xref: /netbsd-src/sys/dev/pci/if_mcx.c (revision 627f7eb200a4419d89b531d55fccd2ee3ffdcde0)
1 /*	$NetBSD: if_mcx.c,v 1.17 2021/02/20 13:31:35 jmcneill Exp $ */
2 /*	$OpenBSD: if_mcx.c,v 1.99 2021/02/15 03:42:00 dlg 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 1.17 2021/02/20 13:31:35 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 structres
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 	uint8_t enaddr[ETHER_ADDR_LEN];
2747 	int counts[PCI_INTR_TYPE_SIZE];
2748 	char intrxname[32];
2749 	pcireg_t memtype;
2750 	uint32_t r;
2751 	unsigned int cq_stride;
2752 	unsigned int cq_size;
2753 	int i, msix;
2754 	kcpuset_t *affinity;
2755 
2756 	sc->sc_dev = self;
2757 	sc->sc_pc = pa->pa_pc;
2758 	sc->sc_tag = pa->pa_tag;
2759 	if (pci_dma64_available(pa))
2760 		sc->sc_dmat = pa->pa_dmat64;
2761 	else
2762 		sc->sc_dmat = pa->pa_dmat;
2763 
2764 	/* Map the PCI memory space */
2765 	memtype = pci_mapreg_type(sc->sc_pc, sc->sc_tag, MCX_HCA_BAR);
2766 	if (pci_mapreg_map(pa, MCX_HCA_BAR, memtype,
2767 	    0 /*BUS_SPACE_MAP_PREFETCHABLE*/, &sc->sc_memt, &sc->sc_memh,
2768 	    NULL, &sc->sc_mems)) {
2769 		aprint_error(": unable to map register memory\n");
2770 		return;
2771 	}
2772 
2773 	pci_aprint_devinfo(pa, "Ethernet controller");
2774 
2775 	mutex_init(&sc->sc_media_mutex, MUTEX_DEFAULT, IPL_SOFTNET);
2776 
2777 	if (mcx_version(sc) != 0) {
2778 		/* error printed by mcx_version */
2779 		goto unmap;
2780 	}
2781 
2782 	r = mcx_rd(sc, MCX_CMDQ_ADDR_LO);
2783 	cq_stride = 1 << MCX_CMDQ_LOG_STRIDE(r); /* size of the entries */
2784 	cq_size = 1 << MCX_CMDQ_LOG_SIZE(r); /* number of entries */
2785 	if (cq_size > MCX_MAX_CQE) {
2786 		aprint_error_dev(self,
2787 		    "command queue size overflow %u\n", cq_size);
2788 		goto unmap;
2789 	}
2790 	if (cq_stride < sizeof(struct mcx_cmdq_entry)) {
2791 		aprint_error_dev(self,
2792 		    "command queue entry size underflow %u\n", cq_stride);
2793 		goto unmap;
2794 	}
2795 	if (cq_stride * cq_size > MCX_PAGE_SIZE) {
2796 		aprint_error_dev(self, "command queue page overflow\n");
2797 		goto unmap;
2798 	}
2799 
2800 	if (mcx_dmamem_alloc(sc, &sc->sc_doorbell_mem, MCX_DOORBELL_AREA_SIZE,
2801 	    MCX_PAGE_SIZE) != 0) {
2802 		aprint_error_dev(self, "unable to allocate doorbell memory\n");
2803 		goto unmap;
2804 	}
2805 
2806 	if (mcx_dmamem_alloc(sc, &sc->sc_cmdq_mem, MCX_PAGE_SIZE,
2807 	    MCX_PAGE_SIZE) != 0) {
2808 		aprint_error_dev(self, "unable to allocate command queue\n");
2809 		goto dbfree;
2810 	}
2811 
2812 	mcx_wr(sc, MCX_CMDQ_ADDR_HI, MCX_DMA_DVA(&sc->sc_cmdq_mem) >> 32);
2813 	mcx_bar(sc, MCX_CMDQ_ADDR_HI, sizeof(uint32_t),
2814 	    BUS_SPACE_BARRIER_WRITE);
2815 	mcx_wr(sc, MCX_CMDQ_ADDR_LO, MCX_DMA_DVA(&sc->sc_cmdq_mem));
2816 	mcx_bar(sc, MCX_CMDQ_ADDR_LO, sizeof(uint32_t),
2817 	    BUS_SPACE_BARRIER_WRITE);
2818 
2819 	if (mcx_init_wait(sc) != 0) {
2820 		aprint_error_dev(self, "timeout waiting for init\n");
2821 		goto cqfree;
2822 	}
2823 
2824 	sc->sc_cmdq_mask = cq_size - 1;
2825 	sc->sc_cmdq_size = cq_stride;
2826 
2827 	if (mcx_enable_hca(sc) != 0) {
2828 		/* error printed by mcx_enable_hca */
2829 		goto cqfree;
2830 	}
2831 
2832 	if (mcx_issi(sc) != 0) {
2833 		/* error printed by mcx_issi */
2834 		goto teardown;
2835 	}
2836 
2837 	if (mcx_pages(sc, &sc->sc_boot_pages,
2838 	    htobe16(MCX_CMD_QUERY_PAGES_BOOT)) != 0) {
2839 		/* error printed by mcx_pages */
2840 		goto teardown;
2841 	}
2842 
2843 	if (mcx_hca_max_caps(sc) != 0) {
2844 		/* error printed by mcx_hca_max_caps */
2845 		goto teardown;
2846 	}
2847 
2848 	if (mcx_hca_set_caps(sc) != 0) {
2849 		/* error printed by mcx_hca_set_caps */
2850 		goto teardown;
2851 	}
2852 
2853 	if (mcx_pages(sc, &sc->sc_init_pages,
2854 	    htobe16(MCX_CMD_QUERY_PAGES_INIT)) != 0) {
2855 		/* error printed by mcx_pages */
2856 		goto teardown;
2857 	}
2858 
2859 	if (mcx_init_hca(sc) != 0) {
2860 		/* error printed by mcx_init_hca */
2861 		goto teardown;
2862 	}
2863 
2864 	if (mcx_pages(sc, &sc->sc_regular_pages,
2865 	    htobe16(MCX_CMD_QUERY_PAGES_REGULAR)) != 0) {
2866 		/* error printed by mcx_pages */
2867 		goto teardown;
2868 	}
2869 
2870 	/* apparently not necessary? */
2871 	if (mcx_set_driver_version(sc) != 0) {
2872 		/* error printed by mcx_set_driver_version */
2873 		goto teardown;
2874 	}
2875 
2876 	if (mcx_iff(sc) != 0) {	/* modify nic vport context */
2877 		/* error printed by mcx_iff? */
2878 		goto teardown;
2879 	}
2880 
2881 	if (mcx_alloc_uar(sc, &sc->sc_uar) != 0) {
2882 		/* error printed by mcx_alloc_uar */
2883 		goto teardown;
2884 	}
2885 
2886 	if (mcx_alloc_pd(sc) != 0) {
2887 		/* error printed by mcx_alloc_pd */
2888 		goto teardown;
2889 	}
2890 
2891 	if (mcx_alloc_tdomain(sc) != 0) {
2892 		/* error printed by mcx_alloc_tdomain */
2893 		goto teardown;
2894 	}
2895 
2896 	/*
2897 	 * PRM makes no mention of msi interrupts, just legacy and msi-x.
2898 	 * mellanox support tells me legacy interrupts are not supported,
2899 	 * so we're stuck with just msi-x.
2900 	 */
2901 	counts[PCI_INTR_TYPE_MSIX] = -1;
2902 	counts[PCI_INTR_TYPE_MSI] = 0;
2903 	counts[PCI_INTR_TYPE_INTX] = 0;
2904 	if (pci_intr_alloc(pa, &sc->sc_intrs, counts, PCI_INTR_TYPE_MSIX) != 0) {
2905 		aprint_error_dev(self, "unable to allocate interrupt\n");
2906 		goto teardown;
2907 	}
2908 	if (counts[PCI_INTR_TYPE_MSIX] < 2) {
2909 		aprint_error_dev(self, "not enough MSI-X vectors\n");
2910 		goto teardown;
2911 	}
2912 	KASSERT(pci_intr_type(sc->sc_pc, sc->sc_intrs[0]) == PCI_INTR_TYPE_MSIX);
2913 	snprintf(intrxname, sizeof(intrxname), "%s adminq", DEVNAME(sc));
2914 	sc->sc_ihc = mcx_establish_intr(sc, 0, NULL, mcx_admin_intr, sc,
2915 	    intrxname);
2916 	if (sc->sc_ihc == NULL) {
2917 		aprint_error_dev(self, "couldn't establish adminq interrupt\n");
2918 		goto teardown;
2919 	}
2920 
2921 	if (mcx_create_eq(sc, &sc->sc_admin_eq, sc->sc_uar,
2922 	    (1ull << MCX_EVENT_TYPE_INTERNAL_ERROR) |
2923 	    (1ull << MCX_EVENT_TYPE_PORT_CHANGE) |
2924 	    (1ull << MCX_EVENT_TYPE_CMD_COMPLETION) |
2925 	    (1ull << MCX_EVENT_TYPE_PAGE_REQUEST), 0) != 0) {
2926 		/* error printed by mcx_create_eq */
2927 		goto teardown;
2928 	}
2929 
2930 	if (mcx_query_nic_vport_context(sc, enaddr) != 0) {
2931 		/* error printed by mcx_query_nic_vport_context */
2932 		goto teardown;
2933 	}
2934 
2935 	if (mcx_query_special_contexts(sc) != 0) {
2936 		/* error printed by mcx_query_special_contexts */
2937 		goto teardown;
2938 	}
2939 
2940 	if (mcx_set_port_mtu(sc, MCX_HARDMTU) != 0) {
2941 		/* error printed by mcx_set_port_mtu */
2942 		goto teardown;
2943 	}
2944 
2945 	aprint_normal_dev(sc->sc_dev, "Ethernet address %s\n",
2946 	    ether_sprintf(enaddr));
2947 
2948 	msix = counts[PCI_INTR_TYPE_MSIX];
2949 	msix--; /* admin ops took one */
2950 
2951 	sc->sc_nqueues = uimin(MCX_MAX_QUEUES, msix);
2952 	sc->sc_nqueues = uimin(sc->sc_nqueues, ncpu);
2953 	sc->sc_queues = kmem_zalloc(sc->sc_nqueues * sizeof(*sc->sc_queues),
2954 	    KM_SLEEP);
2955 
2956 	strlcpy(ifp->if_xname, DEVNAME(sc), IFNAMSIZ);
2957 	ifp->if_softc = sc;
2958 	ifp->if_flags = IFF_BROADCAST | IFF_MULTICAST | IFF_SIMPLEX;
2959 #ifdef MCX_MPSAFE
2960 	ifp->if_extflags = IFEF_MPSAFE;
2961 #endif
2962 	ifp->if_init = mcx_init;
2963 	ifp->if_stop = mcx_stop;
2964 	ifp->if_ioctl = mcx_ioctl;
2965 	ifp->if_start = mcx_start;
2966 	if (sc->sc_nqueues > 1) {
2967 		ifp->if_transmit = mcx_transmit;
2968 	}
2969 	ifp->if_watchdog = mcx_watchdog;
2970 	ifp->if_mtu = sc->sc_hardmtu;
2971 	ifp->if_capabilities = IFCAP_CSUM_IPv4_Rx | IFCAP_CSUM_IPv4_Tx |
2972 	    IFCAP_CSUM_UDPv4_Rx | IFCAP_CSUM_UDPv4_Tx |
2973 	    IFCAP_CSUM_UDPv6_Rx | IFCAP_CSUM_UDPv6_Tx |
2974 	    IFCAP_CSUM_TCPv4_Rx | IFCAP_CSUM_TCPv4_Tx |
2975 	    IFCAP_CSUM_TCPv6_Rx | IFCAP_CSUM_TCPv6_Tx;
2976 	IFQ_SET_MAXLEN(&ifp->if_snd, 1024);
2977 	IFQ_SET_READY(&ifp->if_snd);
2978 
2979 	sc->sc_ec.ec_capabilities = ETHERCAP_JUMBO_MTU |
2980 	    ETHERCAP_VLAN_MTU | ETHERCAP_VLAN_HWTAGGING;
2981 	sc->sc_ec.ec_capenable |= ETHERCAP_VLAN_HWTAGGING;
2982 
2983 	sc->sc_ec.ec_ifmedia = &sc->sc_media;
2984 	ifmedia_init_with_lock(&sc->sc_media, IFM_IMASK, mcx_media_change,
2985 	    mcx_media_status, &sc->sc_media_mutex);
2986 	mcx_media_add_types(sc);
2987 	ifmedia_add(&sc->sc_media, IFM_ETHER | IFM_AUTO, 0, NULL);
2988 	ifmedia_set(&sc->sc_media, IFM_ETHER | IFM_AUTO);
2989 
2990 	if_attach(ifp);
2991 	if_deferred_start_init(ifp, NULL);
2992 
2993 	ether_ifattach(ifp, enaddr);
2994 
2995 	kcpuset_create(&affinity, false);
2996 	kcpuset_set(affinity, 0);
2997 
2998 	for (i = 0; i < sc->sc_nqueues; i++) {
2999 		struct mcx_queues *q = &sc->sc_queues[i];
3000 		struct mcx_rx *rx = &q->q_rx;
3001 		struct mcx_tx *tx = &q->q_tx;
3002 		int vec;
3003 
3004 		vec = i + 1;
3005 		q->q_sc = sc;
3006 		q->q_index = i;
3007 
3008 		if (mcx_alloc_uar(sc, &q->q_uar) != 0) {
3009 			aprint_error_dev(self, "unable to alloc uar %d\n", i);
3010 			goto teardown;
3011 		}
3012 
3013 		if (mcx_create_eq(sc, &q->q_eq, q->q_uar, 0, vec) != 0) {
3014 			aprint_error_dev(self,
3015 			    "unable to create event queue %d\n", i);
3016 			goto teardown;
3017 		}
3018 
3019 		rx->rx_softc = sc;
3020 		callout_init(&rx->rx_refill, CALLOUT_FLAGS);
3021 		callout_setfunc(&rx->rx_refill, mcx_refill, rx);
3022 
3023 		tx->tx_softc = sc;
3024 		mutex_init(&tx->tx_lock, MUTEX_DEFAULT, IPL_NET);
3025 		tx->tx_pcq = pcq_create(MCX_TXQ_NUM, KM_SLEEP);
3026 		tx->tx_softint = softint_establish(SOFTINT_NET|SOFTINT_MPSAFE,
3027 		    mcx_deferred_transmit, tx);
3028 
3029 		snprintf(intrxname, sizeof(intrxname), "%s queue %d",
3030 		    DEVNAME(sc), i);
3031 		q->q_ihc = mcx_establish_intr(sc, vec, affinity, mcx_cq_intr,
3032 		    q, intrxname);
3033 	}
3034 
3035 	callout_init(&sc->sc_calibrate, CALLOUT_FLAGS);
3036 	callout_setfunc(&sc->sc_calibrate, mcx_calibrate, sc);
3037 
3038 	if (workqueue_create(&sc->sc_workq, "mcxportchg", mcx_port_change, sc,
3039 	    PRI_NONE, IPL_NET, 0) != 0) {
3040 		aprint_error_dev(self, "couldn't create port change workq\n");
3041 		goto teardown;
3042 	}
3043 
3044 	mcx_port_change(&sc->sc_port_change, sc);
3045 
3046 	sc->sc_mac_flow_table_id = -1;
3047 	sc->sc_rss_flow_table_id = -1;
3048 	sc->sc_rqt = -1;
3049 	for (i = 0; i < MCX_NUM_FLOW_GROUPS; i++) {
3050 		struct mcx_flow_group *mfg = &sc->sc_flow_group[i];
3051 		mfg->g_id = -1;
3052 		mfg->g_table = -1;
3053 		mfg->g_size = 0;
3054 		mfg->g_start = 0;
3055 	}
3056 	sc->sc_extra_mcast = 0;
3057 	memset(sc->sc_mcast_flows, 0, sizeof(sc->sc_mcast_flows));
3058 
3059 #if NKSTAT > 0
3060 	mcx_kstat_attach(sc);
3061 #endif
3062 	mcx_timecounter_attach(sc);
3063 	return;
3064 
3065 teardown:
3066 	mcx_teardown_hca(sc, htobe16(MCX_CMD_TEARDOWN_HCA_GRACEFUL));
3067 	/* error printed by mcx_teardown_hca, and we're already unwinding */
3068 cqfree:
3069 	mcx_wr(sc, MCX_CMDQ_ADDR_HI, MCX_DMA_DVA(&sc->sc_cmdq_mem) >> 32);
3070 	mcx_bar(sc, MCX_CMDQ_ADDR_HI, sizeof(uint64_t),
3071 	    BUS_SPACE_BARRIER_WRITE);
3072 	mcx_wr(sc, MCX_CMDQ_ADDR_LO, MCX_DMA_DVA(&sc->sc_cmdq_mem) |
3073 	    MCX_CMDQ_INTERFACE_DISABLED);
3074 	mcx_bar(sc, MCX_CMDQ_ADDR_LO, sizeof(uint64_t),
3075 	    BUS_SPACE_BARRIER_WRITE);
3076 
3077 	mcx_wr(sc, MCX_CMDQ_ADDR_HI, 0);
3078 	mcx_bar(sc, MCX_CMDQ_ADDR_HI, sizeof(uint64_t),
3079 	    BUS_SPACE_BARRIER_WRITE);
3080 	mcx_wr(sc, MCX_CMDQ_ADDR_LO, MCX_CMDQ_INTERFACE_DISABLED);
3081 
3082 	mcx_dmamem_free(sc, &sc->sc_cmdq_mem);
3083 dbfree:
3084 	mcx_dmamem_free(sc, &sc->sc_doorbell_mem);
3085 unmap:
3086 	bus_space_unmap(sc->sc_memt, sc->sc_memh, sc->sc_mems);
3087 	sc->sc_mems = 0;
3088 }
3089 
3090 static void *
3091 mcx_establish_intr(struct mcx_softc *sc, int index, kcpuset_t *affinity,
3092     int (*func)(void *), void *arg, const char *xname)
3093 {
3094 	char intrbuf[PCI_INTRSTR_LEN];
3095 	const char *intrstr;
3096 	void *ih;
3097 
3098 	pci_intr_setattr(sc->sc_pc, &sc->sc_intrs[index], PCI_INTR_MPSAFE,
3099 	    true);
3100 
3101 	intrstr = pci_intr_string(sc->sc_pc, sc->sc_intrs[index], intrbuf,
3102 	    sizeof(intrbuf));
3103 	ih = pci_intr_establish_xname(sc->sc_pc, sc->sc_intrs[index], IPL_NET,
3104 	    func, arg, xname);
3105 	if (ih == NULL) {
3106 		aprint_error_dev(sc->sc_dev,
3107 		    "unable to establish interrupt%s%s\n",
3108 		    intrstr ? " at " : "",
3109 		    intrstr ? intrstr : "");
3110 		return NULL;
3111 	}
3112 
3113 	if (affinity != NULL && index > 0) {
3114 		/* Round-robin affinity */
3115 		kcpuset_zero(affinity);
3116 		kcpuset_set(affinity, (index - 1) % ncpu);
3117 		interrupt_distribute(ih, affinity, NULL);
3118 	}
3119 
3120 	return ih;
3121 }
3122 
3123 static void
3124 mcx_rxr_init(struct mcx_rxring *rxr, u_int lwm __unused, u_int hwm)
3125 {
3126 	rxr->rxr_total = hwm;
3127 	rxr->rxr_inuse = 0;
3128 }
3129 
3130 static u_int
3131 mcx_rxr_get(struct mcx_rxring *rxr, u_int max)
3132 {
3133 	const u_int taken = MIN(max, rxr->rxr_total - rxr->rxr_inuse);
3134 
3135 	rxr->rxr_inuse += taken;
3136 
3137 	return taken;
3138 }
3139 
3140 static void
3141 mcx_rxr_put(struct mcx_rxring *rxr, u_int n)
3142 {
3143 	rxr->rxr_inuse -= n;
3144 }
3145 
3146 static u_int
3147 mcx_rxr_inuse(struct mcx_rxring *rxr)
3148 {
3149 	return rxr->rxr_inuse;
3150 }
3151 
3152 static int
3153 mcx_version(struct mcx_softc *sc)
3154 {
3155 	uint32_t fw0, fw1;
3156 	uint16_t cmdif;
3157 
3158 	fw0 = mcx_rd(sc, MCX_FW_VER);
3159 	fw1 = mcx_rd(sc, MCX_CMDIF_FW_SUBVER);
3160 
3161 	aprint_normal_dev(sc->sc_dev, "FW %u.%u.%04u\n", MCX_FW_VER_MAJOR(fw0),
3162 	    MCX_FW_VER_MINOR(fw0), MCX_FW_VER_SUBMINOR(fw1));
3163 
3164 	cmdif = MCX_CMDIF(fw1);
3165 	if (cmdif != MCX_CMD_IF_SUPPORTED) {
3166 		aprint_error_dev(sc->sc_dev,
3167 		    "unsupported command interface %u\n", cmdif);
3168 		return (-1);
3169 	}
3170 
3171 	return (0);
3172 }
3173 
3174 static int
3175 mcx_init_wait(struct mcx_softc *sc)
3176 {
3177 	unsigned int i;
3178 	uint32_t r;
3179 
3180 	for (i = 0; i < 2000; i++) {
3181 		r = mcx_rd(sc, MCX_STATE);
3182 		if ((r & MCX_STATE_MASK) == MCX_STATE_READY)
3183 			return (0);
3184 
3185 		delay(1000);
3186 		mcx_bar(sc, MCX_STATE, sizeof(uint32_t),
3187 		    BUS_SPACE_BARRIER_READ);
3188 	}
3189 
3190 	return (-1);
3191 }
3192 
3193 static uint8_t
3194 mcx_cmdq_poll(struct mcx_softc *sc, struct mcx_cmdq_entry *cqe,
3195     unsigned int msec)
3196 {
3197 	unsigned int i;
3198 
3199 	for (i = 0; i < msec; i++) {
3200 		bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_cmdq_mem),
3201 		    0, MCX_DMA_LEN(&sc->sc_cmdq_mem), BUS_DMASYNC_POSTRW);
3202 
3203 		if ((cqe->cq_status & MCX_CQ_STATUS_OWN_MASK) ==
3204 		    MCX_CQ_STATUS_OWN_SW)
3205 			return (0);
3206 
3207 		delay(1000);
3208 	}
3209 
3210 	return (ETIMEDOUT);
3211 }
3212 
3213 static uint32_t
3214 mcx_mix_u64(uint32_t xor, uint64_t u64)
3215 {
3216 	xor ^= u64 >> 32;
3217 	xor ^= u64;
3218 
3219 	return (xor);
3220 }
3221 
3222 static uint32_t
3223 mcx_mix_u32(uint32_t xor, uint32_t u32)
3224 {
3225 	xor ^= u32;
3226 
3227 	return (xor);
3228 }
3229 
3230 static uint32_t
3231 mcx_mix_u8(uint32_t xor, uint8_t u8)
3232 {
3233 	xor ^= u8;
3234 
3235 	return (xor);
3236 }
3237 
3238 static uint8_t
3239 mcx_mix_done(uint32_t xor)
3240 {
3241 	xor ^= xor >> 16;
3242 	xor ^= xor >> 8;
3243 
3244 	return (xor);
3245 }
3246 
3247 static uint8_t
3248 mcx_xor(const void *buf, size_t len)
3249 {
3250 	const uint32_t *dwords = buf;
3251 	uint32_t xor = 0xff;
3252 	size_t i;
3253 
3254 	len /= sizeof(*dwords);
3255 
3256 	for (i = 0; i < len; i++)
3257 		xor ^= dwords[i];
3258 
3259 	return (mcx_mix_done(xor));
3260 }
3261 
3262 static uint8_t
3263 mcx_cmdq_token(struct mcx_softc *sc)
3264 {
3265 	uint8_t token;
3266 
3267 	do {
3268 		token = ++sc->sc_cmdq_token;
3269 	} while (token == 0);
3270 
3271 	return (token);
3272 }
3273 
3274 static void
3275 mcx_cmdq_init(struct mcx_softc *sc, struct mcx_cmdq_entry *cqe,
3276     uint32_t ilen, uint32_t olen, uint8_t token)
3277 {
3278 	memset(cqe, 0, sc->sc_cmdq_size);
3279 
3280 	cqe->cq_type = MCX_CMDQ_TYPE_PCIE;
3281 	be32enc(&cqe->cq_input_length, ilen);
3282 	be32enc(&cqe->cq_output_length, olen);
3283 	cqe->cq_token = token;
3284 	cqe->cq_status = MCX_CQ_STATUS_OWN_HW;
3285 }
3286 
3287 static void
3288 mcx_cmdq_sign(struct mcx_cmdq_entry *cqe)
3289 {
3290 	cqe->cq_signature = ~mcx_xor(cqe, sizeof(*cqe));
3291 }
3292 
3293 static int
3294 mcx_cmdq_verify(const struct mcx_cmdq_entry *cqe)
3295 {
3296 	/* return (mcx_xor(cqe, sizeof(*cqe)) ? -1 :  0); */
3297 	return (0);
3298 }
3299 
3300 static void *
3301 mcx_cmdq_in(struct mcx_cmdq_entry *cqe)
3302 {
3303 	return (&cqe->cq_input_data);
3304 }
3305 
3306 static void *
3307 mcx_cmdq_out(struct mcx_cmdq_entry *cqe)
3308 {
3309 	return (&cqe->cq_output_data);
3310 }
3311 
3312 static void
3313 mcx_cmdq_post(struct mcx_softc *sc, struct mcx_cmdq_entry *cqe,
3314     unsigned int slot)
3315 {
3316 	mcx_cmdq_sign(cqe);
3317 
3318 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_cmdq_mem),
3319 	    0, MCX_DMA_LEN(&sc->sc_cmdq_mem), BUS_DMASYNC_PRERW);
3320 
3321 	mcx_wr(sc, MCX_CMDQ_DOORBELL, 1U << slot);
3322 	mcx_bar(sc, MCX_CMDQ_DOORBELL, sizeof(uint32_t),
3323 	    BUS_SPACE_BARRIER_WRITE);
3324 }
3325 
3326 static int
3327 mcx_enable_hca(struct mcx_softc *sc)
3328 {
3329 	struct mcx_cmdq_entry *cqe;
3330 	struct mcx_cmd_enable_hca_in *in;
3331 	struct mcx_cmd_enable_hca_out *out;
3332 	int error;
3333 	uint8_t status;
3334 
3335 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
3336 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
3337 
3338 	in = mcx_cmdq_in(cqe);
3339 	in->cmd_opcode = htobe16(MCX_CMD_ENABLE_HCA);
3340 	in->cmd_op_mod = htobe16(0);
3341 	in->cmd_function_id = htobe16(0);
3342 
3343 	mcx_cmdq_post(sc, cqe, 0);
3344 
3345 	error = mcx_cmdq_poll(sc, cqe, 1000);
3346 	if (error != 0) {
3347 		printf(", hca enable timeout\n");
3348 		return (-1);
3349 	}
3350 	if (mcx_cmdq_verify(cqe) != 0) {
3351 		printf(", hca enable command corrupt\n");
3352 		return (-1);
3353 	}
3354 
3355 	status = cqe->cq_output_data[0];
3356 	if (status != MCX_CQ_STATUS_OK) {
3357 		printf(", hca enable failed (%x)\n", status);
3358 		return (-1);
3359 	}
3360 
3361 	return (0);
3362 }
3363 
3364 static int
3365 mcx_teardown_hca(struct mcx_softc *sc, uint16_t profile)
3366 {
3367 	struct mcx_cmdq_entry *cqe;
3368 	struct mcx_cmd_teardown_hca_in *in;
3369 	struct mcx_cmd_teardown_hca_out *out;
3370 	int error;
3371 	uint8_t status;
3372 
3373 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
3374 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
3375 
3376 	in = mcx_cmdq_in(cqe);
3377 	in->cmd_opcode = htobe16(MCX_CMD_TEARDOWN_HCA);
3378 	in->cmd_op_mod = htobe16(0);
3379 	in->cmd_profile = profile;
3380 
3381 	mcx_cmdq_post(sc, cqe, 0);
3382 
3383 	error = mcx_cmdq_poll(sc, cqe, 1000);
3384 	if (error != 0) {
3385 		printf(", hca teardown timeout\n");
3386 		return (-1);
3387 	}
3388 	if (mcx_cmdq_verify(cqe) != 0) {
3389 		printf(", hca teardown command corrupt\n");
3390 		return (-1);
3391 	}
3392 
3393 	status = cqe->cq_output_data[0];
3394 	if (status != MCX_CQ_STATUS_OK) {
3395 		printf(", hca teardown failed (%x)\n", status);
3396 		return (-1);
3397 	}
3398 
3399 	return (0);
3400 }
3401 
3402 static int
3403 mcx_cmdq_mboxes_alloc(struct mcx_softc *sc, struct mcx_dmamem *mxm,
3404     unsigned int nmb, uint64_t *ptr, uint8_t token)
3405 {
3406 	uint8_t *kva;
3407 	uint64_t dva;
3408 	int i;
3409 	int error;
3410 
3411 	error = mcx_dmamem_alloc(sc, mxm,
3412 	    nmb * MCX_CMDQ_MAILBOX_SIZE, MCX_CMDQ_MAILBOX_ALIGN);
3413 	if (error != 0)
3414 		return (error);
3415 
3416 	mcx_dmamem_zero(mxm);
3417 
3418 	dva = MCX_DMA_DVA(mxm);
3419 	kva = MCX_DMA_KVA(mxm);
3420 	for (i = 0; i < nmb; i++) {
3421 		struct mcx_cmdq_mailbox *mbox = (struct mcx_cmdq_mailbox *)kva;
3422 
3423 		/* patch the cqe or mbox pointing at this one */
3424 		be64enc(ptr, dva);
3425 
3426 		/* fill in this mbox */
3427 		be32enc(&mbox->mb_block_number, i);
3428 		mbox->mb_token = token;
3429 
3430 		/* move to the next one */
3431 		ptr = &mbox->mb_next_ptr;
3432 
3433 		dva += MCX_CMDQ_MAILBOX_SIZE;
3434 		kva += MCX_CMDQ_MAILBOX_SIZE;
3435 	}
3436 
3437 	return (0);
3438 }
3439 
3440 static uint32_t
3441 mcx_cmdq_mbox_ctrl_sig(const struct mcx_cmdq_mailbox *mb)
3442 {
3443 	uint32_t xor = 0xff;
3444 
3445 	/* only 3 fields get set, so mix them directly */
3446 	xor = mcx_mix_u64(xor, mb->mb_next_ptr);
3447 	xor = mcx_mix_u32(xor, mb->mb_block_number);
3448 	xor = mcx_mix_u8(xor, mb->mb_token);
3449 
3450 	return (mcx_mix_done(xor));
3451 }
3452 
3453 static void
3454 mcx_cmdq_mboxes_sign(struct mcx_dmamem *mxm, unsigned int nmb)
3455 {
3456 	uint8_t *kva;
3457 	int i;
3458 
3459 	kva = MCX_DMA_KVA(mxm);
3460 
3461 	for (i = 0; i < nmb; i++) {
3462 		struct mcx_cmdq_mailbox *mb = (struct mcx_cmdq_mailbox *)kva;
3463 		uint8_t sig = mcx_cmdq_mbox_ctrl_sig(mb);
3464 		mb->mb_ctrl_signature = sig;
3465 		mb->mb_signature = sig ^
3466 		    mcx_xor(mb->mb_data, sizeof(mb->mb_data));
3467 
3468 		kva += MCX_CMDQ_MAILBOX_SIZE;
3469 	}
3470 }
3471 
3472 static void
3473 mcx_cmdq_mboxes_sync(struct mcx_softc *sc, struct mcx_dmamem *mxm, int ops)
3474 {
3475 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(mxm),
3476 	    0, MCX_DMA_LEN(mxm), ops);
3477 }
3478 
3479 static struct mcx_cmdq_mailbox *
3480 mcx_cq_mbox(struct mcx_dmamem *mxm, unsigned int i)
3481 {
3482 	uint8_t *kva;
3483 
3484 	kva = MCX_DMA_KVA(mxm);
3485 	kva += i * MCX_CMDQ_MAILBOX_SIZE;
3486 
3487 	return ((struct mcx_cmdq_mailbox *)kva);
3488 }
3489 
3490 static inline void *
3491 mcx_cq_mbox_data(struct mcx_cmdq_mailbox *mb)
3492 {
3493 	return (&mb->mb_data);
3494 }
3495 
3496 static void
3497 mcx_cmdq_mboxes_copyin(struct mcx_dmamem *mxm, unsigned int nmb,
3498     void *b, size_t len)
3499 {
3500 	uint8_t *buf = b;
3501 	struct mcx_cmdq_mailbox *mb;
3502 	int i;
3503 
3504 	mb = (struct mcx_cmdq_mailbox *)MCX_DMA_KVA(mxm);
3505 	for (i = 0; i < nmb; i++) {
3506 
3507 		memcpy(mb->mb_data, buf, uimin(sizeof(mb->mb_data), len));
3508 
3509 		if (sizeof(mb->mb_data) >= len)
3510 			break;
3511 
3512 		buf += sizeof(mb->mb_data);
3513 		len -= sizeof(mb->mb_data);
3514 		mb++;
3515 	}
3516 }
3517 
3518 static void
