xref: /openbmc/linux/drivers/net/ethernet/sfc/mcdi.c (revision 47ebd031)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /****************************************************************************
3  * Driver for Solarflare network controllers and boards
4  * Copyright 2008-2013 Solarflare Communications Inc.
5  */
6 
7 #include <linux/delay.h>
8 #include <linux/moduleparam.h>
9 #include <linux/atomic.h>
10 #include "net_driver.h"
11 #include "nic.h"
12 #include "io.h"
13 #include "farch_regs.h"
14 #include "mcdi_pcol.h"
15 
16 /**************************************************************************
17  *
18  * Management-Controller-to-Driver Interface
19  *
20  **************************************************************************
21  */
22 
23 #define MCDI_RPC_TIMEOUT       (10 * HZ)
24 
25 /* A reboot/assertion causes the MCDI status word to be set after the
26  * command word is set or a REBOOT event is sent. If we notice a reboot
27  * via these mechanisms then wait 250ms for the status word to be set.
28  */
29 #define MCDI_STATUS_DELAY_US		100
30 #define MCDI_STATUS_DELAY_COUNT		2500
31 #define MCDI_STATUS_SLEEP_MS						\
32 	(MCDI_STATUS_DELAY_US * MCDI_STATUS_DELAY_COUNT / 1000)
33 
34 #define SEQ_MASK							\
35 	EFX_MASK32(EFX_WIDTH(MCDI_HEADER_SEQ))
36 
37 struct efx_mcdi_async_param {
38 	struct list_head list;
39 	unsigned int cmd;
40 	size_t inlen;
41 	size_t outlen;
42 	bool quiet;
43 	efx_mcdi_async_completer *complete;
44 	unsigned long cookie;
45 	/* followed by request/response buffer */
46 };
47 
48 static void efx_mcdi_timeout_async(struct timer_list *t);
49 static int efx_mcdi_drv_attach(struct efx_nic *efx, bool driver_operating,
50 			       bool *was_attached_out);
51 static bool efx_mcdi_poll_once(struct efx_nic *efx);
52 static void efx_mcdi_abandon(struct efx_nic *efx);
53 
54 #ifdef CONFIG_SFC_MCDI_LOGGING
55 static bool mcdi_logging_default;
56 module_param(mcdi_logging_default, bool, 0644);
57 MODULE_PARM_DESC(mcdi_logging_default,
58 		 "Enable MCDI logging on newly-probed functions");
59 #endif
60 
61 int efx_mcdi_init(struct efx_nic *efx)
62 {
63 	struct efx_mcdi_iface *mcdi;
64 	bool already_attached;
65 	int rc = -ENOMEM;
66 
67 	efx->mcdi = kzalloc(sizeof(*efx->mcdi), GFP_KERNEL);
68 	if (!efx->mcdi)
69 		goto fail;
70 
71 	mcdi = efx_mcdi(efx);
72 	mcdi->efx = efx;
73 #ifdef CONFIG_SFC_MCDI_LOGGING
74 	/* consuming code assumes buffer is page-sized */
75 	mcdi->logging_buffer = (char *)__get_free_page(GFP_KERNEL);
76 	if (!mcdi->logging_buffer)
77 		goto fail1;
78 	mcdi->logging_enabled = mcdi_logging_default;
79 #endif
80 	init_waitqueue_head(&mcdi->wq);
81 	init_waitqueue_head(&mcdi->proxy_rx_wq);
82 	spin_lock_init(&mcdi->iface_lock);
83 	mcdi->state = MCDI_STATE_QUIESCENT;
84 	mcdi->mode = MCDI_MODE_POLL;
85 	spin_lock_init(&mcdi->async_lock);
86 	INIT_LIST_HEAD(&mcdi->async_list);
87 	timer_setup(&mcdi->async_timer, efx_mcdi_timeout_async, 0);
88 
89 	(void) efx_mcdi_poll_reboot(efx);
90 	mcdi->new_epoch = true;
91 
92 	/* Recover from a failed assertion before probing */
93 	rc = efx_mcdi_handle_assertion(efx);
94 	if (rc)
95 		goto fail2;
96 
97 	/* Let the MC (and BMC, if this is a LOM) know that the driver
98 	 * is loaded. We should do this before we reset the NIC.
99 	 */
100 	rc = efx_mcdi_drv_attach(efx, true, &already_attached);
101 	if (rc) {
102 		pci_err(efx->pci_dev, "Unable to register driver with MCPU\n");
103 		goto fail2;
104 	}
105 	if (already_attached)
106 		/* Not a fatal error */
107 		pci_err(efx->pci_dev, "Host already registered with MCPU\n");
108 
109 	if (efx->mcdi->fn_flags &
110 	    (1 << MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_PRIMARY))
111 		efx->primary = efx;
112 
113 	return 0;
114 fail2:
115 #ifdef CONFIG_SFC_MCDI_LOGGING
116 	free_page((unsigned long)mcdi->logging_buffer);
117 fail1:
118 #endif
119 	kfree(efx->mcdi);
120 	efx->mcdi = NULL;
121 fail:
122 	return rc;
123 }
124 
125 void efx_mcdi_detach(struct efx_nic *efx)
126 {
127 	if (!efx->mcdi)
128 		return;
129 
130 	BUG_ON(efx->mcdi->iface.state != MCDI_STATE_QUIESCENT);
131 
132 	/* Relinquish the device (back to the BMC, if this is a LOM) */
133 	efx_mcdi_drv_attach(efx, false, NULL);
134 }
135 
136 void efx_mcdi_fini(struct efx_nic *efx)
137 {
138 	if (!efx->mcdi)
139 		return;
140 
141 #ifdef CONFIG_SFC_MCDI_LOGGING
142 	free_page((unsigned long)efx->mcdi->iface.logging_buffer);
143 #endif
144 
145 	kfree(efx->mcdi);
146 }
147 
148 static void efx_mcdi_send_request(struct efx_nic *efx, unsigned cmd,
149 				  const efx_dword_t *inbuf, size_t inlen)
150 {
151 	struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
152 #ifdef CONFIG_SFC_MCDI_LOGGING
153 	char *buf = mcdi->logging_buffer; /* page-sized */
154 #endif
155 	efx_dword_t hdr[2];
156 	size_t hdr_len;
157 	u32 xflags, seqno;
158 
159 	BUG_ON(mcdi->state == MCDI_STATE_QUIESCENT);
160 
161 	/* Serialise with efx_mcdi_ev_cpl() and efx_mcdi_ev_death() */
162 	spin_lock_bh(&mcdi->iface_lock);
163 	++mcdi->seqno;
164 	seqno = mcdi->seqno & SEQ_MASK;
165 	spin_unlock_bh(&mcdi->iface_lock);
166 
167 	xflags = 0;
168 	if (mcdi->mode == MCDI_MODE_EVENTS)
169 		xflags |= MCDI_HEADER_XFLAGS_EVREQ;
170 
171 	if (efx->type->mcdi_max_ver == 1) {
172 		/* MCDI v1 */
173 		EFX_POPULATE_DWORD_7(hdr[0],
174 				     MCDI_HEADER_RESPONSE, 0,
175 				     MCDI_HEADER_RESYNC, 1,
176 				     MCDI_HEADER_CODE, cmd,
177 				     MCDI_HEADER_DATALEN, inlen,
178 				     MCDI_HEADER_SEQ, seqno,
179 				     MCDI_HEADER_XFLAGS, xflags,
180 				     MCDI_HEADER_NOT_EPOCH, !mcdi->new_epoch);
181 		hdr_len = 4;
182 	} else {
183 		/* MCDI v2 */
184 		BUG_ON(inlen > MCDI_CTL_SDU_LEN_MAX_V2);
185 		EFX_POPULATE_DWORD_7(hdr[0],
186 				     MCDI_HEADER_RESPONSE, 0,
187 				     MCDI_HEADER_RESYNC, 1,
188 				     MCDI_HEADER_CODE, MC_CMD_V2_EXTN,
189 				     MCDI_HEADER_DATALEN, 0,
190 				     MCDI_HEADER_SEQ, seqno,
191 				     MCDI_HEADER_XFLAGS, xflags,
192 				     MCDI_HEADER_NOT_EPOCH, !mcdi->new_epoch);
193 		EFX_POPULATE_DWORD_2(hdr[1],
194 				     MC_CMD_V2_EXTN_IN_EXTENDED_CMD, cmd,
195 				     MC_CMD_V2_EXTN_IN_ACTUAL_LEN, inlen);
196 		hdr_len = 8;
197 	}
198 
199 #ifdef CONFIG_SFC_MCDI_LOGGING
200 	if (mcdi->logging_enabled && !WARN_ON_ONCE(!buf)) {
201 		int bytes = 0;
202 		int i;
203 		/* Lengths should always be a whole number of dwords, so scream
204 		 * if they're not.
205 		 */
206 		WARN_ON_ONCE(hdr_len % 4);
207 		WARN_ON_ONCE(inlen % 4);
208 
209 		/* We own the logging buffer, as only one MCDI can be in
210 		 * progress on a NIC at any one time.  So no need for locking.
211 		 */
212 		for (i = 0; i < hdr_len / 4 && bytes < PAGE_SIZE; i++)
213 			bytes += scnprintf(buf + bytes, PAGE_SIZE - bytes,
214 					   " %08x",
215 					   le32_to_cpu(hdr[i].u32[0]));
216 
217 		for (i = 0; i < inlen / 4 && bytes < PAGE_SIZE; i++)
218 			bytes += scnprintf(buf + bytes, PAGE_SIZE - bytes,
219 					   " %08x",
220 					   le32_to_cpu(inbuf[i].u32[0]));
221 
222 		netif_info(efx, hw, efx->net_dev, "MCDI RPC REQ:%s\n", buf);
223 	}
224 #endif
225 
226 	efx->type->mcdi_request(efx, hdr, hdr_len, inbuf, inlen);
227 
228 	mcdi->new_epoch = false;
229 }
230 
231 static int efx_mcdi_errno(unsigned int mcdi_err)
232 {
233 	switch (mcdi_err) {
234 	case 0:
235 		return 0;
236 #define TRANSLATE_ERROR(name)					\
237 	case MC_CMD_ERR_ ## name:				\
238 		return -name;
239 	TRANSLATE_ERROR(EPERM);
240 	TRANSLATE_ERROR(ENOENT);
241 	TRANSLATE_ERROR(EINTR);
242 	TRANSLATE_ERROR(EAGAIN);
243 	TRANSLATE_ERROR(EACCES);
244 	TRANSLATE_ERROR(EBUSY);
245 	TRANSLATE_ERROR(EINVAL);
246 	TRANSLATE_ERROR(EDEADLK);
247 	TRANSLATE_ERROR(ENOSYS);
248 	TRANSLATE_ERROR(ETIME);
249 	TRANSLATE_ERROR(EALREADY);
250 	TRANSLATE_ERROR(ENOSPC);
251 #undef TRANSLATE_ERROR
252 	case MC_CMD_ERR_ENOTSUP:
253 		return -EOPNOTSUPP;
254 	case MC_CMD_ERR_ALLOC_FAIL:
255 		return -ENOBUFS;
256 	case MC_CMD_ERR_MAC_EXIST:
257 		return -EADDRINUSE;
258 	default:
259 		return -EPROTO;
260 	}
261 }
262 
263 static void efx_mcdi_read_response_header(struct efx_nic *efx)
264 {
265 	struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
266 	unsigned int respseq, respcmd, error;
267 #ifdef CONFIG_SFC_MCDI_LOGGING
268 	char *buf = mcdi->logging_buffer; /* page-sized */
269 #endif
270 	efx_dword_t hdr;
271 
272 	efx->type->mcdi_read_response(efx, &hdr, 0, 4);
273 	respseq = EFX_DWORD_FIELD(hdr, MCDI_HEADER_SEQ);
274 	respcmd = EFX_DWORD_FIELD(hdr, MCDI_HEADER_CODE);
275 	error = EFX_DWORD_FIELD(hdr, MCDI_HEADER_ERROR);
276 
277 	if (respcmd != MC_CMD_V2_EXTN) {
278 		mcdi->resp_hdr_len = 4;
279 		mcdi->resp_data_len = EFX_DWORD_FIELD(hdr, MCDI_HEADER_DATALEN);
280 	} else {
281 		efx->type->mcdi_read_response(efx, &hdr, 4, 4);
282 		mcdi->resp_hdr_len = 8;
283 		mcdi->resp_data_len =
284 			EFX_DWORD_FIELD(hdr, MC_CMD_V2_EXTN_IN_ACTUAL_LEN);
285 	}
286 
287 #ifdef CONFIG_SFC_MCDI_LOGGING
288 	if (mcdi->logging_enabled && !WARN_ON_ONCE(!buf)) {
289 		size_t hdr_len, data_len;
290 		int bytes = 0;
291 		int i;
292 
293 		WARN_ON_ONCE(mcdi->resp_hdr_len % 4);
294 		hdr_len = mcdi->resp_hdr_len / 4;
295 		/* MCDI_DECLARE_BUF ensures that underlying buffer is padded
296 		 * to dword size, and the MCDI buffer is always dword size
297 		 */
298 		data_len = DIV_ROUND_UP(mcdi->resp_data_len, 4);
299 
300 		/* We own the logging buffer, as only one MCDI can be in
301 		 * progress on a NIC at any one time.  So no need for locking.
