xref: /netbsd-src/sys/dev/pci/if_mcx.c (revision 122b5006ee1bd67145794b4cde92f4fe4781a5ec)
1 /*	$NetBSD: if_mcx.c,v 1.22 2021/09/26 20:14:07 jmcneill 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 1.22 2021/09/26 20:14:07 jmcneill 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 
47 #include <machine/intr.h>
48 
49 #include <net/if.h>
50 #include <net/if_dl.h>
51 #include <net/if_ether.h>
52 #include <net/if_media.h>
53 #include <net/if_vlanvar.h>
54 #include <net/toeplitz.h>
55 
56 #include <net/bpf.h>
57 
58 #include <netinet/in.h>
59 
60 #include <dev/pci/pcireg.h>
61 #include <dev/pci/pcivar.h>
62 #include <dev/pci/pcidevs.h>
63 
64 /* TODO: Port kstat key/value stuff to evcnt/sysmon */
65 #define	NKSTAT		0
66 
67 /* XXX This driver is not yet MP-safe; don't claim to be! */
68 /* #ifdef NET_MPSAFE */
69 /* #define	MCX_MPSAFE	1 */
70 /* #define	CALLOUT_FLAGS	CALLOUT_MPSAFE */
71 /* #else */
72 #define	CALLOUT_FLAGS	0
73 /* #endif */
74 
75 #define	MCX_TXQ_NUM		2048
76 
77 #define BUS_DMASYNC_PRERW	(BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE)
78 #define BUS_DMASYNC_POSTRW	(BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE)
79 
80 #define MCX_HCA_BAR	PCI_MAPREG_START /* BAR 0 */
81 
82 #define MCX_FW_VER			0x0000
83 #define  MCX_FW_VER_MAJOR(_v)			((_v) & 0xffff)
84 #define  MCX_FW_VER_MINOR(_v)			((_v) >> 16)
85 #define MCX_CMDIF_FW_SUBVER		0x0004
86 #define  MCX_FW_VER_SUBMINOR(_v)		((_v) & 0xffff)
87 #define  MCX_CMDIF(_v)				((_v) >> 16)
88 
89 #define MCX_ISSI			1 /* as per the PRM */
90 #define MCX_CMD_IF_SUPPORTED		5
91 
92 #define MCX_HARDMTU			9500
93 
94 #define MCX_PAGE_SHIFT			12
95 #define MCX_PAGE_SIZE			(1 << MCX_PAGE_SHIFT)
96 
97 /* queue sizes */
98 #define MCX_LOG_EQ_SIZE			7
99 #define MCX_LOG_CQ_SIZE			12
100 #define MCX_LOG_RQ_SIZE			10
101 #define MCX_LOG_SQ_SIZE			11
102 
103 #define MCX_MAX_QUEUES			16
104 
105 /* completion event moderation - about 10khz, or 90% of the cq */
106 #define MCX_CQ_MOD_PERIOD		50
107 #define MCX_CQ_MOD_COUNTER		\
108 	(((1 << (MCX_LOG_CQ_SIZE - 1)) * 9) / 10)
109 
110 #define MCX_LOG_SQ_ENTRY_SIZE		6
111 #define MCX_SQ_ENTRY_MAX_SLOTS		4
112 #define MCX_SQ_SEGS_PER_SLOT		\
113 	(sizeof(struct mcx_sq_entry) / sizeof(struct mcx_sq_entry_seg))
114 #define MCX_SQ_MAX_SEGMENTS		\
115 	1 + ((MCX_SQ_ENTRY_MAX_SLOTS-1) * MCX_SQ_SEGS_PER_SLOT)
116 
117 #define MCX_LOG_FLOW_TABLE_SIZE		5
118 #define MCX_NUM_STATIC_FLOWS		4 /* promisc, allmulti, ucast, bcast */
119 #define MCX_NUM_MCAST_FLOWS 		\
120 	((1 << MCX_LOG_FLOW_TABLE_SIZE) - MCX_NUM_STATIC_FLOWS)
121 
122 #define MCX_SQ_INLINE_SIZE		18
123 CTASSERT(ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN == MCX_SQ_INLINE_SIZE);
124 
125 /* doorbell offsets */
126 #define MCX_DOORBELL_AREA_SIZE		MCX_PAGE_SIZE
127 
128 #define MCX_CQ_DOORBELL_BASE		0
129 #define MCX_CQ_DOORBELL_STRIDE		64
130 
131 #define MCX_WQ_DOORBELL_BASE		MCX_PAGE_SIZE/2
132 #define MCX_WQ_DOORBELL_STRIDE		64
133 /* make sure the doorbells fit */
134 CTASSERT(MCX_MAX_QUEUES * MCX_CQ_DOORBELL_STRIDE < MCX_WQ_DOORBELL_BASE);
135 CTASSERT(MCX_MAX_QUEUES * MCX_WQ_DOORBELL_STRIDE <
136     MCX_DOORBELL_AREA_SIZE - MCX_WQ_DOORBELL_BASE);
137 
138 #define MCX_WQ_DOORBELL_MASK		0xffff
139 
140 /* uar registers */
141 #define MCX_UAR_CQ_DOORBELL		0x20
142 #define MCX_UAR_EQ_DOORBELL_ARM		0x40
143 #define MCX_UAR_EQ_DOORBELL		0x48
144 #define MCX_UAR_BF			0x800
145 
146 #define MCX_CMDQ_ADDR_HI		0x0010
147 #define MCX_CMDQ_ADDR_LO		0x0014
148 #define MCX_CMDQ_ADDR_NMASK		0xfff
149 #define MCX_CMDQ_LOG_SIZE(_v)		((_v) >> 4 & 0xf)
150 #define MCX_CMDQ_LOG_STRIDE(_v)		((_v) >> 0 & 0xf)
151 #define MCX_CMDQ_INTERFACE_MASK		(0x3 << 8)
152 #define MCX_CMDQ_INTERFACE_FULL_DRIVER	(0x0 << 8)
153 #define MCX_CMDQ_INTERFACE_DISABLED	(0x1 << 8)
154 
155 #define MCX_CMDQ_DOORBELL		0x0018
156 
157 #define MCX_STATE			0x01fc
158 #define MCX_STATE_MASK				(1U << 31)
159 #define MCX_STATE_INITIALIZING			(1 << 31)
160 #define MCX_STATE_READY				(0 << 31)
161 #define MCX_STATE_INTERFACE_MASK		(0x3 << 24)
162 #define MCX_STATE_INTERFACE_FULL_DRIVER		(0x0 << 24)
163 #define MCX_STATE_INTERFACE_DISABLED		(0x1 << 24)
164 
165 #define MCX_INTERNAL_TIMER		0x1000
166 #define MCX_INTERNAL_TIMER_H		0x1000
167 #define MCX_INTERNAL_TIMER_L		0x1004
168 
169 #define MCX_CLEAR_INT			0x100c
170 
171 #define MCX_REG_OP_WRITE		0
172 #define MCX_REG_OP_READ			1
173 
174 #define MCX_REG_PMLP			0x5002
175 #define MCX_REG_PMTU			0x5003
176 #define MCX_REG_PTYS			0x5004
177 #define MCX_REG_PAOS			0x5006
178 #define MCX_REG_PFCC			0x5007
179 #define MCX_REG_PPCNT			0x5008
180 #define MCX_REG_MTCAP			0x9009 /* mgmt temp capabilities */
181 #define MCX_REG_MTMP			0x900a /* mgmt temp */
182 #define MCX_REG_MCIA			0x9014
183 #define MCX_REG_MCAM			0x907f
184 
185 #define MCX_ETHER_CAP_SGMII		0
186 #define MCX_ETHER_CAP_1000_KX		1
187 #define MCX_ETHER_CAP_10G_CX4		2
188 #define MCX_ETHER_CAP_10G_KX4		3
189 #define MCX_ETHER_CAP_10G_KR		4
190 #define MCX_ETHER_CAP_20G_KR2		5
191 #define MCX_ETHER_CAP_40G_CR4		6
192 #define MCX_ETHER_CAP_40G_KR4		7
193 #define MCX_ETHER_CAP_56G_R4		8
194 #define MCX_ETHER_CAP_10G_CR		12
195 #define MCX_ETHER_CAP_10G_SR		13
196 #define MCX_ETHER_CAP_10G_LR		14
197 #define MCX_ETHER_CAP_40G_SR4		15
198 #define MCX_ETHER_CAP_40G_LR4		16
199 #define MCX_ETHER_CAP_50G_SR2		18
200 #define MCX_ETHER_CAP_100G_CR4		20
201 #define MCX_ETHER_CAP_100G_SR4		21
202 #define MCX_ETHER_CAP_100G_KR4		22
203 #define MCX_ETHER_CAP_100G_LR4		23
204 #define MCX_ETHER_CAP_100_TX		24
205 #define MCX_ETHER_CAP_1000_T		25
206 #define MCX_ETHER_CAP_10G_T		26
207 #define MCX_ETHER_CAP_25G_CR		27
208 #define MCX_ETHER_CAP_25G_KR		28
209 #define MCX_ETHER_CAP_25G_SR		29
210 #define MCX_ETHER_CAP_50G_CR2		30
211 #define MCX_ETHER_CAP_50G_KR2		31
212 
213 #define MCX_MAX_CQE			32
214 
215 #define MCX_CMD_QUERY_HCA_CAP		0x100
216 #define MCX_CMD_QUERY_ADAPTER		0x101
217 #define MCX_CMD_INIT_HCA		0x102
218 #define MCX_CMD_TEARDOWN_HCA		0x103
219 #define MCX_CMD_ENABLE_HCA		0x104
220 #define MCX_CMD_DISABLE_HCA		0x105
221 #define MCX_CMD_QUERY_PAGES		0x107
222 #define MCX_CMD_MANAGE_PAGES		0x108
223 #define MCX_CMD_SET_HCA_CAP		0x109
224 #define MCX_CMD_QUERY_ISSI		0x10a
225 #define MCX_CMD_SET_ISSI		0x10b
226 #define MCX_CMD_SET_DRIVER_VERSION	0x10d
227 #define MCX_CMD_QUERY_SPECIAL_CONTEXTS	0x203
228 #define MCX_CMD_CREATE_EQ		0x301
229 #define MCX_CMD_DESTROY_EQ		0x302
230 #define MCX_CMD_QUERY_EQ		0x303
231 #define MCX_CMD_CREATE_CQ		0x400
232 #define MCX_CMD_DESTROY_CQ		0x401
233 #define MCX_CMD_QUERY_CQ		0x402
234 #define MCX_CMD_QUERY_NIC_VPORT_CONTEXT 0x754
235 #define MCX_CMD_MODIFY_NIC_VPORT_CONTEXT \
236 					0x755
237 #define MCX_CMD_QUERY_VPORT_COUNTERS 	0x770
238 #define MCX_CMD_ALLOC_PD		0x800
239 #define MCX_CMD_ALLOC_UAR		0x802
240 #define MCX_CMD_ACCESS_REG		0x805
241 #define MCX_CMD_ALLOC_TRANSPORT_DOMAIN	0x816
242 #define MCX_CMD_CREATE_TIR		0x900
243 #define MCX_CMD_DESTROY_TIR		0x902
244 #define MCX_CMD_CREATE_SQ		0x904
245 #define MCX_CMD_MODIFY_SQ		0x905
246 #define MCX_CMD_DESTROY_SQ		0x906
247 #define MCX_CMD_QUERY_SQ		0x907
248 #define MCX_CMD_CREATE_RQ		0x908
249 #define MCX_CMD_MODIFY_RQ		0x909
250 #define MCX_CMD_DESTROY_RQ		0x90a
251 #define MCX_CMD_QUERY_RQ		0x90b
252 #define MCX_CMD_CREATE_TIS		0x912
253 #define MCX_CMD_DESTROY_TIS		0x914
254 #define MCX_CMD_CREATE_RQT		0x916
255 #define MCX_CMD_DESTROY_RQT		0x918
256 #define MCX_CMD_SET_FLOW_TABLE_ROOT	0x92f
257 #define MCX_CMD_CREATE_FLOW_TABLE	0x930
258 #define MCX_CMD_DESTROY_FLOW_TABLE	0x931
259 #define MCX_CMD_QUERY_FLOW_TABLE	0x932
260 #define MCX_CMD_CREATE_FLOW_GROUP	0x933
261 #define MCX_CMD_DESTROY_FLOW_GROUP	0x934
262 #define MCX_CMD_QUERY_FLOW_GROUP	0x935
263 #define MCX_CMD_SET_FLOW_TABLE_ENTRY	0x936
264 #define MCX_CMD_QUERY_FLOW_TABLE_ENTRY	0x937
265 #define MCX_CMD_DELETE_FLOW_TABLE_ENTRY	0x938
266 #define MCX_CMD_ALLOC_FLOW_COUNTER	0x939
267 #define MCX_CMD_QUERY_FLOW_COUNTER	0x93b
268 
269 #define MCX_QUEUE_STATE_RST		0
270 #define MCX_QUEUE_STATE_RDY		1
271 #define MCX_QUEUE_STATE_ERR		3
272 
273 #define MCX_FLOW_TABLE_TYPE_RX		0
274 #define MCX_FLOW_TABLE_TYPE_TX		1
275 
276 #define MCX_CMDQ_INLINE_DATASIZE	16
277 
278 struct mcx_cmdq_entry {
279 	uint8_t			cq_type;
280 #define MCX_CMDQ_TYPE_PCIE		0x7
281 	uint8_t			cq_reserved0[3];
282 
283 	uint32_t		cq_input_length;
284 	uint64_t		cq_input_ptr;
285 	uint8_t			cq_input_data[MCX_CMDQ_INLINE_DATASIZE];
286 
287 	uint8_t			cq_output_data[MCX_CMDQ_INLINE_DATASIZE];
288 	uint64_t		cq_output_ptr;
289 	uint32_t		cq_output_length;
290 
291 	uint8_t			cq_token;
292 	uint8_t			cq_signature;
293 	uint8_t			cq_reserved1[1];
294 	uint8_t			cq_status;
295 #define MCX_CQ_STATUS_SHIFT		1
296 #define MCX_CQ_STATUS_MASK		(0x7f << MCX_CQ_STATUS_SHIFT)
297 #define MCX_CQ_STATUS_OK		(0x00 << MCX_CQ_STATUS_SHIFT)
298 #define MCX_CQ_STATUS_INT_ERR		(0x01 << MCX_CQ_STATUS_SHIFT)
299 #define MCX_CQ_STATUS_BAD_OPCODE	(0x02 << MCX_CQ_STATUS_SHIFT)
300 #define MCX_CQ_STATUS_BAD_PARAM		(0x03 << MCX_CQ_STATUS_SHIFT)
301 #define MCX_CQ_STATUS_BAD_SYS_STATE	(0x04 << MCX_CQ_STATUS_SHIFT)
302 #define MCX_CQ_STATUS_BAD_RESOURCE	(0x05 << MCX_CQ_STATUS_SHIFT)
303 #define MCX_CQ_STATUS_RESOURCE_BUSY	(0x06 << MCX_CQ_STATUS_SHIFT)
304 #define MCX_CQ_STATUS_EXCEED_LIM	(0x08 << MCX_CQ_STATUS_SHIFT)
305 #define MCX_CQ_STATUS_BAD_RES_STATE	(0x09 << MCX_CQ_STATUS_SHIFT)
306 #define MCX_CQ_STATUS_BAD_INDEX		(0x0a << MCX_CQ_STATUS_SHIFT)
307 #define MCX_CQ_STATUS_NO_RESOURCES	(0x0f << MCX_CQ_STATUS_SHIFT)
308 #define MCX_CQ_STATUS_BAD_INPUT_LEN	(0x50 << MCX_CQ_STATUS_SHIFT)
309 #define MCX_CQ_STATUS_BAD_OUTPUT_LEN	(0x51 << MCX_CQ_STATUS_SHIFT)
310 #define MCX_CQ_STATUS_BAD_RESOURCE_STATE \
311 					(0x10 << MCX_CQ_STATUS_SHIFT)
312 #define MCX_CQ_STATUS_BAD_SIZE		(0x40 << MCX_CQ_STATUS_SHIFT)
313 #define MCX_CQ_STATUS_OWN_MASK		0x1
314 #define MCX_CQ_STATUS_OWN_SW		0x0
315 #define MCX_CQ_STATUS_OWN_HW		0x1
316 } __packed __aligned(8);
317 
318 #define MCX_CMDQ_MAILBOX_DATASIZE	512
319 
320 struct mcx_cmdq_mailbox {
321 	uint8_t			mb_data[MCX_CMDQ_MAILBOX_DATASIZE];
322 	uint8_t			mb_reserved0[48];
323 	uint64_t		mb_next_ptr;
324 	uint32_t		mb_block_number;
325 	uint8_t			mb_reserved1[1];
326 	uint8_t			mb_token;
327 	uint8_t			mb_ctrl_signature;
328 	uint8_t			mb_signature;
329 } __packed __aligned(8);
330 
331 #define MCX_CMDQ_MAILBOX_ALIGN	(1 << 10)
332 #define MCX_CMDQ_MAILBOX_SIZE	roundup(sizeof(struct mcx_cmdq_mailbox), \
333 				    MCX_CMDQ_MAILBOX_ALIGN)
334 /*
335  * command mailbox structures
336  */
337 
338 struct mcx_cmd_enable_hca_in {
339 	uint16_t		cmd_opcode;
340 	uint8_t			cmd_reserved0[4];
341 	uint16_t		cmd_op_mod;
342 	uint8_t			cmd_reserved1[2];
343 	uint16_t		cmd_function_id;
344 	uint8_t			cmd_reserved2[4];
345 } __packed __aligned(4);
346 
347 struct mcx_cmd_enable_hca_out {
348 	uint8_t			cmd_status;
349 	uint8_t			cmd_reserved0[3];
350 	uint32_t		cmd_syndrome;
351 	uint8_t			cmd_reserved1[4];
352 } __packed __aligned(4);
353 
354 struct mcx_cmd_init_hca_in {
355 	uint16_t		cmd_opcode;
356 	uint8_t			cmd_reserved0[4];
357 	uint16_t		cmd_op_mod;
358 	uint8_t			cmd_reserved1[8];
359 } __packed __aligned(4);
360 
361 struct mcx_cmd_init_hca_out {
362 	uint8_t			cmd_status;
363 	uint8_t			cmd_reserved0[3];
364 	uint32_t		cmd_syndrome;
365 	uint8_t			cmd_reserved1[8];
366 } __packed __aligned(4);
367 
368 struct mcx_cmd_teardown_hca_in {
369 	uint16_t		cmd_opcode;
370 	uint8_t			cmd_reserved0[4];
371 	uint16_t		cmd_op_mod;
372 	uint8_t			cmd_reserved1[2];
373 #define MCX_CMD_TEARDOWN_HCA_GRACEFUL	0x0
374 #define MCX_CMD_TEARDOWN_HCA_PANIC	0x1
375 	uint16_t		cmd_profile;
376 	uint8_t			cmd_reserved2[4];
377 } __packed __aligned(4);
378 
379 struct mcx_cmd_teardown_hca_out {
380 	uint8_t			cmd_status;
381 	uint8_t			cmd_reserved0[3];
382 	uint32_t		cmd_syndrome;
383 	uint8_t			cmd_reserved1[8];
384 } __packed __aligned(4);
385 
386 struct mcx_cmd_access_reg_in {
387 	uint16_t		cmd_opcode;
388 	uint8_t			cmd_reserved0[4];
389 	uint16_t		cmd_op_mod;
390 	uint8_t			cmd_reserved1[2];
391 	uint16_t		cmd_register_id;
392 	uint32_t		cmd_argument;
393 } __packed __aligned(4);
394 
395 struct mcx_cmd_access_reg_out {
396 	uint8_t			cmd_status;
397 	uint8_t			cmd_reserved0[3];
398 	uint32_t		cmd_syndrome;
399 	uint8_t			cmd_reserved1[8];
400 } __packed __aligned(4);
401 
402 struct mcx_reg_pmtu {
403 	uint8_t			rp_reserved1;
404 	uint8_t			rp_local_port;
405 	uint8_t			rp_reserved2[2];
406 	uint16_t		rp_max_mtu;
407 	uint8_t			rp_reserved3[2];
408 	uint16_t		rp_admin_mtu;
409 	uint8_t			rp_reserved4[2];
410 	uint16_t		rp_oper_mtu;
411 	uint8_t			rp_reserved5[2];
412 } __packed __aligned(4);
413 
414 struct mcx_reg_ptys {
415 	uint8_t			rp_reserved1;
416 	uint8_t			rp_local_port;
417 	uint8_t			rp_reserved2;
418 	uint8_t			rp_proto_mask;
419 #define MCX_REG_PTYS_PROTO_MASK_ETH		(1 << 2)
420 	uint8_t			rp_reserved3[8];
421 	uint32_t		rp_eth_proto_cap;
422 	uint8_t			rp_reserved4[8];
423 	uint32_t		rp_eth_proto_admin;
424 	uint8_t			rp_reserved5[8];
425 	uint32_t		rp_eth_proto_oper;
426 	uint8_t			rp_reserved6[24];
427 } __packed __aligned(4);
428 
429 struct mcx_reg_paos {
430 	uint8_t			rp_reserved1;
431 	uint8_t			rp_local_port;
432 	uint8_t			rp_admin_status;
433 #define MCX_REG_PAOS_ADMIN_STATUS_UP		1
434 #define MCX_REG_PAOS_ADMIN_STATUS_DOWN		2
435 #define MCX_REG_PAOS_ADMIN_STATUS_UP_ONCE	3
436 #define MCX_REG_PAOS_ADMIN_STATUS_DISABLED	4
437 	uint8_t			rp_oper_status;
438 #define MCX_REG_PAOS_OPER_STATUS_UP		1
439 #define MCX_REG_PAOS_OPER_STATUS_DOWN		2
440 #define MCX_REG_PAOS_OPER_STATUS_FAILED		4
441 	uint8_t			rp_admin_state_update;
442 #define MCX_REG_PAOS_ADMIN_STATE_UPDATE_EN	(1 << 7)
443 	uint8_t			rp_reserved2[11];
444 } __packed __aligned(4);
445 
446 struct mcx_reg_pfcc {
447 	uint8_t			rp_reserved1;
448 	uint8_t			rp_local_port;
449 	uint8_t			rp_reserved2[3];
450 	uint8_t			rp_prio_mask_tx;
451 	uint8_t			rp_reserved3;
452 	uint8_t			rp_prio_mask_rx;
453 	uint8_t			rp_pptx_aptx;
454 	uint8_t			rp_pfctx;
455 	uint8_t			rp_fctx_dis;
456 	uint8_t			rp_reserved4;
457 	uint8_t			rp_pprx_aprx;
458 	uint8_t			rp_pfcrx;
459 	uint8_t			rp_reserved5[2];
460 	uint16_t		rp_dev_stall_min;
461 	uint16_t		rp_dev_stall_crit;
462 	uint8_t			rp_reserved6[12];
463 } __packed __aligned(4);
464 
465 #define MCX_PMLP_MODULE_NUM_MASK	0xff
466 struct mcx_reg_pmlp {
467 	uint8_t			rp_rxtx;
468 	uint8_t			rp_local_port;
469 	uint8_t			rp_reserved0;
470 	uint8_t			rp_width;
471 	uint32_t		rp_lane0_mapping;
472 	uint32_t		rp_lane1_mapping;
473 	uint32_t		rp_lane2_mapping;
474 	uint32_t		rp_lane3_mapping;
475 	uint8_t			rp_reserved1[44];
476 } __packed __aligned(4);
477 
478 struct mcx_reg_ppcnt {
479 	uint8_t			ppcnt_swid;
480 	uint8_t			ppcnt_local_port;
481 	uint8_t			ppcnt_pnat;
482 	uint8_t			ppcnt_grp;
483 #define MCX_REG_PPCNT_GRP_IEEE8023		0x00
484 #define MCX_REG_PPCNT_GRP_RFC2863		0x01
485 #define MCX_REG_PPCNT_GRP_RFC2819		0x02
486 #define MCX_REG_PPCNT_GRP_RFC3635		0x03
487 #define MCX_REG_PPCNT_GRP_PER_PRIO		0x10
488 #define MCX_REG_PPCNT_GRP_PER_TC		0x11
489 #define MCX_REG_PPCNT_GRP_PER_RX_BUFFER		0x11
490 
491 	uint8_t			ppcnt_clr;
492 	uint8_t			ppcnt_reserved1[2];
493 	uint8_t			ppcnt_prio_tc;
494 #define MCX_REG_PPCNT_CLR			(1 << 7)
495 
496 	uint8_t			ppcnt_counter_set[248];
497 } __packed __aligned(8);
498 CTASSERT(sizeof(struct mcx_reg_ppcnt) == 256);
499 CTASSERT((offsetof(struct mcx_reg_ppcnt, ppcnt_counter_set) %
500     sizeof(uint64_t)) == 0);
501 
502 enum mcx_ppcnt_ieee8023 {
503 	frames_transmitted_ok,
504 	frames_received_ok,
505 	frame_check_sequence_errors,
506 	alignment_errors,
507 	octets_transmitted_ok,
508 	octets_received_ok,
509 	multicast_frames_xmitted_ok,
510 	broadcast_frames_xmitted_ok,
511 	multicast_frames_received_ok,
512 	broadcast_frames_received_ok,
513 	in_range_length_errors,
514 	out_of_range_length_field,
515 	frame_too_long_errors,
516 	symbol_error_during_carrier,
517 	mac_control_frames_transmitted,
518 	mac_control_frames_received,
519 	unsupported_opcodes_received,
520 	pause_mac_ctrl_frames_received,
521 	pause_mac_ctrl_frames_transmitted,
522 
523 	mcx_ppcnt_ieee8023_count
524 };
525 CTASSERT(mcx_ppcnt_ieee8023_count * sizeof(uint64_t) == 0x98);
526 
527 enum mcx_ppcnt_rfc2863 {
528 	in_octets,
529 	in_ucast_pkts,
530 	in_discards,
531 	in_errors,
532 	in_unknown_protos,
533 	out_octets,
534 	out_ucast_pkts,
535 	out_discards,
536 	out_errors,
537 	in_multicast_pkts,
538 	in_broadcast_pkts,
539 	out_multicast_pkts,
540 	out_broadcast_pkts,
541 
542 	mcx_ppcnt_rfc2863_count
543 };
544 CTASSERT(mcx_ppcnt_rfc2863_count * sizeof(uint64_t) == 0x68);
545 
546 enum mcx_ppcnt_rfc2819 {
547 	drop_events,
548 	octets,
549 	pkts,
550 	broadcast_pkts,
551 	multicast_pkts,
552 	crc_align_errors,
553 	undersize_pkts,
554 	oversize_pkts,
555 	fragments,
556 	jabbers,
557 	collisions,
558 	pkts64octets,
559 	pkts65to127octets,
560 	pkts128to255octets,
561 	pkts256to511octets,
562 	pkts512to1023octets,
563 	pkts1024to1518octets,
564 	pkts1519to2047octets,
565 	pkts2048to4095octets,
566 	pkts4096to8191octets,
567 	pkts8192to10239octets,
568 
569 	mcx_ppcnt_rfc2819_count
570 };
571 CTASSERT((mcx_ppcnt_rfc2819_count * sizeof(uint64_t)) == 0xa8);
572 
573 enum mcx_ppcnt_rfc3635 {
574 	dot3stats_alignment_errors,
575 	dot3stats_fcs_errors,
576 	dot3stats_single_collision_frames,
577 	dot3stats_multiple_collision_frames,
578 	dot3stats_sqe_test_errors,
579 	dot3stats_deferred_transmissions,
580 	dot3stats_late_collisions,
581 	dot3stats_excessive_collisions,
582 	dot3stats_internal_mac_transmit_errors,
583 	dot3stats_carrier_sense_errors,
584 	dot3stats_frame_too_longs,
585 	dot3stats_internal_mac_receive_errors,
586 	dot3stats_symbol_errors,
587 	dot3control_in_unknown_opcodes,
588 	dot3in_pause_frames,
589 	dot3out_pause_frames,
590 
591 	mcx_ppcnt_rfc3635_count
592 };
593 CTASSERT((mcx_ppcnt_rfc3635_count * sizeof(uint64_t)) == 0x80);
594 
595 struct mcx_reg_mcam {
596 	uint8_t			_reserved1[1];
597 	uint8_t			mcam_feature_group;
598 	uint8_t			_reserved2[1];
599 	uint8_t			mcam_access_reg_group;
600 	uint8_t			_reserved3[4];
601 	uint8_t			mcam_access_reg_cap_mask[16];
602 	uint8_t			_reserved4[16];
603 	uint8_t			mcam_feature_cap_mask[16];
604 	uint8_t			_reserved5[16];
605 } __packed __aligned(4);
606 
607 #define MCX_BITFIELD_BIT(bf, b)	(bf[(sizeof bf - 1) - (b / 8)] & (b % 8))
608 
609 #define MCX_MCAM_FEATURE_CAP_SENSOR_MAP	6
610 
611 struct mcx_reg_mtcap {
612 	uint8_t			_reserved1[3];
613 	uint8_t			mtcap_sensor_count;
614 	uint8_t			_reserved2[4];
615 
616 	uint64_t		mtcap_sensor_map;
617 };
618 
619 struct mcx_reg_mtmp {
620 	uint8_t			_reserved1[2];
621 	uint16_t		mtmp_sensor_index;
622 
623 	uint8_t			_reserved2[2];
624 	uint16_t		mtmp_temperature;
625 
626 	uint16_t		mtmp_mte_mtr;
627 #define MCX_REG_MTMP_MTE		(1 << 15)
628 #define MCX_REG_MTMP_MTR		(1 << 14)
629 	uint16_t		mtmp_max_temperature;
630 
631 	uint16_t		mtmp_tee;
632 #define MCX_REG_MTMP_TEE_NOPE		(0 << 14)
633 #define MCX_REG_MTMP_TEE_GENERATE	(1 << 14)
634 #define MCX_REG_MTMP_TEE_GENERATE_ONE	(2 << 14)
635 	uint16_t		mtmp_temperature_threshold_hi;
636 
637 	uint8_t			_reserved3[2];
638 	uint16_t		mtmp_temperature_threshold_lo;
639 
640 	uint8_t			_reserved4[4];
641 
642 	uint8_t			mtmp_sensor_name[8];
643 };
644 CTASSERT(sizeof(struct mcx_reg_mtmp) == 0x20);
645 CTASSERT(offsetof(struct mcx_reg_mtmp, mtmp_sensor_name) == 0x18);
646 
647 #define MCX_MCIA_EEPROM_BYTES	32
648 struct mcx_reg_mcia {
649 	uint8_t			rm_l;
650 	uint8_t			rm_module;
651 	uint8_t			rm_reserved0;
652 	uint8_t			rm_status;
653 	uint8_t			rm_i2c_addr;
654 	uint8_t			rm_page_num;
655 	uint16_t		rm_dev_addr;
656 	uint16_t		rm_reserved1;
657 	uint16_t		rm_size;
658 	uint32_t		rm_reserved2;
659 	uint8_t			rm_data[48];
660 } __packed __aligned(4);
661 
662 struct mcx_cmd_query_issi_in {
663 	uint16_t		cmd_opcode;
664 	uint8_t			cmd_reserved0[4];
665 	uint16_t		cmd_op_mod;
666 	uint8_t			cmd_reserved1[8];
667 } __packed __aligned(4);
668 
669 struct mcx_cmd_query_issi_il_out {
670 	uint8_t			cmd_status;
671 	uint8_t			cmd_reserved0[3];
672 	uint32_t		cmd_syndrome;
673 	uint8_t			cmd_reserved1[2];
674 	uint16_t		cmd_current_issi;
675 	uint8_t			cmd_reserved2[4];
676 } __packed __aligned(4);
677 
678 CTASSERT(sizeof(struct mcx_cmd_query_issi_il_out) == MCX_CMDQ_INLINE_DATASIZE);
679 
680 struct mcx_cmd_query_issi_mb_out {
681 	uint8_t			cmd_reserved2[16];
682 	uint8_t			cmd_supported_issi[80]; /* very big endian */
683 } __packed __aligned(4);
684 
685 CTASSERT(sizeof(struct mcx_cmd_query_issi_mb_out) <= MCX_CMDQ_MAILBOX_DATASIZE);
686 
687 struct mcx_cmd_set_issi_in {
688 	uint16_t		cmd_opcode;
689 	uint8_t			cmd_reserved0[4];
690 	uint16_t		cmd_op_mod;
691 	uint8_t			cmd_reserved1[2];
692 	uint16_t		cmd_current_issi;
693 	uint8_t			cmd_reserved2[4];
694 } __packed __aligned(4);
695 
696 CTASSERT(sizeof(struct mcx_cmd_set_issi_in) <= MCX_CMDQ_INLINE_DATASIZE);
697 
698 struct mcx_cmd_set_issi_out {
699 	uint8_t			cmd_status;
700 	uint8_t			cmd_reserved0[3];
701 	uint32_t		cmd_syndrome;
702 	uint8_t			cmd_reserved1[8];
703 } __packed __aligned(4);
704 
705 CTASSERT(sizeof(struct mcx_cmd_set_issi_out) <= MCX_CMDQ_INLINE_DATASIZE);
706 
707 struct mcx_cmd_query_pages_in {
708 	uint16_t		cmd_opcode;
709 	uint8_t			cmd_reserved0[4];
710 	uint16_t		cmd_op_mod;
711 #define MCX_CMD_QUERY_PAGES_BOOT	0x01
712 #define MCX_CMD_QUERY_PAGES_INIT	0x02
713 #define MCX_CMD_QUERY_PAGES_REGULAR	0x03
714 	uint8_t			cmd_reserved1[8];
715 } __packed __aligned(4);
716 
717 struct mcx_cmd_query_pages_out {
718 	uint8_t			cmd_status;
719 	uint8_t			cmd_reserved0[3];
720 	uint32_t		cmd_syndrome;
721 	uint8_t			cmd_reserved1[2];
722 	uint16_t		cmd_func_id;
723 	int32_t			cmd_num_pages;
724 } __packed __aligned(4);
725 
726 struct mcx_cmd_manage_pages_in {
727 	uint16_t		cmd_opcode;
728 	uint8_t			cmd_reserved0[4];
729 	uint16_t		cmd_op_mod;
730 #define MCX_CMD_MANAGE_PAGES_ALLOC_FAIL \
731 					0x00
732 #define MCX_CMD_MANAGE_PAGES_ALLOC_SUCCESS \
733 					0x01
734 #define MCX_CMD_MANAGE_PAGES_HCA_RETURN_PAGES \
735 					0x02
736 	uint8_t			cmd_reserved1[2];
737 	uint16_t		cmd_func_id;
738 	uint32_t		cmd_input_num_entries;
739 } __packed __aligned(4);
740 
741 CTASSERT(sizeof(struct mcx_cmd_manage_pages_in) == MCX_CMDQ_INLINE_DATASIZE);
742 
743 struct mcx_cmd_manage_pages_out {
744 	uint8_t			cmd_status;
745 	uint8_t			cmd_reserved0[3];
746 	uint32_t		cmd_syndrome;
747 	uint32_t		cmd_output_num_entries;
748 	uint8_t			cmd_reserved1[4];
749 } __packed __aligned(4);
750 
751 CTASSERT(sizeof(struct mcx_cmd_manage_pages_out) == MCX_CMDQ_INLINE_DATASIZE);
752 
753 struct mcx_cmd_query_hca_cap_in {
754 	uint16_t		cmd_opcode;
755 	uint8_t			cmd_reserved0[4];
756 	uint16_t		cmd_op_mod;
757 #define MCX_CMD_QUERY_HCA_CAP_MAX	(0x0 << 0)
758 #define MCX_CMD_QUERY_HCA_CAP_CURRENT	(0x1 << 0)
759 #define MCX_CMD_QUERY_HCA_CAP_DEVICE	(0x0 << 1)
760 #define MCX_CMD_QUERY_HCA_CAP_OFFLOAD	(0x1 << 1)
761 #define MCX_CMD_QUERY_HCA_CAP_FLOW	(0x7 << 1)
762 	uint8_t			cmd_reserved1[8];
763 } __packed __aligned(4);
764 
765 struct mcx_cmd_query_hca_cap_out {
766 	uint8_t			cmd_status;
767 	uint8_t			cmd_reserved0[3];
768 	uint32_t		cmd_syndrome;
769 	uint8_t			cmd_reserved1[8];
770 } __packed __aligned(4);
771 
772 #define MCX_HCA_CAP_LEN			0x1000
773 #define MCX_HCA_CAP_NMAILBOXES		\
774 	(MCX_HCA_CAP_LEN / MCX_CMDQ_MAILBOX_DATASIZE)
775 
776 #if __GNUC_PREREQ__(4, 3)
777 #define __counter__		__COUNTER__
778 #else
779 #define __counter__		__LINE__
780 #endif
781 
782 #define __token(_tok, _num)	_tok##_num
783 #define _token(_tok, _num)	__token(_tok, _num)
784 #define __reserved__		_token(__reserved, __counter__)
785 
786 struct mcx_cap_device {
787 	uint8_t			reserved0[16];
788 
789 	uint8_t			log_max_srq_sz;
790 	uint8_t			log_max_qp_sz;
791 	uint8_t			__reserved__[1];
792 	uint8_t			log_max_qp; /* 5 bits */
793 #define MCX_CAP_DEVICE_LOG_MAX_QP	0x1f
794 
795 	uint8_t			__reserved__[1];
796 	uint8_t			log_max_srq; /* 5 bits */
797 #define MCX_CAP_DEVICE_LOG_MAX_SRQ	0x1f
798 	uint8_t			__reserved__[2];
799 
800 	uint8_t			__reserved__[1];
801 	uint8_t			log_max_cq_sz;
802 	uint8_t			__reserved__[1];
803 	uint8_t			log_max_cq; /* 5 bits */
804 #define MCX_CAP_DEVICE_LOG_MAX_CQ	0x1f
805 
806 	uint8_t			log_max_eq_sz;
807 	uint8_t			log_max_mkey; /* 6 bits */
808 #define MCX_CAP_DEVICE_LOG_MAX_MKEY	0x3f
809 	uint8_t			__reserved__[1];
810 	uint8_t			log_max_eq; /* 4 bits */
811 #define MCX_CAP_DEVICE_LOG_MAX_EQ	0x0f
812 
813 	uint8_t			max_indirection;
814 	uint8_t			log_max_mrw_sz; /* 7 bits */
815 #define MCX_CAP_DEVICE_LOG_MAX_MRW_SZ	0x7f
816 	uint8_t			teardown_log_max_msf_list_size;
817 #define MCX_CAP_DEVICE_FORCE_TEARDOWN	0x80
818 #define MCX_CAP_DEVICE_LOG_MAX_MSF_LIST_SIZE \
819 					0x3f
820 	uint8_t			log_max_klm_list_size; /* 6 bits */
821 #define MCX_CAP_DEVICE_LOG_MAX_KLM_LIST_SIZE \
822 					0x3f
823 
824 	uint8_t			__reserved__[1];
825 	uint8_t			log_max_ra_req_dc; /* 6 bits */
826 #define MCX_CAP_DEVICE_LOG_MAX_REQ_DC	0x3f
827 	uint8_t			__reserved__[1];
828 	uint8_t			log_max_ra_res_dc; /* 6 bits */
829 #define MCX_CAP_DEVICE_LOG_MAX_RA_RES_DC \
830 					0x3f
831 
832 	uint8_t			__reserved__[1];
833 	uint8_t			log_max_ra_req_qp; /* 6 bits */
834 #define MCX_CAP_DEVICE_LOG_MAX_RA_REQ_QP \
835 					0x3f
836 	uint8_t			__reserved__[1];
837 	uint8_t			log_max_ra_res_qp; /* 6 bits */
838 #define MCX_CAP_DEVICE_LOG_MAX_RA_RES_QP \
839 					0x3f
840 
841 	uint8_t			flags1;
842 #define MCX_CAP_DEVICE_END_PAD		0x80
843 #define MCX_CAP_DEVICE_CC_QUERY_ALLOWED	0x40
844 #define MCX_CAP_DEVICE_CC_MODIFY_ALLOWED \
845 					0x20
846 #define MCX_CAP_DEVICE_START_PAD	0x10
847 #define MCX_CAP_DEVICE_128BYTE_CACHELINE \
848 					0x08
849 	uint8_t			__reserved__[1];
850 	uint16_t		gid_table_size;
851 
852 	uint16_t		flags2;
853 #define MCX_CAP_DEVICE_OUT_OF_SEQ_CNT	0x8000
854 #define MCX_CAP_DEVICE_VPORT_COUNTERS	0x4000
855 #define MCX_CAP_DEVICE_RETRANSMISSION_Q_COUNTERS \
856 					0x2000
857 #define MCX_CAP_DEVICE_DEBUG		0x1000
858 #define MCX_CAP_DEVICE_MODIFY_RQ_COUNTERS_SET_ID \
859 					0x8000
860 #define MCX_CAP_DEVICE_RQ_DELAY_DROP	0x4000
861 #define MCX_CAP_DEVICe_MAX_QP_CNT_MASK	0x03ff
862 	uint16_t		pkey_table_size;
863 
864 	uint8_t			flags3;
865 #define MCX_CAP_DEVICE_VPORT_GROUP_MANAGER \
866 					0x80
867 #define MCX_CAP_DEVICE_VHCA_GROUP_MANAGER \
868 					0x40
869 #define MCX_CAP_DEVICE_IB_VIRTUAL	0x20
870 #define MCX_CAP_DEVICE_ETH_VIRTUAL	0x10
871 #define MCX_CAP_DEVICE_ETS		0x04
872 #define MCX_CAP_DEVICE_NIC_FLOW_TABLE	0x02
873 #define MCX_CAP_DEVICE_ESWITCH_FLOW_TABLE \
874 					0x01
875 	uint8_t			local_ca_ack_delay; /* 5 bits */
876 #define MCX_CAP_DEVICE_LOCAL_CA_ACK_DELAY \
877 					0x1f
878 #define MCX_CAP_DEVICE_MCAM_REG		0x40
879 	uint8_t			port_type;
880 #define MCX_CAP_DEVICE_PORT_MODULE_EVENT \
881 					0x80
882 #define MCX_CAP_DEVICE_PORT_TYPE	0x03
883 #define MCX_CAP_DEVICE_PORT_TYPE_ETH	0x01
884 	uint8_t			num_ports;
885 
886 	uint8_t			snapshot_log_max_msg;
887 #define MCX_CAP_DEVICE_SNAPSHOT		0x80
888 #define MCX_CAP_DEVICE_LOG_MAX_MSG	0x1f
889 	uint8_t			max_tc; /* 4 bits */
890 #define MCX_CAP_DEVICE_MAX_TC		0x0f
891 	uint8_t			flags4;
892 #define MCX_CAP_DEVICE_TEMP_WARN_EVENT	0x80
893 #define MCX_CAP_DEVICE_DCBX		0x40
894 #define MCX_CAP_DEVICE_ROL_S		0x02
895 #define MCX_CAP_DEVICE_ROL_G		0x01
896 	uint8_t			wol;
897 #define MCX_CAP_DEVICE_WOL_S		0x40
898 #define MCX_CAP_DEVICE_WOL_G		0x20
899 #define MCX_CAP_DEVICE_WOL_A		0x10
900 #define MCX_CAP_DEVICE_WOL_B		0x08
901 #define MCX_CAP_DEVICE_WOL_M		0x04
902 #define MCX_CAP_DEVICE_WOL_U		0x02
903 #define MCX_CAP_DEVICE_WOL_P		0x01
904 
905 	uint16_t		stat_rate_support;
906 	uint8_t			__reserved__[1];
907 	uint8_t			cqe_version; /* 4 bits */
908 #define MCX_CAP_DEVICE_CQE_VERSION	0x0f
909 
910 	uint32_t		flags5;
911 #define MCX_CAP_DEVICE_COMPACT_ADDRESS_VECTOR \
912 					0x80000000
913 #define MCX_CAP_DEVICE_STRIDING_RQ	0x40000000
914 #define MCX_CAP_DEVICE_IPOIP_ENHANCED_OFFLOADS \
915 					0x10000000
916 #define MCX_CAP_DEVICE_IPOIP_IPOIP_OFFLOADS \
917 					0x08000000
918 #define MCX_CAP_DEVICE_DC_CONNECT_CP	0x00040000
919 #define MCX_CAP_DEVICE_DC_CNAK_DRACE	0x00020000
920 #define MCX_CAP_DEVICE_DRAIN_SIGERR	0x00010000
921 #define MCX_CAP_DEVICE_DRAIN_SIGERR	0x00010000
922 #define MCX_CAP_DEVICE_CMDIF_CHECKSUM	0x0000c000
923 #define MCX_CAP_DEVICE_SIGERR_QCE	0x00002000
924 #define MCX_CAP_DEVICE_WQ_SIGNATURE	0x00000800
925 #define MCX_CAP_DEVICE_SCTR_DATA_CQE	0x00000400
926 #define MCX_CAP_DEVICE_SHO		0x00000100
927 #define MCX_CAP_DEVICE_TPH		0x00000080
928 #define MCX_CAP_DEVICE_RF		0x00000040
929 #define MCX_CAP_DEVICE_DCT		0x00000020
930 #define MCX_CAP_DEVICE_QOS		0x00000010
931 #define MCX_CAP_DEVICe_ETH_NET_OFFLOADS	0x00000008
932 #define MCX_CAP_DEVICE_ROCE		0x00000004
933 #define MCX_CAP_DEVICE_ATOMIC		0x00000002
934 
935 	uint32_t		flags6;
936 #define MCX_CAP_DEVICE_CQ_OI		0x80000000
937 #define MCX_CAP_DEVICE_CQ_RESIZE	0x40000000
938 #define MCX_CAP_DEVICE_CQ_MODERATION	0x20000000
939 #define MCX_CAP_DEVICE_CQ_PERIOD_MODE_MODIFY \
940 					0x10000000
941 #define MCX_CAP_DEVICE_CQ_INVALIDATE	0x08000000
942 #define MCX_CAP_DEVICE_RESERVED_AT_255	0x04000000
943 #define MCX_CAP_DEVICE_CQ_EQ_REMAP	0x02000000
944 #define MCX_CAP_DEVICE_PG		0x01000000
945 #define MCX_CAP_DEVICE_BLOCK_LB_MC	0x00800000
946 #define MCX_CAP_DEVICE_EXPONENTIAL_BACKOFF \
947 					0x00400000
948 #define MCX_CAP_DEVICE_SCQE_BREAK_MODERATION \
949 					0x00200000
950 #define MCX_CAP_DEVICE_CQ_PERIOD_START_FROM_CQE \
951 					0x00100000
952 #define MCX_CAP_DEVICE_CD		0x00080000
953 #define MCX_CAP_DEVICE_ATM		0x00040000
954 #define MCX_CAP_DEVICE_APM		0x00020000
955 #define MCX_CAP_DEVICE_IMAICL		0x00010000
956 #define MCX_CAP_DEVICE_QKV		0x00000200
957 #define MCX_CAP_DEVICE_PKV		0x00000100
958 #define MCX_CAP_DEVICE_SET_DETH_SQPN	0x00000080
959 #define MCX_CAP_DEVICE_XRC		0x00000008
960 #define MCX_CAP_DEVICE_UD		0x00000004
961 #define MCX_CAP_DEVICE_UC		0x00000002
962 #define MCX_CAP_DEVICE_RC		0x00000001
963 
964 	uint8_t			uar_flags;
965 #define MCX_CAP_DEVICE_UAR_4K		0x80
966 	uint8_t			uar_sz;	/* 6 bits */
967 #define MCX_CAP_DEVICE_UAR_SZ		0x3f
968 	uint8_t			__reserved__[1];
969 	uint8_t			log_pg_sz;
970 
971 	uint8_t			flags7;
972 #define MCX_CAP_DEVICE_BF		0x80
973 #define MCX_CAP_DEVICE_DRIVER_VERSION	0x40
974 #define MCX_CAP_DEVICE_PAD_TX_ETH_PACKET \
975 					0x20
976 	uint8_t			log_bf_reg_size; /* 5 bits */
977 #define MCX_CAP_DEVICE_LOG_BF_REG_SIZE	0x1f
978 	uint8_t			__reserved__[2];
979 
980 	uint16_t		num_of_diagnostic_counters;
