xref: /openbmc/linux/drivers/net/ethernet/sfc/ef10.c (revision 9cfc5c90)
1 /****************************************************************************
2  * Driver for Solarflare network controllers and boards
3  * Copyright 2012-2013 Solarflare Communications Inc.
4  *
5  * This program is free software; you can redistribute it and/or modify it
6  * under the terms of the GNU General Public License version 2 as published
7  * by the Free Software Foundation, incorporated herein by reference.
8  */
9 
10 #include "net_driver.h"
11 #include "ef10_regs.h"
12 #include "io.h"
13 #include "mcdi.h"
14 #include "mcdi_pcol.h"
15 #include "nic.h"
16 #include "workarounds.h"
17 #include "selftest.h"
18 #include "ef10_sriov.h"
19 #include <linux/in.h>
20 #include <linux/jhash.h>
21 #include <linux/wait.h>
22 #include <linux/workqueue.h>
23 
24 /* Hardware control for EF10 architecture including 'Huntington'. */
25 
26 #define EFX_EF10_DRVGEN_EV		7
27 enum {
28 	EFX_EF10_TEST = 1,
29 	EFX_EF10_REFILL,
30 };
31 
32 /* The reserved RSS context value */
33 #define EFX_EF10_RSS_CONTEXT_INVALID	0xffffffff
34 /* The maximum size of a shared RSS context */
35 /* TODO: this should really be from the mcdi protocol export */
36 #define EFX_EF10_MAX_SHARED_RSS_CONTEXT_SIZE 64UL
37 
38 /* The filter table(s) are managed by firmware and we have write-only
39  * access.  When removing filters we must identify them to the
40  * firmware by a 64-bit handle, but this is too wide for Linux kernel
41  * interfaces (32-bit for RX NFC, 16-bit for RFS).  Also, we need to
42  * be able to tell in advance whether a requested insertion will
43  * replace an existing filter.  Therefore we maintain a software hash
44  * table, which should be at least as large as the hardware hash
45  * table.
46  *
47  * Huntington has a single 8K filter table shared between all filter
48  * types and both ports.
49  */
50 #define HUNT_FILTER_TBL_ROWS 8192
51 
52 #define EFX_EF10_FILTER_ID_INVALID 0xffff
53 struct efx_ef10_dev_addr {
54 	u8 addr[ETH_ALEN];
55 	u16 id;
56 };
57 
58 struct efx_ef10_filter_table {
59 /* The RX match field masks supported by this fw & hw, in order of priority */
60 	enum efx_filter_match_flags rx_match_flags[
61 		MC_CMD_GET_PARSER_DISP_INFO_OUT_SUPPORTED_MATCHES_MAXNUM];
62 	unsigned int rx_match_count;
63 
64 	struct {
65 		unsigned long spec;	/* pointer to spec plus flag bits */
66 /* BUSY flag indicates that an update is in progress.  AUTO_OLD is
67  * used to mark and sweep MAC filters for the device address lists.
68  */
69 #define EFX_EF10_FILTER_FLAG_BUSY	1UL
70 #define EFX_EF10_FILTER_FLAG_AUTO_OLD	2UL
71 #define EFX_EF10_FILTER_FLAGS		3UL
72 		u64 handle;		/* firmware handle */
73 	} *entry;
74 	wait_queue_head_t waitq;
75 /* Shadow of net_device address lists, guarded by mac_lock */
76 #define EFX_EF10_FILTER_DEV_UC_MAX	32
77 #define EFX_EF10_FILTER_DEV_MC_MAX	256
78 	struct efx_ef10_dev_addr dev_uc_list[EFX_EF10_FILTER_DEV_UC_MAX];
79 	struct efx_ef10_dev_addr dev_mc_list[EFX_EF10_FILTER_DEV_MC_MAX];
80 	int dev_uc_count;
81 	int dev_mc_count;
82 /* Indices (like efx_ef10_dev_addr.id) for promisc/allmulti filters */
83 	u16 ucdef_id;
84 	u16 bcast_id;
85 	u16 mcdef_id;
86 };
87 
88 /* An arbitrary search limit for the software hash table */
89 #define EFX_EF10_FILTER_SEARCH_LIMIT 200
90 
91 static void efx_ef10_rx_free_indir_table(struct efx_nic *efx);
92 static void efx_ef10_filter_table_remove(struct efx_nic *efx);
93 
94 static int efx_ef10_get_warm_boot_count(struct efx_nic *efx)
95 {
96 	efx_dword_t reg;
97 
98 	efx_readd(efx, &reg, ER_DZ_BIU_MC_SFT_STATUS);
99 	return EFX_DWORD_FIELD(reg, EFX_WORD_1) == 0xb007 ?
100 		EFX_DWORD_FIELD(reg, EFX_WORD_0) : -EIO;
101 }
102 
103 static unsigned int efx_ef10_mem_map_size(struct efx_nic *efx)
104 {
105 	int bar;
106 
107 	bar = efx->type->mem_bar;
108 	return resource_size(&efx->pci_dev->resource[bar]);
109 }
110 
111 static bool efx_ef10_is_vf(struct efx_nic *efx)
112 {
113 	return efx->type->is_vf;
114 }
115 
116 static int efx_ef10_get_pf_index(struct efx_nic *efx)
117 {
118 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_FUNCTION_INFO_OUT_LEN);
119 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
120 	size_t outlen;
121 	int rc;
122 
123 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_FUNCTION_INFO, NULL, 0, outbuf,
124 			  sizeof(outbuf), &outlen);
125 	if (rc)
126 		return rc;
127 	if (outlen < sizeof(outbuf))
128 		return -EIO;
129 
130 	nic_data->pf_index = MCDI_DWORD(outbuf, GET_FUNCTION_INFO_OUT_PF);
131 	return 0;
132 }
133 
134 #ifdef CONFIG_SFC_SRIOV
135 static int efx_ef10_get_vf_index(struct efx_nic *efx)
136 {
137 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_FUNCTION_INFO_OUT_LEN);
138 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
139 	size_t outlen;
140 	int rc;
141 
142 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_FUNCTION_INFO, NULL, 0, outbuf,
143 			  sizeof(outbuf), &outlen);
144 	if (rc)
145 		return rc;
146 	if (outlen < sizeof(outbuf))
147 		return -EIO;
148 
149 	nic_data->vf_index = MCDI_DWORD(outbuf, GET_FUNCTION_INFO_OUT_VF);
150 	return 0;
151 }
152 #endif
153 
154 static int efx_ef10_init_datapath_caps(struct efx_nic *efx)
155 {
156 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_CAPABILITIES_OUT_LEN);
157 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
158 	size_t outlen;
159 	int rc;
160 
161 	BUILD_BUG_ON(MC_CMD_GET_CAPABILITIES_IN_LEN != 0);
162 
163 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_CAPABILITIES, NULL, 0,
164 			  outbuf, sizeof(outbuf), &outlen);
165 	if (rc)
166 		return rc;
167 	if (outlen < sizeof(outbuf)) {
168 		netif_err(efx, drv, efx->net_dev,
169 			  "unable to read datapath firmware capabilities\n");
170 		return -EIO;
171 	}
172 
173 	nic_data->datapath_caps =
174 		MCDI_DWORD(outbuf, GET_CAPABILITIES_OUT_FLAGS1);
175 
176 	/* record the DPCPU firmware IDs to determine VEB vswitching support.
177 	 */
178 	nic_data->rx_dpcpu_fw_id =
179 		MCDI_WORD(outbuf, GET_CAPABILITIES_OUT_RX_DPCPU_FW_ID);
180 	nic_data->tx_dpcpu_fw_id =
181 		MCDI_WORD(outbuf, GET_CAPABILITIES_OUT_TX_DPCPU_FW_ID);
182 
183 	if (!(nic_data->datapath_caps &
184 	      (1 << MC_CMD_GET_CAPABILITIES_OUT_TX_TSO_LBN))) {
185 		netif_err(efx, drv, efx->net_dev,
186 			  "current firmware does not support TSO\n");
187 		return -ENODEV;
188 	}
189 
190 	if (!(nic_data->datapath_caps &
191 	      (1 << MC_CMD_GET_CAPABILITIES_OUT_RX_PREFIX_LEN_14_LBN))) {
192 		netif_err(efx, probe, efx->net_dev,
193 			  "current firmware does not support an RX prefix\n");
194 		return -ENODEV;
195 	}
196 
197 	return 0;
198 }
199 
200 static int efx_ef10_get_sysclk_freq(struct efx_nic *efx)
201 {
202 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_CLOCK_OUT_LEN);
203 	int rc;
204 
205 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_CLOCK, NULL, 0,
206 			  outbuf, sizeof(outbuf), NULL);
207 	if (rc)
208 		return rc;
209 	rc = MCDI_DWORD(outbuf, GET_CLOCK_OUT_SYS_FREQ);
210 	return rc > 0 ? rc : -ERANGE;
211 }
212 
213 static int efx_ef10_get_mac_address_pf(struct efx_nic *efx, u8 *mac_address)
214 {
215 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_MAC_ADDRESSES_OUT_LEN);
216 	size_t outlen;
217 	int rc;
218 
219 	BUILD_BUG_ON(MC_CMD_GET_MAC_ADDRESSES_IN_LEN != 0);
220 
221 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_MAC_ADDRESSES, NULL, 0,
222 			  outbuf, sizeof(outbuf), &outlen);
223 	if (rc)
224 		return rc;
225 	if (outlen < MC_CMD_GET_MAC_ADDRESSES_OUT_LEN)
226 		return -EIO;
227 
228 	ether_addr_copy(mac_address,
229 			MCDI_PTR(outbuf, GET_MAC_ADDRESSES_OUT_MAC_ADDR_BASE));
230 	return 0;
231 }
232 
233 static int efx_ef10_get_mac_address_vf(struct efx_nic *efx, u8 *mac_address)
234 {
235 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VPORT_GET_MAC_ADDRESSES_IN_LEN);
236 	MCDI_DECLARE_BUF(outbuf, MC_CMD_VPORT_GET_MAC_ADDRESSES_OUT_LENMAX);
237 	size_t outlen;
238 	int num_addrs, rc;
239 
240 	MCDI_SET_DWORD(inbuf, VPORT_GET_MAC_ADDRESSES_IN_VPORT_ID,
241 		       EVB_PORT_ID_ASSIGNED);
242 	rc = efx_mcdi_rpc(efx, MC_CMD_VPORT_GET_MAC_ADDRESSES, inbuf,
243 			  sizeof(inbuf), outbuf, sizeof(outbuf), &outlen);
244 
245 	if (rc)
246 		return rc;
247 	if (outlen < MC_CMD_VPORT_GET_MAC_ADDRESSES_OUT_LENMIN)
248 		return -EIO;
249 
250 	num_addrs = MCDI_DWORD(outbuf,
251 			       VPORT_GET_MAC_ADDRESSES_OUT_MACADDR_COUNT);
252 
253 	WARN_ON(num_addrs != 1);
254 
255 	ether_addr_copy(mac_address,
256 			MCDI_PTR(outbuf, VPORT_GET_MAC_ADDRESSES_OUT_MACADDR));
257 
258 	return 0;
259 }
260 
261 static ssize_t efx_ef10_show_link_control_flag(struct device *dev,
262 					       struct device_attribute *attr,
263 					       char *buf)
264 {
265 	struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));
266 
267 	return sprintf(buf, "%d\n",
268 		       ((efx->mcdi->fn_flags) &
269 			(1 << MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_LINKCTRL))
270 		       ? 1 : 0);
271 }
272 
273 static ssize_t efx_ef10_show_primary_flag(struct device *dev,
274 					  struct device_attribute *attr,
275 					  char *buf)
276 {
277 	struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));
278 
279 	return sprintf(buf, "%d\n",
280 		       ((efx->mcdi->fn_flags) &
281 			(1 << MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_PRIMARY))
282 		       ? 1 : 0);
283 }
284 
285 static DEVICE_ATTR(link_control_flag, 0444, efx_ef10_show_link_control_flag,
286 		   NULL);
287 static DEVICE_ATTR(primary_flag, 0444, efx_ef10_show_primary_flag, NULL);
288 
289 static int efx_ef10_probe(struct efx_nic *efx)
290 {
291 	struct efx_ef10_nic_data *nic_data;
292 	struct net_device *net_dev = efx->net_dev;
293 	int i, rc;
294 
295 	/* We can have one VI for each 8K region.  However, until we
296 	 * use TX option descriptors we need two TX queues per channel.
297 	 */
298 	efx->max_channels = min_t(unsigned int,
299 				  EFX_MAX_CHANNELS,
300 				  efx_ef10_mem_map_size(efx) /
301 				  (EFX_VI_PAGE_SIZE * EFX_TXQ_TYPES));
302 	efx->max_tx_channels = efx->max_channels;
303 	if (WARN_ON(efx->max_channels == 0))
304 		return -EIO;
305 
306 	nic_data = kzalloc(sizeof(*nic_data), GFP_KERNEL);
307 	if (!nic_data)
308 		return -ENOMEM;
309 	efx->nic_data = nic_data;
310 
311 	/* we assume later that we can copy from this buffer in dwords */
312 	BUILD_BUG_ON(MCDI_CTL_SDU_LEN_MAX_V2 % 4);
313 
314 	rc = efx_nic_alloc_buffer(efx, &nic_data->mcdi_buf,
315 				  8 + MCDI_CTL_SDU_LEN_MAX_V2, GFP_KERNEL);
316 	if (rc)
317 		goto fail1;
318 
319 	/* Get the MC's warm boot count.  In case it's rebooting right
320 	 * now, be prepared to retry.
321 	 */
322 	i = 0;
323 	for (;;) {
324 		rc = efx_ef10_get_warm_boot_count(efx);
325 		if (rc >= 0)
326 			break;
327 		if (++i == 5)
328 			goto fail2;
329 		ssleep(1);
330 	}
331 	nic_data->warm_boot_count = rc;
332 
333 	nic_data->rx_rss_context = EFX_EF10_RSS_CONTEXT_INVALID;
334 
335 	nic_data->vport_id = EVB_PORT_ID_ASSIGNED;
336 
337 	/* In case we're recovering from a crash (kexec), we want to
338 	 * cancel any outstanding request by the previous user of this
339 	 * function.  We send a special message using the least
340 	 * significant bits of the 'high' (doorbell) register.
341 	 */
342 	_efx_writed(efx, cpu_to_le32(1), ER_DZ_MC_DB_HWRD);
343 
344 	rc = efx_mcdi_init(efx);
345 	if (rc)
346 		goto fail2;
347 
348 	/* Reset (most) configuration for this function */
349 	rc = efx_mcdi_reset(efx, RESET_TYPE_ALL);
350 	if (rc)
351 		goto fail3;
352 
353 	/* Enable event logging */
354 	rc = efx_mcdi_log_ctrl(efx, true, false, 0);
355 	if (rc)
356 		goto fail3;
357 
358 	rc = device_create_file(&efx->pci_dev->dev,
359 				&dev_attr_link_control_flag);
360 	if (rc)
361 		goto fail3;
362 
363 	rc = device_create_file(&efx->pci_dev->dev, &dev_attr_primary_flag);
364 	if (rc)
365 		goto fail4;
366 
367 	rc = efx_ef10_get_pf_index(efx);
368 	if (rc)
369 		goto fail5;
370 
371 	rc = efx_ef10_init_datapath_caps(efx);
372 	if (rc < 0)
373 		goto fail5;
374 
375 	efx->rx_packet_len_offset =
376 		ES_DZ_RX_PREFIX_PKTLEN_OFST - ES_DZ_RX_PREFIX_SIZE;
377 
378 	rc = efx_mcdi_port_get_number(efx);
379 	if (rc < 0)
380 		goto fail5;
381 	efx->port_num = rc;
382 	net_dev->dev_port = rc;
383 
384 	rc = efx->type->get_mac_address(efx, efx->net_dev->perm_addr);
385 	if (rc)
386 		goto fail5;
387 
388 	rc = efx_ef10_get_sysclk_freq(efx);
389 	if (rc < 0)
390 		goto fail5;
391 	efx->timer_quantum_ns = 1536000 / rc; /* 1536 cycles */
392 
393 	/* Check whether firmware supports bug 35388 workaround.
394 	 * First try to enable it, then if we get EPERM, just
395 	 * ask if it's already enabled
396 	 */
397 	rc = efx_mcdi_set_workaround(efx, MC_CMD_WORKAROUND_BUG35388, true, NULL);
398 	if (rc == 0) {
399 		nic_data->workaround_35388 = true;
400 	} else if (rc == -EPERM) {
401 		unsigned int enabled;
402 
403 		rc = efx_mcdi_get_workarounds(efx, NULL, &enabled);
404 		if (rc)
405 			goto fail3;
406 		nic_data->workaround_35388 = enabled &
407 			MC_CMD_GET_WORKAROUNDS_OUT_BUG35388;
408 	} else if (rc != -ENOSYS && rc != -ENOENT) {
409 		goto fail5;
410 	}
411 	netif_dbg(efx, probe, efx->net_dev,
412 		  "workaround for bug 35388 is %sabled\n",
413 		  nic_data->workaround_35388 ? "en" : "dis");
414 
415 	rc = efx_mcdi_mon_probe(efx);
416 	if (rc && rc != -EPERM)
417 		goto fail5;
418 
419 	efx_ptp_probe(efx, NULL);
420 
421 #ifdef CONFIG_SFC_SRIOV
422 	if ((efx->pci_dev->physfn) && (!efx->pci_dev->is_physfn)) {
423 		struct pci_dev *pci_dev_pf = efx->pci_dev->physfn;
424 		struct efx_nic *efx_pf = pci_get_drvdata(pci_dev_pf);
425 
426 		efx_pf->type->get_mac_address(efx_pf, nic_data->port_id);
427 	} else
428 #endif
429 		ether_addr_copy(nic_data->port_id, efx->net_dev->perm_addr);
430 
431 	return 0;
432 
433 fail5:
434 	device_remove_file(&efx->pci_dev->dev, &dev_attr_primary_flag);
435 fail4:
436 	device_remove_file(&efx->pci_dev->dev, &dev_attr_link_control_flag);
437 fail3:
438 	efx_mcdi_fini(efx);
439 fail2:
440 	efx_nic_free_buffer(efx, &nic_data->mcdi_buf);
441 fail1:
442 	kfree(nic_data);
443 	efx->nic_data = NULL;
444 	return rc;
445 }
446 
447 static int efx_ef10_free_vis(struct efx_nic *efx)
448 {
449 	MCDI_DECLARE_BUF_ERR(outbuf);
450 	size_t outlen;
451 	int rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FREE_VIS, NULL, 0,
452 				    outbuf, sizeof(outbuf), &outlen);
453 
454 	/* -EALREADY means nothing to free, so ignore */
455 	if (rc == -EALREADY)
456 		rc = 0;
457 	if (rc)
458 		efx_mcdi_display_error(efx, MC_CMD_FREE_VIS, 0, outbuf, outlen,
459 				       rc);
460 	return rc;
461 }
462 
463 #ifdef EFX_USE_PIO
464 
465 static void efx_ef10_free_piobufs(struct efx_nic *efx)
466 {
467 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
468 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FREE_PIOBUF_IN_LEN);
469 	unsigned int i;
470 	int rc;
471 
472 	BUILD_BUG_ON(MC_CMD_FREE_PIOBUF_OUT_LEN != 0);
473 
474 	for (i = 0; i < nic_data->n_piobufs; i++) {
475 		MCDI_SET_DWORD(inbuf, FREE_PIOBUF_IN_PIOBUF_HANDLE,
476 			       nic_data->piobuf_handle[i]);
477 		rc = efx_mcdi_rpc(efx, MC_CMD_FREE_PIOBUF, inbuf, sizeof(inbuf),
478 				  NULL, 0, NULL);
479 		WARN_ON(rc);
480 	}
481 
482 	nic_data->n_piobufs = 0;
483 }
484 
485 static int efx_ef10_alloc_piobufs(struct efx_nic *efx, unsigned int n)
486 {
487 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
488 	MCDI_DECLARE_BUF(outbuf, MC_CMD_ALLOC_PIOBUF_OUT_LEN);
489 	unsigned int i;
490 	size_t outlen;
491 	int rc = 0;
492 
493 	BUILD_BUG_ON(MC_CMD_ALLOC_PIOBUF_IN_LEN != 0);
494 
495 	for (i = 0; i < n; i++) {
496 		rc = efx_mcdi_rpc(efx, MC_CMD_ALLOC_PIOBUF, NULL, 0,
497 				  outbuf, sizeof(outbuf), &outlen);
498 		if (rc)
499 			break;
500 		if (outlen < MC_CMD_ALLOC_PIOBUF_OUT_LEN) {
501 			rc = -EIO;
502 			break;
503 		}
504 		nic_data->piobuf_handle[i] =
505 			MCDI_DWORD(outbuf, ALLOC_PIOBUF_OUT_PIOBUF_HANDLE);
506 		netif_dbg(efx, probe, efx->net_dev,
507 			  "allocated PIO buffer %u handle %x\n", i,
508 			  nic_data->piobuf_handle[i]);
509 	}
510 
511 	nic_data->n_piobufs = i;
512 	if (rc)
513 		efx_ef10_free_piobufs(efx);
514 	return rc;
515 }
516 
517 static int efx_ef10_link_piobufs(struct efx_nic *efx)
518 {
519 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
520 	_MCDI_DECLARE_BUF(inbuf,
521 			  max(MC_CMD_LINK_PIOBUF_IN_LEN,
522 			      MC_CMD_UNLINK_PIOBUF_IN_LEN));
523 	struct efx_channel *channel;
524 	struct efx_tx_queue *tx_queue;
525 	unsigned int offset, index;
526 	int rc;
527 
528 	BUILD_BUG_ON(MC_CMD_LINK_PIOBUF_OUT_LEN != 0);
529 	BUILD_BUG_ON(MC_CMD_UNLINK_PIOBUF_OUT_LEN != 0);
530 
531 	memset(inbuf, 0, sizeof(inbuf));
532 
533 	/* Link a buffer to each VI in the write-combining mapping */
534 	for (index = 0; index < nic_data->n_piobufs; ++index) {
535 		MCDI_SET_DWORD(inbuf, LINK_PIOBUF_IN_PIOBUF_HANDLE,
536 			       nic_data->piobuf_handle[index]);
537 		MCDI_SET_DWORD(inbuf, LINK_PIOBUF_IN_TXQ_INSTANCE,
538 			       nic_data->pio_write_vi_base + index);
539 		rc = efx_mcdi_rpc(efx, MC_CMD_LINK_PIOBUF,
540 				  inbuf, MC_CMD_LINK_PIOBUF_IN_LEN,
541 				  NULL, 0, NULL);
542 		if (rc) {
543 			netif_err(efx, drv, efx->net_dev,
544 				  "failed to link VI %u to PIO buffer %u (%d)\n",
545 				  nic_data->pio_write_vi_base + index, index,
546 				  rc);
547 			goto fail;
548 		}
549 		netif_dbg(efx, probe, efx->net_dev,
550 			  "linked VI %u to PIO buffer %u\n",
551 			  nic_data->pio_write_vi_base + index, index);
552 	}
553 
554 	/* Link a buffer to each TX queue */
555 	efx_for_each_channel(channel, efx) {
556 		efx_for_each_channel_tx_queue(tx_queue, channel) {
557 			/* We assign the PIO buffers to queues in
558 			 * reverse order to allow for the following
559 			 * special case.
560 			 */
561 			offset = ((efx->tx_channel_offset + efx->n_tx_channels -
562 				   tx_queue->channel->channel - 1) *
563 				  efx_piobuf_size);
564 			index = offset / ER_DZ_TX_PIOBUF_SIZE;
565 			offset = offset % ER_DZ_TX_PIOBUF_SIZE;
566 
567 			/* When the host page size is 4K, the first
568 			 * host page in the WC mapping may be within
569 			 * the same VI page as the last TX queue.  We
570 			 * can only link one buffer to each VI.
571 			 */
572 			if (tx_queue->queue == nic_data->pio_write_vi_base) {
573 				BUG_ON(index != 0);
574 				rc = 0;
575 			} else {
576 				MCDI_SET_DWORD(inbuf,
577 					       LINK_PIOBUF_IN_PIOBUF_HANDLE,
578 					       nic_data->piobuf_handle[index]);
579 				MCDI_SET_DWORD(inbuf,
580 					       LINK_PIOBUF_IN_TXQ_INSTANCE,
581 					       tx_queue->queue);
582 				rc = efx_mcdi_rpc(efx, MC_CMD_LINK_PIOBUF,
583 						  inbuf, MC_CMD_LINK_PIOBUF_IN_LEN,
584 						  NULL, 0, NULL);
585 			}
586 
587 			if (rc) {
588 				/* This is non-fatal; the TX path just
589 				 * won't use PIO for this queue
590 				 */
591 				netif_err(efx, drv, efx->net_dev,
592 					  "failed to link VI %u to PIO buffer %u (%d)\n",
593 					  tx_queue->queue, index, rc);
594 				tx_queue->piobuf = NULL;
595 			} else {
596 				tx_queue->piobuf =
597 					nic_data->pio_write_base +
598 					index * EFX_VI_PAGE_SIZE + offset;
599 				tx_queue->piobuf_offset = offset;
600 				netif_dbg(efx, probe, efx->net_dev,
601 					  "linked VI %u to PIO buffer %u offset %x addr %p\n",
602 					  tx_queue->queue, index,
603 					  tx_queue->piobuf_offset,
604 					  tx_queue->piobuf);
605 			}
606 		}
607 	}
608 
609 	return 0;
610 
611 fail:
612 	while (index--) {
613 		MCDI_SET_DWORD(inbuf, UNLINK_PIOBUF_IN_TXQ_INSTANCE,
614 			       nic_data->pio_write_vi_base + index);
615 		efx_mcdi_rpc(efx, MC_CMD_UNLINK_PIOBUF,
616 			     inbuf, MC_CMD_UNLINK_PIOBUF_IN_LEN,
617 			     NULL, 0, NULL);
618 	}
619 	return rc;
620 }
621 
622 #else /* !EFX_USE_PIO */
623 
624 static int efx_ef10_alloc_piobufs(struct efx_nic *efx, unsigned int n)
625 {
626 	return n == 0 ? 0 : -ENOBUFS;
627 }
628 
629 static int efx_ef10_link_piobufs(struct efx_nic *efx)
630 {
631 	return 0;
632 }
633 
634 static void efx_ef10_free_piobufs(struct efx_nic *efx)
635 {
636 }
637 
638 #endif /* EFX_USE_PIO */
639 
640 static void efx_ef10_remove(struct efx_nic *efx)
641 {
642 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
643 	int rc;
644 
645 #ifdef CONFIG_SFC_SRIOV
646 	struct efx_ef10_nic_data *nic_data_pf;
647 	struct pci_dev *pci_dev_pf;
648 	struct efx_nic *efx_pf;
649 	struct ef10_vf *vf;
650 
651 	if (efx->pci_dev->is_virtfn) {
652 		pci_dev_pf = efx->pci_dev->physfn;
653 		if (pci_dev_pf) {
654 			efx_pf = pci_get_drvdata(pci_dev_pf);
655 			nic_data_pf = efx_pf->nic_data;
656 			vf = nic_data_pf->vf + nic_data->vf_index;
657 			vf->efx = NULL;
658 		} else
659 			netif_info(efx, drv, efx->net_dev,
660 				   "Could not get the PF id from VF\n");
661 	}
662 #endif
663 
664 	efx_ptp_remove(efx);
665 
666 	efx_mcdi_mon_remove(efx);
667 
668 	efx_ef10_rx_free_indir_table(efx);
669 
670 	if (nic_data->wc_membase)
671 		iounmap(nic_data->wc_membase);
672 
673 	rc = efx_ef10_free_vis(efx);
674 	WARN_ON(rc != 0);
675 
676 	if (!nic_data->must_restore_piobufs)
677 		efx_ef10_free_piobufs(efx);
678 
679 	device_remove_file(&efx->pci_dev->dev, &dev_attr_primary_flag);
680 	device_remove_file(&efx->pci_dev->dev, &dev_attr_link_control_flag);
681 
682 	efx_mcdi_fini(efx);
683 	efx_nic_free_buffer(efx, &nic_data->mcdi_buf);
684 	kfree(nic_data);
685 }
686 
687 static int efx_ef10_probe_pf(struct efx_nic *efx)
688 {
689 	return efx_ef10_probe(efx);
690 }
691 
692 int efx_ef10_vadaptor_alloc(struct efx_nic *efx, unsigned int port_id)
693 {
694 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VADAPTOR_ALLOC_IN_LEN);
695 
696 	MCDI_SET_DWORD(inbuf, VADAPTOR_ALLOC_IN_UPSTREAM_PORT_ID, port_id);
697 	return efx_mcdi_rpc(efx, MC_CMD_VADAPTOR_ALLOC, inbuf, sizeof(inbuf),
698 			    NULL, 0, NULL);
699 }
700 
701 int efx_ef10_vadaptor_free(struct efx_nic *efx, unsigned int port_id)
702 {
703 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VADAPTOR_FREE_IN_LEN);
704 
705 	MCDI_SET_DWORD(inbuf, VADAPTOR_FREE_IN_UPSTREAM_PORT_ID, port_id);
706 	return efx_mcdi_rpc(efx, MC_CMD_VADAPTOR_FREE, inbuf, sizeof(inbuf),
707 			    NULL, 0, NULL);
708 }
709 
710 int efx_ef10_vport_add_mac(struct efx_nic *efx,
711 			   unsigned int port_id, u8 *mac)
712 {
713 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VPORT_ADD_MAC_ADDRESS_IN_LEN);
714 
715 	MCDI_SET_DWORD(inbuf, VPORT_ADD_MAC_ADDRESS_IN_VPORT_ID, port_id);
716 	ether_addr_copy(MCDI_PTR(inbuf, VPORT_ADD_MAC_ADDRESS_IN_MACADDR), mac);
717 
718 	return efx_mcdi_rpc(efx, MC_CMD_VPORT_ADD_MAC_ADDRESS, inbuf,
719 			    sizeof(inbuf), NULL, 0, NULL);
720 }
721 
722 int efx_ef10_vport_del_mac(struct efx_nic *efx,
723 			   unsigned int port_id, u8 *mac)
724 {
725 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VPORT_DEL_MAC_ADDRESS_IN_LEN);
726 
727 	MCDI_SET_DWORD(inbuf, VPORT_DEL_MAC_ADDRESS_IN_VPORT_ID, port_id);
728 	ether_addr_copy(MCDI_PTR(inbuf, VPORT_DEL_MAC_ADDRESS_IN_MACADDR), mac);
729 
730 	return efx_mcdi_rpc(efx, MC_CMD_VPORT_DEL_MAC_ADDRESS, inbuf,
731 			    sizeof(inbuf), NULL, 0, NULL);
732 }
733 
734 #ifdef CONFIG_SFC_SRIOV
735 static int efx_ef10_probe_vf(struct efx_nic *efx)
736 {
737 	int rc;
738 	struct pci_dev *pci_dev_pf;
739 
740 	/* If the parent PF has no VF data structure, it doesn't know about this
741 	 * VF so fail probe.  The VF needs to be re-created.  This can happen
742 	 * if the PF driver is unloaded while the VF is assigned to a guest.
