xref: /openbmc/linux/drivers/net/ethernet/sfc/ef10.c (revision cd6d421e)
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 efx_ef10_show_link_control_flag(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 efx_ef10_show_primary_flag(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(link_control_flag, 0444, efx_ef10_show_link_control_flag,
523 		   NULL);
524 static DEVICE_ATTR(primary_flag, 0444, efx_ef10_show_primary_flag, NULL);
525 
526 static int efx_ef10_probe(struct efx_nic *efx)
527 {
528 	struct efx_ef10_nic_data *nic_data;
529 	int i, rc;
530 
531 	nic_data = kzalloc(sizeof(*nic_data), GFP_KERNEL);
532 	if (!nic_data)
533 		return -ENOMEM;
534 	efx->nic_data = nic_data;
535 
536 	/* we assume later that we can copy from this buffer in dwords */
537 	BUILD_BUG_ON(MCDI_CTL_SDU_LEN_MAX_V2 % 4);
538 
539 	rc = efx_nic_alloc_buffer(efx, &nic_data->mcdi_buf,
540 				  8 + MCDI_CTL_SDU_LEN_MAX_V2, GFP_KERNEL);
541 	if (rc)
542 		goto fail1;
543 
544 	/* Get the MC's warm boot count.  In case it's rebooting right
545 	 * now, be prepared to retry.
546 	 */
547 	i = 0;
548 	for (;;) {
549 		rc = efx_ef10_get_warm_boot_count(efx);
550 		if (rc >= 0)
551 			break;
552 		if (++i == 5)
553 			goto fail2;
554 		ssleep(1);
555 	}
556 	nic_data->warm_boot_count = rc;
557 
558 	/* In case we're recovering from a crash (kexec), we want to
559 	 * cancel any outstanding request by the previous user of this
560 	 * function.  We send a special message using the least
561 	 * significant bits of the 'high' (doorbell) register.
562 	 */
563 	_efx_writed(efx, cpu_to_le32(1), ER_DZ_MC_DB_HWRD);
564 
565 	rc = efx_mcdi_init(efx);
566 	if (rc)
567 		goto fail2;
568 
569 	mutex_init(&nic_data->udp_tunnels_lock);
570 	for (i = 0; i < ARRAY_SIZE(nic_data->udp_tunnels); ++i)
571 		nic_data->udp_tunnels[i].type =
572 			TUNNEL_ENCAP_UDP_PORT_ENTRY_INVALID;
573 
574 	/* Reset (most) configuration for this function */
575 	rc = efx_mcdi_reset(efx, RESET_TYPE_ALL);
576 	if (rc)
577 		goto fail3;
578 
579 	/* Enable event logging */
580 	rc = efx_mcdi_log_ctrl(efx, true, false, 0);
581 	if (rc)
582 		goto fail3;
583 
584 	rc = device_create_file(&efx->pci_dev->dev,
585 				&dev_attr_link_control_flag);
586 	if (rc)
587 		goto fail3;
588 
589 	rc = device_create_file(&efx->pci_dev->dev, &dev_attr_primary_flag);
590 	if (rc)
591 		goto fail4;
592 
593 	rc = efx_get_pf_index(efx, &nic_data->pf_index);
594 	if (rc)
595 		goto fail5;
596 
597 	rc = efx_ef10_init_datapath_caps(efx);
598 	if (rc < 0)
599 		goto fail5;
600 
601 	efx_ef10_read_licensed_features(efx);
602 
603 	/* We can have one VI for each vi_stride-byte region.
604 	 * However, until we use TX option descriptors we need up to four
605 	 * TX queues per channel for different checksumming combinations.
606 	 */
607 	if (nic_data->datapath_caps &
608 	    (1 << MC_CMD_GET_CAPABILITIES_OUT_VXLAN_NVGRE_LBN))
609 		efx->tx_queues_per_channel = 4;
610 	else
611 		efx->tx_queues_per_channel = 2;
612 	efx->max_vis = efx_ef10_mem_map_size(efx) / efx->vi_stride;
613 	if (!efx->max_vis) {
614 		netif_err(efx, drv, efx->net_dev, "error determining max VIs\n");
615 		rc = -EIO;
616 		goto fail5;
617 	}
618 	efx->max_channels = min_t(unsigned int, EFX_MAX_CHANNELS,
619 				  efx->max_vis / efx->tx_queues_per_channel);
620 	efx->max_tx_channels = efx->max_channels;
621 	if (WARN_ON(efx->max_channels == 0)) {
622 		rc = -EIO;
623 		goto fail5;
624 	}
625 
626 	efx->rx_packet_len_offset =
627 		ES_DZ_RX_PREFIX_PKTLEN_OFST - ES_DZ_RX_PREFIX_SIZE;
628 
629 	if (nic_data->datapath_caps &
630 	    (1 << MC_CMD_GET_CAPABILITIES_OUT_RX_INCLUDE_FCS_LBN))
631 		efx->net_dev->hw_features |= NETIF_F_RXFCS;
632 
633 	rc = efx_mcdi_port_get_number(efx);
634 	if (rc < 0)
635 		goto fail5;
636 	efx->port_num = rc;
637 
638 	rc = efx->type->get_mac_address(efx, efx->net_dev->perm_addr);
639 	if (rc)
640 		goto fail5;
641 
642 	rc = efx_ef10_get_timer_config(efx);
643 	if (rc < 0)
644 		goto fail5;
645 
646 	rc = efx_mcdi_mon_probe(efx);
647 	if (rc && rc != -EPERM)
648 		goto fail5;
649 
650 	efx_ptp_defer_probe_with_channel(efx);
651 
652 #ifdef CONFIG_SFC_SRIOV
653 	if ((efx->pci_dev->physfn) && (!efx->pci_dev->is_physfn)) {
654 		struct pci_dev *pci_dev_pf = efx->pci_dev->physfn;
655 		struct efx_nic *efx_pf = pci_get_drvdata(pci_dev_pf);
656 
657 		efx_pf->type->get_mac_address(efx_pf, nic_data->port_id);
658 	} else
659 #endif
660 		ether_addr_copy(nic_data->port_id, efx->net_dev->perm_addr);
661 
662 	INIT_LIST_HEAD(&nic_data->vlan_list);
663 	mutex_init(&nic_data->vlan_lock);
664 
665 	/* Add unspecified VID to support VLAN filtering being disabled */
666 	rc = efx_ef10_add_vlan(efx, EFX_FILTER_VID_UNSPEC);
667 	if (rc)
668 		goto fail_add_vid_unspec;
669 
670 	/* If VLAN filtering is enabled, we need VID 0 to get untagged
671 	 * traffic.  It is added automatically if 8021q module is loaded,
672 	 * but we can't rely on it since module may be not loaded.
673 	 */
674 	rc = efx_ef10_add_vlan(efx, 0);
675 	if (rc)
676 		goto fail_add_vid_0;
677 
678 	if (nic_data->datapath_caps &
679 	    (1 << MC_CMD_GET_CAPABILITIES_OUT_VXLAN_NVGRE_LBN) &&
680 	    efx->mcdi->fn_flags &
681 	    (1 << MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_TRUSTED))
682 		efx->net_dev->udp_tunnel_nic_info = &efx_ef10_udp_tunnels;
683 
684 	return 0;
685 
686 fail_add_vid_0:
687 	efx_ef10_cleanup_vlans(efx);
688 fail_add_vid_unspec:
689 	mutex_destroy(&nic_data->vlan_lock);
690 	efx_ptp_remove(efx);
691 	efx_mcdi_mon_remove(efx);
692 fail5:
693 	device_remove_file(&efx->pci_dev->dev, &dev_attr_primary_flag);
694 fail4:
695 	device_remove_file(&efx->pci_dev->dev, &dev_attr_link_control_flag);
696 fail3:
697 	efx_mcdi_detach(efx);
698 
699 	mutex_lock(&nic_data->udp_tunnels_lock);
700 	memset(nic_data->udp_tunnels, 0, sizeof(nic_data->udp_tunnels));
701 	(void)efx_ef10_set_udp_tnl_ports(efx, true);
702 	mutex_unlock(&nic_data->udp_tunnels_lock);
703 	mutex_destroy(&nic_data->udp_tunnels_lock);
704 
705 	efx_mcdi_fini(efx);
706 fail2:
707 	efx_nic_free_buffer(efx, &nic_data->mcdi_buf);
708 fail1:
709 	kfree(nic_data);
710 	efx->nic_data = NULL;
711 	return rc;
712 }
713 
714 #ifdef EFX_USE_PIO
715 
716 static void efx_ef10_free_piobufs(struct efx_nic *efx)
717 {
718 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
719 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FREE_PIOBUF_IN_LEN);
720 	unsigned int i;
721 	int rc;
722 
723 	BUILD_BUG_ON(MC_CMD_FREE_PIOBUF_OUT_LEN != 0);
724 
725 	for (i = 0; i < nic_data->n_piobufs; i++) {
726 		MCDI_SET_DWORD(inbuf, FREE_PIOBUF_IN_PIOBUF_HANDLE,
727 			       nic_data->piobuf_handle[i]);
728 		rc = efx_mcdi_rpc(efx, MC_CMD_FREE_PIOBUF, inbuf, sizeof(inbuf),
729 				  NULL, 0, NULL);
730 		WARN_ON(rc);
731 	}
732 
733 	nic_data->n_piobufs = 0;
734 }
735 
736 static int efx_ef10_alloc_piobufs(struct efx_nic *efx, unsigned int n)
737 {
738 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
739 	MCDI_DECLARE_BUF(outbuf, MC_CMD_ALLOC_PIOBUF_OUT_LEN);
740 	unsigned int i;
741 	size_t outlen;
742 	int rc = 0;
743 
744 	BUILD_BUG_ON(MC_CMD_ALLOC_PIOBUF_IN_LEN != 0);
745 
746 	for (i = 0; i < n; i++) {
747 		rc = efx_mcdi_rpc_quiet(efx, MC_CMD_ALLOC_PIOBUF, NULL, 0,
748 					outbuf, sizeof(outbuf), &outlen);
749 		if (rc) {
750 			/* Don't display the MC error if we didn't have space
751 			 * for a VF.
752 			 */
753 			if (!(efx_ef10_is_vf(efx) && rc == -ENOSPC))
754 				efx_mcdi_display_error(efx, MC_CMD_ALLOC_PIOBUF,
755 						       0, outbuf, outlen, rc);
756 			break;
757 		}
758 		if (outlen < MC_CMD_ALLOC_PIOBUF_OUT_LEN) {
759 			rc = -EIO;
760 			break;
761 		}
762 		nic_data->piobuf_handle[i] =
763 			MCDI_DWORD(outbuf, ALLOC_PIOBUF_OUT_PIOBUF_HANDLE);
764 		netif_dbg(efx, probe, efx->net_dev,
765 			  "allocated PIO buffer %u handle %x\n", i,
766 			  nic_data->piobuf_handle[i]);
767 	}
768 
769 	nic_data->n_piobufs = i;
770 	if (rc)
771 		efx_ef10_free_piobufs(efx);
772 	return rc;
773 }
774 
775 static int efx_ef10_link_piobufs(struct efx_nic *efx)
776 {
777 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
778 	MCDI_DECLARE_BUF(inbuf, MC_CMD_LINK_PIOBUF_IN_LEN);
779 	struct efx_channel *channel;
780 	struct efx_tx_queue *tx_queue;
781 	unsigned int offset, index;
782 	int rc;
783 
784 	BUILD_BUG_ON(MC_CMD_LINK_PIOBUF_OUT_LEN != 0);
785 	BUILD_BUG_ON(MC_CMD_UNLINK_PIOBUF_OUT_LEN != 0);
786 
787 	/* Link a buffer to each VI in the write-combining mapping */
788 	for (index = 0; index < nic_data->n_piobufs; ++index) {
789 		MCDI_SET_DWORD(inbuf, LINK_PIOBUF_IN_PIOBUF_HANDLE,
790 			       nic_data->piobuf_handle[index]);
791 		MCDI_SET_DWORD(inbuf, LINK_PIOBUF_IN_TXQ_INSTANCE,
792 			       nic_data->pio_write_vi_base + index);
793 		rc = efx_mcdi_rpc(efx, MC_CMD_LINK_PIOBUF,
794 				  inbuf, MC_CMD_LINK_PIOBUF_IN_LEN,
795 				  NULL, 0, NULL);
796 		if (rc) {
797 			netif_err(efx, drv, efx->net_dev,
798 				  "failed to link VI %u to PIO buffer %u (%d)\n",
799 				  nic_data->pio_write_vi_base + index, index,
800 				  rc);
801 			goto fail;
802 		}
803 		netif_dbg(efx, probe, efx->net_dev,
804 			  "linked VI %u to PIO buffer %u\n",
805 			  nic_data->pio_write_vi_base + index, index);
806 	}
807 
808 	/* Link a buffer to each TX queue */
809 	efx_for_each_channel(channel, efx) {
810 		/* Extra channels, even those with TXQs (PTP), do not require
811 		 * PIO resources.
812 		 */
813 		if (!channel->type->want_pio ||
814 		    channel->channel >= efx->xdp_channel_offset)
815 			continue;
816 
817 		efx_for_each_channel_tx_queue(tx_queue, channel) {
818 			/* We assign the PIO buffers to queues in
819 			 * reverse order to allow for the following
820 			 * special case.
821 			 */
822 			offset = ((efx->tx_channel_offset + efx->n_tx_channels -
823 				   tx_queue->channel->channel - 1) *
824 				  efx_piobuf_size);
825 			index = offset / nic_data->piobuf_size;
826 			offset = offset % nic_data->piobuf_size;
827 
828 			/* When the host page size is 4K, the first
829 			 * host page in the WC mapping may be within
830 			 * the same VI page as the last TX queue.  We
831 			 * can only link one buffer to each VI.
832 			 */
833 			if (tx_queue->queue == nic_data->pio_write_vi_base) {
834 				BUG_ON(index != 0);
835 				rc = 0;
836 			} else {
837 				MCDI_SET_DWORD(inbuf,
838 					       LINK_PIOBUF_IN_PIOBUF_HANDLE,
839 					       nic_data->piobuf_handle[index]);
840 				MCDI_SET_DWORD(inbuf,
841 					       LINK_PIOBUF_IN_TXQ_INSTANCE,
842 					       tx_queue->queue);
843 				rc = efx_mcdi_rpc(efx, MC_CMD_LINK_PIOBUF,
844 						  inbuf, MC_CMD_LINK_PIOBUF_IN_LEN,
845 						  NULL, 0, NULL);
846 			}
847 
848 			if (rc) {
849 				/* This is non-fatal; the TX path just
850 				 * won't use PIO for this queue
851 				 */
852 				netif_err(efx, drv, efx->net_dev,
853 					  "failed to link VI %u to PIO buffer %u (%d)\n",
854 					  tx_queue->queue, index, rc);
855 				tx_queue->piobuf = NULL;
856 			} else {
857 				tx_queue->piobuf =
858 					nic_data->pio_write_base +
859 					index * efx->vi_stride + offset;
860 				tx_queue->piobuf_offset = offset;
861 				netif_dbg(efx, probe, efx->net_dev,
862 					  "linked VI %u to PIO buffer %u offset %x addr %p\n",
863 					  tx_queue->queue, index,
864 					  tx_queue->piobuf_offset,
865 					  tx_queue->piobuf);
866 			}
867 		}
868 	}
869 
870 	return 0;
871 
872 fail:
873 	/* inbuf was defined for MC_CMD_LINK_PIOBUF.  We can use the same
874 	 * buffer for MC_CMD_UNLINK_PIOBUF because it's shorter.
