xref: /openbmc/linux/drivers/net/ethernet/sfc/ef10.c (revision 3805e6a1)
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 	if (nic_data->datapath_caps &
1924 	    1 << MC_CMD_GET_CAPABILITIES_OUT_RX_RSS_LIMITED_LBN)
1925 		return -EOPNOTSUPP;
1926 
1927 	MCDI_SET_DWORD(inbuf, RSS_CONTEXT_ALLOC_IN_UPSTREAM_PORT_ID,
1928 		       nic_data->vport_id);
1929 	MCDI_SET_DWORD(inbuf, RSS_CONTEXT_ALLOC_IN_TYPE, alloc_type);
1930 	MCDI_SET_DWORD(inbuf, RSS_CONTEXT_ALLOC_IN_NUM_QUEUES, rss_spread);
1931 
1932 	rc = efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_ALLOC, inbuf, sizeof(inbuf),
1933 		outbuf, sizeof(outbuf), &outlen);
1934 	if (rc != 0)
1935 		return rc;
1936 
1937 	if (outlen < MC_CMD_RSS_CONTEXT_ALLOC_OUT_LEN)
1938 		return -EIO;
1939 
1940 	*context = MCDI_DWORD(outbuf, RSS_CONTEXT_ALLOC_OUT_RSS_CONTEXT_ID);
1941 
1942 	if (context_size)
1943 		*context_size = rss_spread;
1944 
1945 	return 0;
1946 }
1947 
1948 static void efx_ef10_free_rss_context(struct efx_nic *efx, u32 context)
1949 {
1950 	MCDI_DECLARE_BUF(inbuf, MC_CMD_RSS_CONTEXT_FREE_IN_LEN);
1951 	int rc;
1952 
1953 	MCDI_SET_DWORD(inbuf, RSS_CONTEXT_FREE_IN_RSS_CONTEXT_ID,
1954 		       context);
1955 
1956 	rc = efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_FREE, inbuf, sizeof(inbuf),
1957 			    NULL, 0, NULL);
1958 	WARN_ON(rc != 0);
1959 }
1960 
1961 static int efx_ef10_populate_rss_table(struct efx_nic *efx, u32 context,
1962 				       const u32 *rx_indir_table)
1963 {
1964 	MCDI_DECLARE_BUF(tablebuf, MC_CMD_RSS_CONTEXT_SET_TABLE_IN_LEN);
1965 	MCDI_DECLARE_BUF(keybuf, MC_CMD_RSS_CONTEXT_SET_KEY_IN_LEN);
1966 	int i, rc;
1967 
1968 	MCDI_SET_DWORD(tablebuf, RSS_CONTEXT_SET_TABLE_IN_RSS_CONTEXT_ID,
1969 		       context);
1970 	BUILD_BUG_ON(ARRAY_SIZE(efx->rx_indir_table) !=
1971 		     MC_CMD_RSS_CONTEXT_SET_TABLE_IN_INDIRECTION_TABLE_LEN);
1972 
1973 	for (i = 0; i < ARRAY_SIZE(efx->rx_indir_table); ++i)
1974 		MCDI_PTR(tablebuf,
1975 			 RSS_CONTEXT_SET_TABLE_IN_INDIRECTION_TABLE)[i] =
1976 				(u8) rx_indir_table[i];
1977 
1978 	rc = efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_SET_TABLE, tablebuf,
1979 			  sizeof(tablebuf), NULL, 0, NULL);
1980 	if (rc != 0)
1981 		return rc;
1982 
1983 	MCDI_SET_DWORD(keybuf, RSS_CONTEXT_SET_KEY_IN_RSS_CONTEXT_ID,
1984 		       context);
1985 	BUILD_BUG_ON(ARRAY_SIZE(efx->rx_hash_key) !=
1986 		     MC_CMD_RSS_CONTEXT_SET_KEY_IN_TOEPLITZ_KEY_LEN);
1987 	for (i = 0; i < ARRAY_SIZE(efx->rx_hash_key); ++i)
1988 		MCDI_PTR(keybuf, RSS_CONTEXT_SET_KEY_IN_TOEPLITZ_KEY)[i] =
1989 			efx->rx_hash_key[i];
1990 
1991 	return efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_SET_KEY, keybuf,
1992 			    sizeof(keybuf), NULL, 0, NULL);
1993 }
1994 
1995 static void efx_ef10_rx_free_indir_table(struct efx_nic *efx)
1996 {
1997 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1998 
1999 	if (nic_data->rx_rss_context != EFX_EF10_RSS_CONTEXT_INVALID)
2000 		efx_ef10_free_rss_context(efx, nic_data->rx_rss_context);
2001 	nic_data->rx_rss_context = EFX_EF10_RSS_CONTEXT_INVALID;
2002 }
2003 
2004 static int efx_ef10_rx_push_shared_rss_config(struct efx_nic *efx,
2005 					      unsigned *context_size)
2006 {
2007 	u32 new_rx_rss_context;
2008 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2009 	int rc = efx_ef10_alloc_rss_context(efx, &new_rx_rss_context,
2010 					    false, context_size);
2011 
2012 	if (rc != 0)
2013 		return rc;
2014 
2015 	nic_data->rx_rss_context = new_rx_rss_context;
2016 	nic_data->rx_rss_context_exclusive = false;
2017 	efx_set_default_rx_indir_table(efx);
2018 	return 0;
2019 }
2020 
2021 static int efx_ef10_rx_push_exclusive_rss_config(struct efx_nic *efx,
2022 						 const u32 *rx_indir_table)
2023 {
2024 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2025 	int rc;
2026 	u32 new_rx_rss_context;
2027 
2028 	if (nic_data->rx_rss_context == EFX_EF10_RSS_CONTEXT_INVALID ||
2029 	    !nic_data->rx_rss_context_exclusive) {
2030 		rc = efx_ef10_alloc_rss_context(efx, &new_rx_rss_context,
2031 						true, NULL);
2032 		if (rc == -EOPNOTSUPP)
2033 			return rc;
2034 		else if (rc != 0)
2035 			goto fail1;
2036 	} else {
2037 		new_rx_rss_context = nic_data->rx_rss_context;
2038 	}
2039 
2040 	rc = efx_ef10_populate_rss_table(efx, new_rx_rss_context,
2041 					 rx_indir_table);
2042 	if (rc != 0)
2043 		goto fail2;
2044 
2045 	if (nic_data->rx_rss_context != new_rx_rss_context)
2046 		efx_ef10_rx_free_indir_table(efx);
2047 	nic_data->rx_rss_context = new_rx_rss_context;
2048 	nic_data->rx_rss_context_exclusive = true;
2049 	if (rx_indir_table != efx->rx_indir_table)
2050 		memcpy(efx->rx_indir_table, rx_indir_table,
2051 		       sizeof(efx->rx_indir_table));
2052 	return 0;
2053 
2054 fail2:
2055 	if (new_rx_rss_context != nic_data->rx_rss_context)
2056 		efx_ef10_free_rss_context(efx, new_rx_rss_context);
2057 fail1:
2058 	netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
2059 	return rc;
2060 }
2061 
2062 static int efx_ef10_pf_rx_push_rss_config(struct efx_nic *efx, bool user,
2063 					  const u32 *rx_indir_table)
2064 {
2065 	int rc;
2066 
2067 	if (efx->rss_spread == 1)
2068 		return 0;
2069 
2070 	rc = efx_ef10_rx_push_exclusive_rss_config(efx, rx_indir_table);
2071 
2072 	if (rc == -ENOBUFS && !user) {
2073 		unsigned context_size;
2074 		bool mismatch = false;
2075 		size_t i;
2076 
2077 		for (i = 0; i < ARRAY_SIZE(efx->rx_indir_table) && !mismatch;
2078 		     i++)
2079 			mismatch = rx_indir_table[i] !=
2080 				ethtool_rxfh_indir_default(i, efx->rss_spread);
2081 
2082 		rc = efx_ef10_rx_push_shared_rss_config(efx, &context_size);
2083 		if (rc == 0) {
2084 			if (context_size != efx->rss_spread)
2085 				netif_warn(efx, probe, efx->net_dev,
2086 					   "Could not allocate an exclusive RSS"
2087 					   " context; allocated a shared one of"
2088 					   " different size."
2089 					   " Wanted %u, got %u.\n",
2090 					   efx->rss_spread, context_size);
2091 			else if (mismatch)
2092 				netif_warn(efx, probe, efx->net_dev,
2093 					   "Could not allocate an exclusive RSS"
2094 					   " context; allocated a shared one but"
2095 					   " could not apply custom"
2096 					   " indirection.\n");
2097 			else
2098 				netif_info(efx, probe, efx->net_dev,
2099 					   "Could not allocate an exclusive RSS"
2100 					   " context; allocated a shared one.\n");
2101 		}
2102 	}
2103 	return rc;
2104 }
2105 
2106 static int efx_ef10_vf_rx_push_rss_config(struct efx_nic *efx, bool user,
2107 					  const u32 *rx_indir_table
2108 					  __attribute__ ((unused)))
2109 {
2110 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2111 
2112 	if (user)
2113 		return -EOPNOTSUPP;
2114 	if (nic_data->rx_rss_context != EFX_EF10_RSS_CONTEXT_INVALID)
2115 		return 0;
2116 	return efx_ef10_rx_push_shared_rss_config(efx, NULL);
2117 }
2118 
2119 static int efx_ef10_rx_probe(struct efx_rx_queue *rx_queue)
2120 {
2121 	return efx_nic_alloc_buffer(rx_queue->efx, &rx_queue->rxd.buf,
2122 				    (rx_queue->ptr_mask + 1) *
2123 				    sizeof(efx_qword_t),
2124 				    GFP_KERNEL);
2125 }
2126 
2127 static void efx_ef10_rx_init(struct efx_rx_queue *rx_queue)
2128 {
2129 	MCDI_DECLARE_BUF(inbuf,
2130 			 MC_CMD_INIT_RXQ_IN_LEN(EFX_MAX_DMAQ_SIZE * 8 /
2131 						EFX_BUF_SIZE));
2132 	struct efx_channel *channel = efx_rx_queue_channel(rx_queue);
2133 	size_t entries = rx_queue->rxd.buf.len / EFX_BUF_SIZE;
2134 	struct efx_nic *efx = rx_queue->efx;
2135 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2136 	size_t inlen;
2137 	dma_addr_t dma_addr;
2138 	int rc;
2139 	int i;
2140 	BUILD_BUG_ON(MC_CMD_INIT_RXQ_OUT_LEN != 0);
2141 
2142 	rx_queue->scatter_n = 0;
2143 	rx_queue->scatter_len = 0;
2144 
2145 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_SIZE, rx_queue->ptr_mask + 1);
2146 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_TARGET_EVQ, channel->channel);
2147 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_LABEL, efx_rx_queue_index(rx_queue));
2148 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_INSTANCE,
2149 		       efx_rx_queue_index(rx_queue));
2150 	MCDI_POPULATE_DWORD_2(inbuf, INIT_RXQ_IN_FLAGS,
2151 			      INIT_RXQ_IN_FLAG_PREFIX, 1,
2152 			      INIT_RXQ_IN_FLAG_TIMESTAMP, 1);
2153 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_OWNER_ID, 0);
2154 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_PORT_ID, nic_data->vport_id);
2155 
2156 	dma_addr = rx_queue->rxd.buf.dma_addr;
2157 
2158 	netif_dbg(efx, hw, efx->net_dev, "pushing RXQ %d. %zu entries (%llx)\n",
2159 		  efx_rx_queue_index(rx_queue), entries, (u64)dma_addr);
2160 
2161 	for (i = 0; i < entries; ++i) {
2162 		MCDI_SET_ARRAY_QWORD(inbuf, INIT_RXQ_IN_DMA_ADDR, i, dma_addr);
2163 		dma_addr += EFX_BUF_SIZE;
2164 	}
2165 
2166 	inlen = MC_CMD_INIT_RXQ_IN_LEN(entries);
2167 
2168 	rc = efx_mcdi_rpc(efx, MC_CMD_INIT_RXQ, inbuf, inlen,
2169 			  NULL, 0, NULL);
2170 	if (rc)
2171 		netdev_WARN(efx->net_dev, "failed to initialise RXQ %d\n",
2172 			    efx_rx_queue_index(rx_queue));
2173 }
2174 
2175 static void efx_ef10_rx_fini(struct efx_rx_queue *rx_queue)
2176 {
2177 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FINI_RXQ_IN_LEN);
2178 	MCDI_DECLARE_BUF_ERR(outbuf);
2179 	struct efx_nic *efx = rx_queue->efx;
2180 	size_t outlen;
2181 	int rc;
2182 
2183 	MCDI_SET_DWORD(inbuf, FINI_RXQ_IN_INSTANCE,
2184 		       efx_rx_queue_index(rx_queue));
2185 
2186 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FINI_RXQ, inbuf, sizeof(inbuf),
2187 			  outbuf, sizeof(outbuf), &outlen);
2188 
2189 	if (rc && rc != -EALREADY)
2190 		goto fail;
2191 
2192 	return;
2193 
2194 fail:
2195 	efx_mcdi_display_error(efx, MC_CMD_FINI_RXQ, MC_CMD_FINI_RXQ_IN_LEN,
2196 			       outbuf, outlen, rc);
2197 }
2198 
2199 static void efx_ef10_rx_remove(struct efx_rx_queue *rx_queue)
2200 {
2201 	efx_nic_free_buffer(rx_queue->efx, &rx_queue->rxd.buf);
2202 }
2203 
2204 /* This creates an entry in the RX descriptor queue */
2205 static inline void
2206 efx_ef10_build_rx_desc(struct efx_rx_queue *rx_queue, unsigned int index)
2207 {
2208 	struct efx_rx_buffer *rx_buf;
2209 	efx_qword_t *rxd;
2210 
2211 	rxd = efx_rx_desc(rx_queue, index);
2212 	rx_buf = efx_rx_buffer(rx_queue, index);
2213 	EFX_POPULATE_QWORD_2(*rxd,
2214 			     ESF_DZ_RX_KER_BYTE_CNT, rx_buf->len,
2215 			     ESF_DZ_RX_KER_BUF_ADDR, rx_buf->dma_addr);
2216 }
2217 
2218 static void efx_ef10_rx_write(struct efx_rx_queue *rx_queue)
2219 {
2220 	struct efx_nic *efx = rx_queue->efx;
2221 	unsigned int write_count;
2222 	efx_dword_t reg;
2223 
2224 	/* Firmware requires that RX_DESC_WPTR be a multiple of 8 */
2225 	write_count = rx_queue->added_count & ~7;
2226 	if (rx_queue->notified_count == write_count)
2227 		return;
2228 
2229 	do
2230 		efx_ef10_build_rx_desc(
2231 			rx_queue,
2232 			rx_queue->notified_count & rx_queue->ptr_mask);
2233 	while (++rx_queue->notified_count != write_count);
2234 
2235 	wmb();
2236 	EFX_POPULATE_DWORD_1(reg, ERF_DZ_RX_DESC_WPTR,
2237 			     write_count & rx_queue->ptr_mask);
2238 	efx_writed_page(efx, &reg, ER_DZ_RX_DESC_UPD,
2239 			efx_rx_queue_index(rx_queue));
2240 }
2241 
2242 static efx_mcdi_async_completer efx_ef10_rx_defer_refill_complete;
2243 
2244 static void efx_ef10_rx_defer_refill(struct efx_rx_queue *rx_queue)
2245 {
2246 	struct efx_channel *channel = efx_rx_queue_channel(rx_queue);
2247 	MCDI_DECLARE_BUF(inbuf, MC_CMD_DRIVER_EVENT_IN_LEN);
2248 	efx_qword_t event;
2249 
2250 	EFX_POPULATE_QWORD_2(event,
2251 			     ESF_DZ_EV_CODE, EFX_EF10_DRVGEN_EV,
2252 			     ESF_DZ_EV_DATA, EFX_EF10_REFILL);
2253 
2254 	MCDI_SET_DWORD(inbuf, DRIVER_EVENT_IN_EVQ, channel->channel);
2255 
2256 	/* MCDI_SET_QWORD is not appropriate here since EFX_POPULATE_* has
2257 	 * already swapped the data to little-endian order.