302 		 */
303 		for (i = 0; i < hdr_len && bytes < PAGE_SIZE; i++) {
304 			efx->type->mcdi_read_response(efx, &hdr, (i * 4), 4);
305 			bytes += scnprintf(buf + bytes, PAGE_SIZE - bytes,
306 					   " %08x", le32_to_cpu(hdr.u32[0]));
307 		}
308 
309 		for (i = 0; i < data_len && bytes < PAGE_SIZE; i++) {
310 			efx->type->mcdi_read_response(efx, &hdr,
311 					mcdi->resp_hdr_len + (i * 4), 4);
312 			bytes += scnprintf(buf + bytes, PAGE_SIZE - bytes,
313 					   " %08x", le32_to_cpu(hdr.u32[0]));
314 		}
315 
316 		netif_info(efx, hw, efx->net_dev, "MCDI RPC RESP:%s\n", buf);
317 	}
318 #endif
319 
320 	mcdi->resprc_raw = 0;
321 	if (error && mcdi->resp_data_len == 0) {
322 		netif_err(efx, hw, efx->net_dev, "MC rebooted\n");
323 		mcdi->resprc = -EIO;
324 	} else if ((respseq ^ mcdi->seqno) & SEQ_MASK) {
325 		netif_err(efx, hw, efx->net_dev,
326 			  "MC response mismatch tx seq 0x%x rx seq 0x%x\n",
327 			  respseq, mcdi->seqno);
328 		mcdi->resprc = -EIO;
329 	} else if (error) {
330 		efx->type->mcdi_read_response(efx, &hdr, mcdi->resp_hdr_len, 4);
331 		mcdi->resprc_raw = EFX_DWORD_FIELD(hdr, EFX_DWORD_0);
332 		mcdi->resprc = efx_mcdi_errno(mcdi->resprc_raw);
333 	} else {
334 		mcdi->resprc = 0;
335 	}
336 }
337 
338 static bool efx_mcdi_poll_once(struct efx_nic *efx)
339 {
340 	struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
341 
342 	rmb();
343 	if (!efx->type->mcdi_poll_response(efx))
344 		return false;
345 
346 	spin_lock_bh(&mcdi->iface_lock);
347 	efx_mcdi_read_response_header(efx);
348 	spin_unlock_bh(&mcdi->iface_lock);
349 
350 	return true;
351 }
352 
353 static int efx_mcdi_poll(struct efx_nic *efx)
354 {
355 	struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
356 	unsigned long time, finish;
357 	unsigned int spins;
358 	int rc;
359 
360 	/* Check for a reboot atomically with respect to efx_mcdi_copyout() */
361 	rc = efx_mcdi_poll_reboot(efx);
362 	if (rc) {
363 		spin_lock_bh(&mcdi->iface_lock);
364 		mcdi->resprc = rc;
365 		mcdi->resp_hdr_len = 0;
366 		mcdi->resp_data_len = 0;
367 		spin_unlock_bh(&mcdi->iface_lock);
368 		return 0;
369 	}
370 
371 	/* Poll for completion. Poll quickly (once a us) for the 1st jiffy,
372 	 * because generally mcdi responses are fast. After that, back off
373 	 * and poll once a jiffy (approximately)
374 	 */
375 	spins = USER_TICK_USEC;
376 	finish = jiffies + MCDI_RPC_TIMEOUT;
377 
378 	while (1) {
379 		if (spins != 0) {
380 			--spins;
381 			udelay(1);
382 		} else {
383 			schedule_timeout_uninterruptible(1);
384 		}
385 
386 		time = jiffies;
387 
388 		if (efx_mcdi_poll_once(efx))
389 			break;
390 
391 		if (time_after(time, finish))
392 			return -ETIMEDOUT;
393 	}
394 
395 	/* Return rc=0 like wait_event_timeout() */
396 	return 0;
397 }
398 
399 /* Test and clear MC-rebooted flag for this port/function; reset
400  * software state as necessary.
401  */
402 int efx_mcdi_poll_reboot(struct efx_nic *efx)
403 {
404 	if (!efx->mcdi)
405 		return 0;
406 
407 	return efx->type->mcdi_poll_reboot(efx);
408 }
409 
410 static bool efx_mcdi_acquire_async(struct efx_mcdi_iface *mcdi)
411 {
412 	return cmpxchg(&mcdi->state,
413 		       MCDI_STATE_QUIESCENT, MCDI_STATE_RUNNING_ASYNC) ==
414 		MCDI_STATE_QUIESCENT;
415 }
416 
417 static void efx_mcdi_acquire_sync(struct efx_mcdi_iface *mcdi)
418 {
419 	/* Wait until the interface becomes QUIESCENT and we win the race
420 	 * to mark it RUNNING_SYNC.
421 	 */
422 	wait_event(mcdi->wq,
423 		   cmpxchg(&mcdi->state,
424 			   MCDI_STATE_QUIESCENT, MCDI_STATE_RUNNING_SYNC) ==
425 		   MCDI_STATE_QUIESCENT);
426 }
427 
428 static int efx_mcdi_await_completion(struct efx_nic *efx)
429 {
430 	struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
431 
432 	if (wait_event_timeout(mcdi->wq, mcdi->state == MCDI_STATE_COMPLETED,
433 			       MCDI_RPC_TIMEOUT) == 0)
434 		return -ETIMEDOUT;
435 
436 	/* Check if efx_mcdi_set_mode() switched us back to polled completions.
437 	 * In which case, poll for completions directly. If efx_mcdi_ev_cpl()
438 	 * completed the request first, then we'll just end up completing the
439 	 * request again, which is safe.
440 	 *
441 	 * We need an smp_rmb() to synchronise with efx_mcdi_mode_poll(), which
442 	 * wait_event_timeout() implicitly provides.
443 	 */
444 	if (mcdi->mode == MCDI_MODE_POLL)
445 		return efx_mcdi_poll(efx);
446 
447 	return 0;
448 }
449 
450 /* If the interface is RUNNING_SYNC, switch to COMPLETED and wake the
451  * requester.  Return whether this was done.  Does not take any locks.
452  */
453 static bool efx_mcdi_complete_sync(struct efx_mcdi_iface *mcdi)
454 {
455 	if (cmpxchg(&mcdi->state,
456 		    MCDI_STATE_RUNNING_SYNC, MCDI_STATE_COMPLETED) ==
457 	    MCDI_STATE_RUNNING_SYNC) {
458 		wake_up(&mcdi->wq);
459 		return true;
460 	}
461 
462 	return false;
463 }
464 
465 static void efx_mcdi_release(struct efx_mcdi_iface *mcdi)
466 {
467 	if (mcdi->mode == MCDI_MODE_EVENTS) {
468 		struct efx_mcdi_async_param *async;
469 		struct efx_nic *efx = mcdi->efx;
470 
471 		/* Process the asynchronous request queue */
472 		spin_lock_bh(&mcdi->async_lock);
473 		async = list_first_entry_or_null(
474 			&mcdi->async_list, struct efx_mcdi_async_param, list);
475 		if (async) {
476 			mcdi->state = MCDI_STATE_RUNNING_ASYNC;
477 			efx_mcdi_send_request(efx, async->cmd,
478 					      (const efx_dword_t *)(async + 1),
479 					      async->inlen);
480 			mod_timer(&mcdi->async_timer,
481 				  jiffies + MCDI_RPC_TIMEOUT);
482 		}
483 		spin_unlock_bh(&mcdi->async_lock);
484 
485 		if (async)
486 			return;
487 	}
488 
489 	mcdi->state = MCDI_STATE_QUIESCENT;
490 	wake_up(&mcdi->wq);
491 }
492 
493 /* If the interface is RUNNING_ASYNC, switch to COMPLETED, call the
494  * asynchronous completion function, and release the interface.
495  * Return whether this was done.  Must be called in bh-disabled
496  * context.  Will take iface_lock and async_lock.
497  */
498 static bool efx_mcdi_complete_async(struct efx_mcdi_iface *mcdi, bool timeout)
499 {
500 	struct efx_nic *efx = mcdi->efx;
501 	struct efx_mcdi_async_param *async;
502 	size_t hdr_len, data_len, err_len;
503 	efx_dword_t *outbuf;
504 	MCDI_DECLARE_BUF_ERR(errbuf);
505 	int rc;
506 
507 	if (cmpxchg(&mcdi->state,
508 		    MCDI_STATE_RUNNING_ASYNC, MCDI_STATE_COMPLETED) !=
509 	    MCDI_STATE_RUNNING_ASYNC)
510 		return false;
511 
512 	spin_lock(&mcdi->iface_lock);
513 	if (timeout) {
514 		/* Ensure that if the completion event arrives later,
515 		 * the seqno check in efx_mcdi_ev_cpl() will fail
516 		 */
517 		++mcdi->seqno;
518 		++mcdi->credits;
519 		rc = -ETIMEDOUT;
520 		hdr_len = 0;
521 		data_len = 0;
522 	} else {
523 		rc = mcdi->resprc;
524 		hdr_len = mcdi->resp_hdr_len;
525 		data_len = mcdi->resp_data_len;
526 	}
527 	spin_unlock(&mcdi->iface_lock);
528 
529 	/* Stop the timer.  In case the timer function is running, we
530 	 * must wait for it to return so that there is no possibility
531 	 * of it aborting the next request.
532 	 */
533 	if (!timeout)
534 		del_timer_sync(&mcdi->async_timer);
535 
536 	spin_lock(&mcdi->async_lock);
537 	async = list_first_entry(&mcdi->async_list,
538 				 struct efx_mcdi_async_param, list);
539 	list_del(&async->list);
540 	spin_unlock(&mcdi->async_lock);
541 
542 	outbuf = (efx_dword_t *)(async + 1);
543 	efx->type->mcdi_read_response(efx, outbuf, hdr_len,
544 				      min(async->outlen, data_len));
545 	if (!timeout && rc && !async->quiet) {
546 		err_len = min(sizeof(errbuf), data_len);
547 		efx->type->mcdi_read_response(efx, errbuf, hdr_len,
548 					      sizeof(errbuf));
549 		efx_mcdi_display_error(efx, async->cmd, async->inlen, errbuf,
550 				       err_len, rc);
551 	}
552 
553 	if (async->complete)
554 		async->complete(efx, async->cookie, rc, outbuf,
555 				min(async->outlen, data_len));
556 	kfree(async);
557 
558 	efx_mcdi_release(mcdi);
559 
560 	return true;
561 }
562 
563 static void efx_mcdi_ev_cpl(struct efx_nic *efx, unsigned int seqno,
564 			    unsigned int datalen, unsigned int mcdi_err)
565 {
566 	struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
567 	bool wake = false;
568 
569 	spin_lock(&mcdi->iface_lock);
570 
571 	if ((seqno ^ mcdi->seqno) & SEQ_MASK) {
572 		if (mcdi->credits)
573 			/* The request has been cancelled */
574 			--mcdi->credits;
575 		else
576 			netif_err(efx, hw, efx->net_dev,
577 				  "MC response mismatch tx seq 0x%x rx "
578 				  "seq 0x%x\n", seqno, mcdi->seqno);
579 	} else {
580 		if (efx->type->mcdi_max_ver >= 2) {
581 			/* MCDI v2 responses don't fit in an event */
582 			efx_mcdi_read_response_header(efx);
583 		} else {
584 			mcdi->resprc = efx_mcdi_errno(mcdi_err);
585 			mcdi->resp_hdr_len = 4;
586 			mcdi->resp_data_len = datalen;
587 		}
588 
589 		wake = true;
590 	}
591 
592 	spin_unlock(&mcdi->iface_lock);
593 
594 	if (wake) {
595 		if (!efx_mcdi_complete_async(mcdi, false))
596 			(void) efx_mcdi_complete_sync(mcdi);
597 
598 		/* If the interface isn't RUNNING_ASYNC or
599 		 * RUNNING_SYNC then we've received a duplicate
600 		 * completion after we've already transitioned back to
601 		 * QUIESCENT. [A subsequent invocation would increment
602 		 * seqno, so would have failed the seqno check].
603 		 */
604 	}
605 }
606 
607 static void efx_mcdi_timeout_async(struct timer_list *t)
608 {
609 	struct efx_mcdi_iface *mcdi = from_timer(mcdi, t, async_timer);
610 
611 	efx_mcdi_complete_async(mcdi, true);
612 }
613 
614 static int
615 efx_mcdi_check_supported(struct efx_nic *efx, unsigned int cmd, size_t inlen)
616 {
617 	if (efx->type->mcdi_max_ver < 0 ||
618 	     (efx->type->mcdi_max_ver < 2 &&
619 	      cmd > MC_CMD_CMD_SPACE_ESCAPE_7))
620 		return -EINVAL;
621 
622 	if (inlen > MCDI_CTL_SDU_LEN_MAX_V2 ||
623 	    (efx->type->mcdi_max_ver < 2 &&
624 	     inlen > MCDI_CTL_SDU_LEN_MAX_V1))
625 		return -EMSGSIZE;
626 
627 	return 0;
628 }
629 
630 static bool efx_mcdi_get_proxy_handle(struct efx_nic *efx,
631 				      size_t hdr_len, size_t data_len,
632 				      u32 *proxy_handle)
633 {
634 	MCDI_DECLARE_BUF_ERR(testbuf);
635 	const size_t buflen = sizeof(testbuf);
636 
637 	if (!proxy_handle || data_len < buflen)
638 		return false;
639 
640 	efx->type->mcdi_read_response(efx, testbuf, hdr_len, buflen);
641 	if (MCDI_DWORD(testbuf, ERR_CODE) == MC_CMD_ERR_PROXY_PENDING) {
642 		*proxy_handle = MCDI_DWORD(testbuf, ERR_PROXY_PENDING_HANDLE);
643 		return true;
644 	}
645 
646 	return false;
647 }
648 
649 static int _efx_mcdi_rpc_finish(struct efx_nic *efx, unsigned int cmd,
650 				size_t inlen,
651 				efx_dword_t *outbuf, size_t outlen,
652 				size_t *outlen_actual, bool quiet,
653 				u32 *proxy_handle, int *raw_rc)
654 {
655 	struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
656 	MCDI_DECLARE_BUF_ERR(errbuf);
657 	int rc;
658 
659 	if (mcdi->mode == MCDI_MODE_POLL)
660 		rc = efx_mcdi_poll(efx);
661 	else
662 		rc = efx_mcdi_await_completion(efx);
663 
664 	if (rc != 0) {
665 		netif_err(efx, hw, efx->net_dev,
666 			  "MC command 0x%x inlen %d mode %d timed out\n",
667 			  cmd, (int)inlen, mcdi->mode);
668 
669 		if (mcdi->mode == MCDI_MODE_EVENTS && efx_mcdi_poll_once(efx)) {
670 			netif_err(efx, hw, efx->net_dev,
671 				  "MCDI request was completed without an event\n");
672 			rc = 0;
673 		}
674 
675 		efx_mcdi_abandon(efx);
676 
677 		/* Close the race with efx_mcdi_ev_cpl() executing just too late
678 		 * and completing a request we've just cancelled, by ensuring
679 		 * that the seqno check therein fails.