3519 mcx_cmdq_mboxes_pas(struct mcx_dmamem *mxm, int offset, int npages,
3520     struct mcx_dmamem *buf)
3521 {
3522 	uint64_t *pas;
3523 	int mbox, mbox_pages, i;
3524 
3525 	mbox = offset / MCX_CMDQ_MAILBOX_DATASIZE;
3526 	offset %= MCX_CMDQ_MAILBOX_DATASIZE;
3527 
3528 	pas = mcx_cq_mbox_data(mcx_cq_mbox(mxm, mbox));
3529 	pas += (offset / sizeof(*pas));
3530 	mbox_pages = (MCX_CMDQ_MAILBOX_DATASIZE - offset) / sizeof(*pas);
3531 	for (i = 0; i < npages; i++) {
3532 		if (i == mbox_pages) {
3533 			mbox++;
3534 			pas = mcx_cq_mbox_data(mcx_cq_mbox(mxm, mbox));
3535 			mbox_pages += MCX_CMDQ_MAILBOX_DATASIZE / sizeof(*pas);
3536 		}
3537 		*pas = htobe64(MCX_DMA_DVA(buf) + (i * MCX_PAGE_SIZE));
3538 		pas++;
3539 	}
3540 }
3541 
3542 static void
3543 mcx_cmdq_mboxes_copyout(struct mcx_dmamem *mxm, int nmb, void *b, size_t len)
3544 {
3545 	uint8_t *buf = b;
3546 	struct mcx_cmdq_mailbox *mb;
3547 	int i;
3548 
3549 	mb = (struct mcx_cmdq_mailbox *)MCX_DMA_KVA(mxm);
3550 	for (i = 0; i < nmb; i++) {
3551 		memcpy(buf, mb->mb_data, uimin(sizeof(mb->mb_data), len));
3552 
3553 		if (sizeof(mb->mb_data) >= len)
3554 			break;
3555 
3556 		buf += sizeof(mb->mb_data);
3557 		len -= sizeof(mb->mb_data);
3558 		mb++;
3559 	}
3560 }
3561 
3562 static void
3563 mcx_cq_mboxes_free(struct mcx_softc *sc, struct mcx_dmamem *mxm)
3564 {
3565 	mcx_dmamem_free(sc, mxm);
3566 }
3567 
3568 #if 0
3569 static void
3570 mcx_cmdq_dump(const struct mcx_cmdq_entry *cqe)
3571 {
3572 	unsigned int i;
3573 
3574 	printf(" type %02x, ilen %u, iptr %016llx", cqe->cq_type,
3575 	    be32dec(&cqe->cq_input_length), be64dec(&cqe->cq_input_ptr));
3576 
3577 	printf(", idata ");
3578 	for (i = 0; i < sizeof(cqe->cq_input_data); i++)
3579 		printf("%02x", cqe->cq_input_data[i]);
3580 
3581 	printf(", odata ");
3582 	for (i = 0; i < sizeof(cqe->cq_output_data); i++)
3583 		printf("%02x", cqe->cq_output_data[i]);
3584 
3585 	printf(", optr %016llx, olen %u, token %02x, sig %02x, status %02x",
3586 	    be64dec(&cqe->cq_output_ptr), be32dec(&cqe->cq_output_length),
3587 	    cqe->cq_token, cqe->cq_signature, cqe->cq_status);
3588 }
3589 
3590 static void
3591 mcx_cmdq_mbox_dump(struct mcx_dmamem *mboxes, int num)
3592 {
3593 	int i, j;
3594 	uint8_t *d;
3595 
3596 	for (i = 0; i < num; i++) {
3597 		struct mcx_cmdq_mailbox *mbox;
3598 		mbox = mcx_cq_mbox(mboxes, i);
3599 
3600 		d = mcx_cq_mbox_data(mbox);
3601 		for (j = 0; j < MCX_CMDQ_MAILBOX_DATASIZE; j++) {
3602 			if (j != 0 && (j % 16 == 0))
3603 				printf("\n");
3604 			printf("%.2x ", d[j]);
3605 		}
3606 	}
3607 }
3608 #endif
3609 
3610 static int
3611 mcx_access_hca_reg(struct mcx_softc *sc, uint16_t reg, int op, void *data,
3612     int len)
3613 {
3614 	struct mcx_dmamem mxm;
3615 	struct mcx_cmdq_entry *cqe;
3616 	struct mcx_cmd_access_reg_in *in;
3617 	struct mcx_cmd_access_reg_out *out;
3618 	uint8_t token = mcx_cmdq_token(sc);
3619 	int error, nmb;
3620 
3621 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
3622 	mcx_cmdq_init(sc, cqe, sizeof(*in) + len, sizeof(*out) + len,
3623 	    token);
3624 
3625 	in = mcx_cmdq_in(cqe);
3626 	in->cmd_opcode = htobe16(MCX_CMD_ACCESS_REG);
3627 	in->cmd_op_mod = htobe16(op);
3628 	in->cmd_register_id = htobe16(reg);
3629 
3630 	nmb = howmany(len, MCX_CMDQ_MAILBOX_DATASIZE);
3631 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, nmb,
3632 	    &cqe->cq_output_ptr, token) != 0) {
3633 		printf(", unable to allocate access reg mailboxen\n");
3634 		return (-1);
3635 	}
3636 	cqe->cq_input_ptr = cqe->cq_output_ptr;
3637 	mcx_cmdq_mboxes_copyin(&mxm, nmb, data, len);
3638 	mcx_cmdq_mboxes_sign(&mxm, nmb);
3639 	mcx_cmdq_mboxes_sync(sc, &mxm, BUS_DMASYNC_PRERW);
3640 
3641 	mcx_cmdq_post(sc, cqe, 0);
3642 	error = mcx_cmdq_poll(sc, cqe, 1000);
3643 	mcx_cmdq_mboxes_sync(sc, &mxm, BUS_DMASYNC_POSTRW);
3644 
3645 	if (error != 0) {
3646 		printf("%s: access reg (%s %x) timeout\n", DEVNAME(sc),
3647 		    (op == MCX_REG_OP_WRITE ? "write" : "read"), reg);
3648 		goto free;
3649 	}
3650 	error = mcx_cmdq_verify(cqe);
3651 	if (error != 0) {
3652 		printf("%s: access reg (%s %x) reply corrupt\n",
3653 		    (op == MCX_REG_OP_WRITE ? "write" : "read"), DEVNAME(sc),
3654 		    reg);
3655 		goto free;
3656 	}
3657 
3658 	out = mcx_cmdq_out(cqe);
3659 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
3660 		printf("%s: access reg (%s %x) failed (%x, %.6x)\n",
3661 		    DEVNAME(sc), (op == MCX_REG_OP_WRITE ? "write" : "read"),
3662 		    reg, out->cmd_status, be32toh(out->cmd_syndrome));
3663 		error = -1;
3664 		goto free;
3665 	}
3666 
3667 	mcx_cmdq_mboxes_copyout(&mxm, nmb, data, len);
3668 free:
3669 	mcx_dmamem_free(sc, &mxm);
3670 
3671 	return (error);
3672 }
3673 
3674 static int
3675 mcx_set_issi(struct mcx_softc *sc, struct mcx_cmdq_entry *cqe,
3676     unsigned int slot)
3677 {
3678 	struct mcx_cmd_set_issi_in *in;
3679 	struct mcx_cmd_set_issi_out *out;
3680 	uint8_t status;
3681 
3682 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
3683 
3684 	in = mcx_cmdq_in(cqe);
3685 	in->cmd_opcode = htobe16(MCX_CMD_SET_ISSI);
3686 	in->cmd_op_mod = htobe16(0);
3687 	in->cmd_current_issi = htobe16(MCX_ISSI);
3688 
3689 	mcx_cmdq_post(sc, cqe, slot);
3690 	if (mcx_cmdq_poll(sc, cqe, 1000) != 0)
3691 		return (-1);
3692 	if (mcx_cmdq_verify(cqe) != 0)
3693 		return (-1);
3694 
3695 	status = cqe->cq_output_data[0];
3696 	if (status != MCX_CQ_STATUS_OK)
3697 		return (-1);
3698 
3699 	return (0);
3700 }
3701 
3702 static int
3703 mcx_issi(struct mcx_softc *sc)
3704 {
3705 	struct mcx_dmamem mxm;
3706 	struct mcx_cmdq_entry *cqe;
3707 	struct mcx_cmd_query_issi_in *in;
3708 	struct mcx_cmd_query_issi_il_out *out;
3709 	struct mcx_cmd_query_issi_mb_out *mb;
3710 	uint8_t token = mcx_cmdq_token(sc);
3711 	uint8_t status;
3712 	int error;
3713 
3714 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
3715 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out) + sizeof(*mb), token);
3716 
3717 	in = mcx_cmdq_in(cqe);
3718 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_ISSI);
3719 	in->cmd_op_mod = htobe16(0);
3720 
3721 	CTASSERT(sizeof(*mb) <= MCX_CMDQ_MAILBOX_DATASIZE);
3722 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
3723 	    &cqe->cq_output_ptr, token) != 0) {
3724 		printf(", unable to allocate query issi mailbox\n");
3725 		return (-1);
3726 	}
3727 	mcx_cmdq_mboxes_sign(&mxm, 1);
3728 
3729 	mcx_cmdq_post(sc, cqe, 0);
3730 	error = mcx_cmdq_poll(sc, cqe, 1000);
3731 	if (error != 0) {
3732 		printf(", query issi timeout\n");
3733 		goto free;
3734 	}
3735 	error = mcx_cmdq_verify(cqe);
3736 	if (error != 0) {
3737 		printf(", query issi reply corrupt\n");
3738 		goto free;
3739 	}
3740 
3741 	status = cqe->cq_output_data[0];
3742 	switch (status) {
3743 	case MCX_CQ_STATUS_OK:
3744 		break;
3745 	case MCX_CQ_STATUS_BAD_OPCODE:
3746 		/* use ISSI 0 */
3747 		goto free;
3748 	default:
3749 		printf(", query issi failed (%x)\n", status);
3750 		error = -1;
3751 		goto free;
3752 	}
3753 
3754 	out = mcx_cmdq_out(cqe);
3755 	if (out->cmd_current_issi == htobe16(MCX_ISSI)) {
3756 		/* use ISSI 1 */
3757 		goto free;
3758 	}
3759 
3760 	/* don't need to read cqe anymore, can be used for SET ISSI */
3761 
3762 	mb = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
3763 	CTASSERT(MCX_ISSI < NBBY);
3764 	 /* XXX math is hard */
3765 	if (!ISSET(mb->cmd_supported_issi[79], 1 << MCX_ISSI)) {
3766 		/* use ISSI 0 */
3767 		goto free;
3768 	}
3769 
3770 	if (mcx_set_issi(sc, cqe, 0) != 0) {
3771 		/* ignore the error, just use ISSI 0 */
3772 	} else {
3773 		/* use ISSI 1 */
3774 	}
3775 
3776 free:
3777 	mcx_cq_mboxes_free(sc, &mxm);
3778 	return (error);
3779 }
3780 
3781 static int
3782 mcx_query_pages(struct mcx_softc *sc, uint16_t type,
3783     int32_t *npages, uint16_t *func_id)
3784 {
3785 	struct mcx_cmdq_entry *cqe;
3786 	struct mcx_cmd_query_pages_in *in;
3787 	struct mcx_cmd_query_pages_out *out;
3788 
3789 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
3790 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
3791 
3792 	in = mcx_cmdq_in(cqe);
3793 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_PAGES);
3794 	in->cmd_op_mod = type;
3795 
3796 	mcx_cmdq_post(sc, cqe, 0);
3797 	if (mcx_cmdq_poll(sc, cqe, 1000) != 0) {
3798 		printf(", query pages timeout\n");
3799 		return (-1);
3800 	}
3801 	if (mcx_cmdq_verify(cqe) != 0) {
3802 		printf(", query pages reply corrupt\n");
3803 		return (-1);
3804 	}
3805 
3806 	out = mcx_cmdq_out(cqe);
3807 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
3808 		printf(", query pages failed (%x)\n", out->cmd_status);
3809 		return (-1);
3810 	}
3811 
3812 	*func_id = out->cmd_func_id;
3813 	*npages = be32dec(&out->cmd_num_pages);
3814 
3815 	return (0);
3816 }
3817 
3818 struct bus_dma_iter {
3819 	bus_dmamap_t		i_map;
3820 	bus_size_t		i_offset;
3821 	unsigned int		i_index;
3822 };
3823 
3824 static void
3825 bus_dma_iter_init(struct bus_dma_iter *i, bus_dmamap_t map)
3826 {
3827 	i->i_map = map;
3828 	i->i_offset = 0;
3829 	i->i_index = 0;
3830 }
3831 
3832 static bus_addr_t
3833 bus_dma_iter_addr(struct bus_dma_iter *i)
3834 {
3835 	return (i->i_map->dm_segs[i->i_index].ds_addr + i->i_offset);
3836 }
3837 
3838 static void
3839 bus_dma_iter_add(struct bus_dma_iter *i, bus_size_t size)
3840 {
3841 	bus_dma_segment_t *seg = i->i_map->dm_segs + i->i_index;
3842 	bus_size_t diff;
3843 
3844 	do {
3845 		diff = seg->ds_len - i->i_offset;
3846 		if (size < diff)
3847 			break;
3848 
3849 		size -= diff;
3850 
3851 		seg++;
3852 
3853 		i->i_offset = 0;
3854 		i->i_index++;
3855 	} while (size > 0);
3856 
3857 	i->i_offset += size;
3858 }
3859 
3860 static int
3861 mcx_add_pages(struct mcx_softc *sc, struct mcx_hwmem *mhm, uint16_t func_id)
3862 {
3863 	struct mcx_dmamem mxm;
3864 	struct mcx_cmdq_entry *cqe;
3865 	struct mcx_cmd_manage_pages_in *in;
3866 	struct mcx_cmd_manage_pages_out *out;
3867 	unsigned int paslen, nmb, i, j, npages;
3868 	struct bus_dma_iter iter;
3869 	uint64_t *pas;
3870 	uint8_t status;
3871 	uint8_t token = mcx_cmdq_token(sc);
3872 	int error;
3873 
3874 	npages = mhm->mhm_npages;
3875 
3876 	paslen = sizeof(*pas) * npages;
3877 	nmb = howmany(paslen, MCX_CMDQ_MAILBOX_DATASIZE);
3878 
3879 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
3880 	mcx_cmdq_init(sc, cqe, sizeof(*in) + paslen, sizeof(*out), token);
3881 
3882 	in = mcx_cmdq_in(cqe);
3883 	in->cmd_opcode = htobe16(MCX_CMD_MANAGE_PAGES);
3884 	in->cmd_op_mod = htobe16(MCX_CMD_MANAGE_PAGES_ALLOC_SUCCESS);
3885 	in->cmd_func_id = func_id;
3886 	be32enc(&in->cmd_input_num_entries, npages);
3887 
3888 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, nmb,
3889 	    &cqe->cq_input_ptr, token) != 0) {
3890 		printf(", unable to allocate manage pages mailboxen\n");
3891 		return (-1);
3892 	}
3893 
3894 	bus_dma_iter_init(&iter, mhm->mhm_map);
3895 	for (i = 0; i < nmb; i++) {
3896 		unsigned int lim;
3897 
3898 		pas = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, i));
3899 		lim = uimin(MCX_CMDQ_MAILBOX_DATASIZE / sizeof(*pas), npages);
3900 
3901 		for (j = 0; j < lim; j++) {
3902 			be64enc(&pas[j], bus_dma_iter_addr(&iter));
3903 			bus_dma_iter_add(&iter, MCX_PAGE_SIZE);
3904 		}
3905 
3906 		npages -= lim;
3907 	}
3908 
3909 	mcx_cmdq_mboxes_sign(&mxm, nmb);
3910 
3911 	mcx_cmdq_post(sc, cqe, 0);
3912 	error = mcx_cmdq_poll(sc, cqe, 1000);
3913 	if (error != 0) {
3914 		printf(", manage pages timeout\n");
3915 		goto free;
3916 	}
3917 	error = mcx_cmdq_verify(cqe);
3918 	if (error != 0) {
3919 		printf(", manage pages reply corrupt\n");
3920 		goto free;
3921 	}
3922 
3923 	status = cqe->cq_output_data[0];
3924 	if (status != MCX_CQ_STATUS_OK) {
3925 		printf(", manage pages failed (%x)\n", status);
3926 		error = -1;
3927 		goto free;
3928 	}
3929 
3930 free:
3931 	mcx_dmamem_free(sc, &mxm);
3932 
3933 	return (error);
3934 }
3935 
3936 static int
3937 mcx_pages(struct mcx_softc *sc, struct mcx_hwmem *mhm, uint16_t type)
3938 {
3939 	int32_t npages;
3940 	uint16_t func_id;
3941 
3942 	if (mcx_query_pages(sc, type, &npages, &func_id) != 0) {
3943 		/* error printed by mcx_query_pages */
3944 		return (-1);
3945 	}
3946 
3947 	if (npages < 1)
3948 		return (0);
3949 
3950 	if (mcx_hwmem_alloc(sc, mhm, npages) != 0) {
3951 		printf(", unable to allocate hwmem\n");
3952 		return (-1);
3953 	}
3954 
3955 	if (mcx_add_pages(sc, mhm, func_id) != 0) {
3956 		printf(", unable to add hwmem\n");
3957 		goto free;
3958 	}
3959 
3960 	return (0);
3961 
3962 free:
3963 	mcx_hwmem_free(sc, mhm);
3964 
3965 	return (-1);
3966 }
3967 
3968 static int
3969 mcx_hca_max_caps(struct mcx_softc *sc)
3970 {
3971 	struct mcx_dmamem mxm;
3972 	struct mcx_cmdq_entry *cqe;
3973 	struct mcx_cmd_query_hca_cap_in *in;
3974 	struct mcx_cmd_query_hca_cap_out *out;
3975 	struct mcx_cmdq_mailbox *mb;
3976 	struct mcx_cap_device *hca;
3977 	uint8_t status;
3978 	uint8_t token = mcx_cmdq_token(sc);
3979 	int error;
3980 
3981 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
3982 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out) + MCX_HCA_CAP_LEN,
3983 	    token);
3984 
3985 	in = mcx_cmdq_in(cqe);
3986 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_HCA_CAP);
3987 	in->cmd_op_mod = htobe16(MCX_CMD_QUERY_HCA_CAP_MAX |
3988 	    MCX_CMD_QUERY_HCA_CAP_DEVICE);
3989 
3990 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, MCX_HCA_CAP_NMAILBOXES,
3991 	    &cqe->cq_output_ptr, token) != 0) {
3992 		printf(", unable to allocate query hca caps mailboxen\n");
3993 		return (-1);
3994 	}
3995 	mcx_cmdq_mboxes_sign(&mxm, MCX_HCA_CAP_NMAILBOXES);
3996 	mcx_cmdq_mboxes_sync(sc, &mxm, BUS_DMASYNC_PRERW);
3997 
3998 	mcx_cmdq_post(sc, cqe, 0);
3999 	error = mcx_cmdq_poll(sc, cqe, 1000);
4000 	mcx_cmdq_mboxes_sync(sc, &mxm, BUS_DMASYNC_POSTRW);
4001 
4002 	if (error != 0) {
4003 		printf(", query hca caps timeout\n");
4004 		goto free;
4005 	}
4006 	error = mcx_cmdq_verify(cqe);
4007 	if (error != 0) {
4008 		printf(", query hca caps reply corrupt\n");
4009 		goto free;
4010 	}
4011 
4012 	status = cqe->cq_output_data[0];
4013 	if (status != MCX_CQ_STATUS_OK) {
4014 		printf(", query hca caps failed (%x)\n", status);
4015 		error = -1;
4016 		goto free;
4017 	}
4018 
4019 	mb = mcx_cq_mbox(&mxm, 0);
4020 	hca = mcx_cq_mbox_data(mb);
4021 
4022 	if ((hca->port_type & MCX_CAP_DEVICE_PORT_TYPE)
4023 	    != MCX_CAP_DEVICE_PORT_TYPE_ETH) {
4024 		printf(", not in ethernet mode\n");
4025 		error = -1;
4026 		goto free;
4027 	}
4028 	if (hca->log_pg_sz > PAGE_SHIFT) {
4029 		printf(", minimum system page shift %u is too large\n",
4030 		    hca->log_pg_sz);
4031 		error = -1;
4032 		goto free;
4033 	}
4034 	/*
4035 	 * blueflame register is split into two buffers, and we must alternate
4036 	 * between the two of them.
4037 	 */
4038 	sc->sc_bf_size = (1 << hca->log_bf_reg_size) / 2;
4039 	sc->sc_max_rqt_size = (1 << hca->log_max_rqt_size);
4040 
4041 	if (hca->local_ca_ack_delay & MCX_CAP_DEVICE_MCAM_REG)
4042 		sc->sc_mcam_reg = 1;
4043 
4044 	sc->sc_mhz = be32dec(&hca->device_frequency_mhz);
4045 	sc->sc_khz = be32dec(&hca->device_frequency_khz);
4046 
4047 free:
4048 	mcx_dmamem_free(sc, &mxm);
4049 
4050 	return (error);
4051 }
4052 
4053 static int
4054 mcx_hca_set_caps(struct mcx_softc *sc)
4055 {
4056 	struct mcx_dmamem mxm;
4057 	struct mcx_cmdq_entry *cqe;
4058 	struct mcx_cmd_query_hca_cap_in *in;
4059 	struct mcx_cmd_query_hca_cap_out *out;
4060 	struct mcx_cmdq_mailbox *mb;
4061 	struct mcx_cap_device *hca;
4062 	uint8_t status;
4063 	uint8_t token = mcx_cmdq_token(sc);
4064 	int error;
4065 
4066 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4067 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out) + MCX_HCA_CAP_LEN,
4068 	    token);
4069 
4070 	in = mcx_cmdq_in(cqe);
4071 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_HCA_CAP);
4072 	in->cmd_op_mod = htobe16(MCX_CMD_QUERY_HCA_CAP_CURRENT |
4073 	    MCX_CMD_QUERY_HCA_CAP_DEVICE);
4074 
4075 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, MCX_HCA_CAP_NMAILBOXES,
4076 	    &cqe->cq_output_ptr, token) != 0) {
4077 		printf(", unable to allocate manage pages mailboxen\n");
4078 		return (-1);
4079 	}
4080 	mcx_cmdq_mboxes_sign(&mxm, MCX_HCA_CAP_NMAILBOXES);
4081 	mcx_cmdq_mboxes_sync(sc, &mxm, BUS_DMASYNC_PRERW);
4082 
4083 	mcx_cmdq_post(sc, cqe, 0);
4084 	error = mcx_cmdq_poll(sc, cqe, 1000);
4085 	mcx_cmdq_mboxes_sync(sc, &mxm, BUS_DMASYNC_POSTRW);
4086 
4087 	if (error != 0) {
4088 		printf(", query hca caps timeout\n");
4089 		goto free;
4090 	}
4091 	error = mcx_cmdq_verify(cqe);
4092 	if (error != 0) {
4093 		printf(", query hca caps reply corrupt\n");
4094 		goto free;
4095 	}
4096 
4097 	status = cqe->cq_output_data[0];
4098 	if (status != MCX_CQ_STATUS_OK) {
4099 		printf(", query hca caps failed (%x)\n", status);
4100 		error = -1;
4101 		goto free;
4102 	}
4103 
4104 	mb = mcx_cq_mbox(&mxm, 0);
4105 	hca = mcx_cq_mbox_data(mb);
4106 
4107 	hca->log_pg_sz = PAGE_SHIFT;
4108 
4109 free:
4110 	mcx_dmamem_free(sc, &mxm);
4111 
4112 	return (error);
4113 }
4114 
4115 
4116 static int
4117 mcx_init_hca(struct mcx_softc *sc)
4118 {
4119 	struct mcx_cmdq_entry *cqe;
4120 	struct mcx_cmd_init_hca_in *in;
4121 	struct mcx_cmd_init_hca_out *out;
4122 	int error;
4123 	uint8_t status;
4124 
4125 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4126 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
4127 
4128 	in = mcx_cmdq_in(cqe);
4129 	in->cmd_opcode = htobe16(MCX_CMD_INIT_HCA);
4130 	in->cmd_op_mod = htobe16(0);
4131 
4132 	mcx_cmdq_post(sc, cqe, 0);
4133 
4134 	error = mcx_cmdq_poll(sc, cqe, 1000);
4135 	if (error != 0) {
4136 		printf(", hca init timeout\n");
4137 		return (-1);
4138 	}
4139 	if (mcx_cmdq_verify(cqe) != 0) {
4140 		printf(", hca init command corrupt\n");
4141 		return (-1);
4142 	}
4143 
4144 	status = cqe->cq_output_data[0];
4145 	if (status != MCX_CQ_STATUS_OK) {
4146 		printf(", hca init failed (%x)\n", status);
4147 		return (-1);
4148 	}
4149 
4150 	return (0);
4151 }
4152 
4153 static int
4154 mcx_set_driver_version(struct mcx_softc *sc)
4155 {
4156 	struct mcx_dmamem mxm;
4157 	struct mcx_cmdq_entry *cqe;
4158 	struct mcx_cmd_set_driver_version_in *in;
4159 	struct mcx_cmd_set_driver_version_out *out;
4160 	int error;
4161 	int token;
4162 	uint8_t status;
4163 
4164 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4165 	token = mcx_cmdq_token(sc);
4166 	mcx_cmdq_init(sc, cqe, sizeof(*in) +
4167 	    sizeof(struct mcx_cmd_set_driver_version), sizeof(*out), token);
4168 
4169 	in = mcx_cmdq_in(cqe);
4170 	in->cmd_opcode = htobe16(MCX_CMD_SET_DRIVER_VERSION);
4171 	in->cmd_op_mod = htobe16(0);
4172 
4173 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
4174 	    &cqe->cq_input_ptr, token) != 0) {
4175 		printf(", unable to allocate set driver version mailboxen\n");
4176 		return (-1);
4177 	}
4178 	strlcpy(mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)),
4179 	    "OpenBSD,mcx,1.000.000000", MCX_CMDQ_MAILBOX_DATASIZE);
4180 
4181 	mcx_cmdq_mboxes_sign(&mxm, 1);
4182 	mcx_cmdq_post(sc, cqe, 0);
4183 
4184 	error = mcx_cmdq_poll(sc, cqe, 1000);
4185 	if (error != 0) {
4186 		printf(", set driver version timeout\n");
4187 		goto free;
4188 	}
4189 	if (mcx_cmdq_verify(cqe) != 0) {
4190 		printf(", set driver version command corrupt\n");
4191 		goto free;
4192 	}
4193 
4194 	status = cqe->cq_output_data[0];
4195 	if (status != MCX_CQ_STATUS_OK) {
4196 		printf(", set driver version failed (%x)\n", status);
4197 		error = -1;
4198 		goto free;
4199 	}
4200 
4201 free:
4202 	mcx_dmamem_free(sc, &mxm);
4203 
4204 	return (error);
4205 }
4206 
4207 static int
4208 mcx_iff(struct mcx_softc *sc)
4209 {
4210 	struct ifnet *ifp = &sc->sc_ec.ec_if;
4211 	struct mcx_dmamem mxm;
4212 	struct mcx_cmdq_entry *cqe;
4213 	struct mcx_cmd_modify_nic_vport_context_in *in;
4214 	struct mcx_cmd_modify_nic_vport_context_out *out;
4215 	struct mcx_nic_vport_ctx *ctx;
4216 	int error;
4217 	int token;
4218 	int insize;
4219 	uint32_t dest;
4220 
4221 	dest = MCX_FLOW_CONTEXT_DEST_TYPE_TABLE |
4222 	    sc->sc_rss_flow_table_id;
4223 
4224 	/* enable or disable the promisc flow */
4225 	if (ISSET(ifp->if_flags, IFF_PROMISC)) {
4226 		if (sc->sc_promisc_flow_enabled == 0) {
4227 			mcx_set_flow_table_entry_mac(sc,
4228 			    MCX_FLOW_GROUP_PROMISC, 0, NULL, dest);
4229 			sc->sc_promisc_flow_enabled = 1;
4230 		}
4231 	} else if (sc->sc_promisc_flow_enabled != 0) {
4232 		mcx_delete_flow_table_entry(sc, MCX_FLOW_GROUP_PROMISC, 0);
4233 		sc->sc_promisc_flow_enabled = 0;
4234 	}
4235 
4236 	/* enable or disable the all-multicast flow */
4237 	if (ISSET(ifp->if_flags, IFF_ALLMULTI)) {
4238 		if (sc->sc_allmulti_flow_enabled == 0) {
4239 			uint8_t mcast[ETHER_ADDR_LEN];
4240 
4241 			memset(mcast, 0, sizeof(mcast));
4242 			mcast[0] = 0x01;
4243 			mcx_set_flow_table_entry_mac(sc,
4244 			    MCX_FLOW_GROUP_ALLMULTI, 0, mcast, dest);
4245 			sc->sc_allmulti_flow_enabled = 1;
4246 		}
4247 	} else if (sc->sc_allmulti_flow_enabled != 0) {
4248 		mcx_delete_flow_table_entry(sc, MCX_FLOW_GROUP_ALLMULTI, 0);
4249 		sc->sc_allmulti_flow_enabled = 0;
4250 	}
4251 
4252 	insize = sizeof(struct mcx_nic_vport_ctx) + 240;
4253 
4254 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4255 	token = mcx_cmdq_token(sc);
4256 	mcx_cmdq_init(sc, cqe, sizeof(*in) + insize, sizeof(*out), token);
4257 
4258 	in = mcx_cmdq_in(cqe);
4259 	in->cmd_opcode = htobe16(MCX_CMD_MODIFY_NIC_VPORT_CONTEXT);
4260 	in->cmd_op_mod = htobe16(0);
4261 	in->cmd_field_select = htobe32(
4262 	    MCX_CMD_MODIFY_NIC_VPORT_CONTEXT_FIELD_PROMISC |
4263 	    MCX_CMD_MODIFY_NIC_VPORT_CONTEXT_FIELD_MTU);
4264 
4265 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1, &cqe->cq_input_ptr, token) != 0) {
4266 		printf(", unable to allocate modify "
4267 		    "nic vport context mailboxen\n");
4268 		return (-1);
4269 	}
4270 	ctx = (struct mcx_nic_vport_ctx *)
4271 	    (((char *)mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0))) + 240);
4272 	ctx->vp_mtu = htobe32(sc->sc_hardmtu);
4273 	/*
4274          * always leave promisc-all enabled on the vport since we
4275          * can't give it a vlan list, and we're already doing multicast
4276          * filtering in the flow table.
