xref: /openbmc/linux/drivers/net/ethernet/sfc/ef10.c (revision 930beb5a)
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 <linux/in.h>
18 #include <linux/jhash.h>
19 #include <linux/wait.h>
20 #include <linux/workqueue.h>
21 
22 /* Hardware control for EF10 architecture including 'Huntington'. */
23 
24 #define EFX_EF10_DRVGEN_EV		7
25 enum {
26 	EFX_EF10_TEST = 1,
27 	EFX_EF10_REFILL,
28 };
29 
30 /* The reserved RSS context value */
31 #define EFX_EF10_RSS_CONTEXT_INVALID	0xffffffff
32 
33 /* The filter table(s) are managed by firmware and we have write-only
34  * access.  When removing filters we must identify them to the
35  * firmware by a 64-bit handle, but this is too wide for Linux kernel
36  * interfaces (32-bit for RX NFC, 16-bit for RFS).  Also, we need to
37  * be able to tell in advance whether a requested insertion will
38  * replace an existing filter.  Therefore we maintain a software hash
39  * table, which should be at least as large as the hardware hash
40  * table.
41  *
42  * Huntington has a single 8K filter table shared between all filter
43  * types and both ports.
44  */
45 #define HUNT_FILTER_TBL_ROWS 8192
46 
47 struct efx_ef10_filter_table {
48 /* The RX match field masks supported by this fw & hw, in order of priority */
49 	enum efx_filter_match_flags rx_match_flags[
50 		MC_CMD_GET_PARSER_DISP_INFO_OUT_SUPPORTED_MATCHES_MAXNUM];
51 	unsigned int rx_match_count;
52 
53 	struct {
54 		unsigned long spec;	/* pointer to spec plus flag bits */
55 /* BUSY flag indicates that an update is in progress.  STACK_OLD is
56  * used to mark and sweep stack-owned MAC filters.
57  */
58 #define EFX_EF10_FILTER_FLAG_BUSY	1UL
59 #define EFX_EF10_FILTER_FLAG_STACK_OLD	2UL
60 #define EFX_EF10_FILTER_FLAGS		3UL
61 		u64 handle;		/* firmware handle */
62 	} *entry;
63 	wait_queue_head_t waitq;
64 /* Shadow of net_device address lists, guarded by mac_lock */
65 #define EFX_EF10_FILTER_STACK_UC_MAX	32
66 #define EFX_EF10_FILTER_STACK_MC_MAX	256
67 	struct {
68 		u8 addr[ETH_ALEN];
69 		u16 id;
70 	} stack_uc_list[EFX_EF10_FILTER_STACK_UC_MAX],
71 	  stack_mc_list[EFX_EF10_FILTER_STACK_MC_MAX];
72 	int stack_uc_count;		/* negative for PROMISC */
73 	int stack_mc_count;		/* negative for PROMISC/ALLMULTI */
74 };
75 
76 /* An arbitrary search limit for the software hash table */
77 #define EFX_EF10_FILTER_SEARCH_LIMIT 200
78 
79 static void efx_ef10_rx_push_indir_table(struct efx_nic *efx);
80 static void efx_ef10_rx_free_indir_table(struct efx_nic *efx);
81 static void efx_ef10_filter_table_remove(struct efx_nic *efx);
82 
83 static int efx_ef10_get_warm_boot_count(struct efx_nic *efx)
84 {
85 	efx_dword_t reg;
86 
87 	efx_readd(efx, &reg, ER_DZ_BIU_MC_SFT_STATUS);
88 	return EFX_DWORD_FIELD(reg, EFX_WORD_1) == 0xb007 ?
89 		EFX_DWORD_FIELD(reg, EFX_WORD_0) : -EIO;
90 }
91 
92 static unsigned int efx_ef10_mem_map_size(struct efx_nic *efx)
93 {
94 	return resource_size(&efx->pci_dev->resource[EFX_MEM_BAR]);
95 }
96 
97 static int efx_ef10_init_datapath_caps(struct efx_nic *efx)
98 {
99 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_CAPABILITIES_OUT_LEN);
100 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
101 	size_t outlen;
102 	int rc;
103 
104 	BUILD_BUG_ON(MC_CMD_GET_CAPABILITIES_IN_LEN != 0);
105 
106 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_CAPABILITIES, NULL, 0,
107 			  outbuf, sizeof(outbuf), &outlen);
108 	if (rc)
109 		return rc;
110 	if (outlen < sizeof(outbuf)) {
111 		netif_err(efx, drv, efx->net_dev,
112 			  "unable to read datapath firmware capabilities\n");
113 		return -EIO;
114 	}
115 
116 	nic_data->datapath_caps =
117 		MCDI_DWORD(outbuf, GET_CAPABILITIES_OUT_FLAGS1);
118 
119 	if (!(nic_data->datapath_caps &
120 	      (1 << MC_CMD_GET_CAPABILITIES_OUT_TX_TSO_LBN))) {
121 		netif_err(efx, drv, efx->net_dev,
122 			  "current firmware does not support TSO\n");
123 		return -ENODEV;
124 	}
125 
126 	if (!(nic_data->datapath_caps &
127 	      (1 << MC_CMD_GET_CAPABILITIES_OUT_RX_PREFIX_LEN_14_LBN))) {
128 		netif_err(efx, probe, efx->net_dev,
129 			  "current firmware does not support an RX prefix\n");
130 		return -ENODEV;
131 	}
132 
133 	return 0;
134 }
135 
136 static int efx_ef10_get_sysclk_freq(struct efx_nic *efx)
137 {
138 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_CLOCK_OUT_LEN);
139 	int rc;
140 
141 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_CLOCK, NULL, 0,
142 			  outbuf, sizeof(outbuf), NULL);
143 	if (rc)
144 		return rc;
145 	rc = MCDI_DWORD(outbuf, GET_CLOCK_OUT_SYS_FREQ);
146 	return rc > 0 ? rc : -ERANGE;
147 }
148 
149 static int efx_ef10_get_mac_address(struct efx_nic *efx, u8 *mac_address)
150 {
151 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_MAC_ADDRESSES_OUT_LEN);
152 	size_t outlen;
153 	int rc;
154 
155 	BUILD_BUG_ON(MC_CMD_GET_MAC_ADDRESSES_IN_LEN != 0);
156 
157 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_MAC_ADDRESSES, NULL, 0,
158 			  outbuf, sizeof(outbuf), &outlen);
159 	if (rc)
160 		return rc;
161 	if (outlen < MC_CMD_GET_MAC_ADDRESSES_OUT_LEN)
162 		return -EIO;
163 
164 	memcpy(mac_address,
165 	       MCDI_PTR(outbuf, GET_MAC_ADDRESSES_OUT_MAC_ADDR_BASE), ETH_ALEN);
166 	return 0;
167 }
168 
169 static int efx_ef10_probe(struct efx_nic *efx)
170 {
171 	struct efx_ef10_nic_data *nic_data;
172 	int i, rc;
173 
174 	/* We can have one VI for each 8K region.  However we need
175 	 * multiple TX queues per channel.
176 	 */
177 	efx->max_channels =
178 		min_t(unsigned int,
179 		      EFX_MAX_CHANNELS,
180 		      resource_size(&efx->pci_dev->resource[EFX_MEM_BAR]) /
181 		      (EFX_VI_PAGE_SIZE * EFX_TXQ_TYPES));
182 	BUG_ON(efx->max_channels == 0);
183 
184 	nic_data = kzalloc(sizeof(*nic_data), GFP_KERNEL);
185 	if (!nic_data)
186 		return -ENOMEM;
187 	efx->nic_data = nic_data;
188 
189 	rc = efx_nic_alloc_buffer(efx, &nic_data->mcdi_buf,
190 				  8 + MCDI_CTL_SDU_LEN_MAX_V2, GFP_KERNEL);
191 	if (rc)
192 		goto fail1;
193 
194 	/* Get the MC's warm boot count.  In case it's rebooting right
195 	 * now, be prepared to retry.
196 	 */
197 	i = 0;
198 	for (;;) {
199 		rc = efx_ef10_get_warm_boot_count(efx);
200 		if (rc >= 0)
201 			break;
202 		if (++i == 5)
203 			goto fail2;
204 		ssleep(1);
205 	}
206 	nic_data->warm_boot_count = rc;
207 
208 	nic_data->rx_rss_context = EFX_EF10_RSS_CONTEXT_INVALID;
209 
210 	/* In case we're recovering from a crash (kexec), we want to
211 	 * cancel any outstanding request by the previous user of this
212 	 * function.  We send a special message using the least
213 	 * significant bits of the 'high' (doorbell) register.
214 	 */
215 	_efx_writed(efx, cpu_to_le32(1), ER_DZ_MC_DB_HWRD);
216 
217 	rc = efx_mcdi_init(efx);
218 	if (rc)
219 		goto fail2;
220 
221 	/* Reset (most) configuration for this function */
222 	rc = efx_mcdi_reset(efx, RESET_TYPE_ALL);
223 	if (rc)
224 		goto fail3;
225 
226 	/* Enable event logging */
227 	rc = efx_mcdi_log_ctrl(efx, true, false, 0);
228 	if (rc)
229 		goto fail3;
230 
231 	rc = efx_ef10_init_datapath_caps(efx);
232 	if (rc < 0)
233 		goto fail3;
234 
235 	efx->rx_packet_len_offset =
236 		ES_DZ_RX_PREFIX_PKTLEN_OFST - ES_DZ_RX_PREFIX_SIZE;
237 
238 	rc = efx_mcdi_port_get_number(efx);
239 	if (rc < 0)
240 		goto fail3;
241 	efx->port_num = rc;
242 
243 	rc = efx_ef10_get_mac_address(efx, efx->net_dev->perm_addr);
244 	if (rc)
245 		goto fail3;
246 
247 	rc = efx_ef10_get_sysclk_freq(efx);
248 	if (rc < 0)
249 		goto fail3;
250 	efx->timer_quantum_ns = 1536000 / rc; /* 1536 cycles */
251 
252 	/* Check whether firmware supports bug 35388 workaround */
253 	rc = efx_mcdi_set_workaround(efx, MC_CMD_WORKAROUND_BUG35388, true);
254 	if (rc == 0)
255 		nic_data->workaround_35388 = true;
256 	else if (rc != -ENOSYS && rc != -ENOENT)
257 		goto fail3;
258 	netif_dbg(efx, probe, efx->net_dev,
259 		  "workaround for bug 35388 is %sabled\n",
260 		  nic_data->workaround_35388 ? "en" : "dis");
261 
262 	rc = efx_mcdi_mon_probe(efx);
263 	if (rc)
264 		goto fail3;
265 
266 	return 0;
267 
268 fail3:
269 	efx_mcdi_fini(efx);
270 fail2:
271 	efx_nic_free_buffer(efx, &nic_data->mcdi_buf);
272 fail1:
273 	kfree(nic_data);
274 	efx->nic_data = NULL;
275 	return rc;
276 }
277 
278 static int efx_ef10_free_vis(struct efx_nic *efx)
279 {
280 	int rc = efx_mcdi_rpc(efx, MC_CMD_FREE_VIS, NULL, 0, NULL, 0, NULL);
281 
282 	/* -EALREADY means nothing to free, so ignore */
283 	if (rc == -EALREADY)
284 		rc = 0;
285 	return rc;
286 }
287 
288 #ifdef EFX_USE_PIO
289 
290 static void efx_ef10_free_piobufs(struct efx_nic *efx)
291 {
292 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
293 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FREE_PIOBUF_IN_LEN);
294 	unsigned int i;
295 	int rc;
296 
297 	BUILD_BUG_ON(MC_CMD_FREE_PIOBUF_OUT_LEN != 0);
298 
299 	for (i = 0; i < nic_data->n_piobufs; i++) {
300 		MCDI_SET_DWORD(inbuf, FREE_PIOBUF_IN_PIOBUF_HANDLE,
301 			       nic_data->piobuf_handle[i]);
302 		rc = efx_mcdi_rpc(efx, MC_CMD_FREE_PIOBUF, inbuf, sizeof(inbuf),
303 				  NULL, 0, NULL);
304 		WARN_ON(rc);
305 	}
306 
307 	nic_data->n_piobufs = 0;
308 }
309 
310 static int efx_ef10_alloc_piobufs(struct efx_nic *efx, unsigned int n)
311 {
312 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
313 	MCDI_DECLARE_BUF(outbuf, MC_CMD_ALLOC_PIOBUF_OUT_LEN);
314 	unsigned int i;
315 	size_t outlen;
316 	int rc = 0;
317 
318 	BUILD_BUG_ON(MC_CMD_ALLOC_PIOBUF_IN_LEN != 0);
319 
320 	for (i = 0; i < n; i++) {
321 		rc = efx_mcdi_rpc(efx, MC_CMD_ALLOC_PIOBUF, NULL, 0,
322 				  outbuf, sizeof(outbuf), &outlen);
323 		if (rc)
324 			break;
325 		if (outlen < MC_CMD_ALLOC_PIOBUF_OUT_LEN) {
326 			rc = -EIO;
327 			break;
328 		}
329 		nic_data->piobuf_handle[i] =
330 			MCDI_DWORD(outbuf, ALLOC_PIOBUF_OUT_PIOBUF_HANDLE);
331 		netif_dbg(efx, probe, efx->net_dev,
332 			  "allocated PIO buffer %u handle %x\n", i,
333 			  nic_data->piobuf_handle[i]);
334 	}
335 
336 	nic_data->n_piobufs = i;
337 	if (rc)
338 		efx_ef10_free_piobufs(efx);
339 	return rc;
340 }
341 
342 static int efx_ef10_link_piobufs(struct efx_nic *efx)
343 {
344 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
345 	MCDI_DECLARE_BUF(inbuf,
346 			 max(MC_CMD_LINK_PIOBUF_IN_LEN,
347 			     MC_CMD_UNLINK_PIOBUF_IN_LEN));
348 	struct efx_channel *channel;
349 	struct efx_tx_queue *tx_queue;
350 	unsigned int offset, index;
351 	int rc;
352 
353 	BUILD_BUG_ON(MC_CMD_LINK_PIOBUF_OUT_LEN != 0);
354 	BUILD_BUG_ON(MC_CMD_UNLINK_PIOBUF_OUT_LEN != 0);
355 
356 	/* Link a buffer to each VI in the write-combining mapping */
357 	for (index = 0; index < nic_data->n_piobufs; ++index) {
358 		MCDI_SET_DWORD(inbuf, LINK_PIOBUF_IN_PIOBUF_HANDLE,
359 			       nic_data->piobuf_handle[index]);
360 		MCDI_SET_DWORD(inbuf, LINK_PIOBUF_IN_TXQ_INSTANCE,
361 			       nic_data->pio_write_vi_base + index);
362 		rc = efx_mcdi_rpc(efx, MC_CMD_LINK_PIOBUF,
363 				  inbuf, MC_CMD_LINK_PIOBUF_IN_LEN,
364 				  NULL, 0, NULL);
365 		if (rc) {
366 			netif_err(efx, drv, efx->net_dev,
367 				  "failed to link VI %u to PIO buffer %u (%d)\n",
368 				  nic_data->pio_write_vi_base + index, index,
369 				  rc);
370 			goto fail;
371 		}
372 		netif_dbg(efx, probe, efx->net_dev,
373 			  "linked VI %u to PIO buffer %u\n",
374 			  nic_data->pio_write_vi_base + index, index);
375 	}
376 
377 	/* Link a buffer to each TX queue */
378 	efx_for_each_channel(channel, efx) {
379 		efx_for_each_channel_tx_queue(tx_queue, channel) {
380 			/* We assign the PIO buffers to queues in
381 			 * reverse order to allow for the following
382 			 * special case.
383 			 */
384 			offset = ((efx->tx_channel_offset + efx->n_tx_channels -
385 				   tx_queue->channel->channel - 1) *
386 				  efx_piobuf_size);
387 			index = offset / ER_DZ_TX_PIOBUF_SIZE;
388 			offset = offset % ER_DZ_TX_PIOBUF_SIZE;
389 
390 			/* When the host page size is 4K, the first
391 			 * host page in the WC mapping may be within
392 			 * the same VI page as the last TX queue.  We
393 			 * can only link one buffer to each VI.
394 			 */
395 			if (tx_queue->queue == nic_data->pio_write_vi_base) {
396 				BUG_ON(index != 0);
397 				rc = 0;
398 			} else {
399 				MCDI_SET_DWORD(inbuf,
400 					       LINK_PIOBUF_IN_PIOBUF_HANDLE,
401 					       nic_data->piobuf_handle[index]);
402 				MCDI_SET_DWORD(inbuf,
403 					       LINK_PIOBUF_IN_TXQ_INSTANCE,
404 					       tx_queue->queue);
405 				rc = efx_mcdi_rpc(efx, MC_CMD_LINK_PIOBUF,
406 						  inbuf, MC_CMD_LINK_PIOBUF_IN_LEN,
407 						  NULL, 0, NULL);
408 			}
409 
410 			if (rc) {
411 				/* This is non-fatal; the TX path just
412 				 * won't use PIO for this queue
413 				 */
414 				netif_err(efx, drv, efx->net_dev,
415 					  "failed to link VI %u to PIO buffer %u (%d)\n",
416 					  tx_queue->queue, index, rc);
417 				tx_queue->piobuf = NULL;
418 			} else {
419 				tx_queue->piobuf =
420 					nic_data->pio_write_base +
421 					index * EFX_VI_PAGE_SIZE + offset;
422 				tx_queue->piobuf_offset = offset;
423 				netif_dbg(efx, probe, efx->net_dev,
424 					  "linked VI %u to PIO buffer %u offset %x addr %p\n",
425 					  tx_queue->queue, index,
426 					  tx_queue->piobuf_offset,
427 					  tx_queue->piobuf);
428 			}
429 		}
430 	}
431 
432 	return 0;
433 
434 fail:
435 	while (index--) {
436 		MCDI_SET_DWORD(inbuf, UNLINK_PIOBUF_IN_TXQ_INSTANCE,
437 			       nic_data->pio_write_vi_base + index);
438 		efx_mcdi_rpc(efx, MC_CMD_UNLINK_PIOBUF,
439 			     inbuf, MC_CMD_UNLINK_PIOBUF_IN_LEN,
440 			     NULL, 0, NULL);
441 	}
442 	return rc;
443 }
444 
445 #else /* !EFX_USE_PIO */
446 
447 static int efx_ef10_alloc_piobufs(struct efx_nic *efx, unsigned int n)
448 {
449 	return n == 0 ? 0 : -ENOBUFS;
450 }
451 
452 static int efx_ef10_link_piobufs(struct efx_nic *efx)
453 {
454 	return 0;
455 }
456 
457 static void efx_ef10_free_piobufs(struct efx_nic *efx)
458 {
459 }
460 
461 #endif /* EFX_USE_PIO */
462 
463 static void efx_ef10_remove(struct efx_nic *efx)
464 {
465 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
466 	int rc;
467 
468 	efx_mcdi_mon_remove(efx);
469 
470 	/* This needs to be after efx_ptp_remove_channel() with no filters */
471 	efx_ef10_rx_free_indir_table(efx);
472 
473 	if (nic_data->wc_membase)
474 		iounmap(nic_data->wc_membase);
475 
476 	rc = efx_ef10_free_vis(efx);
477 	WARN_ON(rc != 0);
478 
479 	if (!nic_data->must_restore_piobufs)
480 		efx_ef10_free_piobufs(efx);
481 
482 	efx_mcdi_fini(efx);
483 	efx_nic_free_buffer(efx, &nic_data->mcdi_buf);
484 	kfree(nic_data);
485 }
486 
487 static int efx_ef10_alloc_vis(struct efx_nic *efx,
488 			      unsigned int min_vis, unsigned int max_vis)
489 {
490 	MCDI_DECLARE_BUF(inbuf, MC_CMD_ALLOC_VIS_IN_LEN);
491 	MCDI_DECLARE_BUF(outbuf, MC_CMD_ALLOC_VIS_OUT_LEN);
492 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
493 	size_t outlen;
494 	int rc;
495 
496 	MCDI_SET_DWORD(inbuf, ALLOC_VIS_IN_MIN_VI_COUNT, min_vis);
497 	MCDI_SET_DWORD(inbuf, ALLOC_VIS_IN_MAX_VI_COUNT, max_vis);
498 	rc = efx_mcdi_rpc(efx, MC_CMD_ALLOC_VIS, inbuf, sizeof(inbuf),
499 			  outbuf, sizeof(outbuf), &outlen);
500 	if (rc != 0)
501 		return rc;
502 
503 	if (outlen < MC_CMD_ALLOC_VIS_OUT_LEN)
504 		return -EIO;
505 
506 	netif_dbg(efx, drv, efx->net_dev, "base VI is A0x%03x\n",
507 		  MCDI_DWORD(outbuf, ALLOC_VIS_OUT_VI_BASE));
508 
509 	nic_data->vi_base = MCDI_DWORD(outbuf, ALLOC_VIS_OUT_VI_BASE);
510 	nic_data->n_allocated_vis = MCDI_DWORD(outbuf, ALLOC_VIS_OUT_VI_COUNT);
511 	return 0;
512 }
513 
514 /* Note that the failure path of this function does not free
515  * resources, as this will be done by efx_ef10_remove().
