
4814 	int error;
4815 	uint64_t *pas;
4816 	uint32_t rq_flags;
4817 	int insize, npages, paslen, token;
4818 
4819 	rx->rx_doorbell = MCX_WQ_DOORBELL_BASE +
4820 	    (db * MCX_WQ_DOORBELL_STRIDE);
4821 
4822 	npages = howmany((1 << MCX_LOG_RQ_SIZE) * sizeof(struct mcx_rq_entry),
4823 	    MCX_PAGE_SIZE);
4824 	paslen = npages * sizeof(*pas);
4825 	insize = 0x10 + sizeof(struct mcx_rq_ctx) + paslen;
4826 
4827 	if (mcx_dmamem_alloc(sc, &rx->rx_rq_mem, npages * MCX_PAGE_SIZE,
4828 	    MCX_PAGE_SIZE) != 0) {
4829 		printf("%s: unable to allocate receive queue memory\n",
4830 		    DEVNAME(sc));
4831 		return (-1);
4832 	}
4833 
4834 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4835 	token = mcx_cmdq_token(sc);
4836 	mcx_cmdq_init(sc, cqe, sizeof(*in) + insize, sizeof(*out), token);
4837 
4838 	in = mcx_cmdq_in(cqe);
4839 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_RQ);
4840 	in->cmd_op_mod = htobe16(0);
4841 
4842 	if (mcx_cmdq_mboxes_alloc(sc, &mxm,
4843 	    howmany(insize, MCX_CMDQ_MAILBOX_DATASIZE),
4844 	    &cqe->cq_input_ptr, token) != 0) {
4845 		printf("%s: unable to allocate create rq mailboxen\n",
4846 		    DEVNAME(sc));
4847 		goto free_rq;
4848 	}
4849 	mbin = (struct mcx_rq_ctx *)
4850 	    (((char *)mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0))) + 0x10);
4851 	rq_flags = MCX_RQ_CTX_RLKEY;
4852 	mbin->rq_flags = htobe32(rq_flags);
4853 	mbin->rq_cqn = htobe32(cqn);
4854 	mbin->rq_wq.wq_type = MCX_WQ_CTX_TYPE_CYCLIC;
4855 	mbin->rq_wq.wq_pd = htobe32(sc->sc_pd);
4856 	mbin->rq_wq.wq_doorbell = htobe64(MCX_DMA_DVA(&sc->sc_doorbell_mem) +
4857 	    rx->rx_doorbell);
4858 	mbin->rq_wq.wq_log_stride = htobe16(4);
4859 	mbin->rq_wq.wq_log_size = MCX_LOG_RQ_SIZE;
4860 
4861 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&rx->rx_rq_mem),
4862 	    0, MCX_DMA_LEN(&rx->rx_rq_mem), BUS_DMASYNC_PREWRITE);
4863 
4864 	/* physical addresses follow the mailbox in data */
4865 	mcx_cmdq_mboxes_pas(&mxm, sizeof(*mbin) + 0x10, npages, &rx->rx_rq_mem);
4866 	mcx_cmdq_post(sc, cqe, 0);
4867 
4868 	error = mcx_cmdq_poll(sc, cqe, 1000);
4869 	if (error != 0) {
4870 		printf("%s: create rq timeout\n", DEVNAME(sc));
4871 		goto free_mxm;
4872 	}
4873 	if (mcx_cmdq_verify(cqe) != 0) {
4874 		printf("%s: create rq command corrupt\n", DEVNAME(sc));
4875 		goto free_mxm;
4876 	}
4877 
4878 	out = mcx_cmdq_out(cqe);
4879 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4880 		printf("%s: create rq failed (%x, %x)\n", DEVNAME(sc),
4881 		    out->cmd_status, be32toh(out->cmd_syndrome));
4882 		goto free_mxm;
4883 	}
4884 
4885 	rx->rx_rqn = mcx_get_id(out->cmd_rqn);
4886 
4887 	mcx_dmamem_free(sc, &mxm);
4888 
4889 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_doorbell_mem),
4890 	    rx->rx_doorbell, sizeof(uint32_t), BUS_DMASYNC_PREWRITE);
4891 
4892 	return (0);
4893 
4894 free_mxm:
4895 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&rx->rx_rq_mem),
4896 	    0, MCX_DMA_LEN(&rx->rx_rq_mem), BUS_DMASYNC_POSTWRITE);
4897 	mcx_dmamem_free(sc, &mxm);
4898 free_rq:
4899 	mcx_dmamem_free(sc, &rx->rx_rq_mem);
4900 	return (-1);
4901 }
4902 
4903 static int
4904 mcx_ready_rq(struct mcx_softc *sc, struct mcx_rx *rx)
4905 {
4906 	struct mcx_cmdq_entry *cqe;
4907 	struct mcx_dmamem mxm;
4908 	struct mcx_cmd_modify_rq_in *in;
4909 	struct mcx_cmd_modify_rq_mb_in *mbin;
4910 	struct mcx_cmd_modify_rq_out *out;
4911 	int error;
4912 	int token;
4913 
4914 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4915 	token = mcx_cmdq_token(sc);
4916 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
4917 	    sizeof(*out), token);
4918 
4919 	in = mcx_cmdq_in(cqe);
4920 	in->cmd_opcode = htobe16(MCX_CMD_MODIFY_RQ);
4921 	in->cmd_op_mod = htobe16(0);
4922 	in->cmd_rq_state = htobe32((MCX_QUEUE_STATE_RST << 28) | rx->rx_rqn);
4923 
4924 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
4925 	    &cqe->cq_input_ptr, token) != 0) {
4926 		printf("%s: unable to allocate modify rq mailbox\n",
4927 		    DEVNAME(sc));
4928 		return (-1);
4929 	}
4930 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
4931 	mbin->cmd_rq_ctx.rq_flags = htobe32(
4932 	    MCX_QUEUE_STATE_RDY << MCX_RQ_CTX_STATE_SHIFT);
4933 
4934 	mcx_cmdq_mboxes_sign(&mxm, 1);
4935 	mcx_cmdq_post(sc, cqe, 0);
4936 	error = mcx_cmdq_poll(sc, cqe, 1000);
4937 	if (error != 0) {
4938 		printf("%s: modify rq timeout\n", DEVNAME(sc));
4939 		goto free;
4940 	}
4941 	if (mcx_cmdq_verify(cqe) != 0) {
4942 		printf("%s: modify rq command corrupt\n", DEVNAME(sc));
4943 		goto free;
4944 	}
4945 
4946 	out = mcx_cmdq_out(cqe);
4947 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4948 		printf("%s: modify rq failed (%x, %x)\n", DEVNAME(sc),
4949 		    out->cmd_status, be32toh(out->cmd_syndrome));
4950 		error = -1;
4951 		goto free;
4952 	}
4953 
4954 free:
4955 	mcx_dmamem_free(sc, &mxm);
4956 	return (error);
4957 }
4958 
4959 static int
4960 mcx_destroy_rq(struct mcx_softc *sc, struct mcx_rx *rx)
4961 {
4962 	struct mcx_cmdq_entry *cqe;
4963 	struct mcx_cmd_destroy_rq_in *in;
4964 	struct mcx_cmd_destroy_rq_out *out;
4965 	int error;
4966 	int token;
4967 
4968 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4969 	token = mcx_cmdq_token(sc);
4970 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), token);
4971 
4972 	in = mcx_cmdq_in(cqe);
4973 	in->cmd_opcode = htobe16(MCX_CMD_DESTROY_RQ);
4974 	in->cmd_op_mod = htobe16(0);
4975 	in->cmd_rqn = htobe32(rx->rx_rqn);
4976 
4977 	mcx_cmdq_post(sc, cqe, 0);
4978 	error = mcx_cmdq_poll(sc, cqe, 1000);
4979 	if (error != 0) {
4980 		printf("%s: destroy rq timeout\n", DEVNAME(sc));
4981 		return error;
4982 	}
4983 	if (mcx_cmdq_verify(cqe) != 0) {
4984 		printf("%s: destroy rq command corrupt\n", DEVNAME(sc));
4985 		return error;
4986 	}
4987 
4988 	out = mcx_cmdq_out(cqe);
4989 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4990 		printf("%s: destroy rq failed (%x, %x)\n", DEVNAME(sc),
4991 		    out->cmd_status, be32toh(out->cmd_syndrome));
4992 		return -1;
4993 	}
4994 
4995 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_doorbell_mem),
4996 	    rx->rx_doorbell, sizeof(uint32_t), BUS_DMASYNC_POSTWRITE);
4997 
4998 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&rx->rx_rq_mem),
4999 	    0, MCX_DMA_LEN(&rx->rx_rq_mem), BUS_DMASYNC_POSTWRITE);
5000 	mcx_dmamem_free(sc, &rx->rx_rq_mem);
5001 
5002 	rx->rx_rqn = 0;
5003 	return 0;
5004 }
5005 
5006 static int
5007 mcx_create_tir_direct(struct mcx_softc *sc, struct mcx_rx *rx, int *tirn)
5008 {
5009 	struct mcx_cmdq_entry *cqe;
5010 	struct mcx_dmamem mxm;
5011 	struct mcx_cmd_create_tir_in *in;
5012 	struct mcx_cmd_create_tir_mb_in *mbin;
5013 	struct mcx_cmd_create_tir_out *out;
5014 	int error;
5015 	int token;
5016 
5017 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5018 	token = mcx_cmdq_token(sc);
5019 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
5020 	    sizeof(*out), token);
5021 
5022 	in = mcx_cmdq_in(cqe);
5023 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_TIR);
5024 	in->cmd_op_mod = htobe16(0);
5025 
5026 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
5027 	    &cqe->cq_input_ptr, token) != 0) {
5028 		printf("%s: unable to allocate create tir mailbox\n",
5029 		    DEVNAME(sc));
5030 		return (-1);
5031 	}
5032 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5033 	/* leave disp_type = 0, so packets get sent to the inline rqn */
5034 	mbin->cmd_inline_rqn = htobe32(rx->rx_rqn);
5035 	mbin->cmd_tdomain = htobe32(sc->sc_tdomain);
5036 
5037 	mcx_cmdq_post(sc, cqe, 0);
5038 	error = mcx_cmdq_poll(sc, cqe, 1000);
5039 	if (error != 0) {
5040 		printf("%s: create tir timeout\n", DEVNAME(sc));
5041 		goto free;
5042 	}
5043 	if (mcx_cmdq_verify(cqe) != 0) {
5044 		printf("%s: create tir command corrupt\n", DEVNAME(sc));
5045 		goto free;
5046 	}
5047 
5048 	out = mcx_cmdq_out(cqe);
5049 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5050 		printf("%s: create tir failed (%x, %x)\n", DEVNAME(sc),
5051 		    out->cmd_status, be32toh(out->cmd_syndrome));
5052 		error = -1;
5053 		goto free;
5054 	}
5055 
5056 	*tirn = mcx_get_id(out->cmd_tirn);
5057 free:
5058 	mcx_dmamem_free(sc, &mxm);
5059 	return (error);
5060 }
5061 
5062 static int
5063 mcx_create_tir_indirect(struct mcx_softc *sc, int rqtn, uint32_t hash_sel,
5064     int *tirn)
5065 {
5066 	struct mcx_cmdq_entry *cqe;
5067 	struct mcx_dmamem mxm;
5068 	struct mcx_cmd_create_tir_in *in;
5069 	struct mcx_cmd_create_tir_mb_in *mbin;
5070 	struct mcx_cmd_create_tir_out *out;
5071 	int error;
5072 	int token;
5073 
5074 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5075 	token = mcx_cmdq_token(sc);
5076 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
5077 	    sizeof(*out), token);
5078 
5079 	in = mcx_cmdq_in(cqe);
5080 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_TIR);
5081 	in->cmd_op_mod = htobe16(0);
5082 
5083 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
5084 	    &cqe->cq_input_ptr, token) != 0) {
5085 		printf("%s: unable to allocate create tir mailbox\n",
5086 		    DEVNAME(sc));
5087 		return (-1);
5088 	}
5089 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5090 	mbin->cmd_disp_type = htobe32(MCX_TIR_CTX_DISP_TYPE_INDIRECT
5091 	    << MCX_TIR_CTX_DISP_TYPE_SHIFT);
5092 	mbin->cmd_indir_table = htobe32(rqtn);
5093 	mbin->cmd_tdomain = htobe32(sc->sc_tdomain |
5094 	    MCX_TIR_CTX_HASH_TOEPLITZ << MCX_TIR_CTX_HASH_SHIFT);
5095 	mbin->cmd_rx_hash_sel_outer = htobe32(hash_sel);
5096 	stoeplitz_to_key(&mbin->cmd_rx_hash_key,
5097 	    sizeof(mbin->cmd_rx_hash_key));
5098 
5099 	mcx_cmdq_post(sc, cqe, 0);
5100 	error = mcx_cmdq_poll(sc, cqe, 1000);
5101 	if (error != 0) {
5102 		printf("%s: create tir timeout\n", DEVNAME(sc));
5103 		goto free;
5104 	}
5105 	if (mcx_cmdq_verify(cqe) != 0) {
5106 		printf("%s: create tir command corrupt\n", DEVNAME(sc));
5107 		goto free;
5108 	}
5109 
5110 	out = mcx_cmdq_out(cqe);
5111 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5112 		printf("%s: create tir failed (%x, %x)\n", DEVNAME(sc),
5113 		    out->cmd_status, be32toh(out->cmd_syndrome));
5114 		error = -1;
5115 		goto free;
5116 	}
5117 
5118 	*tirn = mcx_get_id(out->cmd_tirn);
5119 free:
5120 	mcx_dmamem_free(sc, &mxm);
5121 	return (error);
5122 }
5123 
5124 static int
5125 mcx_destroy_tir(struct mcx_softc *sc, int tirn)
5126 {
5127 	struct mcx_cmdq_entry *cqe;
5128 	struct mcx_cmd_destroy_tir_in *in;
5129 	struct mcx_cmd_destroy_tir_out *out;
5130 	int error;
5131 	int token;
5132 
5133 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5134 	token = mcx_cmdq_token(sc);
5135 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), token);
5136 
5137 	in = mcx_cmdq_in(cqe);
5138 	in->cmd_opcode = htobe16(MCX_CMD_DESTROY_TIR);
5139 	in->cmd_op_mod = htobe16(0);
5140 	in->cmd_tirn = htobe32(tirn);
5141 
5142 	mcx_cmdq_post(sc, cqe, 0);
5143 	error = mcx_cmdq_poll(sc, cqe, 1000);
5144 	if (error != 0) {
5145 		printf("%s: destroy tir timeout\n", DEVNAME(sc));
5146 		return error;
5147 	}
5148 	if (mcx_cmdq_verify(cqe) != 0) {
5149 		printf("%s: destroy tir command corrupt\n", DEVNAME(sc));
5150 		return error;
5151 	}
5152 
5153 	out = mcx_cmdq_out(cqe);
5154 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5155 		printf("%s: destroy tir failed (%x, %x)\n", DEVNAME(sc),
5156 		    out->cmd_status, be32toh(out->cmd_syndrome));
5157 		return -1;
5158 	}
5159 
5160 	return (0);
5161 }
5162 
5163 static int
5164 mcx_create_sq(struct mcx_softc *sc, struct mcx_tx *tx, int uar, int db,
5165     int cqn)
5166 {
5167 	struct mcx_cmdq_entry *cqe;
5168 	struct mcx_dmamem mxm;
5169 	struct mcx_cmd_create_sq_in *in;
5170 	struct mcx_sq_ctx *mbin;
5171 	struct mcx_cmd_create_sq_out *out;
5172 	int error;
5173 	uint64_t *pas;
5174 	int insize, npages, paslen, token;
5175 
5176 	tx->tx_doorbell = MCX_WQ_DOORBELL_BASE +
5177 	    (db * MCX_WQ_DOORBELL_STRIDE) + 4;
5178 
5179 	npages = howmany((1 << MCX_LOG_SQ_SIZE) * sizeof(struct mcx_sq_entry),
5180 	    MCX_PAGE_SIZE);
5181 	paslen = npages * sizeof(*pas);
5182 	insize = sizeof(struct mcx_sq_ctx) + paslen;
5183 
5184 	if (mcx_dmamem_alloc(sc, &tx->tx_sq_mem, npages * MCX_PAGE_SIZE,
5185 	    MCX_PAGE_SIZE) != 0) {
5186 		printf("%s: unable to allocate send queue memory\n",
5187 		    DEVNAME(sc));
5188 		return (-1);
5189 	}
5190 
5191 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5192 	token = mcx_cmdq_token(sc);
5193 	mcx_cmdq_init(sc, cqe, sizeof(*in) + insize + paslen, sizeof(*out),
5194 	    token);
5195 
5196 	in = mcx_cmdq_in(cqe);
5197 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_SQ);
5198 	in->cmd_op_mod = htobe16(0);
5199 
5200 	if (mcx_cmdq_mboxes_alloc(sc, &mxm,
5201 	    howmany(insize, MCX_CMDQ_MAILBOX_DATASIZE),
5202 	    &cqe->cq_input_ptr, token) != 0) {
5203 		printf("%s: unable to allocate create sq mailboxen\n",
5204 		    DEVNAME(sc));
5205 		goto free_sq;
5206 	}
5207 	mbin = (struct mcx_sq_ctx *)
5208 	    (((char *)mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0))) + 0x10);
5209 	mbin->sq_flags = htobe32(MCX_SQ_CTX_RLKEY |
5210 	    (1 << MCX_SQ_CTX_MIN_WQE_INLINE_SHIFT));
5211 	mbin->sq_cqn = htobe32(cqn);
5212 	mbin->sq_tis_lst_sz = htobe32(1 << MCX_SQ_CTX_TIS_LST_SZ_SHIFT);
5213 	mbin->sq_tis_num = htobe32(sc->sc_tis);
5214 	mbin->sq_wq.wq_type = MCX_WQ_CTX_TYPE_CYCLIC;
5215 	mbin->sq_wq.wq_pd = htobe32(sc->sc_pd);
5216 	mbin->sq_wq.wq_uar_page = htobe32(uar);
5217 	mbin->sq_wq.wq_doorbell = htobe64(MCX_DMA_DVA(&sc->sc_doorbell_mem) +
5218 	    tx->tx_doorbell);
5219 	mbin->sq_wq.wq_log_stride = htobe16(MCX_LOG_SQ_ENTRY_SIZE);
5220 	mbin->sq_wq.wq_log_size = MCX_LOG_SQ_SIZE;
5221 
5222 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&tx->tx_sq_mem),
5223 	    0, MCX_DMA_LEN(&tx->tx_sq_mem), BUS_DMASYNC_PREWRITE);
5224 
5225 	/* physical addresses follow the mailbox in data */
5226 	mcx_cmdq_mboxes_pas(&mxm, sizeof(*mbin) + 0x10,
5227 	    npages, &tx->tx_sq_mem);
5228 	mcx_cmdq_post(sc, cqe, 0);
5229 
5230 	error = mcx_cmdq_poll(sc, cqe, 1000);
5231 	if (error != 0) {
5232 		printf("%s: create sq timeout\n", DEVNAME(sc));
5233 		goto free_mxm;
5234 	}
5235 	if (mcx_cmdq_verify(cqe) != 0) {
5236 		printf("%s: create sq command corrupt\n", DEVNAME(sc));
5237 		goto free_mxm;
5238 	}
5239 
5240 	out = mcx_cmdq_out(cqe);
5241 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5242 		printf("%s: create sq failed (%x, %x)\n", DEVNAME(sc),
5243 		    out->cmd_status, be32toh(out->cmd_syndrome));
5244 		goto free_mxm;
5245 	}
5246 
5247 	tx->tx_uar = uar;
5248 	tx->tx_sqn = mcx_get_id(out->cmd_sqn);
5249 
5250 	mcx_dmamem_free(sc, &mxm);
5251 
5252 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_doorbell_mem),
5253 	    tx->tx_doorbell, sizeof(uint32_t), BUS_DMASYNC_PREWRITE);
5254 
5255 	return (0);
5256 
5257 free_mxm:
5258 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&tx->tx_sq_mem),
5259 	    0, MCX_DMA_LEN(&tx->tx_sq_mem), BUS_DMASYNC_POSTWRITE);
5260 	mcx_dmamem_free(sc, &mxm);
5261 free_sq:
5262 	mcx_dmamem_free(sc, &tx->tx_sq_mem);
5263 	return (-1);
5264 }
5265 
5266 static int
5267 mcx_destroy_sq(struct mcx_softc *sc, struct mcx_tx *tx)
5268 {
5269 	struct mcx_cmdq_entry *cqe;
5270 	struct mcx_cmd_destroy_sq_in *in;
5271 	struct mcx_cmd_destroy_sq_out *out;
5272 	int error;
5273 	int token;
5274 
5275 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5276 	token = mcx_cmdq_token(sc);
5277 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), token);
5278 
5279 	in = mcx_cmdq_in(cqe);
5280 	in->cmd_opcode = htobe16(MCX_CMD_DESTROY_SQ);
5281 	in->cmd_op_mod = htobe16(0);
5282 	in->cmd_sqn = htobe32(tx->tx_sqn);
5283 
5284 	mcx_cmdq_post(sc, cqe, 0);
5285 	error = mcx_cmdq_poll(sc, cqe, 1000);
5286 	if (error != 0) {
5287 		printf("%s: destroy sq timeout\n", DEVNAME(sc));
5288 		return error;
5289 	}
5290 	if (mcx_cmdq_verify(cqe) != 0) {
5291 		printf("%s: destroy sq command corrupt\n", DEVNAME(sc));
5292 		return error;
5293 	}
5294 
5295 	out = mcx_cmdq_out(cqe);
5296 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5297 		printf("%s: destroy sq failed (%x, %x)\n", DEVNAME(sc),
5298 		    out->cmd_status, be32toh(out->cmd_syndrome));
5299 		return -1;
5300 	}
5301 
5302 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_doorbell_mem),
5303 	    tx->tx_doorbell, sizeof(uint32_t), BUS_DMASYNC_POSTWRITE);
5304 
5305 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&tx->tx_sq_mem),
5306 	    0, MCX_DMA_LEN(&tx->tx_sq_mem), BUS_DMASYNC_POSTWRITE);
5307 	mcx_dmamem_free(sc, &tx->tx_sq_mem);
5308 
5309 	tx->tx_sqn = 0;
5310 	return 0;
5311 }
5312 
5313 static int
5314 mcx_ready_sq(struct mcx_softc *sc, struct mcx_tx *tx)
5315 {
5316 	struct mcx_cmdq_entry *cqe;
5317 	struct mcx_dmamem mxm;
5318 	struct mcx_cmd_modify_sq_in *in;
5319 	struct mcx_cmd_modify_sq_mb_in *mbin;
5320 	struct mcx_cmd_modify_sq_out *out;
5321 	int error;
5322 	int token;
5323 
5324 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5325 	token = mcx_cmdq_token(sc);