981 	uint16_t		max_wqe_sz_sq;
982 
983 	uint8_t			__reserved__[2];
984 	uint16_t		max_wqe_sz_rq;
985 
986 	uint8_t			__reserved__[2];
987 	uint16_t		max_wqe_sz_sq_dc;
988 
989 	uint32_t		max_qp_mcg; /* 25 bits */
990 #define MCX_CAP_DEVICE_MAX_QP_MCG	0x1ffffff
991 
992 	uint8_t			__reserved__[3];
993 	uint8_t			log_max_mcq;
994 
995 	uint8_t			log_max_transport_domain; /* 5 bits */
996 #define MCX_CAP_DEVICE_LOG_MAX_TRANSORT_DOMAIN \
997 					0x1f
998 	uint8_t			log_max_pd; /* 5 bits */
999 #define MCX_CAP_DEVICE_LOG_MAX_PD	0x1f
1000 	uint8_t			__reserved__[1];
1001 	uint8_t			log_max_xrcd; /* 5 bits */
1002 #define MCX_CAP_DEVICE_LOG_MAX_XRCD	0x1f
1003 
1004 	uint8_t			__reserved__[2];
1005 	uint16_t		max_flow_counter;
1006 
1007 	uint8_t			log_max_rq; /* 5 bits */
1008 #define MCX_CAP_DEVICE_LOG_MAX_RQ	0x1f
1009 	uint8_t			log_max_sq; /* 5 bits */
1010 #define MCX_CAP_DEVICE_LOG_MAX_SQ	0x1f
1011 	uint8_t			log_max_tir; /* 5 bits */
1012 #define MCX_CAP_DEVICE_LOG_MAX_TIR	0x1f
1013 	uint8_t			log_max_tis; /* 5 bits */
1014 #define MCX_CAP_DEVICE_LOG_MAX_TIS	0x1f
1015 
1016 	uint8_t 		flags8;
1017 #define MCX_CAP_DEVICE_BASIC_CYCLIC_RCV_WQE \
1018 					0x80
1019 #define MCX_CAP_DEVICE_LOG_MAX_RMP	0x1f
1020 	uint8_t			log_max_rqt; /* 5 bits */
1021 #define MCX_CAP_DEVICE_LOG_MAX_RQT	0x1f
1022 	uint8_t			log_max_rqt_size; /* 5 bits */
1023 #define MCX_CAP_DEVICE_LOG_MAX_RQT_SIZE	0x1f
1024 	uint8_t			log_max_tis_per_sq; /* 5 bits */
1025 #define MCX_CAP_DEVICE_LOG_MAX_TIS_PER_SQ \
1026 					0x1f
1027 
1028 	uint8_t			flags9;
1029 #define MXC_CAP_DEVICE_EXT_STRIDE_NUM_RANGES \
1030 					0x80
1031 #define MXC_CAP_DEVICE_LOG_MAX_STRIDE_SZ_RQ \
1032 					0x1f
1033 	uint8_t			log_min_stride_sz_rq; /* 5 bits */
1034 #define MXC_CAP_DEVICE_LOG_MIN_STRIDE_SZ_RQ \
1035 					0x1f
1036 	uint8_t			log_max_stride_sz_sq; /* 5 bits */
1037 #define MXC_CAP_DEVICE_LOG_MAX_STRIDE_SZ_SQ \
1038 					0x1f
1039 	uint8_t			log_min_stride_sz_sq; /* 5 bits */
1040 #define MXC_CAP_DEVICE_LOG_MIN_STRIDE_SZ_SQ \
1041 					0x1f
1042 
1043 	uint8_t			log_max_hairpin_queues;
1044 #define MXC_CAP_DEVICE_HAIRPIN		0x80
1045 #define MXC_CAP_DEVICE_LOG_MAX_HAIRPIN_QUEUES \
1046 					0x1f
1047 	uint8_t			log_min_hairpin_queues;
1048 #define MXC_CAP_DEVICE_LOG_MIN_HAIRPIN_QUEUES \
1049 					0x1f
1050 	uint8_t			log_max_hairpin_num_packets;
1051 #define MXC_CAP_DEVICE_LOG_MAX_HAIRPIN_NUM_PACKETS \
1052 					0x1f
1053 	uint8_t			log_max_mq_sz;
1054 #define MXC_CAP_DEVICE_LOG_MAX_WQ_SZ \
1055 					0x1f
1056 
1057 	uint8_t			log_min_hairpin_wq_data_sz;
1058 #define MXC_CAP_DEVICE_NIC_VPORT_CHANGE_EVENT \
1059 					0x80
1060 #define MXC_CAP_DEVICE_DISABLE_LOCAL_LB_UC \
1061 					0x40
1062 #define MXC_CAP_DEVICE_DISABLE_LOCAL_LB_MC \
1063 					0x20
1064 #define MCX_CAP_DEVICE_LOG_MIN_HAIRPIN_WQ_DATA_SZ \
1065 					0x1f
1066 	uint8_t			log_max_vlan_list;
1067 #define MXC_CAP_DEVICE_SYSTEM_IMAGE_GUID_MODIFIABLE \
1068 					0x80
1069 #define MXC_CAP_DEVICE_LOG_MAX_VLAN_LIST \
1070 					0x1f
1071 	uint8_t			log_max_current_mc_list;
1072 #define MXC_CAP_DEVICE_LOG_MAX_CURRENT_MC_LIST \
1073 					0x1f
1074 	uint8_t			log_max_current_uc_list;
1075 #define MXC_CAP_DEVICE_LOG_MAX_CURRENT_UC_LIST \
1076 					0x1f
1077 
1078 	uint8_t			__reserved__[4];
1079 
1080 	uint32_t		create_qp_start_hint; /* 24 bits */
1081 
1082 	uint8_t			log_max_uctx; /* 5 bits */
1083 #define MXC_CAP_DEVICE_LOG_MAX_UCTX	0x1f
1084 	uint8_t			log_max_umem; /* 5 bits */
1085 #define MXC_CAP_DEVICE_LOG_MAX_UMEM	0x1f
1086 	uint16_t		max_num_eqs;
1087 
1088 	uint8_t			log_max_l2_table; /* 5 bits */
1089 #define MXC_CAP_DEVICE_LOG_MAX_L2_TABLE	0x1f
1090 	uint8_t			__reserved__[1];
1091 	uint16_t		log_uar_page_sz;
1092 
1093 	uint8_t			__reserved__[8];
1094 
1095 	uint32_t		device_frequency_mhz;
1096 	uint32_t		device_frequency_khz;
1097 } __packed __aligned(8);
1098 
1099 CTASSERT(offsetof(struct mcx_cap_device, max_indirection) == 0x20);
1100 CTASSERT(offsetof(struct mcx_cap_device, flags1) == 0x2c);
1101 CTASSERT(offsetof(struct mcx_cap_device, flags2) == 0x30);
1102 CTASSERT(offsetof(struct mcx_cap_device, snapshot_log_max_msg) == 0x38);
1103 CTASSERT(offsetof(struct mcx_cap_device, flags5) == 0x40);
1104 CTASSERT(offsetof(struct mcx_cap_device, flags7) == 0x4c);
1105 CTASSERT(offsetof(struct mcx_cap_device, device_frequency_mhz) == 0x98);
1106 CTASSERT(offsetof(struct mcx_cap_device, device_frequency_khz) == 0x9c);
1107 CTASSERT(sizeof(struct mcx_cap_device) <= MCX_CMDQ_MAILBOX_DATASIZE);
1108 
1109 struct mcx_cmd_set_driver_version_in {
1110 	uint16_t		cmd_opcode;
1111 	uint8_t			cmd_reserved0[4];
1112 	uint16_t		cmd_op_mod;
1113 	uint8_t			cmd_reserved1[8];
1114 } __packed __aligned(4);
1115 
1116 struct mcx_cmd_set_driver_version_out {
1117 	uint8_t			cmd_status;
1118 	uint8_t			cmd_reserved0[3];
1119 	uint32_t		cmd_syndrome;
1120 	uint8_t			cmd_reserved1[8];
1121 } __packed __aligned(4);
1122 
1123 struct mcx_cmd_set_driver_version {
1124 	uint8_t			cmd_driver_version[64];
1125 } __packed __aligned(8);
1126 
1127 struct mcx_cmd_modify_nic_vport_context_in {
1128 	uint16_t		cmd_opcode;
1129 	uint8_t			cmd_reserved0[4];
1130 	uint16_t		cmd_op_mod;
1131 	uint8_t			cmd_reserved1[4];
1132 	uint32_t		cmd_field_select;
1133 #define MCX_CMD_MODIFY_NIC_VPORT_CONTEXT_FIELD_ADDR	0x04
1134 #define MCX_CMD_MODIFY_NIC_VPORT_CONTEXT_FIELD_PROMISC	0x10
1135 #define MCX_CMD_MODIFY_NIC_VPORT_CONTEXT_FIELD_MTU	0x40
1136 } __packed __aligned(4);
1137 
1138 struct mcx_cmd_modify_nic_vport_context_out {
1139 	uint8_t			cmd_status;
1140 	uint8_t			cmd_reserved0[3];
1141 	uint32_t		cmd_syndrome;
1142 	uint8_t			cmd_reserved1[8];
1143 } __packed __aligned(4);
1144 
1145 struct mcx_cmd_query_nic_vport_context_in {
1146 	uint16_t		cmd_opcode;
1147 	uint8_t			cmd_reserved0[4];
1148 	uint16_t		cmd_op_mod;
1149 	uint8_t			cmd_reserved1[4];
1150 	uint8_t			cmd_allowed_list_type;
1151 	uint8_t			cmd_reserved2[3];
1152 } __packed __aligned(4);
1153 
1154 struct mcx_cmd_query_nic_vport_context_out {
1155 	uint8_t			cmd_status;
1156 	uint8_t			cmd_reserved0[3];
1157 	uint32_t		cmd_syndrome;
1158 	uint8_t			cmd_reserved1[8];
1159 } __packed __aligned(4);
1160 
1161 struct mcx_nic_vport_ctx {
1162 	uint32_t		vp_min_wqe_inline_mode;
1163 	uint8_t			vp_reserved0[32];
1164 	uint32_t		vp_mtu;
1165 	uint8_t			vp_reserved1[200];
1166 	uint16_t		vp_flags;
1167 #define MCX_NIC_VPORT_CTX_LIST_UC_MAC			(0)
1168 #define MCX_NIC_VPORT_CTX_LIST_MC_MAC			(1 << 24)
1169 #define MCX_NIC_VPORT_CTX_LIST_VLAN			(2 << 24)
1170 #define MCX_NIC_VPORT_CTX_PROMISC_ALL			(1 << 13)
1171 #define MCX_NIC_VPORT_CTX_PROMISC_MCAST			(1 << 14)
1172 #define MCX_NIC_VPORT_CTX_PROMISC_UCAST			(1 << 15)
1173 	uint16_t		vp_allowed_list_size;
1174 	uint64_t		vp_perm_addr;
1175 	uint8_t			vp_reserved2[4];
1176 	/* allowed list follows */
1177 } __packed __aligned(4);
1178 
1179 struct mcx_counter {
1180 	uint64_t		packets;
1181 	uint64_t		octets;
1182 } __packed __aligned(4);
1183 
1184 struct mcx_nic_vport_counters {
1185 	struct mcx_counter	rx_err;
1186 	struct mcx_counter	tx_err;
1187 	uint8_t			reserved0[64]; /* 0x30 */
1188 	struct mcx_counter	rx_bcast;
1189 	struct mcx_counter	tx_bcast;
1190 	struct mcx_counter	rx_ucast;
1191 	struct mcx_counter	tx_ucast;
1192 	struct mcx_counter	rx_mcast;
1193 	struct mcx_counter	tx_mcast;
1194 	uint8_t			reserved1[0x210 - 0xd0];
1195 } __packed __aligned(4);
1196 
1197 struct mcx_cmd_query_vport_counters_in {
1198 	uint16_t		cmd_opcode;
1199 	uint8_t			cmd_reserved0[4];
1200 	uint16_t		cmd_op_mod;
1201 	uint8_t			cmd_reserved1[8];
1202 } __packed __aligned(4);
1203 
1204 struct mcx_cmd_query_vport_counters_mb_in {
1205 	uint8_t			cmd_reserved0[8];
1206 	uint8_t			cmd_clear;
1207 	uint8_t			cmd_reserved1[7];
1208 } __packed __aligned(4);
1209 
1210 struct mcx_cmd_query_vport_counters_out {
1211 	uint8_t			cmd_status;
1212 	uint8_t			cmd_reserved0[3];
1213 	uint32_t		cmd_syndrome;
1214 	uint8_t			cmd_reserved1[8];
1215 } __packed __aligned(4);
1216 
1217 struct mcx_cmd_query_flow_counter_in {
1218 	uint16_t		cmd_opcode;
1219 	uint8_t			cmd_reserved0[4];
1220 	uint16_t		cmd_op_mod;
1221 	uint8_t			cmd_reserved1[8];
1222 } __packed __aligned(4);
1223 
1224 struct mcx_cmd_query_flow_counter_mb_in {
1225 	uint8_t			cmd_reserved0[8];
1226 	uint8_t			cmd_clear;
1227 	uint8_t			cmd_reserved1[5];
1228 	uint16_t		cmd_flow_counter_id;
1229 } __packed __aligned(4);
1230 
1231 struct mcx_cmd_query_flow_counter_out {
1232 	uint8_t			cmd_status;
1233 	uint8_t			cmd_reserved0[3];
1234 	uint32_t		cmd_syndrome;
1235 	uint8_t			cmd_reserved1[8];
1236 } __packed __aligned(4);
1237 
1238 struct mcx_cmd_alloc_uar_in {
1239 	uint16_t		cmd_opcode;
1240 	uint8_t			cmd_reserved0[4];
1241 	uint16_t		cmd_op_mod;
1242 	uint8_t			cmd_reserved1[8];
1243 } __packed __aligned(4);
1244 
1245 struct mcx_cmd_alloc_uar_out {
1246 	uint8_t			cmd_status;
1247 	uint8_t			cmd_reserved0[3];
1248 	uint32_t		cmd_syndrome;
1249 	uint32_t		cmd_uar;
1250 	uint8_t			cmd_reserved1[4];
1251 } __packed __aligned(4);
1252 
1253 struct mcx_cmd_query_special_ctx_in {
1254 	uint16_t		cmd_opcode;
1255 	uint8_t			cmd_reserved0[4];
1256 	uint16_t		cmd_op_mod;
1257 	uint8_t			cmd_reserved1[8];
1258 } __packed __aligned(4);
1259 
1260 struct mcx_cmd_query_special_ctx_out {
1261 	uint8_t			cmd_status;
1262 	uint8_t			cmd_reserved0[3];
1263 	uint32_t		cmd_syndrome;
1264 	uint8_t			cmd_reserved1[4];
1265 	uint32_t		cmd_resd_lkey;
1266 } __packed __aligned(4);
1267 
1268 struct mcx_eq_ctx {
1269 	uint32_t		eq_status;
1270 #define MCX_EQ_CTX_STATE_SHIFT		8
1271 #define MCX_EQ_CTX_STATE_MASK		(0xf << MCX_EQ_CTX_STATE_SHIFT)
1272 #define MCX_EQ_CTX_STATE_ARMED		0x9
1273 #define MCX_EQ_CTX_STATE_FIRED		0xa
1274 #define MCX_EQ_CTX_OI_SHIFT		17
1275 #define MCX_EQ_CTX_OI			(1 << MCX_EQ_CTX_OI_SHIFT)
1276 #define MCX_EQ_CTX_EC_SHIFT		18
1277 #define MCX_EQ_CTX_EC			(1 << MCX_EQ_CTX_EC_SHIFT)
1278 #define MCX_EQ_CTX_STATUS_SHIFT		28
1279 #define MCX_EQ_CTX_STATUS_MASK		(0xf << MCX_EQ_CTX_STATUS_SHIFT)
1280 #define MCX_EQ_CTX_STATUS_OK		0x0
1281 #define MCX_EQ_CTX_STATUS_EQ_WRITE_FAILURE 0xa
1282 	uint32_t		eq_reserved1;
1283 	uint32_t		eq_page_offset;
1284 #define MCX_EQ_CTX_PAGE_OFFSET_SHIFT	5
1285 	uint32_t		eq_uar_size;
1286 #define MCX_EQ_CTX_UAR_PAGE_MASK	0xffffff
1287 #define MCX_EQ_CTX_LOG_EQ_SIZE_SHIFT	24
1288 	uint32_t		eq_reserved2;
1289 	uint8_t			eq_reserved3[3];
1290 	uint8_t			eq_intr;
1291 	uint32_t		eq_log_page_size;
1292 #define MCX_EQ_CTX_LOG_PAGE_SIZE_SHIFT	24
1293 	uint32_t		eq_reserved4[3];
1294 	uint32_t		eq_consumer_counter;
1295 	uint32_t		eq_producer_counter;
1296 #define MCX_EQ_CTX_COUNTER_MASK		0xffffff
1297 	uint32_t		eq_reserved5[4];
1298 } __packed __aligned(4);
1299 
1300 CTASSERT(sizeof(struct mcx_eq_ctx) == 64);
1301 
1302 struct mcx_cmd_create_eq_in {
1303 	uint16_t		cmd_opcode;
1304 	uint8_t			cmd_reserved0[4];
1305 	uint16_t		cmd_op_mod;
1306 	uint8_t			cmd_reserved1[8];
1307 } __packed __aligned(4);
1308 
1309 struct mcx_cmd_create_eq_mb_in {
1310 	struct mcx_eq_ctx	cmd_eq_ctx;
1311 	uint8_t			cmd_reserved0[8];
1312 	uint64_t		cmd_event_bitmask;
1313 #define MCX_EVENT_TYPE_COMPLETION	0x00
1314 #define MCX_EVENT_TYPE_CQ_ERROR		0x04
1315 #define MCX_EVENT_TYPE_INTERNAL_ERROR	0x08
1316 #define MCX_EVENT_TYPE_PORT_CHANGE	0x09
1317 #define MCX_EVENT_TYPE_CMD_COMPLETION	0x0a
1318 #define MCX_EVENT_TYPE_PAGE_REQUEST	0x0b
1319 #define MCX_EVENT_TYPE_LAST_WQE		0x13
1320 	uint8_t			cmd_reserved1[176];
1321 } __packed __aligned(4);
1322 
1323 struct mcx_cmd_create_eq_out {
1324 	uint8_t			cmd_status;
1325 	uint8_t			cmd_reserved0[3];
1326 	uint32_t		cmd_syndrome;
1327 	uint32_t		cmd_eqn;
1328 	uint8_t			cmd_reserved1[4];
1329 } __packed __aligned(4);
1330 
1331 struct mcx_cmd_query_eq_in {
1332 	uint16_t		cmd_opcode;
1333 	uint8_t			cmd_reserved0[4];
1334 	uint16_t		cmd_op_mod;
1335 	uint32_t		cmd_eqn;
1336 	uint8_t			cmd_reserved1[4];
1337 } __packed __aligned(4);
1338 
1339 struct mcx_cmd_query_eq_out {
1340 	uint8_t			cmd_status;
1341 	uint8_t			cmd_reserved0[3];
1342 	uint32_t		cmd_syndrome;
1343 	uint8_t			cmd_reserved1[8];
1344 } __packed __aligned(4);
1345 
1346 struct mcx_eq_entry {
1347 	uint8_t			eq_reserved1;
1348 	uint8_t			eq_event_type;
1349 	uint8_t			eq_reserved2;
1350 	uint8_t			eq_event_sub_type;
1351 
1352 	uint8_t			eq_reserved3[28];
1353 	uint32_t		eq_event_data[7];
1354 	uint8_t			eq_reserved4[2];
1355 	uint8_t			eq_signature;
1356 	uint8_t			eq_owner;
1357 #define MCX_EQ_ENTRY_OWNER_INIT			1
1358 } __packed __aligned(4);
1359 
1360 CTASSERT(sizeof(struct mcx_eq_entry) == 64);
1361 
1362 struct mcx_cmd_alloc_pd_in {
1363 	uint16_t		cmd_opcode;
1364 	uint8_t			cmd_reserved0[4];
1365 	uint16_t		cmd_op_mod;
1366 	uint8_t			cmd_reserved1[8];
1367 } __packed __aligned(4);
1368 
1369 struct mcx_cmd_alloc_pd_out {
1370 	uint8_t			cmd_status;
1371 	uint8_t			cmd_reserved0[3];
1372 	uint32_t		cmd_syndrome;
1373 	uint32_t		cmd_pd;
1374 	uint8_t			cmd_reserved1[4];
1375 } __packed __aligned(4);
1376 
1377 struct mcx_cmd_alloc_td_in {
1378 	uint16_t		cmd_opcode;
1379 	uint8_t			cmd_reserved0[4];
1380 	uint16_t		cmd_op_mod;
1381 	uint8_t			cmd_reserved1[8];
1382 } __packed __aligned(4);
1383 
1384 struct mcx_cmd_alloc_td_out {
1385 	uint8_t			cmd_status;
1386 	uint8_t			cmd_reserved0[3];
1387 	uint32_t		cmd_syndrome;
1388 	uint32_t		cmd_tdomain;
1389 	uint8_t			cmd_reserved1[4];
1390 } __packed __aligned(4);
1391 
1392 struct mcx_cmd_create_tir_in {
1393 	uint16_t		cmd_opcode;
1394 	uint8_t			cmd_reserved0[4];
1395 	uint16_t		cmd_op_mod;
1396 	uint8_t			cmd_reserved1[8];
1397 } __packed __aligned(4);
1398 
1399 struct mcx_cmd_create_tir_mb_in {
1400 	uint8_t			cmd_reserved0[20];
1401 	uint32_t		cmd_disp_type;
1402 #define MCX_TIR_CTX_DISP_TYPE_DIRECT	0
1403 #define MCX_TIR_CTX_DISP_TYPE_INDIRECT	1
1404 #define MCX_TIR_CTX_DISP_TYPE_SHIFT	28
1405 	uint8_t			cmd_reserved1[8];
1406 	uint32_t		cmd_lro;
1407 	uint8_t			cmd_reserved2[8];
1408 	uint32_t		cmd_inline_rqn;
1409 	uint32_t		cmd_indir_table;
1410 	uint32_t		cmd_tdomain;
1411 #define MCX_TIR_CTX_HASH_TOEPLITZ	2
1412 #define MCX_TIR_CTX_HASH_SHIFT		28
1413 	uint8_t			cmd_rx_hash_key[40];
1414 	uint32_t		cmd_rx_hash_sel_outer;
1415 #define MCX_TIR_CTX_HASH_SEL_SRC_IP	(1 << 0)
1416 #define MCX_TIR_CTX_HASH_SEL_DST_IP	(1 << 1)
1417 #define MCX_TIR_CTX_HASH_SEL_SPORT	(1 << 2)
1418 #define MCX_TIR_CTX_HASH_SEL_DPORT	(1 << 3)
1419 #define MCX_TIR_CTX_HASH_SEL_IPV4	(0 << 31)
1420 #define MCX_TIR_CTX_HASH_SEL_IPV6	(1 << 31)
1421 #define MCX_TIR_CTX_HASH_SEL_TCP	(0 << 30)
1422 #define MCX_TIR_CTX_HASH_SEL_UDP	(1 << 30)
1423 	uint32_t		cmd_rx_hash_sel_inner;
1424 	uint8_t			cmd_reserved3[152];
1425 } __packed __aligned(4);
1426 
1427 struct mcx_cmd_create_tir_out {
1428 	uint8_t			cmd_status;
1429 	uint8_t			cmd_reserved0[3];
1430 	uint32_t		cmd_syndrome;
1431 	uint32_t		cmd_tirn;
1432 	uint8_t			cmd_reserved1[4];
1433 } __packed __aligned(4);
1434 
1435 struct mcx_cmd_destroy_tir_in {
1436 	uint16_t		cmd_opcode;
1437 	uint8_t			cmd_reserved0[4];
1438 	uint16_t		cmd_op_mod;
1439 	uint32_t		cmd_tirn;
1440 	uint8_t			cmd_reserved1[4];
1441 } __packed __aligned(4);
1442 
1443 struct mcx_cmd_destroy_tir_out {
1444 	uint8_t			cmd_status;
1445 	uint8_t			cmd_reserved0[3];
1446 	uint32_t		cmd_syndrome;
1447 	uint8_t			cmd_reserved1[8];
1448 } __packed __aligned(4);
1449 
1450 struct mcx_cmd_create_tis_in {
1451 	uint16_t		cmd_opcode;
1452 	uint8_t			cmd_reserved0[4];
1453 	uint16_t		cmd_op_mod;
1454 	uint8_t			cmd_reserved1[8];
1455 } __packed __aligned(4);
1456 
1457 struct mcx_cmd_create_tis_mb_in {
1458 	uint8_t			cmd_reserved[16];
1459 	uint32_t		cmd_prio;
1460 	uint8_t			cmd_reserved1[32];
1461 	uint32_t		cmd_tdomain;
1462 	uint8_t			cmd_reserved2[120];
1463 } __packed __aligned(4);
1464 
1465 struct mcx_cmd_create_tis_out {
1466 	uint8_t			cmd_status;
1467 	uint8_t			cmd_reserved0[3];
1468 	uint32_t		cmd_syndrome;
1469 	uint32_t		cmd_tisn;
1470 	uint8_t			cmd_reserved1[4];
1471 } __packed __aligned(4);
1472 
1473 struct mcx_cmd_destroy_tis_in {
1474 	uint16_t		cmd_opcode;
1475 	uint8_t			cmd_reserved0[4];
1476 	uint16_t		cmd_op_mod;
1477 	uint32_t		cmd_tisn;
1478 	uint8_t			cmd_reserved1[4];
1479 } __packed __aligned(4);
1480 
1481 struct mcx_cmd_destroy_tis_out {
1482 	uint8_t			cmd_status;
1483 	uint8_t			cmd_reserved0[3];
1484 	uint32_t		cmd_syndrome;
1485 	uint8_t			cmd_reserved1[8];
1486 } __packed __aligned(4);
1487 
1488 struct mcx_cmd_create_rqt_in {
1489 	uint16_t		cmd_opcode;
1490 	uint8_t			cmd_reserved0[4];
1491 	uint16_t		cmd_op_mod;
1492 	uint8_t			cmd_reserved1[8];
1493 } __packed __aligned(4);
1494 
1495 struct mcx_rqt_ctx {
1496 	uint8_t			cmd_reserved0[20];
1497 	uint16_t		cmd_reserved1;
1498 	uint16_t		cmd_rqt_max_size;
1499 	uint16_t		cmd_reserved2;
1500 	uint16_t		cmd_rqt_actual_size;
1501 	uint8_t			cmd_reserved3[212];
1502 } __packed __aligned(4);
1503 
1504 struct mcx_cmd_create_rqt_mb_in {
1505 	uint8_t			cmd_reserved0[16];
1506 	struct mcx_rqt_ctx	cmd_rqt;
1507 } __packed __aligned(4);
1508 
1509 struct mcx_cmd_create_rqt_out {
1510 	uint8_t			cmd_status;
1511 	uint8_t			cmd_reserved0[3];
1512 	uint32_t		cmd_syndrome;
1513 	uint32_t		cmd_rqtn;
1514 	uint8_t			cmd_reserved1[4];
1515 } __packed __aligned(4);
1516 
1517 struct mcx_cmd_destroy_rqt_in {
1518 	uint16_t		cmd_opcode;
1519 	uint8_t			cmd_reserved0[4];
1520 	uint16_t		cmd_op_mod;
1521 	uint32_t		cmd_rqtn;
1522 	uint8_t			cmd_reserved1[4];
1523 } __packed __aligned(4);
1524 
1525 struct mcx_cmd_destroy_rqt_out {
1526 	uint8_t			cmd_status;
1527 	uint8_t			cmd_reserved0[3];
1528 	uint32_t		cmd_syndrome;
1529 	uint8_t			cmd_reserved1[8];
1530 } __packed __aligned(4);
1531 
1532 struct mcx_cq_ctx {
1533 	uint32_t		cq_status;
1534 #define MCX_CQ_CTX_STATUS_SHIFT		28
1535 #define MCX_CQ_CTX_STATUS_MASK		(0xf << MCX_CQ_CTX_STATUS_SHIFT)
1536 #define MCX_CQ_CTX_STATUS_OK		0x0
1537 #define MCX_CQ_CTX_STATUS_OVERFLOW	0x9
1538 #define MCX_CQ_CTX_STATUS_WRITE_FAIL	0xa
1539 #define MCX_CQ_CTX_STATE_SHIFT		8
1540 #define MCX_CQ_CTX_STATE_MASK		(0xf << MCX_CQ_CTX_STATE_SHIFT)
1541 #define MCX_CQ_CTX_STATE_SOLICITED	0x6
1542 #define MCX_CQ_CTX_STATE_ARMED		0x9
1543 #define MCX_CQ_CTX_STATE_FIRED		0xa
1544 	uint32_t		cq_reserved1;
1545 	uint32_t		cq_page_offset;
1546 	uint32_t		cq_uar_size;
1547 #define MCX_CQ_CTX_UAR_PAGE_MASK	0xffffff
1548 #define MCX_CQ_CTX_LOG_CQ_SIZE_SHIFT	24
1549 	uint32_t		cq_period_max_count;
1550 #define MCX_CQ_CTX_PERIOD_SHIFT		16
1551 	uint32_t		cq_eqn;
1552 	uint32_t		cq_log_page_size;
1553 #define MCX_CQ_CTX_LOG_PAGE_SIZE_SHIFT	24
1554 	uint32_t		cq_reserved2;
1555 	uint32_t		cq_last_notified;
1556 	uint32_t		cq_last_solicit;
1557 	uint32_t		cq_consumer_counter;
1558 	uint32_t		cq_producer_counter;
1559 	uint8_t			cq_reserved3[8];
1560 	uint64_t		cq_doorbell;
1561 } __packed __aligned(4);
1562 
1563 CTASSERT(sizeof(struct mcx_cq_ctx) == 64);
1564 
1565 struct mcx_cmd_create_cq_in {
1566 	uint16_t		cmd_opcode;
1567 	uint8_t			cmd_reserved0[4];
1568 	uint16_t		cmd_op_mod;
1569 	uint8_t			cmd_reserved1[8];
1570 } __packed __aligned(4);
1571 
1572 struct mcx_cmd_create_cq_mb_in {
1573 	struct mcx_cq_ctx	cmd_cq_ctx;
1574 	uint8_t			cmd_reserved1[192];
1575 } __packed __aligned(4);
1576 
1577 struct mcx_cmd_create_cq_out {
1578 	uint8_t			cmd_status;
1579 	uint8_t			cmd_reserved0[3];
1580 	uint32_t		cmd_syndrome;
1581 	uint32_t		cmd_cqn;
1582 	uint8_t			cmd_reserved1[4];
1583 } __packed __aligned(4);
1584 
1585 struct mcx_cmd_destroy_cq_in {
1586 	uint16_t		cmd_opcode;
1587 	uint8_t			cmd_reserved0[4];
1588 	uint16_t		cmd_op_mod;
1589 	uint32_t		cmd_cqn;
1590 	uint8_t			cmd_reserved1[4];
1591 } __packed __aligned(4);
1592 
1593 struct mcx_cmd_destroy_cq_out {
1594 	uint8_t			cmd_status;
1595 	uint8_t			cmd_reserved0[3];
1596 	uint32_t		cmd_syndrome;
1597 	uint8_t			cmd_reserved1[8];
1598 } __packed __aligned(4);
1599 
1600 struct mcx_cmd_query_cq_in {
1601 	uint16_t		cmd_opcode;
1602 	uint8_t			cmd_reserved0[4];
1603 	uint16_t		cmd_op_mod;
1604 	uint32_t		cmd_cqn;
1605 	uint8_t			cmd_reserved1[4];
1606 } __packed __aligned(4);
1607 
1608 struct mcx_cmd_query_cq_out {
1609 	uint8_t			cmd_status;
1610 	uint8_t			cmd_reserved0[3];
1611 	uint32_t		cmd_syndrome;
1612 	uint8_t			cmd_reserved1[8];
1613 } __packed __aligned(4);
1614 
1615 struct mcx_cq_entry {
1616 	uint32_t		__reserved__;
1617 	uint32_t		cq_lro;
1618 	uint32_t		cq_lro_ack_seq_num;
1619 	uint32_t		cq_rx_hash;
1620 	uint8_t			cq_rx_hash_type;
1621 	uint8_t			cq_ml_path;
1622 	uint16_t		__reserved__;
1623 	uint32_t		cq_checksum;
1624 	uint32_t		__reserved__;
1625 	uint32_t		cq_flags;
1626 #define MCX_CQ_ENTRY_FLAGS_L4_OK		(1 << 26)
1627 #define MCX_CQ_ENTRY_FLAGS_L3_OK		(1 << 25)
1628 #define MCX_CQ_ENTRY_FLAGS_L2_OK		(1 << 24)
1629 #define MCX_CQ_ENTRY_FLAGS_CV			(1 << 16)
1630 #define MCX_CQ_ENTRY_FLAGS_VLAN_MASK		(0xffff)
1631 
1632 	uint32_t		cq_lro_srqn;
1633 	uint32_t		__reserved__[2];
1634 	uint32_t		cq_byte_cnt;
1635 	uint64_t		cq_timestamp;
1636 	uint8_t			cq_rx_drops;
1637 	uint8_t			cq_flow_tag[3];
1638 	uint16_t		cq_wqe_count;
1639 	uint8_t			cq_signature;
1640 	uint8_t			cq_opcode_owner;
1641 #define MCX_CQ_ENTRY_FLAG_OWNER			(1 << 0)
1642 #define MCX_CQ_ENTRY_FLAG_SE			(1 << 1)
1643 #define MCX_CQ_ENTRY_FORMAT_SHIFT		2
1644 #define MCX_CQ_ENTRY_OPCODE_SHIFT		4
1645 
1646 #define MCX_CQ_ENTRY_FORMAT_NO_INLINE		0
1647 #define MCX_CQ_ENTRY_FORMAT_INLINE_32		1
1648 #define MCX_CQ_ENTRY_FORMAT_INLINE_64		2
1649 #define MCX_CQ_ENTRY_FORMAT_COMPRESSED		3
1650 
1651 #define MCX_CQ_ENTRY_OPCODE_REQ			0
1652 #define MCX_CQ_ENTRY_OPCODE_SEND		2
1653 #define MCX_CQ_ENTRY_OPCODE_REQ_ERR		13
1654 #define MCX_CQ_ENTRY_OPCODE_SEND_ERR		14
1655 #define MCX_CQ_ENTRY_OPCODE_INVALID		15
1656 
1657 } __packed __aligned(4);
1658 
1659 CTASSERT(sizeof(struct mcx_cq_entry) == 64);
1660 
1661 struct mcx_cq_doorbell {
1662 	uint32_t		 db_update_ci;
1663 	uint32_t		 db_arm_ci;
1664 #define MCX_CQ_DOORBELL_ARM_CMD_SN_SHIFT	28
1665 #define MCX_CQ_DOORBELL_ARM_CMD			(1 << 24)
1666 #define MCX_CQ_DOORBELL_ARM_CI_MASK		(0xffffff)
1667 } __packed __aligned(8);
1668 
1669 struct mcx_wq_ctx {
1670 	uint8_t			 wq_type;
1671 #define MCX_WQ_CTX_TYPE_CYCLIC			(1 << 4)
1672 #define MCX_WQ_CTX_TYPE_SIGNATURE		(1 << 3)
1673 	uint8_t			 wq_reserved0[5];
1674 	uint16_t		 wq_lwm;
1675 	uint32_t		 wq_pd;
1676 	uint32_t		 wq_uar_page;
1677 	uint64_t		 wq_doorbell;
1678 	uint32_t		 wq_hw_counter;
1679 	uint32_t		 wq_sw_counter;
1680 	uint16_t		 wq_log_stride;
1681 	uint8_t			 wq_log_page_sz;
1682 	uint8_t			 wq_log_size;
1683 	uint8_t			 wq_reserved1[156];
1684 } __packed __aligned(4);
1685 
1686 CTASSERT(sizeof(struct mcx_wq_ctx) == 0xC0);
1687 
1688 struct mcx_sq_ctx {
1689 	uint32_t		sq_flags;
1690 #define MCX_SQ_CTX_RLKEY			(1U << 31)
1691 #define MCX_SQ_CTX_FRE_SHIFT			(1 << 29)
1692 #define MCX_SQ_CTX_FLUSH_IN_ERROR		(1 << 28)
1693 #define MCX_SQ_CTX_MIN_WQE_INLINE_SHIFT		24
1694 #define MCX_SQ_CTX_STATE_SHIFT			20
1695 #define MCX_SQ_CTX_STATE_MASK			(0xf << 20)
1696 #define MCX_SQ_CTX_STATE_RST			0
1697 #define MCX_SQ_CTX_STATE_RDY			1
1698 #define MCX_SQ_CTX_STATE_ERR			3
1699 	uint32_t		sq_user_index;
1700 	uint32_t		sq_cqn;
1701 	uint32_t		sq_reserved1[5];
1702 	uint32_t		sq_tis_lst_sz;
1703 #define MCX_SQ_CTX_TIS_LST_SZ_SHIFT		16
1704 	uint32_t		sq_reserved2[2];
1705 	uint32_t		sq_tis_num;
1706 	struct mcx_wq_ctx	sq_wq;
1707 } __packed __aligned(4);
1708 
1709 struct mcx_sq_entry_seg {
1710 	uint32_t		sqs_byte_count;
1711 	uint32_t		sqs_lkey;
1712 	uint64_t		sqs_addr;
1713 } __packed __aligned(4);
1714 
1715 struct mcx_sq_entry {
1716 	/* control segment */
1717 	uint32_t		sqe_opcode_index;
1718 #define MCX_SQE_WQE_INDEX_SHIFT			8
1719 #define MCX_SQE_WQE_OPCODE_NOP			0x00
1720 #define MCX_SQE_WQE_OPCODE_SEND			0x0a
1721 	uint32_t		sqe_ds_sq_num;
1722 #define MCX_SQE_SQ_NUM_SHIFT			8
1723 	uint32_t		sqe_signature;
1724 #define MCX_SQE_SIGNATURE_SHIFT			24
1725 #define MCX_SQE_SOLICITED_EVENT			0x02
1726 #define MCX_SQE_CE_CQE_ON_ERR			0x00
1727 #define MCX_SQE_CE_CQE_FIRST_ERR		0x04
1728 #define MCX_SQE_CE_CQE_ALWAYS			0x08
1729 #define MCX_SQE_CE_CQE_SOLICIT			0x0C
1730 #define MCX_SQE_FM_NO_FENCE			0x00
1731 #define MCX_SQE_FM_SMALL_FENCE			0x40
1732 	uint32_t		sqe_mkey;
1733 
1734 	/* ethernet segment */
1735 	uint32_t		sqe_reserved1;
1736 	uint32_t		sqe_mss_csum;
1737 #define MCX_SQE_L4_CSUM				(1 << 31)
1738 #define MCX_SQE_L3_CSUM				(1 << 30)
1739 	uint32_t		sqe_reserved2;
1740 	uint16_t		sqe_inline_header_size;
1741 	uint16_t		sqe_inline_headers[9];
1742 
1743 	/* data segment */
1744 	struct mcx_sq_entry_seg sqe_segs[1];
1745 } __packed __aligned(64);
1746 
1747 CTASSERT(sizeof(struct mcx_sq_entry) == 64);
1748 
1749 struct mcx_cmd_create_sq_in {
1750 	uint16_t		cmd_opcode;
1751 	uint8_t			cmd_reserved0[4];
1752 	uint16_t		cmd_op_mod;
1753 	uint8_t			cmd_reserved1[8];
1754 } __packed __aligned(4);
1755 
1756 struct mcx_cmd_create_sq_out {
1757 	uint8_t			cmd_status;
1758 	uint8_t			cmd_reserved0[3];
1759 	uint32_t		cmd_syndrome;
1760 	uint32_t		cmd_sqn;
1761 	uint8_t			cmd_reserved1[4];
1762 } __packed __aligned(4);
1763 
1764 struct mcx_cmd_modify_sq_in {
1765 	uint16_t		cmd_opcode;
1766 	uint8_t			cmd_reserved0[4];
1767 	uint16_t		cmd_op_mod;
1768 	uint32_t		cmd_sq_state;
1769 	uint8_t			cmd_reserved1[4];
1770 } __packed __aligned(4);
1771 
1772 struct mcx_cmd_modify_sq_mb_in {
1773 	uint32_t		cmd_modify_hi;
1774 	uint32_t		cmd_modify_lo;
1775 	uint8_t			cmd_reserved0[8];
1776 	struct mcx_sq_ctx	cmd_sq_ctx;
1777 } __packed __aligned(4);
1778 
1779 struct mcx_cmd_modify_sq_out {
1780 	uint8_t			cmd_status;
1781 	uint8_t			cmd_reserved0[3];
1782 	uint32_t		cmd_syndrome;
1783 	uint8_t			cmd_reserved1[8];
1784 } __packed __aligned(4);
1785 
1786 struct mcx_cmd_destroy_sq_in {
1787 	uint16_t		cmd_opcode;
1788 	uint8_t			cmd_reserved0[4];
1789 	uint16_t		cmd_op_mod;
1790 	uint32_t		cmd_sqn;
1791 	uint8_t			cmd_reserved1[4];
1792 } __packed __aligned(4);
1793 
1794 struct mcx_cmd_destroy_sq_out {
1795 	uint8_t			cmd_status;
1796 	uint8_t			cmd_reserved0[3];
1797 	uint32_t		cmd_syndrome;
1798 	uint8_t			cmd_reserved1[8];
1799 } __packed __aligned(4);
1800 
1801 
1802 struct mcx_rq_ctx {
1803 	uint32_t		rq_flags;
1804 #define MCX_RQ_CTX_RLKEY			(1U << 31)
1805 #define MCX_RQ_CTX_VLAN_STRIP_DIS		(1 << 28)
1806 #define MCX_RQ_CTX_MEM_RQ_TYPE_SHIFT		24
1807 #define MCX_RQ_CTX_STATE_SHIFT			20
1808 #define MCX_RQ_CTX_STATE_MASK			(0xf << 20)
1809 #define MCX_RQ_CTX_STATE_RST			0
1810 #define MCX_RQ_CTX_STATE_RDY			1
1811 #define MCX_RQ_CTX_STATE_ERR			3
1812 #define MCX_RQ_CTX_FLUSH_IN_ERROR		(1 << 18)
1813 	uint32_t		rq_user_index;
1814 	uint32_t		rq_cqn;
1815 	uint32_t		rq_reserved1;
1816 	uint32_t		rq_rmpn;
1817 	uint32_t		rq_reserved2[7];
1818 	struct mcx_wq_ctx	rq_wq;
1819 } __packed __aligned(4);
1820 
1821 struct mcx_rq_entry {
1822 	uint32_t		rqe_byte_count;
1823 	uint32_t		rqe_lkey;
1824 	uint64_t		rqe_addr;
1825 } __packed __aligned(16);
1826 
1827 struct mcx_cmd_create_rq_in {
1828 	uint16_t		cmd_opcode;
1829 	uint8_t			cmd_reserved0[4];
1830 	uint16_t		cmd_op_mod;
1831 	uint8_t			cmd_reserved1[8];
1832 } __packed __aligned(4);
1833 
1834 struct mcx_cmd_create_rq_out {
1835 	uint8_t			cmd_status;
1836 	uint8_t			cmd_reserved0[3];
1837 	uint32_t		cmd_syndrome;
1838 	uint32_t		cmd_rqn;
1839 	uint8_t			cmd_reserved1[4];
1840 } __packed __aligned(4);
1841 
1842 struct mcx_cmd_modify_rq_in {
1843 	uint16_t		cmd_opcode;
1844 	uint8_t			cmd_reserved0[4];
1845 	uint16_t		cmd_op_mod;
1846 	uint32_t		cmd_rq_state;
1847 	uint8_t			cmd_reserved1[4];
1848 } __packed __aligned(4);
1849 
1850 struct mcx_cmd_modify_rq_mb_in {
1851 	uint32_t		cmd_modify_hi;
1852 	uint32_t		cmd_modify_lo;
1853 	uint8_t			cmd_reserved0[8];
1854 	struct mcx_rq_ctx	cmd_rq_ctx;
1855 } __packed __aligned(4);
1856 
1857 struct mcx_cmd_modify_rq_out {
1858 	uint8_t			cmd_status;
1859 	uint8_t			cmd_reserved0[3];
1860 	uint32_t		cmd_syndrome;
1861 	uint8_t			cmd_reserved1[8];
1862 } __packed __aligned(4);
1863 
1864 struct mcx_cmd_destroy_rq_in {
1865 	uint16_t		cmd_opcode;
1866 	uint8_t			cmd_reserved0[4];
1867 	uint16_t		cmd_op_mod;
1868 	uint32_t		cmd_rqn;
1869 	uint8_t			cmd_reserved1[4];
1870 } __packed __aligned(4);
1871 
1872 struct mcx_cmd_destroy_rq_out {
1873 	uint8_t			cmd_status;
1874 	uint8_t			cmd_reserved0[3];
1875 	uint32_t		cmd_syndrome;
1876 	uint8_t			cmd_reserved1[8];
1877 } __packed __aligned(4);
1878 
1879 struct mcx_cmd_create_flow_table_in {
1880 	uint16_t		cmd_opcode;
1881 	uint8_t			cmd_reserved0[4];
1882 	uint16_t		cmd_op_mod;
1883 	uint8_t			cmd_reserved1[8];
1884 } __packed __aligned(4);
1885 
1886 struct mcx_flow_table_ctx {
1887 	uint8_t			ft_miss_action;
1888 	uint8_t			ft_level;
1889 	uint8_t			ft_reserved0;
1890 	uint8_t			ft_log_size;
1891 	uint32_t		ft_table_miss_id;
1892 	uint8_t			ft_reserved1[28];
1893 } __packed __aligned(4);
1894 
1895 struct mcx_cmd_create_flow_table_mb_in {
1896 	uint8_t			cmd_table_type;
1897 	uint8_t			cmd_reserved0[7];
1898 	struct mcx_flow_table_ctx cmd_ctx;
1899 } __packed __aligned(4);
1900 
1901 struct mcx_cmd_create_flow_table_out {
1902 	uint8_t			cmd_status;
1903 	uint8_t			cmd_reserved0[3];
1904 	uint32_t		cmd_syndrome;
1905 	uint32_t		cmd_table_id;
1906 	uint8_t			cmd_reserved1[4];
1907 } __packed __aligned(4);
1908 
1909 struct mcx_cmd_destroy_flow_table_in {
1910 	uint16_t		cmd_opcode;
1911 	uint8_t			cmd_reserved0[4];
1912 	uint16_t		cmd_op_mod;
1913 	uint8_t			cmd_reserved1[8];
1914 } __packed __aligned(4);
1915 
1916 struct mcx_cmd_destroy_flow_table_mb_in {
1917 	uint8_t			cmd_table_type;
1918 	uint8_t			cmd_reserved0[3];
1919 	uint32_t		cmd_table_id;
1920 	uint8_t			cmd_reserved1[40];
1921 } __packed __aligned(4);
1922 
1923 struct mcx_cmd_destroy_flow_table_out {
1924 	uint8_t			cmd_status;
1925 	uint8_t			cmd_reserved0[3];
1926 	uint32_t		cmd_syndrome;
1927 	uint8_t			cmd_reserved1[8];
1928 } __packed __aligned(4);
1929 
1930 struct mcx_cmd_set_flow_table_root_in {
1931 	uint16_t		cmd_opcode;
1932 	uint8_t			cmd_reserved0[4];
1933 	uint16_t		cmd_op_mod;
1934 	uint8_t			cmd_reserved1[8];
1935 } __packed __aligned(4);
1936 
1937 struct mcx_cmd_set_flow_table_root_mb_in {
1938 	uint8_t			cmd_table_type;
1939 	uint8_t			cmd_reserved0[3];
1940 	uint32_t		cmd_table_id;
1941 	uint8_t			cmd_reserved1[56];
1942 } __packed __aligned(4);
1943 
1944 struct mcx_cmd_set_flow_table_root_out {
1945 	uint8_t			cmd_status;
1946 	uint8_t			cmd_reserved0[3];
1947 	uint32_t		cmd_syndrome;
1948 	uint8_t			cmd_reserved1[8];
1949 } __packed __aligned(4);
1950 
1951 struct mcx_flow_match {
1952 	/* outer headers */
1953 	uint8_t			mc_src_mac[6];
1954 	uint16_t		mc_ethertype;
