xref: /openbmc/linux/drivers/net/ethernet/sfc/ef10.c (revision bc5aa3a0)
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 	BUILD_BUG_ON(BITS_TO_LONGS(EF10_STAT_COUNT) != 2);
1521 	mask[0] = raw_mask[0];
1522 	mask[1] = raw_mask[1];
1523 #else
1524 	BUILD_BUG_ON(BITS_TO_LONGS(EF10_STAT_COUNT) != 3);
1525 	mask[0] = raw_mask[0] & 0xffffffff;
1526 	mask[1] = raw_mask[0] >> 32;
1527 	mask[2] = raw_mask[1] & 0xffffffff;
1528 #endif
1529 }
1530 
1531 static size_t efx_ef10_describe_stats(struct efx_nic *efx, u8 *names)
1532 {
1533 	DECLARE_BITMAP(mask, EF10_STAT_COUNT);
1534 
1535 	efx_ef10_get_stat_mask(efx, mask);
1536 	return efx_nic_describe_stats(efx_ef10_stat_desc, EF10_STAT_COUNT,
1537 				      mask, names);
1538 }
1539 
1540 static size_t efx_ef10_update_stats_common(struct efx_nic *efx, u64 *full_stats,
1541 					   struct rtnl_link_stats64 *core_stats)
1542 {
1543 	DECLARE_BITMAP(mask, EF10_STAT_COUNT);
1544 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1545 	u64 *stats = nic_data->stats;
1546 	size_t stats_count = 0, index;
1547 
1548 	efx_ef10_get_stat_mask(efx, mask);
1549 
1550 	if (full_stats) {
1551 		for_each_set_bit(index, mask, EF10_STAT_COUNT) {
1552 			if (efx_ef10_stat_desc[index].name) {
1553 				*full_stats++ = stats[index];
1554 				++stats_count;
1555 			}
1556 		}
1557 	}
1558 
1559 	if (!core_stats)
1560 		return stats_count;
1561 
1562 	if (nic_data->datapath_caps &
1563 			1 << MC_CMD_GET_CAPABILITIES_OUT_EVB_LBN) {
1564 		/* Use vadaptor stats. */
1565 		core_stats->rx_packets = stats[EF10_STAT_rx_unicast] +
1566 					 stats[EF10_STAT_rx_multicast] +
1567 					 stats[EF10_STAT_rx_broadcast];
1568 		core_stats->tx_packets = stats[EF10_STAT_tx_unicast] +
1569 					 stats[EF10_STAT_tx_multicast] +
1570 					 stats[EF10_STAT_tx_broadcast];
1571 		core_stats->rx_bytes = stats[EF10_STAT_rx_unicast_bytes] +
1572 				       stats[EF10_STAT_rx_multicast_bytes] +
1573 				       stats[EF10_STAT_rx_broadcast_bytes];
1574 		core_stats->tx_bytes = stats[EF10_STAT_tx_unicast_bytes] +
1575 				       stats[EF10_STAT_tx_multicast_bytes] +
1576 				       stats[EF10_STAT_tx_broadcast_bytes];
1577 		core_stats->rx_dropped = stats[GENERIC_STAT_rx_nodesc_trunc] +
1578 					 stats[GENERIC_STAT_rx_noskb_drops];
1579 		core_stats->multicast = stats[EF10_STAT_rx_multicast];
1580 		core_stats->rx_crc_errors = stats[EF10_STAT_rx_bad];
1581 		core_stats->rx_fifo_errors = stats[EF10_STAT_rx_overflow];
1582 		core_stats->rx_errors = core_stats->rx_crc_errors;
1583 		core_stats->tx_errors = stats[EF10_STAT_tx_bad];
1584 	} else {
1585 		/* Use port stats. */
1586 		core_stats->rx_packets = stats[EF10_STAT_port_rx_packets];
1587 		core_stats->tx_packets = stats[EF10_STAT_port_tx_packets];
1588 		core_stats->rx_bytes = stats[EF10_STAT_port_rx_bytes];
1589 		core_stats->tx_bytes = stats[EF10_STAT_port_tx_bytes];
1590 		core_stats->rx_dropped = stats[EF10_STAT_port_rx_nodesc_drops] +
1591 					 stats[GENERIC_STAT_rx_nodesc_trunc] +
1592 					 stats[GENERIC_STAT_rx_noskb_drops];
1593 		core_stats->multicast = stats[EF10_STAT_port_rx_multicast];
1594 		core_stats->rx_length_errors =
1595 				stats[EF10_STAT_port_rx_gtjumbo] +
1596 				stats[EF10_STAT_port_rx_length_error];
1597 		core_stats->rx_crc_errors = stats[EF10_STAT_port_rx_bad];
1598 		core_stats->rx_frame_errors =
1599 				stats[EF10_STAT_port_rx_align_error];
1600 		core_stats->rx_fifo_errors = stats[EF10_STAT_port_rx_overflow];
1601 		core_stats->rx_errors = (core_stats->rx_length_errors +
1602 					 core_stats->rx_crc_errors +
1603 					 core_stats->rx_frame_errors);
1604 	}
1605 
1606 	return stats_count;
1607 }
1608 
1609 static int efx_ef10_try_update_nic_stats_pf(struct efx_nic *efx)
1610 {
1611 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1612 	DECLARE_BITMAP(mask, EF10_STAT_COUNT);
1613 	__le64 generation_start, generation_end;
1614 	u64 *stats = nic_data->stats;
1615 	__le64 *dma_stats;
1616 
1617 	efx_ef10_get_stat_mask(efx, mask);
1618 
1619 	dma_stats = efx->stats_buffer.addr;
1620 	nic_data = efx->nic_data;
1621 
1622 	generation_end = dma_stats[MC_CMD_MAC_GENERATION_END];
1623 	if (generation_end == EFX_MC_STATS_GENERATION_INVALID)
1624 		return 0;
1625 	rmb();
1626 	efx_nic_update_stats(efx_ef10_stat_desc, EF10_STAT_COUNT, mask,
1627 			     stats, efx->stats_buffer.addr, false);
1628 	rmb();
1629 	generation_start = dma_stats[MC_CMD_MAC_GENERATION_START];
1630 	if (generation_end != generation_start)
1631 		return -EAGAIN;
1632 
1633 	/* Update derived statistics */
1634 	efx_nic_fix_nodesc_drop_stat(efx,
1635 				     &stats[EF10_STAT_port_rx_nodesc_drops]);
1636 	stats[EF10_STAT_port_rx_good_bytes] =
1637 		stats[EF10_STAT_port_rx_bytes] -
1638 		stats[EF10_STAT_port_rx_bytes_minus_good_bytes];
1639 	efx_update_diff_stat(&stats[EF10_STAT_port_rx_bad_bytes],
1640 			     stats[EF10_STAT_port_rx_bytes_minus_good_bytes]);
1641 	efx_update_sw_stats(efx, stats);
1642 	return 0;
1643 }
1644 
1645 
1646 static size_t efx_ef10_update_stats_pf(struct efx_nic *efx, u64 *full_stats,
1647 				       struct rtnl_link_stats64 *core_stats)
1648 {
1649 	int retry;
1650 
1651 	/* If we're unlucky enough to read statistics during the DMA, wait
1652 	 * up to 10ms for it to finish (typically takes <500us)
1653 	 */
1654 	for (retry = 0; retry < 100; ++retry) {
1655 		if (efx_ef10_try_update_nic_stats_pf(efx) == 0)
1656 			break;
1657 		udelay(100);
1658 	}
1659 
1660 	return efx_ef10_update_stats_common(efx, full_stats, core_stats);
1661 }
1662 
1663 static int efx_ef10_try_update_nic_stats_vf(struct efx_nic *efx)
1664 {
1665 	MCDI_DECLARE_BUF(inbuf, MC_CMD_MAC_STATS_IN_LEN);
1666 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1667 	DECLARE_BITMAP(mask, EF10_STAT_COUNT);
1668 	__le64 generation_start, generation_end;
1669 	u64 *stats = nic_data->stats;
1670 	u32 dma_len = MC_CMD_MAC_NSTATS * sizeof(u64);
1671 	struct efx_buffer stats_buf;
1672 	__le64 *dma_stats;
1673 	int rc;
1674 
1675 	spin_unlock_bh(&efx->stats_lock);
1676 
1677 	if (in_interrupt()) {
1678 		/* If in atomic context, cannot update stats.  Just update the
1679 		 * software stats and return so the caller can continue.
1680 		 */
1681 		spin_lock_bh(&efx->stats_lock);
1682 		efx_update_sw_stats(efx, stats);
1683 		return 0;
1684 	}
1685 
1686 	efx_ef10_get_stat_mask(efx, mask);
1687 
1688 	rc = efx_nic_alloc_buffer(efx, &stats_buf, dma_len, GFP_ATOMIC);
1689 	if (rc) {
1690 		spin_lock_bh(&efx->stats_lock);
1691 		return rc;
1692 	}
1693 
1694 	dma_stats = stats_buf.addr;
1695 	dma_stats[MC_CMD_MAC_GENERATION_END] = EFX_MC_STATS_GENERATION_INVALID;
1696 
1697 	MCDI_SET_QWORD(inbuf, MAC_STATS_IN_DMA_ADDR, stats_buf.dma_addr);
1698 	MCDI_POPULATE_DWORD_1(inbuf, MAC_STATS_IN_CMD,
1699 			      MAC_STATS_IN_DMA, 1);
1700 	MCDI_SET_DWORD(inbuf, MAC_STATS_IN_DMA_LEN, dma_len);
1701 	MCDI_SET_DWORD(inbuf, MAC_STATS_IN_PORT_ID, EVB_PORT_ID_ASSIGNED);
1702 
1703 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_MAC_STATS, inbuf, sizeof(inbuf),
1704 				NULL, 0, NULL);
1705 	spin_lock_bh(&efx->stats_lock);
1706 	if (rc) {
1707 		/* Expect ENOENT if DMA queues have not been set up */
1708 		if (rc != -ENOENT || atomic_read(&efx->active_queues))
1709 			efx_mcdi_display_error(efx, MC_CMD_MAC_STATS,
1710 					       sizeof(inbuf), NULL, 0, rc);
1711 		goto out;
1712 	}
1713 
1714 	generation_end = dma_stats[MC_CMD_MAC_GENERATION_END];
1715 	if (generation_end == EFX_MC_STATS_GENERATION_INVALID) {
1716 		WARN_ON_ONCE(1);
1717 		goto out;
1718 	}
1719 	rmb();
1720 	efx_nic_update_stats(efx_ef10_stat_desc, EF10_STAT_COUNT, mask,
1721 			     stats, stats_buf.addr, false);
1722 	rmb();
1723 	generation_start = dma_stats[MC_CMD_MAC_GENERATION_START];
1724 	if (generation_end != generation_start) {
1725 		rc = -EAGAIN;
1726 		goto out;
1727 	}
1728 
1729 	efx_update_sw_stats(efx, stats);
1730 out:
1731 	efx_nic_free_buffer(efx, &stats_buf);
1732 	return rc;
1733 }
1734 
1735 static size_t efx_ef10_update_stats_vf(struct efx_nic *efx, u64 *full_stats,
1736 				       struct rtnl_link_stats64 *core_stats)
1737 {
1738 	if (efx_ef10_try_update_nic_stats_vf(efx))
1739 		return 0;
1740 
1741 	return efx_ef10_update_stats_common(efx, full_stats, core_stats);
1742 }
1743 
1744 static void efx_ef10_push_irq_moderation(struct efx_channel *channel)
1745 {
1746 	struct efx_nic *efx = channel->efx;
1747 	unsigned int mode, value;
1748 	efx_dword_t timer_cmd;
1749 
1750 	if (channel->irq_moderation) {
1751 		mode = 3;
1752 		value = channel->irq_moderation - 1;
1753 	} else {
1754 		mode = 0;
1755 		value = 0;
1756 	}
1757 
1758 	if (EFX_EF10_WORKAROUND_35388(efx)) {
1759 		EFX_POPULATE_DWORD_3(timer_cmd, ERF_DD_EVQ_IND_TIMER_FLAGS,
1760 				     EFE_DD_EVQ_IND_TIMER_FLAGS,
1761 				     ERF_DD_EVQ_IND_TIMER_MODE, mode,
1762 				     ERF_DD_EVQ_IND_TIMER_VAL, value);
1763 		efx_writed_page(efx, &timer_cmd, ER_DD_EVQ_INDIRECT,
1764 				channel->channel);
1765 	} else {
1766 		EFX_POPULATE_DWORD_2(timer_cmd, ERF_DZ_TC_TIMER_MODE, mode,
1767 				     ERF_DZ_TC_TIMER_VAL, value);
1768 		efx_writed_page(efx, &timer_cmd, ER_DZ_EVQ_TMR,
1769 				channel->channel);
1770 	}
1771 }
1772 
1773 static void efx_ef10_get_wol_vf(struct efx_nic *efx,
1774 				struct ethtool_wolinfo *wol) {}
1775 
1776 static int efx_ef10_set_wol_vf(struct efx_nic *efx, u32 type)
1777 {
1778 	return -EOPNOTSUPP;
1779 }
1780 
1781 static void efx_ef10_get_wol(struct efx_nic *efx, struct ethtool_wolinfo *wol)
1782 {
1783 	wol->supported = 0;
1784 	wol->wolopts = 0;
1785 	memset(&wol->sopass, 0, sizeof(wol->sopass));
1786 }
1787 
1788 static int efx_ef10_set_wol(struct efx_nic *efx, u32 type)
1789 {
1790 	if (type != 0)
1791 		return -EINVAL;
1792 	return 0;
1793 }
1794 
1795 static void efx_ef10_mcdi_request(struct efx_nic *efx,
1796 				  const efx_dword_t *hdr, size_t hdr_len,
1797 				  const efx_dword_t *sdu, size_t sdu_len)
1798 {
1799 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1800 	u8 *pdu = nic_data->mcdi_buf.addr;
1801 
1802 	memcpy(pdu, hdr, hdr_len);
1803 	memcpy(pdu + hdr_len, sdu, sdu_len);
1804 	wmb();
1805 
1806 	/* The hardware provides 'low' and 'high' (doorbell) registers
1807 	 * for passing the 64-bit address of an MCDI request to
1808 	 * firmware.  However the dwords are swapped by firmware.  The
1809 	 * least significant bits of the doorbell are then 0 for all
1810 	 * MCDI requests due to alignment.
1811 	 */
1812 	_efx_writed(efx, cpu_to_le32((u64)nic_data->mcdi_buf.dma_addr >> 32),
1813 		    ER_DZ_MC_DB_LWRD);
1814 	_efx_writed(efx, cpu_to_le32((u32)nic_data->mcdi_buf.dma_addr),
1815 		    ER_DZ_MC_DB_HWRD);
1816 }
1817 
1818 static bool efx_ef10_mcdi_poll_response(struct efx_nic *efx)
1819 {
1820 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1821 	const efx_dword_t hdr = *(const efx_dword_t *)nic_data->mcdi_buf.addr;
1822 
1823 	rmb();
1824 	return EFX_DWORD_FIELD(hdr, MCDI_HEADER_RESPONSE);
1825 }
1826 
1827 static void
1828 efx_ef10_mcdi_read_response(struct efx_nic *efx, efx_dword_t *outbuf,
1829 			    size_t offset, size_t outlen)
1830 {
1831 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1832 	const u8 *pdu = nic_data->mcdi_buf.addr;
1833 
1834 	memcpy(outbuf, pdu + offset, outlen);
1835 }
1836 
1837 static void efx_ef10_mcdi_reboot_detected(struct efx_nic *efx)
1838 {
1839 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1840 
1841 	/* All our allocations have been reset */
1842 	efx_ef10_reset_mc_allocations(efx);
1843 
1844 	/* The datapath firmware might have been changed */
1845 	nic_data->must_check_datapath_caps = true;
1846 
1847 	/* MAC statistics have been cleared on the NIC; clear the local
1848 	 * statistic that we update with efx_update_diff_stat().
1849 	 */
1850 	nic_data->stats[EF10_STAT_port_rx_bad_bytes] = 0;
1851 }
1852 
1853 static int efx_ef10_mcdi_poll_reboot(struct efx_nic *efx)
1854 {
1855 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1856 	int rc;
1857 
1858 	rc = efx_ef10_get_warm_boot_count(efx);
1859 	if (rc < 0) {
1860 		/* The firmware is presumably in the process of
1861 		 * rebooting.  However, we are supposed to report each
1862 		 * reboot just once, so we must only do that once we
1863 		 * can read and store the updated warm boot count.
1864 		 */
1865 		return 0;
1866 	}
1867 
1868 	if (rc == nic_data->warm_boot_count)
1869 		return 0;
1870 
1871 	nic_data->warm_boot_count = rc;
1872 	efx_ef10_mcdi_reboot_detected(efx);
1873 
1874 	return -EIO;
1875 }
1876 
1877 /* Handle an MSI interrupt
1878  *
1879  * Handle an MSI hardware interrupt.  This routine schedules event
1880  * queue processing.  No interrupt acknowledgement cycle is necessary.
1881  * Also, we never need to check that the interrupt is for us, since
1882  * MSI interrupts cannot be shared.
1883  */
1884 static irqreturn_t efx_ef10_msi_interrupt(int irq, void *dev_id)
1885 {
1886 	struct efx_msi_context *context = dev_id;
1887 	struct efx_nic *efx = context->efx;
1888 
1889 	netif_vdbg(efx, intr, efx->net_dev,
1890 		   "IRQ %d on CPU %d\n", irq, raw_smp_processor_id());
1891 
1892 	if (likely(ACCESS_ONCE(efx->irq_soft_enabled))) {
1893 		/* Note test interrupts */
1894 		if (context->index == efx->irq_level)
1895 			efx->last_irq_cpu = raw_smp_processor_id();
1896 
1897 		/* Schedule processing of the channel */
1898 		efx_schedule_channel_irq(efx->channel[context->index]);
1899 	}
1900 
1901 	return IRQ_HANDLED;
1902 }
1903 
1904 static irqreturn_t efx_ef10_legacy_interrupt(int irq, void *dev_id)
1905 {
1906 	struct efx_nic *efx = dev_id;
1907 	bool soft_enabled = ACCESS_ONCE(efx->irq_soft_enabled);
1908 	struct efx_channel *channel;
1909 	efx_dword_t reg;
1910 	u32 queues;
1911 
1912 	/* Read the ISR which also ACKs the interrupts */
1913 	efx_readd(efx, &reg, ER_DZ_BIU_INT_ISR);
1914 	queues = EFX_DWORD_FIELD(reg, ERF_DZ_ISR_REG);
1915 
1916 	if (queues == 0)
1917 		return IRQ_NONE;
1918 
1919 	if (likely(soft_enabled)) {
1920 		/* Note test interrupts */
1921 		if (queues & (1U << efx->irq_level))
1922 			efx->last_irq_cpu = raw_smp_processor_id();
1923 
1924 		efx_for_each_channel(channel, efx) {
1925 			if (queues & 1)
1926 				efx_schedule_channel_irq(channel);
1927 			queues >>= 1;
1928 		}
1929 	}
1930 
1931 	netif_vdbg(efx, intr, efx->net_dev,
1932 		   "IRQ %d on CPU %d status " EFX_DWORD_FMT "\n",
1933 		   irq, raw_smp_processor_id(), EFX_DWORD_VAL(reg));
1934 
1935 	return IRQ_HANDLED;
1936 }
1937 
1938 static void efx_ef10_irq_test_generate(struct efx_nic *efx)
1939 {
1940 	MCDI_DECLARE_BUF(inbuf, MC_CMD_TRIGGER_INTERRUPT_IN_LEN);
1941 
1942 	BUILD_BUG_ON(MC_CMD_TRIGGER_INTERRUPT_OUT_LEN != 0);
1943 
1944 	MCDI_SET_DWORD(inbuf, TRIGGER_INTERRUPT_IN_INTR_LEVEL, efx->irq_level);
1945 	(void) efx_mcdi_rpc(efx, MC_CMD_TRIGGER_INTERRUPT,
1946 			    inbuf, sizeof(inbuf), NULL, 0, NULL);
1947 }
1948 
1949 static int efx_ef10_tx_probe(struct efx_tx_queue *tx_queue)
1950 {
1951 	return efx_nic_alloc_buffer(tx_queue->efx, &tx_queue->txd.buf,
1952 				    (tx_queue->ptr_mask + 1) *
1953 				    sizeof(efx_qword_t),
1954 				    GFP_KERNEL);
1955 }
1956 
1957 /* This writes to the TX_DESC_WPTR and also pushes data */
1958 static inline void efx_ef10_push_tx_desc(struct efx_tx_queue *tx_queue,
1959 					 const efx_qword_t *txd)
1960 {
1961 	unsigned int write_ptr;
1962 	efx_oword_t reg;
1963 
1964 	write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
1965 	EFX_POPULATE_OWORD_1(reg, ERF_DZ_TX_DESC_WPTR, write_ptr);
1966 	reg.qword[0] = *txd;
1967 	efx_writeo_page(tx_queue->efx, &reg,
1968 			ER_DZ_TX_DESC_UPD, tx_queue->queue);
1969 }
1970 
1971 static void efx_ef10_tx_init(struct efx_tx_queue *tx_queue)
1972 {
1973 	MCDI_DECLARE_BUF(inbuf, MC_CMD_INIT_TXQ_IN_LEN(EFX_MAX_DMAQ_SIZE * 8 /
1974 						       EFX_BUF_SIZE));
1975 	bool csum_offload = tx_queue->queue & EFX_TXQ_TYPE_OFFLOAD;
1976 	size_t entries = tx_queue->txd.buf.len / EFX_BUF_SIZE;
1977 	struct efx_channel *channel = tx_queue->channel;
1978 	struct efx_nic *efx = tx_queue->efx;
1979 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1980 	size_t inlen;
1981 	dma_addr_t dma_addr;
1982 	efx_qword_t *txd;
1983 	int rc;
1984 	int i;
1985 	BUILD_BUG_ON(MC_CMD_INIT_TXQ_OUT_LEN != 0);
1986 
1987 	MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_SIZE, tx_queue->ptr_mask + 1);
1988 	MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_TARGET_EVQ, channel->channel);
1989 	MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_LABEL, tx_queue->queue);
1990 	MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_INSTANCE, tx_queue->queue);
1991 	MCDI_POPULATE_DWORD_2(inbuf, INIT_TXQ_IN_FLAGS,
1992 			      INIT_TXQ_IN_FLAG_IP_CSUM_DIS, !csum_offload,
1993 			      INIT_TXQ_IN_FLAG_TCP_CSUM_DIS, !csum_offload);
1994 	MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_OWNER_ID, 0);
1995 	MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_PORT_ID, nic_data->vport_id);
1996 
1997 	dma_addr = tx_queue->txd.buf.dma_addr;
1998 
1999 	netif_dbg(efx, hw, efx->net_dev, "pushing TXQ %d. %zu entries (%llx)\n",
2000 		  tx_queue->queue, entries, (u64)dma_addr);
2001 
2002 	for (i = 0; i < entries; ++i) {
2003 		MCDI_SET_ARRAY_QWORD(inbuf, INIT_TXQ_IN_DMA_ADDR, i, dma_addr);
2004 		dma_addr += EFX_BUF_SIZE;
2005 	}
2006 
2007 	inlen = MC_CMD_INIT_TXQ_IN_LEN(entries);
2008 
2009 	rc = efx_mcdi_rpc(efx, MC_CMD_INIT_TXQ, inbuf, inlen,
2010 			  NULL, 0, NULL);
2011 	if (rc)
2012 		goto fail;
2013 
2014 	/* A previous user of this TX queue might have set us up the
2015 	 * bomb by writing a descriptor to the TX push collector but
2016 	 * not the doorbell.  (Each collector belongs to a port, not a
2017 	 * queue or function, so cannot easily be reset.)  We must
2018 	 * attempt to push a no-op descriptor in its place.
