xref: /openbmc/linux/drivers/net/ethernet/sfc/ef10.c (revision a06c488d)
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_RX_PREFIX_LEN_14_LBN))) {
185 		netif_err(efx, probe, efx->net_dev,
186 			  "current firmware does not support an RX prefix\n");
187 		return -ENODEV;
188 	}
189 
190 	return 0;
191 }
192 
193 static int efx_ef10_get_sysclk_freq(struct efx_nic *efx)
194 {
195 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_CLOCK_OUT_LEN);
196 	int rc;
197 
198 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_CLOCK, NULL, 0,
199 			  outbuf, sizeof(outbuf), NULL);
200 	if (rc)
201 		return rc;
202 	rc = MCDI_DWORD(outbuf, GET_CLOCK_OUT_SYS_FREQ);
203 	return rc > 0 ? rc : -ERANGE;
204 }
205 
206 static int efx_ef10_get_mac_address_pf(struct efx_nic *efx, u8 *mac_address)
207 {
208 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_MAC_ADDRESSES_OUT_LEN);
209 	size_t outlen;
210 	int rc;
211 
212 	BUILD_BUG_ON(MC_CMD_GET_MAC_ADDRESSES_IN_LEN != 0);
213 
214 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_MAC_ADDRESSES, NULL, 0,
215 			  outbuf, sizeof(outbuf), &outlen);
216 	if (rc)
217 		return rc;
218 	if (outlen < MC_CMD_GET_MAC_ADDRESSES_OUT_LEN)
219 		return -EIO;
220 
221 	ether_addr_copy(mac_address,
222 			MCDI_PTR(outbuf, GET_MAC_ADDRESSES_OUT_MAC_ADDR_BASE));
223 	return 0;
224 }
225 
226 static int efx_ef10_get_mac_address_vf(struct efx_nic *efx, u8 *mac_address)
227 {
228 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VPORT_GET_MAC_ADDRESSES_IN_LEN);
229 	MCDI_DECLARE_BUF(outbuf, MC_CMD_VPORT_GET_MAC_ADDRESSES_OUT_LENMAX);
230 	size_t outlen;
231 	int num_addrs, rc;
232 
233 	MCDI_SET_DWORD(inbuf, VPORT_GET_MAC_ADDRESSES_IN_VPORT_ID,
234 		       EVB_PORT_ID_ASSIGNED);
235 	rc = efx_mcdi_rpc(efx, MC_CMD_VPORT_GET_MAC_ADDRESSES, inbuf,
236 			  sizeof(inbuf), outbuf, sizeof(outbuf), &outlen);
237 
238 	if (rc)
239 		return rc;
240 	if (outlen < MC_CMD_VPORT_GET_MAC_ADDRESSES_OUT_LENMIN)
241 		return -EIO;
242 
243 	num_addrs = MCDI_DWORD(outbuf,
244 			       VPORT_GET_MAC_ADDRESSES_OUT_MACADDR_COUNT);
245 
246 	WARN_ON(num_addrs != 1);
247 
248 	ether_addr_copy(mac_address,
249 			MCDI_PTR(outbuf, VPORT_GET_MAC_ADDRESSES_OUT_MACADDR));
250 
251 	return 0;
252 }
253 
254 static ssize_t efx_ef10_show_link_control_flag(struct device *dev,
255 					       struct device_attribute *attr,
256 					       char *buf)
257 {
258 	struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));
259 
260 	return sprintf(buf, "%d\n",
261 		       ((efx->mcdi->fn_flags) &
262 			(1 << MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_LINKCTRL))
263 		       ? 1 : 0);
264 }
265 
266 static ssize_t efx_ef10_show_primary_flag(struct device *dev,
267 					  struct device_attribute *attr,
268 					  char *buf)
269 {
270 	struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));
271 
272 	return sprintf(buf, "%d\n",
273 		       ((efx->mcdi->fn_flags) &
274 			(1 << MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_PRIMARY))
275 		       ? 1 : 0);
276 }
277 
278 static DEVICE_ATTR(link_control_flag, 0444, efx_ef10_show_link_control_flag,
279 		   NULL);
280 static DEVICE_ATTR(primary_flag, 0444, efx_ef10_show_primary_flag, NULL);
281 
282 static int efx_ef10_probe(struct efx_nic *efx)
283 {
284 	struct efx_ef10_nic_data *nic_data;
285 	struct net_device *net_dev = efx->net_dev;
286 	int i, rc;
287 
288 	/* We can have one VI for each 8K region.  However, until we
289 	 * use TX option descriptors we need two TX queues per channel.
290 	 */
291 	efx->max_channels = min_t(unsigned int,
292 				  EFX_MAX_CHANNELS,
293 				  efx_ef10_mem_map_size(efx) /
294 				  (EFX_VI_PAGE_SIZE * EFX_TXQ_TYPES));
295 	efx->max_tx_channels = efx->max_channels;
296 	if (WARN_ON(efx->max_channels == 0))
297 		return -EIO;
298 
299 	nic_data = kzalloc(sizeof(*nic_data), GFP_KERNEL);
300 	if (!nic_data)
301 		return -ENOMEM;
302 	efx->nic_data = nic_data;
303 
304 	/* we assume later that we can copy from this buffer in dwords */
305 	BUILD_BUG_ON(MCDI_CTL_SDU_LEN_MAX_V2 % 4);
306 
307 	rc = efx_nic_alloc_buffer(efx, &nic_data->mcdi_buf,
308 				  8 + MCDI_CTL_SDU_LEN_MAX_V2, GFP_KERNEL);
309 	if (rc)
310 		goto fail1;
311 
312 	/* Get the MC's warm boot count.  In case it's rebooting right
313 	 * now, be prepared to retry.
314 	 */
315 	i = 0;
316 	for (;;) {
317 		rc = efx_ef10_get_warm_boot_count(efx);
318 		if (rc >= 0)
319 			break;
320 		if (++i == 5)
321 			goto fail2;
322 		ssleep(1);
323 	}
324 	nic_data->warm_boot_count = rc;
325 
326 	nic_data->rx_rss_context = EFX_EF10_RSS_CONTEXT_INVALID;
327 
328 	nic_data->vport_id = EVB_PORT_ID_ASSIGNED;
329 
330 	/* In case we're recovering from a crash (kexec), we want to
331 	 * cancel any outstanding request by the previous user of this
332 	 * function.  We send a special message using the least
333 	 * significant bits of the 'high' (doorbell) register.
334 	 */
335 	_efx_writed(efx, cpu_to_le32(1), ER_DZ_MC_DB_HWRD);
336 
337 	rc = efx_mcdi_init(efx);
338 	if (rc)
339 		goto fail2;
340 
341 	/* Reset (most) configuration for this function */
342 	rc = efx_mcdi_reset(efx, RESET_TYPE_ALL);
343 	if (rc)
344 		goto fail3;
345 
346 	/* Enable event logging */
347 	rc = efx_mcdi_log_ctrl(efx, true, false, 0);
348 	if (rc)
349 		goto fail3;
350 
351 	rc = device_create_file(&efx->pci_dev->dev,
352 				&dev_attr_link_control_flag);
353 	if (rc)
354 		goto fail3;
355 
356 	rc = device_create_file(&efx->pci_dev->dev, &dev_attr_primary_flag);
357 	if (rc)
358 		goto fail4;
359 
360 	rc = efx_ef10_get_pf_index(efx);
361 	if (rc)
362 		goto fail5;
363 
364 	rc = efx_ef10_init_datapath_caps(efx);
365 	if (rc < 0)
366 		goto fail5;
367 
368 	efx->rx_packet_len_offset =
369 		ES_DZ_RX_PREFIX_PKTLEN_OFST - ES_DZ_RX_PREFIX_SIZE;
370 
371 	rc = efx_mcdi_port_get_number(efx);
372 	if (rc < 0)
373 		goto fail5;
374 	efx->port_num = rc;
375 	net_dev->dev_port = rc;
376 
377 	rc = efx->type->get_mac_address(efx, efx->net_dev->perm_addr);
378 	if (rc)
379 		goto fail5;
380 
381 	rc = efx_ef10_get_sysclk_freq(efx);
382 	if (rc < 0)
383 		goto fail5;
384 	efx->timer_quantum_ns = 1536000 / rc; /* 1536 cycles */
385 
386 	/* Check whether firmware supports bug 35388 workaround.
387 	 * First try to enable it, then if we get EPERM, just
388 	 * ask if it's already enabled
389 	 */
390 	rc = efx_mcdi_set_workaround(efx, MC_CMD_WORKAROUND_BUG35388, true, NULL);
391 	if (rc == 0) {
392 		nic_data->workaround_35388 = true;
393 	} else if (rc == -EPERM) {
394 		unsigned int enabled;
395 
396 		rc = efx_mcdi_get_workarounds(efx, NULL, &enabled);
397 		if (rc)
398 			goto fail3;
399 		nic_data->workaround_35388 = enabled &
400 			MC_CMD_GET_WORKAROUNDS_OUT_BUG35388;
401 	} else if (rc != -ENOSYS && rc != -ENOENT) {
402 		goto fail5;
403 	}
404 	netif_dbg(efx, probe, efx->net_dev,
405 		  "workaround for bug 35388 is %sabled\n",
406 		  nic_data->workaround_35388 ? "en" : "dis");
407 
408 	rc = efx_mcdi_mon_probe(efx);
409 	if (rc && rc != -EPERM)
410 		goto fail5;
411 
412 	efx_ptp_probe(efx, NULL);
413 
414 #ifdef CONFIG_SFC_SRIOV
415 	if ((efx->pci_dev->physfn) && (!efx->pci_dev->is_physfn)) {
416 		struct pci_dev *pci_dev_pf = efx->pci_dev->physfn;
417 		struct efx_nic *efx_pf = pci_get_drvdata(pci_dev_pf);
418 
419 		efx_pf->type->get_mac_address(efx_pf, nic_data->port_id);
420 	} else
421 #endif
422 		ether_addr_copy(nic_data->port_id, efx->net_dev->perm_addr);
423 
424 	return 0;
425 
426 fail5:
427 	device_remove_file(&efx->pci_dev->dev, &dev_attr_primary_flag);
428 fail4:
429 	device_remove_file(&efx->pci_dev->dev, &dev_attr_link_control_flag);
430 fail3:
431 	efx_mcdi_fini(efx);
432 fail2:
433 	efx_nic_free_buffer(efx, &nic_data->mcdi_buf);
434 fail1:
435 	kfree(nic_data);
436 	efx->nic_data = NULL;
437 	return rc;
438 }
439 
440 static int efx_ef10_free_vis(struct efx_nic *efx)
441 {
442 	MCDI_DECLARE_BUF_ERR(outbuf);
443 	size_t outlen;
444 	int rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FREE_VIS, NULL, 0,
445 				    outbuf, sizeof(outbuf), &outlen);
446 
447 	/* -EALREADY means nothing to free, so ignore */
448 	if (rc == -EALREADY)
449 		rc = 0;
450 	if (rc)
451 		efx_mcdi_display_error(efx, MC_CMD_FREE_VIS, 0, outbuf, outlen,
452 				       rc);
453 	return rc;
454 }
455 
456 #ifdef EFX_USE_PIO
457 
458 static void efx_ef10_free_piobufs(struct efx_nic *efx)
459 {
460 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
461 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FREE_PIOBUF_IN_LEN);
462 	unsigned int i;
463 	int rc;
464 
465 	BUILD_BUG_ON(MC_CMD_FREE_PIOBUF_OUT_LEN != 0);
466 
467 	for (i = 0; i < nic_data->n_piobufs; i++) {
468 		MCDI_SET_DWORD(inbuf, FREE_PIOBUF_IN_PIOBUF_HANDLE,
469 			       nic_data->piobuf_handle[i]);
470 		rc = efx_mcdi_rpc(efx, MC_CMD_FREE_PIOBUF, inbuf, sizeof(inbuf),
471 				  NULL, 0, NULL);
472 		WARN_ON(rc);
473 	}
474 
475 	nic_data->n_piobufs = 0;
476 }
477 
478 static int efx_ef10_alloc_piobufs(struct efx_nic *efx, unsigned int n)
479 {
480 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
481 	MCDI_DECLARE_BUF(outbuf, MC_CMD_ALLOC_PIOBUF_OUT_LEN);
482 	unsigned int i;
483 	size_t outlen;
484 	int rc = 0;
485 
486 	BUILD_BUG_ON(MC_CMD_ALLOC_PIOBUF_IN_LEN != 0);
487 
488 	for (i = 0; i < n; i++) {
489 		rc = efx_mcdi_rpc_quiet(efx, MC_CMD_ALLOC_PIOBUF, NULL, 0,
490 					outbuf, sizeof(outbuf), &outlen);
491 		if (rc) {
492 			/* Don't display the MC error if we didn't have space
493 			 * for a VF.
494 			 */
495 			if (!(efx_ef10_is_vf(efx) && rc == -ENOSPC))
496 				efx_mcdi_display_error(efx, MC_CMD_ALLOC_PIOBUF,
497 						       0, outbuf, outlen, rc);
498 			break;
499 		}
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 void efx_ef10_mcdi_reboot_detected(struct efx_nic *efx)
1608 {
1609 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1610 
1611 	/* All our allocations have been reset */
1612 	efx_ef10_reset_mc_allocations(efx);
1613 
1614 	/* The datapath firmware might have been changed */
1615 	nic_data->must_check_datapath_caps = true;
1616 
1617 	/* MAC statistics have been cleared on the NIC; clear the local
1618 	 * statistic that we update with efx_update_diff_stat().
1619 	 */
1620 	nic_data->stats[EF10_STAT_port_rx_bad_bytes] = 0;
1621 }
1622 
1623 static int efx_ef10_mcdi_poll_reboot(struct efx_nic *efx)
1624 {
1625 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1626 	int rc;
1627 
1628 	rc = efx_ef10_get_warm_boot_count(efx);
1629 	if (rc < 0) {
1630 		/* The firmware is presumably in the process of
1631 		 * rebooting.  However, we are supposed to report each
1632 		 * reboot just once, so we must only do that once we
1633 		 * can read and store the updated warm boot count.
1634 		 */
1635 		return 0;
1636 	}
1637 
1638 	if (rc == nic_data->warm_boot_count)
1639 		return 0;
1640 
1641 	nic_data->warm_boot_count = rc;
1642 	efx_ef10_mcdi_reboot_detected(efx);
1643 
1644 	return -EIO;
1645 }
1646 
1647 /* Handle an MSI interrupt
1648  *
1649  * Handle an MSI hardware interrupt.  This routine schedules event
1650  * queue processing.  No interrupt acknowledgement cycle is necessary.
1651  * Also, we never need to check that the interrupt is for us, since
1652  * MSI interrupts cannot be shared.
1653  */
1654 static irqreturn_t efx_ef10_msi_interrupt(int irq, void *dev_id)
1655 {
1656 	struct efx_msi_context *context = dev_id;
1657 	struct efx_nic *efx = context->efx;
1658 
1659 	netif_vdbg(efx, intr, efx->net_dev,
1660 		   "IRQ %d on CPU %d\n", irq, raw_smp_processor_id());
1661 
1662 	if (likely(ACCESS_ONCE(efx->irq_soft_enabled))) {
1663 		/* Note test interrupts */
1664 		if (context->index == efx->irq_level)
1665 			efx->last_irq_cpu = raw_smp_processor_id();
1666 
1667 		/* Schedule processing of the channel */
1668 		efx_schedule_channel_irq(efx->channel[context->index]);
1669 	}
1670 
1671 	return IRQ_HANDLED;
1672 }
1673 
1674 static irqreturn_t efx_ef10_legacy_interrupt(int irq, void *dev_id)
1675 {
1676 	struct efx_nic *efx = dev_id;
1677 	bool soft_enabled = ACCESS_ONCE(efx->irq_soft_enabled);
1678 	struct efx_channel *channel;
1679 	efx_dword_t reg;
1680 	u32 queues;
1681 
1682 	/* Read the ISR which also ACKs the interrupts */
1683 	efx_readd(efx, &reg, ER_DZ_BIU_INT_ISR);
1684 	queues = EFX_DWORD_FIELD(reg, ERF_DZ_ISR_REG);
1685 
1686 	if (queues == 0)
1687 		return IRQ_NONE;
1688 
1689 	if (likely(soft_enabled)) {
1690 		/* Note test interrupts */
1691 		if (queues & (1U << efx->irq_level))
1692 			efx->last_irq_cpu = raw_smp_processor_id();
1693 
1694 		efx_for_each_channel(channel, efx) {
1695 			if (queues & 1)
1696 				efx_schedule_channel_irq(channel);
1697 			queues >>= 1;
1698 		}
1699 	}
1700 
1701 	netif_vdbg(efx, intr, efx->net_dev,
1702 		   "IRQ %d on CPU %d status " EFX_DWORD_FMT "\n",
1703 		   irq, raw_smp_processor_id(), EFX_DWORD_VAL(reg));
1704 
1705 	return IRQ_HANDLED;
1706 }
1707 
1708 static void efx_ef10_irq_test_generate(struct efx_nic *efx)
1709 {
1710 	MCDI_DECLARE_BUF(inbuf, MC_CMD_TRIGGER_INTERRUPT_IN_LEN);
1711 
1712 	BUILD_BUG_ON(MC_CMD_TRIGGER_INTERRUPT_OUT_LEN != 0);
1713 
1714 	MCDI_SET_DWORD(inbuf, TRIGGER_INTERRUPT_IN_INTR_LEVEL, efx->irq_level);
1715 	(void) efx_mcdi_rpc(efx, MC_CMD_TRIGGER_INTERRUPT,
1716 			    inbuf, sizeof(inbuf), NULL, 0, NULL);
1717 }
1718 
1719 static int efx_ef10_tx_probe(struct efx_tx_queue *tx_queue)
1720 {
1721 	return efx_nic_alloc_buffer(tx_queue->efx, &tx_queue->txd.buf,
1722 				    (tx_queue->ptr_mask + 1) *
1723 				    sizeof(efx_qword_t),
1724 				    GFP_KERNEL);
1725 }
1726 
1727 /* This writes to the TX_DESC_WPTR and also pushes data */
1728 static inline void efx_ef10_push_tx_desc(struct efx_tx_queue *tx_queue,
1729 					 const efx_qword_t *txd)
1730 {
1731 	unsigned int write_ptr;
1732 	efx_oword_t reg;
1733 
1734 	write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
1735 	EFX_POPULATE_OWORD_1(reg, ERF_DZ_TX_DESC_WPTR, write_ptr);
1736 	reg.qword[0] = *txd;
1737 	efx_writeo_page(tx_queue->efx, &reg,
1738 			ER_DZ_TX_DESC_UPD, tx_queue->queue);
1739 }
1740 
1741 static void efx_ef10_tx_init(struct efx_tx_queue *tx_queue)
1742 {
1743 	MCDI_DECLARE_BUF(inbuf, MC_CMD_INIT_TXQ_IN_LEN(EFX_MAX_DMAQ_SIZE * 8 /
1744 						       EFX_BUF_SIZE));
1745 	bool csum_offload = tx_queue->queue & EFX_TXQ_TYPE_OFFLOAD;
1746 	size_t entries = tx_queue->txd.buf.len / EFX_BUF_SIZE;
1747 	struct efx_channel *channel = tx_queue->channel;
1748 	struct efx_nic *efx = tx_queue->efx;
1749 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1750 	size_t inlen;
1751 	dma_addr_t dma_addr;
1752 	efx_qword_t *txd;
1753 	int rc;
1754 	int i;
1755 	BUILD_BUG_ON(MC_CMD_INIT_TXQ_OUT_LEN != 0);
1756 
1757 	MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_SIZE, tx_queue->ptr_mask + 1);
1758 	MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_TARGET_EVQ, channel->channel);
1759 	MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_LABEL, tx_queue->queue);
1760 	MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_INSTANCE, tx_queue->queue);
1761 	MCDI_POPULATE_DWORD_2(inbuf, INIT_TXQ_IN_FLAGS,
1762 			      INIT_TXQ_IN_FLAG_IP_CSUM_DIS, !csum_offload,
1763 			      INIT_TXQ_IN_FLAG_TCP_CSUM_DIS, !csum_offload);
1764 	MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_OWNER_ID, 0);
1765 	MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_PORT_ID, nic_data->vport_id);
1766 
1767 	dma_addr = tx_queue->txd.buf.dma_addr;
1768 
1769 	netif_dbg(efx, hw, efx->net_dev, "pushing TXQ %d. %zu entries (%llx)\n",
1770 		  tx_queue->queue, entries, (u64)dma_addr);
1771 
1772 	for (i = 0; i < entries; ++i) {
1773 		MCDI_SET_ARRAY_QWORD(inbuf, INIT_TXQ_IN_DMA_ADDR, i, dma_addr);
1774 		dma_addr += EFX_BUF_SIZE;
1775 	}
1776 
1777 	inlen = MC_CMD_INIT_TXQ_IN_LEN(entries);
1778 
1779 	rc = efx_mcdi_rpc(efx, MC_CMD_INIT_TXQ, inbuf, inlen,
1780 			  NULL, 0, NULL);
1781 	if (rc)
1782 		goto fail;
1783 
1784 	/* A previous user of this TX queue might have set us up the
1785 	 * bomb by writing a descriptor to the TX push collector but
1786 	 * not the doorbell.  (Each collector belongs to a port, not a
1787 	 * queue or function, so cannot easily be reset.)  We must
1788 	 * attempt to push a no-op descriptor in its place.
