xref: /freebsd/sys/dev/sfxge/common/ef10_ev.c (revision edf8578117e8844e02c0121147f45e4609b30680)
1 /*-
2  * Copyright (c) 2012-2016 Solarflare Communications Inc.
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions are met:
7  *
8  * 1. Redistributions of source code must retain the above copyright notice,
9  *    this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright notice,
11  *    this list of conditions and the following disclaimer in the documentation
12  *    and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
15  * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,
16  * THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
17  * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
18  * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
19  * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
20  * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
21  * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
22  * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
23  * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE,
24  * EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
25  *
26  * The views and conclusions contained in the software and documentation are
27  * those of the authors and should not be interpreted as representing official
28  * policies, either expressed or implied, of the FreeBSD Project.
29  */
30 
31 #include <sys/cdefs.h>
32 #include "efx.h"
33 #include "efx_impl.h"
34 #if EFSYS_OPT_MON_STATS
35 #include "mcdi_mon.h"
36 #endif
37 
38 #if EFSYS_OPT_HUNTINGTON || EFSYS_OPT_MEDFORD || EFSYS_OPT_MEDFORD2
39 
40 #if EFSYS_OPT_QSTATS
41 #define	EFX_EV_QSTAT_INCR(_eep, _stat)					\
42 	do {								\
43 		(_eep)->ee_stat[_stat]++;				\
44 	_NOTE(CONSTANTCONDITION)					\
45 	} while (B_FALSE)
46 #else
47 #define	EFX_EV_QSTAT_INCR(_eep, _stat)
48 #endif
49 
50 /*
51  * Non-interrupting event queue requires interrrupting event queue to
52  * refer to for wake-up events even if wake ups are never used.
53  * It could be even non-allocated event queue.
54  */
55 #define	EFX_EF10_ALWAYS_INTERRUPTING_EVQ_INDEX	(0)
56 
57 static	__checkReturn	boolean_t
58 ef10_ev_rx(
59 	__in		efx_evq_t *eep,
60 	__in		efx_qword_t *eqp,
61 	__in		const efx_ev_callbacks_t *eecp,
62 	__in_opt	void *arg);
63 
64 static	__checkReturn	boolean_t
65 ef10_ev_tx(
66 	__in		efx_evq_t *eep,
67 	__in		efx_qword_t *eqp,
68 	__in		const efx_ev_callbacks_t *eecp,
69 	__in_opt	void *arg);
70 
71 static	__checkReturn	boolean_t
72 ef10_ev_driver(
73 	__in		efx_evq_t *eep,
74 	__in		efx_qword_t *eqp,
75 	__in		const efx_ev_callbacks_t *eecp,
76 	__in_opt	void *arg);
77 
78 static	__checkReturn	boolean_t
79 ef10_ev_drv_gen(
80 	__in		efx_evq_t *eep,
81 	__in		efx_qword_t *eqp,
82 	__in		const efx_ev_callbacks_t *eecp,
83 	__in_opt	void *arg);
84 
85 static	__checkReturn	boolean_t
86 ef10_ev_mcdi(
87 	__in		efx_evq_t *eep,
88 	__in		efx_qword_t *eqp,
89 	__in		const efx_ev_callbacks_t *eecp,
90 	__in_opt	void *arg);
91 
92 static	__checkReturn	efx_rc_t
93 efx_mcdi_set_evq_tmr(
94 	__in		efx_nic_t *enp,
95 	__in		uint32_t instance,
96 	__in		uint32_t mode,
97 	__in		uint32_t timer_ns)
98 {
99 	efx_mcdi_req_t req;
100 	EFX_MCDI_DECLARE_BUF(payload, MC_CMD_SET_EVQ_TMR_IN_LEN,
101 		MC_CMD_SET_EVQ_TMR_OUT_LEN);
102 	efx_rc_t rc;
103 
104 	req.emr_cmd = MC_CMD_SET_EVQ_TMR;
105 	req.emr_in_buf = payload;
106 	req.emr_in_length = MC_CMD_SET_EVQ_TMR_IN_LEN;
107 	req.emr_out_buf = payload;
108 	req.emr_out_length = MC_CMD_SET_EVQ_TMR_OUT_LEN;
109 
110 	MCDI_IN_SET_DWORD(req, SET_EVQ_TMR_IN_INSTANCE, instance);
111 	MCDI_IN_SET_DWORD(req, SET_EVQ_TMR_IN_TMR_LOAD_REQ_NS, timer_ns);
112 	MCDI_IN_SET_DWORD(req, SET_EVQ_TMR_IN_TMR_RELOAD_REQ_NS, timer_ns);
113 	MCDI_IN_SET_DWORD(req, SET_EVQ_TMR_IN_TMR_MODE, mode);
114 
115 	efx_mcdi_execute(enp, &req);
116 
117 	if (req.emr_rc != 0) {
118 		rc = req.emr_rc;
119 		goto fail1;
120 	}
121 
122 	if (req.emr_out_length_used < MC_CMD_SET_EVQ_TMR_OUT_LEN) {
123 		rc = EMSGSIZE;
124 		goto fail2;
125 	}
126 
127 	return (0);
128 
129 fail2:
130 	EFSYS_PROBE(fail2);
131 fail1:
132 	EFSYS_PROBE1(fail1, efx_rc_t, rc);
133 
134 	return (rc);
135 }
136 
137 static	__checkReturn	efx_rc_t
138 efx_mcdi_init_evq(
139 	__in		efx_nic_t *enp,
140 	__in		unsigned int instance,
141 	__in		efsys_mem_t *esmp,
142 	__in		size_t nevs,
143 	__in		uint32_t irq,
144 	__in		uint32_t us,
145 	__in		uint32_t flags,
146 	__in		boolean_t low_latency)
147 {
148 	efx_mcdi_req_t req;
149 	EFX_MCDI_DECLARE_BUF(payload,
150 		MC_CMD_INIT_EVQ_IN_LEN(EFX_EVQ_NBUFS(EFX_EVQ_MAXNEVS)),
151 		MC_CMD_INIT_EVQ_OUT_LEN);
152 	efx_qword_t *dma_addr;
153 	uint64_t addr;
154 	int npages;
155 	int i;
156 	boolean_t interrupting;
157 	int ev_cut_through;
158 	efx_rc_t rc;
159 
160 	npages = EFX_EVQ_NBUFS(nevs);
161 	if (MC_CMD_INIT_EVQ_IN_LEN(npages) > MC_CMD_INIT_EVQ_IN_LENMAX) {
162 		rc = EINVAL;
163 		goto fail1;
164 	}
165 
166 	req.emr_cmd = MC_CMD_INIT_EVQ;
167 	req.emr_in_buf = payload;
168 	req.emr_in_length = MC_CMD_INIT_EVQ_IN_LEN(npages);
169 	req.emr_out_buf = payload;
170 	req.emr_out_length = MC_CMD_INIT_EVQ_OUT_LEN;
171 
172 	MCDI_IN_SET_DWORD(req, INIT_EVQ_IN_SIZE, nevs);
173 	MCDI_IN_SET_DWORD(req, INIT_EVQ_IN_INSTANCE, instance);
174 	MCDI_IN_SET_DWORD(req, INIT_EVQ_IN_IRQ_NUM, irq);
175 
176 	interrupting = ((flags & EFX_EVQ_FLAGS_NOTIFY_MASK) ==
177 	    EFX_EVQ_FLAGS_NOTIFY_INTERRUPT);
178 
179 	/*
180 	 * On Huntington RX and TX event batching can only be requested together
181 	 * (even if the datapath firmware doesn't actually support RX
182 	 * batching). If event cut through is enabled no RX batching will occur.
183 	 *
184 	 * So always enable RX and TX event batching, and enable event cut
185 	 * through if we want low latency operation.
