xref: /openbmc/linux/drivers/net/ethernet/sfc/ef10.c (revision 77a87824)
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 
54 #define EFX_EF10_FILTER_DEV_UC_MAX	32
55 #define EFX_EF10_FILTER_DEV_MC_MAX	256
56 
57 /* VLAN list entry */
58 struct efx_ef10_vlan {
59 	struct list_head list;
60 	u16 vid;
61 };
62 
63 /* Per-VLAN filters information */
64 struct efx_ef10_filter_vlan {
65 	struct list_head list;
66 	u16 vid;
67 	u16 uc[EFX_EF10_FILTER_DEV_UC_MAX];
68 	u16 mc[EFX_EF10_FILTER_DEV_MC_MAX];
69 	u16 ucdef;
70 	u16 bcast;
71 	u16 mcdef;
72 };
73 
74 struct efx_ef10_dev_addr {
75 	u8 addr[ETH_ALEN];
76 };
77 
78 struct efx_ef10_filter_table {
79 /* The MCDI match masks supported by this fw & hw, in order of priority */
80 	u32 rx_match_mcdi_flags[
81 		MC_CMD_GET_PARSER_DISP_INFO_OUT_SUPPORTED_MATCHES_MAXNUM];
82 	unsigned int rx_match_count;
83 
84 	struct {
85 		unsigned long spec;	/* pointer to spec plus flag bits */
86 /* BUSY flag indicates that an update is in progress.  AUTO_OLD is
87  * used to mark and sweep MAC filters for the device address lists.
88  */
89 #define EFX_EF10_FILTER_FLAG_BUSY	1UL
90 #define EFX_EF10_FILTER_FLAG_AUTO_OLD	2UL
91 #define EFX_EF10_FILTER_FLAGS		3UL
92 		u64 handle;		/* firmware handle */
93 	} *entry;
94 	wait_queue_head_t waitq;
95 /* Shadow of net_device address lists, guarded by mac_lock */
96 	struct efx_ef10_dev_addr dev_uc_list[EFX_EF10_FILTER_DEV_UC_MAX];
97 	struct efx_ef10_dev_addr dev_mc_list[EFX_EF10_FILTER_DEV_MC_MAX];
98 	int dev_uc_count;
99 	int dev_mc_count;
100 	bool uc_promisc;
101 	bool mc_promisc;
102 /* Whether in multicast promiscuous mode when last changed */
103 	bool mc_promisc_last;
104 	bool vlan_filter;
105 	struct list_head vlan_list;
106 };
107 
108 /* An arbitrary search limit for the software hash table */
109 #define EFX_EF10_FILTER_SEARCH_LIMIT 200
110 
111 static void efx_ef10_rx_free_indir_table(struct efx_nic *efx);
112 static void efx_ef10_filter_table_remove(struct efx_nic *efx);
113 static int efx_ef10_filter_add_vlan(struct efx_nic *efx, u16 vid);
114 static void efx_ef10_filter_del_vlan_internal(struct efx_nic *efx,
115 					      struct efx_ef10_filter_vlan *vlan);
116 static void efx_ef10_filter_del_vlan(struct efx_nic *efx, u16 vid);
117 
118 static int efx_ef10_get_warm_boot_count(struct efx_nic *efx)
119 {
120 	efx_dword_t reg;
121 
122 	efx_readd(efx, &reg, ER_DZ_BIU_MC_SFT_STATUS);
123 	return EFX_DWORD_FIELD(reg, EFX_WORD_1) == 0xb007 ?
124 		EFX_DWORD_FIELD(reg, EFX_WORD_0) : -EIO;
125 }
126 
127 static unsigned int efx_ef10_mem_map_size(struct efx_nic *efx)
128 {
129 	int bar;
130 
131 	bar = efx->type->mem_bar;
132 	return resource_size(&efx->pci_dev->resource[bar]);
133 }
134 
135 static bool efx_ef10_is_vf(struct efx_nic *efx)
136 {
137 	return efx->type->is_vf;
138 }
139 
140 static int efx_ef10_get_pf_index(struct efx_nic *efx)
141 {
142 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_FUNCTION_INFO_OUT_LEN);
143 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
144 	size_t outlen;
145 	int rc;
146 
147 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_FUNCTION_INFO, NULL, 0, outbuf,
148 			  sizeof(outbuf), &outlen);
149 	if (rc)
150 		return rc;
151 	if (outlen < sizeof(outbuf))
152 		return -EIO;
153 
154 	nic_data->pf_index = MCDI_DWORD(outbuf, GET_FUNCTION_INFO_OUT_PF);
155 	return 0;
156 }
157 
158 #ifdef CONFIG_SFC_SRIOV
159 static int efx_ef10_get_vf_index(struct efx_nic *efx)
160 {
161 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_FUNCTION_INFO_OUT_LEN);
162 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
163 	size_t outlen;
164 	int rc;
165 
166 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_FUNCTION_INFO, NULL, 0, outbuf,
167 			  sizeof(outbuf), &outlen);
168 	if (rc)
169 		return rc;
170 	if (outlen < sizeof(outbuf))
171 		return -EIO;
172 
173 	nic_data->vf_index = MCDI_DWORD(outbuf, GET_FUNCTION_INFO_OUT_VF);
174 	return 0;
175 }
176 #endif
177 
178 static int efx_ef10_init_datapath_caps(struct efx_nic *efx)
179 {
180 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_CAPABILITIES_OUT_LEN);
181 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
182 	size_t outlen;
183 	int rc;
184 
185 	BUILD_BUG_ON(MC_CMD_GET_CAPABILITIES_IN_LEN != 0);
186 
187 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_CAPABILITIES, NULL, 0,
188 			  outbuf, sizeof(outbuf), &outlen);
189 	if (rc)
190 		return rc;
191 	if (outlen < sizeof(outbuf)) {
192 		netif_err(efx, drv, efx->net_dev,
193 			  "unable to read datapath firmware capabilities\n");
194 		return -EIO;
195 	}
196 
197 	nic_data->datapath_caps =
198 		MCDI_DWORD(outbuf, GET_CAPABILITIES_OUT_FLAGS1);
199 
200 	/* record the DPCPU firmware IDs to determine VEB vswitching support.
201 	 */
202 	nic_data->rx_dpcpu_fw_id =
203 		MCDI_WORD(outbuf, GET_CAPABILITIES_OUT_RX_DPCPU_FW_ID);
204 	nic_data->tx_dpcpu_fw_id =
205 		MCDI_WORD(outbuf, GET_CAPABILITIES_OUT_TX_DPCPU_FW_ID);
206 
207 	if (!(nic_data->datapath_caps &
208 	      (1 << MC_CMD_GET_CAPABILITIES_OUT_RX_PREFIX_LEN_14_LBN))) {
209 		netif_err(efx, probe, efx->net_dev,
210 			  "current firmware does not support an RX prefix\n");
211 		return -ENODEV;
212 	}
213 
214 	return 0;
215 }
216 
217 static int efx_ef10_get_sysclk_freq(struct efx_nic *efx)
218 {
219 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_CLOCK_OUT_LEN);
220 	int rc;
221 
222 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_CLOCK, NULL, 0,
223 			  outbuf, sizeof(outbuf), NULL);
224 	if (rc)
225 		return rc;
226 	rc = MCDI_DWORD(outbuf, GET_CLOCK_OUT_SYS_FREQ);
227 	return rc > 0 ? rc : -ERANGE;
228 }
229 
230 static int efx_ef10_get_mac_address_pf(struct efx_nic *efx, u8 *mac_address)
231 {
232 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_MAC_ADDRESSES_OUT_LEN);
233 	size_t outlen;
234 	int rc;
235 
236 	BUILD_BUG_ON(MC_CMD_GET_MAC_ADDRESSES_IN_LEN != 0);
237 
238 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_MAC_ADDRESSES, NULL, 0,
239 			  outbuf, sizeof(outbuf), &outlen);
240 	if (rc)
241 		return rc;
242 	if (outlen < MC_CMD_GET_MAC_ADDRESSES_OUT_LEN)
243 		return -EIO;
244 
245 	ether_addr_copy(mac_address,
246 			MCDI_PTR(outbuf, GET_MAC_ADDRESSES_OUT_MAC_ADDR_BASE));
247 	return 0;
248 }
249 
250 static int efx_ef10_get_mac_address_vf(struct efx_nic *efx, u8 *mac_address)
251 {
252 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VPORT_GET_MAC_ADDRESSES_IN_LEN);
253 	MCDI_DECLARE_BUF(outbuf, MC_CMD_VPORT_GET_MAC_ADDRESSES_OUT_LENMAX);
254 	size_t outlen;
255 	int num_addrs, rc;
256 
257 	MCDI_SET_DWORD(inbuf, VPORT_GET_MAC_ADDRESSES_IN_VPORT_ID,
258 		       EVB_PORT_ID_ASSIGNED);
259 	rc = efx_mcdi_rpc(efx, MC_CMD_VPORT_GET_MAC_ADDRESSES, inbuf,
260 			  sizeof(inbuf), outbuf, sizeof(outbuf), &outlen);
261 
262 	if (rc)
263 		return rc;
264 	if (outlen < MC_CMD_VPORT_GET_MAC_ADDRESSES_OUT_LENMIN)
265 		return -EIO;
266 
267 	num_addrs = MCDI_DWORD(outbuf,
268 			       VPORT_GET_MAC_ADDRESSES_OUT_MACADDR_COUNT);
269 
270 	WARN_ON(num_addrs != 1);
271 
272 	ether_addr_copy(mac_address,
273 			MCDI_PTR(outbuf, VPORT_GET_MAC_ADDRESSES_OUT_MACADDR));
274 
275 	return 0;
276 }
277 
278 static ssize_t efx_ef10_show_link_control_flag(struct device *dev,
279 					       struct device_attribute *attr,
280 					       char *buf)
281 {
282 	struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));
283 
284 	return sprintf(buf, "%d\n",
285 		       ((efx->mcdi->fn_flags) &
286 			(1 << MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_LINKCTRL))
287 		       ? 1 : 0);
288 }
289 
290 static ssize_t efx_ef10_show_primary_flag(struct device *dev,
291 					  struct device_attribute *attr,
292 					  char *buf)
293 {
294 	struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));
295 
296 	return sprintf(buf, "%d\n",
297 		       ((efx->mcdi->fn_flags) &
298 			(1 << MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_PRIMARY))
299 		       ? 1 : 0);
300 }
301 
302 static struct efx_ef10_vlan *efx_ef10_find_vlan(struct efx_nic *efx, u16 vid)
303 {
304 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
305 	struct efx_ef10_vlan *vlan;
306 
307 	WARN_ON(!mutex_is_locked(&nic_data->vlan_lock));
308 
309 	list_for_each_entry(vlan, &nic_data->vlan_list, list) {
310 		if (vlan->vid == vid)
311 			return vlan;
312 	}
313 
314 	return NULL;
315 }
316 
317 static int efx_ef10_add_vlan(struct efx_nic *efx, u16 vid)
318 {
319 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
320 	struct efx_ef10_vlan *vlan;
321 	int rc;
322 
323 	mutex_lock(&nic_data->vlan_lock);
324 
325 	vlan = efx_ef10_find_vlan(efx, vid);
326 	if (vlan) {
327 		/* We add VID 0 on init. 8021q adds it on module init
328 		 * for all interfaces with VLAN filtring feature.
329 		 */
330 		if (vid == 0)
331 			goto done_unlock;
332 		netif_warn(efx, drv, efx->net_dev,
333 			   "VLAN %u already added\n", vid);
334 		rc = -EALREADY;
335 		goto fail_exist;
336 	}
337 
338 	rc = -ENOMEM;
339 	vlan = kzalloc(sizeof(*vlan), GFP_KERNEL);
340 	if (!vlan)
341 		goto fail_alloc;
342 
343 	vlan->vid = vid;
344 
345 	list_add_tail(&vlan->list, &nic_data->vlan_list);
346 
347 	if (efx->filter_state) {
348 		mutex_lock(&efx->mac_lock);
349 		down_write(&efx->filter_sem);
350 		rc = efx_ef10_filter_add_vlan(efx, vlan->vid);
351 		up_write(&efx->filter_sem);
352 		mutex_unlock(&efx->mac_lock);
353 		if (rc)
354 			goto fail_filter_add_vlan;
355 	}
356 
357 done_unlock:
358 	mutex_unlock(&nic_data->vlan_lock);
359 	return 0;
360 
361 fail_filter_add_vlan:
362 	list_del(&vlan->list);
363 	kfree(vlan);
364 fail_alloc:
365 fail_exist:
366 	mutex_unlock(&nic_data->vlan_lock);
367 	return rc;
368 }
369 
370 static void efx_ef10_del_vlan_internal(struct efx_nic *efx,
371 				       struct efx_ef10_vlan *vlan)
372 {
373 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
374 
375 	WARN_ON(!mutex_is_locked(&nic_data->vlan_lock));
376 
377 	if (efx->filter_state) {
378 		down_write(&efx->filter_sem);
379 		efx_ef10_filter_del_vlan(efx, vlan->vid);
380 		up_write(&efx->filter_sem);
381 	}
382 
383 	list_del(&vlan->list);
384 	kfree(vlan);
385 }
386 
387 static int efx_ef10_del_vlan(struct efx_nic *efx, u16 vid)
388 {
389 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
390 	struct efx_ef10_vlan *vlan;
391 	int rc = 0;
392 
393 	/* 8021q removes VID 0 on module unload for all interfaces
394 	 * with VLAN filtering feature. We need to keep it to receive
395 	 * untagged traffic.
396 	 */
397 	if (vid == 0)
398 		return 0;
399 
400 	mutex_lock(&nic_data->vlan_lock);
401 
402 	vlan = efx_ef10_find_vlan(efx, vid);
403 	if (!vlan) {
404 		netif_err(efx, drv, efx->net_dev,
405 			  "VLAN %u to be deleted not found\n", vid);
406 		rc = -ENOENT;
407 	} else {
408 		efx_ef10_del_vlan_internal(efx, vlan);
409 	}
410 
411 	mutex_unlock(&nic_data->vlan_lock);
412 
413 	return rc;
414 }
415 
416 static void efx_ef10_cleanup_vlans(struct efx_nic *efx)
417 {
418 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
419 	struct efx_ef10_vlan *vlan, *next_vlan;
420 
421 	mutex_lock(&nic_data->vlan_lock);
422 	list_for_each_entry_safe(vlan, next_vlan, &nic_data->vlan_list, list)
423 		efx_ef10_del_vlan_internal(efx, vlan);
424 	mutex_unlock(&nic_data->vlan_lock);
425 }
426 
427 static DEVICE_ATTR(link_control_flag, 0444, efx_ef10_show_link_control_flag,
428 		   NULL);
429 static DEVICE_ATTR(primary_flag, 0444, efx_ef10_show_primary_flag, NULL);
430 
431 static int efx_ef10_probe(struct efx_nic *efx)
432 {
433 	struct efx_ef10_nic_data *nic_data;
434 	struct net_device *net_dev = efx->net_dev;
435 	int i, rc;
436 
437 	/* We can have one VI for each 8K region.  However, until we
438 	 * use TX option descriptors we need two TX queues per channel.
439 	 */
440 	efx->max_channels = min_t(unsigned int,
441 				  EFX_MAX_CHANNELS,
442 				  efx_ef10_mem_map_size(efx) /
443 				  (EFX_VI_PAGE_SIZE * EFX_TXQ_TYPES));
444 	efx->max_tx_channels = efx->max_channels;
445 	if (WARN_ON(efx->max_channels == 0))
446 		return -EIO;
447 
448 	nic_data = kzalloc(sizeof(*nic_data), GFP_KERNEL);
449 	if (!nic_data)
450 		return -ENOMEM;
451 	efx->nic_data = nic_data;
452 
453 	/* we assume later that we can copy from this buffer in dwords */
454 	BUILD_BUG_ON(MCDI_CTL_SDU_LEN_MAX_V2 % 4);
455 
456 	rc = efx_nic_alloc_buffer(efx, &nic_data->mcdi_buf,
457 				  8 + MCDI_CTL_SDU_LEN_MAX_V2, GFP_KERNEL);
458 	if (rc)
459 		goto fail1;
460 
461 	/* Get the MC's warm boot count.  In case it's rebooting right
462 	 * now, be prepared to retry.
463 	 */
464 	i = 0;
465 	for (;;) {
466 		rc = efx_ef10_get_warm_boot_count(efx);
467 		if (rc >= 0)
468 			break;
469 		if (++i == 5)
470 			goto fail2;
471 		ssleep(1);
472 	}
473 	nic_data->warm_boot_count = rc;
474 
475 	nic_data->rx_rss_context = EFX_EF10_RSS_CONTEXT_INVALID;
476 
477 	nic_data->vport_id = EVB_PORT_ID_ASSIGNED;
478 
479 	/* In case we're recovering from a crash (kexec), we want to
480 	 * cancel any outstanding request by the previous user of this
481 	 * function.  We send a special message using the least
482 	 * significant bits of the 'high' (doorbell) register.
483 	 */
484 	_efx_writed(efx, cpu_to_le32(1), ER_DZ_MC_DB_HWRD);
485 
486 	rc = efx_mcdi_init(efx);
487 	if (rc)
488 		goto fail2;
489 
490 	/* Reset (most) configuration for this function */
491 	rc = efx_mcdi_reset(efx, RESET_TYPE_ALL);
492 	if (rc)
493 		goto fail3;
494 
495 	/* Enable event logging */
496 	rc = efx_mcdi_log_ctrl(efx, true, false, 0);
497 	if (rc)
498 		goto fail3;
499 
500 	rc = device_create_file(&efx->pci_dev->dev,
501 				&dev_attr_link_control_flag);
502 	if (rc)
503 		goto fail3;
504 
505 	rc = device_create_file(&efx->pci_dev->dev, &dev_attr_primary_flag);
506 	if (rc)
507 		goto fail4;
508 
509 	rc = efx_ef10_get_pf_index(efx);
510 	if (rc)
511 		goto fail5;
512 
513 	rc = efx_ef10_init_datapath_caps(efx);
514 	if (rc < 0)
515 		goto fail5;
516 
517 	efx->rx_packet_len_offset =
518 		ES_DZ_RX_PREFIX_PKTLEN_OFST - ES_DZ_RX_PREFIX_SIZE;
519 
520 	rc = efx_mcdi_port_get_number(efx);
521 	if (rc < 0)
522 		goto fail5;
523 	efx->port_num = rc;
524 	net_dev->dev_port = rc;
525 
526 	rc = efx->type->get_mac_address(efx, efx->net_dev->perm_addr);
527 	if (rc)
528 		goto fail5;
529 
530 	rc = efx_ef10_get_sysclk_freq(efx);
531 	if (rc < 0)
532 		goto fail5;
533 	efx->timer_quantum_ns = 1536000 / rc; /* 1536 cycles */
534 
535 	/* Check whether firmware supports bug 35388 workaround.
536 	 * First try to enable it, then if we get EPERM, just
537 	 * ask if it's already enabled
538 	 */
539 	rc = efx_mcdi_set_workaround(efx, MC_CMD_WORKAROUND_BUG35388, true, NULL);
540 	if (rc == 0) {
541 		nic_data->workaround_35388 = true;
542 	} else if (rc == -EPERM) {
543 		unsigned int enabled;
544 
545 		rc = efx_mcdi_get_workarounds(efx, NULL, &enabled);
546 		if (rc)
547 			goto fail3;
548 		nic_data->workaround_35388 = enabled &
549 			MC_CMD_GET_WORKAROUNDS_OUT_BUG35388;
550 	} else if (rc != -ENOSYS && rc != -ENOENT) {
551 		goto fail5;
552 	}
553 	netif_dbg(efx, probe, efx->net_dev,
554 		  "workaround for bug 35388 is %sabled\n",
555 		  nic_data->workaround_35388 ? "en" : "dis");
556 
557 	rc = efx_mcdi_mon_probe(efx);
558 	if (rc && rc != -EPERM)
559 		goto fail5;
560 
561 	efx_ptp_probe(efx, NULL);
562 
563 #ifdef CONFIG_SFC_SRIOV
564 	if ((efx->pci_dev->physfn) && (!efx->pci_dev->is_physfn)) {
565 		struct pci_dev *pci_dev_pf = efx->pci_dev->physfn;
566 		struct efx_nic *efx_pf = pci_get_drvdata(pci_dev_pf);
567 
568 		efx_pf->type->get_mac_address(efx_pf, nic_data->port_id);
569 	} else
570 #endif
571 		ether_addr_copy(nic_data->port_id, efx->net_dev->perm_addr);
572 
573 	INIT_LIST_HEAD(&nic_data->vlan_list);
574 	mutex_init(&nic_data->vlan_lock);
575 
576 	/* Add unspecified VID to support VLAN filtering being disabled */
577 	rc = efx_ef10_add_vlan(efx, EFX_FILTER_VID_UNSPEC);
578 	if (rc)
579 		goto fail_add_vid_unspec;
580 
581 	/* If VLAN filtering is enabled, we need VID 0 to get untagged
582 	 * traffic.  It is added automatically if 8021q module is loaded,
583 	 * but we can't rely on it since module may be not loaded.
584 	 */
585 	rc = efx_ef10_add_vlan(efx, 0);
586 	if (rc)
587 		goto fail_add_vid_0;
588 
589 	return 0;
590 
591 fail_add_vid_0:
592 	efx_ef10_cleanup_vlans(efx);
593 fail_add_vid_unspec:
594 	mutex_destroy(&nic_data->vlan_lock);
595 	efx_ptp_remove(efx);
596 	efx_mcdi_mon_remove(efx);
597 fail5:
598 	device_remove_file(&efx->pci_dev->dev, &dev_attr_primary_flag);
599 fail4:
600 	device_remove_file(&efx->pci_dev->dev, &dev_attr_link_control_flag);
601 fail3:
602 	efx_mcdi_fini(efx);
603 fail2:
604 	efx_nic_free_buffer(efx, &nic_data->mcdi_buf);
605 fail1:
606 	kfree(nic_data);
607 	efx->nic_data = NULL;
608 	return rc;
609 }
610 
611 static int efx_ef10_free_vis(struct efx_nic *efx)
612 {
613 	MCDI_DECLARE_BUF_ERR(outbuf);
614 	size_t outlen;
615 	int rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FREE_VIS, NULL, 0,
616 				    outbuf, sizeof(outbuf), &outlen);
617 
618 	/* -EALREADY means nothing to free, so ignore */
619 	if (rc == -EALREADY)
620 		rc = 0;
621 	if (rc)
622 		efx_mcdi_display_error(efx, MC_CMD_FREE_VIS, 0, outbuf, outlen,
623 				       rc);
624 	return rc;
625 }
626 
627 #ifdef EFX_USE_PIO
628 
629 static void efx_ef10_free_piobufs(struct efx_nic *efx)
630 {
631 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
632 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FREE_PIOBUF_IN_LEN);
633 	unsigned int i;
634 	int rc;
635 
636 	BUILD_BUG_ON(MC_CMD_FREE_PIOBUF_OUT_LEN != 0);
637 
638 	for (i = 0; i < nic_data->n_piobufs; i++) {
639 		MCDI_SET_DWORD(inbuf, FREE_PIOBUF_IN_PIOBUF_HANDLE,
640 			       nic_data->piobuf_handle[i]);
641 		rc = efx_mcdi_rpc(efx, MC_CMD_FREE_PIOBUF, inbuf, sizeof(inbuf),
642 				  NULL, 0, NULL);
643 		WARN_ON(rc);
644 	}
645 
646 	nic_data->n_piobufs = 0;
647 }
648 
649 static int efx_ef10_alloc_piobufs(struct efx_nic *efx, unsigned int n)
650 {
651 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
652 	MCDI_DECLARE_BUF(outbuf, MC_CMD_ALLOC_PIOBUF_OUT_LEN);
653 	unsigned int i;
654 	size_t outlen;
655 	int rc = 0;
656 
657 	BUILD_BUG_ON(MC_CMD_ALLOC_PIOBUF_IN_LEN != 0);
658 
659 	for (i = 0; i < n; i++) {
660 		rc = efx_mcdi_rpc_quiet(efx, MC_CMD_ALLOC_PIOBUF, NULL, 0,
661 					outbuf, sizeof(outbuf), &outlen);
662 		if (rc) {
663 			/* Don't display the MC error if we didn't have space
664 			 * for a VF.
665 			 */
666 			if (!(efx_ef10_is_vf(efx) && rc == -ENOSPC))
667 				efx_mcdi_display_error(efx, MC_CMD_ALLOC_PIOBUF,
668 						       0, outbuf, outlen, rc);
669 			break;
670 		}
671 		if (outlen < MC_CMD_ALLOC_PIOBUF_OUT_LEN) {
672 			rc = -EIO;
673 			break;
674 		}
675 		nic_data->piobuf_handle[i] =
676 			MCDI_DWORD(outbuf, ALLOC_PIOBUF_OUT_PIOBUF_HANDLE);
677 		netif_dbg(efx, probe, efx->net_dev,
678 			  "allocated PIO buffer %u handle %x\n", i,
679 			  nic_data->piobuf_handle[i]);
680 	}
681 
682 	nic_data->n_piobufs = i;
683 	if (rc)
684 		efx_ef10_free_piobufs(efx);
685 	return rc;
686 }
687 
688 static int efx_ef10_link_piobufs(struct efx_nic *efx)
689 {
690 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
691 	_MCDI_DECLARE_BUF(inbuf,
692 			  max(MC_CMD_LINK_PIOBUF_IN_LEN,
693 			      MC_CMD_UNLINK_PIOBUF_IN_LEN));
694 	struct efx_channel *channel;
695 	struct efx_tx_queue *tx_queue;
696 	unsigned int offset, index;
697 	int rc;
698 
699 	BUILD_BUG_ON(MC_CMD_LINK_PIOBUF_OUT_LEN != 0);
700 	BUILD_BUG_ON(MC_CMD_UNLINK_PIOBUF_OUT_LEN != 0);
701 
702 	memset(inbuf, 0, sizeof(inbuf));
703 
704 	/* Link a buffer to each VI in the write-combining mapping */
705 	for (index = 0; index < nic_data->n_piobufs; ++index) {
706 		MCDI_SET_DWORD(inbuf, LINK_PIOBUF_IN_PIOBUF_HANDLE,
707 			       nic_data->piobuf_handle[index]);
708 		MCDI_SET_DWORD(inbuf, LINK_PIOBUF_IN_TXQ_INSTANCE,
709 			       nic_data->pio_write_vi_base + index);
710 		rc = efx_mcdi_rpc(efx, MC_CMD_LINK_PIOBUF,
711 				  inbuf, MC_CMD_LINK_PIOBUF_IN_LEN,
712 				  NULL, 0, NULL);
713 		if (rc) {
714 			netif_err(efx, drv, efx->net_dev,
715 				  "failed to link VI %u to PIO buffer %u (%d)\n",
716 				  nic_data->pio_write_vi_base + index, index,
717 				  rc);
718 			goto fail;
719 		}
720 		netif_dbg(efx, probe, efx->net_dev,
721 			  "linked VI %u to PIO buffer %u\n",
722 			  nic_data->pio_write_vi_base + index, index);
723 	}
724 
725 	/* Link a buffer to each TX queue */
726 	efx_for_each_channel(channel, efx) {
727 		efx_for_each_channel_tx_queue(tx_queue, channel) {
728 			/* We assign the PIO buffers to queues in
729 			 * reverse order to allow for the following
730 			 * special case.
731 			 */
732 			offset = ((efx->tx_channel_offset + efx->n_tx_channels -
733 				   tx_queue->channel->channel - 1) *
734 				  efx_piobuf_size);
735 			index = offset / ER_DZ_TX_PIOBUF_SIZE;
736 			offset = offset % ER_DZ_TX_PIOBUF_SIZE;
737 
738 			/* When the host page size is 4K, the first
739 			 * host page in the WC mapping may be within
740 			 * the same VI page as the last TX queue.  We
741 			 * can only link one buffer to each VI.
