xref: /linux/drivers/net/ethernet/sfc/ef10.c (revision 2697b79a469b68e3ad3640f55284359c1396278d)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /****************************************************************************
3  * Driver for Solarflare network controllers and boards
4  * Copyright 2012-2013 Solarflare Communications Inc.
5  */
6 
7 #include "net_driver.h"
8 #include "rx_common.h"
9 #include "tx_common.h"
10 #include "ef10_regs.h"
11 #include "io.h"
12 #include "mcdi.h"
13 #include "mcdi_pcol.h"
14 #include "mcdi_port.h"
15 #include "mcdi_port_common.h"
16 #include "mcdi_functions.h"
17 #include "nic.h"
18 #include "mcdi_filters.h"
19 #include "workarounds.h"
20 #include "selftest.h"
21 #include "ef10_sriov.h"
22 #include <linux/in.h>
23 #include <linux/jhash.h>
24 #include <linux/wait.h>
25 #include <linux/workqueue.h>
26 #include <net/udp_tunnel.h>
27 
28 /* Hardware control for EF10 architecture including 'Huntington'. */
29 
30 #define EFX_EF10_DRVGEN_EV		7
31 enum {
32 	EFX_EF10_TEST = 1,
33 	EFX_EF10_REFILL,
34 };
35 
36 /* VLAN list entry */
37 struct efx_ef10_vlan {
38 	struct list_head list;
39 	u16 vid;
40 };
41 
42 static int efx_ef10_set_udp_tnl_ports(struct efx_nic *efx, bool unloading);
43 static const struct udp_tunnel_nic_info efx_ef10_udp_tunnels;
44 
45 static int efx_ef10_get_warm_boot_count(struct efx_nic *efx)
46 {
47 	efx_dword_t reg;
48 
49 	efx_readd(efx, &reg, ER_DZ_BIU_MC_SFT_STATUS);
50 	return EFX_DWORD_FIELD(reg, EFX_WORD_1) == 0xb007 ?
51 		EFX_DWORD_FIELD(reg, EFX_WORD_0) : -EIO;
52 }
53 
54 /* On all EF10s up to and including SFC9220 (Medford1), all PFs use BAR 0 for
55  * I/O space and BAR 2(&3) for memory.  On SFC9250 (Medford2), there is no I/O
56  * bar; PFs use BAR 0/1 for memory.
57  */
58 static unsigned int efx_ef10_pf_mem_bar(struct efx_nic *efx)
59 {
60 	switch (efx->pci_dev->device) {
61 	case 0x0b03: /* SFC9250 PF */
62 		return 0;
63 	default:
64 		return 2;
65 	}
66 }
67 
68 /* All VFs use BAR 0/1 for memory */
69 static unsigned int efx_ef10_vf_mem_bar(struct efx_nic *efx)
70 {
71 	return 0;
72 }
73 
74 static unsigned int efx_ef10_mem_map_size(struct efx_nic *efx)
75 {
76 	int bar;
77 
78 	bar = efx->type->mem_bar(efx);
79 	return resource_size(&efx->pci_dev->resource[bar]);
80 }
81 
82 static bool efx_ef10_is_vf(struct efx_nic *efx)
83 {
84 	return efx->type->is_vf;
85 }
86 
87 #ifdef CONFIG_SFC_SRIOV
88 static int efx_ef10_get_vf_index(struct efx_nic *efx)
89 {
90 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_FUNCTION_INFO_OUT_LEN);
91 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
92 	size_t outlen;
93 	int rc;
94 
95 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_FUNCTION_INFO, NULL, 0, outbuf,
96 			  sizeof(outbuf), &outlen);
97 	if (rc)
98 		return rc;
99 	if (outlen < sizeof(outbuf))
100 		return -EIO;
101 
102 	nic_data->vf_index = MCDI_DWORD(outbuf, GET_FUNCTION_INFO_OUT_VF);
103 	return 0;
104 }
105 #endif
106 
107 static int efx_ef10_init_datapath_caps(struct efx_nic *efx)
108 {
109 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_CAPABILITIES_V4_OUT_LEN);
110 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
111 	size_t outlen;
112 	int rc;
113 
114 	BUILD_BUG_ON(MC_CMD_GET_CAPABILITIES_IN_LEN != 0);
115 
116 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_CAPABILITIES, NULL, 0,
117 			  outbuf, sizeof(outbuf), &outlen);
118 	if (rc)
119 		return rc;
120 	if (outlen < MC_CMD_GET_CAPABILITIES_OUT_LEN) {
121 		netif_err(efx, drv, efx->net_dev,
122 			  "unable to read datapath firmware capabilities\n");
123 		return -EIO;
124 	}
125 
126 	nic_data->datapath_caps =
127 		MCDI_DWORD(outbuf, GET_CAPABILITIES_OUT_FLAGS1);
128 
129 	if (outlen >= MC_CMD_GET_CAPABILITIES_V2_OUT_LEN) {
130 		nic_data->datapath_caps2 = MCDI_DWORD(outbuf,
131 				GET_CAPABILITIES_V2_OUT_FLAGS2);
132 		nic_data->piobuf_size = MCDI_WORD(outbuf,
133 				GET_CAPABILITIES_V2_OUT_SIZE_PIO_BUFF);
134 	} else {
135 		nic_data->datapath_caps2 = 0;
136 		nic_data->piobuf_size = ER_DZ_TX_PIOBUF_SIZE;
137 	}
138 
139 	/* record the DPCPU firmware IDs to determine VEB vswitching support.
140 	 */
141 	nic_data->rx_dpcpu_fw_id =
142 		MCDI_WORD(outbuf, GET_CAPABILITIES_OUT_RX_DPCPU_FW_ID);
143 	nic_data->tx_dpcpu_fw_id =
144 		MCDI_WORD(outbuf, GET_CAPABILITIES_OUT_TX_DPCPU_FW_ID);
145 
146 	if (!(nic_data->datapath_caps &
147 	      (1 << MC_CMD_GET_CAPABILITIES_OUT_RX_PREFIX_LEN_14_LBN))) {
148 		netif_err(efx, probe, efx->net_dev,
149 			  "current firmware does not support an RX prefix\n");
150 		return -ENODEV;
151 	}
152 
153 	if (outlen >= MC_CMD_GET_CAPABILITIES_V3_OUT_LEN) {
154 		u8 vi_window_mode = MCDI_BYTE(outbuf,
155 				GET_CAPABILITIES_V3_OUT_VI_WINDOW_MODE);
156 
157 		rc = efx_mcdi_window_mode_to_stride(efx, vi_window_mode);
158 		if (rc)
159 			return rc;
160 	} else {
161 		/* keep default VI stride */
162 		netif_dbg(efx, probe, efx->net_dev,
163 			  "firmware did not report VI window mode, assuming vi_stride = %u\n",
164 			  efx->vi_stride);
165 	}
166 
167 	if (outlen >= MC_CMD_GET_CAPABILITIES_V4_OUT_LEN) {
168 		efx->num_mac_stats = MCDI_WORD(outbuf,
169 				GET_CAPABILITIES_V4_OUT_MAC_STATS_NUM_STATS);
170 		netif_dbg(efx, probe, efx->net_dev,
171 			  "firmware reports num_mac_stats = %u\n",
172 			  efx->num_mac_stats);
173 	} else {
174 		/* leave num_mac_stats as the default value, MC_CMD_MAC_NSTATS */
175 		netif_dbg(efx, probe, efx->net_dev,
176 			  "firmware did not report num_mac_stats, assuming %u\n",
177 			  efx->num_mac_stats);
178 	}
179 
180 	return 0;
181 }
182 
183 static void efx_ef10_read_licensed_features(struct efx_nic *efx)
184 {
185 	MCDI_DECLARE_BUF(inbuf, MC_CMD_LICENSING_V3_IN_LEN);
186 	MCDI_DECLARE_BUF(outbuf, MC_CMD_LICENSING_V3_OUT_LEN);
187 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
188 	size_t outlen;
189 	int rc;
190 
191 	MCDI_SET_DWORD(inbuf, LICENSING_V3_IN_OP,
192 		       MC_CMD_LICENSING_V3_IN_OP_REPORT_LICENSE);
193 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_LICENSING_V3, inbuf, sizeof(inbuf),
194 				outbuf, sizeof(outbuf), &outlen);
195 	if (rc || (outlen < MC_CMD_LICENSING_V3_OUT_LEN))
196 		return;
197 
198 	nic_data->licensed_features = MCDI_QWORD(outbuf,
199 					 LICENSING_V3_OUT_LICENSED_FEATURES);
200 }
201 
202 static int efx_ef10_get_sysclk_freq(struct efx_nic *efx)
203 {
204 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_CLOCK_OUT_LEN);
205 	int rc;
206 
207 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_CLOCK, NULL, 0,
208 			  outbuf, sizeof(outbuf), NULL);
209 	if (rc)
210 		return rc;
211 	rc = MCDI_DWORD(outbuf, GET_CLOCK_OUT_SYS_FREQ);
212 	return rc > 0 ? rc : -ERANGE;
213 }
214 
215 static int efx_ef10_get_timer_workarounds(struct efx_nic *efx)
216 {
217 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
218 	unsigned int implemented;
219 	unsigned int enabled;
220 	int rc;
221 
222 	nic_data->workaround_35388 = false;
223 	nic_data->workaround_61265 = false;
224 
225 	rc = efx_mcdi_get_workarounds(efx, &implemented, &enabled);
226 
227 	if (rc == -ENOSYS) {
228 		/* Firmware without GET_WORKAROUNDS - not a problem. */
229 		rc = 0;
230 	} else if (rc == 0) {
231 		/* Bug61265 workaround is always enabled if implemented. */
232 		if (enabled & MC_CMD_GET_WORKAROUNDS_OUT_BUG61265)
233 			nic_data->workaround_61265 = true;
234 
235 		if (enabled & MC_CMD_GET_WORKAROUNDS_OUT_BUG35388) {
236 			nic_data->workaround_35388 = true;
237 		} else if (implemented & MC_CMD_GET_WORKAROUNDS_OUT_BUG35388) {
238 			/* Workaround is implemented but not enabled.
239 			 * Try to enable it.
240 			 */
241 			rc = efx_mcdi_set_workaround(efx,
242 						     MC_CMD_WORKAROUND_BUG35388,
243 						     true, NULL);
244 			if (rc == 0)
245 				nic_data->workaround_35388 = true;
246 			/* If we failed to set the workaround just carry on. */
247 			rc = 0;
248 		}
249 	}
250 
251 	netif_dbg(efx, probe, efx->net_dev,
252 		  "workaround for bug 35388 is %sabled\n",
253 		  nic_data->workaround_35388 ? "en" : "dis");
254 	netif_dbg(efx, probe, efx->net_dev,
255 		  "workaround for bug 61265 is %sabled\n",
256 		  nic_data->workaround_61265 ? "en" : "dis");
257 
258 	return rc;
259 }
260 
261 static void efx_ef10_process_timer_config(struct efx_nic *efx,
262 					  const efx_dword_t *data)
263 {
264 	unsigned int max_count;
265 
266 	if (EFX_EF10_WORKAROUND_61265(efx)) {
267 		efx->timer_quantum_ns = MCDI_DWORD(data,
268 			GET_EVQ_TMR_PROPERTIES_OUT_MCDI_TMR_STEP_NS);
269 		efx->timer_max_ns = MCDI_DWORD(data,
270 			GET_EVQ_TMR_PROPERTIES_OUT_MCDI_TMR_MAX_NS);
271 	} else if (EFX_EF10_WORKAROUND_35388(efx)) {
272 		efx->timer_quantum_ns = MCDI_DWORD(data,
273 			GET_EVQ_TMR_PROPERTIES_OUT_BUG35388_TMR_NS_PER_COUNT);
274 		max_count = MCDI_DWORD(data,
275 			GET_EVQ_TMR_PROPERTIES_OUT_BUG35388_TMR_MAX_COUNT);
276 		efx->timer_max_ns = max_count * efx->timer_quantum_ns;
277 	} else {
278 		efx->timer_quantum_ns = MCDI_DWORD(data,
279 			GET_EVQ_TMR_PROPERTIES_OUT_TMR_REG_NS_PER_COUNT);
280 		max_count = MCDI_DWORD(data,
281 			GET_EVQ_TMR_PROPERTIES_OUT_TMR_REG_MAX_COUNT);
282 		efx->timer_max_ns = max_count * efx->timer_quantum_ns;
283 	}
284 
285 	netif_dbg(efx, probe, efx->net_dev,
286 		  "got timer properties from MC: quantum %u ns; max %u ns\n",
287 		  efx->timer_quantum_ns, efx->timer_max_ns);
288 }
289 
290 static int efx_ef10_get_timer_config(struct efx_nic *efx)
291 {
292 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_EVQ_TMR_PROPERTIES_OUT_LEN);
293 	int rc;
294 
295 	rc = efx_ef10_get_timer_workarounds(efx);
296 	if (rc)
297 		return rc;
298 
299 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_GET_EVQ_TMR_PROPERTIES, NULL, 0,
300 				outbuf, sizeof(outbuf), NULL);
301 
302 	if (rc == 0) {
303 		efx_ef10_process_timer_config(efx, outbuf);
304 	} else if (rc == -ENOSYS || rc == -EPERM) {
305 		/* Not available - fall back to Huntington defaults. */
306 		unsigned int quantum;
307 
308 		rc = efx_ef10_get_sysclk_freq(efx);
309 		if (rc < 0)
310 			return rc;
311 
312 		quantum = 1536000 / rc; /* 1536 cycles */
313 		efx->timer_quantum_ns = quantum;
314 		efx->timer_max_ns = efx->type->timer_period_max * quantum;
315 		rc = 0;
316 	} else {
317 		efx_mcdi_display_error(efx, MC_CMD_GET_EVQ_TMR_PROPERTIES,
318 				       MC_CMD_GET_EVQ_TMR_PROPERTIES_OUT_LEN,
319 				       NULL, 0, rc);
320 	}
321 
322 	return rc;
323 }
324 
325 static int efx_ef10_get_mac_address_pf(struct efx_nic *efx, u8 *mac_address)
326 {
327 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_MAC_ADDRESSES_OUT_LEN);
328 	size_t outlen;
329 	int rc;
330 
331 	BUILD_BUG_ON(MC_CMD_GET_MAC_ADDRESSES_IN_LEN != 0);
332 
333 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_MAC_ADDRESSES, NULL, 0,
334 			  outbuf, sizeof(outbuf), &outlen);
335 	if (rc)
336 		return rc;
337 	if (outlen < MC_CMD_GET_MAC_ADDRESSES_OUT_LEN)
338 		return -EIO;
339 
340 	ether_addr_copy(mac_address,
341 			MCDI_PTR(outbuf, GET_MAC_ADDRESSES_OUT_MAC_ADDR_BASE));
342 	return 0;
343 }
344 
345 static int efx_ef10_get_mac_address_vf(struct efx_nic *efx, u8 *mac_address)
346 {
347 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VPORT_GET_MAC_ADDRESSES_IN_LEN);
348 	MCDI_DECLARE_BUF(outbuf, MC_CMD_VPORT_GET_MAC_ADDRESSES_OUT_LENMAX);
349 	size_t outlen;
350 	int num_addrs, rc;
351 
352 	MCDI_SET_DWORD(inbuf, VPORT_GET_MAC_ADDRESSES_IN_VPORT_ID,
353 		       EVB_PORT_ID_ASSIGNED);
354 	rc = efx_mcdi_rpc(efx, MC_CMD_VPORT_GET_MAC_ADDRESSES, inbuf,
355 			  sizeof(inbuf), outbuf, sizeof(outbuf), &outlen);
356 
357 	if (rc)
358 		return rc;
359 	if (outlen < MC_CMD_VPORT_GET_MAC_ADDRESSES_OUT_LENMIN)
360 		return -EIO;
361 
362 	num_addrs = MCDI_DWORD(outbuf,
363 			       VPORT_GET_MAC_ADDRESSES_OUT_MACADDR_COUNT);
364 
365 	WARN_ON(num_addrs != 1);
366 
367 	ether_addr_copy(mac_address,
368 			MCDI_PTR(outbuf, VPORT_GET_MAC_ADDRESSES_OUT_MACADDR));
369 
370 	return 0;
371 }
372 
373 static ssize_t link_control_flag_show(struct device *dev,
374 				      struct device_attribute *attr,
375 				      char *buf)
376 {
377 	struct efx_nic *efx = dev_get_drvdata(dev);
378 
379 	return sprintf(buf, "%d\n",
380 		       ((efx->mcdi->fn_flags) &
381 			(1 << MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_LINKCTRL))
382 		       ? 1 : 0);
383 }
384 
385 static ssize_t primary_flag_show(struct device *dev,
386 				 struct device_attribute *attr,
387 				 char *buf)
388 {
389 	struct efx_nic *efx = dev_get_drvdata(dev);
390 
391 	return sprintf(buf, "%d\n",
392 		       ((efx->mcdi->fn_flags) &
393 			(1 << MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_PRIMARY))
394 		       ? 1 : 0);
395 }
396 
397 static struct efx_ef10_vlan *efx_ef10_find_vlan(struct efx_nic *efx, u16 vid)
398 {
399 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
400 	struct efx_ef10_vlan *vlan;
401 
402 	WARN_ON(!mutex_is_locked(&nic_data->vlan_lock));
403 
404 	list_for_each_entry(vlan, &nic_data->vlan_list, list) {
405 		if (vlan->vid == vid)
406 			return vlan;
407 	}
408 
409 	return NULL;
410 }
411 
412 static int efx_ef10_add_vlan(struct efx_nic *efx, u16 vid)
413 {
414 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
415 	struct efx_ef10_vlan *vlan;
416 	int rc;
417 
418 	mutex_lock(&nic_data->vlan_lock);
419 
420 	vlan = efx_ef10_find_vlan(efx, vid);
421 	if (vlan) {
422 		/* We add VID 0 on init. 8021q adds it on module init
423 		 * for all interfaces with VLAN filtring feature.
424 		 */
425 		if (vid == 0)
426 			goto done_unlock;
427 		netif_warn(efx, drv, efx->net_dev,
428 			   "VLAN %u already added\n", vid);
429 		rc = -EALREADY;
430 		goto fail_exist;
431 	}
432 
433 	rc = -ENOMEM;
434 	vlan = kzalloc(sizeof(*vlan), GFP_KERNEL);
435 	if (!vlan)
436 		goto fail_alloc;
437 
438 	vlan->vid = vid;
439 
440 	list_add_tail(&vlan->list, &nic_data->vlan_list);
441 
442 	if (efx->filter_state) {
443 		mutex_lock(&efx->mac_lock);
444 		down_write(&efx->filter_sem);
445 		rc = efx_mcdi_filter_add_vlan(efx, vlan->vid);
446 		up_write(&efx->filter_sem);
447 		mutex_unlock(&efx->mac_lock);
448 		if (rc)
449 			goto fail_filter_add_vlan;
450 	}
451 
452 done_unlock:
453 	mutex_unlock(&nic_data->vlan_lock);
454 	return 0;
455 
456 fail_filter_add_vlan:
457 	list_del(&vlan->list);
458 	kfree(vlan);
459 fail_alloc:
460 fail_exist:
461 	mutex_unlock(&nic_data->vlan_lock);
462 	return rc;
463 }
464 
465 static void efx_ef10_del_vlan_internal(struct efx_nic *efx,
466 				       struct efx_ef10_vlan *vlan)
467 {
468 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
469 
470 	WARN_ON(!mutex_is_locked(&nic_data->vlan_lock));
471 
472 	if (efx->filter_state) {
473 		down_write(&efx->filter_sem);
474 		efx_mcdi_filter_del_vlan(efx, vlan->vid);
475 		up_write(&efx->filter_sem);
476 	}
477 
478 	list_del(&vlan->list);
479 	kfree(vlan);
480 }
481 
482 static int efx_ef10_del_vlan(struct efx_nic *efx, u16 vid)
483 {
484 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
485 	struct efx_ef10_vlan *vlan;
486 	int rc = 0;
487 
488 	/* 8021q removes VID 0 on module unload for all interfaces
489 	 * with VLAN filtering feature. We need to keep it to receive
490 	 * untagged traffic.
491 	 */
492 	if (vid == 0)
493 		return 0;
494 
495 	mutex_lock(&nic_data->vlan_lock);
496 
497 	vlan = efx_ef10_find_vlan(efx, vid);
498 	if (!vlan) {
499 		netif_err(efx, drv, efx->net_dev,
500 			  "VLAN %u to be deleted not found\n", vid);
501 		rc = -ENOENT;
502 	} else {
503 		efx_ef10_del_vlan_internal(efx, vlan);
504 	}
505 
506 	mutex_unlock(&nic_data->vlan_lock);
507 
508 	return rc;
509 }
510 
511 static void efx_ef10_cleanup_vlans(struct efx_nic *efx)
512 {
513 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
514 	struct efx_ef10_vlan *vlan, *next_vlan;
515 
516 	mutex_lock(&nic_data->vlan_lock);
517 	list_for_each_entry_safe(vlan, next_vlan, &nic_data->vlan_list, list)
518 		efx_ef10_del_vlan_internal(efx, vlan);
519 	mutex_unlock(&nic_data->vlan_lock);
520 }
521 
522 static DEVICE_ATTR_RO(link_control_flag);
523 static DEVICE_ATTR_RO(primary_flag);
524 
525 static int efx_ef10_probe(struct efx_nic *efx)
526 {
527 	struct efx_ef10_nic_data *nic_data;
528 	int i, rc;
529 
530 	nic_data = kzalloc(sizeof(*nic_data), GFP_KERNEL);
531 	if (!nic_data)
532 		return -ENOMEM;
533 	efx->nic_data = nic_data;
534 
535 	/* we assume later that we can copy from this buffer in dwords */
536 	BUILD_BUG_ON(MCDI_CTL_SDU_LEN_MAX_V2 % 4);
537 
538 	rc = efx_nic_alloc_buffer(efx, &nic_data->mcdi_buf,
539 				  8 + MCDI_CTL_SDU_LEN_MAX_V2, GFP_KERNEL);
540 	if (rc)
541 		goto fail1;
542 
543 	/* Get the MC's warm boot count.  In case it's rebooting right
544 	 * now, be prepared to retry.
545 	 */
546 	i = 0;
547 	for (;;) {
548 		rc = efx_ef10_get_warm_boot_count(efx);
549 		if (rc >= 0)
550 			break;
551 		if (++i == 5)
552 			goto fail2;
553 		ssleep(1);
554 	}
555 	nic_data->warm_boot_count = rc;
556 
557 	/* In case we're recovering from a crash (kexec), we want to
558 	 * cancel any outstanding request by the previous user of this
559 	 * function.  We send a special message using the least
560 	 * significant bits of the 'high' (doorbell) register.
561 	 */
562 	_efx_writed(efx, cpu_to_le32(1), ER_DZ_MC_DB_HWRD);
563 
564 	rc = efx_mcdi_init(efx);
565 	if (rc)
566 		goto fail2;
567 
568 	mutex_init(&nic_data->udp_tunnels_lock);
569 	for (i = 0; i < ARRAY_SIZE(nic_data->udp_tunnels); ++i)
570 		nic_data->udp_tunnels[i].type =
571 			TUNNEL_ENCAP_UDP_PORT_ENTRY_INVALID;
572 
573 	/* Reset (most) configuration for this function */
574 	rc = efx_mcdi_reset(efx, RESET_TYPE_ALL);
575 	if (rc)
576 		goto fail3;
577 
578 	/* Enable event logging */
579 	rc = efx_mcdi_log_ctrl(efx, true, false, 0);
580 	if (rc)
581 		goto fail3;
582 
583 	rc = device_create_file(&efx->pci_dev->dev,
584 				&dev_attr_link_control_flag);
585 	if (rc)
586 		goto fail3;
587 
588 	rc = device_create_file(&efx->pci_dev->dev, &dev_attr_primary_flag);
589 	if (rc)
590 		goto fail4;
591 
592 	rc = efx_get_pf_index(efx, &nic_data->pf_index);
593 	if (rc)
594 		goto fail5;
595 
596 	rc = efx_ef10_init_datapath_caps(efx);
597 	if (rc < 0)
598 		goto fail5;
599 
600 	efx_ef10_read_licensed_features(efx);
601 
602 	/* We can have one VI for each vi_stride-byte region.
603 	 * However, until we use TX option descriptors we need up to four
604 	 * TX queues per channel for different checksumming combinations.
605 	 */
606 	if (nic_data->datapath_caps &
607 	    (1 << MC_CMD_GET_CAPABILITIES_OUT_VXLAN_NVGRE_LBN))
608 		efx->tx_queues_per_channel = 4;
609 	else
610 		efx->tx_queues_per_channel = 2;
611 	efx->max_vis = efx_ef10_mem_map_size(efx) / efx->vi_stride;
612 	if (!efx->max_vis) {
613 		netif_err(efx, drv, efx->net_dev, "error determining max VIs\n");
614 		rc = -EIO;
615 		goto fail5;
616 	}
617 	efx->max_channels = min_t(unsigned int, EFX_MAX_CHANNELS,
618 				  efx->max_vis / efx->tx_queues_per_channel);
619 	efx->max_tx_channels = efx->max_channels;
620 	if (WARN_ON(efx->max_channels == 0)) {
621 		rc = -EIO;
622 		goto fail5;
623 	}
624 
625 	efx->rx_packet_len_offset =
626 		ES_DZ_RX_PREFIX_PKTLEN_OFST - ES_DZ_RX_PREFIX_SIZE;
627 
628 	if (nic_data->datapath_caps &
629 	    (1 << MC_CMD_GET_CAPABILITIES_OUT_RX_INCLUDE_FCS_LBN))
630 		efx->net_dev->hw_features |= NETIF_F_RXFCS;
631 
632 	rc = efx_mcdi_port_get_number(efx);
633 	if (rc < 0)
634 		goto fail5;
635 	efx->port_num = rc;
636 
637 	rc = efx->type->get_mac_address(efx, efx->net_dev->perm_addr);
638 	if (rc)
639 		goto fail5;
640 
641 	rc = efx_ef10_get_timer_config(efx);
642 	if (rc < 0)
643 		goto fail5;
644 
645 	rc = efx_mcdi_mon_probe(efx);
646 	if (rc && rc != -EPERM)
647 		goto fail5;
648 
649 	efx_ptp_defer_probe_with_channel(efx);
650 
651 #ifdef CONFIG_SFC_SRIOV
652 	if ((efx->pci_dev->physfn) && (!efx->pci_dev->is_physfn)) {
653 		struct pci_dev *pci_dev_pf = efx->pci_dev->physfn;
654 		struct efx_nic *efx_pf = pci_get_drvdata(pci_dev_pf);
655 
656 		efx_pf->type->get_mac_address(efx_pf, nic_data->port_id);
657 	} else
658 #endif
659 		ether_addr_copy(nic_data->port_id, efx->net_dev->perm_addr);
660 
661 	INIT_LIST_HEAD(&nic_data->vlan_list);
662 	mutex_init(&nic_data->vlan_lock);
663 
664 	/* Add unspecified VID to support VLAN filtering being disabled */
665 	rc = efx_ef10_add_vlan(efx, EFX_FILTER_VID_UNSPEC);
666 	if (rc)
667 		goto fail_add_vid_unspec;
668 
669 	/* If VLAN filtering is enabled, we need VID 0 to get untagged
670 	 * traffic.  It is added automatically if 8021q module is loaded,
671 	 * but we can't rely on it since module may be not loaded.