2258 	 */
2259 	memcpy(MCDI_PTR(inbuf, DRIVER_EVENT_IN_DATA), &event.u64[0],
2260 	       sizeof(efx_qword_t));
2261 
2262 	efx_mcdi_rpc_async(channel->efx, MC_CMD_DRIVER_EVENT,
2263 			   inbuf, sizeof(inbuf), 0,
2264 			   efx_ef10_rx_defer_refill_complete, 0);
2265 }
2266 
2267 static void
2268 efx_ef10_rx_defer_refill_complete(struct efx_nic *efx, unsigned long cookie,
2269 				  int rc, efx_dword_t *outbuf,
2270 				  size_t outlen_actual)
2271 {
2272 	/* nothing to do */
2273 }
2274 
2275 static int efx_ef10_ev_probe(struct efx_channel *channel)
2276 {
2277 	return efx_nic_alloc_buffer(channel->efx, &channel->eventq.buf,
2278 				    (channel->eventq_mask + 1) *
2279 				    sizeof(efx_qword_t),
2280 				    GFP_KERNEL);
2281 }
2282 
2283 static void efx_ef10_ev_fini(struct efx_channel *channel)
2284 {
2285 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FINI_EVQ_IN_LEN);
2286 	MCDI_DECLARE_BUF_ERR(outbuf);
2287 	struct efx_nic *efx = channel->efx;
2288 	size_t outlen;
2289 	int rc;
2290 
2291 	MCDI_SET_DWORD(inbuf, FINI_EVQ_IN_INSTANCE, channel->channel);
2292 
2293 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FINI_EVQ, inbuf, sizeof(inbuf),
2294 			  outbuf, sizeof(outbuf), &outlen);
2295 
2296 	if (rc && rc != -EALREADY)
2297 		goto fail;
2298 
2299 	return;
2300 
2301 fail:
2302 	efx_mcdi_display_error(efx, MC_CMD_FINI_EVQ, MC_CMD_FINI_EVQ_IN_LEN,
2303 			       outbuf, outlen, rc);
2304 }
2305 
2306 static int efx_ef10_ev_init(struct efx_channel *channel)
2307 {
2308 	MCDI_DECLARE_BUF(inbuf,
2309 			 MC_CMD_INIT_EVQ_IN_LEN(EFX_MAX_EVQ_SIZE * 8 /
2310 						EFX_BUF_SIZE));
2311 	MCDI_DECLARE_BUF(outbuf, MC_CMD_INIT_EVQ_OUT_LEN);
2312 	size_t entries = channel->eventq.buf.len / EFX_BUF_SIZE;
2313 	struct efx_nic *efx = channel->efx;
2314 	struct efx_ef10_nic_data *nic_data;
2315 	bool supports_rx_merge;
2316 	size_t inlen, outlen;
2317 	unsigned int enabled, implemented;
2318 	dma_addr_t dma_addr;
2319 	int rc;
2320 	int i;
2321 
2322 	nic_data = efx->nic_data;
2323 	supports_rx_merge =
2324 		!!(nic_data->datapath_caps &
2325 		   1 << MC_CMD_GET_CAPABILITIES_OUT_RX_BATCHING_LBN);
2326 
2327 	/* Fill event queue with all ones (i.e. empty events) */
2328 	memset(channel->eventq.buf.addr, 0xff, channel->eventq.buf.len);
2329 
2330 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_SIZE, channel->eventq_mask + 1);
2331 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_INSTANCE, channel->channel);
2332 	/* INIT_EVQ expects index in vector table, not absolute */
2333 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_IRQ_NUM, channel->channel);
2334 	MCDI_POPULATE_DWORD_4(inbuf, INIT_EVQ_IN_FLAGS,
2335 			      INIT_EVQ_IN_FLAG_INTERRUPTING, 1,
2336 			      INIT_EVQ_IN_FLAG_RX_MERGE, 1,
2337 			      INIT_EVQ_IN_FLAG_TX_MERGE, 1,
2338 			      INIT_EVQ_IN_FLAG_CUT_THRU, !supports_rx_merge);
2339 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_TMR_MODE,
2340 		       MC_CMD_INIT_EVQ_IN_TMR_MODE_DIS);
2341 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_TMR_LOAD, 0);
2342 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_TMR_RELOAD, 0);
2343 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_COUNT_MODE,
2344 		       MC_CMD_INIT_EVQ_IN_COUNT_MODE_DIS);
2345 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_COUNT_THRSHLD, 0);
2346 
2347 	dma_addr = channel->eventq.buf.dma_addr;
2348 	for (i = 0; i < entries; ++i) {
2349 		MCDI_SET_ARRAY_QWORD(inbuf, INIT_EVQ_IN_DMA_ADDR, i, dma_addr);
2350 		dma_addr += EFX_BUF_SIZE;
2351 	}
2352 
2353 	inlen = MC_CMD_INIT_EVQ_IN_LEN(entries);
2354 
2355 	rc = efx_mcdi_rpc(efx, MC_CMD_INIT_EVQ, inbuf, inlen,
2356 			  outbuf, sizeof(outbuf), &outlen);
2357 	/* IRQ return is ignored */
2358 	if (channel->channel || rc)
2359 		return rc;
2360 
2361 	/* Successfully created event queue on channel 0 */
2362 	rc = efx_mcdi_get_workarounds(efx, &implemented, &enabled);
2363 	if (rc == -ENOSYS) {
2364 		/* GET_WORKAROUNDS was implemented before the bug26807
2365 		 * workaround, thus the latter must be unavailable in this fw
2366 		 */
2367 		nic_data->workaround_26807 = false;
2368 		rc = 0;
2369 	} else if (rc) {
2370 		goto fail;
2371 	} else {
2372 		nic_data->workaround_26807 =
2373 			!!(enabled & MC_CMD_GET_WORKAROUNDS_OUT_BUG26807);
2374 
2375 		if (implemented & MC_CMD_GET_WORKAROUNDS_OUT_BUG26807 &&
2376 		    !nic_data->workaround_26807) {
2377 			unsigned int flags;
2378 
2379 			rc = efx_mcdi_set_workaround(efx,
2380 						     MC_CMD_WORKAROUND_BUG26807,
2381 						     true, &flags);
2382 
2383 			if (!rc) {
2384 				if (flags &
2385 				    1 << MC_CMD_WORKAROUND_EXT_OUT_FLR_DONE_LBN) {
2386 					netif_info(efx, drv, efx->net_dev,
2387 						   "other functions on NIC have been reset\n");
2388 
2389 					/* With MCFW v4.6.x and earlier, the
2390 					 * boot count will have incremented,
2391 					 * so re-read the warm_boot_count
2392 					 * value now to ensure this function
2393 					 * doesn't think it has changed next
2394 					 * time it checks.
2395 					 */
2396 					rc = efx_ef10_get_warm_boot_count(efx);
2397 					if (rc >= 0) {
2398 						nic_data->warm_boot_count = rc;
2399 						rc = 0;
2400 					}
2401 				}
2402 				nic_data->workaround_26807 = true;
2403 			} else if (rc == -EPERM) {
2404 				rc = 0;
2405 			}
2406 		}
2407 	}
2408 
2409 	if (!rc)
2410 		return 0;
2411 
2412 fail:
2413 	efx_ef10_ev_fini(channel);
2414 	return rc;
2415 }
2416 
2417 static void efx_ef10_ev_remove(struct efx_channel *channel)
2418 {
2419 	efx_nic_free_buffer(channel->efx, &channel->eventq.buf);
2420 }
2421 
2422 static void efx_ef10_handle_rx_wrong_queue(struct efx_rx_queue *rx_queue,
2423 					   unsigned int rx_queue_label)
2424 {
2425 	struct efx_nic *efx = rx_queue->efx;
2426 
2427 	netif_info(efx, hw, efx->net_dev,
2428 		   "rx event arrived on queue %d labeled as queue %u\n",
2429 		   efx_rx_queue_index(rx_queue), rx_queue_label);
2430 
2431 	efx_schedule_reset(efx, RESET_TYPE_DISABLE);
2432 }
2433 
2434 static void
2435 efx_ef10_handle_rx_bad_lbits(struct efx_rx_queue *rx_queue,
2436 			     unsigned int actual, unsigned int expected)
2437 {
2438 	unsigned int dropped = (actual - expected) & rx_queue->ptr_mask;
2439 	struct efx_nic *efx = rx_queue->efx;
2440 
2441 	netif_info(efx, hw, efx->net_dev,
2442 		   "dropped %d events (index=%d expected=%d)\n",
2443 		   dropped, actual, expected);
2444 
2445 	efx_schedule_reset(efx, RESET_TYPE_DISABLE);
2446 }
2447 
2448 /* partially received RX was aborted. clean up. */
2449 static void efx_ef10_handle_rx_abort(struct efx_rx_queue *rx_queue)
2450 {
2451 	unsigned int rx_desc_ptr;
2452 
2453 	netif_dbg(rx_queue->efx, hw, rx_queue->efx->net_dev,
2454 		  "scattered RX aborted (dropping %u buffers)\n",
2455 		  rx_queue->scatter_n);
2456 
2457 	rx_desc_ptr = rx_queue->removed_count & rx_queue->ptr_mask;
2458 
2459 	efx_rx_packet(rx_queue, rx_desc_ptr, rx_queue->scatter_n,
2460 		      0, EFX_RX_PKT_DISCARD);
2461 
2462 	rx_queue->removed_count += rx_queue->scatter_n;
2463 	rx_queue->scatter_n = 0;
2464 	rx_queue->scatter_len = 0;
2465 	++efx_rx_queue_channel(rx_queue)->n_rx_nodesc_trunc;
2466 }
2467 
2468 static int efx_ef10_handle_rx_event(struct efx_channel *channel,
2469 				    const efx_qword_t *event)
2470 {
2471 	unsigned int rx_bytes, next_ptr_lbits, rx_queue_label, rx_l4_class;
2472 	unsigned int n_descs, n_packets, i;
2473 	struct efx_nic *efx = channel->efx;
2474 	struct efx_rx_queue *rx_queue;
2475 	bool rx_cont;
2476 	u16 flags = 0;
2477 
2478 	if (unlikely(ACCESS_ONCE(efx->reset_pending)))
2479 		return 0;
2480 
2481 	/* Basic packet information */
2482 	rx_bytes = EFX_QWORD_FIELD(*event, ESF_DZ_RX_BYTES);
2483 	next_ptr_lbits = EFX_QWORD_FIELD(*event, ESF_DZ_RX_DSC_PTR_LBITS);
2484 	rx_queue_label = EFX_QWORD_FIELD(*event, ESF_DZ_RX_QLABEL);
2485 	rx_l4_class = EFX_QWORD_FIELD(*event, ESF_DZ_RX_L4_CLASS);
2486 	rx_cont = EFX_QWORD_FIELD(*event, ESF_DZ_RX_CONT);
2487 
2488 	if (EFX_QWORD_FIELD(*event, ESF_DZ_RX_DROP_EVENT))
2489 		netdev_WARN(efx->net_dev, "saw RX_DROP_EVENT: event="
2490 			    EFX_QWORD_FMT "\n",
2491 			    EFX_QWORD_VAL(*event));
2492 
2493 	rx_queue = efx_channel_get_rx_queue(channel);
2494 
2495 	if (unlikely(rx_queue_label != efx_rx_queue_index(rx_queue)))
2496 		efx_ef10_handle_rx_wrong_queue(rx_queue, rx_queue_label);
2497 
2498 	n_descs = ((next_ptr_lbits - rx_queue->removed_count) &
2499 		   ((1 << ESF_DZ_RX_DSC_PTR_LBITS_WIDTH) - 1));
2500 
2501 	if (n_descs != rx_queue->scatter_n + 1) {
2502 		struct efx_ef10_nic_data *nic_data = efx->nic_data;
2503 
2504 		/* detect rx abort */
2505 		if (unlikely(n_descs == rx_queue->scatter_n)) {
2506 			if (rx_queue->scatter_n == 0 || rx_bytes != 0)
2507 				netdev_WARN(efx->net_dev,
2508 					    "invalid RX abort: scatter_n=%u event="
2509 					    EFX_QWORD_FMT "\n",
2510 					    rx_queue->scatter_n,
2511 					    EFX_QWORD_VAL(*event));
2512 			efx_ef10_handle_rx_abort(rx_queue);
2513 			return 0;
2514 		}
2515 
2516 		/* Check that RX completion merging is valid, i.e.
2517 		 * the current firmware supports it and this is a
2518 		 * non-scattered packet.
2519 		 */
2520 		if (!(nic_data->datapath_caps &
2521 		      (1 << MC_CMD_GET_CAPABILITIES_OUT_RX_BATCHING_LBN)) ||
2522 		    rx_queue->scatter_n != 0 || rx_cont) {
2523 			efx_ef10_handle_rx_bad_lbits(
2524 				rx_queue, next_ptr_lbits,
2525 				(rx_queue->removed_count +
2526 				 rx_queue->scatter_n + 1) &
2527 				((1 << ESF_DZ_RX_DSC_PTR_LBITS_WIDTH) - 1));
2528 			return 0;
2529 		}
2530 
2531 		/* Merged completion for multiple non-scattered packets */
2532 		rx_queue->scatter_n = 1;
2533 		rx_queue->scatter_len = 0;
2534 		n_packets = n_descs;
2535 		++channel->n_rx_merge_events;
2536 		channel->n_rx_merge_packets += n_packets;
2537 		flags |= EFX_RX_PKT_PREFIX_LEN;
2538 	} else {
2539 		++rx_queue->scatter_n;
2540 		rx_queue->scatter_len += rx_bytes;
2541 		if (rx_cont)
2542 			return 0;
2543 		n_packets = 1;
2544 	}
2545 
2546 	if (unlikely(EFX_QWORD_FIELD(*event, ESF_DZ_RX_ECRC_ERR)))
2547 		flags |= EFX_RX_PKT_DISCARD;
2548 
2549 	if (unlikely(EFX_QWORD_FIELD(*event, ESF_DZ_RX_IPCKSUM_ERR))) {
2550 		channel->n_rx_ip_hdr_chksum_err += n_packets;
2551 	} else if (unlikely(EFX_QWORD_FIELD(*event,
2552 					    ESF_DZ_RX_TCPUDP_CKSUM_ERR))) {
2553 		channel->n_rx_tcp_udp_chksum_err += n_packets;
2554 	} else if (rx_l4_class == ESE_DZ_L4_CLASS_TCP ||
2555 		   rx_l4_class == ESE_DZ_L4_CLASS_UDP) {
2556 		flags |= EFX_RX_PKT_CSUMMED;
2557 	}
2558 
2559 	if (rx_l4_class == ESE_DZ_L4_CLASS_TCP)
2560 		flags |= EFX_RX_PKT_TCP;
2561 
2562 	channel->irq_mod_score += 2 * n_packets;
2563 
2564 	/* Handle received packet(s) */
2565 	for (i = 0; i < n_packets; i++) {
2566 		efx_rx_packet(rx_queue,
2567 			      rx_queue->removed_count & rx_queue->ptr_mask,
2568 			      rx_queue->scatter_n, rx_queue->scatter_len,
2569 			      flags);
2570 		rx_queue->removed_count += rx_queue->scatter_n;
2571 	}
2572 
2573 	rx_queue->scatter_n = 0;
2574 	rx_queue->scatter_len = 0;
2575 
2576 	return n_packets;
2577 }
2578 
2579 static int
2580 efx_ef10_handle_tx_event(struct efx_channel *channel, efx_qword_t *event)
2581 {
2582 	struct efx_nic *efx = channel->efx;
2583 	struct efx_tx_queue *tx_queue;
2584 	unsigned int tx_ev_desc_ptr;
2585 	unsigned int tx_ev_q_label;
2586 	int tx_descs = 0;
2587 
2588 	if (unlikely(ACCESS_ONCE(efx->reset_pending)))
2589 		return 0;
2590 
2591 	if (unlikely(EFX_QWORD_FIELD(*event, ESF_DZ_TX_DROP_EVENT)))
2592 		return 0;
2593 
2594 	/* Transmit completion */
2595 	tx_ev_desc_ptr = EFX_QWORD_FIELD(*event, ESF_DZ_TX_DESCR_INDX);
2596 	tx_ev_q_label = EFX_QWORD_FIELD(*event, ESF_DZ_TX_QLABEL);
2597 	tx_queue = efx_channel_get_tx_queue(channel,
2598 					    tx_ev_q_label % EFX_TXQ_TYPES);
2599 	tx_descs = ((tx_ev_desc_ptr + 1 - tx_queue->read_count) &
2600 		    tx_queue->ptr_mask);
2601 	efx_xmit_done(tx_queue, tx_ev_desc_ptr & tx_queue->ptr_mask);
2602 
2603 	return tx_descs;
2604 }
2605 
2606 static void
2607 efx_ef10_handle_driver_event(struct efx_channel *channel, efx_qword_t *event)
2608 {
2609 	struct efx_nic *efx = channel->efx;
2610 	int subcode;
2611 
2612 	subcode = EFX_QWORD_FIELD(*event, ESF_DZ_DRV_SUB_CODE);
2613 
2614 	switch (subcode) {
2615 	case ESE_DZ_DRV_TIMER_EV:
2616 	case ESE_DZ_DRV_WAKE_UP_EV:
2617 		break;
2618 	case ESE_DZ_DRV_START_UP_EV:
2619 		/* event queue init complete. ok. */
2620 		break;
2621 	default:
2622 		netif_err(efx, hw, efx->net_dev,
2623 			  "channel %d unknown driver event type %d"
2624 			  " (data " EFX_QWORD_FMT ")\n",
2625 			  channel->channel, subcode,
2626 			  EFX_QWORD_VAL(*event));
2627 
2628 	}
2629 }
2630 
2631 static void efx_ef10_handle_driver_generated_event(struct efx_channel *channel,
2632 						   efx_qword_t *event)
2633 {
2634 	struct efx_nic *efx = channel->efx;
2635 	u32 subcode;
2636 
2637 	subcode = EFX_QWORD_FIELD(*event, EFX_DWORD_0);
2638 
2639 	switch (subcode) {
2640 	case EFX_EF10_TEST:
2641 		channel->event_test_cpu = raw_smp_processor_id();
2642 		break;
2643 	case EFX_EF10_REFILL:
2644 		/* The queue must be empty, so we won't receive any rx
2645 		 * events, so efx_process_channel() won't refill the
2646 		 * queue. Refill it here
2647 		 */
2648 		efx_fast_push_rx_descriptors(&channel->rx_queue, true);
2649 		break;
2650 	default:
2651 		netif_err(efx, hw, efx->net_dev,
2652 			  "channel %d unknown driver event type %u"
2653 			  " (data " EFX_QWORD_FMT ")\n",
2654 			  channel->channel, (unsigned) subcode,
2655 			  EFX_QWORD_VAL(*event));
2656 	}
2657 }
2658 
2659 static int efx_ef10_ev_process(struct efx_channel *channel, int quota)
2660 {
2661 	struct efx_nic *efx = channel->efx;
2662 	efx_qword_t event, *p_event;
2663 	unsigned int read_ptr;
2664 	int ev_code;
2665 	int tx_descs = 0;
2666 	int spent = 0;
2667 
2668 	if (quota <= 0)
2669 		return spent;
2670 
2671 	read_ptr = channel->eventq_read_ptr;
2672 
2673 	for (;;) {
2674 		p_event = efx_event(channel, read_ptr);
2675 		event = *p_event;
2676 
2677 		if (!efx_event_present(&event))
2678 			break;
2679 
2680 		EFX_SET_QWORD(*p_event);
2681 
2682 		++read_ptr;
2683 
2684 		ev_code = EFX_QWORD_FIELD(event, ESF_DZ_EV_CODE);
2685 
2686 		netif_vdbg(efx, drv, efx->net_dev,
2687 			   "processing event on %d " EFX_QWORD_FMT "\n",
2688 			   channel->channel, EFX_QWORD_VAL(event));
2689 
2690 		switch (ev_code) {
2691 		case ESE_DZ_EV_CODE_MCDI_EV:
2692 			efx_mcdi_process_event(channel, &event);
2693 			break;
2694 		case ESE_DZ_EV_CODE_RX_EV:
2695 			spent += efx_ef10_handle_rx_event(channel, &event);
2696 			if (spent >= quota) {
2697 				/* XXX can we split a merged event to
2698 				 * avoid going over-quota?