742 			 */
743 			if (tx_queue->queue == nic_data->pio_write_vi_base) {
744 				BUG_ON(index != 0);
745 				rc = 0;
746 			} else {
747 				MCDI_SET_DWORD(inbuf,
748 					       LINK_PIOBUF_IN_PIOBUF_HANDLE,
749 					       nic_data->piobuf_handle[index]);
750 				MCDI_SET_DWORD(inbuf,
751 					       LINK_PIOBUF_IN_TXQ_INSTANCE,
752 					       tx_queue->queue);
753 				rc = efx_mcdi_rpc(efx, MC_CMD_LINK_PIOBUF,
754 						  inbuf, MC_CMD_LINK_PIOBUF_IN_LEN,
755 						  NULL, 0, NULL);
756 			}
757 
758 			if (rc) {
759 				/* This is non-fatal; the TX path just
760 				 * won't use PIO for this queue
761 				 */
762 				netif_err(efx, drv, efx->net_dev,
763 					  "failed to link VI %u to PIO buffer %u (%d)\n",
764 					  tx_queue->queue, index, rc);
765 				tx_queue->piobuf = NULL;
766 			} else {
767 				tx_queue->piobuf =
768 					nic_data->pio_write_base +
769 					index * EFX_VI_PAGE_SIZE + offset;
770 				tx_queue->piobuf_offset = offset;
771 				netif_dbg(efx, probe, efx->net_dev,
772 					  "linked VI %u to PIO buffer %u offset %x addr %p\n",
773 					  tx_queue->queue, index,
774 					  tx_queue->piobuf_offset,
775 					  tx_queue->piobuf);
776 			}
777 		}
778 	}
779 
780 	return 0;
781 
782 fail:
783 	while (index--) {
784 		MCDI_SET_DWORD(inbuf, UNLINK_PIOBUF_IN_TXQ_INSTANCE,
785 			       nic_data->pio_write_vi_base + index);
786 		efx_mcdi_rpc(efx, MC_CMD_UNLINK_PIOBUF,
787 			     inbuf, MC_CMD_UNLINK_PIOBUF_IN_LEN,
788 			     NULL, 0, NULL);
789 	}
790 	return rc;
791 }
792 
793 static void efx_ef10_forget_old_piobufs(struct efx_nic *efx)
794 {
795 	struct efx_channel *channel;
796 	struct efx_tx_queue *tx_queue;
797 
798 	/* All our existing PIO buffers went away */
799 	efx_for_each_channel(channel, efx)
800 		efx_for_each_channel_tx_queue(tx_queue, channel)
801 			tx_queue->piobuf = NULL;
802 }
803 
804 #else /* !EFX_USE_PIO */
805 
806 static int efx_ef10_alloc_piobufs(struct efx_nic *efx, unsigned int n)
807 {
808 	return n == 0 ? 0 : -ENOBUFS;
809 }
810 
811 static int efx_ef10_link_piobufs(struct efx_nic *efx)
812 {
813 	return 0;
814 }
815 
816 static void efx_ef10_free_piobufs(struct efx_nic *efx)
817 {
818 }
819 
820 static void efx_ef10_forget_old_piobufs(struct efx_nic *efx)
821 {
822 }
823 
824 #endif /* EFX_USE_PIO */
825 
826 static void efx_ef10_remove(struct efx_nic *efx)
827 {
828 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
829 	int rc;
830 
831 #ifdef CONFIG_SFC_SRIOV
832 	struct efx_ef10_nic_data *nic_data_pf;
833 	struct pci_dev *pci_dev_pf;
834 	struct efx_nic *efx_pf;
835 	struct ef10_vf *vf;
836 
837 	if (efx->pci_dev->is_virtfn) {
838 		pci_dev_pf = efx->pci_dev->physfn;
839 		if (pci_dev_pf) {
840 			efx_pf = pci_get_drvdata(pci_dev_pf);
841 			nic_data_pf = efx_pf->nic_data;
842 			vf = nic_data_pf->vf + nic_data->vf_index;
843 			vf->efx = NULL;
844 		} else
845 			netif_info(efx, drv, efx->net_dev,
846 				   "Could not get the PF id from VF\n");
847 	}
848 #endif
849 
850 	efx_ef10_cleanup_vlans(efx);
851 	mutex_destroy(&nic_data->vlan_lock);
852 
853 	efx_ptp_remove(efx);
854 
855 	efx_mcdi_mon_remove(efx);
856 
857 	efx_ef10_rx_free_indir_table(efx);
858 
859 	if (nic_data->wc_membase)
860 		iounmap(nic_data->wc_membase);
861 
862 	rc = efx_ef10_free_vis(efx);
863 	WARN_ON(rc != 0);
864 
865 	if (!nic_data->must_restore_piobufs)
866 		efx_ef10_free_piobufs(efx);
867 
868 	device_remove_file(&efx->pci_dev->dev, &dev_attr_primary_flag);
869 	device_remove_file(&efx->pci_dev->dev, &dev_attr_link_control_flag);
870 
871 	efx_mcdi_fini(efx);
872 	efx_nic_free_buffer(efx, &nic_data->mcdi_buf);
873 	kfree(nic_data);
874 }
875 
876 static int efx_ef10_probe_pf(struct efx_nic *efx)
877 {
878 	return efx_ef10_probe(efx);
879 }
880 
881 int efx_ef10_vadaptor_query(struct efx_nic *efx, unsigned int port_id,
882 			    u32 *port_flags, u32 *vadaptor_flags,
883 			    unsigned int *vlan_tags)
884 {
885 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
886 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VADAPTOR_QUERY_IN_LEN);
887 	MCDI_DECLARE_BUF(outbuf, MC_CMD_VADAPTOR_QUERY_OUT_LEN);
888 	size_t outlen;
889 	int rc;
890 
891 	if (nic_data->datapath_caps &
892 	    (1 << MC_CMD_GET_CAPABILITIES_OUT_VADAPTOR_QUERY_LBN)) {
893 		MCDI_SET_DWORD(inbuf, VADAPTOR_QUERY_IN_UPSTREAM_PORT_ID,
894 			       port_id);
895 
896 		rc = efx_mcdi_rpc(efx, MC_CMD_VADAPTOR_QUERY, inbuf, sizeof(inbuf),
897 				  outbuf, sizeof(outbuf), &outlen);
898 		if (rc)
899 			return rc;
900 
901 		if (outlen < sizeof(outbuf)) {
902 			rc = -EIO;
903 			return rc;
904 		}
905 	}
906 
907 	if (port_flags)
908 		*port_flags = MCDI_DWORD(outbuf, VADAPTOR_QUERY_OUT_PORT_FLAGS);
909 	if (vadaptor_flags)
910 		*vadaptor_flags =
911 			MCDI_DWORD(outbuf, VADAPTOR_QUERY_OUT_VADAPTOR_FLAGS);
912 	if (vlan_tags)
913 		*vlan_tags =
914 			MCDI_DWORD(outbuf,
915 				   VADAPTOR_QUERY_OUT_NUM_AVAILABLE_VLAN_TAGS);
916 
917 	return 0;
918 }
919 
920 int efx_ef10_vadaptor_alloc(struct efx_nic *efx, unsigned int port_id)
921 {
922 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VADAPTOR_ALLOC_IN_LEN);
923 
924 	MCDI_SET_DWORD(inbuf, VADAPTOR_ALLOC_IN_UPSTREAM_PORT_ID, port_id);
925 	return efx_mcdi_rpc(efx, MC_CMD_VADAPTOR_ALLOC, inbuf, sizeof(inbuf),
926 			    NULL, 0, NULL);
927 }
928 
929 int efx_ef10_vadaptor_free(struct efx_nic *efx, unsigned int port_id)
930 {
931 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VADAPTOR_FREE_IN_LEN);
932 
933 	MCDI_SET_DWORD(inbuf, VADAPTOR_FREE_IN_UPSTREAM_PORT_ID, port_id);
934 	return efx_mcdi_rpc(efx, MC_CMD_VADAPTOR_FREE, inbuf, sizeof(inbuf),
935 			    NULL, 0, NULL);
936 }
937 
938 int efx_ef10_vport_add_mac(struct efx_nic *efx,
939 			   unsigned int port_id, u8 *mac)
940 {
941 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VPORT_ADD_MAC_ADDRESS_IN_LEN);
942 
943 	MCDI_SET_DWORD(inbuf, VPORT_ADD_MAC_ADDRESS_IN_VPORT_ID, port_id);
944 	ether_addr_copy(MCDI_PTR(inbuf, VPORT_ADD_MAC_ADDRESS_IN_MACADDR), mac);
945 
946 	return efx_mcdi_rpc(efx, MC_CMD_VPORT_ADD_MAC_ADDRESS, inbuf,
947 			    sizeof(inbuf), NULL, 0, NULL);
948 }
949 
950 int efx_ef10_vport_del_mac(struct efx_nic *efx,
951 			   unsigned int port_id, u8 *mac)
952 {
953 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VPORT_DEL_MAC_ADDRESS_IN_LEN);
954 
955 	MCDI_SET_DWORD(inbuf, VPORT_DEL_MAC_ADDRESS_IN_VPORT_ID, port_id);
956 	ether_addr_copy(MCDI_PTR(inbuf, VPORT_DEL_MAC_ADDRESS_IN_MACADDR), mac);
957 
958 	return efx_mcdi_rpc(efx, MC_CMD_VPORT_DEL_MAC_ADDRESS, inbuf,
959 			    sizeof(inbuf), NULL, 0, NULL);
960 }
961 
962 #ifdef CONFIG_SFC_SRIOV
963 static int efx_ef10_probe_vf(struct efx_nic *efx)
964 {
965 	int rc;
966 	struct pci_dev *pci_dev_pf;
967 
968 	/* If the parent PF has no VF data structure, it doesn't know about this
969 	 * VF so fail probe.  The VF needs to be re-created.  This can happen
970 	 * if the PF driver is unloaded while the VF is assigned to a guest.
971 	 */
972 	pci_dev_pf = efx->pci_dev->physfn;
973 	if (pci_dev_pf) {
974 		struct efx_nic *efx_pf = pci_get_drvdata(pci_dev_pf);
975 		struct efx_ef10_nic_data *nic_data_pf = efx_pf->nic_data;
976 
977 		if (!nic_data_pf->vf) {
978 			netif_info(efx, drv, efx->net_dev,
979 				   "The VF cannot link to its parent PF; "
980 				   "please destroy and re-create the VF\n");
981 			return -EBUSY;
982 		}
983 	}
984 
985 	rc = efx_ef10_probe(efx);
986 	if (rc)
987 		return rc;
988 
989 	rc = efx_ef10_get_vf_index(efx);
990 	if (rc)
991 		goto fail;
992 
993 	if (efx->pci_dev->is_virtfn) {
994 		if (efx->pci_dev->physfn) {
995 			struct efx_nic *efx_pf =
996 				pci_get_drvdata(efx->pci_dev->physfn);
997 			struct efx_ef10_nic_data *nic_data_p = efx_pf->nic_data;
998 			struct efx_ef10_nic_data *nic_data = efx->nic_data;
999 
1000 			nic_data_p->vf[nic_data->vf_index].efx = efx;
1001 			nic_data_p->vf[nic_data->vf_index].pci_dev =
1002 				efx->pci_dev;
1003 		} else
1004 			netif_info(efx, drv, efx->net_dev,
1005 				   "Could not get the PF id from VF\n");
1006 	}
1007 
1008 	return 0;
1009 
1010 fail:
1011 	efx_ef10_remove(efx);
1012 	return rc;
1013 }
1014 #else
1015 static int efx_ef10_probe_vf(struct efx_nic *efx __attribute__ ((unused)))
1016 {
1017 	return 0;
1018 }
1019 #endif
1020 
1021 static int efx_ef10_alloc_vis(struct efx_nic *efx,
1022 			      unsigned int min_vis, unsigned int max_vis)
1023 {
1024 	MCDI_DECLARE_BUF(inbuf, MC_CMD_ALLOC_VIS_IN_LEN);
1025 	MCDI_DECLARE_BUF(outbuf, MC_CMD_ALLOC_VIS_OUT_LEN);
1026 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1027 	size_t outlen;
1028 	int rc;
1029 
1030 	MCDI_SET_DWORD(inbuf, ALLOC_VIS_IN_MIN_VI_COUNT, min_vis);
1031 	MCDI_SET_DWORD(inbuf, ALLOC_VIS_IN_MAX_VI_COUNT, max_vis);
1032 	rc = efx_mcdi_rpc(efx, MC_CMD_ALLOC_VIS, inbuf, sizeof(inbuf),
1033 			  outbuf, sizeof(outbuf), &outlen);
1034 	if (rc != 0)
1035 		return rc;
1036 
1037 	if (outlen < MC_CMD_ALLOC_VIS_OUT_LEN)
1038 		return -EIO;
1039 
1040 	netif_dbg(efx, drv, efx->net_dev, "base VI is A0x%03x\n",
1041 		  MCDI_DWORD(outbuf, ALLOC_VIS_OUT_VI_BASE));
1042 
1043 	nic_data->vi_base = MCDI_DWORD(outbuf, ALLOC_VIS_OUT_VI_BASE);
1044 	nic_data->n_allocated_vis = MCDI_DWORD(outbuf, ALLOC_VIS_OUT_VI_COUNT);
1045 	return 0;
1046 }
1047 
1048 /* Note that the failure path of this function does not free
1049  * resources, as this will be done by efx_ef10_remove().
1050  */
1051 static int efx_ef10_dimension_resources(struct efx_nic *efx)
1052 {
1053 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1054 	unsigned int uc_mem_map_size, wc_mem_map_size;
1055 	unsigned int min_vis = max(EFX_TXQ_TYPES,
1056 				   efx_separate_tx_channels ? 2 : 1);
1057 	unsigned int channel_vis, pio_write_vi_base, max_vis;
1058 	void __iomem *membase;
1059 	int rc;
1060 
1061 	channel_vis = max(efx->n_channels, efx->n_tx_channels * EFX_TXQ_TYPES);
1062 
1063 #ifdef EFX_USE_PIO
1064 	/* Try to allocate PIO buffers if wanted and if the full
1065 	 * number of PIO buffers would be sufficient to allocate one
1066 	 * copy-buffer per TX channel.  Failure is non-fatal, as there
1067 	 * are only a small number of PIO buffers shared between all
1068 	 * functions of the controller.
1069 	 */
1070 	if (efx_piobuf_size != 0 &&
1071 	    ER_DZ_TX_PIOBUF_SIZE / efx_piobuf_size * EF10_TX_PIOBUF_COUNT >=
1072 	    efx->n_tx_channels) {
1073 		unsigned int n_piobufs =
1074 			DIV_ROUND_UP(efx->n_tx_channels,
1075 				     ER_DZ_TX_PIOBUF_SIZE / efx_piobuf_size);
1076 
1077 		rc = efx_ef10_alloc_piobufs(efx, n_piobufs);
1078 		if (rc)
1079 			netif_err(efx, probe, efx->net_dev,
1080 				  "failed to allocate PIO buffers (%d)\n", rc);
1081 		else
1082 			netif_dbg(efx, probe, efx->net_dev,
1083 				  "allocated %u PIO buffers\n", n_piobufs);
1084 	}
1085 #else
1086 	nic_data->n_piobufs = 0;
1087 #endif
1088 
1089 	/* PIO buffers should be mapped with write-combining enabled,
1090 	 * and we want to make single UC and WC mappings rather than
1091 	 * several of each (in fact that's the only option if host
1092 	 * page size is >4K).  So we may allocate some extra VIs just
1093 	 * for writing PIO buffers through.
1094 	 *
1095 	 * The UC mapping contains (channel_vis - 1) complete VIs and the
1096 	 * first half of the next VI.  Then the WC mapping begins with
1097 	 * the second half of this last VI.
1098 	 */
1099 	uc_mem_map_size = PAGE_ALIGN((channel_vis - 1) * EFX_VI_PAGE_SIZE +
1100 				     ER_DZ_TX_PIOBUF);
1101 	if (nic_data->n_piobufs) {
1102 		/* pio_write_vi_base rounds down to give the number of complete
1103 		 * VIs inside the UC mapping.
1104 		 */
1105 		pio_write_vi_base = uc_mem_map_size / EFX_VI_PAGE_SIZE;
1106 		wc_mem_map_size = (PAGE_ALIGN((pio_write_vi_base +
1107 					       nic_data->n_piobufs) *
1108 					      EFX_VI_PAGE_SIZE) -
1109 				   uc_mem_map_size);
1110 		max_vis = pio_write_vi_base + nic_data->n_piobufs;
1111 	} else {
1112 		pio_write_vi_base = 0;
1113 		wc_mem_map_size = 0;
1114 		max_vis = channel_vis;
1115 	}
1116 
1117 	/* In case the last attached driver failed to free VIs, do it now */
1118 	rc = efx_ef10_free_vis(efx);
1119 	if (rc != 0)
1120 		return rc;
1121 
1122 	rc = efx_ef10_alloc_vis(efx, min_vis, max_vis);
1123 	if (rc != 0)
1124 		return rc;
1125 
1126 	if (nic_data->n_allocated_vis < channel_vis) {
1127 		netif_info(efx, drv, efx->net_dev,
1128 			   "Could not allocate enough VIs to satisfy RSS"
1129 			   " requirements. Performance may not be optimal.\n");
1130 		/* We didn't get the VIs to populate our channels.
1131 		 * We could keep what we got but then we'd have more
1132 		 * interrupts than we need.
1133 		 * Instead calculate new max_channels and restart
1134 		 */
1135 		efx->max_channels = nic_data->n_allocated_vis;
1136 		efx->max_tx_channels =
1137 			nic_data->n_allocated_vis / EFX_TXQ_TYPES;
1138 
1139 		efx_ef10_free_vis(efx);
1140 		return -EAGAIN;
1141 	}
1142 
1143 	/* If we didn't get enough VIs to map all the PIO buffers, free the
1144 	 * PIO buffers
1145 	 */
1146 	if (nic_data->n_piobufs &&
1147 	    nic_data->n_allocated_vis <
1148 	    pio_write_vi_base + nic_data->n_piobufs) {
1149 		netif_dbg(efx, probe, efx->net_dev,
1150 			  "%u VIs are not sufficient to map %u PIO buffers\n",
1151 			  nic_data->n_allocated_vis, nic_data->n_piobufs);
1152 		efx_ef10_free_piobufs(efx);
1153 	}
1154 
1155 	/* Shrink the original UC mapping of the memory BAR */
1156 	membase = ioremap_nocache(efx->membase_phys, uc_mem_map_size);
1157 	if (!membase) {
1158 		netif_err(efx, probe, efx->net_dev,
1159 			  "could not shrink memory BAR to %x\n",
1160 			  uc_mem_map_size);
1161 		return -ENOMEM;
1162 	}
1163 	iounmap(efx->membase);
1164 	efx->membase = membase;
1165 
1166 	/* Set up the WC mapping if needed */
1167 	if (wc_mem_map_size) {
1168 		nic_data->wc_membase = ioremap_wc(efx->membase_phys +
1169 						  uc_mem_map_size,
1170 						  wc_mem_map_size);
1171 		if (!nic_data->wc_membase) {
1172 			netif_err(efx, probe, efx->net_dev,
1173 				  "could not allocate WC mapping of size %x\n",
1174 				  wc_mem_map_size);
1175 			return -ENOMEM;
1176 		}
1177 		nic_data->pio_write_vi_base = pio_write_vi_base;
1178 		nic_data->pio_write_base =
1179 			nic_data->wc_membase +
1180 			(pio_write_vi_base * EFX_VI_PAGE_SIZE + ER_DZ_TX_PIOBUF -
1181 			 uc_mem_map_size);
1182 
1183 		rc = efx_ef10_link_piobufs(efx);
1184 		if (rc)
1185 			efx_ef10_free_piobufs(efx);
1186 	}
1187 
1188 	netif_dbg(efx, probe, efx->net_dev,
1189 		  "memory BAR at %pa (virtual %p+%x UC, %p+%x WC)\n",
1190 		  &efx->membase_phys, efx->membase, uc_mem_map_size,
1191 		  nic_data->wc_membase, wc_mem_map_size);
1192 
1193 	return 0;
1194 }
1195 
1196 static int efx_ef10_init_nic(struct efx_nic *efx)
1197 {
1198 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1199 	int rc;
1200 
1201 	if (nic_data->must_check_datapath_caps) {
1202 		rc = efx_ef10_init_datapath_caps(efx);
1203 		if (rc)
1204 			return rc;
1205 		nic_data->must_check_datapath_caps = false;
1206 	}
1207 
1208 	if (nic_data->must_realloc_vis) {
1209 		/* We cannot let the number of VIs change now */
1210 		rc = efx_ef10_alloc_vis(efx, nic_data->n_allocated_vis,
1211 					nic_data->n_allocated_vis);
1212 		if (rc)
1213 			return rc;
1214 		nic_data->must_realloc_vis = false;
1215 	}
1216 
1217 	if (nic_data->must_restore_piobufs && nic_data->n_piobufs) {
1218 		rc = efx_ef10_alloc_piobufs(efx, nic_data->n_piobufs);
1219 		if (rc == 0) {
1220 			rc = efx_ef10_link_piobufs(efx);
1221 			if (rc)
1222 				efx_ef10_free_piobufs(efx);
1223 		}
1224 
1225 		/* Log an error on failure, but this is non-fatal */
1226 		if (rc)
1227 			netif_err(efx, drv, efx->net_dev,
1228 				  "failed to restore PIO buffers (%d)\n", rc);
1229 		nic_data->must_restore_piobufs = false;
1230 	}
1231 
1232 	/* don't fail init if RSS setup doesn't work */
1233 	efx->type->rx_push_rss_config(efx, false, efx->rx_indir_table);
1234 
1235 	return 0;
1236 }
1237 
1238 static void efx_ef10_reset_mc_allocations(struct efx_nic *efx)
1239 {
1240 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1241 #ifdef CONFIG_SFC_SRIOV
1242 	unsigned int i;
1243 #endif
1244 
1245 	/* All our allocations have been reset */
1246 	nic_data->must_realloc_vis = true;
1247 	nic_data->must_restore_filters = true;
1248 	nic_data->must_restore_piobufs = true;
1249 	efx_ef10_forget_old_piobufs(efx);
1250 	nic_data->rx_rss_context = EFX_EF10_RSS_CONTEXT_INVALID;
1251 
1252 	/* Driver-created vswitches and vports must be re-created */
1253 	nic_data->must_probe_vswitching = true;
1254 	nic_data->vport_id = EVB_PORT_ID_ASSIGNED;
1255 #ifdef CONFIG_SFC_SRIOV
1256 	if (nic_data->vf)
1257 		for (i = 0; i < efx->vf_count; i++)
1258 			nic_data->vf[i].vport_id = 0;
1259 #endif
1260 }
1261 
1262 static enum reset_type efx_ef10_map_reset_reason(enum reset_type reason)
1263 {
1264 	if (reason == RESET_TYPE_MC_FAILURE)
1265 		return RESET_TYPE_DATAPATH;
1266 
1267 	return efx_mcdi_map_reset_reason(reason);
1268 }
1269 
1270 static int efx_ef10_map_reset_flags(u32 *flags)
1271 {
1272 	enum {
1273 		EF10_RESET_PORT = ((ETH_RESET_MAC | ETH_RESET_PHY) <<
1274 				   ETH_RESET_SHARED_SHIFT),
1275 		EF10_RESET_MC = ((ETH_RESET_DMA | ETH_RESET_FILTER |
1276 				  ETH_RESET_OFFLOAD | ETH_RESET_MAC |
1277 				  ETH_RESET_PHY | ETH_RESET_MGMT) <<
1278 				 ETH_RESET_SHARED_SHIFT)
1279 	};
1280 
1281 	/* We assume for now that our PCI function is permitted to
1282 	 * reset everything.
1283 	 */
1284 
1285 	if ((*flags & EF10_RESET_MC) == EF10_RESET_MC) {
1286 		*flags &= ~EF10_RESET_MC;
1287 		return RESET_TYPE_WORLD;
1288 	}
1289 
1290 	if ((*flags & EF10_RESET_PORT) == EF10_RESET_PORT) {
1291 		*flags &= ~EF10_RESET_PORT;
1292 		return RESET_TYPE_ALL;
1293 	}
1294 
1295 	/* no invisible reset implemented */
1296 
1297 	return -EINVAL;
1298 }
1299 
1300 static int efx_ef10_reset(struct efx_nic *efx, enum reset_type reset_type)
1301 {
1302 	int rc = efx_mcdi_reset(efx, reset_type);
1303 
1304 	/* Unprivileged functions return -EPERM, but need to return success
1305 	 * here so that the datapath is brought back up.
1306 	 */
1307 	if (reset_type == RESET_TYPE_WORLD && rc == -EPERM)
1308 		rc = 0;
1309 
1310 	/* If it was a port reset, trigger reallocation of MC resources.
1311 	 * Note that on an MC reset nothing needs to be done now because we'll
1312 	 * detect the MC reset later and handle it then.
1313 	 * For an FLR, we never get an MC reset event, but the MC has reset all
1314 	 * resources assigned to us, so we have to trigger reallocation now.
1315 	 */
1316 	if ((reset_type == RESET_TYPE_ALL ||
1317 	     reset_type == RESET_TYPE_MCDI_TIMEOUT) && !rc)
1318 		efx_ef10_reset_mc_allocations(efx);
1319 	return rc;
1320 }
1321 
1322 #define EF10_DMA_STAT(ext_name, mcdi_name)			\
1323 	[EF10_STAT_ ## ext_name] =				\
1324 	{ #ext_name, 64, 8 * MC_CMD_MAC_ ## mcdi_name }
1325 #define EF10_DMA_INVIS_STAT(int_name, mcdi_name)		\
1326 	[EF10_STAT_ ## int_name] =				\
1327 	{ NULL, 64, 8 * MC_CMD_MAC_ ## mcdi_name }
1328 #define EF10_OTHER_STAT(ext_name)				\
1329 	[EF10_STAT_ ## ext_name] = { #ext_name, 0, 0 }
1330 #define GENERIC_SW_STAT(ext_name)				\
1331 	[GENERIC_STAT_ ## ext_name] = { #ext_name, 0, 0 }
1332 
1333 static const struct efx_hw_stat_desc efx_ef10_stat_desc[EF10_STAT_COUNT] = {
1334 	EF10_DMA_STAT(port_tx_bytes, TX_BYTES),
1335 	EF10_DMA_STAT(port_tx_packets, TX_PKTS),
1336 	EF10_DMA_STAT(port_tx_pause, TX_PAUSE_PKTS),
1337 	EF10_DMA_STAT(port_tx_control, TX_CONTROL_PKTS),
1338 	EF10_DMA_STAT(port_tx_unicast, TX_UNICAST_PKTS),
1339 	EF10_DMA_STAT(port_tx_multicast, TX_MULTICAST_PKTS),
1340 	EF10_DMA_STAT(port_tx_broadcast, TX_BROADCAST_PKTS),
1341 	EF10_DMA_STAT(port_tx_lt64, TX_LT64_PKTS),
1342 	EF10_DMA_STAT(port_tx_64, TX_64_PKTS),
1343 	EF10_DMA_STAT(port_tx_65_to_127, TX_65_TO_127_PKTS),
1344 	EF10_DMA_STAT(port_tx_128_to_255, TX_128_TO_255_PKTS),
1345 	EF10_DMA_STAT(port_tx_256_to_511, TX_256_TO_511_PKTS),
1346 	EF10_DMA_STAT(port_tx_512_to_1023, TX_512_TO_1023_PKTS),
1347 	EF10_DMA_STAT(port_tx_1024_to_15xx, TX_1024_TO_15XX_PKTS),
1348 	EF10_DMA_STAT(port_tx_15xx_to_jumbo, TX_15XX_TO_JUMBO_PKTS),
1349 	EF10_DMA_STAT(port_rx_bytes, RX_BYTES),
1350 	EF10_DMA_INVIS_STAT(port_rx_bytes_minus_good_bytes, RX_BAD_BYTES),
1351 	EF10_OTHER_STAT(port_rx_good_bytes),
1352 	EF10_OTHER_STAT(port_rx_bad_bytes),
1353 	EF10_DMA_STAT(port_rx_packets, RX_PKTS),
1354 	EF10_DMA_STAT(port_rx_good, RX_GOOD_PKTS),
1355 	EF10_DMA_STAT(port_rx_bad, RX_BAD_FCS_PKTS),
1356 	EF10_DMA_STAT(port_rx_pause, RX_PAUSE_PKTS),
1357 	EF10_DMA_STAT(port_rx_control, RX_CONTROL_PKTS),
1358 	EF10_DMA_STAT(port_rx_unicast, RX_UNICAST_PKTS),
1359 	EF10_DMA_STAT(port_rx_multicast, RX_MULTICAST_PKTS),
1360 	EF10_DMA_STAT(port_rx_broadcast, RX_BROADCAST_PKTS),
1361 	EF10_DMA_STAT(port_rx_lt64, RX_UNDERSIZE_PKTS),
1362 	EF10_DMA_STAT(port_rx_64, RX_64_PKTS),
1363 	EF10_DMA_STAT(port_rx_65_to_127, RX_65_TO_127_PKTS),
1364 	EF10_DMA_STAT(port_rx_128_to_255, RX_128_TO_255_PKTS),
1365 	EF10_DMA_STAT(port_rx_256_to_511, RX_256_TO_511_PKTS),
1366 	EF10_DMA_STAT(port_rx_512_to_1023, RX_512_TO_1023_PKTS),
1367 	EF10_DMA_STAT(port_rx_1024_to_15xx, RX_1024_TO_15XX_PKTS),
1368 	EF10_DMA_STAT(port_rx_15xx_to_jumbo, RX_15XX_TO_JUMBO_PKTS),
1369 	EF10_DMA_STAT(port_rx_gtjumbo, RX_GTJUMBO_PKTS),
1370 	EF10_DMA_STAT(port_rx_bad_gtjumbo, RX_JABBER_PKTS),
1371 	EF10_DMA_STAT(port_rx_overflow, RX_OVERFLOW_PKTS),
1372 	EF10_DMA_STAT(port_rx_align_error, RX_ALIGN_ERROR_PKTS),
1373 	EF10_DMA_STAT(port_rx_length_error, RX_LENGTH_ERROR_PKTS),
1374 	EF10_DMA_STAT(port_rx_nodesc_drops, RX_NODESC_DROPS),
1375 	GENERIC_SW_STAT(rx_nodesc_trunc),
1376 	GENERIC_SW_STAT(rx_noskb_drops),
1377 	EF10_DMA_STAT(port_rx_pm_trunc_bb_overflow, PM_TRUNC_BB_OVERFLOW),
1378 	EF10_DMA_STAT(port_rx_pm_discard_bb_overflow, PM_DISCARD_BB_OVERFLOW),
1379 	EF10_DMA_STAT(port_rx_pm_trunc_vfifo_full, PM_TRUNC_VFIFO_FULL),
1380 	EF10_DMA_STAT(port_rx_pm_discard_vfifo_full, PM_DISCARD_VFIFO_FULL),
1381 	EF10_DMA_STAT(port_rx_pm_trunc_qbb, PM_TRUNC_QBB),
1382 	EF10_DMA_STAT(port_rx_pm_discard_qbb, PM_DISCARD_QBB),
1383 	EF10_DMA_STAT(port_rx_pm_discard_mapping, PM_DISCARD_MAPPING),
1384 	EF10_DMA_STAT(port_rx_dp_q_disabled_packets, RXDP_Q_DISABLED_PKTS),
1385 	EF10_DMA_STAT(port_rx_dp_di_dropped_packets, RXDP_DI_DROPPED_PKTS),
1386 	EF10_DMA_STAT(port_rx_dp_streaming_packets, RXDP_STREAMING_PKTS),
1387 	EF10_DMA_STAT(port_rx_dp_hlb_fetch, RXDP_HLB_FETCH_CONDITIONS),
1388 	EF10_DMA_STAT(port_rx_dp_hlb_wait, RXDP_HLB_WAIT_CONDITIONS),
1389 	EF10_DMA_STAT(rx_unicast, VADAPTER_RX_UNICAST_PACKETS),
1390 	EF10_DMA_STAT(rx_unicast_bytes, VADAPTER_RX_UNICAST_BYTES),
1391 	EF10_DMA_STAT(rx_multicast, VADAPTER_RX_MULTICAST_PACKETS),
1392 	EF10_DMA_STAT(rx_multicast_bytes, VADAPTER_RX_MULTICAST_BYTES),
1393 	EF10_DMA_STAT(rx_broadcast, VADAPTER_RX_BROADCAST_PACKETS),
1394 	EF10_DMA_STAT(rx_broadcast_bytes, VADAPTER_RX_BROADCAST_BYTES),
1395 	EF10_DMA_STAT(rx_bad, VADAPTER_RX_BAD_PACKETS),
1396 	EF10_DMA_STAT(rx_bad_bytes, VADAPTER_RX_BAD_BYTES),
1397 	EF10_DMA_STAT(rx_overflow, VADAPTER_RX_OVERFLOW),
1398 	EF10_DMA_STAT(tx_unicast, VADAPTER_TX_UNICAST_PACKETS),
1399 	EF10_DMA_STAT(tx_unicast_bytes, VADAPTER_TX_UNICAST_BYTES),
1400 	EF10_DMA_STAT(tx_multicast, VADAPTER_TX_MULTICAST_PACKETS),
1401 	EF10_DMA_STAT(tx_multicast_bytes, VADAPTER_TX_MULTICAST_BYTES),
1402 	EF10_DMA_STAT(tx_broadcast, VADAPTER_TX_BROADCAST_PACKETS),
1403 	EF10_DMA_STAT(tx_broadcast_bytes, VADAPTER_TX_BROADCAST_BYTES),
1404 	EF10_DMA_STAT(tx_bad, VADAPTER_TX_BAD_PACKETS),
1405 	EF10_DMA_STAT(tx_bad_bytes, VADAPTER_TX_BAD_BYTES),
1406 	EF10_DMA_STAT(tx_overflow, VADAPTER_TX_OVERFLOW),
1407 };
1408 
1409 #define HUNT_COMMON_STAT_MASK ((1ULL << EF10_STAT_port_tx_bytes) |	\
1410 			       (1ULL << EF10_STAT_port_tx_packets) |	\
1411 			       (1ULL << EF10_STAT_port_tx_pause) |	\
1412 			       (1ULL << EF10_STAT_port_tx_unicast) |	\
1413 			       (1ULL << EF10_STAT_port_tx_multicast) |	\
1414 			       (1ULL << EF10_STAT_port_tx_broadcast) |	\
1415 			       (1ULL << EF10_STAT_port_rx_bytes) |	\
1416 			       (1ULL <<                                 \
1417 				EF10_STAT_port_rx_bytes_minus_good_bytes) | \
1418 			       (1ULL << EF10_STAT_port_rx_good_bytes) |	\
1419 			       (1ULL << EF10_STAT_port_rx_bad_bytes) |	\
1420 			       (1ULL << EF10_STAT_port_rx_packets) |	\
1421 			       (1ULL << EF10_STAT_port_rx_good) |	\
1422 			       (1ULL << EF10_STAT_port_rx_bad) |	\
1423 			       (1ULL << EF10_STAT_port_rx_pause) |	\
1424 			       (1ULL << EF10_STAT_port_rx_control) |	\
1425 			       (1ULL << EF10_STAT_port_rx_unicast) |	\
1426 			       (1ULL << EF10_STAT_port_rx_multicast) |	\
1427 			       (1ULL << EF10_STAT_port_rx_broadcast) |	\
1428 			       (1ULL << EF10_STAT_port_rx_lt64) |	\
1429 			       (1ULL << EF10_STAT_port_rx_64) |		\
1430 			       (1ULL << EF10_STAT_port_rx_65_to_127) |	\
1431 			       (1ULL << EF10_STAT_port_rx_128_to_255) |	\
1432 			       (1ULL << EF10_STAT_port_rx_256_to_511) |	\
1433 			       (1ULL << EF10_STAT_port_rx_512_to_1023) |\
1434 			       (1ULL << EF10_STAT_port_rx_1024_to_15xx) |\
1435 			       (1ULL << EF10_STAT_port_rx_15xx_to_jumbo) |\
1436 			       (1ULL << EF10_STAT_port_rx_gtjumbo) |	\
1437 			       (1ULL << EF10_STAT_port_rx_bad_gtjumbo) |\
1438 			       (1ULL << EF10_STAT_port_rx_overflow) |	\
1439 			       (1ULL << EF10_STAT_port_rx_nodesc_drops) |\
1440 			       (1ULL << GENERIC_STAT_rx_nodesc_trunc) |	\
1441 			       (1ULL << GENERIC_STAT_rx_noskb_drops))
1442 
1443 /* These statistics are only provided by the 10G MAC.  For a 10G/40G
1444  * switchable port we do not expose these because they might not
1445  * include all the packets they should.