875 	 */
876 	BUILD_BUG_ON(MC_CMD_LINK_PIOBUF_IN_LEN < MC_CMD_UNLINK_PIOBUF_IN_LEN);
877 	while (index--) {
878 		MCDI_SET_DWORD(inbuf, UNLINK_PIOBUF_IN_TXQ_INSTANCE,
879 			       nic_data->pio_write_vi_base + index);
880 		efx_mcdi_rpc(efx, MC_CMD_UNLINK_PIOBUF,
881 			     inbuf, MC_CMD_UNLINK_PIOBUF_IN_LEN,
882 			     NULL, 0, NULL);
883 	}
884 	return rc;
885 }
886 
887 static void efx_ef10_forget_old_piobufs(struct efx_nic *efx)
888 {
889 	struct efx_channel *channel;
890 	struct efx_tx_queue *tx_queue;
891 
892 	/* All our existing PIO buffers went away */
893 	efx_for_each_channel(channel, efx)
894 		efx_for_each_channel_tx_queue(tx_queue, channel)
895 			tx_queue->piobuf = NULL;
896 }
897 
898 #else /* !EFX_USE_PIO */
899 
900 static int efx_ef10_alloc_piobufs(struct efx_nic *efx, unsigned int n)
901 {
902 	return n == 0 ? 0 : -ENOBUFS;
903 }
904 
905 static int efx_ef10_link_piobufs(struct efx_nic *efx)
906 {
907 	return 0;
908 }
909 
910 static void efx_ef10_free_piobufs(struct efx_nic *efx)
911 {
912 }
913 
914 static void efx_ef10_forget_old_piobufs(struct efx_nic *efx)
915 {
916 }
917 
918 #endif /* EFX_USE_PIO */
919 
920 static void efx_ef10_remove(struct efx_nic *efx)
921 {
922 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
923 	int rc;
924 
925 #ifdef CONFIG_SFC_SRIOV
926 	struct efx_ef10_nic_data *nic_data_pf;
927 	struct pci_dev *pci_dev_pf;
928 	struct efx_nic *efx_pf;
929 	struct ef10_vf *vf;
930 
931 	if (efx->pci_dev->is_virtfn) {
932 		pci_dev_pf = efx->pci_dev->physfn;
933 		if (pci_dev_pf) {
934 			efx_pf = pci_get_drvdata(pci_dev_pf);
935 			nic_data_pf = efx_pf->nic_data;
936 			vf = nic_data_pf->vf + nic_data->vf_index;
937 			vf->efx = NULL;
938 		} else
939 			netif_info(efx, drv, efx->net_dev,
940 				   "Could not get the PF id from VF\n");
941 	}
942 #endif
943 
944 	efx_ef10_cleanup_vlans(efx);
945 	mutex_destroy(&nic_data->vlan_lock);
946 
947 	efx_ptp_remove(efx);
948 
949 	efx_mcdi_mon_remove(efx);
950 
951 	efx_mcdi_rx_free_indir_table(efx);
952 
953 	if (nic_data->wc_membase)
954 		iounmap(nic_data->wc_membase);
955 
956 	rc = efx_mcdi_free_vis(efx);
957 	WARN_ON(rc != 0);
958 
959 	if (!nic_data->must_restore_piobufs)
960 		efx_ef10_free_piobufs(efx);
961 
962 	device_remove_file(&efx->pci_dev->dev, &dev_attr_primary_flag);
963 	device_remove_file(&efx->pci_dev->dev, &dev_attr_link_control_flag);
964 
965 	efx_mcdi_detach(efx);
966 
967 	memset(nic_data->udp_tunnels, 0, sizeof(nic_data->udp_tunnels));
968 	mutex_lock(&nic_data->udp_tunnels_lock);
969 	(void)efx_ef10_set_udp_tnl_ports(efx, true);
970 	mutex_unlock(&nic_data->udp_tunnels_lock);
971 
972 	mutex_destroy(&nic_data->udp_tunnels_lock);
973 
974 	efx_mcdi_fini(efx);
975 	efx_nic_free_buffer(efx, &nic_data->mcdi_buf);
976 	kfree(nic_data);
977 }
978 
979 static int efx_ef10_probe_pf(struct efx_nic *efx)
980 {
981 	return efx_ef10_probe(efx);
982 }
983 
984 int efx_ef10_vadaptor_query(struct efx_nic *efx, unsigned int port_id,
985 			    u32 *port_flags, u32 *vadaptor_flags,
986 			    unsigned int *vlan_tags)
987 {
988 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
989 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VADAPTOR_QUERY_IN_LEN);
990 	MCDI_DECLARE_BUF(outbuf, MC_CMD_VADAPTOR_QUERY_OUT_LEN);
991 	size_t outlen;
992 	int rc;
993 
994 	if (nic_data->datapath_caps &
995 	    (1 << MC_CMD_GET_CAPABILITIES_OUT_VADAPTOR_QUERY_LBN)) {
996 		MCDI_SET_DWORD(inbuf, VADAPTOR_QUERY_IN_UPSTREAM_PORT_ID,
997 			       port_id);
998 
999 		rc = efx_mcdi_rpc(efx, MC_CMD_VADAPTOR_QUERY, inbuf, sizeof(inbuf),
1000 				  outbuf, sizeof(outbuf), &outlen);
1001 		if (rc)
1002 			return rc;
1003 
1004 		if (outlen < sizeof(outbuf)) {
1005 			rc = -EIO;
1006 			return rc;
1007 		}
1008 	}
1009 
1010 	if (port_flags)
1011 		*port_flags = MCDI_DWORD(outbuf, VADAPTOR_QUERY_OUT_PORT_FLAGS);
1012 	if (vadaptor_flags)
1013 		*vadaptor_flags =
1014 			MCDI_DWORD(outbuf, VADAPTOR_QUERY_OUT_VADAPTOR_FLAGS);
1015 	if (vlan_tags)
1016 		*vlan_tags =
1017 			MCDI_DWORD(outbuf,
1018 				   VADAPTOR_QUERY_OUT_NUM_AVAILABLE_VLAN_TAGS);
1019 
1020 	return 0;
1021 }
1022 
1023 int efx_ef10_vadaptor_alloc(struct efx_nic *efx, unsigned int port_id)
1024 {
1025 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VADAPTOR_ALLOC_IN_LEN);
1026 
1027 	MCDI_SET_DWORD(inbuf, VADAPTOR_ALLOC_IN_UPSTREAM_PORT_ID, port_id);
1028 	return efx_mcdi_rpc(efx, MC_CMD_VADAPTOR_ALLOC, inbuf, sizeof(inbuf),
1029 			    NULL, 0, NULL);
1030 }
1031 
1032 int efx_ef10_vadaptor_free(struct efx_nic *efx, unsigned int port_id)
1033 {
1034 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VADAPTOR_FREE_IN_LEN);
1035 
1036 	MCDI_SET_DWORD(inbuf, VADAPTOR_FREE_IN_UPSTREAM_PORT_ID, port_id);
1037 	return efx_mcdi_rpc(efx, MC_CMD_VADAPTOR_FREE, inbuf, sizeof(inbuf),
1038 			    NULL, 0, NULL);
1039 }
1040 
1041 int efx_ef10_vport_add_mac(struct efx_nic *efx,
1042 			   unsigned int port_id, u8 *mac)
1043 {
1044 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VPORT_ADD_MAC_ADDRESS_IN_LEN);
1045 
1046 	MCDI_SET_DWORD(inbuf, VPORT_ADD_MAC_ADDRESS_IN_VPORT_ID, port_id);
1047 	ether_addr_copy(MCDI_PTR(inbuf, VPORT_ADD_MAC_ADDRESS_IN_MACADDR), mac);
1048 
1049 	return efx_mcdi_rpc(efx, MC_CMD_VPORT_ADD_MAC_ADDRESS, inbuf,
1050 			    sizeof(inbuf), NULL, 0, NULL);
1051 }
1052 
1053 int efx_ef10_vport_del_mac(struct efx_nic *efx,
1054 			   unsigned int port_id, u8 *mac)
1055 {
1056 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VPORT_DEL_MAC_ADDRESS_IN_LEN);
1057 
1058 	MCDI_SET_DWORD(inbuf, VPORT_DEL_MAC_ADDRESS_IN_VPORT_ID, port_id);
1059 	ether_addr_copy(MCDI_PTR(inbuf, VPORT_DEL_MAC_ADDRESS_IN_MACADDR), mac);
1060 
1061 	return efx_mcdi_rpc(efx, MC_CMD_VPORT_DEL_MAC_ADDRESS, inbuf,
1062 			    sizeof(inbuf), NULL, 0, NULL);
1063 }
1064 
1065 #ifdef CONFIG_SFC_SRIOV
1066 static int efx_ef10_probe_vf(struct efx_nic *efx)
1067 {
1068 	int rc;
1069 	struct pci_dev *pci_dev_pf;
1070 
1071 	/* If the parent PF has no VF data structure, it doesn't know about this
1072 	 * VF so fail probe.  The VF needs to be re-created.  This can happen
1073 	 * if the PF driver is unloaded while the VF is assigned to a guest.
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 	kfree(nic_data->mc_stats);
1301 	nic_data->mc_stats = NULL;
1302 }
1303 
1304 static int efx_ef10_init_nic(struct efx_nic *efx)
1305 {
1306 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1307 	netdev_features_t hw_enc_features = 0;
1308 	int rc;
1309 
1310 	if (nic_data->must_check_datapath_caps) {
1311 		rc = efx_ef10_init_datapath_caps(efx);
1312 		if (rc)
1313 			return rc;
1314 		nic_data->must_check_datapath_caps = false;
1315 	}
1316 
1317 	if (efx->must_realloc_vis) {
1318 		/* We cannot let the number of VIs change now */
1319 		rc = efx_ef10_alloc_vis(efx, nic_data->n_allocated_vis,
1320 					nic_data->n_allocated_vis);
1321 		if (rc)
1322 			return rc;
1323 		efx->must_realloc_vis = false;
1324 	}
1325 
1326 	nic_data->mc_stats = kmalloc(efx->num_mac_stats * sizeof(__le64),
1327 				     GFP_KERNEL);
1328 	if (!nic_data->mc_stats)
1329 		return -ENOMEM;
1330 
1331 	if (nic_data->must_restore_piobufs && nic_data->n_piobufs) {
1332 		rc = efx_ef10_alloc_piobufs(efx, nic_data->n_piobufs);
1333 		if (rc == 0) {
1334 			rc = efx_ef10_link_piobufs(efx);
1335 			if (rc)
1336 				efx_ef10_free_piobufs(efx);
1337 		}
1338 
1339 		/* Log an error on failure, but this is non-fatal.
1340 		 * Permission errors are less important - we've presumably
1341 		 * had the PIO buffer licence removed.
1342 		 */
1343 		if (rc == -EPERM)
1344 			netif_dbg(efx, drv, efx->net_dev,
1345 				  "not permitted to restore PIO buffers\n");
1346 		else if (rc)
1347 			netif_err(efx, drv, efx->net_dev,
1348 				  "failed to restore PIO buffers (%d)\n", rc);
1349 		nic_data->must_restore_piobufs = false;
1350 	}
1351 
1352 	/* add encapsulated checksum offload features */
1353 	if (efx_has_cap(efx, VXLAN_NVGRE) && !efx_ef10_is_vf(efx))
1354 		hw_enc_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
1355 	/* add encapsulated TSO features */
1356 	if (efx_has_cap(efx, TX_TSO_V2_ENCAP)) {
1357 		netdev_features_t encap_tso_features;
1358 
1359 		encap_tso_features = NETIF_F_GSO_UDP_TUNNEL | NETIF_F_GSO_GRE |
1360 			NETIF_F_GSO_UDP_TUNNEL_CSUM | NETIF_F_GSO_GRE_CSUM;
1361 
1362 		hw_enc_features |= encap_tso_features | NETIF_F_TSO;
1363 		efx->net_dev->features |= encap_tso_features;
1364 	}
1365 	efx->net_dev->hw_enc_features = hw_enc_features;
1366 
1367 	/* don't fail init if RSS setup doesn't work */
1368 	rc = efx->type->rx_push_rss_config(efx, false,
1369 					   efx->rss_context.rx_indir_table, NULL);
1370 
1371 	return 0;
1372 }
1373 
1374 static void efx_ef10_table_reset_mc_allocations(struct efx_nic *efx)
1375 {
1376 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1377 #ifdef CONFIG_SFC_SRIOV
1378 	unsigned int i;
1379 #endif
1380 
1381 	/* All our allocations have been reset */
1382 	efx->must_realloc_vis = true;
1383 	efx_mcdi_filter_table_reset_mc_allocations(efx);
1384 	nic_data->must_restore_piobufs = true;
1385 	efx_ef10_forget_old_piobufs(efx);
1386 	efx->rss_context.context_id = EFX_MCDI_RSS_CONTEXT_INVALID;
1387 
1388 	/* Driver-created vswitches and vports must be re-created */
1389 	nic_data->must_probe_vswitching = true;
1390 	efx->vport_id = EVB_PORT_ID_ASSIGNED;
1391 #ifdef CONFIG_SFC_SRIOV
1392 	if (nic_data->vf)
1393 		for (i = 0; i < efx->vf_count; i++)
1394 			nic_data->vf[i].vport_id = 0;
1395 #endif
1396 }
1397 
1398 static enum reset_type efx_ef10_map_reset_reason(enum reset_type reason)
1399 {
1400 	if (reason == RESET_TYPE_MC_FAILURE)
1401 		return RESET_TYPE_DATAPATH;
1402 
1403 	return efx_mcdi_map_reset_reason(reason);
1404 }
1405 
1406 static int efx_ef10_map_reset_flags(u32 *flags)
1407 {
1408 	enum {
1409 		EF10_RESET_PORT = ((ETH_RESET_MAC | ETH_RESET_PHY) <<
1410 				   ETH_RESET_SHARED_SHIFT),
1411 		EF10_RESET_MC = ((ETH_RESET_DMA | ETH_RESET_FILTER |
1412 				  ETH_RESET_OFFLOAD | ETH_RESET_MAC |
1413 				  ETH_RESET_PHY | ETH_RESET_MGMT) <<
1414 				 ETH_RESET_SHARED_SHIFT)
1415 	};
1416 
1417 	/* We assume for now that our PCI function is permitted to
1418 	 * reset everything.
1419 	 */
1420 
1421 	if ((*flags & EF10_RESET_MC) == EF10_RESET_MC) {
1422 		*flags &= ~EF10_RESET_MC;
1423 		return RESET_TYPE_WORLD;
1424 	}
1425 
1426 	if ((*flags & EF10_RESET_PORT) == EF10_RESET_PORT) {
1427 		*flags &= ~EF10_RESET_PORT;
1428 		return RESET_TYPE_ALL;
1429 	}
1430 
1431 	/* no invisible reset implemented */
1432 
1433 	return -EINVAL;
1434 }
1435 
1436 static int efx_ef10_reset(struct efx_nic *efx, enum reset_type reset_type)
1437 {
1438 	int rc = efx_mcdi_reset(efx, reset_type);
1439 
1440 	/* Unprivileged functions return -EPERM, but need to return success
1441 	 * here so that the datapath is brought back up.
1442 	 */
1443 	if (reset_type == RESET_TYPE_WORLD && rc == -EPERM)
1444 		rc = 0;
1445 
1446 	/* If it was a port reset, trigger reallocation of MC resources.
1447 	 * Note that on an MC reset nothing needs to be done now because we'll
1448 	 * detect the MC reset later and handle it then.
1449 	 * For an FLR, we never get an MC reset event, but the MC has reset all
1450 	 * resources assigned to us, so we have to trigger reallocation now.