2019 	 */
2020 	tx_queue->buffer[0].flags = EFX_TX_BUF_OPTION;
2021 	tx_queue->insert_count = 1;
2022 	txd = efx_tx_desc(tx_queue, 0);
2023 	EFX_POPULATE_QWORD_4(*txd,
2024 			     ESF_DZ_TX_DESC_IS_OPT, true,
2025 			     ESF_DZ_TX_OPTION_TYPE,
2026 			     ESE_DZ_TX_OPTION_DESC_CRC_CSUM,
2027 			     ESF_DZ_TX_OPTION_UDP_TCP_CSUM, csum_offload,
2028 			     ESF_DZ_TX_OPTION_IP_CSUM, csum_offload);
2029 	tx_queue->write_count = 1;
2030 
2031 	if (nic_data->datapath_caps &
2032 	    (1 << MC_CMD_GET_CAPABILITIES_OUT_TX_TSO_LBN)) {
2033 		tx_queue->tso_version = 1;
2034 	}
2035 
2036 	wmb();
2037 	efx_ef10_push_tx_desc(tx_queue, txd);
2038 
2039 	return;
2040 
2041 fail:
2042 	netdev_WARN(efx->net_dev, "failed to initialise TXQ %d\n",
2043 		    tx_queue->queue);
2044 }
2045 
2046 static void efx_ef10_tx_fini(struct efx_tx_queue *tx_queue)
2047 {
2048 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FINI_TXQ_IN_LEN);
2049 	MCDI_DECLARE_BUF_ERR(outbuf);
2050 	struct efx_nic *efx = tx_queue->efx;
2051 	size_t outlen;
2052 	int rc;
2053 
2054 	MCDI_SET_DWORD(inbuf, FINI_TXQ_IN_INSTANCE,
2055 		       tx_queue->queue);
2056 
2057 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FINI_TXQ, inbuf, sizeof(inbuf),
2058 			  outbuf, sizeof(outbuf), &outlen);
2059 
2060 	if (rc && rc != -EALREADY)
2061 		goto fail;
2062 
2063 	return;
2064 
2065 fail:
2066 	efx_mcdi_display_error(efx, MC_CMD_FINI_TXQ, MC_CMD_FINI_TXQ_IN_LEN,
2067 			       outbuf, outlen, rc);
2068 }
2069 
2070 static void efx_ef10_tx_remove(struct efx_tx_queue *tx_queue)
2071 {
2072 	efx_nic_free_buffer(tx_queue->efx, &tx_queue->txd.buf);
2073 }
2074 
2075 /* This writes to the TX_DESC_WPTR; write pointer for TX descriptor ring */
2076 static inline void efx_ef10_notify_tx_desc(struct efx_tx_queue *tx_queue)
2077 {
2078 	unsigned int write_ptr;
2079 	efx_dword_t reg;
2080 
2081 	write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
2082 	EFX_POPULATE_DWORD_1(reg, ERF_DZ_TX_DESC_WPTR_DWORD, write_ptr);
2083 	efx_writed_page(tx_queue->efx, &reg,
2084 			ER_DZ_TX_DESC_UPD_DWORD, tx_queue->queue);
2085 }
2086 
2087 static void efx_ef10_tx_write(struct efx_tx_queue *tx_queue)
2088 {
2089 	unsigned int old_write_count = tx_queue->write_count;
2090 	struct efx_tx_buffer *buffer;
2091 	unsigned int write_ptr;
2092 	efx_qword_t *txd;
2093 
2094 	tx_queue->xmit_more_available = false;
2095 	if (unlikely(tx_queue->write_count == tx_queue->insert_count))
2096 		return;
2097 
2098 	do {
2099 		write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
2100 		buffer = &tx_queue->buffer[write_ptr];
2101 		txd = efx_tx_desc(tx_queue, write_ptr);
2102 		++tx_queue->write_count;
2103 
2104 		/* Create TX descriptor ring entry */
2105 		if (buffer->flags & EFX_TX_BUF_OPTION) {
2106 			*txd = buffer->option;
2107 		} else {
2108 			BUILD_BUG_ON(EFX_TX_BUF_CONT != 1);
2109 			EFX_POPULATE_QWORD_3(
2110 				*txd,
2111 				ESF_DZ_TX_KER_CONT,
2112 				buffer->flags & EFX_TX_BUF_CONT,
2113 				ESF_DZ_TX_KER_BYTE_CNT, buffer->len,
2114 				ESF_DZ_TX_KER_BUF_ADDR, buffer->dma_addr);
2115 		}
2116 	} while (tx_queue->write_count != tx_queue->insert_count);
2117 
2118 	wmb(); /* Ensure descriptors are written before they are fetched */
2119 
2120 	if (efx_nic_may_push_tx_desc(tx_queue, old_write_count)) {
2121 		txd = efx_tx_desc(tx_queue,
2122 				  old_write_count & tx_queue->ptr_mask);
2123 		efx_ef10_push_tx_desc(tx_queue, txd);
2124 		++tx_queue->pushes;
2125 	} else {
2126 		efx_ef10_notify_tx_desc(tx_queue);
2127 	}
2128 }
2129 
2130 static int efx_ef10_alloc_rss_context(struct efx_nic *efx, u32 *context,
2131 				      bool exclusive, unsigned *context_size)
2132 {
2133 	MCDI_DECLARE_BUF(inbuf, MC_CMD_RSS_CONTEXT_ALLOC_IN_LEN);
2134 	MCDI_DECLARE_BUF(outbuf, MC_CMD_RSS_CONTEXT_ALLOC_OUT_LEN);
2135 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2136 	size_t outlen;
2137 	int rc;
2138 	u32 alloc_type = exclusive ?
2139 				MC_CMD_RSS_CONTEXT_ALLOC_IN_TYPE_EXCLUSIVE :
2140 				MC_CMD_RSS_CONTEXT_ALLOC_IN_TYPE_SHARED;
2141 	unsigned rss_spread = exclusive ?
2142 				efx->rss_spread :
2143 				min(rounddown_pow_of_two(efx->rss_spread),
2144 				    EFX_EF10_MAX_SHARED_RSS_CONTEXT_SIZE);
2145 
2146 	if (!exclusive && rss_spread == 1) {
2147 		*context = EFX_EF10_RSS_CONTEXT_INVALID;
2148 		if (context_size)
2149 			*context_size = 1;
2150 		return 0;
2151 	}
2152 
2153 	if (nic_data->datapath_caps &
2154 	    1 << MC_CMD_GET_CAPABILITIES_OUT_RX_RSS_LIMITED_LBN)
2155 		return -EOPNOTSUPP;
2156 
2157 	MCDI_SET_DWORD(inbuf, RSS_CONTEXT_ALLOC_IN_UPSTREAM_PORT_ID,
2158 		       nic_data->vport_id);
2159 	MCDI_SET_DWORD(inbuf, RSS_CONTEXT_ALLOC_IN_TYPE, alloc_type);
2160 	MCDI_SET_DWORD(inbuf, RSS_CONTEXT_ALLOC_IN_NUM_QUEUES, rss_spread);
2161 
2162 	rc = efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_ALLOC, inbuf, sizeof(inbuf),
2163 		outbuf, sizeof(outbuf), &outlen);
2164 	if (rc != 0)
2165 		return rc;
2166 
2167 	if (outlen < MC_CMD_RSS_CONTEXT_ALLOC_OUT_LEN)
2168 		return -EIO;
2169 
2170 	*context = MCDI_DWORD(outbuf, RSS_CONTEXT_ALLOC_OUT_RSS_CONTEXT_ID);
2171 
2172 	if (context_size)
2173 		*context_size = rss_spread;
2174 
2175 	return 0;
2176 }
2177 
2178 static void efx_ef10_free_rss_context(struct efx_nic *efx, u32 context)
2179 {
2180 	MCDI_DECLARE_BUF(inbuf, MC_CMD_RSS_CONTEXT_FREE_IN_LEN);
2181 	int rc;
2182 
2183 	MCDI_SET_DWORD(inbuf, RSS_CONTEXT_FREE_IN_RSS_CONTEXT_ID,
2184 		       context);
2185 
2186 	rc = efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_FREE, inbuf, sizeof(inbuf),
2187 			    NULL, 0, NULL);
2188 	WARN_ON(rc != 0);
2189 }
2190 
2191 static int efx_ef10_populate_rss_table(struct efx_nic *efx, u32 context,
2192 				       const u32 *rx_indir_table)
2193 {
2194 	MCDI_DECLARE_BUF(tablebuf, MC_CMD_RSS_CONTEXT_SET_TABLE_IN_LEN);
2195 	MCDI_DECLARE_BUF(keybuf, MC_CMD_RSS_CONTEXT_SET_KEY_IN_LEN);
2196 	int i, rc;
2197 
2198 	MCDI_SET_DWORD(tablebuf, RSS_CONTEXT_SET_TABLE_IN_RSS_CONTEXT_ID,
2199 		       context);
2200 	BUILD_BUG_ON(ARRAY_SIZE(efx->rx_indir_table) !=
2201 		     MC_CMD_RSS_CONTEXT_SET_TABLE_IN_INDIRECTION_TABLE_LEN);
2202 
2203 	for (i = 0; i < ARRAY_SIZE(efx->rx_indir_table); ++i)
2204 		MCDI_PTR(tablebuf,
2205 			 RSS_CONTEXT_SET_TABLE_IN_INDIRECTION_TABLE)[i] =
2206 				(u8) rx_indir_table[i];
2207 
2208 	rc = efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_SET_TABLE, tablebuf,
2209 			  sizeof(tablebuf), NULL, 0, NULL);
2210 	if (rc != 0)
2211 		return rc;
2212 
2213 	MCDI_SET_DWORD(keybuf, RSS_CONTEXT_SET_KEY_IN_RSS_CONTEXT_ID,
2214 		       context);
2215 	BUILD_BUG_ON(ARRAY_SIZE(efx->rx_hash_key) !=
2216 		     MC_CMD_RSS_CONTEXT_SET_KEY_IN_TOEPLITZ_KEY_LEN);
2217 	for (i = 0; i < ARRAY_SIZE(efx->rx_hash_key); ++i)
2218 		MCDI_PTR(keybuf, RSS_CONTEXT_SET_KEY_IN_TOEPLITZ_KEY)[i] =
2219 			efx->rx_hash_key[i];
2220 
2221 	return efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_SET_KEY, keybuf,
2222 			    sizeof(keybuf), NULL, 0, NULL);
2223 }
2224 
2225 static void efx_ef10_rx_free_indir_table(struct efx_nic *efx)
2226 {
2227 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2228 
2229 	if (nic_data->rx_rss_context != EFX_EF10_RSS_CONTEXT_INVALID)
2230 		efx_ef10_free_rss_context(efx, nic_data->rx_rss_context);
2231 	nic_data->rx_rss_context = EFX_EF10_RSS_CONTEXT_INVALID;
2232 }
2233 
2234 static int efx_ef10_rx_push_shared_rss_config(struct efx_nic *efx,
2235 					      unsigned *context_size)
2236 {
2237 	u32 new_rx_rss_context;
2238 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2239 	int rc = efx_ef10_alloc_rss_context(efx, &new_rx_rss_context,
2240 					    false, context_size);
2241 
2242 	if (rc != 0)
2243 		return rc;
2244 
2245 	nic_data->rx_rss_context = new_rx_rss_context;
2246 	nic_data->rx_rss_context_exclusive = false;
2247 	efx_set_default_rx_indir_table(efx);
2248 	return 0;
2249 }
2250 
2251 static int efx_ef10_rx_push_exclusive_rss_config(struct efx_nic *efx,
2252 						 const u32 *rx_indir_table)
2253 {
2254 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2255 	int rc;
2256 	u32 new_rx_rss_context;
2257 
2258 	if (nic_data->rx_rss_context == EFX_EF10_RSS_CONTEXT_INVALID ||
2259 	    !nic_data->rx_rss_context_exclusive) {
2260 		rc = efx_ef10_alloc_rss_context(efx, &new_rx_rss_context,
2261 						true, NULL);
2262 		if (rc == -EOPNOTSUPP)
2263 			return rc;
2264 		else if (rc != 0)
2265 			goto fail1;
2266 	} else {
2267 		new_rx_rss_context = nic_data->rx_rss_context;
2268 	}
2269 
2270 	rc = efx_ef10_populate_rss_table(efx, new_rx_rss_context,
2271 					 rx_indir_table);
2272 	if (rc != 0)
2273 		goto fail2;
2274 
2275 	if (nic_data->rx_rss_context != new_rx_rss_context)
2276 		efx_ef10_rx_free_indir_table(efx);
2277 	nic_data->rx_rss_context = new_rx_rss_context;
2278 	nic_data->rx_rss_context_exclusive = true;
2279 	if (rx_indir_table != efx->rx_indir_table)
2280 		memcpy(efx->rx_indir_table, rx_indir_table,
2281 		       sizeof(efx->rx_indir_table));
2282 	return 0;
2283 
2284 fail2:
2285 	if (new_rx_rss_context != nic_data->rx_rss_context)
2286 		efx_ef10_free_rss_context(efx, new_rx_rss_context);
2287 fail1:
2288 	netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
2289 	return rc;
2290 }
2291 
2292 static int efx_ef10_pf_rx_push_rss_config(struct efx_nic *efx, bool user,
2293 					  const u32 *rx_indir_table)
2294 {
2295 	int rc;
2296 
2297 	if (efx->rss_spread == 1)
2298 		return 0;
2299 
2300 	rc = efx_ef10_rx_push_exclusive_rss_config(efx, rx_indir_table);
2301 
2302 	if (rc == -ENOBUFS && !user) {
2303 		unsigned context_size;
2304 		bool mismatch = false;
2305 		size_t i;
2306 
2307 		for (i = 0; i < ARRAY_SIZE(efx->rx_indir_table) && !mismatch;
2308 		     i++)
2309 			mismatch = rx_indir_table[i] !=
2310 				ethtool_rxfh_indir_default(i, efx->rss_spread);
2311 
2312 		rc = efx_ef10_rx_push_shared_rss_config(efx, &context_size);
2313 		if (rc == 0) {
2314 			if (context_size != efx->rss_spread)
2315 				netif_warn(efx, probe, efx->net_dev,
2316 					   "Could not allocate an exclusive RSS"
2317 					   " context; allocated a shared one of"
2318 					   " different size."
2319 					   " Wanted %u, got %u.\n",
2320 					   efx->rss_spread, context_size);
2321 			else if (mismatch)
2322 				netif_warn(efx, probe, efx->net_dev,
2323 					   "Could not allocate an exclusive RSS"
2324 					   " context; allocated a shared one but"
2325 					   " could not apply custom"
2326 					   " indirection.\n");
2327 			else
2328 				netif_info(efx, probe, efx->net_dev,
2329 					   "Could not allocate an exclusive RSS"
2330 					   " context; allocated a shared one.\n");
2331 		}
2332 	}
2333 	return rc;
2334 }
2335 
2336 static int efx_ef10_vf_rx_push_rss_config(struct efx_nic *efx, bool user,
2337 					  const u32 *rx_indir_table
2338 					  __attribute__ ((unused)))
2339 {
2340 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2341 
2342 	if (user)
2343 		return -EOPNOTSUPP;
2344 	if (nic_data->rx_rss_context != EFX_EF10_RSS_CONTEXT_INVALID)
2345 		return 0;
2346 	return efx_ef10_rx_push_shared_rss_config(efx, NULL);
2347 }
2348 
2349 static int efx_ef10_rx_probe(struct efx_rx_queue *rx_queue)
2350 {
2351 	return efx_nic_alloc_buffer(rx_queue->efx, &rx_queue->rxd.buf,
2352 				    (rx_queue->ptr_mask + 1) *
2353 				    sizeof(efx_qword_t),
2354 				    GFP_KERNEL);
2355 }
2356 
2357 static void efx_ef10_rx_init(struct efx_rx_queue *rx_queue)
2358 {
2359 	MCDI_DECLARE_BUF(inbuf,
2360 			 MC_CMD_INIT_RXQ_IN_LEN(EFX_MAX_DMAQ_SIZE * 8 /
2361 						EFX_BUF_SIZE));
2362 	struct efx_channel *channel = efx_rx_queue_channel(rx_queue);
2363 	size_t entries = rx_queue->rxd.buf.len / EFX_BUF_SIZE;
2364 	struct efx_nic *efx = rx_queue->efx;
2365 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2366 	size_t inlen;
2367 	dma_addr_t dma_addr;
2368 	int rc;
2369 	int i;
2370 	BUILD_BUG_ON(MC_CMD_INIT_RXQ_OUT_LEN != 0);
2371 
2372 	rx_queue->scatter_n = 0;
2373 	rx_queue->scatter_len = 0;
2374 
2375 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_SIZE, rx_queue->ptr_mask + 1);
2376 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_TARGET_EVQ, channel->channel);
2377 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_LABEL, efx_rx_queue_index(rx_queue));
2378 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_INSTANCE,
2379 		       efx_rx_queue_index(rx_queue));
2380 	MCDI_POPULATE_DWORD_2(inbuf, INIT_RXQ_IN_FLAGS,
2381 			      INIT_RXQ_IN_FLAG_PREFIX, 1,
2382 			      INIT_RXQ_IN_FLAG_TIMESTAMP, 1);
2383 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_OWNER_ID, 0);
2384 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_PORT_ID, nic_data->vport_id);
2385 
2386 	dma_addr = rx_queue->rxd.buf.dma_addr;
2387 
2388 	netif_dbg(efx, hw, efx->net_dev, "pushing RXQ %d. %zu entries (%llx)\n",
2389 		  efx_rx_queue_index(rx_queue), entries, (u64)dma_addr);
2390 
2391 	for (i = 0; i < entries; ++i) {
2392 		MCDI_SET_ARRAY_QWORD(inbuf, INIT_RXQ_IN_DMA_ADDR, i, dma_addr);
2393 		dma_addr += EFX_BUF_SIZE;
2394 	}
2395 
2396 	inlen = MC_CMD_INIT_RXQ_IN_LEN(entries);
2397 
2398 	rc = efx_mcdi_rpc(efx, MC_CMD_INIT_RXQ, inbuf, inlen,
2399 			  NULL, 0, NULL);
2400 	if (rc)
2401 		netdev_WARN(efx->net_dev, "failed to initialise RXQ %d\n",
2402 			    efx_rx_queue_index(rx_queue));
2403 }
2404 
2405 static void efx_ef10_rx_fini(struct efx_rx_queue *rx_queue)
2406 {
2407 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FINI_RXQ_IN_LEN);
2408 	MCDI_DECLARE_BUF_ERR(outbuf);
2409 	struct efx_nic *efx = rx_queue->efx;
2410 	size_t outlen;
2411 	int rc;
2412 
2413 	MCDI_SET_DWORD(inbuf, FINI_RXQ_IN_INSTANCE,
2414 		       efx_rx_queue_index(rx_queue));
2415 
2416 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FINI_RXQ, inbuf, sizeof(inbuf),
2417 			  outbuf, sizeof(outbuf), &outlen);
2418 
2419 	if (rc && rc != -EALREADY)
2420 		goto fail;
2421 
2422 	return;
2423 
2424 fail:
2425 	efx_mcdi_display_error(efx, MC_CMD_FINI_RXQ, MC_CMD_FINI_RXQ_IN_LEN,
2426 			       outbuf, outlen, rc);
2427 }
2428 
2429 static void efx_ef10_rx_remove(struct efx_rx_queue *rx_queue)
2430 {
2431 	efx_nic_free_buffer(rx_queue->efx, &rx_queue->rxd.buf);
2432 }
2433 
2434 /* This creates an entry in the RX descriptor queue */
2435 static inline void
2436 efx_ef10_build_rx_desc(struct efx_rx_queue *rx_queue, unsigned int index)
2437 {
2438 	struct efx_rx_buffer *rx_buf;
2439 	efx_qword_t *rxd;
2440 
2441 	rxd = efx_rx_desc(rx_queue, index);
2442 	rx_buf = efx_rx_buffer(rx_queue, index);
2443 	EFX_POPULATE_QWORD_2(*rxd,
2444 			     ESF_DZ_RX_KER_BYTE_CNT, rx_buf->len,
2445 			     ESF_DZ_RX_KER_BUF_ADDR, rx_buf->dma_addr);
2446 }
2447 
2448 static void efx_ef10_rx_write(struct efx_rx_queue *rx_queue)
2449 {
2450 	struct efx_nic *efx = rx_queue->efx;
2451 	unsigned int write_count;
2452 	efx_dword_t reg;
2453 
2454 	/* Firmware requires that RX_DESC_WPTR be a multiple of 8 */
2455 	write_count = rx_queue->added_count & ~7;
2456 	if (rx_queue->notified_count == write_count)
2457 		return;
2458 
2459 	do
2460 		efx_ef10_build_rx_desc(
2461 			rx_queue,
2462 			rx_queue->notified_count & rx_queue->ptr_mask);
2463 	while (++rx_queue->notified_count != write_count);
2464 
2465 	wmb();
2466 	EFX_POPULATE_DWORD_1(reg, ERF_DZ_RX_DESC_WPTR,
2467 			     write_count & rx_queue->ptr_mask);
2468 	efx_writed_page(efx, &reg, ER_DZ_RX_DESC_UPD,
2469 			efx_rx_queue_index(rx_queue));
2470 }
2471 
2472 static efx_mcdi_async_completer efx_ef10_rx_defer_refill_complete;
2473 
2474 static void efx_ef10_rx_defer_refill(struct efx_rx_queue *rx_queue)
2475 {
2476 	struct efx_channel *channel = efx_rx_queue_channel(rx_queue);
2477 	MCDI_DECLARE_BUF(inbuf, MC_CMD_DRIVER_EVENT_IN_LEN);
2478 	efx_qword_t event;
2479 
2480 	EFX_POPULATE_QWORD_2(event,
2481 			     ESF_DZ_EV_CODE, EFX_EF10_DRVGEN_EV,
2482 			     ESF_DZ_EV_DATA, EFX_EF10_REFILL);
2483 
2484 	MCDI_SET_DWORD(inbuf, DRIVER_EVENT_IN_EVQ, channel->channel);
2485 
2486 	/* MCDI_SET_QWORD is not appropriate here since EFX_POPULATE_* has
2487 	 * already swapped the data to little-endian order.