516  */
517 static int efx_ef10_dimension_resources(struct efx_nic *efx)
518 {
519 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
520 	unsigned int uc_mem_map_size, wc_mem_map_size;
521 	unsigned int min_vis, pio_write_vi_base, max_vis;
522 	void __iomem *membase;
523 	int rc;
524 
525 	min_vis = max(efx->n_channels, efx->n_tx_channels * EFX_TXQ_TYPES);
526 
527 #ifdef EFX_USE_PIO
528 	/* Try to allocate PIO buffers if wanted and if the full
529 	 * number of PIO buffers would be sufficient to allocate one
530 	 * copy-buffer per TX channel.  Failure is non-fatal, as there
531 	 * are only a small number of PIO buffers shared between all
532 	 * functions of the controller.
533 	 */
534 	if (efx_piobuf_size != 0 &&
535 	    ER_DZ_TX_PIOBUF_SIZE / efx_piobuf_size * EF10_TX_PIOBUF_COUNT >=
536 	    efx->n_tx_channels) {
537 		unsigned int n_piobufs =
538 			DIV_ROUND_UP(efx->n_tx_channels,
539 				     ER_DZ_TX_PIOBUF_SIZE / efx_piobuf_size);
540 
541 		rc = efx_ef10_alloc_piobufs(efx, n_piobufs);
542 		if (rc)
543 			netif_err(efx, probe, efx->net_dev,
544 				  "failed to allocate PIO buffers (%d)\n", rc);
545 		else
546 			netif_dbg(efx, probe, efx->net_dev,
547 				  "allocated %u PIO buffers\n", n_piobufs);
548 	}
549 #else
550 	nic_data->n_piobufs = 0;
551 #endif
552 
553 	/* PIO buffers should be mapped with write-combining enabled,
554 	 * and we want to make single UC and WC mappings rather than
555 	 * several of each (in fact that's the only option if host
556 	 * page size is >4K).  So we may allocate some extra VIs just
557 	 * for writing PIO buffers through.
558 	 */
559 	uc_mem_map_size = PAGE_ALIGN((min_vis - 1) * EFX_VI_PAGE_SIZE +
560 				     ER_DZ_TX_PIOBUF);
561 	if (nic_data->n_piobufs) {
562 		pio_write_vi_base = uc_mem_map_size / EFX_VI_PAGE_SIZE;
563 		wc_mem_map_size = (PAGE_ALIGN((pio_write_vi_base +
564 					       nic_data->n_piobufs) *
565 					      EFX_VI_PAGE_SIZE) -
566 				   uc_mem_map_size);
567 		max_vis = pio_write_vi_base + nic_data->n_piobufs;
568 	} else {
569 		pio_write_vi_base = 0;
570 		wc_mem_map_size = 0;
571 		max_vis = min_vis;
572 	}
573 
574 	/* In case the last attached driver failed to free VIs, do it now */
575 	rc = efx_ef10_free_vis(efx);
576 	if (rc != 0)
577 		return rc;
578 
579 	rc = efx_ef10_alloc_vis(efx, min_vis, max_vis);
580 	if (rc != 0)
581 		return rc;
582 
583 	/* If we didn't get enough VIs to map all the PIO buffers, free the
584 	 * PIO buffers
585 	 */
586 	if (nic_data->n_piobufs &&
587 	    nic_data->n_allocated_vis <
588 	    pio_write_vi_base + nic_data->n_piobufs) {
589 		netif_dbg(efx, probe, efx->net_dev,
590 			  "%u VIs are not sufficient to map %u PIO buffers\n",
591 			  nic_data->n_allocated_vis, nic_data->n_piobufs);
592 		efx_ef10_free_piobufs(efx);
593 	}
594 
595 	/* Shrink the original UC mapping of the memory BAR */
596 	membase = ioremap_nocache(efx->membase_phys, uc_mem_map_size);
597 	if (!membase) {
598 		netif_err(efx, probe, efx->net_dev,
599 			  "could not shrink memory BAR to %x\n",
600 			  uc_mem_map_size);
601 		return -ENOMEM;
602 	}
603 	iounmap(efx->membase);
604 	efx->membase = membase;
605 
606 	/* Set up the WC mapping if needed */
607 	if (wc_mem_map_size) {
608 		nic_data->wc_membase = ioremap_wc(efx->membase_phys +
609 						  uc_mem_map_size,
610 						  wc_mem_map_size);
611 		if (!nic_data->wc_membase) {
612 			netif_err(efx, probe, efx->net_dev,
613 				  "could not allocate WC mapping of size %x\n",
614 				  wc_mem_map_size);
615 			return -ENOMEM;
616 		}
617 		nic_data->pio_write_vi_base = pio_write_vi_base;
618 		nic_data->pio_write_base =
619 			nic_data->wc_membase +
620 			(pio_write_vi_base * EFX_VI_PAGE_SIZE + ER_DZ_TX_PIOBUF -
621 			 uc_mem_map_size);
622 
623 		rc = efx_ef10_link_piobufs(efx);
624 		if (rc)
625 			efx_ef10_free_piobufs(efx);
626 	}
627 
628 	netif_dbg(efx, probe, efx->net_dev,
629 		  "memory BAR at %pa (virtual %p+%x UC, %p+%x WC)\n",
630 		  &efx->membase_phys, efx->membase, uc_mem_map_size,
631 		  nic_data->wc_membase, wc_mem_map_size);
632 
633 	return 0;
634 }
635 
636 static int efx_ef10_init_nic(struct efx_nic *efx)
637 {
638 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
639 	int rc;
640 
641 	if (nic_data->must_check_datapath_caps) {
642 		rc = efx_ef10_init_datapath_caps(efx);
643 		if (rc)
644 			return rc;
645 		nic_data->must_check_datapath_caps = false;
646 	}
647 
648 	if (nic_data->must_realloc_vis) {
649 		/* We cannot let the number of VIs change now */
650 		rc = efx_ef10_alloc_vis(efx, nic_data->n_allocated_vis,
651 					nic_data->n_allocated_vis);
652 		if (rc)
653 			return rc;
654 		nic_data->must_realloc_vis = false;
655 	}
656 
657 	if (nic_data->must_restore_piobufs && nic_data->n_piobufs) {
658 		rc = efx_ef10_alloc_piobufs(efx, nic_data->n_piobufs);
659 		if (rc == 0) {
660 			rc = efx_ef10_link_piobufs(efx);
661 			if (rc)
662 				efx_ef10_free_piobufs(efx);
663 		}
664 
665 		/* Log an error on failure, but this is non-fatal */
666 		if (rc)
667 			netif_err(efx, drv, efx->net_dev,
668 				  "failed to restore PIO buffers (%d)\n", rc);
669 		nic_data->must_restore_piobufs = false;
670 	}
671 
672 	efx_ef10_rx_push_indir_table(efx);
673 	return 0;
674 }
675 
676 static int efx_ef10_map_reset_flags(u32 *flags)
677 {
678 	enum {
679 		EF10_RESET_PORT = ((ETH_RESET_MAC | ETH_RESET_PHY) <<
680 				   ETH_RESET_SHARED_SHIFT),
681 		EF10_RESET_MC = ((ETH_RESET_DMA | ETH_RESET_FILTER |
682 				  ETH_RESET_OFFLOAD | ETH_RESET_MAC |
683 				  ETH_RESET_PHY | ETH_RESET_MGMT) <<
684 				 ETH_RESET_SHARED_SHIFT)
685 	};
686 
687 	/* We assume for now that our PCI function is permitted to
688 	 * reset everything.
689 	 */
690 
691 	if ((*flags & EF10_RESET_MC) == EF10_RESET_MC) {
692 		*flags &= ~EF10_RESET_MC;
693 		return RESET_TYPE_WORLD;
694 	}
695 
696 	if ((*flags & EF10_RESET_PORT) == EF10_RESET_PORT) {
697 		*flags &= ~EF10_RESET_PORT;
698 		return RESET_TYPE_ALL;
699 	}
700 
701 	/* no invisible reset implemented */
702 
703 	return -EINVAL;
704 }
705 
706 #define EF10_DMA_STAT(ext_name, mcdi_name)			\
707 	[EF10_STAT_ ## ext_name] =				\
708 	{ #ext_name, 64, 8 * MC_CMD_MAC_ ## mcdi_name }
709 #define EF10_DMA_INVIS_STAT(int_name, mcdi_name)		\
710 	[EF10_STAT_ ## int_name] =				\
711 	{ NULL, 64, 8 * MC_CMD_MAC_ ## mcdi_name }
712 #define EF10_OTHER_STAT(ext_name)				\
713 	[EF10_STAT_ ## ext_name] = { #ext_name, 0, 0 }
714 
715 static const struct efx_hw_stat_desc efx_ef10_stat_desc[EF10_STAT_COUNT] = {
716 	EF10_DMA_STAT(tx_bytes, TX_BYTES),
717 	EF10_DMA_STAT(tx_packets, TX_PKTS),
718 	EF10_DMA_STAT(tx_pause, TX_PAUSE_PKTS),
719 	EF10_DMA_STAT(tx_control, TX_CONTROL_PKTS),
720 	EF10_DMA_STAT(tx_unicast, TX_UNICAST_PKTS),
721 	EF10_DMA_STAT(tx_multicast, TX_MULTICAST_PKTS),
722 	EF10_DMA_STAT(tx_broadcast, TX_BROADCAST_PKTS),
723 	EF10_DMA_STAT(tx_lt64, TX_LT64_PKTS),
724 	EF10_DMA_STAT(tx_64, TX_64_PKTS),
725 	EF10_DMA_STAT(tx_65_to_127, TX_65_TO_127_PKTS),
726 	EF10_DMA_STAT(tx_128_to_255, TX_128_TO_255_PKTS),
727 	EF10_DMA_STAT(tx_256_to_511, TX_256_TO_511_PKTS),
728 	EF10_DMA_STAT(tx_512_to_1023, TX_512_TO_1023_PKTS),
729 	EF10_DMA_STAT(tx_1024_to_15xx, TX_1024_TO_15XX_PKTS),
730 	EF10_DMA_STAT(tx_15xx_to_jumbo, TX_15XX_TO_JUMBO_PKTS),
731 	EF10_DMA_STAT(rx_bytes, RX_BYTES),
732 	EF10_DMA_INVIS_STAT(rx_bytes_minus_good_bytes, RX_BAD_BYTES),
733 	EF10_OTHER_STAT(rx_good_bytes),
734 	EF10_OTHER_STAT(rx_bad_bytes),
735 	EF10_DMA_STAT(rx_packets, RX_PKTS),
736 	EF10_DMA_STAT(rx_good, RX_GOOD_PKTS),
737 	EF10_DMA_STAT(rx_bad, RX_BAD_FCS_PKTS),
738 	EF10_DMA_STAT(rx_pause, RX_PAUSE_PKTS),
739 	EF10_DMA_STAT(rx_control, RX_CONTROL_PKTS),
740 	EF10_DMA_STAT(rx_unicast, RX_UNICAST_PKTS),
741 	EF10_DMA_STAT(rx_multicast, RX_MULTICAST_PKTS),
742 	EF10_DMA_STAT(rx_broadcast, RX_BROADCAST_PKTS),
743 	EF10_DMA_STAT(rx_lt64, RX_UNDERSIZE_PKTS),
744 	EF10_DMA_STAT(rx_64, RX_64_PKTS),
745 	EF10_DMA_STAT(rx_65_to_127, RX_65_TO_127_PKTS),
746 	EF10_DMA_STAT(rx_128_to_255, RX_128_TO_255_PKTS),
747 	EF10_DMA_STAT(rx_256_to_511, RX_256_TO_511_PKTS),
748 	EF10_DMA_STAT(rx_512_to_1023, RX_512_TO_1023_PKTS),
749 	EF10_DMA_STAT(rx_1024_to_15xx, RX_1024_TO_15XX_PKTS),
750 	EF10_DMA_STAT(rx_15xx_to_jumbo, RX_15XX_TO_JUMBO_PKTS),
751 	EF10_DMA_STAT(rx_gtjumbo, RX_GTJUMBO_PKTS),
752 	EF10_DMA_STAT(rx_bad_gtjumbo, RX_JABBER_PKTS),
753 	EF10_DMA_STAT(rx_overflow, RX_OVERFLOW_PKTS),
754 	EF10_DMA_STAT(rx_align_error, RX_ALIGN_ERROR_PKTS),
755 	EF10_DMA_STAT(rx_length_error, RX_LENGTH_ERROR_PKTS),
756 	EF10_DMA_STAT(rx_nodesc_drops, RX_NODESC_DROPS),
757 	EF10_DMA_STAT(rx_pm_trunc_bb_overflow, PM_TRUNC_BB_OVERFLOW),
758 	EF10_DMA_STAT(rx_pm_discard_bb_overflow, PM_DISCARD_BB_OVERFLOW),
759 	EF10_DMA_STAT(rx_pm_trunc_vfifo_full, PM_TRUNC_VFIFO_FULL),
760 	EF10_DMA_STAT(rx_pm_discard_vfifo_full, PM_DISCARD_VFIFO_FULL),
761 	EF10_DMA_STAT(rx_pm_trunc_qbb, PM_TRUNC_QBB),
762 	EF10_DMA_STAT(rx_pm_discard_qbb, PM_DISCARD_QBB),
763 	EF10_DMA_STAT(rx_pm_discard_mapping, PM_DISCARD_MAPPING),
764 	EF10_DMA_STAT(rx_dp_q_disabled_packets, RXDP_Q_DISABLED_PKTS),
765 	EF10_DMA_STAT(rx_dp_di_dropped_packets, RXDP_DI_DROPPED_PKTS),
766 	EF10_DMA_STAT(rx_dp_streaming_packets, RXDP_STREAMING_PKTS),
767 	EF10_DMA_STAT(rx_dp_emerg_fetch, RXDP_EMERGENCY_FETCH_CONDITIONS),
768 	EF10_DMA_STAT(rx_dp_emerg_wait, RXDP_EMERGENCY_WAIT_CONDITIONS),
769 };
770 
771 #define HUNT_COMMON_STAT_MASK ((1ULL << EF10_STAT_tx_bytes) |		\
772 			       (1ULL << EF10_STAT_tx_packets) |		\
773 			       (1ULL << EF10_STAT_tx_pause) |		\
774 			       (1ULL << EF10_STAT_tx_unicast) |		\
775 			       (1ULL << EF10_STAT_tx_multicast) |	\
776 			       (1ULL << EF10_STAT_tx_broadcast) |	\
777 			       (1ULL << EF10_STAT_rx_bytes) |		\
778 			       (1ULL << EF10_STAT_rx_bytes_minus_good_bytes) | \
779 			       (1ULL << EF10_STAT_rx_good_bytes) |	\
780 			       (1ULL << EF10_STAT_rx_bad_bytes) |	\
781 			       (1ULL << EF10_STAT_rx_packets) |		\
782 			       (1ULL << EF10_STAT_rx_good) |		\
783 			       (1ULL << EF10_STAT_rx_bad) |		\
784 			       (1ULL << EF10_STAT_rx_pause) |		\
785 			       (1ULL << EF10_STAT_rx_control) |		\
786 			       (1ULL << EF10_STAT_rx_unicast) |		\
787 			       (1ULL << EF10_STAT_rx_multicast) |	\
788 			       (1ULL << EF10_STAT_rx_broadcast) |	\
789 			       (1ULL << EF10_STAT_rx_lt64) |		\
790 			       (1ULL << EF10_STAT_rx_64) |		\
791 			       (1ULL << EF10_STAT_rx_65_to_127) |	\
792 			       (1ULL << EF10_STAT_rx_128_to_255) |	\
793 			       (1ULL << EF10_STAT_rx_256_to_511) |	\
794 			       (1ULL << EF10_STAT_rx_512_to_1023) |	\
795 			       (1ULL << EF10_STAT_rx_1024_to_15xx) |	\
796 			       (1ULL << EF10_STAT_rx_15xx_to_jumbo) |	\
797 			       (1ULL << EF10_STAT_rx_gtjumbo) |		\
798 			       (1ULL << EF10_STAT_rx_bad_gtjumbo) |	\
799 			       (1ULL << EF10_STAT_rx_overflow) |	\
800 			       (1ULL << EF10_STAT_rx_nodesc_drops))
801 
802 /* These statistics are only provided by the 10G MAC.  For a 10G/40G
803  * switchable port we do not expose these because they might not
804  * include all the packets they should.
805  */
806 #define HUNT_10G_ONLY_STAT_MASK ((1ULL << EF10_STAT_tx_control) |	\
807 				 (1ULL << EF10_STAT_tx_lt64) |		\
808 				 (1ULL << EF10_STAT_tx_64) |		\
809 				 (1ULL << EF10_STAT_tx_65_to_127) |	\
810 				 (1ULL << EF10_STAT_tx_128_to_255) |	\
811 				 (1ULL << EF10_STAT_tx_256_to_511) |	\
812 				 (1ULL << EF10_STAT_tx_512_to_1023) |	\
813 				 (1ULL << EF10_STAT_tx_1024_to_15xx) |	\
814 				 (1ULL << EF10_STAT_tx_15xx_to_jumbo))
815 
816 /* These statistics are only provided by the 40G MAC.  For a 10G/40G
817  * switchable port we do expose these because the errors will otherwise
818  * be silent.
819  */
820 #define HUNT_40G_EXTRA_STAT_MASK ((1ULL << EF10_STAT_rx_align_error) |	\
821 				  (1ULL << EF10_STAT_rx_length_error))
822 
823 /* These statistics are only provided if the firmware supports the
824  * capability PM_AND_RXDP_COUNTERS.