4277 	 */
4278 	ctx->vp_flags = htobe16(MCX_NIC_VPORT_CTX_PROMISC_ALL);
4279 
4280 	mcx_cmdq_mboxes_sign(&mxm, 1);
4281 	mcx_cmdq_post(sc, cqe, 0);
4282 
4283 	error = mcx_cmdq_poll(sc, cqe, 1000);
4284 	if (error != 0) {
4285 		printf(", modify nic vport context timeout\n");
4286 		goto free;
4287 	}
4288 	if (mcx_cmdq_verify(cqe) != 0) {
4289 		printf(", modify nic vport context command corrupt\n");
4290 		goto free;
4291 	}
4292 
4293 	out = mcx_cmdq_out(cqe);
4294 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4295 		printf(", modify nic vport context failed (%x, %x)\n",
4296 		    out->cmd_status, be32toh(out->cmd_syndrome));
4297 		error = -1;
4298 		goto free;
4299 	}
4300 
4301 free:
4302 	mcx_dmamem_free(sc, &mxm);
4303 
4304 	return (error);
4305 }
4306 
4307 static int
4308 mcx_alloc_uar(struct mcx_softc *sc, int *uar)
4309 {
4310 	struct mcx_cmdq_entry *cqe;
4311 	struct mcx_cmd_alloc_uar_in *in;
4312 	struct mcx_cmd_alloc_uar_out *out;
4313 	int error;
4314 
4315 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4316 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
4317 
4318 	in = mcx_cmdq_in(cqe);
4319 	in->cmd_opcode = htobe16(MCX_CMD_ALLOC_UAR);
4320 	in->cmd_op_mod = htobe16(0);
4321 
4322 	mcx_cmdq_post(sc, cqe, 0);
4323 
4324 	error = mcx_cmdq_poll(sc, cqe, 1000);
4325 	if (error != 0) {
4326 		printf(", alloc uar timeout\n");
4327 		return (-1);
4328 	}
4329 	if (mcx_cmdq_verify(cqe) != 0) {
4330 		printf(", alloc uar command corrupt\n");
4331 		return (-1);
4332 	}
4333 
4334 	out = mcx_cmdq_out(cqe);
4335 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4336 		printf(", alloc uar failed (%x)\n", out->cmd_status);
4337 		return (-1);
4338 	}
4339 
4340 	*uar = mcx_get_id(out->cmd_uar);
4341 	return (0);
4342 }
4343 
4344 static int
4345 mcx_create_eq(struct mcx_softc *sc, struct mcx_eq *eq, int uar,
4346     uint64_t events, int vector)
4347 {
4348 	struct mcx_cmdq_entry *cqe;
4349 	struct mcx_dmamem mxm;
4350 	struct mcx_cmd_create_eq_in *in;
4351 	struct mcx_cmd_create_eq_mb_in *mbin;
4352 	struct mcx_cmd_create_eq_out *out;
4353 	struct mcx_eq_entry *eqe;
4354 	int error;
4355 	uint64_t *pas;
4356 	int insize, npages, paslen, i, token;
4357 
4358 	eq->eq_cons = 0;
4359 
4360 	npages = howmany((1 << MCX_LOG_EQ_SIZE) * sizeof(struct mcx_eq_entry),
4361 	    MCX_PAGE_SIZE);
4362 	paslen = npages * sizeof(*pas);
4363 	insize = sizeof(struct mcx_cmd_create_eq_mb_in) + paslen;
4364 
4365 	if (mcx_dmamem_alloc(sc, &eq->eq_mem, npages * MCX_PAGE_SIZE,
4366 	    MCX_PAGE_SIZE) != 0) {
4367 		printf(", unable to allocate event queue memory\n");
4368 		return (-1);
4369 	}
4370 
4371 	eqe = (struct mcx_eq_entry *)MCX_DMA_KVA(&eq->eq_mem);
4372 	for (i = 0; i < (1 << MCX_LOG_EQ_SIZE); i++) {
4373 		eqe[i].eq_owner = MCX_EQ_ENTRY_OWNER_INIT;
4374 	}
4375 
4376 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4377 	token = mcx_cmdq_token(sc);
4378 	mcx_cmdq_init(sc, cqe, sizeof(*in) + insize, sizeof(*out), token);
4379 
4380 	in = mcx_cmdq_in(cqe);
4381 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_EQ);
4382 	in->cmd_op_mod = htobe16(0);
4383 
4384 	if (mcx_cmdq_mboxes_alloc(sc, &mxm,
4385 	    howmany(insize, MCX_CMDQ_MAILBOX_DATASIZE),
4386 	    &cqe->cq_input_ptr, token) != 0) {
4387 		printf(", unable to allocate create eq mailboxen\n");
4388 		goto free_eq;
4389 	}
4390 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
4391 	mbin->cmd_eq_ctx.eq_uar_size = htobe32(
4392 	    (MCX_LOG_EQ_SIZE << MCX_EQ_CTX_LOG_EQ_SIZE_SHIFT) | uar);
4393 	mbin->cmd_eq_ctx.eq_intr = vector;
4394 	mbin->cmd_event_bitmask = htobe64(events);
4395 
4396 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&eq->eq_mem),
4397 	    0, MCX_DMA_LEN(&eq->eq_mem), BUS_DMASYNC_PREREAD);
4398 
4399 	/* physical addresses follow the mailbox in data */
4400 	mcx_cmdq_mboxes_pas(&mxm, sizeof(*mbin), npages, &eq->eq_mem);
4401 	mcx_cmdq_mboxes_sign(&mxm, howmany(insize, MCX_CMDQ_MAILBOX_DATASIZE));
4402 	mcx_cmdq_post(sc, cqe, 0);
4403 
4404 	error = mcx_cmdq_poll(sc, cqe, 1000);
4405 	if (error != 0) {
4406 		printf(", create eq timeout\n");
4407 		goto free_mxm;
4408 	}
4409 	if (mcx_cmdq_verify(cqe) != 0) {
4410 		printf(", create eq command corrupt\n");
4411 		goto free_mxm;
4412 	}
4413 
4414 	out = mcx_cmdq_out(cqe);
4415 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4416 		printf(", create eq failed (%x, %x)\n", out->cmd_status,
4417 		    be32toh(out->cmd_syndrome));
4418 		goto free_mxm;
4419 	}
4420 
4421 	eq->eq_n = mcx_get_id(out->cmd_eqn);
4422 
4423 	mcx_dmamem_free(sc, &mxm);
4424 
4425 	mcx_arm_eq(sc, eq, uar);
4426 
4427 	return (0);
4428 
4429 free_mxm:
4430 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&eq->eq_mem),
4431 	    0, MCX_DMA_LEN(&eq->eq_mem), BUS_DMASYNC_POSTREAD);
4432 	mcx_dmamem_free(sc, &mxm);
4433 free_eq:
4434 	mcx_dmamem_free(sc, &eq->eq_mem);
4435 	return (-1);
4436 }
4437 
4438 static int
4439 mcx_alloc_pd(struct mcx_softc *sc)
4440 {
4441 	struct mcx_cmdq_entry *cqe;
4442 	struct mcx_cmd_alloc_pd_in *in;
4443 	struct mcx_cmd_alloc_pd_out *out;
4444 	int error;
4445 
4446 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4447 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
4448 
4449 	in = mcx_cmdq_in(cqe);
4450 	in->cmd_opcode = htobe16(MCX_CMD_ALLOC_PD);
4451 	in->cmd_op_mod = htobe16(0);
4452 
4453 	mcx_cmdq_post(sc, cqe, 0);
4454 
4455 	error = mcx_cmdq_poll(sc, cqe, 1000);
4456 	if (error != 0) {
4457 		printf(", alloc pd timeout\n");
4458 		return (-1);
4459 	}
4460 	if (mcx_cmdq_verify(cqe) != 0) {
4461 		printf(", alloc pd command corrupt\n");
4462 		return (-1);
4463 	}
4464 
4465 	out = mcx_cmdq_out(cqe);
4466 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4467 		printf(", alloc pd failed (%x)\n", out->cmd_status);
4468 		return (-1);
4469 	}
4470 
4471 	sc->sc_pd = mcx_get_id(out->cmd_pd);
4472 	return (0);
4473 }
4474 
4475 static int
4476 mcx_alloc_tdomain(struct mcx_softc *sc)
4477 {
4478 	struct mcx_cmdq_entry *cqe;
4479 	struct mcx_cmd_alloc_td_in *in;
4480 	struct mcx_cmd_alloc_td_out *out;
4481 	int error;
4482 
4483 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4484 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
4485 
4486 	in = mcx_cmdq_in(cqe);
4487 	in->cmd_opcode = htobe16(MCX_CMD_ALLOC_TRANSPORT_DOMAIN);
4488 	in->cmd_op_mod = htobe16(0);
4489 
4490 	mcx_cmdq_post(sc, cqe, 0);
4491 
4492 	error = mcx_cmdq_poll(sc, cqe, 1000);
4493 	if (error != 0) {
4494 		printf(", alloc transport domain timeout\n");
4495 		return (-1);
4496 	}
4497 	if (mcx_cmdq_verify(cqe) != 0) {
4498 		printf(", alloc transport domain command corrupt\n");
4499 		return (-1);
4500 	}
4501 
4502 	out = mcx_cmdq_out(cqe);
4503 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4504 		printf(", alloc transport domain failed (%x)\n",
4505 		    out->cmd_status);
4506 		return (-1);
4507 	}
4508 
4509 	sc->sc_tdomain = mcx_get_id(out->cmd_tdomain);
4510 	return (0);
4511 }
4512 
4513 static int
4514 mcx_query_nic_vport_context(struct mcx_softc *sc, uint8_t *enaddr)
4515 {
4516 	struct mcx_dmamem mxm;
4517 	struct mcx_cmdq_entry *cqe;
4518 	struct mcx_cmd_query_nic_vport_context_in *in;
4519 	struct mcx_cmd_query_nic_vport_context_out *out;
4520 	struct mcx_nic_vport_ctx *ctx;
4521 	uint8_t *addr;
4522 	int error, token, i;
4523 
4524 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4525 	token = mcx_cmdq_token(sc);
4526 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out) + sizeof(*ctx), token);
4527 
4528 	in = mcx_cmdq_in(cqe);
4529 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_NIC_VPORT_CONTEXT);
4530 	in->cmd_op_mod = htobe16(0);
4531 	in->cmd_allowed_list_type = 0;
4532 
4533 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
4534 	    &cqe->cq_output_ptr, token) != 0) {
4535 		printf(", unable to allocate "
4536 		    "query nic vport context mailboxen\n");
4537 		return (-1);
4538 	}
4539 	mcx_cmdq_mboxes_sign(&mxm, 1);
4540 	mcx_cmdq_post(sc, cqe, 0);
4541 
4542 	error = mcx_cmdq_poll(sc, cqe, 1000);
4543 	if (error != 0) {
4544 		printf(", query nic vport context timeout\n");
4545 		goto free;
4546 	}
4547 	if (mcx_cmdq_verify(cqe) != 0) {
4548 		printf(", query nic vport context command corrupt\n");
4549 		goto free;
4550 	}
4551 
4552 	out = mcx_cmdq_out(cqe);
4553 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4554 		printf(", query nic vport context failed (%x, %x)\n",
4555 		    out->cmd_status, be32toh(out->cmd_syndrome));
4556 		error = -1;
4557 		goto free;
4558 	}
4559 
4560 	ctx = (struct mcx_nic_vport_ctx *)
4561 	    mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
4562 	addr = (uint8_t *)&ctx->vp_perm_addr;
4563 	for (i = 0; i < ETHER_ADDR_LEN; i++) {
4564 		enaddr[i] = addr[i + 2];
4565 	}
4566 free:
4567 	mcx_dmamem_free(sc, &mxm);
4568 
4569 	return (error);
4570 }
4571 
4572 static int
4573 mcx_query_special_contexts(struct mcx_softc *sc)
4574 {
4575 	struct mcx_cmdq_entry *cqe;
4576 	struct mcx_cmd_query_special_ctx_in *in;
4577 	struct mcx_cmd_query_special_ctx_out *out;
4578 	int error;
4579 
4580 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4581 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
4582 
4583 	in = mcx_cmdq_in(cqe);
4584 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_SPECIAL_CONTEXTS);
4585 	in->cmd_op_mod = htobe16(0);
4586 
4587 	mcx_cmdq_post(sc, cqe, 0);
4588 
4589 	error = mcx_cmdq_poll(sc, cqe, 1000);
4590 	if (error != 0) {
4591 		printf(", query special contexts timeout\n");
4592 		return (-1);
4593 	}
4594 	if (mcx_cmdq_verify(cqe) != 0) {
4595 		printf(", query special contexts command corrupt\n");
4596 		return (-1);
4597 	}
4598 
4599 	out = mcx_cmdq_out(cqe);
4600 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4601 		printf(", query special contexts failed (%x)\n",
4602 		    out->cmd_status);
4603 		return (-1);
4604 	}
4605 
4606 	sc->sc_lkey = be32toh(out->cmd_resd_lkey);
4607 	return (0);
4608 }
4609 
4610 static int
4611 mcx_set_port_mtu(struct mcx_softc *sc, int mtu)
4612 {
4613 	struct mcx_reg_pmtu pmtu;
4614 	int error;
4615 
4616 	/* read max mtu */
4617 	memset(&pmtu, 0, sizeof(pmtu));
4618 	pmtu.rp_local_port = 1;
4619 	error = mcx_access_hca_reg(sc, MCX_REG_PMTU, MCX_REG_OP_READ, &pmtu,
4620 	    sizeof(pmtu));
4621 	if (error != 0) {
4622 		printf(", unable to get port MTU\n");
4623 		return error;
4624 	}
4625 
4626 	mtu = uimin(mtu, be16toh(pmtu.rp_max_mtu));
4627 	pmtu.rp_admin_mtu = htobe16(mtu);
4628 	error = mcx_access_hca_reg(sc, MCX_REG_PMTU, MCX_REG_OP_WRITE, &pmtu,
4629 	    sizeof(pmtu));
4630 	if (error != 0) {
4631 		printf(", unable to set port MTU\n");
4632 		return error;
4633 	}
4634 
4635 	sc->sc_hardmtu = mtu;
4636 	sc->sc_rxbufsz = roundup(mtu + ETHER_ALIGN, sizeof(long));
4637 	return 0;
4638 }
4639 
4640 static int
4641 mcx_create_cq(struct mcx_softc *sc, struct mcx_cq *cq, int uar, int db, int eqn)
4642 {
4643 	struct mcx_cmdq_entry *cmde;
4644 	struct mcx_cq_entry *cqe;
4645 	struct mcx_dmamem mxm;
4646 	struct mcx_cmd_create_cq_in *in;
4647 	struct mcx_cmd_create_cq_mb_in *mbin;
4648 	struct mcx_cmd_create_cq_out *out;
4649 	int error;
4650 	uint64_t *pas;
4651 	int insize, npages, paslen, i, token;
4652 
4653 	cq->cq_doorbell = MCX_CQ_DOORBELL_BASE + (MCX_CQ_DOORBELL_STRIDE * db);
4654 
4655 	npages = howmany((1 << MCX_LOG_CQ_SIZE) * sizeof(struct mcx_cq_entry),
4656 	    MCX_PAGE_SIZE);
4657 	paslen = npages * sizeof(*pas);
4658 	insize = sizeof(struct mcx_cmd_create_cq_mb_in) + paslen;
4659 
4660 	if (mcx_dmamem_alloc(sc, &cq->cq_mem, npages * MCX_PAGE_SIZE,
4661 	    MCX_PAGE_SIZE) != 0) {
4662 		printf("%s: unable to allocate completion queue memory\n",
4663 		    DEVNAME(sc));
4664 		return (-1);
4665 	}
4666 	cqe = MCX_DMA_KVA(&cq->cq_mem);
4667 	for (i = 0; i < (1 << MCX_LOG_CQ_SIZE); i++) {
4668 		cqe[i].cq_opcode_owner = MCX_CQ_ENTRY_FLAG_OWNER;
4669 	}
4670 
4671 	cmde = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4672 	token = mcx_cmdq_token(sc);
4673 	mcx_cmdq_init(sc, cmde, sizeof(*in) + insize, sizeof(*out), token);
4674 
4675 	in = mcx_cmdq_in(cmde);
4676 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_CQ);
4677 	in->cmd_op_mod = htobe16(0);
4678 
4679 	if (mcx_cmdq_mboxes_alloc(sc, &mxm,
4680 	    howmany(insize, MCX_CMDQ_MAILBOX_DATASIZE),
4681 	    &cmde->cq_input_ptr, token) != 0) {
4682 		printf("%s: unable to allocate create cq mailboxen\n",
4683 		    DEVNAME(sc));
4684 		goto free_cq;
4685 	}
4686 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
4687 	mbin->cmd_cq_ctx.cq_uar_size = htobe32(
4688 	    (MCX_LOG_CQ_SIZE << MCX_CQ_CTX_LOG_CQ_SIZE_SHIFT) | uar);
4689 	mbin->cmd_cq_ctx.cq_eqn = htobe32(eqn);
4690 	mbin->cmd_cq_ctx.cq_period_max_count = htobe32(
4691 	    (MCX_CQ_MOD_PERIOD << MCX_CQ_CTX_PERIOD_SHIFT) |
4692 	    MCX_CQ_MOD_COUNTER);
4693 	mbin->cmd_cq_ctx.cq_doorbell = htobe64(
4694 	    MCX_DMA_DVA(&sc->sc_doorbell_mem) + cq->cq_doorbell);
4695 
4696 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&cq->cq_mem),
4697 	    0, MCX_DMA_LEN(&cq->cq_mem), BUS_DMASYNC_PREREAD);
4698 
4699 	/* physical addresses follow the mailbox in data */
4700 	mcx_cmdq_mboxes_pas(&mxm, sizeof(*mbin), npages, &cq->cq_mem);
4701 	mcx_cmdq_post(sc, cmde, 0);
4702 
4703 	error = mcx_cmdq_poll(sc, cmde, 1000);
4704 	if (error != 0) {
4705 		printf("%s: create cq timeout\n", DEVNAME(sc));
4706 		goto free_mxm;
4707 	}
4708 	if (mcx_cmdq_verify(cmde) != 0) {
4709 		printf("%s: create cq command corrupt\n", DEVNAME(sc));
4710 		goto free_mxm;
4711 	}
4712 
4713 	out = mcx_cmdq_out(cmde);
4714 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4715 		printf("%s: create cq failed (%x, %x)\n", DEVNAME(sc),
4716 		    out->cmd_status, be32toh(out->cmd_syndrome));
4717 		goto free_mxm;
4718 	}
4719 
4720 	cq->cq_n = mcx_get_id(out->cmd_cqn);
4721 	cq->cq_cons = 0;
4722 	cq->cq_count = 0;
4723 
4724 	mcx_dmamem_free(sc, &mxm);
4725 
4726 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_doorbell_mem),
4727 	    cq->cq_doorbell, sizeof(struct mcx_cq_doorbell),
4728 	    BUS_DMASYNC_PREWRITE);
4729 
4730 	mcx_arm_cq(sc, cq, uar);
4731 
4732 	return (0);
4733 
4734 free_mxm:
4735 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&cq->cq_mem),
4736 	    0, MCX_DMA_LEN(&cq->cq_mem), BUS_DMASYNC_POSTREAD);
4737 	mcx_dmamem_free(sc, &mxm);
4738 free_cq:
4739 	mcx_dmamem_free(sc, &cq->cq_mem);
4740 	return (-1);
4741 }
4742 
4743 static int
4744 mcx_destroy_cq(struct mcx_softc *sc, struct mcx_cq *cq)
4745 {
4746 	struct mcx_cmdq_entry *cqe;
4747 	struct mcx_cmd_destroy_cq_in *in;
4748 	struct mcx_cmd_destroy_cq_out *out;
4749 	int error;
4750 	int token;
4751 
4752 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4753 	token = mcx_cmdq_token(sc);
4754 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), token);
4755 
4756 	in = mcx_cmdq_in(cqe);
4757 	in->cmd_opcode = htobe16(MCX_CMD_DESTROY_CQ);
4758 	in->cmd_op_mod = htobe16(0);
4759 	in->cmd_cqn = htobe32(cq->cq_n);
4760 
4761 	mcx_cmdq_post(sc, cqe, 0);
4762 	error = mcx_cmdq_poll(sc, cqe, 1000);
4763 	if (error != 0) {
4764 		printf("%s: destroy cq timeout\n", DEVNAME(sc));
4765 		return error;
4766 	}
4767 	if (mcx_cmdq_verify(cqe) != 0) {
4768 		printf("%s: destroy cq command corrupt\n", DEVNAME(sc));
4769 		return error;
4770 	}
4771 
4772 	out = mcx_cmdq_out(cqe);
4773 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4774 		printf("%s: destroy cq failed (%x, %x)\n", DEVNAME(sc),
4775 		    out->cmd_status, be32toh(out->cmd_syndrome));
4776 		return -1;
4777 	}
4778 
4779 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_doorbell_mem),
4780 	    cq->cq_doorbell, sizeof(struct mcx_cq_doorbell),
4781 	    BUS_DMASYNC_POSTWRITE);
4782 
4783 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&cq->cq_mem),
4784 	    0, MCX_DMA_LEN(&cq->cq_mem), BUS_DMASYNC_POSTREAD);
4785 	mcx_dmamem_free(sc, &cq->cq_mem);
4786 
4787 	cq->cq_n = 0;
4788 	cq->cq_cons = 0;
4789 	cq->cq_count = 0;
4790 	return 0;
4791 }
4792 
4793 static int
4794 mcx_create_rq(struct mcx_softc *sc, struct mcx_rx *rx, int db, int cqn)
4795 {
4796 	struct mcx_cmdq_entry *cqe;
4797 	struct mcx_dmamem mxm;
4798 	struct mcx_cmd_create_rq_in *in;
4799 	struct mcx_cmd_create_rq_out *out;
4800 	struct mcx_rq_ctx *mbin;
4801 	int error;
4802 	uint64_t *pas;
4803 	uint32_t rq_flags;
4804 	int insize, npages, paslen, token;
4805 
4806 	rx->rx_doorbell = MCX_WQ_DOORBELL_BASE +
4807 	    (db * MCX_WQ_DOORBELL_STRIDE);
4808 
4809 	npages = howmany((1 << MCX_LOG_RQ_SIZE) * sizeof(struct mcx_rq_entry),
4810 	    MCX_PAGE_SIZE);
4811 	paslen = npages * sizeof(*pas);
4812 	insize = 0x10 + sizeof(struct mcx_rq_ctx) + paslen;
4813 
4814 	if (mcx_dmamem_alloc(sc, &rx->rx_rq_mem, npages * MCX_PAGE_SIZE,
4815 	    MCX_PAGE_SIZE) != 0) {
4816 		printf("%s: unable to allocate receive queue memory\n",
4817 		    DEVNAME(sc));
4818 		return (-1);
4819 	}
4820 
4821 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4822 	token = mcx_cmdq_token(sc);
4823 	mcx_cmdq_init(sc, cqe, sizeof(*in) + insize, sizeof(*out), token);
4824 
4825 	in = mcx_cmdq_in(cqe);
4826 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_RQ);
4827 	in->cmd_op_mod = htobe16(0);
4828 
4829 	if (mcx_cmdq_mboxes_alloc(sc, &mxm,
4830 	    howmany(insize, MCX_CMDQ_MAILBOX_DATASIZE),
4831 	    &cqe->cq_input_ptr, token) != 0) {
4832 		printf("%s: unable to allocate create rq mailboxen\n",
4833 		    DEVNAME(sc));
4834 		goto free_rq;
4835 	}
4836 	mbin = (struct mcx_rq_ctx *)
4837 	    (((char *)mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0))) + 0x10);
4838 	rq_flags = MCX_RQ_CTX_RLKEY;
4839 	mbin->rq_flags = htobe32(rq_flags);
4840 	mbin->rq_cqn = htobe32(cqn);
4841 	mbin->rq_wq.wq_type = MCX_WQ_CTX_TYPE_CYCLIC;
4842 	mbin->rq_wq.wq_pd = htobe32(sc->sc_pd);
4843 	mbin->rq_wq.wq_doorbell = htobe64(MCX_DMA_DVA(&sc->sc_doorbell_mem) +
4844 	    rx->rx_doorbell);
4845 	mbin->rq_wq.wq_log_stride = htobe16(4);
4846 	mbin->rq_wq.wq_log_size = MCX_LOG_RQ_SIZE;
4847 
4848 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&rx->rx_rq_mem),
4849 	    0, MCX_DMA_LEN(&rx->rx_rq_mem), BUS_DMASYNC_PREWRITE);
4850 
4851 	/* physical addresses follow the mailbox in data */
4852 	mcx_cmdq_mboxes_pas(&mxm, sizeof(*mbin) + 0x10, npages, &rx->rx_rq_mem);
4853 	mcx_cmdq_post(sc, cqe, 0);
4854 
4855 	error = mcx_cmdq_poll(sc, cqe, 1000);
4856 	if (error != 0) {
4857 		printf("%s: create rq timeout\n", DEVNAME(sc));
4858 		goto free_mxm;
4859 	}
4860 	if (mcx_cmdq_verify(cqe) != 0) {
4861 		printf("%s: create rq command corrupt\n", DEVNAME(sc));
4862 		goto free_mxm;
4863 	}
4864 
4865 	out = mcx_cmdq_out(cqe);
4866 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4867 		printf("%s: create rq failed (%x, %x)\n", DEVNAME(sc),
4868 		    out->cmd_status, be32toh(out->cmd_syndrome));
4869 		goto free_mxm;
4870 	}
4871 
4872 	rx->rx_rqn = mcx_get_id(out->cmd_rqn);
4873 
4874 	mcx_dmamem_free(sc, &mxm);
4875 
4876 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_doorbell_mem),
4877 	    rx->rx_doorbell, sizeof(uint32_t), BUS_DMASYNC_PREWRITE);
4878 
4879 	return (0);
4880 
4881 free_mxm:
4882 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&rx->rx_rq_mem),
4883 	    0, MCX_DMA_LEN(&rx->rx_rq_mem), BUS_DMASYNC_POSTWRITE);
4884 	mcx_dmamem_free(sc, &mxm);
4885 free_rq:
4886 	mcx_dmamem_free(sc, &rx->rx_rq_mem);
4887 	return (-1);
4888 }
4889 
4890 static int
4891 mcx_ready_rq(struct mcx_softc *sc, struct mcx_rx *rx)
4892 {
4893 	struct mcx_cmdq_entry *cqe;
4894 	struct mcx_dmamem mxm;
4895 	struct mcx_cmd_modify_rq_in *in;
4896 	struct mcx_cmd_modify_rq_mb_in *mbin;
4897 	struct mcx_cmd_modify_rq_out *out;
4898 	int error;
4899 	int token;
4900 
4901 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4902 	token = mcx_cmdq_token(sc);
4903 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
4904 	    sizeof(*out), token);
4905 
4906 	in = mcx_cmdq_in(cqe);
4907 	in->cmd_opcode = htobe16(MCX_CMD_MODIFY_RQ);
4908 	in->cmd_op_mod = htobe16(0);
4909 	in->cmd_rq_state = htobe32((MCX_QUEUE_STATE_RST << 28) | rx->rx_rqn);
4910 
4911 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
4912 	    &cqe->cq_input_ptr, token) != 0) {
4913 		printf("%s: unable to allocate modify rq mailbox\n",
4914 		    DEVNAME(sc));
4915 		return (-1);
4916 	}
4917 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
4918 	mbin->cmd_rq_ctx.rq_flags = htobe32(
4919 	    MCX_QUEUE_STATE_RDY << MCX_RQ_CTX_STATE_SHIFT);
4920 
4921 	mcx_cmdq_mboxes_sign(&mxm, 1);
4922 	mcx_cmdq_post(sc, cqe, 0);
4923 	error = mcx_cmdq_poll(sc, cqe, 1000);
4924 	if (error != 0) {
4925 		printf("%s: modify rq timeout\n", DEVNAME(sc));
4926 		goto free;
4927 	}
4928 	if (mcx_cmdq_verify(cqe) != 0) {
4929 		printf("%s: modify rq command corrupt\n", DEVNAME(sc));
4930 		goto free;
4931 	}
4932 
4933 	out = mcx_cmdq_out(cqe);
4934 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4935 		printf("%s: modify rq failed (%x, %x)\n", DEVNAME(sc),
4936 		    out->cmd_status, be32toh(out->cmd_syndrome));
4937 		error = -1;
4938 		goto free;
4939 	}
4940 
4941 free:
4942 	mcx_dmamem_free(sc, &mxm);
4943 	return (error);
4944 }
4945 
4946 static int
4947 mcx_destroy_rq(struct mcx_softc *sc, struct mcx_rx *rx)
4948 {
4949 	struct mcx_cmdq_entry *cqe;
4950 	struct mcx_cmd_destroy_rq_in *in;
4951 	struct mcx_cmd_destroy_rq_out *out;
4952 	int error;
4953 	int token;
4954 
4955 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
4956 	token = mcx_cmdq_token(sc);
4957 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), token);
4958 
4959 	in = mcx_cmdq_in(cqe);
4960 	in->cmd_opcode = htobe16(MCX_CMD_DESTROY_RQ);
4961 	in->cmd_op_mod = htobe16(0);
4962 	in->cmd_rqn = htobe32(rx->rx_rqn);
4963 
4964 	mcx_cmdq_post(sc, cqe, 0);
4965 	error = mcx_cmdq_poll(sc, cqe, 1000);
4966 	if (error != 0) {
4967 		printf("%s: destroy rq timeout\n", DEVNAME(sc));
4968 		return error;
4969 	}
4970 	if (mcx_cmdq_verify(cqe) != 0) {
4971 		printf("%s: destroy rq command corrupt\n", DEVNAME(sc));
4972 		return error;
4973 	}
4974 
4975 	out = mcx_cmdq_out(cqe);
4976 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
4977 		printf("%s: destroy rq failed (%x, %x)\n", DEVNAME(sc),
4978 		    out->cmd_status, be32toh(out->cmd_syndrome));
4979 		return -1;
4980 	}
4981 
4982 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_doorbell_mem),
4983 	    rx->rx_doorbell, sizeof(uint32_t), BUS_DMASYNC_POSTWRITE);
4984 
4985 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&rx->rx_rq_mem),
4986 	    0, MCX_DMA_LEN(&rx->rx_rq_mem), BUS_DMASYNC_POSTWRITE);
4987 	mcx_dmamem_free(sc, &rx->rx_rq_mem);
4988 
4989 	rx->rx_rqn = 0;
4990 	return 0;
4991 }
4992 
4993 static int
4994 mcx_create_tir_direct(struct mcx_softc *sc, struct mcx_rx *rx, int *tirn)
4995 {
4996 	struct mcx_cmdq_entry *cqe;
4997 	struct mcx_dmamem mxm;
4998 	struct mcx_cmd_create_tir_in *in;
4999 	struct mcx_cmd_create_tir_mb_in *mbin;
5000 	struct mcx_cmd_create_tir_out *out;
5001 	int error;
5002 	int token;
5003 
5004 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5005 	token = mcx_cmdq_token(sc);
5006 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
5007 	    sizeof(*out), token);
5008 
5009 	in = mcx_cmdq_in(cqe);
5010 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_TIR);
5011 	in->cmd_op_mod = htobe16(0);
5012 
5013 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
5014 	    &cqe->cq_input_ptr, token) != 0) {
5015 		printf("%s: unable to allocate create tir mailbox\n",
5016 		    DEVNAME(sc));
5017 		return (-1);
5018 	}
5019 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5020 	/* leave disp_type = 0, so packets get sent to the inline rqn */
5021 	mbin->cmd_inline_rqn = htobe32(rx->rx_rqn);
5022 	mbin->cmd_tdomain = htobe32(sc->sc_tdomain);
5023 
5024 	mcx_cmdq_post(sc, cqe, 0);
5025 	error = mcx_cmdq_poll(sc, cqe, 1000);
5026 	if (error != 0) {
5027 		printf("%s: create tir timeout\n", DEVNAME(sc));
5028 		goto free;
5029 	}
5030 	if (mcx_cmdq_verify(cqe) != 0) {
5031 		printf("%s: create tir command corrupt\n", DEVNAME(sc));
5032 		goto free;
5033 	}
5034 
5035 	out = mcx_cmdq_out(cqe);
5036 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5037 		printf("%s: create tir failed (%x, %x)\n", DEVNAME(sc),
5038 		    out->cmd_status, be32toh(out->cmd_syndrome));
5039 		error = -1;
5040 		goto free;
5041 	}
5042 
5043 	*tirn = mcx_get_id(out->cmd_tirn);
5044 free:
5045 	mcx_dmamem_free(sc, &mxm);
5046 	return (error);
5047 }
5048 
5049 static int
5050 mcx_create_tir_indirect(struct mcx_softc *sc, int rqtn, uint32_t hash_sel,
5051     int *tirn)
5052 {
5053 	struct mcx_cmdq_entry *cqe;
5054 	struct mcx_dmamem mxm;
5055 	struct mcx_cmd_create_tir_in *in;
5056 	struct mcx_cmd_create_tir_mb_in *mbin;
5057 	struct mcx_cmd_create_tir_out *out;
5058 	int error;
5059 	int token;
5060 
5061 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5062 	token = mcx_cmdq_token(sc);
5063 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
5064 	    sizeof(*out), token);
5065 
5066 	in = mcx_cmdq_in(cqe);
5067 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_TIR);
5068 	in->cmd_op_mod = htobe16(0);
5069 
5070 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
5071 	    &cqe->cq_input_ptr, token) != 0) {
5072 		printf("%s: unable to allocate create tir mailbox\n",
5073 		    DEVNAME(sc));
5074 		return (-1);
5075 	}
5076 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5077 	mbin->cmd_disp_type = htobe32(MCX_TIR_CTX_DISP_TYPE_INDIRECT
5078 	    << MCX_TIR_CTX_DISP_TYPE_SHIFT);
5079 	mbin->cmd_indir_table = htobe32(rqtn);
5080 	mbin->cmd_tdomain = htobe32(sc->sc_tdomain |
5081 	    MCX_TIR_CTX_HASH_TOEPLITZ << MCX_TIR_CTX_HASH_SHIFT);
5082 	mbin->cmd_rx_hash_sel_outer = htobe32(hash_sel);
5083 	stoeplitz_to_key(&mbin->cmd_rx_hash_key,
5084 	    sizeof(mbin->cmd_rx_hash_key));
5085 
5086 	mcx_cmdq_post(sc, cqe, 0);
5087 	error = mcx_cmdq_poll(sc, cqe, 1000);
5088 	if (error != 0) {
5089 		printf("%s: create tir timeout\n", DEVNAME(sc));
5090 		goto free;
5091 	}
5092 	if (mcx_cmdq_verify(cqe) != 0) {
5093 		printf("%s: create tir command corrupt\n", DEVNAME(sc));
5094 		goto free;
5095 	}
5096 
5097 	out = mcx_cmdq_out(cqe);
5098 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5099 		printf("%s: create tir failed (%x, %x)\n", DEVNAME(sc),
5100 		    out->cmd_status, be32toh(out->cmd_syndrome));
5101 		error = -1;
5102 		goto free;
5103 	}
5104 
5105 	*tirn = mcx_get_id(out->cmd_tirn);
5106 free:
5107 	mcx_dmamem_free(sc, &mxm);
5108 	return (error);
5109 }
5110 
5111 static int
5112 mcx_destroy_tir(struct mcx_softc *sc, int tirn)
5113 {
5114 	struct mcx_cmdq_entry *cqe;
5115 	struct mcx_cmd_destroy_tir_in *in;
5116 	struct mcx_cmd_destroy_tir_out *out;
5117 	int error;
5118 	int token;
5119 
5120 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5121 	token = mcx_cmdq_token(sc);
5122 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), token);
5123 
5124 	in = mcx_cmdq_in(cqe);
5125 	in->cmd_opcode = htobe16(MCX_CMD_DESTROY_TIR);
5126 	in->cmd_op_mod = htobe16(0);
5127 	in->cmd_tirn = htobe32(tirn);
5128 
5129 	mcx_cmdq_post(sc, cqe, 0);
5130 	error = mcx_cmdq_poll(sc, cqe, 1000);
5131 	if (error != 0) {
5132 		printf("%s: destroy tir timeout\n", DEVNAME(sc));
5133 		return error;
5134 	}
5135 	if (mcx_cmdq_verify(cqe) != 0) {
5136 		printf("%s: destroy tir command corrupt\n", DEVNAME(sc));
5137 		return error;
5138 	}
5139 
5140 	out = mcx_cmdq_out(cqe);
5141 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5142 		printf("%s: destroy tir failed (%x, %x)\n", DEVNAME(sc),
5143 		    out->cmd_status, be32toh(out->cmd_syndrome));
5144 		return -1;
5145 	}
5146 
5147 	return (0);
5148 }
5149 
5150 static int
5151 mcx_create_sq(struct mcx_softc *sc, struct mcx_tx *tx, int uar, int db,
5152     int cqn)
5153 {
5154 	struct mcx_cmdq_entry *cqe;
5155 	struct mcx_dmamem mxm;
5156 	struct mcx_cmd_create_sq_in *in;
5157 	struct mcx_sq_ctx *mbin;
5158 	struct mcx_cmd_create_sq_out *out;
5159 	int error;
5160 	uint64_t *pas;