2699 				 */
2700 				spent = quota;
2701 				goto out;
2702 			}
2703 			break;
2704 		case ESE_DZ_EV_CODE_TX_EV:
2705 			tx_descs += efx_ef10_handle_tx_event(channel, &event);
2706 			if (tx_descs > efx->txq_entries) {
2707 				spent = quota;
2708 				goto out;
2709 			} else if (++spent == quota) {
2710 				goto out;
2711 			}
2712 			break;
2713 		case ESE_DZ_EV_CODE_DRIVER_EV:
2714 			efx_ef10_handle_driver_event(channel, &event);
2715 			if (++spent == quota)
2716 				goto out;
2717 			break;
2718 		case EFX_EF10_DRVGEN_EV:
2719 			efx_ef10_handle_driver_generated_event(channel, &event);
2720 			break;
2721 		default:
2722 			netif_err(efx, hw, efx->net_dev,
2723 				  "channel %d unknown event type %d"
2724 				  " (data " EFX_QWORD_FMT ")\n",
2725 				  channel->channel, ev_code,
2726 				  EFX_QWORD_VAL(event));
2727 		}
2728 	}
2729 
2730 out:
2731 	channel->eventq_read_ptr = read_ptr;
2732 	return spent;
2733 }
2734 
2735 static void efx_ef10_ev_read_ack(struct efx_channel *channel)
2736 {
2737 	struct efx_nic *efx = channel->efx;
2738 	efx_dword_t rptr;
2739 
2740 	if (EFX_EF10_WORKAROUND_35388(efx)) {
2741 		BUILD_BUG_ON(EFX_MIN_EVQ_SIZE <
2742 			     (1 << ERF_DD_EVQ_IND_RPTR_WIDTH));
2743 		BUILD_BUG_ON(EFX_MAX_EVQ_SIZE >
2744 			     (1 << 2 * ERF_DD_EVQ_IND_RPTR_WIDTH));
2745 
2746 		EFX_POPULATE_DWORD_2(rptr, ERF_DD_EVQ_IND_RPTR_FLAGS,
2747 				     EFE_DD_EVQ_IND_RPTR_FLAGS_HIGH,
2748 				     ERF_DD_EVQ_IND_RPTR,
2749 				     (channel->eventq_read_ptr &
2750 				      channel->eventq_mask) >>
2751 				     ERF_DD_EVQ_IND_RPTR_WIDTH);
2752 		efx_writed_page(efx, &rptr, ER_DD_EVQ_INDIRECT,
2753 				channel->channel);
2754 		EFX_POPULATE_DWORD_2(rptr, ERF_DD_EVQ_IND_RPTR_FLAGS,
2755 				     EFE_DD_EVQ_IND_RPTR_FLAGS_LOW,
2756 				     ERF_DD_EVQ_IND_RPTR,
2757 				     channel->eventq_read_ptr &
2758 				     ((1 << ERF_DD_EVQ_IND_RPTR_WIDTH) - 1));
2759 		efx_writed_page(efx, &rptr, ER_DD_EVQ_INDIRECT,
2760 				channel->channel);
2761 	} else {
2762 		EFX_POPULATE_DWORD_1(rptr, ERF_DZ_EVQ_RPTR,
2763 				     channel->eventq_read_ptr &
2764 				     channel->eventq_mask);
2765 		efx_writed_page(efx, &rptr, ER_DZ_EVQ_RPTR, channel->channel);
2766 	}
2767 }
2768 
2769 static void efx_ef10_ev_test_generate(struct efx_channel *channel)
2770 {
2771 	MCDI_DECLARE_BUF(inbuf, MC_CMD_DRIVER_EVENT_IN_LEN);
2772 	struct efx_nic *efx = channel->efx;
2773 	efx_qword_t event;
2774 	int rc;
2775 
2776 	EFX_POPULATE_QWORD_2(event,
2777 			     ESF_DZ_EV_CODE, EFX_EF10_DRVGEN_EV,
2778 			     ESF_DZ_EV_DATA, EFX_EF10_TEST);
2779 
2780 	MCDI_SET_DWORD(inbuf, DRIVER_EVENT_IN_EVQ, channel->channel);
2781 
2782 	/* MCDI_SET_QWORD is not appropriate here since EFX_POPULATE_* has
2783 	 * already swapped the data to little-endian order.
2784 	 */
2785 	memcpy(MCDI_PTR(inbuf, DRIVER_EVENT_IN_DATA), &event.u64[0],
2786 	       sizeof(efx_qword_t));
2787 
2788 	rc = efx_mcdi_rpc(efx, MC_CMD_DRIVER_EVENT, inbuf, sizeof(inbuf),
2789 			  NULL, 0, NULL);
2790 	if (rc != 0)
2791 		goto fail;
2792 
2793 	return;
2794 
2795 fail:
2796 	WARN_ON(true);
2797 	netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
2798 }
2799 
2800 void efx_ef10_handle_drain_event(struct efx_nic *efx)
2801 {
2802 	if (atomic_dec_and_test(&efx->active_queues))
2803 		wake_up(&efx->flush_wq);
2804 
2805 	WARN_ON(atomic_read(&efx->active_queues) < 0);
2806 }
2807 
2808 static int efx_ef10_fini_dmaq(struct efx_nic *efx)
2809 {
2810 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2811 	struct efx_channel *channel;
2812 	struct efx_tx_queue *tx_queue;
2813 	struct efx_rx_queue *rx_queue;
2814 	int pending;
2815 
2816 	/* If the MC has just rebooted, the TX/RX queues will have already been
2817 	 * torn down, but efx->active_queues needs to be set to zero.
2818 	 */
2819 	if (nic_data->must_realloc_vis) {
2820 		atomic_set(&efx->active_queues, 0);
2821 		return 0;
2822 	}
2823 
2824 	/* Do not attempt to write to the NIC during EEH recovery */
2825 	if (efx->state != STATE_RECOVERY) {
2826 		efx_for_each_channel(channel, efx) {
2827 			efx_for_each_channel_rx_queue(rx_queue, channel)
2828 				efx_ef10_rx_fini(rx_queue);
2829 			efx_for_each_channel_tx_queue(tx_queue, channel)
2830 				efx_ef10_tx_fini(tx_queue);
2831 		}
2832 
2833 		wait_event_timeout(efx->flush_wq,
2834 				   atomic_read(&efx->active_queues) == 0,
2835 				   msecs_to_jiffies(EFX_MAX_FLUSH_TIME));
2836 		pending = atomic_read(&efx->active_queues);
2837 		if (pending) {
2838 			netif_err(efx, hw, efx->net_dev, "failed to flush %d queues\n",
2839 				  pending);
2840 			return -ETIMEDOUT;
2841 		}
2842 	}
2843 
2844 	return 0;
2845 }
2846 
2847 static void efx_ef10_prepare_flr(struct efx_nic *efx)
2848 {
2849 	atomic_set(&efx->active_queues, 0);
2850 }
2851 
2852 static bool efx_ef10_filter_equal(const struct efx_filter_spec *left,
2853 				  const struct efx_filter_spec *right)
2854 {
2855 	if ((left->match_flags ^ right->match_flags) |
2856 	    ((left->flags ^ right->flags) &
2857 	     (EFX_FILTER_FLAG_RX | EFX_FILTER_FLAG_TX)))
2858 		return false;
2859 
2860 	return memcmp(&left->outer_vid, &right->outer_vid,
2861 		      sizeof(struct efx_filter_spec) -
2862 		      offsetof(struct efx_filter_spec, outer_vid)) == 0;
2863 }
2864 
2865 static unsigned int efx_ef10_filter_hash(const struct efx_filter_spec *spec)
2866 {
2867 	BUILD_BUG_ON(offsetof(struct efx_filter_spec, outer_vid) & 3);
2868 	return jhash2((const u32 *)&spec->outer_vid,
2869 		      (sizeof(struct efx_filter_spec) -
2870 		       offsetof(struct efx_filter_spec, outer_vid)) / 4,
2871 		      0);
2872 	/* XXX should we randomise the initval? */
2873 }
2874 
2875 /* Decide whether a filter should be exclusive or else should allow
2876  * delivery to additional recipients.  Currently we decide that
2877  * filters for specific local unicast MAC and IP addresses are
2878  * exclusive.
2879  */
2880 static bool efx_ef10_filter_is_exclusive(const struct efx_filter_spec *spec)
2881 {
2882 	if (spec->match_flags & EFX_FILTER_MATCH_LOC_MAC &&
2883 	    !is_multicast_ether_addr(spec->loc_mac))
2884 		return true;
2885 
2886 	if ((spec->match_flags &
2887 	     (EFX_FILTER_MATCH_ETHER_TYPE | EFX_FILTER_MATCH_LOC_HOST)) ==
2888 	    (EFX_FILTER_MATCH_ETHER_TYPE | EFX_FILTER_MATCH_LOC_HOST)) {
2889 		if (spec->ether_type == htons(ETH_P_IP) &&
2890 		    !ipv4_is_multicast(spec->loc_host[0]))
2891 			return true;
2892 		if (spec->ether_type == htons(ETH_P_IPV6) &&
2893 		    ((const u8 *)spec->loc_host)[0] != 0xff)
2894 			return true;
2895 	}
2896 
2897 	return false;
2898 }
2899 
2900 static struct efx_filter_spec *
2901 efx_ef10_filter_entry_spec(const struct efx_ef10_filter_table *table,
2902 			   unsigned int filter_idx)
2903 {
2904 	return (struct efx_filter_spec *)(table->entry[filter_idx].spec &
2905 					  ~EFX_EF10_FILTER_FLAGS);
2906 }
2907 
2908 static unsigned int
2909 efx_ef10_filter_entry_flags(const struct efx_ef10_filter_table *table,
2910 			   unsigned int filter_idx)
2911 {
2912 	return table->entry[filter_idx].spec & EFX_EF10_FILTER_FLAGS;
2913 }
2914 
2915 static void
2916 efx_ef10_filter_set_entry(struct efx_ef10_filter_table *table,
2917 			  unsigned int filter_idx,
2918 			  const struct efx_filter_spec *spec,
2919 			  unsigned int flags)
2920 {
2921 	table->entry[filter_idx].spec =	(unsigned long)spec | flags;
2922 }
2923 
2924 static void efx_ef10_filter_push_prep(struct efx_nic *efx,
2925 				      const struct efx_filter_spec *spec,
2926 				      efx_dword_t *inbuf, u64 handle,
2927 				      bool replacing)
2928 {
2929 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2930 	u32 flags = spec->flags;
2931 
2932 	memset(inbuf, 0, MC_CMD_FILTER_OP_IN_LEN);
2933 
2934 	/* Remove RSS flag if we don't have an RSS context. */
2935 	if (flags & EFX_FILTER_FLAG_RX_RSS &&
2936 	    spec->rss_context == EFX_FILTER_RSS_CONTEXT_DEFAULT &&
2937 	    nic_data->rx_rss_context == EFX_EF10_RSS_CONTEXT_INVALID)
2938 		flags &= ~EFX_FILTER_FLAG_RX_RSS;
2939 
2940 	if (replacing) {
2941 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
2942 			       MC_CMD_FILTER_OP_IN_OP_REPLACE);
2943 		MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE, handle);
2944 	} else {
2945 		u32 match_fields = 0;
2946 
2947 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
2948 			       efx_ef10_filter_is_exclusive(spec) ?
2949 			       MC_CMD_FILTER_OP_IN_OP_INSERT :
2950 			       MC_CMD_FILTER_OP_IN_OP_SUBSCRIBE);
2951 
2952 		/* Convert match flags and values.  Unlike almost
2953 		 * everything else in MCDI, these fields are in
2954 		 * network byte order.
2955 		 */
2956 		if (spec->match_flags & EFX_FILTER_MATCH_LOC_MAC_IG)
2957 			match_fields |=
2958 				is_multicast_ether_addr(spec->loc_mac) ?
2959 				1 << MC_CMD_FILTER_OP_IN_MATCH_UNKNOWN_MCAST_DST_LBN :
2960 				1 << MC_CMD_FILTER_OP_IN_MATCH_UNKNOWN_UCAST_DST_LBN;
2961 #define COPY_FIELD(gen_flag, gen_field, mcdi_field)			     \
2962 		if (spec->match_flags & EFX_FILTER_MATCH_ ## gen_flag) {     \
2963 			match_fields |=					     \
2964 				1 << MC_CMD_FILTER_OP_IN_MATCH_ ##	     \
2965 				mcdi_field ## _LBN;			     \
2966 			BUILD_BUG_ON(					     \
2967 				MC_CMD_FILTER_OP_IN_ ## mcdi_field ## _LEN < \
2968 				sizeof(spec->gen_field));		     \
2969 			memcpy(MCDI_PTR(inbuf, FILTER_OP_IN_ ##	mcdi_field), \
2970 			       &spec->gen_field, sizeof(spec->gen_field));   \
2971 		}
2972 		COPY_FIELD(REM_HOST, rem_host, SRC_IP);
2973 		COPY_FIELD(LOC_HOST, loc_host, DST_IP);
2974 		COPY_FIELD(REM_MAC, rem_mac, SRC_MAC);
2975 		COPY_FIELD(REM_PORT, rem_port, SRC_PORT);
2976 		COPY_FIELD(LOC_MAC, loc_mac, DST_MAC);
2977 		COPY_FIELD(LOC_PORT, loc_port, DST_PORT);
2978 		COPY_FIELD(ETHER_TYPE, ether_type, ETHER_TYPE);
2979 		COPY_FIELD(INNER_VID, inner_vid, INNER_VLAN);
2980 		COPY_FIELD(OUTER_VID, outer_vid, OUTER_VLAN);
2981 		COPY_FIELD(IP_PROTO, ip_proto, IP_PROTO);
2982 #undef COPY_FIELD
2983 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_MATCH_FIELDS,
2984 			       match_fields);
2985 	}
2986 
2987 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_PORT_ID, nic_data->vport_id);
2988 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_DEST,
2989 		       spec->dmaq_id == EFX_FILTER_RX_DMAQ_ID_DROP ?
2990 		       MC_CMD_FILTER_OP_IN_RX_DEST_DROP :
2991 		       MC_CMD_FILTER_OP_IN_RX_DEST_HOST);
2992 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_TX_DOMAIN, 0);
2993 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_TX_DEST,
2994 		       MC_CMD_FILTER_OP_IN_TX_DEST_DEFAULT);
2995 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_QUEUE,
2996 		       spec->dmaq_id == EFX_FILTER_RX_DMAQ_ID_DROP ?
2997 		       0 : spec->dmaq_id);
2998 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_MODE,
2999 		       (flags & EFX_FILTER_FLAG_RX_RSS) ?
