mcdi.c revision 5731d7b35e5b87157a9b9973cc2eff70c50aec58
1/****************************************************************************
2 * Driver for Solarflare network controllers and boards
3 * Copyright 2008-2013 Solarflare Communications Inc.
4 *
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 as published
7 * by the Free Software Foundation, incorporated herein by reference.
8 */
9
10#include <linux/delay.h>
11#include <asm/cmpxchg.h>
12#include "net_driver.h"
13#include "nic.h"
14#include "io.h"
15#include "farch_regs.h"
16#include "mcdi_pcol.h"
17#include "phy.h"
18
19/**************************************************************************
20 *
21 * Management-Controller-to-Driver Interface
22 *
23 **************************************************************************
24 */
25
26#define MCDI_RPC_TIMEOUT       (10 * HZ)
27
28/* A reboot/assertion causes the MCDI status word to be set after the
29 * command word is set or a REBOOT event is sent. If we notice a reboot
30 * via these mechanisms then wait 250ms for the status word to be set.
31 */
32#define MCDI_STATUS_DELAY_US		100
33#define MCDI_STATUS_DELAY_COUNT		2500
34#define MCDI_STATUS_SLEEP_MS						\
35	(MCDI_STATUS_DELAY_US * MCDI_STATUS_DELAY_COUNT / 1000)
36
37#define SEQ_MASK							\
38	EFX_MASK32(EFX_WIDTH(MCDI_HEADER_SEQ))
39
40struct efx_mcdi_async_param {
41	struct list_head list;
42	unsigned int cmd;
43	size_t inlen;
44	size_t outlen;
45	efx_mcdi_async_completer *complete;
46	unsigned long cookie;
47	/* followed by request/response buffer */
48};
49
50static void efx_mcdi_timeout_async(unsigned long context);
51static int efx_mcdi_drv_attach(struct efx_nic *efx, bool driver_operating,
52			       bool *was_attached_out);
53static bool efx_mcdi_poll_once(struct efx_nic *efx);
54
55static inline struct efx_mcdi_iface *efx_mcdi(struct efx_nic *efx)
56{
57	EFX_BUG_ON_PARANOID(!efx->mcdi);
58	return &efx->mcdi->iface;
59}
60
61int efx_mcdi_init(struct efx_nic *efx)
62{
63	struct efx_mcdi_iface *mcdi;
64	bool already_attached;
65	int rc;
66
67	efx->mcdi = kzalloc(sizeof(*efx->mcdi), GFP_KERNEL);
68	if (!efx->mcdi)
69		return -ENOMEM;
70
71	mcdi = efx_mcdi(efx);
72	mcdi->efx = efx;
73	init_waitqueue_head(&mcdi->wq);
74	spin_lock_init(&mcdi->iface_lock);
75	mcdi->state = MCDI_STATE_QUIESCENT;
76	mcdi->mode = MCDI_MODE_POLL;
77	spin_lock_init(&mcdi->async_lock);
78	INIT_LIST_HEAD(&mcdi->async_list);
79	setup_timer(&mcdi->async_timer, efx_mcdi_timeout_async,
80		    (unsigned long)mcdi);
81
82	(void) efx_mcdi_poll_reboot(efx);
83	mcdi->new_epoch = true;
84
85	/* Recover from a failed assertion before probing */
86	rc = efx_mcdi_handle_assertion(efx);
87	if (rc)
88		return rc;
89
90	/* Let the MC (and BMC, if this is a LOM) know that the driver
91	 * is loaded. We should do this before we reset the NIC.
92	 */
93	rc = efx_mcdi_drv_attach(efx, true, &already_attached);
94	if (rc) {
95		netif_err(efx, probe, efx->net_dev,
96			  "Unable to register driver with MCPU\n");
97		return rc;
98	}
99	if (already_attached)
100		/* Not a fatal error */
101		netif_err(efx, probe, efx->net_dev,
102			  "Host already registered with MCPU\n");
103
104	return 0;
105}
106
107void efx_mcdi_fini(struct efx_nic *efx)
108{
109	if (!efx->mcdi)
110		return;
111
112	BUG_ON(efx->mcdi->iface.state != MCDI_STATE_QUIESCENT);
113
114	/* Relinquish the device (back to the BMC, if this is a LOM) */
115	efx_mcdi_drv_attach(efx, false, NULL);
116
117	kfree(efx->mcdi);
118}
119
120static void efx_mcdi_send_request(struct efx_nic *efx, unsigned cmd,
121				  const efx_dword_t *inbuf, size_t inlen)
122{
123	struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
124	efx_dword_t hdr[2];
125	size_t hdr_len;
126	u32 xflags, seqno;
127
128	BUG_ON(mcdi->state == MCDI_STATE_QUIESCENT);
129
130	/* Serialise with efx_mcdi_ev_cpl() and efx_mcdi_ev_death() */
131	spin_lock_bh(&mcdi->iface_lock);
132	++mcdi->seqno;
133	spin_unlock_bh(&mcdi->iface_lock);
134
135	seqno = mcdi->seqno & SEQ_MASK;
136	xflags = 0;
137	if (mcdi->mode == MCDI_MODE_EVENTS)
138		xflags |= MCDI_HEADER_XFLAGS_EVREQ;
139
140	if (efx->type->mcdi_max_ver == 1) {
141		/* MCDI v1 */
142		EFX_POPULATE_DWORD_7(hdr[0],
143				     MCDI_HEADER_RESPONSE, 0,
144				     MCDI_HEADER_RESYNC, 1,
145				     MCDI_HEADER_CODE, cmd,
146				     MCDI_HEADER_DATALEN, inlen,
147				     MCDI_HEADER_SEQ, seqno,
148				     MCDI_HEADER_XFLAGS, xflags,
149				     MCDI_HEADER_NOT_EPOCH, !mcdi->new_epoch);
150		hdr_len = 4;
151	} else {
152		/* MCDI v2 */
153		BUG_ON(inlen > MCDI_CTL_SDU_LEN_MAX_V2);
154		EFX_POPULATE_DWORD_7(hdr[0],
155				     MCDI_HEADER_RESPONSE, 0,
156				     MCDI_HEADER_RESYNC, 1,
157				     MCDI_HEADER_CODE, MC_CMD_V2_EXTN,
158				     MCDI_HEADER_DATALEN, 0,
159				     MCDI_HEADER_SEQ, seqno,
160				     MCDI_HEADER_XFLAGS, xflags,
161				     MCDI_HEADER_NOT_EPOCH, !mcdi->new_epoch);
162		EFX_POPULATE_DWORD_2(hdr[1],
163				     MC_CMD_V2_EXTN_IN_EXTENDED_CMD, cmd,
164				     MC_CMD_V2_EXTN_IN_ACTUAL_LEN, inlen);
165		hdr_len = 8;
166	}
167
168	efx->type->mcdi_request(efx, hdr, hdr_len, inbuf, inlen);
169
170	mcdi->new_epoch = false;
171}
172
173static int efx_mcdi_errno(unsigned int mcdi_err)
174{
175	switch (mcdi_err) {
176	case 0:
177		return 0;
178#define TRANSLATE_ERROR(name)					\
179	case MC_CMD_ERR_ ## name:				\
180		return -name;
181	TRANSLATE_ERROR(EPERM);
182	TRANSLATE_ERROR(ENOENT);
183	TRANSLATE_ERROR(EINTR);
184	TRANSLATE_ERROR(EAGAIN);
185	TRANSLATE_ERROR(EACCES);
186	TRANSLATE_ERROR(EBUSY);
187	TRANSLATE_ERROR(EINVAL);
188	TRANSLATE_ERROR(EDEADLK);
189	TRANSLATE_ERROR(ENOSYS);
190	TRANSLATE_ERROR(ETIME);
191	TRANSLATE_ERROR(EALREADY);
192	TRANSLATE_ERROR(ENOSPC);
193#undef TRANSLATE_ERROR
194	case MC_CMD_ERR_ALLOC_FAIL:
195		return -ENOBUFS;
196	case MC_CMD_ERR_MAC_EXIST:
197		return -EADDRINUSE;
198	default:
199		return -EPROTO;
200	}
201}
202
203static void efx_mcdi_read_response_header(struct efx_nic *efx)
204{
205	struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
206	unsigned int respseq, respcmd, error;
207	efx_dword_t hdr;
208
209	efx->type->mcdi_read_response(efx, &hdr, 0, 4);
210	respseq = EFX_DWORD_FIELD(hdr, MCDI_HEADER_SEQ);
211	respcmd = EFX_DWORD_FIELD(hdr, MCDI_HEADER_CODE);
212	error = EFX_DWORD_FIELD(hdr, MCDI_HEADER_ERROR);
213
214	if (respcmd != MC_CMD_V2_EXTN) {
215		mcdi->resp_hdr_len = 4;