186 	 */
187 	switch (flags & EFX_EVQ_FLAGS_TYPE_MASK) {
188 	case EFX_EVQ_FLAGS_TYPE_AUTO:
189 		ev_cut_through = low_latency ? 1 : 0;
190 		break;
191 	case EFX_EVQ_FLAGS_TYPE_THROUGHPUT:
192 		ev_cut_through = 0;
193 		break;
194 	case EFX_EVQ_FLAGS_TYPE_LOW_LATENCY:
195 		ev_cut_through = 1;
196 		break;
197 	default:
198 		rc = EINVAL;
199 		goto fail2;
200 	}
201 	MCDI_IN_POPULATE_DWORD_6(req, INIT_EVQ_IN_FLAGS,
202 	    INIT_EVQ_IN_FLAG_INTERRUPTING, interrupting,
203 	    INIT_EVQ_IN_FLAG_RPTR_DOS, 0,
204 	    INIT_EVQ_IN_FLAG_INT_ARMD, 0,
205 	    INIT_EVQ_IN_FLAG_CUT_THRU, ev_cut_through,
206 	    INIT_EVQ_IN_FLAG_RX_MERGE, 1,
207 	    INIT_EVQ_IN_FLAG_TX_MERGE, 1);
208 
209 	/* If the value is zero then disable the timer */
210 	if (us == 0) {
211 		MCDI_IN_SET_DWORD(req, INIT_EVQ_IN_TMR_MODE,
212 		    MC_CMD_INIT_EVQ_IN_TMR_MODE_DIS);
213 		MCDI_IN_SET_DWORD(req, INIT_EVQ_IN_TMR_LOAD, 0);
214 		MCDI_IN_SET_DWORD(req, INIT_EVQ_IN_TMR_RELOAD, 0);
215 	} else {
216 		unsigned int ticks;
217 
218 		if ((rc = efx_ev_usecs_to_ticks(enp, us, &ticks)) != 0)
219 			goto fail3;
220 
221 		MCDI_IN_SET_DWORD(req, INIT_EVQ_IN_TMR_MODE,
222 		    MC_CMD_INIT_EVQ_IN_TMR_INT_HLDOFF);
223 		MCDI_IN_SET_DWORD(req, INIT_EVQ_IN_TMR_LOAD, ticks);
224 		MCDI_IN_SET_DWORD(req, INIT_EVQ_IN_TMR_RELOAD, ticks);
225 	}
226 
227 	MCDI_IN_SET_DWORD(req, INIT_EVQ_IN_COUNT_MODE,
228 	    MC_CMD_INIT_EVQ_IN_COUNT_MODE_DIS);
229 	MCDI_IN_SET_DWORD(req, INIT_EVQ_IN_COUNT_THRSHLD, 0);
230 
231 	dma_addr = MCDI_IN2(req, efx_qword_t, INIT_EVQ_IN_DMA_ADDR);
232 	addr = EFSYS_MEM_ADDR(esmp);
233 
234 	for (i = 0; i < npages; i++) {
235 		EFX_POPULATE_QWORD_2(*dma_addr,
236 		    EFX_DWORD_1, (uint32_t)(addr >> 32),
237 		    EFX_DWORD_0, (uint32_t)(addr & 0xffffffff));
238 
239 		dma_addr++;
240 		addr += EFX_BUF_SIZE;
241 	}
242 
243 	efx_mcdi_execute(enp, &req);
244 
245 	if (req.emr_rc != 0) {
246 		rc = req.emr_rc;
247 		goto fail4;
248 	}
249 
250 	if (req.emr_out_length_used < MC_CMD_INIT_EVQ_OUT_LEN) {
251 		rc = EMSGSIZE;
252 		goto fail5;
253 	}
254 
255 	/* NOTE: ignore the returned IRQ param as firmware does not set it. */
256 
257 	return (0);
258 
259 fail5:
260 	EFSYS_PROBE(fail5);
261 fail4:
262 	EFSYS_PROBE(fail4);
263 fail3:
264 	EFSYS_PROBE(fail3);
265 fail2:
266 	EFSYS_PROBE(fail2);
267 fail1:
268 	EFSYS_PROBE1(fail1, efx_rc_t, rc);
269 
270 	return (rc);
271 }
272 
273 static	__checkReturn	efx_rc_t
274 efx_mcdi_init_evq_v2(
275 	__in		efx_nic_t *enp,
276 	__in		unsigned int instance,
277 	__in		efsys_mem_t *esmp,
278 	__in		size_t nevs,
279 	__in		uint32_t irq,
280 	__in		uint32_t us,
281 	__in		uint32_t flags)
282 {
283 	efx_mcdi_req_t req;
284 	EFX_MCDI_DECLARE_BUF(payload,
285 		MC_CMD_INIT_EVQ_V2_IN_LEN(EFX_EVQ_NBUFS(EFX_EVQ_MAXNEVS)),
286 		MC_CMD_INIT_EVQ_V2_OUT_LEN);
287 	boolean_t interrupting;
288 	unsigned int evq_type;
289 	efx_qword_t *dma_addr;
290 	uint64_t addr;
291 	int npages;
292 	int i;
293 	efx_rc_t rc;
294 
295 	npages = EFX_EVQ_NBUFS(nevs);
296 	if (MC_CMD_INIT_EVQ_V2_IN_LEN(npages) > MC_CMD_INIT_EVQ_V2_IN_LENMAX) {
297 		rc = EINVAL;
298 		goto fail1;
299 	}
300 
301 	req.emr_cmd = MC_CMD_INIT_EVQ;
302 	req.emr_in_buf = payload;
303 	req.emr_in_length = MC_CMD_INIT_EVQ_V2_IN_LEN(npages);
304 	req.emr_out_buf = payload;
305 	req.emr_out_length = MC_CMD_INIT_EVQ_V2_OUT_LEN;
306 
307 	MCDI_IN_SET_DWORD(req, INIT_EVQ_V2_IN_SIZE, nevs);
308 	MCDI_IN_SET_DWORD(req, INIT_EVQ_V2_IN_INSTANCE, instance);
309 	MCDI_IN_SET_DWORD(req, INIT_EVQ_V2_IN_IRQ_NUM, irq);
310 
311 	interrupting = ((flags & EFX_EVQ_FLAGS_NOTIFY_MASK) ==
312 	    EFX_EVQ_FLAGS_NOTIFY_INTERRUPT);
313 
314 	switch (flags & EFX_EVQ_FLAGS_TYPE_MASK) {
315 	case EFX_EVQ_FLAGS_TYPE_AUTO:
316 		evq_type = MC_CMD_INIT_EVQ_V2_IN_FLAG_TYPE_AUTO;
317 		break;
318 	case EFX_EVQ_FLAGS_TYPE_THROUGHPUT:
319 		evq_type = MC_CMD_INIT_EVQ_V2_IN_FLAG_TYPE_THROUGHPUT;
320 		break;
321 	case EFX_EVQ_FLAGS_TYPE_LOW_LATENCY:
322 		evq_type = MC_CMD_INIT_EVQ_V2_IN_FLAG_TYPE_LOW_LATENCY;
323 		break;
324 	default:
325 		rc = EINVAL;
326 		goto fail2;
327 	}
328 	MCDI_IN_POPULATE_DWORD_4(req, INIT_EVQ_V2_IN_FLAGS,
329 	    INIT_EVQ_V2_IN_FLAG_INTERRUPTING, interrupting,
330 	    INIT_EVQ_V2_IN_FLAG_RPTR_DOS, 0,
331 	    INIT_EVQ_V2_IN_FLAG_INT_ARMD, 0,
332 	    INIT_EVQ_V2_IN_FLAG_TYPE, evq_type);
333 
334 	/* If the value is zero then disable the timer */
335 	if (us == 0) {
336 		MCDI_IN_SET_DWORD(req, INIT_EVQ_V2_IN_TMR_MODE,
337 		    MC_CMD_INIT_EVQ_V2_IN_TMR_MODE_DIS);
338 		MCDI_IN_SET_DWORD(req, INIT_EVQ_V2_IN_TMR_LOAD, 0);
339 		MCDI_IN_SET_DWORD(req, INIT_EVQ_V2_IN_TMR_RELOAD, 0);
340 	} else {
341 		unsigned int ticks;
342 
343 		if ((rc = efx_ev_usecs_to_ticks(enp, us, &ticks)) != 0)
344 			goto fail3;
345 
346 		MCDI_IN_SET_DWORD(req, INIT_EVQ_V2_IN_TMR_MODE,
347 		    MC_CMD_INIT_EVQ_V2_IN_TMR_INT_HLDOFF);
348 		MCDI_IN_SET_DWORD(req, INIT_EVQ_V2_IN_TMR_LOAD, ticks);
349 		MCDI_IN_SET_DWORD(req, INIT_EVQ_V2_IN_TMR_RELOAD, ticks);
350 	}
351 
352 	MCDI_IN_SET_DWORD(req, INIT_EVQ_V2_IN_COUNT_MODE,
353 	    MC_CMD_INIT_EVQ_V2_IN_COUNT_MODE_DIS);
354 	MCDI_IN_SET_DWORD(req, INIT_EVQ_V2_IN_COUNT_THRSHLD, 0);
355 
356 	dma_addr = MCDI_IN2(req, efx_qword_t, INIT_EVQ_V2_IN_DMA_ADDR);
357 	addr = EFSYS_MEM_ADDR(esmp);
358 
359 	for (i = 0; i < npages; i++) {
360 		EFX_POPULATE_QWORD_2(*dma_addr,
361 		    EFX_DWORD_1, (uint32_t)(addr >> 32),
362 		    EFX_DWORD_0, (uint32_t)(addr & 0xffffffff));
363 
364 		dma_addr++;
365 		addr += EFX_BUF_SIZE;
366 	}
367 
368 	efx_mcdi_execute(enp, &req);
369 
370 	if (req.emr_rc != 0) {
371 		rc = req.emr_rc;
372 		goto fail4;
373 	}
374 
375 	if (req.emr_out_length_used < MC_CMD_INIT_EVQ_V2_OUT_LEN) {
376 		rc = EMSGSIZE;
377 		goto fail5;
378 	}
379 
380 	/* NOTE: ignore the returned IRQ param as firmware does not set it. */
381 
382 	EFSYS_PROBE1(mcdi_evq_flags, uint32_t,
383 		    MCDI_OUT_DWORD(req, INIT_EVQ_V2_OUT_FLAGS));
384 
385 	return (0);
386 
387 fail5:
388 	EFSYS_PROBE(fail5);
389 fail4:
390 	EFSYS_PROBE(fail4);
391 fail3:
392 	EFSYS_PROBE(fail3);
393 fail2:
394 	EFSYS_PROBE(fail2);
395 fail1:
396 	EFSYS_PROBE1(fail1, efx_rc_t, rc);
397 
398 	return (rc);
399 }
400 
401 static	__checkReturn	efx_rc_t
402 efx_mcdi_fini_evq(
403 	__in		efx_nic_t *enp,
404 	__in		uint32_t instance)
405 {
406 	efx_mcdi_req_t req;
407 	EFX_MCDI_DECLARE_BUF(payload, MC_CMD_FINI_EVQ_IN_LEN,
408 		MC_CMD_FINI_EVQ_OUT_LEN);
409 	efx_rc_t rc;
410 
411 	req.emr_cmd = MC_CMD_FINI_EVQ;
412 	req.emr_in_buf = payload;
413 	req.emr_in_length = MC_CMD_FINI_EVQ_IN_LEN;
414 	req.emr_out_buf = payload;
415 	req.emr_out_length = MC_CMD_FINI_EVQ_OUT_LEN;
416 
417 	MCDI_IN_SET_DWORD(req, FINI_EVQ_IN_INSTANCE, instance);
418 
419 	efx_mcdi_execute_quiet(enp, &req);
420 
421 	if (req.emr_rc != 0) {
422 		rc = req.emr_rc;
423 		goto fail1;
424 	}
425 
426 	return (0);
427 
428 fail1:
429 	/*
430 	 * EALREADY is not an error, but indicates that the MC has rebooted and
431 	 * that the EVQ has already been destroyed.
