xref: /freebsd/sys/dev/sfxge/common/ef10_ev.c (revision cab6a39d7b343596a5823e65c0f7b426551ec22d)
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 __FBSDID("$FreeBSD$");
33 
34 #include "efx.h"
35 #include "efx_impl.h"
36 #if EFSYS_OPT_MON_STATS
37 #include "mcdi_mon.h"
38 #endif
39 
40 #if EFSYS_OPT_HUNTINGTON || EFSYS_OPT_MEDFORD || EFSYS_OPT_MEDFORD2
41 
42 #if EFSYS_OPT_QSTATS
43 #define	EFX_EV_QSTAT_INCR(_eep, _stat)					\
44 	do {								\
45 		(_eep)->ee_stat[_stat]++;				\
46 	_NOTE(CONSTANTCONDITION)					\
47 	} while (B_FALSE)
48 #else
49 #define	EFX_EV_QSTAT_INCR(_eep, _stat)
50 #endif
51 
52 /*
53  * Non-interrupting event queue requires interrrupting event queue to
54  * refer to for wake-up events even if wake ups are never used.
55  * It could be even non-allocated event queue.
56  */
57 #define	EFX_EF10_ALWAYS_INTERRUPTING_EVQ_INDEX	(0)
58 
59 static	__checkReturn	boolean_t
60 ef10_ev_rx(
61 	__in		efx_evq_t *eep,
62 	__in		efx_qword_t *eqp,
63 	__in		const efx_ev_callbacks_t *eecp,
64 	__in_opt	void *arg);
65 
66 static	__checkReturn	boolean_t
67 ef10_ev_tx(
68 	__in		efx_evq_t *eep,
69 	__in		efx_qword_t *eqp,
70 	__in		const efx_ev_callbacks_t *eecp,
71 	__in_opt	void *arg);
72 
73 static	__checkReturn	boolean_t
74 ef10_ev_driver(
75 	__in		efx_evq_t *eep,
76 	__in		efx_qword_t *eqp,
77 	__in		const efx_ev_callbacks_t *eecp,
78 	__in_opt	void *arg);
79 
80 static	__checkReturn	boolean_t
81 ef10_ev_drv_gen(
82 	__in		efx_evq_t *eep,
83 	__in		efx_qword_t *eqp,
84 	__in		const efx_ev_callbacks_t *eecp,
85 	__in_opt	void *arg);
86 
87 static	__checkReturn	boolean_t
88 ef10_ev_mcdi(
89 	__in		efx_evq_t *eep,
90 	__in		efx_qword_t *eqp,
91 	__in		const efx_ev_callbacks_t *eecp,
92 	__in_opt	void *arg);
93 
94 static	__checkReturn	efx_rc_t
95 efx_mcdi_set_evq_tmr(
96 	__in		efx_nic_t *enp,
97 	__in		uint32_t instance,
98 	__in		uint32_t mode,
99 	__in		uint32_t timer_ns)
100 {
101 	efx_mcdi_req_t req;
102 	EFX_MCDI_DECLARE_BUF(payload, MC_CMD_SET_EVQ_TMR_IN_LEN,
103 		MC_CMD_SET_EVQ_TMR_OUT_LEN);
104 	efx_rc_t rc;
105 
106 	req.emr_cmd = MC_CMD_SET_EVQ_TMR;
107 	req.emr_in_buf = payload;
108 	req.emr_in_length = MC_CMD_SET_EVQ_TMR_IN_LEN;
109 	req.emr_out_buf = payload;
110 	req.emr_out_length = MC_CMD_SET_EVQ_TMR_OUT_LEN;
111 
112 	MCDI_IN_SET_DWORD(req, SET_EVQ_TMR_IN_INSTANCE, instance);
113 	MCDI_IN_SET_DWORD(req, SET_EVQ_TMR_IN_TMR_LOAD_REQ_NS, timer_ns);
114 	MCDI_IN_SET_DWORD(req, SET_EVQ_TMR_IN_TMR_RELOAD_REQ_NS, timer_ns);
115 	MCDI_IN_SET_DWORD(req, SET_EVQ_TMR_IN_TMR_MODE, mode);
116 
117 	efx_mcdi_execute(enp, &req);
118 
119 	if (req.emr_rc != 0) {
120 		rc = req.emr_rc;
121 		goto fail1;
122 	}
123 
124 	if (req.emr_out_length_used < MC_CMD_SET_EVQ_TMR_OUT_LEN) {
125 		rc = EMSGSIZE;
126 		goto fail2;
127 	}
128 
129 	return (0);
130 
131 fail2:
132 	EFSYS_PROBE(fail2);
133 fail1:
134 	EFSYS_PROBE1(fail1, efx_rc_t, rc);
135 
136 	return (rc);
137 }
138 
139 static	__checkReturn	efx_rc_t
140 efx_mcdi_init_evq(
141 	__in		efx_nic_t *enp,
142 	__in		unsigned int instance,
143 	__in		efsys_mem_t *esmp,
144 	__in		size_t nevs,
145 	__in		uint32_t irq,
146 	__in		uint32_t us,
147 	__in		uint32_t flags,
148 	__in		boolean_t low_latency)
149 {
150 	efx_mcdi_req_t req;
151 	EFX_MCDI_DECLARE_BUF(payload,
152 		MC_CMD_INIT_EVQ_IN_LEN(EFX_EVQ_NBUFS(EFX_EVQ_MAXNEVS)),
153 		MC_CMD_INIT_EVQ_OUT_LEN);
154 	efx_qword_t *dma_addr;
155 	uint64_t addr;
156 	int npages;
157 	int i;
158 	boolean_t interrupting;
159 	int ev_cut_through;
160 	efx_rc_t rc;
161 
162 	npages = EFX_EVQ_NBUFS(nevs);
163 	if (MC_CMD_INIT_EVQ_IN_LEN(npages) > MC_CMD_INIT_EVQ_IN_LENMAX) {
164 		rc = EINVAL;
165 		goto fail1;
166 	}
167 
168 	req.emr_cmd = MC_CMD_INIT_EVQ;
169 	req.emr_in_buf = payload;
170 	req.emr_in_length = MC_CMD_INIT_EVQ_IN_LEN(npages);
171 	req.emr_out_buf = payload;
172 	req.emr_out_length = MC_CMD_INIT_EVQ_OUT_LEN;
173 
174 	MCDI_IN_SET_DWORD(req, INIT_EVQ_IN_SIZE, nevs);
175 	MCDI_IN_SET_DWORD(req, INIT_EVQ_IN_INSTANCE, instance);
176 	MCDI_IN_SET_DWORD(req, INIT_EVQ_IN_IRQ_NUM, irq);
177 
178 	interrupting = ((flags & EFX_EVQ_FLAGS_NOTIFY_MASK) ==
179 	    EFX_EVQ_FLAGS_NOTIFY_INTERRUPT);
180 
181 	/*
182 	 * On Huntington RX and TX event batching can only be requested together
183 	 * (even if the datapath firmware doesn't actually support RX
184 	 * batching). If event cut through is enabled no RX batching will occur.
185 	 *
186 	 * So always enable RX and TX event batching, and enable event cut
187 	 * through if we want low latency operation.
