xref: /openbmc/linux/drivers/net/ethernet/sfc/ef10.c (revision 206204a1)
1 /****************************************************************************
2  * Driver for Solarflare network controllers and boards
3  * Copyright 2012-2013 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 "net_driver.h"
11 #include "ef10_regs.h"
12 #include "io.h"
13 #include "mcdi.h"
14 #include "mcdi_pcol.h"
15 #include "nic.h"
16 #include "workarounds.h"
17 #include "selftest.h"
18 #include <linux/in.h>
19 #include <linux/jhash.h>
20 #include <linux/wait.h>
21 #include <linux/workqueue.h>
22 
23 /* Hardware control for EF10 architecture including 'Huntington'. */
24 
25 #define EFX_EF10_DRVGEN_EV		7
26 enum {
27 	EFX_EF10_TEST = 1,
28 	EFX_EF10_REFILL,
29 };
30 
31 /* The reserved RSS context value */
32 #define EFX_EF10_RSS_CONTEXT_INVALID	0xffffffff
33 
34 /* The filter table(s) are managed by firmware and we have write-only
35  * access.  When removing filters we must identify them to the
36  * firmware by a 64-bit handle, but this is too wide for Linux kernel
37  * interfaces (32-bit for RX NFC, 16-bit for RFS).  Also, we need to
38  * be able to tell in advance whether a requested insertion will
39  * replace an existing filter.  Therefore we maintain a software hash
40  * table, which should be at least as large as the hardware hash
41  * table.
42  *
43  * Huntington has a single 8K filter table shared between all filter
44  * types and both ports.
45  */
46 #define HUNT_FILTER_TBL_ROWS 8192
47 
48 struct efx_ef10_filter_table {
49 /* The RX match field masks supported by this fw & hw, in order of priority */
50 	enum efx_filter_match_flags rx_match_flags[
51 		MC_CMD_GET_PARSER_DISP_INFO_OUT_SUPPORTED_MATCHES_MAXNUM];
52 	unsigned int rx_match_count;
53 
54 	struct {
55 		unsigned long spec;	/* pointer to spec plus flag bits */
56 /* BUSY flag indicates that an update is in progress.  AUTO_OLD is
57  * used to mark and sweep MAC filters for the device address lists.
58  */
59 #define EFX_EF10_FILTER_FLAG_BUSY	1UL
60 #define EFX_EF10_FILTER_FLAG_AUTO_OLD	2UL
61 #define EFX_EF10_FILTER_FLAGS		3UL
62 		u64 handle;		/* firmware handle */
63 	} *entry;
64 	wait_queue_head_t waitq;
65 /* Shadow of net_device address lists, guarded by mac_lock */
66 #define EFX_EF10_FILTER_DEV_UC_MAX	32
67 #define EFX_EF10_FILTER_DEV_MC_MAX	256
68 	struct {
69 		u8 addr[ETH_ALEN];
70 		u16 id;
71 	} dev_uc_list[EFX_EF10_FILTER_DEV_UC_MAX],
72 	  dev_mc_list[EFX_EF10_FILTER_DEV_MC_MAX];
73 	int dev_uc_count;		/* negative for PROMISC */
74 	int dev_mc_count;		/* negative for PROMISC/ALLMULTI */
75 };
76 
77 /* An arbitrary search limit for the software hash table */
78 #define EFX_EF10_FILTER_SEARCH_LIMIT 200
79 
80 static void efx_ef10_rx_push_rss_config(struct efx_nic *efx);
81 static void efx_ef10_rx_free_indir_table(struct efx_nic *efx);
82 static void efx_ef10_filter_table_remove(struct efx_nic *efx);
83 
84 static int efx_ef10_get_warm_boot_count(struct efx_nic *efx)
85 {
86 	efx_dword_t reg;
87 
88 	efx_readd(efx, &reg, ER_DZ_BIU_MC_SFT_STATUS);
89 	return EFX_DWORD_FIELD(reg, EFX_WORD_1) == 0xb007 ?
90 		EFX_DWORD_FIELD(reg, EFX_WORD_0) : -EIO;
91 }
92 
93 static unsigned int efx_ef10_mem_map_size(struct efx_nic *efx)
94 {
95 	return resource_size(&efx->pci_dev->resource[EFX_MEM_BAR]);
96 }
97 
98 static int efx_ef10_init_datapath_caps(struct efx_nic *efx)
99 {
100 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_CAPABILITIES_OUT_LEN);
101 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
102 	size_t outlen;
103 	int rc;
104 
105 	BUILD_BUG_ON(MC_CMD_GET_CAPABILITIES_IN_LEN != 0);
106 
107 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_CAPABILITIES, NULL, 0,
108 			  outbuf, sizeof(outbuf), &outlen);
109 	if (rc)
110 		return rc;
111 	if (outlen < sizeof(outbuf)) {
112 		netif_err(efx, drv, efx->net_dev,
113 			  "unable to read datapath firmware capabilities\n");
114 		return -EIO;
115 	}
116 
117 	nic_data->datapath_caps =
118 		MCDI_DWORD(outbuf, GET_CAPABILITIES_OUT_FLAGS1);
119 
120 	if (!(nic_data->datapath_caps &
121 	      (1 << MC_CMD_GET_CAPABILITIES_OUT_TX_TSO_LBN))) {
122 		netif_err(efx, drv, efx->net_dev,
123 			  "current firmware does not support TSO\n");
124 		return -ENODEV;
125 	}
126 
127 	if (!(nic_data->datapath_caps &
128 	      (1 << MC_CMD_GET_CAPABILITIES_OUT_RX_PREFIX_LEN_14_LBN))) {
129 		netif_err(efx, probe, efx->net_dev,
130 			  "current firmware does not support an RX prefix\n");
131 		return -ENODEV;
132 	}
133 
134 	return 0;
135 }
136 
137 static int efx_ef10_get_sysclk_freq(struct efx_nic *efx)
138 {
139 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_CLOCK_OUT_LEN);
140 	int rc;
141 
142 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_CLOCK, NULL, 0,
143 			  outbuf, sizeof(outbuf), NULL);
144 	if (rc)
145 		return rc;
146 	rc = MCDI_DWORD(outbuf, GET_CLOCK_OUT_SYS_FREQ);
147 	return rc > 0 ? rc : -ERANGE;
148 }
149 
150 static int efx_ef10_get_mac_address(struct efx_nic *efx, u8 *mac_address)
151 {
152 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_MAC_ADDRESSES_OUT_LEN);
153 	size_t outlen;
154 	int rc;
155 
156 	BUILD_BUG_ON(MC_CMD_GET_MAC_ADDRESSES_IN_LEN != 0);
157 
158 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_MAC_ADDRESSES, NULL, 0,
159 			  outbuf, sizeof(outbuf), &outlen);
160 	if (rc)
161 		return rc;
162 	if (outlen < MC_CMD_GET_MAC_ADDRESSES_OUT_LEN)
163 		return -EIO;
164 
165 	ether_addr_copy(mac_address,
166 			MCDI_PTR(outbuf, GET_MAC_ADDRESSES_OUT_MAC_ADDR_BASE));
167 	return 0;
168 }
169 
170 static int efx_ef10_probe(struct efx_nic *efx)
171 {
172 	struct efx_ef10_nic_data *nic_data;
173 	int i, rc;
174 
175 	/* We can have one VI for each 8K region.  However, until we
176 	 * use TX option descriptors we need two TX queues per channel.
177 	 */
178 	efx->max_channels =
179 		min_t(unsigned int,
180 		      EFX_MAX_CHANNELS,
181 		      resource_size(&efx->pci_dev->resource[EFX_MEM_BAR]) /
182 		      (EFX_VI_PAGE_SIZE * EFX_TXQ_TYPES));
183 	BUG_ON(efx->max_channels == 0);
184 
185 	nic_data = kzalloc(sizeof(*nic_data), GFP_KERNEL);
186 	if (!nic_data)
187 		return -ENOMEM;
188 	efx->nic_data = nic_data;
189 
190 	rc = efx_nic_alloc_buffer(efx, &nic_data->mcdi_buf,
191 				  8 + MCDI_CTL_SDU_LEN_MAX_V2, GFP_KERNEL);
192 	if (rc)
193 		goto fail1;
194 
195 	/* Get the MC's warm boot count.  In case it's rebooting right
196 	 * now, be prepared to retry.
197 	 */
198 	i = 0;
199 	for (;;) {
200 		rc = efx_ef10_get_warm_boot_count(efx);
201 		if (rc >= 0)
202 			break;
203 		if (++i == 5)
204 			goto fail2;
205 		ssleep(1);
206 	}
207 	nic_data->warm_boot_count = rc;
208 
209 	nic_data->rx_rss_context = EFX_EF10_RSS_CONTEXT_INVALID;
210 
211 	/* In case we're recovering from a crash (kexec), we want to
212 	 * cancel any outstanding request by the previous user of this
213 	 * function.  We send a special message using the least
214 	 * significant bits of the 'high' (doorbell) register.
215 	 */
216 	_efx_writed(efx, cpu_to_le32(1), ER_DZ_MC_DB_HWRD);
217 
218 	rc = efx_mcdi_init(efx);
219 	if (rc)
220 		goto fail2;
221 
222 	/* Reset (most) configuration for this function */
223 	rc = efx_mcdi_reset(efx, RESET_TYPE_ALL);
224 	if (rc)
225 		goto fail3;
226 
227 	/* Enable event logging */
228 	rc = efx_mcdi_log_ctrl(efx, true, false, 0);
229 	if (rc)
230 		goto fail3;
231 
232 	rc = efx_ef10_init_datapath_caps(efx);
233 	if (rc < 0)
234 		goto fail3;
235 
236 	efx->rx_packet_len_offset =
237 		ES_DZ_RX_PREFIX_PKTLEN_OFST - ES_DZ_RX_PREFIX_SIZE;
238 
239 	rc = efx_mcdi_port_get_number(efx);
240 	if (rc < 0)
241 		goto fail3;
242 	efx->port_num = rc;
243 
244 	rc = efx_ef10_get_mac_address(efx, efx->net_dev->perm_addr);
245 	if (rc)
246 		goto fail3;
247 
248 	rc = efx_ef10_get_sysclk_freq(efx);
249 	if (rc < 0)
250 		goto fail3;
251 	efx->timer_quantum_ns = 1536000 / rc; /* 1536 cycles */
252 
253 	/* Check whether firmware supports bug 35388 workaround */
254 	rc = efx_mcdi_set_workaround(efx, MC_CMD_WORKAROUND_BUG35388, true);
255 	if (rc == 0)
256 		nic_data->workaround_35388 = true;
257 	else if (rc != -ENOSYS && rc != -ENOENT)
258 		goto fail3;
259 	netif_dbg(efx, probe, efx->net_dev,
260 		  "workaround for bug 35388 is %sabled\n",
261 		  nic_data->workaround_35388 ? "en" : "dis");
262 
263 	rc = efx_mcdi_mon_probe(efx);
264 	if (rc)
265 		goto fail3;
266 
267 	efx_ptp_probe(efx, NULL);
268 
269 	return 0;
270 
271 fail3:
272 	efx_mcdi_fini(efx);
273 fail2:
274 	efx_nic_free_buffer(efx, &nic_data->mcdi_buf);
275 fail1:
276 	kfree(nic_data);
277 	efx->nic_data = NULL;
278 	return rc;
279 }
280 
281 static int efx_ef10_free_vis(struct efx_nic *efx)
282 {
283 	MCDI_DECLARE_BUF_OUT_OR_ERR(outbuf, 0);
284 	size_t outlen;
285 	int rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FREE_VIS, NULL, 0,
286 				    outbuf, sizeof(outbuf), &outlen);
287 
288 	/* -EALREADY means nothing to free, so ignore */
289 	if (rc == -EALREADY)
290 		rc = 0;
291 	if (rc)
292 		efx_mcdi_display_error(efx, MC_CMD_FREE_VIS, 0, outbuf, outlen,
293 				       rc);
294 	return rc;
295 }
296 
297 #ifdef EFX_USE_PIO
298 
299 static void efx_ef10_free_piobufs(struct efx_nic *efx)
300 {
301 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
302 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FREE_PIOBUF_IN_LEN);
303 	unsigned int i;
304 	int rc;
305 
306 	BUILD_BUG_ON(MC_CMD_FREE_PIOBUF_OUT_LEN != 0);
307 
308 	for (i = 0; i < nic_data->n_piobufs; i++) {
309 		MCDI_SET_DWORD(inbuf, FREE_PIOBUF_IN_PIOBUF_HANDLE,
310 			       nic_data->piobuf_handle[i]);
311 		rc = efx_mcdi_rpc(efx, MC_CMD_FREE_PIOBUF, inbuf, sizeof(inbuf),
312 				  NULL, 0, NULL);
313 		WARN_ON(rc);
314 	}
315 
316 	nic_data->n_piobufs = 0;
317 }
318 
319 static int efx_ef10_alloc_piobufs(struct efx_nic *efx, unsigned int n)
320 {
321 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
322 	MCDI_DECLARE_BUF(outbuf, MC_CMD_ALLOC_PIOBUF_OUT_LEN);
323 	unsigned int i;
324 	size_t outlen;
325 	int rc = 0;
326 
327 	BUILD_BUG_ON(MC_CMD_ALLOC_PIOBUF_IN_LEN != 0);
328 
329 	for (i = 0; i < n; i++) {
330 		rc = efx_mcdi_rpc(efx, MC_CMD_ALLOC_PIOBUF, NULL, 0,
331 				  outbuf, sizeof(outbuf), &outlen);
332 		if (rc)
333 			break;
334 		if (outlen < MC_CMD_ALLOC_PIOBUF_OUT_LEN) {
335 			rc = -EIO;
336 			break;
337 		}
338 		nic_data->piobuf_handle[i] =
339 			MCDI_DWORD(outbuf, ALLOC_PIOBUF_OUT_PIOBUF_HANDLE);
340 		netif_dbg(efx, probe, efx->net_dev,
341 			  "allocated PIO buffer %u handle %x\n", i,
342 			  nic_data->piobuf_handle[i]);
343 	}
344 
345 	nic_data->n_piobufs = i;
346 	if (rc)
347 		efx_ef10_free_piobufs(efx);
348 	return rc;
349 }
350 
351 static int efx_ef10_link_piobufs(struct efx_nic *efx)
352 {
353 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
354 	MCDI_DECLARE_BUF(inbuf,
355 			 max(MC_CMD_LINK_PIOBUF_IN_LEN,
356 			     MC_CMD_UNLINK_PIOBUF_IN_LEN));
357 	struct efx_channel *channel;
358 	struct efx_tx_queue *tx_queue;
359 	unsigned int offset, index;
360 	int rc;
361 
362 	BUILD_BUG_ON(MC_CMD_LINK_PIOBUF_OUT_LEN != 0);
363 	BUILD_BUG_ON(MC_CMD_UNLINK_PIOBUF_OUT_LEN != 0);
364 
365 	/* Link a buffer to each VI in the write-combining mapping */
366 	for (index = 0; index < nic_data->n_piobufs; ++index) {
367 		MCDI_SET_DWORD(inbuf, LINK_PIOBUF_IN_PIOBUF_HANDLE,
368 			       nic_data->piobuf_handle[index]);
369 		MCDI_SET_DWORD(inbuf, LINK_PIOBUF_IN_TXQ_INSTANCE,
370 			       nic_data->pio_write_vi_base + index);
371 		rc = efx_mcdi_rpc(efx, MC_CMD_LINK_PIOBUF,
372 				  inbuf, MC_CMD_LINK_PIOBUF_IN_LEN,
373 				  NULL, 0, NULL);
374 		if (rc) {
375 			netif_err(efx, drv, efx->net_dev,
376 				  "failed to link VI %u to PIO buffer %u (%d)\n",
377 				  nic_data->pio_write_vi_base + index, index,
378 				  rc);
379 			goto fail;
380 		}
381 		netif_dbg(efx, probe, efx->net_dev,
382 			  "linked VI %u to PIO buffer %u\n",
383 			  nic_data->pio_write_vi_base + index, index);
384 	}
385 
386 	/* Link a buffer to each TX queue */
387 	efx_for_each_channel(channel, efx) {
388 		efx_for_each_channel_tx_queue(tx_queue, channel) {
389 			/* We assign the PIO buffers to queues in
390 			 * reverse order to allow for the following
391 			 * special case.
392 			 */
393 			offset = ((efx->tx_channel_offset + efx->n_tx_channels -
394 				   tx_queue->channel->channel - 1) *
395 				  efx_piobuf_size);
396 			index = offset / ER_DZ_TX_PIOBUF_SIZE;
397 			offset = offset % ER_DZ_TX_PIOBUF_SIZE;
398 
399 			/* When the host page size is 4K, the first
400 			 * host page in the WC mapping may be within
401 			 * the same VI page as the last TX queue.  We
402 			 * can only link one buffer to each VI.
403 			 */
404 			if (tx_queue->queue == nic_data->pio_write_vi_base) {
405 				BUG_ON(index != 0);
406 				rc = 0;
407 			} else {
408 				MCDI_SET_DWORD(inbuf,
409 					       LINK_PIOBUF_IN_PIOBUF_HANDLE,
410 					       nic_data->piobuf_handle[index]);
411 				MCDI_SET_DWORD(inbuf,
412 					       LINK_PIOBUF_IN_TXQ_INSTANCE,
413 					       tx_queue->queue);
414 				rc = efx_mcdi_rpc(efx, MC_CMD_LINK_PIOBUF,
415 						  inbuf, MC_CMD_LINK_PIOBUF_IN_LEN,
416 						  NULL, 0, NULL);
417 			}
418 
419 			if (rc) {
420 				/* This is non-fatal; the TX path just
421 				 * won't use PIO for this queue
422 				 */
423 				netif_err(efx, drv, efx->net_dev,
424 					  "failed to link VI %u to PIO buffer %u (%d)\n",
425 					  tx_queue->queue, index, rc);
426 				tx_queue->piobuf = NULL;
427 			} else {
428 				tx_queue->piobuf =
429 					nic_data->pio_write_base +
430 					index * EFX_VI_PAGE_SIZE + offset;
431 				tx_queue->piobuf_offset = offset;
432 				netif_dbg(efx, probe, efx->net_dev,
433 					  "linked VI %u to PIO buffer %u offset %x addr %p\n",
434 					  tx_queue->queue, index,
435 					  tx_queue->piobuf_offset,
436 					  tx_queue->piobuf);
437 			}
438 		}
439 	}
440 
441 	return 0;
442 
443 fail:
444 	while (index--) {
445 		MCDI_SET_DWORD(inbuf, UNLINK_PIOBUF_IN_TXQ_INSTANCE,
446 			       nic_data->pio_write_vi_base + index);
447 		efx_mcdi_rpc(efx, MC_CMD_UNLINK_PIOBUF,
448 			     inbuf, MC_CMD_UNLINK_PIOBUF_IN_LEN,
449 			     NULL, 0, NULL);
450 	}
451 	return rc;
452 }
453 
454 #else /* !EFX_USE_PIO */
455 
456 static int efx_ef10_alloc_piobufs(struct efx_nic *efx, unsigned int n)
457 {
458 	return n == 0 ? 0 : -ENOBUFS;
459 }
460 
461 static int efx_ef10_link_piobufs(struct efx_nic *efx)
462 {
463 	return 0;
464 }
465 
466 static void efx_ef10_free_piobufs(struct efx_nic *efx)
467 {
468 }
469 
470 #endif /* EFX_USE_PIO */
471 
472 static void efx_ef10_remove(struct efx_nic *efx)
473 {
474 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
475 	int rc;
476 
477 	efx_ptp_remove(efx);
478 
479 	efx_mcdi_mon_remove(efx);
480 
481 	efx_ef10_rx_free_indir_table(efx);
482 
483 	if (nic_data->wc_membase)
484 		iounmap(nic_data->wc_membase);
485 
486 	rc = efx_ef10_free_vis(efx);
487 	WARN_ON(rc != 0);
488 
489 	if (!nic_data->must_restore_piobufs)
490 		efx_ef10_free_piobufs(efx);
491 
492 	efx_mcdi_fini(efx);
493 	efx_nic_free_buffer(efx, &nic_data->mcdi_buf);
494 	kfree(nic_data);
495 }
496 
497 static int efx_ef10_alloc_vis(struct efx_nic *efx,
498 			      unsigned int min_vis, unsigned int max_vis)
499 {
500 	MCDI_DECLARE_BUF(inbuf, MC_CMD_ALLOC_VIS_IN_LEN);
501 	MCDI_DECLARE_BUF(outbuf, MC_CMD_ALLOC_VIS_OUT_LEN);
502 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
503 	size_t outlen;
504 	int rc;
505 
506 	MCDI_SET_DWORD(inbuf, ALLOC_VIS_IN_MIN_VI_COUNT, min_vis);
507 	MCDI_SET_DWORD(inbuf, ALLOC_VIS_IN_MAX_VI_COUNT, max_vis);
508 	rc = efx_mcdi_rpc(efx, MC_CMD_ALLOC_VIS, inbuf, sizeof(inbuf),
509 			  outbuf, sizeof(outbuf), &outlen);
510 	if (rc != 0)
511 		return rc;
512 
513 	if (outlen < MC_CMD_ALLOC_VIS_OUT_LEN)
514 		return -EIO;
515 
516 	netif_dbg(efx, drv, efx->net_dev, "base VI is A0x%03x\n",
517 		  MCDI_DWORD(outbuf, ALLOC_VIS_OUT_VI_BASE));
518 
519 	nic_data->vi_base = MCDI_DWORD(outbuf, ALLOC_VIS_OUT_VI_BASE);
520 	nic_data->n_allocated_vis = MCDI_DWORD(outbuf, ALLOC_VIS_OUT_VI_COUNT);
521 	return 0;
522 }
523 
524 /* Note that the failure path of this function does not free
525  * resources, as this will be done by efx_ef10_remove().