743 	 */
744 	pci_dev_pf = efx->pci_dev->physfn;
745 	if (pci_dev_pf) {
746 		struct efx_nic *efx_pf = pci_get_drvdata(pci_dev_pf);
747 		struct efx_ef10_nic_data *nic_data_pf = efx_pf->nic_data;
748 
749 		if (!nic_data_pf->vf) {
750 			netif_info(efx, drv, efx->net_dev,
751 				   "The VF cannot link to its parent PF; "
752 				   "please destroy and re-create the VF\n");
753 			return -EBUSY;
754 		}
755 	}
756 
757 	rc = efx_ef10_probe(efx);
758 	if (rc)
759 		return rc;
760 
761 	rc = efx_ef10_get_vf_index(efx);
762 	if (rc)
763 		goto fail;
764 
765 	if (efx->pci_dev->is_virtfn) {
766 		if (efx->pci_dev->physfn) {
767 			struct efx_nic *efx_pf =
768 				pci_get_drvdata(efx->pci_dev->physfn);
769 			struct efx_ef10_nic_data *nic_data_p = efx_pf->nic_data;
770 			struct efx_ef10_nic_data *nic_data = efx->nic_data;
771 
772 			nic_data_p->vf[nic_data->vf_index].efx = efx;
773 			nic_data_p->vf[nic_data->vf_index].pci_dev =
774 				efx->pci_dev;
775 		} else
776 			netif_info(efx, drv, efx->net_dev,
777 				   "Could not get the PF id from VF\n");
778 	}
779 
780 	return 0;
781 
782 fail:
783 	efx_ef10_remove(efx);
784 	return rc;
785 }
786 #else
787 static int efx_ef10_probe_vf(struct efx_nic *efx __attribute__ ((unused)))
788 {
789 	return 0;
790 }
791 #endif
792 
793 static int efx_ef10_alloc_vis(struct efx_nic *efx,
794 			      unsigned int min_vis, unsigned int max_vis)
795 {
796 	MCDI_DECLARE_BUF(inbuf, MC_CMD_ALLOC_VIS_IN_LEN);
797 	MCDI_DECLARE_BUF(outbuf, MC_CMD_ALLOC_VIS_OUT_LEN);
798 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
799 	size_t outlen;
800 	int rc;
801 
802 	MCDI_SET_DWORD(inbuf, ALLOC_VIS_IN_MIN_VI_COUNT, min_vis);
803 	MCDI_SET_DWORD(inbuf, ALLOC_VIS_IN_MAX_VI_COUNT, max_vis);
804 	rc = efx_mcdi_rpc(efx, MC_CMD_ALLOC_VIS, inbuf, sizeof(inbuf),
805 			  outbuf, sizeof(outbuf), &outlen);
806 	if (rc != 0)
807 		return rc;
808 
809 	if (outlen < MC_CMD_ALLOC_VIS_OUT_LEN)
810 		return -EIO;
811 
812 	netif_dbg(efx, drv, efx->net_dev, "base VI is A0x%03x\n",
813 		  MCDI_DWORD(outbuf, ALLOC_VIS_OUT_VI_BASE));
814 
815 	nic_data->vi_base = MCDI_DWORD(outbuf, ALLOC_VIS_OUT_VI_BASE);
816 	nic_data->n_allocated_vis = MCDI_DWORD(outbuf, ALLOC_VIS_OUT_VI_COUNT);
817 	return 0;
818 }
819 
820 /* Note that the failure path of this function does not free
821  * resources, as this will be done by efx_ef10_remove().
822  */
823 static int efx_ef10_dimension_resources(struct efx_nic *efx)
824 {
825 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
826 	unsigned int uc_mem_map_size, wc_mem_map_size;
827 	unsigned int min_vis = max(EFX_TXQ_TYPES,
828 				   efx_separate_tx_channels ? 2 : 1);
829 	unsigned int channel_vis, pio_write_vi_base, max_vis;
830 	void __iomem *membase;
831 	int rc;
832 
833 	channel_vis = max(efx->n_channels, efx->n_tx_channels * EFX_TXQ_TYPES);
834 
835 #ifdef EFX_USE_PIO
836 	/* Try to allocate PIO buffers if wanted and if the full
837 	 * number of PIO buffers would be sufficient to allocate one
838 	 * copy-buffer per TX channel.  Failure is non-fatal, as there
839 	 * are only a small number of PIO buffers shared between all
840 	 * functions of the controller.
841 	 */
842 	if (efx_piobuf_size != 0 &&
843 	    ER_DZ_TX_PIOBUF_SIZE / efx_piobuf_size * EF10_TX_PIOBUF_COUNT >=
844 	    efx->n_tx_channels) {
845 		unsigned int n_piobufs =
846 			DIV_ROUND_UP(efx->n_tx_channels,
847 				     ER_DZ_TX_PIOBUF_SIZE / efx_piobuf_size);
848 
849 		rc = efx_ef10_alloc_piobufs(efx, n_piobufs);
850 		if (rc)
851 			netif_err(efx, probe, efx->net_dev,
852 				  "failed to allocate PIO buffers (%d)\n", rc);
853 		else
854 			netif_dbg(efx, probe, efx->net_dev,
855 				  "allocated %u PIO buffers\n", n_piobufs);
856 	}
857 #else
858 	nic_data->n_piobufs = 0;
859 #endif
860 
861 	/* PIO buffers should be mapped with write-combining enabled,
862 	 * and we want to make single UC and WC mappings rather than
863 	 * several of each (in fact that's the only option if host
864 	 * page size is >4K).  So we may allocate some extra VIs just
865 	 * for writing PIO buffers through.
866 	 *
867 	 * The UC mapping contains (channel_vis - 1) complete VIs and the
868 	 * first half of the next VI.  Then the WC mapping begins with
869 	 * the second half of this last VI.
870 	 */
871 	uc_mem_map_size = PAGE_ALIGN((channel_vis - 1) * EFX_VI_PAGE_SIZE +
872 				     ER_DZ_TX_PIOBUF);
873 	if (nic_data->n_piobufs) {
874 		/* pio_write_vi_base rounds down to give the number of complete
875 		 * VIs inside the UC mapping.
876 		 */
877 		pio_write_vi_base = uc_mem_map_size / EFX_VI_PAGE_SIZE;
878 		wc_mem_map_size = (PAGE_ALIGN((pio_write_vi_base +
879 					       nic_data->n_piobufs) *
880 					      EFX_VI_PAGE_SIZE) -
881 				   uc_mem_map_size);
882 		max_vis = pio_write_vi_base + nic_data->n_piobufs;
883 	} else {
884 		pio_write_vi_base = 0;
885 		wc_mem_map_size = 0;
886 		max_vis = channel_vis;
887 	}
888 
889 	/* In case the last attached driver failed to free VIs, do it now */
890 	rc = efx_ef10_free_vis(efx);
891 	if (rc != 0)
892 		return rc;
893 
894 	rc = efx_ef10_alloc_vis(efx, min_vis, max_vis);
895 	if (rc != 0)
896 		return rc;
897 
898 	if (nic_data->n_allocated_vis < channel_vis) {
899 		netif_info(efx, drv, efx->net_dev,
900 			   "Could not allocate enough VIs to satisfy RSS"
901 			   " requirements. Performance may not be optimal.\n");
902 		/* We didn't get the VIs to populate our channels.
903 		 * We could keep what we got but then we'd have more
904 		 * interrupts than we need.
905 		 * Instead calculate new max_channels and restart
906 		 */
907 		efx->max_channels = nic_data->n_allocated_vis;
908 		efx->max_tx_channels =
909 			nic_data->n_allocated_vis / EFX_TXQ_TYPES;
910 
911 		efx_ef10_free_vis(efx);
912 		return -EAGAIN;
913 	}
914 
915 	/* If we didn't get enough VIs to map all the PIO buffers, free the
916 	 * PIO buffers
917 	 */
918 	if (nic_data->n_piobufs &&
919 	    nic_data->n_allocated_vis <
920 	    pio_write_vi_base + nic_data->n_piobufs) {
921 		netif_dbg(efx, probe, efx->net_dev,
922 			  "%u VIs are not sufficient to map %u PIO buffers\n",
923 			  nic_data->n_allocated_vis, nic_data->n_piobufs);
924 		efx_ef10_free_piobufs(efx);
925 	}
926 
927 	/* Shrink the original UC mapping of the memory BAR */
928 	membase = ioremap_nocache(efx->membase_phys, uc_mem_map_size);
929 	if (!membase) {
930 		netif_err(efx, probe, efx->net_dev,
931 			  "could not shrink memory BAR to %x\n",
932 			  uc_mem_map_size);
933 		return -ENOMEM;
934 	}
935 	iounmap(efx->membase);
936 	efx->membase = membase;
937 
938 	/* Set up the WC mapping if needed */
939 	if (wc_mem_map_size) {
940 		nic_data->wc_membase = ioremap_wc(efx->membase_phys +
941 						  uc_mem_map_size,
942 						  wc_mem_map_size);
943 		if (!nic_data->wc_membase) {
944 			netif_err(efx, probe, efx->net_dev,
945 				  "could not allocate WC mapping of size %x\n",
946 				  wc_mem_map_size);
947 			return -ENOMEM;
948 		}
949 		nic_data->pio_write_vi_base = pio_write_vi_base;
950 		nic_data->pio_write_base =
951 			nic_data->wc_membase +
952 			(pio_write_vi_base * EFX_VI_PAGE_SIZE + ER_DZ_TX_PIOBUF -
953 			 uc_mem_map_size);
954 
955 		rc = efx_ef10_link_piobufs(efx);
956 		if (rc)
957 			efx_ef10_free_piobufs(efx);
958 	}
959 
960 	netif_dbg(efx, probe, efx->net_dev,
961 		  "memory BAR at %pa (virtual %p+%x UC, %p+%x WC)\n",
962 		  &efx->membase_phys, efx->membase, uc_mem_map_size,
963 		  nic_data->wc_membase, wc_mem_map_size);
964 
965 	return 0;
966 }
967 
968 static int efx_ef10_init_nic(struct efx_nic *efx)
969 {
970 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
971 	int rc;
972 
973 	if (nic_data->must_check_datapath_caps) {
974 		rc = efx_ef10_init_datapath_caps(efx);
975 		if (rc)
976 			return rc;
977 		nic_data->must_check_datapath_caps = false;
978 	}
979 
980 	if (nic_data->must_realloc_vis) {
981 		/* We cannot let the number of VIs change now */
982 		rc = efx_ef10_alloc_vis(efx, nic_data->n_allocated_vis,
983 					nic_data->n_allocated_vis);
984 		if (rc)
985 			return rc;
986 		nic_data->must_realloc_vis = false;
987 	}
988 
989 	if (nic_data->must_restore_piobufs && nic_data->n_piobufs) {
990 		rc = efx_ef10_alloc_piobufs(efx, nic_data->n_piobufs);
991 		if (rc == 0) {
992 			rc = efx_ef10_link_piobufs(efx);
993 			if (rc)
994 				efx_ef10_free_piobufs(efx);
995 		}
996 
997 		/* Log an error on failure, but this is non-fatal */
998 		if (rc)
999 			netif_err(efx, drv, efx->net_dev,
1000 				  "failed to restore PIO buffers (%d)\n", rc);
1001 		nic_data->must_restore_piobufs = false;
1002 	}
1003 
1004 	/* don't fail init if RSS setup doesn't work */
1005 	efx->type->rx_push_rss_config(efx, false, efx->rx_indir_table);
1006 
1007 	return 0;
1008 }
1009 
1010 static void efx_ef10_reset_mc_allocations(struct efx_nic *efx)
1011 {
1012 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1013 #ifdef CONFIG_SFC_SRIOV
1014 	unsigned int i;
1015 #endif
1016 
1017 	/* All our allocations have been reset */
1018 	nic_data->must_realloc_vis = true;
1019 	nic_data->must_restore_filters = true;
1020 	nic_data->must_restore_piobufs = true;
1021 	nic_data->rx_rss_context = EFX_EF10_RSS_CONTEXT_INVALID;
1022 
1023 	/* Driver-created vswitches and vports must be re-created */
1024 	nic_data->must_probe_vswitching = true;
1025 	nic_data->vport_id = EVB_PORT_ID_ASSIGNED;
1026 #ifdef CONFIG_SFC_SRIOV
1027 	if (nic_data->vf)
1028 		for (i = 0; i < efx->vf_count; i++)
1029 			nic_data->vf[i].vport_id = 0;
1030 #endif
1031 }
1032 
1033 static enum reset_type efx_ef10_map_reset_reason(enum reset_type reason)
1034 {
1035 	if (reason == RESET_TYPE_MC_FAILURE)
1036 		return RESET_TYPE_DATAPATH;
1037 
1038 	return efx_mcdi_map_reset_reason(reason);
1039 }
1040 
1041 static int efx_ef10_map_reset_flags(u32 *flags)
1042 {
1043 	enum {
1044 		EF10_RESET_PORT = ((ETH_RESET_MAC | ETH_RESET_PHY) <<
1045 				   ETH_RESET_SHARED_SHIFT),
1046 		EF10_RESET_MC = ((ETH_RESET_DMA | ETH_RESET_FILTER |
1047 				  ETH_RESET_OFFLOAD | ETH_RESET_MAC |
1048 				  ETH_RESET_PHY | ETH_RESET_MGMT) <<
1049 				 ETH_RESET_SHARED_SHIFT)
1050 	};
1051 
1052 	/* We assume for now that our PCI function is permitted to
1053 	 * reset everything.
1054 	 */
1055 
1056 	if ((*flags & EF10_RESET_MC) == EF10_RESET_MC) {
1057 		*flags &= ~EF10_RESET_MC;
1058 		return RESET_TYPE_WORLD;
1059 	}
1060 
1061 	if ((*flags & EF10_RESET_PORT) == EF10_RESET_PORT) {
1062 		*flags &= ~EF10_RESET_PORT;
1063 		return RESET_TYPE_ALL;
1064 	}
1065 
1066 	/* no invisible reset implemented */
1067 
1068 	return -EINVAL;
1069 }
1070 
1071 static int efx_ef10_reset(struct efx_nic *efx, enum reset_type reset_type)
1072 {
1073 	int rc = efx_mcdi_reset(efx, reset_type);
1074 
1075 	/* Unprivileged functions return -EPERM, but need to return success
1076 	 * here so that the datapath is brought back up.
1077 	 */
1078 	if (reset_type == RESET_TYPE_WORLD && rc == -EPERM)
1079 		rc = 0;
1080 
1081 	/* If it was a port reset, trigger reallocation of MC resources.
1082 	 * Note that on an MC reset nothing needs to be done now because we'll
1083 	 * detect the MC reset later and handle it then.
1084 	 * For an FLR, we never get an MC reset event, but the MC has reset all
1085 	 * resources assigned to us, so we have to trigger reallocation now.
1086 	 */
1087 	if ((reset_type == RESET_TYPE_ALL ||
1088 	     reset_type == RESET_TYPE_MCDI_TIMEOUT) && !rc)
1089 		efx_ef10_reset_mc_allocations(efx);
1090 	return rc;
1091 }
1092 
1093 #define EF10_DMA_STAT(ext_name, mcdi_name)			\
1094 	[EF10_STAT_ ## ext_name] =				\
1095 	{ #ext_name, 64, 8 * MC_CMD_MAC_ ## mcdi_name }
1096 #define EF10_DMA_INVIS_STAT(int_name, mcdi_name)		\
1097 	[EF10_STAT_ ## int_name] =				\
1098 	{ NULL, 64, 8 * MC_CMD_MAC_ ## mcdi_name }
1099 #define EF10_OTHER_STAT(ext_name)				\
1100 	[EF10_STAT_ ## ext_name] = { #ext_name, 0, 0 }
1101 #define GENERIC_SW_STAT(ext_name)				\
1102 	[GENERIC_STAT_ ## ext_name] = { #ext_name, 0, 0 }
1103 
1104 static const struct efx_hw_stat_desc efx_ef10_stat_desc[EF10_STAT_COUNT] = {
1105 	EF10_DMA_STAT(port_tx_bytes, TX_BYTES),
1106 	EF10_DMA_STAT(port_tx_packets, TX_PKTS),
1107 	EF10_DMA_STAT(port_tx_pause, TX_PAUSE_PKTS),
1108 	EF10_DMA_STAT(port_tx_control, TX_CONTROL_PKTS),
1109 	EF10_DMA_STAT(port_tx_unicast, TX_UNICAST_PKTS),
1110 	EF10_DMA_STAT(port_tx_multicast, TX_MULTICAST_PKTS),
1111 	EF10_DMA_STAT(port_tx_broadcast, TX_BROADCAST_PKTS),
1112 	EF10_DMA_STAT(port_tx_lt64, TX_LT64_PKTS),
1113 	EF10_DMA_STAT(port_tx_64, TX_64_PKTS),
1114 	EF10_DMA_STAT(port_tx_65_to_127, TX_65_TO_127_PKTS),
1115 	EF10_DMA_STAT(port_tx_128_to_255, TX_128_TO_255_PKTS),
1116 	EF10_DMA_STAT(port_tx_256_to_511, TX_256_TO_511_PKTS),
1117 	EF10_DMA_STAT(port_tx_512_to_1023, TX_512_TO_1023_PKTS),
1118 	EF10_DMA_STAT(port_tx_1024_to_15xx, TX_1024_TO_15XX_PKTS),
1119 	EF10_DMA_STAT(port_tx_15xx_to_jumbo, TX_15XX_TO_JUMBO_PKTS),
1120 	EF10_DMA_STAT(port_rx_bytes, RX_BYTES),
1121 	EF10_DMA_INVIS_STAT(port_rx_bytes_minus_good_bytes, RX_BAD_BYTES),
1122 	EF10_OTHER_STAT(port_rx_good_bytes),
1123 	EF10_OTHER_STAT(port_rx_bad_bytes),
1124 	EF10_DMA_STAT(port_rx_packets, RX_PKTS),
1125 	EF10_DMA_STAT(port_rx_good, RX_GOOD_PKTS),
1126 	EF10_DMA_STAT(port_rx_bad, RX_BAD_FCS_PKTS),
1127 	EF10_DMA_STAT(port_rx_pause, RX_PAUSE_PKTS),
1128 	EF10_DMA_STAT(port_rx_control, RX_CONTROL_PKTS),
1129 	EF10_DMA_STAT(port_rx_unicast, RX_UNICAST_PKTS),
1130 	EF10_DMA_STAT(port_rx_multicast, RX_MULTICAST_PKTS),
1131 	EF10_DMA_STAT(port_rx_broadcast, RX_BROADCAST_PKTS),
1132 	EF10_DMA_STAT(port_rx_lt64, RX_UNDERSIZE_PKTS),
1133 	EF10_DMA_STAT(port_rx_64, RX_64_PKTS),
1134 	EF10_DMA_STAT(port_rx_65_to_127, RX_65_TO_127_PKTS),
1135 	EF10_DMA_STAT(port_rx_128_to_255, RX_128_TO_255_PKTS),
1136 	EF10_DMA_STAT(port_rx_256_to_511, RX_256_TO_511_PKTS),
1137 	EF10_DMA_STAT(port_rx_512_to_1023, RX_512_TO_1023_PKTS),
1138 	EF10_DMA_STAT(port_rx_1024_to_15xx, RX_1024_TO_15XX_PKTS),
1139 	EF10_DMA_STAT(port_rx_15xx_to_jumbo, RX_15XX_TO_JUMBO_PKTS),
1140 	EF10_DMA_STAT(port_rx_gtjumbo, RX_GTJUMBO_PKTS),
1141 	EF10_DMA_STAT(port_rx_bad_gtjumbo, RX_JABBER_PKTS),
1142 	EF10_DMA_STAT(port_rx_overflow, RX_OVERFLOW_PKTS),
1143 	EF10_DMA_STAT(port_rx_align_error, RX_ALIGN_ERROR_PKTS),
1144 	EF10_DMA_STAT(port_rx_length_error, RX_LENGTH_ERROR_PKTS),
1145 	EF10_DMA_STAT(port_rx_nodesc_drops, RX_NODESC_DROPS),
1146 	GENERIC_SW_STAT(rx_nodesc_trunc),
1147 	GENERIC_SW_STAT(rx_noskb_drops),
1148 	EF10_DMA_STAT(port_rx_pm_trunc_bb_overflow, PM_TRUNC_BB_OVERFLOW),
1149 	EF10_DMA_STAT(port_rx_pm_discard_bb_overflow, PM_DISCARD_BB_OVERFLOW),
1150 	EF10_DMA_STAT(port_rx_pm_trunc_vfifo_full, PM_TRUNC_VFIFO_FULL),
1151 	EF10_DMA_STAT(port_rx_pm_discard_vfifo_full, PM_DISCARD_VFIFO_FULL),
1152 	EF10_DMA_STAT(port_rx_pm_trunc_qbb, PM_TRUNC_QBB),
1153 	EF10_DMA_STAT(port_rx_pm_discard_qbb, PM_DISCARD_QBB),
1154 	EF10_DMA_STAT(port_rx_pm_discard_mapping, PM_DISCARD_MAPPING),
1155 	EF10_DMA_STAT(port_rx_dp_q_disabled_packets, RXDP_Q_DISABLED_PKTS),
1156 	EF10_DMA_STAT(port_rx_dp_di_dropped_packets, RXDP_DI_DROPPED_PKTS),
1157 	EF10_DMA_STAT(port_rx_dp_streaming_packets, RXDP_STREAMING_PKTS),
1158 	EF10_DMA_STAT(port_rx_dp_hlb_fetch, RXDP_HLB_FETCH_CONDITIONS),
1159 	EF10_DMA_STAT(port_rx_dp_hlb_wait, RXDP_HLB_WAIT_CONDITIONS),
1160 	EF10_DMA_STAT(rx_unicast, VADAPTER_RX_UNICAST_PACKETS),
1161 	EF10_DMA_STAT(rx_unicast_bytes, VADAPTER_RX_UNICAST_BYTES),
1162 	EF10_DMA_STAT(rx_multicast, VADAPTER_RX_MULTICAST_PACKETS),
1163 	EF10_DMA_STAT(rx_multicast_bytes, VADAPTER_RX_MULTICAST_BYTES),
1164 	EF10_DMA_STAT(rx_broadcast, VADAPTER_RX_BROADCAST_PACKETS),
1165 	EF10_DMA_STAT(rx_broadcast_bytes, VADAPTER_RX_BROADCAST_BYTES),
1166 	EF10_DMA_STAT(rx_bad, VADAPTER_RX_BAD_PACKETS),
1167 	EF10_DMA_STAT(rx_bad_bytes, VADAPTER_RX_BAD_BYTES),
1168 	EF10_DMA_STAT(rx_overflow, VADAPTER_RX_OVERFLOW),
1169 	EF10_DMA_STAT(tx_unicast, VADAPTER_TX_UNICAST_PACKETS),
1170 	EF10_DMA_STAT(tx_unicast_bytes, VADAPTER_TX_UNICAST_BYTES),
1171 	EF10_DMA_STAT(tx_multicast, VADAPTER_TX_MULTICAST_PACKETS),
1172 	EF10_DMA_STAT(tx_multicast_bytes, VADAPTER_TX_MULTICAST_BYTES),
1173 	EF10_DMA_STAT(tx_broadcast, VADAPTER_TX_BROADCAST_PACKETS),
1174 	EF10_DMA_STAT(tx_broadcast_bytes, VADAPTER_TX_BROADCAST_BYTES),
1175 	EF10_DMA_STAT(tx_bad, VADAPTER_TX_BAD_PACKETS),
1176 	EF10_DMA_STAT(tx_bad_bytes, VADAPTER_TX_BAD_BYTES),
1177 	EF10_DMA_STAT(tx_overflow, VADAPTER_TX_OVERFLOW),
1178 };
1179 
1180 #define HUNT_COMMON_STAT_MASK ((1ULL << EF10_STAT_port_tx_bytes) |	\
1181 			       (1ULL << EF10_STAT_port_tx_packets) |	\
1182 			       (1ULL << EF10_STAT_port_tx_pause) |	\
1183 			       (1ULL << EF10_STAT_port_tx_unicast) |	\
1184 			       (1ULL << EF10_STAT_port_tx_multicast) |	\
1185 			       (1ULL << EF10_STAT_port_tx_broadcast) |	\
1186 			       (1ULL << EF10_STAT_port_rx_bytes) |	\
1187 			       (1ULL <<                                 \
1188 				EF10_STAT_port_rx_bytes_minus_good_bytes) | \
1189 			       (1ULL << EF10_STAT_port_rx_good_bytes) |	\
1190 			       (1ULL << EF10_STAT_port_rx_bad_bytes) |	\
1191 			       (1ULL << EF10_STAT_port_rx_packets) |	\
1192 			       (1ULL << EF10_STAT_port_rx_good) |	\
1193 			       (1ULL << EF10_STAT_port_rx_bad) |	\
1194 			       (1ULL << EF10_STAT_port_rx_pause) |	\
1195 			       (1ULL << EF10_STAT_port_rx_control) |	\
1196 			       (1ULL << EF10_STAT_port_rx_unicast) |	\
1197 			       (1ULL << EF10_STAT_port_rx_multicast) |	\
1198 			       (1ULL << EF10_STAT_port_rx_broadcast) |	\
1199 			       (1ULL << EF10_STAT_port_rx_lt64) |	\
1200 			       (1ULL << EF10_STAT_port_rx_64) |		\
1201 			       (1ULL << EF10_STAT_port_rx_65_to_127) |	\
1202 			       (1ULL << EF10_STAT_port_rx_128_to_255) |	\
1203 			       (1ULL << EF10_STAT_port_rx_256_to_511) |	\
1204 			       (1ULL << EF10_STAT_port_rx_512_to_1023) |\
1205 			       (1ULL << EF10_STAT_port_rx_1024_to_15xx) |\
1206 			       (1ULL << EF10_STAT_port_rx_15xx_to_jumbo) |\
1207 			       (1ULL << EF10_STAT_port_rx_gtjumbo) |	\
1208 			       (1ULL << EF10_STAT_port_rx_bad_gtjumbo) |\
1209 			       (1ULL << EF10_STAT_port_rx_overflow) |	\
1210 			       (1ULL << EF10_STAT_port_rx_nodesc_drops) |\
1211 			       (1ULL << GENERIC_STAT_rx_nodesc_trunc) |	\
1212 			       (1ULL << GENERIC_STAT_rx_noskb_drops))
1213 
1214 /* These statistics are only provided by the 10G MAC.  For a 10G/40G
1215  * switchable port we do not expose these because they might not
1216  * include all the packets they should.