432 	 */
433 	if (rc != EALREADY)
434 		EFSYS_PROBE1(fail1, efx_rc_t, rc);
435 
436 	return (rc);
437 }
438 
439 	__checkReturn	efx_rc_t
440 ef10_ev_init(
441 	__in		efx_nic_t *enp)
442 {
443 	_NOTE(ARGUNUSED(enp))
444 	return (0);
445 }
446 
447 			void
448 ef10_ev_fini(
449 	__in		efx_nic_t *enp)
450 {
451 	_NOTE(ARGUNUSED(enp))
452 }
453 
454 	__checkReturn	efx_rc_t
455 ef10_ev_qcreate(
456 	__in		efx_nic_t *enp,
457 	__in		unsigned int index,
458 	__in		efsys_mem_t *esmp,
459 	__in		size_t ndescs,
460 	__in		uint32_t id,
461 	__in		uint32_t us,
462 	__in		uint32_t flags,
463 	__in		efx_evq_t *eep)
464 {
465 	efx_nic_cfg_t *encp = &(enp->en_nic_cfg);
466 	uint32_t irq;
467 	efx_rc_t rc;
468 
469 	_NOTE(ARGUNUSED(id))	/* buftbl id managed by MC */
470 	EFX_STATIC_ASSERT(ISP2(EFX_EVQ_MAXNEVS));
471 	EFX_STATIC_ASSERT(ISP2(EFX_EVQ_MINNEVS));
472 
473 	if (!ISP2(ndescs) ||
474 	    (ndescs < EFX_EVQ_MINNEVS) || (ndescs > EFX_EVQ_MAXNEVS)) {
475 		rc = EINVAL;
476 		goto fail1;
477 	}
478 
479 	if (index >= encp->enc_evq_limit) {
480 		rc = EINVAL;
481 		goto fail2;
482 	}
483 
484 	if (us > encp->enc_evq_timer_max_us) {
485 		rc = EINVAL;
486 		goto fail3;
487 	}
488 
489 	/* Set up the handler table */
490 	eep->ee_rx	= ef10_ev_rx;
491 	eep->ee_tx	= ef10_ev_tx;
492 	eep->ee_driver	= ef10_ev_driver;
493 	eep->ee_drv_gen	= ef10_ev_drv_gen;
494 	eep->ee_mcdi	= ef10_ev_mcdi;
495 
496 	/* Set up the event queue */
497 	/* INIT_EVQ expects function-relative vector number */
498 	if ((flags & EFX_EVQ_FLAGS_NOTIFY_MASK) ==
499 	    EFX_EVQ_FLAGS_NOTIFY_INTERRUPT) {
500 		irq = index;
501 	} else if (index == EFX_EF10_ALWAYS_INTERRUPTING_EVQ_INDEX) {
502 		irq = index;
503 		flags = (flags & ~EFX_EVQ_FLAGS_NOTIFY_MASK) |
504 		    EFX_EVQ_FLAGS_NOTIFY_INTERRUPT;
505 	} else {
506 		irq = EFX_EF10_ALWAYS_INTERRUPTING_EVQ_INDEX;
507 	}
508 
509 	/*
510 	 * Interrupts may be raised for events immediately after the queue is
511 	 * created. See bug58606.
512 	 */
513 
514 	if (encp->enc_init_evq_v2_supported) {
515 		/*
516 		 * On Medford the low latency license is required to enable RX
517 		 * and event cut through and to disable RX batching.  If event
518 		 * queue type in flags is auto, we let the firmware decide the
519 		 * settings to use. If the adapter has a low latency license,
520 		 * it will choose the best settings for low latency, otherwise
521 		 * it will choose the best settings for throughput.
522 		 */
523 		rc = efx_mcdi_init_evq_v2(enp, index, esmp, ndescs, irq, us,
524 		    flags);
525 		if (rc != 0)
526 			goto fail4;
527 	} else {
528 		/*
529 		 * On Huntington we need to specify the settings to use.
530 		 * If event queue type in flags is auto, we favour throughput
531 		 * if the adapter is running virtualization supporting firmware
532 		 * (i.e. the full featured firmware variant)
533 		 * and latency otherwise. The Ethernet Virtual Bridging
534 		 * capability is used to make this decision. (Note though that
535 		 * the low latency firmware variant is also best for
536 		 * throughput and corresponding type should be specified
537 		 * to choose it.)