1451 	 */
1452 	if ((reset_type == RESET_TYPE_ALL ||
1453 	     reset_type == RESET_TYPE_MCDI_TIMEOUT) && !rc)
1454 		efx_ef10_table_reset_mc_allocations(efx);
1455 	return rc;
1456 }
1457 
1458 #define EF10_DMA_STAT(ext_name, mcdi_name)			\
1459 	[EF10_STAT_ ## ext_name] =				\
1460 	{ #ext_name, 64, 8 * MC_CMD_MAC_ ## mcdi_name }
1461 #define EF10_DMA_INVIS_STAT(int_name, mcdi_name)		\
1462 	[EF10_STAT_ ## int_name] =				\
1463 	{ NULL, 64, 8 * MC_CMD_MAC_ ## mcdi_name }
1464 #define EF10_OTHER_STAT(ext_name)				\
1465 	[EF10_STAT_ ## ext_name] = { #ext_name, 0, 0 }
1466 
1467 static const struct efx_hw_stat_desc efx_ef10_stat_desc[EF10_STAT_COUNT] = {
1468 	EF10_DMA_STAT(port_tx_bytes, TX_BYTES),
1469 	EF10_DMA_STAT(port_tx_packets, TX_PKTS),
1470 	EF10_DMA_STAT(port_tx_pause, TX_PAUSE_PKTS),
1471 	EF10_DMA_STAT(port_tx_control, TX_CONTROL_PKTS),
1472 	EF10_DMA_STAT(port_tx_unicast, TX_UNICAST_PKTS),
1473 	EF10_DMA_STAT(port_tx_multicast, TX_MULTICAST_PKTS),
1474 	EF10_DMA_STAT(port_tx_broadcast, TX_BROADCAST_PKTS),
1475 	EF10_DMA_STAT(port_tx_lt64, TX_LT64_PKTS),
1476 	EF10_DMA_STAT(port_tx_64, TX_64_PKTS),
1477 	EF10_DMA_STAT(port_tx_65_to_127, TX_65_TO_127_PKTS),
1478 	EF10_DMA_STAT(port_tx_128_to_255, TX_128_TO_255_PKTS),
1479 	EF10_DMA_STAT(port_tx_256_to_511, TX_256_TO_511_PKTS),
1480 	EF10_DMA_STAT(port_tx_512_to_1023, TX_512_TO_1023_PKTS),
1481 	EF10_DMA_STAT(port_tx_1024_to_15xx, TX_1024_TO_15XX_PKTS),
1482 	EF10_DMA_STAT(port_tx_15xx_to_jumbo, TX_15XX_TO_JUMBO_PKTS),
1483 	EF10_DMA_STAT(port_rx_bytes, RX_BYTES),
1484 	EF10_DMA_INVIS_STAT(port_rx_bytes_minus_good_bytes, RX_BAD_BYTES),
1485 	EF10_OTHER_STAT(port_rx_good_bytes),
1486 	EF10_OTHER_STAT(port_rx_bad_bytes),
1487 	EF10_DMA_STAT(port_rx_packets, RX_PKTS),
1488 	EF10_DMA_STAT(port_rx_good, RX_GOOD_PKTS),
1489 	EF10_DMA_STAT(port_rx_bad, RX_BAD_FCS_PKTS),
1490 	EF10_DMA_STAT(port_rx_pause, RX_PAUSE_PKTS),
1491 	EF10_DMA_STAT(port_rx_control, RX_CONTROL_PKTS),
1492 	EF10_DMA_STAT(port_rx_unicast, RX_UNICAST_PKTS),
1493 	EF10_DMA_STAT(port_rx_multicast, RX_MULTICAST_PKTS),
1494 	EF10_DMA_STAT(port_rx_broadcast, RX_BROADCAST_PKTS),
1495 	EF10_DMA_STAT(port_rx_lt64, RX_UNDERSIZE_PKTS),
1496 	EF10_DMA_STAT(port_rx_64, RX_64_PKTS),
1497 	EF10_DMA_STAT(port_rx_65_to_127, RX_65_TO_127_PKTS),
1498 	EF10_DMA_STAT(port_rx_128_to_255, RX_128_TO_255_PKTS),
1499 	EF10_DMA_STAT(port_rx_256_to_511, RX_256_TO_511_PKTS),
1500 	EF10_DMA_STAT(port_rx_512_to_1023, RX_512_TO_1023_PKTS),
1501 	EF10_DMA_STAT(port_rx_1024_to_15xx, RX_1024_TO_15XX_PKTS),
1502 	EF10_DMA_STAT(port_rx_15xx_to_jumbo, RX_15XX_TO_JUMBO_PKTS),
1503 	EF10_DMA_STAT(port_rx_gtjumbo, RX_GTJUMBO_PKTS),
1504 	EF10_DMA_STAT(port_rx_bad_gtjumbo, RX_JABBER_PKTS),
1505 	EF10_DMA_STAT(port_rx_overflow, RX_OVERFLOW_PKTS),
1506 	EF10_DMA_STAT(port_rx_align_error, RX_ALIGN_ERROR_PKTS),
1507 	EF10_DMA_STAT(port_rx_length_error, RX_LENGTH_ERROR_PKTS),
1508 	EF10_DMA_STAT(port_rx_nodesc_drops, RX_NODESC_DROPS),
1509 	EFX_GENERIC_SW_STAT(rx_nodesc_trunc),
1510 	EFX_GENERIC_SW_STAT(rx_noskb_drops),
1511 	EF10_DMA_STAT(port_rx_pm_trunc_bb_overflow, PM_TRUNC_BB_OVERFLOW),
1512 	EF10_DMA_STAT(port_rx_pm_discard_bb_overflow, PM_DISCARD_BB_OVERFLOW),
1513 	EF10_DMA_STAT(port_rx_pm_trunc_vfifo_full, PM_TRUNC_VFIFO_FULL),
1514 	EF10_DMA_STAT(port_rx_pm_discard_vfifo_full, PM_DISCARD_VFIFO_FULL),
1515 	EF10_DMA_STAT(port_rx_pm_trunc_qbb, PM_TRUNC_QBB),
1516 	EF10_DMA_STAT(port_rx_pm_discard_qbb, PM_DISCARD_QBB),
1517 	EF10_DMA_STAT(port_rx_pm_discard_mapping, PM_DISCARD_MAPPING),
1518 	EF10_DMA_STAT(port_rx_dp_q_disabled_packets, RXDP_Q_DISABLED_PKTS),
1519 	EF10_DMA_STAT(port_rx_dp_di_dropped_packets, RXDP_DI_DROPPED_PKTS),
1520 	EF10_DMA_STAT(port_rx_dp_streaming_packets, RXDP_STREAMING_PKTS),
1521 	EF10_DMA_STAT(port_rx_dp_hlb_fetch, RXDP_HLB_FETCH_CONDITIONS),
1522 	EF10_DMA_STAT(port_rx_dp_hlb_wait, RXDP_HLB_WAIT_CONDITIONS),
1523 	EF10_DMA_STAT(rx_unicast, VADAPTER_RX_UNICAST_PACKETS),
1524 	EF10_DMA_STAT(rx_unicast_bytes, VADAPTER_RX_UNICAST_BYTES),
1525 	EF10_DMA_STAT(rx_multicast, VADAPTER_RX_MULTICAST_PACKETS),
1526 	EF10_DMA_STAT(rx_multicast_bytes, VADAPTER_RX_MULTICAST_BYTES),
1527 	EF10_DMA_STAT(rx_broadcast, VADAPTER_RX_BROADCAST_PACKETS),
1528 	EF10_DMA_STAT(rx_broadcast_bytes, VADAPTER_RX_BROADCAST_BYTES),
1529 	EF10_DMA_STAT(rx_bad, VADAPTER_RX_BAD_PACKETS),
1530 	EF10_DMA_STAT(rx_bad_bytes, VADAPTER_RX_BAD_BYTES),
1531 	EF10_DMA_STAT(rx_overflow, VADAPTER_RX_OVERFLOW),
1532 	EF10_DMA_STAT(tx_unicast, VADAPTER_TX_UNICAST_PACKETS),
1533 	EF10_DMA_STAT(tx_unicast_bytes, VADAPTER_TX_UNICAST_BYTES),
1534 	EF10_DMA_STAT(tx_multicast, VADAPTER_TX_MULTICAST_PACKETS),
1535 	EF10_DMA_STAT(tx_multicast_bytes, VADAPTER_TX_MULTICAST_BYTES),
1536 	EF10_DMA_STAT(tx_broadcast, VADAPTER_TX_BROADCAST_PACKETS),
1537 	EF10_DMA_STAT(tx_broadcast_bytes, VADAPTER_TX_BROADCAST_BYTES),
1538 	EF10_DMA_STAT(tx_bad, VADAPTER_TX_BAD_PACKETS),
1539 	EF10_DMA_STAT(tx_bad_bytes, VADAPTER_TX_BAD_BYTES),
1540 	EF10_DMA_STAT(tx_overflow, VADAPTER_TX_OVERFLOW),
1541 	EF10_DMA_STAT(fec_uncorrected_errors, FEC_UNCORRECTED_ERRORS),
1542 	EF10_DMA_STAT(fec_corrected_errors, FEC_CORRECTED_ERRORS),
1543 	EF10_DMA_STAT(fec_corrected_symbols_lane0, FEC_CORRECTED_SYMBOLS_LANE0),
1544 	EF10_DMA_STAT(fec_corrected_symbols_lane1, FEC_CORRECTED_SYMBOLS_LANE1),
1545 	EF10_DMA_STAT(fec_corrected_symbols_lane2, FEC_CORRECTED_SYMBOLS_LANE2),
1546 	EF10_DMA_STAT(fec_corrected_symbols_lane3, FEC_CORRECTED_SYMBOLS_LANE3),
1547 	EF10_DMA_STAT(ctpio_vi_busy_fallback, CTPIO_VI_BUSY_FALLBACK),
1548 	EF10_DMA_STAT(ctpio_long_write_success, CTPIO_LONG_WRITE_SUCCESS),
1549 	EF10_DMA_STAT(ctpio_missing_dbell_fail, CTPIO_MISSING_DBELL_FAIL),
1550 	EF10_DMA_STAT(ctpio_overflow_fail, CTPIO_OVERFLOW_FAIL),
1551 	EF10_DMA_STAT(ctpio_underflow_fail, CTPIO_UNDERFLOW_FAIL),
1552 	EF10_DMA_STAT(ctpio_timeout_fail, CTPIO_TIMEOUT_FAIL),
1553 	EF10_DMA_STAT(ctpio_noncontig_wr_fail, CTPIO_NONCONTIG_WR_FAIL),
1554 	EF10_DMA_STAT(ctpio_frm_clobber_fail, CTPIO_FRM_CLOBBER_FAIL),
1555 	EF10_DMA_STAT(ctpio_invalid_wr_fail, CTPIO_INVALID_WR_FAIL),
1556 	EF10_DMA_STAT(ctpio_vi_clobber_fallback, CTPIO_VI_CLOBBER_FALLBACK),
1557 	EF10_DMA_STAT(ctpio_unqualified_fallback, CTPIO_UNQUALIFIED_FALLBACK),
1558 	EF10_DMA_STAT(ctpio_runt_fallback, CTPIO_RUNT_FALLBACK),
1559 	EF10_DMA_STAT(ctpio_success, CTPIO_SUCCESS),
1560 	EF10_DMA_STAT(ctpio_fallback, CTPIO_FALLBACK),
1561 	EF10_DMA_STAT(ctpio_poison, CTPIO_POISON),
1562 	EF10_DMA_STAT(ctpio_erase, CTPIO_ERASE),
1563 };
1564 
1565 #define HUNT_COMMON_STAT_MASK ((1ULL << EF10_STAT_port_tx_bytes) |	\
1566 			       (1ULL << EF10_STAT_port_tx_packets) |	\
1567 			       (1ULL << EF10_STAT_port_tx_pause) |	\
1568 			       (1ULL << EF10_STAT_port_tx_unicast) |	\
1569 			       (1ULL << EF10_STAT_port_tx_multicast) |	\
1570 			       (1ULL << EF10_STAT_port_tx_broadcast) |	\
1571 			       (1ULL << EF10_STAT_port_rx_bytes) |	\
1572 			       (1ULL <<                                 \
1573 				EF10_STAT_port_rx_bytes_minus_good_bytes) | \
1574 			       (1ULL << EF10_STAT_port_rx_good_bytes) |	\
1575 			       (1ULL << EF10_STAT_port_rx_bad_bytes) |	\
1576 			       (1ULL << EF10_STAT_port_rx_packets) |	\
1577 			       (1ULL << EF10_STAT_port_rx_good) |	\
1578 			       (1ULL << EF10_STAT_port_rx_bad) |	\
1579 			       (1ULL << EF10_STAT_port_rx_pause) |	\
1580 			       (1ULL << EF10_STAT_port_rx_control) |	\
1581 			       (1ULL << EF10_STAT_port_rx_unicast) |	\
1582 			       (1ULL << EF10_STAT_port_rx_multicast) |	\
1583 			       (1ULL << EF10_STAT_port_rx_broadcast) |	\
1584 			       (1ULL << EF10_STAT_port_rx_lt64) |	\
1585 			       (1ULL << EF10_STAT_port_rx_64) |		\
1586 			       (1ULL << EF10_STAT_port_rx_65_to_127) |	\
1587 			       (1ULL << EF10_STAT_port_rx_128_to_255) |	\
1588 			       (1ULL << EF10_STAT_port_rx_256_to_511) |	\
1589 			       (1ULL << EF10_STAT_port_rx_512_to_1023) |\
1590 			       (1ULL << EF10_STAT_port_rx_1024_to_15xx) |\
1591 			       (1ULL << EF10_STAT_port_rx_15xx_to_jumbo) |\
1592 			       (1ULL << EF10_STAT_port_rx_gtjumbo) |	\
1593 			       (1ULL << EF10_STAT_port_rx_bad_gtjumbo) |\
1594 			       (1ULL << EF10_STAT_port_rx_overflow) |	\
1595 			       (1ULL << EF10_STAT_port_rx_nodesc_drops) |\
1596 			       (1ULL << GENERIC_STAT_rx_nodesc_trunc) |	\
1597 			       (1ULL << GENERIC_STAT_rx_noskb_drops))
1598 
1599 /* On 7000 series NICs, these statistics are only provided by the 10G MAC.
1600  * For a 10G/40G switchable port we do not expose these because they might
1601  * not include all the packets they should.
1602  * On 8000 series NICs these statistics are always provided.
1603  */
1604 #define HUNT_10G_ONLY_STAT_MASK ((1ULL << EF10_STAT_port_tx_control) |	\
1605 				 (1ULL << EF10_STAT_port_tx_lt64) |	\
1606 				 (1ULL << EF10_STAT_port_tx_64) |	\
1607 				 (1ULL << EF10_STAT_port_tx_65_to_127) |\
1608 				 (1ULL << EF10_STAT_port_tx_128_to_255) |\
1609 				 (1ULL << EF10_STAT_port_tx_256_to_511) |\
1610 				 (1ULL << EF10_STAT_port_tx_512_to_1023) |\
1611 				 (1ULL << EF10_STAT_port_tx_1024_to_15xx) |\
1612 				 (1ULL << EF10_STAT_port_tx_15xx_to_jumbo))
1613 
1614 /* These statistics are only provided by the 40G MAC.  For a 10G/40G
1615  * switchable port we do expose these because the errors will otherwise
1616  * be silent.
1617  */
1618 #define HUNT_40G_EXTRA_STAT_MASK ((1ULL << EF10_STAT_port_rx_align_error) |\
1619 				  (1ULL << EF10_STAT_port_rx_length_error))
1620 
1621 /* These statistics are only provided if the firmware supports the
1622  * capability PM_AND_RXDP_COUNTERS.