1217  */
1218 #define HUNT_10G_ONLY_STAT_MASK ((1ULL << EF10_STAT_port_tx_control) |	\
1219 				 (1ULL << EF10_STAT_port_tx_lt64) |	\
1220 				 (1ULL << EF10_STAT_port_tx_64) |	\
1221 				 (1ULL << EF10_STAT_port_tx_65_to_127) |\
1222 				 (1ULL << EF10_STAT_port_tx_128_to_255) |\
1223 				 (1ULL << EF10_STAT_port_tx_256_to_511) |\
1224 				 (1ULL << EF10_STAT_port_tx_512_to_1023) |\
1225 				 (1ULL << EF10_STAT_port_tx_1024_to_15xx) |\
1226 				 (1ULL << EF10_STAT_port_tx_15xx_to_jumbo))
1227 
1228 /* These statistics are only provided by the 40G MAC.  For a 10G/40G
1229  * switchable port we do expose these because the errors will otherwise
1230  * be silent.
1231  */
1232 #define HUNT_40G_EXTRA_STAT_MASK ((1ULL << EF10_STAT_port_rx_align_error) |\
1233 				  (1ULL << EF10_STAT_port_rx_length_error))
1234 
1235 /* These statistics are only provided if the firmware supports the
1236  * capability PM_AND_RXDP_COUNTERS.
1237  */
1238 #define HUNT_PM_AND_RXDP_STAT_MASK (					\
1239 	(1ULL << EF10_STAT_port_rx_pm_trunc_bb_overflow) |		\
1240 	(1ULL << EF10_STAT_port_rx_pm_discard_bb_overflow) |		\
1241 	(1ULL << EF10_STAT_port_rx_pm_trunc_vfifo_full) |		\
1242 	(1ULL << EF10_STAT_port_rx_pm_discard_vfifo_full) |		\
1243 	(1ULL << EF10_STAT_port_rx_pm_trunc_qbb) |			\
1244 	(1ULL << EF10_STAT_port_rx_pm_discard_qbb) |			\
1245 	(1ULL << EF10_STAT_port_rx_pm_discard_mapping) |		\
1246 	(1ULL << EF10_STAT_port_rx_dp_q_disabled_packets) |		\
1247 	(1ULL << EF10_STAT_port_rx_dp_di_dropped_packets) |		\
1248 	(1ULL << EF10_STAT_port_rx_dp_streaming_packets) |		\
1249 	(1ULL << EF10_STAT_port_rx_dp_hlb_fetch) |			\
1250 	(1ULL << EF10_STAT_port_rx_dp_hlb_wait))
1251 
1252 static u64 efx_ef10_raw_stat_mask(struct efx_nic *efx)
1253 {
1254 	u64 raw_mask = HUNT_COMMON_STAT_MASK;
1255 	u32 port_caps = efx_mcdi_phy_get_caps(efx);
1256 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1257 
1258 	if (!(efx->mcdi->fn_flags &
1259 	      1 << MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_LINKCTRL))
1260 		return 0;
1261 
1262 	if (port_caps & (1 << MC_CMD_PHY_CAP_40000FDX_LBN))
1263 		raw_mask |= HUNT_40G_EXTRA_STAT_MASK;
1264 	else
1265 		raw_mask |= HUNT_10G_ONLY_STAT_MASK;
1266 
1267 	if (nic_data->datapath_caps &
1268 	    (1 << MC_CMD_GET_CAPABILITIES_OUT_PM_AND_RXDP_COUNTERS_LBN))
1269 		raw_mask |= HUNT_PM_AND_RXDP_STAT_MASK;
1270 
1271 	return raw_mask;
1272 }
1273 
1274 static void efx_ef10_get_stat_mask(struct efx_nic *efx, unsigned long *mask)
1275 {
1276 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1277 	u64 raw_mask[2];
1278 
1279 	raw_mask[0] = efx_ef10_raw_stat_mask(efx);
1280 
1281 	/* Only show vadaptor stats when EVB capability is present */
1282 	if (nic_data->datapath_caps &
1283 	    (1 << MC_CMD_GET_CAPABILITIES_OUT_EVB_LBN)) {
1284 		raw_mask[0] |= ~((1ULL << EF10_STAT_rx_unicast) - 1);
1285 		raw_mask[1] = (1ULL << (EF10_STAT_COUNT - 63)) - 1;
1286 	} else {
1287 		raw_mask[1] = 0;
1288 	}
1289 
1290 #if BITS_PER_LONG == 64
1291 	mask[0] = raw_mask[0];
1292 	mask[1] = raw_mask[1];
1293 #else
1294 	mask[0] = raw_mask[0] & 0xffffffff;
1295 	mask[1] = raw_mask[0] >> 32;
1296 	mask[2] = raw_mask[1] & 0xffffffff;
1297 	mask[3] = raw_mask[1] >> 32;
1298 #endif
1299 }
1300 
1301 static size_t efx_ef10_describe_stats(struct efx_nic *efx, u8 *names)
1302 {
1303 	DECLARE_BITMAP(mask, EF10_STAT_COUNT);
1304 
1305 	efx_ef10_get_stat_mask(efx, mask);
1306 	return efx_nic_describe_stats(efx_ef10_stat_desc, EF10_STAT_COUNT,
1307 				      mask, names);
1308 }
1309 
1310 static size_t efx_ef10_update_stats_common(struct efx_nic *efx, u64 *full_stats,
1311 					   struct rtnl_link_stats64 *core_stats)
1312 {
1313 	DECLARE_BITMAP(mask, EF10_STAT_COUNT);
1314 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1315 	u64 *stats = nic_data->stats;
1316 	size_t stats_count = 0, index;
1317 
1318 	efx_ef10_get_stat_mask(efx, mask);
1319 
1320 	if (full_stats) {
1321 		for_each_set_bit(index, mask, EF10_STAT_COUNT) {
1322 			if (efx_ef10_stat_desc[index].name) {
1323 				*full_stats++ = stats[index];
1324 				++stats_count;
1325 			}
1326 		}
1327 	}
1328 
1329 	if (!core_stats)
1330 		return stats_count;
1331 
1332 	if (nic_data->datapath_caps &
1333 			1 << MC_CMD_GET_CAPABILITIES_OUT_EVB_LBN) {
1334 		/* Use vadaptor stats. */
1335 		core_stats->rx_packets = stats[EF10_STAT_rx_unicast] +
1336 					 stats[EF10_STAT_rx_multicast] +
1337 					 stats[EF10_STAT_rx_broadcast];
1338 		core_stats->tx_packets = stats[EF10_STAT_tx_unicast] +
1339 					 stats[EF10_STAT_tx_multicast] +
1340 					 stats[EF10_STAT_tx_broadcast];
1341 		core_stats->rx_bytes = stats[EF10_STAT_rx_unicast_bytes] +
1342 				       stats[EF10_STAT_rx_multicast_bytes] +
1343 				       stats[EF10_STAT_rx_broadcast_bytes];
1344 		core_stats->tx_bytes = stats[EF10_STAT_tx_unicast_bytes] +
1345 				       stats[EF10_STAT_tx_multicast_bytes] +
1346 				       stats[EF10_STAT_tx_broadcast_bytes];
1347 		core_stats->rx_dropped = stats[GENERIC_STAT_rx_nodesc_trunc] +
1348 					 stats[GENERIC_STAT_rx_noskb_drops];
1349 		core_stats->multicast = stats[EF10_STAT_rx_multicast];
1350 		core_stats->rx_crc_errors = stats[EF10_STAT_rx_bad];
1351 		core_stats->rx_fifo_errors = stats[EF10_STAT_rx_overflow];
1352 		core_stats->rx_errors = core_stats->rx_crc_errors;
1353 		core_stats->tx_errors = stats[EF10_STAT_tx_bad];
1354 	} else {
1355 		/* Use port stats. */
1356 		core_stats->rx_packets = stats[EF10_STAT_port_rx_packets];
1357 		core_stats->tx_packets = stats[EF10_STAT_port_tx_packets];
1358 		core_stats->rx_bytes = stats[EF10_STAT_port_rx_bytes];
1359 		core_stats->tx_bytes = stats[EF10_STAT_port_tx_bytes];
1360 		core_stats->rx_dropped = stats[EF10_STAT_port_rx_nodesc_drops] +
1361 					 stats[GENERIC_STAT_rx_nodesc_trunc] +
1362 					 stats[GENERIC_STAT_rx_noskb_drops];
1363 		core_stats->multicast = stats[EF10_STAT_port_rx_multicast];
1364 		core_stats->rx_length_errors =
1365 				stats[EF10_STAT_port_rx_gtjumbo] +
1366 				stats[EF10_STAT_port_rx_length_error];
1367 		core_stats->rx_crc_errors = stats[EF10_STAT_port_rx_bad];
1368 		core_stats->rx_frame_errors =
1369 				stats[EF10_STAT_port_rx_align_error];
1370 		core_stats->rx_fifo_errors = stats[EF10_STAT_port_rx_overflow];
1371 		core_stats->rx_errors = (core_stats->rx_length_errors +
1372 					 core_stats->rx_crc_errors +
1373 					 core_stats->rx_frame_errors);
1374 	}
1375 
1376 	return stats_count;
1377 }
1378 
1379 static int efx_ef10_try_update_nic_stats_pf(struct efx_nic *efx)
1380 {
1381 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1382 	DECLARE_BITMAP(mask, EF10_STAT_COUNT);
1383 	__le64 generation_start, generation_end;
1384 	u64 *stats = nic_data->stats;
1385 	__le64 *dma_stats;
1386 
1387 	efx_ef10_get_stat_mask(efx, mask);
1388 
1389 	dma_stats = efx->stats_buffer.addr;
1390 	nic_data = efx->nic_data;
1391 
1392 	generation_end = dma_stats[MC_CMD_MAC_GENERATION_END];
1393 	if (generation_end == EFX_MC_STATS_GENERATION_INVALID)
1394 		return 0;
1395 	rmb();
1396 	efx_nic_update_stats(efx_ef10_stat_desc, EF10_STAT_COUNT, mask,
1397 			     stats, efx->stats_buffer.addr, false);
1398 	rmb();
1399 	generation_start = dma_stats[MC_CMD_MAC_GENERATION_START];
1400 	if (generation_end != generation_start)
1401 		return -EAGAIN;
1402 
1403 	/* Update derived statistics */
1404 	efx_nic_fix_nodesc_drop_stat(efx,
1405 				     &stats[EF10_STAT_port_rx_nodesc_drops]);
1406 	stats[EF10_STAT_port_rx_good_bytes] =
1407 		stats[EF10_STAT_port_rx_bytes] -
1408 		stats[EF10_STAT_port_rx_bytes_minus_good_bytes];
1409 	efx_update_diff_stat(&stats[EF10_STAT_port_rx_bad_bytes],
1410 			     stats[EF10_STAT_port_rx_bytes_minus_good_bytes]);
1411 	efx_update_sw_stats(efx, stats);
1412 	return 0;
1413 }
1414 
1415 
1416 static size_t efx_ef10_update_stats_pf(struct efx_nic *efx, u64 *full_stats,
1417 				       struct rtnl_link_stats64 *core_stats)
1418 {
1419 	int retry;
1420 
1421 	/* If we're unlucky enough to read statistics during the DMA, wait
1422 	 * up to 10ms for it to finish (typically takes <500us)
1423 	 */
1424 	for (retry = 0; retry < 100; ++retry) {
1425 		if (efx_ef10_try_update_nic_stats_pf(efx) == 0)
1426 			break;
1427 		udelay(100);
1428 	}
1429 
1430 	return efx_ef10_update_stats_common(efx, full_stats, core_stats);
1431 }
1432 
1433 static int efx_ef10_try_update_nic_stats_vf(struct efx_nic *efx)
1434 {
1435 	MCDI_DECLARE_BUF(inbuf, MC_CMD_MAC_STATS_IN_LEN);
1436 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1437 	DECLARE_BITMAP(mask, EF10_STAT_COUNT);
1438 	__le64 generation_start, generation_end;
1439 	u64 *stats = nic_data->stats;
1440 	u32 dma_len = MC_CMD_MAC_NSTATS * sizeof(u64);
1441 	struct efx_buffer stats_buf;
1442 	__le64 *dma_stats;
1443 	int rc;
1444 
1445 	spin_unlock_bh(&efx->stats_lock);
1446 
1447 	if (in_interrupt()) {
1448 		/* If in atomic context, cannot update stats.  Just update the
1449 		 * software stats and return so the caller can continue.
1450 		 */
1451 		spin_lock_bh(&efx->stats_lock);
1452 		efx_update_sw_stats(efx, stats);
1453 		return 0;
1454 	}
1455 
1456 	efx_ef10_get_stat_mask(efx, mask);
1457 
1458 	rc = efx_nic_alloc_buffer(efx, &stats_buf, dma_len, GFP_ATOMIC);
1459 	if (rc) {
1460 		spin_lock_bh(&efx->stats_lock);
1461 		return rc;
1462 	}
1463 
1464 	dma_stats = stats_buf.addr;
1465 	dma_stats[MC_CMD_MAC_GENERATION_END] = EFX_MC_STATS_GENERATION_INVALID;
1466 
1467 	MCDI_SET_QWORD(inbuf, MAC_STATS_IN_DMA_ADDR, stats_buf.dma_addr);
1468 	MCDI_POPULATE_DWORD_1(inbuf, MAC_STATS_IN_CMD,
1469 			      MAC_STATS_IN_DMA, 1);
1470 	MCDI_SET_DWORD(inbuf, MAC_STATS_IN_DMA_LEN, dma_len);
1471 	MCDI_SET_DWORD(inbuf, MAC_STATS_IN_PORT_ID, EVB_PORT_ID_ASSIGNED);
1472 
1473 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_MAC_STATS, inbuf, sizeof(inbuf),
1474 				NULL, 0, NULL);
1475 	spin_lock_bh(&efx->stats_lock);
1476 	if (rc) {
1477 		/* Expect ENOENT if DMA queues have not been set up */
1478 		if (rc != -ENOENT || atomic_read(&efx->active_queues))
1479 			efx_mcdi_display_error(efx, MC_CMD_MAC_STATS,
1480 					       sizeof(inbuf), NULL, 0, rc);
1481 		goto out;
1482 	}
1483 
1484 	generation_end = dma_stats[MC_CMD_MAC_GENERATION_END];
1485 	if (generation_end == EFX_MC_STATS_GENERATION_INVALID) {
1486 		WARN_ON_ONCE(1);
1487 		goto out;
1488 	}
1489 	rmb();
1490 	efx_nic_update_stats(efx_ef10_stat_desc, EF10_STAT_COUNT, mask,
1491 			     stats, stats_buf.addr, false);
1492 	rmb();
1493 	generation_start = dma_stats[MC_CMD_MAC_GENERATION_START];
1494 	if (generation_end != generation_start) {
1495 		rc = -EAGAIN;
1496 		goto out;
1497 	}
1498 
1499 	efx_update_sw_stats(efx, stats);
1500 out:
1501 	efx_nic_free_buffer(efx, &stats_buf);
1502 	return rc;
1503 }
1504 
1505 static size_t efx_ef10_update_stats_vf(struct efx_nic *efx, u64 *full_stats,
1506 				       struct rtnl_link_stats64 *core_stats)
1507 {
1508 	if (efx_ef10_try_update_nic_stats_vf(efx))
1509 		return 0;
1510 
1511 	return efx_ef10_update_stats_common(efx, full_stats, core_stats);
1512 }
1513 
1514 static void efx_ef10_push_irq_moderation(struct efx_channel *channel)
1515 {
1516 	struct efx_nic *efx = channel->efx;
1517 	unsigned int mode, value;
1518 	efx_dword_t timer_cmd;
1519 
1520 	if (channel->irq_moderation) {
1521 		mode = 3;
1522 		value = channel->irq_moderation - 1;
1523 	} else {
1524 		mode = 0;
1525 		value = 0;
1526 	}
1527 
1528 	if (EFX_EF10_WORKAROUND_35388(efx)) {
1529 		EFX_POPULATE_DWORD_3(timer_cmd, ERF_DD_EVQ_IND_TIMER_FLAGS,
1530 				     EFE_DD_EVQ_IND_TIMER_FLAGS,
1531 				     ERF_DD_EVQ_IND_TIMER_MODE, mode,
1532 				     ERF_DD_EVQ_IND_TIMER_VAL, value);
1533 		efx_writed_page(efx, &timer_cmd, ER_DD_EVQ_INDIRECT,
1534 				channel->channel);
1535 	} else {
1536 		EFX_POPULATE_DWORD_2(timer_cmd, ERF_DZ_TC_TIMER_MODE, mode,
1537 				     ERF_DZ_TC_TIMER_VAL, value);
1538 		efx_writed_page(efx, &timer_cmd, ER_DZ_EVQ_TMR,
1539 				channel->channel);
1540 	}
1541 }
1542 
1543 static void efx_ef10_get_wol_vf(struct efx_nic *efx,
1544 				struct ethtool_wolinfo *wol) {}
1545 
1546 static int efx_ef10_set_wol_vf(struct efx_nic *efx, u32 type)
1547 {
1548 	return -EOPNOTSUPP;
1549 }
1550 
1551 static void efx_ef10_get_wol(struct efx_nic *efx, struct ethtool_wolinfo *wol)
1552 {
1553 	wol->supported = 0;
1554 	wol->wolopts = 0;
1555 	memset(&wol->sopass, 0, sizeof(wol->sopass));
1556 }
1557 
1558 static int efx_ef10_set_wol(struct efx_nic *efx, u32 type)
1559 {
1560 	if (type != 0)
1561 		return -EINVAL;
1562 	return 0;
1563 }
1564 
1565 static void efx_ef10_mcdi_request(struct efx_nic *efx,
1566 				  const efx_dword_t *hdr, size_t hdr_len,
1567 				  const efx_dword_t *sdu, size_t sdu_len)
1568 {
1569 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1570 	u8 *pdu = nic_data->mcdi_buf.addr;
1571 
1572 	memcpy(pdu, hdr, hdr_len);
1573 	memcpy(pdu + hdr_len, sdu, sdu_len);
1574 	wmb();
1575 
1576 	/* The hardware provides 'low' and 'high' (doorbell) registers
1577 	 * for passing the 64-bit address of an MCDI request to
1578 	 * firmware.  However the dwords are swapped by firmware.  The
1579 	 * least significant bits of the doorbell are then 0 for all
1580 	 * MCDI requests due to alignment.
1581 	 */
1582 	_efx_writed(efx, cpu_to_le32((u64)nic_data->mcdi_buf.dma_addr >> 32),
1583 		    ER_DZ_MC_DB_LWRD);
1584 	_efx_writed(efx, cpu_to_le32((u32)nic_data->mcdi_buf.dma_addr),
1585 		    ER_DZ_MC_DB_HWRD);
1586 }
1587 
1588 static bool efx_ef10_mcdi_poll_response(struct efx_nic *efx)
1589 {
1590 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1591 	const efx_dword_t hdr = *(const efx_dword_t *)nic_data->mcdi_buf.addr;
1592 
1593 	rmb();
1594 	return EFX_DWORD_FIELD(hdr, MCDI_HEADER_RESPONSE);
1595 }
1596 
1597 static void
1598 efx_ef10_mcdi_read_response(struct efx_nic *efx, efx_dword_t *outbuf,
1599 			    size_t offset, size_t outlen)
1600 {
1601 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1602 	const u8 *pdu = nic_data->mcdi_buf.addr;
1603 
1604 	memcpy(outbuf, pdu + offset, outlen);
1605 }
1606 
1607 static 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 	wmb();
1801 	efx_ef10_push_tx_desc(tx_queue, txd);
1802 
1803 	return;
1804 
1805 fail:
1806 	netdev_WARN(efx->net_dev, "failed to initialise TXQ %d\n",
1807 		    tx_queue->queue);
1808 }
1809 
1810 static void efx_ef10_tx_fini(struct efx_tx_queue *tx_queue)
1811 {
1812 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FINI_TXQ_IN_LEN);
1813 	MCDI_DECLARE_BUF_ERR(outbuf);
1814 	struct efx_nic *efx = tx_queue->efx;
1815 	size_t outlen;
1816 	int rc;
1817 
1818 	MCDI_SET_DWORD(inbuf, FINI_TXQ_IN_INSTANCE,
1819 		       tx_queue->queue);
1820 
1821 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FINI_TXQ, inbuf, sizeof(inbuf),
1822 			  outbuf, sizeof(outbuf), &outlen);
1823 
1824 	if (rc && rc != -EALREADY)
1825 		goto fail;
1826 
1827 	return;
1828 
1829 fail:
1830 	efx_mcdi_display_error(efx, MC_CMD_FINI_TXQ, MC_CMD_FINI_TXQ_IN_LEN,
1831 			       outbuf, outlen, rc);
1832 }
1833 
1834 static void efx_ef10_tx_remove(struct efx_tx_queue *tx_queue)
1835 {
1836 	efx_nic_free_buffer(tx_queue->efx, &tx_queue->txd.buf);
1837 }
1838 
1839 /* This writes to the TX_DESC_WPTR; write pointer for TX descriptor ring */
1840 static inline void efx_ef10_notify_tx_desc(struct efx_tx_queue *tx_queue)
1841 {
1842 	unsigned int write_ptr;
1843 	efx_dword_t reg;
1844 
1845 	write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
1846 	EFX_POPULATE_DWORD_1(reg, ERF_DZ_TX_DESC_WPTR_DWORD, write_ptr);
1847 	efx_writed_page(tx_queue->efx, &reg,
1848 			ER_DZ_TX_DESC_UPD_DWORD, tx_queue->queue);
1849 }
1850 
1851 static void efx_ef10_tx_write(struct efx_tx_queue *tx_queue)
1852 {
1853 	unsigned int old_write_count = tx_queue->write_count;
1854 	struct efx_tx_buffer *buffer;
1855 	unsigned int write_ptr;
1856 	efx_qword_t *txd;
1857 
1858 	tx_queue->xmit_more_available = false;
1859 	if (unlikely(tx_queue->write_count == tx_queue->insert_count))
1860 		return;
1861 
1862 	do {
1863 		write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
1864 		buffer = &tx_queue->buffer[write_ptr];
1865 		txd = efx_tx_desc(tx_queue, write_ptr);
1866 		++tx_queue->write_count;
1867 
1868 		/* Create TX descriptor ring entry */
1869 		if (buffer->flags & EFX_TX_BUF_OPTION) {
1870 			*txd = buffer->option;
1871 		} else {
1872 			BUILD_BUG_ON(EFX_TX_BUF_CONT != 1);
1873 			EFX_POPULATE_QWORD_3(
1874 				*txd,
1875 				ESF_DZ_TX_KER_CONT,
1876 				buffer->flags & EFX_TX_BUF_CONT,
1877 				ESF_DZ_TX_KER_BYTE_CNT, buffer->len,
1878 				ESF_DZ_TX_KER_BUF_ADDR, buffer->dma_addr);
1879 		}
1880 	} while (tx_queue->write_count != tx_queue->insert_count);
1881 
1882 	wmb(); /* Ensure descriptors are written before they are fetched */
1883 
1884 	if (efx_nic_may_push_tx_desc(tx_queue, old_write_count)) {
1885 		txd = efx_tx_desc(tx_queue,
1886 				  old_write_count & tx_queue->ptr_mask);
1887 		efx_ef10_push_tx_desc(tx_queue, txd);
1888 		++tx_queue->pushes;
1889 	} else {
1890 		efx_ef10_notify_tx_desc(tx_queue);
1891 	}
1892 }
1893 
1894 static int efx_ef10_alloc_rss_context(struct efx_nic *efx, u32 *context,
1895 				      bool exclusive, unsigned *context_size)
1896 {
1897 	MCDI_DECLARE_BUF(inbuf, MC_CMD_RSS_CONTEXT_ALLOC_IN_LEN);
1898 	MCDI_DECLARE_BUF(outbuf, MC_CMD_RSS_CONTEXT_ALLOC_OUT_LEN);
1899 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1900 	size_t outlen;
1901 	int rc;
1902 	u32 alloc_type = exclusive ?
1903 				MC_CMD_RSS_CONTEXT_ALLOC_IN_TYPE_EXCLUSIVE :
1904 				MC_CMD_RSS_CONTEXT_ALLOC_IN_TYPE_SHARED;
1905 	unsigned rss_spread = exclusive ?