1789 	 */
1790 	tx_queue->buffer[0].flags = EFX_TX_BUF_OPTION;
1791 	tx_queue->insert_count = 1;
1792 	txd = efx_tx_desc(tx_queue, 0);
1793 	EFX_POPULATE_QWORD_4(*txd,
1794 			     ESF_DZ_TX_DESC_IS_OPT, true,
1795 			     ESF_DZ_TX_OPTION_TYPE,
1796 			     ESE_DZ_TX_OPTION_DESC_CRC_CSUM,
1797 			     ESF_DZ_TX_OPTION_UDP_TCP_CSUM, csum_offload,
1798 			     ESF_DZ_TX_OPTION_IP_CSUM, csum_offload);
1799 	tx_queue->write_count = 1;
1800 
1801 	if (nic_data->datapath_caps &
1802 	    (1 << MC_CMD_GET_CAPABILITIES_OUT_TX_TSO_LBN)) {
1803 		tx_queue->tso_version = 1;
1804 	}
1805 
1806 	wmb();
1807 	efx_ef10_push_tx_desc(tx_queue, txd);
1808 
1809 	return;
1810 
1811 fail:
1812 	netdev_WARN(efx->net_dev, "failed to initialise TXQ %d\n",
1813 		    tx_queue->queue);
1814 }
1815 
1816 static void efx_ef10_tx_fini(struct efx_tx_queue *tx_queue)
1817 {
1818 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FINI_TXQ_IN_LEN);
1819 	MCDI_DECLARE_BUF_ERR(outbuf);
1820 	struct efx_nic *efx = tx_queue->efx;
1821 	size_t outlen;
1822 	int rc;
1823 
1824 	MCDI_SET_DWORD(inbuf, FINI_TXQ_IN_INSTANCE,
1825 		       tx_queue->queue);
1826 
1827 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FINI_TXQ, inbuf, sizeof(inbuf),
1828 			  outbuf, sizeof(outbuf), &outlen);
1829 
1830 	if (rc && rc != -EALREADY)
1831 		goto fail;
1832 
1833 	return;
1834 
1835 fail:
1836 	efx_mcdi_display_error(efx, MC_CMD_FINI_TXQ, MC_CMD_FINI_TXQ_IN_LEN,
1837 			       outbuf, outlen, rc);
1838 }
1839 
1840 static void efx_ef10_tx_remove(struct efx_tx_queue *tx_queue)
1841 {
1842 	efx_nic_free_buffer(tx_queue->efx, &tx_queue->txd.buf);
1843 }
1844 
1845 /* This writes to the TX_DESC_WPTR; write pointer for TX descriptor ring */
1846 static inline void efx_ef10_notify_tx_desc(struct efx_tx_queue *tx_queue)
1847 {
1848 	unsigned int write_ptr;
1849 	efx_dword_t reg;
1850 
1851 	write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
1852 	EFX_POPULATE_DWORD_1(reg, ERF_DZ_TX_DESC_WPTR_DWORD, write_ptr);
1853 	efx_writed_page(tx_queue->efx, &reg,
1854 			ER_DZ_TX_DESC_UPD_DWORD, tx_queue->queue);
1855 }
1856 
1857 static void efx_ef10_tx_write(struct efx_tx_queue *tx_queue)
1858 {
1859 	unsigned int old_write_count = tx_queue->write_count;
1860 	struct efx_tx_buffer *buffer;
1861 	unsigned int write_ptr;
1862 	efx_qword_t *txd;
1863 
1864 	tx_queue->xmit_more_available = false;
1865 	if (unlikely(tx_queue->write_count == tx_queue->insert_count))
1866 		return;
1867 
1868 	do {
1869 		write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
1870 		buffer = &tx_queue->buffer[write_ptr];
1871 		txd = efx_tx_desc(tx_queue, write_ptr);
1872 		++tx_queue->write_count;
1873 
1874 		/* Create TX descriptor ring entry */
1875 		if (buffer->flags & EFX_TX_BUF_OPTION) {
1876 			*txd = buffer->option;
1877 		} else {
1878 			BUILD_BUG_ON(EFX_TX_BUF_CONT != 1);
1879 			EFX_POPULATE_QWORD_3(
1880 				*txd,
1881 				ESF_DZ_TX_KER_CONT,
1882 				buffer->flags & EFX_TX_BUF_CONT,
1883 				ESF_DZ_TX_KER_BYTE_CNT, buffer->len,
1884 				ESF_DZ_TX_KER_BUF_ADDR, buffer->dma_addr);
1885 		}
1886 	} while (tx_queue->write_count != tx_queue->insert_count);
1887 
1888 	wmb(); /* Ensure descriptors are written before they are fetched */
1889 
1890 	if (efx_nic_may_push_tx_desc(tx_queue, old_write_count)) {
1891 		txd = efx_tx_desc(tx_queue,
1892 				  old_write_count & tx_queue->ptr_mask);
1893 		efx_ef10_push_tx_desc(tx_queue, txd);
1894 		++tx_queue->pushes;
1895 	} else {
1896 		efx_ef10_notify_tx_desc(tx_queue);
1897 	}
1898 }
1899 
1900 static int efx_ef10_alloc_rss_context(struct efx_nic *efx, u32 *context,
1901 				      bool exclusive, unsigned *context_size)
1902 {
1903 	MCDI_DECLARE_BUF(inbuf, MC_CMD_RSS_CONTEXT_ALLOC_IN_LEN);
1904 	MCDI_DECLARE_BUF(outbuf, MC_CMD_RSS_CONTEXT_ALLOC_OUT_LEN);
1905 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1906 	size_t outlen;
1907 	int rc;
1908 	u32 alloc_type = exclusive ?
1909 				MC_CMD_RSS_CONTEXT_ALLOC_IN_TYPE_EXCLUSIVE :
1910 				MC_CMD_RSS_CONTEXT_ALLOC_IN_TYPE_SHARED;
1911 	unsigned rss_spread = exclusive ?
1912 				efx->rss_spread :
1913 				min(rounddown_pow_of_two(efx->rss_spread),
1914 				    EFX_EF10_MAX_SHARED_RSS_CONTEXT_SIZE);
1915 
1916 	if (!exclusive && rss_spread == 1) {
1917 		*context = EFX_EF10_RSS_CONTEXT_INVALID;
1918 		if (context_size)
1919 			*context_size = 1;
1920 		return 0;
1921 	}
1922 
1923 	MCDI_SET_DWORD(inbuf, RSS_CONTEXT_ALLOC_IN_UPSTREAM_PORT_ID,
1924 		       nic_data->vport_id);
1925 	MCDI_SET_DWORD(inbuf, RSS_CONTEXT_ALLOC_IN_TYPE, alloc_type);
1926 	MCDI_SET_DWORD(inbuf, RSS_CONTEXT_ALLOC_IN_NUM_QUEUES, rss_spread);
1927 
1928 	rc = efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_ALLOC, inbuf, sizeof(inbuf),
1929 		outbuf, sizeof(outbuf), &outlen);
1930 	if (rc != 0)
1931 		return rc;
1932 
1933 	if (outlen < MC_CMD_RSS_CONTEXT_ALLOC_OUT_LEN)
1934 		return -EIO;
1935 
1936 	*context = MCDI_DWORD(outbuf, RSS_CONTEXT_ALLOC_OUT_RSS_CONTEXT_ID);
1937 
1938 	if (context_size)
1939 		*context_size = rss_spread;
1940 
1941 	return 0;
1942 }
1943 
1944 static void efx_ef10_free_rss_context(struct efx_nic *efx, u32 context)
1945 {
1946 	MCDI_DECLARE_BUF(inbuf, MC_CMD_RSS_CONTEXT_FREE_IN_LEN);
1947 	int rc;
1948 
1949 	MCDI_SET_DWORD(inbuf, RSS_CONTEXT_FREE_IN_RSS_CONTEXT_ID,
1950 		       context);
1951 
1952 	rc = efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_FREE, inbuf, sizeof(inbuf),
1953 			    NULL, 0, NULL);
1954 	WARN_ON(rc != 0);
1955 }
1956 
1957 static int efx_ef10_populate_rss_table(struct efx_nic *efx, u32 context,
1958 				       const u32 *rx_indir_table)
1959 {
1960 	MCDI_DECLARE_BUF(tablebuf, MC_CMD_RSS_CONTEXT_SET_TABLE_IN_LEN);
1961 	MCDI_DECLARE_BUF(keybuf, MC_CMD_RSS_CONTEXT_SET_KEY_IN_LEN);
1962 	int i, rc;
1963 
1964 	MCDI_SET_DWORD(tablebuf, RSS_CONTEXT_SET_TABLE_IN_RSS_CONTEXT_ID,
1965 		       context);
1966 	BUILD_BUG_ON(ARRAY_SIZE(efx->rx_indir_table) !=
1967 		     MC_CMD_RSS_CONTEXT_SET_TABLE_IN_INDIRECTION_TABLE_LEN);
1968 
1969 	for (i = 0; i < ARRAY_SIZE(efx->rx_indir_table); ++i)
1970 		MCDI_PTR(tablebuf,
1971 			 RSS_CONTEXT_SET_TABLE_IN_INDIRECTION_TABLE)[i] =
1972 				(u8) rx_indir_table[i];
1973 
1974 	rc = efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_SET_TABLE, tablebuf,
1975 			  sizeof(tablebuf), NULL, 0, NULL);
1976 	if (rc != 0)
1977 		return rc;
1978 
1979 	MCDI_SET_DWORD(keybuf, RSS_CONTEXT_SET_KEY_IN_RSS_CONTEXT_ID,
1980 		       context);
1981 	BUILD_BUG_ON(ARRAY_SIZE(efx->rx_hash_key) !=
1982 		     MC_CMD_RSS_CONTEXT_SET_KEY_IN_TOEPLITZ_KEY_LEN);
1983 	for (i = 0; i < ARRAY_SIZE(efx->rx_hash_key); ++i)
1984 		MCDI_PTR(keybuf, RSS_CONTEXT_SET_KEY_IN_TOEPLITZ_KEY)[i] =
1985 			efx->rx_hash_key[i];
1986 
1987 	return efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_SET_KEY, keybuf,
1988 			    sizeof(keybuf), NULL, 0, NULL);
1989 }
1990 
1991 static void efx_ef10_rx_free_indir_table(struct efx_nic *efx)
1992 {
1993 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1994 
1995 	if (nic_data->rx_rss_context != EFX_EF10_RSS_CONTEXT_INVALID)
1996 		efx_ef10_free_rss_context(efx, nic_data->rx_rss_context);
1997 	nic_data->rx_rss_context = EFX_EF10_RSS_CONTEXT_INVALID;
1998 }
1999 
2000 static int efx_ef10_rx_push_shared_rss_config(struct efx_nic *efx,
2001 					      unsigned *context_size)
2002 {
2003 	u32 new_rx_rss_context;
2004 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2005 	int rc = efx_ef10_alloc_rss_context(efx, &new_rx_rss_context,
2006 					    false, context_size);
2007 
2008 	if (rc != 0)
2009 		return rc;
2010 
2011 	nic_data->rx_rss_context = new_rx_rss_context;
2012 	nic_data->rx_rss_context_exclusive = false;
2013 	efx_set_default_rx_indir_table(efx);
2014 	return 0;
2015 }
2016 
2017 static int efx_ef10_rx_push_exclusive_rss_config(struct efx_nic *efx,
2018 						 const u32 *rx_indir_table)
2019 {
2020 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2021 	int rc;
2022 	u32 new_rx_rss_context;
2023 
2024 	if (nic_data->rx_rss_context == EFX_EF10_RSS_CONTEXT_INVALID ||
2025 	    !nic_data->rx_rss_context_exclusive) {
2026 		rc = efx_ef10_alloc_rss_context(efx, &new_rx_rss_context,
2027 						true, NULL);
2028 		if (rc == -EOPNOTSUPP)
2029 			return rc;
2030 		else if (rc != 0)
2031 			goto fail1;
2032 	} else {
2033 		new_rx_rss_context = nic_data->rx_rss_context;
2034 	}
2035 
2036 	rc = efx_ef10_populate_rss_table(efx, new_rx_rss_context,
2037 					 rx_indir_table);
2038 	if (rc != 0)
2039 		goto fail2;
2040 
2041 	if (nic_data->rx_rss_context != new_rx_rss_context)
2042 		efx_ef10_rx_free_indir_table(efx);
2043 	nic_data->rx_rss_context = new_rx_rss_context;
2044 	nic_data->rx_rss_context_exclusive = true;
2045 	if (rx_indir_table != efx->rx_indir_table)
2046 		memcpy(efx->rx_indir_table, rx_indir_table,
2047 		       sizeof(efx->rx_indir_table));
2048 	return 0;
2049 
2050 fail2:
2051 	if (new_rx_rss_context != nic_data->rx_rss_context)
2052 		efx_ef10_free_rss_context(efx, new_rx_rss_context);
2053 fail1:
2054 	netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
2055 	return rc;
2056 }
2057 
2058 static int efx_ef10_pf_rx_push_rss_config(struct efx_nic *efx, bool user,
2059 					  const u32 *rx_indir_table)
2060 {
2061 	int rc;
2062 
2063 	if (efx->rss_spread == 1)
2064 		return 0;
2065 
2066 	rc = efx_ef10_rx_push_exclusive_rss_config(efx, rx_indir_table);
2067 
2068 	if (rc == -ENOBUFS && !user) {
2069 		unsigned context_size;
2070 		bool mismatch = false;
2071 		size_t i;
2072 
2073 		for (i = 0; i < ARRAY_SIZE(efx->rx_indir_table) && !mismatch;
2074 		     i++)
2075 			mismatch = rx_indir_table[i] !=
2076 				ethtool_rxfh_indir_default(i, efx->rss_spread);
2077 
2078 		rc = efx_ef10_rx_push_shared_rss_config(efx, &context_size);
2079 		if (rc == 0) {
2080 			if (context_size != efx->rss_spread)
2081 				netif_warn(efx, probe, efx->net_dev,
2082 					   "Could not allocate an exclusive RSS"
2083 					   " context; allocated a shared one of"
2084 					   " different size."
2085 					   " Wanted %u, got %u.\n",
2086 					   efx->rss_spread, context_size);
2087 			else if (mismatch)
2088 				netif_warn(efx, probe, efx->net_dev,
2089 					   "Could not allocate an exclusive RSS"
2090 					   " context; allocated a shared one but"
2091 					   " could not apply custom"
2092 					   " indirection.\n");
2093 			else
2094 				netif_info(efx, probe, efx->net_dev,
2095 					   "Could not allocate an exclusive RSS"
2096 					   " context; allocated a shared one.\n");
2097 		}
2098 	}
2099 	return rc;
2100 }
2101 
2102 static int efx_ef10_vf_rx_push_rss_config(struct efx_nic *efx, bool user,
2103 					  const u32 *rx_indir_table
2104 					  __attribute__ ((unused)))
2105 {
2106 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2107 
2108 	if (user)
2109 		return -EOPNOTSUPP;
2110 	if (nic_data->rx_rss_context != EFX_EF10_RSS_CONTEXT_INVALID)
2111 		return 0;
2112 	return efx_ef10_rx_push_shared_rss_config(efx, NULL);
2113 }
2114 
2115 static int efx_ef10_rx_probe(struct efx_rx_queue *rx_queue)
2116 {
2117 	return efx_nic_alloc_buffer(rx_queue->efx, &rx_queue->rxd.buf,
2118 				    (rx_queue->ptr_mask + 1) *
2119 				    sizeof(efx_qword_t),
2120 				    GFP_KERNEL);
2121 }
2122 
2123 static void efx_ef10_rx_init(struct efx_rx_queue *rx_queue)
2124 {
2125 	MCDI_DECLARE_BUF(inbuf,
2126 			 MC_CMD_INIT_RXQ_IN_LEN(EFX_MAX_DMAQ_SIZE * 8 /
2127 						EFX_BUF_SIZE));
2128 	struct efx_channel *channel = efx_rx_queue_channel(rx_queue);
2129 	size_t entries = rx_queue->rxd.buf.len / EFX_BUF_SIZE;
2130 	struct efx_nic *efx = rx_queue->efx;
2131 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2132 	size_t inlen;
2133 	dma_addr_t dma_addr;
2134 	int rc;
2135 	int i;
2136 	BUILD_BUG_ON(MC_CMD_INIT_RXQ_OUT_LEN != 0);
2137 
2138 	rx_queue->scatter_n = 0;
2139 	rx_queue->scatter_len = 0;
2140 
2141 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_SIZE, rx_queue->ptr_mask + 1);
2142 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_TARGET_EVQ, channel->channel);
2143 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_LABEL, efx_rx_queue_index(rx_queue));
2144 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_INSTANCE,
2145 		       efx_rx_queue_index(rx_queue));
2146 	MCDI_POPULATE_DWORD_2(inbuf, INIT_RXQ_IN_FLAGS,
2147 			      INIT_RXQ_IN_FLAG_PREFIX, 1,
2148 			      INIT_RXQ_IN_FLAG_TIMESTAMP, 1);
2149 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_OWNER_ID, 0);
2150 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_PORT_ID, nic_data->vport_id);
2151 
2152 	dma_addr = rx_queue->rxd.buf.dma_addr;
2153 
2154 	netif_dbg(efx, hw, efx->net_dev, "pushing RXQ %d. %zu entries (%llx)\n",
2155 		  efx_rx_queue_index(rx_queue), entries, (u64)dma_addr);
2156 
2157 	for (i = 0; i < entries; ++i) {
2158 		MCDI_SET_ARRAY_QWORD(inbuf, INIT_RXQ_IN_DMA_ADDR, i, dma_addr);
2159 		dma_addr += EFX_BUF_SIZE;
2160 	}
2161 
2162 	inlen = MC_CMD_INIT_RXQ_IN_LEN(entries);
2163 
2164 	rc = efx_mcdi_rpc(efx, MC_CMD_INIT_RXQ, inbuf, inlen,
2165 			  NULL, 0, NULL);
2166 	if (rc)
2167 		netdev_WARN(efx->net_dev, "failed to initialise RXQ %d\n",
2168 			    efx_rx_queue_index(rx_queue));
2169 }
2170 
2171 static void efx_ef10_rx_fini(struct efx_rx_queue *rx_queue)
2172 {
2173 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FINI_RXQ_IN_LEN);
2174 	MCDI_DECLARE_BUF_ERR(outbuf);
2175 	struct efx_nic *efx = rx_queue->efx;
2176 	size_t outlen;
2177 	int rc;
2178 
2179 	MCDI_SET_DWORD(inbuf, FINI_RXQ_IN_INSTANCE,
2180 		       efx_rx_queue_index(rx_queue));
2181 
2182 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FINI_RXQ, inbuf, sizeof(inbuf),
2183 			  outbuf, sizeof(outbuf), &outlen);
2184 
2185 	if (rc && rc != -EALREADY)
2186 		goto fail;
2187 
2188 	return;
2189 
2190 fail:
2191 	efx_mcdi_display_error(efx, MC_CMD_FINI_RXQ, MC_CMD_FINI_RXQ_IN_LEN,
2192 			       outbuf, outlen, rc);
2193 }
2194 
2195 static void efx_ef10_rx_remove(struct efx_rx_queue *rx_queue)
2196 {
2197 	efx_nic_free_buffer(rx_queue->efx, &rx_queue->rxd.buf);
2198 }
2199 
2200 /* This creates an entry in the RX descriptor queue */
2201 static inline void
2202 efx_ef10_build_rx_desc(struct efx_rx_queue *rx_queue, unsigned int index)
2203 {
2204 	struct efx_rx_buffer *rx_buf;
2205 	efx_qword_t *rxd;
2206 
2207 	rxd = efx_rx_desc(rx_queue, index);
2208 	rx_buf = efx_rx_buffer(rx_queue, index);
2209 	EFX_POPULATE_QWORD_2(*rxd,
2210 			     ESF_DZ_RX_KER_BYTE_CNT, rx_buf->len,
2211 			     ESF_DZ_RX_KER_BUF_ADDR, rx_buf->dma_addr);
2212 }
2213 
2214 static void efx_ef10_rx_write(struct efx_rx_queue *rx_queue)
2215 {
2216 	struct efx_nic *efx = rx_queue->efx;
2217 	unsigned int write_count;
2218 	efx_dword_t reg;
2219 
2220 	/* Firmware requires that RX_DESC_WPTR be a multiple of 8 */
2221 	write_count = rx_queue->added_count & ~7;
2222 	if (rx_queue->notified_count == write_count)
2223 		return;
2224 
2225 	do
2226 		efx_ef10_build_rx_desc(
2227 			rx_queue,
2228 			rx_queue->notified_count & rx_queue->ptr_mask);
2229 	while (++rx_queue->notified_count != write_count);
2230 
2231 	wmb();
2232 	EFX_POPULATE_DWORD_1(reg, ERF_DZ_RX_DESC_WPTR,
2233 			     write_count & rx_queue->ptr_mask);
2234 	efx_writed_page(efx, &reg, ER_DZ_RX_DESC_UPD,
2235 			efx_rx_queue_index(rx_queue));
2236 }
2237 
2238 static efx_mcdi_async_completer efx_ef10_rx_defer_refill_complete;
2239 
2240 static void efx_ef10_rx_defer_refill(struct efx_rx_queue *rx_queue)
2241 {
2242 	struct efx_channel *channel = efx_rx_queue_channel(rx_queue);
2243 	MCDI_DECLARE_BUF(inbuf, MC_CMD_DRIVER_EVENT_IN_LEN);
2244 	efx_qword_t event;
2245 
2246 	EFX_POPULATE_QWORD_2(event,
2247 			     ESF_DZ_EV_CODE, EFX_EF10_DRVGEN_EV,
2248 			     ESF_DZ_EV_DATA, EFX_EF10_REFILL);
2249 
2250 	MCDI_SET_DWORD(inbuf, DRIVER_EVENT_IN_EVQ, channel->channel);
2251 
2252 	/* MCDI_SET_QWORD is not appropriate here since EFX_POPULATE_* has
2253 	 * already swapped the data to little-endian order.