216		mcdi->resp_data_len = EFX_DWORD_FIELD(hdr, MCDI_HEADER_DATALEN);
217	} else {
218		efx->type->mcdi_read_response(efx, &hdr, 4, 4);
219		mcdi->resp_hdr_len = 8;
220		mcdi->resp_data_len =
221			EFX_DWORD_FIELD(hdr, MC_CMD_V2_EXTN_IN_ACTUAL_LEN);
222	}
223
224	if (error && mcdi->resp_data_len == 0) {
225		netif_err(efx, hw, efx->net_dev, "MC rebooted\n");
226		mcdi->resprc = -EIO;
227	} else if ((respseq ^ mcdi->seqno) & SEQ_MASK) {
228		netif_err(efx, hw, efx->net_dev,
229			  "MC response mismatch tx seq 0x%x rx seq 0x%x\n",
230			  respseq, mcdi->seqno);
231		mcdi->resprc = -EIO;
232	} else if (error) {
233		efx->type->mcdi_read_response(efx, &hdr, mcdi->resp_hdr_len, 4);
234		mcdi->resprc =
235			efx_mcdi_errno(EFX_DWORD_FIELD(hdr, EFX_DWORD_0));
236	} else {
237		mcdi->resprc = 0;
238	}
239}
240
241static bool efx_mcdi_poll_once(struct efx_nic *efx)
242{
243	struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
244
245	rmb();
246	if (!efx->type->mcdi_poll_response(efx))
247		return false;
248
249	spin_lock_bh(&mcdi->iface_lock);
250	efx_mcdi_read_response_header(efx);
251	spin_unlock_bh(&mcdi->iface_lock);
252
253	return true;
254}
255
256static int efx_mcdi_poll(struct efx_nic *efx)
257{
258	struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
259	unsigned long time, finish;
260	unsigned int spins;
261	int rc;
262
263	/* Check for a reboot atomically with respect to efx_mcdi_copyout() */
264	rc = efx_mcdi_poll_reboot(efx);
265	if (rc) {
266		spin_lock_bh(&mcdi->iface_lock);
267		mcdi->resprc = rc;
268		mcdi->resp_hdr_len = 0;
269		mcdi->resp_data_len = 0;
270		spin_unlock_bh(&mcdi->iface_lock);
271		return 0;
272	}
273
274	/* Poll for completion. Poll quickly (once a us) for the 1st jiffy,
275	 * because generally mcdi responses are fast. After that, back off
276	 * and poll once a jiffy (approximately)
277	 */
278	spins = TICK_USEC;
279	finish = jiffies + MCDI_RPC_TIMEOUT;
280
281	while (1) {
282		if (spins != 0) {
283			--spins;
284			udelay(1);
285		} else {
286			schedule_timeout_uninterruptible(1);
287		}
288
289		time = jiffies;
290
291		if (efx_mcdi_poll_once(efx))
292			break;
293
294		if (time_after(time, finish))
295			return -ETIMEDOUT;
296	}
297
298	/* Return rc=0 like wait_event_timeout() */
299	return 0;
300}
301
302/* Test and clear MC-rebooted flag for this port/function; reset
303 * software state as necessary.
304 */
305int efx_mcdi_poll_reboot(struct efx_nic *efx)
306{
307	if (!efx->mcdi)
308		return 0;
309
310	return efx->type->mcdi_poll_reboot(efx);
311}
312
313static bool efx_mcdi_acquire_async(struct efx_mcdi_iface *mcdi)
314{
315	return cmpxchg(&mcdi->state,
316		       MCDI_STATE_QUIESCENT, MCDI_STATE_RUNNING_ASYNC) ==
317		MCDI_STATE_QUIESCENT;
318}
319
320static void efx_mcdi_acquire_sync(struct efx_mcdi_iface *mcdi)
321{
322	/* Wait until the interface becomes QUIESCENT and we win the race
323	 * to mark it RUNNING_SYNC.
324	 */
325	wait_event(mcdi->wq,
326		   cmpxchg(&mcdi->state,
327			   MCDI_STATE_QUIESCENT, MCDI_STATE_RUNNING_SYNC) ==
328		   MCDI_STATE_QUIESCENT);
329}
330
331static int efx_mcdi_await_completion(struct efx_nic *efx)
332{
333	struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
334
335	if (wait_event_timeout(mcdi->wq, mcdi->state == MCDI_STATE_COMPLETED,
336			       MCDI_RPC_TIMEOUT) == 0)
337		return -ETIMEDOUT;
338
339	/* Check if efx_mcdi_set_mode() switched us back to polled completions.
340	 * In which case, poll for completions directly. If efx_mcdi_ev_cpl()
341	 * completed the request first, then we'll just end up completing the
342	 * request again, which is safe.
343	 *
344	 * We need an smp_rmb() to synchronise with efx_mcdi_mode_poll(), which
345	 * wait_event_timeout() implicitly provides.
346	 */
347	if (mcdi->mode == MCDI_MODE_POLL)
348		return efx_mcdi_poll(efx);
349
350	return 0;
351}
352
353/* If the interface is RUNNING_SYNC, switch to COMPLETED and wake the
354 * requester.  Return whether this was done.  Does not take any locks.
355 */
356static bool efx_mcdi_complete_sync(struct efx_mcdi_iface *mcdi)
357{
358	if (cmpxchg(&mcdi->state,
359		    MCDI_STATE_RUNNING_SYNC, MCDI_STATE_COMPLETED) ==
360	    MCDI_STATE_RUNNING_SYNC) {
361		wake_up(&mcdi->wq);
362		return true;
363	}
364
365	return false;
366}
367
368static void efx_mcdi_release(struct efx_mcdi_iface *mcdi)
369{
370	if (mcdi->mode == MCDI_MODE_EVENTS) {
371		struct efx_mcdi_async_param *async;
372		struct efx_nic *efx = mcdi->efx;
373
374		/* Process the asynchronous request queue */
375		spin_lock_bh(&mcdi->async_lock);
376		async = list_first_entry_or_null(
377			&mcdi->async_list, struct efx_mcdi_async_param, list);
378		if (async) {
379			mcdi->state = MCDI_STATE_RUNNING_ASYNC;
380			efx_mcdi_send_request(efx, async->cmd,
381					      (const efx_dword_t *)(async + 1),
382					      async->inlen);
383			mod_timer(&mcdi->async_timer,
384				  jiffies + MCDI_RPC_TIMEOUT);
385		}
386		spin_unlock_bh(&mcdi->async_lock);
387
388		if (async)
389			return;
390	}
391
392	mcdi->state = MCDI_STATE_QUIESCENT;
393	wake_up(&mcdi->wq);
394}
395
396/* If the interface is RUNNING_ASYNC, switch to COMPLETED, call the
397 * asynchronous completion function, and release the interface.
398 * Return whether this was done.  Must be called in bh-disabled
399 * context.  Will take iface_lock and async_lock.
400 */
401static bool efx_mcdi_complete_async(struct efx_mcdi_iface *mcdi, bool timeout)
402{
403	struct efx_nic *efx = mcdi->efx;
404	struct efx_mcdi_async_param *async;
405	size_t hdr_len, data_len;
406	efx_dword_t *outbuf;
407	int rc;
408
409	if (cmpxchg(&mcdi->state,
410		    MCDI_STATE_RUNNING_ASYNC, MCDI_STATE_COMPLETED) !=
411	    MCDI_STATE_RUNNING_ASYNC)
412		return false;
413
414	spin_lock(&mcdi->iface_lock);
415	if (timeout) {
416		/* Ensure that if the completion event arrives later,
417		 * the seqno check in efx_mcdi_ev_cpl() will fail
418		 */
419		++mcdi->seqno;
420		++mcdi->credits;
421		rc = -ETIMEDOUT;
422		hdr_len = 0;
423		data_len = 0;
424	} else {
425		rc = mcdi->resprc;
426		hdr_len = mcdi->resp_hdr_len;
427		data_len = mcdi->resp_data_len;
428	}
429	spin_unlock(&mcdi->iface_lock);
430
431	/* Stop the timer.  In case the timer function is running, we
432	 * must wait for it to return so that there is no possibility
433	 * of it aborting the next request.