1906 				efx->rss_spread :
1907 				min(rounddown_pow_of_two(efx->rss_spread),
1908 				    EFX_EF10_MAX_SHARED_RSS_CONTEXT_SIZE);
1909 
1910 	if (!exclusive && rss_spread == 1) {
1911 		*context = EFX_EF10_RSS_CONTEXT_INVALID;
1912 		if (context_size)
1913 			*context_size = 1;
1914 		return 0;
1915 	}
1916 
1917 	MCDI_SET_DWORD(inbuf, RSS_CONTEXT_ALLOC_IN_UPSTREAM_PORT_ID,
1918 		       nic_data->vport_id);
1919 	MCDI_SET_DWORD(inbuf, RSS_CONTEXT_ALLOC_IN_TYPE, alloc_type);
1920 	MCDI_SET_DWORD(inbuf, RSS_CONTEXT_ALLOC_IN_NUM_QUEUES, rss_spread);
1921 
1922 	rc = efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_ALLOC, inbuf, sizeof(inbuf),
1923 		outbuf, sizeof(outbuf), &outlen);
1924 	if (rc != 0)
1925 		return rc;
1926 
1927 	if (outlen < MC_CMD_RSS_CONTEXT_ALLOC_OUT_LEN)
1928 		return -EIO;
1929 
1930 	*context = MCDI_DWORD(outbuf, RSS_CONTEXT_ALLOC_OUT_RSS_CONTEXT_ID);
1931 
1932 	if (context_size)
1933 		*context_size = rss_spread;
1934 
1935 	return 0;
1936 }
1937 
1938 static void efx_ef10_free_rss_context(struct efx_nic *efx, u32 context)
1939 {
1940 	MCDI_DECLARE_BUF(inbuf, MC_CMD_RSS_CONTEXT_FREE_IN_LEN);
1941 	int rc;
1942 
1943 	MCDI_SET_DWORD(inbuf, RSS_CONTEXT_FREE_IN_RSS_CONTEXT_ID,
1944 		       context);
1945 
1946 	rc = efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_FREE, inbuf, sizeof(inbuf),
1947 			    NULL, 0, NULL);
1948 	WARN_ON(rc != 0);
1949 }
1950 
1951 static int efx_ef10_populate_rss_table(struct efx_nic *efx, u32 context,
1952 				       const u32 *rx_indir_table)
1953 {
1954 	MCDI_DECLARE_BUF(tablebuf, MC_CMD_RSS_CONTEXT_SET_TABLE_IN_LEN);
1955 	MCDI_DECLARE_BUF(keybuf, MC_CMD_RSS_CONTEXT_SET_KEY_IN_LEN);
1956 	int i, rc;
1957 
1958 	MCDI_SET_DWORD(tablebuf, RSS_CONTEXT_SET_TABLE_IN_RSS_CONTEXT_ID,
1959 		       context);
1960 	BUILD_BUG_ON(ARRAY_SIZE(efx->rx_indir_table) !=
1961 		     MC_CMD_RSS_CONTEXT_SET_TABLE_IN_INDIRECTION_TABLE_LEN);
1962 
1963 	for (i = 0; i < ARRAY_SIZE(efx->rx_indir_table); ++i)
1964 		MCDI_PTR(tablebuf,
1965 			 RSS_CONTEXT_SET_TABLE_IN_INDIRECTION_TABLE)[i] =
1966 				(u8) rx_indir_table[i];
1967 
1968 	rc = efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_SET_TABLE, tablebuf,
1969 			  sizeof(tablebuf), NULL, 0, NULL);
1970 	if (rc != 0)
1971 		return rc;
1972 
1973 	MCDI_SET_DWORD(keybuf, RSS_CONTEXT_SET_KEY_IN_RSS_CONTEXT_ID,
1974 		       context);
1975 	BUILD_BUG_ON(ARRAY_SIZE(efx->rx_hash_key) !=
1976 		     MC_CMD_RSS_CONTEXT_SET_KEY_IN_TOEPLITZ_KEY_LEN);
1977 	for (i = 0; i < ARRAY_SIZE(efx->rx_hash_key); ++i)
1978 		MCDI_PTR(keybuf, RSS_CONTEXT_SET_KEY_IN_TOEPLITZ_KEY)[i] =
1979 			efx->rx_hash_key[i];
1980 
1981 	return efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_SET_KEY, keybuf,
1982 			    sizeof(keybuf), NULL, 0, NULL);
1983 }
1984 
1985 static void efx_ef10_rx_free_indir_table(struct efx_nic *efx)
1986 {
1987 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1988 
1989 	if (nic_data->rx_rss_context != EFX_EF10_RSS_CONTEXT_INVALID)
1990 		efx_ef10_free_rss_context(efx, nic_data->rx_rss_context);
1991 	nic_data->rx_rss_context = EFX_EF10_RSS_CONTEXT_INVALID;
1992 }
1993 
1994 static int efx_ef10_rx_push_shared_rss_config(struct efx_nic *efx,
1995 					      unsigned *context_size)
1996 {
1997 	u32 new_rx_rss_context;
1998 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1999 	int rc = efx_ef10_alloc_rss_context(efx, &new_rx_rss_context,
2000 					    false, context_size);
2001 
2002 	if (rc != 0)
2003 		return rc;
2004 
2005 	nic_data->rx_rss_context = new_rx_rss_context;
2006 	nic_data->rx_rss_context_exclusive = false;
2007 	efx_set_default_rx_indir_table(efx);
2008 	return 0;
2009 }
2010 
2011 static int efx_ef10_rx_push_exclusive_rss_config(struct efx_nic *efx,
2012 						 const u32 *rx_indir_table)
2013 {
2014 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2015 	int rc;
2016 	u32 new_rx_rss_context;
2017 
2018 	if (nic_data->rx_rss_context == EFX_EF10_RSS_CONTEXT_INVALID ||
2019 	    !nic_data->rx_rss_context_exclusive) {
2020 		rc = efx_ef10_alloc_rss_context(efx, &new_rx_rss_context,
2021 						true, NULL);
2022 		if (rc == -EOPNOTSUPP)
2023 			return rc;
2024 		else if (rc != 0)
2025 			goto fail1;
2026 	} else {
2027 		new_rx_rss_context = nic_data->rx_rss_context;
2028 	}
2029 
2030 	rc = efx_ef10_populate_rss_table(efx, new_rx_rss_context,
2031 					 rx_indir_table);
2032 	if (rc != 0)
2033 		goto fail2;
2034 
2035 	if (nic_data->rx_rss_context != new_rx_rss_context)
2036 		efx_ef10_rx_free_indir_table(efx);
2037 	nic_data->rx_rss_context = new_rx_rss_context;
2038 	nic_data->rx_rss_context_exclusive = true;
2039 	if (rx_indir_table != efx->rx_indir_table)
2040 		memcpy(efx->rx_indir_table, rx_indir_table,
2041 		       sizeof(efx->rx_indir_table));
2042 	return 0;
2043 
2044 fail2:
2045 	if (new_rx_rss_context != nic_data->rx_rss_context)
2046 		efx_ef10_free_rss_context(efx, new_rx_rss_context);
2047 fail1:
2048 	netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
2049 	return rc;
2050 }
2051 
2052 static int efx_ef10_pf_rx_push_rss_config(struct efx_nic *efx, bool user,
2053 					  const u32 *rx_indir_table)
2054 {
2055 	int rc;
2056 
2057 	if (efx->rss_spread == 1)
2058 		return 0;
2059 
2060 	rc = efx_ef10_rx_push_exclusive_rss_config(efx, rx_indir_table);
2061 
2062 	if (rc == -ENOBUFS && !user) {
2063 		unsigned context_size;
2064 		bool mismatch = false;
2065 		size_t i;
2066 
2067 		for (i = 0; i < ARRAY_SIZE(efx->rx_indir_table) && !mismatch;
2068 		     i++)
2069 			mismatch = rx_indir_table[i] !=
2070 				ethtool_rxfh_indir_default(i, efx->rss_spread);
2071 
2072 		rc = efx_ef10_rx_push_shared_rss_config(efx, &context_size);
2073 		if (rc == 0) {
2074 			if (context_size != efx->rss_spread)
2075 				netif_warn(efx, probe, efx->net_dev,
2076 					   "Could not allocate an exclusive RSS"
2077 					   " context; allocated a shared one of"
2078 					   " different size."
2079 					   " Wanted %u, got %u.\n",
2080 					   efx->rss_spread, context_size);
2081 			else if (mismatch)
2082 				netif_warn(efx, probe, efx->net_dev,
2083 					   "Could not allocate an exclusive RSS"
2084 					   " context; allocated a shared one but"
2085 					   " could not apply custom"
2086 					   " indirection.\n");
2087 			else
2088 				netif_info(efx, probe, efx->net_dev,
2089 					   "Could not allocate an exclusive RSS"
2090 					   " context; allocated a shared one.\n");
2091 		}
2092 	}
2093 	return rc;
2094 }
2095 
2096 static int efx_ef10_vf_rx_push_rss_config(struct efx_nic *efx, bool user,
2097 					  const u32 *rx_indir_table
2098 					  __attribute__ ((unused)))
2099 {
2100 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2101 
2102 	if (user)
2103 		return -EOPNOTSUPP;
2104 	if (nic_data->rx_rss_context != EFX_EF10_RSS_CONTEXT_INVALID)
2105 		return 0;
2106 	return efx_ef10_rx_push_shared_rss_config(efx, NULL);
2107 }
2108 
2109 static int efx_ef10_rx_probe(struct efx_rx_queue *rx_queue)
2110 {
2111 	return efx_nic_alloc_buffer(rx_queue->efx, &rx_queue->rxd.buf,
2112 				    (rx_queue->ptr_mask + 1) *
2113 				    sizeof(efx_qword_t),
2114 				    GFP_KERNEL);
2115 }
2116 
2117 static void efx_ef10_rx_init(struct efx_rx_queue *rx_queue)
2118 {
2119 	MCDI_DECLARE_BUF(inbuf,
2120 			 MC_CMD_INIT_RXQ_IN_LEN(EFX_MAX_DMAQ_SIZE * 8 /
2121 						EFX_BUF_SIZE));
2122 	struct efx_channel *channel = efx_rx_queue_channel(rx_queue);
2123 	size_t entries = rx_queue->rxd.buf.len / EFX_BUF_SIZE;
2124 	struct efx_nic *efx = rx_queue->efx;
2125 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2126 	size_t inlen;
2127 	dma_addr_t dma_addr;
2128 	int rc;
2129 	int i;
2130 	BUILD_BUG_ON(MC_CMD_INIT_RXQ_OUT_LEN != 0);
2131 
2132 	rx_queue->scatter_n = 0;
2133 	rx_queue->scatter_len = 0;
2134 
2135 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_SIZE, rx_queue->ptr_mask + 1);
2136 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_TARGET_EVQ, channel->channel);
2137 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_LABEL, efx_rx_queue_index(rx_queue));
2138 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_INSTANCE,
2139 		       efx_rx_queue_index(rx_queue));
2140 	MCDI_POPULATE_DWORD_2(inbuf, INIT_RXQ_IN_FLAGS,
2141 			      INIT_RXQ_IN_FLAG_PREFIX, 1,
2142 			      INIT_RXQ_IN_FLAG_TIMESTAMP, 1);
2143 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_OWNER_ID, 0);
2144 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_PORT_ID, nic_data->vport_id);
2145 
2146 	dma_addr = rx_queue->rxd.buf.dma_addr;
2147 
2148 	netif_dbg(efx, hw, efx->net_dev, "pushing RXQ %d. %zu entries (%llx)\n",
2149 		  efx_rx_queue_index(rx_queue), entries, (u64)dma_addr);
2150 
2151 	for (i = 0; i < entries; ++i) {
2152 		MCDI_SET_ARRAY_QWORD(inbuf, INIT_RXQ_IN_DMA_ADDR, i, dma_addr);
2153 		dma_addr += EFX_BUF_SIZE;
2154 	}
2155 
2156 	inlen = MC_CMD_INIT_RXQ_IN_LEN(entries);
2157 
2158 	rc = efx_mcdi_rpc(efx, MC_CMD_INIT_RXQ, inbuf, inlen,
2159 			  NULL, 0, NULL);
2160 	if (rc)
2161 		netdev_WARN(efx->net_dev, "failed to initialise RXQ %d\n",
2162 			    efx_rx_queue_index(rx_queue));
2163 }
2164 
2165 static void efx_ef10_rx_fini(struct efx_rx_queue *rx_queue)
2166 {
2167 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FINI_RXQ_IN_LEN);
2168 	MCDI_DECLARE_BUF_ERR(outbuf);
2169 	struct efx_nic *efx = rx_queue->efx;
2170 	size_t outlen;
2171 	int rc;
2172 
2173 	MCDI_SET_DWORD(inbuf, FINI_RXQ_IN_INSTANCE,
2174 		       efx_rx_queue_index(rx_queue));
2175 
2176 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FINI_RXQ, inbuf, sizeof(inbuf),
2177 			  outbuf, sizeof(outbuf), &outlen);
2178 
2179 	if (rc && rc != -EALREADY)
2180 		goto fail;
2181 
2182 	return;
2183 
2184 fail:
2185 	efx_mcdi_display_error(efx, MC_CMD_FINI_RXQ, MC_CMD_FINI_RXQ_IN_LEN,
2186 			       outbuf, outlen, rc);
2187 }
2188 
2189 static void efx_ef10_rx_remove(struct efx_rx_queue *rx_queue)
2190 {
2191 	efx_nic_free_buffer(rx_queue->efx, &rx_queue->rxd.buf);
2192 }
2193 
2194 /* This creates an entry in the RX descriptor queue */
2195 static inline void
2196 efx_ef10_build_rx_desc(struct efx_rx_queue *rx_queue, unsigned int index)
2197 {
2198 	struct efx_rx_buffer *rx_buf;
2199 	efx_qword_t *rxd;
2200 
2201 	rxd = efx_rx_desc(rx_queue, index);
2202 	rx_buf = efx_rx_buffer(rx_queue, index);
2203 	EFX_POPULATE_QWORD_2(*rxd,
2204 			     ESF_DZ_RX_KER_BYTE_CNT, rx_buf->len,
2205 			     ESF_DZ_RX_KER_BUF_ADDR, rx_buf->dma_addr);
2206 }
2207 
2208 static void efx_ef10_rx_write(struct efx_rx_queue *rx_queue)
2209 {
2210 	struct efx_nic *efx = rx_queue->efx;
2211 	unsigned int write_count;
2212 	efx_dword_t reg;
2213 
2214 	/* Firmware requires that RX_DESC_WPTR be a multiple of 8 */
2215 	write_count = rx_queue->added_count & ~7;
2216 	if (rx_queue->notified_count == write_count)
2217 		return;
2218 
2219 	do
2220 		efx_ef10_build_rx_desc(
2221 			rx_queue,
2222 			rx_queue->notified_count & rx_queue->ptr_mask);
2223 	while (++rx_queue->notified_count != write_count);
2224 
2225 	wmb();
2226 	EFX_POPULATE_DWORD_1(reg, ERF_DZ_RX_DESC_WPTR,
2227 			     write_count & rx_queue->ptr_mask);
2228 	efx_writed_page(efx, &reg, ER_DZ_RX_DESC_UPD,
2229 			efx_rx_queue_index(rx_queue));
2230 }
2231 
2232 static efx_mcdi_async_completer efx_ef10_rx_defer_refill_complete;
2233 
2234 static void efx_ef10_rx_defer_refill(struct efx_rx_queue *rx_queue)
2235 {
2236 	struct efx_channel *channel = efx_rx_queue_channel(rx_queue);
2237 	MCDI_DECLARE_BUF(inbuf, MC_CMD_DRIVER_EVENT_IN_LEN);
2238 	efx_qword_t event;
2239 
2240 	EFX_POPULATE_QWORD_2(event,
2241 			     ESF_DZ_EV_CODE, EFX_EF10_DRVGEN_EV,
2242 			     ESF_DZ_EV_DATA, EFX_EF10_REFILL);
2243 
2244 	MCDI_SET_DWORD(inbuf, DRIVER_EVENT_IN_EVQ, channel->channel);
2245 
2246 	/* MCDI_SET_QWORD is not appropriate here since EFX_POPULATE_* has
2247 	 * already swapped the data to little-endian order.
2248 	 */
2249 	memcpy(MCDI_PTR(inbuf, DRIVER_EVENT_IN_DATA), &event.u64[0],
2250 	       sizeof(efx_qword_t));
2251 
2252 	efx_mcdi_rpc_async(channel->efx, MC_CMD_DRIVER_EVENT,
2253 			   inbuf, sizeof(inbuf), 0,
2254 			   efx_ef10_rx_defer_refill_complete, 0);
2255 }
2256 
2257 static void
2258 efx_ef10_rx_defer_refill_complete(struct efx_nic *efx, unsigned long cookie,
2259 				  int rc, efx_dword_t *outbuf,
2260 				  size_t outlen_actual)
2261 {
2262 	/* nothing to do */
2263 }
2264 
2265 static int efx_ef10_ev_probe(struct efx_channel *channel)
2266 {
2267 	return efx_nic_alloc_buffer(channel->efx, &channel->eventq.buf,
2268 				    (channel->eventq_mask + 1) *
2269 				    sizeof(efx_qword_t),
2270 				    GFP_KERNEL);
2271 }
2272 
2273 static void efx_ef10_ev_fini(struct efx_channel *channel)
2274 {
2275 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FINI_EVQ_IN_LEN);
2276 	MCDI_DECLARE_BUF_ERR(outbuf);
2277 	struct efx_nic *efx = channel->efx;
2278 	size_t outlen;
2279 	int rc;
2280 
2281 	MCDI_SET_DWORD(inbuf, FINI_EVQ_IN_INSTANCE, channel->channel);
2282 
2283 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FINI_EVQ, inbuf, sizeof(inbuf),
2284 			  outbuf, sizeof(outbuf), &outlen);
2285 
2286 	if (rc && rc != -EALREADY)
2287 		goto fail;
2288 
2289 	return;
2290 
2291 fail:
2292 	efx_mcdi_display_error(efx, MC_CMD_FINI_EVQ, MC_CMD_FINI_EVQ_IN_LEN,
2293 			       outbuf, outlen, rc);
2294 }
2295 
2296 static int efx_ef10_ev_init(struct efx_channel *channel)
2297 {
2298 	MCDI_DECLARE_BUF(inbuf,
2299 			 MC_CMD_INIT_EVQ_IN_LEN(EFX_MAX_EVQ_SIZE * 8 /
2300 						EFX_BUF_SIZE));
2301 	MCDI_DECLARE_BUF(outbuf, MC_CMD_INIT_EVQ_OUT_LEN);
2302 	size_t entries = channel->eventq.buf.len / EFX_BUF_SIZE;
2303 	struct efx_nic *efx = channel->efx;
2304 	struct efx_ef10_nic_data *nic_data;
2305 	bool supports_rx_merge;
2306 	size_t inlen, outlen;
2307 	unsigned int enabled, implemented;
2308 	dma_addr_t dma_addr;
2309 	int rc;
2310 	int i;
2311 
2312 	nic_data = efx->nic_data;
2313 	supports_rx_merge =
2314 		!!(nic_data->datapath_caps &
2315 		   1 << MC_CMD_GET_CAPABILITIES_OUT_RX_BATCHING_LBN);
2316 
2317 	/* Fill event queue with all ones (i.e. empty events) */
2318 	memset(channel->eventq.buf.addr, 0xff, channel->eventq.buf.len);
2319 
2320 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_SIZE, channel->eventq_mask + 1);
2321 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_INSTANCE, channel->channel);
2322 	/* INIT_EVQ expects index in vector table, not absolute */
2323 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_IRQ_NUM, channel->channel);
2324 	MCDI_POPULATE_DWORD_4(inbuf, INIT_EVQ_IN_FLAGS,
2325 			      INIT_EVQ_IN_FLAG_INTERRUPTING, 1,
2326 			      INIT_EVQ_IN_FLAG_RX_MERGE, 1,
2327 			      INIT_EVQ_IN_FLAG_TX_MERGE, 1,
2328 			      INIT_EVQ_IN_FLAG_CUT_THRU, !supports_rx_merge);
2329 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_TMR_MODE,
2330 		       MC_CMD_INIT_EVQ_IN_TMR_MODE_DIS);
2331 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_TMR_LOAD, 0);
2332 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_TMR_RELOAD, 0);
2333 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_COUNT_MODE,
2334 		       MC_CMD_INIT_EVQ_IN_COUNT_MODE_DIS);
2335 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_COUNT_THRSHLD, 0);
2336 
2337 	dma_addr = channel->eventq.buf.dma_addr;
2338 	for (i = 0; i < entries; ++i) {
2339 		MCDI_SET_ARRAY_QWORD(inbuf, INIT_EVQ_IN_DMA_ADDR, i, dma_addr);
2340 		dma_addr += EFX_BUF_SIZE;
2341 	}
2342 
2343 	inlen = MC_CMD_INIT_EVQ_IN_LEN(entries);
2344 
2345 	rc = efx_mcdi_rpc(efx, MC_CMD_INIT_EVQ, inbuf, inlen,
2346 			  outbuf, sizeof(outbuf), &outlen);
2347 	/* IRQ return is ignored */
2348 	if (channel->channel || rc)
2349 		return rc;
2350 
2351 	/* Successfully created event queue on channel 0 */
2352 	rc = efx_mcdi_get_workarounds(efx, &implemented, &enabled);
2353 	if (rc == -ENOSYS) {
2354 		/* GET_WORKAROUNDS was implemented before the bug26807
2355 		 * workaround, thus the latter must be unavailable in this fw
2356 		 */
2357 		nic_data->workaround_26807 = false;
2358 		rc = 0;
2359 	} else if (rc) {
2360 		goto fail;
2361 	} else {
2362 		nic_data->workaround_26807 =
2363 			!!(enabled & MC_CMD_GET_WORKAROUNDS_OUT_BUG26807);
2364 
2365 		if (implemented & MC_CMD_GET_WORKAROUNDS_OUT_BUG26807 &&
2366 		    !nic_data->workaround_26807) {
2367 			unsigned int flags;
2368 
2369 			rc = efx_mcdi_set_workaround(efx,
2370 						     MC_CMD_WORKAROUND_BUG26807,
2371 						     true, &flags);
2372 
2373 			if (!rc) {
2374 				if (flags &
2375 				    1 << MC_CMD_WORKAROUND_EXT_OUT_FLR_DONE_LBN) {
2376 					netif_info(efx, drv, efx->net_dev,
2377 						   "other functions on NIC have been reset\n");
2378 					/* MC's boot count has incremented */
2379 					++nic_data->warm_boot_count;
2380 				}
2381 				nic_data->workaround_26807 = true;
2382 			} else if (rc == -EPERM) {
2383 				rc = 0;
2384 			}
2385 		}
2386 	}
2387 
2388 	if (!rc)
2389 		return 0;
2390 
2391 fail:
2392 	efx_ef10_ev_fini(channel);
2393 	return rc;
2394 }
2395 
2396 static void efx_ef10_ev_remove(struct efx_channel *channel)
2397 {
2398 	efx_nic_free_buffer(channel->efx, &channel->eventq.buf);
2399 }
2400 
2401 static void efx_ef10_handle_rx_wrong_queue(struct efx_rx_queue *rx_queue,
2402 					   unsigned int rx_queue_label)
2403 {
2404 	struct efx_nic *efx = rx_queue->efx;
2405 
2406 	netif_info(efx, hw, efx->net_dev,
2407 		   "rx event arrived on queue %d labeled as queue %u\n",
2408 		   efx_rx_queue_index(rx_queue), rx_queue_label);
2409 
2410 	efx_schedule_reset(efx, RESET_TYPE_DISABLE);
2411 }
2412 
2413 static void
2414 efx_ef10_handle_rx_bad_lbits(struct efx_rx_queue *rx_queue,
2415 			     unsigned int actual, unsigned int expected)
2416 {
2417 	unsigned int dropped = (actual - expected) & rx_queue->ptr_mask;
2418 	struct efx_nic *efx = rx_queue->efx;
2419 
2420 	netif_info(efx, hw, efx->net_dev,
2421 		   "dropped %d events (index=%d expected=%d)\n",
2422 		   dropped, actual, expected);
2423 
2424 	efx_schedule_reset(efx, RESET_TYPE_DISABLE);
2425 }
2426 
2427 /* partially received RX was aborted. clean up. */
2428 static void efx_ef10_handle_rx_abort(struct efx_rx_queue *rx_queue)
2429 {
2430 	unsigned int rx_desc_ptr;
2431 
2432 	netif_dbg(rx_queue->efx, hw, rx_queue->efx->net_dev,
2433 		  "scattered RX aborted (dropping %u buffers)\n",
2434 		  rx_queue->scatter_n);
2435 
2436 	rx_desc_ptr = rx_queue->removed_count & rx_queue->ptr_mask;
2437 
2438 	efx_rx_packet(rx_queue, rx_desc_ptr, rx_queue->scatter_n,
2439 		      0, EFX_RX_PKT_DISCARD);
2440 
2441 	rx_queue->removed_count += rx_queue->scatter_n;
2442 	rx_queue->scatter_n = 0;
2443 	rx_queue->scatter_len = 0;
2444 	++efx_rx_queue_channel(rx_queue)->n_rx_nodesc_trunc;
2445 }
2446 
2447 static int efx_ef10_handle_rx_event(struct efx_channel *channel,
2448 				    const efx_qword_t *event)
2449 {
2450 	unsigned int rx_bytes, next_ptr_lbits, rx_queue_label, rx_l4_class;
2451 	unsigned int n_descs, n_packets, i;
2452 	struct efx_nic *efx = channel->efx;
2453 	struct efx_rx_queue *rx_queue;
2454 	bool rx_cont;
2455 	u16 flags = 0;
2456 
2457 	if (unlikely(ACCESS_ONCE(efx->reset_pending)))
2458 		return 0;
2459 
2460 	/* Basic packet information */
2461 	rx_bytes = EFX_QWORD_FIELD(*event, ESF_DZ_RX_BYTES);
2462 	next_ptr_lbits = EFX_QWORD_FIELD(*event, ESF_DZ_RX_DSC_PTR_LBITS);
2463 	rx_queue_label = EFX_QWORD_FIELD(*event, ESF_DZ_RX_QLABEL);
2464 	rx_l4_class = EFX_QWORD_FIELD(*event, ESF_DZ_RX_L4_CLASS);
2465 	rx_cont = EFX_QWORD_FIELD(*event, ESF_DZ_RX_CONT);
2466 
2467 	if (EFX_QWORD_FIELD(*event, ESF_DZ_RX_DROP_EVENT))
2468 		netdev_WARN(efx->net_dev, "saw RX_DROP_EVENT: event="
2469 			    EFX_QWORD_FMT "\n",
2470 			    EFX_QWORD_VAL(*event));
2471 
2472 	rx_queue = efx_channel_get_rx_queue(channel);
2473 
2474 	if (unlikely(rx_queue_label != efx_rx_queue_index(rx_queue)))
2475 		efx_ef10_handle_rx_wrong_queue(rx_queue, rx_queue_label);
2476 
2477 	n_descs = ((next_ptr_lbits - rx_queue->removed_count) &
2478 		   ((1 << ESF_DZ_RX_DSC_PTR_LBITS_WIDTH) - 1));
2479 
2480 	if (n_descs != rx_queue->scatter_n + 1) {
2481 		struct efx_ef10_nic_data *nic_data = efx->nic_data;
2482 
2483 		/* detect rx abort */
2484 		if (unlikely(n_descs == rx_queue->scatter_n)) {
2485 			if (rx_queue->scatter_n == 0 || rx_bytes != 0)
2486 				netdev_WARN(efx->net_dev,
2487 					    "invalid RX abort: scatter_n=%u event="
2488 					    EFX_QWORD_FMT "\n",
2489 					    rx_queue->scatter_n,
2490 					    EFX_QWORD_VAL(*event));
2491 			efx_ef10_handle_rx_abort(rx_queue);
2492 			return 0;
2493 		}
2494 
2495 		/* Check that RX completion merging is valid, i.e.
2496 		 * the current firmware supports it and this is a
2497 		 * non-scattered packet.