3000 		       MC_CMD_FILTER_OP_IN_RX_MODE_RSS :
3001 		       MC_CMD_FILTER_OP_IN_RX_MODE_SIMPLE);
3002 	if (flags & EFX_FILTER_FLAG_RX_RSS)
3003 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_CONTEXT,
3004 			       spec->rss_context !=
3005 			       EFX_FILTER_RSS_CONTEXT_DEFAULT ?
3006 			       spec->rss_context : nic_data->rx_rss_context);
3007 }
3008 
3009 static int efx_ef10_filter_push(struct efx_nic *efx,
3010 				const struct efx_filter_spec *spec,
3011 				u64 *handle, bool replacing)
3012 {
3013 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_IN_LEN);
3014 	MCDI_DECLARE_BUF(outbuf, MC_CMD_FILTER_OP_OUT_LEN);
3015 	int rc;
3016 
3017 	efx_ef10_filter_push_prep(efx, spec, inbuf, *handle, replacing);
3018 	rc = efx_mcdi_rpc(efx, MC_CMD_FILTER_OP, inbuf, sizeof(inbuf),
3019 			  outbuf, sizeof(outbuf), NULL);
3020 	if (rc == 0)
3021 		*handle = MCDI_QWORD(outbuf, FILTER_OP_OUT_HANDLE);
3022 	if (rc == -ENOSPC)
3023 		rc = -EBUSY; /* to match efx_farch_filter_insert() */
3024 	return rc;
3025 }
3026 
3027 static int efx_ef10_filter_rx_match_pri(struct efx_ef10_filter_table *table,
3028 					enum efx_filter_match_flags match_flags)
3029 {
3030 	unsigned int match_pri;
3031 
3032 	for (match_pri = 0;
3033 	     match_pri < table->rx_match_count;
3034 	     match_pri++)
3035 		if (table->rx_match_flags[match_pri] == match_flags)
3036 			return match_pri;
3037 
3038 	return -EPROTONOSUPPORT;
3039 }
3040 
3041 static s32 efx_ef10_filter_insert(struct efx_nic *efx,
3042 				  struct efx_filter_spec *spec,
3043 				  bool replace_equal)
3044 {
3045 	struct efx_ef10_filter_table *table = efx->filter_state;
3046 	DECLARE_BITMAP(mc_rem_map, EFX_EF10_FILTER_SEARCH_LIMIT);
3047 	struct efx_filter_spec *saved_spec;
3048 	unsigned int match_pri, hash;
3049 	unsigned int priv_flags;
3050 	bool replacing = false;
3051 	int ins_index = -1;
3052 	DEFINE_WAIT(wait);
3053 	bool is_mc_recip;
3054 	s32 rc;
3055 
3056 	/* For now, only support RX filters */
3057 	if ((spec->flags & (EFX_FILTER_FLAG_RX | EFX_FILTER_FLAG_TX)) !=
3058 	    EFX_FILTER_FLAG_RX)
3059 		return -EINVAL;
3060 
3061 	rc = efx_ef10_filter_rx_match_pri(table, spec->match_flags);
3062 	if (rc < 0)
3063 		return rc;
3064 	match_pri = rc;
3065 
3066 	hash = efx_ef10_filter_hash(spec);
3067 	is_mc_recip = efx_filter_is_mc_recipient(spec);
3068 	if (is_mc_recip)
3069 		bitmap_zero(mc_rem_map, EFX_EF10_FILTER_SEARCH_LIMIT);
3070 
3071 	/* Find any existing filters with the same match tuple or
3072 	 * else a free slot to insert at.  If any of them are busy,
3073 	 * we have to wait and retry.
3074 	 */
3075 	for (;;) {
3076 		unsigned int depth = 1;
3077 		unsigned int i;
3078 
3079 		spin_lock_bh(&efx->filter_lock);
3080 
3081 		for (;;) {
3082 			i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
3083 			saved_spec = efx_ef10_filter_entry_spec(table, i);
3084 
3085 			if (!saved_spec) {
3086 				if (ins_index < 0)
3087 					ins_index = i;
3088 			} else if (efx_ef10_filter_equal(spec, saved_spec)) {
3089 				if (table->entry[i].spec &
3090 				    EFX_EF10_FILTER_FLAG_BUSY)
3091 					break;
3092 				if (spec->priority < saved_spec->priority &&
3093 				    spec->priority != EFX_FILTER_PRI_AUTO) {
3094 					rc = -EPERM;
3095 					goto out_unlock;
3096 				}
3097 				if (!is_mc_recip) {
3098 					/* This is the only one */
3099 					if (spec->priority ==
3100 					    saved_spec->priority &&
3101 					    !replace_equal) {
3102 						rc = -EEXIST;
3103 						goto out_unlock;
3104 					}
3105 					ins_index = i;
3106 					goto found;
3107 				} else if (spec->priority >
3108 					   saved_spec->priority ||
3109 					   (spec->priority ==
3110 					    saved_spec->priority &&
3111 					    replace_equal)) {
3112 					if (ins_index < 0)
3113 						ins_index = i;
3114 					else
3115 						__set_bit(depth, mc_rem_map);
3116 				}
3117 			}
3118 
3119 			/* Once we reach the maximum search depth, use
3120 			 * the first suitable slot or return -EBUSY if
3121 			 * there was none
3122 			 */
3123 			if (depth == EFX_EF10_FILTER_SEARCH_LIMIT) {
3124 				if (ins_index < 0) {
3125 					rc = -EBUSY;
3126 					goto out_unlock;
3127 				}
3128 				goto found;
3129 			}
3130 
3131 			++depth;
3132 		}
3133 
3134 		prepare_to_wait(&table->waitq, &wait, TASK_UNINTERRUPTIBLE);
3135 		spin_unlock_bh(&efx->filter_lock);
3136 		schedule();
3137 	}
3138 
3139 found:
3140 	/* Create a software table entry if necessary, and mark it
3141 	 * busy.  We might yet fail to insert, but any attempt to
3142 	 * insert a conflicting filter while we're waiting for the
3143 	 * firmware must find the busy entry.
3144 	 */
3145 	saved_spec = efx_ef10_filter_entry_spec(table, ins_index);
3146 	if (saved_spec) {
3147 		if (spec->priority == EFX_FILTER_PRI_AUTO &&
3148 		    saved_spec->priority >= EFX_FILTER_PRI_AUTO) {
3149 			/* Just make sure it won't be removed */
3150 			if (saved_spec->priority > EFX_FILTER_PRI_AUTO)
3151 				saved_spec->flags |= EFX_FILTER_FLAG_RX_OVER_AUTO;
3152 			table->entry[ins_index].spec &=
3153 				~EFX_EF10_FILTER_FLAG_AUTO_OLD;
3154 			rc = ins_index;
3155 			goto out_unlock;
3156 		}
3157 		replacing = true;
3158 		priv_flags = efx_ef10_filter_entry_flags(table, ins_index);
3159 	} else {
3160 		saved_spec = kmalloc(sizeof(*spec), GFP_ATOMIC);
3161 		if (!saved_spec) {
3162 			rc = -ENOMEM;
3163 			goto out_unlock;
3164 		}
3165 		*saved_spec = *spec;
3166 		priv_flags = 0;
3167 	}
3168 	efx_ef10_filter_set_entry(table, ins_index, saved_spec,
3169 				  priv_flags | EFX_EF10_FILTER_FLAG_BUSY);
3170 
3171 	/* Mark lower-priority multicast recipients busy prior to removal */
3172 	if (is_mc_recip) {
3173 		unsigned int depth, i;
3174 
3175 		for (depth = 0; depth < EFX_EF10_FILTER_SEARCH_LIMIT; depth++) {
3176 			i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
3177 			if (test_bit(depth, mc_rem_map))
3178 				table->entry[i].spec |=
3179 					EFX_EF10_FILTER_FLAG_BUSY;
3180 		}
3181 	}
3182 
3183 	spin_unlock_bh(&efx->filter_lock);
3184 
3185 	rc = efx_ef10_filter_push(efx, spec, &table->entry[ins_index].handle,
3186 				  replacing);
3187 
3188 	/* Finalise the software table entry */
3189 	spin_lock_bh(&efx->filter_lock);
3190 	if (rc == 0) {
3191 		if (replacing) {
3192 			/* Update the fields that may differ */
3193 			if (saved_spec->priority == EFX_FILTER_PRI_AUTO)
3194 				saved_spec->flags |=
3195 					EFX_FILTER_FLAG_RX_OVER_AUTO;
3196 			saved_spec->priority = spec->priority;
3197 			saved_spec->flags &= EFX_FILTER_FLAG_RX_OVER_AUTO;
3198 			saved_spec->flags |= spec->flags;
3199 			saved_spec->rss_context = spec->rss_context;
3200 			saved_spec->dmaq_id = spec->dmaq_id;
3201 		}
3202 	} else if (!replacing) {
3203 		kfree(saved_spec);
3204 		saved_spec = NULL;
3205 	}
3206 	efx_ef10_filter_set_entry(table, ins_index, saved_spec, priv_flags);
3207 
3208 	/* Remove and finalise entries for lower-priority multicast
3209 	 * recipients
3210 	 */
3211 	if (is_mc_recip) {
3212 		MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_IN_LEN);
3213 		unsigned int depth, i;
3214 
3215 		memset(inbuf, 0, sizeof(inbuf));
3216 
3217 		for (depth = 0; depth < EFX_EF10_FILTER_SEARCH_LIMIT; depth++) {
3218 			if (!test_bit(depth, mc_rem_map))
3219 				continue;
3220 
3221 			i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
3222 			saved_spec = efx_ef10_filter_entry_spec(table, i);
3223 			priv_flags = efx_ef10_filter_entry_flags(table, i);
3224 
3225 			if (rc == 0) {
3226 				spin_unlock_bh(&efx->filter_lock);
3227 				MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
3228 					       MC_CMD_FILTER_OP_IN_OP_UNSUBSCRIBE);
3229 				MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
3230 					       table->entry[i].handle);
3231 				rc = efx_mcdi_rpc(efx, MC_CMD_FILTER_OP,
3232 						  inbuf, sizeof(inbuf),
3233 						  NULL, 0, NULL);
3234 				spin_lock_bh(&efx->filter_lock);
3235 			}
3236 
3237 			if (rc == 0) {
3238 				kfree(saved_spec);
3239 				saved_spec = NULL;
3240 				priv_flags = 0;
3241 			} else {
3242 				priv_flags &= ~EFX_EF10_FILTER_FLAG_BUSY;
3243 			}
3244 			efx_ef10_filter_set_entry(table, i, saved_spec,
3245 						  priv_flags);
3246 		}
3247 	}
3248 
3249 	/* If successful, return the inserted filter ID */
3250 	if (rc == 0)
3251 		rc = match_pri * HUNT_FILTER_TBL_ROWS + ins_index;
3252 
3253 	wake_up_all(&table->waitq);
3254 out_unlock:
3255 	spin_unlock_bh(&efx->filter_lock);
3256 	finish_wait(&table->waitq, &wait);
3257 	return rc;
3258 }
3259 
3260 static void efx_ef10_filter_update_rx_scatter(struct efx_nic *efx)
3261 {
3262 	/* no need to do anything here on EF10 */
3263 }
3264 
3265 /* Remove a filter.
3266  * If !by_index, remove by ID
3267  * If by_index, remove by index
3268  * Filter ID may come from userland and must be range-checked.
3269  */
3270 static int efx_ef10_filter_remove_internal(struct efx_nic *efx,
3271 					   unsigned int priority_mask,
3272 					   u32 filter_id, bool by_index)
3273 {
3274 	unsigned int filter_idx = filter_id % HUNT_FILTER_TBL_ROWS;
3275 	struct efx_ef10_filter_table *table = efx->filter_state;
3276 	MCDI_DECLARE_BUF(inbuf,
3277 			 MC_CMD_FILTER_OP_IN_HANDLE_OFST +
3278 			 MC_CMD_FILTER_OP_IN_HANDLE_LEN);
3279 	struct efx_filter_spec *spec;
3280 	DEFINE_WAIT(wait);
3281 	int rc;
3282 
3283 	/* Find the software table entry and mark it busy.  Don't
3284 	 * remove it yet; any attempt to update while we're waiting
3285 	 * for the firmware must find the busy entry.
3286 	 */
3287 	for (;;) {
3288 		spin_lock_bh(&efx->filter_lock);
3289 		if (!(table->entry[filter_idx].spec &
3290 		      EFX_EF10_FILTER_FLAG_BUSY))
3291 			break;
3292 		prepare_to_wait(&table->waitq, &wait, TASK_UNINTERRUPTIBLE);
3293 		spin_unlock_bh(&efx->filter_lock);
3294 		schedule();
3295 	}
3296 
3297 	spec = efx_ef10_filter_entry_spec(table, filter_idx);
3298 	if (!spec ||
3299 	    (!by_index &&
3300 	     efx_ef10_filter_rx_match_pri(table, spec->match_flags) !=
3301 	     filter_id / HUNT_FILTER_TBL_ROWS)) {
3302 		rc = -ENOENT;
3303 		goto out_unlock;
3304 	}
3305 
3306 	if (spec->flags & EFX_FILTER_FLAG_RX_OVER_AUTO &&
3307 	    priority_mask == (1U << EFX_FILTER_PRI_AUTO)) {
3308 		/* Just remove flags */
3309 		spec->flags &= ~EFX_FILTER_FLAG_RX_OVER_AUTO;
3310 		table->entry[filter_idx].spec &= ~EFX_EF10_FILTER_FLAG_AUTO_OLD;
3311 		rc = 0;
3312 		goto out_unlock;
3313 	}
3314 
3315 	if (!(priority_mask & (1U << spec->priority))) {
3316 		rc = -ENOENT;
3317 		goto out_unlock;
3318 	}
3319 
3320 	table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_BUSY;
3321 	spin_unlock_bh(&efx->filter_lock);
3322 
3323 	if (spec->flags & EFX_FILTER_FLAG_RX_OVER_AUTO) {
3324 		/* Reset to an automatic filter */
3325 
3326 		struct efx_filter_spec new_spec = *spec;
3327 
3328 		new_spec.priority = EFX_FILTER_PRI_AUTO;
3329 		new_spec.flags = (EFX_FILTER_FLAG_RX |
3330 				  (efx_rss_enabled(efx) ?
3331 				   EFX_FILTER_FLAG_RX_RSS : 0));
3332 		new_spec.dmaq_id = 0;
3333 		new_spec.rss_context = EFX_FILTER_RSS_CONTEXT_DEFAULT;
3334 		rc = efx_ef10_filter_push(efx, &new_spec,
3335 					  &table->entry[filter_idx].handle,
3336 					  true);
3337 
3338 		spin_lock_bh(&efx->filter_lock);
3339 		if (rc == 0)
3340 			*spec = new_spec;
3341 	} else {
3342 		/* Really remove the filter */
3343 
3344 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
3345 			       efx_ef10_filter_is_exclusive(spec) ?
3346 			       MC_CMD_FILTER_OP_IN_OP_REMOVE :
3347 			       MC_CMD_FILTER_OP_IN_OP_UNSUBSCRIBE);
3348 		MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
3349 			       table->entry[filter_idx].handle);
3350 		rc = efx_mcdi_rpc(efx, MC_CMD_FILTER_OP,
3351 				  inbuf, sizeof(inbuf), NULL, 0, NULL);
3352 
3353 		spin_lock_bh(&efx->filter_lock);
3354 		if (rc == 0) {
3355 			kfree(spec);
3356 			efx_ef10_filter_set_entry(table, filter_idx, NULL, 0);
3357 		}
3358 	}
3359 
3360 	table->entry[filter_idx].spec &= ~EFX_EF10_FILTER_FLAG_BUSY;
3361 	wake_up_all(&table->waitq);
3362 out_unlock:
3363 	spin_unlock_bh(&efx->filter_lock);
3364 	finish_wait(&table->waitq, &wait);
3365 	return rc;
3366 }
3367 
3368 static int efx_ef10_filter_remove_safe(struct efx_nic *efx,
3369 				       enum efx_filter_priority priority,
3370 				       u32 filter_id)
3371 {
3372 	return efx_ef10_filter_remove_internal(efx, 1U << priority,
3373 					       filter_id, false);
3374 }
3375 
3376 static u32 efx_ef10_filter_get_unsafe_id(struct efx_nic *efx, u32 filter_id)
3377 {
3378 	return filter_id % HUNT_FILTER_TBL_ROWS;
3379 }
3380 
3381 static int efx_ef10_filter_remove_unsafe(struct efx_nic *efx,
3382 					 enum efx_filter_priority priority,
3383 					 u32 filter_id)
3384 {
3385 	return efx_ef10_filter_remove_internal(efx, 1U << priority,
3386 					       filter_id, true);
3387 }
3388 
3389 static int efx_ef10_filter_get_safe(struct efx_nic *efx,
3390 				    enum efx_filter_priority priority,
3391 				    u32 filter_id, struct efx_filter_spec *spec)
3392 {
3393 	unsigned int filter_idx = filter_id % HUNT_FILTER_TBL_ROWS;
3394 	struct efx_ef10_filter_table *table = efx->filter_state;
3395 	const struct efx_filter_spec *saved_spec;
3396 	int rc;
3397 
3398 	spin_lock_bh(&efx->filter_lock);
3399 	saved_spec = efx_ef10_filter_entry_spec(table, filter_idx);
3400 	if (saved_spec && saved_spec->priority == priority &&
3401 	    efx_ef10_filter_rx_match_pri(table, saved_spec->match_flags) ==
3402 	    filter_id / HUNT_FILTER_TBL_ROWS) {
3403 		*spec = *saved_spec;
3404 		rc = 0;
3405 	} else {
3406 		rc = -ENOENT;
3407 	}
3408 	spin_unlock_bh(&efx->filter_lock);
3409 	return rc;
3410 }
3411 
3412 static int efx_ef10_filter_clear_rx(struct efx_nic *efx,
3413 				     enum efx_filter_priority priority)
3414 {
3415 	unsigned int priority_mask;
3416 	unsigned int i;
3417 	int rc;
3418 
3419 	priority_mask = (((1U << (priority + 1)) - 1) &
3420 			 ~(1U << EFX_FILTER_PRI_AUTO));
3421 
3422 	for (i = 0; i < HUNT_FILTER_TBL_ROWS; i++) {
3423 		rc = efx_ef10_filter_remove_internal(efx, priority_mask,
3424 						     i, true);
3425 		if (rc && rc != -ENOENT)
3426 			return rc;
3427 	}
3428 
3429 	return 0;
3430 }
3431 
3432 static u32 efx_ef10_filter_count_rx_used(struct efx_nic *efx,
3433 					 enum efx_filter_priority priority)
3434 {
3435 	struct efx_ef10_filter_table *table = efx->filter_state;
3436 	unsigned int filter_idx;
3437 	s32 count = 0;
3438 
3439 	spin_lock_bh(&efx->filter_lock);
3440 	for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
3441 		if (table->entry[filter_idx].spec &&
3442 		    efx_ef10_filter_entry_spec(table, filter_idx)->priority ==
3443 		    priority)
3444 			++count;
3445 	}
3446 	spin_unlock_bh(&efx->filter_lock);
3447 	return count;
3448 }
3449 
3450 static u32 efx_ef10_filter_get_rx_id_limit(struct efx_nic *efx)
3451 {
3452 	struct efx_ef10_filter_table *table = efx->filter_state;
3453 
3454 	return table->rx_match_count * HUNT_FILTER_TBL_ROWS;
3455 }
3456 
3457 static s32 efx_ef10_filter_get_rx_ids(struct efx_nic *efx,
3458 				      enum efx_filter_priority priority,
3459 				      u32 *buf, u32 size)
3460 {
3461 	struct efx_ef10_filter_table *table = efx->filter_state;
3462 	struct efx_filter_spec *spec;
3463 	unsigned int filter_idx;
3464 	s32 count = 0;
3465 
3466 	spin_lock_bh(&efx->filter_lock);
3467 	for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
3468 		spec = efx_ef10_filter_entry_spec(table, filter_idx);
3469 		if (spec && spec->priority == priority) {
3470 			if (count == size) {
3471 				count = -EMSGSIZE;
3472 				break;
3473 			}
3474 			buf[count++] = (efx_ef10_filter_rx_match_pri(
3475 						table, spec->match_flags) *
3476 					HUNT_FILTER_TBL_ROWS +
3477 					filter_idx);
3478 		}
3479 	}
3480 	spin_unlock_bh(&efx->filter_lock);
3481 	return count;
3482 }
3483 
3484 #ifdef CONFIG_RFS_ACCEL
3485 
3486 static efx_mcdi_async_completer efx_ef10_filter_rfs_insert_complete;
3487 
3488 static s32 efx_ef10_filter_rfs_insert(struct efx_nic *efx,
3489 				      struct efx_filter_spec *spec)
3490 {
3491 	struct efx_ef10_filter_table *table = efx->filter_state;
3492 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_IN_LEN);
3493 	struct efx_filter_spec *saved_spec;
3494 	unsigned int hash, i, depth = 1;
3495 	bool replacing = false;
3496 	int ins_index = -1;
3497 	u64 cookie;
3498 	s32 rc;
3499 
3500 	/* Must be an RX filter without RSS and not for a multicast
3501 	 * destination address (RFS only works for connected sockets).