5161 	int insize, npages, paslen, token;
5162 
5163 	tx->tx_doorbell = MCX_WQ_DOORBELL_BASE +
5164 	    (db * MCX_WQ_DOORBELL_STRIDE) + 4;
5165 
5166 	npages = howmany((1 << MCX_LOG_SQ_SIZE) * sizeof(struct mcx_sq_entry),
5167 	    MCX_PAGE_SIZE);
5168 	paslen = npages * sizeof(*pas);
5169 	insize = sizeof(struct mcx_sq_ctx) + paslen;
5170 
5171 	if (mcx_dmamem_alloc(sc, &tx->tx_sq_mem, npages * MCX_PAGE_SIZE,
5172 	    MCX_PAGE_SIZE) != 0) {
5173 		printf("%s: unable to allocate send queue memory\n",
5174 		    DEVNAME(sc));
5175 		return (-1);
5176 	}
5177 
5178 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5179 	token = mcx_cmdq_token(sc);
5180 	mcx_cmdq_init(sc, cqe, sizeof(*in) + insize + paslen, sizeof(*out),
5181 	    token);
5182 
5183 	in = mcx_cmdq_in(cqe);
5184 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_SQ);
5185 	in->cmd_op_mod = htobe16(0);
5186 
5187 	if (mcx_cmdq_mboxes_alloc(sc, &mxm,
5188 	    howmany(insize, MCX_CMDQ_MAILBOX_DATASIZE),
5189 	    &cqe->cq_input_ptr, token) != 0) {
5190 		printf("%s: unable to allocate create sq mailboxen\n",
5191 		    DEVNAME(sc));
5192 		goto free_sq;
5193 	}
5194 	mbin = (struct mcx_sq_ctx *)
5195 	    (((char *)mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0))) + 0x10);
5196 	mbin->sq_flags = htobe32(MCX_SQ_CTX_RLKEY |
5197 	    (1 << MCX_SQ_CTX_MIN_WQE_INLINE_SHIFT));
5198 	mbin->sq_cqn = htobe32(cqn);
5199 	mbin->sq_tis_lst_sz = htobe32(1 << MCX_SQ_CTX_TIS_LST_SZ_SHIFT);
5200 	mbin->sq_tis_num = htobe32(sc->sc_tis);
5201 	mbin->sq_wq.wq_type = MCX_WQ_CTX_TYPE_CYCLIC;
5202 	mbin->sq_wq.wq_pd = htobe32(sc->sc_pd);
5203 	mbin->sq_wq.wq_uar_page = htobe32(uar);
5204 	mbin->sq_wq.wq_doorbell = htobe64(MCX_DMA_DVA(&sc->sc_doorbell_mem) +
5205 	    tx->tx_doorbell);
5206 	mbin->sq_wq.wq_log_stride = htobe16(MCX_LOG_SQ_ENTRY_SIZE);
5207 	mbin->sq_wq.wq_log_size = MCX_LOG_SQ_SIZE;
5208 
5209 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&tx->tx_sq_mem),
5210 	    0, MCX_DMA_LEN(&tx->tx_sq_mem), BUS_DMASYNC_PREWRITE);
5211 
5212 	/* physical addresses follow the mailbox in data */
5213 	mcx_cmdq_mboxes_pas(&mxm, sizeof(*mbin) + 0x10,
5214 	    npages, &tx->tx_sq_mem);
5215 	mcx_cmdq_post(sc, cqe, 0);
5216 
5217 	error = mcx_cmdq_poll(sc, cqe, 1000);
5218 	if (error != 0) {
5219 		printf("%s: create sq timeout\n", DEVNAME(sc));
5220 		goto free_mxm;
5221 	}
5222 	if (mcx_cmdq_verify(cqe) != 0) {
5223 		printf("%s: create sq command corrupt\n", DEVNAME(sc));
5224 		goto free_mxm;
5225 	}
5226 
5227 	out = mcx_cmdq_out(cqe);
5228 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5229 		printf("%s: create sq failed (%x, %x)\n", DEVNAME(sc),
5230 		    out->cmd_status, be32toh(out->cmd_syndrome));
5231 		goto free_mxm;
5232 	}
5233 
5234 	tx->tx_uar = uar;
5235 	tx->tx_sqn = mcx_get_id(out->cmd_sqn);
5236 
5237 	mcx_dmamem_free(sc, &mxm);
5238 
5239 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_doorbell_mem),
5240 	    tx->tx_doorbell, sizeof(uint32_t), BUS_DMASYNC_PREWRITE);
5241 
5242 	return (0);
5243 
5244 free_mxm:
5245 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&tx->tx_sq_mem),
5246 	    0, MCX_DMA_LEN(&tx->tx_sq_mem), BUS_DMASYNC_POSTWRITE);
5247 	mcx_dmamem_free(sc, &mxm);
5248 free_sq:
5249 	mcx_dmamem_free(sc, &tx->tx_sq_mem);
5250 	return (-1);
5251 }
5252 
5253 static int
5254 mcx_destroy_sq(struct mcx_softc *sc, struct mcx_tx *tx)
5255 {
5256 	struct mcx_cmdq_entry *cqe;
5257 	struct mcx_cmd_destroy_sq_in *in;
5258 	struct mcx_cmd_destroy_sq_out *out;
5259 	int error;
5260 	int token;
5261 
5262 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5263 	token = mcx_cmdq_token(sc);
5264 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), token);
5265 
5266 	in = mcx_cmdq_in(cqe);
5267 	in->cmd_opcode = htobe16(MCX_CMD_DESTROY_SQ);
5268 	in->cmd_op_mod = htobe16(0);
5269 	in->cmd_sqn = htobe32(tx->tx_sqn);
5270 
5271 	mcx_cmdq_post(sc, cqe, 0);
5272 	error = mcx_cmdq_poll(sc, cqe, 1000);
5273 	if (error != 0) {
5274 		printf("%s: destroy sq timeout\n", DEVNAME(sc));
5275 		return error;
5276 	}
5277 	if (mcx_cmdq_verify(cqe) != 0) {
5278 		printf("%s: destroy sq command corrupt\n", DEVNAME(sc));
5279 		return error;
5280 	}
5281 
5282 	out = mcx_cmdq_out(cqe);
5283 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5284 		printf("%s: destroy sq failed (%x, %x)\n", DEVNAME(sc),
5285 		    out->cmd_status, be32toh(out->cmd_syndrome));
5286 		return -1;
5287 	}
5288 
5289 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_doorbell_mem),
5290 	    tx->tx_doorbell, sizeof(uint32_t), BUS_DMASYNC_POSTWRITE);
5291 
5292 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&tx->tx_sq_mem),
5293 	    0, MCX_DMA_LEN(&tx->tx_sq_mem), BUS_DMASYNC_POSTWRITE);
5294 	mcx_dmamem_free(sc, &tx->tx_sq_mem);
5295 
5296 	tx->tx_sqn = 0;
5297 	return 0;
5298 }
5299 
5300 static int
5301 mcx_ready_sq(struct mcx_softc *sc, struct mcx_tx *tx)
5302 {
5303 	struct mcx_cmdq_entry *cqe;
5304 	struct mcx_dmamem mxm;
5305 	struct mcx_cmd_modify_sq_in *in;
5306 	struct mcx_cmd_modify_sq_mb_in *mbin;
5307 	struct mcx_cmd_modify_sq_out *out;
5308 	int error;
5309 	int token;
5310 
5311 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5312 	token = mcx_cmdq_token(sc);
5313 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
5314 	    sizeof(*out), token);
5315 
5316 	in = mcx_cmdq_in(cqe);
5317 	in->cmd_opcode = htobe16(MCX_CMD_MODIFY_SQ);
5318 	in->cmd_op_mod = htobe16(0);
5319 	in->cmd_sq_state = htobe32((MCX_QUEUE_STATE_RST << 28) | tx->tx_sqn);
5320 
5321 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
5322 	    &cqe->cq_input_ptr, token) != 0) {
5323 		printf("%s: unable to allocate modify sq mailbox\n",
5324 		    DEVNAME(sc));
5325 		return (-1);
5326 	}
5327 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5328 	mbin->cmd_sq_ctx.sq_flags = htobe32(
5329 	    MCX_QUEUE_STATE_RDY << MCX_SQ_CTX_STATE_SHIFT);
5330 
5331 	mcx_cmdq_mboxes_sign(&mxm, 1);
5332 	mcx_cmdq_post(sc, cqe, 0);
5333 	error = mcx_cmdq_poll(sc, cqe, 1000);
5334 	if (error != 0) {
5335 		printf("%s: modify sq timeout\n", DEVNAME(sc));
5336 		goto free;
5337 	}
5338 	if (mcx_cmdq_verify(cqe) != 0) {
5339 		printf("%s: modify sq command corrupt\n", DEVNAME(sc));
5340 		goto free;
5341 	}
5342 
5343 	out = mcx_cmdq_out(cqe);
5344 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5345 		printf("%s: modify sq failed (%x, %x)\n", DEVNAME(sc),
5346 		    out->cmd_status, be32toh(out->cmd_syndrome));
5347 		error = -1;
5348 		goto free;
5349 	}
5350 
5351 free:
5352 	mcx_dmamem_free(sc, &mxm);
5353 	return (error);
5354 }
5355 
5356 static int
5357 mcx_create_tis(struct mcx_softc *sc, int *tis)
5358 {
5359 	struct mcx_cmdq_entry *cqe;
5360 	struct mcx_dmamem mxm;
5361 	struct mcx_cmd_create_tis_in *in;
5362 	struct mcx_cmd_create_tis_mb_in *mbin;
5363 	struct mcx_cmd_create_tis_out *out;
5364 	int error;
5365 	int token;
5366 
5367 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5368 	token = mcx_cmdq_token(sc);
5369 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
5370 	    sizeof(*out), token);
5371 
5372 	in = mcx_cmdq_in(cqe);
5373 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_TIS);
5374 	in->cmd_op_mod = htobe16(0);
5375 
5376 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
5377 	    &cqe->cq_input_ptr, token) != 0) {
5378 		printf("%s: unable to allocate create tis mailbox\n",
5379 		    DEVNAME(sc));
5380 		return (-1);
5381 	}
5382 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5383 	mbin->cmd_tdomain = htobe32(sc->sc_tdomain);
5384 
5385 	mcx_cmdq_mboxes_sign(&mxm, 1);
5386 	mcx_cmdq_post(sc, cqe, 0);
5387 	error = mcx_cmdq_poll(sc, cqe, 1000);
5388 	if (error != 0) {
5389 		printf("%s: create tis timeout\n", DEVNAME(sc));
5390 		goto free;
5391 	}
5392 	if (mcx_cmdq_verify(cqe) != 0) {
5393 		printf("%s: create tis command corrupt\n", DEVNAME(sc));
5394 		goto free;
5395 	}
5396 
5397 	out = mcx_cmdq_out(cqe);
5398 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5399 		printf("%s: create tis failed (%x, %x)\n", DEVNAME(sc),
5400 		    out->cmd_status, be32toh(out->cmd_syndrome));
5401 		error = -1;
5402 		goto free;
5403 	}
5404 
5405 	*tis = mcx_get_id(out->cmd_tisn);
5406 free:
5407 	mcx_dmamem_free(sc, &mxm);
5408 	return (error);
5409 }
5410 
5411 static int
5412 mcx_destroy_tis(struct mcx_softc *sc, int tis)
5413 {
5414 	struct mcx_cmdq_entry *cqe;
5415 	struct mcx_cmd_destroy_tis_in *in;
5416 	struct mcx_cmd_destroy_tis_out *out;
5417 	int error;
5418 	int token;
5419 
5420 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5421 	token = mcx_cmdq_token(sc);
5422 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), token);
5423 
5424 	in = mcx_cmdq_in(cqe);
5425 	in->cmd_opcode = htobe16(MCX_CMD_DESTROY_TIS);
5426 	in->cmd_op_mod = htobe16(0);
5427 	in->cmd_tisn = htobe32(tis);
5428 
5429 	mcx_cmdq_post(sc, cqe, 0);
5430 	error = mcx_cmdq_poll(sc, cqe, 1000);
5431 	if (error != 0) {
5432 		printf("%s: destroy tis timeout\n", DEVNAME(sc));
5433 		return error;
5434 	}
5435 	if (mcx_cmdq_verify(cqe) != 0) {
5436 		printf("%s: destroy tis command corrupt\n", DEVNAME(sc));
5437 		return error;
5438 	}
5439 
5440 	out = mcx_cmdq_out(cqe);
5441 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5442 		printf("%s: destroy tis failed (%x, %x)\n", DEVNAME(sc),
5443 		    out->cmd_status, be32toh(out->cmd_syndrome));
5444 		return -1;
5445 	}
5446 
5447 	return 0;
5448 }
5449 
5450 static int
5451 mcx_create_rqt(struct mcx_softc *sc, int size, int *rqns, int *rqt)
5452 {
5453 	struct mcx_cmdq_entry *cqe;
5454 	struct mcx_dmamem mxm;
5455 	struct mcx_cmd_create_rqt_in *in;
5456 	struct mcx_cmd_create_rqt_mb_in *mbin;
5457 	struct mcx_cmd_create_rqt_out *out;
5458 	struct mcx_rqt_ctx *rqt_ctx;
5459 	int *rqtn;
5460 	int error;
5461 	int token;
5462 	int i;
5463 
5464 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5465 	token = mcx_cmdq_token(sc);
5466 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin) +
5467 	    (size * sizeof(int)), sizeof(*out), token);
5468 
5469 	in = mcx_cmdq_in(cqe);
5470 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_RQT);
5471 	in->cmd_op_mod = htobe16(0);
5472 
5473 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
5474 	    &cqe->cq_input_ptr, token) != 0) {
5475 		printf("%s: unable to allocate create rqt mailbox\n",
5476 		    DEVNAME(sc));
5477 		return (-1);
5478 	}
5479 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5480 	rqt_ctx = &mbin->cmd_rqt;
5481 	rqt_ctx->cmd_rqt_max_size = htobe16(sc->sc_max_rqt_size);
5482 	rqt_ctx->cmd_rqt_actual_size = htobe16(size);
5483 
5484 	/* rqt list follows the rqt context */
5485 	rqtn = (int *)(rqt_ctx + 1);
5486 	for (i = 0; i < size; i++) {
5487 		rqtn[i] = htobe32(rqns[i]);
5488 	}
5489 
5490 	mcx_cmdq_mboxes_sign(&mxm, 1);
5491 	mcx_cmdq_post(sc, cqe, 0);
5492 	error = mcx_cmdq_poll(sc, cqe, 1000);
5493 	if (error != 0) {
5494 		printf("%s: create rqt timeout\n", DEVNAME(sc));
5495 		goto free;
5496 	}
5497 	if (mcx_cmdq_verify(cqe) != 0) {
5498 		printf("%s: create rqt command corrupt\n", DEVNAME(sc));
5499 		goto free;
5500 	}
5501 
5502 	out = mcx_cmdq_out(cqe);
5503 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5504 		printf("%s: create rqt failed (%x, %x)\n", DEVNAME(sc),
5505 		    out->cmd_status, be32toh(out->cmd_syndrome));
5506 		error = -1;
5507 		goto free;
5508 	}
5509 
5510 	*rqt = mcx_get_id(out->cmd_rqtn);
5511 	return (0);
5512 free:
5513 	mcx_dmamem_free(sc, &mxm);
5514 	return (error);
5515 }
5516 
5517 static int
5518 mcx_destroy_rqt(struct mcx_softc *sc, int rqt)
5519 {
5520 	struct mcx_cmdq_entry *cqe;
5521 	struct mcx_cmd_destroy_rqt_in *in;
5522 	struct mcx_cmd_destroy_rqt_out *out;
5523 	int error;
5524 	int token;
5525 
5526 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5527 	token = mcx_cmdq_token(sc);
5528 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), token);
5529 
5530 	in = mcx_cmdq_in(cqe);
5531 	in->cmd_opcode = htobe16(MCX_CMD_DESTROY_RQT);
5532 	in->cmd_op_mod = htobe16(0);
5533 	in->cmd_rqtn = htobe32(rqt);
5534 
5535 	mcx_cmdq_post(sc, cqe, 0);
5536 	error = mcx_cmdq_poll(sc, cqe, 1000);
5537 	if (error != 0) {
5538 		printf("%s: destroy rqt timeout\n", DEVNAME(sc));
5539 		return error;
5540 	}
5541 	if (mcx_cmdq_verify(cqe) != 0) {
5542 		printf("%s: destroy rqt command corrupt\n", DEVNAME(sc));
5543 		return error;
5544 	}
5545 
5546 	out = mcx_cmdq_out(cqe);
5547 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5548 		printf("%s: destroy rqt failed (%x, %x)\n", DEVNAME(sc),
5549 		    out->cmd_status, be32toh(out->cmd_syndrome));
5550 		return -1;
5551 	}
5552 
5553 	return 0;
5554 }
5555 
5556 #if 0
5557 static int
5558 mcx_alloc_flow_counter(struct mcx_softc *sc, int i)
5559 {
5560 	struct mcx_cmdq_entry *cqe;
5561 	struct mcx_cmd_alloc_flow_counter_in *in;
5562 	struct mcx_cmd_alloc_flow_counter_out *out;
5563 	int error;
5564 
5565 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5566 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out), mcx_cmdq_token(sc));
5567 
5568 	in = mcx_cmdq_in(cqe);
5569 	in->cmd_opcode = htobe16(MCX_CMD_ALLOC_FLOW_COUNTER);
5570 	in->cmd_op_mod = htobe16(0);
5571 
5572 	mcx_cmdq_post(sc, cqe, 0);
5573 
5574 	error = mcx_cmdq_poll(sc, cqe, 1000);
5575 	if (error != 0) {
5576 		printf("%s: alloc flow counter timeout\n", DEVNAME(sc));
5577 		return (-1);
5578 	}
5579 	if (mcx_cmdq_verify(cqe) != 0) {
5580 		printf("%s: alloc flow counter command corrupt\n", DEVNAME(sc));
5581 		return (-1);
5582 	}
5583 
5584 	out = (struct mcx_cmd_alloc_flow_counter_out *)cqe->cq_output_data;
5585 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5586 		printf("%s: alloc flow counter failed (%x)\n", DEVNAME(sc),
5587 		    out->cmd_status);
5588 		return (-1);
5589 	}
5590 
5591 	sc->sc_flow_counter_id[i]  = be16toh(out->cmd_flow_counter_id);
5592 	printf("flow counter id %d = %d\n", i, sc->sc_flow_counter_id[i]);
5593 
5594 	return (0);
5595 }
5596 #endif
5597 
5598 static int
5599 mcx_create_flow_table(struct mcx_softc *sc, int log_size, int level,
5600     int *flow_table_id)
5601 {
5602 	struct mcx_cmdq_entry *cqe;
5603 	struct mcx_dmamem mxm;
5604 	struct mcx_cmd_create_flow_table_in *in;
5605 	struct mcx_cmd_create_flow_table_mb_in *mbin;
5606 	struct mcx_cmd_create_flow_table_out *out;
5607 	int error;
5608 	int token;
5609 
5610 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5611 	token = mcx_cmdq_token(sc);
5612 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
5613 	    sizeof(*out), token);
5614 
5615 	in = mcx_cmdq_in(cqe);
5616 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_FLOW_TABLE);
5617 	in->cmd_op_mod = htobe16(0);
5618 
5619 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
5620 	    &cqe->cq_input_ptr, token) != 0) {
5621 		printf("%s: unable to allocate create flow table mailbox\n",
5622 		    DEVNAME(sc));
5623 		return (-1);
5624 	}
5625 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5626 	mbin->cmd_table_type = MCX_FLOW_TABLE_TYPE_RX;
5627 	mbin->cmd_ctx.ft_log_size = log_size;
5628 	mbin->cmd_ctx.ft_level = level;
5629 
5630 	mcx_cmdq_mboxes_sign(&mxm, 1);
5631 	mcx_cmdq_post(sc, cqe, 0);
5632 	error = mcx_cmdq_poll(sc, cqe, 1000);
5633 	if (error != 0) {
5634 		printf("%s: create flow table timeout\n", DEVNAME(sc));
5635 		goto free;
5636 	}
5637 	if (mcx_cmdq_verify(cqe) != 0) {
5638 		printf("%s: create flow table command corrupt\n", DEVNAME(sc));
5639 		goto free;
5640 	}
5641 
5642 	out = mcx_cmdq_out(cqe);
5643 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5644 		printf("%s: create flow table failed (%x, %x)\n", DEVNAME(sc),
5645 		    out->cmd_status, be32toh(out->cmd_syndrome));
5646 		error = -1;
5647 		goto free;
5648 	}
5649 
5650 	*flow_table_id = mcx_get_id(out->cmd_table_id);
5651 free:
5652 	mcx_dmamem_free(sc, &mxm);
5653 	return (error);
5654 }
5655 
5656 static int
5657 mcx_set_flow_table_root(struct mcx_softc *sc, int flow_table_id)
5658 {
5659 	struct mcx_cmdq_entry *cqe;
5660 	struct mcx_dmamem mxm;
5661 	struct mcx_cmd_set_flow_table_root_in *in;
5662 	struct mcx_cmd_set_flow_table_root_mb_in *mbin;
5663 	struct mcx_cmd_set_flow_table_root_out *out;
5664 	int error;
5665 	int token;
5666 
5667 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5668 	token = mcx_cmdq_token(sc);
5669 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
5670 	    sizeof(*out), token);
5671 
5672 	in = mcx_cmdq_in(cqe);
5673 	in->cmd_opcode = htobe16(MCX_CMD_SET_FLOW_TABLE_ROOT);
5674 	in->cmd_op_mod = htobe16(0);
5675 
5676 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
5677 	    &cqe->cq_input_ptr, token) != 0) {
5678 		printf("%s: unable to allocate set flow table root mailbox\n",
5679 		    DEVNAME(sc));
5680 		return (-1);
5681 	}
5682 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5683 	mbin->cmd_table_type = MCX_FLOW_TABLE_TYPE_RX;
5684 	mbin->cmd_table_id = htobe32(flow_table_id);
5685 
5686 	mcx_cmdq_mboxes_sign(&mxm, 1);
5687 	mcx_cmdq_post(sc, cqe, 0);
5688 	error = mcx_cmdq_poll(sc, cqe, 1000);
5689 	if (error != 0) {
5690 		printf("%s: set flow table root timeout\n", DEVNAME(sc));
5691 		goto free;
5692 	}
5693 	if (mcx_cmdq_verify(cqe) != 0) {
5694 		printf("%s: set flow table root command corrupt\n",
5695 		    DEVNAME(sc));
5696 		goto free;
5697 	}
5698 
5699 	out = mcx_cmdq_out(cqe);
5700 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5701 		printf("%s: set flow table root failed (%x, %x)\n",
5702 		    DEVNAME(sc), out->cmd_status, be32toh(out->cmd_syndrome));
5703 		error = -1;
5704 		goto free;
5705 	}
5706 
5707 free:
5708 	mcx_dmamem_free(sc, &mxm);
5709 	return (error);
5710 }
5711 
5712 static int
5713 mcx_destroy_flow_table(struct mcx_softc *sc, int flow_table_id)
5714 {
5715 	struct mcx_cmdq_entry *cqe;
5716 	struct mcx_dmamem mxm;
5717 	struct mcx_cmd_destroy_flow_table_in *in;
5718 	struct mcx_cmd_destroy_flow_table_mb_in *mb;
5719 	struct mcx_cmd_destroy_flow_table_out *out;
5720 	int error;
5721 	int token;
5722 
5723 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5724 	token = mcx_cmdq_token(sc);
5725 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mb), sizeof(*out), token);
5726 
5727 	in = mcx_cmdq_in(cqe);
5728 	in->cmd_opcode = htobe16(MCX_CMD_DESTROY_FLOW_TABLE);
5729 	in->cmd_op_mod = htobe16(0);
5730 
5731 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
5732 	    &cqe->cq_input_ptr, token) != 0) {
5733 		printf("%s: unable to allocate destroy flow table mailbox\n",
5734 		    DEVNAME(sc));
5735 		return (-1);
5736 	}
5737 	mb = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5738 	mb->cmd_table_type = MCX_FLOW_TABLE_TYPE_RX;
5739 	mb->cmd_table_id = htobe32(flow_table_id);
5740 
5741 	mcx_cmdq_mboxes_sign(&mxm, 1);
5742 	mcx_cmdq_post(sc, cqe, 0);
5743 	error = mcx_cmdq_poll(sc, cqe, 1000);
5744 	if (error != 0) {
5745 		printf("%s: destroy flow table timeout\n", DEVNAME(sc));
5746 		goto free;
5747 	}
5748 	if (mcx_cmdq_verify(cqe) != 0) {
5749 		printf("%s: destroy flow table command corrupt\n",
5750 		    DEVNAME(sc));
5751 		goto free;
5752 	}
5753 
5754 	out = mcx_cmdq_out(cqe);
5755 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5756 		printf("%s: destroy flow table failed (%x, %x)\n", DEVNAME(sc),
5757 		    out->cmd_status, be32toh(out->cmd_syndrome));
5758 		error = -1;
5759 		goto free;
5760 	}
5761 
5762 free:
5763 	mcx_dmamem_free(sc, &mxm);
5764 	return (error);
5765 }
5766 
5767 
5768 static int
5769 mcx_create_flow_group(struct mcx_softc *sc, int flow_table_id, int group,
5770     int start, int size, int match_enable, struct mcx_flow_match *match)
5771 {
5772 	struct mcx_cmdq_entry *cqe;
5773 	struct mcx_dmamem mxm;
5774 	struct mcx_cmd_create_flow_group_in *in;
5775 	struct mcx_cmd_create_flow_group_mb_in *mbin;
5776 	struct mcx_cmd_create_flow_group_out *out;
5777 	struct mcx_flow_group *mfg;
5778 	int error;
5779 	int token;
5780 
5781 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5782 	token = mcx_cmdq_token(sc);
5783 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin), sizeof(*out),
5784 	    token);
5785 
5786 	in = mcx_cmdq_in(cqe);
5787 	in->cmd_opcode = htobe16(MCX_CMD_CREATE_FLOW_GROUP);
5788 	in->cmd_op_mod = htobe16(0);
5789 
5790 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2, &cqe->cq_input_ptr, token)
5791 	    != 0) {
5792 		printf("%s: unable to allocate create flow group mailbox\n",
5793 		    DEVNAME(sc));
5794 		return (-1);
5795 	}
5796 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5797 	mbin->cmd_table_type = MCX_FLOW_TABLE_TYPE_RX;
5798 	mbin->cmd_table_id = htobe32(flow_table_id);
5799 	mbin->cmd_start_flow_index = htobe32(start);
5800 	mbin->cmd_end_flow_index = htobe32(start + (size - 1));
5801 
5802 	mbin->cmd_match_criteria_enable = match_enable;
5803 	memcpy(&mbin->cmd_match_criteria, match, sizeof(*match));
5804 
5805 	mcx_cmdq_mboxes_sign(&mxm, 2);
5806 	mcx_cmdq_post(sc, cqe, 0);
5807 	error = mcx_cmdq_poll(sc, cqe, 1000);
5808 	if (error != 0) {
5809 		printf("%s: create flow group timeout\n", DEVNAME(sc));
5810 		goto free;
5811 	}
5812 	if (mcx_cmdq_verify(cqe) != 0) {
5813 		printf("%s: create flow group command corrupt\n", DEVNAME(sc));
5814 		goto free;
5815 	}
5816 
5817 	out = mcx_cmdq_out(cqe);
5818 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5819 		printf("%s: create flow group failed (%x, %x)\n", DEVNAME(sc),
5820 		    out->cmd_status, be32toh(out->cmd_syndrome));
5821 		error = -1;
5822 		goto free;
5823 	}
5824 
5825 	mfg = &sc->sc_flow_group[group];
5826 	mfg->g_id = mcx_get_id(out->cmd_group_id);
5827 	mfg->g_table = flow_table_id;
5828 	mfg->g_start = start;
5829 	mfg->g_size = size;
5830 
5831 free:
5832 	mcx_dmamem_free(sc, &mxm);
5833 	return (error);
5834 }
5835 
5836 static int
5837 mcx_destroy_flow_group(struct mcx_softc *sc, int group)
5838 {
5839 	struct mcx_cmdq_entry *cqe;
5840 	struct mcx_dmamem mxm;
5841 	struct mcx_cmd_destroy_flow_group_in *in;
5842 	struct mcx_cmd_destroy_flow_group_mb_in *mb;
5843 	struct mcx_cmd_destroy_flow_group_out *out;
5844 	struct mcx_flow_group *mfg;
5845 	int error;
5846 	int token;
5847 
5848 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5849 	token = mcx_cmdq_token(sc);
5850 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mb), sizeof(*out), token);
5851 
5852 	in = mcx_cmdq_in(cqe);
5853 	in->cmd_opcode = htobe16(MCX_CMD_DESTROY_FLOW_GROUP);
5854 	in->cmd_op_mod = htobe16(0);
5855 
5856 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2,
5857 	    &cqe->cq_input_ptr, token) != 0) {
5858 		printf("%s: unable to allocate destroy flow group mailbox\n",
5859 		    DEVNAME(sc));
5860 		return (-1);
5861 	}
5862 	mb = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5863 	mb->cmd_table_type = MCX_FLOW_TABLE_TYPE_RX;
5864 	mfg = &sc->sc_flow_group[group];
5865 	mb->cmd_table_id = htobe32(mfg->g_table);
5866 	mb->cmd_group_id = htobe32(mfg->g_id);
5867 
5868 	mcx_cmdq_mboxes_sign(&mxm, 2);
5869 	mcx_cmdq_post(sc, cqe, 0);
5870 	error = mcx_cmdq_poll(sc, cqe, 1000);
5871 	if (error != 0) {
5872 		printf("%s: destroy flow group timeout\n", DEVNAME(sc));
5873 		goto free;
5874 	}
5875 	if (mcx_cmdq_verify(cqe) != 0) {
5876 		printf("%s: destroy flow group command corrupt\n", DEVNAME(sc));
5877 		goto free;
5878 	}
5879 
5880 	out = mcx_cmdq_out(cqe);
5881 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5882 		printf("%s: destroy flow group failed (%x, %x)\n", DEVNAME(sc),
5883 		    out->cmd_status, be32toh(out->cmd_syndrome));
5884 		error = -1;
5885 		goto free;
5886 	}
5887 
5888 	mfg->g_id = -1;
5889 	mfg->g_table = -1;
5890 	mfg->g_size = 0;
5891 	mfg->g_start = 0;
5892 free:
5893 	mcx_dmamem_free(sc, &mxm);
5894 	return (error);
5895 }
5896 
5897 static int
5898 mcx_set_flow_table_entry_mac(struct mcx_softc *sc, int group, int index,
5899     const uint8_t *macaddr, uint32_t dest)
5900 {
5901 	struct mcx_cmdq_entry *cqe;
5902 	struct mcx_dmamem mxm;
5903 	struct mcx_cmd_set_flow_table_entry_in *in;
5904 	struct mcx_cmd_set_flow_table_entry_mb_in *mbin;
5905 	struct mcx_cmd_set_flow_table_entry_out *out;
5906 	struct mcx_flow_group *mfg;
5907 	uint32_t *pdest;
5908 	int error;
5909 	int token;
5910 
5911 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5912 	token = mcx_cmdq_token(sc);
5913 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin) + sizeof(*pdest),
5914 	    sizeof(*out), token);
5915 
5916 	in = mcx_cmdq_in(cqe);
5917 	in->cmd_opcode = htobe16(MCX_CMD_SET_FLOW_TABLE_ENTRY);
5918 	in->cmd_op_mod = htobe16(0);
5919 
5920 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2, &cqe->cq_input_ptr, token)
5921 	    != 0) {
5922 		printf("%s: unable to allocate set flow table entry mailbox\n",
5923 		    DEVNAME(sc));
5924 		return (-1);
5925 	}
5926 
5927 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
5928 	mbin->cmd_table_type = MCX_FLOW_TABLE_TYPE_RX;
5929 
5930 	mfg = &sc->sc_flow_group[group];
5931 	mbin->cmd_table_id = htobe32(mfg->g_table);
5932 	mbin->cmd_flow_index = htobe32(mfg->g_start + index);
5933 	mbin->cmd_flow_ctx.fc_group_id = htobe32(mfg->g_id);
5934 
5935 	/* flow context ends at offset 0x330, 0x130 into the second mbox */
5936 	pdest = (uint32_t *)
5937 	    (((char *)mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 1))) + 0x130);
5938 	mbin->cmd_flow_ctx.fc_action = htobe32(MCX_FLOW_CONTEXT_ACTION_FORWARD);
5939 	mbin->cmd_flow_ctx.fc_dest_list_size = htobe32(1);
5940 	*pdest = htobe32(dest);
5941 
5942 	/* the only thing we match on at the moment is the dest mac address */
5943 	if (macaddr != NULL) {
5944 		memcpy(mbin->cmd_flow_ctx.fc_match_value.mc_dest_mac, macaddr,
5945 		    ETHER_ADDR_LEN);
5946 	}
5947 
5948 	mcx_cmdq_mboxes_sign(&mxm, 2);
5949 	mcx_cmdq_post(sc, cqe, 0);
5950 	error = mcx_cmdq_poll(sc, cqe, 1000);
5951 	if (error != 0) {
5952 		printf("%s: set flow table entry timeout\n", DEVNAME(sc));
5953 		goto free;
5954 	}
5955 	if (mcx_cmdq_verify(cqe) != 0) {
5956 		printf("%s: set flow table entry command corrupt\n",
5957 		    DEVNAME(sc));
5958 		goto free;
5959 	}
5960 
5961 	out = mcx_cmdq_out(cqe);
5962 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
5963 		printf("%s: set flow table entry failed (%x, %x)\n",
5964 		    DEVNAME(sc), out->cmd_status, be32toh(out->cmd_syndrome));
5965 		error = -1;
5966 		goto free;
5967 	}
5968 
5969 free:
5970 	mcx_dmamem_free(sc, &mxm);
5971 	return (error);
5972 }
5973 
5974 static int
5975 mcx_set_flow_table_entry_proto(struct mcx_softc *sc, int group, int index,
5976     int ethertype, int ip_proto, uint32_t dest)
5977 {
5978 	struct mcx_cmdq_entry *cqe;
5979 	struct mcx_dmamem mxm;
5980 	struct mcx_cmd_set_flow_table_entry_in *in;
5981 	struct mcx_cmd_set_flow_table_entry_mb_in *mbin;
5982 	struct mcx_cmd_set_flow_table_entry_out *out;
5983 	struct mcx_flow_group *mfg;
5984 	uint32_t *pdest;
5985 	int error;
5986 	int token;
5987 
5988 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
5989 	token = mcx_cmdq_token(sc);
5990 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin) + sizeof(*pdest),
5991 	    sizeof(*out), token);
5992 
5993 	in = mcx_cmdq_in(cqe);
5994 	in->cmd_opcode = htobe16(MCX_CMD_SET_FLOW_TABLE_ENTRY);
5995 	in->cmd_op_mod = htobe16(0);
5996 
5997 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2, &cqe->cq_input_ptr, token)
5998 	    != 0) {
5999 		printf("%s: unable to allocate set flow table entry mailbox\n",
6000 		    DEVNAME(sc));
6001 		return (-1);
6002 	}
6003 
6004 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
6005 	mbin->cmd_table_type = MCX_FLOW_TABLE_TYPE_RX;
6006 
6007 	mfg = &sc->sc_flow_group[group];
6008 	mbin->cmd_table_id = htobe32(mfg->g_table);
6009 	mbin->cmd_flow_index = htobe32(mfg->g_start + index);
6010 	mbin->cmd_flow_ctx.fc_group_id = htobe32(mfg->g_id);
6011 
6012 	/* flow context ends at offset 0x330, 0x130 into the second mbox */
6013 	pdest = (uint32_t *)
6014 	    (((char *)mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 1))) + 0x130);
6015 	mbin->cmd_flow_ctx.fc_action = htobe32(MCX_FLOW_CONTEXT_ACTION_FORWARD);