538 		 */
539 		boolean_t low_latency = encp->enc_datapath_cap_evb ? 0 : 1;
540 		rc = efx_mcdi_init_evq(enp, index, esmp, ndescs, irq, us, flags,
541 		    low_latency);
542 		if (rc != 0)
543 			goto fail5;
544 	}
545 
546 	return (0);
547 
548 fail5:
549 	EFSYS_PROBE(fail5);
550 fail4:
551 	EFSYS_PROBE(fail4);
552 fail3:
553 	EFSYS_PROBE(fail3);
554 fail2:
555 	EFSYS_PROBE(fail2);
556 fail1:
557 	EFSYS_PROBE1(fail1, efx_rc_t, rc);
558 
559 	return (rc);
560 }
561 
562 			void
563 ef10_ev_qdestroy(
564 	__in		efx_evq_t *eep)
565 {
566 	efx_nic_t *enp = eep->ee_enp;
567 
568 	EFSYS_ASSERT(enp->en_family == EFX_FAMILY_HUNTINGTON ||
569 	    enp->en_family == EFX_FAMILY_MEDFORD ||
570 	    enp->en_family == EFX_FAMILY_MEDFORD2);
571 
572 	(void) efx_mcdi_fini_evq(enp, eep->ee_index);
573 }
574 
575 	__checkReturn	efx_rc_t
576 ef10_ev_qprime(
577 	__in		efx_evq_t *eep,
578 	__in		unsigned int count)
579 {
580 	efx_nic_t *enp = eep->ee_enp;
581 	uint32_t rptr;
582 	efx_dword_t dword;
583 
584 	rptr = count & eep->ee_mask;
585 
586 	if (enp->en_nic_cfg.enc_bug35388_workaround) {
587 		EFX_STATIC_ASSERT(EFX_EVQ_MINNEVS >
588 		    (1 << ERF_DD_EVQ_IND_RPTR_WIDTH));
589 		EFX_STATIC_ASSERT(EFX_EVQ_MAXNEVS <
590 		    (1 << 2 * ERF_DD_EVQ_IND_RPTR_WIDTH));
591 
592 		EFX_POPULATE_DWORD_2(dword,
593 		    ERF_DD_EVQ_IND_RPTR_FLAGS,
594 		    EFE_DD_EVQ_IND_RPTR_FLAGS_HIGH,
595 		    ERF_DD_EVQ_IND_RPTR,
596 		    (rptr >> ERF_DD_EVQ_IND_RPTR_WIDTH));
597 		EFX_BAR_VI_WRITED(enp, ER_DD_EVQ_INDIRECT, eep->ee_index,
598 		    &dword, B_FALSE);
599 
600 		EFX_POPULATE_DWORD_2(dword,
601 		    ERF_DD_EVQ_IND_RPTR_FLAGS,
602 		    EFE_DD_EVQ_IND_RPTR_FLAGS_LOW,
603 		    ERF_DD_EVQ_IND_RPTR,
604 		    rptr & ((1 << ERF_DD_EVQ_IND_RPTR_WIDTH) - 1));
605 		EFX_BAR_VI_WRITED(enp, ER_DD_EVQ_INDIRECT, eep->ee_index,
606 		    &dword, B_FALSE);
607 	} else {
608 		EFX_POPULATE_DWORD_1(dword, ERF_DZ_EVQ_RPTR, rptr);
609 		EFX_BAR_VI_WRITED(enp, ER_DZ_EVQ_RPTR_REG, eep->ee_index,
610 		    &dword, B_FALSE);
611 	}
612 
613 	return (0);
614 }
615 
616 static	__checkReturn	efx_rc_t
617 efx_mcdi_driver_event(
618 	__in		efx_nic_t *enp,
619 	__in		uint32_t evq,
620 	__in		efx_qword_t data)
621 {
622 	efx_mcdi_req_t req;
623 	EFX_MCDI_DECLARE_BUF(payload, MC_CMD_DRIVER_EVENT_IN_LEN,
624 		MC_CMD_DRIVER_EVENT_OUT_LEN);
625 	efx_rc_t rc;
626 
627 	req.emr_cmd = MC_CMD_DRIVER_EVENT;
628 	req.emr_in_buf = payload;
629 	req.emr_in_length = MC_CMD_DRIVER_EVENT_IN_LEN;
630 	req.emr_out_buf = payload;
631 	req.emr_out_length = MC_CMD_DRIVER_EVENT_OUT_LEN;
632 
633 	MCDI_IN_SET_DWORD(req, DRIVER_EVENT_IN_EVQ, evq);
634 
635 	MCDI_IN_SET_DWORD(req, DRIVER_EVENT_IN_DATA_LO,
636 	    EFX_QWORD_FIELD(data, EFX_DWORD_0));
637 	MCDI_IN_SET_DWORD(req, DRIVER_EVENT_IN_DATA_HI,
638 	    EFX_QWORD_FIELD(data, EFX_DWORD_1));
639 
640 	efx_mcdi_execute(enp, &req);
641 
642 	if (req.emr_rc != 0) {
643 		rc = req.emr_rc;
644 		goto fail1;
645 	}
646 
647 	return (0);
648 
649 fail1:
650 	EFSYS_PROBE1(fail1, efx_rc_t, rc);
651 
652 	return (rc);
653 }
654 
655 			void
656 ef10_ev_qpost(
657 	__in	efx_evq_t *eep,
658 	__in	uint16_t data)
659 {
660 	efx_nic_t *enp = eep->ee_enp;
661 	efx_qword_t event;
662 
663 	EFX_POPULATE_QWORD_3(event,
664 	    ESF_DZ_DRV_CODE, ESE_DZ_EV_CODE_DRV_GEN_EV,
665 	    ESF_DZ_DRV_SUB_CODE, 0,
666 	    ESF_DZ_DRV_SUB_DATA_DW0, (uint32_t)data);
667 
668 	(void) efx_mcdi_driver_event(enp, eep->ee_index, event);
669 }
670 
671 	__checkReturn	efx_rc_t
672 ef10_ev_qmoderate(
673 	__in		efx_evq_t *eep,
674 	__in		unsigned int us)
675 {
676 	efx_nic_t *enp = eep->ee_enp;
677 	efx_nic_cfg_t *encp = &(enp->en_nic_cfg);
678 	efx_dword_t dword;
679 	uint32_t mode;
680 	efx_rc_t rc;
681 
682 	/* Check that hardware and MCDI use the same timer MODE values */
683 	EFX_STATIC_ASSERT(FFE_CZ_TIMER_MODE_DIS ==
684 	    MC_CMD_SET_EVQ_TMR_IN_TIMER_MODE_DIS);
685 	EFX_STATIC_ASSERT(FFE_CZ_TIMER_MODE_IMMED_START ==
686 	    MC_CMD_SET_EVQ_TMR_IN_TIMER_MODE_IMMED_START);
687 	EFX_STATIC_ASSERT(FFE_CZ_TIMER_MODE_TRIG_START ==
688 	    MC_CMD_SET_EVQ_TMR_IN_TIMER_MODE_TRIG_START);
689 	EFX_STATIC_ASSERT(FFE_CZ_TIMER_MODE_INT_HLDOFF ==
690 	    MC_CMD_SET_EVQ_TMR_IN_TIMER_MODE_INT_HLDOFF);
691 
692 	if (us > encp->enc_evq_timer_max_us) {
693 		rc = EINVAL;
694 		goto fail1;
695 	}
696 
697 	/* If the value is zero then disable the timer */
698 	if (us == 0) {
699 		mode = FFE_CZ_TIMER_MODE_DIS;
700 	} else {
701 		mode = FFE_CZ_TIMER_MODE_INT_HLDOFF;
702 	}
703 
704 	if (encp->enc_bug61265_workaround) {
705 		uint32_t ns = us * 1000;
706 
707 		rc = efx_mcdi_set_evq_tmr(enp, eep->ee_index, mode, ns);
708 		if (rc != 0)
709 			goto fail2;
710 	} else {
711 		unsigned int ticks;
712 
713 		if ((rc = efx_ev_usecs_to_ticks(enp, us, &ticks)) != 0)
714 			goto fail3;
715 
716 		if (encp->enc_bug35388_workaround) {
717 			EFX_POPULATE_DWORD_3(dword,
718 			    ERF_DD_EVQ_IND_TIMER_FLAGS,
719 			    EFE_DD_EVQ_IND_TIMER_FLAGS,
720 			    ERF_DD_EVQ_IND_TIMER_MODE, mode,
721 			    ERF_DD_EVQ_IND_TIMER_VAL, ticks);
722 			EFX_BAR_VI_WRITED(enp, ER_DD_EVQ_INDIRECT,
723 			    eep->ee_index, &dword, 0);
724 		} else {
725 			/*
726 			 * NOTE: The TMR_REL field introduced in Medford2 is
727 			 * ignored on earlier EF10 controllers. See bug66418
728 			 * comment 9 for details.
729 			 */
730 			EFX_POPULATE_DWORD_3(dword,
731 			    ERF_DZ_TC_TIMER_MODE, mode,
732 			    ERF_DZ_TC_TIMER_VAL, ticks,
733 			    ERF_FZ_TC_TMR_REL_VAL, ticks);
734 			EFX_BAR_VI_WRITED(enp, ER_DZ_EVQ_TMR_REG,
735 			    eep->ee_index, &dword, 0);
736 		}
737 	}
738 
739 	return (0);
740 
741 fail3:
742 	EFSYS_PROBE(fail3);
743 fail2:
744 	EFSYS_PROBE(fail2);
745 fail1:
746 	EFSYS_PROBE1(fail1, efx_rc_t, rc);
747 
748 	return (rc);
749 }
750 
751 #if EFSYS_OPT_QSTATS
752 			void
753 ef10_ev_qstats_update(
754 	__in				efx_evq_t *eep,
755 	__inout_ecount(EV_NQSTATS)	efsys_stat_t *stat)
756 {
757 	unsigned int id;
758 
759 	for (id = 0; id < EV_NQSTATS; id++) {
760 		efsys_stat_t *essp = &stat[id];
761 
762 		EFSYS_STAT_INCR(essp, eep->ee_stat[id]);
763 		eep->ee_stat[id] = 0;
764 	}
765 }
766 #endif /* EFSYS_OPT_QSTATS */
767 
768 #if EFSYS_OPT_RX_PACKED_STREAM || EFSYS_OPT_RX_ES_SUPER_BUFFER
769 
770 static	__checkReturn	boolean_t
771 ef10_ev_rx_packed_stream(
772 	__in		efx_evq_t *eep,
773 	__in		efx_qword_t *eqp,
774 	__in		const efx_ev_callbacks_t *eecp,
775 	__in_opt	void *arg)
776 {
777 	uint32_t label;
778 	uint32_t pkt_count_lbits;
779 	uint16_t flags;
780 	boolean_t should_abort;
781 	efx_evq_rxq_state_t *eersp;
782 	unsigned int pkt_count;
783 	unsigned int current_id;
784 	boolean_t new_buffer;
785 
786 	pkt_count_lbits = EFX_QWORD_FIELD(*eqp, ESF_DZ_RX_DSC_PTR_LBITS);
787 	label = EFX_QWORD_FIELD(*eqp, ESF_DZ_RX_QLABEL);
788 	new_buffer = EFX_QWORD_FIELD(*eqp, ESF_DZ_RX_EV_ROTATE);
789 
790 	flags = 0;
791 
792 	eersp = &eep->ee_rxq_state[label];
793 
794 	/*
795 	 * RX_DSC_PTR_LBITS has least significant bits of the global
796 	 * (not per-buffer) packet counter. It is guaranteed that
797 	 * maximum number of completed packets fits in lbits-mask.