188 	 */
189 	switch (flags & EFX_EVQ_FLAGS_TYPE_MASK) {
190 	case EFX_EVQ_FLAGS_TYPE_AUTO:
191 		ev_cut_through = low_latency ? 1 : 0;
192 		break;
193 	case EFX_EVQ_FLAGS_TYPE_THROUGHPUT:
194 		ev_cut_through = 0;
195 		break;
196 	case EFX_EVQ_FLAGS_TYPE_LOW_LATENCY:
197 		ev_cut_through = 1;
198 		break;
199 	default:
200 		rc = EINVAL;
201 		goto fail2;
202 	}
203 	MCDI_IN_POPULATE_DWORD_6(req, INIT_EVQ_IN_FLAGS,
204 	    INIT_EVQ_IN_FLAG_INTERRUPTING, interrupting,
205 	    INIT_EVQ_IN_FLAG_RPTR_DOS, 0,
206 	    INIT_EVQ_IN_FLAG_INT_ARMD, 0,
207 	    INIT_EVQ_IN_FLAG_CUT_THRU, ev_cut_through,
208 	    INIT_EVQ_IN_FLAG_RX_MERGE, 1,
209 	    INIT_EVQ_IN_FLAG_TX_MERGE, 1);
210 
211 	/* If the value is zero then disable the timer */
212 	if (us == 0) {
213 		MCDI_IN_SET_DWORD(req, INIT_EVQ_IN_TMR_MODE,
214 		    MC_CMD_INIT_EVQ_IN_TMR_MODE_DIS);
215 		MCDI_IN_SET_DWORD(req, INIT_EVQ_IN_TMR_LOAD, 0);
216 		MCDI_IN_SET_DWORD(req, INIT_EVQ_IN_TMR_RELOAD, 0);
217 	} else {
218 		unsigned int ticks;
219 
220 		if ((rc = efx_ev_usecs_to_ticks(enp, us, &ticks)) != 0)
221 			goto fail3;
222 
223 		MCDI_IN_SET_DWORD(req, INIT_EVQ_IN_TMR_MODE,
224 		    MC_CMD_INIT_EVQ_IN_TMR_INT_HLDOFF);
225 		MCDI_IN_SET_DWORD(req, INIT_EVQ_IN_TMR_LOAD, ticks);
226 		MCDI_IN_SET_DWORD(req, INIT_EVQ_IN_TMR_RELOAD, ticks);
227 	}
228 
229 	MCDI_IN_SET_DWORD(req, INIT_EVQ_IN_COUNT_MODE,
230 	    MC_CMD_INIT_EVQ_IN_COUNT_MODE_DIS);
231 	MCDI_IN_SET_DWORD(req, INIT_EVQ_IN_COUNT_THRSHLD, 0);
232 
233 	dma_addr = MCDI_IN2(req, efx_qword_t, INIT_EVQ_IN_DMA_ADDR);
234 	addr = EFSYS_MEM_ADDR(esmp);
235 
236 	for (i = 0; i < npages; i++) {
237 		EFX_POPULATE_QWORD_2(*dma_addr,
238 		    EFX_DWORD_1, (uint32_t)(addr >> 32),
239 		    EFX_DWORD_0, (uint32_t)(addr & 0xffffffff));
240 
241 		dma_addr++;
242 		addr += EFX_BUF_SIZE;
243 	}
244 
245 	efx_mcdi_execute(enp, &req);
246 
247 	if (req.emr_rc != 0) {
248 		rc = req.emr_rc;
249 		goto fail4;
250 	}
251 
252 	if (req.emr_out_length_used < MC_CMD_INIT_EVQ_OUT_LEN) {
253 		rc = EMSGSIZE;
254 		goto fail5;
255 	}
256 
257 	/* NOTE: ignore the returned IRQ param as firmware does not set it. */
258 
259 	return (0);
260 
261 fail5:
262 	EFSYS_PROBE(fail5);
263 fail4:
264 	EFSYS_PROBE(fail4);
265 fail3:
266 	EFSYS_PROBE(fail3);
267 fail2:
268 	EFSYS_PROBE(fail2);
269 fail1:
270 	EFSYS_PROBE1(fail1, efx_rc_t, rc);
271 
272 	return (rc);
273 }
274 
275 static	__checkReturn	efx_rc_t
276 efx_mcdi_init_evq_v2(
277 	__in		efx_nic_t *enp,
278 	__in		unsigned int instance,
279 	__in		efsys_mem_t *esmp,
280 	__in		size_t nevs,
281 	__in		uint32_t irq,
282 	__in		uint32_t us,
283 	__in		uint32_t flags)
284 {
285 	efx_mcdi_req_t req;
286 	EFX_MCDI_DECLARE_BUF(payload,
287 		MC_CMD_INIT_EVQ_V2_IN_LEN(EFX_EVQ_NBUFS(EFX_EVQ_MAXNEVS)),
288 		MC_CMD_INIT_EVQ_V2_OUT_LEN);
289 	boolean_t interrupting;
290 	unsigned int evq_type;
291 	efx_qword_t *dma_addr;
292 	uint64_t addr;
293 	int npages;
294 	int i;
295 	efx_rc_t rc;
296 
297 	npages = EFX_EVQ_NBUFS(nevs);
298 	if (MC_CMD_INIT_EVQ_V2_IN_LEN(npages) > MC_CMD_INIT_EVQ_V2_IN_LENMAX) {
299 		rc = EINVAL;
300 		goto fail1;
301 	}
302 
303 	req.emr_cmd = MC_CMD_INIT_EVQ;
304 	req.emr_in_buf = payload;
305 	req.emr_in_length = MC_CMD_INIT_EVQ_V2_IN_LEN(npages);
306 	req.emr_out_buf = payload;
307 	req.emr_out_length = MC_CMD_INIT_EVQ_V2_OUT_LEN;
308 
309 	MCDI_IN_SET_DWORD(req, INIT_EVQ_V2_IN_SIZE, nevs);
310 	MCDI_IN_SET_DWORD(req, INIT_EVQ_V2_IN_INSTANCE, instance);
311 	MCDI_IN_SET_DWORD(req, INIT_EVQ_V2_IN_IRQ_NUM, irq);
312 
313 	interrupting = ((flags & EFX_EVQ_FLAGS_NOTIFY_MASK) ==
314 	    EFX_EVQ_FLAGS_NOTIFY_INTERRUPT);
315 
316 	switch (flags & EFX_EVQ_FLAGS_TYPE_MASK) {
317 	case EFX_EVQ_FLAGS_TYPE_AUTO:
318 		evq_type = MC_CMD_INIT_EVQ_V2_IN_FLAG_TYPE_AUTO;
319 		break;
320 	case EFX_EVQ_FLAGS_TYPE_THROUGHPUT:
321 		evq_type = MC_CMD_INIT_EVQ_V2_IN_FLAG_TYPE_THROUGHPUT;
322 		break;
323 	case EFX_EVQ_FLAGS_TYPE_LOW_LATENCY:
324 		evq_type = MC_CMD_INIT_EVQ_V2_IN_FLAG_TYPE_LOW_LATENCY;
325 		break;
326 	default:
327 		rc = EINVAL;
328 		goto fail2;
329 	}
330 	MCDI_IN_POPULATE_DWORD_4(req, INIT_EVQ_V2_IN_FLAGS,
331 	    INIT_EVQ_V2_IN_FLAG_INTERRUPTING, interrupting,
332 	    INIT_EVQ_V2_IN_FLAG_RPTR_DOS, 0,
333 	    INIT_EVQ_V2_IN_FLAG_INT_ARMD, 0,
334 	    INIT_EVQ_V2_IN_FLAG_TYPE, evq_type);
335 
336 	/* If the value is zero then disable the timer */
337 	if (us == 0) {
338 		MCDI_IN_SET_DWORD(req, INIT_EVQ_V2_IN_TMR_MODE,
339 		    MC_CMD_INIT_EVQ_V2_IN_TMR_MODE_DIS);
340 		MCDI_IN_SET_DWORD(req, INIT_EVQ_V2_IN_TMR_LOAD, 0);
341 		MCDI_IN_SET_DWORD(req, INIT_EVQ_V2_IN_TMR_RELOAD, 0);
342 	} else {
343 		unsigned int ticks;
344 
345 		if ((rc = efx_ev_usecs_to_ticks(enp, us, &ticks)) != 0)
346 			goto fail3;
347 
348 		MCDI_IN_SET_DWORD(req, INIT_EVQ_V2_IN_TMR_MODE,
349 		    MC_CMD_INIT_EVQ_V2_IN_TMR_INT_HLDOFF);
350 		MCDI_IN_SET_DWORD(req, INIT_EVQ_V2_IN_TMR_LOAD, ticks);
351 		MCDI_IN_SET_DWORD(req, INIT_EVQ_V2_IN_TMR_RELOAD, ticks);
352 	}
353 
354 	MCDI_IN_SET_DWORD(req, INIT_EVQ_V2_IN_COUNT_MODE,
355 	    MC_CMD_INIT_EVQ_V2_IN_COUNT_MODE_DIS);
356 	MCDI_IN_SET_DWORD(req, INIT_EVQ_V2_IN_COUNT_THRSHLD, 0);
357 
358 	dma_addr = MCDI_IN2(req, efx_qword_t, INIT_EVQ_V2_IN_DMA_ADDR);
359 	addr = EFSYS_MEM_ADDR(esmp);
360 
361 	for (i = 0; i < npages; i++) {
362 		EFX_POPULATE_QWORD_2(*dma_addr,
363 		    EFX_DWORD_1, (uint32_t)(addr >> 32),
364 		    EFX_DWORD_0, (uint32_t)(addr & 0xffffffff));
365 
366 		dma_addr++;
367 		addr += EFX_BUF_SIZE;
368 	}
369 
370 	efx_mcdi_execute(enp, &req);
371 
372 	if (req.emr_rc != 0) {
373 		rc = req.emr_rc;
374 		goto fail4;
375 	}
376 
377 	if (req.emr_out_length_used < MC_CMD_INIT_EVQ_V2_OUT_LEN) {
378 		rc = EMSGSIZE;
379 		goto fail5;
380 	}
381 
382 	/* NOTE: ignore the returned IRQ param as firmware does not set it. */
383 
384 	EFSYS_PROBE1(mcdi_evq_flags, uint32_t,
385 		    MCDI_OUT_DWORD(req, INIT_EVQ_V2_OUT_FLAGS));
386 
387 	return (0);
388 
389 fail5:
390 	EFSYS_PROBE(fail5);
391 fail4:
392 	EFSYS_PROBE(fail4);
393 fail3:
394 	EFSYS_PROBE(fail3);
395 fail2:
396 	EFSYS_PROBE(fail2);
397 fail1:
398 	EFSYS_PROBE1(fail1, efx_rc_t, rc);
399 
400 	return (rc);
401 }
402 
403 static	__checkReturn	efx_rc_t
404 efx_mcdi_fini_evq(
405 	__in		efx_nic_t *enp,
406 	__in		uint32_t instance)
407 {
408 	efx_mcdi_req_t req;
409 	EFX_MCDI_DECLARE_BUF(payload, MC_CMD_FINI_EVQ_IN_LEN,
410 		MC_CMD_FINI_EVQ_OUT_LEN);
411 	efx_rc_t rc;
412 
413 	req.emr_cmd = MC_CMD_FINI_EVQ;
414 	req.emr_in_buf = payload;
415 	req.emr_in_length = MC_CMD_FINI_EVQ_IN_LEN;
416 	req.emr_out_buf = payload;
417 	req.emr_out_length = MC_CMD_FINI_EVQ_OUT_LEN;
418 
419 	MCDI_IN_SET_DWORD(req, FINI_EVQ_IN_INSTANCE, instance);
420 
421 	efx_mcdi_execute_quiet(enp, &req);
422 
423 	if (req.emr_rc != 0) {
424 		rc = req.emr_rc;
425 		goto fail1;
426 	}
427 
428 	return (0);
429 
430 fail1:
431 	/*
432 	 * EALREADY is not an error, but indicates that the MC has rebooted and
433 	 * that the EVQ has already been destroyed.