2254 	 */
2255 	memcpy(MCDI_PTR(inbuf, DRIVER_EVENT_IN_DATA), &event.u64[0],
2256 	       sizeof(efx_qword_t));
2257 
2258 	efx_mcdi_rpc_async(channel->efx, MC_CMD_DRIVER_EVENT,
2259 			   inbuf, sizeof(inbuf), 0,
2260 			   efx_ef10_rx_defer_refill_complete, 0);
2261 }
2262 
2263 static void
2264 efx_ef10_rx_defer_refill_complete(struct efx_nic *efx, unsigned long cookie,
2265 				  int rc, efx_dword_t *outbuf,
2266 				  size_t outlen_actual)
2267 {
2268 	/* nothing to do */
2269 }
2270 
2271 static int efx_ef10_ev_probe(struct efx_channel *channel)
2272 {
2273 	return efx_nic_alloc_buffer(channel->efx, &channel->eventq.buf,
2274 				    (channel->eventq_mask + 1) *
2275 				    sizeof(efx_qword_t),
2276 				    GFP_KERNEL);
2277 }
2278 
2279 static void efx_ef10_ev_fini(struct efx_channel *channel)
2280 {
2281 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FINI_EVQ_IN_LEN);
2282 	MCDI_DECLARE_BUF_ERR(outbuf);
2283 	struct efx_nic *efx = channel->efx;
2284 	size_t outlen;
2285 	int rc;
2286 
2287 	MCDI_SET_DWORD(inbuf, FINI_EVQ_IN_INSTANCE, channel->channel);
2288 
2289 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FINI_EVQ, inbuf, sizeof(inbuf),
2290 			  outbuf, sizeof(outbuf), &outlen);
2291 
2292 	if (rc && rc != -EALREADY)
2293 		goto fail;
2294 
2295 	return;
2296 
2297 fail:
2298 	efx_mcdi_display_error(efx, MC_CMD_FINI_EVQ, MC_CMD_FINI_EVQ_IN_LEN,
2299 			       outbuf, outlen, rc);
2300 }
2301 
2302 static int efx_ef10_ev_init(struct efx_channel *channel)
2303 {
2304 	MCDI_DECLARE_BUF(inbuf,
2305 			 MC_CMD_INIT_EVQ_IN_LEN(EFX_MAX_EVQ_SIZE * 8 /
2306 						EFX_BUF_SIZE));
2307 	MCDI_DECLARE_BUF(outbuf, MC_CMD_INIT_EVQ_OUT_LEN);
2308 	size_t entries = channel->eventq.buf.len / EFX_BUF_SIZE;
2309 	struct efx_nic *efx = channel->efx;
2310 	struct efx_ef10_nic_data *nic_data;
2311 	bool supports_rx_merge;
2312 	size_t inlen, outlen;
2313 	unsigned int enabled, implemented;
2314 	dma_addr_t dma_addr;
2315 	int rc;
2316 	int i;
2317 
2318 	nic_data = efx->nic_data;
2319 	supports_rx_merge =
2320 		!!(nic_data->datapath_caps &
2321 		   1 << MC_CMD_GET_CAPABILITIES_OUT_RX_BATCHING_LBN);
2322 
2323 	/* Fill event queue with all ones (i.e. empty events) */
2324 	memset(channel->eventq.buf.addr, 0xff, channel->eventq.buf.len);
2325 
2326 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_SIZE, channel->eventq_mask + 1);
2327 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_INSTANCE, channel->channel);
2328 	/* INIT_EVQ expects index in vector table, not absolute */
2329 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_IRQ_NUM, channel->channel);
2330 	MCDI_POPULATE_DWORD_4(inbuf, INIT_EVQ_IN_FLAGS,
2331 			      INIT_EVQ_IN_FLAG_INTERRUPTING, 1,
2332 			      INIT_EVQ_IN_FLAG_RX_MERGE, 1,
2333 			      INIT_EVQ_IN_FLAG_TX_MERGE, 1,
2334 			      INIT_EVQ_IN_FLAG_CUT_THRU, !supports_rx_merge);
2335 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_TMR_MODE,
2336 		       MC_CMD_INIT_EVQ_IN_TMR_MODE_DIS);
2337 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_TMR_LOAD, 0);
2338 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_TMR_RELOAD, 0);
2339 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_COUNT_MODE,
2340 		       MC_CMD_INIT_EVQ_IN_COUNT_MODE_DIS);
2341 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_COUNT_THRSHLD, 0);
2342 
2343 	dma_addr = channel->eventq.buf.dma_addr;
2344 	for (i = 0; i < entries; ++i) {
2345 		MCDI_SET_ARRAY_QWORD(inbuf, INIT_EVQ_IN_DMA_ADDR, i, dma_addr);
2346 		dma_addr += EFX_BUF_SIZE;
2347 	}
2348 
2349 	inlen = MC_CMD_INIT_EVQ_IN_LEN(entries);
2350 
2351 	rc = efx_mcdi_rpc(efx, MC_CMD_INIT_EVQ, inbuf, inlen,
2352 			  outbuf, sizeof(outbuf), &outlen);
2353 	/* IRQ return is ignored */
2354 	if (channel->channel || rc)
2355 		return rc;
2356 
2357 	/* Successfully created event queue on channel 0 */
2358 	rc = efx_mcdi_get_workarounds(efx, &implemented, &enabled);
2359 	if (rc == -ENOSYS) {
2360 		/* GET_WORKAROUNDS was implemented before the bug26807
2361 		 * workaround, thus the latter must be unavailable in this fw
2362 		 */
2363 		nic_data->workaround_26807 = false;
2364 		rc = 0;
2365 	} else if (rc) {
2366 		goto fail;
2367 	} else {
2368 		nic_data->workaround_26807 =
2369 			!!(enabled & MC_CMD_GET_WORKAROUNDS_OUT_BUG26807);
2370 
2371 		if (implemented & MC_CMD_GET_WORKAROUNDS_OUT_BUG26807 &&
2372 		    !nic_data->workaround_26807) {
2373 			unsigned int flags;
2374 
2375 			rc = efx_mcdi_set_workaround(efx,
2376 						     MC_CMD_WORKAROUND_BUG26807,
2377 						     true, &flags);
2378 
2379 			if (!rc) {
2380 				if (flags &
2381 				    1 << MC_CMD_WORKAROUND_EXT_OUT_FLR_DONE_LBN) {
2382 					netif_info(efx, drv, efx->net_dev,
2383 						   "other functions on NIC have been reset\n");
2384 
2385 					/* With MCFW v4.6.x and earlier, the
2386 					 * boot count will have incremented,
2387 					 * so re-read the warm_boot_count
2388 					 * value now to ensure this function
2389 					 * doesn't think it has changed next
2390 					 * time it checks.
2391 					 */
2392 					rc = efx_ef10_get_warm_boot_count(efx);
2393 					if (rc >= 0) {
2394 						nic_data->warm_boot_count = rc;
2395 						rc = 0;
2396 					}
2397 				}
2398 				nic_data->workaround_26807 = true;
2399 			} else if (rc == -EPERM) {
2400 				rc = 0;
2401 			}
2402 		}
2403 	}
2404 
2405 	if (!rc)
2406 		return 0;
2407 
2408 fail:
2409 	efx_ef10_ev_fini(channel);
2410 	return rc;
2411 }
2412 
2413 static void efx_ef10_ev_remove(struct efx_channel *channel)
2414 {
2415 	efx_nic_free_buffer(channel->efx, &channel->eventq.buf);
2416 }
2417 
2418 static void efx_ef10_handle_rx_wrong_queue(struct efx_rx_queue *rx_queue,
2419 					   unsigned int rx_queue_label)
2420 {
2421 	struct efx_nic *efx = rx_queue->efx;
2422 
2423 	netif_info(efx, hw, efx->net_dev,
2424 		   "rx event arrived on queue %d labeled as queue %u\n",
2425 		   efx_rx_queue_index(rx_queue), rx_queue_label);
2426 
2427 	efx_schedule_reset(efx, RESET_TYPE_DISABLE);
2428 }
2429 
2430 static void
2431 efx_ef10_handle_rx_bad_lbits(struct efx_rx_queue *rx_queue,
2432 			     unsigned int actual, unsigned int expected)
2433 {
2434 	unsigned int dropped = (actual - expected) & rx_queue->ptr_mask;
2435 	struct efx_nic *efx = rx_queue->efx;
2436 
2437 	netif_info(efx, hw, efx->net_dev,
2438 		   "dropped %d events (index=%d expected=%d)\n",
2439 		   dropped, actual, expected);
2440 
2441 	efx_schedule_reset(efx, RESET_TYPE_DISABLE);
2442 }
2443 
2444 /* partially received RX was aborted. clean up. */
2445 static void efx_ef10_handle_rx_abort(struct efx_rx_queue *rx_queue)
2446 {
2447 	unsigned int rx_desc_ptr;
2448 
2449 	netif_dbg(rx_queue->efx, hw, rx_queue->efx->net_dev,
2450 		  "scattered RX aborted (dropping %u buffers)\n",
2451 		  rx_queue->scatter_n);
2452 
2453 	rx_desc_ptr = rx_queue->removed_count & rx_queue->ptr_mask;
2454 
2455 	efx_rx_packet(rx_queue, rx_desc_ptr, rx_queue->scatter_n,
2456 		      0, EFX_RX_PKT_DISCARD);
2457 
2458 	rx_queue->removed_count += rx_queue->scatter_n;
2459 	rx_queue->scatter_n = 0;
2460 	rx_queue->scatter_len = 0;
2461 	++efx_rx_queue_channel(rx_queue)->n_rx_nodesc_trunc;
2462 }
2463 
2464 static int efx_ef10_handle_rx_event(struct efx_channel *channel,
2465 				    const efx_qword_t *event)
2466 {
2467 	unsigned int rx_bytes, next_ptr_lbits, rx_queue_label, rx_l4_class;
2468 	unsigned int n_descs, n_packets, i;
2469 	struct efx_nic *efx = channel->efx;
2470 	struct efx_rx_queue *rx_queue;
2471 	bool rx_cont;
2472 	u16 flags = 0;
2473 
2474 	if (unlikely(ACCESS_ONCE(efx->reset_pending)))
2475 		return 0;
2476 
2477 	/* Basic packet information */
2478 	rx_bytes = EFX_QWORD_FIELD(*event, ESF_DZ_RX_BYTES);
2479 	next_ptr_lbits = EFX_QWORD_FIELD(*event, ESF_DZ_RX_DSC_PTR_LBITS);
2480 	rx_queue_label = EFX_QWORD_FIELD(*event, ESF_DZ_RX_QLABEL);
2481 	rx_l4_class = EFX_QWORD_FIELD(*event, ESF_DZ_RX_L4_CLASS);
2482 	rx_cont = EFX_QWORD_FIELD(*event, ESF_DZ_RX_CONT);
2483 
2484 	if (EFX_QWORD_FIELD(*event, ESF_DZ_RX_DROP_EVENT))
2485 		netdev_WARN(efx->net_dev, "saw RX_DROP_EVENT: event="
2486 			    EFX_QWORD_FMT "\n",
2487 			    EFX_QWORD_VAL(*event));
2488 
2489 	rx_queue = efx_channel_get_rx_queue(channel);
2490 
2491 	if (unlikely(rx_queue_label != efx_rx_queue_index(rx_queue)))
2492 		efx_ef10_handle_rx_wrong_queue(rx_queue, rx_queue_label);
2493 
2494 	n_descs = ((next_ptr_lbits - rx_queue->removed_count) &
2495 		   ((1 << ESF_DZ_RX_DSC_PTR_LBITS_WIDTH) - 1));
2496 
2497 	if (n_descs != rx_queue->scatter_n + 1) {
2498 		struct efx_ef10_nic_data *nic_data = efx->nic_data;
2499 
2500 		/* detect rx abort */
2501 		if (unlikely(n_descs == rx_queue->scatter_n)) {
2502 			if (rx_queue->scatter_n == 0 || rx_bytes != 0)
2503 				netdev_WARN(efx->net_dev,
2504 					    "invalid RX abort: scatter_n=%u event="
2505 					    EFX_QWORD_FMT "\n",
2506 					    rx_queue->scatter_n,
2507 					    EFX_QWORD_VAL(*event));
2508 			efx_ef10_handle_rx_abort(rx_queue);
2509 			return 0;
2510 		}
2511 
2512 		/* Check that RX completion merging is valid, i.e.
2513 		 * the current firmware supports it and this is a
2514 		 * non-scattered packet.
2515 		 */
2516 		if (!(nic_data->datapath_caps &
2517 		      (1 << MC_CMD_GET_CAPABILITIES_OUT_RX_BATCHING_LBN)) ||
2518 		    rx_queue->scatter_n != 0 || rx_cont) {
2519 			efx_ef10_handle_rx_bad_lbits(
2520 				rx_queue, next_ptr_lbits,
2521 				(rx_queue->removed_count +
2522 				 rx_queue->scatter_n + 1) &
2523 				((1 << ESF_DZ_RX_DSC_PTR_LBITS_WIDTH) - 1));
2524 			return 0;
2525 		}
2526 
2527 		/* Merged completion for multiple non-scattered packets */
2528 		rx_queue->scatter_n = 1;
2529 		rx_queue->scatter_len = 0;
2530 		n_packets = n_descs;
2531 		++channel->n_rx_merge_events;
2532 		channel->n_rx_merge_packets += n_packets;
2533 		flags |= EFX_RX_PKT_PREFIX_LEN;
2534 	} else {
2535 		++rx_queue->scatter_n;
2536 		rx_queue->scatter_len += rx_bytes;
2537 		if (rx_cont)
2538 			return 0;
2539 		n_packets = 1;
2540 	}
2541 
2542 	if (unlikely(EFX_QWORD_FIELD(*event, ESF_DZ_RX_ECRC_ERR)))
2543 		flags |= EFX_RX_PKT_DISCARD;
2544 
2545 	if (unlikely(EFX_QWORD_FIELD(*event, ESF_DZ_RX_IPCKSUM_ERR))) {
2546 		channel->n_rx_ip_hdr_chksum_err += n_packets;
2547 	} else if (unlikely(EFX_QWORD_FIELD(*event,
2548 					    ESF_DZ_RX_TCPUDP_CKSUM_ERR))) {
2549 		channel->n_rx_tcp_udp_chksum_err += n_packets;
2550 	} else if (rx_l4_class == ESE_DZ_L4_CLASS_TCP ||
2551 		   rx_l4_class == ESE_DZ_L4_CLASS_UDP) {
2552 		flags |= EFX_RX_PKT_CSUMMED;
2553 	}
2554 
2555 	if (rx_l4_class == ESE_DZ_L4_CLASS_TCP)
2556 		flags |= EFX_RX_PKT_TCP;
2557 
2558 	channel->irq_mod_score += 2 * n_packets;
2559 
2560 	/* Handle received packet(s) */
2561 	for (i = 0; i < n_packets; i++) {
2562 		efx_rx_packet(rx_queue,
2563 			      rx_queue->removed_count & rx_queue->ptr_mask,
2564 			      rx_queue->scatter_n, rx_queue->scatter_len,
2565 			      flags);
2566 		rx_queue->removed_count += rx_queue->scatter_n;
2567 	}
2568 
2569 	rx_queue->scatter_n = 0;
2570 	rx_queue->scatter_len = 0;
2571 
2572 	return n_packets;
2573 }
2574 
2575 static int
2576 efx_ef10_handle_tx_event(struct efx_channel *channel, efx_qword_t *event)
2577 {
2578 	struct efx_nic *efx = channel->efx;
2579 	struct efx_tx_queue *tx_queue;
2580 	unsigned int tx_ev_desc_ptr;
2581 	unsigned int tx_ev_q_label;
2582 	int tx_descs = 0;
2583 
2584 	if (unlikely(ACCESS_ONCE(efx->reset_pending)))
2585 		return 0;
2586 
2587 	if (unlikely(EFX_QWORD_FIELD(*event, ESF_DZ_TX_DROP_EVENT)))
2588 		return 0;
2589 
2590 	/* Transmit completion */
2591 	tx_ev_desc_ptr = EFX_QWORD_FIELD(*event, ESF_DZ_TX_DESCR_INDX);
2592 	tx_ev_q_label = EFX_QWORD_FIELD(*event, ESF_DZ_TX_QLABEL);
2593 	tx_queue = efx_channel_get_tx_queue(channel,
2594 					    tx_ev_q_label % EFX_TXQ_TYPES);
2595 	tx_descs = ((tx_ev_desc_ptr + 1 - tx_queue->read_count) &
2596 		    tx_queue->ptr_mask);
2597 	efx_xmit_done(tx_queue, tx_ev_desc_ptr & tx_queue->ptr_mask);
2598 
2599 	return tx_descs;
2600 }
2601 
2602 static void
2603 efx_ef10_handle_driver_event(struct efx_channel *channel, efx_qword_t *event)
2604 {
2605 	struct efx_nic *efx = channel->efx;
2606 	int subcode;
2607 
2608 	subcode = EFX_QWORD_FIELD(*event, ESF_DZ_DRV_SUB_CODE);
2609 
2610 	switch (subcode) {
2611 	case ESE_DZ_DRV_TIMER_EV:
2612 	case ESE_DZ_DRV_WAKE_UP_EV:
2613 		break;
2614 	case ESE_DZ_DRV_START_UP_EV:
2615 		/* event queue init complete. ok. */
2616 		break;
2617 	default:
2618 		netif_err(efx, hw, efx->net_dev,
2619 			  "channel %d unknown driver event type %d"
2620 			  " (data " EFX_QWORD_FMT ")\n",
2621 			  channel->channel, subcode,
2622 			  EFX_QWORD_VAL(*event));
2623 
2624 	}
2625 }
2626 
2627 static void efx_ef10_handle_driver_generated_event(struct efx_channel *channel,
2628 						   efx_qword_t *event)
2629 {
2630 	struct efx_nic *efx = channel->efx;
2631 	u32 subcode;
2632 
2633 	subcode = EFX_QWORD_FIELD(*event, EFX_DWORD_0);
2634 
2635 	switch (subcode) {
2636 	case EFX_EF10_TEST:
2637 		channel->event_test_cpu = raw_smp_processor_id();
2638 		break;
2639 	case EFX_EF10_REFILL:
2640 		/* The queue must be empty, so we won't receive any rx
2641 		 * events, so efx_process_channel() won't refill the
2642 		 * queue. Refill it here
2643 		 */
2644 		efx_fast_push_rx_descriptors(&channel->rx_queue, true);
2645 		break;
2646 	default:
2647 		netif_err(efx, hw, efx->net_dev,
2648 			  "channel %d unknown driver event type %u"
2649 			  " (data " EFX_QWORD_FMT ")\n",
2650 			  channel->channel, (unsigned) subcode,
2651 			  EFX_QWORD_VAL(*event));
2652 	}
2653 }
2654 
2655 static int efx_ef10_ev_process(struct efx_channel *channel, int quota)
2656 {
2657 	struct efx_nic *efx = channel->efx;
2658 	efx_qword_t event, *p_event;
2659 	unsigned int read_ptr;
2660 	int ev_code;
2661 	int tx_descs = 0;
2662 	int spent = 0;
2663 
2664 	if (quota <= 0)
2665 		return spent;
2666 
2667 	read_ptr = channel->eventq_read_ptr;
2668 
2669 	for (;;) {
2670 		p_event = efx_event(channel, read_ptr);
2671 		event = *p_event;
2672 
2673 		if (!efx_event_present(&event))
2674 			break;
2675 
2676 		EFX_SET_QWORD(*p_event);
2677 
2678 		++read_ptr;
2679 
2680 		ev_code = EFX_QWORD_FIELD(event, ESF_DZ_EV_CODE);
2681 
2682 		netif_vdbg(efx, drv, efx->net_dev,
2683 			   "processing event on %d " EFX_QWORD_FMT "\n",
2684 			   channel->channel, EFX_QWORD_VAL(event));
2685 
2686 		switch (ev_code) {
2687 		case ESE_DZ_EV_CODE_MCDI_EV:
2688 			efx_mcdi_process_event(channel, &event);
2689 			break;
2690 		case ESE_DZ_EV_CODE_RX_EV:
2691 			spent += efx_ef10_handle_rx_event(channel, &event);
2692 			if (spent >= quota) {
2693 				/* XXX can we split a merged event to
2694 				 * avoid going over-quota?