672 	 */
673 	rc = efx_ef10_add_vlan(efx, 0);
674 	if (rc)
675 		goto fail_add_vid_0;
676 
677 	if (nic_data->datapath_caps &
678 	    (1 << MC_CMD_GET_CAPABILITIES_OUT_VXLAN_NVGRE_LBN) &&
679 	    efx->mcdi->fn_flags &
680 	    (1 << MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_TRUSTED))
681 		efx->net_dev->udp_tunnel_nic_info = &efx_ef10_udp_tunnels;
682 
683 	return 0;
684 
685 fail_add_vid_0:
686 	efx_ef10_cleanup_vlans(efx);
687 fail_add_vid_unspec:
688 	mutex_destroy(&nic_data->vlan_lock);
689 	efx_ptp_remove(efx);
690 	efx_mcdi_mon_remove(efx);
691 fail5:
692 	device_remove_file(&efx->pci_dev->dev, &dev_attr_primary_flag);
693 fail4:
694 	device_remove_file(&efx->pci_dev->dev, &dev_attr_link_control_flag);
695 fail3:
696 	efx_mcdi_detach(efx);
697 
698 	mutex_lock(&nic_data->udp_tunnels_lock);
699 	memset(nic_data->udp_tunnels, 0, sizeof(nic_data->udp_tunnels));
700 	(void)efx_ef10_set_udp_tnl_ports(efx, true);
701 	mutex_unlock(&nic_data->udp_tunnels_lock);
702 	mutex_destroy(&nic_data->udp_tunnels_lock);
703 
704 	efx_mcdi_fini(efx);
705 fail2:
706 	efx_nic_free_buffer(efx, &nic_data->mcdi_buf);
707 fail1:
708 	kfree(nic_data);
709 	efx->nic_data = NULL;
710 	return rc;
711 }
712 
713 #ifdef EFX_USE_PIO
714 
715 static void efx_ef10_free_piobufs(struct efx_nic *efx)
716 {
717 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
718 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FREE_PIOBUF_IN_LEN);
719 	unsigned int i;
720 	int rc;
721 
722 	BUILD_BUG_ON(MC_CMD_FREE_PIOBUF_OUT_LEN != 0);
723 
724 	for (i = 0; i < nic_data->n_piobufs; i++) {
725 		MCDI_SET_DWORD(inbuf, FREE_PIOBUF_IN_PIOBUF_HANDLE,
726 			       nic_data->piobuf_handle[i]);
727 		rc = efx_mcdi_rpc(efx, MC_CMD_FREE_PIOBUF, inbuf, sizeof(inbuf),
728 				  NULL, 0, NULL);
729 		WARN_ON(rc);
730 	}
731 
732 	nic_data->n_piobufs = 0;
733 }
734 
735 static int efx_ef10_alloc_piobufs(struct efx_nic *efx, unsigned int n)
736 {
737 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
738 	MCDI_DECLARE_BUF(outbuf, MC_CMD_ALLOC_PIOBUF_OUT_LEN);
739 	unsigned int i;
740 	size_t outlen;
741 	int rc = 0;
742 
743 	BUILD_BUG_ON(MC_CMD_ALLOC_PIOBUF_IN_LEN != 0);
744 
745 	for (i = 0; i < n; i++) {
746 		rc = efx_mcdi_rpc_quiet(efx, MC_CMD_ALLOC_PIOBUF, NULL, 0,
747 					outbuf, sizeof(outbuf), &outlen);
748 		if (rc) {
749 			/* Don't display the MC error if we didn't have space
750 			 * for a VF.
751 			 */
752 			if (!(efx_ef10_is_vf(efx) && rc == -ENOSPC))
753 				efx_mcdi_display_error(efx, MC_CMD_ALLOC_PIOBUF,
754 						       0, outbuf, outlen, rc);
755 			break;
756 		}
757 		if (outlen < MC_CMD_ALLOC_PIOBUF_OUT_LEN) {
758 			rc = -EIO;
759 			break;
760 		}
761 		nic_data->piobuf_handle[i] =
762 			MCDI_DWORD(outbuf, ALLOC_PIOBUF_OUT_PIOBUF_HANDLE);
763 		netif_dbg(efx, probe, efx->net_dev,
764 			  "allocated PIO buffer %u handle %x\n", i,
765 			  nic_data->piobuf_handle[i]);
766 	}
767 
768 	nic_data->n_piobufs = i;
769 	if (rc)
770 		efx_ef10_free_piobufs(efx);
771 	return rc;
772 }
773 
774 static int efx_ef10_link_piobufs(struct efx_nic *efx)
775 {
776 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
777 	MCDI_DECLARE_BUF(inbuf, MC_CMD_LINK_PIOBUF_IN_LEN);
778 	struct efx_channel *channel;
779 	struct efx_tx_queue *tx_queue;
780 	unsigned int offset, index;
781 	int rc;
782 
783 	BUILD_BUG_ON(MC_CMD_LINK_PIOBUF_OUT_LEN != 0);
784 	BUILD_BUG_ON(MC_CMD_UNLINK_PIOBUF_OUT_LEN != 0);
785 
786 	/* Link a buffer to each VI in the write-combining mapping */
787 	for (index = 0; index < nic_data->n_piobufs; ++index) {
788 		MCDI_SET_DWORD(inbuf, LINK_PIOBUF_IN_PIOBUF_HANDLE,
789 			       nic_data->piobuf_handle[index]);
790 		MCDI_SET_DWORD(inbuf, LINK_PIOBUF_IN_TXQ_INSTANCE,
791 			       nic_data->pio_write_vi_base + index);
792 		rc = efx_mcdi_rpc(efx, MC_CMD_LINK_PIOBUF,
793 				  inbuf, MC_CMD_LINK_PIOBUF_IN_LEN,
794 				  NULL, 0, NULL);
795 		if (rc) {
796 			netif_err(efx, drv, efx->net_dev,
797 				  "failed to link VI %u to PIO buffer %u (%d)\n",
798 				  nic_data->pio_write_vi_base + index, index,
799 				  rc);
800 			goto fail;
801 		}
802 		netif_dbg(efx, probe, efx->net_dev,
803 			  "linked VI %u to PIO buffer %u\n",
804 			  nic_data->pio_write_vi_base + index, index);
805 	}
806 
807 	/* Link a buffer to each TX queue */
808 	efx_for_each_channel(channel, efx) {
809 		/* Extra channels, even those with TXQs (PTP), do not require
810 		 * PIO resources.
811 		 */
812 		if (!channel->type->want_pio ||
813 		    channel->channel >= efx->xdp_channel_offset)
814 			continue;
815 
816 		efx_for_each_channel_tx_queue(tx_queue, channel) {
817 			/* We assign the PIO buffers to queues in
818 			 * reverse order to allow for the following
819 			 * special case.
820 			 */
821 			offset = ((efx->tx_channel_offset + efx->n_tx_channels -
822 				   tx_queue->channel->channel - 1) *
823 				  efx_piobuf_size);
824 			index = offset / nic_data->piobuf_size;
825 			offset = offset % nic_data->piobuf_size;
826 
827 			/* When the host page size is 4K, the first
828 			 * host page in the WC mapping may be within
829 			 * the same VI page as the last TX queue.  We
830 			 * can only link one buffer to each VI.
831 			 */
832 			if (tx_queue->queue == nic_data->pio_write_vi_base) {
833 				BUG_ON(index != 0);
834 				rc = 0;
835 			} else {
836 				MCDI_SET_DWORD(inbuf,
837 					       LINK_PIOBUF_IN_PIOBUF_HANDLE,
838 					       nic_data->piobuf_handle[index]);
839 				MCDI_SET_DWORD(inbuf,
840 					       LINK_PIOBUF_IN_TXQ_INSTANCE,
841 					       tx_queue->queue);
842 				rc = efx_mcdi_rpc(efx, MC_CMD_LINK_PIOBUF,
843 						  inbuf, MC_CMD_LINK_PIOBUF_IN_LEN,
844 						  NULL, 0, NULL);
845 			}
846 
847 			if (rc) {
848 				/* This is non-fatal; the TX path just
849 				 * won't use PIO for this queue
850 				 */
851 				netif_err(efx, drv, efx->net_dev,
852 					  "failed to link VI %u to PIO buffer %u (%d)\n",
853 					  tx_queue->queue, index, rc);
854 				tx_queue->piobuf = NULL;
855 			} else {
856 				tx_queue->piobuf =
857 					nic_data->pio_write_base +
858 					index * efx->vi_stride + offset;
859 				tx_queue->piobuf_offset = offset;
860 				netif_dbg(efx, probe, efx->net_dev,
861 					  "linked VI %u to PIO buffer %u offset %x addr %p\n",
862 					  tx_queue->queue, index,
863 					  tx_queue->piobuf_offset,
864 					  tx_queue->piobuf);
865 			}
866 		}
867 	}
868 
869 	return 0;
870 
871 fail:
872 	/* inbuf was defined for MC_CMD_LINK_PIOBUF.  We can use the same
873 	 * buffer for MC_CMD_UNLINK_PIOBUF because it's shorter.
874 	 */
875 	BUILD_BUG_ON(MC_CMD_LINK_PIOBUF_IN_LEN < MC_CMD_UNLINK_PIOBUF_IN_LEN);
876 	while (index--) {
877 		MCDI_SET_DWORD(inbuf, UNLINK_PIOBUF_IN_TXQ_INSTANCE,
878 			       nic_data->pio_write_vi_base + index);
879 		efx_mcdi_rpc(efx, MC_CMD_UNLINK_PIOBUF,
880 			     inbuf, MC_CMD_UNLINK_PIOBUF_IN_LEN,
881 			     NULL, 0, NULL);
882 	}
883 	return rc;
884 }
885 
886 static void efx_ef10_forget_old_piobufs(struct efx_nic *efx)
887 {
888 	struct efx_channel *channel;
889 	struct efx_tx_queue *tx_queue;
890 
891 	/* All our existing PIO buffers went away */
892 	efx_for_each_channel(channel, efx)
893 		efx_for_each_channel_tx_queue(tx_queue, channel)
894 			tx_queue->piobuf = NULL;
895 }
896 
897 #else /* !EFX_USE_PIO */
898 
899 static int efx_ef10_alloc_piobufs(struct efx_nic *efx, unsigned int n)
900 {
901 	return n == 0 ? 0 : -ENOBUFS;
902 }
903 
904 static int efx_ef10_link_piobufs(struct efx_nic *efx)
905 {
906 	return 0;
907 }
908 
909 static void efx_ef10_free_piobufs(struct efx_nic *efx)
910 {
911 }
912 
913 static void efx_ef10_forget_old_piobufs(struct efx_nic *efx)
914 {
915 }
916 
917 #endif /* EFX_USE_PIO */
918 
919 static void efx_ef10_remove(struct efx_nic *efx)
920 {
921 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
922 	int rc;
923 
924 #ifdef CONFIG_SFC_SRIOV
925 	struct efx_ef10_nic_data *nic_data_pf;
926 	struct pci_dev *pci_dev_pf;
927 	struct efx_nic *efx_pf;
928 	struct ef10_vf *vf;
929 
930 	if (efx->pci_dev->is_virtfn) {
931 		pci_dev_pf = efx->pci_dev->physfn;
932 		if (pci_dev_pf) {
933 			efx_pf = pci_get_drvdata(pci_dev_pf);
934 			nic_data_pf = efx_pf->nic_data;
935 			vf = nic_data_pf->vf + nic_data->vf_index;
936 			vf->efx = NULL;
937 		} else
938 			netif_info(efx, drv, efx->net_dev,
939 				   "Could not get the PF id from VF\n");
940 	}
941 #endif
942 
943 	efx_ef10_cleanup_vlans(efx);
944 	mutex_destroy(&nic_data->vlan_lock);
945 
946 	efx_ptp_remove(efx);
947 
948 	efx_mcdi_mon_remove(efx);
949 
950 	efx_mcdi_rx_free_indir_table(efx);
951 
952 	if (nic_data->wc_membase)
953 		iounmap(nic_data->wc_membase);
954 
955 	rc = efx_mcdi_free_vis(efx);
956 	WARN_ON(rc != 0);
957 
958 	if (!nic_data->must_restore_piobufs)
959 		efx_ef10_free_piobufs(efx);
960 
961 	device_remove_file(&efx->pci_dev->dev, &dev_attr_primary_flag);
962 	device_remove_file(&efx->pci_dev->dev, &dev_attr_link_control_flag);
963 
964 	efx_mcdi_detach(efx);
965 
966 	memset(nic_data->udp_tunnels, 0, sizeof(nic_data->udp_tunnels));
967 	mutex_lock(&nic_data->udp_tunnels_lock);
968 	(void)efx_ef10_set_udp_tnl_ports(efx, true);
969 	mutex_unlock(&nic_data->udp_tunnels_lock);
970 
971 	mutex_destroy(&nic_data->udp_tunnels_lock);
972 
973 	efx_mcdi_fini(efx);
974 	efx_nic_free_buffer(efx, &nic_data->mcdi_buf);
975 	kfree(nic_data);
976 }
977 
978 static int efx_ef10_probe_pf(struct efx_nic *efx)
979 {
980 	return efx_ef10_probe(efx);
981 }
982 
983 int efx_ef10_vadaptor_query(struct efx_nic *efx, unsigned int port_id,
984 			    u32 *port_flags, u32 *vadaptor_flags,
985 			    unsigned int *vlan_tags)
986 {
987 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
988 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VADAPTOR_QUERY_IN_LEN);
989 	MCDI_DECLARE_BUF(outbuf, MC_CMD_VADAPTOR_QUERY_OUT_LEN);
990 	size_t outlen;
991 	int rc;
992 
993 	if (nic_data->datapath_caps &
994 	    (1 << MC_CMD_GET_CAPABILITIES_OUT_VADAPTOR_QUERY_LBN)) {
995 		MCDI_SET_DWORD(inbuf, VADAPTOR_QUERY_IN_UPSTREAM_PORT_ID,
996 			       port_id);
997 
998 		rc = efx_mcdi_rpc(efx, MC_CMD_VADAPTOR_QUERY, inbuf, sizeof(inbuf),
999 				  outbuf, sizeof(outbuf), &outlen);
1000 		if (rc)
1001 			return rc;
1002 
1003 		if (outlen < sizeof(outbuf)) {
1004 			rc = -EIO;
1005 			return rc;
1006 		}
1007 	}
1008 
1009 	if (port_flags)
1010 		*port_flags = MCDI_DWORD(outbuf, VADAPTOR_QUERY_OUT_PORT_FLAGS);
1011 	if (vadaptor_flags)
1012 		*vadaptor_flags =
1013 			MCDI_DWORD(outbuf, VADAPTOR_QUERY_OUT_VADAPTOR_FLAGS);
1014 	if (vlan_tags)
1015 		*vlan_tags =
1016 			MCDI_DWORD(outbuf,
1017 				   VADAPTOR_QUERY_OUT_NUM_AVAILABLE_VLAN_TAGS);
1018 
1019 	return 0;
1020 }
1021 
1022 int efx_ef10_vadaptor_alloc(struct efx_nic *efx, unsigned int port_id)
1023 {
1024 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VADAPTOR_ALLOC_IN_LEN);
1025 
1026 	MCDI_SET_DWORD(inbuf, VADAPTOR_ALLOC_IN_UPSTREAM_PORT_ID, port_id);
1027 	return efx_mcdi_rpc(efx, MC_CMD_VADAPTOR_ALLOC, inbuf, sizeof(inbuf),
1028 			    NULL, 0, NULL);
1029 }
1030 
1031 int efx_ef10_vadaptor_free(struct efx_nic *efx, unsigned int port_id)
1032 {
1033 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VADAPTOR_FREE_IN_LEN);
1034 
1035 	MCDI_SET_DWORD(inbuf, VADAPTOR_FREE_IN_UPSTREAM_PORT_ID, port_id);
1036 	return efx_mcdi_rpc(efx, MC_CMD_VADAPTOR_FREE, inbuf, sizeof(inbuf),
1037 			    NULL, 0, NULL);
1038 }
1039 
1040 int efx_ef10_vport_add_mac(struct efx_nic *efx,
1041 			   unsigned int port_id, const u8 *mac)
1042 {
1043 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VPORT_ADD_MAC_ADDRESS_IN_LEN);
1044 
1045 	MCDI_SET_DWORD(inbuf, VPORT_ADD_MAC_ADDRESS_IN_VPORT_ID, port_id);
1046 	ether_addr_copy(MCDI_PTR(inbuf, VPORT_ADD_MAC_ADDRESS_IN_MACADDR), mac);
1047 
1048 	return efx_mcdi_rpc(efx, MC_CMD_VPORT_ADD_MAC_ADDRESS, inbuf,
1049 			    sizeof(inbuf), NULL, 0, NULL);
1050 }
1051 
1052 int efx_ef10_vport_del_mac(struct efx_nic *efx,
1053 			   unsigned int port_id, const u8 *mac)
1054 {
1055 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VPORT_DEL_MAC_ADDRESS_IN_LEN);
1056 
1057 	MCDI_SET_DWORD(inbuf, VPORT_DEL_MAC_ADDRESS_IN_VPORT_ID, port_id);
1058 	ether_addr_copy(MCDI_PTR(inbuf, VPORT_DEL_MAC_ADDRESS_IN_MACADDR), mac);
1059 
1060 	return efx_mcdi_rpc(efx, MC_CMD_VPORT_DEL_MAC_ADDRESS, inbuf,
1061 			    sizeof(inbuf), NULL, 0, NULL);
1062 }
1063 
1064 #ifdef CONFIG_SFC_SRIOV
1065 static int efx_ef10_probe_vf(struct efx_nic *efx)
1066 {
1067 	int rc;
1068 	struct pci_dev *pci_dev_pf;
1069 
1070 	/* If the parent PF has no VF data structure, it doesn't know about this
1071 	 * VF so fail probe.  The VF needs to be re-created.  This can happen
1072 	 * if the PF driver was unloaded while any VF was assigned to a guest
1073 	 * (using Xen, only).
1074 	 */
1075 	pci_dev_pf = efx->pci_dev->physfn;
1076 	if (pci_dev_pf) {
1077 		struct efx_nic *efx_pf = pci_get_drvdata(pci_dev_pf);
1078 		struct efx_ef10_nic_data *nic_data_pf = efx_pf->nic_data;
1079 
1080 		if (!nic_data_pf->vf) {
1081 			netif_info(efx, drv, efx->net_dev,
1082 				   "The VF cannot link to its parent PF; "
1083 				   "please destroy and re-create the VF\n");
1084 			return -EBUSY;
1085 		}
1086 	}
1087 
1088 	rc = efx_ef10_probe(efx);
1089 	if (rc)
1090 		return rc;
1091 
1092 	rc = efx_ef10_get_vf_index(efx);
1093 	if (rc)
1094 		goto fail;
1095 
1096 	if (efx->pci_dev->is_virtfn) {
1097 		if (efx->pci_dev->physfn) {
1098 			struct efx_nic *efx_pf =
1099 				pci_get_drvdata(efx->pci_dev->physfn);
1100 			struct efx_ef10_nic_data *nic_data_p = efx_pf->nic_data;
1101 			struct efx_ef10_nic_data *nic_data = efx->nic_data;
1102 
1103 			nic_data_p->vf[nic_data->vf_index].efx = efx;
1104 			nic_data_p->vf[nic_data->vf_index].pci_dev =
1105 				efx->pci_dev;
1106 		} else
1107 			netif_info(efx, drv, efx->net_dev,
1108 				   "Could not get the PF id from VF\n");
1109 	}
1110 
1111 	return 0;
1112 
1113 fail:
1114 	efx_ef10_remove(efx);
1115 	return rc;
1116 }
1117 #else
1118 static int efx_ef10_probe_vf(struct efx_nic *efx __attribute__ ((unused)))
1119 {
1120 	return 0;
1121 }
1122 #endif
1123 
1124 static int efx_ef10_alloc_vis(struct efx_nic *efx,
1125 			      unsigned int min_vis, unsigned int max_vis)
1126 {
1127 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1128 
1129 	return efx_mcdi_alloc_vis(efx, min_vis, max_vis, &nic_data->vi_base,
1130 				  &nic_data->n_allocated_vis);
1131 }
1132 
1133 /* Note that the failure path of this function does not free
1134  * resources, as this will be done by efx_ef10_remove().
1135  */
1136 static int efx_ef10_dimension_resources(struct efx_nic *efx)
1137 {
1138 	unsigned int min_vis = max_t(unsigned int, efx->tx_queues_per_channel,
1139 				     efx_separate_tx_channels ? 2 : 1);
1140 	unsigned int channel_vis, pio_write_vi_base, max_vis;
1141 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1142 	unsigned int uc_mem_map_size, wc_mem_map_size;
1143 	void __iomem *membase;
1144 	int rc;
1145 
1146 	channel_vis = max(efx->n_channels,
1147 			  ((efx->n_tx_channels + efx->n_extra_tx_channels) *
1148 			   efx->tx_queues_per_channel) +
1149 			   efx->n_xdp_channels * efx->xdp_tx_per_channel);
1150 	if (efx->max_vis && efx->max_vis < channel_vis) {
1151 		netif_dbg(efx, drv, efx->net_dev,
1152 			  "Reducing channel VIs from %u to %u\n",
1153 			  channel_vis, efx->max_vis);
1154 		channel_vis = efx->max_vis;
1155 	}
1156 
1157 #ifdef EFX_USE_PIO
1158 	/* Try to allocate PIO buffers if wanted and if the full
1159 	 * number of PIO buffers would be sufficient to allocate one
1160 	 * copy-buffer per TX channel.  Failure is non-fatal, as there
1161 	 * are only a small number of PIO buffers shared between all
1162 	 * functions of the controller.
1163 	 */
1164 	if (efx_piobuf_size != 0 &&
1165 	    nic_data->piobuf_size / efx_piobuf_size * EF10_TX_PIOBUF_COUNT >=
1166 	    efx->n_tx_channels) {
1167 		unsigned int n_piobufs =
1168 			DIV_ROUND_UP(efx->n_tx_channels,
1169 				     nic_data->piobuf_size / efx_piobuf_size);
1170 
1171 		rc = efx_ef10_alloc_piobufs(efx, n_piobufs);
1172 		if (rc == -ENOSPC)
1173 			netif_dbg(efx, probe, efx->net_dev,
1174 				  "out of PIO buffers; cannot allocate more\n");
1175 		else if (rc == -EPERM)
1176 			netif_dbg(efx, probe, efx->net_dev,
1177 				  "not permitted to allocate PIO buffers\n");
1178 		else if (rc)
1179 			netif_err(efx, probe, efx->net_dev,
1180 				  "failed to allocate PIO buffers (%d)\n", rc);
1181 		else
1182 			netif_dbg(efx, probe, efx->net_dev,
1183 				  "allocated %u PIO buffers\n", n_piobufs);
1184 	}
1185 #else
1186 	nic_data->n_piobufs = 0;
1187 #endif
1188 
1189 	/* PIO buffers should be mapped with write-combining enabled,
1190 	 * and we want to make single UC and WC mappings rather than
1191 	 * several of each (in fact that's the only option if host
1192 	 * page size is >4K).  So we may allocate some extra VIs just
1193 	 * for writing PIO buffers through.
1194 	 *
1195 	 * The UC mapping contains (channel_vis - 1) complete VIs and the
1196 	 * first 4K of the next VI.  Then the WC mapping begins with
1197 	 * the remainder of this last VI.
1198 	 */
1199 	uc_mem_map_size = PAGE_ALIGN((channel_vis - 1) * efx->vi_stride +
1200 				     ER_DZ_TX_PIOBUF);
1201 	if (nic_data->n_piobufs) {
1202 		/* pio_write_vi_base rounds down to give the number of complete
1203 		 * VIs inside the UC mapping.
1204 		 */
1205 		pio_write_vi_base = uc_mem_map_size / efx->vi_stride;
1206 		wc_mem_map_size = (PAGE_ALIGN((pio_write_vi_base +
1207 					       nic_data->n_piobufs) *
1208 					      efx->vi_stride) -
1209 				   uc_mem_map_size);
1210 		max_vis = pio_write_vi_base + nic_data->n_piobufs;
1211 	} else {
1212 		pio_write_vi_base = 0;
1213 		wc_mem_map_size = 0;
1214 		max_vis = channel_vis;
1215 	}
1216 
1217 	/* In case the last attached driver failed to free VIs, do it now */
1218 	rc = efx_mcdi_free_vis(efx);
1219 	if (rc != 0)
1220 		return rc;
1221 
1222 	rc = efx_ef10_alloc_vis(efx, min_vis, max_vis);
1223 	if (rc != 0)
1224 		return rc;
1225 
1226 	if (nic_data->n_allocated_vis < channel_vis) {
1227 		netif_info(efx, drv, efx->net_dev,
1228 			   "Could not allocate enough VIs to satisfy RSS"
1229 			   " requirements. Performance may not be optimal.\n");
1230 		/* We didn't get the VIs to populate our channels.
1231 		 * We could keep what we got but then we'd have more
1232 		 * interrupts than we need.