434	 */
435	if (!timeout)
436		del_timer_sync(&mcdi->async_timer);
437
438	spin_lock(&mcdi->async_lock);
439	async = list_first_entry(&mcdi->async_list,
440				 struct efx_mcdi_async_param, list);
441	list_del(&async->list);
442	spin_unlock(&mcdi->async_lock);
443
444	outbuf = (efx_dword_t *)(async + 1);
445	efx->type->mcdi_read_response(efx, outbuf, hdr_len,
446				      min(async->outlen, data_len));
447	async->complete(efx, async->cookie, rc, outbuf, data_len);
448	kfree(async);
449
450	efx_mcdi_release(mcdi);
451
452	return true;
453}
454
455static void efx_mcdi_ev_cpl(struct efx_nic *efx, unsigned int seqno,
456			    unsigned int datalen, unsigned int mcdi_err)
457{
458	struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
459	bool wake = false;
460
461	spin_lock(&mcdi->iface_lock);
462
463	if ((seqno ^ mcdi->seqno) & SEQ_MASK) {
464		if (mcdi->credits)
465			/* The request has been cancelled */
466			--mcdi->credits;
467		else
468			netif_err(efx, hw, efx->net_dev,
469				  "MC response mismatch tx seq 0x%x rx "
470				  "seq 0x%x\n", seqno, mcdi->seqno);
471	} else {
472		if (efx->type->mcdi_max_ver >= 2) {
473			/* MCDI v2 responses don't fit in an event */
474			efx_mcdi_read_response_header(efx);
475		} else {
476			mcdi->resprc = efx_mcdi_errno(mcdi_err);
477			mcdi->resp_hdr_len = 4;
478			mcdi->resp_data_len = datalen;
479		}
480
481		wake = true;
482	}
483
484	spin_unlock(&mcdi->iface_lock);
485
486	if (wake) {
487		if (!efx_mcdi_complete_async(mcdi, false))
488			(void) efx_mcdi_complete_sync(mcdi);
489
490		/* If the interface isn't RUNNING_ASYNC or
491		 * RUNNING_SYNC then we've received a duplicate
492		 * completion after we've already transitioned back to
493		 * QUIESCENT. [A subsequent invocation would increment
494		 * seqno, so would have failed the seqno check].
495		 */
496	}
497}
498
499static void efx_mcdi_timeout_async(unsigned long context)
500{
501	struct efx_mcdi_iface *mcdi = (struct efx_mcdi_iface *)context;
502
503	efx_mcdi_complete_async(mcdi, true);
504}
505
506static int
507efx_mcdi_check_supported(struct efx_nic *efx, unsigned int cmd, size_t inlen)
508{
509	if (efx->type->mcdi_max_ver < 0 ||
510	     (efx->type->mcdi_max_ver < 2 &&
511	      cmd > MC_CMD_CMD_SPACE_ESCAPE_7))
512		return -EINVAL;
513
514	if (inlen > MCDI_CTL_SDU_LEN_MAX_V2 ||
515	    (efx->type->mcdi_max_ver < 2 &&
516	     inlen > MCDI_CTL_SDU_LEN_MAX_V1))
517		return -EMSGSIZE;
518
519	return 0;
520}
521
522int efx_mcdi_rpc(struct efx_nic *efx, unsigned cmd,
523		 const efx_dword_t *inbuf, size_t inlen,
524		 efx_dword_t *outbuf, size_t outlen,
525		 size_t *outlen_actual)
526{
527	int rc;
528
529	rc = efx_mcdi_rpc_start(efx, cmd, inbuf, inlen);
530	if (rc)
531		return rc;
532	return efx_mcdi_rpc_finish(efx, cmd, inlen,
533				   outbuf, outlen, outlen_actual);
534}
535
536int efx_mcdi_rpc_start(struct efx_nic *efx, unsigned cmd,
537		       const efx_dword_t *inbuf, size_t inlen)
538{
539	struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
540	int rc;
541
542	rc = efx_mcdi_check_supported(efx, cmd, inlen);
543	if (rc)
544		return rc;
545
546	efx_mcdi_acquire_sync(mcdi);
547	efx_mcdi_send_request(efx, cmd, inbuf, inlen);
548	return 0;
549}
550
551/**
552 * efx_mcdi_rpc_async - Schedule an MCDI command to run asynchronously
553 * @efx: NIC through which to issue the command
554 * @cmd: Command type number
555 * @inbuf: Command parameters
556 * @inlen: Length of command parameters, in bytes
557 * @outlen: Length to allocate for response buffer, in bytes
558 * @complete: Function to be called on completion or cancellation.
559 * @cookie: Arbitrary value to be passed to @complete.
560 *
561 * This function does not sleep and therefore may be called in atomic
562 * context.  It will fail if event queues are disabled or if MCDI
563 * event completions have been disabled due to an error.
564 *
565 * If it succeeds, the @complete function will be called exactly once
566 * in atomic context, when one of the following occurs:
567 * (a) the completion event is received (in NAPI context)
568 * (b) event queues are disabled (in the process that disables them)
569 * (c) the request times-out (in timer context)
570 */
571int
572efx_mcdi_rpc_async(struct efx_nic *efx, unsigned int cmd,
573		   const efx_dword_t *inbuf, size_t inlen, size_t outlen,
574		   efx_mcdi_async_completer *complete, unsigned long cookie)
575{
576	struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
577	struct efx_mcdi_async_param *async;
578	int rc;
579
580	rc = efx_mcdi_check_supported(efx, cmd, inlen);
581	if (rc)
582		return rc;
583
584	async = kmalloc(sizeof(*async) + ALIGN(max(inlen, outlen), 4),
585			GFP_ATOMIC);
586	if (!async)
587		return -ENOMEM;
588
589	async->cmd = cmd;
590	async->inlen = inlen;
591	async->outlen = outlen;
592	async->complete = complete;
593	async->cookie = cookie;
594	memcpy(async + 1, inbuf, inlen);
595
596	spin_lock_bh(&mcdi->async_lock);
597
598	if (mcdi->mode == MCDI_MODE_EVENTS) {
599		list_add_tail(&async->list, &mcdi->async_list);
600
601		/* If this is at the front of the queue, try to start it
602		 * immediately
603		 */
604		if (mcdi->async_list.next == &async->list &&
605		    efx_mcdi_acquire_async(mcdi)) {
606			efx_mcdi_send_request(efx, cmd, inbuf, inlen);
607			mod_timer(&mcdi->async_timer,
608				  jiffies + MCDI_RPC_TIMEOUT);
609		}
610	} else {
611		kfree(async);
612		rc = -ENETDOWN;
613	}
614
615	spin_unlock_bh(&mcdi->async_lock);
616
617	return rc;
618}
619
620int efx_mcdi_rpc_finish(struct efx_nic *efx, unsigned cmd, size_t inlen,
621			efx_dword_t *outbuf, size_t outlen,
622			size_t *outlen_actual)
623{
624	struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
625	int rc;
626
627	if (mcdi->mode == MCDI_MODE_POLL)
628		rc = efx_mcdi_poll(efx);
629	else
630		rc = efx_mcdi_await_completion(efx);
631
632	if (rc != 0) {
633		/* Close the race with efx_mcdi_ev_cpl() executing just too late
634		 * and completing a request we've just cancelled, by ensuring
635		 * that the seqno check therein fails.
636		 */
637		spin_lock_bh(&mcdi->iface_lock);
638		++mcdi->seqno;
639		++mcdi->credits;
640		spin_unlock_bh(&mcdi->iface_lock);
641
642		netif_err(efx, hw, efx->net_dev,
643			  "MC command 0x%x inlen %d mode %d timed out\n",
644			  cmd, (int)inlen, mcdi->mode);
645	} else {
646		size_t hdr_len, data_len;
647
648		/* At the very least we need a memory barrier here to ensure
649		 * we pick up changes from efx_mcdi_ev_cpl(). Protect against
650		 * a spurious efx_mcdi_ev_cpl() running concurrently by
651		 * acquiring the iface_lock. */
652		spin_lock_bh(&mcdi->iface_lock);
653		rc = mcdi->resprc;
654		hdr_len = mcdi->resp_hdr_len;
655		data_len = mcdi->resp_data_len;
656		spin_unlock_bh(&mcdi->iface_lock);
657
658		BUG_ON(rc > 0);
659
660		if (rc == 0) {
661			efx->type->mcdi_read_response(efx, outbuf, hdr_len,
662						      min(outlen, data_len));
663			if (outlen_actual != NULL)
664				*outlen_actual = data_len;
665		} else if (cmd == MC_CMD_REBOOT && rc == -EIO)
666			; /* Don't reset if MC_CMD_REBOOT returns EIO */
667		else if (rc == -EIO || rc == -EINTR) {
668			netif_err(efx, hw, efx->net_dev, "MC fatal error %d\n",
669				  -rc);
670			efx_schedule_reset(efx, RESET_TYPE_MC_FAILURE);
671		} else
672			netif_dbg(efx, hw, efx->net_dev,
673				  "MC command 0x%x inlen %d failed rc=%d\n",
674				  cmd, (int)inlen, -rc);
675
676		if (rc == -EIO || rc == -EINTR) {
677			msleep(MCDI_STATUS_SLEEP_MS);
678			efx_mcdi_poll_reboot(efx);
679			mcdi->new_epoch = true;
680		}
681	}
682
683	efx_mcdi_release(mcdi);
684	return rc;
685}
686
687/* Switch to polled MCDI completions.  This can be called in various
688 * error conditions with various locks held, so it must be lockless.
689 * Caller is responsible for flushing asynchronous requests later.
690 */
691void efx_mcdi_mode_poll(struct efx_nic *efx)
692{
693	struct efx_mcdi_iface *mcdi;
694
695	if (!efx->mcdi)
696		return;
697
698	mcdi = efx_mcdi(efx);
699	if (mcdi->mode == MCDI_MODE_POLL)
700		return;
701
702	/* We can switch from event completion to polled completion, because
703	 * mcdi requests are always completed in shared memory. We do this by
704	 * switching the mode to POLL'd then completing the request.