2488 	 */
2489 	memcpy(MCDI_PTR(inbuf, DRIVER_EVENT_IN_DATA), &event.u64[0],
2490 	       sizeof(efx_qword_t));
2491 
2492 	efx_mcdi_rpc_async(channel->efx, MC_CMD_DRIVER_EVENT,
2493 			   inbuf, sizeof(inbuf), 0,
2494 			   efx_ef10_rx_defer_refill_complete, 0);
2495 }
2496 
2497 static void
2498 efx_ef10_rx_defer_refill_complete(struct efx_nic *efx, unsigned long cookie,
2499 				  int rc, efx_dword_t *outbuf,
2500 				  size_t outlen_actual)
2501 {
2502 	/* nothing to do */
2503 }
2504 
2505 static int efx_ef10_ev_probe(struct efx_channel *channel)
2506 {
2507 	return efx_nic_alloc_buffer(channel->efx, &channel->eventq.buf,
2508 				    (channel->eventq_mask + 1) *
2509 				    sizeof(efx_qword_t),
2510 				    GFP_KERNEL);
2511 }
2512 
2513 static void efx_ef10_ev_fini(struct efx_channel *channel)
2514 {
2515 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FINI_EVQ_IN_LEN);
2516 	MCDI_DECLARE_BUF_ERR(outbuf);
2517 	struct efx_nic *efx = channel->efx;
2518 	size_t outlen;
2519 	int rc;
2520 
2521 	MCDI_SET_DWORD(inbuf, FINI_EVQ_IN_INSTANCE, channel->channel);
2522 
2523 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FINI_EVQ, inbuf, sizeof(inbuf),
2524 			  outbuf, sizeof(outbuf), &outlen);
2525 
2526 	if (rc && rc != -EALREADY)
2527 		goto fail;
2528 
2529 	return;
2530 
2531 fail:
2532 	efx_mcdi_display_error(efx, MC_CMD_FINI_EVQ, MC_CMD_FINI_EVQ_IN_LEN,
2533 			       outbuf, outlen, rc);
2534 }
2535 
2536 static int efx_ef10_ev_init(struct efx_channel *channel)
2537 {
2538 	MCDI_DECLARE_BUF(inbuf,
2539 			 MC_CMD_INIT_EVQ_IN_LEN(EFX_MAX_EVQ_SIZE * 8 /
2540 						EFX_BUF_SIZE));
2541 	MCDI_DECLARE_BUF(outbuf, MC_CMD_INIT_EVQ_OUT_LEN);
2542 	size_t entries = channel->eventq.buf.len / EFX_BUF_SIZE;
2543 	struct efx_nic *efx = channel->efx;
2544 	struct efx_ef10_nic_data *nic_data;
2545 	bool supports_rx_merge;
2546 	size_t inlen, outlen;
2547 	unsigned int enabled, implemented;
2548 	dma_addr_t dma_addr;
2549 	int rc;
2550 	int i;
2551 
2552 	nic_data = efx->nic_data;
2553 	supports_rx_merge =
2554 		!!(nic_data->datapath_caps &
2555 		   1 << MC_CMD_GET_CAPABILITIES_OUT_RX_BATCHING_LBN);
2556 
2557 	/* Fill event queue with all ones (i.e. empty events) */
2558 	memset(channel->eventq.buf.addr, 0xff, channel->eventq.buf.len);
2559 
2560 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_SIZE, channel->eventq_mask + 1);
2561 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_INSTANCE, channel->channel);
2562 	/* INIT_EVQ expects index in vector table, not absolute */
2563 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_IRQ_NUM, channel->channel);
2564 	MCDI_POPULATE_DWORD_4(inbuf, INIT_EVQ_IN_FLAGS,
2565 			      INIT_EVQ_IN_FLAG_INTERRUPTING, 1,
2566 			      INIT_EVQ_IN_FLAG_RX_MERGE, 1,
2567 			      INIT_EVQ_IN_FLAG_TX_MERGE, 1,
2568 			      INIT_EVQ_IN_FLAG_CUT_THRU, !supports_rx_merge);
2569 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_TMR_MODE,
2570 		       MC_CMD_INIT_EVQ_IN_TMR_MODE_DIS);
2571 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_TMR_LOAD, 0);
2572 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_TMR_RELOAD, 0);
2573 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_COUNT_MODE,
2574 		       MC_CMD_INIT_EVQ_IN_COUNT_MODE_DIS);
2575 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_COUNT_THRSHLD, 0);
2576 
2577 	dma_addr = channel->eventq.buf.dma_addr;
2578 	for (i = 0; i < entries; ++i) {
2579 		MCDI_SET_ARRAY_QWORD(inbuf, INIT_EVQ_IN_DMA_ADDR, i, dma_addr);
2580 		dma_addr += EFX_BUF_SIZE;
2581 	}
2582 
2583 	inlen = MC_CMD_INIT_EVQ_IN_LEN(entries);
2584 
2585 	rc = efx_mcdi_rpc(efx, MC_CMD_INIT_EVQ, inbuf, inlen,
2586 			  outbuf, sizeof(outbuf), &outlen);
2587 	/* IRQ return is ignored */
2588 	if (channel->channel || rc)
2589 		return rc;
2590 
2591 	/* Successfully created event queue on channel 0 */
2592 	rc = efx_mcdi_get_workarounds(efx, &implemented, &enabled);
2593 	if (rc == -ENOSYS) {
2594 		/* GET_WORKAROUNDS was implemented before the bug26807
2595 		 * workaround, thus the latter must be unavailable in this fw
2596 		 */
2597 		nic_data->workaround_26807 = false;
2598 		rc = 0;
2599 	} else if (rc) {
2600 		goto fail;
2601 	} else {
2602 		nic_data->workaround_26807 =
2603 			!!(enabled & MC_CMD_GET_WORKAROUNDS_OUT_BUG26807);
2604 
2605 		if (implemented & MC_CMD_GET_WORKAROUNDS_OUT_BUG26807 &&
2606 		    !nic_data->workaround_26807) {
2607 			unsigned int flags;
2608 
2609 			rc = efx_mcdi_set_workaround(efx,
2610 						     MC_CMD_WORKAROUND_BUG26807,
2611 						     true, &flags);
2612 
2613 			if (!rc) {
2614 				if (flags &
2615 				    1 << MC_CMD_WORKAROUND_EXT_OUT_FLR_DONE_LBN) {
2616 					netif_info(efx, drv, efx->net_dev,
2617 						   "other functions on NIC have been reset\n");
2618 
2619 					/* With MCFW v4.6.x and earlier, the
2620 					 * boot count will have incremented,
2621 					 * so re-read the warm_boot_count
2622 					 * value now to ensure this function
2623 					 * doesn't think it has changed next
2624 					 * time it checks.
2625 					 */
2626 					rc = efx_ef10_get_warm_boot_count(efx);
2627 					if (rc >= 0) {
2628 						nic_data->warm_boot_count = rc;
2629 						rc = 0;
2630 					}
2631 				}
2632 				nic_data->workaround_26807 = true;
2633 			} else if (rc == -EPERM) {
2634 				rc = 0;
2635 			}
2636 		}
2637 	}
2638 
2639 	if (!rc)
2640 		return 0;
2641 
2642 fail:
2643 	efx_ef10_ev_fini(channel);
2644 	return rc;
2645 }
2646 
2647 static void efx_ef10_ev_remove(struct efx_channel *channel)
2648 {
2649 	efx_nic_free_buffer(channel->efx, &channel->eventq.buf);
2650 }
2651 
2652 static void efx_ef10_handle_rx_wrong_queue(struct efx_rx_queue *rx_queue,
2653 					   unsigned int rx_queue_label)
2654 {
2655 	struct efx_nic *efx = rx_queue->efx;
2656 
2657 	netif_info(efx, hw, efx->net_dev,
2658 		   "rx event arrived on queue %d labeled as queue %u\n",
2659 		   efx_rx_queue_index(rx_queue), rx_queue_label);
2660 
2661 	efx_schedule_reset(efx, RESET_TYPE_DISABLE);
2662 }
2663 
2664 static void
2665 efx_ef10_handle_rx_bad_lbits(struct efx_rx_queue *rx_queue,
2666 			     unsigned int actual, unsigned int expected)
2667 {
2668 	unsigned int dropped = (actual - expected) & rx_queue->ptr_mask;
2669 	struct efx_nic *efx = rx_queue->efx;
2670 
2671 	netif_info(efx, hw, efx->net_dev,
2672 		   "dropped %d events (index=%d expected=%d)\n",
2673 		   dropped, actual, expected);
2674 
2675 	efx_schedule_reset(efx, RESET_TYPE_DISABLE);
2676 }
2677 
2678 /* partially received RX was aborted. clean up. */
2679 static void efx_ef10_handle_rx_abort(struct efx_rx_queue *rx_queue)
2680 {
2681 	unsigned int rx_desc_ptr;
2682 
2683 	netif_dbg(rx_queue->efx, hw, rx_queue->efx->net_dev,
2684 		  "scattered RX aborted (dropping %u buffers)\n",
2685 		  rx_queue->scatter_n);
2686 
2687 	rx_desc_ptr = rx_queue->removed_count & rx_queue->ptr_mask;
2688 
2689 	efx_rx_packet(rx_queue, rx_desc_ptr, rx_queue->scatter_n,
2690 		      0, EFX_RX_PKT_DISCARD);
2691 
2692 	rx_queue->removed_count += rx_queue->scatter_n;
2693 	rx_queue->scatter_n = 0;
2694 	rx_queue->scatter_len = 0;
2695 	++efx_rx_queue_channel(rx_queue)->n_rx_nodesc_trunc;
2696 }
2697 
2698 static int efx_ef10_handle_rx_event(struct efx_channel *channel,
2699 				    const efx_qword_t *event)
2700 {
2701 	unsigned int rx_bytes, next_ptr_lbits, rx_queue_label, rx_l4_class;
2702 	unsigned int n_descs, n_packets, i;
2703 	struct efx_nic *efx = channel->efx;
2704 	struct efx_rx_queue *rx_queue;
2705 	bool rx_cont;
2706 	u16 flags = 0;
2707 
2708 	if (unlikely(ACCESS_ONCE(efx->reset_pending)))
2709 		return 0;
2710 
2711 	/* Basic packet information */
2712 	rx_bytes = EFX_QWORD_FIELD(*event, ESF_DZ_RX_BYTES);
2713 	next_ptr_lbits = EFX_QWORD_FIELD(*event, ESF_DZ_RX_DSC_PTR_LBITS);
2714 	rx_queue_label = EFX_QWORD_FIELD(*event, ESF_DZ_RX_QLABEL);
2715 	rx_l4_class = EFX_QWORD_FIELD(*event, ESF_DZ_RX_L4_CLASS);
2716 	rx_cont = EFX_QWORD_FIELD(*event, ESF_DZ_RX_CONT);
2717 
2718 	if (EFX_QWORD_FIELD(*event, ESF_DZ_RX_DROP_EVENT))
2719 		netdev_WARN(efx->net_dev, "saw RX_DROP_EVENT: event="
2720 			    EFX_QWORD_FMT "\n",
2721 			    EFX_QWORD_VAL(*event));
2722 
2723 	rx_queue = efx_channel_get_rx_queue(channel);
2724 
2725 	if (unlikely(rx_queue_label != efx_rx_queue_index(rx_queue)))
2726 		efx_ef10_handle_rx_wrong_queue(rx_queue, rx_queue_label);
2727 
2728 	n_descs = ((next_ptr_lbits - rx_queue->removed_count) &
2729 		   ((1 << ESF_DZ_RX_DSC_PTR_LBITS_WIDTH) - 1));
2730 
2731 	if (n_descs != rx_queue->scatter_n + 1) {
2732 		struct efx_ef10_nic_data *nic_data = efx->nic_data;
2733 
2734 		/* detect rx abort */
2735 		if (unlikely(n_descs == rx_queue->scatter_n)) {
2736 			if (rx_queue->scatter_n == 0 || rx_bytes != 0)
2737 				netdev_WARN(efx->net_dev,
2738 					    "invalid RX abort: scatter_n=%u event="
2739 					    EFX_QWORD_FMT "\n",
2740 					    rx_queue->scatter_n,
2741 					    EFX_QWORD_VAL(*event));
2742 			efx_ef10_handle_rx_abort(rx_queue);
2743 			return 0;
2744 		}
2745 
2746 		/* Check that RX completion merging is valid, i.e.
2747 		 * the current firmware supports it and this is a
2748 		 * non-scattered packet.
2749 		 */
2750 		if (!(nic_data->datapath_caps &
2751 		      (1 << MC_CMD_GET_CAPABILITIES_OUT_RX_BATCHING_LBN)) ||
2752 		    rx_queue->scatter_n != 0 || rx_cont) {
2753 			efx_ef10_handle_rx_bad_lbits(
2754 				rx_queue, next_ptr_lbits,
2755 				(rx_queue->removed_count +
2756 				 rx_queue->scatter_n + 1) &
2757 				((1 << ESF_DZ_RX_DSC_PTR_LBITS_WIDTH) - 1));
2758 			return 0;
2759 		}
2760 
2761 		/* Merged completion for multiple non-scattered packets */
2762 		rx_queue->scatter_n = 1;
2763 		rx_queue->scatter_len = 0;
2764 		n_packets = n_descs;
2765 		++channel->n_rx_merge_events;
2766 		channel->n_rx_merge_packets += n_packets;
2767 		flags |= EFX_RX_PKT_PREFIX_LEN;
2768 	} else {
2769 		++rx_queue->scatter_n;
2770 		rx_queue->scatter_len += rx_bytes;
2771 		if (rx_cont)
2772 			return 0;
2773 		n_packets = 1;
2774 	}
2775 
2776 	if (unlikely(EFX_QWORD_FIELD(*event, ESF_DZ_RX_ECRC_ERR)))
2777 		flags |= EFX_RX_PKT_DISCARD;
2778 
2779 	if (unlikely(EFX_QWORD_FIELD(*event, ESF_DZ_RX_IPCKSUM_ERR))) {
2780 		channel->n_rx_ip_hdr_chksum_err += n_packets;
2781 	} else if (unlikely(EFX_QWORD_FIELD(*event,
2782 					    ESF_DZ_RX_TCPUDP_CKSUM_ERR))) {
2783 		channel->n_rx_tcp_udp_chksum_err += n_packets;
2784 	} else if (rx_l4_class == ESE_DZ_L4_CLASS_TCP ||
2785 		   rx_l4_class == ESE_DZ_L4_CLASS_UDP) {
2786 		flags |= EFX_RX_PKT_CSUMMED;
2787 	}
2788 
2789 	if (rx_l4_class == ESE_DZ_L4_CLASS_TCP)
2790 		flags |= EFX_RX_PKT_TCP;
2791 
2792 	channel->irq_mod_score += 2 * n_packets;
2793 
2794 	/* Handle received packet(s) */
2795 	for (i = 0; i < n_packets; i++) {
2796 		efx_rx_packet(rx_queue,
2797 			      rx_queue->removed_count & rx_queue->ptr_mask,
2798 			      rx_queue->scatter_n, rx_queue->scatter_len,
2799 			      flags);
2800 		rx_queue->removed_count += rx_queue->scatter_n;
2801 	}
2802 
2803 	rx_queue->scatter_n = 0;
2804 	rx_queue->scatter_len = 0;
2805 
2806 	return n_packets;
2807 }
2808 
2809 static int
2810 efx_ef10_handle_tx_event(struct efx_channel *channel, efx_qword_t *event)
2811 {
2812 	struct efx_nic *efx = channel->efx;
2813 	struct efx_tx_queue *tx_queue;
2814 	unsigned int tx_ev_desc_ptr;
2815 	unsigned int tx_ev_q_label;
2816 	int tx_descs = 0;
2817 
2818 	if (unlikely(ACCESS_ONCE(efx->reset_pending)))
2819 		return 0;
2820 
2821 	if (unlikely(EFX_QWORD_FIELD(*event, ESF_DZ_TX_DROP_EVENT)))
2822 		return 0;
2823 
2824 	/* Transmit completion */
2825 	tx_ev_desc_ptr = EFX_QWORD_FIELD(*event, ESF_DZ_TX_DESCR_INDX);
2826 	tx_ev_q_label = EFX_QWORD_FIELD(*event, ESF_DZ_TX_QLABEL);
2827 	tx_queue = efx_channel_get_tx_queue(channel,
2828 					    tx_ev_q_label % EFX_TXQ_TYPES);
2829 	tx_descs = ((tx_ev_desc_ptr + 1 - tx_queue->read_count) &
2830 		    tx_queue->ptr_mask);
2831 	efx_xmit_done(tx_queue, tx_ev_desc_ptr & tx_queue->ptr_mask);
2832 
2833 	return tx_descs;
2834 }
2835 
2836 static void
2837 efx_ef10_handle_driver_event(struct efx_channel *channel, efx_qword_t *event)
2838 {
2839 	struct efx_nic *efx = channel->efx;
2840 	int subcode;
2841 
2842 	subcode = EFX_QWORD_FIELD(*event, ESF_DZ_DRV_SUB_CODE);
2843 
2844 	switch (subcode) {
2845 	case ESE_DZ_DRV_TIMER_EV:
2846 	case ESE_DZ_DRV_WAKE_UP_EV:
2847 		break;
2848 	case ESE_DZ_DRV_START_UP_EV:
2849 		/* event queue init complete. ok. */
2850 		break;
2851 	default:
2852 		netif_err(efx, hw, efx->net_dev,
2853 			  "channel %d unknown driver event type %d"
2854 			  " (data " EFX_QWORD_FMT ")\n",
2855 			  channel->channel, subcode,
2856 			  EFX_QWORD_VAL(*event));
2857 
2858 	}
2859 }
2860 
2861 static void efx_ef10_handle_driver_generated_event(struct efx_channel *channel,
2862 						   efx_qword_t *event)
2863 {
2864 	struct efx_nic *efx = channel->efx;
2865 	u32 subcode;
2866 
2867 	subcode = EFX_QWORD_FIELD(*event, EFX_DWORD_0);
2868 
2869 	switch (subcode) {
2870 	case EFX_EF10_TEST:
2871 		channel->event_test_cpu = raw_smp_processor_id();
2872 		break;
2873 	case EFX_EF10_REFILL:
2874 		/* The queue must be empty, so we won't receive any rx
2875 		 * events, so efx_process_channel() won't refill the
2876 		 * queue. Refill it here
2877 		 */
2878 		efx_fast_push_rx_descriptors(&channel->rx_queue, true);
2879 		break;
2880 	default:
2881 		netif_err(efx, hw, efx->net_dev,
2882 			  "channel %d unknown driver event type %u"
2883 			  " (data " EFX_QWORD_FMT ")\n",
2884 			  channel->channel, (unsigned) subcode,
2885 			  EFX_QWORD_VAL(*event));
2886 	}
2887 }
2888 
2889 static int efx_ef10_ev_process(struct efx_channel *channel, int quota)
2890 {
2891 	struct efx_nic *efx = channel->efx;
2892 	efx_qword_t event, *p_event;
2893 	unsigned int read_ptr;
2894 	int ev_code;
2895 	int tx_descs = 0;
2896 	int spent = 0;
2897 
2898 	if (quota <= 0)
2899 		return spent;
2900 
2901 	read_ptr = channel->eventq_read_ptr;
2902 
2903 	for (;;) {
2904 		p_event = efx_event(channel, read_ptr);
2905 		event = *p_event;
2906 
2907 		if (!efx_event_present(&event))
2908 			break;
2909 
2910 		EFX_SET_QWORD(*p_event);
2911 
2912 		++read_ptr;
2913 
2914 		ev_code = EFX_QWORD_FIELD(event, ESF_DZ_EV_CODE);
2915 
2916 		netif_vdbg(efx, drv, efx->net_dev,
2917 			   "processing event on %d " EFX_QWORD_FMT "\n",
2918 			   channel->channel, EFX_QWORD_VAL(event));
2919 
2920 		switch (ev_code) {
2921 		case ESE_DZ_EV_CODE_MCDI_EV:
2922 			efx_mcdi_process_event(channel, &event);
2923 			break;
2924 		case ESE_DZ_EV_CODE_RX_EV:
2925 			spent += efx_ef10_handle_rx_event(channel, &event);
2926 			if (spent >= quota) {
2927 				/* XXX can we split a merged event to
2928 				 * avoid going over-quota?
2929 				 */
2930 				spent = quota;
2931 				goto out;
2932 			}
2933 			break;
2934 		case ESE_DZ_EV_CODE_TX_EV:
2935 			tx_descs += efx_ef10_handle_tx_event(channel, &event);
2936 			if (tx_descs > efx->txq_entries) {
2937 				spent = quota;
2938 				goto out;
2939 			} else if (++spent == quota) {
2940 				goto out;
2941 			}
2942 			break;
2943 		case ESE_DZ_EV_CODE_DRIVER_EV:
2944 			efx_ef10_handle_driver_event(channel, &event);
2945 			if (++spent == quota)
2946 				goto out;
2947 			break;
2948 		case EFX_EF10_DRVGEN_EV:
2949 			efx_ef10_handle_driver_generated_event(channel, &event);
2950 			break;
2951 		default:
2952 			netif_err(efx, hw, efx->net_dev,
2953 				  "channel %d unknown event type %d"
2954 				  " (data " EFX_QWORD_FMT ")\n",
2955 				  channel->channel, ev_code,
2956 				  EFX_QWORD_VAL(event));
2957 		}
2958 	}
2959 
2960 out:
2961 	channel->eventq_read_ptr = read_ptr;
2962 	return spent;
2963 }
2964 
2965 static void efx_ef10_ev_read_ack(struct efx_channel *channel)
2966 {
2967 	struct efx_nic *efx = channel->efx;
2968 	efx_dword_t rptr;
2969 
2970 	if (EFX_EF10_WORKAROUND_35388(efx)) {
2971 		BUILD_BUG_ON(EFX_MIN_EVQ_SIZE <
2972 			     (1 << ERF_DD_EVQ_IND_RPTR_WIDTH));
2973 		BUILD_BUG_ON(EFX_MAX_EVQ_SIZE >
2974 			     (1 << 2 * ERF_DD_EVQ_IND_RPTR_WIDTH));
2975 
2976 		EFX_POPULATE_DWORD_2(rptr, ERF_DD_EVQ_IND_RPTR_FLAGS,
2977 				     EFE_DD_EVQ_IND_RPTR_FLAGS_HIGH,
2978 				     ERF_DD_EVQ_IND_RPTR,
2979 				     (channel->eventq_read_ptr &
2980 				      channel->eventq_mask) >>
2981 				     ERF_DD_EVQ_IND_RPTR_WIDTH);
2982 		efx_writed_page(efx, &rptr, ER_DD_EVQ_INDIRECT,
2983 				channel->channel);
2984 		EFX_POPULATE_DWORD_2(rptr, ERF_DD_EVQ_IND_RPTR_FLAGS,
2985 				     EFE_DD_EVQ_IND_RPTR_FLAGS_LOW,
2986 				     ERF_DD_EVQ_IND_RPTR,
2987 				     channel->eventq_read_ptr &
2988 				     ((1 << ERF_DD_EVQ_IND_RPTR_WIDTH) - 1));
2989 		efx_writed_page(efx, &rptr, ER_DD_EVQ_INDIRECT,
2990 				channel->channel);
2991 	} else {
2992 		EFX_POPULATE_DWORD_1(rptr, ERF_DZ_EVQ_RPTR,
2993 				     channel->eventq_read_ptr &
2994 				     channel->eventq_mask);
2995 		efx_writed_page(efx, &rptr, ER_DZ_EVQ_RPTR, channel->channel);
2996 	}
2997 }
2998 
2999 static void efx_ef10_ev_test_generate(struct efx_channel *channel)
3000 {
3001 	MCDI_DECLARE_BUF(inbuf, MC_CMD_DRIVER_EVENT_IN_LEN);
3002 	struct efx_nic *efx = channel->efx;
3003 	efx_qword_t event;
3004 	int rc;
3005 
3006 	EFX_POPULATE_QWORD_2(event,
3007 			     ESF_DZ_EV_CODE, EFX_EF10_DRVGEN_EV,
3008 			     ESF_DZ_EV_DATA, EFX_EF10_TEST);
3009 
3010 	MCDI_SET_DWORD(inbuf, DRIVER_EVENT_IN_EVQ, channel->channel);
3011 
3012 	/* MCDI_SET_QWORD is not appropriate here since EFX_POPULATE_* has
3013 	 * already swapped the data to little-endian order.