798 	 * So, modulo lbits-mask arithmetic should be used to calculate
799 	 * packet counter increment.
800 	 */
801 	pkt_count = (pkt_count_lbits - eersp->eers_rx_stream_npackets) &
802 	    EFX_MASK32(ESF_DZ_RX_DSC_PTR_LBITS);
803 	eersp->eers_rx_stream_npackets += pkt_count;
804 
805 	if (new_buffer) {
806 		flags |= EFX_PKT_PACKED_STREAM_NEW_BUFFER;
807 #if EFSYS_OPT_RX_PACKED_STREAM
808 		/*
809 		 * If both packed stream and equal stride super-buffer
810 		 * modes are compiled in, in theory credits should be
811 		 * be maintained for packed stream only, but right now
812 		 * these modes are not distinguished in the event queue
813 		 * Rx queue state and it is OK to increment the counter
814 		 * regardless (it might be event cheaper than branching
815 		 * since neighbour structure member are updated as well).
816 		 */
817 		eersp->eers_rx_packed_stream_credits++;
818 #endif
819 		eersp->eers_rx_read_ptr++;
820 	}
821 	current_id = eersp->eers_rx_read_ptr & eersp->eers_rx_mask;
822 
823 	/* Check for errors that invalidate checksum and L3/L4 fields */
824 	if (EFX_QWORD_FIELD(*eqp, ESF_DZ_RX_TRUNC_ERR) != 0) {
825 		/* RX frame truncated */
826 		EFX_EV_QSTAT_INCR(eep, EV_RX_FRM_TRUNC);
827 		flags |= EFX_DISCARD;
828 		goto deliver;
829 	}
830 	if (EFX_QWORD_FIELD(*eqp, ESF_DZ_RX_ECRC_ERR) != 0) {
831 		/* Bad Ethernet frame CRC */
832 		EFX_EV_QSTAT_INCR(eep, EV_RX_ETH_CRC_ERR);
833 		flags |= EFX_DISCARD;
834 		goto deliver;
835 	}
836 
837 	if (EFX_QWORD_FIELD(*eqp, ESF_DZ_RX_PARSE_INCOMPLETE)) {
838 		flags |= EFX_PKT_PACKED_STREAM_PARSE_INCOMPLETE;
839 		goto deliver;
840 	}
841 
842 	if (EFX_QWORD_FIELD(*eqp, ESF_DZ_RX_IPCKSUM_ERR))
843 		EFX_EV_QSTAT_INCR(eep, EV_RX_IPV4_HDR_CHKSUM_ERR);
844 
845 	if (EFX_QWORD_FIELD(*eqp, ESF_DZ_RX_TCPUDP_CKSUM_ERR))
846 		EFX_EV_QSTAT_INCR(eep, EV_RX_TCP_UDP_CHKSUM_ERR);
847 
848 deliver:
849 	/* If we're not discarding the packet then it is ok */
850 	if (~flags & EFX_DISCARD)
851 		EFX_EV_QSTAT_INCR(eep, EV_RX_OK);
852 
853 	EFSYS_ASSERT(eecp->eec_rx_ps != NULL);
854 	should_abort = eecp->eec_rx_ps(arg, label, current_id, pkt_count,
855 	    flags);
856 
857 	return (should_abort);
858 }
859 
860 #endif /* EFSYS_OPT_RX_PACKED_STREAM || EFSYS_OPT_RX_ES_SUPER_BUFFER */
861 
862 static	__checkReturn	boolean_t
863 ef10_ev_rx(
864 	__in		efx_evq_t *eep,
865 	__in		efx_qword_t *eqp,
866 	__in		const efx_ev_callbacks_t *eecp,
867 	__in_opt	void *arg)
868 {
869 	efx_nic_t *enp = eep->ee_enp;
870 	uint32_t size;
871 	uint32_t label;
872 	uint32_t mac_class;
873 	uint32_t eth_tag_class;
874 	uint32_t l3_class;
875 	uint32_t l4_class;
876 	uint32_t next_read_lbits;
877 	uint16_t flags;
878 	boolean_t cont;
879 	boolean_t should_abort;
880 	efx_evq_rxq_state_t *eersp;
881 	unsigned int desc_count;
882 	unsigned int last_used_id;
883 
884 	EFX_EV_QSTAT_INCR(eep, EV_RX);
885 
886 	/* Discard events after RXQ/TXQ errors, or hardware not available */
887 	if (enp->en_reset_flags &
888 	    (EFX_RESET_RXQ_ERR | EFX_RESET_TXQ_ERR | EFX_RESET_HW_UNAVAIL))
889 		return (B_FALSE);
890 
891 	/* Basic packet information */
892 	label = EFX_QWORD_FIELD(*eqp, ESF_DZ_RX_QLABEL);
893 	eersp = &eep->ee_rxq_state[label];
894 
895 #if EFSYS_OPT_RX_PACKED_STREAM || EFSYS_OPT_RX_ES_SUPER_BUFFER
896 	/*
897 	 * Packed stream events are very different,
898 	 * so handle them separately
899 	 */
900 	if (eersp->eers_rx_packed_stream)
901 	    return (ef10_ev_rx_packed_stream(eep, eqp, eecp, arg));
902 #endif
903 
904 	size = EFX_QWORD_FIELD(*eqp, ESF_DZ_RX_BYTES);
905 	cont = EFX_QWORD_FIELD(*eqp, ESF_DZ_RX_CONT);
906 	next_read_lbits = EFX_QWORD_FIELD(*eqp, ESF_DZ_RX_DSC_PTR_LBITS);
907 	eth_tag_class = EFX_QWORD_FIELD(*eqp, ESF_DZ_RX_ETH_TAG_CLASS);
908 	mac_class = EFX_QWORD_FIELD(*eqp, ESF_DZ_RX_MAC_CLASS);
909 	l3_class = EFX_QWORD_FIELD(*eqp, ESF_DZ_RX_L3_CLASS);
910 
911 	/*
912 	 * RX_L4_CLASS is 3 bits wide on Huntington and Medford, but is only
913 	 * 2 bits wide on Medford2. Check it is safe to use the Medford2 field
914 	 * and values for all EF10 controllers.
915 	 */
916 	EFX_STATIC_ASSERT(ESF_FZ_RX_L4_CLASS_LBN == ESF_DE_RX_L4_CLASS_LBN);
917 	EFX_STATIC_ASSERT(ESE_FZ_L4_CLASS_TCP == ESE_DE_L4_CLASS_TCP);
918 	EFX_STATIC_ASSERT(ESE_FZ_L4_CLASS_UDP == ESE_DE_L4_CLASS_UDP);
919 	EFX_STATIC_ASSERT(ESE_FZ_L4_CLASS_UNKNOWN == ESE_DE_L4_CLASS_UNKNOWN);
920 
921 	l4_class = EFX_QWORD_FIELD(*eqp, ESF_FZ_RX_L4_CLASS);
922 
923 	if (EFX_QWORD_FIELD(*eqp, ESF_DZ_RX_DROP_EVENT) != 0) {
924 		/* Drop this event */
925 		return (B_FALSE);
926 	}
927 	flags = 0;
928 
929 	if (cont != 0) {
930 		/*
931 		 * This may be part of a scattered frame, or it may be a
932 		 * truncated frame if scatter is disabled on this RXQ.
933 		 * Overlength frames can be received if e.g. a VF is configured
934 		 * for 1500 MTU but connected to a port set to 9000 MTU
935 		 * (see bug56567).
936 		 * FIXME: There is not yet any driver that supports scatter on
937 		 * Huntington.  Scatter support is required for OSX.
938 		 */
939 		flags |= EFX_PKT_CONT;
940 	}
941 
942 	if (mac_class == ESE_DZ_MAC_CLASS_UCAST)
943 		flags |= EFX_PKT_UNICAST;
944 
945 	/* Increment the count of descriptors read */
946 	desc_count = (next_read_lbits - eersp->eers_rx_read_ptr) &
947 	    EFX_MASK32(ESF_DZ_RX_DSC_PTR_LBITS);
948 	eersp->eers_rx_read_ptr += desc_count;
949 
950 	/*
951 	 * FIXME: add error checking to make sure this a batched event.