1955 	uint8_t			mc_dest_mac[6];
1956 	uint16_t		mc_first_vlan;
1957 	uint8_t			mc_ip_proto;
1958 	uint8_t			mc_ip_dscp_ecn;
1959 	uint8_t			mc_vlan_flags;
1960 #define MCX_FLOW_MATCH_IP_FRAG	(1 << 5)
1961 	uint8_t			mc_tcp_flags;
1962 	uint16_t		mc_tcp_sport;
1963 	uint16_t		mc_tcp_dport;
1964 	uint32_t		mc_reserved0;
1965 	uint16_t		mc_udp_sport;
1966 	uint16_t		mc_udp_dport;
1967 	uint8_t			mc_src_ip[16];
1968 	uint8_t			mc_dest_ip[16];
1969 
1970 	/* misc parameters */
1971 	uint8_t			mc_reserved1[8];
1972 	uint16_t		mc_second_vlan;
1973 	uint8_t			mc_reserved2[2];
1974 	uint8_t			mc_second_vlan_flags;
1975 	uint8_t			mc_reserved3[15];
1976 	uint32_t		mc_outer_ipv6_flow_label;
1977 	uint8_t			mc_reserved4[32];
1978 
1979 	uint8_t			mc_reserved[384];
1980 } __packed __aligned(4);
1981 
1982 CTASSERT(sizeof(struct mcx_flow_match) == 512);
1983 
1984 struct mcx_cmd_create_flow_group_in {
1985 	uint16_t		cmd_opcode;
1986 	uint8_t			cmd_reserved0[4];
1987 	uint16_t		cmd_op_mod;
1988 	uint8_t			cmd_reserved1[8];
1989 } __packed __aligned(4);
1990 
1991 struct mcx_cmd_create_flow_group_mb_in {
1992 	uint8_t			cmd_table_type;
1993 	uint8_t			cmd_reserved0[3];
1994 	uint32_t		cmd_table_id;
1995 	uint8_t			cmd_reserved1[4];
1996 	uint32_t		cmd_start_flow_index;
1997 	uint8_t			cmd_reserved2[4];
1998 	uint32_t		cmd_end_flow_index;
1999 	uint8_t			cmd_reserved3[23];
2000 	uint8_t			cmd_match_criteria_enable;
2001 #define MCX_CREATE_FLOW_GROUP_CRIT_OUTER	(1 << 0)
2002 #define MCX_CREATE_FLOW_GROUP_CRIT_MISC		(1 << 1)
2003 #define MCX_CREATE_FLOW_GROUP_CRIT_INNER	(1 << 2)
2004 	struct mcx_flow_match	cmd_match_criteria;
2005 	uint8_t			cmd_reserved4[448];
2006 } __packed __aligned(4);
2007 
2008 struct mcx_cmd_create_flow_group_out {
2009 	uint8_t			cmd_status;
2010 	uint8_t			cmd_reserved0[3];
2011 	uint32_t		cmd_syndrome;
2012 	uint32_t		cmd_group_id;
2013 	uint8_t			cmd_reserved1[4];
2014 } __packed __aligned(4);
2015 
2016 struct mcx_flow_ctx {
2017 	uint8_t			fc_reserved0[4];
2018 	uint32_t		fc_group_id;
2019 	uint32_t		fc_flow_tag;
2020 	uint32_t		fc_action;
2021 #define MCX_FLOW_CONTEXT_ACTION_ALLOW		(1 << 0)
2022 #define MCX_FLOW_CONTEXT_ACTION_DROP		(1 << 1)
2023 #define MCX_FLOW_CONTEXT_ACTION_FORWARD		(1 << 2)
2024 #define MCX_FLOW_CONTEXT_ACTION_COUNT		(1 << 3)
2025 	uint32_t		fc_dest_list_size;
2026 	uint32_t		fc_counter_list_size;
2027 	uint8_t			fc_reserved1[40];
2028 	struct mcx_flow_match	fc_match_value;
2029 	uint8_t			fc_reserved2[192];
2030 } __packed __aligned(4);
2031 
2032 #define MCX_FLOW_CONTEXT_DEST_TYPE_TABLE	(1 << 24)
2033 #define MCX_FLOW_CONTEXT_DEST_TYPE_TIR		(2 << 24)
2034 
2035 struct mcx_cmd_destroy_flow_group_in {
2036 	uint16_t		cmd_opcode;
2037 	uint8_t			cmd_reserved0[4];
2038 	uint16_t		cmd_op_mod;
2039 	uint8_t			cmd_reserved1[8];
2040 } __packed __aligned(4);
2041 
2042 struct mcx_cmd_destroy_flow_group_mb_in {
2043 	uint8_t			cmd_table_type;
2044 	uint8_t			cmd_reserved0[3];
2045 	uint32_t		cmd_table_id;
2046 	uint32_t		cmd_group_id;
2047 	uint8_t			cmd_reserved1[36];
2048 } __packed __aligned(4);
2049 
2050 struct mcx_cmd_destroy_flow_group_out {
2051 	uint8_t			cmd_status;
2052 	uint8_t			cmd_reserved0[3];
2053 	uint32_t		cmd_syndrome;
2054 	uint8_t			cmd_reserved1[8];
2055 } __packed __aligned(4);
2056 
2057 struct mcx_cmd_set_flow_table_entry_in {
2058 	uint16_t		cmd_opcode;
2059 	uint8_t			cmd_reserved0[4];
2060 	uint16_t		cmd_op_mod;
2061 	uint8_t			cmd_reserved1[8];
2062 } __packed __aligned(4);
2063 
2064 struct mcx_cmd_set_flow_table_entry_mb_in {
2065 	uint8_t			cmd_table_type;
2066 	uint8_t			cmd_reserved0[3];
2067 	uint32_t		cmd_table_id;
2068 	uint32_t		cmd_modify_enable_mask;
2069 	uint8_t			cmd_reserved1[4];
2070 	uint32_t		cmd_flow_index;
2071 	uint8_t			cmd_reserved2[28];
2072 	struct mcx_flow_ctx	cmd_flow_ctx;
2073 } __packed __aligned(4);
2074 
2075 struct mcx_cmd_set_flow_table_entry_out {
2076 	uint8_t			cmd_status;
2077 	uint8_t			cmd_reserved0[3];
2078 	uint32_t		cmd_syndrome;
2079 	uint8_t			cmd_reserved1[8];
2080 } __packed __aligned(4);
2081 
2082 struct mcx_cmd_query_flow_table_entry_in {
2083 	uint16_t		cmd_opcode;
2084 	uint8_t			cmd_reserved0[4];
2085 	uint16_t		cmd_op_mod;
2086 	uint8_t			cmd_reserved1[8];
2087 } __packed __aligned(4);
2088 
2089 struct mcx_cmd_query_flow_table_entry_mb_in {
2090 	uint8_t			cmd_table_type;
2091 	uint8_t			cmd_reserved0[3];
2092 	uint32_t		cmd_table_id;
2093 	uint8_t			cmd_reserved1[8];
2094 	uint32_t		cmd_flow_index;
2095 	uint8_t			cmd_reserved2[28];
2096 } __packed __aligned(4);
2097 
2098 struct mcx_cmd_query_flow_table_entry_out {
2099 	uint8_t			cmd_status;
2100 	uint8_t			cmd_reserved0[3];
2101 	uint32_t		cmd_syndrome;
2102 	uint8_t			cmd_reserved1[8];
2103 } __packed __aligned(4);
2104 
2105 struct mcx_cmd_query_flow_table_entry_mb_out {
2106 	uint8_t			cmd_reserved0[48];
2107 	struct mcx_flow_ctx	cmd_flow_ctx;
2108 } __packed __aligned(4);
2109 
2110 struct mcx_cmd_delete_flow_table_entry_in {
2111 	uint16_t		cmd_opcode;
2112 	uint8_t			cmd_reserved0[4];
2113 	uint16_t		cmd_op_mod;
2114 	uint8_t			cmd_reserved1[8];
2115 } __packed __aligned(4);
2116 
2117 struct mcx_cmd_delete_flow_table_entry_mb_in {
2118 	uint8_t			cmd_table_type;
2119 	uint8_t			cmd_reserved0[3];
2120 	uint32_t		cmd_table_id;
2121 	uint8_t			cmd_reserved1[8];
2122 	uint32_t		cmd_flow_index;
2123 	uint8_t			cmd_reserved2[28];
2124 } __packed __aligned(4);
2125 
2126 struct mcx_cmd_delete_flow_table_entry_out {
2127 	uint8_t			cmd_status;
2128 	uint8_t			cmd_reserved0[3];
2129 	uint32_t		cmd_syndrome;
2130 	uint8_t			cmd_reserved1[8];
2131 } __packed __aligned(4);
2132 
2133 struct mcx_cmd_query_flow_group_in {
2134 	uint16_t		cmd_opcode;
2135 	uint8_t			cmd_reserved0[4];
2136 	uint16_t		cmd_op_mod;
2137 	uint8_t			cmd_reserved1[8];
2138 } __packed __aligned(4);
2139 
2140 struct mcx_cmd_query_flow_group_mb_in {
2141 	uint8_t			cmd_table_type;
2142 	uint8_t			cmd_reserved0[3];
2143 	uint32_t		cmd_table_id;
2144 	uint32_t		cmd_group_id;
2145 	uint8_t			cmd_reserved1[36];
2146 } __packed __aligned(4);
2147 
2148 struct mcx_cmd_query_flow_group_out {
2149 	uint8_t			cmd_status;
2150 	uint8_t			cmd_reserved0[3];
2151 	uint32_t		cmd_syndrome;
2152 	uint8_t			cmd_reserved1[8];
2153 } __packed __aligned(4);
2154 
2155 struct mcx_cmd_query_flow_group_mb_out {
2156 	uint8_t			cmd_reserved0[12];
2157 	uint32_t		cmd_start_flow_index;
2158 	uint8_t			cmd_reserved1[4];
2159 	uint32_t		cmd_end_flow_index;
2160 	uint8_t			cmd_reserved2[20];
2161 	uint32_t		cmd_match_criteria_enable;
2162 	uint8_t			cmd_match_criteria[512];
2163 	uint8_t			cmd_reserved4[448];
2164 } __packed __aligned(4);
2165 
2166 struct mcx_cmd_query_flow_table_in {
2167 	uint16_t		cmd_opcode;
2168 	uint8_t			cmd_reserved0[4];
2169 	uint16_t		cmd_op_mod;
2170 	uint8_t			cmd_reserved1[8];
2171 } __packed __aligned(4);
2172 
2173 struct mcx_cmd_query_flow_table_mb_in {
2174 	uint8_t			cmd_table_type;
2175 	uint8_t			cmd_reserved0[3];
2176 	uint32_t		cmd_table_id;
2177 	uint8_t			cmd_reserved1[40];
2178 } __packed __aligned(4);
2179 
2180 struct mcx_cmd_query_flow_table_out {
2181 	uint8_t			cmd_status;
2182 	uint8_t			cmd_reserved0[3];
2183 	uint32_t		cmd_syndrome;
2184 	uint8_t			cmd_reserved1[8];
2185 } __packed __aligned(4);
2186 
2187 struct mcx_cmd_query_flow_table_mb_out {
2188 	uint8_t			cmd_reserved0[4];
2189 	struct mcx_flow_table_ctx cmd_ctx;
2190 } __packed __aligned(4);
2191 
2192 struct mcx_cmd_alloc_flow_counter_in {
2193 	uint16_t		cmd_opcode;
2194 	uint8_t			cmd_reserved0[4];
2195 	uint16_t		cmd_op_mod;
2196 	uint8_t			cmd_reserved1[8];
2197 } __packed __aligned(4);
2198 
2199 struct mcx_cmd_query_rq_in {
2200 	uint16_t		cmd_opcode;
2201 	uint8_t			cmd_reserved0[4];
2202 	uint16_t		cmd_op_mod;
2203 	uint32_t		cmd_rqn;
2204 	uint8_t			cmd_reserved1[4];
2205 } __packed __aligned(4);
2206 
2207 struct mcx_cmd_query_rq_out {
2208 	uint8_t			cmd_status;
2209 	uint8_t			cmd_reserved0[3];
2210 	uint32_t		cmd_syndrome;
2211 	uint8_t			cmd_reserved1[8];
2212 } __packed __aligned(4);
2213 
2214 struct mcx_cmd_query_rq_mb_out {
2215 	uint8_t			cmd_reserved0[16];
2216 	struct mcx_rq_ctx	cmd_ctx;
2217 };
2218 
2219 struct mcx_cmd_query_sq_in {
2220 	uint16_t		cmd_opcode;
2221 	uint8_t			cmd_reserved0[4];
2222 	uint16_t		cmd_op_mod;
2223 	uint32_t		cmd_sqn;
2224 	uint8_t			cmd_reserved1[4];
2225 } __packed __aligned(4);
2226 
2227 struct mcx_cmd_query_sq_out {
2228 	uint8_t			cmd_status;
2229 	uint8_t			cmd_reserved0[3];
2230 	uint32_t		cmd_syndrome;
2231 	uint8_t			cmd_reserved1[8];
2232 } __packed __aligned(4);
2233 
2234 struct mcx_cmd_query_sq_mb_out {
2235 	uint8_t			cmd_reserved0[16];
2236 	struct mcx_sq_ctx	cmd_ctx;
2237 };
2238 
2239 struct mcx_cmd_alloc_flow_counter_out {
2240 	uint8_t			cmd_status;
2241 	uint8_t			cmd_reserved0[3];
2242 	uint32_t		cmd_syndrome;
2243 	uint8_t			cmd_reserved1[2];
2244 	uint16_t		cmd_flow_counter_id;
2245 	uint8_t			cmd_reserved2[4];
2246 } __packed __aligned(4);
2247 
2248 struct mcx_wq_doorbell {
2249 	uint32_t		 db_recv_counter;
2250 	uint32_t		 db_send_counter;
2251 } __packed __aligned(8);
2252 
2253 struct mcx_dmamem {
2254 	bus_dmamap_t		 mxm_map;
2255 	bus_dma_segment_t	 mxm_seg;
2256 	int			 mxm_nsegs;
2257 	size_t			 mxm_size;
2258 	void			*mxm_kva;
2259 };
2260 #define MCX_DMA_MAP(_mxm)	((_mxm)->mxm_map)
2261 #define MCX_DMA_DVA(_mxm)	((_mxm)->mxm_map->dm_segs[0].ds_addr)
2262 #define MCX_DMA_KVA(_mxm)	((void *)(_mxm)->mxm_kva)
2263 #define MCX_DMA_OFF(_mxm, _off)	((void *)((char *)(_mxm)->mxm_kva + (_off)))
2264 #define MCX_DMA_LEN(_mxm)	((_mxm)->mxm_size)
2265 
2266 struct mcx_hwmem {
2267 	bus_dmamap_t		 mhm_map;
2268 	bus_dma_segment_t	*mhm_segs;
2269 	unsigned int		 mhm_seg_count;
2270 	unsigned int		 mhm_npages;
2271 };
2272 
2273 struct mcx_slot {
2274 	bus_dmamap_t		 ms_map;
2275 	struct mbuf		*ms_m;
2276 };
2277 
2278 struct mcx_eq {
2279 	int			 eq_n;
2280 	uint32_t		 eq_cons;
2281 	struct mcx_dmamem	 eq_mem;
2282 };
2283 
2284 struct mcx_cq {
2285 	int			 cq_n;
2286 	struct mcx_dmamem	 cq_mem;
2287 	bus_addr_t		 cq_doorbell;
2288 	uint32_t		 cq_cons;
2289 	uint32_t		 cq_count;
2290 };
2291 
2292 struct mcx_calibration {
2293 	uint64_t		 c_timestamp;	/* previous mcx chip time */
2294 	uint64_t		 c_uptime;	/* previous kernel nanouptime */
2295 	uint64_t		 c_tbase;	/* mcx chip time */
2296 	uint64_t		 c_ubase;	/* kernel nanouptime */
2297 	uint64_t		 c_ratio;
2298 };
2299 
2300 #define MCX_CALIBRATE_FIRST    2
2301 #define MCX_CALIBRATE_NORMAL   32
2302 
2303 struct mcx_rxring {
2304 	u_int			 rxr_total;
2305 	u_int			 rxr_inuse;
2306 };
2307 
2308 MBUFQ_HEAD(mcx_mbufq);
2309 
2310 struct mcx_rx {
2311 	struct mcx_softc	*rx_softc;
2312 
2313 	int			 rx_rqn;
2314 	struct mcx_dmamem	 rx_rq_mem;
2315 	struct mcx_slot		*rx_slots;
2316 	bus_addr_t		 rx_doorbell;
2317 
2318 	uint32_t		 rx_prod;
2319 	callout_t		 rx_refill;
2320 	struct mcx_rxring	 rx_rxr;
2321 } __aligned(64);
2322 
2323 struct mcx_tx {
2324 	struct mcx_softc	*tx_softc;
2325 	kmutex_t		 tx_lock;
2326 	pcq_t			*tx_pcq;
2327 	void			*tx_softint;
2328 
2329 	int			 tx_uar;
2330 	int			 tx_sqn;
2331 	struct mcx_dmamem	 tx_sq_mem;
2332 	struct mcx_slot		*tx_slots;
2333 	bus_addr_t		 tx_doorbell;
2334 	int			 tx_bf_offset;
2335 
2336 	uint32_t		 tx_cons;
2337 	uint32_t		 tx_prod;
2338 } __aligned(64);
2339 
2340 struct mcx_queues {
2341 	void			*q_ihc;
2342 	struct mcx_softc	*q_sc;
2343 	int			 q_uar;
2344 	int			 q_index;
2345 	struct mcx_rx		 q_rx;
2346 	struct mcx_tx		 q_tx;
2347 	struct mcx_cq		 q_cq;
2348 	struct mcx_eq		 q_eq;
2349 #if NKSTAT > 0
2350 	struct kstat		*q_kstat;
2351 #endif
2352 };
2353 
2354 struct mcx_flow_group {
2355 	int			 g_id;
2356 	int			 g_table;
2357 	int			 g_start;
2358 	int			 g_size;
2359 };
2360 
2361 #define MCX_FLOW_GROUP_PROMISC		0
2362 #define MCX_FLOW_GROUP_ALLMULTI		1
2363 #define MCX_FLOW_GROUP_MAC		2
2364 #define MCX_FLOW_GROUP_RSS_L4		3
2365 #define MCX_FLOW_GROUP_RSS_L3		4
2366 #define MCX_FLOW_GROUP_RSS_NONE		5
2367 #define	MCX_NUM_FLOW_GROUPS		6
2368 
2369 #define MCX_HASH_SEL_L3		MCX_TIR_CTX_HASH_SEL_SRC_IP | \
2370 				MCX_TIR_CTX_HASH_SEL_DST_IP
2371 #define MCX_HASH_SEL_L4		MCX_HASH_SEL_L3 | MCX_TIR_CTX_HASH_SEL_SPORT | \
2372 				MCX_TIR_CTX_HASH_SEL_DPORT
2373 
2374 #define MCX_RSS_HASH_SEL_V4_TCP MCX_HASH_SEL_L4 | MCX_TIR_CTX_HASH_SEL_TCP  |\
2375 				MCX_TIR_CTX_HASH_SEL_IPV4
2376 #define MCX_RSS_HASH_SEL_V6_TCP	MCX_HASH_SEL_L4 | MCX_TIR_CTX_HASH_SEL_TCP | \
2377 				MCX_TIR_CTX_HASH_SEL_IPV6
2378 #define MCX_RSS_HASH_SEL_V4_UDP	MCX_HASH_SEL_L4 | MCX_TIR_CTX_HASH_SEL_UDP | \
2379 				MCX_TIR_CTX_HASH_SEL_IPV4
2380 #define MCX_RSS_HASH_SEL_V6_UDP	MCX_HASH_SEL_L4 | MCX_TIR_CTX_HASH_SEL_UDP | \
2381 				MCX_TIR_CTX_HASH_SEL_IPV6
2382 #define MCX_RSS_HASH_SEL_V4	MCX_HASH_SEL_L3 | MCX_TIR_CTX_HASH_SEL_IPV4
2383 #define MCX_RSS_HASH_SEL_V6	MCX_HASH_SEL_L3 | MCX_TIR_CTX_HASH_SEL_IPV6
2384 
2385 /*
2386  * There are a few different pieces involved in configuring RSS.
2387  * A Receive Queue Table (RQT) is the indirection table that maps packets to
2388  * different rx queues based on a hash value.  We only create one, because
2389  * we want to scatter any traffic we can apply RSS to across all our rx
2390  * queues.  Anything else will only be delivered to the first rx queue,
2391  * which doesn't require an RQT.
2392  *
2393  * A Transport Interface Receive (TIR) delivers packets to either a single rx
2394  * queue or an RQT, and in the latter case, specifies the set of fields
2395  * hashed, the hash function, and the hash key.  We need one of these for each
2396  * type of RSS traffic - v4 TCP, v6 TCP, v4 UDP, v6 UDP, other v4, other v6,
2397  * and one for non-RSS traffic.
2398  *
2399  * Flow tables hold flow table entries in sequence.  The first entry that
2400  * matches a packet is applied, sending the packet to either another flow
2401  * table or a TIR.  We use one flow table to select packets based on
2402  * destination MAC address, and a second to apply RSS.  The entries in the
2403  * first table send matching packets to the second, and the entries in the
2404  * RSS table send packets to RSS TIRs if possible, or the non-RSS TIR.
2405  *
2406  * The flow table entry that delivers packets to an RSS TIR must include match
2407  * criteria that ensure packets delivered to the TIR include all the fields
2408  * that the TIR hashes on - so for a v4 TCP TIR, the flow table entry must
2409  * only accept v4 TCP packets.  Accordingly, we need flow table entries for
2410  * each TIR.
2411  *
2412  * All of this is a lot more flexible than we need, and we can describe most
2413  * of the stuff we need with a simple array.
2414  *
2415  * An RSS config creates a TIR with hashing enabled on a set of fields,
2416  * pointing to either the first rx queue or the RQT containing all the rx
2417  * queues, and a flow table entry that matches on an ether type and
2418  * optionally an ip proto, that delivers packets to the TIR.
2419  */
2420 static struct mcx_rss_rule {
2421 	int			hash_sel;
2422 	int			flow_group;
2423 	int			ethertype;
2424 	int			ip_proto;
2425 } mcx_rss_config[] = {
2426 	/* udp and tcp for v4/v6 */
2427 	{ MCX_RSS_HASH_SEL_V4_TCP, MCX_FLOW_GROUP_RSS_L4,
2428 	  ETHERTYPE_IP, IPPROTO_TCP },
2429 	{ MCX_RSS_HASH_SEL_V6_TCP, MCX_FLOW_GROUP_RSS_L4,
2430 	  ETHERTYPE_IPV6, IPPROTO_TCP },
2431 	{ MCX_RSS_HASH_SEL_V4_UDP, MCX_FLOW_GROUP_RSS_L4,
2432 	  ETHERTYPE_IP, IPPROTO_UDP },
2433 	{ MCX_RSS_HASH_SEL_V6_UDP, MCX_FLOW_GROUP_RSS_L4,
2434 	  ETHERTYPE_IPV6, IPPROTO_UDP },
2435 
2436 	/* other v4/v6 */
2437 	{ MCX_RSS_HASH_SEL_V4, MCX_FLOW_GROUP_RSS_L3,
2438 	  ETHERTYPE_IP, 0 },
2439 	{ MCX_RSS_HASH_SEL_V6, MCX_FLOW_GROUP_RSS_L3,
2440 	  ETHERTYPE_IPV6, 0 },
2441 
2442 	/* non v4/v6 */
2443 	{ 0, MCX_FLOW_GROUP_RSS_NONE, 0, 0 }
2444 };
2445 
2446 struct mcx_softc {
2447 	device_t		 sc_dev;
2448 	struct ethercom		 sc_ec;
2449 	struct ifmedia		 sc_media;
2450 	uint64_t		 sc_media_status;
2451 	uint64_t		 sc_media_active;
2452 	kmutex_t		 sc_media_mutex;
2453 
2454 	pci_chipset_tag_t	 sc_pc;
2455 	pci_intr_handle_t	*sc_intrs;
2456 	void			*sc_ihc;
2457 	pcitag_t		 sc_tag;
2458 
2459 	bus_dma_tag_t		 sc_dmat;
2460 	bus_space_tag_t		 sc_memt;
2461 	bus_space_handle_t	 sc_memh;
2462 	bus_size_t		 sc_mems;
2463 
2464 	struct mcx_dmamem	 sc_cmdq_mem;
2465 	unsigned int		 sc_cmdq_mask;
2466 	unsigned int		 sc_cmdq_size;
2467 
2468 	unsigned int		 sc_cmdq_token;
2469 
2470 	struct mcx_hwmem	 sc_boot_pages;
2471 	struct mcx_hwmem	 sc_init_pages;
2472 	struct mcx_hwmem	 sc_regular_pages;
2473 
2474 	int			 sc_uar;
2475 	int			 sc_pd;
2476 	int			 sc_tdomain;
2477 	uint32_t		 sc_lkey;
2478 	int			 sc_tis;
2479 	int			 sc_tir[__arraycount(mcx_rss_config)];
2480 	int			 sc_rqt;
2481 
2482 	struct mcx_dmamem	 sc_doorbell_mem;
2483 
2484 	struct mcx_eq		 sc_admin_eq;
2485 	struct mcx_eq		 sc_queue_eq;
2486 
2487 	int			 sc_hardmtu;
2488 	int			 sc_rxbufsz;
2489 
2490 	int			 sc_bf_size;
2491 	int			 sc_max_rqt_size;
2492 
2493 	struct workqueue	*sc_workq;
2494 	struct work		 sc_port_change;
2495 
2496 	int			 sc_mac_flow_table_id;
2497 	int			 sc_rss_flow_table_id;
2498 	struct mcx_flow_group	 sc_flow_group[MCX_NUM_FLOW_GROUPS];
2499 	int			 sc_promisc_flow_enabled;
2500 	int			 sc_allmulti_flow_enabled;
2501 	int			 sc_mcast_flow_base;
2502 	int			 sc_extra_mcast;
2503 	uint8_t			 sc_mcast_flows[MCX_NUM_MCAST_FLOWS][ETHER_ADDR_LEN];
2504 
2505 	struct mcx_calibration	 sc_calibration[2];
2506 	unsigned int		 sc_calibration_gen;
2507 	callout_t		 sc_calibrate;
2508 	uint32_t		 sc_mhz;
2509 	uint32_t		 sc_khz;
2510 
2511 	struct mcx_queues	*sc_queues;
2512 	unsigned int		 sc_nqueues;
2513 
2514 	int			 sc_mcam_reg;
2515 
2516 #if NKSTAT > 0
2517 	struct kstat		*sc_kstat_ieee8023;
2518 	struct kstat		*sc_kstat_rfc2863;
2519 	struct kstat		*sc_kstat_rfc2819;
2520 	struct kstat		*sc_kstat_rfc3635;
2521 	unsigned int		 sc_kstat_mtmp_count;
2522 	struct kstat		**sc_kstat_mtmp;
2523 #endif
2524 
2525 	struct timecounter	 sc_timecounter;
2526 };
2527 #define DEVNAME(_sc) device_xname((_sc)->sc_dev)
2528 
2529 static int	mcx_match(device_t, cfdata_t, void *);
2530 static void	mcx_attach(device_t, device_t, void *);
2531 
2532 static void *	mcx_establish_intr(struct mcx_softc *, int, kcpuset_t *,
2533 		    int (*)(void *), void *, const char *);
2534 
2535 static void	mcx_rxr_init(struct mcx_rxring *, u_int, u_int);
2536 static u_int	mcx_rxr_get(struct mcx_rxring *, u_int);
2537 static void	mcx_rxr_put(struct mcx_rxring *, u_int);
2538 static u_int	mcx_rxr_inuse(struct mcx_rxring *);
2539 
2540 #if NKSTAT > 0
2541 static void	mcx_kstat_attach(struct mcx_softc *);
2542 #endif
2543 
2544 static void	mcx_timecounter_attach(struct mcx_softc *);
2545 
2546 static int	mcx_version(struct mcx_softc *);
2547 static int	mcx_init_wait(struct mcx_softc *);
2548 static int	mcx_enable_hca(struct mcx_softc *);
2549 static int	mcx_teardown_hca(struct mcx_softc *, uint16_t);
2550 static int	mcx_access_hca_reg(struct mcx_softc *, uint16_t, int, void *,
2551 		    int);
2552 static int	mcx_issi(struct mcx_softc *);
2553 static int	mcx_pages(struct mcx_softc *, struct mcx_hwmem *, uint16_t);
2554 static int	mcx_hca_max_caps(struct mcx_softc *);
2555 static int	mcx_hca_set_caps(struct mcx_softc *);
2556 static int	mcx_init_hca(struct mcx_softc *);
2557 static int	mcx_set_driver_version(struct mcx_softc *);
2558 static int	mcx_iff(struct mcx_softc *);
2559 static int	mcx_alloc_uar(struct mcx_softc *, int *);
2560 static int	mcx_alloc_pd(struct mcx_softc *);
2561 static int	mcx_alloc_tdomain(struct mcx_softc *);
2562 static int	mcx_create_eq(struct mcx_softc *, struct mcx_eq *, int,
2563 		    uint64_t, int);
2564 static int	mcx_query_nic_vport_context(struct mcx_softc *, uint8_t *);
2565 static int	mcx_query_special_contexts(struct mcx_softc *);
2566 static int	mcx_set_port_mtu(struct mcx_softc *, int);
2567 static int	mcx_create_cq(struct mcx_softc *, struct mcx_cq *, int, int,
2568 		    int);
2569 static int	mcx_destroy_cq(struct mcx_softc *, struct mcx_cq *);
2570 static int	mcx_create_sq(struct mcx_softc *, struct mcx_tx *, int, int,
2571 		    int);
2572 static int	mcx_destroy_sq(struct mcx_softc *, struct mcx_tx *);
2573 static int	mcx_ready_sq(struct mcx_softc *, struct mcx_tx *);
2574 static int	mcx_create_rq(struct mcx_softc *, struct mcx_rx *, int, int);
2575 static int	mcx_destroy_rq(struct mcx_softc *, struct mcx_rx *);
2576 static int	mcx_ready_rq(struct mcx_softc *, struct mcx_rx *);
2577 static int	mcx_create_tir_direct(struct mcx_softc *, struct mcx_rx *,
2578 		    int *);
2579 static int	mcx_create_tir_indirect(struct mcx_softc *, int, uint32_t,
2580 		    int *);
2581 static int	mcx_destroy_tir(struct mcx_softc *, int);
2582 static int	mcx_create_tis(struct mcx_softc *, int *);
2583 static int	mcx_destroy_tis(struct mcx_softc *, int);
2584 static int	mcx_create_rqt(struct mcx_softc *, int, int *, int *);
2585 static int	mcx_destroy_rqt(struct mcx_softc *, int);
2586 static int	mcx_create_flow_table(struct mcx_softc *, int, int, int *);
2587 static int	mcx_set_flow_table_root(struct mcx_softc *, int);
2588 static int	mcx_destroy_flow_table(struct mcx_softc *, int);
2589 static int	mcx_create_flow_group(struct mcx_softc *, int, int, int,
2590 		    int, int, struct mcx_flow_match *);
2591 static int	mcx_destroy_flow_group(struct mcx_softc *, int);
2592 static int	mcx_set_flow_table_entry_mac(struct mcx_softc *, int, int,
2593 		    const uint8_t *, uint32_t);
2594 static int	mcx_set_flow_table_entry_proto(struct mcx_softc *, int, int,
2595 		    int, int, uint32_t);
2596 static int	mcx_delete_flow_table_entry(struct mcx_softc *, int, int);
2597 
2598 #if NKSTAT > 0
2599 static int	mcx_query_rq(struct mcx_softc *, struct mcx_rx *, struct mcx_rq_ctx *);
2600 static int	mcx_query_sq(struct mcx_softc *, struct mcx_tx *, struct mcx_sq_ctx *);
2601 static int	mcx_query_cq(struct mcx_softc *, struct mcx_cq *, struct mcx_cq_ctx *);
2602 static int	mcx_query_eq(struct mcx_softc *, struct mcx_eq *, struct mcx_eq_ctx *);
2603 #endif
2604 
2605 #if 0
2606 static int	mcx_dump_flow_table(struct mcx_softc *, int);
2607 static int	mcx_dump_flow_table_entry(struct mcx_softc *, int, int);
2608 static int	mcx_dump_flow_group(struct mcx_softc *, int);
2609 #endif
2610 
2611 
2612 /*
2613 static void	mcx_cmdq_dump(const struct mcx_cmdq_entry *);
2614 static void	mcx_cmdq_mbox_dump(struct mcx_dmamem *, int);
2615 */
2616 static void	mcx_refill(void *);
2617 static int	mcx_process_rx(struct mcx_softc *, struct mcx_rx *,
2618 		    struct mcx_cq_entry *, struct mcx_mbufq *,
2619 		    const struct mcx_calibration *);
2620 static int	mcx_process_txeof(struct mcx_softc *, struct mcx_tx *,
2621 		    struct mcx_cq_entry *);
2622 static void	mcx_process_cq(struct mcx_softc *, struct mcx_queues *,
2623 		    struct mcx_cq *);
2624 
2625 static void	mcx_arm_cq(struct mcx_softc *, struct mcx_cq *, int);
2626 static void	mcx_arm_eq(struct mcx_softc *, struct mcx_eq *, int);
2627 static int	mcx_admin_intr(void *);
2628 static int	mcx_cq_intr(void *);
2629 
2630 static int	mcx_init(struct ifnet *);
2631 static void	mcx_stop(struct ifnet *, int);
2632 static int	mcx_ioctl(struct ifnet *, u_long, void *);
2633 static void	mcx_start(struct ifnet *);
2634 static int	mcx_transmit(struct ifnet *, struct mbuf *);
2635 static void	mcx_deferred_transmit(void *);
2636 static void	mcx_watchdog(struct ifnet *);
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 	sc->sc_queues = kmem_zalloc(sc->sc_nqueues * sizeof(*sc->sc_queues),
2960 	    KM_SLEEP);
2961 
2962 	strlcpy(ifp->if_xname, DEVNAME(sc), IFNAMSIZ);
2963 	ifp->if_softc = sc;
2964 	ifp->if_flags = IFF_BROADCAST | IFF_MULTICAST | IFF_SIMPLEX;
2965 #ifdef MCX_MPSAFE
2966 	ifp->if_extflags = IFEF_MPSAFE;
2967 #endif
2968 	ifp->if_init = mcx_init;
2969 	ifp->if_stop = mcx_stop;
2970 	ifp->if_ioctl = mcx_ioctl;
2971 	ifp->if_start = mcx_start;
2972 	if (sc->sc_nqueues > 1) {
2973 		ifp->if_transmit = mcx_transmit;
2974 	}
2975 	ifp->if_watchdog = mcx_watchdog;
2976 	ifp->if_mtu = sc->sc_hardmtu;
2977 	ifp->if_capabilities = IFCAP_CSUM_IPv4_Rx | IFCAP_CSUM_IPv4_Tx |
2978 	    IFCAP_CSUM_UDPv4_Rx | IFCAP_CSUM_UDPv4_Tx |
2979 	    IFCAP_CSUM_UDPv6_Rx | IFCAP_CSUM_UDPv6_Tx |
2980 	    IFCAP_CSUM_TCPv4_Rx | IFCAP_CSUM_TCPv4_Tx |
2981 	    IFCAP_CSUM_TCPv6_Rx | IFCAP_CSUM_TCPv6_Tx;
2982 	IFQ_SET_MAXLEN(&ifp->if_snd, 1024);
2983 	IFQ_SET_READY(&ifp->if_snd);
2984 
2985 	sc->sc_ec.ec_capabilities = ETHERCAP_JUMBO_MTU |
2986 	    ETHERCAP_VLAN_MTU | ETHERCAP_VLAN_HWTAGGING;
2987 	sc->sc_ec.ec_capenable |= ETHERCAP_VLAN_HWTAGGING;
2988 
2989 	sc->sc_ec.ec_ifmedia = &sc->sc_media;
2990 	ifmedia_init_with_lock(&sc->sc_media, IFM_IMASK, mcx_media_change,
2991 	    mcx_media_status, &sc->sc_media_mutex);
2992 	mcx_media_add_types(sc);
2993 	ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL);
2994 	ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO);
2995 
2996 	if_attach(ifp);
2997 
2998 	/* Enable hardware offload by default */
2999 	memset(&ifcr, 0, sizeof(ifcr));
3000 	ifcr.ifcr_capenable = ifp->if_capabilities;
3001 	ifioctl_common(ifp, SIOCSIFCAP, &ifcr);
3002 
3003 	if_deferred_start_init(ifp, NULL);
3004 
3005 	ether_ifattach(ifp, enaddr);
3006 
3007 	kcpuset_create(&affinity, false);
3008 	kcpuset_set(affinity, 0);
3009 
3010 	for (i = 0; i < sc->sc_nqueues; i++) {
3011 		struct mcx_queues *q = &sc->sc_queues[i];
3012 		struct mcx_rx *rx = &q->q_rx;
3013 		struct mcx_tx *tx = &q->q_tx;
3014 		int vec;
3015 
3016 		vec = i + 1;
3017 		q->q_sc = sc;
3018 		q->q_index = i;
3019 
3020 		if (mcx_alloc_uar(sc, &q->q_uar) != 0) {
3021 			aprint_error_dev(self, "unable to alloc uar %d\n", i);
3022 			goto teardown;
3023 		}
3024 
3025 		if (mcx_create_eq(sc, &q->q_eq, q->q_uar, 0, vec) != 0) {
3026 			aprint_error_dev(self,
3027 			    "unable to create event queue %d\n", i);
3028 			goto teardown;
3029 		}
3030 
3031 		rx->rx_softc = sc;
3032 		callout_init(&rx->rx_refill, CALLOUT_FLAGS);
3033 		callout_setfunc(&rx->rx_refill, mcx_refill, rx);
3034 
3035 		tx->tx_softc = sc;
3036 		mutex_init(&tx->tx_lock, MUTEX_DEFAULT, IPL_NET);
3037 		tx->tx_pcq = pcq_create(MCX_TXQ_NUM, KM_SLEEP);
3038 		tx->tx_softint = softint_establish(SOFTINT_NET|SOFTINT_MPSAFE,
3039 		    mcx_deferred_transmit, tx);
3040 
3041 		snprintf(intrxname, sizeof(intrxname), "%s queue %d",
3042 		    DEVNAME(sc), i);
3043 		q->q_ihc = mcx_establish_intr(sc, vec, affinity, mcx_cq_intr,
3044 		    q, intrxname);
3045 	}
3046 
3047 	callout_init(&sc->sc_calibrate, CALLOUT_FLAGS);
3048 	callout_setfunc(&sc->sc_calibrate, mcx_calibrate, sc);
3049 
3050 	if (workqueue_create(&sc->sc_workq, "mcxportchg", mcx_port_change, sc,
3051 	    PRI_NONE, IPL_NET, 0) != 0) {
3052 		aprint_error_dev(self, "couldn't create port change workq\n");
3053 		goto teardown;
3054 	}
3055 
3056 	mcx_port_change(&sc->sc_port_change, sc);
3057 
3058 	sc->sc_mac_flow_table_id = -1;
3059 	sc->sc_rss_flow_table_id = -1;
3060 	sc->sc_rqt = -1;
3061 	for (i = 0; i < MCX_NUM_FLOW_GROUPS; i++) {
3062 		struct mcx_flow_group *mfg = &sc->sc_flow_group[i];
3063 		mfg->g_id = -1;
3064 		mfg->g_table = -1;
3065 		mfg->g_size = 0;
3066 		mfg->g_start = 0;
3067 	}
3068 	sc->sc_extra_mcast = 0;
3069 	memset(sc->sc_mcast_flows, 0, sizeof(sc->sc_mcast_flows));
3070 
3071 #if NKSTAT > 0
3072 	mcx_kstat_attach(sc);
3073 #endif
3074 	mcx_timecounter_attach(sc);
3075 	return;
3076 
3077 teardown:
3078 	mcx_teardown_hca(sc, htobe16(MCX_CMD_TEARDOWN_HCA_GRACEFUL));
3079 	/* error printed by mcx_teardown_hca, and we're already unwinding */
3080 cqfree:
3081 	mcx_wr(sc, MCX_CMDQ_ADDR_HI, MCX_DMA_DVA(&sc->sc_cmdq_mem) >> 32);
3082 	mcx_bar(sc, MCX_CMDQ_ADDR_HI, sizeof(uint64_t),
3083 	    BUS_SPACE_BARRIER_WRITE);
3084 	mcx_wr(sc, MCX_CMDQ_ADDR_LO, MCX_DMA_DVA(&sc->sc_cmdq_mem) |
3085 	    MCX_CMDQ_INTERFACE_DISABLED);
3086 	mcx_bar(sc, MCX_CMDQ_ADDR_LO, sizeof(uint64_t),
3087 	    BUS_SPACE_BARRIER_WRITE);
3088 
3089 	mcx_wr(sc, MCX_CMDQ_ADDR_HI, 0);
3090 	mcx_bar(sc, MCX_CMDQ_ADDR_HI, sizeof(uint64_t),
3091 	    BUS_SPACE_BARRIER_WRITE);
3092 	mcx_wr(sc, MCX_CMDQ_ADDR_LO, MCX_CMDQ_INTERFACE_DISABLED);
3093 
3094 	mcx_dmamem_free(sc, &sc->sc_cmdq_mem);
3095 dbfree:
3096 	mcx_dmamem_free(sc, &sc->sc_doorbell_mem);
3097 unmap:
3098 	bus_space_unmap(sc->sc_memt, sc->sc_memh, sc->sc_mems);
3099 	sc->sc_mems = 0;
3100 }
3101 
3102 static void *
3103 mcx_establish_intr(struct mcx_softc *sc, int index, kcpuset_t *affinity,
3104     int (*func)(void *), void *arg, const char *xname)
3105 {
3106 	char intrbuf[PCI_INTRSTR_LEN];
3107 	const char *intrstr;
3108 	void *ih;
3109 
3110 	pci_intr_setattr(sc->sc_pc, &sc->sc_intrs[index], PCI_INTR_MPSAFE,
3111 	    true);
3112 
3113 	intrstr = pci_intr_string(sc->sc_pc, sc->sc_intrs[index], intrbuf,
3114 	    sizeof(intrbuf));
3115 	ih = pci_intr_establish_xname(sc->sc_pc, sc->sc_intrs[index], IPL_NET,
3116 	    func, arg, xname);
3117 	if (ih == NULL) {
3118 		aprint_error_dev(sc->sc_dev,
3119 		    "unable to establish interrupt%s%s\n",
3120 		    intrstr ? " at " : "",
3121 		    intrstr ? intrstr : "");
3122 		return NULL;
3123 	}
3124 
3125 	if (affinity != NULL && index > 0) {
3126 		/* Round-robin affinity */
3127 		kcpuset_zero(affinity);
3128 		kcpuset_set(affinity, (index - 1) % ncpu);
3129 		interrupt_distribute(ih, affinity, NULL);
3130 	}
3131 
3132 	return ih;
3133 }
3134 
3135 static void
3136 mcx_rxr_init(struct mcx_rxring *rxr, u_int lwm __unused, u_int hwm)
3137 {
3138 	rxr->rxr_total = hwm;
3139 	rxr->rxr_inuse = 0;
3140 }
3141 
3142 static u_int
3143 mcx_rxr_get(struct mcx_rxring *rxr, u_int max)
3144 {
3145 	const u_int taken = MIN(max, rxr->rxr_total - rxr->rxr_inuse);
3146 
3147 	rxr->rxr_inuse += taken;
3148 
3149 	return taken;
3150 }
3151 
3152 static void
3153 mcx_rxr_put(struct mcx_rxring *rxr, u_int n)
3154 {
3155 	rxr->rxr_inuse -= n;
3156 }
3157 
3158 static u_int
3159 mcx_rxr_inuse(struct mcx_rxring *rxr)
3160 {
3161 	return rxr->rxr_inuse;
3162 }
3163 
3164 static int
3165 mcx_version(struct mcx_softc *sc)
3166 {
3167 	uint32_t fw0, fw1;
3168 	uint16_t cmdif;
3169 
3170 	fw0 = mcx_rd(sc, MCX_FW_VER);
3171 	fw1 = mcx_rd(sc, MCX_CMDIF_FW_SUBVER);
3172 
3173 	aprint_normal_dev(sc->sc_dev, "FW %u.%u.%04u\n", MCX_FW_VER_MAJOR(fw0),
3174 	    MCX_FW_VER_MINOR(fw0), MCX_FW_VER_SUBMINOR(fw1));
3175 
3176 	cmdif = MCX_CMDIF(fw1);
3177 	if (cmdif != MCX_CMD_IF_SUPPORTED) {
3178 		aprint_error_dev(sc->sc_dev,
3179 		    "unsupported command interface %u\n", cmdif);
3180 		return (-1);
3181 	}
3182 
3183 	return (0);
3184 }
3185 
3186 static int
3187 mcx_init_wait(struct mcx_softc *sc)