6016 	mbin->cmd_flow_ctx.fc_dest_list_size = htobe32(1);
6017 	*pdest = htobe32(dest);
6018 
6019 	mbin->cmd_flow_ctx.fc_match_value.mc_ethertype = htobe16(ethertype);
6020 	mbin->cmd_flow_ctx.fc_match_value.mc_ip_proto = ip_proto;
6021 
6022 	mcx_cmdq_mboxes_sign(&mxm, 2);
6023 	mcx_cmdq_post(sc, cqe, 0);
6024 	error = mcx_cmdq_poll(sc, cqe, 1000);
6025 	if (error != 0) {
6026 		printf("%s: set flow table entry timeout\n", DEVNAME(sc));
6027 		goto free;
6028 	}
6029 	if (mcx_cmdq_verify(cqe) != 0) {
6030 		printf("%s: set flow table entry command corrupt\n",
6031 		    DEVNAME(sc));
6032 		goto free;
6033 	}
6034 
6035 	out = mcx_cmdq_out(cqe);
6036 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
6037 		printf("%s: set flow table entry failed (%x, %x)\n",
6038 		    DEVNAME(sc), out->cmd_status, be32toh(out->cmd_syndrome));
6039 		error = -1;
6040 		goto free;
6041 	}
6042 
6043 free:
6044 	mcx_dmamem_free(sc, &mxm);
6045 	return (error);
6046 }
6047 
6048 static int
6049 mcx_delete_flow_table_entry(struct mcx_softc *sc, int group, int index)
6050 {
6051 	struct mcx_cmdq_entry *cqe;
6052 	struct mcx_dmamem mxm;
6053 	struct mcx_cmd_delete_flow_table_entry_in *in;
6054 	struct mcx_cmd_delete_flow_table_entry_mb_in *mbin;
6055 	struct mcx_cmd_delete_flow_table_entry_out *out;
6056 	struct mcx_flow_group *mfg;
6057 	int error;
6058 	int token;
6059 
6060 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
6061 	token = mcx_cmdq_token(sc);
6062 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin), sizeof(*out),
6063 	    token);
6064 
6065 	in = mcx_cmdq_in(cqe);
6066 	in->cmd_opcode = htobe16(MCX_CMD_DELETE_FLOW_TABLE_ENTRY);
6067 	in->cmd_op_mod = htobe16(0);
6068 
6069 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2,
6070 	    &cqe->cq_input_ptr, token) != 0) {
6071 		printf("%s: unable to allocate "
6072 		    "delete flow table entry mailbox\n", DEVNAME(sc));
6073 		return (-1);
6074 	}
6075 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
6076 	mbin->cmd_table_type = MCX_FLOW_TABLE_TYPE_RX;
6077 
6078 	mfg = &sc->sc_flow_group[group];
6079 	mbin->cmd_table_id = htobe32(mfg->g_table);
6080 	mbin->cmd_flow_index = htobe32(mfg->g_start + index);
6081 
6082 	mcx_cmdq_mboxes_sign(&mxm, 2);
6083 	mcx_cmdq_post(sc, cqe, 0);
6084 	error = mcx_cmdq_poll(sc, cqe, 1000);
6085 	if (error != 0) {
6086 		printf("%s: delete flow table entry timeout\n", DEVNAME(sc));
6087 		goto free;
6088 	}
6089 	if (mcx_cmdq_verify(cqe) != 0) {
6090 		printf("%s: delete flow table entry command corrupt\n",
6091 		    DEVNAME(sc));
6092 		goto free;
6093 	}
6094 
6095 	out = mcx_cmdq_out(cqe);
6096 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
6097 		printf("%s: delete flow table entry %d:%d failed (%x, %x)\n",
6098 		    DEVNAME(sc), group, index, out->cmd_status,
6099 		    be32toh(out->cmd_syndrome));
6100 		error = -1;
6101 		goto free;
6102 	}
6103 
6104 free:
6105 	mcx_dmamem_free(sc, &mxm);
6106 	return (error);
6107 }
6108 
6109 #if 0
6110 int
6111 mcx_dump_flow_table(struct mcx_softc *sc, int flow_table_id)
6112 {
6113 	struct mcx_dmamem mxm;
6114 	struct mcx_cmdq_entry *cqe;
6115 	struct mcx_cmd_query_flow_table_in *in;
6116 	struct mcx_cmd_query_flow_table_mb_in *mbin;
6117 	struct mcx_cmd_query_flow_table_out *out;
6118 	struct mcx_cmd_query_flow_table_mb_out *mbout;
6119 	uint8_t token = mcx_cmdq_token(sc);
6120 	int error;
6121 	int i;
6122 	uint8_t *dump;
6123 
6124 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
6125 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
6126 	    sizeof(*out) + sizeof(*mbout) + 16, token);
6127 
6128 	in = mcx_cmdq_in(cqe);
6129 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_FLOW_TABLE);
6130 	in->cmd_op_mod = htobe16(0);
6131 
6132 	CTASSERT(sizeof(*mbin) <= MCX_CMDQ_MAILBOX_DATASIZE);
6133 	CTASSERT(sizeof(*mbout) <= MCX_CMDQ_MAILBOX_DATASIZE);
6134 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2,
6135 	    &cqe->cq_output_ptr, token) != 0) {
6136 		printf(", unable to allocate query flow table mailboxes\n");
6137 		return (-1);
6138 	}
6139 	cqe->cq_input_ptr = cqe->cq_output_ptr;
6140 
6141 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
6142 	mbin->cmd_table_type = 0;
6143 	mbin->cmd_table_id = htobe32(flow_table_id);
6144 
6145 	mcx_cmdq_mboxes_sign(&mxm, 1);
6146 
6147 	mcx_cmdq_post(sc, cqe, 0);
6148 	error = mcx_cmdq_poll(sc, cqe, 1000);
6149 	if (error != 0) {
6150 		printf("%s: query flow table timeout\n", DEVNAME(sc));
6151 		goto free;
6152 	}
6153 	error = mcx_cmdq_verify(cqe);
6154 	if (error != 0) {
6155 		printf("%s: query flow table reply corrupt\n", DEVNAME(sc));
6156 		goto free;
6157 	}
6158 
6159 	out = mcx_cmdq_out(cqe);
6160 	switch (out->cmd_status) {
6161 	case MCX_CQ_STATUS_OK:
6162 		break;
6163 	default:
6164 		printf("%s: query flow table failed (%x/%x)\n", DEVNAME(sc),
6165 		    out->cmd_status, be32toh(out->cmd_syndrome));
6166 		error = -1;
6167 		goto free;
6168 	}
6169 
6170         mbout = (struct mcx_cmd_query_flow_table_mb_out *)
6171 	    (mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)));
6172 	dump = (uint8_t *)mbout + 8;
6173 	for (i = 0; i < sizeof(struct mcx_flow_table_ctx); i++) {
6174 		printf("%.2x ", dump[i]);
6175 		if (i % 16 == 15)
6176 			printf("\n");
6177 	}
6178 free:
6179 	mcx_cq_mboxes_free(sc, &mxm);
6180 	return (error);
6181 }
6182 int
6183 mcx_dump_flow_table_entry(struct mcx_softc *sc, int flow_table_id, int index)
6184 {
6185 	struct mcx_dmamem mxm;
6186 	struct mcx_cmdq_entry *cqe;
6187 	struct mcx_cmd_query_flow_table_entry_in *in;
6188 	struct mcx_cmd_query_flow_table_entry_mb_in *mbin;
6189 	struct mcx_cmd_query_flow_table_entry_out *out;
6190 	struct mcx_cmd_query_flow_table_entry_mb_out *mbout;
6191 	uint8_t token = mcx_cmdq_token(sc);
6192 	int error;
6193 	int i;
6194 	uint8_t *dump;
6195 
6196 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
6197 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
6198 	    sizeof(*out) + sizeof(*mbout) + 16, token);
6199 
6200 	in = mcx_cmdq_in(cqe);
6201 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_FLOW_TABLE_ENTRY);
6202 	in->cmd_op_mod = htobe16(0);
6203 
6204 	CTASSERT(sizeof(*mbin) <= MCX_CMDQ_MAILBOX_DATASIZE);
6205 	CTASSERT(sizeof(*mbout) <= MCX_CMDQ_MAILBOX_DATASIZE*2);
6206 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2,
6207 	    &cqe->cq_output_ptr, token) != 0) {
6208 		printf(", unable to allocate "
6209 		    "query flow table entry mailboxes\n");
6210 		return (-1);
6211 	}
6212 	cqe->cq_input_ptr = cqe->cq_output_ptr;
6213 
6214 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
6215 	mbin->cmd_table_type = 0;
6216 	mbin->cmd_table_id = htobe32(flow_table_id);
6217 	mbin->cmd_flow_index = htobe32(index);
6218 
6219 	mcx_cmdq_mboxes_sign(&mxm, 1);
6220 
6221 	mcx_cmdq_post(sc, cqe, 0);
6222 	error = mcx_cmdq_poll(sc, cqe, 1000);
6223 	if (error != 0) {
6224 		printf("%s: query flow table entry timeout\n", DEVNAME(sc));
6225 		goto free;
6226 	}
6227 	error = mcx_cmdq_verify(cqe);
6228 	if (error != 0) {
6229 		printf("%s: query flow table entry reply corrupt\n",
6230 		    DEVNAME(sc));
6231 		goto free;
6232 	}
6233 
6234 	out = mcx_cmdq_out(cqe);
6235 	switch (out->cmd_status) {
6236 	case MCX_CQ_STATUS_OK:
6237 		break;
6238 	default:
6239 		printf("%s: query flow table entry failed (%x/%x)\n",
6240 		    DEVNAME(sc), out->cmd_status, be32toh(out->cmd_syndrome));
6241 		error = -1;
6242 		goto free;
6243 	}
6244 
6245         mbout = (struct mcx_cmd_query_flow_table_entry_mb_out *)
6246 	    (mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)));
6247 	dump = (uint8_t *)mbout;
6248 	for (i = 0; i < MCX_CMDQ_MAILBOX_DATASIZE; i++) {
6249 		printf("%.2x ", dump[i]);
6250 		if (i % 16 == 15)
6251 			printf("\n");
6252 	}
6253 
6254 free:
6255 	mcx_cq_mboxes_free(sc, &mxm);
6256 	return (error);
6257 }
6258 
6259 int
6260 mcx_dump_flow_group(struct mcx_softc *sc, int flow_table_id)
6261 {
6262 	struct mcx_dmamem mxm;
6263 	struct mcx_cmdq_entry *cqe;
6264 	struct mcx_cmd_query_flow_group_in *in;
6265 	struct mcx_cmd_query_flow_group_mb_in *mbin;
6266 	struct mcx_cmd_query_flow_group_out *out;
6267 	struct mcx_cmd_query_flow_group_mb_out *mbout;
6268 	uint8_t token = mcx_cmdq_token(sc);
6269 	int error;
6270 	int i;
6271 	uint8_t *dump;
6272 
6273 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
6274 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
6275 	    sizeof(*out) + sizeof(*mbout) + 16, token);
6276 
6277 	in = mcx_cmdq_in(cqe);
6278 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_FLOW_GROUP);
6279 	in->cmd_op_mod = htobe16(0);
6280 
6281 	CTASSERT(sizeof(*mbin) <= MCX_CMDQ_MAILBOX_DATASIZE);
6282 	CTASSERT(sizeof(*mbout) <= MCX_CMDQ_MAILBOX_DATASIZE*2);
6283 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2,
6284 	    &cqe->cq_output_ptr, token) != 0) {
6285 		printf(", unable to allocate query flow group mailboxes\n");
6286 		return (-1);
6287 	}
6288 	cqe->cq_input_ptr = cqe->cq_output_ptr;
6289 
6290 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
6291 	mbin->cmd_table_type = 0;
6292 	mbin->cmd_table_id = htobe32(flow_table_id);
6293 	mbin->cmd_group_id = htobe32(sc->sc_flow_group_id);
6294 
6295 	mcx_cmdq_mboxes_sign(&mxm, 1);
6296 
6297 	mcx_cmdq_post(sc, cqe, 0);
6298 	error = mcx_cmdq_poll(sc, cqe, 1000);
6299 	if (error != 0) {
6300 		printf("%s: query flow group timeout\n", DEVNAME(sc));
6301 		goto free;
6302 	}
6303 	error = mcx_cmdq_verify(cqe);
6304 	if (error != 0) {
6305 		printf("%s: query flow group reply corrupt\n", DEVNAME(sc));
6306 		goto free;
6307 	}
6308 
6309 	out = mcx_cmdq_out(cqe);
6310 	switch (out->cmd_status) {
6311 	case MCX_CQ_STATUS_OK:
6312 		break;
6313 	default:
6314 		printf("%s: query flow group failed (%x/%x)\n", DEVNAME(sc),
6315 		    out->cmd_status, be32toh(out->cmd_syndrome));
6316 		error = -1;
6317 		goto free;
6318 	}
6319 
6320         mbout = (struct mcx_cmd_query_flow_group_mb_out *)
6321 	    (mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)));
6322 	dump = (uint8_t *)mbout;
6323 	for (i = 0; i < MCX_CMDQ_MAILBOX_DATASIZE; i++) {
6324 		printf("%.2x ", dump[i]);
6325 		if (i % 16 == 15)
6326 			printf("\n");
6327 	}
6328 	dump = (uint8_t *)(mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 1)));
6329 	for (i = 0; i < MCX_CMDQ_MAILBOX_DATASIZE; i++) {
6330 		printf("%.2x ", dump[i]);
6331 		if (i % 16 == 15)
6332 			printf("\n");
6333 	}
6334 
6335 free:
6336 	mcx_cq_mboxes_free(sc, &mxm);
6337 	return (error);
6338 }
6339 
6340 static int
6341 mcx_dump_counters(struct mcx_softc *sc)
6342 {
6343 	struct mcx_dmamem mxm;
6344 	struct mcx_cmdq_entry *cqe;
6345 	struct mcx_cmd_query_vport_counters_in *in;
6346 	struct mcx_cmd_query_vport_counters_mb_in *mbin;
6347 	struct mcx_cmd_query_vport_counters_out *out;
6348 	struct mcx_nic_vport_counters *counters;
6349 	int error, token;
6350 
6351 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
6352 	token = mcx_cmdq_token(sc);
6353 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin),
6354 	    sizeof(*out) + sizeof(*counters), token);
6355 
6356 	in = mcx_cmdq_in(cqe);
6357 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_VPORT_COUNTERS);
6358 	in->cmd_op_mod = htobe16(0);
6359 
6360 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
6361 	    &cqe->cq_output_ptr, token) != 0) {
6362 		printf(", unable to allocate "
6363 		    "query nic vport counters mailboxen\n");
6364 		return (-1);
6365 	}
6366 	cqe->cq_input_ptr = cqe->cq_output_ptr;
6367 
6368 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
6369 	mbin->cmd_clear = 0x80;
6370 
6371 	mcx_cmdq_mboxes_sign(&mxm, 1);
6372 	mcx_cmdq_post(sc, cqe, 0);
6373 
6374 	error = mcx_cmdq_poll(sc, cqe, 1000);
6375 	if (error != 0) {
6376 		printf("%s: query nic vport counters timeout\n", DEVNAME(sc));
6377 		goto free;
6378 	}
6379 	if (mcx_cmdq_verify(cqe) != 0) {
6380 		printf("%s: query nic vport counters command corrupt\n",
6381 		    DEVNAME(sc));
6382 		goto free;
6383 	}
6384 
6385 	out = mcx_cmdq_out(cqe);
6386 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
6387 		printf("%s: query nic vport counters failed (%x, %x)\n",
6388 		    DEVNAME(sc), out->cmd_status, betoh32(out->cmd_syndrome));
6389 		error = -1;
6390 		goto free;
6391 	}
6392 
6393 	counters = (struct mcx_nic_vport_counters *)
6394 	    (mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)));
6395 	if (counters->rx_bcast.packets + counters->tx_bcast.packets +
6396 	    counters->rx_ucast.packets + counters->tx_ucast.packets +
6397 	    counters->rx_err.packets + counters->tx_err.packets)
6398 		printf("%s: err %llx/%llx uc %llx/%llx bc %llx/%llx\n",
6399 		    DEVNAME(sc),
6400 		    betoh64(counters->tx_err.packets),
6401 		    betoh64(counters->rx_err.packets),
6402 		    betoh64(counters->tx_ucast.packets),
6403 		    betoh64(counters->rx_ucast.packets),
6404 		    betoh64(counters->tx_bcast.packets),
6405 		    betoh64(counters->rx_bcast.packets));
6406 free:
6407 	mcx_dmamem_free(sc, &mxm);
6408 
6409 	return (error);
6410 }
6411 
6412 static int
6413 mcx_dump_flow_counter(struct mcx_softc *sc, int index, const char *what)
6414 {
6415 	struct mcx_dmamem mxm;
6416 	struct mcx_cmdq_entry *cqe;
6417 	struct mcx_cmd_query_flow_counter_in *in;
6418 	struct mcx_cmd_query_flow_counter_mb_in *mbin;
6419 	struct mcx_cmd_query_flow_counter_out *out;
6420 	struct mcx_counter *counters;
6421 	int error, token;
6422 
6423 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
6424 	token = mcx_cmdq_token(sc);
6425 	mcx_cmdq_init(sc, cqe, sizeof(*in) + sizeof(*mbin), sizeof(*out) +
6426 	    sizeof(*counters), token);
6427 
6428 	in = mcx_cmdq_in(cqe);
6429 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_FLOW_COUNTER);
6430 	in->cmd_op_mod = htobe16(0);
6431 
6432 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 1,
6433 	    &cqe->cq_output_ptr, token) != 0) {
6434 		printf(", unable to allocate query flow counter mailboxen\n");
6435 		return (-1);
6436 	}
6437 	cqe->cq_input_ptr = cqe->cq_output_ptr;
6438 	mbin = mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0));
6439 	mbin->cmd_flow_counter_id = htobe16(sc->sc_flow_counter_id[index]);
6440 	mbin->cmd_clear = 0x80;
6441 
6442 	mcx_cmdq_mboxes_sign(&mxm, 1);
6443 	mcx_cmdq_post(sc, cqe, 0);
6444 
6445 	error = mcx_cmdq_poll(sc, cqe, 1000);
6446 	if (error != 0) {
6447 		printf("%s: query flow counter timeout\n", DEVNAME(sc));
6448 		goto free;
6449 	}
6450 	if (mcx_cmdq_verify(cqe) != 0) {
6451 		printf("%s: query flow counter command corrupt\n", DEVNAME(sc));
6452 		goto free;
6453 	}
6454 
6455 	out = mcx_cmdq_out(cqe);
6456 	if (out->cmd_status != MCX_CQ_STATUS_OK) {
6457 		printf("%s: query flow counter failed (%x, %x)\n", DEVNAME(sc),
6458 		    out->cmd_status, betoh32(out->cmd_syndrome));
6459 		error = -1;
6460 		goto free;
6461 	}
6462 
6463 	counters = (struct mcx_counter *)
6464 	    (mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)));
6465 	if (counters->packets)
6466 		printf("%s: %s inflow %llx\n", DEVNAME(sc), what,
6467 		    betoh64(counters->packets));
6468 free:
6469 	mcx_dmamem_free(sc, &mxm);
6470 
6471 	return (error);
6472 }
6473 
6474 #endif
6475 
6476 #if NKSTAT > 0
6477 
6478 int
6479 mcx_query_rq(struct mcx_softc *sc, struct mcx_rx *rx, struct mcx_rq_ctx *rq_ctx)
6480 {
6481 	struct mcx_dmamem mxm;
6482 	struct mcx_cmdq_entry *cqe;
6483 	struct mcx_cmd_query_rq_in *in;
6484 	struct mcx_cmd_query_rq_out *out;
6485 	struct mcx_cmd_query_rq_mb_out *mbout;
6486 	uint8_t token = mcx_cmdq_token(sc);
6487 	int error;
6488 
6489 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
6490 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out) + sizeof(*mbout) + 16,
6491 	    token);
6492 
6493 	in = mcx_cmdq_in(cqe);
6494 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_RQ);
6495 	in->cmd_op_mod = htobe16(0);
6496 	in->cmd_rqn = htobe32(rx->rx_rqn);
6497 
6498 	CTASSERT(sizeof(*mbout) <= MCX_CMDQ_MAILBOX_DATASIZE*2);
6499 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2,
6500 	    &cqe->cq_output_ptr, token) != 0) {
6501 		printf("%s: unable to allocate query rq mailboxes\n", DEVNAME(sc));
6502 		return (-1);
6503 	}
6504 
6505 	mcx_cmdq_mboxes_sign(&mxm, 1);
6506 
6507 	mcx_cmdq_post(sc, cqe, 0);
6508 	error = mcx_cmdq_poll(sc, cqe, 1000);
6509 	if (error != 0) {
6510 		printf("%s: query rq timeout\n", DEVNAME(sc));
6511 		goto free;
6512 	}
6513 	error = mcx_cmdq_verify(cqe);
6514 	if (error != 0) {
6515 		printf("%s: query rq reply corrupt\n", DEVNAME(sc));
6516 		goto free;
6517 	}
6518 
6519 	out = mcx_cmdq_out(cqe);
6520 	switch (out->cmd_status) {
6521 	case MCX_CQ_STATUS_OK:
6522 		break;
6523 	default:
6524 		printf("%s: query rq failed (%x/%x)\n", DEVNAME(sc),
6525 		    out->cmd_status, be32toh(out->cmd_syndrome));
6526 		error = -1;
6527 		goto free;
6528 	}
6529 
6530         mbout = (struct mcx_cmd_query_rq_mb_out *)
6531 	    (mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)));
6532 	memcpy(rq_ctx, &mbout->cmd_ctx, sizeof(*rq_ctx));
6533 
6534 free:
6535 	mcx_cq_mboxes_free(sc, &mxm);
6536 	return (error);
6537 }
6538 
6539 int
6540 mcx_query_sq(struct mcx_softc *sc, struct mcx_tx *tx, struct mcx_sq_ctx *sq_ctx)
6541 {
6542 	struct mcx_dmamem mxm;
6543 	struct mcx_cmdq_entry *cqe;
6544 	struct mcx_cmd_query_sq_in *in;
6545 	struct mcx_cmd_query_sq_out *out;
6546 	struct mcx_cmd_query_sq_mb_out *mbout;
6547 	uint8_t token = mcx_cmdq_token(sc);
6548 	int error;
6549 
6550 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
6551 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out) + sizeof(*mbout) + 16,
6552 	    token);
6553 
6554 	in = mcx_cmdq_in(cqe);
6555 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_SQ);
6556 	in->cmd_op_mod = htobe16(0);
6557 	in->cmd_sqn = htobe32(tx->tx_sqn);
6558 
6559 	CTASSERT(sizeof(*mbout) <= MCX_CMDQ_MAILBOX_DATASIZE*2);
6560 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2,
6561 	    &cqe->cq_output_ptr, token) != 0) {
6562 		printf("%s: unable to allocate query sq mailboxes\n", DEVNAME(sc));
6563 		return (-1);
6564 	}
6565 
6566 	mcx_cmdq_mboxes_sign(&mxm, 1);
6567 
6568 	mcx_cmdq_post(sc, cqe, 0);
6569 	error = mcx_cmdq_poll(sc, cqe, 1000);
6570 	if (error != 0) {
6571 		printf("%s: query sq timeout\n", DEVNAME(sc));
6572 		goto free;
6573 	}
6574 	error = mcx_cmdq_verify(cqe);
6575 	if (error != 0) {
6576 		printf("%s: query sq reply corrupt\n", DEVNAME(sc));
6577 		goto free;
6578 	}
6579 
6580 	out = mcx_cmdq_out(cqe);
6581 	switch (out->cmd_status) {
6582 	case MCX_CQ_STATUS_OK:
6583 		break;
6584 	default:
6585 		printf("%s: query sq failed (%x/%x)\n", DEVNAME(sc),
6586 		    out->cmd_status, be32toh(out->cmd_syndrome));
6587 		error = -1;
6588 		goto free;
6589 	}
6590 
6591         mbout = (struct mcx_cmd_query_sq_mb_out *)
6592 	    (mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)));
6593 	memcpy(sq_ctx, &mbout->cmd_ctx, sizeof(*sq_ctx));
6594 
6595 free:
6596 	mcx_cq_mboxes_free(sc, &mxm);
6597 	return (error);
6598 }
6599 
6600 int
6601 mcx_query_cq(struct mcx_softc *sc, struct mcx_cq *cq, struct mcx_cq_ctx *cq_ctx)
6602 {
6603 	struct mcx_dmamem mxm;
6604 	struct mcx_cmdq_entry *cqe;
6605 	struct mcx_cmd_query_cq_in *in;
6606 	struct mcx_cmd_query_cq_out *out;
6607 	struct mcx_cq_ctx *ctx;
6608 	uint8_t token = mcx_cmdq_token(sc);
6609 	int error;
6610 
6611 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
6612 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out) + sizeof(*ctx) + 16,
6613 	    token);
6614 
6615 	in = mcx_cmdq_in(cqe);
6616 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_CQ);
6617 	in->cmd_op_mod = htobe16(0);
6618 	in->cmd_cqn = htobe32(cq->cq_n);
6619 
6620 	CTASSERT(sizeof(*ctx) <= MCX_CMDQ_MAILBOX_DATASIZE*2);
6621 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2,
6622 	    &cqe->cq_output_ptr, token) != 0) {
6623 		printf("%s: unable to allocate query cq mailboxes\n",
6624 		    DEVNAME(sc));
6625 		return (-1);
6626 	}
6627 
6628 	mcx_cmdq_mboxes_sign(&mxm, 1);
6629 
6630 	mcx_cmdq_post(sc, cqe, 0);
6631 	error = mcx_cmdq_poll(sc, cqe, 1000);
6632 	if (error != 0) {
6633 		printf("%s: query cq timeout\n", DEVNAME(sc));
6634 		goto free;
6635 	}
6636 	if (mcx_cmdq_verify(cqe) != 0) {
6637 		printf("%s: query cq reply corrupt\n", DEVNAME(sc));
6638 		goto free;
6639 	}
6640 
6641 	out = mcx_cmdq_out(cqe);
6642 	switch (out->cmd_status) {
6643 	case MCX_CQ_STATUS_OK:
6644 		break;
6645 	default:
6646 		printf("%s: query qc failed (%x/%x)\n", DEVNAME(sc),
6647 		    out->cmd_status, be32toh(out->cmd_syndrome));
6648 		error = -1;
6649 		goto free;
6650 	}
6651 
6652 	ctx = (struct mcx_cq_ctx *)(mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)));
6653 	memcpy(cq_ctx, ctx, sizeof(*cq_ctx));
6654 free:
6655 	mcx_dmamem_free(sc, &mxm);
6656 	return (error);
6657 }
6658 
6659 int
6660 mcx_query_eq(struct mcx_softc *sc, struct mcx_eq *eq, struct mcx_eq_ctx *eq_ctx)
6661 {
6662 	struct mcx_dmamem mxm;
6663 	struct mcx_cmdq_entry *cqe;
6664 	struct mcx_cmd_query_eq_in *in;
6665 	struct mcx_cmd_query_eq_out *out;
6666 	struct mcq_eq_ctx *ctx;
6667 	uint8_t token = mcx_cmdq_token(sc);
6668 	int error;
6669 
6670 	cqe = MCX_DMA_KVA(&sc->sc_cmdq_mem);
6671 	mcx_cmdq_init(sc, cqe, sizeof(*in), sizeof(*out) + sizeof(*ctx) + 16,
6672 	    token);
6673 
6674 	in = mcx_cmdq_in(cqe);
6675 	in->cmd_opcode = htobe16(MCX_CMD_QUERY_EQ);
6676 	in->cmd_op_mod = htobe16(0);
6677 	in->cmd_eqn = htobe32(eq->eq_n);
6678 
6679 	CTASSERT(sizeof(*ctx) <= MCX_CMDQ_MAILBOX_DATASIZE*2);
6680 	if (mcx_cmdq_mboxes_alloc(sc, &mxm, 2,
6681 	    &cqe->cq_output_ptr, token) != 0) {
6682 		printf("%s: unable to allocate query eq mailboxes\n",
6683 		    DEVNAME(sc));
6684 		return (-1);
6685 	}
6686 
6687 	mcx_cmdq_mboxes_sign(&mxm, 1);
6688 
6689 	mcx_cmdq_post(sc, cqe, 0);
6690 	error = mcx_cmdq_poll(sc, cqe, 1000);
6691 	if (error != 0) {
6692 		printf("%s: query eq timeout\n", DEVNAME(sc));
6693 		goto free;
6694 	}
6695 	if (mcx_cmdq_verify(cqe) != 0) {
6696 		printf("%s: query eq reply corrupt\n", DEVNAME(sc));
6697 		goto free;
6698 	}
6699 
6700 	out = mcx_cmdq_out(cqe);
6701 	switch (out->cmd_status) {
6702 	case MCX_CQ_STATUS_OK:
6703 		break;
6704 	default:
6705 		printf("%s: query eq failed (%x/%x)\n", DEVNAME(sc),
6706 		    out->cmd_status, be32toh(out->cmd_syndrome));
6707 		error = -1;
6708 		goto free;
6709 	}
6710 
6711 	ctx = (struct mcx_eq_ctx *)(mcx_cq_mbox_data(mcx_cq_mbox(&mxm, 0)));
6712 	memcpy(eq_ctx, ctx, sizeof(*eq_ctx));
6713 free:
6714 	mcx_dmamem_free(sc, &mxm);
6715 	return (error);
6716 }
6717 
6718 #endif /* NKSTAT > 0 */
6719 
6720 
6721 static inline unsigned int
6722 mcx_rx_fill_slots(struct mcx_softc *sc, struct mcx_rx *rx, uint nslots)
6723 {
6724 	struct mcx_rq_entry *ring, *rqe;
6725 	struct mcx_slot *ms;
6726 	struct mbuf *m;
6727 	uint slot, p, fills;
6728 
6729 	ring = MCX_DMA_KVA(&rx->rx_rq_mem);
6730 	p = rx->rx_prod;
6731 
6732 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&rx->rx_rq_mem),
6733 	    0, MCX_DMA_LEN(&rx->rx_rq_mem), BUS_DMASYNC_POSTWRITE);
6734 
6735 	slot = (p % (1 << MCX_LOG_RQ_SIZE));
6736 	rqe = ring;
6737 	for (fills = 0; fills < nslots; fills++) {
6738 		slot = p % (1 << MCX_LOG_RQ_SIZE);
6739 
6740 		ms = &rx->rx_slots[slot];
6741 		rqe = &ring[slot];
6742 
6743 		m = NULL;
6744 		MGETHDR(m, M_DONTWAIT, MT_DATA);
6745 		if (m == NULL)
6746 			break;
6747 
6748 		MCLGET(m, M_DONTWAIT);
6749 		if ((m->m_flags & M_EXT) == 0) {
6750 			m_freem(m);
6751 			break;
6752 		}
6753 
6754 		m->m_len = m->m_pkthdr.len = sc->sc_hardmtu;
6755 		m_adj(m, m->m_ext.ext_size - sc->sc_rxbufsz);
6756 		m_adj(m, ETHER_ALIGN);
6757 
6758 		if (bus_dmamap_load_mbuf(sc->sc_dmat, ms->ms_map, m,
6759 		    BUS_DMA_NOWAIT) != 0) {
6760 			m_freem(m);
6761 			break;
6762 		}
6763 		bus_dmamap_sync(sc->sc_dmat, ms->ms_map, 0, ms->ms_map->dm_mapsize, BUS_DMASYNC_PREREAD);
6764 		ms->ms_m = m;
6765 
6766 		be32enc(&rqe->rqe_byte_count, ms->ms_map->dm_segs[0].ds_len);
6767 		be64enc(&rqe->rqe_addr, ms->ms_map->dm_segs[0].ds_addr);
6768 		be32enc(&rqe->rqe_lkey, sc->sc_lkey);
6769 
6770 		p++;
6771 	}
6772 
6773 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&rx->rx_rq_mem),
6774 	    0, MCX_DMA_LEN(&rx->rx_rq_mem), BUS_DMASYNC_PREWRITE);
6775 
6776 	rx->rx_prod = p;
6777 
6778 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_doorbell_mem),
6779 	    rx->rx_doorbell, sizeof(uint32_t), BUS_DMASYNC_POSTWRITE);
6780 	be32enc(MCX_DMA_OFF(&sc->sc_doorbell_mem, rx->rx_doorbell),
6781 	    p & MCX_WQ_DOORBELL_MASK);
6782 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_doorbell_mem),
6783 	    rx->rx_doorbell, sizeof(uint32_t), BUS_DMASYNC_PREWRITE);
6784 
6785 	return (nslots - fills);
6786 }
6787 
6788 static int
6789 mcx_rx_fill(struct mcx_softc *sc, struct mcx_rx *rx)
6790 {
6791 	u_int slots;
6792 
6793 	slots = mcx_rxr_get(&rx->rx_rxr, (1 << MCX_LOG_RQ_SIZE));
6794 	if (slots == 0)
6795 		return (1);
6796 
6797 	slots = mcx_rx_fill_slots(sc, rx, slots);
6798 	mcx_rxr_put(&rx->rx_rxr, slots);
6799 	return (0);
6800 }
6801 
6802 void
6803 mcx_refill(void *xrx)
6804 {
6805 	struct mcx_rx *rx = xrx;
6806 	struct mcx_softc *sc = rx->rx_softc;
6807 
6808 	mcx_rx_fill(sc, rx);
6809 
6810 	if (mcx_rxr_inuse(&rx->rx_rxr) == 0)
6811 		callout_schedule(&rx->rx_refill, 1);
6812 }
6813 
6814 static int
6815 mcx_process_txeof(struct mcx_softc *sc, struct mcx_tx *tx,
6816     struct mcx_cq_entry *cqe)
6817 {
6818 	struct mcx_slot *ms;
6819 	bus_dmamap_t map;
6820 	int slot, slots;
6821 
6822 	slot = be16toh(cqe->cq_wqe_count) % (1 << MCX_LOG_SQ_SIZE);
6823 
6824 	ms = &tx->tx_slots[slot];
6825 	map = ms->ms_map;
6826 	bus_dmamap_sync(sc->sc_dmat, map, 0, map->dm_mapsize,
6827 	    BUS_DMASYNC_POSTWRITE);
6828 
6829 	slots = 1;
6830 	if (map->dm_nsegs > 1)
6831 		slots += (map->dm_nsegs+2) / MCX_SQ_SEGS_PER_SLOT;
6832 
6833 	bus_dmamap_unload(sc->sc_dmat, map);
6834 	m_freem(ms->ms_m);
6835 	ms->ms_m = NULL;
6836 
6837 	return (slots);
6838 }
6839 
6840 static uint64_t
6841 mcx_uptime(void)
6842 {
6843 	struct timespec ts;
6844 
6845 	nanouptime(&ts);
6846 
6847 	return ((uint64_t)ts.tv_sec * 1000000000 + (uint64_t)ts.tv_nsec);
6848 }
6849 
6850 static void
6851 mcx_calibrate_first(struct mcx_softc *sc)
6852 {
6853 	struct mcx_calibration *c = &sc->sc_calibration[0];
6854 	int s;
6855 
6856 	sc->sc_calibration_gen = 0;
6857 
6858 	s = splhigh(); /* crit_enter? */
6859 	c->c_ubase = mcx_uptime();
6860 	c->c_tbase = mcx_timer(sc);
6861 	splx(s);
6862 	c->c_ratio = 0;
6863 
6864 #if notyet
6865 	callout_schedule(&sc->sc_calibrate, MCX_CALIBRATE_FIRST * hz);
6866 #endif
6867 }
6868 
6869 #define MCX_TIMESTAMP_SHIFT 24
6870 
6871 static void
6872 mcx_calibrate(void *arg)
6873 {
6874 	struct mcx_softc *sc = arg;
6875 	struct mcx_calibration *nc, *pc;
6876 	uint64_t udiff, tdiff;
6877 	unsigned int gen;
6878 	int s;
6879 
6880 	if (!ISSET(sc->sc_ec.ec_if.if_flags, IFF_RUNNING))
6881 		return;
6882 
6883 	callout_schedule(&sc->sc_calibrate, MCX_CALIBRATE_NORMAL * hz);
6884 
6885 	gen = sc->sc_calibration_gen;
6886 	pc = &sc->sc_calibration[gen % __arraycount(sc->sc_calibration)];
6887 	gen++;
6888 	nc = &sc->sc_calibration[gen % __arraycount(sc->sc_calibration)];
6889 
6890 	nc->c_uptime = pc->c_ubase;
6891 	nc->c_timestamp = pc->c_tbase;
6892 
6893 	s = splhigh(); /* crit_enter? */
6894 	nc->c_ubase = mcx_uptime();
6895 	nc->c_tbase = mcx_timer(sc);
6896 	splx(s);
6897 
6898 	udiff = nc->c_ubase - nc->c_uptime;
6899 	tdiff = nc->c_tbase - nc->c_timestamp;
6900 
6901 	/*
6902 	 * udiff is the wall clock time between calibration ticks,
6903 	 * which should be 32 seconds or 32 billion nanoseconds. if
6904 	 * we squint, 1 billion nanoseconds is kind of like a 32 bit
6905 	 * number, so 32 billion should still have a lot of high bits
6906 	 * spare. we use this space by shifting the nanoseconds up
6907 	 * 24 bits so we have a nice big number to divide by the
6908 	 * number of mcx timer ticks.
