xref: /linux/drivers/net/ethernet/sfc/ef10.c (revision ca55b2fef3a9373fcfc30f82fd26bc7fccbda732)
1 /****************************************************************************
2  * Driver for Solarflare network controllers and boards
3  * Copyright 2012-2013 Solarflare Communications Inc.
4  *
5  * This program is free software; you can redistribute it and/or modify it
6  * under the terms of the GNU General Public License version 2 as published
7  * by the Free Software Foundation, incorporated herein by reference.
8  */
9 
10 #include "net_driver.h"
11 #include "ef10_regs.h"
12 #include "io.h"
13 #include "mcdi.h"
14 #include "mcdi_pcol.h"
15 #include "nic.h"
16 #include "workarounds.h"
17 #include "selftest.h"
18 #include "ef10_sriov.h"
19 #include <linux/in.h>
20 #include <linux/jhash.h>
21 #include <linux/wait.h>
22 #include <linux/workqueue.h>
23 
24 /* Hardware control for EF10 architecture including 'Huntington'. */
25 
26 #define EFX_EF10_DRVGEN_EV		7
27 enum {
28 	EFX_EF10_TEST = 1,
29 	EFX_EF10_REFILL,
30 };
31 
32 /* The reserved RSS context value */
33 #define EFX_EF10_RSS_CONTEXT_INVALID	0xffffffff
34 /* The maximum size of a shared RSS context */
35 /* TODO: this should really be from the mcdi protocol export */
36 #define EFX_EF10_MAX_SHARED_RSS_CONTEXT_SIZE 64UL
37 
38 /* The filter table(s) are managed by firmware and we have write-only
39  * access.  When removing filters we must identify them to the
40  * firmware by a 64-bit handle, but this is too wide for Linux kernel
41  * interfaces (32-bit for RX NFC, 16-bit for RFS).  Also, we need to
42  * be able to tell in advance whether a requested insertion will
43  * replace an existing filter.  Therefore we maintain a software hash
44  * table, which should be at least as large as the hardware hash
45  * table.
46  *
47  * Huntington has a single 8K filter table shared between all filter
48  * types and both ports.
49  */
50 #define HUNT_FILTER_TBL_ROWS 8192
51 
52 #define EFX_EF10_FILTER_ID_INVALID 0xffff
53 struct efx_ef10_dev_addr {
54 	u8 addr[ETH_ALEN];
55 	u16 id;
56 };
57 
58 struct efx_ef10_filter_table {
59 /* The RX match field masks supported by this fw & hw, in order of priority */
60 	enum efx_filter_match_flags rx_match_flags[
61 		MC_CMD_GET_PARSER_DISP_INFO_OUT_SUPPORTED_MATCHES_MAXNUM];
62 	unsigned int rx_match_count;
63 
64 	struct {
65 		unsigned long spec;	/* pointer to spec plus flag bits */
66 /* BUSY flag indicates that an update is in progress.  AUTO_OLD is
67  * used to mark and sweep MAC filters for the device address lists.
68  */
69 #define EFX_EF10_FILTER_FLAG_BUSY	1UL
70 #define EFX_EF10_FILTER_FLAG_AUTO_OLD	2UL
71 #define EFX_EF10_FILTER_FLAGS		3UL
72 		u64 handle;		/* firmware handle */
73 	} *entry;
74 	wait_queue_head_t waitq;
75 /* Shadow of net_device address lists, guarded by mac_lock */
76 #define EFX_EF10_FILTER_DEV_UC_MAX	32
77 #define EFX_EF10_FILTER_DEV_MC_MAX	256
78 	struct efx_ef10_dev_addr dev_uc_list[EFX_EF10_FILTER_DEV_UC_MAX];
79 	struct efx_ef10_dev_addr dev_mc_list[EFX_EF10_FILTER_DEV_MC_MAX];
80 	int dev_uc_count;
81 	int dev_mc_count;
82 /* Indices (like efx_ef10_dev_addr.id) for promisc/allmulti filters */
83 	u16 ucdef_id;
84 	u16 bcast_id;
85 	u16 mcdef_id;
86 };
87 
88 /* An arbitrary search limit for the software hash table */
89 #define EFX_EF10_FILTER_SEARCH_LIMIT 200
90 
91 static void efx_ef10_rx_free_indir_table(struct efx_nic *efx);
92 static void efx_ef10_filter_table_remove(struct efx_nic *efx);
93 
94 static int efx_ef10_get_warm_boot_count(struct efx_nic *efx)
95 {
96 	efx_dword_t reg;
97 
98 	efx_readd(efx, &reg, ER_DZ_BIU_MC_SFT_STATUS);
99 	return EFX_DWORD_FIELD(reg, EFX_WORD_1) == 0xb007 ?
100 		EFX_DWORD_FIELD(reg, EFX_WORD_0) : -EIO;
101 }
102 
103 static unsigned int efx_ef10_mem_map_size(struct efx_nic *efx)
104 {
105 	int bar;
106 
107 	bar = efx->type->mem_bar;
108 	return resource_size(&efx->pci_dev->resource[bar]);
109 }
110 
111 static bool efx_ef10_is_vf(struct efx_nic *efx)
112 {
113 	return efx->type->is_vf;
114 }
115 
116 static int efx_ef10_get_pf_index(struct efx_nic *efx)
117 {
118 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_FUNCTION_INFO_OUT_LEN);
119 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
120 	size_t outlen;
121 	int rc;
122 
123 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_FUNCTION_INFO, NULL, 0, outbuf,
124 			  sizeof(outbuf), &outlen);
125 	if (rc)
126 		return rc;
127 	if (outlen < sizeof(outbuf))
128 		return -EIO;
129 
130 	nic_data->pf_index = MCDI_DWORD(outbuf, GET_FUNCTION_INFO_OUT_PF);
131 	return 0;
132 }
133 
134 #ifdef CONFIG_SFC_SRIOV
135 static int efx_ef10_get_vf_index(struct efx_nic *efx)
136 {
137 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_FUNCTION_INFO_OUT_LEN);
138 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
139 	size_t outlen;
140 	int rc;
141 
142 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_FUNCTION_INFO, NULL, 0, outbuf,
143 			  sizeof(outbuf), &outlen);
144 	if (rc)
145 		return rc;
146 	if (outlen < sizeof(outbuf))
147 		return -EIO;
148 
149 	nic_data->vf_index = MCDI_DWORD(outbuf, GET_FUNCTION_INFO_OUT_VF);
150 	return 0;
151 }
152 #endif
153 
154 static int efx_ef10_init_datapath_caps(struct efx_nic *efx)
155 {
156 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_CAPABILITIES_OUT_LEN);
157 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
158 	size_t outlen;
159 	int rc;
160 
161 	BUILD_BUG_ON(MC_CMD_GET_CAPABILITIES_IN_LEN != 0);
162 
163 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_CAPABILITIES, NULL, 0,
164 			  outbuf, sizeof(outbuf), &outlen);
165 	if (rc)
166 		return rc;
167 	if (outlen < sizeof(outbuf)) {
168 		netif_err(efx, drv, efx->net_dev,
169 			  "unable to read datapath firmware capabilities\n");
170 		return -EIO;
171 	}
172 
173 	nic_data->datapath_caps =
174 		MCDI_DWORD(outbuf, GET_CAPABILITIES_OUT_FLAGS1);
175 
176 	/* record the DPCPU firmware IDs to determine VEB vswitching support.
177 	 */
178 	nic_data->rx_dpcpu_fw_id =
179 		MCDI_WORD(outbuf, GET_CAPABILITIES_OUT_RX_DPCPU_FW_ID);
180 	nic_data->tx_dpcpu_fw_id =
181 		MCDI_WORD(outbuf, GET_CAPABILITIES_OUT_TX_DPCPU_FW_ID);
182 
183 	if (!(nic_data->datapath_caps &
184 	      (1 << MC_CMD_GET_CAPABILITIES_OUT_TX_TSO_LBN))) {
185 		netif_err(efx, drv, efx->net_dev,
186 			  "current firmware does not support TSO\n");
187 		return -ENODEV;
188 	}
189 
190 	if (!(nic_data->datapath_caps &
191 	      (1 << MC_CMD_GET_CAPABILITIES_OUT_RX_PREFIX_LEN_14_LBN))) {
192 		netif_err(efx, probe, efx->net_dev,
193 			  "current firmware does not support an RX prefix\n");
194 		return -ENODEV;
195 	}
196 
197 	return 0;
198 }
199 
200 static int efx_ef10_get_sysclk_freq(struct efx_nic *efx)
201 {
202 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_CLOCK_OUT_LEN);
203 	int rc;
204 
205 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_CLOCK, NULL, 0,
206 			  outbuf, sizeof(outbuf), NULL);
207 	if (rc)
208 		return rc;
209 	rc = MCDI_DWORD(outbuf, GET_CLOCK_OUT_SYS_FREQ);
210 	return rc > 0 ? rc : -ERANGE;
211 }
212 
213 static int efx_ef10_get_mac_address_pf(struct efx_nic *efx, u8 *mac_address)
214 {
215 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_MAC_ADDRESSES_OUT_LEN);
216 	size_t outlen;
217 	int rc;
218 
219 	BUILD_BUG_ON(MC_CMD_GET_MAC_ADDRESSES_IN_LEN != 0);
220 
221 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_MAC_ADDRESSES, NULL, 0,
222 			  outbuf, sizeof(outbuf), &outlen);
223 	if (rc)
224 		return rc;
225 	if (outlen < MC_CMD_GET_MAC_ADDRESSES_OUT_LEN)
226 		return -EIO;
227 
228 	ether_addr_copy(mac_address,
229 			MCDI_PTR(outbuf, GET_MAC_ADDRESSES_OUT_MAC_ADDR_BASE));
230 	return 0;
231 }
232 
233 static int efx_ef10_get_mac_address_vf(struct efx_nic *efx, u8 *mac_address)
234 {
235 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VPORT_GET_MAC_ADDRESSES_IN_LEN);
236 	MCDI_DECLARE_BUF(outbuf, MC_CMD_VPORT_GET_MAC_ADDRESSES_OUT_LENMAX);
237 	size_t outlen;
238 	int num_addrs, rc;
239 
240 	MCDI_SET_DWORD(inbuf, VPORT_GET_MAC_ADDRESSES_IN_VPORT_ID,
241 		       EVB_PORT_ID_ASSIGNED);
242 	rc = efx_mcdi_rpc(efx, MC_CMD_VPORT_GET_MAC_ADDRESSES, inbuf,
243 			  sizeof(inbuf), outbuf, sizeof(outbuf), &outlen);
244 
245 	if (rc)
246 		return rc;
247 	if (outlen < MC_CMD_VPORT_GET_MAC_ADDRESSES_OUT_LENMIN)
248 		return -EIO;
249 
250 	num_addrs = MCDI_DWORD(outbuf,
251 			       VPORT_GET_MAC_ADDRESSES_OUT_MACADDR_COUNT);
252 
253 	WARN_ON(num_addrs != 1);
254 
255 	ether_addr_copy(mac_address,
256 			MCDI_PTR(outbuf, VPORT_GET_MAC_ADDRESSES_OUT_MACADDR));
257 
258 	return 0;
259 }
260 
261 static ssize_t efx_ef10_show_link_control_flag(struct device *dev,
262 					       struct device_attribute *attr,
263 					       char *buf)
264 {
265 	struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));
266 
267 	return sprintf(buf, "%d\n",
268 		       ((efx->mcdi->fn_flags) &
269 			(1 << MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_LINKCTRL))
270 		       ? 1 : 0);
271 }
272 
273 static ssize_t efx_ef10_show_primary_flag(struct device *dev,
274 					  struct device_attribute *attr,
275 					  char *buf)
276 {
277 	struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));
278 
279 	return sprintf(buf, "%d\n",
280 		       ((efx->mcdi->fn_flags) &
281 			(1 << MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_PRIMARY))
282 		       ? 1 : 0);
283 }
284 
285 static DEVICE_ATTR(link_control_flag, 0444, efx_ef10_show_link_control_flag,
286 		   NULL);
287 static DEVICE_ATTR(primary_flag, 0444, efx_ef10_show_primary_flag, NULL);
288 
289 static int efx_ef10_probe(struct efx_nic *efx)
290 {
291 	struct efx_ef10_nic_data *nic_data;
292 	struct net_device *net_dev = efx->net_dev;
293 	int i, rc;
294 
295 	/* We can have one VI for each 8K region.  However, until we
296 	 * use TX option descriptors we need two TX queues per channel.
297 	 */
298 	efx->max_channels = min_t(unsigned int,
299 				  EFX_MAX_CHANNELS,
300 				  efx_ef10_mem_map_size(efx) /
301 				  (EFX_VI_PAGE_SIZE * EFX_TXQ_TYPES));
302 	efx->max_tx_channels = efx->max_channels;
303 	if (WARN_ON(efx->max_channels == 0))
304 		return -EIO;
305 
306 	nic_data = kzalloc(sizeof(*nic_data), GFP_KERNEL);
307 	if (!nic_data)
308 		return -ENOMEM;
309 	efx->nic_data = nic_data;
310 
311 	/* we assume later that we can copy from this buffer in dwords */
312 	BUILD_BUG_ON(MCDI_CTL_SDU_LEN_MAX_V2 % 4);
313 
314 	rc = efx_nic_alloc_buffer(efx, &nic_data->mcdi_buf,
315 				  8 + MCDI_CTL_SDU_LEN_MAX_V2, GFP_KERNEL);
316 	if (rc)
317 		goto fail1;
318 
319 	/* Get the MC's warm boot count.  In case it's rebooting right
320 	 * now, be prepared to retry.
321 	 */
322 	i = 0;
323 	for (;;) {
324 		rc = efx_ef10_get_warm_boot_count(efx);
325 		if (rc >= 0)
326 			break;
327 		if (++i == 5)
328 			goto fail2;
329 		ssleep(1);
330 	}
331 	nic_data->warm_boot_count = rc;
332 
333 	nic_data->rx_rss_context = EFX_EF10_RSS_CONTEXT_INVALID;
334 
335 	nic_data->vport_id = EVB_PORT_ID_ASSIGNED;
336 
337 	/* In case we're recovering from a crash (kexec), we want to
338 	 * cancel any outstanding request by the previous user of this
339 	 * function.  We send a special message using the least
340 	 * significant bits of the 'high' (doorbell) register.
341 	 */
342 	_efx_writed(efx, cpu_to_le32(1), ER_DZ_MC_DB_HWRD);
343 
344 	rc = efx_mcdi_init(efx);
345 	if (rc)
346 		goto fail2;
347 
348 	/* Reset (most) configuration for this function */
349 	rc = efx_mcdi_reset(efx, RESET_TYPE_ALL);
350 	if (rc)
351 		goto fail3;
352 
353 	/* Enable event logging */
354 	rc = efx_mcdi_log_ctrl(efx, true, false, 0);
355 	if (rc)
356 		goto fail3;
357 
358 	rc = device_create_file(&efx->pci_dev->dev,
359 				&dev_attr_link_control_flag);
360 	if (rc)
361 		goto fail3;
362 
363 	rc = device_create_file(&efx->pci_dev->dev, &dev_attr_primary_flag);
364 	if (rc)
365 		goto fail4;
366 
367 	rc = efx_ef10_get_pf_index(efx);
368 	if (rc)
369 		goto fail5;
370 
371 	rc = efx_ef10_init_datapath_caps(efx);
372 	if (rc < 0)
373 		goto fail5;
374 
375 	efx->rx_packet_len_offset =
376 		ES_DZ_RX_PREFIX_PKTLEN_OFST - ES_DZ_RX_PREFIX_SIZE;
377 
378 	rc = efx_mcdi_port_get_number(efx);
379 	if (rc < 0)
380 		goto fail5;
381 	efx->port_num = rc;
382 	net_dev->dev_port = rc;
383 
384 	rc = efx->type->get_mac_address(efx, efx->net_dev->perm_addr);
385 	if (rc)
386 		goto fail5;
387 
388 	rc = efx_ef10_get_sysclk_freq(efx);
389 	if (rc < 0)
390 		goto fail5;
391 	efx->timer_quantum_ns = 1536000 / rc; /* 1536 cycles */
392 
393 	/* Check whether firmware supports bug 35388 workaround.
394 	 * First try to enable it, then if we get EPERM, just
395 	 * ask if it's already enabled
396 	 */
397 	rc = efx_mcdi_set_workaround(efx, MC_CMD_WORKAROUND_BUG35388, true, NULL);
398 	if (rc == 0) {
399 		nic_data->workaround_35388 = true;
400 	} else if (rc == -EPERM) {
401 		unsigned int enabled;
402 
403 		rc = efx_mcdi_get_workarounds(efx, NULL, &enabled);
404 		if (rc)
405 			goto fail3;
406 		nic_data->workaround_35388 = enabled &
407 			MC_CMD_GET_WORKAROUNDS_OUT_BUG35388;
408 	} else if (rc != -ENOSYS && rc != -ENOENT) {
409 		goto fail5;
410 	}
411 	netif_dbg(efx, probe, efx->net_dev,
412 		  "workaround for bug 35388 is %sabled\n",
413 		  nic_data->workaround_35388 ? "en" : "dis");
414 
415 	rc = efx_mcdi_mon_probe(efx);
416 	if (rc && rc != -EPERM)
417 		goto fail5;
418 
419 	efx_ptp_probe(efx, NULL);
420 
421 #ifdef CONFIG_SFC_SRIOV
422 	if ((efx->pci_dev->physfn) && (!efx->pci_dev->is_physfn)) {
423 		struct pci_dev *pci_dev_pf = efx->pci_dev->physfn;
424 		struct efx_nic *efx_pf = pci_get_drvdata(pci_dev_pf);
425 
426 		efx_pf->type->get_mac_address(efx_pf, nic_data->port_id);
427 	} else
428 #endif
429 		ether_addr_copy(nic_data->port_id, efx->net_dev->perm_addr);
430 
431 	return 0;
432 
433 fail5:
434 	device_remove_file(&efx->pci_dev->dev, &dev_attr_primary_flag);
435 fail4:
436 	device_remove_file(&efx->pci_dev->dev, &dev_attr_link_control_flag);
437 fail3:
438 	efx_mcdi_fini(efx);
439 fail2:
440 	efx_nic_free_buffer(efx, &nic_data->mcdi_buf);
441 fail1:
442 	kfree(nic_data);
443 	efx->nic_data = NULL;
444 	return rc;
445 }
446 
447 static int efx_ef10_free_vis(struct efx_nic *efx)
448 {
449 	MCDI_DECLARE_BUF_ERR(outbuf);
450 	size_t outlen;
451 	int rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FREE_VIS, NULL, 0,
452 				    outbuf, sizeof(outbuf), &outlen);
453 
454 	/* -EALREADY means nothing to free, so ignore */
455 	if (rc == -EALREADY)
456 		rc = 0;
457 	if (rc)
458 		efx_mcdi_display_error(efx, MC_CMD_FREE_VIS, 0, outbuf, outlen,
459 				       rc);
460 	return rc;
461 }
462 
463 #ifdef EFX_USE_PIO
464 
465 static void efx_ef10_free_piobufs(struct efx_nic *efx)
466 {
467 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
468 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FREE_PIOBUF_IN_LEN);
469 	unsigned int i;
470 	int rc;
471 
472 	BUILD_BUG_ON(MC_CMD_FREE_PIOBUF_OUT_LEN != 0);
473 
474 	for (i = 0; i < nic_data->n_piobufs; i++) {
475 		MCDI_SET_DWORD(inbuf, FREE_PIOBUF_IN_PIOBUF_HANDLE,
476 			       nic_data->piobuf_handle[i]);
477 		rc = efx_mcdi_rpc(efx, MC_CMD_FREE_PIOBUF, inbuf, sizeof(inbuf),
478 				  NULL, 0, NULL);
479 		WARN_ON(rc);
480 	}
481 
482 	nic_data->n_piobufs = 0;
483 }
484 
485 static int efx_ef10_alloc_piobufs(struct efx_nic *efx, unsigned int n)
486 {
487 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
488 	MCDI_DECLARE_BUF(outbuf, MC_CMD_ALLOC_PIOBUF_OUT_LEN);
489 	unsigned int i;
490 	size_t outlen;
491 	int rc = 0;
492 
493 	BUILD_BUG_ON(MC_CMD_ALLOC_PIOBUF_IN_LEN != 0);
494 
495 	for (i = 0; i < n; i++) {
496 		rc = efx_mcdi_rpc(efx, MC_CMD_ALLOC_PIOBUF, NULL, 0,
497 				  outbuf, sizeof(outbuf), &outlen);
498 		if (rc)
499 			break;
500 		if (outlen < MC_CMD_ALLOC_PIOBUF_OUT_LEN) {
501 			rc = -EIO;
502 			break;
503 		}
504 		nic_data->piobuf_handle[i] =
505 			MCDI_DWORD(outbuf, ALLOC_PIOBUF_OUT_PIOBUF_HANDLE);
506 		netif_dbg(efx, probe, efx->net_dev,
507 			  "allocated PIO buffer %u handle %x\n", i,
508 			  nic_data->piobuf_handle[i]);
509 	}
510 
511 	nic_data->n_piobufs = i;
512 	if (rc)
513 		efx_ef10_free_piobufs(efx);
514 	return rc;
515 }
516 
517 static int efx_ef10_link_piobufs(struct efx_nic *efx)
518 {
519 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
520 	_MCDI_DECLARE_BUF(inbuf,
521 			  max(MC_CMD_LINK_PIOBUF_IN_LEN,
522 			      MC_CMD_UNLINK_PIOBUF_IN_LEN));
523 	struct efx_channel *channel;
524 	struct efx_tx_queue *tx_queue;
525 	unsigned int offset, index;
526 	int rc;
527 
528 	BUILD_BUG_ON(MC_CMD_LINK_PIOBUF_OUT_LEN != 0);
529 	BUILD_BUG_ON(MC_CMD_UNLINK_PIOBUF_OUT_LEN != 0);
530 
531 	memset(inbuf, 0, sizeof(inbuf));
532 
533 	/* Link a buffer to each VI in the write-combining mapping */
534 	for (index = 0; index < nic_data->n_piobufs; ++index) {
535 		MCDI_SET_DWORD(inbuf, LINK_PIOBUF_IN_PIOBUF_HANDLE,
536 			       nic_data->piobuf_handle[index]);
537 		MCDI_SET_DWORD(inbuf, LINK_PIOBUF_IN_TXQ_INSTANCE,
538 			       nic_data->pio_write_vi_base + index);
539 		rc = efx_mcdi_rpc(efx, MC_CMD_LINK_PIOBUF,
540 				  inbuf, MC_CMD_LINK_PIOBUF_IN_LEN,
541 				  NULL, 0, NULL);
542 		if (rc) {
543 			netif_err(efx, drv, efx->net_dev,
544 				  "failed to link VI %u to PIO buffer %u (%d)\n",
545 				  nic_data->pio_write_vi_base + index, index,
546 				  rc);
547 			goto fail;
548 		}
549 		netif_dbg(efx, probe, efx->net_dev,
550 			  "linked VI %u to PIO buffer %u\n",
551 			  nic_data->pio_write_vi_base + index, index);
552 	}
553 
554 	/* Link a buffer to each TX queue */
555 	efx_for_each_channel(channel, efx) {
556 		efx_for_each_channel_tx_queue(tx_queue, channel) {
557 			/* We assign the PIO buffers to queues in
558 			 * reverse order to allow for the following
559 			 * special case.
560 			 */
561 			offset = ((efx->tx_channel_offset + efx->n_tx_channels -
562 				   tx_queue->channel->channel - 1) *
563 				  efx_piobuf_size);
564 			index = offset / ER_DZ_TX_PIOBUF_SIZE;
565 			offset = offset % ER_DZ_TX_PIOBUF_SIZE;
566 
567 			/* When the host page size is 4K, the first
568 			 * host page in the WC mapping may be within
569 			 * the same VI page as the last TX queue.  We
570 			 * can only link one buffer to each VI.
571 			 */
572 			if (tx_queue->queue == nic_data->pio_write_vi_base) {
573 				BUG_ON(index != 0);
574 				rc = 0;
575 			} else {
576 				MCDI_SET_DWORD(inbuf,
577 					       LINK_PIOBUF_IN_PIOBUF_HANDLE,
578 					       nic_data->piobuf_handle[index]);
579 				MCDI_SET_DWORD(inbuf,
580 					       LINK_PIOBUF_IN_TXQ_INSTANCE,
581 					       tx_queue->queue);
582 				rc = efx_mcdi_rpc(efx, MC_CMD_LINK_PIOBUF,
583 						  inbuf, MC_CMD_LINK_PIOBUF_IN_LEN,
584 						  NULL, 0, NULL);
585 			}
586 
587 			if (rc) {
588 				/* This is non-fatal; the TX path just
589 				 * won't use PIO for this queue
590 				 */
591 				netif_err(efx, drv, efx->net_dev,
592 					  "failed to link VI %u to PIO buffer %u (%d)\n",
593 					  tx_queue->queue, index, rc);
594 				tx_queue->piobuf = NULL;
595 			} else {
596 				tx_queue->piobuf =
597 					nic_data->pio_write_base +
598 					index * EFX_VI_PAGE_SIZE + offset;
599 				tx_queue->piobuf_offset = offset;
600 				netif_dbg(efx, probe, efx->net_dev,
601 					  "linked VI %u to PIO buffer %u offset %x addr %p\n",
602 					  tx_queue->queue, index,
603 					  tx_queue->piobuf_offset,
604 					  tx_queue->piobuf);
605 			}
606 		}
607 	}
608 
609 	return 0;
610 
611 fail:
612 	while (index--) {
613 		MCDI_SET_DWORD(inbuf, UNLINK_PIOBUF_IN_TXQ_INSTANCE,
614 			       nic_data->pio_write_vi_base + index);
615 		efx_mcdi_rpc(efx, MC_CMD_UNLINK_PIOBUF,
616 			     inbuf, MC_CMD_UNLINK_PIOBUF_IN_LEN,
617 			     NULL, 0, NULL);
618 	}
619 	return rc;
620 }
621 
622 #else /* !EFX_USE_PIO */
623 
624 static int efx_ef10_alloc_piobufs(struct efx_nic *efx, unsigned int n)
625 {
626 	return n == 0 ? 0 : -ENOBUFS;
627 }
628 
629 static int efx_ef10_link_piobufs(struct efx_nic *efx)
630 {
631 	return 0;
632 }
633 
634 static void efx_ef10_free_piobufs(struct efx_nic *efx)
635 {
636 }
637 
638 #endif /* EFX_USE_PIO */
639 
640 static void efx_ef10_remove(struct efx_nic *efx)
641 {
642 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
643 	int rc;
644 
645 #ifdef CONFIG_SFC_SRIOV
646 	struct efx_ef10_nic_data *nic_data_pf;
647 	struct pci_dev *pci_dev_pf;
648 	struct efx_nic *efx_pf;
649 	struct ef10_vf *vf;
650 
651 	if (efx->pci_dev->is_virtfn) {
652 		pci_dev_pf = efx->pci_dev->physfn;
653 		if (pci_dev_pf) {
654 			efx_pf = pci_get_drvdata(pci_dev_pf);
655 			nic_data_pf = efx_pf->nic_data;
656 			vf = nic_data_pf->vf + nic_data->vf_index;
657 			vf->efx = NULL;
658 		} else
659 			netif_info(efx, drv, efx->net_dev,
660 				   "Could not get the PF id from VF\n");
661 	}
662 #endif
663 
664 	efx_ptp_remove(efx);
665 
666 	efx_mcdi_mon_remove(efx);
667 
668 	efx_ef10_rx_free_indir_table(efx);
669 
670 	if (nic_data->wc_membase)
671 		iounmap(nic_data->wc_membase);
672 
673 	rc = efx_ef10_free_vis(efx);
674 	WARN_ON(rc != 0);
675 
676 	if (!nic_data->must_restore_piobufs)
677 		efx_ef10_free_piobufs(efx);
678 
679 	device_remove_file(&efx->pci_dev->dev, &dev_attr_primary_flag);
680 	device_remove_file(&efx->pci_dev->dev, &dev_attr_link_control_flag);
681 
682 	efx_mcdi_fini(efx);
683 	efx_nic_free_buffer(efx, &nic_data->mcdi_buf);
684 	kfree(nic_data);
685 }
686 
687 static int efx_ef10_probe_pf(struct efx_nic *efx)
688 {
689 	return efx_ef10_probe(efx);
690 }
691 
692 int efx_ef10_vadaptor_alloc(struct efx_nic *efx, unsigned int port_id)
693 {
694 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VADAPTOR_ALLOC_IN_LEN);
695 
696 	MCDI_SET_DWORD(inbuf, VADAPTOR_ALLOC_IN_UPSTREAM_PORT_ID, port_id);
697 	return efx_mcdi_rpc(efx, MC_CMD_VADAPTOR_ALLOC, inbuf, sizeof(inbuf),
698 			    NULL, 0, NULL);
699 }
700 
701 int efx_ef10_vadaptor_free(struct efx_nic *efx, unsigned int port_id)
702 {
703 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VADAPTOR_FREE_IN_LEN);
704 
705 	MCDI_SET_DWORD(inbuf, VADAPTOR_FREE_IN_UPSTREAM_PORT_ID, port_id);
706 	return efx_mcdi_rpc(efx, MC_CMD_VADAPTOR_FREE, inbuf, sizeof(inbuf),
707 			    NULL, 0, NULL);
708 }
709 
710 int efx_ef10_vport_add_mac(struct efx_nic *efx,
711 			   unsigned int port_id, u8 *mac)
712 {
713 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VPORT_ADD_MAC_ADDRESS_IN_LEN);
714 
715 	MCDI_SET_DWORD(inbuf, VPORT_ADD_MAC_ADDRESS_IN_VPORT_ID, port_id);
716 	ether_addr_copy(MCDI_PTR(inbuf, VPORT_ADD_MAC_ADDRESS_IN_MACADDR), mac);
717 
718 	return efx_mcdi_rpc(efx, MC_CMD_VPORT_ADD_MAC_ADDRESS, inbuf,
719 			    sizeof(inbuf), NULL, 0, NULL);
720 }
721 
722 int efx_ef10_vport_del_mac(struct efx_nic *efx,
723 			   unsigned int port_id, u8 *mac)
724 {
725 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VPORT_DEL_MAC_ADDRESS_IN_LEN);
726 
727 	MCDI_SET_DWORD(inbuf, VPORT_DEL_MAC_ADDRESS_IN_VPORT_ID, port_id);
728 	ether_addr_copy(MCDI_PTR(inbuf, VPORT_DEL_MAC_ADDRESS_IN_MACADDR), mac);
729 
730 	return efx_mcdi_rpc(efx, MC_CMD_VPORT_DEL_MAC_ADDRESS, inbuf,
731 			    sizeof(inbuf), NULL, 0, NULL);
732 }
733 
734 #ifdef CONFIG_SFC_SRIOV
735 static int efx_ef10_probe_vf(struct efx_nic *efx)
736 {
737 	int rc;
738 	struct pci_dev *pci_dev_pf;
739 
740 	/* If the parent PF has no VF data structure, it doesn't know about this
741 	 * VF so fail probe.  The VF needs to be re-created.  This can happen
742 	 * if the PF driver is unloaded while the VF is assigned to a guest.