5326 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
5327 	    sizeof(*out), token);
5328 
5329 	in = mcx_cmdq_in(cqe);
5330 	in->cmd_opcode = htobe16(MCX_CMD_MODIFY_SQ);
5331 	in->cmd_op_mod = htobe16(0);
5332 	in->cmd_sq_state = htobe32((MCX_QUEUE_STATE_RST << 28) | tx->tx_sqn);
5333 
5334 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
5335 	    &cqe->cq_input_ptr, token) != 0) {
5336 		printf("%s: unable to allocate modify sq mailbox\n",
5337 		    DEVNAME(sc));
5338 		return (-1);
5339 	}
5340 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5341 	mbin->cmd_sq_ctx.sq_flags = htobe32(
5342 	    MCX_QUEUE_STATE_RDY << MCX_SQ_CTX_STATE_SHIFT);
5343 
5344 	mcx_cmdq_mboxes_sign(&mxm, 1);
5345 	mcx_cmdq_post(sc, cqe, 0);
5346 	error = mcx_cmdq_poll(sc, cqe, 1000);
5347 	if (error != 0) {
5348 		printf("%s: modify sq timeout\n", DEVNAME(sc));
5349 		goto free;
5350 	}
5351 	if (mcx_cmdq_verify(cqe) != 0) {
5352 		printf("%s: modify sq command corrupt\n", DEVNAME(sc));
5353 		goto free;
5354 	}
5355 
5356 	out = mcx_cmdq_out(cqe);
5357 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5358 		printf("%s: modify sq failed (%x, %x)\n", DEVNAME(sc),
5359 		    out->cmd_status, be32toh(out->cmd_syndrome));
5360 		error = -1;
5361 		goto free;
5362 	}
5363 
5364 free:
5365 	mcx_dmamem_free(sc, &mxm);
5366 	return (error);
5367 }
5368 
5369 static int
5370 mcx_create_tis(struct mcx_softc *sc, int *tis)
5371 {
5372 	struct mcx_cmdq_entry *cqe;
5373 	struct mcx_dmamem mxm;
5374 	struct mcx_cmd_create_tis_in *in;
5375 	struct mcx_cmd_create_tis_mb_in *mbin;
5376 	struct mcx_cmd_create_tis_out *out;
5377 	int error;
5378 	int token;
5379 
5380 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5381 	token = mcx_cmdq_token(sc);
5382 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
5383 	    sizeof(*out), token);
5384 
5385 	in = mcx_cmdq_in(cqe);
5386 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_TIS);
5387 	in->cmd_op_mod = htobe16(0);
5388 
5389 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
5390 	    &cqe->cq_input_ptr, token) != 0) {
5391 		printf("%s: unable to allocate create tis mailbox\n",
5392 		    DEVNAME(sc));
5393 		return (-1);
5394 	}
5395 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5396 	mbin->cmd_tdomain = htobe32(sc->sc_tdomain);
5397 
5398 	mcx_cmdq_mboxes_sign(&mxm, 1);
5399 	mcx_cmdq_post(sc, cqe, 0);
5400 	error = mcx_cmdq_poll(sc, cqe, 1000);
5401 	if (error != 0) {
5402 		printf("%s: create tis timeout\n", DEVNAME(sc));
5403 		goto free;
5404 	}
5405 	if (mcx_cmdq_verify(cqe) != 0) {
5406 		printf("%s: create tis command corrupt\n", DEVNAME(sc));
5407 		goto free;
5408 	}
5409 
5410 	out = mcx_cmdq_out(cqe);
5411 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5412 		printf("%s: create tis failed (%x, %x)\n", DEVNAME(sc),
5413 		    out->cmd_status, be32toh(out->cmd_syndrome));
5414 		error = -1;
5415 		goto free;
5416 	}
5417 
5418 	*tis = mcx_get_id(out->cmd_tisn);
5419 free:
5420 	mcx_dmamem_free(sc, &mxm);
5421 	return (error);
5422 }
5423 
5424 static int
5425 mcx_destroy_tis(struct mcx_softc *sc, int tis)
5426 {
5427 	struct mcx_cmdq_entry *cqe;
5428 	struct mcx_cmd_destroy_tis_in *in;
5429 	struct mcx_cmd_destroy_tis_out *out;
5430 	int error;
5431 	int token;
5432 
5433 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5434 	token = mcx_cmdq_token(sc);
5435 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), token);
5436 
5437 	in = mcx_cmdq_in(cqe);
5438 	in->cmd_opcode = htobe16(MCX_CMD_DESTROY_TIS);
5439 	in->cmd_op_mod = htobe16(0);
5440 	in->cmd_tisn = htobe32(tis);
5441 
5442 	mcx_cmdq_post(sc, cqe, 0);
5443 	error = mcx_cmdq_poll(sc, cqe, 1000);
5444 	if (error != 0) {
5445 		printf("%s: destroy tis timeout\n", DEVNAME(sc));
5446 		return error;
5447 	}
5448 	if (mcx_cmdq_verify(cqe) != 0) {
5449 		printf("%s: destroy tis command corrupt\n", DEVNAME(sc));
5450 		return error;
5451 	}
5452 
5453 	out = mcx_cmdq_out(cqe);
5454 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5455 		printf("%s: destroy tis failed (%x, %x)\n", DEVNAME(sc),
5456 		    out->cmd_status, be32toh(out->cmd_syndrome));
5457 		return -1;
5458 	}
5459 
5460 	return 0;
5461 }
5462 
5463 static int
5464 mcx_create_rqt(struct mcx_softc *sc, int size, int *rqns, int *rqt)
5465 {
5466 	struct mcx_cmdq_entry *cqe;
5467 	struct mcx_dmamem mxm;
5468 	struct mcx_cmd_create_rqt_in *in;
5469 	struct mcx_cmd_create_rqt_mb_in *mbin;
5470 	struct mcx_cmd_create_rqt_out *out;
5471 	struct mcx_rqt_ctx *rqt_ctx;
5472 	int *rqtn;
5473 	int error;
5474 	int token;
5475 	int i;
5476 
5477 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5478 	token = mcx_cmdq_token(sc);
5479 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin) +
5480 	    (size * sizeof(int)), sizeof(*out), token);
5481 
5482 	in = mcx_cmdq_in(cqe);
5483 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_RQT);
5484 	in->cmd_op_mod = htobe16(0);
5485 
5486 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
5487 	    &cqe->cq_input_ptr, token) != 0) {
5488 		printf("%s: unable to allocate create rqt mailbox\n",
5489 		    DEVNAME(sc));
5490 		return (-1);
5491 	}
5492 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5493 	rqt_ctx = &mbin->cmd_rqt;
5494 	rqt_ctx->cmd_rqt_max_size = htobe16(sc->sc_max_rqt_size);
5495 	rqt_ctx->cmd_rqt_actual_size = htobe16(size);
5496 
5497 	/* rqt list follows the rqt context */
5498 	rqtn = (int *)(rqt_ctx + 1);
5499 	for (i = 0; i < size; i++) {
5500 		rqtn[i] = htobe32(rqns[i]);
5501 	}
5502 
5503 	mcx_cmdq_mboxes_sign(&mxm, 1);
5504 	mcx_cmdq_post(sc, cqe, 0);
5505 	error = mcx_cmdq_poll(sc, cqe, 1000);
5506 	if (error != 0) {
5507 		printf("%s: create rqt timeout\n", DEVNAME(sc));
5508 		goto free;
5509 	}
5510 	if (mcx_cmdq_verify(cqe) != 0) {
5511 		printf("%s: create rqt command corrupt\n", DEVNAME(sc));
5512 		goto free;
5513 	}
5514 
5515 	out = mcx_cmdq_out(cqe);
5516 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5517 		printf("%s: create rqt failed (%x, %x)\n", DEVNAME(sc),
5518 		    out->cmd_status, be32toh(out->cmd_syndrome));
5519 		error = -1;
5520 		goto free;
5521 	}
5522 
5523 	*rqt = mcx_get_id(out->cmd_rqtn);
5524 	return (0);
5525 free:
5526 	mcx_dmamem_free(sc, &mxm);
5527 	return (error);
5528 }
5529 
5530 static int
5531 mcx_destroy_rqt(struct mcx_softc *sc, int rqt)
5532 {
5533 	struct mcx_cmdq_entry *cqe;
5534 	struct mcx_cmd_destroy_rqt_in *in;
5535 	struct mcx_cmd_destroy_rqt_out *out;
5536 	int error;
5537 	int token;
5538 
5539 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5540 	token = mcx_cmdq_token(sc);
5541 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), token);
5542 
5543 	in = mcx_cmdq_in(cqe);
5544 	in->cmd_opcode = htobe16(MCX_CMD_DESTROY_RQT);
5545 	in->cmd_op_mod = htobe16(0);
5546 	in->cmd_rqtn = htobe32(rqt);
5547 
5548 	mcx_cmdq_post(sc, cqe, 0);
5549 	error = mcx_cmdq_poll(sc, cqe, 1000);
5550 	if (error != 0) {
5551 		printf("%s: destroy rqt timeout\n", DEVNAME(sc));
5552 		return error;
5553 	}
5554 	if (mcx_cmdq_verify(cqe) != 0) {
5555 		printf("%s: destroy rqt command corrupt\n", DEVNAME(sc));
5556 		return error;
5557 	}
5558 
5559 	out = mcx_cmdq_out(cqe);
5560 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5561 		printf("%s: destroy rqt failed (%x, %x)\n", DEVNAME(sc),
5562 		    out->cmd_status, be32toh(out->cmd_syndrome));
5563 		return -1;
5564 	}
5565 
5566 	return 0;
5567 }
5568 
5569 #if 0
5570 static int
5571 mcx_alloc_flow_counter(struct mcx_softc *sc, int i)
5572 {
5573 	struct mcx_cmdq_entry *cqe;
5574 	struct mcx_cmd_alloc_flow_counter_in *in;
5575 	struct mcx_cmd_alloc_flow_counter_out *out;
5576 	int error;
5577 
5578 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5579 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
5580 
5581 	in = mcx_cmdq_in(cqe);
5582 	in->cmd_opcode = htobe16(MCX_CMD_ALLOC_FLOW_COUNTER);
5583 	in->cmd_op_mod = htobe16(0);
5584 
5585 	mcx_cmdq_post(sc, cqe, 0);
5586 
5587 	error = mcx_cmdq_poll(sc, cqe, 1000);
5588 	if (error != 0) {
5589 		printf("%s: alloc flow counter timeout\n", DEVNAME(sc));
5590 		return (-1);
5591 	}
5592 	if (mcx_cmdq_verify(cqe) != 0) {
5593 		printf("%s: alloc flow counter command corrupt\n", DEVNAME(sc));
5594 		return (-1);
5595 	}
5596 
5597 	out = (struct mcx_cmd_alloc_flow_counter_out *)cqe->cq_output_data;
5598 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5599 		printf("%s: alloc flow counter failed (%x)\n", DEVNAME(sc),
5600 		    out->cmd_status);
5601 		return (-1);
5602 	}
5603 
5604 	sc->sc_flow_counter_id[i]  = be16toh(out->cmd_flow_counter_id);
5605 	printf("flow counter id %d = %d\n", i, sc->sc_flow_counter_id[i]);
5606 
5607 	return (0);
5608 }
5609 #endif
5610 
5611 static int
5612 mcx_create_flow_table(struct mcx_softc *sc, int log_size, int level,
5613     int *flow_table_id)
5614 {
5615 	struct mcx_cmdq_entry *cqe;
5616 	struct mcx_dmamem mxm;
5617 	struct mcx_cmd_create_flow_table_in *in;
5618 	struct mcx_cmd_create_flow_table_mb_in *mbin;
5619 	struct mcx_cmd_create_flow_table_out *out;
5620 	int error;
5621 	int token;
5622 
5623 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5624 	token = mcx_cmdq_token(sc);
5625 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
5626 	    sizeof(*out), token);
5627 
5628 	in = mcx_cmdq_in(cqe);
5629 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_FLOW_TABLE);
5630 	in->cmd_op_mod = htobe16(0);
5631 
5632 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
5633 	    &cqe->cq_input_ptr, token) != 0) {
5634 		printf("%s: unable to allocate create flow table mailbox\n",
5635 		    DEVNAME(sc));
5636 		return (-1);
5637 	}
5638 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5639 	mbin->cmd_table_type = MCX_FLOW_TABLE_TYPE_RX;
5640 	mbin->cmd_ctx.ft_log_size = log_size;
5641 	mbin->cmd_ctx.ft_level = level;
5642 
5643 	mcx_cmdq_mboxes_sign(&mxm, 1);
5644 	mcx_cmdq_post(sc, cqe, 0);
5645 	error = mcx_cmdq_poll(sc, cqe, 1000);
5646 	if (error != 0) {
5647 		printf("%s: create flow table timeout\n", DEVNAME(sc));
5648 		goto free;
5649 	}
5650 	if (mcx_cmdq_verify(cqe) != 0) {
5651 		printf("%s: create flow table command corrupt\n", DEVNAME(sc));
5652 		goto free;
5653 	}
5654 
5655 	out = mcx_cmdq_out(cqe);
5656 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5657 		printf("%s: create flow table failed (%x, %x)\n", DEVNAME(sc),
5658 		    out->cmd_status, be32toh(out->cmd_syndrome));
5659 		error = -1;
5660 		goto free;
5661 	}
5662 
5663 	*flow_table_id = mcx_get_id(out->cmd_table_id);
5664 free:
5665 	mcx_dmamem_free(sc, &mxm);
5666 	return (error);
5667 }
5668 
5669 static int
5670 mcx_set_flow_table_root(struct mcx_softc *sc, int flow_table_id)
5671 {
5672 	struct mcx_cmdq_entry *cqe;
5673 	struct mcx_dmamem mxm;
5674 	struct mcx_cmd_set_flow_table_root_in *in;
5675 	struct mcx_cmd_set_flow_table_root_mb_in *mbin;
5676 	struct mcx_cmd_set_flow_table_root_out *out;
5677 	int error;
5678 	int token;
5679 
5680 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5681 	token = mcx_cmdq_token(sc);
5682 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
5683 	    sizeof(*out), token);
5684 
5685 	in = mcx_cmdq_in(cqe);
5686 	in->cmd_opcode = htobe16(MCX_CMD_SET_FLOW_TABLE_ROOT);
5687 	in->cmd_op_mod = htobe16(0);
5688 
5689 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
5690 	    &cqe->cq_input_ptr, token) != 0) {
5691 		printf("%s: unable to allocate set flow table root mailbox\n",
5692 		    DEVNAME(sc));
5693 		return (-1);
5694 	}
5695 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5696 	mbin->cmd_table_type = MCX_FLOW_TABLE_TYPE_RX;
5697 	mbin->cmd_table_id = htobe32(flow_table_id);
5698 
5699 	mcx_cmdq_mboxes_sign(&mxm, 1);
5700 	mcx_cmdq_post(sc, cqe, 0);
5701 	error = mcx_cmdq_poll(sc, cqe, 1000);
5702 	if (error != 0) {
5703 		printf("%s: set flow table root timeout\n", DEVNAME(sc));
5704 		goto free;
5705 	}
5706 	if (mcx_cmdq_verify(cqe) != 0) {
5707 		printf("%s: set flow table root command corrupt\n",
5708 		    DEVNAME(sc));
5709 		goto free;
5710 	}
5711 
5712 	out = mcx_cmdq_out(cqe);
5713 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5714 		printf("%s: set flow table root failed (%x, %x)\n",
5715 		    DEVNAME(sc), out->cmd_status, be32toh(out->cmd_syndrome));
5716 		error = -1;
5717 		goto free;
5718 	}
5719 
5720 free:
5721 	mcx_dmamem_free(sc, &mxm);
5722 	return (error);
5723 }
5724 
5725 static int
5726 mcx_destroy_flow_table(struct mcx_softc *sc, int flow_table_id)
5727 {
5728 	struct mcx_cmdq_entry *cqe;
5729 	struct mcx_dmamem mxm;
5730 	struct mcx_cmd_destroy_flow_table_in *in;
5731 	struct mcx_cmd_destroy_flow_table_mb_in *mb;
5732 	struct mcx_cmd_destroy_flow_table_out *out;
5733 	int error;
5734 	int token;
5735 
5736 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5737 	token = mcx_cmdq_token(sc);
5738 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mb), sizeof(*out), token);
5739 
5740 	in = mcx_cmdq_in(cqe);
5741 	in->cmd_opcode = htobe16(MCX_CMD_DESTROY_FLOW_TABLE);
5742 	in->cmd_op_mod = htobe16(0);
5743 
5744 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
5745 	    &cqe->cq_input_ptr, token) != 0) {
5746 		printf("%s: unable to allocate destroy flow table mailbox\n",
5747 		    DEVNAME(sc));
5748 		return (-1);
5749 	}
5750 	mb = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5751 	mb->cmd_table_type = MCX_FLOW_TABLE_TYPE_RX;
5752 	mb->cmd_table_id = htobe32(flow_table_id);
5753 
5754 	mcx_cmdq_mboxes_sign(&mxm, 1);
5755 	mcx_cmdq_post(sc, cqe, 0);
5756 	error = mcx_cmdq_poll(sc, cqe, 1000);
5757 	if (error != 0) {
5758 		printf("%s: destroy flow table timeout\n", DEVNAME(sc));
5759 		goto free;
5760 	}
5761 	if (mcx_cmdq_verify(cqe) != 0) {
5762 		printf("%s: destroy flow table command corrupt\n",
5763 		    DEVNAME(sc));
5764 		goto free;
5765 	}
5766 
5767 	out = mcx_cmdq_out(cqe);
5768 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5769 		printf("%s: destroy flow table failed (%x, %x)\n", DEVNAME(sc),
5770 		    out->cmd_status, be32toh(out->cmd_syndrome));
5771 		error = -1;
5772 		goto free;
5773 	}
5774 
5775 free:
5776 	mcx_dmamem_free(sc, &mxm);
5777 	return (error);
5778 }
5779 
5780 
5781 static int
5782 mcx_create_flow_group(struct mcx_softc *sc, int flow_table_id, int group,
5783     int start, int size, int match_enable, struct mcx_flow_match *match)
5784 {
5785 	struct mcx_cmdq_entry *cqe;
5786 	struct mcx_dmamem mxm;
5787 	struct mcx_cmd_create_flow_group_in *in;
5788 	struct mcx_cmd_create_flow_group_mb_in *mbin;
5789 	struct mcx_cmd_create_flow_group_out *out;
5790 	struct mcx_flow_group *mfg;
5791 	int error;
5792 	int token;
5793 
5794 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5795 	token = mcx_cmdq_token(sc);
5796 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin), sizeof(*out),
5797 	    token);
5798 
5799 	in = mcx_cmdq_in(cqe);
5800 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_FLOW_GROUP);
5801 	in->cmd_op_mod = htobe16(0);
5802 
5803 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2, &cqe->cq_input_ptr, token)
5804 	    != 0) {
5805 		printf("%s: unable to allocate create flow group mailbox\n",
5806 		    DEVNAME(sc));
5807 		return (-1);
5808 	}
5809 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5810 	mbin->cmd_table_type = MCX_FLOW_TABLE_TYPE_RX;
5811 	mbin->cmd_table_id = htobe32(flow_table_id);
5812 	mbin->cmd_start_flow_index = htobe32(start);
5813 	mbin->cmd_end_flow_index = htobe32(start + (size - 1));
5814 
5815 	mbin->cmd_match_criteria_enable = match_enable;
5816 	memcpy(&mbin->cmd_match_criteria, match, sizeof(*match));
5817 
5818 	mcx_cmdq_mboxes_sign(&mxm, 2);
5819 	mcx_cmdq_post(sc, cqe, 0);
5820 	error = mcx_cmdq_poll(sc, cqe, 1000);
5821 	if (error != 0) {
5822 		printf("%s: create flow group timeout\n", DEVNAME(sc));
5823 		goto free;
5824 	}
5825 	if (mcx_cmdq_verify(cqe) != 0) {
5826 		printf("%s: create flow group command corrupt\n", DEVNAME(sc));
5827 		goto free;
5828 	}
5829 
5830 	out = mcx_cmdq_out(cqe);
5831 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5832 		printf("%s: create flow group failed (%x, %x)\n", DEVNAME(sc),
5833 		    out->cmd_status, be32toh(out->cmd_syndrome));
5834 		error = -1;
5835 		goto free;
5836 	}
5837 
5838 	mfg = &sc->sc_flow_group[group];
5839 	mfg->g_id = mcx_get_id(out->cmd_group_id);
5840 	mfg->g_table = flow_table_id;
5841 	mfg->g_start = start;
5842 	mfg->g_size = size;
5843 
5844 free:
5845 	mcx_dmamem_free(sc, &mxm);
5846 	return (error);
5847 }
5848 
5849 static int
5850 mcx_destroy_flow_group(struct mcx_softc *sc, int group)
5851 {
5852 	struct mcx_cmdq_entry *cqe;
5853 	struct mcx_dmamem mxm;
5854 	struct mcx_cmd_destroy_flow_group_in *in;
5855 	struct mcx_cmd_destroy_flow_group_mb_in *mb;
5856 	struct mcx_cmd_destroy_flow_group_out *out;
5857 	struct mcx_flow_group *mfg;
5858 	int error;
5859 	int token;
5860 
5861 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5862 	token = mcx_cmdq_token(sc);