3188 {
3189 	unsigned int i;
3190 	uint32_t r;
3191 
3192 	for (i = 0; i < 2000; i++) {
3193 		r = mcx_rd(sc, MCX_STATE);
3194 		if ((r & MCX_STATE_MASK) == MCX_STATE_READY)
3195 			return (0);
3196 
3197 		delay(1000);
3198 		mcx_bar(sc, MCX_STATE, sizeof(uint32_t),
3199 		    BUS_SPACE_BARRIER_READ);
3200 	}
3201 
3202 	return (-1);
3203 }
3204 
3205 static uint8_t
3206 mcx_cmdq_poll(struct mcx_softc *sc, struct mcx_cmdq_entry *cqe,
3207     unsigned int msec)
3208 {
3209 	unsigned int i;
3210 
3211 	for (i = 0; i < msec; i++) {
3212 		bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_cmdq_mem),
3213 		    0, MCX_DMA_LEN(&sc->sc_cmdq_mem), BUS_DMASYNC_POSTRW);
3214 
3215 		if ((cqe->cq_status & MCX_CQ_STATUS_OWN_MASK) ==
3216 		    MCX_CQ_STATUS_OWN_SW)
3217 			return (0);
3218 
3219 		delay(1000);
3220 	}
3221 
3222 	return (ETIMEDOUT);
3223 }
3224 
3225 static uint32_t
3226 mcx_mix_u64(uint32_t xor, uint64_t u64)
3227 {
3228 	xor ^= u64 >> 32;
3229 	xor ^= u64;
3230 
3231 	return (xor);
3232 }
3233 
3234 static uint32_t
3235 mcx_mix_u32(uint32_t xor, uint32_t u32)
3236 {
3237 	xor ^= u32;
3238 
3239 	return (xor);
3240 }
3241 
3242 static uint32_t
3243 mcx_mix_u8(uint32_t xor, uint8_t u8)
3244 {
3245 	xor ^= u8;
3246 
3247 	return (xor);
3248 }
3249 
3250 static uint8_t
3251 mcx_mix_done(uint32_t xor)
3252 {
3253 	xor ^= xor >> 16;
3254 	xor ^= xor >> 8;
3255 
3256 	return (xor);
3257 }
3258 
3259 static uint8_t
3260 mcx_xor(const void *buf, size_t len)
3261 {
3262 	const uint32_t *dwords = buf;
3263 	uint32_t xor = 0xff;
3264 	size_t i;
3265 
3266 	len /= sizeof(*dwords);
3267 
3268 	for (i = 0; i < len; i++)
3269 		xor ^= dwords[i];
3270 
3271 	return (mcx_mix_done(xor));
3272 }
3273 
3274 static uint8_t
3275 mcx_cmdq_token(struct mcx_softc *sc)
3276 {
3277 	uint8_t token;
3278 
3279 	do {
3280 		token = ++sc->sc_cmdq_token;
3281 	} while (token == 0);
3282 
3283 	return (token);
3284 }
3285 
3286 static void
3287 mcx_cmdq_init(struct mcx_softc *sc, struct mcx_cmdq_entry *cqe,
3288     uint32_t ilen, uint32_t olen, uint8_t token)
3289 {
3290 	memset(cqe, 0, sc->sc_cmdq_size);
3291 
3292 	cqe->cq_type = MCX_CMDQ_TYPE_PCIE;
3293 	be32enc(&cqe->cq_input_length, ilen);
3294 	be32enc(&cqe->cq_output_length, olen);
3295 	cqe->cq_token = token;
3296 	cqe->cq_status = MCX_CQ_STATUS_OWN_HW;
3297 }
3298 
3299 static void
3300 mcx_cmdq_sign(struct mcx_cmdq_entry *cqe)
3301 {
3302 	cqe->cq_signature = ~mcx_xor(cqe, sizeof(*cqe));
3303 }
3304 
3305 static int
3306 mcx_cmdq_verify(const struct mcx_cmdq_entry *cqe)
3307 {
3308 	/* return (mcx_xor(cqe, sizeof(*cqe)) ? -1 :  0); */
3309 	return (0);
3310 }
3311 
3312 static void *
3313 mcx_cmdq_in(struct mcx_cmdq_entry *cqe)
3314 {
3315 	return (&cqe->cq_input_data);
3316 }
3317 
3318 static void *
3319 mcx_cmdq_out(struct mcx_cmdq_entry *cqe)
3320 {
3321 	return (&cqe->cq_output_data);
3322 }
3323 
3324 static void
3325 mcx_cmdq_post(struct mcx_softc *sc, struct mcx_cmdq_entry *cqe,
3326     unsigned int slot)
3327 {
3328 	mcx_cmdq_sign(cqe);
3329 
3330 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_cmdq_mem),
3331 	    0, MCX_DMA_LEN(&sc->sc_cmdq_mem), BUS_DMASYNC_PRERW);
3332 
3333 	mcx_wr(sc, MCX_CMDQ_DOORBELL, 1U << slot);
3334 	mcx_bar(sc, MCX_CMDQ_DOORBELL, sizeof(uint32_t),
3335 	    BUS_SPACE_BARRIER_WRITE);
3336 }
3337 
3338 static int
3339 mcx_enable_hca(struct mcx_softc *sc)
3340 {
3341 	struct mcx_cmdq_entry *cqe;
3342 	struct mcx_cmd_enable_hca_in *in;
3343 	struct mcx_cmd_enable_hca_out *out;
3344 	int error;
3345 	uint8_t status;
3346 
3347 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
3348 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
3349 
3350 	in = mcx_cmdq_in(cqe);
3351 	in->cmd_opcode = htobe16(MCX_CMD_ENABLE_HCA);
3352 	in->cmd_op_mod = htobe16(0);
3353 	in->cmd_function_id = htobe16(0);
3354 
3355 	mcx_cmdq_post(sc, cqe, 0);
3356 
3357 	error = mcx_cmdq_poll(sc, cqe, 1000);
3358 	if (error != 0) {
3359 		printf(", hca enable timeout\n");
3360 		return (-1);
3361 	}
3362 	if (mcx_cmdq_verify(cqe) != 0) {
3363 		printf(", hca enable command corrupt\n");
3364 		return (-1);
3365 	}
3366 
3367 	status = cqe->cq_output_data[0];
3368 	if (status != MCX_CQ_STATUS_OK) {
3369 		printf(", hca enable failed (%x)\n", status);
3370 		return (-1);
3371 	}
3372 
3373 	return (0);
3374 }
3375 
3376 static int
3377 mcx_teardown_hca(struct mcx_softc *sc, uint16_t profile)
3378 {
3379 	struct mcx_cmdq_entry *cqe;
3380 	struct mcx_cmd_teardown_hca_in *in;
3381 	struct mcx_cmd_teardown_hca_out *out;
3382 	int error;
3383 	uint8_t status;
3384 
3385 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
3386 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
3387 
3388 	in = mcx_cmdq_in(cqe);
3389 	in->cmd_opcode = htobe16(MCX_CMD_TEARDOWN_HCA);
3390 	in->cmd_op_mod = htobe16(0);
3391 	in->cmd_profile = profile;
3392 
3393 	mcx_cmdq_post(sc, cqe, 0);
3394 
3395 	error = mcx_cmdq_poll(sc, cqe, 1000);
3396 	if (error != 0) {
3397 		printf(", hca teardown timeout\n");
3398 		return (-1);
3399 	}
3400 	if (mcx_cmdq_verify(cqe) != 0) {
3401 		printf(", hca teardown command corrupt\n");
3402 		return (-1);
3403 	}
3404 
3405 	status = cqe->cq_output_data[0];
3406 	if (status != MCX_CQ_STATUS_OK) {
3407 		printf(", hca teardown failed (%x)\n", status);
3408 		return (-1);
3409 	}
3410 
3411 	return (0);
3412 }
3413 
3414 static int
3415 mcx_cmdq_mboxes_alloc(struct mcx_softc *sc, struct mcx_dmamem *mxm,
3416     unsigned int nmb, uint64_t *ptr, uint8_t token)
3417 {
3418 	uint8_t *kva;
3419 	uint64_t dva;
3420 	int i;
3421 	int error;
3422 
3423 	error = mcx_dmamem_alloc(sc, mxm,
3424 	    nmb * MCX_CMDQ_MAILBOX_SIZE, MCX_CMDQ_MAILBOX_ALIGN);
3425 	if (error != 0)
3426 		return (error);
3427 
3428 	mcx_dmamem_zero(mxm);
3429 
3430 	dva = MCX_DMA_DVA(mxm);
3431 	kva = MCX_DMA_KVA(mxm);
3432 	for (i = 0; i < nmb; i++) {
3433 		struct mcx_cmdq_mailbox *mbox = (struct mcx_cmdq_mailbox *)kva;
3434 
3435 		/* patch the cqe or mbox pointing at this one */
3436 		be64enc(ptr, dva);
3437 
3438 		/* fill in this mbox */
3439 		be32enc(&mbox->mb_block_number, i);
3440 		mbox->mb_token = token;
3441 
3442 		/* move to the next one */
3443 		ptr = &mbox->mb_next_ptr;
3444 
3445 		dva += MCX_CMDQ_MAILBOX_SIZE;
3446 		kva += MCX_CMDQ_MAILBOX_SIZE;
3447 	}
3448 
3449 	return (0);
3450 }
3451 
3452 static uint32_t
3453 mcx_cmdq_mbox_ctrl_sig(const struct mcx_cmdq_mailbox *mb)
3454 {
3455 	uint32_t xor = 0xff;
3456 
3457 	/* only 3 fields get set, so mix them directly */
3458 	xor = mcx_mix_u64(xor, mb->mb_next_ptr);
3459 	xor = mcx_mix_u32(xor, mb->mb_block_number);
3460 	xor = mcx_mix_u8(xor, mb->mb_token);
3461 
3462 	return (mcx_mix_done(xor));
3463 }
3464 
3465 static void
3466 mcx_cmdq_mboxes_sign(struct mcx_dmamem *mxm, unsigned int nmb)
3467 {
3468 	uint8_t *kva;
3469 	int i;
3470 
3471 	kva = MCX_DMA_KVA(mxm);
3472 
3473 	for (i = 0; i < nmb; i++) {
3474 		struct mcx_cmdq_mailbox *mb = (struct mcx_cmdq_mailbox *)kva;
3475 		uint8_t sig = mcx_cmdq_mbox_ctrl_sig(mb);
3476 		mb->mb_ctrl_signature = sig;
3477 		mb->mb_signature = sig ^
3478 		    mcx_xor(mb->mb_data, sizeof(mb->mb_data));
3479 
3480 		kva += MCX_CMDQ_MAILBOX_SIZE;
3481 	}
3482 }
3483 
3484 static void
3485 mcx_cmdq_mboxes_sync(struct mcx_softc *sc, struct mcx_dmamem *mxm, int ops)
3486 {
3487 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(mxm),
3488 	    0, MCX_DMA_LEN(mxm), ops);
3489 }
3490 
3491 static struct mcx_cmdq_mailbox *
3492 mcx_cq_mbox(struct mcx_dmamem *mxm, unsigned int i)
3493 {
3494 	uint8_t *kva;
3495 
3496 	kva = MCX_DMA_KVA(mxm);
3497 	kva += i * MCX_CMDQ_MAILBOX_SIZE;
3498 
3499 	return ((struct mcx_cmdq_mailbox *)kva);
3500 }
3501 
3502 static inline void *
3503 mcx_cq_mbox_data(struct mcx_cmdq_mailbox *mb)
3504 {
3505 	return (&mb->mb_data);
3506 }
3507 
3508 static void
3509 mcx_cmdq_mboxes_copyin(struct mcx_dmamem *mxm, unsigned int nmb,
3510     void *b, size_t len)
3511 {
3512 	uint8_t *buf = b;
3513 	struct mcx_cmdq_mailbox *mb;
3514 	int i;
3515 
3516 	mb = (struct mcx_cmdq_mailbox *)MCX_DMA_KVA(mxm);
3517 	for (i = 0; i < nmb; i++) {
3518 
3519 		memcpy(mb->mb_data, buf, uimin(sizeof(mb->mb_data), len));
3520 
3521 		if (sizeof(mb->mb_data) >= len)
3522 			break;
3523 
3524 		buf += sizeof(mb->mb_data);
3525 		len -= sizeof(mb->mb_data);
3526 		mb++;
3527 	}
3528 }
3529 
3530 static void
3531 mcx_cmdq_mboxes_pas(struct mcx_dmamem *mxm, int offset, int npages,
3532     struct mcx_dmamem *buf)
3533 {
3534 	uint64_t *pas;
3535 	int mbox, mbox_pages, i;
3536 
3537 	mbox = offset / MCX_CMDQ_MAILBOX_DATASIZE;
3538 	offset %= MCX_CMDQ_MAILBOX_DATASIZE;
3539 
3540 	pas = mcx_cq_mbox_data(mcx_cq_mbox(mxm, mbox));
3541 	pas += (offset / sizeof(*pas));
3542 	mbox_pages = (MCX_CMDQ_MAILBOX_DATASIZE - offset) / sizeof(*pas);
3543 	for (i = 0; i < npages; i++) {
3544 		if (i == mbox_pages) {
3545 			mbox++;
3546 			pas = mcx_cq_mbox_data(mcx_cq_mbox(mxm, mbox));
3547 			mbox_pages += MCX_CMDQ_MAILBOX_DATASIZE / sizeof(*pas);
3548 		}
3549 		*pas = htobe64(MCX_DMA_DVA(buf) + (i * MCX_PAGE_SIZE));
3550 		pas++;
3551 	}
3552 }
3553 
3554 static void
3555 mcx_cmdq_mboxes_copyout(struct mcx_dmamem *mxm, int nmb, void *b, size_t len)
3556 {
3557 	uint8_t *buf = b;
3558 	struct mcx_cmdq_mailbox *mb;
3559 	int i;
3560 
3561 	mb = (struct mcx_cmdq_mailbox *)MCX_DMA_KVA(mxm);
3562 	for (i = 0; i < nmb; i++) {
3563 		memcpy(buf, mb->mb_data, uimin(sizeof(mb->mb_data), len));
3564 
3565 		if (sizeof(mb->mb_data) >= len)
3566 			break;
3567 
3568 		buf += sizeof(mb->mb_data);
3569 		len -= sizeof(mb->mb_data);
3570 		mb++;
3571 	}
3572 }
3573 
3574 static void
3575 mcx_cq_mboxes_free(struct mcx_softc *sc, struct mcx_dmamem *mxm)
3576 {
3577 	mcx_dmamem_free(sc, mxm);
3578 }
3579 
3580 #if 0
3581 static void
3582 mcx_cmdq_dump(const struct mcx_cmdq_entry *cqe)
3583 {
3584 	unsigned int i;
3585 
3586 	printf(" type %02x, ilen %u, iptr %016llx", cqe->cq_type,
3587 	    be32dec(&cqe->cq_input_length), be64dec(&cqe->cq_input_ptr));
3588 
3589 	printf(", idata ");
3590 	for (i = 0; i < sizeof(cqe->cq_input_data); i++)
3591 		printf("%02x", cqe->cq_input_data[i]);
3592 
3593 	printf(", odata ");
3594 	for (i = 0; i < sizeof(cqe->cq_output_data); i++)
3595 		printf("%02x", cqe->cq_output_data[i]);
3596 
3597 	printf(", optr %016llx, olen %u, token %02x, sig %02x, status %02x",
3598 	    be64dec(&cqe->cq_output_ptr), be32dec(&cqe->cq_output_length),
3599 	    cqe->cq_token, cqe->cq_signature, cqe->cq_status);
3600 }
3601 
3602 static void
3603 mcx_cmdq_mbox_dump(struct mcx_dmamem *mboxes, int num)
3604 {
3605 	int i, j;
3606 	uint8_t *d;
3607 
3608 	for (i = 0; i < num; i++) {
3609 		struct mcx_cmdq_mailbox *mbox;
3610 		mbox = mcx_cq_mbox(mboxes, i);
3611 
3612 		d = mcx_cq_mbox_data(mbox);
3613 		for (j = 0; j < MCX_CMDQ_MAILBOX_DATASIZE; j++) {
3614 			if (j != 0 && (j % 16 == 0))
3615 				printf("\n");
3616 			printf("%.2x ", d[j]);
3617 		}
3618 	}
3619 }
3620 #endif
3621 
3622 static int
3623 mcx_access_hca_reg(struct mcx_softc *sc, uint16_t reg, int op, void *data,
3624     int len)
3625 {
3626 	struct mcx_dmamem mxm;
3627 	struct mcx_cmdq_entry *cqe;
3628 	struct mcx_cmd_access_reg_in *in;
3629 	struct mcx_cmd_access_reg_out *out;
3630 	uint8_t token = mcx_cmdq_token(sc);
3631 	int error, nmb;
3632 
3633 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
3634 	mcx_cmdq_init(sc, cqe, sizeof(*in) + len, sizeof(*out) + len,
3635 	    token);
3636 
3637 	in = mcx_cmdq_in(cqe);
3638 	in->cmd_opcode = htobe16(MCX_CMD_ACCESS_REG);
3639 	in->cmd_op_mod = htobe16(op);
3640 	in->cmd_register_id = htobe16(reg);
3641 
3642 	nmb = howmany(len, MCX_CMDQ_MAILBOX_DATASIZE);
3643 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, nmb,
3644 	    &cqe->cq_output_ptr, token) != 0) {
3645 		printf(", unable to allocate access reg mailboxen\n");
3646 		return (-1);
3647 	}
3648 	cqe->cq_input_ptr = cqe->cq_output_ptr;
3649 	mcx_cmdq_mboxes_copyin(&mxm, nmb, data, len);
3650 	mcx_cmdq_mboxes_sign(&mxm, nmb);
3651 	mcx_cmdq_mboxes_sync(sc, &mxm, BUS_DMASYNC_PRERW);
3652 
3653 	mcx_cmdq_post(sc, cqe, 0);
3654 	error = mcx_cmdq_poll(sc, cqe, 1000);
3655 	mcx_cmdq_mboxes_sync(sc, &mxm, BUS_DMASYNC_POSTRW);
3656 
3657 	if (error != 0) {
3658 		printf("%s: access reg (%s %x) timeout\n", DEVNAME(sc),
3659 		    (op == MCX_REG_OP_WRITE ? "write" : "read"), reg);
3660 		goto free;
3661 	}
3662 	error = mcx_cmdq_verify(cqe);
3663 	if (error != 0) {
3664 		printf("%s: access reg (%s %x) reply corrupt\n",
3665 		    (op == MCX_REG_OP_WRITE ? "write" : "read"), DEVNAME(sc),
3666 		    reg);
3667 		goto free;
3668 	}
3669 
3670 	out = mcx_cmdq_out(cqe);
3671 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
3672 		printf("%s: access reg (%s %x) failed (%x, %.6x)\n",
3673 		    DEVNAME(sc), (op == MCX_REG_OP_WRITE ? "write" : "read"),
3674 		    reg, out->cmd_status, be32toh(out->cmd_syndrome));
3675 		error = -1;
3676 		goto free;
3677 	}
3678 
3679 	mcx_cmdq_mboxes_copyout(&mxm, nmb, data, len);
3680 free:
3681 	mcx_dmamem_free(sc, &mxm);
3682 
3683 	return (error);
3684 }
3685 
3686 static int
3687 mcx_set_issi(struct mcx_softc *sc, struct mcx_cmdq_entry *cqe,
3688     unsigned int slot)
3689 {
3690 	struct mcx_cmd_set_issi_in *in;
3691 	struct mcx_cmd_set_issi_out *out;
3692 	uint8_t status;
3693 
3694 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
3695 
3696 	in = mcx_cmdq_in(cqe);
3697 	in->cmd_opcode = htobe16(MCX_CMD_SET_ISSI);
3698 	in->cmd_op_mod = htobe16(0);
3699 	in->cmd_current_issi = htobe16(MCX_ISSI);
3700 
3701 	mcx_cmdq_post(sc, cqe, slot);
3702 	if (mcx_cmdq_poll(sc, cqe, 1000) != 0)
3703 		return (-1);
3704 	if (mcx_cmdq_verify(cqe) != 0)
3705 		return (-1);
3706 
3707 	status = cqe->cq_output_data[0];
3708 	if (status != MCX_CQ_STATUS_OK)
3709 		return (-1);
3710 
3711 	return (0);
3712 }
3713 
3714 static int
3715 mcx_issi(struct mcx_softc *sc)
3716 {
3717 	struct mcx_dmamem mxm;
3718 	struct mcx_cmdq_entry *cqe;
3719 	struct mcx_cmd_query_issi_in *in;
3720 	struct mcx_cmd_query_issi_il_out *out;
3721 	struct mcx_cmd_query_issi_mb_out *mb;
3722 	uint8_t token = mcx_cmdq_token(sc);
3723 	uint8_t status;
3724 	int error;
3725 
3726 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
3727 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out) + sizeof(*mb), token);
3728 
3729 	in = mcx_cmdq_in(cqe);
3730 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_ISSI);
3731 	in->cmd_op_mod = htobe16(0);
3732 
3733 	CTASSERT(sizeof(*mb) <= MCX_CMDQ_MAILBOX_DATASIZE);
3734 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
3735 	    &cqe->cq_output_ptr, token) != 0) {
3736 		printf(", unable to allocate query issi mailbox\n");
3737 		return (-1);
3738 	}
3739 	mcx_cmdq_mboxes_sign(&mxm, 1);
3740 
3741 	mcx_cmdq_post(sc, cqe, 0);
3742 	error = mcx_cmdq_poll(sc, cqe, 1000);
3743 	if (error != 0) {
3744 		printf(", query issi timeout\n");
3745 		goto free;
3746 	}
3747 	error = mcx_cmdq_verify(cqe);
3748 	if (error != 0) {
3749 		printf(", query issi reply corrupt\n");
3750 		goto free;
3751 	}
3752 
3753 	status = cqe->cq_output_data[0];
3754 	switch (status) {
3755 	case MCX_CQ_STATUS_OK:
3756 		break;
3757 	case MCX_CQ_STATUS_BAD_OPCODE:
3758 		/* use ISSI 0 */
3759 		goto free;
3760 	default:
3761 		printf(", query issi failed (%x)\n", status);
3762 		error = -1;
3763 		goto free;
3764 	}
3765 
3766 	out = mcx_cmdq_out(cqe);
3767 	if (out->cmd_current_issi == htobe16(MCX_ISSI)) {
3768 		/* use ISSI 1 */
3769 		goto free;
3770 	}
3771 
3772 	/* don't need to read cqe anymore, can be used for SET ISSI */
3773 
3774 	mb = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
3775 	CTASSERT(MCX_ISSI < NBBY);
3776 	 /* XXX math is hard */
3777 	if (!ISSET(mb->cmd_supported_issi[79], 1 << MCX_ISSI)) {
3778 		/* use ISSI 0 */
3779 		goto free;
3780 	}
3781 
3782 	if (mcx_set_issi(sc, cqe, 0) != 0) {
3783 		/* ignore the error, just use ISSI 0 */
3784 	} else {
3785 		/* use ISSI 1 */
3786 	}
3787 
3788 free:
3789 	mcx_cq_mboxes_free(sc, &mxm);
3790 	return (error);
3791 }
3792 
3793 static int
3794 mcx_query_pages(struct mcx_softc *sc, uint16_t type,
3795     int32_t *npages, uint16_t *func_id)
3796 {
3797 	struct mcx_cmdq_entry *cqe;
3798 	struct mcx_cmd_query_pages_in *in;
3799 	struct mcx_cmd_query_pages_out *out;
3800 
3801 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
3802 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
3803 
3804 	in = mcx_cmdq_in(cqe);
3805 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_PAGES);
3806 	in->cmd_op_mod = type;
3807 
3808 	mcx_cmdq_post(sc, cqe, 0);
3809 	if (mcx_cmdq_poll(sc, cqe, 1000) != 0) {
3810 		printf(", query pages timeout\n");
3811 		return (-1);
3812 	}
3813 	if (mcx_cmdq_verify(cqe) != 0) {
3814 		printf(", query pages reply corrupt\n");
3815 		return (-1);
3816 	}
3817 
3818 	out = mcx_cmdq_out(cqe);
3819 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
3820 		printf(", query pages failed (%x)\n", out->cmd_status);
3821 		return (-1);
3822 	}
3823 
3824 	*func_id = out->cmd_func_id;
3825 	*npages = be32dec(&out->cmd_num_pages);
3826 
3827 	return (0);
3828 }
3829 
3830 struct bus_dma_iter {
3831 	bus_dmamap_t		i_map;
3832 	bus_size_t		i_offset;
3833 	unsigned int		i_index;
3834 };
3835 
3836 static void
3837 bus_dma_iter_init(struct bus_dma_iter *i, bus_dmamap_t map)
3838 {
3839 	i->i_map = map;
3840 	i->i_offset = 0;
3841 	i->i_index = 0;
3842 }
3843 
3844 static bus_addr_t
3845 bus_dma_iter_addr(struct bus_dma_iter *i)
3846 {
3847 	return (i->i_map->dm_segs[i->i_index].ds_addr + i->i_offset);
3848 }
3849 
3850 static void
3851 bus_dma_iter_add(struct bus_dma_iter *i, bus_size_t size)
3852 {
3853 	bus_dma_segment_t *seg = i->i_map->dm_segs + i->i_index;
3854 	bus_size_t diff;
3855 
3856 	do {
3857 		diff = seg->ds_len - i->i_offset;
3858 		if (size < diff)
3859 			break;
3860 
3861 		size -= diff;
3862 
3863 		seg++;
3864 
3865 		i->i_offset = 0;
3866 		i->i_index++;
3867 	} while (size > 0);
3868 
3869 	i->i_offset += size;
3870 }
3871 
3872 static int
3873 mcx_add_pages(struct mcx_softc *sc, struct mcx_hwmem *mhm, uint16_t func_id)
3874 {
3875 	struct mcx_dmamem mxm;
3876 	struct mcx_cmdq_entry *cqe;
3877 	struct mcx_cmd_manage_pages_in *in;
3878 	struct mcx_cmd_manage_pages_out *out;
3879 	unsigned int paslen, nmb, i, j, npages;
3880 	struct bus_dma_iter iter;
3881 	uint64_t *pas;
3882 	uint8_t status;
3883 	uint8_t token = mcx_cmdq_token(sc);
3884 	int error;
3885 
3886 	npages = mhm->mhm_npages;
3887 
3888 	paslen = sizeof(*pas) * npages;
3889 	nmb = howmany(paslen, MCX_CMDQ_MAILBOX_DATASIZE);
3890 
3891 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
3892 	mcx_cmdq_init(sc, cqe, sizeof(*in) + paslen, sizeof(*out), token);
3893 
3894 	in = mcx_cmdq_in(cqe);
3895 	in->cmd_opcode = htobe16(MCX_CMD_MANAGE_PAGES);
3896 	in->cmd_op_mod = htobe16(MCX_CMD_MANAGE_PAGES_ALLOC_SUCCESS);
3897 	in->cmd_func_id = func_id;
3898 	be32enc(&in->cmd_input_num_entries, npages);
3899 
3900 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, nmb,
3901 	    &cqe->cq_input_ptr, token) != 0) {
3902 		printf(", unable to allocate manage pages mailboxen\n");
3903 		return (-1);
3904 	}
3905 
3906 	bus_dma_iter_init(&iter, mhm->mhm_map);
3907 	for (i = 0; i < nmb; i++) {
3908 		unsigned int lim;
3909 
3910 		pas = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, i));
3911 		lim = uimin(MCX_CMDQ_MAILBOX_DATASIZE / sizeof(*pas), npages);
3912 
3913 		for (j = 0; j < lim; j++) {
3914 			be64enc(&pas[j], bus_dma_iter_addr(&iter));
3915 			bus_dma_iter_add(&iter, MCX_PAGE_SIZE);
3916 		}
3917 
3918 		npages -= lim;
3919 	}
3920 
3921 	mcx_cmdq_mboxes_sign(&mxm, nmb);
3922 
3923 	mcx_cmdq_post(sc, cqe, 0);
3924 	error = mcx_cmdq_poll(sc, cqe, 1000);
3925 	if (error != 0) {
3926 		printf(", manage pages timeout\n");
3927 		goto free;
3928 	}
3929 	error = mcx_cmdq_verify(cqe);
3930 	if (error != 0) {
3931 		printf(", manage pages reply corrupt\n");
3932 		goto free;
3933 	}
3934 
3935 	status = cqe->cq_output_data[0];
3936 	if (status != MCX_CQ_STATUS_OK) {
3937 		printf(", manage pages failed (%x)\n", status);
3938 		error = -1;
3939 		goto free;
3940 	}
3941 
3942 free:
3943 	mcx_dmamem_free(sc, &mxm);
3944 
3945 	return (error);
3946 }
3947 
3948 static int
3949 mcx_pages(struct mcx_softc *sc, struct mcx_hwmem *mhm, uint16_t type)
3950 {
3951 	int32_t npages;
3952 	uint16_t func_id;
3953 
3954 	if (mcx_query_pages(sc, type, &npages, &func_id) != 0) {
3955 		/* error printed by mcx_query_pages */
3956 		return (-1);
3957 	}
3958 
3959 	if (npages < 1)
3960 		return (0);
3961 
3962 	if (mcx_hwmem_alloc(sc, mhm, npages) != 0) {
3963 		printf(", unable to allocate hwmem\n");
3964 		return (-1);
3965 	}
3966 
3967 	if (mcx_add_pages(sc, mhm, func_id) != 0) {
3968 		printf(", unable to add hwmem\n");
3969 		goto free;
3970 	}
3971 
3972 	return (0);
3973 
3974 free:
3975 	mcx_hwmem_free(sc, mhm);
3976 
3977 	return (-1);
3978 }
3979 
3980 static int
3981 mcx_hca_max_caps(struct mcx_softc *sc)
3982 {
3983 	struct mcx_dmamem mxm;
3984 	struct mcx_cmdq_entry *cqe;
3985 	struct mcx_cmd_query_hca_cap_in *in;
3986 	struct mcx_cmd_query_hca_cap_out *out;
3987 	struct mcx_cmdq_mailbox *mb;
3988 	struct mcx_cap_device *hca;
3989 	uint8_t status;
3990 	uint8_t token = mcx_cmdq_token(sc);
3991 	int error;
3992 
3993 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
3994 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out) + MCX_HCA_CAP_LEN,
3995 	    token);
3996 
3997 	in = mcx_cmdq_in(cqe);
3998 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_HCA_CAP);
3999 	in->cmd_op_mod = htobe16(MCX_CMD_QUERY_HCA_CAP_MAX |
4000 	    MCX_CMD_QUERY_HCA_CAP_DEVICE);
4001 
4002 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, MCX_HCA_CAP_NMAILBOXES,
4003 	    &cqe->cq_output_ptr, token) != 0) {
4004 		printf(", unable to allocate query hca caps mailboxen\n");
4005 		return (-1);
4006 	}
4007 	mcx_cmdq_mboxes_sign(&mxm, MCX_HCA_CAP_NMAILBOXES);
4008 	mcx_cmdq_mboxes_sync(sc, &mxm, BUS_DMASYNC_PRERW);
4009 
4010 	mcx_cmdq_post(sc, cqe, 0);
4011 	error = mcx_cmdq_poll(sc, cqe, 1000);
4012 	mcx_cmdq_mboxes_sync(sc, &mxm, BUS_DMASYNC_POSTRW);
4013 
4014 	if (error != 0) {
4015 		printf(", query hca caps timeout\n");
4016 		goto free;
4017 	}
4018 	error = mcx_cmdq_verify(cqe);
4019 	if (error != 0) {
4020 		printf(", query hca caps reply corrupt\n");
4021 		goto free;
4022 	}
4023 
4024 	status = cqe->cq_output_data[0];
4025 	if (status != MCX_CQ_STATUS_OK) {
4026 		printf(", query hca caps failed (%x)\n", status);
4027 		error = -1;
4028 		goto free;
4029 	}
4030 
4031 	mb = mcx_cq_mbox(&mxm, 0);
4032 	hca = mcx_cq_mbox_data(mb);
4033 
4034 	if ((hca->port_type & MCX_CAP_DEVICE_PORT_TYPE)
4035 	    != MCX_CAP_DEVICE_PORT_TYPE_ETH) {
4036 		printf(", not in ethernet mode\n");
4037 		error = -1;
4038 		goto free;
4039 	}
4040 	if (hca->log_pg_sz > PAGE_SHIFT) {
4041 		printf(", minimum system page shift %u is too large\n",
4042 		    hca->log_pg_sz);
4043 		error = -1;
4044 		goto free;
4045 	}
4046 	/*
4047 	 * blueflame register is split into two buffers, and we must alternate
4048 	 * between the two of them.
4049 	 */
4050 	sc->sc_bf_size = (1 << hca->log_bf_reg_size) / 2;
4051 	sc->sc_max_rqt_size = (1 << hca->log_max_rqt_size);
4052 
4053 	if (hca->local_ca_ack_delay & MCX_CAP_DEVICE_MCAM_REG)
4054 		sc->sc_mcam_reg = 1;
4055 
4056 	sc->sc_mhz = be32dec(&hca->device_frequency_mhz);
4057 	sc->sc_khz = be32dec(&hca->device_frequency_khz);
4058 
4059 free:
4060 	mcx_dmamem_free(sc, &mxm);
4061 
4062 	return (error);
4063 }
4064 
4065 static int
4066 mcx_hca_set_caps(struct mcx_softc *sc)
4067 {
4068 	struct mcx_dmamem mxm;
4069 	struct mcx_cmdq_entry *cqe;
4070 	struct mcx_cmd_query_hca_cap_in *in;
4071 	struct mcx_cmd_query_hca_cap_out *out;
4072 	struct mcx_cmdq_mailbox *mb;
4073 	struct mcx_cap_device *hca;
4074 	uint8_t status;
4075 	uint8_t token = mcx_cmdq_token(sc);
4076 	int error;
4077 
4078 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4079 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out) + MCX_HCA_CAP_LEN,
4080 	    token);
4081 
4082 	in = mcx_cmdq_in(cqe);
4083 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_HCA_CAP);
4084 	in->cmd_op_mod = htobe16(MCX_CMD_QUERY_HCA_CAP_CURRENT |
4085 	    MCX_CMD_QUERY_HCA_CAP_DEVICE);
4086 
4087 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, MCX_HCA_CAP_NMAILBOXES,
4088 	    &cqe->cq_output_ptr, token) != 0) {
4089 		printf(", unable to allocate manage pages mailboxen\n");
4090 		return (-1);
4091 	}
4092 	mcx_cmdq_mboxes_sign(&mxm, MCX_HCA_CAP_NMAILBOXES);
4093 	mcx_cmdq_mboxes_sync(sc, &mxm, BUS_DMASYNC_PRERW);
4094 
4095 	mcx_cmdq_post(sc, cqe, 0);
4096 	error = mcx_cmdq_poll(sc, cqe, 1000);
4097 	mcx_cmdq_mboxes_sync(sc, &mxm, BUS_DMASYNC_POSTRW);
4098 
4099 	if (error != 0) {
4100 		printf(", query hca caps timeout\n");
4101 		goto free;
4102 	}
4103 	error = mcx_cmdq_verify(cqe);
4104 	if (error != 0) {
4105 		printf(", query hca caps reply corrupt\n");
4106 		goto free;
4107 	}
4108 
4109 	status = cqe->cq_output_data[0];
4110 	if (status != MCX_CQ_STATUS_OK) {
4111 		printf(", query hca caps failed (%x)\n", status);
4112 		error = -1;
4113 		goto free;
4114 	}
4115 
4116 	mb = mcx_cq_mbox(&mxm, 0);
4117 	hca = mcx_cq_mbox_data(mb);
4118 
4119 	hca->log_pg_sz = PAGE_SHIFT;
4120 
4121 free:
4122 	mcx_dmamem_free(sc, &mxm);
4123 
4124 	return (error);
4125 }
4126 
4127 
4128 static int
4129 mcx_init_hca(struct mcx_softc *sc)
4130 {
4131 	struct mcx_cmdq_entry *cqe;
4132 	struct mcx_cmd_init_hca_in *in;
4133 	struct mcx_cmd_init_hca_out *out;
4134 	int error;
4135 	uint8_t status;
4136 
4137 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4138 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
4139 
4140 	in = mcx_cmdq_in(cqe);
4141 	in->cmd_opcode = htobe16(MCX_CMD_INIT_HCA);
4142 	in->cmd_op_mod = htobe16(0);
4143 
4144 	mcx_cmdq_post(sc, cqe, 0);
4145 
4146 	error = mcx_cmdq_poll(sc, cqe, 1000);
4147 	if (error != 0) {
4148 		printf(", hca init timeout\n");
4149 		return (-1);
4150 	}
4151 	if (mcx_cmdq_verify(cqe) != 0) {
4152 		printf(", hca init command corrupt\n");
4153 		return (-1);
4154 	}
4155 
4156 	status = cqe->cq_output_data[0];
4157 	if (status != MCX_CQ_STATUS_OK) {
4158 		printf(", hca init failed (%x)\n", status);
4159 		return (-1);
4160 	}
4161 
4162 	return (0);
4163 }
4164 
4165 static int
4166 mcx_set_driver_version(struct mcx_softc *sc)
4167 {
4168 	struct mcx_dmamem mxm;
4169 	struct mcx_cmdq_entry *cqe;
4170 	struct mcx_cmd_set_driver_version_in *in;
4171 	struct mcx_cmd_set_driver_version_out *out;
4172 	int error;
4173 	int token;
4174 	uint8_t status;
4175 
4176 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4177 	token = mcx_cmdq_token(sc);
4178 	mcx_cmdq_init(sc, cqe, sizeof(*in) +
4179 	    sizeof(struct mcx_cmd_set_driver_version), sizeof(*out), token);
4180 
4181 	in = mcx_cmdq_in(cqe);
4182 	in->cmd_opcode = htobe16(MCX_CMD_SET_DRIVER_VERSION);
4183 	in->cmd_op_mod = htobe16(0);
4184 
4185 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
4186 	    &cqe->cq_input_ptr, token) != 0) {
4187 		printf(", unable to allocate set driver version mailboxen\n");
4188 		return (-1);
4189 	}
4190 	strlcpy(mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)),
4191 	    "OpenBSD,mcx,1.000.000000", MCX_CMDQ_MAILBOX_DATASIZE);
4192 
4193 	mcx_cmdq_mboxes_sign(&mxm, 1);
4194 	mcx_cmdq_post(sc, cqe, 0);
4195 
4196 	error = mcx_cmdq_poll(sc, cqe, 1000);
4197 	if (error != 0) {
4198 		printf(", set driver version timeout\n");
4199 		goto free;
4200 	}
4201 	if (mcx_cmdq_verify(cqe) != 0) {
4202 		printf(", set driver version command corrupt\n");
4203 		goto free;
4204 	}
4205 
4206 	status = cqe->cq_output_data[0];
4207 	if (status != MCX_CQ_STATUS_OK) {
4208 		printf(", set driver version failed (%x)\n", status);
4209 		error = -1;
4210 		goto free;
4211 	}
4212 
4213 free:
4214 	mcx_dmamem_free(sc, &mxm);
4215 
4216 	return (error);
4217 }
4218 
4219 static int
4220 mcx_iff(struct mcx_softc *sc)
4221 {
4222 	struct ifnet *ifp = &sc->sc_ec.ec_if;
4223 	struct mcx_dmamem mxm;
4224 	struct mcx_cmdq_entry *cqe;
4225 	struct mcx_cmd_modify_nic_vport_context_in *in;
4226 	struct mcx_cmd_modify_nic_vport_context_out *out;
4227 	struct mcx_nic_vport_ctx *ctx;
4228 	int error;
4229 	int token;
4230 	int insize;
4231 	uint32_t dest;
4232 
4233 	dest = MCX_FLOW_CONTEXT_DEST_TYPE_TABLE |
4234 	    sc->sc_rss_flow_table_id;
4235 
4236 	/* enable or disable the promisc flow */
4237 	if (ISSET(ifp->if_flags, IFF_PROMISC)) {
4238 		if (sc->sc_promisc_flow_enabled == 0) {
4239 			mcx_set_flow_table_entry_mac(sc,
4240 			    MCX_FLOW_GROUP_PROMISC, 0, NULL, dest);
4241 			sc->sc_promisc_flow_enabled = 1;
4242 		}
4243 	} else if (sc->sc_promisc_flow_enabled != 0) {
4244 		mcx_delete_flow_table_entry(sc, MCX_FLOW_GROUP_PROMISC, 0);
4245 		sc->sc_promisc_flow_enabled = 0;
4246 	}
4247 
4248 	/* enable or disable the all-multicast flow */
4249 	if (ISSET(ifp->if_flags, IFF_ALLMULTI)) {
4250 		if (sc->sc_allmulti_flow_enabled == 0) {
4251 			uint8_t mcast[ETHER_ADDR_LEN];
4252 
4253 			memset(mcast, 0, sizeof(mcast));
4254 			mcast[0] = 0x01;
4255 			mcx_set_flow_table_entry_mac(sc,
4256 			    MCX_FLOW_GROUP_ALLMULTI, 0, mcast, dest);
4257 			sc->sc_allmulti_flow_enabled = 1;
4258 		}
4259 	} else if (sc->sc_allmulti_flow_enabled != 0) {
4260 		mcx_delete_flow_table_entry(sc, MCX_FLOW_GROUP_ALLMULTI, 0);
4261 		sc->sc_allmulti_flow_enabled = 0;
4262 	}
4263 
4264 	insize = sizeof(struct mcx_nic_vport_ctx) + 240;
4265 
4266 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4267 	token = mcx_cmdq_token(sc);
4268 	mcx_cmdq_init(sc, cqe, sizeof(*in) + insize, sizeof(*out), token);
4269 
4270 	in = mcx_cmdq_in(cqe);
4271 	in->cmd_opcode = htobe16(MCX_CMD_MODIFY_NIC_VPORT_CONTEXT);
4272 	in->cmd_op_mod = htobe16(0);
4273 	in->cmd_field_select = htobe32(
4274 	    MCX_CMD_MODIFY_NIC_VPORT_CONTEXT_FIELD_PROMISC |
4275 	    MCX_CMD_MODIFY_NIC_VPORT_CONTEXT_FIELD_MTU);
4276 
4277 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1, &cqe->cq_input_ptr, token) != 0) {
4278 		printf(", unable to allocate modify "
4279 		    "nic vport context mailboxen\n");
4280 		return (-1);
4281 	}
4282 	ctx = (struct mcx_nic_vport_ctx *)
4283 	    (((char *)mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0))) + 240);
4284 	ctx->vp_mtu = htobe32(sc->sc_hardmtu);
4285 	/*
4286          * always leave promisc-all enabled on the vport since we
4287          * can't give it a vlan list, and we're already doing multicast
4288          * filtering in the flow table.