1623  */
1624 #define HUNT_PM_AND_RXDP_STAT_MASK (					\
1625 	(1ULL << EF10_STAT_port_rx_pm_trunc_bb_overflow) |		\
1626 	(1ULL << EF10_STAT_port_rx_pm_discard_bb_overflow) |		\
1627 	(1ULL << EF10_STAT_port_rx_pm_trunc_vfifo_full) |		\
1628 	(1ULL << EF10_STAT_port_rx_pm_discard_vfifo_full) |		\
1629 	(1ULL << EF10_STAT_port_rx_pm_trunc_qbb) |			\
1630 	(1ULL << EF10_STAT_port_rx_pm_discard_qbb) |			\
1631 	(1ULL << EF10_STAT_port_rx_pm_discard_mapping) |		\
1632 	(1ULL << EF10_STAT_port_rx_dp_q_disabled_packets) |		\
1633 	(1ULL << EF10_STAT_port_rx_dp_di_dropped_packets) |		\
1634 	(1ULL << EF10_STAT_port_rx_dp_streaming_packets) |		\
1635 	(1ULL << EF10_STAT_port_rx_dp_hlb_fetch) |			\
1636 	(1ULL << EF10_STAT_port_rx_dp_hlb_wait))
1637 
1638 /* These statistics are only provided if the NIC supports MC_CMD_MAC_STATS_V2,
1639  * indicated by returning a value >= MC_CMD_MAC_NSTATS_V2 in
1640  * MC_CMD_GET_CAPABILITIES_V4_OUT_MAC_STATS_NUM_STATS.
1641  * These bits are in the second u64 of the raw mask.
1642  */
1643 #define EF10_FEC_STAT_MASK (						\
1644 	(1ULL << (EF10_STAT_fec_uncorrected_errors - 64)) |		\
1645 	(1ULL << (EF10_STAT_fec_corrected_errors - 64)) |		\
1646 	(1ULL << (EF10_STAT_fec_corrected_symbols_lane0 - 64)) |	\
1647 	(1ULL << (EF10_STAT_fec_corrected_symbols_lane1 - 64)) |	\
1648 	(1ULL << (EF10_STAT_fec_corrected_symbols_lane2 - 64)) |	\
1649 	(1ULL << (EF10_STAT_fec_corrected_symbols_lane3 - 64)))
1650 
1651 /* These statistics are only provided if the NIC supports MC_CMD_MAC_STATS_V3,
1652  * indicated by returning a value >= MC_CMD_MAC_NSTATS_V3 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_CTPIO_STAT_MASK (						\
1657 	(1ULL << (EF10_STAT_ctpio_vi_busy_fallback - 64)) |		\
1658 	(1ULL << (EF10_STAT_ctpio_long_write_success - 64)) |		\
1659 	(1ULL << (EF10_STAT_ctpio_missing_dbell_fail - 64)) |		\
1660 	(1ULL << (EF10_STAT_ctpio_overflow_fail - 64)) |		\
1661 	(1ULL << (EF10_STAT_ctpio_underflow_fail - 64)) |		\
1662 	(1ULL << (EF10_STAT_ctpio_timeout_fail - 64)) |			\
1663 	(1ULL << (EF10_STAT_ctpio_noncontig_wr_fail - 64)) |		\
1664 	(1ULL << (EF10_STAT_ctpio_frm_clobber_fail - 64)) |		\
1665 	(1ULL << (EF10_STAT_ctpio_invalid_wr_fail - 64)) |		\
1666 	(1ULL << (EF10_STAT_ctpio_vi_clobber_fallback - 64)) |		\
1667 	(1ULL << (EF10_STAT_ctpio_unqualified_fallback - 64)) |		\
1668 	(1ULL << (EF10_STAT_ctpio_runt_fallback - 64)) |		\
1669 	(1ULL << (EF10_STAT_ctpio_success - 64)) |			\
1670 	(1ULL << (EF10_STAT_ctpio_fallback - 64)) |			\
1671 	(1ULL << (EF10_STAT_ctpio_poison - 64)) |			\
1672 	(1ULL << (EF10_STAT_ctpio_erase - 64)))
1673 
1674 static u64 efx_ef10_raw_stat_mask(struct efx_nic *efx)
1675 {
1676 	u64 raw_mask = HUNT_COMMON_STAT_MASK;
1677 	u32 port_caps = efx_mcdi_phy_get_caps(efx);
1678 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1679 
1680 	if (!(efx->mcdi->fn_flags &
1681 	      1 << MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_LINKCTRL))
1682 		return 0;
1683 
1684 	if (port_caps & (1 << MC_CMD_PHY_CAP_40000FDX_LBN)) {
1685 		raw_mask |= HUNT_40G_EXTRA_STAT_MASK;
1686 		/* 8000 series have everything even at 40G */
1687 		if (nic_data->datapath_caps2 &
1688 		    (1 << MC_CMD_GET_CAPABILITIES_V2_OUT_MAC_STATS_40G_TX_SIZE_BINS_LBN))
1689 			raw_mask |= HUNT_10G_ONLY_STAT_MASK;
1690 	} else {
1691 		raw_mask |= HUNT_10G_ONLY_STAT_MASK;
1692 	}
1693 
1694 	if (nic_data->datapath_caps &
1695 	    (1 << MC_CMD_GET_CAPABILITIES_OUT_PM_AND_RXDP_COUNTERS_LBN))
1696 		raw_mask |= HUNT_PM_AND_RXDP_STAT_MASK;
1697 
1698 	return raw_mask;
1699 }
1700 
1701 static void efx_ef10_get_stat_mask(struct efx_nic *efx, unsigned long *mask)
1702 {
1703 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1704 	u64 raw_mask[2];
1705 
1706 	raw_mask[0] = efx_ef10_raw_stat_mask(efx);
1707 
1708 	/* Only show vadaptor stats when EVB capability is present */
1709 	if (nic_data->datapath_caps &
1710 	    (1 << MC_CMD_GET_CAPABILITIES_OUT_EVB_LBN)) {
1711 		raw_mask[0] |= ~((1ULL << EF10_STAT_rx_unicast) - 1);
1712 		raw_mask[1] = (1ULL << (EF10_STAT_V1_COUNT - 64)) - 1;
1713 	} else {
1714 		raw_mask[1] = 0;
1715 	}
1716 	/* Only show FEC stats when NIC supports MC_CMD_MAC_STATS_V2 */
1717 	if (efx->num_mac_stats >= MC_CMD_MAC_NSTATS_V2)
1718 		raw_mask[1] |= EF10_FEC_STAT_MASK;
1719 
1720 	/* CTPIO stats appear in V3. Only show them on devices that actually
1721 	 * support CTPIO. Although this driver doesn't use CTPIO others might,
1722 	 * and we may be reporting the stats for the underlying port.
1723 	 */
1724 	if (efx->num_mac_stats >= MC_CMD_MAC_NSTATS_V3 &&
1725 	    (nic_data->datapath_caps2 &
1726 	     (1 << MC_CMD_GET_CAPABILITIES_V4_OUT_CTPIO_LBN)))
1727 		raw_mask[1] |= EF10_CTPIO_STAT_MASK;
1728 
1729 #if BITS_PER_LONG == 64
1730 	BUILD_BUG_ON(BITS_TO_LONGS(EF10_STAT_COUNT) != 2);
1731 	mask[0] = raw_mask[0];
1732 	mask[1] = raw_mask[1];
1733 #else
1734 	BUILD_BUG_ON(BITS_TO_LONGS(EF10_STAT_COUNT) != 3);
1735 	mask[0] = raw_mask[0] & 0xffffffff;
1736 	mask[1] = raw_mask[0] >> 32;
1737 	mask[2] = raw_mask[1] & 0xffffffff;
1738 #endif
1739 }
1740 
1741 static size_t efx_ef10_describe_stats(struct efx_nic *efx, u8 *names)
1742 {
1743 	DECLARE_BITMAP(mask, EF10_STAT_COUNT);
1744 
1745 	efx_ef10_get_stat_mask(efx, mask);
1746 	return efx_nic_describe_stats(efx_ef10_stat_desc, EF10_STAT_COUNT,
1747 				      mask, names);
1748 }
1749 
1750 static size_t efx_ef10_update_stats_common(struct efx_nic *efx, u64 *full_stats,
1751 					   struct rtnl_link_stats64 *core_stats)
1752 {
1753 	DECLARE_BITMAP(mask, EF10_STAT_COUNT);
1754 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1755 	u64 *stats = nic_data->stats;
1756 	size_t stats_count = 0, index;
1757 
1758 	efx_ef10_get_stat_mask(efx, mask);
1759 
1760 	if (full_stats) {
1761 		for_each_set_bit(index, mask, EF10_STAT_COUNT) {
1762 			if (efx_ef10_stat_desc[index].name) {
1763 				*full_stats++ = stats[index];
1764 				++stats_count;
1765 			}
1766 		}
1767 	}
1768 
1769 	if (!core_stats)
1770 		return stats_count;
1771 
1772 	if (nic_data->datapath_caps &
1773 			1 << MC_CMD_GET_CAPABILITIES_OUT_EVB_LBN) {
1774 		/* Use vadaptor stats. */
1775 		core_stats->rx_packets = stats[EF10_STAT_rx_unicast] +
1776 					 stats[EF10_STAT_rx_multicast] +
1777 					 stats[EF10_STAT_rx_broadcast];
1778 		core_stats->tx_packets = stats[EF10_STAT_tx_unicast] +
1779 					 stats[EF10_STAT_tx_multicast] +
1780 					 stats[EF10_STAT_tx_broadcast];
1781 		core_stats->rx_bytes = stats[EF10_STAT_rx_unicast_bytes] +
1782 				       stats[EF10_STAT_rx_multicast_bytes] +
1783 				       stats[EF10_STAT_rx_broadcast_bytes];
1784 		core_stats->tx_bytes = stats[EF10_STAT_tx_unicast_bytes] +
1785 				       stats[EF10_STAT_tx_multicast_bytes] +
1786 				       stats[EF10_STAT_tx_broadcast_bytes];
1787 		core_stats->rx_dropped = stats[GENERIC_STAT_rx_nodesc_trunc] +
1788 					 stats[GENERIC_STAT_rx_noskb_drops];
1789 		core_stats->multicast = stats[EF10_STAT_rx_multicast];
1790 		core_stats->rx_crc_errors = stats[EF10_STAT_rx_bad];
1791 		core_stats->rx_fifo_errors = stats[EF10_STAT_rx_overflow];
1792 		core_stats->rx_errors = core_stats->rx_crc_errors;
1793 		core_stats->tx_errors = stats[EF10_STAT_tx_bad];
1794 	} else {
1795 		/* Use port stats. */
1796 		core_stats->rx_packets = stats[EF10_STAT_port_rx_packets];
1797 		core_stats->tx_packets = stats[EF10_STAT_port_tx_packets];
1798 		core_stats->rx_bytes = stats[EF10_STAT_port_rx_bytes];
1799 		core_stats->tx_bytes = stats[EF10_STAT_port_tx_bytes];
1800 		core_stats->rx_dropped = stats[EF10_STAT_port_rx_nodesc_drops] +
1801 					 stats[GENERIC_STAT_rx_nodesc_trunc] +
1802 					 stats[GENERIC_STAT_rx_noskb_drops];
1803 		core_stats->multicast = stats[EF10_STAT_port_rx_multicast];
1804 		core_stats->rx_length_errors =
1805 				stats[EF10_STAT_port_rx_gtjumbo] +
1806 				stats[EF10_STAT_port_rx_length_error];
1807 		core_stats->rx_crc_errors = stats[EF10_STAT_port_rx_bad];
1808 		core_stats->rx_frame_errors =
1809 				stats[EF10_STAT_port_rx_align_error];
1810 		core_stats->rx_fifo_errors = stats[EF10_STAT_port_rx_overflow];
1811 		core_stats->rx_errors = (core_stats->rx_length_errors +
1812 					 core_stats->rx_crc_errors +
1813 					 core_stats->rx_frame_errors);
1814 	}
1815 
1816 	return stats_count;
1817 }
1818 
1819 static size_t efx_ef10_update_stats_pf(struct efx_nic *efx, u64 *full_stats,
1820 				       struct rtnl_link_stats64 *core_stats)
1821 {
1822 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1823 	DECLARE_BITMAP(mask, EF10_STAT_COUNT);
1824 	u64 *stats = nic_data->stats;
1825 
1826 	efx_ef10_get_stat_mask(efx, mask);
1827 
1828 	efx_nic_copy_stats(efx, nic_data->mc_stats);
1829 	efx_nic_update_stats(efx_ef10_stat_desc, EF10_STAT_COUNT,
1830 			     mask, stats, nic_data->mc_stats, false);
1831 
1832 	/* Update derived statistics */
1833 	efx_nic_fix_nodesc_drop_stat(efx,
1834 				     &stats[EF10_STAT_port_rx_nodesc_drops]);
1835 	/* MC Firmware reads RX_BYTES and RX_GOOD_BYTES from the MAC.
1836 	 * It then calculates RX_BAD_BYTES and DMAs it to us with RX_BYTES.
1837 	 * We report these as port_rx_ stats. We are not given RX_GOOD_BYTES.
1838 	 * Here we calculate port_rx_good_bytes.
1839 	 */
1840 	stats[EF10_STAT_port_rx_good_bytes] =
1841 		stats[EF10_STAT_port_rx_bytes] -
1842 		stats[EF10_STAT_port_rx_bytes_minus_good_bytes];
1843 
1844 	/* The asynchronous reads used to calculate RX_BAD_BYTES in
1845 	 * MC Firmware are done such that we should not see an increase in
1846 	 * RX_BAD_BYTES when a good packet has arrived. Unfortunately this
1847 	 * does mean that the stat can decrease at times. Here we do not
1848 	 * update the stat unless it has increased or has gone to zero
1849 	 * (In the case of the NIC rebooting).
1850 	 * Please see Bug 33781 for a discussion of why things work this way.
1851 	 */
1852 	efx_update_diff_stat(&stats[EF10_STAT_port_rx_bad_bytes],
1853 			     stats[EF10_STAT_port_rx_bytes_minus_good_bytes]);
1854 	efx_update_sw_stats(efx, stats);
1855 
1856 	return efx_ef10_update_stats_common(efx, full_stats, core_stats);
1857 }
1858 
1859 static int efx_ef10_try_update_nic_stats_vf(struct efx_nic *efx)
1860 	__must_hold(&efx->stats_lock)
1861 {
1862 	MCDI_DECLARE_BUF(inbuf, MC_CMD_MAC_STATS_IN_LEN);
1863 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1864 	DECLARE_BITMAP(mask, EF10_STAT_COUNT);
1865 	__le64 generation_start, generation_end;
1866 	u64 *stats = nic_data->stats;
1867 	u32 dma_len = efx->num_mac_stats * sizeof(u64);
1868 	struct efx_buffer stats_buf;
1869 	__le64 *dma_stats;
1870 	int rc;
1871 
1872 	spin_unlock_bh(&efx->stats_lock);
1873 
1874 	efx_ef10_get_stat_mask(efx, mask);
1875 
1876 	rc = efx_nic_alloc_buffer(efx, &stats_buf, dma_len, GFP_KERNEL);
1877 	if (rc) {
1878 		spin_lock_bh(&efx->stats_lock);
1879 		return rc;
1880 	}
1881 
1882 	dma_stats = stats_buf.addr;
1883 	dma_stats[efx->num_mac_stats - 1] = EFX_MC_STATS_GENERATION_INVALID;
1884 
1885 	MCDI_SET_QWORD(inbuf, MAC_STATS_IN_DMA_ADDR, stats_buf.dma_addr);
1886 	MCDI_POPULATE_DWORD_1(inbuf, MAC_STATS_IN_CMD,
1887 			      MAC_STATS_IN_DMA, 1);
1888 	MCDI_SET_DWORD(inbuf, MAC_STATS_IN_DMA_LEN, dma_len);
1889 	MCDI_SET_DWORD(inbuf, MAC_STATS_IN_PORT_ID, EVB_PORT_ID_ASSIGNED);
1890 
1891 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_MAC_STATS, inbuf, sizeof(inbuf),
1892 				NULL, 0, NULL);
1893 	spin_lock_bh(&efx->stats_lock);
1894 	if (rc) {
1895 		/* Expect ENOENT if DMA queues have not been set up */
1896 		if (rc != -ENOENT || atomic_read(&efx->active_queues))
1897 			efx_mcdi_display_error(efx, MC_CMD_MAC_STATS,
1898 					       sizeof(inbuf), NULL, 0, rc);
1899 		goto out;
1900 	}
1901 
1902 	generation_end = dma_stats[efx->num_mac_stats - 1];
1903 	if (generation_end == EFX_MC_STATS_GENERATION_INVALID) {
1904 		WARN_ON_ONCE(1);
1905 		goto out;
1906 	}
1907 	rmb();
1908 	efx_nic_update_stats(efx_ef10_stat_desc, EF10_STAT_COUNT, mask,
1909 			     stats, stats_buf.addr, false);
1910 	rmb();
1911 	generation_start = dma_stats[MC_CMD_MAC_GENERATION_START];
1912 	if (generation_end != generation_start) {
1913 		rc = -EAGAIN;
1914 		goto out;
1915 	}
1916 
1917 	efx_update_sw_stats(efx, stats);
1918 out:
1919 	efx_nic_free_buffer(efx, &stats_buf);
1920 	return rc;
1921 }
1922 
1923 static size_t efx_ef10_update_stats_vf(struct efx_nic *efx, u64 *full_stats,
1924 				       struct rtnl_link_stats64 *core_stats)
1925 {
1926 	if (efx_ef10_try_update_nic_stats_vf(efx))
1927 		return 0;
1928 
1929 	return efx_ef10_update_stats_common(efx, full_stats, core_stats);
1930 }
1931 
1932 static size_t efx_ef10_update_stats_atomic_vf(struct efx_nic *efx, u64 *full_stats,
1933 					      struct rtnl_link_stats64 *core_stats)
1934 {
1935 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1936 
1937 	/* In atomic context, cannot update HW stats.  Just update the
1938 	 * software stats and return so the caller can continue.