1446  */
1447 #define HUNT_10G_ONLY_STAT_MASK ((1ULL << EF10_STAT_port_tx_control) |	\
1448 				 (1ULL << EF10_STAT_port_tx_lt64) |	\
1449 				 (1ULL << EF10_STAT_port_tx_64) |	\
1450 				 (1ULL << EF10_STAT_port_tx_65_to_127) |\
1451 				 (1ULL << EF10_STAT_port_tx_128_to_255) |\
1452 				 (1ULL << EF10_STAT_port_tx_256_to_511) |\
1453 				 (1ULL << EF10_STAT_port_tx_512_to_1023) |\
1454 				 (1ULL << EF10_STAT_port_tx_1024_to_15xx) |\
1455 				 (1ULL << EF10_STAT_port_tx_15xx_to_jumbo))
1456 
1457 /* These statistics are only provided by the 40G MAC.  For a 10G/40G
1458  * switchable port we do expose these because the errors will otherwise
1459  * be silent.
1460  */
1461 #define HUNT_40G_EXTRA_STAT_MASK ((1ULL << EF10_STAT_port_rx_align_error) |\
1462 				  (1ULL << EF10_STAT_port_rx_length_error))
1463 
1464 /* These statistics are only provided if the firmware supports the
1465  * capability PM_AND_RXDP_COUNTERS.
1466  */
1467 #define HUNT_PM_AND_RXDP_STAT_MASK (					\
1468 	(1ULL << EF10_STAT_port_rx_pm_trunc_bb_overflow) |		\
1469 	(1ULL << EF10_STAT_port_rx_pm_discard_bb_overflow) |		\
1470 	(1ULL << EF10_STAT_port_rx_pm_trunc_vfifo_full) |		\
1471 	(1ULL << EF10_STAT_port_rx_pm_discard_vfifo_full) |		\
1472 	(1ULL << EF10_STAT_port_rx_pm_trunc_qbb) |			\
1473 	(1ULL << EF10_STAT_port_rx_pm_discard_qbb) |			\
1474 	(1ULL << EF10_STAT_port_rx_pm_discard_mapping) |		\
1475 	(1ULL << EF10_STAT_port_rx_dp_q_disabled_packets) |		\
1476 	(1ULL << EF10_STAT_port_rx_dp_di_dropped_packets) |		\
1477 	(1ULL << EF10_STAT_port_rx_dp_streaming_packets) |		\
1478 	(1ULL << EF10_STAT_port_rx_dp_hlb_fetch) |			\
1479 	(1ULL << EF10_STAT_port_rx_dp_hlb_wait))
1480 
1481 static u64 efx_ef10_raw_stat_mask(struct efx_nic *efx)
1482 {
1483 	u64 raw_mask = HUNT_COMMON_STAT_MASK;
1484 	u32 port_caps = efx_mcdi_phy_get_caps(efx);
1485 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1486 
1487 	if (!(efx->mcdi->fn_flags &
1488 	      1 << MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_LINKCTRL))
1489 		return 0;
1490 
1491 	if (port_caps & (1 << MC_CMD_PHY_CAP_40000FDX_LBN))
1492 		raw_mask |= HUNT_40G_EXTRA_STAT_MASK;
1493 	else
1494 		raw_mask |= HUNT_10G_ONLY_STAT_MASK;
1495 
1496 	if (nic_data->datapath_caps &
1497 	    (1 << MC_CMD_GET_CAPABILITIES_OUT_PM_AND_RXDP_COUNTERS_LBN))
1498 		raw_mask |= HUNT_PM_AND_RXDP_STAT_MASK;
1499 
1500 	return raw_mask;
1501 }
1502 
1503 static void efx_ef10_get_stat_mask(struct efx_nic *efx, unsigned long *mask)
1504 {
1505 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1506 	u64 raw_mask[2];
1507 
1508 	raw_mask[0] = efx_ef10_raw_stat_mask(efx);
1509 
1510 	/* Only show vadaptor stats when EVB capability is present */
1511 	if (nic_data->datapath_caps &
1512 	    (1 << MC_CMD_GET_CAPABILITIES_OUT_EVB_LBN)) {
1513 		raw_mask[0] |= ~((1ULL << EF10_STAT_rx_unicast) - 1);
1514 		raw_mask[1] = (1ULL << (EF10_STAT_COUNT - 63)) - 1;
1515 	} else {
1516 		raw_mask[1] = 0;
1517 	}
1518 
1519 #if BITS_PER_LONG == 64
1520 	mask[0] = raw_mask[0];
1521 	mask[1] = raw_mask[1];
1522 #else
1523 	mask[0] = raw_mask[0] & 0xffffffff;
1524 	mask[1] = raw_mask[0] >> 32;
1525 	mask[2] = raw_mask[1] & 0xffffffff;
1526 	mask[3] = raw_mask[1] >> 32;
1527 #endif
1528 }
1529 
1530 static size_t efx_ef10_describe_stats(struct efx_nic *efx, u8 *names)
1531 {
1532 	DECLARE_BITMAP(mask, EF10_STAT_COUNT);
1533 
1534 	efx_ef10_get_stat_mask(efx, mask);
1535 	return efx_nic_describe_stats(efx_ef10_stat_desc, EF10_STAT_COUNT,
1536 				      mask, names);
1537 }
1538 
1539 static size_t efx_ef10_update_stats_common(struct efx_nic *efx, u64 *full_stats,
1540 					   struct rtnl_link_stats64 *core_stats)
1541 {
1542 	DECLARE_BITMAP(mask, EF10_STAT_COUNT);
1543 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1544 	u64 *stats = nic_data->stats;
1545 	size_t stats_count = 0, index;
1546 
1547 	efx_ef10_get_stat_mask(efx, mask);
1548 
1549 	if (full_stats) {
1550 		for_each_set_bit(index, mask, EF10_STAT_COUNT) {
1551 			if (efx_ef10_stat_desc[index].name) {
1552 				*full_stats++ = stats[index];
1553 				++stats_count;
1554 			}
1555 		}
1556 	}
1557 
1558 	if (!core_stats)
1559 		return stats_count;
1560 
1561 	if (nic_data->datapath_caps &
1562 			1 << MC_CMD_GET_CAPABILITIES_OUT_EVB_LBN) {
1563 		/* Use vadaptor stats. */
1564 		core_stats->rx_packets = stats[EF10_STAT_rx_unicast] +
1565 					 stats[EF10_STAT_rx_multicast] +
1566 					 stats[EF10_STAT_rx_broadcast];
1567 		core_stats->tx_packets = stats[EF10_STAT_tx_unicast] +
1568 					 stats[EF10_STAT_tx_multicast] +
1569 					 stats[EF10_STAT_tx_broadcast];
1570 		core_stats->rx_bytes = stats[EF10_STAT_rx_unicast_bytes] +
1571 				       stats[EF10_STAT_rx_multicast_bytes] +
1572 				       stats[EF10_STAT_rx_broadcast_bytes];
1573 		core_stats->tx_bytes = stats[EF10_STAT_tx_unicast_bytes] +
1574 				       stats[EF10_STAT_tx_multicast_bytes] +
1575 				       stats[EF10_STAT_tx_broadcast_bytes];
1576 		core_stats->rx_dropped = stats[GENERIC_STAT_rx_nodesc_trunc] +
1577 					 stats[GENERIC_STAT_rx_noskb_drops];
1578 		core_stats->multicast = stats[EF10_STAT_rx_multicast];
1579 		core_stats->rx_crc_errors = stats[EF10_STAT_rx_bad];
1580 		core_stats->rx_fifo_errors = stats[EF10_STAT_rx_overflow];
1581 		core_stats->rx_errors = core_stats->rx_crc_errors;
1582 		core_stats->tx_errors = stats[EF10_STAT_tx_bad];
1583 	} else {
1584 		/* Use port stats. */
1585 		core_stats->rx_packets = stats[EF10_STAT_port_rx_packets];
1586 		core_stats->tx_packets = stats[EF10_STAT_port_tx_packets];
1587 		core_stats->rx_bytes = stats[EF10_STAT_port_rx_bytes];
1588 		core_stats->tx_bytes = stats[EF10_STAT_port_tx_bytes];
1589 		core_stats->rx_dropped = stats[EF10_STAT_port_rx_nodesc_drops] +
1590 					 stats[GENERIC_STAT_rx_nodesc_trunc] +
1591 					 stats[GENERIC_STAT_rx_noskb_drops];
1592 		core_stats->multicast = stats[EF10_STAT_port_rx_multicast];
1593 		core_stats->rx_length_errors =
1594 				stats[EF10_STAT_port_rx_gtjumbo] +
1595 				stats[EF10_STAT_port_rx_length_error];
1596 		core_stats->rx_crc_errors = stats[EF10_STAT_port_rx_bad];
1597 		core_stats->rx_frame_errors =
1598 				stats[EF10_STAT_port_rx_align_error];
1599 		core_stats->rx_fifo_errors = stats[EF10_STAT_port_rx_overflow];
1600 		core_stats->rx_errors = (core_stats->rx_length_errors +
1601 					 core_stats->rx_crc_errors +
1602 					 core_stats->rx_frame_errors);
1603 	}
1604 
1605 	return stats_count;
1606 }
1607 
1608 static int efx_ef10_try_update_nic_stats_pf(struct efx_nic *efx)
1609 {
1610 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1611 	DECLARE_BITMAP(mask, EF10_STAT_COUNT);
1612 	__le64 generation_start, generation_end;
1613 	u64 *stats = nic_data->stats;
1614 	__le64 *dma_stats;
1615 
1616 	efx_ef10_get_stat_mask(efx, mask);
1617 
1618 	dma_stats = efx->stats_buffer.addr;
1619 	nic_data = efx->nic_data;
1620 
1621 	generation_end = dma_stats[MC_CMD_MAC_GENERATION_END];
1622 	if (generation_end == EFX_MC_STATS_GENERATION_INVALID)
1623 		return 0;
1624 	rmb();
1625 	efx_nic_update_stats(efx_ef10_stat_desc, EF10_STAT_COUNT, mask,
1626 			     stats, efx->stats_buffer.addr, false);
1627 	rmb();
1628 	generation_start = dma_stats[MC_CMD_MAC_GENERATION_START];
1629 	if (generation_end != generation_start)
1630 		return -EAGAIN;
1631 
1632 	/* Update derived statistics */
1633 	efx_nic_fix_nodesc_drop_stat(efx,
1634 				     &stats[EF10_STAT_port_rx_nodesc_drops]);
1635 	stats[EF10_STAT_port_rx_good_bytes] =
1636 		stats[EF10_STAT_port_rx_bytes] -
1637 		stats[EF10_STAT_port_rx_bytes_minus_good_bytes];
1638 	efx_update_diff_stat(&stats[EF10_STAT_port_rx_bad_bytes],
1639 			     stats[EF10_STAT_port_rx_bytes_minus_good_bytes]);
1640 	efx_update_sw_stats(efx, stats);
1641 	return 0;
1642 }
1643 
1644 
1645 static size_t efx_ef10_update_stats_pf(struct efx_nic *efx, u64 *full_stats,
1646 				       struct rtnl_link_stats64 *core_stats)
1647 {
1648 	int retry;
1649 
1650 	/* If we're unlucky enough to read statistics during the DMA, wait
1651 	 * up to 10ms for it to finish (typically takes <500us)
1652 	 */
1653 	for (retry = 0; retry < 100; ++retry) {
1654 		if (efx_ef10_try_update_nic_stats_pf(efx) == 0)
1655 			break;
1656 		udelay(100);
1657 	}
1658 
1659 	return efx_ef10_update_stats_common(efx, full_stats, core_stats);
1660 }
1661 
1662 static int efx_ef10_try_update_nic_stats_vf(struct efx_nic *efx)
1663 {
1664 	MCDI_DECLARE_BUF(inbuf, MC_CMD_MAC_STATS_IN_LEN);
1665 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1666 	DECLARE_BITMAP(mask, EF10_STAT_COUNT);
1667 	__le64 generation_start, generation_end;
1668 	u64 *stats = nic_data->stats;
1669 	u32 dma_len = MC_CMD_MAC_NSTATS * sizeof(u64);
1670 	struct efx_buffer stats_buf;
1671 	__le64 *dma_stats;
1672 	int rc;
1673 
1674 	spin_unlock_bh(&efx->stats_lock);
1675 
1676 	if (in_interrupt()) {
1677 		/* If in atomic context, cannot update stats.  Just update the
1678 		 * software stats and return so the caller can continue.
1679 		 */
1680 		spin_lock_bh(&efx->stats_lock);
1681 		efx_update_sw_stats(efx, stats);
1682 		return 0;
1683 	}
1684 
1685 	efx_ef10_get_stat_mask(efx, mask);
1686 
1687 	rc = efx_nic_alloc_buffer(efx, &stats_buf, dma_len, GFP_ATOMIC);
1688 	if (rc) {
1689 		spin_lock_bh(&efx->stats_lock);
1690 		return rc;
1691 	}
1692 
1693 	dma_stats = stats_buf.addr;
1694 	dma_stats[MC_CMD_MAC_GENERATION_END] = EFX_MC_STATS_GENERATION_INVALID;
1695 
1696 	MCDI_SET_QWORD(inbuf, MAC_STATS_IN_DMA_ADDR, stats_buf.dma_addr);
1697 	MCDI_POPULATE_DWORD_1(inbuf, MAC_STATS_IN_CMD,
1698 			      MAC_STATS_IN_DMA, 1);
1699 	MCDI_SET_DWORD(inbuf, MAC_STATS_IN_DMA_LEN, dma_len);
1700 	MCDI_SET_DWORD(inbuf, MAC_STATS_IN_PORT_ID, EVB_PORT_ID_ASSIGNED);
1701 
1702 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_MAC_STATS, inbuf, sizeof(inbuf),
1703 				NULL, 0, NULL);
1704 	spin_lock_bh(&efx->stats_lock);
1705 	if (rc) {
1706 		/* Expect ENOENT if DMA queues have not been set up */
1707 		if (rc != -ENOENT || atomic_read(&efx->active_queues))
1708 			efx_mcdi_display_error(efx, MC_CMD_MAC_STATS,
1709 					       sizeof(inbuf), NULL, 0, rc);
1710 		goto out;
1711 	}
1712 
1713 	generation_end = dma_stats[MC_CMD_MAC_GENERATION_END];
1714 	if (generation_end == EFX_MC_STATS_GENERATION_INVALID) {
1715 		WARN_ON_ONCE(1);
1716 		goto out;
1717 	}
1718 	rmb();
1719 	efx_nic_update_stats(efx_ef10_stat_desc, EF10_STAT_COUNT, mask,
1720 			     stats, stats_buf.addr, false);
1721 	rmb();
1722 	generation_start = dma_stats[MC_CMD_MAC_GENERATION_START];
1723 	if (generation_end != generation_start) {
1724 		rc = -EAGAIN;
1725 		goto out;
1726 	}
1727 
1728 	efx_update_sw_stats(efx, stats);
1729 out:
1730 	efx_nic_free_buffer(efx, &stats_buf);
1731 	return rc;
1732 }
1733 
1734 static size_t efx_ef10_update_stats_vf(struct efx_nic *efx, u64 *full_stats,
1735 				       struct rtnl_link_stats64 *core_stats)
1736 {
1737 	if (efx_ef10_try_update_nic_stats_vf(efx))
1738 		return 0;
1739 
1740 	return efx_ef10_update_stats_common(efx, full_stats, core_stats);
1741 }
1742 
1743 static void efx_ef10_push_irq_moderation(struct efx_channel *channel)
1744 {
1745 	struct efx_nic *efx = channel->efx;
1746 	unsigned int mode, value;
1747 	efx_dword_t timer_cmd;
1748 
1749 	if (channel->irq_moderation) {
1750 		mode = 3;
1751 		value = channel->irq_moderation - 1;
1752 	} else {
1753 		mode = 0;
1754 		value = 0;
1755 	}
1756 
1757 	if (EFX_EF10_WORKAROUND_35388(efx)) {
1758 		EFX_POPULATE_DWORD_3(timer_cmd, ERF_DD_EVQ_IND_TIMER_FLAGS,
1759 				     EFE_DD_EVQ_IND_TIMER_FLAGS,
1760 				     ERF_DD_EVQ_IND_TIMER_MODE, mode,
1761 				     ERF_DD_EVQ_IND_TIMER_VAL, value);
1762 		efx_writed_page(efx, &timer_cmd, ER_DD_EVQ_INDIRECT,
1763 				channel->channel);
1764 	} else {
1765 		EFX_POPULATE_DWORD_2(timer_cmd, ERF_DZ_TC_TIMER_MODE, mode,
1766 				     ERF_DZ_TC_TIMER_VAL, value);
1767 		efx_writed_page(efx, &timer_cmd, ER_DZ_EVQ_TMR,
1768 				channel->channel);
1769 	}
1770 }
1771 
1772 static void efx_ef10_get_wol_vf(struct efx_nic *efx,
1773 				struct ethtool_wolinfo *wol) {}
1774 
1775 static int efx_ef10_set_wol_vf(struct efx_nic *efx, u32 type)
1776 {
1777 	return -EOPNOTSUPP;
1778 }
1779 
1780 static void efx_ef10_get_wol(struct efx_nic *efx, struct ethtool_wolinfo *wol)
1781 {
1782 	wol->supported = 0;
1783 	wol->wolopts = 0;
1784 	memset(&wol->sopass, 0, sizeof(wol->sopass));
1785 }
1786 
1787 static int efx_ef10_set_wol(struct efx_nic *efx, u32 type)
1788 {
1789 	if (type != 0)
1790 		return -EINVAL;
1791 	return 0;
1792 }
1793 
1794 static void efx_ef10_mcdi_request(struct efx_nic *efx,
1795 				  const efx_dword_t *hdr, size_t hdr_len,
1796 				  const efx_dword_t *sdu, size_t sdu_len)
1797 {
1798 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1799 	u8 *pdu = nic_data->mcdi_buf.addr;
1800 
1801 	memcpy(pdu, hdr, hdr_len);
1802 	memcpy(pdu + hdr_len, sdu, sdu_len);
1803 	wmb();
1804 
1805 	/* The hardware provides 'low' and 'high' (doorbell) registers
1806 	 * for passing the 64-bit address of an MCDI request to
1807 	 * firmware.  However the dwords are swapped by firmware.  The
1808 	 * least significant bits of the doorbell are then 0 for all
1809 	 * MCDI requests due to alignment.
1810 	 */
1811 	_efx_writed(efx, cpu_to_le32((u64)nic_data->mcdi_buf.dma_addr >> 32),
1812 		    ER_DZ_MC_DB_LWRD);
1813 	_efx_writed(efx, cpu_to_le32((u32)nic_data->mcdi_buf.dma_addr),
1814 		    ER_DZ_MC_DB_HWRD);
1815 }
1816 
1817 static bool efx_ef10_mcdi_poll_response(struct efx_nic *efx)
1818 {
1819 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1820 	const efx_dword_t hdr = *(const efx_dword_t *)nic_data->mcdi_buf.addr;
1821 
1822 	rmb();
1823 	return EFX_DWORD_FIELD(hdr, MCDI_HEADER_RESPONSE);
1824 }
1825 
1826 static void
1827 efx_ef10_mcdi_read_response(struct efx_nic *efx, efx_dword_t *outbuf,
1828 			    size_t offset, size_t outlen)
1829 {
1830 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1831 	const u8 *pdu = nic_data->mcdi_buf.addr;
1832 
1833 	memcpy(outbuf, pdu + offset, outlen);
1834 }
1835 
1836 static void efx_ef10_mcdi_reboot_detected(struct efx_nic *efx)
1837 {
1838 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1839 
1840 	/* All our allocations have been reset */
1841 	efx_ef10_reset_mc_allocations(efx);
1842 
1843 	/* The datapath firmware might have been changed */
1844 	nic_data->must_check_datapath_caps = true;
1845 
1846 	/* MAC statistics have been cleared on the NIC; clear the local
1847 	 * statistic that we update with efx_update_diff_stat().
1848 	 */
1849 	nic_data->stats[EF10_STAT_port_rx_bad_bytes] = 0;
1850 }
1851 
1852 static int efx_ef10_mcdi_poll_reboot(struct efx_nic *efx)
1853 {
1854 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1855 	int rc;
1856 
1857 	rc = efx_ef10_get_warm_boot_count(efx);
1858 	if (rc < 0) {
1859 		/* The firmware is presumably in the process of
1860 		 * rebooting.  However, we are supposed to report each
1861 		 * reboot just once, so we must only do that once we
1862 		 * can read and store the updated warm boot count.
1863 		 */
1864 		return 0;
1865 	}
1866 
1867 	if (rc == nic_data->warm_boot_count)
1868 		return 0;
1869 
1870 	nic_data->warm_boot_count = rc;
1871 	efx_ef10_mcdi_reboot_detected(efx);
1872 
1873 	return -EIO;
1874 }
1875 
1876 /* Handle an MSI interrupt
1877  *
1878  * Handle an MSI hardware interrupt.  This routine schedules event
1879  * queue processing.  No interrupt acknowledgement cycle is necessary.
1880  * Also, we never need to check that the interrupt is for us, since
1881  * MSI interrupts cannot be shared.
1882  */
1883 static irqreturn_t efx_ef10_msi_interrupt(int irq, void *dev_id)
1884 {
1885 	struct efx_msi_context *context = dev_id;
1886 	struct efx_nic *efx = context->efx;
1887 
1888 	netif_vdbg(efx, intr, efx->net_dev,
1889 		   "IRQ %d on CPU %d\n", irq, raw_smp_processor_id());
1890 
1891 	if (likely(ACCESS_ONCE(efx->irq_soft_enabled))) {
1892 		/* Note test interrupts */
1893 		if (context->index == efx->irq_level)
1894 			efx->last_irq_cpu = raw_smp_processor_id();
1895 
1896 		/* Schedule processing of the channel */
1897 		efx_schedule_channel_irq(efx->channel[context->index]);
1898 	}
1899 
1900 	return IRQ_HANDLED;
1901 }
1902 
1903 static irqreturn_t efx_ef10_legacy_interrupt(int irq, void *dev_id)
1904 {
1905 	struct efx_nic *efx = dev_id;
1906 	bool soft_enabled = ACCESS_ONCE(efx->irq_soft_enabled);
1907 	struct efx_channel *channel;
1908 	efx_dword_t reg;
1909 	u32 queues;
1910 
1911 	/* Read the ISR which also ACKs the interrupts */
1912 	efx_readd(efx, &reg, ER_DZ_BIU_INT_ISR);
1913 	queues = EFX_DWORD_FIELD(reg, ERF_DZ_ISR_REG);
1914 
1915 	if (queues == 0)
1916 		return IRQ_NONE;
1917 
1918 	if (likely(soft_enabled)) {
1919 		/* Note test interrupts */
1920 		if (queues & (1U << efx->irq_level))
1921 			efx->last_irq_cpu = raw_smp_processor_id();
1922 
1923 		efx_for_each_channel(channel, efx) {
1924 			if (queues & 1)
1925 				efx_schedule_channel_irq(channel);
1926 			queues >>= 1;
1927 		}
1928 	}
1929 
1930 	netif_vdbg(efx, intr, efx->net_dev,
1931 		   "IRQ %d on CPU %d status " EFX_DWORD_FMT "\n",
1932 		   irq, raw_smp_processor_id(), EFX_DWORD_VAL(reg));
1933 
1934 	return IRQ_HANDLED;
1935 }
1936 
1937 static void efx_ef10_irq_test_generate(struct efx_nic *efx)
1938 {
1939 	MCDI_DECLARE_BUF(inbuf, MC_CMD_TRIGGER_INTERRUPT_IN_LEN);
1940 
1941 	BUILD_BUG_ON(MC_CMD_TRIGGER_INTERRUPT_OUT_LEN != 0);
1942 
1943 	MCDI_SET_DWORD(inbuf, TRIGGER_INTERRUPT_IN_INTR_LEVEL, efx->irq_level);
1944 	(void) efx_mcdi_rpc(efx, MC_CMD_TRIGGER_INTERRUPT,
1945 			    inbuf, sizeof(inbuf), NULL, 0, NULL);
1946 }
1947 
1948 static int efx_ef10_tx_probe(struct efx_tx_queue *tx_queue)
1949 {
1950 	return efx_nic_alloc_buffer(tx_queue->efx, &tx_queue->txd.buf,
1951 				    (tx_queue->ptr_mask + 1) *
1952 				    sizeof(efx_qword_t),
1953 				    GFP_KERNEL);
1954 }
1955 
1956 /* This writes to the TX_DESC_WPTR and also pushes data */
1957 static inline void efx_ef10_push_tx_desc(struct efx_tx_queue *tx_queue,
1958 					 const efx_qword_t *txd)
1959 {
1960 	unsigned int write_ptr;
1961 	efx_oword_t reg;
1962 
1963 	write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
1964 	EFX_POPULATE_OWORD_1(reg, ERF_DZ_TX_DESC_WPTR, write_ptr);
1965 	reg.qword[0] = *txd;
1966 	efx_writeo_page(tx_queue->efx, &reg,
1967 			ER_DZ_TX_DESC_UPD, tx_queue->queue);
1968 }
1969 
1970 static void efx_ef10_tx_init(struct efx_tx_queue *tx_queue)
1971 {
1972 	MCDI_DECLARE_BUF(inbuf, MC_CMD_INIT_TXQ_IN_LEN(EFX_MAX_DMAQ_SIZE * 8 /
1973 						       EFX_BUF_SIZE));
1974 	bool csum_offload = tx_queue->queue & EFX_TXQ_TYPE_OFFLOAD;
1975 	size_t entries = tx_queue->txd.buf.len / EFX_BUF_SIZE;
1976 	struct efx_channel *channel = tx_queue->channel;
1977 	struct efx_nic *efx = tx_queue->efx;
1978 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1979 	size_t inlen;
1980 	dma_addr_t dma_addr;
1981 	efx_qword_t *txd;
1982 	int rc;
1983 	int i;
1984 	BUILD_BUG_ON(MC_CMD_INIT_TXQ_OUT_LEN != 0);
1985 
1986 	MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_SIZE, tx_queue->ptr_mask + 1);
1987 	MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_TARGET_EVQ, channel->channel);
1988 	MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_LABEL, tx_queue->queue);
1989 	MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_INSTANCE, tx_queue->queue);
1990 	MCDI_POPULATE_DWORD_2(inbuf, INIT_TXQ_IN_FLAGS,
1991 			      INIT_TXQ_IN_FLAG_IP_CSUM_DIS, !csum_offload,
1992 			      INIT_TXQ_IN_FLAG_TCP_CSUM_DIS, !csum_offload);
1993 	MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_OWNER_ID, 0);
1994 	MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_PORT_ID, nic_data->vport_id);
1995 
1996 	dma_addr = tx_queue->txd.buf.dma_addr;
1997 
1998 	netif_dbg(efx, hw, efx->net_dev, "pushing TXQ %d. %zu entries (%llx)\n",
1999 		  tx_queue->queue, entries, (u64)dma_addr);
2000 
2001 	for (i = 0; i < entries; ++i) {
2002 		MCDI_SET_ARRAY_QWORD(inbuf, INIT_TXQ_IN_DMA_ADDR, i, dma_addr);
2003 		dma_addr += EFX_BUF_SIZE;
2004 	}
2005 
2006 	inlen = MC_CMD_INIT_TXQ_IN_LEN(entries);
2007 
2008 	rc = efx_mcdi_rpc(efx, MC_CMD_INIT_TXQ, inbuf, inlen,
2009 			  NULL, 0, NULL);
2010 	if (rc)
2011 		goto fail;
2012 
2013 	/* A previous user of this TX queue might have set us up the
2014 	 * bomb by writing a descriptor to the TX push collector but
2015 	 * not the doorbell.  (Each collector belongs to a port, not a
2016 	 * queue or function, so cannot easily be reset.)  We must
2017 	 * attempt to push a no-op descriptor in its place.