5863 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mb), sizeof(*out), token);
5864 
5865 	in = mcx_cmdq_in(cqe);
5866 	in->cmd_opcode = htobe16(MCX_CMD_DESTROY_FLOW_GROUP);
5867 	in->cmd_op_mod = htobe16(0);
5868 
5869 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2,
5870 	    &cqe->cq_input_ptr, token) != 0) {
5871 		printf("%s: unable to allocate destroy flow group mailbox\n",
5872 		    DEVNAME(sc));
5873 		return (-1);
5874 	}
5875 	mb = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5876 	mb->cmd_table_type = MCX_FLOW_TABLE_TYPE_RX;
5877 	mfg = &sc->sc_flow_group[group];
5878 	mb->cmd_table_id = htobe32(mfg->g_table);
5879 	mb->cmd_group_id = htobe32(mfg->g_id);
5880 
5881 	mcx_cmdq_mboxes_sign(&mxm, 2);
5882 	mcx_cmdq_post(sc, cqe, 0);
5883 	error = mcx_cmdq_poll(sc, cqe, 1000);
5884 	if (error != 0) {
5885 		printf("%s: destroy flow group timeout\n", DEVNAME(sc));
5886 		goto free;
5887 	}
5888 	if (mcx_cmdq_verify(cqe) != 0) {
5889 		printf("%s: destroy flow group command corrupt\n", DEVNAME(sc));
5890 		goto free;
5891 	}
5892 
5893 	out = mcx_cmdq_out(cqe);
5894 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5895 		printf("%s: destroy flow group failed (%x, %x)\n", DEVNAME(sc),
5896 		    out->cmd_status, be32toh(out->cmd_syndrome));
5897 		error = -1;
5898 		goto free;
5899 	}
5900 
5901 	mfg->g_id = -1;
5902 	mfg->g_table = -1;
5903 	mfg->g_size = 0;
5904 	mfg->g_start = 0;
5905 free:
5906 	mcx_dmamem_free(sc, &mxm);
5907 	return (error);
5908 }
5909 
5910 static int
5911 mcx_set_flow_table_entry_mac(struct mcx_softc *sc, int group, int index,
5912     const uint8_t *macaddr, uint32_t dest)
5913 {
5914 	struct mcx_cmdq_entry *cqe;
5915 	struct mcx_dmamem mxm;
5916 	struct mcx_cmd_set_flow_table_entry_in *in;
5917 	struct mcx_cmd_set_flow_table_entry_mb_in *mbin;
5918 	struct mcx_cmd_set_flow_table_entry_out *out;
5919 	struct mcx_flow_group *mfg;
5920 	uint32_t *pdest;
5921 	int error;
5922 	int token;
5923 
5924 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5925 	token = mcx_cmdq_token(sc);
5926 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin) + sizeof(*pdest),
5927 	    sizeof(*out), token);
5928 
5929 	in = mcx_cmdq_in(cqe);
5930 	in->cmd_opcode = htobe16(MCX_CMD_SET_FLOW_TABLE_ENTRY);
5931 	in->cmd_op_mod = htobe16(0);
5932 
5933 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2, &cqe->cq_input_ptr, token)
5934 	    != 0) {
5935 		printf("%s: unable to allocate set flow table entry mailbox\n",
5936 		    DEVNAME(sc));
5937 		return (-1);
5938 	}
5939 
5940 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5941 	mbin->cmd_table_type = MCX_FLOW_TABLE_TYPE_RX;
5942 
5943 	mfg = &sc->sc_flow_group[group];
5944 	mbin->cmd_table_id = htobe32(mfg->g_table);
5945 	mbin->cmd_flow_index = htobe32(mfg->g_start + index);
5946 	mbin->cmd_flow_ctx.fc_group_id = htobe32(mfg->g_id);
5947 
5948 	/* flow context ends at offset 0x330, 0x130 into the second mbox */
5949 	pdest = (uint32_t *)
5950 	    (((char *)mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 1))) + 0x130);
5951 	mbin->cmd_flow_ctx.fc_action = htobe32(MCX_FLOW_CONTEXT_ACTION_FORWARD);
5952 	mbin->cmd_flow_ctx.fc_dest_list_size = htobe32(1);
5953 	*pdest = htobe32(dest);
5954 
5955 	/* the only thing we match on at the moment is the dest mac address */
5956 	if (macaddr != NULL) {
5957 		memcpy(mbin->cmd_flow_ctx.fc_match_value.mc_dest_mac, macaddr,
5958 		    ETHER_ADDR_LEN);
5959 	}
5960 
5961 	mcx_cmdq_mboxes_sign(&mxm, 2);
5962 	mcx_cmdq_post(sc, cqe, 0);
5963 	error = mcx_cmdq_poll(sc, cqe, 1000);
5964 	if (error != 0) {
5965 		printf("%s: set flow table entry timeout\n", DEVNAME(sc));
5966 		goto free;
5967 	}
5968 	if (mcx_cmdq_verify(cqe) != 0) {
5969 		printf("%s: set flow table entry command corrupt\n",
5970 		    DEVNAME(sc));
5971 		goto free;
5972 	}
5973 
5974 	out = mcx_cmdq_out(cqe);
5975 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5976 		printf("%s: set flow table entry failed (%x, %x)\n",
5977 		    DEVNAME(sc), out->cmd_status, be32toh(out->cmd_syndrome));
5978 		error = -1;
5979 		goto free;
5980 	}
5981 
5982 free:
5983 	mcx_dmamem_free(sc, &mxm);
5984 	return (error);
5985 }
5986 
5987 static int
5988 mcx_set_flow_table_entry_proto(struct mcx_softc *sc, int group, int index,
5989     int ethertype, int ip_proto, uint32_t dest)
5990 {
5991 	struct mcx_cmdq_entry *cqe;
5992 	struct mcx_dmamem mxm;
5993 	struct mcx_cmd_set_flow_table_entry_in *in;
5994 	struct mcx_cmd_set_flow_table_entry_mb_in *mbin;
5995 	struct mcx_cmd_set_flow_table_entry_out *out;
5996 	struct mcx_flow_group *mfg;
5997 	uint32_t *pdest;
5998 	int error;
5999 	int token;
6000 
6001 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
6002 	token = mcx_cmdq_token(sc);
6003 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin) + sizeof(*pdest),
6004 	    sizeof(*out), token);
6005 
6006 	in = mcx_cmdq_in(cqe);
6007 	in->cmd_opcode = htobe16(MCX_CMD_SET_FLOW_TABLE_ENTRY);
6008 	in->cmd_op_mod = htobe16(0);
6009 
6010 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2, &cqe->cq_input_ptr, token)
6011 	    != 0) {
6012 		printf("%s: unable to allocate set flow table entry mailbox\n",
6013 		    DEVNAME(sc));
6014 		return (-1);
6015 	}
6016 
6017 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
6018 	mbin->cmd_table_type = MCX_FLOW_TABLE_TYPE_RX;
6019 
6020 	mfg = &sc->sc_flow_group[group];
6021 	mbin->cmd_table_id = htobe32(mfg->g_table);
6022 	mbin->cmd_flow_index = htobe32(mfg->g_start + index);
6023 	mbin->cmd_flow_ctx.fc_group_id = htobe32(mfg->g_id);
6024 
6025 	/* flow context ends at offset 0x330, 0x130 into the second mbox */
6026 	pdest = (uint32_t *)
6027 	    (((char *)mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 1))) + 0x130);
6028 	mbin->cmd_flow_ctx.fc_action = htobe32(MCX_FLOW_CONTEXT_ACTION_FORWARD);
6029 	mbin->cmd_flow_ctx.fc_dest_list_size = htobe32(1);
6030 	*pdest = htobe32(dest);
6031 
6032 	mbin->cmd_flow_ctx.fc_match_value.mc_ethertype = htobe16(ethertype);
6033 	mbin->cmd_flow_ctx.fc_match_value.mc_ip_proto = ip_proto;
6034 
6035 	mcx_cmdq_mboxes_sign(&mxm, 2);
6036 	mcx_cmdq_post(sc, cqe, 0);
6037 	error = mcx_cmdq_poll(sc, cqe, 1000);
6038 	if (error != 0) {
6039 		printf("%s: set flow table entry timeout\n", DEVNAME(sc));
6040 		goto free;
6041 	}
6042 	if (mcx_cmdq_verify(cqe) != 0) {
6043 		printf("%s: set flow table entry command corrupt\n",
6044 		    DEVNAME(sc));
6045 		goto free;
6046 	}
6047 
6048 	out = mcx_cmdq_out(cqe);
6049 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
6050 		printf("%s: set flow table entry failed (%x, %x)\n",
6051 		    DEVNAME(sc), out->cmd_status, be32toh(out->cmd_syndrome));
6052 		error = -1;
6053 		goto free;
6054 	}
6055 
6056 free:
6057 	mcx_dmamem_free(sc, &mxm);
6058 	return (error);
6059 }
6060 
6061 static int
6062 mcx_delete_flow_table_entry(struct mcx_softc *sc, int group, int index)
6063 {
6064 	struct mcx_cmdq_entry *cqe;
6065 	struct mcx_dmamem mxm;
6066 	struct mcx_cmd_delete_flow_table_entry_in *in;
6067 	struct mcx_cmd_delete_flow_table_entry_mb_in *mbin;
6068 	struct mcx_cmd_delete_flow_table_entry_out *out;
6069 	struct mcx_flow_group *mfg;
6070 	int error;
6071 	int token;
6072 
6073 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
6074 	token = mcx_cmdq_token(sc);
6075 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin), sizeof(*out),
6076 	    token);
6077 
6078 	in = mcx_cmdq_in(cqe);
6079 	in->cmd_opcode = htobe16(MCX_CMD_DELETE_FLOW_TABLE_ENTRY);
6080 	in->cmd_op_mod = htobe16(0);
6081 
6082 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2,
6083 	    &cqe->cq_input_ptr, token) != 0) {
6084 		printf("%s: unable to allocate "
6085 		    "delete flow table entry mailbox\n", DEVNAME(sc));
6086 		return (-1);
6087 	}
6088 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
6089 	mbin->cmd_table_type = MCX_FLOW_TABLE_TYPE_RX;
6090 
6091 	mfg = &sc->sc_flow_group[group];
6092 	mbin->cmd_table_id = htobe32(mfg->g_table);
6093 	mbin->cmd_flow_index = htobe32(mfg->g_start + index);
6094 
6095 	mcx_cmdq_mboxes_sign(&mxm, 2);
6096 	mcx_cmdq_post(sc, cqe, 0);
6097 	error = mcx_cmdq_poll(sc, cqe, 1000);
6098 	if (error != 0) {
6099 		printf("%s: delete flow table entry timeout\n", DEVNAME(sc));
6100 		goto free;
6101 	}
6102 	if (mcx_cmdq_verify(cqe) != 0) {
6103 		printf("%s: delete flow table entry command corrupt\n",
6104 		    DEVNAME(sc));
6105 		goto free;
6106 	}
6107 
6108 	out = mcx_cmdq_out(cqe);
6109 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
6110 		printf("%s: delete flow table entry %d:%d failed (%x, %x)\n",
6111 		    DEVNAME(sc), group, index, out->cmd_status,
6112 		    be32toh(out->cmd_syndrome));
6113 		error = -1;
6114 		goto free;
6115 	}
6116 
6117 free:
6118 	mcx_dmamem_free(sc, &mxm);
6119 	return (error);
6120 }
6121 
6122 #if 0
6123 int
6124 mcx_dump_flow_table(struct mcx_softc *sc, int flow_table_id)
6125 {
6126 	struct mcx_dmamem mxm;
6127 	struct mcx_cmdq_entry *cqe;
6128 	struct mcx_cmd_query_flow_table_in *in;
6129 	struct mcx_cmd_query_flow_table_mb_in *mbin;
6130 	struct mcx_cmd_query_flow_table_out *out;
6131 	struct mcx_cmd_query_flow_table_mb_out *mbout;
6132 	uint8_t token = mcx_cmdq_token(sc);
6133 	int error;
6134 	int i;
6135 	uint8_t *dump;
6136 
6137 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
6138 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
6139 	    sizeof(*out) + sizeof(*mbout) + 16, token);
6140 
6141 	in = mcx_cmdq_in(cqe);
6142 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_FLOW_TABLE);
6143 	in->cmd_op_mod = htobe16(0);
6144 
6145 	CTASSERT(sizeof(*mbin) <= MCX_CMDQ_MAILBOX_DATASIZE);
6146 	CTASSERT(sizeof(*mbout) <= MCX_CMDQ_MAILBOX_DATASIZE);
6147 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2,
6148 	    &cqe->cq_output_ptr, token) != 0) {
6149 		printf(", unable to allocate query flow table mailboxes\n");
6150 		return (-1);
6151 	}
6152 	cqe->cq_input_ptr = cqe->cq_output_ptr;
6153 
6154 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
6155 	mbin->cmd_table_type = 0;
6156 	mbin->cmd_table_id = htobe32(flow_table_id);
6157 
6158 	mcx_cmdq_mboxes_sign(&mxm, 1);
6159 
6160 	mcx_cmdq_post(sc, cqe, 0);
6161 	error = mcx_cmdq_poll(sc, cqe, 1000);
6162 	if (error != 0) {
6163 		printf("%s: query flow table timeout\n", DEVNAME(sc));
6164 		goto free;
6165 	}
6166 	error = mcx_cmdq_verify(cqe);
6167 	if (error != 0) {
6168 		printf("%s: query flow table reply corrupt\n", DEVNAME(sc));
6169 		goto free;
6170 	}
6171 
6172 	out = mcx_cmdq_out(cqe);
6173 	switch (out->cmd_status) {
6174 	case MCX_CQ_STATUS_OK:
6175 		break;
6176 	default:
6177 		printf("%s: query flow table failed (%x/%x)\n", DEVNAME(sc),
6178 		    out->cmd_status, be32toh(out->cmd_syndrome));
6179 		error = -1;
6180 		goto free;
6181 	}
6182 
6183         mbout = (struct mcx_cmd_query_flow_table_mb_out *)
6184 	    (mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)));
6185 	dump = (uint8_t *)mbout + 8;
6186 	for (i = 0; i < sizeof(struct mcx_flow_table_ctx); i++) {
6187 		printf("%.2x ", dump[i]);
6188 		if (i % 16 == 15)
6189 			printf("\n");
6190 	}
6191 free:
6192 	mcx_cq_mboxes_free(sc, &mxm);
6193 	return (error);
6194 }
6195 int
6196 mcx_dump_flow_table_entry(struct mcx_softc *sc, int flow_table_id, int index)
6197 {
6198 	struct mcx_dmamem mxm;
6199 	struct mcx_cmdq_entry *cqe;
6200 	struct mcx_cmd_query_flow_table_entry_in *in;
6201 	struct mcx_cmd_query_flow_table_entry_mb_in *mbin;
6202 	struct mcx_cmd_query_flow_table_entry_out *out;
6203 	struct mcx_cmd_query_flow_table_entry_mb_out *mbout;
6204 	uint8_t token = mcx_cmdq_token(sc);
6205 	int error;
6206 	int i;
6207 	uint8_t *dump;
6208 
6209 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
6210 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
6211 	    sizeof(*out) + sizeof(*mbout) + 16, token);
6212 
6213 	in = mcx_cmdq_in(cqe);
6214 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_FLOW_TABLE_ENTRY);
6215 	in->cmd_op_mod = htobe16(0);
6216 
6217 	CTASSERT(sizeof(*mbin) <= MCX_CMDQ_MAILBOX_DATASIZE);
6218 	CTASSERT(sizeof(*mbout) <= MCX_CMDQ_MAILBOX_DATASIZE*2);
6219 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2,
6220 	    &cqe->cq_output_ptr, token) != 0) {
6221 		printf(", unable to allocate "
6222 		    "query flow table entry mailboxes\n");
6223 		return (-1);
6224 	}
6225 	cqe->cq_input_ptr = cqe->cq_output_ptr;
6226 
6227 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
6228 	mbin->cmd_table_type = 0;
6229 	mbin->cmd_table_id = htobe32(flow_table_id);
6230 	mbin->cmd_flow_index = htobe32(index);
6231 
6232 	mcx_cmdq_mboxes_sign(&mxm, 1);
6233 
6234 	mcx_cmdq_post(sc, cqe, 0);
6235 	error = mcx_cmdq_poll(sc, cqe, 1000);
6236 	if (error != 0) {
6237 		printf("%s: query flow table entry timeout\n", DEVNAME(sc));
6238 		goto free;
6239 	}
6240 	error = mcx_cmdq_verify(cqe);
6241 	if (error != 0) {
6242 		printf("%s: query flow table entry reply corrupt\n",
6243 		    DEVNAME(sc));
6244 		goto free;
6245 	}
6246 
6247 	out = mcx_cmdq_out(cqe);
6248 	switch (out->cmd_status) {
6249 	case MCX_CQ_STATUS_OK:
6250 		break;
6251 	default:
6252 		printf("%s: query flow table entry failed (%x/%x)\n",
6253 		    DEVNAME(sc), out->cmd_status, be32toh(out->cmd_syndrome));
6254 		error = -1;
6255 		goto free;
6256 	}
6257 
6258         mbout = (struct mcx_cmd_query_flow_table_entry_mb_out *)
6259 	    (mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)));
6260 	dump = (uint8_t *)mbout;
6261 	for (i = 0; i < MCX_CMDQ_MAILBOX_DATASIZE; i++) {
6262 		printf("%.2x ", dump[i]);
6263 		if (i % 16 == 15)
6264 			printf("\n");
6265 	}
6266 
6267 free:
6268 	mcx_cq_mboxes_free(sc, &mxm);
6269 	return (error);
6270 }
6271 
6272 int
6273 mcx_dump_flow_group(struct mcx_softc *sc, int flow_table_id)
6274 {
6275 	struct mcx_dmamem mxm;
6276 	struct mcx_cmdq_entry *cqe;
6277 	struct mcx_cmd_query_flow_group_in *in;
6278 	struct mcx_cmd_query_flow_group_mb_in *mbin;
6279 	struct mcx_cmd_query_flow_group_out *out;
6280 	struct mcx_cmd_query_flow_group_mb_out *mbout;
6281 	uint8_t token = mcx_cmdq_token(sc);
6282 	int error;
6283 	int i;
6284 	uint8_t *dump;
6285 
6286 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
6287 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
6288 	    sizeof(*out) + sizeof(*mbout) + 16, token);
6289 
6290 	in = mcx_cmdq_in(cqe);
6291 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_FLOW_GROUP);
6292 	in->cmd_op_mod = htobe16(0);
6293 
6294 	CTASSERT(sizeof(*mbin) <= MCX_CMDQ_MAILBOX_DATASIZE);
6295 	CTASSERT(sizeof(*mbout) <= MCX_CMDQ_MAILBOX_DATASIZE*2);
6296 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2,
6297 	    &cqe->cq_output_ptr, token) != 0) {
6298 		printf(", unable to allocate query flow group mailboxes\n");
6299 		return (-1);
6300 	}
6301 	cqe->cq_input_ptr = cqe->cq_output_ptr;
6302 
6303 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
6304 	mbin->cmd_table_type = 0;
6305 	mbin->cmd_table_id = htobe32(flow_table_id);
6306 	mbin->cmd_group_id = htobe32(sc->sc_flow_group_id);
6307 
6308 	mcx_cmdq_mboxes_sign(&mxm, 1);
6309 
6310 	mcx_cmdq_post(sc, cqe, 0);
6311 	error = mcx_cmdq_poll(sc, cqe, 1000);
6312 	if (error != 0) {
6313 		printf("%s: query flow group timeout\n", DEVNAME(sc));
6314 		goto free;
6315 	}
6316 	error = mcx_cmdq_verify(cqe);
6317 	if (error != 0) {
6318 		printf("%s: query flow group reply corrupt\n", DEVNAME(sc));
6319 		goto free;
6320 	}
6321 
6322 	out = mcx_cmdq_out(cqe);
6323 	switch (out->cmd_status) {
6324 	case MCX_CQ_STATUS_OK:
6325 		break;
6326 	default:
6327 		printf("%s: query flow group failed (%x/%x)\n", DEVNAME(sc),
6328 		    out->cmd_status, be32toh(out->cmd_syndrome));
6329 		error = -1;
6330 		goto free;
6331 	}
6332 
6333         mbout = (struct mcx_cmd_query_flow_group_mb_out *)
6334 	    (mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)));
6335 	dump = (uint8_t *)mbout;
6336 	for (i = 0; i < MCX_CMDQ_MAILBOX_DATASIZE; i++) {
6337 		printf("%.2x ", dump[i]);
6338 		if (i % 16 == 15)
6339 			printf("\n");
6340 	}
6341 	dump = (uint8_t *)(mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 1)));
6342 	for (i = 0; i < MCX_CMDQ_MAILBOX_DATASIZE; i++) {
6343 		printf("%.2x ", dump[i]);
6344 		if (i % 16 == 15)
6345 			printf("\n");
6346 	}
6347 
6348 free:
6349 	mcx_cq_mboxes_free(sc, &mxm);
6350 	return (error);
6351 }
6352 
6353 static int
6354 mcx_dump_counters(struct mcx_softc *sc)
6355 {
6356 	struct mcx_dmamem mxm;
6357 	struct mcx_cmdq_entry *cqe;
6358 	struct mcx_cmd_query_vport_counters_in *in;
6359 	struct mcx_cmd_query_vport_counters_mb_in *mbin;
6360 	struct mcx_cmd_query_vport_counters_out *out;
6361 	struct mcx_nic_vport_counters *counters;
6362 	int error, token;
6363 
6364 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
6365 	token = mcx_cmdq_token(sc);
6366 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
6367 	    sizeof(*out) + sizeof(*counters), token);