3502 	 * These restrictions allow us to pass only a tiny amount of
3503 	 * data through to the completion function.
3504 	 */
3505 	EFX_WARN_ON_PARANOID(spec->flags !=
3506 			     (EFX_FILTER_FLAG_RX | EFX_FILTER_FLAG_RX_SCATTER));
3507 	EFX_WARN_ON_PARANOID(spec->priority != EFX_FILTER_PRI_HINT);
3508 	EFX_WARN_ON_PARANOID(efx_filter_is_mc_recipient(spec));
3509 
3510 	hash = efx_ef10_filter_hash(spec);
3511 
3512 	spin_lock_bh(&efx->filter_lock);
3513 
3514 	/* Find any existing filter with the same match tuple or else
3515 	 * a free slot to insert at.  If an existing filter is busy,
3516 	 * we have to give up.
3517 	 */
3518 	for (;;) {
3519 		i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
3520 		saved_spec = efx_ef10_filter_entry_spec(table, i);
3521 
3522 		if (!saved_spec) {
3523 			if (ins_index < 0)
3524 				ins_index = i;
3525 		} else if (efx_ef10_filter_equal(spec, saved_spec)) {
3526 			if (table->entry[i].spec & EFX_EF10_FILTER_FLAG_BUSY) {
3527 				rc = -EBUSY;
3528 				goto fail_unlock;
3529 			}
3530 			if (spec->priority < saved_spec->priority) {
3531 				rc = -EPERM;
3532 				goto fail_unlock;
3533 			}
3534 			ins_index = i;
3535 			break;
3536 		}
3537 
3538 		/* Once we reach the maximum search depth, use the
3539 		 * first suitable slot or return -EBUSY if there was
3540 		 * none
3541 		 */
3542 		if (depth == EFX_EF10_FILTER_SEARCH_LIMIT) {
3543 			if (ins_index < 0) {
3544 				rc = -EBUSY;
3545 				goto fail_unlock;
3546 			}
3547 			break;
3548 		}
3549 
3550 		++depth;
3551 	}
3552 
3553 	/* Create a software table entry if necessary, and mark it
3554 	 * busy.  We might yet fail to insert, but any attempt to
3555 	 * insert a conflicting filter while we're waiting for the
3556 	 * firmware must find the busy entry.
3557 	 */
3558 	saved_spec = efx_ef10_filter_entry_spec(table, ins_index);
3559 	if (saved_spec) {
3560 		replacing = true;
3561 	} else {
3562 		saved_spec = kmalloc(sizeof(*spec), GFP_ATOMIC);
3563 		if (!saved_spec) {
3564 			rc = -ENOMEM;
3565 			goto fail_unlock;
3566 		}
3567 		*saved_spec = *spec;
3568 	}
3569 	efx_ef10_filter_set_entry(table, ins_index, saved_spec,
3570 				  EFX_EF10_FILTER_FLAG_BUSY);
3571 
3572 	spin_unlock_bh(&efx->filter_lock);
3573 
3574 	/* Pack up the variables needed on completion */
3575 	cookie = replacing << 31 | ins_index << 16 | spec->dmaq_id;
3576 
3577 	efx_ef10_filter_push_prep(efx, spec, inbuf,
3578 				  table->entry[ins_index].handle, replacing);
3579 	efx_mcdi_rpc_async(efx, MC_CMD_FILTER_OP, inbuf, sizeof(inbuf),
3580 			   MC_CMD_FILTER_OP_OUT_LEN,
3581 			   efx_ef10_filter_rfs_insert_complete, cookie);
3582 
3583 	return ins_index;
3584 
3585 fail_unlock:
3586 	spin_unlock_bh(&efx->filter_lock);
3587 	return rc;
3588 }
3589 
3590 static void
3591 efx_ef10_filter_rfs_insert_complete(struct efx_nic *efx, unsigned long cookie,
3592 				    int rc, efx_dword_t *outbuf,
3593 				    size_t outlen_actual)
3594 {
3595 	struct efx_ef10_filter_table *table = efx->filter_state;
3596 	unsigned int ins_index, dmaq_id;
3597 	struct efx_filter_spec *spec;
3598 	bool replacing;
3599 
3600 	/* Unpack the cookie */
3601 	replacing = cookie >> 31;
3602 	ins_index = (cookie >> 16) & (HUNT_FILTER_TBL_ROWS - 1);
3603 	dmaq_id = cookie & 0xffff;
3604 
3605 	spin_lock_bh(&efx->filter_lock);
3606 	spec = efx_ef10_filter_entry_spec(table, ins_index);
3607 	if (rc == 0) {
3608 		table->entry[ins_index].handle =
3609 			MCDI_QWORD(outbuf, FILTER_OP_OUT_HANDLE);
3610 		if (replacing)
3611 			spec->dmaq_id = dmaq_id;
3612 	} else if (!replacing) {
3613 		kfree(spec);
3614 		spec = NULL;
3615 	}
3616 	efx_ef10_filter_set_entry(table, ins_index, spec, 0);
3617 	spin_unlock_bh(&efx->filter_lock);
3618 
3619 	wake_up_all(&table->waitq);
3620 }
3621 
3622 static void
3623 efx_ef10_filter_rfs_expire_complete(struct efx_nic *efx,
3624 				    unsigned long filter_idx,
3625 				    int rc, efx_dword_t *outbuf,
3626 				    size_t outlen_actual);
3627 
3628 static bool efx_ef10_filter_rfs_expire_one(struct efx_nic *efx, u32 flow_id,
3629 					   unsigned int filter_idx)
3630 {
3631 	struct efx_ef10_filter_table *table = efx->filter_state;
3632 	struct efx_filter_spec *spec =
3633 		efx_ef10_filter_entry_spec(table, filter_idx);
3634 	MCDI_DECLARE_BUF(inbuf,
3635 			 MC_CMD_FILTER_OP_IN_HANDLE_OFST +
3636 			 MC_CMD_FILTER_OP_IN_HANDLE_LEN);
3637 
3638 	if (!spec ||
3639 	    (table->entry[filter_idx].spec & EFX_EF10_FILTER_FLAG_BUSY) ||
3640 	    spec->priority != EFX_FILTER_PRI_HINT ||
3641 	    !rps_may_expire_flow(efx->net_dev, spec->dmaq_id,
3642 				 flow_id, filter_idx))
3643 		return false;
3644 
3645 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
3646 		       MC_CMD_FILTER_OP_IN_OP_REMOVE);
3647 	MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
3648 		       table->entry[filter_idx].handle);
3649 	if (efx_mcdi_rpc_async(efx, MC_CMD_FILTER_OP, inbuf, sizeof(inbuf), 0,
3650 			       efx_ef10_filter_rfs_expire_complete, filter_idx))
3651 		return false;
3652 
3653 	table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_BUSY;
3654 	return true;
3655 }
3656 
3657 static void
3658 efx_ef10_filter_rfs_expire_complete(struct efx_nic *efx,
3659 				    unsigned long filter_idx,
3660 				    int rc, efx_dword_t *outbuf,
3661 				    size_t outlen_actual)
3662 {
3663 	struct efx_ef10_filter_table *table = efx->filter_state;
3664 	struct efx_filter_spec *spec =
3665 		efx_ef10_filter_entry_spec(table, filter_idx);
3666 
3667 	spin_lock_bh(&efx->filter_lock);
3668 	if (rc == 0) {
3669 		kfree(spec);
3670 		efx_ef10_filter_set_entry(table, filter_idx, NULL, 0);
3671 	}
3672 	table->entry[filter_idx].spec &= ~EFX_EF10_FILTER_FLAG_BUSY;
3673 	wake_up_all(&table->waitq);
3674 	spin_unlock_bh(&efx->filter_lock);
3675 }
3676 
3677 #endif /* CONFIG_RFS_ACCEL */
3678 
3679 static int efx_ef10_filter_match_flags_from_mcdi(u32 mcdi_flags)
3680 {
3681 	int match_flags = 0;
3682 
3683 #define MAP_FLAG(gen_flag, mcdi_field) {				\
3684 		u32 old_mcdi_flags = mcdi_flags;			\
3685 		mcdi_flags &= ~(1 << MC_CMD_FILTER_OP_IN_MATCH_ ##	\
3686 				mcdi_field ## _LBN);			\
3687 		if (mcdi_flags != old_mcdi_flags)			\
3688 			match_flags |= EFX_FILTER_MATCH_ ## gen_flag;	\
3689 	}
3690 	MAP_FLAG(LOC_MAC_IG, UNKNOWN_UCAST_DST);
3691 	MAP_FLAG(LOC_MAC_IG, UNKNOWN_MCAST_DST);
3692 	MAP_FLAG(REM_HOST, SRC_IP);
3693 	MAP_FLAG(LOC_HOST, DST_IP);
3694 	MAP_FLAG(REM_MAC, SRC_MAC);
3695 	MAP_FLAG(REM_PORT, SRC_PORT);
3696 	MAP_FLAG(LOC_MAC, DST_MAC);
3697 	MAP_FLAG(LOC_PORT, DST_PORT);
3698 	MAP_FLAG(ETHER_TYPE, ETHER_TYPE);
3699 	MAP_FLAG(INNER_VID, INNER_VLAN);
3700 	MAP_FLAG(OUTER_VID, OUTER_VLAN);
3701 	MAP_FLAG(IP_PROTO, IP_PROTO);
3702 #undef MAP_FLAG
3703 
3704 	/* Did we map them all? */
3705 	if (mcdi_flags)
3706 		return -EINVAL;
3707 
3708 	return match_flags;
3709 }
3710 
3711 static int efx_ef10_filter_table_probe(struct efx_nic *efx)
3712 {
3713 	MCDI_DECLARE_BUF(inbuf, MC_CMD_GET_PARSER_DISP_INFO_IN_LEN);
3714 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_PARSER_DISP_INFO_OUT_LENMAX);
3715 	unsigned int pd_match_pri, pd_match_count;
3716 	struct efx_ef10_filter_table *table;
3717 	size_t outlen;
3718 	int rc;
3719 
3720 	table = kzalloc(sizeof(*table), GFP_KERNEL);
3721 	if (!table)
3722 		return -ENOMEM;
3723 
3724 	/* Find out which RX filter types are supported, and their priorities */
3725 	MCDI_SET_DWORD(inbuf, GET_PARSER_DISP_INFO_IN_OP,
3726 		       MC_CMD_GET_PARSER_DISP_INFO_IN_OP_GET_SUPPORTED_RX_MATCHES);
3727 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_PARSER_DISP_INFO,
3728 			  inbuf, sizeof(inbuf), outbuf, sizeof(outbuf),
3729 			  &outlen);
3730 	if (rc)
3731 		goto fail;
3732 	pd_match_count = MCDI_VAR_ARRAY_LEN(
3733 		outlen, GET_PARSER_DISP_INFO_OUT_SUPPORTED_MATCHES);
3734 	table->rx_match_count = 0;
3735 
3736 	for (pd_match_pri = 0; pd_match_pri < pd_match_count; pd_match_pri++) {
3737 		u32 mcdi_flags =
3738 			MCDI_ARRAY_DWORD(
3739 				outbuf,
3740 				GET_PARSER_DISP_INFO_OUT_SUPPORTED_MATCHES,
3741 				pd_match_pri);
3742 		rc = efx_ef10_filter_match_flags_from_mcdi(mcdi_flags);
3743 		if (rc < 0) {
3744 			netif_dbg(efx, probe, efx->net_dev,
3745 				  "%s: fw flags %#x pri %u not supported in driver\n",
3746 				  __func__, mcdi_flags, pd_match_pri);
3747 		} else {
3748 			netif_dbg(efx, probe, efx->net_dev,
3749 				  "%s: fw flags %#x pri %u supported as driver flags %#x pri %u\n",
3750 				  __func__, mcdi_flags, pd_match_pri,
3751 				  rc, table->rx_match_count);
3752 			table->rx_match_flags[table->rx_match_count++] = rc;
3753 		}
3754 	}
3755 
3756 	table->entry = vzalloc(HUNT_FILTER_TBL_ROWS * sizeof(*table->entry));
3757 	if (!table->entry) {
3758 		rc = -ENOMEM;
3759 		goto fail;
3760 	}
3761 
3762 	table->ucdef_id = EFX_EF10_FILTER_ID_INVALID;
3763 	table->bcast_id = EFX_EF10_FILTER_ID_INVALID;
3764 	table->mcdef_id = EFX_EF10_FILTER_ID_INVALID;
3765 
3766 	efx->filter_state = table;
3767 	init_waitqueue_head(&table->waitq);
3768 	return 0;
3769 
3770 fail:
3771 	kfree(table);
3772 	return rc;
3773 }
3774 
3775 /* Caller must hold efx->filter_sem for read if race against
3776  * efx_ef10_filter_table_remove() is possible
3777  */
3778 static void efx_ef10_filter_table_restore(struct efx_nic *efx)
3779 {
3780 	struct efx_ef10_filter_table *table = efx->filter_state;
3781 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
3782 	struct efx_filter_spec *spec;
3783 	unsigned int filter_idx;
3784 	bool failed = false;
3785 	int rc;
3786 
3787 	WARN_ON(!rwsem_is_locked(&efx->filter_sem));
3788 
3789 	if (!nic_data->must_restore_filters)
3790 		return;
3791 
3792 	if (!table)
3793 		return;
3794 
3795 	spin_lock_bh(&efx->filter_lock);
3796 
3797 	for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
3798 		spec = efx_ef10_filter_entry_spec(table, filter_idx);
3799 		if (!spec)
3800 			continue;
3801 
3802 		table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_BUSY;
3803 		spin_unlock_bh(&efx->filter_lock);
3804 
3805 		rc = efx_ef10_filter_push(efx, spec,
3806 					  &table->entry[filter_idx].handle,
3807 					  false);
3808 		if (rc)
3809 			failed = true;
3810 
3811 		spin_lock_bh(&efx->filter_lock);
3812 		if (rc) {
3813 			kfree(spec);
3814 			efx_ef10_filter_set_entry(table, filter_idx, NULL, 0);
3815 		} else {
3816 			table->entry[filter_idx].spec &=
3817 				~EFX_EF10_FILTER_FLAG_BUSY;
3818 		}
3819 	}
3820 
3821 	spin_unlock_bh(&efx->filter_lock);
3822 
3823 	if (failed)
3824 		netif_err(efx, hw, efx->net_dev,
3825 			  "unable to restore all filters\n");
3826 	else
3827 		nic_data->must_restore_filters = false;
3828 }
3829 
3830 /* Caller must hold efx->filter_sem for write */
3831 static void efx_ef10_filter_table_remove(struct efx_nic *efx)
3832 {
3833 	struct efx_ef10_filter_table *table = efx->filter_state;
3834 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_IN_LEN);
3835 	struct efx_filter_spec *spec;
3836 	unsigned int filter_idx;
3837 	int rc;
3838 
3839 	efx->filter_state = NULL;
3840 	if (!table)
3841 		return;
3842 
3843 	for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
3844 		spec = efx_ef10_filter_entry_spec(table, filter_idx);
3845 		if (!spec)
3846 			continue;
3847 
3848 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
3849 			       efx_ef10_filter_is_exclusive(spec) ?