680 		 */
681 		spin_lock_bh(&mcdi->iface_lock);
682 		++mcdi->seqno;
683 		++mcdi->credits;
684 		spin_unlock_bh(&mcdi->iface_lock);
685 	}
686 
687 	if (proxy_handle)
688 		*proxy_handle = 0;
689 
690 	if (rc != 0) {
691 		if (outlen_actual)
692 			*outlen_actual = 0;
693 	} else {
694 		size_t hdr_len, data_len, err_len;
695 
696 		/* At the very least we need a memory barrier here to ensure
697 		 * we pick up changes from efx_mcdi_ev_cpl(). Protect against
698 		 * a spurious efx_mcdi_ev_cpl() running concurrently by
699 		 * acquiring the iface_lock. */
700 		spin_lock_bh(&mcdi->iface_lock);
701 		rc = mcdi->resprc;
702 		if (raw_rc)
703 			*raw_rc = mcdi->resprc_raw;
704 		hdr_len = mcdi->resp_hdr_len;
705 		data_len = mcdi->resp_data_len;
706 		err_len = min(sizeof(errbuf), data_len);
707 		spin_unlock_bh(&mcdi->iface_lock);
708 
709 		BUG_ON(rc > 0);
710 
711 		efx->type->mcdi_read_response(efx, outbuf, hdr_len,
712 					      min(outlen, data_len));
713 		if (outlen_actual)
714 			*outlen_actual = data_len;
715 
716 		efx->type->mcdi_read_response(efx, errbuf, hdr_len, err_len);
717 
718 		if (cmd == MC_CMD_REBOOT && rc == -EIO) {
719 			/* Don't reset if MC_CMD_REBOOT returns EIO */
720 		} else if (rc == -EIO || rc == -EINTR) {
721 			netif_err(efx, hw, efx->net_dev, "MC reboot detected\n");
722 			netif_dbg(efx, hw, efx->net_dev, "MC rebooted during command %d rc %d\n",
723 				  cmd, -rc);
724 			if (efx->type->mcdi_reboot_detected)
725 				efx->type->mcdi_reboot_detected(efx);
726 			efx_schedule_reset(efx, RESET_TYPE_MC_FAILURE);
727 		} else if (proxy_handle && (rc == -EPROTO) &&
728 			   efx_mcdi_get_proxy_handle(efx, hdr_len, data_len,
729 						     proxy_handle)) {
730 			mcdi->proxy_rx_status = 0;
731 			mcdi->proxy_rx_handle = 0;
732 			mcdi->state = MCDI_STATE_PROXY_WAIT;
733 		} else if (rc && !quiet) {
734 			efx_mcdi_display_error(efx, cmd, inlen, errbuf, err_len,
735 					       rc);
736 		}
737 
738 		if (rc == -EIO || rc == -EINTR) {
739 			msleep(MCDI_STATUS_SLEEP_MS);
740 			efx_mcdi_poll_reboot(efx);
741 			mcdi->new_epoch = true;
742 		}
743 	}
744 
745 	if (!proxy_handle || !*proxy_handle)
746 		efx_mcdi_release(mcdi);
747 	return rc;
748 }
749 
750 static void efx_mcdi_proxy_abort(struct efx_mcdi_iface *mcdi)
751 {
752 	if (mcdi->state == MCDI_STATE_PROXY_WAIT) {
753 		/* Interrupt the proxy wait. */
754 		mcdi->proxy_rx_status = -EINTR;
755 		wake_up(&mcdi->proxy_rx_wq);
756 	}
757 }
758 
759 static void efx_mcdi_ev_proxy_response(struct efx_nic *efx,
760 				       u32 handle, int status)
761 {
762 	struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
763 
764 	WARN_ON(mcdi->state != MCDI_STATE_PROXY_WAIT);
765 
766 	mcdi->proxy_rx_status = efx_mcdi_errno(status);
767 	/* Ensure the status is written before we update the handle, since the
768 	 * latter is used to check if we've finished.
769 	 */
770 	wmb();
771 	mcdi->proxy_rx_handle = handle;
772 	wake_up(&mcdi->proxy_rx_wq);
773 }
774 
775 static int efx_mcdi_proxy_wait(struct efx_nic *efx, u32 handle, bool quiet)
776 {
777 	struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
778 	int rc;
779 
780 	/* Wait for a proxy event, or timeout. */
781 	rc = wait_event_timeout(mcdi->proxy_rx_wq,
782 				mcdi->proxy_rx_handle != 0 ||
783 				mcdi->proxy_rx_status == -EINTR,
784 				MCDI_RPC_TIMEOUT);
785 
786 	if (rc <= 0) {
787 		netif_dbg(efx, hw, efx->net_dev,
788 			  "MCDI proxy timeout %d\n", handle);
789 		return -ETIMEDOUT;
790 	} else if (mcdi->proxy_rx_handle != handle) {
791 		netif_warn(efx, hw, efx->net_dev,
792 			   "MCDI proxy unexpected handle %d (expected %d)\n",
793 			   mcdi->proxy_rx_handle, handle);
794 		return -EINVAL;
795 	}
796 
797 	return mcdi->proxy_rx_status;
798 }
799 
800 static int _efx_mcdi_rpc(struct efx_nic *efx, unsigned int cmd,
801 			 const efx_dword_t *inbuf, size_t inlen,
802 			 efx_dword_t *outbuf, size_t outlen,
803 			 size_t *outlen_actual, bool quiet, int *raw_rc)
804 {
805 	u32 proxy_handle = 0; /* Zero is an invalid proxy handle. */
806 	int rc;
807 
808 	if (inbuf && inlen && (inbuf == outbuf)) {
809 		/* The input buffer can't be aliased with the output. */
810 		WARN_ON(1);
811 		return -EINVAL;
812 	}
813 
814 	rc = efx_mcdi_rpc_start(efx, cmd, inbuf, inlen);
815 	if (rc)
816 		return rc;
817 
818 	rc = _efx_mcdi_rpc_finish(efx, cmd, inlen, outbuf, outlen,
819 				  outlen_actual, quiet, &proxy_handle, raw_rc);
820 
821 	if (proxy_handle) {
822 		/* Handle proxy authorisation. This allows approval of MCDI
823 		 * operations to be delegated to the admin function, allowing
824 		 * fine control over (eg) multicast subscriptions.
825 		 */
826 		struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
827 
828 		netif_dbg(efx, hw, efx->net_dev,
829 			  "MCDI waiting for proxy auth %d\n",
830 			  proxy_handle);
831 		rc = efx_mcdi_proxy_wait(efx, proxy_handle, quiet);
832 
833 		if (rc == 0) {
834 			netif_dbg(efx, hw, efx->net_dev,
835 				  "MCDI proxy retry %d\n", proxy_handle);
836 
837 			/* We now retry the original request. */
838 			mcdi->state = MCDI_STATE_RUNNING_SYNC;
839 			efx_mcdi_send_request(efx, cmd, inbuf, inlen);
840 
841 			rc = _efx_mcdi_rpc_finish(efx, cmd, inlen,
842 						  outbuf, outlen, outlen_actual,
843 						  quiet, NULL, raw_rc);
844 		} else {
845 			netif_cond_dbg(efx, hw, efx->net_dev, rc == -EPERM, err,
846 				       "MC command 0x%x failed after proxy auth rc=%d\n",
847 				       cmd, rc);
848 
849 			if (rc == -EINTR || rc == -EIO)
850 				efx_schedule_reset(efx, RESET_TYPE_MC_FAILURE);
851 			efx_mcdi_release(mcdi);
852 		}
853 	}
854 
855 	return rc;
856 }
857 
858 static int _efx_mcdi_rpc_evb_retry(struct efx_nic *efx, unsigned cmd,
859 				   const efx_dword_t *inbuf, size_t inlen,
860 				   efx_dword_t *outbuf, size_t outlen,
861 				   size_t *outlen_actual, bool quiet)
862 {
863 	int raw_rc = 0;
864 	int rc;
865 
866 	rc = _efx_mcdi_rpc(efx, cmd, inbuf, inlen,
867 			   outbuf, outlen, outlen_actual, true, &raw_rc);
868 
869 	if ((rc == -EPROTO) && (raw_rc == MC_CMD_ERR_NO_EVB_PORT) &&
870 	    efx->type->is_vf) {
871 		/* If the EVB port isn't available within a VF this may
872 		 * mean the PF is still bringing the switch up. We should
873 		 * retry our request shortly.
874 		 */
875 		unsigned long abort_time = jiffies + MCDI_RPC_TIMEOUT;
876 		unsigned int delay_us = 10000;
877 
878 		netif_dbg(efx, hw, efx->net_dev,
879 			  "%s: NO_EVB_PORT; will retry request\n",
880 			  __func__);
881 
882 		do {
883 			usleep_range(delay_us, delay_us + 10000);
884 			rc = _efx_mcdi_rpc(efx, cmd, inbuf, inlen,
885 					   outbuf, outlen, outlen_actual,
886 					   true, &raw_rc);
887 			if (delay_us < 100000)
888 				delay_us <<= 1;
889 		} while ((rc == -EPROTO) &&
890 			 (raw_rc == MC_CMD_ERR_NO_EVB_PORT) &&
891 			 time_before(jiffies, abort_time));
892 	}
893 
894 	if (rc && !quiet && !(cmd == MC_CMD_REBOOT && rc == -EIO))
895 		efx_mcdi_display_error(efx, cmd, inlen,
896 				       outbuf, outlen, rc);
897 
898 	return rc;
899 }
900 
901 /**
902  * efx_mcdi_rpc - Issue an MCDI command and wait for completion
903  * @efx: NIC through which to issue the command
904  * @cmd: Command type number
905  * @inbuf: Command parameters
906  * @inlen: Length of command parameters, in bytes.  Must be a multiple
907  *	of 4 and no greater than %MCDI_CTL_SDU_LEN_MAX_V1.
908  * @outbuf: Response buffer.  May be %NULL if @outlen is 0.
909  * @outlen: Length of response buffer, in bytes.  If the actual
910  *	response is longer than @outlen & ~3, it will be truncated
911  *	to that length.
912  * @outlen_actual: Pointer through which to return the actual response
913  *	length.  May be %NULL if this is not needed.
914  *
915  * This function may sleep and therefore must be called in an appropriate
916  * context.
917  *
918  * Return: A negative error code, or zero if successful.  The error
919  *	code may come from the MCDI response or may indicate a failure
920  *	to communicate with the MC.  In the former case, the response
921  *	will still be copied to @outbuf and *@outlen_actual will be
922  *	set accordingly.  In the latter case, *@outlen_actual will be
923  *	set to zero.
924  */
925 int efx_mcdi_rpc(struct efx_nic *efx, unsigned cmd,
926 		 const efx_dword_t *inbuf, size_t inlen,
927 		 efx_dword_t *outbuf, size_t outlen,
928 		 size_t *outlen_actual)
929 {
930 	return _efx_mcdi_rpc_evb_retry(efx, cmd, inbuf, inlen, outbuf, outlen,
931 				       outlen_actual, false);
932 }
933 
934 /* Normally, on receiving an error code in the MCDI response,
935  * efx_mcdi_rpc will log an error message containing (among other
936  * things) the raw error code, by means of efx_mcdi_display_error.
937  * This _quiet version suppresses that; if the caller wishes to log
938  * the error conditionally on the return code, it should call this
939  * function and is then responsible for calling efx_mcdi_display_error
940  * as needed.
941  */
942 int efx_mcdi_rpc_quiet(struct efx_nic *efx, unsigned cmd,
943 		       const efx_dword_t *inbuf, size_t inlen,
944 		       efx_dword_t *outbuf, size_t outlen,
945 		       size_t *outlen_actual)
946 {
947 	return _efx_mcdi_rpc_evb_retry(efx, cmd, inbuf, inlen, outbuf, outlen,
948 				       outlen_actual, true);
949 }
950 
951 int efx_mcdi_rpc_start(struct efx_nic *efx, unsigned cmd,
952 		       const efx_dword_t *inbuf, size_t inlen)
953 {
954 	struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
955 	int rc;
956 
957 	rc = efx_mcdi_check_supported(efx, cmd, inlen);
958 	if (rc)
959 		return rc;
960 
961 	if (efx->mc_bist_for_other_fn)
962 		return -ENETDOWN;
963 
964 	if (mcdi->mode == MCDI_MODE_FAIL)
965 		return -ENETDOWN;
966 
967 	efx_mcdi_acquire_sync(mcdi);
968 	efx_mcdi_send_request(efx, cmd, inbuf, inlen);
969 	return 0;
970 }
971 
972 static int _efx_mcdi_rpc_async(struct efx_nic *efx, unsigned int cmd,
973 			       const efx_dword_t *inbuf, size_t inlen,
974 			       size_t outlen,
975 			       efx_mcdi_async_completer *complete,
976 			       unsigned long cookie, bool quiet)
977 {
978 	struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
979 	struct efx_mcdi_async_param *async;
980 	int rc;
981 
982 	rc = efx_mcdi_check_supported(efx, cmd, inlen);
983 	if (rc)
984 		return rc;
985 
986 	if (efx->mc_bist_for_other_fn)
987 		return -ENETDOWN;
988 
989 	async = kmalloc(sizeof(*async) + ALIGN(max(inlen, outlen), 4),
990 			GFP_ATOMIC);
991 	if (!async)
992 		return -ENOMEM;
993 
994 	async->cmd = cmd;
995 	async->inlen = inlen;
996 	async->outlen = outlen;
997 	async->quiet = quiet;
998 	async->complete = complete;
999 	async->cookie = cookie;
1000 	memcpy(async + 1, inbuf, inlen);
1001 
1002 	spin_lock_bh(&mcdi->async_lock);
1003 
1004 	if (mcdi->mode == MCDI_MODE_EVENTS) {
1005 		list_add_tail(&async->list, &mcdi->async_list);
1006 
1007 		/* If this is at the front of the queue, try to start it
1008 		 * immediately
1009 		 */
1010 		if (mcdi->async_list.next == &async->list &&
1011 		    efx_mcdi_acquire_async(mcdi)) {
1012 			efx_mcdi_send_request(efx, cmd, inbuf, inlen);
1013 			mod_timer(&mcdi->async_timer,
1014 				  jiffies + MCDI_RPC_TIMEOUT);
1015 		}
1016 	} else {
1017 		kfree(async);
1018 		rc = -ENETDOWN;
1019 	}
1020 
1021 	spin_unlock_bh(&mcdi->async_lock);
1022 
1023 	return rc;
1024 }
1025 
1026 /**
1027  * efx_mcdi_rpc_async - Schedule an MCDI command to run asynchronously
1028  * @efx: NIC through which to issue the command
1029  * @cmd: Command type number
1030  * @inbuf: Command parameters
1031  * @inlen: Length of command parameters, in bytes
1032  * @outlen: Length to allocate for response buffer, in bytes
1033  * @complete: Function to be called on completion or cancellation.