2498 		 */
2499 		if (!(nic_data->datapath_caps &
2500 		      (1 << MC_CMD_GET_CAPABILITIES_OUT_RX_BATCHING_LBN)) ||
2501 		    rx_queue->scatter_n != 0 || rx_cont) {
2502 			efx_ef10_handle_rx_bad_lbits(
2503 				rx_queue, next_ptr_lbits,
2504 				(rx_queue->removed_count +
2505 				 rx_queue->scatter_n + 1) &
2506 				((1 << ESF_DZ_RX_DSC_PTR_LBITS_WIDTH) - 1));
2507 			return 0;
2508 		}
2509 
2510 		/* Merged completion for multiple non-scattered packets */
2511 		rx_queue->scatter_n = 1;
2512 		rx_queue->scatter_len = 0;
2513 		n_packets = n_descs;
2514 		++channel->n_rx_merge_events;
2515 		channel->n_rx_merge_packets += n_packets;
2516 		flags |= EFX_RX_PKT_PREFIX_LEN;
2517 	} else {
2518 		++rx_queue->scatter_n;
2519 		rx_queue->scatter_len += rx_bytes;
2520 		if (rx_cont)
2521 			return 0;
2522 		n_packets = 1;
2523 	}
2524 
2525 	if (unlikely(EFX_QWORD_FIELD(*event, ESF_DZ_RX_ECRC_ERR)))
2526 		flags |= EFX_RX_PKT_DISCARD;
2527 
2528 	if (unlikely(EFX_QWORD_FIELD(*event, ESF_DZ_RX_IPCKSUM_ERR))) {
2529 		channel->n_rx_ip_hdr_chksum_err += n_packets;
2530 	} else if (unlikely(EFX_QWORD_FIELD(*event,
2531 					    ESF_DZ_RX_TCPUDP_CKSUM_ERR))) {
2532 		channel->n_rx_tcp_udp_chksum_err += n_packets;
2533 	} else if (rx_l4_class == ESE_DZ_L4_CLASS_TCP ||
2534 		   rx_l4_class == ESE_DZ_L4_CLASS_UDP) {
2535 		flags |= EFX_RX_PKT_CSUMMED;
2536 	}
2537 
2538 	if (rx_l4_class == ESE_DZ_L4_CLASS_TCP)
2539 		flags |= EFX_RX_PKT_TCP;
2540 
2541 	channel->irq_mod_score += 2 * n_packets;
2542 
2543 	/* Handle received packet(s) */
2544 	for (i = 0; i < n_packets; i++) {
2545 		efx_rx_packet(rx_queue,
2546 			      rx_queue->removed_count & rx_queue->ptr_mask,
2547 			      rx_queue->scatter_n, rx_queue->scatter_len,
2548 			      flags);
2549 		rx_queue->removed_count += rx_queue->scatter_n;
2550 	}
2551 
2552 	rx_queue->scatter_n = 0;
2553 	rx_queue->scatter_len = 0;
2554 
2555 	return n_packets;
2556 }
2557 
2558 static int
2559 efx_ef10_handle_tx_event(struct efx_channel *channel, efx_qword_t *event)
2560 {
2561 	struct efx_nic *efx = channel->efx;
2562 	struct efx_tx_queue *tx_queue;
2563 	unsigned int tx_ev_desc_ptr;
2564 	unsigned int tx_ev_q_label;
2565 	int tx_descs = 0;
2566 
2567 	if (unlikely(ACCESS_ONCE(efx->reset_pending)))
2568 		return 0;
2569 
2570 	if (unlikely(EFX_QWORD_FIELD(*event, ESF_DZ_TX_DROP_EVENT)))
2571 		return 0;
2572 
2573 	/* Transmit completion */
2574 	tx_ev_desc_ptr = EFX_QWORD_FIELD(*event, ESF_DZ_TX_DESCR_INDX);
2575 	tx_ev_q_label = EFX_QWORD_FIELD(*event, ESF_DZ_TX_QLABEL);
2576 	tx_queue = efx_channel_get_tx_queue(channel,
2577 					    tx_ev_q_label % EFX_TXQ_TYPES);
2578 	tx_descs = ((tx_ev_desc_ptr + 1 - tx_queue->read_count) &
2579 		    tx_queue->ptr_mask);
2580 	efx_xmit_done(tx_queue, tx_ev_desc_ptr & tx_queue->ptr_mask);
2581 
2582 	return tx_descs;
2583 }
2584 
2585 static void
2586 efx_ef10_handle_driver_event(struct efx_channel *channel, efx_qword_t *event)
2587 {
2588 	struct efx_nic *efx = channel->efx;
2589 	int subcode;
2590 
2591 	subcode = EFX_QWORD_FIELD(*event, ESF_DZ_DRV_SUB_CODE);
2592 
2593 	switch (subcode) {
2594 	case ESE_DZ_DRV_TIMER_EV:
2595 	case ESE_DZ_DRV_WAKE_UP_EV:
2596 		break;
2597 	case ESE_DZ_DRV_START_UP_EV:
2598 		/* event queue init complete. ok. */
2599 		break;
2600 	default:
2601 		netif_err(efx, hw, efx->net_dev,
2602 			  "channel %d unknown driver event type %d"
2603 			  " (data " EFX_QWORD_FMT ")\n",
2604 			  channel->channel, subcode,
2605 			  EFX_QWORD_VAL(*event));
2606 
2607 	}
2608 }
2609 
2610 static void efx_ef10_handle_driver_generated_event(struct efx_channel *channel,
2611 						   efx_qword_t *event)
2612 {
2613 	struct efx_nic *efx = channel->efx;
2614 	u32 subcode;
2615 
2616 	subcode = EFX_QWORD_FIELD(*event, EFX_DWORD_0);
2617 
2618 	switch (subcode) {
2619 	case EFX_EF10_TEST:
2620 		channel->event_test_cpu = raw_smp_processor_id();
2621 		break;
2622 	case EFX_EF10_REFILL:
2623 		/* The queue must be empty, so we won't receive any rx
2624 		 * events, so efx_process_channel() won't refill the
2625 		 * queue. Refill it here
2626 		 */
2627 		efx_fast_push_rx_descriptors(&channel->rx_queue, true);
2628 		break;
2629 	default:
2630 		netif_err(efx, hw, efx->net_dev,
2631 			  "channel %d unknown driver event type %u"
2632 			  " (data " EFX_QWORD_FMT ")\n",
2633 			  channel->channel, (unsigned) subcode,
2634 			  EFX_QWORD_VAL(*event));
2635 	}
2636 }
2637 
2638 static int efx_ef10_ev_process(struct efx_channel *channel, int quota)
2639 {
2640 	struct efx_nic *efx = channel->efx;
2641 	efx_qword_t event, *p_event;
2642 	unsigned int read_ptr;
2643 	int ev_code;
2644 	int tx_descs = 0;
2645 	int spent = 0;
2646 
2647 	if (quota <= 0)
2648 		return spent;
2649 
2650 	read_ptr = channel->eventq_read_ptr;
2651 
2652 	for (;;) {
2653 		p_event = efx_event(channel, read_ptr);
2654 		event = *p_event;
2655 
2656 		if (!efx_event_present(&event))
2657 			break;
2658 
2659 		EFX_SET_QWORD(*p_event);
2660 
2661 		++read_ptr;
2662 
2663 		ev_code = EFX_QWORD_FIELD(event, ESF_DZ_EV_CODE);
2664 
2665 		netif_vdbg(efx, drv, efx->net_dev,
2666 			   "processing event on %d " EFX_QWORD_FMT "\n",
2667 			   channel->channel, EFX_QWORD_VAL(event));
2668 
2669 		switch (ev_code) {
2670 		case ESE_DZ_EV_CODE_MCDI_EV:
2671 			efx_mcdi_process_event(channel, &event);
2672 			break;
2673 		case ESE_DZ_EV_CODE_RX_EV:
2674 			spent += efx_ef10_handle_rx_event(channel, &event);
2675 			if (spent >= quota) {
2676 				/* XXX can we split a merged event to
2677 				 * avoid going over-quota?
2678 				 */
2679 				spent = quota;
2680 				goto out;
2681 			}
2682 			break;
2683 		case ESE_DZ_EV_CODE_TX_EV:
2684 			tx_descs += efx_ef10_handle_tx_event(channel, &event);
2685 			if (tx_descs > efx->txq_entries) {
2686 				spent = quota;
2687 				goto out;
2688 			} else if (++spent == quota) {
2689 				goto out;
2690 			}
2691 			break;
2692 		case ESE_DZ_EV_CODE_DRIVER_EV:
2693 			efx_ef10_handle_driver_event(channel, &event);
2694 			if (++spent == quota)
2695 				goto out;
2696 			break;
2697 		case EFX_EF10_DRVGEN_EV:
2698 			efx_ef10_handle_driver_generated_event(channel, &event);
2699 			break;
2700 		default:
2701 			netif_err(efx, hw, efx->net_dev,
2702 				  "channel %d unknown event type %d"
2703 				  " (data " EFX_QWORD_FMT ")\n",
2704 				  channel->channel, ev_code,
2705 				  EFX_QWORD_VAL(event));
2706 		}
2707 	}
2708 
2709 out:
2710 	channel->eventq_read_ptr = read_ptr;
2711 	return spent;
2712 }
2713 
2714 static void efx_ef10_ev_read_ack(struct efx_channel *channel)
2715 {
2716 	struct efx_nic *efx = channel->efx;
2717 	efx_dword_t rptr;
2718 
2719 	if (EFX_EF10_WORKAROUND_35388(efx)) {
2720 		BUILD_BUG_ON(EFX_MIN_EVQ_SIZE <
2721 			     (1 << ERF_DD_EVQ_IND_RPTR_WIDTH));
2722 		BUILD_BUG_ON(EFX_MAX_EVQ_SIZE >
2723 			     (1 << 2 * ERF_DD_EVQ_IND_RPTR_WIDTH));
2724 
2725 		EFX_POPULATE_DWORD_2(rptr, ERF_DD_EVQ_IND_RPTR_FLAGS,
2726 				     EFE_DD_EVQ_IND_RPTR_FLAGS_HIGH,
2727 				     ERF_DD_EVQ_IND_RPTR,
2728 				     (channel->eventq_read_ptr &
2729 				      channel->eventq_mask) >>
2730 				     ERF_DD_EVQ_IND_RPTR_WIDTH);
2731 		efx_writed_page(efx, &rptr, ER_DD_EVQ_INDIRECT,
2732 				channel->channel);
2733 		EFX_POPULATE_DWORD_2(rptr, ERF_DD_EVQ_IND_RPTR_FLAGS,
2734 				     EFE_DD_EVQ_IND_RPTR_FLAGS_LOW,
2735 				     ERF_DD_EVQ_IND_RPTR,
2736 				     channel->eventq_read_ptr &
2737 				     ((1 << ERF_DD_EVQ_IND_RPTR_WIDTH) - 1));
2738 		efx_writed_page(efx, &rptr, ER_DD_EVQ_INDIRECT,
2739 				channel->channel);
2740 	} else {
2741 		EFX_POPULATE_DWORD_1(rptr, ERF_DZ_EVQ_RPTR,
2742 				     channel->eventq_read_ptr &
2743 				     channel->eventq_mask);
2744 		efx_writed_page(efx, &rptr, ER_DZ_EVQ_RPTR, channel->channel);
2745 	}
2746 }
2747 
2748 static void efx_ef10_ev_test_generate(struct efx_channel *channel)
2749 {
2750 	MCDI_DECLARE_BUF(inbuf, MC_CMD_DRIVER_EVENT_IN_LEN);
2751 	struct efx_nic *efx = channel->efx;
2752 	efx_qword_t event;
2753 	int rc;
2754 
2755 	EFX_POPULATE_QWORD_2(event,
2756 			     ESF_DZ_EV_CODE, EFX_EF10_DRVGEN_EV,
2757 			     ESF_DZ_EV_DATA, EFX_EF10_TEST);
2758 
2759 	MCDI_SET_DWORD(inbuf, DRIVER_EVENT_IN_EVQ, channel->channel);
2760 
2761 	/* MCDI_SET_QWORD is not appropriate here since EFX_POPULATE_* has
2762 	 * already swapped the data to little-endian order.
2763 	 */
2764 	memcpy(MCDI_PTR(inbuf, DRIVER_EVENT_IN_DATA), &event.u64[0],
2765 	       sizeof(efx_qword_t));
2766 
2767 	rc = efx_mcdi_rpc(efx, MC_CMD_DRIVER_EVENT, inbuf, sizeof(inbuf),
2768 			  NULL, 0, NULL);
2769 	if (rc != 0)
2770 		goto fail;
2771 
2772 	return;
2773 
2774 fail:
2775 	WARN_ON(true);
2776 	netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
2777 }
2778 
2779 void efx_ef10_handle_drain_event(struct efx_nic *efx)
2780 {
2781 	if (atomic_dec_and_test(&efx->active_queues))
2782 		wake_up(&efx->flush_wq);
2783 
2784 	WARN_ON(atomic_read(&efx->active_queues) < 0);
2785 }
2786 
2787 static int efx_ef10_fini_dmaq(struct efx_nic *efx)
2788 {
2789 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2790 	struct efx_channel *channel;
2791 	struct efx_tx_queue *tx_queue;
2792 	struct efx_rx_queue *rx_queue;
2793 	int pending;
2794 
2795 	/* If the MC has just rebooted, the TX/RX queues will have already been
2796 	 * torn down, but efx->active_queues needs to be set to zero.
2797 	 */
2798 	if (nic_data->must_realloc_vis) {
2799 		atomic_set(&efx->active_queues, 0);
2800 		return 0;
2801 	}
2802 
2803 	/* Do not attempt to write to the NIC during EEH recovery */
2804 	if (efx->state != STATE_RECOVERY) {
2805 		efx_for_each_channel(channel, efx) {
2806 			efx_for_each_channel_rx_queue(rx_queue, channel)
2807 				efx_ef10_rx_fini(rx_queue);
2808 			efx_for_each_channel_tx_queue(tx_queue, channel)
2809 				efx_ef10_tx_fini(tx_queue);
2810 		}
2811 
2812 		wait_event_timeout(efx->flush_wq,
2813 				   atomic_read(&efx->active_queues) == 0,
2814 				   msecs_to_jiffies(EFX_MAX_FLUSH_TIME));
2815 		pending = atomic_read(&efx->active_queues);
2816 		if (pending) {
2817 			netif_err(efx, hw, efx->net_dev, "failed to flush %d queues\n",
2818 				  pending);
2819 			return -ETIMEDOUT;
2820 		}
2821 	}
2822 
2823 	return 0;
2824 }
2825 
2826 static void efx_ef10_prepare_flr(struct efx_nic *efx)
2827 {
2828 	atomic_set(&efx->active_queues, 0);
2829 }
2830 
2831 static bool efx_ef10_filter_equal(const struct efx_filter_spec *left,
2832 				  const struct efx_filter_spec *right)
2833 {
2834 	if ((left->match_flags ^ right->match_flags) |
2835 	    ((left->flags ^ right->flags) &
2836 	     (EFX_FILTER_FLAG_RX | EFX_FILTER_FLAG_TX)))
2837 		return false;
2838 
2839 	return memcmp(&left->outer_vid, &right->outer_vid,
2840 		      sizeof(struct efx_filter_spec) -
2841 		      offsetof(struct efx_filter_spec, outer_vid)) == 0;
2842 }
2843 
2844 static unsigned int efx_ef10_filter_hash(const struct efx_filter_spec *spec)
2845 {
2846 	BUILD_BUG_ON(offsetof(struct efx_filter_spec, outer_vid) & 3);
2847 	return jhash2((const u32 *)&spec->outer_vid,
2848 		      (sizeof(struct efx_filter_spec) -
2849 		       offsetof(struct efx_filter_spec, outer_vid)) / 4,
2850 		      0);
2851 	/* XXX should we randomise the initval? */
2852 }
2853 
2854 /* Decide whether a filter should be exclusive or else should allow
2855  * delivery to additional recipients.  Currently we decide that
2856  * filters for specific local unicast MAC and IP addresses are
2857  * exclusive.
2858  */
2859 static bool efx_ef10_filter_is_exclusive(const struct efx_filter_spec *spec)
2860 {
2861 	if (spec->match_flags & EFX_FILTER_MATCH_LOC_MAC &&
2862 	    !is_multicast_ether_addr(spec->loc_mac))
2863 		return true;
2864 
2865 	if ((spec->match_flags &
2866 	     (EFX_FILTER_MATCH_ETHER_TYPE | EFX_FILTER_MATCH_LOC_HOST)) ==
2867 	    (EFX_FILTER_MATCH_ETHER_TYPE | EFX_FILTER_MATCH_LOC_HOST)) {
2868 		if (spec->ether_type == htons(ETH_P_IP) &&
2869 		    !ipv4_is_multicast(spec->loc_host[0]))
2870 			return true;
2871 		if (spec->ether_type == htons(ETH_P_IPV6) &&
2872 		    ((const u8 *)spec->loc_host)[0] != 0xff)
2873 			return true;
2874 	}
2875 
2876 	return false;
2877 }
2878 
2879 static struct efx_filter_spec *
2880 efx_ef10_filter_entry_spec(const struct efx_ef10_filter_table *table,
2881 			   unsigned int filter_idx)
2882 {
2883 	return (struct efx_filter_spec *)(table->entry[filter_idx].spec &
2884 					  ~EFX_EF10_FILTER_FLAGS);
2885 }
2886 
2887 static unsigned int
2888 efx_ef10_filter_entry_flags(const struct efx_ef10_filter_table *table,
2889 			   unsigned int filter_idx)
2890 {
2891 	return table->entry[filter_idx].spec & EFX_EF10_FILTER_FLAGS;
2892 }
2893 
2894 static void
2895 efx_ef10_filter_set_entry(struct efx_ef10_filter_table *table,
2896 			  unsigned int filter_idx,
2897 			  const struct efx_filter_spec *spec,
2898 			  unsigned int flags)
2899 {
2900 	table->entry[filter_idx].spec =	(unsigned long)spec | flags;
2901 }
2902 
2903 static void efx_ef10_filter_push_prep(struct efx_nic *efx,
2904 				      const struct efx_filter_spec *spec,
2905 				      efx_dword_t *inbuf, u64 handle,
2906 				      bool replacing)
2907 {
2908 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2909 
2910 	memset(inbuf, 0, MC_CMD_FILTER_OP_IN_LEN);
2911 
2912 	if (replacing) {
2913 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
2914 			       MC_CMD_FILTER_OP_IN_OP_REPLACE);
2915 		MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE, handle);
2916 	} else {
2917 		u32 match_fields = 0;
2918 
2919 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
2920 			       efx_ef10_filter_is_exclusive(spec) ?
2921 			       MC_CMD_FILTER_OP_IN_OP_INSERT :
2922 			       MC_CMD_FILTER_OP_IN_OP_SUBSCRIBE);
2923 
2924 		/* Convert match flags and values.  Unlike almost
2925 		 * everything else in MCDI, these fields are in
2926 		 * network byte order.
2927 		 */
2928 		if (spec->match_flags & EFX_FILTER_MATCH_LOC_MAC_IG)
2929 			match_fields |=
2930 				is_multicast_ether_addr(spec->loc_mac) ?
2931 				1 << MC_CMD_FILTER_OP_IN_MATCH_UNKNOWN_MCAST_DST_LBN :
2932 				1 << MC_CMD_FILTER_OP_IN_MATCH_UNKNOWN_UCAST_DST_LBN;
2933 #define COPY_FIELD(gen_flag, gen_field, mcdi_field)			     \
2934 		if (spec->match_flags & EFX_FILTER_MATCH_ ## gen_flag) {     \
2935 			match_fields |=					     \
2936 				1 << MC_CMD_FILTER_OP_IN_MATCH_ ##	     \
2937 				mcdi_field ## _LBN;			     \
2938 			BUILD_BUG_ON(					     \
2939 				MC_CMD_FILTER_OP_IN_ ## mcdi_field ## _LEN < \
2940 				sizeof(spec->gen_field));		     \
2941 			memcpy(MCDI_PTR(inbuf, FILTER_OP_IN_ ##	mcdi_field), \
2942 			       &spec->gen_field, sizeof(spec->gen_field));   \
2943 		}
2944 		COPY_FIELD(REM_HOST, rem_host, SRC_IP);
2945 		COPY_FIELD(LOC_HOST, loc_host, DST_IP);
2946 		COPY_FIELD(REM_MAC, rem_mac, SRC_MAC);
2947 		COPY_FIELD(REM_PORT, rem_port, SRC_PORT);
2948 		COPY_FIELD(LOC_MAC, loc_mac, DST_MAC);
2949 		COPY_FIELD(LOC_PORT, loc_port, DST_PORT);
2950 		COPY_FIELD(ETHER_TYPE, ether_type, ETHER_TYPE);
2951 		COPY_FIELD(INNER_VID, inner_vid, INNER_VLAN);
2952 		COPY_FIELD(OUTER_VID, outer_vid, OUTER_VLAN);
2953 		COPY_FIELD(IP_PROTO, ip_proto, IP_PROTO);
2954 #undef COPY_FIELD
2955 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_MATCH_FIELDS,
2956 			       match_fields);
2957 	}
2958 
2959 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_PORT_ID, nic_data->vport_id);
2960 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_DEST,
2961 		       spec->dmaq_id == EFX_FILTER_RX_DMAQ_ID_DROP ?
2962 		       MC_CMD_FILTER_OP_IN_RX_DEST_DROP :
2963 		       MC_CMD_FILTER_OP_IN_RX_DEST_HOST);
2964 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_TX_DOMAIN, 0);
2965 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_TX_DEST,
2966 		       MC_CMD_FILTER_OP_IN_TX_DEST_DEFAULT);
2967 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_QUEUE,
2968 		       spec->dmaq_id == EFX_FILTER_RX_DMAQ_ID_DROP ?
2969 		       0 : spec->dmaq_id);
2970 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_MODE,
2971 		       (spec->flags & EFX_FILTER_FLAG_RX_RSS) ?
2972 		       MC_CMD_FILTER_OP_IN_RX_MODE_RSS :
2973 		       MC_CMD_FILTER_OP_IN_RX_MODE_SIMPLE);
2974 	if (spec->flags & EFX_FILTER_FLAG_RX_RSS)
2975 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_CONTEXT,
2976 			       spec->rss_context !=
2977 			       EFX_FILTER_RSS_CONTEXT_DEFAULT ?
2978 			       spec->rss_context : nic_data->rx_rss_context);
2979 }
2980 
2981 static int efx_ef10_filter_push(struct efx_nic *efx,
2982 				const struct efx_filter_spec *spec,
2983 				u64 *handle, bool replacing)
2984 {
2985 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_IN_LEN);
2986 	MCDI_DECLARE_BUF(outbuf, MC_CMD_FILTER_OP_OUT_LEN);
2987 	int rc;
2988 
2989 	efx_ef10_filter_push_prep(efx, spec, inbuf, *handle, replacing);
2990 	rc = efx_mcdi_rpc(efx, MC_CMD_FILTER_OP, inbuf, sizeof(inbuf),
2991 			  outbuf, sizeof(outbuf), NULL);
2992 	if (rc == 0)
2993 		*handle = MCDI_QWORD(outbuf, FILTER_OP_OUT_HANDLE);
2994 	if (rc == -ENOSPC)
2995 		rc = -EBUSY; /* to match efx_farch_filter_insert() */
2996 	return rc;
2997 }
2998 
2999 static int efx_ef10_filter_rx_match_pri(struct efx_ef10_filter_table *table,
3000 					enum efx_filter_match_flags match_flags)
3001 {
3002 	unsigned int match_pri;
3003 
3004 	for (match_pri = 0;
3005 	     match_pri < table->rx_match_count;
3006 	     match_pri++)
3007 		if (table->rx_match_flags[match_pri] == match_flags)
3008 			return match_pri;
3009 
3010 	return -EPROTONOSUPPORT;
3011 }
3012 
3013 static s32 efx_ef10_filter_insert(struct efx_nic *efx,
3014 				  struct efx_filter_spec *spec,
3015 				  bool replace_equal)
3016 {
3017 	struct efx_ef10_filter_table *table = efx->filter_state;
3018 	DECLARE_BITMAP(mc_rem_map, EFX_EF10_FILTER_SEARCH_LIMIT);
3019 	struct efx_filter_spec *saved_spec;
3020 	unsigned int match_pri, hash;
3021 	unsigned int priv_flags;
3022 	bool replacing = false;
3023 	int ins_index = -1;
3024 	DEFINE_WAIT(wait);
3025 	bool is_mc_recip;
3026 	s32 rc;
3027 
3028 	/* For now, only support RX filters */
3029 	if ((spec->flags & (EFX_FILTER_FLAG_RX | EFX_FILTER_FLAG_TX)) !=
3030 	    EFX_FILTER_FLAG_RX)
3031 		return -EINVAL;
3032 
3033 	rc = efx_ef10_filter_rx_match_pri(table, spec->match_flags);
3034 	if (rc < 0)
3035 		return rc;
3036 	match_pri = rc;
3037 
3038 	hash = efx_ef10_filter_hash(spec);
3039 	is_mc_recip = efx_filter_is_mc_recipient(spec);
3040 	if (is_mc_recip)
3041 		bitmap_zero(mc_rem_map, EFX_EF10_FILTER_SEARCH_LIMIT);
3042 
3043 	/* Find any existing filters with the same match tuple or
3044 	 * else a free slot to insert at.  If any of them are busy,
3045 	 * we have to wait and retry.
3046 	 */
3047 	for (;;) {
3048 		unsigned int depth = 1;
3049 		unsigned int i;
3050 
3051 		spin_lock_bh(&efx->filter_lock);
3052 
3053 		for (;;) {
3054 			i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
3055 			saved_spec = efx_ef10_filter_entry_spec(table, i);
3056 
3057 			if (!saved_spec) {
3058 				if (ins_index < 0)
3059 					ins_index = i;
3060 			} else if (efx_ef10_filter_equal(spec, saved_spec)) {
3061 				if (table->entry[i].spec &
3062 				    EFX_EF10_FILTER_FLAG_BUSY)
3063 					break;
3064 				if (spec->priority < saved_spec->priority &&
3065 				    spec->priority != EFX_FILTER_PRI_AUTO) {
3066 					rc = -EPERM;
3067 					goto out_unlock;
3068 				}
3069 				if (!is_mc_recip) {
3070 					/* This is the only one */
3071 					if (spec->priority ==
3072 					    saved_spec->priority &&
3073 					    !replace_equal) {
3074 						rc = -EEXIST;
3075 						goto out_unlock;
3076 					}
3077 					ins_index = i;
3078 					goto found;
3079 				} else if (spec->priority >
3080 					   saved_spec->priority ||
3081 					   (spec->priority ==
3082 					    saved_spec->priority &&
3083 					    replace_equal)) {
3084 					if (ins_index < 0)
3085 						ins_index = i;
3086 					else
3087 						__set_bit(depth, mc_rem_map);
3088 				}
3089 			}
3090 
3091 			/* Once we reach the maximum search depth, use
3092 			 * the first suitable slot or return -EBUSY if
3093 			 * there was none
3094 			 */
3095 			if (depth == EFX_EF10_FILTER_SEARCH_LIMIT) {
3096 				if (ins_index < 0) {
3097 					rc = -EBUSY;
3098 					goto out_unlock;
3099 				}
3100 				goto found;
3101 			}
3102 
3103 			++depth;
3104 		}
3105 
3106 		prepare_to_wait(&table->waitq, &wait, TASK_UNINTERRUPTIBLE);
3107 		spin_unlock_bh(&efx->filter_lock);
3108 		schedule();
3109 	}
3110 
3111 found:
3112 	/* Create a software table entry if necessary, and mark it
3113 	 * busy.  We might yet fail to insert, but any attempt to
3114 	 * insert a conflicting filter while we're waiting for the
3115 	 * firmware must find the busy entry.