952 	 * This could also be an aborted scatter, see Bug36629.
953 	 */
954 	if (desc_count > 1) {
955 		EFX_EV_QSTAT_INCR(eep, EV_RX_BATCH);
956 		flags |= EFX_PKT_PREFIX_LEN;
957 	}
958 
959 	/* Calculate the index of the last descriptor consumed */
960 	last_used_id = (eersp->eers_rx_read_ptr - 1) & eersp->eers_rx_mask;
961 
962 	/* Check for errors that invalidate checksum and L3/L4 fields */
963 	if (EFX_QWORD_FIELD(*eqp, ESF_DZ_RX_TRUNC_ERR) != 0) {
964 		/* RX frame truncated */
965 		EFX_EV_QSTAT_INCR(eep, EV_RX_FRM_TRUNC);
966 		flags |= EFX_DISCARD;
967 		goto deliver;
968 	}
969 	if (EFX_QWORD_FIELD(*eqp, ESF_DZ_RX_ECRC_ERR) != 0) {
970 		/* Bad Ethernet frame CRC */
971 		EFX_EV_QSTAT_INCR(eep, EV_RX_ETH_CRC_ERR);
972 		flags |= EFX_DISCARD;
973 		goto deliver;
974 	}
975 	if (EFX_QWORD_FIELD(*eqp, ESF_DZ_RX_PARSE_INCOMPLETE)) {
976 		/*
977 		 * Hardware parse failed, due to malformed headers
978 		 * or headers that are too long for the parser.
979 		 * Headers and checksums must be validated by the host.
980 		 */
981 		/* TODO: EFX_EV_QSTAT_INCR(eep, EV_RX_PARSE_INCOMPLETE); */
982 		goto deliver;
983 	}
984 
985 	if ((eth_tag_class == ESE_DZ_ETH_TAG_CLASS_VLAN1) ||
986 	    (eth_tag_class == ESE_DZ_ETH_TAG_CLASS_VLAN2)) {
987 		flags |= EFX_PKT_VLAN_TAGGED;
988 	}
989 
990 	switch (l3_class) {
991 	case ESE_DZ_L3_CLASS_IP4:
992 	case ESE_DZ_L3_CLASS_IP4_FRAG:
993 		flags |= EFX_PKT_IPV4;
994 		if (EFX_QWORD_FIELD(*eqp, ESF_DZ_RX_IPCKSUM_ERR)) {
995 			EFX_EV_QSTAT_INCR(eep, EV_RX_IPV4_HDR_CHKSUM_ERR);
996 		} else {
997 			flags |= EFX_CKSUM_IPV4;
998 		}
999 
1000 		/*
1001 		 * RX_L4_CLASS is 3 bits wide on Huntington and Medford, but is
1002 		 * only 2 bits wide on Medford2. Check it is safe to use the
1003 		 * Medford2 field and values for all EF10 controllers.
1004 		 */
1005 		EFX_STATIC_ASSERT(ESF_FZ_RX_L4_CLASS_LBN ==
1006 		    ESF_DE_RX_L4_CLASS_LBN);
1007 		EFX_STATIC_ASSERT(ESE_FZ_L4_CLASS_TCP == ESE_DE_L4_CLASS_TCP);
1008 		EFX_STATIC_ASSERT(ESE_FZ_L4_CLASS_UDP == ESE_DE_L4_CLASS_UDP);
1009 		EFX_STATIC_ASSERT(ESE_FZ_L4_CLASS_UNKNOWN ==
1010 		    ESE_DE_L4_CLASS_UNKNOWN);
1011 
1012 		if (l4_class == ESE_FZ_L4_CLASS_TCP) {
1013 			EFX_EV_QSTAT_INCR(eep, EV_RX_TCP_IPV4);
1014 			flags |= EFX_PKT_TCP;
1015 		} else if (l4_class == ESE_FZ_L4_CLASS_UDP) {
1016 			EFX_EV_QSTAT_INCR(eep, EV_RX_UDP_IPV4);
1017 			flags |= EFX_PKT_UDP;
1018 		} else {
1019 			EFX_EV_QSTAT_INCR(eep, EV_RX_OTHER_IPV4);
1020 		}
1021 		break;
1022 
1023 	case ESE_DZ_L3_CLASS_IP6:
1024 	case ESE_DZ_L3_CLASS_IP6_FRAG:
1025 		flags |= EFX_PKT_IPV6;
1026 
1027 		/*
1028 		 * RX_L4_CLASS is 3 bits wide on Huntington and Medford, but is
1029 		 * only 2 bits wide on Medford2. Check it is safe to use the
1030 		 * Medford2 field and values for all EF10 controllers.
1031 		 */
1032 		EFX_STATIC_ASSERT(ESF_FZ_RX_L4_CLASS_LBN ==
1033 		    ESF_DE_RX_L4_CLASS_LBN);
1034 		EFX_STATIC_ASSERT(ESE_FZ_L4_CLASS_TCP == ESE_DE_L4_CLASS_TCP);
1035 		EFX_STATIC_ASSERT(ESE_FZ_L4_CLASS_UDP == ESE_DE_L4_CLASS_UDP);
1036 		EFX_STATIC_ASSERT(ESE_FZ_L4_CLASS_UNKNOWN ==
1037 		    ESE_DE_L4_CLASS_UNKNOWN);
1038 
1039 		if (l4_class == ESE_FZ_L4_CLASS_TCP) {
1040 			EFX_EV_QSTAT_INCR(eep, EV_RX_TCP_IPV6);
1041 			flags |= EFX_PKT_TCP;
1042 		} else if (l4_class == ESE_FZ_L4_CLASS_UDP) {
1043 			EFX_EV_QSTAT_INCR(eep, EV_RX_UDP_IPV6);
1044 			flags |= EFX_PKT_UDP;
1045 		} else {
1046 			EFX_EV_QSTAT_INCR(eep, EV_RX_OTHER_IPV6);
1047 		}
1048 		break;
1049 
1050 	default:
1051 		EFX_EV_QSTAT_INCR(eep, EV_RX_NON_IP);
1052 		break;
1053 	}
1054 
1055 	if (flags & (EFX_PKT_TCP | EFX_PKT_UDP)) {
1056 		if (EFX_QWORD_FIELD(*eqp, ESF_DZ_RX_TCPUDP_CKSUM_ERR)) {
1057 			EFX_EV_QSTAT_INCR(eep, EV_RX_TCP_UDP_CHKSUM_ERR);
1058 		} else {
1059 			flags |= EFX_CKSUM_TCPUDP;
1060 		}
1061 	}
1062 
1063 deliver:
1064 	/* If we're not discarding the packet then it is ok */
1065 	if (~flags & EFX_DISCARD)
1066 		EFX_EV_QSTAT_INCR(eep, EV_RX_OK);
1067 
1068 	EFSYS_ASSERT(eecp->eec_rx != NULL);
1069 	should_abort = eecp->eec_rx(arg, label, last_used_id, size, flags);
1070 
1071 	return (should_abort);
1072 }
1073 
1074 static	__checkReturn	boolean_t
1075 ef10_ev_tx(
1076 	__in		efx_evq_t *eep,
1077 	__in		efx_qword_t *eqp,
1078 	__in		const efx_ev_callbacks_t *eecp,
1079 	__in_opt	void *arg)
1080 {
1081 	efx_nic_t *enp = eep->ee_enp;
1082 	uint32_t id;
1083 	uint32_t label;
1084 	boolean_t should_abort;
1085 
1086 	EFX_EV_QSTAT_INCR(eep, EV_TX);
1087 
1088 	/* Discard events after RXQ/TXQ errors, or hardware not available */
1089 	if (enp->en_reset_flags &
1090 	    (EFX_RESET_RXQ_ERR | EFX_RESET_TXQ_ERR | EFX_RESET_HW_UNAVAIL))
1091 		return (B_FALSE);
1092 
1093 	if (EFX_QWORD_FIELD(*eqp, ESF_DZ_TX_DROP_EVENT) != 0) {
1094 		/* Drop this event */
1095 		return (B_FALSE);
1096 	}
1097 
1098 	/* Per-packet TX completion (was per-descriptor for Falcon/Siena) */
1099 	id = EFX_QWORD_FIELD(*eqp, ESF_DZ_TX_DESCR_INDX);
1100 	label = EFX_QWORD_FIELD(*eqp, ESF_DZ_TX_QLABEL);
1101 
1102 	EFSYS_PROBE2(tx_complete, uint32_t, label, uint32_t, id);
1103 
1104 	EFSYS_ASSERT(eecp->eec_tx != NULL);
1105 	should_abort = eecp->eec_tx(arg, label, id);
1106 
1107 	return (should_abort);
1108 }
1109 
1110 static	__checkReturn	boolean_t
1111 ef10_ev_driver(
1112 	__in		efx_evq_t *eep,
1113 	__in		efx_qword_t *eqp,
1114 	__in		const efx_ev_callbacks_t *eecp,
1115 	__in_opt	void *arg)
1116 {
1117 	unsigned int code;
1118 	boolean_t should_abort;
1119 