825  */
826 #define HUNT_PM_AND_RXDP_STAT_MASK (					\
827 	(1ULL << EF10_STAT_rx_pm_trunc_bb_overflow) |			\
828 	(1ULL << EF10_STAT_rx_pm_discard_bb_overflow) |			\
829 	(1ULL << EF10_STAT_rx_pm_trunc_vfifo_full) |			\
830 	(1ULL << EF10_STAT_rx_pm_discard_vfifo_full) |			\
831 	(1ULL << EF10_STAT_rx_pm_trunc_qbb) |				\
832 	(1ULL << EF10_STAT_rx_pm_discard_qbb) |				\
833 	(1ULL << EF10_STAT_rx_pm_discard_mapping) |			\
834 	(1ULL << EF10_STAT_rx_dp_q_disabled_packets) |			\
835 	(1ULL << EF10_STAT_rx_dp_di_dropped_packets) |			\
836 	(1ULL << EF10_STAT_rx_dp_streaming_packets) |			\
837 	(1ULL << EF10_STAT_rx_dp_emerg_fetch) |				\
838 	(1ULL << EF10_STAT_rx_dp_emerg_wait))
839 
840 static u64 efx_ef10_raw_stat_mask(struct efx_nic *efx)
841 {
842 	u64 raw_mask = HUNT_COMMON_STAT_MASK;
843 	u32 port_caps = efx_mcdi_phy_get_caps(efx);
844 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
845 
846 	if (port_caps & (1 << MC_CMD_PHY_CAP_40000FDX_LBN))
847 		raw_mask |= HUNT_40G_EXTRA_STAT_MASK;
848 	else
849 		raw_mask |= HUNT_10G_ONLY_STAT_MASK;
850 
851 	if (nic_data->datapath_caps &
852 	    (1 << MC_CMD_GET_CAPABILITIES_OUT_PM_AND_RXDP_COUNTERS_LBN))
853 		raw_mask |= HUNT_PM_AND_RXDP_STAT_MASK;
854 
855 	return raw_mask;
856 }
857 
858 static void efx_ef10_get_stat_mask(struct efx_nic *efx, unsigned long *mask)
859 {
860 	u64 raw_mask = efx_ef10_raw_stat_mask(efx);
861 
862 #if BITS_PER_LONG == 64
863 	mask[0] = raw_mask;
864 #else
865 	mask[0] = raw_mask & 0xffffffff;
866 	mask[1] = raw_mask >> 32;
867 #endif
868 }
869 
870 static size_t efx_ef10_describe_stats(struct efx_nic *efx, u8 *names)
871 {
872 	DECLARE_BITMAP(mask, EF10_STAT_COUNT);
873 
874 	efx_ef10_get_stat_mask(efx, mask);
875 	return efx_nic_describe_stats(efx_ef10_stat_desc, EF10_STAT_COUNT,
876 				      mask, names);
877 }
878 
879 static int efx_ef10_try_update_nic_stats(struct efx_nic *efx)
880 {
881 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
882 	DECLARE_BITMAP(mask, EF10_STAT_COUNT);
883 	__le64 generation_start, generation_end;
884 	u64 *stats = nic_data->stats;
885 	__le64 *dma_stats;
886 
887 	efx_ef10_get_stat_mask(efx, mask);
888 
889 	dma_stats = efx->stats_buffer.addr;
890 	nic_data = efx->nic_data;
891 
892 	generation_end = dma_stats[MC_CMD_MAC_GENERATION_END];
893 	if (generation_end == EFX_MC_STATS_GENERATION_INVALID)
894 		return 0;
895 	rmb();
896 	efx_nic_update_stats(efx_ef10_stat_desc, EF10_STAT_COUNT, mask,
897 			     stats, efx->stats_buffer.addr, false);
898 	rmb();
899 	generation_start = dma_stats[MC_CMD_MAC_GENERATION_START];
900 	if (generation_end != generation_start)
901 		return -EAGAIN;
902 
903 	/* Update derived statistics */
904 	stats[EF10_STAT_rx_good_bytes] =
905 		stats[EF10_STAT_rx_bytes] -
906 		stats[EF10_STAT_rx_bytes_minus_good_bytes];
907 	efx_update_diff_stat(&stats[EF10_STAT_rx_bad_bytes],
908 			     stats[EF10_STAT_rx_bytes_minus_good_bytes]);
909 
910 	return 0;
911 }
912 
913 
914 static size_t efx_ef10_update_stats(struct efx_nic *efx, u64 *full_stats,
915 				    struct rtnl_link_stats64 *core_stats)
916 {
917 	DECLARE_BITMAP(mask, EF10_STAT_COUNT);
918 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
919 	u64 *stats = nic_data->stats;
920 	size_t stats_count = 0, index;
921 	int retry;
922 
923 	efx_ef10_get_stat_mask(efx, mask);
924 
925 	/* If we're unlucky enough to read statistics during the DMA, wait
926 	 * up to 10ms for it to finish (typically takes <500us)
927 	 */
928 	for (retry = 0; retry < 100; ++retry) {
929 		if (efx_ef10_try_update_nic_stats(efx) == 0)
930 			break;
931 		udelay(100);
932 	}
933 
934 	if (full_stats) {
935 		for_each_set_bit(index, mask, EF10_STAT_COUNT) {
936 			if (efx_ef10_stat_desc[index].name) {
937 				*full_stats++ = stats[index];
938 				++stats_count;
939 			}
940 		}
941 	}
942 
943 	if (core_stats) {
944 		core_stats->rx_packets = stats[EF10_STAT_rx_packets];
945 		core_stats->tx_packets = stats[EF10_STAT_tx_packets];
946 		core_stats->rx_bytes = stats[EF10_STAT_rx_bytes];
947 		core_stats->tx_bytes = stats[EF10_STAT_tx_bytes];
948 		core_stats->rx_dropped = stats[EF10_STAT_rx_nodesc_drops];
949 		core_stats->multicast = stats[EF10_STAT_rx_multicast];
950 		core_stats->rx_length_errors =
951 			stats[EF10_STAT_rx_gtjumbo] +
952 			stats[EF10_STAT_rx_length_error];
953 		core_stats->rx_crc_errors = stats[EF10_STAT_rx_bad];
954 		core_stats->rx_frame_errors = stats[EF10_STAT_rx_align_error];
955 		core_stats->rx_fifo_errors = stats[EF10_STAT_rx_overflow];
956 		core_stats->rx_errors = (core_stats->rx_length_errors +
957 					 core_stats->rx_crc_errors +
958 					 core_stats->rx_frame_errors);
959 	}
960 
961 	return stats_count;
962 }
963 
964 static void efx_ef10_push_irq_moderation(struct efx_channel *channel)
965 {
966 	struct efx_nic *efx = channel->efx;
967 	unsigned int mode, value;
968 	efx_dword_t timer_cmd;
969 
970 	if (channel->irq_moderation) {
971 		mode = 3;
972 		value = channel->irq_moderation - 1;
973 	} else {
974 		mode = 0;
975 		value = 0;
976 	}
977 
978 	if (EFX_EF10_WORKAROUND_35388(efx)) {
979 		EFX_POPULATE_DWORD_3(timer_cmd, ERF_DD_EVQ_IND_TIMER_FLAGS,
980 				     EFE_DD_EVQ_IND_TIMER_FLAGS,
981 				     ERF_DD_EVQ_IND_TIMER_MODE, mode,
982 				     ERF_DD_EVQ_IND_TIMER_VAL, value);
983 		efx_writed_page(efx, &timer_cmd, ER_DD_EVQ_INDIRECT,
984 				channel->channel);
985 	} else {
986 		EFX_POPULATE_DWORD_2(timer_cmd, ERF_DZ_TC_TIMER_MODE, mode,
987 				     ERF_DZ_TC_TIMER_VAL, value);
988 		efx_writed_page(efx, &timer_cmd, ER_DZ_EVQ_TMR,
989 				channel->channel);
990 	}
991 }
992 
993 static void efx_ef10_get_wol(struct efx_nic *efx, struct ethtool_wolinfo *wol)
994 {
995 	wol->supported = 0;
996 	wol->wolopts = 0;
997 	memset(&wol->sopass, 0, sizeof(wol->sopass));
998 }
999 
1000 static int efx_ef10_set_wol(struct efx_nic *efx, u32 type)
1001 {
1002 	if (type != 0)
1003 		return -EINVAL;
1004 	return 0;
1005 }
1006 
1007 static void efx_ef10_mcdi_request(struct efx_nic *efx,
1008 				  const efx_dword_t *hdr, size_t hdr_len,
1009 				  const efx_dword_t *sdu, size_t sdu_len)
1010 {
1011 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1012 	u8 *pdu = nic_data->mcdi_buf.addr;
1013 
1014 	memcpy(pdu, hdr, hdr_len);
1015 	memcpy(pdu + hdr_len, sdu, sdu_len);
1016 	wmb();
1017 
1018 	/* The hardware provides 'low' and 'high' (doorbell) registers
1019 	 * for passing the 64-bit address of an MCDI request to
1020 	 * firmware.  However the dwords are swapped by firmware.  The
1021 	 * least significant bits of the doorbell are then 0 for all
1022 	 * MCDI requests due to alignment.
1023 	 */
1024 	_efx_writed(efx, cpu_to_le32((u64)nic_data->mcdi_buf.dma_addr >> 32),
1025 		    ER_DZ_MC_DB_LWRD);
1026 	_efx_writed(efx, cpu_to_le32((u32)nic_data->mcdi_buf.dma_addr),
1027 		    ER_DZ_MC_DB_HWRD);
1028 }
1029 
1030 static bool efx_ef10_mcdi_poll_response(struct efx_nic *efx)
1031 {
1032 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1033 	const efx_dword_t hdr = *(const efx_dword_t *)nic_data->mcdi_buf.addr;
1034 
1035 	rmb();
1036 	return EFX_DWORD_FIELD(hdr, MCDI_HEADER_RESPONSE);
1037 }
1038 
1039 static void
1040 efx_ef10_mcdi_read_response(struct efx_nic *efx, efx_dword_t *outbuf,
1041 			    size_t offset, size_t outlen)
1042 {
1043 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1044 	const u8 *pdu = nic_data->mcdi_buf.addr;
1045 
1046 	memcpy(outbuf, pdu + offset, outlen);
1047 }
1048 
1049 static int efx_ef10_mcdi_poll_reboot(struct efx_nic *efx)
1050 {
1051 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1052 	int rc;
1053 
1054 	rc = efx_ef10_get_warm_boot_count(efx);
1055 	if (rc < 0) {
1056 		/* The firmware is presumably in the process of
1057 		 * rebooting.  However, we are supposed to report each
1058 		 * reboot just once, so we must only do that once we
1059 		 * can read and store the updated warm boot count.
1060 		 */
1061 		return 0;
1062 	}
1063 
1064 	if (rc == nic_data->warm_boot_count)
1065 		return 0;
1066 
1067 	nic_data->warm_boot_count = rc;
1068 
1069 	/* All our allocations have been reset */
1070 	nic_data->must_realloc_vis = true;
1071 	nic_data->must_restore_filters = true;
1072 	nic_data->must_restore_piobufs = true;
1073 	nic_data->rx_rss_context = EFX_EF10_RSS_CONTEXT_INVALID;
1074 
1075 	/* The datapath firmware might have been changed */
1076 	nic_data->must_check_datapath_caps = true;
1077 
1078 	/* MAC statistics have been cleared on the NIC; clear the local
1079 	 * statistic that we update with efx_update_diff_stat().
1080 	 */
1081 	nic_data->stats[EF10_STAT_rx_bad_bytes] = 0;
1082 
1083 	return -EIO;
1084 }
1085 
1086 /* Handle an MSI interrupt
1087  *
1088  * Handle an MSI hardware interrupt.  This routine schedules event
1089  * queue processing.  No interrupt acknowledgement cycle is necessary.
1090  * Also, we never need to check that the interrupt is for us, since
1091  * MSI interrupts cannot be shared.
1092  */
1093 static irqreturn_t efx_ef10_msi_interrupt(int irq, void *dev_id)
1094 {
1095 	struct efx_msi_context *context = dev_id;
1096 	struct efx_nic *efx = context->efx;
1097 
1098 	netif_vdbg(efx, intr, efx->net_dev,
1099 		   "IRQ %d on CPU %d\n", irq, raw_smp_processor_id());
1100 
1101 	if (likely(ACCESS_ONCE(efx->irq_soft_enabled))) {
1102 		/* Note test interrupts */
1103 		if (context->index == efx->irq_level)
1104 			efx->last_irq_cpu = raw_smp_processor_id();
1105 
1106 		/* Schedule processing of the channel */
1107 		efx_schedule_channel_irq(efx->channel[context->index]);
1108 	}
1109 
1110 	return IRQ_HANDLED;
1111 }
1112 
1113 static irqreturn_t efx_ef10_legacy_interrupt(int irq, void *dev_id)
1114 {
1115 	struct efx_nic *efx = dev_id;
1116 	bool soft_enabled = ACCESS_ONCE(efx->irq_soft_enabled);
1117 	struct efx_channel *channel;
1118 	efx_dword_t reg;
1119 	u32 queues;
1120 
1121 	/* Read the ISR which also ACKs the interrupts */
1122 	efx_readd(efx, &reg, ER_DZ_BIU_INT_ISR);
1123 	queues = EFX_DWORD_FIELD(reg, ERF_DZ_ISR_REG);
1124 
1125 	if (queues == 0)
1126 		return IRQ_NONE;
1127 
1128 	if (likely(soft_enabled)) {
1129 		/* Note test interrupts */
1130 		if (queues & (1U << efx->irq_level))
1131 			efx->last_irq_cpu = raw_smp_processor_id();
1132 
1133 		efx_for_each_channel(channel, efx) {
1134 			if (queues & 1)
1135 				efx_schedule_channel_irq(channel);
1136 			queues >>= 1;
1137 		}
1138 	}
1139 
1140 	netif_vdbg(efx, intr, efx->net_dev,
1141 		   "IRQ %d on CPU %d status " EFX_DWORD_FMT "\n",
1142 		   irq, raw_smp_processor_id(), EFX_DWORD_VAL(reg));
1143 
1144 	return IRQ_HANDLED;
1145 }
1146 
1147 static void efx_ef10_irq_test_generate(struct efx_nic *efx)
1148 {
1149 	MCDI_DECLARE_BUF(inbuf, MC_CMD_TRIGGER_INTERRUPT_IN_LEN);
1150 
1151 	BUILD_BUG_ON(MC_CMD_TRIGGER_INTERRUPT_OUT_LEN != 0);
1152 
1153 	MCDI_SET_DWORD(inbuf, TRIGGER_INTERRUPT_IN_INTR_LEVEL, efx->irq_level);
1154 	(void) efx_mcdi_rpc(efx, MC_CMD_TRIGGER_INTERRUPT,
1155 			    inbuf, sizeof(inbuf), NULL, 0, NULL);
1156 }
1157 
1158 static int efx_ef10_tx_probe(struct efx_tx_queue *tx_queue)
1159 {
1160 	return efx_nic_alloc_buffer(tx_queue->efx, &tx_queue->txd.buf,
1161 				    (tx_queue->ptr_mask + 1) *
1162 				    sizeof(efx_qword_t),
1163 				    GFP_KERNEL);
1164 }
1165 
1166 /* This writes to the TX_DESC_WPTR and also pushes data */
1167 static inline void efx_ef10_push_tx_desc(struct efx_tx_queue *tx_queue,
1168 					 const efx_qword_t *txd)
1169 {
1170 	unsigned int write_ptr;
1171 	efx_oword_t reg;
1172 
1173 	write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
1174 	EFX_POPULATE_OWORD_1(reg, ERF_DZ_TX_DESC_WPTR, write_ptr);
1175 	reg.qword[0] = *txd;
1176 	efx_writeo_page(tx_queue->efx, &reg,
1177 			ER_DZ_TX_DESC_UPD, tx_queue->queue);
1178 }
1179 
1180 static void efx_ef10_tx_init(struct efx_tx_queue *tx_queue)
1181 {
1182 	MCDI_DECLARE_BUF(inbuf, MC_CMD_INIT_TXQ_IN_LEN(EFX_MAX_DMAQ_SIZE * 8 /
1183 						       EFX_BUF_SIZE));
1184 	MCDI_DECLARE_BUF(outbuf, MC_CMD_INIT_TXQ_OUT_LEN);
1185 	bool csum_offload = tx_queue->queue & EFX_TXQ_TYPE_OFFLOAD;
1186 	size_t entries = tx_queue->txd.buf.len / EFX_BUF_SIZE;
1187 	struct efx_channel *channel = tx_queue->channel;
1188 	struct efx_nic *efx = tx_queue->efx;
1189 	size_t inlen, outlen;
1190 	dma_addr_t dma_addr;
1191 	efx_qword_t *txd;
1192 	int rc;
1193 	int i;
1194 
1195 	MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_SIZE, tx_queue->ptr_mask + 1);
1196 	MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_TARGET_EVQ, channel->channel);
1197 	MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_LABEL, tx_queue->queue);
1198 	MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_INSTANCE, tx_queue->queue);
1199 	MCDI_POPULATE_DWORD_2(inbuf, INIT_TXQ_IN_FLAGS,
1200 			      INIT_TXQ_IN_FLAG_IP_CSUM_DIS, !csum_offload,
1201 			      INIT_TXQ_IN_FLAG_TCP_CSUM_DIS, !csum_offload);
1202 	MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_OWNER_ID, 0);
1203 	MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_PORT_ID, EVB_PORT_ID_ASSIGNED);
1204 
1205 	dma_addr = tx_queue->txd.buf.dma_addr;
1206 
1207 	netif_dbg(efx, hw, efx->net_dev, "pushing TXQ %d. %zu entries (%llx)\n",
1208 		  tx_queue->queue, entries, (u64)dma_addr);
1209 
1210 	for (i = 0; i < entries; ++i) {
1211 		MCDI_SET_ARRAY_QWORD(inbuf, INIT_TXQ_IN_DMA_ADDR, i, dma_addr);
1212 		dma_addr += EFX_BUF_SIZE;
1213 	}
1214 
1215 	inlen = MC_CMD_INIT_TXQ_IN_LEN(entries);
1216 
1217 	rc = efx_mcdi_rpc(efx, MC_CMD_INIT_TXQ, inbuf, inlen,
1218 			  outbuf, sizeof(outbuf), &outlen);
1219 	if (rc)
1220 		goto fail;
1221 
1222 	/* A previous user of this TX queue might have set us up the
1223 	 * bomb by writing a descriptor to the TX push collector but
1224 	 * not the doorbell.  (Each collector belongs to a port, not a
1225 	 * queue or function, so cannot easily be reset.)  We must
1226 	 * attempt to push a no-op descriptor in its place.