3116 	 */
3117 	saved_spec = efx_ef10_filter_entry_spec(table, ins_index);
3118 	if (saved_spec) {
3119 		if (spec->priority == EFX_FILTER_PRI_AUTO &&
3120 		    saved_spec->priority >= EFX_FILTER_PRI_AUTO) {
3121 			/* Just make sure it won't be removed */
3122 			if (saved_spec->priority > EFX_FILTER_PRI_AUTO)
3123 				saved_spec->flags |= EFX_FILTER_FLAG_RX_OVER_AUTO;
3124 			table->entry[ins_index].spec &=
3125 				~EFX_EF10_FILTER_FLAG_AUTO_OLD;
3126 			rc = ins_index;
3127 			goto out_unlock;
3128 		}
3129 		replacing = true;
3130 		priv_flags = efx_ef10_filter_entry_flags(table, ins_index);
3131 	} else {
3132 		saved_spec = kmalloc(sizeof(*spec), GFP_ATOMIC);
3133 		if (!saved_spec) {
3134 			rc = -ENOMEM;
3135 			goto out_unlock;
3136 		}
3137 		*saved_spec = *spec;
3138 		priv_flags = 0;
3139 	}
3140 	efx_ef10_filter_set_entry(table, ins_index, saved_spec,
3141 				  priv_flags | EFX_EF10_FILTER_FLAG_BUSY);
3142 
3143 	/* Mark lower-priority multicast recipients busy prior to removal */
3144 	if (is_mc_recip) {
3145 		unsigned int depth, i;
3146 
3147 		for (depth = 0; depth < EFX_EF10_FILTER_SEARCH_LIMIT; depth++) {
3148 			i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
3149 			if (test_bit(depth, mc_rem_map))
3150 				table->entry[i].spec |=
3151 					EFX_EF10_FILTER_FLAG_BUSY;
3152 		}
3153 	}
3154 
3155 	spin_unlock_bh(&efx->filter_lock);
3156 
3157 	rc = efx_ef10_filter_push(efx, spec, &table->entry[ins_index].handle,
3158 				  replacing);
3159 
3160 	/* Finalise the software table entry */
3161 	spin_lock_bh(&efx->filter_lock);
3162 	if (rc == 0) {
3163 		if (replacing) {
3164 			/* Update the fields that may differ */
3165 			if (saved_spec->priority == EFX_FILTER_PRI_AUTO)
3166 				saved_spec->flags |=
3167 					EFX_FILTER_FLAG_RX_OVER_AUTO;
3168 			saved_spec->priority = spec->priority;
3169 			saved_spec->flags &= EFX_FILTER_FLAG_RX_OVER_AUTO;
3170 			saved_spec->flags |= spec->flags;
3171 			saved_spec->rss_context = spec->rss_context;
3172 			saved_spec->dmaq_id = spec->dmaq_id;
3173 		}
3174 	} else if (!replacing) {
3175 		kfree(saved_spec);
3176 		saved_spec = NULL;
3177 	}
3178 	efx_ef10_filter_set_entry(table, ins_index, saved_spec, priv_flags);
3179 
3180 	/* Remove and finalise entries for lower-priority multicast
3181 	 * recipients
3182 	 */
3183 	if (is_mc_recip) {
3184 		MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_IN_LEN);
3185 		unsigned int depth, i;
3186 
3187 		memset(inbuf, 0, sizeof(inbuf));
3188 
3189 		for (depth = 0; depth < EFX_EF10_FILTER_SEARCH_LIMIT; depth++) {
3190 			if (!test_bit(depth, mc_rem_map))
3191 				continue;
3192 
3193 			i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
3194 			saved_spec = efx_ef10_filter_entry_spec(table, i);
3195 			priv_flags = efx_ef10_filter_entry_flags(table, i);
3196 
3197 			if (rc == 0) {
3198 				spin_unlock_bh(&efx->filter_lock);
3199 				MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
3200 					       MC_CMD_FILTER_OP_IN_OP_UNSUBSCRIBE);
3201 				MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
3202 					       table->entry[i].handle);
3203 				rc = efx_mcdi_rpc(efx, MC_CMD_FILTER_OP,
3204 						  inbuf, sizeof(inbuf),
3205 						  NULL, 0, NULL);
3206 				spin_lock_bh(&efx->filter_lock);
3207 			}
3208 
3209 			if (rc == 0) {
3210 				kfree(saved_spec);
3211 				saved_spec = NULL;
3212 				priv_flags = 0;
3213 			} else {
3214 				priv_flags &= ~EFX_EF10_FILTER_FLAG_BUSY;
3215 			}
3216 			efx_ef10_filter_set_entry(table, i, saved_spec,
3217 						  priv_flags);
3218 		}
3219 	}
3220 
3221 	/* If successful, return the inserted filter ID */
3222 	if (rc == 0)
3223 		rc = match_pri * HUNT_FILTER_TBL_ROWS + ins_index;
3224 
3225 	wake_up_all(&table->waitq);
3226 out_unlock:
3227 	spin_unlock_bh(&efx->filter_lock);
3228 	finish_wait(&table->waitq, &wait);
3229 	return rc;
3230 }
3231 
3232 static void efx_ef10_filter_update_rx_scatter(struct efx_nic *efx)
3233 {
3234 	/* no need to do anything here on EF10 */
3235 }
3236 
3237 /* Remove a filter.
3238  * If !by_index, remove by ID
3239  * If by_index, remove by index
3240  * Filter ID may come from userland and must be range-checked.
3241  */
3242 static int efx_ef10_filter_remove_internal(struct efx_nic *efx,
3243 					   unsigned int priority_mask,
3244 					   u32 filter_id, bool by_index)
3245 {
3246 	unsigned int filter_idx = filter_id % HUNT_FILTER_TBL_ROWS;
3247 	struct efx_ef10_filter_table *table = efx->filter_state;
3248 	MCDI_DECLARE_BUF(inbuf,
3249 			 MC_CMD_FILTER_OP_IN_HANDLE_OFST +
3250 			 MC_CMD_FILTER_OP_IN_HANDLE_LEN);
3251 	struct efx_filter_spec *spec;
3252 	DEFINE_WAIT(wait);
3253 	int rc;
3254 
3255 	/* Find the software table entry and mark it busy.  Don't
3256 	 * remove it yet; any attempt to update while we're waiting
3257 	 * for the firmware must find the busy entry.
3258 	 */
3259 	for (;;) {
3260 		spin_lock_bh(&efx->filter_lock);
3261 		if (!(table->entry[filter_idx].spec &
3262 		      EFX_EF10_FILTER_FLAG_BUSY))
3263 			break;
3264 		prepare_to_wait(&table->waitq, &wait, TASK_UNINTERRUPTIBLE);
3265 		spin_unlock_bh(&efx->filter_lock);
3266 		schedule();
3267 	}
3268 
3269 	spec = efx_ef10_filter_entry_spec(table, filter_idx);
3270 	if (!spec ||
3271 	    (!by_index &&
3272 	     efx_ef10_filter_rx_match_pri(table, spec->match_flags) !=
3273 	     filter_id / HUNT_FILTER_TBL_ROWS)) {
3274 		rc = -ENOENT;
3275 		goto out_unlock;
3276 	}
3277 
3278 	if (spec->flags & EFX_FILTER_FLAG_RX_OVER_AUTO &&
3279 	    priority_mask == (1U << EFX_FILTER_PRI_AUTO)) {
3280 		/* Just remove flags */
3281 		spec->flags &= ~EFX_FILTER_FLAG_RX_OVER_AUTO;
3282 		table->entry[filter_idx].spec &= ~EFX_EF10_FILTER_FLAG_AUTO_OLD;
3283 		rc = 0;
3284 		goto out_unlock;
3285 	}
3286 
3287 	if (!(priority_mask & (1U << spec->priority))) {
3288 		rc = -ENOENT;
3289 		goto out_unlock;
3290 	}
3291 
3292 	table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_BUSY;
3293 	spin_unlock_bh(&efx->filter_lock);
3294 
3295 	if (spec->flags & EFX_FILTER_FLAG_RX_OVER_AUTO) {
3296 		/* Reset to an automatic filter */
3297 
3298 		struct efx_filter_spec new_spec = *spec;
3299 
3300 		new_spec.priority = EFX_FILTER_PRI_AUTO;
3301 		new_spec.flags = (EFX_FILTER_FLAG_RX |
3302 				  EFX_FILTER_FLAG_RX_RSS);
3303 		new_spec.dmaq_id = 0;
3304 		new_spec.rss_context = EFX_FILTER_RSS_CONTEXT_DEFAULT;
3305 		rc = efx_ef10_filter_push(efx, &new_spec,
3306 					  &table->entry[filter_idx].handle,
3307 					  true);
3308 
3309 		spin_lock_bh(&efx->filter_lock);
3310 		if (rc == 0)
3311 			*spec = new_spec;
3312 	} else {
3313 		/* Really remove the filter */
3314 
3315 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
3316 			       efx_ef10_filter_is_exclusive(spec) ?
3317 			       MC_CMD_FILTER_OP_IN_OP_REMOVE :
3318 			       MC_CMD_FILTER_OP_IN_OP_UNSUBSCRIBE);
3319 		MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
3320 			       table->entry[filter_idx].handle);
3321 		rc = efx_mcdi_rpc(efx, MC_CMD_FILTER_OP,
3322 				  inbuf, sizeof(inbuf), NULL, 0, NULL);
3323 
3324 		spin_lock_bh(&efx->filter_lock);
3325 		if (rc == 0) {
3326 			kfree(spec);
3327 			efx_ef10_filter_set_entry(table, filter_idx, NULL, 0);
3328 		}
3329 	}
3330 
3331 	table->entry[filter_idx].spec &= ~EFX_EF10_FILTER_FLAG_BUSY;
3332 	wake_up_all(&table->waitq);
3333 out_unlock:
3334 	spin_unlock_bh(&efx->filter_lock);
3335 	finish_wait(&table->waitq, &wait);
3336 	return rc;
3337 }
3338 
3339 static int efx_ef10_filter_remove_safe(struct efx_nic *efx,
3340 				       enum efx_filter_priority priority,
3341 				       u32 filter_id)
3342 {
3343 	return efx_ef10_filter_remove_internal(efx, 1U << priority,
3344 					       filter_id, false);
3345 }
3346 
3347 static u32 efx_ef10_filter_get_unsafe_id(struct efx_nic *efx, u32 filter_id)
3348 {
3349 	return filter_id % HUNT_FILTER_TBL_ROWS;
3350 }
3351 
3352 static int efx_ef10_filter_remove_unsafe(struct efx_nic *efx,
3353 					 enum efx_filter_priority priority,
3354 					 u32 filter_id)
3355 {
3356 	return efx_ef10_filter_remove_internal(efx, 1U << priority,
3357 					       filter_id, true);
3358 }
3359 
3360 static int efx_ef10_filter_get_safe(struct efx_nic *efx,
3361 				    enum efx_filter_priority priority,
3362 				    u32 filter_id, struct efx_filter_spec *spec)
3363 {
3364 	unsigned int filter_idx = filter_id % HUNT_FILTER_TBL_ROWS;
3365 	struct efx_ef10_filter_table *table = efx->filter_state;
3366 	const struct efx_filter_spec *saved_spec;
3367 	int rc;
3368 
3369 	spin_lock_bh(&efx->filter_lock);
3370 	saved_spec = efx_ef10_filter_entry_spec(table, filter_idx);
3371 	if (saved_spec && saved_spec->priority == priority &&
3372 	    efx_ef10_filter_rx_match_pri(table, saved_spec->match_flags) ==
3373 	    filter_id / HUNT_FILTER_TBL_ROWS) {
3374 		*spec = *saved_spec;
3375 		rc = 0;
3376 	} else {
3377 		rc = -ENOENT;
3378 	}
3379 	spin_unlock_bh(&efx->filter_lock);
3380 	return rc;
3381 }
3382 
3383 static int efx_ef10_filter_clear_rx(struct efx_nic *efx,
3384 				     enum efx_filter_priority priority)
3385 {
3386 	unsigned int priority_mask;
3387 	unsigned int i;
3388 	int rc;
3389 
3390 	priority_mask = (((1U << (priority + 1)) - 1) &
3391 			 ~(1U << EFX_FILTER_PRI_AUTO));
3392 
3393 	for (i = 0; i < HUNT_FILTER_TBL_ROWS; i++) {
3394 		rc = efx_ef10_filter_remove_internal(efx, priority_mask,
3395 						     i, true);
3396 		if (rc && rc != -ENOENT)
3397 			return rc;
3398 	}
3399 
3400 	return 0;
3401 }
3402 
3403 static u32 efx_ef10_filter_count_rx_used(struct efx_nic *efx,
3404 					 enum efx_filter_priority priority)
3405 {
3406 	struct efx_ef10_filter_table *table = efx->filter_state;
3407 	unsigned int filter_idx;
3408 	s32 count = 0;
3409 
3410 	spin_lock_bh(&efx->filter_lock);
3411 	for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
3412 		if (table->entry[filter_idx].spec &&
3413 		    efx_ef10_filter_entry_spec(table, filter_idx)->priority ==
3414 		    priority)
3415 			++count;
3416 	}
3417 	spin_unlock_bh(&efx->filter_lock);
3418 	return count;
3419 }
3420 
3421 static u32 efx_ef10_filter_get_rx_id_limit(struct efx_nic *efx)
3422 {
3423 	struct efx_ef10_filter_table *table = efx->filter_state;
3424 
3425 	return table->rx_match_count * HUNT_FILTER_TBL_ROWS;
3426 }
3427 
3428 static s32 efx_ef10_filter_get_rx_ids(struct efx_nic *efx,
3429 				      enum efx_filter_priority priority,
3430 				      u32 *buf, u32 size)
3431 {
3432 	struct efx_ef10_filter_table *table = efx->filter_state;
3433 	struct efx_filter_spec *spec;
3434 	unsigned int filter_idx;
3435 	s32 count = 0;
3436 
3437 	spin_lock_bh(&efx->filter_lock);
3438 	for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
3439 		spec = efx_ef10_filter_entry_spec(table, filter_idx);
3440 		if (spec && spec->priority == priority) {
3441 			if (count == size) {
3442 				count = -EMSGSIZE;
3443 				break;
3444 			}
3445 			buf[count++] = (efx_ef10_filter_rx_match_pri(
3446 						table, spec->match_flags) *
3447 					HUNT_FILTER_TBL_ROWS +
3448 					filter_idx);
3449 		}
3450 	}
3451 	spin_unlock_bh(&efx->filter_lock);
3452 	return count;
3453 }
3454 
3455 #ifdef CONFIG_RFS_ACCEL
3456 
3457 static efx_mcdi_async_completer efx_ef10_filter_rfs_insert_complete;
3458 
3459 static s32 efx_ef10_filter_rfs_insert(struct efx_nic *efx,
3460 				      struct efx_filter_spec *spec)
3461 {
3462 	struct efx_ef10_filter_table *table = efx->filter_state;
3463 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_IN_LEN);
3464 	struct efx_filter_spec *saved_spec;
3465 	unsigned int hash, i, depth = 1;
3466 	bool replacing = false;
3467 	int ins_index = -1;
3468 	u64 cookie;
3469 	s32 rc;
3470 
3471 	/* Must be an RX filter without RSS and not for a multicast
3472 	 * destination address (RFS only works for connected sockets).
3473 	 * These restrictions allow us to pass only a tiny amount of
3474 	 * data through to the completion function.
3475 	 */
3476 	EFX_WARN_ON_PARANOID(spec->flags !=
3477 			     (EFX_FILTER_FLAG_RX | EFX_FILTER_FLAG_RX_SCATTER));
3478 	EFX_WARN_ON_PARANOID(spec->priority != EFX_FILTER_PRI_HINT);
3479 	EFX_WARN_ON_PARANOID(efx_filter_is_mc_recipient(spec));
3480 
3481 	hash = efx_ef10_filter_hash(spec);
3482 
3483 	spin_lock_bh(&efx->filter_lock);
3484 
3485 	/* Find any existing filter with the same match tuple or else
3486 	 * a free slot to insert at.  If an existing filter is busy,
3487 	 * we have to give up.
3488 	 */
3489 	for (;;) {
3490 		i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
3491 		saved_spec = efx_ef10_filter_entry_spec(table, i);
3492 
3493 		if (!saved_spec) {
3494 			if (ins_index < 0)
3495 				ins_index = i;
3496 		} else if (efx_ef10_filter_equal(spec, saved_spec)) {
3497 			if (table->entry[i].spec & EFX_EF10_FILTER_FLAG_BUSY) {
3498 				rc = -EBUSY;
3499 				goto fail_unlock;
3500 			}
3501 			if (spec->priority < saved_spec->priority) {
3502 				rc = -EPERM;
3503 				goto fail_unlock;
3504 			}
3505 			ins_index = i;
3506 			break;
3507 		}
3508 
3509 		/* Once we reach the maximum search depth, use the
3510 		 * first suitable slot or return -EBUSY if there was
3511 		 * none
3512 		 */
3513 		if (depth == EFX_EF10_FILTER_SEARCH_LIMIT) {
3514 			if (ins_index < 0) {
3515 				rc = -EBUSY;
3516 				goto fail_unlock;
3517 			}
3518 			break;
3519 		}
3520 
3521 		++depth;
3522 	}
3523 
3524 	/* Create a software table entry if necessary, and mark it
3525 	 * busy.  We might yet fail to insert, but any attempt to
3526 	 * insert a conflicting filter while we're waiting for the
3527 	 * firmware must find the busy entry.
3528 	 */
3529 	saved_spec = efx_ef10_filter_entry_spec(table, ins_index);
3530 	if (saved_spec) {
3531 		replacing = true;
3532 	} else {
3533 		saved_spec = kmalloc(sizeof(*spec), GFP_ATOMIC);
3534 		if (!saved_spec) {
3535 			rc = -ENOMEM;
3536 			goto fail_unlock;
3537 		}
3538 		*saved_spec = *spec;
3539 	}
3540 	efx_ef10_filter_set_entry(table, ins_index, saved_spec,
3541 				  EFX_EF10_FILTER_FLAG_BUSY);
3542 
3543 	spin_unlock_bh(&efx->filter_lock);
3544 
3545 	/* Pack up the variables needed on completion */
3546 	cookie = replacing << 31 | ins_index << 16 | spec->dmaq_id;
3547 
3548 	efx_ef10_filter_push_prep(efx, spec, inbuf,
3549 				  table->entry[ins_index].handle, replacing);
3550 	efx_mcdi_rpc_async(efx, MC_CMD_FILTER_OP, inbuf, sizeof(inbuf),
3551 			   MC_CMD_FILTER_OP_OUT_LEN,
3552 			   efx_ef10_filter_rfs_insert_complete, cookie);
3553 
3554 	return ins_index;
3555 
3556 fail_unlock:
3557 	spin_unlock_bh(&efx->filter_lock);
3558 	return rc;
3559 }
3560 
3561 static void
3562 efx_ef10_filter_rfs_insert_complete(struct efx_nic *efx, unsigned long cookie,
3563 				    int rc, efx_dword_t *outbuf,
3564 				    size_t outlen_actual)
3565 {
3566 	struct efx_ef10_filter_table *table = efx->filter_state;
3567 	unsigned int ins_index, dmaq_id;
3568 	struct efx_filter_spec *spec;
3569 	bool replacing;
3570 
3571 	/* Unpack the cookie */
3572 	replacing = cookie >> 31;
3573 	ins_index = (cookie >> 16) & (HUNT_FILTER_TBL_ROWS - 1);
3574 	dmaq_id = cookie & 0xffff;
3575 
3576 	spin_lock_bh(&efx->filter_lock);
3577 	spec = efx_ef10_filter_entry_spec(table, ins_index);
3578 	if (rc == 0) {
3579 		table->entry[ins_index].handle =
3580 			MCDI_QWORD(outbuf, FILTER_OP_OUT_HANDLE);
3581 		if (replacing)
3582 			spec->dmaq_id = dmaq_id;
3583 	} else if (!replacing) {
3584 		kfree(spec);
3585 		spec = NULL;
3586 	}
3587 	efx_ef10_filter_set_entry(table, ins_index, spec, 0);
3588 	spin_unlock_bh(&efx->filter_lock);
3589 
3590 	wake_up_all(&table->waitq);
3591 }
3592 
3593 static void
3594 efx_ef10_filter_rfs_expire_complete(struct efx_nic *efx,
3595 				    unsigned long filter_idx,
3596 				    int rc, efx_dword_t *outbuf,
3597 				    size_t outlen_actual);
3598 
3599 static bool efx_ef10_filter_rfs_expire_one(struct efx_nic *efx, u32 flow_id,
3600 					   unsigned int filter_idx)
3601 {
3602 	struct efx_ef10_filter_table *table = efx->filter_state;
3603 	struct efx_filter_spec *spec =
3604 		efx_ef10_filter_entry_spec(table, filter_idx);
3605 	MCDI_DECLARE_BUF(inbuf,
3606 			 MC_CMD_FILTER_OP_IN_HANDLE_OFST +
3607 			 MC_CMD_FILTER_OP_IN_HANDLE_LEN);
3608 
3609 	if (!spec ||
3610 	    (table->entry[filter_idx].spec & EFX_EF10_FILTER_FLAG_BUSY) ||
3611 	    spec->priority != EFX_FILTER_PRI_HINT ||
3612 	    !rps_may_expire_flow(efx->net_dev, spec->dmaq_id,
3613 				 flow_id, filter_idx))
3614 		return false;
3615 
3616 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
3617 		       MC_CMD_FILTER_OP_IN_OP_REMOVE);
3618 	MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
3619 		       table->entry[filter_idx].handle);
3620 	if (efx_mcdi_rpc_async(efx, MC_CMD_FILTER_OP, inbuf, sizeof(inbuf), 0,
3621 			       efx_ef10_filter_rfs_expire_complete, filter_idx))
3622 		return false;
3623 
3624 	table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_BUSY;
3625 	return true;
3626 }
3627 
3628 static void
3629 efx_ef10_filter_rfs_expire_complete(struct efx_nic *efx,
3630 				    unsigned long filter_idx,
3631 				    int rc, efx_dword_t *outbuf,
3632 				    size_t outlen_actual)
3633 {
3634 	struct efx_ef10_filter_table *table = efx->filter_state;
3635 	struct efx_filter_spec *spec =
3636 		efx_ef10_filter_entry_spec(table, filter_idx);
3637 
3638 	spin_lock_bh(&efx->filter_lock);
3639 	if (rc == 0) {
3640 		kfree(spec);
3641 		efx_ef10_filter_set_entry(table, filter_idx, NULL, 0);
3642 	}
3643 	table->entry[filter_idx].spec &= ~EFX_EF10_FILTER_FLAG_BUSY;
3644 	wake_up_all(&table->waitq);
3645 	spin_unlock_bh(&efx->filter_lock);
3646 }
3647 
3648 #endif /* CONFIG_RFS_ACCEL */
3649 
3650 static int efx_ef10_filter_match_flags_from_mcdi(u32 mcdi_flags)
3651 {
3652 	int match_flags = 0;
3653 
3654 #define MAP_FLAG(gen_flag, mcdi_field) {				\
3655 		u32 old_mcdi_flags = mcdi_flags;			\
3656 		mcdi_flags &= ~(1 << MC_CMD_FILTER_OP_IN_MATCH_ ##	\
3657 				mcdi_field ## _LBN);			\
3658 		if (mcdi_flags != old_mcdi_flags)			\
3659 			match_flags |= EFX_FILTER_MATCH_ ## gen_flag;	\
3660 	}
3661 	MAP_FLAG(LOC_MAC_IG, UNKNOWN_UCAST_DST);
3662 	MAP_FLAG(LOC_MAC_IG, UNKNOWN_MCAST_DST);
3663 	MAP_FLAG(REM_HOST, SRC_IP);
3664 	MAP_FLAG(LOC_HOST, DST_IP);
3665 	MAP_FLAG(REM_MAC, SRC_MAC);
3666 	MAP_FLAG(REM_PORT, SRC_PORT);
3667 	MAP_FLAG(LOC_MAC, DST_MAC);
3668 	MAP_FLAG(LOC_PORT, DST_PORT);
3669 	MAP_FLAG(ETHER_TYPE, ETHER_TYPE);
3670 	MAP_FLAG(INNER_VID, INNER_VLAN);
3671 	MAP_FLAG(OUTER_VID, OUTER_VLAN);
3672 	MAP_FLAG(IP_PROTO, IP_PROTO);
3673 #undef MAP_FLAG
3674 
3675 	/* Did we map them all? */
3676 	if (mcdi_flags)
3677 		return -EINVAL;
3678 
3679 	return match_flags;
3680 }
3681 
3682 static int efx_ef10_filter_table_probe(struct efx_nic *efx)
3683 {
3684 	MCDI_DECLARE_BUF(inbuf, MC_CMD_GET_PARSER_DISP_INFO_IN_LEN);
3685 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_PARSER_DISP_INFO_OUT_LENMAX);
3686 	unsigned int pd_match_pri, pd_match_count;
3687 	struct efx_ef10_filter_table *table;
3688 	size_t outlen;
3689 	int rc;
3690 
3691 	table = kzalloc(sizeof(*table), GFP_KERNEL);
3692 	if (!table)
3693 		return -ENOMEM;
3694 
3695 	/* Find out which RX filter types are supported, and their priorities */
3696 	MCDI_SET_DWORD(inbuf, GET_PARSER_DISP_INFO_IN_OP,
3697 		       MC_CMD_GET_PARSER_DISP_INFO_IN_OP_GET_SUPPORTED_RX_MATCHES);
3698 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_PARSER_DISP_INFO,
3699 			  inbuf, sizeof(inbuf), outbuf, sizeof(outbuf),
3700 			  &outlen);
3701 	if (rc)
3702 		goto fail;
3703 	pd_match_count = MCDI_VAR_ARRAY_LEN(
3704 		outlen, GET_PARSER_DISP_INFO_OUT_SUPPORTED_MATCHES);
3705 	table->rx_match_count = 0;
3706 
3707 	for (pd_match_pri = 0; pd_match_pri < pd_match_count; pd_match_pri++) {
3708 		u32 mcdi_flags =
3709 			MCDI_ARRAY_DWORD(
3710 				outbuf,
3711 				GET_PARSER_DISP_INFO_OUT_SUPPORTED_MATCHES,
3712 				pd_match_pri);
3713 		rc = efx_ef10_filter_match_flags_from_mcdi(mcdi_flags);
3714 		if (rc < 0) {
3715 			netif_dbg(efx, probe, efx->net_dev,
3716 				  "%s: fw flags %#x pri %u not supported in driver\n",
3717 				  __func__, mcdi_flags, pd_match_pri);
3718 		} else {
3719 			netif_dbg(efx, probe, efx->net_dev,
3720 				  "%s: fw flags %#x pri %u supported as driver flags %#x pri %u\n",
3721 				  __func__, mcdi_flags, pd_match_pri,
3722 				  rc, table->rx_match_count);
3723 			table->rx_match_flags[table->rx_match_count++] = rc;
3724 		}
3725 	}
3726 
3727 	table->entry = vzalloc(HUNT_FILTER_TBL_ROWS * sizeof(*table->entry));
3728 	if (!table->entry) {
3729 		rc = -ENOMEM;
3730 		goto fail;
3731 	}
3732 
3733 	table->ucdef_id = EFX_EF10_FILTER_ID_INVALID;
3734 	table->bcast_id = EFX_EF10_FILTER_ID_INVALID;
3735 	table->mcdef_id = EFX_EF10_FILTER_ID_INVALID;
3736 
3737 	efx->filter_state = table;
3738 	init_waitqueue_head(&table->waitq);
3739 	return 0;
3740 
3741 fail:
3742 	kfree(table);
3743 	return rc;
3744 }
3745 
3746 /* Caller must hold efx->filter_sem for read if race against
3747  * efx_ef10_filter_table_remove() is possible
3748  */
3749 static void efx_ef10_filter_table_restore(struct efx_nic *efx)
3750 {
3751 	struct efx_ef10_filter_table *table = efx->filter_state;
3752 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
3753 	struct efx_filter_spec *spec;
3754 	unsigned int filter_idx;
3755 	bool failed = false;
3756 	int rc;
3757 
3758 	WARN_ON(!rwsem_is_locked(&efx->filter_sem));
3759 
3760 	if (!nic_data->must_restore_filters)
3761 		return;
3762 
3763 	if (!table)
3764 		return;
3765 
3766 	spin_lock_bh(&efx->filter_lock);
3767 
3768 	for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
3769 		spec = efx_ef10_filter_entry_spec(table, filter_idx);
3770 		if (!spec)
3771 			continue;
3772 
3773 		table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_BUSY;
3774 		spin_unlock_bh(&efx->filter_lock);
3775 
3776 		rc = efx_ef10_filter_push(efx, spec,
3777 					  &table->entry[filter_idx].handle,
3778 					  false);
3779 		if (rc)
3780 			failed = true;
3781 
3782 		spin_lock_bh(&efx->filter_lock);
3783 		if (rc) {
3784 			kfree(spec);
3785 			efx_ef10_filter_set_entry(table, filter_idx, NULL, 0);
3786 		} else {
3787 			table->entry[filter_idx].spec &=
3788 				~EFX_EF10_FILTER_FLAG_BUSY;
3789 		}
3790 	}
3791 
3792 	spin_unlock_bh(&efx->filter_lock);
3793 
3794 	if (failed)
3795 		netif_err(efx, hw, efx->net_dev,
3796 			  "unable to restore all filters\n");
3797 	else
3798 		nic_data->must_restore_filters = false;
3799 }
3800 
3801 /* Caller must hold efx->filter_sem for write */
3802 static void efx_ef10_filter_table_remove(struct efx_nic *efx)
3803 {
3804 	struct efx_ef10_filter_table *table = efx->filter_state;
3805 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_IN_LEN);
3806 	struct efx_filter_spec *spec;
3807 	unsigned int filter_idx;
3808 	int rc;
3809 
3810 	efx->filter_state = NULL;
3811 	if (!table)
3812 		return;
3813 
3814 	for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
3815 		spec = efx_ef10_filter_entry_spec(table, filter_idx);
3816 		if (!spec)
3817 			continue;
3818 
3819 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
3820 			       efx_ef10_filter_is_exclusive(spec) ?