434 	 */
435 	if (rc != EALREADY)
436 		EFSYS_PROBE1(fail1, efx_rc_t, rc);
437 
438 	return (rc);
439 }
440 
441 	__checkReturn	efx_rc_t
442 ef10_ev_init(
443 	__in		efx_nic_t *enp)
444 {
445 	_NOTE(ARGUNUSED(enp))
446 	return (0);
447 }
448 
449 			void
450 ef10_ev_fini(
451 	__in		efx_nic_t *enp)
452 {
453 	_NOTE(ARGUNUSED(enp))
454 }
455 
456 	__checkReturn	efx_rc_t
457 ef10_ev_qcreate(
458 	__in		efx_nic_t *enp,
459 	__in		unsigned int index,
460 	__in		efsys_mem_t *esmp,
461 	__in		size_t ndescs,
462 	__in		uint32_t id,
463 	__in		uint32_t us,
464 	__in		uint32_t flags,
465 	__in		efx_evq_t *eep)
466 {
467 	efx_nic_cfg_t *encp = &(enp->en_nic_cfg);
468 	uint32_t irq;
469 	efx_rc_t rc;
470 
471 	_NOTE(ARGUNUSED(id))	/* buftbl id managed by MC */
472 	EFX_STATIC_ASSERT(ISP2(EFX_EVQ_MAXNEVS));
473 	EFX_STATIC_ASSERT(ISP2(EFX_EVQ_MINNEVS));
474 
475 	if (!ISP2(ndescs) ||
476 	    (ndescs < EFX_EVQ_MINNEVS) || (ndescs > EFX_EVQ_MAXNEVS)) {
477 		rc = EINVAL;
478 		goto fail1;
479 	}
480 
481 	if (index >= encp->enc_evq_limit) {
482 		rc = EINVAL;
483 		goto fail2;
484 	}
485 
486 	if (us > encp->enc_evq_timer_max_us) {
487 		rc = EINVAL;
488 		goto fail3;
489 	}
490 
491 	/* Set up the handler table */
492 	eep->ee_rx	= ef10_ev_rx;
493 	eep->ee_tx	= ef10_ev_tx;
494 	eep->ee_driver	= ef10_ev_driver;
495 	eep->ee_drv_gen	= ef10_ev_drv_gen;
496 	eep->ee_mcdi	= ef10_ev_mcdi;
497 
498 	/* Set up the event queue */
499 	/* INIT_EVQ expects function-relative vector number */
500 	if ((flags & EFX_EVQ_FLAGS_NOTIFY_MASK) ==
501 	    EFX_EVQ_FLAGS_NOTIFY_INTERRUPT) {
502 		irq = index;
503 	} else if (index == EFX_EF10_ALWAYS_INTERRUPTING_EVQ_INDEX) {
504 		irq = index;
505 		flags = (flags & ~EFX_EVQ_FLAGS_NOTIFY_MASK) |
506 		    EFX_EVQ_FLAGS_NOTIFY_INTERRUPT;
507 	} else {
508 		irq = EFX_EF10_ALWAYS_INTERRUPTING_EVQ_INDEX;
509 	}
510 
511 	/*
512 	 * Interrupts may be raised for events immediately after the queue is
513 	 * created. See bug58606.
514 	 */
515 
516 	if (encp->enc_init_evq_v2_supported) {
517 		/*
518 		 * On Medford the low latency license is required to enable RX
519 		 * and event cut through and to disable RX batching.  If event
520 		 * queue type in flags is auto, we let the firmware decide the
521 		 * settings to use. If the adapter has a low latency license,
522 		 * it will choose the best settings for low latency, otherwise
523 		 * it will choose the best settings for throughput.
524 		 */
525 		rc = efx_mcdi_init_evq_v2(enp, index, esmp, ndescs, irq, us,
526 		    flags);
527 		if (rc != 0)
528 			goto fail4;
529 	} else {
530 		/*
531 		 * On Huntington we need to specify the settings to use.
532 		 * If event queue type in flags is auto, we favour throughput
533 		 * if the adapter is running virtualization supporting firmware
534 		 * (i.e. the full featured firmware variant)
535 		 * and latency otherwise. The Ethernet Virtual Bridging
536 		 * capability is used to make this decision. (Note though that
537 		 * the low latency firmware variant is also best for
538 		 * throughput and corresponding type should be specified
539 		 * to choose it.)