3014 	 */
3015 	memcpy(MCDI_PTR(inbuf, DRIVER_EVENT_IN_DATA), &event.u64[0],
3016 	       sizeof(efx_qword_t));
3017 
3018 	rc = efx_mcdi_rpc(efx, MC_CMD_DRIVER_EVENT, inbuf, sizeof(inbuf),
3019 			  NULL, 0, NULL);
3020 	if (rc != 0)
3021 		goto fail;
3022 
3023 	return;
3024 
3025 fail:
3026 	WARN_ON(true);
3027 	netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
3028 }
3029 
3030 void efx_ef10_handle_drain_event(struct efx_nic *efx)
3031 {
3032 	if (atomic_dec_and_test(&efx->active_queues))
3033 		wake_up(&efx->flush_wq);
3034 
3035 	WARN_ON(atomic_read(&efx->active_queues) < 0);
3036 }
3037 
3038 static int efx_ef10_fini_dmaq(struct efx_nic *efx)
3039 {
3040 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
3041 	struct efx_channel *channel;
3042 	struct efx_tx_queue *tx_queue;
3043 	struct efx_rx_queue *rx_queue;
3044 	int pending;
3045 
3046 	/* If the MC has just rebooted, the TX/RX queues will have already been
3047 	 * torn down, but efx->active_queues needs to be set to zero.
3048 	 */
3049 	if (nic_data->must_realloc_vis) {
3050 		atomic_set(&efx->active_queues, 0);
3051 		return 0;
3052 	}
3053 
3054 	/* Do not attempt to write to the NIC during EEH recovery */
3055 	if (efx->state != STATE_RECOVERY) {
3056 		efx_for_each_channel(channel, efx) {
3057 			efx_for_each_channel_rx_queue(rx_queue, channel)
3058 				efx_ef10_rx_fini(rx_queue);
3059 			efx_for_each_channel_tx_queue(tx_queue, channel)
3060 				efx_ef10_tx_fini(tx_queue);
3061 		}
3062 
3063 		wait_event_timeout(efx->flush_wq,
3064 				   atomic_read(&efx->active_queues) == 0,
3065 				   msecs_to_jiffies(EFX_MAX_FLUSH_TIME));
3066 		pending = atomic_read(&efx->active_queues);
3067 		if (pending) {
3068 			netif_err(efx, hw, efx->net_dev, "failed to flush %d queues\n",
3069 				  pending);
3070 			return -ETIMEDOUT;
3071 		}
3072 	}
3073 
3074 	return 0;
3075 }
3076 
3077 static void efx_ef10_prepare_flr(struct efx_nic *efx)
3078 {
3079 	atomic_set(&efx->active_queues, 0);
3080 }
3081 
3082 static bool efx_ef10_filter_equal(const struct efx_filter_spec *left,
3083 				  const struct efx_filter_spec *right)
3084 {
3085 	if ((left->match_flags ^ right->match_flags) |
3086 	    ((left->flags ^ right->flags) &
3087 	     (EFX_FILTER_FLAG_RX | EFX_FILTER_FLAG_TX)))
3088 		return false;
3089 
3090 	return memcmp(&left->outer_vid, &right->outer_vid,
3091 		      sizeof(struct efx_filter_spec) -
3092 		      offsetof(struct efx_filter_spec, outer_vid)) == 0;
3093 }
3094 
3095 static unsigned int efx_ef10_filter_hash(const struct efx_filter_spec *spec)
3096 {
3097 	BUILD_BUG_ON(offsetof(struct efx_filter_spec, outer_vid) & 3);
3098 	return jhash2((const u32 *)&spec->outer_vid,
3099 		      (sizeof(struct efx_filter_spec) -
3100 		       offsetof(struct efx_filter_spec, outer_vid)) / 4,
3101 		      0);
3102 	/* XXX should we randomise the initval? */
3103 }
3104 
3105 /* Decide whether a filter should be exclusive or else should allow
3106  * delivery to additional recipients.  Currently we decide that
3107  * filters for specific local unicast MAC and IP addresses are
3108  * exclusive.
3109  */
3110 static bool efx_ef10_filter_is_exclusive(const struct efx_filter_spec *spec)
3111 {
3112 	if (spec->match_flags & EFX_FILTER_MATCH_LOC_MAC &&
3113 	    !is_multicast_ether_addr(spec->loc_mac))
3114 		return true;
3115 
3116 	if ((spec->match_flags &
3117 	     (EFX_FILTER_MATCH_ETHER_TYPE | EFX_FILTER_MATCH_LOC_HOST)) ==
3118 	    (EFX_FILTER_MATCH_ETHER_TYPE | EFX_FILTER_MATCH_LOC_HOST)) {
3119 		if (spec->ether_type == htons(ETH_P_IP) &&
3120 		    !ipv4_is_multicast(spec->loc_host[0]))
3121 			return true;
3122 		if (spec->ether_type == htons(ETH_P_IPV6) &&
3123 		    ((const u8 *)spec->loc_host)[0] != 0xff)
3124 			return true;
3125 	}
3126 
3127 	return false;
3128 }
3129 
3130 static struct efx_filter_spec *
3131 efx_ef10_filter_entry_spec(const struct efx_ef10_filter_table *table,
3132 			   unsigned int filter_idx)
3133 {
3134 	return (struct efx_filter_spec *)(table->entry[filter_idx].spec &
3135 					  ~EFX_EF10_FILTER_FLAGS);
3136 }
3137 
3138 static unsigned int
3139 efx_ef10_filter_entry_flags(const struct efx_ef10_filter_table *table,
3140 			   unsigned int filter_idx)
3141 {
3142 	return table->entry[filter_idx].spec & EFX_EF10_FILTER_FLAGS;
3143 }
3144 
3145 static void
3146 efx_ef10_filter_set_entry(struct efx_ef10_filter_table *table,
3147 			  unsigned int filter_idx,
3148 			  const struct efx_filter_spec *spec,
3149 			  unsigned int flags)
3150 {
3151 	table->entry[filter_idx].spec =	(unsigned long)spec | flags;
3152 }
3153 
3154 static void efx_ef10_filter_push_prep(struct efx_nic *efx,
3155 				      const struct efx_filter_spec *spec,
3156 				      efx_dword_t *inbuf, u64 handle,
3157 				      bool replacing)
3158 {
3159 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
3160 	u32 flags = spec->flags;
3161 
3162 	memset(inbuf, 0, MC_CMD_FILTER_OP_IN_LEN);
3163 
3164 	/* Remove RSS flag if we don't have an RSS context. */
3165 	if (flags & EFX_FILTER_FLAG_RX_RSS &&
3166 	    spec->rss_context == EFX_FILTER_RSS_CONTEXT_DEFAULT &&
3167 	    nic_data->rx_rss_context == EFX_EF10_RSS_CONTEXT_INVALID)
3168 		flags &= ~EFX_FILTER_FLAG_RX_RSS;
3169 
3170 	if (replacing) {
3171 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
3172 			       MC_CMD_FILTER_OP_IN_OP_REPLACE);
3173 		MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE, handle);
3174 	} else {
3175 		u32 match_fields = 0;
3176 
3177 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
3178 			       efx_ef10_filter_is_exclusive(spec) ?
3179 			       MC_CMD_FILTER_OP_IN_OP_INSERT :
3180 			       MC_CMD_FILTER_OP_IN_OP_SUBSCRIBE);
3181 
3182 		/* Convert match flags and values.  Unlike almost
3183 		 * everything else in MCDI, these fields are in
3184 		 * network byte order.
3185 		 */
3186 		if (spec->match_flags & EFX_FILTER_MATCH_LOC_MAC_IG)
3187 			match_fields |=
3188 				is_multicast_ether_addr(spec->loc_mac) ?
3189 				1 << MC_CMD_FILTER_OP_IN_MATCH_UNKNOWN_MCAST_DST_LBN :
3190 				1 << MC_CMD_FILTER_OP_IN_MATCH_UNKNOWN_UCAST_DST_LBN;
3191 #define COPY_FIELD(gen_flag, gen_field, mcdi_field)			     \
3192 		if (spec->match_flags & EFX_FILTER_MATCH_ ## gen_flag) {     \
3193 			match_fields |=					     \
3194 				1 << MC_CMD_FILTER_OP_IN_MATCH_ ##	     \
3195 				mcdi_field ## _LBN;			     \
3196 			BUILD_BUG_ON(					     \
3197 				MC_CMD_FILTER_OP_IN_ ## mcdi_field ## _LEN < \
3198 				sizeof(spec->gen_field));		     \
3199 			memcpy(MCDI_PTR(inbuf, FILTER_OP_IN_ ##	mcdi_field), \
3200 			       &spec->gen_field, sizeof(spec->gen_field));   \
3201 		}
3202 		COPY_FIELD(REM_HOST, rem_host, SRC_IP);
3203 		COPY_FIELD(LOC_HOST, loc_host, DST_IP);
3204 		COPY_FIELD(REM_MAC, rem_mac, SRC_MAC);
3205 		COPY_FIELD(REM_PORT, rem_port, SRC_PORT);
3206 		COPY_FIELD(LOC_MAC, loc_mac, DST_MAC);
3207 		COPY_FIELD(LOC_PORT, loc_port, DST_PORT);
3208 		COPY_FIELD(ETHER_TYPE, ether_type, ETHER_TYPE);
3209 		COPY_FIELD(INNER_VID, inner_vid, INNER_VLAN);
3210 		COPY_FIELD(OUTER_VID, outer_vid, OUTER_VLAN);
3211 		COPY_FIELD(IP_PROTO, ip_proto, IP_PROTO);
3212 #undef COPY_FIELD
3213 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_MATCH_FIELDS,
3214 			       match_fields);
3215 	}
3216 
3217 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_PORT_ID, nic_data->vport_id);
3218 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_DEST,
3219 		       spec->dmaq_id == EFX_FILTER_RX_DMAQ_ID_DROP ?
3220 		       MC_CMD_FILTER_OP_IN_RX_DEST_DROP :
3221 		       MC_CMD_FILTER_OP_IN_RX_DEST_HOST);
3222 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_TX_DOMAIN, 0);
3223 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_TX_DEST,
3224 		       MC_CMD_FILTER_OP_IN_TX_DEST_DEFAULT);
3225 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_QUEUE,
3226 		       spec->dmaq_id == EFX_FILTER_RX_DMAQ_ID_DROP ?
3227 		       0 : spec->dmaq_id);
3228 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_MODE,
3229 		       (flags & EFX_FILTER_FLAG_RX_RSS) ?
3230 		       MC_CMD_FILTER_OP_IN_RX_MODE_RSS :
3231 		       MC_CMD_FILTER_OP_IN_RX_MODE_SIMPLE);
3232 	if (flags & EFX_FILTER_FLAG_RX_RSS)
3233 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_CONTEXT,
3234 			       spec->rss_context !=
3235 			       EFX_FILTER_RSS_CONTEXT_DEFAULT ?
3236 			       spec->rss_context : nic_data->rx_rss_context);
3237 }
3238 
3239 static int efx_ef10_filter_push(struct efx_nic *efx,
3240 				const struct efx_filter_spec *spec,
3241 				u64 *handle, bool replacing)
3242 {
3243 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_IN_LEN);
3244 	MCDI_DECLARE_BUF(outbuf, MC_CMD_FILTER_OP_OUT_LEN);
3245 	int rc;
3246 
3247 	efx_ef10_filter_push_prep(efx, spec, inbuf, *handle, replacing);
3248 	rc = efx_mcdi_rpc(efx, MC_CMD_FILTER_OP, inbuf, sizeof(inbuf),
3249 			  outbuf, sizeof(outbuf), NULL);
3250 	if (rc == 0)
3251 		*handle = MCDI_QWORD(outbuf, FILTER_OP_OUT_HANDLE);
3252 	if (rc == -ENOSPC)
3253 		rc = -EBUSY; /* to match efx_farch_filter_insert() */
3254 	return rc;
3255 }
3256 
3257 static u32 efx_ef10_filter_mcdi_flags_from_spec(const struct efx_filter_spec *spec)
3258 {
3259 	unsigned int match_flags = spec->match_flags;
3260 	u32 mcdi_flags = 0;
3261 
3262 	if (match_flags & EFX_FILTER_MATCH_LOC_MAC_IG) {
3263 		match_flags &= ~EFX_FILTER_MATCH_LOC_MAC_IG;
3264 		mcdi_flags |=
3265 			is_multicast_ether_addr(spec->loc_mac) ?
3266 			(1 << MC_CMD_FILTER_OP_IN_MATCH_UNKNOWN_MCAST_DST_LBN) :
3267 			(1 << MC_CMD_FILTER_OP_IN_MATCH_UNKNOWN_UCAST_DST_LBN);
3268 	}
3269 
3270 #define MAP_FILTER_TO_MCDI_FLAG(gen_flag, mcdi_field) {			\
3271 		unsigned int old_match_flags = match_flags;		\
3272 		match_flags &= ~EFX_FILTER_MATCH_ ## gen_flag;		\
3273 		if (match_flags != old_match_flags)			\
3274 			mcdi_flags |=					\
3275 				(1 << MC_CMD_FILTER_OP_IN_MATCH_ ##	\
3276 				 mcdi_field ## _LBN);			\
3277 	}
3278 	MAP_FILTER_TO_MCDI_FLAG(REM_HOST, SRC_IP);
3279 	MAP_FILTER_TO_MCDI_FLAG(LOC_HOST, DST_IP);
3280 	MAP_FILTER_TO_MCDI_FLAG(REM_MAC, SRC_MAC);
3281 	MAP_FILTER_TO_MCDI_FLAG(REM_PORT, SRC_PORT);
3282 	MAP_FILTER_TO_MCDI_FLAG(LOC_MAC, DST_MAC);
3283 	MAP_FILTER_TO_MCDI_FLAG(LOC_PORT, DST_PORT);
3284 	MAP_FILTER_TO_MCDI_FLAG(ETHER_TYPE, ETHER_TYPE);
3285 	MAP_FILTER_TO_MCDI_FLAG(INNER_VID, INNER_VLAN);
3286 	MAP_FILTER_TO_MCDI_FLAG(OUTER_VID, OUTER_VLAN);
3287 	MAP_FILTER_TO_MCDI_FLAG(IP_PROTO, IP_PROTO);
3288 #undef MAP_FILTER_TO_MCDI_FLAG
3289 
3290 	/* Did we map them all? */
3291 	WARN_ON_ONCE(match_flags);
3292 
3293 	return mcdi_flags;
3294 }
3295 
3296 static int efx_ef10_filter_pri(struct efx_ef10_filter_table *table,
3297 			       const struct efx_filter_spec *spec)
3298 {
3299 	u32 mcdi_flags = efx_ef10_filter_mcdi_flags_from_spec(spec);
3300 	unsigned int match_pri;
3301 
3302 	for (match_pri = 0;
3303 	     match_pri < table->rx_match_count;
3304 	     match_pri++)
3305 		if (table->rx_match_mcdi_flags[match_pri] == mcdi_flags)
3306 			return match_pri;
3307 
3308 	return -EPROTONOSUPPORT;
3309 }
3310 
3311 static s32 efx_ef10_filter_insert(struct efx_nic *efx,
3312 				  struct efx_filter_spec *spec,
3313 				  bool replace_equal)
3314 {
3315 	struct efx_ef10_filter_table *table = efx->filter_state;
3316 	DECLARE_BITMAP(mc_rem_map, EFX_EF10_FILTER_SEARCH_LIMIT);
3317 	struct efx_filter_spec *saved_spec;
3318 	unsigned int match_pri, hash;
3319 	unsigned int priv_flags;
3320 	bool replacing = false;
3321 	int ins_index = -1;
3322 	DEFINE_WAIT(wait);
3323 	bool is_mc_recip;
3324 	s32 rc;
3325 
3326 	/* For now, only support RX filters */
3327 	if ((spec->flags & (EFX_FILTER_FLAG_RX | EFX_FILTER_FLAG_TX)) !=
3328 	    EFX_FILTER_FLAG_RX)
3329 		return -EINVAL;
3330 
3331 	rc = efx_ef10_filter_pri(table, spec);
3332 	if (rc < 0)
3333 		return rc;
3334 	match_pri = rc;
3335 
3336 	hash = efx_ef10_filter_hash(spec);
3337 	is_mc_recip = efx_filter_is_mc_recipient(spec);
3338 	if (is_mc_recip)
3339 		bitmap_zero(mc_rem_map, EFX_EF10_FILTER_SEARCH_LIMIT);
3340 
3341 	/* Find any existing filters with the same match tuple or
3342 	 * else a free slot to insert at.  If any of them are busy,
3343 	 * we have to wait and retry.
3344 	 */
3345 	for (;;) {
3346 		unsigned int depth = 1;
3347 		unsigned int i;
3348 
3349 		spin_lock_bh(&efx->filter_lock);
3350 
3351 		for (;;) {
3352 			i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
3353 			saved_spec = efx_ef10_filter_entry_spec(table, i);
3354 
3355 			if (!saved_spec) {
3356 				if (ins_index < 0)
3357 					ins_index = i;
3358 			} else if (efx_ef10_filter_equal(spec, saved_spec)) {
3359 				if (table->entry[i].spec &
3360 				    EFX_EF10_FILTER_FLAG_BUSY)
3361 					break;
3362 				if (spec->priority < saved_spec->priority &&
3363 				    spec->priority != EFX_FILTER_PRI_AUTO) {
3364 					rc = -EPERM;
3365 					goto out_unlock;
3366 				}
3367 				if (!is_mc_recip) {
3368 					/* This is the only one */
3369 					if (spec->priority ==
3370 					    saved_spec->priority &&
3371 					    !replace_equal) {
3372 						rc = -EEXIST;
3373 						goto out_unlock;
3374 					}
3375 					ins_index = i;
3376 					goto found;
3377 				} else if (spec->priority >
3378 					   saved_spec->priority ||
3379 					   (spec->priority ==
3380 					    saved_spec->priority &&
3381 					    replace_equal)) {
3382 					if (ins_index < 0)
3383 						ins_index = i;
3384 					else
3385 						__set_bit(depth, mc_rem_map);
3386 				}
3387 			}
3388 
3389 			/* Once we reach the maximum search depth, use
3390 			 * the first suitable slot or return -EBUSY if
3391 			 * there was none
3392 			 */
3393 			if (depth == EFX_EF10_FILTER_SEARCH_LIMIT) {
3394 				if (ins_index < 0) {
3395 					rc = -EBUSY;
3396 					goto out_unlock;
3397 				}
3398 				goto found;
3399 			}
3400 
3401 			++depth;
3402 		}
3403 
3404 		prepare_to_wait(&table->waitq, &wait, TASK_UNINTERRUPTIBLE);
3405 		spin_unlock_bh(&efx->filter_lock);
3406 		schedule();
3407 	}
3408 
3409 found:
3410 	/* Create a software table entry if necessary, and mark it
3411 	 * busy.  We might yet fail to insert, but any attempt to
3412 	 * insert a conflicting filter while we're waiting for the
3413 	 * firmware must find the busy entry.
3414 	 */
3415 	saved_spec = efx_ef10_filter_entry_spec(table, ins_index);
3416 	if (saved_spec) {
3417 		if (spec->priority == EFX_FILTER_PRI_AUTO &&
3418 		    saved_spec->priority >= EFX_FILTER_PRI_AUTO) {
3419 			/* Just make sure it won't be removed */
3420 			if (saved_spec->priority > EFX_FILTER_PRI_AUTO)
3421 				saved_spec->flags |= EFX_FILTER_FLAG_RX_OVER_AUTO;
3422 			table->entry[ins_index].spec &=
3423 				~EFX_EF10_FILTER_FLAG_AUTO_OLD;
3424 			rc = ins_index;
3425 			goto out_unlock;
3426 		}
3427 		replacing = true;
3428 		priv_flags = efx_ef10_filter_entry_flags(table, ins_index);
3429 	} else {
3430 		saved_spec = kmalloc(sizeof(*spec), GFP_ATOMIC);
3431 		if (!saved_spec) {
3432 			rc = -ENOMEM;
3433 			goto out_unlock;
3434 		}
3435 		*saved_spec = *spec;
3436 		priv_flags = 0;
3437 	}
3438 	efx_ef10_filter_set_entry(table, ins_index, saved_spec,
3439 				  priv_flags | EFX_EF10_FILTER_FLAG_BUSY);
3440 
3441 	/* Mark lower-priority multicast recipients busy prior to removal */
3442 	if (is_mc_recip) {
3443 		unsigned int depth, i;
3444 
3445 		for (depth = 0; depth < EFX_EF10_FILTER_SEARCH_LIMIT; depth++) {
3446 			i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
3447 			if (test_bit(depth, mc_rem_map))
3448 				table->entry[i].spec |=
3449 					EFX_EF10_FILTER_FLAG_BUSY;
3450 		}
3451 	}
3452 
3453 	spin_unlock_bh(&efx->filter_lock);
3454 
3455 	rc = efx_ef10_filter_push(efx, spec, &table->entry[ins_index].handle,
3456 				  replacing);
3457 
3458 	/* Finalise the software table entry */
3459 	spin_lock_bh(&efx->filter_lock);
3460 	if (rc == 0) {
3461 		if (replacing) {
3462 			/* Update the fields that may differ */
3463 			if (saved_spec->priority == EFX_FILTER_PRI_AUTO)
3464 				saved_spec->flags |=
3465 					EFX_FILTER_FLAG_RX_OVER_AUTO;
3466 			saved_spec->priority = spec->priority;
3467 			saved_spec->flags &= EFX_FILTER_FLAG_RX_OVER_AUTO;
3468 			saved_spec->flags |= spec->flags;
3469 			saved_spec->rss_context = spec->rss_context;
3470 			saved_spec->dmaq_id = spec->dmaq_id;
3471 		}
3472 	} else if (!replacing) {
3473 		kfree(saved_spec);
3474 		saved_spec = NULL;
3475 	}
3476 	efx_ef10_filter_set_entry(table, ins_index, saved_spec, priv_flags);
3477 
3478 	/* Remove and finalise entries for lower-priority multicast
3479 	 * recipients
3480 	 */
3481 	if (is_mc_recip) {
3482 		MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_IN_LEN);
3483 		unsigned int depth, i;
3484 
3485 		memset(inbuf, 0, sizeof(inbuf));
3486 
3487 		for (depth = 0; depth < EFX_EF10_FILTER_SEARCH_LIMIT; depth++) {
3488 			if (!test_bit(depth, mc_rem_map))
3489 				continue;
3490 
3491 			i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
3492 			saved_spec = efx_ef10_filter_entry_spec(table, i);
3493 			priv_flags = efx_ef10_filter_entry_flags(table, i);
3494 
3495 			if (rc == 0) {
3496 				spin_unlock_bh(&efx->filter_lock);
3497 				MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
3498 					       MC_CMD_FILTER_OP_IN_OP_UNSUBSCRIBE);
3499 				MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
3500 					       table->entry[i].handle);
3501 				rc = efx_mcdi_rpc(efx, MC_CMD_FILTER_OP,
3502 						  inbuf, sizeof(inbuf),
3503 						  NULL, 0, NULL);
3504 				spin_lock_bh(&efx->filter_lock);
3505 			}
3506 
3507 			if (rc == 0) {
3508 				kfree(saved_spec);
3509 				saved_spec = NULL;
3510 				priv_flags = 0;
3511 			} else {
3512 				priv_flags &= ~EFX_EF10_FILTER_FLAG_BUSY;
3513 			}
3514 			efx_ef10_filter_set_entry(table, i, saved_spec,
3515 						  priv_flags);
3516 		}
3517 	}
3518 
3519 	/* If successful, return the inserted filter ID */
3520 	if (rc == 0)
3521 		rc = match_pri * HUNT_FILTER_TBL_ROWS + ins_index;
3522 
3523 	wake_up_all(&table->waitq);
3524 out_unlock:
3525 	spin_unlock_bh(&efx->filter_lock);
3526 	finish_wait(&table->waitq, &wait);
3527 	return rc;
3528 }
3529 
3530 static void efx_ef10_filter_update_rx_scatter(struct efx_nic *efx)
3531 {
3532 	/* no need to do anything here on EF10 */
3533 }
3534 
3535 /* Remove a filter.