3821 			       MC_CMD_FILTER_OP_IN_OP_REMOVE :
3822 			       MC_CMD_FILTER_OP_IN_OP_UNSUBSCRIBE);
3823 		MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
3824 			       table->entry[filter_idx].handle);
3825 		rc = efx_mcdi_rpc(efx, MC_CMD_FILTER_OP, inbuf, sizeof(inbuf),
3826 				  NULL, 0, NULL);
3827 		if (rc)
3828 			netdev_WARN(efx->net_dev,
3829 				    "filter_idx=%#x handle=%#llx\n",
3830 				    filter_idx,
3831 				    table->entry[filter_idx].handle);
3832 		kfree(spec);
3833 	}
3834 
3835 	vfree(table->entry);
3836 	kfree(table);
3837 }
3838 
3839 #define EFX_EF10_FILTER_DO_MARK_OLD(id) \
3840 		if (id != EFX_EF10_FILTER_ID_INVALID) { \
3841 			filter_idx = efx_ef10_filter_get_unsafe_id(efx, id); \
3842 			WARN_ON(!table->entry[filter_idx].spec); \
3843 			table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_AUTO_OLD; \
3844 		}
3845 static void efx_ef10_filter_mark_old(struct efx_nic *efx)
3846 {
3847 	struct efx_ef10_filter_table *table = efx->filter_state;
3848 	unsigned int filter_idx, i;
3849 
3850 	if (!table)
3851 		return;
3852 
3853 	/* Mark old filters that may need to be removed */
3854 	spin_lock_bh(&efx->filter_lock);
3855 	for (i = 0; i < table->dev_uc_count; i++)
3856 		EFX_EF10_FILTER_DO_MARK_OLD(table->dev_uc_list[i].id);
3857 	for (i = 0; i < table->dev_mc_count; i++)
3858 		EFX_EF10_FILTER_DO_MARK_OLD(table->dev_mc_list[i].id);
3859 	EFX_EF10_FILTER_DO_MARK_OLD(table->ucdef_id);
3860 	EFX_EF10_FILTER_DO_MARK_OLD(table->bcast_id);
3861 	EFX_EF10_FILTER_DO_MARK_OLD(table->mcdef_id);
3862 	spin_unlock_bh(&efx->filter_lock);
3863 }
3864 #undef EFX_EF10_FILTER_DO_MARK_OLD
3865 
3866 static void efx_ef10_filter_uc_addr_list(struct efx_nic *efx, bool *promisc)
3867 {
3868 	struct efx_ef10_filter_table *table = efx->filter_state;
3869 	struct net_device *net_dev = efx->net_dev;
3870 	struct netdev_hw_addr *uc;
3871 	int addr_count;
3872 	unsigned int i;
3873 
3874 	table->ucdef_id = EFX_EF10_FILTER_ID_INVALID;
3875 	addr_count = netdev_uc_count(net_dev);
3876 	if (net_dev->flags & IFF_PROMISC)
3877 		*promisc = true;
3878 	table->dev_uc_count = 1 + addr_count;
3879 	ether_addr_copy(table->dev_uc_list[0].addr, net_dev->dev_addr);
3880 	i = 1;
3881 	netdev_for_each_uc_addr(uc, net_dev) {
3882 		if (i >= EFX_EF10_FILTER_DEV_UC_MAX) {
3883 			*promisc = true;
3884 			break;
3885 		}
3886 		ether_addr_copy(table->dev_uc_list[i].addr, uc->addr);
3887 		table->dev_uc_list[i].id = EFX_EF10_FILTER_ID_INVALID;
3888 		i++;
3889 	}
3890 }
3891 
3892 static void efx_ef10_filter_mc_addr_list(struct efx_nic *efx, bool *promisc)
3893 {
3894 	struct efx_ef10_filter_table *table = efx->filter_state;
3895 	struct net_device *net_dev = efx->net_dev;
3896 	struct netdev_hw_addr *mc;
3897 	unsigned int i, addr_count;
3898 
3899 	table->mcdef_id = EFX_EF10_FILTER_ID_INVALID;
3900 	table->bcast_id = EFX_EF10_FILTER_ID_INVALID;
3901 	if (net_dev->flags & (IFF_PROMISC | IFF_ALLMULTI))
3902 		*promisc = true;
3903 
3904 	addr_count = netdev_mc_count(net_dev);
3905 	i = 0;
3906 	netdev_for_each_mc_addr(mc, net_dev) {
3907 		if (i >= EFX_EF10_FILTER_DEV_MC_MAX) {
3908 			*promisc = true;
3909 			break;
3910 		}
3911 		ether_addr_copy(table->dev_mc_list[i].addr, mc->addr);
3912 		table->dev_mc_list[i].id = EFX_EF10_FILTER_ID_INVALID;
3913 		i++;
3914 	}
3915 
3916 	table->dev_mc_count = i;
3917 }
3918 
3919 static int efx_ef10_filter_insert_addr_list(struct efx_nic *efx,
3920 					     bool multicast, bool rollback)
3921 {
3922 	struct efx_ef10_filter_table *table = efx->filter_state;
3923 	struct efx_ef10_dev_addr *addr_list;
3924 	struct efx_filter_spec spec;
3925 	u8 baddr[ETH_ALEN];
3926 	unsigned int i, j;
3927 	int addr_count;
3928 	int rc;
3929 
3930 	if (multicast) {
3931 		addr_list = table->dev_mc_list;
3932 		addr_count = table->dev_mc_count;
3933 	} else {
3934 		addr_list = table->dev_uc_list;
3935 		addr_count = table->dev_uc_count;
3936 	}
3937 
3938 	/* Insert/renew filters */
3939 	for (i = 0; i < addr_count; i++) {
3940 		efx_filter_init_rx(&spec, EFX_FILTER_PRI_AUTO,
3941 				   EFX_FILTER_FLAG_RX_RSS,
3942 				   0);
3943 		efx_filter_set_eth_local(&spec, EFX_FILTER_VID_UNSPEC,
3944 					 addr_list[i].addr);
3945 		rc = efx_ef10_filter_insert(efx, &spec, true);
3946 		if (rc < 0) {
3947 			if (rollback) {
3948 				netif_info(efx, drv, efx->net_dev,
3949 					   "efx_ef10_filter_insert failed rc=%d\n",
3950 					   rc);
3951 				/* Fall back to promiscuous */
3952 				for (j = 0; j < i; j++) {
3953 					if (addr_list[j].id == EFX_EF10_FILTER_ID_INVALID)
3954 						continue;
3955 					efx_ef10_filter_remove_unsafe(
3956 						efx, EFX_FILTER_PRI_AUTO,
3957 						addr_list[j].id);
3958 					addr_list[j].id = EFX_EF10_FILTER_ID_INVALID;
3959 				}
3960 				return rc;
3961 			} else {
3962 				/* mark as not inserted, and carry on */
3963 				rc = EFX_EF10_FILTER_ID_INVALID;
3964 			}
3965 		}
3966 		addr_list[i].id = efx_ef10_filter_get_unsafe_id(efx, rc);
3967 	}
3968 
3969 	if (multicast && rollback) {
3970 		/* Also need an Ethernet broadcast filter */
3971 		efx_filter_init_rx(&spec, EFX_FILTER_PRI_AUTO,
3972 				   EFX_FILTER_FLAG_RX_RSS,
3973 				   0);
3974 		eth_broadcast_addr(baddr);
3975 		efx_filter_set_eth_local(&spec, EFX_FILTER_VID_UNSPEC, baddr);
3976 		rc = efx_ef10_filter_insert(efx, &spec, true);
3977 		if (rc < 0) {
3978 			netif_warn(efx, drv, efx->net_dev,
3979 				   "Broadcast filter insert failed rc=%d\n", rc);
3980 			/* Fall back to promiscuous */
3981 			for (j = 0; j < i; j++) {
3982 				if (addr_list[j].id == EFX_EF10_FILTER_ID_INVALID)
3983 					continue;
3984 				efx_ef10_filter_remove_unsafe(
3985 					efx, EFX_FILTER_PRI_AUTO,
3986 					addr_list[j].id);
3987 				addr_list[j].id = EFX_EF10_FILTER_ID_INVALID;
3988 			}
3989 			return rc;
3990 		} else {
3991 			table->bcast_id = efx_ef10_filter_get_unsafe_id(efx, rc);
3992 		}
3993 	}
3994 
3995 	return 0;
3996 }
3997 
3998 static int efx_ef10_filter_insert_def(struct efx_nic *efx, bool multicast,
3999 				      bool rollback)
4000 {
4001 	struct efx_ef10_filter_table *table = efx->filter_state;
4002 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
4003 	struct efx_filter_spec spec;
4004 	u8 baddr[ETH_ALEN];
4005 	int rc;
4006 
4007 	efx_filter_init_rx(&spec, EFX_FILTER_PRI_AUTO,
4008 			   EFX_FILTER_FLAG_RX_RSS,
4009 			   0);
4010 
4011 	if (multicast)
4012 		efx_filter_set_mc_def(&spec);
4013 	else
4014 		efx_filter_set_uc_def(&spec);
4015 
4016 	rc = efx_ef10_filter_insert(efx, &spec, true);
4017 	if (rc < 0) {
4018 		netif_warn(efx, drv, efx->net_dev,
4019 			   "%scast mismatch filter insert failed rc=%d\n",
4020 			   multicast ? "Multi" : "Uni", rc);
4021 	} else if (multicast) {
4022 		table->mcdef_id = efx_ef10_filter_get_unsafe_id(efx, rc);
4023 		if (!nic_data->workaround_26807) {
4024 			/* Also need an Ethernet broadcast filter */
4025 			efx_filter_init_rx(&spec, EFX_FILTER_PRI_AUTO,
4026 					   EFX_FILTER_FLAG_RX_RSS,
4027 					   0);
4028 			eth_broadcast_addr(baddr);
4029 			efx_filter_set_eth_local(&spec, EFX_FILTER_VID_UNSPEC,
4030 						 baddr);
4031 			rc = efx_ef10_filter_insert(efx, &spec, true);
4032 			if (rc < 0) {
4033 				netif_warn(efx, drv, efx->net_dev,
4034 					   "Broadcast filter insert failed rc=%d\n",
4035 					   rc);
4036 				if (rollback) {
4037 					/* Roll back the mc_def filter */
4038 					efx_ef10_filter_remove_unsafe(
4039 							efx, EFX_FILTER_PRI_AUTO,
4040 							table->mcdef_id);
4041 					table->mcdef_id = EFX_EF10_FILTER_ID_INVALID;
4042 					return rc;
4043 				}
4044 			} else {
4045 				table->bcast_id = efx_ef10_filter_get_unsafe_id(efx, rc);
4046 			}
4047 		}
4048 		rc = 0;
4049 	} else {
4050 		table->ucdef_id = rc;
4051 		rc = 0;
4052 	}
4053 	return rc;
4054 }
4055 
4056 /* Remove filters that weren't renewed.  Since nothing else changes the AUTO_OLD
4057  * flag or removes these filters, we don't need to hold the filter_lock while
4058  * scanning for these filters.
4059  */
4060 static void efx_ef10_filter_remove_old(struct efx_nic *efx)
4061 {
4062 	struct efx_ef10_filter_table *table = efx->filter_state;
4063 	bool remove_failed = false;
4064 	int i;
4065 
4066 	for (i = 0; i < HUNT_FILTER_TBL_ROWS; i++) {
4067 		if (ACCESS_ONCE(table->entry[i].spec) &
4068 		    EFX_EF10_FILTER_FLAG_AUTO_OLD) {
4069 			if (efx_ef10_filter_remove_internal(
4070 				    efx, 1U << EFX_FILTER_PRI_AUTO,
4071 				    i, true) < 0)
4072 				remove_failed = true;
4073 		}
4074 	}
4075 	WARN_ON(remove_failed);
4076 }
4077 
4078 static int efx_ef10_vport_set_mac_address(struct efx_nic *efx)
4079 {
4080 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
4081 	u8 mac_old[ETH_ALEN];
4082 	int rc, rc2;
4083 
4084 	/* Only reconfigure a PF-created vport */
4085 	if (is_zero_ether_addr(nic_data->vport_mac))
4086 		return 0;
4087 
4088 	efx_device_detach_sync(efx);
4089 	efx_net_stop(efx->net_dev);
4090 	down_write(&efx->filter_sem);
4091 	efx_ef10_filter_table_remove(efx);
4092 	up_write(&efx->filter_sem);
4093 
4094 	rc = efx_ef10_vadaptor_free(efx, nic_data->vport_id);
4095 	if (rc)
4096 		goto restore_filters;
4097 
4098 	ether_addr_copy(mac_old, nic_data->vport_mac);
4099 	rc = efx_ef10_vport_del_mac(efx, nic_data->vport_id,
4100 				    nic_data->vport_mac);
4101 	if (rc)
4102 		goto restore_vadaptor;
4103 
4104 	rc = efx_ef10_vport_add_mac(efx, nic_data->vport_id,
4105 				    efx->net_dev->dev_addr);
4106 	if (!rc) {
4107 		ether_addr_copy(nic_data->vport_mac, efx->net_dev->dev_addr);
4108 	} else {
4109 		rc2 = efx_ef10_vport_add_mac(efx, nic_data->vport_id, mac_old);
4110 		if (rc2) {
4111 			/* Failed to add original MAC, so clear vport_mac */
4112 			eth_zero_addr(nic_data->vport_mac);
4113 			goto reset_nic;
4114 		}
4115 	}
4116 
4117 restore_vadaptor:
4118 	rc2 = efx_ef10_vadaptor_alloc(efx, nic_data->vport_id);
4119 	if (rc2)
4120 		goto reset_nic;
4121 restore_filters:
4122 	down_write(&efx->filter_sem);
4123 	rc2 = efx_ef10_filter_table_probe(efx);
4124 	up_write(&efx->filter_sem);
4125 	if (rc2)
4126 		goto reset_nic;
4127 
4128 	rc2 = efx_net_open(efx->net_dev);
4129 	if (rc2)
4130 		goto reset_nic;
4131 
4132 	netif_device_attach(efx->net_dev);
4133 
4134 	return rc;
4135 
4136 reset_nic:
4137 	netif_err(efx, drv, efx->net_dev,
4138 		  "Failed to restore when changing MAC address - scheduling reset\n");
4139 	efx_schedule_reset(efx, RESET_TYPE_DATAPATH);
4140 
4141 	return rc ? rc : rc2;
4142 }
4143 
4144 /* Caller must hold efx->filter_sem for read if race against
4145  * efx_ef10_filter_table_remove() is possible
4146  */
4147 static void efx_ef10_filter_sync_rx_mode(struct efx_nic *efx)
4148 {
4149 	struct efx_ef10_filter_table *table = efx->filter_state;
4150 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
4151 	struct net_device *net_dev = efx->net_dev;
4152 	bool uc_promisc = false, mc_promisc = false;
4153 
4154 	if (!efx_dev_registered(efx))
4155 		return;
4156 
4157 	if (!table)
4158 		return;
4159 
4160 	efx_ef10_filter_mark_old(efx);
4161 
4162 	/* Copy/convert the address lists; add the primary station
4163 	 * address and broadcast address
4164 	 */
4165 	netif_addr_lock_bh(net_dev);
4166 	efx_ef10_filter_uc_addr_list(efx, &uc_promisc);
4167 	efx_ef10_filter_mc_addr_list(efx, &mc_promisc);
4168 	netif_addr_unlock_bh(net_dev);
4169 
4170 	/* Insert/renew unicast filters */
4171 	if (uc_promisc) {
4172 		efx_ef10_filter_insert_def(efx, false, false);
4173 		efx_ef10_filter_insert_addr_list(efx, false, false);
4174 	} else {
4175 		/* If any of the filters failed to insert, fall back to
4176 		 * promiscuous mode - add in the uc_def filter.  But keep
4177 		 * our individual unicast filters.
4178 		 */
4179 		if (efx_ef10_filter_insert_addr_list(efx, false, false))
4180 			efx_ef10_filter_insert_def(efx, false, false);
4181 	}
4182 
4183 	/* Insert/renew multicast filters */
4184 	/* If changing promiscuous state with cascaded multicast filters, remove
4185 	 * old filters first, so that packets are dropped rather than duplicated
4186 	 */
4187 	if (nic_data->workaround_26807 && efx->mc_promisc != mc_promisc)
4188 		efx_ef10_filter_remove_old(efx);
4189 	if (mc_promisc) {
4190 		if (nic_data->workaround_26807) {
4191 			/* If we failed to insert promiscuous filters, rollback
4192 			 * and fall back to individual multicast filters
4193 			 */
4194 			if (efx_ef10_filter_insert_def(efx, true, true)) {
4195 				/* Changing promisc state, so remove old filters */
4196 				efx_ef10_filter_remove_old(efx);
4197 				efx_ef10_filter_insert_addr_list(efx, true, false);
4198 			}
4199 		} else {
4200 			/* If we failed to insert promiscuous filters, don't
4201 			 * rollback.  Regardless, also insert the mc_list
4202 			 */
4203 			efx_ef10_filter_insert_def(efx, true, false);
4204 			efx_ef10_filter_insert_addr_list(efx, true, false);
4205 		}
4206 	} else {
4207 		/* If any filters failed to insert, rollback and fall back to
4208 		 * promiscuous mode - mc_def filter and maybe broadcast.  If
4209 		 * that fails, roll back again and insert as many of our
4210 		 * individual multicast filters as we can.
4211 		 */
4212 		if (efx_ef10_filter_insert_addr_list(efx, true, true)) {
4213 			/* Changing promisc state, so remove old filters */
4214 			if (nic_data->workaround_26807)
4215 				efx_ef10_filter_remove_old(efx);
4216 			if (efx_ef10_filter_insert_def(efx, true, true))
4217 				efx_ef10_filter_insert_addr_list(efx, true, false);
4218 		}
4219 	}
4220 
4221 	efx_ef10_filter_remove_old(efx);
4222 	efx->mc_promisc = mc_promisc;
4223 }
4224 
4225 static int efx_ef10_set_mac_address(struct efx_nic *efx)
4226 {
4227 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VADAPTOR_SET_MAC_IN_LEN);
4228 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
4229 	bool was_enabled = efx->port_enabled;
4230 	int rc;
4231 
4232 	efx_device_detach_sync(efx);
4233 	efx_net_stop(efx->net_dev);
4234 	down_write(&efx->filter_sem);
4235 	efx_ef10_filter_table_remove(efx);
4236 
4237 	ether_addr_copy(MCDI_PTR(inbuf, VADAPTOR_SET_MAC_IN_MACADDR),
4238 			efx->net_dev->dev_addr);
4239 	MCDI_SET_DWORD(inbuf, VADAPTOR_SET_MAC_IN_UPSTREAM_PORT_ID,
4240 		       nic_data->vport_id);
4241 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_VADAPTOR_SET_MAC, inbuf,
4242 				sizeof(inbuf), NULL, 0, NULL);
4243 
4244 	efx_ef10_filter_table_probe(efx);
4245 	up_write(&efx->filter_sem);
4246 	if (was_enabled)
4247 		efx_net_open(efx->net_dev);
4248 	netif_device_attach(efx->net_dev);
4249 
4250 #ifdef CONFIG_SFC_SRIOV
4251 	if (efx->pci_dev->is_virtfn && efx->pci_dev->physfn) {
4252 		struct pci_dev *pci_dev_pf = efx->pci_dev->physfn;
4253 
4254 		if (rc == -EPERM) {
4255 			struct efx_nic *efx_pf;
4256 
4257 			/* Switch to PF and change MAC address on vport */
4258 			efx_pf = pci_get_drvdata(pci_dev_pf);
4259 
4260 			rc = efx_ef10_sriov_set_vf_mac(efx_pf,
4261 						       nic_data->vf_index,
4262 						       efx->net_dev->dev_addr);
4263 		} else if (!rc) {
4264 			struct efx_nic *efx_pf = pci_get_drvdata(pci_dev_pf);
4265 			struct efx_ef10_nic_data *nic_data = efx_pf->nic_data;
4266 			unsigned int i;
4267 
4268 			/* MAC address successfully changed by VF (with MAC
4269 			 * spoofing) so update the parent PF if possible.
4270 			 */
4271 			for (i = 0; i < efx_pf->vf_count; ++i) {
4272 				struct ef10_vf *vf = nic_data->vf + i;
4273 
4274 				if (vf->efx == efx) {
4275 					ether_addr_copy(vf->mac,
4276 							efx->net_dev->dev_addr);
4277 					return 0;
4278 				}
4279 			}
4280 		}
4281 	} else
4282 #endif
4283 	if (rc == -EPERM) {
4284 		netif_err(efx, drv, efx->net_dev,
4285 			  "Cannot change MAC address; use sfboot to enable"
4286 			  " mac-spoofing on this interface\n");
4287 	} else if (rc == -ENOSYS && !efx_ef10_is_vf(efx)) {
4288 		/* If the active MCFW does not support MC_CMD_VADAPTOR_SET_MAC
4289 		 * fall-back to the method of changing the MAC address on the
4290 		 * vport.  This only applies to PFs because such versions of
4291 		 * MCFW do not support VFs.
4292 		 */
4293 		rc = efx_ef10_vport_set_mac_address(efx);
4294 	} else {
4295 		efx_mcdi_display_error(efx, MC_CMD_VADAPTOR_SET_MAC,
4296 				       sizeof(inbuf), NULL, 0, rc);
4297 	}
4298 
4299 	return rc;
4300 }
4301 
4302 static int efx_ef10_mac_reconfigure(struct efx_nic *efx)
4303 {
4304 	efx_ef10_filter_sync_rx_mode(efx);
4305 
4306 	return efx_mcdi_set_mac(efx);
4307 }
4308 
4309 static int efx_ef10_mac_reconfigure_vf(struct efx_nic *efx)
4310 {
4311 	efx_ef10_filter_sync_rx_mode(efx);
4312 
4313 	return 0;
4314 }
4315 
4316 static int efx_ef10_start_bist(struct efx_nic *efx, u32 bist_type)
4317 {
4318 	MCDI_DECLARE_BUF(inbuf, MC_CMD_START_BIST_IN_LEN);
4319 
4320 	MCDI_SET_DWORD(inbuf, START_BIST_IN_TYPE, bist_type);
4321 	return efx_mcdi_rpc(efx, MC_CMD_START_BIST, inbuf, sizeof(inbuf),
4322 			    NULL, 0, NULL);
4323 }
4324 
4325 /* MC BISTs follow a different poll mechanism to phy BISTs.
4326  * The BIST is done in the poll handler on the MC, and the MCDI command
4327  * will block until the BIST is done.