2695 				 */
2696 				spent = quota;
2697 				goto out;
2698 			}
2699 			break;
2700 		case ESE_DZ_EV_CODE_TX_EV:
2701 			tx_descs += efx_ef10_handle_tx_event(channel, &event);
2702 			if (tx_descs > efx->txq_entries) {
2703 				spent = quota;
2704 				goto out;
2705 			} else if (++spent == quota) {
2706 				goto out;
2707 			}
2708 			break;
2709 		case ESE_DZ_EV_CODE_DRIVER_EV:
2710 			efx_ef10_handle_driver_event(channel, &event);
2711 			if (++spent == quota)
2712 				goto out;
2713 			break;
2714 		case EFX_EF10_DRVGEN_EV:
2715 			efx_ef10_handle_driver_generated_event(channel, &event);
2716 			break;
2717 		default:
2718 			netif_err(efx, hw, efx->net_dev,
2719 				  "channel %d unknown event type %d"
2720 				  " (data " EFX_QWORD_FMT ")\n",
2721 				  channel->channel, ev_code,
2722 				  EFX_QWORD_VAL(event));
2723 		}
2724 	}
2725 
2726 out:
2727 	channel->eventq_read_ptr = read_ptr;
2728 	return spent;
2729 }
2730 
2731 static void efx_ef10_ev_read_ack(struct efx_channel *channel)
2732 {
2733 	struct efx_nic *efx = channel->efx;
2734 	efx_dword_t rptr;
2735 
2736 	if (EFX_EF10_WORKAROUND_35388(efx)) {
2737 		BUILD_BUG_ON(EFX_MIN_EVQ_SIZE <
2738 			     (1 << ERF_DD_EVQ_IND_RPTR_WIDTH));
2739 		BUILD_BUG_ON(EFX_MAX_EVQ_SIZE >
2740 			     (1 << 2 * ERF_DD_EVQ_IND_RPTR_WIDTH));
2741 
2742 		EFX_POPULATE_DWORD_2(rptr, ERF_DD_EVQ_IND_RPTR_FLAGS,
2743 				     EFE_DD_EVQ_IND_RPTR_FLAGS_HIGH,
2744 				     ERF_DD_EVQ_IND_RPTR,
2745 				     (channel->eventq_read_ptr &
2746 				      channel->eventq_mask) >>
2747 				     ERF_DD_EVQ_IND_RPTR_WIDTH);
2748 		efx_writed_page(efx, &rptr, ER_DD_EVQ_INDIRECT,
2749 				channel->channel);
2750 		EFX_POPULATE_DWORD_2(rptr, ERF_DD_EVQ_IND_RPTR_FLAGS,
2751 				     EFE_DD_EVQ_IND_RPTR_FLAGS_LOW,
2752 				     ERF_DD_EVQ_IND_RPTR,
2753 				     channel->eventq_read_ptr &
2754 				     ((1 << ERF_DD_EVQ_IND_RPTR_WIDTH) - 1));
2755 		efx_writed_page(efx, &rptr, ER_DD_EVQ_INDIRECT,
2756 				channel->channel);
2757 	} else {
2758 		EFX_POPULATE_DWORD_1(rptr, ERF_DZ_EVQ_RPTR,
2759 				     channel->eventq_read_ptr &
2760 				     channel->eventq_mask);
2761 		efx_writed_page(efx, &rptr, ER_DZ_EVQ_RPTR, channel->channel);
2762 	}
2763 }
2764 
2765 static void efx_ef10_ev_test_generate(struct efx_channel *channel)
2766 {
2767 	MCDI_DECLARE_BUF(inbuf, MC_CMD_DRIVER_EVENT_IN_LEN);
2768 	struct efx_nic *efx = channel->efx;
2769 	efx_qword_t event;
2770 	int rc;
2771 
2772 	EFX_POPULATE_QWORD_2(event,
2773 			     ESF_DZ_EV_CODE, EFX_EF10_DRVGEN_EV,
2774 			     ESF_DZ_EV_DATA, EFX_EF10_TEST);
2775 
2776 	MCDI_SET_DWORD(inbuf, DRIVER_EVENT_IN_EVQ, channel->channel);
2777 
2778 	/* MCDI_SET_QWORD is not appropriate here since EFX_POPULATE_* has
2779 	 * already swapped the data to little-endian order.
2780 	 */
2781 	memcpy(MCDI_PTR(inbuf, DRIVER_EVENT_IN_DATA), &event.u64[0],
2782 	       sizeof(efx_qword_t));
2783 
2784 	rc = efx_mcdi_rpc(efx, MC_CMD_DRIVER_EVENT, inbuf, sizeof(inbuf),
2785 			  NULL, 0, NULL);
2786 	if (rc != 0)
2787 		goto fail;
2788 
2789 	return;
2790 
2791 fail:
2792 	WARN_ON(true);
2793 	netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
2794 }
2795 
2796 void efx_ef10_handle_drain_event(struct efx_nic *efx)
2797 {
2798 	if (atomic_dec_and_test(&efx->active_queues))
2799 		wake_up(&efx->flush_wq);
2800 
2801 	WARN_ON(atomic_read(&efx->active_queues) < 0);
2802 }
2803 
2804 static int efx_ef10_fini_dmaq(struct efx_nic *efx)
2805 {
2806 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2807 	struct efx_channel *channel;
2808 	struct efx_tx_queue *tx_queue;
2809 	struct efx_rx_queue *rx_queue;
2810 	int pending;
2811 
2812 	/* If the MC has just rebooted, the TX/RX queues will have already been
2813 	 * torn down, but efx->active_queues needs to be set to zero.
2814 	 */
2815 	if (nic_data->must_realloc_vis) {
2816 		atomic_set(&efx->active_queues, 0);
2817 		return 0;
2818 	}
2819 
2820 	/* Do not attempt to write to the NIC during EEH recovery */
2821 	if (efx->state != STATE_RECOVERY) {
2822 		efx_for_each_channel(channel, efx) {
2823 			efx_for_each_channel_rx_queue(rx_queue, channel)
2824 				efx_ef10_rx_fini(rx_queue);
2825 			efx_for_each_channel_tx_queue(tx_queue, channel)
2826 				efx_ef10_tx_fini(tx_queue);
2827 		}
2828 
2829 		wait_event_timeout(efx->flush_wq,
2830 				   atomic_read(&efx->active_queues) == 0,
2831 				   msecs_to_jiffies(EFX_MAX_FLUSH_TIME));
2832 		pending = atomic_read(&efx->active_queues);
2833 		if (pending) {
2834 			netif_err(efx, hw, efx->net_dev, "failed to flush %d queues\n",
2835 				  pending);
2836 			return -ETIMEDOUT;
2837 		}
2838 	}
2839 
2840 	return 0;
2841 }
2842 
2843 static void efx_ef10_prepare_flr(struct efx_nic *efx)
2844 {
2845 	atomic_set(&efx->active_queues, 0);
2846 }
2847 
2848 static bool efx_ef10_filter_equal(const struct efx_filter_spec *left,
2849 				  const struct efx_filter_spec *right)
2850 {
2851 	if ((left->match_flags ^ right->match_flags) |
2852 	    ((left->flags ^ right->flags) &
2853 	     (EFX_FILTER_FLAG_RX | EFX_FILTER_FLAG_TX)))
2854 		return false;
2855 
2856 	return memcmp(&left->outer_vid, &right->outer_vid,
2857 		      sizeof(struct efx_filter_spec) -
2858 		      offsetof(struct efx_filter_spec, outer_vid)) == 0;
2859 }
2860 
2861 static unsigned int efx_ef10_filter_hash(const struct efx_filter_spec *spec)
2862 {
2863 	BUILD_BUG_ON(offsetof(struct efx_filter_spec, outer_vid) & 3);
2864 	return jhash2((const u32 *)&spec->outer_vid,
2865 		      (sizeof(struct efx_filter_spec) -
2866 		       offsetof(struct efx_filter_spec, outer_vid)) / 4,
2867 		      0);
2868 	/* XXX should we randomise the initval? */
2869 }
2870 
2871 /* Decide whether a filter should be exclusive or else should allow
2872  * delivery to additional recipients.  Currently we decide that
2873  * filters for specific local unicast MAC and IP addresses are
2874  * exclusive.
2875  */
2876 static bool efx_ef10_filter_is_exclusive(const struct efx_filter_spec *spec)
2877 {
2878 	if (spec->match_flags & EFX_FILTER_MATCH_LOC_MAC &&
2879 	    !is_multicast_ether_addr(spec->loc_mac))
2880 		return true;
2881 
2882 	if ((spec->match_flags &
2883 	     (EFX_FILTER_MATCH_ETHER_TYPE | EFX_FILTER_MATCH_LOC_HOST)) ==
2884 	    (EFX_FILTER_MATCH_ETHER_TYPE | EFX_FILTER_MATCH_LOC_HOST)) {
2885 		if (spec->ether_type == htons(ETH_P_IP) &&
2886 		    !ipv4_is_multicast(spec->loc_host[0]))
2887 			return true;
2888 		if (spec->ether_type == htons(ETH_P_IPV6) &&
2889 		    ((const u8 *)spec->loc_host)[0] != 0xff)
2890 			return true;
2891 	}
2892 
2893 	return false;
2894 }
2895 
2896 static struct efx_filter_spec *
2897 efx_ef10_filter_entry_spec(const struct efx_ef10_filter_table *table,
2898 			   unsigned int filter_idx)
2899 {
2900 	return (struct efx_filter_spec *)(table->entry[filter_idx].spec &
2901 					  ~EFX_EF10_FILTER_FLAGS);
2902 }
2903 
2904 static unsigned int
2905 efx_ef10_filter_entry_flags(const struct efx_ef10_filter_table *table,
2906 			   unsigned int filter_idx)
2907 {
2908 	return table->entry[filter_idx].spec & EFX_EF10_FILTER_FLAGS;
2909 }
2910 
2911 static void
2912 efx_ef10_filter_set_entry(struct efx_ef10_filter_table *table,
2913 			  unsigned int filter_idx,
2914 			  const struct efx_filter_spec *spec,
2915 			  unsigned int flags)
2916 {
2917 	table->entry[filter_idx].spec =	(unsigned long)spec | flags;
2918 }
2919 
2920 static void efx_ef10_filter_push_prep(struct efx_nic *efx,
2921 				      const struct efx_filter_spec *spec,
2922 				      efx_dword_t *inbuf, u64 handle,
2923 				      bool replacing)
2924 {
2925 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2926 
2927 	memset(inbuf, 0, MC_CMD_FILTER_OP_IN_LEN);
2928 
2929 	if (replacing) {
2930 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
2931 			       MC_CMD_FILTER_OP_IN_OP_REPLACE);
2932 		MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE, handle);
2933 	} else {
2934 		u32 match_fields = 0;
2935 
2936 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
2937 			       efx_ef10_filter_is_exclusive(spec) ?
2938 			       MC_CMD_FILTER_OP_IN_OP_INSERT :
2939 			       MC_CMD_FILTER_OP_IN_OP_SUBSCRIBE);
2940 
2941 		/* Convert match flags and values.  Unlike almost
2942 		 * everything else in MCDI, these fields are in
2943 		 * network byte order.
2944 		 */
2945 		if (spec->match_flags & EFX_FILTER_MATCH_LOC_MAC_IG)
2946 			match_fields |=
2947 				is_multicast_ether_addr(spec->loc_mac) ?
2948 				1 << MC_CMD_FILTER_OP_IN_MATCH_UNKNOWN_MCAST_DST_LBN :
2949 				1 << MC_CMD_FILTER_OP_IN_MATCH_UNKNOWN_UCAST_DST_LBN;
2950 #define COPY_FIELD(gen_flag, gen_field, mcdi_field)			     \
2951 		if (spec->match_flags & EFX_FILTER_MATCH_ ## gen_flag) {     \
2952 			match_fields |=					     \
2953 				1 << MC_CMD_FILTER_OP_IN_MATCH_ ##	     \
2954 				mcdi_field ## _LBN;			     \
2955 			BUILD_BUG_ON(					     \
2956 				MC_CMD_FILTER_OP_IN_ ## mcdi_field ## _LEN < \
2957 				sizeof(spec->gen_field));		     \
2958 			memcpy(MCDI_PTR(inbuf, FILTER_OP_IN_ ##	mcdi_field), \
2959 			       &spec->gen_field, sizeof(spec->gen_field));   \
2960 		}
2961 		COPY_FIELD(REM_HOST, rem_host, SRC_IP);
2962 		COPY_FIELD(LOC_HOST, loc_host, DST_IP);
2963 		COPY_FIELD(REM_MAC, rem_mac, SRC_MAC);
2964 		COPY_FIELD(REM_PORT, rem_port, SRC_PORT);
2965 		COPY_FIELD(LOC_MAC, loc_mac, DST_MAC);
2966 		COPY_FIELD(LOC_PORT, loc_port, DST_PORT);
2967 		COPY_FIELD(ETHER_TYPE, ether_type, ETHER_TYPE);
2968 		COPY_FIELD(INNER_VID, inner_vid, INNER_VLAN);
2969 		COPY_FIELD(OUTER_VID, outer_vid, OUTER_VLAN);
2970 		COPY_FIELD(IP_PROTO, ip_proto, IP_PROTO);
2971 #undef COPY_FIELD
2972 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_MATCH_FIELDS,
2973 			       match_fields);
2974 	}
2975 
2976 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_PORT_ID, nic_data->vport_id);
2977 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_DEST,
2978 		       spec->dmaq_id == EFX_FILTER_RX_DMAQ_ID_DROP ?
2979 		       MC_CMD_FILTER_OP_IN_RX_DEST_DROP :
2980 		       MC_CMD_FILTER_OP_IN_RX_DEST_HOST);
2981 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_TX_DOMAIN, 0);
2982 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_TX_DEST,
2983 		       MC_CMD_FILTER_OP_IN_TX_DEST_DEFAULT);
2984 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_QUEUE,
2985 		       spec->dmaq_id == EFX_FILTER_RX_DMAQ_ID_DROP ?
2986 		       0 : spec->dmaq_id);
2987 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_MODE,
2988 		       (spec->flags & EFX_FILTER_FLAG_RX_RSS) ?
2989 		       MC_CMD_FILTER_OP_IN_RX_MODE_RSS :
2990 		       MC_CMD_FILTER_OP_IN_RX_MODE_SIMPLE);
2991 	if (spec->flags & EFX_FILTER_FLAG_RX_RSS)
2992 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_CONTEXT,
2993 			       spec->rss_context !=
2994 			       EFX_FILTER_RSS_CONTEXT_DEFAULT ?
2995 			       spec->rss_context : nic_data->rx_rss_context);
2996 }
2997 
2998 static int efx_ef10_filter_push(struct efx_nic *efx,
2999 				const struct efx_filter_spec *spec,
3000 				u64 *handle, bool replacing)
3001 {
3002 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_IN_LEN);
3003 	MCDI_DECLARE_BUF(outbuf, MC_CMD_FILTER_OP_OUT_LEN);
3004 	int rc;
3005 
3006 	efx_ef10_filter_push_prep(efx, spec, inbuf, *handle, replacing);
3007 	rc = efx_mcdi_rpc(efx, MC_CMD_FILTER_OP, inbuf, sizeof(inbuf),
3008 			  outbuf, sizeof(outbuf), NULL);
3009 	if (rc == 0)
3010 		*handle = MCDI_QWORD(outbuf, FILTER_OP_OUT_HANDLE);
3011 	if (rc == -ENOSPC)
3012 		rc = -EBUSY; /* to match efx_farch_filter_insert() */
3013 	return rc;
3014 }
3015 
3016 static int efx_ef10_filter_rx_match_pri(struct efx_ef10_filter_table *table,
3017 					enum efx_filter_match_flags match_flags)
3018 {
3019 	unsigned int match_pri;
3020 
3021 	for (match_pri = 0;
3022 	     match_pri < table->rx_match_count;
3023 	     match_pri++)
3024 		if (table->rx_match_flags[match_pri] == match_flags)
3025 			return match_pri;
3026 
3027 	return -EPROTONOSUPPORT;
3028 }
3029 
3030 static s32 efx_ef10_filter_insert(struct efx_nic *efx,
3031 				  struct efx_filter_spec *spec,
3032 				  bool replace_equal)
3033 {
3034 	struct efx_ef10_filter_table *table = efx->filter_state;
3035 	DECLARE_BITMAP(mc_rem_map, EFX_EF10_FILTER_SEARCH_LIMIT);
3036 	struct efx_filter_spec *saved_spec;
3037 	unsigned int match_pri, hash;
3038 	unsigned int priv_flags;
3039 	bool replacing = false;
3040 	int ins_index = -1;
3041 	DEFINE_WAIT(wait);
3042 	bool is_mc_recip;
3043 	s32 rc;
3044 
3045 	/* For now, only support RX filters */
3046 	if ((spec->flags & (EFX_FILTER_FLAG_RX | EFX_FILTER_FLAG_TX)) !=
3047 	    EFX_FILTER_FLAG_RX)
3048 		return -EINVAL;
3049 
3050 	rc = efx_ef10_filter_rx_match_pri(table, spec->match_flags);
3051 	if (rc < 0)
3052 		return rc;
3053 	match_pri = rc;
3054 
3055 	hash = efx_ef10_filter_hash(spec);
3056 	is_mc_recip = efx_filter_is_mc_recipient(spec);
3057 	if (is_mc_recip)
3058 		bitmap_zero(mc_rem_map, EFX_EF10_FILTER_SEARCH_LIMIT);
3059 
3060 	/* Find any existing filters with the same match tuple or
3061 	 * else a free slot to insert at.  If any of them are busy,
3062 	 * we have to wait and retry.
3063 	 */
3064 	for (;;) {
3065 		unsigned int depth = 1;
3066 		unsigned int i;
3067 
3068 		spin_lock_bh(&efx->filter_lock);
3069 
3070 		for (;;) {
3071 			i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
3072 			saved_spec = efx_ef10_filter_entry_spec(table, i);
3073 
3074 			if (!saved_spec) {
3075 				if (ins_index < 0)
3076 					ins_index = i;
3077 			} else if (efx_ef10_filter_equal(spec, saved_spec)) {
3078 				if (table->entry[i].spec &
3079 				    EFX_EF10_FILTER_FLAG_BUSY)
3080 					break;
3081 				if (spec->priority < saved_spec->priority &&
3082 				    spec->priority != EFX_FILTER_PRI_AUTO) {
3083 					rc = -EPERM;
3084 					goto out_unlock;
3085 				}
3086 				if (!is_mc_recip) {
3087 					/* This is the only one */
3088 					if (spec->priority ==
3089 					    saved_spec->priority &&
3090 					    !replace_equal) {
3091 						rc = -EEXIST;
3092 						goto out_unlock;
3093 					}
3094 					ins_index = i;
3095 					goto found;
3096 				} else if (spec->priority >
3097 					   saved_spec->priority ||
3098 					   (spec->priority ==
3099 					    saved_spec->priority &&
3100 					    replace_equal)) {
3101 					if (ins_index < 0)
3102 						ins_index = i;
3103 					else
3104 						__set_bit(depth, mc_rem_map);
3105 				}
3106 			}
3107 
3108 			/* Once we reach the maximum search depth, use
3109 			 * the first suitable slot or return -EBUSY if
3110 			 * there was none
3111 			 */
3112 			if (depth == EFX_EF10_FILTER_SEARCH_LIMIT) {
3113 				if (ins_index < 0) {
3114 					rc = -EBUSY;
3115 					goto out_unlock;
3116 				}
3117 				goto found;
3118 			}
3119 
3120 			++depth;
3121 		}
3122 
3123 		prepare_to_wait(&table->waitq, &wait, TASK_UNINTERRUPTIBLE);
3124 		spin_unlock_bh(&efx->filter_lock);
3125 		schedule();
3126 	}
3127 
3128 found:
3129 	/* Create a software table entry if necessary, and mark it
3130 	 * busy.  We might yet fail to insert, but any attempt to
3131 	 * insert a conflicting filter while we're waiting for the
3132 	 * firmware must find the busy entry.