6909 	 */
6910 	nc->c_ratio = (udiff << MCX_TIMESTAMP_SHIFT) / tdiff;
6911 
6912 	membar_producer();
6913 	sc->sc_calibration_gen = gen;
6914 }
6915 
6916 static int
6917 mcx_process_rx(struct mcx_softc *sc, struct mcx_rx *rx,
6918     struct mcx_cq_entry *cqe, struct mcx_mbufq *mq,
6919     const struct mcx_calibration *c)
6920 {
6921 	struct ifnet *ifp = &sc->sc_ec.ec_if;
6922 	struct mcx_slot *ms;
6923 	struct mbuf *m;
6924 	uint32_t flags;
6925 	int slot;
6926 
6927 	slot = be16toh(cqe->cq_wqe_count) % (1 << MCX_LOG_RQ_SIZE);
6928 
6929 	ms = &rx->rx_slots[slot];
6930 	bus_dmamap_sync(sc->sc_dmat, ms->ms_map, 0, ms->ms_map->dm_mapsize,
6931 	    BUS_DMASYNC_POSTREAD);
6932 	bus_dmamap_unload(sc->sc_dmat, ms->ms_map);
6933 
6934 	m = ms->ms_m;
6935 	ms->ms_m = NULL;
6936 
6937 	m_set_rcvif(m, &sc->sc_ec.ec_if);
6938 	m->m_pkthdr.len = m->m_len = be32dec(&cqe->cq_byte_cnt);
6939 
6940 #if 0
6941 	if (cqe->cq_rx_hash_type) {
6942 		m->m_pkthdr.ph_flowid = be32toh(cqe->cq_rx_hash);
6943 		m->m_pkthdr.csum_flags |= M_FLOWID;
6944 	}
6945 #endif
6946 
6947 	flags = be32dec(&cqe->cq_flags);
6948 	if (flags & MCX_CQ_ENTRY_FLAGS_L3_OK) {
6949 		if (ifp->if_capenable & IFCAP_CSUM_IPv4_Rx)
6950 			m->m_pkthdr.csum_flags |= M_CSUM_IPv4;
6951 	}
6952 	if (flags & MCX_CQ_ENTRY_FLAGS_L4_OK) {
6953 		if (ifp->if_capenable & IFCAP_CSUM_TCPv4_Rx)
6954 			m->m_pkthdr.csum_flags |= M_CSUM_TCPv4;
6955 		if (ifp->if_capenable & IFCAP_CSUM_TCPv6_Rx)
6956 			m->m_pkthdr.csum_flags |= M_CSUM_TCPv6;
6957 		if (ifp->if_capenable & IFCAP_CSUM_UDPv4_Rx)
6958 			m->m_pkthdr.csum_flags |= M_CSUM_UDPv4;
6959 		if (ifp->if_capenable & IFCAP_CSUM_UDPv6_Rx)
6960 			m->m_pkthdr.csum_flags |= M_CSUM_UDPv6;
6961 	}
6962 	if (flags & MCX_CQ_ENTRY_FLAGS_CV) {
6963 		vlan_set_tag(m, flags & MCX_CQ_ENTRY_FLAGS_VLAN_MASK);
6964 	}
6965 
6966 #if notyet
6967 	if (ISSET(sc->sc_ec.ec_if.if_flags, IFF_LINK0) && c->c_ratio) {
6968 		uint64_t t = be64dec(&cqe->cq_timestamp);
6969 		t -= c->c_timestamp;
6970 		t *= c->c_ratio;
6971 		t >>= MCX_TIMESTAMP_SHIFT;
6972 		t += c->c_uptime;
6973 
6974 		m->m_pkthdr.ph_timestamp = t;
6975 		SET(m->m_pkthdr.csum_flags, M_TIMESTAMP);
6976 	}
6977 #endif
6978 
6979 	MBUFQ_ENQUEUE(mq, m);
6980 
6981 	return (1);
6982 }
6983 
6984 static struct mcx_cq_entry *
6985 mcx_next_cq_entry(struct mcx_softc *sc, struct mcx_cq *cq)
6986 {
6987 	struct mcx_cq_entry *cqe;
6988 	int next;
6989 
6990 	cqe = (struct mcx_cq_entry *)MCX_DMA_KVA(&cq->cq_mem);
6991 	next = cq->cq_cons % (1 << MCX_LOG_CQ_SIZE);
6992 
6993 	if ((cqe[next].cq_opcode_owner & MCX_CQ_ENTRY_FLAG_OWNER) ==
6994 	    ((cq->cq_cons >> MCX_LOG_CQ_SIZE) & 1)) {
6995 		return (&cqe[next]);
6996 	}
6997 
6998 	return (NULL);
6999 }
7000 
7001 static void
7002 mcx_arm_cq(struct mcx_softc *sc, struct mcx_cq *cq, int uar)
7003 {
7004 	struct mcx_cq_doorbell *db;
7005 	bus_size_t offset;
7006 	uint32_t val;
7007 	uint64_t uval;
7008 
7009 	val = ((cq->cq_count) & 3) << MCX_CQ_DOORBELL_ARM_CMD_SN_SHIFT;
7010 	val |= (cq->cq_cons & MCX_CQ_DOORBELL_ARM_CI_MASK);
7011 
7012 	db = MCX_DMA_OFF(&sc->sc_doorbell_mem, cq->cq_doorbell);
7013 
7014 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_doorbell_mem),
7015 	    cq->cq_doorbell, sizeof(*db), BUS_DMASYNC_POSTWRITE);
7016 
7017 	be32enc(&db->db_update_ci, cq->cq_cons & MCX_CQ_DOORBELL_ARM_CI_MASK);
7018 	be32enc(&db->db_arm_ci, val);
7019 
7020 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_doorbell_mem),
7021 	    cq->cq_doorbell, sizeof(*db), BUS_DMASYNC_PREWRITE);
7022 
7023 	offset = (MCX_PAGE_SIZE * uar) + MCX_UAR_CQ_DOORBELL;
7024 
7025 	uval = (uint64_t)val << 32;
7026 	uval |= cq->cq_n;
7027 
7028 	bus_space_write_8(sc->sc_memt, sc->sc_memh, offset, htobe64(uval));
7029 	mcx_bar(sc, offset, sizeof(uval), BUS_SPACE_BARRIER_WRITE);
7030 }
7031 
7032 void
7033 mcx_process_cq(struct mcx_softc *sc, struct mcx_queues *q, struct mcx_cq *cq)
7034 {
7035 	struct mcx_rx *rx = &q->q_rx;
7036 	struct mcx_tx *tx = &q->q_tx;
7037 	struct ifnet *ifp = &sc->sc_ec.ec_if;
7038 	const struct mcx_calibration *c;
7039 	unsigned int gen;
7040 	struct mcx_cq_entry *cqe;
7041 	struct mcx_mbufq mq;
7042 	struct mbuf *m;
7043 	int rxfree, txfree;
7044 
7045 	MBUFQ_INIT(&mq);
7046 
7047 	gen = sc->sc_calibration_gen;
7048 	membar_consumer();
7049 	c = &sc->sc_calibration[gen % __arraycount(sc->sc_calibration)];
7050 
7051 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&cq->cq_mem),
7052 	    0, MCX_DMA_LEN(&cq->cq_mem), BUS_DMASYNC_POSTREAD);
7053 
7054 	rxfree = 0;
7055 	txfree = 0;
7056 	while ((cqe = mcx_next_cq_entry(sc, cq))) {
7057 		uint8_t opcode;
7058 		opcode = (cqe->cq_opcode_owner >> MCX_CQ_ENTRY_OPCODE_SHIFT);
7059 		switch (opcode) {
7060 		case MCX_CQ_ENTRY_OPCODE_REQ:
7061 			txfree += mcx_process_txeof(sc, tx, cqe);
7062 			break;
7063 		case MCX_CQ_ENTRY_OPCODE_SEND:
7064 			rxfree += mcx_process_rx(sc, rx, cqe, &mq, c);
7065 			break;
7066 		case MCX_CQ_ENTRY_OPCODE_REQ_ERR:
7067 		case MCX_CQ_ENTRY_OPCODE_SEND_ERR:
7068 			/* uint8_t *cqp = (uint8_t *)cqe; */
7069 			/* printf("%s: cq completion error: %x\n",
7070 			    DEVNAME(sc), cqp[0x37]); */
7071 			break;
7072 
7073 		default:
7074 			/* printf("%s: cq completion opcode %x??\n",
7075 			    DEVNAME(sc), opcode); */
7076 			break;
7077 		}
7078 
7079 		cq->cq_cons++;
7080 	}
7081 
7082 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&cq->cq_mem),
7083 	    0, MCX_DMA_LEN(&cq->cq_mem), BUS_DMASYNC_PREREAD);
7084 
7085 	if (rxfree > 0) {
7086 		mcx_rxr_put(&rx->rx_rxr, rxfree);
7087 		while (MBUFQ_FIRST(&mq) != NULL) {
7088 			MBUFQ_DEQUEUE(&mq, m);
7089 			if_percpuq_enqueue(ifp->if_percpuq, m);
7090 		}
7091 
7092 		mcx_rx_fill(sc, rx);
7093 		if (mcx_rxr_inuse(&rx->rx_rxr) == 0)
7094 			callout_schedule(&rx->rx_refill, 1);
7095 	}
7096 
7097 	cq->cq_count++;
7098 	mcx_arm_cq(sc, cq, q->q_uar);
7099 
7100 	if (txfree > 0) {
7101 		tx->tx_cons += txfree;
7102 		if_schedule_deferred_start(ifp);
7103 	}
7104 }
7105 
7106 
7107 static void
7108 mcx_arm_eq(struct mcx_softc *sc, struct mcx_eq *eq, int uar)
7109 {
7110 	bus_size_t offset;
7111 	uint32_t val;
7112 
7113 	offset = (MCX_PAGE_SIZE * uar) + MCX_UAR_EQ_DOORBELL_ARM;
7114 	val = (eq->eq_n << 24) | (eq->eq_cons & 0xffffff);
7115 
7116 	mcx_wr(sc, offset, val);
7117 	mcx_bar(sc, offset, sizeof(val), BUS_SPACE_BARRIER_WRITE);
7118 }
7119 
7120 static struct mcx_eq_entry *
7121 mcx_next_eq_entry(struct mcx_softc *sc, struct mcx_eq *eq)
7122 {
7123 	struct mcx_eq_entry *eqe;
7124 	int next;
7125 
7126 	eqe = (struct mcx_eq_entry *)MCX_DMA_KVA(&eq->eq_mem);
7127 	next = eq->eq_cons % (1 << MCX_LOG_EQ_SIZE);
7128 	if ((eqe[next].eq_owner & 1) ==
7129 	    ((eq->eq_cons >> MCX_LOG_EQ_SIZE) & 1)) {
7130 		eq->eq_cons++;
7131 		return (&eqe[next]);
7132 	}
7133 	return (NULL);
7134 }
7135 
7136 int
7137 mcx_admin_intr(void *xsc)
7138 {
7139 	struct mcx_softc *sc = (struct mcx_softc *)xsc;
7140 	struct mcx_eq *eq = &sc->sc_admin_eq;
7141 	struct mcx_eq_entry *eqe;
7142 
7143 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&eq->eq_mem),
7144 	    0, MCX_DMA_LEN(&eq->eq_mem), BUS_DMASYNC_POSTREAD);
7145 
7146 	while ((eqe = mcx_next_eq_entry(sc, eq)) != NULL) {
7147 		switch (eqe->eq_event_type) {
7148 		case MCX_EVENT_TYPE_LAST_WQE:
7149 			/* printf("%s: last wqe reached?\n", DEVNAME(sc)); */
7150 			break;
7151 
7152 		case MCX_EVENT_TYPE_CQ_ERROR:
7153 			/* printf("%s: cq error\n", DEVNAME(sc)); */
7154 			break;
7155 
7156 		case MCX_EVENT_TYPE_CMD_COMPLETION:
7157 			/* wakeup probably */
7158 			break;
7159 
7160 		case MCX_EVENT_TYPE_PORT_CHANGE:
7161 			workqueue_enqueue(sc->sc_workq, &sc->sc_port_change, NULL);
7162 			break;
7163 
7164 		default:
7165 			/* printf("%s: something happened\n", DEVNAME(sc)); */
7166 			break;
7167 		}
7168 	}
7169 
7170 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&eq->eq_mem),
7171 	    0, MCX_DMA_LEN(&eq->eq_mem), BUS_DMASYNC_PREREAD);
7172 
7173 	mcx_arm_eq(sc, eq, sc->sc_uar);
7174 
7175 	return (1);
7176 }
7177 
7178 int
7179 mcx_cq_intr(void *xq)
7180 {
7181 	struct mcx_queues *q = (struct mcx_queues *)xq;
7182 	struct mcx_softc *sc = q->q_sc;
7183 	struct mcx_eq *eq = &q->q_eq;
7184 	struct mcx_eq_entry *eqe;
7185 	int cqn;
7186 
7187 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&eq->eq_mem),
7188 	    0, MCX_DMA_LEN(&eq->eq_mem), BUS_DMASYNC_POSTREAD);
7189 
7190 	while ((eqe = mcx_next_eq_entry(sc, eq)) != NULL) {
7191 		switch (eqe->eq_event_type) {
7192 		case MCX_EVENT_TYPE_COMPLETION:
7193 			cqn = be32toh(eqe->eq_event_data[6]);
7194 			if (cqn == q->q_cq.cq_n)
7195 				mcx_process_cq(sc, q, &q->q_cq);
7196 			break;
7197 		}
7198 	}
7199 
7200 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&eq->eq_mem),
7201 	    0, MCX_DMA_LEN(&eq->eq_mem), BUS_DMASYNC_PREREAD);
7202 
7203 	mcx_arm_eq(sc, eq, q->q_uar);
7204 
7205 	return (1);
7206 }
7207 
7208 static void
7209 mcx_free_slots(struct mcx_softc *sc, struct mcx_slot *slots, int allocated,
7210     int total)
7211 {
7212 	struct mcx_slot *ms;
7213 
7214 	int i = allocated;
7215 	while (i-- > 0) {
7216 		ms = &slots[i];
7217 		bus_dmamap_destroy(sc->sc_dmat, ms->ms_map);
7218 		if (ms->ms_m != NULL)
7219 			m_freem(ms->ms_m);
7220 	}
7221 	kmem_free(slots, total * sizeof(*ms));
7222 }
7223 
7224 static int
7225 mcx_queue_up(struct mcx_softc *sc, struct mcx_queues *q)
7226 {
7227 	struct mcx_rx *rx;
7228 	struct mcx_tx *tx;
7229 	struct mcx_slot *ms;
7230 	int i;
7231 
7232 	rx = &q->q_rx;
7233 	rx->rx_slots = kmem_zalloc(sizeof(*ms) * (1 << MCX_LOG_RQ_SIZE),
7234 	    KM_SLEEP);
7235 
7236 	for (i = 0; i < (1 << MCX_LOG_RQ_SIZE); i++) {
7237 		ms = &rx->rx_slots[i];
7238 		if (bus_dmamap_create(sc->sc_dmat, sc->sc_hardmtu, 1,
7239 		    sc->sc_hardmtu, 0,
7240 		    BUS_DMA_WAITOK | BUS_DMA_ALLOCNOW,
7241 		    &ms->ms_map) != 0) {
7242 			printf("%s: failed to allocate rx dma maps\n",
7243 			    DEVNAME(sc));
7244 			goto destroy_rx_slots;
7245 		}
7246 	}
7247 
7248 	tx = &q->q_tx;
7249 	tx->tx_slots = kmem_zalloc(sizeof(*ms) * (1 << MCX_LOG_SQ_SIZE),
7250 	     KM_SLEEP);
7251 
7252 	for (i = 0; i < (1 << MCX_LOG_SQ_SIZE); i++) {
7253 		ms = &tx->tx_slots[i];
7254 		if (bus_dmamap_create(sc->sc_dmat, sc->sc_hardmtu,
7255 		    MCX_SQ_MAX_SEGMENTS, sc->sc_hardmtu, 0,
7256 		    BUS_DMA_WAITOK | BUS_DMA_ALLOCNOW,
7257 		    &ms->ms_map) != 0) {
7258 			printf("%s: failed to allocate tx dma maps\n",
7259 			    DEVNAME(sc));
7260 			goto destroy_tx_slots;
7261 		}
7262 	}
7263 
7264 	if (mcx_create_cq(sc, &q->q_cq, q->q_uar, q->q_index,
7265 	    q->q_eq.eq_n) != 0)
7266 		goto destroy_tx_slots;
7267 
7268 	if (mcx_create_sq(sc, tx, q->q_uar, q->q_index, q->q_cq.cq_n)
7269 	    != 0)
7270 		goto destroy_cq;
7271 
7272 	if (mcx_create_rq(sc, rx, q->q_index, q->q_cq.cq_n) != 0)
7273 		goto destroy_sq;
7274 
7275 	return 0;
7276 
7277 destroy_sq:
7278 	mcx_destroy_sq(sc, tx);
7279 destroy_cq:
7280 	mcx_destroy_cq(sc, &q->q_cq);
7281 destroy_tx_slots:
7282 	mcx_free_slots(sc, tx->tx_slots, i, (1 << MCX_LOG_SQ_SIZE));
7283 	tx->tx_slots = NULL;
7284 
7285 	i = (1 << MCX_LOG_RQ_SIZE);
7286 destroy_rx_slots:
7287 	mcx_free_slots(sc, rx->rx_slots, i, (1 << MCX_LOG_RQ_SIZE));
7288 	rx->rx_slots = NULL;
7289 	return ENOMEM;
7290 }
7291 
7292 static int
7293 mcx_rss_group_entry_count(struct mcx_softc *sc, int group)
7294 {
7295 	int i;
7296 	int count;
7297 
7298 	count = 0;
7299 	for (i = 0; i < __arraycount(mcx_rss_config); i++) {
7300 		if (mcx_rss_config[i].flow_group == group)
7301 			count++;
7302 	}
7303 
7304 	return count;
7305 }
7306 
7307 static int
7308 mcx_init(struct ifnet *ifp)
7309 {
7310 	struct mcx_softc *sc = ifp->if_softc;
7311 	struct mcx_rx *rx;
7312 	struct mcx_tx *tx;
7313 	int i, start, count, flow_group, flow_index;
7314 	struct mcx_flow_match match_crit;
7315 	struct mcx_rss_rule *rss;
7316 	uint32_t dest;
7317 	int rqns[MCX_MAX_QUEUES] = { 0 };
7318 
7319 	if (ISSET(ifp->if_flags, IFF_RUNNING))
7320 		mcx_stop(ifp, 0);
7321 
7322 	if (mcx_create_tis(sc, &sc->sc_tis) != 0)
7323 		goto down;
7324 
7325 	for (i = 0; i < sc->sc_nqueues; i++) {
7326 		if (mcx_queue_up(sc, &sc->sc_queues[i]) != 0) {
7327 			goto down;
7328 		}
7329 	}
7330 
7331 	/* RSS flow table and flow groups */
7332 	if (mcx_create_flow_table(sc, MCX_LOG_FLOW_TABLE_SIZE, 1,
7333 	    &sc->sc_rss_flow_table_id) != 0)
7334 		goto down;
7335 
7336 	dest = MCX_FLOW_CONTEXT_DEST_TYPE_TABLE |
7337 	    sc->sc_rss_flow_table_id;
7338 
7339 	/* L4 RSS flow group (v4/v6 tcp/udp, no fragments) */
7340 	memset(&match_crit, 0, sizeof(match_crit));
7341 	match_crit.mc_ethertype = 0xffff;
7342 	match_crit.mc_ip_proto = 0xff;
7343 	match_crit.mc_vlan_flags = MCX_FLOW_MATCH_IP_FRAG;
7344 	start = 0;
7345 	count = mcx_rss_group_entry_count(sc, MCX_FLOW_GROUP_RSS_L4);
7346 	if (count != 0) {
7347 		if (mcx_create_flow_group(sc, sc->sc_rss_flow_table_id,
7348 		    MCX_FLOW_GROUP_RSS_L4, start, count,
7349 		    MCX_CREATE_FLOW_GROUP_CRIT_OUTER, &match_crit) != 0)
7350 			goto down;
7351 		start += count;
7352 	}
7353 
7354 	/* L3 RSS flow group (v4/v6, including fragments) */
7355 	memset(&match_crit, 0, sizeof(match_crit));
7356 	match_crit.mc_ethertype = 0xffff;
7357 	count = mcx_rss_group_entry_count(sc, MCX_FLOW_GROUP_RSS_L3);
7358 	if (mcx_create_flow_group(sc, sc->sc_rss_flow_table_id,
7359 	    MCX_FLOW_GROUP_RSS_L3, start, count,
7360 	    MCX_CREATE_FLOW_GROUP_CRIT_OUTER, &match_crit) != 0)
7361 		goto down;
7362 	start += count;
7363 
7364 	/* non-RSS flow group */
7365 	count = mcx_rss_group_entry_count(sc, MCX_FLOW_GROUP_RSS_NONE);
7366 	memset(&match_crit, 0, sizeof(match_crit));
7367 	if (mcx_create_flow_group(sc, sc->sc_rss_flow_table_id,
7368 	    MCX_FLOW_GROUP_RSS_NONE, start, count, 0, &match_crit) != 0)
7369 		goto down;
7370 
7371 	/* Root flow table, matching packets based on mac address */
7372 	if (mcx_create_flow_table(sc, MCX_LOG_FLOW_TABLE_SIZE, 0,
7373 	    &sc->sc_mac_flow_table_id) != 0)
7374 		goto down;
7375 
7376 	/* promisc flow group */
7377 	start = 0;
7378 	memset(&match_crit, 0, sizeof(match_crit));
7379 	if (mcx_create_flow_group(sc, sc->sc_mac_flow_table_id,
7380 	    MCX_FLOW_GROUP_PROMISC, start, 1, 0, &match_crit) != 0)
7381 		goto down;
7382 	sc->sc_promisc_flow_enabled = 0;
7383 	start++;
7384 
7385 	/* all multicast flow group */
7386 	match_crit.mc_dest_mac[0] = 0x01;
7387 	if (mcx_create_flow_group(sc, sc->sc_mac_flow_table_id,
7388 	    MCX_FLOW_GROUP_ALLMULTI, start, 1,
7389 	    MCX_CREATE_FLOW_GROUP_CRIT_OUTER, &match_crit) != 0)
7390 		goto down;
7391 	sc->sc_allmulti_flow_enabled = 0;
7392 	start++;
7393 
7394 	/* mac address matching flow group */
7395 	memset(&match_crit.mc_dest_mac, 0xff, sizeof(match_crit.mc_dest_mac));
7396 	if (mcx_create_flow_group(sc, sc->sc_mac_flow_table_id,
7397 	    MCX_FLOW_GROUP_MAC, start, (1 << MCX_LOG_FLOW_TABLE_SIZE) - start,
7398 	    MCX_CREATE_FLOW_GROUP_CRIT_OUTER, &match_crit) != 0)
7399 		goto down;
7400 
7401 	/* flow table entries for unicast and broadcast */
7402 	start = 0;
7403 	if (mcx_set_flow_table_entry_mac(sc, MCX_FLOW_GROUP_MAC, start,
7404 	    LLADDR(satosdl(ifp->if_dl->ifa_addr)), dest) != 0)
7405 		goto down;
7406 	start++;
7407 
7408 	if (mcx_set_flow_table_entry_mac(sc, MCX_FLOW_GROUP_MAC, start,
7409 	    etherbroadcastaddr, dest) != 0)
7410 		goto down;
7411 	start++;
7412 
7413 	/* multicast entries go after that */
7414 	sc->sc_mcast_flow_base = start;
7415 
7416 	/* re-add any existing multicast flows */
7417 	for (i = 0; i < MCX_NUM_MCAST_FLOWS; i++) {
7418 		if (sc->sc_mcast_flows[i][0] != 0) {
7419 			mcx_set_flow_table_entry_mac(sc, MCX_FLOW_GROUP_MAC,
7420 			    sc->sc_mcast_flow_base + i,
7421 			    sc->sc_mcast_flows[i], dest);
7422 		}
7423 	}
7424 
7425 	if (mcx_set_flow_table_root(sc, sc->sc_mac_flow_table_id) != 0)
7426 		goto down;
7427 
7428 	/*
7429 	 * the RQT can be any size as long as it's a power of two.
7430 	 * since we also restrict the number of queues to a power of two,
7431 	 * we can just put each rx queue in once.