6368 
6369 	in = mcx_cmdq_in(cqe);
6370 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_VPORT_COUNTERS);
6371 	in->cmd_op_mod = htobe16(0);
6372 
6373 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
6374 	    &cqe->cq_output_ptr, token) != 0) {
6375 		printf(", unable to allocate "
6376 		    "query nic vport counters mailboxen\n");
6377 		return (-1);
6378 	}
6379 	cqe->cq_input_ptr = cqe->cq_output_ptr;
6380 
6381 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
6382 	mbin->cmd_clear = 0x80;
6383 
6384 	mcx_cmdq_mboxes_sign(&mxm, 1);
6385 	mcx_cmdq_post(sc, cqe, 0);
6386 
6387 	error = mcx_cmdq_poll(sc, cqe, 1000);
6388 	if (error != 0) {
6389 		printf("%s: query nic vport counters timeout\n", DEVNAME(sc));
6390 		goto free;
6391 	}
6392 	if (mcx_cmdq_verify(cqe) != 0) {
6393 		printf("%s: query nic vport counters command corrupt\n",
6394 		    DEVNAME(sc));
6395 		goto free;
6396 	}
6397 
6398 	out = mcx_cmdq_out(cqe);
6399 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
6400 		printf("%s: query nic vport counters failed (%x, %x)\n",
6401 		    DEVNAME(sc), out->cmd_status, betoh32(out->cmd_syndrome));
6402 		error = -1;
6403 		goto free;
6404 	}
6405 
6406 	counters = (struct mcx_nic_vport_counters *)
6407 	    (mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)));
6408 	if (counters->rx_bcast.packets + counters->tx_bcast.packets +
6409 	    counters->rx_ucast.packets + counters->tx_ucast.packets +
6410 	    counters->rx_err.packets + counters->tx_err.packets)
6411 		printf("%s: err %llx/%llx uc %llx/%llx bc %llx/%llx\n",
6412 		    DEVNAME(sc),
6413 		    betoh64(counters->tx_err.packets),
6414 		    betoh64(counters->rx_err.packets),
6415 		    betoh64(counters->tx_ucast.packets),
6416 		    betoh64(counters->rx_ucast.packets),
6417 		    betoh64(counters->tx_bcast.packets),
6418 		    betoh64(counters->rx_bcast.packets));
6419 free:
6420 	mcx_dmamem_free(sc, &mxm);
6421 
6422 	return (error);
6423 }
6424 
6425 static int
6426 mcx_dump_flow_counter(struct mcx_softc *sc, int index, const char *what)
6427 {
6428 	struct mcx_dmamem mxm;
6429 	struct mcx_cmdq_entry *cqe;
6430 	struct mcx_cmd_query_flow_counter_in *in;
6431 	struct mcx_cmd_query_flow_counter_mb_in *mbin;
6432 	struct mcx_cmd_query_flow_counter_out *out;
6433 	struct mcx_counter *counters;
6434 	int error, token;
6435 
6436 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
6437 	token = mcx_cmdq_token(sc);
6438 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin), sizeof(*out) +
6439 	    sizeof(*counters), token);
6440 
6441 	in = mcx_cmdq_in(cqe);
6442 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_FLOW_COUNTER);
6443 	in->cmd_op_mod = htobe16(0);
6444 
6445 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
6446 	    &cqe->cq_output_ptr, token) != 0) {
6447 		printf(", unable to allocate query flow counter mailboxen\n");
6448 		return (-1);
6449 	}
6450 	cqe->cq_input_ptr = cqe->cq_output_ptr;
6451 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
6452 	mbin->cmd_flow_counter_id = htobe16(sc->sc_flow_counter_id[index]);
6453 	mbin->cmd_clear = 0x80;
6454 
6455 	mcx_cmdq_mboxes_sign(&mxm, 1);
6456 	mcx_cmdq_post(sc, cqe, 0);
6457 
6458 	error = mcx_cmdq_poll(sc, cqe, 1000);
6459 	if (error != 0) {
6460 		printf("%s: query flow counter timeout\n", DEVNAME(sc));
6461 		goto free;
6462 	}
6463 	if (mcx_cmdq_verify(cqe) != 0) {
6464 		printf("%s: query flow counter command corrupt\n", DEVNAME(sc));
6465 		goto free;
6466 	}
6467 
6468 	out = mcx_cmdq_out(cqe);
6469 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
6470 		printf("%s: query flow counter failed (%x, %x)\n", DEVNAME(sc),
6471 		    out->cmd_status, betoh32(out->cmd_syndrome));
6472 		error = -1;
6473 		goto free;
6474 	}
6475 
6476 	counters = (struct mcx_counter *)
6477 	    (mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)));
6478 	if (counters->packets)
6479 		printf("%s: %s inflow %llx\n", DEVNAME(sc), what,
6480 		    betoh64(counters->packets));
6481 free:
6482 	mcx_dmamem_free(sc, &mxm);
6483 
6484 	return (error);
6485 }
6486 
6487 #endif
6488 
6489 #if NKSTAT > 0
6490 
6491 int
6492 mcx_query_rq(struct mcx_softc *sc, struct mcx_rx *rx, struct mcx_rq_ctx *rq_ctx)
6493 {
6494 	struct mcx_dmamem mxm;
6495 	struct mcx_cmdq_entry *cqe;
6496 	struct mcx_cmd_query_rq_in *in;
6497 	struct mcx_cmd_query_rq_out *out;
6498 	struct mcx_cmd_query_rq_mb_out *mbout;
6499 	uint8_t token = mcx_cmdq_token(sc);
6500 	int error;
6501 
6502 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
6503 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out) + sizeof(*mbout) + 16,
6504 	    token);
6505 
6506 	in = mcx_cmdq_in(cqe);
6507 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_RQ);
6508 	in->cmd_op_mod = htobe16(0);
6509 	in->cmd_rqn = htobe32(rx->rx_rqn);
6510 
6511 	CTASSERT(sizeof(*mbout) <= MCX_CMDQ_MAILBOX_DATASIZE*2);
6512 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2,
6513 	    &cqe->cq_output_ptr, token) != 0) {
6514 		printf("%s: unable to allocate query rq mailboxes\n", DEVNAME(sc));
6515 		return (-1);
6516 	}
6517 
6518 	mcx_cmdq_mboxes_sign(&mxm, 1);
6519 
6520 	mcx_cmdq_post(sc, cqe, 0);
6521 	error = mcx_cmdq_poll(sc, cqe, 1000);
6522 	if (error != 0) {
6523 		printf("%s: query rq timeout\n", DEVNAME(sc));
6524 		goto free;
6525 	}
6526 	error = mcx_cmdq_verify(cqe);
6527 	if (error != 0) {
6528 		printf("%s: query rq reply corrupt\n", DEVNAME(sc));
6529 		goto free;
6530 	}
6531 
6532 	out = mcx_cmdq_out(cqe);
6533 	switch (out->cmd_status) {
6534 	case MCX_CQ_STATUS_OK:
6535 		break;
6536 	default:
6537 		printf("%s: query rq failed (%x/%x)\n", DEVNAME(sc),
6538 		    out->cmd_status, be32toh(out->cmd_syndrome));
6539 		error = -1;
6540 		goto free;
6541 	}
6542 
6543         mbout = (struct mcx_cmd_query_rq_mb_out *)
6544 	    (mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)));
6545 	memcpy(rq_ctx, &mbout->cmd_ctx, sizeof(*rq_ctx));
6546 
6547 free:
6548 	mcx_cq_mboxes_free(sc, &mxm);
6549 	return (error);
6550 }
6551 
6552 int
6553 mcx_query_sq(struct mcx_softc *sc, struct mcx_tx *tx, struct mcx_sq_ctx *sq_ctx)
6554 {
6555 	struct mcx_dmamem mxm;
6556 	struct mcx_cmdq_entry *cqe;
6557 	struct mcx_cmd_query_sq_in *in;
6558 	struct mcx_cmd_query_sq_out *out;
6559 	struct mcx_cmd_query_sq_mb_out *mbout;
6560 	uint8_t token = mcx_cmdq_token(sc);
6561 	int error;
6562 
6563 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
6564 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out) + sizeof(*mbout) + 16,
6565 	    token);
6566 
6567 	in = mcx_cmdq_in(cqe);
6568 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_SQ);
6569 	in->cmd_op_mod = htobe16(0);
6570 	in->cmd_sqn = htobe32(tx->tx_sqn);
6571 
6572 	CTASSERT(sizeof(*mbout) <= MCX_CMDQ_MAILBOX_DATASIZE*2);
6573 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2,
6574 	    &cqe->cq_output_ptr, token) != 0) {
6575 		printf("%s: unable to allocate query sq mailboxes\n", DEVNAME(sc));
6576 		return (-1);
6577 	}
6578 
6579 	mcx_cmdq_mboxes_sign(&mxm, 1);
6580 
6581 	mcx_cmdq_post(sc, cqe, 0);
6582 	error = mcx_cmdq_poll(sc, cqe, 1000);
6583 	if (error != 0) {
6584 		printf("%s: query sq timeout\n", DEVNAME(sc));
6585 		goto free;
6586 	}
6587 	error = mcx_cmdq_verify(cqe);
6588 	if (error != 0) {
6589 		printf("%s: query sq reply corrupt\n", DEVNAME(sc));
6590 		goto free;
6591 	}
6592 
6593 	out = mcx_cmdq_out(cqe);
6594 	switch (out->cmd_status) {
6595 	case MCX_CQ_STATUS_OK:
6596 		break;
6597 	default:
6598 		printf("%s: query sq failed (%x/%x)\n", DEVNAME(sc),
6599 		    out->cmd_status, be32toh(out->cmd_syndrome));
6600 		error = -1;
6601 		goto free;
6602 	}
6603 
6604         mbout = (struct mcx_cmd_query_sq_mb_out *)
6605 	    (mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)));
6606 	memcpy(sq_ctx, &mbout->cmd_ctx, sizeof(*sq_ctx));
6607 
6608 free:
6609 	mcx_cq_mboxes_free(sc, &mxm);
6610 	return (error);
6611 }
6612 
6613 int
6614 mcx_query_cq(struct mcx_softc *sc, struct mcx_cq *cq, struct mcx_cq_ctx *cq_ctx)
6615 {
6616 	struct mcx_dmamem mxm;
6617 	struct mcx_cmdq_entry *cqe;
6618 	struct mcx_cmd_query_cq_in *in;
6619 	struct mcx_cmd_query_cq_out *out;
6620 	struct mcx_cq_ctx *ctx;
6621 	uint8_t token = mcx_cmdq_token(sc);
6622 	int error;
6623 
6624 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
6625 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out) + sizeof(*ctx) + 16,
6626 	    token);
6627 
6628 	in = mcx_cmdq_in(cqe);
6629 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_CQ);
6630 	in->cmd_op_mod = htobe16(0);
6631 	in->cmd_cqn = htobe32(cq->cq_n);
6632 
6633 	CTASSERT(sizeof(*ctx) <= MCX_CMDQ_MAILBOX_DATASIZE*2);
6634 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2,
6635 	    &cqe->cq_output_ptr, token) != 0) {
6636 		printf("%s: unable to allocate query cq mailboxes\n",
6637 		    DEVNAME(sc));
6638 		return (-1);
6639 	}
6640 
6641 	mcx_cmdq_mboxes_sign(&mxm, 1);
6642 
6643 	mcx_cmdq_post(sc, cqe, 0);
6644 	error = mcx_cmdq_poll(sc, cqe, 1000);
6645 	if (error != 0) {
6646 		printf("%s: query cq timeout\n", DEVNAME(sc));
6647 		goto free;
6648 	}
6649 	if (mcx_cmdq_verify(cqe) != 0) {
6650 		printf("%s: query cq reply corrupt\n", DEVNAME(sc));
6651 		goto free;
6652 	}
6653 
6654 	out = mcx_cmdq_out(cqe);
6655 	switch (out->cmd_status) {
6656 	case MCX_CQ_STATUS_OK:
6657 		break;
6658 	default:
6659 		printf("%s: query qc failed (%x/%x)\n", DEVNAME(sc),
6660 		    out->cmd_status, be32toh(out->cmd_syndrome));
6661 		error = -1;
6662 		goto free;
6663 	}
6664 
6665 	ctx = (struct mcx_cq_ctx *)(mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)));
6666 	memcpy(cq_ctx, ctx, sizeof(*cq_ctx));
6667 free:
6668 	mcx_dmamem_free(sc, &mxm);
6669 	return (error);
6670 }
6671 
6672 int
6673 mcx_query_eq(struct mcx_softc *sc, struct mcx_eq *eq, struct mcx_eq_ctx *eq_ctx)
6674 {
6675 	struct mcx_dmamem mxm;
6676 	struct mcx_cmdq_entry *cqe;
6677 	struct mcx_cmd_query_eq_in *in;
6678 	struct mcx_cmd_query_eq_out *out;
6679 	struct mcq_eq_ctx *ctx;
6680 	uint8_t token = mcx_cmdq_token(sc);
6681 	int error;
6682 
6683 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
6684 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out) + sizeof(*ctx) + 16,
6685 	    token);
6686 
6687 	in = mcx_cmdq_in(cqe);
6688 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_EQ);
6689 	in->cmd_op_mod = htobe16(0);
6690 	in->cmd_eqn = htobe32(eq->eq_n);
6691 
6692 	CTASSERT(sizeof(*ctx) <= MCX_CMDQ_MAILBOX_DATASIZE*2);
6693 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2,
6694 	    &cqe->cq_output_ptr, token) != 0) {
6695 		printf("%s: unable to allocate query eq mailboxes\n",
6696 		    DEVNAME(sc));
6697 		return (-1);
6698 	}
6699 
6700 	mcx_cmdq_mboxes_sign(&mxm, 1);
6701 
6702 	mcx_cmdq_post(sc, cqe, 0);
6703 	error = mcx_cmdq_poll(sc, cqe, 1000);
6704 	if (error != 0) {
6705 		printf("%s: query eq timeout\n", DEVNAME(sc));
6706 		goto free;
6707 	}
6708 	if (mcx_cmdq_verify(cqe) != 0) {
6709 		printf("%s: query eq reply corrupt\n", DEVNAME(sc));
6710 		goto free;
6711 	}
6712 
6713 	out = mcx_cmdq_out(cqe);
6714 	switch (out->cmd_status) {
6715 	case MCX_CQ_STATUS_OK:
6716 		break;
6717 	default:
6718 		printf("%s: query eq failed (%x/%x)\n", DEVNAME(sc),
6719 		    out->cmd_status, be32toh(out->cmd_syndrome));
6720 		error = -1;
6721 		goto free;
6722 	}
6723 
6724 	ctx = (struct mcx_eq_ctx *)(mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)));
6725 	memcpy(eq_ctx, ctx, sizeof(*eq_ctx));
6726 free:
6727 	mcx_dmamem_free(sc, &mxm);
6728 	return (error);
6729 }
6730 
6731 #endif /* NKSTAT > 0 */
6732 
6733 
6734 static inline unsigned int
6735 mcx_rx_fill_slots(struct mcx_softc *sc, struct mcx_rx *rx, uint nslots)
6736 {
6737 	struct mcx_rq_entry *ring, *rqe;
6738 	struct mcx_slot *ms;
6739 	struct mbuf *m;
6740 	uint slot, p, fills;
6741 
6742 	ring = MCX_DMA_KVA(&rx->rx_rq_mem);
6743 	p = rx->rx_prod;
6744 
6745 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&rx->rx_rq_mem),
6746 	    0, MCX_DMA_LEN(&rx->rx_rq_mem), BUS_DMASYNC_POSTWRITE);
6747 
6748 	slot = (p % (1 << MCX_LOG_RQ_SIZE));
6749 	rqe = ring;
6750 	for (fills = 0; fills < nslots; fills++) {
6751 		slot = p % (1 << MCX_LOG_RQ_SIZE);
6752 
6753 		ms = &rx->rx_slots[slot];
6754 		rqe = &ring[slot];
6755 
6756 		m = NULL;
6757 		MGETHDR(m, M_DONTWAIT, MT_DATA);
6758 		if (m == NULL)
6759 			break;
6760 
6761 		MCLGET(m, M_DONTWAIT);
6762 		if ((m->m_flags & M_EXT) == 0) {
6763 			m_freem(m);
6764 			break;
6765 		}
6766 
6767 		m->m_len = m->m_pkthdr.len = sc->sc_hardmtu;
6768 		m_adj(m, m->m_ext.ext_size - sc->sc_rxbufsz);
6769 		m_adj(m, ETHER_ALIGN);
6770 
6771 		if (bus_dmamap_load_mbuf(sc->sc_dmat, ms->ms_map, m,
6772 		    BUS_DMA_NOWAIT) != 0) {
6773 			m_freem(m);
6774 			break;
6775 		}
6776 		bus_dmamap_sync(sc->sc_dmat, ms->ms_map, 0, ms->ms_map->dm_mapsize, BUS_DMASYNC_PREREAD);
6777 		ms->ms_m = m;
6778 
6779 		be32enc(&rqe->rqe_byte_count, ms->ms_map->dm_segs[0].ds_len);
6780 		be64enc(&rqe->rqe_addr, ms->ms_map->dm_segs[0].ds_addr);
6781 		be32enc(&rqe->rqe_lkey, sc->sc_lkey);
6782 
6783 		p++;
6784 	}
6785 
6786 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&rx->rx_rq_mem),
6787 	    0, MCX_DMA_LEN(&rx->rx_rq_mem), BUS_DMASYNC_PREWRITE);
6788 
6789 	rx->rx_prod = p;
6790 
6791 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_doorbell_mem),
6792 	    rx->rx_doorbell, sizeof(uint32_t), BUS_DMASYNC_POSTWRITE);
6793 	be32enc(MCX_DMA_OFF(&sc->sc_doorbell_mem, rx->rx_doorbell),
6794 	    p & MCX_WQ_DOORBELL_MASK);
6795 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_doorbell_mem),
6796 	    rx->rx_doorbell, sizeof(uint32_t), BUS_DMASYNC_PREWRITE);
6797 
6798 	return (nslots - fills);
6799 }
6800 
6801 static int
6802 mcx_rx_fill(struct mcx_softc *sc, struct mcx_rx *rx)
6803 {
6804 	u_int slots;
6805 
6806 	slots = mcx_rxr_get(&rx->rx_rxr, (1 << MCX_LOG_RQ_SIZE));
6807 	if (slots == 0)
6808 		return (1);
6809 
6810 	slots = mcx_rx_fill_slots(sc, rx, slots);
6811 	mcx_rxr_put(&rx->rx_rxr, slots);
6812 	return (0);
6813 }
6814 
6815 void
6816 mcx_refill(void *xrx)
6817 {
6818 	struct mcx_rx *rx = xrx;
6819 	struct mcx_softc *sc = rx->rx_softc;
6820 
6821 	mcx_rx_fill(sc, rx);
6822 
6823 	if (mcx_rxr_inuse(&rx->rx_rxr) == 0)
6824 		callout_schedule(&rx->rx_refill, 1);
6825 }
6826 
6827 static int
6828 mcx_process_txeof(struct mcx_softc *sc, struct mcx_tx *tx,
6829     struct mcx_cq_entry *cqe)
6830 {
6831 	struct mcx_slot *ms;
6832 	bus_dmamap_t map;
6833 	int slot, slots;
6834 
6835 	slot = be16toh(cqe->cq_wqe_count) % (1 << MCX_LOG_SQ_SIZE);
6836 
6837 	ms = &tx->tx_slots[slot];
6838 	map = ms->ms_map;
6839 	bus_dmamap_sync(sc->sc_dmat, map, 0, map->dm_mapsize,
6840 	    BUS_DMASYNC_POSTWRITE);
6841 
6842 	slots = 1;
6843 	if (map->dm_nsegs > 1)
6844 		slots += (map->dm_nsegs+2) / MCX_SQ_SEGS_PER_SLOT;
6845 
6846 	bus_dmamap_unload(sc->sc_dmat, map);
6847 	m_freem(ms->ms_m);
6848 	ms->ms_m = NULL;
6849 
6850 	return (slots);
6851 }
6852 
6853 static uint64_t
6854 mcx_uptime(void)
6855 {
6856 	struct timespec ts;
6857 
6858 	nanouptime(&ts);
6859 
6860 	return ((uint64_t)ts.tv_sec * 1000000000 + (uint64_t)ts.tv_nsec);
6861 }
6862 
6863 static void
6864 mcx_calibrate_first(struct mcx_softc *sc)
6865 {
6866 	struct mcx_calibration *c = &sc->sc_calibration[0];
6867 	int s;
6868 
6869 	sc->sc_calibration_gen = 0;
6870 
6871 	s = splhigh(); /* crit_enter? */
6872 	c->c_ubase = mcx_uptime();
6873 	c->c_tbase = mcx_timer(sc);
6874 	splx(s);
6875 	c->c_ratio = 0;
6876 
6877 #if notyet
6878 	callout_schedule(&sc->sc_calibrate, MCX_CALIBRATE_FIRST * hz);
6879 #endif
6880 }
6881 
6882 #define MCX_TIMESTAMP_SHIFT 24
6883 
6884 static void
6885 mcx_calibrate(void *arg)
6886 {
6887 	struct mcx_softc *sc = arg;
6888 	struct mcx_calibration *nc, *pc;
6889 	uint64_t udiff, tdiff;
6890 	unsigned int gen;
6891 	int s;
6892 
6893 	if (!ISSET(sc->sc_ec.ec_if.if_flags, IFF_RUNNING))
6894 		return;
6895 
6896 	callout_schedule(&sc->sc_calibrate, MCX_CALIBRATE_NORMAL * hz);
6897 
6898 	gen = sc->sc_calibration_gen;
6899 	pc = &sc->sc_calibration[gen % __arraycount(sc->sc_calibration)];
6900 	gen++;
6901 	nc = &sc->sc_calibration[gen % __arraycount(sc->sc_calibration)];
6902 
6903 	nc->c_uptime = pc->c_ubase;
6904 	nc->c_timestamp = pc->c_tbase;
6905 
6906 	s = splhigh(); /* crit_enter? */
6907 	nc->c_ubase = mcx_uptime();
6908 	nc->c_tbase = mcx_timer(sc);
6909 	splx(s);
6910 
6911 	udiff = nc->c_ubase - nc->c_uptime;
6912 	tdiff = nc->c_tbase - nc->c_timestamp;
6913 
6914 	/*
6915 	 * udiff is the wall clock time between calibration ticks,
6916 	 * which should be 32 seconds or 32 billion nanoseconds. if
6917 	 * we squint, 1 billion nanoseconds is kind of like a 32 bit
6918 	 * number, so 32 billion should still have a lot of high bits
6919 	 * spare. we use this space by shifting the nanoseconds up
6920 	 * 24 bits so we have a nice big number to divide by the
6921 	 * number of mcx timer ticks.