1939 	 */
1940 	efx_update_sw_stats(efx, nic_data->stats);
1941 	return efx_ef10_update_stats_common(efx, full_stats, core_stats);
1942 }
1943 
1944 static void efx_ef10_push_irq_moderation(struct efx_channel *channel)
1945 {
1946 	struct efx_nic *efx = channel->efx;
1947 	unsigned int mode, usecs;
1948 	efx_dword_t timer_cmd;
1949 
1950 	if (channel->irq_moderation_us) {
1951 		mode = 3;
1952 		usecs = channel->irq_moderation_us;
1953 	} else {
1954 		mode = 0;
1955 		usecs = 0;
1956 	}
1957 
1958 	if (EFX_EF10_WORKAROUND_61265(efx)) {
1959 		MCDI_DECLARE_BUF(inbuf, MC_CMD_SET_EVQ_TMR_IN_LEN);
1960 		unsigned int ns = usecs * 1000;
1961 
1962 		MCDI_SET_DWORD(inbuf, SET_EVQ_TMR_IN_INSTANCE,
1963 			       channel->channel);
1964 		MCDI_SET_DWORD(inbuf, SET_EVQ_TMR_IN_TMR_LOAD_REQ_NS, ns);
1965 		MCDI_SET_DWORD(inbuf, SET_EVQ_TMR_IN_TMR_RELOAD_REQ_NS, ns);
1966 		MCDI_SET_DWORD(inbuf, SET_EVQ_TMR_IN_TMR_MODE, mode);
1967 
1968 		efx_mcdi_rpc_async(efx, MC_CMD_SET_EVQ_TMR,
1969 				   inbuf, sizeof(inbuf), 0, NULL, 0);
1970 	} else if (EFX_EF10_WORKAROUND_35388(efx)) {
1971 		unsigned int ticks = efx_usecs_to_ticks(efx, usecs);
1972 
1973 		EFX_POPULATE_DWORD_3(timer_cmd, ERF_DD_EVQ_IND_TIMER_FLAGS,
1974 				     EFE_DD_EVQ_IND_TIMER_FLAGS,
1975 				     ERF_DD_EVQ_IND_TIMER_MODE, mode,
1976 				     ERF_DD_EVQ_IND_TIMER_VAL, ticks);
1977 		efx_writed_page(efx, &timer_cmd, ER_DD_EVQ_INDIRECT,
1978 				channel->channel);
1979 	} else {
1980 		unsigned int ticks = efx_usecs_to_ticks(efx, usecs);
1981 
1982 		EFX_POPULATE_DWORD_3(timer_cmd, ERF_DZ_TC_TIMER_MODE, mode,
1983 				     ERF_DZ_TC_TIMER_VAL, ticks,
1984 				     ERF_FZ_TC_TMR_REL_VAL, ticks);
1985 		efx_writed_page(efx, &timer_cmd, ER_DZ_EVQ_TMR,
1986 				channel->channel);
1987 	}
1988 }
1989 
1990 static void efx_ef10_get_wol_vf(struct efx_nic *efx,
1991 				struct ethtool_wolinfo *wol) {}
1992 
1993 static int efx_ef10_set_wol_vf(struct efx_nic *efx, u32 type)
1994 {
1995 	return -EOPNOTSUPP;
1996 }
1997 
1998 static void efx_ef10_get_wol(struct efx_nic *efx, struct ethtool_wolinfo *wol)
1999 {
2000 	wol->supported = 0;
2001 	wol->wolopts = 0;
2002 	memset(&wol->sopass, 0, sizeof(wol->sopass));
2003 }
2004 
2005 static int efx_ef10_set_wol(struct efx_nic *efx, u32 type)
2006 {
2007 	if (type != 0)
2008 		return -EINVAL;
2009 	return 0;
2010 }
2011 
2012 static void efx_ef10_mcdi_request(struct efx_nic *efx,
2013 				  const efx_dword_t *hdr, size_t hdr_len,
2014 				  const efx_dword_t *sdu, size_t sdu_len)
2015 {
2016 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2017 	u8 *pdu = nic_data->mcdi_buf.addr;
2018 
2019 	memcpy(pdu, hdr, hdr_len);
2020 	memcpy(pdu + hdr_len, sdu, sdu_len);
2021 	wmb();
2022 
2023 	/* The hardware provides 'low' and 'high' (doorbell) registers
2024 	 * for passing the 64-bit address of an MCDI request to
2025 	 * firmware.  However the dwords are swapped by firmware.  The
2026 	 * least significant bits of the doorbell are then 0 for all
2027 	 * MCDI requests due to alignment.
2028 	 */
2029 	_efx_writed(efx, cpu_to_le32((u64)nic_data->mcdi_buf.dma_addr >> 32),
2030 		    ER_DZ_MC_DB_LWRD);
2031 	_efx_writed(efx, cpu_to_le32((u32)nic_data->mcdi_buf.dma_addr),
2032 		    ER_DZ_MC_DB_HWRD);
2033 }
2034 
2035 static bool efx_ef10_mcdi_poll_response(struct efx_nic *efx)
2036 {
2037 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2038 	const efx_dword_t hdr = *(const efx_dword_t *)nic_data->mcdi_buf.addr;
2039 
2040 	rmb();
2041 	return EFX_DWORD_FIELD(hdr, MCDI_HEADER_RESPONSE);
2042 }
2043 
2044 static void
2045 efx_ef10_mcdi_read_response(struct efx_nic *efx, efx_dword_t *outbuf,
2046 			    size_t offset, size_t outlen)
2047 {
2048 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2049 	const u8 *pdu = nic_data->mcdi_buf.addr;
2050 
2051 	memcpy(outbuf, pdu + offset, outlen);
2052 }
2053 
2054 static void efx_ef10_mcdi_reboot_detected(struct efx_nic *efx)
2055 {
2056 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2057 
2058 	/* All our allocations have been reset */
2059 	efx_ef10_table_reset_mc_allocations(efx);
2060 
2061 	/* The datapath firmware might have been changed */
2062 	nic_data->must_check_datapath_caps = true;
2063 
2064 	/* MAC statistics have been cleared on the NIC; clear the local
2065 	 * statistic that we update with efx_update_diff_stat().
2066 	 */
2067 	nic_data->stats[EF10_STAT_port_rx_bad_bytes] = 0;
2068 }
2069 
2070 static int efx_ef10_mcdi_poll_reboot(struct efx_nic *efx)
2071 {
2072 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2073 	int rc;
2074 
2075 	rc = efx_ef10_get_warm_boot_count(efx);
2076 	if (rc < 0) {
2077 		/* The firmware is presumably in the process of
2078 		 * rebooting.  However, we are supposed to report each
2079 		 * reboot just once, so we must only do that once we
2080 		 * can read and store the updated warm boot count.
2081 		 */
2082 		return 0;
2083 	}
2084 
2085 	if (rc == nic_data->warm_boot_count)
2086 		return 0;
2087 
2088 	nic_data->warm_boot_count = rc;
2089 	efx_ef10_mcdi_reboot_detected(efx);
2090 
2091 	return -EIO;
2092 }
2093 
2094 /* Handle an MSI interrupt
2095  *
2096  * Handle an MSI hardware interrupt.  This routine schedules event
2097  * queue processing.  No interrupt acknowledgement cycle is necessary.
2098  * Also, we never need to check that the interrupt is for us, since
2099  * MSI interrupts cannot be shared.
2100  */
2101 static irqreturn_t efx_ef10_msi_interrupt(int irq, void *dev_id)
2102 {
2103 	struct efx_msi_context *context = dev_id;
2104 	struct efx_nic *efx = context->efx;
2105 
2106 	netif_vdbg(efx, intr, efx->net_dev,
2107 		   "IRQ %d on CPU %d\n", irq, raw_smp_processor_id());
2108 
2109 	if (likely(READ_ONCE(efx->irq_soft_enabled))) {
2110 		/* Note test interrupts */
2111 		if (context->index == efx->irq_level)
2112 			efx->last_irq_cpu = raw_smp_processor_id();
2113 
2114 		/* Schedule processing of the channel */
2115 		efx_schedule_channel_irq(efx->channel[context->index]);
2116 	}
2117 
2118 	return IRQ_HANDLED;
2119 }
2120 
2121 static irqreturn_t efx_ef10_legacy_interrupt(int irq, void *dev_id)
2122 {
2123 	struct efx_nic *efx = dev_id;
2124 	bool soft_enabled = READ_ONCE(efx->irq_soft_enabled);
2125 	struct efx_channel *channel;
2126 	efx_dword_t reg;
2127 	u32 queues;
2128 
2129 	/* Read the ISR which also ACKs the interrupts */
2130 	efx_readd(efx, &reg, ER_DZ_BIU_INT_ISR);
2131 	queues = EFX_DWORD_FIELD(reg, ERF_DZ_ISR_REG);
2132 
2133 	if (queues == 0)
2134 		return IRQ_NONE;
2135 
2136 	if (likely(soft_enabled)) {
2137 		/* Note test interrupts */
2138 		if (queues & (1U << efx->irq_level))
2139 			efx->last_irq_cpu = raw_smp_processor_id();
2140 
2141 		efx_for_each_channel(channel, efx) {
2142 			if (queues & 1)
2143 				efx_schedule_channel_irq(channel);
2144 			queues >>= 1;
2145 		}
2146 	}
2147 
2148 	netif_vdbg(efx, intr, efx->net_dev,
2149 		   "IRQ %d on CPU %d status " EFX_DWORD_FMT "\n",
2150 		   irq, raw_smp_processor_id(), EFX_DWORD_VAL(reg));
2151 
2152 	return IRQ_HANDLED;
2153 }
2154 
2155 static int efx_ef10_irq_test_generate(struct efx_nic *efx)
2156 {
2157 	MCDI_DECLARE_BUF(inbuf, MC_CMD_TRIGGER_INTERRUPT_IN_LEN);
2158 
2159 	if (efx_mcdi_set_workaround(efx, MC_CMD_WORKAROUND_BUG41750, true,
2160 				    NULL) == 0)
2161 		return -ENOTSUPP;
2162 
2163 	BUILD_BUG_ON(MC_CMD_TRIGGER_INTERRUPT_OUT_LEN != 0);
2164 
2165 	MCDI_SET_DWORD(inbuf, TRIGGER_INTERRUPT_IN_INTR_LEVEL, efx->irq_level);
2166 	return efx_mcdi_rpc(efx, MC_CMD_TRIGGER_INTERRUPT,
2167 			    inbuf, sizeof(inbuf), NULL, 0, NULL);
2168 }
2169 
2170 static int efx_ef10_tx_probe(struct efx_tx_queue *tx_queue)
2171 {
2172 	/* low two bits of label are what we want for type */
2173 	BUILD_BUG_ON((EFX_TXQ_TYPE_OUTER_CSUM | EFX_TXQ_TYPE_INNER_CSUM) != 3);
2174 	tx_queue->type = tx_queue->label & 3;
2175 	return efx_nic_alloc_buffer(tx_queue->efx, &tx_queue->txd.buf,
2176 				    (tx_queue->ptr_mask + 1) *
2177 				    sizeof(efx_qword_t),
2178 				    GFP_KERNEL);
2179 }
2180 
2181 /* This writes to the TX_DESC_WPTR and also pushes data */
2182 static inline void efx_ef10_push_tx_desc(struct efx_tx_queue *tx_queue,
2183 					 const efx_qword_t *txd)
2184 {
2185 	unsigned int write_ptr;
2186 	efx_oword_t reg;
2187 
2188 	write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
2189 	EFX_POPULATE_OWORD_1(reg, ERF_DZ_TX_DESC_WPTR, write_ptr);
2190 	reg.qword[0] = *txd;
2191 	efx_writeo_page(tx_queue->efx, &reg,
2192 			ER_DZ_TX_DESC_UPD, tx_queue->queue);
2193 }
2194 
2195 /* Add Firmware-Assisted TSO v2 option descriptors to a queue.
2196  */
2197 int efx_ef10_tx_tso_desc(struct efx_tx_queue *tx_queue, struct sk_buff *skb,
2198 			 bool *data_mapped)
2199 {
2200 	struct efx_tx_buffer *buffer;
2201 	u16 inner_ipv4_id = 0;
2202 	u16 outer_ipv4_id = 0;
2203 	struct tcphdr *tcp;
2204 	struct iphdr *ip;
2205 	u16 ip_tot_len;
2206 	u32 seqnum;
2207 	u32 mss;
2208 
2209 	EFX_WARN_ON_ONCE_PARANOID(tx_queue->tso_version != 2);
2210 
2211 	mss = skb_shinfo(skb)->gso_size;
2212 
2213 	if (unlikely(mss < 4)) {
2214 		WARN_ONCE(1, "MSS of %u is too small for TSO v2\n", mss);
2215 		return -EINVAL;
2216 	}
2217 
2218 	if (skb->encapsulation) {
2219 		if (!tx_queue->tso_encap)
2220 			return -EINVAL;
2221 		ip = ip_hdr(skb);
2222 		if (ip->version == 4)
2223 			outer_ipv4_id = ntohs(ip->id);
2224 
2225 		ip = inner_ip_hdr(skb);
2226 		tcp = inner_tcp_hdr(skb);
2227 	} else {
2228 		ip = ip_hdr(skb);
2229 		tcp = tcp_hdr(skb);
2230 	}
2231 
2232 	/* 8000-series EF10 hardware requires that IP Total Length be
2233 	 * greater than or equal to the value it will have in each segment
2234 	 * (which is at most mss + 208 + TCP header length), but also less
2235 	 * than (0x10000 - inner_network_header).  Otherwise the TCP
2236 	 * checksum calculation will be broken for encapsulated packets.
2237 	 * We fill in ip->tot_len with 0xff30, which should satisfy the
2238 	 * first requirement unless the MSS is ridiculously large (which
2239 	 * should be impossible as the driver max MTU is 9216); it is
2240 	 * guaranteed to satisfy the second as we only attempt TSO if
2241 	 * inner_network_header <= 208.