4289 	 */
4290 	ctx->vp_flags = htobe16(MCX_NIC_VPORT_CTX_PROMISC_ALL);
4291 
4292 	mcx_cmdq_mboxes_sign(&mxm, 1);
4293 	mcx_cmdq_post(sc, cqe, 0);
4294 
4295 	error = mcx_cmdq_poll(sc, cqe, 1000);
4296 	if (error != 0) {
4297 		printf(", modify nic vport context timeout\n");
4298 		goto free;
4299 	}
4300 	if (mcx_cmdq_verify(cqe) != 0) {
4301 		printf(", modify nic vport context command corrupt\n");
4302 		goto free;
4303 	}
4304 
4305 	out = mcx_cmdq_out(cqe);
4306 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4307 		printf(", modify nic vport context failed (%x, %x)\n",
4308 		    out->cmd_status, be32toh(out->cmd_syndrome));
4309 		error = -1;
4310 		goto free;
4311 	}
4312 
4313 free:
4314 	mcx_dmamem_free(sc, &mxm);
4315 
4316 	return (error);
4317 }
4318 
4319 static int
4320 mcx_alloc_uar(struct mcx_softc *sc, int *uar)
4321 {
4322 	struct mcx_cmdq_entry *cqe;
4323 	struct mcx_cmd_alloc_uar_in *in;
4324 	struct mcx_cmd_alloc_uar_out *out;
4325 	int error;
4326 
4327 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4328 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
4329 
4330 	in = mcx_cmdq_in(cqe);
4331 	in->cmd_opcode = htobe16(MCX_CMD_ALLOC_UAR);
4332 	in->cmd_op_mod = htobe16(0);
4333 
4334 	mcx_cmdq_post(sc, cqe, 0);
4335 
4336 	error = mcx_cmdq_poll(sc, cqe, 1000);
4337 	if (error != 0) {
4338 		printf(", alloc uar timeout\n");
4339 		return (-1);
4340 	}
4341 	if (mcx_cmdq_verify(cqe) != 0) {
4342 		printf(", alloc uar command corrupt\n");
4343 		return (-1);
4344 	}
4345 
4346 	out = mcx_cmdq_out(cqe);
4347 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4348 		printf(", alloc uar failed (%x)\n", out->cmd_status);
4349 		return (-1);
4350 	}
4351 
4352 	*uar = mcx_get_id(out->cmd_uar);
4353 	return (0);
4354 }
4355 
4356 static int
4357 mcx_create_eq(struct mcx_softc *sc, struct mcx_eq *eq, int uar,
4358     uint64_t events, int vector)
4359 {
4360 	struct mcx_cmdq_entry *cqe;
4361 	struct mcx_dmamem mxm;
4362 	struct mcx_cmd_create_eq_in *in;
4363 	struct mcx_cmd_create_eq_mb_in *mbin;
4364 	struct mcx_cmd_create_eq_out *out;
4365 	struct mcx_eq_entry *eqe;
4366 	int error;
4367 	uint64_t *pas;
4368 	int insize, npages, paslen, i, token;
4369 
4370 	eq->eq_cons = 0;
4371 
4372 	npages = howmany((1 << MCX_LOG_EQ_SIZE) * sizeof(struct mcx_eq_entry),
4373 	    MCX_PAGE_SIZE);
4374 	paslen = npages * sizeof(*pas);
4375 	insize = sizeof(struct mcx_cmd_create_eq_mb_in) + paslen;
4376 
4377 	if (mcx_dmamem_alloc(sc, &eq->eq_mem, npages * MCX_PAGE_SIZE,
4378 	    MCX_PAGE_SIZE) != 0) {
4379 		printf(", unable to allocate event queue memory\n");
4380 		return (-1);
4381 	}
4382 
4383 	eqe = (struct mcx_eq_entry *)MCX_DMA_KVA(&eq->eq_mem);
4384 	for (i = 0; i < (1 << MCX_LOG_EQ_SIZE); i++) {
4385 		eqe[i].eq_owner = MCX_EQ_ENTRY_OWNER_INIT;
4386 	}
4387 
4388 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4389 	token = mcx_cmdq_token(sc);
4390 	mcx_cmdq_init(sc, cqe, sizeof(*in) + insize, sizeof(*out), token);
4391 
4392 	in = mcx_cmdq_in(cqe);
4393 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_EQ);
4394 	in->cmd_op_mod = htobe16(0);
4395 
4396 	if (mcx_cmdq_mboxes_alloc(sc, &mxm,
4397 	    howmany(insize, MCX_CMDQ_MAILBOX_DATASIZE),
4398 	    &cqe->cq_input_ptr, token) != 0) {
4399 		printf(", unable to allocate create eq mailboxen\n");
4400 		goto free_eq;
4401 	}
4402 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
4403 	mbin->cmd_eq_ctx.eq_uar_size = htobe32(
4404 	    (MCX_LOG_EQ_SIZE << MCX_EQ_CTX_LOG_EQ_SIZE_SHIFT) | uar);
4405 	mbin->cmd_eq_ctx.eq_intr = vector;
4406 	mbin->cmd_event_bitmask = htobe64(events);
4407 
4408 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&eq->eq_mem),
4409 	    0, MCX_DMA_LEN(&eq->eq_mem), BUS_DMASYNC_PREREAD);
4410 
4411 	/* physical addresses follow the mailbox in data */
4412 	mcx_cmdq_mboxes_pas(&mxm, sizeof(*mbin), npages, &eq->eq_mem);
4413 	mcx_cmdq_mboxes_sign(&mxm, howmany(insize, MCX_CMDQ_MAILBOX_DATASIZE));
4414 	mcx_cmdq_post(sc, cqe, 0);
4415 
4416 	error = mcx_cmdq_poll(sc, cqe, 1000);
4417 	if (error != 0) {
4418 		printf(", create eq timeout\n");
4419 		goto free_mxm;
4420 	}
4421 	if (mcx_cmdq_verify(cqe) != 0) {
4422 		printf(", create eq command corrupt\n");
4423 		goto free_mxm;
4424 	}
4425 
4426 	out = mcx_cmdq_out(cqe);
4427 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4428 		printf(", create eq failed (%x, %x)\n", out->cmd_status,
4429 		    be32toh(out->cmd_syndrome));
4430 		goto free_mxm;
4431 	}
4432 
4433 	eq->eq_n = mcx_get_id(out->cmd_eqn);
4434 
4435 	mcx_dmamem_free(sc, &mxm);
4436 
4437 	mcx_arm_eq(sc, eq, uar);
4438 
4439 	return (0);
4440 
4441 free_mxm:
4442 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&eq->eq_mem),
4443 	    0, MCX_DMA_LEN(&eq->eq_mem), BUS_DMASYNC_POSTREAD);
4444 	mcx_dmamem_free(sc, &mxm);
4445 free_eq:
4446 	mcx_dmamem_free(sc, &eq->eq_mem);
4447 	return (-1);
4448 }
4449 
4450 static int
4451 mcx_alloc_pd(struct mcx_softc *sc)
4452 {
4453 	struct mcx_cmdq_entry *cqe;
4454 	struct mcx_cmd_alloc_pd_in *in;
4455 	struct mcx_cmd_alloc_pd_out *out;
4456 	int error;
4457 
4458 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4459 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
4460 
4461 	in = mcx_cmdq_in(cqe);
4462 	in->cmd_opcode = htobe16(MCX_CMD_ALLOC_PD);
4463 	in->cmd_op_mod = htobe16(0);
4464 
4465 	mcx_cmdq_post(sc, cqe, 0);
4466 
4467 	error = mcx_cmdq_poll(sc, cqe, 1000);
4468 	if (error != 0) {
4469 		printf(", alloc pd timeout\n");
4470 		return (-1);
4471 	}
4472 	if (mcx_cmdq_verify(cqe) != 0) {
4473 		printf(", alloc pd command corrupt\n");
4474 		return (-1);
4475 	}
4476 
4477 	out = mcx_cmdq_out(cqe);
4478 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4479 		printf(", alloc pd failed (%x)\n", out->cmd_status);
4480 		return (-1);
4481 	}
4482 
4483 	sc->sc_pd = mcx_get_id(out->cmd_pd);
4484 	return (0);
4485 }
4486 
4487 static int
4488 mcx_alloc_tdomain(struct mcx_softc *sc)
4489 {
4490 	struct mcx_cmdq_entry *cqe;
4491 	struct mcx_cmd_alloc_td_in *in;
4492 	struct mcx_cmd_alloc_td_out *out;
4493 	int error;
4494 
4495 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4496 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
4497 
4498 	in = mcx_cmdq_in(cqe);
4499 	in->cmd_opcode = htobe16(MCX_CMD_ALLOC_TRANSPORT_DOMAIN);
4500 	in->cmd_op_mod = htobe16(0);
4501 
4502 	mcx_cmdq_post(sc, cqe, 0);
4503 
4504 	error = mcx_cmdq_poll(sc, cqe, 1000);
4505 	if (error != 0) {
4506 		printf(", alloc transport domain timeout\n");
4507 		return (-1);
4508 	}
4509 	if (mcx_cmdq_verify(cqe) != 0) {
4510 		printf(", alloc transport domain command corrupt\n");
4511 		return (-1);
4512 	}
4513 
4514 	out = mcx_cmdq_out(cqe);
4515 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4516 		printf(", alloc transport domain failed (%x)\n",
4517 		    out->cmd_status);
4518 		return (-1);
4519 	}
4520 
4521 	sc->sc_tdomain = mcx_get_id(out->cmd_tdomain);
4522 	return (0);
4523 }
4524 
4525 static int
4526 mcx_query_nic_vport_context(struct mcx_softc *sc, uint8_t *enaddr)
4527 {
4528 	struct mcx_dmamem mxm;
4529 	struct mcx_cmdq_entry *cqe;
4530 	struct mcx_cmd_query_nic_vport_context_in *in;
4531 	struct mcx_cmd_query_nic_vport_context_out *out;
4532 	struct mcx_nic_vport_ctx *ctx;
4533 	uint8_t *addr;
4534 	int error, token, i;
4535 
4536 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4537 	token = mcx_cmdq_token(sc);
4538 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out) + sizeof(*ctx), token);
4539 
4540 	in = mcx_cmdq_in(cqe);
4541 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_NIC_VPORT_CONTEXT);
4542 	in->cmd_op_mod = htobe16(0);
4543 	in->cmd_allowed_list_type = 0;
4544 
4545 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
4546 	    &cqe->cq_output_ptr, token) != 0) {
4547 		printf(", unable to allocate "
4548 		    "query nic vport context mailboxen\n");
4549 		return (-1);
4550 	}
4551 	mcx_cmdq_mboxes_sign(&mxm, 1);
4552 	mcx_cmdq_post(sc, cqe, 0);
4553 
4554 	error = mcx_cmdq_poll(sc, cqe, 1000);
4555 	if (error != 0) {
4556 		printf(", query nic vport context timeout\n");
4557 		goto free;
4558 	}
4559 	if (mcx_cmdq_verify(cqe) != 0) {
4560 		printf(", query nic vport context command corrupt\n");
4561 		goto free;
4562 	}
4563 
4564 	out = mcx_cmdq_out(cqe);
4565 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4566 		printf(", query nic vport context failed (%x, %x)\n",
4567 		    out->cmd_status, be32toh(out->cmd_syndrome));
4568 		error = -1;
4569 		goto free;
4570 	}
4571 
4572 	ctx = (struct mcx_nic_vport_ctx *)
4573 	    mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
4574 	addr = (uint8_t *)&ctx->vp_perm_addr;
4575 	for (i = 0; i < ETHER_ADDR_LEN; i++) {
4576 		enaddr[i] = addr[i + 2];
4577 	}
4578 free:
4579 	mcx_dmamem_free(sc, &mxm);
4580 
4581 	return (error);
4582 }
4583 
4584 static int
4585 mcx_query_special_contexts(struct mcx_softc *sc)
4586 {
4587 	struct mcx_cmdq_entry *cqe;
4588 	struct mcx_cmd_query_special_ctx_in *in;
4589 	struct mcx_cmd_query_special_ctx_out *out;
4590 	int error;
4591 
4592 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4593 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
4594 
4595 	in = mcx_cmdq_in(cqe);
4596 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_SPECIAL_CONTEXTS);
4597 	in->cmd_op_mod = htobe16(0);
4598 
4599 	mcx_cmdq_post(sc, cqe, 0);
4600 
4601 	error = mcx_cmdq_poll(sc, cqe, 1000);
4602 	if (error != 0) {
4603 		printf(", query special contexts timeout\n");
4604 		return (-1);
4605 	}
4606 	if (mcx_cmdq_verify(cqe) != 0) {
4607 		printf(", query special contexts command corrupt\n");
4608 		return (-1);
4609 	}
4610 
4611 	out = mcx_cmdq_out(cqe);
4612 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4613 		printf(", query special contexts failed (%x)\n",
4614 		    out->cmd_status);
4615 		return (-1);
4616 	}
4617 
4618 	sc->sc_lkey = be32toh(out->cmd_resd_lkey);
4619 	return (0);
4620 }
4621 
4622 static int
4623 mcx_set_port_mtu(struct mcx_softc *sc, int mtu)
4624 {
4625 	struct mcx_reg_pmtu pmtu;
4626 	int error;
4627 
4628 	/* read max mtu */
4629 	memset(&pmtu, 0, sizeof(pmtu));
4630 	pmtu.rp_local_port = 1;
4631 	error = mcx_access_hca_reg(sc, MCX_REG_PMTU, MCX_REG_OP_READ, &pmtu,
4632 	    sizeof(pmtu));
4633 	if (error != 0) {
4634 		printf(", unable to get port MTU\n");
4635 		return error;
4636 	}
4637 
4638 	mtu = uimin(mtu, be16toh(pmtu.rp_max_mtu));
4639 	pmtu.rp_admin_mtu = htobe16(mtu);
4640 	error = mcx_access_hca_reg(sc, MCX_REG_PMTU, MCX_REG_OP_WRITE, &pmtu,
4641 	    sizeof(pmtu));
4642 	if (error != 0) {
4643 		printf(", unable to set port MTU\n");
4644 		return error;
4645 	}
4646 
4647 	sc->sc_hardmtu = mtu;
4648 	sc->sc_rxbufsz = roundup(mtu + ETHER_ALIGN, sizeof(long));
4649 	return 0;
4650 }
4651 
4652 static int
4653 mcx_create_cq(struct mcx_softc *sc, struct mcx_cq *cq, int uar, int db, int eqn)
4654 {
4655 	struct mcx_cmdq_entry *cmde;
4656 	struct mcx_cq_entry *cqe;
4657 	struct mcx_dmamem mxm;
4658 	struct mcx_cmd_create_cq_in *in;
4659 	struct mcx_cmd_create_cq_mb_in *mbin;
4660 	struct mcx_cmd_create_cq_out *out;
4661 	int error;
4662 	uint64_t *pas;
4663 	int insize, npages, paslen, i, token;
4664 
4665 	cq->cq_doorbell = MCX_CQ_DOORBELL_BASE + (MCX_CQ_DOORBELL_STRIDE * db);
4666 
4667 	npages = howmany((1 << MCX_LOG_CQ_SIZE) * sizeof(struct mcx_cq_entry),
4668 	    MCX_PAGE_SIZE);
4669 	paslen = npages * sizeof(*pas);
4670 	insize = sizeof(struct mcx_cmd_create_cq_mb_in) + paslen;
4671 
4672 	if (mcx_dmamem_alloc(sc, &cq->cq_mem, npages * MCX_PAGE_SIZE,
4673 	    MCX_PAGE_SIZE) != 0) {
4674 		printf("%s: unable to allocate completion queue memory\n",
4675 		    DEVNAME(sc));
4676 		return (-1);
4677 	}
4678 	cqe = MCX_DMA_KVA(&cq->cq_mem);
4679 	for (i = 0; i < (1 << MCX_LOG_CQ_SIZE); i++) {
4680 		cqe[i].cq_opcode_owner = MCX_CQ_ENTRY_FLAG_OWNER;
4681 	}
4682 
4683 	cmde = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4684 	token = mcx_cmdq_token(sc);
4685 	mcx_cmdq_init(sc, cmde, sizeof(*in) + insize, sizeof(*out), token);
4686 
4687 	in = mcx_cmdq_in(cmde);
4688 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_CQ);
4689 	in->cmd_op_mod = htobe16(0);
4690 
4691 	if (mcx_cmdq_mboxes_alloc(sc, &mxm,
4692 	    howmany(insize, MCX_CMDQ_MAILBOX_DATASIZE),
4693 	    &cmde->cq_input_ptr, token) != 0) {
4694 		printf("%s: unable to allocate create cq mailboxen\n",
4695 		    DEVNAME(sc));
4696 		goto free_cq;
4697 	}
4698 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
4699 	mbin->cmd_cq_ctx.cq_uar_size = htobe32(
4700 	    (MCX_LOG_CQ_SIZE << MCX_CQ_CTX_LOG_CQ_SIZE_SHIFT) | uar);
4701 	mbin->cmd_cq_ctx.cq_eqn = htobe32(eqn);
4702 	mbin->cmd_cq_ctx.cq_period_max_count = htobe32(
4703 	    (MCX_CQ_MOD_PERIOD << MCX_CQ_CTX_PERIOD_SHIFT) |
4704 	    MCX_CQ_MOD_COUNTER);
4705 	mbin->cmd_cq_ctx.cq_doorbell = htobe64(
4706 	    MCX_DMA_DVA(&sc->sc_doorbell_mem) + cq->cq_doorbell);
4707 
4708 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&cq->cq_mem),
4709 	    0, MCX_DMA_LEN(&cq->cq_mem), BUS_DMASYNC_PREREAD);
4710 
4711 	/* physical addresses follow the mailbox in data */
4712 	mcx_cmdq_mboxes_pas(&mxm, sizeof(*mbin), npages, &cq->cq_mem);
4713 	mcx_cmdq_post(sc, cmde, 0);
4714 
4715 	error = mcx_cmdq_poll(sc, cmde, 1000);
4716 	if (error != 0) {
4717 		printf("%s: create cq timeout\n", DEVNAME(sc));
4718 		goto free_mxm;
4719 	}
4720 	if (mcx_cmdq_verify(cmde) != 0) {
4721 		printf("%s: create cq command corrupt\n", DEVNAME(sc));
4722 		goto free_mxm;
4723 	}
4724 
4725 	out = mcx_cmdq_out(cmde);
4726 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4727 		printf("%s: create cq failed (%x, %x)\n", DEVNAME(sc),
4728 		    out->cmd_status, be32toh(out->cmd_syndrome));
4729 		goto free_mxm;
4730 	}
4731 
4732 	cq->cq_n = mcx_get_id(out->cmd_cqn);
4733 	cq->cq_cons = 0;
4734 	cq->cq_count = 0;
4735 
4736 	mcx_dmamem_free(sc, &mxm);
4737 
4738 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_doorbell_mem),
4739 	    cq->cq_doorbell, sizeof(struct mcx_cq_doorbell),
4740 	    BUS_DMASYNC_PREWRITE);
4741 
4742 	mcx_arm_cq(sc, cq, uar);
4743 
4744 	return (0);
4745 
4746 free_mxm:
4747 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&cq->cq_mem),
4748 	    0, MCX_DMA_LEN(&cq->cq_mem), BUS_DMASYNC_POSTREAD);
4749 	mcx_dmamem_free(sc, &mxm);
4750 free_cq:
4751 	mcx_dmamem_free(sc, &cq->cq_mem);
4752 	return (-1);
4753 }
4754 
4755 static int
4756 mcx_destroy_cq(struct mcx_softc *sc, struct mcx_cq *cq)
4757 {
4758 	struct mcx_cmdq_entry *cqe;
4759 	struct mcx_cmd_destroy_cq_in *in;
4760 	struct mcx_cmd_destroy_cq_out *out;
4761 	int error;
4762 	int token;
4763 
4764 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4765 	token = mcx_cmdq_token(sc);
4766 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), token);
4767 
4768 	in = mcx_cmdq_in(cqe);
4769 	in->cmd_opcode = htobe16(MCX_CMD_DESTROY_CQ);
4770 	in->cmd_op_mod = htobe16(0);
4771 	in->cmd_cqn = htobe32(cq->cq_n);
4772 
4773 	mcx_cmdq_post(sc, cqe, 0);
4774 	error = mcx_cmdq_poll(sc, cqe, 1000);
4775 	if (error != 0) {
4776 		printf("%s: destroy cq timeout\n", DEVNAME(sc));
4777 		return error;
4778 	}
4779 	if (mcx_cmdq_verify(cqe) != 0) {
4780 		printf("%s: destroy cq command corrupt\n", DEVNAME(sc));
4781 		return error;
4782 	}
4783 
4784 	out = mcx_cmdq_out(cqe);
4785 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4786 		printf("%s: destroy cq failed (%x, %x)\n", DEVNAME(sc),
4787 		    out->cmd_status, be32toh(out->cmd_syndrome));
4788 		return -1;
4789 	}
4790 
4791 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_doorbell_mem),
4792 	    cq->cq_doorbell, sizeof(struct mcx_cq_doorbell),
4793 	    BUS_DMASYNC_POSTWRITE);
4794 
4795 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&cq->cq_mem),
4796 	    0, MCX_DMA_LEN(&cq->cq_mem), BUS_DMASYNC_POSTREAD);
4797 	mcx_dmamem_free(sc, &cq->cq_mem);
4798 
4799 	cq->cq_n = 0;
4800 	cq->cq_cons = 0;
4801 	cq->cq_count = 0;
4802 	return 0;
4803 }
4804 
4805 static int
4806 mcx_create_rq(struct mcx_softc *sc, struct mcx_rx *rx, int db, int cqn)
4807 {
4808 	struct mcx_cmdq_entry *cqe;
4809 	struct mcx_dmamem mxm;
4810 	struct mcx_cmd_create_rq_in *in;
4811 	struct mcx_cmd_create_rq_out *out;
4812 	struct mcx_rq_ctx *mbin;
4813 	int error;
4814 	uint64_t *pas;
4815 	uint32_t rq_flags;
4816 	int insize, npages, paslen, token;
4817 
4818 	rx->rx_doorbell = MCX_WQ_DOORBELL_BASE +
4819 	    (db * MCX_WQ_DOORBELL_STRIDE);
4820 
4821 	npages = howmany((1 << MCX_LOG_RQ_SIZE) * sizeof(struct mcx_rq_entry),
4822 	    MCX_PAGE_SIZE);
4823 	paslen = npages * sizeof(*pas);
4824 	insize = 0x10 + sizeof(struct mcx_rq_ctx) + paslen;
4825 
4826 	if (mcx_dmamem_alloc(sc, &rx->rx_rq_mem, npages * MCX_PAGE_SIZE,
4827 	    MCX_PAGE_SIZE) != 0) {
4828 		printf("%s: unable to allocate receive queue memory\n",
4829 		    DEVNAME(sc));
4830 		return (-1);
4831 	}
4832 
4833 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4834 	token = mcx_cmdq_token(sc);
4835 	mcx_cmdq_init(sc, cqe, sizeof(*in) + insize, sizeof(*out), token);
4836 
4837 	in = mcx_cmdq_in(cqe);
4838 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_RQ);
4839 	in->cmd_op_mod = htobe16(0);
4840 
4841 	if (mcx_cmdq_mboxes_alloc(sc, &mxm,
4842 	    howmany(insize, MCX_CMDQ_MAILBOX_DATASIZE),
4843 	    &cqe->cq_input_ptr, token) != 0) {
4844 		printf("%s: unable to allocate create rq mailboxen\n",
4845 		    DEVNAME(sc));
4846 		goto free_rq;
4847 	}
4848 	mbin = (struct mcx_rq_ctx *)
4849 	    (((char *)mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0))) + 0x10);
4850 	rq_flags = MCX_RQ_CTX_RLKEY;
4851 	mbin->rq_flags = htobe32(rq_flags);
4852 	mbin->rq_cqn = htobe32(cqn);
4853 	mbin->rq_wq.wq_type = MCX_WQ_CTX_TYPE_CYCLIC;
4854 	mbin->rq_wq.wq_pd = htobe32(sc->sc_pd);
4855 	mbin->rq_wq.wq_doorbell = htobe64(MCX_DMA_DVA(&sc->sc_doorbell_mem) +
4856 	    rx->rx_doorbell);
4857 	mbin->rq_wq.wq_log_stride = htobe16(4);
4858 	mbin->rq_wq.wq_log_size = MCX_LOG_RQ_SIZE;
4859 
4860 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&rx->rx_rq_mem),
4861 	    0, MCX_DMA_LEN(&rx->rx_rq_mem), BUS_DMASYNC_PREWRITE);
4862 
4863 	/* physical addresses follow the mailbox in data */
4864 	mcx_cmdq_mboxes_pas(&mxm, sizeof(*mbin) + 0x10, npages, &rx->rx_rq_mem);
4865 	mcx_cmdq_post(sc, cqe, 0);
4866 
4867 	error = mcx_cmdq_poll(sc, cqe, 1000);
4868 	if (error != 0) {
4869 		printf("%s: create rq timeout\n", DEVNAME(sc));
4870 		goto free_mxm;
4871 	}
4872 	if (mcx_cmdq_verify(cqe) != 0) {
4873 		printf("%s: create rq command corrupt\n", DEVNAME(sc));
4874 		goto free_mxm;
4875 	}
4876 
4877 	out = mcx_cmdq_out(cqe);
4878 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4879 		printf("%s: create rq failed (%x, %x)\n", DEVNAME(sc),
4880 		    out->cmd_status, be32toh(out->cmd_syndrome));
4881 		goto free_mxm;
4882 	}
4883 
4884 	rx->rx_rqn = mcx_get_id(out->cmd_rqn);
4885 
4886 	mcx_dmamem_free(sc, &mxm);
4887 
4888 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_doorbell_mem),
4889 	    rx->rx_doorbell, sizeof(uint32_t), BUS_DMASYNC_PREWRITE);
4890 
4891 	return (0);
4892 
4893 free_mxm:
4894 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&rx->rx_rq_mem),
4895 	    0, MCX_DMA_LEN(&rx->rx_rq_mem), BUS_DMASYNC_POSTWRITE);
4896 	mcx_dmamem_free(sc, &mxm);
4897 free_rq:
4898 	mcx_dmamem_free(sc, &rx->rx_rq_mem);
4899 	return (-1);
4900 }
4901 
4902 static int
4903 mcx_ready_rq(struct mcx_softc *sc, struct mcx_rx *rx)
4904 {
4905 	struct mcx_cmdq_entry *cqe;
4906 	struct mcx_dmamem mxm;
4907 	struct mcx_cmd_modify_rq_in *in;
4908 	struct mcx_cmd_modify_rq_mb_in *mbin;
4909 	struct mcx_cmd_modify_rq_out *out;
4910 	int error;
4911 	int token;
4912 
4913 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4914 	token = mcx_cmdq_token(sc);
4915 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
4916 	    sizeof(*out), token);
4917 
4918 	in = mcx_cmdq_in(cqe);
4919 	in->cmd_opcode = htobe16(MCX_CMD_MODIFY_RQ);
4920 	in->cmd_op_mod = htobe16(0);
4921 	in->cmd_rq_state = htobe32((MCX_QUEUE_STATE_RST << 28) | rx->rx_rqn);
4922 
4923 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
4924 	    &cqe->cq_input_ptr, token) != 0) {
4925 		printf("%s: unable to allocate modify rq mailbox\n",
4926 		    DEVNAME(sc));
4927 		return (-1);
4928 	}
4929 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
4930 	mbin->cmd_rq_ctx.rq_flags = htobe32(
4931 	    MCX_QUEUE_STATE_RDY << MCX_RQ_CTX_STATE_SHIFT);
4932 
4933 	mcx_cmdq_mboxes_sign(&mxm, 1);
4934 	mcx_cmdq_post(sc, cqe, 0);
4935 	error = mcx_cmdq_poll(sc, cqe, 1000);
4936 	if (error != 0) {
4937 		printf("%s: modify rq timeout\n", DEVNAME(sc));
4938 		goto free;
4939 	}
4940 	if (mcx_cmdq_verify(cqe) != 0) {
4941 		printf("%s: modify rq command corrupt\n", DEVNAME(sc));
4942 		goto free;
4943 	}
4944 
4945 	out = mcx_cmdq_out(cqe);
4946 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4947 		printf("%s: modify rq failed (%x, %x)\n", DEVNAME(sc),
4948 		    out->cmd_status, be32toh(out->cmd_syndrome));
4949 		error = -1;
4950 		goto free;
4951 	}
4952 
4953 free:
4954 	mcx_dmamem_free(sc, &mxm);
4955 	return (error);
4956 }
4957 
4958 static int
4959 mcx_destroy_rq(struct mcx_softc *sc, struct mcx_rx *rx)
4960 {
4961 	struct mcx_cmdq_entry *cqe;
4962 	struct mcx_cmd_destroy_rq_in *in;
4963 	struct mcx_cmd_destroy_rq_out *out;
4964 	int error;
4965 	int token;
4966 
4967 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4968 	token = mcx_cmdq_token(sc);
4969 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), token);
4970 
4971 	in = mcx_cmdq_in(cqe);
4972 	in->cmd_opcode = htobe16(MCX_CMD_DESTROY_RQ);
4973 	in->cmd_op_mod = htobe16(0);
4974 	in->cmd_rqn = htobe32(rx->rx_rqn);
4975 
4976 	mcx_cmdq_post(sc, cqe, 0);
4977 	error = mcx_cmdq_poll(sc, cqe, 1000);
4978 	if (error != 0) {
4979 		printf("%s: destroy rq timeout\n", DEVNAME(sc));
4980 		return error;
4981 	}
4982 	if (mcx_cmdq_verify(cqe) != 0) {
4983 		printf("%s: destroy rq command corrupt\n", DEVNAME(sc));
4984 		return error;
4985 	}
4986 
4987 	out = mcx_cmdq_out(cqe);
4988 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4989 		printf("%s: destroy rq failed (%x, %x)\n", DEVNAME(sc),
4990 		    out->cmd_status, be32toh(out->cmd_syndrome));
4991 		return -1;
4992 	}
4993 
4994 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_doorbell_mem),
4995 	    rx->rx_doorbell, sizeof(uint32_t), BUS_DMASYNC_POSTWRITE);
4996 
4997 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&rx->rx_rq_mem),
4998 	    0, MCX_DMA_LEN(&rx->rx_rq_mem), BUS_DMASYNC_POSTWRITE);
4999 	mcx_dmamem_free(sc, &rx->rx_rq_mem);
5000 
5001 	rx->rx_rqn = 0;
5002 	return 0;
5003 }
5004 
5005 static int
5006 mcx_create_tir_direct(struct mcx_softc *sc, struct mcx_rx *rx, int *tirn)
5007 {
5008 	struct mcx_cmdq_entry *cqe;
5009 	struct mcx_dmamem mxm;
5010 	struct mcx_cmd_create_tir_in *in;
5011 	struct mcx_cmd_create_tir_mb_in *mbin;
5012 	struct mcx_cmd_create_tir_out *out;
5013 	int error;
5014 	int token;
5015 
5016 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5017 	token = mcx_cmdq_token(sc);
5018 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
5019 	    sizeof(*out), token);
5020 
5021 	in = mcx_cmdq_in(cqe);
5022 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_TIR);
5023 	in->cmd_op_mod = htobe16(0);
5024 
5025 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
5026 	    &cqe->cq_input_ptr, token) != 0) {
5027 		printf("%s: unable to allocate create tir mailbox\n",
5028 		    DEVNAME(sc));
5029 		return (-1);
5030 	}
5031 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5032 	/* leave disp_type = 0, so packets get sent to the inline rqn */
5033 	mbin->cmd_inline_rqn = htobe32(rx->rx_rqn);
5034 	mbin->cmd_tdomain = htobe32(sc->sc_tdomain);
5035 
5036 	mcx_cmdq_post(sc, cqe, 0);
5037 	error = mcx_cmdq_poll(sc, cqe, 1000);
5038 	if (error != 0) {
5039 		printf("%s: create tir timeout\n", DEVNAME(sc));
5040 		goto free;
5041 	}
5042 	if (mcx_cmdq_verify(cqe) != 0) {
5043 		printf("%s: create tir command corrupt\n", DEVNAME(sc));
5044 		goto free;
5045 	}
5046 
5047 	out = mcx_cmdq_out(cqe);
5048 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5049 		printf("%s: create tir failed (%x, %x)\n", DEVNAME(sc),
5050 		    out->cmd_status, be32toh(out->cmd_syndrome));
5051 		error = -1;
5052 		goto free;
5053 	}
5054 
5055 	*tirn = mcx_get_id(out->cmd_tirn);
5056 free:
5057 	mcx_dmamem_free(sc, &mxm);
5058 	return (error);
5059 }
5060 
5061 static int
5062 mcx_create_tir_indirect(struct mcx_softc *sc, int rqtn, uint32_t hash_sel,
5063     int *tirn)
5064 {
5065 	struct mcx_cmdq_entry *cqe;
5066 	struct mcx_dmamem mxm;
5067 	struct mcx_cmd_create_tir_in *in;
5068 	struct mcx_cmd_create_tir_mb_in *mbin;
5069 	struct mcx_cmd_create_tir_out *out;
5070 	int error;
5071 	int token;
5072 
5073 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5074 	token = mcx_cmdq_token(sc);
5075 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
5076 	    sizeof(*out), token);
5077 
5078 	in = mcx_cmdq_in(cqe);
5079 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_TIR);
5080 	in->cmd_op_mod = htobe16(0);
5081 
5082 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
5083 	    &cqe->cq_input_ptr, token) != 0) {
5084 		printf("%s: unable to allocate create tir mailbox\n",
5085 		    DEVNAME(sc));
5086 		return (-1);
5087 	}
5088 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5089 	mbin->cmd_disp_type = htobe32(MCX_TIR_CTX_DISP_TYPE_INDIRECT
5090 	    << MCX_TIR_CTX_DISP_TYPE_SHIFT);
5091 	mbin->cmd_indir_table = htobe32(rqtn);
5092 	mbin->cmd_tdomain = htobe32(sc->sc_tdomain |
5093 	    MCX_TIR_CTX_HASH_TOEPLITZ << MCX_TIR_CTX_HASH_SHIFT);
5094 	mbin->cmd_rx_hash_sel_outer = htobe32(hash_sel);
5095 	stoeplitz_to_key(&mbin->cmd_rx_hash_key,
5096 	    sizeof(mbin->cmd_rx_hash_key));
5097 
5098 	mcx_cmdq_post(sc, cqe, 0);
5099 	error = mcx_cmdq_poll(sc, cqe, 1000);
5100 	if (error != 0) {
5101 		printf("%s: create tir timeout\n", DEVNAME(sc));
5102 		goto free;
5103 	}
5104 	if (mcx_cmdq_verify(cqe) != 0) {
5105 		printf("%s: create tir command corrupt\n", DEVNAME(sc));
5106 		goto free;
5107 	}
5108 
5109 	out = mcx_cmdq_out(cqe);
5110 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5111 		printf("%s: create tir failed (%x, %x)\n", DEVNAME(sc),
5112 		    out->cmd_status, be32toh(out->cmd_syndrome));
5113 		error = -1;
5114 		goto free;
5115 	}
5116 
5117 	*tirn = mcx_get_id(out->cmd_tirn);
5118 free:
5119 	mcx_dmamem_free(sc, &mxm);
5120 	return (error);
5121 }
5122 
5123 static int
5124 mcx_destroy_tir(struct mcx_softc *sc, int tirn)
5125 {
5126 	struct mcx_cmdq_entry *cqe;
5127 	struct mcx_cmd_destroy_tir_in *in;
5128 	struct mcx_cmd_destroy_tir_out *out;
5129 	int error;
5130 	int token;
5131 
5132 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5133 	token = mcx_cmdq_token(sc);
5134 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), token);
5135 
5136 	in = mcx_cmdq_in(cqe);
5137 	in->cmd_opcode = htobe16(MCX_CMD_DESTROY_TIR);
5138 	in->cmd_op_mod = htobe16(0);
5139 	in->cmd_tirn = htobe32(tirn);
5140 
5141 	mcx_cmdq_post(sc, cqe, 0);
5142 	error = mcx_cmdq_poll(sc, cqe, 1000);
5143 	if (error != 0) {
5144 		printf("%s: destroy tir timeout\n", DEVNAME(sc));
5145 		return error;
5146 	}
5147 	if (mcx_cmdq_verify(cqe) != 0) {
5148 		printf("%s: destroy tir command corrupt\n", DEVNAME(sc));
5149 		return error;
5150 	}
5151 
5152 	out = mcx_cmdq_out(cqe);
5153 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5154 		printf("%s: destroy tir failed (%x, %x)\n", DEVNAME(sc),
5155 		    out->cmd_status, be32toh(out->cmd_syndrome));
5156 		return -1;
5157 	}
5158 
5159 	return (0);
5160 }
5161 
5162 static int
5163 mcx_create_sq(struct mcx_softc *sc, struct mcx_tx *tx, int uar, int db,
5164     int cqn)
5165 {
5166 	struct mcx_cmdq_entry *cqe;
5167 	struct mcx_dmamem mxm;
5168 	struct mcx_cmd_create_sq_in *in;
5169 	struct mcx_sq_ctx *mbin;
5170 	struct mcx_cmd_create_sq_out *out;
5171 	int error;
5172 	uint64_t *pas;
5173 	int insize, npages, paslen, token;
5174 
5175 	tx->tx_doorbell = MCX_WQ_DOORBELL_BASE +
5176 	    (db * MCX_WQ_DOORBELL_STRIDE) + 4;
5177 
5178 	npages = howmany((1 << MCX_LOG_SQ_SIZE) * sizeof(struct mcx_sq_entry),
5179 	    MCX_PAGE_SIZE);
5180 	paslen = npages * sizeof(*pas);
5181 	insize = sizeof(struct mcx_sq_ctx) + paslen;
5182 
5183 	if (mcx_dmamem_alloc(sc, &tx->tx_sq_mem, npages * MCX_PAGE_SIZE,
5184 	    MCX_PAGE_SIZE) != 0) {
5185 		printf("%s: unable to allocate send queue memory\n",
5186 		    DEVNAME(sc));
5187 		return (-1);
5188 	}
5189 
5190 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5191 	token = mcx_cmdq_token(sc);
5192 	mcx_cmdq_init(sc, cqe, sizeof(*in) + insize + paslen, sizeof(*out),
5193 	    token);
5194 
5195 	in = mcx_cmdq_in(cqe);
5196 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_SQ);
5197 	in->cmd_op_mod = htobe16(0);
5198 
5199 	if (mcx_cmdq_mboxes_alloc(sc, &mxm,
5200 	    howmany(insize, MCX_CMDQ_MAILBOX_DATASIZE),
5201 	    &cqe->cq_input_ptr, token) != 0) {
5202 		printf("%s: unable to allocate create sq mailboxen\n",
5203 		    DEVNAME(sc));
5204 		goto free_sq;
5205 	}
5206 	mbin = (struct mcx_sq_ctx *)
5207 	    (((char *)mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0))) + 0x10);
5208 	mbin->sq_flags = htobe32(MCX_SQ_CTX_RLKEY |
5209 	    (1 << MCX_SQ_CTX_MIN_WQE_INLINE_SHIFT));
5210 	mbin->sq_cqn = htobe32(cqn);
5211 	mbin->sq_tis_lst_sz = htobe32(1 << MCX_SQ_CTX_TIS_LST_SZ_SHIFT);
5212 	mbin->sq_tis_num = htobe32(sc->sc_tis);
5213 	mbin->sq_wq.wq_type = MCX_WQ_CTX_TYPE_CYCLIC;
5214 	mbin->sq_wq.wq_pd = htobe32(sc->sc_pd);
5215 	mbin->sq_wq.wq_uar_page = htobe32(uar);
5216 	mbin->sq_wq.wq_doorbell = htobe64(MCX_DMA_DVA(&sc->sc_doorbell_mem) +
5217 	    tx->tx_doorbell);
5218 	mbin->sq_wq.wq_log_stride = htobe16(MCX_LOG_SQ_ENTRY_SIZE);
5219 	mbin->sq_wq.wq_log_size = MCX_LOG_SQ_SIZE;
5220 
5221 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&tx->tx_sq_mem),
5222 	    0, MCX_DMA_LEN(&tx->tx_sq_mem), BUS_DMASYNC_PREWRITE);
5223 
5224 	/* physical addresses follow the mailbox in data */
5225 	mcx_cmdq_mboxes_pas(&mxm, sizeof(*mbin) + 0x10,
5226 	    npages, &tx->tx_sq_mem);
5227 	mcx_cmdq_post(sc, cqe, 0);
5228 
5229 	error = mcx_cmdq_poll(sc, cqe, 1000);
5230 	if (error != 0) {
5231 		printf("%s: create sq timeout\n", DEVNAME(sc));
5232 		goto free_mxm;
5233 	}
5234 	if (mcx_cmdq_verify(cqe) != 0) {
5235 		printf("%s: create sq command corrupt\n", DEVNAME(sc));
5236 		goto free_mxm;
5237 	}
5238 
5239 	out = mcx_cmdq_out(cqe);
5240 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5241 		printf("%s: create sq failed (%x, %x)\n", DEVNAME(sc),
5242 		    out->cmd_status, be32toh(out->cmd_syndrome));
5243 		goto free_mxm;
5244 	}
5245 
5246 	tx->tx_uar = uar;
5247 	tx->tx_sqn = mcx_get_id(out->cmd_sqn);
5248 
5249 	mcx_dmamem_free(sc, &mxm);
5250 
5251 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_doorbell_mem),
5252 	    tx->tx_doorbell, sizeof(uint32_t), BUS_DMASYNC_PREWRITE);
5253 
5254 	return (0);
5255 
5256 free_mxm:
5257 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&tx->tx_sq_mem),
5258 	    0, MCX_DMA_LEN(&tx->tx_sq_mem), BUS_DMASYNC_POSTWRITE);
5259 	mcx_dmamem_free(sc, &mxm);
5260 free_sq:
5261 	mcx_dmamem_free(sc, &tx->tx_sq_mem);
5262 	return (-1);
5263 }
5264 
5265 static int
5266 mcx_destroy_sq(struct mcx_softc *sc, struct mcx_tx *tx)
5267 {
5268 	struct mcx_cmdq_entry *cqe;
5269 	struct mcx_cmd_destroy_sq_in *in;
5270 	struct mcx_cmd_destroy_sq_out *out;
5271 	int error;
5272 	int token;
5273 
5274 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5275 	token = mcx_cmdq_token(sc);
5276 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), token);
5277 
5278 	in = mcx_cmdq_in(cqe);
5279 	in->cmd_opcode = htobe16(MCX_CMD_DESTROY_SQ);
5280 	in->cmd_op_mod = htobe16(0);
5281 	in->cmd_sqn = htobe32(tx->tx_sqn);
5282 
5283 	mcx_cmdq_post(sc, cqe, 0);
5284 	error = mcx_cmdq_poll(sc, cqe, 1000);
5285 	if (error != 0) {
5286 		printf("%s: destroy sq timeout\n", DEVNAME(sc));
5287 		return error;
5288 	}
5289 	if (mcx_cmdq_verify(cqe) != 0) {
5290 		printf("%s: destroy sq command corrupt\n", DEVNAME(sc));
5291 		return error;
5292 	}
5293 
5294 	out = mcx_cmdq_out(cqe);
5295 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5296 		printf("%s: destroy sq failed (%x, %x)\n", DEVNAME(sc),
5297 		    out->cmd_status, be32toh(out->cmd_syndrome));
5298 		return -1;
5299 	}
5300 
5301 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_doorbell_mem),
5302 	    tx->tx_doorbell, sizeof(uint32_t), BUS_DMASYNC_POSTWRITE);
5303 
5304 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&tx->tx_sq_mem),
5305 	    0, MCX_DMA_LEN(&tx->tx_sq_mem), BUS_DMASYNC_POSTWRITE);
5306 	mcx_dmamem_free(sc, &tx->tx_sq_mem);
5307 
5308 	tx->tx_sqn = 0;
5309 	return 0;
5310 }
5311 
5312 static int
5313 mcx_ready_sq(struct mcx_softc *sc, struct mcx_tx *tx)
5314 {
5315 	struct mcx_cmdq_entry *cqe;
5316 	struct mcx_dmamem mxm;
5317 	struct mcx_cmd_modify_sq_in *in;
5318 	struct mcx_cmd_modify_sq_mb_in *mbin;
5319 	struct mcx_cmd_modify_sq_out *out;
5320 	int error;
5321 	int token;
5322 
5323 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5324 	token = mcx_cmdq_token(sc);
5325 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
5326 	    sizeof(*out), token);
5327 
5328 	in = mcx_cmdq_in(cqe);
5329 	in->cmd_opcode = htobe16(MCX_CMD_MODIFY_SQ);
5330 	in->cmd_op_mod = htobe16(0);
5331 	in->cmd_sq_state = htobe32((MCX_QUEUE_STATE_RST << 28) | tx->tx_sqn);
5332 
5333 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
5334 	    &cqe->cq_input_ptr, token) != 0) {
5335 		printf("%s: unable to allocate modify sq mailbox\n",
5336 		    DEVNAME(sc));
5337 		return (-1);
5338 	}
5339 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5340 	mbin->cmd_sq_ctx.sq_flags = htobe32(
5341 	    MCX_QUEUE_STATE_RDY << MCX_SQ_CTX_STATE_SHIFT);
5342 
5343 	mcx_cmdq_mboxes_sign(&mxm, 1);
5344 	mcx_cmdq_post(sc, cqe, 0);
5345 	error = mcx_cmdq_poll(sc, cqe, 1000);
5346 	if (error != 0) {
5347 		printf("%s: modify sq timeout\n", DEVNAME(sc));
5348 		goto free;
5349 	}
5350 	if (mcx_cmdq_verify(cqe) != 0) {
5351 		printf("%s: modify sq command corrupt\n", DEVNAME(sc));
5352 		goto free;
5353 	}
5354 
5355 	out = mcx_cmdq_out(cqe);
5356 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5357 		printf("%s: modify sq failed (%x, %x)\n", DEVNAME(sc),
5358 		    out->cmd_status, be32toh(out->cmd_syndrome));
5359 		error = -1;
5360 		goto free;
5361 	}
5362 
5363 free:
5364 	mcx_dmamem_free(sc, &mxm);
5365 	return (error);
5366 }
5367 
5368 static int
5369 mcx_create_tis(struct mcx_softc *sc, int *tis)
5370 {
5371 	struct mcx_cmdq_entry *cqe;
5372 	struct mcx_dmamem mxm;
5373 	struct mcx_cmd_create_tis_in *in;
5374 	struct mcx_cmd_create_tis_mb_in *mbin;
5375 	struct mcx_cmd_create_tis_out *out;
5376 	int error;
5377 	int token;
5378 
5379 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5380 	token = mcx_cmdq_token(sc);
5381 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
5382 	    sizeof(*out), token);
5383 
5384 	in = mcx_cmdq_in(cqe);
5385 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_TIS);
5386 	in->cmd_op_mod = htobe16(0);
5387 
5388 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
5389 	    &cqe->cq_input_ptr, token) != 0) {
5390 		printf("%s: unable to allocate create tis mailbox\n",
5391 		    DEVNAME(sc));
5392 		return (-1);
5393 	}
5394 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5395 	mbin->cmd_tdomain = htobe32(sc->sc_tdomain);
5396 
5397 	mcx_cmdq_mboxes_sign(&mxm, 1);
5398 	mcx_cmdq_post(sc, cqe, 0);
5399 	error = mcx_cmdq_poll(sc, cqe, 1000);
5400 	if (error != 0) {
5401 		printf("%s: create tis timeout\n", DEVNAME(sc));
5402 		goto free;
5403 	}
5404 	if (mcx_cmdq_verify(cqe) != 0) {
5405 		printf("%s: create tis command corrupt\n", DEVNAME(sc));
5406 		goto free;
5407 	}
5408 
5409 	out = mcx_cmdq_out(cqe);
5410 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5411 		printf("%s: create tis failed (%x, %x)\n", DEVNAME(sc),
5412 		    out->cmd_status, be32toh(out->cmd_syndrome));
5413 		error = -1;
5414 		goto free;
5415 	}
5416 
5417 	*tis = mcx_get_id(out->cmd_tisn);
5418 free:
5419 	mcx_dmamem_free(sc, &mxm);
5420 	return (error);
5421 }
5422 
5423 static int
5424 mcx_destroy_tis(struct mcx_softc *sc, int tis)
5425 {
5426 	struct mcx_cmdq_entry *cqe;
5427 	struct mcx_cmd_destroy_tis_in *in;
5428 	struct mcx_cmd_destroy_tis_out *out;
5429 	int error;
5430 	int token;
5431 
5432 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5433 	token = mcx_cmdq_token(sc);
5434 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), token);
5435 
5436 	in = mcx_cmdq_in(cqe);
5437 	in->cmd_opcode = htobe16(MCX_CMD_DESTROY_TIS);
5438 	in->cmd_op_mod = htobe16(0);
5439 	in->cmd_tisn = htobe32(tis);
5440 
5441 	mcx_cmdq_post(sc, cqe, 0);
5442 	error = mcx_cmdq_poll(sc, cqe, 1000);
5443 	if (error != 0) {
5444 		printf("%s: destroy tis timeout\n", DEVNAME(sc));
5445 		return error;
5446 	}
5447 	if (mcx_cmdq_verify(cqe) != 0) {
5448 		printf("%s: destroy tis command corrupt\n", DEVNAME(sc));
5449 		return error;
5450 	}
5451 
5452 	out = mcx_cmdq_out(cqe);
5453 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5454 		printf("%s: destroy tis failed (%x, %x)\n", DEVNAME(sc),
5455 		    out->cmd_status, be32toh(out->cmd_syndrome));
5456 		return -1;
5457 	}
5458 
5459 	return 0;
5460 }
5461 
5462 static int
5463 mcx_create_rqt(struct mcx_softc *sc, int size, int *rqns, int *rqt)
5464 {
5465 	struct mcx_cmdq_entry *cqe;
5466 	struct mcx_dmamem mxm;
5467 	struct mcx_cmd_create_rqt_in *in;
5468 	struct mcx_cmd_create_rqt_mb_in *mbin;
5469 	struct mcx_cmd_create_rqt_out *out;
5470 	struct mcx_rqt_ctx *rqt_ctx;
5471 	int *rqtn;
5472 	int error;
5473 	int token;
5474 	int i;
5475 
5476 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5477 	token = mcx_cmdq_token(sc);
5478 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin) +
5479 	    (size * sizeof(int)), sizeof(*out), token);
5480 
5481 	in = mcx_cmdq_in(cqe);
5482 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_RQT);
5483 	in->cmd_op_mod = htobe16(0);
5484 
5485 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
5486 	    &cqe->cq_input_ptr, token) != 0) {
5487 		printf("%s: unable to allocate create rqt mailbox\n",
5488 		    DEVNAME(sc));
5489 		return (-1);
5490 	}
5491 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5492 	rqt_ctx = &mbin->cmd_rqt;
5493 	rqt_ctx->cmd_rqt_max_size = htobe16(sc->sc_max_rqt_size);
5494 	rqt_ctx->cmd_rqt_actual_size = htobe16(size);
5495 
5496 	/* rqt list follows the rqt context */
5497 	rqtn = (int *)(rqt_ctx + 1);
5498 	for (i = 0; i < size; i++) {
5499 		rqtn[i] = htobe32(rqns[i]);
5500 	}
5501 
5502 	mcx_cmdq_mboxes_sign(&mxm, 1);
5503 	mcx_cmdq_post(sc, cqe, 0);
5504 	error = mcx_cmdq_poll(sc, cqe, 1000);
5505 	if (error != 0) {
5506 		printf("%s: create rqt timeout\n", DEVNAME(sc));
5507 		goto free;
5508 	}
5509 	if (mcx_cmdq_verify(cqe) != 0) {
5510 		printf("%s: create rqt command corrupt\n", DEVNAME(sc));
5511 		goto free;
5512 	}
5513 
5514 	out = mcx_cmdq_out(cqe);
5515 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5516 		printf("%s: create rqt failed (%x, %x)\n", DEVNAME(sc),
5517 		    out->cmd_status, be32toh(out->cmd_syndrome));
5518 		error = -1;
5519 		goto free;
5520 	}
5521 
5522 	*rqt = mcx_get_id(out->cmd_rqtn);
5523 	return (0);
5524 free:
5525 	mcx_dmamem_free(sc, &mxm);
5526 	return (error);
5527 }
5528 
5529 static int
5530 mcx_destroy_rqt(struct mcx_softc *sc, int rqt)
5531 {
5532 	struct mcx_cmdq_entry *cqe;
5533 	struct mcx_cmd_destroy_rqt_in *in;
5534 	struct mcx_cmd_destroy_rqt_out *out;
5535 	int error;
5536 	int token;
5537 
5538 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5539 	token = mcx_cmdq_token(sc);
5540 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), token);
5541 
5542 	in = mcx_cmdq_in(cqe);
5543 	in->cmd_opcode = htobe16(MCX_CMD_DESTROY_RQT);
5544 	in->cmd_op_mod = htobe16(0);
5545 	in->cmd_rqtn = htobe32(rqt);
5546 
5547 	mcx_cmdq_post(sc, cqe, 0);
5548 	error = mcx_cmdq_poll(sc, cqe, 1000);
5549 	if (error != 0) {
5550 		printf("%s: destroy rqt timeout\n", DEVNAME(sc));
5551 		return error;
5552 	}
5553 	if (mcx_cmdq_verify(cqe) != 0) {
5554 		printf("%s: destroy rqt command corrupt\n", DEVNAME(sc));
5555 		return error;
5556 	}
5557 
5558 	out = mcx_cmdq_out(cqe);
5559 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5560 		printf("%s: destroy rqt failed (%x, %x)\n", DEVNAME(sc),
5561 		    out->cmd_status, be32toh(out->cmd_syndrome));
5562 		return -1;
5563 	}
5564 
5565 	return 0;
5566 }
5567 
5568 #if 0
5569 static int
5570 mcx_alloc_flow_counter(struct mcx_softc *sc, int i)
5571 {
5572 	struct mcx_cmdq_entry *cqe;
5573 	struct mcx_cmd_alloc_flow_counter_in *in;
5574 	struct mcx_cmd_alloc_flow_counter_out *out;
5575 	int error;
5576 
5577 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5578 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
5579 
5580 	in = mcx_cmdq_in(cqe);
5581 	in->cmd_opcode = htobe16(MCX_CMD_ALLOC_FLOW_COUNTER);
5582 	in->cmd_op_mod = htobe16(0);
5583 
5584 	mcx_cmdq_post(sc, cqe, 0);
5585 
5586 	error = mcx_cmdq_poll(sc, cqe, 1000);
5587 	if (error != 0) {
5588 		printf("%s: alloc flow counter timeout\n", DEVNAME(sc));
5589 		return (-1);
5590 	}
5591 	if (mcx_cmdq_verify(cqe) != 0) {
5592 		printf("%s: alloc flow counter command corrupt\n", DEVNAME(sc));
5593 		return (-1);
5594 	}
5595 
5596 	out = (struct mcx_cmd_alloc_flow_counter_out *)cqe->cq_output_data;
5597 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5598 		printf("%s: alloc flow counter failed (%x)\n", DEVNAME(sc),
5599 		    out->cmd_status);
5600 		return (-1);
5601 	}
5602 
5603 	sc->sc_flow_counter_id[i]  = be16toh(out->cmd_flow_counter_id);
5604 	printf("flow counter id %d = %d\n", i, sc->sc_flow_counter_id[i]);
5605 
5606 	return (0);
5607 }
5608 #endif
5609 
5610 static int
5611 mcx_create_flow_table(struct mcx_softc *sc, int log_size, int level,