3536  * If !by_index, remove by ID
3537  * If by_index, remove by index
3538  * Filter ID may come from userland and must be range-checked.
3539  */
3540 static int efx_ef10_filter_remove_internal(struct efx_nic *efx,
3541 					   unsigned int priority_mask,
3542 					   u32 filter_id, bool by_index)
3543 {
3544 	unsigned int filter_idx = filter_id % HUNT_FILTER_TBL_ROWS;
3545 	struct efx_ef10_filter_table *table = efx->filter_state;
3546 	MCDI_DECLARE_BUF(inbuf,
3547 			 MC_CMD_FILTER_OP_IN_HANDLE_OFST +
3548 			 MC_CMD_FILTER_OP_IN_HANDLE_LEN);
3549 	struct efx_filter_spec *spec;
3550 	DEFINE_WAIT(wait);
3551 	int rc;
3552 
3553 	/* Find the software table entry and mark it busy.  Don't
3554 	 * remove it yet; any attempt to update while we're waiting
3555 	 * for the firmware must find the busy entry.
3556 	 */
3557 	for (;;) {
3558 		spin_lock_bh(&efx->filter_lock);
3559 		if (!(table->entry[filter_idx].spec &
3560 		      EFX_EF10_FILTER_FLAG_BUSY))
3561 			break;
3562 		prepare_to_wait(&table->waitq, &wait, TASK_UNINTERRUPTIBLE);
3563 		spin_unlock_bh(&efx->filter_lock);
3564 		schedule();
3565 	}
3566 
3567 	spec = efx_ef10_filter_entry_spec(table, filter_idx);
3568 	if (!spec ||
3569 	    (!by_index &&
3570 	     efx_ef10_filter_pri(table, spec) !=
3571 	     filter_id / HUNT_FILTER_TBL_ROWS)) {
3572 		rc = -ENOENT;
3573 		goto out_unlock;
3574 	}
3575 
3576 	if (spec->flags & EFX_FILTER_FLAG_RX_OVER_AUTO &&
3577 	    priority_mask == (1U << EFX_FILTER_PRI_AUTO)) {
3578 		/* Just remove flags */
3579 		spec->flags &= ~EFX_FILTER_FLAG_RX_OVER_AUTO;
3580 		table->entry[filter_idx].spec &= ~EFX_EF10_FILTER_FLAG_AUTO_OLD;
3581 		rc = 0;
3582 		goto out_unlock;
3583 	}
3584 
3585 	if (!(priority_mask & (1U << spec->priority))) {
3586 		rc = -ENOENT;
3587 		goto out_unlock;
3588 	}
3589 
3590 	table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_BUSY;
3591 	spin_unlock_bh(&efx->filter_lock);
3592 
3593 	if (spec->flags & EFX_FILTER_FLAG_RX_OVER_AUTO) {
3594 		/* Reset to an automatic filter */
3595 
3596 		struct efx_filter_spec new_spec = *spec;
3597 
3598 		new_spec.priority = EFX_FILTER_PRI_AUTO;
3599 		new_spec.flags = (EFX_FILTER_FLAG_RX |
3600 				  (efx_rss_enabled(efx) ?
3601 				   EFX_FILTER_FLAG_RX_RSS : 0));
3602 		new_spec.dmaq_id = 0;
3603 		new_spec.rss_context = EFX_FILTER_RSS_CONTEXT_DEFAULT;
3604 		rc = efx_ef10_filter_push(efx, &new_spec,
3605 					  &table->entry[filter_idx].handle,
3606 					  true);
3607 
3608 		spin_lock_bh(&efx->filter_lock);
3609 		if (rc == 0)
3610 			*spec = new_spec;
3611 	} else {
3612 		/* Really remove the filter */
3613 
3614 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
3615 			       efx_ef10_filter_is_exclusive(spec) ?
3616 			       MC_CMD_FILTER_OP_IN_OP_REMOVE :
3617 			       MC_CMD_FILTER_OP_IN_OP_UNSUBSCRIBE);
3618 		MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
3619 			       table->entry[filter_idx].handle);
3620 		rc = efx_mcdi_rpc(efx, MC_CMD_FILTER_OP,
3621 				  inbuf, sizeof(inbuf), NULL, 0, NULL);
3622 
3623 		spin_lock_bh(&efx->filter_lock);
3624 		if (rc == 0) {
3625 			kfree(spec);
3626 			efx_ef10_filter_set_entry(table, filter_idx, NULL, 0);
3627 		}
3628 	}
3629 
3630 	table->entry[filter_idx].spec &= ~EFX_EF10_FILTER_FLAG_BUSY;
3631 	wake_up_all(&table->waitq);
3632 out_unlock:
3633 	spin_unlock_bh(&efx->filter_lock);
3634 	finish_wait(&table->waitq, &wait);
3635 	return rc;
3636 }
3637 
3638 static int efx_ef10_filter_remove_safe(struct efx_nic *efx,
3639 				       enum efx_filter_priority priority,
3640 				       u32 filter_id)
3641 {
3642 	return efx_ef10_filter_remove_internal(efx, 1U << priority,
3643 					       filter_id, false);
3644 }
3645 
3646 static u32 efx_ef10_filter_get_unsafe_id(struct efx_nic *efx, u32 filter_id)
3647 {
3648 	return filter_id % HUNT_FILTER_TBL_ROWS;
3649 }
3650 
3651 static void efx_ef10_filter_remove_unsafe(struct efx_nic *efx,
3652 					  enum efx_filter_priority priority,
3653 					  u32 filter_id)
3654 {
3655 	if (filter_id == EFX_EF10_FILTER_ID_INVALID)
3656 		return;
3657 	efx_ef10_filter_remove_internal(efx, 1U << priority, filter_id, true);
3658 }
3659 
3660 static int efx_ef10_filter_get_safe(struct efx_nic *efx,
3661 				    enum efx_filter_priority priority,
3662 				    u32 filter_id, struct efx_filter_spec *spec)
3663 {
3664 	unsigned int filter_idx = filter_id % HUNT_FILTER_TBL_ROWS;
3665 	struct efx_ef10_filter_table *table = efx->filter_state;
3666 	const struct efx_filter_spec *saved_spec;
3667 	int rc;
3668 
3669 	spin_lock_bh(&efx->filter_lock);
3670 	saved_spec = efx_ef10_filter_entry_spec(table, filter_idx);
3671 	if (saved_spec && saved_spec->priority == priority &&
3672 	    efx_ef10_filter_pri(table, saved_spec) ==
3673 	    filter_id / HUNT_FILTER_TBL_ROWS) {
3674 		*spec = *saved_spec;
3675 		rc = 0;
3676 	} else {
3677 		rc = -ENOENT;
3678 	}
3679 	spin_unlock_bh(&efx->filter_lock);
3680 	return rc;
3681 }
3682 
3683 static int efx_ef10_filter_clear_rx(struct efx_nic *efx,
3684 				     enum efx_filter_priority priority)
3685 {
3686 	unsigned int priority_mask;
3687 	unsigned int i;
3688 	int rc;
3689 
3690 	priority_mask = (((1U << (priority + 1)) - 1) &
3691 			 ~(1U << EFX_FILTER_PRI_AUTO));
3692 
3693 	for (i = 0; i < HUNT_FILTER_TBL_ROWS; i++) {
3694 		rc = efx_ef10_filter_remove_internal(efx, priority_mask,
3695 						     i, true);
3696 		if (rc && rc != -ENOENT)
3697 			return rc;
3698 	}
3699 
3700 	return 0;
3701 }
3702 
3703 static u32 efx_ef10_filter_count_rx_used(struct efx_nic *efx,
3704 					 enum efx_filter_priority priority)
3705 {
3706 	struct efx_ef10_filter_table *table = efx->filter_state;
3707 	unsigned int filter_idx;
3708 	s32 count = 0;
3709 
3710 	spin_lock_bh(&efx->filter_lock);
3711 	for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
3712 		if (table->entry[filter_idx].spec &&
3713 		    efx_ef10_filter_entry_spec(table, filter_idx)->priority ==
3714 		    priority)
3715 			++count;
3716 	}
3717 	spin_unlock_bh(&efx->filter_lock);
3718 	return count;
3719 }
3720 
3721 static u32 efx_ef10_filter_get_rx_id_limit(struct efx_nic *efx)
3722 {
3723 	struct efx_ef10_filter_table *table = efx->filter_state;
3724 
3725 	return table->rx_match_count * HUNT_FILTER_TBL_ROWS;
3726 }
3727 
3728 static s32 efx_ef10_filter_get_rx_ids(struct efx_nic *efx,
3729 				      enum efx_filter_priority priority,
3730 				      u32 *buf, u32 size)
3731 {
3732 	struct efx_ef10_filter_table *table = efx->filter_state;
3733 	struct efx_filter_spec *spec;
3734 	unsigned int filter_idx;
3735 	s32 count = 0;
3736 
3737 	spin_lock_bh(&efx->filter_lock);
3738 	for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
3739 		spec = efx_ef10_filter_entry_spec(table, filter_idx);
3740 		if (spec && spec->priority == priority) {
3741 			if (count == size) {
3742 				count = -EMSGSIZE;
3743 				break;
3744 			}
3745 			buf[count++] = (efx_ef10_filter_pri(table, spec) *
3746 					HUNT_FILTER_TBL_ROWS +
3747 					filter_idx);
3748 		}
3749 	}
3750 	spin_unlock_bh(&efx->filter_lock);
3751 	return count;
3752 }
3753 
3754 #ifdef CONFIG_RFS_ACCEL
3755 
3756 static efx_mcdi_async_completer efx_ef10_filter_rfs_insert_complete;
3757 
3758 static s32 efx_ef10_filter_rfs_insert(struct efx_nic *efx,
3759 				      struct efx_filter_spec *spec)
3760 {
3761 	struct efx_ef10_filter_table *table = efx->filter_state;
3762 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_IN_LEN);
3763 	struct efx_filter_spec *saved_spec;
3764 	unsigned int hash, i, depth = 1;
3765 	bool replacing = false;
3766 	int ins_index = -1;
3767 	u64 cookie;
3768 	s32 rc;
3769 
3770 	/* Must be an RX filter without RSS and not for a multicast
3771 	 * destination address (RFS only works for connected sockets).
3772 	 * These restrictions allow us to pass only a tiny amount of
3773 	 * data through to the completion function.
3774 	 */
3775 	EFX_WARN_ON_PARANOID(spec->flags !=
3776 			     (EFX_FILTER_FLAG_RX | EFX_FILTER_FLAG_RX_SCATTER));
3777 	EFX_WARN_ON_PARANOID(spec->priority != EFX_FILTER_PRI_HINT);
3778 	EFX_WARN_ON_PARANOID(efx_filter_is_mc_recipient(spec));
3779 
3780 	hash = efx_ef10_filter_hash(spec);
3781 
3782 	spin_lock_bh(&efx->filter_lock);
3783 
3784 	/* Find any existing filter with the same match tuple or else
3785 	 * a free slot to insert at.  If an existing filter is busy,
3786 	 * we have to give up.
3787 	 */
3788 	for (;;) {
3789 		i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
3790 		saved_spec = efx_ef10_filter_entry_spec(table, i);
3791 
3792 		if (!saved_spec) {
3793 			if (ins_index < 0)
3794 				ins_index = i;
3795 		} else if (efx_ef10_filter_equal(spec, saved_spec)) {
3796 			if (table->entry[i].spec & EFX_EF10_FILTER_FLAG_BUSY) {
3797 				rc = -EBUSY;
3798 				goto fail_unlock;
3799 			}
3800 			if (spec->priority < saved_spec->priority) {
3801 				rc = -EPERM;
3802 				goto fail_unlock;
3803 			}
3804 			ins_index = i;
3805 			break;
3806 		}
3807 
3808 		/* Once we reach the maximum search depth, use the
3809 		 * first suitable slot or return -EBUSY if there was
3810 		 * none
3811 		 */
3812 		if (depth == EFX_EF10_FILTER_SEARCH_LIMIT) {
3813 			if (ins_index < 0) {
3814 				rc = -EBUSY;
3815 				goto fail_unlock;
3816 			}
3817 			break;
3818 		}
3819 
3820 		++depth;
3821 	}
3822 
3823 	/* Create a software table entry if necessary, and mark it
3824 	 * busy.  We might yet fail to insert, but any attempt to
3825 	 * insert a conflicting filter while we're waiting for the
3826 	 * firmware must find the busy entry.
3827 	 */
3828 	saved_spec = efx_ef10_filter_entry_spec(table, ins_index);
3829 	if (saved_spec) {
3830 		replacing = true;
3831 	} else {
3832 		saved_spec = kmalloc(sizeof(*spec), GFP_ATOMIC);
3833 		if (!saved_spec) {
3834 			rc = -ENOMEM;
3835 			goto fail_unlock;
3836 		}
3837 		*saved_spec = *spec;
3838 	}
3839 	efx_ef10_filter_set_entry(table, ins_index, saved_spec,
3840 				  EFX_EF10_FILTER_FLAG_BUSY);
3841 
3842 	spin_unlock_bh(&efx->filter_lock);
3843 
3844 	/* Pack up the variables needed on completion */
3845 	cookie = replacing << 31 | ins_index << 16 | spec->dmaq_id;
3846 
3847 	efx_ef10_filter_push_prep(efx, spec, inbuf,
3848 				  table->entry[ins_index].handle, replacing);
3849 	efx_mcdi_rpc_async(efx, MC_CMD_FILTER_OP, inbuf, sizeof(inbuf),
3850 			   MC_CMD_FILTER_OP_OUT_LEN,
3851 			   efx_ef10_filter_rfs_insert_complete, cookie);
3852 
3853 	return ins_index;
3854 
3855 fail_unlock:
3856 	spin_unlock_bh(&efx->filter_lock);
3857 	return rc;
3858 }
3859 
3860 static void
3861 efx_ef10_filter_rfs_insert_complete(struct efx_nic *efx, unsigned long cookie,
3862 				    int rc, efx_dword_t *outbuf,
3863 				    size_t outlen_actual)
3864 {
3865 	struct efx_ef10_filter_table *table = efx->filter_state;
3866 	unsigned int ins_index, dmaq_id;
3867 	struct efx_filter_spec *spec;
3868 	bool replacing;
3869 
3870 	/* Unpack the cookie */
3871 	replacing = cookie >> 31;
3872 	ins_index = (cookie >> 16) & (HUNT_FILTER_TBL_ROWS - 1);
3873 	dmaq_id = cookie & 0xffff;
3874 
3875 	spin_lock_bh(&efx->filter_lock);
3876 	spec = efx_ef10_filter_entry_spec(table, ins_index);
3877 	if (rc == 0) {
3878 		table->entry[ins_index].handle =
3879 			MCDI_QWORD(outbuf, FILTER_OP_OUT_HANDLE);
3880 		if (replacing)
3881 			spec->dmaq_id = dmaq_id;
3882 	} else if (!replacing) {
3883 		kfree(spec);
3884 		spec = NULL;
3885 	}
3886 	efx_ef10_filter_set_entry(table, ins_index, spec, 0);
3887 	spin_unlock_bh(&efx->filter_lock);
3888 
3889 	wake_up_all(&table->waitq);
3890 }
3891 
3892 static void
3893 efx_ef10_filter_rfs_expire_complete(struct efx_nic *efx,
3894 				    unsigned long filter_idx,
3895 				    int rc, efx_dword_t *outbuf,
3896 				    size_t outlen_actual);
3897 
3898 static bool efx_ef10_filter_rfs_expire_one(struct efx_nic *efx, u32 flow_id,
3899 					   unsigned int filter_idx)
3900 {
3901 	struct efx_ef10_filter_table *table = efx->filter_state;
3902 	struct efx_filter_spec *spec =
3903 		efx_ef10_filter_entry_spec(table, filter_idx);
3904 	MCDI_DECLARE_BUF(inbuf,
3905 			 MC_CMD_FILTER_OP_IN_HANDLE_OFST +
3906 			 MC_CMD_FILTER_OP_IN_HANDLE_LEN);
3907 
3908 	if (!spec ||
3909 	    (table->entry[filter_idx].spec & EFX_EF10_FILTER_FLAG_BUSY) ||
3910 	    spec->priority != EFX_FILTER_PRI_HINT ||
3911 	    !rps_may_expire_flow(efx->net_dev, spec->dmaq_id,
3912 				 flow_id, filter_idx))
3913 		return false;
3914 
3915 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
3916 		       MC_CMD_FILTER_OP_IN_OP_REMOVE);
3917 	MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
3918 		       table->entry[filter_idx].handle);
3919 	if (efx_mcdi_rpc_async(efx, MC_CMD_FILTER_OP, inbuf, sizeof(inbuf), 0,
3920 			       efx_ef10_filter_rfs_expire_complete, filter_idx))
3921 		return false;
3922 
3923 	table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_BUSY;
3924 	return true;
3925 }
3926 
3927 static void
3928 efx_ef10_filter_rfs_expire_complete(struct efx_nic *efx,
3929 				    unsigned long filter_idx,
3930 				    int rc, efx_dword_t *outbuf,
3931 				    size_t outlen_actual)
3932 {
3933 	struct efx_ef10_filter_table *table = efx->filter_state;
3934 	struct efx_filter_spec *spec =
3935 		efx_ef10_filter_entry_spec(table, filter_idx);
3936 
3937 	spin_lock_bh(&efx->filter_lock);
3938 	if (rc == 0) {
3939 		kfree(spec);
3940 		efx_ef10_filter_set_entry(table, filter_idx, NULL, 0);
3941 	}
3942 	table->entry[filter_idx].spec &= ~EFX_EF10_FILTER_FLAG_BUSY;
3943 	wake_up_all(&table->waitq);
3944 	spin_unlock_bh(&efx->filter_lock);
3945 }
3946 
3947 #endif /* CONFIG_RFS_ACCEL */
3948 
3949 static int efx_ef10_filter_match_flags_from_mcdi(u32 mcdi_flags)
3950 {
3951 	int match_flags = 0;
3952 
3953 #define MAP_FLAG(gen_flag, mcdi_field) {				\
3954 		u32 old_mcdi_flags = mcdi_flags;			\
3955 		mcdi_flags &= ~(1 << MC_CMD_FILTER_OP_IN_MATCH_ ##	\
3956 				mcdi_field ## _LBN);			\
3957 		if (mcdi_flags != old_mcdi_flags)			\
3958 			match_flags |= EFX_FILTER_MATCH_ ## gen_flag;	\
3959 	}
3960 	MAP_FLAG(LOC_MAC_IG, UNKNOWN_UCAST_DST);
3961 	MAP_FLAG(LOC_MAC_IG, UNKNOWN_MCAST_DST);
3962 	MAP_FLAG(REM_HOST, SRC_IP);
3963 	MAP_FLAG(LOC_HOST, DST_IP);
3964 	MAP_FLAG(REM_MAC, SRC_MAC);
3965 	MAP_FLAG(REM_PORT, SRC_PORT);
3966 	MAP_FLAG(LOC_MAC, DST_MAC);
3967 	MAP_FLAG(LOC_PORT, DST_PORT);
3968 	MAP_FLAG(ETHER_TYPE, ETHER_TYPE);
3969 	MAP_FLAG(INNER_VID, INNER_VLAN);
3970 	MAP_FLAG(OUTER_VID, OUTER_VLAN);
3971 	MAP_FLAG(IP_PROTO, IP_PROTO);
3972 #undef MAP_FLAG
3973 
3974 	/* Did we map them all? */
3975 	if (mcdi_flags)
3976 		return -EINVAL;
3977 
3978 	return match_flags;
3979 }
3980 
3981 static void efx_ef10_filter_cleanup_vlans(struct efx_nic *efx)
3982 {
3983 	struct efx_ef10_filter_table *table = efx->filter_state;
3984 	struct efx_ef10_filter_vlan *vlan, *next_vlan;
3985 
3986 	/* See comment in efx_ef10_filter_table_remove() */
3987 	if (!efx_rwsem_assert_write_locked(&efx->filter_sem))
3988 		return;
3989 
3990 	if (!table)
3991 		return;
3992 
3993 	list_for_each_entry_safe(vlan, next_vlan, &table->vlan_list, list)
3994 		efx_ef10_filter_del_vlan_internal(efx, vlan);
3995 }
3996 
3997 static bool efx_ef10_filter_match_supported(struct efx_ef10_filter_table *table,
3998 					    enum efx_filter_match_flags match_flags)
3999 {
4000 	unsigned int match_pri;
4001 	int mf;
4002 
4003 	for (match_pri = 0;
4004 	     match_pri < table->rx_match_count;
4005 	     match_pri++) {
4006 		mf = efx_ef10_filter_match_flags_from_mcdi(
4007 				table->rx_match_mcdi_flags[match_pri]);
4008 		if (mf == match_flags)
4009 			return true;
4010 	}
4011 
4012 	return false;
4013 }
4014 
4015 static int efx_ef10_filter_table_probe(struct efx_nic *efx)
4016 {
4017 	MCDI_DECLARE_BUF(inbuf, MC_CMD_GET_PARSER_DISP_INFO_IN_LEN);
4018 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_PARSER_DISP_INFO_OUT_LENMAX);
4019 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
4020 	struct net_device *net_dev = efx->net_dev;
4021 	unsigned int pd_match_pri, pd_match_count;
4022 	struct efx_ef10_filter_table *table;
4023 	struct efx_ef10_vlan *vlan;
4024 	size_t outlen;
4025 	int rc;
4026 
4027 	if (!efx_rwsem_assert_write_locked(&efx->filter_sem))
4028 		return -EINVAL;
4029 
4030 	if (efx->filter_state) /* already probed */
4031 		return 0;
4032 
4033 	table = kzalloc(sizeof(*table), GFP_KERNEL);
4034 	if (!table)
4035 		return -ENOMEM;
4036 
4037 	/* Find out which RX filter types are supported, and their priorities */
4038 	MCDI_SET_DWORD(inbuf, GET_PARSER_DISP_INFO_IN_OP,
4039 		       MC_CMD_GET_PARSER_DISP_INFO_IN_OP_GET_SUPPORTED_RX_MATCHES);
4040 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_PARSER_DISP_INFO,
4041 			  inbuf, sizeof(inbuf), outbuf, sizeof(outbuf),
4042 			  &outlen);
4043 	if (rc)
4044 		goto fail;
4045 	pd_match_count = MCDI_VAR_ARRAY_LEN(
4046 		outlen, GET_PARSER_DISP_INFO_OUT_SUPPORTED_MATCHES);
4047 	table->rx_match_count = 0;
4048 