705	 * efx_mcdi_await_completion() will then call efx_mcdi_poll().
706	 *
707	 * We need an smp_wmb() to synchronise with efx_mcdi_await_completion(),
708	 * which efx_mcdi_complete_sync() provides for us.
709	 */
710	mcdi->mode = MCDI_MODE_POLL;
711
712	efx_mcdi_complete_sync(mcdi);
713}
714
715/* Flush any running or queued asynchronous requests, after event processing
716 * is stopped
717 */
718void efx_mcdi_flush_async(struct efx_nic *efx)
719{
720	struct efx_mcdi_async_param *async, *next;
721	struct efx_mcdi_iface *mcdi;
722
723	if (!efx->mcdi)
724		return;
725
726	mcdi = efx_mcdi(efx);
727
728	/* We must be in polling mode so no more requests can be queued */
729	BUG_ON(mcdi->mode != MCDI_MODE_POLL);
730
731	del_timer_sync(&mcdi->async_timer);
732
733	/* If a request is still running, make sure we give the MC
734	 * time to complete it so that the response won't overwrite our
735	 * next request.
736	 */
737	if (mcdi->state == MCDI_STATE_RUNNING_ASYNC) {
738		efx_mcdi_poll(efx);
739		mcdi->state = MCDI_STATE_QUIESCENT;
740	}
741
742	/* Nothing else will access the async list now, so it is safe
743	 * to walk it without holding async_lock.  If we hold it while
744	 * calling a completer then lockdep may warn that we have
745	 * acquired locks in the wrong order.
746	 */
747	list_for_each_entry_safe(async, next, &mcdi->async_list, list) {
748		async->complete(efx, async->cookie, -ENETDOWN, NULL, 0);
749		list_del(&async->list);
750		kfree(async);
751	}
752}
753
754void efx_mcdi_mode_event(struct efx_nic *efx)
755{
756	struct efx_mcdi_iface *mcdi;
757
758	if (!efx->mcdi)
759		return;
760
761	mcdi = efx_mcdi(efx);
762
763	if (mcdi->mode == MCDI_MODE_EVENTS)
764		return;
765
766	/* We can't switch from polled to event completion in the middle of a
767	 * request, because the completion method is specified in the request.
768	 * So acquire the interface to serialise the requestors. We don't need
769	 * to acquire the iface_lock to change the mode here, but we do need a
770	 * write memory barrier ensure that efx_mcdi_rpc() sees it, which
771	 * efx_mcdi_acquire() provides.
772	 */
773	efx_mcdi_acquire_sync(mcdi);
774	mcdi->mode = MCDI_MODE_EVENTS;
775	efx_mcdi_release(mcdi);
776}
777
778static void efx_mcdi_ev_death(struct efx_nic *efx, int rc)
779{
780	struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
781
782	/* If there is an outstanding MCDI request, it has been terminated
783	 * either by a BADASSERT or REBOOT event. If the mcdi interface is
784	 * in polled mode, then do nothing because the MC reboot handler will
785	 * set the header correctly. However, if the mcdi interface is waiting
786	 * for a CMDDONE event it won't receive it [and since all MCDI events
787	 * are sent to the same queue, we can't be racing with
788	 * efx_mcdi_ev_cpl()]
789	 *
790	 * If there is an outstanding asynchronous request, we can't
791	 * complete it now (efx_mcdi_complete() would deadlock).  The
792	 * reset process will take care of this.
793	 *
794	 * There's a race here with efx_mcdi_send_request(), because
795	 * we might receive a REBOOT event *before* the request has
796	 * been copied out. In polled mode (during startup) this is
797	 * irrelevant, because efx_mcdi_complete_sync() is ignored. In
798	 * event mode, this condition is just an edge-case of
799	 * receiving a REBOOT event after posting the MCDI
800	 * request. Did the mc reboot before or after the copyout? The
801	 * best we can do always is just return failure.
802	 */
803	spin_lock(&mcdi->iface_lock);
804	if (efx_mcdi_complete_sync(mcdi)) {
805		if (mcdi->mode == MCDI_MODE_EVENTS) {
806			mcdi->resprc = rc;
807			mcdi->resp_hdr_len = 0;
808			mcdi->resp_data_len = 0;
809			++mcdi->credits;
810		}
811	} else {
812		int count;
813
814		/* Consume the status word since efx_mcdi_rpc_finish() won't */
815		for (count = 0; count < MCDI_STATUS_DELAY_COUNT; ++count) {
816			if (efx_mcdi_poll_reboot(efx))
817				break;
818			udelay(MCDI_STATUS_DELAY_US);
819		}
820		mcdi->new_epoch = true;
821
822		/* Nobody was waiting for an MCDI request, so trigger a reset */
823		efx_schedule_reset(efx, RESET_TYPE_MC_FAILURE);
824	}
825
826	spin_unlock(&mcdi->iface_lock);
827}
828
829/* Called from  falcon_process_eventq for MCDI events */
830void efx_mcdi_process_event(struct efx_channel *channel,
831			    efx_qword_t *event)
832{
833	struct efx_nic *efx = channel->efx;
834	int code = EFX_QWORD_FIELD(*event, MCDI_EVENT_CODE);
835	u32 data = EFX_QWORD_FIELD(*event, MCDI_EVENT_DATA);
836
837	switch (code) {
838	case MCDI_EVENT_CODE_BADSSERT:
839		netif_err(efx, hw, efx->net_dev,
840			  "MC watchdog or assertion failure at 0x%x\n", data);
841		efx_mcdi_ev_death(efx, -EINTR);
842		break;
843
844	case MCDI_EVENT_CODE_PMNOTICE:
845		netif_info(efx, wol, efx->net_dev, "MCDI PM event.\n");
846		break;
847
848	case MCDI_EVENT_CODE_CMDDONE:
849		efx_mcdi_ev_cpl(efx,
850				MCDI_EVENT_FIELD(*event, CMDDONE_SEQ),
851				MCDI_EVENT_FIELD(*event, CMDDONE_DATALEN),
852				MCDI_EVENT_FIELD(*event, CMDDONE_ERRNO));
853		break;
854
855	case MCDI_EVENT_CODE_LINKCHANGE:
856		efx_mcdi_process_link_change(efx, event);
857		break;
858	case MCDI_EVENT_CODE_SENSOREVT:
859		efx_mcdi_sensor_event(efx, event);
860		break;
861	case MCDI_EVENT_CODE_SCHEDERR:
862		netif_info(efx, hw, efx->net_dev,
863			   "MC Scheduler error address=0x%x\n", data);
864		break;
865	case MCDI_EVENT_CODE_REBOOT:
866	case MCDI_EVENT_CODE_MC_REBOOT:
867		netif_info(efx, hw, efx->net_dev, "MC Reboot\n");
868		efx_mcdi_ev_death(efx, -EIO);
869		break;
870	case MCDI_EVENT_CODE_MAC_STATS_DMA:
871		/* MAC stats are gather lazily.  We can ignore this. */
872		break;
873	case MCDI_EVENT_CODE_FLR:
874		efx_sriov_flr(efx, MCDI_EVENT_FIELD(*event, FLR_VF));
875		break;
876	case MCDI_EVENT_CODE_PTP_RX:
877	case MCDI_EVENT_CODE_PTP_FAULT:
878	case MCDI_EVENT_CODE_PTP_PPS:
879		efx_ptp_event(efx, event);
880		break;
881	case MCDI_EVENT_CODE_TX_FLUSH:
882	case MCDI_EVENT_CODE_RX_FLUSH:
883		/* Two flush events will be sent: one to the same event
884		 * queue as completions, and one to event queue 0.
885		 * In the latter case the {RX,TX}_FLUSH_TO_DRIVER
886		 * flag will be set, and we should ignore the event
887		 * because we want to wait for all completions.