3133 	 */
3134 	saved_spec = efx_ef10_filter_entry_spec(table, ins_index);
3135 	if (saved_spec) {
3136 		if (spec->priority == EFX_FILTER_PRI_AUTO &&
3137 		    saved_spec->priority >= EFX_FILTER_PRI_AUTO) {
3138 			/* Just make sure it won't be removed */
3139 			if (saved_spec->priority > EFX_FILTER_PRI_AUTO)
3140 				saved_spec->flags |= EFX_FILTER_FLAG_RX_OVER_AUTO;
3141 			table->entry[ins_index].spec &=
3142 				~EFX_EF10_FILTER_FLAG_AUTO_OLD;
3143 			rc = ins_index;
3144 			goto out_unlock;
3145 		}
3146 		replacing = true;
3147 		priv_flags = efx_ef10_filter_entry_flags(table, ins_index);
3148 	} else {
3149 		saved_spec = kmalloc(sizeof(*spec), GFP_ATOMIC);
3150 		if (!saved_spec) {
3151 			rc = -ENOMEM;
3152 			goto out_unlock;
3153 		}
3154 		*saved_spec = *spec;
3155 		priv_flags = 0;
3156 	}
3157 	efx_ef10_filter_set_entry(table, ins_index, saved_spec,
3158 				  priv_flags | EFX_EF10_FILTER_FLAG_BUSY);
3159 
3160 	/* Mark lower-priority multicast recipients busy prior to removal */
3161 	if (is_mc_recip) {
3162 		unsigned int depth, i;
3163 
3164 		for (depth = 0; depth < EFX_EF10_FILTER_SEARCH_LIMIT; depth++) {
3165 			i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
3166 			if (test_bit(depth, mc_rem_map))
3167 				table->entry[i].spec |=
3168 					EFX_EF10_FILTER_FLAG_BUSY;
3169 		}
3170 	}
3171 
3172 	spin_unlock_bh(&efx->filter_lock);
3173 
3174 	rc = efx_ef10_filter_push(efx, spec, &table->entry[ins_index].handle,
3175 				  replacing);
3176 
3177 	/* Finalise the software table entry */
3178 	spin_lock_bh(&efx->filter_lock);
3179 	if (rc == 0) {
3180 		if (replacing) {
3181 			/* Update the fields that may differ */
3182 			if (saved_spec->priority == EFX_FILTER_PRI_AUTO)
3183 				saved_spec->flags |=
3184 					EFX_FILTER_FLAG_RX_OVER_AUTO;
3185 			saved_spec->priority = spec->priority;
3186 			saved_spec->flags &= EFX_FILTER_FLAG_RX_OVER_AUTO;
3187 			saved_spec->flags |= spec->flags;
3188 			saved_spec->rss_context = spec->rss_context;
3189 			saved_spec->dmaq_id = spec->dmaq_id;
3190 		}
3191 	} else if (!replacing) {
3192 		kfree(saved_spec);
3193 		saved_spec = NULL;
3194 	}
3195 	efx_ef10_filter_set_entry(table, ins_index, saved_spec, priv_flags);
3196 
3197 	/* Remove and finalise entries for lower-priority multicast
3198 	 * recipients
3199 	 */
3200 	if (is_mc_recip) {
3201 		MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_IN_LEN);
3202 		unsigned int depth, i;
3203 
3204 		memset(inbuf, 0, sizeof(inbuf));
3205 
3206 		for (depth = 0; depth < EFX_EF10_FILTER_SEARCH_LIMIT; depth++) {
3207 			if (!test_bit(depth, mc_rem_map))
3208 				continue;
3209 
3210 			i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
3211 			saved_spec = efx_ef10_filter_entry_spec(table, i);
3212 			priv_flags = efx_ef10_filter_entry_flags(table, i);
3213 
3214 			if (rc == 0) {
3215 				spin_unlock_bh(&efx->filter_lock);
3216 				MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
3217 					       MC_CMD_FILTER_OP_IN_OP_UNSUBSCRIBE);
3218 				MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
3219 					       table->entry[i].handle);
3220 				rc = efx_mcdi_rpc(efx, MC_CMD_FILTER_OP,
3221 						  inbuf, sizeof(inbuf),
3222 						  NULL, 0, NULL);
3223 				spin_lock_bh(&efx->filter_lock);
3224 			}
3225 
3226 			if (rc == 0) {
3227 				kfree(saved_spec);
3228 				saved_spec = NULL;
3229 				priv_flags = 0;
3230 			} else {
3231 				priv_flags &= ~EFX_EF10_FILTER_FLAG_BUSY;
3232 			}
3233 			efx_ef10_filter_set_entry(table, i, saved_spec,
3234 						  priv_flags);
3235 		}
3236 	}
3237 
3238 	/* If successful, return the inserted filter ID */
3239 	if (rc == 0)
3240 		rc = match_pri * HUNT_FILTER_TBL_ROWS + ins_index;
3241 
3242 	wake_up_all(&table->waitq);
3243 out_unlock:
3244 	spin_unlock_bh(&efx->filter_lock);
3245 	finish_wait(&table->waitq, &wait);
3246 	return rc;
3247 }
3248 
3249 static void efx_ef10_filter_update_rx_scatter(struct efx_nic *efx)
3250 {
3251 	/* no need to do anything here on EF10 */
3252 }
3253 
3254 /* Remove a filter.
3255  * If !by_index, remove by ID
3256  * If by_index, remove by index
3257  * Filter ID may come from userland and must be range-checked.
3258  */
3259 static int efx_ef10_filter_remove_internal(struct efx_nic *efx,
3260 					   unsigned int priority_mask,
3261 					   u32 filter_id, bool by_index)
3262 {
3263 	unsigned int filter_idx = filter_id % HUNT_FILTER_TBL_ROWS;
3264 	struct efx_ef10_filter_table *table = efx->filter_state;
3265 	MCDI_DECLARE_BUF(inbuf,
3266 			 MC_CMD_FILTER_OP_IN_HANDLE_OFST +
3267 			 MC_CMD_FILTER_OP_IN_HANDLE_LEN);
3268 	struct efx_filter_spec *spec;
3269 	DEFINE_WAIT(wait);
3270 	int rc;
3271 
3272 	/* Find the software table entry and mark it busy.  Don't
3273 	 * remove it yet; any attempt to update while we're waiting
3274 	 * for the firmware must find the busy entry.
3275 	 */
3276 	for (;;) {
3277 		spin_lock_bh(&efx->filter_lock);
3278 		if (!(table->entry[filter_idx].spec &
3279 		      EFX_EF10_FILTER_FLAG_BUSY))
3280 			break;
3281 		prepare_to_wait(&table->waitq, &wait, TASK_UNINTERRUPTIBLE);
3282 		spin_unlock_bh(&efx->filter_lock);
3283 		schedule();
3284 	}
3285 
3286 	spec = efx_ef10_filter_entry_spec(table, filter_idx);
3287 	if (!spec ||
3288 	    (!by_index &&
3289 	     efx_ef10_filter_rx_match_pri(table, spec->match_flags) !=
3290 	     filter_id / HUNT_FILTER_TBL_ROWS)) {
3291 		rc = -ENOENT;
3292 		goto out_unlock;
3293 	}
3294 
3295 	if (spec->flags & EFX_FILTER_FLAG_RX_OVER_AUTO &&
3296 	    priority_mask == (1U << EFX_FILTER_PRI_AUTO)) {
3297 		/* Just remove flags */
3298 		spec->flags &= ~EFX_FILTER_FLAG_RX_OVER_AUTO;
3299 		table->entry[filter_idx].spec &= ~EFX_EF10_FILTER_FLAG_AUTO_OLD;
3300 		rc = 0;
3301 		goto out_unlock;
3302 	}
3303 
3304 	if (!(priority_mask & (1U << spec->priority))) {
3305 		rc = -ENOENT;
3306 		goto out_unlock;
3307 	}
3308 
3309 	table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_BUSY;
3310 	spin_unlock_bh(&efx->filter_lock);
3311 
3312 	if (spec->flags & EFX_FILTER_FLAG_RX_OVER_AUTO) {
3313 		/* Reset to an automatic filter */
3314 
3315 		struct efx_filter_spec new_spec = *spec;
3316 
3317 		new_spec.priority = EFX_FILTER_PRI_AUTO;
3318 		new_spec.flags = (EFX_FILTER_FLAG_RX |
3319 				  (efx_rss_enabled(efx) ?
3320 				   EFX_FILTER_FLAG_RX_RSS : 0));
3321 		new_spec.dmaq_id = 0;
3322 		new_spec.rss_context = EFX_FILTER_RSS_CONTEXT_DEFAULT;
3323 		rc = efx_ef10_filter_push(efx, &new_spec,
3324 					  &table->entry[filter_idx].handle,
3325 					  true);
3326 
3327 		spin_lock_bh(&efx->filter_lock);
3328 		if (rc == 0)
3329 			*spec = new_spec;
3330 	} else {
3331 		/* Really remove the filter */
3332 
3333 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
3334 			       efx_ef10_filter_is_exclusive(spec) ?
3335 			       MC_CMD_FILTER_OP_IN_OP_REMOVE :
3336 			       MC_CMD_FILTER_OP_IN_OP_UNSUBSCRIBE);
3337 		MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
3338 			       table->entry[filter_idx].handle);
3339 		rc = efx_mcdi_rpc(efx, MC_CMD_FILTER_OP,
3340 				  inbuf, sizeof(inbuf), NULL, 0, NULL);
3341 
3342 		spin_lock_bh(&efx->filter_lock);
3343 		if (rc == 0) {
3344 			kfree(spec);
3345 			efx_ef10_filter_set_entry(table, filter_idx, NULL, 0);
3346 		}
3347 	}
3348 
3349 	table->entry[filter_idx].spec &= ~EFX_EF10_FILTER_FLAG_BUSY;
3350 	wake_up_all(&table->waitq);
3351 out_unlock:
3352 	spin_unlock_bh(&efx->filter_lock);
3353 	finish_wait(&table->waitq, &wait);
3354 	return rc;
3355 }
3356 
3357 static int efx_ef10_filter_remove_safe(struct efx_nic *efx,
3358 				       enum efx_filter_priority priority,
3359 				       u32 filter_id)
3360 {
3361 	return efx_ef10_filter_remove_internal(efx, 1U << priority,
3362 					       filter_id, false);
3363 }
3364 
3365 static u32 efx_ef10_filter_get_unsafe_id(struct efx_nic *efx, u32 filter_id)
3366 {
3367 	return filter_id % HUNT_FILTER_TBL_ROWS;
3368 }
3369 
3370 static int efx_ef10_filter_remove_unsafe(struct efx_nic *efx,
3371 					 enum efx_filter_priority priority,
3372 					 u32 filter_id)
3373 {
3374 	return efx_ef10_filter_remove_internal(efx, 1U << priority,
3375 					       filter_id, true);
3376 }
3377 
3378 static int efx_ef10_filter_get_safe(struct efx_nic *efx,
3379 				    enum efx_filter_priority priority,
3380 				    u32 filter_id, struct efx_filter_spec *spec)
3381 {
3382 	unsigned int filter_idx = filter_id % HUNT_FILTER_TBL_ROWS;
3383 	struct efx_ef10_filter_table *table = efx->filter_state;
3384 	const struct efx_filter_spec *saved_spec;
3385 	int rc;
3386 
3387 	spin_lock_bh(&efx->filter_lock);
3388 	saved_spec = efx_ef10_filter_entry_spec(table, filter_idx);
3389 	if (saved_spec && saved_spec->priority == priority &&
3390 	    efx_ef10_filter_rx_match_pri(table, saved_spec->match_flags) ==
3391 	    filter_id / HUNT_FILTER_TBL_ROWS) {
3392 		*spec = *saved_spec;
3393 		rc = 0;
3394 	} else {
3395 		rc = -ENOENT;
3396 	}
3397 	spin_unlock_bh(&efx->filter_lock);
3398 	return rc;
3399 }
3400 
3401 static int efx_ef10_filter_clear_rx(struct efx_nic *efx,
3402 				     enum efx_filter_priority priority)
3403 {
3404 	unsigned int priority_mask;
3405 	unsigned int i;
3406 	int rc;
3407 
3408 	priority_mask = (((1U << (priority + 1)) - 1) &
3409 			 ~(1U << EFX_FILTER_PRI_AUTO));
3410 
3411 	for (i = 0; i < HUNT_FILTER_TBL_ROWS; i++) {
3412 		rc = efx_ef10_filter_remove_internal(efx, priority_mask,
3413 						     i, true);
3414 		if (rc && rc != -ENOENT)
3415 			return rc;
3416 	}
3417 
3418 	return 0;
3419 }
3420 
3421 static u32 efx_ef10_filter_count_rx_used(struct efx_nic *efx,
3422 					 enum efx_filter_priority priority)
3423 {
3424 	struct efx_ef10_filter_table *table = efx->filter_state;
3425 	unsigned int filter_idx;
3426 	s32 count = 0;
3427 
3428 	spin_lock_bh(&efx->filter_lock);
3429 	for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
3430 		if (table->entry[filter_idx].spec &&
3431 		    efx_ef10_filter_entry_spec(table, filter_idx)->priority ==
3432 		    priority)
3433 			++count;
3434 	}
3435 	spin_unlock_bh(&efx->filter_lock);
3436 	return count;
3437 }
3438 
3439 static u32 efx_ef10_filter_get_rx_id_limit(struct efx_nic *efx)
3440 {
3441 	struct efx_ef10_filter_table *table = efx->filter_state;
3442 
3443 	return table->rx_match_count * HUNT_FILTER_TBL_ROWS;
3444 }
3445 
3446 static s32 efx_ef10_filter_get_rx_ids(struct efx_nic *efx,
3447 				      enum efx_filter_priority priority,
3448 				      u32 *buf, u32 size)
3449 {
3450 	struct efx_ef10_filter_table *table = efx->filter_state;
3451 	struct efx_filter_spec *spec;
3452 	unsigned int filter_idx;
3453 	s32 count = 0;
3454 
3455 	spin_lock_bh(&efx->filter_lock);
3456 	for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
3457 		spec = efx_ef10_filter_entry_spec(table, filter_idx);
3458 		if (spec && spec->priority == priority) {
3459 			if (count == size) {
3460 				count = -EMSGSIZE;
3461 				break;
3462 			}
3463 			buf[count++] = (efx_ef10_filter_rx_match_pri(
3464 						table, spec->match_flags) *
3465 					HUNT_FILTER_TBL_ROWS +
3466 					filter_idx);
3467 		}
3468 	}
3469 	spin_unlock_bh(&efx->filter_lock);
3470 	return count;
3471 }
3472 
3473 #ifdef CONFIG_RFS_ACCEL
3474 
3475 static efx_mcdi_async_completer efx_ef10_filter_rfs_insert_complete;
3476 
3477 static s32 efx_ef10_filter_rfs_insert(struct efx_nic *efx,
3478 				      struct efx_filter_spec *spec)
3479 {
3480 	struct efx_ef10_filter_table *table = efx->filter_state;
3481 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_IN_LEN);
3482 	struct efx_filter_spec *saved_spec;
3483 	unsigned int hash, i, depth = 1;
3484 	bool replacing = false;
3485 	int ins_index = -1;
3486 	u64 cookie;
3487 	s32 rc;
3488 
3489 	/* Must be an RX filter without RSS and not for a multicast
3490 	 * destination address (RFS only works for connected sockets).
3491 	 * These restrictions allow us to pass only a tiny amount of
3492 	 * data through to the completion function.
3493 	 */
3494 	EFX_WARN_ON_PARANOID(spec->flags !=
3495 			     (EFX_FILTER_FLAG_RX | EFX_FILTER_FLAG_RX_SCATTER));
3496 	EFX_WARN_ON_PARANOID(spec->priority != EFX_FILTER_PRI_HINT);
3497 	EFX_WARN_ON_PARANOID(efx_filter_is_mc_recipient(spec));
3498 
3499 	hash = efx_ef10_filter_hash(spec);
3500 
3501 	spin_lock_bh(&efx->filter_lock);
3502 
3503 	/* Find any existing filter with the same match tuple or else
3504 	 * a free slot to insert at.  If an existing filter is busy,
3505 	 * we have to give up.
3506 	 */
3507 	for (;;) {
3508 		i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
3509 		saved_spec = efx_ef10_filter_entry_spec(table, i);
3510 
3511 		if (!saved_spec) {
3512 			if (ins_index < 0)
3513 				ins_index = i;
3514 		} else if (efx_ef10_filter_equal(spec, saved_spec)) {
3515 			if (table->entry[i].spec & EFX_EF10_FILTER_FLAG_BUSY) {
3516 				rc = -EBUSY;
3517 				goto fail_unlock;
3518 			}
3519 			if (spec->priority < saved_spec->priority) {
3520 				rc = -EPERM;
3521 				goto fail_unlock;
3522 			}
3523 			ins_index = i;
3524 			break;
3525 		}
3526 
3527 		/* Once we reach the maximum search depth, use the
3528 		 * first suitable slot or return -EBUSY if there was
3529 		 * none
3530 		 */
3531 		if (depth == EFX_EF10_FILTER_SEARCH_LIMIT) {
3532 			if (ins_index < 0) {
3533 				rc = -EBUSY;
3534 				goto fail_unlock;
3535 			}
3536 			break;
3537 		}
3538 
3539 		++depth;
3540 	}
3541 
3542 	/* Create a software table entry if necessary, and mark it
3543 	 * busy.  We might yet fail to insert, but any attempt to
3544 	 * insert a conflicting filter while we're waiting for the
3545 	 * firmware must find the busy entry.