1227 	 */
1228 	tx_queue->buffer[0].flags = EFX_TX_BUF_OPTION;
1229 	tx_queue->insert_count = 1;
1230 	txd = efx_tx_desc(tx_queue, 0);
1231 	EFX_POPULATE_QWORD_4(*txd,
1232 			     ESF_DZ_TX_DESC_IS_OPT, true,
1233 			     ESF_DZ_TX_OPTION_TYPE,
1234 			     ESE_DZ_TX_OPTION_DESC_CRC_CSUM,
1235 			     ESF_DZ_TX_OPTION_UDP_TCP_CSUM, csum_offload,
1236 			     ESF_DZ_TX_OPTION_IP_CSUM, csum_offload);
1237 	tx_queue->write_count = 1;
1238 	wmb();
1239 	efx_ef10_push_tx_desc(tx_queue, txd);
1240 
1241 	return;
1242 
1243 fail:
1244 	WARN_ON(true);
1245 	netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
1246 }
1247 
1248 static void efx_ef10_tx_fini(struct efx_tx_queue *tx_queue)
1249 {
1250 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FINI_TXQ_IN_LEN);
1251 	MCDI_DECLARE_BUF(outbuf, MC_CMD_FINI_TXQ_OUT_LEN);
1252 	struct efx_nic *efx = tx_queue->efx;
1253 	size_t outlen;
1254 	int rc;
1255 
1256 	MCDI_SET_DWORD(inbuf, FINI_TXQ_IN_INSTANCE,
1257 		       tx_queue->queue);
1258 
1259 	rc = efx_mcdi_rpc(efx, MC_CMD_FINI_TXQ, inbuf, sizeof(inbuf),
1260 			  outbuf, sizeof(outbuf), &outlen);
1261 
1262 	if (rc && rc != -EALREADY)
1263 		goto fail;
1264 
1265 	return;
1266 
1267 fail:
1268 	netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
1269 }
1270 
1271 static void efx_ef10_tx_remove(struct efx_tx_queue *tx_queue)
1272 {
1273 	efx_nic_free_buffer(tx_queue->efx, &tx_queue->txd.buf);
1274 }
1275 
1276 /* This writes to the TX_DESC_WPTR; write pointer for TX descriptor ring */
1277 static inline void efx_ef10_notify_tx_desc(struct efx_tx_queue *tx_queue)
1278 {
1279 	unsigned int write_ptr;
1280 	efx_dword_t reg;
1281 
1282 	write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
1283 	EFX_POPULATE_DWORD_1(reg, ERF_DZ_TX_DESC_WPTR_DWORD, write_ptr);
1284 	efx_writed_page(tx_queue->efx, &reg,
1285 			ER_DZ_TX_DESC_UPD_DWORD, tx_queue->queue);
1286 }
1287 
1288 static void efx_ef10_tx_write(struct efx_tx_queue *tx_queue)
1289 {
1290 	unsigned int old_write_count = tx_queue->write_count;
1291 	struct efx_tx_buffer *buffer;
1292 	unsigned int write_ptr;
1293 	efx_qword_t *txd;
1294 
1295 	BUG_ON(tx_queue->write_count == tx_queue->insert_count);
1296 
1297 	do {
1298 		write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
1299 		buffer = &tx_queue->buffer[write_ptr];
1300 		txd = efx_tx_desc(tx_queue, write_ptr);
1301 		++tx_queue->write_count;
1302 
1303 		/* Create TX descriptor ring entry */
1304 		if (buffer->flags & EFX_TX_BUF_OPTION) {
1305 			*txd = buffer->option;
1306 		} else {
1307 			BUILD_BUG_ON(EFX_TX_BUF_CONT != 1);
1308 			EFX_POPULATE_QWORD_3(
1309 				*txd,
1310 				ESF_DZ_TX_KER_CONT,
1311 				buffer->flags & EFX_TX_BUF_CONT,
1312 				ESF_DZ_TX_KER_BYTE_CNT, buffer->len,
1313 				ESF_DZ_TX_KER_BUF_ADDR, buffer->dma_addr);
1314 		}
1315 	} while (tx_queue->write_count != tx_queue->insert_count);
1316 
1317 	wmb(); /* Ensure descriptors are written before they are fetched */
1318 
1319 	if (efx_nic_may_push_tx_desc(tx_queue, old_write_count)) {
1320 		txd = efx_tx_desc(tx_queue,
1321 				  old_write_count & tx_queue->ptr_mask);
1322 		efx_ef10_push_tx_desc(tx_queue, txd);
1323 		++tx_queue->pushes;
1324 	} else {
1325 		efx_ef10_notify_tx_desc(tx_queue);
1326 	}
1327 }
1328 
1329 static int efx_ef10_alloc_rss_context(struct efx_nic *efx, u32 *context)
1330 {
1331 	MCDI_DECLARE_BUF(inbuf, MC_CMD_RSS_CONTEXT_ALLOC_IN_LEN);
1332 	MCDI_DECLARE_BUF(outbuf, MC_CMD_RSS_CONTEXT_ALLOC_OUT_LEN);
1333 	size_t outlen;
1334 	int rc;
1335 
1336 	MCDI_SET_DWORD(inbuf, RSS_CONTEXT_ALLOC_IN_UPSTREAM_PORT_ID,
1337 		       EVB_PORT_ID_ASSIGNED);
1338 	MCDI_SET_DWORD(inbuf, RSS_CONTEXT_ALLOC_IN_TYPE,
1339 		       MC_CMD_RSS_CONTEXT_ALLOC_IN_TYPE_EXCLUSIVE);
1340 	MCDI_SET_DWORD(inbuf, RSS_CONTEXT_ALLOC_IN_NUM_QUEUES,
1341 		       EFX_MAX_CHANNELS);
1342 
1343 	rc = efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_ALLOC, inbuf, sizeof(inbuf),
1344 		outbuf, sizeof(outbuf), &outlen);
1345 	if (rc != 0)
1346 		return rc;
1347 
1348 	if (outlen < MC_CMD_RSS_CONTEXT_ALLOC_OUT_LEN)
1349 		return -EIO;
1350 
1351 	*context = MCDI_DWORD(outbuf, RSS_CONTEXT_ALLOC_OUT_RSS_CONTEXT_ID);
1352 
1353 	return 0;
1354 }
1355 
1356 static void efx_ef10_free_rss_context(struct efx_nic *efx, u32 context)
1357 {
1358 	MCDI_DECLARE_BUF(inbuf, MC_CMD_RSS_CONTEXT_FREE_IN_LEN);
1359 	int rc;
1360 
1361 	MCDI_SET_DWORD(inbuf, RSS_CONTEXT_FREE_IN_RSS_CONTEXT_ID,
1362 		       context);
1363 
1364 	rc = efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_FREE, inbuf, sizeof(inbuf),
1365 			    NULL, 0, NULL);
1366 	WARN_ON(rc != 0);
1367 }
1368 
1369 static int efx_ef10_populate_rss_table(struct efx_nic *efx, u32 context)
1370 {
1371 	MCDI_DECLARE_BUF(tablebuf, MC_CMD_RSS_CONTEXT_SET_TABLE_IN_LEN);
1372 	MCDI_DECLARE_BUF(keybuf, MC_CMD_RSS_CONTEXT_SET_KEY_IN_LEN);
1373 	int i, rc;
1374 
1375 	MCDI_SET_DWORD(tablebuf, RSS_CONTEXT_SET_TABLE_IN_RSS_CONTEXT_ID,
1376 		       context);
1377 	BUILD_BUG_ON(ARRAY_SIZE(efx->rx_indir_table) !=
1378 		     MC_CMD_RSS_CONTEXT_SET_TABLE_IN_INDIRECTION_TABLE_LEN);
1379 
1380 	for (i = 0; i < ARRAY_SIZE(efx->rx_indir_table); ++i)
1381 		MCDI_PTR(tablebuf,
1382 			 RSS_CONTEXT_SET_TABLE_IN_INDIRECTION_TABLE)[i] =
1383 				(u8) efx->rx_indir_table[i];
1384 
1385 	rc = efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_SET_TABLE, tablebuf,
1386 			  sizeof(tablebuf), NULL, 0, NULL);
1387 	if (rc != 0)
1388 		return rc;
1389 
1390 	MCDI_SET_DWORD(keybuf, RSS_CONTEXT_SET_KEY_IN_RSS_CONTEXT_ID,
1391 		       context);
1392 	BUILD_BUG_ON(ARRAY_SIZE(efx->rx_hash_key) !=
1393 		     MC_CMD_RSS_CONTEXT_SET_KEY_IN_TOEPLITZ_KEY_LEN);
1394 	for (i = 0; i < ARRAY_SIZE(efx->rx_hash_key); ++i)
1395 		MCDI_PTR(keybuf, RSS_CONTEXT_SET_KEY_IN_TOEPLITZ_KEY)[i] =
1396 			efx->rx_hash_key[i];
1397 
1398 	return efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_SET_KEY, keybuf,
1399 			    sizeof(keybuf), NULL, 0, NULL);
1400 }
1401 
1402 static void efx_ef10_rx_free_indir_table(struct efx_nic *efx)
1403 {
1404 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1405 
1406 	if (nic_data->rx_rss_context != EFX_EF10_RSS_CONTEXT_INVALID)
1407 		efx_ef10_free_rss_context(efx, nic_data->rx_rss_context);
1408 	nic_data->rx_rss_context = EFX_EF10_RSS_CONTEXT_INVALID;
1409 }
1410 
1411 static void efx_ef10_rx_push_indir_table(struct efx_nic *efx)
1412 {
1413 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1414 	int rc;
1415 
1416 	netif_dbg(efx, drv, efx->net_dev, "pushing RX indirection table\n");
1417 
1418 	if (nic_data->rx_rss_context == EFX_EF10_RSS_CONTEXT_INVALID) {
1419 		rc = efx_ef10_alloc_rss_context(efx, &nic_data->rx_rss_context);
1420 		if (rc != 0)
1421 			goto fail;
1422 	}
1423 
1424 	rc = efx_ef10_populate_rss_table(efx, nic_data->rx_rss_context);
1425 	if (rc != 0)
1426 		goto fail;
1427 
1428 	return;
1429 
1430 fail:
1431 	netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
1432 }
1433 
1434 static int efx_ef10_rx_probe(struct efx_rx_queue *rx_queue)
1435 {
1436 	return efx_nic_alloc_buffer(rx_queue->efx, &rx_queue->rxd.buf,
1437 				    (rx_queue->ptr_mask + 1) *
1438 				    sizeof(efx_qword_t),
1439 				    GFP_KERNEL);
1440 }
1441 
1442 static void efx_ef10_rx_init(struct efx_rx_queue *rx_queue)
1443 {
1444 	MCDI_DECLARE_BUF(inbuf,
1445 			 MC_CMD_INIT_RXQ_IN_LEN(EFX_MAX_DMAQ_SIZE * 8 /
1446 						EFX_BUF_SIZE));
1447 	MCDI_DECLARE_BUF(outbuf, MC_CMD_INIT_RXQ_OUT_LEN);
1448 	struct efx_channel *channel = efx_rx_queue_channel(rx_queue);
1449 	size_t entries = rx_queue->rxd.buf.len / EFX_BUF_SIZE;
1450 	struct efx_nic *efx = rx_queue->efx;
1451 	size_t inlen, outlen;
1452 	dma_addr_t dma_addr;
1453 	int rc;
1454 	int i;
1455 
1456 	rx_queue->scatter_n = 0;
1457 	rx_queue->scatter_len = 0;
1458 
1459 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_SIZE, rx_queue->ptr_mask + 1);
1460 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_TARGET_EVQ, channel->channel);
1461 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_LABEL, efx_rx_queue_index(rx_queue));
1462 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_INSTANCE,
1463 		       efx_rx_queue_index(rx_queue));
1464 	MCDI_POPULATE_DWORD_1(inbuf, INIT_RXQ_IN_FLAGS,
1465 			      INIT_RXQ_IN_FLAG_PREFIX, 1);
1466 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_OWNER_ID, 0);
1467 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_PORT_ID, EVB_PORT_ID_ASSIGNED);
1468 
1469 	dma_addr = rx_queue->rxd.buf.dma_addr;
1470 
1471 	netif_dbg(efx, hw, efx->net_dev, "pushing RXQ %d. %zu entries (%llx)\n",
1472 		  efx_rx_queue_index(rx_queue), entries, (u64)dma_addr);
1473 
1474 	for (i = 0; i < entries; ++i) {
1475 		MCDI_SET_ARRAY_QWORD(inbuf, INIT_RXQ_IN_DMA_ADDR, i, dma_addr);
1476 		dma_addr += EFX_BUF_SIZE;
1477 	}
1478 
1479 	inlen = MC_CMD_INIT_RXQ_IN_LEN(entries);
1480 
1481 	rc = efx_mcdi_rpc(efx, MC_CMD_INIT_RXQ, inbuf, inlen,
1482 			  outbuf, sizeof(outbuf), &outlen);
1483 	if (rc)
1484 		goto fail;
1485 
1486 	return;
1487 
1488 fail:
1489 	WARN_ON(true);
1490 	netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
1491 }
1492 
1493 static void efx_ef10_rx_fini(struct efx_rx_queue *rx_queue)
1494 {
1495 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FINI_RXQ_IN_LEN);
1496 	MCDI_DECLARE_BUF(outbuf, MC_CMD_FINI_RXQ_OUT_LEN);
1497 	struct efx_nic *efx = rx_queue->efx;
1498 	size_t outlen;
1499 	int rc;
1500 
1501 	MCDI_SET_DWORD(inbuf, FINI_RXQ_IN_INSTANCE,
1502 		       efx_rx_queue_index(rx_queue));
1503 
1504 	rc = efx_mcdi_rpc(efx, MC_CMD_FINI_RXQ, inbuf, sizeof(inbuf),
1505 			  outbuf, sizeof(outbuf), &outlen);
1506 
1507 	if (rc && rc != -EALREADY)
1508 		goto fail;
1509 
1510 	return;
1511 
1512 fail:
1513 	netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
1514 }
1515 
1516 static void efx_ef10_rx_remove(struct efx_rx_queue *rx_queue)
1517 {
1518 	efx_nic_free_buffer(rx_queue->efx, &rx_queue->rxd.buf);
1519 }
1520 
1521 /* This creates an entry in the RX descriptor queue */
1522 static inline void
1523 efx_ef10_build_rx_desc(struct efx_rx_queue *rx_queue, unsigned int index)
1524 {
1525 	struct efx_rx_buffer *rx_buf;
1526 	efx_qword_t *rxd;
1527 
1528 	rxd = efx_rx_desc(rx_queue, index);
1529 	rx_buf = efx_rx_buffer(rx_queue, index);
1530 	EFX_POPULATE_QWORD_2(*rxd,
1531 			     ESF_DZ_RX_KER_BYTE_CNT, rx_buf->len,
1532 			     ESF_DZ_RX_KER_BUF_ADDR, rx_buf->dma_addr);
1533 }
1534 
1535 static void efx_ef10_rx_write(struct efx_rx_queue *rx_queue)
1536 {
1537 	struct efx_nic *efx = rx_queue->efx;
1538 	unsigned int write_count;
1539 	efx_dword_t reg;
1540 
1541 	/* Firmware requires that RX_DESC_WPTR be a multiple of 8 */
1542 	write_count = rx_queue->added_count & ~7;
1543 	if (rx_queue->notified_count == write_count)
1544 		return;
1545 
1546 	do
1547 		efx_ef10_build_rx_desc(
1548 			rx_queue,
1549 			rx_queue->notified_count & rx_queue->ptr_mask);
1550 	while (++rx_queue->notified_count != write_count);
1551 
1552 	wmb();
1553 	EFX_POPULATE_DWORD_1(reg, ERF_DZ_RX_DESC_WPTR,
1554 			     write_count & rx_queue->ptr_mask);
1555 	efx_writed_page(efx, &reg, ER_DZ_RX_DESC_UPD,
1556 			efx_rx_queue_index(rx_queue));
1557 }
1558 
1559 static efx_mcdi_async_completer efx_ef10_rx_defer_refill_complete;
1560 
1561 static void efx_ef10_rx_defer_refill(struct efx_rx_queue *rx_queue)
1562 {
1563 	struct efx_channel *channel = efx_rx_queue_channel(rx_queue);
1564 	MCDI_DECLARE_BUF(inbuf, MC_CMD_DRIVER_EVENT_IN_LEN);
1565 	efx_qword_t event;
1566 
1567 	EFX_POPULATE_QWORD_2(event,
1568 			     ESF_DZ_EV_CODE, EFX_EF10_DRVGEN_EV,
1569 			     ESF_DZ_EV_DATA, EFX_EF10_REFILL);
1570 
1571 	MCDI_SET_DWORD(inbuf, DRIVER_EVENT_IN_EVQ, channel->channel);
1572 
1573 	/* MCDI_SET_QWORD is not appropriate here since EFX_POPULATE_* has
1574 	 * already swapped the data to little-endian order.
1575 	 */
1576 	memcpy(MCDI_PTR(inbuf, DRIVER_EVENT_IN_DATA), &event.u64[0],
1577 	       sizeof(efx_qword_t));
1578 
1579 	efx_mcdi_rpc_async(channel->efx, MC_CMD_DRIVER_EVENT,
1580 			   inbuf, sizeof(inbuf), 0,
1581 			   efx_ef10_rx_defer_refill_complete, 0);
1582 }
1583 
1584 static void
1585 efx_ef10_rx_defer_refill_complete(struct efx_nic *efx, unsigned long cookie,
1586 				  int rc, efx_dword_t *outbuf,
1587 				  size_t outlen_actual)
1588 {
1589 	/* nothing to do */
1590 }
1591 
1592 static int efx_ef10_ev_probe(struct efx_channel *channel)
1593 {
1594 	return efx_nic_alloc_buffer(channel->efx, &channel->eventq.buf,
1595 				    (channel->eventq_mask + 1) *
1596 				    sizeof(efx_qword_t),
1597 				    GFP_KERNEL);
1598 }
1599 
1600 static int efx_ef10_ev_init(struct efx_channel *channel)
1601 {
1602 	MCDI_DECLARE_BUF(inbuf,
1603 			 MC_CMD_INIT_EVQ_IN_LEN(EFX_MAX_EVQ_SIZE * 8 /
1604 						EFX_BUF_SIZE));
1605 	MCDI_DECLARE_BUF(outbuf, MC_CMD_INIT_EVQ_OUT_LEN);
1606 	size_t entries = channel->eventq.buf.len / EFX_BUF_SIZE;
1607 	struct efx_nic *efx = channel->efx;
1608 	struct efx_ef10_nic_data *nic_data;
1609 	bool supports_rx_merge;
1610 	size_t inlen, outlen;
1611 	dma_addr_t dma_addr;
1612 	int rc;
1613 	int i;
1614 
1615 	nic_data = efx->nic_data;
1616 	supports_rx_merge =
1617 		!!(nic_data->datapath_caps &
1618 		   1 << MC_CMD_GET_CAPABILITIES_OUT_RX_BATCHING_LBN);
1619 
1620 	/* Fill event queue with all ones (i.e. empty events) */
1621 	memset(channel->eventq.buf.addr, 0xff, channel->eventq.buf.len);
1622 
1623 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_SIZE, channel->eventq_mask + 1);
1624 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_INSTANCE, channel->channel);
1625 	/* INIT_EVQ expects index in vector table, not absolute */
1626 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_IRQ_NUM, channel->channel);
1627 	MCDI_POPULATE_DWORD_4(inbuf, INIT_EVQ_IN_FLAGS,
1628 			      INIT_EVQ_IN_FLAG_INTERRUPTING, 1,
1629 			      INIT_EVQ_IN_FLAG_RX_MERGE, 1,
1630 			      INIT_EVQ_IN_FLAG_TX_MERGE, 1,
1631 			      INIT_EVQ_IN_FLAG_CUT_THRU, !supports_rx_merge);
1632 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_TMR_MODE,
1633 		       MC_CMD_INIT_EVQ_IN_TMR_MODE_DIS);
1634 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_TMR_LOAD, 0);
1635 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_TMR_RELOAD, 0);
1636 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_COUNT_MODE,
1637 		       MC_CMD_INIT_EVQ_IN_COUNT_MODE_DIS);
1638 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_COUNT_THRSHLD, 0);
1639 
1640 	dma_addr = channel->eventq.buf.dma_addr;
1641 	for (i = 0; i < entries; ++i) {
1642 		MCDI_SET_ARRAY_QWORD(inbuf, INIT_EVQ_IN_DMA_ADDR, i, dma_addr);
1643 		dma_addr += EFX_BUF_SIZE;
1644 	}
1645 
1646 	inlen = MC_CMD_INIT_EVQ_IN_LEN(entries);
1647 
1648 	rc = efx_mcdi_rpc(efx, MC_CMD_INIT_EVQ, inbuf, inlen,
1649 			  outbuf, sizeof(outbuf), &outlen);
1650 	if (rc)
1651 		goto fail;
1652 
1653 	/* IRQ return is ignored */
1654 
1655 	return 0;
1656 
1657 fail:
1658 	netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
1659 	return rc;
1660 }
1661 
1662 static void efx_ef10_ev_fini(struct efx_channel *channel)
1663 {
1664 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FINI_EVQ_IN_LEN);
1665 	MCDI_DECLARE_BUF(outbuf, MC_CMD_FINI_EVQ_OUT_LEN);
1666 	struct efx_nic *efx = channel->efx;
1667 	size_t outlen;
1668 	int rc;
1669 
1670 	MCDI_SET_DWORD(inbuf, FINI_EVQ_IN_INSTANCE, channel->channel);
1671 
1672 	rc = efx_mcdi_rpc(efx, MC_CMD_FINI_EVQ, inbuf, sizeof(inbuf),
1673 			  outbuf, sizeof(outbuf), &outlen);
1674 
1675 	if (rc && rc != -EALREADY)
1676 		goto fail;
1677 
1678 	return;
1679 
1680 fail:
1681 	netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
1682 }
1683 
1684 static void efx_ef10_ev_remove(struct efx_channel *channel)
1685 {
1686 	efx_nic_free_buffer(channel->efx, &channel->eventq.buf);
1687 }
1688 
1689 static void efx_ef10_handle_rx_wrong_queue(struct efx_rx_queue *rx_queue,
1690 					   unsigned int rx_queue_label)
1691 {
1692 	struct efx_nic *efx = rx_queue->efx;
1693 
1694 	netif_info(efx, hw, efx->net_dev,
1695 		   "rx event arrived on queue %d labeled as queue %u\n",
1696 		   efx_rx_queue_index(rx_queue), rx_queue_label);
1697 
1698 	efx_schedule_reset(efx, RESET_TYPE_DISABLE);
1699 }
1700 
1701 static void
1702 efx_ef10_handle_rx_bad_lbits(struct efx_rx_queue *rx_queue,
1703 			     unsigned int actual, unsigned int expected)
1704 {
1705 	unsigned int dropped = (actual - expected) & rx_queue->ptr_mask;
1706 	struct efx_nic *efx = rx_queue->efx;
1707 
1708 	netif_info(efx, hw, efx->net_dev,
1709 		   "dropped %d events (index=%d expected=%d)\n",
1710 		   dropped, actual, expected);
1711 
1712 	efx_schedule_reset(efx, RESET_TYPE_DISABLE);
1713 }
1714 
1715 /* partially received RX was aborted. clean up. */
1716 static void efx_ef10_handle_rx_abort(struct efx_rx_queue *rx_queue)
1717 {
1718 	unsigned int rx_desc_ptr;
1719 
1720 	WARN_ON(rx_queue->scatter_n == 0);
1721 
1722 	netif_dbg(rx_queue->efx, hw, rx_queue->efx->net_dev,
1723 		  "scattered RX aborted (dropping %u buffers)\n",
1724 		  rx_queue->scatter_n);
1725 
1726 	rx_desc_ptr = rx_queue->removed_count & rx_queue->ptr_mask;
1727 
1728 	efx_rx_packet(rx_queue, rx_desc_ptr, rx_queue->scatter_n,
1729 		      0, EFX_RX_PKT_DISCARD);
1730 
1731 	rx_queue->removed_count += rx_queue->scatter_n;
1732 	rx_queue->scatter_n = 0;
1733 	rx_queue->scatter_len = 0;
1734 	++efx_rx_queue_channel(rx_queue)->n_rx_nodesc_trunc;
1735 }
1736 
1737 static int efx_ef10_handle_rx_event(struct efx_channel *channel,
1738 				    const efx_qword_t *event)
1739 {
1740 	unsigned int rx_bytes, next_ptr_lbits, rx_queue_label, rx_l4_class;
1741 	unsigned int n_descs, n_packets, i;
1742 	struct efx_nic *efx = channel->efx;
1743 	struct efx_rx_queue *rx_queue;
1744 	bool rx_cont;
1745 	u16 flags = 0;
1746 
1747 	if (unlikely(ACCESS_ONCE(efx->reset_pending)))
1748 		return 0;
1749 
1750 	/* Basic packet information */
1751 	rx_bytes = EFX_QWORD_FIELD(*event, ESF_DZ_RX_BYTES);
1752 	next_ptr_lbits = EFX_QWORD_FIELD(*event, ESF_DZ_RX_DSC_PTR_LBITS);
1753 	rx_queue_label = EFX_QWORD_FIELD(*event, ESF_DZ_RX_QLABEL);
1754 	rx_l4_class = EFX_QWORD_FIELD(*event, ESF_DZ_RX_L4_CLASS);
1755 	rx_cont = EFX_QWORD_FIELD(*event, ESF_DZ_RX_CONT);
1756 
1757 	WARN_ON(EFX_QWORD_FIELD(*event, ESF_DZ_RX_DROP_EVENT));
1758 
1759 	rx_queue = efx_channel_get_rx_queue(channel);
1760 
1761 	if (unlikely(rx_queue_label != efx_rx_queue_index(rx_queue)))
1762 		efx_ef10_handle_rx_wrong_queue(rx_queue, rx_queue_label);
1763 
1764 	n_descs = ((next_ptr_lbits - rx_queue->removed_count) &
1765 		   ((1 << ESF_DZ_RX_DSC_PTR_LBITS_WIDTH) - 1));
1766 
1767 	if (n_descs != rx_queue->scatter_n + 1) {
1768 		/* detect rx abort */
1769 		if (unlikely(n_descs == rx_queue->scatter_n)) {
1770 			WARN_ON(rx_bytes != 0);
1771 			efx_ef10_handle_rx_abort(rx_queue);
1772 			return 0;
1773 		}
1774 
1775 		if (unlikely(rx_queue->scatter_n != 0)) {
1776 			/* Scattered packet completions cannot be
1777 			 * merged, so something has gone wrong.