4049 	for (pd_match_pri = 0; pd_match_pri < pd_match_count; pd_match_pri++) {
4050 		u32 mcdi_flags =
4051 			MCDI_ARRAY_DWORD(
4052 				outbuf,
4053 				GET_PARSER_DISP_INFO_OUT_SUPPORTED_MATCHES,
4054 				pd_match_pri);
4055 		rc = efx_ef10_filter_match_flags_from_mcdi(mcdi_flags);
4056 		if (rc < 0) {
4057 			netif_dbg(efx, probe, efx->net_dev,
4058 				  "%s: fw flags %#x pri %u not supported in driver\n",
4059 				  __func__, mcdi_flags, pd_match_pri);
4060 		} else {
4061 			netif_dbg(efx, probe, efx->net_dev,
4062 				  "%s: fw flags %#x pri %u supported as driver flags %#x pri %u\n",
4063 				  __func__, mcdi_flags, pd_match_pri,
4064 				  rc, table->rx_match_count);
4065 			table->rx_match_mcdi_flags[table->rx_match_count] = mcdi_flags;
4066 			table->rx_match_count++;
4067 		}
4068 	}
4069 
4070 	if ((efx_supported_features(efx) & NETIF_F_HW_VLAN_CTAG_FILTER) &&
4071 	    !(efx_ef10_filter_match_supported(table,
4072 		(EFX_FILTER_MATCH_OUTER_VID | EFX_FILTER_MATCH_LOC_MAC)) &&
4073 	      efx_ef10_filter_match_supported(table,
4074 		(EFX_FILTER_MATCH_OUTER_VID | EFX_FILTER_MATCH_LOC_MAC_IG)))) {
4075 		netif_info(efx, probe, net_dev,
4076 			   "VLAN filters are not supported in this firmware variant\n");
4077 		net_dev->features &= ~NETIF_F_HW_VLAN_CTAG_FILTER;
4078 		efx->fixed_features &= ~NETIF_F_HW_VLAN_CTAG_FILTER;
4079 		net_dev->hw_features &= ~NETIF_F_HW_VLAN_CTAG_FILTER;
4080 	}
4081 
4082 	table->entry = vzalloc(HUNT_FILTER_TBL_ROWS * sizeof(*table->entry));
4083 	if (!table->entry) {
4084 		rc = -ENOMEM;
4085 		goto fail;
4086 	}
4087 
4088 	table->mc_promisc_last = false;
4089 	table->vlan_filter =
4090 		!!(efx->net_dev->features & NETIF_F_HW_VLAN_CTAG_FILTER);
4091 	INIT_LIST_HEAD(&table->vlan_list);
4092 
4093 	efx->filter_state = table;
4094 	init_waitqueue_head(&table->waitq);
4095 
4096 	list_for_each_entry(vlan, &nic_data->vlan_list, list) {
4097 		rc = efx_ef10_filter_add_vlan(efx, vlan->vid);
4098 		if (rc)
4099 			goto fail_add_vlan;
4100 	}
4101 
4102 	return 0;
4103 
4104 fail_add_vlan:
4105 	efx_ef10_filter_cleanup_vlans(efx);
4106 	efx->filter_state = NULL;
4107 fail:
4108 	kfree(table);
4109 	return rc;
4110 }
4111 
4112 /* Caller must hold efx->filter_sem for read if race against
4113  * efx_ef10_filter_table_remove() is possible
4114  */
4115 static void efx_ef10_filter_table_restore(struct efx_nic *efx)
4116 {
4117 	struct efx_ef10_filter_table *table = efx->filter_state;
4118 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
4119 	struct efx_filter_spec *spec;
4120 	unsigned int filter_idx;
4121 	bool failed = false;
4122 	int rc;
4123 
4124 	WARN_ON(!rwsem_is_locked(&efx->filter_sem));
4125 
4126 	if (!nic_data->must_restore_filters)
4127 		return;
4128 
4129 	if (!table)
4130 		return;
4131 
4132 	spin_lock_bh(&efx->filter_lock);
4133 
4134 	for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
4135 		spec = efx_ef10_filter_entry_spec(table, filter_idx);
4136 		if (!spec)
4137 			continue;
4138 
4139 		table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_BUSY;
4140 		spin_unlock_bh(&efx->filter_lock);
4141 
4142 		rc = efx_ef10_filter_push(efx, spec,
4143 					  &table->entry[filter_idx].handle,
4144 					  false);
4145 		if (rc)
4146 			failed = true;
4147 
4148 		spin_lock_bh(&efx->filter_lock);
4149 		if (rc) {
4150 			kfree(spec);
4151 			efx_ef10_filter_set_entry(table, filter_idx, NULL, 0);
4152 		} else {
4153 			table->entry[filter_idx].spec &=
4154 				~EFX_EF10_FILTER_FLAG_BUSY;
4155 		}
4156 	}
4157 
4158 	spin_unlock_bh(&efx->filter_lock);
4159 
4160 	if (failed)
4161 		netif_err(efx, hw, efx->net_dev,
4162 			  "unable to restore all filters\n");
4163 	else
4164 		nic_data->must_restore_filters = false;
4165 }
4166 
4167 static void efx_ef10_filter_table_remove(struct efx_nic *efx)
4168 {
4169 	struct efx_ef10_filter_table *table = efx->filter_state;
4170 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_IN_LEN);
4171 	struct efx_filter_spec *spec;
4172 	unsigned int filter_idx;
4173 	int rc;
4174 
4175 	efx_ef10_filter_cleanup_vlans(efx);
4176 	efx->filter_state = NULL;
4177 	/* If we were called without locking, then it's not safe to free
4178 	 * the table as others might be using it.  So we just WARN, leak
4179 	 * the memory, and potentially get an inconsistent filter table
4180 	 * state.
4181 	 * This should never actually happen.
4182 	 */
4183 	if (!efx_rwsem_assert_write_locked(&efx->filter_sem))
4184 		return;
4185 
4186 	if (!table)
4187 		return;
4188 
4189 	for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
4190 		spec = efx_ef10_filter_entry_spec(table, filter_idx);
4191 		if (!spec)
4192 			continue;
4193 
4194 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
4195 			       efx_ef10_filter_is_exclusive(spec) ?
4196 			       MC_CMD_FILTER_OP_IN_OP_REMOVE :
4197 			       MC_CMD_FILTER_OP_IN_OP_UNSUBSCRIBE);
4198 		MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
4199 			       table->entry[filter_idx].handle);
4200 		rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FILTER_OP, inbuf,
4201 					sizeof(inbuf), NULL, 0, NULL);
4202 		if (rc)
4203 			netif_info(efx, drv, efx->net_dev,
4204 				   "%s: filter %04x remove failed\n",
4205 				   __func__, filter_idx);
4206 		kfree(spec);
4207 	}
4208 
4209 	vfree(table->entry);
4210 	kfree(table);
4211 }
4212 
4213 static void efx_ef10_filter_mark_one_old(struct efx_nic *efx, uint16_t *id)
4214 {
4215 	struct efx_ef10_filter_table *table = efx->filter_state;
4216 	unsigned int filter_idx;
4217 
4218 	if (*id != EFX_EF10_FILTER_ID_INVALID) {
4219 		filter_idx = efx_ef10_filter_get_unsafe_id(efx, *id);
4220 		if (!table->entry[filter_idx].spec)
4221 			netif_dbg(efx, drv, efx->net_dev,
4222 				  "marked null spec old %04x:%04x\n", *id,
4223 				  filter_idx);
4224 		table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_AUTO_OLD;
4225 		*id = EFX_EF10_FILTER_ID_INVALID;
4226 	}
4227 }
4228 
4229 /* Mark old per-VLAN filters that may need to be removed */
4230 static void _efx_ef10_filter_vlan_mark_old(struct efx_nic *efx,
4231 					   struct efx_ef10_filter_vlan *vlan)
4232 {
4233 	struct efx_ef10_filter_table *table = efx->filter_state;
4234 	unsigned int i;
4235 
4236 	for (i = 0; i < table->dev_uc_count; i++)
4237 		efx_ef10_filter_mark_one_old(efx, &vlan->uc[i]);
4238 	for (i = 0; i < table->dev_mc_count; i++)
4239 		efx_ef10_filter_mark_one_old(efx, &vlan->mc[i]);
4240 	efx_ef10_filter_mark_one_old(efx, &vlan->ucdef);
4241 	efx_ef10_filter_mark_one_old(efx, &vlan->bcast);
4242 	efx_ef10_filter_mark_one_old(efx, &vlan->mcdef);
4243 }
4244 
4245 /* Mark old filters that may need to be removed.
4246  * Caller must hold efx->filter_sem for read if race against
4247  * efx_ef10_filter_table_remove() is possible
4248  */
4249 static void efx_ef10_filter_mark_old(struct efx_nic *efx)
4250 {
4251 	struct efx_ef10_filter_table *table = efx->filter_state;
4252 	struct efx_ef10_filter_vlan *vlan;
4253 
4254 	spin_lock_bh(&efx->filter_lock);
4255 	list_for_each_entry(vlan, &table->vlan_list, list)
4256 		_efx_ef10_filter_vlan_mark_old(efx, vlan);
4257 	spin_unlock_bh(&efx->filter_lock);
4258 }
4259 
4260 static void efx_ef10_filter_uc_addr_list(struct efx_nic *efx)
4261 {
4262 	struct efx_ef10_filter_table *table = efx->filter_state;
4263 	struct net_device *net_dev = efx->net_dev;
4264 	struct netdev_hw_addr *uc;
4265 	int addr_count;
4266 	unsigned int i;
4267 
4268 	addr_count = netdev_uc_count(net_dev);
4269 	table->uc_promisc = !!(net_dev->flags & IFF_PROMISC);
4270 	table->dev_uc_count = 1 + addr_count;
4271 	ether_addr_copy(table->dev_uc_list[0].addr, net_dev->dev_addr);
4272 	i = 1;
4273 	netdev_for_each_uc_addr(uc, net_dev) {
4274 		if (i >= EFX_EF10_FILTER_DEV_UC_MAX) {
4275 			table->uc_promisc = true;
4276 			break;
4277 		}
4278 		ether_addr_copy(table->dev_uc_list[i].addr, uc->addr);
4279 		i++;
4280 	}
4281 }
4282 
4283 static void efx_ef10_filter_mc_addr_list(struct efx_nic *efx)
4284 {
4285 	struct efx_ef10_filter_table *table = efx->filter_state;
4286 	struct net_device *net_dev = efx->net_dev;
4287 	struct netdev_hw_addr *mc;
4288 	unsigned int i, addr_count;
4289 
4290 	table->mc_promisc = !!(net_dev->flags & (IFF_PROMISC | IFF_ALLMULTI));
4291 
4292 	addr_count = netdev_mc_count(net_dev);
4293 	i = 0;
4294 	netdev_for_each_mc_addr(mc, net_dev) {
4295 		if (i >= EFX_EF10_FILTER_DEV_MC_MAX) {
4296 			table->mc_promisc = true;
4297 			break;
4298 		}
4299 		ether_addr_copy(table->dev_mc_list[i].addr, mc->addr);
4300 		i++;
4301 	}
4302 
4303 	table->dev_mc_count = i;
4304 }
4305 
4306 static int efx_ef10_filter_insert_addr_list(struct efx_nic *efx,
4307 					    struct efx_ef10_filter_vlan *vlan,
4308 					    bool multicast, bool rollback)
4309 {
4310 	struct efx_ef10_filter_table *table = efx->filter_state;
4311 	struct efx_ef10_dev_addr *addr_list;
4312 	enum efx_filter_flags filter_flags;
4313 	struct efx_filter_spec spec;
4314 	u8 baddr[ETH_ALEN];
4315 	unsigned int i, j;
4316 	int addr_count;
4317 	u16 *ids;
4318 	int rc;
4319 
4320 	if (multicast) {
4321 		addr_list = table->dev_mc_list;
4322 		addr_count = table->dev_mc_count;
4323 		ids = vlan->mc;
4324 	} else {
4325 		addr_list = table->dev_uc_list;
4326 		addr_count = table->dev_uc_count;
4327 		ids = vlan->uc;
4328 	}
4329 
4330 	filter_flags = efx_rss_enabled(efx) ? EFX_FILTER_FLAG_RX_RSS : 0;
4331 
4332 	/* Insert/renew filters */
4333 	for (i = 0; i < addr_count; i++) {
4334 		efx_filter_init_rx(&spec, EFX_FILTER_PRI_AUTO, filter_flags, 0);
4335 		efx_filter_set_eth_local(&spec, vlan->vid, addr_list[i].addr);
4336 		rc = efx_ef10_filter_insert(efx, &spec, true);
4337 		if (rc < 0) {
4338 			if (rollback) {
4339 				netif_info(efx, drv, efx->net_dev,
4340 					   "efx_ef10_filter_insert failed rc=%d\n",
4341 					   rc);
4342 				/* Fall back to promiscuous */
4343 				for (j = 0; j < i; j++) {
4344 					efx_ef10_filter_remove_unsafe(
4345 						efx, EFX_FILTER_PRI_AUTO,
4346 						ids[j]);
4347 					ids[j] = EFX_EF10_FILTER_ID_INVALID;
4348 				}
4349 				return rc;
4350 			} else {
4351 				/* mark as not inserted, and carry on */
4352 				rc = EFX_EF10_FILTER_ID_INVALID;
4353 			}
4354 		}
4355 		ids[i] = efx_ef10_filter_get_unsafe_id(efx, rc);
4356 	}
4357 
4358 	if (multicast && rollback) {
4359 		/* Also need an Ethernet broadcast filter */
4360 		efx_filter_init_rx(&spec, EFX_FILTER_PRI_AUTO, filter_flags, 0);
4361 		eth_broadcast_addr(baddr);
4362 		efx_filter_set_eth_local(&spec, vlan->vid, baddr);
4363 		rc = efx_ef10_filter_insert(efx, &spec, true);
4364 		if (rc < 0) {
4365 			netif_warn(efx, drv, efx->net_dev,
4366 				   "Broadcast filter insert failed rc=%d\n", rc);
4367 			/* Fall back to promiscuous */
4368 			for (j = 0; j < i; j++) {
4369 				efx_ef10_filter_remove_unsafe(
4370 					efx, EFX_FILTER_PRI_AUTO,
4371 					ids[j]);
4372 				ids[j] = EFX_EF10_FILTER_ID_INVALID;
4373 			}
4374 			return rc;
4375 		} else {
4376 			EFX_WARN_ON_PARANOID(vlan->bcast !=
4377 					     EFX_EF10_FILTER_ID_INVALID);
4378 			vlan->bcast = efx_ef10_filter_get_unsafe_id(efx, rc);
4379 		}
4380 	}
4381 
4382 	return 0;
4383 }
4384 
4385 static int efx_ef10_filter_insert_def(struct efx_nic *efx,
4386 				      struct efx_ef10_filter_vlan *vlan,
4387 				      bool multicast, bool rollback)
4388 {
4389 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
4390 	enum efx_filter_flags filter_flags;
4391 	struct efx_filter_spec spec;
4392 	u8 baddr[ETH_ALEN];
4393 	int rc;
4394 
4395 	filter_flags = efx_rss_enabled(efx) ? EFX_FILTER_FLAG_RX_RSS : 0;
4396 
4397 	efx_filter_init_rx(&spec, EFX_FILTER_PRI_AUTO, filter_flags, 0);
4398 
4399 	if (multicast)
4400 		efx_filter_set_mc_def(&spec);
4401 	else
4402 		efx_filter_set_uc_def(&spec);
4403 
4404 	if (vlan->vid != EFX_FILTER_VID_UNSPEC)
4405 		efx_filter_set_eth_local(&spec, vlan->vid, NULL);
4406 
4407 	rc = efx_ef10_filter_insert(efx, &spec, true);
4408 	if (rc < 0) {
4409 		netif_printk(efx, drv, rc == -EPERM ? KERN_DEBUG : KERN_WARNING,
4410 			     efx->net_dev,
4411 			     "%scast mismatch filter insert failed rc=%d\n",
4412 			     multicast ? "Multi" : "Uni", rc);
4413 	} else if (multicast) {
4414 		EFX_WARN_ON_PARANOID(vlan->mcdef != EFX_EF10_FILTER_ID_INVALID);
4415 		vlan->mcdef = efx_ef10_filter_get_unsafe_id(efx, rc);
4416 		if (!nic_data->workaround_26807) {
4417 			/* Also need an Ethernet broadcast filter */
4418 			efx_filter_init_rx(&spec, EFX_FILTER_PRI_AUTO,
4419 					   filter_flags, 0);
4420 			eth_broadcast_addr(baddr);
4421 			efx_filter_set_eth_local(&spec, vlan->vid, baddr);
4422 			rc = efx_ef10_filter_insert(efx, &spec, true);
4423 			if (rc < 0) {
4424 				netif_warn(efx, drv, efx->net_dev,
4425 					   "Broadcast filter insert failed rc=%d\n",
4426 					   rc);
4427 				if (rollback) {
4428 					/* Roll back the mc_def filter */
4429 					efx_ef10_filter_remove_unsafe(
4430 							efx, EFX_FILTER_PRI_AUTO,
4431 							vlan->mcdef);
4432 					vlan->mcdef = EFX_EF10_FILTER_ID_INVALID;
4433 					return rc;
4434 				}
4435 			} else {
4436 				EFX_WARN_ON_PARANOID(vlan->bcast !=
4437 						     EFX_EF10_FILTER_ID_INVALID);
4438 				vlan->bcast = efx_ef10_filter_get_unsafe_id(efx, rc);
4439 			}
4440 		}
4441 		rc = 0;
4442 	} else {
4443 		EFX_WARN_ON_PARANOID(vlan->ucdef != EFX_EF10_FILTER_ID_INVALID);
4444 		vlan->ucdef = rc;
4445 		rc = 0;
4446 	}
4447 	return rc;
4448 }
4449 
4450 /* Remove filters that weren't renewed.  Since nothing else changes the AUTO_OLD
4451  * flag or removes these filters, we don't need to hold the filter_lock while
4452  * scanning for these filters.
4453  */
4454 static void efx_ef10_filter_remove_old(struct efx_nic *efx)
4455 {
4456 	struct efx_ef10_filter_table *table = efx->filter_state;
4457 	int remove_failed = 0;
4458 	int remove_noent = 0;
4459 	int rc;
4460 	int i;
4461 
4462 	for (i = 0; i < HUNT_FILTER_TBL_ROWS; i++) {
4463 		if (ACCESS_ONCE(table->entry[i].spec) &
4464 		    EFX_EF10_FILTER_FLAG_AUTO_OLD) {
4465 			rc = efx_ef10_filter_remove_internal(efx,
4466 					1U << EFX_FILTER_PRI_AUTO, i, true);
4467 			if (rc == -ENOENT)
4468 				remove_noent++;
4469 			else if (rc)
4470 				remove_failed++;
4471 		}
4472 	}
4473 
4474 	if (remove_failed)
4475 		netif_info(efx, drv, efx->net_dev,
4476 			   "%s: failed to remove %d filters\n",
4477 			   __func__, remove_failed);
4478 	if (remove_noent)
4479 		netif_info(efx, drv, efx->net_dev,
4480 			   "%s: failed to remove %d non-existent filters\n",
4481 			   __func__, remove_noent);
4482 }
4483 
4484 static int efx_ef10_vport_set_mac_address(struct efx_nic *efx)
4485 {
4486 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
4487 	u8 mac_old[ETH_ALEN];
4488 	int rc, rc2;
4489 
4490 	/* Only reconfigure a PF-created vport */
4491 	if (is_zero_ether_addr(nic_data->vport_mac))
4492 		return 0;
4493 
4494 	efx_device_detach_sync(efx);
4495 	efx_net_stop(efx->net_dev);
4496 	down_write(&efx->filter_sem);
4497 	efx_ef10_filter_table_remove(efx);
4498 	up_write(&efx->filter_sem);
4499 
4500 	rc = efx_ef10_vadaptor_free(efx, nic_data->vport_id);
4501 	if (rc)
4502 		goto restore_filters;
4503 
4504 	ether_addr_copy(mac_old, nic_data->vport_mac);
4505 	rc = efx_ef10_vport_del_mac(efx, nic_data->vport_id,
4506 				    nic_data->vport_mac);
4507 	if (rc)
4508 		goto restore_vadaptor;
4509 
4510 	rc = efx_ef10_vport_add_mac(efx, nic_data->vport_id,
4511 				    efx->net_dev->dev_addr);
4512 	if (!rc) {
4513 		ether_addr_copy(nic_data->vport_mac, efx->net_dev->dev_addr);
4514 	} else {
4515 		rc2 = efx_ef10_vport_add_mac(efx, nic_data->vport_id, mac_old);
4516 		if (rc2) {
4517 			/* Failed to add original MAC, so clear vport_mac */
4518 			eth_zero_addr(nic_data->vport_mac);
4519 			goto reset_nic;
4520 		}
4521 	}
4522 
4523 restore_vadaptor:
4524 	rc2 = efx_ef10_vadaptor_alloc(efx, nic_data->vport_id);
4525 	if (rc2)
4526 		goto reset_nic;
4527 restore_filters:
4528 	down_write(&efx->filter_sem);
4529 	rc2 = efx_ef10_filter_table_probe(efx);
4530 	up_write(&efx->filter_sem);
4531 	if (rc2)
4532 		goto reset_nic;
4533 
4534 	rc2 = efx_net_open(efx->net_dev);
4535 	if (rc2)
4536 		goto reset_nic;
4537 
4538 	netif_device_attach(efx->net_dev);
4539 
4540 	return rc;
4541 
4542 reset_nic:
4543 	netif_err(efx, drv, efx->net_dev,
4544 		  "Failed to restore when changing MAC address - scheduling reset\n");
4545 	efx_schedule_reset(efx, RESET_TYPE_DATAPATH);
4546 
4547 	return rc ? rc : rc2;
4548 }
4549 
4550 /* Caller must hold efx->filter_sem for read if race against
4551  * efx_ef10_filter_table_remove() is possible
4552  */
4553 static void efx_ef10_filter_vlan_sync_rx_mode(struct efx_nic *efx,
4554 					      struct efx_ef10_filter_vlan *vlan)
4555 {
4556 	struct efx_ef10_filter_table *table = efx->filter_state;
4557 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
4558 
4559 	/* Do not install unspecified VID if VLAN filtering is enabled.