3850 			       MC_CMD_FILTER_OP_IN_OP_REMOVE :
3851 			       MC_CMD_FILTER_OP_IN_OP_UNSUBSCRIBE);
3852 		MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
3853 			       table->entry[filter_idx].handle);
3854 		rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FILTER_OP, inbuf,
3855 					sizeof(inbuf), NULL, 0, NULL);
3856 		if (rc)
3857 			netif_info(efx, drv, efx->net_dev,
3858 				   "%s: filter %04x remove failed\n",
3859 				   __func__, filter_idx);
3860 		kfree(spec);
3861 	}
3862 
3863 	vfree(table->entry);
3864 	kfree(table);
3865 }
3866 
3867 #define EFX_EF10_FILTER_DO_MARK_OLD(id) \
3868 	if (id != EFX_EF10_FILTER_ID_INVALID) { \
3869 		filter_idx = efx_ef10_filter_get_unsafe_id(efx, id); \
3870 		if (!table->entry[filter_idx].spec) \
3871 			netif_dbg(efx, drv, efx->net_dev, \
3872 				  "%s: marked null spec old %04x:%04x\n", \
3873 				  __func__, id, filter_idx); \
3874 		table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_AUTO_OLD;\
3875 	}
3876 static void efx_ef10_filter_mark_old(struct efx_nic *efx)
3877 {
3878 	struct efx_ef10_filter_table *table = efx->filter_state;
3879 	unsigned int filter_idx, i;
3880 
3881 	if (!table)
3882 		return;
3883 
3884 	/* Mark old filters that may need to be removed */
3885 	spin_lock_bh(&efx->filter_lock);
3886 	for (i = 0; i < table->dev_uc_count; i++)
3887 		EFX_EF10_FILTER_DO_MARK_OLD(table->dev_uc_list[i].id);
3888 	for (i = 0; i < table->dev_mc_count; i++)
3889 		EFX_EF10_FILTER_DO_MARK_OLD(table->dev_mc_list[i].id);
3890 	EFX_EF10_FILTER_DO_MARK_OLD(table->ucdef_id);
3891 	EFX_EF10_FILTER_DO_MARK_OLD(table->bcast_id);
3892 	EFX_EF10_FILTER_DO_MARK_OLD(table->mcdef_id);
3893 	spin_unlock_bh(&efx->filter_lock);
3894 }
3895 #undef EFX_EF10_FILTER_DO_MARK_OLD
3896 
3897 static void efx_ef10_filter_uc_addr_list(struct efx_nic *efx, bool *promisc)
3898 {
3899 	struct efx_ef10_filter_table *table = efx->filter_state;
3900 	struct net_device *net_dev = efx->net_dev;
3901 	struct netdev_hw_addr *uc;
3902 	int addr_count;
3903 	unsigned int i;
3904 
3905 	table->ucdef_id = EFX_EF10_FILTER_ID_INVALID;
3906 	addr_count = netdev_uc_count(net_dev);
3907 	if (net_dev->flags & IFF_PROMISC)
3908 		*promisc = true;
3909 	table->dev_uc_count = 1 + addr_count;
3910 	ether_addr_copy(table->dev_uc_list[0].addr, net_dev->dev_addr);
3911 	i = 1;
3912 	netdev_for_each_uc_addr(uc, net_dev) {
3913 		if (i >= EFX_EF10_FILTER_DEV_UC_MAX) {
3914 			*promisc = true;
3915 			break;
3916 		}
3917 		ether_addr_copy(table->dev_uc_list[i].addr, uc->addr);
3918 		table->dev_uc_list[i].id = EFX_EF10_FILTER_ID_INVALID;
3919 		i++;
3920 	}
3921 }
3922 
3923 static void efx_ef10_filter_mc_addr_list(struct efx_nic *efx, bool *promisc)
3924 {
3925 	struct efx_ef10_filter_table *table = efx->filter_state;
3926 	struct net_device *net_dev = efx->net_dev;
3927 	struct netdev_hw_addr *mc;
3928 	unsigned int i, addr_count;
3929 
3930 	table->mcdef_id = EFX_EF10_FILTER_ID_INVALID;
3931 	table->bcast_id = EFX_EF10_FILTER_ID_INVALID;
3932 	if (net_dev->flags & (IFF_PROMISC | IFF_ALLMULTI))
3933 		*promisc = true;
3934 
3935 	addr_count = netdev_mc_count(net_dev);
3936 	i = 0;
3937 	netdev_for_each_mc_addr(mc, net_dev) {
3938 		if (i >= EFX_EF10_FILTER_DEV_MC_MAX) {
3939 			*promisc = true;
3940 			break;
3941 		}
3942 		ether_addr_copy(table->dev_mc_list[i].addr, mc->addr);
3943 		table->dev_mc_list[i].id = EFX_EF10_FILTER_ID_INVALID;
3944 		i++;
3945 	}
3946 
3947 	table->dev_mc_count = i;
3948 }
3949 
3950 static int efx_ef10_filter_insert_addr_list(struct efx_nic *efx,
3951 					     bool multicast, bool rollback)
3952 {
3953 	struct efx_ef10_filter_table *table = efx->filter_state;
3954 	struct efx_ef10_dev_addr *addr_list;
3955 	enum efx_filter_flags filter_flags;
3956 	struct efx_filter_spec spec;
3957 	u8 baddr[ETH_ALEN];
3958 	unsigned int i, j;
3959 	int addr_count;
3960 	int rc;
3961 
3962 	if (multicast) {
3963 		addr_list = table->dev_mc_list;
3964 		addr_count = table->dev_mc_count;
3965 	} else {
3966 		addr_list = table->dev_uc_list;
3967 		addr_count = table->dev_uc_count;
3968 	}
3969 
3970 	filter_flags = efx_rss_enabled(efx) ? EFX_FILTER_FLAG_RX_RSS : 0;
3971 
3972 	/* Insert/renew filters */
3973 	for (i = 0; i < addr_count; i++) {
3974 		efx_filter_init_rx(&spec, EFX_FILTER_PRI_AUTO, filter_flags, 0);
3975 		efx_filter_set_eth_local(&spec, EFX_FILTER_VID_UNSPEC,
3976 					 addr_list[i].addr);
3977 		rc = efx_ef10_filter_insert(efx, &spec, true);
3978 		if (rc < 0) {
3979 			if (rollback) {
3980 				netif_info(efx, drv, efx->net_dev,
3981 					   "efx_ef10_filter_insert failed rc=%d\n",
3982 					   rc);
3983 				/* Fall back to promiscuous */
3984 				for (j = 0; j < i; j++) {
3985 					if (addr_list[j].id == EFX_EF10_FILTER_ID_INVALID)
3986 						continue;
3987 					efx_ef10_filter_remove_unsafe(
3988 						efx, EFX_FILTER_PRI_AUTO,
3989 						addr_list[j].id);
3990 					addr_list[j].id = EFX_EF10_FILTER_ID_INVALID;
3991 				}
3992 				return rc;
3993 			} else {
3994 				/* mark as not inserted, and carry on */
3995 				rc = EFX_EF10_FILTER_ID_INVALID;
3996 			}
3997 		}
3998 		addr_list[i].id = efx_ef10_filter_get_unsafe_id(efx, rc);
3999 	}
4000 
4001 	if (multicast && rollback) {
4002 		/* Also need an Ethernet broadcast filter */
4003 		efx_filter_init_rx(&spec, EFX_FILTER_PRI_AUTO, filter_flags, 0);
4004 		eth_broadcast_addr(baddr);
4005 		efx_filter_set_eth_local(&spec, EFX_FILTER_VID_UNSPEC, baddr);
4006 		rc = efx_ef10_filter_insert(efx, &spec, true);
4007 		if (rc < 0) {
4008 			netif_warn(efx, drv, efx->net_dev,
4009 				   "Broadcast filter insert failed rc=%d\n", rc);
4010 			/* Fall back to promiscuous */
4011 			for (j = 0; j < i; j++) {
4012 				if (addr_list[j].id == EFX_EF10_FILTER_ID_INVALID)
4013 					continue;
4014 				efx_ef10_filter_remove_unsafe(
4015 					efx, EFX_FILTER_PRI_AUTO,
4016 					addr_list[j].id);
4017 				addr_list[j].id = EFX_EF10_FILTER_ID_INVALID;
4018 			}
4019 			return rc;
4020 		} else {
4021 			table->bcast_id = efx_ef10_filter_get_unsafe_id(efx, rc);
4022 		}
4023 	}
4024 
4025 	return 0;
4026 }
4027 
4028 static int efx_ef10_filter_insert_def(struct efx_nic *efx, bool multicast,
4029 				      bool rollback)
4030 {
4031 	struct efx_ef10_filter_table *table = efx->filter_state;
4032 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
4033 	enum efx_filter_flags filter_flags;
4034 	struct efx_filter_spec spec;
4035 	u8 baddr[ETH_ALEN];
4036 	int rc;
4037 
4038 	filter_flags = efx_rss_enabled(efx) ? EFX_FILTER_FLAG_RX_RSS : 0;
4039 
4040 	efx_filter_init_rx(&spec, EFX_FILTER_PRI_AUTO, filter_flags, 0);
4041 
4042 	if (multicast)
4043 		efx_filter_set_mc_def(&spec);
4044 	else
4045 		efx_filter_set_uc_def(&spec);
4046 
4047 	rc = efx_ef10_filter_insert(efx, &spec, true);
4048 	if (rc < 0) {
4049 		netif_printk(efx, drv, rc == -EPERM ? KERN_DEBUG : KERN_WARNING,
4050 			     efx->net_dev,
4051 			     "%scast mismatch filter insert failed rc=%d\n",
4052 			     multicast ? "Multi" : "Uni", rc);
4053 	} else if (multicast) {
4054 		table->mcdef_id = efx_ef10_filter_get_unsafe_id(efx, rc);
4055 		if (!nic_data->workaround_26807) {
4056 			/* Also need an Ethernet broadcast filter */
4057 			efx_filter_init_rx(&spec, EFX_FILTER_PRI_AUTO,
4058 					   filter_flags, 0);
4059 			eth_broadcast_addr(baddr);
4060 			efx_filter_set_eth_local(&spec, EFX_FILTER_VID_UNSPEC,
4061 						 baddr);
4062 			rc = efx_ef10_filter_insert(efx, &spec, true);
4063 			if (rc < 0) {
4064 				netif_warn(efx, drv, efx->net_dev,
4065 					   "Broadcast filter insert failed rc=%d\n",
4066 					   rc);
4067 				if (rollback) {
4068 					/* Roll back the mc_def filter */
4069 					efx_ef10_filter_remove_unsafe(
4070 							efx, EFX_FILTER_PRI_AUTO,
4071 							table->mcdef_id);
4072 					table->mcdef_id = EFX_EF10_FILTER_ID_INVALID;
4073 					return rc;
4074 				}
4075 			} else {
4076 				table->bcast_id = efx_ef10_filter_get_unsafe_id(efx, rc);
4077 			}
4078 		}
4079 		rc = 0;
4080 	} else {
4081 		table->ucdef_id = rc;
4082 		rc = 0;
4083 	}
4084 	return rc;
4085 }
4086 
4087 /* Remove filters that weren't renewed.  Since nothing else changes the AUTO_OLD
4088  * flag or removes these filters, we don't need to hold the filter_lock while
4089  * scanning for these filters.
4090  */
4091 static void efx_ef10_filter_remove_old(struct efx_nic *efx)
4092 {
4093 	struct efx_ef10_filter_table *table = efx->filter_state;
4094 	int remove_failed = 0;
4095 	int remove_noent = 0;
4096 	int rc;
4097 	int i;
4098 
4099 	for (i = 0; i < HUNT_FILTER_TBL_ROWS; i++) {
4100 		if (ACCESS_ONCE(table->entry[i].spec) &
4101 		    EFX_EF10_FILTER_FLAG_AUTO_OLD) {
4102 			rc = efx_ef10_filter_remove_internal(efx,
4103 					1U << EFX_FILTER_PRI_AUTO, i, true);
4104 			if (rc == -ENOENT)
4105 				remove_noent++;
4106 			else if (rc)
4107 				remove_failed++;
4108 		}
4109 	}
4110 
4111 	if (remove_failed)
4112 		netif_info(efx, drv, efx->net_dev,
4113 			   "%s: failed to remove %d filters\n",
4114 			   __func__, remove_failed);
4115 	if (remove_noent)
4116 		netif_info(efx, drv, efx->net_dev,
4117 			   "%s: failed to remove %d non-existent filters\n",
4118 			   __func__, remove_noent);
4119 }
4120 
4121 static int efx_ef10_vport_set_mac_address(struct efx_nic *efx)
4122 {
4123 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
4124 	u8 mac_old[ETH_ALEN];
4125 	int rc, rc2;
4126 
4127 	/* Only reconfigure a PF-created vport */
4128 	if (is_zero_ether_addr(nic_data->vport_mac))
4129 		return 0;
4130 
4131 	efx_device_detach_sync(efx);
4132 	efx_net_stop(efx->net_dev);
4133 	down_write(&efx->filter_sem);
4134 	efx_ef10_filter_table_remove(efx);
4135 	up_write(&efx->filter_sem);
4136 
4137 	rc = efx_ef10_vadaptor_free(efx, nic_data->vport_id);
4138 	if (rc)
4139 		goto restore_filters;
4140 
4141 	ether_addr_copy(mac_old, nic_data->vport_mac);
4142 	rc = efx_ef10_vport_del_mac(efx, nic_data->vport_id,
4143 				    nic_data->vport_mac);
4144 	if (rc)
4145 		goto restore_vadaptor;
4146 
4147 	rc = efx_ef10_vport_add_mac(efx, nic_data->vport_id,
4148 				    efx->net_dev->dev_addr);
4149 	if (!rc) {
4150 		ether_addr_copy(nic_data->vport_mac, efx->net_dev->dev_addr);
4151 	} else {
4152 		rc2 = efx_ef10_vport_add_mac(efx, nic_data->vport_id, mac_old);
4153 		if (rc2) {
4154 			/* Failed to add original MAC, so clear vport_mac */
4155 			eth_zero_addr(nic_data->vport_mac);
4156 			goto reset_nic;
4157 		}
4158 	}
4159 
4160 restore_vadaptor:
4161 	rc2 = efx_ef10_vadaptor_alloc(efx, nic_data->vport_id);
4162 	if (rc2)
4163 		goto reset_nic;
4164 restore_filters:
4165 	down_write(&efx->filter_sem);
4166 	rc2 = efx_ef10_filter_table_probe(efx);
4167 	up_write(&efx->filter_sem);
4168 	if (rc2)
4169 		goto reset_nic;
4170 
4171 	rc2 = efx_net_open(efx->net_dev);
4172 	if (rc2)
4173 		goto reset_nic;
4174 
4175 	netif_device_attach(efx->net_dev);
4176 
4177 	return rc;
4178 
4179 reset_nic:
4180 	netif_err(efx, drv, efx->net_dev,
4181 		  "Failed to restore when changing MAC address - scheduling reset\n");
4182 	efx_schedule_reset(efx, RESET_TYPE_DATAPATH);
4183 
4184 	return rc ? rc : rc2;
4185 }
4186 
4187 /* Caller must hold efx->filter_sem for read if race against
4188  * efx_ef10_filter_table_remove() is possible
4189  */
4190 static void efx_ef10_filter_sync_rx_mode(struct efx_nic *efx)
4191 {
4192 	struct efx_ef10_filter_table *table = efx->filter_state;
4193 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
4194 	struct net_device *net_dev = efx->net_dev;
4195 	bool uc_promisc = false, mc_promisc = false;
4196 
4197 	if (!efx_dev_registered(efx))
4198 		return;
4199 
4200 	if (!table)
4201 		return;
4202 
4203 	efx_ef10_filter_mark_old(efx);
4204 
4205 	/* Copy/convert the address lists; add the primary station
4206 	 * address and broadcast address
4207 	 */
4208 	netif_addr_lock_bh(net_dev);
4209 	efx_ef10_filter_uc_addr_list(efx, &uc_promisc);
4210 	efx_ef10_filter_mc_addr_list(efx, &mc_promisc);
4211 	netif_addr_unlock_bh(net_dev);
4212 
4213 	/* Insert/renew unicast filters */
4214 	if (uc_promisc) {
4215 		efx_ef10_filter_insert_def(efx, false, false);
4216 		efx_ef10_filter_insert_addr_list(efx, false, false);
4217 	} else {
4218 		/* If any of the filters failed to insert, fall back to
4219 		 * promiscuous mode - add in the uc_def filter.  But keep
4220 		 * our individual unicast filters.
4221 		 */
4222 		if (efx_ef10_filter_insert_addr_list(efx, false, false))
4223 			efx_ef10_filter_insert_def(efx, false, false);
4224 	}
4225 
4226 	/* Insert/renew multicast filters */
4227 	/* If changing promiscuous state with cascaded multicast filters, remove
4228 	 * old filters first, so that packets are dropped rather than duplicated
4229 	 */
4230 	if (nic_data->workaround_26807 && efx->mc_promisc != mc_promisc)
4231 		efx_ef10_filter_remove_old(efx);
4232 	if (mc_promisc) {
4233 		if (nic_data->workaround_26807) {
4234 			/* If we failed to insert promiscuous filters, rollback
4235 			 * and fall back to individual multicast filters
4236 			 */
4237 			if (efx_ef10_filter_insert_def(efx, true, true)) {
4238 				/* Changing promisc state, so remove old filters */
4239 				efx_ef10_filter_remove_old(efx);
4240 				efx_ef10_filter_insert_addr_list(efx, true, false);
4241 			}
4242 		} else {
4243 			/* If we failed to insert promiscuous filters, don't
4244 			 * rollback.  Regardless, also insert the mc_list
4245 			 */
4246 			efx_ef10_filter_insert_def(efx, true, false);
4247 			efx_ef10_filter_insert_addr_list(efx, true, false);
4248 		}
4249 	} else {
4250 		/* If any filters failed to insert, rollback and fall back to
4251 		 * promiscuous mode - mc_def filter and maybe broadcast.  If
4252 		 * that fails, roll back again and insert as many of our
4253 		 * individual multicast filters as we can.