7432 	 */
7433 	for (i = 0; i < sc->sc_nqueues; i++)
7434 		rqns[i] = sc->sc_queues[i].q_rx.rx_rqn;
7435 
7436 	if (mcx_create_rqt(sc, sc->sc_nqueues, rqns, &sc->sc_rqt) != 0)
7437 		goto down;
7438 
7439 	start = 0;
7440 	flow_index = 0;
7441 	flow_group = -1;
7442 	for (i = 0; i < __arraycount(mcx_rss_config); i++) {
7443 		rss = &mcx_rss_config[i];
7444 		if (rss->flow_group != flow_group) {
7445 			flow_group = rss->flow_group;
7446 			flow_index = 0;
7447 		}
7448 
7449 		if (rss->hash_sel == 0) {
7450 			if (mcx_create_tir_direct(sc, &sc->sc_queues[0].q_rx,
7451 			    &sc->sc_tir[i]) != 0)
7452 				goto down;
7453 		} else {
7454 			if (mcx_create_tir_indirect(sc, sc->sc_rqt,
7455 			    rss->hash_sel, &sc->sc_tir[i]) != 0)
7456 				goto down;
7457 		}
7458 
7459 		if (mcx_set_flow_table_entry_proto(sc, flow_group,
7460 		    flow_index, rss->ethertype, rss->ip_proto,
7461 		    MCX_FLOW_CONTEXT_DEST_TYPE_TIR | sc->sc_tir[i]) != 0)
7462 			goto down;
7463 		flow_index++;
7464 	}
7465 
7466 	for (i = 0; i < sc->sc_nqueues; i++) {
7467 		struct mcx_queues *q = &sc->sc_queues[i];
7468 		rx = &q->q_rx;
7469 		tx = &q->q_tx;
7470 
7471 		/* start the queues */
7472 		if (mcx_ready_sq(sc, tx) != 0)
7473 			goto down;
7474 
7475 		if (mcx_ready_rq(sc, rx) != 0)
7476 			goto down;
7477 
7478 		mcx_rxr_init(&rx->rx_rxr, 1, (1 << MCX_LOG_RQ_SIZE));
7479 		rx->rx_prod = 0;
7480 		mcx_rx_fill(sc, rx);
7481 
7482 		tx->tx_cons = 0;
7483 		tx->tx_prod = 0;
7484 	}
7485 
7486 	mcx_calibrate_first(sc);
7487 
7488 	SET(ifp->if_flags, IFF_RUNNING);
7489 	CLR(ifp->if_flags, IFF_OACTIVE);
7490 	if_schedule_deferred_start(ifp);
7491 
7492 	return 0;
7493 down:
7494 	mcx_stop(ifp, 0);
7495 	return EIO;
7496 }
7497 
7498 static void
7499 mcx_stop(struct ifnet *ifp, int disable)
7500 {
7501 	struct mcx_softc *sc = ifp->if_softc;
7502 	struct mcx_rss_rule *rss;
7503 	int group, i, flow_group, flow_index;
7504 
7505 	CLR(ifp->if_flags, IFF_RUNNING);
7506 
7507 	/*
7508 	 * delete flow table entries first, so no packets can arrive
7509 	 * after the barriers
7510 	 */
7511 	if (sc->sc_promisc_flow_enabled)
7512 		mcx_delete_flow_table_entry(sc, MCX_FLOW_GROUP_PROMISC, 0);
7513 	if (sc->sc_allmulti_flow_enabled)
7514 		mcx_delete_flow_table_entry(sc, MCX_FLOW_GROUP_ALLMULTI, 0);
7515 	mcx_delete_flow_table_entry(sc, MCX_FLOW_GROUP_MAC, 0);
7516 	mcx_delete_flow_table_entry(sc, MCX_FLOW_GROUP_MAC, 1);
7517 	for (i = 0; i < MCX_NUM_MCAST_FLOWS; i++) {
7518 		if (sc->sc_mcast_flows[i][0] != 0) {
7519 			mcx_delete_flow_table_entry(sc, MCX_FLOW_GROUP_MAC,
7520 			    sc->sc_mcast_flow_base + i);
7521 		}
7522 	}
7523 
7524 	flow_group = -1;
7525 	flow_index = 0;
7526 	for (i = 0; i < __arraycount(mcx_rss_config); i++) {
7527 		rss = &mcx_rss_config[i];
7528 		if (rss->flow_group != flow_group) {
7529 			flow_group = rss->flow_group;
7530 			flow_index = 0;
7531 		}
7532 
7533 		mcx_delete_flow_table_entry(sc, flow_group, flow_index);
7534 
7535 		mcx_destroy_tir(sc, sc->sc_tir[i]);
7536 		sc->sc_tir[i] = 0;
7537 
7538 		flow_index++;
7539 	}
7540 
7541 	for (i = 0; i < sc->sc_nqueues; i++) {
7542 		callout_halt(&sc->sc_queues[i].q_rx.rx_refill, NULL);
7543 	}
7544 
7545 	callout_halt(&sc->sc_calibrate, NULL);
7546 
7547 	for (group = 0; group < MCX_NUM_FLOW_GROUPS; group++) {
7548 		if (sc->sc_flow_group[group].g_id != -1)
7549 			mcx_destroy_flow_group(sc, group);
7550 	}
7551 
7552 	if (sc->sc_mac_flow_table_id != -1) {
7553 		mcx_destroy_flow_table(sc, sc->sc_mac_flow_table_id);
7554 		sc->sc_mac_flow_table_id = -1;
7555 	}
7556 	if (sc->sc_rss_flow_table_id != -1) {
7557 		mcx_destroy_flow_table(sc, sc->sc_rss_flow_table_id);
7558 		sc->sc_rss_flow_table_id = -1;
7559 	}
7560 	if (sc->sc_rqt != -1) {
7561 		mcx_destroy_rqt(sc, sc->sc_rqt);
7562 		sc->sc_rqt = -1;
7563 	}
7564 
7565 	for (i = 0; i < sc->sc_nqueues; i++) {
7566 		struct mcx_queues *q = &sc->sc_queues[i];
7567 		struct mcx_rx *rx = &q->q_rx;
7568 		struct mcx_tx *tx = &q->q_tx;
7569 		struct mcx_cq *cq = &q->q_cq;
7570 
7571 		if (rx->rx_rqn != 0)
7572 			mcx_destroy_rq(sc, rx);
7573 
7574 		if (tx->tx_sqn != 0)
7575 			mcx_destroy_sq(sc, tx);
7576 
7577 		if (tx->tx_slots != NULL) {
7578 			mcx_free_slots(sc, tx->tx_slots,
7579 			    (1 << MCX_LOG_SQ_SIZE), (1 << MCX_LOG_SQ_SIZE));
7580 			tx->tx_slots = NULL;
7581 		}
7582 		if (rx->rx_slots != NULL) {
7583 			mcx_free_slots(sc, rx->rx_slots,
7584 			    (1 << MCX_LOG_RQ_SIZE), (1 << MCX_LOG_RQ_SIZE));
7585 			rx->rx_slots = NULL;
7586 		}
7587 
7588 		if (cq->cq_n != 0)
7589 			mcx_destroy_cq(sc, cq);
7590 	}
7591 	if (sc->sc_tis != 0) {
7592 		mcx_destroy_tis(sc, sc->sc_tis);
7593 		sc->sc_tis = 0;
7594 	}
7595 }
7596 
7597 static int
7598 mcx_ioctl(struct ifnet *ifp, u_long cmd, void *data)
7599 {
7600 	struct mcx_softc *sc = (struct mcx_softc *)ifp->if_softc;
7601 	struct ifreq *ifr = (struct ifreq *)data;
7602 	struct ethercom *ec = &sc->sc_ec;
7603 	uint8_t addrhi[ETHER_ADDR_LEN], addrlo[ETHER_ADDR_LEN];
7604 	struct ether_multi *enm;
7605 	struct ether_multistep step;
7606 	int s, i, flags, error = 0;
7607 	uint32_t dest;
7608 
7609 	s = splnet();
7610 	switch (cmd) {
7611 
7612 	case SIOCADDMULTI:
7613 		if (ether_addmulti(ifreq_getaddr(cmd, ifr), &sc->sc_ec) == ENETRESET) {
7614 			error = ether_multiaddr(&ifr->ifr_addr, addrlo, addrhi);
7615 			if (error != 0) {
7616 				splx(s);
7617 				return (error);
7618 			}
7619 
7620 			dest = MCX_FLOW_CONTEXT_DEST_TYPE_TABLE |
7621 			    sc->sc_rss_flow_table_id;
7622 
7623 			for (i = 0; i < MCX_NUM_MCAST_FLOWS; i++) {
7624 				if (sc->sc_mcast_flows[i][0] == 0) {
7625 					memcpy(sc->sc_mcast_flows[i], addrlo,
7626 					    ETHER_ADDR_LEN);
7627 					if (ISSET(ifp->if_flags, IFF_RUNNING)) {
7628 						mcx_set_flow_table_entry_mac(sc,
7629 						    MCX_FLOW_GROUP_MAC,
7630 						    sc->sc_mcast_flow_base + i,
7631 						    sc->sc_mcast_flows[i], dest);
7632 					}
7633 					break;
7634 				}
7635 			}
7636 
7637 			if (!ISSET(ifp->if_flags, IFF_ALLMULTI)) {
7638 				if (i == MCX_NUM_MCAST_FLOWS) {
7639 					SET(ifp->if_flags, IFF_ALLMULTI);
7640 					sc->sc_extra_mcast++;
7641 					error = ENETRESET;
7642 				}
7643 
7644 				if (memcmp(addrlo, addrhi, ETHER_ADDR_LEN)) {
7645 					SET(ifp->if_flags, IFF_ALLMULTI);
7646 					error = ENETRESET;
7647 				}
7648 			}
7649 		}
7650 		break;
7651 
7652 	case SIOCDELMULTI:
7653 		if (ether_delmulti(ifreq_getaddr(cmd, ifr), &sc->sc_ec) == ENETRESET) {
7654 			error = ether_multiaddr(&ifr->ifr_addr, addrlo, addrhi);
7655 			if (error != 0) {
7656 				splx(s);
7657 				return (error);
7658 			}
7659 
7660 			for (i = 0; i < MCX_NUM_MCAST_FLOWS; i++) {
7661 				if (memcmp(sc->sc_mcast_flows[i], addrlo,
7662 				    ETHER_ADDR_LEN) == 0) {
7663 					if (ISSET(ifp->if_flags, IFF_RUNNING)) {
7664 						mcx_delete_flow_table_entry(sc,
7665 						    MCX_FLOW_GROUP_MAC,
7666 						    sc->sc_mcast_flow_base + i);
7667 					}
7668 					sc->sc_mcast_flows[i][0] = 0;
7669 					break;
7670 				}
7671 			}
7672 
7673 			if (i == MCX_NUM_MCAST_FLOWS)
7674 				sc->sc_extra_mcast--;
7675 
7676 			if (ISSET(ifp->if_flags, IFF_ALLMULTI) &&
7677 			    sc->sc_extra_mcast == 0) {
7678 				flags = 0;
7679 				ETHER_LOCK(ec);
7680 				ETHER_FIRST_MULTI(step, ec, enm);
7681 				while (enm != NULL) {
7682 					if (memcmp(enm->enm_addrlo,
7683 					    enm->enm_addrhi, ETHER_ADDR_LEN)) {
7684 						SET(flags, IFF_ALLMULTI);
7685 						break;
7686 					}
7687 					ETHER_NEXT_MULTI(step, enm);
7688 				}
7689 				ETHER_UNLOCK(ec);
7690 				if (!ISSET(flags, IFF_ALLMULTI)) {
7691 					CLR(ifp->if_flags, IFF_ALLMULTI);
7692 					error = ENETRESET;
7693 				}
7694 			}
7695 		}
7696 		break;
7697 
7698 	default:
7699 		error = ether_ioctl(ifp, cmd, data);
7700 	}
7701 
7702 	if (error == ENETRESET) {
7703 		if ((ifp->if_flags & (IFF_UP | IFF_RUNNING)) ==
7704 		    (IFF_UP | IFF_RUNNING))
7705 			mcx_iff(sc);
7706 		error = 0;
7707 	}
7708 	splx(s);
7709 
7710 	return (error);
7711 }
7712 
7713 #if 0
7714 static int
7715 mcx_get_sffpage(struct ifnet *ifp, struct if_sffpage *sff)
7716 {
7717 	struct mcx_softc *sc = (struct mcx_softc *)ifp->if_softc;
7718 	struct mcx_reg_mcia mcia;
7719 	struct mcx_reg_pmlp pmlp;
7720 	int offset, error;
7721 
7722 	/* get module number */
7723 	memset(&pmlp, 0, sizeof(pmlp));
7724 	pmlp.rp_local_port = 1;
7725 	error = mcx_access_hca_reg(sc, MCX_REG_PMLP, MCX_REG_OP_READ, &pmlp,
7726 	    sizeof(pmlp));
7727 	if (error != 0) {
7728 		printf("%s: unable to get eeprom module number\n",
7729 		    DEVNAME(sc));
7730 		return error;
7731 	}
7732 
7733 	for (offset = 0; offset < 256; offset += MCX_MCIA_EEPROM_BYTES) {
7734 		memset(&mcia, 0, sizeof(mcia));
7735 		mcia.rm_l = 0;
7736 		mcia.rm_module = be32toh(pmlp.rp_lane0_mapping) &
7737 		    MCX_PMLP_MODULE_NUM_MASK;
7738 		mcia.rm_i2c_addr = sff->sff_addr / 2;	/* apparently */
7739 		mcia.rm_page_num = sff->sff_page;
7740 		mcia.rm_dev_addr = htobe16(offset);
7741 		mcia.rm_size = htobe16(MCX_MCIA_EEPROM_BYTES);
7742 
7743 		error = mcx_access_hca_reg(sc, MCX_REG_MCIA, MCX_REG_OP_READ,
7744 		    &mcia, sizeof(mcia));
7745 		if (error != 0) {
7746 			printf("%s: unable to read eeprom at %x\n",
7747 			    DEVNAME(sc), offset);
7748 			return error;
7749 		}
7750 
7751 		memcpy(sff->sff_data + offset, mcia.rm_data,
7752 		    MCX_MCIA_EEPROM_BYTES);
7753 	}
7754 
7755 	return 0;
7756 }
7757 #endif
7758 
7759 static int
7760 mcx_load_mbuf(struct mcx_softc *sc, struct mcx_slot *ms, struct mbuf *m)
7761 {
7762 	switch (bus_dmamap_load_mbuf(sc->sc_dmat, ms->ms_map, m,
7763 	    BUS_DMA_STREAMING | BUS_DMA_NOWAIT)) {
7764 	case 0:
7765 		break;
7766 
7767 	case EFBIG:
7768 		if (m_defrag(m, M_DONTWAIT) != NULL &&
7769 		    bus_dmamap_load_mbuf(sc->sc_dmat, ms->ms_map, m,
7770 		    BUS_DMA_STREAMING | BUS_DMA_NOWAIT) == 0)
7771 			break;
7772 
7773 		/* FALLTHROUGH */
7774 	default:
7775 		return (1);
7776 	}
7777 
7778 	ms->ms_m = m;
7779 	return (0);
7780 }
7781 
7782 static void
7783 mcx_send_common_locked(struct ifnet *ifp, struct mcx_tx *tx, bool is_transmit)
7784 {
7785 	struct mcx_softc *sc = ifp->if_softc;
7786 	struct mcx_sq_entry *sq, *sqe;
7787 	struct mcx_sq_entry_seg *sqs;
7788 	struct mcx_slot *ms;
7789 	bus_dmamap_t map;
7790 	struct mbuf *m;
7791 	u_int idx, free, used;
7792 	uint64_t *bf;
7793 	uint32_t csum;
7794 	size_t bf_base;
7795 	int i, seg, nseg;
7796 
7797 	KASSERT(mutex_owned(&tx->tx_lock));
7798 
7799 	if ((ifp->if_flags & IFF_RUNNING) == 0)
7800 		return;
7801 
7802 	bf_base = (tx->tx_uar * MCX_PAGE_SIZE) + MCX_UAR_BF;
7803 
7804 	idx = tx->tx_prod % (1 << MCX_LOG_SQ_SIZE);
7805 	free = (tx->tx_cons + (1 << MCX_LOG_SQ_SIZE)) - tx->tx_prod;
7806 
7807 	used = 0;
7808 	bf = NULL;
7809 
7810 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&tx->tx_sq_mem),
7811 	    0, MCX_DMA_LEN(&tx->tx_sq_mem), BUS_DMASYNC_POSTWRITE);
7812 
7813 	sq = (struct mcx_sq_entry *)MCX_DMA_KVA(&tx->tx_sq_mem);
7814 
7815 	for (;;) {
7816 		if (used + MCX_SQ_ENTRY_MAX_SLOTS >= free) {
7817 			SET(ifp->if_flags, IFF_OACTIVE);
7818 			break;
7819 		}
7820 
7821 		if (is_transmit) {
7822 			m = pcq_get(tx->tx_pcq);
7823 		} else {
7824 			IFQ_DEQUEUE(&ifp->if_snd, m);
7825 		}
7826 		if (m == NULL) {
7827 			break;
7828 		}
7829 
7830 		sqe = sq + idx;
7831 		ms = &tx->tx_slots[idx];
7832 		memset(sqe, 0, sizeof(*sqe));
7833 
7834 		/* ctrl segment */
7835 		sqe->sqe_opcode_index = htobe32(MCX_SQE_WQE_OPCODE_SEND |
7836 		    ((tx->tx_prod & 0xffff) << MCX_SQE_WQE_INDEX_SHIFT));
7837 		/* always generate a completion event */
7838 		sqe->sqe_signature = htobe32(MCX_SQE_CE_CQE_ALWAYS);
7839 
7840 		/* eth segment */
7841 		csum = 0;
7842 		if (m->m_pkthdr.csum_flags & M_CSUM_IPv4)
7843 			csum |= MCX_SQE_L3_CSUM;
7844 		if (m->m_pkthdr.csum_flags &
7845 		    (M_CSUM_TCPv4 | M_CSUM_UDPv4 | M_CSUM_TCPv6 | M_CSUM_UDPv6))
7846 			csum |= MCX_SQE_L4_CSUM;
7847 		sqe->sqe_mss_csum = htobe32(csum);
7848 		sqe->sqe_inline_header_size = htobe16(MCX_SQ_INLINE_SIZE);
7849 		if (vlan_has_tag(m)) {
7850 			struct ether_vlan_header *evh;
7851 			evh = (struct ether_vlan_header *)
7852 			    &sqe->sqe_inline_headers;
7853 
7854 			m_copydata(m, 0, ETHER_HDR_LEN, evh);
7855 			evh->evl_proto = evh->evl_encap_proto;
7856 			evh->evl_encap_proto = htons(ETHERTYPE_VLAN);
7857 			evh->evl_tag = htons(vlan_get_tag(m));
7858 			m_adj(m, ETHER_HDR_LEN);
7859 		} else {
7860 			m_copydata(m, 0, MCX_SQ_INLINE_SIZE,
7861 			    sqe->sqe_inline_headers);
7862 			m_adj(m, MCX_SQ_INLINE_SIZE);
7863 		}
7864 
7865 		if (mcx_load_mbuf(sc, ms, m) != 0) {
7866 			m_freem(m);
7867 			if_statinc(ifp, if_oerrors);
7868 			continue;
7869 		}
7870 		bf = (uint64_t *)sqe;
7871 
7872 		if (ifp->if_bpf != NULL)
7873 			bpf_mtap2(ifp->if_bpf, sqe->sqe_inline_headers,
7874 			    MCX_SQ_INLINE_SIZE, m, BPF_D_OUT);
7875 
7876 		map = ms->ms_map;
7877 		bus_dmamap_sync(sc->sc_dmat, map, 0, map->dm_mapsize,
7878 		    BUS_DMASYNC_PREWRITE);
7879 
7880 		sqe->sqe_ds_sq_num =
7881 		    htobe32((tx->tx_sqn << MCX_SQE_SQ_NUM_SHIFT) |
7882 		    (map->dm_nsegs + 3));
7883 
7884 		/* data segment - first wqe has one segment */
7885 		sqs = sqe->sqe_segs;
7886 		seg = 0;
7887 		nseg = 1;
7888 		for (i = 0; i < map->dm_nsegs; i++) {
7889 			if (seg == nseg) {
7890 				/* next slot */
7891 				idx++;
7892 				if (idx == (1 << MCX_LOG_SQ_SIZE))
7893 					idx = 0;
7894 				tx->tx_prod++;
7895 				used++;
7896 
7897 				sqs = (struct mcx_sq_entry_seg *)(sq + idx);
7898 				seg = 0;
7899 				nseg = MCX_SQ_SEGS_PER_SLOT;
7900 			}
7901 			sqs[seg].sqs_byte_count =
7902 			    htobe32(map->dm_segs[i].ds_len);
7903 			sqs[seg].sqs_lkey = htobe32(sc->sc_lkey);
7904 			sqs[seg].sqs_addr = htobe64(map->dm_segs[i].ds_addr);
7905 			seg++;
7906 		}
7907 
7908 		idx++;
7909 		if (idx == (1 << MCX_LOG_SQ_SIZE))
7910 			idx = 0;
7911 		tx->tx_prod++;
7912 		used++;
7913 	}
7914 
7915 	bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&tx->tx_sq_mem),
7916 	    0, MCX_DMA_LEN(&tx->tx_sq_mem), BUS_DMASYNC_PREWRITE);
7917 
7918 	if (used) {
7919 		bus_size_t blueflame;
7920 
7921 		bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_doorbell_mem),
7922 		    tx->tx_doorbell, sizeof(uint32_t), BUS_DMASYNC_POSTWRITE);
7923 		be32enc(MCX_DMA_OFF(&sc->sc_doorbell_mem, tx->tx_doorbell),
7924 		    tx->tx_prod & MCX_WQ_DOORBELL_MASK);
7925 		bus_dmamap_sync(sc->sc_dmat, MCX_DMA_MAP(&sc->sc_doorbell_mem),
7926 		    tx->tx_doorbell, sizeof(uint32_t), BUS_DMASYNC_PREWRITE);
7927 
7928 		/*
7929 		 * write the first 64 bits of the last sqe we produced
7930 		 * to the blue flame buffer
7931 		 */
7932 
7933 		blueflame = bf_base + tx->tx_bf_offset;
7934 		bus_space_write_8(sc->sc_memt, sc->sc_memh,
7935 		    blueflame, *bf);
7936 		mcx_bar(sc, blueflame, sizeof(*bf), BUS_SPACE_BARRIER_WRITE);
7937 
7938 		/* next write goes to the other buffer */
7939 		tx->tx_bf_offset ^= sc->sc_bf_size;
7940 	}
7941 }
7942 
7943 static void
7944 mcx_start(struct ifnet *ifp)
7945 {
7946 	struct mcx_softc *sc = ifp->if_softc;
7947 	/* mcx_start() always uses TX ring[0] */
7948 	struct mcx_tx *tx = &sc->sc_queues[0].q_tx;
7949 
7950 	mutex_enter(&tx->tx_lock);
7951 	if (!ISSET(ifp->if_flags, IFF_OACTIVE)) {
7952 		mcx_send_common_locked(ifp, tx, false);
7953 	}
7954 	mutex_exit(&tx->tx_lock);
7955 }
7956 
7957 static int
7958 mcx_transmit(struct ifnet *ifp, struct mbuf *m)
7959 {
7960 	struct mcx_softc *sc = ifp->if_softc;
7961 	struct mcx_tx *tx;
7962 
7963 	tx = &sc->sc_queues[cpu_index(curcpu()) % sc->sc_nqueues].q_tx;
7964 	if (__predict_false(!pcq_put(tx->tx_pcq, m))) {
7965 		m_freem(m);
7966 		return ENOBUFS;
7967 	}
7968 
7969 	if (mutex_tryenter(&tx->tx_lock)) {
7970 		mcx_send_common_locked(ifp, tx, true);
7971 		mutex_exit(&tx->tx_lock);
7972 	} else {
7973 		softint_schedule(tx->tx_softint);
7974 	}
7975 
7976 	return 0;
7977 }
7978 
7979 static void
7980 mcx_deferred_transmit(void *arg)
7981 {
7982 	struct mcx_tx *tx = arg;
7983 	struct mcx_softc *sc = tx->tx_softc;
7984 	struct ifnet *ifp = &sc->sc_ec.ec_if;
7985 
7986 	mutex_enter(&tx->tx_lock);
7987 	if (pcq_peek(tx->tx_pcq) != NULL) {
7988 		mcx_send_common_locked(ifp, tx, true);
7989 	}
7990 	mutex_exit(&tx->tx_lock);
7991 }
7992 
7993 static void
7994 mcx_watchdog(struct ifnet *ifp)
7995 {
7996 }
7997 
7998 static void
7999 mcx_media_add_types(struct mcx_softc *sc)
8000 {
8001 	struct mcx_reg_ptys ptys;
8002 	int i;
8003 	uint32_t proto_cap;
8004 
8005 	memset(&ptys, 0, sizeof(ptys));
8006 	ptys.rp_local_port = 1;
8007 	ptys.rp_proto_mask = MCX_REG_PTYS_PROTO_MASK_ETH;
8008 	if (mcx_access_hca_reg(sc, MCX_REG_PTYS, MCX_REG_OP_READ, &ptys,
8009 	    sizeof(ptys)) != 0) {
8010 		printf("%s: unable to read port type/speed\n", DEVNAME(sc));
8011 		return;
8012 	}
8013 
8014 	proto_cap = be32toh(ptys.rp_eth_proto_cap);
8015 	for (i = 0; i < __arraycount(mcx_eth_cap_map); i++) {
8016 		const struct mcx_eth_proto_capability *cap;
8017 		if (!ISSET(proto_cap, 1U << i))
8018 			continue;
8019 
8020 		cap = &mcx_eth_cap_map[i];
8021 		if (cap->cap_media == 0)
8022 			continue;
8023 
8024 		ifmedia_add(&sc->sc_media, IFM_ETHER | cap->cap_media, 0, NULL);
8025 	}
8026 }
8027 
8028 static void
8029 mcx_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
8030 {
8031 	struct mcx_softc *sc = (struct mcx_softc *)ifp->if_softc;
8032 	struct mcx_reg_ptys ptys;
8033 	int i;
8034 	uint32_t proto_oper;
8035 	uint64_t media_oper;
8036 
8037 	memset(&ptys, 0, sizeof(ptys));
8038 	ptys.rp_local_port = 1;
8039 	ptys.rp_proto_mask = MCX_REG_PTYS_PROTO_MASK_ETH;
8040 
8041 	if (mcx_access_hca_reg(sc, MCX_REG_PTYS, MCX_REG_OP_READ, &ptys,
8042 	    sizeof(ptys)) != 0) {
8043 		printf("%s: unable to read port type/speed\n", DEVNAME(sc));
8044 		return;
8045 	}
8046 
8047 	proto_oper = be32toh(ptys.rp_eth_proto_oper);
8048 
8049 	media_oper = 0;
8050 
8051 	for (i = 0; i < __arraycount(mcx_eth_cap_map); i++) {
8052 		const struct mcx_eth_proto_capability *cap;
8053 		if (!ISSET(proto_oper, 1U << i))
8054 			continue;
8055 
8056 		cap = &mcx_eth_cap_map[i];
8057 
8058 		if (cap->cap_media != 0)
8059 			media_oper = cap->cap_media;
8060 	}
8061 
8062 	ifmr->ifm_status = IFM_AVALID;
8063 	if (proto_oper != 0) {
8064 		ifmr->ifm_status |= IFM_ACTIVE;
8065 		ifmr->ifm_active = IFM_ETHER | IFM_AUTO | media_oper;
8066 		/* txpause, rxpause, duplex? */
8067 	}
8068 }
8069 
8070 static int
8071 mcx_media_change(struct ifnet *ifp)
8072 {
8073 	struct mcx_softc *sc = (struct mcx_softc *)ifp->if_softc;
8074 	struct mcx_reg_ptys ptys;
8075 	struct mcx_reg_paos paos;
8076 	uint32_t media;
8077 	int i, error;
8078 
8079 	if (IFM_TYPE(sc->sc_media.ifm_media) != IFM_ETHER)
8080 		return EINVAL;
8081 
8082 	error = 0;
8083 
8084 	if (IFM_SUBTYPE(sc->sc_media.ifm_media) == IFM_AUTO) {
8085 		/* read ptys to get supported media */
8086 		memset(&ptys, 0, sizeof(ptys));
8087 		ptys.rp_local_port = 1;
8088 		ptys.rp_proto_mask = MCX_REG_PTYS_PROTO_MASK_ETH;
8089 		if (mcx_access_hca_reg(sc, MCX_REG_PTYS, MCX_REG_OP_READ,
8090 		    &ptys, sizeof(ptys)) != 0) {
8091 			printf("%s: unable to read port type/speed\n",
8092 			    DEVNAME(sc));
8093 			return EIO;
8094 		}
8095 
8096 		media = be32toh(ptys.rp_eth_proto_cap);
8097 	} else {
8098 		/* map media type */
8099 		media = 0;
8100 		for (i = 0; i < __arraycount(mcx_eth_cap_map); i++) {
8101 			const struct mcx_eth_proto_capability *cap;
8102 
8103 			cap = &mcx_eth_cap_map[i];
8104 			if (cap->cap_media ==
8105 			    IFM_SUBTYPE(sc->sc_media.ifm_media)) {
8106 				media = (1 << i);
8107 				break;
8108 			}
8109 		}
8110 	}
8111 
8112 	/* disable the port */
8113 	memset(&paos, 0, sizeof(paos));
8114 	paos.rp_local_port = 1;
8115 	paos.rp_admin_status = MCX_REG_PAOS_ADMIN_STATUS_DOWN;
8116 	paos.rp_admin_state_update = MCX_REG_PAOS_ADMIN_STATE_UPDATE_EN;
8117 	if (mcx_access_hca_reg(sc, MCX_REG_PAOS, MCX_REG_OP_WRITE, &paos,
8118 	    sizeof(paos)) != 0) {
8119 		printf("%s: unable to set port state to down\n", DEVNAME(sc));
8120 		return EIO;
8121 	}
8122 
8123 	memset(&ptys, 0, sizeof(ptys));
8124 	ptys.rp_local_port = 1;
8125 	ptys.rp_proto_mask = MCX_REG_PTYS_PROTO_MASK_ETH;
8126 	ptys.rp_eth_proto_admin = htobe32(media);
8127 	if (mcx_access_hca_reg(sc, MCX_REG_PTYS, MCX_REG_OP_WRITE, &ptys,
8128 	    sizeof(ptys)) != 0) {
8129 		printf("%s: unable to set port media type/speed\n",
8130 		    DEVNAME(sc));
8131 		error = EIO;
8132 	}
8133 
8134 	/* re-enable the port to start negotiation */
8135 	memset(&paos, 0, sizeof(paos));
8136 	paos.rp_local_port = 1;
8137 	paos.rp_admin_status = MCX_REG_PAOS_ADMIN_STATUS_UP;
8138 	paos.rp_admin_state_update = MCX_REG_PAOS_ADMIN_STATE_UPDATE_EN;
8139 	if (mcx_access_hca_reg(sc, MCX_REG_PAOS, MCX_REG_OP_WRITE, &paos,
8140 	    sizeof(paos)) != 0) {
8141 		printf("%s: unable to set port state to up\n", DEVNAME(sc));
8142 		error = EIO;
8143 	}
8144 
8145 	return error;
8146 }
8147 
8148 static void
8149 mcx_port_change(struct work *wk, void *xsc)
8150 {
8151 	struct mcx_softc *sc = xsc;
8152 	struct ifnet *ifp = &sc->sc_ec.ec_if;
8153 	struct mcx_reg_ptys ptys = {
8154 		.rp_local_port = 1,
8155 		.rp_proto_mask = MCX_REG_PTYS_PROTO_MASK_ETH,
8156 	};
8157 	int link_state = LINK_STATE_DOWN;
8158 
8159 	if (mcx_access_hca_reg(sc, MCX_REG_PTYS, MCX_REG_OP_READ, &ptys,
8160 	    sizeof(ptys)) == 0) {
8161 		uint32_t proto_oper = be32toh(ptys.rp_eth_proto_oper);
8162 		uint64_t baudrate = 0;
8163 		unsigned int i;
8164 
8165 		if (proto_oper != 0)
8166 			link_state = LINK_STATE_UP;
8167 
8168 		for (i = 0; i < __arraycount(mcx_eth_cap_map); i++) {
8169 			const struct mcx_eth_proto_capability *cap;
8170 			if (!ISSET(proto_oper, 1U << i))
8171 				continue;
8172 
8173 			cap = &mcx_eth_cap_map[i];
8174 			if (cap->cap_baudrate == 0)
8175 				continue;
8176 
8177 			baudrate = cap->cap_baudrate;
8178 			break;
8179 		}
8180 
8181 		ifp->if_baudrate = baudrate;
8182 	}
8183 
8184 	if (link_state != ifp->if_link_state) {
8185 		if_link_state_change(ifp, link_state);
8186 	}
8187 }
8188 
8189 
8190 static inline uint32_t
8191 mcx_rd(struct mcx_softc *sc, bus_size_t r)
8192 {
8193 	uint32_t word;
8194 
8195 	word = bus_space_read_4(sc->sc_memt, sc->sc_memh, r);
8196 
8197 	return (be32toh(word));
8198 }
8199 
8200 static inline void
8201 mcx_wr(struct mcx_softc *sc, bus_size_t r, uint32_t v)
8202 {
8203 	bus_space_write_4(sc->sc_memt, sc->sc_memh, r, htobe32(v));
8204 }
8205 
8206 static inline void
8207 mcx_bar(struct mcx_softc *sc, bus_size_t r, bus_size_t l, int f)
8208 {
8209 	bus_space_barrier(sc->sc_memt, sc->sc_memh, r, l, f);
8210 }
8211 
8212 static uint64_t
8213 mcx_timer(struct mcx_softc *sc)
8214 {
8215 	uint32_t hi, lo, ni;
8216 
8217 	hi = mcx_rd(sc, MCX_INTERNAL_TIMER_H);
8218 	for (;;) {
8219 		lo = mcx_rd(sc, MCX_INTERNAL_TIMER_L);