6922 	 */
6923 	nc->c_ratio = (udiff << MCX_TIMESTAMP_SHIFT) / tdiff;
6924 
6925 	membar_producer();
6926 	sc->sc_calibration_gen = gen;
6927 }
6928 
6929 static int
6930 mcx_process_rx(struct mcx_softc *sc, struct mcx_rx *rx,
6931     struct mcx_cq_entry *cqe, struct mcx_mbufq *mq,
6932     const struct mcx_calibration *c)
6933 {
6934 	struct ifnet *ifp = &sc->sc_ec.ec_if;
6935 	struct mcx_slot *ms;
6936 	struct mbuf *m;
6937 	uint32_t flags, len;
6938 	int slot;
6939 
6940 	len = be32dec(&cqe->cq_byte_cnt);
6941 	slot = be16toh(cqe->cq_wqe_count) % (1 << MCX_LOG_RQ_SIZE);
6942 
6943 	ms = &rx->rx_slots[slot];
6944 	bus_dmamap_sync(sc->sc_dmat, ms->ms_map, 0, len, BUS_DMASYNC_POSTREAD);
6945 	bus_dmamap_unload(sc->sc_dmat, ms->ms_map);
6946 
6947 	m = ms->ms_m;
6948 	ms->ms_m = NULL;
6949 
6950 	m_set_rcvif(m, &sc->sc_ec.ec_if);
6951 	m->m_pkthdr.len = m->m_len = len;
6952 
6953 #if 0
6954 	if (cqe->cq_rx_hash_type) {
6955 		m->m_pkthdr.ph_flowid = be32toh(cqe->cq_rx_hash);
6956 		m->m_pkthdr.csum_flags |= M_FLOWID;
6957 	}
6958 #endif
6959 
6960 	flags = be32dec(&cqe->cq_flags);
6961 	if (flags & MCX_CQ_ENTRY_FLAGS_L3_OK) {
6962 		if (ifp->if_capenable & IFCAP_CSUM_IPv4_Rx)
6963 			m->m_pkthdr.csum_flags |= M_CSUM_IPv4;
6964 	}
6965 	if (flags & MCX_CQ_ENTRY_FLAGS_L4_OK) {
6966 		if (ifp->if_capenable & IFCAP_CSUM_TCPv4_Rx)
6967 			m->m_pkthdr.csum_flags |= M_CSUM_TCPv4;
6968 		if (ifp->if_capenable & IFCAP_CSUM_TCPv6_Rx)
6969 			m->m_pkthdr.csum_flags |= M_CSUM_TCPv6;
6970 		if (ifp->if_capenable & IFCAP_CSUM_UDPv4_Rx)
6971 			m->m_pkthdr.csum_flags |= M_CSUM_UDPv4;
6972 		if (ifp->if_capenable & IFCAP_CSUM_UDPv6_Rx)
6973 			m->m_pkthdr.csum_flags |= M_CSUM_UDPv6;
6974 	}
6975 	if (flags & MCX_CQ_ENTRY_FLAGS_CV) {
6976 		vlan_set_tag(m, flags & MCX_CQ_ENTRY_FLAGS_VLAN_MASK);
6977 	}
6978 
6979 #if notyet
6980 	if (ISSET(sc->sc_ec.ec_if.if_flags, IFF_LINK0) && c->c_ratio) {
6981 		uint64_t t = be64dec(&cqe->cq_timestamp);
6982 		t -= c->c_timestamp;
6983 		t *= c->c_ratio;
6984 		t >>= MCX_TIMESTAMP_SHIFT;
6985 		t += c->c_uptime;
6986 
6987 		m->m_pkthdr.ph_timestamp = t;
6988 		SET(m->m_pkthdr.csum_flags, M_TIMESTAMP);
6989 	}
6990 #endif
6991 
6992 	MBUFQ_ENQUEUE(mq, m);
6993 
6994 	return (1);
6995 }
6996 
6997 static struct mcx_cq_entry *
6998 mcx_next_cq_entry(struct mcx_softc *sc, struct mcx_cq *cq)
6999 {
7000 	struct mcx_cq_entry *cqe;
7001 	int next;
7002 
7003 	cqe = (struct mcx_cq_entry *)MCX_DMA_KVA(&cq->cq_mem);
7004 	next = cq->cq_cons % (1 << MCX_LOG_CQ_SIZE);
7005 
7006 	if ((cqe[next].cq_opcode_owner & MCX_CQ_ENTRY_FLAG_OWNER) ==
7007 	    ((cq->cq_cons >> MCX_LOG_CQ_SIZE) & 1)) {
7008 		return (&cqe[next]);
7009 	}
7010 
7011 	return (NULL);
7012 }
7013 
7014 static void
7015 mcx_arm_cq(struct mcx_softc *sc, struct mcx_cq *cq, int uar)
7016 {
7017 	struct mcx_cq_doorbell *db;
7018 	bus_size_t offset;
7019 	uint32_t val;
7020 	uint64_t uval;
7021 
7022 	val = ((cq->cq_count) & 3) << MCX_CQ_DOORBELL_ARM_CMD_SN_SHIFT;
7023 	val |= (cq->cq_cons & MCX_CQ_DOORBELL_ARM_CI_MASK);
7024 
7025 	db = MCX_DMA_OFF(&sc->sc_doorbell_mem, cq->cq_doorbell);
7026 
7027 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_doorbell_mem),
7028 	    cq->cq_doorbell, sizeof(*db), BUS_DMASYNC_POSTWRITE);
7029 
7030 	be32enc(&db->db_update_ci, cq->cq_cons & MCX_CQ_DOORBELL_ARM_CI_MASK);
7031 	be32enc(&db->db_arm_ci, val);
7032 
7033 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_doorbell_mem),
7034 	    cq->cq_doorbell, sizeof(*db), BUS_DMASYNC_PREWRITE);
7035 
7036 	offset = (MCX_PAGE_SIZE * uar) + MCX_UAR_CQ_DOORBELL;
7037 
7038 	uval = (uint64_t)val << 32;
7039 	uval |= cq->cq_n;
7040 
7041 	bus_space_write_8(sc->sc_memt, sc->sc_memh, offset, htobe64(uval));
7042 	mcx_bar(sc, offset, sizeof(uval), BUS_SPACE_BARRIER_WRITE);
7043 }
7044 
7045 void
7046 mcx_process_cq(struct mcx_softc *sc, struct mcx_queues *q, struct mcx_cq *cq)
7047 {
7048 	struct mcx_rx *rx = &q->q_rx;
7049 	struct mcx_tx *tx = &q->q_tx;
7050 	struct ifnet *ifp = &sc->sc_ec.ec_if;
7051 	const struct mcx_calibration *c;
7052 	unsigned int gen;
7053 	struct mcx_cq_entry *cqe;
7054 	struct mcx_mbufq mq;
7055 	struct mbuf *m;
7056 	int rxfree, txfree;
7057 
7058 	MBUFQ_INIT(&mq);
7059 
7060 	gen = sc->sc_calibration_gen;
7061 	membar_consumer();
7062 	c = &sc->sc_calibration[gen % __arraycount(sc->sc_calibration)];
7063 
7064 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&cq->cq_mem),
7065 	    0, MCX_DMA_LEN(&cq->cq_mem), BUS_DMASYNC_POSTREAD);
7066 
7067 	rxfree = 0;
7068 	txfree = 0;
7069 	while ((cqe = mcx_next_cq_entry(sc, cq))) {
7070 		uint8_t opcode;
7071 		opcode = (cqe->cq_opcode_owner >> MCX_CQ_ENTRY_OPCODE_SHIFT);
7072 		switch (opcode) {
7073 		case MCX_CQ_ENTRY_OPCODE_REQ:
7074 			txfree += mcx_process_txeof(sc, tx, cqe);
7075 			break;
7076 		case MCX_CQ_ENTRY_OPCODE_SEND:
7077 			rxfree += mcx_process_rx(sc, rx, cqe, &mq, c);
7078 			break;
7079 		case MCX_CQ_ENTRY_OPCODE_REQ_ERR:
7080 		case MCX_CQ_ENTRY_OPCODE_SEND_ERR:
7081 			/* uint8_t *cqp = (uint8_t *)cqe; */
7082 			/* printf("%s: cq completion error: %x\n",
7083 			    DEVNAME(sc), cqp[0x37]); */
7084 			break;
7085 
7086 		default:
7087 			/* printf("%s: cq completion opcode %x??\n",
7088 			    DEVNAME(sc), opcode); */
7089 			break;
7090 		}
7091 
7092 		cq->cq_cons++;
7093 	}
7094 
7095 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&cq->cq_mem),
7096 	    0, MCX_DMA_LEN(&cq->cq_mem), BUS_DMASYNC_PREREAD);
7097 
7098 	if (rxfree > 0) {
7099 		mcx_rxr_put(&rx->rx_rxr, rxfree);
7100 		while (MBUFQ_FIRST(&mq) != NULL) {
7101 			MBUFQ_DEQUEUE(&mq, m);
7102 			if_percpuq_enqueue(ifp->if_percpuq, m);
7103 		}
7104 
7105 		mcx_rx_fill(sc, rx);
7106 		if (mcx_rxr_inuse(&rx->rx_rxr) == 0)
7107 			callout_schedule(&rx->rx_refill, 1);
7108 	}
7109 
7110 	cq->cq_count++;
7111 	mcx_arm_cq(sc, cq, q->q_uar);
7112 
7113 	if (txfree > 0) {
7114 		tx->tx_cons += txfree;
7115 		if_schedule_deferred_start(ifp);
7116 	}
7117 }
7118 
7119 
7120 static void
7121 mcx_arm_eq(struct mcx_softc *sc, struct mcx_eq *eq, int uar)
7122 {
7123 	bus_size_t offset;
7124 	uint32_t val;
7125 
7126 	offset = (MCX_PAGE_SIZE * uar) + MCX_UAR_EQ_DOORBELL_ARM;
7127 	val = (eq->eq_n << 24) | (eq->eq_cons & 0xffffff);
7128 
7129 	mcx_wr(sc, offset, val);
7130 	mcx_bar(sc, offset, sizeof(val), BUS_SPACE_BARRIER_WRITE);
7131 }
7132 
7133 static struct mcx_eq_entry *
7134 mcx_next_eq_entry(struct mcx_softc *sc, struct mcx_eq *eq)
7135 {
7136 	struct mcx_eq_entry *eqe;
7137 	int next;
7138 
7139 	eqe = (struct mcx_eq_entry *)MCX_DMA_KVA(&eq->eq_mem);
7140 	next = eq->eq_cons % (1 << MCX_LOG_EQ_SIZE);
7141 	if ((eqe[next].eq_owner & 1) ==
7142 	    ((eq->eq_cons >> MCX_LOG_EQ_SIZE) & 1)) {
7143 		eq->eq_cons++;
7144 		return (&eqe[next]);
7145 	}
7146 	return (NULL);
7147 }
7148 
7149 int
7150 mcx_admin_intr(void *xsc)
7151 {
7152 	struct mcx_softc *sc = (struct mcx_softc *)xsc;
7153 	struct mcx_eq *eq = &sc->sc_admin_eq;
7154 	struct mcx_eq_entry *eqe;
7155 
7156 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&eq->eq_mem),
7157 	    0, MCX_DMA_LEN(&eq->eq_mem), BUS_DMASYNC_POSTREAD);
7158 
7159 	while ((eqe = mcx_next_eq_entry(sc, eq)) != NULL) {
7160 		switch (eqe->eq_event_type) {
7161 		case MCX_EVENT_TYPE_LAST_WQE:
7162 			/* printf("%s: last wqe reached?\n", DEVNAME(sc)); */
7163 			break;
7164 
7165 		case MCX_EVENT_TYPE_CQ_ERROR:
7166 			/* printf("%s: cq error\n", DEVNAME(sc)); */
7167 			break;
7168 
7169 		case MCX_EVENT_TYPE_CMD_COMPLETION:
7170 			/* wakeup probably */
7171 			break;
7172 
7173 		case MCX_EVENT_TYPE_PORT_CHANGE:
7174 			workqueue_enqueue(sc->sc_workq, &sc->sc_port_change, NULL);
7175 			break;
7176 
7177 		default:
7178 			/* printf("%s: something happened\n", DEVNAME(sc)); */
7179 			break;
7180 		}
7181 	}
7182 
7183 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&eq->eq_mem),
7184 	    0, MCX_DMA_LEN(&eq->eq_mem), BUS_DMASYNC_PREREAD);
7185 
7186 	mcx_arm_eq(sc, eq, sc->sc_uar);
7187 
7188 	return (1);
7189 }
7190 
7191 int
7192 mcx_cq_intr(void *xq)
7193 {
7194 	struct mcx_queues *q = (struct mcx_queues *)xq;
7195 	struct mcx_softc *sc = q->q_sc;
7196 	struct mcx_eq *eq = &q->q_eq;
7197 	struct mcx_eq_entry *eqe;
7198 	int cqn;
7199 
7200 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&eq->eq_mem),
7201 	    0, MCX_DMA_LEN(&eq->eq_mem), BUS_DMASYNC_POSTREAD);
7202 
7203 	while ((eqe = mcx_next_eq_entry(sc, eq)) != NULL) {
7204 		switch (eqe->eq_event_type) {
7205 		case MCX_EVENT_TYPE_COMPLETION:
7206 			cqn = be32toh(eqe->eq_event_data[6]);
7207 			if (cqn == q->q_cq.cq_n)
7208 				mcx_process_cq(sc, q, &q->q_cq);
7209 			break;
7210 		}
7211 	}
7212 
7213 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&eq->eq_mem),
7214 	    0, MCX_DMA_LEN(&eq->eq_mem), BUS_DMASYNC_PREREAD);
7215 
7216 	mcx_arm_eq(sc, eq, q->q_uar);
7217 
7218 	return (1);
7219 }
7220 
7221 static void
7222 mcx_free_slots(struct mcx_softc *sc, struct mcx_slot *slots, int allocated,
7223     int total)
7224 {
7225 	struct mcx_slot *ms;
7226 
7227 	int i = allocated;
7228 	while (i-- > 0) {
7229 		ms = &slots[i];
7230 		bus_dmamap_destroy(sc->sc_dmat, ms->ms_map);
7231 		if (ms->ms_m != NULL)
7232 			m_freem(ms->ms_m);
7233 	}
7234 	kmem_free(slots, total * sizeof(*ms));
7235 }
7236 
7237 static int
7238 mcx_queue_up(struct mcx_softc *sc, struct mcx_queues *q)
7239 {
7240 	struct mcx_rx *rx;
7241 	struct mcx_tx *tx;
7242 	struct mcx_slot *ms;
7243 	int i;
7244 
7245 	rx = &q->q_rx;
7246 	rx->rx_slots = kmem_zalloc(sizeof(*ms) * (1 << MCX_LOG_RQ_SIZE),
7247 	    KM_SLEEP);
7248 
7249 	for (i = 0; i < (1 << MCX_LOG_RQ_SIZE); i++) {
7250 		ms = &rx->rx_slots[i];
7251 		if (bus_dmamap_create(sc->sc_dmat, sc->sc_hardmtu, 1,
7252 		    sc->sc_hardmtu, 0,
7253 		    BUS_DMA_WAITOK | BUS_DMA_ALLOCNOW,
7254 		    &ms->ms_map) != 0) {
7255 			printf("%s: failed to allocate rx dma maps\n",
7256 			    DEVNAME(sc));
7257 			goto destroy_rx_slots;
7258 		}
7259 	}
7260 
7261 	tx = &q->q_tx;
7262 	tx->tx_slots = kmem_zalloc(sizeof(*ms) * (1 << MCX_LOG_SQ_SIZE),
7263 	     KM_SLEEP);
7264 
7265 	for (i = 0; i < (1 << MCX_LOG_SQ_SIZE); i++) {
7266 		ms = &tx->tx_slots[i];
7267 		if (bus_dmamap_create(sc->sc_dmat, sc->sc_hardmtu,
7268 		    MCX_SQ_MAX_SEGMENTS, sc->sc_hardmtu, 0,
7269 		    BUS_DMA_WAITOK | BUS_DMA_ALLOCNOW,
7270 		    &ms->ms_map) != 0) {
7271 			printf("%s: failed to allocate tx dma maps\n",
7272 			    DEVNAME(sc));
7273 			goto destroy_tx_slots;
7274 		}
7275 	}
7276 
7277 	if (mcx_create_cq(sc, &q->q_cq, q->q_uar, q->q_index,
7278 	    q->q_eq.eq_n) != 0)
7279 		goto destroy_tx_slots;
7280 
7281 	if (mcx_create_sq(sc, tx, q->q_uar, q->q_index, q->q_cq.cq_n)
7282 	    != 0)
7283 		goto destroy_cq;
7284 
7285 	if (mcx_create_rq(sc, rx, q->q_index, q->q_cq.cq_n) != 0)
7286 		goto destroy_sq;
7287 
7288 	return 0;
7289 
7290 destroy_sq:
7291 	mcx_destroy_sq(sc, tx);
7292 destroy_cq:
7293 	mcx_destroy_cq(sc, &q->q_cq);
7294 destroy_tx_slots:
7295 	mcx_free_slots(sc, tx->tx_slots, i, (1 << MCX_LOG_SQ_SIZE));
7296 	tx->tx_slots = NULL;
7297 
7298 	i = (1 << MCX_LOG_RQ_SIZE);
7299 destroy_rx_slots:
7300 	mcx_free_slots(sc, rx->rx_slots, i, (1 << MCX_LOG_RQ_SIZE));
7301 	rx->rx_slots = NULL;
7302 	return ENOMEM;
7303 }
7304 
7305 static int
7306 mcx_rss_group_entry_count(struct mcx_softc *sc, int group)
7307 {
7308 	int i;
7309 	int count;
7310 
7311 	count = 0;
7312 	for (i = 0; i < __arraycount(mcx_rss_config); i++) {
7313 		if (mcx_rss_config[i].flow_group == group)
7314 			count++;
7315 	}
7316 
7317 	return count;
7318 }
7319 
7320 static int
7321 mcx_init(struct ifnet *ifp)
7322 {
7323 	struct mcx_softc *sc = ifp->if_softc;
7324 	struct mcx_rx *rx;
7325 	struct mcx_tx *tx;
7326 	int i, start, count, flow_group, flow_index;
7327 	struct mcx_flow_match match_crit;
7328 	struct mcx_rss_rule *rss;
7329 	uint32_t dest;
7330 	int rqns[MCX_MAX_QUEUES] = { 0 };
7331 
7332 	if (ISSET(ifp->if_flags, IFF_RUNNING))
7333 		mcx_stop(ifp, 0);
7334 
7335 	if (mcx_create_tis(sc, &sc->sc_tis) != 0)
7336 		goto down;
7337 
7338 	for (i = 0; i < sc->sc_nqueues; i++) {
7339 		if (mcx_queue_up(sc, &sc->sc_queues[i]) != 0) {
7340 			goto down;
7341 		}
7342 	}
7343 
7344 	/* RSS flow table and flow groups */
7345 	if (mcx_create_flow_table(sc, MCX_LOG_FLOW_TABLE_SIZE, 1,
7346 	    &sc->sc_rss_flow_table_id) != 0)
7347 		goto down;
7348 
7349 	dest = MCX_FLOW_CONTEXT_DEST_TYPE_TABLE |
7350 	    sc->sc_rss_flow_table_id;
7351 
7352 	/* L4 RSS flow group (v4/v6 tcp/udp, no fragments) */
7353 	memset(&match_crit, 0, sizeof(match_crit));
7354 	match_crit.mc_ethertype = 0xffff;
7355 	match_crit.mc_ip_proto = 0xff;
7356 	match_crit.mc_vlan_flags = MCX_FLOW_MATCH_IP_FRAG;
7357 	start = 0;
7358 	count = mcx_rss_group_entry_count(sc, MCX_FLOW_GROUP_RSS_L4);
7359 	if (count != 0) {
7360 		if (mcx_create_flow_group(sc, sc->sc_rss_flow_table_id,
7361 		    MCX_FLOW_GROUP_RSS_L4, start, count,
7362 		    MCX_CREATE_FLOW_GROUP_CRIT_OUTER, &match_crit) != 0)
7363 			goto down;
7364 		start += count;
7365 	}
7366 
7367 	/* L3 RSS flow group (v4/v6, including fragments) */
7368 	memset(&match_crit, 0, sizeof(match_crit));
7369 	match_crit.mc_ethertype = 0xffff;
7370 	count = mcx_rss_group_entry_count(sc, MCX_FLOW_GROUP_RSS_L3);
7371 	if (mcx_create_flow_group(sc, sc->sc_rss_flow_table_id,
7372 	    MCX_FLOW_GROUP_RSS_L3, start, count,
7373 	    MCX_CREATE_FLOW_GROUP_CRIT_OUTER, &match_crit) != 0)
7374 		goto down;
7375 	start += count;
7376 
7377 	/* non-RSS flow group */
7378 	count = mcx_rss_group_entry_count(sc, MCX_FLOW_GROUP_RSS_NONE);
7379 	memset(&match_crit, 0, sizeof(match_crit));
7380 	if (mcx_create_flow_group(sc, sc->sc_rss_flow_table_id,
7381 	    MCX_FLOW_GROUP_RSS_NONE, start, count, 0, &match_crit) != 0)
7382 		goto down;
7383 
7384 	/* Root flow table, matching packets based on mac address */
7385 	if (mcx_create_flow_table(sc, MCX_LOG_FLOW_TABLE_SIZE, 0,
7386 	    &sc->sc_mac_flow_table_id) != 0)
7387 		goto down;
7388 
7389 	/* promisc flow group */
7390 	start = 0;
7391 	memset(&match_crit, 0, sizeof(match_crit));
7392 	if (mcx_create_flow_group(sc, sc->sc_mac_flow_table_id,
7393 	    MCX_FLOW_GROUP_PROMISC, start, 1, 0, &match_crit) != 0)
7394 		goto down;
7395 	sc->sc_promisc_flow_enabled = 0;
7396 	start++;
7397 
7398 	/* all multicast flow group */
7399 	match_crit.mc_dest_mac[0] = 0x01;
7400 	if (mcx_create_flow_group(sc, sc->sc_mac_flow_table_id,
7401 	    MCX_FLOW_GROUP_ALLMULTI, start, 1,
7402 	    MCX_CREATE_FLOW_GROUP_CRIT_OUTER, &match_crit) != 0)
7403 		goto down;
7404 	sc->sc_allmulti_flow_enabled = 0;
7405 	start++;
7406 
7407 	/* mac address matching flow group */
7408 	memset(&match_crit.mc_dest_mac, 0xff, sizeof(match_crit.mc_dest_mac));
7409 	if (mcx_create_flow_group(sc, sc->sc_mac_flow_table_id,
7410 	    MCX_FLOW_GROUP_MAC, start, (1 << MCX_LOG_FLOW_TABLE_SIZE) - start,
7411 	    MCX_CREATE_FLOW_GROUP_CRIT_OUTER, &match_crit) != 0)
7412 		goto down;
7413 
7414 	/* flow table entries for unicast and broadcast */
7415 	start = 0;
7416 	if (mcx_set_flow_table_entry_mac(sc, MCX_FLOW_GROUP_MAC, start,
7417 	    LLADDR(satosdl(ifp->if_dl->ifa_addr)), dest) != 0)
7418 		goto down;
7419 	start++;
7420 
7421 	if (mcx_set_flow_table_entry_mac(sc, MCX_FLOW_GROUP_MAC, start,
7422 	    etherbroadcastaddr, dest) != 0)
7423 		goto down;
7424 	start++;
7425 
7426 	/* multicast entries go after that */
7427 	sc->sc_mcast_flow_base = start;
7428 
7429 	/* re-add any existing multicast flows */
7430 	for (i = 0; i < MCX_NUM_MCAST_FLOWS; i++) {
7431 		if (sc->sc_mcast_flows[i][0] != 0) {
7432 			mcx_set_flow_table_entry_mac(sc, MCX_FLOW_GROUP_MAC,
7433 			    sc->sc_mcast_flow_base + i,
7434 			    sc->sc_mcast_flows[i], dest);
7435 		}
7436 	}
7437 
7438 	if (mcx_set_flow_table_root(sc, sc->sc_mac_flow_table_id) != 0)
7439 		goto down;
7440 
7441 	/*
7442 	 * the RQT can be any size as long as it's a power of two.
7443 	 * since we also restrict the number of queues to a power of two,
7444 	 * we can just put each rx queue in once.