4328  */
4329 static int efx_ef10_poll_bist(struct efx_nic *efx)
4330 {
4331 	int rc;
4332 	MCDI_DECLARE_BUF(outbuf, MC_CMD_POLL_BIST_OUT_LEN);
4333 	size_t outlen;
4334 	u32 result;
4335 
4336 	rc = efx_mcdi_rpc(efx, MC_CMD_POLL_BIST, NULL, 0,
4337 			   outbuf, sizeof(outbuf), &outlen);
4338 	if (rc != 0)
4339 		return rc;
4340 
4341 	if (outlen < MC_CMD_POLL_BIST_OUT_LEN)
4342 		return -EIO;
4343 
4344 	result = MCDI_DWORD(outbuf, POLL_BIST_OUT_RESULT);
4345 	switch (result) {
4346 	case MC_CMD_POLL_BIST_PASSED:
4347 		netif_dbg(efx, hw, efx->net_dev, "BIST passed.\n");
4348 		return 0;
4349 	case MC_CMD_POLL_BIST_TIMEOUT:
4350 		netif_err(efx, hw, efx->net_dev, "BIST timed out\n");
4351 		return -EIO;
4352 	case MC_CMD_POLL_BIST_FAILED:
4353 		netif_err(efx, hw, efx->net_dev, "BIST failed.\n");
4354 		return -EIO;
4355 	default:
4356 		netif_err(efx, hw, efx->net_dev,
4357 			  "BIST returned unknown result %u", result);
4358 		return -EIO;
4359 	}
4360 }
4361 
4362 static int efx_ef10_run_bist(struct efx_nic *efx, u32 bist_type)
4363 {
4364 	int rc;
4365 
4366 	netif_dbg(efx, drv, efx->net_dev, "starting BIST type %u\n", bist_type);
4367 
4368 	rc = efx_ef10_start_bist(efx, bist_type);
4369 	if (rc != 0)
4370 		return rc;
4371 
4372 	return efx_ef10_poll_bist(efx);
4373 }
4374 
4375 static int
4376 efx_ef10_test_chip(struct efx_nic *efx, struct efx_self_tests *tests)
4377 {
4378 	int rc, rc2;
4379 
4380 	efx_reset_down(efx, RESET_TYPE_WORLD);
4381 
4382 	rc = efx_mcdi_rpc(efx, MC_CMD_ENABLE_OFFLINE_BIST,
4383 			  NULL, 0, NULL, 0, NULL);
4384 	if (rc != 0)
4385 		goto out;
4386 
4387 	tests->memory = efx_ef10_run_bist(efx, MC_CMD_MC_MEM_BIST) ? -1 : 1;
4388 	tests->registers = efx_ef10_run_bist(efx, MC_CMD_REG_BIST) ? -1 : 1;
4389 
4390 	rc = efx_mcdi_reset(efx, RESET_TYPE_WORLD);
4391 
4392 out:
4393 	if (rc == -EPERM)
4394 		rc = 0;
4395 	rc2 = efx_reset_up(efx, RESET_TYPE_WORLD, rc == 0);
4396 	return rc ? rc : rc2;
4397 }
4398 
4399 #ifdef CONFIG_SFC_MTD
4400 
4401 struct efx_ef10_nvram_type_info {
4402 	u16 type, type_mask;
4403 	u8 port;
4404 	const char *name;
4405 };
4406 
4407 static const struct efx_ef10_nvram_type_info efx_ef10_nvram_types[] = {
4408 	{ NVRAM_PARTITION_TYPE_MC_FIRMWARE,	   0,    0, "sfc_mcfw" },
4409 	{ NVRAM_PARTITION_TYPE_MC_FIRMWARE_BACKUP, 0,    0, "sfc_mcfw_backup" },
4410 	{ NVRAM_PARTITION_TYPE_EXPANSION_ROM,	   0,    0, "sfc_exp_rom" },
4411 	{ NVRAM_PARTITION_TYPE_STATIC_CONFIG,	   0,    0, "sfc_static_cfg" },
4412 	{ NVRAM_PARTITION_TYPE_DYNAMIC_CONFIG,	   0,    0, "sfc_dynamic_cfg" },
4413 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT0, 0,   0, "sfc_exp_rom_cfg" },
4414 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT1, 0,   1, "sfc_exp_rom_cfg" },
4415 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT2, 0,   2, "sfc_exp_rom_cfg" },
4416 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT3, 0,   3, "sfc_exp_rom_cfg" },
4417 	{ NVRAM_PARTITION_TYPE_LICENSE,		   0,    0, "sfc_license" },
4418 	{ NVRAM_PARTITION_TYPE_PHY_MIN,		   0xff, 0, "sfc_phy_fw" },
4419 };
4420 
4421 static int efx_ef10_mtd_probe_partition(struct efx_nic *efx,
4422 					struct efx_mcdi_mtd_partition *part,
4423 					unsigned int type)
4424 {
4425 	MCDI_DECLARE_BUF(inbuf, MC_CMD_NVRAM_METADATA_IN_LEN);
4426 	MCDI_DECLARE_BUF(outbuf, MC_CMD_NVRAM_METADATA_OUT_LENMAX);
4427 	const struct efx_ef10_nvram_type_info *info;
4428 	size_t size, erase_size, outlen;
4429 	bool protected;
4430 	int rc;
4431 
4432 	for (info = efx_ef10_nvram_types; ; info++) {
4433 		if (info ==
4434 		    efx_ef10_nvram_types + ARRAY_SIZE(efx_ef10_nvram_types))
4435 			return -ENODEV;
4436 		if ((type & ~info->type_mask) == info->type)
4437 			break;
4438 	}
4439 	if (info->port != efx_port_num(efx))
4440 		return -ENODEV;
4441 
4442 	rc = efx_mcdi_nvram_info(efx, type, &size, &erase_size, &protected);
4443 	if (rc)
4444 		return rc;
4445 	if (protected)
4446 		return -ENODEV; /* hide it */
4447 
4448 	part->nvram_type = type;
4449 
4450 	MCDI_SET_DWORD(inbuf, NVRAM_METADATA_IN_TYPE, type);
4451 	rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_METADATA, inbuf, sizeof(inbuf),
4452 			  outbuf, sizeof(outbuf), &outlen);
4453 	if (rc)
4454 		return rc;
4455 	if (outlen < MC_CMD_NVRAM_METADATA_OUT_LENMIN)
4456 		return -EIO;
4457 	if (MCDI_DWORD(outbuf, NVRAM_METADATA_OUT_FLAGS) &
4458 	    (1 << MC_CMD_NVRAM_METADATA_OUT_SUBTYPE_VALID_LBN))
4459 		part->fw_subtype = MCDI_DWORD(outbuf,
4460 					      NVRAM_METADATA_OUT_SUBTYPE);
4461 
4462 	part->common.dev_type_name = "EF10 NVRAM manager";
4463 	part->common.type_name = info->name;
4464 
4465 	part->common.mtd.type = MTD_NORFLASH;
4466 	part->common.mtd.flags = MTD_CAP_NORFLASH;
4467 	part->common.mtd.size = size;
4468 	part->common.mtd.erasesize = erase_size;
4469 
4470 	return 0;
4471 }
4472 
4473 static int efx_ef10_mtd_probe(struct efx_nic *efx)
4474 {
4475 	MCDI_DECLARE_BUF(outbuf, MC_CMD_NVRAM_PARTITIONS_OUT_LENMAX);
4476 	struct efx_mcdi_mtd_partition *parts;
4477 	size_t outlen, n_parts_total, i, n_parts;
4478 	unsigned int type;
4479 	int rc;
4480 
4481 	ASSERT_RTNL();
4482 
4483 	BUILD_BUG_ON(MC_CMD_NVRAM_PARTITIONS_IN_LEN != 0);
4484 	rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_PARTITIONS, NULL, 0,
4485 			  outbuf, sizeof(outbuf), &outlen);
4486 	if (rc)
4487 		return rc;
4488 	if (outlen < MC_CMD_NVRAM_PARTITIONS_OUT_LENMIN)
4489 		return -EIO;
4490 
4491 	n_parts_total = MCDI_DWORD(outbuf, NVRAM_PARTITIONS_OUT_NUM_PARTITIONS);
4492 	if (n_parts_total >
4493 	    MCDI_VAR_ARRAY_LEN(outlen, NVRAM_PARTITIONS_OUT_TYPE_ID))
4494 		return -EIO;
4495 
4496 	parts = kcalloc(n_parts_total, sizeof(*parts), GFP_KERNEL);
4497 	if (!parts)
4498 		return -ENOMEM;
4499 
4500 	n_parts = 0;
4501 	for (i = 0; i < n_parts_total; i++) {
4502 		type = MCDI_ARRAY_DWORD(outbuf, NVRAM_PARTITIONS_OUT_TYPE_ID,
4503 					i);
4504 		rc = efx_ef10_mtd_probe_partition(efx, &parts[n_parts], type);
4505 		if (rc == 0)
4506 			n_parts++;
4507 		else if (rc != -ENODEV)
4508 			goto fail;
4509 	}
4510 
4511 	rc = efx_mtd_add(efx, &parts[0].common, n_parts, sizeof(*parts));
4512 fail:
4513 	if (rc)
4514 		kfree(parts);
4515 	return rc;
4516 }
4517 
4518 #endif /* CONFIG_SFC_MTD */
4519 
4520 static void efx_ef10_ptp_write_host_time(struct efx_nic *efx, u32 host_time)
4521 {
4522 	_efx_writed(efx, cpu_to_le32(host_time), ER_DZ_MC_DB_LWRD);
4523 }
4524 
4525 static void efx_ef10_ptp_write_host_time_vf(struct efx_nic *efx,
4526 					    u32 host_time) {}
4527 
4528 static int efx_ef10_rx_enable_timestamping(struct efx_channel *channel,
4529 					   bool temp)
4530 {
4531 	MCDI_DECLARE_BUF(inbuf, MC_CMD_PTP_IN_TIME_EVENT_SUBSCRIBE_LEN);
4532 	int rc;
4533 
4534 	if (channel->sync_events_state == SYNC_EVENTS_REQUESTED ||
4535 	    channel->sync_events_state == SYNC_EVENTS_VALID ||
4536 	    (temp && channel->sync_events_state == SYNC_EVENTS_DISABLED))
4537 		return 0;
4538 	channel->sync_events_state = SYNC_EVENTS_REQUESTED;
4539 
4540 	MCDI_SET_DWORD(inbuf, PTP_IN_OP, MC_CMD_PTP_OP_TIME_EVENT_SUBSCRIBE);
4541 	MCDI_SET_DWORD(inbuf, PTP_IN_PERIPH_ID, 0);
4542 	MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_SUBSCRIBE_QUEUE,
4543 		       channel->channel);
4544 
4545 	rc = efx_mcdi_rpc(channel->efx, MC_CMD_PTP,
4546 			  inbuf, sizeof(inbuf), NULL, 0, NULL);
4547 
4548 	if (rc != 0)
4549 		channel->sync_events_state = temp ? SYNC_EVENTS_QUIESCENT :
4550 						    SYNC_EVENTS_DISABLED;
4551 
4552 	return rc;
4553 }
4554 
4555 static int efx_ef10_rx_disable_timestamping(struct efx_channel *channel,
4556 					    bool temp)
4557 {
4558 	MCDI_DECLARE_BUF(inbuf, MC_CMD_PTP_IN_TIME_EVENT_UNSUBSCRIBE_LEN);
4559 	int rc;
4560 
4561 	if (channel->sync_events_state == SYNC_EVENTS_DISABLED ||
4562 	    (temp && channel->sync_events_state == SYNC_EVENTS_QUIESCENT))
4563 		return 0;
4564 	if (channel->sync_events_state == SYNC_EVENTS_QUIESCENT) {
4565 		channel->sync_events_state = SYNC_EVENTS_DISABLED;
4566 		return 0;
4567 	}
4568 	channel->sync_events_state = temp ? SYNC_EVENTS_QUIESCENT :
4569 					    SYNC_EVENTS_DISABLED;
4570 
4571 	MCDI_SET_DWORD(inbuf, PTP_IN_OP, MC_CMD_PTP_OP_TIME_EVENT_UNSUBSCRIBE);
4572 	MCDI_SET_DWORD(inbuf, PTP_IN_PERIPH_ID, 0);
4573 	MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_UNSUBSCRIBE_CONTROL,
4574 		       MC_CMD_PTP_IN_TIME_EVENT_UNSUBSCRIBE_SINGLE);
4575 	MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_UNSUBSCRIBE_QUEUE,
4576 		       channel->channel);
4577 
4578 	rc = efx_mcdi_rpc(channel->efx, MC_CMD_PTP,
4579 			  inbuf, sizeof(inbuf), NULL, 0, NULL);
4580 
4581 	return rc;
4582 }
4583 
4584 static int efx_ef10_ptp_set_ts_sync_events(struct efx_nic *efx, bool en,
4585 					   bool temp)
4586 {
4587 	int (*set)(struct efx_channel *channel, bool temp);
4588 	struct efx_channel *channel;
4589 
4590 	set = en ?
4591 	      efx_ef10_rx_enable_timestamping :
4592 	      efx_ef10_rx_disable_timestamping;
4593 
4594 	efx_for_each_channel(channel, efx) {
4595 		int rc = set(channel, temp);
4596 		if (en && rc != 0) {
4597 			efx_ef10_ptp_set_ts_sync_events(efx, false, temp);
4598 			return rc;
4599 		}
4600 	}
4601 
4602 	return 0;
4603 }
4604 
4605 static int efx_ef10_ptp_set_ts_config_vf(struct efx_nic *efx,
4606 					 struct hwtstamp_config *init)
4607 {
4608 	return -EOPNOTSUPP;
4609 }
4610 
4611 static int efx_ef10_ptp_set_ts_config(struct efx_nic *efx,
4612 				      struct hwtstamp_config *init)
4613 {
4614 	int rc;
4615 
4616 	switch (init->rx_filter) {
4617 	case HWTSTAMP_FILTER_NONE:
4618 		efx_ef10_ptp_set_ts_sync_events(efx, false, false);
4619 		/* if TX timestamping is still requested then leave PTP on */
4620 		return efx_ptp_change_mode(efx,
4621 					   init->tx_type != HWTSTAMP_TX_OFF, 0);
4622 	case HWTSTAMP_FILTER_ALL:
4623 	case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
4624 	case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
4625 	case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
4626 	case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
4627 	case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
4628 	case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
4629 	case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
4630 	case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
4631 	case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
4632 	case HWTSTAMP_FILTER_PTP_V2_EVENT:
4633 	case HWTSTAMP_FILTER_PTP_V2_SYNC:
4634 	case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
4635 		init->rx_filter = HWTSTAMP_FILTER_ALL;
4636 		rc = efx_ptp_change_mode(efx, true, 0);
4637 		if (!rc)
4638 			rc = efx_ef10_ptp_set_ts_sync_events(efx, true, false);
4639 		if (rc)
4640 			efx_ptp_change_mode(efx, false, 0);
4641 		return rc;
4642 	default:
4643 		return -ERANGE;
4644 	}
4645 }
4646 
4647 const struct efx_nic_type efx_hunt_a0_vf_nic_type = {
4648 	.is_vf = true,
4649 	.mem_bar = EFX_MEM_VF_BAR,
4650 	.mem_map_size = efx_ef10_mem_map_size,
4651 	.probe = efx_ef10_probe_vf,
4652 	.remove = efx_ef10_remove,
4653 	.dimension_resources = efx_ef10_dimension_resources,
4654 	.init = efx_ef10_init_nic,
4655 	.fini = efx_port_dummy_op_void,
4656 	.map_reset_reason = efx_ef10_map_reset_reason,
4657 	.map_reset_flags = efx_ef10_map_reset_flags,
4658 	.reset = efx_ef10_reset,
4659 	.probe_port = efx_mcdi_port_probe,
4660 	.remove_port = efx_mcdi_port_remove,
4661 	.fini_dmaq = efx_ef10_fini_dmaq,
4662 	.prepare_flr = efx_ef10_prepare_flr,
4663 	.finish_flr = efx_port_dummy_op_void,
4664 	.describe_stats = efx_ef10_describe_stats,
4665 	.update_stats = efx_ef10_update_stats_vf,
4666 	.start_stats = efx_port_dummy_op_void,
4667 	.pull_stats = efx_port_dummy_op_void,
4668 	.stop_stats = efx_port_dummy_op_void,
4669 	.set_id_led = efx_mcdi_set_id_led,
4670 	.push_irq_moderation = efx_ef10_push_irq_moderation,
4671 	.reconfigure_mac = efx_ef10_mac_reconfigure_vf,
4672 	.check_mac_fault = efx_mcdi_mac_check_fault,
4673 	.reconfigure_port = efx_mcdi_port_reconfigure,
4674 	.get_wol = efx_ef10_get_wol_vf,
4675 	.set_wol = efx_ef10_set_wol_vf,
4676 	.resume_wol = efx_port_dummy_op_void,
4677 	.mcdi_request = efx_ef10_mcdi_request,
4678 	.mcdi_poll_response = efx_ef10_mcdi_poll_response,
4679 	.mcdi_read_response = efx_ef10_mcdi_read_response,
4680 	.mcdi_poll_reboot = efx_ef10_mcdi_poll_reboot,
4681 	.mcdi_reboot_detected = efx_ef10_mcdi_reboot_detected,
4682 	.irq_enable_master = efx_port_dummy_op_void,
4683 	.irq_test_generate = efx_ef10_irq_test_generate,
4684 	.irq_disable_non_ev = efx_port_dummy_op_void,
4685 	.irq_handle_msi = efx_ef10_msi_interrupt,
4686 	.irq_handle_legacy = efx_ef10_legacy_interrupt,
4687 	.tx_probe = efx_ef10_tx_probe,
4688 	.tx_init = efx_ef10_tx_init,
4689 	.tx_remove = efx_ef10_tx_remove,
4690 	.tx_write = efx_ef10_tx_write,
4691 	.rx_push_rss_config = efx_ef10_vf_rx_push_rss_config,
4692 	.rx_probe = efx_ef10_rx_probe,
4693 	.rx_init = efx_ef10_rx_init,
4694 	.rx_remove = efx_ef10_rx_remove,
4695 	.rx_write = efx_ef10_rx_write,
4696 	.rx_defer_refill = efx_ef10_rx_defer_refill,
4697 	.ev_probe = efx_ef10_ev_probe,
4698 	.ev_init = efx_ef10_ev_init,
4699 	.ev_fini = efx_ef10_ev_fini,
4700 	.ev_remove = efx_ef10_ev_remove,
4701 	.ev_process = efx_ef10_ev_process,
4702 	.ev_read_ack = efx_ef10_ev_read_ack,
4703 	.ev_test_generate = efx_ef10_ev_test_generate,
4704 	.filter_table_probe = efx_ef10_filter_table_probe,
4705 	.filter_table_restore = efx_ef10_filter_table_restore,
4706 	.filter_table_remove = efx_ef10_filter_table_remove,
4707 	.filter_update_rx_scatter = efx_ef10_filter_update_rx_scatter,
4708 	.filter_insert = efx_ef10_filter_insert,
4709 	.filter_remove_safe = efx_ef10_filter_remove_safe,
4710 	.filter_get_safe = efx_ef10_filter_get_safe,
4711 	.filter_clear_rx = efx_ef10_filter_clear_rx,
4712 	.filter_count_rx_used = efx_ef10_filter_count_rx_used,
4713 	.filter_get_rx_id_limit = efx_ef10_filter_get_rx_id_limit,
4714 	.filter_get_rx_ids = efx_ef10_filter_get_rx_ids,
4715 #ifdef CONFIG_RFS_ACCEL
4716 	.filter_rfs_insert = efx_ef10_filter_rfs_insert,
4717 	.filter_rfs_expire_one = efx_ef10_filter_rfs_expire_one,
4718 #endif
4719 #ifdef CONFIG_SFC_MTD
4720 	.mtd_probe = efx_port_dummy_op_int,
4721 #endif
4722 	.ptp_write_host_time = efx_ef10_ptp_write_host_time_vf,
4723 	.ptp_set_ts_config = efx_ef10_ptp_set_ts_config_vf,
4724 #ifdef CONFIG_SFC_SRIOV
4725 	.vswitching_probe = efx_ef10_vswitching_probe_vf,
4726 	.vswitching_restore = efx_ef10_vswitching_restore_vf,
4727 	.vswitching_remove = efx_ef10_vswitching_remove_vf,
4728 	.sriov_get_phys_port_id = efx_ef10_sriov_get_phys_port_id,
4729 #endif
4730 	.get_mac_address = efx_ef10_get_mac_address_vf,
4731 	.set_mac_address = efx_ef10_set_mac_address,
4732 
4733 	.revision = EFX_REV_HUNT_A0,
4734 	.max_dma_mask = DMA_BIT_MASK(ESF_DZ_TX_KER_BUF_ADDR_WIDTH),
4735 	.rx_prefix_size = ES_DZ_RX_PREFIX_SIZE,
4736 	.rx_hash_offset = ES_DZ_RX_PREFIX_HASH_OFST,
4737 	.rx_ts_offset = ES_DZ_RX_PREFIX_TSTAMP_OFST,
4738 	.can_rx_scatter = true,
4739 	.always_rx_scatter = true,
4740 	.max_interrupt_mode = EFX_INT_MODE_MSIX,
4741 	.timer_period_max = 1 << ERF_DD_EVQ_IND_TIMER_VAL_WIDTH,
4742 	.offload_features = (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
4743 			     NETIF_F_RXHASH | NETIF_F_NTUPLE),
4744 	.mcdi_max_ver = 2,
4745 	.max_rx_ip_filters = HUNT_FILTER_TBL_ROWS,
4746 	.hwtstamp_filters = 1 << HWTSTAMP_FILTER_NONE |
4747 			    1 << HWTSTAMP_FILTER_ALL,
4748 };
4749 
4750 const struct efx_nic_type efx_hunt_a0_nic_type = {
4751 	.is_vf = false,
4752 	.mem_bar = EFX_MEM_BAR,
4753 	.mem_map_size = efx_ef10_mem_map_size,
4754 	.probe = efx_ef10_probe_pf,
4755 	.remove = efx_ef10_remove,
4756 	.dimension_resources = efx_ef10_dimension_resources,
4757 	.init = efx_ef10_init_nic,
4758 	.fini = efx_port_dummy_op_void,
4759 	.map_reset_reason = efx_ef10_map_reset_reason,
4760 	.map_reset_flags = efx_ef10_map_reset_flags,
4761 	.reset = efx_ef10_reset,
4762 	.probe_port = efx_mcdi_port_probe,
4763 	.remove_port = efx_mcdi_port_remove,
4764 	.fini_dmaq = efx_ef10_fini_dmaq,
4765 	.prepare_flr = efx_ef10_prepare_flr,
4766 	.finish_flr = efx_port_dummy_op_void,
4767 	.describe_stats = efx_ef10_describe_stats,
4768 	.update_stats = efx_ef10_update_stats_pf,
4769 	.start_stats = efx_mcdi_mac_start_stats,
4770 	.pull_stats = efx_mcdi_mac_pull_stats,
4771 	.stop_stats = efx_mcdi_mac_stop_stats,
4772 	.set_id_led = efx_mcdi_set_id_led,
4773 	.push_irq_moderation = efx_ef10_push_irq_moderation,
4774 	.reconfigure_mac = efx_ef10_mac_reconfigure,
4775 	.check_mac_fault = efx_mcdi_mac_check_fault,
4776 	.reconfigure_port = efx_mcdi_port_reconfigure,
4777 	.get_wol = efx_ef10_get_wol,
4778 	.set_wol = efx_ef10_set_wol,
4779 	.resume_wol = efx_port_dummy_op_void,
4780 	.test_chip = efx_ef10_test_chip,
4781 	.test_nvram = efx_mcdi_nvram_test_all,
4782 	.mcdi_request = efx_ef10_mcdi_request,
4783 	.mcdi_poll_response = efx_ef10_mcdi_poll_response,
4784 	.mcdi_read_response = efx_ef10_mcdi_read_response,
4785 	.mcdi_poll_reboot = efx_ef10_mcdi_poll_reboot,
4786 	.mcdi_reboot_detected = efx_ef10_mcdi_reboot_detected,
4787 	.irq_enable_master = efx_port_dummy_op_void,
4788 	.irq_test_generate = efx_ef10_irq_test_generate,
4789 	.irq_disable_non_ev = efx_port_dummy_op_void,
4790 	.irq_handle_msi = efx_ef10_msi_interrupt,
4791 	.irq_handle_legacy = efx_ef10_legacy_interrupt,
4792 	.tx_probe = efx_ef10_tx_probe,
4793 	.tx_init = efx_ef10_tx_init,
4794 	.tx_remove = efx_ef10_tx_remove,
4795 	.tx_write = efx_ef10_tx_write,
4796 	.rx_push_rss_config = efx_ef10_pf_rx_push_rss_config,
4797 	.rx_probe = efx_ef10_rx_probe,
4798 	.rx_init = efx_ef10_rx_init,
4799 	.rx_remove = efx_ef10_rx_remove,
4800 	.rx_write = efx_ef10_rx_write,
4801 	.rx_defer_refill = efx_ef10_rx_defer_refill,
4802 	.ev_probe = efx_ef10_ev_probe,
4803 	.ev_init = efx_ef10_ev_init,
4804 	.ev_fini = efx_ef10_ev_fini,
4805 	.ev_remove = efx_ef10_ev_remove,
4806 	.ev_process = efx_ef10_ev_process,
4807 	.ev_read_ack = efx_ef10_ev_read_ack,
4808 	.ev_test_generate = efx_ef10_ev_test_generate,
4809 	.filter_table_probe = efx_ef10_filter_table_probe,
4810 	.filter_table_restore = efx_ef10_filter_table_restore,
4811 	.filter_table_remove = efx_ef10_filter_table_remove,
4812 	.filter_update_rx_scatter = efx_ef10_filter_update_rx_scatter,
4813 	.filter_insert = efx_ef10_filter_insert,
4814 	.filter_remove_safe = efx_ef10_filter_remove_safe,
4815 	.filter_get_safe = efx_ef10_filter_get_safe,
4816 	.filter_clear_rx = efx_ef10_filter_clear_rx,
4817 	.filter_count_rx_used = efx_ef10_filter_count_rx_used,
4818 	.filter_get_rx_id_limit = efx_ef10_filter_get_rx_id_limit,
4819 	.filter_get_rx_ids = efx_ef10_filter_get_rx_ids,
4820 #ifdef CONFIG_RFS_ACCEL
4821 	.filter_rfs_insert = efx_ef10_filter_rfs_insert,
4822 	.filter_rfs_expire_one = efx_ef10_filter_rfs_expire_one,
4823 #endif
4824 #ifdef CONFIG_SFC_MTD
4825 	.mtd_probe = efx_ef10_mtd_probe,
4826 	.mtd_rename = efx_mcdi_mtd_rename,
4827 	.mtd_read = efx_mcdi_mtd_read,
4828 	.mtd_erase = efx_mcdi_mtd_erase,
4829 	.mtd_write = efx_mcdi_mtd_write,
4830 	.mtd_sync = efx_mcdi_mtd_sync,
4831 #endif
4832 	.ptp_write_host_time = efx_ef10_ptp_write_host_time,
4833 	.ptp_set_ts_sync_events = efx_ef10_ptp_set_ts_sync_events,
4834 	.ptp_set_ts_config = efx_ef10_ptp_set_ts_config,
4835 #ifdef CONFIG_SFC_SRIOV
4836 	.sriov_configure = efx_ef10_sriov_configure,
4837 	.sriov_init = efx_ef10_sriov_init,
4838 	.sriov_fini = efx_ef10_sriov_fini,
4839 	.sriov_wanted = efx_ef10_sriov_wanted,
4840 	.sriov_reset = efx_ef10_sriov_reset,
4841 	.sriov_flr = efx_ef10_sriov_flr,
4842 	.sriov_set_vf_mac = efx_ef10_sriov_set_vf_mac,
4843 	.sriov_set_vf_vlan = efx_ef10_sriov_set_vf_vlan,
4844 	.sriov_set_vf_spoofchk = efx_ef10_sriov_set_vf_spoofchk,
4845 	.sriov_get_vf_config = efx_ef10_sriov_get_vf_config,
4846 	.sriov_set_vf_link_state = efx_ef10_sriov_set_vf_link_state,
4847 	.vswitching_probe = efx_ef10_vswitching_probe_pf,
4848 	.vswitching_restore = efx_ef10_vswitching_restore_pf,
4849 	.vswitching_remove = efx_ef10_vswitching_remove_pf,
4850 #endif
4851 	.get_mac_address = efx_ef10_get_mac_address_pf,
4852 	.set_mac_address = efx_ef10_set_mac_address,
4853 
4854 	.revision = EFX_REV_HUNT_A0,
4855 	.max_dma_mask = DMA_BIT_MASK(ESF_DZ_TX_KER_BUF_ADDR_WIDTH),
4856 	.rx_prefix_size = ES_DZ_RX_PREFIX_SIZE,
4857 	.rx_hash_offset = ES_DZ_RX_PREFIX_HASH_OFST,
4858 	.rx_ts_offset = ES_DZ_RX_PREFIX_TSTAMP_OFST,
4859 	.can_rx_scatter = true,
4860 	.always_rx_scatter = true,
4861 	.max_interrupt_mode = EFX_INT_MODE_MSIX,
4862 	.timer_period_max = 1 << ERF_DD_EVQ_IND_TIMER_VAL_WIDTH,
4863 	.offload_features = (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
4864 			     NETIF_F_RXHASH | NETIF_F_NTUPLE),
4865 	.mcdi_max_ver = 2,
4866 	.max_rx_ip_filters = HUNT_FILTER_TBL_ROWS,
4867 	.hwtstamp_filters = 1 << HWTSTAMP_FILTER_NONE |
4868 			    1 << HWTSTAMP_FILTER_ALL,
4869 };
4870