888		 */
889		BUILD_BUG_ON(MCDI_EVENT_TX_FLUSH_TO_DRIVER_LBN !=
890			     MCDI_EVENT_RX_FLUSH_TO_DRIVER_LBN);
891		if (!MCDI_EVENT_FIELD(*event, TX_FLUSH_TO_DRIVER))
892			efx_ef10_handle_drain_event(efx);
893		break;
894	case MCDI_EVENT_CODE_TX_ERR:
895	case MCDI_EVENT_CODE_RX_ERR:
896		netif_err(efx, hw, efx->net_dev,
897			  "%s DMA error (event: "EFX_QWORD_FMT")\n",
898			  code == MCDI_EVENT_CODE_TX_ERR ? "TX" : "RX",
899			  EFX_QWORD_VAL(*event));
900		efx_schedule_reset(efx, RESET_TYPE_DMA_ERROR);
901		break;
902	default:
903		netif_err(efx, hw, efx->net_dev, "Unknown MCDI event 0x%x\n",
904			  code);
905	}
906}
907
908/**************************************************************************
909 *
910 * Specific request functions
911 *
912 **************************************************************************
913 */
914
915void efx_mcdi_print_fwver(struct efx_nic *efx, char *buf, size_t len)
916{
917	MCDI_DECLARE_BUF(outbuf,
918			 max(MC_CMD_GET_VERSION_OUT_LEN,
919			     MC_CMD_GET_CAPABILITIES_OUT_LEN));
920	size_t outlength;
921	const __le16 *ver_words;
922	size_t offset;
923	int rc;
924
925	BUILD_BUG_ON(MC_CMD_GET_VERSION_IN_LEN != 0);
926	rc = efx_mcdi_rpc(efx, MC_CMD_GET_VERSION, NULL, 0,
927			  outbuf, sizeof(outbuf), &outlength);
928	if (rc)
929		goto fail;
930	if (outlength < MC_CMD_GET_VERSION_OUT_LEN) {
931		rc = -EIO;
932		goto fail;
933	}
934
935	ver_words = (__le16 *)MCDI_PTR(outbuf, GET_VERSION_OUT_VERSION);
936	offset = snprintf(buf, len, "%u.%u.%u.%u",
937			  le16_to_cpu(ver_words[0]), le16_to_cpu(ver_words[1]),
938			  le16_to_cpu(ver_words[2]), le16_to_cpu(ver_words[3]));
939
940	/* EF10 may have multiple datapath firmware variants within a
941	 * single version.  Report which variants are running.
942	 */
943	if (efx_nic_rev(efx) >= EFX_REV_HUNT_A0) {
944		BUILD_BUG_ON(MC_CMD_GET_CAPABILITIES_IN_LEN != 0);
945		rc = efx_mcdi_rpc(efx, MC_CMD_GET_CAPABILITIES, NULL, 0,
946				  outbuf, sizeof(outbuf), &outlength);
947		if (rc || outlength < MC_CMD_GET_CAPABILITIES_OUT_LEN)
948			offset += snprintf(
949				buf + offset, len - offset, " rx? tx?");
950		else
951			offset += snprintf(
952				buf + offset, len - offset, " rx%x tx%x",
953				MCDI_WORD(outbuf,
954					  GET_CAPABILITIES_OUT_RX_DPCPU_FW_ID),
955				MCDI_WORD(outbuf,
956					  GET_CAPABILITIES_OUT_TX_DPCPU_FW_ID));
957
958		/* It's theoretically possible for the string to exceed 31
959		 * characters, though in practice the first three version
960		 * components are short enough that this doesn't happen.
961		 */
962		if (WARN_ON(offset >= len))
963			buf[0] = 0;
964	}
965
966	return;
967
968fail:
969	netif_err(efx, probe, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
970	buf[0] = 0;
971}
972
973static int efx_mcdi_drv_attach(struct efx_nic *efx, bool driver_operating,
974			       bool *was_attached)
975{
976	MCDI_DECLARE_BUF(inbuf, MC_CMD_DRV_ATTACH_IN_LEN);
977	MCDI_DECLARE_BUF(outbuf, MC_CMD_DRV_ATTACH_EXT_OUT_LEN);
978	size_t outlen;
979	int rc;
980
981	MCDI_SET_DWORD(inbuf, DRV_ATTACH_IN_NEW_STATE,
982		       driver_operating ? 1 : 0);
983	MCDI_SET_DWORD(inbuf, DRV_ATTACH_IN_UPDATE, 1);
984	MCDI_SET_DWORD(inbuf, DRV_ATTACH_IN_FIRMWARE_ID, MC_CMD_FW_LOW_LATENCY);
985
986	rc = efx_mcdi_rpc(efx, MC_CMD_DRV_ATTACH, inbuf, sizeof(inbuf),
987			  outbuf, sizeof(outbuf), &outlen);
988	if (rc)
989		goto fail;
990	if (outlen < MC_CMD_DRV_ATTACH_OUT_LEN) {
991		rc = -EIO;
992		goto fail;
993	}
994
995	/* We currently assume we have control of the external link
996	 * and are completely trusted by firmware.  Abort probing
997	 * if that's not true for this function.
998	 */
999	if (driver_operating &&
1000	    outlen >= MC_CMD_DRV_ATTACH_EXT_OUT_LEN &&
1001	    (MCDI_DWORD(outbuf, DRV_ATTACH_EXT_OUT_FUNC_FLAGS) &
1002	     (1 << MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_LINKCTRL |
1003	      1 << MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_TRUSTED)) !=
1004	    (1 << MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_LINKCTRL |
1005	     1 << MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_TRUSTED)) {
1006		netif_err(efx, probe, efx->net_dev,
1007			  "This driver version only supports one function per port\n");
1008		return -ENODEV;
1009	}
1010
1011	if (was_attached != NULL)
1012		*was_attached = MCDI_DWORD(outbuf, DRV_ATTACH_OUT_OLD_STATE);
1013	return 0;
1014
1015fail:
1016	netif_err(efx, probe, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
1017	return rc;
1018}
1019
1020int efx_mcdi_get_board_cfg(struct efx_nic *efx, u8 *mac_address,
1021			   u16 *fw_subtype_list, u32 *capabilities)
1022{
1023	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_BOARD_CFG_OUT_LENMAX);
1024	size_t outlen, i;
1025	int port_num = efx_port_num(efx);
1026	int rc;
1027
1028	BUILD_BUG_ON(MC_CMD_GET_BOARD_CFG_IN_LEN != 0);
1029
1030	rc = efx_mcdi_rpc(efx, MC_CMD_GET_BOARD_CFG, NULL, 0,
1031			  outbuf, sizeof(outbuf), &outlen);
1032	if (rc)
1033		goto fail;
1034
1035	if (outlen < MC_CMD_GET_BOARD_CFG_OUT_LENMIN) {
1036		rc = -EIO;
1037		goto fail;
1038	}
1039
1040	if (mac_address)
1041		memcpy(mac_address,
1042		       port_num ?
1043		       MCDI_PTR(outbuf, GET_BOARD_CFG_OUT_MAC_ADDR_BASE_PORT1) :
1044		       MCDI_PTR(outbuf, GET_BOARD_CFG_OUT_MAC_ADDR_BASE_PORT0),
1045		       ETH_ALEN);
1046	if (fw_subtype_list) {
1047		for (i = 0;
1048		     i < MCDI_VAR_ARRAY_LEN(outlen,
1049					    GET_BOARD_CFG_OUT_FW_SUBTYPE_LIST);
1050		     i++)
1051			fw_subtype_list[i] = MCDI_ARRAY_WORD(
1052				outbuf, GET_BOARD_CFG_OUT_FW_SUBTYPE_LIST, i);
1053		for (; i < MC_CMD_GET_BOARD_CFG_OUT_FW_SUBTYPE_LIST_MAXNUM; i++)
1054			fw_subtype_list[i] = 0;
1055	}
1056	if (capabilities) {
1057		if (port_num)
1058			*capabilities = MCDI_DWORD(outbuf,
1059					GET_BOARD_CFG_OUT_CAPABILITIES_PORT1);
1060		else
1061			*capabilities = MCDI_DWORD(outbuf,
1062					GET_BOARD_CFG_OUT_CAPABILITIES_PORT0);
1063	}
1064
1065	return 0;
1066
1067fail:
1068	netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d len=%d\n",
1069		  __func__, rc, (int)outlen);
1070
1071	return rc;
1072}
1073