4560 	 * Do not install all specified VIDs if VLAN filtering is disabled.
4561 	 */
4562 	if ((vlan->vid == EFX_FILTER_VID_UNSPEC) == table->vlan_filter)
4563 		return;
4564 
4565 	/* Insert/renew unicast filters */
4566 	if (table->uc_promisc) {
4567 		efx_ef10_filter_insert_def(efx, vlan, false, false);
4568 		efx_ef10_filter_insert_addr_list(efx, vlan, false, false);
4569 	} else {
4570 		/* If any of the filters failed to insert, fall back to
4571 		 * promiscuous mode - add in the uc_def filter.  But keep
4572 		 * our individual unicast filters.
4573 		 */
4574 		if (efx_ef10_filter_insert_addr_list(efx, vlan, false, false))
4575 			efx_ef10_filter_insert_def(efx, vlan, false, false);
4576 	}
4577 
4578 	/* Insert/renew multicast filters */
4579 	/* If changing promiscuous state with cascaded multicast filters, remove
4580 	 * old filters first, so that packets are dropped rather than duplicated
4581 	 */
4582 	if (nic_data->workaround_26807 &&
4583 	    table->mc_promisc_last != table->mc_promisc)
4584 		efx_ef10_filter_remove_old(efx);
4585 	if (table->mc_promisc) {
4586 		if (nic_data->workaround_26807) {
4587 			/* If we failed to insert promiscuous filters, rollback
4588 			 * and fall back to individual multicast filters
4589 			 */
4590 			if (efx_ef10_filter_insert_def(efx, vlan, true, true)) {
4591 				/* Changing promisc state, so remove old filters */
4592 				efx_ef10_filter_remove_old(efx);
4593 				efx_ef10_filter_insert_addr_list(efx, vlan,
4594 								 true, false);
4595 			}
4596 		} else {
4597 			/* If we failed to insert promiscuous filters, don't
4598 			 * rollback.  Regardless, also insert the mc_list
4599 			 */
4600 			efx_ef10_filter_insert_def(efx, vlan, true, false);
4601 			efx_ef10_filter_insert_addr_list(efx, vlan, true, false);
4602 		}
4603 	} else {
4604 		/* If any filters failed to insert, rollback and fall back to
4605 		 * promiscuous mode - mc_def filter and maybe broadcast.  If
4606 		 * that fails, roll back again and insert as many of our
4607 		 * individual multicast filters as we can.
4608 		 */
4609 		if (efx_ef10_filter_insert_addr_list(efx, vlan, true, true)) {
4610 			/* Changing promisc state, so remove old filters */
4611 			if (nic_data->workaround_26807)
4612 				efx_ef10_filter_remove_old(efx);
4613 			if (efx_ef10_filter_insert_def(efx, vlan, true, true))
4614 				efx_ef10_filter_insert_addr_list(efx, vlan,
4615 								 true, false);
4616 		}
4617 	}
4618 }
4619 
4620 /* Caller must hold efx->filter_sem for read if race against
4621  * efx_ef10_filter_table_remove() is possible
4622  */
4623 static void efx_ef10_filter_sync_rx_mode(struct efx_nic *efx)
4624 {
4625 	struct efx_ef10_filter_table *table = efx->filter_state;
4626 	struct net_device *net_dev = efx->net_dev;
4627 	struct efx_ef10_filter_vlan *vlan;
4628 	bool vlan_filter;
4629 
4630 	if (!efx_dev_registered(efx))
4631 		return;
4632 
4633 	if (!table)
4634 		return;
4635 
4636 	efx_ef10_filter_mark_old(efx);
4637 
4638 	/* Copy/convert the address lists; add the primary station
4639 	 * address and broadcast address
4640 	 */
4641 	netif_addr_lock_bh(net_dev);
4642 	efx_ef10_filter_uc_addr_list(efx);
4643 	efx_ef10_filter_mc_addr_list(efx);
4644 	netif_addr_unlock_bh(net_dev);
4645 
4646 	/* If VLAN filtering changes, all old filters are finally removed.
4647 	 * Do it in advance to avoid conflicts for unicast untagged and
4648 	 * VLAN 0 tagged filters.
4649 	 */
4650 	vlan_filter = !!(net_dev->features & NETIF_F_HW_VLAN_CTAG_FILTER);
4651 	if (table->vlan_filter != vlan_filter) {
4652 		table->vlan_filter = vlan_filter;
4653 		efx_ef10_filter_remove_old(efx);
4654 	}
4655 
4656 	list_for_each_entry(vlan, &table->vlan_list, list)
4657 		efx_ef10_filter_vlan_sync_rx_mode(efx, vlan);
4658 
4659 	efx_ef10_filter_remove_old(efx);
4660 	table->mc_promisc_last = table->mc_promisc;
4661 }
4662 
4663 static struct efx_ef10_filter_vlan *efx_ef10_filter_find_vlan(struct efx_nic *efx, u16 vid)
4664 {
4665 	struct efx_ef10_filter_table *table = efx->filter_state;
4666 	struct efx_ef10_filter_vlan *vlan;
4667 
4668 	WARN_ON(!rwsem_is_locked(&efx->filter_sem));
4669 
4670 	list_for_each_entry(vlan, &table->vlan_list, list) {
4671 		if (vlan->vid == vid)
4672 			return vlan;
4673 	}
4674 
4675 	return NULL;
4676 }
4677 
4678 static int efx_ef10_filter_add_vlan(struct efx_nic *efx, u16 vid)
4679 {
4680 	struct efx_ef10_filter_table *table = efx->filter_state;
4681 	struct efx_ef10_filter_vlan *vlan;
4682 	unsigned int i;
4683 
4684 	if (!efx_rwsem_assert_write_locked(&efx->filter_sem))
4685 		return -EINVAL;
4686 
4687 	vlan = efx_ef10_filter_find_vlan(efx, vid);
4688 	if (WARN_ON(vlan)) {
4689 		netif_err(efx, drv, efx->net_dev,
4690 			  "VLAN %u already added\n", vid);
4691 		return -EALREADY;
4692 	}
4693 
4694 	vlan = kzalloc(sizeof(*vlan), GFP_KERNEL);
4695 	if (!vlan)
4696 		return -ENOMEM;
4697 
4698 	vlan->vid = vid;
4699 
4700 	for (i = 0; i < ARRAY_SIZE(vlan->uc); i++)
4701 		vlan->uc[i] = EFX_EF10_FILTER_ID_INVALID;
4702 	for (i = 0; i < ARRAY_SIZE(vlan->mc); i++)
4703 		vlan->mc[i] = EFX_EF10_FILTER_ID_INVALID;
4704 	vlan->ucdef = EFX_EF10_FILTER_ID_INVALID;
4705 	vlan->bcast = EFX_EF10_FILTER_ID_INVALID;
4706 	vlan->mcdef = EFX_EF10_FILTER_ID_INVALID;
4707 
4708 	list_add_tail(&vlan->list, &table->vlan_list);
4709 
4710 	if (efx_dev_registered(efx))
4711 		efx_ef10_filter_vlan_sync_rx_mode(efx, vlan);
4712 
4713 	return 0;
4714 }
4715 
4716 static void efx_ef10_filter_del_vlan_internal(struct efx_nic *efx,
4717 					      struct efx_ef10_filter_vlan *vlan)
4718 {
4719 	unsigned int i;
4720 
4721 	/* See comment in efx_ef10_filter_table_remove() */
4722 	if (!efx_rwsem_assert_write_locked(&efx->filter_sem))
4723 		return;
4724 
4725 	list_del(&vlan->list);
4726 
4727 	for (i = 0; i < ARRAY_SIZE(vlan->uc); i++)
4728 		efx_ef10_filter_remove_unsafe(efx, EFX_FILTER_PRI_AUTO,
4729 					      vlan->uc[i]);
4730 	for (i = 0; i < ARRAY_SIZE(vlan->mc); i++)
4731 		efx_ef10_filter_remove_unsafe(efx, EFX_FILTER_PRI_AUTO,
4732 					      vlan->mc[i]);
4733 	efx_ef10_filter_remove_unsafe(efx, EFX_FILTER_PRI_AUTO, vlan->ucdef);
4734 	efx_ef10_filter_remove_unsafe(efx, EFX_FILTER_PRI_AUTO, vlan->bcast);
4735 	efx_ef10_filter_remove_unsafe(efx, EFX_FILTER_PRI_AUTO, vlan->mcdef);
4736 
4737 	kfree(vlan);
4738 }
4739 
4740 static void efx_ef10_filter_del_vlan(struct efx_nic *efx, u16 vid)
4741 {
4742 	struct efx_ef10_filter_vlan *vlan;
4743 
4744 	/* See comment in efx_ef10_filter_table_remove() */
4745 	if (!efx_rwsem_assert_write_locked(&efx->filter_sem))
4746 		return;
4747 
4748 	vlan = efx_ef10_filter_find_vlan(efx, vid);
4749 	if (!vlan) {
4750 		netif_err(efx, drv, efx->net_dev,
4751 			  "VLAN %u not found in filter state\n", vid);
4752 		return;
4753 	}
4754 
4755 	efx_ef10_filter_del_vlan_internal(efx, vlan);
4756 }
4757 
4758 static int efx_ef10_set_mac_address(struct efx_nic *efx)
4759 {
4760 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VADAPTOR_SET_MAC_IN_LEN);
4761 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
4762 	bool was_enabled = efx->port_enabled;
4763 	int rc;
4764 
4765 	efx_device_detach_sync(efx);
4766 	efx_net_stop(efx->net_dev);
4767 
4768 	mutex_lock(&efx->mac_lock);
4769 	down_write(&efx->filter_sem);
4770 	efx_ef10_filter_table_remove(efx);
4771 
4772 	ether_addr_copy(MCDI_PTR(inbuf, VADAPTOR_SET_MAC_IN_MACADDR),
4773 			efx->net_dev->dev_addr);
4774 	MCDI_SET_DWORD(inbuf, VADAPTOR_SET_MAC_IN_UPSTREAM_PORT_ID,
4775 		       nic_data->vport_id);
4776 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_VADAPTOR_SET_MAC, inbuf,
4777 				sizeof(inbuf), NULL, 0, NULL);
4778 
4779 	efx_ef10_filter_table_probe(efx);
4780 	up_write(&efx->filter_sem);
4781 	mutex_unlock(&efx->mac_lock);
4782 
4783 	if (was_enabled)
4784 		efx_net_open(efx->net_dev);
4785 	netif_device_attach(efx->net_dev);
4786 
4787 #ifdef CONFIG_SFC_SRIOV
4788 	if (efx->pci_dev->is_virtfn && efx->pci_dev->physfn) {
4789 		struct pci_dev *pci_dev_pf = efx->pci_dev->physfn;
4790 
4791 		if (rc == -EPERM) {
4792 			struct efx_nic *efx_pf;
4793 
4794 			/* Switch to PF and change MAC address on vport */
4795 			efx_pf = pci_get_drvdata(pci_dev_pf);
4796 
4797 			rc = efx_ef10_sriov_set_vf_mac(efx_pf,
4798 						       nic_data->vf_index,
4799 						       efx->net_dev->dev_addr);
4800 		} else if (!rc) {
4801 			struct efx_nic *efx_pf = pci_get_drvdata(pci_dev_pf);
4802 			struct efx_ef10_nic_data *nic_data = efx_pf->nic_data;
4803 			unsigned int i;
4804 
4805 			/* MAC address successfully changed by VF (with MAC
4806 			 * spoofing) so update the parent PF if possible.
4807 			 */
4808 			for (i = 0; i < efx_pf->vf_count; ++i) {
4809 				struct ef10_vf *vf = nic_data->vf + i;
4810 
4811 				if (vf->efx == efx) {
4812 					ether_addr_copy(vf->mac,
4813 							efx->net_dev->dev_addr);
4814 					return 0;
4815 				}
4816 			}
4817 		}
4818 	} else
4819 #endif
4820 	if (rc == -EPERM) {
4821 		netif_err(efx, drv, efx->net_dev,
4822 			  "Cannot change MAC address; use sfboot to enable"
4823 			  " mac-spoofing on this interface\n");
4824 	} else if (rc == -ENOSYS && !efx_ef10_is_vf(efx)) {
4825 		/* If the active MCFW does not support MC_CMD_VADAPTOR_SET_MAC
4826 		 * fall-back to the method of changing the MAC address on the
4827 		 * vport.  This only applies to PFs because such versions of
4828 		 * MCFW do not support VFs.
4829 		 */
4830 		rc = efx_ef10_vport_set_mac_address(efx);
4831 	} else {
4832 		efx_mcdi_display_error(efx, MC_CMD_VADAPTOR_SET_MAC,
4833 				       sizeof(inbuf), NULL, 0, rc);
4834 	}
4835 
4836 	return rc;
4837 }
4838 
4839 static int efx_ef10_mac_reconfigure(struct efx_nic *efx)
4840 {
4841 	efx_ef10_filter_sync_rx_mode(efx);
4842 
4843 	return efx_mcdi_set_mac(efx);
4844 }
4845 
4846 static int efx_ef10_mac_reconfigure_vf(struct efx_nic *efx)
4847 {
4848 	efx_ef10_filter_sync_rx_mode(efx);
4849 
4850 	return 0;
4851 }
4852 
4853 static int efx_ef10_start_bist(struct efx_nic *efx, u32 bist_type)
4854 {
4855 	MCDI_DECLARE_BUF(inbuf, MC_CMD_START_BIST_IN_LEN);
4856 
4857 	MCDI_SET_DWORD(inbuf, START_BIST_IN_TYPE, bist_type);
4858 	return efx_mcdi_rpc(efx, MC_CMD_START_BIST, inbuf, sizeof(inbuf),
4859 			    NULL, 0, NULL);
4860 }
4861 
4862 /* MC BISTs follow a different poll mechanism to phy BISTs.
4863  * The BIST is done in the poll handler on the MC, and the MCDI command
4864  * will block until the BIST is done.
4865  */
4866 static int efx_ef10_poll_bist(struct efx_nic *efx)
4867 {
4868 	int rc;
4869 	MCDI_DECLARE_BUF(outbuf, MC_CMD_POLL_BIST_OUT_LEN);
4870 	size_t outlen;
4871 	u32 result;
4872 
4873 	rc = efx_mcdi_rpc(efx, MC_CMD_POLL_BIST, NULL, 0,
4874 			   outbuf, sizeof(outbuf), &outlen);
4875 	if (rc != 0)
4876 		return rc;
4877 
4878 	if (outlen < MC_CMD_POLL_BIST_OUT_LEN)
4879 		return -EIO;
4880 
4881 	result = MCDI_DWORD(outbuf, POLL_BIST_OUT_RESULT);
4882 	switch (result) {
4883 	case MC_CMD_POLL_BIST_PASSED:
4884 		netif_dbg(efx, hw, efx->net_dev, "BIST passed.\n");
4885 		return 0;
4886 	case MC_CMD_POLL_BIST_TIMEOUT:
4887 		netif_err(efx, hw, efx->net_dev, "BIST timed out\n");
4888 		return -EIO;
4889 	case MC_CMD_POLL_BIST_FAILED:
4890 		netif_err(efx, hw, efx->net_dev, "BIST failed.\n");
4891 		return -EIO;
4892 	default:
4893 		netif_err(efx, hw, efx->net_dev,
4894 			  "BIST returned unknown result %u", result);
4895 		return -EIO;
4896 	}
4897 }
4898 
4899 static int efx_ef10_run_bist(struct efx_nic *efx, u32 bist_type)
4900 {
4901 	int rc;
4902 
4903 	netif_dbg(efx, drv, efx->net_dev, "starting BIST type %u\n", bist_type);
4904 
4905 	rc = efx_ef10_start_bist(efx, bist_type);
4906 	if (rc != 0)
4907 		return rc;
4908 
4909 	return efx_ef10_poll_bist(efx);
4910 }
4911 
4912 static int
4913 efx_ef10_test_chip(struct efx_nic *efx, struct efx_self_tests *tests)
4914 {
4915 	int rc, rc2;
4916 
4917 	efx_reset_down(efx, RESET_TYPE_WORLD);
4918 
4919 	rc = efx_mcdi_rpc(efx, MC_CMD_ENABLE_OFFLINE_BIST,
4920 			  NULL, 0, NULL, 0, NULL);
4921 	if (rc != 0)
4922 		goto out;
4923 
4924 	tests->memory = efx_ef10_run_bist(efx, MC_CMD_MC_MEM_BIST) ? -1 : 1;
4925 	tests->registers = efx_ef10_run_bist(efx, MC_CMD_REG_BIST) ? -1 : 1;
4926 
4927 	rc = efx_mcdi_reset(efx, RESET_TYPE_WORLD);
4928 
4929 out:
4930 	if (rc == -EPERM)
4931 		rc = 0;
4932 	rc2 = efx_reset_up(efx, RESET_TYPE_WORLD, rc == 0);
4933 	return rc ? rc : rc2;
4934 }
4935 
4936 #ifdef CONFIG_SFC_MTD
4937 
4938 struct efx_ef10_nvram_type_info {
4939 	u16 type, type_mask;
4940 	u8 port;
4941 	const char *name;
4942 };
4943 
4944 static const struct efx_ef10_nvram_type_info efx_ef10_nvram_types[] = {
4945 	{ NVRAM_PARTITION_TYPE_MC_FIRMWARE,	   0,    0, "sfc_mcfw" },
4946 	{ NVRAM_PARTITION_TYPE_MC_FIRMWARE_BACKUP, 0,    0, "sfc_mcfw_backup" },
4947 	{ NVRAM_PARTITION_TYPE_EXPANSION_ROM,	   0,    0, "sfc_exp_rom" },
4948 	{ NVRAM_PARTITION_TYPE_STATIC_CONFIG,	   0,    0, "sfc_static_cfg" },
4949 	{ NVRAM_PARTITION_TYPE_DYNAMIC_CONFIG,	   0,    0, "sfc_dynamic_cfg" },
4950 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT0, 0,   0, "sfc_exp_rom_cfg" },
4951 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT1, 0,   1, "sfc_exp_rom_cfg" },
4952 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT2, 0,   2, "sfc_exp_rom_cfg" },
4953 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT3, 0,   3, "sfc_exp_rom_cfg" },
4954 	{ NVRAM_PARTITION_TYPE_LICENSE,		   0,    0, "sfc_license" },
4955 	{ NVRAM_PARTITION_TYPE_PHY_MIN,		   0xff, 0, "sfc_phy_fw" },
4956 };
4957 
4958 static int efx_ef10_mtd_probe_partition(struct efx_nic *efx,
4959 					struct efx_mcdi_mtd_partition *part,
4960 					unsigned int type)
4961 {
4962 	MCDI_DECLARE_BUF(inbuf, MC_CMD_NVRAM_METADATA_IN_LEN);
4963 	MCDI_DECLARE_BUF(outbuf, MC_CMD_NVRAM_METADATA_OUT_LENMAX);
4964 	const struct efx_ef10_nvram_type_info *info;
4965 	size_t size, erase_size, outlen;
4966 	bool protected;
4967 	int rc;
4968 
4969 	for (info = efx_ef10_nvram_types; ; info++) {
4970 		if (info ==
4971 		    efx_ef10_nvram_types + ARRAY_SIZE(efx_ef10_nvram_types))
4972 			return -ENODEV;
4973 		if ((type & ~info->type_mask) == info->type)
4974 			break;
4975 	}
4976 	if (info->port != efx_port_num(efx))
4977 		return -ENODEV;
4978 
4979 	rc = efx_mcdi_nvram_info(efx, type, &size, &erase_size, &protected);
4980 	if (rc)
4981 		return rc;
4982 	if (protected)
4983 		return -ENODEV; /* hide it */
4984 
4985 	part->nvram_type = type;
4986 
4987 	MCDI_SET_DWORD(inbuf, NVRAM_METADATA_IN_TYPE, type);
4988 	rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_METADATA, inbuf, sizeof(inbuf),
4989 			  outbuf, sizeof(outbuf), &outlen);
4990 	if (rc)
4991 		return rc;
4992 	if (outlen < MC_CMD_NVRAM_METADATA_OUT_LENMIN)
4993 		return -EIO;
4994 	if (MCDI_DWORD(outbuf, NVRAM_METADATA_OUT_FLAGS) &
4995 	    (1 << MC_CMD_NVRAM_METADATA_OUT_SUBTYPE_VALID_LBN))
4996 		part->fw_subtype = MCDI_DWORD(outbuf,
4997 					      NVRAM_METADATA_OUT_SUBTYPE);
4998 
4999 	part->common.dev_type_name = "EF10 NVRAM manager";
5000 	part->common.type_name = info->name;
5001 
5002 	part->common.mtd.type = MTD_NORFLASH;
5003 	part->common.mtd.flags = MTD_CAP_NORFLASH;
5004 	part->common.mtd.size = size;
5005 	part->common.mtd.erasesize = erase_size;
5006 
5007 	return 0;
5008 }
5009 
5010 static int efx_ef10_mtd_probe(struct efx_nic *efx)
5011 {
5012 	MCDI_DECLARE_BUF(outbuf, MC_CMD_NVRAM_PARTITIONS_OUT_LENMAX);
5013 	struct efx_mcdi_mtd_partition *parts;
5014 	size_t outlen, n_parts_total, i, n_parts;
5015 	unsigned int type;
5016 	int rc;
5017 
5018 	ASSERT_RTNL();
5019 
5020 	BUILD_BUG_ON(MC_CMD_NVRAM_PARTITIONS_IN_LEN != 0);
5021 	rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_PARTITIONS, NULL, 0,
5022 			  outbuf, sizeof(outbuf), &outlen);
5023 	if (rc)
5024 		return rc;
5025 	if (outlen < MC_CMD_NVRAM_PARTITIONS_OUT_LENMIN)
5026 		return -EIO;
5027 
5028 	n_parts_total = MCDI_DWORD(outbuf, NVRAM_PARTITIONS_OUT_NUM_PARTITIONS);
5029 	if (n_parts_total >
5030 	    MCDI_VAR_ARRAY_LEN(outlen, NVRAM_PARTITIONS_OUT_TYPE_ID))
5031 		return -EIO;
5032 
5033 	parts = kcalloc(n_parts_total, sizeof(*parts), GFP_KERNEL);
5034 	if (!parts)
5035 		return -ENOMEM;
5036 
5037 	n_parts = 0;
5038 	for (i = 0; i < n_parts_total; i++) {
5039 		type = MCDI_ARRAY_DWORD(outbuf, NVRAM_PARTITIONS_OUT_TYPE_ID,
5040 					i);
5041 		rc = efx_ef10_mtd_probe_partition(efx, &parts[n_parts], type);
5042 		if (rc == 0)
5043 			n_parts++;
5044 		else if (rc != -ENODEV)
5045 			goto fail;
5046 	}
5047 
5048 	rc = efx_mtd_add(efx, &parts[0].common, n_parts, sizeof(*parts));
5049 fail:
5050 	if (rc)
5051 		kfree(parts);
5052 	return rc;
5053 }
5054 
5055 #endif /* CONFIG_SFC_MTD */
5056 
5057 static void efx_ef10_ptp_write_host_time(struct efx_nic *efx, u32 host_time)
5058 {
5059 	_efx_writed(efx, cpu_to_le32(host_time), ER_DZ_MC_DB_LWRD);
5060 }
5061 
5062 static void efx_ef10_ptp_write_host_time_vf(struct efx_nic *efx,
5063 					    u32 host_time) {}
5064 
5065 static int efx_ef10_rx_enable_timestamping(struct efx_channel *channel,
5066 					   bool temp)
5067 {
5068 	MCDI_DECLARE_BUF(inbuf, MC_CMD_PTP_IN_TIME_EVENT_SUBSCRIBE_LEN);
5069 	int rc;
5070 
5071 	if (channel->sync_events_state == SYNC_EVENTS_REQUESTED ||
5072 	    channel->sync_events_state == SYNC_EVENTS_VALID ||
5073 	    (temp && channel->sync_events_state == SYNC_EVENTS_DISABLED))
5074 		return 0;
5075 	channel->sync_events_state = SYNC_EVENTS_REQUESTED;
5076 
5077 	MCDI_SET_DWORD(inbuf, PTP_IN_OP, MC_CMD_PTP_OP_TIME_EVENT_SUBSCRIBE);
5078 	MCDI_SET_DWORD(inbuf, PTP_IN_PERIPH_ID, 0);
5079 	MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_SUBSCRIBE_QUEUE,
5080 		       channel->channel);
5081 
5082 	rc = efx_mcdi_rpc(channel->efx, MC_CMD_PTP,
5083 			  inbuf, sizeof(inbuf), NULL, 0, NULL);
5084 
5085 	if (rc != 0)
5086 		channel->sync_events_state = temp ? SYNC_EVENTS_QUIESCENT :
5087 						    SYNC_EVENTS_DISABLED;
5088 
5089 	return rc;
5090 }
5091 
5092 static int efx_ef10_rx_disable_timestamping(struct efx_channel *channel,
5093 					    bool temp)
5094 {
5095 	MCDI_DECLARE_BUF(inbuf, MC_CMD_PTP_IN_TIME_EVENT_UNSUBSCRIBE_LEN);
5096 	int rc;
5097 
5098 	if (channel->sync_events_state == SYNC_EVENTS_DISABLED ||
5099 	    (temp && channel->sync_events_state == SYNC_EVENTS_QUIESCENT))
5100 		return 0;
5101 	if (channel->sync_events_state == SYNC_EVENTS_QUIESCENT) {
5102 		channel->sync_events_state = SYNC_EVENTS_DISABLED;
5103 		return 0;
5104 	}
5105 	channel->sync_events_state = temp ? SYNC_EVENTS_QUIESCENT :
5106 					    SYNC_EVENTS_DISABLED;
5107 
5108 	MCDI_SET_DWORD(inbuf, PTP_IN_OP, MC_CMD_PTP_OP_TIME_EVENT_UNSUBSCRIBE);
5109 	MCDI_SET_DWORD(inbuf, PTP_IN_PERIPH_ID, 0);
5110 	MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_UNSUBSCRIBE_CONTROL,
5111 		       MC_CMD_PTP_IN_TIME_EVENT_UNSUBSCRIBE_SINGLE);
5112 	MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_UNSUBSCRIBE_QUEUE,
5113 		       channel->channel);
5114 
5115 	rc = efx_mcdi_rpc(channel->efx, MC_CMD_PTP,
5116 			  inbuf, sizeof(inbuf), NULL, 0, NULL);
5117 
5118 	return rc;
5119 }
5120 
5121 static int efx_ef10_ptp_set_ts_sync_events(struct efx_nic *efx, bool en,
5122 					   bool temp)
5123 {
5124 	int (*set)(struct efx_channel *channel, bool temp);
5125 	struct efx_channel *channel;
5126 
5127 	set = en ?