2018 	 */
2019 	tx_queue->buffer[0].flags = EFX_TX_BUF_OPTION;
2020 	tx_queue->insert_count = 1;
2021 	txd = efx_tx_desc(tx_queue, 0);
2022 	EFX_POPULATE_QWORD_4(*txd,
2023 			     ESF_DZ_TX_DESC_IS_OPT, true,
2024 			     ESF_DZ_TX_OPTION_TYPE,
2025 			     ESE_DZ_TX_OPTION_DESC_CRC_CSUM,
2026 			     ESF_DZ_TX_OPTION_UDP_TCP_CSUM, csum_offload,
2027 			     ESF_DZ_TX_OPTION_IP_CSUM, csum_offload);
2028 	tx_queue->write_count = 1;
2029 
2030 	if (nic_data->datapath_caps &
2031 	    (1 << MC_CMD_GET_CAPABILITIES_OUT_TX_TSO_LBN)) {
2032 		tx_queue->tso_version = 1;
2033 	}
2034 
2035 	wmb();
2036 	efx_ef10_push_tx_desc(tx_queue, txd);
2037 
2038 	return;
2039 
2040 fail:
2041 	netdev_WARN(efx->net_dev, "failed to initialise TXQ %d\n",
2042 		    tx_queue->queue);
2043 }
2044 
2045 static void efx_ef10_tx_fini(struct efx_tx_queue *tx_queue)
2046 {
2047 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FINI_TXQ_IN_LEN);
2048 	MCDI_DECLARE_BUF_ERR(outbuf);
2049 	struct efx_nic *efx = tx_queue->efx;
2050 	size_t outlen;
2051 	int rc;
2052 
2053 	MCDI_SET_DWORD(inbuf, FINI_TXQ_IN_INSTANCE,
2054 		       tx_queue->queue);
2055 
2056 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FINI_TXQ, inbuf, sizeof(inbuf),
2057 			  outbuf, sizeof(outbuf), &outlen);
2058 
2059 	if (rc && rc != -EALREADY)
2060 		goto fail;
2061 
2062 	return;
2063 
2064 fail:
2065 	efx_mcdi_display_error(efx, MC_CMD_FINI_TXQ, MC_CMD_FINI_TXQ_IN_LEN,
2066 			       outbuf, outlen, rc);
2067 }
2068 
2069 static void efx_ef10_tx_remove(struct efx_tx_queue *tx_queue)
2070 {
2071 	efx_nic_free_buffer(tx_queue->efx, &tx_queue->txd.buf);
2072 }
2073 
2074 /* This writes to the TX_DESC_WPTR; write pointer for TX descriptor ring */
2075 static inline void efx_ef10_notify_tx_desc(struct efx_tx_queue *tx_queue)
2076 {
2077 	unsigned int write_ptr;
2078 	efx_dword_t reg;
2079 
2080 	write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
2081 	EFX_POPULATE_DWORD_1(reg, ERF_DZ_TX_DESC_WPTR_DWORD, write_ptr);
2082 	efx_writed_page(tx_queue->efx, &reg,
2083 			ER_DZ_TX_DESC_UPD_DWORD, tx_queue->queue);
2084 }
2085 
2086 static void efx_ef10_tx_write(struct efx_tx_queue *tx_queue)
2087 {
2088 	unsigned int old_write_count = tx_queue->write_count;
2089 	struct efx_tx_buffer *buffer;
2090 	unsigned int write_ptr;
2091 	efx_qword_t *txd;
2092 
2093 	tx_queue->xmit_more_available = false;
2094 	if (unlikely(tx_queue->write_count == tx_queue->insert_count))
2095 		return;
2096 
2097 	do {
2098 		write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
2099 		buffer = &tx_queue->buffer[write_ptr];
2100 		txd = efx_tx_desc(tx_queue, write_ptr);
2101 		++tx_queue->write_count;
2102 
2103 		/* Create TX descriptor ring entry */
2104 		if (buffer->flags & EFX_TX_BUF_OPTION) {
2105 			*txd = buffer->option;
2106 		} else {
2107 			BUILD_BUG_ON(EFX_TX_BUF_CONT != 1);
2108 			EFX_POPULATE_QWORD_3(
2109 				*txd,
2110 				ESF_DZ_TX_KER_CONT,
2111 				buffer->flags & EFX_TX_BUF_CONT,
2112 				ESF_DZ_TX_KER_BYTE_CNT, buffer->len,
2113 				ESF_DZ_TX_KER_BUF_ADDR, buffer->dma_addr);
2114 		}
2115 	} while (tx_queue->write_count != tx_queue->insert_count);
2116 
2117 	wmb(); /* Ensure descriptors are written before they are fetched */
2118 
2119 	if (efx_nic_may_push_tx_desc(tx_queue, old_write_count)) {
2120 		txd = efx_tx_desc(tx_queue,
2121 				  old_write_count & tx_queue->ptr_mask);
2122 		efx_ef10_push_tx_desc(tx_queue, txd);
2123 		++tx_queue->pushes;
2124 	} else {
2125 		efx_ef10_notify_tx_desc(tx_queue);
2126 	}
2127 }
2128 
2129 static int efx_ef10_alloc_rss_context(struct efx_nic *efx, u32 *context,
2130 				      bool exclusive, unsigned *context_size)
2131 {
2132 	MCDI_DECLARE_BUF(inbuf, MC_CMD_RSS_CONTEXT_ALLOC_IN_LEN);
2133 	MCDI_DECLARE_BUF(outbuf, MC_CMD_RSS_CONTEXT_ALLOC_OUT_LEN);
2134 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2135 	size_t outlen;
2136 	int rc;
2137 	u32 alloc_type = exclusive ?
2138 				MC_CMD_RSS_CONTEXT_ALLOC_IN_TYPE_EXCLUSIVE :
2139 				MC_CMD_RSS_CONTEXT_ALLOC_IN_TYPE_SHARED;
2140 	unsigned rss_spread = exclusive ?
2141 				efx->rss_spread :
2142 				min(rounddown_pow_of_two(efx->rss_spread),
2143 				    EFX_EF10_MAX_SHARED_RSS_CONTEXT_SIZE);
2144 
2145 	if (!exclusive && rss_spread == 1) {
2146 		*context = EFX_EF10_RSS_CONTEXT_INVALID;
2147 		if (context_size)
2148 			*context_size = 1;
2149 		return 0;
2150 	}
2151 
2152 	if (nic_data->datapath_caps &
2153 	    1 << MC_CMD_GET_CAPABILITIES_OUT_RX_RSS_LIMITED_LBN)
2154 		return -EOPNOTSUPP;
2155 
2156 	MCDI_SET_DWORD(inbuf, RSS_CONTEXT_ALLOC_IN_UPSTREAM_PORT_ID,
2157 		       nic_data->vport_id);
2158 	MCDI_SET_DWORD(inbuf, RSS_CONTEXT_ALLOC_IN_TYPE, alloc_type);
2159 	MCDI_SET_DWORD(inbuf, RSS_CONTEXT_ALLOC_IN_NUM_QUEUES, rss_spread);
2160 
2161 	rc = efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_ALLOC, inbuf, sizeof(inbuf),
2162 		outbuf, sizeof(outbuf), &outlen);
2163 	if (rc != 0)
2164 		return rc;
2165 
2166 	if (outlen < MC_CMD_RSS_CONTEXT_ALLOC_OUT_LEN)
2167 		return -EIO;
2168 
2169 	*context = MCDI_DWORD(outbuf, RSS_CONTEXT_ALLOC_OUT_RSS_CONTEXT_ID);
2170 
2171 	if (context_size)
2172 		*context_size = rss_spread;
2173 
2174 	return 0;
2175 }
2176 
2177 static void efx_ef10_free_rss_context(struct efx_nic *efx, u32 context)
2178 {
2179 	MCDI_DECLARE_BUF(inbuf, MC_CMD_RSS_CONTEXT_FREE_IN_LEN);
2180 	int rc;
2181 
2182 	MCDI_SET_DWORD(inbuf, RSS_CONTEXT_FREE_IN_RSS_CONTEXT_ID,
2183 		       context);
2184 
2185 	rc = efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_FREE, inbuf, sizeof(inbuf),
2186 			    NULL, 0, NULL);
2187 	WARN_ON(rc != 0);
2188 }
2189 
2190 static int efx_ef10_populate_rss_table(struct efx_nic *efx, u32 context,
2191 				       const u32 *rx_indir_table)
2192 {
2193 	MCDI_DECLARE_BUF(tablebuf, MC_CMD_RSS_CONTEXT_SET_TABLE_IN_LEN);
2194 	MCDI_DECLARE_BUF(keybuf, MC_CMD_RSS_CONTEXT_SET_KEY_IN_LEN);
2195 	int i, rc;
2196 
2197 	MCDI_SET_DWORD(tablebuf, RSS_CONTEXT_SET_TABLE_IN_RSS_CONTEXT_ID,
2198 		       context);
2199 	BUILD_BUG_ON(ARRAY_SIZE(efx->rx_indir_table) !=
2200 		     MC_CMD_RSS_CONTEXT_SET_TABLE_IN_INDIRECTION_TABLE_LEN);
2201 
2202 	for (i = 0; i < ARRAY_SIZE(efx->rx_indir_table); ++i)
2203 		MCDI_PTR(tablebuf,
2204 			 RSS_CONTEXT_SET_TABLE_IN_INDIRECTION_TABLE)[i] =
2205 				(u8) rx_indir_table[i];
2206 
2207 	rc = efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_SET_TABLE, tablebuf,
2208 			  sizeof(tablebuf), NULL, 0, NULL);
2209 	if (rc != 0)
2210 		return rc;
2211 
2212 	MCDI_SET_DWORD(keybuf, RSS_CONTEXT_SET_KEY_IN_RSS_CONTEXT_ID,
2213 		       context);
2214 	BUILD_BUG_ON(ARRAY_SIZE(efx->rx_hash_key) !=
2215 		     MC_CMD_RSS_CONTEXT_SET_KEY_IN_TOEPLITZ_KEY_LEN);
2216 	for (i = 0; i < ARRAY_SIZE(efx->rx_hash_key); ++i)
2217 		MCDI_PTR(keybuf, RSS_CONTEXT_SET_KEY_IN_TOEPLITZ_KEY)[i] =
2218 			efx->rx_hash_key[i];
2219 
2220 	return efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_SET_KEY, keybuf,
2221 			    sizeof(keybuf), NULL, 0, NULL);
2222 }
2223 
2224 static void efx_ef10_rx_free_indir_table(struct efx_nic *efx)
2225 {
2226 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2227 
2228 	if (nic_data->rx_rss_context != EFX_EF10_RSS_CONTEXT_INVALID)
2229 		efx_ef10_free_rss_context(efx, nic_data->rx_rss_context);
2230 	nic_data->rx_rss_context = EFX_EF10_RSS_CONTEXT_INVALID;
2231 }
2232 
2233 static int efx_ef10_rx_push_shared_rss_config(struct efx_nic *efx,
2234 					      unsigned *context_size)
2235 {
2236 	u32 new_rx_rss_context;
2237 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2238 	int rc = efx_ef10_alloc_rss_context(efx, &new_rx_rss_context,
2239 					    false, context_size);
2240 
2241 	if (rc != 0)
2242 		return rc;
2243 
2244 	nic_data->rx_rss_context = new_rx_rss_context;
2245 	nic_data->rx_rss_context_exclusive = false;
2246 	efx_set_default_rx_indir_table(efx);
2247 	return 0;
2248 }
2249 
2250 static int efx_ef10_rx_push_exclusive_rss_config(struct efx_nic *efx,
2251 						 const u32 *rx_indir_table)
2252 {
2253 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2254 	int rc;
2255 	u32 new_rx_rss_context;
2256 
2257 	if (nic_data->rx_rss_context == EFX_EF10_RSS_CONTEXT_INVALID ||
2258 	    !nic_data->rx_rss_context_exclusive) {
2259 		rc = efx_ef10_alloc_rss_context(efx, &new_rx_rss_context,
2260 						true, NULL);
2261 		if (rc == -EOPNOTSUPP)
2262 			return rc;
2263 		else if (rc != 0)
2264 			goto fail1;
2265 	} else {
2266 		new_rx_rss_context = nic_data->rx_rss_context;
2267 	}
2268 
2269 	rc = efx_ef10_populate_rss_table(efx, new_rx_rss_context,
2270 					 rx_indir_table);
2271 	if (rc != 0)
2272 		goto fail2;
2273 
2274 	if (nic_data->rx_rss_context != new_rx_rss_context)
2275 		efx_ef10_rx_free_indir_table(efx);
2276 	nic_data->rx_rss_context = new_rx_rss_context;
2277 	nic_data->rx_rss_context_exclusive = true;
2278 	if (rx_indir_table != efx->rx_indir_table)
2279 		memcpy(efx->rx_indir_table, rx_indir_table,
2280 		       sizeof(efx->rx_indir_table));
2281 	return 0;
2282 
2283 fail2:
2284 	if (new_rx_rss_context != nic_data->rx_rss_context)
2285 		efx_ef10_free_rss_context(efx, new_rx_rss_context);
2286 fail1:
2287 	netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
2288 	return rc;
2289 }
2290 
2291 static int efx_ef10_pf_rx_push_rss_config(struct efx_nic *efx, bool user,
2292 					  const u32 *rx_indir_table)
2293 {
2294 	int rc;
2295 
2296 	if (efx->rss_spread == 1)
2297 		return 0;
2298 
2299 	rc = efx_ef10_rx_push_exclusive_rss_config(efx, rx_indir_table);
2300 
2301 	if (rc == -ENOBUFS && !user) {
2302 		unsigned context_size;
2303 		bool mismatch = false;
2304 		size_t i;
2305 
2306 		for (i = 0; i < ARRAY_SIZE(efx->rx_indir_table) && !mismatch;
2307 		     i++)
2308 			mismatch = rx_indir_table[i] !=
2309 				ethtool_rxfh_indir_default(i, efx->rss_spread);
2310 
2311 		rc = efx_ef10_rx_push_shared_rss_config(efx, &context_size);
2312 		if (rc == 0) {
2313 			if (context_size != efx->rss_spread)
2314 				netif_warn(efx, probe, efx->net_dev,
2315 					   "Could not allocate an exclusive RSS"
2316 					   " context; allocated a shared one of"
2317 					   " different size."
2318 					   " Wanted %u, got %u.\n",
2319 					   efx->rss_spread, context_size);
2320 			else if (mismatch)
2321 				netif_warn(efx, probe, efx->net_dev,
2322 					   "Could not allocate an exclusive RSS"
2323 					   " context; allocated a shared one but"
2324 					   " could not apply custom"
2325 					   " indirection.\n");
2326 			else
2327 				netif_info(efx, probe, efx->net_dev,
2328 					   "Could not allocate an exclusive RSS"
2329 					   " context; allocated a shared one.\n");
2330 		}
2331 	}
2332 	return rc;
2333 }
2334 
2335 static int efx_ef10_vf_rx_push_rss_config(struct efx_nic *efx, bool user,
2336 					  const u32 *rx_indir_table
2337 					  __attribute__ ((unused)))
2338 {
2339 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2340 
2341 	if (user)
2342 		return -EOPNOTSUPP;
2343 	if (nic_data->rx_rss_context != EFX_EF10_RSS_CONTEXT_INVALID)
2344 		return 0;
2345 	return efx_ef10_rx_push_shared_rss_config(efx, NULL);
2346 }
2347 
2348 static int efx_ef10_rx_probe(struct efx_rx_queue *rx_queue)
2349 {
2350 	return efx_nic_alloc_buffer(rx_queue->efx, &rx_queue->rxd.buf,
2351 				    (rx_queue->ptr_mask + 1) *
2352 				    sizeof(efx_qword_t),
2353 				    GFP_KERNEL);
2354 }
2355 
2356 static void efx_ef10_rx_init(struct efx_rx_queue *rx_queue)
2357 {
2358 	MCDI_DECLARE_BUF(inbuf,
2359 			 MC_CMD_INIT_RXQ_IN_LEN(EFX_MAX_DMAQ_SIZE * 8 /
2360 						EFX_BUF_SIZE));
2361 	struct efx_channel *channel = efx_rx_queue_channel(rx_queue);
2362 	size_t entries = rx_queue->rxd.buf.len / EFX_BUF_SIZE;
2363 	struct efx_nic *efx = rx_queue->efx;
2364 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2365 	size_t inlen;
2366 	dma_addr_t dma_addr;
2367 	int rc;
2368 	int i;
2369 	BUILD_BUG_ON(MC_CMD_INIT_RXQ_OUT_LEN != 0);
2370 
2371 	rx_queue->scatter_n = 0;
2372 	rx_queue->scatter_len = 0;
2373 
2374 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_SIZE, rx_queue->ptr_mask + 1);
2375 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_TARGET_EVQ, channel->channel);
2376 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_LABEL, efx_rx_queue_index(rx_queue));
2377 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_INSTANCE,
2378 		       efx_rx_queue_index(rx_queue));
2379 	MCDI_POPULATE_DWORD_2(inbuf, INIT_RXQ_IN_FLAGS,
2380 			      INIT_RXQ_IN_FLAG_PREFIX, 1,
2381 			      INIT_RXQ_IN_FLAG_TIMESTAMP, 1);
2382 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_OWNER_ID, 0);
2383 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_PORT_ID, nic_data->vport_id);
2384 
2385 	dma_addr = rx_queue->rxd.buf.dma_addr;
2386 
2387 	netif_dbg(efx, hw, efx->net_dev, "pushing RXQ %d. %zu entries (%llx)\n",
2388 		  efx_rx_queue_index(rx_queue), entries, (u64)dma_addr);
2389 
2390 	for (i = 0; i < entries; ++i) {
2391 		MCDI_SET_ARRAY_QWORD(inbuf, INIT_RXQ_IN_DMA_ADDR, i, dma_addr);
2392 		dma_addr += EFX_BUF_SIZE;
2393 	}
2394 
2395 	inlen = MC_CMD_INIT_RXQ_IN_LEN(entries);
2396 
2397 	rc = efx_mcdi_rpc(efx, MC_CMD_INIT_RXQ, inbuf, inlen,
2398 			  NULL, 0, NULL);
2399 	if (rc)
2400 		netdev_WARN(efx->net_dev, "failed to initialise RXQ %d\n",
2401 			    efx_rx_queue_index(rx_queue));
2402 }
2403 
2404 static void efx_ef10_rx_fini(struct efx_rx_queue *rx_queue)
2405 {
2406 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FINI_RXQ_IN_LEN);
2407 	MCDI_DECLARE_BUF_ERR(outbuf);
2408 	struct efx_nic *efx = rx_queue->efx;
2409 	size_t outlen;
2410 	int rc;
2411 
2412 	MCDI_SET_DWORD(inbuf, FINI_RXQ_IN_INSTANCE,
2413 		       efx_rx_queue_index(rx_queue));
2414 
2415 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FINI_RXQ, inbuf, sizeof(inbuf),
2416 			  outbuf, sizeof(outbuf), &outlen);
2417 
2418 	if (rc && rc != -EALREADY)
2419 		goto fail;
2420 
2421 	return;
2422 
2423 fail:
2424 	efx_mcdi_display_error(efx, MC_CMD_FINI_RXQ, MC_CMD_FINI_RXQ_IN_LEN,
2425 			       outbuf, outlen, rc);
2426 }
2427 
2428 static void efx_ef10_rx_remove(struct efx_rx_queue *rx_queue)
2429 {
2430 	efx_nic_free_buffer(rx_queue->efx, &rx_queue->rxd.buf);
2431 }
2432 
2433 /* This creates an entry in the RX descriptor queue */
2434 static inline void
2435 efx_ef10_build_rx_desc(struct efx_rx_queue *rx_queue, unsigned int index)
2436 {
2437 	struct efx_rx_buffer *rx_buf;
2438 	efx_qword_t *rxd;
2439 
2440 	rxd = efx_rx_desc(rx_queue, index);
2441 	rx_buf = efx_rx_buffer(rx_queue, index);
2442 	EFX_POPULATE_QWORD_2(*rxd,
2443 			     ESF_DZ_RX_KER_BYTE_CNT, rx_buf->len,
2444 			     ESF_DZ_RX_KER_BUF_ADDR, rx_buf->dma_addr);
2445 }
2446 
2447 static void efx_ef10_rx_write(struct efx_rx_queue *rx_queue)
2448 {
2449 	struct efx_nic *efx = rx_queue->efx;
2450 	unsigned int write_count;
2451 	efx_dword_t reg;
2452 
2453 	/* Firmware requires that RX_DESC_WPTR be a multiple of 8 */
2454 	write_count = rx_queue->added_count & ~7;
2455 	if (rx_queue->notified_count == write_count)
2456 		return;
2457 
2458 	do
2459 		efx_ef10_build_rx_desc(
2460 			rx_queue,
2461 			rx_queue->notified_count & rx_queue->ptr_mask);
2462 	while (++rx_queue->notified_count != write_count);
2463 
2464 	wmb();
2465 	EFX_POPULATE_DWORD_1(reg, ERF_DZ_RX_DESC_WPTR,
2466 			     write_count & rx_queue->ptr_mask);
2467 	efx_writed_page(efx, &reg, ER_DZ_RX_DESC_UPD,
2468 			efx_rx_queue_index(rx_queue));
2469 }
2470 
2471 static efx_mcdi_async_completer efx_ef10_rx_defer_refill_complete;
2472 
2473 static void efx_ef10_rx_defer_refill(struct efx_rx_queue *rx_queue)
2474 {
2475 	struct efx_channel *channel = efx_rx_queue_channel(rx_queue);
2476 	MCDI_DECLARE_BUF(inbuf, MC_CMD_DRIVER_EVENT_IN_LEN);
2477 	efx_qword_t event;
2478 
2479 	EFX_POPULATE_QWORD_2(event,
2480 			     ESF_DZ_EV_CODE, EFX_EF10_DRVGEN_EV,
2481 			     ESF_DZ_EV_DATA, EFX_EF10_REFILL);
2482 
2483 	MCDI_SET_DWORD(inbuf, DRIVER_EVENT_IN_EVQ, channel->channel);
2484 
2485 	/* MCDI_SET_QWORD is not appropriate here since EFX_POPULATE_* has
2486 	 * already swapped the data to little-endian order.
2487 	 */
2488 	memcpy(MCDI_PTR(inbuf, DRIVER_EVENT_IN_DATA), &event.u64[0],
2489 	       sizeof(efx_qword_t));
2490 
2491 	efx_mcdi_rpc_async(channel->efx, MC_CMD_DRIVER_EVENT,
2492 			   inbuf, sizeof(inbuf), 0,
2493 			   efx_ef10_rx_defer_refill_complete, 0);
2494 }
2495 
2496 static void
2497 efx_ef10_rx_defer_refill_complete(struct efx_nic *efx, unsigned long cookie,
2498 				  int rc, efx_dword_t *outbuf,
2499 				  size_t outlen_actual)
2500 {
2501 	/* nothing to do */
2502 }
2503 
2504 static int efx_ef10_ev_probe(struct efx_channel *channel)
2505 {
2506 	return efx_nic_alloc_buffer(channel->efx, &channel->eventq.buf,
2507 				    (channel->eventq_mask + 1) *
2508 				    sizeof(efx_qword_t),
2509 				    GFP_KERNEL);
2510 }
2511 
2512 static void efx_ef10_ev_fini(struct efx_channel *channel)
2513 {
2514 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FINI_EVQ_IN_LEN);
2515 	MCDI_DECLARE_BUF_ERR(outbuf);
2516 	struct efx_nic *efx = channel->efx;
2517 	size_t outlen;
2518 	int rc;
2519 
2520 	MCDI_SET_DWORD(inbuf, FINI_EVQ_IN_INSTANCE, channel->channel);
2521 
2522 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FINI_EVQ, inbuf, sizeof(inbuf),
2523 			  outbuf, sizeof(outbuf), &outlen);
2524 
2525 	if (rc && rc != -EALREADY)
2526 		goto fail;
2527 
2528 	return;
2529 
2530 fail:
2531 	efx_mcdi_display_error(efx, MC_CMD_FINI_EVQ, MC_CMD_FINI_EVQ_IN_LEN,
2532 			       outbuf, outlen, rc);
2533 }
2534 
2535 static int efx_ef10_ev_init(struct efx_channel *channel)
2536 {
2537 	MCDI_DECLARE_BUF(inbuf,
2538 			 MC_CMD_INIT_EVQ_IN_LEN(EFX_MAX_EVQ_SIZE * 8 /
2539 						EFX_BUF_SIZE));
2540 	MCDI_DECLARE_BUF(outbuf, MC_CMD_INIT_EVQ_OUT_LEN);
2541 	size_t entries = channel->eventq.buf.len / EFX_BUF_SIZE;
2542 	struct efx_nic *efx = channel->efx;
2543 	struct efx_ef10_nic_data *nic_data;
2544 	bool supports_rx_merge;
2545 	size_t inlen, outlen;
2546 	unsigned int enabled, implemented;
2547 	dma_addr_t dma_addr;
2548 	int rc;
2549 	int i;
2550 
2551 	nic_data = efx->nic_data;
2552 	supports_rx_merge =
2553 		!!(nic_data->datapath_caps &
2554 		   1 << MC_CMD_GET_CAPABILITIES_OUT_RX_BATCHING_LBN);
2555 
2556 	/* Fill event queue with all ones (i.e. empty events) */
2557 	memset(channel->eventq.buf.addr, 0xff, channel->eventq.buf.len);
2558 
2559 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_SIZE, channel->eventq_mask + 1);
2560 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_INSTANCE, channel->channel);
2561 	/* INIT_EVQ expects index in vector table, not absolute */
2562 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_IRQ_NUM, channel->channel);
2563 	MCDI_POPULATE_DWORD_4(inbuf, INIT_EVQ_IN_FLAGS,
2564 			      INIT_EVQ_IN_FLAG_INTERRUPTING, 1,
2565 			      INIT_EVQ_IN_FLAG_RX_MERGE, 1,
2566 			      INIT_EVQ_IN_FLAG_TX_MERGE, 1,
2567 			      INIT_EVQ_IN_FLAG_CUT_THRU, !supports_rx_merge);
2568 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_TMR_MODE,
2569 		       MC_CMD_INIT_EVQ_IN_TMR_MODE_DIS);
2570 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_TMR_LOAD, 0);
2571 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_TMR_RELOAD, 0);
2572 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_COUNT_MODE,
2573 		       MC_CMD_INIT_EVQ_IN_COUNT_MODE_DIS);
2574 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_COUNT_THRSHLD, 0);
2575 
2576 	dma_addr = channel->eventq.buf.dma_addr;
2577 	for (i = 0; i < entries; ++i) {
2578 		MCDI_SET_ARRAY_QWORD(inbuf, INIT_EVQ_IN_DMA_ADDR, i, dma_addr);
2579 		dma_addr += EFX_BUF_SIZE;
2580 	}
2581 
2582 	inlen = MC_CMD_INIT_EVQ_IN_LEN(entries);
2583 
2584 	rc = efx_mcdi_rpc(efx, MC_CMD_INIT_EVQ, inbuf, inlen,
2585 			  outbuf, sizeof(outbuf), &outlen);
2586 	/* IRQ return is ignored */
2587 	if (channel->channel || rc)
2588 		return rc;
2589 
2590 	/* Successfully created event queue on channel 0 */
2591 	rc = efx_mcdi_get_workarounds(efx, &implemented, &enabled);
2592 	if (rc == -ENOSYS) {
2593 		/* GET_WORKAROUNDS was implemented before the bug26807
2594 		 * workaround, thus the latter must be unavailable in this fw
2595 		 */
2596 		nic_data->workaround_26807 = false;
2597 		rc = 0;
2598 	} else if (rc) {
2599 		goto fail;
2600 	} else {
2601 		nic_data->workaround_26807 =
2602 			!!(enabled & MC_CMD_GET_WORKAROUNDS_OUT_BUG26807);
2603 
2604 		if (implemented & MC_CMD_GET_WORKAROUNDS_OUT_BUG26807 &&
2605 		    !nic_data->workaround_26807) {
2606 			unsigned int flags;
2607 
2608 			rc = efx_mcdi_set_workaround(efx,
2609 						     MC_CMD_WORKAROUND_BUG26807,
2610 						     true, &flags);
2611 
2612 			if (!rc) {
2613 				if (flags &
2614 				    1 << MC_CMD_WORKAROUND_EXT_OUT_FLR_DONE_LBN) {
2615 					netif_info(efx, drv, efx->net_dev,
2616 						   "other functions on NIC have been reset\n");
2617 
2618 					/* With MCFW v4.6.x and earlier, the
2619 					 * boot count will have incremented,
2620 					 * so re-read the warm_boot_count
2621 					 * value now to ensure this function
2622 					 * doesn't think it has changed next
2623 					 * time it checks.
2624 					 */
2625 					rc = efx_ef10_get_warm_boot_count(efx);
2626 					if (rc >= 0) {
2627 						nic_data->warm_boot_count = rc;
2628 						rc = 0;
2629 					}
2630 				}
2631 				nic_data->workaround_26807 = true;
2632 			} else if (rc == -EPERM) {
2633 				rc = 0;
2634 			}
2635 		}
2636 	}
2637 
2638 	if (!rc)
2639 		return 0;
2640 
2641 fail:
2642 	efx_ef10_ev_fini(channel);
2643 	return rc;
2644 }
2645 
2646 static void efx_ef10_ev_remove(struct efx_channel *channel)
2647 {
2648 	efx_nic_free_buffer(channel->efx, &channel->eventq.buf);
2649 }
2650 
2651 static void efx_ef10_handle_rx_wrong_queue(struct efx_rx_queue *rx_queue,
2652 					   unsigned int rx_queue_label)
2653 {
2654 	struct efx_nic *efx = rx_queue->efx;
2655 
2656 	netif_info(efx, hw, efx->net_dev,
2657 		   "rx event arrived on queue %d labeled as queue %u\n",
2658 		   efx_rx_queue_index(rx_queue), rx_queue_label);
2659 
2660 	efx_schedule_reset(efx, RESET_TYPE_DISABLE);
2661 }
2662 
2663 static void
2664 efx_ef10_handle_rx_bad_lbits(struct efx_rx_queue *rx_queue,
2665 			     unsigned int actual, unsigned int expected)
2666 {
2667 	unsigned int dropped = (actual - expected) & rx_queue->ptr_mask;
2668 	struct efx_nic *efx = rx_queue->efx;
2669 
2670 	netif_info(efx, hw, efx->net_dev,
2671 		   "dropped %d events (index=%d expected=%d)\n",
2672 		   dropped, actual, expected);
2673 
2674 	efx_schedule_reset(efx, RESET_TYPE_DISABLE);
2675 }
2676 
2677 /* partially received RX was aborted. clean up. */
2678 static void efx_ef10_handle_rx_abort(struct efx_rx_queue *rx_queue)
2679 {
2680 	unsigned int rx_desc_ptr;
2681 
2682 	netif_dbg(rx_queue->efx, hw, rx_queue->efx->net_dev,
2683 		  "scattered RX aborted (dropping %u buffers)\n",
2684 		  rx_queue->scatter_n);
2685 
2686 	rx_desc_ptr = rx_queue->removed_count & rx_queue->ptr_mask;
2687 
2688 	efx_rx_packet(rx_queue, rx_desc_ptr, rx_queue->scatter_n,
2689 		      0, EFX_RX_PKT_DISCARD);
2690 
2691 	rx_queue->removed_count += rx_queue->scatter_n;
2692 	rx_queue->scatter_n = 0;
2693 	rx_queue->scatter_len = 0;
2694 	++efx_rx_queue_channel(rx_queue)->n_rx_nodesc_trunc;
2695 }
2696 
2697 static int efx_ef10_handle_rx_event(struct efx_channel *channel,
2698 				    const efx_qword_t *event)
2699 {
2700 	unsigned int rx_bytes, next_ptr_lbits, rx_queue_label, rx_l4_class;
2701 	unsigned int n_descs, n_packets, i;
2702 	struct efx_nic *efx = channel->efx;
2703 	struct efx_rx_queue *rx_queue;
2704 	bool rx_cont;
2705 	u16 flags = 0;
2706 
2707 	if (unlikely(ACCESS_ONCE(efx->reset_pending)))
2708 		return 0;
2709 
2710 	/* Basic packet information */
2711 	rx_bytes = EFX_QWORD_FIELD(*event, ESF_DZ_RX_BYTES);
2712 	next_ptr_lbits = EFX_QWORD_FIELD(*event, ESF_DZ_RX_DSC_PTR_LBITS);
2713 	rx_queue_label = EFX_QWORD_FIELD(*event, ESF_DZ_RX_QLABEL);
2714 	rx_l4_class = EFX_QWORD_FIELD(*event, ESF_DZ_RX_L4_CLASS);
2715 	rx_cont = EFX_QWORD_FIELD(*event, ESF_DZ_RX_CONT);
2716 
2717 	if (EFX_QWORD_FIELD(*event, ESF_DZ_RX_DROP_EVENT))
2718 		netdev_WARN(efx->net_dev, "saw RX_DROP_EVENT: event="
2719 			    EFX_QWORD_FMT "\n",
2720 			    EFX_QWORD_VAL(*event));
2721 
2722 	rx_queue = efx_channel_get_rx_queue(channel);
2723 
2724 	if (unlikely(rx_queue_label != efx_rx_queue_index(rx_queue)))
2725 		efx_ef10_handle_rx_wrong_queue(rx_queue, rx_queue_label);
2726 
2727 	n_descs = ((next_ptr_lbits - rx_queue->removed_count) &
2728 		   ((1 << ESF_DZ_RX_DSC_PTR_LBITS_WIDTH) - 1));
2729 
2730 	if (n_descs != rx_queue->scatter_n + 1) {
2731 		struct efx_ef10_nic_data *nic_data = efx->nic_data;
2732 
2733 		/* detect rx abort */
2734 		if (unlikely(n_descs == rx_queue->scatter_n)) {
2735 			if (rx_queue->scatter_n == 0 || rx_bytes != 0)
2736 				netdev_WARN(efx->net_dev,
2737 					    "invalid RX abort: scatter_n=%u event="
2738 					    EFX_QWORD_FMT "\n",
2739 					    rx_queue->scatter_n,
2740 					    EFX_QWORD_VAL(*event));
2741 			efx_ef10_handle_rx_abort(rx_queue);
2742 			return 0;
2743 		}
2744 
2745 		/* Check that RX completion merging is valid, i.e.