1074int efx_mcdi_log_ctrl(struct efx_nic *efx, bool evq, bool uart, u32 dest_evq)
1075{
1076	MCDI_DECLARE_BUF(inbuf, MC_CMD_LOG_CTRL_IN_LEN);
1077	u32 dest = 0;
1078	int rc;
1079
1080	if (uart)
1081		dest |= MC_CMD_LOG_CTRL_IN_LOG_DEST_UART;
1082	if (evq)
1083		dest |= MC_CMD_LOG_CTRL_IN_LOG_DEST_EVQ;
1084
1085	MCDI_SET_DWORD(inbuf, LOG_CTRL_IN_LOG_DEST, dest);
1086	MCDI_SET_DWORD(inbuf, LOG_CTRL_IN_LOG_DEST_EVQ, dest_evq);
1087
1088	BUILD_BUG_ON(MC_CMD_LOG_CTRL_OUT_LEN != 0);
1089
1090	rc = efx_mcdi_rpc(efx, MC_CMD_LOG_CTRL, inbuf, sizeof(inbuf),
1091			  NULL, 0, NULL);
1092	if (rc)
1093		goto fail;
1094
1095	return 0;
1096
1097fail:
1098	netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
1099	return rc;
1100}
1101
1102int efx_mcdi_nvram_types(struct efx_nic *efx, u32 *nvram_types_out)
1103{
1104	MCDI_DECLARE_BUF(outbuf, MC_CMD_NVRAM_TYPES_OUT_LEN);
1105	size_t outlen;
1106	int rc;
1107
1108	BUILD_BUG_ON(MC_CMD_NVRAM_TYPES_IN_LEN != 0);
1109
1110	rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_TYPES, NULL, 0,
1111			  outbuf, sizeof(outbuf), &outlen);
1112	if (rc)
1113		goto fail;
1114	if (outlen < MC_CMD_NVRAM_TYPES_OUT_LEN) {
1115		rc = -EIO;
1116		goto fail;
1117	}
1118
1119	*nvram_types_out = MCDI_DWORD(outbuf, NVRAM_TYPES_OUT_TYPES);
1120	return 0;
1121
1122fail:
1123	netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n",
1124		  __func__, rc);
1125	return rc;
1126}
1127
1128int efx_mcdi_nvram_info(struct efx_nic *efx, unsigned int type,
1129			size_t *size_out, size_t *erase_size_out,
1130			bool *protected_out)
1131{
1132	MCDI_DECLARE_BUF(inbuf, MC_CMD_NVRAM_INFO_IN_LEN);
1133	MCDI_DECLARE_BUF(outbuf, MC_CMD_NVRAM_INFO_OUT_LEN);
1134	size_t outlen;
1135	int rc;
1136
1137	MCDI_SET_DWORD(inbuf, NVRAM_INFO_IN_TYPE, type);
1138
1139	rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_INFO, inbuf, sizeof(inbuf),
1140			  outbuf, sizeof(outbuf), &outlen);
1141	if (rc)
1142		goto fail;
1143	if (outlen < MC_CMD_NVRAM_INFO_OUT_LEN) {
1144		rc = -EIO;
1145		goto fail;
1146	}
1147
1148	*size_out = MCDI_DWORD(outbuf, NVRAM_INFO_OUT_SIZE);
1149	*erase_size_out = MCDI_DWORD(outbuf, NVRAM_INFO_OUT_ERASESIZE);
1150	*protected_out = !!(MCDI_DWORD(outbuf, NVRAM_INFO_OUT_FLAGS) &
1151				(1 << MC_CMD_NVRAM_INFO_OUT_PROTECTED_LBN));
1152	return 0;
1153
1154fail:
1155	netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
1156	return rc;
1157}
1158
1159static int efx_mcdi_nvram_test(struct efx_nic *efx, unsigned int type)
1160{
1161	MCDI_DECLARE_BUF(inbuf, MC_CMD_NVRAM_TEST_IN_LEN);
1162	MCDI_DECLARE_BUF(outbuf, MC_CMD_NVRAM_TEST_OUT_LEN);
1163	int rc;
1164
1165	MCDI_SET_DWORD(inbuf, NVRAM_TEST_IN_TYPE, type);
1166
1167	rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_TEST, inbuf, sizeof(inbuf),
1168			  outbuf, sizeof(outbuf), NULL);
1169	if (rc)
1170		return rc;
1171
1172	switch (MCDI_DWORD(outbuf, NVRAM_TEST_OUT_RESULT)) {
1173	case MC_CMD_NVRAM_TEST_PASS:
1174	case MC_CMD_NVRAM_TEST_NOTSUPP:
1175		return 0;
1176	default:
1177		return -EIO;
1178	}
1179}
1180
1181int efx_mcdi_nvram_test_all(struct efx_nic *efx)
1182{
1183	u32 nvram_types;
1184	unsigned int type;
1185	int rc;
1186
1187	rc = efx_mcdi_nvram_types(efx, &nvram_types);
1188	if (rc)
1189		goto fail1;
1190
1191	type = 0;
1192	while (nvram_types != 0) {
1193		if (nvram_types & 1) {
1194			rc = efx_mcdi_nvram_test(efx, type);
1195			if (rc)
1196				goto fail2;
1197		}
1198		type++;
1199		nvram_types >>= 1;
1200	}
1201
1202	return 0;
1203
1204fail2:
1205	netif_err(efx, hw, efx->net_dev, "%s: failed type=%u\n",
1206		  __func__, type);
1207fail1:
1208	netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
1209	return rc;
1210}
1211
1212static int efx_mcdi_read_assertion(struct efx_nic *efx)
1213{
1214	MCDI_DECLARE_BUF(inbuf, MC_CMD_GET_ASSERTS_IN_LEN);
1215	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_ASSERTS_OUT_LEN);
1216	unsigned int flags, index;
1217	const char *reason;
1218	size_t outlen;
1219	int retry;
1220	int rc;
1221
1222	/* Attempt to read any stored assertion state before we reboot
1223	 * the mcfw out of the assertion handler. Retry twice, once
1224	 * because a boot-time assertion might cause this command to fail
1225	 * with EINTR. And once again because GET_ASSERTS can race with
1226	 * MC_CMD_REBOOT running on the other port. */
1227	retry = 2;
1228	do {
1229		MCDI_SET_DWORD(inbuf, GET_ASSERTS_IN_CLEAR, 1);
1230		rc = efx_mcdi_rpc(efx, MC_CMD_GET_ASSERTS,
1231				  inbuf, MC_CMD_GET_ASSERTS_IN_LEN,
1232				  outbuf, sizeof(outbuf), &outlen);
1233	} while ((rc == -EINTR || rc == -EIO) && retry-- > 0);
1234
1235	if (rc)
1236		return rc;
1237	if (outlen < MC_CMD_GET_ASSERTS_OUT_LEN)
1238		return -EIO;
1239
1240	/* Print out any recorded assertion state */
1241	flags = MCDI_DWORD(outbuf, GET_ASSERTS_OUT_GLOBAL_FLAGS);
1242	if (flags == MC_CMD_GET_ASSERTS_FLAGS_NO_FAILS)
1243		return 0;
1244
1245	reason = (flags == MC_CMD_GET_ASSERTS_FLAGS_SYS_FAIL)
1246		? "system-level assertion"
1247		: (flags == MC_CMD_GET_ASSERTS_FLAGS_THR_FAIL)
1248		? "thread-level assertion"
1249		: (flags == MC_CMD_GET_ASSERTS_FLAGS_WDOG_FIRED)
1250		? "watchdog reset"
1251		: "unknown assertion";
1252	netif_err(efx, hw, efx->net_dev,
1253		  "MCPU %s at PC = 0x%.8x in thread 0x%.8x\n", reason,
1254		  MCDI_DWORD(outbuf, GET_ASSERTS_OUT_SAVED_PC_OFFS),
1255		  MCDI_DWORD(outbuf, GET_ASSERTS_OUT_THREAD_OFFS));
1256
1257	/* Print out the registers */
1258	for (index = 0;
1259	     index < MC_CMD_GET_ASSERTS_OUT_GP_REGS_OFFS_NUM;
1260	     index++)
1261		netif_err(efx, hw, efx->net_dev, "R%.2d (?): 0x%.8x\n",
1262			  1 + index,
1263			  MCDI_ARRAY_DWORD(outbuf, GET_ASSERTS_OUT_GP_REGS_OFFS,
1264					   index));
1265
1266	return 0;
1267}
1268
1269static void efx_mcdi_exit_assertion(struct efx_nic *efx)
1270{
1271	MCDI_DECLARE_BUF(inbuf, MC_CMD_REBOOT_IN_LEN);
1272
1273	/* If the MC is running debug firmware, it might now be
1274	 * waiting for a debugger to attach, but we just want it to
1275	 * reboot.  We set a flag that makes the command a no-op if it
1276	 * has already done so.  We don't know what return code to
1277	 * expect (0 or -EIO), so ignore it.