2242 	 */
2243 	ip_tot_len = -EFX_TSO2_MAX_HDRLEN;
2244 	EFX_WARN_ON_ONCE_PARANOID(mss + EFX_TSO2_MAX_HDRLEN +
2245 				  (tcp->doff << 2u) > ip_tot_len);
2246 
2247 	if (ip->version == 4) {
2248 		ip->tot_len = htons(ip_tot_len);
2249 		ip->check = 0;
2250 		inner_ipv4_id = ntohs(ip->id);
2251 	} else {
2252 		((struct ipv6hdr *)ip)->payload_len = htons(ip_tot_len);
2253 	}
2254 
2255 	seqnum = ntohl(tcp->seq);
2256 
2257 	buffer = efx_tx_queue_get_insert_buffer(tx_queue);
2258 
2259 	buffer->flags = EFX_TX_BUF_OPTION;
2260 	buffer->len = 0;
2261 	buffer->unmap_len = 0;
2262 	EFX_POPULATE_QWORD_5(buffer->option,
2263 			ESF_DZ_TX_DESC_IS_OPT, 1,
2264 			ESF_DZ_TX_OPTION_TYPE, ESE_DZ_TX_OPTION_DESC_TSO,
2265 			ESF_DZ_TX_TSO_OPTION_TYPE,
2266 			ESE_DZ_TX_TSO_OPTION_DESC_FATSO2A,
2267 			ESF_DZ_TX_TSO_IP_ID, inner_ipv4_id,
2268 			ESF_DZ_TX_TSO_TCP_SEQNO, seqnum
2269 			);
2270 	++tx_queue->insert_count;
2271 
2272 	buffer = efx_tx_queue_get_insert_buffer(tx_queue);
2273 
2274 	buffer->flags = EFX_TX_BUF_OPTION;
2275 	buffer->len = 0;
2276 	buffer->unmap_len = 0;
2277 	EFX_POPULATE_QWORD_5(buffer->option,
2278 			ESF_DZ_TX_DESC_IS_OPT, 1,
2279 			ESF_DZ_TX_OPTION_TYPE, ESE_DZ_TX_OPTION_DESC_TSO,
2280 			ESF_DZ_TX_TSO_OPTION_TYPE,
2281 			ESE_DZ_TX_TSO_OPTION_DESC_FATSO2B,
2282 			ESF_DZ_TX_TSO_OUTER_IPID, outer_ipv4_id,
2283 			ESF_DZ_TX_TSO_TCP_MSS, mss
2284 			);
2285 	++tx_queue->insert_count;
2286 
2287 	return 0;
2288 }
2289 
2290 static u32 efx_ef10_tso_versions(struct efx_nic *efx)
2291 {
2292 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2293 	u32 tso_versions = 0;
2294 
2295 	if (nic_data->datapath_caps &
2296 	    (1 << MC_CMD_GET_CAPABILITIES_OUT_TX_TSO_LBN))
2297 		tso_versions |= BIT(1);
2298 	if (nic_data->datapath_caps2 &
2299 	    (1 << MC_CMD_GET_CAPABILITIES_V2_OUT_TX_TSO_V2_LBN))
2300 		tso_versions |= BIT(2);
2301 	return tso_versions;
2302 }
2303 
2304 static void efx_ef10_tx_init(struct efx_tx_queue *tx_queue)
2305 {
2306 	bool csum_offload = tx_queue->type & EFX_TXQ_TYPE_OUTER_CSUM;
2307 	bool inner_csum = tx_queue->type & EFX_TXQ_TYPE_INNER_CSUM;
2308 	struct efx_channel *channel = tx_queue->channel;
2309 	struct efx_nic *efx = tx_queue->efx;
2310 	struct efx_ef10_nic_data *nic_data;
2311 	efx_qword_t *txd;
2312 	int rc;
2313 
2314 	nic_data = efx->nic_data;
2315 
2316 	/* Only attempt to enable TX timestamping if we have the license for it,
2317 	 * otherwise TXQ init will fail
2318 	 */
2319 	if (!(nic_data->licensed_features &
2320 	      (1 << LICENSED_V3_FEATURES_TX_TIMESTAMPS_LBN))) {
2321 		tx_queue->timestamping = false;
2322 		/* Disable sync events on this channel. */
2323 		if (efx->type->ptp_set_ts_sync_events)
2324 			efx->type->ptp_set_ts_sync_events(efx, false, false);
2325 	}
2326 
2327 	/* TSOv2 is a limited resource that can only be configured on a limited
2328 	 * number of queues. TSO without checksum offload is not really a thing,
2329 	 * so we only enable it for those queues.
2330 	 * TSOv2 cannot be used with Hardware timestamping, and is never needed
2331 	 * for XDP tx.
2332 	 */
2333 	if (efx_has_cap(efx, TX_TSO_V2)) {
2334 		if ((csum_offload || inner_csum) &&
2335 		    !tx_queue->timestamping && !tx_queue->xdp_tx) {
2336 			tx_queue->tso_version = 2;
2337 			netif_dbg(efx, hw, efx->net_dev, "Using TSOv2 for channel %u\n",
2338 				  channel->channel);
2339 		}
2340 	} else if (efx_has_cap(efx, TX_TSO)) {
2341 		tx_queue->tso_version = 1;
2342 	}
2343 
2344 	rc = efx_mcdi_tx_init(tx_queue);
2345 	if (rc)
2346 		goto fail;
2347 
2348 	/* A previous user of this TX queue might have set us up the
2349 	 * bomb by writing a descriptor to the TX push collector but
2350 	 * not the doorbell.  (Each collector belongs to a port, not a
2351 	 * queue or function, so cannot easily be reset.)  We must
2352 	 * attempt to push a no-op descriptor in its place.
2353 	 */
2354 	tx_queue->buffer[0].flags = EFX_TX_BUF_OPTION;
2355 	tx_queue->insert_count = 1;
2356 	txd = efx_tx_desc(tx_queue, 0);
2357 	EFX_POPULATE_QWORD_7(*txd,
2358 			     ESF_DZ_TX_DESC_IS_OPT, true,
2359 			     ESF_DZ_TX_OPTION_TYPE,
2360 			     ESE_DZ_TX_OPTION_DESC_CRC_CSUM,
2361 			     ESF_DZ_TX_OPTION_UDP_TCP_CSUM, csum_offload,
2362 			     ESF_DZ_TX_OPTION_IP_CSUM, csum_offload && tx_queue->tso_version != 2,
2363 			     ESF_DZ_TX_OPTION_INNER_UDP_TCP_CSUM, inner_csum,
2364 			     ESF_DZ_TX_OPTION_INNER_IP_CSUM, inner_csum && tx_queue->tso_version != 2,
2365 			     ESF_DZ_TX_TIMESTAMP, tx_queue->timestamping);
2366 	tx_queue->write_count = 1;
2367 
2368 	if (tx_queue->tso_version == 2 && efx_has_cap(efx, TX_TSO_V2_ENCAP))
2369 		tx_queue->tso_encap = true;
2370 
2371 	wmb();
2372 	efx_ef10_push_tx_desc(tx_queue, txd);
2373 
2374 	return;
2375 
2376 fail:
2377 	netdev_WARN(efx->net_dev, "failed to initialise TXQ %d\n",
2378 		    tx_queue->queue);
2379 }
2380 
2381 /* This writes to the TX_DESC_WPTR; write pointer for TX descriptor ring */
2382 static inline void efx_ef10_notify_tx_desc(struct efx_tx_queue *tx_queue)
2383 {
2384 	unsigned int write_ptr;
2385 	efx_dword_t reg;
2386 
2387 	write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
2388 	EFX_POPULATE_DWORD_1(reg, ERF_DZ_TX_DESC_WPTR_DWORD, write_ptr);
2389 	efx_writed_page(tx_queue->efx, &reg,
2390 			ER_DZ_TX_DESC_UPD_DWORD, tx_queue->queue);
2391 }
2392 
2393 #define EFX_EF10_MAX_TX_DESCRIPTOR_LEN 0x3fff
2394 
2395 static unsigned int efx_ef10_tx_limit_len(struct efx_tx_queue *tx_queue,
2396 					  dma_addr_t dma_addr, unsigned int len)
2397 {
2398 	if (len > EFX_EF10_MAX_TX_DESCRIPTOR_LEN) {
2399 		/* If we need to break across multiple descriptors we should
2400 		 * stop at a page boundary. This assumes the length limit is
2401 		 * greater than the page size.
2402 		 */
2403 		dma_addr_t end = dma_addr + EFX_EF10_MAX_TX_DESCRIPTOR_LEN;
2404 
2405 		BUILD_BUG_ON(EFX_EF10_MAX_TX_DESCRIPTOR_LEN < EFX_PAGE_SIZE);
2406 		len = (end & (~(EFX_PAGE_SIZE - 1))) - dma_addr;
2407 	}
2408 
2409 	return len;
2410 }
2411 
2412 static void efx_ef10_tx_write(struct efx_tx_queue *tx_queue)
2413 {
2414 	unsigned int old_write_count = tx_queue->write_count;
2415 	struct efx_tx_buffer *buffer;
2416 	unsigned int write_ptr;
2417 	efx_qword_t *txd;
2418 
2419 	tx_queue->xmit_pending = false;
2420 	if (unlikely(tx_queue->write_count == tx_queue->insert_count))
2421 		return;
2422 
2423 	do {
2424 		write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
2425 		buffer = &tx_queue->buffer[write_ptr];
2426 		txd = efx_tx_desc(tx_queue, write_ptr);
2427 		++tx_queue->write_count;
2428 
2429 		/* Create TX descriptor ring entry */
2430 		if (buffer->flags & EFX_TX_BUF_OPTION) {
2431 			*txd = buffer->option;
2432 			if (EFX_QWORD_FIELD(*txd, ESF_DZ_TX_OPTION_TYPE) == 1)
2433 				/* PIO descriptor */
2434 				tx_queue->packet_write_count = tx_queue->write_count;
2435 		} else {
2436 			tx_queue->packet_write_count = tx_queue->write_count;
2437 			BUILD_BUG_ON(EFX_TX_BUF_CONT != 1);
2438 			EFX_POPULATE_QWORD_3(
2439 				*txd,
2440 				ESF_DZ_TX_KER_CONT,
2441 				buffer->flags & EFX_TX_BUF_CONT,
2442 				ESF_DZ_TX_KER_BYTE_CNT, buffer->len,
2443 				ESF_DZ_TX_KER_BUF_ADDR, buffer->dma_addr);
2444 		}
2445 	} while (tx_queue->write_count != tx_queue->insert_count);
2446 
2447 	wmb(); /* Ensure descriptors are written before they are fetched */
2448 
2449 	if (efx_nic_may_push_tx_desc(tx_queue, old_write_count)) {
2450 		txd = efx_tx_desc(tx_queue,
2451 				  old_write_count & tx_queue->ptr_mask);
2452 		efx_ef10_push_tx_desc(tx_queue, txd);
2453 		++tx_queue->pushes;
2454 	} else {
2455 		efx_ef10_notify_tx_desc(tx_queue);
2456 	}
2457 }
2458 
2459 static int efx_ef10_probe_multicast_chaining(struct efx_nic *efx)
2460 {
2461 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2462 	unsigned int enabled, implemented;
2463 	bool want_workaround_26807;
2464 	int rc;
2465 
2466 	rc = efx_mcdi_get_workarounds(efx, &implemented, &enabled);
2467 	if (rc == -ENOSYS) {
2468 		/* GET_WORKAROUNDS was implemented before this workaround,
2469 		 * thus it must be unavailable in this firmware.
2470 		 */
2471 		nic_data->workaround_26807 = false;
2472 		return 0;
2473 	}
2474 	if (rc)
2475 		return rc;
2476 	want_workaround_26807 =
2477 		implemented & MC_CMD_GET_WORKAROUNDS_OUT_BUG26807;
2478 	nic_data->workaround_26807 =
2479 		!!(enabled & MC_CMD_GET_WORKAROUNDS_OUT_BUG26807);
2480 
2481 	if (want_workaround_26807 && !nic_data->workaround_26807) {
2482 		unsigned int flags;
2483 
2484 		rc = efx_mcdi_set_workaround(efx,
2485 					     MC_CMD_WORKAROUND_BUG26807,
2486 					     true, &flags);
2487 		if (!rc) {
2488 			if (flags &
2489 			    1 << MC_CMD_WORKAROUND_EXT_OUT_FLR_DONE_LBN) {
2490 				netif_info(efx, drv, efx->net_dev,
2491 					   "other functions on NIC have been reset\n");
2492 
2493 				/* With MCFW v4.6.x and earlier, the
2494 				 * boot count will have incremented,
2495 				 * so re-read the warm_boot_count
2496 				 * value now to ensure this function
2497 				 * doesn't think it has changed next
2498 				 * time it checks.
2499 				 */
2500 				rc = efx_ef10_get_warm_boot_count(efx);
2501 				if (rc >= 0) {
2502 					nic_data->warm_boot_count = rc;
2503 					rc = 0;
2504 				}
2505 			}
2506 			nic_data->workaround_26807 = true;
2507 		} else if (rc == -EPERM) {
2508 			rc = 0;
2509 		}
2510 	}
2511 	return rc;
2512 }
2513 
2514 static int efx_ef10_filter_table_probe(struct efx_nic *efx)
2515 {
2516 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2517 	int rc = efx_ef10_probe_multicast_chaining(efx);
2518 	struct efx_mcdi_filter_vlan *vlan;
2519 
2520 	if (rc)
2521 		return rc;
2522 	rc = efx_mcdi_filter_table_probe(efx, nic_data->workaround_26807);
2523 
2524 	if (rc)
2525 		return rc;
2526 
2527 	list_for_each_entry(vlan, &nic_data->vlan_list, list) {
2528 		rc = efx_mcdi_filter_add_vlan(efx, vlan->vid);
2529 		if (rc)
2530 			goto fail_add_vlan;
2531 	}
2532 	return 0;
2533 
2534 fail_add_vlan:
2535 	efx_mcdi_filter_table_remove(efx);
2536 	return rc;
2537 }
2538 
2539 /* This creates an entry in the RX descriptor queue */
2540 static inline void
2541 efx_ef10_build_rx_desc(struct efx_rx_queue *rx_queue, unsigned int index)
2542 {
2543 	struct efx_rx_buffer *rx_buf;
2544 	efx_qword_t *rxd;
2545 
2546 	rxd = efx_rx_desc(rx_queue, index);
2547 	rx_buf = efx_rx_buffer(rx_queue, index);
2548 	EFX_POPULATE_QWORD_2(*rxd,
2549 			     ESF_DZ_RX_KER_BYTE_CNT, rx_buf->len,
2550 			     ESF_DZ_RX_KER_BUF_ADDR, rx_buf->dma_addr);
2551 }
2552 
2553 static void efx_ef10_rx_write(struct efx_rx_queue *rx_queue)
2554 {
2555 	struct efx_nic *efx = rx_queue->efx;
2556 	unsigned int write_count;
2557 	efx_dword_t reg;
2558 
2559 	/* Firmware requires that RX_DESC_WPTR be a multiple of 8 */
2560 	write_count = rx_queue->added_count & ~7;
2561 	if (rx_queue->notified_count == write_count)
2562 		return;
2563 
2564 	do
2565 		efx_ef10_build_rx_desc(
2566 			rx_queue,
2567 			rx_queue->notified_count & rx_queue->ptr_mask);
2568 	while (++rx_queue->notified_count != write_count);
2569 
2570 	wmb();
2571 	EFX_POPULATE_DWORD_1(reg, ERF_DZ_RX_DESC_WPTR,
2572 			     write_count & rx_queue->ptr_mask);
2573 	efx_writed_page(efx, &reg, ER_DZ_RX_DESC_UPD,
2574 			efx_rx_queue_index(rx_queue));
2575 }
2576 
2577 static efx_mcdi_async_completer efx_ef10_rx_defer_refill_complete;
2578 
2579 static void efx_ef10_rx_defer_refill(struct efx_rx_queue *rx_queue)
2580 {
2581 	struct efx_channel *channel = efx_rx_queue_channel(rx_queue);
2582 	MCDI_DECLARE_BUF(inbuf, MC_CMD_DRIVER_EVENT_IN_LEN);
2583 	efx_qword_t event;
2584 
2585 	EFX_POPULATE_QWORD_2(event,
2586 			     ESF_DZ_EV_CODE, EFX_EF10_DRVGEN_EV,
2587 			     ESF_DZ_EV_DATA, EFX_EF10_REFILL);
2588 
2589 	MCDI_SET_DWORD(inbuf, DRIVER_EVENT_IN_EVQ, channel->channel);
2590 
2591 	/* MCDI_SET_QWORD is not appropriate here since EFX_POPULATE_* has
2592 	 * already swapped the data to little-endian order.