2746 		 * the current firmware supports it and this is a
2747 		 * non-scattered packet.
2748 		 */
2749 		if (!(nic_data->datapath_caps &
2750 		      (1 << MC_CMD_GET_CAPABILITIES_OUT_RX_BATCHING_LBN)) ||
2751 		    rx_queue->scatter_n != 0 || rx_cont) {
2752 			efx_ef10_handle_rx_bad_lbits(
2753 				rx_queue, next_ptr_lbits,
2754 				(rx_queue->removed_count +
2755 				 rx_queue->scatter_n + 1) &
2756 				((1 << ESF_DZ_RX_DSC_PTR_LBITS_WIDTH) - 1));
2757 			return 0;
2758 		}
2759 
2760 		/* Merged completion for multiple non-scattered packets */
2761 		rx_queue->scatter_n = 1;
2762 		rx_queue->scatter_len = 0;
2763 		n_packets = n_descs;
2764 		++channel->n_rx_merge_events;
2765 		channel->n_rx_merge_packets += n_packets;
2766 		flags |= EFX_RX_PKT_PREFIX_LEN;
2767 	} else {
2768 		++rx_queue->scatter_n;
2769 		rx_queue->scatter_len += rx_bytes;
2770 		if (rx_cont)
2771 			return 0;
2772 		n_packets = 1;
2773 	}
2774 
2775 	if (unlikely(EFX_QWORD_FIELD(*event, ESF_DZ_RX_ECRC_ERR)))
2776 		flags |= EFX_RX_PKT_DISCARD;
2777 
2778 	if (unlikely(EFX_QWORD_FIELD(*event, ESF_DZ_RX_IPCKSUM_ERR))) {
2779 		channel->n_rx_ip_hdr_chksum_err += n_packets;
2780 	} else if (unlikely(EFX_QWORD_FIELD(*event,
2781 					    ESF_DZ_RX_TCPUDP_CKSUM_ERR))) {
2782 		channel->n_rx_tcp_udp_chksum_err += n_packets;
2783 	} else if (rx_l4_class == ESE_DZ_L4_CLASS_TCP ||
2784 		   rx_l4_class == ESE_DZ_L4_CLASS_UDP) {
2785 		flags |= EFX_RX_PKT_CSUMMED;
2786 	}
2787 
2788 	if (rx_l4_class == ESE_DZ_L4_CLASS_TCP)
2789 		flags |= EFX_RX_PKT_TCP;
2790 
2791 	channel->irq_mod_score += 2 * n_packets;
2792 
2793 	/* Handle received packet(s) */
2794 	for (i = 0; i < n_packets; i++) {
2795 		efx_rx_packet(rx_queue,
2796 			      rx_queue->removed_count & rx_queue->ptr_mask,
2797 			      rx_queue->scatter_n, rx_queue->scatter_len,
2798 			      flags);
2799 		rx_queue->removed_count += rx_queue->scatter_n;
2800 	}
2801 
2802 	rx_queue->scatter_n = 0;
2803 	rx_queue->scatter_len = 0;
2804 
2805 	return n_packets;
2806 }
2807 
2808 static int
2809 efx_ef10_handle_tx_event(struct efx_channel *channel, efx_qword_t *event)
2810 {
2811 	struct efx_nic *efx = channel->efx;
2812 	struct efx_tx_queue *tx_queue;
2813 	unsigned int tx_ev_desc_ptr;
2814 	unsigned int tx_ev_q_label;
2815 	int tx_descs = 0;
2816 
2817 	if (unlikely(ACCESS_ONCE(efx->reset_pending)))
2818 		return 0;
2819 
2820 	if (unlikely(EFX_QWORD_FIELD(*event, ESF_DZ_TX_DROP_EVENT)))
2821 		return 0;
2822 
2823 	/* Transmit completion */
2824 	tx_ev_desc_ptr = EFX_QWORD_FIELD(*event, ESF_DZ_TX_DESCR_INDX);
2825 	tx_ev_q_label = EFX_QWORD_FIELD(*event, ESF_DZ_TX_QLABEL);
2826 	tx_queue = efx_channel_get_tx_queue(channel,
2827 					    tx_ev_q_label % EFX_TXQ_TYPES);
2828 	tx_descs = ((tx_ev_desc_ptr + 1 - tx_queue->read_count) &
2829 		    tx_queue->ptr_mask);
2830 	efx_xmit_done(tx_queue, tx_ev_desc_ptr & tx_queue->ptr_mask);
2831 
2832 	return tx_descs;
2833 }
2834 
2835 static void
2836 efx_ef10_handle_driver_event(struct efx_channel *channel, efx_qword_t *event)
2837 {
2838 	struct efx_nic *efx = channel->efx;
2839 	int subcode;
2840 
2841 	subcode = EFX_QWORD_FIELD(*event, ESF_DZ_DRV_SUB_CODE);
2842 
2843 	switch (subcode) {
2844 	case ESE_DZ_DRV_TIMER_EV:
2845 	case ESE_DZ_DRV_WAKE_UP_EV:
2846 		break;
2847 	case ESE_DZ_DRV_START_UP_EV:
2848 		/* event queue init complete. ok. */
2849 		break;
2850 	default:
2851 		netif_err(efx, hw, efx->net_dev,
2852 			  "channel %d unknown driver event type %d"
2853 			  " (data " EFX_QWORD_FMT ")\n",
2854 			  channel->channel, subcode,
2855 			  EFX_QWORD_VAL(*event));
2856 
2857 	}
2858 }
2859 
2860 static void efx_ef10_handle_driver_generated_event(struct efx_channel *channel,
2861 						   efx_qword_t *event)
2862 {
2863 	struct efx_nic *efx = channel->efx;
2864 	u32 subcode;
2865 
2866 	subcode = EFX_QWORD_FIELD(*event, EFX_DWORD_0);
2867 
2868 	switch (subcode) {
2869 	case EFX_EF10_TEST:
2870 		channel->event_test_cpu = raw_smp_processor_id();
2871 		break;
2872 	case EFX_EF10_REFILL:
2873 		/* The queue must be empty, so we won't receive any rx
2874 		 * events, so efx_process_channel() won't refill the
2875 		 * queue. Refill it here
2876 		 */
2877 		efx_fast_push_rx_descriptors(&channel->rx_queue, true);
2878 		break;
2879 	default:
2880 		netif_err(efx, hw, efx->net_dev,
2881 			  "channel %d unknown driver event type %u"
2882 			  " (data " EFX_QWORD_FMT ")\n",
2883 			  channel->channel, (unsigned) subcode,
2884 			  EFX_QWORD_VAL(*event));
2885 	}
2886 }
2887 
2888 static int efx_ef10_ev_process(struct efx_channel *channel, int quota)
2889 {
2890 	struct efx_nic *efx = channel->efx;
2891 	efx_qword_t event, *p_event;
2892 	unsigned int read_ptr;
2893 	int ev_code;
2894 	int tx_descs = 0;
2895 	int spent = 0;
2896 
2897 	if (quota <= 0)
2898 		return spent;
2899 
2900 	read_ptr = channel->eventq_read_ptr;
2901 
2902 	for (;;) {
2903 		p_event = efx_event(channel, read_ptr);
2904 		event = *p_event;
2905 
2906 		if (!efx_event_present(&event))
2907 			break;
2908 
2909 		EFX_SET_QWORD(*p_event);
2910 
2911 		++read_ptr;
2912 
2913 		ev_code = EFX_QWORD_FIELD(event, ESF_DZ_EV_CODE);
2914 
2915 		netif_vdbg(efx, drv, efx->net_dev,
2916 			   "processing event on %d " EFX_QWORD_FMT "\n",
2917 			   channel->channel, EFX_QWORD_VAL(event));
2918 
2919 		switch (ev_code) {
2920 		case ESE_DZ_EV_CODE_MCDI_EV:
2921 			efx_mcdi_process_event(channel, &event);
2922 			break;
2923 		case ESE_DZ_EV_CODE_RX_EV:
2924 			spent += efx_ef10_handle_rx_event(channel, &event);
2925 			if (spent >= quota) {
2926 				/* XXX can we split a merged event to
2927 				 * avoid going over-quota?
2928 				 */
2929 				spent = quota;
2930 				goto out;
2931 			}
2932 			break;
2933 		case ESE_DZ_EV_CODE_TX_EV:
2934 			tx_descs += efx_ef10_handle_tx_event(channel, &event);
2935 			if (tx_descs > efx->txq_entries) {
2936 				spent = quota;
2937 				goto out;
2938 			} else if (++spent == quota) {
2939 				goto out;
2940 			}
2941 			break;
2942 		case ESE_DZ_EV_CODE_DRIVER_EV:
2943 			efx_ef10_handle_driver_event(channel, &event);
2944 			if (++spent == quota)
2945 				goto out;
2946 			break;
2947 		case EFX_EF10_DRVGEN_EV:
2948 			efx_ef10_handle_driver_generated_event(channel, &event);
2949 			break;
2950 		default:
2951 			netif_err(efx, hw, efx->net_dev,
2952 				  "channel %d unknown event type %d"
2953 				  " (data " EFX_QWORD_FMT ")\n",
2954 				  channel->channel, ev_code,
2955 				  EFX_QWORD_VAL(event));
2956 		}
2957 	}
2958 
2959 out:
2960 	channel->eventq_read_ptr = read_ptr;
2961 	return spent;
2962 }
2963 
2964 static void efx_ef10_ev_read_ack(struct efx_channel *channel)
2965 {
2966 	struct efx_nic *efx = channel->efx;
2967 	efx_dword_t rptr;
2968 
2969 	if (EFX_EF10_WORKAROUND_35388(efx)) {
2970 		BUILD_BUG_ON(EFX_MIN_EVQ_SIZE <
2971 			     (1 << ERF_DD_EVQ_IND_RPTR_WIDTH));
2972 		BUILD_BUG_ON(EFX_MAX_EVQ_SIZE >
2973 			     (1 << 2 * ERF_DD_EVQ_IND_RPTR_WIDTH));
2974 
2975 		EFX_POPULATE_DWORD_2(rptr, ERF_DD_EVQ_IND_RPTR_FLAGS,
2976 				     EFE_DD_EVQ_IND_RPTR_FLAGS_HIGH,
2977 				     ERF_DD_EVQ_IND_RPTR,
2978 				     (channel->eventq_read_ptr &
2979 				      channel->eventq_mask) >>
2980 				     ERF_DD_EVQ_IND_RPTR_WIDTH);
2981 		efx_writed_page(efx, &rptr, ER_DD_EVQ_INDIRECT,
2982 				channel->channel);
2983 		EFX_POPULATE_DWORD_2(rptr, ERF_DD_EVQ_IND_RPTR_FLAGS,
2984 				     EFE_DD_EVQ_IND_RPTR_FLAGS_LOW,
2985 				     ERF_DD_EVQ_IND_RPTR,
2986 				     channel->eventq_read_ptr &
2987 				     ((1 << ERF_DD_EVQ_IND_RPTR_WIDTH) - 1));
2988 		efx_writed_page(efx, &rptr, ER_DD_EVQ_INDIRECT,
2989 				channel->channel);
2990 	} else {
2991 		EFX_POPULATE_DWORD_1(rptr, ERF_DZ_EVQ_RPTR,
2992 				     channel->eventq_read_ptr &
2993 				     channel->eventq_mask);
2994 		efx_writed_page(efx, &rptr, ER_DZ_EVQ_RPTR, channel->channel);
2995 	}
2996 }
2997 
2998 static void efx_ef10_ev_test_generate(struct efx_channel *channel)
2999 {
3000 	MCDI_DECLARE_BUF(inbuf, MC_CMD_DRIVER_EVENT_IN_LEN);
3001 	struct efx_nic *efx = channel->efx;
3002 	efx_qword_t event;
3003 	int rc;
3004 
3005 	EFX_POPULATE_QWORD_2(event,
3006 			     ESF_DZ_EV_CODE, EFX_EF10_DRVGEN_EV,
3007 			     ESF_DZ_EV_DATA, EFX_EF10_TEST);
3008 
3009 	MCDI_SET_DWORD(inbuf, DRIVER_EVENT_IN_EVQ, channel->channel);
3010 
3011 	/* MCDI_SET_QWORD is not appropriate here since EFX_POPULATE_* has
3012 	 * already swapped the data to little-endian order.
3013 	 */
3014 	memcpy(MCDI_PTR(inbuf, DRIVER_EVENT_IN_DATA), &event.u64[0],
3015 	       sizeof(efx_qword_t));
3016 
3017 	rc = efx_mcdi_rpc(efx, MC_CMD_DRIVER_EVENT, inbuf, sizeof(inbuf),
3018 			  NULL, 0, NULL);
3019 	if (rc != 0)
3020 		goto fail;
3021 
3022 	return;
3023 
3024 fail:
3025 	WARN_ON(true);
3026 	netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
3027 }
3028 
3029 void efx_ef10_handle_drain_event(struct efx_nic *efx)
3030 {
3031 	if (atomic_dec_and_test(&efx->active_queues))
3032 		wake_up(&efx->flush_wq);
3033 
3034 	WARN_ON(atomic_read(&efx->active_queues) < 0);
3035 }
3036 
3037 static int efx_ef10_fini_dmaq(struct efx_nic *efx)
3038 {
3039 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
3040 	struct efx_channel *channel;
3041 	struct efx_tx_queue *tx_queue;
3042 	struct efx_rx_queue *rx_queue;
3043 	int pending;
3044 
3045 	/* If the MC has just rebooted, the TX/RX queues will have already been
3046 	 * torn down, but efx->active_queues needs to be set to zero.
3047 	 */
3048 	if (nic_data->must_realloc_vis) {
3049 		atomic_set(&efx->active_queues, 0);
3050 		return 0;
3051 	}
3052 
3053 	/* Do not attempt to write to the NIC during EEH recovery */
3054 	if (efx->state != STATE_RECOVERY) {
3055 		efx_for_each_channel(channel, efx) {
3056 			efx_for_each_channel_rx_queue(rx_queue, channel)
3057 				efx_ef10_rx_fini(rx_queue);
3058 			efx_for_each_channel_tx_queue(tx_queue, channel)
3059 				efx_ef10_tx_fini(tx_queue);
3060 		}
3061 
3062 		wait_event_timeout(efx->flush_wq,
3063 				   atomic_read(&efx->active_queues) == 0,
3064 				   msecs_to_jiffies(EFX_MAX_FLUSH_TIME));
3065 		pending = atomic_read(&efx->active_queues);
3066 		if (pending) {
3067 			netif_err(efx, hw, efx->net_dev, "failed to flush %d queues\n",
3068 				  pending);
3069 			return -ETIMEDOUT;
3070 		}
3071 	}
3072 
3073 	return 0;
3074 }
3075 
3076 static void efx_ef10_prepare_flr(struct efx_nic *efx)
3077 {
3078 	atomic_set(&efx->active_queues, 0);
3079 }
3080 
3081 static bool efx_ef10_filter_equal(const struct efx_filter_spec *left,
3082 				  const struct efx_filter_spec *right)
3083 {
3084 	if ((left->match_flags ^ right->match_flags) |
3085 	    ((left->flags ^ right->flags) &
3086 	     (EFX_FILTER_FLAG_RX | EFX_FILTER_FLAG_TX)))
3087 		return false;
3088 
3089 	return memcmp(&left->outer_vid, &right->outer_vid,
3090 		      sizeof(struct efx_filter_spec) -
3091 		      offsetof(struct efx_filter_spec, outer_vid)) == 0;
3092 }
3093 
3094 static unsigned int efx_ef10_filter_hash(const struct efx_filter_spec *spec)
3095 {
3096 	BUILD_BUG_ON(offsetof(struct efx_filter_spec, outer_vid) & 3);
3097 	return jhash2((const u32 *)&spec->outer_vid,
3098 		      (sizeof(struct efx_filter_spec) -
3099 		       offsetof(struct efx_filter_spec, outer_vid)) / 4,
3100 		      0);
3101 	/* XXX should we randomise the initval? */
3102 }
3103 
3104 /* Decide whether a filter should be exclusive or else should allow
3105  * delivery to additional recipients.  Currently we decide that
3106  * filters for specific local unicast MAC and IP addresses are
3107  * exclusive.
3108  */
3109 static bool efx_ef10_filter_is_exclusive(const struct efx_filter_spec *spec)
3110 {
3111 	if (spec->match_flags & EFX_FILTER_MATCH_LOC_MAC &&
3112 	    !is_multicast_ether_addr(spec->loc_mac))
3113 		return true;
3114 
3115 	if ((spec->match_flags &
3116 	     (EFX_FILTER_MATCH_ETHER_TYPE | EFX_FILTER_MATCH_LOC_HOST)) ==
3117 	    (EFX_FILTER_MATCH_ETHER_TYPE | EFX_FILTER_MATCH_LOC_HOST)) {
3118 		if (spec->ether_type == htons(ETH_P_IP) &&
3119 		    !ipv4_is_multicast(spec->loc_host[0]))
3120 			return true;
3121 		if (spec->ether_type == htons(ETH_P_IPV6) &&
3122 		    ((const u8 *)spec->loc_host)[0] != 0xff)
3123 			return true;
3124 	}
3125 
3126 	return false;
3127 }
3128 
3129 static struct efx_filter_spec *
3130 efx_ef10_filter_entry_spec(const struct efx_ef10_filter_table *table,
3131 			   unsigned int filter_idx)
3132 {
3133 	return (struct efx_filter_spec *)(table->entry[filter_idx].spec &
3134 					  ~EFX_EF10_FILTER_FLAGS);
3135 }
3136 
3137 static unsigned int
3138 efx_ef10_filter_entry_flags(const struct efx_ef10_filter_table *table,
3139 			   unsigned int filter_idx)
3140 {
3141 	return table->entry[filter_idx].spec & EFX_EF10_FILTER_FLAGS;
3142 }
3143 
3144 static void
3145 efx_ef10_filter_set_entry(struct efx_ef10_filter_table *table,
3146 			  unsigned int filter_idx,
3147 			  const struct efx_filter_spec *spec,
3148 			  unsigned int flags)
3149 {
3150 	table->entry[filter_idx].spec =	(unsigned long)spec | flags;
3151 }
3152 
3153 static void efx_ef10_filter_push_prep(struct efx_nic *efx,
3154 				      const struct efx_filter_spec *spec,
3155 				      efx_dword_t *inbuf, u64 handle,
3156 				      bool replacing)
3157 {
3158 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
3159 	u32 flags = spec->flags;
3160 
3161 	memset(inbuf, 0, MC_CMD_FILTER_OP_IN_LEN);
3162 
3163 	/* Remove RSS flag if we don't have an RSS context. */
3164 	if (flags & EFX_FILTER_FLAG_RX_RSS &&
3165 	    spec->rss_context == EFX_FILTER_RSS_CONTEXT_DEFAULT &&
3166 	    nic_data->rx_rss_context == EFX_EF10_RSS_CONTEXT_INVALID)
3167 		flags &= ~EFX_FILTER_FLAG_RX_RSS;
3168 
3169 	if (replacing) {
3170 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
3171 			       MC_CMD_FILTER_OP_IN_OP_REPLACE);
3172 		MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE, handle);
3173 	} else {
3174 		u32 match_fields = 0;
3175 
3176 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
3177 			       efx_ef10_filter_is_exclusive(spec) ?
3178 			       MC_CMD_FILTER_OP_IN_OP_INSERT :
3179 			       MC_CMD_FILTER_OP_IN_OP_SUBSCRIBE);
3180 
3181 		/* Convert match flags and values.  Unlike almost
3182 		 * everything else in MCDI, these fields are in
3183 		 * network byte order.
3184 		 */
3185 		if (spec->match_flags & EFX_FILTER_MATCH_LOC_MAC_IG)
3186 			match_fields |=
3187 				is_multicast_ether_addr(spec->loc_mac) ?
3188 				1 << MC_CMD_FILTER_OP_IN_MATCH_UNKNOWN_MCAST_DST_LBN :
3189 				1 << MC_CMD_FILTER_OP_IN_MATCH_UNKNOWN_UCAST_DST_LBN;
3190 #define COPY_FIELD(gen_flag, gen_field, mcdi_field)			     \
3191 		if (spec->match_flags & EFX_FILTER_MATCH_ ## gen_flag) {     \
3192 			match_fields |=					     \
3193 				1 << MC_CMD_FILTER_OP_IN_MATCH_ ##	     \
3194 				mcdi_field ## _LBN;			     \
3195 			BUILD_BUG_ON(					     \
3196 				MC_CMD_FILTER_OP_IN_ ## mcdi_field ## _LEN < \
3197 				sizeof(spec->gen_field));		     \
3198 			memcpy(MCDI_PTR(inbuf, FILTER_OP_IN_ ##	mcdi_field), \
3199 			       &spec->gen_field, sizeof(spec->gen_field));   \
3200 		}
3201 		COPY_FIELD(REM_HOST, rem_host, SRC_IP);
3202 		COPY_FIELD(LOC_HOST, loc_host, DST_IP);
3203 		COPY_FIELD(REM_MAC, rem_mac, SRC_MAC);
3204 		COPY_FIELD(REM_PORT, rem_port, SRC_PORT);
3205 		COPY_FIELD(LOC_MAC, loc_mac, DST_MAC);
3206 		COPY_FIELD(LOC_PORT, loc_port, DST_PORT);
3207 		COPY_FIELD(ETHER_TYPE, ether_type, ETHER_TYPE);
3208 		COPY_FIELD(INNER_VID, inner_vid, INNER_VLAN);
3209 		COPY_FIELD(OUTER_VID, outer_vid, OUTER_VLAN);
3210 		COPY_FIELD(IP_PROTO, ip_proto, IP_PROTO);
3211 #undef COPY_FIELD
3212 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_MATCH_FIELDS,
3213 			       match_fields);
3214 	}
3215 
3216 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_PORT_ID, nic_data->vport_id);
3217 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_DEST,
3218 		       spec->dmaq_id == EFX_FILTER_RX_DMAQ_ID_DROP ?
3219 		       MC_CMD_FILTER_OP_IN_RX_DEST_DROP :
3220 		       MC_CMD_FILTER_OP_IN_RX_DEST_HOST);
3221 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_TX_DOMAIN, 0);
3222 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_TX_DEST,
3223 		       MC_CMD_FILTER_OP_IN_TX_DEST_DEFAULT);
3224 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_QUEUE,
3225 		       spec->dmaq_id == EFX_FILTER_RX_DMAQ_ID_DROP ?
3226 		       0 : spec->dmaq_id);
3227 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_MODE,
3228 		       (flags & EFX_FILTER_FLAG_RX_RSS) ?
3229 		       MC_CMD_FILTER_OP_IN_RX_MODE_RSS :
3230 		       MC_CMD_FILTER_OP_IN_RX_MODE_SIMPLE);
3231 	if (flags & EFX_FILTER_FLAG_RX_RSS)
3232 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_CONTEXT,
3233 			       spec->rss_context !=
3234 			       EFX_FILTER_RSS_CONTEXT_DEFAULT ?
3235 			       spec->rss_context : nic_data->rx_rss_context);
3236 }
3237 
3238 static int efx_ef10_filter_push(struct efx_nic *efx,
3239 				const struct efx_filter_spec *spec,
3240 				u64 *handle, bool replacing)
3241 {
3242 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_IN_LEN);
3243 	MCDI_DECLARE_BUF(outbuf, MC_CMD_FILTER_OP_OUT_LEN);
3244 	int rc;
3245 
3246 	efx_ef10_filter_push_prep(efx, spec, inbuf, *handle, replacing);
3247 	rc = efx_mcdi_rpc(efx, MC_CMD_FILTER_OP, inbuf, sizeof(inbuf),
3248 			  outbuf, sizeof(outbuf), NULL);
3249 	if (rc == 0)
3250 		*handle = MCDI_QWORD(outbuf, FILTER_OP_OUT_HANDLE);
3251 	if (rc == -ENOSPC)
3252 		rc = -EBUSY; /* to match efx_farch_filter_insert() */
3253 	return rc;
3254 }
3255 
3256 static u32 efx_ef10_filter_mcdi_flags_from_spec(const struct efx_filter_spec *spec)
3257 {
3258 	unsigned int match_flags = spec->match_flags;
3259 	u32 mcdi_flags = 0;
3260 
3261 	if (match_flags & EFX_FILTER_MATCH_LOC_MAC_IG) {
3262 		match_flags &= ~EFX_FILTER_MATCH_LOC_MAC_IG;
3263 		mcdi_flags |=
3264 			is_multicast_ether_addr(spec->loc_mac) ?
3265 			(1 << MC_CMD_FILTER_OP_IN_MATCH_UNKNOWN_MCAST_DST_LBN) :
3266 			(1 << MC_CMD_FILTER_OP_IN_MATCH_UNKNOWN_UCAST_DST_LBN);
3267 	}
3268 
3269 #define MAP_FILTER_TO_MCDI_FLAG(gen_flag, mcdi_field) {			\
3270 		unsigned int old_match_flags = match_flags;		\
3271 		match_flags &= ~EFX_FILTER_MATCH_ ## gen_flag;		\
3272 		if (match_flags != old_match_flags)			\
3273 			mcdi_flags |=					\
3274 				(1 << MC_CMD_FILTER_OP_IN_MATCH_ ##	\
3275 				 mcdi_field ## _LBN);			\
3276 	}
3277 	MAP_FILTER_TO_MCDI_FLAG(REM_HOST, SRC_IP);
3278 	MAP_FILTER_TO_MCDI_FLAG(LOC_HOST, DST_IP);
3279 	MAP_FILTER_TO_MCDI_FLAG(REM_MAC, SRC_MAC);
3280 	MAP_FILTER_TO_MCDI_FLAG(REM_PORT, SRC_PORT);
3281 	MAP_FILTER_TO_MCDI_FLAG(LOC_MAC, DST_MAC);
3282 	MAP_FILTER_TO_MCDI_FLAG(LOC_PORT, DST_PORT);
3283 	MAP_FILTER_TO_MCDI_FLAG(ETHER_TYPE, ETHER_TYPE);
3284 	MAP_FILTER_TO_MCDI_FLAG(INNER_VID, INNER_VLAN);
3285 	MAP_FILTER_TO_MCDI_FLAG(OUTER_VID, OUTER_VLAN);
3286 	MAP_FILTER_TO_MCDI_FLAG(IP_PROTO, IP_PROTO);
3287 #undef MAP_FILTER_TO_MCDI_FLAG
3288 
3289 	/* Did we map them all? */
3290 	WARN_ON_ONCE(match_flags);
3291 
3292 	return mcdi_flags;
3293 }
3294 
3295 static int efx_ef10_filter_pri(struct efx_ef10_filter_table *table,
3296 			       const struct efx_filter_spec *spec)
3297 {
3298 	u32 mcdi_flags = efx_ef10_filter_mcdi_flags_from_spec(spec);
3299 	unsigned int match_pri;
3300 
3301 	for (match_pri = 0;
3302 	     match_pri < table->rx_match_count;
3303 	     match_pri++)
3304 		if (table->rx_match_mcdi_flags[match_pri] == mcdi_flags)
3305 			return match_pri;
3306 
3307 	return -EPROTONOSUPPORT;
3308 }
3309 
3310 static s32 efx_ef10_filter_insert(struct efx_nic *efx,
3311 				  struct efx_filter_spec *spec,
3312 				  bool replace_equal)
3313 {
3314 	struct efx_ef10_filter_table *table = efx->filter_state;
3315 	DECLARE_BITMAP(mc_rem_map, EFX_EF10_FILTER_SEARCH_LIMIT);
3316 	struct efx_filter_spec *saved_spec;
3317 	unsigned int match_pri, hash;
3318 	unsigned int priv_flags;
3319 	bool replacing = false;
3320 	int ins_index = -1;
3321 	DEFINE_WAIT(wait);
3322 	bool is_mc_recip;
3323 	s32 rc;
3324 
3325 	/* For now, only support RX filters */
3326 	if ((spec->flags & (EFX_FILTER_FLAG_RX | EFX_FILTER_FLAG_TX)) !=
3327 	    EFX_FILTER_FLAG_RX)
3328 		return -EINVAL;
3329 
3330 	rc = efx_ef10_filter_pri(table, spec);
3331 	if (rc < 0)
3332 		return rc;
3333 	match_pri = rc;
3334 
3335 	hash = efx_ef10_filter_hash(spec);
3336 	is_mc_recip = efx_filter_is_mc_recipient(spec);
3337 	if (is_mc_recip)
3338 		bitmap_zero(mc_rem_map, EFX_EF10_FILTER_SEARCH_LIMIT);
3339 
3340 	/* Find any existing filters with the same match tuple or
3341 	 * else a free slot to insert at.  If any of them are busy,
3342 	 * we have to wait and retry.
3343 	 */
3344 	for (;;) {
3345 		unsigned int depth = 1;
3346 		unsigned int i;
3347 
3348 		spin_lock_bh(&efx->filter_lock);
3349 
3350 		for (;;) {
3351 			i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
3352 			saved_spec = efx_ef10_filter_entry_spec(table, i);
3353 
3354 			if (!saved_spec) {
3355 				if (ins_index < 0)
3356 					ins_index = i;
3357 			} else if (efx_ef10_filter_equal(spec, saved_spec)) {
3358 				if (table->entry[i].spec &
3359 				    EFX_EF10_FILTER_FLAG_BUSY)
3360 					break;
3361 				if (spec->priority < saved_spec->priority &&
3362 				    spec->priority != EFX_FILTER_PRI_AUTO) {
3363 					rc = -EPERM;
3364 					goto out_unlock;
3365 				}
3366 				if (!is_mc_recip) {
3367 					/* This is the only one */
3368 					if (spec->priority ==
3369 					    saved_spec->priority &&
3370 					    !replace_equal) {
3371 						rc = -EEXIST;
3372 						goto out_unlock;
3373 					}
3374 					ins_index = i;
3375 					goto found;
3376 				} else if (spec->priority >
3377 					   saved_spec->priority ||
3378 					   (spec->priority ==
3379 					    saved_spec->priority &&
3380 					    replace_equal)) {
3381 					if (ins_index < 0)
3382 						ins_index = i;
3383 					else
3384 						__set_bit(depth, mc_rem_map);
3385 				}
3386 			}
3387 
3388 			/* Once we reach the maximum search depth, use
3389 			 * the first suitable slot or return -EBUSY if
3390 			 * there was none
3391 			 */
3392 			if (depth == EFX_EF10_FILTER_SEARCH_LIMIT) {
3393 				if (ins_index < 0) {
3394 					rc = -EBUSY;
3395 					goto out_unlock;
3396 				}
3397 				goto found;
3398 			}
3399 
3400 			++depth;
3401 		}
3402 
3403 		prepare_to_wait(&table->waitq, &wait, TASK_UNINTERRUPTIBLE);
3404 		spin_unlock_bh(&efx->filter_lock);
3405 		schedule();
3406 	}
3407 
3408 found:
3409 	/* Create a software table entry if necessary, and mark it
3410 	 * busy.  We might yet fail to insert, but any attempt to
3411 	 * insert a conflicting filter while we're waiting for the
3412 	 * firmware must find the busy entry.