1278	 */
1279	BUILD_BUG_ON(MC_CMD_REBOOT_OUT_LEN != 0);
1280	MCDI_SET_DWORD(inbuf, REBOOT_IN_FLAGS,
1281		       MC_CMD_REBOOT_FLAGS_AFTER_ASSERTION);
1282	(void) efx_mcdi_rpc(efx, MC_CMD_REBOOT, inbuf, MC_CMD_REBOOT_IN_LEN,
1283			    NULL, 0, NULL);
1284}
1285
1286int efx_mcdi_handle_assertion(struct efx_nic *efx)
1287{
1288	int rc;
1289
1290	rc = efx_mcdi_read_assertion(efx);
1291	if (rc)
1292		return rc;
1293
1294	efx_mcdi_exit_assertion(efx);
1295
1296	return 0;
1297}
1298
1299void efx_mcdi_set_id_led(struct efx_nic *efx, enum efx_led_mode mode)
1300{
1301	MCDI_DECLARE_BUF(inbuf, MC_CMD_SET_ID_LED_IN_LEN);
1302	int rc;
1303
1304	BUILD_BUG_ON(EFX_LED_OFF != MC_CMD_LED_OFF);
1305	BUILD_BUG_ON(EFX_LED_ON != MC_CMD_LED_ON);
1306	BUILD_BUG_ON(EFX_LED_DEFAULT != MC_CMD_LED_DEFAULT);
1307
1308	BUILD_BUG_ON(MC_CMD_SET_ID_LED_OUT_LEN != 0);
1309
1310	MCDI_SET_DWORD(inbuf, SET_ID_LED_IN_STATE, mode);
1311
1312	rc = efx_mcdi_rpc(efx, MC_CMD_SET_ID_LED, inbuf, sizeof(inbuf),
1313			  NULL, 0, NULL);
1314	if (rc)
1315		netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n",
1316			  __func__, rc);
1317}
1318
1319static int efx_mcdi_reset_port(struct efx_nic *efx)
1320{
1321	int rc = efx_mcdi_rpc(efx, MC_CMD_ENTITY_RESET, NULL, 0, NULL, 0, NULL);
1322	if (rc)
1323		netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n",
1324			  __func__, rc);
1325	return rc;
1326}
1327
1328static int efx_mcdi_reset_mc(struct efx_nic *efx)
1329{
1330	MCDI_DECLARE_BUF(inbuf, MC_CMD_REBOOT_IN_LEN);
1331	int rc;
1332
1333	BUILD_BUG_ON(MC_CMD_REBOOT_OUT_LEN != 0);
1334	MCDI_SET_DWORD(inbuf, REBOOT_IN_FLAGS, 0);
1335	rc = efx_mcdi_rpc(efx, MC_CMD_REBOOT, inbuf, sizeof(inbuf),
1336			  NULL, 0, NULL);
1337	/* White is black, and up is down */
1338	if (rc == -EIO)
1339		return 0;
1340	if (rc == 0)
1341		rc = -EIO;
1342	netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
1343	return rc;
1344}
1345
1346enum reset_type efx_mcdi_map_reset_reason(enum reset_type reason)
1347{
1348	return RESET_TYPE_RECOVER_OR_ALL;
1349}
1350
1351int efx_mcdi_reset(struct efx_nic *efx, enum reset_type method)
1352{
1353	int rc;
1354
1355	/* Recover from a failed assertion pre-reset */
1356	rc = efx_mcdi_handle_assertion(efx);
1357	if (rc)
1358		return rc;
1359
1360	if (method == RESET_TYPE_WORLD)
1361		return efx_mcdi_reset_mc(efx);
1362	else
1363		return efx_mcdi_reset_port(efx);
1364}
1365
1366static int efx_mcdi_wol_filter_set(struct efx_nic *efx, u32 type,
1367				   const u8 *mac, int *id_out)
1368{
1369	MCDI_DECLARE_BUF(inbuf, MC_CMD_WOL_FILTER_SET_IN_LEN);
1370	MCDI_DECLARE_BUF(outbuf, MC_CMD_WOL_FILTER_SET_OUT_LEN);
1371	size_t outlen;
1372	int rc;
1373
1374	MCDI_SET_DWORD(inbuf, WOL_FILTER_SET_IN_WOL_TYPE, type);
1375	MCDI_SET_DWORD(inbuf, WOL_FILTER_SET_IN_FILTER_MODE,
1376		       MC_CMD_FILTER_MODE_SIMPLE);
1377	memcpy(MCDI_PTR(inbuf, WOL_FILTER_SET_IN_MAGIC_MAC), mac, ETH_ALEN);
1378
1379	rc = efx_mcdi_rpc(efx, MC_CMD_WOL_FILTER_SET, inbuf, sizeof(inbuf),
1380			  outbuf, sizeof(outbuf), &outlen);
1381	if (rc)
1382		goto fail;
1383
1384	if (outlen < MC_CMD_WOL_FILTER_SET_OUT_LEN) {
1385		rc = -EIO;
1386		goto fail;
1387	}
1388
1389	*id_out = (int)MCDI_DWORD(outbuf, WOL_FILTER_SET_OUT_FILTER_ID);
1390
1391	return 0;
1392
1393fail:
1394	*id_out = -1;
1395	netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
1396	return rc;
1397
1398}
1399
1400
1401int
1402efx_mcdi_wol_filter_set_magic(struct efx_nic *efx,  const u8 *mac, int *id_out)
1403{
1404	return efx_mcdi_wol_filter_set(efx, MC_CMD_WOL_TYPE_MAGIC, mac, id_out);
1405}
1406
1407
1408int efx_mcdi_wol_filter_get_magic(struct efx_nic *efx, int *id_out)
1409{
1410	MCDI_DECLARE_BUF(outbuf, MC_CMD_WOL_FILTER_GET_OUT_LEN);
1411	size_t outlen;
1412	int rc;
1413
1414	rc = efx_mcdi_rpc(efx, MC_CMD_WOL_FILTER_GET, NULL, 0,
1415			  outbuf, sizeof(outbuf), &outlen);
1416	if (rc)
1417		goto fail;
1418
1419	if (outlen < MC_CMD_WOL_FILTER_GET_OUT_LEN) {
1420		rc = -EIO;
1421		goto fail;
1422	}
1423
1424	*id_out = (int)MCDI_DWORD(outbuf, WOL_FILTER_GET_OUT_FILTER_ID);
1425
1426	return 0;
1427
1428fail:
1429	*id_out = -1;
1430	netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
1431	return rc;
1432}
1433
1434
1435int efx_mcdi_wol_filter_remove(struct efx_nic *efx, int id)
1436{
1437	MCDI_DECLARE_BUF(inbuf, MC_CMD_WOL_FILTER_REMOVE_IN_LEN);
1438	int rc;
1439
1440	MCDI_SET_DWORD(inbuf, WOL_FILTER_REMOVE_IN_FILTER_ID, (u32)id);
1441
1442	rc = efx_mcdi_rpc(efx, MC_CMD_WOL_FILTER_REMOVE, inbuf, sizeof(inbuf),
1443			  NULL, 0, NULL);
1444	if (rc)
1445		goto fail;
1446
1447	return 0;
1448
1449fail:
1450	netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
1451	return rc;
1452}
1453
1454int efx_mcdi_flush_rxqs(struct efx_nic *efx)
1455{
1456	struct efx_channel *channel;
1457	struct efx_rx_queue *rx_queue;
1458	MCDI_DECLARE_BUF(inbuf,
1459			 MC_CMD_FLUSH_RX_QUEUES_IN_LEN(EFX_MAX_CHANNELS));
1460	int rc, count;
1461
1462	BUILD_BUG_ON(EFX_MAX_CHANNELS >
1463		     MC_CMD_FLUSH_RX_QUEUES_IN_QID_OFST_MAXNUM);
1464
1465	count = 0;
1466	efx_for_each_channel(channel, efx) {
1467		efx_for_each_channel_rx_queue(rx_queue, channel) {
1468			if (rx_queue->flush_pending) {
1469				rx_queue->flush_pending = false;
1470				atomic_dec(&efx->rxq_flush_pending);
1471				MCDI_SET_ARRAY_DWORD(
1472					inbuf, FLUSH_RX_QUEUES_IN_QID_OFST,
1473					count, efx_rx_queue_index(rx_queue));
1474				count++;
1475			}
1476		}
1477	}
1478
1479	rc = efx_mcdi_rpc(efx, MC_CMD_FLUSH_RX_QUEUES, inbuf,
1480			  MC_CMD_FLUSH_RX_QUEUES_IN_LEN(count), NULL, 0, NULL);
1481	WARN_ON(rc < 0);
1482
1483	return rc;
1484}
1485
1486int efx_mcdi_wol_filter_reset(struct efx_nic *efx)
1487{
1488	int rc;
1489
1490	rc = efx_mcdi_rpc(efx, MC_CMD_WOL_FILTER_RESET, NULL, 0, NULL, 0, NULL);
1491	if (rc)
1492		goto fail;
1493
1494	return 0;
1495
1496fail:
1497	netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
1498	return rc;
1499}
1500
1501int efx_mcdi_set_workaround(struct efx_nic *efx, u32 type, bool enabled)
1502{
1503	MCDI_DECLARE_BUF(inbuf, MC_CMD_WORKAROUND_IN_LEN);
1504
1505	BUILD_BUG_ON(MC_CMD_WORKAROUND_OUT_LEN != 0);
1506	MCDI_SET_DWORD(inbuf, WORKAROUND_IN_TYPE, type);
1507	MCDI_SET_DWORD(inbuf, WORKAROUND_IN_ENABLED, enabled);
1508	return efx_mcdi_rpc(efx, MC_CMD_WORKAROUND, inbuf, sizeof(inbuf),
1509			    NULL, 0, NULL);
1510}
1511
1512#ifdef CONFIG_SFC_MTD
1513
1514#define EFX_MCDI_NVRAM_LEN_MAX 128
1515
1516static int efx_mcdi_nvram_update_start(struct efx_nic *efx, unsigned int type)
1517{
1518	MCDI_DECLARE_BUF(inbuf, MC_CMD_NVRAM_UPDATE_START_IN_LEN);
1519	int rc;
1520
1521	MCDI_SET_DWORD(inbuf, NVRAM_UPDATE_START_IN_TYPE, type);