2593 	 */
2594 	memcpy(MCDI_PTR(inbuf, DRIVER_EVENT_IN_DATA), &event.u64[0],
2595 	       sizeof(efx_qword_t));
2596 
2597 	efx_mcdi_rpc_async(channel->efx, MC_CMD_DRIVER_EVENT,
2598 			   inbuf, sizeof(inbuf), 0,
2599 			   efx_ef10_rx_defer_refill_complete, 0);
2600 }
2601 
2602 static void
2603 efx_ef10_rx_defer_refill_complete(struct efx_nic *efx, unsigned long cookie,
2604 				  int rc, efx_dword_t *outbuf,
2605 				  size_t outlen_actual)
2606 {
2607 	/* nothing to do */
2608 }
2609 
2610 static int efx_ef10_ev_init(struct efx_channel *channel)
2611 {
2612 	struct efx_nic *efx = channel->efx;
2613 	struct efx_ef10_nic_data *nic_data;
2614 	bool use_v2, cut_thru;
2615 
2616 	nic_data = efx->nic_data;
2617 	use_v2 = nic_data->datapath_caps2 &
2618 			    1 << MC_CMD_GET_CAPABILITIES_V2_OUT_INIT_EVQ_V2_LBN;
2619 	cut_thru = !(nic_data->datapath_caps &
2620 			      1 << MC_CMD_GET_CAPABILITIES_OUT_RX_BATCHING_LBN);
2621 	return efx_mcdi_ev_init(channel, cut_thru, use_v2);
2622 }
2623 
2624 static void efx_ef10_handle_rx_wrong_queue(struct efx_rx_queue *rx_queue,
2625 					   unsigned int rx_queue_label)
2626 {
2627 	struct efx_nic *efx = rx_queue->efx;
2628 
2629 	netif_info(efx, hw, efx->net_dev,
2630 		   "rx event arrived on queue %d labeled as queue %u\n",
2631 		   efx_rx_queue_index(rx_queue), rx_queue_label);
2632 
2633 	efx_schedule_reset(efx, RESET_TYPE_DISABLE);
2634 }
2635 
2636 static void
2637 efx_ef10_handle_rx_bad_lbits(struct efx_rx_queue *rx_queue,
2638 			     unsigned int actual, unsigned int expected)
2639 {
2640 	unsigned int dropped = (actual - expected) & rx_queue->ptr_mask;
2641 	struct efx_nic *efx = rx_queue->efx;
2642 
2643 	netif_info(efx, hw, efx->net_dev,
2644 		   "dropped %d events (index=%d expected=%d)\n",
2645 		   dropped, actual, expected);
2646 
2647 	efx_schedule_reset(efx, RESET_TYPE_DISABLE);
2648 }
2649 
2650 /* partially received RX was aborted. clean up. */
2651 static void efx_ef10_handle_rx_abort(struct efx_rx_queue *rx_queue)
2652 {
2653 	unsigned int rx_desc_ptr;
2654 
2655 	netif_dbg(rx_queue->efx, hw, rx_queue->efx->net_dev,
2656 		  "scattered RX aborted (dropping %u buffers)\n",
2657 		  rx_queue->scatter_n);
2658 
2659 	rx_desc_ptr = rx_queue->removed_count & rx_queue->ptr_mask;
2660 
2661 	efx_rx_packet(rx_queue, rx_desc_ptr, rx_queue->scatter_n,
2662 		      0, EFX_RX_PKT_DISCARD);
2663 
2664 	rx_queue->removed_count += rx_queue->scatter_n;
2665 	rx_queue->scatter_n = 0;
2666 	rx_queue->scatter_len = 0;
2667 	++efx_rx_queue_channel(rx_queue)->n_rx_nodesc_trunc;
2668 }
2669 
2670 static u16 efx_ef10_handle_rx_event_errors(struct efx_channel *channel,
2671 					   unsigned int n_packets,
2672 					   unsigned int rx_encap_hdr,
2673 					   unsigned int rx_l3_class,
2674 					   unsigned int rx_l4_class,
2675 					   const efx_qword_t *event)
2676 {
2677 	struct efx_nic *efx = channel->efx;
2678 	bool handled = false;
2679 
2680 	if (EFX_QWORD_FIELD(*event, ESF_DZ_RX_ECRC_ERR)) {
2681 		if (!(efx->net_dev->features & NETIF_F_RXALL)) {
2682 			if (!efx->loopback_selftest)
2683 				channel->n_rx_eth_crc_err += n_packets;
2684 			return EFX_RX_PKT_DISCARD;
2685 		}
2686 		handled = true;
2687 	}
2688 	if (EFX_QWORD_FIELD(*event, ESF_DZ_RX_IPCKSUM_ERR)) {
2689 		if (unlikely(rx_encap_hdr != ESE_EZ_ENCAP_HDR_VXLAN &&
2690 			     rx_l3_class != ESE_DZ_L3_CLASS_IP4 &&
2691 			     rx_l3_class != ESE_DZ_L3_CLASS_IP4_FRAG &&
2692 			     rx_l3_class != ESE_DZ_L3_CLASS_IP6 &&
2693 			     rx_l3_class != ESE_DZ_L3_CLASS_IP6_FRAG))
2694 			netdev_WARN(efx->net_dev,
2695 				    "invalid class for RX_IPCKSUM_ERR: event="
2696 				    EFX_QWORD_FMT "\n",
2697 				    EFX_QWORD_VAL(*event));
2698 		if (!efx->loopback_selftest)
2699 			*(rx_encap_hdr ?
2700 			  &channel->n_rx_outer_ip_hdr_chksum_err :
2701 			  &channel->n_rx_ip_hdr_chksum_err) += n_packets;
2702 		return 0;
2703 	}
2704 	if (EFX_QWORD_FIELD(*event, ESF_DZ_RX_TCPUDP_CKSUM_ERR)) {
2705 		if (unlikely(rx_encap_hdr != ESE_EZ_ENCAP_HDR_VXLAN &&
2706 			     ((rx_l3_class != ESE_DZ_L3_CLASS_IP4 &&
2707 			       rx_l3_class != ESE_DZ_L3_CLASS_IP6) ||
2708 			      (rx_l4_class != ESE_FZ_L4_CLASS_TCP &&
2709 			       rx_l4_class != ESE_FZ_L4_CLASS_UDP))))
2710 			netdev_WARN(efx->net_dev,
2711 				    "invalid class for RX_TCPUDP_CKSUM_ERR: event="
2712 				    EFX_QWORD_FMT "\n",
2713 				    EFX_QWORD_VAL(*event));
2714 		if (!efx->loopback_selftest)
2715 			*(rx_encap_hdr ?
2716 			  &channel->n_rx_outer_tcp_udp_chksum_err :
2717 			  &channel->n_rx_tcp_udp_chksum_err) += n_packets;
2718 		return 0;
2719 	}
2720 	if (EFX_QWORD_FIELD(*event, ESF_EZ_RX_IP_INNER_CHKSUM_ERR)) {
2721 		if (unlikely(!rx_encap_hdr))
2722 			netdev_WARN(efx->net_dev,
2723 				    "invalid encapsulation type for RX_IP_INNER_CHKSUM_ERR: event="
2724 				    EFX_QWORD_FMT "\n",
2725 				    EFX_QWORD_VAL(*event));
2726 		else if (unlikely(rx_l3_class != ESE_DZ_L3_CLASS_IP4 &&
2727 				  rx_l3_class != ESE_DZ_L3_CLASS_IP4_FRAG &&
2728 				  rx_l3_class != ESE_DZ_L3_CLASS_IP6 &&
2729 				  rx_l3_class != ESE_DZ_L3_CLASS_IP6_FRAG))
2730 			netdev_WARN(efx->net_dev,
2731 				    "invalid class for RX_IP_INNER_CHKSUM_ERR: event="
2732 				    EFX_QWORD_FMT "\n",
2733 				    EFX_QWORD_VAL(*event));
2734 		if (!efx->loopback_selftest)
2735 			channel->n_rx_inner_ip_hdr_chksum_err += n_packets;
2736 		return 0;
2737 	}
2738 	if (EFX_QWORD_FIELD(*event, ESF_EZ_RX_TCP_UDP_INNER_CHKSUM_ERR)) {
2739 		if (unlikely(!rx_encap_hdr))
2740 			netdev_WARN(efx->net_dev,
2741 				    "invalid encapsulation type for RX_TCP_UDP_INNER_CHKSUM_ERR: event="
2742 				    EFX_QWORD_FMT "\n",
2743 				    EFX_QWORD_VAL(*event));
2744 		else if (unlikely((rx_l3_class != ESE_DZ_L3_CLASS_IP4 &&
2745 				   rx_l3_class != ESE_DZ_L3_CLASS_IP6) ||
2746 				  (rx_l4_class != ESE_FZ_L4_CLASS_TCP &&
2747 				   rx_l4_class != ESE_FZ_L4_CLASS_UDP)))
2748 			netdev_WARN(efx->net_dev,
2749 				    "invalid class for RX_TCP_UDP_INNER_CHKSUM_ERR: event="
2750 				    EFX_QWORD_FMT "\n",
2751 				    EFX_QWORD_VAL(*event));
2752 		if (!efx->loopback_selftest)
2753 			channel->n_rx_inner_tcp_udp_chksum_err += n_packets;
2754 		return 0;
2755 	}
2756 
2757 	WARN_ON(!handled); /* No error bits were recognised */
2758 	return 0;
2759 }
2760 
2761 static int efx_ef10_handle_rx_event(struct efx_channel *channel,
2762 				    const efx_qword_t *event)
2763 {
2764 	unsigned int rx_bytes, next_ptr_lbits, rx_queue_label;
2765 	unsigned int rx_l3_class, rx_l4_class, rx_encap_hdr;
2766 	unsigned int n_descs, n_packets, i;
2767 	struct efx_nic *efx = channel->efx;
2768 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2769 	struct efx_rx_queue *rx_queue;
2770 	efx_qword_t errors;
2771 	bool rx_cont;
2772 	u16 flags = 0;
2773 
2774 	if (unlikely(READ_ONCE(efx->reset_pending)))
2775 		return 0;
2776 
2777 	/* Basic packet information */
2778 	rx_bytes = EFX_QWORD_FIELD(*event, ESF_DZ_RX_BYTES);
2779 	next_ptr_lbits = EFX_QWORD_FIELD(*event, ESF_DZ_RX_DSC_PTR_LBITS);
2780 	rx_queue_label = EFX_QWORD_FIELD(*event, ESF_DZ_RX_QLABEL);
2781 	rx_l3_class = EFX_QWORD_FIELD(*event, ESF_DZ_RX_L3_CLASS);
2782 	rx_l4_class = EFX_QWORD_FIELD(*event, ESF_FZ_RX_L4_CLASS);
2783 	rx_cont = EFX_QWORD_FIELD(*event, ESF_DZ_RX_CONT);
2784 	rx_encap_hdr =
2785 		nic_data->datapath_caps &
2786 			(1 << MC_CMD_GET_CAPABILITIES_OUT_VXLAN_NVGRE_LBN) ?
2787 		EFX_QWORD_FIELD(*event, ESF_EZ_RX_ENCAP_HDR) :
2788 		ESE_EZ_ENCAP_HDR_NONE;
2789 
2790 	if (EFX_QWORD_FIELD(*event, ESF_DZ_RX_DROP_EVENT))
2791 		netdev_WARN(efx->net_dev, "saw RX_DROP_EVENT: event="
2792 			    EFX_QWORD_FMT "\n",
2793 			    EFX_QWORD_VAL(*event));
2794 
2795 	rx_queue = efx_channel_get_rx_queue(channel);
2796 
2797 	if (unlikely(rx_queue_label != efx_rx_queue_index(rx_queue)))
2798 		efx_ef10_handle_rx_wrong_queue(rx_queue, rx_queue_label);
2799 
2800 	n_descs = ((next_ptr_lbits - rx_queue->removed_count) &
2801 		   ((1 << ESF_DZ_RX_DSC_PTR_LBITS_WIDTH) - 1));
2802 
2803 	if (n_descs != rx_queue->scatter_n + 1) {
2804 		struct efx_ef10_nic_data *nic_data = efx->nic_data;
2805 
2806 		/* detect rx abort */
2807 		if (unlikely(n_descs == rx_queue->scatter_n)) {
2808 			if (rx_queue->scatter_n == 0 || rx_bytes != 0)
2809 				netdev_WARN(efx->net_dev,
2810 					    "invalid RX abort: scatter_n=%u event="
2811 					    EFX_QWORD_FMT "\n",
2812 					    rx_queue->scatter_n,
2813 					    EFX_QWORD_VAL(*event));
2814 			efx_ef10_handle_rx_abort(rx_queue);
2815 			return 0;
2816 		}
2817 
2818 		/* Check that RX completion merging is valid, i.e.
2819 		 * the current firmware supports it and this is a
2820 		 * non-scattered packet.