1034  * @cookie: Arbitrary value to be passed to @complete.
1035  *
1036  * This function does not sleep and therefore may be called in atomic
1037  * context.  It will fail if event queues are disabled or if MCDI
1038  * event completions have been disabled due to an error.
1039  *
1040  * If it succeeds, the @complete function will be called exactly once
1041  * in atomic context, when one of the following occurs:
1042  * (a) the completion event is received (in NAPI context)
1043  * (b) event queues are disabled (in the process that disables them)
1044  * (c) the request times-out (in timer context)
1045  */
1046 int
1047 efx_mcdi_rpc_async(struct efx_nic *efx, unsigned int cmd,
1048 		   const efx_dword_t *inbuf, size_t inlen, size_t outlen,
1049 		   efx_mcdi_async_completer *complete, unsigned long cookie)
1050 {
1051 	return _efx_mcdi_rpc_async(efx, cmd, inbuf, inlen, outlen, complete,
1052 				   cookie, false);
1053 }
1054 
1055 int efx_mcdi_rpc_async_quiet(struct efx_nic *efx, unsigned int cmd,
1056 			     const efx_dword_t *inbuf, size_t inlen,
1057 			     size_t outlen, efx_mcdi_async_completer *complete,
1058 			     unsigned long cookie)
1059 {
1060 	return _efx_mcdi_rpc_async(efx, cmd, inbuf, inlen, outlen, complete,
1061 				   cookie, true);
1062 }
1063 
1064 int efx_mcdi_rpc_finish(struct efx_nic *efx, unsigned cmd, size_t inlen,
1065 			efx_dword_t *outbuf, size_t outlen,
1066 			size_t *outlen_actual)
1067 {
1068 	return _efx_mcdi_rpc_finish(efx, cmd, inlen, outbuf, outlen,
1069 				    outlen_actual, false, NULL, NULL);
1070 }
1071 
1072 int efx_mcdi_rpc_finish_quiet(struct efx_nic *efx, unsigned cmd, size_t inlen,
1073 			      efx_dword_t *outbuf, size_t outlen,
1074 			      size_t *outlen_actual)
1075 {
1076 	return _efx_mcdi_rpc_finish(efx, cmd, inlen, outbuf, outlen,
1077 				    outlen_actual, true, NULL, NULL);
1078 }
1079 
1080 void efx_mcdi_display_error(struct efx_nic *efx, unsigned cmd,
1081 			    size_t inlen, efx_dword_t *outbuf,
1082 			    size_t outlen, int rc)
1083 {
1084 	int code = 0, err_arg = 0;
1085 
1086 	if (outlen >= MC_CMD_ERR_CODE_OFST + 4)
1087 		code = MCDI_DWORD(outbuf, ERR_CODE);
1088 	if (outlen >= MC_CMD_ERR_ARG_OFST + 4)
1089 		err_arg = MCDI_DWORD(outbuf, ERR_ARG);
1090 	netif_cond_dbg(efx, hw, efx->net_dev, rc == -EPERM, err,
1091 		       "MC command 0x%x inlen %zu failed rc=%d (raw=%d) arg=%d\n",
1092 		       cmd, inlen, rc, code, err_arg);
1093 }
1094 
1095 /* Switch to polled MCDI completions.  This can be called in various
1096  * error conditions with various locks held, so it must be lockless.
1097  * Caller is responsible for flushing asynchronous requests later.
1098  */
1099 void efx_mcdi_mode_poll(struct efx_nic *efx)
1100 {
1101 	struct efx_mcdi_iface *mcdi;
1102 
1103 	if (!efx->mcdi)
1104 		return;
1105 
1106 	mcdi = efx_mcdi(efx);
1107 	/* If already in polling mode, nothing to do.
1108 	 * If in fail-fast state, don't switch to polled completion.
1109 	 * FLR recovery will do that later.
1110 	 */
1111 	if (mcdi->mode == MCDI_MODE_POLL || mcdi->mode == MCDI_MODE_FAIL)
1112 		return;
1113 
1114 	/* We can switch from event completion to polled completion, because
1115 	 * mcdi requests are always completed in shared memory. We do this by
1116 	 * switching the mode to POLL'd then completing the request.
1117 	 * efx_mcdi_await_completion() will then call efx_mcdi_poll().
1118 	 *
1119 	 * We need an smp_wmb() to synchronise with efx_mcdi_await_completion(),
1120 	 * which efx_mcdi_complete_sync() provides for us.
1121 	 */
1122 	mcdi->mode = MCDI_MODE_POLL;
1123 
1124 	efx_mcdi_complete_sync(mcdi);
1125 }
1126 
1127 /* Flush any running or queued asynchronous requests, after event processing
1128  * is stopped
1129  */
1130 void efx_mcdi_flush_async(struct efx_nic *efx)
1131 {
1132 	struct efx_mcdi_async_param *async, *next;
1133 	struct efx_mcdi_iface *mcdi;
1134 
1135 	if (!efx->mcdi)
1136 		return;
1137 
1138 	mcdi = efx_mcdi(efx);
1139 
1140 	/* We must be in poll or fail mode so no more requests can be queued */
1141 	BUG_ON(mcdi->mode == MCDI_MODE_EVENTS);
1142 
1143 	del_timer_sync(&mcdi->async_timer);
1144 
1145 	/* If a request is still running, make sure we give the MC
1146 	 * time to complete it so that the response won't overwrite our
1147 	 * next request.
1148 	 */
1149 	if (mcdi->state == MCDI_STATE_RUNNING_ASYNC) {
1150 		efx_mcdi_poll(efx);
1151 		mcdi->state = MCDI_STATE_QUIESCENT;
1152 	}
1153 
1154 	/* Nothing else will access the async list now, so it is safe
1155 	 * to walk it without holding async_lock.  If we hold it while
1156 	 * calling a completer then lockdep may warn that we have
1157 	 * acquired locks in the wrong order.
1158 	 */
1159 	list_for_each_entry_safe(async, next, &mcdi->async_list, list) {
1160 		if (async->complete)
1161 			async->complete(efx, async->cookie, -ENETDOWN, NULL, 0);
1162 		list_del(&async->list);
1163 		kfree(async);
1164 	}
1165 }
1166 
1167 void efx_mcdi_mode_event(struct efx_nic *efx)
1168 {
1169 	struct efx_mcdi_iface *mcdi;
1170 
1171 	if (!efx->mcdi)
1172 		return;
1173 
1174 	mcdi = efx_mcdi(efx);
1175 	/* If already in event completion mode, nothing to do.
1176 	 * If in fail-fast state, don't switch to event completion.  FLR
1177 	 * recovery will do that later.
1178 	 */
1179 	if (mcdi->mode == MCDI_MODE_EVENTS || mcdi->mode == MCDI_MODE_FAIL)
1180 		return;
1181 
1182 	/* We can't switch from polled to event completion in the middle of a
1183 	 * request, because the completion method is specified in the request.
1184 	 * So acquire the interface to serialise the requestors. We don't need
1185 	 * to acquire the iface_lock to change the mode here, but we do need a
1186 	 * write memory barrier ensure that efx_mcdi_rpc() sees it, which
1187 	 * efx_mcdi_acquire() provides.
1188 	 */
1189 	efx_mcdi_acquire_sync(mcdi);
1190 	mcdi->mode = MCDI_MODE_EVENTS;
1191 	efx_mcdi_release(mcdi);
1192 }
1193 
1194 static void efx_mcdi_ev_death(struct efx_nic *efx, int rc)
1195 {
1196 	struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
1197 
1198 	/* If there is an outstanding MCDI request, it has been terminated
1199 	 * either by a BADASSERT or REBOOT event. If the mcdi interface is
1200 	 * in polled mode, then do nothing because the MC reboot handler will
1201 	 * set the header correctly. However, if the mcdi interface is waiting
1202 	 * for a CMDDONE event it won't receive it [and since all MCDI events
1203 	 * are sent to the same queue, we can't be racing with
1204 	 * efx_mcdi_ev_cpl()]
1205 	 *
1206 	 * If there is an outstanding asynchronous request, we can't
1207 	 * complete it now (efx_mcdi_complete() would deadlock).  The
1208 	 * reset process will take care of this.
1209 	 *
1210 	 * There's a race here with efx_mcdi_send_request(), because
1211 	 * we might receive a REBOOT event *before* the request has
1212 	 * been copied out. In polled mode (during startup) this is
1213 	 * irrelevant, because efx_mcdi_complete_sync() is ignored. In
1214 	 * event mode, this condition is just an edge-case of
1215 	 * receiving a REBOOT event after posting the MCDI
1216 	 * request. Did the mc reboot before or after the copyout? The
1217 	 * best we can do always is just return failure.
1218 	 *
1219 	 * If there is an outstanding proxy response expected it is not going
1220 	 * to arrive. We should thus abort it.
1221 	 */
1222 	spin_lock(&mcdi->iface_lock);
1223 	efx_mcdi_proxy_abort(mcdi);
1224 
1225 	if (efx_mcdi_complete_sync(mcdi)) {
1226 		if (mcdi->mode == MCDI_MODE_EVENTS) {
1227 			mcdi->resprc = rc;
1228 			mcdi->resp_hdr_len = 0;
1229 			mcdi->resp_data_len = 0;
1230 			++mcdi->credits;
1231 		}
1232 	} else {
1233 		int count;
1234 
1235 		/* Consume the status word since efx_mcdi_rpc_finish() won't */
1236 		for (count = 0; count < MCDI_STATUS_DELAY_COUNT; ++count) {
1237 			rc = efx_mcdi_poll_reboot(efx);
1238 			if (rc)
1239 				break;
1240 			udelay(MCDI_STATUS_DELAY_US);
1241 		}
1242 
1243 		/* On EF10, a CODE_MC_REBOOT event can be received without the
1244 		 * reboot detection in efx_mcdi_poll_reboot() being triggered.
1245 		 * If zero was returned from the final call to
1246 		 * efx_mcdi_poll_reboot(), the MC reboot wasn't noticed but the
1247 		 * MC has definitely rebooted so prepare for the reset.
1248 		 */
1249 		if (!rc && efx->type->mcdi_reboot_detected)
1250 			efx->type->mcdi_reboot_detected(efx);
1251 
1252 		mcdi->new_epoch = true;
1253 
1254 		/* Nobody was waiting for an MCDI request, so trigger a reset */
1255 		efx_schedule_reset(efx, RESET_TYPE_MC_FAILURE);
1256 	}
1257 
1258 	spin_unlock(&mcdi->iface_lock);
1259 }
1260 
1261 /* The MC is going down in to BIST mode. set the BIST flag to block
1262  * new MCDI, cancel any outstanding MCDI and schedule a BIST-type reset
1263  * (which doesn't actually execute a reset, it waits for the controlling
1264  * function to reset it).
1265  */
1266 static void efx_mcdi_ev_bist(struct efx_nic *efx)
1267 {
1268 	struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
1269 
1270 	spin_lock(&mcdi->iface_lock);
1271 	efx->mc_bist_for_other_fn = true;
1272 	efx_mcdi_proxy_abort(mcdi);
1273 
1274 	if (efx_mcdi_complete_sync(mcdi)) {
1275 		if (mcdi->mode == MCDI_MODE_EVENTS) {
1276 			mcdi->resprc = -EIO;
1277 			mcdi->resp_hdr_len = 0;
1278 			mcdi->resp_data_len = 0;
1279 			++mcdi->credits;
1280 		}
1281 	}
1282 	mcdi->new_epoch = true;
1283 	efx_schedule_reset(efx, RESET_TYPE_MC_BIST);
1284 	spin_unlock(&mcdi->iface_lock);
1285 }
1286 
1287 /* MCDI timeouts seen, so make all MCDI calls fail-fast and issue an FLR to try
1288  * to recover.