5612     int *flow_table_id)
5613 {
5614 	struct mcx_cmdq_entry *cqe;
5615 	struct mcx_dmamem mxm;
5616 	struct mcx_cmd_create_flow_table_in *in;
5617 	struct mcx_cmd_create_flow_table_mb_in *mbin;
5618 	struct mcx_cmd_create_flow_table_out *out;
5619 	int error;
5620 	int token;
5621 
5622 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5623 	token = mcx_cmdq_token(sc);
5624 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
5625 	    sizeof(*out), token);
5626 
5627 	in = mcx_cmdq_in(cqe);
5628 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_FLOW_TABLE);
5629 	in->cmd_op_mod = htobe16(0);
5630 
5631 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
5632 	    &cqe->cq_input_ptr, token) != 0) {
5633 		printf("%s: unable to allocate create flow table mailbox\n",
5634 		    DEVNAME(sc));
5635 		return (-1);
5636 	}
5637 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5638 	mbin->cmd_table_type = MCX_FLOW_TABLE_TYPE_RX;
5639 	mbin->cmd_ctx.ft_log_size = log_size;
5640 	mbin->cmd_ctx.ft_level = level;
5641 
5642 	mcx_cmdq_mboxes_sign(&mxm, 1);
5643 	mcx_cmdq_post(sc, cqe, 0);
5644 	error = mcx_cmdq_poll(sc, cqe, 1000);
5645 	if (error != 0) {
5646 		printf("%s: create flow table timeout\n", DEVNAME(sc));
5647 		goto free;
5648 	}
5649 	if (mcx_cmdq_verify(cqe) != 0) {
5650 		printf("%s: create flow table command corrupt\n", DEVNAME(sc));
5651 		goto free;
5652 	}
5653 
5654 	out = mcx_cmdq_out(cqe);
5655 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5656 		printf("%s: create flow table failed (%x, %x)\n", DEVNAME(sc),
5657 		    out->cmd_status, be32toh(out->cmd_syndrome));
5658 		error = -1;
5659 		goto free;
5660 	}
5661 
5662 	*flow_table_id = mcx_get_id(out->cmd_table_id);
5663 free:
5664 	mcx_dmamem_free(sc, &mxm);
5665 	return (error);
5666 }
5667 
5668 static int
5669 mcx_set_flow_table_root(struct mcx_softc *sc, int flow_table_id)
5670 {
5671 	struct mcx_cmdq_entry *cqe;
5672 	struct mcx_dmamem mxm;
5673 	struct mcx_cmd_set_flow_table_root_in *in;
5674 	struct mcx_cmd_set_flow_table_root_mb_in *mbin;
5675 	struct mcx_cmd_set_flow_table_root_out *out;
5676 	int error;
5677 	int token;
5678 
5679 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5680 	token = mcx_cmdq_token(sc);
5681 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
5682 	    sizeof(*out), token);
5683 
5684 	in = mcx_cmdq_in(cqe);
5685 	in->cmd_opcode = htobe16(MCX_CMD_SET_FLOW_TABLE_ROOT);
5686 	in->cmd_op_mod = htobe16(0);
5687 
5688 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
5689 	    &cqe->cq_input_ptr, token) != 0) {
5690 		printf("%s: unable to allocate set flow table root mailbox\n",
5691 		    DEVNAME(sc));
5692 		return (-1);
5693 	}
5694 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5695 	mbin->cmd_table_type = MCX_FLOW_TABLE_TYPE_RX;
5696 	mbin->cmd_table_id = htobe32(flow_table_id);
5697 
5698 	mcx_cmdq_mboxes_sign(&mxm, 1);
5699 	mcx_cmdq_post(sc, cqe, 0);
5700 	error = mcx_cmdq_poll(sc, cqe, 1000);
5701 	if (error != 0) {
5702 		printf("%s: set flow table root timeout\n", DEVNAME(sc));
5703 		goto free;
5704 	}
5705 	if (mcx_cmdq_verify(cqe) != 0) {
5706 		printf("%s: set flow table root command corrupt\n",
5707 		    DEVNAME(sc));
5708 		goto free;
5709 	}
5710 
5711 	out = mcx_cmdq_out(cqe);
5712 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5713 		printf("%s: set flow table root failed (%x, %x)\n",
5714 		    DEVNAME(sc), out->cmd_status, be32toh(out->cmd_syndrome));
5715 		error = -1;
5716 		goto free;
5717 	}
5718 
5719 free:
5720 	mcx_dmamem_free(sc, &mxm);
5721 	return (error);
5722 }
5723 
5724 static int
5725 mcx_destroy_flow_table(struct mcx_softc *sc, int flow_table_id)
5726 {
5727 	struct mcx_cmdq_entry *cqe;
5728 	struct mcx_dmamem mxm;
5729 	struct mcx_cmd_destroy_flow_table_in *in;
5730 	struct mcx_cmd_destroy_flow_table_mb_in *mb;
5731 	struct mcx_cmd_destroy_flow_table_out *out;
5732 	int error;
5733 	int token;
5734 
5735 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5736 	token = mcx_cmdq_token(sc);
5737 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mb), sizeof(*out), token);
5738 
5739 	in = mcx_cmdq_in(cqe);
5740 	in->cmd_opcode = htobe16(MCX_CMD_DESTROY_FLOW_TABLE);
5741 	in->cmd_op_mod = htobe16(0);
5742 
5743 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
5744 	    &cqe->cq_input_ptr, token) != 0) {
5745 		printf("%s: unable to allocate destroy flow table mailbox\n",
5746 		    DEVNAME(sc));
5747 		return (-1);
5748 	}
5749 	mb = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5750 	mb->cmd_table_type = MCX_FLOW_TABLE_TYPE_RX;
5751 	mb->cmd_table_id = htobe32(flow_table_id);
5752 
5753 	mcx_cmdq_mboxes_sign(&mxm, 1);
5754 	mcx_cmdq_post(sc, cqe, 0);
5755 	error = mcx_cmdq_poll(sc, cqe, 1000);
5756 	if (error != 0) {
5757 		printf("%s: destroy flow table timeout\n", DEVNAME(sc));
5758 		goto free;
5759 	}
5760 	if (mcx_cmdq_verify(cqe) != 0) {
5761 		printf("%s: destroy flow table command corrupt\n",
5762 		    DEVNAME(sc));
5763 		goto free;
5764 	}
5765 
5766 	out = mcx_cmdq_out(cqe);
5767 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5768 		printf("%s: destroy flow table failed (%x, %x)\n", DEVNAME(sc),
5769 		    out->cmd_status, be32toh(out->cmd_syndrome));
5770 		error = -1;
5771 		goto free;
5772 	}
5773 
5774 free:
5775 	mcx_dmamem_free(sc, &mxm);
5776 	return (error);
5777 }
5778 
5779 
5780 static int
5781 mcx_create_flow_group(struct mcx_softc *sc, int flow_table_id, int group,
5782     int start, int size, int match_enable, struct mcx_flow_match *match)
5783 {
5784 	struct mcx_cmdq_entry *cqe;
5785 	struct mcx_dmamem mxm;
5786 	struct mcx_cmd_create_flow_group_in *in;
5787 	struct mcx_cmd_create_flow_group_mb_in *mbin;
5788 	struct mcx_cmd_create_flow_group_out *out;
5789 	struct mcx_flow_group *mfg;
5790 	int error;
5791 	int token;
5792 
5793 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5794 	token = mcx_cmdq_token(sc);
5795 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin), sizeof(*out),
5796 	    token);
5797 
5798 	in = mcx_cmdq_in(cqe);
5799 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_FLOW_GROUP);
5800 	in->cmd_op_mod = htobe16(0);
5801 
5802 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2, &cqe->cq_input_ptr, token)
5803 	    != 0) {
5804 		printf("%s: unable to allocate create flow group mailbox\n",
5805 		    DEVNAME(sc));
5806 		return (-1);
5807 	}
5808 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5809 	mbin->cmd_table_type = MCX_FLOW_TABLE_TYPE_RX;
5810 	mbin->cmd_table_id = htobe32(flow_table_id);
5811 	mbin->cmd_start_flow_index = htobe32(start);
5812 	mbin->cmd_end_flow_index = htobe32(start + (size - 1));
5813 
5814 	mbin->cmd_match_criteria_enable = match_enable;
5815 	memcpy(&mbin->cmd_match_criteria, match, sizeof(*match));
5816 
5817 	mcx_cmdq_mboxes_sign(&mxm, 2);
5818 	mcx_cmdq_post(sc, cqe, 0);
5819 	error = mcx_cmdq_poll(sc, cqe, 1000);
5820 	if (error != 0) {
5821 		printf("%s: create flow group timeout\n", DEVNAME(sc));
5822 		goto free;
5823 	}
5824 	if (mcx_cmdq_verify(cqe) != 0) {
5825 		printf("%s: create flow group command corrupt\n", DEVNAME(sc));
5826 		goto free;
5827 	}
5828 
5829 	out = mcx_cmdq_out(cqe);
5830 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5831 		printf("%s: create flow group failed (%x, %x)\n", DEVNAME(sc),
5832 		    out->cmd_status, be32toh(out->cmd_syndrome));
5833 		error = -1;
5834 		goto free;
5835 	}
5836 
5837 	mfg = &sc->sc_flow_group[group];
5838 	mfg->g_id = mcx_get_id(out->cmd_group_id);
5839 	mfg->g_table = flow_table_id;
5840 	mfg->g_start = start;
5841 	mfg->g_size = size;
5842 
5843 free:
5844 	mcx_dmamem_free(sc, &mxm);
5845 	return (error);
5846 }
5847 
5848 static int
5849 mcx_destroy_flow_group(struct mcx_softc *sc, int group)
5850 {
5851 	struct mcx_cmdq_entry *cqe;
5852 	struct mcx_dmamem mxm;
5853 	struct mcx_cmd_destroy_flow_group_in *in;
5854 	struct mcx_cmd_destroy_flow_group_mb_in *mb;
5855 	struct mcx_cmd_destroy_flow_group_out *out;
5856 	struct mcx_flow_group *mfg;
5857 	int error;
5858 	int token;
5859 
5860 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5861 	token = mcx_cmdq_token(sc);
5862 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mb), sizeof(*out), token);
5863 
5864 	in = mcx_cmdq_in(cqe);
5865 	in->cmd_opcode = htobe16(MCX_CMD_DESTROY_FLOW_GROUP);
5866 	in->cmd_op_mod = htobe16(0);
5867 
5868 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2,
5869 	    &cqe->cq_input_ptr, token) != 0) {
5870 		printf("%s: unable to allocate destroy flow group mailbox\n",
5871 		    DEVNAME(sc));
5872 		return (-1);
5873 	}
5874 	mb = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5875 	mb->cmd_table_type = MCX_FLOW_TABLE_TYPE_RX;
5876 	mfg = &sc->sc_flow_group[group];
5877 	mb->cmd_table_id = htobe32(mfg->g_table);
5878 	mb->cmd_group_id = htobe32(mfg->g_id);
5879 
5880 	mcx_cmdq_mboxes_sign(&mxm, 2);
5881 	mcx_cmdq_post(sc, cqe, 0);
5882 	error = mcx_cmdq_poll(sc, cqe, 1000);
5883 	if (error != 0) {
5884 		printf("%s: destroy flow group timeout\n", DEVNAME(sc));
5885 		goto free;
5886 	}
5887 	if (mcx_cmdq_verify(cqe) != 0) {
5888 		printf("%s: destroy flow group command corrupt\n", DEVNAME(sc));
5889 		goto free;
5890 	}
5891 
5892 	out = mcx_cmdq_out(cqe);
5893 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5894 		printf("%s: destroy flow group failed (%x, %x)\n", DEVNAME(sc),
5895 		    out->cmd_status, be32toh(out->cmd_syndrome));
5896 		error = -1;
5897 		goto free;
5898 	}
5899 
5900 	mfg->g_id = -1;
5901 	mfg->g_table = -1;
5902 	mfg->g_size = 0;
5903 	mfg->g_start = 0;
5904 free:
5905 	mcx_dmamem_free(sc, &mxm);
5906 	return (error);
5907 }
5908 
5909 static int
5910 mcx_set_flow_table_entry_mac(struct mcx_softc *sc, int group, int index,
5911     const uint8_t *macaddr, uint32_t dest)
5912 {
5913 	struct mcx_cmdq_entry *cqe;
5914 	struct mcx_dmamem mxm;
5915 	struct mcx_cmd_set_flow_table_entry_in *in;
5916 	struct mcx_cmd_set_flow_table_entry_mb_in *mbin;
5917 	struct mcx_cmd_set_flow_table_entry_out *out;
5918 	struct mcx_flow_group *mfg;
5919 	uint32_t *pdest;
5920 	int error;
5921 	int token;
5922 
5923 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5924 	token = mcx_cmdq_token(sc);
5925 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin) + sizeof(*pdest),
5926 	    sizeof(*out), token);
5927 
5928 	in = mcx_cmdq_in(cqe);
5929 	in->cmd_opcode = htobe16(MCX_CMD_SET_FLOW_TABLE_ENTRY);
5930 	in->cmd_op_mod = htobe16(0);
5931 
5932 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2, &cqe->cq_input_ptr, token)
5933 	    != 0) {
5934 		printf("%s: unable to allocate set flow table entry mailbox\n",
5935 		    DEVNAME(sc));
5936 		return (-1);
5937 	}
5938 
5939 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5940 	mbin->cmd_table_type = MCX_FLOW_TABLE_TYPE_RX;
5941 
5942 	mfg = &sc->sc_flow_group[group];
5943 	mbin->cmd_table_id = htobe32(mfg->g_table);
5944 	mbin->cmd_flow_index = htobe32(mfg->g_start + index);
5945 	mbin->cmd_flow_ctx.fc_group_id = htobe32(mfg->g_id);
5946 
5947 	/* flow context ends at offset 0x330, 0x130 into the second mbox */
5948 	pdest = (uint32_t *)
5949 	    (((char *)mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 1))) + 0x130);
5950 	mbin->cmd_flow_ctx.fc_action = htobe32(MCX_FLOW_CONTEXT_ACTION_FORWARD);
5951 	mbin->cmd_flow_ctx.fc_dest_list_size = htobe32(1);
5952 	*pdest = htobe32(dest);
5953 
5954 	/* the only thing we match on at the moment is the dest mac address */
5955 	if (macaddr != NULL) {
5956 		memcpy(mbin->cmd_flow_ctx.fc_match_value.mc_dest_mac, macaddr,
5957 		    ETHER_ADDR_LEN);
5958 	}
5959 
5960 	mcx_cmdq_mboxes_sign(&mxm, 2);
5961 	mcx_cmdq_post(sc, cqe, 0);
5962 	error = mcx_cmdq_poll(sc, cqe, 1000);
5963 	if (error != 0) {
5964 		printf("%s: set flow table entry timeout\n", DEVNAME(sc));
5965 		goto free;
5966 	}
5967 	if (mcx_cmdq_verify(cqe) != 0) {
5968 		printf("%s: set flow table entry command corrupt\n",
5969 		    DEVNAME(sc));
5970 		goto free;
5971 	}
5972 
5973 	out = mcx_cmdq_out(cqe);
5974 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5975 		printf("%s: set flow table entry failed (%x, %x)\n",
5976 		    DEVNAME(sc), out->cmd_status, be32toh(out->cmd_syndrome));
5977 		error = -1;
5978 		goto free;
5979 	}
5980 
5981 free:
5982 	mcx_dmamem_free(sc, &mxm);
5983 	return (error);
5984 }
5985 
5986 static int
5987 mcx_set_flow_table_entry_proto(struct mcx_softc *sc, int group, int index,
5988     int ethertype, int ip_proto, uint32_t dest)
5989 {
5990 	struct mcx_cmdq_entry *cqe;
5991 	struct mcx_dmamem mxm;
5992 	struct mcx_cmd_set_flow_table_entry_in *in;
5993 	struct mcx_cmd_set_flow_table_entry_mb_in *mbin;
5994 	struct mcx_cmd_set_flow_table_entry_out *out;
5995 	struct mcx_flow_group *mfg;
5996 	uint32_t *pdest;
5997 	int error;
5998 	int token;
5999 
6000 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
6001 	token = mcx_cmdq_token(sc);
6002 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin) + sizeof(*pdest),
6003 	    sizeof(*out), token);
6004 
6005 	in = mcx_cmdq_in(cqe);
6006 	in->cmd_opcode = htobe16(MCX_CMD_SET_FLOW_TABLE_ENTRY);
6007 	in->cmd_op_mod = htobe16(0);
6008 
6009 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2, &cqe->cq_input_ptr, token)
6010 	    != 0) {
6011 		printf("%s: unable to allocate set flow table entry mailbox\n",
6012 		    DEVNAME(sc));
6013 		return (-1);
6014 	}
6015 
6016 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
6017 	mbin->cmd_table_type = MCX_FLOW_TABLE_TYPE_RX;
6018 
6019 	mfg = &sc->sc_flow_group[group];
6020 	mbin->cmd_table_id = htobe32(mfg->g_table);
6021 	mbin->cmd_flow_index = htobe32(mfg->g_start + index);
6022 	mbin->cmd_flow_ctx.fc_group_id = htobe32(mfg->g_id);
6023 
6024 	/* flow context ends at offset 0x330, 0x130 into the second mbox */
6025 	pdest = (uint32_t *)
6026 	    (((char *)mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 1))) + 0x130);
6027 	mbin->cmd_flow_ctx.fc_action = htobe32(MCX_FLOW_CONTEXT_ACTION_FORWARD);
6028 	mbin->cmd_flow_ctx.fc_dest_list_size = htobe32(1);
6029 	*pdest = htobe32(dest);
6030 
6031 	mbin->cmd_flow_ctx.fc_match_value.mc_ethertype = htobe16(ethertype);
6032 	mbin->cmd_flow_ctx.fc_match_value.mc_ip_proto = ip_proto;
6033 
6034 	mcx_cmdq_mboxes_sign(&mxm, 2);
6035 	mcx_cmdq_post(sc, cqe, 0);
6036 	error = mcx_cmdq_poll(sc, cqe, 1000);
6037 	if (error != 0) {
6038 		printf("%s: set flow table entry timeout\n", DEVNAME(sc));
6039 		goto free;
6040 	}
6041 	if (mcx_cmdq_verify(cqe) != 0) {
6042 		printf("%s: set flow table entry command corrupt\n",
6043 		    DEVNAME(sc));
6044 		goto free;
6045 	}
6046 
6047 	out = mcx_cmdq_out(cqe);
6048 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
6049 		printf("%s: set flow table entry failed (%x, %x)\n",
6050 		    DEVNAME(sc), out->cmd_status, be32toh(out->cmd_syndrome));
6051 		error = -1;
6052 		goto free;
6053 	}
6054 
6055 free:
6056 	mcx_dmamem_free(sc, &mxm);
6057 	return (error);
6058 }
6059 
6060 static int
6061 mcx_delete_flow_table_entry(struct mcx_softc *sc, int group, int index)
6062 {
6063 	struct mcx_cmdq_entry *cqe;
6064 	struct mcx_dmamem mxm;
6065 	struct mcx_cmd_delete_flow_table_entry_in *in;
6066 	struct mcx_cmd_delete_flow_table_entry_mb_in *mbin;
6067 	struct mcx_cmd_delete_flow_table_entry_out *out;
6068 	struct mcx_flow_group *mfg;
6069 	int error;
6070 	int token;
6071 
6072 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
6073 	token = mcx_cmdq_token(sc);
6074 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin), sizeof(*out),
6075 	    token);
6076 
6077 	in = mcx_cmdq_in(cqe);
6078 	in->cmd_opcode = htobe16(MCX_CMD_DELETE_FLOW_TABLE_ENTRY);
6079 	in->cmd_op_mod = htobe16(0);
6080 
6081 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2,
6082 	    &cqe->cq_input_ptr, token) != 0) {
6083 		printf("%s: unable to allocate "
6084 		    "delete flow table entry mailbox\n", DEVNAME(sc));
6085 		return (-1);
6086 	}
6087 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
6088 	mbin->cmd_table_type = MCX_FLOW_TABLE_TYPE_RX;
6089 
6090 	mfg = &sc->sc_flow_group[group];
6091 	mbin->cmd_table_id = htobe32(mfg->g_table);
6092 	mbin->cmd_flow_index = htobe32(mfg->g_start + index);
6093 
6094 	mcx_cmdq_mboxes_sign(&mxm, 2);
6095 	mcx_cmdq_post(sc, cqe, 0);
6096 	error = mcx_cmdq_poll(sc, cqe, 1000);
6097 	if (error != 0) {
6098 		printf("%s: delete flow table entry timeout\n", DEVNAME(sc));
6099 		goto free;
6100 	}
6101 	if (mcx_cmdq_verify(cqe) != 0) {
6102 		printf("%s: delete flow table entry command corrupt\n",
6103 		    DEVNAME(sc));
6104 		goto free;
6105 	}
6106 
6107 	out = mcx_cmdq_out(cqe);
6108 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
6109 		printf("%s: delete flow table entry %d:%d failed (%x, %x)\n",
6110 		    DEVNAME(sc), group, index, out->cmd_status,
6111 		    be32toh(out->cmd_syndrome));
6112 		error = -1;
6113 		goto free;
6114 	}
6115 
6116 free:
6117 	mcx_dmamem_free(sc, &mxm);
6118 	return (error);
6119 }
6120 
6121 #if 0
6122 int
6123 mcx_dump_flow_table(struct mcx_softc *sc, int flow_table_id)
6124 {
6125 	struct mcx_dmamem mxm;
6126 	struct mcx_cmdq_entry *cqe;
6127 	struct mcx_cmd_query_flow_table_in *in;
6128 	struct mcx_cmd_query_flow_table_mb_in *mbin;
6129 	struct mcx_cmd_query_flow_table_out *out;
6130 	struct mcx_cmd_query_flow_table_mb_out *mbout;
6131 	uint8_t token = mcx_cmdq_token(sc);
6132 	int error;
6133 	int i;
6134 	uint8_t *dump;
6135 
6136 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
6137 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
6138 	    sizeof(*out) + sizeof(*mbout) + 16, token);
6139 
6140 	in = mcx_cmdq_in(cqe);
6141 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_FLOW_TABLE);
6142 	in->cmd_op_mod = htobe16(0);
6143 
6144 	CTASSERT(sizeof(*mbin) <= MCX_CMDQ_MAILBOX_DATASIZE);
6145 	CTASSERT(sizeof(*mbout) <= MCX_CMDQ_MAILBOX_DATASIZE);
6146 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2,
6147 	    &cqe->cq_output_ptr, token) != 0) {
6148 		printf(", unable to allocate query flow table mailboxes\n");
6149 		return (-1);
6150 	}
6151 	cqe->cq_input_ptr = cqe->cq_output_ptr;
6152 
6153 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
6154 	mbin->cmd_table_type = 0;
6155 	mbin->cmd_table_id = htobe32(flow_table_id);
6156 
6157 	mcx_cmdq_mboxes_sign(&mxm, 1);
6158 
6159 	mcx_cmdq_post(sc, cqe, 0);
6160 	error = mcx_cmdq_poll(sc, cqe, 1000);
6161 	if (error != 0) {
6162 		printf("%s: query flow table timeout\n", DEVNAME(sc));
6163 		goto free;
6164 	}
6165 	error = mcx_cmdq_verify(cqe);
6166 	if (error != 0) {
6167 		printf("%s: query flow table reply corrupt\n", DEVNAME(sc));
6168 		goto free;
6169 	}
6170 
6171 	out = mcx_cmdq_out(cqe);
6172 	switch (out->cmd_status) {
6173 	case MCX_CQ_STATUS_OK:
6174 		break;
6175 	default:
6176 		printf("%s: query flow table failed (%x/%x)\n", DEVNAME(sc),
6177 		    out->cmd_status, be32toh(out->cmd_syndrome));
6178 		error = -1;
6179 		goto free;
6180 	}
6181 
6182         mbout = (struct mcx_cmd_query_flow_table_mb_out *)
6183 	    (mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)));
6184 	dump = (uint8_t *)mbout + 8;
6185 	for (i = 0; i < sizeof(struct mcx_flow_table_ctx); i++) {
6186 		printf("%.2x ", dump[i]);
6187 		if (i % 16 == 15)
6188 			printf("\n");
6189 	}
6190 free:
6191 	mcx_cq_mboxes_free(sc, &mxm);
6192 	return (error);
6193 }
6194 int
6195 mcx_dump_flow_table_entry(struct mcx_softc *sc, int flow_table_id, int index)
6196 {
6197 	struct mcx_dmamem mxm;
6198 	struct mcx_cmdq_entry *cqe;
6199 	struct mcx_cmd_query_flow_table_entry_in *in;
6200 	struct mcx_cmd_query_flow_table_entry_mb_in *mbin;
6201 	struct mcx_cmd_query_flow_table_entry_out *out;
6202 	struct mcx_cmd_query_flow_table_entry_mb_out *mbout;
6203 	uint8_t token = mcx_cmdq_token(sc);
6204 	int error;
6205 	int i;
6206 	uint8_t *dump;
6207 
6208 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
6209 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
6210 	    sizeof(*out) + sizeof(*mbout) + 16, token);
6211 
6212 	in = mcx_cmdq_in(cqe);
6213 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_FLOW_TABLE_ENTRY);
6214 	in->cmd_op_mod = htobe16(0);
6215 
6216 	CTASSERT(sizeof(*mbin) <= MCX_CMDQ_MAILBOX_DATASIZE);
6217 	CTASSERT(sizeof(*mbout) <= MCX_CMDQ_MAILBOX_DATASIZE*2);
6218 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2,
6219 	    &cqe->cq_output_ptr, token) != 0) {
6220 		printf(", unable to allocate "
6221 		    "query flow table entry mailboxes\n");
6222 		return (-1);
6223 	}
6224 	cqe->cq_input_ptr = cqe->cq_output_ptr;
6225 
6226 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
6227 	mbin->cmd_table_type = 0;
6228 	mbin->cmd_table_id = htobe32(flow_table_id);
6229 	mbin->cmd_flow_index = htobe32(index);
6230 
6231 	mcx_cmdq_mboxes_sign(&mxm, 1);
6232 
6233 	mcx_cmdq_post(sc, cqe, 0);
6234 	error = mcx_cmdq_poll(sc, cqe, 1000);
6235 	if (error != 0) {
6236 		printf("%s: query flow table entry timeout\n", DEVNAME(sc));
6237 		goto free;
6238 	}
6239 	error = mcx_cmdq_verify(cqe);
6240 	if (error != 0) {
6241 		printf("%s: query flow table entry reply corrupt\n",
6242 		    DEVNAME(sc));
6243 		goto free;
6244 	}
6245 
6246 	out = mcx_cmdq_out(cqe);
6247 	switch (out->cmd_status) {
6248 	case MCX_CQ_STATUS_OK:
6249 		break;
6250 	default:
6251 		printf("%s: query flow table entry failed (%x/%x)\n",
6252 		    DEVNAME(sc), out->cmd_status, be32toh(out->cmd_syndrome));
6253 		error = -1;
6254 		goto free;
6255 	}
6256 
6257         mbout = (struct mcx_cmd_query_flow_table_entry_mb_out *)
6258 	    (mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)));
6259 	dump = (uint8_t *)mbout;
6260 	for (i = 0; i < MCX_CMDQ_MAILBOX_DATASIZE; i++) {
6261 		printf("%.2x ", dump[i]);
6262 		if (i % 16 == 15)
6263 			printf("\n");
6264 	}
6265 
6266 free:
6267 	mcx_cq_mboxes_free(sc, &mxm);
6268 	return (error);
6269 }
6270 
6271 int
6272 mcx_dump_flow_group(struct mcx_softc *sc, int flow_table_id)
6273 {
6274 	struct mcx_dmamem mxm;
6275 	struct mcx_cmdq_entry *cqe;
6276 	struct mcx_cmd_query_flow_group_in *in;
6277 	struct mcx_cmd_query_flow_group_mb_in *mbin;
6278 	struct mcx_cmd_query_flow_group_out *out;
6279 	struct mcx_cmd_query_flow_group_mb_out *mbout;
6280 	uint8_t token = mcx_cmdq_token(sc);
6281 	int error;
6282 	int i;
6283 	uint8_t *dump;
6284 
6285 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
6286 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
6287 	    sizeof(*out) + sizeof(*mbout) + 16, token);
6288 
6289 	in = mcx_cmdq_in(cqe);
6290 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_FLOW_GROUP);
6291 	in->cmd_op_mod = htobe16(0);
6292 
6293 	CTASSERT(sizeof(*mbin) <= MCX_CMDQ_MAILBOX_DATASIZE);
6294 	CTASSERT(sizeof(*mbout) <= MCX_CMDQ_MAILBOX_DATASIZE*2);
6295 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2,
6296 	    &cqe->cq_output_ptr, token) != 0) {
6297 		printf(", unable to allocate query flow group mailboxes\n");
6298 		return (-1);
6299 	}
6300 	cqe->cq_input_ptr = cqe->cq_output_ptr;
6301 
6302 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
6303 	mbin->cmd_table_type = 0;
6304 	mbin->cmd_table_id = htobe32(flow_table_id);
6305 	mbin->cmd_group_id = htobe32(sc->sc_flow_group_id);
6306 
6307 	mcx_cmdq_mboxes_sign(&mxm, 1);
6308 
6309 	mcx_cmdq_post(sc, cqe, 0);
6310 	error = mcx_cmdq_poll(sc, cqe, 1000);
6311 	if (error != 0) {
6312 		printf("%s: query flow group timeout\n", DEVNAME(sc));
6313 		goto free;
6314 	}
6315 	error = mcx_cmdq_verify(cqe);
6316 	if (error != 0) {
6317 		printf("%s: query flow group reply corrupt\n", DEVNAME(sc));
6318 		goto free;
6319 	}
6320 
6321 	out = mcx_cmdq_out(cqe);
6322 	switch (out->cmd_status) {
6323 	case MCX_CQ_STATUS_OK:
6324 		break;
6325 	default:
6326 		printf("%s: query flow group failed (%x/%x)\n", DEVNAME(sc),
6327 		    out->cmd_status, be32toh(out->cmd_syndrome));
6328 		error = -1;
6329 		goto free;
6330 	}
6331 
6332         mbout = (struct mcx_cmd_query_flow_group_mb_out *)
6333 	    (mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)));
6334 	dump = (uint8_t *)mbout;
6335 	for (i = 0; i < MCX_CMDQ_MAILBOX_DATASIZE; i++) {
6336 		printf("%.2x ", dump[i]);
6337 		if (i % 16 == 15)
6338 			printf("\n");
6339 	}
6340 	dump = (uint8_t *)(mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 1)));
6341 	for (i = 0; i < MCX_CMDQ_MAILBOX_DATASIZE; i++) {
6342 		printf("%.2x ", dump[i]);
6343 		if (i % 16 == 15)
6344 			printf("\n");
6345 	}
6346 
6347 free:
6348 	mcx_cq_mboxes_free(sc, &mxm);
6349 	return (error);
6350 }
6351 
6352 static int
6353 mcx_dump_counters(struct mcx_softc *sc)
6354 {
6355 	struct mcx_dmamem mxm;
6356 	struct mcx_cmdq_entry *cqe;
6357 	struct mcx_cmd_query_vport_counters_in *in;
6358 	struct mcx_cmd_query_vport_counters_mb_in *mbin;
6359 	struct mcx_cmd_query_vport_counters_out *out;
6360 	struct mcx_nic_vport_counters *counters;
6361 	int error, token;
6362 
6363 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
6364 	token = mcx_cmdq_token(sc);
6365 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
6366 	    sizeof(*out) + sizeof(*counters), token);
6367 
6368 	in = mcx_cmdq_in(cqe);
6369 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_VPORT_COUNTERS);
6370 	in->cmd_op_mod = htobe16(0);
6371 
6372 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
6373 	    &cqe->cq_output_ptr, token) != 0) {
6374 		printf(", unable to allocate "
6375 		    "query nic vport counters mailboxen\n");
6376 		return (-1);
6377 	}
6378 	cqe->cq_input_ptr = cqe->cq_output_ptr;
6379 
6380 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
6381 	mbin->cmd_clear = 0x80;
6382 
6383 	mcx_cmdq_mboxes_sign(&mxm, 1);
6384 	mcx_cmdq_post(sc, cqe, 0);
6385 
6386 	error = mcx_cmdq_poll(sc, cqe, 1000);
6387 	if (error != 0) {
6388 		printf("%s: query nic vport counters timeout\n", DEVNAME(sc));
6389 		goto free;
6390 	}
6391 	if (mcx_cmdq_verify(cqe) != 0) {
6392 		printf("%s: query nic vport counters command corrupt\n",
6393 		    DEVNAME(sc));
6394 		goto free;
6395 	}
6396 
6397 	out = mcx_cmdq_out(cqe);
6398 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
6399 		printf("%s: query nic vport counters failed (%x, %x)\n",
6400 		    DEVNAME(sc), out->cmd_status, betoh32(out->cmd_syndrome));
6401 		error = -1;
6402 		goto free;
6403 	}
6404 
6405 	counters = (struct mcx_nic_vport_counters *)
6406 	    (mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)));
6407 	if (counters->rx_bcast.packets + counters->tx_bcast.packets +
6408 	    counters->rx_ucast.packets + counters->tx_ucast.packets +
6409 	    counters->rx_err.packets + counters->tx_err.packets)
6410 		printf("%s: err %llx/%llx uc %llx/%llx bc %llx/%llx\n",
6411 		    DEVNAME(sc),
6412 		    betoh64(counters->tx_err.packets),
6413 		    betoh64(counters->rx_err.packets),
6414 		    betoh64(counters->tx_ucast.packets),
6415 		    betoh64(counters->rx_ucast.packets),
6416 		    betoh64(counters->tx_bcast.packets),
6417 		    betoh64(counters->rx_bcast.packets));
6418 free:
6419 	mcx_dmamem_free(sc, &mxm);
6420 
6421 	return (error);
6422 }
6423 
6424 static int
6425 mcx_dump_flow_counter(struct mcx_softc *sc, int index, const char *what)
6426 {
6427 	struct mcx_dmamem mxm;
6428 	struct mcx_cmdq_entry *cqe;
6429 	struct mcx_cmd_query_flow_counter_in *in;
6430 	struct mcx_cmd_query_flow_counter_mb_in *mbin;
6431 	struct mcx_cmd_query_flow_counter_out *out;
6432 	struct mcx_counter *counters;
6433 	int error, token;
6434 
6435 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
6436 	token = mcx_cmdq_token(sc);
6437 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin), sizeof(*out) +
6438 	    sizeof(*counters), token);
6439 
6440 	in = mcx_cmdq_in(cqe);
6441 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_FLOW_COUNTER);
6442 	in->cmd_op_mod = htobe16(0);
6443 
6444 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
6445 	    &cqe->cq_output_ptr, token) != 0) {
6446 		printf(", unable to allocate query flow counter mailboxen\n");
6447 		return (-1);
6448 	}
6449 	cqe->cq_input_ptr = cqe->cq_output_ptr;
6450 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
6451 	mbin->cmd_flow_counter_id = htobe16(sc->sc_flow_counter_id[index]);
6452 	mbin->cmd_clear = 0x80;
6453 
6454 	mcx_cmdq_mboxes_sign(&mxm, 1);
6455 	mcx_cmdq_post(sc, cqe, 0);
6456 
6457 	error = mcx_cmdq_poll(sc, cqe, 1000);
6458 	if (error != 0) {
6459 		printf("%s: query flow counter timeout\n", DEVNAME(sc));
6460 		goto free;
6461 	}
6462 	if (mcx_cmdq_verify(cqe) != 0) {
6463 		printf("%s: query flow counter command corrupt\n", DEVNAME(sc));
6464 		goto free;
6465 	}
6466 
6467 	out = mcx_cmdq_out(cqe);
6468 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
6469 		printf("%s: query flow counter failed (%x, %x)\n", DEVNAME(sc),
6470 		    out->cmd_status, betoh32(out->cmd_syndrome));
6471 		error = -1;
6472 		goto free;
6473 	}
6474 
6475 	counters = (struct mcx_counter *)
6476 	    (mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)));
6477 	if (counters->packets)
6478 		printf("%s: %s inflow %llx\n", DEVNAME(sc), what,
6479 		    betoh64(counters->packets));
6480 free:
6481 	mcx_dmamem_free(sc, &mxm);
6482 
6483 	return (error);
6484 }
6485 
6486 #endif
6487 
6488 #if NKSTAT > 0
6489 
6490 int
6491 mcx_query_rq(struct mcx_softc *sc, struct mcx_rx *rx, struct mcx_rq_ctx *rq_ctx)
6492 {
6493 	struct mcx_dmamem mxm;
6494 	struct mcx_cmdq_entry *cqe;
6495 	struct mcx_cmd_query_rq_in *in;
6496 	struct mcx_cmd_query_rq_out *out;
6497 	struct mcx_cmd_query_rq_mb_out *mbout;
6498 	uint8_t token = mcx_cmdq_token(sc);
6499 	int error;
6500 
6501 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
6502 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out) + sizeof(*mbout) + 16,
6503 	    token);
6504 
6505 	in = mcx_cmdq_in(cqe);
6506 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_RQ);
6507 	in->cmd_op_mod = htobe16(0);
6508 	in->cmd_rqn = htobe32(rx->rx_rqn);
6509 
6510 	CTASSERT(sizeof(*mbout) <= MCX_CMDQ_MAILBOX_DATASIZE*2);
6511 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2,
6512 	    &cqe->cq_output_ptr, token) != 0) {
6513 		printf("%s: unable to allocate query rq mailboxes\n", DEVNAME(sc));
6514 		return (-1);
6515 	}
6516 
6517 	mcx_cmdq_mboxes_sign(&mxm, 1);
6518 
6519 	mcx_cmdq_post(sc, cqe, 0);
6520 	error = mcx_cmdq_poll(sc, cqe, 1000);
6521 	if (error != 0) {
6522 		printf("%s: query rq timeout\n", DEVNAME(sc));
6523 		goto free;
6524 	}
6525 	error = mcx_cmdq_verify(cqe);
6526 	if (error != 0) {
6527 		printf("%s: query rq reply corrupt\n", DEVNAME(sc));
6528 		goto free;
6529 	}
6530 
6531 	out = mcx_cmdq_out(cqe);
6532 	switch (out->cmd_status) {
6533 	case MCX_CQ_STATUS_OK:
6534 		break;
6535 	default:
6536 		printf("%s: query rq failed (%x/%x)\n", DEVNAME(sc),
6537 		    out->cmd_status, be32toh(out->cmd_syndrome));
6538 		error = -1;
6539 		goto free;
6540 	}
6541 
6542         mbout = (struct mcx_cmd_query_rq_mb_out *)
6543 	    (mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)));
6544 	memcpy(rq_ctx, &mbout->cmd_ctx, sizeof(*rq_ctx));
6545 
6546 free:
6547 	mcx_cq_mboxes_free(sc, &mxm);
6548 	return (error);
6549 }
6550 
6551 int
6552 mcx_query_sq(struct mcx_softc *sc, struct mcx_tx *tx, struct mcx_sq_ctx *sq_ctx)
6553 {
6554 	struct mcx_dmamem mxm;
6555 	struct mcx_cmdq_entry *cqe;
6556 	struct mcx_cmd_query_sq_in *in;
6557 	struct mcx_cmd_query_sq_out *out;
6558 	struct mcx_cmd_query_sq_mb_out *mbout;
6559 	uint8_t token = mcx_cmdq_token(sc);
6560 	int error;
6561 
6562 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
6563 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out) + sizeof(*mbout) + 16,
6564 	    token);
6565 
6566 	in = mcx_cmdq_in(cqe);
6567 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_SQ);
6568 	in->cmd_op_mod = htobe16(0);
6569 	in->cmd_sqn = htobe32(tx->tx_sqn);
6570 
6571 	CTASSERT(sizeof(*mbout) <= MCX_CMDQ_MAILBOX_DATASIZE*2);
6572 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2,
6573 	    &cqe->cq_output_ptr, token) != 0) {
6574 		printf("%s: unable to allocate query sq mailboxes\n", DEVNAME(sc));
6575 		return (-1);
6576 	}
6577 
6578 	mcx_cmdq_mboxes_sign(&mxm, 1);
6579 
6580 	mcx_cmdq_post(sc, cqe, 0);
6581 	error = mcx_cmdq_poll(sc, cqe, 1000);
6582 	if (error != 0) {
6583 		printf("%s: query sq timeout\n", DEVNAME(sc));
6584 		goto free;
6585 	}
6586 	error = mcx_cmdq_verify(cqe);
6587 	if (error != 0) {
6588 		printf("%s: query sq reply corrupt\n", DEVNAME(sc));
6589 		goto free;
6590 	}
6591 
6592 	out = mcx_cmdq_out(cqe);
6593 	switch (out->cmd_status) {
6594 	case MCX_CQ_STATUS_OK:
6595 		break;
6596 	default:
6597 		printf("%s: query sq failed (%x/%x)\n", DEVNAME(sc),
6598 		    out->cmd_status, be32toh(out->cmd_syndrome));
6599 		error = -1;
6600 		goto free;
6601 	}
6602 
6603         mbout = (struct mcx_cmd_query_sq_mb_out *)
6604 	    (mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)));
6605 	memcpy(sq_ctx, &mbout->cmd_ctx, sizeof(*sq_ctx));
6606 
6607 free:
6608 	mcx_cq_mboxes_free(sc, &mxm);
6609 	return (error);
6610 }
6611 
6612 int
6613 mcx_query_cq(struct mcx_softc *sc, struct mcx_cq *cq, struct mcx_cq_ctx *cq_ctx)
6614 {
6615 	struct mcx_dmamem mxm;
6616 	struct mcx_cmdq_entry *cqe;
6617 	struct mcx_cmd_query_cq_in *in;
6618 	struct mcx_cmd_query_cq_out *out;
6619 	struct mcx_cq_ctx *ctx;
6620 	uint8_t token = mcx_cmdq_token(sc);
6621 	int error;
6622 
6623 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
6624 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out) + sizeof(*ctx) + 16,
6625 	    token);
6626 
6627 	in = mcx_cmdq_in(cqe);
6628 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_CQ);
6629 	in->cmd_op_mod = htobe16(0);
6630 	in->cmd_cqn = htobe32(cq->cq_n);
6631 
6632 	CTASSERT(sizeof(*ctx) <= MCX_CMDQ_MAILBOX_DATASIZE*2);
6633 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2,
6634 	    &cqe->cq_output_ptr, token) != 0) {
6635 		printf("%s: unable to allocate query cq mailboxes\n",
6636 		    DEVNAME(sc));
6637 		return (-1);
6638 	}
6639 
6640 	mcx_cmdq_mboxes_sign(&mxm, 1);
6641 
6642 	mcx_cmdq_post(sc, cqe, 0);
6643 	error = mcx_cmdq_poll(sc, cqe, 1000);
6644 	if (error != 0) {
6645 		printf("%s: query cq timeout\n", DEVNAME(sc));
6646 		goto free;
6647 	}
6648 	if (mcx_cmdq_verify(cqe) != 0) {
6649 		printf("%s: query cq reply corrupt\n", DEVNAME(sc));
6650 		goto free;
6651 	}
6652 
6653 	out = mcx_cmdq_out(cqe);
6654 	switch (out->cmd_status) {
6655 	case MCX_CQ_STATUS_OK:
6656 		break;
6657 	default:
6658 		printf("%s: query qc failed (%x/%x)\n", DEVNAME(sc),
6659 		    out->cmd_status, be32toh(out->cmd_syndrome));
6660 		error = -1;
6661 		goto free;
6662 	}
6663 
6664 	ctx = (struct mcx_cq_ctx *)(mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)));
6665 	memcpy(cq_ctx, ctx, sizeof(*cq_ctx));
6666 free:
6667 	mcx_dmamem_free(sc, &mxm);
6668 	return (error);
6669 }
6670 
6671 int
6672 mcx_query_eq(struct mcx_softc *sc, struct mcx_eq *eq, struct mcx_eq_ctx *eq_ctx)
6673 {
6674 	struct mcx_dmamem mxm;
6675 	struct mcx_cmdq_entry *cqe;
6676 	struct mcx_cmd_query_eq_in *in;
6677 	struct mcx_cmd_query_eq_out *out;
6678 	struct mcq_eq_ctx *ctx;
6679 	uint8_t token = mcx_cmdq_token(sc);
6680 	int error;
6681 
6682 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
6683 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out) + sizeof(*ctx) + 16,
6684 	    token);
6685 
6686 	in = mcx_cmdq_in(cqe);
6687 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_EQ);
6688 	in->cmd_op_mod = htobe16(0);
6689 	in->cmd_eqn = htobe32(eq->eq_n);
6690 
6691 	CTASSERT(sizeof(*ctx) <= MCX_CMDQ_MAILBOX_DATASIZE*2);
6692 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2,
6693 	    &cqe->cq_output_ptr, token) != 0) {
6694 		printf("%s: unable to allocate query eq mailboxes\n",
6695 		    DEVNAME(sc));
6696 		return (-1);
6697 	}
6698 
6699 	mcx_cmdq_mboxes_sign(&mxm, 1);
6700 
6701 	mcx_cmdq_post(sc, cqe, 0);
6702 	error = mcx_cmdq_poll(sc, cqe, 1000);
6703 	if (error != 0) {
6704 		printf("%s: query eq timeout\n", DEVNAME(sc));
6705 		goto free;
6706 	}
6707 	if (mcx_cmdq_verify(cqe) != 0) {
6708 		printf("%s: query eq reply corrupt\n", DEVNAME(sc));
6709 		goto free;
6710 	}
6711 
6712 	out = mcx_cmdq_out(cqe);
6713 	switch (out->cmd_status) {
6714 	case MCX_CQ_STATUS_OK:
6715 		break;
6716 	default:
6717 		printf("%s: query eq failed (%x/%x)\n", DEVNAME(sc),
6718 		    out->cmd_status, be32toh(out->cmd_syndrome));
6719 		error = -1;
6720 		goto free;
6721 	}
6722 
6723 	ctx = (struct mcx_eq_ctx *)(mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)));
6724 	memcpy(eq_ctx, ctx, sizeof(*eq_ctx));
6725 free:
6726 	mcx_dmamem_free(sc, &mxm);
6727 	return (error);
6728 }
6729 
6730 #endif /* NKSTAT > 0 */
6731 
6732 
6733 static inline unsigned int
6734 mcx_rx_fill_slots(struct mcx_softc *sc, struct mcx_rx *rx, uint nslots)
6735 {
6736 	struct mcx_rq_entry *ring, *rqe;
6737 	struct mcx_slot *ms;
6738 	struct mbuf *m;
6739 	uint slot, p, fills;
6740 
6741 	ring = MCX_DMA_KVA(&rx->rx_rq_mem);
6742 	p = rx->rx_prod;
6743 
6744 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&rx->rx_rq_mem),
6745 	    0, MCX_DMA_LEN(&rx->rx_rq_mem), BUS_DMASYNC_POSTWRITE);
6746 
6747 	slot = (p % (1 << MCX_LOG_RQ_SIZE));
6748 	rqe = ring;
6749 	for (fills = 0; fills < nslots; fills++) {
6750 		slot = p % (1 << MCX_LOG_RQ_SIZE);
6751 
6752 		ms = &rx->rx_slots[slot];
6753 		rqe = &ring[slot];
6754 
6755 		m = NULL;
6756 		MGETHDR(m, M_DONTWAIT, MT_DATA);
6757 		if (m == NULL)
6758 			break;
6759 
6760 		MCLGET(m, M_DONTWAIT);
6761 		if ((m->m_flags & M_EXT) == 0) {
6762 			m_freem(m);
6763 			break;
6764 		}
6765 
6766 		m->m_len = m->m_pkthdr.len = sc->sc_hardmtu;
6767 		m_adj(m, m->m_ext.ext_size - sc->sc_rxbufsz);
6768 		m_adj(m, ETHER_ALIGN);
6769 
6770 		if (bus_dmamap_load_mbuf(sc->sc_dmat, ms->ms_map, m,
6771 		    BUS_DMA_NOWAIT) != 0) {
6772 			m_freem(m);
6773 			break;
6774 		}
6775 		bus_dmamap_sync(sc->sc_dmat, ms->ms_map, 0, ms->ms_map->dm_mapsize, BUS_DMASYNC_PREREAD);
6776 		ms->ms_m = m;
6777 
6778 		be32enc(&rqe->rqe_byte_count, ms->ms_map->dm_segs[0].ds_len);
6779 		be64enc(&rqe->rqe_addr, ms->ms_map->dm_segs[0].ds_addr);
6780 		be32enc(&rqe->rqe_lkey, sc->sc_lkey);
6781 
6782 		p++;
6783 	}
6784 
6785 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&rx->rx_rq_mem),
6786 	    0, MCX_DMA_LEN(&rx->rx_rq_mem), BUS_DMASYNC_PREWRITE);
6787 
6788 	rx->rx_prod = p;
6789 
6790 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_doorbell_mem),
6791 	    rx->rx_doorbell, sizeof(uint32_t), BUS_DMASYNC_POSTWRITE);
6792 	be32enc(MCX_DMA_OFF(&sc->sc_doorbell_mem, rx->rx_doorbell),
6793 	    p & MCX_WQ_DOORBELL_MASK);
6794 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_doorbell_mem),
6795 	    rx->rx_doorbell, sizeof(uint32_t), BUS_DMASYNC_PREWRITE);
6796 
6797 	return (nslots - fills);
6798 }
6799 
6800 static int
6801 mcx_rx_fill(struct mcx_softc *sc, struct mcx_rx *rx)
6802 {
6803 	u_int slots;
6804 
6805 	slots = mcx_rxr_get(&rx->rx_rxr, (1 << MCX_LOG_RQ_SIZE));
6806 	if (slots == 0)
6807 		return (1);
6808 
6809 	slots = mcx_rx_fill_slots(sc, rx, slots);
6810 	mcx_rxr_put(&rx->rx_rxr, slots);
6811 	return (0);
6812 }
6813 
6814 void
6815 mcx_refill(void *xrx)
6816 {
6817 	struct mcx_rx *rx = xrx;
6818 	struct mcx_softc *sc = rx->rx_softc;
6819 
6820 	mcx_rx_fill(sc, rx);
6821 
6822 	if (mcx_rxr_inuse(&rx->rx_rxr) == 0)
6823 		callout_schedule(&rx->rx_refill, 1);
6824 }
6825 
6826 static int
6827 mcx_process_txeof(struct mcx_softc *sc, struct mcx_tx *tx,
6828     struct mcx_cq_entry *cqe)
6829 {
6830 	struct mcx_slot *ms;
6831 	bus_dmamap_t map;
6832 	int slot, slots;
6833 
6834 	slot = be16toh(cqe->cq_wqe_count) % (1 << MCX_LOG_SQ_SIZE);
6835 
6836 	ms = &tx->tx_slots[slot];
6837 	map = ms->ms_map;
6838 	bus_dmamap_sync(sc->sc_dmat, map, 0, map->dm_mapsize,
6839 	    BUS_DMASYNC_POSTWRITE);
6840 
6841 	slots = 1;
6842 	if (map->dm_nsegs > 1)
6843 		slots += (map->dm_nsegs+2) / MCX_SQ_SEGS_PER_SLOT;
6844 
6845 	bus_dmamap_unload(sc->sc_dmat, map);
6846 	m_freem(ms->ms_m);
6847 	ms->ms_m = NULL;
6848 
6849 	return (slots);
6850 }
6851 
6852 static uint64_t
6853 mcx_uptime(void)
6854 {
6855 	struct timespec ts;
6856 
6857 	nanouptime(&ts);
6858 
6859 	return ((uint64_t)ts.tv_sec * 1000000000 + (uint64_t)ts.tv_nsec);
6860 }
6861 
6862 static void
6863 mcx_calibrate_first(struct mcx_softc *sc)
6864 {
6865 	struct mcx_calibration *c = &sc->sc_calibration[0];
6866 	int s;
6867 
6868 	sc->sc_calibration_gen = 0;
6869 
6870 	s = splhigh(); /* crit_enter? */
6871 	c->c_ubase = mcx_uptime();
6872 	c->c_tbase = mcx_timer(sc);
6873 	splx(s);
6874 	c->c_ratio = 0;
6875 
6876 #if notyet
6877 	callout_schedule(&sc->sc_calibrate, MCX_CALIBRATE_FIRST * hz);
6878 #endif
6879 }
6880 
6881 #define MCX_TIMESTAMP_SHIFT 24
6882 
6883 static void
6884 mcx_calibrate(void *arg)
6885 {
6886 	struct mcx_softc *sc = arg;
6887 	struct mcx_calibration *nc, *pc;
6888 	uint64_t udiff, tdiff;
6889 	unsigned int gen;
6890 	int s;
6891 
6892 	if (!ISSET(sc->sc_ec.ec_if.if_flags, IFF_RUNNING))
6893 		return;
6894 
6895 	callout_schedule(&sc->sc_calibrate, MCX_CALIBRATE_NORMAL * hz);
6896 
6897 	gen = sc->sc_calibration_gen;
6898 	pc = &sc->sc_calibration[gen % __arraycount(sc->sc_calibration)];
6899 	gen++;
6900 	nc = &sc->sc_calibration[gen % __arraycount(sc->sc_calibration)];
6901 
6902 	nc->c_uptime = pc->c_ubase;
6903 	nc->c_timestamp = pc->c_tbase;
6904 
6905 	s = splhigh(); /* crit_enter? */
6906 	nc->c_ubase = mcx_uptime();
6907 	nc->c_tbase = mcx_timer(sc);
6908 	splx(s);
6909 
6910 	udiff = nc->c_ubase - nc->c_uptime;
6911 	tdiff = nc->c_tbase - nc->c_timestamp;
6912 
6913 	/*
6914 	 * udiff is the wall clock time between calibration ticks,
6915 	 * which should be 32 seconds or 32 billion nanoseconds. if
6916 	 * we squint, 1 billion nanoseconds is kind of like a 32 bit
6917 	 * number, so 32 billion should still have a lot of high bits
6918 	 * spare. we use this space by shifting the nanoseconds up
6919 	 * 24 bits so we have a nice big number to divide by the
6920 	 * number of mcx timer ticks.