1233 		 * Instead calculate new max_channels and restart
1234 		 */
1235 		efx->max_channels = nic_data->n_allocated_vis;
1236 		efx->max_tx_channels =
1237 			nic_data->n_allocated_vis / efx->tx_queues_per_channel;
1238 
1239 		efx_mcdi_free_vis(efx);
1240 		return -EAGAIN;
1241 	}
1242 
1243 	/* If we didn't get enough VIs to map all the PIO buffers, free the
1244 	 * PIO buffers
1245 	 */
1246 	if (nic_data->n_piobufs &&
1247 	    nic_data->n_allocated_vis <
1248 	    pio_write_vi_base + nic_data->n_piobufs) {
1249 		netif_dbg(efx, probe, efx->net_dev,
1250 			  "%u VIs are not sufficient to map %u PIO buffers\n",
1251 			  nic_data->n_allocated_vis, nic_data->n_piobufs);
1252 		efx_ef10_free_piobufs(efx);
1253 	}
1254 
1255 	/* Shrink the original UC mapping of the memory BAR */
1256 	membase = ioremap(efx->membase_phys, uc_mem_map_size);
1257 	if (!membase) {
1258 		netif_err(efx, probe, efx->net_dev,
1259 			  "could not shrink memory BAR to %x\n",
1260 			  uc_mem_map_size);
1261 		return -ENOMEM;
1262 	}
1263 	iounmap(efx->membase);
1264 	efx->membase = membase;
1265 
1266 	/* Set up the WC mapping if needed */
1267 	if (wc_mem_map_size) {
1268 		nic_data->wc_membase = ioremap_wc(efx->membase_phys +
1269 						  uc_mem_map_size,
1270 						  wc_mem_map_size);
1271 		if (!nic_data->wc_membase) {
1272 			netif_err(efx, probe, efx->net_dev,
1273 				  "could not allocate WC mapping of size %x\n",
1274 				  wc_mem_map_size);
1275 			return -ENOMEM;
1276 		}
1277 		nic_data->pio_write_vi_base = pio_write_vi_base;
1278 		nic_data->pio_write_base =
1279 			nic_data->wc_membase +
1280 			(pio_write_vi_base * efx->vi_stride + ER_DZ_TX_PIOBUF -
1281 			 uc_mem_map_size);
1282 
1283 		rc = efx_ef10_link_piobufs(efx);
1284 		if (rc)
1285 			efx_ef10_free_piobufs(efx);
1286 	}
1287 
1288 	netif_dbg(efx, probe, efx->net_dev,
1289 		  "memory BAR at %pa (virtual %p+%x UC, %p+%x WC)\n",
1290 		  &efx->membase_phys, efx->membase, uc_mem_map_size,
1291 		  nic_data->wc_membase, wc_mem_map_size);
1292 
1293 	return 0;
1294 }
1295 
1296 static void efx_ef10_fini_nic(struct efx_nic *efx)
1297 {
1298 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1299 
1300 	spin_lock_bh(&efx->stats_lock);
1301 	kfree(nic_data->mc_stats);
1302 	nic_data->mc_stats = NULL;
1303 	spin_unlock_bh(&efx->stats_lock);
1304 }
1305 
1306 static int efx_ef10_init_nic(struct efx_nic *efx)
1307 {
1308 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1309 	struct net_device *net_dev = efx->net_dev;
1310 	netdev_features_t tun_feats, tso_feats;
1311 	int rc;
1312 
1313 	if (nic_data->must_check_datapath_caps) {
1314 		rc = efx_ef10_init_datapath_caps(efx);
1315 		if (rc)
1316 			return rc;
1317 		nic_data->must_check_datapath_caps = false;
1318 	}
1319 
1320 	if (efx->must_realloc_vis) {
1321 		/* We cannot let the number of VIs change now */
1322 		rc = efx_ef10_alloc_vis(efx, nic_data->n_allocated_vis,
1323 					nic_data->n_allocated_vis);
1324 		if (rc)
1325 			return rc;
1326 		efx->must_realloc_vis = false;
1327 	}
1328 
1329 	nic_data->mc_stats = kmalloc(efx->num_mac_stats * sizeof(__le64),
1330 				     GFP_KERNEL);
1331 	if (!nic_data->mc_stats)
1332 		return -ENOMEM;
1333 
1334 	if (nic_data->must_restore_piobufs && nic_data->n_piobufs) {
1335 		rc = efx_ef10_alloc_piobufs(efx, nic_data->n_piobufs);
1336 		if (rc == 0) {
1337 			rc = efx_ef10_link_piobufs(efx);
1338 			if (rc)
1339 				efx_ef10_free_piobufs(efx);
1340 		}
1341 
1342 		/* Log an error on failure, but this is non-fatal.
1343 		 * Permission errors are less important - we've presumably
1344 		 * had the PIO buffer licence removed.
1345 		 */
1346 		if (rc == -EPERM)
1347 			netif_dbg(efx, drv, efx->net_dev,
1348 				  "not permitted to restore PIO buffers\n");
1349 		else if (rc)
1350 			netif_err(efx, drv, efx->net_dev,
1351 				  "failed to restore PIO buffers (%d)\n", rc);
1352 		nic_data->must_restore_piobufs = false;
1353 	}
1354 
1355 	/* encap features might change during reset if fw variant changed */
1356 	if (efx_has_cap(efx, VXLAN_NVGRE) && !efx_ef10_is_vf(efx))
1357 		net_dev->hw_enc_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
1358 	else
1359 		net_dev->hw_enc_features &= ~(NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
1360 
1361 	tun_feats = NETIF_F_GSO_UDP_TUNNEL | NETIF_F_GSO_GRE |
1362 		    NETIF_F_GSO_UDP_TUNNEL_CSUM | NETIF_F_GSO_GRE_CSUM;
1363 	tso_feats = NETIF_F_TSO | NETIF_F_TSO6;
1364 
1365 	if (efx_has_cap(efx, TX_TSO_V2_ENCAP)) {
1366 		/* If this is first nic_init, or if it is a reset and a new fw
1367 		 * variant has added new features, enable them by default.
1368 		 * If the features are not new, maintain their current value.
1369 		 */
1370 		if (!(net_dev->hw_features & tun_feats))
1371 			net_dev->features |= tun_feats;
1372 		net_dev->hw_enc_features |= tun_feats | tso_feats;
1373 		net_dev->hw_features |= tun_feats;
1374 	} else {
1375 		net_dev->hw_enc_features &= ~(tun_feats | tso_feats);
1376 		net_dev->hw_features &= ~tun_feats;
1377 		net_dev->features &= ~tun_feats;
1378 	}
1379 
1380 	/* don't fail init if RSS setup doesn't work */
1381 	rc = efx->type->rx_push_rss_config(efx, false,
1382 					   efx->rss_context.rx_indir_table, NULL);
1383 
1384 	return 0;
1385 }
1386 
1387 static void efx_ef10_table_reset_mc_allocations(struct efx_nic *efx)
1388 {
1389 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1390 #ifdef CONFIG_SFC_SRIOV
1391 	unsigned int i;
1392 #endif
1393 
1394 	/* All our allocations have been reset */
1395 	efx->must_realloc_vis = true;
1396 	efx_mcdi_filter_table_reset_mc_allocations(efx);
1397 	nic_data->must_restore_piobufs = true;
1398 	efx_ef10_forget_old_piobufs(efx);
1399 	efx->rss_context.context_id = EFX_MCDI_RSS_CONTEXT_INVALID;
1400 
1401 	/* Driver-created vswitches and vports must be re-created */
1402 	nic_data->must_probe_vswitching = true;
1403 	efx->vport_id = EVB_PORT_ID_ASSIGNED;
1404 #ifdef CONFIG_SFC_SRIOV
1405 	if (nic_data->vf)
1406 		for (i = 0; i < efx->vf_count; i++)
1407 			nic_data->vf[i].vport_id = 0;
1408 #endif
1409 }
1410 
1411 static enum reset_type efx_ef10_map_reset_reason(enum reset_type reason)
1412 {
1413 	if (reason == RESET_TYPE_MC_FAILURE)
1414 		return RESET_TYPE_DATAPATH;
1415 
1416 	return efx_mcdi_map_reset_reason(reason);
1417 }
1418 
1419 static int efx_ef10_map_reset_flags(u32 *flags)
1420 {
1421 	enum {
1422 		EF10_RESET_PORT = ((ETH_RESET_MAC | ETH_RESET_PHY) <<
1423 				   ETH_RESET_SHARED_SHIFT),
1424 		EF10_RESET_MC = ((ETH_RESET_DMA | ETH_RESET_FILTER |
1425 				  ETH_RESET_OFFLOAD | ETH_RESET_MAC |
1426 				  ETH_RESET_PHY | ETH_RESET_MGMT) <<
1427 				 ETH_RESET_SHARED_SHIFT)
1428 	};
1429 
1430 	/* We assume for now that our PCI function is permitted to
1431 	 * reset everything.
1432 	 */
1433 
1434 	if ((*flags & EF10_RESET_MC) == EF10_RESET_MC) {
1435 		*flags &= ~EF10_RESET_MC;
1436 		return RESET_TYPE_WORLD;
1437 	}
1438 
1439 	if ((*flags & EF10_RESET_PORT) == EF10_RESET_PORT) {
1440 		*flags &= ~EF10_RESET_PORT;
1441 		return RESET_TYPE_ALL;
1442 	}
1443 
1444 	/* no invisible reset implemented */
1445 
1446 	return -EINVAL;
1447 }
1448 
1449 static int efx_ef10_reset(struct efx_nic *efx, enum reset_type reset_type)
1450 {
1451 	int rc = efx_mcdi_reset(efx, reset_type);
1452 
1453 	/* Unprivileged functions return -EPERM, but need to return success
1454 	 * here so that the datapath is brought back up.
1455 	 */
1456 	if (reset_type == RESET_TYPE_WORLD && rc == -EPERM)
1457 		rc = 0;
1458 
1459 	/* If it was a port reset, trigger reallocation of MC resources.
1460 	 * Note that on an MC reset nothing needs to be done now because we'll
1461 	 * detect the MC reset later and handle it then.
1462 	 * For an FLR, we never get an MC reset event, but the MC has reset all
1463 	 * resources assigned to us, so we have to trigger reallocation now.
1464 	 */
1465 	if ((reset_type == RESET_TYPE_ALL ||
1466 	     reset_type == RESET_TYPE_MCDI_TIMEOUT) && !rc)
1467 		efx_ef10_table_reset_mc_allocations(efx);
1468 	return rc;
1469 }
1470 
1471 #define EF10_DMA_STAT(ext_name, mcdi_name)			\
1472 	[EF10_STAT_ ## ext_name] =				\
1473 	{ #ext_name, 64, 8 * MC_CMD_MAC_ ## mcdi_name }
1474 #define EF10_DMA_INVIS_STAT(int_name, mcdi_name)		\
1475 	[EF10_STAT_ ## int_name] =				\
1476 	{ NULL, 64, 8 * MC_CMD_MAC_ ## mcdi_name }
1477 #define EF10_OTHER_STAT(ext_name)				\
1478 	[EF10_STAT_ ## ext_name] = { #ext_name, 0, 0 }
1479 
1480 static const struct efx_hw_stat_desc efx_ef10_stat_desc[EF10_STAT_COUNT] = {
1481 	EF10_DMA_STAT(port_tx_bytes, TX_BYTES),
1482 	EF10_DMA_STAT(port_tx_packets, TX_PKTS),
1483 	EF10_DMA_STAT(port_tx_pause, TX_PAUSE_PKTS),
1484 	EF10_DMA_STAT(port_tx_control, TX_CONTROL_PKTS),
1485 	EF10_DMA_STAT(port_tx_unicast, TX_UNICAST_PKTS),
1486 	EF10_DMA_STAT(port_tx_multicast, TX_MULTICAST_PKTS),
1487 	EF10_DMA_STAT(port_tx_broadcast, TX_BROADCAST_PKTS),
1488 	EF10_DMA_STAT(port_tx_lt64, TX_LT64_PKTS),
1489 	EF10_DMA_STAT(port_tx_64, TX_64_PKTS),
1490 	EF10_DMA_STAT(port_tx_65_to_127, TX_65_TO_127_PKTS),
1491 	EF10_DMA_STAT(port_tx_128_to_255, TX_128_TO_255_PKTS),
1492 	EF10_DMA_STAT(port_tx_256_to_511, TX_256_TO_511_PKTS),
1493 	EF10_DMA_STAT(port_tx_512_to_1023, TX_512_TO_1023_PKTS),
1494 	EF10_DMA_STAT(port_tx_1024_to_15xx, TX_1024_TO_15XX_PKTS),
1495 	EF10_DMA_STAT(port_tx_15xx_to_jumbo, TX_15XX_TO_JUMBO_PKTS),
1496 	EF10_DMA_STAT(port_rx_bytes, RX_BYTES),
1497 	EF10_DMA_INVIS_STAT(port_rx_bytes_minus_good_bytes, RX_BAD_BYTES),
1498 	EF10_OTHER_STAT(port_rx_good_bytes),
1499 	EF10_OTHER_STAT(port_rx_bad_bytes),
1500 	EF10_DMA_STAT(port_rx_packets, RX_PKTS),
1501 	EF10_DMA_STAT(port_rx_good, RX_GOOD_PKTS),
1502 	EF10_DMA_STAT(port_rx_bad, RX_BAD_FCS_PKTS),
1503 	EF10_DMA_STAT(port_rx_pause, RX_PAUSE_PKTS),
1504 	EF10_DMA_STAT(port_rx_control, RX_CONTROL_PKTS),
1505 	EF10_DMA_STAT(port_rx_unicast, RX_UNICAST_PKTS),
1506 	EF10_DMA_STAT(port_rx_multicast, RX_MULTICAST_PKTS),
1507 	EF10_DMA_STAT(port_rx_broadcast, RX_BROADCAST_PKTS),
1508 	EF10_DMA_STAT(port_rx_lt64, RX_UNDERSIZE_PKTS),
1509 	EF10_DMA_STAT(port_rx_64, RX_64_PKTS),
1510 	EF10_DMA_STAT(port_rx_65_to_127, RX_65_TO_127_PKTS),
1511 	EF10_DMA_STAT(port_rx_128_to_255, RX_128_TO_255_PKTS),
1512 	EF10_DMA_STAT(port_rx_256_to_511, RX_256_TO_511_PKTS),
1513 	EF10_DMA_STAT(port_rx_512_to_1023, RX_512_TO_1023_PKTS),
1514 	EF10_DMA_STAT(port_rx_1024_to_15xx, RX_1024_TO_15XX_PKTS),
1515 	EF10_DMA_STAT(port_rx_15xx_to_jumbo, RX_15XX_TO_JUMBO_PKTS),
1516 	EF10_DMA_STAT(port_rx_gtjumbo, RX_GTJUMBO_PKTS),
1517 	EF10_DMA_STAT(port_rx_bad_gtjumbo, RX_JABBER_PKTS),
1518 	EF10_DMA_STAT(port_rx_overflow, RX_OVERFLOW_PKTS),
1519 	EF10_DMA_STAT(port_rx_align_error, RX_ALIGN_ERROR_PKTS),
1520 	EF10_DMA_STAT(port_rx_length_error, RX_LENGTH_ERROR_PKTS),
1521 	EF10_DMA_STAT(port_rx_nodesc_drops, RX_NODESC_DROPS),
1522 	EFX_GENERIC_SW_STAT(rx_nodesc_trunc),
1523 	EFX_GENERIC_SW_STAT(rx_noskb_drops),
1524 	EF10_DMA_STAT(port_rx_pm_trunc_bb_overflow, PM_TRUNC_BB_OVERFLOW),
1525 	EF10_DMA_STAT(port_rx_pm_discard_bb_overflow, PM_DISCARD_BB_OVERFLOW),
1526 	EF10_DMA_STAT(port_rx_pm_trunc_vfifo_full, PM_TRUNC_VFIFO_FULL),
1527 	EF10_DMA_STAT(port_rx_pm_discard_vfifo_full, PM_DISCARD_VFIFO_FULL),
1528 	EF10_DMA_STAT(port_rx_pm_trunc_qbb, PM_TRUNC_QBB),
1529 	EF10_DMA_STAT(port_rx_pm_discard_qbb, PM_DISCARD_QBB),
1530 	EF10_DMA_STAT(port_rx_pm_discard_mapping, PM_DISCARD_MAPPING),
1531 	EF10_DMA_STAT(port_rx_dp_q_disabled_packets, RXDP_Q_DISABLED_PKTS),
1532 	EF10_DMA_STAT(port_rx_dp_di_dropped_packets, RXDP_DI_DROPPED_PKTS),
1533 	EF10_DMA_STAT(port_rx_dp_streaming_packets, RXDP_STREAMING_PKTS),
1534 	EF10_DMA_STAT(port_rx_dp_hlb_fetch, RXDP_HLB_FETCH_CONDITIONS),
1535 	EF10_DMA_STAT(port_rx_dp_hlb_wait, RXDP_HLB_WAIT_CONDITIONS),
1536 	EF10_DMA_STAT(rx_unicast, VADAPTER_RX_UNICAST_PACKETS),
1537 	EF10_DMA_STAT(rx_unicast_bytes, VADAPTER_RX_UNICAST_BYTES),
1538 	EF10_DMA_STAT(rx_multicast, VADAPTER_RX_MULTICAST_PACKETS),
1539 	EF10_DMA_STAT(rx_multicast_bytes, VADAPTER_RX_MULTICAST_BYTES),
1540 	EF10_DMA_STAT(rx_broadcast, VADAPTER_RX_BROADCAST_PACKETS),
1541 	EF10_DMA_STAT(rx_broadcast_bytes, VADAPTER_RX_BROADCAST_BYTES),
1542 	EF10_DMA_STAT(rx_bad, VADAPTER_RX_BAD_PACKETS),
1543 	EF10_DMA_STAT(rx_bad_bytes, VADAPTER_RX_BAD_BYTES),
1544 	EF10_DMA_STAT(rx_overflow, VADAPTER_RX_OVERFLOW),
1545 	EF10_DMA_STAT(tx_unicast, VADAPTER_TX_UNICAST_PACKETS),
1546 	EF10_DMA_STAT(tx_unicast_bytes, VADAPTER_TX_UNICAST_BYTES),
1547 	EF10_DMA_STAT(tx_multicast, VADAPTER_TX_MULTICAST_PACKETS),
1548 	EF10_DMA_STAT(tx_multicast_bytes, VADAPTER_TX_MULTICAST_BYTES),
1549 	EF10_DMA_STAT(tx_broadcast, VADAPTER_TX_BROADCAST_PACKETS),
1550 	EF10_DMA_STAT(tx_broadcast_bytes, VADAPTER_TX_BROADCAST_BYTES),
1551 	EF10_DMA_STAT(tx_bad, VADAPTER_TX_BAD_PACKETS),
1552 	EF10_DMA_STAT(tx_bad_bytes, VADAPTER_TX_BAD_BYTES),
1553 	EF10_DMA_STAT(tx_overflow, VADAPTER_TX_OVERFLOW),
1554 	EF10_DMA_STAT(fec_uncorrected_errors, FEC_UNCORRECTED_ERRORS),
1555 	EF10_DMA_STAT(fec_corrected_errors, FEC_CORRECTED_ERRORS),
1556 	EF10_DMA_STAT(fec_corrected_symbols_lane0, FEC_CORRECTED_SYMBOLS_LANE0),
1557 	EF10_DMA_STAT(fec_corrected_symbols_lane1, FEC_CORRECTED_SYMBOLS_LANE1),
1558 	EF10_DMA_STAT(fec_corrected_symbols_lane2, FEC_CORRECTED_SYMBOLS_LANE2),
1559 	EF10_DMA_STAT(fec_corrected_symbols_lane3, FEC_CORRECTED_SYMBOLS_LANE3),
1560 	EF10_DMA_STAT(ctpio_vi_busy_fallback, CTPIO_VI_BUSY_FALLBACK),
1561 	EF10_DMA_STAT(ctpio_long_write_success, CTPIO_LONG_WRITE_SUCCESS),
1562 	EF10_DMA_STAT(ctpio_missing_dbell_fail, CTPIO_MISSING_DBELL_FAIL),
1563 	EF10_DMA_STAT(ctpio_overflow_fail, CTPIO_OVERFLOW_FAIL),
1564 	EF10_DMA_STAT(ctpio_underflow_fail, CTPIO_UNDERFLOW_FAIL),
1565 	EF10_DMA_STAT(ctpio_timeout_fail, CTPIO_TIMEOUT_FAIL),
1566 	EF10_DMA_STAT(ctpio_noncontig_wr_fail, CTPIO_NONCONTIG_WR_FAIL),
1567 	EF10_DMA_STAT(ctpio_frm_clobber_fail, CTPIO_FRM_CLOBBER_FAIL),
1568 	EF10_DMA_STAT(ctpio_invalid_wr_fail, CTPIO_INVALID_WR_FAIL),
1569 	EF10_DMA_STAT(ctpio_vi_clobber_fallback, CTPIO_VI_CLOBBER_FALLBACK),
1570 	EF10_DMA_STAT(ctpio_unqualified_fallback, CTPIO_UNQUALIFIED_FALLBACK),
1571 	EF10_DMA_STAT(ctpio_runt_fallback, CTPIO_RUNT_FALLBACK),
1572 	EF10_DMA_STAT(ctpio_success, CTPIO_SUCCESS),
1573 	EF10_DMA_STAT(ctpio_fallback, CTPIO_FALLBACK),
1574 	EF10_DMA_STAT(ctpio_poison, CTPIO_POISON),
1575 	EF10_DMA_STAT(ctpio_erase, CTPIO_ERASE),
1576 };
1577 
1578 #define HUNT_COMMON_STAT_MASK ((1ULL << EF10_STAT_port_tx_bytes) |	\
1579 			       (1ULL << EF10_STAT_port_tx_packets) |	\
1580 			       (1ULL << EF10_STAT_port_tx_pause) |	\
1581 			       (1ULL << EF10_STAT_port_tx_unicast) |	\
1582 			       (1ULL << EF10_STAT_port_tx_multicast) |	\
1583 			       (1ULL << EF10_STAT_port_tx_broadcast) |	\
1584 			       (1ULL << EF10_STAT_port_rx_bytes) |	\
1585 			       (1ULL <<                                 \
1586 				EF10_STAT_port_rx_bytes_minus_good_bytes) | \
1587 			       (1ULL << EF10_STAT_port_rx_good_bytes) |	\
1588 			       (1ULL << EF10_STAT_port_rx_bad_bytes) |	\
1589 			       (1ULL << EF10_STAT_port_rx_packets) |	\
1590 			       (1ULL << EF10_STAT_port_rx_good) |	\
1591 			       (1ULL << EF10_STAT_port_rx_bad) |	\
1592 			       (1ULL << EF10_STAT_port_rx_pause) |	\
1593 			       (1ULL << EF10_STAT_port_rx_control) |	\
1594 			       (1ULL << EF10_STAT_port_rx_unicast) |	\
1595 			       (1ULL << EF10_STAT_port_rx_multicast) |	\
1596 			       (1ULL << EF10_STAT_port_rx_broadcast) |	\
1597 			       (1ULL << EF10_STAT_port_rx_lt64) |	\
1598 			       (1ULL << EF10_STAT_port_rx_64) |		\
1599 			       (1ULL << EF10_STAT_port_rx_65_to_127) |	\
1600 			       (1ULL << EF10_STAT_port_rx_128_to_255) |	\
1601 			       (1ULL << EF10_STAT_port_rx_256_to_511) |	\
1602 			       (1ULL << EF10_STAT_port_rx_512_to_1023) |\
1603 			       (1ULL << EF10_STAT_port_rx_1024_to_15xx) |\
1604 			       (1ULL << EF10_STAT_port_rx_15xx_to_jumbo) |\
1605 			       (1ULL << EF10_STAT_port_rx_gtjumbo) |	\
1606 			       (1ULL << EF10_STAT_port_rx_bad_gtjumbo) |\
1607 			       (1ULL << EF10_STAT_port_rx_overflow) |	\
1608 			       (1ULL << EF10_STAT_port_rx_nodesc_drops) |\
1609 			       (1ULL << GENERIC_STAT_rx_nodesc_trunc) |	\
1610 			       (1ULL << GENERIC_STAT_rx_noskb_drops))
1611 
1612 /* On 7000 series NICs, these statistics are only provided by the 10G MAC.
1613  * For a 10G/40G switchable port we do not expose these because they might
1614  * not include all the packets they should.
1615  * On 8000 series NICs these statistics are always provided.
1616  */
1617 #define HUNT_10G_ONLY_STAT_MASK ((1ULL << EF10_STAT_port_tx_control) |	\
1618 				 (1ULL << EF10_STAT_port_tx_lt64) |	\
1619 				 (1ULL << EF10_STAT_port_tx_64) |	\
1620 				 (1ULL << EF10_STAT_port_tx_65_to_127) |\
1621 				 (1ULL << EF10_STAT_port_tx_128_to_255) |\
1622 				 (1ULL << EF10_STAT_port_tx_256_to_511) |\
1623 				 (1ULL << EF10_STAT_port_tx_512_to_1023) |\
1624 				 (1ULL << EF10_STAT_port_tx_1024_to_15xx) |\
1625 				 (1ULL << EF10_STAT_port_tx_15xx_to_jumbo))
1626 
1627 /* These statistics are only provided by the 40G MAC.  For a 10G/40G
1628  * switchable port we do expose these because the errors will otherwise
1629  * be silent.
1630  */
1631 #define HUNT_40G_EXTRA_STAT_MASK ((1ULL << EF10_STAT_port_rx_align_error) |\
1632 				  (1ULL << EF10_STAT_port_rx_length_error))
1633 
1634 /* These statistics are only provided if the firmware supports the
1635  * capability PM_AND_RXDP_COUNTERS.
1636  */
1637 #define HUNT_PM_AND_RXDP_STAT_MASK (					\
1638 	(1ULL << EF10_STAT_port_rx_pm_trunc_bb_overflow) |		\
1639 	(1ULL << EF10_STAT_port_rx_pm_discard_bb_overflow) |		\
1640 	(1ULL << EF10_STAT_port_rx_pm_trunc_vfifo_full) |		\
1641 	(1ULL << EF10_STAT_port_rx_pm_discard_vfifo_full) |		\
1642 	(1ULL << EF10_STAT_port_rx_pm_trunc_qbb) |			\
1643 	(1ULL << EF10_STAT_port_rx_pm_discard_qbb) |			\
1644 	(1ULL << EF10_STAT_port_rx_pm_discard_mapping) |		\
1645 	(1ULL << EF10_STAT_port_rx_dp_q_disabled_packets) |		\
1646 	(1ULL << EF10_STAT_port_rx_dp_di_dropped_packets) |		\
1647 	(1ULL << EF10_STAT_port_rx_dp_streaming_packets) |		\
1648 	(1ULL << EF10_STAT_port_rx_dp_hlb_fetch) |			\
1649 	(1ULL << EF10_STAT_port_rx_dp_hlb_wait))
1650 
1651 /* These statistics are only provided if the NIC supports MC_CMD_MAC_STATS_V2,
1652  * indicated by returning a value >= MC_CMD_MAC_NSTATS_V2 in
1653  * MC_CMD_GET_CAPABILITIES_V4_OUT_MAC_STATS_NUM_STATS.