743 	 */
744 	pci_dev_pf = efx->pci_dev->physfn;
745 	if (pci_dev_pf) {
746 		struct efx_nic *efx_pf = pci_get_drvdata(pci_dev_pf);
747 		struct efx_ef10_nic_data *nic_data_pf = efx_pf->nic_data;
748 
749 		if (!nic_data_pf->vf) {
750 			netif_info(efx, drv, efx->net_dev,
751 				   "The VF cannot link to its parent PF; "
752 				   "please destroy and re-create the VF\n");
753 			return -EBUSY;
754 		}
755 	}
756 
757 	rc = efx_ef10_probe(efx);
758 	if (rc)
759 		return rc;
760 
761 	rc = efx_ef10_get_vf_index(efx);
762 	if (rc)
763 		goto fail;
764 
765 	if (efx->pci_dev->is_virtfn) {
766 		if (efx->pci_dev->physfn) {
767 			struct efx_nic *efx_pf =
768 				pci_get_drvdata(efx->pci_dev->physfn);
769 			struct efx_ef10_nic_data *nic_data_p = efx_pf->nic_data;
770 			struct efx_ef10_nic_data *nic_data = efx->nic_data;
771 
772 			nic_data_p->vf[nic_data->vf_index].efx = efx;
773 			nic_data_p->vf[nic_data->vf_index].pci_dev =
774 				efx->pci_dev;
775 		} else
776 			netif_info(efx, drv, efx->net_dev,
777 				   "Could not get the PF id from VF\n");
778 	}
779 
780 	return 0;
781 
782 fail:
783 	efx_ef10_remove(efx);
784 	return rc;
785 }
786 #else
787 static int efx_ef10_probe_vf(struct efx_nic *efx __attribute__ ((unused)))
788 {
789 	return 0;
790 }
791 #endif
792 
793 static int efx_ef10_alloc_vis(struct efx_nic *efx,
794 			      unsigned int min_vis, unsigned int max_vis)
795 {
796 	MCDI_DECLARE_BUF(inbuf, MC_CMD_ALLOC_VIS_IN_LEN);
797 	MCDI_DECLARE_BUF(outbuf, MC_CMD_ALLOC_VIS_OUT_LEN);
798 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
799 	size_t outlen;
800 	int rc;
801 
802 	MCDI_SET_DWORD(inbuf, ALLOC_VIS_IN_MIN_VI_COUNT, min_vis);
803 	MCDI_SET_DWORD(inbuf, ALLOC_VIS_IN_MAX_VI_COUNT, max_vis);
804 	rc = efx_mcdi_rpc(efx, MC_CMD_ALLOC_VIS, inbuf, sizeof(inbuf),
805 			  outbuf, sizeof(outbuf), &outlen);
806 	if (rc != 0)
807 		return rc;
808 
809 	if (outlen < MC_CMD_ALLOC_VIS_OUT_LEN)
810 		return -EIO;
811 
812 	netif_dbg(efx, drv, efx->net_dev, "base VI is A0x%03x\n",
813 		  MCDI_DWORD(outbuf, ALLOC_VIS_OUT_VI_BASE));
814 
815 	nic_data->vi_base = MCDI_DWORD(outbuf, ALLOC_VIS_OUT_VI_BASE);
816 	nic_data->n_allocated_vis = MCDI_DWORD(outbuf, ALLOC_VIS_OUT_VI_COUNT);
817 	return 0;
818 }
819 
820 /* Note that the failure path of this function does not free
821  * resources, as this will be done by efx_ef10_remove().
822  */
823 static int efx_ef10_dimension_resources(struct efx_nic *efx)
824 {
825 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
826 	unsigned int uc_mem_map_size, wc_mem_map_size;
827 	unsigned int min_vis = max(EFX_TXQ_TYPES,
828 				   efx_separate_tx_channels ? 2 : 1);
829 	unsigned int channel_vis, pio_write_vi_base, max_vis;
830 	void __iomem *membase;
831 	int rc;
832 
833 	channel_vis = max(efx->n_channels, efx->n_tx_channels * EFX_TXQ_TYPES);
834 
835 #ifdef EFX_USE_PIO
836 	/* Try to allocate PIO buffers if wanted and if the full
837 	 * number of PIO buffers would be sufficient to allocate one
838 	 * copy-buffer per TX channel.  Failure is non-fatal, as there
839 	 * are only a small number of PIO buffers shared between all
840 	 * functions of the controller.
841 	 */
842 	if (efx_piobuf_size != 0 &&
843 	    ER_DZ_TX_PIOBUF_SIZE / efx_piobuf_size * EF10_TX_PIOBUF_COUNT >=
844 	    efx->n_tx_channels) {
845 		unsigned int n_piobufs =
846 			DIV_ROUND_UP(efx->n_tx_channels,
847 				     ER_DZ_TX_PIOBUF_SIZE / efx_piobuf_size);
848 
849 		rc = efx_ef10_alloc_piobufs(efx, n_piobufs);
850 		if (rc)
851 			netif_err(efx, probe, efx->net_dev,
852 				  "failed to allocate PIO buffers (%d)\n", rc);
853 		else
854 			netif_dbg(efx, probe, efx->net_dev,
855 				  "allocated %u PIO buffers\n", n_piobufs);
856 	}
857 #else
858 	nic_data->n_piobufs = 0;
859 #endif
860 
861 	/* PIO buffers should be mapped with write-combining enabled,
862 	 * and we want to make single UC and WC mappings rather than
863 	 * several of each (in fact that's the only option if host
864 	 * page size is >4K).  So we may allocate some extra VIs just
865 	 * for writing PIO buffers through.
866 	 *
867 	 * The UC mapping contains (channel_vis - 1) complete VIs and the
868 	 * first half of the next VI.  Then the WC mapping begins with
869 	 * the second half of this last VI.
870 	 */
871 	uc_mem_map_size = PAGE_ALIGN((channel_vis - 1) * EFX_VI_PAGE_SIZE +
872 				     ER_DZ_TX_PIOBUF);
873 	if (nic_data->n_piobufs) {
874 		/* pio_write_vi_base rounds down to give the number of complete
875 		 * VIs inside the UC mapping.
876 		 */
877 		pio_write_vi_base = uc_mem_map_size / EFX_VI_PAGE_SIZE;
878 		wc_mem_map_size = (PAGE_ALIGN((pio_write_vi_base +
879 					       nic_data->n_piobufs) *
880 					      EFX_VI_PAGE_SIZE) -
881 				   uc_mem_map_size);
882 		max_vis = pio_write_vi_base + nic_data->n_piobufs;
883 	} else {
884 		pio_write_vi_base = 0;
885 		wc_mem_map_size = 0;
886 		max_vis = channel_vis;
887 	}
888 
889 	/* In case the last attached driver failed to free VIs, do it now */
890 	rc = efx_ef10_free_vis(efx);
891 	if (rc != 0)
892 		return rc;
893 
894 	rc = efx_ef10_alloc_vis(efx, min_vis, max_vis);
895 	if (rc != 0)
896 		return rc;
897 
898 	if (nic_data->n_allocated_vis < channel_vis) {
899 		netif_info(efx, drv, efx->net_dev,
900 			   "Could not allocate enough VIs to satisfy RSS"
901 			   " requirements. Performance may not be optimal.\n");
902 		/* We didn't get the VIs to populate our channels.
903 		 * We could keep what we got but then we'd have more
904 		 * interrupts than we need.
905 		 * Instead calculate new max_channels and restart
906 		 */
907 		efx->max_channels = nic_data->n_allocated_vis;
908 		efx->max_tx_channels =
909 			nic_data->n_allocated_vis / EFX_TXQ_TYPES;
910 
911 		efx_ef10_free_vis(efx);
912 		return -EAGAIN;
913 	}
914 
915 	/* If we didn't get enough VIs to map all the PIO buffers, free the
916 	 * PIO buffers
917 	 */
918 	if (nic_data->n_piobufs &&
919 	    nic_data->n_allocated_vis <
920 	    pio_write_vi_base + nic_data->n_piobufs) {
921 		netif_dbg(efx, probe, efx->net_dev,
922 			  "%u VIs are not sufficient to map %u PIO buffers\n",
923 			  nic_data->n_allocated_vis, nic_data->n_piobufs);
924 		efx_ef10_free_piobufs(efx);
925 	}
926 
927 	/* Shrink the original UC mapping of the memory BAR */
928 	membase = ioremap_nocache(efx->membase_phys, uc_mem_map_size);
929 	if (!membase) {
930 		netif_err(efx, probe, efx->net_dev,
931 			  "could not shrink memory BAR to %x\n",
932 			  uc_mem_map_size);
933 		return -ENOMEM;
934 	}
935 	iounmap(efx->membase);
936 	efx->membase = membase;
937 
938 	/* Set up the WC mapping if needed */
939 	if (wc_mem_map_size) {
940 		nic_data->wc_membase = ioremap_wc(efx->membase_phys +
941 						  uc_mem_map_size,
942 						  wc_mem_map_size);
943 		if (!nic_data->wc_membase) {
944 			netif_err(efx, probe, efx->net_dev,
945 				  "could not allocate WC mapping of size %x\n",
946 				  wc_mem_map_size);
947 			return -ENOMEM;
948 		}
949 		nic_data->pio_write_vi_base = pio_write_vi_base;
950 		nic_data->pio_write_base =
951 			nic_data->wc_membase +
952 			(pio_write_vi_base * EFX_VI_PAGE_SIZE + ER_DZ_TX_PIOBUF -
953 			 uc_mem_map_size);
954 
955 		rc = efx_ef10_link_piobufs(efx);
956 		if (rc)
957 			efx_ef10_free_piobufs(efx);
958 	}
959 
960 	netif_dbg(efx, probe, efx->net_dev,
961 		  "memory BAR at %pa (virtual %p+%x UC, %p+%x WC)\n",
962 		  &efx->membase_phys, efx->membase, uc_mem_map_size,
963 		  nic_data->wc_membase, wc_mem_map_size);
964 
965 	return 0;
966 }
967 
968 static int efx_ef10_init_nic(struct efx_nic *efx)
969 {
970 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
971 	int rc;
972 
973 	if (nic_data->must_check_datapath_caps) {
974 		rc = efx_ef10_init_datapath_caps(efx);
975 		if (rc)
976 			return rc;
977 		nic_data->must_check_datapath_caps = false;
978 	}
979 
980 	if (nic_data->must_realloc_vis) {
981 		/* We cannot let the number of VIs change now */
982 		rc = efx_ef10_alloc_vis(efx, nic_data->n_allocated_vis,
983 					nic_data->n_allocated_vis);
984 		if (rc)
985 			return rc;
986 		nic_data->must_realloc_vis = false;
987 	}
988 
989 	if (nic_data->must_restore_piobufs && nic_data->n_piobufs) {
990 		rc = efx_ef10_alloc_piobufs(efx, nic_data->n_piobufs);
991 		if (rc == 0) {
992 			rc = efx_ef10_link_piobufs(efx);
993 			if (rc)
994 				efx_ef10_free_piobufs(efx);
995 		}
996 
997 		/* Log an error on failure, but this is non-fatal */
998 		if (rc)
999 			netif_err(efx, drv, efx->net_dev,
1000 				  "failed to restore PIO buffers (%d)\n", rc);
1001 		nic_data->must_restore_piobufs = false;
1002 	}
1003 
1004 	/* don't fail init if RSS setup doesn't work */
1005 	efx->type->rx_push_rss_config(efx, false, efx->rx_indir_table);
1006 
1007 	return 0;
1008 }
1009 
1010 static void efx_ef10_reset_mc_allocations(struct efx_nic *efx)
1011 {
1012 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1013 #ifdef CONFIG_SFC_SRIOV
1014 	unsigned int i;
1015 #endif
1016 
1017 	/* All our allocations have been reset */
1018 	nic_data->must_realloc_vis = true;
1019 	nic_data->must_restore_filters = true;
1020 	nic_data->must_restore_piobufs = true;
1021 	nic_data->rx_rss_context = EFX_EF10_RSS_CONTEXT_INVALID;
1022 
1023 	/* Driver-created vswitches and vports must be re-created */
1024 	nic_data->must_probe_vswitching = true;
1025 	nic_data->vport_id = EVB_PORT_ID_ASSIGNED;
1026 #ifdef CONFIG_SFC_SRIOV
1027 	if (nic_data->vf)
1028 		for (i = 0; i < efx->vf_count; i++)
1029 			nic_data->vf[i].vport_id = 0;
1030 #endif
1031 }
1032 
1033 static enum reset_type efx_ef10_map_reset_reason(enum reset_type reason)
1034 {
1035 	if (reason == RESET_TYPE_MC_FAILURE)
1036 		return RESET_TYPE_DATAPATH;
1037 
1038 	return efx_mcdi_map_reset_reason(reason);
1039 }
1040 
1041 static int efx_ef10_map_reset_flags(u32 *flags)
1042 {
1043 	enum {
1044 		EF10_RESET_PORT = ((ETH_RESET_MAC | ETH_RESET_PHY) <<
1045 				   ETH_RESET_SHARED_SHIFT),
1046 		EF10_RESET_MC = ((ETH_RESET_DMA | ETH_RESET_FILTER |
1047 				  ETH_RESET_OFFLOAD | ETH_RESET_MAC |
1048 				  ETH_RESET_PHY | ETH_RESET_MGMT) <<
1049 				 ETH_RESET_SHARED_SHIFT)
1050 	};
1051 
1052 	/* We assume for now that our PCI function is permitted to
1053 	 * reset everything.
1054 	 */
1055 
1056 	if ((*flags & EF10_RESET_MC) == EF10_RESET_MC) {
1057 		*flags &= ~EF10_RESET_MC;
1058 		return RESET_TYPE_WORLD;
1059 	}
1060 
1061 	if ((*flags & EF10_RESET_PORT) == EF10_RESET_PORT) {
1062 		*flags &= ~EF10_RESET_PORT;
1063 		return RESET_TYPE_ALL;
1064 	}
1065 
1066 	/* no invisible reset implemented */
1067 
1068 	return -EINVAL;
1069 }
1070 
1071 static int efx_ef10_reset(struct efx_nic *efx, enum reset_type reset_type)
1072 {
1073 	int rc = efx_mcdi_reset(efx, reset_type);
1074 
1075 	/* Unprivileged functions return -EPERM, but need to return success
1076 	 * here so that the datapath is brought back up.
1077 	 */
1078 	if (reset_type == RESET_TYPE_WORLD && rc == -EPERM)
1079 		rc = 0;
1080 
1081 	/* If it was a port reset, trigger reallocation of MC resources.
1082 	 * Note that on an MC reset nothing needs to be done now because we'll
1083 	 * detect the MC reset later and handle it then.
1084 	 * For an FLR, we never get an MC reset event, but the MC has reset all
1085 	 * resources assigned to us, so we have to trigger reallocation now.
1086 	 */
1087 	if ((reset_type == RESET_TYPE_ALL ||
1088 	     reset_type == RESET_TYPE_MCDI_TIMEOUT) && !rc)
1089 		efx_ef10_reset_mc_allocations(efx);
1090 	return rc;
1091 }
1092 
1093 #define EF10_DMA_STAT(ext_name, mcdi_name)			\
1094 	[EF10_STAT_ ## ext_name] =				\
1095 	{ #ext_name, 64, 8 * MC_CMD_MAC_ ## mcdi_name }
1096 #define EF10_DMA_INVIS_STAT(int_name, mcdi_name)		\
1097 	[EF10_STAT_ ## int_name] =				\
1098 	{ NULL, 64, 8 * MC_CMD_MAC_ ## mcdi_name }
1099 #define EF10_OTHER_STAT(ext_name)				\
1100 	[EF10_STAT_ ## ext_name] = { #ext_name, 0, 0 }
1101 #define GENERIC_SW_STAT(ext_name)				\
1102 	[GENERIC_STAT_ ## ext_name] = { #ext_name, 0, 0 }
1103 
1104 static const struct efx_hw_stat_desc efx_ef10_stat_desc[EF10_STAT_COUNT] = {
1105 	EF10_DMA_STAT(port_tx_bytes, TX_BYTES),
1106 	EF10_DMA_STAT(port_tx_packets, TX_PKTS),
1107 	EF10_DMA_STAT(port_tx_pause, TX_PAUSE_PKTS),
1108 	EF10_DMA_STAT(port_tx_control, TX_CONTROL_PKTS),
1109 	EF10_DMA_STAT(port_tx_unicast, TX_UNICAST_PKTS),
1110 	EF10_DMA_STAT(port_tx_multicast, TX_MULTICAST_PKTS),
1111 	EF10_DMA_STAT(port_tx_broadcast, TX_BROADCAST_PKTS),
1112 	EF10_DMA_STAT(port_tx_lt64, TX_LT64_PKTS),
1113 	EF10_DMA_STAT(port_tx_64, TX_64_PKTS),
1114 	EF10_DMA_STAT(port_tx_65_to_127, TX_65_TO_127_PKTS),
1115 	EF10_DMA_STAT(port_tx_128_to_255, TX_128_TO_255_PKTS),
1116 	EF10_DMA_STAT(port_tx_256_to_511, TX_256_TO_511_PKTS),
1117 	EF10_DMA_STAT(port_tx_512_to_1023, TX_512_TO_1023_PKTS),
1118 	EF10_DMA_STAT(port_tx_1024_to_15xx, TX_1024_TO_15XX_PKTS),
1119 	EF10_DMA_STAT(port_tx_15xx_to_jumbo, TX_15XX_TO_JUMBO_PKTS),
1120 	EF10_DMA_STAT(port_rx_bytes, RX_BYTES),
1121 	EF10_DMA_INVIS_STAT(port_rx_bytes_minus_good_bytes, RX_BAD_BYTES),
1122 	EF10_OTHER_STAT(port_rx_good_bytes),
1123 	EF10_OTHER_STAT(port_rx_bad_bytes),
1124 	EF10_DMA_STAT(port_rx_packets, RX_PKTS),
1125 	EF10_DMA_STAT(port_rx_good, RX_GOOD_PKTS),
1126 	EF10_DMA_STAT(port_rx_bad, RX_BAD_FCS_PKTS),
1127 	EF10_DMA_STAT(port_rx_pause, RX_PAUSE_PKTS),
1128 	EF10_DMA_STAT(port_rx_control, RX_CONTROL_PKTS),
1129 	EF10_DMA_STAT(port_rx_unicast, RX_UNICAST_PKTS),
1130 	EF10_DMA_STAT(port_rx_multicast, RX_MULTICAST_PKTS),
1131 	EF10_DMA_STAT(port_rx_broadcast, RX_BROADCAST_PKTS),
1132 	EF10_DMA_STAT(port_rx_lt64, RX_UNDERSIZE_PKTS),
1133 	EF10_DMA_STAT(port_rx_64, RX_64_PKTS),
1134 	EF10_DMA_STAT(port_rx_65_to_127, RX_65_TO_127_PKTS),
1135 	EF10_DMA_STAT(port_rx_128_to_255, RX_128_TO_255_PKTS),
1136 	EF10_DMA_STAT(port_rx_256_to_511, RX_256_TO_511_PKTS),
1137 	EF10_DMA_STAT(port_rx_512_to_1023, RX_512_TO_1023_PKTS),
1138 	EF10_DMA_STAT(port_rx_1024_to_15xx, RX_1024_TO_15XX_PKTS),
1139 	EF10_DMA_STAT(port_rx_15xx_to_jumbo, RX_15XX_TO_JUMBO_PKTS),
1140 	EF10_DMA_STAT(port_rx_gtjumbo, RX_GTJUMBO_PKTS),
1141 	EF10_DMA_STAT(port_rx_bad_gtjumbo, RX_JABBER_PKTS),
1142 	EF10_DMA_STAT(port_rx_overflow, RX_OVERFLOW_PKTS),
1143 	EF10_DMA_STAT(port_rx_align_error, RX_ALIGN_ERROR_PKTS),
1144 	EF10_DMA_STAT(port_rx_length_error, RX_LENGTH_ERROR_PKTS),
1145 	EF10_DMA_STAT(port_rx_nodesc_drops, RX_NODESC_DROPS),
1146 	GENERIC_SW_STAT(rx_nodesc_trunc),
1147 	GENERIC_SW_STAT(rx_noskb_drops),
1148 	EF10_DMA_STAT(port_rx_pm_trunc_bb_overflow, PM_TRUNC_BB_OVERFLOW),
1149 	EF10_DMA_STAT(port_rx_pm_discard_bb_overflow, PM_DISCARD_BB_OVERFLOW),
1150 	EF10_DMA_STAT(port_rx_pm_trunc_vfifo_full, PM_TRUNC_VFIFO_FULL),
1151 	EF10_DMA_STAT(port_rx_pm_discard_vfifo_full, PM_DISCARD_VFIFO_FULL),
1152 	EF10_DMA_STAT(port_rx_pm_trunc_qbb, PM_TRUNC_QBB),
1153 	EF10_DMA_STAT(port_rx_pm_discard_qbb, PM_DISCARD_QBB),
1154 	EF10_DMA_STAT(port_rx_pm_discard_mapping, PM_DISCARD_MAPPING),
1155 	EF10_DMA_STAT(port_rx_dp_q_disabled_packets, RXDP_Q_DISABLED_PKTS),
1156 	EF10_DMA_STAT(port_rx_dp_di_dropped_packets, RXDP_DI_DROPPED_PKTS),
1157 	EF10_DMA_STAT(port_rx_dp_streaming_packets, RXDP_STREAMING_PKTS),
1158 	EF10_DMA_STAT(port_rx_dp_hlb_fetch, RXDP_HLB_FETCH_CONDITIONS),
1159 	EF10_DMA_STAT(port_rx_dp_hlb_wait, RXDP_HLB_WAIT_CONDITIONS),
1160 	EF10_DMA_STAT(rx_unicast, VADAPTER_RX_UNICAST_PACKETS),
1161 	EF10_DMA_STAT(rx_unicast_bytes, VADAPTER_RX_UNICAST_BYTES),
1162 	EF10_DMA_STAT(rx_multicast, VADAPTER_RX_MULTICAST_PACKETS),
1163 	EF10_DMA_STAT(rx_multicast_bytes, VADAPTER_RX_MULTICAST_BYTES),
1164 	EF10_DMA_STAT(rx_broadcast, VADAPTER_RX_BROADCAST_PACKETS),
1165 	EF10_DMA_STAT(rx_broadcast_bytes, VADAPTER_RX_BROADCAST_BYTES),
1166 	EF10_DMA_STAT(rx_bad, VADAPTER_RX_BAD_PACKETS),
1167 	EF10_DMA_STAT(rx_bad_bytes, VADAPTER_RX_BAD_BYTES),
1168 	EF10_DMA_STAT(rx_overflow, VADAPTER_RX_OVERFLOW),
1169 	EF10_DMA_STAT(tx_unicast, VADAPTER_TX_UNICAST_PACKETS),
1170 	EF10_DMA_STAT(tx_unicast_bytes, VADAPTER_TX_UNICAST_BYTES),
1171 	EF10_DMA_STAT(tx_multicast, VADAPTER_TX_MULTICAST_PACKETS),
1172 	EF10_DMA_STAT(tx_multicast_bytes, VADAPTER_TX_MULTICAST_BYTES),
1173 	EF10_DMA_STAT(tx_broadcast, VADAPTER_TX_BROADCAST_PACKETS),
1174 	EF10_DMA_STAT(tx_broadcast_bytes, VADAPTER_TX_BROADCAST_BYTES),
1175 	EF10_DMA_STAT(tx_bad, VADAPTER_TX_BAD_PACKETS),
1176 	EF10_DMA_STAT(tx_bad_bytes, VADAPTER_TX_BAD_BYTES),
1177 	EF10_DMA_STAT(tx_overflow, VADAPTER_TX_OVERFLOW),
1178 };
1179 
1180 #define HUNT_COMMON_STAT_MASK ((1ULL << EF10_STAT_port_tx_bytes) |	\
1181 			       (1ULL << EF10_STAT_port_tx_packets) |	\
1182 			       (1ULL << EF10_STAT_port_tx_pause) |	\
1183 			       (1ULL << EF10_STAT_port_tx_unicast) |	\
1184 			       (1ULL << EF10_STAT_port_tx_multicast) |	\
1185 			       (1ULL << EF10_STAT_port_tx_broadcast) |	\
1186 			       (1ULL << EF10_STAT_port_rx_bytes) |	\
1187 			       (1ULL <<                                 \
1188 				EF10_STAT_port_rx_bytes_minus_good_bytes) | \
1189 			       (1ULL << EF10_STAT_port_rx_good_bytes) |	\
1190 			       (1ULL << EF10_STAT_port_rx_bad_bytes) |	\
1191 			       (1ULL << EF10_STAT_port_rx_packets) |	\
1192 			       (1ULL << EF10_STAT_port_rx_good) |	\
1193 			       (1ULL << EF10_STAT_port_rx_bad) |	\
1194 			       (1ULL << EF10_STAT_port_rx_pause) |	\
1195 			       (1ULL << EF10_STAT_port_rx_control) |	\
1196 			       (1ULL << EF10_STAT_port_rx_unicast) |	\
1197 			       (1ULL << EF10_STAT_port_rx_multicast) |	\
1198 			       (1ULL << EF10_STAT_port_rx_broadcast) |	\
1199 			       (1ULL << EF10_STAT_port_rx_lt64) |	\
1200 			       (1ULL << EF10_STAT_port_rx_64) |		\
1201 			       (1ULL << EF10_STAT_port_rx_65_to_127) |	\
1202 			       (1ULL << EF10_STAT_port_rx_128_to_255) |	\
1203 			       (1ULL << EF10_STAT_port_rx_256_to_511) |	\
1204 			       (1ULL << EF10_STAT_port_rx_512_to_1023) |\
1205 			       (1ULL << EF10_STAT_port_rx_1024_to_15xx) |\
1206 			       (1ULL << EF10_STAT_port_rx_15xx_to_jumbo) |\
1207 			       (1ULL << EF10_STAT_port_rx_gtjumbo) |	\
1208 			       (1ULL << EF10_STAT_port_rx_bad_gtjumbo) |\
1209 			       (1ULL << EF10_STAT_port_rx_overflow) |	\
1210 			       (1ULL << EF10_STAT_port_rx_nodesc_drops) |\
1211 			       (1ULL << GENERIC_STAT_rx_nodesc_trunc) |	\
1212 			       (1ULL << GENERIC_STAT_rx_noskb_drops))
1213 
1214 /* These statistics are only provided by the 10G MAC.  For a 10G/40G
1215  * switchable port we do not expose these because they might not
1216  * include all the packets they should.