3413 	 */
3414 	saved_spec = efx_ef10_filter_entry_spec(table, ins_index);
3415 	if (saved_spec) {
3416 		if (spec->priority == EFX_FILTER_PRI_AUTO &&
3417 		    saved_spec->priority >= EFX_FILTER_PRI_AUTO) {
3418 			/* Just make sure it won't be removed */
3419 			if (saved_spec->priority > EFX_FILTER_PRI_AUTO)
3420 				saved_spec->flags |= EFX_FILTER_FLAG_RX_OVER_AUTO;
3421 			table->entry[ins_index].spec &=
3422 				~EFX_EF10_FILTER_FLAG_AUTO_OLD;
3423 			rc = ins_index;
3424 			goto out_unlock;
3425 		}
3426 		replacing = true;
3427 		priv_flags = efx_ef10_filter_entry_flags(table, ins_index);
3428 	} else {
3429 		saved_spec = kmalloc(sizeof(*spec), GFP_ATOMIC);
3430 		if (!saved_spec) {
3431 			rc = -ENOMEM;
3432 			goto out_unlock;
3433 		}
3434 		*saved_spec = *spec;
3435 		priv_flags = 0;
3436 	}
3437 	efx_ef10_filter_set_entry(table, ins_index, saved_spec,
3438 				  priv_flags | EFX_EF10_FILTER_FLAG_BUSY);
3439 
3440 	/* Mark lower-priority multicast recipients busy prior to removal */
3441 	if (is_mc_recip) {
3442 		unsigned int depth, i;
3443 
3444 		for (depth = 0; depth < EFX_EF10_FILTER_SEARCH_LIMIT; depth++) {
3445 			i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
3446 			if (test_bit(depth, mc_rem_map))
3447 				table->entry[i].spec |=
3448 					EFX_EF10_FILTER_FLAG_BUSY;
3449 		}
3450 	}
3451 
3452 	spin_unlock_bh(&efx->filter_lock);
3453 
3454 	rc = efx_ef10_filter_push(efx, spec, &table->entry[ins_index].handle,
3455 				  replacing);
3456 
3457 	/* Finalise the software table entry */
3458 	spin_lock_bh(&efx->filter_lock);
3459 	if (rc == 0) {
3460 		if (replacing) {
3461 			/* Update the fields that may differ */
3462 			if (saved_spec->priority == EFX_FILTER_PRI_AUTO)
3463 				saved_spec->flags |=
3464 					EFX_FILTER_FLAG_RX_OVER_AUTO;
3465 			saved_spec->priority = spec->priority;
3466 			saved_spec->flags &= EFX_FILTER_FLAG_RX_OVER_AUTO;
3467 			saved_spec->flags |= spec->flags;
3468 			saved_spec->rss_context = spec->rss_context;
3469 			saved_spec->dmaq_id = spec->dmaq_id;
3470 		}
3471 	} else if (!replacing) {
3472 		kfree(saved_spec);
3473 		saved_spec = NULL;
3474 	}
3475 	efx_ef10_filter_set_entry(table, ins_index, saved_spec, priv_flags);
3476 
3477 	/* Remove and finalise entries for lower-priority multicast
3478 	 * recipients
3479 	 */
3480 	if (is_mc_recip) {
3481 		MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_IN_LEN);
3482 		unsigned int depth, i;
3483 
3484 		memset(inbuf, 0, sizeof(inbuf));
3485 
3486 		for (depth = 0; depth < EFX_EF10_FILTER_SEARCH_LIMIT; depth++) {
3487 			if (!test_bit(depth, mc_rem_map))
3488 				continue;
3489 
3490 			i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
3491 			saved_spec = efx_ef10_filter_entry_spec(table, i);
3492 			priv_flags = efx_ef10_filter_entry_flags(table, i);
3493 
3494 			if (rc == 0) {
3495 				spin_unlock_bh(&efx->filter_lock);
3496 				MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
3497 					       MC_CMD_FILTER_OP_IN_OP_UNSUBSCRIBE);
3498 				MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
3499 					       table->entry[i].handle);
3500 				rc = efx_mcdi_rpc(efx, MC_CMD_FILTER_OP,
3501 						  inbuf, sizeof(inbuf),
3502 						  NULL, 0, NULL);
3503 				spin_lock_bh(&efx->filter_lock);
3504 			}
3505 
3506 			if (rc == 0) {
3507 				kfree(saved_spec);
3508 				saved_spec = NULL;
3509 				priv_flags = 0;
3510 			} else {
3511 				priv_flags &= ~EFX_EF10_FILTER_FLAG_BUSY;
3512 			}
3513 			efx_ef10_filter_set_entry(table, i, saved_spec,
3514 						  priv_flags);
3515 		}
3516 	}
3517 
3518 	/* If successful, return the inserted filter ID */
3519 	if (rc == 0)
3520 		rc = match_pri * HUNT_FILTER_TBL_ROWS + ins_index;
3521 
3522 	wake_up_all(&table->waitq);
3523 out_unlock:
3524 	spin_unlock_bh(&efx->filter_lock);
3525 	finish_wait(&table->waitq, &wait);
3526 	return rc;
3527 }
3528 
3529 static void efx_ef10_filter_update_rx_scatter(struct efx_nic *efx)
3530 {
3531 	/* no need to do anything here on EF10 */
3532 }
3533 
3534 /* Remove a filter.
3535  * If !by_index, remove by ID
3536  * If by_index, remove by index
3537  * Filter ID may come from userland and must be range-checked.
3538  */
3539 static int efx_ef10_filter_remove_internal(struct efx_nic *efx,
3540 					   unsigned int priority_mask,
3541 					   u32 filter_id, bool by_index)
3542 {
3543 	unsigned int filter_idx = filter_id % HUNT_FILTER_TBL_ROWS;
3544 	struct efx_ef10_filter_table *table = efx->filter_state;
3545 	MCDI_DECLARE_BUF(inbuf,
3546 			 MC_CMD_FILTER_OP_IN_HANDLE_OFST +
3547 			 MC_CMD_FILTER_OP_IN_HANDLE_LEN);
3548 	struct efx_filter_spec *spec;
3549 	DEFINE_WAIT(wait);
3550 	int rc;
3551 
3552 	/* Find the software table entry and mark it busy.  Don't
3553 	 * remove it yet; any attempt to update while we're waiting
3554 	 * for the firmware must find the busy entry.
3555 	 */
3556 	for (;;) {
3557 		spin_lock_bh(&efx->filter_lock);
3558 		if (!(table->entry[filter_idx].spec &
3559 		      EFX_EF10_FILTER_FLAG_BUSY))
3560 			break;
3561 		prepare_to_wait(&table->waitq, &wait, TASK_UNINTERRUPTIBLE);
3562 		spin_unlock_bh(&efx->filter_lock);
3563 		schedule();
3564 	}
3565 
3566 	spec = efx_ef10_filter_entry_spec(table, filter_idx);
3567 	if (!spec ||
3568 	    (!by_index &&
3569 	     efx_ef10_filter_pri(table, spec) !=
3570 	     filter_id / HUNT_FILTER_TBL_ROWS)) {
3571 		rc = -ENOENT;
3572 		goto out_unlock;
3573 	}
3574 
3575 	if (spec->flags & EFX_FILTER_FLAG_RX_OVER_AUTO &&
3576 	    priority_mask == (1U << EFX_FILTER_PRI_AUTO)) {
3577 		/* Just remove flags */
3578 		spec->flags &= ~EFX_FILTER_FLAG_RX_OVER_AUTO;
3579 		table->entry[filter_idx].spec &= ~EFX_EF10_FILTER_FLAG_AUTO_OLD;
3580 		rc = 0;
3581 		goto out_unlock;
3582 	}
3583 
3584 	if (!(priority_mask & (1U << spec->priority))) {
3585 		rc = -ENOENT;
3586 		goto out_unlock;
3587 	}
3588 
3589 	table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_BUSY;
3590 	spin_unlock_bh(&efx->filter_lock);
3591 
3592 	if (spec->flags & EFX_FILTER_FLAG_RX_OVER_AUTO) {
3593 		/* Reset to an automatic filter */
3594 
3595 		struct efx_filter_spec new_spec = *spec;
3596 
3597 		new_spec.priority = EFX_FILTER_PRI_AUTO;
3598 		new_spec.flags = (EFX_FILTER_FLAG_RX |
3599 				  (efx_rss_enabled(efx) ?
3600 				   EFX_FILTER_FLAG_RX_RSS : 0));
3601 		new_spec.dmaq_id = 0;
3602 		new_spec.rss_context = EFX_FILTER_RSS_CONTEXT_DEFAULT;
3603 		rc = efx_ef10_filter_push(efx, &new_spec,
3604 					  &table->entry[filter_idx].handle,
3605 					  true);
3606 
3607 		spin_lock_bh(&efx->filter_lock);
3608 		if (rc == 0)
3609 			*spec = new_spec;
3610 	} else {
3611 		/* Really remove the filter */
3612 
3613 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
3614 			       efx_ef10_filter_is_exclusive(spec) ?
3615 			       MC_CMD_FILTER_OP_IN_OP_REMOVE :
3616 			       MC_CMD_FILTER_OP_IN_OP_UNSUBSCRIBE);
3617 		MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
3618 			       table->entry[filter_idx].handle);
3619 		rc = efx_mcdi_rpc(efx, MC_CMD_FILTER_OP,
3620 				  inbuf, sizeof(inbuf), NULL, 0, NULL);
3621 
3622 		spin_lock_bh(&efx->filter_lock);
3623 		if (rc == 0) {
3624 			kfree(spec);
3625 			efx_ef10_filter_set_entry(table, filter_idx, NULL, 0);
3626 		}
3627 	}
3628 
3629 	table->entry[filter_idx].spec &= ~EFX_EF10_FILTER_FLAG_BUSY;
3630 	wake_up_all(&table->waitq);
3631 out_unlock:
3632 	spin_unlock_bh(&efx->filter_lock);
3633 	finish_wait(&table->waitq, &wait);
3634 	return rc;
3635 }
3636 
3637 static int efx_ef10_filter_remove_safe(struct efx_nic *efx,
3638 				       enum efx_filter_priority priority,
3639 				       u32 filter_id)
3640 {
3641 	return efx_ef10_filter_remove_internal(efx, 1U << priority,
3642 					       filter_id, false);
3643 }
3644 
3645 static u32 efx_ef10_filter_get_unsafe_id(struct efx_nic *efx, u32 filter_id)
3646 {
3647 	return filter_id % HUNT_FILTER_TBL_ROWS;
3648 }
3649 
3650 static void efx_ef10_filter_remove_unsafe(struct efx_nic *efx,
3651 					  enum efx_filter_priority priority,
3652 					  u32 filter_id)
3653 {
3654 	if (filter_id == EFX_EF10_FILTER_ID_INVALID)
3655 		return;
3656 	efx_ef10_filter_remove_internal(efx, 1U << priority, filter_id, true);
3657 }
3658 
3659 static int efx_ef10_filter_get_safe(struct efx_nic *efx,
3660 				    enum efx_filter_priority priority,
3661 				    u32 filter_id, struct efx_filter_spec *spec)
3662 {
3663 	unsigned int filter_idx = filter_id % HUNT_FILTER_TBL_ROWS;
3664 	struct efx_ef10_filter_table *table = efx->filter_state;
3665 	const struct efx_filter_spec *saved_spec;
3666 	int rc;
3667 
3668 	spin_lock_bh(&efx->filter_lock);
3669 	saved_spec = efx_ef10_filter_entry_spec(table, filter_idx);
3670 	if (saved_spec && saved_spec->priority == priority &&
3671 	    efx_ef10_filter_pri(table, saved_spec) ==
3672 	    filter_id / HUNT_FILTER_TBL_ROWS) {
3673 		*spec = *saved_spec;
3674 		rc = 0;
3675 	} else {
3676 		rc = -ENOENT;
3677 	}
3678 	spin_unlock_bh(&efx->filter_lock);
3679 	return rc;
3680 }
3681 
3682 static int efx_ef10_filter_clear_rx(struct efx_nic *efx,
3683 				     enum efx_filter_priority priority)
3684 {
3685 	unsigned int priority_mask;
3686 	unsigned int i;
3687 	int rc;
3688 
3689 	priority_mask = (((1U << (priority + 1)) - 1) &
3690 			 ~(1U << EFX_FILTER_PRI_AUTO));
3691 
3692 	for (i = 0; i < HUNT_FILTER_TBL_ROWS; i++) {
3693 		rc = efx_ef10_filter_remove_internal(efx, priority_mask,
3694 						     i, true);
3695 		if (rc && rc != -ENOENT)
3696 			return rc;
3697 	}
3698 
3699 	return 0;
3700 }
3701 
3702 static u32 efx_ef10_filter_count_rx_used(struct efx_nic *efx,
3703 					 enum efx_filter_priority priority)
3704 {
3705 	struct efx_ef10_filter_table *table = efx->filter_state;
3706 	unsigned int filter_idx;
3707 	s32 count = 0;
3708 
3709 	spin_lock_bh(&efx->filter_lock);
3710 	for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
3711 		if (table->entry[filter_idx].spec &&
3712 		    efx_ef10_filter_entry_spec(table, filter_idx)->priority ==
3713 		    priority)
3714 			++count;
3715 	}
3716 	spin_unlock_bh(&efx->filter_lock);
3717 	return count;
3718 }
3719 
3720 static u32 efx_ef10_filter_get_rx_id_limit(struct efx_nic *efx)
3721 {
3722 	struct efx_ef10_filter_table *table = efx->filter_state;
3723 
3724 	return table->rx_match_count * HUNT_FILTER_TBL_ROWS;
3725 }
3726 
3727 static s32 efx_ef10_filter_get_rx_ids(struct efx_nic *efx,
3728 				      enum efx_filter_priority priority,
3729 				      u32 *buf, u32 size)
3730 {
3731 	struct efx_ef10_filter_table *table = efx->filter_state;
3732 	struct efx_filter_spec *spec;
3733 	unsigned int filter_idx;
3734 	s32 count = 0;
3735 
3736 	spin_lock_bh(&efx->filter_lock);
3737 	for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
3738 		spec = efx_ef10_filter_entry_spec(table, filter_idx);
3739 		if (spec && spec->priority == priority) {
3740 			if (count == size) {
3741 				count = -EMSGSIZE;
3742 				break;
3743 			}
3744 			buf[count++] = (efx_ef10_filter_pri(table, spec) *
3745 					HUNT_FILTER_TBL_ROWS +
3746 					filter_idx);
3747 		}
3748 	}
3749 	spin_unlock_bh(&efx->filter_lock);
3750 	return count;
3751 }
3752 
3753 #ifdef CONFIG_RFS_ACCEL
3754 
3755 static efx_mcdi_async_completer efx_ef10_filter_rfs_insert_complete;
3756 
3757 static s32 efx_ef10_filter_rfs_insert(struct efx_nic *efx,
3758 				      struct efx_filter_spec *spec)
3759 {
3760 	struct efx_ef10_filter_table *table = efx->filter_state;
3761 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_IN_LEN);
3762 	struct efx_filter_spec *saved_spec;
3763 	unsigned int hash, i, depth = 1;
3764 	bool replacing = false;
3765 	int ins_index = -1;
3766 	u64 cookie;
3767 	s32 rc;
3768 
3769 	/* Must be an RX filter without RSS and not for a multicast
3770 	 * destination address (RFS only works for connected sockets).
3771 	 * These restrictions allow us to pass only a tiny amount of
3772 	 * data through to the completion function.
3773 	 */
3774 	EFX_WARN_ON_PARANOID(spec->flags !=
3775 			     (EFX_FILTER_FLAG_RX | EFX_FILTER_FLAG_RX_SCATTER));
3776 	EFX_WARN_ON_PARANOID(spec->priority != EFX_FILTER_PRI_HINT);
3777 	EFX_WARN_ON_PARANOID(efx_filter_is_mc_recipient(spec));
3778 
3779 	hash = efx_ef10_filter_hash(spec);
3780 
3781 	spin_lock_bh(&efx->filter_lock);
3782 
3783 	/* Find any existing filter with the same match tuple or else
3784 	 * a free slot to insert at.  If an existing filter is busy,
3785 	 * we have to give up.
3786 	 */
3787 	for (;;) {
3788 		i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
3789 		saved_spec = efx_ef10_filter_entry_spec(table, i);
3790 
3791 		if (!saved_spec) {
3792 			if (ins_index < 0)
3793 				ins_index = i;
3794 		} else if (efx_ef10_filter_equal(spec, saved_spec)) {
3795 			if (table->entry[i].spec & EFX_EF10_FILTER_FLAG_BUSY) {
3796 				rc = -EBUSY;
3797 				goto fail_unlock;
3798 			}
3799 			if (spec->priority < saved_spec->priority) {
3800 				rc = -EPERM;
3801 				goto fail_unlock;
3802 			}
3803 			ins_index = i;
3804 			break;
3805 		}
3806 
3807 		/* Once we reach the maximum search depth, use the
3808 		 * first suitable slot or return -EBUSY if there was
3809 		 * none
3810 		 */
3811 		if (depth == EFX_EF10_FILTER_SEARCH_LIMIT) {
3812 			if (ins_index < 0) {
3813 				rc = -EBUSY;
3814 				goto fail_unlock;
3815 			}
3816 			break;
3817 		}
3818 
3819 		++depth;
3820 	}
3821 
3822 	/* Create a software table entry if necessary, and mark it
3823 	 * busy.  We might yet fail to insert, but any attempt to
3824 	 * insert a conflicting filter while we're waiting for the
3825 	 * firmware must find the busy entry.
3826 	 */
3827 	saved_spec = efx_ef10_filter_entry_spec(table, ins_index);
3828 	if (saved_spec) {
3829 		replacing = true;
3830 	} else {
3831 		saved_spec = kmalloc(sizeof(*spec), GFP_ATOMIC);
3832 		if (!saved_spec) {
3833 			rc = -ENOMEM;
3834 			goto fail_unlock;
3835 		}
3836 		*saved_spec = *spec;
3837 	}
3838 	efx_ef10_filter_set_entry(table, ins_index, saved_spec,
3839 				  EFX_EF10_FILTER_FLAG_BUSY);
3840 
3841 	spin_unlock_bh(&efx->filter_lock);
3842 
3843 	/* Pack up the variables needed on completion */
3844 	cookie = replacing << 31 | ins_index << 16 | spec->dmaq_id;
3845 
3846 	efx_ef10_filter_push_prep(efx, spec, inbuf,
3847 				  table->entry[ins_index].handle, replacing);
3848 	efx_mcdi_rpc_async(efx, MC_CMD_FILTER_OP, inbuf, sizeof(inbuf),
3849 			   MC_CMD_FILTER_OP_OUT_LEN,
3850 			   efx_ef10_filter_rfs_insert_complete, cookie);
3851 
3852 	return ins_index;
3853 
3854 fail_unlock:
3855 	spin_unlock_bh(&efx->filter_lock);
3856 	return rc;
3857 }
3858 
3859 static void
3860 efx_ef10_filter_rfs_insert_complete(struct efx_nic *efx, unsigned long cookie,
3861 				    int rc, efx_dword_t *outbuf,
3862 				    size_t outlen_actual)
3863 {
3864 	struct efx_ef10_filter_table *table = efx->filter_state;
3865 	unsigned int ins_index, dmaq_id;
3866 	struct efx_filter_spec *spec;
3867 	bool replacing;
3868 
3869 	/* Unpack the cookie */
3870 	replacing = cookie >> 31;
3871 	ins_index = (cookie >> 16) & (HUNT_FILTER_TBL_ROWS - 1);
3872 	dmaq_id = cookie & 0xffff;
3873 
3874 	spin_lock_bh(&efx->filter_lock);
3875 	spec = efx_ef10_filter_entry_spec(table, ins_index);
3876 	if (rc == 0) {
3877 		table->entry[ins_index].handle =
3878 			MCDI_QWORD(outbuf, FILTER_OP_OUT_HANDLE);
3879 		if (replacing)
3880 			spec->dmaq_id = dmaq_id;
3881 	} else if (!replacing) {
3882 		kfree(spec);
3883 		spec = NULL;
3884 	}
3885 	efx_ef10_filter_set_entry(table, ins_index, spec, 0);
3886 	spin_unlock_bh(&efx->filter_lock);
3887 
3888 	wake_up_all(&table->waitq);
3889 }
3890 
3891 static void
3892 efx_ef10_filter_rfs_expire_complete(struct efx_nic *efx,
3893 				    unsigned long filter_idx,
3894 				    int rc, efx_dword_t *outbuf,
3895 				    size_t outlen_actual);
3896 
3897 static bool efx_ef10_filter_rfs_expire_one(struct efx_nic *efx, u32 flow_id,
3898 					   unsigned int filter_idx)
3899 {
3900 	struct efx_ef10_filter_table *table = efx->filter_state;
3901 	struct efx_filter_spec *spec =
3902 		efx_ef10_filter_entry_spec(table, filter_idx);
3903 	MCDI_DECLARE_BUF(inbuf,
3904 			 MC_CMD_FILTER_OP_IN_HANDLE_OFST +
3905 			 MC_CMD_FILTER_OP_IN_HANDLE_LEN);
3906 
3907 	if (!spec ||
3908 	    (table->entry[filter_idx].spec & EFX_EF10_FILTER_FLAG_BUSY) ||
3909 	    spec->priority != EFX_FILTER_PRI_HINT ||
3910 	    !rps_may_expire_flow(efx->net_dev, spec->dmaq_id,
3911 				 flow_id, filter_idx))
3912 		return false;
3913 
3914 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
3915 		       MC_CMD_FILTER_OP_IN_OP_REMOVE);
3916 	MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
3917 		       table->entry[filter_idx].handle);
3918 	if (efx_mcdi_rpc_async(efx, MC_CMD_FILTER_OP, inbuf, sizeof(inbuf), 0,
3919 			       efx_ef10_filter_rfs_expire_complete, filter_idx))
3920 		return false;
3921 
3922 	table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_BUSY;
3923 	return true;
3924 }
3925 
3926 static void
3927 efx_ef10_filter_rfs_expire_complete(struct efx_nic *efx,
3928 				    unsigned long filter_idx,
3929 				    int rc, efx_dword_t *outbuf,
3930 				    size_t outlen_actual)
3931 {
3932 	struct efx_ef10_filter_table *table = efx->filter_state;
3933 	struct efx_filter_spec *spec =
3934 		efx_ef10_filter_entry_spec(table, filter_idx);
3935 
3936 	spin_lock_bh(&efx->filter_lock);
3937 	if (rc == 0) {
3938 		kfree(spec);
3939 		efx_ef10_filter_set_entry(table, filter_idx, NULL, 0);
3940 	}
3941 	table->entry[filter_idx].spec &= ~EFX_EF10_FILTER_FLAG_BUSY;
3942 	wake_up_all(&table->waitq);
3943 	spin_unlock_bh(&efx->filter_lock);
3944 }
3945 
3946 #endif /* CONFIG_RFS_ACCEL */
3947 
3948 static int efx_ef10_filter_match_flags_from_mcdi(u32 mcdi_flags)
3949 {
3950 	int match_flags = 0;
3951 
3952 #define MAP_FLAG(gen_flag, mcdi_field) {				\
3953 		u32 old_mcdi_flags = mcdi_flags;			\
3954 		mcdi_flags &= ~(1 << MC_CMD_FILTER_OP_IN_MATCH_ ##	\
3955 				mcdi_field ## _LBN);			\
3956 		if (mcdi_flags != old_mcdi_flags)			\
3957 			match_flags |= EFX_FILTER_MATCH_ ## gen_flag;	\
3958 	}
3959 	MAP_FLAG(LOC_MAC_IG, UNKNOWN_UCAST_DST);
3960 	MAP_FLAG(LOC_MAC_IG, UNKNOWN_MCAST_DST);
3961 	MAP_FLAG(REM_HOST, SRC_IP);
3962 	MAP_FLAG(LOC_HOST, DST_IP);
3963 	MAP_FLAG(REM_MAC, SRC_MAC);
3964 	MAP_FLAG(REM_PORT, SRC_PORT);
3965 	MAP_FLAG(LOC_MAC, DST_MAC);
3966 	MAP_FLAG(LOC_PORT, DST_PORT);
3967 	MAP_FLAG(ETHER_TYPE, ETHER_TYPE);
3968 	MAP_FLAG(INNER_VID, INNER_VLAN);
3969 	MAP_FLAG(OUTER_VID, OUTER_VLAN);
3970 	MAP_FLAG(IP_PROTO, IP_PROTO);
3971 #undef MAP_FLAG
3972 
3973 	/* Did we map them all? */
3974 	if (mcdi_flags)
3975 		return -EINVAL;
3976 
3977 	return match_flags;
3978 }
3979 
3980 static void efx_ef10_filter_cleanup_vlans(struct efx_nic *efx)
3981 {
3982 	struct efx_ef10_filter_table *table = efx->filter_state;
3983 	struct efx_ef10_filter_vlan *vlan, *next_vlan;
3984 
3985 	/* See comment in efx_ef10_filter_table_remove() */
3986 	if (!efx_rwsem_assert_write_locked(&efx->filter_sem))
3987 		return;
3988 
3989 	if (!table)
3990 		return;
3991 
3992 	list_for_each_entry_safe(vlan, next_vlan, &table->vlan_list, list)
3993 		efx_ef10_filter_del_vlan_internal(efx, vlan);
3994 }
3995 
3996 static bool efx_ef10_filter_match_supported(struct efx_ef10_filter_table *table,
3997 					    enum efx_filter_match_flags match_flags)
3998 {
3999 	unsigned int match_pri;
4000 	int mf;
4001 
4002 	for (match_pri = 0;
4003 	     match_pri < table->rx_match_count;
4004 	     match_pri++) {
4005 		mf = efx_ef10_filter_match_flags_from_mcdi(
4006 				table->rx_match_mcdi_flags[match_pri]);
4007 		if (mf == match_flags)
4008 			return true;
4009 	}
4010 
4011 	return false;
4012 }
4013 
4014 static int efx_ef10_filter_table_probe(struct efx_nic *efx)
4015 {
4016 	MCDI_DECLARE_BUF(inbuf, MC_CMD_GET_PARSER_DISP_INFO_IN_LEN);
4017 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_PARSER_DISP_INFO_OUT_LENMAX);
4018 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
4019 	struct net_device *net_dev = efx->net_dev;
4020 	unsigned int pd_match_pri, pd_match_count;
4021 	struct efx_ef10_filter_table *table;
4022 	struct efx_ef10_vlan *vlan;
4023 	size_t outlen;
4024 	int rc;
4025 
4026 	if (!efx_rwsem_assert_write_locked(&efx->filter_sem))
4027 		return -EINVAL;
4028 
4029 	if (efx->filter_state) /* already probed */
4030 		return 0;
4031 
4032 	table = kzalloc(sizeof(*table), GFP_KERNEL);
4033 	if (!table)
4034 		return -ENOMEM;
4035 
4036 	/* Find out which RX filter types are supported, and their priorities */
4037 	MCDI_SET_DWORD(inbuf, GET_PARSER_DISP_INFO_IN_OP,
4038 		       MC_CMD_GET_PARSER_DISP_INFO_IN_OP_GET_SUPPORTED_RX_MATCHES);
4039 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_PARSER_DISP_INFO,
4040 			  inbuf, sizeof(inbuf), outbuf, sizeof(outbuf),
4041 			  &outlen);
4042 	if (rc)
4043 		goto fail;
4044 	pd_match_count = MCDI_VAR_ARRAY_LEN(
4045 		outlen, GET_PARSER_DISP_INFO_OUT_SUPPORTED_MATCHES);
4046 	table->rx_match_count = 0;
4047 
4048 	for (pd_match_pri = 0; pd_match_pri < pd_match_count; pd_match_pri++) {
4049 		u32 mcdi_flags =
4050 			MCDI_ARRAY_DWORD(
4051 				outbuf,
4052 				GET_PARSER_DISP_INFO_OUT_SUPPORTED_MATCHES,
4053 				pd_match_pri);
4054 		rc = efx_ef10_filter_match_flags_from_mcdi(mcdi_flags);
4055 		if (rc < 0) {
4056 			netif_dbg(efx, probe, efx->net_dev,
4057 				  "%s: fw flags %#x pri %u not supported in driver\n",
4058 				  __func__, mcdi_flags, pd_match_pri);
4059 		} else {
4060 			netif_dbg(efx, probe, efx->net_dev,
4061 				  "%s: fw flags %#x pri %u supported as driver flags %#x pri %u\n",
4062 				  __func__, mcdi_flags, pd_match_pri,
4063 				  rc, table->rx_match_count);
4064 			table->rx_match_mcdi_flags[table->rx_match_count] = mcdi_flags;
4065 			table->rx_match_count++;
4066 		}
4067 	}
4068 
4069 	if ((efx_supported_features(efx) & NETIF_F_HW_VLAN_CTAG_FILTER) &&
4070 	    !(efx_ef10_filter_match_supported(table,
4071 		(EFX_FILTER_MATCH_OUTER_VID | EFX_FILTER_MATCH_LOC_MAC)) &&
4072 	      efx_ef10_filter_match_supported(table,
4073 		(EFX_FILTER_MATCH_OUTER_VID | EFX_FILTER_MATCH_LOC_MAC_IG)))) {
4074 		netif_info(efx, probe, net_dev,
4075 			   "VLAN filters are not supported in this firmware variant\n");
4076 		net_dev->features &= ~NETIF_F_HW_VLAN_CTAG_FILTER;
4077 		efx->fixed_features &= ~NETIF_F_HW_VLAN_CTAG_FILTER;
4078 		net_dev->hw_features &= ~NETIF_F_HW_VLAN_CTAG_FILTER;
4079 	}
4080 
4081 	table->entry = vzalloc(HUNT_FILTER_TBL_ROWS * sizeof(*table->entry));
4082 	if (!table->entry) {
4083 		rc = -ENOMEM;
4084 		goto fail;
4085 	}
4086 
4087 	table->mc_promisc_last = false;
4088 	table->vlan_filter =
4089 		!!(efx->net_dev->features & NETIF_F_HW_VLAN_CTAG_FILTER);
4090 	INIT_LIST_HEAD(&table->vlan_list);
4091 
4092 	efx->filter_state = table;
4093 	init_waitqueue_head(&table->waitq);
4094 
4095 	list_for_each_entry(vlan, &nic_data->vlan_list, list) {
4096 		rc = efx_ef10_filter_add_vlan(efx, vlan->vid);
4097 		if (rc)
4098 			goto fail_add_vlan;
4099 	}
4100 
4101 	return 0;
4102 
4103 fail_add_vlan:
4104 	efx_ef10_filter_cleanup_vlans(efx);
4105 	efx->filter_state = NULL;
4106 fail:
4107 	kfree(table);
4108 	return rc;
4109 }
4110 
4111 /* Caller must hold efx->filter_sem for read if race against
4112  * efx_ef10_filter_table_remove() is possible
4113  */
4114 static void efx_ef10_filter_table_restore(struct efx_nic *efx)
4115 {
4116 	struct efx_ef10_filter_table *table = efx->filter_state;
4117 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
4118 	struct efx_filter_spec *spec;
4119 	unsigned int filter_idx;
4120 	bool failed = false;
4121 	int rc;
4122 
4123 	WARN_ON(!rwsem_is_locked(&efx->filter_sem));
4124 
4125 	if (!nic_data->must_restore_filters)
4126 		return;
4127 
4128 	if (!table)
4129 		return;
4130 
4131 	spin_lock_bh(&efx->filter_lock);
4132 
4133 	for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
4134 		spec = efx_ef10_filter_entry_spec(table, filter_idx);
4135 		if (!spec)
4136 			continue;
4137 
4138 		table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_BUSY;
4139 		spin_unlock_bh(&efx->filter_lock);
4140 
4141 		rc = efx_ef10_filter_push(efx, spec,
4142 					  &table->entry[filter_idx].handle,
4143 					  false);
4144 		if (rc)
4145 			failed = true;
4146 
4147 		spin_lock_bh(&efx->filter_lock);
4148 		if (rc) {
4149 			kfree(spec);
4150 			efx_ef10_filter_set_entry(table, filter_idx, NULL, 0);
4151 		} else {
4152 			table->entry[filter_idx].spec &=
4153 				~EFX_EF10_FILTER_FLAG_BUSY;
4154 		}
4155 	}
4156 
4157 	spin_unlock_bh(&efx->filter_lock);
4158 
4159 	if (failed)
4160 		netif_err(efx, hw, efx->net_dev,
4161 			  "unable to restore all filters\n");
4162 	else
4163 		nic_data->must_restore_filters = false;
4164 }
4165 
4166 static void efx_ef10_filter_table_remove(struct efx_nic *efx)
4167 {
4168 	struct efx_ef10_filter_table *table = efx->filter_state;
4169 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_IN_LEN);
4170 	struct efx_filter_spec *spec;
4171 	unsigned int filter_idx;
4172 	int rc;
4173 
4174 	efx_ef10_filter_cleanup_vlans(efx);
4175 	efx->filter_state = NULL;
4176 	/* If we were called without locking, then it's not safe to free
4177 	 * the table as others might be using it.  So we just WARN, leak
4178 	 * the memory, and potentially get an inconsistent filter table
4179 	 * state.