1654  * These bits are in the second u64 of the raw mask.
1655  */
1656 #define EF10_FEC_STAT_MASK (						\
1657 	(1ULL << (EF10_STAT_fec_uncorrected_errors - 64)) |		\
1658 	(1ULL << (EF10_STAT_fec_corrected_errors - 64)) |		\
1659 	(1ULL << (EF10_STAT_fec_corrected_symbols_lane0 - 64)) |	\
1660 	(1ULL << (EF10_STAT_fec_corrected_symbols_lane1 - 64)) |	\
1661 	(1ULL << (EF10_STAT_fec_corrected_symbols_lane2 - 64)) |	\
1662 	(1ULL << (EF10_STAT_fec_corrected_symbols_lane3 - 64)))
1663 
1664 /* These statistics are only provided if the NIC supports MC_CMD_MAC_STATS_V3,
1665  * indicated by returning a value >= MC_CMD_MAC_NSTATS_V3 in
1666  * MC_CMD_GET_CAPABILITIES_V4_OUT_MAC_STATS_NUM_STATS.
1667  * These bits are in the second u64 of the raw mask.
1668  */
1669 #define EF10_CTPIO_STAT_MASK (						\
1670 	(1ULL << (EF10_STAT_ctpio_vi_busy_fallback - 64)) |		\
1671 	(1ULL << (EF10_STAT_ctpio_long_write_success - 64)) |		\
1672 	(1ULL << (EF10_STAT_ctpio_missing_dbell_fail - 64)) |		\
1673 	(1ULL << (EF10_STAT_ctpio_overflow_fail - 64)) |		\
1674 	(1ULL << (EF10_STAT_ctpio_underflow_fail - 64)) |		\
1675 	(1ULL << (EF10_STAT_ctpio_timeout_fail - 64)) |			\
1676 	(1ULL << (EF10_STAT_ctpio_noncontig_wr_fail - 64)) |		\
1677 	(1ULL << (EF10_STAT_ctpio_frm_clobber_fail - 64)) |		\
1678 	(1ULL << (EF10_STAT_ctpio_invalid_wr_fail - 64)) |		\
1679 	(1ULL << (EF10_STAT_ctpio_vi_clobber_fallback - 64)) |		\
1680 	(1ULL << (EF10_STAT_ctpio_unqualified_fallback - 64)) |		\
1681 	(1ULL << (EF10_STAT_ctpio_runt_fallback - 64)) |		\
1682 	(1ULL << (EF10_STAT_ctpio_success - 64)) |			\
1683 	(1ULL << (EF10_STAT_ctpio_fallback - 64)) |			\
1684 	(1ULL << (EF10_STAT_ctpio_poison - 64)) |			\
1685 	(1ULL << (EF10_STAT_ctpio_erase - 64)))
1686 
1687 static u64 efx_ef10_raw_stat_mask(struct efx_nic *efx)
1688 {
1689 	u64 raw_mask = HUNT_COMMON_STAT_MASK;
1690 	u32 port_caps = efx_mcdi_phy_get_caps(efx);
1691 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1692 
1693 	if (!(efx->mcdi->fn_flags &
1694 	      1 << MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_LINKCTRL))
1695 		return 0;
1696 
1697 	if (port_caps & (1 << MC_CMD_PHY_CAP_40000FDX_LBN)) {
1698 		raw_mask |= HUNT_40G_EXTRA_STAT_MASK;
1699 		/* 8000 series have everything even at 40G */
1700 		if (nic_data->datapath_caps2 &
1701 		    (1 << MC_CMD_GET_CAPABILITIES_V2_OUT_MAC_STATS_40G_TX_SIZE_BINS_LBN))
1702 			raw_mask |= HUNT_10G_ONLY_STAT_MASK;
1703 	} else {
1704 		raw_mask |= HUNT_10G_ONLY_STAT_MASK;
1705 	}
1706 
1707 	if (nic_data->datapath_caps &
1708 	    (1 << MC_CMD_GET_CAPABILITIES_OUT_PM_AND_RXDP_COUNTERS_LBN))
1709 		raw_mask |= HUNT_PM_AND_RXDP_STAT_MASK;
1710 
1711 	return raw_mask;
1712 }
1713 
1714 static void efx_ef10_get_stat_mask(struct efx_nic *efx, unsigned long *mask)
1715 {
1716 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1717 	u64 raw_mask[2];
1718 
1719 	raw_mask[0] = efx_ef10_raw_stat_mask(efx);
1720 
1721 	/* Only show vadaptor stats when EVB capability is present */
1722 	if (nic_data->datapath_caps &
1723 	    (1 << MC_CMD_GET_CAPABILITIES_OUT_EVB_LBN)) {
1724 		raw_mask[0] |= ~((1ULL << EF10_STAT_rx_unicast) - 1);
1725 		raw_mask[1] = (1ULL << (EF10_STAT_V1_COUNT - 64)) - 1;
1726 	} else {
1727 		raw_mask[1] = 0;
1728 	}
1729 	/* Only show FEC stats when NIC supports MC_CMD_MAC_STATS_V2 */
1730 	if (efx->num_mac_stats >= MC_CMD_MAC_NSTATS_V2)
1731 		raw_mask[1] |= EF10_FEC_STAT_MASK;
1732 
1733 	/* CTPIO stats appear in V3. Only show them on devices that actually
1734 	 * support CTPIO. Although this driver doesn't use CTPIO others might,
1735 	 * and we may be reporting the stats for the underlying port.
1736 	 */
1737 	if (efx->num_mac_stats >= MC_CMD_MAC_NSTATS_V3 &&
1738 	    (nic_data->datapath_caps2 &
1739 	     (1 << MC_CMD_GET_CAPABILITIES_V4_OUT_CTPIO_LBN)))
1740 		raw_mask[1] |= EF10_CTPIO_STAT_MASK;
1741 
1742 #if BITS_PER_LONG == 64
1743 	BUILD_BUG_ON(BITS_TO_LONGS(EF10_STAT_COUNT) != 2);
1744 	mask[0] = raw_mask[0];
1745 	mask[1] = raw_mask[1];
1746 #else
1747 	BUILD_BUG_ON(BITS_TO_LONGS(EF10_STAT_COUNT) != 3);
1748 	mask[0] = raw_mask[0] & 0xffffffff;
1749 	mask[1] = raw_mask[0] >> 32;
1750 	mask[2] = raw_mask[1] & 0xffffffff;
1751 #endif
1752 }
1753 
1754 static size_t efx_ef10_describe_stats(struct efx_nic *efx, u8 *names)
1755 {
1756 	DECLARE_BITMAP(mask, EF10_STAT_COUNT);
1757 
1758 	efx_ef10_get_stat_mask(efx, mask);
1759 	return efx_nic_describe_stats(efx_ef10_stat_desc, EF10_STAT_COUNT,
1760 				      mask, names);
1761 }
1762 
1763 static void efx_ef10_get_fec_stats(struct efx_nic *efx,
1764 				   struct ethtool_fec_stats *fec_stats)
1765 {
1766 	DECLARE_BITMAP(mask, EF10_STAT_COUNT);
1767 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1768 	u64 *stats = nic_data->stats;
1769 
1770 	efx_ef10_get_stat_mask(efx, mask);
1771 	if (test_bit(EF10_STAT_fec_corrected_errors, mask))
1772 		fec_stats->corrected_blocks.total =
1773 			stats[EF10_STAT_fec_corrected_errors];
1774 	if (test_bit(EF10_STAT_fec_uncorrected_errors, mask))
1775 		fec_stats->uncorrectable_blocks.total =
1776 			stats[EF10_STAT_fec_uncorrected_errors];
1777 }
1778 
1779 static size_t efx_ef10_update_stats_common(struct efx_nic *efx, u64 *full_stats,
1780 					   struct rtnl_link_stats64 *core_stats)
1781 {
1782 	DECLARE_BITMAP(mask, EF10_STAT_COUNT);
1783 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1784 	u64 *stats = nic_data->stats;
1785 	size_t stats_count = 0, index;
1786 
1787 	efx_ef10_get_stat_mask(efx, mask);
1788 
1789 	if (full_stats) {
1790 		for_each_set_bit(index, mask, EF10_STAT_COUNT) {
1791 			if (efx_ef10_stat_desc[index].name) {
1792 				*full_stats++ = stats[index];
1793 				++stats_count;
1794 			}
1795 		}
1796 	}
1797 
1798 	if (!core_stats)
1799 		return stats_count;
1800 
1801 	if (nic_data->datapath_caps &
1802 			1 << MC_CMD_GET_CAPABILITIES_OUT_EVB_LBN) {
1803 		/* Use vadaptor stats. */
1804 		core_stats->rx_packets = stats[EF10_STAT_rx_unicast] +
1805 					 stats[EF10_STAT_rx_multicast] +
1806 					 stats[EF10_STAT_rx_broadcast];
1807 		core_stats->tx_packets = stats[EF10_STAT_tx_unicast] +
1808 					 stats[EF10_STAT_tx_multicast] +
1809 					 stats[EF10_STAT_tx_broadcast];
1810 		core_stats->rx_bytes = stats[EF10_STAT_rx_unicast_bytes] +
1811 				       stats[EF10_STAT_rx_multicast_bytes] +
1812 				       stats[EF10_STAT_rx_broadcast_bytes];
1813 		core_stats->tx_bytes = stats[EF10_STAT_tx_unicast_bytes] +
1814 				       stats[EF10_STAT_tx_multicast_bytes] +
1815 				       stats[EF10_STAT_tx_broadcast_bytes];
1816 		core_stats->rx_dropped = stats[GENERIC_STAT_rx_nodesc_trunc] +
1817 					 stats[GENERIC_STAT_rx_noskb_drops];
1818 		core_stats->multicast = stats[EF10_STAT_rx_multicast];
1819 		core_stats->rx_crc_errors = stats[EF10_STAT_rx_bad];
1820 		core_stats->rx_fifo_errors = stats[EF10_STAT_rx_overflow];
1821 		core_stats->rx_errors = core_stats->rx_crc_errors;
1822 		core_stats->tx_errors = stats[EF10_STAT_tx_bad];
1823 	} else {
1824 		/* Use port stats. */
1825 		core_stats->rx_packets = stats[EF10_STAT_port_rx_packets];
1826 		core_stats->tx_packets = stats[EF10_STAT_port_tx_packets];
1827 		core_stats->rx_bytes = stats[EF10_STAT_port_rx_bytes];
1828 		core_stats->tx_bytes = stats[EF10_STAT_port_tx_bytes];
1829 		core_stats->rx_dropped = stats[EF10_STAT_port_rx_nodesc_drops] +
1830 					 stats[GENERIC_STAT_rx_nodesc_trunc] +
1831 					 stats[GENERIC_STAT_rx_noskb_drops];
1832 		core_stats->multicast = stats[EF10_STAT_port_rx_multicast];
1833 		core_stats->rx_length_errors =
1834 				stats[EF10_STAT_port_rx_gtjumbo] +
1835 				stats[EF10_STAT_port_rx_length_error];
1836 		core_stats->rx_crc_errors = stats[EF10_STAT_port_rx_bad];
1837 		core_stats->rx_frame_errors =
1838 				stats[EF10_STAT_port_rx_align_error];
1839 		core_stats->rx_fifo_errors = stats[EF10_STAT_port_rx_overflow];
1840 		core_stats->rx_errors = (core_stats->rx_length_errors +
1841 					 core_stats->rx_crc_errors +
1842 					 core_stats->rx_frame_errors);
1843 	}
1844 
1845 	return stats_count;
1846 }
1847 
1848 static size_t efx_ef10_update_stats_pf(struct efx_nic *efx, u64 *full_stats,
1849 				       struct rtnl_link_stats64 *core_stats)
1850 {
1851 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1852 	DECLARE_BITMAP(mask, EF10_STAT_COUNT);
1853 	u64 *stats = nic_data->stats;
1854 
1855 	efx_ef10_get_stat_mask(efx, mask);
1856 
1857 	/* If NIC was fini'd (probably resetting), then we can't read
1858 	 * updated stats right now.
1859 	 */
1860 	if (nic_data->mc_stats) {
1861 		efx_nic_copy_stats(efx, nic_data->mc_stats);
1862 		efx_nic_update_stats(efx_ef10_stat_desc, EF10_STAT_COUNT,
1863 				     mask, stats, nic_data->mc_stats, false);
1864 	}
1865 
1866 	/* Update derived statistics */
1867 	efx_nic_fix_nodesc_drop_stat(efx,
1868 				     &stats[EF10_STAT_port_rx_nodesc_drops]);
1869 	/* MC Firmware reads RX_BYTES and RX_GOOD_BYTES from the MAC.
1870 	 * It then calculates RX_BAD_BYTES and DMAs it to us with RX_BYTES.
1871 	 * We report these as port_rx_ stats. We are not given RX_GOOD_BYTES.
1872 	 * Here we calculate port_rx_good_bytes.
1873 	 */
1874 	stats[EF10_STAT_port_rx_good_bytes] =
1875 		stats[EF10_STAT_port_rx_bytes] -
1876 		stats[EF10_STAT_port_rx_bytes_minus_good_bytes];
1877 
1878 	/* The asynchronous reads used to calculate RX_BAD_BYTES in
1879 	 * MC Firmware are done such that we should not see an increase in
1880 	 * RX_BAD_BYTES when a good packet has arrived. Unfortunately this
1881 	 * does mean that the stat can decrease at times. Here we do not
1882 	 * update the stat unless it has increased or has gone to zero
1883 	 * (In the case of the NIC rebooting).
1884 	 * Please see Bug 33781 for a discussion of why things work this way.
1885 	 */
1886 	efx_update_diff_stat(&stats[EF10_STAT_port_rx_bad_bytes],
1887 			     stats[EF10_STAT_port_rx_bytes_minus_good_bytes]);
1888 	efx_update_sw_stats(efx, stats);
1889 
1890 	return efx_ef10_update_stats_common(efx, full_stats, core_stats);
1891 }
1892 
1893 static int efx_ef10_try_update_nic_stats_vf(struct efx_nic *efx)
1894 	__must_hold(&efx->stats_lock)
1895 {
1896 	MCDI_DECLARE_BUF(inbuf, MC_CMD_MAC_STATS_IN_LEN);
1897 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1898 	DECLARE_BITMAP(mask, EF10_STAT_COUNT);
1899 	__le64 generation_start, generation_end;
1900 	u64 *stats = nic_data->stats;
1901 	u32 dma_len = efx->num_mac_stats * sizeof(u64);
1902 	struct efx_buffer stats_buf;
1903 	__le64 *dma_stats;
1904 	int rc;
1905 
1906 	spin_unlock_bh(&efx->stats_lock);
1907 
1908 	efx_ef10_get_stat_mask(efx, mask);
1909 
1910 	rc = efx_nic_alloc_buffer(efx, &stats_buf, dma_len, GFP_KERNEL);
1911 	if (rc) {
1912 		spin_lock_bh(&efx->stats_lock);
1913 		return rc;
1914 	}
1915 
1916 	dma_stats = stats_buf.addr;
1917 	dma_stats[efx->num_mac_stats - 1] = EFX_MC_STATS_GENERATION_INVALID;
1918 
1919 	MCDI_SET_QWORD(inbuf, MAC_STATS_IN_DMA_ADDR, stats_buf.dma_addr);
1920 	MCDI_POPULATE_DWORD_1(inbuf, MAC_STATS_IN_CMD,
1921 			      MAC_STATS_IN_DMA, 1);
1922 	MCDI_SET_DWORD(inbuf, MAC_STATS_IN_DMA_LEN, dma_len);
1923 	MCDI_SET_DWORD(inbuf, MAC_STATS_IN_PORT_ID, EVB_PORT_ID_ASSIGNED);
1924 
1925 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_MAC_STATS, inbuf, sizeof(inbuf),
1926 				NULL, 0, NULL);
1927 	spin_lock_bh(&efx->stats_lock);
1928 	if (rc) {
1929 		/* Expect ENOENT if DMA queues have not been set up */
1930 		if (rc != -ENOENT || atomic_read(&efx->active_queues))
1931 			efx_mcdi_display_error(efx, MC_CMD_MAC_STATS,
1932 					       sizeof(inbuf), NULL, 0, rc);
1933 		goto out;
1934 	}
1935 
1936 	generation_end = dma_stats[efx->num_mac_stats - 1];
1937 	if (generation_end == EFX_MC_STATS_GENERATION_INVALID) {
1938 		WARN_ON_ONCE(1);
1939 		goto out;
1940 	}
1941 	rmb();
1942 	efx_nic_update_stats(efx_ef10_stat_desc, EF10_STAT_COUNT, mask,
1943 			     stats, stats_buf.addr, false);
1944 	rmb();
1945 	generation_start = dma_stats[MC_CMD_MAC_GENERATION_START];
1946 	if (generation_end != generation_start) {
1947 		rc = -EAGAIN;
1948 		goto out;
1949 	}
1950 
1951 	efx_update_sw_stats(efx, stats);
1952 out:
1953 	/* releasing a DMA coherent buffer with BH disabled can panic */
1954 	spin_unlock_bh(&efx->stats_lock);
1955 	efx_nic_free_buffer(efx, &stats_buf);
1956 	spin_lock_bh(&efx->stats_lock);
1957 	return rc;
1958 }
1959 
1960 static size_t efx_ef10_update_stats_vf(struct efx_nic *efx, u64 *full_stats,
1961 				       struct rtnl_link_stats64 *core_stats)
1962 {
1963 	if (efx_ef10_try_update_nic_stats_vf(efx))
1964 		return 0;
1965 
1966 	return efx_ef10_update_stats_common(efx, full_stats, core_stats);
1967 }
1968 
1969 static size_t efx_ef10_update_stats_atomic_vf(struct efx_nic *efx, u64 *full_stats,
1970 					      struct rtnl_link_stats64 *core_stats)
1971 {
1972 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1973 
1974 	/* In atomic context, cannot update HW stats.  Just update the
1975 	 * software stats and return so the caller can continue.
1976 	 */
1977 	efx_update_sw_stats(efx, nic_data->stats);
1978 	return efx_ef10_update_stats_common(efx, full_stats, core_stats);
1979 }
1980 
1981 static void efx_ef10_push_irq_moderation(struct efx_channel *channel)
1982 {
1983 	struct efx_nic *efx = channel->efx;
1984 	unsigned int mode, usecs;
1985 	efx_dword_t timer_cmd;
1986 
1987 	if (channel->irq_moderation_us) {
1988 		mode = 3;
1989 		usecs = channel->irq_moderation_us;
1990 	} else {
1991 		mode = 0;
1992 		usecs = 0;
1993 	}
1994 
1995 	if (EFX_EF10_WORKAROUND_61265(efx)) {
1996 		MCDI_DECLARE_BUF(inbuf, MC_CMD_SET_EVQ_TMR_IN_LEN);
1997 		unsigned int ns = usecs * 1000;
1998 
1999 		MCDI_SET_DWORD(inbuf, SET_EVQ_TMR_IN_INSTANCE,
2000 			       channel->channel);
2001 		MCDI_SET_DWORD(inbuf, SET_EVQ_TMR_IN_TMR_LOAD_REQ_NS, ns);
2002 		MCDI_SET_DWORD(inbuf, SET_EVQ_TMR_IN_TMR_RELOAD_REQ_NS, ns);
2003 		MCDI_SET_DWORD(inbuf, SET_EVQ_TMR_IN_TMR_MODE, mode);
2004 
2005 		efx_mcdi_rpc_async(efx, MC_CMD_SET_EVQ_TMR,
2006 				   inbuf, sizeof(inbuf), 0, NULL, 0);
2007 	} else if (EFX_EF10_WORKAROUND_35388(efx)) {
2008 		unsigned int ticks = efx_usecs_to_ticks(efx, usecs);
2009 
2010 		EFX_POPULATE_DWORD_3(timer_cmd, ERF_DD_EVQ_IND_TIMER_FLAGS,
2011 				     EFE_DD_EVQ_IND_TIMER_FLAGS,
2012 				     ERF_DD_EVQ_IND_TIMER_MODE, mode,
2013 				     ERF_DD_EVQ_IND_TIMER_VAL, ticks);
2014 		efx_writed_page(efx, &timer_cmd, ER_DD_EVQ_INDIRECT,
2015 				channel->channel);
2016 	} else {
2017 		unsigned int ticks = efx_usecs_to_ticks(efx, usecs);
2018 
2019 		EFX_POPULATE_DWORD_3(timer_cmd, ERF_DZ_TC_TIMER_MODE, mode,
2020 				     ERF_DZ_TC_TIMER_VAL, ticks,
2021 				     ERF_FZ_TC_TMR_REL_VAL, ticks);
2022 		efx_writed_page(efx, &timer_cmd, ER_DZ_EVQ_TMR,
2023 				channel->channel);
2024 	}
2025 }
2026 
2027 static void efx_ef10_get_wol_vf(struct efx_nic *efx,
2028 				struct ethtool_wolinfo *wol) {}
2029 
2030 static int efx_ef10_set_wol_vf(struct efx_nic *efx, u32 type)
2031 {
2032 	return -EOPNOTSUPP;
2033 }
2034 
2035 static void efx_ef10_get_wol(struct efx_nic *efx, struct ethtool_wolinfo *wol)
2036 {
2037 	wol->supported = 0;
2038 	wol->wolopts = 0;
2039 	memset(&wol->sopass, 0, sizeof(wol->sopass));
2040 }
2041 
2042 static int efx_ef10_set_wol(struct efx_nic *efx, u32 type)
2043 {
2044 	if (type != 0)
2045 		return -EINVAL;
2046 	return 0;
2047 }
2048 
2049 static void efx_ef10_mcdi_request(struct efx_nic *efx,
2050 				  const efx_dword_t *hdr, size_t hdr_len,
2051 				  const efx_dword_t *sdu, size_t sdu_len)
2052 {
2053 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2054 	u8 *pdu = nic_data->mcdi_buf.addr;
2055 
2056 	memcpy(pdu, hdr, hdr_len);
2057 	memcpy(pdu + hdr_len, sdu, sdu_len);
2058 	wmb();
2059 
2060 	/* The hardware provides 'low' and 'high' (doorbell) registers
2061 	 * for passing the 64-bit address of an MCDI request to
2062 	 * firmware.  However the dwords are swapped by firmware.  The
2063 	 * least significant bits of the doorbell are then 0 for all
2064 	 * MCDI requests due to alignment.
2065 	 */
2066 	_efx_writed(efx, cpu_to_le32((u64)nic_data->mcdi_buf.dma_addr >> 32),
2067 		    ER_DZ_MC_DB_LWRD);
2068 	_efx_writed(efx, cpu_to_le32((u32)nic_data->mcdi_buf.dma_addr),
2069 		    ER_DZ_MC_DB_HWRD);
2070 }
2071 
2072 static bool efx_ef10_mcdi_poll_response(struct efx_nic *efx)
2073 {
2074 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2075 	const efx_dword_t hdr = *(const efx_dword_t *)nic_data->mcdi_buf.addr;
2076 
2077 	rmb();
2078 	return EFX_DWORD_FIELD(hdr, MCDI_HEADER_RESPONSE);
2079 }
2080 
2081 static void
2082 efx_ef10_mcdi_read_response(struct efx_nic *efx, efx_dword_t *outbuf,
2083 			    size_t offset, size_t outlen)
2084 {
2085 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2086 	const u8 *pdu = nic_data->mcdi_buf.addr;
2087 
2088 	memcpy(outbuf, pdu + offset, outlen);
2089 }
2090 
2091 static void efx_ef10_mcdi_reboot_detected(struct efx_nic *efx)
2092 {
2093 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2094 
2095 	/* All our allocations have been reset */
2096 	efx_ef10_table_reset_mc_allocations(efx);
2097 
2098 	/* The datapath firmware might have been changed */
2099 	nic_data->must_check_datapath_caps = true;
2100 
2101 	/* MAC statistics have been cleared on the NIC; clear the local
2102 	 * statistic that we update with efx_update_diff_stat().
2103 	 */
2104 	nic_data->stats[EF10_STAT_port_rx_bad_bytes] = 0;
2105 }
2106 
2107 static int efx_ef10_mcdi_poll_reboot(struct efx_nic *efx)
2108 {
2109 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2110 	int rc;
2111 
2112 	rc = efx_ef10_get_warm_boot_count(efx);
2113 	if (rc < 0) {
2114 		/* The firmware is presumably in the process of
2115 		 * rebooting.  However, we are supposed to report each
2116 		 * reboot just once, so we must only do that once we
2117 		 * can read and store the updated warm boot count.
2118 		 */
2119 		return 0;
2120 	}
2121 
2122 	if (rc == nic_data->warm_boot_count)
2123 		return 0;
2124 
2125 	nic_data->warm_boot_count = rc;
2126 	efx_ef10_mcdi_reboot_detected(efx);
2127 
2128 	return -EIO;
2129 }
2130 
2131 /* Handle an MSI interrupt
2132  *
2133  * Handle an MSI hardware interrupt.  This routine schedules event
2134  * queue processing.  No interrupt acknowledgement cycle is necessary.
2135  * Also, we never need to check that the interrupt is for us, since
2136  * MSI interrupts cannot be shared.