7445 	 */
7446 	for (i = 0; i < sc->sc_nqueues; i++)
7447 		rqns[i] = sc->sc_queues[i].q_rx.rx_rqn;
7448 
7449 	if (mcx_create_rqt(sc, sc->sc_nqueues, rqns, &sc->sc_rqt) != 0)
7450 		goto down;
7451 
7452 	start = 0;
7453 	flow_index = 0;
7454 	flow_group = -1;
7455 	for (i = 0; i < __arraycount(mcx_rss_config); i++) {
7456 		rss = &mcx_rss_config[i];
7457 		if (rss->flow_group != flow_group) {
7458 			flow_group = rss->flow_group;
7459 			flow_index = 0;
7460 		}
7461 
7462 		if (rss->hash_sel == 0) {
7463 			if (mcx_create_tir_direct(sc, &sc->sc_queues[0].q_rx,
7464 			    &sc->sc_tir[i]) != 0)
7465 				goto down;
7466 		} else {
7467 			if (mcx_create_tir_indirect(sc, sc->sc_rqt,
7468 			    rss->hash_sel, &sc->sc_tir[i]) != 0)
7469 				goto down;
7470 		}
7471 
7472 		if (mcx_set_flow_table_entry_proto(sc, flow_group,
7473 		    flow_index, rss->ethertype, rss->ip_proto,
7474 		    MCX_FLOW_CONTEXT_DEST_TYPE_TIR | sc->sc_tir[i]) != 0)
7475 			goto down;
7476 		flow_index++;
7477 	}
7478 
7479 	for (i = 0; i < sc->sc_nqueues; i++) {
7480 		struct mcx_queues *q = &sc->sc_queues[i];
7481 		rx = &q->q_rx;
7482 		tx = &q->q_tx;
7483 
7484 		/* start the queues */
7485 		if (mcx_ready_sq(sc, tx) != 0)
7486 			goto down;
7487 
7488 		if (mcx_ready_rq(sc, rx) != 0)
7489 			goto down;
7490 
7491 		mcx_rxr_init(&rx->rx_rxr, 1, (1 << MCX_LOG_RQ_SIZE));
7492 		rx->rx_prod = 0;
7493 		mcx_rx_fill(sc, rx);
7494 
7495 		tx->tx_cons = 0;
7496 		tx->tx_prod = 0;
7497 	}
7498 
7499 	mcx_calibrate_first(sc);
7500 
7501 	SET(ifp->if_flags, IFF_RUNNING);
7502 	CLR(ifp->if_flags, IFF_OACTIVE);
7503 	if_schedule_deferred_start(ifp);
7504 
7505 	return 0;
7506 down:
7507 	mcx_stop(ifp, 0);
7508 	return EIO;
7509 }
7510 
7511 static void
7512 mcx_stop(struct ifnet *ifp, int disable)
7513 {
7514 	struct mcx_softc *sc = ifp->if_softc;
7515 	struct mcx_rss_rule *rss;
7516 	int group, i, flow_group, flow_index;
7517 
7518 	CLR(ifp->if_flags, IFF_RUNNING);
7519 
7520 	/*
7521 	 * delete flow table entries first, so no packets can arrive
7522 	 * after the barriers
7523 	 */
7524 	if (sc->sc_promisc_flow_enabled)
7525 		mcx_delete_flow_table_entry(sc, MCX_FLOW_GROUP_PROMISC, 0);
7526 	if (sc->sc_allmulti_flow_enabled)
7527 		mcx_delete_flow_table_entry(sc, MCX_FLOW_GROUP_ALLMULTI, 0);
7528 	mcx_delete_flow_table_entry(sc, MCX_FLOW_GROUP_MAC, 0);
7529 	mcx_delete_flow_table_entry(sc, MCX_FLOW_GROUP_MAC, 1);
7530 	for (i = 0; i < MCX_NUM_MCAST_FLOWS; i++) {
7531 		if (sc->sc_mcast_flows[i][0] != 0) {
7532 			mcx_delete_flow_table_entry(sc, MCX_FLOW_GROUP_MAC,
7533 			    sc->sc_mcast_flow_base + i);
7534 		}
7535 	}
7536 
7537 	flow_group = -1;
7538 	flow_index = 0;
7539 	for (i = 0; i < __arraycount(mcx_rss_config); i++) {
7540 		rss = &mcx_rss_config[i];
7541 		if (rss->flow_group != flow_group) {
7542 			flow_group = rss->flow_group;
7543 			flow_index = 0;
7544 		}
7545 
7546 		mcx_delete_flow_table_entry(sc, flow_group, flow_index);
7547 
7548 		mcx_destroy_tir(sc, sc->sc_tir[i]);
7549 		sc->sc_tir[i] = 0;
7550 
7551 		flow_index++;
7552 	}
7553 
7554 	for (i = 0; i < sc->sc_nqueues; i++) {
7555 		callout_halt(&sc->sc_queues[i].q_rx.rx_refill, NULL);
7556 	}
7557 
7558 	callout_halt(&sc->sc_calibrate, NULL);
7559 
7560 	for (group = 0; group < MCX_NUM_FLOW_GROUPS; group++) {
7561 		if (sc->sc_flow_group[group].g_id != -1)
7562 			mcx_destroy_flow_group(sc, group);
7563 	}
7564 
7565 	if (sc->sc_mac_flow_table_id != -1) {
7566 		mcx_destroy_flow_table(sc, sc->sc_mac_flow_table_id);
7567 		sc->sc_mac_flow_table_id = -1;
7568 	}
7569 	if (sc->sc_rss_flow_table_id != -1) {
7570 		mcx_destroy_flow_table(sc, sc->sc_rss_flow_table_id);
7571 		sc->sc_rss_flow_table_id = -1;
7572 	}
7573 	if (sc->sc_rqt != -1) {
7574 		mcx_destroy_rqt(sc, sc->sc_rqt);
7575 		sc->sc_rqt = -1;
7576 	}
7577 
7578 	for (i = 0; i < sc->sc_nqueues; i++) {
7579 		struct mcx_queues *q = &sc->sc_queues[i];
7580 		struct mcx_rx *rx = &q->q_rx;
7581 		struct mcx_tx *tx = &q->q_tx;
7582 		struct mcx_cq *cq = &q->q_cq;
7583 
7584 		if (rx->rx_rqn != 0)
7585 			mcx_destroy_rq(sc, rx);
7586 
7587 		if (tx->tx_sqn != 0)
7588 			mcx_destroy_sq(sc, tx);
7589 
7590 		if (tx->tx_slots != NULL) {
7591 			mcx_free_slots(sc, tx->tx_slots,
7592 			    (1 << MCX_LOG_SQ_SIZE), (1 << MCX_LOG_SQ_SIZE));
7593 			tx->tx_slots = NULL;
7594 		}
7595 		if (rx->rx_slots != NULL) {
7596 			mcx_free_slots(sc, rx->rx_slots,
7597 			    (1 << MCX_LOG_RQ_SIZE), (1 << MCX_LOG_RQ_SIZE));
7598 			rx->rx_slots = NULL;
7599 		}
7600 
7601 		if (cq->cq_n != 0)
7602 			mcx_destroy_cq(sc, cq);
7603 	}
7604 	if (sc->sc_tis != 0) {
7605 		mcx_destroy_tis(sc, sc->sc_tis);
7606 		sc->sc_tis = 0;
7607 	}
7608 }
7609 
7610 static int
7611 mcx_ioctl(struct ifnet *ifp, u_long cmd, void *data)
7612 {
7613 	struct mcx_softc *sc = (struct mcx_softc *)ifp->if_softc;
7614 	struct ifreq *ifr = (struct ifreq *)data;
7615 	struct ethercom *ec = &sc->sc_ec;
7616 	uint8_t addrhi[ETHER_ADDR_LEN], addrlo[ETHER_ADDR_LEN];
7617 	struct ether_multi *enm;
7618 	struct ether_multistep step;
7619 	int s, i, flags, error = 0;
7620 	uint32_t dest;
7621 
7622 	s = splnet();
7623 	switch (cmd) {
7624 
7625 	case SIOCADDMULTI:
7626 		if (ether_addmulti(ifreq_getaddr(cmd, ifr), &sc->sc_ec) == ENETRESET) {
7627 			error = ether_multiaddr(&ifr->ifr_addr, addrlo, addrhi);
7628 			if (error != 0) {
7629 				splx(s);
7630 				return (error);
7631 			}
7632 
7633 			dest = MCX_FLOW_CONTEXT_DEST_TYPE_TABLE |
7634 			    sc->sc_rss_flow_table_id;
7635 
7636 			for (i = 0; i < MCX_NUM_MCAST_FLOWS; i++) {
7637 				if (sc->sc_mcast_flows[i][0] == 0) {
7638 					memcpy(sc->sc_mcast_flows[i], addrlo,
7639 					    ETHER_ADDR_LEN);
7640 					if (ISSET(ifp->if_flags, IFF_RUNNING)) {
7641 						mcx_set_flow_table_entry_mac(sc,
7642 						    MCX_FLOW_GROUP_MAC,
7643 						    sc->sc_mcast_flow_base + i,
7644 						    sc->sc_mcast_flows[i], dest);
7645 					}
7646 					break;
7647 				}
7648 			}
7649 
7650 			if (!ISSET(ifp->if_flags, IFF_ALLMULTI)) {
7651 				if (i == MCX_NUM_MCAST_FLOWS) {
7652 					SET(ifp->if_flags, IFF_ALLMULTI);
7653 					sc->sc_extra_mcast++;
7654 					error = ENETRESET;
7655 				}
7656 
7657 				if (memcmp(addrlo, addrhi, ETHER_ADDR_LEN)) {
7658 					SET(ifp->if_flags, IFF_ALLMULTI);
7659 					error = ENETRESET;
7660 				}
7661 			}
7662 		}
7663 		break;
7664 
7665 	case SIOCDELMULTI:
7666 		if (ether_delmulti(ifreq_getaddr(cmd, ifr), &sc->sc_ec) == ENETRESET) {
7667 			error = ether_multiaddr(&ifr->ifr_addr, addrlo, addrhi);
7668 			if (error != 0) {
7669 				splx(s);
7670 				return (error);
7671 			}
7672 
7673 			for (i = 0; i < MCX_NUM_MCAST_FLOWS; i++) {
7674 				if (memcmp(sc->sc_mcast_flows[i], addrlo,
7675 				    ETHER_ADDR_LEN) == 0) {
7676 					if (ISSET(ifp->if_flags, IFF_RUNNING)) {
7677 						mcx_delete_flow_table_entry(sc,
7678 						    MCX_FLOW_GROUP_MAC,
7679 						    sc->sc_mcast_flow_base + i);
7680 					}
7681 					sc->sc_mcast_flows[i][0] = 0;
7682 					break;
7683 				}
7684 			}
7685 
7686 			if (i == MCX_NUM_MCAST_FLOWS)
7687 				sc->sc_extra_mcast--;
7688 
7689 			if (ISSET(ifp->if_flags, IFF_ALLMULTI) &&
7690 			    sc->sc_extra_mcast == 0) {
7691 				flags = 0;
7692 				ETHER_LOCK(ec);
7693 				ETHER_FIRST_MULTI(step, ec, enm);
7694 				while (enm != NULL) {
7695 					if (memcmp(enm->enm_addrlo,
7696 					    enm->enm_addrhi, ETHER_ADDR_LEN)) {
7697 						SET(flags, IFF_ALLMULTI);
7698 						break;
7699 					}
7700 					ETHER_NEXT_MULTI(step, enm);
7701 				}
7702 				ETHER_UNLOCK(ec);
7703 				if (!ISSET(flags, IFF_ALLMULTI)) {
7704 					CLR(ifp->if_flags, IFF_ALLMULTI);
7705 					error = ENETRESET;
7706 				}
7707 			}
7708 		}
7709 		break;
7710 
7711 	default:
7712 		error = ether_ioctl(ifp, cmd, data);
7713 	}
7714 
7715 	if (error == ENETRESET) {
7716 		if ((ifp->if_flags & (IFF_UP | IFF_RUNNING)) ==
7717 		    (IFF_UP | IFF_RUNNING))
7718 			mcx_iff(sc);
7719 		error = 0;
7720 	}
7721 	splx(s);
7722 
7723 	return (error);
7724 }
7725 
7726 #if 0
7727 static int
7728 mcx_get_sffpage(struct ifnet *ifp, struct if_sffpage *sff)
7729 {
7730 	struct mcx_softc *sc = (struct mcx_softc *)ifp->if_softc;
7731 	struct mcx_reg_mcia mcia;
7732 	struct mcx_reg_pmlp pmlp;
7733 	int offset, error;
7734 
7735 	/* get module number */
7736 	memset(&pmlp, 0, sizeof(pmlp));
7737 	pmlp.rp_local_port = 1;
7738 	error = mcx_access_hca_reg(sc, MCX_REG_PMLP, MCX_REG_OP_READ, &pmlp,
7739 	    sizeof(pmlp));
7740 	if (error != 0) {
7741 		printf("%s: unable to get eeprom module number\n",
7742 		    DEVNAME(sc));
7743 		return error;
7744 	}
7745 
7746 	for (offset = 0; offset < 256; offset += MCX_MCIA_EEPROM_BYTES) {
7747 		memset(&mcia, 0, sizeof(mcia));
7748 		mcia.rm_l = 0;
7749 		mcia.rm_module = be32toh(pmlp.rp_lane0_mapping) &
7750 		    MCX_PMLP_MODULE_NUM_MASK;
7751 		mcia.rm_i2c_addr = sff->sff_addr / 2;	/* apparently */
7752 		mcia.rm_page_num = sff->sff_page;
7753 		mcia.rm_dev_addr = htobe16(offset);
7754 		mcia.rm_size = htobe16(MCX_MCIA_EEPROM_BYTES);
7755 
7756 		error = mcx_access_hca_reg(sc, MCX_REG_MCIA, MCX_REG_OP_READ,
7757 		    &mcia, sizeof(mcia));
7758 		if (error != 0) {
7759 			printf("%s: unable to read eeprom at %x\n",
7760 			    DEVNAME(sc), offset);
7761 			return error;
7762 		}
7763 
7764 		memcpy(sff->sff_data + offset, mcia.rm_data,
7765 		    MCX_MCIA_EEPROM_BYTES);
7766 	}
7767 
7768 	return 0;
7769 }
7770 #endif
7771 
7772 static int
7773 mcx_load_mbuf(struct mcx_softc *sc, struct mcx_slot *ms, struct mbuf *m)
7774 {
7775 	switch (bus_dmamap_load_mbuf(sc->sc_dmat, ms->ms_map, m,
7776 	    BUS_DMA_STREAMING | BUS_DMA_NOWAIT)) {
7777 	case 0:
7778 		break;
7779 
7780 	case EFBIG:
7781 		if (m_defrag(m, M_DONTWAIT) != NULL &&
7782 		    bus_dmamap_load_mbuf(sc->sc_dmat, ms->ms_map, m,
7783 		    BUS_DMA_STREAMING | BUS_DMA_NOWAIT) == 0)
7784 			break;
7785 
7786 		/* FALLTHROUGH */
7787 	default:
7788 		return (1);
7789 	}
7790 
7791 	ms->ms_m = m;
7792 	return (0);
7793 }
7794 
7795 static void
7796 mcx_send_common_locked(struct ifnet *ifp, struct mcx_tx *tx, bool is_transmit)
7797 {
7798 	struct mcx_softc *sc = ifp->if_softc;
7799 	struct mcx_sq_entry *sq, *sqe;
7800 	struct mcx_sq_entry_seg *sqs;
7801 	struct mcx_slot *ms;
7802 	bus_dmamap_t map;
7803 	struct mbuf *m;
7804 	u_int idx, free, used;
7805 	uint64_t *bf;
7806 	uint32_t csum;
7807 	size_t bf_base;
7808 	int i, seg, nseg;
7809 
7810 	KASSERT(mutex_owned(&tx->tx_lock));
7811 
7812 	if ((ifp->if_flags & IFF_RUNNING) == 0)
7813 		return;
7814 
7815 	bf_base = (tx->tx_uar * MCX_PAGE_SIZE) + MCX_UAR_BF;
7816 
7817 	idx = tx->tx_prod % (1 << MCX_LOG_SQ_SIZE);
7818 	free = (tx->tx_cons + (1 << MCX_LOG_SQ_SIZE)) - tx->tx_prod;
7819 
7820 	used = 0;
7821 	bf = NULL;
7822 
7823 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&tx->tx_sq_mem),
7824 	    0, MCX_DMA_LEN(&tx->tx_sq_mem), BUS_DMASYNC_POSTWRITE);
7825 
7826 	sq = (struct mcx_sq_entry *)MCX_DMA_KVA(&tx->tx_sq_mem);
7827 
7828 	for (;;) {
7829 		if (used + MCX_SQ_ENTRY_MAX_SLOTS >= free) {
7830 			SET(ifp->if_flags, IFF_OACTIVE);
7831 			break;
7832 		}
7833 
7834 		if (is_transmit) {
7835 			m = pcq_get(tx->tx_pcq);
7836 		} else {
7837 			IFQ_DEQUEUE(&ifp->if_snd, m);
7838 		}
7839 		if (m == NULL) {
7840 			break;
7841 		}
7842 
7843 		sqe = sq + idx;
7844 		ms = &tx->tx_slots[idx];
7845 		memset(sqe, 0, sizeof(*sqe));
7846 
7847 		/* ctrl segment */
7848 		sqe->sqe_opcode_index = htobe32(MCX_SQE_WQE_OPCODE_SEND |
7849 		    ((tx->tx_prod & 0xffff) << MCX_SQE_WQE_INDEX_SHIFT));
7850 		/* always generate a completion event */
7851 		sqe->sqe_signature = htobe32(MCX_SQE_CE_CQE_ALWAYS);
7852 
7853 		/* eth segment */
7854 		csum = 0;
7855 		if (m->m_pkthdr.csum_flags & M_CSUM_IPv4)
7856 			csum |= MCX_SQE_L3_CSUM;
7857 		if (m->m_pkthdr.csum_flags &
7858 		    (M_CSUM_TCPv4 | M_CSUM_UDPv4 | M_CSUM_TCPv6 | M_CSUM_UDPv6))
7859 			csum |= MCX_SQE_L4_CSUM;
7860 		sqe->sqe_mss_csum = htobe32(csum);
7861 		sqe->sqe_inline_header_size = htobe16(MCX_SQ_INLINE_SIZE);
7862 		if (vlan_has_tag(m)) {
7863 			struct ether_vlan_header *evh;
7864 			evh = (struct ether_vlan_header *)
7865 			    &sqe->sqe_inline_headers;
7866 
7867 			m_copydata(m, 0, ETHER_HDR_LEN, evh);
7868 			evh->evl_proto = evh->evl_encap_proto;
7869 			evh->evl_encap_proto = htons(ETHERTYPE_VLAN);
7870 			evh->evl_tag = htons(vlan_get_tag(m));
7871 			m_adj(m, ETHER_HDR_LEN);
7872 		} else {
7873 			m_copydata(m, 0, MCX_SQ_INLINE_SIZE,
7874 			    sqe->sqe_inline_headers);
7875 			m_adj(m, MCX_SQ_INLINE_SIZE);
7876 		}
7877 
7878 		if (mcx_load_mbuf(sc, ms, m) != 0) {
7879 			m_freem(m);
7880 			if_statinc(ifp, if_oerrors);
7881 			continue;
7882 		}
7883 		bf = (uint64_t *)sqe;
7884 
7885 		if (ifp->if_bpf != NULL)
7886 			bpf_mtap2(ifp->if_bpf, sqe->sqe_inline_headers,
7887 			    MCX_SQ_INLINE_SIZE, m, BPF_D_OUT);
7888 
7889 		map = ms->ms_map;
7890 		bus_dmamap_sync(sc->sc_dmat, map, 0, map->dm_mapsize,
7891 		    BUS_DMASYNC_PREWRITE);
7892 
7893 		sqe->sqe_ds_sq_num =
7894 		    htobe32((tx->tx_sqn << MCX_SQE_SQ_NUM_SHIFT) |
7895 		    (map->dm_nsegs + 3));
7896 
7897 		/* data segment - first wqe has one segment */
7898 		sqs = sqe->sqe_segs;
7899 		seg = 0;
7900 		nseg = 1;
7901 		for (i = 0; i < map->dm_nsegs; i++) {
7902 			if (seg == nseg) {
7903 				/* next slot */
7904 				idx++;
7905 				if (idx == (1 << MCX_LOG_SQ_SIZE))
7906 					idx = 0;
7907 				tx->tx_prod++;
7908 				used++;
7909 
7910 				sqs = (struct mcx_sq_entry_seg *)(sq + idx);
7911 				seg = 0;
7912 				nseg = MCX_SQ_SEGS_PER_SLOT;
7913 			}
7914 			sqs[seg].sqs_byte_count =
7915 			    htobe32(map->dm_segs[i].ds_len);
7916 			sqs[seg].sqs_lkey = htobe32(sc->sc_lkey);
7917 			sqs[seg].sqs_addr = htobe64(map->dm_segs[i].ds_addr);
7918 			seg++;
7919 		}
7920 
7921 		idx++;
7922 		if (idx == (1 << MCX_LOG_SQ_SIZE))
7923 			idx = 0;
7924 		tx->tx_prod++;
7925 		used++;
7926 	}
7927 
7928 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&tx->tx_sq_mem),
7929 	    0, MCX_DMA_LEN(&tx->tx_sq_mem), BUS_DMASYNC_PREWRITE);
7930 
7931 	if (used) {
7932 		bus_size_t blueflame;
7933 
7934 		bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_doorbell_mem),
7935 		    tx->tx_doorbell, sizeof(uint32_t), BUS_DMASYNC_POSTWRITE);
7936 		be32enc(MCX_DMA_OFF(&sc->sc_doorbell_mem, tx->tx_doorbell),
7937 		    tx->tx_prod & MCX_WQ_DOORBELL_MASK);
7938 		bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_doorbell_mem),
7939 		    tx->tx_doorbell, sizeof(uint32_t), BUS_DMASYNC_PREWRITE);
7940 
7941 		/*
7942 		 * write the first 64 bits of the last sqe we produced
7943 		 * to the blue flame buffer
7944 		 */
7945 
7946 		blueflame = bf_base + tx->tx_bf_offset;
7947 		bus_space_write_8(sc->sc_memt, sc->sc_memh,
7948 		    blueflame, *bf);
7949 		mcx_bar(sc, blueflame, sizeof(*bf), BUS_SPACE_BARRIER_WRITE);
7950 
7951 		/* next write goes to the other buffer */
7952 		tx->tx_bf_offset ^= sc->sc_bf_size;
7953 	}
7954 }
7955 
7956 static void
7957 mcx_start(struct ifnet *ifp)
7958 {
7959 	struct mcx_softc *sc = ifp->if_softc;
7960 	/* mcx_start() always uses TX ring[0] */
7961 	struct mcx_tx *tx = &sc->sc_queues[0].q_tx;
7962 
7963 	mutex_enter(&tx->tx_lock);
7964 	if (!ISSET(ifp->if_flags, IFF_OACTIVE)) {
7965 		mcx_send_common_locked(ifp, tx, false);