1120 	EFX_EV_QSTAT_INCR(eep, EV_DRIVER);
1121 	should_abort = B_FALSE;
1122 
1123 	code = EFX_QWORD_FIELD(*eqp, ESF_DZ_DRV_SUB_CODE);
1124 	switch (code) {
1125 	case ESE_DZ_DRV_TIMER_EV: {
1126 		uint32_t id;
1127 
1128 		id = EFX_QWORD_FIELD(*eqp, ESF_DZ_DRV_TMR_ID);
1129 
1130 		EFSYS_ASSERT(eecp->eec_timer != NULL);
1131 		should_abort = eecp->eec_timer(arg, id);
1132 		break;
1133 	}
1134 
1135 	case ESE_DZ_DRV_WAKE_UP_EV: {
1136 		uint32_t id;
1137 
1138 		id = EFX_QWORD_FIELD(*eqp, ESF_DZ_DRV_EVQ_ID);
1139 
1140 		EFSYS_ASSERT(eecp->eec_wake_up != NULL);
1141 		should_abort = eecp->eec_wake_up(arg, id);
1142 		break;
1143 	}
1144 
1145 	case ESE_DZ_DRV_START_UP_EV:
1146 		EFSYS_ASSERT(eecp->eec_initialized != NULL);
1147 		should_abort = eecp->eec_initialized(arg);
1148 		break;
1149 
1150 	default:
1151 		EFSYS_PROBE3(bad_event, unsigned int, eep->ee_index,
1152 		    uint32_t, EFX_QWORD_FIELD(*eqp, EFX_DWORD_1),
1153 		    uint32_t, EFX_QWORD_FIELD(*eqp, EFX_DWORD_0));
1154 		break;
1155 	}
1156 
1157 	return (should_abort);
1158 }
1159 
1160 static	__checkReturn	boolean_t
1161 ef10_ev_drv_gen(
1162 	__in		efx_evq_t *eep,
1163 	__in		efx_qword_t *eqp,
1164 	__in		const efx_ev_callbacks_t *eecp,
1165 	__in_opt	void *arg)
1166 {
1167 	uint32_t data;
1168 	boolean_t should_abort;
1169 
1170 	EFX_EV_QSTAT_INCR(eep, EV_DRV_GEN);
1171 	should_abort = B_FALSE;
1172 
1173 	data = EFX_QWORD_FIELD(*eqp, ESF_DZ_DRV_SUB_DATA_DW0);
1174 	if (data >= ((uint32_t)1 << 16)) {
1175 		EFSYS_PROBE3(bad_event, unsigned int, eep->ee_index,
1176 		    uint32_t, EFX_QWORD_FIELD(*eqp, EFX_DWORD_1),
1177 		    uint32_t, EFX_QWORD_FIELD(*eqp, EFX_DWORD_0));
1178 
1179 		return (B_TRUE);
1180 	}
1181 
1182 	EFSYS_ASSERT(eecp->eec_software != NULL);
1183 	should_abort = eecp->eec_software(arg, (uint16_t)data);
1184 
1185 	return (should_abort);
1186 }
1187 
1188 static	__checkReturn	boolean_t
1189 ef10_ev_mcdi(
1190 	__in		efx_evq_t *eep,
1191 	__in		efx_qword_t *eqp,
1192 	__in		const efx_ev_callbacks_t *eecp,
1193 	__in_opt	void *arg)
1194 {
1195 	efx_nic_t *enp = eep->ee_enp;
1196 	unsigned int code;
1197 	boolean_t should_abort = B_FALSE;
1198 
1199 	EFX_EV_QSTAT_INCR(eep, EV_MCDI_RESPONSE);
1200 
1201 	code = EFX_QWORD_FIELD(*eqp, MCDI_EVENT_CODE);
1202 	switch (code) {
1203 	case MCDI_EVENT_CODE_BADSSERT:
1204 		efx_mcdi_ev_death(enp, EINTR);
1205 		break;
1206 
1207 	case MCDI_EVENT_CODE_CMDDONE:
1208 		efx_mcdi_ev_cpl(enp,
1209 		    MCDI_EV_FIELD(eqp, CMDDONE_SEQ),
1210 		    MCDI_EV_FIELD(eqp, CMDDONE_DATALEN),
1211 		    MCDI_EV_FIELD(eqp, CMDDONE_ERRNO));
1212 		break;
1213 
1214 #if EFSYS_OPT_MCDI_PROXY_AUTH
1215 	case MCDI_EVENT_CODE_PROXY_RESPONSE:
1216 		/*
1217 		 * This event notifies a function that an authorization request
1218 		 * has been processed. If the request was authorized then the
1219 		 * function can now re-send the original MCDI request.
1220 		 * See SF-113652-SW "SR-IOV Proxied Network Access Control".
1221 		 */
1222 		efx_mcdi_ev_proxy_response(enp,
1223 		    MCDI_EV_FIELD(eqp, PROXY_RESPONSE_HANDLE),
1224 		    MCDI_EV_FIELD(eqp, PROXY_RESPONSE_RC));
1225 		break;
1226 #endif /* EFSYS_OPT_MCDI_PROXY_AUTH */
1227 
1228 	case MCDI_EVENT_CODE_LINKCHANGE: {
1229 		efx_link_mode_t link_mode;
1230 
1231 		ef10_phy_link_ev(enp, eqp, &link_mode);
1232 		should_abort = eecp->eec_link_change(arg, link_mode);
1233 		break;
1234 	}
1235 
1236 	case MCDI_EVENT_CODE_SENSOREVT: {
1237 #if EFSYS_OPT_MON_STATS
1238 		efx_mon_stat_t id;
1239 		efx_mon_stat_value_t value;
1240 		efx_rc_t rc;
1241 
1242 		/* Decode monitor stat for MCDI sensor (if supported) */
1243 		if ((rc = mcdi_mon_ev(enp, eqp, &id, &value)) == 0) {
1244 			/* Report monitor stat change */
1245 			should_abort = eecp->eec_monitor(arg, id, value);
1246 		} else if (rc == ENOTSUP) {
1247 			should_abort = eecp->eec_exception(arg,
1248 				EFX_EXCEPTION_UNKNOWN_SENSOREVT,
1249 				MCDI_EV_FIELD(eqp, DATA));
1250 		} else {
1251 			EFSYS_ASSERT(rc == ENODEV);	/* Wrong port */
1252 		}
1253 #endif
1254 		break;
1255 	}
1256 
1257 	case MCDI_EVENT_CODE_SCHEDERR:
1258 		/* Informational only */
1259 		break;
1260 
1261 	case MCDI_EVENT_CODE_REBOOT:
1262 		/* Falcon/Siena only (should not been seen with Huntington). */
1263 		efx_mcdi_ev_death(enp, EIO);
1264 		break;
1265 
1266 	case MCDI_EVENT_CODE_MC_REBOOT:
1267 		/* MC_REBOOT event is used for Huntington (EF10) and later. */
1268 		efx_mcdi_ev_death(enp, EIO);
1269 		break;
1270 
1271 	case MCDI_EVENT_CODE_MAC_STATS_DMA:
1272 #if EFSYS_OPT_MAC_STATS
1273 		if (eecp->eec_mac_stats != NULL) {
1274 			eecp->eec_mac_stats(arg,
1275 			    MCDI_EV_FIELD(eqp, MAC_STATS_DMA_GENERATION));
1276 		}
1277 #endif
1278 		break;
1279 
1280 	case MCDI_EVENT_CODE_FWALERT: {
1281 		uint32_t reason = MCDI_EV_FIELD(eqp, FWALERT_REASON);
1282 
1283 		if (reason == MCDI_EVENT_FWALERT_REASON_SRAM_ACCESS)
1284 			should_abort = eecp->eec_exception(arg,
1285 				EFX_EXCEPTION_FWALERT_SRAM,
1286 				MCDI_EV_FIELD(eqp, FWALERT_DATA));
1287 		else
1288 			should_abort = eecp->eec_exception(arg,
1289 				EFX_EXCEPTION_UNKNOWN_FWALERT,
1290 				MCDI_EV_FIELD(eqp, DATA));
1291 		break;
1292 	}
1293 
1294 	case MCDI_EVENT_CODE_TX_ERR: {
1295 		/*
1296 		 * After a TXQ error is detected, firmware sends a TX_ERR event.
1297 		 * This may be followed by TX completions (which we discard),
1298 		 * and then finally by a TX_FLUSH event. Firmware destroys the
1299 		 * TXQ automatically after sending the TX_FLUSH event.