526  */
527 static int efx_ef10_dimension_resources(struct efx_nic *efx)
528 {
529 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
530 	unsigned int uc_mem_map_size, wc_mem_map_size;
531 	unsigned int min_vis, pio_write_vi_base, max_vis;
532 	void __iomem *membase;
533 	int rc;
534 
535 	min_vis = max(efx->n_channels, efx->n_tx_channels * EFX_TXQ_TYPES);
536 
537 #ifdef EFX_USE_PIO
538 	/* Try to allocate PIO buffers if wanted and if the full
539 	 * number of PIO buffers would be sufficient to allocate one
540 	 * copy-buffer per TX channel.  Failure is non-fatal, as there
541 	 * are only a small number of PIO buffers shared between all
542 	 * functions of the controller.
543 	 */
544 	if (efx_piobuf_size != 0 &&
545 	    ER_DZ_TX_PIOBUF_SIZE / efx_piobuf_size * EF10_TX_PIOBUF_COUNT >=
546 	    efx->n_tx_channels) {
547 		unsigned int n_piobufs =
548 			DIV_ROUND_UP(efx->n_tx_channels,
549 				     ER_DZ_TX_PIOBUF_SIZE / efx_piobuf_size);
550 
551 		rc = efx_ef10_alloc_piobufs(efx, n_piobufs);
552 		if (rc)
553 			netif_err(efx, probe, efx->net_dev,
554 				  "failed to allocate PIO buffers (%d)\n", rc);
555 		else
556 			netif_dbg(efx, probe, efx->net_dev,
557 				  "allocated %u PIO buffers\n", n_piobufs);
558 	}
559 #else
560 	nic_data->n_piobufs = 0;
561 #endif
562 
563 	/* PIO buffers should be mapped with write-combining enabled,
564 	 * and we want to make single UC and WC mappings rather than
565 	 * several of each (in fact that's the only option if host
566 	 * page size is >4K).  So we may allocate some extra VIs just
567 	 * for writing PIO buffers through.
568 	 *
569 	 * The UC mapping contains (min_vis - 1) complete VIs and the
570 	 * first half of the next VI.  Then the WC mapping begins with
571 	 * the second half of this last VI.
572 	 */
573 	uc_mem_map_size = PAGE_ALIGN((min_vis - 1) * EFX_VI_PAGE_SIZE +
574 				     ER_DZ_TX_PIOBUF);
575 	if (nic_data->n_piobufs) {
576 		/* pio_write_vi_base rounds down to give the number of complete
577 		 * VIs inside the UC mapping.
578 		 */
579 		pio_write_vi_base = uc_mem_map_size / EFX_VI_PAGE_SIZE;
580 		wc_mem_map_size = (PAGE_ALIGN((pio_write_vi_base +
581 					       nic_data->n_piobufs) *
582 					      EFX_VI_PAGE_SIZE) -
583 				   uc_mem_map_size);
584 		max_vis = pio_write_vi_base + nic_data->n_piobufs;
585 	} else {
586 		pio_write_vi_base = 0;
587 		wc_mem_map_size = 0;
588 		max_vis = min_vis;
589 	}
590 
591 	/* In case the last attached driver failed to free VIs, do it now */
592 	rc = efx_ef10_free_vis(efx);
593 	if (rc != 0)
594 		return rc;
595 
596 	rc = efx_ef10_alloc_vis(efx, min_vis, max_vis);
597 	if (rc != 0)
598 		return rc;
599 
600 	/* If we didn't get enough VIs to map all the PIO buffers, free the
601 	 * PIO buffers
602 	 */
603 	if (nic_data->n_piobufs &&
604 	    nic_data->n_allocated_vis <
605 	    pio_write_vi_base + nic_data->n_piobufs) {
606 		netif_dbg(efx, probe, efx->net_dev,
607 			  "%u VIs are not sufficient to map %u PIO buffers\n",
608 			  nic_data->n_allocated_vis, nic_data->n_piobufs);
609 		efx_ef10_free_piobufs(efx);
610 	}
611 
612 	/* Shrink the original UC mapping of the memory BAR */
613 	membase = ioremap_nocache(efx->membase_phys, uc_mem_map_size);
614 	if (!membase) {
615 		netif_err(efx, probe, efx->net_dev,
616 			  "could not shrink memory BAR to %x\n",
617 			  uc_mem_map_size);
618 		return -ENOMEM;
619 	}
620 	iounmap(efx->membase);
621 	efx->membase = membase;
622 
623 	/* Set up the WC mapping if needed */
624 	if (wc_mem_map_size) {
625 		nic_data->wc_membase = ioremap_wc(efx->membase_phys +
626 						  uc_mem_map_size,
627 						  wc_mem_map_size);
628 		if (!nic_data->wc_membase) {
629 			netif_err(efx, probe, efx->net_dev,
630 				  "could not allocate WC mapping of size %x\n",
631 				  wc_mem_map_size);
632 			return -ENOMEM;
633 		}
634 		nic_data->pio_write_vi_base = pio_write_vi_base;
635 		nic_data->pio_write_base =
636 			nic_data->wc_membase +
637 			(pio_write_vi_base * EFX_VI_PAGE_SIZE + ER_DZ_TX_PIOBUF -
638 			 uc_mem_map_size);
639 
640 		rc = efx_ef10_link_piobufs(efx);
641 		if (rc)
642 			efx_ef10_free_piobufs(efx);
643 	}
644 
645 	netif_dbg(efx, probe, efx->net_dev,
646 		  "memory BAR at %pa (virtual %p+%x UC, %p+%x WC)\n",
647 		  &efx->membase_phys, efx->membase, uc_mem_map_size,
648 		  nic_data->wc_membase, wc_mem_map_size);
649 
650 	return 0;
651 }
652 
653 static int efx_ef10_init_nic(struct efx_nic *efx)
654 {
655 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
656 	int rc;
657 
658 	if (nic_data->must_check_datapath_caps) {
659 		rc = efx_ef10_init_datapath_caps(efx);
660 		if (rc)
661 			return rc;
662 		nic_data->must_check_datapath_caps = false;
663 	}
664 
665 	if (nic_data->must_realloc_vis) {
666 		/* We cannot let the number of VIs change now */
667 		rc = efx_ef10_alloc_vis(efx, nic_data->n_allocated_vis,
668 					nic_data->n_allocated_vis);
669 		if (rc)
670 			return rc;
671 		nic_data->must_realloc_vis = false;
672 	}
673 
674 	if (nic_data->must_restore_piobufs && nic_data->n_piobufs) {
675 		rc = efx_ef10_alloc_piobufs(efx, nic_data->n_piobufs);
676 		if (rc == 0) {
677 			rc = efx_ef10_link_piobufs(efx);
678 			if (rc)
679 				efx_ef10_free_piobufs(efx);
680 		}
681 
682 		/* Log an error on failure, but this is non-fatal */
683 		if (rc)
684 			netif_err(efx, drv, efx->net_dev,
685 				  "failed to restore PIO buffers (%d)\n", rc);
686 		nic_data->must_restore_piobufs = false;
687 	}
688 
689 	efx_ef10_rx_push_rss_config(efx);
690 	return 0;
691 }
692 
693 static void efx_ef10_reset_mc_allocations(struct efx_nic *efx)
694 {
695 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
696 
697 	/* All our allocations have been reset */
698 	nic_data->must_realloc_vis = true;
699 	nic_data->must_restore_filters = true;
700 	nic_data->must_restore_piobufs = true;
701 	nic_data->rx_rss_context = EFX_EF10_RSS_CONTEXT_INVALID;
702 }
703 
704 static int efx_ef10_map_reset_flags(u32 *flags)
705 {
706 	enum {
707 		EF10_RESET_PORT = ((ETH_RESET_MAC | ETH_RESET_PHY) <<
708 				   ETH_RESET_SHARED_SHIFT),
709 		EF10_RESET_MC = ((ETH_RESET_DMA | ETH_RESET_FILTER |
710 				  ETH_RESET_OFFLOAD | ETH_RESET_MAC |
711 				  ETH_RESET_PHY | ETH_RESET_MGMT) <<
712 				 ETH_RESET_SHARED_SHIFT)
713 	};
714 
715 	/* We assume for now that our PCI function is permitted to
716 	 * reset everything.
717 	 */
718 
719 	if ((*flags & EF10_RESET_MC) == EF10_RESET_MC) {
720 		*flags &= ~EF10_RESET_MC;
721 		return RESET_TYPE_WORLD;
722 	}
723 
724 	if ((*flags & EF10_RESET_PORT) == EF10_RESET_PORT) {
725 		*flags &= ~EF10_RESET_PORT;
726 		return RESET_TYPE_ALL;
727 	}
728 
729 	/* no invisible reset implemented */
730 
731 	return -EINVAL;
732 }
733 
734 static int efx_ef10_reset(struct efx_nic *efx, enum reset_type reset_type)
735 {
736 	int rc = efx_mcdi_reset(efx, reset_type);
737 
738 	/* If it was a port reset, trigger reallocation of MC resources.
739 	 * Note that on an MC reset nothing needs to be done now because we'll
740 	 * detect the MC reset later and handle it then.
741 	 * For an FLR, we never get an MC reset event, but the MC has reset all
742 	 * resources assigned to us, so we have to trigger reallocation now.
743 	 */
744 	if ((reset_type == RESET_TYPE_ALL ||
745 	     reset_type == RESET_TYPE_MCDI_TIMEOUT) && !rc)
746 		efx_ef10_reset_mc_allocations(efx);
747 	return rc;
748 }
749 
750 #define EF10_DMA_STAT(ext_name, mcdi_name)			\
751 	[EF10_STAT_ ## ext_name] =				\
752 	{ #ext_name, 64, 8 * MC_CMD_MAC_ ## mcdi_name }
753 #define EF10_DMA_INVIS_STAT(int_name, mcdi_name)		\
754 	[EF10_STAT_ ## int_name] =				\
755 	{ NULL, 64, 8 * MC_CMD_MAC_ ## mcdi_name }
756 #define EF10_OTHER_STAT(ext_name)				\
757 	[EF10_STAT_ ## ext_name] = { #ext_name, 0, 0 }
758 
759 static const struct efx_hw_stat_desc efx_ef10_stat_desc[EF10_STAT_COUNT] = {
760 	EF10_DMA_STAT(tx_bytes, TX_BYTES),
761 	EF10_DMA_STAT(tx_packets, TX_PKTS),
762 	EF10_DMA_STAT(tx_pause, TX_PAUSE_PKTS),
763 	EF10_DMA_STAT(tx_control, TX_CONTROL_PKTS),
764 	EF10_DMA_STAT(tx_unicast, TX_UNICAST_PKTS),
765 	EF10_DMA_STAT(tx_multicast, TX_MULTICAST_PKTS),
766 	EF10_DMA_STAT(tx_broadcast, TX_BROADCAST_PKTS),
767 	EF10_DMA_STAT(tx_lt64, TX_LT64_PKTS),
768 	EF10_DMA_STAT(tx_64, TX_64_PKTS),
769 	EF10_DMA_STAT(tx_65_to_127, TX_65_TO_127_PKTS),
770 	EF10_DMA_STAT(tx_128_to_255, TX_128_TO_255_PKTS),
771 	EF10_DMA_STAT(tx_256_to_511, TX_256_TO_511_PKTS),
772 	EF10_DMA_STAT(tx_512_to_1023, TX_512_TO_1023_PKTS),
773 	EF10_DMA_STAT(tx_1024_to_15xx, TX_1024_TO_15XX_PKTS),
774 	EF10_DMA_STAT(tx_15xx_to_jumbo, TX_15XX_TO_JUMBO_PKTS),
775 	EF10_DMA_STAT(rx_bytes, RX_BYTES),
776 	EF10_DMA_INVIS_STAT(rx_bytes_minus_good_bytes, RX_BAD_BYTES),
777 	EF10_OTHER_STAT(rx_good_bytes),
778 	EF10_OTHER_STAT(rx_bad_bytes),
779 	EF10_DMA_STAT(rx_packets, RX_PKTS),
780 	EF10_DMA_STAT(rx_good, RX_GOOD_PKTS),
781 	EF10_DMA_STAT(rx_bad, RX_BAD_FCS_PKTS),
782 	EF10_DMA_STAT(rx_pause, RX_PAUSE_PKTS),
783 	EF10_DMA_STAT(rx_control, RX_CONTROL_PKTS),
784 	EF10_DMA_STAT(rx_unicast, RX_UNICAST_PKTS),
785 	EF10_DMA_STAT(rx_multicast, RX_MULTICAST_PKTS),
786 	EF10_DMA_STAT(rx_broadcast, RX_BROADCAST_PKTS),
787 	EF10_DMA_STAT(rx_lt64, RX_UNDERSIZE_PKTS),
788 	EF10_DMA_STAT(rx_64, RX_64_PKTS),
789 	EF10_DMA_STAT(rx_65_to_127, RX_65_TO_127_PKTS),
790 	EF10_DMA_STAT(rx_128_to_255, RX_128_TO_255_PKTS),
791 	EF10_DMA_STAT(rx_256_to_511, RX_256_TO_511_PKTS),
792 	EF10_DMA_STAT(rx_512_to_1023, RX_512_TO_1023_PKTS),
793 	EF10_DMA_STAT(rx_1024_to_15xx, RX_1024_TO_15XX_PKTS),
794 	EF10_DMA_STAT(rx_15xx_to_jumbo, RX_15XX_TO_JUMBO_PKTS),
795 	EF10_DMA_STAT(rx_gtjumbo, RX_GTJUMBO_PKTS),
796 	EF10_DMA_STAT(rx_bad_gtjumbo, RX_JABBER_PKTS),
797 	EF10_DMA_STAT(rx_overflow, RX_OVERFLOW_PKTS),
798 	EF10_DMA_STAT(rx_align_error, RX_ALIGN_ERROR_PKTS),
799 	EF10_DMA_STAT(rx_length_error, RX_LENGTH_ERROR_PKTS),
800 	EF10_DMA_STAT(rx_nodesc_drops, RX_NODESC_DROPS),
801 	EF10_DMA_STAT(rx_pm_trunc_bb_overflow, PM_TRUNC_BB_OVERFLOW),
802 	EF10_DMA_STAT(rx_pm_discard_bb_overflow, PM_DISCARD_BB_OVERFLOW),
803 	EF10_DMA_STAT(rx_pm_trunc_vfifo_full, PM_TRUNC_VFIFO_FULL),
804 	EF10_DMA_STAT(rx_pm_discard_vfifo_full, PM_DISCARD_VFIFO_FULL),
805 	EF10_DMA_STAT(rx_pm_trunc_qbb, PM_TRUNC_QBB),
806 	EF10_DMA_STAT(rx_pm_discard_qbb, PM_DISCARD_QBB),
807 	EF10_DMA_STAT(rx_pm_discard_mapping, PM_DISCARD_MAPPING),
808 	EF10_DMA_STAT(rx_dp_q_disabled_packets, RXDP_Q_DISABLED_PKTS),
809 	EF10_DMA_STAT(rx_dp_di_dropped_packets, RXDP_DI_DROPPED_PKTS),
810 	EF10_DMA_STAT(rx_dp_streaming_packets, RXDP_STREAMING_PKTS),
811 	EF10_DMA_STAT(rx_dp_hlb_fetch, RXDP_EMERGENCY_FETCH_CONDITIONS),
812 	EF10_DMA_STAT(rx_dp_hlb_wait, RXDP_EMERGENCY_WAIT_CONDITIONS),
813 };
814 
815 #define HUNT_COMMON_STAT_MASK ((1ULL << EF10_STAT_tx_bytes) |		\
816 			       (1ULL << EF10_STAT_tx_packets) |		\
817 			       (1ULL << EF10_STAT_tx_pause) |		\
818 			       (1ULL << EF10_STAT_tx_unicast) |		\
819 			       (1ULL << EF10_STAT_tx_multicast) |	\
820 			       (1ULL << EF10_STAT_tx_broadcast) |	\
821 			       (1ULL << EF10_STAT_rx_bytes) |		\
822 			       (1ULL << EF10_STAT_rx_bytes_minus_good_bytes) | \
823 			       (1ULL << EF10_STAT_rx_good_bytes) |	\
824 			       (1ULL << EF10_STAT_rx_bad_bytes) |	\
825 			       (1ULL << EF10_STAT_rx_packets) |		\
826 			       (1ULL << EF10_STAT_rx_good) |		\
827 			       (1ULL << EF10_STAT_rx_bad) |		\
828 			       (1ULL << EF10_STAT_rx_pause) |		\
829 			       (1ULL << EF10_STAT_rx_control) |		\
830 			       (1ULL << EF10_STAT_rx_unicast) |		\
831 			       (1ULL << EF10_STAT_rx_multicast) |	\
832 			       (1ULL << EF10_STAT_rx_broadcast) |	\
833 			       (1ULL << EF10_STAT_rx_lt64) |		\
834 			       (1ULL << EF10_STAT_rx_64) |		\
835 			       (1ULL << EF10_STAT_rx_65_to_127) |	\
836 			       (1ULL << EF10_STAT_rx_128_to_255) |	\
837 			       (1ULL << EF10_STAT_rx_256_to_511) |	\
838 			       (1ULL << EF10_STAT_rx_512_to_1023) |	\
839 			       (1ULL << EF10_STAT_rx_1024_to_15xx) |	\
840 			       (1ULL << EF10_STAT_rx_15xx_to_jumbo) |	\
841 			       (1ULL << EF10_STAT_rx_gtjumbo) |		\
842 			       (1ULL << EF10_STAT_rx_bad_gtjumbo) |	\
843 			       (1ULL << EF10_STAT_rx_overflow) |	\
844 			       (1ULL << EF10_STAT_rx_nodesc_drops))
845 
846 /* These statistics are only provided by the 10G MAC.  For a 10G/40G
847  * switchable port we do not expose these because they might not
848  * include all the packets they should.
849  */
850 #define HUNT_10G_ONLY_STAT_MASK ((1ULL << EF10_STAT_tx_control) |	\
851 				 (1ULL << EF10_STAT_tx_lt64) |		\
852 				 (1ULL << EF10_STAT_tx_64) |		\
853 				 (1ULL << EF10_STAT_tx_65_to_127) |	\
854 				 (1ULL << EF10_STAT_tx_128_to_255) |	\
855 				 (1ULL << EF10_STAT_tx_256_to_511) |	\
856 				 (1ULL << EF10_STAT_tx_512_to_1023) |	\
857 				 (1ULL << EF10_STAT_tx_1024_to_15xx) |	\
858 				 (1ULL << EF10_STAT_tx_15xx_to_jumbo))
859 
860 /* These statistics are only provided by the 40G MAC.  For a 10G/40G
861  * switchable port we do expose these because the errors will otherwise
862  * be silent.
863  */
864 #define HUNT_40G_EXTRA_STAT_MASK ((1ULL << EF10_STAT_rx_align_error) |	\
865 				  (1ULL << EF10_STAT_rx_length_error))
866 
867 /* These statistics are only provided if the firmware supports the
868  * capability PM_AND_RXDP_COUNTERS.