1778 			 */
1779 			efx_ef10_handle_rx_bad_lbits(
1780 				rx_queue, next_ptr_lbits,
1781 				(rx_queue->removed_count +
1782 				 rx_queue->scatter_n + 1) &
1783 				((1 << ESF_DZ_RX_DSC_PTR_LBITS_WIDTH) - 1));
1784 			return 0;
1785 		}
1786 
1787 		/* Merged completion for multiple non-scattered packets */
1788 		rx_queue->scatter_n = 1;
1789 		rx_queue->scatter_len = 0;
1790 		n_packets = n_descs;
1791 		++channel->n_rx_merge_events;
1792 		channel->n_rx_merge_packets += n_packets;
1793 		flags |= EFX_RX_PKT_PREFIX_LEN;
1794 	} else {
1795 		++rx_queue->scatter_n;
1796 		rx_queue->scatter_len += rx_bytes;
1797 		if (rx_cont)
1798 			return 0;
1799 		n_packets = 1;
1800 	}
1801 
1802 	if (unlikely(EFX_QWORD_FIELD(*event, ESF_DZ_RX_ECRC_ERR)))
1803 		flags |= EFX_RX_PKT_DISCARD;
1804 
1805 	if (unlikely(EFX_QWORD_FIELD(*event, ESF_DZ_RX_IPCKSUM_ERR))) {
1806 		channel->n_rx_ip_hdr_chksum_err += n_packets;
1807 	} else if (unlikely(EFX_QWORD_FIELD(*event,
1808 					    ESF_DZ_RX_TCPUDP_CKSUM_ERR))) {
1809 		channel->n_rx_tcp_udp_chksum_err += n_packets;
1810 	} else if (rx_l4_class == ESE_DZ_L4_CLASS_TCP ||
1811 		   rx_l4_class == ESE_DZ_L4_CLASS_UDP) {
1812 		flags |= EFX_RX_PKT_CSUMMED;
1813 	}
1814 
1815 	if (rx_l4_class == ESE_DZ_L4_CLASS_TCP)
1816 		flags |= EFX_RX_PKT_TCP;
1817 
1818 	channel->irq_mod_score += 2 * n_packets;
1819 
1820 	/* Handle received packet(s) */
1821 	for (i = 0; i < n_packets; i++) {
1822 		efx_rx_packet(rx_queue,
1823 			      rx_queue->removed_count & rx_queue->ptr_mask,
1824 			      rx_queue->scatter_n, rx_queue->scatter_len,
1825 			      flags);
1826 		rx_queue->removed_count += rx_queue->scatter_n;
1827 	}
1828 
1829 	rx_queue->scatter_n = 0;
1830 	rx_queue->scatter_len = 0;
1831 
1832 	return n_packets;
1833 }
1834 
1835 static int
1836 efx_ef10_handle_tx_event(struct efx_channel *channel, efx_qword_t *event)
1837 {
1838 	struct efx_nic *efx = channel->efx;
1839 	struct efx_tx_queue *tx_queue;
1840 	unsigned int tx_ev_desc_ptr;
1841 	unsigned int tx_ev_q_label;
1842 	int tx_descs = 0;
1843 
1844 	if (unlikely(ACCESS_ONCE(efx->reset_pending)))
1845 		return 0;
1846 
1847 	if (unlikely(EFX_QWORD_FIELD(*event, ESF_DZ_TX_DROP_EVENT)))
1848 		return 0;
1849 
1850 	/* Transmit completion */
1851 	tx_ev_desc_ptr = EFX_QWORD_FIELD(*event, ESF_DZ_TX_DESCR_INDX);
1852 	tx_ev_q_label = EFX_QWORD_FIELD(*event, ESF_DZ_TX_QLABEL);
1853 	tx_queue = efx_channel_get_tx_queue(channel,
1854 					    tx_ev_q_label % EFX_TXQ_TYPES);
1855 	tx_descs = ((tx_ev_desc_ptr + 1 - tx_queue->read_count) &
1856 		    tx_queue->ptr_mask);
1857 	efx_xmit_done(tx_queue, tx_ev_desc_ptr & tx_queue->ptr_mask);
1858 
1859 	return tx_descs;
1860 }
1861 
1862 static void
1863 efx_ef10_handle_driver_event(struct efx_channel *channel, efx_qword_t *event)
1864 {
1865 	struct efx_nic *efx = channel->efx;
1866 	int subcode;
1867 
1868 	subcode = EFX_QWORD_FIELD(*event, ESF_DZ_DRV_SUB_CODE);
1869 
1870 	switch (subcode) {
1871 	case ESE_DZ_DRV_TIMER_EV:
1872 	case ESE_DZ_DRV_WAKE_UP_EV:
1873 		break;
1874 	case ESE_DZ_DRV_START_UP_EV:
1875 		/* event queue init complete. ok. */
1876 		break;
1877 	default:
1878 		netif_err(efx, hw, efx->net_dev,
1879 			  "channel %d unknown driver event type %d"
1880 			  " (data " EFX_QWORD_FMT ")\n",
1881 			  channel->channel, subcode,
1882 			  EFX_QWORD_VAL(*event));
1883 
1884 	}
1885 }
1886 
1887 static void efx_ef10_handle_driver_generated_event(struct efx_channel *channel,
1888 						   efx_qword_t *event)
1889 {
1890 	struct efx_nic *efx = channel->efx;
1891 	u32 subcode;
1892 
1893 	subcode = EFX_QWORD_FIELD(*event, EFX_DWORD_0);
1894 
1895 	switch (subcode) {
1896 	case EFX_EF10_TEST:
1897 		channel->event_test_cpu = raw_smp_processor_id();
1898 		break;
1899 	case EFX_EF10_REFILL:
1900 		/* The queue must be empty, so we won't receive any rx
1901 		 * events, so efx_process_channel() won't refill the
1902 		 * queue. Refill it here
1903 		 */
1904 		efx_fast_push_rx_descriptors(&channel->rx_queue);
1905 		break;
1906 	default:
1907 		netif_err(efx, hw, efx->net_dev,
1908 			  "channel %d unknown driver event type %u"
1909 			  " (data " EFX_QWORD_FMT ")\n",
1910 			  channel->channel, (unsigned) subcode,
1911 			  EFX_QWORD_VAL(*event));
1912 	}
1913 }
1914 
1915 static int efx_ef10_ev_process(struct efx_channel *channel, int quota)
1916 {
1917 	struct efx_nic *efx = channel->efx;
1918 	efx_qword_t event, *p_event;
1919 	unsigned int read_ptr;
1920 	int ev_code;
1921 	int tx_descs = 0;
1922 	int spent = 0;
1923 
1924 	read_ptr = channel->eventq_read_ptr;
1925 
1926 	for (;;) {
1927 		p_event = efx_event(channel, read_ptr);
1928 		event = *p_event;
1929 
1930 		if (!efx_event_present(&event))
1931 			break;
1932 
1933 		EFX_SET_QWORD(*p_event);
1934 
1935 		++read_ptr;
1936 
1937 		ev_code = EFX_QWORD_FIELD(event, ESF_DZ_EV_CODE);
1938 
1939 		netif_vdbg(efx, drv, efx->net_dev,
1940 			   "processing event on %d " EFX_QWORD_FMT "\n",
1941 			   channel->channel, EFX_QWORD_VAL(event));
1942 
1943 		switch (ev_code) {
1944 		case ESE_DZ_EV_CODE_MCDI_EV:
1945 			efx_mcdi_process_event(channel, &event);
1946 			break;
1947 		case ESE_DZ_EV_CODE_RX_EV:
1948 			spent += efx_ef10_handle_rx_event(channel, &event);
1949 			if (spent >= quota) {
1950 				/* XXX can we split a merged event to
1951 				 * avoid going over-quota?
1952 				 */
1953 				spent = quota;
1954 				goto out;
1955 			}
1956 			break;
1957 		case ESE_DZ_EV_CODE_TX_EV:
1958 			tx_descs += efx_ef10_handle_tx_event(channel, &event);
1959 			if (tx_descs > efx->txq_entries) {
1960 				spent = quota;
1961 				goto out;
1962 			} else if (++spent == quota) {
1963 				goto out;
1964 			}
1965 			break;
1966 		case ESE_DZ_EV_CODE_DRIVER_EV:
1967 			efx_ef10_handle_driver_event(channel, &event);
1968 			if (++spent == quota)
1969 				goto out;
1970 			break;
1971 		case EFX_EF10_DRVGEN_EV:
1972 			efx_ef10_handle_driver_generated_event(channel, &event);
1973 			break;
1974 		default:
1975 			netif_err(efx, hw, efx->net_dev,
1976 				  "channel %d unknown event type %d"
1977 				  " (data " EFX_QWORD_FMT ")\n",
1978 				  channel->channel, ev_code,
1979 				  EFX_QWORD_VAL(event));
1980 		}
1981 	}
1982 
1983 out:
1984 	channel->eventq_read_ptr = read_ptr;
1985 	return spent;
1986 }
1987 
1988 static void efx_ef10_ev_read_ack(struct efx_channel *channel)
1989 {
1990 	struct efx_nic *efx = channel->efx;
1991 	efx_dword_t rptr;
1992 
1993 	if (EFX_EF10_WORKAROUND_35388(efx)) {
1994 		BUILD_BUG_ON(EFX_MIN_EVQ_SIZE <
1995 			     (1 << ERF_DD_EVQ_IND_RPTR_WIDTH));
1996 		BUILD_BUG_ON(EFX_MAX_EVQ_SIZE >
1997 			     (1 << 2 * ERF_DD_EVQ_IND_RPTR_WIDTH));
1998 
1999 		EFX_POPULATE_DWORD_2(rptr, ERF_DD_EVQ_IND_RPTR_FLAGS,
2000 				     EFE_DD_EVQ_IND_RPTR_FLAGS_HIGH,
2001 				     ERF_DD_EVQ_IND_RPTR,
2002 				     (channel->eventq_read_ptr &
2003 				      channel->eventq_mask) >>
2004 				     ERF_DD_EVQ_IND_RPTR_WIDTH);
2005 		efx_writed_page(efx, &rptr, ER_DD_EVQ_INDIRECT,
2006 				channel->channel);
2007 		EFX_POPULATE_DWORD_2(rptr, ERF_DD_EVQ_IND_RPTR_FLAGS,
2008 				     EFE_DD_EVQ_IND_RPTR_FLAGS_LOW,
2009 				     ERF_DD_EVQ_IND_RPTR,
2010 				     channel->eventq_read_ptr &
2011 				     ((1 << ERF_DD_EVQ_IND_RPTR_WIDTH) - 1));
2012 		efx_writed_page(efx, &rptr, ER_DD_EVQ_INDIRECT,
2013 				channel->channel);
2014 	} else {
2015 		EFX_POPULATE_DWORD_1(rptr, ERF_DZ_EVQ_RPTR,
2016 				     channel->eventq_read_ptr &
2017 				     channel->eventq_mask);
2018 		efx_writed_page(efx, &rptr, ER_DZ_EVQ_RPTR, channel->channel);
2019 	}
2020 }
2021 
2022 static void efx_ef10_ev_test_generate(struct efx_channel *channel)
2023 {
2024 	MCDI_DECLARE_BUF(inbuf, MC_CMD_DRIVER_EVENT_IN_LEN);
2025 	struct efx_nic *efx = channel->efx;
2026 	efx_qword_t event;
2027 	int rc;
2028 
2029 	EFX_POPULATE_QWORD_2(event,
2030 			     ESF_DZ_EV_CODE, EFX_EF10_DRVGEN_EV,
2031 			     ESF_DZ_EV_DATA, EFX_EF10_TEST);
2032 
2033 	MCDI_SET_DWORD(inbuf, DRIVER_EVENT_IN_EVQ, channel->channel);
2034 
2035 	/* MCDI_SET_QWORD is not appropriate here since EFX_POPULATE_* has
2036 	 * already swapped the data to little-endian order.
2037 	 */
2038 	memcpy(MCDI_PTR(inbuf, DRIVER_EVENT_IN_DATA), &event.u64[0],
2039 	       sizeof(efx_qword_t));
2040 
2041 	rc = efx_mcdi_rpc(efx, MC_CMD_DRIVER_EVENT, inbuf, sizeof(inbuf),
2042 			  NULL, 0, NULL);
2043 	if (rc != 0)
2044 		goto fail;
2045 
2046 	return;
2047 
2048 fail:
2049 	WARN_ON(true);
2050 	netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
2051 }
2052 
2053 void efx_ef10_handle_drain_event(struct efx_nic *efx)
2054 {
2055 	if (atomic_dec_and_test(&efx->active_queues))
2056 		wake_up(&efx->flush_wq);
2057 
2058 	WARN_ON(atomic_read(&efx->active_queues) < 0);
2059 }
2060 
2061 static int efx_ef10_fini_dmaq(struct efx_nic *efx)
2062 {
2063 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2064 	struct efx_channel *channel;
2065 	struct efx_tx_queue *tx_queue;
2066 	struct efx_rx_queue *rx_queue;
2067 	int pending;
2068 
2069 	/* If the MC has just rebooted, the TX/RX queues will have already been
2070 	 * torn down, but efx->active_queues needs to be set to zero.
2071 	 */
2072 	if (nic_data->must_realloc_vis) {
2073 		atomic_set(&efx->active_queues, 0);
2074 		return 0;
2075 	}
2076 
2077 	/* Do not attempt to write to the NIC during EEH recovery */
2078 	if (efx->state != STATE_RECOVERY) {
2079 		efx_for_each_channel(channel, efx) {
2080 			efx_for_each_channel_rx_queue(rx_queue, channel)
2081 				efx_ef10_rx_fini(rx_queue);
2082 			efx_for_each_channel_tx_queue(tx_queue, channel)
2083 				efx_ef10_tx_fini(tx_queue);
2084 		}
2085 
2086 		wait_event_timeout(efx->flush_wq,
2087 				   atomic_read(&efx->active_queues) == 0,
2088 				   msecs_to_jiffies(EFX_MAX_FLUSH_TIME));
2089 		pending = atomic_read(&efx->active_queues);
2090 		if (pending) {
2091 			netif_err(efx, hw, efx->net_dev, "failed to flush %d queues\n",
2092 				  pending);
2093 			return -ETIMEDOUT;
2094 		}
2095 	}
2096 
2097 	return 0;
2098 }
2099 
2100 static bool efx_ef10_filter_equal(const struct efx_filter_spec *left,
2101 				  const struct efx_filter_spec *right)
2102 {
2103 	if ((left->match_flags ^ right->match_flags) |
2104 	    ((left->flags ^ right->flags) &
2105 	     (EFX_FILTER_FLAG_RX | EFX_FILTER_FLAG_TX)))
2106 		return false;
2107 
2108 	return memcmp(&left->outer_vid, &right->outer_vid,
2109 		      sizeof(struct efx_filter_spec) -
2110 		      offsetof(struct efx_filter_spec, outer_vid)) == 0;
2111 }
2112 
2113 static unsigned int efx_ef10_filter_hash(const struct efx_filter_spec *spec)
2114 {
2115 	BUILD_BUG_ON(offsetof(struct efx_filter_spec, outer_vid) & 3);
2116 	return jhash2((const u32 *)&spec->outer_vid,
2117 		      (sizeof(struct efx_filter_spec) -
2118 		       offsetof(struct efx_filter_spec, outer_vid)) / 4,
2119 		      0);
2120 	/* XXX should we randomise the initval? */
2121 }
2122 
2123 /* Decide whether a filter should be exclusive or else should allow
2124  * delivery to additional recipients.  Currently we decide that
2125  * filters for specific local unicast MAC and IP addresses are
2126  * exclusive.