1289  */
1290 static void efx_mcdi_abandon(struct efx_nic *efx)
1291 {
1292 	struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
1293 
1294 	if (xchg(&mcdi->mode, MCDI_MODE_FAIL) == MCDI_MODE_FAIL)
1295 		return; /* it had already been done */
1296 	netif_dbg(efx, hw, efx->net_dev, "MCDI is timing out; trying to recover\n");
1297 	efx_schedule_reset(efx, RESET_TYPE_MCDI_TIMEOUT);
1298 }
1299 
1300 static void efx_handle_drain_event(struct efx_nic *efx)
1301 {
1302 	if (atomic_dec_and_test(&efx->active_queues))
1303 		wake_up(&efx->flush_wq);
1304 
1305 	WARN_ON(atomic_read(&efx->active_queues) < 0);
1306 }
1307 
1308 /* Called from efx_farch_ev_process and efx_ef10_ev_process for MCDI events */
1309 void efx_mcdi_process_event(struct efx_channel *channel,
1310 			    efx_qword_t *event)
1311 {
1312 	struct efx_nic *efx = channel->efx;
1313 	int code = EFX_QWORD_FIELD(*event, MCDI_EVENT_CODE);
1314 	u32 data = EFX_QWORD_FIELD(*event, MCDI_EVENT_DATA);
1315 
1316 	switch (code) {
1317 	case MCDI_EVENT_CODE_BADSSERT:
1318 		netif_err(efx, hw, efx->net_dev,
1319 			  "MC watchdog or assertion failure at 0x%x\n", data);
1320 		efx_mcdi_ev_death(efx, -EINTR);
1321 		break;
1322 
1323 	case MCDI_EVENT_CODE_PMNOTICE:
1324 		netif_info(efx, wol, efx->net_dev, "MCDI PM event.\n");
1325 		break;
1326 
1327 	case MCDI_EVENT_CODE_CMDDONE:
1328 		efx_mcdi_ev_cpl(efx,
1329 				MCDI_EVENT_FIELD(*event, CMDDONE_SEQ),
1330 				MCDI_EVENT_FIELD(*event, CMDDONE_DATALEN),
1331 				MCDI_EVENT_FIELD(*event, CMDDONE_ERRNO));
1332 		break;
1333 
1334 	case MCDI_EVENT_CODE_LINKCHANGE:
1335 		efx_mcdi_process_link_change(efx, event);
1336 		break;
1337 	case MCDI_EVENT_CODE_SENSOREVT:
1338 		efx_sensor_event(efx, event);
1339 		break;
1340 	case MCDI_EVENT_CODE_SCHEDERR:
1341 		netif_dbg(efx, hw, efx->net_dev,
1342 			  "MC Scheduler alert (0x%x)\n", data);
1343 		break;
1344 	case MCDI_EVENT_CODE_REBOOT:
1345 	case MCDI_EVENT_CODE_MC_REBOOT:
1346 		netif_info(efx, hw, efx->net_dev, "MC Reboot\n");
1347 		efx_mcdi_ev_death(efx, -EIO);
1348 		break;
1349 	case MCDI_EVENT_CODE_MC_BIST:
1350 		netif_info(efx, hw, efx->net_dev, "MC entered BIST mode\n");
1351 		efx_mcdi_ev_bist(efx);
1352 		break;
1353 	case MCDI_EVENT_CODE_MAC_STATS_DMA:
1354 		/* MAC stats are gather lazily.  We can ignore this. */
1355 		break;
1356 	case MCDI_EVENT_CODE_FLR:
1357 		if (efx->type->sriov_flr)
1358 			efx->type->sriov_flr(efx,
1359 					     MCDI_EVENT_FIELD(*event, FLR_VF));
1360 		break;
1361 	case MCDI_EVENT_CODE_PTP_RX:
1362 	case MCDI_EVENT_CODE_PTP_FAULT:
1363 	case MCDI_EVENT_CODE_PTP_PPS:
1364 		efx_ptp_event(efx, event);
1365 		break;
1366 	case MCDI_EVENT_CODE_PTP_TIME:
1367 		efx_time_sync_event(channel, event);
1368 		break;
1369 	case MCDI_EVENT_CODE_TX_FLUSH:
1370 	case MCDI_EVENT_CODE_RX_FLUSH:
1371 		/* Two flush events will be sent: one to the same event
1372 		 * queue as completions, and one to event queue 0.
1373 		 * In the latter case the {RX,TX}_FLUSH_TO_DRIVER
1374 		 * flag will be set, and we should ignore the event
1375 		 * because we want to wait for all completions.
1376 		 */
1377 		BUILD_BUG_ON(MCDI_EVENT_TX_FLUSH_TO_DRIVER_LBN !=
1378 			     MCDI_EVENT_RX_FLUSH_TO_DRIVER_LBN);
1379 		if (!MCDI_EVENT_FIELD(*event, TX_FLUSH_TO_DRIVER))
1380 			efx_handle_drain_event(efx);
1381 		break;
1382 	case MCDI_EVENT_CODE_TX_ERR:
1383 	case MCDI_EVENT_CODE_RX_ERR:
1384 		netif_err(efx, hw, efx->net_dev,
1385 			  "%s DMA error (event: "EFX_QWORD_FMT")\n",
1386 			  code == MCDI_EVENT_CODE_TX_ERR ? "TX" : "RX",
1387 			  EFX_QWORD_VAL(*event));
1388 		efx_schedule_reset(efx, RESET_TYPE_DMA_ERROR);
1389 		break;
1390 	case MCDI_EVENT_CODE_PROXY_RESPONSE:
1391 		efx_mcdi_ev_proxy_response(efx,
1392 				MCDI_EVENT_FIELD(*event, PROXY_RESPONSE_HANDLE),
1393 				MCDI_EVENT_FIELD(*event, PROXY_RESPONSE_RC));
1394 		break;
1395 	default:
1396 		netif_err(efx, hw, efx->net_dev,
1397 			  "Unknown MCDI event " EFX_QWORD_FMT "\n",
1398 			  EFX_QWORD_VAL(*event));
1399 	}
1400 }
1401 
1402 /**************************************************************************
1403  *
1404  * Specific request functions
1405  *
1406  **************************************************************************
1407  */
1408 
1409 void efx_mcdi_print_fwver(struct efx_nic *efx, char *buf, size_t len)
1410 {
1411 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_VERSION_OUT_LEN);
1412 	size_t outlength;
1413 	const __le16 *ver_words;
1414 	size_t offset;
1415 	int rc;
1416 
1417 	BUILD_BUG_ON(MC_CMD_GET_VERSION_IN_LEN != 0);
1418 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_VERSION, NULL, 0,
1419 			  outbuf, sizeof(outbuf), &outlength);
1420 	if (rc)
1421 		goto fail;
1422 	if (outlength < MC_CMD_GET_VERSION_OUT_LEN) {
1423 		rc = -EIO;
1424 		goto fail;
1425 	}
1426 
1427 	ver_words = (__le16 *)MCDI_PTR(outbuf, GET_VERSION_OUT_VERSION);
1428 	offset = scnprintf(buf, len, "%u.%u.%u.%u",
1429 			   le16_to_cpu(ver_words[0]),
1430 			   le16_to_cpu(ver_words[1]),
1431 			   le16_to_cpu(ver_words[2]),
1432 			   le16_to_cpu(ver_words[3]));
1433 
1434 	if (efx->type->print_additional_fwver)
1435 		offset += efx->type->print_additional_fwver(efx, buf + offset,
1436 							    len - offset);
1437 
1438 	/* It's theoretically possible for the string to exceed 31
1439 	 * characters, though in practice the first three version
1440 	 * components are short enough that this doesn't happen.
1441 	 */
1442 	if (WARN_ON(offset >= len))
1443 		buf[0] = 0;
1444 
1445 	return;
1446 
1447 fail:
1448 	pci_err(efx->pci_dev, "%s: failed rc=%d\n", __func__, rc);
1449 	buf[0] = 0;
1450 }
1451 
1452 static int efx_mcdi_drv_attach(struct efx_nic *efx, bool driver_operating,
1453 			       bool *was_attached)
1454 {
1455 	MCDI_DECLARE_BUF(inbuf, MC_CMD_DRV_ATTACH_IN_LEN);
1456 	MCDI_DECLARE_BUF(outbuf, MC_CMD_DRV_ATTACH_EXT_OUT_LEN);
1457 	size_t outlen;
1458 	int rc;
1459 
1460 	MCDI_SET_DWORD(inbuf, DRV_ATTACH_IN_NEW_STATE,
1461 		       driver_operating ? 1 : 0);
1462 	MCDI_SET_DWORD(inbuf, DRV_ATTACH_IN_UPDATE, 1);
1463 	MCDI_SET_DWORD(inbuf, DRV_ATTACH_IN_FIRMWARE_ID, MC_CMD_FW_LOW_LATENCY);
1464 
1465 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_DRV_ATTACH, inbuf, sizeof(inbuf),
1466 				outbuf, sizeof(outbuf), &outlen);
1467 	/* If we're not the primary PF, trying to ATTACH with a FIRMWARE_ID
1468 	 * specified will fail with EPERM, and we have to tell the MC we don't
1469 	 * care what firmware we get.
1470 	 */
1471 	if (rc == -EPERM) {
1472 		pci_dbg(efx->pci_dev,
1473 			"%s with fw-variant setting failed EPERM, trying without it\n",
1474 			__func__);
1475 		MCDI_SET_DWORD(inbuf, DRV_ATTACH_IN_FIRMWARE_ID,
1476 			       MC_CMD_FW_DONT_CARE);
1477 		rc = efx_mcdi_rpc_quiet(efx, MC_CMD_DRV_ATTACH, inbuf,
1478 					sizeof(inbuf), outbuf, sizeof(outbuf),
1479 					&outlen);
1480 	}
1481 	if (rc) {
1482 		efx_mcdi_display_error(efx, MC_CMD_DRV_ATTACH, sizeof(inbuf),
1483 				       outbuf, outlen, rc);
1484 		goto fail;
1485 	}
1486 	if (outlen < MC_CMD_DRV_ATTACH_OUT_LEN) {
1487 		rc = -EIO;
1488 		goto fail;
1489 	}
1490 
1491 	if (driver_operating) {
1492 		if (outlen >= MC_CMD_DRV_ATTACH_EXT_OUT_LEN) {
1493 			efx->mcdi->fn_flags =
1494 				MCDI_DWORD(outbuf,
1495 					   DRV_ATTACH_EXT_OUT_FUNC_FLAGS);
1496 		} else {
1497 			/* Synthesise flags for Siena */
1498 			efx->mcdi->fn_flags =
1499 				1 << MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_LINKCTRL |
1500 				1 << MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_TRUSTED |
1501 				(efx_port_num(efx) == 0) <<
1502 				MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_PRIMARY;
1503 		}
1504 	}
1505 
1506 	/* We currently assume we have control of the external link
1507 	 * and are completely trusted by firmware.  Abort probing
1508 	 * if that's not true for this function.
1509 	 */
1510 
1511 	if (was_attached != NULL)
1512 		*was_attached = MCDI_DWORD(outbuf, DRV_ATTACH_OUT_OLD_STATE);
1513 	return 0;
1514 
1515 fail:
1516 	pci_err(efx->pci_dev, "%s: failed rc=%d\n", __func__, rc);
1517 	return rc;
1518 }
1519 
1520 int efx_mcdi_get_board_cfg(struct efx_nic *efx, u8 *mac_address,
1521 			   u16 *fw_subtype_list, u32 *capabilities)
1522 {
1523 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_BOARD_CFG_OUT_LENMAX);
1524 	size_t outlen, i;
1525 	int port_num = efx_port_num(efx);
1526 	int rc;
1527 
1528 	BUILD_BUG_ON(MC_CMD_GET_BOARD_CFG_IN_LEN != 0);
1529 	/* we need __aligned(2) for ether_addr_copy */
1530 	BUILD_BUG_ON(MC_CMD_GET_BOARD_CFG_OUT_MAC_ADDR_BASE_PORT0_OFST & 1);
1531 	BUILD_BUG_ON(MC_CMD_GET_BOARD_CFG_OUT_MAC_ADDR_BASE_PORT1_OFST & 1);
1532 
1533 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_BOARD_CFG, NULL, 0,
1534 			  outbuf, sizeof(outbuf), &outlen);
1535 	if (rc)
1536 		goto fail;
1537 
1538 	if (outlen < MC_CMD_GET_BOARD_CFG_OUT_LENMIN) {
1539 		rc = -EIO;
1540 		goto fail;
1541 	}
1542 
1543 	if (mac_address)
1544 		ether_addr_copy(mac_address,
1545 				port_num ?