869  */
870 #define HUNT_PM_AND_RXDP_STAT_MASK (					\
871 	(1ULL << EF10_STAT_rx_pm_trunc_bb_overflow) |			\
872 	(1ULL << EF10_STAT_rx_pm_discard_bb_overflow) |			\
873 	(1ULL << EF10_STAT_rx_pm_trunc_vfifo_full) |			\
874 	(1ULL << EF10_STAT_rx_pm_discard_vfifo_full) |			\
875 	(1ULL << EF10_STAT_rx_pm_trunc_qbb) |				\
876 	(1ULL << EF10_STAT_rx_pm_discard_qbb) |				\
877 	(1ULL << EF10_STAT_rx_pm_discard_mapping) |			\
878 	(1ULL << EF10_STAT_rx_dp_q_disabled_packets) |			\
879 	(1ULL << EF10_STAT_rx_dp_di_dropped_packets) |			\
880 	(1ULL << EF10_STAT_rx_dp_streaming_packets) |			\
881 	(1ULL << EF10_STAT_rx_dp_hlb_fetch) |				\
882 	(1ULL << EF10_STAT_rx_dp_hlb_wait))
883 
884 static u64 efx_ef10_raw_stat_mask(struct efx_nic *efx)
885 {
886 	u64 raw_mask = HUNT_COMMON_STAT_MASK;
887 	u32 port_caps = efx_mcdi_phy_get_caps(efx);
888 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
889 
890 	if (port_caps & (1 << MC_CMD_PHY_CAP_40000FDX_LBN))
891 		raw_mask |= HUNT_40G_EXTRA_STAT_MASK;
892 	else
893 		raw_mask |= HUNT_10G_ONLY_STAT_MASK;
894 
895 	if (nic_data->datapath_caps &
896 	    (1 << MC_CMD_GET_CAPABILITIES_OUT_PM_AND_RXDP_COUNTERS_LBN))
897 		raw_mask |= HUNT_PM_AND_RXDP_STAT_MASK;
898 
899 	return raw_mask;
900 }
901 
902 static void efx_ef10_get_stat_mask(struct efx_nic *efx, unsigned long *mask)
903 {
904 	u64 raw_mask = efx_ef10_raw_stat_mask(efx);
905 
906 #if BITS_PER_LONG == 64
907 	mask[0] = raw_mask;
908 #else
909 	mask[0] = raw_mask & 0xffffffff;
910 	mask[1] = raw_mask >> 32;
911 #endif
912 }
913 
914 static size_t efx_ef10_describe_stats(struct efx_nic *efx, u8 *names)
915 {
916 	DECLARE_BITMAP(mask, EF10_STAT_COUNT);
917 
918 	efx_ef10_get_stat_mask(efx, mask);
919 	return efx_nic_describe_stats(efx_ef10_stat_desc, EF10_STAT_COUNT,
920 				      mask, names);
921 }
922 
923 static int efx_ef10_try_update_nic_stats(struct efx_nic *efx)
924 {
925 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
926 	DECLARE_BITMAP(mask, EF10_STAT_COUNT);
927 	__le64 generation_start, generation_end;
928 	u64 *stats = nic_data->stats;
929 	__le64 *dma_stats;
930 
931 	efx_ef10_get_stat_mask(efx, mask);
932 
933 	dma_stats = efx->stats_buffer.addr;
934 	nic_data = efx->nic_data;
935 
936 	generation_end = dma_stats[MC_CMD_MAC_GENERATION_END];
937 	if (generation_end == EFX_MC_STATS_GENERATION_INVALID)
938 		return 0;
939 	rmb();
940 	efx_nic_update_stats(efx_ef10_stat_desc, EF10_STAT_COUNT, mask,
941 			     stats, efx->stats_buffer.addr, false);
942 	rmb();
943 	generation_start = dma_stats[MC_CMD_MAC_GENERATION_START];
944 	if (generation_end != generation_start)
945 		return -EAGAIN;
946 
947 	/* Update derived statistics */
948 	efx_nic_fix_nodesc_drop_stat(efx, &stats[EF10_STAT_rx_nodesc_drops]);
949 	stats[EF10_STAT_rx_good_bytes] =
950 		stats[EF10_STAT_rx_bytes] -
951 		stats[EF10_STAT_rx_bytes_minus_good_bytes];
952 	efx_update_diff_stat(&stats[EF10_STAT_rx_bad_bytes],
953 			     stats[EF10_STAT_rx_bytes_minus_good_bytes]);
954 
955 	return 0;
956 }
957 
958 
959 static size_t efx_ef10_update_stats(struct efx_nic *efx, u64 *full_stats,
960 				    struct rtnl_link_stats64 *core_stats)
961 {
962 	DECLARE_BITMAP(mask, EF10_STAT_COUNT);
963 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
964 	u64 *stats = nic_data->stats;
965 	size_t stats_count = 0, index;
966 	int retry;
967 
968 	efx_ef10_get_stat_mask(efx, mask);
969 
970 	/* If we're unlucky enough to read statistics during the DMA, wait
971 	 * up to 10ms for it to finish (typically takes <500us)
972 	 */
973 	for (retry = 0; retry < 100; ++retry) {
974 		if (efx_ef10_try_update_nic_stats(efx) == 0)
975 			break;
976 		udelay(100);
977 	}
978 
979 	if (full_stats) {
980 		for_each_set_bit(index, mask, EF10_STAT_COUNT) {
981 			if (efx_ef10_stat_desc[index].name) {
982 				*full_stats++ = stats[index];
983 				++stats_count;
984 			}
985 		}
986 	}
987 
988 	if (core_stats) {
989 		core_stats->rx_packets = stats[EF10_STAT_rx_packets];
990 		core_stats->tx_packets = stats[EF10_STAT_tx_packets];
991 		core_stats->rx_bytes = stats[EF10_STAT_rx_bytes];
992 		core_stats->tx_bytes = stats[EF10_STAT_tx_bytes];
993 		core_stats->rx_dropped = stats[EF10_STAT_rx_nodesc_drops];
994 		core_stats->multicast = stats[EF10_STAT_rx_multicast];
995 		core_stats->rx_length_errors =
996 			stats[EF10_STAT_rx_gtjumbo] +
997 			stats[EF10_STAT_rx_length_error];
998 		core_stats->rx_crc_errors = stats[EF10_STAT_rx_bad];
999 		core_stats->rx_frame_errors = stats[EF10_STAT_rx_align_error];
1000 		core_stats->rx_fifo_errors = stats[EF10_STAT_rx_overflow];
1001 		core_stats->rx_errors = (core_stats->rx_length_errors +
1002 					 core_stats->rx_crc_errors +
1003 					 core_stats->rx_frame_errors);
1004 	}
1005 
1006 	return stats_count;
1007 }
1008 
1009 static void efx_ef10_push_irq_moderation(struct efx_channel *channel)
1010 {
1011 	struct efx_nic *efx = channel->efx;
1012 	unsigned int mode, value;
1013 	efx_dword_t timer_cmd;
1014 
1015 	if (channel->irq_moderation) {
1016 		mode = 3;
1017 		value = channel->irq_moderation - 1;
1018 	} else {
1019 		mode = 0;
1020 		value = 0;
1021 	}
1022 
1023 	if (EFX_EF10_WORKAROUND_35388(efx)) {
1024 		EFX_POPULATE_DWORD_3(timer_cmd, ERF_DD_EVQ_IND_TIMER_FLAGS,
1025 				     EFE_DD_EVQ_IND_TIMER_FLAGS,
1026 				     ERF_DD_EVQ_IND_TIMER_MODE, mode,
1027 				     ERF_DD_EVQ_IND_TIMER_VAL, value);
1028 		efx_writed_page(efx, &timer_cmd, ER_DD_EVQ_INDIRECT,
1029 				channel->channel);
1030 	} else {
1031 		EFX_POPULATE_DWORD_2(timer_cmd, ERF_DZ_TC_TIMER_MODE, mode,
1032 				     ERF_DZ_TC_TIMER_VAL, value);
1033 		efx_writed_page(efx, &timer_cmd, ER_DZ_EVQ_TMR,
1034 				channel->channel);
1035 	}
1036 }
1037 
1038 static void efx_ef10_get_wol(struct efx_nic *efx, struct ethtool_wolinfo *wol)
1039 {
1040 	wol->supported = 0;
1041 	wol->wolopts = 0;
1042 	memset(&wol->sopass, 0, sizeof(wol->sopass));
1043 }
1044 
1045 static int efx_ef10_set_wol(struct efx_nic *efx, u32 type)
1046 {
1047 	if (type != 0)
1048 		return -EINVAL;
1049 	return 0;
1050 }
1051 
1052 static void efx_ef10_mcdi_request(struct efx_nic *efx,
1053 				  const efx_dword_t *hdr, size_t hdr_len,
1054 				  const efx_dword_t *sdu, size_t sdu_len)
1055 {
1056 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1057 	u8 *pdu = nic_data->mcdi_buf.addr;
1058 
1059 	memcpy(pdu, hdr, hdr_len);
1060 	memcpy(pdu + hdr_len, sdu, sdu_len);
1061 	wmb();
1062 
1063 	/* The hardware provides 'low' and 'high' (doorbell) registers
1064 	 * for passing the 64-bit address of an MCDI request to
1065 	 * firmware.  However the dwords are swapped by firmware.  The
1066 	 * least significant bits of the doorbell are then 0 for all
1067 	 * MCDI requests due to alignment.
1068 	 */
1069 	_efx_writed(efx, cpu_to_le32((u64)nic_data->mcdi_buf.dma_addr >> 32),
1070 		    ER_DZ_MC_DB_LWRD);
1071 	_efx_writed(efx, cpu_to_le32((u32)nic_data->mcdi_buf.dma_addr),
1072 		    ER_DZ_MC_DB_HWRD);
1073 }
1074 
1075 static bool efx_ef10_mcdi_poll_response(struct efx_nic *efx)
1076 {
1077 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1078 	const efx_dword_t hdr = *(const efx_dword_t *)nic_data->mcdi_buf.addr;
1079 
1080 	rmb();
1081 	return EFX_DWORD_FIELD(hdr, MCDI_HEADER_RESPONSE);
1082 }
1083 
1084 static void
1085 efx_ef10_mcdi_read_response(struct efx_nic *efx, efx_dword_t *outbuf,
1086 			    size_t offset, size_t outlen)
1087 {
1088 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1089 	const u8 *pdu = nic_data->mcdi_buf.addr;
1090 
1091 	memcpy(outbuf, pdu + offset, outlen);
1092 }
1093 
1094 static int efx_ef10_mcdi_poll_reboot(struct efx_nic *efx)
1095 {
1096 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1097 	int rc;
1098 
1099 	rc = efx_ef10_get_warm_boot_count(efx);
1100 	if (rc < 0) {
1101 		/* The firmware is presumably in the process of
1102 		 * rebooting.  However, we are supposed to report each
1103 		 * reboot just once, so we must only do that once we
1104 		 * can read and store the updated warm boot count.
1105 		 */
1106 		return 0;
1107 	}
1108 
1109 	if (rc == nic_data->warm_boot_count)
1110 		return 0;
1111 
1112 	nic_data->warm_boot_count = rc;
1113 
1114 	/* All our allocations have been reset */
1115 	efx_ef10_reset_mc_allocations(efx);
1116 
1117 	/* The datapath firmware might have been changed */
1118 	nic_data->must_check_datapath_caps = true;
1119 
1120 	/* MAC statistics have been cleared on the NIC; clear the local
1121 	 * statistic that we update with efx_update_diff_stat().
1122 	 */
1123 	nic_data->stats[EF10_STAT_rx_bad_bytes] = 0;
1124 
1125 	return -EIO;
1126 }
1127 
1128 /* Handle an MSI interrupt
1129  *
1130  * Handle an MSI hardware interrupt.  This routine schedules event
1131  * queue processing.  No interrupt acknowledgement cycle is necessary.
1132  * Also, we never need to check that the interrupt is for us, since
1133  * MSI interrupts cannot be shared.
1134  */
1135 static irqreturn_t efx_ef10_msi_interrupt(int irq, void *dev_id)
1136 {
1137 	struct efx_msi_context *context = dev_id;
1138 	struct efx_nic *efx = context->efx;
1139 
1140 	netif_vdbg(efx, intr, efx->net_dev,
1141 		   "IRQ %d on CPU %d\n", irq, raw_smp_processor_id());
1142 
1143 	if (likely(ACCESS_ONCE(efx->irq_soft_enabled))) {
1144 		/* Note test interrupts */
1145 		if (context->index == efx->irq_level)
1146 			efx->last_irq_cpu = raw_smp_processor_id();
1147 
1148 		/* Schedule processing of the channel */
1149 		efx_schedule_channel_irq(efx->channel[context->index]);
1150 	}
1151 
1152 	return IRQ_HANDLED;
1153 }
1154 
1155 static irqreturn_t efx_ef10_legacy_interrupt(int irq, void *dev_id)
1156 {
1157 	struct efx_nic *efx = dev_id;
1158 	bool soft_enabled = ACCESS_ONCE(efx->irq_soft_enabled);
1159 	struct efx_channel *channel;
1160 	efx_dword_t reg;
1161 	u32 queues;
1162 
1163 	/* Read the ISR which also ACKs the interrupts */
1164 	efx_readd(efx, &reg, ER_DZ_BIU_INT_ISR);
1165 	queues = EFX_DWORD_FIELD(reg, ERF_DZ_ISR_REG);
1166 
1167 	if (queues == 0)
1168 		return IRQ_NONE;
1169 
1170 	if (likely(soft_enabled)) {
1171 		/* Note test interrupts */
1172 		if (queues & (1U << efx->irq_level))
1173 			efx->last_irq_cpu = raw_smp_processor_id();
1174 
1175 		efx_for_each_channel(channel, efx) {
1176 			if (queues & 1)
1177 				efx_schedule_channel_irq(channel);
1178 			queues >>= 1;
1179 		}
1180 	}
1181 
1182 	netif_vdbg(efx, intr, efx->net_dev,
1183 		   "IRQ %d on CPU %d status " EFX_DWORD_FMT "\n",
1184 		   irq, raw_smp_processor_id(), EFX_DWORD_VAL(reg));
1185 
1186 	return IRQ_HANDLED;
1187 }
1188 
1189 static void efx_ef10_irq_test_generate(struct efx_nic *efx)
1190 {
1191 	MCDI_DECLARE_BUF(inbuf, MC_CMD_TRIGGER_INTERRUPT_IN_LEN);
1192 
1193 	BUILD_BUG_ON(MC_CMD_TRIGGER_INTERRUPT_OUT_LEN != 0);
1194 
1195 	MCDI_SET_DWORD(inbuf, TRIGGER_INTERRUPT_IN_INTR_LEVEL, efx->irq_level);
1196 	(void) efx_mcdi_rpc(efx, MC_CMD_TRIGGER_INTERRUPT,
1197 			    inbuf, sizeof(inbuf), NULL, 0, NULL);
1198 }
1199 
1200 static int efx_ef10_tx_probe(struct efx_tx_queue *tx_queue)
1201 {
1202 	return efx_nic_alloc_buffer(tx_queue->efx, &tx_queue->txd.buf,
1203 				    (tx_queue->ptr_mask + 1) *
1204 				    sizeof(efx_qword_t),
1205 				    GFP_KERNEL);
1206 }
1207 
1208 /* This writes to the TX_DESC_WPTR and also pushes data */
1209 static inline void efx_ef10_push_tx_desc(struct efx_tx_queue *tx_queue,
1210 					 const efx_qword_t *txd)
1211 {
1212 	unsigned int write_ptr;
1213 	efx_oword_t reg;
1214 
1215 	write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
1216 	EFX_POPULATE_OWORD_1(reg, ERF_DZ_TX_DESC_WPTR, write_ptr);
1217 	reg.qword[0] = *txd;
1218 	efx_writeo_page(tx_queue->efx, &reg,
1219 			ER_DZ_TX_DESC_UPD, tx_queue->queue);
1220 }
1221 
1222 static void efx_ef10_tx_init(struct efx_tx_queue *tx_queue)
1223 {
1224 	MCDI_DECLARE_BUF(inbuf, MC_CMD_INIT_TXQ_IN_LEN(EFX_MAX_DMAQ_SIZE * 8 /
1225 						       EFX_BUF_SIZE));
1226 	MCDI_DECLARE_BUF(outbuf, MC_CMD_INIT_TXQ_OUT_LEN);
1227 	bool csum_offload = tx_queue->queue & EFX_TXQ_TYPE_OFFLOAD;
1228 	size_t entries = tx_queue->txd.buf.len / EFX_BUF_SIZE;
1229 	struct efx_channel *channel = tx_queue->channel;
1230 	struct efx_nic *efx = tx_queue->efx;
1231 	size_t inlen, outlen;
1232 	dma_addr_t dma_addr;
1233 	efx_qword_t *txd;
1234 	int rc;
1235 	int i;
1236 
1237 	MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_SIZE, tx_queue->ptr_mask + 1);
1238 	MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_TARGET_EVQ, channel->channel);
1239 	MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_LABEL, tx_queue->queue);
1240 	MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_INSTANCE, tx_queue->queue);
1241 	MCDI_POPULATE_DWORD_2(inbuf, INIT_TXQ_IN_FLAGS,
1242 			      INIT_TXQ_IN_FLAG_IP_CSUM_DIS, !csum_offload,
1243 			      INIT_TXQ_IN_FLAG_TCP_CSUM_DIS, !csum_offload);
1244 	MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_OWNER_ID, 0);
1245 	MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_PORT_ID, EVB_PORT_ID_ASSIGNED);
1246 
1247 	dma_addr = tx_queue->txd.buf.dma_addr;
1248 
1249 	netif_dbg(efx, hw, efx->net_dev, "pushing TXQ %d. %zu entries (%llx)\n",
1250 		  tx_queue->queue, entries, (u64)dma_addr);
1251 
1252 	for (i = 0; i < entries; ++i) {
1253 		MCDI_SET_ARRAY_QWORD(inbuf, INIT_TXQ_IN_DMA_ADDR, i, dma_addr);
1254 		dma_addr += EFX_BUF_SIZE;
1255 	}
1256 
1257 	inlen = MC_CMD_INIT_TXQ_IN_LEN(entries);
1258 
1259 	rc = efx_mcdi_rpc(efx, MC_CMD_INIT_TXQ, inbuf, inlen,
1260 			  outbuf, sizeof(outbuf), &outlen);
1261 	if (rc)
1262 		goto fail;
1263 
1264 	/* A previous user of this TX queue might have set us up the
1265 	 * bomb by writing a descriptor to the TX push collector but
1266 	 * not the doorbell.  (Each collector belongs to a port, not a
1267 	 * queue or function, so cannot easily be reset.)  We must
1268 	 * attempt to push a no-op descriptor in its place.