2137  */
2138 static irqreturn_t efx_ef10_msi_interrupt(int irq, void *dev_id)
2139 {
2140 	struct efx_msi_context *context = dev_id;
2141 	struct efx_nic *efx = context->efx;
2142 
2143 	netif_vdbg(efx, intr, efx->net_dev,
2144 		   "IRQ %d on CPU %d\n", irq, raw_smp_processor_id());
2145 
2146 	if (likely(READ_ONCE(efx->irq_soft_enabled))) {
2147 		/* Note test interrupts */
2148 		if (context->index == efx->irq_level)
2149 			efx->last_irq_cpu = raw_smp_processor_id();
2150 
2151 		/* Schedule processing of the channel */
2152 		efx_schedule_channel_irq(efx->channel[context->index]);
2153 	}
2154 
2155 	return IRQ_HANDLED;
2156 }
2157 
2158 static irqreturn_t efx_ef10_legacy_interrupt(int irq, void *dev_id)
2159 {
2160 	struct efx_nic *efx = dev_id;
2161 	bool soft_enabled = READ_ONCE(efx->irq_soft_enabled);
2162 	struct efx_channel *channel;
2163 	efx_dword_t reg;
2164 	u32 queues;
2165 
2166 	/* Read the ISR which also ACKs the interrupts */
2167 	efx_readd(efx, &reg, ER_DZ_BIU_INT_ISR);
2168 	queues = EFX_DWORD_FIELD(reg, ERF_DZ_ISR_REG);
2169 
2170 	if (queues == 0)
2171 		return IRQ_NONE;
2172 
2173 	if (likely(soft_enabled)) {
2174 		/* Note test interrupts */
2175 		if (queues & (1U << efx->irq_level))
2176 			efx->last_irq_cpu = raw_smp_processor_id();
2177 
2178 		efx_for_each_channel(channel, efx) {
2179 			if (queues & 1)
2180 				efx_schedule_channel_irq(channel);
2181 			queues >>= 1;
2182 		}
2183 	}
2184 
2185 	netif_vdbg(efx, intr, efx->net_dev,
2186 		   "IRQ %d on CPU %d status " EFX_DWORD_FMT "\n",
2187 		   irq, raw_smp_processor_id(), EFX_DWORD_VAL(reg));
2188 
2189 	return IRQ_HANDLED;
2190 }
2191 
2192 static int efx_ef10_irq_test_generate(struct efx_nic *efx)
2193 {
2194 	MCDI_DECLARE_BUF(inbuf, MC_CMD_TRIGGER_INTERRUPT_IN_LEN);
2195 
2196 	if (efx_mcdi_set_workaround(efx, MC_CMD_WORKAROUND_BUG41750, true,
2197 				    NULL) == 0)
2198 		return -ENOTSUPP;
2199 
2200 	BUILD_BUG_ON(MC_CMD_TRIGGER_INTERRUPT_OUT_LEN != 0);
2201 
2202 	MCDI_SET_DWORD(inbuf, TRIGGER_INTERRUPT_IN_INTR_LEVEL, efx->irq_level);
2203 	return efx_mcdi_rpc(efx, MC_CMD_TRIGGER_INTERRUPT,
2204 			    inbuf, sizeof(inbuf), NULL, 0, NULL);
2205 }
2206 
2207 static int efx_ef10_tx_probe(struct efx_tx_queue *tx_queue)
2208 {
2209 	/* low two bits of label are what we want for type */
2210 	BUILD_BUG_ON((EFX_TXQ_TYPE_OUTER_CSUM | EFX_TXQ_TYPE_INNER_CSUM) != 3);
2211 	tx_queue->type = tx_queue->label & 3;
2212 	return efx_nic_alloc_buffer(tx_queue->efx, &tx_queue->txd,
2213 				    (tx_queue->ptr_mask + 1) *
2214 				    sizeof(efx_qword_t),
2215 				    GFP_KERNEL);
2216 }
2217 
2218 /* This writes to the TX_DESC_WPTR and also pushes data */
2219 static inline void efx_ef10_push_tx_desc(struct efx_tx_queue *tx_queue,
2220 					 const efx_qword_t *txd)
2221 {
2222 	unsigned int write_ptr;
2223 	efx_oword_t reg;
2224 
2225 	write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
2226 	EFX_POPULATE_OWORD_1(reg, ERF_DZ_TX_DESC_WPTR, write_ptr);
2227 	reg.qword[0] = *txd;
2228 	efx_writeo_page(tx_queue->efx, &reg,
2229 			ER_DZ_TX_DESC_UPD, tx_queue->queue);
2230 }
2231 
2232 /* Add Firmware-Assisted TSO v2 option descriptors to a queue.
2233  */
2234 int efx_ef10_tx_tso_desc(struct efx_tx_queue *tx_queue, struct sk_buff *skb,
2235 			 bool *data_mapped)
2236 {
2237 	struct efx_tx_buffer *buffer;
2238 	u16 inner_ipv4_id = 0;
2239 	u16 outer_ipv4_id = 0;
2240 	struct tcphdr *tcp;
2241 	struct iphdr *ip;
2242 	u16 ip_tot_len;
2243 	u32 seqnum;
2244 	u32 mss;
2245 
2246 	EFX_WARN_ON_ONCE_PARANOID(tx_queue->tso_version != 2);
2247 
2248 	mss = skb_shinfo(skb)->gso_size;
2249 
2250 	if (unlikely(mss < 4)) {
2251 		WARN_ONCE(1, "MSS of %u is too small for TSO v2\n", mss);
2252 		return -EINVAL;
2253 	}
2254 
2255 	if (skb->encapsulation) {
2256 		if (!tx_queue->tso_encap)
2257 			return -EINVAL;
2258 		ip = ip_hdr(skb);
2259 		if (ip->version == 4)
2260 			outer_ipv4_id = ntohs(ip->id);
2261 
2262 		ip = inner_ip_hdr(skb);
2263 		tcp = inner_tcp_hdr(skb);
2264 	} else {
2265 		ip = ip_hdr(skb);
2266 		tcp = tcp_hdr(skb);
2267 	}
2268 
2269 	/* 8000-series EF10 hardware requires that IP Total Length be
2270 	 * greater than or equal to the value it will have in each segment
2271 	 * (which is at most mss + 208 + TCP header length), but also less
2272 	 * than (0x10000 - inner_network_header).  Otherwise the TCP
2273 	 * checksum calculation will be broken for encapsulated packets.
2274 	 * We fill in ip->tot_len with 0xff30, which should satisfy the
2275 	 * first requirement unless the MSS is ridiculously large (which
2276 	 * should be impossible as the driver max MTU is 9216); it is
2277 	 * guaranteed to satisfy the second as we only attempt TSO if
2278 	 * inner_network_header <= 208.
2279 	 */
2280 	ip_tot_len = 0x10000 - EFX_TSO2_MAX_HDRLEN;
2281 	EFX_WARN_ON_ONCE_PARANOID(mss + EFX_TSO2_MAX_HDRLEN +
2282 				  (tcp->doff << 2u) > ip_tot_len);
2283 
2284 	if (ip->version == 4) {
2285 		ip->tot_len = htons(ip_tot_len);
2286 		ip->check = 0;
2287 		inner_ipv4_id = ntohs(ip->id);
2288 	} else {
2289 		((struct ipv6hdr *)ip)->payload_len = htons(ip_tot_len);
2290 	}
2291 
2292 	seqnum = ntohl(tcp->seq);
2293 
2294 	buffer = efx_tx_queue_get_insert_buffer(tx_queue);
2295 
2296 	buffer->flags = EFX_TX_BUF_OPTION;
2297 	buffer->len = 0;
2298 	buffer->unmap_len = 0;
2299 	EFX_POPULATE_QWORD_5(buffer->option,
2300 			ESF_DZ_TX_DESC_IS_OPT, 1,
2301 			ESF_DZ_TX_OPTION_TYPE, ESE_DZ_TX_OPTION_DESC_TSO,
2302 			ESF_DZ_TX_TSO_OPTION_TYPE,
2303 			ESE_DZ_TX_TSO_OPTION_DESC_FATSO2A,
2304 			ESF_DZ_TX_TSO_IP_ID, inner_ipv4_id,
2305 			ESF_DZ_TX_TSO_TCP_SEQNO, seqnum
2306 			);
2307 	++tx_queue->insert_count;
2308 
2309 	buffer = efx_tx_queue_get_insert_buffer(tx_queue);
2310 
2311 	buffer->flags = EFX_TX_BUF_OPTION;
2312 	buffer->len = 0;
2313 	buffer->unmap_len = 0;
2314 	EFX_POPULATE_QWORD_5(buffer->option,
2315 			ESF_DZ_TX_DESC_IS_OPT, 1,
2316 			ESF_DZ_TX_OPTION_TYPE, ESE_DZ_TX_OPTION_DESC_TSO,
2317 			ESF_DZ_TX_TSO_OPTION_TYPE,
2318 			ESE_DZ_TX_TSO_OPTION_DESC_FATSO2B,
2319 			ESF_DZ_TX_TSO_OUTER_IPID, outer_ipv4_id,
2320 			ESF_DZ_TX_TSO_TCP_MSS, mss
2321 			);
2322 	++tx_queue->insert_count;
2323 
2324 	return 0;
2325 }
2326 
2327 static u32 efx_ef10_tso_versions(struct efx_nic *efx)
2328 {
2329 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2330 	u32 tso_versions = 0;
2331 
2332 	if (nic_data->datapath_caps &
2333 	    (1 << MC_CMD_GET_CAPABILITIES_OUT_TX_TSO_LBN))
2334 		tso_versions |= BIT(1);
2335 	if (nic_data->datapath_caps2 &
2336 	    (1 << MC_CMD_GET_CAPABILITIES_V2_OUT_TX_TSO_V2_LBN))
2337 		tso_versions |= BIT(2);
2338 	return tso_versions;
2339 }
2340 
2341 static void efx_ef10_tx_init(struct efx_tx_queue *tx_queue)
2342 {
2343 	bool csum_offload = tx_queue->type & EFX_TXQ_TYPE_OUTER_CSUM;
2344 	bool inner_csum = tx_queue->type & EFX_TXQ_TYPE_INNER_CSUM;
2345 	struct efx_channel *channel = tx_queue->channel;
2346 	struct efx_nic *efx = tx_queue->efx;
2347 	struct efx_ef10_nic_data *nic_data;
2348 	efx_qword_t *txd;
2349 	int rc;
2350 
2351 	nic_data = efx->nic_data;
2352 
2353 	/* Only attempt to enable TX timestamping if we have the license for it,
2354 	 * otherwise TXQ init will fail
2355 	 */
2356 	if (!(nic_data->licensed_features &
2357 	      (1 << LICENSED_V3_FEATURES_TX_TIMESTAMPS_LBN))) {
2358 		tx_queue->timestamping = false;
2359 		/* Disable sync events on this channel. */
2360 		if (efx->type->ptp_set_ts_sync_events)
2361 			efx->type->ptp_set_ts_sync_events(efx, false, false);
2362 	}
2363 
2364 	/* TSOv2 is a limited resource that can only be configured on a limited
2365 	 * number of queues. TSO without checksum offload is not really a thing,
2366 	 * so we only enable it for those queues.
2367 	 * TSOv2 cannot be used with Hardware timestamping, and is never needed
2368 	 * for XDP tx.
2369 	 */
2370 	if (efx_has_cap(efx, TX_TSO_V2)) {
2371 		if ((csum_offload || inner_csum) &&
2372 		    !tx_queue->timestamping && !tx_queue->xdp_tx) {
2373 			tx_queue->tso_version = 2;
2374 			netif_dbg(efx, hw, efx->net_dev, "Using TSOv2 for channel %u\n",
2375 				  channel->channel);
2376 		}
2377 	} else if (efx_has_cap(efx, TX_TSO)) {
2378 		tx_queue->tso_version = 1;
2379 	}
2380 
2381 	rc = efx_mcdi_tx_init(tx_queue);
2382 	if (rc)
2383 		goto fail;
2384 
2385 	/* A previous user of this TX queue might have set us up the
2386 	 * bomb by writing a descriptor to the TX push collector but
2387 	 * not the doorbell.  (Each collector belongs to a port, not a
2388 	 * queue or function, so cannot easily be reset.)  We must
2389 	 * attempt to push a no-op descriptor in its place.
2390 	 */
2391 	tx_queue->buffer[0].flags = EFX_TX_BUF_OPTION;
2392 	tx_queue->insert_count = 1;
2393 	txd = efx_tx_desc(tx_queue, 0);
2394 	EFX_POPULATE_QWORD_7(*txd,
2395 			     ESF_DZ_TX_DESC_IS_OPT, true,
2396 			     ESF_DZ_TX_OPTION_TYPE,
2397 			     ESE_DZ_TX_OPTION_DESC_CRC_CSUM,
2398 			     ESF_DZ_TX_OPTION_UDP_TCP_CSUM, csum_offload,
2399 			     ESF_DZ_TX_OPTION_IP_CSUM, csum_offload && tx_queue->tso_version != 2,
2400 			     ESF_DZ_TX_OPTION_INNER_UDP_TCP_CSUM, inner_csum,
2401 			     ESF_DZ_TX_OPTION_INNER_IP_CSUM, inner_csum && tx_queue->tso_version != 2,
2402 			     ESF_DZ_TX_TIMESTAMP, tx_queue->timestamping);
2403 	tx_queue->write_count = 1;
2404 
2405 	if (tx_queue->tso_version == 2 && efx_has_cap(efx, TX_TSO_V2_ENCAP))
2406 		tx_queue->tso_encap = true;
2407 
2408 	wmb();
2409 	efx_ef10_push_tx_desc(tx_queue, txd);
2410 
2411 	return;
2412 
2413 fail:
2414 	netdev_WARN(efx->net_dev, "failed to initialise TXQ %d\n",
2415 		    tx_queue->queue);
2416 }
2417 
2418 /* This writes to the TX_DESC_WPTR; write pointer for TX descriptor ring */
2419 static inline void efx_ef10_notify_tx_desc(struct efx_tx_queue *tx_queue)
2420 {
2421 	unsigned int write_ptr;
2422 	efx_dword_t reg;
2423 
2424 	write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
2425 	EFX_POPULATE_DWORD_1(reg, ERF_DZ_TX_DESC_WPTR_DWORD, write_ptr);
2426 	efx_writed_page(tx_queue->efx, &reg,
2427 			ER_DZ_TX_DESC_UPD_DWORD, tx_queue->queue);
2428 }
2429 
2430 #define EFX_EF10_MAX_TX_DESCRIPTOR_LEN 0x3fff
2431 
2432 static unsigned int efx_ef10_tx_limit_len(struct efx_tx_queue *tx_queue,
2433 					  dma_addr_t dma_addr, unsigned int len)
2434 {
2435 	if (len > EFX_EF10_MAX_TX_DESCRIPTOR_LEN) {
2436 		/* If we need to break across multiple descriptors we should
2437 		 * stop at a page boundary. This assumes the length limit is
2438 		 * greater than the page size.
2439 		 */
2440 		dma_addr_t end = dma_addr + EFX_EF10_MAX_TX_DESCRIPTOR_LEN;
2441 
2442 		BUILD_BUG_ON(EFX_EF10_MAX_TX_DESCRIPTOR_LEN < EFX_PAGE_SIZE);
2443 		len = (end & (~(EFX_PAGE_SIZE - 1))) - dma_addr;
2444 	}
2445 
2446 	return len;
2447 }
2448 
2449 static void efx_ef10_tx_write(struct efx_tx_queue *tx_queue)
2450 {
2451 	unsigned int old_write_count = tx_queue->write_count;
2452 	struct efx_tx_buffer *buffer;
2453 	unsigned int write_ptr;
2454 	efx_qword_t *txd;
2455 
2456 	tx_queue->xmit_pending = false;
2457 	if (unlikely(tx_queue->write_count == tx_queue->insert_count))
2458 		return;
2459 
2460 	do {
2461 		write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
2462 		buffer = &tx_queue->buffer[write_ptr];
2463 		txd = efx_tx_desc(tx_queue, write_ptr);
2464 		++tx_queue->write_count;
2465 
2466 		/* Create TX descriptor ring entry */
2467 		if (buffer->flags & EFX_TX_BUF_OPTION) {
2468 			*txd = buffer->option;
2469 			if (EFX_QWORD_FIELD(*txd, ESF_DZ_TX_OPTION_TYPE) == 1)
2470 				/* PIO descriptor */
2471 				tx_queue->packet_write_count = tx_queue->write_count;
2472 		} else {
2473 			tx_queue->packet_write_count = tx_queue->write_count;
2474 			BUILD_BUG_ON(EFX_TX_BUF_CONT != 1);
2475 			EFX_POPULATE_QWORD_3(
2476 				*txd,
2477 				ESF_DZ_TX_KER_CONT,
2478 				buffer->flags & EFX_TX_BUF_CONT,
2479 				ESF_DZ_TX_KER_BYTE_CNT, buffer->len,
2480 				ESF_DZ_TX_KER_BUF_ADDR, buffer->dma_addr);
2481 		}
2482 	} while (tx_queue->write_count != tx_queue->insert_count);
2483 
2484 	wmb(); /* Ensure descriptors are written before they are fetched */
2485 
2486 	if (efx_nic_may_push_tx_desc(tx_queue, old_write_count)) {
2487 		txd = efx_tx_desc(tx_queue,
2488 				  old_write_count & tx_queue->ptr_mask);
2489 		efx_ef10_push_tx_desc(tx_queue, txd);
2490 		++tx_queue->pushes;
2491 	} else {
2492 		efx_ef10_notify_tx_desc(tx_queue);
2493 	}
2494 }
2495 
2496 static int efx_ef10_probe_multicast_chaining(struct efx_nic *efx)
2497 {
2498 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2499 	unsigned int enabled, implemented;
2500 	bool want_workaround_26807;
2501 	int rc;
2502 
2503 	rc = efx_mcdi_get_workarounds(efx, &implemented, &enabled);
2504 	if (rc == -ENOSYS) {
2505 		/* GET_WORKAROUNDS was implemented before this workaround,
2506 		 * thus it must be unavailable in this firmware.
2507 		 */
2508 		nic_data->workaround_26807 = false;
2509 		return 0;
2510 	}
2511 	if (rc)
2512 		return rc;
2513 	want_workaround_26807 =
2514 		implemented & MC_CMD_GET_WORKAROUNDS_OUT_BUG26807;
2515 	nic_data->workaround_26807 =
2516 		!!(enabled & MC_CMD_GET_WORKAROUNDS_OUT_BUG26807);
2517 
2518 	if (want_workaround_26807 && !nic_data->workaround_26807) {
2519 		unsigned int flags;
2520 
2521 		rc = efx_mcdi_set_workaround(efx,
2522 					     MC_CMD_WORKAROUND_BUG26807,
2523 					     true, &flags);
2524 		if (!rc) {
2525 			if (flags &
2526 			    1 << MC_CMD_WORKAROUND_EXT_OUT_FLR_DONE_LBN) {
2527 				netif_info(efx, drv, efx->net_dev,
2528 					   "other functions on NIC have been reset\n");
2529 
2530 				/* With MCFW v4.6.x and earlier, the
2531 				 * boot count will have incremented,
2532 				 * so re-read the warm_boot_count
2533 				 * value now to ensure this function
2534 				 * doesn't think it has changed next
2535 				 * time it checks.
2536 				 */
2537 				rc = efx_ef10_get_warm_boot_count(efx);
2538 				if (rc >= 0) {
2539 					nic_data->warm_boot_count = rc;
2540 					rc = 0;
2541 				}
2542 			}
2543 			nic_data->workaround_26807 = true;
2544 		} else if (rc == -EPERM) {
2545 			rc = 0;
2546 		}
2547 	}
2548 	return rc;
2549 }
2550 
2551 static int efx_ef10_filter_table_probe(struct efx_nic *efx)
2552 {
2553 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2554 	int rc = efx_ef10_probe_multicast_chaining(efx);
2555 	struct efx_mcdi_filter_vlan *vlan;
2556 
2557 	if (rc)
2558 		return rc;
2559 	down_write(&efx->filter_sem);
2560 	rc = efx_mcdi_filter_table_probe(efx, nic_data->workaround_26807);
2561 
2562 	if (rc)
2563 		goto out_unlock;
2564 
2565 	list_for_each_entry(vlan, &nic_data->vlan_list, list) {
2566 		rc = efx_mcdi_filter_add_vlan(efx, vlan->vid);
2567 		if (rc)
2568 			goto fail_add_vlan;
2569 	}
2570 	goto out_unlock;
2571 
2572 fail_add_vlan:
2573 	efx_mcdi_filter_table_remove(efx);
2574 out_unlock:
2575 	up_write(&efx->filter_sem);
2576 	return rc;
2577 }
2578 
2579 static void efx_ef10_filter_table_remove(struct efx_nic *efx)
2580 {
2581 	down_write(&efx->filter_sem);
2582 	efx_mcdi_filter_table_remove(efx);
2583 	up_write(&efx->filter_sem);
2584 }
2585 
2586 /* This creates an entry in the RX descriptor queue */
2587 static inline void
2588 efx_ef10_build_rx_desc(struct efx_rx_queue *rx_queue, unsigned int index)
2589 {
2590 	struct efx_rx_buffer *rx_buf;
2591 	efx_qword_t *rxd;
2592 
2593 	rxd = efx_rx_desc(rx_queue, index);
2594 	rx_buf = efx_rx_buffer(rx_queue, index);
2595 	EFX_POPULATE_QWORD_2(*rxd,
2596 			     ESF_DZ_RX_KER_BYTE_CNT, rx_buf->len,
2597 			     ESF_DZ_RX_KER_BUF_ADDR, rx_buf->dma_addr);
2598 }
2599 
2600 static void efx_ef10_rx_write(struct efx_rx_queue *rx_queue)
2601 {
2602 	struct efx_nic *efx = rx_queue->efx;
2603 	unsigned int write_count;
2604 	efx_dword_t reg;
2605 
2606 	/* Firmware requires that RX_DESC_WPTR be a multiple of 8 */
2607 	write_count = rx_queue->added_count & ~7;
2608 	if (rx_queue->notified_count == write_count)
2609 		return;
2610 
2611 	do
2612 		efx_ef10_build_rx_desc(
2613 			rx_queue,
2614 			rx_queue->notified_count & rx_queue->ptr_mask);
2615 	while (++rx_queue->notified_count != write_count);
2616 
2617 	wmb();
2618 	EFX_POPULATE_DWORD_1(reg, ERF_DZ_RX_DESC_WPTR,
2619 			     write_count & rx_queue->ptr_mask);
2620 	efx_writed_page(efx, &reg, ER_DZ_RX_DESC_UPD,
2621 			efx_rx_queue_index(rx_queue));
2622 }
2623 
2624 static efx_mcdi_async_completer efx_ef10_rx_defer_refill_complete;
2625 
2626 static void efx_ef10_rx_defer_refill(struct efx_rx_queue *rx_queue)
2627 {
2628 	struct efx_channel *channel = efx_rx_queue_channel(rx_queue);
2629 	MCDI_DECLARE_BUF(inbuf, MC_CMD_DRIVER_EVENT_IN_LEN);
2630 	efx_qword_t event;
2631 
2632 	EFX_POPULATE_QWORD_2(event,
2633 			     ESF_DZ_EV_CODE, EFX_EF10_DRVGEN_EV,
2634 			     ESF_DZ_EV_DATA, EFX_EF10_REFILL);
2635 
2636 	MCDI_SET_DWORD(inbuf, DRIVER_EVENT_IN_EVQ, channel->channel);
2637 
2638 	/* MCDI_SET_QWORD is not appropriate here since EFX_POPULATE_* has
2639 	 * already swapped the data to little-endian order.
2640 	 */
2641 	memcpy(MCDI_PTR(inbuf, DRIVER_EVENT_IN_DATA), &event.u64[0],
2642 	       sizeof(efx_qword_t));
2643 
2644 	efx_mcdi_rpc_async(channel->efx, MC_CMD_DRIVER_EVENT,
2645 			   inbuf, sizeof(inbuf), 0,
2646 			   efx_ef10_rx_defer_refill_complete, 0);
2647 }
2648 
2649 static void
2650 efx_ef10_rx_defer_refill_complete(struct efx_nic *efx, unsigned long cookie,
2651 				  int rc, efx_dword_t *outbuf,
2652 				  size_t outlen_actual)
2653 {
2654 	/* nothing to do */
2655 }
2656 
2657 static int efx_ef10_ev_init(struct efx_channel *channel)
2658 {
2659 	struct efx_nic *efx = channel->efx;
2660 	struct efx_ef10_nic_data *nic_data;
2661 	bool use_v2, cut_thru;
2662 
2663 	nic_data = efx->nic_data;
2664 	use_v2 = nic_data->datapath_caps2 &
2665 			    1 << MC_CMD_GET_CAPABILITIES_V2_OUT_INIT_EVQ_V2_LBN;
2666 	cut_thru = !(nic_data->datapath_caps &
2667 			      1 << MC_CMD_GET_CAPABILITIES_OUT_RX_BATCHING_LBN);
2668 	return efx_mcdi_ev_init(channel, cut_thru, use_v2);
2669 }
2670 
2671 static void efx_ef10_handle_rx_wrong_queue(struct efx_rx_queue *rx_queue,
2672 					   unsigned int rx_queue_label)
2673 {
2674 	struct efx_nic *efx = rx_queue->efx;
2675 
2676 	netif_info(efx, hw, efx->net_dev,
2677 		   "rx event arrived on queue %d labeled as queue %u\n",
2678 		   efx_rx_queue_index(rx_queue), rx_queue_label);
2679 
2680 	efx_schedule_reset(efx, RESET_TYPE_DISABLE);
2681 }
2682 
2683 static void
2684 efx_ef10_handle_rx_bad_lbits(struct efx_rx_queue *rx_queue,
2685 			     unsigned int actual, unsigned int expected)
2686 {
2687 	unsigned int dropped = (actual - expected) & rx_queue->ptr_mask;
2688 	struct efx_nic *efx = rx_queue->efx;
2689 
2690 	netif_info(efx, hw, efx->net_dev,
2691 		   "dropped %d events (index=%d expected=%d)\n",
2692 		   dropped, actual, expected);
2693 
2694 	efx_schedule_reset(efx, RESET_TYPE_DISABLE);
2695 }
2696 
2697 /* partially received RX was aborted. clean up. */
2698 static void efx_ef10_handle_rx_abort(struct efx_rx_queue *rx_queue)
2699 {
2700 	unsigned int rx_desc_ptr;
2701 
2702 	netif_dbg(rx_queue->efx, hw, rx_queue->efx->net_dev,
2703 		  "scattered RX aborted (dropping %u buffers)\n",
2704 		  rx_queue->scatter_n);
2705 
2706 	rx_desc_ptr = rx_queue->removed_count & rx_queue->ptr_mask;
2707 
2708 	efx_rx_packet(rx_queue, rx_desc_ptr, rx_queue->scatter_n,
2709 		      0, EFX_RX_PKT_DISCARD);
2710 
2711 	rx_queue->removed_count += rx_queue->scatter_n;
2712 	rx_queue->scatter_n = 0;
2713 	rx_queue->scatter_len = 0;
2714 	++efx_rx_queue_channel(rx_queue)->n_rx_nodesc_trunc;
2715 }
2716 
2717 static u16 efx_ef10_handle_rx_event_errors(struct efx_channel *channel,
2718 					   unsigned int n_packets,
2719 					   unsigned int rx_encap_hdr,
2720 					   unsigned int rx_l3_class,
2721 					   unsigned int rx_l4_class,
2722 					   const efx_qword_t *event)
2723 {
2724 	struct efx_nic *efx = channel->efx;
2725 	bool handled = false;
2726 
2727 	if (EFX_QWORD_FIELD(*event, ESF_DZ_RX_ECRC_ERR)) {
2728 		if (!(efx->net_dev->features & NETIF_F_RXALL)) {
2729 			if (!efx->loopback_selftest)
2730 				channel->n_rx_eth_crc_err += n_packets;
2731 			return EFX_RX_PKT_DISCARD;
2732 		}
2733 		handled = true;
2734 	}
2735 	if (EFX_QWORD_FIELD(*event, ESF_DZ_RX_IPCKSUM_ERR)) {
2736 		if (unlikely(rx_encap_hdr != ESE_EZ_ENCAP_HDR_VXLAN &&
2737 			     rx_l3_class != ESE_DZ_L3_CLASS_IP4 &&
2738 			     rx_l3_class != ESE_DZ_L3_CLASS_IP4_FRAG &&
2739 			     rx_l3_class != ESE_DZ_L3_CLASS_IP6 &&
2740 			     rx_l3_class != ESE_DZ_L3_CLASS_IP6_FRAG))
2741 			netdev_WARN(efx->net_dev,
2742 				    "invalid class for RX_IPCKSUM_ERR: event="
2743 				    EFX_QWORD_FMT "\n",
2744 				    EFX_QWORD_VAL(*event));
2745 		if (!efx->loopback_selftest)
2746 			*(rx_encap_hdr ?