1217  */
1218 #define HUNT_10G_ONLY_STAT_MASK ((1ULL << EF10_STAT_port_tx_control) |	\
1219 				 (1ULL << EF10_STAT_port_tx_lt64) |	\
1220 				 (1ULL << EF10_STAT_port_tx_64) |	\
1221 				 (1ULL << EF10_STAT_port_tx_65_to_127) |\
1222 				 (1ULL << EF10_STAT_port_tx_128_to_255) |\
1223 				 (1ULL << EF10_STAT_port_tx_256_to_511) |\
1224 				 (1ULL << EF10_STAT_port_tx_512_to_1023) |\
1225 				 (1ULL << EF10_STAT_port_tx_1024_to_15xx) |\
1226 				 (1ULL << EF10_STAT_port_tx_15xx_to_jumbo))
1227 
1228 /* These statistics are only provided by the 40G MAC.  For a 10G/40G
1229  * switchable port we do expose these because the errors will otherwise
1230  * be silent.
1231  */
1232 #define HUNT_40G_EXTRA_STAT_MASK ((1ULL << EF10_STAT_port_rx_align_error) |\
1233 				  (1ULL << EF10_STAT_port_rx_length_error))
1234 
1235 /* These statistics are only provided if the firmware supports the
1236  * capability PM_AND_RXDP_COUNTERS.
1237  */
1238 #define HUNT_PM_AND_RXDP_STAT_MASK (					\
1239 	(1ULL << EF10_STAT_port_rx_pm_trunc_bb_overflow) |		\
1240 	(1ULL << EF10_STAT_port_rx_pm_discard_bb_overflow) |		\
1241 	(1ULL << EF10_STAT_port_rx_pm_trunc_vfifo_full) |		\
1242 	(1ULL << EF10_STAT_port_rx_pm_discard_vfifo_full) |		\
1243 	(1ULL << EF10_STAT_port_rx_pm_trunc_qbb) |			\
1244 	(1ULL << EF10_STAT_port_rx_pm_discard_qbb) |			\
1245 	(1ULL << EF10_STAT_port_rx_pm_discard_mapping) |		\
1246 	(1ULL << EF10_STAT_port_rx_dp_q_disabled_packets) |		\
1247 	(1ULL << EF10_STAT_port_rx_dp_di_dropped_packets) |		\
1248 	(1ULL << EF10_STAT_port_rx_dp_streaming_packets) |		\
1249 	(1ULL << EF10_STAT_port_rx_dp_hlb_fetch) |			\
1250 	(1ULL << EF10_STAT_port_rx_dp_hlb_wait))
1251 
1252 static u64 efx_ef10_raw_stat_mask(struct efx_nic *efx)
1253 {
1254 	u64 raw_mask = HUNT_COMMON_STAT_MASK;
1255 	u32 port_caps = efx_mcdi_phy_get_caps(efx);
1256 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1257 
1258 	if (!(efx->mcdi->fn_flags &
1259 	      1 << MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_LINKCTRL))
1260 		return 0;
1261 
1262 	if (port_caps & (1 << MC_CMD_PHY_CAP_40000FDX_LBN))
1263 		raw_mask |= HUNT_40G_EXTRA_STAT_MASK;
1264 	else
1265 		raw_mask |= HUNT_10G_ONLY_STAT_MASK;
1266 
1267 	if (nic_data->datapath_caps &
1268 	    (1 << MC_CMD_GET_CAPABILITIES_OUT_PM_AND_RXDP_COUNTERS_LBN))
1269 		raw_mask |= HUNT_PM_AND_RXDP_STAT_MASK;
1270 
1271 	return raw_mask;
1272 }
1273 
1274 static void efx_ef10_get_stat_mask(struct efx_nic *efx, unsigned long *mask)
1275 {
1276 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1277 	u64 raw_mask[2];
1278 
1279 	raw_mask[0] = efx_ef10_raw_stat_mask(efx);
1280 
1281 	/* Only show vadaptor stats when EVB capability is present */
1282 	if (nic_data->datapath_caps &
1283 	    (1 << MC_CMD_GET_CAPABILITIES_OUT_EVB_LBN)) {
1284 		raw_mask[0] |= ~((1ULL << EF10_STAT_rx_unicast) - 1);
1285 		raw_mask[1] = (1ULL << (EF10_STAT_COUNT - 63)) - 1;
1286 	} else {
1287 		raw_mask[1] = 0;
1288 	}
1289 
1290 #if BITS_PER_LONG == 64
1291 	mask[0] = raw_mask[0];
1292 	mask[1] = raw_mask[1];
1293 #else
1294 	mask[0] = raw_mask[0] & 0xffffffff;
1295 	mask[1] = raw_mask[0] >> 32;
1296 	mask[2] = raw_mask[1] & 0xffffffff;
1297 	mask[3] = raw_mask[1] >> 32;
1298 #endif
1299 }
1300 
1301 static size_t efx_ef10_describe_stats(struct efx_nic *efx, u8 *names)
1302 {
1303 	DECLARE_BITMAP(mask, EF10_STAT_COUNT);
1304 
1305 	efx_ef10_get_stat_mask(efx, mask);
1306 	return efx_nic_describe_stats(efx_ef10_stat_desc, EF10_STAT_COUNT,
1307 				      mask, names);
1308 }
1309 
1310 static size_t efx_ef10_update_stats_common(struct efx_nic *efx, u64 *full_stats,
1311 					   struct rtnl_link_stats64 *core_stats)
1312 {
1313 	DECLARE_BITMAP(mask, EF10_STAT_COUNT);
1314 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1315 	u64 *stats = nic_data->stats;
1316 	size_t stats_count = 0, index;
1317 
1318 	efx_ef10_get_stat_mask(efx, mask);
1319 
1320 	if (full_stats) {
1321 		for_each_set_bit(index, mask, EF10_STAT_COUNT) {
1322 			if (efx_ef10_stat_desc[index].name) {
1323 				*full_stats++ = stats[index];
1324 				++stats_count;
1325 			}
1326 		}
1327 	}
1328 
1329 	if (!core_stats)
1330 		return stats_count;
1331 
1332 	if (nic_data->datapath_caps &
1333 			1 << MC_CMD_GET_CAPABILITIES_OUT_EVB_LBN) {
1334 		/* Use vadaptor stats. */
1335 		core_stats->rx_packets = stats[EF10_STAT_rx_unicast] +
1336 					 stats[EF10_STAT_rx_multicast] +
1337 					 stats[EF10_STAT_rx_broadcast];
1338 		core_stats->tx_packets = stats[EF10_STAT_tx_unicast] +
1339 					 stats[EF10_STAT_tx_multicast] +
1340 					 stats[EF10_STAT_tx_broadcast];
1341 		core_stats->rx_bytes = stats[EF10_STAT_rx_unicast_bytes] +
1342 				       stats[EF10_STAT_rx_multicast_bytes] +
1343 				       stats[EF10_STAT_rx_broadcast_bytes];
1344 		core_stats->tx_bytes = stats[EF10_STAT_tx_unicast_bytes] +
1345 				       stats[EF10_STAT_tx_multicast_bytes] +
1346 				       stats[EF10_STAT_tx_broadcast_bytes];
1347 		core_stats->rx_dropped = stats[GENERIC_STAT_rx_nodesc_trunc] +
1348 					 stats[GENERIC_STAT_rx_noskb_drops];
1349 		core_stats->multicast = stats[EF10_STAT_rx_multicast];
1350 		core_stats->rx_crc_errors = stats[EF10_STAT_rx_bad];
1351 		core_stats->rx_fifo_errors = stats[EF10_STAT_rx_overflow];
1352 		core_stats->rx_errors = core_stats->rx_crc_errors;
1353 		core_stats->tx_errors = stats[EF10_STAT_tx_bad];
1354 	} else {
1355 		/* Use port stats. */
1356 		core_stats->rx_packets = stats[EF10_STAT_port_rx_packets];
1357 		core_stats->tx_packets = stats[EF10_STAT_port_tx_packets];
1358 		core_stats->rx_bytes = stats[EF10_STAT_port_rx_bytes];
1359 		core_stats->tx_bytes = stats[EF10_STAT_port_tx_bytes];
1360 		core_stats->rx_dropped = stats[EF10_STAT_port_rx_nodesc_drops] +
1361 					 stats[GENERIC_STAT_rx_nodesc_trunc] +
1362 					 stats[GENERIC_STAT_rx_noskb_drops];
1363 		core_stats->multicast = stats[EF10_STAT_port_rx_multicast];
1364 		core_stats->rx_length_errors =
1365 				stats[EF10_STAT_port_rx_gtjumbo] +
1366 				stats[EF10_STAT_port_rx_length_error];
1367 		core_stats->rx_crc_errors = stats[EF10_STAT_port_rx_bad];
1368 		core_stats->rx_frame_errors =
1369 				stats[EF10_STAT_port_rx_align_error];
1370 		core_stats->rx_fifo_errors = stats[EF10_STAT_port_rx_overflow];
1371 		core_stats->rx_errors = (core_stats->rx_length_errors +
1372 					 core_stats->rx_crc_errors +
1373 					 core_stats->rx_frame_errors);
1374 	}
1375 
1376 	return stats_count;
1377 }
1378 
1379 static int efx_ef10_try_update_nic_stats_pf(struct efx_nic *efx)
1380 {
1381 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1382 	DECLARE_BITMAP(mask, EF10_STAT_COUNT);
1383 	__le64 generation_start, generation_end;
1384 	u64 *stats = nic_data->stats;
1385 	__le64 *dma_stats;
1386 
1387 	efx_ef10_get_stat_mask(efx, mask);
1388 
1389 	dma_stats = efx->stats_buffer.addr;
1390 	nic_data = efx->nic_data;
1391 
1392 	generation_end = dma_stats[MC_CMD_MAC_GENERATION_END];
1393 	if (generation_end == EFX_MC_STATS_GENERATION_INVALID)
1394 		return 0;
1395 	rmb();
1396 	efx_nic_update_stats(efx_ef10_stat_desc, EF10_STAT_COUNT, mask,
1397 			     stats, efx->stats_buffer.addr, false);
1398 	rmb();
1399 	generation_start = dma_stats[MC_CMD_MAC_GENERATION_START];
1400 	if (generation_end != generation_start)
1401 		return -EAGAIN;
1402 
1403 	/* Update derived statistics */
1404 	efx_nic_fix_nodesc_drop_stat(efx,
1405 				     &stats[EF10_STAT_port_rx_nodesc_drops]);
1406 	stats[EF10_STAT_port_rx_good_bytes] =
1407 		stats[EF10_STAT_port_rx_bytes] -
1408 		stats[EF10_STAT_port_rx_bytes_minus_good_bytes];
1409 	efx_update_diff_stat(&stats[EF10_STAT_port_rx_bad_bytes],
1410 			     stats[EF10_STAT_port_rx_bytes_minus_good_bytes]);
1411 	efx_update_sw_stats(efx, stats);
1412 	return 0;
1413 }
1414 
1415 
1416 static size_t efx_ef10_update_stats_pf(struct efx_nic *efx, u64 *full_stats,
1417 				       struct rtnl_link_stats64 *core_stats)
1418 {
1419 	int retry;
1420 
1421 	/* If we're unlucky enough to read statistics during the DMA, wait
1422 	 * up to 10ms for it to finish (typically takes <500us)
1423 	 */
1424 	for (retry = 0; retry < 100; ++retry) {
1425 		if (efx_ef10_try_update_nic_stats_pf(efx) == 0)
1426 			break;
1427 		udelay(100);
1428 	}
1429 
1430 	return efx_ef10_update_stats_common(efx, full_stats, core_stats);
1431 }
1432 
1433 static int efx_ef10_try_update_nic_stats_vf(struct efx_nic *efx)
1434 {
1435 	MCDI_DECLARE_BUF(inbuf, MC_CMD_MAC_STATS_IN_LEN);
1436 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1437 	DECLARE_BITMAP(mask, EF10_STAT_COUNT);
1438 	__le64 generation_start, generation_end;
1439 	u64 *stats = nic_data->stats;
1440 	u32 dma_len = MC_CMD_MAC_NSTATS * sizeof(u64);
1441 	struct efx_buffer stats_buf;
1442 	__le64 *dma_stats;
1443 	int rc;
1444 
1445 	spin_unlock_bh(&efx->stats_lock);
1446 
1447 	if (in_interrupt()) {
1448 		/* If in atomic context, cannot update stats.  Just update the
1449 		 * software stats and return so the caller can continue.
1450 		 */
1451 		spin_lock_bh(&efx->stats_lock);
1452 		efx_update_sw_stats(efx, stats);
1453 		return 0;
1454 	}
1455 
1456 	efx_ef10_get_stat_mask(efx, mask);
1457 
1458 	rc = efx_nic_alloc_buffer(efx, &stats_buf, dma_len, GFP_ATOMIC);
1459 	if (rc) {
1460 		spin_lock_bh(&efx->stats_lock);
1461 		return rc;
1462 	}
1463 
1464 	dma_stats = stats_buf.addr;
1465 	dma_stats[MC_CMD_MAC_GENERATION_END] = EFX_MC_STATS_GENERATION_INVALID;
1466 
1467 	MCDI_SET_QWORD(inbuf, MAC_STATS_IN_DMA_ADDR, stats_buf.dma_addr);
1468 	MCDI_POPULATE_DWORD_1(inbuf, MAC_STATS_IN_CMD,
1469 			      MAC_STATS_IN_DMA, 1);
1470 	MCDI_SET_DWORD(inbuf, MAC_STATS_IN_DMA_LEN, dma_len);
1471 	MCDI_SET_DWORD(inbuf, MAC_STATS_IN_PORT_ID, EVB_PORT_ID_ASSIGNED);
1472 
1473 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_MAC_STATS, inbuf, sizeof(inbuf),
1474 				NULL, 0, NULL);
1475 	spin_lock_bh(&efx->stats_lock);
1476 	if (rc) {
1477 		/* Expect ENOENT if DMA queues have not been set up */
1478 		if (rc != -ENOENT || atomic_read(&efx->active_queues))
1479 			efx_mcdi_display_error(efx, MC_CMD_MAC_STATS,
1480 					       sizeof(inbuf), NULL, 0, rc);
1481 		goto out;
1482 	}
1483 
1484 	generation_end = dma_stats[MC_CMD_MAC_GENERATION_END];
1485 	if (generation_end == EFX_MC_STATS_GENERATION_INVALID) {
1486 		WARN_ON_ONCE(1);
1487 		goto out;
1488 	}
1489 	rmb();
1490 	efx_nic_update_stats(efx_ef10_stat_desc, EF10_STAT_COUNT, mask,
1491 			     stats, stats_buf.addr, false);
1492 	rmb();
1493 	generation_start = dma_stats[MC_CMD_MAC_GENERATION_START];
1494 	if (generation_end != generation_start) {
1495 		rc = -EAGAIN;
1496 		goto out;
1497 	}
1498 
1499 	efx_update_sw_stats(efx, stats);
1500 out:
1501 	efx_nic_free_buffer(efx, &stats_buf);
1502 	return rc;
1503 }
1504 
1505 static size_t efx_ef10_update_stats_vf(struct efx_nic *efx, u64 *full_stats,
1506 				       struct rtnl_link_stats64 *core_stats)
1507 {
1508 	if (efx_ef10_try_update_nic_stats_vf(efx))
1509 		return 0;
1510 
1511 	return efx_ef10_update_stats_common(efx, full_stats, core_stats);
1512 }
1513 
1514 static void efx_ef10_push_irq_moderation(struct efx_channel *channel)
1515 {
1516 	struct efx_nic *efx = channel->efx;
1517 	unsigned int mode, value;
1518 	efx_dword_t timer_cmd;
1519 
1520 	if (channel->irq_moderation) {
1521 		mode = 3;
1522 		value = channel->irq_moderation - 1;
1523 	} else {
1524 		mode = 0;
1525 		value = 0;
1526 	}
1527 
1528 	if (EFX_EF10_WORKAROUND_35388(efx)) {
1529 		EFX_POPULATE_DWORD_3(timer_cmd, ERF_DD_EVQ_IND_TIMER_FLAGS,
1530 				     EFE_DD_EVQ_IND_TIMER_FLAGS,
1531 				     ERF_DD_EVQ_IND_TIMER_MODE, mode,
1532 				     ERF_DD_EVQ_IND_TIMER_VAL, value);
1533 		efx_writed_page(efx, &timer_cmd, ER_DD_EVQ_INDIRECT,
1534 				channel->channel);
1535 	} else {
1536 		EFX_POPULATE_DWORD_2(timer_cmd, ERF_DZ_TC_TIMER_MODE, mode,
1537 				     ERF_DZ_TC_TIMER_VAL, value);
1538 		efx_writed_page(efx, &timer_cmd, ER_DZ_EVQ_TMR,
1539 				channel->channel);
1540 	}
1541 }
1542 
1543 static void efx_ef10_get_wol_vf(struct efx_nic *efx,
1544 				struct ethtool_wolinfo *wol) {}
1545 
1546 static int efx_ef10_set_wol_vf(struct efx_nic *efx, u32 type)
1547 {
1548 	return -EOPNOTSUPP;
1549 }
1550 
1551 static void efx_ef10_get_wol(struct efx_nic *efx, struct ethtool_wolinfo *wol)
1552 {
1553 	wol->supported = 0;
1554 	wol->wolopts = 0;
1555 	memset(&wol->sopass, 0, sizeof(wol->sopass));
1556 }
1557 
1558 static int efx_ef10_set_wol(struct efx_nic *efx, u32 type)
1559 {
1560 	if (type != 0)
1561 		return -EINVAL;
1562 	return 0;
1563 }
1564 
1565 static void efx_ef10_mcdi_request(struct efx_nic *efx,
1566 				  const efx_dword_t *hdr, size_t hdr_len,
1567 				  const efx_dword_t *sdu, size_t sdu_len)
1568 {
1569 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1570 	u8 *pdu = nic_data->mcdi_buf.addr;
1571 
1572 	memcpy(pdu, hdr, hdr_len);
1573 	memcpy(pdu + hdr_len, sdu, sdu_len);
1574 	wmb();
1575 
1576 	/* The hardware provides 'low' and 'high' (doorbell) registers
1577 	 * for passing the 64-bit address of an MCDI request to
1578 	 * firmware.  However the dwords are swapped by firmware.  The
1579 	 * least significant bits of the doorbell are then 0 for all
1580 	 * MCDI requests due to alignment.
1581 	 */
1582 	_efx_writed(efx, cpu_to_le32((u64)nic_data->mcdi_buf.dma_addr >> 32),
1583 		    ER_DZ_MC_DB_LWRD);
1584 	_efx_writed(efx, cpu_to_le32((u32)nic_data->mcdi_buf.dma_addr),
1585 		    ER_DZ_MC_DB_HWRD);
1586 }
1587 
1588 static bool efx_ef10_mcdi_poll_response(struct efx_nic *efx)
1589 {
1590 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1591 	const efx_dword_t hdr = *(const efx_dword_t *)nic_data->mcdi_buf.addr;
1592 
1593 	rmb();
1594 	return EFX_DWORD_FIELD(hdr, MCDI_HEADER_RESPONSE);
1595 }
1596 
1597 static void
1598 efx_ef10_mcdi_read_response(struct efx_nic *efx, efx_dword_t *outbuf,
1599 			    size_t offset, size_t outlen)
1600 {
1601 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1602 	const u8 *pdu = nic_data->mcdi_buf.addr;
1603 
1604 	memcpy(outbuf, pdu + offset, outlen);
1605 }
1606 
1607 static int efx_ef10_mcdi_poll_reboot(struct efx_nic *efx)
1608 {
1609 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1610 	int rc;
1611 
1612 	rc = efx_ef10_get_warm_boot_count(efx);
1613 	if (rc < 0) {
1614 		/* The firmware is presumably in the process of
1615 		 * rebooting.  However, we are supposed to report each
1616 		 * reboot just once, so we must only do that once we
1617 		 * can read and store the updated warm boot count.
1618 		 */
1619 		return 0;
1620 	}
1621 
1622 	if (rc == nic_data->warm_boot_count)
1623 		return 0;
1624 
1625 	nic_data->warm_boot_count = rc;
1626 
1627 	/* All our allocations have been reset */
1628 	efx_ef10_reset_mc_allocations(efx);
1629 
1630 	/* The datapath firmware might have been changed */
1631 	nic_data->must_check_datapath_caps = true;
1632 
1633 	/* MAC statistics have been cleared on the NIC; clear the local
1634 	 * statistic that we update with efx_update_diff_stat().
1635 	 */
1636 	nic_data->stats[EF10_STAT_port_rx_bad_bytes] = 0;
1637 
1638 	return -EIO;
1639 }
1640 
1641 /* Handle an MSI interrupt
1642  *
1643  * Handle an MSI hardware interrupt.  This routine schedules event
1644  * queue processing.  No interrupt acknowledgement cycle is necessary.
1645  * Also, we never need to check that the interrupt is for us, since
1646  * MSI interrupts cannot be shared.
1647  */
1648 static irqreturn_t efx_ef10_msi_interrupt(int irq, void *dev_id)
1649 {
1650 	struct efx_msi_context *context = dev_id;
1651 	struct efx_nic *efx = context->efx;
1652 
1653 	netif_vdbg(efx, intr, efx->net_dev,
1654 		   "IRQ %d on CPU %d\n", irq, raw_smp_processor_id());
1655 
1656 	if (likely(ACCESS_ONCE(efx->irq_soft_enabled))) {
1657 		/* Note test interrupts */
1658 		if (context->index == efx->irq_level)
1659 			efx->last_irq_cpu = raw_smp_processor_id();
1660 
1661 		/* Schedule processing of the channel */
1662 		efx_schedule_channel_irq(efx->channel[context->index]);
1663 	}
1664 
1665 	return IRQ_HANDLED;
1666 }
1667 
1668 static irqreturn_t efx_ef10_legacy_interrupt(int irq, void *dev_id)
1669 {
1670 	struct efx_nic *efx = dev_id;
1671 	bool soft_enabled = ACCESS_ONCE(efx->irq_soft_enabled);
1672 	struct efx_channel *channel;
1673 	efx_dword_t reg;
1674 	u32 queues;
1675 
1676 	/* Read the ISR which also ACKs the interrupts */
1677 	efx_readd(efx, &reg, ER_DZ_BIU_INT_ISR);
1678 	queues = EFX_DWORD_FIELD(reg, ERF_DZ_ISR_REG);
1679 
1680 	if (queues == 0)
1681 		return IRQ_NONE;
1682 
1683 	if (likely(soft_enabled)) {
1684 		/* Note test interrupts */
1685 		if (queues & (1U << efx->irq_level))
1686 			efx->last_irq_cpu = raw_smp_processor_id();
1687 
1688 		efx_for_each_channel(channel, efx) {
1689 			if (queues & 1)
1690 				efx_schedule_channel_irq(channel);
1691 			queues >>= 1;
1692 		}
1693 	}
1694 
1695 	netif_vdbg(efx, intr, efx->net_dev,
1696 		   "IRQ %d on CPU %d status " EFX_DWORD_FMT "\n",
1697 		   irq, raw_smp_processor_id(), EFX_DWORD_VAL(reg));
1698 
1699 	return IRQ_HANDLED;
1700 }
1701 
1702 static void efx_ef10_irq_test_generate(struct efx_nic *efx)
1703 {
1704 	MCDI_DECLARE_BUF(inbuf, MC_CMD_TRIGGER_INTERRUPT_IN_LEN);
1705 
1706 	BUILD_BUG_ON(MC_CMD_TRIGGER_INTERRUPT_OUT_LEN != 0);
1707 
1708 	MCDI_SET_DWORD(inbuf, TRIGGER_INTERRUPT_IN_INTR_LEVEL, efx->irq_level);
1709 	(void) efx_mcdi_rpc(efx, MC_CMD_TRIGGER_INTERRUPT,
1710 			    inbuf, sizeof(inbuf), NULL, 0, NULL);
1711 }
1712 
1713 static int efx_ef10_tx_probe(struct efx_tx_queue *tx_queue)
1714 {
1715 	return efx_nic_alloc_buffer(tx_queue->efx, &tx_queue->txd.buf,
1716 				    (tx_queue->ptr_mask + 1) *
1717 				    sizeof(efx_qword_t),
1718 				    GFP_KERNEL);
1719 }
1720 
1721 /* This writes to the TX_DESC_WPTR and also pushes data */
1722 static inline void efx_ef10_push_tx_desc(struct efx_tx_queue *tx_queue,
1723 					 const efx_qword_t *txd)
1724 {
1725 	unsigned int write_ptr;
1726 	efx_oword_t reg;
1727 
1728 	write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
1729 	EFX_POPULATE_OWORD_1(reg, ERF_DZ_TX_DESC_WPTR, write_ptr);
1730 	reg.qword[0] = *txd;
1731 	efx_writeo_page(tx_queue->efx, &reg,
1732 			ER_DZ_TX_DESC_UPD, tx_queue->queue);
1733 }
1734 
1735 static void efx_ef10_tx_init(struct efx_tx_queue *tx_queue)
1736 {
1737 	MCDI_DECLARE_BUF(inbuf, MC_CMD_INIT_TXQ_IN_LEN(EFX_MAX_DMAQ_SIZE * 8 /
1738 						       EFX_BUF_SIZE));
1739 	bool csum_offload = tx_queue->queue & EFX_TXQ_TYPE_OFFLOAD;
1740 	size_t entries = tx_queue->txd.buf.len / EFX_BUF_SIZE;
1741 	struct efx_channel *channel = tx_queue->channel;
1742 	struct efx_nic *efx = tx_queue->efx;
1743 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1744 	size_t inlen;
1745 	dma_addr_t dma_addr;
1746 	efx_qword_t *txd;
1747 	int rc;
1748 	int i;
1749 	BUILD_BUG_ON(MC_CMD_INIT_TXQ_OUT_LEN != 0);
1750 
1751 	MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_SIZE, tx_queue->ptr_mask + 1);
1752 	MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_TARGET_EVQ, channel->channel);
1753 	MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_LABEL, tx_queue->queue);
1754 	MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_INSTANCE, tx_queue->queue);
1755 	MCDI_POPULATE_DWORD_2(inbuf, INIT_TXQ_IN_FLAGS,
1756 			      INIT_TXQ_IN_FLAG_IP_CSUM_DIS, !csum_offload,
1757 			      INIT_TXQ_IN_FLAG_TCP_CSUM_DIS, !csum_offload);
1758 	MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_OWNER_ID, 0);
1759 	MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_PORT_ID, nic_data->vport_id);
1760 
1761 	dma_addr = tx_queue->txd.buf.dma_addr;
1762 
1763 	netif_dbg(efx, hw, efx->net_dev, "pushing TXQ %d. %zu entries (%llx)\n",
1764 		  tx_queue->queue, entries, (u64)dma_addr);
1765 
1766 	for (i = 0; i < entries; ++i) {
1767 		MCDI_SET_ARRAY_QWORD(inbuf, INIT_TXQ_IN_DMA_ADDR, i, dma_addr);
1768 		dma_addr += EFX_BUF_SIZE;
1769 	}
1770 
1771 	inlen = MC_CMD_INIT_TXQ_IN_LEN(entries);
1772 
1773 	rc = efx_mcdi_rpc(efx, MC_CMD_INIT_TXQ, inbuf, inlen,
1774 			  NULL, 0, NULL);
1775 	if (rc)
1776 		goto fail;
1777 
1778 	/* A previous user of this TX queue might have set us up the
1779 	 * bomb by writing a descriptor to the TX push collector but
1780 	 * not the doorbell.  (Each collector belongs to a port, not a
1781 	 * queue or function, so cannot easily be reset.)  We must
1782 	 * attempt to push a no-op descriptor in its place.