1269 	 */
1270 	tx_queue->buffer[0].flags = EFX_TX_BUF_OPTION;
1271 	tx_queue->insert_count = 1;
1272 	txd = efx_tx_desc(tx_queue, 0);
1273 	EFX_POPULATE_QWORD_4(*txd,
1274 			     ESF_DZ_TX_DESC_IS_OPT, true,
1275 			     ESF_DZ_TX_OPTION_TYPE,
1276 			     ESE_DZ_TX_OPTION_DESC_CRC_CSUM,
1277 			     ESF_DZ_TX_OPTION_UDP_TCP_CSUM, csum_offload,
1278 			     ESF_DZ_TX_OPTION_IP_CSUM, csum_offload);
1279 	tx_queue->write_count = 1;
1280 	wmb();
1281 	efx_ef10_push_tx_desc(tx_queue, txd);
1282 
1283 	return;
1284 
1285 fail:
1286 	netdev_WARN(efx->net_dev, "failed to initialise TXQ %d\n",
1287 		    tx_queue->queue);
1288 }
1289 
1290 static void efx_ef10_tx_fini(struct efx_tx_queue *tx_queue)
1291 {
1292 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FINI_TXQ_IN_LEN);
1293 	MCDI_DECLARE_BUF(outbuf, MC_CMD_FINI_TXQ_OUT_LEN);
1294 	struct efx_nic *efx = tx_queue->efx;
1295 	size_t outlen;
1296 	int rc;
1297 
1298 	MCDI_SET_DWORD(inbuf, FINI_TXQ_IN_INSTANCE,
1299 		       tx_queue->queue);
1300 
1301 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FINI_TXQ, inbuf, sizeof(inbuf),
1302 			  outbuf, sizeof(outbuf), &outlen);
1303 
1304 	if (rc && rc != -EALREADY)
1305 		goto fail;
1306 
1307 	return;
1308 
1309 fail:
1310 	efx_mcdi_display_error(efx, MC_CMD_FINI_TXQ, MC_CMD_FINI_TXQ_IN_LEN,
1311 			       outbuf, outlen, rc);
1312 }
1313 
1314 static void efx_ef10_tx_remove(struct efx_tx_queue *tx_queue)
1315 {
1316 	efx_nic_free_buffer(tx_queue->efx, &tx_queue->txd.buf);
1317 }
1318 
1319 /* This writes to the TX_DESC_WPTR; write pointer for TX descriptor ring */
1320 static inline void efx_ef10_notify_tx_desc(struct efx_tx_queue *tx_queue)
1321 {
1322 	unsigned int write_ptr;
1323 	efx_dword_t reg;
1324 
1325 	write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
1326 	EFX_POPULATE_DWORD_1(reg, ERF_DZ_TX_DESC_WPTR_DWORD, write_ptr);
1327 	efx_writed_page(tx_queue->efx, &reg,
1328 			ER_DZ_TX_DESC_UPD_DWORD, tx_queue->queue);
1329 }
1330 
1331 static void efx_ef10_tx_write(struct efx_tx_queue *tx_queue)
1332 {
1333 	unsigned int old_write_count = tx_queue->write_count;
1334 	struct efx_tx_buffer *buffer;
1335 	unsigned int write_ptr;
1336 	efx_qword_t *txd;
1337 
1338 	BUG_ON(tx_queue->write_count == tx_queue->insert_count);
1339 
1340 	do {
1341 		write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
1342 		buffer = &tx_queue->buffer[write_ptr];
1343 		txd = efx_tx_desc(tx_queue, write_ptr);
1344 		++tx_queue->write_count;
1345 
1346 		/* Create TX descriptor ring entry */
1347 		if (buffer->flags & EFX_TX_BUF_OPTION) {
1348 			*txd = buffer->option;
1349 		} else {
1350 			BUILD_BUG_ON(EFX_TX_BUF_CONT != 1);
1351 			EFX_POPULATE_QWORD_3(
1352 				*txd,
1353 				ESF_DZ_TX_KER_CONT,
1354 				buffer->flags & EFX_TX_BUF_CONT,
1355 				ESF_DZ_TX_KER_BYTE_CNT, buffer->len,
1356 				ESF_DZ_TX_KER_BUF_ADDR, buffer->dma_addr);
1357 		}
1358 	} while (tx_queue->write_count != tx_queue->insert_count);
1359 
1360 	wmb(); /* Ensure descriptors are written before they are fetched */
1361 
1362 	if (efx_nic_may_push_tx_desc(tx_queue, old_write_count)) {
1363 		txd = efx_tx_desc(tx_queue,
1364 				  old_write_count & tx_queue->ptr_mask);
1365 		efx_ef10_push_tx_desc(tx_queue, txd);
1366 		++tx_queue->pushes;
1367 	} else {
1368 		efx_ef10_notify_tx_desc(tx_queue);
1369 	}
1370 }
1371 
1372 static int efx_ef10_alloc_rss_context(struct efx_nic *efx, u32 *context)
1373 {
1374 	MCDI_DECLARE_BUF(inbuf, MC_CMD_RSS_CONTEXT_ALLOC_IN_LEN);
1375 	MCDI_DECLARE_BUF(outbuf, MC_CMD_RSS_CONTEXT_ALLOC_OUT_LEN);
1376 	size_t outlen;
1377 	int rc;
1378 
1379 	MCDI_SET_DWORD(inbuf, RSS_CONTEXT_ALLOC_IN_UPSTREAM_PORT_ID,
1380 		       EVB_PORT_ID_ASSIGNED);
1381 	MCDI_SET_DWORD(inbuf, RSS_CONTEXT_ALLOC_IN_TYPE,
1382 		       MC_CMD_RSS_CONTEXT_ALLOC_IN_TYPE_EXCLUSIVE);
1383 	MCDI_SET_DWORD(inbuf, RSS_CONTEXT_ALLOC_IN_NUM_QUEUES,
1384 		       EFX_MAX_CHANNELS);
1385 
1386 	rc = efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_ALLOC, inbuf, sizeof(inbuf),
1387 		outbuf, sizeof(outbuf), &outlen);
1388 	if (rc != 0)
1389 		return rc;
1390 
1391 	if (outlen < MC_CMD_RSS_CONTEXT_ALLOC_OUT_LEN)
1392 		return -EIO;
1393 
1394 	*context = MCDI_DWORD(outbuf, RSS_CONTEXT_ALLOC_OUT_RSS_CONTEXT_ID);
1395 
1396 	return 0;
1397 }
1398 
1399 static void efx_ef10_free_rss_context(struct efx_nic *efx, u32 context)
1400 {
1401 	MCDI_DECLARE_BUF(inbuf, MC_CMD_RSS_CONTEXT_FREE_IN_LEN);
1402 	int rc;
1403 
1404 	MCDI_SET_DWORD(inbuf, RSS_CONTEXT_FREE_IN_RSS_CONTEXT_ID,
1405 		       context);
1406 
1407 	rc = efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_FREE, inbuf, sizeof(inbuf),
1408 			    NULL, 0, NULL);
1409 	WARN_ON(rc != 0);
1410 }
1411 
1412 static int efx_ef10_populate_rss_table(struct efx_nic *efx, u32 context)
1413 {
1414 	MCDI_DECLARE_BUF(tablebuf, MC_CMD_RSS_CONTEXT_SET_TABLE_IN_LEN);
1415 	MCDI_DECLARE_BUF(keybuf, MC_CMD_RSS_CONTEXT_SET_KEY_IN_LEN);
1416 	int i, rc;
1417 
1418 	MCDI_SET_DWORD(tablebuf, RSS_CONTEXT_SET_TABLE_IN_RSS_CONTEXT_ID,
1419 		       context);
1420 	BUILD_BUG_ON(ARRAY_SIZE(efx->rx_indir_table) !=
1421 		     MC_CMD_RSS_CONTEXT_SET_TABLE_IN_INDIRECTION_TABLE_LEN);
1422 
1423 	for (i = 0; i < ARRAY_SIZE(efx->rx_indir_table); ++i)
1424 		MCDI_PTR(tablebuf,
1425 			 RSS_CONTEXT_SET_TABLE_IN_INDIRECTION_TABLE)[i] =
1426 				(u8) efx->rx_indir_table[i];
1427 
1428 	rc = efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_SET_TABLE, tablebuf,
1429 			  sizeof(tablebuf), NULL, 0, NULL);
1430 	if (rc != 0)
1431 		return rc;
1432 
1433 	MCDI_SET_DWORD(keybuf, RSS_CONTEXT_SET_KEY_IN_RSS_CONTEXT_ID,
1434 		       context);
1435 	BUILD_BUG_ON(ARRAY_SIZE(efx->rx_hash_key) !=
1436 		     MC_CMD_RSS_CONTEXT_SET_KEY_IN_TOEPLITZ_KEY_LEN);
1437 	for (i = 0; i < ARRAY_SIZE(efx->rx_hash_key); ++i)
1438 		MCDI_PTR(keybuf, RSS_CONTEXT_SET_KEY_IN_TOEPLITZ_KEY)[i] =
1439 			efx->rx_hash_key[i];
1440 
1441 	return efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_SET_KEY, keybuf,
1442 			    sizeof(keybuf), NULL, 0, NULL);
1443 }
1444 
1445 static void efx_ef10_rx_free_indir_table(struct efx_nic *efx)
1446 {
1447 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1448 
1449 	if (nic_data->rx_rss_context != EFX_EF10_RSS_CONTEXT_INVALID)
1450 		efx_ef10_free_rss_context(efx, nic_data->rx_rss_context);
1451 	nic_data->rx_rss_context = EFX_EF10_RSS_CONTEXT_INVALID;
1452 }
1453 
1454 static void efx_ef10_rx_push_rss_config(struct efx_nic *efx)
1455 {
1456 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1457 	int rc;
1458 
1459 	netif_dbg(efx, drv, efx->net_dev, "pushing RSS config\n");
1460 
1461 	if (nic_data->rx_rss_context == EFX_EF10_RSS_CONTEXT_INVALID) {
1462 		rc = efx_ef10_alloc_rss_context(efx, &nic_data->rx_rss_context);
1463 		if (rc != 0)
1464 			goto fail;
1465 	}
1466 
1467 	rc = efx_ef10_populate_rss_table(efx, nic_data->rx_rss_context);
1468 	if (rc != 0)
1469 		goto fail;
1470 
1471 	return;
1472 
1473 fail:
1474 	netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
1475 }
1476 
1477 static int efx_ef10_rx_probe(struct efx_rx_queue *rx_queue)
1478 {
1479 	return efx_nic_alloc_buffer(rx_queue->efx, &rx_queue->rxd.buf,
1480 				    (rx_queue->ptr_mask + 1) *
1481 				    sizeof(efx_qword_t),
1482 				    GFP_KERNEL);
1483 }
1484 
1485 static void efx_ef10_rx_init(struct efx_rx_queue *rx_queue)
1486 {
1487 	MCDI_DECLARE_BUF(inbuf,
1488 			 MC_CMD_INIT_RXQ_IN_LEN(EFX_MAX_DMAQ_SIZE * 8 /
1489 						EFX_BUF_SIZE));
1490 	MCDI_DECLARE_BUF(outbuf, MC_CMD_INIT_RXQ_OUT_LEN);
1491 	struct efx_channel *channel = efx_rx_queue_channel(rx_queue);
1492 	size_t entries = rx_queue->rxd.buf.len / EFX_BUF_SIZE;
1493 	struct efx_nic *efx = rx_queue->efx;
1494 	size_t inlen, outlen;
1495 	dma_addr_t dma_addr;
1496 	int rc;
1497 	int i;
1498 
1499 	rx_queue->scatter_n = 0;
1500 	rx_queue->scatter_len = 0;
1501 
1502 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_SIZE, rx_queue->ptr_mask + 1);
1503 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_TARGET_EVQ, channel->channel);
1504 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_LABEL, efx_rx_queue_index(rx_queue));
1505 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_INSTANCE,
1506 		       efx_rx_queue_index(rx_queue));
1507 	MCDI_POPULATE_DWORD_2(inbuf, INIT_RXQ_IN_FLAGS,
1508 			      INIT_RXQ_IN_FLAG_PREFIX, 1,
1509 			      INIT_RXQ_IN_FLAG_TIMESTAMP, 1);
1510 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_OWNER_ID, 0);
1511 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_PORT_ID, EVB_PORT_ID_ASSIGNED);
1512 
1513 	dma_addr = rx_queue->rxd.buf.dma_addr;
1514 
1515 	netif_dbg(efx, hw, efx->net_dev, "pushing RXQ %d. %zu entries (%llx)\n",
1516 		  efx_rx_queue_index(rx_queue), entries, (u64)dma_addr);
1517 
1518 	for (i = 0; i < entries; ++i) {
1519 		MCDI_SET_ARRAY_QWORD(inbuf, INIT_RXQ_IN_DMA_ADDR, i, dma_addr);
1520 		dma_addr += EFX_BUF_SIZE;
1521 	}
1522 
1523 	inlen = MC_CMD_INIT_RXQ_IN_LEN(entries);
1524 
1525 	rc = efx_mcdi_rpc(efx, MC_CMD_INIT_RXQ, inbuf, inlen,
1526 			  outbuf, sizeof(outbuf), &outlen);
1527 	if (rc)
1528 		netdev_WARN(efx->net_dev, "failed to initialise RXQ %d\n",
1529 			    efx_rx_queue_index(rx_queue));
1530 }
1531 
1532 static void efx_ef10_rx_fini(struct efx_rx_queue *rx_queue)
1533 {
1534 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FINI_RXQ_IN_LEN);
1535 	MCDI_DECLARE_BUF(outbuf, MC_CMD_FINI_RXQ_OUT_LEN);
1536 	struct efx_nic *efx = rx_queue->efx;
1537 	size_t outlen;
1538 	int rc;
1539 
1540 	MCDI_SET_DWORD(inbuf, FINI_RXQ_IN_INSTANCE,
1541 		       efx_rx_queue_index(rx_queue));
1542 
1543 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FINI_RXQ, inbuf, sizeof(inbuf),
1544 			  outbuf, sizeof(outbuf), &outlen);
1545 
1546 	if (rc && rc != -EALREADY)
1547 		goto fail;
1548 
1549 	return;
1550 
1551 fail:
1552 	efx_mcdi_display_error(efx, MC_CMD_FINI_RXQ, MC_CMD_FINI_RXQ_IN_LEN,
1553 			       outbuf, outlen, rc);
1554 }
1555 
1556 static void efx_ef10_rx_remove(struct efx_rx_queue *rx_queue)
1557 {
1558 	efx_nic_free_buffer(rx_queue->efx, &rx_queue->rxd.buf);
1559 }
1560 
1561 /* This creates an entry in the RX descriptor queue */
1562 static inline void
1563 efx_ef10_build_rx_desc(struct efx_rx_queue *rx_queue, unsigned int index)
1564 {
1565 	struct efx_rx_buffer *rx_buf;
1566 	efx_qword_t *rxd;
1567 
1568 	rxd = efx_rx_desc(rx_queue, index);
1569 	rx_buf = efx_rx_buffer(rx_queue, index);
1570 	EFX_POPULATE_QWORD_2(*rxd,
1571 			     ESF_DZ_RX_KER_BYTE_CNT, rx_buf->len,
1572 			     ESF_DZ_RX_KER_BUF_ADDR, rx_buf->dma_addr);
1573 }
1574 
1575 static void efx_ef10_rx_write(struct efx_rx_queue *rx_queue)
1576 {
1577 	struct efx_nic *efx = rx_queue->efx;
1578 	unsigned int write_count;
1579 	efx_dword_t reg;
1580 
1581 	/* Firmware requires that RX_DESC_WPTR be a multiple of 8 */
1582 	write_count = rx_queue->added_count & ~7;
1583 	if (rx_queue->notified_count == write_count)
1584 		return;
1585 
1586 	do
1587 		efx_ef10_build_rx_desc(
1588 			rx_queue,
1589 			rx_queue->notified_count & rx_queue->ptr_mask);
1590 	while (++rx_queue->notified_count != write_count);
1591 
1592 	wmb();
1593 	EFX_POPULATE_DWORD_1(reg, ERF_DZ_RX_DESC_WPTR,
1594 			     write_count & rx_queue->ptr_mask);
1595 	efx_writed_page(efx, &reg, ER_DZ_RX_DESC_UPD,
1596 			efx_rx_queue_index(rx_queue));
1597 }
1598 
1599 static efx_mcdi_async_completer efx_ef10_rx_defer_refill_complete;
1600 
1601 static void efx_ef10_rx_defer_refill(struct efx_rx_queue *rx_queue)
1602 {
1603 	struct efx_channel *channel = efx_rx_queue_channel(rx_queue);
1604 	MCDI_DECLARE_BUF(inbuf, MC_CMD_DRIVER_EVENT_IN_LEN);
1605 	efx_qword_t event;
1606 
1607 	EFX_POPULATE_QWORD_2(event,
1608 			     ESF_DZ_EV_CODE, EFX_EF10_DRVGEN_EV,
1609 			     ESF_DZ_EV_DATA, EFX_EF10_REFILL);
1610 
1611 	MCDI_SET_DWORD(inbuf, DRIVER_EVENT_IN_EVQ, channel->channel);
1612 
1613 	/* MCDI_SET_QWORD is not appropriate here since EFX_POPULATE_* has
1614 	 * already swapped the data to little-endian order.
1615 	 */
1616 	memcpy(MCDI_PTR(inbuf, DRIVER_EVENT_IN_DATA), &event.u64[0],
1617 	       sizeof(efx_qword_t));
1618 
1619 	efx_mcdi_rpc_async(channel->efx, MC_CMD_DRIVER_EVENT,
1620 			   inbuf, sizeof(inbuf), 0,
1621 			   efx_ef10_rx_defer_refill_complete, 0);
1622 }
1623 
1624 static void
1625 efx_ef10_rx_defer_refill_complete(struct efx_nic *efx, unsigned long cookie,
1626 				  int rc, efx_dword_t *outbuf,
1627 				  size_t outlen_actual)
1628 {
1629 	/* nothing to do */
1630 }
1631 
1632 static int efx_ef10_ev_probe(struct efx_channel *channel)
1633 {
1634 	return efx_nic_alloc_buffer(channel->efx, &channel->eventq.buf,
1635 				    (channel->eventq_mask + 1) *
1636 				    sizeof(efx_qword_t),
1637 				    GFP_KERNEL);
1638 }
1639 
1640 static int efx_ef10_ev_init(struct efx_channel *channel)
1641 {
1642 	MCDI_DECLARE_BUF(inbuf,
1643 			 MC_CMD_INIT_EVQ_IN_LEN(EFX_MAX_EVQ_SIZE * 8 /
1644 						EFX_BUF_SIZE));
1645 	MCDI_DECLARE_BUF(outbuf, MC_CMD_INIT_EVQ_OUT_LEN);
1646 	size_t entries = channel->eventq.buf.len / EFX_BUF_SIZE;
1647 	struct efx_nic *efx = channel->efx;
1648 	struct efx_ef10_nic_data *nic_data;
1649 	bool supports_rx_merge;
1650 	size_t inlen, outlen;
1651 	dma_addr_t dma_addr;
1652 	int rc;
1653 	int i;
1654 
1655 	nic_data = efx->nic_data;
1656 	supports_rx_merge =
1657 		!!(nic_data->datapath_caps &
1658 		   1 << MC_CMD_GET_CAPABILITIES_OUT_RX_BATCHING_LBN);
1659 
1660 	/* Fill event queue with all ones (i.e. empty events) */
1661 	memset(channel->eventq.buf.addr, 0xff, channel->eventq.buf.len);
1662 
1663 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_SIZE, channel->eventq_mask + 1);
1664 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_INSTANCE, channel->channel);
1665 	/* INIT_EVQ expects index in vector table, not absolute */
1666 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_IRQ_NUM, channel->channel);
1667 	MCDI_POPULATE_DWORD_4(inbuf, INIT_EVQ_IN_FLAGS,
1668 			      INIT_EVQ_IN_FLAG_INTERRUPTING, 1,
1669 			      INIT_EVQ_IN_FLAG_RX_MERGE, 1,
1670 			      INIT_EVQ_IN_FLAG_TX_MERGE, 1,
1671 			      INIT_EVQ_IN_FLAG_CUT_THRU, !supports_rx_merge);
1672 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_TMR_MODE,
1673 		       MC_CMD_INIT_EVQ_IN_TMR_MODE_DIS);
1674 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_TMR_LOAD, 0);
1675 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_TMR_RELOAD, 0);
1676 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_COUNT_MODE,
1677 		       MC_CMD_INIT_EVQ_IN_COUNT_MODE_DIS);
1678 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_COUNT_THRSHLD, 0);
1679 
1680 	dma_addr = channel->eventq.buf.dma_addr;
1681 	for (i = 0; i < entries; ++i) {
1682 		MCDI_SET_ARRAY_QWORD(inbuf, INIT_EVQ_IN_DMA_ADDR, i, dma_addr);
1683 		dma_addr += EFX_BUF_SIZE;
1684 	}
1685 
1686 	inlen = MC_CMD_INIT_EVQ_IN_LEN(entries);
1687 
1688 	rc = efx_mcdi_rpc(efx, MC_CMD_INIT_EVQ, inbuf, inlen,
1689 			  outbuf, sizeof(outbuf), &outlen);
1690 	/* IRQ return is ignored */
1691 	return rc;
1692 }
1693 
1694 static void efx_ef10_ev_fini(struct efx_channel *channel)
1695 {
1696 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FINI_EVQ_IN_LEN);
1697 	MCDI_DECLARE_BUF(outbuf, MC_CMD_FINI_EVQ_OUT_LEN);
1698 	struct efx_nic *efx = channel->efx;
1699 	size_t outlen;
1700 	int rc;
1701 
1702 	MCDI_SET_DWORD(inbuf, FINI_EVQ_IN_INSTANCE, channel->channel);
1703 
1704 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FINI_EVQ, inbuf, sizeof(inbuf),
1705 			  outbuf, sizeof(outbuf), &outlen);
1706 
1707 	if (rc && rc != -EALREADY)
1708 		goto fail;
1709 
1710 	return;
1711 
1712 fail:
1713 	efx_mcdi_display_error(efx, MC_CMD_FINI_EVQ, MC_CMD_FINI_EVQ_IN_LEN,
1714 			       outbuf, outlen, rc);
1715 }
1716 
1717 static void efx_ef10_ev_remove(struct efx_channel *channel)
1718 {
1719 	efx_nic_free_buffer(channel->efx, &channel->eventq.buf);
1720 }
1721 
1722 static void efx_ef10_handle_rx_wrong_queue(struct efx_rx_queue *rx_queue,
1723 					   unsigned int rx_queue_label)
1724 {
1725 	struct efx_nic *efx = rx_queue->efx;
1726 
1727 	netif_info(efx, hw, efx->net_dev,
1728 		   "rx event arrived on queue %d labeled as queue %u\n",
1729 		   efx_rx_queue_index(rx_queue), rx_queue_label);
1730 
1731 	efx_schedule_reset(efx, RESET_TYPE_DISABLE);
1732 }
1733 
1734 static void
1735 efx_ef10_handle_rx_bad_lbits(struct efx_rx_queue *rx_queue,
1736 			     unsigned int actual, unsigned int expected)
1737 {
1738 	unsigned int dropped = (actual - expected) & rx_queue->ptr_mask;
1739 	struct efx_nic *efx = rx_queue->efx;
1740 
1741 	netif_info(efx, hw, efx->net_dev,
1742 		   "dropped %d events (index=%d expected=%d)\n",
1743 		   dropped, actual, expected);
1744 
1745 	efx_schedule_reset(efx, RESET_TYPE_DISABLE);
1746 }
1747 
1748 /* partially received RX was aborted. clean up. */
1749 static void efx_ef10_handle_rx_abort(struct efx_rx_queue *rx_queue)
1750 {
1751 	unsigned int rx_desc_ptr;
1752 
1753 	netif_dbg(rx_queue->efx, hw, rx_queue->efx->net_dev,
1754 		  "scattered RX aborted (dropping %u buffers)\n",
1755 		  rx_queue->scatter_n);
1756 
1757 	rx_desc_ptr = rx_queue->removed_count & rx_queue->ptr_mask;
1758 
1759 	efx_rx_packet(rx_queue, rx_desc_ptr, rx_queue->scatter_n,
1760 		      0, EFX_RX_PKT_DISCARD);
1761 
1762 	rx_queue->removed_count += rx_queue->scatter_n;
1763 	rx_queue->scatter_n = 0;
1764 	rx_queue->scatter_len = 0;
1765 	++efx_rx_queue_channel(rx_queue)->n_rx_nodesc_trunc;
1766 }
1767 
1768 static int efx_ef10_handle_rx_event(struct efx_channel *channel,
1769 				    const efx_qword_t *event)
1770 {
1771 	unsigned int rx_bytes, next_ptr_lbits, rx_queue_label, rx_l4_class;
1772 	unsigned int n_descs, n_packets, i;
1773 	struct efx_nic *efx = channel->efx;
1774 	struct efx_rx_queue *rx_queue;
1775 	bool rx_cont;
1776 	u16 flags = 0;
1777 
1778 	if (unlikely(ACCESS_ONCE(efx->reset_pending)))
1779 		return 0;
1780 
1781 	/* Basic packet information */
1782 	rx_bytes = EFX_QWORD_FIELD(*event, ESF_DZ_RX_BYTES);
1783 	next_ptr_lbits = EFX_QWORD_FIELD(*event, ESF_DZ_RX_DSC_PTR_LBITS);
1784 	rx_queue_label = EFX_QWORD_FIELD(*event, ESF_DZ_RX_QLABEL);
1785 	rx_l4_class = EFX_QWORD_FIELD(*event, ESF_DZ_RX_L4_CLASS);
1786 	rx_cont = EFX_QWORD_FIELD(*event, ESF_DZ_RX_CONT);
1787 
1788 	if (EFX_QWORD_FIELD(*event, ESF_DZ_RX_DROP_EVENT))
1789 		netdev_WARN(efx->net_dev, "saw RX_DROP_EVENT: event="
1790 			    EFX_QWORD_FMT "\n",
1791 			    EFX_QWORD_VAL(*event));
1792 
1793 	rx_queue = efx_channel_get_rx_queue(channel);
1794 
1795 	if (unlikely(rx_queue_label != efx_rx_queue_index(rx_queue)))
1796 		efx_ef10_handle_rx_wrong_queue(rx_queue, rx_queue_label);
1797 
1798 	n_descs = ((next_ptr_lbits - rx_queue->removed_count) &
1799 		   ((1 << ESF_DZ_RX_DSC_PTR_LBITS_WIDTH) - 1));
1800 
1801 	if (n_descs != rx_queue->scatter_n + 1) {
1802 		struct efx_ef10_nic_data *nic_data = efx->nic_data;
1803 
1804 		/* detect rx abort */
1805 		if (unlikely(n_descs == rx_queue->scatter_n)) {
1806 			if (rx_queue->scatter_n == 0 || rx_bytes != 0)
1807 				netdev_WARN(efx->net_dev,
1808 					    "invalid RX abort: scatter_n=%u event="
1809 					    EFX_QWORD_FMT "\n",
1810 					    rx_queue->scatter_n,
1811 					    EFX_QWORD_VAL(*event));
1812 			efx_ef10_handle_rx_abort(rx_queue);
1813 			return 0;
1814 		}
1815 
1816 		/* Check that RX completion merging is valid, i.e.