4180 	 * This should never actually happen.
4181 	 */
4182 	if (!efx_rwsem_assert_write_locked(&efx->filter_sem))
4183 		return;
4184 
4185 	if (!table)
4186 		return;
4187 
4188 	for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
4189 		spec = efx_ef10_filter_entry_spec(table, filter_idx);
4190 		if (!spec)
4191 			continue;
4192 
4193 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
4194 			       efx_ef10_filter_is_exclusive(spec) ?
4195 			       MC_CMD_FILTER_OP_IN_OP_REMOVE :
4196 			       MC_CMD_FILTER_OP_IN_OP_UNSUBSCRIBE);
4197 		MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
4198 			       table->entry[filter_idx].handle);
4199 		rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FILTER_OP, inbuf,
4200 					sizeof(inbuf), NULL, 0, NULL);
4201 		if (rc)
4202 			netif_info(efx, drv, efx->net_dev,
4203 				   "%s: filter %04x remove failed\n",
4204 				   __func__, filter_idx);
4205 		kfree(spec);
4206 	}
4207 
4208 	vfree(table->entry);
4209 	kfree(table);
4210 }
4211 
4212 static void efx_ef10_filter_mark_one_old(struct efx_nic *efx, uint16_t *id)
4213 {
4214 	struct efx_ef10_filter_table *table = efx->filter_state;
4215 	unsigned int filter_idx;
4216 
4217 	if (*id != EFX_EF10_FILTER_ID_INVALID) {
4218 		filter_idx = efx_ef10_filter_get_unsafe_id(efx, *id);
4219 		if (!table->entry[filter_idx].spec)
4220 			netif_dbg(efx, drv, efx->net_dev,
4221 				  "marked null spec old %04x:%04x\n", *id,
4222 				  filter_idx);
4223 		table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_AUTO_OLD;
4224 		*id = EFX_EF10_FILTER_ID_INVALID;
4225 	}
4226 }
4227 
4228 /* Mark old per-VLAN filters that may need to be removed */
4229 static void _efx_ef10_filter_vlan_mark_old(struct efx_nic *efx,
4230 					   struct efx_ef10_filter_vlan *vlan)
4231 {
4232 	struct efx_ef10_filter_table *table = efx->filter_state;
4233 	unsigned int i;
4234 
4235 	for (i = 0; i < table->dev_uc_count; i++)
4236 		efx_ef10_filter_mark_one_old(efx, &vlan->uc[i]);
4237 	for (i = 0; i < table->dev_mc_count; i++)
4238 		efx_ef10_filter_mark_one_old(efx, &vlan->mc[i]);
4239 	efx_ef10_filter_mark_one_old(efx, &vlan->ucdef);
4240 	efx_ef10_filter_mark_one_old(efx, &vlan->bcast);
4241 	efx_ef10_filter_mark_one_old(efx, &vlan->mcdef);
4242 }
4243 
4244 /* Mark old filters that may need to be removed.
4245  * Caller must hold efx->filter_sem for read if race against
4246  * efx_ef10_filter_table_remove() is possible
4247  */
4248 static void efx_ef10_filter_mark_old(struct efx_nic *efx)
4249 {
4250 	struct efx_ef10_filter_table *table = efx->filter_state;
4251 	struct efx_ef10_filter_vlan *vlan;
4252 
4253 	spin_lock_bh(&efx->filter_lock);
4254 	list_for_each_entry(vlan, &table->vlan_list, list)
4255 		_efx_ef10_filter_vlan_mark_old(efx, vlan);
4256 	spin_unlock_bh(&efx->filter_lock);
4257 }
4258 
4259 static void efx_ef10_filter_uc_addr_list(struct efx_nic *efx)
4260 {
4261 	struct efx_ef10_filter_table *table = efx->filter_state;
4262 	struct net_device *net_dev = efx->net_dev;
4263 	struct netdev_hw_addr *uc;
4264 	int addr_count;
4265 	unsigned int i;
4266 
4267 	addr_count = netdev_uc_count(net_dev);
4268 	table->uc_promisc = !!(net_dev->flags & IFF_PROMISC);
4269 	table->dev_uc_count = 1 + addr_count;
4270 	ether_addr_copy(table->dev_uc_list[0].addr, net_dev->dev_addr);
4271 	i = 1;
4272 	netdev_for_each_uc_addr(uc, net_dev) {
4273 		if (i >= EFX_EF10_FILTER_DEV_UC_MAX) {
4274 			table->uc_promisc = true;
4275 			break;
4276 		}
4277 		ether_addr_copy(table->dev_uc_list[i].addr, uc->addr);
4278 		i++;
4279 	}
4280 }
4281 
4282 static void efx_ef10_filter_mc_addr_list(struct efx_nic *efx)
4283 {
4284 	struct efx_ef10_filter_table *table = efx->filter_state;
4285 	struct net_device *net_dev = efx->net_dev;
4286 	struct netdev_hw_addr *mc;
4287 	unsigned int i, addr_count;
4288 
4289 	table->mc_promisc = !!(net_dev->flags & (IFF_PROMISC | IFF_ALLMULTI));
4290 
4291 	addr_count = netdev_mc_count(net_dev);
4292 	i = 0;
4293 	netdev_for_each_mc_addr(mc, net_dev) {
4294 		if (i >= EFX_EF10_FILTER_DEV_MC_MAX) {
4295 			table->mc_promisc = true;
4296 			break;
4297 		}
4298 		ether_addr_copy(table->dev_mc_list[i].addr, mc->addr);
4299 		i++;
4300 	}
4301 
4302 	table->dev_mc_count = i;
4303 }
4304 
4305 static int efx_ef10_filter_insert_addr_list(struct efx_nic *efx,
4306 					    struct efx_ef10_filter_vlan *vlan,
4307 					    bool multicast, bool rollback)
4308 {
4309 	struct efx_ef10_filter_table *table = efx->filter_state;
4310 	struct efx_ef10_dev_addr *addr_list;
4311 	enum efx_filter_flags filter_flags;
4312 	struct efx_filter_spec spec;
4313 	u8 baddr[ETH_ALEN];
4314 	unsigned int i, j;
4315 	int addr_count;
4316 	u16 *ids;
4317 	int rc;
4318 
4319 	if (multicast) {
4320 		addr_list = table->dev_mc_list;
4321 		addr_count = table->dev_mc_count;
4322 		ids = vlan->mc;
4323 	} else {
4324 		addr_list = table->dev_uc_list;
4325 		addr_count = table->dev_uc_count;
4326 		ids = vlan->uc;
4327 	}
4328 
4329 	filter_flags = efx_rss_enabled(efx) ? EFX_FILTER_FLAG_RX_RSS : 0;
4330 
4331 	/* Insert/renew filters */
4332 	for (i = 0; i < addr_count; i++) {
4333 		efx_filter_init_rx(&spec, EFX_FILTER_PRI_AUTO, filter_flags, 0);
4334 		efx_filter_set_eth_local(&spec, vlan->vid, addr_list[i].addr);
4335 		rc = efx_ef10_filter_insert(efx, &spec, true);
4336 		if (rc < 0) {
4337 			if (rollback) {
4338 				netif_info(efx, drv, efx->net_dev,
4339 					   "efx_ef10_filter_insert failed rc=%d\n",
4340 					   rc);
4341 				/* Fall back to promiscuous */
4342 				for (j = 0; j < i; j++) {
4343 					efx_ef10_filter_remove_unsafe(
4344 						efx, EFX_FILTER_PRI_AUTO,
4345 						ids[j]);
4346 					ids[j] = EFX_EF10_FILTER_ID_INVALID;
4347 				}
4348 				return rc;
4349 			} else {
4350 				/* mark as not inserted, and carry on */
4351 				rc = EFX_EF10_FILTER_ID_INVALID;
4352 			}
4353 		}
4354 		ids[i] = efx_ef10_filter_get_unsafe_id(efx, rc);
4355 	}
4356 
4357 	if (multicast && rollback) {
4358 		/* Also need an Ethernet broadcast filter */
4359 		efx_filter_init_rx(&spec, EFX_FILTER_PRI_AUTO, filter_flags, 0);
4360 		eth_broadcast_addr(baddr);
4361 		efx_filter_set_eth_local(&spec, vlan->vid, baddr);
4362 		rc = efx_ef10_filter_insert(efx, &spec, true);
4363 		if (rc < 0) {
4364 			netif_warn(efx, drv, efx->net_dev,
4365 				   "Broadcast filter insert failed rc=%d\n", rc);
4366 			/* Fall back to promiscuous */
4367 			for (j = 0; j < i; j++) {
4368 				efx_ef10_filter_remove_unsafe(
4369 					efx, EFX_FILTER_PRI_AUTO,
4370 					ids[j]);
4371 				ids[j] = EFX_EF10_FILTER_ID_INVALID;
4372 			}
4373 			return rc;
4374 		} else {
4375 			EFX_WARN_ON_PARANOID(vlan->bcast !=
4376 					     EFX_EF10_FILTER_ID_INVALID);
4377 			vlan->bcast = efx_ef10_filter_get_unsafe_id(efx, rc);
4378 		}
4379 	}
4380 
4381 	return 0;
4382 }
4383 
4384 static int efx_ef10_filter_insert_def(struct efx_nic *efx,
4385 				      struct efx_ef10_filter_vlan *vlan,
4386 				      bool multicast, bool rollback)
4387 {
4388 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
4389 	enum efx_filter_flags filter_flags;
4390 	struct efx_filter_spec spec;
4391 	u8 baddr[ETH_ALEN];
4392 	int rc;
4393 
4394 	filter_flags = efx_rss_enabled(efx) ? EFX_FILTER_FLAG_RX_RSS : 0;
4395 
4396 	efx_filter_init_rx(&spec, EFX_FILTER_PRI_AUTO, filter_flags, 0);
4397 
4398 	if (multicast)
4399 		efx_filter_set_mc_def(&spec);
4400 	else
4401 		efx_filter_set_uc_def(&spec);
4402 
4403 	if (vlan->vid != EFX_FILTER_VID_UNSPEC)
4404 		efx_filter_set_eth_local(&spec, vlan->vid, NULL);
4405 
4406 	rc = efx_ef10_filter_insert(efx, &spec, true);
4407 	if (rc < 0) {
4408 		netif_printk(efx, drv, rc == -EPERM ? KERN_DEBUG : KERN_WARNING,
4409 			     efx->net_dev,
4410 			     "%scast mismatch filter insert failed rc=%d\n",
4411 			     multicast ? "Multi" : "Uni", rc);
4412 	} else if (multicast) {
4413 		EFX_WARN_ON_PARANOID(vlan->mcdef != EFX_EF10_FILTER_ID_INVALID);
4414 		vlan->mcdef = efx_ef10_filter_get_unsafe_id(efx, rc);
4415 		if (!nic_data->workaround_26807) {
4416 			/* Also need an Ethernet broadcast filter */
4417 			efx_filter_init_rx(&spec, EFX_FILTER_PRI_AUTO,
4418 					   filter_flags, 0);
4419 			eth_broadcast_addr(baddr);
4420 			efx_filter_set_eth_local(&spec, vlan->vid, baddr);
4421 			rc = efx_ef10_filter_insert(efx, &spec, true);
4422 			if (rc < 0) {
4423 				netif_warn(efx, drv, efx->net_dev,
4424 					   "Broadcast filter insert failed rc=%d\n",
4425 					   rc);
4426 				if (rollback) {
4427 					/* Roll back the mc_def filter */
4428 					efx_ef10_filter_remove_unsafe(
4429 							efx, EFX_FILTER_PRI_AUTO,
4430 							vlan->mcdef);
4431 					vlan->mcdef = EFX_EF10_FILTER_ID_INVALID;
4432 					return rc;
4433 				}
4434 			} else {
4435 				EFX_WARN_ON_PARANOID(vlan->bcast !=
4436 						     EFX_EF10_FILTER_ID_INVALID);
4437 				vlan->bcast = efx_ef10_filter_get_unsafe_id(efx, rc);
4438 			}
4439 		}
4440 		rc = 0;
4441 	} else {
4442 		EFX_WARN_ON_PARANOID(vlan->ucdef != EFX_EF10_FILTER_ID_INVALID);
4443 		vlan->ucdef = rc;
4444 		rc = 0;
4445 	}
4446 	return rc;
4447 }
4448 
4449 /* Remove filters that weren't renewed.  Since nothing else changes the AUTO_OLD
4450  * flag or removes these filters, we don't need to hold the filter_lock while
4451  * scanning for these filters.
4452  */
4453 static void efx_ef10_filter_remove_old(struct efx_nic *efx)
4454 {
4455 	struct efx_ef10_filter_table *table = efx->filter_state;
4456 	int remove_failed = 0;
4457 	int remove_noent = 0;
4458 	int rc;
4459 	int i;
4460 
4461 	for (i = 0; i < HUNT_FILTER_TBL_ROWS; i++) {
4462 		if (ACCESS_ONCE(table->entry[i].spec) &
4463 		    EFX_EF10_FILTER_FLAG_AUTO_OLD) {
4464 			rc = efx_ef10_filter_remove_internal(efx,
4465 					1U << EFX_FILTER_PRI_AUTO, i, true);
4466 			if (rc == -ENOENT)
4467 				remove_noent++;
4468 			else if (rc)
4469 				remove_failed++;
4470 		}
4471 	}
4472 
4473 	if (remove_failed)
4474 		netif_info(efx, drv, efx->net_dev,
4475 			   "%s: failed to remove %d filters\n",
4476 			   __func__, remove_failed);
4477 	if (remove_noent)
4478 		netif_info(efx, drv, efx->net_dev,
4479 			   "%s: failed to remove %d non-existent filters\n",
4480 			   __func__, remove_noent);
4481 }
4482 
4483 static int efx_ef10_vport_set_mac_address(struct efx_nic *efx)
4484 {
4485 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
4486 	u8 mac_old[ETH_ALEN];
4487 	int rc, rc2;
4488 
4489 	/* Only reconfigure a PF-created vport */
4490 	if (is_zero_ether_addr(nic_data->vport_mac))
4491 		return 0;
4492 
4493 	efx_device_detach_sync(efx);
4494 	efx_net_stop(efx->net_dev);
4495 	down_write(&efx->filter_sem);
4496 	efx_ef10_filter_table_remove(efx);
4497 	up_write(&efx->filter_sem);
4498 
4499 	rc = efx_ef10_vadaptor_free(efx, nic_data->vport_id);
4500 	if (rc)
4501 		goto restore_filters;
4502 
4503 	ether_addr_copy(mac_old, nic_data->vport_mac);
4504 	rc = efx_ef10_vport_del_mac(efx, nic_data->vport_id,
4505 				    nic_data->vport_mac);
4506 	if (rc)
4507 		goto restore_vadaptor;
4508 
4509 	rc = efx_ef10_vport_add_mac(efx, nic_data->vport_id,
4510 				    efx->net_dev->dev_addr);
4511 	if (!rc) {
4512 		ether_addr_copy(nic_data->vport_mac, efx->net_dev->dev_addr);
4513 	} else {
4514 		rc2 = efx_ef10_vport_add_mac(efx, nic_data->vport_id, mac_old);
4515 		if (rc2) {
4516 			/* Failed to add original MAC, so clear vport_mac */
4517 			eth_zero_addr(nic_data->vport_mac);
4518 			goto reset_nic;
4519 		}
4520 	}
4521 
4522 restore_vadaptor:
4523 	rc2 = efx_ef10_vadaptor_alloc(efx, nic_data->vport_id);
4524 	if (rc2)
4525 		goto reset_nic;
4526 restore_filters:
4527 	down_write(&efx->filter_sem);
4528 	rc2 = efx_ef10_filter_table_probe(efx);
4529 	up_write(&efx->filter_sem);
4530 	if (rc2)
4531 		goto reset_nic;
4532 
4533 	rc2 = efx_net_open(efx->net_dev);
4534 	if (rc2)
4535 		goto reset_nic;
4536 
4537 	netif_device_attach(efx->net_dev);
4538 
4539 	return rc;
4540 
4541 reset_nic:
4542 	netif_err(efx, drv, efx->net_dev,
4543 		  "Failed to restore when changing MAC address - scheduling reset\n");
4544 	efx_schedule_reset(efx, RESET_TYPE_DATAPATH);
4545 
4546 	return rc ? rc : rc2;
4547 }
4548 
4549 /* Caller must hold efx->filter_sem for read if race against
4550  * efx_ef10_filter_table_remove() is possible
4551  */
4552 static void efx_ef10_filter_vlan_sync_rx_mode(struct efx_nic *efx,
4553 					      struct efx_ef10_filter_vlan *vlan)
4554 {
4555 	struct efx_ef10_filter_table *table = efx->filter_state;
4556 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
4557 
4558 	/* Do not install unspecified VID if VLAN filtering is enabled.
4559 	 * Do not install all specified VIDs if VLAN filtering is disabled.
4560 	 */
4561 	if ((vlan->vid == EFX_FILTER_VID_UNSPEC) == table->vlan_filter)
4562 		return;
4563 
4564 	/* Insert/renew unicast filters */
4565 	if (table->uc_promisc) {
4566 		efx_ef10_filter_insert_def(efx, vlan, false, false);
4567 		efx_ef10_filter_insert_addr_list(efx, vlan, false, false);
4568 	} else {
4569 		/* If any of the filters failed to insert, fall back to
4570 		 * promiscuous mode - add in the uc_def filter.  But keep
4571 		 * our individual unicast filters.
4572 		 */
4573 		if (efx_ef10_filter_insert_addr_list(efx, vlan, false, false))
4574 			efx_ef10_filter_insert_def(efx, vlan, false, false);
4575 	}
4576 
4577 	/* Insert/renew multicast filters */
4578 	/* If changing promiscuous state with cascaded multicast filters, remove
4579 	 * old filters first, so that packets are dropped rather than duplicated
4580 	 */
4581 	if (nic_data->workaround_26807 &&
4582 	    table->mc_promisc_last != table->mc_promisc)
4583 		efx_ef10_filter_remove_old(efx);
4584 	if (table->mc_promisc) {
4585 		if (nic_data->workaround_26807) {
4586 			/* If we failed to insert promiscuous filters, rollback
4587 			 * and fall back to individual multicast filters
4588 			 */
4589 			if (efx_ef10_filter_insert_def(efx, vlan, true, true)) {
4590 				/* Changing promisc state, so remove old filters */
4591 				efx_ef10_filter_remove_old(efx);
4592 				efx_ef10_filter_insert_addr_list(efx, vlan,
4593 								 true, false);
4594 			}
4595 		} else {
4596 			/* If we failed to insert promiscuous filters, don't
4597 			 * rollback.  Regardless, also insert the mc_list
4598 			 */
4599 			efx_ef10_filter_insert_def(efx, vlan, true, false);
4600 			efx_ef10_filter_insert_addr_list(efx, vlan, true, false);
4601 		}
4602 	} else {
4603 		/* If any filters failed to insert, rollback and fall back to
4604 		 * promiscuous mode - mc_def filter and maybe broadcast.  If
4605 		 * that fails, roll back again and insert as many of our
4606 		 * individual multicast filters as we can.
4607 		 */
4608 		if (efx_ef10_filter_insert_addr_list(efx, vlan, true, true)) {
4609 			/* Changing promisc state, so remove old filters */
4610 			if (nic_data->workaround_26807)
4611 				efx_ef10_filter_remove_old(efx);
4612 			if (efx_ef10_filter_insert_def(efx, vlan, true, true))
4613 				efx_ef10_filter_insert_addr_list(efx, vlan,
4614 								 true, false);
4615 		}
4616 	}
4617 }
4618 
4619 /* Caller must hold efx->filter_sem for read if race against
4620  * efx_ef10_filter_table_remove() is possible
4621  */
4622 static void efx_ef10_filter_sync_rx_mode(struct efx_nic *efx)
4623 {
4624 	struct efx_ef10_filter_table *table = efx->filter_state;
4625 	struct net_device *net_dev = efx->net_dev;
4626 	struct efx_ef10_filter_vlan *vlan;
4627 	bool vlan_filter;
4628 
4629 	if (!efx_dev_registered(efx))
4630 		return;
4631 
4632 	if (!table)
4633 		return;
4634 
4635 	efx_ef10_filter_mark_old(efx);
4636 
4637 	/* Copy/convert the address lists; add the primary station
4638 	 * address and broadcast address
4639 	 */
4640 	netif_addr_lock_bh(net_dev);
4641 	efx_ef10_filter_uc_addr_list(efx);
4642 	efx_ef10_filter_mc_addr_list(efx);
4643 	netif_addr_unlock_bh(net_dev);
4644 
4645 	/* If VLAN filtering changes, all old filters are finally removed.
4646 	 * Do it in advance to avoid conflicts for unicast untagged and
4647 	 * VLAN 0 tagged filters.
4648 	 */
4649 	vlan_filter = !!(net_dev->features & NETIF_F_HW_VLAN_CTAG_FILTER);
4650 	if (table->vlan_filter != vlan_filter) {
4651 		table->vlan_filter = vlan_filter;
4652 		efx_ef10_filter_remove_old(efx);
4653 	}
4654 
4655 	list_for_each_entry(vlan, &table->vlan_list, list)
4656 		efx_ef10_filter_vlan_sync_rx_mode(efx, vlan);
4657 
4658 	efx_ef10_filter_remove_old(efx);
4659 	table->mc_promisc_last = table->mc_promisc;
4660 }
4661 
4662 static struct efx_ef10_filter_vlan *efx_ef10_filter_find_vlan(struct efx_nic *efx, u16 vid)
4663 {
4664 	struct efx_ef10_filter_table *table = efx->filter_state;
4665 	struct efx_ef10_filter_vlan *vlan;
4666 
4667 	WARN_ON(!rwsem_is_locked(&efx->filter_sem));
4668 
4669 	list_for_each_entry(vlan, &table->vlan_list, list) {
4670 		if (vlan->vid == vid)
4671 			return vlan;
4672 	}
4673 
4674 	return NULL;
4675 }
4676 
4677 static int efx_ef10_filter_add_vlan(struct efx_nic *efx, u16 vid)
4678 {
4679 	struct efx_ef10_filter_table *table = efx->filter_state;
4680 	struct efx_ef10_filter_vlan *vlan;
4681 	unsigned int i;
4682 
4683 	if (!efx_rwsem_assert_write_locked(&efx->filter_sem))
4684 		return -EINVAL;
4685 
4686 	vlan = efx_ef10_filter_find_vlan(efx, vid);
4687 	if (WARN_ON(vlan)) {
4688 		netif_err(efx, drv, efx->net_dev,
4689 			  "VLAN %u already added\n", vid);
4690 		return -EALREADY;
4691 	}
4692 
4693 	vlan = kzalloc(sizeof(*vlan), GFP_KERNEL);
4694 	if (!vlan)
4695 		return -ENOMEM;
4696 
4697 	vlan->vid = vid;
4698 
4699 	for (i = 0; i < ARRAY_SIZE(vlan->uc); i++)
4700 		vlan->uc[i] = EFX_EF10_FILTER_ID_INVALID;
4701 	for (i = 0; i < ARRAY_SIZE(vlan->mc); i++)
4702 		vlan->mc[i] = EFX_EF10_FILTER_ID_INVALID;
4703 	vlan->ucdef = EFX_EF10_FILTER_ID_INVALID;
4704 	vlan->bcast = EFX_EF10_FILTER_ID_INVALID;
4705 	vlan->mcdef = EFX_EF10_FILTER_ID_INVALID;
4706 
4707 	list_add_tail(&vlan->list, &table->vlan_list);
4708 
4709 	if (efx_dev_registered(efx))
4710 		efx_ef10_filter_vlan_sync_rx_mode(efx, vlan);
4711 
4712 	return 0;
4713 }
4714 
4715 static void efx_ef10_filter_del_vlan_internal(struct efx_nic *efx,
4716 					      struct efx_ef10_filter_vlan *vlan)
4717 {
4718 	unsigned int i;
4719 
4720 	/* See comment in efx_ef10_filter_table_remove() */
4721 	if (!efx_rwsem_assert_write_locked(&efx->filter_sem))
4722 		return;
4723 
4724 	list_del(&vlan->list);
4725 
4726 	for (i = 0; i < ARRAY_SIZE(vlan->uc); i++)
4727 		efx_ef10_filter_remove_unsafe(efx, EFX_FILTER_PRI_AUTO,
4728 					      vlan->uc[i]);
4729 	for (i = 0; i < ARRAY_SIZE(vlan->mc); i++)
4730 		efx_ef10_filter_remove_unsafe(efx, EFX_FILTER_PRI_AUTO,
4731 					      vlan->mc[i]);
4732 	efx_ef10_filter_remove_unsafe(efx, EFX_FILTER_PRI_AUTO, vlan->ucdef);
4733 	efx_ef10_filter_remove_unsafe(efx, EFX_FILTER_PRI_AUTO, vlan->bcast);
4734 	efx_ef10_filter_remove_unsafe(efx, EFX_FILTER_PRI_AUTO, vlan->mcdef);
4735 
4736 	kfree(vlan);
4737 }
4738 
4739 static void efx_ef10_filter_del_vlan(struct efx_nic *efx, u16 vid)
4740 {
4741 	struct efx_ef10_filter_vlan *vlan;
4742 
4743 	/* See comment in efx_ef10_filter_table_remove() */
4744 	if (!efx_rwsem_assert_write_locked(&efx->filter_sem))
4745 		return;
4746 
4747 	vlan = efx_ef10_filter_find_vlan(efx, vid);
4748 	if (!vlan) {
4749 		netif_err(efx, drv, efx->net_dev,
4750 			  "VLAN %u not found in filter state\n", vid);
4751 		return;
4752 	}
4753 
4754 	efx_ef10_filter_del_vlan_internal(efx, vlan);
4755 }
4756 
4757 static int efx_ef10_set_mac_address(struct efx_nic *efx)
4758 {
4759 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VADAPTOR_SET_MAC_IN_LEN);
4760 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
4761 	bool was_enabled = efx->port_enabled;
4762 	int rc;
4763 
4764 	efx_device_detach_sync(efx);
4765 	efx_net_stop(efx->net_dev);
4766 
4767 	mutex_lock(&efx->mac_lock);
4768 	down_write(&efx->filter_sem);
4769 	efx_ef10_filter_table_remove(efx);
4770 
4771 	ether_addr_copy(MCDI_PTR(inbuf, VADAPTOR_SET_MAC_IN_MACADDR),
4772 			efx->net_dev->dev_addr);
4773 	MCDI_SET_DWORD(inbuf, VADAPTOR_SET_MAC_IN_UPSTREAM_PORT_ID,
4774 		       nic_data->vport_id);
4775 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_VADAPTOR_SET_MAC, inbuf,
4776 				sizeof(inbuf), NULL, 0, NULL);
4777 
4778 	efx_ef10_filter_table_probe(efx);
4779 	up_write(&efx->filter_sem);
4780 	mutex_unlock(&efx->mac_lock);
4781 
4782 	if (was_enabled)
4783 		efx_net_open(efx->net_dev);
4784 	netif_device_attach(efx->net_dev);
4785 
4786 #ifdef CONFIG_SFC_SRIOV
4787 	if (efx->pci_dev->is_virtfn && efx->pci_dev->physfn) {
4788 		struct pci_dev *pci_dev_pf = efx->pci_dev->physfn;
4789 
4790 		if (rc == -EPERM) {
4791 			struct efx_nic *efx_pf;
4792 
4793 			/* Switch to PF and change MAC address on vport */
4794 			efx_pf = pci_get_drvdata(pci_dev_pf);
4795 
4796 			rc = efx_ef10_sriov_set_vf_mac(efx_pf,
4797 						       nic_data->vf_index,
4798 						       efx->net_dev->dev_addr);
4799 		} else if (!rc) {
4800 			struct efx_nic *efx_pf = pci_get_drvdata(pci_dev_pf);
4801 			struct efx_ef10_nic_data *nic_data = efx_pf->nic_data;
4802 			unsigned int i;
4803 
4804 			/* MAC address successfully changed by VF (with MAC
4805 			 * spoofing) so update the parent PF if possible.
4806 			 */
4807 			for (i = 0; i < efx_pf->vf_count; ++i) {
4808 				struct ef10_vf *vf = nic_data->vf + i;
4809 
4810 				if (vf->efx == efx) {
4811 					ether_addr_copy(vf->mac,
4812 							efx->net_dev->dev_addr);
4813 					return 0;
4814 				}
4815 			}
4816 		}
4817 	} else
4818 #endif
4819 	if (rc == -EPERM) {
4820 		netif_err(efx, drv, efx->net_dev,
4821 			  "Cannot change MAC address; use sfboot to enable"
4822 			  " mac-spoofing on this interface\n");
4823 	} else if (rc == -ENOSYS && !efx_ef10_is_vf(efx)) {
4824 		/* If the active MCFW does not support MC_CMD_VADAPTOR_SET_MAC
4825 		 * fall-back to the method of changing the MAC address on the
4826 		 * vport.  This only applies to PFs because such versions of
4827 		 * MCFW do not support VFs.
4828 		 */
4829 		rc = efx_ef10_vport_set_mac_address(efx);
4830 	} else {
4831 		efx_mcdi_display_error(efx, MC_CMD_VADAPTOR_SET_MAC,
4832 				       sizeof(inbuf), NULL, 0, rc);
4833 	}
4834 
4835 	return rc;
4836 }
4837 
4838 static int efx_ef10_mac_reconfigure(struct efx_nic *efx)
4839 {
4840 	efx_ef10_filter_sync_rx_mode(efx);
4841 
4842 	return efx_mcdi_set_mac(efx);
4843 }
4844 
4845 static int efx_ef10_mac_reconfigure_vf(struct efx_nic *efx)
4846 {
4847 	efx_ef10_filter_sync_rx_mode(efx);
4848 
4849 	return 0;
4850 }
4851 
4852 static int efx_ef10_start_bist(struct efx_nic *efx, u32 bist_type)
4853 {
4854 	MCDI_DECLARE_BUF(inbuf, MC_CMD_START_BIST_IN_LEN);
4855 
4856 	MCDI_SET_DWORD(inbuf, START_BIST_IN_TYPE, bist_type);
4857 	return efx_mcdi_rpc(efx, MC_CMD_START_BIST, inbuf, sizeof(inbuf),
4858 			    NULL, 0, NULL);
4859 }
4860 
4861 /* MC BISTs follow a different poll mechanism to phy BISTs.