2127  */
2128 static bool efx_ef10_filter_is_exclusive(const struct efx_filter_spec *spec)
2129 {
2130 	if (spec->match_flags & EFX_FILTER_MATCH_LOC_MAC &&
2131 	    !is_multicast_ether_addr(spec->loc_mac))
2132 		return true;
2133 
2134 	if ((spec->match_flags &
2135 	     (EFX_FILTER_MATCH_ETHER_TYPE | EFX_FILTER_MATCH_LOC_HOST)) ==
2136 	    (EFX_FILTER_MATCH_ETHER_TYPE | EFX_FILTER_MATCH_LOC_HOST)) {
2137 		if (spec->ether_type == htons(ETH_P_IP) &&
2138 		    !ipv4_is_multicast(spec->loc_host[0]))
2139 			return true;
2140 		if (spec->ether_type == htons(ETH_P_IPV6) &&
2141 		    ((const u8 *)spec->loc_host)[0] != 0xff)
2142 			return true;
2143 	}
2144 
2145 	return false;
2146 }
2147 
2148 static struct efx_filter_spec *
2149 efx_ef10_filter_entry_spec(const struct efx_ef10_filter_table *table,
2150 			   unsigned int filter_idx)
2151 {
2152 	return (struct efx_filter_spec *)(table->entry[filter_idx].spec &
2153 					  ~EFX_EF10_FILTER_FLAGS);
2154 }
2155 
2156 static unsigned int
2157 efx_ef10_filter_entry_flags(const struct efx_ef10_filter_table *table,
2158 			   unsigned int filter_idx)
2159 {
2160 	return table->entry[filter_idx].spec & EFX_EF10_FILTER_FLAGS;
2161 }
2162 
2163 static void
2164 efx_ef10_filter_set_entry(struct efx_ef10_filter_table *table,
2165 			  unsigned int filter_idx,
2166 			  const struct efx_filter_spec *spec,
2167 			  unsigned int flags)
2168 {
2169 	table->entry[filter_idx].spec =	(unsigned long)spec | flags;
2170 }
2171 
2172 static void efx_ef10_filter_push_prep(struct efx_nic *efx,
2173 				      const struct efx_filter_spec *spec,
2174 				      efx_dword_t *inbuf, u64 handle,
2175 				      bool replacing)
2176 {
2177 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2178 
2179 	memset(inbuf, 0, MC_CMD_FILTER_OP_IN_LEN);
2180 
2181 	if (replacing) {
2182 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
2183 			       MC_CMD_FILTER_OP_IN_OP_REPLACE);
2184 		MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE, handle);
2185 	} else {
2186 		u32 match_fields = 0;
2187 
2188 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
2189 			       efx_ef10_filter_is_exclusive(spec) ?
2190 			       MC_CMD_FILTER_OP_IN_OP_INSERT :
2191 			       MC_CMD_FILTER_OP_IN_OP_SUBSCRIBE);
2192 
2193 		/* Convert match flags and values.  Unlike almost
2194 		 * everything else in MCDI, these fields are in
2195 		 * network byte order.
2196 		 */
2197 		if (spec->match_flags & EFX_FILTER_MATCH_LOC_MAC_IG)
2198 			match_fields |=
2199 				is_multicast_ether_addr(spec->loc_mac) ?
2200 				1 << MC_CMD_FILTER_OP_IN_MATCH_UNKNOWN_MCAST_DST_LBN :
2201 				1 << MC_CMD_FILTER_OP_IN_MATCH_UNKNOWN_UCAST_DST_LBN;
2202 #define COPY_FIELD(gen_flag, gen_field, mcdi_field)			     \
2203 		if (spec->match_flags & EFX_FILTER_MATCH_ ## gen_flag) {     \
2204 			match_fields |=					     \
2205 				1 << MC_CMD_FILTER_OP_IN_MATCH_ ##	     \
2206 				mcdi_field ## _LBN;			     \
2207 			BUILD_BUG_ON(					     \
2208 				MC_CMD_FILTER_OP_IN_ ## mcdi_field ## _LEN < \
2209 				sizeof(spec->gen_field));		     \
2210 			memcpy(MCDI_PTR(inbuf, FILTER_OP_IN_ ##	mcdi_field), \
2211 			       &spec->gen_field, sizeof(spec->gen_field));   \
2212 		}
2213 		COPY_FIELD(REM_HOST, rem_host, SRC_IP);
2214 		COPY_FIELD(LOC_HOST, loc_host, DST_IP);
2215 		COPY_FIELD(REM_MAC, rem_mac, SRC_MAC);
2216 		COPY_FIELD(REM_PORT, rem_port, SRC_PORT);
2217 		COPY_FIELD(LOC_MAC, loc_mac, DST_MAC);
2218 		COPY_FIELD(LOC_PORT, loc_port, DST_PORT);
2219 		COPY_FIELD(ETHER_TYPE, ether_type, ETHER_TYPE);
2220 		COPY_FIELD(INNER_VID, inner_vid, INNER_VLAN);
2221 		COPY_FIELD(OUTER_VID, outer_vid, OUTER_VLAN);
2222 		COPY_FIELD(IP_PROTO, ip_proto, IP_PROTO);
2223 #undef COPY_FIELD
2224 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_MATCH_FIELDS,
2225 			       match_fields);
2226 	}
2227 
2228 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_PORT_ID, EVB_PORT_ID_ASSIGNED);
2229 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_DEST,
2230 		       spec->dmaq_id == EFX_FILTER_RX_DMAQ_ID_DROP ?
2231 		       MC_CMD_FILTER_OP_IN_RX_DEST_DROP :
2232 		       MC_CMD_FILTER_OP_IN_RX_DEST_HOST);
2233 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_TX_DEST,
2234 		       MC_CMD_FILTER_OP_IN_TX_DEST_DEFAULT);
2235 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_QUEUE, spec->dmaq_id);
2236 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_MODE,
2237 		       (spec->flags & EFX_FILTER_FLAG_RX_RSS) ?
2238 		       MC_CMD_FILTER_OP_IN_RX_MODE_RSS :
2239 		       MC_CMD_FILTER_OP_IN_RX_MODE_SIMPLE);
2240 	if (spec->flags & EFX_FILTER_FLAG_RX_RSS)
2241 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_CONTEXT,
2242 			       spec->rss_context !=
2243 			       EFX_FILTER_RSS_CONTEXT_DEFAULT ?
2244 			       spec->rss_context : nic_data->rx_rss_context);
2245 }
2246 
2247 static int efx_ef10_filter_push(struct efx_nic *efx,
2248 				const struct efx_filter_spec *spec,
2249 				u64 *handle, bool replacing)
2250 {
2251 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_IN_LEN);
2252 	MCDI_DECLARE_BUF(outbuf, MC_CMD_FILTER_OP_OUT_LEN);
2253 	int rc;
2254 
2255 	efx_ef10_filter_push_prep(efx, spec, inbuf, *handle, replacing);
2256 	rc = efx_mcdi_rpc(efx, MC_CMD_FILTER_OP, inbuf, sizeof(inbuf),
2257 			  outbuf, sizeof(outbuf), NULL);
2258 	if (rc == 0)
2259 		*handle = MCDI_QWORD(outbuf, FILTER_OP_OUT_HANDLE);
2260 	return rc;
2261 }
2262 
2263 static int efx_ef10_filter_rx_match_pri(struct efx_ef10_filter_table *table,
2264 					enum efx_filter_match_flags match_flags)
2265 {
2266 	unsigned int match_pri;
2267 
2268 	for (match_pri = 0;
2269 	     match_pri < table->rx_match_count;
2270 	     match_pri++)
2271 		if (table->rx_match_flags[match_pri] == match_flags)
2272 			return match_pri;
2273 
2274 	return -EPROTONOSUPPORT;
2275 }
2276 
2277 static s32 efx_ef10_filter_insert(struct efx_nic *efx,
2278 				  struct efx_filter_spec *spec,
2279 				  bool replace_equal)
2280 {
2281 	struct efx_ef10_filter_table *table = efx->filter_state;
2282 	DECLARE_BITMAP(mc_rem_map, EFX_EF10_FILTER_SEARCH_LIMIT);
2283 	struct efx_filter_spec *saved_spec;
2284 	unsigned int match_pri, hash;
2285 	unsigned int priv_flags;
2286 	bool replacing = false;
2287 	int ins_index = -1;
2288 	DEFINE_WAIT(wait);
2289 	bool is_mc_recip;
2290 	s32 rc;
2291 
2292 	/* For now, only support RX filters */
2293 	if ((spec->flags & (EFX_FILTER_FLAG_RX | EFX_FILTER_FLAG_TX)) !=
2294 	    EFX_FILTER_FLAG_RX)
2295 		return -EINVAL;
2296 
2297 	rc = efx_ef10_filter_rx_match_pri(table, spec->match_flags);
2298 	if (rc < 0)
2299 		return rc;
2300 	match_pri = rc;
2301 
2302 	hash = efx_ef10_filter_hash(spec);
2303 	is_mc_recip = efx_filter_is_mc_recipient(spec);
2304 	if (is_mc_recip)
2305 		bitmap_zero(mc_rem_map, EFX_EF10_FILTER_SEARCH_LIMIT);
2306 
2307 	/* Find any existing filters with the same match tuple or
2308 	 * else a free slot to insert at.  If any of them are busy,
2309 	 * we have to wait and retry.
2310 	 */
2311 	for (;;) {
2312 		unsigned int depth = 1;
2313 		unsigned int i;
2314 
2315 		spin_lock_bh(&efx->filter_lock);
2316 
2317 		for (;;) {
2318 			i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
2319 			saved_spec = efx_ef10_filter_entry_spec(table, i);
2320 
2321 			if (!saved_spec) {
2322 				if (ins_index < 0)
2323 					ins_index = i;
2324 			} else if (efx_ef10_filter_equal(spec, saved_spec)) {
2325 				if (table->entry[i].spec &
2326 				    EFX_EF10_FILTER_FLAG_BUSY)
2327 					break;
2328 				if (spec->priority < saved_spec->priority &&
2329 				    !(saved_spec->priority ==
2330 				      EFX_FILTER_PRI_REQUIRED &&
2331 				      saved_spec->flags &
2332 				      EFX_FILTER_FLAG_RX_STACK)) {
2333 					rc = -EPERM;
2334 					goto out_unlock;
2335 				}
2336 				if (!is_mc_recip) {
2337 					/* This is the only one */
2338 					if (spec->priority ==
2339 					    saved_spec->priority &&
2340 					    !replace_equal) {
2341 						rc = -EEXIST;
2342 						goto out_unlock;
2343 					}
2344 					ins_index = i;
2345 					goto found;
2346 				} else if (spec->priority >
2347 					   saved_spec->priority ||
2348 					   (spec->priority ==
2349 					    saved_spec->priority &&
2350 					    replace_equal)) {
2351 					if (ins_index < 0)
2352 						ins_index = i;
2353 					else
2354 						__set_bit(depth, mc_rem_map);
2355 				}
2356 			}
2357 
2358 			/* Once we reach the maximum search depth, use
2359 			 * the first suitable slot or return -EBUSY if
2360 			 * there was none
2361 			 */
2362 			if (depth == EFX_EF10_FILTER_SEARCH_LIMIT) {
2363 				if (ins_index < 0) {
2364 					rc = -EBUSY;
2365 					goto out_unlock;
2366 				}
2367 				goto found;
2368 			}
2369 
2370 			++depth;
2371 		}
2372 
2373 		prepare_to_wait(&table->waitq, &wait, TASK_UNINTERRUPTIBLE);
2374 		spin_unlock_bh(&efx->filter_lock);
2375 		schedule();
2376 	}
2377 
2378 found:
2379 	/* Create a software table entry if necessary, and mark it
2380 	 * busy.  We might yet fail to insert, but any attempt to
2381 	 * insert a conflicting filter while we're waiting for the
2382 	 * firmware must find the busy entry.
2383 	 */
2384 	saved_spec = efx_ef10_filter_entry_spec(table, ins_index);
2385 	if (saved_spec) {
2386 		if (spec->flags & EFX_FILTER_FLAG_RX_STACK) {
2387 			/* Just make sure it won't be removed */
2388 			saved_spec->flags |= EFX_FILTER_FLAG_RX_STACK;
2389 			table->entry[ins_index].spec &=
2390 				~EFX_EF10_FILTER_FLAG_STACK_OLD;
2391 			rc = ins_index;
2392 			goto out_unlock;
2393 		}
2394 		replacing = true;
2395 		priv_flags = efx_ef10_filter_entry_flags(table, ins_index);
2396 	} else {
2397 		saved_spec = kmalloc(sizeof(*spec), GFP_ATOMIC);
2398 		if (!saved_spec) {
2399 			rc = -ENOMEM;
2400 			goto out_unlock;
2401 		}
2402 		*saved_spec = *spec;
2403 		priv_flags = 0;
2404 	}
2405 	efx_ef10_filter_set_entry(table, ins_index, saved_spec,
2406 				  priv_flags | EFX_EF10_FILTER_FLAG_BUSY);
2407 
2408 	/* Mark lower-priority multicast recipients busy prior to removal */
2409 	if (is_mc_recip) {
2410 		unsigned int depth, i;
2411 
2412 		for (depth = 0; depth < EFX_EF10_FILTER_SEARCH_LIMIT; depth++) {
2413 			i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
2414 			if (test_bit(depth, mc_rem_map))
2415 				table->entry[i].spec |=
2416 					EFX_EF10_FILTER_FLAG_BUSY;
2417 		}
2418 	}
2419 
2420 	spin_unlock_bh(&efx->filter_lock);
2421 
2422 	rc = efx_ef10_filter_push(efx, spec, &table->entry[ins_index].handle,
2423 				  replacing);
2424 
2425 	/* Finalise the software table entry */
2426 	spin_lock_bh(&efx->filter_lock);
2427 	if (rc == 0) {
2428 		if (replacing) {
2429 			/* Update the fields that may differ */
2430 			saved_spec->priority = spec->priority;
2431 			saved_spec->flags &= EFX_FILTER_FLAG_RX_STACK;
2432 			saved_spec->flags |= spec->flags;
2433 			saved_spec->rss_context = spec->rss_context;
2434 			saved_spec->dmaq_id = spec->dmaq_id;
2435 		}
2436 	} else if (!replacing) {
2437 		kfree(saved_spec);
2438 		saved_spec = NULL;
2439 	}
2440 	efx_ef10_filter_set_entry(table, ins_index, saved_spec, priv_flags);
2441 
2442 	/* Remove and finalise entries for lower-priority multicast
2443 	 * recipients
2444 	 */
2445 	if (is_mc_recip) {
2446 		MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_IN_LEN);
2447 		unsigned int depth, i;
2448 
2449 		memset(inbuf, 0, sizeof(inbuf));
2450 
2451 		for (depth = 0; depth < EFX_EF10_FILTER_SEARCH_LIMIT; depth++) {
2452 			if (!test_bit(depth, mc_rem_map))
2453 				continue;
2454 
2455 			i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
2456 			saved_spec = efx_ef10_filter_entry_spec(table, i);
2457 			priv_flags = efx_ef10_filter_entry_flags(table, i);
2458 
2459 			if (rc == 0) {
2460 				spin_unlock_bh(&efx->filter_lock);
2461 				MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
2462 					       MC_CMD_FILTER_OP_IN_OP_UNSUBSCRIBE);
2463 				MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
2464 					       table->entry[i].handle);
2465 				rc = efx_mcdi_rpc(efx, MC_CMD_FILTER_OP,
2466 						  inbuf, sizeof(inbuf),
2467 						  NULL, 0, NULL);
2468 				spin_lock_bh(&efx->filter_lock);
2469 			}
2470 
2471 			if (rc == 0) {
2472 				kfree(saved_spec);
2473 				saved_spec = NULL;
2474 				priv_flags = 0;
2475 			} else {
2476 				priv_flags &= ~EFX_EF10_FILTER_FLAG_BUSY;
2477 			}
2478 			efx_ef10_filter_set_entry(table, i, saved_spec,
2479 						  priv_flags);
2480 		}
2481 	}
2482 
2483 	/* If successful, return the inserted filter ID */
2484 	if (rc == 0)
2485 		rc = match_pri * HUNT_FILTER_TBL_ROWS + ins_index;
2486 
2487 	wake_up_all(&table->waitq);
2488 out_unlock:
2489 	spin_unlock_bh(&efx->filter_lock);
2490 	finish_wait(&table->waitq, &wait);
2491 	return rc;
2492 }
2493 
2494 static void efx_ef10_filter_update_rx_scatter(struct efx_nic *efx)
2495 {
2496 	/* no need to do anything here on EF10 */
2497 }
2498 
2499 /* Remove a filter.
2500  * If !stack_requested, remove by ID
2501  * If stack_requested, remove by index
2502  * Filter ID may come from userland and must be range-checked.
2503  */
2504 static int efx_ef10_filter_remove_internal(struct efx_nic *efx,
2505 					   enum efx_filter_priority priority,
2506 					   u32 filter_id, bool stack_requested)
2507 {
2508 	unsigned int filter_idx = filter_id % HUNT_FILTER_TBL_ROWS;
2509 	struct efx_ef10_filter_table *table = efx->filter_state;
2510 	MCDI_DECLARE_BUF(inbuf,
2511 			 MC_CMD_FILTER_OP_IN_HANDLE_OFST +
2512 			 MC_CMD_FILTER_OP_IN_HANDLE_LEN);
2513 	struct efx_filter_spec *spec;
2514 	DEFINE_WAIT(wait);
2515 	int rc;
2516 
2517 	/* Find the software table entry and mark it busy.  Don't
2518 	 * remove it yet; any attempt to update while we're waiting
2519 	 * for the firmware must find the busy entry.
2520 	 */
2521 	for (;;) {
2522 		spin_lock_bh(&efx->filter_lock);
2523 		if (!(table->entry[filter_idx].spec &
2524 		      EFX_EF10_FILTER_FLAG_BUSY))
2525 			break;
2526 		prepare_to_wait(&table->waitq, &wait, TASK_UNINTERRUPTIBLE);
2527 		spin_unlock_bh(&efx->filter_lock);
2528 		schedule();
2529 	}
2530 	spec = efx_ef10_filter_entry_spec(table, filter_idx);
2531 	if (!spec || spec->priority > priority ||
2532 	    (!stack_requested &&
2533 	     efx_ef10_filter_rx_match_pri(table, spec->match_flags) !=
2534 	     filter_id / HUNT_FILTER_TBL_ROWS)) {
2535 		rc = -ENOENT;
2536 		goto out_unlock;
2537 	}
2538 	table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_BUSY;
2539 	spin_unlock_bh(&efx->filter_lock);
2540 
2541 	if (spec->flags & EFX_FILTER_FLAG_RX_STACK && !stack_requested) {
2542 		/* Reset steering of a stack-owned filter */
2543 
2544 		struct efx_filter_spec new_spec = *spec;
2545 
2546 		new_spec.priority = EFX_FILTER_PRI_REQUIRED;
2547 		new_spec.flags = (EFX_FILTER_FLAG_RX |
2548 				  EFX_FILTER_FLAG_RX_RSS |
2549 				  EFX_FILTER_FLAG_RX_STACK);
2550 		new_spec.dmaq_id = 0;
2551 		new_spec.rss_context = EFX_FILTER_RSS_CONTEXT_DEFAULT;
2552 		rc = efx_ef10_filter_push(efx, &new_spec,
2553 					  &table->entry[filter_idx].handle,
2554 					  true);
2555 
2556 		spin_lock_bh(&efx->filter_lock);
2557 		if (rc == 0)
2558 			*spec = new_spec;
2559 	} else {
2560 		/* Really remove the filter */
2561 
2562 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
2563 			       efx_ef10_filter_is_exclusive(spec) ?