540 		 */
541 		boolean_t low_latency = encp->enc_datapath_cap_evb ? 0 : 1;
542 		rc = efx_mcdi_init_evq(enp, index, esmp, ndescs, irq, us, flags,
543 		    low_latency);
544 		if (rc != 0)
545 			goto fail5;
546 	}
547 
548 	return (0);
549 
550 fail5:
551 	EFSYS_PROBE(fail5);
552 fail4:
553 	EFSYS_PROBE(fail4);
554 fail3:
555 	EFSYS_PROBE(fail3);
556 fail2:
557 	EFSYS_PROBE(fail2);
558 fail1:
559 	EFSYS_PROBE1(fail1, efx_rc_t, rc);
560 
561 	return (rc);
562 }
563 
564 			void
565 ef10_ev_qdestroy(
566 	__in		efx_evq_t *eep)
567 {
568 	efx_nic_t *enp = eep->ee_enp;
569 
570 	EFSYS_ASSERT(enp->en_family == EFX_FAMILY_HUNTINGTON ||
571 	    enp->en_family == EFX_FAMILY_MEDFORD ||
572 	    enp->en_family == EFX_FAMILY_MEDFORD2);
573 
574 	(void) efx_mcdi_fini_evq(enp, eep->ee_index);
575 }
576 
577 	__checkReturn	efx_rc_t
578 ef10_ev_qprime(
579 	__in		efx_evq_t *eep,
580 	__in		unsigned int count)
581 {
582 	efx_nic_t *enp = eep->ee_enp;
583 	uint32_t rptr;
584 	efx_dword_t dword;
585 
586 	rptr = count & eep->ee_mask;
587 
588 	if (enp->en_nic_cfg.enc_bug35388_workaround) {
589 		EFX_STATIC_ASSERT(EFX_EVQ_MINNEVS >
590 		    (1 << ERF_DD_EVQ_IND_RPTR_WIDTH));
591 		EFX_STATIC_ASSERT(EFX_EVQ_MAXNEVS <
592 		    (1 << 2 * ERF_DD_EVQ_IND_RPTR_WIDTH));
593 
594 		EFX_POPULATE_DWORD_2(dword,
595 		    ERF_DD_EVQ_IND_RPTR_FLAGS,
596 		    EFE_DD_EVQ_IND_RPTR_FLAGS_HIGH,
597 		    ERF_DD_EVQ_IND_RPTR,
598 		    (rptr >> ERF_DD_EVQ_IND_RPTR_WIDTH));
599 		EFX_BAR_VI_WRITED(enp, ER_DD_EVQ_INDIRECT, eep->ee_index,
600 		    &dword, B_FALSE);
601 
602 		EFX_POPULATE_DWORD_2(dword,
603 		    ERF_DD_EVQ_IND_RPTR_FLAGS,
604 		    EFE_DD_EVQ_IND_RPTR_FLAGS_LOW,
605 		    ERF_DD_EVQ_IND_RPTR,
606 		    rptr & ((1 << ERF_DD_EVQ_IND_RPTR_WIDTH) - 1));
607 		EFX_BAR_VI_WRITED(enp, ER_DD_EVQ_INDIRECT, eep->ee_index,
608 		    &dword, B_FALSE);
609 	} else {
610 		EFX_POPULATE_DWORD_1(dword, ERF_DZ_EVQ_RPTR, rptr);
611 		EFX_BAR_VI_WRITED(enp, ER_DZ_EVQ_RPTR_REG, eep->ee_index,
612 		    &dword, B_FALSE);
613 	}
614 
615 	return (0);
616 }
617 
618 static	__checkReturn	efx_rc_t
619 efx_mcdi_driver_event(
620 	__in		efx_nic_t *enp,
621 	__in		uint32_t evq,
622 	__in		efx_qword_t data)
623 {
624 	efx_mcdi_req_t req;
625 	EFX_MCDI_DECLARE_BUF(payload, MC_CMD_DRIVER_EVENT_IN_LEN,
626 		MC_CMD_DRIVER_EVENT_OUT_LEN);
627 	efx_rc_t rc;
628 
629 	req.emr_cmd = MC_CMD_DRIVER_EVENT;
630 	req.emr_in_buf = payload;
631 	req.emr_in_length = MC_CMD_DRIVER_EVENT_IN_LEN;
632 	req.emr_out_buf = payload;
633 	req.emr_out_length = MC_CMD_DRIVER_EVENT_OUT_LEN;
634 
635 	MCDI_IN_SET_DWORD(req, DRIVER_EVENT_IN_EVQ, evq);
636 
637 	MCDI_IN_SET_DWORD(req, DRIVER_EVENT_IN_DATA_LO,
638 	    EFX_QWORD_FIELD(data, EFX_DWORD_0));
639 	MCDI_IN_SET_DWORD(req, DRIVER_EVENT_IN_DATA_HI,
640 	    EFX_QWORD_FIELD(data, EFX_DWORD_1));
641 
642 	efx_mcdi_execute(enp, &req);
643 
644 	if (req.emr_rc != 0) {
645 		rc = req.emr_rc;
646 		goto fail1;
647 	}
648 
649 	return (0);
650 
651 fail1:
652 	EFSYS_PROBE1(fail1, efx_rc_t, rc);
653 
654 	return (rc);
655 }
656 
657 			void
658 ef10_ev_qpost(
659 	__in	efx_evq_t *eep,
660 	__in	uint16_t data)
661 {
662 	efx_nic_t *enp = eep->ee_enp;
663 	efx_qword_t event;
664 
665 	EFX_POPULATE_QWORD_3(event,
666 	    ESF_DZ_DRV_CODE, ESE_DZ_EV_CODE_DRV_GEN_EV,
667 	    ESF_DZ_DRV_SUB_CODE, 0,
668 	    ESF_DZ_DRV_SUB_DATA_DW0, (uint32_t)data);
669 
670 	(void) efx_mcdi_driver_event(enp, eep->ee_index, event);
671 }
672 
673 	__checkReturn	efx_rc_t
674 ef10_ev_qmoderate(
675 	__in		efx_evq_t *eep,
676 	__in		unsigned int us)
677 {
678 	efx_nic_t *enp = eep->ee_enp;
679 	efx_nic_cfg_t *encp = &(enp->en_nic_cfg);
680 	efx_dword_t dword;
681 	uint32_t mode;
682 	efx_rc_t rc;
683 
684 	/* Check that hardware and MCDI use the same timer MODE values */
685 	EFX_STATIC_ASSERT(FFE_CZ_TIMER_MODE_DIS ==
686 	    MC_CMD_SET_EVQ_TMR_IN_TIMER_MODE_DIS);
687 	EFX_STATIC_ASSERT(FFE_CZ_TIMER_MODE_IMMED_START ==
688 	    MC_CMD_SET_EVQ_TMR_IN_TIMER_MODE_IMMED_START);
689 	EFX_STATIC_ASSERT(FFE_CZ_TIMER_MODE_TRIG_START ==
690 	    MC_CMD_SET_EVQ_TMR_IN_TIMER_MODE_TRIG_START);
691 	EFX_STATIC_ASSERT(FFE_CZ_TIMER_MODE_INT_HLDOFF ==
692 	    MC_CMD_SET_EVQ_TMR_IN_TIMER_MODE_INT_HLDOFF);
693 
694 	if (us > encp->enc_evq_timer_max_us) {
695 		rc = EINVAL;
696 		goto fail1;
697 	}
698 
699 	/* If the value is zero then disable the timer */
700 	if (us == 0) {
701 		mode = FFE_CZ_TIMER_MODE_DIS;
702 	} else {
703 		mode = FFE_CZ_TIMER_MODE_INT_HLDOFF;
704 	}
705 
706 	if (encp->enc_bug61265_workaround) {
707 		uint32_t ns = us * 1000;
708 
709 		rc = efx_mcdi_set_evq_tmr(enp, eep->ee_index, mode, ns);
710 		if (rc != 0)
711 			goto fail2;
712 	} else {
713 		unsigned int ticks;
714 
715 		if ((rc = efx_ev_usecs_to_ticks(enp, us, &ticks)) != 0)
716 			goto fail3;
717 
718 		if (encp->enc_bug35388_workaround) {
719 			EFX_POPULATE_DWORD_3(dword,
720 			    ERF_DD_EVQ_IND_TIMER_FLAGS,
721 			    EFE_DD_EVQ_IND_TIMER_FLAGS,
722 			    ERF_DD_EVQ_IND_TIMER_MODE, mode,
723 			    ERF_DD_EVQ_IND_TIMER_VAL, ticks);
724 			EFX_BAR_VI_WRITED(enp, ER_DD_EVQ_INDIRECT,
725 			    eep->ee_index, &dword, 0);
726 		} else {
727 			/*
728 			 * NOTE: The TMR_REL field introduced in Medford2 is
729 			 * ignored on earlier EF10 controllers. See bug66418
730 			 * comment 9 for details.
731 			 */
732 			EFX_POPULATE_DWORD_3(dword,
733 			    ERF_DZ_TC_TIMER_MODE, mode,
734 			    ERF_DZ_TC_TIMER_VAL, ticks,
735 			    ERF_FZ_TC_TMR_REL_VAL, ticks);
736 			EFX_BAR_VI_WRITED(enp, ER_DZ_EVQ_TMR_REG,
737 			    eep->ee_index, &dword, 0);
738 		}
739 	}
740 
741 	return (0);
742 
743 fail3:
744 	EFSYS_PROBE(fail3);
745 fail2:
746 	EFSYS_PROBE(fail2);
747 fail1:
748 	EFSYS_PROBE1(fail1, efx_rc_t, rc);
749 
750 	return (rc);
751 }
752 
753 #if EFSYS_OPT_QSTATS
754 			void
755 ef10_ev_qstats_update(
756 	__in				efx_evq_t *eep,
757 	__inout_ecount(EV_NQSTATS)	efsys_stat_t *stat)
758 {
759 	unsigned int id;
760 
761 	for (id = 0; id < EV_NQSTATS; id++) {
762 		efsys_stat_t *essp = &stat[id];
763 
764 		EFSYS_STAT_INCR(essp, eep->ee_stat[id]);
765 		eep->ee_stat[id] = 0;
766 	}
767 }
768 #endif /* EFSYS_OPT_QSTATS */
769 
770 #if EFSYS_OPT_RX_PACKED_STREAM || EFSYS_OPT_RX_ES_SUPER_BUFFER
771 
772 static	__checkReturn	boolean_t
773 ef10_ev_rx_packed_stream(
774 	__in		efx_evq_t *eep,
775 	__in		efx_qword_t *eqp,
776 	__in		const efx_ev_callbacks_t *eecp,
777 	__in_opt	void *arg)
778 {
779 	uint32_t label;
780 	uint32_t pkt_count_lbits;
781 	uint16_t flags;
782 	boolean_t should_abort;
783 	efx_evq_rxq_state_t *eersp;
784 	unsigned int pkt_count;
785 	unsigned int current_id;
786 	boolean_t new_buffer;
787 
788 	pkt_count_lbits = EFX_QWORD_FIELD(*eqp, ESF_DZ_RX_DSC_PTR_LBITS);
789 	label = EFX_QWORD_FIELD(*eqp, ESF_DZ_RX_QLABEL);
790 	new_buffer = EFX_QWORD_FIELD(*eqp, ESF_DZ_RX_EV_ROTATE);
791 
792 	flags = 0;
793 
794 	eersp = &eep->ee_rxq_state[label];
795 
796 	/*
797 	 * RX_DSC_PTR_LBITS has least significant bits of the global
798 	 * (not per-buffer) packet counter. It is guaranteed that
799 	 * maximum number of completed packets fits in lbits-mask.