1817 		 * the current firmware supports it and this is a
1818 		 * non-scattered packet.
1819 		 */
1820 		if (!(nic_data->datapath_caps &
1821 		      (1 << MC_CMD_GET_CAPABILITIES_OUT_RX_BATCHING_LBN)) ||
1822 		    rx_queue->scatter_n != 0 || rx_cont) {
1823 			efx_ef10_handle_rx_bad_lbits(
1824 				rx_queue, next_ptr_lbits,
1825 				(rx_queue->removed_count +
1826 				 rx_queue->scatter_n + 1) &
1827 				((1 << ESF_DZ_RX_DSC_PTR_LBITS_WIDTH) - 1));
1828 			return 0;
1829 		}
1830 
1831 		/* Merged completion for multiple non-scattered packets */
1832 		rx_queue->scatter_n = 1;
1833 		rx_queue->scatter_len = 0;
1834 		n_packets = n_descs;
1835 		++channel->n_rx_merge_events;
1836 		channel->n_rx_merge_packets += n_packets;
1837 		flags |= EFX_RX_PKT_PREFIX_LEN;
1838 	} else {
1839 		++rx_queue->scatter_n;
1840 		rx_queue->scatter_len += rx_bytes;
1841 		if (rx_cont)
1842 			return 0;
1843 		n_packets = 1;
1844 	}
1845 
1846 	if (unlikely(EFX_QWORD_FIELD(*event, ESF_DZ_RX_ECRC_ERR)))
1847 		flags |= EFX_RX_PKT_DISCARD;
1848 
1849 	if (unlikely(EFX_QWORD_FIELD(*event, ESF_DZ_RX_IPCKSUM_ERR))) {
1850 		channel->n_rx_ip_hdr_chksum_err += n_packets;
1851 	} else if (unlikely(EFX_QWORD_FIELD(*event,
1852 					    ESF_DZ_RX_TCPUDP_CKSUM_ERR))) {
1853 		channel->n_rx_tcp_udp_chksum_err += n_packets;
1854 	} else if (rx_l4_class == ESE_DZ_L4_CLASS_TCP ||
1855 		   rx_l4_class == ESE_DZ_L4_CLASS_UDP) {
1856 		flags |= EFX_RX_PKT_CSUMMED;
1857 	}
1858 
1859 	if (rx_l4_class == ESE_DZ_L4_CLASS_TCP)
1860 		flags |= EFX_RX_PKT_TCP;
1861 
1862 	channel->irq_mod_score += 2 * n_packets;
1863 
1864 	/* Handle received packet(s) */
1865 	for (i = 0; i < n_packets; i++) {
1866 		efx_rx_packet(rx_queue,
1867 			      rx_queue->removed_count & rx_queue->ptr_mask,
1868 			      rx_queue->scatter_n, rx_queue->scatter_len,
1869 			      flags);
1870 		rx_queue->removed_count += rx_queue->scatter_n;
1871 	}
1872 
1873 	rx_queue->scatter_n = 0;
1874 	rx_queue->scatter_len = 0;
1875 
1876 	return n_packets;
1877 }
1878 
1879 static int
1880 efx_ef10_handle_tx_event(struct efx_channel *channel, efx_qword_t *event)
1881 {
1882 	struct efx_nic *efx = channel->efx;
1883 	struct efx_tx_queue *tx_queue;
1884 	unsigned int tx_ev_desc_ptr;
1885 	unsigned int tx_ev_q_label;
1886 	int tx_descs = 0;
1887 
1888 	if (unlikely(ACCESS_ONCE(efx->reset_pending)))
1889 		return 0;
1890 
1891 	if (unlikely(EFX_QWORD_FIELD(*event, ESF_DZ_TX_DROP_EVENT)))
1892 		return 0;
1893 
1894 	/* Transmit completion */
1895 	tx_ev_desc_ptr = EFX_QWORD_FIELD(*event, ESF_DZ_TX_DESCR_INDX);
1896 	tx_ev_q_label = EFX_QWORD_FIELD(*event, ESF_DZ_TX_QLABEL);
1897 	tx_queue = efx_channel_get_tx_queue(channel,
1898 					    tx_ev_q_label % EFX_TXQ_TYPES);
1899 	tx_descs = ((tx_ev_desc_ptr + 1 - tx_queue->read_count) &
1900 		    tx_queue->ptr_mask);
1901 	efx_xmit_done(tx_queue, tx_ev_desc_ptr & tx_queue->ptr_mask);
1902 
1903 	return tx_descs;
1904 }
1905 
1906 static void
1907 efx_ef10_handle_driver_event(struct efx_channel *channel, efx_qword_t *event)
1908 {
1909 	struct efx_nic *efx = channel->efx;
1910 	int subcode;
1911 
1912 	subcode = EFX_QWORD_FIELD(*event, ESF_DZ_DRV_SUB_CODE);
1913 
1914 	switch (subcode) {
1915 	case ESE_DZ_DRV_TIMER_EV:
1916 	case ESE_DZ_DRV_WAKE_UP_EV:
1917 		break;
1918 	case ESE_DZ_DRV_START_UP_EV:
1919 		/* event queue init complete. ok. */
1920 		break;
1921 	default:
1922 		netif_err(efx, hw, efx->net_dev,
1923 			  "channel %d unknown driver event type %d"
1924 			  " (data " EFX_QWORD_FMT ")\n",
1925 			  channel->channel, subcode,
1926 			  EFX_QWORD_VAL(*event));
1927 
1928 	}
1929 }
1930 
1931 static void efx_ef10_handle_driver_generated_event(struct efx_channel *channel,
1932 						   efx_qword_t *event)
1933 {
1934 	struct efx_nic *efx = channel->efx;
1935 	u32 subcode;
1936 
1937 	subcode = EFX_QWORD_FIELD(*event, EFX_DWORD_0);
1938 
1939 	switch (subcode) {
1940 	case EFX_EF10_TEST:
1941 		channel->event_test_cpu = raw_smp_processor_id();
1942 		break;
1943 	case EFX_EF10_REFILL:
1944 		/* The queue must be empty, so we won't receive any rx
1945 		 * events, so efx_process_channel() won't refill the
1946 		 * queue. Refill it here
1947 		 */
1948 		efx_fast_push_rx_descriptors(&channel->rx_queue, true);
1949 		break;
1950 	default:
1951 		netif_err(efx, hw, efx->net_dev,
1952 			  "channel %d unknown driver event type %u"
1953 			  " (data " EFX_QWORD_FMT ")\n",
1954 			  channel->channel, (unsigned) subcode,
1955 			  EFX_QWORD_VAL(*event));
1956 	}
1957 }
1958 
1959 static int efx_ef10_ev_process(struct efx_channel *channel, int quota)
1960 {
1961 	struct efx_nic *efx = channel->efx;
1962 	efx_qword_t event, *p_event;
1963 	unsigned int read_ptr;
1964 	int ev_code;
1965 	int tx_descs = 0;
1966 	int spent = 0;
1967 
1968 	if (quota <= 0)
1969 		return spent;
1970 
1971 	read_ptr = channel->eventq_read_ptr;
1972 
1973 	for (;;) {
1974 		p_event = efx_event(channel, read_ptr);
1975 		event = *p_event;
1976 
1977 		if (!efx_event_present(&event))
1978 			break;
1979 
1980 		EFX_SET_QWORD(*p_event);
1981 
1982 		++read_ptr;
1983 
1984 		ev_code = EFX_QWORD_FIELD(event, ESF_DZ_EV_CODE);
1985 
1986 		netif_vdbg(efx, drv, efx->net_dev,
1987 			   "processing event on %d " EFX_QWORD_FMT "\n",
1988 			   channel->channel, EFX_QWORD_VAL(event));
1989 
1990 		switch (ev_code) {
1991 		case ESE_DZ_EV_CODE_MCDI_EV:
1992 			efx_mcdi_process_event(channel, &event);
1993 			break;
1994 		case ESE_DZ_EV_CODE_RX_EV:
1995 			spent += efx_ef10_handle_rx_event(channel, &event);
1996 			if (spent >= quota) {
1997 				/* XXX can we split a merged event to
1998 				 * avoid going over-quota?
1999 				 */
2000 				spent = quota;
2001 				goto out;
2002 			}
2003 			break;
2004 		case ESE_DZ_EV_CODE_TX_EV:
2005 			tx_descs += efx_ef10_handle_tx_event(channel, &event);
2006 			if (tx_descs > efx->txq_entries) {
2007 				spent = quota;
2008 				goto out;
2009 			} else if (++spent == quota) {
2010 				goto out;
2011 			}
2012 			break;
2013 		case ESE_DZ_EV_CODE_DRIVER_EV:
2014 			efx_ef10_handle_driver_event(channel, &event);
2015 			if (++spent == quota)
2016 				goto out;
2017 			break;
2018 		case EFX_EF10_DRVGEN_EV:
2019 			efx_ef10_handle_driver_generated_event(channel, &event);
2020 			break;
2021 		default:
2022 			netif_err(efx, hw, efx->net_dev,
2023 				  "channel %d unknown event type %d"
2024 				  " (data " EFX_QWORD_FMT ")\n",
2025 				  channel->channel, ev_code,
2026 				  EFX_QWORD_VAL(event));
2027 		}
2028 	}
2029 
2030 out:
2031 	channel->eventq_read_ptr = read_ptr;
2032 	return spent;
2033 }
2034 
2035 static void efx_ef10_ev_read_ack(struct efx_channel *channel)
2036 {
2037 	struct efx_nic *efx = channel->efx;
2038 	efx_dword_t rptr;
2039 
2040 	if (EFX_EF10_WORKAROUND_35388(efx)) {
2041 		BUILD_BUG_ON(EFX_MIN_EVQ_SIZE <
2042 			     (1 << ERF_DD_EVQ_IND_RPTR_WIDTH));
2043 		BUILD_BUG_ON(EFX_MAX_EVQ_SIZE >
2044 			     (1 << 2 * ERF_DD_EVQ_IND_RPTR_WIDTH));
2045 
2046 		EFX_POPULATE_DWORD_2(rptr, ERF_DD_EVQ_IND_RPTR_FLAGS,
2047 				     EFE_DD_EVQ_IND_RPTR_FLAGS_HIGH,
2048 				     ERF_DD_EVQ_IND_RPTR,
2049 				     (channel->eventq_read_ptr &
2050 				      channel->eventq_mask) >>
2051 				     ERF_DD_EVQ_IND_RPTR_WIDTH);
2052 		efx_writed_page(efx, &rptr, ER_DD_EVQ_INDIRECT,
2053 				channel->channel);
2054 		EFX_POPULATE_DWORD_2(rptr, ERF_DD_EVQ_IND_RPTR_FLAGS,
2055 				     EFE_DD_EVQ_IND_RPTR_FLAGS_LOW,
2056 				     ERF_DD_EVQ_IND_RPTR,
2057 				     channel->eventq_read_ptr &
2058 				     ((1 << ERF_DD_EVQ_IND_RPTR_WIDTH) - 1));
2059 		efx_writed_page(efx, &rptr, ER_DD_EVQ_INDIRECT,
2060 				channel->channel);
2061 	} else {
2062 		EFX_POPULATE_DWORD_1(rptr, ERF_DZ_EVQ_RPTR,
2063 				     channel->eventq_read_ptr &
2064 				     channel->eventq_mask);
2065 		efx_writed_page(efx, &rptr, ER_DZ_EVQ_RPTR, channel->channel);
2066 	}
2067 }
2068 
2069 static void efx_ef10_ev_test_generate(struct efx_channel *channel)
2070 {
2071 	MCDI_DECLARE_BUF(inbuf, MC_CMD_DRIVER_EVENT_IN_LEN);
2072 	struct efx_nic *efx = channel->efx;
2073 	efx_qword_t event;
2074 	int rc;
2075 
2076 	EFX_POPULATE_QWORD_2(event,
2077 			     ESF_DZ_EV_CODE, EFX_EF10_DRVGEN_EV,
2078 			     ESF_DZ_EV_DATA, EFX_EF10_TEST);
2079 
2080 	MCDI_SET_DWORD(inbuf, DRIVER_EVENT_IN_EVQ, channel->channel);
2081 
2082 	/* MCDI_SET_QWORD is not appropriate here since EFX_POPULATE_* has
2083 	 * already swapped the data to little-endian order.
2084 	 */
2085 	memcpy(MCDI_PTR(inbuf, DRIVER_EVENT_IN_DATA), &event.u64[0],
2086 	       sizeof(efx_qword_t));
2087 
2088 	rc = efx_mcdi_rpc(efx, MC_CMD_DRIVER_EVENT, inbuf, sizeof(inbuf),
2089 			  NULL, 0, NULL);
2090 	if (rc != 0)
2091 		goto fail;
2092 
2093 	return;
2094 
2095 fail:
2096 	WARN_ON(true);
2097 	netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
2098 }
2099 
2100 void efx_ef10_handle_drain_event(struct efx_nic *efx)
2101 {
2102 	if (atomic_dec_and_test(&efx->active_queues))
2103 		wake_up(&efx->flush_wq);
2104 
2105 	WARN_ON(atomic_read(&efx->active_queues) < 0);
2106 }
2107 
2108 static int efx_ef10_fini_dmaq(struct efx_nic *efx)
2109 {
2110 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2111 	struct efx_channel *channel;
2112 	struct efx_tx_queue *tx_queue;
2113 	struct efx_rx_queue *rx_queue;
2114 	int pending;
2115 
2116 	/* If the MC has just rebooted, the TX/RX queues will have already been
2117 	 * torn down, but efx->active_queues needs to be set to zero.
2118 	 */
2119 	if (nic_data->must_realloc_vis) {
2120 		atomic_set(&efx->active_queues, 0);
2121 		return 0;
2122 	}
2123 
2124 	/* Do not attempt to write to the NIC during EEH recovery */
2125 	if (efx->state != STATE_RECOVERY) {
2126 		efx_for_each_channel(channel, efx) {
2127 			efx_for_each_channel_rx_queue(rx_queue, channel)
2128 				efx_ef10_rx_fini(rx_queue);
2129 			efx_for_each_channel_tx_queue(tx_queue, channel)
2130 				efx_ef10_tx_fini(tx_queue);
2131 		}
2132 
2133 		wait_event_timeout(efx->flush_wq,
2134 				   atomic_read(&efx->active_queues) == 0,
2135 				   msecs_to_jiffies(EFX_MAX_FLUSH_TIME));
2136 		pending = atomic_read(&efx->active_queues);
2137 		if (pending) {
2138 			netif_err(efx, hw, efx->net_dev, "failed to flush %d queues\n",
2139 				  pending);
2140 			return -ETIMEDOUT;
2141 		}
2142 	}
2143 
2144 	return 0;
2145 }
2146 
2147 static void efx_ef10_prepare_flr(struct efx_nic *efx)
2148 {
2149 	atomic_set(&efx->active_queues, 0);
2150 }
2151 
2152 static bool efx_ef10_filter_equal(const struct efx_filter_spec *left,
2153 				  const struct efx_filter_spec *right)
2154 {
2155 	if ((left->match_flags ^ right->match_flags) |
2156 	    ((left->flags ^ right->flags) &
2157 	     (EFX_FILTER_FLAG_RX | EFX_FILTER_FLAG_TX)))
2158 		return false;
2159 
2160 	return memcmp(&left->outer_vid, &right->outer_vid,
2161 		      sizeof(struct efx_filter_spec) -
2162 		      offsetof(struct efx_filter_spec, outer_vid)) == 0;
2163 }
2164 
2165 static unsigned int efx_ef10_filter_hash(const struct efx_filter_spec *spec)
2166 {
2167 	BUILD_BUG_ON(offsetof(struct efx_filter_spec, outer_vid) & 3);
2168 	return jhash2((const u32 *)&spec->outer_vid,
2169 		      (sizeof(struct efx_filter_spec) -
2170 		       offsetof(struct efx_filter_spec, outer_vid)) / 4,
2171 		      0);
2172 	/* XXX should we randomise the initval? */
2173 }
2174 
2175 /* Decide whether a filter should be exclusive or else should allow
2176  * delivery to additional recipients.  Currently we decide that
2177  * filters for specific local unicast MAC and IP addresses are
2178  * exclusive.
2179  */
2180 static bool efx_ef10_filter_is_exclusive(const struct efx_filter_spec *spec)
2181 {
2182 	if (spec->match_flags & EFX_FILTER_MATCH_LOC_MAC &&
2183 	    !is_multicast_ether_addr(spec->loc_mac))
2184 		return true;
2185 
2186 	if ((spec->match_flags &
2187 	     (EFX_FILTER_MATCH_ETHER_TYPE | EFX_FILTER_MATCH_LOC_HOST)) ==
2188 	    (EFX_FILTER_MATCH_ETHER_TYPE | EFX_FILTER_MATCH_LOC_HOST)) {
2189 		if (spec->ether_type == htons(ETH_P_IP) &&
2190 		    !ipv4_is_multicast(spec->loc_host[0]))
2191 			return true;
2192 		if (spec->ether_type == htons(ETH_P_IPV6) &&
2193 		    ((const u8 *)spec->loc_host)[0] != 0xff)
2194 			return true;
2195 	}
2196 
2197 	return false;
2198 }
2199 
2200 static struct efx_filter_spec *
2201 efx_ef10_filter_entry_spec(const struct efx_ef10_filter_table *table,
2202 			   unsigned int filter_idx)
2203 {
2204 	return (struct efx_filter_spec *)(table->entry[filter_idx].spec &
2205 					  ~EFX_EF10_FILTER_FLAGS);
2206 }
2207 
2208 static unsigned int
2209 efx_ef10_filter_entry_flags(const struct efx_ef10_filter_table *table,
2210 			   unsigned int filter_idx)
2211 {
2212 	return table->entry[filter_idx].spec & EFX_EF10_FILTER_FLAGS;
2213 }
2214 
2215 static void
2216 efx_ef10_filter_set_entry(struct efx_ef10_filter_table *table,
2217 			  unsigned int filter_idx,
2218 			  const struct efx_filter_spec *spec,
2219 			  unsigned int flags)
2220 {
2221 	table->entry[filter_idx].spec =	(unsigned long)spec | flags;
2222 }
2223 
2224 static void efx_ef10_filter_push_prep(struct efx_nic *efx,
2225 				      const struct efx_filter_spec *spec,
2226 				      efx_dword_t *inbuf, u64 handle,
2227 				      bool replacing)
2228 {
2229 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2230 
2231 	memset(inbuf, 0, MC_CMD_FILTER_OP_IN_LEN);
2232 
2233 	if (replacing) {
2234 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
2235 			       MC_CMD_FILTER_OP_IN_OP_REPLACE);
2236 		MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE, handle);
2237 	} else {
2238 		u32 match_fields = 0;
2239 
2240 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
2241 			       efx_ef10_filter_is_exclusive(spec) ?
2242 			       MC_CMD_FILTER_OP_IN_OP_INSERT :
2243 			       MC_CMD_FILTER_OP_IN_OP_SUBSCRIBE);
2244 
2245 		/* Convert match flags and values.  Unlike almost
2246 		 * everything else in MCDI, these fields are in
2247 		 * network byte order.
2248 		 */
2249 		if (spec->match_flags & EFX_FILTER_MATCH_LOC_MAC_IG)
2250 			match_fields |=
2251 				is_multicast_ether_addr(spec->loc_mac) ?
2252 				1 << MC_CMD_FILTER_OP_IN_MATCH_UNKNOWN_MCAST_DST_LBN :
2253 				1 << MC_CMD_FILTER_OP_IN_MATCH_UNKNOWN_UCAST_DST_LBN;
2254 #define COPY_FIELD(gen_flag, gen_field, mcdi_field)			     \
2255 		if (spec->match_flags & EFX_FILTER_MATCH_ ## gen_flag) {     \
2256 			match_fields |=					     \
2257 				1 << MC_CMD_FILTER_OP_IN_MATCH_ ##	     \
2258 				mcdi_field ## _LBN;			     \
2259 			BUILD_BUG_ON(					     \
2260 				MC_CMD_FILTER_OP_IN_ ## mcdi_field ## _LEN < \
2261 				sizeof(spec->gen_field));		     \
2262 			memcpy(MCDI_PTR(inbuf, FILTER_OP_IN_ ##	mcdi_field), \
2263 			       &spec->gen_field, sizeof(spec->gen_field));   \
2264 		}
2265 		COPY_FIELD(REM_HOST, rem_host, SRC_IP);
2266 		COPY_FIELD(LOC_HOST, loc_host, DST_IP);
2267 		COPY_FIELD(REM_MAC, rem_mac, SRC_MAC);
2268 		COPY_FIELD(REM_PORT, rem_port, SRC_PORT);
2269 		COPY_FIELD(LOC_MAC, loc_mac, DST_MAC);
2270 		COPY_FIELD(LOC_PORT, loc_port, DST_PORT);
2271 		COPY_FIELD(ETHER_TYPE, ether_type, ETHER_TYPE);
2272 		COPY_FIELD(INNER_VID, inner_vid, INNER_VLAN);
2273 		COPY_FIELD(OUTER_VID, outer_vid, OUTER_VLAN);
2274 		COPY_FIELD(IP_PROTO, ip_proto, IP_PROTO);
2275 #undef COPY_FIELD
2276 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_MATCH_FIELDS,
2277 			       match_fields);
2278 	}
2279 
2280 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_PORT_ID, EVB_PORT_ID_ASSIGNED);
2281 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_DEST,
2282 		       spec->dmaq_id == EFX_FILTER_RX_DMAQ_ID_DROP ?