1546 				MCDI_PTR(outbuf, GET_BOARD_CFG_OUT_MAC_ADDR_BASE_PORT1) :
1547 				MCDI_PTR(outbuf, GET_BOARD_CFG_OUT_MAC_ADDR_BASE_PORT0));
1548 	if (fw_subtype_list) {
1549 		for (i = 0;
1550 		     i < MCDI_VAR_ARRAY_LEN(outlen,
1551 					    GET_BOARD_CFG_OUT_FW_SUBTYPE_LIST);
1552 		     i++)
1553 			fw_subtype_list[i] = MCDI_ARRAY_WORD(
1554 				outbuf, GET_BOARD_CFG_OUT_FW_SUBTYPE_LIST, i);
1555 		for (; i < MC_CMD_GET_BOARD_CFG_OUT_FW_SUBTYPE_LIST_MAXNUM; i++)
1556 			fw_subtype_list[i] = 0;
1557 	}
1558 	if (capabilities) {
1559 		if (port_num)
1560 			*capabilities = MCDI_DWORD(outbuf,
1561 					GET_BOARD_CFG_OUT_CAPABILITIES_PORT1);
1562 		else
1563 			*capabilities = MCDI_DWORD(outbuf,
1564 					GET_BOARD_CFG_OUT_CAPABILITIES_PORT0);
1565 	}
1566 
1567 	return 0;
1568 
1569 fail:
1570 	netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d len=%d\n",
1571 		  __func__, rc, (int)outlen);
1572 
1573 	return rc;
1574 }
1575 
1576 int efx_mcdi_log_ctrl(struct efx_nic *efx, bool evq, bool uart, u32 dest_evq)
1577 {
1578 	MCDI_DECLARE_BUF(inbuf, MC_CMD_LOG_CTRL_IN_LEN);
1579 	u32 dest = 0;
1580 	int rc;
1581 
1582 	if (uart)
1583 		dest |= MC_CMD_LOG_CTRL_IN_LOG_DEST_UART;
1584 	if (evq)
1585 		dest |= MC_CMD_LOG_CTRL_IN_LOG_DEST_EVQ;
1586 
1587 	MCDI_SET_DWORD(inbuf, LOG_CTRL_IN_LOG_DEST, dest);
1588 	MCDI_SET_DWORD(inbuf, LOG_CTRL_IN_LOG_DEST_EVQ, dest_evq);
1589 
1590 	BUILD_BUG_ON(MC_CMD_LOG_CTRL_OUT_LEN != 0);
1591 
1592 	rc = efx_mcdi_rpc(efx, MC_CMD_LOG_CTRL, inbuf, sizeof(inbuf),
1593 			  NULL, 0, NULL);
1594 	return rc;
1595 }
1596 
1597 int efx_mcdi_nvram_types(struct efx_nic *efx, u32 *nvram_types_out)
1598 {
1599 	MCDI_DECLARE_BUF(outbuf, MC_CMD_NVRAM_TYPES_OUT_LEN);
1600 	size_t outlen;
1601 	int rc;
1602 
1603 	BUILD_BUG_ON(MC_CMD_NVRAM_TYPES_IN_LEN != 0);
1604 
1605 	rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_TYPES, NULL, 0,
1606 			  outbuf, sizeof(outbuf), &outlen);
1607 	if (rc)
1608 		goto fail;
1609 	if (outlen < MC_CMD_NVRAM_TYPES_OUT_LEN) {
1610 		rc = -EIO;
1611 		goto fail;
1612 	}
1613 
1614 	*nvram_types_out = MCDI_DWORD(outbuf, NVRAM_TYPES_OUT_TYPES);
1615 	return 0;
1616 
1617 fail:
1618 	netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n",
1619 		  __func__, rc);
1620 	return rc;
1621 }
1622 
1623 /* This function finds types using the new NVRAM_PARTITIONS mcdi. */
1624 static int efx_new_mcdi_nvram_types(struct efx_nic *efx, u32 *number,
1625 				    u32 *nvram_types)
1626 {
1627 	efx_dword_t *outbuf = kzalloc(MC_CMD_NVRAM_PARTITIONS_OUT_LENMAX_MCDI2,
1628 				      GFP_KERNEL);
1629 	size_t outlen;
1630 	int rc;
1631 
1632 	if (!outbuf)
1633 		return -ENOMEM;
1634 
1635 	BUILD_BUG_ON(MC_CMD_NVRAM_PARTITIONS_IN_LEN != 0);
1636 
1637 	rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_PARTITIONS, NULL, 0,
1638 			  outbuf, MC_CMD_NVRAM_PARTITIONS_OUT_LENMAX_MCDI2, &outlen);
1639 	if (rc)
1640 		goto fail;
1641 
1642 	*number = MCDI_DWORD(outbuf, NVRAM_PARTITIONS_OUT_NUM_PARTITIONS);
1643 
1644 	memcpy(nvram_types, MCDI_PTR(outbuf, NVRAM_PARTITIONS_OUT_TYPE_ID),
1645 	       *number * sizeof(u32));
1646 
1647 fail:
1648 	kfree(outbuf);
1649 	return rc;
1650 }
1651 
1652 int efx_mcdi_nvram_info(struct efx_nic *efx, unsigned int type,
1653 			size_t *size_out, size_t *erase_size_out,
1654 			bool *protected_out)
1655 {
1656 	MCDI_DECLARE_BUF(inbuf, MC_CMD_NVRAM_INFO_IN_LEN);
1657 	MCDI_DECLARE_BUF(outbuf, MC_CMD_NVRAM_INFO_OUT_LEN);
1658 	size_t outlen;
1659 	int rc;
1660 
1661 	MCDI_SET_DWORD(inbuf, NVRAM_INFO_IN_TYPE, type);
1662 
1663 	rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_INFO, inbuf, sizeof(inbuf),
1664 			  outbuf, sizeof(outbuf), &outlen);
1665 	if (rc)
1666 		goto fail;
1667 	if (outlen < MC_CMD_NVRAM_INFO_OUT_LEN) {
1668 		rc = -EIO;
1669 		goto fail;
1670 	}
1671 
1672 	*size_out = MCDI_DWORD(outbuf, NVRAM_INFO_OUT_SIZE);
1673 	*erase_size_out = MCDI_DWORD(outbuf, NVRAM_INFO_OUT_ERASESIZE);
1674 	*protected_out = !!(MCDI_DWORD(outbuf, NVRAM_INFO_OUT_FLAGS) &
1675 				(1 << MC_CMD_NVRAM_INFO_OUT_PROTECTED_LBN));
1676 	return 0;
1677 
1678 fail:
1679 	netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
1680 	return rc;
1681 }
1682 
1683 static int efx_mcdi_nvram_test(struct efx_nic *efx, unsigned int type)
1684 {
1685 	MCDI_DECLARE_BUF(inbuf, MC_CMD_NVRAM_TEST_IN_LEN);
1686 	MCDI_DECLARE_BUF(outbuf, MC_CMD_NVRAM_TEST_OUT_LEN);
1687 	int rc;
1688 
1689 	MCDI_SET_DWORD(inbuf, NVRAM_TEST_IN_TYPE, type);
1690 
1691 	rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_TEST, inbuf, sizeof(inbuf),
1692 			  outbuf, sizeof(outbuf), NULL);
1693 	if (rc)
1694 		return rc;
1695 
1696 	switch (MCDI_DWORD(outbuf, NVRAM_TEST_OUT_RESULT)) {
1697 	case MC_CMD_NVRAM_TEST_PASS:
1698 	case MC_CMD_NVRAM_TEST_NOTSUPP:
1699 		return 0;
1700 	default:
1701 		return -EIO;
1702 	}
1703 }
1704 
1705 /* This function tests nvram partitions using the new mcdi partition lookup scheme */
1706 int efx_new_mcdi_nvram_test_all(struct efx_nic *efx)
1707 {
1708 	u32 *nvram_types = kzalloc(MC_CMD_NVRAM_PARTITIONS_OUT_LENMAX_MCDI2,
1709 				   GFP_KERNEL);
1710 	unsigned int number;
1711 	int rc, i;
1712 
1713 	if (!nvram_types)
1714 		return -ENOMEM;
1715 
1716 	rc = efx_new_mcdi_nvram_types(efx, &number, nvram_types);
1717 	if (rc)
1718 		goto fail;
1719 
1720 	/* Require at least one check */
1721 	rc = -EAGAIN;
1722 
1723 	for (i = 0; i < number; i++) {
1724 		if (nvram_types[i] == NVRAM_PARTITION_TYPE_PARTITION_MAP ||
1725 		    nvram_types[i] == NVRAM_PARTITION_TYPE_DYNAMIC_CONFIG)
1726 			continue;
1727 
1728 		rc = efx_mcdi_nvram_test(efx, nvram_types[i]);
1729 		if (rc)
1730 			goto fail;
1731 	}
1732 
1733 fail:
1734 	kfree(nvram_types);
1735 	return rc;
1736 }
1737 
1738 int efx_mcdi_nvram_test_all(struct efx_nic *efx)
1739 {
1740 	u32 nvram_types;
1741 	unsigned int type;
1742 	int rc;
1743 
1744 	rc = efx_mcdi_nvram_types(efx, &nvram_types);
1745 	if (rc)
1746 		goto fail1;
1747 
1748 	type = 0;
1749 	while (nvram_types != 0) {
1750 		if (nvram_types & 1) {
1751 			rc = efx_mcdi_nvram_test(efx, type);
1752 			if (rc)
1753 				goto fail2;
1754 		}
1755 		type++;
1756 		nvram_types >>= 1;
1757 	}
1758 
1759 	return 0;
1760 
1761 fail2:
1762 	netif_err(efx, hw, efx->net_dev, "%s: failed type=%u\n",
1763 		  __func__, type);
1764 fail1:
1765 	netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
1766 	return rc;
1767 }
1768 
1769 /* Returns 1 if an assertion was read, 0 if no assertion had fired,
1770  * negative on error.
1771  */
1772 static int efx_mcdi_read_assertion(struct efx_nic *efx)
1773 {
1774 	MCDI_DECLARE_BUF(inbuf, MC_CMD_GET_ASSERTS_IN_LEN);
1775 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_ASSERTS_OUT_LEN);
1776 	unsigned int flags, index;
1777 	const char *reason;
1778 	size_t outlen;
1779 	int retry;
1780 	int rc;
1781 
1782 	/* Attempt to read any stored assertion state before we reboot
1783 	 * the mcfw out of the assertion handler. Retry twice, once
1784 	 * because a boot-time assertion might cause this command to fail
1785 	 * with EINTR. And once again because GET_ASSERTS can race with
1786 	 * MC_CMD_REBOOT running on the other port. */
1787 	retry = 2;
1788 	do {
1789 		MCDI_SET_DWORD(inbuf, GET_ASSERTS_IN_CLEAR, 1);
1790 		rc = efx_mcdi_rpc_quiet(efx, MC_CMD_GET_ASSERTS,
1791 					inbuf, MC_CMD_GET_ASSERTS_IN_LEN,
1792 					outbuf, sizeof(outbuf), &outlen);
1793 		if (rc == -EPERM)
1794 			return 0;
1795 	} while ((rc == -EINTR || rc == -EIO) && retry-- > 0);
1796 
1797 	if (rc) {
1798 		efx_mcdi_display_error(efx, MC_CMD_GET_ASSERTS,
1799 				       MC_CMD_GET_ASSERTS_IN_LEN, outbuf,
1800 				       outlen, rc);
1801 		return rc;
1802 	}
1803 	if (outlen < MC_CMD_GET_ASSERTS_OUT_LEN)
1804 		return -EIO;
1805 
1806 	/* Print out any recorded assertion state */
1807 	flags = MCDI_DWORD(outbuf, GET_ASSERTS_OUT_GLOBAL_FLAGS);
1808 	if (flags == MC_CMD_GET_ASSERTS_FLAGS_NO_FAILS)
1809 		return 0;
1810 
1811 	reason = (flags == MC_CMD_GET_ASSERTS_FLAGS_SYS_FAIL)
1812 		? "system-level assertion"
1813 		: (flags == MC_CMD_GET_ASSERTS_FLAGS_THR_FAIL)
1814 		? "thread-level assertion"
1815 		: (flags == MC_CMD_GET_ASSERTS_FLAGS_WDOG_FIRED)
1816 		? "watchdog reset"
1817 		: "unknown assertion";
1818 	netif_err(efx, hw, efx->net_dev,
1819 		  "MCPU %s at PC = 0x%.8x in thread 0x%.8x\n", reason,
1820 		  MCDI_DWORD(outbuf, GET_ASSERTS_OUT_SAVED_PC_OFFS),
1821 		  MCDI_DWORD(outbuf, GET_ASSERTS_OUT_THREAD_OFFS));
1822 
1823 	/* Print out the registers */
1824 	for (index = 0;
1825 	     index < MC_CMD_GET_ASSERTS_OUT_GP_REGS_OFFS_NUM;
1826 	     index++)
1827 		netif_err(efx, hw, efx->net_dev, "R%.2d (?): 0x%.8x\n",
1828 			  1 + index,
1829 			  MCDI_ARRAY_DWORD(outbuf, GET_ASSERTS_OUT_GP_REGS_OFFS,
1830 					   index));
1831 
1832 	return 1;
1833 }
1834 
1835 static int efx_mcdi_exit_assertion(struct efx_nic *efx)
1836 {
1837 	MCDI_DECLARE_BUF(inbuf, MC_CMD_REBOOT_IN_LEN);
1838 	int rc;
1839 
1840 	/* If the MC is running debug firmware, it might now be
1841 	 * waiting for a debugger to attach, but we just want it to
1842 	 * reboot.  We set a flag that makes the command a no-op if it
1843 	 * has already done so.
1844 	 * The MCDI will thus return either 0 or -EIO.