4862  * The BIST is done in the poll handler on the MC, and the MCDI command
4863  * will block until the BIST is done.
4864  */
4865 static int efx_ef10_poll_bist(struct efx_nic *efx)
4866 {
4867 	int rc;
4868 	MCDI_DECLARE_BUF(outbuf, MC_CMD_POLL_BIST_OUT_LEN);
4869 	size_t outlen;
4870 	u32 result;
4871 
4872 	rc = efx_mcdi_rpc(efx, MC_CMD_POLL_BIST, NULL, 0,
4873 			   outbuf, sizeof(outbuf), &outlen);
4874 	if (rc != 0)
4875 		return rc;
4876 
4877 	if (outlen < MC_CMD_POLL_BIST_OUT_LEN)
4878 		return -EIO;
4879 
4880 	result = MCDI_DWORD(outbuf, POLL_BIST_OUT_RESULT);
4881 	switch (result) {
4882 	case MC_CMD_POLL_BIST_PASSED:
4883 		netif_dbg(efx, hw, efx->net_dev, "BIST passed.\n");
4884 		return 0;
4885 	case MC_CMD_POLL_BIST_TIMEOUT:
4886 		netif_err(efx, hw, efx->net_dev, "BIST timed out\n");
4887 		return -EIO;
4888 	case MC_CMD_POLL_BIST_FAILED:
4889 		netif_err(efx, hw, efx->net_dev, "BIST failed.\n");
4890 		return -EIO;
4891 	default:
4892 		netif_err(efx, hw, efx->net_dev,
4893 			  "BIST returned unknown result %u", result);
4894 		return -EIO;
4895 	}
4896 }
4897 
4898 static int efx_ef10_run_bist(struct efx_nic *efx, u32 bist_type)
4899 {
4900 	int rc;
4901 
4902 	netif_dbg(efx, drv, efx->net_dev, "starting BIST type %u\n", bist_type);
4903 
4904 	rc = efx_ef10_start_bist(efx, bist_type);
4905 	if (rc != 0)
4906 		return rc;
4907 
4908 	return efx_ef10_poll_bist(efx);
4909 }
4910 
4911 static int
4912 efx_ef10_test_chip(struct efx_nic *efx, struct efx_self_tests *tests)
4913 {
4914 	int rc, rc2;
4915 
4916 	efx_reset_down(efx, RESET_TYPE_WORLD);
4917 
4918 	rc = efx_mcdi_rpc(efx, MC_CMD_ENABLE_OFFLINE_BIST,
4919 			  NULL, 0, NULL, 0, NULL);
4920 	if (rc != 0)
4921 		goto out;
4922 
4923 	tests->memory = efx_ef10_run_bist(efx, MC_CMD_MC_MEM_BIST) ? -1 : 1;
4924 	tests->registers = efx_ef10_run_bist(efx, MC_CMD_REG_BIST) ? -1 : 1;
4925 
4926 	rc = efx_mcdi_reset(efx, RESET_TYPE_WORLD);
4927 
4928 out:
4929 	if (rc == -EPERM)
4930 		rc = 0;
4931 	rc2 = efx_reset_up(efx, RESET_TYPE_WORLD, rc == 0);
4932 	return rc ? rc : rc2;
4933 }
4934 
4935 #ifdef CONFIG_SFC_MTD
4936 
4937 struct efx_ef10_nvram_type_info {
4938 	u16 type, type_mask;
4939 	u8 port;
4940 	const char *name;
4941 };
4942 
4943 static const struct efx_ef10_nvram_type_info efx_ef10_nvram_types[] = {
4944 	{ NVRAM_PARTITION_TYPE_MC_FIRMWARE,	   0,    0, "sfc_mcfw" },
4945 	{ NVRAM_PARTITION_TYPE_MC_FIRMWARE_BACKUP, 0,    0, "sfc_mcfw_backup" },
4946 	{ NVRAM_PARTITION_TYPE_EXPANSION_ROM,	   0,    0, "sfc_exp_rom" },
4947 	{ NVRAM_PARTITION_TYPE_STATIC_CONFIG,	   0,    0, "sfc_static_cfg" },
4948 	{ NVRAM_PARTITION_TYPE_DYNAMIC_CONFIG,	   0,    0, "sfc_dynamic_cfg" },
4949 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT0, 0,   0, "sfc_exp_rom_cfg" },
4950 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT1, 0,   1, "sfc_exp_rom_cfg" },
4951 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT2, 0,   2, "sfc_exp_rom_cfg" },
4952 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT3, 0,   3, "sfc_exp_rom_cfg" },
4953 	{ NVRAM_PARTITION_TYPE_LICENSE,		   0,    0, "sfc_license" },
4954 	{ NVRAM_PARTITION_TYPE_PHY_MIN,		   0xff, 0, "sfc_phy_fw" },
4955 };
4956 
4957 static int efx_ef10_mtd_probe_partition(struct efx_nic *efx,
4958 					struct efx_mcdi_mtd_partition *part,
4959 					unsigned int type)
4960 {
4961 	MCDI_DECLARE_BUF(inbuf, MC_CMD_NVRAM_METADATA_IN_LEN);
4962 	MCDI_DECLARE_BUF(outbuf, MC_CMD_NVRAM_METADATA_OUT_LENMAX);
4963 	const struct efx_ef10_nvram_type_info *info;
4964 	size_t size, erase_size, outlen;
4965 	bool protected;
4966 	int rc;
4967 
4968 	for (info = efx_ef10_nvram_types; ; info++) {
4969 		if (info ==
4970 		    efx_ef10_nvram_types + ARRAY_SIZE(efx_ef10_nvram_types))
4971 			return -ENODEV;
4972 		if ((type & ~info->type_mask) == info->type)
4973 			break;
4974 	}
4975 	if (info->port != efx_port_num(efx))
4976 		return -ENODEV;
4977 
4978 	rc = efx_mcdi_nvram_info(efx, type, &size, &erase_size, &protected);
4979 	if (rc)
4980 		return rc;
4981 	if (protected)
4982 		return -ENODEV; /* hide it */
4983 
4984 	part->nvram_type = type;
4985 
4986 	MCDI_SET_DWORD(inbuf, NVRAM_METADATA_IN_TYPE, type);
4987 	rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_METADATA, inbuf, sizeof(inbuf),
4988 			  outbuf, sizeof(outbuf), &outlen);
4989 	if (rc)
4990 		return rc;
4991 	if (outlen < MC_CMD_NVRAM_METADATA_OUT_LENMIN)
4992 		return -EIO;
4993 	if (MCDI_DWORD(outbuf, NVRAM_METADATA_OUT_FLAGS) &
4994 	    (1 << MC_CMD_NVRAM_METADATA_OUT_SUBTYPE_VALID_LBN))
4995 		part->fw_subtype = MCDI_DWORD(outbuf,
4996 					      NVRAM_METADATA_OUT_SUBTYPE);
4997 
4998 	part->common.dev_type_name = "EF10 NVRAM manager";
4999 	part->common.type_name = info->name;
5000 
5001 	part->common.mtd.type = MTD_NORFLASH;
5002 	part->common.mtd.flags = MTD_CAP_NORFLASH;
5003 	part->common.mtd.size = size;
5004 	part->common.mtd.erasesize = erase_size;
5005 
5006 	return 0;
5007 }
5008 
5009 static int efx_ef10_mtd_probe(struct efx_nic *efx)
5010 {
5011 	MCDI_DECLARE_BUF(outbuf, MC_CMD_NVRAM_PARTITIONS_OUT_LENMAX);
5012 	struct efx_mcdi_mtd_partition *parts;
5013 	size_t outlen, n_parts_total, i, n_parts;
5014 	unsigned int type;
5015 	int rc;
5016 
5017 	ASSERT_RTNL();
5018 
5019 	BUILD_BUG_ON(MC_CMD_NVRAM_PARTITIONS_IN_LEN != 0);
5020 	rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_PARTITIONS, NULL, 0,
5021 			  outbuf, sizeof(outbuf), &outlen);
5022 	if (rc)
5023 		return rc;
5024 	if (outlen < MC_CMD_NVRAM_PARTITIONS_OUT_LENMIN)
5025 		return -EIO;
5026 
5027 	n_parts_total = MCDI_DWORD(outbuf, NVRAM_PARTITIONS_OUT_NUM_PARTITIONS);
5028 	if (n_parts_total >
5029 	    MCDI_VAR_ARRAY_LEN(outlen, NVRAM_PARTITIONS_OUT_TYPE_ID))
5030 		return -EIO;
5031 
5032 	parts = kcalloc(n_parts_total, sizeof(*parts), GFP_KERNEL);
5033 	if (!parts)
5034 		return -ENOMEM;
5035 
5036 	n_parts = 0;
5037 	for (i = 0; i < n_parts_total; i++) {
5038 		type = MCDI_ARRAY_DWORD(outbuf, NVRAM_PARTITIONS_OUT_TYPE_ID,
5039 					i);
5040 		rc = efx_ef10_mtd_probe_partition(efx, &parts[n_parts], type);
5041 		if (rc == 0)
5042 			n_parts++;
5043 		else if (rc != -ENODEV)
5044 			goto fail;
5045 	}
5046 
5047 	rc = efx_mtd_add(efx, &parts[0].common, n_parts, sizeof(*parts));
5048 fail:
5049 	if (rc)
5050 		kfree(parts);
5051 	return rc;
5052 }
5053 
5054 #endif /* CONFIG_SFC_MTD */
5055 
5056 static void efx_ef10_ptp_write_host_time(struct efx_nic *efx, u32 host_time)
5057 {
5058 	_efx_writed(efx, cpu_to_le32(host_time), ER_DZ_MC_DB_LWRD);
5059 }
5060 
5061 static void efx_ef10_ptp_write_host_time_vf(struct efx_nic *efx,
5062 					    u32 host_time) {}
5063 
5064 static int efx_ef10_rx_enable_timestamping(struct efx_channel *channel,
5065 					   bool temp)
5066 {
5067 	MCDI_DECLARE_BUF(inbuf, MC_CMD_PTP_IN_TIME_EVENT_SUBSCRIBE_LEN);
5068 	int rc;
5069 
5070 	if (channel->sync_events_state == SYNC_EVENTS_REQUESTED ||
5071 	    channel->sync_events_state == SYNC_EVENTS_VALID ||
5072 	    (temp && channel->sync_events_state == SYNC_EVENTS_DISABLED))
5073 		return 0;
5074 	channel->sync_events_state = SYNC_EVENTS_REQUESTED;
5075 
5076 	MCDI_SET_DWORD(inbuf, PTP_IN_OP, MC_CMD_PTP_OP_TIME_EVENT_SUBSCRIBE);
5077 	MCDI_SET_DWORD(inbuf, PTP_IN_PERIPH_ID, 0);
5078 	MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_SUBSCRIBE_QUEUE,
5079 		       channel->channel);
5080 
5081 	rc = efx_mcdi_rpc(channel->efx, MC_CMD_PTP,
5082 			  inbuf, sizeof(inbuf), NULL, 0, NULL);
5083 
5084 	if (rc != 0)
5085 		channel->sync_events_state = temp ? SYNC_EVENTS_QUIESCENT :
5086 						    SYNC_EVENTS_DISABLED;
5087 
5088 	return rc;
5089 }
5090 
5091 static int efx_ef10_rx_disable_timestamping(struct efx_channel *channel,
5092 					    bool temp)
5093 {
5094 	MCDI_DECLARE_BUF(inbuf, MC_CMD_PTP_IN_TIME_EVENT_UNSUBSCRIBE_LEN);
5095 	int rc;
5096 
5097 	if (channel->sync_events_state == SYNC_EVENTS_DISABLED ||
5098 	    (temp && channel->sync_events_state == SYNC_EVENTS_QUIESCENT))
5099 		return 0;
5100 	if (channel->sync_events_state == SYNC_EVENTS_QUIESCENT) {
5101 		channel->sync_events_state = SYNC_EVENTS_DISABLED;
5102 		return 0;
5103 	}
5104 	channel->sync_events_state = temp ? SYNC_EVENTS_QUIESCENT :
5105 					    SYNC_EVENTS_DISABLED;
5106 
5107 	MCDI_SET_DWORD(inbuf, PTP_IN_OP, MC_CMD_PTP_OP_TIME_EVENT_UNSUBSCRIBE);
5108 	MCDI_SET_DWORD(inbuf, PTP_IN_PERIPH_ID, 0);
5109 	MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_UNSUBSCRIBE_CONTROL,
5110 		       MC_CMD_PTP_IN_TIME_EVENT_UNSUBSCRIBE_SINGLE);
5111 	MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_UNSUBSCRIBE_QUEUE,
5112 		       channel->channel);
5113 
5114 	rc = efx_mcdi_rpc(channel->efx, MC_CMD_PTP,
5115 			  inbuf, sizeof(inbuf), NULL, 0, NULL);
5116 
5117 	return rc;
5118 }
5119 
5120 static int efx_ef10_ptp_set_ts_sync_events(struct efx_nic *efx, bool en,
5121 					   bool temp)
5122 {
5123 	int (*set)(struct efx_channel *channel, bool temp);
5124 	struct efx_channel *channel;
5125 
5126 	set = en ?
5127 	      efx_ef10_rx_enable_timestamping :
5128 	      efx_ef10_rx_disable_timestamping;
5129 
5130 	efx_for_each_channel(channel, efx) {
5131 		int rc = set(channel, temp);
5132 		if (en && rc != 0) {
5133 			efx_ef10_ptp_set_ts_sync_events(efx, false, temp);
5134 			return rc;
5135 		}
5136 	}
5137 
5138 	return 0;
5139 }
5140 
5141 static int efx_ef10_ptp_set_ts_config_vf(struct efx_nic *efx,
5142 					 struct hwtstamp_config *init)
5143 {
5144 	return -EOPNOTSUPP;
5145 }
5146 
5147 static int efx_ef10_ptp_set_ts_config(struct efx_nic *efx,
5148 				      struct hwtstamp_config *init)
5149 {
5150 	int rc;
5151 
5152 	switch (init->rx_filter) {
5153 	case HWTSTAMP_FILTER_NONE:
5154 		efx_ef10_ptp_set_ts_sync_events(efx, false, false);
5155 		/* if TX timestamping is still requested then leave PTP on */
5156 		return efx_ptp_change_mode(efx,
5157 					   init->tx_type != HWTSTAMP_TX_OFF, 0);
5158 	case HWTSTAMP_FILTER_ALL:
5159 	case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
5160 	case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
5161 	case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
5162 	case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
5163 	case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
5164 	case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
5165 	case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
5166 	case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
5167 	case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
5168 	case HWTSTAMP_FILTER_PTP_V2_EVENT:
5169 	case HWTSTAMP_FILTER_PTP_V2_SYNC:
5170 	case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
5171 		init->rx_filter = HWTSTAMP_FILTER_ALL;
5172 		rc = efx_ptp_change_mode(efx, true, 0);
5173 		if (!rc)
5174 			rc = efx_ef10_ptp_set_ts_sync_events(efx, true, false);
5175 		if (rc)
5176 			efx_ptp_change_mode(efx, false, 0);
5177 		return rc;
5178 	default:
5179 		return -ERANGE;
5180 	}
5181 }
5182 
5183 static int efx_ef10_vlan_rx_add_vid(struct efx_nic *efx, __be16 proto, u16 vid)
5184 {
5185 	if (proto != htons(ETH_P_8021Q))
5186 		return -EINVAL;
5187 
5188 	return efx_ef10_add_vlan(efx, vid);
5189 }
5190 
5191 static int efx_ef10_vlan_rx_kill_vid(struct efx_nic *efx, __be16 proto, u16 vid)
5192 {
5193 	if (proto != htons(ETH_P_8021Q))
5194 		return -EINVAL;
5195 
5196 	return efx_ef10_del_vlan(efx, vid);
5197 }
5198 
5199 #define EF10_OFFLOAD_FEATURES		\
5200 	(NETIF_F_IP_CSUM |		\
5201 	 NETIF_F_HW_VLAN_CTAG_FILTER |	\
5202 	 NETIF_F_IPV6_CSUM |		\
5203 	 NETIF_F_RXHASH |		\
5204 	 NETIF_F_NTUPLE)
5205 
5206 const struct efx_nic_type efx_hunt_a0_vf_nic_type = {
5207 	.is_vf = true,
5208 	.mem_bar = EFX_MEM_VF_BAR,
5209 	.mem_map_size = efx_ef10_mem_map_size,
5210 	.probe = efx_ef10_probe_vf,
5211 	.remove = efx_ef10_remove,
5212 	.dimension_resources = efx_ef10_dimension_resources,
5213 	.init = efx_ef10_init_nic,
5214 	.fini = efx_port_dummy_op_void,
5215 	.map_reset_reason = efx_ef10_map_reset_reason,
5216 	.map_reset_flags = efx_ef10_map_reset_flags,
5217 	.reset = efx_ef10_reset,
5218 	.probe_port = efx_mcdi_port_probe,
5219 	.remove_port = efx_mcdi_port_remove,
5220 	.fini_dmaq = efx_ef10_fini_dmaq,
5221 	.prepare_flr = efx_ef10_prepare_flr,
5222 	.finish_flr = efx_port_dummy_op_void,
5223 	.describe_stats = efx_ef10_describe_stats,
5224 	.update_stats = efx_ef10_update_stats_vf,
5225 	.start_stats = efx_port_dummy_op_void,
5226 	.pull_stats = efx_port_dummy_op_void,
5227 	.stop_stats = efx_port_dummy_op_void,
5228 	.set_id_led = efx_mcdi_set_id_led,
5229 	.push_irq_moderation = efx_ef10_push_irq_moderation,
5230 	.reconfigure_mac = efx_ef10_mac_reconfigure_vf,
5231 	.check_mac_fault = efx_mcdi_mac_check_fault,
5232 	.reconfigure_port = efx_mcdi_port_reconfigure,
5233 	.get_wol = efx_ef10_get_wol_vf,
5234 	.set_wol = efx_ef10_set_wol_vf,
5235 	.resume_wol = efx_port_dummy_op_void,
5236 	.mcdi_request = efx_ef10_mcdi_request,
5237 	.mcdi_poll_response = efx_ef10_mcdi_poll_response,
5238 	.mcdi_read_response = efx_ef10_mcdi_read_response,
5239 	.mcdi_poll_reboot = efx_ef10_mcdi_poll_reboot,
5240 	.mcdi_reboot_detected = efx_ef10_mcdi_reboot_detected,
5241 	.irq_enable_master = efx_port_dummy_op_void,
5242 	.irq_test_generate = efx_ef10_irq_test_generate,
5243 	.irq_disable_non_ev = efx_port_dummy_op_void,
5244 	.irq_handle_msi = efx_ef10_msi_interrupt,
5245 	.irq_handle_legacy = efx_ef10_legacy_interrupt,
5246 	.tx_probe = efx_ef10_tx_probe,
5247 	.tx_init = efx_ef10_tx_init,
5248 	.tx_remove = efx_ef10_tx_remove,
5249 	.tx_write = efx_ef10_tx_write,
5250 	.rx_push_rss_config = efx_ef10_vf_rx_push_rss_config,
5251 	.rx_probe = efx_ef10_rx_probe,
5252 	.rx_init = efx_ef10_rx_init,
5253 	.rx_remove = efx_ef10_rx_remove,
5254 	.rx_write = efx_ef10_rx_write,
5255 	.rx_defer_refill = efx_ef10_rx_defer_refill,
5256 	.ev_probe = efx_ef10_ev_probe,
5257 	.ev_init = efx_ef10_ev_init,
5258 	.ev_fini = efx_ef10_ev_fini,
5259 	.ev_remove = efx_ef10_ev_remove,
5260 	.ev_process = efx_ef10_ev_process,
5261 	.ev_read_ack = efx_ef10_ev_read_ack,
5262 	.ev_test_generate = efx_ef10_ev_test_generate,
5263 	.filter_table_probe = efx_ef10_filter_table_probe,
5264 	.filter_table_restore = efx_ef10_filter_table_restore,
5265 	.filter_table_remove = efx_ef10_filter_table_remove,
5266 	.filter_update_rx_scatter = efx_ef10_filter_update_rx_scatter,
5267 	.filter_insert = efx_ef10_filter_insert,
5268 	.filter_remove_safe = efx_ef10_filter_remove_safe,
5269 	.filter_get_safe = efx_ef10_filter_get_safe,
5270 	.filter_clear_rx = efx_ef10_filter_clear_rx,
5271 	.filter_count_rx_used = efx_ef10_filter_count_rx_used,
5272 	.filter_get_rx_id_limit = efx_ef10_filter_get_rx_id_limit,
5273 	.filter_get_rx_ids = efx_ef10_filter_get_rx_ids,
5274 #ifdef CONFIG_RFS_ACCEL
5275 	.filter_rfs_insert = efx_ef10_filter_rfs_insert,
5276 	.filter_rfs_expire_one = efx_ef10_filter_rfs_expire_one,
5277 #endif
5278 #ifdef CONFIG_SFC_MTD
5279 	.mtd_probe = efx_port_dummy_op_int,
5280 #endif
5281 	.ptp_write_host_time = efx_ef10_ptp_write_host_time_vf,
5282 	.ptp_set_ts_config = efx_ef10_ptp_set_ts_config_vf,
5283 	.vlan_rx_add_vid = efx_ef10_vlan_rx_add_vid,
5284 	.vlan_rx_kill_vid = efx_ef10_vlan_rx_kill_vid,
5285 #ifdef CONFIG_SFC_SRIOV
5286 	.vswitching_probe = efx_ef10_vswitching_probe_vf,
5287 	.vswitching_restore = efx_ef10_vswitching_restore_vf,
5288 	.vswitching_remove = efx_ef10_vswitching_remove_vf,
5289 	.sriov_get_phys_port_id = efx_ef10_sriov_get_phys_port_id,
5290 #endif
5291 	.get_mac_address = efx_ef10_get_mac_address_vf,
5292 	.set_mac_address = efx_ef10_set_mac_address,
5293 
5294 	.revision = EFX_REV_HUNT_A0,
5295 	.max_dma_mask = DMA_BIT_MASK(ESF_DZ_TX_KER_BUF_ADDR_WIDTH),
5296 	.rx_prefix_size = ES_DZ_RX_PREFIX_SIZE,
5297 	.rx_hash_offset = ES_DZ_RX_PREFIX_HASH_OFST,
5298 	.rx_ts_offset = ES_DZ_RX_PREFIX_TSTAMP_OFST,
5299 	.can_rx_scatter = true,
5300 	.always_rx_scatter = true,
5301 	.max_interrupt_mode = EFX_INT_MODE_MSIX,
5302 	.timer_period_max = 1 << ERF_DD_EVQ_IND_TIMER_VAL_WIDTH,
5303 	.offload_features = EF10_OFFLOAD_FEATURES,
5304 	.mcdi_max_ver = 2,
5305 	.max_rx_ip_filters = HUNT_FILTER_TBL_ROWS,
5306 	.hwtstamp_filters = 1 << HWTSTAMP_FILTER_NONE |
5307 			    1 << HWTSTAMP_FILTER_ALL,
5308 };
5309 
5310 const struct efx_nic_type efx_hunt_a0_nic_type = {
5311 	.is_vf = false,
5312 	.mem_bar = EFX_MEM_BAR,
5313 	.mem_map_size = efx_ef10_mem_map_size,
5314 	.probe = efx_ef10_probe_pf,
5315 	.remove = efx_ef10_remove,
5316 	.dimension_resources = efx_ef10_dimension_resources,
5317 	.init = efx_ef10_init_nic,
5318 	.fini = efx_port_dummy_op_void,
5319 	.map_reset_reason = efx_ef10_map_reset_reason,
5320 	.map_reset_flags = efx_ef10_map_reset_flags,
5321 	.reset = efx_ef10_reset,
5322 	.probe_port = efx_mcdi_port_probe,
5323 	.remove_port = efx_mcdi_port_remove,
5324 	.fini_dmaq = efx_ef10_fini_dmaq,
5325 	.prepare_flr = efx_ef10_prepare_flr,
5326 	.finish_flr = efx_port_dummy_op_void,
5327 	.describe_stats = efx_ef10_describe_stats,
5328 	.update_stats = efx_ef10_update_stats_pf,
5329 	.start_stats = efx_mcdi_mac_start_stats,
5330 	.pull_stats = efx_mcdi_mac_pull_stats,
5331 	.stop_stats = efx_mcdi_mac_stop_stats,
5332 	.set_id_led = efx_mcdi_set_id_led,
5333 	.push_irq_moderation = efx_ef10_push_irq_moderation,
5334 	.reconfigure_mac = efx_ef10_mac_reconfigure,
5335 	.check_mac_fault = efx_mcdi_mac_check_fault,
5336 	.reconfigure_port = efx_mcdi_port_reconfigure,
5337 	.get_wol = efx_ef10_get_wol,
5338 	.set_wol = efx_ef10_set_wol,
5339 	.resume_wol = efx_port_dummy_op_void,
5340 	.test_chip = efx_ef10_test_chip,
5341 	.test_nvram = efx_mcdi_nvram_test_all,
5342 	.mcdi_request = efx_ef10_mcdi_request,
5343 	.mcdi_poll_response = efx_ef10_mcdi_poll_response,
5344 	.mcdi_read_response = efx_ef10_mcdi_read_response,
5345 	.mcdi_poll_reboot = efx_ef10_mcdi_poll_reboot,
5346 	.mcdi_reboot_detected = efx_ef10_mcdi_reboot_detected,
5347 	.irq_enable_master = efx_port_dummy_op_void,
5348 	.irq_test_generate = efx_ef10_irq_test_generate,
5349 	.irq_disable_non_ev = efx_port_dummy_op_void,
5350 	.irq_handle_msi = efx_ef10_msi_interrupt,
5351 	.irq_handle_legacy = efx_ef10_legacy_interrupt,
5352 	.tx_probe = efx_ef10_tx_probe,
5353 	.tx_init = efx_ef10_tx_init,
5354 	.tx_remove = efx_ef10_tx_remove,
5355 	.tx_write = efx_ef10_tx_write,
5356 	.rx_push_rss_config = efx_ef10_pf_rx_push_rss_config,
5357 	.rx_probe = efx_ef10_rx_probe,
5358 	.rx_init = efx_ef10_rx_init,
5359 	.rx_remove = efx_ef10_rx_remove,
5360 	.rx_write = efx_ef10_rx_write,
5361 	.rx_defer_refill = efx_ef10_rx_defer_refill,
5362 	.ev_probe = efx_ef10_ev_probe,
5363 	.ev_init = efx_ef10_ev_init,
5364 	.ev_fini = efx_ef10_ev_fini,
5365 	.ev_remove = efx_ef10_ev_remove,
5366 	.ev_process = efx_ef10_ev_process,
5367 	.ev_read_ack = efx_ef10_ev_read_ack,
5368 	.ev_test_generate = efx_ef10_ev_test_generate,
5369 	.filter_table_probe = efx_ef10_filter_table_probe,
5370 	.filter_table_restore = efx_ef10_filter_table_restore,
5371 	.filter_table_remove = efx_ef10_filter_table_remove,
5372 	.filter_update_rx_scatter = efx_ef10_filter_update_rx_scatter,
5373 	.filter_insert = efx_ef10_filter_insert,
5374 	.filter_remove_safe = efx_ef10_filter_remove_safe,
5375 	.filter_get_safe = efx_ef10_filter_get_safe,
5376 	.filter_clear_rx = efx_ef10_filter_clear_rx,
5377 	.filter_count_rx_used = efx_ef10_filter_count_rx_used,
5378 	.filter_get_rx_id_limit = efx_ef10_filter_get_rx_id_limit,
5379 	.filter_get_rx_ids = efx_ef10_filter_get_rx_ids,
5380 #ifdef CONFIG_RFS_ACCEL
5381 	.filter_rfs_insert = efx_ef10_filter_rfs_insert,
5382 	.filter_rfs_expire_one = efx_ef10_filter_rfs_expire_one,
5383 #endif
5384 #ifdef CONFIG_SFC_MTD
5385 	.mtd_probe = efx_ef10_mtd_probe,
5386 	.mtd_rename = efx_mcdi_mtd_rename,
5387 	.mtd_read = efx_mcdi_mtd_read,
5388 	.mtd_erase = efx_mcdi_mtd_erase,
5389 	.mtd_write = efx_mcdi_mtd_write,
5390 	.mtd_sync = efx_mcdi_mtd_sync,
5391 #endif
5392 	.ptp_write_host_time = efx_ef10_ptp_write_host_time,
5393 	.ptp_set_ts_sync_events = efx_ef10_ptp_set_ts_sync_events,
5394 	.ptp_set_ts_config = efx_ef10_ptp_set_ts_config,
5395 	.vlan_rx_add_vid = efx_ef10_vlan_rx_add_vid,
5396 	.vlan_rx_kill_vid = efx_ef10_vlan_rx_kill_vid,
5397 #ifdef CONFIG_SFC_SRIOV
5398 	.sriov_configure = efx_ef10_sriov_configure,
5399 	.sriov_init = efx_ef10_sriov_init,
5400 	.sriov_fini = efx_ef10_sriov_fini,
5401 	.sriov_wanted = efx_ef10_sriov_wanted,
5402 	.sriov_reset = efx_ef10_sriov_reset,
5403 	.sriov_flr = efx_ef10_sriov_flr,
5404 	.sriov_set_vf_mac = efx_ef10_sriov_set_vf_mac,
5405 	.sriov_set_vf_vlan = efx_ef10_sriov_set_vf_vlan,
5406 	.sriov_set_vf_spoofchk = efx_ef10_sriov_set_vf_spoofchk,
5407 	.sriov_get_vf_config = efx_ef10_sriov_get_vf_config,
5408 	.sriov_set_vf_link_state = efx_ef10_sriov_set_vf_link_state,
5409 	.vswitching_probe = efx_ef10_vswitching_probe_pf,
5410 	.vswitching_restore = efx_ef10_vswitching_restore_pf,
5411 	.vswitching_remove = efx_ef10_vswitching_remove_pf,
5412 #endif
5413 	.get_mac_address = efx_ef10_get_mac_address_pf,
5414 	.set_mac_address = efx_ef10_set_mac_address,
5415 
5416 	.revision = EFX_REV_HUNT_A0,
5417 	.max_dma_mask = DMA_BIT_MASK(ESF_DZ_TX_KER_BUF_ADDR_WIDTH),
5418 	.rx_prefix_size = ES_DZ_RX_PREFIX_SIZE,
5419 	.rx_hash_offset = ES_DZ_RX_PREFIX_HASH_OFST,
5420 	.rx_ts_offset = ES_DZ_RX_PREFIX_TSTAMP_OFST,
5421 	.can_rx_scatter = true,
5422 	.always_rx_scatter = true,
5423 	.max_interrupt_mode = EFX_INT_MODE_MSIX,
5424 	.timer_period_max = 1 << ERF_DD_EVQ_IND_TIMER_VAL_WIDTH,
5425 	.offload_features = EF10_OFFLOAD_FEATURES,
5426 	.mcdi_max_ver = 2,
5427 	.max_rx_ip_filters = HUNT_FILTER_TBL_ROWS,
5428 	.hwtstamp_filters = 1 << HWTSTAMP_FILTER_NONE |
5429 			    1 << HWTSTAMP_FILTER_ALL,
5430 };
5431