2747 			  &channel->n_rx_outer_ip_hdr_chksum_err :
2748 			  &channel->n_rx_ip_hdr_chksum_err) += n_packets;
2749 		return 0;
2750 	}
2751 	if (EFX_QWORD_FIELD(*event, ESF_DZ_RX_TCPUDP_CKSUM_ERR)) {
2752 		if (unlikely(rx_encap_hdr != ESE_EZ_ENCAP_HDR_VXLAN &&
2753 			     ((rx_l3_class != ESE_DZ_L3_CLASS_IP4 &&
2754 			       rx_l3_class != ESE_DZ_L3_CLASS_IP6) ||
2755 			      (rx_l4_class != ESE_FZ_L4_CLASS_TCP &&
2756 			       rx_l4_class != ESE_FZ_L4_CLASS_UDP))))
2757 			netdev_WARN(efx->net_dev,
2758 				    "invalid class for RX_TCPUDP_CKSUM_ERR: event="
2759 				    EFX_QWORD_FMT "\n",
2760 				    EFX_QWORD_VAL(*event));
2761 		if (!efx->loopback_selftest)
2762 			*(rx_encap_hdr ?
2763 			  &channel->n_rx_outer_tcp_udp_chksum_err :
2764 			  &channel->n_rx_tcp_udp_chksum_err) += n_packets;
2765 		return 0;
2766 	}
2767 	if (EFX_QWORD_FIELD(*event, ESF_EZ_RX_IP_INNER_CHKSUM_ERR)) {
2768 		if (unlikely(!rx_encap_hdr))
2769 			netdev_WARN(efx->net_dev,
2770 				    "invalid encapsulation type for RX_IP_INNER_CHKSUM_ERR: event="
2771 				    EFX_QWORD_FMT "\n",
2772 				    EFX_QWORD_VAL(*event));
2773 		else if (unlikely(rx_l3_class != ESE_DZ_L3_CLASS_IP4 &&
2774 				  rx_l3_class != ESE_DZ_L3_CLASS_IP4_FRAG &&
2775 				  rx_l3_class != ESE_DZ_L3_CLASS_IP6 &&
2776 				  rx_l3_class != ESE_DZ_L3_CLASS_IP6_FRAG))
2777 			netdev_WARN(efx->net_dev,
2778 				    "invalid class for RX_IP_INNER_CHKSUM_ERR: event="
2779 				    EFX_QWORD_FMT "\n",
2780 				    EFX_QWORD_VAL(*event));
2781 		if (!efx->loopback_selftest)
2782 			channel->n_rx_inner_ip_hdr_chksum_err += n_packets;
2783 		return 0;
2784 	}
2785 	if (EFX_QWORD_FIELD(*event, ESF_EZ_RX_TCP_UDP_INNER_CHKSUM_ERR)) {
2786 		if (unlikely(!rx_encap_hdr))
2787 			netdev_WARN(efx->net_dev,
2788 				    "invalid encapsulation type for RX_TCP_UDP_INNER_CHKSUM_ERR: event="
2789 				    EFX_QWORD_FMT "\n",
2790 				    EFX_QWORD_VAL(*event));
2791 		else if (unlikely((rx_l3_class != ESE_DZ_L3_CLASS_IP4 &&
2792 				   rx_l3_class != ESE_DZ_L3_CLASS_IP6) ||
2793 				  (rx_l4_class != ESE_FZ_L4_CLASS_TCP &&
2794 				   rx_l4_class != ESE_FZ_L4_CLASS_UDP)))
2795 			netdev_WARN(efx->net_dev,
2796 				    "invalid class for RX_TCP_UDP_INNER_CHKSUM_ERR: event="
2797 				    EFX_QWORD_FMT "\n",
2798 				    EFX_QWORD_VAL(*event));
2799 		if (!efx->loopback_selftest)
2800 			channel->n_rx_inner_tcp_udp_chksum_err += n_packets;
2801 		return 0;
2802 	}
2803 
2804 	WARN_ON(!handled); /* No error bits were recognised */
2805 	return 0;
2806 }
2807 
2808 static int efx_ef10_handle_rx_event(struct efx_channel *channel,
2809 				    const efx_qword_t *event)
2810 {
2811 	unsigned int rx_bytes, next_ptr_lbits, rx_queue_label;
2812 	unsigned int rx_l3_class, rx_l4_class, rx_encap_hdr;
2813 	unsigned int n_descs, n_packets, i;
2814 	struct efx_nic *efx = channel->efx;
2815 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2816 	struct efx_rx_queue *rx_queue;
2817 	efx_qword_t errors;
2818 	bool rx_cont;
2819 	u16 flags = 0;
2820 
2821 	if (unlikely(READ_ONCE(efx->reset_pending)))
2822 		return 0;
2823 
2824 	/* Basic packet information */
2825 	rx_bytes = EFX_QWORD_FIELD(*event, ESF_DZ_RX_BYTES);
2826 	next_ptr_lbits = EFX_QWORD_FIELD(*event, ESF_DZ_RX_DSC_PTR_LBITS);
2827 	rx_queue_label = EFX_QWORD_FIELD(*event, ESF_DZ_RX_QLABEL);
2828 	rx_l3_class = EFX_QWORD_FIELD(*event, ESF_DZ_RX_L3_CLASS);
2829 	rx_l4_class = EFX_QWORD_FIELD(*event, ESF_FZ_RX_L4_CLASS);
2830 	rx_cont = EFX_QWORD_FIELD(*event, ESF_DZ_RX_CONT);
2831 	rx_encap_hdr =
2832 		nic_data->datapath_caps &
2833 			(1 << MC_CMD_GET_CAPABILITIES_OUT_VXLAN_NVGRE_LBN) ?
2834 		EFX_QWORD_FIELD(*event, ESF_EZ_RX_ENCAP_HDR) :
2835 		ESE_EZ_ENCAP_HDR_NONE;
2836 
2837 	if (EFX_QWORD_FIELD(*event, ESF_DZ_RX_DROP_EVENT))
2838 		netdev_WARN(efx->net_dev, "saw RX_DROP_EVENT: event="
2839 			    EFX_QWORD_FMT "\n",
2840 			    EFX_QWORD_VAL(*event));
2841 
2842 	rx_queue = efx_channel_get_rx_queue(channel);
2843 
2844 	if (unlikely(rx_queue_label != efx_rx_queue_index(rx_queue)))
2845 		efx_ef10_handle_rx_wrong_queue(rx_queue, rx_queue_label);
2846 
2847 	n_descs = ((next_ptr_lbits - rx_queue->removed_count) &
2848 		   ((1 << ESF_DZ_RX_DSC_PTR_LBITS_WIDTH) - 1));
2849 
2850 	if (n_descs != rx_queue->scatter_n + 1) {
2851 		struct efx_ef10_nic_data *nic_data = efx->nic_data;
2852 
2853 		/* detect rx abort */
2854 		if (unlikely(n_descs == rx_queue->scatter_n)) {
2855 			if (rx_queue->scatter_n == 0 || rx_bytes != 0)
2856 				netdev_WARN(efx->net_dev,
2857 					    "invalid RX abort: scatter_n=%u event="
2858 					    EFX_QWORD_FMT "\n",
2859 					    rx_queue->scatter_n,
2860 					    EFX_QWORD_VAL(*event));
2861 			efx_ef10_handle_rx_abort(rx_queue);
2862 			return 0;
2863 		}
2864 
2865 		/* Check that RX completion merging is valid, i.e.
2866 		 * the current firmware supports it and this is a
2867 		 * non-scattered packet.
2868 		 */
2869 		if (!(nic_data->datapath_caps &
2870 		      (1 << MC_CMD_GET_CAPABILITIES_OUT_RX_BATCHING_LBN)) ||
2871 		    rx_queue->scatter_n != 0 || rx_cont) {
2872 			efx_ef10_handle_rx_bad_lbits(
2873 				rx_queue, next_ptr_lbits,
2874 				(rx_queue->removed_count +
2875 				 rx_queue->scatter_n + 1) &
2876 				((1 << ESF_DZ_RX_DSC_PTR_LBITS_WIDTH) - 1));
2877 			return 0;
2878 		}
2879 
2880 		/* Merged completion for multiple non-scattered packets */
2881 		rx_queue->scatter_n = 1;
2882 		rx_queue->scatter_len = 0;
2883 		n_packets = n_descs;
2884 		++channel->n_rx_merge_events;
2885 		channel->n_rx_merge_packets += n_packets;
2886 		flags |= EFX_RX_PKT_PREFIX_LEN;
2887 	} else {
2888 		++rx_queue->scatter_n;
2889 		rx_queue->scatter_len += rx_bytes;
2890 		if (rx_cont)
2891 			return 0;
2892 		n_packets = 1;
2893 	}
2894 
2895 	EFX_POPULATE_QWORD_5(errors, ESF_DZ_RX_ECRC_ERR, 1,
2896 				     ESF_DZ_RX_IPCKSUM_ERR, 1,
2897 				     ESF_DZ_RX_TCPUDP_CKSUM_ERR, 1,
2898 				     ESF_EZ_RX_IP_INNER_CHKSUM_ERR, 1,
2899 				     ESF_EZ_RX_TCP_UDP_INNER_CHKSUM_ERR, 1);
2900 	EFX_AND_QWORD(errors, *event, errors);
2901 	if (unlikely(!EFX_QWORD_IS_ZERO(errors))) {
2902 		flags |= efx_ef10_handle_rx_event_errors(channel, n_packets,
2903 							 rx_encap_hdr,
2904 							 rx_l3_class, rx_l4_class,
2905 							 event);
2906 	} else {
2907 		bool tcpudp = rx_l4_class == ESE_FZ_L4_CLASS_TCP ||
2908 			      rx_l4_class == ESE_FZ_L4_CLASS_UDP;
2909 
2910 		switch (rx_encap_hdr) {
2911 		case ESE_EZ_ENCAP_HDR_VXLAN: /* VxLAN or GENEVE */
2912 			flags |= EFX_RX_PKT_CSUMMED; /* outer UDP csum */
2913 			if (tcpudp)
2914 				flags |= EFX_RX_PKT_CSUM_LEVEL; /* inner L4 */
2915 			break;
2916 		case ESE_EZ_ENCAP_HDR_GRE:
2917 		case ESE_EZ_ENCAP_HDR_NONE:
2918 			if (tcpudp)
2919 				flags |= EFX_RX_PKT_CSUMMED;
2920 			break;
2921 		default:
2922 			netdev_WARN(efx->net_dev,
2923 				    "unknown encapsulation type: event="
2924 				    EFX_QWORD_FMT "\n",
2925 				    EFX_QWORD_VAL(*event));
2926 		}
2927 	}
2928 
2929 	if (rx_l4_class == ESE_FZ_L4_CLASS_TCP)
2930 		flags |= EFX_RX_PKT_TCP;
2931 
2932 	channel->irq_mod_score += 2 * n_packets;
2933 
2934 	/* Handle received packet(s) */
2935 	for (i = 0; i < n_packets; i++) {
2936 		efx_rx_packet(rx_queue,
2937 			      rx_queue->removed_count & rx_queue->ptr_mask,
2938 			      rx_queue->scatter_n, rx_queue->scatter_len,
2939 			      flags);
2940 		rx_queue->removed_count += rx_queue->scatter_n;
2941 	}
2942 
2943 	rx_queue->scatter_n = 0;
2944 	rx_queue->scatter_len = 0;
2945 
2946 	return n_packets;
2947 }
2948 
2949 static u32 efx_ef10_extract_event_ts(efx_qword_t *event)
2950 {
2951 	u32 tstamp;
2952 
2953 	tstamp = EFX_QWORD_FIELD(*event, TX_TIMESTAMP_EVENT_TSTAMP_DATA_HI);
2954 	tstamp <<= 16;
2955 	tstamp |= EFX_QWORD_FIELD(*event, TX_TIMESTAMP_EVENT_TSTAMP_DATA_LO);
2956 
2957 	return tstamp;
2958 }
2959 
2960 static int
2961 efx_ef10_handle_tx_event(struct efx_channel *channel, efx_qword_t *event)
2962 {
2963 	struct efx_nic *efx = channel->efx;
2964 	struct efx_tx_queue *tx_queue;
2965 	unsigned int tx_ev_desc_ptr;
2966 	unsigned int tx_ev_q_label;
2967 	unsigned int tx_ev_type;
2968 	int work_done;
2969 	u64 ts_part;
2970 
2971 	if (unlikely(READ_ONCE(efx->reset_pending)))
2972 		return 0;
2973 
2974 	if (unlikely(EFX_QWORD_FIELD(*event, ESF_DZ_TX_DROP_EVENT)))
2975 		return 0;
2976 
2977 	/* Get the transmit queue */
2978 	tx_ev_q_label = EFX_QWORD_FIELD(*event, ESF_DZ_TX_QLABEL);
2979 	tx_queue = channel->tx_queue + (tx_ev_q_label % EFX_MAX_TXQ_PER_CHANNEL);
2980 
2981 	if (!tx_queue->timestamping) {
2982 		/* Transmit completion */
2983 		tx_ev_desc_ptr = EFX_QWORD_FIELD(*event, ESF_DZ_TX_DESCR_INDX);
2984 		return efx_xmit_done(tx_queue, tx_ev_desc_ptr & tx_queue->ptr_mask);
2985 	}
2986 
2987 	/* Transmit timestamps are only available for 8XXX series. They result
2988 	 * in up to three events per packet. These occur in order, and are:
2989 	 *  - the normal completion event (may be omitted)
2990 	 *  - the low part of the timestamp
2991 	 *  - the high part of the timestamp
2992 	 *
2993 	 * It's possible for multiple completion events to appear before the
2994 	 * corresponding timestamps. So we can for example get:
2995 	 *  COMP N
2996 	 *  COMP N+1
2997 	 *  TS_LO N
2998 	 *  TS_HI N
2999 	 *  TS_LO N+1
3000 	 *  TS_HI N+1
3001 	 *
3002 	 * In addition it's also possible for the adjacent completions to be
3003 	 * merged, so we may not see COMP N above. As such, the completion
3004 	 * events are not very useful here.
3005 	 *
3006 	 * Each part of the timestamp is itself split across two 16 bit
3007 	 * fields in the event.
3008 	 */
3009 	tx_ev_type = EFX_QWORD_FIELD(*event, ESF_EZ_TX_SOFT1);
3010 	work_done = 0;
3011 
3012 	switch (tx_ev_type) {
3013 	case TX_TIMESTAMP_EVENT_TX_EV_COMPLETION:
3014 		/* Ignore this event - see above. */
3015 		break;
3016 
3017 	case TX_TIMESTAMP_EVENT_TX_EV_TSTAMP_LO:
3018 		ts_part = efx_ef10_extract_event_ts(event);
3019 		tx_queue->completed_timestamp_minor = ts_part;
3020 		break;
3021 
3022 	case TX_TIMESTAMP_EVENT_TX_EV_TSTAMP_HI:
3023 		ts_part = efx_ef10_extract_event_ts(event);
3024 		tx_queue->completed_timestamp_major = ts_part;
3025 
3026 		efx_xmit_done_single(tx_queue);
3027 		work_done = 1;
3028 		break;
3029 
3030 	default:
3031 		netif_err(efx, hw, efx->net_dev,
3032 			  "channel %d unknown tx event type %d (data "
3033 			  EFX_QWORD_FMT ")\n",
3034 			  channel->channel, tx_ev_type,
3035 			  EFX_QWORD_VAL(*event));
3036 		break;
3037 	}
3038 
3039 	return work_done;
3040 }
3041 
3042 static void
3043 efx_ef10_handle_driver_event(struct efx_channel *channel, efx_qword_t *event)
3044 {
3045 	struct efx_nic *efx = channel->efx;
3046 	int subcode;
3047 
3048 	subcode = EFX_QWORD_FIELD(*event, ESF_DZ_DRV_SUB_CODE);
3049 
3050 	switch (subcode) {
3051 	case ESE_DZ_DRV_TIMER_EV:
3052 	case ESE_DZ_DRV_WAKE_UP_EV:
3053 		break;
3054 	case ESE_DZ_DRV_START_UP_EV:
3055 		/* event queue init complete. ok. */
3056 		break;
3057 	default:
3058 		netif_err(efx, hw, efx->net_dev,
3059 			  "channel %d unknown driver event type %d"
3060 			  " (data " EFX_QWORD_FMT ")\n",
3061 			  channel->channel, subcode,
3062 			  EFX_QWORD_VAL(*event));
3063 
3064 	}
3065 }
3066 
3067 static void efx_ef10_handle_driver_generated_event(struct efx_channel *channel,
3068 						   efx_qword_t *event)
3069 {
3070 	struct efx_nic *efx = channel->efx;
3071 	u32 subcode;
3072 
3073 	subcode = EFX_QWORD_FIELD(*event, EFX_DWORD_0);
3074 
3075 	switch (subcode) {
3076 	case EFX_EF10_TEST:
3077 		channel->event_test_cpu = raw_smp_processor_id();
3078 		break;
3079 	case EFX_EF10_REFILL:
3080 		/* The queue must be empty, so we won't receive any rx
3081 		 * events, so efx_process_channel() won't refill the
3082 		 * queue. Refill it here
3083 		 */
3084 		efx_fast_push_rx_descriptors(&channel->rx_queue, true);
3085 		break;
3086 	default:
3087 		netif_err(efx, hw, efx->net_dev,
3088 			  "channel %d unknown driver event type %u"
3089 			  " (data " EFX_QWORD_FMT ")\n",
3090 			  channel->channel, (unsigned) subcode,
3091 			  EFX_QWORD_VAL(*event));
3092 	}
3093 }
3094 
3095 #define EFX_NAPI_MAX_TX 512
3096 
3097 static int efx_ef10_ev_process(struct efx_channel *channel, int quota)
3098 {
3099 	struct efx_nic *efx = channel->efx;
3100 	efx_qword_t event, *p_event;
3101 	unsigned int read_ptr;
3102 	int spent_tx = 0;
3103 	int spent = 0;
3104 	int ev_code;
3105 
3106 	if (quota <= 0)
3107 		return spent;
3108 
3109 	read_ptr = channel->eventq_read_ptr;
3110 
3111 	for (;;) {
3112 		p_event = efx_event(channel, read_ptr);
3113 		event = *p_event;
3114 
3115 		if (!efx_event_present(&event))
3116 			break;
3117 
3118 		EFX_SET_QWORD(*p_event);
3119 
3120 		++read_ptr;
3121 
3122 		ev_code = EFX_QWORD_FIELD(event, ESF_DZ_EV_CODE);
3123 
3124 		netif_vdbg(efx, drv, efx->net_dev,
3125 			   "processing event on %d " EFX_QWORD_FMT "\n",
3126 			   channel->channel, EFX_QWORD_VAL(event));
3127 
3128 		switch (ev_code) {
3129 		case ESE_DZ_EV_CODE_MCDI_EV:
3130 			efx_mcdi_process_event(channel, &event);
3131 			break;
3132 		case ESE_DZ_EV_CODE_RX_EV:
3133 			spent += efx_ef10_handle_rx_event(channel, &event);
3134 			if (spent >= quota) {
3135 				/* XXX can we split a merged event to
3136 				 * avoid going over-quota?
3137 				 */
3138 				spent = quota;
3139 				goto out;
3140 			}
3141 			break;
3142 		case ESE_DZ_EV_CODE_TX_EV:
3143 			spent_tx += efx_ef10_handle_tx_event(channel, &event);
3144 			if (spent_tx >= EFX_NAPI_MAX_TX) {
3145 				spent = quota;
3146 				goto out;
3147 			}
3148 			break;
3149 		case ESE_DZ_EV_CODE_DRIVER_EV:
3150 			efx_ef10_handle_driver_event(channel, &event);
3151 			if (++spent == quota)
3152 				goto out;
3153 			break;
3154 		case EFX_EF10_DRVGEN_EV:
3155 			efx_ef10_handle_driver_generated_event(channel, &event);
3156 			break;
3157 		default:
3158 			netif_err(efx, hw, efx->net_dev,
3159 				  "channel %d unknown event type %d"
3160 				  " (data " EFX_QWORD_FMT ")\n",
3161 				  channel->channel, ev_code,
3162 				  EFX_QWORD_VAL(event));
3163 		}
3164 	}
3165 
3166 out:
3167 	channel->eventq_read_ptr = read_ptr;
3168 	return spent;
3169 }
3170 
3171 static void efx_ef10_ev_read_ack(struct efx_channel *channel)
3172 {
3173 	struct efx_nic *efx = channel->efx;
3174 	efx_dword_t rptr;
3175 
3176 	if (EFX_EF10_WORKAROUND_35388(efx)) {
3177 		BUILD_BUG_ON(EFX_MIN_EVQ_SIZE <
3178 			     (1 << ERF_DD_EVQ_IND_RPTR_WIDTH));
3179 		BUILD_BUG_ON(EFX_MAX_EVQ_SIZE >
3180 			     (1 << 2 * ERF_DD_EVQ_IND_RPTR_WIDTH));
3181 
3182 		EFX_POPULATE_DWORD_2(rptr, ERF_DD_EVQ_IND_RPTR_FLAGS,
3183 				     EFE_DD_EVQ_IND_RPTR_FLAGS_HIGH,
3184 				     ERF_DD_EVQ_IND_RPTR,
3185 				     (channel->eventq_read_ptr &
3186 				      channel->eventq_mask) >>
3187 				     ERF_DD_EVQ_IND_RPTR_WIDTH);
3188 		efx_writed_page(efx, &rptr, ER_DD_EVQ_INDIRECT,
3189 				channel->channel);
3190 		EFX_POPULATE_DWORD_2(rptr, ERF_DD_EVQ_IND_RPTR_FLAGS,
3191 				     EFE_DD_EVQ_IND_RPTR_FLAGS_LOW,
3192 				     ERF_DD_EVQ_IND_RPTR,
3193 				     channel->eventq_read_ptr &
3194 				     ((1 << ERF_DD_EVQ_IND_RPTR_WIDTH) - 1));
3195 		efx_writed_page(efx, &rptr, ER_DD_EVQ_INDIRECT,
3196 				channel->channel);
3197 	} else {
3198 		EFX_POPULATE_DWORD_1(rptr, ERF_DZ_EVQ_RPTR,
3199 				     channel->eventq_read_ptr &
3200 				     channel->eventq_mask);
3201 		efx_writed_page(efx, &rptr, ER_DZ_EVQ_RPTR, channel->channel);
3202 	}
3203 }
3204 
3205 static void efx_ef10_ev_test_generate(struct efx_channel *channel)
3206 {
3207 	MCDI_DECLARE_BUF(inbuf, MC_CMD_DRIVER_EVENT_IN_LEN);
3208 	struct efx_nic *efx = channel->efx;
3209 	efx_qword_t event;
3210 	int rc;
3211 
3212 	EFX_POPULATE_QWORD_2(event,
3213 			     ESF_DZ_EV_CODE, EFX_EF10_DRVGEN_EV,
3214 			     ESF_DZ_EV_DATA, EFX_EF10_TEST);
3215 
3216 	MCDI_SET_DWORD(inbuf, DRIVER_EVENT_IN_EVQ, channel->channel);
3217 
3218 	/* MCDI_SET_QWORD is not appropriate here since EFX_POPULATE_* has
3219 	 * already swapped the data to little-endian order.
3220 	 */
3221 	memcpy(MCDI_PTR(inbuf, DRIVER_EVENT_IN_DATA), &event.u64[0],
3222 	       sizeof(efx_qword_t));
3223 
3224 	rc = efx_mcdi_rpc(efx, MC_CMD_DRIVER_EVENT, inbuf, sizeof(inbuf),
3225 			  NULL, 0, NULL);
3226 	if (rc != 0)
3227 		goto fail;
3228 
3229 	return;
3230 
3231 fail:
3232 	WARN_ON(true);
3233 	netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
3234 }
3235 
3236 static void efx_ef10_prepare_flr(struct efx_nic *efx)
3237 {
3238 	atomic_set(&efx->active_queues, 0);
3239 }
3240 
3241 static int efx_ef10_vport_set_mac_address(struct efx_nic *efx)
3242 {
3243 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
3244 	u8 mac_old[ETH_ALEN];
3245 	int rc, rc2;
3246 
3247 	/* Only reconfigure a PF-created vport */
3248 	if (is_zero_ether_addr(nic_data->vport_mac))
3249 		return 0;
3250 
3251 	efx_device_detach_sync(efx);
3252 	efx_net_stop(efx->net_dev);
3253 	efx_ef10_filter_table_remove(efx);
3254 
3255 	rc = efx_ef10_vadaptor_free(efx, efx->vport_id);
3256 	if (rc)
3257 		goto restore_filters;
3258 
3259 	ether_addr_copy(mac_old, nic_data->vport_mac);
3260 	rc = efx_ef10_vport_del_mac(efx, efx->vport_id,
3261 				    nic_data->vport_mac);
3262 	if (rc)
3263 		goto restore_vadaptor;
3264 
3265 	rc = efx_ef10_vport_add_mac(efx, efx->vport_id,
3266 				    efx->net_dev->dev_addr);
3267 	if (!rc) {
3268 		ether_addr_copy(nic_data->vport_mac, efx->net_dev->dev_addr);
3269 	} else {
3270 		rc2 = efx_ef10_vport_add_mac(efx, efx->vport_id, mac_old);
3271 		if (rc2) {
3272 			/* Failed to add original MAC, so clear vport_mac */
3273 			eth_zero_addr(nic_data->vport_mac);
3274 			goto reset_nic;
3275 		}
3276 	}
3277 
3278 restore_vadaptor:
3279 	rc2 = efx_ef10_vadaptor_alloc(efx, efx->vport_id);
3280 	if (rc2)
3281 		goto reset_nic;
3282 restore_filters:
3283 	rc2 = efx_ef10_filter_table_probe(efx);
3284 	if (rc2)
3285 		goto reset_nic;
3286 
3287 	rc2 = efx_net_open(efx->net_dev);
3288 	if (rc2)
3289 		goto reset_nic;
3290 
3291 	efx_device_attach_if_not_resetting(efx);
3292 
3293 	return rc;
3294 
3295 reset_nic:
3296 	netif_err(efx, drv, efx->net_dev,
3297 		  "Failed to restore when changing MAC address - scheduling reset\n");
3298 	efx_schedule_reset(efx, RESET_TYPE_DATAPATH);
3299 
3300 	return rc ? rc : rc2;
3301 }
3302 
3303 static int efx_ef10_set_mac_address(struct efx_nic *efx)
3304 {
3305 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VADAPTOR_SET_MAC_IN_LEN);
3306 	bool was_enabled = efx->port_enabled;
3307 	int rc;
3308 
3309 #ifdef CONFIG_SFC_SRIOV
3310 	/* If this function is a VF and we have access to the parent PF,
3311 	 * then use the PF control path to attempt to change the VF MAC address.