6921 	 */
6922 	nc->c_ratio = (udiff << MCX_TIMESTAMP_SHIFT) / tdiff;
6923 
6924 	membar_producer();
6925 	sc->sc_calibration_gen = gen;
6926 }
6927 
6928 static int
6929 mcx_process_rx(struct mcx_softc *sc, struct mcx_rx *rx,
6930     struct mcx_cq_entry *cqe, struct mcx_mbufq *mq,
6931     const struct mcx_calibration *c)
6932 {
6933 	struct ifnet *ifp = &sc->sc_ec.ec_if;
6934 	struct mcx_slot *ms;
6935 	struct mbuf *m;
6936 	uint32_t flags, len;
6937 	int slot;
6938 
6939 	len = be32dec(&cqe->cq_byte_cnt);
6940 	slot = be16toh(cqe->cq_wqe_count) % (1 << MCX_LOG_RQ_SIZE);
6941 
6942 	ms = &rx->rx_slots[slot];
6943 	bus_dmamap_sync(sc->sc_dmat, ms->ms_map, 0, len, BUS_DMASYNC_POSTREAD);
6944 	bus_dmamap_unload(sc->sc_dmat, ms->ms_map);
6945 
6946 	m = ms->ms_m;
6947 	ms->ms_m = NULL;
6948 
6949 	m_set_rcvif(m, &sc->sc_ec.ec_if);
6950 	m->m_pkthdr.len = m->m_len = len;
6951 
6952 #if 0
6953 	if (cqe->cq_rx_hash_type) {
6954 		m->m_pkthdr.ph_flowid = be32toh(cqe->cq_rx_hash);
6955 		m->m_pkthdr.csum_flags |= M_FLOWID;
6956 	}
6957 #endif
6958 
6959 	flags = be32dec(&cqe->cq_flags);
6960 	if (flags & MCX_CQ_ENTRY_FLAGS_L3_OK) {
6961 		if (ifp->if_capenable & IFCAP_CSUM_IPv4_Rx)
6962 			m->m_pkthdr.csum_flags |= M_CSUM_IPv4;
6963 	}
6964 	if (flags & MCX_CQ_ENTRY_FLAGS_L4_OK) {
6965 		if (ifp->if_capenable & IFCAP_CSUM_TCPv4_Rx)
6966 			m->m_pkthdr.csum_flags |= M_CSUM_TCPv4;
6967 		if (ifp->if_capenable & IFCAP_CSUM_TCPv6_Rx)
6968 			m->m_pkthdr.csum_flags |= M_CSUM_TCPv6;
6969 		if (ifp->if_capenable & IFCAP_CSUM_UDPv4_Rx)
6970 			m->m_pkthdr.csum_flags |= M_CSUM_UDPv4;
6971 		if (ifp->if_capenable & IFCAP_CSUM_UDPv6_Rx)
6972 			m->m_pkthdr.csum_flags |= M_CSUM_UDPv6;
6973 	}
6974 	if (flags & MCX_CQ_ENTRY_FLAGS_CV) {
6975 		vlan_set_tag(m, flags & MCX_CQ_ENTRY_FLAGS_VLAN_MASK);
6976 	}
6977 
6978 #if notyet
6979 	if (ISSET(sc->sc_ec.ec_if.if_flags, IFF_LINK0) && c->c_ratio) {
6980 		uint64_t t = be64dec(&cqe->cq_timestamp);
6981 		t -= c->c_timestamp;
6982 		t *= c->c_ratio;
6983 		t >>= MCX_TIMESTAMP_SHIFT;
6984 		t += c->c_uptime;
6985 
6986 		m->m_pkthdr.ph_timestamp = t;
6987 		SET(m->m_pkthdr.csum_flags, M_TIMESTAMP);
6988 	}
6989 #endif
6990 
6991 	MBUFQ_ENQUEUE(mq, m);
6992 
6993 	return (1);
6994 }
6995 
6996 static struct mcx_cq_entry *
6997 mcx_next_cq_entry(struct mcx_softc *sc, struct mcx_cq *cq)
6998 {
6999 	struct mcx_cq_entry *cqe;
7000 	int next;
7001 
7002 	cqe = (struct mcx_cq_entry *)MCX_DMA_KVA(&cq->cq_mem);
7003 	next = cq->cq_cons % (1 << MCX_LOG_CQ_SIZE);
7004 
7005 	if ((cqe[next].cq_opcode_owner & MCX_CQ_ENTRY_FLAG_OWNER) ==
7006 	    ((cq->cq_cons >> MCX_LOG_CQ_SIZE) & 1)) {
7007 		return (&cqe[next]);
7008 	}
7009 
7010 	return (NULL);
7011 }
7012 
7013 static void
7014 mcx_arm_cq(struct mcx_softc *sc, struct mcx_cq *cq, int uar)
7015 {
7016 	struct mcx_cq_doorbell *db;
7017 	bus_size_t offset;
7018 	uint32_t val;
7019 	uint64_t uval;
7020 
7021 	val = ((cq->cq_count) & 3) << MCX_CQ_DOORBELL_ARM_CMD_SN_SHIFT;
7022 	val |= (cq->cq_cons & MCX_CQ_DOORBELL_ARM_CI_MASK);
7023 
7024 	db = MCX_DMA_OFF(&sc->sc_doorbell_mem, cq->cq_doorbell);
7025 
7026 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_doorbell_mem),
7027 	    cq->cq_doorbell, sizeof(*db), BUS_DMASYNC_POSTWRITE);
7028 
7029 	be32enc(&db->db_update_ci, cq->cq_cons & MCX_CQ_DOORBELL_ARM_CI_MASK);
7030 	be32enc(&db->db_arm_ci, val);
7031 
7032 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_doorbell_mem),
7033 	    cq->cq_doorbell, sizeof(*db), BUS_DMASYNC_PREWRITE);
7034 
7035 	offset = (MCX_PAGE_SIZE * uar) + MCX_UAR_CQ_DOORBELL;
7036 
7037 	uval = (uint64_t)val << 32;
7038 	uval |= cq->cq_n;
7039 
7040 	bus_space_write_8(sc->sc_memt, sc->sc_memh, offset, htobe64(uval));
7041 	mcx_bar(sc, offset, sizeof(uval), BUS_SPACE_BARRIER_WRITE);
7042 }
7043 
7044 void
7045 mcx_process_cq(struct mcx_softc *sc, struct mcx_queues *q, struct mcx_cq *cq)
7046 {
7047 	struct mcx_rx *rx = &q->q_rx;
7048 	struct mcx_tx *tx = &q->q_tx;
7049 	struct ifnet *ifp = &sc->sc_ec.ec_if;
7050 	const struct mcx_calibration *c;
7051 	unsigned int gen;
7052 	struct mcx_cq_entry *cqe;
7053 	struct mcx_mbufq mq;
7054 	struct mbuf *m;
7055 	int rxfree, txfree;
7056 
7057 	MBUFQ_INIT(&mq);
7058 
7059 	gen = sc->sc_calibration_gen;
7060 	membar_consumer();
7061 	c = &sc->sc_calibration[gen % __arraycount(sc->sc_calibration)];
7062 
7063 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&cq->cq_mem),
7064 	    0, MCX_DMA_LEN(&cq->cq_mem), BUS_DMASYNC_POSTREAD);
7065 
7066 	rxfree = 0;
7067 	txfree = 0;
7068 	while ((cqe = mcx_next_cq_entry(sc, cq))) {
7069 		uint8_t opcode;
7070 		opcode = (cqe->cq_opcode_owner >> MCX_CQ_ENTRY_OPCODE_SHIFT);
7071 		switch (opcode) {
7072 		case MCX_CQ_ENTRY_OPCODE_REQ:
7073 			txfree += mcx_process_txeof(sc, tx, cqe);
7074 			break;
7075 		case MCX_CQ_ENTRY_OPCODE_SEND:
7076 			rxfree += mcx_process_rx(sc, rx, cqe, &mq, c);
7077 			break;
7078 		case MCX_CQ_ENTRY_OPCODE_REQ_ERR:
7079 		case MCX_CQ_ENTRY_OPCODE_SEND_ERR:
7080 			/* uint8_t *cqp = (uint8_t *)cqe; */
7081 			/* printf("%s: cq completion error: %x\n",
7082 			    DEVNAME(sc), cqp[0x37]); */
7083 			break;
7084 
7085 		default:
7086 			/* printf("%s: cq completion opcode %x??\n",
7087 			    DEVNAME(sc), opcode); */
7088 			break;
7089 		}
7090 
7091 		cq->cq_cons++;
7092 	}
7093 
7094 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&cq->cq_mem),
7095 	    0, MCX_DMA_LEN(&cq->cq_mem), BUS_DMASYNC_PREREAD);
7096 
7097 	if (rxfree > 0) {
7098 		mcx_rxr_put(&rx->rx_rxr, rxfree);
7099 		while (MBUFQ_FIRST(&mq) != NULL) {
7100 			MBUFQ_DEQUEUE(&mq, m);
7101 			if_percpuq_enqueue(ifp->if_percpuq, m);
7102 		}
7103 
7104 		mcx_rx_fill(sc, rx);
7105 		if (mcx_rxr_inuse(&rx->rx_rxr) == 0)
7106 			callout_schedule(&rx->rx_refill, 1);
7107 	}
7108 
7109 	cq->cq_count++;
7110 	mcx_arm_cq(sc, cq, q->q_uar);
7111 
7112 	if (txfree > 0) {
7113 		tx->tx_cons += txfree;
7114 		if_schedule_deferred_start(ifp);
7115 	}
7116 }
7117 
7118 
7119 static void
7120 mcx_arm_eq(struct mcx_softc *sc, struct mcx_eq *eq, int uar)
7121 {
7122 	bus_size_t offset;
7123 	uint32_t val;
7124 
7125 	offset = (MCX_PAGE_SIZE * uar) + MCX_UAR_EQ_DOORBELL_ARM;
7126 	val = (eq->eq_n << 24) | (eq->eq_cons & 0xffffff);
7127 
7128 	mcx_wr(sc, offset, val);
7129 	mcx_bar(sc, offset, sizeof(val), BUS_SPACE_BARRIER_WRITE);
7130 }
7131 
7132 static struct mcx_eq_entry *
7133 mcx_next_eq_entry(struct mcx_softc *sc, struct mcx_eq *eq)
7134 {
7135 	struct mcx_eq_entry *eqe;
7136 	int next;
7137 
7138 	eqe = (struct mcx_eq_entry *)MCX_DMA_KVA(&eq->eq_mem);
7139 	next = eq->eq_cons % (1 << MCX_LOG_EQ_SIZE);
7140 	if ((eqe[next].eq_owner & 1) ==
7141 	    ((eq->eq_cons >> MCX_LOG_EQ_SIZE) & 1)) {
7142 		eq->eq_cons++;
7143 		return (&eqe[next]);
7144 	}
7145 	return (NULL);
7146 }
7147 
7148 int
7149 mcx_admin_intr(void *xsc)
7150 {
7151 	struct mcx_softc *sc = (struct mcx_softc *)xsc;
7152 	struct mcx_eq *eq = &sc->sc_admin_eq;
7153 	struct mcx_eq_entry *eqe;
7154 
7155 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&eq->eq_mem),
7156 	    0, MCX_DMA_LEN(&eq->eq_mem), BUS_DMASYNC_POSTREAD);
7157 
7158 	while ((eqe = mcx_next_eq_entry(sc, eq)) != NULL) {
7159 		switch (eqe->eq_event_type) {
7160 		case MCX_EVENT_TYPE_LAST_WQE:
7161 			/* printf("%s: last wqe reached?\n", DEVNAME(sc)); */
7162 			break;
7163 
7164 		case MCX_EVENT_TYPE_CQ_ERROR:
7165 			/* printf("%s: cq error\n", DEVNAME(sc)); */
7166 			break;
7167 
7168 		case MCX_EVENT_TYPE_CMD_COMPLETION:
7169 			/* wakeup probably */
7170 			break;
7171 
7172 		case MCX_EVENT_TYPE_PORT_CHANGE:
7173 			workqueue_enqueue(sc->sc_workq, &sc->sc_port_change, NULL);
7174 			break;
7175 
7176 		default:
7177 			/* printf("%s: something happened\n", DEVNAME(sc)); */
7178 			break;
7179 		}
7180 	}
7181 
7182 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&eq->eq_mem),
7183 	    0, MCX_DMA_LEN(&eq->eq_mem), BUS_DMASYNC_PREREAD);
7184 
7185 	mcx_arm_eq(sc, eq, sc->sc_uar);
7186 
7187 	return (1);
7188 }
7189 
7190 int
7191 mcx_cq_intr(void *xq)
7192 {
7193 	struct mcx_queues *q = (struct mcx_queues *)xq;
7194 	struct mcx_softc *sc = q->q_sc;
7195 	struct mcx_eq *eq = &q->q_eq;
7196 	struct mcx_eq_entry *eqe;
7197 	int cqn;
7198 
7199 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&eq->eq_mem),
7200 	    0, MCX_DMA_LEN(&eq->eq_mem), BUS_DMASYNC_POSTREAD);
7201 
7202 	while ((eqe = mcx_next_eq_entry(sc, eq)) != NULL) {
7203 		switch (eqe->eq_event_type) {
7204 		case MCX_EVENT_TYPE_COMPLETION:
7205 			cqn = be32toh(eqe->eq_event_data[6]);
7206 			if (cqn == q->q_cq.cq_n)
7207 				mcx_process_cq(sc, q, &q->q_cq);
7208 			break;
7209 		}
7210 	}
7211 
7212 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&eq->eq_mem),
7213 	    0, MCX_DMA_LEN(&eq->eq_mem), BUS_DMASYNC_PREREAD);
7214 
7215 	mcx_arm_eq(sc, eq, q->q_uar);
7216 
7217 	return (1);
7218 }
7219 
7220 static void
7221 mcx_free_slots(struct mcx_softc *sc, struct mcx_slot *slots, int allocated,
7222     int total)
7223 {
7224 	struct mcx_slot *ms;
7225 
7226 	int i = allocated;
7227 	while (i-- > 0) {
7228 		ms = &slots[i];
7229 		bus_dmamap_destroy(sc->sc_dmat, ms->ms_map);
7230 		if (ms->ms_m != NULL)
7231 			m_freem(ms->ms_m);
7232 	}
7233 	kmem_free(slots, total * sizeof(*ms));
7234 }
7235 
7236 static int
7237 mcx_queue_up(struct mcx_softc *sc, struct mcx_queues *q)
7238 {
7239 	struct mcx_rx *rx;
7240 	struct mcx_tx *tx;
7241 	struct mcx_slot *ms;
7242 	int i;
7243 
7244 	rx = &q->q_rx;
7245 	rx->rx_slots = kmem_zalloc(sizeof(*ms) * (1 << MCX_LOG_RQ_SIZE),
7246 	    KM_SLEEP);
7247 
7248 	for (i = 0; i < (1 << MCX_LOG_RQ_SIZE); i++) {
7249 		ms = &rx->rx_slots[i];
7250 		if (bus_dmamap_create(sc->sc_dmat, sc->sc_hardmtu, 1,
7251 		    sc->sc_hardmtu, 0,
7252 		    BUS_DMA_WAITOK | BUS_DMA_ALLOCNOW,
7253 		    &ms->ms_map) != 0) {
7254 			printf("%s: failed to allocate rx dma maps\n",
7255 			    DEVNAME(sc));
7256 			goto destroy_rx_slots;
7257 		}
7258 	}
7259 
7260 	tx = &q->q_tx;
7261 	tx->tx_slots = kmem_zalloc(sizeof(*ms) * (1 << MCX_LOG_SQ_SIZE),
7262 	     KM_SLEEP);
7263 
7264 	for (i = 0; i < (1 << MCX_LOG_SQ_SIZE); i++) {
7265 		ms = &tx->tx_slots[i];
7266 		if (bus_dmamap_create(sc->sc_dmat, sc->sc_hardmtu,
7267 		    MCX_SQ_MAX_SEGMENTS, sc->sc_hardmtu, 0,
7268 		    BUS_DMA_WAITOK | BUS_DMA_ALLOCNOW,
7269 		    &ms->ms_map) != 0) {
7270 			printf("%s: failed to allocate tx dma maps\n",
7271 			    DEVNAME(sc));
7272 			goto destroy_tx_slots;
7273 		}
7274 	}
7275 
7276 	if (mcx_create_cq(sc, &q->q_cq, q->q_uar, q->q_index,
7277 	    q->q_eq.eq_n) != 0)
7278 		goto destroy_tx_slots;
7279 
7280 	if (mcx_create_sq(sc, tx, q->q_uar, q->q_index, q->q_cq.cq_n)
7281 	    != 0)
7282 		goto destroy_cq;
7283 
7284 	if (mcx_create_rq(sc, rx, q->q_index, q->q_cq.cq_n) != 0)
7285 		goto destroy_sq;
7286 
7287 	return 0;
7288 
7289 destroy_sq:
7290 	mcx_destroy_sq(sc, tx);
7291 destroy_cq:
7292 	mcx_destroy_cq(sc, &q->q_cq);
7293 destroy_tx_slots:
7294 	mcx_free_slots(sc, tx->tx_slots, i, (1 << MCX_LOG_SQ_SIZE));
7295 	tx->tx_slots = NULL;
7296 
7297 	i = (1 << MCX_LOG_RQ_SIZE);
7298 destroy_rx_slots:
7299 	mcx_free_slots(sc, rx->rx_slots, i, (1 << MCX_LOG_RQ_SIZE));
7300 	rx->rx_slots = NULL;
7301 	return ENOMEM;
7302 }
7303 
7304 static int
7305 mcx_rss_group_entry_count(struct mcx_softc *sc, int group)
7306 {
7307 	int i;
7308 	int count;
7309 
7310 	count = 0;
7311 	for (i = 0; i < __arraycount(mcx_rss_config); i++) {
7312 		if (mcx_rss_config[i].flow_group == group)
7313 			count++;
7314 	}
7315 
7316 	return count;
7317 }
7318 
7319 static int
7320 mcx_init(struct ifnet *ifp)
7321 {
7322 	struct mcx_softc *sc = ifp->if_softc;
7323 	struct mcx_rx *rx;
7324 	struct mcx_tx *tx;
7325 	int i, start, count, flow_group, flow_index;
7326 	struct mcx_flow_match match_crit;
7327 	struct mcx_rss_rule *rss;
7328 	uint32_t dest;
7329 	int rqns[MCX_MAX_QUEUES] = { 0 };
7330 
7331 	if (ISSET(ifp->if_flags, IFF_RUNNING))
7332 		mcx_stop(ifp, 0);
7333 
7334 	if (mcx_create_tis(sc, &sc->sc_tis) != 0)
7335 		goto down;
7336 
7337 	for (i = 0; i < sc->sc_nqueues; i++) {
7338 		if (mcx_queue_up(sc, &sc->sc_queues[i]) != 0) {
7339 			goto down;
7340 		}
7341 	}
7342 
7343 	/* RSS flow table and flow groups */
7344 	if (mcx_create_flow_table(sc, MCX_LOG_FLOW_TABLE_SIZE, 1,
7345 	    &sc->sc_rss_flow_table_id) != 0)
7346 		goto down;
7347 
7348 	dest = MCX_FLOW_CONTEXT_DEST_TYPE_TABLE |
7349 	    sc->sc_rss_flow_table_id;
7350 
7351 	/* L4 RSS flow group (v4/v6 tcp/udp, no fragments) */
7352 	memset(&match_crit, 0, sizeof(match_crit));
7353 	match_crit.mc_ethertype = 0xffff;
7354 	match_crit.mc_ip_proto = 0xff;
7355 	match_crit.mc_vlan_flags = MCX_FLOW_MATCH_IP_FRAG;
7356 	start = 0;
7357 	count = mcx_rss_group_entry_count(sc, MCX_FLOW_GROUP_RSS_L4);
7358 	if (count != 0) {
7359 		if (mcx_create_flow_group(sc, sc->sc_rss_flow_table_id,
7360 		    MCX_FLOW_GROUP_RSS_L4, start, count,
7361 		    MCX_CREATE_FLOW_GROUP_CRIT_OUTER, &match_crit) != 0)
7362 			goto down;
7363 		start += count;
7364 	}
7365 
7366 	/* L3 RSS flow group (v4/v6, including fragments) */
7367 	memset(&match_crit, 0, sizeof(match_crit));
7368 	match_crit.mc_ethertype = 0xffff;
7369 	count = mcx_rss_group_entry_count(sc, MCX_FLOW_GROUP_RSS_L3);
7370 	if (mcx_create_flow_group(sc, sc->sc_rss_flow_table_id,
7371 	    MCX_FLOW_GROUP_RSS_L3, start, count,
7372 	    MCX_CREATE_FLOW_GROUP_CRIT_OUTER, &match_crit) != 0)
7373 		goto down;
7374 	start += count;
7375 
7376 	/* non-RSS flow group */
7377 	count = mcx_rss_group_entry_count(sc, MCX_FLOW_GROUP_RSS_NONE);
7378 	memset(&match_crit, 0, sizeof(match_crit));
7379 	if (mcx_create_flow_group(sc, sc->sc_rss_flow_table_id,
7380 	    MCX_FLOW_GROUP_RSS_NONE, start, count, 0, &match_crit) != 0)
7381 		goto down;
7382 
7383 	/* Root flow table, matching packets based on mac address */
7384 	if (mcx_create_flow_table(sc, MCX_LOG_FLOW_TABLE_SIZE, 0,
7385 	    &sc->sc_mac_flow_table_id) != 0)
7386 		goto down;
7387 
7388 	/* promisc flow group */
7389 	start = 0;
7390 	memset(&match_crit, 0, sizeof(match_crit));
7391 	if (mcx_create_flow_group(sc, sc->sc_mac_flow_table_id,
7392 	    MCX_FLOW_GROUP_PROMISC, start, 1, 0, &match_crit) != 0)
7393 		goto down;
7394 	sc->sc_promisc_flow_enabled = 0;
7395 	start++;
7396 
7397 	/* all multicast flow group */
7398 	match_crit.mc_dest_mac[0] = 0x01;
7399 	if (mcx_create_flow_group(sc, sc->sc_mac_flow_table_id,
7400 	    MCX_FLOW_GROUP_ALLMULTI, start, 1,
7401 	    MCX_CREATE_FLOW_GROUP_CRIT_OUTER, &match_crit) != 0)
7402 		goto down;
7403 	sc->sc_allmulti_flow_enabled = 0;
7404 	start++;
7405 
7406 	/* mac address matching flow group */
7407 	memset(&match_crit.mc_dest_mac, 0xff, sizeof(match_crit.mc_dest_mac));
7408 	if (mcx_create_flow_group(sc, sc->sc_mac_flow_table_id,
7409 	    MCX_FLOW_GROUP_MAC, start, (1 << MCX_LOG_FLOW_TABLE_SIZE) - start,
7410 	    MCX_CREATE_FLOW_GROUP_CRIT_OUTER, &match_crit) != 0)
7411 		goto down;
7412 
7413 	/* flow table entries for unicast and broadcast */
7414 	start = 0;
7415 	if (mcx_set_flow_table_entry_mac(sc, MCX_FLOW_GROUP_MAC, start,
7416 	    LLADDR(satosdl(ifp->if_dl->ifa_addr)), dest) != 0)
7417 		goto down;
7418 	start++;
7419 
7420 	if (mcx_set_flow_table_entry_mac(sc, MCX_FLOW_GROUP_MAC, start,
7421 	    etherbroadcastaddr, dest) != 0)
7422 		goto down;
7423 	start++;
7424 
7425 	/* multicast entries go after that */
7426 	sc->sc_mcast_flow_base = start;
7427 
7428 	/* re-add any existing multicast flows */
7429 	for (i = 0; i < MCX_NUM_MCAST_FLOWS; i++) {
7430 		if (sc->sc_mcast_flows[i][0] != 0) {
7431 			mcx_set_flow_table_entry_mac(sc, MCX_FLOW_GROUP_MAC,
7432 			    sc->sc_mcast_flow_base + i,
7433 			    sc->sc_mcast_flows[i], dest);
7434 		}
7435 	}
7436 
7437 	if (mcx_set_flow_table_root(sc, sc->sc_mac_flow_table_id) != 0)
7438 		goto down;
7439 
7440 	/*
7441 	 * the RQT can be any size as long as it's a power of two.
7442 	 * since we also restrict the number of queues to a power of two,
7443 	 * we can just put each rx queue in once.
7444 	 */
7445 	for (i = 0; i < sc->sc_nqueues; i++)
7446 		rqns[i] = sc->sc_queues[i].q_rx.rx_rqn;
7447 
7448 	if (mcx_create_rqt(sc, sc->sc_nqueues, rqns, &sc->sc_rqt) != 0)
7449 		goto down;
7450 
7451 	start = 0;
7452 	flow_index = 0;
7453 	flow_group = -1;
7454 	for (i = 0; i < __arraycount(mcx_rss_config); i++) {
7455 		rss = &mcx_rss_config[i];
7456 		if (rss->flow_group != flow_group) {
7457 			flow_group = rss->flow_group;
7458 			flow_index = 0;
7459 		}
7460 
7461 		if (rss->hash_sel == 0) {
7462 			if (mcx_create_tir_direct(sc, &sc->sc_queues[0].q_rx,
7463 			    &sc->sc_tir[i]) != 0)
7464 				goto down;
7465 		} else {
7466 			if (mcx_create_tir_indirect(sc, sc->sc_rqt,
7467 			    rss->hash_sel, &sc->sc_tir[i]) != 0)
7468 				goto down;
7469 		}
7470 
7471 		if (mcx_set_flow_table_entry_proto(sc, flow_group,
7472 		    flow_index, rss->ethertype, rss->ip_proto,
7473 		    MCX_FLOW_CONTEXT_DEST_TYPE_TIR | sc->sc_tir[i]) != 0)
7474 			goto down;
7475 		flow_index++;
7476 	}
7477 
7478 	for (i = 0; i < sc->sc_nqueues; i++) {
7479 		struct mcx_queues *q = &sc->sc_queues[i];
7480 		rx = &q->q_rx;
7481 		tx = &q->q_tx;
7482 
7483 		/* start the queues */
7484 		if (mcx_ready_sq(sc, tx) != 0)
7485 			goto down;
7486 
7487 		if (mcx_ready_rq(sc, rx) != 0)
7488 			goto down;
7489 
7490 		mcx_rxr_init(&rx->rx_rxr, 1, (1 << MCX_LOG_RQ_SIZE));
7491 		rx->rx_prod = 0;
7492 		mcx_rx_fill(sc, rx);
7493 
7494 		tx->tx_cons = 0;
7495 		tx->tx_prod = 0;
7496 	}
7497 
7498 	mcx_calibrate_first(sc);
7499 
7500 	SET(ifp->if_flags, IFF_RUNNING);
7501 	CLR(ifp->if_flags, IFF_OACTIVE);
7502 	if_schedule_deferred_start(ifp);
7503 
7504 	return 0;
7505 down:
7506 	mcx_stop(ifp, 0);
7507 	return EIO;
7508 }
7509 
7510 static void
7511 mcx_stop(struct ifnet *ifp, int disable)
7512 {
7513 	struct mcx_softc *sc = ifp->if_softc;
7514 	struct mcx_rss_rule *rss;
7515 	int group, i, flow_group, flow_index;
7516 
7517 	CLR(ifp->if_flags, IFF_RUNNING);
7518 
7519 	/*
7520 	 * delete flow table entries first, so no packets can arrive
7521 	 * after the barriers
7522 	 */
7523 	if (sc->sc_promisc_flow_enabled)
7524 		mcx_delete_flow_table_entry(sc, MCX_FLOW_GROUP_PROMISC, 0);
7525 	if (sc->sc_allmulti_flow_enabled)
7526 		mcx_delete_flow_table_entry(sc, MCX_FLOW_GROUP_ALLMULTI, 0);
7527 	mcx_delete_flow_table_entry(sc, MCX_FLOW_GROUP_MAC, 0);
7528 	mcx_delete_flow_table_entry(sc, MCX_FLOW_GROUP_MAC, 1);
7529 	for (i = 0; i < MCX_NUM_MCAST_FLOWS; i++) {
7530 		if (sc->sc_mcast_flows[i][0] != 0) {
7531 			mcx_delete_flow_table_entry(sc, MCX_FLOW_GROUP_MAC,
7532 			    sc->sc_mcast_flow_base + i);
7533 		}
7534 	}
7535 
7536 	flow_group = -1;
7537 	flow_index = 0;
7538 	for (i = 0; i < __arraycount(mcx_rss_config); i++) {
7539 		rss = &mcx_rss_config[i];
7540 		if (rss->flow_group != flow_group) {
7541 			flow_group = rss->flow_group;
7542 			flow_index = 0;
7543 		}
7544 
7545 		mcx_delete_flow_table_entry(sc, flow_group, flow_index);
7546 
7547 		mcx_destroy_tir(sc, sc->sc_tir[i]);
7548 		sc->sc_tir[i] = 0;
7549 
7550 		flow_index++;
7551 	}
7552 
7553 	for (i = 0; i < sc->sc_nqueues; i++) {
7554 		callout_halt(&sc->sc_queues[i].q_rx.rx_refill, NULL);
7555 	}
7556 
7557 	callout_halt(&sc->sc_calibrate, NULL);
7558 
7559 	for (group = 0; group < MCX_NUM_FLOW_GROUPS; group++) {
7560 		if (sc->sc_flow_group[group].g_id != -1)
7561 			mcx_destroy_flow_group(sc, group);
7562 	}
7563 
7564 	if (sc->sc_mac_flow_table_id != -1) {
7565 		mcx_destroy_flow_table(sc, sc->sc_mac_flow_table_id);
7566 		sc->sc_mac_flow_table_id = -1;
7567 	}
7568 	if (sc->sc_rss_flow_table_id != -1) {
7569 		mcx_destroy_flow_table(sc, sc->sc_rss_flow_table_id);
7570 		sc->sc_rss_flow_table_id = -1;
7571 	}
7572 	if (sc->sc_rqt != -1) {
7573 		mcx_destroy_rqt(sc, sc->sc_rqt);
7574 		sc->sc_rqt = -1;
7575 	}
7576 
7577 	for (i = 0; i < sc->sc_nqueues; i++) {
7578 		struct mcx_queues *q = &sc->sc_queues[i];
7579 		struct mcx_rx *rx = &q->q_rx;
7580 		struct mcx_tx *tx = &q->q_tx;
7581 		struct mcx_cq *cq = &q->q_cq;
7582 
7583 		if (rx->rx_rqn != 0)
7584 			mcx_destroy_rq(sc, rx);
7585 
7586 		if (tx->tx_sqn != 0)
7587 			mcx_destroy_sq(sc, tx);
7588 
7589 		if (tx->tx_slots != NULL) {
7590 			mcx_free_slots(sc, tx->tx_slots,
7591 			    (1 << MCX_LOG_SQ_SIZE), (1 << MCX_LOG_SQ_SIZE));
7592 			tx->tx_slots = NULL;
7593 		}
7594 		if (rx->rx_slots != NULL) {
7595 			mcx_free_slots(sc, rx->rx_slots,
7596 			    (1 << MCX_LOG_RQ_SIZE), (1 << MCX_LOG_RQ_SIZE));
7597 			rx->rx_slots = NULL;
7598 		}
7599 
7600 		if (cq->cq_n != 0)
7601 			mcx_destroy_cq(sc, cq);
7602 	}
7603 	if (sc->sc_tis != 0) {
7604 		mcx_destroy_tis(sc, sc->sc_tis);
7605 		sc->sc_tis = 0;
7606 	}
7607 }
7608 
7609 static int
7610 mcx_ioctl(struct ifnet *ifp, u_long cmd, void *data)
7611 {
7612 	struct mcx_softc *sc = (struct mcx_softc *)ifp->if_softc;
7613 	struct ifreq *ifr = (struct ifreq *)data;
7614 	struct ethercom *ec = &sc->sc_ec;
7615 	uint8_t addrhi[ETHER_ADDR_LEN], addrlo[ETHER_ADDR_LEN];
7616 	struct ether_multi *enm;
7617 	struct ether_multistep step;
7618 	int s, i, flags, error = 0;
7619 	uint32_t dest;
7620 
7621 	s = splnet();
7622 	switch (cmd) {
7623 
7624 	case SIOCADDMULTI:
7625 		if (ether_addmulti(ifreq_getaddr(cmd, ifr), &sc->sc_ec) == ENETRESET) {
7626 			error = ether_multiaddr(&ifr->ifr_addr, addrlo, addrhi);
7627 			if (error != 0) {
7628 				splx(s);
7629 				return (error);
7630 			}
7631 
7632 			dest = MCX_FLOW_CONTEXT_DEST_TYPE_TABLE |
7633 			    sc->sc_rss_flow_table_id;
7634 
7635 			for (i = 0; i < MCX_NUM_MCAST_FLOWS; i++) {
7636 				if (sc->sc_mcast_flows[i][0] == 0) {
7637 					memcpy(sc->sc_mcast_flows[i], addrlo,
7638 					    ETHER_ADDR_LEN);
7639 					if (ISSET(ifp->if_flags, IFF_RUNNING)) {
7640 						mcx_set_flow_table_entry_mac(sc,
7641 						    MCX_FLOW_GROUP_MAC,
7642 						    sc->sc_mcast_flow_base + i,
7643 						    sc->sc_mcast_flows[i], dest);
7644 					}
7645 					break;
7646 				}
7647 			}
7648 
7649 			if (!ISSET(ifp->if_flags, IFF_ALLMULTI)) {
7650 				if (i == MCX_NUM_MCAST_FLOWS) {
7651 					SET(ifp->if_flags, IFF_ALLMULTI);
7652 					sc->sc_extra_mcast++;
7653 					error = ENETRESET;
7654 				}
7655 
7656 				if (memcmp(addrlo, addrhi, ETHER_ADDR_LEN)) {
7657 					SET(ifp->if_flags, IFF_ALLMULTI);
7658 					error = ENETRESET;
7659 				}
7660 			}
7661 		}
7662 		break;
7663 
7664 	case SIOCDELMULTI:
7665 		if (ether_delmulti(ifreq_getaddr(cmd, ifr), &sc->sc_ec) == ENETRESET) {
7666 			error = ether_multiaddr(&ifr->ifr_addr, addrlo, addrhi);
7667 			if (error != 0) {
7668 				splx(s);
7669 				return (error);
7670 			}
7671 
7672 			for (i = 0; i < MCX_NUM_MCAST_FLOWS; i++) {
7673 				if (memcmp(sc->sc_mcast_flows[i], addrlo,
7674 				    ETHER_ADDR_LEN) == 0) {
7675 					if (ISSET(ifp->if_flags, IFF_RUNNING)) {
7676 						mcx_delete_flow_table_entry(sc,
7677 						    MCX_FLOW_GROUP_MAC,
7678 						    sc->sc_mcast_flow_base + i);
7679 					}
7680 					sc->sc_mcast_flows[i][0] = 0;
7681 					break;
7682 				}
7683 			}
7684 
7685 			if (i == MCX_NUM_MCAST_FLOWS)
7686 				sc->sc_extra_mcast--;
7687 
7688 			if (ISSET(ifp->if_flags, IFF_ALLMULTI) &&
7689 			    sc->sc_extra_mcast == 0) {
7690 				flags = 0;
7691 				ETHER_LOCK(ec);
7692 				ETHER_FIRST_MULTI(step, ec, enm);
7693 				while (enm != NULL) {
7694 					if (memcmp(enm->enm_addrlo,
7695 					    enm->enm_addrhi, ETHER_ADDR_LEN)) {
7696 						SET(flags, IFF_ALLMULTI);
7697 						break;
7698 					}
7699 					ETHER_NEXT_MULTI(step, enm);
7700 				}
7701 				ETHER_UNLOCK(ec);
7702 				if (!ISSET(flags, IFF_ALLMULTI)) {
7703 					CLR(ifp->if_flags, IFF_ALLMULTI);
7704 					error = ENETRESET;
7705 				}
7706 			}
7707 		}
7708 		break;
7709 
7710 	default:
7711 		error = ether_ioctl(ifp, cmd, data);
7712 	}
7713 
7714 	if (error == ENETRESET) {
7715 		if ((ifp->if_flags & (IFF_UP | IFF_RUNNING)) ==
7716 		    (IFF_UP | IFF_RUNNING))
7717 			mcx_iff(sc);
7718 		error = 0;
7719 	}
7720 	splx(s);
7721 
7722 	return (error);
7723 }
7724 
7725 #if 0
7726 static int
7727 mcx_get_sffpage(struct ifnet *ifp, struct if_sffpage *sff)
7728 {
7729 	struct mcx_softc *sc = (struct mcx_softc *)ifp->if_softc;
7730 	struct mcx_reg_mcia mcia;
7731 	struct mcx_reg_pmlp pmlp;
7732 	int offset, error;
7733 
7734 	/* get module number */
7735 	memset(&pmlp, 0, sizeof(pmlp));
7736 	pmlp.rp_local_port = 1;
7737 	error = mcx_access_hca_reg(sc, MCX_REG_PMLP, MCX_REG_OP_READ, &pmlp,
7738 	    sizeof(pmlp));
7739 	if (error != 0) {
7740 		printf("%s: unable to get eeprom module number\n",
7741 		    DEVNAME(sc));
7742 		return error;
7743 	}
7744 
7745 	for (offset = 0; offset < 256; offset += MCX_MCIA_EEPROM_BYTES) {
7746 		memset(&mcia, 0, sizeof(mcia));
7747 		mcia.rm_l = 0;
7748 		mcia.rm_module = be32toh(pmlp.rp_lane0_mapping) &
7749 		    MCX_PMLP_MODULE_NUM_MASK;
7750 		mcia.rm_i2c_addr = sff->sff_addr / 2;	/* apparently */
7751 		mcia.rm_page_num = sff->sff_page;
7752 		mcia.rm_dev_addr = htobe16(offset);
7753 		mcia.rm_size = htobe16(MCX_MCIA_EEPROM_BYTES);
7754 
7755 		error = mcx_access_hca_reg(sc, MCX_REG_MCIA, MCX_REG_OP_READ,
7756 		    &mcia, sizeof(mcia));
7757 		if (error != 0) {
7758 			printf("%s: unable to read eeprom at %x\n",
7759 			    DEVNAME(sc), offset);
7760 			return error;
7761 		}
7762 
7763 		memcpy(sff->sff_data + offset, mcia.rm_data,
7764 		    MCX_MCIA_EEPROM_BYTES);
7765 	}
7766 
7767 	return 0;
7768 }
7769 #endif
7770 
7771 static int
7772 mcx_load_mbuf(struct mcx_softc *sc, struct mcx_slot *ms, struct mbuf *m)
7773 {
7774 	switch (bus_dmamap_load_mbuf(sc->sc_dmat, ms->ms_map, m,
7775 	    BUS_DMA_STREAMING | BUS_DMA_NOWAIT)) {
7776 	case 0:
7777 		break;
7778 
7779 	case EFBIG:
7780 		if (m_defrag(m, M_DONTWAIT) != NULL &&
7781 		    bus_dmamap_load_mbuf(sc->sc_dmat, ms->ms_map, m,
7782 		    BUS_DMA_STREAMING | BUS_DMA_NOWAIT) == 0)
7783 			break;
7784 
7785 		/* FALLTHROUGH */
7786 	default:
7787 		return (1);
7788 	}
7789 
7790 	ms->ms_m = m;
7791 	return (0);
7792 }
7793 
7794 static void
7795 mcx_send_common_locked(struct ifnet *ifp, struct mcx_tx *tx, bool is_transmit)
7796 {
7797 	struct mcx_softc *sc = ifp->if_softc;
7798 	struct mcx_sq_entry *sq, *sqe;
7799 	struct mcx_sq_entry_seg *sqs;
7800 	struct mcx_slot *ms;
7801 	bus_dmamap_t map;
7802 	struct mbuf *m;
7803 	u_int idx, free, used;
7804 	uint64_t *bf;
7805 	uint32_t csum;
7806 	size_t bf_base;
7807 	int i, seg, nseg;
7808 
7809 	KASSERT(mutex_owned(&tx->tx_lock));
7810 
7811 	if ((ifp->if_flags & IFF_RUNNING) == 0)
7812 		return;
7813 
7814 	bf_base = (tx->tx_uar * MCX_PAGE_SIZE) + MCX_UAR_BF;
7815 
7816 	idx = tx->tx_prod % (1 << MCX_LOG_SQ_SIZE);
7817 	free = (tx->tx_cons + (1 << MCX_LOG_SQ_SIZE)) - tx->tx_prod;
7818 
7819 	used = 0;
7820 	bf = NULL;
7821 
7822 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&tx->tx_sq_mem),
7823 	    0, MCX_DMA_LEN(&tx->tx_sq_mem), BUS_DMASYNC_POSTWRITE);
7824 
7825 	sq = (struct mcx_sq_entry *)MCX_DMA_KVA(&tx->tx_sq_mem);
7826 
7827 	for (;;) {
7828 		if (used + MCX_SQ_ENTRY_MAX_SLOTS >= free) {
7829 			SET(ifp->if_flags, IFF_OACTIVE);
7830 			break;
7831 		}
7832 
7833 		if (is_transmit) {
7834 			m = pcq_get(tx->tx_pcq);
7835 		} else {
7836 			IFQ_DEQUEUE(&ifp->if_snd, m);
7837 		}
7838 		if (m == NULL) {
7839 			break;
7840 		}
7841 
7842 		sqe = sq + idx;
7843 		ms = &tx->tx_slots[idx];
7844 		memset(sqe, 0, sizeof(*sqe));
7845 
7846 		/* ctrl segment */
7847 		sqe->sqe_opcode_index = htobe32(MCX_SQE_WQE_OPCODE_SEND |
7848 		    ((tx->tx_prod & 0xffff) << MCX_SQE_WQE_INDEX_SHIFT));
7849 		/* always generate a completion event */
7850 		sqe->sqe_signature = htobe32(MCX_SQE_CE_CQE_ALWAYS);
7851 
7852 		/* eth segment */
7853 		csum = 0;
7854 		if (m->m_pkthdr.csum_flags & M_CSUM_IPv4)
7855 			csum |= MCX_SQE_L3_CSUM;
7856 		if (m->m_pkthdr.csum_flags &
7857 		    (M_CSUM_TCPv4 | M_CSUM_UDPv4 | M_CSUM_TCPv6 | M_CSUM_UDPv6))
7858 			csum |= MCX_SQE_L4_CSUM;
7859 		sqe->sqe_mss_csum = htobe32(csum);
7860 		sqe->sqe_inline_header_size = htobe16(MCX_SQ_INLINE_SIZE);
7861 		if (vlan_has_tag(m)) {
7862 			struct ether_vlan_header *evh;
7863 			evh = (struct ether_vlan_header *)
7864 			    &sqe->sqe_inline_headers;
7865 
7866 			m_copydata(m, 0, ETHER_HDR_LEN, evh);
7867 			evh->evl_proto = evh->evl_encap_proto;
7868 			evh->evl_encap_proto = htons(ETHERTYPE_VLAN);
7869 			evh->evl_tag = htons(vlan_get_tag(m));
7870 			m_adj(m, ETHER_HDR_LEN);
7871 		} else {
7872 			m_copydata(m, 0, MCX_SQ_INLINE_SIZE,
7873 			    sqe->sqe_inline_headers);
7874 			m_adj(m, MCX_SQ_INLINE_SIZE);
7875 		}
7876 
7877 		if (mcx_load_mbuf(sc, ms, m) != 0) {
7878 			m_freem(m);
7879 			if_statinc(ifp, if_oerrors);
7880 			continue;
7881 		}
7882 		bf = (uint64_t *)sqe;
7883 
7884 		if (ifp->if_bpf != NULL)
7885 			bpf_mtap2(ifp->if_bpf, sqe->sqe_inline_headers,
7886 			    MCX_SQ_INLINE_SIZE, m, BPF_D_OUT);
7887 
7888 		map = ms->ms_map;
7889 		bus_dmamap_sync(sc->sc_dmat, map, 0, map->dm_mapsize,
7890 		    BUS_DMASYNC_PREWRITE);
7891 
7892 		sqe->sqe_ds_sq_num =
7893 		    htobe32((tx->tx_sqn << MCX_SQE_SQ_NUM_SHIFT) |
7894 		    (map->dm_nsegs + 3));
7895 
7896 		/* data segment - first wqe has one segment */
7897 		sqs = sqe->sqe_segs;
7898 		seg = 0;
7899 		nseg = 1;
7900 		for (i = 0; i < map->dm_nsegs; i++) {
7901 			if (seg == nseg) {
7902 				/* next slot */
7903 				idx++;
7904 				if (idx == (1 << MCX_LOG_SQ_SIZE))
7905 					idx = 0;
7906 				tx->tx_prod++;
7907 				used++;
7908 
7909 				sqs = (struct mcx_sq_entry_seg *)(sq + idx);
7910 				seg = 0;
7911 				nseg = MCX_SQ_SEGS_PER_SLOT;
7912 			}
7913 			sqs[seg].sqs_byte_count =
7914 			    htobe32(map->dm_segs[i].ds_len);
7915 			sqs[seg].sqs_lkey = htobe32(sc->sc_lkey);
7916 			sqs[seg].sqs_addr = htobe64(map->dm_segs[i].ds_addr);
7917 			seg++;
7918 		}
7919 
7920 		idx++;
7921 		if (idx == (1 << MCX_LOG_SQ_SIZE))
7922 			idx = 0;
7923 		tx->tx_prod++;
7924 		used++;
7925 	}
7926 
7927 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&tx->tx_sq_mem),
7928 	    0, MCX_DMA_LEN(&tx->tx_sq_mem), BUS_DMASYNC_PREWRITE);
7929 
7930 	if (used) {
7931 		bus_size_t blueflame;
7932 
7933 		bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_doorbell_mem),
7934 		    tx->tx_doorbell, sizeof(uint32_t), BUS_DMASYNC_POSTWRITE);
7935 		be32enc(MCX_DMA_OFF(&sc->sc_doorbell_mem, tx->tx_doorbell),
7936 		    tx->tx_prod & MCX_WQ_DOORBELL_MASK);
7937 		bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_doorbell_mem),
7938 		    tx->tx_doorbell, sizeof(uint32_t), BUS_DMASYNC_PREWRITE);
7939 
7940 		/*
7941 		 * write the first 64 bits of the last sqe we produced
7942 		 * to the blue flame buffer
7943 		 */
7944 
7945 		blueflame = bf_base + tx->tx_bf_offset;
7946 		bus_space_write_8(sc->sc_memt, sc->sc_memh,
7947 		    blueflame, *bf);
7948 		mcx_bar(sc, blueflame, sizeof(*bf), BUS_SPACE_BARRIER_WRITE);
7949 
7950 		/* next write goes to the other buffer */
7951 		tx->tx_bf_offset ^= sc->sc_bf_size;