800 	 * So, modulo lbits-mask arithmetic should be used to calculate
801 	 * packet counter increment.
802 	 */
803 	pkt_count = (pkt_count_lbits - eersp->eers_rx_stream_npackets) &
804 	    EFX_MASK32(ESF_DZ_RX_DSC_PTR_LBITS);
805 	eersp->eers_rx_stream_npackets += pkt_count;
806 
807 	if (new_buffer) {
808 		flags |= EFX_PKT_PACKED_STREAM_NEW_BUFFER;
809 #if EFSYS_OPT_RX_PACKED_STREAM
810 		/*
811 		 * If both packed stream and equal stride super-buffer
812 		 * modes are compiled in, in theory credits should be
813 		 * be maintained for packed stream only, but right now
814 		 * these modes are not distinguished in the event queue
815 		 * Rx queue state and it is OK to increment the counter
816 		 * regardless (it might be event cheaper than branching
817 		 * since neighbour structure member are updated as well).
818 		 */
819 		eersp->eers_rx_packed_stream_credits++;
820 #endif
821 		eersp->eers_rx_read_ptr++;
822 	}
823 	current_id = eersp->eers_rx_read_ptr & eersp->eers_rx_mask;
824 
825 	/* Check for errors that invalidate checksum and L3/L4 fields */
826 	if (EFX_QWORD_FIELD(*eqp, ESF_DZ_RX_TRUNC_ERR) != 0) {
827 		/* RX frame truncated */
828 		EFX_EV_QSTAT_INCR(eep, EV_RX_FRM_TRUNC);
829 		flags |= EFX_DISCARD;
830 		goto deliver;
831 	}
832 	if (EFX_QWORD_FIELD(*eqp, ESF_DZ_RX_ECRC_ERR) != 0) {
833 		/* Bad Ethernet frame CRC */
834 		EFX_EV_QSTAT_INCR(eep, EV_RX_ETH_CRC_ERR);
835 		flags |= EFX_DISCARD;
836 		goto deliver;
837 	}
838 
839 	if (EFX_QWORD_FIELD(*eqp, ESF_DZ_RX_PARSE_INCOMPLETE)) {
840 		flags |= EFX_PKT_PACKED_STREAM_PARSE_INCOMPLETE;
841 		goto deliver;
842 	}
843 
844 	if (EFX_QWORD_FIELD(*eqp, ESF_DZ_RX_IPCKSUM_ERR))
845 		EFX_EV_QSTAT_INCR(eep, EV_RX_IPV4_HDR_CHKSUM_ERR);
846 
847 	if (EFX_QWORD_FIELD(*eqp, ESF_DZ_RX_TCPUDP_CKSUM_ERR))
848 		EFX_EV_QSTAT_INCR(eep, EV_RX_TCP_UDP_CHKSUM_ERR);
849 
850 deliver:
851 	/* If we're not discarding the packet then it is ok */
852 	if (~flags & EFX_DISCARD)
853 		EFX_EV_QSTAT_INCR(eep, EV_RX_OK);
854 
855 	EFSYS_ASSERT(eecp->eec_rx_ps != NULL);
856 	should_abort = eecp->eec_rx_ps(arg, label, current_id, pkt_count,
857 	    flags);
858 
859 	return (should_abort);
860 }
861 
862 #endif /* EFSYS_OPT_RX_PACKED_STREAM || EFSYS_OPT_RX_ES_SUPER_BUFFER */
863 
864 static	__checkReturn	boolean_t
865 ef10_ev_rx(
866 	__in		efx_evq_t *eep,
867 	__in		efx_qword_t *eqp,
868 	__in		const efx_ev_callbacks_t *eecp,
869 	__in_opt	void *arg)
870 {
871 	efx_nic_t *enp = eep->ee_enp;
872 	uint32_t size;
873 	uint32_t label;
874 	uint32_t mac_class;
875 	uint32_t eth_tag_class;
876 	uint32_t l3_class;
877 	uint32_t l4_class;
878 	uint32_t next_read_lbits;
879 	uint16_t flags;
880 	boolean_t cont;
881 	boolean_t should_abort;
882 	efx_evq_rxq_state_t *eersp;
883 	unsigned int desc_count;
884 	unsigned int last_used_id;
885 
886 	EFX_EV_QSTAT_INCR(eep, EV_RX);
887 
888 	/* Discard events after RXQ/TXQ errors, or hardware not available */
889 	if (enp->en_reset_flags &
890 	    (EFX_RESET_RXQ_ERR | EFX_RESET_TXQ_ERR | EFX_RESET_HW_UNAVAIL))
891 		return (B_FALSE);
892 
893 	/* Basic packet information */
894 	label = EFX_QWORD_FIELD(*eqp, ESF_DZ_RX_QLABEL);
895 	eersp = &eep->ee_rxq_state[label];
896 
897 #if EFSYS_OPT_RX_PACKED_STREAM || EFSYS_OPT_RX_ES_SUPER_BUFFER
898 	/*
899 	 * Packed stream events are very different,
900 	 * so handle them separately
901 	 */
902 	if (eersp->eers_rx_packed_stream)
903 	    return (ef10_ev_rx_packed_stream(eep, eqp, eecp, arg));
904 #endif
905 
906 	size = EFX_QWORD_FIELD(*eqp, ESF_DZ_RX_BYTES);
907 	cont = EFX_QWORD_FIELD(*eqp, ESF_DZ_RX_CONT);
908 	next_read_lbits = EFX_QWORD_FIELD(*eqp, ESF_DZ_RX_DSC_PTR_LBITS);
909 	eth_tag_class = EFX_QWORD_FIELD(*eqp, ESF_DZ_RX_ETH_TAG_CLASS);
910 	mac_class = EFX_QWORD_FIELD(*eqp, ESF_DZ_RX_MAC_CLASS);
911 	l3_class = EFX_QWORD_FIELD(*eqp, ESF_DZ_RX_L3_CLASS);
912 
913 	/*
914 	 * RX_L4_CLASS is 3 bits wide on Huntington and Medford, but is only
915 	 * 2 bits wide on Medford2. Check it is safe to use the Medford2 field
916 	 * and values for all EF10 controllers.
917 	 */
918 	EFX_STATIC_ASSERT(ESF_FZ_RX_L4_CLASS_LBN == ESF_DE_RX_L4_CLASS_LBN);
919 	EFX_STATIC_ASSERT(ESE_FZ_L4_CLASS_TCP == ESE_DE_L4_CLASS_TCP);
920 	EFX_STATIC_ASSERT(ESE_FZ_L4_CLASS_UDP == ESE_DE_L4_CLASS_UDP);
921 	EFX_STATIC_ASSERT(ESE_FZ_L4_CLASS_UNKNOWN == ESE_DE_L4_CLASS_UNKNOWN);
922 
923 	l4_class = EFX_QWORD_FIELD(*eqp, ESF_FZ_RX_L4_CLASS);
924 
925 	if (EFX_QWORD_FIELD(*eqp, ESF_DZ_RX_DROP_EVENT) != 0) {
926 		/* Drop this event */
927 		return (B_FALSE);
928 	}
929 	flags = 0;
930 
931 	if (cont != 0) {
932 		/*
933 		 * This may be part of a scattered frame, or it may be a
934 		 * truncated frame if scatter is disabled on this RXQ.
935 		 * Overlength frames can be received if e.g. a VF is configured
936 		 * for 1500 MTU but connected to a port set to 9000 MTU
937 		 * (see bug56567).
938 		 * FIXME: There is not yet any driver that supports scatter on
939 		 * Huntington.  Scatter support is required for OSX.
940 		 */
941 		flags |= EFX_PKT_CONT;
942 	}
943 
944 	if (mac_class == ESE_DZ_MAC_CLASS_UCAST)
945 		flags |= EFX_PKT_UNICAST;
946 
947 	/* Increment the count of descriptors read */
948 	desc_count = (next_read_lbits - eersp->eers_rx_read_ptr) &
949 	    EFX_MASK32(ESF_DZ_RX_DSC_PTR_LBITS);
950 	eersp->eers_rx_read_ptr += desc_count;
951 
952 	/*
953 	 * FIXME: add error checking to make sure this a batched event.
954 	 * This could also be an aborted scatter, see Bug36629.