1300 		 */
1301 		enp->en_reset_flags |= EFX_RESET_TXQ_ERR;
1302 
1303 		EFSYS_PROBE2(tx_descq_err,
1304 			    uint32_t, EFX_QWORD_FIELD(*eqp, EFX_DWORD_1),
1305 			    uint32_t, EFX_QWORD_FIELD(*eqp, EFX_DWORD_0));
1306 
1307 		/* Inform the driver that a reset is required. */
1308 		eecp->eec_exception(arg, EFX_EXCEPTION_TX_ERROR,
1309 		    MCDI_EV_FIELD(eqp, TX_ERR_DATA));
1310 		break;
1311 	}
1312 
1313 	case MCDI_EVENT_CODE_TX_FLUSH: {
1314 		uint32_t txq_index = MCDI_EV_FIELD(eqp, TX_FLUSH_TXQ);
1315 
1316 		/*
1317 		 * EF10 firmware sends two TX_FLUSH events: one to the txq's
1318 		 * event queue, and one to evq 0 (with TX_FLUSH_TO_DRIVER set).
1319 		 * We want to wait for all completions, so ignore the events
1320 		 * with TX_FLUSH_TO_DRIVER.
1321 		 */
1322 		if (MCDI_EV_FIELD(eqp, TX_FLUSH_TO_DRIVER) != 0) {
1323 			should_abort = B_FALSE;
1324 			break;
1325 		}
1326 
1327 		EFX_EV_QSTAT_INCR(eep, EV_DRIVER_TX_DESCQ_FLS_DONE);
1328 
1329 		EFSYS_PROBE1(tx_descq_fls_done, uint32_t, txq_index);
1330 
1331 		EFSYS_ASSERT(eecp->eec_txq_flush_done != NULL);
1332 		should_abort = eecp->eec_txq_flush_done(arg, txq_index);
1333 		break;
1334 	}
1335 
1336 	case MCDI_EVENT_CODE_RX_ERR: {
1337 		/*
1338 		 * After an RXQ error is detected, firmware sends an RX_ERR
1339 		 * event. This may be followed by RX events (which we discard),
1340 		 * and then finally by an RX_FLUSH event. Firmware destroys the
1341 		 * RXQ automatically after sending the RX_FLUSH event.
1342 		 */
1343 		enp->en_reset_flags |= EFX_RESET_RXQ_ERR;
1344 
1345 		EFSYS_PROBE2(rx_descq_err,
1346 			    uint32_t, EFX_QWORD_FIELD(*eqp, EFX_DWORD_1),
1347 			    uint32_t, EFX_QWORD_FIELD(*eqp, EFX_DWORD_0));
1348 
1349 		/* Inform the driver that a reset is required. */
1350 		eecp->eec_exception(arg, EFX_EXCEPTION_RX_ERROR,
1351 		    MCDI_EV_FIELD(eqp, RX_ERR_DATA));
1352 		break;
1353 	}
1354 
1355 	case MCDI_EVENT_CODE_RX_FLUSH: {
1356 		uint32_t rxq_index = MCDI_EV_FIELD(eqp, RX_FLUSH_RXQ);
1357 
1358 		/*
1359 		 * EF10 firmware sends two RX_FLUSH events: one to the rxq's
1360 		 * event queue, and one to evq 0 (with RX_FLUSH_TO_DRIVER set).
1361 		 * We want to wait for all completions, so ignore the events
1362 		 * with RX_FLUSH_TO_DRIVER.
1363 		 */
1364 		if (MCDI_EV_FIELD(eqp, RX_FLUSH_TO_DRIVER) != 0) {
1365 			should_abort = B_FALSE;
1366 			break;
1367 		}
1368 
1369 		EFX_EV_QSTAT_INCR(eep, EV_DRIVER_RX_DESCQ_FLS_DONE);
1370 
1371 		EFSYS_PROBE1(rx_descq_fls_done, uint32_t, rxq_index);
1372 
1373 		EFSYS_ASSERT(eecp->eec_rxq_flush_done != NULL);
1374 		should_abort = eecp->eec_rxq_flush_done(arg, rxq_index);
1375 		break;
1376 	}
1377 
1378 	default:
1379 		EFSYS_PROBE3(bad_event, unsigned int, eep->ee_index,
1380 		    uint32_t, EFX_QWORD_FIELD(*eqp, EFX_DWORD_1),
1381 		    uint32_t, EFX_QWORD_FIELD(*eqp, EFX_DWORD_0));
1382 		break;
1383 	}
1384 
1385 	return (should_abort);
1386 }
1387 
1388 		void
1389 ef10_ev_rxlabel_init(
1390 	__in		efx_evq_t *eep,
1391 	__in		efx_rxq_t *erp,
1392 	__in		unsigned int label,
1393 	__in		efx_rxq_type_t type)
1394 {
1395 	efx_evq_rxq_state_t *eersp;
1396 #if EFSYS_OPT_RX_PACKED_STREAM || EFSYS_OPT_RX_ES_SUPER_BUFFER
1397 	boolean_t packed_stream = (type == EFX_RXQ_TYPE_PACKED_STREAM);
1398 	boolean_t es_super_buffer = (type == EFX_RXQ_TYPE_ES_SUPER_BUFFER);
1399 #endif
1400 
1401 	_NOTE(ARGUNUSED(type))
1402 	EFSYS_ASSERT3U(label, <, EFX_ARRAY_SIZE(eep->ee_rxq_state));
1403 	eersp = &eep->ee_rxq_state[label];
1404 
1405 	EFSYS_ASSERT3U(eersp->eers_rx_mask, ==, 0);
1406 
1407 #if EFSYS_OPT_RX_PACKED_STREAM
1408 	/*
1409 	 * For packed stream modes, the very first event will
1410 	 * have a new buffer flag set, so it will be incremented,
1411 	 * yielding the correct pointer. That results in a simpler
1412 	 * code than trying to detect start-of-the-world condition
1413 	 * in the event handler.
1414 	 */
1415 	eersp->eers_rx_read_ptr = packed_stream ? ~0 : 0;
1416 #else
1417 	eersp->eers_rx_read_ptr = 0;
1418 #endif
1419 	eersp->eers_rx_mask = erp->er_mask;
1420 #if EFSYS_OPT_RX_PACKED_STREAM || EFSYS_OPT_RX_ES_SUPER_BUFFER
1421 	eersp->eers_rx_stream_npackets = 0;
1422 	eersp->eers_rx_packed_stream = packed_stream || es_super_buffer;
1423 #endif
1424 #if EFSYS_OPT_RX_PACKED_STREAM
1425 	if (packed_stream) {
1426 		eersp->eers_rx_packed_stream_credits = (eep->ee_mask + 1) /
1427 		    EFX_DIV_ROUND_UP(EFX_RX_PACKED_STREAM_MEM_PER_CREDIT,
1428 		    EFX_RX_PACKED_STREAM_MIN_PACKET_SPACE);
1429 		EFSYS_ASSERT3U(eersp->eers_rx_packed_stream_credits, !=, 0);
1430 		/*
1431 		 * A single credit is allocated to the queue when it is started.
1432 		 * It is immediately spent by the first packet which has NEW
1433 		 * BUFFER flag set, though, but still we shall take into
1434 		 * account, as to not wrap around the maximum number of credits
1435 		 * accidentally
1436 		 */
1437 		eersp->eers_rx_packed_stream_credits--;
1438 		EFSYS_ASSERT3U(eersp->eers_rx_packed_stream_credits, <=,
1439 		    EFX_RX_PACKED_STREAM_MAX_CREDITS);
1440 	}
1441 #endif
1442 }
1443 
1444 		void
1445 ef10_ev_rxlabel_fini(
1446 	__in		efx_evq_t *eep,
1447 	__in		unsigned int label)
1448 {
1449 	efx_evq_rxq_state_t *eersp;
1450 
1451 	EFSYS_ASSERT3U(label, <, EFX_ARRAY_SIZE(eep->ee_rxq_state));
1452 	eersp = &eep->ee_rxq_state[label];
1453 
1454 	EFSYS_ASSERT3U(eersp->eers_rx_mask, !=, 0);
1455 
1456 	eersp->eers_rx_read_ptr = 0;
1457 	eersp->eers_rx_mask = 0;
1458 #if EFSYS_OPT_RX_PACKED_STREAM || EFSYS_OPT_RX_ES_SUPER_BUFFER
1459 	eersp->eers_rx_stream_npackets = 0;
1460 	eersp->eers_rx_packed_stream = B_FALSE;
1461 #endif
1462 #if EFSYS_OPT_RX_PACKED_STREAM
1463 	eersp->eers_rx_packed_stream_credits = 0;
1464 #endif
1465 }
1466 
1467 #endif	/* EFSYS_OPT_HUNTINGTON || EFSYS_OPT_MEDFORD || EFSYS_OPT_MEDFORD2 */
1468