4254 		 */
4255 		if (efx_ef10_filter_insert_addr_list(efx, true, true)) {
4256 			/* Changing promisc state, so remove old filters */
4257 			if (nic_data->workaround_26807)
4258 				efx_ef10_filter_remove_old(efx);
4259 			if (efx_ef10_filter_insert_def(efx, true, true))
4260 				efx_ef10_filter_insert_addr_list(efx, true, false);
4261 		}
4262 	}
4263 
4264 	efx_ef10_filter_remove_old(efx);
4265 	efx->mc_promisc = mc_promisc;
4266 }
4267 
4268 static int efx_ef10_set_mac_address(struct efx_nic *efx)
4269 {
4270 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VADAPTOR_SET_MAC_IN_LEN);
4271 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
4272 	bool was_enabled = efx->port_enabled;
4273 	int rc;
4274 
4275 	efx_device_detach_sync(efx);
4276 	efx_net_stop(efx->net_dev);
4277 	down_write(&efx->filter_sem);
4278 	efx_ef10_filter_table_remove(efx);
4279 
4280 	ether_addr_copy(MCDI_PTR(inbuf, VADAPTOR_SET_MAC_IN_MACADDR),
4281 			efx->net_dev->dev_addr);
4282 	MCDI_SET_DWORD(inbuf, VADAPTOR_SET_MAC_IN_UPSTREAM_PORT_ID,
4283 		       nic_data->vport_id);
4284 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_VADAPTOR_SET_MAC, inbuf,
4285 				sizeof(inbuf), NULL, 0, NULL);
4286 
4287 	efx_ef10_filter_table_probe(efx);
4288 	up_write(&efx->filter_sem);
4289 	if (was_enabled)
4290 		efx_net_open(efx->net_dev);
4291 	netif_device_attach(efx->net_dev);
4292 
4293 #ifdef CONFIG_SFC_SRIOV
4294 	if (efx->pci_dev->is_virtfn && efx->pci_dev->physfn) {
4295 		struct pci_dev *pci_dev_pf = efx->pci_dev->physfn;
4296 
4297 		if (rc == -EPERM) {
4298 			struct efx_nic *efx_pf;
4299 
4300 			/* Switch to PF and change MAC address on vport */
4301 			efx_pf = pci_get_drvdata(pci_dev_pf);
4302 
4303 			rc = efx_ef10_sriov_set_vf_mac(efx_pf,
4304 						       nic_data->vf_index,
4305 						       efx->net_dev->dev_addr);
4306 		} else if (!rc) {
4307 			struct efx_nic *efx_pf = pci_get_drvdata(pci_dev_pf);
4308 			struct efx_ef10_nic_data *nic_data = efx_pf->nic_data;
4309 			unsigned int i;
4310 
4311 			/* MAC address successfully changed by VF (with MAC
4312 			 * spoofing) so update the parent PF if possible.
4313 			 */
4314 			for (i = 0; i < efx_pf->vf_count; ++i) {
4315 				struct ef10_vf *vf = nic_data->vf + i;
4316 
4317 				if (vf->efx == efx) {
4318 					ether_addr_copy(vf->mac,
4319 							efx->net_dev->dev_addr);
4320 					return 0;
4321 				}
4322 			}
4323 		}
4324 	} else
4325 #endif
4326 	if (rc == -EPERM) {
4327 		netif_err(efx, drv, efx->net_dev,
4328 			  "Cannot change MAC address; use sfboot to enable"
4329 			  " mac-spoofing on this interface\n");
4330 	} else if (rc == -ENOSYS && !efx_ef10_is_vf(efx)) {
4331 		/* If the active MCFW does not support MC_CMD_VADAPTOR_SET_MAC
4332 		 * fall-back to the method of changing the MAC address on the
4333 		 * vport.  This only applies to PFs because such versions of
4334 		 * MCFW do not support VFs.
4335 		 */
4336 		rc = efx_ef10_vport_set_mac_address(efx);
4337 	} else {
4338 		efx_mcdi_display_error(efx, MC_CMD_VADAPTOR_SET_MAC,
4339 				       sizeof(inbuf), NULL, 0, rc);
4340 	}
4341 
4342 	return rc;
4343 }
4344 
4345 static int efx_ef10_mac_reconfigure(struct efx_nic *efx)
4346 {
4347 	efx_ef10_filter_sync_rx_mode(efx);
4348 
4349 	return efx_mcdi_set_mac(efx);
4350 }
4351 
4352 static int efx_ef10_mac_reconfigure_vf(struct efx_nic *efx)
4353 {
4354 	efx_ef10_filter_sync_rx_mode(efx);
4355 
4356 	return 0;
4357 }
4358 
4359 static int efx_ef10_start_bist(struct efx_nic *efx, u32 bist_type)
4360 {
4361 	MCDI_DECLARE_BUF(inbuf, MC_CMD_START_BIST_IN_LEN);
4362 
4363 	MCDI_SET_DWORD(inbuf, START_BIST_IN_TYPE, bist_type);
4364 	return efx_mcdi_rpc(efx, MC_CMD_START_BIST, inbuf, sizeof(inbuf),
4365 			    NULL, 0, NULL);
4366 }
4367 
4368 /* MC BISTs follow a different poll mechanism to phy BISTs.
4369  * The BIST is done in the poll handler on the MC, and the MCDI command
4370  * will block until the BIST is done.
4371  */
4372 static int efx_ef10_poll_bist(struct efx_nic *efx)
4373 {
4374 	int rc;
4375 	MCDI_DECLARE_BUF(outbuf, MC_CMD_POLL_BIST_OUT_LEN);
4376 	size_t outlen;
4377 	u32 result;
4378 
4379 	rc = efx_mcdi_rpc(efx, MC_CMD_POLL_BIST, NULL, 0,
4380 			   outbuf, sizeof(outbuf), &outlen);
4381 	if (rc != 0)
4382 		return rc;
4383 
4384 	if (outlen < MC_CMD_POLL_BIST_OUT_LEN)
4385 		return -EIO;
4386 
4387 	result = MCDI_DWORD(outbuf, POLL_BIST_OUT_RESULT);
4388 	switch (result) {
4389 	case MC_CMD_POLL_BIST_PASSED:
4390 		netif_dbg(efx, hw, efx->net_dev, "BIST passed.\n");
4391 		return 0;
4392 	case MC_CMD_POLL_BIST_TIMEOUT:
4393 		netif_err(efx, hw, efx->net_dev, "BIST timed out\n");
4394 		return -EIO;
4395 	case MC_CMD_POLL_BIST_FAILED:
4396 		netif_err(efx, hw, efx->net_dev, "BIST failed.\n");
4397 		return -EIO;
4398 	default:
4399 		netif_err(efx, hw, efx->net_dev,
4400 			  "BIST returned unknown result %u", result);
4401 		return -EIO;
4402 	}
4403 }
4404 
4405 static int efx_ef10_run_bist(struct efx_nic *efx, u32 bist_type)
4406 {
4407 	int rc;
4408 
4409 	netif_dbg(efx, drv, efx->net_dev, "starting BIST type %u\n", bist_type);
4410 
4411 	rc = efx_ef10_start_bist(efx, bist_type);
4412 	if (rc != 0)
4413 		return rc;
4414 
4415 	return efx_ef10_poll_bist(efx);
4416 }
4417 
4418 static int
4419 efx_ef10_test_chip(struct efx_nic *efx, struct efx_self_tests *tests)
4420 {
4421 	int rc, rc2;
4422 
4423 	efx_reset_down(efx, RESET_TYPE_WORLD);
4424 
4425 	rc = efx_mcdi_rpc(efx, MC_CMD_ENABLE_OFFLINE_BIST,
4426 			  NULL, 0, NULL, 0, NULL);
4427 	if (rc != 0)
4428 		goto out;
4429 
4430 	tests->memory = efx_ef10_run_bist(efx, MC_CMD_MC_MEM_BIST) ? -1 : 1;
4431 	tests->registers = efx_ef10_run_bist(efx, MC_CMD_REG_BIST) ? -1 : 1;
4432 
4433 	rc = efx_mcdi_reset(efx, RESET_TYPE_WORLD);
4434 
4435 out:
4436 	if (rc == -EPERM)
4437 		rc = 0;
4438 	rc2 = efx_reset_up(efx, RESET_TYPE_WORLD, rc == 0);
4439 	return rc ? rc : rc2;
4440 }
4441 
4442 #ifdef CONFIG_SFC_MTD
4443 
4444 struct efx_ef10_nvram_type_info {
4445 	u16 type, type_mask;
4446 	u8 port;
4447 	const char *name;
4448 };
4449 
4450 static const struct efx_ef10_nvram_type_info efx_ef10_nvram_types[] = {
4451 	{ NVRAM_PARTITION_TYPE_MC_FIRMWARE,	   0,    0, "sfc_mcfw" },
4452 	{ NVRAM_PARTITION_TYPE_MC_FIRMWARE_BACKUP, 0,    0, "sfc_mcfw_backup" },
4453 	{ NVRAM_PARTITION_TYPE_EXPANSION_ROM,	   0,    0, "sfc_exp_rom" },
4454 	{ NVRAM_PARTITION_TYPE_STATIC_CONFIG,	   0,    0, "sfc_static_cfg" },
4455 	{ NVRAM_PARTITION_TYPE_DYNAMIC_CONFIG,	   0,    0, "sfc_dynamic_cfg" },
4456 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT0, 0,   0, "sfc_exp_rom_cfg" },
4457 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT1, 0,   1, "sfc_exp_rom_cfg" },
4458 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT2, 0,   2, "sfc_exp_rom_cfg" },
4459 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT3, 0,   3, "sfc_exp_rom_cfg" },
4460 	{ NVRAM_PARTITION_TYPE_LICENSE,		   0,    0, "sfc_license" },
4461 	{ NVRAM_PARTITION_TYPE_PHY_MIN,		   0xff, 0, "sfc_phy_fw" },
4462 };
4463 
4464 static int efx_ef10_mtd_probe_partition(struct efx_nic *efx,
4465 					struct efx_mcdi_mtd_partition *part,
4466 					unsigned int type)
4467 {
4468 	MCDI_DECLARE_BUF(inbuf, MC_CMD_NVRAM_METADATA_IN_LEN);
4469 	MCDI_DECLARE_BUF(outbuf, MC_CMD_NVRAM_METADATA_OUT_LENMAX);
4470 	const struct efx_ef10_nvram_type_info *info;
4471 	size_t size, erase_size, outlen;
4472 	bool protected;
4473 	int rc;
4474 
4475 	for (info = efx_ef10_nvram_types; ; info++) {
4476 		if (info ==
4477 		    efx_ef10_nvram_types + ARRAY_SIZE(efx_ef10_nvram_types))
4478 			return -ENODEV;
4479 		if ((type & ~info->type_mask) == info->type)
4480 			break;
4481 	}
4482 	if (info->port != efx_port_num(efx))
4483 		return -ENODEV;
4484 
4485 	rc = efx_mcdi_nvram_info(efx, type, &size, &erase_size, &protected);
4486 	if (rc)
4487 		return rc;
4488 	if (protected)
4489 		return -ENODEV; /* hide it */
4490 
4491 	part->nvram_type = type;
4492 
4493 	MCDI_SET_DWORD(inbuf, NVRAM_METADATA_IN_TYPE, type);
4494 	rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_METADATA, inbuf, sizeof(inbuf),
4495 			  outbuf, sizeof(outbuf), &outlen);
4496 	if (rc)
4497 		return rc;
4498 	if (outlen < MC_CMD_NVRAM_METADATA_OUT_LENMIN)
4499 		return -EIO;
4500 	if (MCDI_DWORD(outbuf, NVRAM_METADATA_OUT_FLAGS) &
4501 	    (1 << MC_CMD_NVRAM_METADATA_OUT_SUBTYPE_VALID_LBN))
4502 		part->fw_subtype = MCDI_DWORD(outbuf,
4503 					      NVRAM_METADATA_OUT_SUBTYPE);
4504 
4505 	part->common.dev_type_name = "EF10 NVRAM manager";
4506 	part->common.type_name = info->name;
4507 
4508 	part->common.mtd.type = MTD_NORFLASH;
4509 	part->common.mtd.flags = MTD_CAP_NORFLASH;
4510 	part->common.mtd.size = size;
4511 	part->common.mtd.erasesize = erase_size;
4512 
4513 	return 0;
4514 }
4515 
4516 static int efx_ef10_mtd_probe(struct efx_nic *efx)
4517 {
4518 	MCDI_DECLARE_BUF(outbuf, MC_CMD_NVRAM_PARTITIONS_OUT_LENMAX);
4519 	struct efx_mcdi_mtd_partition *parts;
4520 	size_t outlen, n_parts_total, i, n_parts;
4521 	unsigned int type;
4522 	int rc;
4523 
4524 	ASSERT_RTNL();
4525 
4526 	BUILD_BUG_ON(MC_CMD_NVRAM_PARTITIONS_IN_LEN != 0);
4527 	rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_PARTITIONS, NULL, 0,
4528 			  outbuf, sizeof(outbuf), &outlen);
4529 	if (rc)
4530 		return rc;
4531 	if (outlen < MC_CMD_NVRAM_PARTITIONS_OUT_LENMIN)
4532 		return -EIO;
4533 
4534 	n_parts_total = MCDI_DWORD(outbuf, NVRAM_PARTITIONS_OUT_NUM_PARTITIONS);
4535 	if (n_parts_total >
4536 	    MCDI_VAR_ARRAY_LEN(outlen, NVRAM_PARTITIONS_OUT_TYPE_ID))
4537 		return -EIO;
4538 
4539 	parts = kcalloc(n_parts_total, sizeof(*parts), GFP_KERNEL);
4540 	if (!parts)
4541 		return -ENOMEM;
4542 
4543 	n_parts = 0;
4544 	for (i = 0; i < n_parts_total; i++) {
4545 		type = MCDI_ARRAY_DWORD(outbuf, NVRAM_PARTITIONS_OUT_TYPE_ID,
4546 					i);
4547 		rc = efx_ef10_mtd_probe_partition(efx, &parts[n_parts], type);
4548 		if (rc == 0)
4549 			n_parts++;
4550 		else if (rc != -ENODEV)
4551 			goto fail;
4552 	}
4553 
4554 	rc = efx_mtd_add(efx, &parts[0].common, n_parts, sizeof(*parts));
4555 fail:
4556 	if (rc)
4557 		kfree(parts);
4558 	return rc;
4559 }
4560 
4561 #endif /* CONFIG_SFC_MTD */
4562 
4563 static void efx_ef10_ptp_write_host_time(struct efx_nic *efx, u32 host_time)
4564 {
4565 	_efx_writed(efx, cpu_to_le32(host_time), ER_DZ_MC_DB_LWRD);
4566 }
4567 
4568 static void efx_ef10_ptp_write_host_time_vf(struct efx_nic *efx,
4569 					    u32 host_time) {}
4570 
4571 static int efx_ef10_rx_enable_timestamping(struct efx_channel *channel,
4572 					   bool temp)
4573 {
4574 	MCDI_DECLARE_BUF(inbuf, MC_CMD_PTP_IN_TIME_EVENT_SUBSCRIBE_LEN);
4575 	int rc;
4576 
4577 	if (channel->sync_events_state == SYNC_EVENTS_REQUESTED ||
4578 	    channel->sync_events_state == SYNC_EVENTS_VALID ||
4579 	    (temp && channel->sync_events_state == SYNC_EVENTS_DISABLED))
4580 		return 0;
4581 	channel->sync_events_state = SYNC_EVENTS_REQUESTED;
4582 
4583 	MCDI_SET_DWORD(inbuf, PTP_IN_OP, MC_CMD_PTP_OP_TIME_EVENT_SUBSCRIBE);
4584 	MCDI_SET_DWORD(inbuf, PTP_IN_PERIPH_ID, 0);
4585 	MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_SUBSCRIBE_QUEUE,
4586 		       channel->channel);
4587 
4588 	rc = efx_mcdi_rpc(channel->efx, MC_CMD_PTP,
4589 			  inbuf, sizeof(inbuf), NULL, 0, NULL);
4590 
4591 	if (rc != 0)
4592 		channel->sync_events_state = temp ? SYNC_EVENTS_QUIESCENT :
4593 						    SYNC_EVENTS_DISABLED;
4594 
4595 	return rc;
4596 }
4597 
4598 static int efx_ef10_rx_disable_timestamping(struct efx_channel *channel,
4599 					    bool temp)
4600 {
4601 	MCDI_DECLARE_BUF(inbuf, MC_CMD_PTP_IN_TIME_EVENT_UNSUBSCRIBE_LEN);
4602 	int rc;
4603 
4604 	if (channel->sync_events_state == SYNC_EVENTS_DISABLED ||
4605 	    (temp && channel->sync_events_state == SYNC_EVENTS_QUIESCENT))
4606 		return 0;
4607 	if (channel->sync_events_state == SYNC_EVENTS_QUIESCENT) {
4608 		channel->sync_events_state = SYNC_EVENTS_DISABLED;
4609 		return 0;
4610 	}
4611 	channel->sync_events_state = temp ? SYNC_EVENTS_QUIESCENT :
4612 					    SYNC_EVENTS_DISABLED;
4613 
4614 	MCDI_SET_DWORD(inbuf, PTP_IN_OP, MC_CMD_PTP_OP_TIME_EVENT_UNSUBSCRIBE);
4615 	MCDI_SET_DWORD(inbuf, PTP_IN_PERIPH_ID, 0);
4616 	MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_UNSUBSCRIBE_CONTROL,
4617 		       MC_CMD_PTP_IN_TIME_EVENT_UNSUBSCRIBE_SINGLE);
4618 	MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_UNSUBSCRIBE_QUEUE,
4619 		       channel->channel);
4620 
4621 	rc = efx_mcdi_rpc(channel->efx, MC_CMD_PTP,
4622 			  inbuf, sizeof(inbuf), NULL, 0, NULL);
4623 
4624 	return rc;
4625 }
4626 
4627 static int efx_ef10_ptp_set_ts_sync_events(struct efx_nic *efx, bool en,
4628 					   bool temp)
4629 {
4630 	int (*set)(struct efx_channel *channel, bool temp);
4631 	struct efx_channel *channel;
4632 
4633 	set = en ?