7966 	}
7967 	mutex_exit(&tx->tx_lock);
7968 }
7969 
7970 static int
7971 mcx_transmit(struct ifnet *ifp, struct mbuf *m)
7972 {
7973 	struct mcx_softc *sc = ifp->if_softc;
7974 	struct mcx_tx *tx;
7975 
7976 	tx = &sc->sc_queues[cpu_index(curcpu()) % sc->sc_nqueues].q_tx;
7977 	if (__predict_false(!pcq_put(tx->tx_pcq, m))) {
7978 		m_freem(m);
7979 		return ENOBUFS;
7980 	}
7981 
7982 	if (mutex_tryenter(&tx->tx_lock)) {
7983 		mcx_send_common_locked(ifp, tx, true);
7984 		mutex_exit(&tx->tx_lock);
7985 	} else {
7986 		softint_schedule(tx->tx_softint);
7987 	}
7988 
7989 	return 0;
7990 }
7991 
7992 static void
7993 mcx_deferred_transmit(void *arg)
7994 {
7995 	struct mcx_tx *tx = arg;
7996 	struct mcx_softc *sc = tx->tx_softc;
7997 	struct ifnet *ifp = &sc->sc_ec.ec_if;
7998 
7999 	mutex_enter(&tx->tx_lock);
8000 	if (pcq_peek(tx->tx_pcq) != NULL) {
8001 		mcx_send_common_locked(ifp, tx, true);
8002 	}
8003 	mutex_exit(&tx->tx_lock);
8004 }
8005 
8006 
8007 static void
8008 mcx_media_add_types(struct mcx_softc *sc)
8009 {
8010 	struct mcx_reg_ptys ptys;
8011 	int i;
8012 	uint32_t proto_cap;
8013 
8014 	memset(&ptys, 0, sizeof(ptys));
8015 	ptys.rp_local_port = 1;
8016 	ptys.rp_proto_mask = MCX_REG_PTYS_PROTO_MASK_ETH;
8017 	if (mcx_access_hca_reg(sc, MCX_REG_PTYS, MCX_REG_OP_READ, &ptys,
8018 	    sizeof(ptys)) != 0) {
8019 		printf("%s: unable to read port type/speed\n", DEVNAME(sc));
8020 		return;
8021 	}
8022 
8023 	proto_cap = be32toh(ptys.rp_eth_proto_cap);
8024 	for (i = 0; i < __arraycount(mcx_eth_cap_map); i++) {
8025 		const struct mcx_eth_proto_capability *cap;
8026 		if (!ISSET(proto_cap, 1U << i))
8027 			continue;
8028 
8029 		cap = &mcx_eth_cap_map[i];
8030 		if (cap->cap_media == 0)
8031 			continue;
8032 
8033 		ifmedia_add(&sc->sc_media, IFM_ETHER | cap->cap_media, 0, NULL);
8034 	}
8035 }
8036 
8037 static void
8038 mcx_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
8039 {
8040 	struct mcx_softc *sc = (struct mcx_softc *)ifp->if_softc;
8041 	struct mcx_reg_ptys ptys;
8042 	int i;
8043 	uint32_t proto_oper;
8044 	uint64_t media_oper;
8045 
8046 	memset(&ptys, 0, sizeof(ptys));
8047 	ptys.rp_local_port = 1;
8048 	ptys.rp_proto_mask = MCX_REG_PTYS_PROTO_MASK_ETH;
8049 
8050 	if (mcx_access_hca_reg(sc, MCX_REG_PTYS, MCX_REG_OP_READ, &ptys,
8051 	    sizeof(ptys)) != 0) {
8052 		printf("%s: unable to read port type/speed\n", DEVNAME(sc));
8053 		return;
8054 	}
8055 
8056 	proto_oper = be32toh(ptys.rp_eth_proto_oper);
8057 
8058 	media_oper = 0;
8059 
8060 	for (i = 0; i < __arraycount(mcx_eth_cap_map); i++) {
8061 		const struct mcx_eth_proto_capability *cap;
8062 		if (!ISSET(proto_oper, 1U << i))
8063 			continue;
8064 
8065 		cap = &mcx_eth_cap_map[i];
8066 
8067 		if (cap->cap_media != 0)
8068 			media_oper = cap->cap_media;
8069 	}
8070 
8071 	ifmr->ifm_status = IFM_AVALID;
8072 	if (proto_oper != 0) {
8073 		ifmr->ifm_status |= IFM_ACTIVE;
8074 		ifmr->ifm_active = IFM_ETHER | IFM_AUTO | media_oper;
8075 		/* txpause, rxpause, duplex? */
8076 	}
8077 }
8078 
8079 static int
8080 mcx_media_change(struct ifnet *ifp)
8081 {
8082 	struct mcx_softc *sc = (struct mcx_softc *)ifp->if_softc;
8083 	struct mcx_reg_ptys ptys;
8084 	struct mcx_reg_paos paos;
8085 	uint32_t media;
8086 	int i, error;
8087 
8088 	if (IFM_TYPE(sc->sc_media.ifm_media) != IFM_ETHER)
8089 		return EINVAL;
8090 
8091 	error = 0;
8092 
8093 	if (IFM_SUBTYPE(sc->sc_media.ifm_media) == IFM_AUTO) {
8094 		/* read ptys to get supported media */
8095 		memset(&ptys, 0, sizeof(ptys));
8096 		ptys.rp_local_port = 1;
8097 		ptys.rp_proto_mask = MCX_REG_PTYS_PROTO_MASK_ETH;
8098 		if (mcx_access_hca_reg(sc, MCX_REG_PTYS, MCX_REG_OP_READ,
8099 		    &ptys, sizeof(ptys)) != 0) {
8100 			printf("%s: unable to read port type/speed\n",
8101 			    DEVNAME(sc));
8102 			return EIO;
8103 		}
8104 
8105 		media = be32toh(ptys.rp_eth_proto_cap);
8106 	} else {
8107 		/* map media type */
8108 		media = 0;
8109 		for (i = 0; i < __arraycount(mcx_eth_cap_map); i++) {
8110 			const struct mcx_eth_proto_capability *cap;
8111 
8112 			cap = &mcx_eth_cap_map[i];
8113 			if (cap->cap_media ==
8114 			    IFM_SUBTYPE(sc->sc_media.ifm_media)) {
8115 				media = (1 << i);
8116 				break;
8117 			}
8118 		}
8119 	}
8120 
8121 	/* disable the port */
8122 	memset(&paos, 0, sizeof(paos));
8123 	paos.rp_local_port = 1;
8124 	paos.rp_admin_status = MCX_REG_PAOS_ADMIN_STATUS_DOWN;
8125 	paos.rp_admin_state_update = MCX_REG_PAOS_ADMIN_STATE_UPDATE_EN;
8126 	if (mcx_access_hca_reg(sc, MCX_REG_PAOS, MCX_REG_OP_WRITE, &paos,
8127 	    sizeof(paos)) != 0) {
8128 		printf("%s: unable to set port state to down\n", DEVNAME(sc));
8129 		return EIO;
8130 	}
8131 
8132 	memset(&ptys, 0, sizeof(ptys));
8133 	ptys.rp_local_port = 1;
8134 	ptys.rp_proto_mask = MCX_REG_PTYS_PROTO_MASK_ETH;
8135 	ptys.rp_eth_proto_admin = htobe32(media);
8136 	if (mcx_access_hca_reg(sc, MCX_REG_PTYS, MCX_REG_OP_WRITE, &ptys,
8137 	    sizeof(ptys)) != 0) {
8138 		printf("%s: unable to set port media type/speed\n",
8139 		    DEVNAME(sc));
8140 		error = EIO;
8141 	}
8142 
8143 	/* re-enable the port to start negotiation */
8144 	memset(&paos, 0, sizeof(paos));
8145 	paos.rp_local_port = 1;
8146 	paos.rp_admin_status = MCX_REG_PAOS_ADMIN_STATUS_UP;
8147 	paos.rp_admin_state_update = MCX_REG_PAOS_ADMIN_STATE_UPDATE_EN;
8148 	if (mcx_access_hca_reg(sc, MCX_REG_PAOS, MCX_REG_OP_WRITE, &paos,
8149 	    sizeof(paos)) != 0) {
8150 		printf("%s: unable to set port state to up\n", DEVNAME(sc));
8151 		error = EIO;
8152 	}
8153 
8154 	return error;
8155 }
8156 
8157 static void
8158 mcx_port_change(struct work *wk, void *xsc)
8159 {
8160 	struct mcx_softc *sc = xsc;
8161 	struct ifnet *ifp = &sc->sc_ec.ec_if;
8162 	struct mcx_reg_ptys ptys = {
8163 		.rp_local_port = 1,
8164 		.rp_proto_mask = MCX_REG_PTYS_PROTO_MASK_ETH,
8165 	};
8166 	int link_state = LINK_STATE_DOWN;
8167 
8168 	if (mcx_access_hca_reg(sc, MCX_REG_PTYS, MCX_REG_OP_READ, &ptys,
8169 	    sizeof(ptys)) == 0) {
8170 		uint32_t proto_oper = be32toh(ptys.rp_eth_proto_oper);
8171 		uint64_t baudrate = 0;
8172 		unsigned int i;
8173 
8174 		if (proto_oper != 0)
8175 			link_state = LINK_STATE_UP;
8176 
8177 		for (i = 0; i < __arraycount(mcx_eth_cap_map); i++) {
8178 			const struct mcx_eth_proto_capability *cap;
8179 			if (!ISSET(proto_oper, 1U << i))
8180 				continue;
8181 
8182 			cap = &mcx_eth_cap_map[i];
8183 			if (cap->cap_baudrate == 0)
8184 				continue;
8185 
8186 			baudrate = cap->cap_baudrate;
8187 			break;
8188 		}
8189 
8190 		ifp->if_baudrate = baudrate;
8191 	}
8192 
8193 	if (link_state != ifp->if_link_state) {
8194 		if_link_state_change(ifp, link_state);
8195 	}
8196 }
8197 
8198 
8199 static inline uint32_t
8200 mcx_rd(struct mcx_softc *sc, bus_size_t r)
8201 {
8202 	uint32_t word;
8203 
8204 	word = bus_space_read_4(sc->sc_memt, sc->sc_memh, r);
8205 
8206 	return (be32toh(word));
8207 }
8208 
8209 static inline void
8210 mcx_wr(struct mcx_softc *sc, bus_size_t r, uint32_t v)
8211 {
8212 	bus_space_write_4(sc->sc_memt, sc->sc_memh, r, htobe32(v));
8213 }
8214 
8215 static inline void
8216 mcx_bar(struct mcx_softc *sc, bus_size_t r, bus_size_t l, int f)
8217 {
8218 #ifndef __NetBSD__
8219 	bus_space_barrier(sc->sc_memt, sc->sc_memh, r, l, f);
8220 #endif
8221 }
8222 
8223 static uint64_t
8224 mcx_timer(struct mcx_softc *sc)
8225 {
8226 	uint32_t hi, lo, ni;
8227 
8228 	hi = mcx_rd(sc, MCX_INTERNAL_TIMER_H);
8229 	for (;;) {
8230 		lo = mcx_rd(sc, MCX_INTERNAL_TIMER_L);
8231 		mcx_bar(sc, MCX_INTERNAL_TIMER_L, 8, BUS_SPACE_BARRIER_READ);
8232 		ni = mcx_rd(sc, MCX_INTERNAL_TIMER_H);
8233 
8234 		if (ni == hi)
8235 			break;
8236 
8237 		hi = ni;
8238 	}
8239 
8240 	return (((uint64_t)hi << 32) | (uint64_t)lo);
8241 }
8242 
8243 static int
8244 mcx_dmamem_alloc(struct mcx_softc *sc, struct mcx_dmamem *mxm,
8245     bus_size_t size, u_int align)
8246 {
8247 	mxm->mxm_size = size;
8248 
8249 	if (bus_dmamap_create(sc->sc_dmat, mxm->mxm_size, 1,
8250 	    mxm->mxm_size, 0,
8251 	    BUS_DMA_WAITOK | BUS_DMA_ALLOCNOW,
8252 	    &mxm->mxm_map) != 0)
8253 		return (1);
8254 	if (bus_dmamem_alloc(sc->sc_dmat, mxm->mxm_size,
8255 	    align, 0, &mxm->mxm_seg, 1, &mxm->mxm_nsegs,
8256 	    BUS_DMA_WAITOK) != 0)
8257 		goto destroy;
8258 	if (bus_dmamem_map(sc->sc_dmat, &mxm->mxm_seg, mxm->mxm_nsegs,
8259 	    mxm->mxm_size, &mxm->mxm_kva,
8260 	    BUS_DMA_WAITOK | BUS_DMA_COHERENT) != 0)
8261 		goto free;
8262 	if (bus_dmamap_load(sc->sc_dmat, mxm->mxm_map, mxm->mxm_kva,
8263 	    mxm->mxm_size, NULL, BUS_DMA_WAITOK) != 0)
8264 		goto unmap;
8265 
8266 	mcx_dmamem_zero(mxm);
8267 
8268 	return (0);
8269 unmap:
8270 	bus_dmamem_unmap(sc->sc_dmat, mxm->mxm_kva, mxm->mxm_size);
8271 free:
8272 	bus_dmamem_free(sc->sc_dmat, &mxm->mxm_seg, 1);
8273 destroy:
8274 	bus_dmamap_destroy(sc->sc_dmat, mxm->mxm_map);
8275 	return (1);
8276 }
8277 
8278 static void
8279 mcx_dmamem_zero(struct mcx_dmamem *mxm)
8280 {
8281 	memset(MCX_DMA_KVA(mxm), 0, MCX_DMA_LEN(mxm));
8282 }
8283 
8284 static void
8285 mcx_dmamem_free(struct mcx_softc *sc, struct mcx_dmamem *mxm)
8286 {
8287 	bus_dmamap_unload(sc->sc_dmat, mxm->mxm_map);
8288 	bus_dmamem_unmap(sc->sc_dmat, mxm->mxm_kva, mxm->mxm_size);
8289 	bus_dmamem_free(sc->sc_dmat, &mxm->mxm_seg, 1);
8290 	bus_dmamap_destroy(sc->sc_dmat, mxm->mxm_map);
8291 }
8292 
8293 static int
8294 mcx_hwmem_alloc(struct mcx_softc *sc, struct mcx_hwmem *mhm, unsigned int pages)
8295 {
8296 	bus_dma_segment_t *segs;
8297 	bus_size_t len = pages * MCX_PAGE_SIZE;
8298 	size_t seglen;
8299 
8300 	segs = kmem_alloc(sizeof(*segs) * pages, KM_SLEEP);
8301 	seglen = sizeof(*segs) * pages;
8302 
8303 	if (bus_dmamem_alloc(sc->sc_dmat, len, MCX_PAGE_SIZE, 0,
8304 	    segs, pages, &mhm->mhm_seg_count, BUS_DMA_NOWAIT) != 0)
8305 		goto free_segs;
8306 
8307 	if (mhm->mhm_seg_count < pages) {
8308 		size_t nseglen;
8309 
8310 		mhm->mhm_segs = kmem_alloc(
8311 		    sizeof(*mhm->mhm_segs) * mhm->mhm_seg_count, KM_SLEEP);
8312 
8313 		nseglen = sizeof(*mhm->mhm_segs) * mhm->mhm_seg_count;
8314 
8315 		memcpy(mhm->mhm_segs, segs, nseglen);
8316 
8317 		kmem_free(segs, seglen);
8318 
8319 		segs = mhm->mhm_segs;
8320 		seglen = nseglen;
8321 	} else
8322 		mhm->mhm_segs = segs;
8323 
8324 	if (bus_dmamap_create(sc->sc_dmat, len, pages, MCX_PAGE_SIZE,
8325 	    MCX_PAGE_SIZE, BUS_DMA_NOWAIT|BUS_DMA_ALLOCNOW /*|BUS_DMA_64BIT*/,
8326 	    &mhm->mhm_map) != 0)
8327 		goto free_dmamem;
8328 
8329 	if (bus_dmamap_load_raw(sc->sc_dmat, mhm->mhm_map,
8330 	    mhm->mhm_segs, mhm->mhm_seg_count, len, BUS_DMA_NOWAIT) != 0)
8331 		goto destroy;
8332 
8333 	bus_dmamap_sync(sc->sc_dmat, mhm->mhm_map,
8334 	    0, mhm->mhm_map->dm_mapsize, BUS_DMASYNC_PRERW);
8335 
8336 	mhm->mhm_npages = pages;
8337 
8338 	return (0);
8339 
8340 destroy:
8341 	bus_dmamap_destroy(sc->sc_dmat, mhm->mhm_map);
8342 free_dmamem:
8343 	bus_dmamem_free(sc->sc_dmat, mhm->mhm_segs, mhm->mhm_seg_count);
8344 free_segs:
8345 	kmem_free(segs, seglen);
8346 	mhm->mhm_segs = NULL;
8347 
8348 	return (-1);
8349 }
8350 
8351 static void
8352 mcx_hwmem_free(struct mcx_softc *sc, struct mcx_hwmem *mhm)
8353 {
8354 	if (mhm->mhm_npages == 0)
8355 		return;
8356 
8357 	bus_dmamap_sync(sc->sc_dmat, mhm->mhm_map,
8358 	    0, mhm->mhm_map->dm_mapsize, BUS_DMASYNC_POSTRW);
8359 
8360 	bus_dmamap_unload(sc->sc_dmat, mhm->mhm_map);
8361 	bus_dmamap_destroy(sc->sc_dmat, mhm->mhm_map);
8362 	bus_dmamem_free(sc->sc_dmat, mhm->mhm_segs, mhm->mhm_seg_count);
8363 	kmem_free(mhm->mhm_segs, sizeof(*mhm->mhm_segs) * mhm->mhm_seg_count);
8364 
8365 	mhm->mhm_npages = 0;
8366 }
8367 
8368 #if NKSTAT > 0
8369 struct mcx_ppcnt {
8370 	char			 name[KSTAT_KV_NAMELEN];
8371 	enum kstat_kv_unit	 unit;
8372 };
8373 
8374 static const struct mcx_ppcnt mcx_ppcnt_ieee8023_tpl[] = {
8375 	{ "Good Tx",		KSTAT_KV_U_PACKETS, },
8376 	{ "Good Rx",		KSTAT_KV_U_PACKETS, },
8377 	{ "FCS errs",		KSTAT_KV_U_PACKETS, },
8378 	{ "Alignment Errs",	KSTAT_KV_U_PACKETS, },
8379 	{ "Good Tx",		KSTAT_KV_U_BYTES, },
8380 	{ "Good Rx",		KSTAT_KV_U_BYTES, },
8381 	{ "Multicast Tx",	KSTAT_KV_U_PACKETS, },
8382 	{ "Broadcast Tx",	KSTAT_KV_U_PACKETS, },
8383 	{ "Multicast Rx",	KSTAT_KV_U_PACKETS, },
8384 	{ "Broadcast Rx",	KSTAT_KV_U_PACKETS, },
8385 	{ "In Range Len",	KSTAT_KV_U_PACKETS, },
8386 	{ "Out Of Range Len",	KSTAT_KV_U_PACKETS, },
8387 	{ "Frame Too Long",	KSTAT_KV_U_PACKETS, },
8388 	{ "Symbol Errs",	KSTAT_KV_U_PACKETS, },
8389 	{ "MAC Ctrl Tx",	KSTAT_KV_U_PACKETS, },
8390 	{ "MAC Ctrl Rx",	KSTAT_KV_U_PACKETS, },
8391 	{ "MAC Ctrl Unsup",	KSTAT_KV_U_PACKETS, },
8392 	{ "Pause Rx",		KSTAT_KV_U_PACKETS, },
8393 	{ "Pause Tx",		KSTAT_KV_U_PACKETS, },
8394 };
8395 CTASSERT(__arraycount(mcx_ppcnt_ieee8023_tpl) == mcx_ppcnt_ieee8023_count);
8396 
8397 static const struct mcx_ppcnt mcx_ppcnt_rfc2863_tpl[] = {
8398 	{ "Rx Bytes",		KSTAT_KV_U_BYTES, },
8399 	{ "Rx Unicast",		KSTAT_KV_U_PACKETS, },
8400 	{ "Rx Discards",	KSTAT_KV_U_PACKETS, },
8401 	{ "Rx Errors",		KSTAT_KV_U_PACKETS, },
8402 	{ "Rx Unknown Proto",	KSTAT_KV_U_PACKETS, },
8403 	{ "Tx Bytes",		KSTAT_KV_U_BYTES, },
8404 	{ "Tx Unicast",		KSTAT_KV_U_PACKETS, },
8405 	{ "Tx Discards",	KSTAT_KV_U_PACKETS, },
8406 	{ "Tx Errors",		KSTAT_KV_U_PACKETS, },
8407 	{ "Rx Multicast",	KSTAT_KV_U_PACKETS, },
8408 	{ "Rx Broadcast",	KSTAT_KV_U_PACKETS, },
8409 	{ "Tx Multicast",	KSTAT_KV_U_PACKETS, },
8410 	{ "Tx Broadcast",	KSTAT_KV_U_PACKETS, },
8411 };
8412 CTASSERT(__arraycount(mcx_ppcnt_rfc2863_tpl) == mcx_ppcnt_rfc2863_count);
8413 
8414 static const struct mcx_ppcnt mcx_ppcnt_rfc2819_tpl[] = {
8415 	{ "Drop Events",	KSTAT_KV_U_PACKETS, },
8416 	{ "Octets",		KSTAT_KV_U_BYTES, },
8417 	{ "Packets",		KSTAT_KV_U_PACKETS, },
8418 	{ "Broadcasts",		KSTAT_KV_U_PACKETS, },
8419 	{ "Multicasts",		KSTAT_KV_U_PACKETS, },
8420 	{ "CRC Align Errs",	KSTAT_KV_U_PACKETS, },
8421 	{ "Undersize",		KSTAT_KV_U_PACKETS, },
8422 	{ "Oversize",		KSTAT_KV_U_PACKETS, },
8423 	{ "Fragments",		KSTAT_KV_U_PACKETS, },
8424 	{ "Jabbers",		KSTAT_KV_U_PACKETS, },
8425 	{ "Collisions",		KSTAT_KV_U_NONE, },
8426 	{ "64B",		KSTAT_KV_U_PACKETS, },
8427 	{ "65-127B",		KSTAT_KV_U_PACKETS, },
8428 	{ "128-255B",		KSTAT_KV_U_PACKETS, },
8429 	{ "256-511B",		KSTAT_KV_U_PACKETS, },
8430 	{ "512-1023B",		KSTAT_KV_U_PACKETS, },
8431 	{ "1024-1518B",		KSTAT_KV_U_PACKETS, },
8432 	{ "1519-2047B",		KSTAT_KV_U_PACKETS, },
8433 	{ "2048-4095B",		KSTAT_KV_U_PACKETS, },
8434 	{ "4096-8191B",		KSTAT_KV_U_PACKETS, },
8435 	{ "8192-10239B",	KSTAT_KV_U_PACKETS, },
8436 };
8437 CTASSERT(__arraycount(mcx_ppcnt_rfc2819_tpl) == mcx_ppcnt_rfc2819_count);
8438 
8439 static const struct mcx_ppcnt mcx_ppcnt_rfc3635_tpl[] = {
8440 	{ "Alignment Errs",	KSTAT_KV_U_PACKETS, },
8441 	{ "FCS Errs",		KSTAT_KV_U_PACKETS, },
8442 	{ "Single Colls",	KSTAT_KV_U_PACKETS, },
8443 	{ "Multiple Colls",	KSTAT_KV_U_PACKETS, },
8444 	{ "SQE Test Errs",	KSTAT_KV_U_NONE, },
8445 	{ "Deferred Tx",	KSTAT_KV_U_PACKETS, },
8446 	{ "Late Colls",		KSTAT_KV_U_NONE, },
8447 	{ "Exess Colls",	KSTAT_KV_U_NONE, },
8448 	{ "Int MAC Tx Errs",	KSTAT_KV_U_PACKETS, },
8449 	{ "CSM Sense Errs",	KSTAT_KV_U_NONE, },
8450 	{ "Too Long",		KSTAT_KV_U_PACKETS, },
8451 	{ "Int MAC Rx Errs",	KSTAT_KV_U_PACKETS, },
8452 	{ "Symbol Errs",	KSTAT_KV_U_NONE, },