955 	 */
956 	if (desc_count > 1) {
957 		EFX_EV_QSTAT_INCR(eep, EV_RX_BATCH);
958 		flags |= EFX_PKT_PREFIX_LEN;
959 	}
960 
961 	/* Calculate the index of the last descriptor consumed */
962 	last_used_id = (eersp->eers_rx_read_ptr - 1) & eersp->eers_rx_mask;
963 
964 	/* Check for errors that invalidate checksum and L3/L4 fields */
965 	if (EFX_QWORD_FIELD(*eqp, ESF_DZ_RX_TRUNC_ERR) != 0) {
966 		/* RX frame truncated */
967 		EFX_EV_QSTAT_INCR(eep, EV_RX_FRM_TRUNC);
968 		flags |= EFX_DISCARD;
969 		goto deliver;
970 	}
971 	if (EFX_QWORD_FIELD(*eqp, ESF_DZ_RX_ECRC_ERR) != 0) {
972 		/* Bad Ethernet frame CRC */
973 		EFX_EV_QSTAT_INCR(eep, EV_RX_ETH_CRC_ERR);
974 		flags |= EFX_DISCARD;
975 		goto deliver;
976 	}
977 	if (EFX_QWORD_FIELD(*eqp, ESF_DZ_RX_PARSE_INCOMPLETE)) {
978 		/*
979 		 * Hardware parse failed, due to malformed headers
980 		 * or headers that are too long for the parser.
981 		 * Headers and checksums must be validated by the host.
982 		 */
983 		/* TODO: EFX_EV_QSTAT_INCR(eep, EV_RX_PARSE_INCOMPLETE); */
984 		goto deliver;
985 	}
986 
987 	if ((eth_tag_class == ESE_DZ_ETH_TAG_CLASS_VLAN1) ||
988 	    (eth_tag_class == ESE_DZ_ETH_TAG_CLASS_VLAN2)) {
989 		flags |= EFX_PKT_VLAN_TAGGED;
990 	}
991 
992 	switch (l3_class) {
993 	case ESE_DZ_L3_CLASS_IP4:
994 	case ESE_DZ_L3_CLASS_IP4_FRAG:
995 		flags |= EFX_PKT_IPV4;
996 		if (EFX_QWORD_FIELD(*eqp, ESF_DZ_RX_IPCKSUM_ERR)) {
997 			EFX_EV_QSTAT_INCR(eep, EV_RX_IPV4_HDR_CHKSUM_ERR);
998 		} else {
999 			flags |= EFX_CKSUM_IPV4;
1000 		}
1001 
1002 		/*
1003 		 * RX_L4_CLASS is 3 bits wide on Huntington and Medford, but is
1004 		 * only 2 bits wide on Medford2. Check it is safe to use the
1005 		 * Medford2 field and values for all EF10 controllers.
1006 		 */
1007 		EFX_STATIC_ASSERT(ESF_FZ_RX_L4_CLASS_LBN ==
1008 		    ESF_DE_RX_L4_CLASS_LBN);
1009 		EFX_STATIC_ASSERT(ESE_FZ_L4_CLASS_TCP == ESE_DE_L4_CLASS_TCP);
1010 		EFX_STATIC_ASSERT(ESE_FZ_L4_CLASS_UDP == ESE_DE_L4_CLASS_UDP);
1011 		EFX_STATIC_ASSERT(ESE_FZ_L4_CLASS_UNKNOWN ==
1012 		    ESE_DE_L4_CLASS_UNKNOWN);
1013 
1014 		if (l4_class == ESE_FZ_L4_CLASS_TCP) {
1015 			EFX_EV_QSTAT_INCR(eep, EV_RX_TCP_IPV4);
1016 			flags |= EFX_PKT_TCP;
1017 		} else if (l4_class == ESE_FZ_L4_CLASS_UDP) {
1018 			EFX_EV_QSTAT_INCR(eep, EV_RX_UDP_IPV4);
1019 			flags |= EFX_PKT_UDP;
1020 		} else {
1021 			EFX_EV_QSTAT_INCR(eep, EV_RX_OTHER_IPV4);
1022 		}
1023 		break;
1024 
1025 	case ESE_DZ_L3_CLASS_IP6:
1026 	case ESE_DZ_L3_CLASS_IP6_FRAG:
1027 		flags |= EFX_PKT_IPV6;
1028 
1029 		/*
1030 		 * RX_L4_CLASS is 3 bits wide on Huntington and Medford, but is
1031 		 * only 2 bits wide on Medford2. Check it is safe to use the
1032 		 * Medford2 field and values for all EF10 controllers.
1033 		 */
1034 		EFX_STATIC_ASSERT(ESF_FZ_RX_L4_CLASS_LBN ==
1035 		    ESF_DE_RX_L4_CLASS_LBN);
1036 		EFX_STATIC_ASSERT(ESE_FZ_L4_CLASS_TCP == ESE_DE_L4_CLASS_TCP);
1037 		EFX_STATIC_ASSERT(ESE_FZ_L4_CLASS_UDP == ESE_DE_L4_CLASS_UDP);
1038 		EFX_STATIC_ASSERT(ESE_FZ_L4_CLASS_UNKNOWN ==
1039 		    ESE_DE_L4_CLASS_UNKNOWN);
1040 
1041 		if (l4_class == ESE_FZ_L4_CLASS_TCP) {
1042 			EFX_EV_QSTAT_INCR(eep, EV_RX_TCP_IPV6);
1043 			flags |= EFX_PKT_TCP;
1044 		} else if (l4_class == ESE_FZ_L4_CLASS_UDP) {
1045 			EFX_EV_QSTAT_INCR(eep, EV_RX_UDP_IPV6);
1046 			flags |= EFX_PKT_UDP;
1047 		} else {
1048 			EFX_EV_QSTAT_INCR(eep, EV_RX_OTHER_IPV6);
1049 		}
1050 		break;
1051 
1052 	default:
1053 		EFX_EV_QSTAT_INCR(eep, EV_RX_NON_IP);
1054 		break;
1055 	}
1056 
1057 	if (flags & (EFX_PKT_TCP | EFX_PKT_UDP)) {
1058 		if (EFX_QWORD_FIELD(*eqp, ESF_DZ_RX_TCPUDP_CKSUM_ERR)) {
1059 			EFX_EV_QSTAT_INCR(eep, EV_RX_TCP_UDP_CHKSUM_ERR);
1060 		} else {
1061 			flags |= EFX_CKSUM_TCPUDP;
1062 		}
1063 	}
1064 
1065 deliver:
1066 	/* If we're not discarding the packet then it is ok */
1067 	if (~flags & EFX_DISCARD)
1068 		EFX_EV_QSTAT_INCR(eep, EV_RX_OK);
1069 
1070 	EFSYS_ASSERT(eecp->eec_rx != NULL);
1071 	should_abort = eecp->eec_rx(arg, label, last_used_id, size, flags);
1072 
1073 	return (should_abort);
1074 }
1075 
1076 static	__checkReturn	boolean_t
1077 ef10_ev_tx(
1078 	__in		efx_evq_t *eep,
1079 	__in		efx_qword_t *eqp,
1080 	__in		const efx_ev_callbacks_t *eecp,
1081 	__in_opt	void *arg)
1082 {
1083 	efx_nic_t *enp = eep->ee_enp;
1084 	uint32_t id;
1085 	uint32_t label;
1086 	boolean_t should_abort;
1087 
1088 	EFX_EV_QSTAT_INCR(eep, EV_TX);
1089 
1090 	/* Discard events after RXQ/TXQ errors, or hardware not available */
1091 	if (enp->en_reset_flags &
1092 	    (EFX_RESET_RXQ_ERR | EFX_RESET_TXQ_ERR | EFX_RESET_HW_UNAVAIL))
1093 		return (B_FALSE);
1094 
1095 	if (EFX_QWORD_FIELD(*eqp, ESF_DZ_TX_DROP_EVENT) != 0) {
1096 		/* Drop this event */
1097 		return (B_FALSE);
1098 	}
1099 
1100 	/* Per-packet TX completion (was per-descriptor for Falcon/Siena) */
1101 	id = EFX_QWORD_FIELD(*eqp, ESF_DZ_TX_DESCR_INDX);
1102 	label = EFX_QWORD_FIELD(*eqp, ESF_DZ_TX_QLABEL);
1103 
1104 	EFSYS_PROBE2(tx_complete, uint32_t, label, uint32_t, id);
1105 
1106 	EFSYS_ASSERT(eecp->eec_tx != NULL);
1107 	should_abort = eecp->eec_tx(arg, label, id);
1108 
1109 	return (should_abort);
1110 }
1111 
1112 static	__checkReturn	boolean_t
1113 ef10_ev_driver(
1114 	__in		efx_evq_t *eep,
1115 	__in		efx_qword_t *eqp,
1116 	__in		const efx_ev_callbacks_t *eecp,
1117 	__in_opt	void *arg)
1118 {
1119 	unsigned int code;
1120 	boolean_t should_abort;
1121 
1122 	EFX_EV_QSTAT_INCR(eep, EV_DRIVER);