4634 	      efx_ef10_rx_enable_timestamping :
4635 	      efx_ef10_rx_disable_timestamping;
4636 
4637 	efx_for_each_channel(channel, efx) {
4638 		int rc = set(channel, temp);
4639 		if (en && rc != 0) {
4640 			efx_ef10_ptp_set_ts_sync_events(efx, false, temp);
4641 			return rc;
4642 		}
4643 	}
4644 
4645 	return 0;
4646 }
4647 
4648 static int efx_ef10_ptp_set_ts_config_vf(struct efx_nic *efx,
4649 					 struct hwtstamp_config *init)
4650 {
4651 	return -EOPNOTSUPP;
4652 }
4653 
4654 static int efx_ef10_ptp_set_ts_config(struct efx_nic *efx,
4655 				      struct hwtstamp_config *init)
4656 {
4657 	int rc;
4658 
4659 	switch (init->rx_filter) {
4660 	case HWTSTAMP_FILTER_NONE:
4661 		efx_ef10_ptp_set_ts_sync_events(efx, false, false);
4662 		/* if TX timestamping is still requested then leave PTP on */
4663 		return efx_ptp_change_mode(efx,
4664 					   init->tx_type != HWTSTAMP_TX_OFF, 0);
4665 	case HWTSTAMP_FILTER_ALL:
4666 	case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
4667 	case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
4668 	case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
4669 	case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
4670 	case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
4671 	case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
4672 	case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
4673 	case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
4674 	case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
4675 	case HWTSTAMP_FILTER_PTP_V2_EVENT:
4676 	case HWTSTAMP_FILTER_PTP_V2_SYNC:
4677 	case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
4678 		init->rx_filter = HWTSTAMP_FILTER_ALL;
4679 		rc = efx_ptp_change_mode(efx, true, 0);
4680 		if (!rc)
4681 			rc = efx_ef10_ptp_set_ts_sync_events(efx, true, false);
4682 		if (rc)
4683 			efx_ptp_change_mode(efx, false, 0);
4684 		return rc;
4685 	default:
4686 		return -ERANGE;
4687 	}
4688 }
4689 
4690 const struct efx_nic_type efx_hunt_a0_vf_nic_type = {
4691 	.is_vf = true,
4692 	.mem_bar = EFX_MEM_VF_BAR,
4693 	.mem_map_size = efx_ef10_mem_map_size,
4694 	.probe = efx_ef10_probe_vf,
4695 	.remove = efx_ef10_remove,
4696 	.dimension_resources = efx_ef10_dimension_resources,
4697 	.init = efx_ef10_init_nic,
4698 	.fini = efx_port_dummy_op_void,
4699 	.map_reset_reason = efx_ef10_map_reset_reason,
4700 	.map_reset_flags = efx_ef10_map_reset_flags,
4701 	.reset = efx_ef10_reset,
4702 	.probe_port = efx_mcdi_port_probe,
4703 	.remove_port = efx_mcdi_port_remove,
4704 	.fini_dmaq = efx_ef10_fini_dmaq,
4705 	.prepare_flr = efx_ef10_prepare_flr,
4706 	.finish_flr = efx_port_dummy_op_void,
4707 	.describe_stats = efx_ef10_describe_stats,
4708 	.update_stats = efx_ef10_update_stats_vf,
4709 	.start_stats = efx_port_dummy_op_void,
4710 	.pull_stats = efx_port_dummy_op_void,
4711 	.stop_stats = efx_port_dummy_op_void,
4712 	.set_id_led = efx_mcdi_set_id_led,
4713 	.push_irq_moderation = efx_ef10_push_irq_moderation,
4714 	.reconfigure_mac = efx_ef10_mac_reconfigure_vf,
4715 	.check_mac_fault = efx_mcdi_mac_check_fault,
4716 	.reconfigure_port = efx_mcdi_port_reconfigure,
4717 	.get_wol = efx_ef10_get_wol_vf,
4718 	.set_wol = efx_ef10_set_wol_vf,
4719 	.resume_wol = efx_port_dummy_op_void,
4720 	.mcdi_request = efx_ef10_mcdi_request,
4721 	.mcdi_poll_response = efx_ef10_mcdi_poll_response,
4722 	.mcdi_read_response = efx_ef10_mcdi_read_response,
4723 	.mcdi_poll_reboot = efx_ef10_mcdi_poll_reboot,
4724 	.mcdi_reboot_detected = efx_ef10_mcdi_reboot_detected,
4725 	.irq_enable_master = efx_port_dummy_op_void,
4726 	.irq_test_generate = efx_ef10_irq_test_generate,
4727 	.irq_disable_non_ev = efx_port_dummy_op_void,
4728 	.irq_handle_msi = efx_ef10_msi_interrupt,
4729 	.irq_handle_legacy = efx_ef10_legacy_interrupt,
4730 	.tx_probe = efx_ef10_tx_probe,
4731 	.tx_init = efx_ef10_tx_init,
4732 	.tx_remove = efx_ef10_tx_remove,
4733 	.tx_write = efx_ef10_tx_write,
4734 	.rx_push_rss_config = efx_ef10_vf_rx_push_rss_config,
4735 	.rx_probe = efx_ef10_rx_probe,
4736 	.rx_init = efx_ef10_rx_init,
4737 	.rx_remove = efx_ef10_rx_remove,
4738 	.rx_write = efx_ef10_rx_write,
4739 	.rx_defer_refill = efx_ef10_rx_defer_refill,
4740 	.ev_probe = efx_ef10_ev_probe,
4741 	.ev_init = efx_ef10_ev_init,
4742 	.ev_fini = efx_ef10_ev_fini,
4743 	.ev_remove = efx_ef10_ev_remove,
4744 	.ev_process = efx_ef10_ev_process,
4745 	.ev_read_ack = efx_ef10_ev_read_ack,
4746 	.ev_test_generate = efx_ef10_ev_test_generate,
4747 	.filter_table_probe = efx_ef10_filter_table_probe,
4748 	.filter_table_restore = efx_ef10_filter_table_restore,
4749 	.filter_table_remove = efx_ef10_filter_table_remove,
4750 	.filter_update_rx_scatter = efx_ef10_filter_update_rx_scatter,
4751 	.filter_insert = efx_ef10_filter_insert,
4752 	.filter_remove_safe = efx_ef10_filter_remove_safe,
4753 	.filter_get_safe = efx_ef10_filter_get_safe,
4754 	.filter_clear_rx = efx_ef10_filter_clear_rx,
4755 	.filter_count_rx_used = efx_ef10_filter_count_rx_used,
4756 	.filter_get_rx_id_limit = efx_ef10_filter_get_rx_id_limit,
4757 	.filter_get_rx_ids = efx_ef10_filter_get_rx_ids,
4758 #ifdef CONFIG_RFS_ACCEL
4759 	.filter_rfs_insert = efx_ef10_filter_rfs_insert,
4760 	.filter_rfs_expire_one = efx_ef10_filter_rfs_expire_one,
4761 #endif
4762 #ifdef CONFIG_SFC_MTD
4763 	.mtd_probe = efx_port_dummy_op_int,
4764 #endif
4765 	.ptp_write_host_time = efx_ef10_ptp_write_host_time_vf,
4766 	.ptp_set_ts_config = efx_ef10_ptp_set_ts_config_vf,
4767 #ifdef CONFIG_SFC_SRIOV
4768 	.vswitching_probe = efx_ef10_vswitching_probe_vf,
4769 	.vswitching_restore = efx_ef10_vswitching_restore_vf,
4770 	.vswitching_remove = efx_ef10_vswitching_remove_vf,
4771 	.sriov_get_phys_port_id = efx_ef10_sriov_get_phys_port_id,
4772 #endif
4773 	.get_mac_address = efx_ef10_get_mac_address_vf,
4774 	.set_mac_address = efx_ef10_set_mac_address,
4775 
4776 	.revision = EFX_REV_HUNT_A0,
4777 	.max_dma_mask = DMA_BIT_MASK(ESF_DZ_TX_KER_BUF_ADDR_WIDTH),
4778 	.rx_prefix_size = ES_DZ_RX_PREFIX_SIZE,
4779 	.rx_hash_offset = ES_DZ_RX_PREFIX_HASH_OFST,
4780 	.rx_ts_offset = ES_DZ_RX_PREFIX_TSTAMP_OFST,
4781 	.can_rx_scatter = true,
4782 	.always_rx_scatter = true,
4783 	.max_interrupt_mode = EFX_INT_MODE_MSIX,
4784 	.timer_period_max = 1 << ERF_DD_EVQ_IND_TIMER_VAL_WIDTH,
4785 	.offload_features = (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
4786 			     NETIF_F_RXHASH | NETIF_F_NTUPLE),
4787 	.mcdi_max_ver = 2,
4788 	.max_rx_ip_filters = HUNT_FILTER_TBL_ROWS,
4789 	.hwtstamp_filters = 1 << HWTSTAMP_FILTER_NONE |
4790 			    1 << HWTSTAMP_FILTER_ALL,
4791 };
4792 
4793 const struct efx_nic_type efx_hunt_a0_nic_type = {
4794 	.is_vf = false,
4795 	.mem_bar = EFX_MEM_BAR,
4796 	.mem_map_size = efx_ef10_mem_map_size,
4797 	.probe = efx_ef10_probe_pf,
4798 	.remove = efx_ef10_remove,
4799 	.dimension_resources = efx_ef10_dimension_resources,
4800 	.init = efx_ef10_init_nic,
4801 	.fini = efx_port_dummy_op_void,
4802 	.map_reset_reason = efx_ef10_map_reset_reason,
4803 	.map_reset_flags = efx_ef10_map_reset_flags,
4804 	.reset = efx_ef10_reset,
4805 	.probe_port = efx_mcdi_port_probe,
4806 	.remove_port = efx_mcdi_port_remove,
4807 	.fini_dmaq = efx_ef10_fini_dmaq,
4808 	.prepare_flr = efx_ef10_prepare_flr,
4809 	.finish_flr = efx_port_dummy_op_void,
4810 	.describe_stats = efx_ef10_describe_stats,
4811 	.update_stats = efx_ef10_update_stats_pf,
4812 	.start_stats = efx_mcdi_mac_start_stats,
4813 	.pull_stats = efx_mcdi_mac_pull_stats,
4814 	.stop_stats = efx_mcdi_mac_stop_stats,
4815 	.set_id_led = efx_mcdi_set_id_led,
4816 	.push_irq_moderation = efx_ef10_push_irq_moderation,
4817 	.reconfigure_mac = efx_ef10_mac_reconfigure,
4818 	.check_mac_fault = efx_mcdi_mac_check_fault,
4819 	.reconfigure_port = efx_mcdi_port_reconfigure,
4820 	.get_wol = efx_ef10_get_wol,
4821 	.set_wol = efx_ef10_set_wol,
4822 	.resume_wol = efx_port_dummy_op_void,
4823 	.test_chip = efx_ef10_test_chip,
4824 	.test_nvram = efx_mcdi_nvram_test_all,
4825 	.mcdi_request = efx_ef10_mcdi_request,
4826 	.mcdi_poll_response = efx_ef10_mcdi_poll_response,
4827 	.mcdi_read_response = efx_ef10_mcdi_read_response,
4828 	.mcdi_poll_reboot = efx_ef10_mcdi_poll_reboot,
4829 	.mcdi_reboot_detected = efx_ef10_mcdi_reboot_detected,
4830 	.irq_enable_master = efx_port_dummy_op_void,
4831 	.irq_test_generate = efx_ef10_irq_test_generate,
4832 	.irq_disable_non_ev = efx_port_dummy_op_void,
4833 	.irq_handle_msi = efx_ef10_msi_interrupt,
4834 	.irq_handle_legacy = efx_ef10_legacy_interrupt,
4835 	.tx_probe = efx_ef10_tx_probe,
4836 	.tx_init = efx_ef10_tx_init,
4837 	.tx_remove = efx_ef10_tx_remove,
4838 	.tx_write = efx_ef10_tx_write,
4839 	.rx_push_rss_config = efx_ef10_pf_rx_push_rss_config,
4840 	.rx_probe = efx_ef10_rx_probe,
4841 	.rx_init = efx_ef10_rx_init,
4842 	.rx_remove = efx_ef10_rx_remove,
4843 	.rx_write = efx_ef10_rx_write,
4844 	.rx_defer_refill = efx_ef10_rx_defer_refill,
4845 	.ev_probe = efx_ef10_ev_probe,
4846 	.ev_init = efx_ef10_ev_init,
4847 	.ev_fini = efx_ef10_ev_fini,
4848 	.ev_remove = efx_ef10_ev_remove,
4849 	.ev_process = efx_ef10_ev_process,
4850 	.ev_read_ack = efx_ef10_ev_read_ack,
4851 	.ev_test_generate = efx_ef10_ev_test_generate,
4852 	.filter_table_probe = efx_ef10_filter_table_probe,
4853 	.filter_table_restore = efx_ef10_filter_table_restore,
4854 	.filter_table_remove = efx_ef10_filter_table_remove,
4855 	.filter_update_rx_scatter = efx_ef10_filter_update_rx_scatter,
4856 	.filter_insert = efx_ef10_filter_insert,
4857 	.filter_remove_safe = efx_ef10_filter_remove_safe,
4858 	.filter_get_safe = efx_ef10_filter_get_safe,
4859 	.filter_clear_rx = efx_ef10_filter_clear_rx,
4860 	.filter_count_rx_used = efx_ef10_filter_count_rx_used,
4861 	.filter_get_rx_id_limit = efx_ef10_filter_get_rx_id_limit,
4862 	.filter_get_rx_ids = efx_ef10_filter_get_rx_ids,
4863 #ifdef CONFIG_RFS_ACCEL
4864 	.filter_rfs_insert = efx_ef10_filter_rfs_insert,
4865 	.filter_rfs_expire_one = efx_ef10_filter_rfs_expire_one,
4866 #endif
4867 #ifdef CONFIG_SFC_MTD
4868 	.mtd_probe = efx_ef10_mtd_probe,
4869 	.mtd_rename = efx_mcdi_mtd_rename,
4870 	.mtd_read = efx_mcdi_mtd_read,
4871 	.mtd_erase = efx_mcdi_mtd_erase,
4872 	.mtd_write = efx_mcdi_mtd_write,
4873 	.mtd_sync = efx_mcdi_mtd_sync,
4874 #endif
4875 	.ptp_write_host_time = efx_ef10_ptp_write_host_time,
4876 	.ptp_set_ts_sync_events = efx_ef10_ptp_set_ts_sync_events,
4877 	.ptp_set_ts_config = efx_ef10_ptp_set_ts_config,
4878 #ifdef CONFIG_SFC_SRIOV
4879 	.sriov_configure = efx_ef10_sriov_configure,
4880 	.sriov_init = efx_ef10_sriov_init,
4881 	.sriov_fini = efx_ef10_sriov_fini,
4882 	.sriov_wanted = efx_ef10_sriov_wanted,
4883 	.sriov_reset = efx_ef10_sriov_reset,
4884 	.sriov_flr = efx_ef10_sriov_flr,
4885 	.sriov_set_vf_mac = efx_ef10_sriov_set_vf_mac,
4886 	.sriov_set_vf_vlan = efx_ef10_sriov_set_vf_vlan,
4887 	.sriov_set_vf_spoofchk = efx_ef10_sriov_set_vf_spoofchk,
4888 	.sriov_get_vf_config = efx_ef10_sriov_get_vf_config,
4889 	.sriov_set_vf_link_state = efx_ef10_sriov_set_vf_link_state,
4890 	.vswitching_probe = efx_ef10_vswitching_probe_pf,
4891 	.vswitching_restore = efx_ef10_vswitching_restore_pf,
4892 	.vswitching_remove = efx_ef10_vswitching_remove_pf,
4893 #endif
4894 	.get_mac_address = efx_ef10_get_mac_address_pf,
4895 	.set_mac_address = efx_ef10_set_mac_address,
4896 
4897 	.revision = EFX_REV_HUNT_A0,
4898 	.max_dma_mask = DMA_BIT_MASK(ESF_DZ_TX_KER_BUF_ADDR_WIDTH),
4899 	.rx_prefix_size = ES_DZ_RX_PREFIX_SIZE,
4900 	.rx_hash_offset = ES_DZ_RX_PREFIX_HASH_OFST,
4901 	.rx_ts_offset = ES_DZ_RX_PREFIX_TSTAMP_OFST,
4902 	.can_rx_scatter = true,
4903 	.always_rx_scatter = true,
4904 	.max_interrupt_mode = EFX_INT_MODE_MSIX,
4905 	.timer_period_max = 1 << ERF_DD_EVQ_IND_TIMER_VAL_WIDTH,
4906 	.offload_features = (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
4907 			     NETIF_F_RXHASH | NETIF_F_NTUPLE),
4908 	.mcdi_max_ver = 2,
4909 	.max_rx_ip_filters = HUNT_FILTER_TBL_ROWS,
4910 	.hwtstamp_filters = 1 << HWTSTAMP_FILTER_NONE |
4911 			    1 << HWTSTAMP_FILTER_ALL,
4912 };
4913