2283 		       MC_CMD_FILTER_OP_IN_RX_DEST_DROP :
2284 		       MC_CMD_FILTER_OP_IN_RX_DEST_HOST);
2285 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_TX_DEST,
2286 		       MC_CMD_FILTER_OP_IN_TX_DEST_DEFAULT);
2287 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_QUEUE,
2288 		       spec->dmaq_id == EFX_FILTER_RX_DMAQ_ID_DROP ?
2289 		       0 : spec->dmaq_id);
2290 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_MODE,
2291 		       (spec->flags & EFX_FILTER_FLAG_RX_RSS) ?
2292 		       MC_CMD_FILTER_OP_IN_RX_MODE_RSS :
2293 		       MC_CMD_FILTER_OP_IN_RX_MODE_SIMPLE);
2294 	if (spec->flags & EFX_FILTER_FLAG_RX_RSS)
2295 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_CONTEXT,
2296 			       spec->rss_context !=
2297 			       EFX_FILTER_RSS_CONTEXT_DEFAULT ?
2298 			       spec->rss_context : nic_data->rx_rss_context);
2299 }
2300 
2301 static int efx_ef10_filter_push(struct efx_nic *efx,
2302 				const struct efx_filter_spec *spec,
2303 				u64 *handle, bool replacing)
2304 {
2305 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_IN_LEN);
2306 	MCDI_DECLARE_BUF(outbuf, MC_CMD_FILTER_OP_OUT_LEN);
2307 	int rc;
2308 
2309 	efx_ef10_filter_push_prep(efx, spec, inbuf, *handle, replacing);
2310 	rc = efx_mcdi_rpc(efx, MC_CMD_FILTER_OP, inbuf, sizeof(inbuf),
2311 			  outbuf, sizeof(outbuf), NULL);
2312 	if (rc == 0)
2313 		*handle = MCDI_QWORD(outbuf, FILTER_OP_OUT_HANDLE);
2314 	if (rc == -ENOSPC)
2315 		rc = -EBUSY; /* to match efx_farch_filter_insert() */
2316 	return rc;
2317 }
2318 
2319 static int efx_ef10_filter_rx_match_pri(struct efx_ef10_filter_table *table,
2320 					enum efx_filter_match_flags match_flags)
2321 {
2322 	unsigned int match_pri;
2323 
2324 	for (match_pri = 0;
2325 	     match_pri < table->rx_match_count;
2326 	     match_pri++)
2327 		if (table->rx_match_flags[match_pri] == match_flags)
2328 			return match_pri;
2329 
2330 	return -EPROTONOSUPPORT;
2331 }
2332 
2333 static s32 efx_ef10_filter_insert(struct efx_nic *efx,
2334 				  struct efx_filter_spec *spec,
2335 				  bool replace_equal)
2336 {
2337 	struct efx_ef10_filter_table *table = efx->filter_state;
2338 	DECLARE_BITMAP(mc_rem_map, EFX_EF10_FILTER_SEARCH_LIMIT);
2339 	struct efx_filter_spec *saved_spec;
2340 	unsigned int match_pri, hash;
2341 	unsigned int priv_flags;
2342 	bool replacing = false;
2343 	int ins_index = -1;
2344 	DEFINE_WAIT(wait);
2345 	bool is_mc_recip;
2346 	s32 rc;
2347 
2348 	/* For now, only support RX filters */
2349 	if ((spec->flags & (EFX_FILTER_FLAG_RX | EFX_FILTER_FLAG_TX)) !=
2350 	    EFX_FILTER_FLAG_RX)
2351 		return -EINVAL;
2352 
2353 	rc = efx_ef10_filter_rx_match_pri(table, spec->match_flags);
2354 	if (rc < 0)
2355 		return rc;
2356 	match_pri = rc;
2357 
2358 	hash = efx_ef10_filter_hash(spec);
2359 	is_mc_recip = efx_filter_is_mc_recipient(spec);
2360 	if (is_mc_recip)
2361 		bitmap_zero(mc_rem_map, EFX_EF10_FILTER_SEARCH_LIMIT);
2362 
2363 	/* Find any existing filters with the same match tuple or
2364 	 * else a free slot to insert at.  If any of them are busy,
2365 	 * we have to wait and retry.
2366 	 */
2367 	for (;;) {
2368 		unsigned int depth = 1;
2369 		unsigned int i;
2370 
2371 		spin_lock_bh(&efx->filter_lock);
2372 
2373 		for (;;) {
2374 			i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
2375 			saved_spec = efx_ef10_filter_entry_spec(table, i);
2376 
2377 			if (!saved_spec) {
2378 				if (ins_index < 0)
2379 					ins_index = i;
2380 			} else if (efx_ef10_filter_equal(spec, saved_spec)) {
2381 				if (table->entry[i].spec &
2382 				    EFX_EF10_FILTER_FLAG_BUSY)
2383 					break;
2384 				if (spec->priority < saved_spec->priority &&
2385 				    spec->priority != EFX_FILTER_PRI_AUTO) {
2386 					rc = -EPERM;
2387 					goto out_unlock;
2388 				}
2389 				if (!is_mc_recip) {
2390 					/* This is the only one */
2391 					if (spec->priority ==
2392 					    saved_spec->priority &&
2393 					    !replace_equal) {
2394 						rc = -EEXIST;
2395 						goto out_unlock;
2396 					}
2397 					ins_index = i;
2398 					goto found;
2399 				} else if (spec->priority >
2400 					   saved_spec->priority ||
2401 					   (spec->priority ==
2402 					    saved_spec->priority &&
2403 					    replace_equal)) {
2404 					if (ins_index < 0)
2405 						ins_index = i;
2406 					else
2407 						__set_bit(depth, mc_rem_map);
2408 				}
2409 			}
2410 
2411 			/* Once we reach the maximum search depth, use
2412 			 * the first suitable slot or return -EBUSY if
2413 			 * there was none
2414 			 */
2415 			if (depth == EFX_EF10_FILTER_SEARCH_LIMIT) {
2416 				if (ins_index < 0) {
2417 					rc = -EBUSY;
2418 					goto out_unlock;
2419 				}
2420 				goto found;
2421 			}
2422 
2423 			++depth;
2424 		}
2425 
2426 		prepare_to_wait(&table->waitq, &wait, TASK_UNINTERRUPTIBLE);
2427 		spin_unlock_bh(&efx->filter_lock);
2428 		schedule();
2429 	}
2430 
2431 found:
2432 	/* Create a software table entry if necessary, and mark it
2433 	 * busy.  We might yet fail to insert, but any attempt to
2434 	 * insert a conflicting filter while we're waiting for the
2435 	 * firmware must find the busy entry.
2436 	 */
2437 	saved_spec = efx_ef10_filter_entry_spec(table, ins_index);
2438 	if (saved_spec) {
2439 		if (spec->priority == EFX_FILTER_PRI_AUTO &&
2440 		    saved_spec->priority >= EFX_FILTER_PRI_AUTO) {
2441 			/* Just make sure it won't be removed */
2442 			if (saved_spec->priority > EFX_FILTER_PRI_AUTO)
2443 				saved_spec->flags |= EFX_FILTER_FLAG_RX_OVER_AUTO;
2444 			table->entry[ins_index].spec &=
2445 				~EFX_EF10_FILTER_FLAG_AUTO_OLD;
2446 			rc = ins_index;
2447 			goto out_unlock;
2448 		}
2449 		replacing = true;
2450 		priv_flags = efx_ef10_filter_entry_flags(table, ins_index);
2451 	} else {
2452 		saved_spec = kmalloc(sizeof(*spec), GFP_ATOMIC);
2453 		if (!saved_spec) {
2454 			rc = -ENOMEM;
2455 			goto out_unlock;
2456 		}
2457 		*saved_spec = *spec;
2458 		priv_flags = 0;
2459 	}
2460 	efx_ef10_filter_set_entry(table, ins_index, saved_spec,
2461 				  priv_flags | EFX_EF10_FILTER_FLAG_BUSY);
2462 
2463 	/* Mark lower-priority multicast recipients busy prior to removal */
2464 	if (is_mc_recip) {
2465 		unsigned int depth, i;
2466 
2467 		for (depth = 0; depth < EFX_EF10_FILTER_SEARCH_LIMIT; depth++) {
2468 			i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
2469 			if (test_bit(depth, mc_rem_map))
2470 				table->entry[i].spec |=
2471 					EFX_EF10_FILTER_FLAG_BUSY;
2472 		}
2473 	}
2474 
2475 	spin_unlock_bh(&efx->filter_lock);
2476 
2477 	rc = efx_ef10_filter_push(efx, spec, &table->entry[ins_index].handle,
2478 				  replacing);
2479 
2480 	/* Finalise the software table entry */
2481 	spin_lock_bh(&efx->filter_lock);
2482 	if (rc == 0) {
2483 		if (replacing) {
2484 			/* Update the fields that may differ */
2485 			if (saved_spec->priority == EFX_FILTER_PRI_AUTO)
2486 				saved_spec->flags |=
2487 					EFX_FILTER_FLAG_RX_OVER_AUTO;
2488 			saved_spec->priority = spec->priority;
2489 			saved_spec->flags &= EFX_FILTER_FLAG_RX_OVER_AUTO;
2490 			saved_spec->flags |= spec->flags;
2491 			saved_spec->rss_context = spec->rss_context;
2492 			saved_spec->dmaq_id = spec->dmaq_id;
2493 		}
2494 	} else if (!replacing) {
2495 		kfree(saved_spec);
2496 		saved_spec = NULL;
2497 	}
2498 	efx_ef10_filter_set_entry(table, ins_index, saved_spec, priv_flags);
2499 
2500 	/* Remove and finalise entries for lower-priority multicast
2501 	 * recipients
2502 	 */
2503 	if (is_mc_recip) {
2504 		MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_IN_LEN);
2505 		unsigned int depth, i;
2506 
2507 		memset(inbuf, 0, sizeof(inbuf));
2508 
2509 		for (depth = 0; depth < EFX_EF10_FILTER_SEARCH_LIMIT; depth++) {
2510 			if (!test_bit(depth, mc_rem_map))
2511 				continue;
2512 
2513 			i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
2514 			saved_spec = efx_ef10_filter_entry_spec(table, i);
2515 			priv_flags = efx_ef10_filter_entry_flags(table, i);
2516 
2517 			if (rc == 0) {
2518 				spin_unlock_bh(&efx->filter_lock);
2519 				MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
2520 					       MC_CMD_FILTER_OP_IN_OP_UNSUBSCRIBE);
2521 				MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
2522 					       table->entry[i].handle);
2523 				rc = efx_mcdi_rpc(efx, MC_CMD_FILTER_OP,
2524 						  inbuf, sizeof(inbuf),
2525 						  NULL, 0, NULL);
2526 				spin_lock_bh(&efx->filter_lock);
2527 			}
2528 
2529 			if (rc == 0) {
2530 				kfree(saved_spec);
2531 				saved_spec = NULL;
2532 				priv_flags = 0;
2533 			} else {
2534 				priv_flags &= ~EFX_EF10_FILTER_FLAG_BUSY;
2535 			}
2536 			efx_ef10_filter_set_entry(table, i, saved_spec,
2537 						  priv_flags);
2538 		}
2539 	}
2540 
2541 	/* If successful, return the inserted filter ID */
2542 	if (rc == 0)
2543 		rc = match_pri * HUNT_FILTER_TBL_ROWS + ins_index;
2544 
2545 	wake_up_all(&table->waitq);
2546 out_unlock:
2547 	spin_unlock_bh(&efx->filter_lock);
2548 	finish_wait(&table->waitq, &wait);
2549 	return rc;
2550 }
2551 
2552 static void efx_ef10_filter_update_rx_scatter(struct efx_nic *efx)
2553 {
2554 	/* no need to do anything here on EF10 */
2555 }
2556 
2557 /* Remove a filter.
2558  * If !by_index, remove by ID
2559  * If by_index, remove by index
2560  * Filter ID may come from userland and must be range-checked.
2561  */
2562 static int efx_ef10_filter_remove_internal(struct efx_nic *efx,
2563 					   unsigned int priority_mask,
2564 					   u32 filter_id, bool by_index)
2565 {
2566 	unsigned int filter_idx = filter_id % HUNT_FILTER_TBL_ROWS;
2567 	struct efx_ef10_filter_table *table = efx->filter_state;
2568 	MCDI_DECLARE_BUF(inbuf,
2569 			 MC_CMD_FILTER_OP_IN_HANDLE_OFST +
2570 			 MC_CMD_FILTER_OP_IN_HANDLE_LEN);
2571 	struct efx_filter_spec *spec;
2572 	DEFINE_WAIT(wait);
2573 	int rc;
2574 
2575 	/* Find the software table entry and mark it busy.  Don't
2576 	 * remove it yet; any attempt to update while we're waiting
2577 	 * for the firmware must find the busy entry.
2578 	 */
2579 	for (;;) {
2580 		spin_lock_bh(&efx->filter_lock);
2581 		if (!(table->entry[filter_idx].spec &
2582 		      EFX_EF10_FILTER_FLAG_BUSY))
2583 			break;
2584 		prepare_to_wait(&table->waitq, &wait, TASK_UNINTERRUPTIBLE);
2585 		spin_unlock_bh(&efx->filter_lock);
2586 		schedule();
2587 	}
2588 
2589 	spec = efx_ef10_filter_entry_spec(table, filter_idx);
2590 	if (!spec ||
2591 	    (!by_index &&
2592 	     efx_ef10_filter_rx_match_pri(table, spec->match_flags) !=
2593 	     filter_id / HUNT_FILTER_TBL_ROWS)) {
2594 		rc = -ENOENT;
2595 		goto out_unlock;
2596 	}
2597 
2598 	if (spec->flags & EFX_FILTER_FLAG_RX_OVER_AUTO &&
2599 	    priority_mask == (1U << EFX_FILTER_PRI_AUTO)) {
2600 		/* Just remove flags */
2601 		spec->flags &= ~EFX_FILTER_FLAG_RX_OVER_AUTO;
2602 		table->entry[filter_idx].spec &= ~EFX_EF10_FILTER_FLAG_AUTO_OLD;
2603 		rc = 0;
2604 		goto out_unlock;
2605 	}
2606 
2607 	if (!(priority_mask & (1U << spec->priority))) {
2608 		rc = -ENOENT;
2609 		goto out_unlock;
2610 	}
2611 
2612 	table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_BUSY;
2613 	spin_unlock_bh(&efx->filter_lock);
2614 
2615 	if (spec->flags & EFX_FILTER_FLAG_RX_OVER_AUTO) {
2616 		/* Reset to an automatic filter */
2617 
2618 		struct efx_filter_spec new_spec = *spec;
2619 
2620 		new_spec.priority = EFX_FILTER_PRI_AUTO;
2621 		new_spec.flags = (EFX_FILTER_FLAG_RX |
2622 				  EFX_FILTER_FLAG_RX_RSS);
2623 		new_spec.dmaq_id = 0;
2624 		new_spec.rss_context = EFX_FILTER_RSS_CONTEXT_DEFAULT;
2625 		rc = efx_ef10_filter_push(efx, &new_spec,
2626 					  &table->entry[filter_idx].handle,
2627 					  true);
2628 
2629 		spin_lock_bh(&efx->filter_lock);
2630 		if (rc == 0)
2631 			*spec = new_spec;
2632 	} else {
2633 		/* Really remove the filter */
2634 
2635 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
2636 			       efx_ef10_filter_is_exclusive(spec) ?
2637 			       MC_CMD_FILTER_OP_IN_OP_REMOVE :
2638 			       MC_CMD_FILTER_OP_IN_OP_UNSUBSCRIBE);
2639 		MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
2640 			       table->entry[filter_idx].handle);
2641 		rc = efx_mcdi_rpc(efx, MC_CMD_FILTER_OP,
2642 				  inbuf, sizeof(inbuf), NULL, 0, NULL);
2643 
2644 		spin_lock_bh(&efx->filter_lock);
2645 		if (rc == 0) {
2646 			kfree(spec);
2647 			efx_ef10_filter_set_entry(table, filter_idx, NULL, 0);
2648 		}
2649 	}
2650 
2651 	table->entry[filter_idx].spec &= ~EFX_EF10_FILTER_FLAG_BUSY;
2652 	wake_up_all(&table->waitq);
2653 out_unlock:
2654 	spin_unlock_bh(&efx->filter_lock);
2655 	finish_wait(&table->waitq, &wait);
2656 	return rc;
2657 }
2658 
2659 static int efx_ef10_filter_remove_safe(struct efx_nic *efx,
2660 				       enum efx_filter_priority priority,
2661 				       u32 filter_id)
2662 {
2663 	return efx_ef10_filter_remove_internal(efx, 1U << priority,
2664 					       filter_id, false);
2665 }
2666 
2667 static int efx_ef10_filter_get_safe(struct efx_nic *efx,
2668 				    enum efx_filter_priority priority,
2669 				    u32 filter_id, struct efx_filter_spec *spec)
2670 {
2671 	unsigned int filter_idx = filter_id % HUNT_FILTER_TBL_ROWS;
2672 	struct efx_ef10_filter_table *table = efx->filter_state;
2673 	const struct efx_filter_spec *saved_spec;
2674 	int rc;
2675 
2676 	spin_lock_bh(&efx->filter_lock);
2677 	saved_spec = efx_ef10_filter_entry_spec(table, filter_idx);
2678 	if (saved_spec && saved_spec->priority == priority &&
2679 	    efx_ef10_filter_rx_match_pri(table, saved_spec->match_flags) ==
2680 	    filter_id / HUNT_FILTER_TBL_ROWS) {
2681 		*spec = *saved_spec;
2682 		rc = 0;
2683 	} else {
2684 		rc = -ENOENT;
2685 	}
2686 	spin_unlock_bh(&efx->filter_lock);
2687 	return rc;
2688 }
2689 
2690 static int efx_ef10_filter_clear_rx(struct efx_nic *efx,
2691 				     enum efx_filter_priority priority)
2692 {
2693 	unsigned int priority_mask;
2694 	unsigned int i;
2695 	int rc;
2696 
2697 	priority_mask = (((1U << (priority + 1)) - 1) &
2698 			 ~(1U << EFX_FILTER_PRI_AUTO));
2699 
2700 	for (i = 0; i < HUNT_FILTER_TBL_ROWS; i++) {
2701 		rc = efx_ef10_filter_remove_internal(efx, priority_mask,
2702 						     i, true);
2703 		if (rc && rc != -ENOENT)
2704 			return rc;
2705 	}
2706 
2707 	return 0;
2708 }
2709 
2710 static u32 efx_ef10_filter_count_rx_used(struct efx_nic *efx,
2711 					 enum efx_filter_priority priority)
2712 {
2713 	struct efx_ef10_filter_table *table = efx->filter_state;
2714 	unsigned int filter_idx;
2715 	s32 count = 0;
2716 
2717 	spin_lock_bh(&efx->filter_lock);
2718 	for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
2719 		if (table->entry[filter_idx].spec &&
2720 		    efx_ef10_filter_entry_spec(table, filter_idx)->priority ==
2721 		    priority)
2722 			++count;
2723 	}
2724 	spin_unlock_bh(&efx->filter_lock);
2725 	return count;
2726 }
2727 
2728 static u32 efx_ef10_filter_get_rx_id_limit(struct efx_nic *efx)
2729 {
2730 	struct efx_ef10_filter_table *table = efx->filter_state;
2731 
2732 	return table->rx_match_count * HUNT_FILTER_TBL_ROWS;
2733 }
2734 
2735 static s32 efx_ef10_filter_get_rx_ids(struct efx_nic *efx,
2736 				      enum efx_filter_priority priority,
2737 				      u32 *buf, u32 size)
2738 {
2739 	struct efx_ef10_filter_table *table = efx->filter_state;
2740 	struct efx_filter_spec *spec;
2741 	unsigned int filter_idx;
2742 	s32 count = 0;
2743 
2744 	spin_lock_bh(&efx->filter_lock);
2745 	for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
2746 		spec = efx_ef10_filter_entry_spec(table, filter_idx);
2747 		if (spec && spec->priority == priority) {
2748 			if (count == size) {
2749 				count = -EMSGSIZE;
2750 				break;
2751 			}
2752 			buf[count++] = (efx_ef10_filter_rx_match_pri(
2753 						table, spec->match_flags) *
2754 					HUNT_FILTER_TBL_ROWS +
2755 					filter_idx);
2756 		}
2757 	}
2758 	spin_unlock_bh(&efx->filter_lock);
2759 	return count;
2760 }
2761 
2762 #ifdef CONFIG_RFS_ACCEL
2763 
2764 static efx_mcdi_async_completer efx_ef10_filter_rfs_insert_complete;
2765 
2766 static s32 efx_ef10_filter_rfs_insert(struct efx_nic *efx,
2767 				      struct efx_filter_spec *spec)
2768 {
2769 	struct efx_ef10_filter_table *table = efx->filter_state;
2770 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_IN_LEN);
2771 	struct efx_filter_spec *saved_spec;
2772 	unsigned int hash, i, depth = 1;
2773 	bool replacing = false;
2774 	int ins_index = -1;
2775 	u64 cookie;
2776 	s32 rc;
2777 
2778 	/* Must be an RX filter without RSS and not for a multicast
2779 	 * destination address (RFS only works for connected sockets).
2780 	 * These restrictions allow us to pass only a tiny amount of
2781 	 * data through to the completion function.
2782 	 */
2783 	EFX_WARN_ON_PARANOID(spec->flags !=
2784 			     (EFX_FILTER_FLAG_RX | EFX_FILTER_FLAG_RX_SCATTER));
2785 	EFX_WARN_ON_PARANOID(spec->priority != EFX_FILTER_PRI_HINT);
2786 	EFX_WARN_ON_PARANOID(efx_filter_is_mc_recipient(spec));
2787 
2788 	hash = efx_ef10_filter_hash(spec);
2789 
2790 	spin_lock_bh(&efx->filter_lock);
2791 
2792 	/* Find any existing filter with the same match tuple or else
2793 	 * a free slot to insert at.  If an existing filter is busy,
2794 	 * we have to give up.