2564 			       MC_CMD_FILTER_OP_IN_OP_REMOVE :
2565 			       MC_CMD_FILTER_OP_IN_OP_UNSUBSCRIBE);
2566 		MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
2567 			       table->entry[filter_idx].handle);
2568 		rc = efx_mcdi_rpc(efx, MC_CMD_FILTER_OP,
2569 				  inbuf, sizeof(inbuf), NULL, 0, NULL);
2570 
2571 		spin_lock_bh(&efx->filter_lock);
2572 		if (rc == 0) {
2573 			kfree(spec);
2574 			efx_ef10_filter_set_entry(table, filter_idx, NULL, 0);
2575 		}
2576 	}
2577 	table->entry[filter_idx].spec &= ~EFX_EF10_FILTER_FLAG_BUSY;
2578 	wake_up_all(&table->waitq);
2579 out_unlock:
2580 	spin_unlock_bh(&efx->filter_lock);
2581 	finish_wait(&table->waitq, &wait);
2582 	return rc;
2583 }
2584 
2585 static int efx_ef10_filter_remove_safe(struct efx_nic *efx,
2586 				       enum efx_filter_priority priority,
2587 				       u32 filter_id)
2588 {
2589 	return efx_ef10_filter_remove_internal(efx, priority, filter_id, false);
2590 }
2591 
2592 static int efx_ef10_filter_get_safe(struct efx_nic *efx,
2593 				    enum efx_filter_priority priority,
2594 				    u32 filter_id, struct efx_filter_spec *spec)
2595 {
2596 	unsigned int filter_idx = filter_id % HUNT_FILTER_TBL_ROWS;
2597 	struct efx_ef10_filter_table *table = efx->filter_state;
2598 	const struct efx_filter_spec *saved_spec;
2599 	int rc;
2600 
2601 	spin_lock_bh(&efx->filter_lock);
2602 	saved_spec = efx_ef10_filter_entry_spec(table, filter_idx);
2603 	if (saved_spec && saved_spec->priority == priority &&
2604 	    efx_ef10_filter_rx_match_pri(table, saved_spec->match_flags) ==
2605 	    filter_id / HUNT_FILTER_TBL_ROWS) {
2606 		*spec = *saved_spec;
2607 		rc = 0;
2608 	} else {
2609 		rc = -ENOENT;
2610 	}
2611 	spin_unlock_bh(&efx->filter_lock);
2612 	return rc;
2613 }
2614 
2615 static void efx_ef10_filter_clear_rx(struct efx_nic *efx,
2616 				     enum efx_filter_priority priority)
2617 {
2618 	/* TODO */
2619 }
2620 
2621 static u32 efx_ef10_filter_count_rx_used(struct efx_nic *efx,
2622 					 enum efx_filter_priority priority)
2623 {
2624 	struct efx_ef10_filter_table *table = efx->filter_state;
2625 	unsigned int filter_idx;
2626 	s32 count = 0;
2627 
2628 	spin_lock_bh(&efx->filter_lock);
2629 	for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
2630 		if (table->entry[filter_idx].spec &&
2631 		    efx_ef10_filter_entry_spec(table, filter_idx)->priority ==
2632 		    priority)
2633 			++count;
2634 	}
2635 	spin_unlock_bh(&efx->filter_lock);
2636 	return count;
2637 }
2638 
2639 static u32 efx_ef10_filter_get_rx_id_limit(struct efx_nic *efx)
2640 {
2641 	struct efx_ef10_filter_table *table = efx->filter_state;
2642 
2643 	return table->rx_match_count * HUNT_FILTER_TBL_ROWS;
2644 }
2645 
2646 static s32 efx_ef10_filter_get_rx_ids(struct efx_nic *efx,
2647 				      enum efx_filter_priority priority,
2648 				      u32 *buf, u32 size)
2649 {
2650 	struct efx_ef10_filter_table *table = efx->filter_state;
2651 	struct efx_filter_spec *spec;
2652 	unsigned int filter_idx;
2653 	s32 count = 0;
2654 
2655 	spin_lock_bh(&efx->filter_lock);
2656 	for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
2657 		spec = efx_ef10_filter_entry_spec(table, filter_idx);
2658 		if (spec && spec->priority == priority) {
2659 			if (count == size) {
2660 				count = -EMSGSIZE;
2661 				break;
2662 			}
2663 			buf[count++] = (efx_ef10_filter_rx_match_pri(
2664 						table, spec->match_flags) *
2665 					HUNT_FILTER_TBL_ROWS +
2666 					filter_idx);
2667 		}
2668 	}
2669 	spin_unlock_bh(&efx->filter_lock);
2670 	return count;
2671 }
2672 
2673 #ifdef CONFIG_RFS_ACCEL
2674 
2675 static efx_mcdi_async_completer efx_ef10_filter_rfs_insert_complete;
2676 
2677 static s32 efx_ef10_filter_rfs_insert(struct efx_nic *efx,
2678 				      struct efx_filter_spec *spec)
2679 {
2680 	struct efx_ef10_filter_table *table = efx->filter_state;
2681 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_IN_LEN);
2682 	struct efx_filter_spec *saved_spec;
2683 	unsigned int hash, i, depth = 1;
2684 	bool replacing = false;
2685 	int ins_index = -1;
2686 	u64 cookie;
2687 	s32 rc;
2688 
2689 	/* Must be an RX filter without RSS and not for a multicast
2690 	 * destination address (RFS only works for connected sockets).
2691 	 * These restrictions allow us to pass only a tiny amount of
2692 	 * data through to the completion function.
2693 	 */
2694 	EFX_WARN_ON_PARANOID(spec->flags !=
2695 			     (EFX_FILTER_FLAG_RX | EFX_FILTER_FLAG_RX_SCATTER));
2696 	EFX_WARN_ON_PARANOID(spec->priority != EFX_FILTER_PRI_HINT);
2697 	EFX_WARN_ON_PARANOID(efx_filter_is_mc_recipient(spec));
2698 
2699 	hash = efx_ef10_filter_hash(spec);
2700 
2701 	spin_lock_bh(&efx->filter_lock);
2702 
2703 	/* Find any existing filter with the same match tuple or else
2704 	 * a free slot to insert at.  If an existing filter is busy,
2705 	 * we have to give up.
2706 	 */
2707 	for (;;) {
2708 		i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
2709 		saved_spec = efx_ef10_filter_entry_spec(table, i);
2710 
2711 		if (!saved_spec) {
2712 			if (ins_index < 0)
2713 				ins_index = i;
2714 		} else if (efx_ef10_filter_equal(spec, saved_spec)) {
2715 			if (table->entry[i].spec & EFX_EF10_FILTER_FLAG_BUSY) {
2716 				rc = -EBUSY;
2717 				goto fail_unlock;
2718 			}
2719 			EFX_WARN_ON_PARANOID(saved_spec->flags &
2720 					     EFX_FILTER_FLAG_RX_STACK);
2721 			if (spec->priority < saved_spec->priority) {
2722 				rc = -EPERM;
2723 				goto fail_unlock;
2724 			}
2725 			ins_index = i;
2726 			break;
2727 		}
2728 
2729 		/* Once we reach the maximum search depth, use the
2730 		 * first suitable slot or return -EBUSY if there was
2731 		 * none
2732 		 */
2733 		if (depth == EFX_EF10_FILTER_SEARCH_LIMIT) {
2734 			if (ins_index < 0) {
2735 				rc = -EBUSY;
2736 				goto fail_unlock;
2737 			}
2738 			break;
2739 		}
2740 
2741 		++depth;
2742 	}
2743 
2744 	/* Create a software table entry if necessary, and mark it
2745 	 * busy.  We might yet fail to insert, but any attempt to
2746 	 * insert a conflicting filter while we're waiting for the
2747 	 * firmware must find the busy entry.
2748 	 */
2749 	saved_spec = efx_ef10_filter_entry_spec(table, ins_index);
2750 	if (saved_spec) {
2751 		replacing = true;
2752 	} else {
2753 		saved_spec = kmalloc(sizeof(*spec), GFP_ATOMIC);
2754 		if (!saved_spec) {
2755 			rc = -ENOMEM;
2756 			goto fail_unlock;
2757 		}
2758 		*saved_spec = *spec;
2759 	}
2760 	efx_ef10_filter_set_entry(table, ins_index, saved_spec,
2761 				  EFX_EF10_FILTER_FLAG_BUSY);
2762 
2763 	spin_unlock_bh(&efx->filter_lock);
2764 
2765 	/* Pack up the variables needed on completion */
2766 	cookie = replacing << 31 | ins_index << 16 | spec->dmaq_id;
2767 
2768 	efx_ef10_filter_push_prep(efx, spec, inbuf,
2769 				  table->entry[ins_index].handle, replacing);
2770 	efx_mcdi_rpc_async(efx, MC_CMD_FILTER_OP, inbuf, sizeof(inbuf),
2771 			   MC_CMD_FILTER_OP_OUT_LEN,
2772 			   efx_ef10_filter_rfs_insert_complete, cookie);
2773 
2774 	return ins_index;
2775 
2776 fail_unlock:
2777 	spin_unlock_bh(&efx->filter_lock);
2778 	return rc;
2779 }
2780 
2781 static void
2782 efx_ef10_filter_rfs_insert_complete(struct efx_nic *efx, unsigned long cookie,
2783 				    int rc, efx_dword_t *outbuf,
2784 				    size_t outlen_actual)
2785 {
2786 	struct efx_ef10_filter_table *table = efx->filter_state;
2787 	unsigned int ins_index, dmaq_id;
2788 	struct efx_filter_spec *spec;
2789 	bool replacing;
2790 
2791 	/* Unpack the cookie */
2792 	replacing = cookie >> 31;
2793 	ins_index = (cookie >> 16) & (HUNT_FILTER_TBL_ROWS - 1);
2794 	dmaq_id = cookie & 0xffff;
2795 
2796 	spin_lock_bh(&efx->filter_lock);
2797 	spec = efx_ef10_filter_entry_spec(table, ins_index);
2798 	if (rc == 0) {
2799 		table->entry[ins_index].handle =
2800 			MCDI_QWORD(outbuf, FILTER_OP_OUT_HANDLE);
2801 		if (replacing)
2802 			spec->dmaq_id = dmaq_id;
2803 	} else if (!replacing) {
2804 		kfree(spec);
2805 		spec = NULL;
2806 	}
2807 	efx_ef10_filter_set_entry(table, ins_index, spec, 0);
2808 	spin_unlock_bh(&efx->filter_lock);
2809 
2810 	wake_up_all(&table->waitq);
2811 }
2812 
2813 static void
2814 efx_ef10_filter_rfs_expire_complete(struct efx_nic *efx,
2815 				    unsigned long filter_idx,
2816 				    int rc, efx_dword_t *outbuf,
2817 				    size_t outlen_actual);
2818 
2819 static bool efx_ef10_filter_rfs_expire_one(struct efx_nic *efx, u32 flow_id,
2820 					   unsigned int filter_idx)
2821 {
2822 	struct efx_ef10_filter_table *table = efx->filter_state;
2823 	struct efx_filter_spec *spec =
2824 		efx_ef10_filter_entry_spec(table, filter_idx);
2825 	MCDI_DECLARE_BUF(inbuf,
2826 			 MC_CMD_FILTER_OP_IN_HANDLE_OFST +
2827 			 MC_CMD_FILTER_OP_IN_HANDLE_LEN);
2828 
2829 	if (!spec ||
2830 	    (table->entry[filter_idx].spec & EFX_EF10_FILTER_FLAG_BUSY) ||
2831 	    spec->priority != EFX_FILTER_PRI_HINT ||
2832 	    !rps_may_expire_flow(efx->net_dev, spec->dmaq_id,
2833 				 flow_id, filter_idx))
2834 		return false;
2835 
2836 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
2837 		       MC_CMD_FILTER_OP_IN_OP_REMOVE);
2838 	MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
2839 		       table->entry[filter_idx].handle);
2840 	if (efx_mcdi_rpc_async(efx, MC_CMD_FILTER_OP, inbuf, sizeof(inbuf), 0,
2841 			       efx_ef10_filter_rfs_expire_complete, filter_idx))
2842 		return false;
2843 
2844 	table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_BUSY;
2845 	return true;
2846 }
2847 
2848 static void
2849 efx_ef10_filter_rfs_expire_complete(struct efx_nic *efx,
2850 				    unsigned long filter_idx,
2851 				    int rc, efx_dword_t *outbuf,
2852 				    size_t outlen_actual)
2853 {
2854 	struct efx_ef10_filter_table *table = efx->filter_state;
2855 	struct efx_filter_spec *spec =
2856 		efx_ef10_filter_entry_spec(table, filter_idx);
2857 
2858 	spin_lock_bh(&efx->filter_lock);
2859 	if (rc == 0) {
2860 		kfree(spec);
2861 		efx_ef10_filter_set_entry(table, filter_idx, NULL, 0);
2862 	}
2863 	table->entry[filter_idx].spec &= ~EFX_EF10_FILTER_FLAG_BUSY;
2864 	wake_up_all(&table->waitq);
2865 	spin_unlock_bh(&efx->filter_lock);
2866 }
2867 
2868 #endif /* CONFIG_RFS_ACCEL */
2869 
2870 static int efx_ef10_filter_match_flags_from_mcdi(u32 mcdi_flags)
2871 {
2872 	int match_flags = 0;
2873 
2874 #define MAP_FLAG(gen_flag, mcdi_field) {				\
2875 		u32 old_mcdi_flags = mcdi_flags;			\
2876 		mcdi_flags &= ~(1 << MC_CMD_FILTER_OP_IN_MATCH_ ##	\
2877 				mcdi_field ## _LBN);			\
2878 		if (mcdi_flags != old_mcdi_flags)			\
2879 			match_flags |= EFX_FILTER_MATCH_ ## gen_flag;	\
2880 	}
2881 	MAP_FLAG(LOC_MAC_IG, UNKNOWN_UCAST_DST);
2882 	MAP_FLAG(LOC_MAC_IG, UNKNOWN_MCAST_DST);
2883 	MAP_FLAG(REM_HOST, SRC_IP);
2884 	MAP_FLAG(LOC_HOST, DST_IP);
2885 	MAP_FLAG(REM_MAC, SRC_MAC);
2886 	MAP_FLAG(REM_PORT, SRC_PORT);
2887 	MAP_FLAG(LOC_MAC, DST_MAC);
2888 	MAP_FLAG(LOC_PORT, DST_PORT);
2889 	MAP_FLAG(ETHER_TYPE, ETHER_TYPE);
2890 	MAP_FLAG(INNER_VID, INNER_VLAN);
2891 	MAP_FLAG(OUTER_VID, OUTER_VLAN);
2892 	MAP_FLAG(IP_PROTO, IP_PROTO);
2893 #undef MAP_FLAG
2894 
2895 	/* Did we map them all? */
2896 	if (mcdi_flags)
2897 		return -EINVAL;
2898 
2899 	return match_flags;
2900 }
2901 
2902 static int efx_ef10_filter_table_probe(struct efx_nic *efx)
2903 {
2904 	MCDI_DECLARE_BUF(inbuf, MC_CMD_GET_PARSER_DISP_INFO_IN_LEN);
2905 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_PARSER_DISP_INFO_OUT_LENMAX);
2906 	unsigned int pd_match_pri, pd_match_count;
2907 	struct efx_ef10_filter_table *table;
2908 	size_t outlen;
2909 	int rc;
2910 
2911 	table = kzalloc(sizeof(*table), GFP_KERNEL);
2912 	if (!table)
2913 		return -ENOMEM;
2914 
2915 	/* Find out which RX filter types are supported, and their priorities */
2916 	MCDI_SET_DWORD(inbuf, GET_PARSER_DISP_INFO_IN_OP,
2917 		       MC_CMD_GET_PARSER_DISP_INFO_IN_OP_GET_SUPPORTED_RX_MATCHES);
2918 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_PARSER_DISP_INFO,
2919 			  inbuf, sizeof(inbuf), outbuf, sizeof(outbuf),
2920 			  &outlen);
2921 	if (rc)
2922 		goto fail;
2923 	pd_match_count = MCDI_VAR_ARRAY_LEN(
2924 		outlen, GET_PARSER_DISP_INFO_OUT_SUPPORTED_MATCHES);
2925 	table->rx_match_count = 0;
2926 
2927 	for (pd_match_pri = 0; pd_match_pri < pd_match_count; pd_match_pri++) {
2928 		u32 mcdi_flags =
2929 			MCDI_ARRAY_DWORD(
2930 				outbuf,
2931 				GET_PARSER_DISP_INFO_OUT_SUPPORTED_MATCHES,
2932 				pd_match_pri);
2933 		rc = efx_ef10_filter_match_flags_from_mcdi(mcdi_flags);
2934 		if (rc < 0) {
2935 			netif_dbg(efx, probe, efx->net_dev,
2936 				  "%s: fw flags %#x pri %u not supported in driver\n",
2937 				  __func__, mcdi_flags, pd_match_pri);
2938 		} else {
2939 			netif_dbg(efx, probe, efx->net_dev,
2940 				  "%s: fw flags %#x pri %u supported as driver flags %#x pri %u\n",
2941 				  __func__, mcdi_flags, pd_match_pri,
2942 				  rc, table->rx_match_count);
2943 			table->rx_match_flags[table->rx_match_count++] = rc;
2944 		}
2945 	}
2946 
2947 	table->entry = vzalloc(HUNT_FILTER_TBL_ROWS * sizeof(*table->entry));
2948 	if (!table->entry) {
2949 		rc = -ENOMEM;
2950 		goto fail;
2951 	}
2952 
2953 	efx->filter_state = table;
2954 	init_waitqueue_head(&table->waitq);
2955 	return 0;
2956 
2957 fail:
2958 	kfree(table);
2959 	return rc;
2960 }
2961 
2962 static void efx_ef10_filter_table_restore(struct efx_nic *efx)
2963 {
2964 	struct efx_ef10_filter_table *table = efx->filter_state;
2965 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2966 	struct efx_filter_spec *spec;
2967 	unsigned int filter_idx;
2968 	bool failed = false;
2969 	int rc;
2970 
2971 	if (!nic_data->must_restore_filters)
2972 		return;
2973 
2974 	spin_lock_bh(&efx->filter_lock);
2975 
2976 	for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
2977 		spec = efx_ef10_filter_entry_spec(table, filter_idx);
2978 		if (!spec)
2979 			continue;
2980 
2981 		table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_BUSY;
2982 		spin_unlock_bh(&efx->filter_lock);
2983 
2984 		rc = efx_ef10_filter_push(efx, spec,
2985 					  &table->entry[filter_idx].handle,
2986 					  false);
2987 		if (rc)
2988 			failed = true;
2989 
2990 		spin_lock_bh(&efx->filter_lock);
2991 		if (rc) {
2992 			kfree(spec);
2993 			efx_ef10_filter_set_entry(table, filter_idx, NULL, 0);
2994 		} else {
2995 			table->entry[filter_idx].spec &=
2996 				~EFX_EF10_FILTER_FLAG_BUSY;
2997 		}
2998 	}
2999 
3000 	spin_unlock_bh(&efx->filter_lock);
3001 
3002 	if (failed)
3003 		netif_err(efx, hw, efx->net_dev,
3004 			  "unable to restore all filters\n");
3005 	else
3006 		nic_data->must_restore_filters = false;
3007 }
3008 
3009 static void efx_ef10_filter_table_remove(struct efx_nic *efx)
3010 {
3011 	struct efx_ef10_filter_table *table = efx->filter_state;
3012 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_IN_LEN);
3013 	struct efx_filter_spec *spec;
3014 	unsigned int filter_idx;
3015 	int rc;
3016 
3017 	for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
3018 		spec = efx_ef10_filter_entry_spec(table, filter_idx);
3019 		if (!spec)
3020 			continue;
3021 
3022 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
3023 			       efx_ef10_filter_is_exclusive(spec) ?