1783 	 */
1784 	tx_queue->buffer[0].flags = EFX_TX_BUF_OPTION;
1785 	tx_queue->insert_count = 1;
1786 	txd = efx_tx_desc(tx_queue, 0);
1787 	EFX_POPULATE_QWORD_4(*txd,
1788 			     ESF_DZ_TX_DESC_IS_OPT, true,
1789 			     ESF_DZ_TX_OPTION_TYPE,
1790 			     ESE_DZ_TX_OPTION_DESC_CRC_CSUM,
1791 			     ESF_DZ_TX_OPTION_UDP_TCP_CSUM, csum_offload,
1792 			     ESF_DZ_TX_OPTION_IP_CSUM, csum_offload);
1793 	tx_queue->write_count = 1;
1794 	wmb();
1795 	efx_ef10_push_tx_desc(tx_queue, txd);
1796 
1797 	return;
1798 
1799 fail:
1800 	netdev_WARN(efx->net_dev, "failed to initialise TXQ %d\n",
1801 		    tx_queue->queue);
1802 }
1803 
1804 static void efx_ef10_tx_fini(struct efx_tx_queue *tx_queue)
1805 {
1806 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FINI_TXQ_IN_LEN);
1807 	MCDI_DECLARE_BUF_ERR(outbuf);
1808 	struct efx_nic *efx = tx_queue->efx;
1809 	size_t outlen;
1810 	int rc;
1811 
1812 	MCDI_SET_DWORD(inbuf, FINI_TXQ_IN_INSTANCE,
1813 		       tx_queue->queue);
1814 
1815 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FINI_TXQ, inbuf, sizeof(inbuf),
1816 			  outbuf, sizeof(outbuf), &outlen);
1817 
1818 	if (rc && rc != -EALREADY)
1819 		goto fail;
1820 
1821 	return;
1822 
1823 fail:
1824 	efx_mcdi_display_error(efx, MC_CMD_FINI_TXQ, MC_CMD_FINI_TXQ_IN_LEN,
1825 			       outbuf, outlen, rc);
1826 }
1827 
1828 static void efx_ef10_tx_remove(struct efx_tx_queue *tx_queue)
1829 {
1830 	efx_nic_free_buffer(tx_queue->efx, &tx_queue->txd.buf);
1831 }
1832 
1833 /* This writes to the TX_DESC_WPTR; write pointer for TX descriptor ring */
1834 static inline void efx_ef10_notify_tx_desc(struct efx_tx_queue *tx_queue)
1835 {
1836 	unsigned int write_ptr;
1837 	efx_dword_t reg;
1838 
1839 	write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
1840 	EFX_POPULATE_DWORD_1(reg, ERF_DZ_TX_DESC_WPTR_DWORD, write_ptr);
1841 	efx_writed_page(tx_queue->efx, &reg,
1842 			ER_DZ_TX_DESC_UPD_DWORD, tx_queue->queue);
1843 }
1844 
1845 static void efx_ef10_tx_write(struct efx_tx_queue *tx_queue)
1846 {
1847 	unsigned int old_write_count = tx_queue->write_count;
1848 	struct efx_tx_buffer *buffer;
1849 	unsigned int write_ptr;
1850 	efx_qword_t *txd;
1851 
1852 	BUG_ON(tx_queue->write_count == tx_queue->insert_count);
1853 
1854 	do {
1855 		write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
1856 		buffer = &tx_queue->buffer[write_ptr];
1857 		txd = efx_tx_desc(tx_queue, write_ptr);
1858 		++tx_queue->write_count;
1859 
1860 		/* Create TX descriptor ring entry */
1861 		if (buffer->flags & EFX_TX_BUF_OPTION) {
1862 			*txd = buffer->option;
1863 		} else {
1864 			BUILD_BUG_ON(EFX_TX_BUF_CONT != 1);
1865 			EFX_POPULATE_QWORD_3(
1866 				*txd,
1867 				ESF_DZ_TX_KER_CONT,
1868 				buffer->flags & EFX_TX_BUF_CONT,
1869 				ESF_DZ_TX_KER_BYTE_CNT, buffer->len,
1870 				ESF_DZ_TX_KER_BUF_ADDR, buffer->dma_addr);
1871 		}
1872 	} while (tx_queue->write_count != tx_queue->insert_count);
1873 
1874 	wmb(); /* Ensure descriptors are written before they are fetched */
1875 
1876 	if (efx_nic_may_push_tx_desc(tx_queue, old_write_count)) {
1877 		txd = efx_tx_desc(tx_queue,
1878 				  old_write_count & tx_queue->ptr_mask);
1879 		efx_ef10_push_tx_desc(tx_queue, txd);
1880 		++tx_queue->pushes;
1881 	} else {
1882 		efx_ef10_notify_tx_desc(tx_queue);
1883 	}
1884 }
1885 
1886 static int efx_ef10_alloc_rss_context(struct efx_nic *efx, u32 *context,
1887 				      bool exclusive, unsigned *context_size)
1888 {
1889 	MCDI_DECLARE_BUF(inbuf, MC_CMD_RSS_CONTEXT_ALLOC_IN_LEN);
1890 	MCDI_DECLARE_BUF(outbuf, MC_CMD_RSS_CONTEXT_ALLOC_OUT_LEN);
1891 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1892 	size_t outlen;
1893 	int rc;
1894 	u32 alloc_type = exclusive ?
1895 				MC_CMD_RSS_CONTEXT_ALLOC_IN_TYPE_EXCLUSIVE :
1896 				MC_CMD_RSS_CONTEXT_ALLOC_IN_TYPE_SHARED;
1897 	unsigned rss_spread = exclusive ?
1898 				efx->rss_spread :
1899 				min(rounddown_pow_of_two(efx->rss_spread),
1900 				    EFX_EF10_MAX_SHARED_RSS_CONTEXT_SIZE);
1901 
1902 	if (!exclusive && rss_spread == 1) {
1903 		*context = EFX_EF10_RSS_CONTEXT_INVALID;
1904 		if (context_size)
1905 			*context_size = 1;
1906 		return 0;
1907 	}
1908 
1909 	MCDI_SET_DWORD(inbuf, RSS_CONTEXT_ALLOC_IN_UPSTREAM_PORT_ID,
1910 		       nic_data->vport_id);
1911 	MCDI_SET_DWORD(inbuf, RSS_CONTEXT_ALLOC_IN_TYPE, alloc_type);
1912 	MCDI_SET_DWORD(inbuf, RSS_CONTEXT_ALLOC_IN_NUM_QUEUES, rss_spread);
1913 
1914 	rc = efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_ALLOC, inbuf, sizeof(inbuf),
1915 		outbuf, sizeof(outbuf), &outlen);
1916 	if (rc != 0)
1917 		return rc;
1918 
1919 	if (outlen < MC_CMD_RSS_CONTEXT_ALLOC_OUT_LEN)
1920 		return -EIO;
1921 
1922 	*context = MCDI_DWORD(outbuf, RSS_CONTEXT_ALLOC_OUT_RSS_CONTEXT_ID);
1923 
1924 	if (context_size)
1925 		*context_size = rss_spread;
1926 
1927 	return 0;
1928 }
1929 
1930 static void efx_ef10_free_rss_context(struct efx_nic *efx, u32 context)
1931 {
1932 	MCDI_DECLARE_BUF(inbuf, MC_CMD_RSS_CONTEXT_FREE_IN_LEN);
1933 	int rc;
1934 
1935 	MCDI_SET_DWORD(inbuf, RSS_CONTEXT_FREE_IN_RSS_CONTEXT_ID,
1936 		       context);
1937 
1938 	rc = efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_FREE, inbuf, sizeof(inbuf),
1939 			    NULL, 0, NULL);
1940 	WARN_ON(rc != 0);
1941 }
1942 
1943 static int efx_ef10_populate_rss_table(struct efx_nic *efx, u32 context,
1944 				       const u32 *rx_indir_table)
1945 {
1946 	MCDI_DECLARE_BUF(tablebuf, MC_CMD_RSS_CONTEXT_SET_TABLE_IN_LEN);
1947 	MCDI_DECLARE_BUF(keybuf, MC_CMD_RSS_CONTEXT_SET_KEY_IN_LEN);
1948 	int i, rc;
1949 
1950 	MCDI_SET_DWORD(tablebuf, RSS_CONTEXT_SET_TABLE_IN_RSS_CONTEXT_ID,
1951 		       context);
1952 	BUILD_BUG_ON(ARRAY_SIZE(efx->rx_indir_table) !=
1953 		     MC_CMD_RSS_CONTEXT_SET_TABLE_IN_INDIRECTION_TABLE_LEN);
1954 
1955 	for (i = 0; i < ARRAY_SIZE(efx->rx_indir_table); ++i)
1956 		MCDI_PTR(tablebuf,
1957 			 RSS_CONTEXT_SET_TABLE_IN_INDIRECTION_TABLE)[i] =
1958 				(u8) rx_indir_table[i];
1959 
1960 	rc = efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_SET_TABLE, tablebuf,
1961 			  sizeof(tablebuf), NULL, 0, NULL);
1962 	if (rc != 0)
1963 		return rc;
1964 
1965 	MCDI_SET_DWORD(keybuf, RSS_CONTEXT_SET_KEY_IN_RSS_CONTEXT_ID,
1966 		       context);
1967 	BUILD_BUG_ON(ARRAY_SIZE(efx->rx_hash_key) !=
1968 		     MC_CMD_RSS_CONTEXT_SET_KEY_IN_TOEPLITZ_KEY_LEN);
1969 	for (i = 0; i < ARRAY_SIZE(efx->rx_hash_key); ++i)
1970 		MCDI_PTR(keybuf, RSS_CONTEXT_SET_KEY_IN_TOEPLITZ_KEY)[i] =
1971 			efx->rx_hash_key[i];
1972 
1973 	return efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_SET_KEY, keybuf,
1974 			    sizeof(keybuf), NULL, 0, NULL);
1975 }
1976 
1977 static void efx_ef10_rx_free_indir_table(struct efx_nic *efx)
1978 {
1979 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1980 
1981 	if (nic_data->rx_rss_context != EFX_EF10_RSS_CONTEXT_INVALID)
1982 		efx_ef10_free_rss_context(efx, nic_data->rx_rss_context);
1983 	nic_data->rx_rss_context = EFX_EF10_RSS_CONTEXT_INVALID;
1984 }
1985 
1986 static int efx_ef10_rx_push_shared_rss_config(struct efx_nic *efx,
1987 					      unsigned *context_size)
1988 {
1989 	u32 new_rx_rss_context;
1990 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1991 	int rc = efx_ef10_alloc_rss_context(efx, &new_rx_rss_context,
1992 					    false, context_size);
1993 
1994 	if (rc != 0)
1995 		return rc;
1996 
1997 	nic_data->rx_rss_context = new_rx_rss_context;
1998 	nic_data->rx_rss_context_exclusive = false;
1999 	efx_set_default_rx_indir_table(efx);
2000 	return 0;
2001 }
2002 
2003 static int efx_ef10_rx_push_exclusive_rss_config(struct efx_nic *efx,
2004 						 const u32 *rx_indir_table)
2005 {
2006 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2007 	int rc;
2008 	u32 new_rx_rss_context;
2009 
2010 	if (nic_data->rx_rss_context == EFX_EF10_RSS_CONTEXT_INVALID ||
2011 	    !nic_data->rx_rss_context_exclusive) {
2012 		rc = efx_ef10_alloc_rss_context(efx, &new_rx_rss_context,
2013 						true, NULL);
2014 		if (rc == -EOPNOTSUPP)
2015 			return rc;
2016 		else if (rc != 0)
2017 			goto fail1;
2018 	} else {
2019 		new_rx_rss_context = nic_data->rx_rss_context;
2020 	}
2021 
2022 	rc = efx_ef10_populate_rss_table(efx, new_rx_rss_context,
2023 					 rx_indir_table);
2024 	if (rc != 0)
2025 		goto fail2;
2026 
2027 	if (nic_data->rx_rss_context != new_rx_rss_context)
2028 		efx_ef10_rx_free_indir_table(efx);
2029 	nic_data->rx_rss_context = new_rx_rss_context;
2030 	nic_data->rx_rss_context_exclusive = true;
2031 	if (rx_indir_table != efx->rx_indir_table)
2032 		memcpy(efx->rx_indir_table, rx_indir_table,
2033 		       sizeof(efx->rx_indir_table));
2034 	return 0;
2035 
2036 fail2:
2037 	if (new_rx_rss_context != nic_data->rx_rss_context)
2038 		efx_ef10_free_rss_context(efx, new_rx_rss_context);
2039 fail1:
2040 	netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
2041 	return rc;
2042 }
2043 
2044 static int efx_ef10_pf_rx_push_rss_config(struct efx_nic *efx, bool user,
2045 					  const u32 *rx_indir_table)
2046 {
2047 	int rc;
2048 
2049 	if (efx->rss_spread == 1)
2050 		return 0;
2051 
2052 	rc = efx_ef10_rx_push_exclusive_rss_config(efx, rx_indir_table);
2053 
2054 	if (rc == -ENOBUFS && !user) {
2055 		unsigned context_size;
2056 		bool mismatch = false;
2057 		size_t i;
2058 
2059 		for (i = 0; i < ARRAY_SIZE(efx->rx_indir_table) && !mismatch;
2060 		     i++)
2061 			mismatch = rx_indir_table[i] !=
2062 				ethtool_rxfh_indir_default(i, efx->rss_spread);
2063 
2064 		rc = efx_ef10_rx_push_shared_rss_config(efx, &context_size);
2065 		if (rc == 0) {
2066 			if (context_size != efx->rss_spread)
2067 				netif_warn(efx, probe, efx->net_dev,
2068 					   "Could not allocate an exclusive RSS"
2069 					   " context; allocated a shared one of"
2070 					   " different size."
2071 					   " Wanted %u, got %u.\n",
2072 					   efx->rss_spread, context_size);
2073 			else if (mismatch)
2074 				netif_warn(efx, probe, efx->net_dev,
2075 					   "Could not allocate an exclusive RSS"
2076 					   " context; allocated a shared one but"
2077 					   " could not apply custom"
2078 					   " indirection.\n");
2079 			else
2080 				netif_info(efx, probe, efx->net_dev,
2081 					   "Could not allocate an exclusive RSS"
2082 					   " context; allocated a shared one.\n");
2083 		}
2084 	}
2085 	return rc;
2086 }
2087 
2088 static int efx_ef10_vf_rx_push_rss_config(struct efx_nic *efx, bool user,
2089 					  const u32 *rx_indir_table
2090 					  __attribute__ ((unused)))
2091 {
2092 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2093 
2094 	if (user)
2095 		return -EOPNOTSUPP;
2096 	if (nic_data->rx_rss_context != EFX_EF10_RSS_CONTEXT_INVALID)
2097 		return 0;
2098 	return efx_ef10_rx_push_shared_rss_config(efx, NULL);
2099 }
2100 
2101 static int efx_ef10_rx_probe(struct efx_rx_queue *rx_queue)
2102 {
2103 	return efx_nic_alloc_buffer(rx_queue->efx, &rx_queue->rxd.buf,
2104 				    (rx_queue->ptr_mask + 1) *
2105 				    sizeof(efx_qword_t),
2106 				    GFP_KERNEL);
2107 }
2108 
2109 static void efx_ef10_rx_init(struct efx_rx_queue *rx_queue)
2110 {
2111 	MCDI_DECLARE_BUF(inbuf,
2112 			 MC_CMD_INIT_RXQ_IN_LEN(EFX_MAX_DMAQ_SIZE * 8 /
2113 						EFX_BUF_SIZE));
2114 	struct efx_channel *channel = efx_rx_queue_channel(rx_queue);
2115 	size_t entries = rx_queue->rxd.buf.len / EFX_BUF_SIZE;
2116 	struct efx_nic *efx = rx_queue->efx;
2117 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2118 	size_t inlen;
2119 	dma_addr_t dma_addr;
2120 	int rc;
2121 	int i;
2122 	BUILD_BUG_ON(MC_CMD_INIT_RXQ_OUT_LEN != 0);
2123 
2124 	rx_queue->scatter_n = 0;
2125 	rx_queue->scatter_len = 0;
2126 
2127 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_SIZE, rx_queue->ptr_mask + 1);
2128 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_TARGET_EVQ, channel->channel);
2129 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_LABEL, efx_rx_queue_index(rx_queue));
2130 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_INSTANCE,
2131 		       efx_rx_queue_index(rx_queue));
2132 	MCDI_POPULATE_DWORD_2(inbuf, INIT_RXQ_IN_FLAGS,
2133 			      INIT_RXQ_IN_FLAG_PREFIX, 1,
2134 			      INIT_RXQ_IN_FLAG_TIMESTAMP, 1);
2135 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_OWNER_ID, 0);
2136 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_PORT_ID, nic_data->vport_id);
2137 
2138 	dma_addr = rx_queue->rxd.buf.dma_addr;
2139 
2140 	netif_dbg(efx, hw, efx->net_dev, "pushing RXQ %d. %zu entries (%llx)\n",
2141 		  efx_rx_queue_index(rx_queue), entries, (u64)dma_addr);
2142 
2143 	for (i = 0; i < entries; ++i) {
2144 		MCDI_SET_ARRAY_QWORD(inbuf, INIT_RXQ_IN_DMA_ADDR, i, dma_addr);
2145 		dma_addr += EFX_BUF_SIZE;
2146 	}
2147 
2148 	inlen = MC_CMD_INIT_RXQ_IN_LEN(entries);
2149 
2150 	rc = efx_mcdi_rpc(efx, MC_CMD_INIT_RXQ, inbuf, inlen,
2151 			  NULL, 0, NULL);
2152 	if (rc)
2153 		netdev_WARN(efx->net_dev, "failed to initialise RXQ %d\n",
2154 			    efx_rx_queue_index(rx_queue));
2155 }
2156 
2157 static void efx_ef10_rx_fini(struct efx_rx_queue *rx_queue)
2158 {
2159 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FINI_RXQ_IN_LEN);
2160 	MCDI_DECLARE_BUF_ERR(outbuf);
2161 	struct efx_nic *efx = rx_queue->efx;
2162 	size_t outlen;
2163 	int rc;
2164 
2165 	MCDI_SET_DWORD(inbuf, FINI_RXQ_IN_INSTANCE,
2166 		       efx_rx_queue_index(rx_queue));
2167 
2168 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FINI_RXQ, inbuf, sizeof(inbuf),
2169 			  outbuf, sizeof(outbuf), &outlen);
2170 
2171 	if (rc && rc != -EALREADY)
2172 		goto fail;
2173 
2174 	return;
2175 
2176 fail:
2177 	efx_mcdi_display_error(efx, MC_CMD_FINI_RXQ, MC_CMD_FINI_RXQ_IN_LEN,
2178 			       outbuf, outlen, rc);
2179 }
2180 
2181 static void efx_ef10_rx_remove(struct efx_rx_queue *rx_queue)
2182 {
2183 	efx_nic_free_buffer(rx_queue->efx, &rx_queue->rxd.buf);
2184 }
2185 
2186 /* This creates an entry in the RX descriptor queue */
2187 static inline void
2188 efx_ef10_build_rx_desc(struct efx_rx_queue *rx_queue, unsigned int index)
2189 {
2190 	struct efx_rx_buffer *rx_buf;
2191 	efx_qword_t *rxd;
2192 
2193 	rxd = efx_rx_desc(rx_queue, index);
2194 	rx_buf = efx_rx_buffer(rx_queue, index);
2195 	EFX_POPULATE_QWORD_2(*rxd,
2196 			     ESF_DZ_RX_KER_BYTE_CNT, rx_buf->len,
2197 			     ESF_DZ_RX_KER_BUF_ADDR, rx_buf->dma_addr);
2198 }
2199 
2200 static void efx_ef10_rx_write(struct efx_rx_queue *rx_queue)
2201 {
2202 	struct efx_nic *efx = rx_queue->efx;
2203 	unsigned int write_count;
2204 	efx_dword_t reg;
2205 
2206 	/* Firmware requires that RX_DESC_WPTR be a multiple of 8 */
2207 	write_count = rx_queue->added_count & ~7;
2208 	if (rx_queue->notified_count == write_count)
2209 		return;
2210 
2211 	do
2212 		efx_ef10_build_rx_desc(
2213 			rx_queue,
2214 			rx_queue->notified_count & rx_queue->ptr_mask);
2215 	while (++rx_queue->notified_count != write_count);
2216 
2217 	wmb();
2218 	EFX_POPULATE_DWORD_1(reg, ERF_DZ_RX_DESC_WPTR,
2219 			     write_count & rx_queue->ptr_mask);
2220 	efx_writed_page(efx, &reg, ER_DZ_RX_DESC_UPD,
2221 			efx_rx_queue_index(rx_queue));
2222 }
2223 
2224 static efx_mcdi_async_completer efx_ef10_rx_defer_refill_complete;
2225 
2226 static void efx_ef10_rx_defer_refill(struct efx_rx_queue *rx_queue)
2227 {
2228 	struct efx_channel *channel = efx_rx_queue_channel(rx_queue);
2229 	MCDI_DECLARE_BUF(inbuf, MC_CMD_DRIVER_EVENT_IN_LEN);
2230 	efx_qword_t event;
2231 
2232 	EFX_POPULATE_QWORD_2(event,
2233 			     ESF_DZ_EV_CODE, EFX_EF10_DRVGEN_EV,
2234 			     ESF_DZ_EV_DATA, EFX_EF10_REFILL);
2235 
2236 	MCDI_SET_DWORD(inbuf, DRIVER_EVENT_IN_EVQ, channel->channel);
2237 
2238 	/* MCDI_SET_QWORD is not appropriate here since EFX_POPULATE_* has
2239 	 * already swapped the data to little-endian order.
2240 	 */
2241 	memcpy(MCDI_PTR(inbuf, DRIVER_EVENT_IN_DATA), &event.u64[0],
2242 	       sizeof(efx_qword_t));
2243 
2244 	efx_mcdi_rpc_async(channel->efx, MC_CMD_DRIVER_EVENT,
2245 			   inbuf, sizeof(inbuf), 0,
2246 			   efx_ef10_rx_defer_refill_complete, 0);
2247 }
2248 
2249 static void
2250 efx_ef10_rx_defer_refill_complete(struct efx_nic *efx, unsigned long cookie,
2251 				  int rc, efx_dword_t *outbuf,
2252 				  size_t outlen_actual)
2253 {
2254 	/* nothing to do */
2255 }
2256 
2257 static int efx_ef10_ev_probe(struct efx_channel *channel)
2258 {
2259 	return efx_nic_alloc_buffer(channel->efx, &channel->eventq.buf,
2260 				    (channel->eventq_mask + 1) *
2261 				    sizeof(efx_qword_t),
2262 				    GFP_KERNEL);
2263 }
2264 
2265 static void efx_ef10_ev_fini(struct efx_channel *channel)
2266 {
2267 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FINI_EVQ_IN_LEN);
2268 	MCDI_DECLARE_BUF_ERR(outbuf);
2269 	struct efx_nic *efx = channel->efx;
2270 	size_t outlen;
2271 	int rc;
2272 
2273 	MCDI_SET_DWORD(inbuf, FINI_EVQ_IN_INSTANCE, channel->channel);
2274 
2275 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FINI_EVQ, inbuf, sizeof(inbuf),
2276 			  outbuf, sizeof(outbuf), &outlen);
2277 
2278 	if (rc && rc != -EALREADY)
2279 		goto fail;
2280 
2281 	return;
2282 
2283 fail:
2284 	efx_mcdi_display_error(efx, MC_CMD_FINI_EVQ, MC_CMD_FINI_EVQ_IN_LEN,
2285 			       outbuf, outlen, rc);
2286 }
2287 
2288 static int efx_ef10_ev_init(struct efx_channel *channel)
2289 {
2290 	MCDI_DECLARE_BUF(inbuf,
2291 			 MC_CMD_INIT_EVQ_IN_LEN(EFX_MAX_EVQ_SIZE * 8 /
2292 						EFX_BUF_SIZE));
2293 	MCDI_DECLARE_BUF(outbuf, MC_CMD_INIT_EVQ_OUT_LEN);
2294 	size_t entries = channel->eventq.buf.len / EFX_BUF_SIZE;
2295 	struct efx_nic *efx = channel->efx;
2296 	struct efx_ef10_nic_data *nic_data;
2297 	bool supports_rx_merge;
2298 	size_t inlen, outlen;
2299 	unsigned int enabled, implemented;
2300 	dma_addr_t dma_addr;
2301 	int rc;
2302 	int i;
2303 
2304 	nic_data = efx->nic_data;
2305 	supports_rx_merge =
2306 		!!(nic_data->datapath_caps &
2307 		   1 << MC_CMD_GET_CAPABILITIES_OUT_RX_BATCHING_LBN);
2308 
2309 	/* Fill event queue with all ones (i.e. empty events) */
2310 	memset(channel->eventq.buf.addr, 0xff, channel->eventq.buf.len);
2311 
2312 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_SIZE, channel->eventq_mask + 1);
2313 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_INSTANCE, channel->channel);
2314 	/* INIT_EVQ expects index in vector table, not absolute */
2315 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_IRQ_NUM, channel->channel);
2316 	MCDI_POPULATE_DWORD_4(inbuf, INIT_EVQ_IN_FLAGS,
2317 			      INIT_EVQ_IN_FLAG_INTERRUPTING, 1,
2318 			      INIT_EVQ_IN_FLAG_RX_MERGE, 1,
2319 			      INIT_EVQ_IN_FLAG_TX_MERGE, 1,
2320 			      INIT_EVQ_IN_FLAG_CUT_THRU, !supports_rx_merge);
2321 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_TMR_MODE,
2322 		       MC_CMD_INIT_EVQ_IN_TMR_MODE_DIS);
2323 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_TMR_LOAD, 0);
2324 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_TMR_RELOAD, 0);
2325 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_COUNT_MODE,
2326 		       MC_CMD_INIT_EVQ_IN_COUNT_MODE_DIS);
2327 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_COUNT_THRSHLD, 0);
2328 
2329 	dma_addr = channel->eventq.buf.dma_addr;
2330 	for (i = 0; i < entries; ++i) {
2331 		MCDI_SET_ARRAY_QWORD(inbuf, INIT_EVQ_IN_DMA_ADDR, i, dma_addr);
2332 		dma_addr += EFX_BUF_SIZE;
2333 	}
2334 
2335 	inlen = MC_CMD_INIT_EVQ_IN_LEN(entries);
2336 
2337 	rc = efx_mcdi_rpc(efx, MC_CMD_INIT_EVQ, inbuf, inlen,
2338 			  outbuf, sizeof(outbuf), &outlen);
2339 	/* IRQ return is ignored */
2340 	if (channel->channel || rc)
2341 		return rc;
2342 
2343 	/* Successfully created event queue on channel 0 */
2344 	rc = efx_mcdi_get_workarounds(efx, &implemented, &enabled);
2345 	if (rc == -ENOSYS) {
2346 		/* GET_WORKAROUNDS was implemented before the bug26807
2347 		 * workaround, thus the latter must be unavailable in this fw
2348 		 */
2349 		nic_data->workaround_26807 = false;
2350 		rc = 0;
2351 	} else if (rc) {
2352 		goto fail;
2353 	} else {
2354 		nic_data->workaround_26807 =
2355 			!!(enabled & MC_CMD_GET_WORKAROUNDS_OUT_BUG26807);
2356 
2357 		if (implemented & MC_CMD_GET_WORKAROUNDS_OUT_BUG26807 &&
2358 		    !nic_data->workaround_26807) {
2359 			unsigned int flags;
2360 
2361 			rc = efx_mcdi_set_workaround(efx,
2362 						     MC_CMD_WORKAROUND_BUG26807,
2363 						     true, &flags);
2364 
2365 			if (!rc) {
2366 				if (flags &
2367 				    1 << MC_CMD_WORKAROUND_EXT_OUT_FLR_DONE_LBN) {
2368 					netif_info(efx, drv, efx->net_dev,
2369 						   "other functions on NIC have been reset\n");
2370 					/* MC's boot count has incremented */
2371 					++nic_data->warm_boot_count;
2372 				}
2373 				nic_data->workaround_26807 = true;
2374 			} else if (rc == -EPERM) {
2375 				rc = 0;
2376 			}
2377 		}
2378 	}
2379 
2380 	if (!rc)
2381 		return 0;
2382 
2383 fail:
2384 	efx_ef10_ev_fini(channel);
2385 	return rc;
2386 }
2387 
2388 static void efx_ef10_ev_remove(struct efx_channel *channel)
2389 {
2390 	efx_nic_free_buffer(channel->efx, &channel->eventq.buf);
2391 }
2392 
2393 static void efx_ef10_handle_rx_wrong_queue(struct efx_rx_queue *rx_queue,
2394 					   unsigned int rx_queue_label)
2395 {
2396 	struct efx_nic *efx = rx_queue->efx;
2397 
2398 	netif_info(efx, hw, efx->net_dev,
2399 		   "rx event arrived on queue %d labeled as queue %u\n",
2400 		   efx_rx_queue_index(rx_queue), rx_queue_label);
2401 
2402 	efx_schedule_reset(efx, RESET_TYPE_DISABLE);
2403 }
2404 
2405 static void
2406 efx_ef10_handle_rx_bad_lbits(struct efx_rx_queue *rx_queue,
2407 			     unsigned int actual, unsigned int expected)
2408 {
2409 	unsigned int dropped = (actual - expected) & rx_queue->ptr_mask;
2410 	struct efx_nic *efx = rx_queue->efx;
2411 
2412 	netif_info(efx, hw, efx->net_dev,
2413 		   "dropped %d events (index=%d expected=%d)\n",
2414 		   dropped, actual, expected);
2415 
2416 	efx_schedule_reset(efx, RESET_TYPE_DISABLE);
2417 }
2418 
2419 /* partially received RX was aborted. clean up. */
2420 static void efx_ef10_handle_rx_abort(struct efx_rx_queue *rx_queue)
2421 {
2422 	unsigned int rx_desc_ptr;
2423 
2424 	netif_dbg(rx_queue->efx, hw, rx_queue->efx->net_dev,
2425 		  "scattered RX aborted (dropping %u buffers)\n",
2426 		  rx_queue->scatter_n);
2427 
2428 	rx_desc_ptr = rx_queue->removed_count & rx_queue->ptr_mask;
2429 
2430 	efx_rx_packet(rx_queue, rx_desc_ptr, rx_queue->scatter_n,
2431 		      0, EFX_RX_PKT_DISCARD);
2432 
2433 	rx_queue->removed_count += rx_queue->scatter_n;
2434 	rx_queue->scatter_n = 0;
2435 	rx_queue->scatter_len = 0;
2436 	++efx_rx_queue_channel(rx_queue)->n_rx_nodesc_trunc;
2437 }
2438 
2439 static int efx_ef10_handle_rx_event(struct efx_channel *channel,
2440 				    const efx_qword_t *event)
2441 {
2442 	unsigned int rx_bytes, next_ptr_lbits, rx_queue_label, rx_l4_class;
2443 	unsigned int n_descs, n_packets, i;
2444 	struct efx_nic *efx = channel->efx;
2445 	struct efx_rx_queue *rx_queue;
2446 	bool rx_cont;
2447 	u16 flags = 0;
2448 
2449 	if (unlikely(ACCESS_ONCE(efx->reset_pending)))
2450 		return 0;
2451 
2452 	/* Basic packet information */
2453 	rx_bytes = EFX_QWORD_FIELD(*event, ESF_DZ_RX_BYTES);
2454 	next_ptr_lbits = EFX_QWORD_FIELD(*event, ESF_DZ_RX_DSC_PTR_LBITS);
2455 	rx_queue_label = EFX_QWORD_FIELD(*event, ESF_DZ_RX_QLABEL);
2456 	rx_l4_class = EFX_QWORD_FIELD(*event, ESF_DZ_RX_L4_CLASS);
2457 	rx_cont = EFX_QWORD_FIELD(*event, ESF_DZ_RX_CONT);
2458 
2459 	if (EFX_QWORD_FIELD(*event, ESF_DZ_RX_DROP_EVENT))
2460 		netdev_WARN(efx->net_dev, "saw RX_DROP_EVENT: event="
2461 			    EFX_QWORD_FMT "\n",
2462 			    EFX_QWORD_VAL(*event));
2463 
2464 	rx_queue = efx_channel_get_rx_queue(channel);
2465 
2466 	if (unlikely(rx_queue_label != efx_rx_queue_index(rx_queue)))
2467 		efx_ef10_handle_rx_wrong_queue(rx_queue, rx_queue_label);
2468 
2469 	n_descs = ((next_ptr_lbits - rx_queue->removed_count) &
2470 		   ((1 << ESF_DZ_RX_DSC_PTR_LBITS_WIDTH) - 1));
2471 
2472 	if (n_descs != rx_queue->scatter_n + 1) {
2473 		struct efx_ef10_nic_data *nic_data = efx->nic_data;
2474 
2475 		/* detect rx abort */
2476 		if (unlikely(n_descs == rx_queue->scatter_n)) {
2477 			if (rx_queue->scatter_n == 0 || rx_bytes != 0)
2478 				netdev_WARN(efx->net_dev,
2479 					    "invalid RX abort: scatter_n=%u event="
2480 					    EFX_QWORD_FMT "\n",
2481 					    rx_queue->scatter_n,
2482 					    EFX_QWORD_VAL(*event));
2483 			efx_ef10_handle_rx_abort(rx_queue);
2484 			return 0;
2485 		}
2486 
2487 		/* Check that RX completion merging is valid, i.e.