1845 	 */
1846 	BUILD_BUG_ON(MC_CMD_REBOOT_OUT_LEN != 0);
1847 	MCDI_SET_DWORD(inbuf, REBOOT_IN_FLAGS,
1848 		       MC_CMD_REBOOT_FLAGS_AFTER_ASSERTION);
1849 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_REBOOT, inbuf, MC_CMD_REBOOT_IN_LEN,
1850 				NULL, 0, NULL);
1851 	if (rc == -EIO)
1852 		rc = 0;
1853 	if (rc)
1854 		efx_mcdi_display_error(efx, MC_CMD_REBOOT, MC_CMD_REBOOT_IN_LEN,
1855 				       NULL, 0, rc);
1856 	return rc;
1857 }
1858 
1859 int efx_mcdi_handle_assertion(struct efx_nic *efx)
1860 {
1861 	int rc;
1862 
1863 	rc = efx_mcdi_read_assertion(efx);
1864 	if (rc <= 0)
1865 		return rc;
1866 
1867 	return efx_mcdi_exit_assertion(efx);
1868 }
1869 
1870 int efx_mcdi_set_id_led(struct efx_nic *efx, enum efx_led_mode mode)
1871 {
1872 	MCDI_DECLARE_BUF(inbuf, MC_CMD_SET_ID_LED_IN_LEN);
1873 
1874 	BUILD_BUG_ON(EFX_LED_OFF != MC_CMD_LED_OFF);
1875 	BUILD_BUG_ON(EFX_LED_ON != MC_CMD_LED_ON);
1876 	BUILD_BUG_ON(EFX_LED_DEFAULT != MC_CMD_LED_DEFAULT);
1877 
1878 	BUILD_BUG_ON(MC_CMD_SET_ID_LED_OUT_LEN != 0);
1879 
1880 	MCDI_SET_DWORD(inbuf, SET_ID_LED_IN_STATE, mode);
1881 
1882 	return efx_mcdi_rpc(efx, MC_CMD_SET_ID_LED, inbuf, sizeof(inbuf), NULL, 0, NULL);
1883 }
1884 
1885 static int efx_mcdi_reset_func(struct efx_nic *efx)
1886 {
1887 	MCDI_DECLARE_BUF(inbuf, MC_CMD_ENTITY_RESET_IN_LEN);
1888 	int rc;
1889 
1890 	BUILD_BUG_ON(MC_CMD_ENTITY_RESET_OUT_LEN != 0);
1891 	MCDI_POPULATE_DWORD_1(inbuf, ENTITY_RESET_IN_FLAG,
1892 			      ENTITY_RESET_IN_FUNCTION_RESOURCE_RESET, 1);
1893 	rc = efx_mcdi_rpc(efx, MC_CMD_ENTITY_RESET, inbuf, sizeof(inbuf),
1894 			  NULL, 0, NULL);
1895 	return rc;
1896 }
1897 
1898 static int efx_mcdi_reset_mc(struct efx_nic *efx)
1899 {
1900 	MCDI_DECLARE_BUF(inbuf, MC_CMD_REBOOT_IN_LEN);
1901 	int rc;
1902 
1903 	BUILD_BUG_ON(MC_CMD_REBOOT_OUT_LEN != 0);
1904 	MCDI_SET_DWORD(inbuf, REBOOT_IN_FLAGS, 0);
1905 	rc = efx_mcdi_rpc(efx, MC_CMD_REBOOT, inbuf, sizeof(inbuf),
1906 			  NULL, 0, NULL);
1907 	/* White is black, and up is down */
1908 	if (rc == -EIO)
1909 		return 0;
1910 	if (rc == 0)
1911 		rc = -EIO;
1912 	return rc;
1913 }
1914 
1915 enum reset_type efx_mcdi_map_reset_reason(enum reset_type reason)
1916 {
1917 	return RESET_TYPE_RECOVER_OR_ALL;
1918 }
1919 
1920 int efx_mcdi_reset(struct efx_nic *efx, enum reset_type method)
1921 {
1922 	int rc;
1923 
1924 	/* If MCDI is down, we can't handle_assertion */
1925 	if (method == RESET_TYPE_MCDI_TIMEOUT) {
1926 		rc = pci_reset_function(efx->pci_dev);
1927 		if (rc)
1928 			return rc;
1929 		/* Re-enable polled MCDI completion */
1930 		if (efx->mcdi) {
1931 			struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
1932 			mcdi->mode = MCDI_MODE_POLL;
1933 		}
1934 		return 0;
1935 	}
1936 
1937 	/* Recover from a failed assertion pre-reset */
1938 	rc = efx_mcdi_handle_assertion(efx);
1939 	if (rc)
1940 		return rc;
1941 
1942 	if (method == RESET_TYPE_DATAPATH)
1943 		return 0;
1944 	else if (method == RESET_TYPE_WORLD)
1945 		return efx_mcdi_reset_mc(efx);
1946 	else
1947 		return efx_mcdi_reset_func(efx);
1948 }
1949 
1950 static int efx_mcdi_wol_filter_set(struct efx_nic *efx, u32 type,
1951 				   const u8 *mac, int *id_out)
1952 {
1953 	MCDI_DECLARE_BUF(inbuf, MC_CMD_WOL_FILTER_SET_IN_LEN);
1954 	MCDI_DECLARE_BUF(outbuf, MC_CMD_WOL_FILTER_SET_OUT_LEN);
1955 	size_t outlen;
1956 	int rc;
1957 
1958 	MCDI_SET_DWORD(inbuf, WOL_FILTER_SET_IN_WOL_TYPE, type);
1959 	MCDI_SET_DWORD(inbuf, WOL_FILTER_SET_IN_FILTER_MODE,
1960 		       MC_CMD_FILTER_MODE_SIMPLE);
1961 	ether_addr_copy(MCDI_PTR(inbuf, WOL_FILTER_SET_IN_MAGIC_MAC), mac);
1962 
1963 	rc = efx_mcdi_rpc(efx, MC_CMD_WOL_FILTER_SET, inbuf, sizeof(inbuf),
1964 			  outbuf, sizeof(outbuf), &outlen);
1965 	if (rc)
1966 		goto fail;
1967 
1968 	if (outlen < MC_CMD_WOL_FILTER_SET_OUT_LEN) {
1969 		rc = -EIO;
1970 		goto fail;
1971 	}
1972 
1973 	*id_out = (int)MCDI_DWORD(outbuf, WOL_FILTER_SET_OUT_FILTER_ID);
1974 
1975 	return 0;
1976 
1977 fail:
1978 	*id_out = -1;
1979 	netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
1980 	return rc;
1981 
1982 }
1983 
1984 
1985 int
1986 efx_mcdi_wol_filter_set_magic(struct efx_nic *efx,  const u8 *mac, int *id_out)
1987 {
1988 	return efx_mcdi_wol_filter_set(efx, MC_CMD_WOL_TYPE_MAGIC, mac, id_out);
1989 }
1990 
1991 
1992 int efx_mcdi_wol_filter_get_magic(struct efx_nic *efx, int *id_out)
1993 {
1994 	MCDI_DECLARE_BUF(outbuf, MC_CMD_WOL_FILTER_GET_OUT_LEN);
1995 	size_t outlen;
1996 	int rc;
1997 
1998 	rc = efx_mcdi_rpc(efx, MC_CMD_WOL_FILTER_GET, NULL, 0,
1999 			  outbuf, sizeof(outbuf), &outlen);
2000 	if (rc)
2001 		goto fail;
2002 
2003 	if (outlen < MC_CMD_WOL_FILTER_GET_OUT_LEN) {
2004 		rc = -EIO;
2005 		goto fail;
2006 	}
2007 
2008 	*id_out = (int)MCDI_DWORD(outbuf, WOL_FILTER_GET_OUT_FILTER_ID);
2009 
2010 	return 0;
2011 
2012 fail:
2013 	*id_out = -1;
2014 	netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
2015 	return rc;
2016 }
2017 
2018 
2019 int efx_mcdi_wol_filter_remove(struct efx_nic *efx, int id)
2020 {
2021 	MCDI_DECLARE_BUF(inbuf, MC_CMD_WOL_FILTER_REMOVE_IN_LEN);
2022 	int rc;
2023 
2024 	MCDI_SET_DWORD(inbuf, WOL_FILTER_REMOVE_IN_FILTER_ID, (u32)id);
2025 
2026 	rc = efx_mcdi_rpc(efx, MC_CMD_WOL_FILTER_REMOVE, inbuf, sizeof(inbuf),
2027 			  NULL, 0, NULL);
2028 	return rc;
2029 }
2030 
2031 int efx_mcdi_flush_rxqs(struct efx_nic *efx)
2032 {
2033 	struct efx_channel *channel;
2034 	struct efx_rx_queue *rx_queue;
2035 	MCDI_DECLARE_BUF(inbuf,
2036 			 MC_CMD_FLUSH_RX_QUEUES_IN_LEN(EFX_MAX_CHANNELS));
2037 	int rc, count;
2038 
2039 	BUILD_BUG_ON(EFX_MAX_CHANNELS >
2040 		     MC_CMD_FLUSH_RX_QUEUES_IN_QID_OFST_MAXNUM);
2041 
2042 	count = 0;
2043 	efx_for_each_channel(channel, efx) {
2044 		efx_for_each_channel_rx_queue(rx_queue, channel) {
2045 			if (rx_queue->flush_pending) {
2046 				rx_queue->flush_pending = false;
2047 				atomic_dec(&efx->rxq_flush_pending);
2048 				MCDI_SET_ARRAY_DWORD(
2049 					inbuf, FLUSH_RX_QUEUES_IN_QID_OFST,
2050 					count, efx_rx_queue_index(rx_queue));
2051 				count++;
2052 			}
2053 		}
2054 	}
2055 
2056 	rc = efx_mcdi_rpc(efx, MC_CMD_FLUSH_RX_QUEUES, inbuf,
2057 			  MC_CMD_FLUSH_RX_QUEUES_IN_LEN(count), NULL, 0, NULL);
2058 	WARN_ON(rc < 0);
2059 
2060 	return rc;
2061 }
2062 
2063 int efx_mcdi_wol_filter_reset(struct efx_nic *efx)
2064 {
2065 	int rc;
2066 
2067 	rc = efx_mcdi_rpc(efx, MC_CMD_WOL_FILTER_RESET, NULL, 0, NULL, 0, NULL);
2068 	return rc;
2069 }
2070 
2071 int efx_mcdi_set_workaround(struct efx_nic *efx, u32 type, bool enabled,
2072 			    unsigned int *flags)
2073 {
2074 	MCDI_DECLARE_BUF(inbuf, MC_CMD_WORKAROUND_IN_LEN);
2075 	MCDI_DECLARE_BUF(outbuf, MC_CMD_WORKAROUND_EXT_OUT_LEN);
2076 	size_t outlen;
2077 	int rc;
2078 
2079 	BUILD_BUG_ON(MC_CMD_WORKAROUND_OUT_LEN != 0);
2080 	MCDI_SET_DWORD(inbuf, WORKAROUND_IN_TYPE, type);
2081 	MCDI_SET_DWORD(inbuf, WORKAROUND_IN_ENABLED, enabled);
2082 	rc = efx_mcdi_rpc(efx, MC_CMD_WORKAROUND, inbuf, sizeof(inbuf),
2083 			  outbuf, sizeof(outbuf), &outlen);
2084 	if (rc)
2085 		return rc;
2086 
2087 	if (!flags)
2088 		return 0;
2089 
2090 	if (outlen >= MC_CMD_WORKAROUND_EXT_OUT_LEN)
2091 		*flags = MCDI_DWORD(outbuf, WORKAROUND_EXT_OUT_FLAGS);
2092 	else
2093 		*flags = 0;
2094 
2095 	return 0;
2096 }
2097 
2098 int efx_mcdi_get_workarounds(struct efx_nic *efx, unsigned int *impl_out,
2099 			     unsigned int *enabled_out)
2100 {
2101 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_WORKAROUNDS_OUT_LEN);
2102 	size_t outlen;
2103 	int rc;
2104 
2105 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_WORKAROUNDS, NULL, 0,
2106 			  outbuf, sizeof(outbuf), &outlen);
2107 	if (rc)
2108 		goto fail;
2109 
2110 	if (outlen < MC_CMD_GET_WORKAROUNDS_OUT_LEN) {
2111 		rc = -EIO;
2112 		goto fail;
2113 	}
2114 
2115 	if (impl_out)
2116 		*impl_out = MCDI_DWORD(outbuf, GET_WORKAROUNDS_OUT_IMPLEMENTED);
2117 
2118 	if (enabled_out)
2119 		*enabled_out = MCDI_DWORD(outbuf, GET_WORKAROUNDS_OUT_ENABLED);
2120 
2121 	return 0;
2122 
2123 fail:
2124 	/* Older firmware lacks GET_WORKAROUNDS and this isn't especially
2125 	 * terrifying.  The call site will have to deal with it though.
2126 	 */
2127 	netif_cond_dbg(efx, hw, efx->net_dev, rc == -ENOSYS, err,
2128 		       "%s: failed rc=%d\n", __func__, rc);
2129 	return rc;
2130 }
2131 
2132 /* Failure to read a privilege mask is never fatal, because we can always
2133  * carry on as though we didn't have the privilege we were interested in.
2134  * So use efx_mcdi_rpc_quiet().