3024 			       MC_CMD_FILTER_OP_IN_OP_REMOVE :
3025 			       MC_CMD_FILTER_OP_IN_OP_UNSUBSCRIBE);
3026 		MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
3027 			       table->entry[filter_idx].handle);
3028 		rc = efx_mcdi_rpc(efx, MC_CMD_FILTER_OP, inbuf, sizeof(inbuf),
3029 				  NULL, 0, NULL);
3030 
3031 		WARN_ON(rc != 0);
3032 		kfree(spec);
3033 	}
3034 
3035 	vfree(table->entry);
3036 	kfree(table);
3037 }
3038 
3039 static void efx_ef10_filter_sync_rx_mode(struct efx_nic *efx)
3040 {
3041 	struct efx_ef10_filter_table *table = efx->filter_state;
3042 	struct net_device *net_dev = efx->net_dev;
3043 	struct efx_filter_spec spec;
3044 	bool remove_failed = false;
3045 	struct netdev_hw_addr *uc;
3046 	struct netdev_hw_addr *mc;
3047 	unsigned int filter_idx;
3048 	int i, n, rc;
3049 
3050 	if (!efx_dev_registered(efx))
3051 		return;
3052 
3053 	/* Mark old filters that may need to be removed */
3054 	spin_lock_bh(&efx->filter_lock);
3055 	n = table->stack_uc_count < 0 ? 1 : table->stack_uc_count;
3056 	for (i = 0; i < n; i++) {
3057 		filter_idx = table->stack_uc_list[i].id % HUNT_FILTER_TBL_ROWS;
3058 		table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_STACK_OLD;
3059 	}
3060 	n = table->stack_mc_count < 0 ? 1 : table->stack_mc_count;
3061 	for (i = 0; i < n; i++) {
3062 		filter_idx = table->stack_mc_list[i].id % HUNT_FILTER_TBL_ROWS;
3063 		table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_STACK_OLD;
3064 	}
3065 	spin_unlock_bh(&efx->filter_lock);
3066 
3067 	/* Copy/convert the address lists; add the primary station
3068 	 * address and broadcast address
3069 	 */
3070 	netif_addr_lock_bh(net_dev);
3071 	if (net_dev->flags & IFF_PROMISC ||
3072 	    netdev_uc_count(net_dev) >= EFX_EF10_FILTER_STACK_UC_MAX) {
3073 		table->stack_uc_count = -1;
3074 	} else {
3075 		table->stack_uc_count = 1 + netdev_uc_count(net_dev);
3076 		memcpy(table->stack_uc_list[0].addr, net_dev->dev_addr,
3077 		       ETH_ALEN);
3078 		i = 1;
3079 		netdev_for_each_uc_addr(uc, net_dev) {
3080 			memcpy(table->stack_uc_list[i].addr,
3081 			       uc->addr, ETH_ALEN);
3082 			i++;
3083 		}
3084 	}
3085 	if (net_dev->flags & (IFF_PROMISC | IFF_ALLMULTI) ||
3086 	    netdev_mc_count(net_dev) >= EFX_EF10_FILTER_STACK_MC_MAX) {
3087 		table->stack_mc_count = -1;
3088 	} else {
3089 		table->stack_mc_count = 1 + netdev_mc_count(net_dev);
3090 		eth_broadcast_addr(table->stack_mc_list[0].addr);
3091 		i = 1;
3092 		netdev_for_each_mc_addr(mc, net_dev) {
3093 			memcpy(table->stack_mc_list[i].addr,
3094 			       mc->addr, ETH_ALEN);
3095 			i++;
3096 		}
3097 	}
3098 	netif_addr_unlock_bh(net_dev);
3099 
3100 	/* Insert/renew unicast filters */
3101 	if (table->stack_uc_count >= 0) {
3102 		for (i = 0; i < table->stack_uc_count; i++) {
3103 			efx_filter_init_rx(&spec, EFX_FILTER_PRI_REQUIRED,
3104 					   EFX_FILTER_FLAG_RX_RSS |
3105 					   EFX_FILTER_FLAG_RX_STACK,
3106 					   0);
3107 			efx_filter_set_eth_local(&spec, EFX_FILTER_VID_UNSPEC,
3108 						 table->stack_uc_list[i].addr);
3109 			rc = efx_ef10_filter_insert(efx, &spec, true);
3110 			if (rc < 0) {
3111 				/* Fall back to unicast-promisc */
3112 				while (i--)
3113 					efx_ef10_filter_remove_safe(
3114 						efx, EFX_FILTER_PRI_REQUIRED,
3115 						table->stack_uc_list[i].id);
3116 				table->stack_uc_count = -1;
3117 				break;
3118 			}
3119 			table->stack_uc_list[i].id = rc;
3120 		}
3121 	}
3122 	if (table->stack_uc_count < 0) {
3123 		efx_filter_init_rx(&spec, EFX_FILTER_PRI_REQUIRED,
3124 				   EFX_FILTER_FLAG_RX_RSS |
3125 				   EFX_FILTER_FLAG_RX_STACK,
3126 				   0);
3127 		efx_filter_set_uc_def(&spec);
3128 		rc = efx_ef10_filter_insert(efx, &spec, true);
3129 		if (rc < 0) {
3130 			WARN_ON(1);
3131 			table->stack_uc_count = 0;
3132 		} else {
3133 			table->stack_uc_list[0].id = rc;
3134 		}
3135 	}
3136 
3137 	/* Insert/renew multicast filters */
3138 	if (table->stack_mc_count >= 0) {
3139 		for (i = 0; i < table->stack_mc_count; i++) {
3140 			efx_filter_init_rx(&spec, EFX_FILTER_PRI_REQUIRED,
3141 					   EFX_FILTER_FLAG_RX_RSS |
3142 					   EFX_FILTER_FLAG_RX_STACK,
3143 					   0);
3144 			efx_filter_set_eth_local(&spec, EFX_FILTER_VID_UNSPEC,
3145 						 table->stack_mc_list[i].addr);
3146 			rc = efx_ef10_filter_insert(efx, &spec, true);
3147 			if (rc < 0) {
3148 				/* Fall back to multicast-promisc */
3149 				while (i--)
3150 					efx_ef10_filter_remove_safe(
3151 						efx, EFX_FILTER_PRI_REQUIRED,
3152 						table->stack_mc_list[i].id);
3153 				table->stack_mc_count = -1;
3154 				break;
3155 			}
3156 			table->stack_mc_list[i].id = rc;
3157 		}
3158 	}
3159 	if (table->stack_mc_count < 0) {
3160 		efx_filter_init_rx(&spec, EFX_FILTER_PRI_REQUIRED,
3161 				   EFX_FILTER_FLAG_RX_RSS |
3162 				   EFX_FILTER_FLAG_RX_STACK,
3163 				   0);
3164 		efx_filter_set_mc_def(&spec);
3165 		rc = efx_ef10_filter_insert(efx, &spec, true);
3166 		if (rc < 0) {
3167 			WARN_ON(1);
3168 			table->stack_mc_count = 0;
3169 		} else {
3170 			table->stack_mc_list[0].id = rc;
3171 		}
3172 	}
3173 
3174 	/* Remove filters that weren't renewed.  Since nothing else
3175 	 * changes the STACK_OLD flag or removes these filters, we
3176 	 * don't need to hold the filter_lock while scanning for
3177 	 * these filters.
3178 	 */
3179 	for (i = 0; i < HUNT_FILTER_TBL_ROWS; i++) {
3180 		if (ACCESS_ONCE(table->entry[i].spec) &
3181 		    EFX_EF10_FILTER_FLAG_STACK_OLD) {
3182 			if (efx_ef10_filter_remove_internal(efx,
3183 					EFX_FILTER_PRI_REQUIRED,
3184 					i, true) < 0)
3185 				remove_failed = true;
3186 		}
3187 	}
3188 	WARN_ON(remove_failed);
3189 }
3190 
3191 static int efx_ef10_mac_reconfigure(struct efx_nic *efx)
3192 {
3193 	efx_ef10_filter_sync_rx_mode(efx);
3194 
3195 	return efx_mcdi_set_mac(efx);
3196 }
3197 
3198 #ifdef CONFIG_SFC_MTD
3199 
3200 struct efx_ef10_nvram_type_info {
3201 	u16 type, type_mask;
3202 	u8 port;
3203 	const char *name;
3204 };
3205 
3206 static const struct efx_ef10_nvram_type_info efx_ef10_nvram_types[] = {
3207 	{ NVRAM_PARTITION_TYPE_MC_FIRMWARE,	   0,    0, "sfc_mcfw" },
3208 	{ NVRAM_PARTITION_TYPE_MC_FIRMWARE_BACKUP, 0,    0, "sfc_mcfw_backup" },
3209 	{ NVRAM_PARTITION_TYPE_EXPANSION_ROM,	   0,    0, "sfc_exp_rom" },
3210 	{ NVRAM_PARTITION_TYPE_STATIC_CONFIG,	   0,    0, "sfc_static_cfg" },
3211 	{ NVRAM_PARTITION_TYPE_DYNAMIC_CONFIG,	   0,    0, "sfc_dynamic_cfg" },
3212 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT0, 0,   0, "sfc_exp_rom_cfg" },
3213 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT1, 0,   1, "sfc_exp_rom_cfg" },
3214 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT2, 0,   2, "sfc_exp_rom_cfg" },
3215 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT3, 0,   3, "sfc_exp_rom_cfg" },
3216 	{ NVRAM_PARTITION_TYPE_PHY_MIN,		   0xff, 0, "sfc_phy_fw" },
3217 };
3218 
3219 static int efx_ef10_mtd_probe_partition(struct efx_nic *efx,
3220 					struct efx_mcdi_mtd_partition *part,
3221 					unsigned int type)
3222 {
3223 	MCDI_DECLARE_BUF(inbuf, MC_CMD_NVRAM_METADATA_IN_LEN);
3224 	MCDI_DECLARE_BUF(outbuf, MC_CMD_NVRAM_METADATA_OUT_LENMAX);
3225 	const struct efx_ef10_nvram_type_info *info;
3226 	size_t size, erase_size, outlen;
3227 	bool protected;
3228 	int rc;
3229 
3230 	for (info = efx_ef10_nvram_types; ; info++) {
3231 		if (info ==
3232 		    efx_ef10_nvram_types + ARRAY_SIZE(efx_ef10_nvram_types))
3233 			return -ENODEV;
3234 		if ((type & ~info->type_mask) == info->type)
3235 			break;
3236 	}
3237 	if (info->port != efx_port_num(efx))
3238 		return -ENODEV;
3239 
3240 	rc = efx_mcdi_nvram_info(efx, type, &size, &erase_size, &protected);
3241 	if (rc)
3242 		return rc;
3243 	if (protected)
3244 		return -ENODEV; /* hide it */
3245 
3246 	part->nvram_type = type;
3247 
3248 	MCDI_SET_DWORD(inbuf, NVRAM_METADATA_IN_TYPE, type);
3249 	rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_METADATA, inbuf, sizeof(inbuf),
3250 			  outbuf, sizeof(outbuf), &outlen);
3251 	if (rc)
3252 		return rc;
3253 	if (outlen < MC_CMD_NVRAM_METADATA_OUT_LENMIN)
3254 		return -EIO;
3255 	if (MCDI_DWORD(outbuf, NVRAM_METADATA_OUT_FLAGS) &
3256 	    (1 << MC_CMD_NVRAM_METADATA_OUT_SUBTYPE_VALID_LBN))
3257 		part->fw_subtype = MCDI_DWORD(outbuf,
3258 					      NVRAM_METADATA_OUT_SUBTYPE);
3259 
3260 	part->common.dev_type_name = "EF10 NVRAM manager";
3261 	part->common.type_name = info->name;
3262 
3263 	part->common.mtd.type = MTD_NORFLASH;
3264 	part->common.mtd.flags = MTD_CAP_NORFLASH;
3265 	part->common.mtd.size = size;
3266 	part->common.mtd.erasesize = erase_size;
3267 
3268 	return 0;
3269 }
3270 
3271 static int efx_ef10_mtd_probe(struct efx_nic *efx)
3272 {
3273 	MCDI_DECLARE_BUF(outbuf, MC_CMD_NVRAM_PARTITIONS_OUT_LENMAX);
3274 	struct efx_mcdi_mtd_partition *parts;
3275 	size_t outlen, n_parts_total, i, n_parts;
3276 	unsigned int type;
3277 	int rc;
3278 
3279 	ASSERT_RTNL();
3280 
3281 	BUILD_BUG_ON(MC_CMD_NVRAM_PARTITIONS_IN_LEN != 0);
3282 	rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_PARTITIONS, NULL, 0,
3283 			  outbuf, sizeof(outbuf), &outlen);
3284 	if (rc)
3285 		return rc;
3286 	if (outlen < MC_CMD_NVRAM_PARTITIONS_OUT_LENMIN)
3287 		return -EIO;
3288 
3289 	n_parts_total = MCDI_DWORD(outbuf, NVRAM_PARTITIONS_OUT_NUM_PARTITIONS);
3290 	if (n_parts_total >
3291 	    MCDI_VAR_ARRAY_LEN(outlen, NVRAM_PARTITIONS_OUT_TYPE_ID))
3292 		return -EIO;
3293 
3294 	parts = kcalloc(n_parts_total, sizeof(*parts), GFP_KERNEL);
3295 	if (!parts)
3296 		return -ENOMEM;
3297 
3298 	n_parts = 0;
3299 	for (i = 0; i < n_parts_total; i++) {
3300 		type = MCDI_ARRAY_DWORD(outbuf, NVRAM_PARTITIONS_OUT_TYPE_ID,
3301 					i);
3302 		rc = efx_ef10_mtd_probe_partition(efx, &parts[n_parts], type);
3303 		if (rc == 0)
3304 			n_parts++;
3305 		else if (rc != -ENODEV)
3306 			goto fail;
3307 	}
3308 
3309 	rc = efx_mtd_add(efx, &parts[0].common, n_parts, sizeof(*parts));
3310 fail:
3311 	if (rc)
3312 		kfree(parts);
3313 	return rc;
3314 }
3315 
3316 #endif /* CONFIG_SFC_MTD */
3317 
3318 static void efx_ef10_ptp_write_host_time(struct efx_nic *efx, u32 host_time)
3319 {
3320 	_efx_writed(efx, cpu_to_le32(host_time), ER_DZ_MC_DB_LWRD);
3321 }
3322 
3323 const struct efx_nic_type efx_hunt_a0_nic_type = {
3324 	.mem_map_size = efx_ef10_mem_map_size,
3325 	.probe = efx_ef10_probe,
3326 	.remove = efx_ef10_remove,
3327 	.dimension_resources = efx_ef10_dimension_resources,
3328 	.init = efx_ef10_init_nic,
3329 	.fini = efx_port_dummy_op_void,
3330 	.map_reset_reason = efx_mcdi_map_reset_reason,
3331 	.map_reset_flags = efx_ef10_map_reset_flags,
3332 	.reset = efx_mcdi_reset,
3333 	.probe_port = efx_mcdi_port_probe,
3334 	.remove_port = efx_mcdi_port_remove,
3335 	.fini_dmaq = efx_ef10_fini_dmaq,
3336 	.describe_stats = efx_ef10_describe_stats,
3337 	.update_stats = efx_ef10_update_stats,
3338 	.start_stats = efx_mcdi_mac_start_stats,
3339 	.stop_stats = efx_mcdi_mac_stop_stats,
3340 	.set_id_led = efx_mcdi_set_id_led,
3341 	.push_irq_moderation = efx_ef10_push_irq_moderation,
3342 	.reconfigure_mac = efx_ef10_mac_reconfigure,
3343 	.check_mac_fault = efx_mcdi_mac_check_fault,
3344 	.reconfigure_port = efx_mcdi_port_reconfigure,
3345 	.get_wol = efx_ef10_get_wol,
3346 	.set_wol = efx_ef10_set_wol,
3347 	.resume_wol = efx_port_dummy_op_void,
3348 	/* TODO: test_chip */
3349 	.test_nvram = efx_mcdi_nvram_test_all,
3350 	.mcdi_request = efx_ef10_mcdi_request,
3351 	.mcdi_poll_response = efx_ef10_mcdi_poll_response,
3352 	.mcdi_read_response = efx_ef10_mcdi_read_response,
3353 	.mcdi_poll_reboot = efx_ef10_mcdi_poll_reboot,
3354 	.irq_enable_master = efx_port_dummy_op_void,
3355 	.irq_test_generate = efx_ef10_irq_test_generate,
3356 	.irq_disable_non_ev = efx_port_dummy_op_void,
3357 	.irq_handle_msi = efx_ef10_msi_interrupt,
3358 	.irq_handle_legacy = efx_ef10_legacy_interrupt,
3359 	.tx_probe = efx_ef10_tx_probe,
3360 	.tx_init = efx_ef10_tx_init,
3361 	.tx_remove = efx_ef10_tx_remove,
3362 	.tx_write = efx_ef10_tx_write,
3363 	.rx_push_indir_table = efx_ef10_rx_push_indir_table,
3364 	.rx_probe = efx_ef10_rx_probe,
3365 	.rx_init = efx_ef10_rx_init,
3366 	.rx_remove = efx_ef10_rx_remove,
3367 	.rx_write = efx_ef10_rx_write,
3368 	.rx_defer_refill = efx_ef10_rx_defer_refill,
3369 	.ev_probe = efx_ef10_ev_probe,
3370 	.ev_init = efx_ef10_ev_init,
3371 	.ev_fini = efx_ef10_ev_fini,
3372 	.ev_remove = efx_ef10_ev_remove,
3373 	.ev_process = efx_ef10_ev_process,
3374 	.ev_read_ack = efx_ef10_ev_read_ack,
3375 	.ev_test_generate = efx_ef10_ev_test_generate,
3376 	.filter_table_probe = efx_ef10_filter_table_probe,
3377 	.filter_table_restore = efx_ef10_filter_table_restore,
3378 	.filter_table_remove = efx_ef10_filter_table_remove,
3379 	.filter_update_rx_scatter = efx_ef10_filter_update_rx_scatter,
3380 	.filter_insert = efx_ef10_filter_insert,
3381 	.filter_remove_safe = efx_ef10_filter_remove_safe,
3382 	.filter_get_safe = efx_ef10_filter_get_safe,
3383 	.filter_clear_rx = efx_ef10_filter_clear_rx,
3384 	.filter_count_rx_used = efx_ef10_filter_count_rx_used,
3385 	.filter_get_rx_id_limit = efx_ef10_filter_get_rx_id_limit,
3386 	.filter_get_rx_ids = efx_ef10_filter_get_rx_ids,
3387 #ifdef CONFIG_RFS_ACCEL
3388 	.filter_rfs_insert = efx_ef10_filter_rfs_insert,
3389 	.filter_rfs_expire_one = efx_ef10_filter_rfs_expire_one,
3390 #endif
3391 #ifdef CONFIG_SFC_MTD
3392 	.mtd_probe = efx_ef10_mtd_probe,
3393 	.mtd_rename = efx_mcdi_mtd_rename,
3394 	.mtd_read = efx_mcdi_mtd_read,
3395 	.mtd_erase = efx_mcdi_mtd_erase,
3396 	.mtd_write = efx_mcdi_mtd_write,
3397 	.mtd_sync = efx_mcdi_mtd_sync,
3398 #endif
3399 	.ptp_write_host_time = efx_ef10_ptp_write_host_time,
3400 
3401 	.revision = EFX_REV_HUNT_A0,
3402 	.max_dma_mask = DMA_BIT_MASK(ESF_DZ_TX_KER_BUF_ADDR_WIDTH),
3403 	.rx_prefix_size = ES_DZ_RX_PREFIX_SIZE,
3404 	.rx_hash_offset = ES_DZ_RX_PREFIX_HASH_OFST,
3405 	.can_rx_scatter = true,
3406 	.always_rx_scatter = true,
3407 	.max_interrupt_mode = EFX_INT_MODE_MSIX,
3408 	.timer_period_max = 1 << ERF_DD_EVQ_IND_TIMER_VAL_WIDTH,
3409 	.offload_features = (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
3410 			     NETIF_F_RXHASH | NETIF_F_NTUPLE),
3411 	.mcdi_max_ver = 2,
3412 	.max_rx_ip_filters = HUNT_FILTER_TBL_ROWS,
3413 };
3414