2488 		 * the current firmware supports it and this is a
2489 		 * non-scattered packet.
2490 		 */
2491 		if (!(nic_data->datapath_caps &
2492 		      (1 << MC_CMD_GET_CAPABILITIES_OUT_RX_BATCHING_LBN)) ||
2493 		    rx_queue->scatter_n != 0 || rx_cont) {
2494 			efx_ef10_handle_rx_bad_lbits(
2495 				rx_queue, next_ptr_lbits,
2496 				(rx_queue->removed_count +
2497 				 rx_queue->scatter_n + 1) &
2498 				((1 << ESF_DZ_RX_DSC_PTR_LBITS_WIDTH) - 1));
2499 			return 0;
2500 		}
2501 
2502 		/* Merged completion for multiple non-scattered packets */
2503 		rx_queue->scatter_n = 1;
2504 		rx_queue->scatter_len = 0;
2505 		n_packets = n_descs;
2506 		++channel->n_rx_merge_events;
2507 		channel->n_rx_merge_packets += n_packets;
2508 		flags |= EFX_RX_PKT_PREFIX_LEN;
2509 	} else {
2510 		++rx_queue->scatter_n;
2511 		rx_queue->scatter_len += rx_bytes;
2512 		if (rx_cont)
2513 			return 0;
2514 		n_packets = 1;
2515 	}
2516 
2517 	if (unlikely(EFX_QWORD_FIELD(*event, ESF_DZ_RX_ECRC_ERR)))
2518 		flags |= EFX_RX_PKT_DISCARD;
2519 
2520 	if (unlikely(EFX_QWORD_FIELD(*event, ESF_DZ_RX_IPCKSUM_ERR))) {
2521 		channel->n_rx_ip_hdr_chksum_err += n_packets;
2522 	} else if (unlikely(EFX_QWORD_FIELD(*event,
2523 					    ESF_DZ_RX_TCPUDP_CKSUM_ERR))) {
2524 		channel->n_rx_tcp_udp_chksum_err += n_packets;
2525 	} else if (rx_l4_class == ESE_DZ_L4_CLASS_TCP ||
2526 		   rx_l4_class == ESE_DZ_L4_CLASS_UDP) {
2527 		flags |= EFX_RX_PKT_CSUMMED;
2528 	}
2529 
2530 	if (rx_l4_class == ESE_DZ_L4_CLASS_TCP)
2531 		flags |= EFX_RX_PKT_TCP;
2532 
2533 	channel->irq_mod_score += 2 * n_packets;
2534 
2535 	/* Handle received packet(s) */
2536 	for (i = 0; i < n_packets; i++) {
2537 		efx_rx_packet(rx_queue,
2538 			      rx_queue->removed_count & rx_queue->ptr_mask,
2539 			      rx_queue->scatter_n, rx_queue->scatter_len,
2540 			      flags);
2541 		rx_queue->removed_count += rx_queue->scatter_n;
2542 	}
2543 
2544 	rx_queue->scatter_n = 0;
2545 	rx_queue->scatter_len = 0;
2546 
2547 	return n_packets;
2548 }
2549 
2550 static int
2551 efx_ef10_handle_tx_event(struct efx_channel *channel, efx_qword_t *event)
2552 {
2553 	struct efx_nic *efx = channel->efx;
2554 	struct efx_tx_queue *tx_queue;
2555 	unsigned int tx_ev_desc_ptr;
2556 	unsigned int tx_ev_q_label;
2557 	int tx_descs = 0;
2558 
2559 	if (unlikely(ACCESS_ONCE(efx->reset_pending)))
2560 		return 0;
2561 
2562 	if (unlikely(EFX_QWORD_FIELD(*event, ESF_DZ_TX_DROP_EVENT)))
2563 		return 0;
2564 
2565 	/* Transmit completion */
2566 	tx_ev_desc_ptr = EFX_QWORD_FIELD(*event, ESF_DZ_TX_DESCR_INDX);
2567 	tx_ev_q_label = EFX_QWORD_FIELD(*event, ESF_DZ_TX_QLABEL);
2568 	tx_queue = efx_channel_get_tx_queue(channel,
2569 					    tx_ev_q_label % EFX_TXQ_TYPES);
2570 	tx_descs = ((tx_ev_desc_ptr + 1 - tx_queue->read_count) &
2571 		    tx_queue->ptr_mask);
2572 	efx_xmit_done(tx_queue, tx_ev_desc_ptr & tx_queue->ptr_mask);
2573 
2574 	return tx_descs;
2575 }
2576 
2577 static void
2578 efx_ef10_handle_driver_event(struct efx_channel *channel, efx_qword_t *event)
2579 {
2580 	struct efx_nic *efx = channel->efx;
2581 	int subcode;
2582 
2583 	subcode = EFX_QWORD_FIELD(*event, ESF_DZ_DRV_SUB_CODE);
2584 
2585 	switch (subcode) {
2586 	case ESE_DZ_DRV_TIMER_EV:
2587 	case ESE_DZ_DRV_WAKE_UP_EV:
2588 		break;
2589 	case ESE_DZ_DRV_START_UP_EV:
2590 		/* event queue init complete. ok. */
2591 		break;
2592 	default:
2593 		netif_err(efx, hw, efx->net_dev,
2594 			  "channel %d unknown driver event type %d"
2595 			  " (data " EFX_QWORD_FMT ")\n",
2596 			  channel->channel, subcode,
2597 			  EFX_QWORD_VAL(*event));
2598 
2599 	}
2600 }
2601 
2602 static void efx_ef10_handle_driver_generated_event(struct efx_channel *channel,
2603 						   efx_qword_t *event)
2604 {
2605 	struct efx_nic *efx = channel->efx;
2606 	u32 subcode;
2607 
2608 	subcode = EFX_QWORD_FIELD(*event, EFX_DWORD_0);
2609 
2610 	switch (subcode) {
2611 	case EFX_EF10_TEST:
2612 		channel->event_test_cpu = raw_smp_processor_id();
2613 		break;
2614 	case EFX_EF10_REFILL:
2615 		/* The queue must be empty, so we won't receive any rx
2616 		 * events, so efx_process_channel() won't refill the
2617 		 * queue. Refill it here
2618 		 */
2619 		efx_fast_push_rx_descriptors(&channel->rx_queue, true);
2620 		break;
2621 	default:
2622 		netif_err(efx, hw, efx->net_dev,
2623 			  "channel %d unknown driver event type %u"
2624 			  " (data " EFX_QWORD_FMT ")\n",
2625 			  channel->channel, (unsigned) subcode,
2626 			  EFX_QWORD_VAL(*event));
2627 	}
2628 }
2629 
2630 static int efx_ef10_ev_process(struct efx_channel *channel, int quota)
2631 {
2632 	struct efx_nic *efx = channel->efx;
2633 	efx_qword_t event, *p_event;
2634 	unsigned int read_ptr;
2635 	int ev_code;
2636 	int tx_descs = 0;
2637 	int spent = 0;
2638 
2639 	if (quota <= 0)
2640 		return spent;
2641 
2642 	read_ptr = channel->eventq_read_ptr;
2643 
2644 	for (;;) {
2645 		p_event = efx_event(channel, read_ptr);
2646 		event = *p_event;
2647 
2648 		if (!efx_event_present(&event))
2649 			break;
2650 
2651 		EFX_SET_QWORD(*p_event);
2652 
2653 		++read_ptr;
2654 
2655 		ev_code = EFX_QWORD_FIELD(event, ESF_DZ_EV_CODE);
2656 
2657 		netif_vdbg(efx, drv, efx->net_dev,
2658 			   "processing event on %d " EFX_QWORD_FMT "\n",
2659 			   channel->channel, EFX_QWORD_VAL(event));
2660 
2661 		switch (ev_code) {
2662 		case ESE_DZ_EV_CODE_MCDI_EV:
2663 			efx_mcdi_process_event(channel, &event);
2664 			break;
2665 		case ESE_DZ_EV_CODE_RX_EV:
2666 			spent += efx_ef10_handle_rx_event(channel, &event);
2667 			if (spent >= quota) {
2668 				/* XXX can we split a merged event to
2669 				 * avoid going over-quota?
2670 				 */
2671 				spent = quota;
2672 				goto out;
2673 			}
2674 			break;
2675 		case ESE_DZ_EV_CODE_TX_EV:
2676 			tx_descs += efx_ef10_handle_tx_event(channel, &event);
2677 			if (tx_descs > efx->txq_entries) {
2678 				spent = quota;
2679 				goto out;
2680 			} else if (++spent == quota) {
2681 				goto out;
2682 			}
2683 			break;
2684 		case ESE_DZ_EV_CODE_DRIVER_EV:
2685 			efx_ef10_handle_driver_event(channel, &event);
2686 			if (++spent == quota)
2687 				goto out;
2688 			break;
2689 		case EFX_EF10_DRVGEN_EV:
2690 			efx_ef10_handle_driver_generated_event(channel, &event);
2691 			break;
2692 		default:
2693 			netif_err(efx, hw, efx->net_dev,
2694 				  "channel %d unknown event type %d"
2695 				  " (data " EFX_QWORD_FMT ")\n",
2696 				  channel->channel, ev_code,
2697 				  EFX_QWORD_VAL(event));
2698 		}
2699 	}
2700 
2701 out:
2702 	channel->eventq_read_ptr = read_ptr;
2703 	return spent;
2704 }
2705 
2706 static void efx_ef10_ev_read_ack(struct efx_channel *channel)
2707 {
2708 	struct efx_nic *efx = channel->efx;
2709 	efx_dword_t rptr;
2710 
2711 	if (EFX_EF10_WORKAROUND_35388(efx)) {
2712 		BUILD_BUG_ON(EFX_MIN_EVQ_SIZE <
2713 			     (1 << ERF_DD_EVQ_IND_RPTR_WIDTH));
2714 		BUILD_BUG_ON(EFX_MAX_EVQ_SIZE >
2715 			     (1 << 2 * ERF_DD_EVQ_IND_RPTR_WIDTH));
2716 
2717 		EFX_POPULATE_DWORD_2(rptr, ERF_DD_EVQ_IND_RPTR_FLAGS,
2718 				     EFE_DD_EVQ_IND_RPTR_FLAGS_HIGH,
2719 				     ERF_DD_EVQ_IND_RPTR,
2720 				     (channel->eventq_read_ptr &
2721 				      channel->eventq_mask) >>
2722 				     ERF_DD_EVQ_IND_RPTR_WIDTH);
2723 		efx_writed_page(efx, &rptr, ER_DD_EVQ_INDIRECT,
2724 				channel->channel);
2725 		EFX_POPULATE_DWORD_2(rptr, ERF_DD_EVQ_IND_RPTR_FLAGS,
2726 				     EFE_DD_EVQ_IND_RPTR_FLAGS_LOW,
2727 				     ERF_DD_EVQ_IND_RPTR,
2728 				     channel->eventq_read_ptr &
2729 				     ((1 << ERF_DD_EVQ_IND_RPTR_WIDTH) - 1));
2730 		efx_writed_page(efx, &rptr, ER_DD_EVQ_INDIRECT,
2731 				channel->channel);
2732 	} else {
2733 		EFX_POPULATE_DWORD_1(rptr, ERF_DZ_EVQ_RPTR,
2734 				     channel->eventq_read_ptr &
2735 				     channel->eventq_mask);
2736 		efx_writed_page(efx, &rptr, ER_DZ_EVQ_RPTR, channel->channel);
2737 	}
2738 }
2739 
2740 static void efx_ef10_ev_test_generate(struct efx_channel *channel)
2741 {
2742 	MCDI_DECLARE_BUF(inbuf, MC_CMD_DRIVER_EVENT_IN_LEN);
2743 	struct efx_nic *efx = channel->efx;
2744 	efx_qword_t event;
2745 	int rc;
2746 
2747 	EFX_POPULATE_QWORD_2(event,
2748 			     ESF_DZ_EV_CODE, EFX_EF10_DRVGEN_EV,
2749 			     ESF_DZ_EV_DATA, EFX_EF10_TEST);
2750 
2751 	MCDI_SET_DWORD(inbuf, DRIVER_EVENT_IN_EVQ, channel->channel);
2752 
2753 	/* MCDI_SET_QWORD is not appropriate here since EFX_POPULATE_* has
2754 	 * already swapped the data to little-endian order.
2755 	 */
2756 	memcpy(MCDI_PTR(inbuf, DRIVER_EVENT_IN_DATA), &event.u64[0],
2757 	       sizeof(efx_qword_t));
2758 
2759 	rc = efx_mcdi_rpc(efx, MC_CMD_DRIVER_EVENT, inbuf, sizeof(inbuf),
2760 			  NULL, 0, NULL);
2761 	if (rc != 0)
2762 		goto fail;
2763 
2764 	return;
2765 
2766 fail:
2767 	WARN_ON(true);
2768 	netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
2769 }
2770 
2771 void efx_ef10_handle_drain_event(struct efx_nic *efx)
2772 {
2773 	if (atomic_dec_and_test(&efx->active_queues))
2774 		wake_up(&efx->flush_wq);
2775 
2776 	WARN_ON(atomic_read(&efx->active_queues) < 0);
2777 }
2778 
2779 static int efx_ef10_fini_dmaq(struct efx_nic *efx)
2780 {
2781 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2782 	struct efx_channel *channel;
2783 	struct efx_tx_queue *tx_queue;
2784 	struct efx_rx_queue *rx_queue;
2785 	int pending;
2786 
2787 	/* If the MC has just rebooted, the TX/RX queues will have already been
2788 	 * torn down, but efx->active_queues needs to be set to zero.
2789 	 */
2790 	if (nic_data->must_realloc_vis) {
2791 		atomic_set(&efx->active_queues, 0);
2792 		return 0;
2793 	}
2794 
2795 	/* Do not attempt to write to the NIC during EEH recovery */
2796 	if (efx->state != STATE_RECOVERY) {
2797 		efx_for_each_channel(channel, efx) {
2798 			efx_for_each_channel_rx_queue(rx_queue, channel)
2799 				efx_ef10_rx_fini(rx_queue);
2800 			efx_for_each_channel_tx_queue(tx_queue, channel)
2801 				efx_ef10_tx_fini(tx_queue);
2802 		}
2803 
2804 		wait_event_timeout(efx->flush_wq,
2805 				   atomic_read(&efx->active_queues) == 0,
2806 				   msecs_to_jiffies(EFX_MAX_FLUSH_TIME));
2807 		pending = atomic_read(&efx->active_queues);
2808 		if (pending) {
2809 			netif_err(efx, hw, efx->net_dev, "failed to flush %d queues\n",
2810 				  pending);
2811 			return -ETIMEDOUT;
2812 		}
2813 	}
2814 
2815 	return 0;
2816 }
2817 
2818 static void efx_ef10_prepare_flr(struct efx_nic *efx)
2819 {
2820 	atomic_set(&efx->active_queues, 0);
2821 }
2822 
2823 static bool efx_ef10_filter_equal(const struct efx_filter_spec *left,
2824 				  const struct efx_filter_spec *right)
2825 {
2826 	if ((left->match_flags ^ right->match_flags) |
2827 	    ((left->flags ^ right->flags) &
2828 	     (EFX_FILTER_FLAG_RX | EFX_FILTER_FLAG_TX)))
2829 		return false;
2830 
2831 	return memcmp(&left->outer_vid, &right->outer_vid,
2832 		      sizeof(struct efx_filter_spec) -
2833 		      offsetof(struct efx_filter_spec, outer_vid)) == 0;
2834 }
2835 
2836 static unsigned int efx_ef10_filter_hash(const struct efx_filter_spec *spec)
2837 {
2838 	BUILD_BUG_ON(offsetof(struct efx_filter_spec, outer_vid) & 3);
2839 	return jhash2((const u32 *)&spec->outer_vid,
2840 		      (sizeof(struct efx_filter_spec) -
2841 		       offsetof(struct efx_filter_spec, outer_vid)) / 4,
2842 		      0);
2843 	/* XXX should we randomise the initval? */
2844 }
2845 
2846 /* Decide whether a filter should be exclusive or else should allow
2847  * delivery to additional recipients.  Currently we decide that
2848  * filters for specific local unicast MAC and IP addresses are
2849  * exclusive.
2850  */
2851 static bool efx_ef10_filter_is_exclusive(const struct efx_filter_spec *spec)
2852 {
2853 	if (spec->match_flags & EFX_FILTER_MATCH_LOC_MAC &&
2854 	    !is_multicast_ether_addr(spec->loc_mac))
2855 		return true;
2856 
2857 	if ((spec->match_flags &
2858 	     (EFX_FILTER_MATCH_ETHER_TYPE | EFX_FILTER_MATCH_LOC_HOST)) ==
2859 	    (EFX_FILTER_MATCH_ETHER_TYPE | EFX_FILTER_MATCH_LOC_HOST)) {
2860 		if (spec->ether_type == htons(ETH_P_IP) &&
2861 		    !ipv4_is_multicast(spec->loc_host[0]))
2862 			return true;
2863 		if (spec->ether_type == htons(ETH_P_IPV6) &&
2864 		    ((const u8 *)spec->loc_host)[0] != 0xff)
2865 			return true;
2866 	}
2867 
2868 	return false;
2869 }
2870 
2871 static struct efx_filter_spec *
2872 efx_ef10_filter_entry_spec(const struct efx_ef10_filter_table *table,
2873 			   unsigned int filter_idx)
2874 {
2875 	return (struct efx_filter_spec *)(table->entry[filter_idx].spec &
2876 					  ~EFX_EF10_FILTER_FLAGS);
2877 }
2878 
2879 static unsigned int
2880 efx_ef10_filter_entry_flags(const struct efx_ef10_filter_table *table,
2881 			   unsigned int filter_idx)
2882 {
2883 	return table->entry[filter_idx].spec & EFX_EF10_FILTER_FLAGS;
2884 }
2885 
2886 static void
2887 efx_ef10_filter_set_entry(struct efx_ef10_filter_table *table,
2888 			  unsigned int filter_idx,
2889 			  const struct efx_filter_spec *spec,
2890 			  unsigned int flags)
2891 {
2892 	table->entry[filter_idx].spec =	(unsigned long)spec | flags;
2893 }
2894 
2895 static void efx_ef10_filter_push_prep(struct efx_nic *efx,
2896 				      const struct efx_filter_spec *spec,
2897 				      efx_dword_t *inbuf, u64 handle,
2898 				      bool replacing)
2899 {
2900 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2901 
2902 	memset(inbuf, 0, MC_CMD_FILTER_OP_IN_LEN);
2903 
2904 	if (replacing) {
2905 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
2906 			       MC_CMD_FILTER_OP_IN_OP_REPLACE);
2907 		MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE, handle);
2908 	} else {
2909 		u32 match_fields = 0;
2910 
2911 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
2912 			       efx_ef10_filter_is_exclusive(spec) ?
2913 			       MC_CMD_FILTER_OP_IN_OP_INSERT :
2914 			       MC_CMD_FILTER_OP_IN_OP_SUBSCRIBE);
2915 
2916 		/* Convert match flags and values.  Unlike almost
2917 		 * everything else in MCDI, these fields are in
2918 		 * network byte order.
2919 		 */
2920 		if (spec->match_flags & EFX_FILTER_MATCH_LOC_MAC_IG)
2921 			match_fields |=
2922 				is_multicast_ether_addr(spec->loc_mac) ?
2923 				1 << MC_CMD_FILTER_OP_IN_MATCH_UNKNOWN_MCAST_DST_LBN :
2924 				1 << MC_CMD_FILTER_OP_IN_MATCH_UNKNOWN_UCAST_DST_LBN;
2925 #define COPY_FIELD(gen_flag, gen_field, mcdi_field)			     \
2926 		if (spec->match_flags & EFX_FILTER_MATCH_ ## gen_flag) {     \
2927 			match_fields |=					     \
2928 				1 << MC_CMD_FILTER_OP_IN_MATCH_ ##	     \
2929 				mcdi_field ## _LBN;			     \
2930 			BUILD_BUG_ON(					     \
2931 				MC_CMD_FILTER_OP_IN_ ## mcdi_field ## _LEN < \
2932 				sizeof(spec->gen_field));		     \
2933 			memcpy(MCDI_PTR(inbuf, FILTER_OP_IN_ ##	mcdi_field), \
2934 			       &spec->gen_field, sizeof(spec->gen_field));   \
2935 		}
2936 		COPY_FIELD(REM_HOST, rem_host, SRC_IP);
2937 		COPY_FIELD(LOC_HOST, loc_host, DST_IP);
2938 		COPY_FIELD(REM_MAC, rem_mac, SRC_MAC);
2939 		COPY_FIELD(REM_PORT, rem_port, SRC_PORT);
2940 		COPY_FIELD(LOC_MAC, loc_mac, DST_MAC);
2941 		COPY_FIELD(LOC_PORT, loc_port, DST_PORT);
2942 		COPY_FIELD(ETHER_TYPE, ether_type, ETHER_TYPE);
2943 		COPY_FIELD(INNER_VID, inner_vid, INNER_VLAN);
2944 		COPY_FIELD(OUTER_VID, outer_vid, OUTER_VLAN);
2945 		COPY_FIELD(IP_PROTO, ip_proto, IP_PROTO);
2946 #undef COPY_FIELD
2947 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_MATCH_FIELDS,
2948 			       match_fields);
2949 	}
2950 
2951 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_PORT_ID, nic_data->vport_id);
2952 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_DEST,
2953 		       spec->dmaq_id == EFX_FILTER_RX_DMAQ_ID_DROP ?
2954 		       MC_CMD_FILTER_OP_IN_RX_DEST_DROP :
2955 		       MC_CMD_FILTER_OP_IN_RX_DEST_HOST);
2956 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_TX_DOMAIN, 0);
2957 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_TX_DEST,
2958 		       MC_CMD_FILTER_OP_IN_TX_DEST_DEFAULT);
2959 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_QUEUE,
2960 		       spec->dmaq_id == EFX_FILTER_RX_DMAQ_ID_DROP ?
2961 		       0 : spec->dmaq_id);
2962 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_MODE,
2963 		       (spec->flags & EFX_FILTER_FLAG_RX_RSS) ?
2964 		       MC_CMD_FILTER_OP_IN_RX_MODE_RSS :
2965 		       MC_CMD_FILTER_OP_IN_RX_MODE_SIMPLE);
2966 	if (spec->flags & EFX_FILTER_FLAG_RX_RSS)
2967 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_CONTEXT,
2968 			       spec->rss_context !=
2969 			       EFX_FILTER_RSS_CONTEXT_DEFAULT ?
2970 			       spec->rss_context : nic_data->rx_rss_context);
2971 }
2972 
2973 static int efx_ef10_filter_push(struct efx_nic *efx,
2974 				const struct efx_filter_spec *spec,
2975 				u64 *handle, bool replacing)
2976 {
2977 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_IN_LEN);
2978 	MCDI_DECLARE_BUF(outbuf, MC_CMD_FILTER_OP_OUT_LEN);
2979 	int rc;
2980 
2981 	efx_ef10_filter_push_prep(efx, spec, inbuf, *handle, replacing);
2982 	rc = efx_mcdi_rpc(efx, MC_CMD_FILTER_OP, inbuf, sizeof(inbuf),
2983 			  outbuf, sizeof(outbuf), NULL);
2984 	if (rc == 0)
2985 		*handle = MCDI_QWORD(outbuf, FILTER_OP_OUT_HANDLE);
2986 	if (rc == -ENOSPC)
2987 		rc = -EBUSY; /* to match efx_farch_filter_insert() */
2988 	return rc;
2989 }
2990 
2991 static int efx_ef10_filter_rx_match_pri(struct efx_ef10_filter_table *table,
2992 					enum efx_filter_match_flags match_flags)
2993 {
2994 	unsigned int match_pri;
2995 
2996 	for (match_pri = 0;
2997 	     match_pri < table->rx_match_count;
2998 	     match_pri++)
2999 		if (table->rx_match_flags[match_pri] == match_flags)
3000 			return match_pri;
3001 
3002 	return -EPROTONOSUPPORT;
3003 }
3004 
3005 static s32 efx_ef10_filter_insert(struct efx_nic *efx,
3006 				  struct efx_filter_spec *spec,
3007 				  bool replace_equal)
3008 {
3009 	struct efx_ef10_filter_table *table = efx->filter_state;
3010 	DECLARE_BITMAP(mc_rem_map, EFX_EF10_FILTER_SEARCH_LIMIT);
3011 	struct efx_filter_spec *saved_spec;
3012 	unsigned int match_pri, hash;
3013 	unsigned int priv_flags;
3014 	bool replacing = false;
3015 	int ins_index = -1;
3016 	DEFINE_WAIT(wait);
3017 	bool is_mc_recip;
3018 	s32 rc;
3019 
3020 	/* For now, only support RX filters */
3021 	if ((spec->flags & (EFX_FILTER_FLAG_RX | EFX_FILTER_FLAG_TX)) !=
3022 	    EFX_FILTER_FLAG_RX)
3023 		return -EINVAL;
3024 
3025 	rc = efx_ef10_filter_rx_match_pri(table, spec->match_flags);
3026 	if (rc < 0)
3027 		return rc;
3028 	match_pri = rc;
3029 
3030 	hash = efx_ef10_filter_hash(spec);
3031 	is_mc_recip = efx_filter_is_mc_recipient(spec);
3032 	if (is_mc_recip)
3033 		bitmap_zero(mc_rem_map, EFX_EF10_FILTER_SEARCH_LIMIT);
3034 
3035 	/* Find any existing filters with the same match tuple or
3036 	 * else a free slot to insert at.  If any of them are busy,
3037 	 * we have to wait and retry.