3546 	 */
3547 	saved_spec = efx_ef10_filter_entry_spec(table, ins_index);
3548 	if (saved_spec) {
3549 		replacing = true;
3550 	} else {
3551 		saved_spec = kmalloc(sizeof(*spec), GFP_ATOMIC);
3552 		if (!saved_spec) {
3553 			rc = -ENOMEM;
3554 			goto fail_unlock;
3555 		}
3556 		*saved_spec = *spec;
3557 	}
3558 	efx_ef10_filter_set_entry(table, ins_index, saved_spec,
3559 				  EFX_EF10_FILTER_FLAG_BUSY);
3560 
3561 	spin_unlock_bh(&efx->filter_lock);
3562 
3563 	/* Pack up the variables needed on completion */
3564 	cookie = replacing << 31 | ins_index << 16 | spec->dmaq_id;
3565 
3566 	efx_ef10_filter_push_prep(efx, spec, inbuf,
3567 				  table->entry[ins_index].handle, replacing);
3568 	efx_mcdi_rpc_async(efx, MC_CMD_FILTER_OP, inbuf, sizeof(inbuf),
3569 			   MC_CMD_FILTER_OP_OUT_LEN,
3570 			   efx_ef10_filter_rfs_insert_complete, cookie);
3571 
3572 	return ins_index;
3573 
3574 fail_unlock:
3575 	spin_unlock_bh(&efx->filter_lock);
3576 	return rc;
3577 }
3578 
3579 static void
3580 efx_ef10_filter_rfs_insert_complete(struct efx_nic *efx, unsigned long cookie,
3581 				    int rc, efx_dword_t *outbuf,
3582 				    size_t outlen_actual)
3583 {
3584 	struct efx_ef10_filter_table *table = efx->filter_state;
3585 	unsigned int ins_index, dmaq_id;
3586 	struct efx_filter_spec *spec;
3587 	bool replacing;
3588 
3589 	/* Unpack the cookie */
3590 	replacing = cookie >> 31;
3591 	ins_index = (cookie >> 16) & (HUNT_FILTER_TBL_ROWS - 1);
3592 	dmaq_id = cookie & 0xffff;
3593 
3594 	spin_lock_bh(&efx->filter_lock);
3595 	spec = efx_ef10_filter_entry_spec(table, ins_index);
3596 	if (rc == 0) {
3597 		table->entry[ins_index].handle =
3598 			MCDI_QWORD(outbuf, FILTER_OP_OUT_HANDLE);
3599 		if (replacing)
3600 			spec->dmaq_id = dmaq_id;
3601 	} else if (!replacing) {
3602 		kfree(spec);
3603 		spec = NULL;
3604 	}
3605 	efx_ef10_filter_set_entry(table, ins_index, spec, 0);
3606 	spin_unlock_bh(&efx->filter_lock);
3607 
3608 	wake_up_all(&table->waitq);
3609 }
3610 
3611 static void
3612 efx_ef10_filter_rfs_expire_complete(struct efx_nic *efx,
3613 				    unsigned long filter_idx,
3614 				    int rc, efx_dword_t *outbuf,
3615 				    size_t outlen_actual);
3616 
3617 static bool efx_ef10_filter_rfs_expire_one(struct efx_nic *efx, u32 flow_id,
3618 					   unsigned int filter_idx)
3619 {
3620 	struct efx_ef10_filter_table *table = efx->filter_state;
3621 	struct efx_filter_spec *spec =
3622 		efx_ef10_filter_entry_spec(table, filter_idx);
3623 	MCDI_DECLARE_BUF(inbuf,
3624 			 MC_CMD_FILTER_OP_IN_HANDLE_OFST +
3625 			 MC_CMD_FILTER_OP_IN_HANDLE_LEN);
3626 
3627 	if (!spec ||
3628 	    (table->entry[filter_idx].spec & EFX_EF10_FILTER_FLAG_BUSY) ||
3629 	    spec->priority != EFX_FILTER_PRI_HINT ||
3630 	    !rps_may_expire_flow(efx->net_dev, spec->dmaq_id,
3631 				 flow_id, filter_idx))
3632 		return false;
3633 
3634 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
3635 		       MC_CMD_FILTER_OP_IN_OP_REMOVE);
3636 	MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
3637 		       table->entry[filter_idx].handle);
3638 	if (efx_mcdi_rpc_async(efx, MC_CMD_FILTER_OP, inbuf, sizeof(inbuf), 0,
3639 			       efx_ef10_filter_rfs_expire_complete, filter_idx))
3640 		return false;
3641 
3642 	table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_BUSY;
3643 	return true;
3644 }
3645 
3646 static void
3647 efx_ef10_filter_rfs_expire_complete(struct efx_nic *efx,
3648 				    unsigned long filter_idx,
3649 				    int rc, efx_dword_t *outbuf,
3650 				    size_t outlen_actual)
3651 {
3652 	struct efx_ef10_filter_table *table = efx->filter_state;
3653 	struct efx_filter_spec *spec =
3654 		efx_ef10_filter_entry_spec(table, filter_idx);
3655 
3656 	spin_lock_bh(&efx->filter_lock);
3657 	if (rc == 0) {
3658 		kfree(spec);
3659 		efx_ef10_filter_set_entry(table, filter_idx, NULL, 0);
3660 	}
3661 	table->entry[filter_idx].spec &= ~EFX_EF10_FILTER_FLAG_BUSY;
3662 	wake_up_all(&table->waitq);
3663 	spin_unlock_bh(&efx->filter_lock);
3664 }
3665 
3666 #endif /* CONFIG_RFS_ACCEL */
3667 
3668 static int efx_ef10_filter_match_flags_from_mcdi(u32 mcdi_flags)
3669 {
3670 	int match_flags = 0;
3671 
3672 #define MAP_FLAG(gen_flag, mcdi_field) {				\
3673 		u32 old_mcdi_flags = mcdi_flags;			\
3674 		mcdi_flags &= ~(1 << MC_CMD_FILTER_OP_IN_MATCH_ ##	\
3675 				mcdi_field ## _LBN);			\
3676 		if (mcdi_flags != old_mcdi_flags)			\
3677 			match_flags |= EFX_FILTER_MATCH_ ## gen_flag;	\
3678 	}
3679 	MAP_FLAG(LOC_MAC_IG, UNKNOWN_UCAST_DST);
3680 	MAP_FLAG(LOC_MAC_IG, UNKNOWN_MCAST_DST);
3681 	MAP_FLAG(REM_HOST, SRC_IP);
3682 	MAP_FLAG(LOC_HOST, DST_IP);
3683 	MAP_FLAG(REM_MAC, SRC_MAC);
3684 	MAP_FLAG(REM_PORT, SRC_PORT);
3685 	MAP_FLAG(LOC_MAC, DST_MAC);
3686 	MAP_FLAG(LOC_PORT, DST_PORT);
3687 	MAP_FLAG(ETHER_TYPE, ETHER_TYPE);
3688 	MAP_FLAG(INNER_VID, INNER_VLAN);
3689 	MAP_FLAG(OUTER_VID, OUTER_VLAN);
3690 	MAP_FLAG(IP_PROTO, IP_PROTO);
3691 #undef MAP_FLAG
3692 
3693 	/* Did we map them all? */
3694 	if (mcdi_flags)
3695 		return -EINVAL;
3696 
3697 	return match_flags;
3698 }
3699 
3700 static int efx_ef10_filter_table_probe(struct efx_nic *efx)
3701 {
3702 	MCDI_DECLARE_BUF(inbuf, MC_CMD_GET_PARSER_DISP_INFO_IN_LEN);
3703 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_PARSER_DISP_INFO_OUT_LENMAX);
3704 	unsigned int pd_match_pri, pd_match_count;
3705 	struct efx_ef10_filter_table *table;
3706 	size_t outlen;
3707 	int rc;
3708 
3709 	table = kzalloc(sizeof(*table), GFP_KERNEL);
3710 	if (!table)
3711 		return -ENOMEM;
3712 
3713 	/* Find out which RX filter types are supported, and their priorities */
3714 	MCDI_SET_DWORD(inbuf, GET_PARSER_DISP_INFO_IN_OP,
3715 		       MC_CMD_GET_PARSER_DISP_INFO_IN_OP_GET_SUPPORTED_RX_MATCHES);
3716 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_PARSER_DISP_INFO,
3717 			  inbuf, sizeof(inbuf), outbuf, sizeof(outbuf),
3718 			  &outlen);
3719 	if (rc)
3720 		goto fail;
3721 	pd_match_count = MCDI_VAR_ARRAY_LEN(
3722 		outlen, GET_PARSER_DISP_INFO_OUT_SUPPORTED_MATCHES);
3723 	table->rx_match_count = 0;
3724 
3725 	for (pd_match_pri = 0; pd_match_pri < pd_match_count; pd_match_pri++) {
3726 		u32 mcdi_flags =
3727 			MCDI_ARRAY_DWORD(
3728 				outbuf,
3729 				GET_PARSER_DISP_INFO_OUT_SUPPORTED_MATCHES,
3730 				pd_match_pri);
3731 		rc = efx_ef10_filter_match_flags_from_mcdi(mcdi_flags);
3732 		if (rc < 0) {
3733 			netif_dbg(efx, probe, efx->net_dev,
3734 				  "%s: fw flags %#x pri %u not supported in driver\n",
3735 				  __func__, mcdi_flags, pd_match_pri);
3736 		} else {
3737 			netif_dbg(efx, probe, efx->net_dev,
3738 				  "%s: fw flags %#x pri %u supported as driver flags %#x pri %u\n",
3739 				  __func__, mcdi_flags, pd_match_pri,
3740 				  rc, table->rx_match_count);
3741 			table->rx_match_flags[table->rx_match_count++] = rc;
3742 		}
3743 	}
3744 
3745 	table->entry = vzalloc(HUNT_FILTER_TBL_ROWS * sizeof(*table->entry));
3746 	if (!table->entry) {
3747 		rc = -ENOMEM;
3748 		goto fail;
3749 	}
3750 
3751 	table->ucdef_id = EFX_EF10_FILTER_ID_INVALID;
3752 	table->bcast_id = EFX_EF10_FILTER_ID_INVALID;
3753 	table->mcdef_id = EFX_EF10_FILTER_ID_INVALID;
3754 
3755 	efx->filter_state = table;
3756 	init_waitqueue_head(&table->waitq);
3757 	return 0;
3758 
3759 fail:
3760 	kfree(table);
3761 	return rc;
3762 }
3763 
3764 /* Caller must hold efx->filter_sem for read if race against
3765  * efx_ef10_filter_table_remove() is possible
3766  */
3767 static void efx_ef10_filter_table_restore(struct efx_nic *efx)
3768 {
3769 	struct efx_ef10_filter_table *table = efx->filter_state;
3770 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
3771 	struct efx_filter_spec *spec;
3772 	unsigned int filter_idx;
3773 	bool failed = false;
3774 	int rc;
3775 
3776 	WARN_ON(!rwsem_is_locked(&efx->filter_sem));
3777 
3778 	if (!nic_data->must_restore_filters)
3779 		return;
3780 
3781 	if (!table)
3782 		return;
3783 
3784 	spin_lock_bh(&efx->filter_lock);
3785 
3786 	for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
3787 		spec = efx_ef10_filter_entry_spec(table, filter_idx);
3788 		if (!spec)
3789 			continue;
3790 
3791 		table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_BUSY;
3792 		spin_unlock_bh(&efx->filter_lock);
3793 
3794 		rc = efx_ef10_filter_push(efx, spec,
3795 					  &table->entry[filter_idx].handle,
3796 					  false);
3797 		if (rc)
3798 			failed = true;
3799 
3800 		spin_lock_bh(&efx->filter_lock);
3801 		if (rc) {
3802 			kfree(spec);
3803 			efx_ef10_filter_set_entry(table, filter_idx, NULL, 0);
3804 		} else {
3805 			table->entry[filter_idx].spec &=
3806 				~EFX_EF10_FILTER_FLAG_BUSY;
3807 		}
3808 	}
3809 
3810 	spin_unlock_bh(&efx->filter_lock);
3811 
3812 	if (failed)
3813 		netif_err(efx, hw, efx->net_dev,
3814 			  "unable to restore all filters\n");
3815 	else
3816 		nic_data->must_restore_filters = false;
3817 }
3818 
3819 /* Caller must hold efx->filter_sem for write */
3820 static void efx_ef10_filter_table_remove(struct efx_nic *efx)
3821 {
3822 	struct efx_ef10_filter_table *table = efx->filter_state;
3823 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_IN_LEN);
3824 	struct efx_filter_spec *spec;
3825 	unsigned int filter_idx;
3826 	int rc;
3827 
3828 	efx->filter_state = NULL;
3829 	if (!table)
3830 		return;
3831 
3832 	for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
3833 		spec = efx_ef10_filter_entry_spec(table, filter_idx);
3834 		if (!spec)
3835 			continue;
3836 
3837 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
3838 			       efx_ef10_filter_is_exclusive(spec) ?
3839 			       MC_CMD_FILTER_OP_IN_OP_REMOVE :
3840 			       MC_CMD_FILTER_OP_IN_OP_UNSUBSCRIBE);
3841 		MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
3842 			       table->entry[filter_idx].handle);
3843 		rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FILTER_OP, inbuf,
3844 					sizeof(inbuf), NULL, 0, NULL);
3845 		if (rc)
3846 			netif_info(efx, drv, efx->net_dev,
3847 				   "%s: filter %04x remove failed\n",
3848 				   __func__, filter_idx);
3849 		kfree(spec);
3850 	}
3851 
3852 	vfree(table->entry);
3853 	kfree(table);
3854 }
3855 
3856 #define EFX_EF10_FILTER_DO_MARK_OLD(id) \
3857 	if (id != EFX_EF10_FILTER_ID_INVALID) { \
3858 		filter_idx = efx_ef10_filter_get_unsafe_id(efx, id); \
3859 		if (!table->entry[filter_idx].spec) \
3860 			netif_dbg(efx, drv, efx->net_dev, \
3861 				  "%s: marked null spec old %04x:%04x\n", \
3862 				  __func__, id, filter_idx); \
3863 		table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_AUTO_OLD;\
3864 	}
3865 static void efx_ef10_filter_mark_old(struct efx_nic *efx)
3866 {
3867 	struct efx_ef10_filter_table *table = efx->filter_state;
3868 	unsigned int filter_idx, i;
3869 
3870 	if (!table)
3871 		return;
3872 
3873 	/* Mark old filters that may need to be removed */
3874 	spin_lock_bh(&efx->filter_lock);
3875 	for (i = 0; i < table->dev_uc_count; i++)
3876 		EFX_EF10_FILTER_DO_MARK_OLD(table->dev_uc_list[i].id);
3877 	for (i = 0; i < table->dev_mc_count; i++)
3878 		EFX_EF10_FILTER_DO_MARK_OLD(table->dev_mc_list[i].id);
3879 	EFX_EF10_FILTER_DO_MARK_OLD(table->ucdef_id);
3880 	EFX_EF10_FILTER_DO_MARK_OLD(table->bcast_id);
3881 	EFX_EF10_FILTER_DO_MARK_OLD(table->mcdef_id);
3882 	spin_unlock_bh(&efx->filter_lock);
3883 }
3884 #undef EFX_EF10_FILTER_DO_MARK_OLD
3885 
3886 static void efx_ef10_filter_uc_addr_list(struct efx_nic *efx, bool *promisc)
3887 {
3888 	struct efx_ef10_filter_table *table = efx->filter_state;
3889 	struct net_device *net_dev = efx->net_dev;
3890 	struct netdev_hw_addr *uc;
3891 	int addr_count;
3892 	unsigned int i;
3893 
3894 	table->ucdef_id = EFX_EF10_FILTER_ID_INVALID;
3895 	addr_count = netdev_uc_count(net_dev);
3896 	if (net_dev->flags & IFF_PROMISC)
3897 		*promisc = true;
3898 	table->dev_uc_count = 1 + addr_count;
3899 	ether_addr_copy(table->dev_uc_list[0].addr, net_dev->dev_addr);
3900 	i = 1;
3901 	netdev_for_each_uc_addr(uc, net_dev) {
3902 		if (i >= EFX_EF10_FILTER_DEV_UC_MAX) {
3903 			*promisc = true;
3904 			break;
3905 		}
3906 		ether_addr_copy(table->dev_uc_list[i].addr, uc->addr);
3907 		table->dev_uc_list[i].id = EFX_EF10_FILTER_ID_INVALID;
3908 		i++;
3909 	}
3910 }
3911 
3912 static void efx_ef10_filter_mc_addr_list(struct efx_nic *efx, bool *promisc)
3913 {
3914 	struct efx_ef10_filter_table *table = efx->filter_state;
3915 	struct net_device *net_dev = efx->net_dev;
3916 	struct netdev_hw_addr *mc;
3917 	unsigned int i, addr_count;
3918 
3919 	table->mcdef_id = EFX_EF10_FILTER_ID_INVALID;
3920 	table->bcast_id = EFX_EF10_FILTER_ID_INVALID;
3921 	if (net_dev->flags & (IFF_PROMISC | IFF_ALLMULTI))
3922 		*promisc = true;
3923 
3924 	addr_count = netdev_mc_count(net_dev);
3925 	i = 0;
3926 	netdev_for_each_mc_addr(mc, net_dev) {
3927 		if (i >= EFX_EF10_FILTER_DEV_MC_MAX) {
3928 			*promisc = true;
3929 			break;
3930 		}
3931 		ether_addr_copy(table->dev_mc_list[i].addr, mc->addr);
3932 		table->dev_mc_list[i].id = EFX_EF10_FILTER_ID_INVALID;
3933 		i++;
3934 	}
3935 
3936 	table->dev_mc_count = i;
3937 }
3938 
3939 static int efx_ef10_filter_insert_addr_list(struct efx_nic *efx,
3940 					     bool multicast, bool rollback)
3941 {
3942 	struct efx_ef10_filter_table *table = efx->filter_state;
3943 	struct efx_ef10_dev_addr *addr_list;
3944 	enum efx_filter_flags filter_flags;
3945 	struct efx_filter_spec spec;
3946 	u8 baddr[ETH_ALEN];
3947 	unsigned int i, j;
3948 	int addr_count;
3949 	int rc;
3950 
3951 	if (multicast) {
3952 		addr_list = table->dev_mc_list;
3953 		addr_count = table->dev_mc_count;
3954 	} else {
3955 		addr_list = table->dev_uc_list;
3956 		addr_count = table->dev_uc_count;
3957 	}
3958 
3959 	filter_flags = efx_rss_enabled(efx) ? EFX_FILTER_FLAG_RX_RSS : 0;
3960 
3961 	/* Insert/renew filters */
3962 	for (i = 0; i < addr_count; i++) {
3963 		efx_filter_init_rx(&spec, EFX_FILTER_PRI_AUTO, filter_flags, 0);
3964 		efx_filter_set_eth_local(&spec, EFX_FILTER_VID_UNSPEC,
3965 					 addr_list[i].addr);
3966 		rc = efx_ef10_filter_insert(efx, &spec, true);
3967 		if (rc < 0) {
3968 			if (rollback) {
3969 				netif_info(efx, drv, efx->net_dev,
3970 					   "efx_ef10_filter_insert failed rc=%d\n",
3971 					   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 				/* mark as not inserted, and carry on */
3984 				rc = EFX_EF10_FILTER_ID_INVALID;
3985 			}
3986 		}
3987 		addr_list[i].id = efx_ef10_filter_get_unsafe_id(efx, rc);
3988 	}
3989 
3990 	if (multicast && rollback) {
3991 		/* Also need an Ethernet broadcast filter */
3992 		efx_filter_init_rx(&spec, EFX_FILTER_PRI_AUTO, filter_flags, 0);
3993 		eth_broadcast_addr(baddr);
3994 		efx_filter_set_eth_local(&spec, EFX_FILTER_VID_UNSPEC, baddr);
3995 		rc = efx_ef10_filter_insert(efx, &spec, true);
3996 		if (rc < 0) {
3997 			netif_warn(efx, drv, efx->net_dev,
3998 				   "Broadcast filter insert failed rc=%d\n", rc);
3999 			/* Fall back to promiscuous */
4000 			for (j = 0; j < i; j++) {
4001 				if (addr_list[j].id == EFX_EF10_FILTER_ID_INVALID)
4002 					continue;
4003 				efx_ef10_filter_remove_unsafe(
4004 					efx, EFX_FILTER_PRI_AUTO,
4005 					addr_list[j].id);
4006 				addr_list[j].id = EFX_EF10_FILTER_ID_INVALID;
4007 			}
4008 			return rc;
4009 		} else {
4010 			table->bcast_id = efx_ef10_filter_get_unsafe_id(efx, rc);
4011 		}
4012 	}
4013 
4014 	return 0;
4015 }
4016 
4017 static int efx_ef10_filter_insert_def(struct efx_nic *efx, bool multicast,
4018 				      bool rollback)
4019 {
4020 	struct efx_ef10_filter_table *table = efx->filter_state;
4021 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
4022 	enum efx_filter_flags filter_flags;
4023 	struct efx_filter_spec spec;
4024 	u8 baddr[ETH_ALEN];
4025 	int rc;
4026 
4027 	filter_flags = efx_rss_enabled(efx) ? EFX_FILTER_FLAG_RX_RSS : 0;
4028 
4029 	efx_filter_init_rx(&spec, EFX_FILTER_PRI_AUTO, filter_flags, 0);
4030 
4031 	if (multicast)
4032 		efx_filter_set_mc_def(&spec);
4033 	else
4034 		efx_filter_set_uc_def(&spec);
4035 
4036 	rc = efx_ef10_filter_insert(efx, &spec, true);
4037 	if (rc < 0) {
4038 		netif_printk(efx, drv, rc == -EPERM ? KERN_DEBUG : KERN_WARNING,
4039 			     efx->net_dev,
4040 			     "%scast mismatch filter insert failed rc=%d\n",
4041 			     multicast ? "Multi" : "Uni", rc);
4042 	} else if (multicast) {
4043 		table->mcdef_id = efx_ef10_filter_get_unsafe_id(efx, rc);
4044 		if (!nic_data->workaround_26807) {
4045 			/* Also need an Ethernet broadcast filter */
4046 			efx_filter_init_rx(&spec, EFX_FILTER_PRI_AUTO,
4047 					   filter_flags, 0);
4048 			eth_broadcast_addr(baddr);
4049 			efx_filter_set_eth_local(&spec, EFX_FILTER_VID_UNSPEC,
4050 						 baddr);
4051 			rc = efx_ef10_filter_insert(efx, &spec, true);
4052 			if (rc < 0) {
4053 				netif_warn(efx, drv, efx->net_dev,
4054 					   "Broadcast filter insert failed rc=%d\n",
4055 					   rc);
4056 				if (rollback) {
4057 					/* Roll back the mc_def filter */
4058 					efx_ef10_filter_remove_unsafe(
4059 							efx, EFX_FILTER_PRI_AUTO,
4060 							table->mcdef_id);
4061 					table->mcdef_id = EFX_EF10_FILTER_ID_INVALID;
4062 					return rc;
4063 				}
4064 			} else {
4065 				table->bcast_id = efx_ef10_filter_get_unsafe_id(efx, rc);
4066 			}
4067 		}
4068 		rc = 0;
4069 	} else {
4070 		table->ucdef_id = rc;
4071 		rc = 0;
4072 	}
4073 	return rc;
4074 }
4075 
4076 /* Remove filters that weren't renewed.  Since nothing else changes the AUTO_OLD
4077  * flag or removes these filters, we don't need to hold the filter_lock while
4078  * scanning for these filters.