2795 	 */
2796 	for (;;) {
2797 		i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
2798 		saved_spec = efx_ef10_filter_entry_spec(table, i);
2799 
2800 		if (!saved_spec) {
2801 			if (ins_index < 0)
2802 				ins_index = i;
2803 		} else if (efx_ef10_filter_equal(spec, saved_spec)) {
2804 			if (table->entry[i].spec & EFX_EF10_FILTER_FLAG_BUSY) {
2805 				rc = -EBUSY;
2806 				goto fail_unlock;
2807 			}
2808 			if (spec->priority < saved_spec->priority) {
2809 				rc = -EPERM;
2810 				goto fail_unlock;
2811 			}
2812 			ins_index = i;
2813 			break;
2814 		}
2815 
2816 		/* Once we reach the maximum search depth, use the
2817 		 * first suitable slot or return -EBUSY if there was
2818 		 * none
2819 		 */
2820 		if (depth == EFX_EF10_FILTER_SEARCH_LIMIT) {
2821 			if (ins_index < 0) {
2822 				rc = -EBUSY;
2823 				goto fail_unlock;
2824 			}
2825 			break;
2826 		}
2827 
2828 		++depth;
2829 	}
2830 
2831 	/* Create a software table entry if necessary, and mark it
2832 	 * busy.  We might yet fail to insert, but any attempt to
2833 	 * insert a conflicting filter while we're waiting for the
2834 	 * firmware must find the busy entry.
2835 	 */
2836 	saved_spec = efx_ef10_filter_entry_spec(table, ins_index);
2837 	if (saved_spec) {
2838 		replacing = true;
2839 	} else {
2840 		saved_spec = kmalloc(sizeof(*spec), GFP_ATOMIC);
2841 		if (!saved_spec) {
2842 			rc = -ENOMEM;
2843 			goto fail_unlock;
2844 		}
2845 		*saved_spec = *spec;
2846 	}
2847 	efx_ef10_filter_set_entry(table, ins_index, saved_spec,
2848 				  EFX_EF10_FILTER_FLAG_BUSY);
2849 
2850 	spin_unlock_bh(&efx->filter_lock);
2851 
2852 	/* Pack up the variables needed on completion */
2853 	cookie = replacing << 31 | ins_index << 16 | spec->dmaq_id;
2854 
2855 	efx_ef10_filter_push_prep(efx, spec, inbuf,
2856 				  table->entry[ins_index].handle, replacing);
2857 	efx_mcdi_rpc_async(efx, MC_CMD_FILTER_OP, inbuf, sizeof(inbuf),
2858 			   MC_CMD_FILTER_OP_OUT_LEN,
2859 			   efx_ef10_filter_rfs_insert_complete, cookie);
2860 
2861 	return ins_index;
2862 
2863 fail_unlock:
2864 	spin_unlock_bh(&efx->filter_lock);
2865 	return rc;
2866 }
2867 
2868 static void
2869 efx_ef10_filter_rfs_insert_complete(struct efx_nic *efx, unsigned long cookie,
2870 				    int rc, efx_dword_t *outbuf,
2871 				    size_t outlen_actual)
2872 {
2873 	struct efx_ef10_filter_table *table = efx->filter_state;
2874 	unsigned int ins_index, dmaq_id;
2875 	struct efx_filter_spec *spec;
2876 	bool replacing;
2877 
2878 	/* Unpack the cookie */
2879 	replacing = cookie >> 31;
2880 	ins_index = (cookie >> 16) & (HUNT_FILTER_TBL_ROWS - 1);
2881 	dmaq_id = cookie & 0xffff;
2882 
2883 	spin_lock_bh(&efx->filter_lock);
2884 	spec = efx_ef10_filter_entry_spec(table, ins_index);
2885 	if (rc == 0) {
2886 		table->entry[ins_index].handle =
2887 			MCDI_QWORD(outbuf, FILTER_OP_OUT_HANDLE);
2888 		if (replacing)
2889 			spec->dmaq_id = dmaq_id;
2890 	} else if (!replacing) {
2891 		kfree(spec);
2892 		spec = NULL;
2893 	}
2894 	efx_ef10_filter_set_entry(table, ins_index, spec, 0);
2895 	spin_unlock_bh(&efx->filter_lock);
2896 
2897 	wake_up_all(&table->waitq);
2898 }
2899 
2900 static void
2901 efx_ef10_filter_rfs_expire_complete(struct efx_nic *efx,
2902 				    unsigned long filter_idx,
2903 				    int rc, efx_dword_t *outbuf,
2904 				    size_t outlen_actual);
2905 
2906 static bool efx_ef10_filter_rfs_expire_one(struct efx_nic *efx, u32 flow_id,
2907 					   unsigned int filter_idx)
2908 {
2909 	struct efx_ef10_filter_table *table = efx->filter_state;
2910 	struct efx_filter_spec *spec =
2911 		efx_ef10_filter_entry_spec(table, filter_idx);
2912 	MCDI_DECLARE_BUF(inbuf,
2913 			 MC_CMD_FILTER_OP_IN_HANDLE_OFST +
2914 			 MC_CMD_FILTER_OP_IN_HANDLE_LEN);
2915 
2916 	if (!spec ||
2917 	    (table->entry[filter_idx].spec & EFX_EF10_FILTER_FLAG_BUSY) ||
2918 	    spec->priority != EFX_FILTER_PRI_HINT ||
2919 	    !rps_may_expire_flow(efx->net_dev, spec->dmaq_id,
2920 				 flow_id, filter_idx))
2921 		return false;
2922 
2923 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
2924 		       MC_CMD_FILTER_OP_IN_OP_REMOVE);
2925 	MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
2926 		       table->entry[filter_idx].handle);
2927 	if (efx_mcdi_rpc_async(efx, MC_CMD_FILTER_OP, inbuf, sizeof(inbuf), 0,
2928 			       efx_ef10_filter_rfs_expire_complete, filter_idx))
2929 		return false;
2930 
2931 	table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_BUSY;
2932 	return true;
2933 }
2934 
2935 static void
2936 efx_ef10_filter_rfs_expire_complete(struct efx_nic *efx,
2937 				    unsigned long filter_idx,
2938 				    int rc, efx_dword_t *outbuf,
2939 				    size_t outlen_actual)
2940 {
2941 	struct efx_ef10_filter_table *table = efx->filter_state;
2942 	struct efx_filter_spec *spec =
2943 		efx_ef10_filter_entry_spec(table, filter_idx);
2944 
2945 	spin_lock_bh(&efx->filter_lock);
2946 	if (rc == 0) {
2947 		kfree(spec);
2948 		efx_ef10_filter_set_entry(table, filter_idx, NULL, 0);
2949 	}
2950 	table->entry[filter_idx].spec &= ~EFX_EF10_FILTER_FLAG_BUSY;
2951 	wake_up_all(&table->waitq);
2952 	spin_unlock_bh(&efx->filter_lock);
2953 }
2954 
2955 #endif /* CONFIG_RFS_ACCEL */
2956 
2957 static int efx_ef10_filter_match_flags_from_mcdi(u32 mcdi_flags)
2958 {
2959 	int match_flags = 0;
2960 
2961 #define MAP_FLAG(gen_flag, mcdi_field) {				\
2962 		u32 old_mcdi_flags = mcdi_flags;			\
2963 		mcdi_flags &= ~(1 << MC_CMD_FILTER_OP_IN_MATCH_ ##	\
2964 				mcdi_field ## _LBN);			\
2965 		if (mcdi_flags != old_mcdi_flags)			\
2966 			match_flags |= EFX_FILTER_MATCH_ ## gen_flag;	\
2967 	}
2968 	MAP_FLAG(LOC_MAC_IG, UNKNOWN_UCAST_DST);
2969 	MAP_FLAG(LOC_MAC_IG, UNKNOWN_MCAST_DST);
2970 	MAP_FLAG(REM_HOST, SRC_IP);
2971 	MAP_FLAG(LOC_HOST, DST_IP);
2972 	MAP_FLAG(REM_MAC, SRC_MAC);
2973 	MAP_FLAG(REM_PORT, SRC_PORT);
2974 	MAP_FLAG(LOC_MAC, DST_MAC);
2975 	MAP_FLAG(LOC_PORT, DST_PORT);
2976 	MAP_FLAG(ETHER_TYPE, ETHER_TYPE);
2977 	MAP_FLAG(INNER_VID, INNER_VLAN);
2978 	MAP_FLAG(OUTER_VID, OUTER_VLAN);
2979 	MAP_FLAG(IP_PROTO, IP_PROTO);
2980 #undef MAP_FLAG
2981 
2982 	/* Did we map them all? */
2983 	if (mcdi_flags)
2984 		return -EINVAL;
2985 
2986 	return match_flags;
2987 }
2988 
2989 static int efx_ef10_filter_table_probe(struct efx_nic *efx)
2990 {
2991 	MCDI_DECLARE_BUF(inbuf, MC_CMD_GET_PARSER_DISP_INFO_IN_LEN);
2992 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_PARSER_DISP_INFO_OUT_LENMAX);
2993 	unsigned int pd_match_pri, pd_match_count;
2994 	struct efx_ef10_filter_table *table;
2995 	size_t outlen;
2996 	int rc;
2997 
2998 	table = kzalloc(sizeof(*table), GFP_KERNEL);
2999 	if (!table)
3000 		return -ENOMEM;
3001 
3002 	/* Find out which RX filter types are supported, and their priorities */
3003 	MCDI_SET_DWORD(inbuf, GET_PARSER_DISP_INFO_IN_OP,
3004 		       MC_CMD_GET_PARSER_DISP_INFO_IN_OP_GET_SUPPORTED_RX_MATCHES);
3005 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_PARSER_DISP_INFO,
3006 			  inbuf, sizeof(inbuf), outbuf, sizeof(outbuf),
3007 			  &outlen);
3008 	if (rc)
3009 		goto fail;
3010 	pd_match_count = MCDI_VAR_ARRAY_LEN(
3011 		outlen, GET_PARSER_DISP_INFO_OUT_SUPPORTED_MATCHES);
3012 	table->rx_match_count = 0;
3013 
3014 	for (pd_match_pri = 0; pd_match_pri < pd_match_count; pd_match_pri++) {
3015 		u32 mcdi_flags =
3016 			MCDI_ARRAY_DWORD(
3017 				outbuf,
3018 				GET_PARSER_DISP_INFO_OUT_SUPPORTED_MATCHES,
3019 				pd_match_pri);
3020 		rc = efx_ef10_filter_match_flags_from_mcdi(mcdi_flags);
3021 		if (rc < 0) {
3022 			netif_dbg(efx, probe, efx->net_dev,
3023 				  "%s: fw flags %#x pri %u not supported in driver\n",
3024 				  __func__, mcdi_flags, pd_match_pri);
3025 		} else {
3026 			netif_dbg(efx, probe, efx->net_dev,
3027 				  "%s: fw flags %#x pri %u supported as driver flags %#x pri %u\n",
3028 				  __func__, mcdi_flags, pd_match_pri,
3029 				  rc, table->rx_match_count);
3030 			table->rx_match_flags[table->rx_match_count++] = rc;
3031 		}
3032 	}
3033 
3034 	table->entry = vzalloc(HUNT_FILTER_TBL_ROWS * sizeof(*table->entry));
3035 	if (!table->entry) {
3036 		rc = -ENOMEM;
3037 		goto fail;
3038 	}
3039 
3040 	efx->filter_state = table;
3041 	init_waitqueue_head(&table->waitq);
3042 	return 0;
3043 
3044 fail:
3045 	kfree(table);
3046 	return rc;
3047 }
3048 
3049 static void efx_ef10_filter_table_restore(struct efx_nic *efx)
3050 {
3051 	struct efx_ef10_filter_table *table = efx->filter_state;
3052 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
3053 	struct efx_filter_spec *spec;
3054 	unsigned int filter_idx;
3055 	bool failed = false;
3056 	int rc;
3057 
3058 	if (!nic_data->must_restore_filters)
3059 		return;
3060 
3061 	spin_lock_bh(&efx->filter_lock);
3062 
3063 	for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
3064 		spec = efx_ef10_filter_entry_spec(table, filter_idx);
3065 		if (!spec)
3066 			continue;
3067 
3068 		table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_BUSY;
3069 		spin_unlock_bh(&efx->filter_lock);
3070 
3071 		rc = efx_ef10_filter_push(efx, spec,
3072 					  &table->entry[filter_idx].handle,
3073 					  false);
3074 		if (rc)
3075 			failed = true;
3076 
3077 		spin_lock_bh(&efx->filter_lock);
3078 		if (rc) {
3079 			kfree(spec);
3080 			efx_ef10_filter_set_entry(table, filter_idx, NULL, 0);
3081 		} else {
3082 			table->entry[filter_idx].spec &=
3083 				~EFX_EF10_FILTER_FLAG_BUSY;
3084 		}
3085 	}
3086 
3087 	spin_unlock_bh(&efx->filter_lock);
3088 
3089 	if (failed)
3090 		netif_err(efx, hw, efx->net_dev,
3091 			  "unable to restore all filters\n");
3092 	else
3093 		nic_data->must_restore_filters = false;
3094 }
3095 
3096 static void efx_ef10_filter_table_remove(struct efx_nic *efx)
3097 {
3098 	struct efx_ef10_filter_table *table = efx->filter_state;
3099 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_IN_LEN);
3100 	struct efx_filter_spec *spec;
3101 	unsigned int filter_idx;
3102 	int rc;
3103 
3104 	for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
3105 		spec = efx_ef10_filter_entry_spec(table, filter_idx);
3106 		if (!spec)
3107 			continue;
3108 
3109 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
3110 			       efx_ef10_filter_is_exclusive(spec) ?
3111 			       MC_CMD_FILTER_OP_IN_OP_REMOVE :
3112 			       MC_CMD_FILTER_OP_IN_OP_UNSUBSCRIBE);
3113 		MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
3114 			       table->entry[filter_idx].handle);
3115 		rc = efx_mcdi_rpc(efx, MC_CMD_FILTER_OP, inbuf, sizeof(inbuf),
3116 				  NULL, 0, NULL);
3117 		if (rc)
3118 			netdev_WARN(efx->net_dev,
3119 				    "filter_idx=%#x handle=%#llx\n",
3120 				    filter_idx,
3121 				    table->entry[filter_idx].handle);
3122 		kfree(spec);
3123 	}
3124 
3125 	vfree(table->entry);
3126 	kfree(table);
3127 }
3128 
3129 static void efx_ef10_filter_sync_rx_mode(struct efx_nic *efx)
3130 {
3131 	struct efx_ef10_filter_table *table = efx->filter_state;
3132 	struct net_device *net_dev = efx->net_dev;
3133 	struct efx_filter_spec spec;
3134 	bool remove_failed = false;
3135 	struct netdev_hw_addr *uc;
3136 	struct netdev_hw_addr *mc;
3137 	unsigned int filter_idx;
3138 	int i, n, rc;
3139 
3140 	if (!efx_dev_registered(efx))
3141 		return;
3142 
3143 	/* Mark old filters that may need to be removed */
3144 	spin_lock_bh(&efx->filter_lock);
3145 	n = table->dev_uc_count < 0 ? 1 : table->dev_uc_count;
3146 	for (i = 0; i < n; i++) {
3147 		filter_idx = table->dev_uc_list[i].id % HUNT_FILTER_TBL_ROWS;
3148 		table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_AUTO_OLD;
3149 	}
3150 	n = table->dev_mc_count < 0 ? 1 : table->dev_mc_count;
3151 	for (i = 0; i < n; i++) {
3152 		filter_idx = table->dev_mc_list[i].id % HUNT_FILTER_TBL_ROWS;
3153 		table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_AUTO_OLD;
3154 	}
3155 	spin_unlock_bh(&efx->filter_lock);
3156 
3157 	/* Copy/convert the address lists; add the primary station
3158 	 * address and broadcast address
3159 	 */
3160 	netif_addr_lock_bh(net_dev);
3161 	if (net_dev->flags & IFF_PROMISC ||
3162 	    netdev_uc_count(net_dev) >= EFX_EF10_FILTER_DEV_UC_MAX) {
3163 		table->dev_uc_count = -1;
3164 	} else {
3165 		table->dev_uc_count = 1 + netdev_uc_count(net_dev);
3166 		ether_addr_copy(table->dev_uc_list[0].addr, net_dev->dev_addr);
3167 		i = 1;
3168 		netdev_for_each_uc_addr(uc, net_dev) {
3169 			ether_addr_copy(table->dev_uc_list[i].addr, uc->addr);
3170 			i++;
3171 		}
3172 	}
3173 	if (net_dev->flags & (IFF_PROMISC | IFF_ALLMULTI) ||
3174 	    netdev_mc_count(net_dev) >= EFX_EF10_FILTER_DEV_MC_MAX) {
3175 		table->dev_mc_count = -1;
3176 	} else {
3177 		table->dev_mc_count = 1 + netdev_mc_count(net_dev);
3178 		eth_broadcast_addr(table->dev_mc_list[0].addr);
3179 		i = 1;
3180 		netdev_for_each_mc_addr(mc, net_dev) {
3181 			ether_addr_copy(table->dev_mc_list[i].addr, mc->addr);
3182 			i++;
3183 		}
3184 	}
3185 	netif_addr_unlock_bh(net_dev);
3186 
3187 	/* Insert/renew unicast filters */
3188 	if (table->dev_uc_count >= 0) {
3189 		for (i = 0; i < table->dev_uc_count; i++) {
3190 			efx_filter_init_rx(&spec, EFX_FILTER_PRI_AUTO,
3191 					   EFX_FILTER_FLAG_RX_RSS,
3192 					   0);
3193 			efx_filter_set_eth_local(&spec, EFX_FILTER_VID_UNSPEC,
3194 						 table->dev_uc_list[i].addr);
3195 			rc = efx_ef10_filter_insert(efx, &spec, true);
3196 			if (rc < 0) {
3197 				/* Fall back to unicast-promisc */
3198 				while (i--)
3199 					efx_ef10_filter_remove_safe(
3200 						efx, EFX_FILTER_PRI_AUTO,
3201 						table->dev_uc_list[i].id);
3202 				table->dev_uc_count = -1;
3203 				break;
3204 			}
3205 			table->dev_uc_list[i].id = rc;
3206 		}
3207 	}
3208 	if (table->dev_uc_count < 0) {
3209 		efx_filter_init_rx(&spec, EFX_FILTER_PRI_AUTO,
3210 				   EFX_FILTER_FLAG_RX_RSS,
3211 				   0);
3212 		efx_filter_set_uc_def(&spec);
3213 		rc = efx_ef10_filter_insert(efx, &spec, true);
3214 		if (rc < 0) {
3215 			WARN_ON(1);
3216 			table->dev_uc_count = 0;
3217 		} else {
3218 			table->dev_uc_list[0].id = rc;
3219 		}
3220 	}
3221 
3222 	/* Insert/renew multicast filters */
3223 	if (table->dev_mc_count >= 0) {
3224 		for (i = 0; i < table->dev_mc_count; i++) {
3225 			efx_filter_init_rx(&spec, EFX_FILTER_PRI_AUTO,
3226 					   EFX_FILTER_FLAG_RX_RSS,
3227 					   0);
3228 			efx_filter_set_eth_local(&spec, EFX_FILTER_VID_UNSPEC,
3229 						 table->dev_mc_list[i].addr);
3230 			rc = efx_ef10_filter_insert(efx, &spec, true);
3231 			if (rc < 0) {
3232 				/* Fall back to multicast-promisc */
3233 				while (i--)
3234 					efx_ef10_filter_remove_safe(
3235 						efx, EFX_FILTER_PRI_AUTO,
3236 						table->dev_mc_list[i].id);
3237 				table->dev_mc_count = -1;
3238 				break;
3239 			}
3240 			table->dev_mc_list[i].id = rc;
3241 		}
3242 	}
3243 	if (table->dev_mc_count < 0) {
3244 		efx_filter_init_rx(&spec, EFX_FILTER_PRI_AUTO,
3245 				   EFX_FILTER_FLAG_RX_RSS,
3246 				   0);
3247 		efx_filter_set_mc_def(&spec);
3248 		rc = efx_ef10_filter_insert(efx, &spec, true);
3249 		if (rc < 0) {
3250 			WARN_ON(1);
3251 			table->dev_mc_count = 0;
3252 		} else {
3253 			table->dev_mc_list[0].id = rc;
3254 		}
3255 	}
3256 
3257 	/* Remove filters that weren't renewed.  Since nothing else
3258 	 * changes the AUTO_OLD flag or removes these filters, we
3259 	 * don't need to hold the filter_lock while scanning for
3260 	 * these filters.
3261 	 */
3262 	for (i = 0; i < HUNT_FILTER_TBL_ROWS; i++) {
3263 		if (ACCESS_ONCE(table->entry[i].spec) &
3264 		    EFX_EF10_FILTER_FLAG_AUTO_OLD) {
3265 			if (efx_ef10_filter_remove_internal(
3266 				    efx, 1U << EFX_FILTER_PRI_AUTO,
3267 				    i, true) < 0)
3268 				remove_failed = true;
3269 		}
3270 	}
3271 	WARN_ON(remove_failed);
3272 }
3273 
3274 static int efx_ef10_mac_reconfigure(struct efx_nic *efx)
3275 {
3276 	efx_ef10_filter_sync_rx_mode(efx);
3277 
3278 	return efx_mcdi_set_mac(efx);
3279 }
3280 
3281 static int efx_ef10_start_bist(struct efx_nic *efx, u32 bist_type)
3282 {
3283 	MCDI_DECLARE_BUF(inbuf, MC_CMD_START_BIST_IN_LEN);
3284 
3285 	MCDI_SET_DWORD(inbuf, START_BIST_IN_TYPE, bist_type);
3286 	return efx_mcdi_rpc(efx, MC_CMD_START_BIST, inbuf, sizeof(inbuf),
3287 			    NULL, 0, NULL);
3288 }
3289 
3290 /* MC BISTs follow a different poll mechanism to phy BISTs.