1522
1523	BUILD_BUG_ON(MC_CMD_NVRAM_UPDATE_START_OUT_LEN != 0);
1524
1525	rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_UPDATE_START, inbuf, sizeof(inbuf),
1526			  NULL, 0, NULL);
1527	if (rc)
1528		goto fail;
1529
1530	return 0;
1531
1532fail:
1533	netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
1534	return rc;
1535}
1536
1537static int efx_mcdi_nvram_read(struct efx_nic *efx, unsigned int type,
1538			       loff_t offset, u8 *buffer, size_t length)
1539{
1540	MCDI_DECLARE_BUF(inbuf, MC_CMD_NVRAM_READ_IN_LEN);
1541	MCDI_DECLARE_BUF(outbuf,
1542			 MC_CMD_NVRAM_READ_OUT_LEN(EFX_MCDI_NVRAM_LEN_MAX));
1543	size_t outlen;
1544	int rc;
1545
1546	MCDI_SET_DWORD(inbuf, NVRAM_READ_IN_TYPE, type);
1547	MCDI_SET_DWORD(inbuf, NVRAM_READ_IN_OFFSET, offset);
1548	MCDI_SET_DWORD(inbuf, NVRAM_READ_IN_LENGTH, length);
1549
1550	rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_READ, inbuf, sizeof(inbuf),
1551			  outbuf, sizeof(outbuf), &outlen);
1552	if (rc)
1553		goto fail;
1554
1555	memcpy(buffer, MCDI_PTR(outbuf, NVRAM_READ_OUT_READ_BUFFER), length);
1556	return 0;
1557
1558fail:
1559	netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
1560	return rc;
1561}
1562
1563static int efx_mcdi_nvram_write(struct efx_nic *efx, unsigned int type,
1564				loff_t offset, const u8 *buffer, size_t length)
1565{
1566	MCDI_DECLARE_BUF(inbuf,
1567			 MC_CMD_NVRAM_WRITE_IN_LEN(EFX_MCDI_NVRAM_LEN_MAX));
1568	int rc;
1569
1570	MCDI_SET_DWORD(inbuf, NVRAM_WRITE_IN_TYPE, type);
1571	MCDI_SET_DWORD(inbuf, NVRAM_WRITE_IN_OFFSET, offset);
1572	MCDI_SET_DWORD(inbuf, NVRAM_WRITE_IN_LENGTH, length);
1573	memcpy(MCDI_PTR(inbuf, NVRAM_WRITE_IN_WRITE_BUFFER), buffer, length);
1574
1575	BUILD_BUG_ON(MC_CMD_NVRAM_WRITE_OUT_LEN != 0);
1576
1577	rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_WRITE, inbuf,
1578			  ALIGN(MC_CMD_NVRAM_WRITE_IN_LEN(length), 4),
1579			  NULL, 0, NULL);
1580	if (rc)
1581		goto fail;
1582
1583	return 0;
1584
1585fail:
1586	netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
1587	return rc;
1588}
1589
1590static int efx_mcdi_nvram_erase(struct efx_nic *efx, unsigned int type,
1591				loff_t offset, size_t length)
1592{
1593	MCDI_DECLARE_BUF(inbuf, MC_CMD_NVRAM_ERASE_IN_LEN);
1594	int rc;
1595
1596	MCDI_SET_DWORD(inbuf, NVRAM_ERASE_IN_TYPE, type);
1597	MCDI_SET_DWORD(inbuf, NVRAM_ERASE_IN_OFFSET, offset);
1598	MCDI_SET_DWORD(inbuf, NVRAM_ERASE_IN_LENGTH, length);
1599
1600	BUILD_BUG_ON(MC_CMD_NVRAM_ERASE_OUT_LEN != 0);
1601
1602	rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_ERASE, inbuf, sizeof(inbuf),
1603			  NULL, 0, NULL);
1604	if (rc)
1605		goto fail;
1606
1607	return 0;
1608
1609fail:
1610	netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
1611	return rc;
1612}
1613
1614static int efx_mcdi_nvram_update_finish(struct efx_nic *efx, unsigned int type)
1615{
1616	MCDI_DECLARE_BUF(inbuf, MC_CMD_NVRAM_UPDATE_FINISH_IN_LEN);
1617	int rc;
1618
1619	MCDI_SET_DWORD(inbuf, NVRAM_UPDATE_FINISH_IN_TYPE, type);
1620
1621	BUILD_BUG_ON(MC_CMD_NVRAM_UPDATE_FINISH_OUT_LEN != 0);
1622
1623	rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_UPDATE_FINISH, inbuf, sizeof(inbuf),
1624			  NULL, 0, NULL);
1625	if (rc)
1626		goto fail;
1627
1628	return 0;
1629
1630fail:
1631	netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
1632	return rc;
1633}
1634
1635int efx_mcdi_mtd_read(struct mtd_info *mtd, loff_t start,
1636		      size_t len, size_t *retlen, u8 *buffer)
1637{
1638	struct efx_mcdi_mtd_partition *part = to_efx_mcdi_mtd_partition(mtd);
1639	struct efx_nic *efx = mtd->priv;
1640	loff_t offset = start;
1641	loff_t end = min_t(loff_t, start + len, mtd->size);
1642	size_t chunk;
1643	int rc = 0;
1644
1645	while (offset < end) {
1646		chunk = min_t(size_t, end - offset, EFX_MCDI_NVRAM_LEN_MAX);
1647		rc = efx_mcdi_nvram_read(efx, part->nvram_type, offset,
1648					 buffer, chunk);
1649		if (rc)
1650			goto out;
1651		offset += chunk;
1652		buffer += chunk;
1653	}
1654out:
1655	*retlen = offset - start;
1656	return rc;
1657}
1658
1659int efx_mcdi_mtd_erase(struct mtd_info *mtd, loff_t start, size_t len)
1660{
1661	struct efx_mcdi_mtd_partition *part = to_efx_mcdi_mtd_partition(mtd);
1662	struct efx_nic *efx = mtd->priv;
1663	loff_t offset = start & ~((loff_t)(mtd->erasesize - 1));
1664	loff_t end = min_t(loff_t, start + len, mtd->size);
1665	size_t chunk = part->common.mtd.erasesize;
1666	int rc = 0;
1667
1668	if (!part->updating) {
1669		rc = efx_mcdi_nvram_update_start(efx, part->nvram_type);
1670		if (rc)
1671			goto out;
1672		part->updating = true;
1673	}
1674
1675	/* The MCDI interface can in fact do multiple erase blocks at once;
1676	 * but erasing may be slow, so we make multiple calls here to avoid
1677	 * tripping the MCDI RPC timeout. */
1678	while (offset < end) {
1679		rc = efx_mcdi_nvram_erase(efx, part->nvram_type, offset,
1680					  chunk);
1681		if (rc)
1682			goto out;
1683		offset += chunk;
1684	}
1685out:
1686	return rc;
1687}
1688
1689int efx_mcdi_mtd_write(struct mtd_info *mtd, loff_t start,
1690		       size_t len, size_t *retlen, const u8 *buffer)
1691{
1692	struct efx_mcdi_mtd_partition *part = to_efx_mcdi_mtd_partition(mtd);
1693	struct efx_nic *efx = mtd->priv;
1694	loff_t offset = start;
1695	loff_t end = min_t(loff_t, start + len, mtd->size);
1696	size_t chunk;
1697	int rc = 0;
1698
1699	if (!part->updating) {
1700		rc = efx_mcdi_nvram_update_start(efx, part->nvram_type);
1701		if (rc)
1702			goto out;
1703		part->updating = true;
1704	}
1705
1706	while (offset < end) {
1707		chunk = min_t(size_t, end - offset, EFX_MCDI_NVRAM_LEN_MAX);
1708		rc = efx_mcdi_nvram_write(efx, part->nvram_type, offset,
1709					  buffer, chunk);
1710		if (rc)
1711			goto out;
1712		offset += chunk;
1713		buffer += chunk;
1714	}
1715out:
1716	*retlen = offset - start;
1717	return rc;
1718}
1719
1720int efx_mcdi_mtd_sync(struct mtd_info *mtd)
1721{
1722	struct efx_mcdi_mtd_partition *part = to_efx_mcdi_mtd_partition(mtd);
1723	struct efx_nic *efx = mtd->priv;
1724	int rc = 0;
1725
1726	if (part->updating) {
1727		part->updating = false;
1728		rc = efx_mcdi_nvram_update_finish(efx, part->nvram_type);
1729	}
1730
1731	return rc;
1732}
1733
1734void efx_mcdi_mtd_rename(struct efx_mtd_partition *part)
1735{
1736	struct efx_mcdi_mtd_partition *mcdi_part =
1737		container_of(part, struct efx_mcdi_mtd_partition, common);
1738	struct efx_nic *efx = part->mtd.priv;
1739
1740	snprintf(part->name, sizeof(part->name), "%s %s:%02x",
1741		 efx->name, part->type_name, mcdi_part->fw_subtype);
1742}
1743
1744#endif /* CONFIG_SFC_MTD */
1745