4079  */
4080 static void efx_ef10_filter_remove_old(struct efx_nic *efx)
4081 {
4082 	struct efx_ef10_filter_table *table = efx->filter_state;
4083 	int remove_failed = 0;
4084 	int remove_noent = 0;
4085 	int rc;
4086 	int i;
4087 
4088 	for (i = 0; i < HUNT_FILTER_TBL_ROWS; i++) {
4089 		if (ACCESS_ONCE(table->entry[i].spec) &
4090 		    EFX_EF10_FILTER_FLAG_AUTO_OLD) {
4091 			rc = efx_ef10_filter_remove_internal(efx,
4092 					1U << EFX_FILTER_PRI_AUTO, i, true);
4093 			if (rc == -ENOENT)
4094 				remove_noent++;
4095 			else if (rc)
4096 				remove_failed++;
4097 		}
4098 	}
4099 
4100 	if (remove_failed)
4101 		netif_info(efx, drv, efx->net_dev,
4102 			   "%s: failed to remove %d filters\n",
4103 			   __func__, remove_failed);
4104 	if (remove_noent)
4105 		netif_info(efx, drv, efx->net_dev,
4106 			   "%s: failed to remove %d non-existent filters\n",
4107 			   __func__, remove_noent);
4108 }
4109 
4110 static int efx_ef10_vport_set_mac_address(struct efx_nic *efx)
4111 {
4112 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
4113 	u8 mac_old[ETH_ALEN];
4114 	int rc, rc2;
4115 
4116 	/* Only reconfigure a PF-created vport */
4117 	if (is_zero_ether_addr(nic_data->vport_mac))
4118 		return 0;
4119 
4120 	efx_device_detach_sync(efx);
4121 	efx_net_stop(efx->net_dev);
4122 	down_write(&efx->filter_sem);
4123 	efx_ef10_filter_table_remove(efx);
4124 	up_write(&efx->filter_sem);
4125 
4126 	rc = efx_ef10_vadaptor_free(efx, nic_data->vport_id);
4127 	if (rc)
4128 		goto restore_filters;
4129 
4130 	ether_addr_copy(mac_old, nic_data->vport_mac);
4131 	rc = efx_ef10_vport_del_mac(efx, nic_data->vport_id,
4132 				    nic_data->vport_mac);
4133 	if (rc)
4134 		goto restore_vadaptor;
4135 
4136 	rc = efx_ef10_vport_add_mac(efx, nic_data->vport_id,
4137 				    efx->net_dev->dev_addr);
4138 	if (!rc) {
4139 		ether_addr_copy(nic_data->vport_mac, efx->net_dev->dev_addr);
4140 	} else {
4141 		rc2 = efx_ef10_vport_add_mac(efx, nic_data->vport_id, mac_old);
4142 		if (rc2) {
4143 			/* Failed to add original MAC, so clear vport_mac */
4144 			eth_zero_addr(nic_data->vport_mac);
4145 			goto reset_nic;
4146 		}
4147 	}
4148 
4149 restore_vadaptor:
4150 	rc2 = efx_ef10_vadaptor_alloc(efx, nic_data->vport_id);
4151 	if (rc2)
4152 		goto reset_nic;
4153 restore_filters:
4154 	down_write(&efx->filter_sem);
4155 	rc2 = efx_ef10_filter_table_probe(efx);
4156 	up_write(&efx->filter_sem);
4157 	if (rc2)
4158 		goto reset_nic;
4159 
4160 	rc2 = efx_net_open(efx->net_dev);
4161 	if (rc2)
4162 		goto reset_nic;
4163 
4164 	netif_device_attach(efx->net_dev);
4165 
4166 	return rc;
4167 
4168 reset_nic:
4169 	netif_err(efx, drv, efx->net_dev,
4170 		  "Failed to restore when changing MAC address - scheduling reset\n");
4171 	efx_schedule_reset(efx, RESET_TYPE_DATAPATH);
4172 
4173 	return rc ? rc : rc2;
4174 }
4175 
4176 /* Caller must hold efx->filter_sem for read if race against
4177  * efx_ef10_filter_table_remove() is possible
4178  */
4179 static void efx_ef10_filter_sync_rx_mode(struct efx_nic *efx)
4180 {
4181 	struct efx_ef10_filter_table *table = efx->filter_state;
4182 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
4183 	struct net_device *net_dev = efx->net_dev;
4184 	bool uc_promisc = false, mc_promisc = false;
4185 
4186 	if (!efx_dev_registered(efx))
4187 		return;
4188 
4189 	if (!table)
4190 		return;
4191 
4192 	efx_ef10_filter_mark_old(efx);
4193 
4194 	/* Copy/convert the address lists; add the primary station
4195 	 * address and broadcast address
4196 	 */
4197 	netif_addr_lock_bh(net_dev);
4198 	efx_ef10_filter_uc_addr_list(efx, &uc_promisc);
4199 	efx_ef10_filter_mc_addr_list(efx, &mc_promisc);
4200 	netif_addr_unlock_bh(net_dev);
4201 
4202 	/* Insert/renew unicast filters */
4203 	if (uc_promisc) {
4204 		efx_ef10_filter_insert_def(efx, false, false);
4205 		efx_ef10_filter_insert_addr_list(efx, false, false);
4206 	} else {
4207 		/* If any of the filters failed to insert, fall back to
4208 		 * promiscuous mode - add in the uc_def filter.  But keep
4209 		 * our individual unicast filters.
4210 		 */
4211 		if (efx_ef10_filter_insert_addr_list(efx, false, false))
4212 			efx_ef10_filter_insert_def(efx, false, false);
4213 	}
4214 
4215 	/* Insert/renew multicast filters */
4216 	/* If changing promiscuous state with cascaded multicast filters, remove
4217 	 * old filters first, so that packets are dropped rather than duplicated
4218 	 */
4219 	if (nic_data->workaround_26807 && efx->mc_promisc != mc_promisc)
4220 		efx_ef10_filter_remove_old(efx);
4221 	if (mc_promisc) {
4222 		if (nic_data->workaround_26807) {
4223 			/* If we failed to insert promiscuous filters, rollback
4224 			 * and fall back to individual multicast filters
4225 			 */
4226 			if (efx_ef10_filter_insert_def(efx, true, true)) {
4227 				/* Changing promisc state, so remove old filters */
4228 				efx_ef10_filter_remove_old(efx);
4229 				efx_ef10_filter_insert_addr_list(efx, true, false);
4230 			}
4231 		} else {
4232 			/* If we failed to insert promiscuous filters, don't
4233 			 * rollback.  Regardless, also insert the mc_list
4234 			 */
4235 			efx_ef10_filter_insert_def(efx, true, false);
4236 			efx_ef10_filter_insert_addr_list(efx, true, false);
4237 		}
4238 	} else {
4239 		/* If any filters failed to insert, rollback and fall back to
4240 		 * promiscuous mode - mc_def filter and maybe broadcast.  If
4241 		 * that fails, roll back again and insert as many of our
4242 		 * individual multicast filters as we can.
4243 		 */
4244 		if (efx_ef10_filter_insert_addr_list(efx, true, true)) {
4245 			/* Changing promisc state, so remove old filters */
4246 			if (nic_data->workaround_26807)
4247 				efx_ef10_filter_remove_old(efx);
4248 			if (efx_ef10_filter_insert_def(efx, true, true))
4249 				efx_ef10_filter_insert_addr_list(efx, true, false);
4250 		}
4251 	}
4252 
4253 	efx_ef10_filter_remove_old(efx);
4254 	efx->mc_promisc = mc_promisc;
4255 }
4256 
4257 static int efx_ef10_set_mac_address(struct efx_nic *efx)
4258 {
4259 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VADAPTOR_SET_MAC_IN_LEN);
4260 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
4261 	bool was_enabled = efx->port_enabled;
4262 	int rc;
4263 
4264 	efx_device_detach_sync(efx);
4265 	efx_net_stop(efx->net_dev);
4266 	down_write(&efx->filter_sem);
4267 	efx_ef10_filter_table_remove(efx);
4268 
4269 	ether_addr_copy(MCDI_PTR(inbuf, VADAPTOR_SET_MAC_IN_MACADDR),
4270 			efx->net_dev->dev_addr);
4271 	MCDI_SET_DWORD(inbuf, VADAPTOR_SET_MAC_IN_UPSTREAM_PORT_ID,
4272 		       nic_data->vport_id);
4273 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_VADAPTOR_SET_MAC, inbuf,
4274 				sizeof(inbuf), NULL, 0, NULL);
4275 
4276 	efx_ef10_filter_table_probe(efx);
4277 	up_write(&efx->filter_sem);
4278 	if (was_enabled)
4279 		efx_net_open(efx->net_dev);
4280 	netif_device_attach(efx->net_dev);
4281 
4282 #ifdef CONFIG_SFC_SRIOV
4283 	if (efx->pci_dev->is_virtfn && efx->pci_dev->physfn) {
4284 		struct pci_dev *pci_dev_pf = efx->pci_dev->physfn;
4285 
4286 		if (rc == -EPERM) {
4287 			struct efx_nic *efx_pf;
4288 
4289 			/* Switch to PF and change MAC address on vport */
4290 			efx_pf = pci_get_drvdata(pci_dev_pf);
4291 
4292 			rc = efx_ef10_sriov_set_vf_mac(efx_pf,
4293 						       nic_data->vf_index,
4294 						       efx->net_dev->dev_addr);
4295 		} else if (!rc) {
4296 			struct efx_nic *efx_pf = pci_get_drvdata(pci_dev_pf);
4297 			struct efx_ef10_nic_data *nic_data = efx_pf->nic_data;
4298 			unsigned int i;
4299 
4300 			/* MAC address successfully changed by VF (with MAC
4301 			 * spoofing) so update the parent PF if possible.
4302 			 */
4303 			for (i = 0; i < efx_pf->vf_count; ++i) {
4304 				struct ef10_vf *vf = nic_data->vf + i;
4305 
4306 				if (vf->efx == efx) {
4307 					ether_addr_copy(vf->mac,
4308 							efx->net_dev->dev_addr);
4309 					return 0;
4310 				}
4311 			}
4312 		}
4313 	} else
4314 #endif
4315 	if (rc == -EPERM) {
4316 		netif_err(efx, drv, efx->net_dev,
4317 			  "Cannot change MAC address; use sfboot to enable"
4318 			  " mac-spoofing on this interface\n");
4319 	} else if (rc == -ENOSYS && !efx_ef10_is_vf(efx)) {
4320 		/* If the active MCFW does not support MC_CMD_VADAPTOR_SET_MAC
4321 		 * fall-back to the method of changing the MAC address on the
4322 		 * vport.  This only applies to PFs because such versions of
4323 		 * MCFW do not support VFs.
4324 		 */
4325 		rc = efx_ef10_vport_set_mac_address(efx);
4326 	} else {
4327 		efx_mcdi_display_error(efx, MC_CMD_VADAPTOR_SET_MAC,
4328 				       sizeof(inbuf), NULL, 0, rc);
4329 	}
4330 
4331 	return rc;
4332 }
4333 
4334 static int efx_ef10_mac_reconfigure(struct efx_nic *efx)
4335 {
4336 	efx_ef10_filter_sync_rx_mode(efx);
4337 
4338 	return efx_mcdi_set_mac(efx);
4339 }
4340 
4341 static int efx_ef10_mac_reconfigure_vf(struct efx_nic *efx)
4342 {
4343 	efx_ef10_filter_sync_rx_mode(efx);
4344 
4345 	return 0;
4346 }
4347 
4348 static int efx_ef10_start_bist(struct efx_nic *efx, u32 bist_type)
4349 {
4350 	MCDI_DECLARE_BUF(inbuf, MC_CMD_START_BIST_IN_LEN);
4351 
4352 	MCDI_SET_DWORD(inbuf, START_BIST_IN_TYPE, bist_type);
4353 	return efx_mcdi_rpc(efx, MC_CMD_START_BIST, inbuf, sizeof(inbuf),
4354 			    NULL, 0, NULL);
4355 }
4356 
4357 /* MC BISTs follow a different poll mechanism to phy BISTs.
4358  * The BIST is done in the poll handler on the MC, and the MCDI command
4359  * will block until the BIST is done.
4360  */
4361 static int efx_ef10_poll_bist(struct efx_nic *efx)
4362 {
4363 	int rc;
4364 	MCDI_DECLARE_BUF(outbuf, MC_CMD_POLL_BIST_OUT_LEN);
4365 	size_t outlen;
4366 	u32 result;
4367 
4368 	rc = efx_mcdi_rpc(efx, MC_CMD_POLL_BIST, NULL, 0,
4369 			   outbuf, sizeof(outbuf), &outlen);
4370 	if (rc != 0)
4371 		return rc;
4372 
4373 	if (outlen < MC_CMD_POLL_BIST_OUT_LEN)
4374 		return -EIO;
4375 
4376 	result = MCDI_DWORD(outbuf, POLL_BIST_OUT_RESULT);
4377 	switch (result) {
4378 	case MC_CMD_POLL_BIST_PASSED:
4379 		netif_dbg(efx, hw, efx->net_dev, "BIST passed.\n");
4380 		return 0;
4381 	case MC_CMD_POLL_BIST_TIMEOUT:
4382 		netif_err(efx, hw, efx->net_dev, "BIST timed out\n");
4383 		return -EIO;
4384 	case MC_CMD_POLL_BIST_FAILED:
4385 		netif_err(efx, hw, efx->net_dev, "BIST failed.\n");
4386 		return -EIO;
4387 	default:
4388 		netif_err(efx, hw, efx->net_dev,
4389 			  "BIST returned unknown result %u", result);
4390 		return -EIO;
4391 	}
4392 }
4393 
4394 static int efx_ef10_run_bist(struct efx_nic *efx, u32 bist_type)
4395 {
4396 	int rc;
4397 
4398 	netif_dbg(efx, drv, efx->net_dev, "starting BIST type %u\n", bist_type);
4399 
4400 	rc = efx_ef10_start_bist(efx, bist_type);
4401 	if (rc != 0)
4402 		return rc;
4403 
4404 	return efx_ef10_poll_bist(efx);
4405 }
4406 
4407 static int
4408 efx_ef10_test_chip(struct efx_nic *efx, struct efx_self_tests *tests)
4409 {
4410 	int rc, rc2;
4411 
4412 	efx_reset_down(efx, RESET_TYPE_WORLD);
4413 
4414 	rc = efx_mcdi_rpc(efx, MC_CMD_ENABLE_OFFLINE_BIST,
4415 			  NULL, 0, NULL, 0, NULL);
4416 	if (rc != 0)
4417 		goto out;
4418 
4419 	tests->memory = efx_ef10_run_bist(efx, MC_CMD_MC_MEM_BIST) ? -1 : 1;
4420 	tests->registers = efx_ef10_run_bist(efx, MC_CMD_REG_BIST) ? -1 : 1;
4421 
4422 	rc = efx_mcdi_reset(efx, RESET_TYPE_WORLD);
4423 
4424 out:
4425 	if (rc == -EPERM)
4426 		rc = 0;
4427 	rc2 = efx_reset_up(efx, RESET_TYPE_WORLD, rc == 0);
4428 	return rc ? rc : rc2;
4429 }
4430 
4431 #ifdef CONFIG_SFC_MTD
4432 
4433 struct efx_ef10_nvram_type_info {
4434 	u16 type, type_mask;
4435 	u8 port;
4436 	const char *name;
4437 };
4438 
4439 static const struct efx_ef10_nvram_type_info efx_ef10_nvram_types[] = {
4440 	{ NVRAM_PARTITION_TYPE_MC_FIRMWARE,	   0,    0, "sfc_mcfw" },
4441 	{ NVRAM_PARTITION_TYPE_MC_FIRMWARE_BACKUP, 0,    0, "sfc_mcfw_backup" },
4442 	{ NVRAM_PARTITION_TYPE_EXPANSION_ROM,	   0,    0, "sfc_exp_rom" },
4443 	{ NVRAM_PARTITION_TYPE_STATIC_CONFIG,	   0,    0, "sfc_static_cfg" },
4444 	{ NVRAM_PARTITION_TYPE_DYNAMIC_CONFIG,	   0,    0, "sfc_dynamic_cfg" },
4445 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT0, 0,   0, "sfc_exp_rom_cfg" },
4446 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT1, 0,   1, "sfc_exp_rom_cfg" },
4447 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT2, 0,   2, "sfc_exp_rom_cfg" },
4448 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT3, 0,   3, "sfc_exp_rom_cfg" },
4449 	{ NVRAM_PARTITION_TYPE_LICENSE,		   0,    0, "sfc_license" },
4450 	{ NVRAM_PARTITION_TYPE_PHY_MIN,		   0xff, 0, "sfc_phy_fw" },
4451 };
4452 
4453 static int efx_ef10_mtd_probe_partition(struct efx_nic *efx,
4454 					struct efx_mcdi_mtd_partition *part,
4455 					unsigned int type)
4456 {
4457 	MCDI_DECLARE_BUF(inbuf, MC_CMD_NVRAM_METADATA_IN_LEN);
4458 	MCDI_DECLARE_BUF(outbuf, MC_CMD_NVRAM_METADATA_OUT_LENMAX);
4459 	const struct efx_ef10_nvram_type_info *info;
4460 	size_t size, erase_size, outlen;
4461 	bool protected;
4462 	int rc;
4463 
4464 	for (info = efx_ef10_nvram_types; ; info++) {
4465 		if (info ==
4466 		    efx_ef10_nvram_types + ARRAY_SIZE(efx_ef10_nvram_types))
4467 			return -ENODEV;
4468 		if ((type & ~info->type_mask) == info->type)
4469 			break;
4470 	}
4471 	if (info->port != efx_port_num(efx))
4472 		return -ENODEV;
4473 
4474 	rc = efx_mcdi_nvram_info(efx, type, &size, &erase_size, &protected);
4475 	if (rc)
4476 		return rc;
4477 	if (protected)
4478 		return -ENODEV; /* hide it */
4479 
4480 	part->nvram_type = type;
4481 
4482 	MCDI_SET_DWORD(inbuf, NVRAM_METADATA_IN_TYPE, type);
4483 	rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_METADATA, inbuf, sizeof(inbuf),
4484 			  outbuf, sizeof(outbuf), &outlen);
4485 	if (rc)
4486 		return rc;
4487 	if (outlen < MC_CMD_NVRAM_METADATA_OUT_LENMIN)
4488 		return -EIO;
4489 	if (MCDI_DWORD(outbuf, NVRAM_METADATA_OUT_FLAGS) &
4490 	    (1 << MC_CMD_NVRAM_METADATA_OUT_SUBTYPE_VALID_LBN))
4491 		part->fw_subtype = MCDI_DWORD(outbuf,
4492 					      NVRAM_METADATA_OUT_SUBTYPE);
4493 
4494 	part->common.dev_type_name = "EF10 NVRAM manager";
4495 	part->common.type_name = info->name;
4496 
4497 	part->common.mtd.type = MTD_NORFLASH;
4498 	part->common.mtd.flags = MTD_CAP_NORFLASH;
4499 	part->common.mtd.size = size;
4500 	part->common.mtd.erasesize = erase_size;
4501 
4502 	return 0;
4503 }
4504 
4505 static int efx_ef10_mtd_probe(struct efx_nic *efx)
4506 {
4507 	MCDI_DECLARE_BUF(outbuf, MC_CMD_NVRAM_PARTITIONS_OUT_LENMAX);
4508 	struct efx_mcdi_mtd_partition *parts;
4509 	size_t outlen, n_parts_total, i, n_parts;
4510 	unsigned int type;
4511 	int rc;
4512 
4513 	ASSERT_RTNL();
4514 
4515 	BUILD_BUG_ON(MC_CMD_NVRAM_PARTITIONS_IN_LEN != 0);
4516 	rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_PARTITIONS, NULL, 0,
4517 			  outbuf, sizeof(outbuf), &outlen);
4518 	if (rc)
4519 		return rc;
4520 	if (outlen < MC_CMD_NVRAM_PARTITIONS_OUT_LENMIN)
4521 		return -EIO;
4522 
4523 	n_parts_total = MCDI_DWORD(outbuf, NVRAM_PARTITIONS_OUT_NUM_PARTITIONS);
4524 	if (n_parts_total >
4525 	    MCDI_VAR_ARRAY_LEN(outlen, NVRAM_PARTITIONS_OUT_TYPE_ID))
4526 		return -EIO;
4527 
4528 	parts = kcalloc(n_parts_total, sizeof(*parts), GFP_KERNEL);
4529 	if (!parts)
4530 		return -ENOMEM;
4531 
4532 	n_parts = 0;
4533 	for (i = 0; i < n_parts_total; i++) {
4534 		type = MCDI_ARRAY_DWORD(outbuf, NVRAM_PARTITIONS_OUT_TYPE_ID,
4535 					i);
4536 		rc = efx_ef10_mtd_probe_partition(efx, &parts[n_parts], type);
4537 		if (rc == 0)
4538 			n_parts++;
4539 		else if (rc != -ENODEV)
4540 			goto fail;
4541 	}
4542 
4543 	rc = efx_mtd_add(efx, &parts[0].common, n_parts, sizeof(*parts));
4544 fail:
4545 	if (rc)
4546 		kfree(parts);
4547 	return rc;
4548 }
4549 
4550 #endif /* CONFIG_SFC_MTD */
4551 
4552 static void efx_ef10_ptp_write_host_time(struct efx_nic *efx, u32 host_time)
4553 {
4554 	_efx_writed(efx, cpu_to_le32(host_time), ER_DZ_MC_DB_LWRD);
4555 }
4556 
4557 static void efx_ef10_ptp_write_host_time_vf(struct efx_nic *efx,
4558 					    u32 host_time) {}
4559 
4560 static int efx_ef10_rx_enable_timestamping(struct efx_channel *channel,
4561 					   bool temp)
4562 {
4563 	MCDI_DECLARE_BUF(inbuf, MC_CMD_PTP_IN_TIME_EVENT_SUBSCRIBE_LEN);
4564 	int rc;
4565 
4566 	if (channel->sync_events_state == SYNC_EVENTS_REQUESTED ||
4567 	    channel->sync_events_state == SYNC_EVENTS_VALID ||
4568 	    (temp && channel->sync_events_state == SYNC_EVENTS_DISABLED))
4569 		return 0;
4570 	channel->sync_events_state = SYNC_EVENTS_REQUESTED;
4571 
4572 	MCDI_SET_DWORD(inbuf, PTP_IN_OP, MC_CMD_PTP_OP_TIME_EVENT_SUBSCRIBE);
4573 	MCDI_SET_DWORD(inbuf, PTP_IN_PERIPH_ID, 0);
4574 	MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_SUBSCRIBE_QUEUE,
4575 		       channel->channel);
4576 
4577 	rc = efx_mcdi_rpc(channel->efx, MC_CMD_PTP,
4578 			  inbuf, sizeof(inbuf), NULL, 0, NULL);
4579 
4580 	if (rc != 0)
4581 		channel->sync_events_state = temp ? SYNC_EVENTS_QUIESCENT :
4582 						    SYNC_EVENTS_DISABLED;
4583 
4584 	return rc;
4585 }
4586 
4587 static int efx_ef10_rx_disable_timestamping(struct efx_channel *channel,
4588 					    bool temp)
4589 {
4590 	MCDI_DECLARE_BUF(inbuf, MC_CMD_PTP_IN_TIME_EVENT_UNSUBSCRIBE_LEN);
4591 	int rc;
4592 
4593 	if (channel->sync_events_state == SYNC_EVENTS_DISABLED ||
4594 	    (temp && channel->sync_events_state == SYNC_EVENTS_QUIESCENT))
4595 		return 0;
4596 	if (channel->sync_events_state == SYNC_EVENTS_QUIESCENT) {
4597 		channel->sync_events_state = SYNC_EVENTS_DISABLED;
4598 		return 0;
4599 	}
4600 	channel->sync_events_state = temp ? SYNC_EVENTS_QUIESCENT :
4601 					    SYNC_EVENTS_DISABLED;
4602 
4603 	MCDI_SET_DWORD(inbuf, PTP_IN_OP, MC_CMD_PTP_OP_TIME_EVENT_UNSUBSCRIBE);
4604 	MCDI_SET_DWORD(inbuf, PTP_IN_PERIPH_ID, 0);
4605 	MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_UNSUBSCRIBE_CONTROL,
4606 		       MC_CMD_PTP_IN_TIME_EVENT_UNSUBSCRIBE_SINGLE);
4607 	MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_UNSUBSCRIBE_QUEUE,
4608 		       channel->channel);
4609 
4610 	rc = efx_mcdi_rpc(channel->efx, MC_CMD_PTP,
4611 			  inbuf, sizeof(inbuf), NULL, 0, NULL);
4612 
4613 	return rc;
4614 }
4615 
4616 static int efx_ef10_ptp_set_ts_sync_events(struct efx_nic *efx, bool en,
4617 					   bool temp)
4618 {
4619 	int (*set)(struct efx_channel *channel, bool temp);
4620 	struct efx_channel *channel;
4621 
4622 	set = en ?