2135  */
2136 int efx_mcdi_get_privilege_mask(struct efx_nic *efx, u32 *mask)
2137 {
2138 	MCDI_DECLARE_BUF(fi_outbuf, MC_CMD_GET_FUNCTION_INFO_OUT_LEN);
2139 	MCDI_DECLARE_BUF(pm_inbuf, MC_CMD_PRIVILEGE_MASK_IN_LEN);
2140 	MCDI_DECLARE_BUF(pm_outbuf, MC_CMD_PRIVILEGE_MASK_OUT_LEN);
2141 	size_t outlen;
2142 	u16 pf, vf;
2143 	int rc;
2144 
2145 	if (!efx || !mask)
2146 		return -EINVAL;
2147 
2148 	/* Get our function number */
2149 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_GET_FUNCTION_INFO, NULL, 0,
2150 				fi_outbuf, MC_CMD_GET_FUNCTION_INFO_OUT_LEN,
2151 				&outlen);
2152 	if (rc != 0)
2153 		return rc;
2154 	if (outlen < MC_CMD_GET_FUNCTION_INFO_OUT_LEN)
2155 		return -EIO;
2156 
2157 	pf = MCDI_DWORD(fi_outbuf, GET_FUNCTION_INFO_OUT_PF);
2158 	vf = MCDI_DWORD(fi_outbuf, GET_FUNCTION_INFO_OUT_VF);
2159 
2160 	MCDI_POPULATE_DWORD_2(pm_inbuf, PRIVILEGE_MASK_IN_FUNCTION,
2161 			      PRIVILEGE_MASK_IN_FUNCTION_PF, pf,
2162 			      PRIVILEGE_MASK_IN_FUNCTION_VF, vf);
2163 
2164 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_PRIVILEGE_MASK,
2165 				pm_inbuf, sizeof(pm_inbuf),
2166 				pm_outbuf, sizeof(pm_outbuf), &outlen);
2167 
2168 	if (rc != 0)
2169 		return rc;
2170 	if (outlen < MC_CMD_PRIVILEGE_MASK_OUT_LEN)
2171 		return -EIO;
2172 
2173 	*mask = MCDI_DWORD(pm_outbuf, PRIVILEGE_MASK_OUT_OLD_MASK);
2174 
2175 	return 0;
2176 }
2177 
2178 int efx_mcdi_nvram_metadata(struct efx_nic *efx, unsigned int type,
2179 			    u32 *subtype, u16 version[4], char *desc,
2180 			    size_t descsize)
2181 {
2182 	MCDI_DECLARE_BUF(inbuf, MC_CMD_NVRAM_METADATA_IN_LEN);
2183 	efx_dword_t *outbuf;
2184 	size_t outlen;
2185 	u32 flags;
2186 	int rc;
2187 
2188 	outbuf = kzalloc(MC_CMD_NVRAM_METADATA_OUT_LENMAX_MCDI2, GFP_KERNEL);
2189 	if (!outbuf)
2190 		return -ENOMEM;
2191 
2192 	MCDI_SET_DWORD(inbuf, NVRAM_METADATA_IN_TYPE, type);
2193 
2194 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_NVRAM_METADATA, inbuf,
2195 				sizeof(inbuf), outbuf,
2196 				MC_CMD_NVRAM_METADATA_OUT_LENMAX_MCDI2,
2197 				&outlen);
2198 	if (rc)
2199 		goto out_free;
2200 	if (outlen < MC_CMD_NVRAM_METADATA_OUT_LENMIN) {
2201 		rc = -EIO;
2202 		goto out_free;
2203 	}
2204 
2205 	flags = MCDI_DWORD(outbuf, NVRAM_METADATA_OUT_FLAGS);
2206 
2207 	if (desc && descsize > 0) {
2208 		if (flags & BIT(MC_CMD_NVRAM_METADATA_OUT_DESCRIPTION_VALID_LBN)) {
2209 			if (descsize <=
2210 			    MC_CMD_NVRAM_METADATA_OUT_DESCRIPTION_NUM(outlen)) {
2211 				rc = -E2BIG;
2212 				goto out_free;
2213 			}
2214 
2215 			strncpy(desc,
2216 				MCDI_PTR(outbuf, NVRAM_METADATA_OUT_DESCRIPTION),
2217 				MC_CMD_NVRAM_METADATA_OUT_DESCRIPTION_NUM(outlen));
2218 			desc[MC_CMD_NVRAM_METADATA_OUT_DESCRIPTION_NUM(outlen)] = '\0';
2219 		} else {
2220 			desc[0] = '\0';
2221 		}
2222 	}
2223 
2224 	if (subtype) {
2225 		if (flags & BIT(MC_CMD_NVRAM_METADATA_OUT_SUBTYPE_VALID_LBN))
2226 			*subtype = MCDI_DWORD(outbuf, NVRAM_METADATA_OUT_SUBTYPE);
2227 		else
2228 			*subtype = 0;
2229 	}
2230 
2231 	if (version) {
2232 		if (flags & BIT(MC_CMD_NVRAM_METADATA_OUT_VERSION_VALID_LBN)) {
2233 			version[0] = MCDI_WORD(outbuf, NVRAM_METADATA_OUT_VERSION_W);
2234 			version[1] = MCDI_WORD(outbuf, NVRAM_METADATA_OUT_VERSION_X);
2235 			version[2] = MCDI_WORD(outbuf, NVRAM_METADATA_OUT_VERSION_Y);
2236 			version[3] = MCDI_WORD(outbuf, NVRAM_METADATA_OUT_VERSION_Z);
2237 		} else {
2238 			version[0] = 0;
2239 			version[1] = 0;
2240 			version[2] = 0;
2241 			version[3] = 0;
2242 		}
2243 	}
2244 
2245 out_free:
2246 	kfree(outbuf);
2247 	return rc;
2248 }
2249 
2250 #ifdef CONFIG_SFC_MTD
2251 
2252 #define EFX_MCDI_NVRAM_LEN_MAX 128
2253 
2254 static int efx_mcdi_nvram_update_start(struct efx_nic *efx, unsigned int type)
2255 {
2256 	MCDI_DECLARE_BUF(inbuf, MC_CMD_NVRAM_UPDATE_START_V2_IN_LEN);
2257 	int rc;
2258 
2259 	MCDI_SET_DWORD(inbuf, NVRAM_UPDATE_START_IN_TYPE, type);
2260 	MCDI_POPULATE_DWORD_1(inbuf, NVRAM_UPDATE_START_V2_IN_FLAGS,
2261 			      NVRAM_UPDATE_START_V2_IN_FLAG_REPORT_VERIFY_RESULT,
2262 			      1);
2263 
2264 	BUILD_BUG_ON(MC_CMD_NVRAM_UPDATE_START_OUT_LEN != 0);
2265 
2266 	rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_UPDATE_START, inbuf, sizeof(inbuf),
2267 			  NULL, 0, NULL);
2268 
2269 	return rc;
2270 }
2271 
2272 static int efx_mcdi_nvram_read(struct efx_nic *efx, unsigned int type,
2273 			       loff_t offset, u8 *buffer, size_t length)
2274 {
2275 	MCDI_DECLARE_BUF(inbuf, MC_CMD_NVRAM_READ_IN_V2_LEN);
2276 	MCDI_DECLARE_BUF(outbuf,
2277 			 MC_CMD_NVRAM_READ_OUT_LEN(EFX_MCDI_NVRAM_LEN_MAX));
2278 	size_t outlen;
2279 	int rc;
2280 
2281 	MCDI_SET_DWORD(inbuf, NVRAM_READ_IN_TYPE, type);
2282 	MCDI_SET_DWORD(inbuf, NVRAM_READ_IN_OFFSET, offset);
2283 	MCDI_SET_DWORD(inbuf, NVRAM_READ_IN_LENGTH, length);
2284 	MCDI_SET_DWORD(inbuf, NVRAM_READ_IN_V2_MODE,
2285 		       MC_CMD_NVRAM_READ_IN_V2_DEFAULT);
2286 
2287 	rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_READ, inbuf, sizeof(inbuf),
2288 			  outbuf, sizeof(outbuf), &outlen);
2289 	if (rc)
2290 		return rc;
2291 
2292 	memcpy(buffer, MCDI_PTR(outbuf, NVRAM_READ_OUT_READ_BUFFER), length);
2293 	return 0;
2294 }
2295 
2296 static int efx_mcdi_nvram_write(struct efx_nic *efx, unsigned int type,
2297 				loff_t offset, const u8 *buffer, size_t length)
2298 {
2299 	MCDI_DECLARE_BUF(inbuf,
2300 			 MC_CMD_NVRAM_WRITE_IN_LEN(EFX_MCDI_NVRAM_LEN_MAX));
2301 	int rc;
2302 
2303 	MCDI_SET_DWORD(inbuf, NVRAM_WRITE_IN_TYPE, type);
2304 	MCDI_SET_DWORD(inbuf, NVRAM_WRITE_IN_OFFSET, offset);
2305 	MCDI_SET_DWORD(inbuf, NVRAM_WRITE_IN_LENGTH, length);
2306 	memcpy(MCDI_PTR(inbuf, NVRAM_WRITE_IN_WRITE_BUFFER), buffer, length);
2307 
2308 	BUILD_BUG_ON(MC_CMD_NVRAM_WRITE_OUT_LEN != 0);
2309 
2310 	rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_WRITE, inbuf,
2311 			  ALIGN(MC_CMD_NVRAM_WRITE_IN_LEN(length), 4),
2312 			  NULL, 0, NULL);
2313 	return rc;
2314 }
2315 
2316 static int efx_mcdi_nvram_erase(struct efx_nic *efx, unsigned int type,
2317 				loff_t offset, size_t length)
2318 {
2319 	MCDI_DECLARE_BUF(inbuf, MC_CMD_NVRAM_ERASE_IN_LEN);
2320 	int rc;
2321 
2322 	MCDI_SET_DWORD(inbuf, NVRAM_ERASE_IN_TYPE, type);
2323 	MCDI_SET_DWORD(inbuf, NVRAM_ERASE_IN_OFFSET, offset);
2324 	MCDI_SET_DWORD(inbuf, NVRAM_ERASE_IN_LENGTH, length);
2325 
2326 	BUILD_BUG_ON(MC_CMD_NVRAM_ERASE_OUT_LEN != 0);
2327 
2328 	rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_ERASE, inbuf, sizeof(inbuf),
2329 			  NULL, 0, NULL);
2330 	return rc;
2331 }
2332 
2333 static int efx_mcdi_nvram_update_finish(struct efx_nic *efx, unsigned int type)
2334 {
2335 	MCDI_DECLARE_BUF(inbuf, MC_CMD_NVRAM_UPDATE_FINISH_V2_IN_LEN);
2336 	MCDI_DECLARE_BUF(outbuf, MC_CMD_NVRAM_UPDATE_FINISH_V2_OUT_LEN);
2337 	size_t outlen;
2338 	int rc, rc2;
2339 
2340 	MCDI_SET_DWORD(inbuf, NVRAM_UPDATE_FINISH_IN_TYPE, type);
2341 	/* Always set this flag. Old firmware ignores it */
2342 	MCDI_POPULATE_DWORD_1(inbuf, NVRAM_UPDATE_FINISH_V2_IN_FLAGS,
2343 			      NVRAM_UPDATE_FINISH_V2_IN_FLAG_REPORT_VERIFY_RESULT,
2344 			      1);
2345 
2346 	rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_UPDATE_FINISH, inbuf, sizeof(inbuf),
2347 			  outbuf, sizeof(outbuf), &outlen);
2348 	if (!rc && outlen >= MC_CMD_NVRAM_UPDATE_FINISH_V2_OUT_LEN) {
2349 		rc2 = MCDI_DWORD(outbuf, NVRAM_UPDATE_FINISH_V2_OUT_RESULT_CODE);
2350 		if (rc2 != MC_CMD_NVRAM_VERIFY_RC_SUCCESS)
2351 			netif_err(efx, drv, efx->net_dev,
2352 				  "NVRAM update failed verification with code 0x%x\n",
2353 				  rc2);
2354 		switch (rc2) {
2355 		case MC_CMD_NVRAM_VERIFY_RC_SUCCESS:
2356 			break;
2357 		case MC_CMD_NVRAM_VERIFY_RC_CMS_CHECK_FAILED:
2358 		case MC_CMD_NVRAM_VERIFY_RC_MESSAGE_DIGEST_CHECK_FAILED:
2359 		case MC_CMD_NVRAM_VERIFY_RC_SIGNATURE_CHECK_FAILED:
2360 		case MC_CMD_NVRAM_VERIFY_RC_TRUSTED_APPROVERS_CHECK_FAILED:
2361 		case MC_CMD_NVRAM_VERIFY_RC_SIGNATURE_CHAIN_CHECK_FAILED:
2362 			rc = -EIO;
2363 			break;
2364 		case MC_CMD_NVRAM_VERIFY_RC_INVALID_CMS_FORMAT:
2365 		case MC_CMD_NVRAM_VERIFY_RC_BAD_MESSAGE_DIGEST:
2366 			rc = -EINVAL;
2367 			break;
2368 		case MC_CMD_NVRAM_VERIFY_RC_NO_VALID_SIGNATURES:
2369 		case MC_CMD_NVRAM_VERIFY_RC_NO_TRUSTED_APPROVERS:
2370 		case MC_CMD_NVRAM_VERIFY_RC_NO_SIGNATURE_MATCH:
2371 			rc = -EPERM;
2372 			break;
2373 		default:
2374 			netif_err(efx, drv, efx->net_dev,
2375 				  "Unknown response to NVRAM_UPDATE_FINISH\n");
2376 			rc = -EIO;
2377 		}
2378 	}
2379 
2380 	return rc;
2381 }
2382 
2383 int efx_mcdi_mtd_read(struct mtd_info *mtd, loff_t start,
2384 		      size_t len, size_t *retlen, u8 *buffer)
2385 {
2386 	struct efx_mcdi_mtd_partition *part = to_efx_mcdi_mtd_partition(mtd);
2387 	struct efx_nic *efx = mtd->priv;
2388 	loff_t offset = start;
2389 	loff_t end = min_t(loff_t, start + len, mtd->size);
2390 	size_t chunk;
2391 	int rc = 0;
2392 
2393 	while (offset < end) {
2394 		chunk = min_t(size_t, end - offset, EFX_MCDI_NVRAM_LEN_MAX);
2395 		rc = efx_mcdi_nvram_read(efx, part->nvram_type, offset,
2396 					 buffer, chunk);
2397 		if (rc)
2398 			goto out;
2399 		offset += chunk;
2400 		buffer += chunk;
2401 	}
2402 out:
2403 	*retlen = offset - start;
2404 	return rc;
2405 }
2406 
2407 int efx_mcdi_mtd_erase(struct mtd_info *mtd, loff_t start, size_t len)
2408 {
2409 	struct efx_mcdi_mtd_partition *part = to_efx_mcdi_mtd_partition(mtd);
2410 	struct efx_nic *efx = mtd->priv;
2411 	loff_t offset = start & ~((loff_t)(mtd->erasesize - 1));
2412 	loff_t end = min_t(loff_t, start + len, mtd->size);
2413 	size_t chunk = part->common.mtd.erasesize;
2414 	int rc = 0;
2415 
2416 	if (!part->updating) {
2417 		rc = efx_mcdi_nvram_update_start(efx, part->nvram_type);
2418 		if (rc)
2419 			goto out;
2420 		part->updating = true;
2421 	}
2422 
2423 	/* The MCDI interface can in fact do multiple erase blocks at once;
2424 	 * but erasing may be slow, so we make multiple calls here to avoid
2425 	 * tripping the MCDI RPC timeout. */
2426 	while (offset < end) {
2427 		rc = efx_mcdi_nvram_erase(efx, part->nvram_type, offset,
2428 					  chunk);
2429 		if (rc)
2430 			goto out;
2431 		offset += chunk;
2432 	}
2433 out:
2434 	return rc;
2435 }
2436 
2437 int efx_mcdi_mtd_write(struct mtd_info *mtd, loff_t start,
2438 		       size_t len, size_t *retlen, const u8 *buffer)
2439 {
2440 	struct efx_mcdi_mtd_partition *part = to_efx_mcdi_mtd_partition(mtd);
2441 	struct efx_nic *efx = mtd->priv;
2442 	loff_t offset = start;
2443 	loff_t end = min_t(loff_t, start + len, mtd->size);
2444 	size_t chunk;
2445 	int rc = 0;
2446 
2447 	if (!part->updating) {
2448 		rc = efx_mcdi_nvram_update_start(efx, part->nvram_type);
2449 		if (rc)
2450 			goto out;
2451 		part->updating = true;
2452 	}
2453 
2454 	while (offset < end) {
2455 		chunk = min_t(size_t, end - offset, EFX_MCDI_NVRAM_LEN_MAX);
2456 		rc = efx_mcdi_nvram_write(efx, part->nvram_type, offset,
2457 					  buffer, chunk);
2458 		if (rc)
2459 			goto out;
2460 		offset += chunk;
2461 		buffer += chunk;
2462 	}
2463 out:
2464 	*retlen = offset - start;
2465 	return rc;
2466 }
2467 
2468 int efx_mcdi_mtd_sync(struct mtd_info *mtd)
2469 {
2470 	struct efx_mcdi_mtd_partition *part = to_efx_mcdi_mtd_partition(mtd);
2471 	struct efx_nic *efx = mtd->priv;
2472 	int rc = 0;
2473 
2474 	if (part->updating) {
2475 		part->updating = false;
2476 		rc = efx_mcdi_nvram_update_finish(efx, part->nvram_type);
2477 	}
2478 
2479 	return rc;
2480 }
2481 
2482 void efx_mcdi_mtd_rename(struct efx_mtd_partition *part)
2483 {
2484 	struct efx_mcdi_mtd_partition *mcdi_part =
2485 		container_of(part, struct efx_mcdi_mtd_partition, common);
2486 	struct efx_nic *efx = part->mtd.priv;
2487 
2488 	snprintf(part->name, sizeof(part->name), "%s %s:%02x",
2489 		 efx->name, part->type_name, mcdi_part->fw_subtype);
2490 }
2491 
2492 #endif /* CONFIG_SFC_MTD */
2493