1123 	should_abort = B_FALSE;
1124 
1125 	code = EFX_QWORD_FIELD(*eqp, ESF_DZ_DRV_SUB_CODE);
1126 	switch (code) {
1127 	case ESE_DZ_DRV_TIMER_EV: {
1128 		uint32_t id;
1129 
1130 		id = EFX_QWORD_FIELD(*eqp, ESF_DZ_DRV_TMR_ID);
1131 
1132 		EFSYS_ASSERT(eecp->eec_timer != NULL);
1133 		should_abort = eecp->eec_timer(arg, id);
1134 		break;
1135 	}
1136 
1137 	case ESE_DZ_DRV_WAKE_UP_EV: {
1138 		uint32_t id;
1139 
1140 		id = EFX_QWORD_FIELD(*eqp, ESF_DZ_DRV_EVQ_ID);
1141 
1142 		EFSYS_ASSERT(eecp->eec_wake_up != NULL);
1143 		should_abort = eecp->eec_wake_up(arg, id);
1144 		break;
1145 	}
1146 
1147 	case ESE_DZ_DRV_START_UP_EV:
1148 		EFSYS_ASSERT(eecp->eec_initialized != NULL);
1149 		should_abort = eecp->eec_initialized(arg);
1150 		break;
1151 
1152 	default:
1153 		EFSYS_PROBE3(bad_event, unsigned int, eep->ee_index,
1154 		    uint32_t, EFX_QWORD_FIELD(*eqp, EFX_DWORD_1),
1155 		    uint32_t, EFX_QWORD_FIELD(*eqp, EFX_DWORD_0));
1156 		break;
1157 	}
1158 
1159 	return (should_abort);
1160 }
1161 
1162 static	__checkReturn	boolean_t
1163 ef10_ev_drv_gen(
1164 	__in		efx_evq_t *eep,
1165 	__in		efx_qword_t *eqp,
1166 	__in		const efx_ev_callbacks_t *eecp,
1167 	__in_opt	void *arg)
1168 {
1169 	uint32_t data;
1170 	boolean_t should_abort;
1171 
1172 	EFX_EV_QSTAT_INCR(eep, EV_DRV_GEN);
1173 	should_abort = B_FALSE;
1174 
1175 	data = EFX_QWORD_FIELD(*eqp, ESF_DZ_DRV_SUB_DATA_DW0);
1176 	if (data >= ((uint32_t)1 << 16)) {
1177 		EFSYS_PROBE3(bad_event, unsigned int, eep->ee_index,
1178 		    uint32_t, EFX_QWORD_FIELD(*eqp, EFX_DWORD_1),
1179 		    uint32_t, EFX_QWORD_FIELD(*eqp, EFX_DWORD_0));
1180 
1181 		return (B_TRUE);
1182 	}
1183 
1184 	EFSYS_ASSERT(eecp->eec_software != NULL);
1185 	should_abort = eecp->eec_software(arg, (uint16_t)data);
1186 
1187 	return (should_abort);
1188 }
1189 
1190 static	__checkReturn	boolean_t
1191 ef10_ev_mcdi(
1192 	__in		efx_evq_t *eep,
1193 	__in		efx_qword_t *eqp,
1194 	__in		const efx_ev_callbacks_t *eecp,
1195 	__in_opt	void *arg)
1196 {
1197 	efx_nic_t *enp = eep->ee_enp;
1198 	unsigned int code;
1199 	boolean_t should_abort = B_FALSE;
1200 
1201 	EFX_EV_QSTAT_INCR(eep, EV_MCDI_RESPONSE);
1202 
1203 	code = EFX_QWORD_FIELD(*eqp, MCDI_EVENT_CODE);
1204 	switch (code) {
1205 	case MCDI_EVENT_CODE_BADSSERT:
1206 		efx_mcdi_ev_death(enp, EINTR);
1207 		break;
1208 
1209 	case MCDI_EVENT_CODE_CMDDONE:
1210 		efx_mcdi_ev_cpl(enp,
1211 		    MCDI_EV_FIELD(eqp, CMDDONE_SEQ),
1212 		    MCDI_EV_FIELD(eqp, CMDDONE_DATALEN),
1213 		    MCDI_EV_FIELD(eqp, CMDDONE_ERRNO));
1214 		break;
1215 
1216 #if EFSYS_OPT_MCDI_PROXY_AUTH
1217 	case MCDI_EVENT_CODE_PROXY_RESPONSE:
1218 		/*
1219 		 * This event notifies a function that an authorization request
1220 		 * has been processed. If the request was authorized then the
1221 		 * function can now re-send the original MCDI request.
1222 		 * See SF-113652-SW "SR-IOV Proxied Network Access Control".
1223 		 */
1224 		efx_mcdi_ev_proxy_response(enp,
1225 		    MCDI_EV_FIELD(eqp, PROXY_RESPONSE_HANDLE),
1226 		    MCDI_EV_FIELD(eqp, PROXY_RESPONSE_RC));
1227 		break;
1228 #endif /* EFSYS_OPT_MCDI_PROXY_AUTH */
1229 
1230 	case MCDI_EVENT_CODE_LINKCHANGE: {
1231 		efx_link_mode_t link_mode;
1232 
1233 		ef10_phy_link_ev(enp, eqp, &link_mode);
1234 		should_abort = eecp->eec_link_change(arg, link_mode);
1235 		break;
1236 	}
1237 
1238 	case MCDI_EVENT_CODE_SENSOREVT: {
1239 #if EFSYS_OPT_MON_STATS
1240 		efx_mon_stat_t id;
1241 		efx_mon_stat_value_t value;
1242 		efx_rc_t rc;
1243 
1244 		/* Decode monitor stat for MCDI sensor (if supported) */
1245 		if ((rc = mcdi_mon_ev(enp, eqp, &id, &value)) == 0) {
1246 			/* Report monitor stat change */
1247 			should_abort = eecp->eec_monitor(arg, id, value);
1248 		} else if (rc == ENOTSUP) {
1249 			should_abort = eecp->eec_exception(arg,
1250 				EFX_EXCEPTION_UNKNOWN_SENSOREVT,
1251 				MCDI_EV_FIELD(eqp, DATA));
1252 		} else {
1253 			EFSYS_ASSERT(rc == ENODEV);	/* Wrong port */
1254 		}
1255 #endif
1256 		break;
1257 	}
1258 
1259 	case MCDI_EVENT_CODE_SCHEDERR:
1260 		/* Informational only */
1261 		break;
1262 
1263 	case MCDI_EVENT_CODE_REBOOT:
1264 		/* Falcon/Siena only (should not been seen with Huntington). */
1265 		efx_mcdi_ev_death(enp, EIO);
1266 		break;
1267 
1268 	case MCDI_EVENT_CODE_MC_REBOOT:
1269 		/* MC_REBOOT event is used for Huntington (EF10) and later. */
1270 		efx_mcdi_ev_death(enp, EIO);
1271 		break;
1272 
1273 	case MCDI_EVENT_CODE_MAC_STATS_DMA:
1274 #if EFSYS_OPT_MAC_STATS
1275 		if (eecp->eec_mac_stats != NULL) {
1276 			eecp->eec_mac_stats(arg,
1277 			    MCDI_EV_FIELD(eqp, MAC_STATS_DMA_GENERATION));
1278 		}
1279 #endif
1280 		break;
1281 
1282 	case MCDI_EVENT_CODE_FWALERT: {
1283 		uint32_t reason = MCDI_EV_FIELD(eqp, FWALERT_REASON);
1284 
1285 		if (reason == MCDI_EVENT_FWALERT_REASON_SRAM_ACCESS)
1286 			should_abort = eecp->eec_exception(arg,
1287 				EFX_EXCEPTION_FWALERT_SRAM,
1288 				MCDI_EV_FIELD(eqp, FWALERT_DATA));
1289 		else
1290 			should_abort = eecp->eec_exception(arg,
1291 				EFX_EXCEPTION_UNKNOWN_FWALERT,
1292 				MCDI_EV_FIELD(eqp, DATA));
1293 		break;
1294 	}
1295 
1296 	case MCDI_EVENT_CODE_TX_ERR: {
1297 		/*
1298 		 * After a TXQ error is detected, firmware sends a TX_ERR event.
1299 		 * This may be followed by TX completions (which we discard),
1300 		 * and then finally by a TX_FLUSH event. Firmware destroys the
1301 		 * TXQ automatically after sending the TX_FLUSH event.