3312 	 */
3313 	if (efx->pci_dev->is_virtfn && efx->pci_dev->physfn) {
3314 		struct efx_nic *efx_pf = pci_get_drvdata(efx->pci_dev->physfn);
3315 		struct efx_ef10_nic_data *nic_data = efx->nic_data;
3316 		u8 mac[ETH_ALEN];
3317 
3318 		/* net_dev->dev_addr can be zeroed by efx_net_stop in
3319 		 * efx_ef10_sriov_set_vf_mac, so pass in a copy.
3320 		 */
3321 		ether_addr_copy(mac, efx->net_dev->dev_addr);
3322 
3323 		rc = efx_ef10_sriov_set_vf_mac(efx_pf, nic_data->vf_index, mac);
3324 		if (!rc)
3325 			return 0;
3326 
3327 		netif_dbg(efx, drv, efx->net_dev,
3328 			  "Updating VF mac via PF failed (%d), setting directly\n",
3329 			  rc);
3330 	}
3331 #endif
3332 
3333 	efx_device_detach_sync(efx);
3334 	efx_net_stop(efx->net_dev);
3335 
3336 	mutex_lock(&efx->mac_lock);
3337 	efx_ef10_filter_table_remove(efx);
3338 
3339 	ether_addr_copy(MCDI_PTR(inbuf, VADAPTOR_SET_MAC_IN_MACADDR),
3340 			efx->net_dev->dev_addr);
3341 	MCDI_SET_DWORD(inbuf, VADAPTOR_SET_MAC_IN_UPSTREAM_PORT_ID,
3342 		       efx->vport_id);
3343 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_VADAPTOR_SET_MAC, inbuf,
3344 				sizeof(inbuf), NULL, 0, NULL);
3345 
3346 	efx_ef10_filter_table_probe(efx);
3347 	mutex_unlock(&efx->mac_lock);
3348 
3349 	if (was_enabled)
3350 		efx_net_open(efx->net_dev);
3351 	efx_device_attach_if_not_resetting(efx);
3352 
3353 	if (rc == -EPERM) {
3354 		netif_err(efx, drv, efx->net_dev,
3355 			  "Cannot change MAC address; use sfboot to enable"
3356 			  " mac-spoofing on this interface\n");
3357 	} else if (rc == -ENOSYS && !efx_ef10_is_vf(efx)) {
3358 		/* If the active MCFW does not support MC_CMD_VADAPTOR_SET_MAC
3359 		 * fall-back to the method of changing the MAC address on the
3360 		 * vport.  This only applies to PFs because such versions of
3361 		 * MCFW do not support VFs.
3362 		 */
3363 		rc = efx_ef10_vport_set_mac_address(efx);
3364 	} else if (rc) {
3365 		efx_mcdi_display_error(efx, MC_CMD_VADAPTOR_SET_MAC,
3366 				       sizeof(inbuf), NULL, 0, rc);
3367 	}
3368 
3369 	return rc;
3370 }
3371 
3372 static int efx_ef10_mac_reconfigure(struct efx_nic *efx, bool mtu_only)
3373 {
3374 	WARN_ON(!mutex_is_locked(&efx->mac_lock));
3375 
3376 	efx_mcdi_filter_sync_rx_mode(efx);
3377 
3378 	if (mtu_only && efx_has_cap(efx, SET_MAC_ENHANCED))
3379 		return efx_mcdi_set_mtu(efx);
3380 	return efx_mcdi_set_mac(efx);
3381 }
3382 
3383 static int efx_ef10_start_bist(struct efx_nic *efx, u32 bist_type)
3384 {
3385 	MCDI_DECLARE_BUF(inbuf, MC_CMD_START_BIST_IN_LEN);
3386 
3387 	MCDI_SET_DWORD(inbuf, START_BIST_IN_TYPE, bist_type);
3388 	return efx_mcdi_rpc(efx, MC_CMD_START_BIST, inbuf, sizeof(inbuf),
3389 			    NULL, 0, NULL);
3390 }
3391 
3392 /* MC BISTs follow a different poll mechanism to phy BISTs.
3393  * The BIST is done in the poll handler on the MC, and the MCDI command
3394  * will block until the BIST is done.
3395  */
3396 static int efx_ef10_poll_bist(struct efx_nic *efx)
3397 {
3398 	int rc;
3399 	MCDI_DECLARE_BUF(outbuf, MC_CMD_POLL_BIST_OUT_LEN);
3400 	size_t outlen;
3401 	u32 result;
3402 
3403 	rc = efx_mcdi_rpc(efx, MC_CMD_POLL_BIST, NULL, 0,
3404 			   outbuf, sizeof(outbuf), &outlen);
3405 	if (rc != 0)
3406 		return rc;
3407 
3408 	if (outlen < MC_CMD_POLL_BIST_OUT_LEN)
3409 		return -EIO;
3410 
3411 	result = MCDI_DWORD(outbuf, POLL_BIST_OUT_RESULT);
3412 	switch (result) {
3413 	case MC_CMD_POLL_BIST_PASSED:
3414 		netif_dbg(efx, hw, efx->net_dev, "BIST passed.\n");
3415 		return 0;
3416 	case MC_CMD_POLL_BIST_TIMEOUT:
3417 		netif_err(efx, hw, efx->net_dev, "BIST timed out\n");
3418 		return -EIO;
3419 	case MC_CMD_POLL_BIST_FAILED:
3420 		netif_err(efx, hw, efx->net_dev, "BIST failed.\n");
3421 		return -EIO;
3422 	default:
3423 		netif_err(efx, hw, efx->net_dev,
3424 			  "BIST returned unknown result %u", result);
3425 		return -EIO;
3426 	}
3427 }
3428 
3429 static int efx_ef10_run_bist(struct efx_nic *efx, u32 bist_type)
3430 {
3431 	int rc;
3432 
3433 	netif_dbg(efx, drv, efx->net_dev, "starting BIST type %u\n", bist_type);
3434 
3435 	rc = efx_ef10_start_bist(efx, bist_type);
3436 	if (rc != 0)
3437 		return rc;
3438 
3439 	return efx_ef10_poll_bist(efx);
3440 }
3441 
3442 static int
3443 efx_ef10_test_chip(struct efx_nic *efx, struct efx_self_tests *tests)
3444 {
3445 	int rc, rc2;
3446 
3447 	efx_reset_down(efx, RESET_TYPE_WORLD);
3448 
3449 	rc = efx_mcdi_rpc(efx, MC_CMD_ENABLE_OFFLINE_BIST,
3450 			  NULL, 0, NULL, 0, NULL);
3451 	if (rc != 0)
3452 		goto out;
3453 
3454 	tests->memory = efx_ef10_run_bist(efx, MC_CMD_MC_MEM_BIST) ? -1 : 1;
3455 	tests->registers = efx_ef10_run_bist(efx, MC_CMD_REG_BIST) ? -1 : 1;
3456 
3457 	rc = efx_mcdi_reset(efx, RESET_TYPE_WORLD);
3458 
3459 out:
3460 	if (rc == -EPERM)
3461 		rc = 0;
3462 	rc2 = efx_reset_up(efx, RESET_TYPE_WORLD, rc == 0);
3463 	return rc ? rc : rc2;
3464 }
3465 
3466 #ifdef CONFIG_SFC_MTD
3467 
3468 struct efx_ef10_nvram_type_info {
3469 	u16 type, type_mask;
3470 	u8 port;
3471 	const char *name;
3472 };
3473 
3474 static const struct efx_ef10_nvram_type_info efx_ef10_nvram_types[] = {
3475 	{ NVRAM_PARTITION_TYPE_MC_FIRMWARE,	   0,    0, "sfc_mcfw" },
3476 	{ NVRAM_PARTITION_TYPE_MC_FIRMWARE_BACKUP, 0,    0, "sfc_mcfw_backup" },
3477 	{ NVRAM_PARTITION_TYPE_EXPANSION_ROM,	   0,    0, "sfc_exp_rom" },
3478 	{ NVRAM_PARTITION_TYPE_STATIC_CONFIG,	   0,    0, "sfc_static_cfg" },
3479 	{ NVRAM_PARTITION_TYPE_DYNAMIC_CONFIG,	   0,    0, "sfc_dynamic_cfg" },
3480 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT0, 0,   0, "sfc_exp_rom_cfg" },
3481 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT1, 0,   1, "sfc_exp_rom_cfg" },
3482 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT2, 0,   2, "sfc_exp_rom_cfg" },
3483 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT3, 0,   3, "sfc_exp_rom_cfg" },
3484 	{ NVRAM_PARTITION_TYPE_LICENSE,		   0,    0, "sfc_license" },
3485 	{ NVRAM_PARTITION_TYPE_PHY_MIN,		   0xff, 0, "sfc_phy_fw" },
3486 	{ NVRAM_PARTITION_TYPE_MUM_FIRMWARE,	   0,    0, "sfc_mumfw" },
3487 	{ NVRAM_PARTITION_TYPE_EXPANSION_UEFI,	   0,    0, "sfc_uefi" },
3488 	{ NVRAM_PARTITION_TYPE_DYNCONFIG_DEFAULTS, 0,    0, "sfc_dynamic_cfg_dflt" },
3489 	{ NVRAM_PARTITION_TYPE_ROMCONFIG_DEFAULTS, 0,    0, "sfc_exp_rom_cfg_dflt" },
3490 	{ NVRAM_PARTITION_TYPE_STATUS,		   0,    0, "sfc_status" },
3491 	{ NVRAM_PARTITION_TYPE_BUNDLE,		   0,    0, "sfc_bundle" },
3492 	{ NVRAM_PARTITION_TYPE_BUNDLE_METADATA,	   0,    0, "sfc_bundle_metadata" },
3493 };
3494 #define EF10_NVRAM_PARTITION_COUNT	ARRAY_SIZE(efx_ef10_nvram_types)
3495 
3496 static int efx_ef10_mtd_probe_partition(struct efx_nic *efx,
3497 					struct efx_mcdi_mtd_partition *part,
3498 					unsigned int type,
3499 					unsigned long *found)
3500 {
3501 	MCDI_DECLARE_BUF(inbuf, MC_CMD_NVRAM_METADATA_IN_LEN);
3502 	MCDI_DECLARE_BUF(outbuf, MC_CMD_NVRAM_METADATA_OUT_LENMAX);
3503 	const struct efx_ef10_nvram_type_info *info;
3504 	size_t size, erase_size, outlen;
3505 	int type_idx = 0;
3506 	bool protected;
3507 	int rc;
3508 
3509 	for (type_idx = 0; ; type_idx++) {
3510 		if (type_idx == EF10_NVRAM_PARTITION_COUNT)
3511 			return -ENODEV;
3512 		info = efx_ef10_nvram_types + type_idx;
3513 		if ((type & ~info->type_mask) == info->type)
3514 			break;
3515 	}
3516 	if (info->port != efx_port_num(efx))
3517 		return -ENODEV;
3518 
3519 	rc = efx_mcdi_nvram_info(efx, type, &size, &erase_size, &protected);
3520 	if (rc)
3521 		return rc;
3522 	if (protected &&
3523 	    (type != NVRAM_PARTITION_TYPE_DYNCONFIG_DEFAULTS &&
3524 	     type != NVRAM_PARTITION_TYPE_ROMCONFIG_DEFAULTS))
3525 		/* Hide protected partitions that don't provide defaults. */
3526 		return -ENODEV;
3527 
3528 	if (protected)
3529 		/* Protected partitions are read only. */
3530 		erase_size = 0;
3531 
3532 	/* If we've already exposed a partition of this type, hide this
3533 	 * duplicate.  All operations on MTDs are keyed by the type anyway,
3534 	 * so we can't act on the duplicate.
3535 	 */
3536 	if (__test_and_set_bit(type_idx, found))
3537 		return -EEXIST;
3538 
3539 	part->nvram_type = type;
3540 
3541 	MCDI_SET_DWORD(inbuf, NVRAM_METADATA_IN_TYPE, type);
3542 	rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_METADATA, inbuf, sizeof(inbuf),
3543 			  outbuf, sizeof(outbuf), &outlen);
3544 	if (rc)
3545 		return rc;
3546 	if (outlen < MC_CMD_NVRAM_METADATA_OUT_LENMIN)
3547 		return -EIO;
3548 	if (MCDI_DWORD(outbuf, NVRAM_METADATA_OUT_FLAGS) &
3549 	    (1 << MC_CMD_NVRAM_METADATA_OUT_SUBTYPE_VALID_LBN))
3550 		part->fw_subtype = MCDI_DWORD(outbuf,
3551 					      NVRAM_METADATA_OUT_SUBTYPE);
3552 
3553 	part->common.dev_type_name = "EF10 NVRAM manager";
3554 	part->common.type_name = info->name;
3555 
3556 	part->common.mtd.type = MTD_NORFLASH;
3557 	part->common.mtd.flags = MTD_CAP_NORFLASH;
3558 	part->common.mtd.size = size;
3559 	part->common.mtd.erasesize = erase_size;
3560 	/* sfc_status is read-only */
3561 	if (!erase_size)
3562 		part->common.mtd.flags |= MTD_NO_ERASE;
3563 
3564 	return 0;
3565 }
3566 
3567 static int efx_ef10_mtd_probe(struct efx_nic *efx)
3568 {
3569 	MCDI_DECLARE_BUF(outbuf, MC_CMD_NVRAM_PARTITIONS_OUT_LENMAX);
3570 	DECLARE_BITMAP(found, EF10_NVRAM_PARTITION_COUNT) = { 0 };
3571 	struct efx_mcdi_mtd_partition *parts;
3572 	size_t outlen, n_parts_total, i, n_parts;
3573 	unsigned int type;
3574 	int rc;
3575 
3576 	ASSERT_RTNL();
3577 
3578 	BUILD_BUG_ON(MC_CMD_NVRAM_PARTITIONS_IN_LEN != 0);
3579 	rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_PARTITIONS, NULL, 0,
3580 			  outbuf, sizeof(outbuf), &outlen);
3581 	if (rc)
3582 		return rc;
3583 	if (outlen < MC_CMD_NVRAM_PARTITIONS_OUT_LENMIN)
3584 		return -EIO;
3585 
3586 	n_parts_total = MCDI_DWORD(outbuf, NVRAM_PARTITIONS_OUT_NUM_PARTITIONS);
3587 	if (n_parts_total >
3588 	    MCDI_VAR_ARRAY_LEN(outlen, NVRAM_PARTITIONS_OUT_TYPE_ID))
3589 		return -EIO;
3590 
3591 	parts = kcalloc(n_parts_total, sizeof(*parts), GFP_KERNEL);
3592 	if (!parts)
3593 		return -ENOMEM;
3594 
3595 	n_parts = 0;
3596 	for (i = 0; i < n_parts_total; i++) {
3597 		type = MCDI_ARRAY_DWORD(outbuf, NVRAM_PARTITIONS_OUT_TYPE_ID,
3598 					i);
3599 		rc = efx_ef10_mtd_probe_partition(efx, &parts[n_parts], type,
3600 						  found);
3601 		if (rc == -EEXIST || rc == -ENODEV)
3602 			continue;
3603 		if (rc)
3604 			goto fail;
3605 		n_parts++;
3606 	}
3607 
3608 	if (!n_parts) {
3609 		kfree(parts);
3610 		return 0;
3611 	}
3612 
3613 	rc = efx_mtd_add(efx, &parts[0].common, n_parts, sizeof(*parts));
3614 fail:
3615 	if (rc)
3616 		kfree(parts);
3617 	return rc;
3618 }
3619 
3620 #endif /* CONFIG_SFC_MTD */
3621 
3622 static void efx_ef10_ptp_write_host_time(struct efx_nic *efx, u32 host_time)
3623 {
3624 	_efx_writed(efx, cpu_to_le32(host_time), ER_DZ_MC_DB_LWRD);
3625 }
3626 
3627 static void efx_ef10_ptp_write_host_time_vf(struct efx_nic *efx,
3628 					    u32 host_time) {}
3629 
3630 static int efx_ef10_rx_enable_timestamping(struct efx_channel *channel,
3631 					   bool temp)
3632 {
3633 	MCDI_DECLARE_BUF(inbuf, MC_CMD_PTP_IN_TIME_EVENT_SUBSCRIBE_LEN);
3634 	int rc;
3635 
3636 	if (channel->sync_events_state == SYNC_EVENTS_REQUESTED ||
3637 	    channel->sync_events_state == SYNC_EVENTS_VALID ||
3638 	    (temp && channel->sync_events_state == SYNC_EVENTS_DISABLED))
3639 		return 0;
3640 	channel->sync_events_state = SYNC_EVENTS_REQUESTED;
3641 
3642 	MCDI_SET_DWORD(inbuf, PTP_IN_OP, MC_CMD_PTP_OP_TIME_EVENT_SUBSCRIBE);
3643 	MCDI_SET_DWORD(inbuf, PTP_IN_PERIPH_ID, 0);
3644 	MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_SUBSCRIBE_QUEUE,
3645 		       channel->channel);
3646 
3647 	rc = efx_mcdi_rpc(channel->efx, MC_CMD_PTP,
3648 			  inbuf, sizeof(inbuf), NULL, 0, NULL);
3649 
3650 	if (rc != 0)
3651 		channel->sync_events_state = temp ? SYNC_EVENTS_QUIESCENT :
3652 						    SYNC_EVENTS_DISABLED;
3653 
3654 	return rc;
3655 }
3656 
3657 static int efx_ef10_rx_disable_timestamping(struct efx_channel *channel,
3658 					    bool temp)
3659 {
3660 	MCDI_DECLARE_BUF(inbuf, MC_CMD_PTP_IN_TIME_EVENT_UNSUBSCRIBE_LEN);
3661 	int rc;
3662 
3663 	if (channel->sync_events_state == SYNC_EVENTS_DISABLED ||
3664 	    (temp && channel->sync_events_state == SYNC_EVENTS_QUIESCENT))
3665 		return 0;
3666 	if (channel->sync_events_state == SYNC_EVENTS_QUIESCENT) {
3667 		channel->sync_events_state = SYNC_EVENTS_DISABLED;
3668 		return 0;
3669 	}
3670 	channel->sync_events_state = temp ? SYNC_EVENTS_QUIESCENT :
3671 					    SYNC_EVENTS_DISABLED;
3672 
3673 	MCDI_SET_DWORD(inbuf, PTP_IN_OP, MC_CMD_PTP_OP_TIME_EVENT_UNSUBSCRIBE);
3674 	MCDI_SET_DWORD(inbuf, PTP_IN_PERIPH_ID, 0);
3675 	MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_UNSUBSCRIBE_CONTROL,
3676 		       MC_CMD_PTP_IN_TIME_EVENT_UNSUBSCRIBE_SINGLE);
3677 	MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_UNSUBSCRIBE_QUEUE,
3678 		       channel->channel);
3679 
3680 	rc = efx_mcdi_rpc(channel->efx, MC_CMD_PTP,
3681 			  inbuf, sizeof(inbuf), NULL, 0, NULL);
3682 
3683 	return rc;
3684 }
3685 
3686 static int efx_ef10_ptp_set_ts_sync_events(struct efx_nic *efx, bool en,
3687 					   bool temp)
3688 {
3689 	int (*set)(struct efx_channel *channel, bool temp);
3690 	struct efx_channel *channel;
3691 
3692 	set = en ?
3693 	      efx_ef10_rx_enable_timestamping :
3694 	      efx_ef10_rx_disable_timestamping;
3695 
3696 	channel = efx_ptp_channel(efx);
3697 	if (channel) {
3698 		int rc = set(channel, temp);
3699 		if (en && rc != 0) {
3700 			efx_ef10_ptp_set_ts_sync_events(efx, false, temp);
3701 			return rc;
3702 		}
3703 	}
3704 
3705 	return 0;
3706 }
3707 
3708 static int efx_ef10_ptp_set_ts_config_vf(struct efx_nic *efx,
3709 					 struct kernel_hwtstamp_config *init)
3710 {
3711 	return -EOPNOTSUPP;
3712 }
3713 
3714 static int efx_ef10_ptp_set_ts_config(struct efx_nic *efx,
3715 				      struct kernel_hwtstamp_config *init)
3716 {
3717 	int rc;
3718 
3719 	switch (init->rx_filter) {
3720 	case HWTSTAMP_FILTER_NONE:
3721 		efx_ef10_ptp_set_ts_sync_events(efx, false, false);
3722 		/* if TX timestamping is still requested then leave PTP on */
3723 		return efx_ptp_change_mode(efx,
3724 					   init->tx_type != HWTSTAMP_TX_OFF, 0);
3725 	case HWTSTAMP_FILTER_ALL:
3726 	case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
3727 	case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
3728 	case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
3729 	case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
3730 	case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
3731 	case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
3732 	case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
3733 	case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
3734 	case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
3735 	case HWTSTAMP_FILTER_PTP_V2_EVENT:
3736 	case HWTSTAMP_FILTER_PTP_V2_SYNC:
3737 	case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
3738 	case HWTSTAMP_FILTER_NTP_ALL:
3739 		init->rx_filter = HWTSTAMP_FILTER_ALL;
3740 		rc = efx_ptp_change_mode(efx, true, 0);
3741 		if (!rc)
3742 			rc = efx_ef10_ptp_set_ts_sync_events(efx, true, false);
3743 		if (rc)
3744 			efx_ptp_change_mode(efx, false, 0);
3745 		return rc;
3746 	default:
3747 		return -ERANGE;
3748 	}
3749 }
3750 
3751 static int efx_ef10_get_phys_port_id(struct efx_nic *efx,
3752 				     struct netdev_phys_item_id *ppid)
3753 {
3754 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
3755 
3756 	if (!is_valid_ether_addr(nic_data->port_id))
3757 		return -EOPNOTSUPP;
3758 
3759 	ppid->id_len = ETH_ALEN;
3760 	memcpy(ppid->id, nic_data->port_id, ppid->id_len);
3761 
3762 	return 0;
3763 }
3764 
3765 static int efx_ef10_vlan_rx_add_vid(struct efx_nic *efx, __be16 proto, u16 vid)
3766 {
3767 	if (proto != htons(ETH_P_8021Q))
3768 		return -EINVAL;
3769 
3770 	return efx_ef10_add_vlan(efx, vid);
3771 }
3772 
3773 static int efx_ef10_vlan_rx_kill_vid(struct efx_nic *efx, __be16 proto, u16 vid)
3774 {
3775 	if (proto != htons(ETH_P_8021Q))
3776 		return -EINVAL;
3777 
3778 	return efx_ef10_del_vlan(efx, vid);
3779 }
3780 
3781 /* We rely on the MCDI wiping out our TX rings if it made any changes to the
3782  * ports table, ensuring that any TSO descriptors that were made on a now-
3783  * removed tunnel port will be blown away and won't break things when we try
3784  * to transmit them using the new ports table.
3785  */
3786 static int efx_ef10_set_udp_tnl_ports(struct efx_nic *efx, bool unloading)
3787 {
3788 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
3789 	MCDI_DECLARE_BUF(inbuf, MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_IN_LENMAX);
3790 	MCDI_DECLARE_BUF(outbuf, MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_OUT_LEN);
3791 	bool will_reset = false;
3792 	size_t num_entries = 0;
3793 	size_t inlen, outlen;
3794 	size_t i;
3795 	int rc;
3796 	efx_dword_t flags_and_num_entries;
3797 
3798 	WARN_ON(!mutex_is_locked(&nic_data->udp_tunnels_lock));
3799 
3800 	nic_data->udp_tunnels_dirty = false;
3801 
3802 	if (!(nic_data->datapath_caps &
3803 	    (1 << MC_CMD_GET_CAPABILITIES_OUT_VXLAN_NVGRE_LBN))) {
3804 		efx_device_attach_if_not_resetting(efx);
3805 		return 0;
3806 	}
3807 
3808 	BUILD_BUG_ON(ARRAY_SIZE(nic_data->udp_tunnels) >
3809 		     MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_IN_ENTRIES_MAXNUM);
3810 
3811 	for (i = 0; i < ARRAY_SIZE(nic_data->udp_tunnels); ++i) {
3812 		if (nic_data->udp_tunnels[i].type !=
3813 		    TUNNEL_ENCAP_UDP_PORT_ENTRY_INVALID) {
3814 			efx_dword_t entry;
3815 
3816 			EFX_POPULATE_DWORD_2(entry,
3817 				TUNNEL_ENCAP_UDP_PORT_ENTRY_UDP_PORT,
3818 					ntohs(nic_data->udp_tunnels[i].port),
3819 				TUNNEL_ENCAP_UDP_PORT_ENTRY_PROTOCOL,
3820 					nic_data->udp_tunnels[i].type);
3821 			*_MCDI_ARRAY_DWORD(inbuf,
3822 				SET_TUNNEL_ENCAP_UDP_PORTS_IN_ENTRIES,
3823 				num_entries++) = entry;
3824 		}
3825 	}
3826 
3827 	BUILD_BUG_ON((MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_IN_NUM_ENTRIES_OFST -
3828 		      MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_IN_FLAGS_OFST) * 8 !=
3829 		     EFX_WORD_1_LBN);
3830 	BUILD_BUG_ON(MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_IN_NUM_ENTRIES_LEN * 8 !=
3831 		     EFX_WORD_1_WIDTH);
3832 	EFX_POPULATE_DWORD_2(flags_and_num_entries,
3833 			     MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_IN_UNLOADING,
3834 				!!unloading,
3835 			     EFX_WORD_1, num_entries);
3836 	*_MCDI_DWORD(inbuf, SET_TUNNEL_ENCAP_UDP_PORTS_IN_FLAGS) =
3837 		flags_and_num_entries;
3838 
3839 	inlen = MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_IN_LEN(num_entries);
3840 
3841 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS,
3842 				inbuf, inlen, outbuf, sizeof(outbuf), &outlen);
3843 	if (rc == -EIO) {
3844 		/* Most likely the MC rebooted due to another function also
3845 		 * setting its tunnel port list. Mark the tunnel port list as
3846 		 * dirty, so it will be pushed upon coming up from the reboot.