3038 	 */
3039 	for (;;) {
3040 		unsigned int depth = 1;
3041 		unsigned int i;
3042 
3043 		spin_lock_bh(&efx->filter_lock);
3044 
3045 		for (;;) {
3046 			i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
3047 			saved_spec = efx_ef10_filter_entry_spec(table, i);
3048 
3049 			if (!saved_spec) {
3050 				if (ins_index < 0)
3051 					ins_index = i;
3052 			} else if (efx_ef10_filter_equal(spec, saved_spec)) {
3053 				if (table->entry[i].spec &
3054 				    EFX_EF10_FILTER_FLAG_BUSY)
3055 					break;
3056 				if (spec->priority < saved_spec->priority &&
3057 				    spec->priority != EFX_FILTER_PRI_AUTO) {
3058 					rc = -EPERM;
3059 					goto out_unlock;
3060 				}
3061 				if (!is_mc_recip) {
3062 					/* This is the only one */
3063 					if (spec->priority ==
3064 					    saved_spec->priority &&
3065 					    !replace_equal) {
3066 						rc = -EEXIST;
3067 						goto out_unlock;
3068 					}
3069 					ins_index = i;
3070 					goto found;
3071 				} else if (spec->priority >
3072 					   saved_spec->priority ||
3073 					   (spec->priority ==
3074 					    saved_spec->priority &&
3075 					    replace_equal)) {
3076 					if (ins_index < 0)
3077 						ins_index = i;
3078 					else
3079 						__set_bit(depth, mc_rem_map);
3080 				}
3081 			}
3082 
3083 			/* Once we reach the maximum search depth, use
3084 			 * the first suitable slot or return -EBUSY if
3085 			 * there was none
3086 			 */
3087 			if (depth == EFX_EF10_FILTER_SEARCH_LIMIT) {
3088 				if (ins_index < 0) {
3089 					rc = -EBUSY;
3090 					goto out_unlock;
3091 				}
3092 				goto found;
3093 			}
3094 
3095 			++depth;
3096 		}
3097 
3098 		prepare_to_wait(&table->waitq, &wait, TASK_UNINTERRUPTIBLE);
3099 		spin_unlock_bh(&efx->filter_lock);
3100 		schedule();
3101 	}
3102 
3103 found:
3104 	/* Create a software table entry if necessary, and mark it
3105 	 * busy.  We might yet fail to insert, but any attempt to
3106 	 * insert a conflicting filter while we're waiting for the
3107 	 * firmware must find the busy entry.
3108 	 */
3109 	saved_spec = efx_ef10_filter_entry_spec(table, ins_index);
3110 	if (saved_spec) {
3111 		if (spec->priority == EFX_FILTER_PRI_AUTO &&
3112 		    saved_spec->priority >= EFX_FILTER_PRI_AUTO) {
3113 			/* Just make sure it won't be removed */
3114 			if (saved_spec->priority > EFX_FILTER_PRI_AUTO)
3115 				saved_spec->flags |= EFX_FILTER_FLAG_RX_OVER_AUTO;
3116 			table->entry[ins_index].spec &=
3117 				~EFX_EF10_FILTER_FLAG_AUTO_OLD;
3118 			rc = ins_index;
3119 			goto out_unlock;
3120 		}
3121 		replacing = true;
3122 		priv_flags = efx_ef10_filter_entry_flags(table, ins_index);
3123 	} else {
3124 		saved_spec = kmalloc(sizeof(*spec), GFP_ATOMIC);
3125 		if (!saved_spec) {
3126 			rc = -ENOMEM;
3127 			goto out_unlock;
3128 		}
3129 		*saved_spec = *spec;
3130 		priv_flags = 0;
3131 	}
3132 	efx_ef10_filter_set_entry(table, ins_index, saved_spec,
3133 				  priv_flags | EFX_EF10_FILTER_FLAG_BUSY);
3134 
3135 	/* Mark lower-priority multicast recipients busy prior to removal */
3136 	if (is_mc_recip) {
3137 		unsigned int depth, i;
3138 
3139 		for (depth = 0; depth < EFX_EF10_FILTER_SEARCH_LIMIT; depth++) {
3140 			i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
3141 			if (test_bit(depth, mc_rem_map))
3142 				table->entry[i].spec |=
3143 					EFX_EF10_FILTER_FLAG_BUSY;
3144 		}
3145 	}
3146 
3147 	spin_unlock_bh(&efx->filter_lock);
3148 
3149 	rc = efx_ef10_filter_push(efx, spec, &table->entry[ins_index].handle,
3150 				  replacing);
3151 
3152 	/* Finalise the software table entry */
3153 	spin_lock_bh(&efx->filter_lock);
3154 	if (rc == 0) {
3155 		if (replacing) {
3156 			/* Update the fields that may differ */
3157 			if (saved_spec->priority == EFX_FILTER_PRI_AUTO)
3158 				saved_spec->flags |=
3159 					EFX_FILTER_FLAG_RX_OVER_AUTO;
3160 			saved_spec->priority = spec->priority;
3161 			saved_spec->flags &= EFX_FILTER_FLAG_RX_OVER_AUTO;
3162 			saved_spec->flags |= spec->flags;
3163 			saved_spec->rss_context = spec->rss_context;
3164 			saved_spec->dmaq_id = spec->dmaq_id;
3165 		}
3166 	} else if (!replacing) {
3167 		kfree(saved_spec);
3168 		saved_spec = NULL;
3169 	}
3170 	efx_ef10_filter_set_entry(table, ins_index, saved_spec, priv_flags);
3171 
3172 	/* Remove and finalise entries for lower-priority multicast
3173 	 * recipients
3174 	 */
3175 	if (is_mc_recip) {
3176 		MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_IN_LEN);
3177 		unsigned int depth, i;
3178 
3179 		memset(inbuf, 0, sizeof(inbuf));
3180 
3181 		for (depth = 0; depth < EFX_EF10_FILTER_SEARCH_LIMIT; depth++) {
3182 			if (!test_bit(depth, mc_rem_map))
3183 				continue;
3184 
3185 			i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
3186 			saved_spec = efx_ef10_filter_entry_spec(table, i);
3187 			priv_flags = efx_ef10_filter_entry_flags(table, i);
3188 
3189 			if (rc == 0) {
3190 				spin_unlock_bh(&efx->filter_lock);
3191 				MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
3192 					       MC_CMD_FILTER_OP_IN_OP_UNSUBSCRIBE);
3193 				MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
3194 					       table->entry[i].handle);
3195 				rc = efx_mcdi_rpc(efx, MC_CMD_FILTER_OP,
3196 						  inbuf, sizeof(inbuf),
3197 						  NULL, 0, NULL);
3198 				spin_lock_bh(&efx->filter_lock);
3199 			}
3200 
3201 			if (rc == 0) {
3202 				kfree(saved_spec);
3203 				saved_spec = NULL;
3204 				priv_flags = 0;
3205 			} else {
3206 				priv_flags &= ~EFX_EF10_FILTER_FLAG_BUSY;
3207 			}
3208 			efx_ef10_filter_set_entry(table, i, saved_spec,
3209 						  priv_flags);
3210 		}
3211 	}
3212 
3213 	/* If successful, return the inserted filter ID */
3214 	if (rc == 0)
3215 		rc = match_pri * HUNT_FILTER_TBL_ROWS + ins_index;
3216 
3217 	wake_up_all(&table->waitq);
3218 out_unlock:
3219 	spin_unlock_bh(&efx->filter_lock);
3220 	finish_wait(&table->waitq, &wait);
3221 	return rc;
3222 }
3223 
3224 static void efx_ef10_filter_update_rx_scatter(struct efx_nic *efx)
3225 {
3226 	/* no need to do anything here on EF10 */
3227 }
3228 
3229 /* Remove a filter.
3230  * If !by_index, remove by ID
3231  * If by_index, remove by index
3232  * Filter ID may come from userland and must be range-checked.
3233  */
3234 static int efx_ef10_filter_remove_internal(struct efx_nic *efx,
3235 					   unsigned int priority_mask,
3236 					   u32 filter_id, bool by_index)
3237 {
3238 	unsigned int filter_idx = filter_id % HUNT_FILTER_TBL_ROWS;
3239 	struct efx_ef10_filter_table *table = efx->filter_state;
3240 	MCDI_DECLARE_BUF(inbuf,
3241 			 MC_CMD_FILTER_OP_IN_HANDLE_OFST +
3242 			 MC_CMD_FILTER_OP_IN_HANDLE_LEN);
3243 	struct efx_filter_spec *spec;
3244 	DEFINE_WAIT(wait);
3245 	int rc;
3246 
3247 	/* Find the software table entry and mark it busy.  Don't
3248 	 * remove it yet; any attempt to update while we're waiting
3249 	 * for the firmware must find the busy entry.
3250 	 */
3251 	for (;;) {
3252 		spin_lock_bh(&efx->filter_lock);
3253 		if (!(table->entry[filter_idx].spec &
3254 		      EFX_EF10_FILTER_FLAG_BUSY))
3255 			break;
3256 		prepare_to_wait(&table->waitq, &wait, TASK_UNINTERRUPTIBLE);
3257 		spin_unlock_bh(&efx->filter_lock);
3258 		schedule();
3259 	}
3260 
3261 	spec = efx_ef10_filter_entry_spec(table, filter_idx);
3262 	if (!spec ||
3263 	    (!by_index &&
3264 	     efx_ef10_filter_rx_match_pri(table, spec->match_flags) !=
3265 	     filter_id / HUNT_FILTER_TBL_ROWS)) {
3266 		rc = -ENOENT;
3267 		goto out_unlock;
3268 	}
3269 
3270 	if (spec->flags & EFX_FILTER_FLAG_RX_OVER_AUTO &&
3271 	    priority_mask == (1U << EFX_FILTER_PRI_AUTO)) {
3272 		/* Just remove flags */
3273 		spec->flags &= ~EFX_FILTER_FLAG_RX_OVER_AUTO;
3274 		table->entry[filter_idx].spec &= ~EFX_EF10_FILTER_FLAG_AUTO_OLD;
3275 		rc = 0;
3276 		goto out_unlock;
3277 	}
3278 
3279 	if (!(priority_mask & (1U << spec->priority))) {
3280 		rc = -ENOENT;
3281 		goto out_unlock;
3282 	}
3283 
3284 	table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_BUSY;
3285 	spin_unlock_bh(&efx->filter_lock);
3286 
3287 	if (spec->flags & EFX_FILTER_FLAG_RX_OVER_AUTO) {
3288 		/* Reset to an automatic filter */
3289 
3290 		struct efx_filter_spec new_spec = *spec;
3291 
3292 		new_spec.priority = EFX_FILTER_PRI_AUTO;
3293 		new_spec.flags = (EFX_FILTER_FLAG_RX |
3294 				  EFX_FILTER_FLAG_RX_RSS);
3295 		new_spec.dmaq_id = 0;
3296 		new_spec.rss_context = EFX_FILTER_RSS_CONTEXT_DEFAULT;
3297 		rc = efx_ef10_filter_push(efx, &new_spec,
3298 					  &table->entry[filter_idx].handle,
3299 					  true);
3300 
3301 		spin_lock_bh(&efx->filter_lock);
3302 		if (rc == 0)
3303 			*spec = new_spec;
3304 	} else {
3305 		/* Really remove the filter */
3306 
3307 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
3308 			       efx_ef10_filter_is_exclusive(spec) ?
3309 			       MC_CMD_FILTER_OP_IN_OP_REMOVE :
3310 			       MC_CMD_FILTER_OP_IN_OP_UNSUBSCRIBE);
3311 		MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
3312 			       table->entry[filter_idx].handle);
3313 		rc = efx_mcdi_rpc(efx, MC_CMD_FILTER_OP,
3314 				  inbuf, sizeof(inbuf), NULL, 0, NULL);
3315 
3316 		spin_lock_bh(&efx->filter_lock);
3317 		if (rc == 0) {
3318 			kfree(spec);
3319 			efx_ef10_filter_set_entry(table, filter_idx, NULL, 0);
3320 		}
3321 	}
3322 
3323 	table->entry[filter_idx].spec &= ~EFX_EF10_FILTER_FLAG_BUSY;
3324 	wake_up_all(&table->waitq);
3325 out_unlock:
3326 	spin_unlock_bh(&efx->filter_lock);
3327 	finish_wait(&table->waitq, &wait);
3328 	return rc;
3329 }
3330 
3331 static int efx_ef10_filter_remove_safe(struct efx_nic *efx,
3332 				       enum efx_filter_priority priority,
3333 				       u32 filter_id)
3334 {
3335 	return efx_ef10_filter_remove_internal(efx, 1U << priority,
3336 					       filter_id, false);
3337 }
3338 
3339 static u32 efx_ef10_filter_get_unsafe_id(struct efx_nic *efx, u32 filter_id)
3340 {
3341 	return filter_id % HUNT_FILTER_TBL_ROWS;
3342 }
3343 
3344 static int efx_ef10_filter_remove_unsafe(struct efx_nic *efx,
3345 					 enum efx_filter_priority priority,
3346 					 u32 filter_id)
3347 {
3348 	return efx_ef10_filter_remove_internal(efx, 1U << priority,
3349 					       filter_id, true);
3350 }
3351 
3352 static int efx_ef10_filter_get_safe(struct efx_nic *efx,
3353 				    enum efx_filter_priority priority,
3354 				    u32 filter_id, struct efx_filter_spec *spec)
3355 {
3356 	unsigned int filter_idx = filter_id % HUNT_FILTER_TBL_ROWS;
3357 	struct efx_ef10_filter_table *table = efx->filter_state;
3358 	const struct efx_filter_spec *saved_spec;
3359 	int rc;
3360 
3361 	spin_lock_bh(&efx->filter_lock);
3362 	saved_spec = efx_ef10_filter_entry_spec(table, filter_idx);
3363 	if (saved_spec && saved_spec->priority == priority &&
3364 	    efx_ef10_filter_rx_match_pri(table, saved_spec->match_flags) ==
3365 	    filter_id / HUNT_FILTER_TBL_ROWS) {
3366 		*spec = *saved_spec;
3367 		rc = 0;
3368 	} else {
3369 		rc = -ENOENT;
3370 	}
3371 	spin_unlock_bh(&efx->filter_lock);
3372 	return rc;
3373 }
3374 
3375 static int efx_ef10_filter_clear_rx(struct efx_nic *efx,
3376 				     enum efx_filter_priority priority)
3377 {
3378 	unsigned int priority_mask;
3379 	unsigned int i;
3380 	int rc;
3381 
3382 	priority_mask = (((1U << (priority + 1)) - 1) &
3383 			 ~(1U << EFX_FILTER_PRI_AUTO));
3384 
3385 	for (i = 0; i < HUNT_FILTER_TBL_ROWS; i++) {
3386 		rc = efx_ef10_filter_remove_internal(efx, priority_mask,
3387 						     i, true);
3388 		if (rc && rc != -ENOENT)
3389 			return rc;
3390 	}
3391 
3392 	return 0;
3393 }
3394 
3395 static u32 efx_ef10_filter_count_rx_used(struct efx_nic *efx,
3396 					 enum efx_filter_priority priority)
3397 {
3398 	struct efx_ef10_filter_table *table = efx->filter_state;
3399 	unsigned int filter_idx;
3400 	s32 count = 0;
3401 
3402 	spin_lock_bh(&efx->filter_lock);
3403 	for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
3404 		if (table->entry[filter_idx].spec &&
3405 		    efx_ef10_filter_entry_spec(table, filter_idx)->priority ==
3406 		    priority)
3407 			++count;
3408 	}
3409 	spin_unlock_bh(&efx->filter_lock);
3410 	return count;
3411 }
3412 
3413 static u32 efx_ef10_filter_get_rx_id_limit(struct efx_nic *efx)
3414 {
3415 	struct efx_ef10_filter_table *table = efx->filter_state;
3416 
3417 	return table->rx_match_count * HUNT_FILTER_TBL_ROWS;
3418 }
3419 
3420 static s32 efx_ef10_filter_get_rx_ids(struct efx_nic *efx,
3421 				      enum efx_filter_priority priority,
3422 				      u32 *buf, u32 size)
3423 {
3424 	struct efx_ef10_filter_table *table = efx->filter_state;
3425 	struct efx_filter_spec *spec;
3426 	unsigned int filter_idx;
3427 	s32 count = 0;
3428 
3429 	spin_lock_bh(&efx->filter_lock);
3430 	for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
3431 		spec = efx_ef10_filter_entry_spec(table, filter_idx);
3432 		if (spec && spec->priority == priority) {
3433 			if (count == size) {
3434 				count = -EMSGSIZE;
3435 				break;
3436 			}
3437 			buf[count++] = (efx_ef10_filter_rx_match_pri(
3438 						table, spec->match_flags) *
3439 					HUNT_FILTER_TBL_ROWS +
3440 					filter_idx);
3441 		}
3442 	}
3443 	spin_unlock_bh(&efx->filter_lock);
3444 	return count;
3445 }
3446 
3447 #ifdef CONFIG_RFS_ACCEL
3448 
3449 static efx_mcdi_async_completer efx_ef10_filter_rfs_insert_complete;
3450 
3451 static s32 efx_ef10_filter_rfs_insert(struct efx_nic *efx,
3452 				      struct efx_filter_spec *spec)
3453 {
3454 	struct efx_ef10_filter_table *table = efx->filter_state;
3455 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_IN_LEN);
3456 	struct efx_filter_spec *saved_spec;
3457 	unsigned int hash, i, depth = 1;
3458 	bool replacing = false;
3459 	int ins_index = -1;
3460 	u64 cookie;
3461 	s32 rc;
3462 
3463 	/* Must be an RX filter without RSS and not for a multicast
3464 	 * destination address (RFS only works for connected sockets).
3465 	 * These restrictions allow us to pass only a tiny amount of
3466 	 * data through to the completion function.
3467 	 */
3468 	EFX_WARN_ON_PARANOID(spec->flags !=
3469 			     (EFX_FILTER_FLAG_RX | EFX_FILTER_FLAG_RX_SCATTER));
3470 	EFX_WARN_ON_PARANOID(spec->priority != EFX_FILTER_PRI_HINT);
3471 	EFX_WARN_ON_PARANOID(efx_filter_is_mc_recipient(spec));
3472 
3473 	hash = efx_ef10_filter_hash(spec);
3474 
3475 	spin_lock_bh(&efx->filter_lock);
3476 
3477 	/* Find any existing filter with the same match tuple or else
3478 	 * a free slot to insert at.  If an existing filter is busy,
3479 	 * we have to give up.
3480 	 */
3481 	for (;;) {
3482 		i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
3483 		saved_spec = efx_ef10_filter_entry_spec(table, i);
3484 
3485 		if (!saved_spec) {
3486 			if (ins_index < 0)
3487 				ins_index = i;
3488 		} else if (efx_ef10_filter_equal(spec, saved_spec)) {
3489 			if (table->entry[i].spec & EFX_EF10_FILTER_FLAG_BUSY) {
3490 				rc = -EBUSY;
3491 				goto fail_unlock;
3492 			}
3493 			if (spec->priority < saved_spec->priority) {
3494 				rc = -EPERM;
3495 				goto fail_unlock;
3496 			}
3497 			ins_index = i;
3498 			break;
3499 		}
3500 
3501 		/* Once we reach the maximum search depth, use the
3502 		 * first suitable slot or return -EBUSY if there was
3503 		 * none
3504 		 */
3505 		if (depth == EFX_EF10_FILTER_SEARCH_LIMIT) {
3506 			if (ins_index < 0) {
3507 				rc = -EBUSY;
3508 				goto fail_unlock;
3509 			}
3510 			break;
3511 		}
3512 
3513 		++depth;
3514 	}
3515 
3516 	/* Create a software table entry if necessary, and mark it
3517 	 * busy.  We might yet fail to insert, but any attempt to
3518 	 * insert a conflicting filter while we're waiting for the
3519 	 * firmware must find the busy entry.
3520 	 */
3521 	saved_spec = efx_ef10_filter_entry_spec(table, ins_index);
3522 	if (saved_spec) {
3523 		replacing = true;
3524 	} else {
3525 		saved_spec = kmalloc(sizeof(*spec), GFP_ATOMIC);
3526 		if (!saved_spec) {
3527 			rc = -ENOMEM;
3528 			goto fail_unlock;
3529 		}
3530 		*saved_spec = *spec;
3531 	}
3532 	efx_ef10_filter_set_entry(table, ins_index, saved_spec,
3533 				  EFX_EF10_FILTER_FLAG_BUSY);
3534 
3535 	spin_unlock_bh(&efx->filter_lock);
3536 
3537 	/* Pack up the variables needed on completion */
3538 	cookie = replacing << 31 | ins_index << 16 | spec->dmaq_id;
3539 
3540 	efx_ef10_filter_push_prep(efx, spec, inbuf,
3541 				  table->entry[ins_index].handle, replacing);
3542 	efx_mcdi_rpc_async(efx, MC_CMD_FILTER_OP, inbuf, sizeof(inbuf),
3543 			   MC_CMD_FILTER_OP_OUT_LEN,
3544 			   efx_ef10_filter_rfs_insert_complete, cookie);
3545 
3546 	return ins_index;
3547 
3548 fail_unlock:
3549 	spin_unlock_bh(&efx->filter_lock);
3550 	return rc;
3551 }
3552 
3553 static void
3554 efx_ef10_filter_rfs_insert_complete(struct efx_nic *efx, unsigned long cookie,
3555 				    int rc, efx_dword_t *outbuf,
3556 				    size_t outlen_actual)
3557 {
3558 	struct efx_ef10_filter_table *table = efx->filter_state;
3559 	unsigned int ins_index, dmaq_id;
3560 	struct efx_filter_spec *spec;
3561 	bool replacing;
3562 
3563 	/* Unpack the cookie */
3564 	replacing = cookie >> 31;
3565 	ins_index = (cookie >> 16) & (HUNT_FILTER_TBL_ROWS - 1);
3566 	dmaq_id = cookie & 0xffff;
3567 
3568 	spin_lock_bh(&efx->filter_lock);
3569 	spec = efx_ef10_filter_entry_spec(table, ins_index);
3570 	if (rc == 0) {
3571 		table->entry[ins_index].handle =
3572 			MCDI_QWORD(outbuf, FILTER_OP_OUT_HANDLE);
3573 		if (replacing)
3574 			spec->dmaq_id = dmaq_id;
3575 	} else if (!replacing) {
3576 		kfree(spec);
3577 		spec = NULL;
3578 	}
3579 	efx_ef10_filter_set_entry(table, ins_index, spec, 0);
3580 	spin_unlock_bh(&efx->filter_lock);
3581 
3582 	wake_up_all(&table->waitq);
3583 }
3584 
3585 static void
3586 efx_ef10_filter_rfs_expire_complete(struct efx_nic *efx,
3587 				    unsigned long filter_idx,
3588 				    int rc, efx_dword_t *outbuf,
3589 				    size_t outlen_actual);
3590 
3591 static bool efx_ef10_filter_rfs_expire_one(struct efx_nic *efx, u32 flow_id,
3592 					   unsigned int filter_idx)
3593 {
3594 	struct efx_ef10_filter_table *table = efx->filter_state;
3595 	struct efx_filter_spec *spec =
3596 		efx_ef10_filter_entry_spec(table, filter_idx);
3597 	MCDI_DECLARE_BUF(inbuf,
3598 			 MC_CMD_FILTER_OP_IN_HANDLE_OFST +
3599 			 MC_CMD_FILTER_OP_IN_HANDLE_LEN);
3600 
3601 	if (!spec ||
3602 	    (table->entry[filter_idx].spec & EFX_EF10_FILTER_FLAG_BUSY) ||
3603 	    spec->priority != EFX_FILTER_PRI_HINT ||
3604 	    !rps_may_expire_flow(efx->net_dev, spec->dmaq_id,
3605 				 flow_id, filter_idx))
3606 		return false;
3607 
3608 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
3609 		       MC_CMD_FILTER_OP_IN_OP_REMOVE);
3610 	MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
3611 		       table->entry[filter_idx].handle);
3612 	if (efx_mcdi_rpc_async(efx, MC_CMD_FILTER_OP, inbuf, sizeof(inbuf), 0,
3613 			       efx_ef10_filter_rfs_expire_complete, filter_idx))
3614 		return false;
3615 
3616 	table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_BUSY;
3617 	return true;
3618 }
3619 
3620 static void
3621 efx_ef10_filter_rfs_expire_complete(struct efx_nic *efx,
3622 				    unsigned long filter_idx,
3623 				    int rc, efx_dword_t *outbuf,
3624 				    size_t outlen_actual)
3625 {
3626 	struct efx_ef10_filter_table *table = efx->filter_state;
3627 	struct efx_filter_spec *spec =
3628 		efx_ef10_filter_entry_spec(table, filter_idx);
3629 
3630 	spin_lock_bh(&efx->filter_lock);
3631 	if (rc == 0) {
3632 		kfree(spec);
3633 		efx_ef10_filter_set_entry(table, filter_idx, NULL, 0);
3634 	}
3635 	table->entry[filter_idx].spec &= ~EFX_EF10_FILTER_FLAG_BUSY;
3636 	wake_up_all(&table->waitq);
3637 	spin_unlock_bh(&efx->filter_lock);
3638 }
3639 
3640 #endif /* CONFIG_RFS_ACCEL */
3641 
3642 static int efx_ef10_filter_match_flags_from_mcdi(u32 mcdi_flags)
3643 {
3644 	int match_flags = 0;
3645 
3646 #define MAP_FLAG(gen_flag, mcdi_field) {				\
3647 		u32 old_mcdi_flags = mcdi_flags;			\
3648 		mcdi_flags &= ~(1 << MC_CMD_FILTER_OP_IN_MATCH_ ##	\
3649 				mcdi_field ## _LBN);			\
3650 		if (mcdi_flags != old_mcdi_flags)			\
3651 			match_flags |= EFX_FILTER_MATCH_ ## gen_flag;	\
3652 	}
3653 	MAP_FLAG(LOC_MAC_IG, UNKNOWN_UCAST_DST);
3654 	MAP_FLAG(LOC_MAC_IG, UNKNOWN_MCAST_DST);
3655 	MAP_FLAG(REM_HOST, SRC_IP);
3656 	MAP_FLAG(LOC_HOST, DST_IP);
3657 	MAP_FLAG(REM_MAC, SRC_MAC);
3658 	MAP_FLAG(REM_PORT, SRC_PORT);
3659 	MAP_FLAG(LOC_MAC, DST_MAC);
3660 	MAP_FLAG(LOC_PORT, DST_PORT);
3661 	MAP_FLAG(ETHER_TYPE, ETHER_TYPE);
3662 	MAP_FLAG(INNER_VID, INNER_VLAN);
3663 	MAP_FLAG(OUTER_VID, OUTER_VLAN);
3664 	MAP_FLAG(IP_PROTO, IP_PROTO);
3665 #undef MAP_FLAG
3666 
3667 	/* Did we map them all? */
3668 	if (mcdi_flags)
3669 		return -EINVAL;
3670 
3671 	return match_flags;
3672 }
3673 
3674 static int efx_ef10_filter_table_probe(struct efx_nic *efx)
3675 {
3676 	MCDI_DECLARE_BUF(inbuf, MC_CMD_GET_PARSER_DISP_INFO_IN_LEN);
3677 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_PARSER_DISP_INFO_OUT_LENMAX);
3678 	unsigned int pd_match_pri, pd_match_count;
3679 	struct efx_ef10_filter_table *table;
3680 	size_t outlen;
3681 	int rc;
3682 
3683 	table = kzalloc(sizeof(*table), GFP_KERNEL);
3684 	if (!table)
3685 		return -ENOMEM;
3686 
3687 	/* Find out which RX filter types are supported, and their priorities */
3688 	MCDI_SET_DWORD(inbuf, GET_PARSER_DISP_INFO_IN_OP,
3689 		       MC_CMD_GET_PARSER_DISP_INFO_IN_OP_GET_SUPPORTED_RX_MATCHES);
3690 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_PARSER_DISP_INFO,
3691 			  inbuf, sizeof(inbuf), outbuf, sizeof(outbuf),
3692 			  &outlen);
3693 	if (rc)
3694 		goto fail;
3695 	pd_match_count = MCDI_VAR_ARRAY_LEN(
3696 		outlen, GET_PARSER_DISP_INFO_OUT_SUPPORTED_MATCHES);
3697 	table->rx_match_count = 0;
3698 
3699 	for (pd_match_pri = 0; pd_match_pri < pd_match_count; pd_match_pri++) {
3700 		u32 mcdi_flags =
3701 			MCDI_ARRAY_DWORD(
3702 				outbuf,
3703 				GET_PARSER_DISP_INFO_OUT_SUPPORTED_MATCHES,
3704 				pd_match_pri);
3705 		rc = efx_ef10_filter_match_flags_from_mcdi(mcdi_flags);
3706 		if (rc < 0) {
3707 			netif_dbg(efx, probe, efx->net_dev,
3708 				  "%s: fw flags %#x pri %u not supported in driver\n",
3709 				  __func__, mcdi_flags, pd_match_pri);
3710 		} else {
3711 			netif_dbg(efx, probe, efx->net_dev,
3712 				  "%s: fw flags %#x pri %u supported as driver flags %#x pri %u\n",
3713 				  __func__, mcdi_flags, pd_match_pri,
3714 				  rc, table->rx_match_count);
3715 			table->rx_match_flags[table->rx_match_count++] = rc;
3716 		}
3717 	}
3718 
3719 	table->entry = vzalloc(HUNT_FILTER_TBL_ROWS * sizeof(*table->entry));
3720 	if (!table->entry) {
3721 		rc = -ENOMEM;
3722 		goto fail;
3723 	}
3724 
3725 	table->ucdef_id = EFX_EF10_FILTER_ID_INVALID;
3726 	table->bcast_id = EFX_EF10_FILTER_ID_INVALID;
3727 	table->mcdef_id = EFX_EF10_FILTER_ID_INVALID;
3728 
3729 	efx->filter_state = table;
3730 	init_waitqueue_head(&table->waitq);
3731 	return 0;
3732 
3733 fail:
3734 	kfree(table);
3735 	return rc;
3736 }
3737 
3738 /* Caller must hold efx->filter_sem for read if race against
3739  * efx_ef10_filter_table_remove() is possible
3740  */
3741 static void efx_ef10_filter_table_restore(struct efx_nic *efx)
3742 {
3743 	struct efx_ef10_filter_table *table = efx->filter_state;
3744 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
3745 	struct efx_filter_spec *spec;
3746 	unsigned int filter_idx;
3747 	bool failed = false;
3748 	int rc;
3749 
3750 	WARN_ON(!rwsem_is_locked(&efx->filter_sem));
3751 
3752 	if (!nic_data->must_restore_filters)
3753 		return;
3754 
3755 	if (!table)
3756 		return;
3757 
3758 	spin_lock_bh(&efx->filter_lock);
3759 
3760 	for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
3761 		spec = efx_ef10_filter_entry_spec(table, filter_idx);
3762 		if (!spec)
3763 			continue;
3764 
3765 		table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_BUSY;
3766 		spin_unlock_bh(&efx->filter_lock);
3767 
3768 		rc = efx_ef10_filter_push(efx, spec,
3769 					  &table->entry[filter_idx].handle,
3770 					  false);
3771 		if (rc)
3772 			failed = true;
3773 
3774 		spin_lock_bh(&efx->filter_lock);
3775 		if (rc) {
3776 			kfree(spec);
3777 			efx_ef10_filter_set_entry(table, filter_idx, NULL, 0);
3778 		} else {
3779 			table->entry[filter_idx].spec &=
3780 				~EFX_EF10_FILTER_FLAG_BUSY;
3781 		}
3782 	}
3783 
3784 	spin_unlock_bh(&efx->filter_lock);
3785 
3786 	if (failed)
3787 		netif_err(efx, hw, efx->net_dev,
3788 			  "unable to restore all filters\n");
3789 	else
3790 		nic_data->must_restore_filters = false;
3791 }
3792 
3793 /* Caller must hold efx->filter_sem for write */
3794 static void efx_ef10_filter_table_remove(struct efx_nic *efx)
3795 {
3796 	struct efx_ef10_filter_table *table = efx->filter_state;
3797 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_IN_LEN);
3798 	struct efx_filter_spec *spec;
3799 	unsigned int filter_idx;
3800 	int rc;
3801 
3802 	efx->filter_state = NULL;
3803 	if (!table)
3804 		return;
3805 
3806 	for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
3807 		spec = efx_ef10_filter_entry_spec(table, filter_idx);
3808 		if (!spec)
3809 			continue;
3810 
3811 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
3812 			       efx_ef10_filter_is_exclusive(spec) ?