4623 	      efx_ef10_rx_enable_timestamping :
4624 	      efx_ef10_rx_disable_timestamping;
4625 
4626 	efx_for_each_channel(channel, efx) {
4627 		int rc = set(channel, temp);
4628 		if (en && rc != 0) {
4629 			efx_ef10_ptp_set_ts_sync_events(efx, false, temp);
4630 			return rc;
4631 		}
4632 	}
4633 
4634 	return 0;
4635 }
4636 
4637 static int efx_ef10_ptp_set_ts_config_vf(struct efx_nic *efx,
4638 					 struct hwtstamp_config *init)
4639 {
4640 	return -EOPNOTSUPP;
4641 }
4642 
4643 static int efx_ef10_ptp_set_ts_config(struct efx_nic *efx,
4644 				      struct hwtstamp_config *init)
4645 {
4646 	int rc;
4647 
4648 	switch (init->rx_filter) {
4649 	case HWTSTAMP_FILTER_NONE:
4650 		efx_ef10_ptp_set_ts_sync_events(efx, false, false);
4651 		/* if TX timestamping is still requested then leave PTP on */
4652 		return efx_ptp_change_mode(efx,
4653 					   init->tx_type != HWTSTAMP_TX_OFF, 0);
4654 	case HWTSTAMP_FILTER_ALL:
4655 	case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
4656 	case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
4657 	case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
4658 	case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
4659 	case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
4660 	case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
4661 	case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
4662 	case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
4663 	case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
4664 	case HWTSTAMP_FILTER_PTP_V2_EVENT:
4665 	case HWTSTAMP_FILTER_PTP_V2_SYNC:
4666 	case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
4667 		init->rx_filter = HWTSTAMP_FILTER_ALL;
4668 		rc = efx_ptp_change_mode(efx, true, 0);
4669 		if (!rc)
4670 			rc = efx_ef10_ptp_set_ts_sync_events(efx, true, false);
4671 		if (rc)
4672 			efx_ptp_change_mode(efx, false, 0);
4673 		return rc;
4674 	default:
4675 		return -ERANGE;
4676 	}
4677 }
4678 
4679 const struct efx_nic_type efx_hunt_a0_vf_nic_type = {
4680 	.is_vf = true,
4681 	.mem_bar = EFX_MEM_VF_BAR,
4682 	.mem_map_size = efx_ef10_mem_map_size,
4683 	.probe = efx_ef10_probe_vf,
4684 	.remove = efx_ef10_remove,
4685 	.dimension_resources = efx_ef10_dimension_resources,
4686 	.init = efx_ef10_init_nic,
4687 	.fini = efx_port_dummy_op_void,
4688 	.map_reset_reason = efx_ef10_map_reset_reason,
4689 	.map_reset_flags = efx_ef10_map_reset_flags,
4690 	.reset = efx_ef10_reset,
4691 	.probe_port = efx_mcdi_port_probe,
4692 	.remove_port = efx_mcdi_port_remove,
4693 	.fini_dmaq = efx_ef10_fini_dmaq,
4694 	.prepare_flr = efx_ef10_prepare_flr,
4695 	.finish_flr = efx_port_dummy_op_void,
4696 	.describe_stats = efx_ef10_describe_stats,
4697 	.update_stats = efx_ef10_update_stats_vf,
4698 	.start_stats = efx_port_dummy_op_void,
4699 	.pull_stats = efx_port_dummy_op_void,
4700 	.stop_stats = efx_port_dummy_op_void,
4701 	.set_id_led = efx_mcdi_set_id_led,
4702 	.push_irq_moderation = efx_ef10_push_irq_moderation,
4703 	.reconfigure_mac = efx_ef10_mac_reconfigure_vf,
4704 	.check_mac_fault = efx_mcdi_mac_check_fault,
4705 	.reconfigure_port = efx_mcdi_port_reconfigure,
4706 	.get_wol = efx_ef10_get_wol_vf,
4707 	.set_wol = efx_ef10_set_wol_vf,
4708 	.resume_wol = efx_port_dummy_op_void,
4709 	.mcdi_request = efx_ef10_mcdi_request,
4710 	.mcdi_poll_response = efx_ef10_mcdi_poll_response,
4711 	.mcdi_read_response = efx_ef10_mcdi_read_response,
4712 	.mcdi_poll_reboot = efx_ef10_mcdi_poll_reboot,
4713 	.mcdi_reboot_detected = efx_ef10_mcdi_reboot_detected,
4714 	.irq_enable_master = efx_port_dummy_op_void,
4715 	.irq_test_generate = efx_ef10_irq_test_generate,
4716 	.irq_disable_non_ev = efx_port_dummy_op_void,
4717 	.irq_handle_msi = efx_ef10_msi_interrupt,
4718 	.irq_handle_legacy = efx_ef10_legacy_interrupt,
4719 	.tx_probe = efx_ef10_tx_probe,
4720 	.tx_init = efx_ef10_tx_init,
4721 	.tx_remove = efx_ef10_tx_remove,
4722 	.tx_write = efx_ef10_tx_write,
4723 	.rx_push_rss_config = efx_ef10_vf_rx_push_rss_config,
4724 	.rx_probe = efx_ef10_rx_probe,
4725 	.rx_init = efx_ef10_rx_init,
4726 	.rx_remove = efx_ef10_rx_remove,
4727 	.rx_write = efx_ef10_rx_write,
4728 	.rx_defer_refill = efx_ef10_rx_defer_refill,
4729 	.ev_probe = efx_ef10_ev_probe,
4730 	.ev_init = efx_ef10_ev_init,
4731 	.ev_fini = efx_ef10_ev_fini,
4732 	.ev_remove = efx_ef10_ev_remove,
4733 	.ev_process = efx_ef10_ev_process,
4734 	.ev_read_ack = efx_ef10_ev_read_ack,
4735 	.ev_test_generate = efx_ef10_ev_test_generate,
4736 	.filter_table_probe = efx_ef10_filter_table_probe,
4737 	.filter_table_restore = efx_ef10_filter_table_restore,
4738 	.filter_table_remove = efx_ef10_filter_table_remove,
4739 	.filter_update_rx_scatter = efx_ef10_filter_update_rx_scatter,
4740 	.filter_insert = efx_ef10_filter_insert,
4741 	.filter_remove_safe = efx_ef10_filter_remove_safe,
4742 	.filter_get_safe = efx_ef10_filter_get_safe,
4743 	.filter_clear_rx = efx_ef10_filter_clear_rx,
4744 	.filter_count_rx_used = efx_ef10_filter_count_rx_used,
4745 	.filter_get_rx_id_limit = efx_ef10_filter_get_rx_id_limit,
4746 	.filter_get_rx_ids = efx_ef10_filter_get_rx_ids,
4747 #ifdef CONFIG_RFS_ACCEL
4748 	.filter_rfs_insert = efx_ef10_filter_rfs_insert,
4749 	.filter_rfs_expire_one = efx_ef10_filter_rfs_expire_one,
4750 #endif
4751 #ifdef CONFIG_SFC_MTD
4752 	.mtd_probe = efx_port_dummy_op_int,
4753 #endif
4754 	.ptp_write_host_time = efx_ef10_ptp_write_host_time_vf,
4755 	.ptp_set_ts_config = efx_ef10_ptp_set_ts_config_vf,
4756 #ifdef CONFIG_SFC_SRIOV
4757 	.vswitching_probe = efx_ef10_vswitching_probe_vf,
4758 	.vswitching_restore = efx_ef10_vswitching_restore_vf,
4759 	.vswitching_remove = efx_ef10_vswitching_remove_vf,
4760 	.sriov_get_phys_port_id = efx_ef10_sriov_get_phys_port_id,
4761 #endif
4762 	.get_mac_address = efx_ef10_get_mac_address_vf,
4763 	.set_mac_address = efx_ef10_set_mac_address,
4764 
4765 	.revision = EFX_REV_HUNT_A0,
4766 	.max_dma_mask = DMA_BIT_MASK(ESF_DZ_TX_KER_BUF_ADDR_WIDTH),
4767 	.rx_prefix_size = ES_DZ_RX_PREFIX_SIZE,
4768 	.rx_hash_offset = ES_DZ_RX_PREFIX_HASH_OFST,
4769 	.rx_ts_offset = ES_DZ_RX_PREFIX_TSTAMP_OFST,
4770 	.can_rx_scatter = true,
4771 	.always_rx_scatter = true,
4772 	.max_interrupt_mode = EFX_INT_MODE_MSIX,
4773 	.timer_period_max = 1 << ERF_DD_EVQ_IND_TIMER_VAL_WIDTH,
4774 	.offload_features = (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
4775 			     NETIF_F_RXHASH | NETIF_F_NTUPLE),
4776 	.mcdi_max_ver = 2,
4777 	.max_rx_ip_filters = HUNT_FILTER_TBL_ROWS,
4778 	.hwtstamp_filters = 1 << HWTSTAMP_FILTER_NONE |
4779 			    1 << HWTSTAMP_FILTER_ALL,
4780 };
4781 
4782 const struct efx_nic_type efx_hunt_a0_nic_type = {
4783 	.is_vf = false,
4784 	.mem_bar = EFX_MEM_BAR,
4785 	.mem_map_size = efx_ef10_mem_map_size,
4786 	.probe = efx_ef10_probe_pf,
4787 	.remove = efx_ef10_remove,
4788 	.dimension_resources = efx_ef10_dimension_resources,
4789 	.init = efx_ef10_init_nic,
4790 	.fini = efx_port_dummy_op_void,
4791 	.map_reset_reason = efx_ef10_map_reset_reason,
4792 	.map_reset_flags = efx_ef10_map_reset_flags,
4793 	.reset = efx_ef10_reset,
4794 	.probe_port = efx_mcdi_port_probe,
4795 	.remove_port = efx_mcdi_port_remove,
4796 	.fini_dmaq = efx_ef10_fini_dmaq,
4797 	.prepare_flr = efx_ef10_prepare_flr,
4798 	.finish_flr = efx_port_dummy_op_void,
4799 	.describe_stats = efx_ef10_describe_stats,
4800 	.update_stats = efx_ef10_update_stats_pf,
4801 	.start_stats = efx_mcdi_mac_start_stats,
4802 	.pull_stats = efx_mcdi_mac_pull_stats,
4803 	.stop_stats = efx_mcdi_mac_stop_stats,
4804 	.set_id_led = efx_mcdi_set_id_led,
4805 	.push_irq_moderation = efx_ef10_push_irq_moderation,
4806 	.reconfigure_mac = efx_ef10_mac_reconfigure,
4807 	.check_mac_fault = efx_mcdi_mac_check_fault,
4808 	.reconfigure_port = efx_mcdi_port_reconfigure,
4809 	.get_wol = efx_ef10_get_wol,
4810 	.set_wol = efx_ef10_set_wol,
4811 	.resume_wol = efx_port_dummy_op_void,
4812 	.test_chip = efx_ef10_test_chip,
4813 	.test_nvram = efx_mcdi_nvram_test_all,
4814 	.mcdi_request = efx_ef10_mcdi_request,
4815 	.mcdi_poll_response = efx_ef10_mcdi_poll_response,
4816 	.mcdi_read_response = efx_ef10_mcdi_read_response,
4817 	.mcdi_poll_reboot = efx_ef10_mcdi_poll_reboot,
4818 	.mcdi_reboot_detected = efx_ef10_mcdi_reboot_detected,
4819 	.irq_enable_master = efx_port_dummy_op_void,
4820 	.irq_test_generate = efx_ef10_irq_test_generate,
4821 	.irq_disable_non_ev = efx_port_dummy_op_void,
4822 	.irq_handle_msi = efx_ef10_msi_interrupt,
4823 	.irq_handle_legacy = efx_ef10_legacy_interrupt,
4824 	.tx_probe = efx_ef10_tx_probe,
4825 	.tx_init = efx_ef10_tx_init,
4826 	.tx_remove = efx_ef10_tx_remove,
4827 	.tx_write = efx_ef10_tx_write,
4828 	.rx_push_rss_config = efx_ef10_pf_rx_push_rss_config,
4829 	.rx_probe = efx_ef10_rx_probe,
4830 	.rx_init = efx_ef10_rx_init,
4831 	.rx_remove = efx_ef10_rx_remove,
4832 	.rx_write = efx_ef10_rx_write,
4833 	.rx_defer_refill = efx_ef10_rx_defer_refill,
4834 	.ev_probe = efx_ef10_ev_probe,
4835 	.ev_init = efx_ef10_ev_init,
4836 	.ev_fini = efx_ef10_ev_fini,
4837 	.ev_remove = efx_ef10_ev_remove,
4838 	.ev_process = efx_ef10_ev_process,
4839 	.ev_read_ack = efx_ef10_ev_read_ack,
4840 	.ev_test_generate = efx_ef10_ev_test_generate,
4841 	.filter_table_probe = efx_ef10_filter_table_probe,
4842 	.filter_table_restore = efx_ef10_filter_table_restore,
4843 	.filter_table_remove = efx_ef10_filter_table_remove,
4844 	.filter_update_rx_scatter = efx_ef10_filter_update_rx_scatter,
4845 	.filter_insert = efx_ef10_filter_insert,
4846 	.filter_remove_safe = efx_ef10_filter_remove_safe,
4847 	.filter_get_safe = efx_ef10_filter_get_safe,
4848 	.filter_clear_rx = efx_ef10_filter_clear_rx,
4849 	.filter_count_rx_used = efx_ef10_filter_count_rx_used,
4850 	.filter_get_rx_id_limit = efx_ef10_filter_get_rx_id_limit,
4851 	.filter_get_rx_ids = efx_ef10_filter_get_rx_ids,
4852 #ifdef CONFIG_RFS_ACCEL
4853 	.filter_rfs_insert = efx_ef10_filter_rfs_insert,
4854 	.filter_rfs_expire_one = efx_ef10_filter_rfs_expire_one,
4855 #endif
4856 #ifdef CONFIG_SFC_MTD
4857 	.mtd_probe = efx_ef10_mtd_probe,
4858 	.mtd_rename = efx_mcdi_mtd_rename,
4859 	.mtd_read = efx_mcdi_mtd_read,
4860 	.mtd_erase = efx_mcdi_mtd_erase,
4861 	.mtd_write = efx_mcdi_mtd_write,
4862 	.mtd_sync = efx_mcdi_mtd_sync,
4863 #endif
4864 	.ptp_write_host_time = efx_ef10_ptp_write_host_time,
4865 	.ptp_set_ts_sync_events = efx_ef10_ptp_set_ts_sync_events,
4866 	.ptp_set_ts_config = efx_ef10_ptp_set_ts_config,
4867 #ifdef CONFIG_SFC_SRIOV
4868 	.sriov_configure = efx_ef10_sriov_configure,
4869 	.sriov_init = efx_ef10_sriov_init,
4870 	.sriov_fini = efx_ef10_sriov_fini,
4871 	.sriov_wanted = efx_ef10_sriov_wanted,
4872 	.sriov_reset = efx_ef10_sriov_reset,
4873 	.sriov_flr = efx_ef10_sriov_flr,
4874 	.sriov_set_vf_mac = efx_ef10_sriov_set_vf_mac,
4875 	.sriov_set_vf_vlan = efx_ef10_sriov_set_vf_vlan,
4876 	.sriov_set_vf_spoofchk = efx_ef10_sriov_set_vf_spoofchk,
4877 	.sriov_get_vf_config = efx_ef10_sriov_get_vf_config,
4878 	.sriov_set_vf_link_state = efx_ef10_sriov_set_vf_link_state,
4879 	.vswitching_probe = efx_ef10_vswitching_probe_pf,
4880 	.vswitching_restore = efx_ef10_vswitching_restore_pf,
4881 	.vswitching_remove = efx_ef10_vswitching_remove_pf,
4882 #endif
4883 	.get_mac_address = efx_ef10_get_mac_address_pf,
4884 	.set_mac_address = efx_ef10_set_mac_address,
4885 
4886 	.revision = EFX_REV_HUNT_A0,
4887 	.max_dma_mask = DMA_BIT_MASK(ESF_DZ_TX_KER_BUF_ADDR_WIDTH),
4888 	.rx_prefix_size = ES_DZ_RX_PREFIX_SIZE,
4889 	.rx_hash_offset = ES_DZ_RX_PREFIX_HASH_OFST,
4890 	.rx_ts_offset = ES_DZ_RX_PREFIX_TSTAMP_OFST,
4891 	.can_rx_scatter = true,
4892 	.always_rx_scatter = true,
4893 	.max_interrupt_mode = EFX_INT_MODE_MSIX,
4894 	.timer_period_max = 1 << ERF_DD_EVQ_IND_TIMER_VAL_WIDTH,
4895 	.offload_features = (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
4896 			     NETIF_F_RXHASH | NETIF_F_NTUPLE),
4897 	.mcdi_max_ver = 2,
4898 	.max_rx_ip_filters = HUNT_FILTER_TBL_ROWS,
4899 	.hwtstamp_filters = 1 << HWTSTAMP_FILTER_NONE |
4900 			    1 << HWTSTAMP_FILTER_ALL,
4901 };
4902