1302 		 */
1303 		enp->en_reset_flags |= EFX_RESET_TXQ_ERR;
1304 
1305 		EFSYS_PROBE2(tx_descq_err,
1306 			    uint32_t, EFX_QWORD_FIELD(*eqp, EFX_DWORD_1),
1307 			    uint32_t, EFX_QWORD_FIELD(*eqp, EFX_DWORD_0));
1308 
1309 		/* Inform the driver that a reset is required. */
1310 		eecp->eec_exception(arg, EFX_EXCEPTION_TX_ERROR,
1311 		    MCDI_EV_FIELD(eqp, TX_ERR_DATA));
1312 		break;
1313 	}
1314 
1315 	case MCDI_EVENT_CODE_TX_FLUSH: {
1316 		uint32_t txq_index = MCDI_EV_FIELD(eqp, TX_FLUSH_TXQ);
1317 
1318 		/*
1319 		 * EF10 firmware sends two TX_FLUSH events: one to the txq's
1320 		 * event queue, and one to evq 0 (with TX_FLUSH_TO_DRIVER set).
1321 		 * We want to wait for all completions, so ignore the events
1322 		 * with TX_FLUSH_TO_DRIVER.
1323 		 */
1324 		if (MCDI_EV_FIELD(eqp, TX_FLUSH_TO_DRIVER) != 0) {
1325 			should_abort = B_FALSE;
1326 			break;
1327 		}
1328 
1329 		EFX_EV_QSTAT_INCR(eep, EV_DRIVER_TX_DESCQ_FLS_DONE);
1330 
1331 		EFSYS_PROBE1(tx_descq_fls_done, uint32_t, txq_index);
1332 
1333 		EFSYS_ASSERT(eecp->eec_txq_flush_done != NULL);
1334 		should_abort = eecp->eec_txq_flush_done(arg, txq_index);
1335 		break;
1336 	}
1337 
1338 	case MCDI_EVENT_CODE_RX_ERR: {
1339 		/*
1340 		 * After an RXQ error is detected, firmware sends an RX_ERR
1341 		 * event. This may be followed by RX events (which we discard),
1342 		 * and then finally by an RX_FLUSH event. Firmware destroys the
1343 		 * RXQ automatically after sending the RX_FLUSH event.
1344 		 */
1345 		enp->en_reset_flags |= EFX_RESET_RXQ_ERR;
1346 
1347 		EFSYS_PROBE2(rx_descq_err,
1348 			    uint32_t, EFX_QWORD_FIELD(*eqp, EFX_DWORD_1),
1349 			    uint32_t, EFX_QWORD_FIELD(*eqp, EFX_DWORD_0));
1350 
1351 		/* Inform the driver that a reset is required. */
1352 		eecp->eec_exception(arg, EFX_EXCEPTION_RX_ERROR,
1353 		    MCDI_EV_FIELD(eqp, RX_ERR_DATA));
1354 		break;
1355 	}
1356 
1357 	case MCDI_EVENT_CODE_RX_FLUSH: {
1358 		uint32_t rxq_index = MCDI_EV_FIELD(eqp, RX_FLUSH_RXQ);
1359 
1360 		/*
1361 		 * EF10 firmware sends two RX_FLUSH events: one to the rxq's
1362 		 * event queue, and one to evq 0 (with RX_FLUSH_TO_DRIVER set).
1363 		 * We want to wait for all completions, so ignore the events
1364 		 * with RX_FLUSH_TO_DRIVER.
1365 		 */
1366 		if (MCDI_EV_FIELD(eqp, RX_FLUSH_TO_DRIVER) != 0) {
1367 			should_abort = B_FALSE;
1368 			break;
1369 		}
1370 
1371 		EFX_EV_QSTAT_INCR(eep, EV_DRIVER_RX_DESCQ_FLS_DONE);
1372 
1373 		EFSYS_PROBE1(rx_descq_fls_done, uint32_t, rxq_index);
1374 
1375 		EFSYS_ASSERT(eecp->eec_rxq_flush_done != NULL);
1376 		should_abort = eecp->eec_rxq_flush_done(arg, rxq_index);
1377 		break;
1378 	}
1379 
1380 	default:
1381 		EFSYS_PROBE3(bad_event, unsigned int, eep->ee_index,
1382 		    uint32_t, EFX_QWORD_FIELD(*eqp, EFX_DWORD_1),
1383 		    uint32_t, EFX_QWORD_FIELD(*eqp, EFX_DWORD_0));
1384 		break;
1385 	}
1386 
1387 	return (should_abort);
1388 }
1389 
1390 		void
1391 ef10_ev_rxlabel_init(
1392 	__in		efx_evq_t *eep,
1393 	__in		efx_rxq_t *erp,
1394 	__in		unsigned int label,
1395 	__in		efx_rxq_type_t type)
1396 {
1397 	efx_evq_rxq_state_t *eersp;
1398 #if EFSYS_OPT_RX_PACKED_STREAM || EFSYS_OPT_RX_ES_SUPER_BUFFER
1399 	boolean_t packed_stream = (type == EFX_RXQ_TYPE_PACKED_STREAM);
1400 	boolean_t es_super_buffer = (type == EFX_RXQ_TYPE_ES_SUPER_BUFFER);
1401 #endif
1402 
1403 	_NOTE(ARGUNUSED(type))
1404 	EFSYS_ASSERT3U(label, <, EFX_ARRAY_SIZE(eep->ee_rxq_state));
1405 	eersp = &eep->ee_rxq_state[label];
1406 
1407 	EFSYS_ASSERT3U(eersp->eers_rx_mask, ==, 0);
1408 
1409 #if EFSYS_OPT_RX_PACKED_STREAM
1410 	/*
1411 	 * For packed stream modes, the very first event will
1412 	 * have a new buffer flag set, so it will be incremented,
1413 	 * yielding the correct pointer. That results in a simpler
1414 	 * code than trying to detect start-of-the-world condition
1415 	 * in the event handler.
1416 	 */
1417 	eersp->eers_rx_read_ptr = packed_stream ? ~0 : 0;
1418 #else
1419 	eersp->eers_rx_read_ptr = 0;
1420 #endif
1421 	eersp->eers_rx_mask = erp->er_mask;
1422 #if EFSYS_OPT_RX_PACKED_STREAM || EFSYS_OPT_RX_ES_SUPER_BUFFER
1423 	eersp->eers_rx_stream_npackets = 0;
1424 	eersp->eers_rx_packed_stream = packed_stream || es_super_buffer;
1425 #endif
1426 #if EFSYS_OPT_RX_PACKED_STREAM
1427 	if (packed_stream) {
1428 		eersp->eers_rx_packed_stream_credits = (eep->ee_mask + 1) /
1429 		    EFX_DIV_ROUND_UP(EFX_RX_PACKED_STREAM_MEM_PER_CREDIT,
1430 		    EFX_RX_PACKED_STREAM_MIN_PACKET_SPACE);
1431 		EFSYS_ASSERT3U(eersp->eers_rx_packed_stream_credits, !=, 0);
1432 		/*
1433 		 * A single credit is allocated to the queue when it is started.
1434 		 * It is immediately spent by the first packet which has NEW
1435 		 * BUFFER flag set, though, but still we shall take into
1436 		 * account, as to not wrap around the maximum number of credits
1437 		 * accidentally
1438 		 */
1439 		eersp->eers_rx_packed_stream_credits--;
1440 		EFSYS_ASSERT3U(eersp->eers_rx_packed_stream_credits, <=,
1441 		    EFX_RX_PACKED_STREAM_MAX_CREDITS);
1442 	}
1443 #endif
1444 }
1445 
1446 		void
1447 ef10_ev_rxlabel_fini(
1448 	__in		efx_evq_t *eep,
1449 	__in		unsigned int label)
1450 {
1451 	efx_evq_rxq_state_t *eersp;
1452 
1453 	EFSYS_ASSERT3U(label, <, EFX_ARRAY_SIZE(eep->ee_rxq_state));
1454 	eersp = &eep->ee_rxq_state[label];
1455 
1456 	EFSYS_ASSERT3U(eersp->eers_rx_mask, !=, 0);
1457 
1458 	eersp->eers_rx_read_ptr = 0;
1459 	eersp->eers_rx_mask = 0;
1460 #if EFSYS_OPT_RX_PACKED_STREAM || EFSYS_OPT_RX_ES_SUPER_BUFFER
1461 	eersp->eers_rx_stream_npackets = 0;
1462 	eersp->eers_rx_packed_stream = B_FALSE;
1463 #endif
1464 #if EFSYS_OPT_RX_PACKED_STREAM
1465 	eersp->eers_rx_packed_stream_credits = 0;
1466 #endif
1467 }
1468 
1469 #endif	/* EFSYS_OPT_HUNTINGTON || EFSYS_OPT_MEDFORD || EFSYS_OPT_MEDFORD2 */
1470