3847 		 */
3848 		nic_data->udp_tunnels_dirty = true;
3849 		return 0;
3850 	}
3851 
3852 	if (rc) {
3853 		/* expected not available on unprivileged functions */
3854 		if (rc != -EPERM)
3855 			netif_warn(efx, drv, efx->net_dev,
3856 				   "Unable to set UDP tunnel ports; rc=%d.\n", rc);
3857 	} else if (MCDI_DWORD(outbuf, SET_TUNNEL_ENCAP_UDP_PORTS_OUT_FLAGS) &
3858 		   (1 << MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_OUT_RESETTING_LBN)) {
3859 		netif_info(efx, drv, efx->net_dev,
3860 			   "Rebooting MC due to UDP tunnel port list change\n");
3861 		will_reset = true;
3862 		if (unloading)
3863 			/* Delay for the MC reset to complete. This will make
3864 			 * unloading other functions a bit smoother. This is a
3865 			 * race, but the other unload will work whichever way
3866 			 * it goes, this just avoids an unnecessary error
3867 			 * message.
3868 			 */
3869 			msleep(100);
3870 	}
3871 	if (!will_reset && !unloading) {
3872 		/* The caller will have detached, relying on the MC reset to
3873 		 * trigger a re-attach.  Since there won't be an MC reset, we
3874 		 * have to do the attach ourselves.
3875 		 */
3876 		efx_device_attach_if_not_resetting(efx);
3877 	}
3878 
3879 	return rc;
3880 }
3881 
3882 static int efx_ef10_udp_tnl_push_ports(struct efx_nic *efx)
3883 {
3884 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
3885 	int rc = 0;
3886 
3887 	mutex_lock(&nic_data->udp_tunnels_lock);
3888 	if (nic_data->udp_tunnels_dirty) {
3889 		/* Make sure all TX are stopped while we modify the table, else
3890 		 * we might race against an efx_features_check().
3891 		 */
3892 		efx_device_detach_sync(efx);
3893 		rc = efx_ef10_set_udp_tnl_ports(efx, false);
3894 	}
3895 	mutex_unlock(&nic_data->udp_tunnels_lock);
3896 	return rc;
3897 }
3898 
3899 static int efx_ef10_udp_tnl_set_port(struct net_device *dev,
3900 				     unsigned int table, unsigned int entry,
3901 				     struct udp_tunnel_info *ti)
3902 {
3903 	struct efx_nic *efx = efx_netdev_priv(dev);
3904 	struct efx_ef10_nic_data *nic_data;
3905 	int efx_tunnel_type, rc;
3906 
3907 	if (ti->type == UDP_TUNNEL_TYPE_VXLAN)
3908 		efx_tunnel_type = TUNNEL_ENCAP_UDP_PORT_ENTRY_VXLAN;
3909 	else
3910 		efx_tunnel_type = TUNNEL_ENCAP_UDP_PORT_ENTRY_GENEVE;
3911 
3912 	nic_data = efx->nic_data;
3913 	if (!(nic_data->datapath_caps &
3914 	      (1 << MC_CMD_GET_CAPABILITIES_OUT_VXLAN_NVGRE_LBN)))
3915 		return -EOPNOTSUPP;
3916 
3917 	mutex_lock(&nic_data->udp_tunnels_lock);
3918 	/* Make sure all TX are stopped while we add to the table, else we
3919 	 * might race against an efx_features_check().
3920 	 */
3921 	efx_device_detach_sync(efx);
3922 	nic_data->udp_tunnels[entry].type = efx_tunnel_type;
3923 	nic_data->udp_tunnels[entry].port = ti->port;
3924 	rc = efx_ef10_set_udp_tnl_ports(efx, false);
3925 	mutex_unlock(&nic_data->udp_tunnels_lock);
3926 
3927 	return rc;
3928 }
3929 
3930 /* Called under the TX lock with the TX queue running, hence no-one can be
3931  * in the middle of updating the UDP tunnels table.  However, they could
3932  * have tried and failed the MCDI, in which case they'll have set the dirty
3933  * flag before dropping their locks.
3934  */
3935 static bool efx_ef10_udp_tnl_has_port(struct efx_nic *efx, __be16 port)
3936 {
3937 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
3938 	size_t i;
3939 
3940 	if (!(nic_data->datapath_caps &
3941 	      (1 << MC_CMD_GET_CAPABILITIES_OUT_VXLAN_NVGRE_LBN)))
3942 		return false;
3943 
3944 	if (nic_data->udp_tunnels_dirty)
3945 		/* SW table may not match HW state, so just assume we can't
3946 		 * use any UDP tunnel offloads.
3947 		 */
3948 		return false;
3949 
3950 	for (i = 0; i < ARRAY_SIZE(nic_data->udp_tunnels); ++i)
3951 		if (nic_data->udp_tunnels[i].type !=
3952 		    TUNNEL_ENCAP_UDP_PORT_ENTRY_INVALID &&
3953 		    nic_data->udp_tunnels[i].port == port)
3954 			return true;
3955 
3956 	return false;
3957 }
3958 
3959 static int efx_ef10_udp_tnl_unset_port(struct net_device *dev,
3960 				       unsigned int table, unsigned int entry,
3961 				       struct udp_tunnel_info *ti)
3962 {
3963 	struct efx_nic *efx = efx_netdev_priv(dev);
3964 	struct efx_ef10_nic_data *nic_data;
3965 	int rc;
3966 
3967 	nic_data = efx->nic_data;
3968 
3969 	mutex_lock(&nic_data->udp_tunnels_lock);
3970 	/* Make sure all TX are stopped while we remove from the table, else we
3971 	 * might race against an efx_features_check().
3972 	 */
3973 	efx_device_detach_sync(efx);
3974 	nic_data->udp_tunnels[entry].type = TUNNEL_ENCAP_UDP_PORT_ENTRY_INVALID;
3975 	nic_data->udp_tunnels[entry].port = 0;
3976 	rc = efx_ef10_set_udp_tnl_ports(efx, false);
3977 	mutex_unlock(&nic_data->udp_tunnels_lock);
3978 
3979 	return rc;
3980 }
3981 
3982 static const struct udp_tunnel_nic_info efx_ef10_udp_tunnels = {
3983 	.set_port	= efx_ef10_udp_tnl_set_port,
3984 	.unset_port	= efx_ef10_udp_tnl_unset_port,
3985 	.flags          = UDP_TUNNEL_NIC_INFO_MAY_SLEEP,
3986 	.tables         = {
3987 		{
3988 			.n_entries = 16,
3989 			.tunnel_types = UDP_TUNNEL_TYPE_VXLAN |
3990 					UDP_TUNNEL_TYPE_GENEVE,
3991 		},
3992 	},
3993 };
3994 
3995 /* EF10 may have multiple datapath firmware variants within a
3996  * single version.  Report which variants are running.
3997  */
3998 static size_t efx_ef10_print_additional_fwver(struct efx_nic *efx, char *buf,
3999 					      size_t len)
4000 {
4001 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
4002 
4003 	return scnprintf(buf, len, " rx%x tx%x",
4004 			 nic_data->rx_dpcpu_fw_id,
4005 			 nic_data->tx_dpcpu_fw_id);
4006 }
4007 
4008 static unsigned int ef10_check_caps(const struct efx_nic *efx,
4009 				    u8 flag,
4010 				    u32 offset)
4011 {
4012 	const struct efx_ef10_nic_data *nic_data = efx->nic_data;
4013 
4014 	switch (offset) {
4015 	case(MC_CMD_GET_CAPABILITIES_V4_OUT_FLAGS1_OFST):
4016 		return nic_data->datapath_caps & BIT_ULL(flag);
4017 	case(MC_CMD_GET_CAPABILITIES_V4_OUT_FLAGS2_OFST):
4018 		return nic_data->datapath_caps2 & BIT_ULL(flag);
4019 	default:
4020 		return 0;
4021 	}
4022 }
4023 
4024 static unsigned int efx_ef10_recycle_ring_size(const struct efx_nic *efx)
4025 {
4026 	unsigned int ret = EFX_RECYCLE_RING_SIZE_10G;
4027 
4028 	/* There is no difference between PFs and VFs. The side is based on
4029 	 * the maximum link speed of a given NIC.
4030 	 */
4031 	switch (efx->pci_dev->device & 0xfff) {
4032 	case 0x0903:	/* Farmingdale can do up to 10G */
4033 		break;
4034 	case 0x0923:	/* Greenport can do up to 40G */
4035 	case 0x0a03:	/* Medford can do up to 40G */
4036 		ret *= 4;
4037 		break;
4038 	default:	/* Medford2 can do up to 100G */
4039 		ret *= 10;
4040 	}
4041 
4042 	if (IS_ENABLED(CONFIG_PPC64))
4043 		ret *= 4;
4044 
4045 	return ret;
4046 }
4047 
4048 #define EF10_OFFLOAD_FEATURES		\
4049 	(NETIF_F_IP_CSUM |		\
4050 	 NETIF_F_HW_VLAN_CTAG_FILTER |	\
4051 	 NETIF_F_IPV6_CSUM |		\
4052 	 NETIF_F_RXHASH |		\
4053 	 NETIF_F_NTUPLE |		\
4054 	 NETIF_F_SG |			\
4055 	 NETIF_F_RXCSUM |		\
4056 	 NETIF_F_RXALL)
4057 
4058 const struct efx_nic_type efx_hunt_a0_vf_nic_type = {
4059 	.is_vf = true,
4060 	.mem_bar = efx_ef10_vf_mem_bar,
4061 	.mem_map_size = efx_ef10_mem_map_size,
4062 	.probe = efx_ef10_probe_vf,
4063 	.remove = efx_ef10_remove,
4064 	.dimension_resources = efx_ef10_dimension_resources,
4065 	.init = efx_ef10_init_nic,
4066 	.fini = efx_ef10_fini_nic,
4067 	.map_reset_reason = efx_ef10_map_reset_reason,
4068 	.map_reset_flags = efx_ef10_map_reset_flags,
4069 	.reset = efx_ef10_reset,
4070 	.probe_port = efx_mcdi_port_probe,
4071 	.remove_port = efx_mcdi_port_remove,
4072 	.fini_dmaq = efx_fini_dmaq,
4073 	.prepare_flr = efx_ef10_prepare_flr,
4074 	.finish_flr = efx_port_dummy_op_void,
4075 	.describe_stats = efx_ef10_describe_stats,
4076 	.update_stats = efx_ef10_update_stats_vf,
4077 	.update_stats_atomic = efx_ef10_update_stats_atomic_vf,
4078 	.start_stats = efx_port_dummy_op_void,
4079 	.pull_stats = efx_port_dummy_op_void,
4080 	.stop_stats = efx_port_dummy_op_void,
4081 	.push_irq_moderation = efx_ef10_push_irq_moderation,
4082 	.reconfigure_mac = efx_ef10_mac_reconfigure,
4083 	.check_mac_fault = efx_mcdi_mac_check_fault,
4084 	.reconfigure_port = efx_mcdi_port_reconfigure,
4085 	.get_wol = efx_ef10_get_wol_vf,
4086 	.set_wol = efx_ef10_set_wol_vf,
4087 	.resume_wol = efx_port_dummy_op_void,
4088 	.mcdi_request = efx_ef10_mcdi_request,
4089 	.mcdi_poll_response = efx_ef10_mcdi_poll_response,
4090 	.mcdi_read_response = efx_ef10_mcdi_read_response,
4091 	.mcdi_poll_reboot = efx_ef10_mcdi_poll_reboot,
4092 	.mcdi_reboot_detected = efx_ef10_mcdi_reboot_detected,
4093 	.irq_enable_master = efx_port_dummy_op_void,
4094 	.irq_test_generate = efx_ef10_irq_test_generate,
4095 	.irq_disable_non_ev = efx_port_dummy_op_void,
4096 	.irq_handle_msi = efx_ef10_msi_interrupt,
4097 	.irq_handle_legacy = efx_ef10_legacy_interrupt,
4098 	.tx_probe = efx_ef10_tx_probe,
4099 	.tx_init = efx_ef10_tx_init,
4100 	.tx_remove = efx_mcdi_tx_remove,
4101 	.tx_write = efx_ef10_tx_write,
4102 	.tx_limit_len = efx_ef10_tx_limit_len,
4103 	.tx_enqueue = __efx_enqueue_skb,
4104 	.rx_push_rss_config = efx_mcdi_vf_rx_push_rss_config,
4105 	.rx_pull_rss_config = efx_mcdi_rx_pull_rss_config,
4106 	.rx_probe = efx_mcdi_rx_probe,
4107 	.rx_init = efx_mcdi_rx_init,
4108 	.rx_remove = efx_mcdi_rx_remove,
4109 	.rx_write = efx_ef10_rx_write,
4110 	.rx_defer_refill = efx_ef10_rx_defer_refill,
4111 	.rx_packet = __efx_rx_packet,
4112 	.ev_probe = efx_mcdi_ev_probe,
4113 	.ev_init = efx_ef10_ev_init,
4114 	.ev_fini = efx_mcdi_ev_fini,
4115 	.ev_remove = efx_mcdi_ev_remove,
4116 	.ev_process = efx_ef10_ev_process,
4117 	.ev_read_ack = efx_ef10_ev_read_ack,
4118 	.ev_test_generate = efx_ef10_ev_test_generate,
4119 	.filter_table_probe = efx_ef10_filter_table_probe,
4120 	.filter_table_restore = efx_mcdi_filter_table_restore,
4121 	.filter_table_remove = efx_ef10_filter_table_remove,
4122 	.filter_update_rx_scatter = efx_mcdi_update_rx_scatter,
4123 	.filter_insert = efx_mcdi_filter_insert,
4124 	.filter_remove_safe = efx_mcdi_filter_remove_safe,
4125 	.filter_get_safe = efx_mcdi_filter_get_safe,
4126 	.filter_clear_rx = efx_mcdi_filter_clear_rx,
4127 	.filter_count_rx_used = efx_mcdi_filter_count_rx_used,
4128 	.filter_get_rx_id_limit = efx_mcdi_filter_get_rx_id_limit,
4129 	.filter_get_rx_ids = efx_mcdi_filter_get_rx_ids,
4130 #ifdef CONFIG_RFS_ACCEL
4131 	.filter_rfs_expire_one = efx_mcdi_filter_rfs_expire_one,
4132 #endif
4133 #ifdef CONFIG_SFC_MTD
4134 	.mtd_probe = efx_port_dummy_op_int,
4135 #endif
4136 	.ptp_write_host_time = efx_ef10_ptp_write_host_time_vf,
4137 	.ptp_set_ts_config = efx_ef10_ptp_set_ts_config_vf,
4138 	.vlan_rx_add_vid = efx_ef10_vlan_rx_add_vid,
4139 	.vlan_rx_kill_vid = efx_ef10_vlan_rx_kill_vid,
4140 #ifdef CONFIG_SFC_SRIOV
4141 	.vswitching_probe = efx_ef10_vswitching_probe_vf,
4142 	.vswitching_restore = efx_ef10_vswitching_restore_vf,
4143 	.vswitching_remove = efx_ef10_vswitching_remove_vf,
4144 #endif
4145 	.get_mac_address = efx_ef10_get_mac_address_vf,
4146 	.set_mac_address = efx_ef10_set_mac_address,
4147 
4148 	.get_phys_port_id = efx_ef10_get_phys_port_id,
4149 	.revision = EFX_REV_HUNT_A0,
4150 	.max_dma_mask = DMA_BIT_MASK(ESF_DZ_TX_KER_BUF_ADDR_WIDTH),
4151 	.rx_prefix_size = ES_DZ_RX_PREFIX_SIZE,
4152 	.rx_hash_offset = ES_DZ_RX_PREFIX_HASH_OFST,
4153 	.rx_ts_offset = ES_DZ_RX_PREFIX_TSTAMP_OFST,
4154 	.can_rx_scatter = true,
4155 	.always_rx_scatter = true,
4156 	.min_interrupt_mode = EFX_INT_MODE_MSIX,
4157 	.timer_period_max = 1 << ERF_DD_EVQ_IND_TIMER_VAL_WIDTH,
4158 	.offload_features = EF10_OFFLOAD_FEATURES,
4159 	.mcdi_max_ver = 2,
4160 	.max_rx_ip_filters = EFX_MCDI_FILTER_TBL_ROWS,
4161 	.hwtstamp_filters = 1 << HWTSTAMP_FILTER_NONE |
4162 			    1 << HWTSTAMP_FILTER_ALL,
4163 	.rx_hash_key_size = 40,
4164 	.check_caps = ef10_check_caps,
4165 	.print_additional_fwver = efx_ef10_print_additional_fwver,
4166 	.sensor_event = efx_mcdi_sensor_event,
4167 	.rx_recycle_ring_size = efx_ef10_recycle_ring_size,
4168 };
4169 
4170 const struct efx_nic_type efx_hunt_a0_nic_type = {
4171 	.is_vf = false,
4172 	.mem_bar = efx_ef10_pf_mem_bar,
4173 	.mem_map_size = efx_ef10_mem_map_size,
4174 	.probe = efx_ef10_probe_pf,
4175 	.remove = efx_ef10_remove,
4176 	.dimension_resources = efx_ef10_dimension_resources,
4177 	.init = efx_ef10_init_nic,
4178 	.fini = efx_ef10_fini_nic,
4179 	.map_reset_reason = efx_ef10_map_reset_reason,
4180 	.map_reset_flags = efx_ef10_map_reset_flags,
4181 	.reset = efx_ef10_reset,
4182 	.probe_port = efx_mcdi_port_probe,
4183 	.remove_port = efx_mcdi_port_remove,
4184 	.fini_dmaq = efx_fini_dmaq,
4185 	.prepare_flr = efx_ef10_prepare_flr,
4186 	.finish_flr = efx_port_dummy_op_void,
4187 	.describe_stats = efx_ef10_describe_stats,
4188 	.update_stats = efx_ef10_update_stats_pf,
4189 	.start_stats = efx_mcdi_mac_start_stats,
4190 	.pull_stats = efx_mcdi_mac_pull_stats,
4191 	.stop_stats = efx_mcdi_mac_stop_stats,
4192 	.push_irq_moderation = efx_ef10_push_irq_moderation,
4193 	.reconfigure_mac = efx_ef10_mac_reconfigure,
4194 	.check_mac_fault = efx_mcdi_mac_check_fault,
4195 	.reconfigure_port = efx_mcdi_port_reconfigure,
4196 	.get_wol = efx_ef10_get_wol,
4197 	.set_wol = efx_ef10_set_wol,
4198 	.resume_wol = efx_port_dummy_op_void,
4199 	.get_fec_stats = efx_ef10_get_fec_stats,
4200 	.test_chip = efx_ef10_test_chip,
4201 	.test_nvram = efx_mcdi_nvram_test_all,
4202 	.mcdi_request = efx_ef10_mcdi_request,
4203 	.mcdi_poll_response = efx_ef10_mcdi_poll_response,
4204 	.mcdi_read_response = efx_ef10_mcdi_read_response,
4205 	.mcdi_poll_reboot = efx_ef10_mcdi_poll_reboot,
4206 	.mcdi_reboot_detected = efx_ef10_mcdi_reboot_detected,
4207 	.irq_enable_master = efx_port_dummy_op_void,
4208 	.irq_test_generate = efx_ef10_irq_test_generate,
4209 	.irq_disable_non_ev = efx_port_dummy_op_void,
4210 	.irq_handle_msi = efx_ef10_msi_interrupt,
4211 	.irq_handle_legacy = efx_ef10_legacy_interrupt,
4212 	.tx_probe = efx_ef10_tx_probe,
4213 	.tx_init = efx_ef10_tx_init,
4214 	.tx_remove = efx_mcdi_tx_remove,
4215 	.tx_write = efx_ef10_tx_write,
4216 	.tx_limit_len = efx_ef10_tx_limit_len,
4217 	.tx_enqueue = __efx_enqueue_skb,
4218 	.rx_push_rss_config = efx_mcdi_pf_rx_push_rss_config,
4219 	.rx_pull_rss_config = efx_mcdi_rx_pull_rss_config,
4220 	.rx_push_rss_context_config = efx_mcdi_rx_push_rss_context_config,
4221 	.rx_pull_rss_context_config = efx_mcdi_rx_pull_rss_context_config,
4222 	.rx_restore_rss_contexts = efx_mcdi_rx_restore_rss_contexts,
4223 	.rx_probe = efx_mcdi_rx_probe,
4224 	.rx_init = efx_mcdi_rx_init,
4225 	.rx_remove = efx_mcdi_rx_remove,
4226 	.rx_write = efx_ef10_rx_write,
4227 	.rx_defer_refill = efx_ef10_rx_defer_refill,
4228 	.rx_packet = __efx_rx_packet,
4229 	.ev_probe = efx_mcdi_ev_probe,
4230 	.ev_init = efx_ef10_ev_init,
4231 	.ev_fini = efx_mcdi_ev_fini,
4232 	.ev_remove = efx_mcdi_ev_remove,
4233 	.ev_process = efx_ef10_ev_process,
4234 	.ev_read_ack = efx_ef10_ev_read_ack,
4235 	.ev_test_generate = efx_ef10_ev_test_generate,
4236 	.filter_table_probe = efx_ef10_filter_table_probe,
4237 	.filter_table_restore = efx_mcdi_filter_table_restore,
4238 	.filter_table_remove = efx_ef10_filter_table_remove,
4239 	.filter_update_rx_scatter = efx_mcdi_update_rx_scatter,
4240 	.filter_insert = efx_mcdi_filter_insert,
4241 	.filter_remove_safe = efx_mcdi_filter_remove_safe,
4242 	.filter_get_safe = efx_mcdi_filter_get_safe,
4243 	.filter_clear_rx = efx_mcdi_filter_clear_rx,
4244 	.filter_count_rx_used = efx_mcdi_filter_count_rx_used,
4245 	.filter_get_rx_id_limit = efx_mcdi_filter_get_rx_id_limit,
4246 	.filter_get_rx_ids = efx_mcdi_filter_get_rx_ids,
4247 #ifdef CONFIG_RFS_ACCEL
4248 	.filter_rfs_expire_one = efx_mcdi_filter_rfs_expire_one,
4249 #endif
4250 #ifdef CONFIG_SFC_MTD
4251 	.mtd_probe = efx_ef10_mtd_probe,
4252 	.mtd_rename = efx_mcdi_mtd_rename,
4253 	.mtd_read = efx_mcdi_mtd_read,
4254 	.mtd_erase = efx_mcdi_mtd_erase,
4255 	.mtd_write = efx_mcdi_mtd_write,
4256 	.mtd_sync = efx_mcdi_mtd_sync,
4257 #endif
4258 	.ptp_write_host_time = efx_ef10_ptp_write_host_time,
4259 	.ptp_set_ts_sync_events = efx_ef10_ptp_set_ts_sync_events,
4260 	.ptp_set_ts_config = efx_ef10_ptp_set_ts_config,
4261 	.vlan_rx_add_vid = efx_ef10_vlan_rx_add_vid,
4262 	.vlan_rx_kill_vid = efx_ef10_vlan_rx_kill_vid,
4263 	.udp_tnl_push_ports = efx_ef10_udp_tnl_push_ports,
4264 	.udp_tnl_has_port = efx_ef10_udp_tnl_has_port,
4265 #ifdef CONFIG_SFC_SRIOV
4266 	.sriov_configure = efx_ef10_sriov_configure,
4267 	.sriov_init = efx_ef10_sriov_init,
4268 	.sriov_fini = efx_ef10_sriov_fini,
4269 	.sriov_wanted = efx_ef10_sriov_wanted,
4270 	.sriov_set_vf_mac = efx_ef10_sriov_set_vf_mac,
4271 	.sriov_set_vf_vlan = efx_ef10_sriov_set_vf_vlan,
4272 	.sriov_set_vf_spoofchk = efx_ef10_sriov_set_vf_spoofchk,
4273 	.sriov_get_vf_config = efx_ef10_sriov_get_vf_config,
4274 	.sriov_set_vf_link_state = efx_ef10_sriov_set_vf_link_state,
4275 	.vswitching_probe = efx_ef10_vswitching_probe_pf,
4276 	.vswitching_restore = efx_ef10_vswitching_restore_pf,
4277 	.vswitching_remove = efx_ef10_vswitching_remove_pf,
4278 #endif
4279 	.get_mac_address = efx_ef10_get_mac_address_pf,
4280 	.set_mac_address = efx_ef10_set_mac_address,
4281 	.tso_versions = efx_ef10_tso_versions,
4282 
4283 	.get_phys_port_id = efx_ef10_get_phys_port_id,
4284 	.revision = EFX_REV_HUNT_A0,
4285 	.max_dma_mask = DMA_BIT_MASK(ESF_DZ_TX_KER_BUF_ADDR_WIDTH),
4286 	.rx_prefix_size = ES_DZ_RX_PREFIX_SIZE,
4287 	.rx_hash_offset = ES_DZ_RX_PREFIX_HASH_OFST,
4288 	.rx_ts_offset = ES_DZ_RX_PREFIX_TSTAMP_OFST,
4289 	.can_rx_scatter = true,
4290 	.always_rx_scatter = true,
4291 	.option_descriptors = true,
4292 	.min_interrupt_mode = EFX_INT_MODE_LEGACY,
4293 	.timer_period_max = 1 << ERF_DD_EVQ_IND_TIMER_VAL_WIDTH,
4294 	.offload_features = EF10_OFFLOAD_FEATURES,
4295 	.mcdi_max_ver = 2,
4296 	.max_rx_ip_filters = EFX_MCDI_FILTER_TBL_ROWS,
4297 	.hwtstamp_filters = 1 << HWTSTAMP_FILTER_NONE |
4298 			    1 << HWTSTAMP_FILTER_ALL,
4299 	.rx_hash_key_size = 40,
4300 	.check_caps = ef10_check_caps,
4301 	.print_additional_fwver = efx_ef10_print_additional_fwver,
4302 	.sensor_event = efx_mcdi_sensor_event,
4303 	.rx_recycle_ring_size = efx_ef10_recycle_ring_size,
4304 };
4305