7952 	}
7953 }
7954 
7955 static void
7956 mcx_start(struct ifnet *ifp)
7957 {
7958 	struct mcx_softc *sc = ifp->if_softc;
7959 	/* mcx_start() always uses TX ring[0] */
7960 	struct mcx_tx *tx = &sc->sc_queues[0].q_tx;
7961 
7962 	mutex_enter(&tx->tx_lock);
7963 	if (!ISSET(ifp->if_flags, IFF_OACTIVE)) {
7964 		mcx_send_common_locked(ifp, tx, false);
7965 	}
7966 	mutex_exit(&tx->tx_lock);
7967 }
7968 
7969 static int
7970 mcx_transmit(struct ifnet *ifp, struct mbuf *m)
7971 {
7972 	struct mcx_softc *sc = ifp->if_softc;
7973 	struct mcx_tx *tx;
7974 
7975 	tx = &sc->sc_queues[cpu_index(curcpu()) % sc->sc_nqueues].q_tx;
7976 	if (__predict_false(!pcq_put(tx->tx_pcq, m))) {
7977 		m_freem(m);
7978 		return ENOBUFS;
7979 	}
7980 
7981 	if (mutex_tryenter(&tx->tx_lock)) {
7982 		mcx_send_common_locked(ifp, tx, true);
7983 		mutex_exit(&tx->tx_lock);
7984 	} else {
7985 		softint_schedule(tx->tx_softint);
7986 	}
7987 
7988 	return 0;
7989 }
7990 
7991 static void
7992 mcx_deferred_transmit(void *arg)
7993 {
7994 	struct mcx_tx *tx = arg;
7995 	struct mcx_softc *sc = tx->tx_softc;
7996 	struct ifnet *ifp = &sc->sc_ec.ec_if;
7997 
7998 	mutex_enter(&tx->tx_lock);
7999 	if (pcq_peek(tx->tx_pcq) != NULL) {
8000 		mcx_send_common_locked(ifp, tx, true);
8001 	}
8002 	mutex_exit(&tx->tx_lock);
8003 }
8004 
8005 static void
8006 mcx_watchdog(struct ifnet *ifp)
8007 {
8008 }
8009 
8010 static void
8011 mcx_media_add_types(struct mcx_softc *sc)
8012 {
8013 	struct mcx_reg_ptys ptys;
8014 	int i;
8015 	uint32_t proto_cap;
8016 
8017 	memset(&ptys, 0, sizeof(ptys));
8018 	ptys.rp_local_port = 1;
8019 	ptys.rp_proto_mask = MCX_REG_PTYS_PROTO_MASK_ETH;
8020 	if (mcx_access_hca_reg(sc, MCX_REG_PTYS, MCX_REG_OP_READ, &ptys,
8021 	    sizeof(ptys)) != 0) {
8022 		printf("%s: unable to read port type/speed\n", DEVNAME(sc));
8023 		return;
8024 	}
8025 
8026 	proto_cap = be32toh(ptys.rp_eth_proto_cap);
8027 	for (i = 0; i < __arraycount(mcx_eth_cap_map); i++) {
8028 		const struct mcx_eth_proto_capability *cap;
8029 		if (!ISSET(proto_cap, 1U << i))
8030 			continue;
8031 
8032 		cap = &mcx_eth_cap_map[i];
8033 		if (cap->cap_media == 0)
8034 			continue;
8035 
8036 		ifmedia_add(&sc->sc_media, IFM_ETHER | cap->cap_media, 0, NULL);
8037 	}
8038 }
8039 
8040 static void
8041 mcx_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
8042 {
8043 	struct mcx_softc *sc = (struct mcx_softc *)ifp->if_softc;
8044 	struct mcx_reg_ptys ptys;
8045 	int i;
8046 	uint32_t proto_oper;
8047 	uint64_t media_oper;
8048 
8049 	memset(&ptys, 0, sizeof(ptys));
8050 	ptys.rp_local_port = 1;
8051 	ptys.rp_proto_mask = MCX_REG_PTYS_PROTO_MASK_ETH;
8052 
8053 	if (mcx_access_hca_reg(sc, MCX_REG_PTYS, MCX_REG_OP_READ, &ptys,
8054 	    sizeof(ptys)) != 0) {
8055 		printf("%s: unable to read port type/speed\n", DEVNAME(sc));
8056 		return;
8057 	}
8058 
8059 	proto_oper = be32toh(ptys.rp_eth_proto_oper);
8060 
8061 	media_oper = 0;
8062 
8063 	for (i = 0; i < __arraycount(mcx_eth_cap_map); i++) {
8064 		const struct mcx_eth_proto_capability *cap;
8065 		if (!ISSET(proto_oper, 1U << i))
8066 			continue;
8067 
8068 		cap = &mcx_eth_cap_map[i];
8069 
8070 		if (cap->cap_media != 0)
8071 			media_oper = cap->cap_media;
8072 	}
8073 
8074 	ifmr->ifm_status = IFM_AVALID;
8075 	if (proto_oper != 0) {
8076 		ifmr->ifm_status |= IFM_ACTIVE;
8077 		ifmr->ifm_active = IFM_ETHER | IFM_AUTO | media_oper;
8078 		/* txpause, rxpause, duplex? */
8079 	}
8080 }
8081 
8082 static int
8083 mcx_media_change(struct ifnet *ifp)
8084 {
8085 	struct mcx_softc *sc = (struct mcx_softc *)ifp->if_softc;
8086 	struct mcx_reg_ptys ptys;
8087 	struct mcx_reg_paos paos;
8088 	uint32_t media;
8089 	int i, error;
8090 
8091 	if (IFM_TYPE(sc->sc_media.ifm_media) != IFM_ETHER)
8092 		return EINVAL;
8093 
8094 	error = 0;
8095 
8096 	if (IFM_SUBTYPE(sc->sc_media.ifm_media) == IFM_AUTO) {
8097 		/* read ptys to get supported media */
8098 		memset(&ptys, 0, sizeof(ptys));
8099 		ptys.rp_local_port = 1;
8100 		ptys.rp_proto_mask = MCX_REG_PTYS_PROTO_MASK_ETH;
8101 		if (mcx_access_hca_reg(sc, MCX_REG_PTYS, MCX_REG_OP_READ,
8102 		    &ptys, sizeof(ptys)) != 0) {
8103 			printf("%s: unable to read port type/speed\n",
8104 			    DEVNAME(sc));
8105 			return EIO;
8106 		}
8107 
8108 		media = be32toh(ptys.rp_eth_proto_cap);
8109 	} else {
8110 		/* map media type */
8111 		media = 0;
8112 		for (i = 0; i < __arraycount(mcx_eth_cap_map); i++) {
8113 			const struct mcx_eth_proto_capability *cap;
8114 
8115 			cap = &mcx_eth_cap_map[i];
8116 			if (cap->cap_media ==
8117 			    IFM_SUBTYPE(sc->sc_media.ifm_media)) {
8118 				media = (1 << i);
8119 				break;
8120 			}
8121 		}
8122 	}
8123 
8124 	/* disable the port */
8125 	memset(&paos, 0, sizeof(paos));
8126 	paos.rp_local_port = 1;
8127 	paos.rp_admin_status = MCX_REG_PAOS_ADMIN_STATUS_DOWN;
8128 	paos.rp_admin_state_update = MCX_REG_PAOS_ADMIN_STATE_UPDATE_EN;
8129 	if (mcx_access_hca_reg(sc, MCX_REG_PAOS, MCX_REG_OP_WRITE, &paos,
8130 	    sizeof(paos)) != 0) {
8131 		printf("%s: unable to set port state to down\n", DEVNAME(sc));
8132 		return EIO;
8133 	}
8134 
8135 	memset(&ptys, 0, sizeof(ptys));
8136 	ptys.rp_local_port = 1;
8137 	ptys.rp_proto_mask = MCX_REG_PTYS_PROTO_MASK_ETH;
8138 	ptys.rp_eth_proto_admin = htobe32(media);
8139 	if (mcx_access_hca_reg(sc, MCX_REG_PTYS, MCX_REG_OP_WRITE, &ptys,
8140 	    sizeof(ptys)) != 0) {
8141 		printf("%s: unable to set port media type/speed\n",
8142 		    DEVNAME(sc));
8143 		error = EIO;
8144 	}
8145 
8146 	/* re-enable the port to start negotiation */
8147 	memset(&paos, 0, sizeof(paos));
8148 	paos.rp_local_port = 1;
8149 	paos.rp_admin_status = MCX_REG_PAOS_ADMIN_STATUS_UP;
8150 	paos.rp_admin_state_update = MCX_REG_PAOS_ADMIN_STATE_UPDATE_EN;
8151 	if (mcx_access_hca_reg(sc, MCX_REG_PAOS, MCX_REG_OP_WRITE, &paos,
8152 	    sizeof(paos)) != 0) {
8153 		printf("%s: unable to set port state to up\n", DEVNAME(sc));
8154 		error = EIO;
8155 	}
8156 
8157 	return error;
8158 }
8159 
8160 static void
8161 mcx_port_change(struct work *wk, void *xsc)
8162 {
8163 	struct mcx_softc *sc = xsc;
8164 	struct ifnet *ifp = &sc->sc_ec.ec_if;
8165 	struct mcx_reg_ptys ptys = {
8166 		.rp_local_port = 1,
8167 		.rp_proto_mask = MCX_REG_PTYS_PROTO_MASK_ETH,
8168 	};
8169 	int link_state = LINK_STATE_DOWN;
8170 
8171 	if (mcx_access_hca_reg(sc, MCX_REG_PTYS, MCX_REG_OP_READ, &ptys,
8172 	    sizeof(ptys)) == 0) {
8173 		uint32_t proto_oper = be32toh(ptys.rp_eth_proto_oper);
8174 		uint64_t baudrate = 0;
8175 		unsigned int i;
8176 
8177 		if (proto_oper != 0)
8178 			link_state = LINK_STATE_UP;
8179 
8180 		for (i = 0; i < __arraycount(mcx_eth_cap_map); i++) {
8181 			const struct mcx_eth_proto_capability *cap;
8182 			if (!ISSET(proto_oper, 1U << i))
8183 				continue;
8184 
8185 			cap = &mcx_eth_cap_map[i];
8186 			if (cap->cap_baudrate == 0)
8187 				continue;
8188 
8189 			baudrate = cap->cap_baudrate;
8190 			break;
8191 		}
8192 
8193 		ifp->if_baudrate = baudrate;
8194 	}
8195 
8196 	if (link_state != ifp->if_link_state) {
8197 		if_link_state_change(ifp, link_state);
8198 	}
8199 }
8200 
8201 
8202 static inline uint32_t
8203 mcx_rd(struct mcx_softc *sc, bus_size_t r)
8204 {
8205 	uint32_t word;
8206 
8207 	word = bus_space_read_4(sc->sc_memt, sc->sc_memh, r);
8208 
8209 	return (be32toh(word));
8210 }
8211 
8212 static inline void
8213 mcx_wr(struct mcx_softc *sc, bus_size_t r, uint32_t v)
8214 {
8215 	bus_space_write_4(sc->sc_memt, sc->sc_memh, r, htobe32(v));
8216 }
8217 
8218 static inline void
8219 mcx_bar(struct mcx_softc *sc, bus_size_t r, bus_size_t l, int f)
8220 {
8221 #ifndef __NetBSD__
8222 	bus_space_barrier(sc->sc_memt, sc->sc_memh, r, l, f);
8223 #endif
8224 }
8225 
8226 static uint64_t
8227 mcx_timer(struct mcx_softc *sc)
8228 {
8229 	uint32_t hi, lo, ni;
8230 
8231 	hi = mcx_rd(sc, MCX_INTERNAL_TIMER_H);
8232 	for (;;) {
8233 		lo = mcx_rd(sc, MCX_INTERNAL_TIMER_L);
8234 		mcx_bar(sc, MCX_INTERNAL_TIMER_L, 8, BUS_SPACE_BARRIER_READ);
8235 		ni = mcx_rd(sc, MCX_INTERNAL_TIMER_H);
8236 
8237 		if (ni == hi)
8238 			break;
8239 
8240 		hi = ni;
8241 	}
8242 
8243 	return (((uint64_t)hi << 32) | (uint64_t)lo);
8244 }
8245 
8246 static int
8247 mcx_dmamem_alloc(struct mcx_softc *sc, struct mcx_dmamem *mxm,
8248     bus_size_t size, u_int align)
8249 {
8250 	mxm->mxm_size = size;
8251 
8252 	if (bus_dmamap_create(sc->sc_dmat, mxm->mxm_size, 1,
8253 	    mxm->mxm_size, 0,
8254 	    BUS_DMA_WAITOK | BUS_DMA_ALLOCNOW,
8255 	    &mxm->mxm_map) != 0)
8256 		return (1);
8257 	if (bus_dmamem_alloc(sc->sc_dmat, mxm->mxm_size,
8258 	    align, 0, &mxm->mxm_seg, 1, &mxm->mxm_nsegs,
8259 	    BUS_DMA_WAITOK) != 0)
8260 		goto destroy;
8261 	if (bus_dmamem_map(sc->sc_dmat, &mxm->mxm_seg, mxm->mxm_nsegs,
8262 	    mxm->mxm_size, &mxm->mxm_kva,
8263 	    BUS_DMA_WAITOK | BUS_DMA_COHERENT) != 0)
8264 		goto free;
8265 	if (bus_dmamap_load(sc->sc_dmat, mxm->mxm_map, mxm->mxm_kva,
8266 	    mxm->mxm_size, NULL, BUS_DMA_WAITOK) != 0)
8267 		goto unmap;
8268 
8269 	mcx_dmamem_zero(mxm);
8270 
8271 	return (0);
8272 unmap:
8273 	bus_dmamem_unmap(sc->sc_dmat, mxm->mxm_kva, mxm->mxm_size);
8274 free:
8275 	bus_dmamem_free(sc->sc_dmat, &mxm->mxm_seg, 1);
8276 destroy:
8277 	bus_dmamap_destroy(sc->sc_dmat, mxm->mxm_map);
8278 	return (1);
8279 }
8280 
8281 static void
8282 mcx_dmamem_zero(struct mcx_dmamem *mxm)
8283 {
8284 	memset(MCX_DMA_KVA(mxm), 0, MCX_DMA_LEN(mxm));
8285 }
8286 
8287 static void
8288 mcx_dmamem_free(struct mcx_softc *sc, struct mcx_dmamem *mxm)
8289 {
8290 	bus_dmamap_unload(sc->sc_dmat, mxm->mxm_map);
8291 	bus_dmamem_unmap(sc->sc_dmat, mxm->mxm_kva, mxm->mxm_size);
8292 	bus_dmamem_free(sc->sc_dmat, &mxm->mxm_seg, 1);
8293 	bus_dmamap_destroy(sc->sc_dmat, mxm->mxm_map);
8294 }
8295 
8296 static int
8297 mcx_hwmem_alloc(struct mcx_softc *sc, struct mcx_hwmem *mhm, unsigned int pages)
8298 {
8299 	bus_dma_segment_t *segs;
8300 	bus_size_t len = pages * MCX_PAGE_SIZE;
8301 	size_t seglen;
8302 
8303 	segs = kmem_alloc(sizeof(*segs) * pages, KM_SLEEP);
8304 	seglen = sizeof(*segs) * pages;
8305 
8306 	if (bus_dmamem_alloc(sc->sc_dmat, len, MCX_PAGE_SIZE, 0,
8307 	    segs, pages, &mhm->mhm_seg_count, BUS_DMA_NOWAIT) != 0)
8308 		goto free_segs;
8309 
8310 	if (mhm->mhm_seg_count < pages) {
8311 		size_t nseglen;
8312 
8313 		mhm->mhm_segs = kmem_alloc(
8314 		    sizeof(*mhm->mhm_segs) * mhm->mhm_seg_count, KM_SLEEP);
8315 
8316 		nseglen = sizeof(*mhm->mhm_segs) * mhm->mhm_seg_count;
8317 
8318 		memcpy(mhm->mhm_segs, segs, nseglen);
8319 
8320 		kmem_free(segs, seglen);
8321 
8322 		segs = mhm->mhm_segs;
8323 		seglen = nseglen;
8324 	} else
8325 		mhm->mhm_segs = segs;
8326 
8327 	if (bus_dmamap_create(sc->sc_dmat, len, pages, MCX_PAGE_SIZE,
8328 	    MCX_PAGE_SIZE, BUS_DMA_NOWAIT|BUS_DMA_ALLOCNOW /*|BUS_DMA_64BIT*/,
8329 	    &mhm->mhm_map) != 0)
8330 		goto free_dmamem;
8331 
8332 	if (bus_dmamap_load_raw(sc->sc_dmat, mhm->mhm_map,
8333 	    mhm->mhm_segs, mhm->mhm_seg_count, len, BUS_DMA_NOWAIT) != 0)
8334 		goto destroy;
8335 
8336 	bus_dmamap_sync(sc->sc_dmat, mhm->mhm_map,
8337 	    0, mhm->mhm_map->dm_mapsize, BUS_DMASYNC_PRERW);
8338 
8339 	mhm->mhm_npages = pages;
8340 
8341 	return (0);
8342 
8343 destroy:
8344 	bus_dmamap_destroy(sc->sc_dmat, mhm->mhm_map);
8345 free_dmamem:
8346 	bus_dmamem_free(sc->sc_dmat, mhm->mhm_segs, mhm->mhm_seg_count);
8347 free_segs:
8348 	kmem_free(segs, seglen);
8349 	mhm->mhm_segs = NULL;
8350 
8351 	return (-1);
8352 }
8353 
8354 static void
8355 mcx_hwmem_free(struct mcx_softc *sc, struct mcx_hwmem *mhm)
8356 {
8357 	if (mhm->mhm_npages == 0)
8358 		return;
8359 
8360 	bus_dmamap_sync(sc->sc_dmat, mhm->mhm_map,
8361 	    0, mhm->mhm_map->dm_mapsize, BUS_DMASYNC_POSTRW);
8362 
8363 	bus_dmamap_unload(sc->sc_dmat, mhm->mhm_map);
8364 	bus_dmamap_destroy(sc->sc_dmat, mhm->mhm_map);
8365 	bus_dmamem_free(sc->sc_dmat, mhm->mhm_segs, mhm->mhm_seg_count);
8366 	kmem_free(mhm->mhm_segs, sizeof(*mhm->mhm_segs) * mhm->mhm_seg_count);
8367 
8368 	mhm->mhm_npages = 0;
8369 }
8370 
8371 #if NKSTAT > 0
8372 struct mcx_ppcnt {
8373 	char			 name[KSTAT_KV_NAMELEN];
8374 	enum kstat_kv_unit	 unit;
8375 };
8376 
8377 static const struct mcx_ppcnt mcx_ppcnt_ieee8023_tpl[] = {
8378 	{ "Good Tx",		KSTAT_KV_U_PACKETS, },
8379 	{ "Good Rx",		KSTAT_KV_U_PACKETS, },
8380 	{ "FCS errs",		KSTAT_KV_U_PACKETS, },
8381 	{ "Alignment Errs",	KSTAT_KV_U_PACKETS, },
8382 	{ "Good Tx",		KSTAT_KV_U_BYTES, },
8383 	{ "Good Rx",		KSTAT_KV_U_BYTES, },
8384 	{ "Multicast Tx",	KSTAT_KV_U_PACKETS, },
8385 	{ "Broadcast Tx",	KSTAT_KV_U_PACKETS, },
8386 	{ "Multicast Rx",	KSTAT_KV_U_PACKETS, },
8387 	{ "Broadcast Rx",	KSTAT_KV_U_PACKETS, },
8388 	{ "In Range Len",	KSTAT_KV_U_PACKETS, },
8389 	{ "Out Of Range Len",	KSTAT_KV_U_PACKETS, },
8390 	{ "Frame Too Long",	KSTAT_KV_U_PACKETS, },
8391 	{ "Symbol Errs",	KSTAT_KV_U_PACKETS, },
8392 	{ "MAC Ctrl Tx",	KSTAT_KV_U_PACKETS, },
8393 	{ "MAC Ctrl Rx",	KSTAT_KV_U_PACKETS, },
8394 	{ "MAC Ctrl Unsup",	KSTAT_KV_U_PACKETS, },
8395 	{ "Pause Rx",		KSTAT_KV_U_PACKETS, },
8396 	{ "Pause Tx",		KSTAT_KV_U_PACKETS, },
8397 };
8398 CTASSERT(__arraycount(mcx_ppcnt_ieee8023_tpl) == mcx_ppcnt_ieee8023_count);
8399 
8400 static const struct mcx_ppcnt mcx_ppcnt_rfc2863_tpl[] = {
8401 	{ "Rx Bytes",		KSTAT_KV_U_BYTES, },
8402 	{ "Rx Unicast",		KSTAT_KV_U_PACKETS, },
8403 	{ "Rx Discards",	KSTAT_KV_U_PACKETS, },
8404 	{ "Rx Errors",		KSTAT_KV_U_PACKETS, },
8405 	{ "Rx Unknown Proto",	KSTAT_KV_U_PACKETS, },
8406 	{ "Tx Bytes",		KSTAT_KV_U_BYTES, },
8407 	{ "Tx Unicast",		KSTAT_KV_U_PACKETS, },
8408 	{ "Tx Discards",	KSTAT_KV_U_PACKETS, },
8409 	{ "Tx Errors",		KSTAT_KV_U_PACKETS, },
8410 	{ "Rx Multicast",	KSTAT_KV_U_PACKETS, },
8411 	{ "Rx Broadcast",	KSTAT_KV_U_PACKETS, },
8412 	{ "Tx Multicast",	KSTAT_KV_U_PACKETS, },
8413 	{ "Tx Broadcast",	KSTAT_KV_U_PACKETS, },
8414 };
8415 CTASSERT(__arraycount(mcx_ppcnt_rfc2863_tpl) == mcx_ppcnt_rfc2863_count);
8416 
8417 static const struct mcx_ppcnt mcx_ppcnt_rfc2819_tpl[] = {
8418 	{ "Drop Events",	KSTAT_KV_U_PACKETS, },
8419 	{ "Octets",		KSTAT_KV_U_BYTES, },
8420 	{ "Packets",		KSTAT_KV_U_PACKETS, },
8421 	{ "Broadcasts",		KSTAT_KV_U_PACKETS, },
8422 	{ "Multicasts",		KSTAT_KV_U_PACKETS, },
8423 	{ "CRC Align Errs",	KSTAT_KV_U_PACKETS, },
8424 	{ "Undersize",		KSTAT_KV_U_PACKETS, },
8425 	{ "Oversize",		KSTAT_KV_U_PACKETS, },
8426 	{ "Fragments",		KSTAT_KV_U_PACKETS, },
8427 	{ "Jabbers",		KSTAT_KV_U_PACKETS, },
8428 	{ "Collisions",		KSTAT_KV_U_NONE, },
8429 	{ "64B",		KSTAT_KV_U_PACKETS, },
8430 	{ "65-127B",		KSTAT_KV_U_PACKETS, },
8431 	{ "128-255B",		KSTAT_KV_U_PACKETS, },
8432 	{ "256-511B",		KSTAT_KV_U_PACKETS, },
8433 	{ "512-1023B",		KSTAT_KV_U_PACKETS, },
8434 	{ "1024-1518B",		KSTAT_KV_U_PACKETS, },
8435 	{ "1519-2047B",		KSTAT_KV_U_PACKETS, },
8436 	{ "2048-4095B",		KSTAT_KV_U_PACKETS, },
8437 	{ "4096-8191B",		KSTAT_KV_U_PACKETS, },
8438 	{ "8192-10239B",	KSTAT_KV_U_PACKETS, },
8439 };
8440 CTASSERT(__arraycount(mcx_ppcnt_rfc2819_tpl) == mcx_ppcnt_rfc2819_count);
8441 
8442 static const struct mcx_ppcnt mcx_ppcnt_rfc3635_tpl[] = {
8443 	{ "Alignment Errs",	KSTAT_KV_U_PACKETS, },
8444 	{ "FCS Errs",		KSTAT_KV_U_PACKETS, },
8445 	{ "Single Colls",	KSTAT_KV_U_PACKETS, },
8446 	{ "Multiple Colls",	KSTAT_KV_U_PACKETS, },
8447 	{ "SQE Test Errs",	KSTAT_KV_U_NONE, },
8448 	{ "Deferred Tx",	KSTAT_KV_U_PACKETS, },
8449 	{ "Late Colls",		KSTAT_KV_U_NONE, },
8450 	{ "Exess Colls",	KSTAT_KV_U_NONE, },
8451 	{ "Int MAC Tx Errs",	KSTAT_KV_U_PACKETS, },
8452 	{ "CSM Sense Errs",	KSTAT_KV_U_NONE, },
8453 	{ "Too Long",		KSTAT_KV_U_PACKETS, },
8454 	{ "Int MAC Rx Errs",	KSTAT_KV_U_PACKETS, },
8455 	{ "Symbol Errs",	KSTAT_KV_U_NONE, },
8456 	{ "Unknown Control",	KSTAT_KV_U_PACKETS, },
8457 	{ "Pause Rx",		KSTAT_KV_U_PACKETS, },
8458 	{ "Pause Tx",		KSTAT_KV_U_PACKETS, },
8459 };
8460 CTASSERT(__arraycount(mcx_ppcnt_rfc3635_tpl) == mcx_ppcnt_rfc3635_count);
8461 
8462 struct mcx_kstat_ppcnt {
8463 	const char		*ksp_name;
8464 	const struct mcx_ppcnt	*ksp_tpl;
8465 	unsigned int		 ksp_n;
8466 	uint8_t			 ksp_grp;
8467 };
8468 
8469 static const struct mcx_kstat_ppcnt mcx_kstat_ppcnt_ieee8023 = {
8470 	.ksp_name =		"ieee802.3",
8471 	.ksp_tpl =		mcx_ppcnt_ieee8023_tpl,
8472 	.ksp_n =		__arraycount(mcx_ppcnt_ieee8023_tpl),
8473 	.ksp_grp =		MCX_REG_PPCNT_GRP_IEEE8023,
8474 };
8475 
8476 static const struct mcx_kstat_ppcnt mcx_kstat_ppcnt_rfc2863 = {
8477 	.ksp_name =		"rfc2863",
8478 	.ksp_tpl =		mcx_ppcnt_rfc2863_tpl,
8479 	.ksp_n =		__arraycount(mcx_ppcnt_rfc2863_tpl),
8480 	.ksp_grp =		MCX_REG_PPCNT_GRP_RFC2863,
8481 };
8482 
8483 static const struct mcx_kstat_ppcnt mcx_kstat_ppcnt_rfc2819 = {
8484 	.ksp_name =		"rfc2819",
8485 	.ksp_tpl =		mcx_ppcnt_rfc2819_tpl,
8486 	.ksp_n =		__arraycount(mcx_ppcnt_rfc2819_tpl),
8487 	.ksp_grp =		MCX_REG_PPCNT_GRP_RFC2819,
8488 };
8489 
8490 static const struct mcx_kstat_ppcnt mcx_kstat_ppcnt_rfc3635 = {
8491 	.ksp_name =		"rfc3635",
8492 	.ksp_tpl =		mcx_ppcnt_rfc3635_tpl,
8493 	.ksp_n =		__arraycount(mcx_ppcnt_rfc3635_tpl),
8494 	.ksp_grp =		MCX_REG_PPCNT_GRP_RFC3635,
8495 };
8496 
8497 static int	mcx_kstat_ppcnt_read(struct kstat *);
8498 
8499 static void	mcx_kstat_attach_tmps(struct mcx_softc *sc);
8500 static void	mcx_kstat_attach_queues(struct mcx_softc *sc);
8501 
8502 static struct kstat *
8503 mcx_kstat_attach_ppcnt(struct mcx_softc *sc,
8504     const struct mcx_kstat_ppcnt *ksp)
8505 {
8506 	struct kstat *ks;
8507 	struct kstat_kv *kvs;
8508 	unsigned int i;
8509 
8510 	ks = kstat_create(DEVNAME(sc), 0, ksp->ksp_name, 0, KSTAT_T_KV, 0);
8511 	if (ks == NULL)
8512 		return (NULL);
8513 
8514 	kvs = mallocarray(ksp->ksp_n, sizeof(*kvs),
8515 	    M_DEVBUF, M_WAITOK);
8516 
8517 	for (i = 0; i < ksp->ksp_n; i++) {
8518 		const struct mcx_ppcnt *tpl = &ksp->ksp_tpl[i];
8519 
8520 		kstat_kv_unit_init(&kvs[i], tpl->name,
8521 		    KSTAT_KV_T_COUNTER64, tpl->unit);
8522 	}
8523 
8524 	ks->ks_softc = sc;
8525 	ks->ks_ptr = (void *)ksp;
8526 	ks->ks_data = kvs;
8527 	ks->ks_datalen = ksp->ksp_n * sizeof(*kvs);
8528 	ks->ks_read = mcx_kstat_ppcnt_read;
8529 
8530 	kstat_install(ks);
8531 
8532 	return (ks);
8533 }
8534 
8535 static void
8536 mcx_kstat_attach(struct mcx_softc *sc)
8537 {
8538 	sc->sc_kstat_ieee8023 = mcx_kstat_attach_ppcnt(sc,
8539 	    &mcx_kstat_ppcnt_ieee8023);
8540 	sc->sc_kstat_rfc2863 = mcx_kstat_attach_ppcnt(sc,
8541 	    &mcx_kstat_ppcnt_rfc2863);
8542 	sc->sc_kstat_rfc2819 = mcx_kstat_attach_ppcnt(sc,
8543 	    &mcx_kstat_ppcnt_rfc2819);
8544 	sc->sc_kstat_rfc3635 = mcx_kstat_attach_ppcnt(sc,
8545 	    &mcx_kstat_ppcnt_rfc3635);
8546 
8547 	mcx_kstat_attach_tmps(sc);
8548 	mcx_kstat_attach_queues(sc);
8549 }
8550 
8551 static int
8552 mcx_kstat_ppcnt_read(struct kstat *ks)
8553 {
8554 	struct mcx_softc *sc = ks->ks_softc;
8555 	struct mcx_kstat_ppcnt *ksp = ks->ks_ptr;
8556 	struct mcx_reg_ppcnt ppcnt = {
8557 		.ppcnt_grp = ksp->ksp_grp,
8558 		.ppcnt_local_port = 1,
8559 	};
8560 	struct kstat_kv *kvs = ks->ks_data;
8561 	uint64_t *vs = (uint64_t *)&ppcnt.ppcnt_counter_set;
8562 	unsigned int i;
8563 	int rv;
8564 
8565 	KERNEL_LOCK(); /* XXX */
8566 	rv = mcx_access_hca_reg(sc, MCX_REG_PPCNT, MCX_REG_OP_READ,
8567 	    &ppcnt, sizeof(ppcnt));
8568 	KERNEL_UNLOCK();
8569 	if (rv != 0)
8570 		return (EIO);
8571 
8572 	nanouptime(&ks->ks_updated);
8573 
8574 	for (i = 0; i < ksp->ksp_n; i++)
8575 		kstat_kv_u64(&kvs[i]) = bemtoh64(&vs[i]);
8576 
8577 	return (0);
8578 }
8579 
8580 struct mcx_kstat_mtmp {
8581 	struct kstat_kv		ktmp_name;
8582 	struct kstat_kv		ktmp_temperature;
8583 	struct kstat_kv		ktmp_threshold_lo;
8584 	struct kstat_kv		ktmp_threshold_hi;
8585 };
8586 
8587 static const struct mcx_kstat_mtmp mcx_kstat_mtmp_tpl = {
8588 	KSTAT_KV_INITIALIZER("name",		KSTAT_KV_T_ISTR),
8589 	KSTAT_KV_INITIALIZER("temperature",	KSTAT_KV_T_TEMP),
8590 	KSTAT_KV_INITIALIZER("lo threshold",	KSTAT_KV_T_TEMP),
8591 	KSTAT_KV_INITIALIZER("hi threshold",	KSTAT_KV_T_TEMP),
8592 };
8593 
8594 static const struct timeval mcx_kstat_mtmp_rate = { 1, 0 };
8595 
8596 static int mcx_kstat_mtmp_read(struct kstat *);
8597 
8598 static void
8599 mcx_kstat_attach_tmps(struct mcx_softc *sc)
8600 {
8601 	struct kstat *ks;
8602 	struct mcx_reg_mcam mcam;
8603 	struct mcx_reg_mtcap mtcap;
8604 	struct mcx_kstat_mtmp *ktmp;
8605 	uint64_t map;
8606 	unsigned int i, n;
8607 
8608 	memset(&mtcap, 0, sizeof(mtcap));
8609 	memset(&mcam, 0, sizeof(mcam));
8610 
8611 	if (sc->sc_mcam_reg == 0) {
8612 		/* no management capabilities */
8613 		return;
8614 	}
8615 
8616 	if (mcx_access_hca_reg(sc, MCX_REG_MCAM, MCX_REG_OP_READ,
8617 	    &mcam, sizeof(mcam)) != 0) {
8618 		/* unable to check management capabilities? */
8619 		return;
8620 	}
8621 
8622 	if (MCX_BITFIELD_BIT(mcam.mcam_feature_cap_mask,
8623 	    MCX_MCAM_FEATURE_CAP_SENSOR_MAP) == 0) {
8624 		/* no sensor map */
8625 		return;
8626 	}
8627 
8628 	if (mcx_access_hca_reg(sc, MCX_REG_MTCAP, MCX_REG_OP_READ,
8629 	    &mtcap, sizeof(mtcap)) != 0) {
8630 		/* unable to find temperature sensors */
8631 		return;
8632 	}
8633 
8634 	sc->sc_kstat_mtmp_count = mtcap.mtcap_sensor_count;
8635 	sc->sc_kstat_mtmp = mallocarray(sc->sc_kstat_mtmp_count,
8636 	    sizeof(*sc->sc_kstat_mtmp), M_DEVBUF, M_WAITOK);
8637 
8638 	n = 0;
8639 	map = bemtoh64(&mtcap.mtcap_sensor_map);
8640 	for (i = 0; i < sizeof(map) * NBBY; i++) {
8641 		if (!ISSET(map, (1ULL << i)))
8642 			continue;
8643 
8644 		ks = kstat_create(DEVNAME(sc), 0, "temperature", i,
8645 		    KSTAT_T_KV, 0);
8646 		if (ks == NULL) {
8647 			/* unable to attach temperature sensor %u, i */
8648 			continue;
8649 		}
8650 
8651 		ktmp = malloc(sizeof(*ktmp), M_DEVBUF, M_WAITOK|M_ZERO);
8652 		*ktmp = mcx_kstat_mtmp_tpl;
8653 
8654 		ks->ks_data = ktmp;
8655 		ks->ks_datalen = sizeof(*ktmp);
8656 		TIMEVAL_TO_TIMESPEC(&mcx_kstat_mtmp_rate, &ks->ks_interval);
8657 		ks->ks_read = mcx_kstat_mtmp_read;
8658 
8659 		ks->ks_softc = sc;
8660 		kstat_install(ks);
8661 
8662 		sc->sc_kstat_mtmp[n++] = ks;
8663 		if (n >= sc->sc_kstat_mtmp_count)
8664 			break;
8665 	}
8666 }
8667 
8668 static uint64_t
8669 mcx_tmp_to_uK(uint16_t *t)
8670 {
8671 	int64_t mt = (int16_t)bemtoh16(t); /* 0.125 C units */
8672 	mt *= 1000000 / 8; /* convert to uC */
8673 	mt += 273150000; /* convert to uK */
8674 
8675 	return (mt);
8676 }
8677 
8678 static int
8679 mcx_kstat_mtmp_read(struct kstat *ks)
8680 {
8681 	struct mcx_softc *sc = ks->ks_softc;
8682 	struct mcx_kstat_mtmp *ktmp = ks->ks_data;
8683 	struct mcx_reg_mtmp mtmp;
8684 	int rv;
8685 	struct timeval updated;
8686 
8687 	TIMESPEC_TO_TIMEVAL(&updated, &ks->ks_updated);
8688 
8689 	if (!ratecheck(&updated, &mcx_kstat_mtmp_rate))
8690 		return (0);
8691 
8692 	memset(&mtmp, 0, sizeof(mtmp));
8693 	htobem16(&mtmp.mtmp_sensor_index, ks->ks_unit);
8694 
8695 	KERNEL_LOCK(); /* XXX */
8696 	rv = mcx_access_hca_reg(sc, MCX_REG_MTMP, MCX_REG_OP_READ,
8697 	    &mtmp, sizeof(mtmp));
8698 	KERNEL_UNLOCK();
8699 	if (rv != 0)
8700 		return (EIO);
8701 
8702 	memset(kstat_kv_istr(&ktmp->ktmp_name), 0,
8703 	    sizeof(kstat_kv_istr(&ktmp->ktmp_name)));
8704 	memcpy(kstat_kv_istr(&ktmp->ktmp_name),
8705 	    mtmp.mtmp_sensor_name, sizeof(mtmp.mtmp_sensor_name));
8706 	kstat_kv_temp(&ktmp->ktmp_temperature) =
8707 	    mcx_tmp_to_uK(&mtmp.mtmp_temperature);
8708 	kstat_kv_temp(&ktmp->ktmp_threshold_lo) =
8709 	    mcx_tmp_to_uK(&mtmp.mtmp_temperature_threshold_lo);
8710 	kstat_kv_temp(&ktmp->ktmp_threshold_hi) =
8711 	    mcx_tmp_to_uK(&mtmp.mtmp_temperature_threshold_hi);
8712 
8713 	TIMEVAL_TO_TIMESPEC(&updated, &ks->ks_updated);
8714 
8715 	return (0);
8716 }
8717 
8718 struct mcx_queuestat {
8719 	char			 name[KSTAT_KV_NAMELEN];
8720 	enum kstat_kv_type	 type;
8721 };
8722 
8723 static const struct mcx_queuestat mcx_queue_kstat_tpl[] = {
8724 	{ "RQ SW prod",		KSTAT_KV_T_COUNTER64 },
8725 	{ "RQ HW prod",		KSTAT_KV_T_COUNTER64 },
8726 	{ "RQ HW cons",		KSTAT_KV_T_COUNTER64 },
8727 	{ "RQ HW state",	KSTAT_KV_T_ISTR },
8728 
8729 	{ "SQ SW prod",		KSTAT_KV_T_COUNTER64 },
8730 	{ "SQ SW cons",		KSTAT_KV_T_COUNTER64 },
8731 	{ "SQ HW prod",		KSTAT_KV_T_COUNTER64 },
8732 	{ "SQ HW cons",		KSTAT_KV_T_COUNTER64 },
8733 	{ "SQ HW state",	KSTAT_KV_T_ISTR },
8734 
8735 	{ "CQ SW cons",		KSTAT_KV_T_COUNTER64 },
8736 	{ "CQ HW prod",		KSTAT_KV_T_COUNTER64 },
8737 	{ "CQ HW cons",		KSTAT_KV_T_COUNTER64 },
8738 	{ "CQ HW notify",	KSTAT_KV_T_COUNTER64 },
8739 	{ "CQ HW solicit",	KSTAT_KV_T_COUNTER64 },
8740 	{ "CQ HW status",	KSTAT_KV_T_ISTR },
8741 	{ "CQ HW state",	KSTAT_KV_T_ISTR },
8742 
8743 	{ "EQ SW cons",		KSTAT_KV_T_COUNTER64 },
8744 	{ "EQ HW prod",		KSTAT_KV_T_COUNTER64 },
8745 	{ "EQ HW cons",		KSTAT_KV_T_COUNTER64 },
8746 	{ "EQ HW status",	KSTAT_KV_T_ISTR },
8747 	{ "EQ HW state",	KSTAT_KV_T_ISTR },
8748 };
8749 
8750 static int	mcx_kstat_queue_read(struct kstat *);
8751 
8752 static void
8753 mcx_kstat_attach_queues(struct mcx_softc *sc)
8754 {
8755 	struct kstat *ks;
8756 	struct kstat_kv *kvs;
8757 	int q, i;
8758 
8759 	for (q = 0; q < sc->sc_nqueues; q++) {
8760 		ks = kstat_create(DEVNAME(sc), 0, "mcx-queues", q,
8761 		    KSTAT_T_KV, 0);
8762 		if (ks == NULL) {
8763 			/* unable to attach queue stats %u, q */
8764 			continue;
8765 		}
8766 
8767 		kvs = mallocarray(nitems(mcx_queue_kstat_tpl),
8768 		    sizeof(*kvs), M_DEVBUF, M_WAITOK);
8769 
8770 		for (i = 0; i < nitems(mcx_queue_kstat_tpl); i++) {
8771 			const struct mcx_queuestat *tpl =
8772 			    &mcx_queue_kstat_tpl[i];
8773 
8774 			kstat_kv_init(&kvs[i], tpl->name, tpl->type);
8775 		}
8776 
8777 		ks->ks_softc = &sc->sc_queues[q];
8778 		ks->ks_data = kvs;
8779 		ks->ks_datalen = nitems(mcx_queue_kstat_tpl) * sizeof(*kvs);
8780 		ks->ks_read = mcx_kstat_queue_read;
8781 
8782 		sc->sc_queues[q].q_kstat = ks;
8783 		kstat_install(ks);
8784 	}
8785 }
8786 
8787 static int
8788 mcx_kstat_queue_read(struct kstat *ks)
8789 {
8790 	struct mcx_queues *q = ks->ks_softc;
8791 	struct mcx_softc *sc = q->q_sc;
8792 	struct kstat_kv *kvs = ks->ks_data;
8793 	union {
8794 		struct mcx_rq_ctx rq;
8795 		struct mcx_sq_ctx sq;
8796 		struct mcx_cq_ctx cq;
8797 		struct mcx_eq_ctx eq;
8798 	} u;
8799 	const char *text;
8800 	int error = 0;
8801 
8802 	KERNEL_LOCK();
8803 
8804 	if (mcx_query_rq(sc, &q->q_rx, &u.rq) != 0) {
8805 		error = EIO;
8806 		goto out;
8807 	}
8808 
8809 	kstat_kv_u64(kvs++) = q->q_rx.rx_prod;
8810 	kstat_kv_u64(kvs++) = bemtoh32(&u.rq.rq_wq.wq_sw_counter);
8811 	kstat_kv_u64(kvs++) = bemtoh32(&u.rq.rq_wq.wq_hw_counter);
8812 	switch ((bemtoh32(&u.rq.rq_flags) & MCX_RQ_CTX_STATE_MASK) >>
8813 	    MCX_RQ_CTX_STATE_SHIFT) {
8814 	case MCX_RQ_CTX_STATE_RST:
8815 		text = "RST";
8816 		break;
8817 	case MCX_RQ_CTX_STATE_RDY:
8818 		text = "RDY";
8819 		break;
8820 	case MCX_RQ_CTX_STATE_ERR:
8821 		text = "ERR";
8822 		break;
8823 	default:
8824 		text = "unknown";
8825 		break;
8826 	}
8827 	strlcpy(kstat_kv_istr(kvs), text, sizeof(kstat_kv_istr(kvs)));
8828 	kvs++;
8829 
8830 	if (mcx_query_sq(sc, &q->q_tx, &u.sq) != 0) {
8831 		error = EIO;
8832 		goto out;
8833 	}
8834 
8835 	kstat_kv_u64(kvs++) = q->q_tx.tx_prod;
8836 	kstat_kv_u64(kvs++) = q->q_tx.tx_cons;
8837 	kstat_kv_u64(kvs++) = bemtoh32(&u.sq.sq_wq.wq_sw_counter);
8838 	kstat_kv_u64(kvs++) = bemtoh32(&u.sq.sq_wq.wq_hw_counter);
8839 	switch ((bemtoh32(&u.sq.sq_flags) & MCX_SQ_CTX_STATE_MASK) >>
8840 	    MCX_SQ_CTX_STATE_SHIFT) {
8841 	case MCX_SQ_CTX_STATE_RST:
8842 		text = "RST";
8843 		break;
8844 	case MCX_SQ_CTX_STATE_RDY:
8845 		text = "RDY";
8846 		break;
8847 	case MCX_SQ_CTX_STATE_ERR:
8848 		text = "ERR";
8849 		break;
8850 	default:
8851 		text = "unknown";
8852 		break;
8853 	}
8854 	strlcpy(kstat_kv_istr(kvs), text, sizeof(kstat_kv_istr(kvs)));
8855 	kvs++;
8856 
8857 	if (mcx_query_cq(sc, &q->q_cq, &u.cq) != 0) {
8858 		error = EIO;
8859 		goto out;
8860 	}
8861 
8862 	kstat_kv_u64(kvs++) = q->q_cq.cq_cons;
8863 	kstat_kv_u64(kvs++) = bemtoh32(&u.cq.cq_producer_counter);
8864 	kstat_kv_u64(kvs++) = bemtoh32(&u.cq.cq_consumer_counter);
8865 	kstat_kv_u64(kvs++) = bemtoh32(&u.cq.cq_last_notified);
8866 	kstat_kv_u64(kvs++) = bemtoh32(&u.cq.cq_last_solicit);
8867 
8868 	switch ((bemtoh32(&u.cq.cq_status) & MCX_CQ_CTX_STATUS_MASK) >>
8869 	    MCX_CQ_CTX_STATUS_SHIFT) {
8870 	case MCX_CQ_CTX_STATUS_OK:
8871 		text = "OK";
8872 		break;
8873 	case MCX_CQ_CTX_STATUS_OVERFLOW:
8874 		text = "overflow";
8875 		break;
8876 	case MCX_CQ_CTX_STATUS_WRITE_FAIL:
8877 		text = "write fail";
8878 		break;
8879 	default:
8880 		text = "unknown";
8881 		break;
8882 	}
8883 	strlcpy(kstat_kv_istr(kvs), text, sizeof(kstat_kv_istr(kvs)));
8884 	kvs++;
8885 
8886 	switch ((bemtoh32(&u.cq.cq_status) & MCX_CQ_CTX_STATE_MASK) >>
8887 	    MCX_CQ_CTX_STATE_SHIFT) {
8888 	case MCX_CQ_CTX_STATE_SOLICITED:
8889 		text = "solicited";
8890 		break;
8891 	case MCX_CQ_CTX_STATE_ARMED:
8892 		text = "armed";
8893 		break;
8894 	case MCX_CQ_CTX_STATE_FIRED:
8895 		text = "fired";
8896 		break;
8897 	default:
8898 		text = "unknown";
8899 		break;
8900 	}
8901 	strlcpy(kstat_kv_istr(kvs), text, sizeof(kstat_kv_istr(kvs)));
8902 	kvs++;
8903 
8904 	if (mcx_query_eq(sc, &q->q_eq, &u.eq) != 0) {
8905 		error = EIO;
8906 		goto out;
8907 	}
8908 
8909 	kstat_kv_u64(kvs++) = q->q_eq.eq_cons;
8910 	kstat_kv_u64(kvs++) = bemtoh32(&u.eq.eq_producer_counter);
8911 	kstat_kv_u64(kvs++) = bemtoh32(&u.eq.eq_consumer_counter);
8912 
8913 	switch ((bemtoh32(&u.eq.eq_status) & MCX_EQ_CTX_STATUS_MASK) >>
8914 	    MCX_EQ_CTX_STATUS_SHIFT) {
8915 	case MCX_EQ_CTX_STATUS_EQ_WRITE_FAILURE:
8916 		text = "write fail";
8917 		break;
8918 	case MCX_EQ_CTX_STATUS_OK:
8919 		text = "OK";
8920 		break;
8921 	default:
8922 		text = "unknown";
8923 		break;
8924 	}
8925 	strlcpy(kstat_kv_istr(kvs), text, sizeof(kstat_kv_istr(kvs)));
8926 	kvs++;
8927 
8928 	switch ((bemtoh32(&u.eq.eq_status) & MCX_EQ_CTX_STATE_MASK) >>
8929 	    MCX_EQ_CTX_STATE_SHIFT) {
8930 	case MCX_EQ_CTX_STATE_ARMED:
8931 		text = "armed";
8932 		break;
8933 	case MCX_EQ_CTX_STATE_FIRED:
8934 		text = "fired";
8935 		break;
8936 	default:
8937 		text = "unknown";
8938 		break;
8939 	}
8940 	strlcpy(kstat_kv_istr(kvs), text, sizeof(kstat_kv_istr(kvs)));
8941 	kvs++;
8942 
8943 	nanouptime(&ks->ks_updated);
8944 out:
8945 	KERNEL_UNLOCK();
8946 	return (error);
8947 }
8948 
8949 #endif /* NKSTAT > 0 */
8950 
8951 static unsigned int
8952 mcx_timecounter_read(struct timecounter *tc)
8953 {
8954 	struct mcx_softc *sc = tc->tc_priv;
8955 
8956 	return (mcx_rd(sc, MCX_INTERNAL_TIMER_L));
8957 }
8958 
8959 static void
8960 mcx_timecounter_attach(struct mcx_softc *sc)
8961 {
8962 	struct timecounter *tc = &sc->sc_timecounter;
8963 
8964 	tc->tc_get_timecount = mcx_timecounter_read;
8965 	tc->tc_counter_mask = ~0U;
8966 	tc->tc_frequency = sc->sc_khz * 1000;
8967 	tc->tc_name = device_xname(sc->sc_dev);
8968 	tc->tc_quality = -100;
8969 	tc->tc_priv = sc;
8970 
8971 	tc_init(tc);
8972 }
8973