3813 			       MC_CMD_FILTER_OP_IN_OP_REMOVE :
3814 			       MC_CMD_FILTER_OP_IN_OP_UNSUBSCRIBE);
3815 		MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
3816 			       table->entry[filter_idx].handle);
3817 		rc = efx_mcdi_rpc(efx, MC_CMD_FILTER_OP, inbuf, sizeof(inbuf),
3818 				  NULL, 0, NULL);
3819 		if (rc)
3820 			netdev_WARN(efx->net_dev,
3821 				    "filter_idx=%#x handle=%#llx\n",
3822 				    filter_idx,
3823 				    table->entry[filter_idx].handle);
3824 		kfree(spec);
3825 	}
3826 
3827 	vfree(table->entry);
3828 	kfree(table);
3829 }
3830 
3831 #define EFX_EF10_FILTER_DO_MARK_OLD(id) \
3832 		if (id != EFX_EF10_FILTER_ID_INVALID) { \
3833 			filter_idx = efx_ef10_filter_get_unsafe_id(efx, id); \
3834 			WARN_ON(!table->entry[filter_idx].spec); \
3835 			table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_AUTO_OLD; \
3836 		}
3837 static void efx_ef10_filter_mark_old(struct efx_nic *efx)
3838 {
3839 	struct efx_ef10_filter_table *table = efx->filter_state;
3840 	unsigned int filter_idx, i;
3841 
3842 	if (!table)
3843 		return;
3844 
3845 	/* Mark old filters that may need to be removed */
3846 	spin_lock_bh(&efx->filter_lock);
3847 	for (i = 0; i < table->dev_uc_count; i++)
3848 		EFX_EF10_FILTER_DO_MARK_OLD(table->dev_uc_list[i].id);
3849 	for (i = 0; i < table->dev_mc_count; i++)
3850 		EFX_EF10_FILTER_DO_MARK_OLD(table->dev_mc_list[i].id);
3851 	EFX_EF10_FILTER_DO_MARK_OLD(table->ucdef_id);
3852 	EFX_EF10_FILTER_DO_MARK_OLD(table->bcast_id);
3853 	EFX_EF10_FILTER_DO_MARK_OLD(table->mcdef_id);
3854 	spin_unlock_bh(&efx->filter_lock);
3855 }
3856 #undef EFX_EF10_FILTER_DO_MARK_OLD
3857 
3858 static void efx_ef10_filter_uc_addr_list(struct efx_nic *efx, bool *promisc)
3859 {
3860 	struct efx_ef10_filter_table *table = efx->filter_state;
3861 	struct net_device *net_dev = efx->net_dev;
3862 	struct netdev_hw_addr *uc;
3863 	int addr_count;
3864 	unsigned int i;
3865 
3866 	table->ucdef_id = EFX_EF10_FILTER_ID_INVALID;
3867 	addr_count = netdev_uc_count(net_dev);
3868 	if (net_dev->flags & IFF_PROMISC)
3869 		*promisc = true;
3870 	table->dev_uc_count = 1 + addr_count;
3871 	ether_addr_copy(table->dev_uc_list[0].addr, net_dev->dev_addr);
3872 	i = 1;
3873 	netdev_for_each_uc_addr(uc, net_dev) {
3874 		if (i >= EFX_EF10_FILTER_DEV_UC_MAX) {
3875 			*promisc = true;
3876 			break;
3877 		}
3878 		ether_addr_copy(table->dev_uc_list[i].addr, uc->addr);
3879 		table->dev_uc_list[i].id = EFX_EF10_FILTER_ID_INVALID;
3880 		i++;
3881 	}
3882 }
3883 
3884 static void efx_ef10_filter_mc_addr_list(struct efx_nic *efx, bool *promisc)
3885 {
3886 	struct efx_ef10_filter_table *table = efx->filter_state;
3887 	struct net_device *net_dev = efx->net_dev;
3888 	struct netdev_hw_addr *mc;
3889 	unsigned int i, addr_count;
3890 
3891 	table->mcdef_id = EFX_EF10_FILTER_ID_INVALID;
3892 	table->bcast_id = EFX_EF10_FILTER_ID_INVALID;
3893 	if (net_dev->flags & (IFF_PROMISC | IFF_ALLMULTI))
3894 		*promisc = true;
3895 
3896 	addr_count = netdev_mc_count(net_dev);
3897 	i = 0;
3898 	netdev_for_each_mc_addr(mc, net_dev) {
3899 		if (i >= EFX_EF10_FILTER_DEV_MC_MAX) {
3900 			*promisc = true;
3901 			break;
3902 		}
3903 		ether_addr_copy(table->dev_mc_list[i].addr, mc->addr);
3904 		table->dev_mc_list[i].id = EFX_EF10_FILTER_ID_INVALID;
3905 		i++;
3906 	}
3907 
3908 	table->dev_mc_count = i;
3909 }
3910 
3911 static int efx_ef10_filter_insert_addr_list(struct efx_nic *efx,
3912 					     bool multicast, bool rollback)
3913 {
3914 	struct efx_ef10_filter_table *table = efx->filter_state;
3915 	struct efx_ef10_dev_addr *addr_list;
3916 	struct efx_filter_spec spec;
3917 	u8 baddr[ETH_ALEN];
3918 	unsigned int i, j;
3919 	int addr_count;
3920 	int rc;
3921 
3922 	if (multicast) {
3923 		addr_list = table->dev_mc_list;
3924 		addr_count = table->dev_mc_count;
3925 	} else {
3926 		addr_list = table->dev_uc_list;
3927 		addr_count = table->dev_uc_count;
3928 	}
3929 
3930 	/* Insert/renew filters */
3931 	for (i = 0; i < addr_count; i++) {
3932 		efx_filter_init_rx(&spec, EFX_FILTER_PRI_AUTO,
3933 				   EFX_FILTER_FLAG_RX_RSS,
3934 				   0);
3935 		efx_filter_set_eth_local(&spec, EFX_FILTER_VID_UNSPEC,
3936 					 addr_list[i].addr);
3937 		rc = efx_ef10_filter_insert(efx, &spec, true);
3938 		if (rc < 0) {
3939 			if (rollback) {
3940 				netif_info(efx, drv, efx->net_dev,
3941 					   "efx_ef10_filter_insert failed rc=%d\n",
3942 					   rc);
3943 				/* Fall back to promiscuous */
3944 				for (j = 0; j < i; j++) {
3945 					if (addr_list[j].id == EFX_EF10_FILTER_ID_INVALID)
3946 						continue;
3947 					efx_ef10_filter_remove_unsafe(
3948 						efx, EFX_FILTER_PRI_AUTO,
3949 						addr_list[j].id);
3950 					addr_list[j].id = EFX_EF10_FILTER_ID_INVALID;
3951 				}
3952 				return rc;
3953 			} else {
3954 				/* mark as not inserted, and carry on */
3955 				rc = EFX_EF10_FILTER_ID_INVALID;
3956 			}
3957 		}
3958 		addr_list[i].id = efx_ef10_filter_get_unsafe_id(efx, rc);
3959 	}
3960 
3961 	if (multicast && rollback) {
3962 		/* Also need an Ethernet broadcast filter */
3963 		efx_filter_init_rx(&spec, EFX_FILTER_PRI_AUTO,
3964 				   EFX_FILTER_FLAG_RX_RSS,
3965 				   0);
3966 		eth_broadcast_addr(baddr);
3967 		efx_filter_set_eth_local(&spec, EFX_FILTER_VID_UNSPEC, baddr);
3968 		rc = efx_ef10_filter_insert(efx, &spec, true);
3969 		if (rc < 0) {
3970 			netif_warn(efx, drv, efx->net_dev,
3971 				   "Broadcast filter insert failed rc=%d\n", rc);
3972 			/* Fall back to promiscuous */
3973 			for (j = 0; j < i; j++) {
3974 				if (addr_list[j].id == EFX_EF10_FILTER_ID_INVALID)
3975 					continue;
3976 				efx_ef10_filter_remove_unsafe(
3977 					efx, EFX_FILTER_PRI_AUTO,
3978 					addr_list[j].id);
3979 				addr_list[j].id = EFX_EF10_FILTER_ID_INVALID;
3980 			}
3981 			return rc;
3982 		} else {
3983 			table->bcast_id = efx_ef10_filter_get_unsafe_id(efx, rc);
3984 		}
3985 	}
3986 
3987 	return 0;
3988 }
3989 
3990 static int efx_ef10_filter_insert_def(struct efx_nic *efx, bool multicast,
3991 				      bool rollback)
3992 {
3993 	struct efx_ef10_filter_table *table = efx->filter_state;
3994 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
3995 	struct efx_filter_spec spec;
3996 	u8 baddr[ETH_ALEN];
3997 	int rc;
3998 
3999 	efx_filter_init_rx(&spec, EFX_FILTER_PRI_AUTO,
4000 			   EFX_FILTER_FLAG_RX_RSS,
4001 			   0);
4002 
4003 	if (multicast)
4004 		efx_filter_set_mc_def(&spec);
4005 	else
4006 		efx_filter_set_uc_def(&spec);
4007 
4008 	rc = efx_ef10_filter_insert(efx, &spec, true);
4009 	if (rc < 0) {
4010 		netif_warn(efx, drv, efx->net_dev,
4011 			   "%scast mismatch filter insert failed rc=%d\n",
4012 			   multicast ? "Multi" : "Uni", rc);
4013 	} else if (multicast) {
4014 		table->mcdef_id = efx_ef10_filter_get_unsafe_id(efx, rc);
4015 		if (!nic_data->workaround_26807) {
4016 			/* Also need an Ethernet broadcast filter */
4017 			efx_filter_init_rx(&spec, EFX_FILTER_PRI_AUTO,
4018 					   EFX_FILTER_FLAG_RX_RSS,
4019 					   0);
4020 			eth_broadcast_addr(baddr);
4021 			efx_filter_set_eth_local(&spec, EFX_FILTER_VID_UNSPEC,
4022 						 baddr);
4023 			rc = efx_ef10_filter_insert(efx, &spec, true);
4024 			if (rc < 0) {
4025 				netif_warn(efx, drv, efx->net_dev,
4026 					   "Broadcast filter insert failed rc=%d\n",
4027 					   rc);
4028 				if (rollback) {
4029 					/* Roll back the mc_def filter */
4030 					efx_ef10_filter_remove_unsafe(
4031 							efx, EFX_FILTER_PRI_AUTO,
4032 							table->mcdef_id);
4033 					table->mcdef_id = EFX_EF10_FILTER_ID_INVALID;
4034 					return rc;
4035 				}
4036 			} else {
4037 				table->bcast_id = efx_ef10_filter_get_unsafe_id(efx, rc);
4038 			}
4039 		}
4040 		rc = 0;
4041 	} else {
4042 		table->ucdef_id = rc;
4043 		rc = 0;
4044 	}
4045 	return rc;
4046 }
4047 
4048 /* Remove filters that weren't renewed.  Since nothing else changes the AUTO_OLD
4049  * flag or removes these filters, we don't need to hold the filter_lock while
4050  * scanning for these filters.
4051  */
4052 static void efx_ef10_filter_remove_old(struct efx_nic *efx)
4053 {
4054 	struct efx_ef10_filter_table *table = efx->filter_state;
4055 	bool remove_failed = false;
4056 	int i;
4057 
4058 	for (i = 0; i < HUNT_FILTER_TBL_ROWS; i++) {
4059 		if (ACCESS_ONCE(table->entry[i].spec) &
4060 		    EFX_EF10_FILTER_FLAG_AUTO_OLD) {
4061 			if (efx_ef10_filter_remove_internal(
4062 				    efx, 1U << EFX_FILTER_PRI_AUTO,
4063 				    i, true) < 0)
4064 				remove_failed = true;
4065 		}
4066 	}
4067 	WARN_ON(remove_failed);
4068 }
4069 
4070 static int efx_ef10_vport_set_mac_address(struct efx_nic *efx)
4071 {
4072 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
4073 	u8 mac_old[ETH_ALEN];
4074 	int rc, rc2;
4075 
4076 	/* Only reconfigure a PF-created vport */
4077 	if (is_zero_ether_addr(nic_data->vport_mac))
4078 		return 0;
4079 
4080 	efx_device_detach_sync(efx);
4081 	efx_net_stop(efx->net_dev);
4082 	down_write(&efx->filter_sem);
4083 	efx_ef10_filter_table_remove(efx);
4084 	up_write(&efx->filter_sem);
4085 
4086 	rc = efx_ef10_vadaptor_free(efx, nic_data->vport_id);
4087 	if (rc)
4088 		goto restore_filters;
4089 
4090 	ether_addr_copy(mac_old, nic_data->vport_mac);
4091 	rc = efx_ef10_vport_del_mac(efx, nic_data->vport_id,
4092 				    nic_data->vport_mac);
4093 	if (rc)
4094 		goto restore_vadaptor;
4095 
4096 	rc = efx_ef10_vport_add_mac(efx, nic_data->vport_id,
4097 				    efx->net_dev->dev_addr);
4098 	if (!rc) {
4099 		ether_addr_copy(nic_data->vport_mac, efx->net_dev->dev_addr);
4100 	} else {
4101 		rc2 = efx_ef10_vport_add_mac(efx, nic_data->vport_id, mac_old);
4102 		if (rc2) {
4103 			/* Failed to add original MAC, so clear vport_mac */
4104 			eth_zero_addr(nic_data->vport_mac);
4105 			goto reset_nic;
4106 		}
4107 	}
4108 
4109 restore_vadaptor:
4110 	rc2 = efx_ef10_vadaptor_alloc(efx, nic_data->vport_id);
4111 	if (rc2)
4112 		goto reset_nic;
4113 restore_filters:
4114 	down_write(&efx->filter_sem);
4115 	rc2 = efx_ef10_filter_table_probe(efx);
4116 	up_write(&efx->filter_sem);
4117 	if (rc2)
4118 		goto reset_nic;
4119 
4120 	rc2 = efx_net_open(efx->net_dev);
4121 	if (rc2)
4122 		goto reset_nic;
4123 
4124 	netif_device_attach(efx->net_dev);
4125 
4126 	return rc;
4127 
4128 reset_nic:
4129 	netif_err(efx, drv, efx->net_dev,
4130 		  "Failed to restore when changing MAC address - scheduling reset\n");
4131 	efx_schedule_reset(efx, RESET_TYPE_DATAPATH);
4132 
4133 	return rc ? rc : rc2;
4134 }
4135 
4136 /* Caller must hold efx->filter_sem for read if race against
4137  * efx_ef10_filter_table_remove() is possible
4138  */
4139 static void efx_ef10_filter_sync_rx_mode(struct efx_nic *efx)
4140 {
4141 	struct efx_ef10_filter_table *table = efx->filter_state;
4142 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
4143 	struct net_device *net_dev = efx->net_dev;
4144 	bool uc_promisc = false, mc_promisc = false;
4145 
4146 	if (!efx_dev_registered(efx))
4147 		return;
4148 
4149 	if (!table)
4150 		return;
4151 
4152 	efx_ef10_filter_mark_old(efx);
4153 
4154 	/* Copy/convert the address lists; add the primary station
4155 	 * address and broadcast address
4156 	 */
4157 	netif_addr_lock_bh(net_dev);
4158 	efx_ef10_filter_uc_addr_list(efx, &uc_promisc);
4159 	efx_ef10_filter_mc_addr_list(efx, &mc_promisc);
4160 	netif_addr_unlock_bh(net_dev);
4161 
4162 	/* Insert/renew unicast filters */
4163 	if (uc_promisc) {
4164 		efx_ef10_filter_insert_def(efx, false, false);
4165 		efx_ef10_filter_insert_addr_list(efx, false, false);
4166 	} else {
4167 		/* If any of the filters failed to insert, fall back to
4168 		 * promiscuous mode - add in the uc_def filter.  But keep
4169 		 * our individual unicast filters.
4170 		 */
4171 		if (efx_ef10_filter_insert_addr_list(efx, false, false))
4172 			efx_ef10_filter_insert_def(efx, false, false);
4173 	}
4174 
4175 	/* Insert/renew multicast filters */
4176 	/* If changing promiscuous state with cascaded multicast filters, remove
4177 	 * old filters first, so that packets are dropped rather than duplicated
4178 	 */
4179 	if (nic_data->workaround_26807 && efx->mc_promisc != mc_promisc)
4180 		efx_ef10_filter_remove_old(efx);
4181 	if (mc_promisc) {
4182 		if (nic_data->workaround_26807) {
4183 			/* If we failed to insert promiscuous filters, rollback
4184 			 * and fall back to individual multicast filters
4185 			 */
4186 			if (efx_ef10_filter_insert_def(efx, true, true)) {
4187 				/* Changing promisc state, so remove old filters */
4188 				efx_ef10_filter_remove_old(efx);
4189 				efx_ef10_filter_insert_addr_list(efx, true, false);
4190 			}
4191 		} else {
4192 			/* If we failed to insert promiscuous filters, don't
4193 			 * rollback.  Regardless, also insert the mc_list
4194 			 */
4195 			efx_ef10_filter_insert_def(efx, true, false);
4196 			efx_ef10_filter_insert_addr_list(efx, true, false);
4197 		}
4198 	} else {
4199 		/* If any filters failed to insert, rollback and fall back to
4200 		 * promiscuous mode - mc_def filter and maybe broadcast.  If
4201 		 * that fails, roll back again and insert as many of our
4202 		 * individual multicast filters as we can.
4203 		 */
4204 		if (efx_ef10_filter_insert_addr_list(efx, true, true)) {
4205 			/* Changing promisc state, so remove old filters */
4206 			if (nic_data->workaround_26807)
4207 				efx_ef10_filter_remove_old(efx);
4208 			if (efx_ef10_filter_insert_def(efx, true, true))
4209 				efx_ef10_filter_insert_addr_list(efx, true, false);
4210 		}
4211 	}
4212 
4213 	efx_ef10_filter_remove_old(efx);
4214 	efx->mc_promisc = mc_promisc;
4215 }
4216 
4217 static int efx_ef10_set_mac_address(struct efx_nic *efx)
4218 {
4219 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VADAPTOR_SET_MAC_IN_LEN);
4220 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
4221 	bool was_enabled = efx->port_enabled;
4222 	int rc;
4223 
4224 	efx_device_detach_sync(efx);
4225 	efx_net_stop(efx->net_dev);
4226 	down_write(&efx->filter_sem);
4227 	efx_ef10_filter_table_remove(efx);
4228 
4229 	ether_addr_copy(MCDI_PTR(inbuf, VADAPTOR_SET_MAC_IN_MACADDR),
4230 			efx->net_dev->dev_addr);
4231 	MCDI_SET_DWORD(inbuf, VADAPTOR_SET_MAC_IN_UPSTREAM_PORT_ID,
4232 		       nic_data->vport_id);
4233 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_VADAPTOR_SET_MAC, inbuf,
4234 				sizeof(inbuf), NULL, 0, NULL);
4235 
4236 	efx_ef10_filter_table_probe(efx);
4237 	up_write(&efx->filter_sem);
4238 	if (was_enabled)
4239 		efx_net_open(efx->net_dev);
4240 	netif_device_attach(efx->net_dev);
4241 
4242 #ifdef CONFIG_SFC_SRIOV
4243 	if (efx->pci_dev->is_virtfn && efx->pci_dev->physfn) {
4244 		struct pci_dev *pci_dev_pf = efx->pci_dev->physfn;
4245 
4246 		if (rc == -EPERM) {
4247 			struct efx_nic *efx_pf;
4248 
4249 			/* Switch to PF and change MAC address on vport */
4250 			efx_pf = pci_get_drvdata(pci_dev_pf);
4251 
4252 			rc = efx_ef10_sriov_set_vf_mac(efx_pf,
4253 						       nic_data->vf_index,
4254 						       efx->net_dev->dev_addr);
4255 		} else if (!rc) {
4256 			struct efx_nic *efx_pf = pci_get_drvdata(pci_dev_pf);
4257 			struct efx_ef10_nic_data *nic_data = efx_pf->nic_data;
4258 			unsigned int i;
4259 
4260 			/* MAC address successfully changed by VF (with MAC
4261 			 * spoofing) so update the parent PF if possible.
4262 			 */
4263 			for (i = 0; i < efx_pf->vf_count; ++i) {
4264 				struct ef10_vf *vf = nic_data->vf + i;
4265 
4266 				if (vf->efx == efx) {
4267 					ether_addr_copy(vf->mac,
4268 							efx->net_dev->dev_addr);
4269 					return 0;
4270 				}
4271 			}
4272 		}
4273 	} else
4274 #endif
4275 	if (rc == -EPERM) {
4276 		netif_err(efx, drv, efx->net_dev,
4277 			  "Cannot change MAC address; use sfboot to enable"
4278 			  " mac-spoofing on this interface\n");
4279 	} else if (rc == -ENOSYS && !efx_ef10_is_vf(efx)) {
4280 		/* If the active MCFW does not support MC_CMD_VADAPTOR_SET_MAC
4281 		 * fall-back to the method of changing the MAC address on the
4282 		 * vport.  This only applies to PFs because such versions of
4283 		 * MCFW do not support VFs.
4284 		 */
4285 		rc = efx_ef10_vport_set_mac_address(efx);
4286 	} else {
4287 		efx_mcdi_display_error(efx, MC_CMD_VADAPTOR_SET_MAC,
4288 				       sizeof(inbuf), NULL, 0, rc);
4289 	}
4290 
4291 	return rc;
4292 }
4293 
4294 static int efx_ef10_mac_reconfigure(struct efx_nic *efx)
4295 {
4296 	efx_ef10_filter_sync_rx_mode(efx);
4297 
4298 	return efx_mcdi_set_mac(efx);
4299 }
4300 
4301 static int efx_ef10_mac_reconfigure_vf(struct efx_nic *efx)
4302 {
4303 	efx_ef10_filter_sync_rx_mode(efx);
4304 
4305 	return 0;
4306 }
4307 
4308 static int efx_ef10_start_bist(struct efx_nic *efx, u32 bist_type)
4309 {
4310 	MCDI_DECLARE_BUF(inbuf, MC_CMD_START_BIST_IN_LEN);
4311 
4312 	MCDI_SET_DWORD(inbuf, START_BIST_IN_TYPE, bist_type);
4313 	return efx_mcdi_rpc(efx, MC_CMD_START_BIST, inbuf, sizeof(inbuf),
4314 			    NULL, 0, NULL);
4315 }
4316 
4317 /* MC BISTs follow a different poll mechanism to phy BISTs.
4318  * The BIST is done in the poll handler on the MC, and the MCDI command
4319  * will block until the BIST is done.