5128 	      efx_ef10_rx_enable_timestamping :
5129 	      efx_ef10_rx_disable_timestamping;
5130 
5131 	efx_for_each_channel(channel, efx) {
5132 		int rc = set(channel, temp);
5133 		if (en && rc != 0) {
5134 			efx_ef10_ptp_set_ts_sync_events(efx, false, temp);
5135 			return rc;
5136 		}
5137 	}
5138 
5139 	return 0;
5140 }
5141 
5142 static int efx_ef10_ptp_set_ts_config_vf(struct efx_nic *efx,
5143 					 struct hwtstamp_config *init)
5144 {
5145 	return -EOPNOTSUPP;
5146 }
5147 
5148 static int efx_ef10_ptp_set_ts_config(struct efx_nic *efx,
5149 				      struct hwtstamp_config *init)
5150 {
5151 	int rc;
5152 
5153 	switch (init->rx_filter) {
5154 	case HWTSTAMP_FILTER_NONE:
5155 		efx_ef10_ptp_set_ts_sync_events(efx, false, false);
5156 		/* if TX timestamping is still requested then leave PTP on */
5157 		return efx_ptp_change_mode(efx,
5158 					   init->tx_type != HWTSTAMP_TX_OFF, 0);
5159 	case HWTSTAMP_FILTER_ALL:
5160 	case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
5161 	case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
5162 	case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
5163 	case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
5164 	case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
5165 	case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
5166 	case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
5167 	case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
5168 	case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
5169 	case HWTSTAMP_FILTER_PTP_V2_EVENT:
5170 	case HWTSTAMP_FILTER_PTP_V2_SYNC:
5171 	case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
5172 		init->rx_filter = HWTSTAMP_FILTER_ALL;
5173 		rc = efx_ptp_change_mode(efx, true, 0);
5174 		if (!rc)
5175 			rc = efx_ef10_ptp_set_ts_sync_events(efx, true, false);
5176 		if (rc)
5177 			efx_ptp_change_mode(efx, false, 0);
5178 		return rc;
5179 	default:
5180 		return -ERANGE;
5181 	}
5182 }
5183 
5184 static int efx_ef10_vlan_rx_add_vid(struct efx_nic *efx, __be16 proto, u16 vid)
5185 {
5186 	if (proto != htons(ETH_P_8021Q))
5187 		return -EINVAL;
5188 
5189 	return efx_ef10_add_vlan(efx, vid);
5190 }
5191 
5192 static int efx_ef10_vlan_rx_kill_vid(struct efx_nic *efx, __be16 proto, u16 vid)
5193 {
5194 	if (proto != htons(ETH_P_8021Q))
5195 		return -EINVAL;
5196 
5197 	return efx_ef10_del_vlan(efx, vid);
5198 }
5199 
5200 #define EF10_OFFLOAD_FEATURES		\
5201 	(NETIF_F_IP_CSUM |		\
5202 	 NETIF_F_HW_VLAN_CTAG_FILTER |	\
5203 	 NETIF_F_IPV6_CSUM |		\
5204 	 NETIF_F_RXHASH |		\
5205 	 NETIF_F_NTUPLE)
5206 
5207 const struct efx_nic_type efx_hunt_a0_vf_nic_type = {
5208 	.is_vf = true,
5209 	.mem_bar = EFX_MEM_VF_BAR,
5210 	.mem_map_size = efx_ef10_mem_map_size,
5211 	.probe = efx_ef10_probe_vf,
5212 	.remove = efx_ef10_remove,
5213 	.dimension_resources = efx_ef10_dimension_resources,
5214 	.init = efx_ef10_init_nic,
5215 	.fini = efx_port_dummy_op_void,
5216 	.map_reset_reason = efx_ef10_map_reset_reason,
5217 	.map_reset_flags = efx_ef10_map_reset_flags,
5218 	.reset = efx_ef10_reset,
5219 	.probe_port = efx_mcdi_port_probe,
5220 	.remove_port = efx_mcdi_port_remove,
5221 	.fini_dmaq = efx_ef10_fini_dmaq,
5222 	.prepare_flr = efx_ef10_prepare_flr,
5223 	.finish_flr = efx_port_dummy_op_void,
5224 	.describe_stats = efx_ef10_describe_stats,
5225 	.update_stats = efx_ef10_update_stats_vf,
5226 	.start_stats = efx_port_dummy_op_void,
5227 	.pull_stats = efx_port_dummy_op_void,
5228 	.stop_stats = efx_port_dummy_op_void,
5229 	.set_id_led = efx_mcdi_set_id_led,
5230 	.push_irq_moderation = efx_ef10_push_irq_moderation,
5231 	.reconfigure_mac = efx_ef10_mac_reconfigure_vf,
5232 	.check_mac_fault = efx_mcdi_mac_check_fault,
5233 	.reconfigure_port = efx_mcdi_port_reconfigure,
5234 	.get_wol = efx_ef10_get_wol_vf,
5235 	.set_wol = efx_ef10_set_wol_vf,
5236 	.resume_wol = efx_port_dummy_op_void,
5237 	.mcdi_request = efx_ef10_mcdi_request,
5238 	.mcdi_poll_response = efx_ef10_mcdi_poll_response,
5239 	.mcdi_read_response = efx_ef10_mcdi_read_response,
5240 	.mcdi_poll_reboot = efx_ef10_mcdi_poll_reboot,
5241 	.mcdi_reboot_detected = efx_ef10_mcdi_reboot_detected,
5242 	.irq_enable_master = efx_port_dummy_op_void,
5243 	.irq_test_generate = efx_ef10_irq_test_generate,
5244 	.irq_disable_non_ev = efx_port_dummy_op_void,
5245 	.irq_handle_msi = efx_ef10_msi_interrupt,
5246 	.irq_handle_legacy = efx_ef10_legacy_interrupt,
5247 	.tx_probe = efx_ef10_tx_probe,
5248 	.tx_init = efx_ef10_tx_init,
5249 	.tx_remove = efx_ef10_tx_remove,
5250 	.tx_write = efx_ef10_tx_write,
5251 	.rx_push_rss_config = efx_ef10_vf_rx_push_rss_config,
5252 	.rx_probe = efx_ef10_rx_probe,
5253 	.rx_init = efx_ef10_rx_init,
5254 	.rx_remove = efx_ef10_rx_remove,
5255 	.rx_write = efx_ef10_rx_write,
5256 	.rx_defer_refill = efx_ef10_rx_defer_refill,
5257 	.ev_probe = efx_ef10_ev_probe,
5258 	.ev_init = efx_ef10_ev_init,
5259 	.ev_fini = efx_ef10_ev_fini,
5260 	.ev_remove = efx_ef10_ev_remove,
5261 	.ev_process = efx_ef10_ev_process,
5262 	.ev_read_ack = efx_ef10_ev_read_ack,
5263 	.ev_test_generate = efx_ef10_ev_test_generate,
5264 	.filter_table_probe = efx_ef10_filter_table_probe,
5265 	.filter_table_restore = efx_ef10_filter_table_restore,
5266 	.filter_table_remove = efx_ef10_filter_table_remove,
5267 	.filter_update_rx_scatter = efx_ef10_filter_update_rx_scatter,
5268 	.filter_insert = efx_ef10_filter_insert,
5269 	.filter_remove_safe = efx_ef10_filter_remove_safe,
5270 	.filter_get_safe = efx_ef10_filter_get_safe,
5271 	.filter_clear_rx = efx_ef10_filter_clear_rx,
5272 	.filter_count_rx_used = efx_ef10_filter_count_rx_used,
5273 	.filter_get_rx_id_limit = efx_ef10_filter_get_rx_id_limit,
5274 	.filter_get_rx_ids = efx_ef10_filter_get_rx_ids,
5275 #ifdef CONFIG_RFS_ACCEL
5276 	.filter_rfs_insert = efx_ef10_filter_rfs_insert,
5277 	.filter_rfs_expire_one = efx_ef10_filter_rfs_expire_one,
5278 #endif
5279 #ifdef CONFIG_SFC_MTD
5280 	.mtd_probe = efx_port_dummy_op_int,
5281 #endif
5282 	.ptp_write_host_time = efx_ef10_ptp_write_host_time_vf,
5283 	.ptp_set_ts_config = efx_ef10_ptp_set_ts_config_vf,
5284 	.vlan_rx_add_vid = efx_ef10_vlan_rx_add_vid,
5285 	.vlan_rx_kill_vid = efx_ef10_vlan_rx_kill_vid,
5286 #ifdef CONFIG_SFC_SRIOV
5287 	.vswitching_probe = efx_ef10_vswitching_probe_vf,
5288 	.vswitching_restore = efx_ef10_vswitching_restore_vf,
5289 	.vswitching_remove = efx_ef10_vswitching_remove_vf,
5290 	.sriov_get_phys_port_id = efx_ef10_sriov_get_phys_port_id,
5291 #endif
5292 	.get_mac_address = efx_ef10_get_mac_address_vf,
5293 	.set_mac_address = efx_ef10_set_mac_address,
5294 
5295 	.revision = EFX_REV_HUNT_A0,
5296 	.max_dma_mask = DMA_BIT_MASK(ESF_DZ_TX_KER_BUF_ADDR_WIDTH),
5297 	.rx_prefix_size = ES_DZ_RX_PREFIX_SIZE,
5298 	.rx_hash_offset = ES_DZ_RX_PREFIX_HASH_OFST,
5299 	.rx_ts_offset = ES_DZ_RX_PREFIX_TSTAMP_OFST,
5300 	.can_rx_scatter = true,
5301 	.always_rx_scatter = true,
5302 	.max_interrupt_mode = EFX_INT_MODE_MSIX,
5303 	.timer_period_max = 1 << ERF_DD_EVQ_IND_TIMER_VAL_WIDTH,
5304 	.offload_features = EF10_OFFLOAD_FEATURES,
5305 	.mcdi_max_ver = 2,
5306 	.max_rx_ip_filters = HUNT_FILTER_TBL_ROWS,
5307 	.hwtstamp_filters = 1 << HWTSTAMP_FILTER_NONE |
5308 			    1 << HWTSTAMP_FILTER_ALL,
5309 };
5310 
5311 const struct efx_nic_type efx_hunt_a0_nic_type = {
5312 	.is_vf = false,
5313 	.mem_bar = EFX_MEM_BAR,
5314 	.mem_map_size = efx_ef10_mem_map_size,
5315 	.probe = efx_ef10_probe_pf,
5316 	.remove = efx_ef10_remove,
5317 	.dimension_resources = efx_ef10_dimension_resources,
5318 	.init = efx_ef10_init_nic,
5319 	.fini = efx_port_dummy_op_void,
5320 	.map_reset_reason = efx_ef10_map_reset_reason,
5321 	.map_reset_flags = efx_ef10_map_reset_flags,
5322 	.reset = efx_ef10_reset,
5323 	.probe_port = efx_mcdi_port_probe,
5324 	.remove_port = efx_mcdi_port_remove,
5325 	.fini_dmaq = efx_ef10_fini_dmaq,
5326 	.prepare_flr = efx_ef10_prepare_flr,
5327 	.finish_flr = efx_port_dummy_op_void,
5328 	.describe_stats = efx_ef10_describe_stats,
5329 	.update_stats = efx_ef10_update_stats_pf,
5330 	.start_stats = efx_mcdi_mac_start_stats,
5331 	.pull_stats = efx_mcdi_mac_pull_stats,
5332 	.stop_stats = efx_mcdi_mac_stop_stats,
5333 	.set_id_led = efx_mcdi_set_id_led,
5334 	.push_irq_moderation = efx_ef10_push_irq_moderation,
5335 	.reconfigure_mac = efx_ef10_mac_reconfigure,
5336 	.check_mac_fault = efx_mcdi_mac_check_fault,
5337 	.reconfigure_port = efx_mcdi_port_reconfigure,
5338 	.get_wol = efx_ef10_get_wol,
5339 	.set_wol = efx_ef10_set_wol,
5340 	.resume_wol = efx_port_dummy_op_void,
5341 	.test_chip = efx_ef10_test_chip,
5342 	.test_nvram = efx_mcdi_nvram_test_all,
5343 	.mcdi_request = efx_ef10_mcdi_request,
5344 	.mcdi_poll_response = efx_ef10_mcdi_poll_response,
5345 	.mcdi_read_response = efx_ef10_mcdi_read_response,
5346 	.mcdi_poll_reboot = efx_ef10_mcdi_poll_reboot,
5347 	.mcdi_reboot_detected = efx_ef10_mcdi_reboot_detected,
5348 	.irq_enable_master = efx_port_dummy_op_void,
5349 	.irq_test_generate = efx_ef10_irq_test_generate,
5350 	.irq_disable_non_ev = efx_port_dummy_op_void,
5351 	.irq_handle_msi = efx_ef10_msi_interrupt,
5352 	.irq_handle_legacy = efx_ef10_legacy_interrupt,
5353 	.tx_probe = efx_ef10_tx_probe,
5354 	.tx_init = efx_ef10_tx_init,
5355 	.tx_remove = efx_ef10_tx_remove,
5356 	.tx_write = efx_ef10_tx_write,
5357 	.rx_push_rss_config = efx_ef10_pf_rx_push_rss_config,
5358 	.rx_probe = efx_ef10_rx_probe,
5359 	.rx_init = efx_ef10_rx_init,
5360 	.rx_remove = efx_ef10_rx_remove,
5361 	.rx_write = efx_ef10_rx_write,
5362 	.rx_defer_refill = efx_ef10_rx_defer_refill,
5363 	.ev_probe = efx_ef10_ev_probe,
5364 	.ev_init = efx_ef10_ev_init,
5365 	.ev_fini = efx_ef10_ev_fini,
5366 	.ev_remove = efx_ef10_ev_remove,
5367 	.ev_process = efx_ef10_ev_process,
5368 	.ev_read_ack = efx_ef10_ev_read_ack,
5369 	.ev_test_generate = efx_ef10_ev_test_generate,
5370 	.filter_table_probe = efx_ef10_filter_table_probe,
5371 	.filter_table_restore = efx_ef10_filter_table_restore,
5372 	.filter_table_remove = efx_ef10_filter_table_remove,
5373 	.filter_update_rx_scatter = efx_ef10_filter_update_rx_scatter,
5374 	.filter_insert = efx_ef10_filter_insert,
5375 	.filter_remove_safe = efx_ef10_filter_remove_safe,
5376 	.filter_get_safe = efx_ef10_filter_get_safe,
5377 	.filter_clear_rx = efx_ef10_filter_clear_rx,
5378 	.filter_count_rx_used = efx_ef10_filter_count_rx_used,
5379 	.filter_get_rx_id_limit = efx_ef10_filter_get_rx_id_limit,
5380 	.filter_get_rx_ids = efx_ef10_filter_get_rx_ids,
5381 #ifdef CONFIG_RFS_ACCEL
5382 	.filter_rfs_insert = efx_ef10_filter_rfs_insert,
5383 	.filter_rfs_expire_one = efx_ef10_filter_rfs_expire_one,
5384 #endif
5385 #ifdef CONFIG_SFC_MTD
5386 	.mtd_probe = efx_ef10_mtd_probe,
5387 	.mtd_rename = efx_mcdi_mtd_rename,
5388 	.mtd_read = efx_mcdi_mtd_read,
5389 	.mtd_erase = efx_mcdi_mtd_erase,
5390 	.mtd_write = efx_mcdi_mtd_write,
5391 	.mtd_sync = efx_mcdi_mtd_sync,
5392 #endif
5393 	.ptp_write_host_time = efx_ef10_ptp_write_host_time,
5394 	.ptp_set_ts_sync_events = efx_ef10_ptp_set_ts_sync_events,
5395 	.ptp_set_ts_config = efx_ef10_ptp_set_ts_config,
5396 	.vlan_rx_add_vid = efx_ef10_vlan_rx_add_vid,
5397 	.vlan_rx_kill_vid = efx_ef10_vlan_rx_kill_vid,
5398 #ifdef CONFIG_SFC_SRIOV
5399 	.sriov_configure = efx_ef10_sriov_configure,
5400 	.sriov_init = efx_ef10_sriov_init,
5401 	.sriov_fini = efx_ef10_sriov_fini,
5402 	.sriov_wanted = efx_ef10_sriov_wanted,
5403 	.sriov_reset = efx_ef10_sriov_reset,
5404 	.sriov_flr = efx_ef10_sriov_flr,
5405 	.sriov_set_vf_mac = efx_ef10_sriov_set_vf_mac,
5406 	.sriov_set_vf_vlan = efx_ef10_sriov_set_vf_vlan,
5407 	.sriov_set_vf_spoofchk = efx_ef10_sriov_set_vf_spoofchk,
5408 	.sriov_get_vf_config = efx_ef10_sriov_get_vf_config,
5409 	.sriov_set_vf_link_state = efx_ef10_sriov_set_vf_link_state,
5410 	.vswitching_probe = efx_ef10_vswitching_probe_pf,
5411 	.vswitching_restore = efx_ef10_vswitching_restore_pf,
5412 	.vswitching_remove = efx_ef10_vswitching_remove_pf,
5413 #endif
5414 	.get_mac_address = efx_ef10_get_mac_address_pf,
5415 	.set_mac_address = efx_ef10_set_mac_address,
5416 
5417 	.revision = EFX_REV_HUNT_A0,
5418 	.max_dma_mask = DMA_BIT_MASK(ESF_DZ_TX_KER_BUF_ADDR_WIDTH),
5419 	.rx_prefix_size = ES_DZ_RX_PREFIX_SIZE,
5420 	.rx_hash_offset = ES_DZ_RX_PREFIX_HASH_OFST,
5421 	.rx_ts_offset = ES_DZ_RX_PREFIX_TSTAMP_OFST,
5422 	.can_rx_scatter = true,
5423 	.always_rx_scatter = true,
5424 	.max_interrupt_mode = EFX_INT_MODE_MSIX,
5425 	.timer_period_max = 1 << ERF_DD_EVQ_IND_TIMER_VAL_WIDTH,
5426 	.offload_features = EF10_OFFLOAD_FEATURES,
5427 	.mcdi_max_ver = 2,
5428 	.max_rx_ip_filters = HUNT_FILTER_TBL_ROWS,
5429 	.hwtstamp_filters = 1 << HWTSTAMP_FILTER_NONE |
5430 			    1 << HWTSTAMP_FILTER_ALL,
5431 };
5432