3291  * The BIST is done in the poll handler on the MC, and the MCDI command
3292  * will block until the BIST is done.
3293  */
3294 static int efx_ef10_poll_bist(struct efx_nic *efx)
3295 {
3296 	int rc;
3297 	MCDI_DECLARE_BUF(outbuf, MC_CMD_POLL_BIST_OUT_LEN);
3298 	size_t outlen;
3299 	u32 result;
3300 
3301 	rc = efx_mcdi_rpc(efx, MC_CMD_POLL_BIST, NULL, 0,
3302 			   outbuf, sizeof(outbuf), &outlen);
3303 	if (rc != 0)
3304 		return rc;
3305 
3306 	if (outlen < MC_CMD_POLL_BIST_OUT_LEN)
3307 		return -EIO;
3308 
3309 	result = MCDI_DWORD(outbuf, POLL_BIST_OUT_RESULT);
3310 	switch (result) {
3311 	case MC_CMD_POLL_BIST_PASSED:
3312 		netif_dbg(efx, hw, efx->net_dev, "BIST passed.\n");
3313 		return 0;
3314 	case MC_CMD_POLL_BIST_TIMEOUT:
3315 		netif_err(efx, hw, efx->net_dev, "BIST timed out\n");
3316 		return -EIO;
3317 	case MC_CMD_POLL_BIST_FAILED:
3318 		netif_err(efx, hw, efx->net_dev, "BIST failed.\n");
3319 		return -EIO;
3320 	default:
3321 		netif_err(efx, hw, efx->net_dev,
3322 			  "BIST returned unknown result %u", result);
3323 		return -EIO;
3324 	}
3325 }
3326 
3327 static int efx_ef10_run_bist(struct efx_nic *efx, u32 bist_type)
3328 {
3329 	int rc;
3330 
3331 	netif_dbg(efx, drv, efx->net_dev, "starting BIST type %u\n", bist_type);
3332 
3333 	rc = efx_ef10_start_bist(efx, bist_type);
3334 	if (rc != 0)
3335 		return rc;
3336 
3337 	return efx_ef10_poll_bist(efx);
3338 }
3339 
3340 static int
3341 efx_ef10_test_chip(struct efx_nic *efx, struct efx_self_tests *tests)
3342 {
3343 	int rc, rc2;
3344 
3345 	efx_reset_down(efx, RESET_TYPE_WORLD);
3346 
3347 	rc = efx_mcdi_rpc(efx, MC_CMD_ENABLE_OFFLINE_BIST,
3348 			  NULL, 0, NULL, 0, NULL);
3349 	if (rc != 0)
3350 		goto out;
3351 
3352 	tests->memory = efx_ef10_run_bist(efx, MC_CMD_MC_MEM_BIST) ? -1 : 1;
3353 	tests->registers = efx_ef10_run_bist(efx, MC_CMD_REG_BIST) ? -1 : 1;
3354 
3355 	rc = efx_mcdi_reset(efx, RESET_TYPE_WORLD);
3356 
3357 out:
3358 	rc2 = efx_reset_up(efx, RESET_TYPE_WORLD, rc == 0);
3359 	return rc ? rc : rc2;
3360 }
3361 
3362 #ifdef CONFIG_SFC_MTD
3363 
3364 struct efx_ef10_nvram_type_info {
3365 	u16 type, type_mask;
3366 	u8 port;
3367 	const char *name;
3368 };
3369 
3370 static const struct efx_ef10_nvram_type_info efx_ef10_nvram_types[] = {
3371 	{ NVRAM_PARTITION_TYPE_MC_FIRMWARE,	   0,    0, "sfc_mcfw" },
3372 	{ NVRAM_PARTITION_TYPE_MC_FIRMWARE_BACKUP, 0,    0, "sfc_mcfw_backup" },
3373 	{ NVRAM_PARTITION_TYPE_EXPANSION_ROM,	   0,    0, "sfc_exp_rom" },
3374 	{ NVRAM_PARTITION_TYPE_STATIC_CONFIG,	   0,    0, "sfc_static_cfg" },
3375 	{ NVRAM_PARTITION_TYPE_DYNAMIC_CONFIG,	   0,    0, "sfc_dynamic_cfg" },
3376 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT0, 0,   0, "sfc_exp_rom_cfg" },
3377 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT1, 0,   1, "sfc_exp_rom_cfg" },
3378 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT2, 0,   2, "sfc_exp_rom_cfg" },
3379 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT3, 0,   3, "sfc_exp_rom_cfg" },
3380 	{ NVRAM_PARTITION_TYPE_LICENSE,		   0,    0, "sfc_license" },
3381 	{ NVRAM_PARTITION_TYPE_PHY_MIN,		   0xff, 0, "sfc_phy_fw" },
3382 };
3383 
3384 static int efx_ef10_mtd_probe_partition(struct efx_nic *efx,
3385 					struct efx_mcdi_mtd_partition *part,
3386 					unsigned int type)
3387 {
3388 	MCDI_DECLARE_BUF(inbuf, MC_CMD_NVRAM_METADATA_IN_LEN);
3389 	MCDI_DECLARE_BUF(outbuf, MC_CMD_NVRAM_METADATA_OUT_LENMAX);
3390 	const struct efx_ef10_nvram_type_info *info;
3391 	size_t size, erase_size, outlen;
3392 	bool protected;
3393 	int rc;
3394 
3395 	for (info = efx_ef10_nvram_types; ; info++) {
3396 		if (info ==
3397 		    efx_ef10_nvram_types + ARRAY_SIZE(efx_ef10_nvram_types))
3398 			return -ENODEV;
3399 		if ((type & ~info->type_mask) == info->type)
3400 			break;
3401 	}
3402 	if (info->port != efx_port_num(efx))
3403 		return -ENODEV;
3404 
3405 	rc = efx_mcdi_nvram_info(efx, type, &size, &erase_size, &protected);
3406 	if (rc)
3407 		return rc;
3408 	if (protected)
3409 		return -ENODEV; /* hide it */
3410 
3411 	part->nvram_type = type;
3412 
3413 	MCDI_SET_DWORD(inbuf, NVRAM_METADATA_IN_TYPE, type);
3414 	rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_METADATA, inbuf, sizeof(inbuf),
3415 			  outbuf, sizeof(outbuf), &outlen);
3416 	if (rc)
3417 		return rc;
3418 	if (outlen < MC_CMD_NVRAM_METADATA_OUT_LENMIN)
3419 		return -EIO;
3420 	if (MCDI_DWORD(outbuf, NVRAM_METADATA_OUT_FLAGS) &
3421 	    (1 << MC_CMD_NVRAM_METADATA_OUT_SUBTYPE_VALID_LBN))
3422 		part->fw_subtype = MCDI_DWORD(outbuf,
3423 					      NVRAM_METADATA_OUT_SUBTYPE);
3424 
3425 	part->common.dev_type_name = "EF10 NVRAM manager";
3426 	part->common.type_name = info->name;
3427 
3428 	part->common.mtd.type = MTD_NORFLASH;
3429 	part->common.mtd.flags = MTD_CAP_NORFLASH;
3430 	part->common.mtd.size = size;
3431 	part->common.mtd.erasesize = erase_size;
3432 
3433 	return 0;
3434 }
3435 
3436 static int efx_ef10_mtd_probe(struct efx_nic *efx)
3437 {
3438 	MCDI_DECLARE_BUF(outbuf, MC_CMD_NVRAM_PARTITIONS_OUT_LENMAX);
3439 	struct efx_mcdi_mtd_partition *parts;
3440 	size_t outlen, n_parts_total, i, n_parts;
3441 	unsigned int type;
3442 	int rc;
3443 
3444 	ASSERT_RTNL();
3445 
3446 	BUILD_BUG_ON(MC_CMD_NVRAM_PARTITIONS_IN_LEN != 0);
3447 	rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_PARTITIONS, NULL, 0,
3448 			  outbuf, sizeof(outbuf), &outlen);
3449 	if (rc)
3450 		return rc;
3451 	if (outlen < MC_CMD_NVRAM_PARTITIONS_OUT_LENMIN)
3452 		return -EIO;
3453 
3454 	n_parts_total = MCDI_DWORD(outbuf, NVRAM_PARTITIONS_OUT_NUM_PARTITIONS);
3455 	if (n_parts_total >
3456 	    MCDI_VAR_ARRAY_LEN(outlen, NVRAM_PARTITIONS_OUT_TYPE_ID))
3457 		return -EIO;
3458 
3459 	parts = kcalloc(n_parts_total, sizeof(*parts), GFP_KERNEL);
3460 	if (!parts)
3461 		return -ENOMEM;
3462 
3463 	n_parts = 0;
3464 	for (i = 0; i < n_parts_total; i++) {
3465 		type = MCDI_ARRAY_DWORD(outbuf, NVRAM_PARTITIONS_OUT_TYPE_ID,
3466 					i);
3467 		rc = efx_ef10_mtd_probe_partition(efx, &parts[n_parts], type);
3468 		if (rc == 0)
3469 			n_parts++;
3470 		else if (rc != -ENODEV)
3471 			goto fail;
3472 	}
3473 
3474 	rc = efx_mtd_add(efx, &parts[0].common, n_parts, sizeof(*parts));
3475 fail:
3476 	if (rc)
3477 		kfree(parts);
3478 	return rc;
3479 }
3480 
3481 #endif /* CONFIG_SFC_MTD */
3482 
3483 static void efx_ef10_ptp_write_host_time(struct efx_nic *efx, u32 host_time)
3484 {
3485 	_efx_writed(efx, cpu_to_le32(host_time), ER_DZ_MC_DB_LWRD);
3486 }
3487 
3488 static int efx_ef10_rx_enable_timestamping(struct efx_channel *channel,
3489 					   bool temp)
3490 {
3491 	MCDI_DECLARE_BUF(inbuf, MC_CMD_PTP_IN_TIME_EVENT_SUBSCRIBE_LEN);
3492 	int rc;
3493 
3494 	if (channel->sync_events_state == SYNC_EVENTS_REQUESTED ||
3495 	    channel->sync_events_state == SYNC_EVENTS_VALID ||
3496 	    (temp && channel->sync_events_state == SYNC_EVENTS_DISABLED))
3497 		return 0;
3498 	channel->sync_events_state = SYNC_EVENTS_REQUESTED;
3499 
3500 	MCDI_SET_DWORD(inbuf, PTP_IN_OP, MC_CMD_PTP_OP_TIME_EVENT_SUBSCRIBE);
3501 	MCDI_SET_DWORD(inbuf, PTP_IN_PERIPH_ID, 0);
3502 	MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_SUBSCRIBE_QUEUE,
3503 		       channel->channel);
3504 
3505 	rc = efx_mcdi_rpc(channel->efx, MC_CMD_PTP,
3506 			  inbuf, sizeof(inbuf), NULL, 0, NULL);
3507 
3508 	if (rc != 0)
3509 		channel->sync_events_state = temp ? SYNC_EVENTS_QUIESCENT :
3510 						    SYNC_EVENTS_DISABLED;
3511 
3512 	return rc;
3513 }
3514 
3515 static int efx_ef10_rx_disable_timestamping(struct efx_channel *channel,
3516 					    bool temp)
3517 {
3518 	MCDI_DECLARE_BUF(inbuf, MC_CMD_PTP_IN_TIME_EVENT_UNSUBSCRIBE_LEN);
3519 	int rc;
3520 
3521 	if (channel->sync_events_state == SYNC_EVENTS_DISABLED ||
3522 	    (temp && channel->sync_events_state == SYNC_EVENTS_QUIESCENT))
3523 		return 0;
3524 	if (channel->sync_events_state == SYNC_EVENTS_QUIESCENT) {
3525 		channel->sync_events_state = SYNC_EVENTS_DISABLED;
3526 		return 0;
3527 	}
3528 	channel->sync_events_state = temp ? SYNC_EVENTS_QUIESCENT :
3529 					    SYNC_EVENTS_DISABLED;
3530 
3531 	MCDI_SET_DWORD(inbuf, PTP_IN_OP, MC_CMD_PTP_OP_TIME_EVENT_UNSUBSCRIBE);
3532 	MCDI_SET_DWORD(inbuf, PTP_IN_PERIPH_ID, 0);
3533 	MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_UNSUBSCRIBE_CONTROL,
3534 		       MC_CMD_PTP_IN_TIME_EVENT_UNSUBSCRIBE_SINGLE);
3535 	MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_UNSUBSCRIBE_QUEUE,
3536 		       channel->channel);
3537 
3538 	rc = efx_mcdi_rpc(channel->efx, MC_CMD_PTP,
3539 			  inbuf, sizeof(inbuf), NULL, 0, NULL);
3540 
3541 	return rc;
3542 }
3543 
3544 static int efx_ef10_ptp_set_ts_sync_events(struct efx_nic *efx, bool en,
3545 					   bool temp)
3546 {
3547 	int (*set)(struct efx_channel *channel, bool temp);
3548 	struct efx_channel *channel;
3549 
3550 	set = en ?
3551 	      efx_ef10_rx_enable_timestamping :
3552 	      efx_ef10_rx_disable_timestamping;
3553 
3554 	efx_for_each_channel(channel, efx) {
3555 		int rc = set(channel, temp);
3556 		if (en && rc != 0) {
3557 			efx_ef10_ptp_set_ts_sync_events(efx, false, temp);
3558 			return rc;
3559 		}
3560 	}
3561 
3562 	return 0;
3563 }
3564 
3565 static int efx_ef10_ptp_set_ts_config(struct efx_nic *efx,
3566 				      struct hwtstamp_config *init)
3567 {
3568 	int rc;
3569 
3570 	switch (init->rx_filter) {
3571 	case HWTSTAMP_FILTER_NONE:
3572 		efx_ef10_ptp_set_ts_sync_events(efx, false, false);
3573 		/* if TX timestamping is still requested then leave PTP on */
3574 		return efx_ptp_change_mode(efx,
3575 					   init->tx_type != HWTSTAMP_TX_OFF, 0);
3576 	case HWTSTAMP_FILTER_ALL:
3577 	case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
3578 	case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
3579 	case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
3580 	case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
3581 	case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
3582 	case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
3583 	case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
3584 	case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
3585 	case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
3586 	case HWTSTAMP_FILTER_PTP_V2_EVENT:
3587 	case HWTSTAMP_FILTER_PTP_V2_SYNC:
3588 	case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
3589 		init->rx_filter = HWTSTAMP_FILTER_ALL;
3590 		rc = efx_ptp_change_mode(efx, true, 0);
3591 		if (!rc)
3592 			rc = efx_ef10_ptp_set_ts_sync_events(efx, true, false);
3593 		if (rc)
3594 			efx_ptp_change_mode(efx, false, 0);
3595 		return rc;
3596 	default:
3597 		return -ERANGE;
3598 	}
3599 }
3600 
3601 const struct efx_nic_type efx_hunt_a0_nic_type = {
3602 	.mem_map_size = efx_ef10_mem_map_size,
3603 	.probe = efx_ef10_probe,
3604 	.remove = efx_ef10_remove,
3605 	.dimension_resources = efx_ef10_dimension_resources,
3606 	.init = efx_ef10_init_nic,
3607 	.fini = efx_port_dummy_op_void,
3608 	.map_reset_reason = efx_mcdi_map_reset_reason,
3609 	.map_reset_flags = efx_ef10_map_reset_flags,
3610 	.reset = efx_ef10_reset,
3611 	.probe_port = efx_mcdi_port_probe,
3612 	.remove_port = efx_mcdi_port_remove,
3613 	.fini_dmaq = efx_ef10_fini_dmaq,
3614 	.prepare_flr = efx_ef10_prepare_flr,
3615 	.finish_flr = efx_port_dummy_op_void,
3616 	.describe_stats = efx_ef10_describe_stats,
3617 	.update_stats = efx_ef10_update_stats,
3618 	.start_stats = efx_mcdi_mac_start_stats,
3619 	.pull_stats = efx_mcdi_mac_pull_stats,
3620 	.stop_stats = efx_mcdi_mac_stop_stats,
3621 	.set_id_led = efx_mcdi_set_id_led,
3622 	.push_irq_moderation = efx_ef10_push_irq_moderation,
3623 	.reconfigure_mac = efx_ef10_mac_reconfigure,
3624 	.check_mac_fault = efx_mcdi_mac_check_fault,
3625 	.reconfigure_port = efx_mcdi_port_reconfigure,
3626 	.get_wol = efx_ef10_get_wol,
3627 	.set_wol = efx_ef10_set_wol,
3628 	.resume_wol = efx_port_dummy_op_void,
3629 	.test_chip = efx_ef10_test_chip,
3630 	.test_nvram = efx_mcdi_nvram_test_all,
3631 	.mcdi_request = efx_ef10_mcdi_request,
3632 	.mcdi_poll_response = efx_ef10_mcdi_poll_response,
3633 	.mcdi_read_response = efx_ef10_mcdi_read_response,
3634 	.mcdi_poll_reboot = efx_ef10_mcdi_poll_reboot,
3635 	.irq_enable_master = efx_port_dummy_op_void,
3636 	.irq_test_generate = efx_ef10_irq_test_generate,
3637 	.irq_disable_non_ev = efx_port_dummy_op_void,
3638 	.irq_handle_msi = efx_ef10_msi_interrupt,
3639 	.irq_handle_legacy = efx_ef10_legacy_interrupt,
3640 	.tx_probe = efx_ef10_tx_probe,
3641 	.tx_init = efx_ef10_tx_init,
3642 	.tx_remove = efx_ef10_tx_remove,
3643 	.tx_write = efx_ef10_tx_write,
3644 	.rx_push_rss_config = efx_ef10_rx_push_rss_config,
3645 	.rx_probe = efx_ef10_rx_probe,
3646 	.rx_init = efx_ef10_rx_init,
3647 	.rx_remove = efx_ef10_rx_remove,
3648 	.rx_write = efx_ef10_rx_write,
3649 	.rx_defer_refill = efx_ef10_rx_defer_refill,
3650 	.ev_probe = efx_ef10_ev_probe,
3651 	.ev_init = efx_ef10_ev_init,
3652 	.ev_fini = efx_ef10_ev_fini,
3653 	.ev_remove = efx_ef10_ev_remove,
3654 	.ev_process = efx_ef10_ev_process,
3655 	.ev_read_ack = efx_ef10_ev_read_ack,
3656 	.ev_test_generate = efx_ef10_ev_test_generate,
3657 	.filter_table_probe = efx_ef10_filter_table_probe,
3658 	.filter_table_restore = efx_ef10_filter_table_restore,
3659 	.filter_table_remove = efx_ef10_filter_table_remove,
3660 	.filter_update_rx_scatter = efx_ef10_filter_update_rx_scatter,
3661 	.filter_insert = efx_ef10_filter_insert,
3662 	.filter_remove_safe = efx_ef10_filter_remove_safe,
3663 	.filter_get_safe = efx_ef10_filter_get_safe,
3664 	.filter_clear_rx = efx_ef10_filter_clear_rx,
3665 	.filter_count_rx_used = efx_ef10_filter_count_rx_used,
3666 	.filter_get_rx_id_limit = efx_ef10_filter_get_rx_id_limit,
3667 	.filter_get_rx_ids = efx_ef10_filter_get_rx_ids,
3668 #ifdef CONFIG_RFS_ACCEL
3669 	.filter_rfs_insert = efx_ef10_filter_rfs_insert,
3670 	.filter_rfs_expire_one = efx_ef10_filter_rfs_expire_one,
3671 #endif
3672 #ifdef CONFIG_SFC_MTD
3673 	.mtd_probe = efx_ef10_mtd_probe,
3674 	.mtd_rename = efx_mcdi_mtd_rename,
3675 	.mtd_read = efx_mcdi_mtd_read,
3676 	.mtd_erase = efx_mcdi_mtd_erase,
3677 	.mtd_write = efx_mcdi_mtd_write,
3678 	.mtd_sync = efx_mcdi_mtd_sync,
3679 #endif
3680 	.ptp_write_host_time = efx_ef10_ptp_write_host_time,
3681 	.ptp_set_ts_sync_events = efx_ef10_ptp_set_ts_sync_events,
3682 	.ptp_set_ts_config = efx_ef10_ptp_set_ts_config,
3683 
3684 	.revision = EFX_REV_HUNT_A0,
3685 	.max_dma_mask = DMA_BIT_MASK(ESF_DZ_TX_KER_BUF_ADDR_WIDTH),
3686 	.rx_prefix_size = ES_DZ_RX_PREFIX_SIZE,
3687 	.rx_hash_offset = ES_DZ_RX_PREFIX_HASH_OFST,
3688 	.rx_ts_offset = ES_DZ_RX_PREFIX_TSTAMP_OFST,
3689 	.can_rx_scatter = true,
3690 	.always_rx_scatter = true,
3691 	.max_interrupt_mode = EFX_INT_MODE_MSIX,
3692 	.timer_period_max = 1 << ERF_DD_EVQ_IND_TIMER_VAL_WIDTH,
3693 	.offload_features = (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
3694 			     NETIF_F_RXHASH | NETIF_F_NTUPLE),
3695 	.mcdi_max_ver = 2,
3696 	.max_rx_ip_filters = HUNT_FILTER_TBL_ROWS,
3697 	.hwtstamp_filters = 1 << HWTSTAMP_FILTER_NONE |
3698 			    1 << HWTSTAMP_FILTER_ALL,
3699 };
3700