8453 	{ "Unknown Control",	KSTAT_KV_U_PACKETS, },
8454 	{ "Pause Rx",		KSTAT_KV_U_PACKETS, },
8455 	{ "Pause Tx",		KSTAT_KV_U_PACKETS, },
8456 };
8457 CTASSERT(__arraycount(mcx_ppcnt_rfc3635_tpl) == mcx_ppcnt_rfc3635_count);
8458 
8459 struct mcx_kstat_ppcnt {
8460 	const char		*ksp_name;
8461 	const struct mcx_ppcnt	*ksp_tpl;
8462 	unsigned int		 ksp_n;
8463 	uint8_t			 ksp_grp;
8464 };
8465 
8466 static const struct mcx_kstat_ppcnt mcx_kstat_ppcnt_ieee8023 = {
8467 	.ksp_name =		"ieee802.3",
8468 	.ksp_tpl =		mcx_ppcnt_ieee8023_tpl,
8469 	.ksp_n =		__arraycount(mcx_ppcnt_ieee8023_tpl),
8470 	.ksp_grp =		MCX_REG_PPCNT_GRP_IEEE8023,
8471 };
8472 
8473 static const struct mcx_kstat_ppcnt mcx_kstat_ppcnt_rfc2863 = {
8474 	.ksp_name =		"rfc2863",
8475 	.ksp_tpl =		mcx_ppcnt_rfc2863_tpl,
8476 	.ksp_n =		__arraycount(mcx_ppcnt_rfc2863_tpl),
8477 	.ksp_grp =		MCX_REG_PPCNT_GRP_RFC2863,
8478 };
8479 
8480 static const struct mcx_kstat_ppcnt mcx_kstat_ppcnt_rfc2819 = {
8481 	.ksp_name =		"rfc2819",
8482 	.ksp_tpl =		mcx_ppcnt_rfc2819_tpl,
8483 	.ksp_n =		__arraycount(mcx_ppcnt_rfc2819_tpl),
8484 	.ksp_grp =		MCX_REG_PPCNT_GRP_RFC2819,
8485 };
8486 
8487 static const struct mcx_kstat_ppcnt mcx_kstat_ppcnt_rfc3635 = {
8488 	.ksp_name =		"rfc3635",
8489 	.ksp_tpl =		mcx_ppcnt_rfc3635_tpl,
8490 	.ksp_n =		__arraycount(mcx_ppcnt_rfc3635_tpl),
8491 	.ksp_grp =		MCX_REG_PPCNT_GRP_RFC3635,
8492 };
8493 
8494 static int	mcx_kstat_ppcnt_read(struct kstat *);
8495 
8496 static void	mcx_kstat_attach_tmps(struct mcx_softc *sc);
8497 static void	mcx_kstat_attach_queues(struct mcx_softc *sc);
8498 
8499 static struct kstat *
8500 mcx_kstat_attach_ppcnt(struct mcx_softc *sc,
8501     const struct mcx_kstat_ppcnt *ksp)
8502 {
8503 	struct kstat *ks;
8504 	struct kstat_kv *kvs;
8505 	unsigned int i;
8506 
8507 	ks = kstat_create(DEVNAME(sc), 0, ksp->ksp_name, 0, KSTAT_T_KV, 0);
8508 	if (ks == NULL)
8509 		return (NULL);
8510 
8511 	kvs = mallocarray(ksp->ksp_n, sizeof(*kvs),
8512 	    M_DEVBUF, M_WAITOK);
8513 
8514 	for (i = 0; i < ksp->ksp_n; i++) {
8515 		const struct mcx_ppcnt *tpl = &ksp->ksp_tpl[i];
8516 
8517 		kstat_kv_unit_init(&kvs[i], tpl->name,
8518 		    KSTAT_KV_T_COUNTER64, tpl->unit);
8519 	}
8520 
8521 	ks->ks_softc = sc;
8522 	ks->ks_ptr = (void *)ksp;
8523 	ks->ks_data = kvs;
8524 	ks->ks_datalen = ksp->ksp_n * sizeof(*kvs);
8525 	ks->ks_read = mcx_kstat_ppcnt_read;
8526 
8527 	kstat_install(ks);
8528 
8529 	return (ks);
8530 }
8531 
8532 static void
8533 mcx_kstat_attach(struct mcx_softc *sc)
8534 {
8535 	sc->sc_kstat_ieee8023 = mcx_kstat_attach_ppcnt(sc,
8536 	    &mcx_kstat_ppcnt_ieee8023);
8537 	sc->sc_kstat_rfc2863 = mcx_kstat_attach_ppcnt(sc,
8538 	    &mcx_kstat_ppcnt_rfc2863);
8539 	sc->sc_kstat_rfc2819 = mcx_kstat_attach_ppcnt(sc,
8540 	    &mcx_kstat_ppcnt_rfc2819);
8541 	sc->sc_kstat_rfc3635 = mcx_kstat_attach_ppcnt(sc,
8542 	    &mcx_kstat_ppcnt_rfc3635);
8543 
8544 	mcx_kstat_attach_tmps(sc);
8545 	mcx_kstat_attach_queues(sc);
8546 }
8547 
8548 static int
8549 mcx_kstat_ppcnt_read(struct kstat *ks)
8550 {
8551 	struct mcx_softc *sc = ks->ks_softc;
8552 	struct mcx_kstat_ppcnt *ksp = ks->ks_ptr;
8553 	struct mcx_reg_ppcnt ppcnt = {
8554 		.ppcnt_grp = ksp->ksp_grp,
8555 		.ppcnt_local_port = 1,
8556 	};
8557 	struct kstat_kv *kvs = ks->ks_data;
8558 	uint64_t *vs = (uint64_t *)&ppcnt.ppcnt_counter_set;
8559 	unsigned int i;
8560 	int rv;
8561 
8562 	KERNEL_LOCK(); /* XXX */
8563 	rv = mcx_access_hca_reg(sc, MCX_REG_PPCNT, MCX_REG_OP_READ,
8564 	    &ppcnt, sizeof(ppcnt));
8565 	KERNEL_UNLOCK();
8566 	if (rv != 0)
8567 		return (EIO);
8568 
8569 	nanouptime(&ks->ks_updated);
8570 
8571 	for (i = 0; i < ksp->ksp_n; i++)
8572 		kstat_kv_u64(&kvs[i]) = bemtoh64(&vs[i]);
8573 
8574 	return (0);
8575 }
8576 
8577 struct mcx_kstat_mtmp {
8578 	struct kstat_kv		ktmp_name;
8579 	struct kstat_kv		ktmp_temperature;
8580 	struct kstat_kv		ktmp_threshold_lo;
8581 	struct kstat_kv		ktmp_threshold_hi;
8582 };
8583 
8584 static const struct mcx_kstat_mtmp mcx_kstat_mtmp_tpl = {
8585 	KSTAT_KV_INITIALIZER("name",		KSTAT_KV_T_ISTR),
8586 	KSTAT_KV_INITIALIZER("temperature",	KSTAT_KV_T_TEMP),
8587 	KSTAT_KV_INITIALIZER("lo threshold",	KSTAT_KV_T_TEMP),
8588 	KSTAT_KV_INITIALIZER("hi threshold",	KSTAT_KV_T_TEMP),
8589 };
8590 
8591 static const struct timeval mcx_kstat_mtmp_rate = { 1, 0 };
8592 
8593 static int mcx_kstat_mtmp_read(struct kstat *);
8594 
8595 static void
8596 mcx_kstat_attach_tmps(struct mcx_softc *sc)
8597 {
8598 	struct kstat *ks;
8599 	struct mcx_reg_mcam mcam;
8600 	struct mcx_reg_mtcap mtcap;
8601 	struct mcx_kstat_mtmp *ktmp;
8602 	uint64_t map;
8603 	unsigned int i, n;
8604 
8605 	memset(&mtcap, 0, sizeof(mtcap));
8606 	memset(&mcam, 0, sizeof(mcam));
8607 
8608 	if (sc->sc_mcam_reg == 0) {
8609 		/* no management capabilities */
8610 		return;
8611 	}
8612 
8613 	if (mcx_access_hca_reg(sc, MCX_REG_MCAM, MCX_REG_OP_READ,
8614 	    &mcam, sizeof(mcam)) != 0) {
8615 		/* unable to check management capabilities? */
8616 		return;
8617 	}
8618 
8619 	if (MCX_BITFIELD_BIT(mcam.mcam_feature_cap_mask,
8620 	    MCX_MCAM_FEATURE_CAP_SENSOR_MAP) == 0) {
8621 		/* no sensor map */
8622 		return;
8623 	}
8624 
8625 	if (mcx_access_hca_reg(sc, MCX_REG_MTCAP, MCX_REG_OP_READ,
8626 	    &mtcap, sizeof(mtcap)) != 0) {
8627 		/* unable to find temperature sensors */
8628 		return;
8629 	}
8630 
8631 	sc->sc_kstat_mtmp_count = mtcap.mtcap_sensor_count;
8632 	sc->sc_kstat_mtmp = mallocarray(sc->sc_kstat_mtmp_count,
8633 	    sizeof(*sc->sc_kstat_mtmp), M_DEVBUF, M_WAITOK);
8634 
8635 	n = 0;
8636 	map = bemtoh64(&mtcap.mtcap_sensor_map);
8637 	for (i = 0; i < sizeof(map) * NBBY; i++) {
8638 		if (!ISSET(map, (1ULL << i)))
8639 			continue;
8640 
8641 		ks = kstat_create(DEVNAME(sc), 0, "temperature", i,
8642 		    KSTAT_T_KV, 0);
8643 		if (ks == NULL) {
8644 			/* unable to attach temperature sensor %u, i */
8645 			continue;
8646 		}
8647 
8648 		ktmp = malloc(sizeof(*ktmp), M_DEVBUF, M_WAITOK|M_ZERO);
8649 		*ktmp = mcx_kstat_mtmp_tpl;
8650 
8651 		ks->ks_data = ktmp;
8652 		ks->ks_datalen = sizeof(*ktmp);
8653 		TIMEVAL_TO_TIMESPEC(&mcx_kstat_mtmp_rate, &ks->ks_interval);
8654 		ks->ks_read = mcx_kstat_mtmp_read;
8655 
8656 		ks->ks_softc = sc;
8657 		kstat_install(ks);
8658 
8659 		sc->sc_kstat_mtmp[n++] = ks;
8660 		if (n >= sc->sc_kstat_mtmp_count)
8661 			break;
8662 	}
8663 }
8664 
8665 static uint64_t
8666 mcx_tmp_to_uK(uint16_t *t)
8667 {
8668 	int64_t mt = (int16_t)bemtoh16(t); /* 0.125 C units */
8669 	mt *= 1000000 / 8; /* convert to uC */
8670 	mt += 273150000; /* convert to uK */
8671 
8672 	return (mt);
8673 }
8674 
8675 static int
8676 mcx_kstat_mtmp_read(struct kstat *ks)
8677 {
8678 	struct mcx_softc *sc = ks->ks_softc;
8679 	struct mcx_kstat_mtmp *ktmp = ks->ks_data;
8680 	struct mcx_reg_mtmp mtmp;
8681 	int rv;
8682 	struct timeval updated;
8683 
8684 	TIMESPEC_TO_TIMEVAL(&updated, &ks->ks_updated);
8685 
8686 	if (!ratecheck(&updated, &mcx_kstat_mtmp_rate))
8687 		return (0);
8688 
8689 	memset(&mtmp, 0, sizeof(mtmp));
8690 	htobem16(&mtmp.mtmp_sensor_index, ks->ks_unit);
8691 
8692 	KERNEL_LOCK(); /* XXX */
8693 	rv = mcx_access_hca_reg(sc, MCX_REG_MTMP, MCX_REG_OP_READ,
8694 	    &mtmp, sizeof(mtmp));
8695 	KERNEL_UNLOCK();
8696 	if (rv != 0)
8697 		return (EIO);
8698 
8699 	memset(kstat_kv_istr(&ktmp->ktmp_name), 0,
8700 	    sizeof(kstat_kv_istr(&ktmp->ktmp_name)));
8701 	memcpy(kstat_kv_istr(&ktmp->ktmp_name),
8702 	    mtmp.mtmp_sensor_name, sizeof(mtmp.mtmp_sensor_name));
8703 	kstat_kv_temp(&ktmp->ktmp_temperature) =
8704 	    mcx_tmp_to_uK(&mtmp.mtmp_temperature);
8705 	kstat_kv_temp(&ktmp->ktmp_threshold_lo) =
8706 	    mcx_tmp_to_uK(&mtmp.mtmp_temperature_threshold_lo);
8707 	kstat_kv_temp(&ktmp->ktmp_threshold_hi) =
8708 	    mcx_tmp_to_uK(&mtmp.mtmp_temperature_threshold_hi);
8709 
8710 	TIMEVAL_TO_TIMESPEC(&updated, &ks->ks_updated);
8711 
8712 	return (0);
8713 }
8714 
8715 struct mcx_queuestat {
8716 	char			 name[KSTAT_KV_NAMELEN];
8717 	enum kstat_kv_type	 type;
8718 };
8719 
8720 static const struct mcx_queuestat mcx_queue_kstat_tpl[] = {
8721 	{ "RQ SW prod",		KSTAT_KV_T_COUNTER64 },
8722 	{ "RQ HW prod",		KSTAT_KV_T_COUNTER64 },
8723 	{ "RQ HW cons",		KSTAT_KV_T_COUNTER64 },
8724 	{ "RQ HW state",	KSTAT_KV_T_ISTR },
8725 
8726 	{ "SQ SW prod",		KSTAT_KV_T_COUNTER64 },
8727 	{ "SQ SW cons",		KSTAT_KV_T_COUNTER64 },
8728 	{ "SQ HW prod",		KSTAT_KV_T_COUNTER64 },
8729 	{ "SQ HW cons",		KSTAT_KV_T_COUNTER64 },
8730 	{ "SQ HW state",	KSTAT_KV_T_ISTR },
8731 
8732 	{ "CQ SW cons",		KSTAT_KV_T_COUNTER64 },
8733 	{ "CQ HW prod",		KSTAT_KV_T_COUNTER64 },
8734 	{ "CQ HW cons",		KSTAT_KV_T_COUNTER64 },
8735 	{ "CQ HW notify",	KSTAT_KV_T_COUNTER64 },
8736 	{ "CQ HW solicit",	KSTAT_KV_T_COUNTER64 },
8737 	{ "CQ HW status",	KSTAT_KV_T_ISTR },
8738 	{ "CQ HW state",	KSTAT_KV_T_ISTR },
8739 
8740 	{ "EQ SW cons",		KSTAT_KV_T_COUNTER64 },
8741 	{ "EQ HW prod",		KSTAT_KV_T_COUNTER64 },
8742 	{ "EQ HW cons",		KSTAT_KV_T_COUNTER64 },
8743 	{ "EQ HW status",	KSTAT_KV_T_ISTR },
8744 	{ "EQ HW state",	KSTAT_KV_T_ISTR },
8745 };
8746 
8747 static int	mcx_kstat_queue_read(struct kstat *);
8748 
8749 static void
8750 mcx_kstat_attach_queues(struct mcx_softc *sc)
8751 {
8752 	struct kstat *ks;
8753 	struct kstat_kv *kvs;
8754 	int q, i;
8755 
8756 	for (q = 0; q < sc->sc_nqueues; q++) {
8757 		ks = kstat_create(DEVNAME(sc), 0, "mcx-queues", q,
8758 		    KSTAT_T_KV, 0);
8759 		if (ks == NULL) {
8760 			/* unable to attach queue stats %u, q */
8761 			continue;
8762 		}
8763 
8764 		kvs = mallocarray(nitems(mcx_queue_kstat_tpl),
8765 		    sizeof(*kvs), M_DEVBUF, M_WAITOK);
8766 
8767 		for (i = 0; i < nitems(mcx_queue_kstat_tpl); i++) {
8768 			const struct mcx_queuestat *tpl =
8769 			    &mcx_queue_kstat_tpl[i];
8770 
8771 			kstat_kv_init(&kvs[i], tpl->name, tpl->type);
8772 		}
8773 
8774 		ks->ks_softc = &sc->sc_queues[q];
8775 		ks->ks_data = kvs;
8776 		ks->ks_datalen = nitems(mcx_queue_kstat_tpl) * sizeof(*kvs);
8777 		ks->ks_read = mcx_kstat_queue_read;
8778 
8779 		sc->sc_queues[q].q_kstat = ks;
8780 		kstat_install(ks);
8781 	}
8782 }
8783 
8784 static int
8785 mcx_kstat_queue_read(struct kstat *ks)
8786 {
8787 	struct mcx_queues *q = ks->ks_softc;
8788 	struct mcx_softc *sc = q->q_sc;
8789 	struct kstat_kv *kvs = ks->ks_data;
8790 	union {
8791 		struct mcx_rq_ctx rq;
8792 		struct mcx_sq_ctx sq;
8793 		struct mcx_cq_ctx cq;
8794 		struct mcx_eq_ctx eq;
8795 	} u;
8796 	const char *text;
8797 	int error = 0;
8798 
8799 	KERNEL_LOCK();
8800 
8801 	if (mcx_query_rq(sc, &q->q_rx, &u.rq) != 0) {
8802 		error = EIO;
8803 		goto out;
8804 	}
8805 
8806 	kstat_kv_u64(kvs++) = q->q_rx.rx_prod;
8807 	kstat_kv_u64(kvs++) = bemtoh32(&u.rq.rq_wq.wq_sw_counter);
8808 	kstat_kv_u64(kvs++) = bemtoh32(&u.rq.rq_wq.wq_hw_counter);
8809 	switch ((bemtoh32(&u.rq.rq_flags) & MCX_RQ_CTX_STATE_MASK) >>
8810 	    MCX_RQ_CTX_STATE_SHIFT) {
8811 	case MCX_RQ_CTX_STATE_RST:
8812 		text = "RST";
8813 		break;
8814 	case MCX_RQ_CTX_STATE_RDY:
8815 		text = "RDY";
8816 		break;
8817 	case MCX_RQ_CTX_STATE_ERR:
8818 		text = "ERR";
8819 		break;
8820 	default:
8821 		text = "unknown";
8822 		break;
8823 	}
8824 	strlcpy(kstat_kv_istr(kvs), text, sizeof(kstat_kv_istr(kvs)));
8825 	kvs++;
8826 
8827 	if (mcx_query_sq(sc, &q->q_tx, &u.sq) != 0) {
8828 		error = EIO;
8829 		goto out;
8830 	}
8831 
8832 	kstat_kv_u64(kvs++) = q->q_tx.tx_prod;
8833 	kstat_kv_u64(kvs++) = q->q_tx.tx_cons;
8834 	kstat_kv_u64(kvs++) = bemtoh32(&u.sq.sq_wq.wq_sw_counter);
8835 	kstat_kv_u64(kvs++) = bemtoh32(&u.sq.sq_wq.wq_hw_counter);
8836 	switch ((bemtoh32(&u.sq.sq_flags) & MCX_SQ_CTX_STATE_MASK) >>
8837 	    MCX_SQ_CTX_STATE_SHIFT) {
8838 	case MCX_SQ_CTX_STATE_RST:
8839 		text = "RST";
8840 		break;
8841 	case MCX_SQ_CTX_STATE_RDY:
8842 		text = "RDY";
8843 		break;
8844 	case MCX_SQ_CTX_STATE_ERR:
8845 		text = "ERR";
8846 		break;
8847 	default:
8848 		text = "unknown";
8849 		break;
8850 	}
8851 	strlcpy(kstat_kv_istr(kvs), text, sizeof(kstat_kv_istr(kvs)));
8852 	kvs++;
8853 
8854 	if (mcx_query_cq(sc, &q->q_cq, &u.cq) != 0) {
8855 		error = EIO;
8856 		goto out;
8857 	}
8858 
8859 	kstat_kv_u64(kvs++) = q->q_cq.cq_cons;
8860 	kstat_kv_u64(kvs++) = bemtoh32(&u.cq.cq_producer_counter);
8861 	kstat_kv_u64(kvs++) = bemtoh32(&u.cq.cq_consumer_counter);
8862 	kstat_kv_u64(kvs++) = bemtoh32(&u.cq.cq_last_notified);
8863 	kstat_kv_u64(kvs++) = bemtoh32(&u.cq.cq_last_solicit);
8864 
8865 	switch ((bemtoh32(&u.cq.cq_status) & MCX_CQ_CTX_STATUS_MASK) >>
8866 	    MCX_CQ_CTX_STATUS_SHIFT) {
8867 	case MCX_CQ_CTX_STATUS_OK:
8868 		text = "OK";
8869 		break;
8870 	case MCX_CQ_CTX_STATUS_OVERFLOW:
8871 		text = "overflow";
8872 		break;
8873 	case MCX_CQ_CTX_STATUS_WRITE_FAIL:
8874 		text = "write fail";
8875 		break;
8876 	default:
8877 		text = "unknown";
8878 		break;
8879 	}
8880 	strlcpy(kstat_kv_istr(kvs), text, sizeof(kstat_kv_istr(kvs)));
8881 	kvs++;
8882 
8883 	switch ((bemtoh32(&u.cq.cq_status) & MCX_CQ_CTX_STATE_MASK) >>
8884 	    MCX_CQ_CTX_STATE_SHIFT) {
8885 	case MCX_CQ_CTX_STATE_SOLICITED:
8886 		text = "solicited";
8887 		break;
8888 	case MCX_CQ_CTX_STATE_ARMED:
8889 		text = "armed";
8890 		break;
8891 	case MCX_CQ_CTX_STATE_FIRED:
8892 		text = "fired";
8893 		break;
8894 	default:
8895 		text = "unknown";
8896 		break;
8897 	}
8898 	strlcpy(kstat_kv_istr(kvs), text, sizeof(kstat_kv_istr(kvs)));
8899 	kvs++;
8900 
8901 	if (mcx_query_eq(sc, &q->q_eq, &u.eq) != 0) {
8902 		error = EIO;
8903 		goto out;
8904 	}
8905 
8906 	kstat_kv_u64(kvs++) = q->q_eq.eq_cons;
8907 	kstat_kv_u64(kvs++) = bemtoh32(&u.eq.eq_producer_counter);
8908 	kstat_kv_u64(kvs++) = bemtoh32(&u.eq.eq_consumer_counter);
8909 
8910 	switch ((bemtoh32(&u.eq.eq_status) & MCX_EQ_CTX_STATUS_MASK) >>
8911 	    MCX_EQ_CTX_STATUS_SHIFT) {
8912 	case MCX_EQ_CTX_STATUS_EQ_WRITE_FAILURE:
8913 		text = "write fail";
8914 		break;
8915 	case MCX_EQ_CTX_STATUS_OK:
8916 		text = "OK";
8917 		break;
8918 	default:
8919 		text = "unknown";
8920 		break;
8921 	}
8922 	strlcpy(kstat_kv_istr(kvs), text, sizeof(kstat_kv_istr(kvs)));
8923 	kvs++;
8924 
8925 	switch ((bemtoh32(&u.eq.eq_status) & MCX_EQ_CTX_STATE_MASK) >>
8926 	    MCX_EQ_CTX_STATE_SHIFT) {
8927 	case MCX_EQ_CTX_STATE_ARMED:
8928 		text = "armed";
8929 		break;
8930 	case MCX_EQ_CTX_STATE_FIRED:
8931 		text = "fired";
8932 		break;
8933 	default:
8934 		text = "unknown";
8935 		break;
8936 	}
8937 	strlcpy(kstat_kv_istr(kvs), text, sizeof(kstat_kv_istr(kvs)));
8938 	kvs++;
8939 
8940 	nanouptime(&ks->ks_updated);
8941 out:
8942 	KERNEL_UNLOCK();
8943 	return (error);
8944 }
8945 
8946 #endif /* NKSTAT > 0 */
8947 
8948 static unsigned int
8949 mcx_timecounter_read(struct timecounter *tc)
8950 {
8951 	struct mcx_softc *sc = tc->tc_priv;
8952 
8953 	return (mcx_rd(sc, MCX_INTERNAL_TIMER_L));
8954 }
8955 
8956 static void
8957 mcx_timecounter_attach(struct mcx_softc *sc)
8958 {
8959 	struct timecounter *tc = &sc->sc_timecounter;
8960 
8961 	tc->tc_get_timecount = mcx_timecounter_read;
8962 	tc->tc_counter_mask = ~0U;
8963 	tc->tc_frequency = sc->sc_khz * 1000;
8964 	tc->tc_name = device_xname(sc->sc_dev);
8965 	tc->tc_quality = -100;
8966 	tc->tc_priv = sc;
8967 
8968 	tc_init(tc);
8969 }
8970