4320  */
4321 static int efx_ef10_poll_bist(struct efx_nic *efx)
4322 {
4323 	int rc;
4324 	MCDI_DECLARE_BUF(outbuf, MC_CMD_POLL_BIST_OUT_LEN);
4325 	size_t outlen;
4326 	u32 result;
4327 
4328 	rc = efx_mcdi_rpc(efx, MC_CMD_POLL_BIST, NULL, 0,
4329 			   outbuf, sizeof(outbuf), &outlen);
4330 	if (rc != 0)
4331 		return rc;
4332 
4333 	if (outlen < MC_CMD_POLL_BIST_OUT_LEN)
4334 		return -EIO;
4335 
4336 	result = MCDI_DWORD(outbuf, POLL_BIST_OUT_RESULT);
4337 	switch (result) {
4338 	case MC_CMD_POLL_BIST_PASSED:
4339 		netif_dbg(efx, hw, efx->net_dev, "BIST passed.\n");
4340 		return 0;
4341 	case MC_CMD_POLL_BIST_TIMEOUT:
4342 		netif_err(efx, hw, efx->net_dev, "BIST timed out\n");
4343 		return -EIO;
4344 	case MC_CMD_POLL_BIST_FAILED:
4345 		netif_err(efx, hw, efx->net_dev, "BIST failed.\n");
4346 		return -EIO;
4347 	default:
4348 		netif_err(efx, hw, efx->net_dev,
4349 			  "BIST returned unknown result %u", result);
4350 		return -EIO;
4351 	}
4352 }
4353 
4354 static int efx_ef10_run_bist(struct efx_nic *efx, u32 bist_type)
4355 {
4356 	int rc;
4357 
4358 	netif_dbg(efx, drv, efx->net_dev, "starting BIST type %u\n", bist_type);
4359 
4360 	rc = efx_ef10_start_bist(efx, bist_type);
4361 	if (rc != 0)
4362 		return rc;
4363 
4364 	return efx_ef10_poll_bist(efx);
4365 }
4366 
4367 static int
4368 efx_ef10_test_chip(struct efx_nic *efx, struct efx_self_tests *tests)
4369 {
4370 	int rc, rc2;
4371 
4372 	efx_reset_down(efx, RESET_TYPE_WORLD);
4373 
4374 	rc = efx_mcdi_rpc(efx, MC_CMD_ENABLE_OFFLINE_BIST,
4375 			  NULL, 0, NULL, 0, NULL);
4376 	if (rc != 0)
4377 		goto out;
4378 
4379 	tests->memory = efx_ef10_run_bist(efx, MC_CMD_MC_MEM_BIST) ? -1 : 1;
4380 	tests->registers = efx_ef10_run_bist(efx, MC_CMD_REG_BIST) ? -1 : 1;
4381 
4382 	rc = efx_mcdi_reset(efx, RESET_TYPE_WORLD);
4383 
4384 out:
4385 	if (rc == -EPERM)
4386 		rc = 0;
4387 	rc2 = efx_reset_up(efx, RESET_TYPE_WORLD, rc == 0);
4388 	return rc ? rc : rc2;
4389 }
4390 
4391 #ifdef CONFIG_SFC_MTD
4392 
4393 struct efx_ef10_nvram_type_info {
4394 	u16 type, type_mask;
4395 	u8 port;
4396 	const char *name;
4397 };
4398 
4399 static const struct efx_ef10_nvram_type_info efx_ef10_nvram_types[] = {
4400 	{ NVRAM_PARTITION_TYPE_MC_FIRMWARE,	   0,    0, "sfc_mcfw" },
4401 	{ NVRAM_PARTITION_TYPE_MC_FIRMWARE_BACKUP, 0,    0, "sfc_mcfw_backup" },
4402 	{ NVRAM_PARTITION_TYPE_EXPANSION_ROM,	   0,    0, "sfc_exp_rom" },
4403 	{ NVRAM_PARTITION_TYPE_STATIC_CONFIG,	   0,    0, "sfc_static_cfg" },
4404 	{ NVRAM_PARTITION_TYPE_DYNAMIC_CONFIG,	   0,    0, "sfc_dynamic_cfg" },
4405 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT0, 0,   0, "sfc_exp_rom_cfg" },
4406 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT1, 0,   1, "sfc_exp_rom_cfg" },
4407 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT2, 0,   2, "sfc_exp_rom_cfg" },
4408 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT3, 0,   3, "sfc_exp_rom_cfg" },
4409 	{ NVRAM_PARTITION_TYPE_LICENSE,		   0,    0, "sfc_license" },
4410 	{ NVRAM_PARTITION_TYPE_PHY_MIN,		   0xff, 0, "sfc_phy_fw" },
4411 };
4412 
4413 static int efx_ef10_mtd_probe_partition(struct efx_nic *efx,
4414 					struct efx_mcdi_mtd_partition *part,
4415 					unsigned int type)
4416 {
4417 	MCDI_DECLARE_BUF(inbuf, MC_CMD_NVRAM_METADATA_IN_LEN);
4418 	MCDI_DECLARE_BUF(outbuf, MC_CMD_NVRAM_METADATA_OUT_LENMAX);
4419 	const struct efx_ef10_nvram_type_info *info;
4420 	size_t size, erase_size, outlen;
4421 	bool protected;
4422 	int rc;
4423 
4424 	for (info = efx_ef10_nvram_types; ; info++) {
4425 		if (info ==
4426 		    efx_ef10_nvram_types + ARRAY_SIZE(efx_ef10_nvram_types))
4427 			return -ENODEV;
4428 		if ((type & ~info->type_mask) == info->type)
4429 			break;
4430 	}
4431 	if (info->port != efx_port_num(efx))
4432 		return -ENODEV;
4433 
4434 	rc = efx_mcdi_nvram_info(efx, type, &size, &erase_size, &protected);
4435 	if (rc)
4436 		return rc;
4437 	if (protected)
4438 		return -ENODEV; /* hide it */
4439 
4440 	part->nvram_type = type;
4441 
4442 	MCDI_SET_DWORD(inbuf, NVRAM_METADATA_IN_TYPE, type);
4443 	rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_METADATA, inbuf, sizeof(inbuf),
4444 			  outbuf, sizeof(outbuf), &outlen);
4445 	if (rc)
4446 		return rc;
4447 	if (outlen < MC_CMD_NVRAM_METADATA_OUT_LENMIN)
4448 		return -EIO;
4449 	if (MCDI_DWORD(outbuf, NVRAM_METADATA_OUT_FLAGS) &
4450 	    (1 << MC_CMD_NVRAM_METADATA_OUT_SUBTYPE_VALID_LBN))
4451 		part->fw_subtype = MCDI_DWORD(outbuf,
4452 					      NVRAM_METADATA_OUT_SUBTYPE);
4453 
4454 	part->common.dev_type_name = "EF10 NVRAM manager";
4455 	part->common.type_name = info->name;
4456 
4457 	part->common.mtd.type = MTD_NORFLASH;
4458 	part->common.mtd.flags = MTD_CAP_NORFLASH;
4459 	part->common.mtd.size = size;
4460 	part->common.mtd.erasesize = erase_size;
4461 
4462 	return 0;
4463 }
4464 
4465 static int efx_ef10_mtd_probe(struct efx_nic *efx)
4466 {
4467 	MCDI_DECLARE_BUF(outbuf, MC_CMD_NVRAM_PARTITIONS_OUT_LENMAX);
4468 	struct efx_mcdi_mtd_partition *parts;
4469 	size_t outlen, n_parts_total, i, n_parts;
4470 	unsigned int type;
4471 	int rc;
4472 
4473 	ASSERT_RTNL();
4474 
4475 	BUILD_BUG_ON(MC_CMD_NVRAM_PARTITIONS_IN_LEN != 0);
4476 	rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_PARTITIONS, NULL, 0,
4477 			  outbuf, sizeof(outbuf), &outlen);
4478 	if (rc)
4479 		return rc;
4480 	if (outlen < MC_CMD_NVRAM_PARTITIONS_OUT_LENMIN)
4481 		return -EIO;
4482 
4483 	n_parts_total = MCDI_DWORD(outbuf, NVRAM_PARTITIONS_OUT_NUM_PARTITIONS);
4484 	if (n_parts_total >
4485 	    MCDI_VAR_ARRAY_LEN(outlen, NVRAM_PARTITIONS_OUT_TYPE_ID))
4486 		return -EIO;
4487 
4488 	parts = kcalloc(n_parts_total, sizeof(*parts), GFP_KERNEL);
4489 	if (!parts)
4490 		return -ENOMEM;
4491 
4492 	n_parts = 0;
4493 	for (i = 0; i < n_parts_total; i++) {
4494 		type = MCDI_ARRAY_DWORD(outbuf, NVRAM_PARTITIONS_OUT_TYPE_ID,
4495 					i);
4496 		rc = efx_ef10_mtd_probe_partition(efx, &parts[n_parts], type);
4497 		if (rc == 0)
4498 			n_parts++;
4499 		else if (rc != -ENODEV)
4500 			goto fail;
4501 	}
4502 
4503 	rc = efx_mtd_add(efx, &parts[0].common, n_parts, sizeof(*parts));
4504 fail:
4505 	if (rc)
4506 		kfree(parts);
4507 	return rc;
4508 }
4509 
4510 #endif /* CONFIG_SFC_MTD */
4511 
4512 static void efx_ef10_ptp_write_host_time(struct efx_nic *efx, u32 host_time)
4513 {
4514 	_efx_writed(efx, cpu_to_le32(host_time), ER_DZ_MC_DB_LWRD);
4515 }
4516 
4517 static void efx_ef10_ptp_write_host_time_vf(struct efx_nic *efx,
4518 					    u32 host_time) {}
4519 
4520 static int efx_ef10_rx_enable_timestamping(struct efx_channel *channel,
4521 					   bool temp)
4522 {
4523 	MCDI_DECLARE_BUF(inbuf, MC_CMD_PTP_IN_TIME_EVENT_SUBSCRIBE_LEN);
4524 	int rc;
4525 
4526 	if (channel->sync_events_state == SYNC_EVENTS_REQUESTED ||
4527 	    channel->sync_events_state == SYNC_EVENTS_VALID ||
4528 	    (temp && channel->sync_events_state == SYNC_EVENTS_DISABLED))
4529 		return 0;
4530 	channel->sync_events_state = SYNC_EVENTS_REQUESTED;
4531 
4532 	MCDI_SET_DWORD(inbuf, PTP_IN_OP, MC_CMD_PTP_OP_TIME_EVENT_SUBSCRIBE);
4533 	MCDI_SET_DWORD(inbuf, PTP_IN_PERIPH_ID, 0);
4534 	MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_SUBSCRIBE_QUEUE,
4535 		       channel->channel);
4536 
4537 	rc = efx_mcdi_rpc(channel->efx, MC_CMD_PTP,
4538 			  inbuf, sizeof(inbuf), NULL, 0, NULL);
4539 
4540 	if (rc != 0)
4541 		channel->sync_events_state = temp ? SYNC_EVENTS_QUIESCENT :
4542 						    SYNC_EVENTS_DISABLED;
4543 
4544 	return rc;
4545 }
4546 
4547 static int efx_ef10_rx_disable_timestamping(struct efx_channel *channel,
4548 					    bool temp)
4549 {
4550 	MCDI_DECLARE_BUF(inbuf, MC_CMD_PTP_IN_TIME_EVENT_UNSUBSCRIBE_LEN);
4551 	int rc;
4552 
4553 	if (channel->sync_events_state == SYNC_EVENTS_DISABLED ||
4554 	    (temp && channel->sync_events_state == SYNC_EVENTS_QUIESCENT))
4555 		return 0;
4556 	if (channel->sync_events_state == SYNC_EVENTS_QUIESCENT) {
4557 		channel->sync_events_state = SYNC_EVENTS_DISABLED;
4558 		return 0;
4559 	}
4560 	channel->sync_events_state = temp ? SYNC_EVENTS_QUIESCENT :
4561 					    SYNC_EVENTS_DISABLED;
4562 
4563 	MCDI_SET_DWORD(inbuf, PTP_IN_OP, MC_CMD_PTP_OP_TIME_EVENT_UNSUBSCRIBE);
4564 	MCDI_SET_DWORD(inbuf, PTP_IN_PERIPH_ID, 0);
4565 	MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_UNSUBSCRIBE_CONTROL,
4566 		       MC_CMD_PTP_IN_TIME_EVENT_UNSUBSCRIBE_SINGLE);
4567 	MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_UNSUBSCRIBE_QUEUE,
4568 		       channel->channel);
4569 
4570 	rc = efx_mcdi_rpc(channel->efx, MC_CMD_PTP,
4571 			  inbuf, sizeof(inbuf), NULL, 0, NULL);
4572 
4573 	return rc;
4574 }
4575 
4576 static int efx_ef10_ptp_set_ts_sync_events(struct efx_nic *efx, bool en,
4577 					   bool temp)
4578 {
4579 	int (*set)(struct efx_channel *channel, bool temp);
4580 	struct efx_channel *channel;
4581 
4582 	set = en ?
4583 	      efx_ef10_rx_enable_timestamping :
4584 	      efx_ef10_rx_disable_timestamping;
4585 
4586 	efx_for_each_channel(channel, efx) {
4587 		int rc = set(channel, temp);
4588 		if (en && rc != 0) {
4589 			efx_ef10_ptp_set_ts_sync_events(efx, false, temp);
4590 			return rc;
4591 		}
4592 	}
4593 
4594 	return 0;
4595 }
4596 
4597 static int efx_ef10_ptp_set_ts_config_vf(struct efx_nic *efx,
4598 					 struct hwtstamp_config *init)
4599 {
4600 	return -EOPNOTSUPP;
4601 }
4602 
4603 static int efx_ef10_ptp_set_ts_config(struct efx_nic *efx,
4604 				      struct hwtstamp_config *init)
4605 {
4606 	int rc;
4607 
4608 	switch (init->rx_filter) {
4609 	case HWTSTAMP_FILTER_NONE:
4610 		efx_ef10_ptp_set_ts_sync_events(efx, false, false);
4611 		/* if TX timestamping is still requested then leave PTP on */
4612 		return efx_ptp_change_mode(efx,
4613 					   init->tx_type != HWTSTAMP_TX_OFF, 0);
4614 	case HWTSTAMP_FILTER_ALL:
4615 	case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
4616 	case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
4617 	case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
4618 	case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
4619 	case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
4620 	case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
4621 	case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
4622 	case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
4623 	case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
4624 	case HWTSTAMP_FILTER_PTP_V2_EVENT:
4625 	case HWTSTAMP_FILTER_PTP_V2_SYNC:
4626 	case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
4627 		init->rx_filter = HWTSTAMP_FILTER_ALL;
4628 		rc = efx_ptp_change_mode(efx, true, 0);
4629 		if (!rc)
4630 			rc = efx_ef10_ptp_set_ts_sync_events(efx, true, false);
4631 		if (rc)
4632 			efx_ptp_change_mode(efx, false, 0);
4633 		return rc;
4634 	default:
4635 		return -ERANGE;
4636 	}
4637 }
4638 
4639 const struct efx_nic_type efx_hunt_a0_vf_nic_type = {
4640 	.is_vf = true,
4641 	.mem_bar = EFX_MEM_VF_BAR,
4642 	.mem_map_size = efx_ef10_mem_map_size,
4643 	.probe = efx_ef10_probe_vf,
4644 	.remove = efx_ef10_remove,
4645 	.dimension_resources = efx_ef10_dimension_resources,
4646 	.init = efx_ef10_init_nic,
4647 	.fini = efx_port_dummy_op_void,
4648 	.map_reset_reason = efx_ef10_map_reset_reason,
4649 	.map_reset_flags = efx_ef10_map_reset_flags,
4650 	.reset = efx_ef10_reset,
4651 	.probe_port = efx_mcdi_port_probe,
4652 	.remove_port = efx_mcdi_port_remove,
4653 	.fini_dmaq = efx_ef10_fini_dmaq,
4654 	.prepare_flr = efx_ef10_prepare_flr,
4655 	.finish_flr = efx_port_dummy_op_void,
4656 	.describe_stats = efx_ef10_describe_stats,
4657 	.update_stats = efx_ef10_update_stats_vf,
4658 	.start_stats = efx_port_dummy_op_void,
4659 	.pull_stats = efx_port_dummy_op_void,
4660 	.stop_stats = efx_port_dummy_op_void,
4661 	.set_id_led = efx_mcdi_set_id_led,
4662 	.push_irq_moderation = efx_ef10_push_irq_moderation,
4663 	.reconfigure_mac = efx_ef10_mac_reconfigure_vf,
4664 	.check_mac_fault = efx_mcdi_mac_check_fault,
4665 	.reconfigure_port = efx_mcdi_port_reconfigure,
4666 	.get_wol = efx_ef10_get_wol_vf,
4667 	.set_wol = efx_ef10_set_wol_vf,
4668 	.resume_wol = efx_port_dummy_op_void,
4669 	.mcdi_request = efx_ef10_mcdi_request,
4670 	.mcdi_poll_response = efx_ef10_mcdi_poll_response,
4671 	.mcdi_read_response = efx_ef10_mcdi_read_response,
4672 	.mcdi_poll_reboot = efx_ef10_mcdi_poll_reboot,
4673 	.irq_enable_master = efx_port_dummy_op_void,
4674 	.irq_test_generate = efx_ef10_irq_test_generate,
4675 	.irq_disable_non_ev = efx_port_dummy_op_void,
4676 	.irq_handle_msi = efx_ef10_msi_interrupt,
4677 	.irq_handle_legacy = efx_ef10_legacy_interrupt,
4678 	.tx_probe = efx_ef10_tx_probe,
4679 	.tx_init = efx_ef10_tx_init,
4680 	.tx_remove = efx_ef10_tx_remove,
4681 	.tx_write = efx_ef10_tx_write,
4682 	.rx_push_rss_config = efx_ef10_vf_rx_push_rss_config,
4683 	.rx_probe = efx_ef10_rx_probe,
4684 	.rx_init = efx_ef10_rx_init,
4685 	.rx_remove = efx_ef10_rx_remove,
4686 	.rx_write = efx_ef10_rx_write,
4687 	.rx_defer_refill = efx_ef10_rx_defer_refill,
4688 	.ev_probe = efx_ef10_ev_probe,
4689 	.ev_init = efx_ef10_ev_init,
4690 	.ev_fini = efx_ef10_ev_fini,
4691 	.ev_remove = efx_ef10_ev_remove,
4692 	.ev_process = efx_ef10_ev_process,
4693 	.ev_read_ack = efx_ef10_ev_read_ack,
4694 	.ev_test_generate = efx_ef10_ev_test_generate,
4695 	.filter_table_probe = efx_ef10_filter_table_probe,
4696 	.filter_table_restore = efx_ef10_filter_table_restore,
4697 	.filter_table_remove = efx_ef10_filter_table_remove,
4698 	.filter_update_rx_scatter = efx_ef10_filter_update_rx_scatter,
4699 	.filter_insert = efx_ef10_filter_insert,
4700 	.filter_remove_safe = efx_ef10_filter_remove_safe,
4701 	.filter_get_safe = efx_ef10_filter_get_safe,
4702 	.filter_clear_rx = efx_ef10_filter_clear_rx,
4703 	.filter_count_rx_used = efx_ef10_filter_count_rx_used,
4704 	.filter_get_rx_id_limit = efx_ef10_filter_get_rx_id_limit,
4705 	.filter_get_rx_ids = efx_ef10_filter_get_rx_ids,
4706 #ifdef CONFIG_RFS_ACCEL
4707 	.filter_rfs_insert = efx_ef10_filter_rfs_insert,
4708 	.filter_rfs_expire_one = efx_ef10_filter_rfs_expire_one,
4709 #endif
4710 #ifdef CONFIG_SFC_MTD
4711 	.mtd_probe = efx_port_dummy_op_int,
4712 #endif
4713 	.ptp_write_host_time = efx_ef10_ptp_write_host_time_vf,
4714 	.ptp_set_ts_config = efx_ef10_ptp_set_ts_config_vf,
4715 #ifdef CONFIG_SFC_SRIOV
4716 	.vswitching_probe = efx_ef10_vswitching_probe_vf,
4717 	.vswitching_restore = efx_ef10_vswitching_restore_vf,
4718 	.vswitching_remove = efx_ef10_vswitching_remove_vf,
4719 	.sriov_get_phys_port_id = efx_ef10_sriov_get_phys_port_id,
4720 #endif
4721 	.get_mac_address = efx_ef10_get_mac_address_vf,
4722 	.set_mac_address = efx_ef10_set_mac_address,
4723 
4724 	.revision = EFX_REV_HUNT_A0,
4725 	.max_dma_mask = DMA_BIT_MASK(ESF_DZ_TX_KER_BUF_ADDR_WIDTH),
4726 	.rx_prefix_size = ES_DZ_RX_PREFIX_SIZE,
4727 	.rx_hash_offset = ES_DZ_RX_PREFIX_HASH_OFST,
4728 	.rx_ts_offset = ES_DZ_RX_PREFIX_TSTAMP_OFST,
4729 	.can_rx_scatter = true,
4730 	.always_rx_scatter = true,
4731 	.max_interrupt_mode = EFX_INT_MODE_MSIX,
4732 	.timer_period_max = 1 << ERF_DD_EVQ_IND_TIMER_VAL_WIDTH,
4733 	.offload_features = (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
4734 			     NETIF_F_RXHASH | NETIF_F_NTUPLE),
4735 	.mcdi_max_ver = 2,
4736 	.max_rx_ip_filters = HUNT_FILTER_TBL_ROWS,
4737 	.hwtstamp_filters = 1 << HWTSTAMP_FILTER_NONE |
4738 			    1 << HWTSTAMP_FILTER_ALL,
4739 };
4740 
4741 const struct efx_nic_type efx_hunt_a0_nic_type = {
4742 	.is_vf = false,
4743 	.mem_bar = EFX_MEM_BAR,
4744 	.mem_map_size = efx_ef10_mem_map_size,
4745 	.probe = efx_ef10_probe_pf,
4746 	.remove = efx_ef10_remove,
4747 	.dimension_resources = efx_ef10_dimension_resources,
4748 	.init = efx_ef10_init_nic,
4749 	.fini = efx_port_dummy_op_void,
4750 	.map_reset_reason = efx_ef10_map_reset_reason,
4751 	.map_reset_flags = efx_ef10_map_reset_flags,
4752 	.reset = efx_ef10_reset,
4753 	.probe_port = efx_mcdi_port_probe,
4754 	.remove_port = efx_mcdi_port_remove,
4755 	.fini_dmaq = efx_ef10_fini_dmaq,
4756 	.prepare_flr = efx_ef10_prepare_flr,
4757 	.finish_flr = efx_port_dummy_op_void,
4758 	.describe_stats = efx_ef10_describe_stats,
4759 	.update_stats = efx_ef10_update_stats_pf,
4760 	.start_stats = efx_mcdi_mac_start_stats,
4761 	.pull_stats = efx_mcdi_mac_pull_stats,
4762 	.stop_stats = efx_mcdi_mac_stop_stats,
4763 	.set_id_led = efx_mcdi_set_id_led,
4764 	.push_irq_moderation = efx_ef10_push_irq_moderation,
4765 	.reconfigure_mac = efx_ef10_mac_reconfigure,
4766 	.check_mac_fault = efx_mcdi_mac_check_fault,
4767 	.reconfigure_port = efx_mcdi_port_reconfigure,
4768 	.get_wol = efx_ef10_get_wol,
4769 	.set_wol = efx_ef10_set_wol,
4770 	.resume_wol = efx_port_dummy_op_void,
4771 	.test_chip = efx_ef10_test_chip,
4772 	.test_nvram = efx_mcdi_nvram_test_all,
4773 	.mcdi_request = efx_ef10_mcdi_request,
4774 	.mcdi_poll_response = efx_ef10_mcdi_poll_response,
4775 	.mcdi_read_response = efx_ef10_mcdi_read_response,
4776 	.mcdi_poll_reboot = efx_ef10_mcdi_poll_reboot,
4777 	.irq_enable_master = efx_port_dummy_op_void,
4778 	.irq_test_generate = efx_ef10_irq_test_generate,
4779 	.irq_disable_non_ev = efx_port_dummy_op_void,
4780 	.irq_handle_msi = efx_ef10_msi_interrupt,
4781 	.irq_handle_legacy = efx_ef10_legacy_interrupt,
4782 	.tx_probe = efx_ef10_tx_probe,
4783 	.tx_init = efx_ef10_tx_init,
4784 	.tx_remove = efx_ef10_tx_remove,
4785 	.tx_write = efx_ef10_tx_write,
4786 	.rx_push_rss_config = efx_ef10_pf_rx_push_rss_config,
4787 	.rx_probe = efx_ef10_rx_probe,
4788 	.rx_init = efx_ef10_rx_init,
4789 	.rx_remove = efx_ef10_rx_remove,
4790 	.rx_write = efx_ef10_rx_write,
4791 	.rx_defer_refill = efx_ef10_rx_defer_refill,
4792 	.ev_probe = efx_ef10_ev_probe,
4793 	.ev_init = efx_ef10_ev_init,
4794 	.ev_fini = efx_ef10_ev_fini,
4795 	.ev_remove = efx_ef10_ev_remove,
4796 	.ev_process = efx_ef10_ev_process,
4797 	.ev_read_ack = efx_ef10_ev_read_ack,
4798 	.ev_test_generate = efx_ef10_ev_test_generate,
4799 	.filter_table_probe = efx_ef10_filter_table_probe,
4800 	.filter_table_restore = efx_ef10_filter_table_restore,
4801 	.filter_table_remove = efx_ef10_filter_table_remove,
4802 	.filter_update_rx_scatter = efx_ef10_filter_update_rx_scatter,
4803 	.filter_insert = efx_ef10_filter_insert,
4804 	.filter_remove_safe = efx_ef10_filter_remove_safe,
4805 	.filter_get_safe = efx_ef10_filter_get_safe,
4806 	.filter_clear_rx = efx_ef10_filter_clear_rx,
4807 	.filter_count_rx_used = efx_ef10_filter_count_rx_used,
4808 	.filter_get_rx_id_limit = efx_ef10_filter_get_rx_id_limit,
4809 	.filter_get_rx_ids = efx_ef10_filter_get_rx_ids,
4810 #ifdef CONFIG_RFS_ACCEL
4811 	.filter_rfs_insert = efx_ef10_filter_rfs_insert,
4812 	.filter_rfs_expire_one = efx_ef10_filter_rfs_expire_one,
4813 #endif
4814 #ifdef CONFIG_SFC_MTD
4815 	.mtd_probe = efx_ef10_mtd_probe,
4816 	.mtd_rename = efx_mcdi_mtd_rename,
4817 	.mtd_read = efx_mcdi_mtd_read,
4818 	.mtd_erase = efx_mcdi_mtd_erase,
4819 	.mtd_write = efx_mcdi_mtd_write,
4820 	.mtd_sync = efx_mcdi_mtd_sync,
4821 #endif
4822 	.ptp_write_host_time = efx_ef10_ptp_write_host_time,
4823 	.ptp_set_ts_sync_events = efx_ef10_ptp_set_ts_sync_events,
4824 	.ptp_set_ts_config = efx_ef10_ptp_set_ts_config,
4825 #ifdef CONFIG_SFC_SRIOV
4826 	.sriov_configure = efx_ef10_sriov_configure,
4827 	.sriov_init = efx_ef10_sriov_init,
4828 	.sriov_fini = efx_ef10_sriov_fini,
4829 	.sriov_wanted = efx_ef10_sriov_wanted,
4830 	.sriov_reset = efx_ef10_sriov_reset,
4831 	.sriov_flr = efx_ef10_sriov_flr,
4832 	.sriov_set_vf_mac = efx_ef10_sriov_set_vf_mac,
4833 	.sriov_set_vf_vlan = efx_ef10_sriov_set_vf_vlan,
4834 	.sriov_set_vf_spoofchk = efx_ef10_sriov_set_vf_spoofchk,
4835 	.sriov_get_vf_config = efx_ef10_sriov_get_vf_config,
4836 	.sriov_set_vf_link_state = efx_ef10_sriov_set_vf_link_state,
4837 	.vswitching_probe = efx_ef10_vswitching_probe_pf,
4838 	.vswitching_restore = efx_ef10_vswitching_restore_pf,
4839 	.vswitching_remove = efx_ef10_vswitching_remove_pf,
4840 #endif
4841 	.get_mac_address = efx_ef10_get_mac_address_pf,
4842 	.set_mac_address = efx_ef10_set_mac_address,
4843 
4844 	.revision = EFX_REV_HUNT_A0,
4845 	.max_dma_mask = DMA_BIT_MASK(ESF_DZ_TX_KER_BUF_ADDR_WIDTH),
4846 	.rx_prefix_size = ES_DZ_RX_PREFIX_SIZE,
4847 	.rx_hash_offset = ES_DZ_RX_PREFIX_HASH_OFST,
4848 	.rx_ts_offset = ES_DZ_RX_PREFIX_TSTAMP_OFST,
4849 	.can_rx_scatter = true,
4850 	.always_rx_scatter = true,
4851 	.max_interrupt_mode = EFX_INT_MODE_MSIX,
4852 	.timer_period_max = 1 << ERF_DD_EVQ_IND_TIMER_VAL_WIDTH,
4853 	.offload_features = (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
4854 			     NETIF_F_RXHASH | NETIF_F_NTUPLE),
4855 	.mcdi_max_ver = 2,
4856 	.max_rx_ip_filters = HUNT_FILTER_TBL_ROWS,
4857 	.hwtstamp_filters = 1 << HWTSTAMP_FILTER_NONE |
4858 			    1 << HWTSTAMP_FILTER_ALL,
4859 };
4860