8220 		mcx_bar(sc, MCX_INTERNAL_TIMER_L, 8, BUS_SPACE_BARRIER_READ);
8221 		ni = mcx_rd(sc, MCX_INTERNAL_TIMER_H);
8222 
8223 		if (ni == hi)
8224 			break;
8225 
8226 		hi = ni;
8227 	}
8228 
8229 	return (((uint64_t)hi << 32) | (uint64_t)lo);
8230 }
8231 
8232 static int
8233 mcx_dmamem_alloc(struct mcx_softc *sc, struct mcx_dmamem *mxm,
8234     bus_size_t size, u_int align)
8235 {
8236 	mxm->mxm_size = size;
8237 
8238 	if (bus_dmamap_create(sc->sc_dmat, mxm->mxm_size, 1,
8239 	    mxm->mxm_size, 0,
8240 	    BUS_DMA_WAITOK | BUS_DMA_ALLOCNOW,
8241 	    &mxm->mxm_map) != 0)
8242 		return (1);
8243 	if (bus_dmamem_alloc(sc->sc_dmat, mxm->mxm_size,
8244 	    align, 0, &mxm->mxm_seg, 1, &mxm->mxm_nsegs,
8245 	    BUS_DMA_WAITOK) != 0)
8246 		goto destroy;
8247 	if (bus_dmamem_map(sc->sc_dmat, &mxm->mxm_seg, mxm->mxm_nsegs,
8248 	    mxm->mxm_size, &mxm->mxm_kva, BUS_DMA_WAITOK) != 0)
8249 		goto free;
8250 	if (bus_dmamap_load(sc->sc_dmat, mxm->mxm_map, mxm->mxm_kva,
8251 	    mxm->mxm_size, NULL, BUS_DMA_WAITOK) != 0)
8252 		goto unmap;
8253 
8254 	mcx_dmamem_zero(mxm);
8255 
8256 	return (0);
8257 unmap:
8258 	bus_dmamem_unmap(sc->sc_dmat, mxm->mxm_kva, mxm->mxm_size);
8259 free:
8260 	bus_dmamem_free(sc->sc_dmat, &mxm->mxm_seg, 1);
8261 destroy:
8262 	bus_dmamap_destroy(sc->sc_dmat, mxm->mxm_map);
8263 	return (1);
8264 }
8265 
8266 static void
8267 mcx_dmamem_zero(struct mcx_dmamem *mxm)
8268 {
8269 	memset(MCX_DMA_KVA(mxm), 0, MCX_DMA_LEN(mxm));
8270 }
8271 
8272 static void
8273 mcx_dmamem_free(struct mcx_softc *sc, struct mcx_dmamem *mxm)
8274 {
8275 	bus_dmamap_unload(sc->sc_dmat, mxm->mxm_map);
8276 	bus_dmamem_unmap(sc->sc_dmat, mxm->mxm_kva, mxm->mxm_size);
8277 	bus_dmamem_free(sc->sc_dmat, &mxm->mxm_seg, 1);
8278 	bus_dmamap_destroy(sc->sc_dmat, mxm->mxm_map);
8279 }
8280 
8281 static int
8282 mcx_hwmem_alloc(struct mcx_softc *sc, struct mcx_hwmem *mhm, unsigned int pages)
8283 {
8284 	bus_dma_segment_t *segs;
8285 	bus_size_t len = pages * MCX_PAGE_SIZE;
8286 	size_t seglen;
8287 
8288 	segs = kmem_alloc(sizeof(*segs) * pages, KM_SLEEP);
8289 	seglen = sizeof(*segs) * pages;
8290 
8291 	if (bus_dmamem_alloc(sc->sc_dmat, len, MCX_PAGE_SIZE, 0,
8292 	    segs, pages, &mhm->mhm_seg_count, BUS_DMA_NOWAIT) != 0)
8293 		goto free_segs;
8294 
8295 	if (mhm->mhm_seg_count < pages) {
8296 		size_t nseglen;
8297 
8298 		mhm->mhm_segs = kmem_alloc(
8299 		    sizeof(*mhm->mhm_segs) * mhm->mhm_seg_count, KM_SLEEP);
8300 
8301 		nseglen = sizeof(*mhm->mhm_segs) * mhm->mhm_seg_count;
8302 
8303 		memcpy(mhm->mhm_segs, segs, nseglen);
8304 
8305 		kmem_free(segs, seglen);
8306 
8307 		segs = mhm->mhm_segs;
8308 		seglen = nseglen;
8309 	} else
8310 		mhm->mhm_segs = segs;
8311 
8312 	if (bus_dmamap_create(sc->sc_dmat, len, pages, MCX_PAGE_SIZE,
8313 	    MCX_PAGE_SIZE, BUS_DMA_NOWAIT|BUS_DMA_ALLOCNOW /*|BUS_DMA_64BIT*/,
8314 	    &mhm->mhm_map) != 0)
8315 		goto free_dmamem;
8316 
8317 	if (bus_dmamap_load_raw(sc->sc_dmat, mhm->mhm_map,
8318 	    mhm->mhm_segs, mhm->mhm_seg_count, len, BUS_DMA_NOWAIT) != 0)
8319 		goto destroy;
8320 
8321 	bus_dmamap_sync(sc->sc_dmat, mhm->mhm_map,
8322 	    0, mhm->mhm_map->dm_mapsize, BUS_DMASYNC_PRERW);
8323 
8324 	mhm->mhm_npages = pages;
8325 
8326 	return (0);
8327 
8328 destroy:
8329 	bus_dmamap_destroy(sc->sc_dmat, mhm->mhm_map);
8330 free_dmamem:
8331 	bus_dmamem_free(sc->sc_dmat, mhm->mhm_segs, mhm->mhm_seg_count);
8332 free_segs:
8333 	kmem_free(segs, seglen);
8334 	mhm->mhm_segs = NULL;
8335 
8336 	return (-1);
8337 }
8338 
8339 static void
8340 mcx_hwmem_free(struct mcx_softc *sc, struct mcx_hwmem *mhm)
8341 {
8342 	if (mhm->mhm_npages == 0)
8343 		return;
8344 
8345 	bus_dmamap_sync(sc->sc_dmat, mhm->mhm_map,
8346 	    0, mhm->mhm_map->dm_mapsize, BUS_DMASYNC_POSTRW);
8347 
8348 	bus_dmamap_unload(sc->sc_dmat, mhm->mhm_map);
8349 	bus_dmamap_destroy(sc->sc_dmat, mhm->mhm_map);
8350 	bus_dmamem_free(sc->sc_dmat, mhm->mhm_segs, mhm->mhm_seg_count);
8351 	kmem_free(mhm->mhm_segs, sizeof(*mhm->mhm_segs) * mhm->mhm_seg_count);
8352 
8353 	mhm->mhm_npages = 0;
8354 }
8355 
8356 #if NKSTAT > 0
8357 struct mcx_ppcnt {
8358 	char			 name[KSTAT_KV_NAMELEN];
8359 	enum kstat_kv_unit	 unit;
8360 };
8361 
8362 static const struct mcx_ppcnt mcx_ppcnt_ieee8023_tpl[] = {
8363 	{ "Good Tx",		KSTAT_KV_U_PACKETS, },
8364 	{ "Good Rx",		KSTAT_KV_U_PACKETS, },
8365 	{ "FCS errs",		KSTAT_KV_U_PACKETS, },
8366 	{ "Alignment Errs",	KSTAT_KV_U_PACKETS, },
8367 	{ "Good Tx",		KSTAT_KV_U_BYTES, },
8368 	{ "Good Rx",		KSTAT_KV_U_BYTES, },
8369 	{ "Multicast Tx",	KSTAT_KV_U_PACKETS, },
8370 	{ "Broadcast Tx",	KSTAT_KV_U_PACKETS, },
8371 	{ "Multicast Rx",	KSTAT_KV_U_PACKETS, },
8372 	{ "Broadcast Rx",	KSTAT_KV_U_PACKETS, },
8373 	{ "In Range Len",	KSTAT_KV_U_PACKETS, },
8374 	{ "Out Of Range Len",	KSTAT_KV_U_PACKETS, },
8375 	{ "Frame Too Long",	KSTAT_KV_U_PACKETS, },
8376 	{ "Symbol Errs",	KSTAT_KV_U_PACKETS, },
8377 	{ "MAC Ctrl Tx",	KSTAT_KV_U_PACKETS, },
8378 	{ "MAC Ctrl Rx",	KSTAT_KV_U_PACKETS, },
8379 	{ "MAC Ctrl Unsup",	KSTAT_KV_U_PACKETS, },
8380 	{ "Pause Rx",		KSTAT_KV_U_PACKETS, },
8381 	{ "Pause Tx",		KSTAT_KV_U_PACKETS, },
8382 };
8383 CTASSERT(__arraycount(mcx_ppcnt_ieee8023_tpl) == mcx_ppcnt_ieee8023_count);
8384 
8385 static const struct mcx_ppcnt mcx_ppcnt_rfc2863_tpl[] = {
8386 	{ "Rx Bytes",		KSTAT_KV_U_BYTES, },
8387 	{ "Rx Unicast",		KSTAT_KV_U_PACKETS, },
8388 	{ "Rx Discards",	KSTAT_KV_U_PACKETS, },
8389 	{ "Rx Errors",		KSTAT_KV_U_PACKETS, },
8390 	{ "Rx Unknown Proto",	KSTAT_KV_U_PACKETS, },
8391 	{ "Tx Bytes",		KSTAT_KV_U_BYTES, },
8392 	{ "Tx Unicast",		KSTAT_KV_U_PACKETS, },
8393 	{ "Tx Discards",	KSTAT_KV_U_PACKETS, },
8394 	{ "Tx Errors",		KSTAT_KV_U_PACKETS, },
8395 	{ "Rx Multicast",	KSTAT_KV_U_PACKETS, },
8396 	{ "Rx Broadcast",	KSTAT_KV_U_PACKETS, },
8397 	{ "Tx Multicast",	KSTAT_KV_U_PACKETS, },
8398 	{ "Tx Broadcast",	KSTAT_KV_U_PACKETS, },
8399 };
8400 CTASSERT(__arraycount(mcx_ppcnt_rfc2863_tpl) == mcx_ppcnt_rfc2863_count);
8401 
8402 static const struct mcx_ppcnt mcx_ppcnt_rfc2819_tpl[] = {
8403 	{ "Drop Events",	KSTAT_KV_U_PACKETS, },
8404 	{ "Octets",		KSTAT_KV_U_BYTES, },
8405 	{ "Packets",		KSTAT_KV_U_PACKETS, },
8406 	{ "Broadcasts",		KSTAT_KV_U_PACKETS, },
8407 	{ "Multicasts",		KSTAT_KV_U_PACKETS, },
8408 	{ "CRC Align Errs",	KSTAT_KV_U_PACKETS, },
8409 	{ "Undersize",		KSTAT_KV_U_PACKETS, },
8410 	{ "Oversize",		KSTAT_KV_U_PACKETS, },
8411 	{ "Fragments",		KSTAT_KV_U_PACKETS, },
8412 	{ "Jabbers",		KSTAT_KV_U_PACKETS, },
8413 	{ "Collisions",		KSTAT_KV_U_NONE, },
8414 	{ "64B",		KSTAT_KV_U_PACKETS, },
8415 	{ "65-127B",		KSTAT_KV_U_PACKETS, },
8416 	{ "128-255B",		KSTAT_KV_U_PACKETS, },
8417 	{ "256-511B",		KSTAT_KV_U_PACKETS, },
8418 	{ "512-1023B",		KSTAT_KV_U_PACKETS, },
8419 	{ "1024-1518B",		KSTAT_KV_U_PACKETS, },
8420 	{ "1519-2047B",		KSTAT_KV_U_PACKETS, },
8421 	{ "2048-4095B",		KSTAT_KV_U_PACKETS, },
8422 	{ "4096-8191B",		KSTAT_KV_U_PACKETS, },
8423 	{ "8192-10239B",	KSTAT_KV_U_PACKETS, },
8424 };
8425 CTASSERT(__arraycount(mcx_ppcnt_rfc2819_tpl) == mcx_ppcnt_rfc2819_count);
8426 
8427 static const struct mcx_ppcnt mcx_ppcnt_rfc3635_tpl[] = {
8428 	{ "Alignment Errs",	KSTAT_KV_U_PACKETS, },
8429 	{ "FCS Errs",		KSTAT_KV_U_PACKETS, },
8430 	{ "Single Colls",	KSTAT_KV_U_PACKETS, },
8431 	{ "Multiple Colls",	KSTAT_KV_U_PACKETS, },
8432 	{ "SQE Test Errs",	KSTAT_KV_U_NONE, },
8433 	{ "Deferred Tx",	KSTAT_KV_U_PACKETS, },
8434 	{ "Late Colls",		KSTAT_KV_U_NONE, },
8435 	{ "Exess Colls",	KSTAT_KV_U_NONE, },
8436 	{ "Int MAC Tx Errs",	KSTAT_KV_U_PACKETS, },
8437 	{ "CSM Sense Errs",	KSTAT_KV_U_NONE, },
8438 	{ "Too Long",		KSTAT_KV_U_PACKETS, },
8439 	{ "Int MAC Rx Errs",	KSTAT_KV_U_PACKETS, },
8440 	{ "Symbol Errs",	KSTAT_KV_U_NONE, },
8441 	{ "Unknown Control",	KSTAT_KV_U_PACKETS, },
8442 	{ "Pause Rx",		KSTAT_KV_U_PACKETS, },
8443 	{ "Pause Tx",		KSTAT_KV_U_PACKETS, },
8444 };
8445 CTASSERT(__arraycount(mcx_ppcnt_rfc3635_tpl) == mcx_ppcnt_rfc3635_count);
8446 
8447 struct mcx_kstat_ppcnt {
8448 	const char		*ksp_name;
8449 	const struct mcx_ppcnt	*ksp_tpl;
8450 	unsigned int		 ksp_n;
8451 	uint8_t			 ksp_grp;
8452 };
8453 
8454 static const struct mcx_kstat_ppcnt mcx_kstat_ppcnt_ieee8023 = {
8455 	.ksp_name =		"ieee802.3",
8456 	.ksp_tpl =		mcx_ppcnt_ieee8023_tpl,
8457 	.ksp_n =		__arraycount(mcx_ppcnt_ieee8023_tpl),
8458 	.ksp_grp =		MCX_REG_PPCNT_GRP_IEEE8023,
8459 };
8460 
8461 static const struct mcx_kstat_ppcnt mcx_kstat_ppcnt_rfc2863 = {
8462 	.ksp_name =		"rfc2863",
8463 	.ksp_tpl =		mcx_ppcnt_rfc2863_tpl,
8464 	.ksp_n =		__arraycount(mcx_ppcnt_rfc2863_tpl),
8465 	.ksp_grp =		MCX_REG_PPCNT_GRP_RFC2863,
8466 };
8467 
8468 static const struct mcx_kstat_ppcnt mcx_kstat_ppcnt_rfc2819 = {
8469 	.ksp_name =		"rfc2819",
8470 	.ksp_tpl =		mcx_ppcnt_rfc2819_tpl,
8471 	.ksp_n =		__arraycount(mcx_ppcnt_rfc2819_tpl),
8472 	.ksp_grp =		MCX_REG_PPCNT_GRP_RFC2819,
8473 };
8474 
8475 static const struct mcx_kstat_ppcnt mcx_kstat_ppcnt_rfc3635 = {
8476 	.ksp_name =		"rfc3635",
8477 	.ksp_tpl =		mcx_ppcnt_rfc3635_tpl,
8478 	.ksp_n =		__arraycount(mcx_ppcnt_rfc3635_tpl),
8479 	.ksp_grp =		MCX_REG_PPCNT_GRP_RFC3635,
8480 };
8481 
8482 static int	mcx_kstat_ppcnt_read(struct kstat *);
8483 
8484 static void	mcx_kstat_attach_tmps(struct mcx_softc *sc);
8485 static void	mcx_kstat_attach_queues(struct mcx_softc *sc);
8486 
8487 static struct kstat *
8488 mcx_kstat_attach_ppcnt(struct mcx_softc *sc,
8489     const struct mcx_kstat_ppcnt *ksp)
8490 {
8491 	struct kstat *ks;
8492 	struct kstat_kv *kvs;
8493 	unsigned int i;
8494 
8495 	ks = kstat_create(DEVNAME(sc), 0, ksp->ksp_name, 0, KSTAT_T_KV, 0);
8496 	if (ks == NULL)
8497 		return (NULL);
8498 
8499 	kvs = mallocarray(ksp->ksp_n, sizeof(*kvs),
8500 	    M_DEVBUF, M_WAITOK);
8501 
8502 	for (i = 0; i < ksp->ksp_n; i++) {
8503 		const struct mcx_ppcnt *tpl = &ksp->ksp_tpl[i];
8504 
8505 		kstat_kv_unit_init(&kvs[i], tpl->name,
8506 		    KSTAT_KV_T_COUNTER64, tpl->unit);
8507 	}
8508 
8509 	ks->ks_softc = sc;
8510 	ks->ks_ptr = (void *)ksp;
8511 	ks->ks_data = kvs;
8512 	ks->ks_datalen = ksp->ksp_n * sizeof(*kvs);
8513 	ks->ks_read = mcx_kstat_ppcnt_read;
8514 
8515 	kstat_install(ks);
8516 
8517 	return (ks);
8518 }
8519 
8520 static void
8521 mcx_kstat_attach(struct mcx_softc *sc)
8522 {
8523 	sc->sc_kstat_ieee8023 = mcx_kstat_attach_ppcnt(sc,
8524 	    &mcx_kstat_ppcnt_ieee8023);
8525 	sc->sc_kstat_rfc2863 = mcx_kstat_attach_ppcnt(sc,
8526 	    &mcx_kstat_ppcnt_rfc2863);
8527 	sc->sc_kstat_rfc2819 = mcx_kstat_attach_ppcnt(sc,
8528 	    &mcx_kstat_ppcnt_rfc2819);
8529 	sc->sc_kstat_rfc3635 = mcx_kstat_attach_ppcnt(sc,
8530 	    &mcx_kstat_ppcnt_rfc3635);
8531 
8532 	mcx_kstat_attach_tmps(sc);
8533 	mcx_kstat_attach_queues(sc);
8534 }
8535 
8536 static int
8537 mcx_kstat_ppcnt_read(struct kstat *ks)
8538 {
8539 	struct mcx_softc *sc = ks->ks_softc;
8540 	struct mcx_kstat_ppcnt *ksp = ks->ks_ptr;
8541 	struct mcx_reg_ppcnt ppcnt = {
8542 		.ppcnt_grp = ksp->ksp_grp,
8543 		.ppcnt_local_port = 1,
8544 	};
8545 	struct kstat_kv *kvs = ks->ks_data;
8546 	uint64_t *vs = (uint64_t *)&ppcnt.ppcnt_counter_set;
8547 	unsigned int i;
8548 	int rv;
8549 
8550 	KERNEL_LOCK(); /* XXX */
8551 	rv = mcx_access_hca_reg(sc, MCX_REG_PPCNT, MCX_REG_OP_READ,
8552 	    &ppcnt, sizeof(ppcnt));
8553 	KERNEL_UNLOCK();
8554 	if (rv != 0)
8555 		return (EIO);
8556 
8557 	nanouptime(&ks->ks_updated);
8558 
8559 	for (i = 0; i < ksp->ksp_n; i++)
8560 		kstat_kv_u64(&kvs[i]) = bemtoh64(&vs[i]);
8561 
8562 	return (0);
8563 }
8564 
8565 struct mcx_kstat_mtmp {
8566 	struct kstat_kv		ktmp_name;
8567 	struct kstat_kv		ktmp_temperature;
8568 	struct kstat_kv		ktmp_threshold_lo;
8569 	struct kstat_kv		ktmp_threshold_hi;
8570 };
8571 
8572 static const struct mcx_kstat_mtmp mcx_kstat_mtmp_tpl = {
8573 	KSTAT_KV_INITIALIZER("name",		KSTAT_KV_T_ISTR),
8574 	KSTAT_KV_INITIALIZER("temperature",	KSTAT_KV_T_TEMP),
8575 	KSTAT_KV_INITIALIZER("lo threshold",	KSTAT_KV_T_TEMP),
8576 	KSTAT_KV_INITIALIZER("hi threshold",	KSTAT_KV_T_TEMP),
8577 };
8578 
8579 static const struct timeval mcx_kstat_mtmp_rate = { 1, 0 };
8580 
8581 static int mcx_kstat_mtmp_read(struct kstat *);
8582 
8583 static void
8584 mcx_kstat_attach_tmps(struct mcx_softc *sc)
8585 {
8586 	struct kstat *ks;
8587 	struct mcx_reg_mcam mcam;
8588 	struct mcx_reg_mtcap mtcap;
8589 	struct mcx_kstat_mtmp *ktmp;
8590 	uint64_t map;
8591 	unsigned int i, n;
8592 
8593 	memset(&mtcap, 0, sizeof(mtcap));
8594 	memset(&mcam, 0, sizeof(mcam));
8595 
8596 	if (sc->sc_mcam_reg == 0) {
8597 		/* no management capabilities */
8598 		return;
8599 	}
8600 
8601 	if (mcx_access_hca_reg(sc, MCX_REG_MCAM, MCX_REG_OP_READ,
8602 	    &mcam, sizeof(mcam)) != 0) {
8603 		/* unable to check management capabilities? */
8604 		return;
8605 	}
8606 
8607 	if (MCX_BITFIELD_BIT(mcam.mcam_feature_cap_mask,
8608 	    MCX_MCAM_FEATURE_CAP_SENSOR_MAP) == 0) {
8609 		/* no sensor map */
8610 		return;
8611 	}
8612 
8613 	if (mcx_access_hca_reg(sc, MCX_REG_MTCAP, MCX_REG_OP_READ,
8614 	    &mtcap, sizeof(mtcap)) != 0) {
8615 		/* unable to find temperature sensors */
8616 		return;
8617 	}
8618 
8619 	sc->sc_kstat_mtmp_count = mtcap.mtcap_sensor_count;
8620 	sc->sc_kstat_mtmp = mallocarray(sc->sc_kstat_mtmp_count,
8621 	    sizeof(*sc->sc_kstat_mtmp), M_DEVBUF, M_WAITOK);
8622 
8623 	n = 0;
8624 	map = bemtoh64(&mtcap.mtcap_sensor_map);
8625 	for (i = 0; i < sizeof(map) * NBBY; i++) {
8626 		if (!ISSET(map, (1ULL << i)))
8627 			continue;
8628 
8629 		ks = kstat_create(DEVNAME(sc), 0, "temperature", i,
8630 		    KSTAT_T_KV, 0);
8631 		if (ks == NULL) {
8632 			/* unable to attach temperature sensor %u, i */
8633 			continue;
8634 		}
8635 
8636 		ktmp = malloc(sizeof(*ktmp), M_DEVBUF, M_WAITOK|M_ZERO);
8637 		*ktmp = mcx_kstat_mtmp_tpl;
8638 
8639 		ks->ks_data = ktmp;
8640 		ks->ks_datalen = sizeof(*ktmp);
8641 		TIMEVAL_TO_TIMESPEC(&mcx_kstat_mtmp_rate, &ks->ks_interval);
8642 		ks->ks_read = mcx_kstat_mtmp_read;
8643 
8644 		ks->ks_softc = sc;
8645 		kstat_install(ks);
8646 
8647 		sc->sc_kstat_mtmp[n++] = ks;
8648 		if (n >= sc->sc_kstat_mtmp_count)
8649 			break;
8650 	}
8651 }
8652 
8653 static uint64_t
8654 mcx_tmp_to_uK(uint16_t *t)
8655 {
8656 	int64_t mt = (int16_t)bemtoh16(t); /* 0.125 C units */
8657 	mt *= 1000000 / 8; /* convert to uC */
8658 	mt += 273150000; /* convert to uK */
8659 
8660 	return (mt);
8661 }
8662 
8663 static int
8664 mcx_kstat_mtmp_read(struct kstat *ks)
8665 {
8666 	struct mcx_softc *sc = ks->ks_softc;
8667 	struct mcx_kstat_mtmp *ktmp = ks->ks_data;
8668 	struct mcx_reg_mtmp mtmp;
8669 	int rv;
8670 	struct timeval updated;
8671 
8672 	TIMESPEC_TO_TIMEVAL(&updated, &ks->ks_updated);
8673 
8674 	if (!ratecheck(&updated, &mcx_kstat_mtmp_rate))
8675 		return (0);
8676 
8677 	memset(&mtmp, 0, sizeof(mtmp));
8678 	htobem16(&mtmp.mtmp_sensor_index, ks->ks_unit);
8679 
8680 	KERNEL_LOCK(); /* XXX */
8681 	rv = mcx_access_hca_reg(sc, MCX_REG_MTMP, MCX_REG_OP_READ,
8682 	    &mtmp, sizeof(mtmp));
8683 	KERNEL_UNLOCK();
8684 	if (rv != 0)
8685 		return (EIO);
8686 
8687 	memset(kstat_kv_istr(&ktmp->ktmp_name), 0,
8688 	    sizeof(kstat_kv_istr(&ktmp->ktmp_name)));
8689 	memcpy(kstat_kv_istr(&ktmp->ktmp_name),
8690 	    mtmp.mtmp_sensor_name, sizeof(mtmp.mtmp_sensor_name));
8691 	kstat_kv_temp(&ktmp->ktmp_temperature) =
8692 	    mcx_tmp_to_uK(&mtmp.mtmp_temperature);
8693 	kstat_kv_temp(&ktmp->ktmp_threshold_lo) =
8694 	    mcx_tmp_to_uK(&mtmp.mtmp_temperature_threshold_lo);
8695 	kstat_kv_temp(&ktmp->ktmp_threshold_hi) =
8696 	    mcx_tmp_to_uK(&mtmp.mtmp_temperature_threshold_hi);
8697 
8698 	TIMEVAL_TO_TIMESPEC(&updated, &ks->ks_updated);
8699 
8700 	return (0);
8701 }
8702 
8703 struct mcx_queuestat {
8704 	char			 name[KSTAT_KV_NAMELEN];
8705 	enum kstat_kv_type	 type;
8706 };
8707 
8708 static const struct mcx_queuestat mcx_queue_kstat_tpl[] = {
8709 	{ "RQ SW prod",		KSTAT_KV_T_COUNTER64 },
8710 	{ "RQ HW prod",		KSTAT_KV_T_COUNTER64 },
8711 	{ "RQ HW cons",		KSTAT_KV_T_COUNTER64 },
8712 	{ "RQ HW state",	KSTAT_KV_T_ISTR },
8713 
8714 	{ "SQ SW prod",		KSTAT_KV_T_COUNTER64 },
8715 	{ "SQ SW cons",		KSTAT_KV_T_COUNTER64 },
8716 	{ "SQ HW prod",		KSTAT_KV_T_COUNTER64 },
8717 	{ "SQ HW cons",		KSTAT_KV_T_COUNTER64 },
8718 	{ "SQ HW state",	KSTAT_KV_T_ISTR },
8719 
8720 	{ "CQ SW cons",		KSTAT_KV_T_COUNTER64 },
8721 	{ "CQ HW prod",		KSTAT_KV_T_COUNTER64 },
8722 	{ "CQ HW cons",		KSTAT_KV_T_COUNTER64 },
8723 	{ "CQ HW notify",	KSTAT_KV_T_COUNTER64 },
8724 	{ "CQ HW solicit",	KSTAT_KV_T_COUNTER64 },
8725 	{ "CQ HW status",	KSTAT_KV_T_ISTR },
8726 	{ "CQ HW state",	KSTAT_KV_T_ISTR },
8727 
8728 	{ "EQ SW cons",		KSTAT_KV_T_COUNTER64 },
8729 	{ "EQ HW prod",		KSTAT_KV_T_COUNTER64 },
8730 	{ "EQ HW cons",		KSTAT_KV_T_COUNTER64 },
8731 	{ "EQ HW status",	KSTAT_KV_T_ISTR },
8732 	{ "EQ HW state",	KSTAT_KV_T_ISTR },
8733 };
8734 
8735 static int	mcx_kstat_queue_read(struct kstat *);
8736 
8737 static void
8738 mcx_kstat_attach_queues(struct mcx_softc *sc)
8739 {
8740 	struct kstat *ks;
8741 	struct kstat_kv *kvs;
8742 	int q, i;
8743 
8744 	for (q = 0; q < sc->sc_nqueues; q++) {
8745 		ks = kstat_create(DEVNAME(sc), 0, "mcx-queues", q,
8746 		    KSTAT_T_KV, 0);
8747 		if (ks == NULL) {
8748 			/* unable to attach queue stats %u, q */
8749 			continue;
8750 		}
8751 
8752 		kvs = mallocarray(nitems(mcx_queue_kstat_tpl),
8753 		    sizeof(*kvs), M_DEVBUF, M_WAITOK);
8754 
8755 		for (i = 0; i < nitems(mcx_queue_kstat_tpl); i++) {
8756 			const struct mcx_queuestat *tpl =
8757 			    &mcx_queue_kstat_tpl[i];
8758 
8759 			kstat_kv_init(&kvs[i], tpl->name, tpl->type);
8760 		}
8761 
8762 		ks->ks_softc = &sc->sc_queues[q];
8763 		ks->ks_data = kvs;
8764 		ks->ks_datalen = nitems(mcx_queue_kstat_tpl) * sizeof(*kvs);
8765 		ks->ks_read = mcx_kstat_queue_read;
8766 
8767 		sc->sc_queues[q].q_kstat = ks;
8768 		kstat_install(ks);
8769 	}
8770 }
8771 
8772 static int
8773 mcx_kstat_queue_read(struct kstat *ks)
8774 {
8775 	struct mcx_queues *q = ks->ks_softc;
8776 	struct mcx_softc *sc = q->q_sc;
8777 	struct kstat_kv *kvs = ks->ks_data;
8778 	union {
8779 		struct mcx_rq_ctx rq;
8780 		struct mcx_sq_ctx sq;
8781 		struct mcx_cq_ctx cq;
8782 		struct mcx_eq_ctx eq;
8783 	} u;
8784 	const char *text;
8785 	int error = 0;
8786 
8787 	KERNEL_LOCK();
8788 
8789 	if (mcx_query_rq(sc, &q->q_rx, &u.rq) != 0) {
8790 		error = EIO;
8791 		goto out;
8792 	}
8793 
8794 	kstat_kv_u64(kvs++) = q->q_rx.rx_prod;
8795 	kstat_kv_u64(kvs++) = bemtoh32(&u.rq.rq_wq.wq_sw_counter);
8796 	kstat_kv_u64(kvs++) = bemtoh32(&u.rq.rq_wq.wq_hw_counter);
8797 	switch ((bemtoh32(&u.rq.rq_flags) & MCX_RQ_CTX_STATE_MASK) >>
8798 	    MCX_RQ_CTX_STATE_SHIFT) {
8799 	case MCX_RQ_CTX_STATE_RST:
8800 		text = "RST";
8801 		break;
8802 	case MCX_RQ_CTX_STATE_RDY:
8803 		text = "RDY";
8804 		break;
8805 	case MCX_RQ_CTX_STATE_ERR:
8806 		text = "ERR";
8807 		break;
8808 	default:
8809 		text = "unknown";
8810 		break;
8811 	}
8812 	strlcpy(kstat_kv_istr(kvs), text, sizeof(kstat_kv_istr(kvs)));
8813 	kvs++;
8814 
8815 	if (mcx_query_sq(sc, &q->q_tx, &u.sq) != 0) {
8816 		error = EIO;
8817 		goto out;
8818 	}
8819 
8820 	kstat_kv_u64(kvs++) = q->q_tx.tx_prod;
8821 	kstat_kv_u64(kvs++) = q->q_tx.tx_cons;
8822 	kstat_kv_u64(kvs++) = bemtoh32(&u.sq.sq_wq.wq_sw_counter);
8823 	kstat_kv_u64(kvs++) = bemtoh32(&u.sq.sq_wq.wq_hw_counter);
8824 	switch ((bemtoh32(&u.sq.sq_flags) & MCX_SQ_CTX_STATE_MASK) >>
8825 	    MCX_SQ_CTX_STATE_SHIFT) {
8826 	case MCX_SQ_CTX_STATE_RST:
8827 		text = "RST";
8828 		break;
8829 	case MCX_SQ_CTX_STATE_RDY:
8830 		text = "RDY";
8831 		break;
8832 	case MCX_SQ_CTX_STATE_ERR:
8833 		text = "ERR";
8834 		break;
8835 	default:
8836 		text = "unknown";
8837 		break;
8838 	}
8839 	strlcpy(kstat_kv_istr(kvs), text, sizeof(kstat_kv_istr(kvs)));
8840 	kvs++;
8841 
8842 	if (mcx_query_cq(sc, &q->q_cq, &u.cq) != 0) {
8843 		error = EIO;
8844 		goto out;
8845 	}
8846 
8847 	kstat_kv_u64(kvs++) = q->q_cq.cq_cons;
8848 	kstat_kv_u64(kvs++) = bemtoh32(&u.cq.cq_producer_counter);
8849 	kstat_kv_u64(kvs++) = bemtoh32(&u.cq.cq_consumer_counter);
8850 	kstat_kv_u64(kvs++) = bemtoh32(&u.cq.cq_last_notified);
8851 	kstat_kv_u64(kvs++) = bemtoh32(&u.cq.cq_last_solicit);
8852 
8853 	switch ((bemtoh32(&u.cq.cq_status) & MCX_CQ_CTX_STATUS_MASK) >>
8854 	    MCX_CQ_CTX_STATUS_SHIFT) {
8855 	case MCX_CQ_CTX_STATUS_OK:
8856 		text = "OK";
8857 		break;
8858 	case MCX_CQ_CTX_STATUS_OVERFLOW:
8859 		text = "overflow";
8860 		break;
8861 	case MCX_CQ_CTX_STATUS_WRITE_FAIL:
8862 		text = "write fail";
8863 		break;
8864 	default:
8865 		text = "unknown";
8866 		break;
8867 	}
8868 	strlcpy(kstat_kv_istr(kvs), text, sizeof(kstat_kv_istr(kvs)));
8869 	kvs++;
8870 
8871 	switch ((bemtoh32(&u.cq.cq_status) & MCX_CQ_CTX_STATE_MASK) >>
8872 	    MCX_CQ_CTX_STATE_SHIFT) {
8873 	case MCX_CQ_CTX_STATE_SOLICITED:
8874 		text = "solicited";
8875 		break;
8876 	case MCX_CQ_CTX_STATE_ARMED:
8877 		text = "armed";
8878 		break;
8879 	case MCX_CQ_CTX_STATE_FIRED:
8880 		text = "fired";
8881 		break;
8882 	default:
8883 		text = "unknown";
8884 		break;
8885 	}
8886 	strlcpy(kstat_kv_istr(kvs), text, sizeof(kstat_kv_istr(kvs)));
8887 	kvs++;
8888 
8889 	if (mcx_query_eq(sc, &q->q_eq, &u.eq) != 0) {
8890 		error = EIO;
8891 		goto out;
8892 	}
8893 
8894 	kstat_kv_u64(kvs++) = q->q_eq.eq_cons;
8895 	kstat_kv_u64(kvs++) = bemtoh32(&u.eq.eq_producer_counter);
8896 	kstat_kv_u64(kvs++) = bemtoh32(&u.eq.eq_consumer_counter);
8897 
8898 	switch ((bemtoh32(&u.eq.eq_status) & MCX_EQ_CTX_STATUS_MASK) >>
8899 	    MCX_EQ_CTX_STATUS_SHIFT) {
8900 	case MCX_EQ_CTX_STATUS_EQ_WRITE_FAILURE:
8901 		text = "write fail";
8902 		break;
8903 	case MCX_EQ_CTX_STATUS_OK:
8904 		text = "OK";
8905 		break;
8906 	default:
8907 		text = "unknown";
8908 		break;
8909 	}
8910 	strlcpy(kstat_kv_istr(kvs), text, sizeof(kstat_kv_istr(kvs)));
8911 	kvs++;
8912 
8913 	switch ((bemtoh32(&u.eq.eq_status) & MCX_EQ_CTX_STATE_MASK) >>
8914 	    MCX_EQ_CTX_STATE_SHIFT) {
8915 	case MCX_EQ_CTX_STATE_ARMED:
8916 		text = "armed";
8917 		break;
8918 	case MCX_EQ_CTX_STATE_FIRED:
8919 		text = "fired";
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 	nanouptime(&ks->ks_updated);
8929 out:
8930 	KERNEL_UNLOCK();
8931 	return (error);
8932 }
8933 
8934 #endif /* NKSTAT > 0 */
8935 
8936 static unsigned int
8937 mcx_timecounter_read(struct timecounter *tc)
8938 {
8939 	struct mcx_softc *sc = tc->tc_priv;
8940 
8941 	return (mcx_rd(sc, MCX_INTERNAL_TIMER_L));
8942 }
8943 
8944 static void
8945 mcx_timecounter_attach(struct mcx_softc *sc)
8946 {
8947 	struct timecounter *tc = &sc->sc_timecounter;
8948 
8949 	tc->tc_get_timecount = mcx_timecounter_read;
8950 	tc->tc_counter_mask = ~0U;
8951 	tc->tc_frequency = sc->sc_khz * 1000;
8952 	tc->tc_name = device_xname(sc->sc_dev);
8953 	tc->tc_quality = -100;
8954 	tc->tc_priv = sc;
8955 
8956 	tc_init(tc);
8957 }
8958