2821 		 */
2822 		if (!(nic_data->datapath_caps &
2823 		      (1 << MC_CMD_GET_CAPABILITIES_OUT_RX_BATCHING_LBN)) ||
2824 		    rx_queue->scatter_n != 0 || rx_cont) {
2825 			efx_ef10_handle_rx_bad_lbits(
2826 				rx_queue, next_ptr_lbits,
2827 				(rx_queue->removed_count +
2828 				 rx_queue->scatter_n + 1) &
2829 				((1 << ESF_DZ_RX_DSC_PTR_LBITS_WIDTH) - 1));
2830 			return 0;
2831 		}
2832 
2833 		/* Merged completion for multiple non-scattered packets */
2834 		rx_queue->scatter_n = 1;
2835 		rx_queue->scatter_len = 0;
2836 		n_packets = n_descs;
2837 		++channel->n_rx_merge_events;
2838 		channel->n_rx_merge_packets += n_packets;
2839 		flags |= EFX_RX_PKT_PREFIX_LEN;
2840 	} else {
2841 		++rx_queue->scatter_n;
2842 		rx_queue->scatter_len += rx_bytes;
2843 		if (rx_cont)
2844 			return 0;
2845 		n_packets = 1;
2846 	}
2847 
2848 	EFX_POPULATE_QWORD_5(errors, ESF_DZ_RX_ECRC_ERR, 1,
2849 				     ESF_DZ_RX_IPCKSUM_ERR, 1,
2850 				     ESF_DZ_RX_TCPUDP_CKSUM_ERR, 1,
2851 				     ESF_EZ_RX_IP_INNER_CHKSUM_ERR, 1,
2852 				     ESF_EZ_RX_TCP_UDP_INNER_CHKSUM_ERR, 1);
2853 	EFX_AND_QWORD(errors, *event, errors);
2854 	if (unlikely(!EFX_QWORD_IS_ZERO(errors))) {
2855 		flags |= efx_ef10_handle_rx_event_errors(channel, n_packets,
2856 							 rx_encap_hdr,
2857 							 rx_l3_class, rx_l4_class,
2858 							 event);
2859 	} else {
2860 		bool tcpudp = rx_l4_class == ESE_FZ_L4_CLASS_TCP ||
2861 			      rx_l4_class == ESE_FZ_L4_CLASS_UDP;
2862 
2863 		switch (rx_encap_hdr) {
2864 		case ESE_EZ_ENCAP_HDR_VXLAN: /* VxLAN or GENEVE */
2865 			flags |= EFX_RX_PKT_CSUMMED; /* outer UDP csum */
2866 			if (tcpudp)
2867 				flags |= EFX_RX_PKT_CSUM_LEVEL; /* inner L4 */
2868 			break;
2869 		case ESE_EZ_ENCAP_HDR_GRE:
2870 		case ESE_EZ_ENCAP_HDR_NONE:
2871 			if (tcpudp)
2872 				flags |= EFX_RX_PKT_CSUMMED;
2873 			break;
2874 		default:
2875 			netdev_WARN(efx->net_dev,
2876 				    "unknown encapsulation type: event="
2877 				    EFX_QWORD_FMT "\n",
2878 				    EFX_QWORD_VAL(*event));
2879 		}
2880 	}
2881 
2882 	if (rx_l4_class == ESE_FZ_L4_CLASS_TCP)
2883 		flags |= EFX_RX_PKT_TCP;
2884 
2885 	channel->irq_mod_score += 2 * n_packets;
2886 
2887 	/* Handle received packet(s) */
2888 	for (i = 0; i < n_packets; i++) {
2889 		efx_rx_packet(rx_queue,
2890 			      rx_queue->removed_count & rx_queue->ptr_mask,
2891 			      rx_queue->scatter_n, rx_queue->scatter_len,
2892 			      flags);
2893 		rx_queue->removed_count += rx_queue->scatter_n;
2894 	}
2895 
2896 	rx_queue->scatter_n = 0;
2897 	rx_queue->scatter_len = 0;
2898 
2899 	return n_packets;
2900 }
2901 
2902 static u32 efx_ef10_extract_event_ts(efx_qword_t *event)
2903 {
2904 	u32 tstamp;
2905 
2906 	tstamp = EFX_QWORD_FIELD(*event, TX_TIMESTAMP_EVENT_TSTAMP_DATA_HI);
2907 	tstamp <<= 16;
2908 	tstamp |= EFX_QWORD_FIELD(*event, TX_TIMESTAMP_EVENT_TSTAMP_DATA_LO);
2909 
2910 	return tstamp;
2911 }
2912 
2913 static void
2914 efx_ef10_handle_tx_event(struct efx_channel *channel, efx_qword_t *event)
2915 {
2916 	struct efx_nic *efx = channel->efx;
2917 	struct efx_tx_queue *tx_queue;
2918 	unsigned int tx_ev_desc_ptr;
2919 	unsigned int tx_ev_q_label;
2920 	unsigned int tx_ev_type;
2921 	u64 ts_part;
2922 
2923 	if (unlikely(READ_ONCE(efx->reset_pending)))
2924 		return;
2925 
2926 	if (unlikely(EFX_QWORD_FIELD(*event, ESF_DZ_TX_DROP_EVENT)))
2927 		return;
2928 
2929 	/* Get the transmit queue */
2930 	tx_ev_q_label = EFX_QWORD_FIELD(*event, ESF_DZ_TX_QLABEL);
2931 	tx_queue = efx_channel_get_tx_queue(channel,
2932 					    tx_ev_q_label % EFX_MAX_TXQ_PER_CHANNEL);
2933 
2934 	if (!tx_queue->timestamping) {
2935 		/* Transmit completion */
2936 		tx_ev_desc_ptr = EFX_QWORD_FIELD(*event, ESF_DZ_TX_DESCR_INDX);
2937 		efx_xmit_done(tx_queue, tx_ev_desc_ptr & tx_queue->ptr_mask);
2938 		return;
2939 	}
2940 
2941 	/* Transmit timestamps are only available for 8XXX series. They result
2942 	 * in up to three events per packet. These occur in order, and are:
2943 	 *  - the normal completion event (may be omitted)
2944 	 *  - the low part of the timestamp
2945 	 *  - the high part of the timestamp
2946 	 *
2947 	 * It's possible for multiple completion events to appear before the
2948 	 * corresponding timestamps. So we can for example get:
2949 	 *  COMP N
2950 	 *  COMP N+1
2951 	 *  TS_LO N
2952 	 *  TS_HI N
2953 	 *  TS_LO N+1
2954 	 *  TS_HI N+1
2955 	 *
2956 	 * In addition it's also possible for the adjacent completions to be
2957 	 * merged, so we may not see COMP N above. As such, the completion
2958 	 * events are not very useful here.
2959 	 *
2960 	 * Each part of the timestamp is itself split across two 16 bit
2961 	 * fields in the event.
2962 	 */
2963 	tx_ev_type = EFX_QWORD_FIELD(*event, ESF_EZ_TX_SOFT1);
2964 
2965 	switch (tx_ev_type) {
2966 	case TX_TIMESTAMP_EVENT_TX_EV_COMPLETION:
2967 		/* Ignore this event - see above. */
2968 		break;
2969 
2970 	case TX_TIMESTAMP_EVENT_TX_EV_TSTAMP_LO:
2971 		ts_part = efx_ef10_extract_event_ts(event);
2972 		tx_queue->completed_timestamp_minor = ts_part;
2973 		break;
2974 
2975 	case TX_TIMESTAMP_EVENT_TX_EV_TSTAMP_HI:
2976 		ts_part = efx_ef10_extract_event_ts(event);
2977 		tx_queue->completed_timestamp_major = ts_part;
2978 
2979 		efx_xmit_done_single(tx_queue);
2980 		break;
2981 
2982 	default:
2983 		netif_err(efx, hw, efx->net_dev,
2984 			  "channel %d unknown tx event type %d (data "
2985 			  EFX_QWORD_FMT ")\n",
2986 			  channel->channel, tx_ev_type,
2987 			  EFX_QWORD_VAL(*event));
2988 		break;
2989 	}
2990 }
2991 
2992 static void
2993 efx_ef10_handle_driver_event(struct efx_channel *channel, efx_qword_t *event)
2994 {
2995 	struct efx_nic *efx = channel->efx;
2996 	int subcode;
2997 
2998 	subcode = EFX_QWORD_FIELD(*event, ESF_DZ_DRV_SUB_CODE);
2999 
3000 	switch (subcode) {
3001 	case ESE_DZ_DRV_TIMER_EV:
3002 	case ESE_DZ_DRV_WAKE_UP_EV:
3003 		break;
3004 	case ESE_DZ_DRV_START_UP_EV:
3005 		/* event queue init complete. ok. */
3006 		break;
3007 	default:
3008 		netif_err(efx, hw, efx->net_dev,
3009 			  "channel %d unknown driver event type %d"
3010 			  " (data " EFX_QWORD_FMT ")\n",
3011 			  channel->channel, subcode,
3012 			  EFX_QWORD_VAL(*event));
3013 
3014 	}
3015 }
3016 
3017 static void efx_ef10_handle_driver_generated_event(struct efx_channel *channel,
3018 						   efx_qword_t *event)
3019 {
3020 	struct efx_nic *efx = channel->efx;
3021 	u32 subcode;
3022 
3023 	subcode = EFX_QWORD_FIELD(*event, EFX_DWORD_0);
3024 
3025 	switch (subcode) {
3026 	case EFX_EF10_TEST:
3027 		channel->event_test_cpu = raw_smp_processor_id();
3028 		break;
3029 	case EFX_EF10_REFILL:
3030 		/* The queue must be empty, so we won't receive any rx
3031 		 * events, so efx_process_channel() won't refill the
3032 		 * queue. Refill it here
3033 		 */
3034 		efx_fast_push_rx_descriptors(&channel->rx_queue, true);
3035 		break;
3036 	default:
3037 		netif_err(efx, hw, efx->net_dev,
3038 			  "channel %d unknown driver event type %u"
3039 			  " (data " EFX_QWORD_FMT ")\n",
3040 			  channel->channel, (unsigned) subcode,
3041 			  EFX_QWORD_VAL(*event));
3042 	}
3043 }
3044 
3045 static int efx_ef10_ev_process(struct efx_channel *channel, int quota)
3046 {
3047 	struct efx_nic *efx = channel->efx;
3048 	efx_qword_t event, *p_event;
3049 	unsigned int read_ptr;
3050 	int ev_code;
3051 	int spent = 0;
3052 
3053 	if (quota <= 0)
3054 		return spent;
3055 
3056 	read_ptr = channel->eventq_read_ptr;
3057 
3058 	for (;;) {
3059 		p_event = efx_event(channel, read_ptr);
3060 		event = *p_event;
3061 
3062 		if (!efx_event_present(&event))
3063 			break;
3064 
3065 		EFX_SET_QWORD(*p_event);
3066 
3067 		++read_ptr;
3068 
3069 		ev_code = EFX_QWORD_FIELD(event, ESF_DZ_EV_CODE);
3070 
3071 		netif_vdbg(efx, drv, efx->net_dev,
3072 			   "processing event on %d " EFX_QWORD_FMT "\n",
3073 			   channel->channel, EFX_QWORD_VAL(event));
3074 
3075 		switch (ev_code) {
3076 		case ESE_DZ_EV_CODE_MCDI_EV:
3077 			efx_mcdi_process_event(channel, &event);
3078 			break;
3079 		case ESE_DZ_EV_CODE_RX_EV:
3080 			spent += efx_ef10_handle_rx_event(channel, &event);
3081 			if (spent >= quota) {
3082 				/* XXX can we split a merged event to
3083 				 * avoid going over-quota?
3084 				 */
3085 				spent = quota;
3086 				goto out;
3087 			}
3088 			break;
3089 		case ESE_DZ_EV_CODE_TX_EV:
3090 			efx_ef10_handle_tx_event(channel, &event);
3091 			break;
3092 		case ESE_DZ_EV_CODE_DRIVER_EV:
3093 			efx_ef10_handle_driver_event(channel, &event);
3094 			if (++spent == quota)
3095 				goto out;
3096 			break;
3097 		case EFX_EF10_DRVGEN_EV:
3098 			efx_ef10_handle_driver_generated_event(channel, &event);
3099 			break;
3100 		default:
3101 			netif_err(efx, hw, efx->net_dev,
3102 				  "channel %d unknown event type %d"
3103 				  " (data " EFX_QWORD_FMT ")\n",
3104 				  channel->channel, ev_code,
3105 				  EFX_QWORD_VAL(event));
3106 		}
3107 	}
3108 
3109 out:
3110 	channel->eventq_read_ptr = read_ptr;
3111 	return spent;
3112 }
3113 
3114 static void efx_ef10_ev_read_ack(struct efx_channel *channel)
3115 {
3116 	struct efx_nic *efx = channel->efx;
3117 	efx_dword_t rptr;
3118 
3119 	if (EFX_EF10_WORKAROUND_35388(efx)) {
3120 		BUILD_BUG_ON(EFX_MIN_EVQ_SIZE <
3121 			     (1 << ERF_DD_EVQ_IND_RPTR_WIDTH));
3122 		BUILD_BUG_ON(EFX_MAX_EVQ_SIZE >
3123 			     (1 << 2 * ERF_DD_EVQ_IND_RPTR_WIDTH));
3124 
3125 		EFX_POPULATE_DWORD_2(rptr, ERF_DD_EVQ_IND_RPTR_FLAGS,
3126 				     EFE_DD_EVQ_IND_RPTR_FLAGS_HIGH,
3127 				     ERF_DD_EVQ_IND_RPTR,
3128 				     (channel->eventq_read_ptr &
3129 				      channel->eventq_mask) >>
3130 				     ERF_DD_EVQ_IND_RPTR_WIDTH);
3131 		efx_writed_page(efx, &rptr, ER_DD_EVQ_INDIRECT,
3132 				channel->channel);
3133 		EFX_POPULATE_DWORD_2(rptr, ERF_DD_EVQ_IND_RPTR_FLAGS,
3134 				     EFE_DD_EVQ_IND_RPTR_FLAGS_LOW,
3135 				     ERF_DD_EVQ_IND_RPTR,
3136 				     channel->eventq_read_ptr &
3137 				     ((1 << ERF_DD_EVQ_IND_RPTR_WIDTH) - 1));
3138 		efx_writed_page(efx, &rptr, ER_DD_EVQ_INDIRECT,
3139 				channel->channel);
3140 	} else {
3141 		EFX_POPULATE_DWORD_1(rptr, ERF_DZ_EVQ_RPTR,
3142 				     channel->eventq_read_ptr &
3143 				     channel->eventq_mask);
3144 		efx_writed_page(efx, &rptr, ER_DZ_EVQ_RPTR, channel->channel);
3145 	}
3146 }
3147 
3148 static void efx_ef10_ev_test_generate(struct efx_channel *channel)
3149 {
3150 	MCDI_DECLARE_BUF(inbuf, MC_CMD_DRIVER_EVENT_IN_LEN);
3151 	struct efx_nic *efx = channel->efx;
3152 	efx_qword_t event;
3153 	int rc;
3154 
3155 	EFX_POPULATE_QWORD_2(event,
3156 			     ESF_DZ_EV_CODE, EFX_EF10_DRVGEN_EV,
3157 			     ESF_DZ_EV_DATA, EFX_EF10_TEST);
3158 
3159 	MCDI_SET_DWORD(inbuf, DRIVER_EVENT_IN_EVQ, channel->channel);
3160 
3161 	/* MCDI_SET_QWORD is not appropriate here since EFX_POPULATE_* has
3162 	 * already swapped the data to little-endian order.
3163 	 */
3164 	memcpy(MCDI_PTR(inbuf, DRIVER_EVENT_IN_DATA), &event.u64[0],
3165 	       sizeof(efx_qword_t));
3166 
3167 	rc = efx_mcdi_rpc(efx, MC_CMD_DRIVER_EVENT, inbuf, sizeof(inbuf),
3168 			  NULL, 0, NULL);
3169 	if (rc != 0)
3170 		goto fail;
3171 
3172 	return;
3173 
3174 fail:
3175 	WARN_ON(true);
3176 	netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
3177 }
3178 
3179 static void efx_ef10_prepare_flr(struct efx_nic *efx)
3180 {
3181 	atomic_set(&efx->active_queues, 0);
3182 }
3183 
3184 static int efx_ef10_vport_set_mac_address(struct efx_nic *efx)
3185 {
3186 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
3187 	u8 mac_old[ETH_ALEN];
3188 	int rc, rc2;
3189 
3190 	/* Only reconfigure a PF-created vport */
3191 	if (is_zero_ether_addr(nic_data->vport_mac))
3192 		return 0;
3193 
3194 	efx_device_detach_sync(efx);
3195 	efx_net_stop(efx->net_dev);
3196 	down_write(&efx->filter_sem);
3197 	efx_mcdi_filter_table_remove(efx);
3198 	up_write(&efx->filter_sem);
3199 
3200 	rc = efx_ef10_vadaptor_free(efx, efx->vport_id);
3201 	if (rc)
3202 		goto restore_filters;
3203 
3204 	ether_addr_copy(mac_old, nic_data->vport_mac);
3205 	rc = efx_ef10_vport_del_mac(efx, efx->vport_id,
3206 				    nic_data->vport_mac);
3207 	if (rc)
3208 		goto restore_vadaptor;
3209 
3210 	rc = efx_ef10_vport_add_mac(efx, efx->vport_id,
3211 				    efx->net_dev->dev_addr);
3212 	if (!rc) {
3213 		ether_addr_copy(nic_data->vport_mac, efx->net_dev->dev_addr);
3214 	} else {
3215 		rc2 = efx_ef10_vport_add_mac(efx, efx->vport_id, mac_old);
3216 		if (rc2) {
3217 			/* Failed to add original MAC, so clear vport_mac */
3218 			eth_zero_addr(nic_data->vport_mac);
3219 			goto reset_nic;
3220 		}
3221 	}
3222 
3223 restore_vadaptor:
3224 	rc2 = efx_ef10_vadaptor_alloc(efx, efx->vport_id);
3225 	if (rc2)
3226 		goto reset_nic;
3227 restore_filters:
3228 	down_write(&efx->filter_sem);
3229 	rc2 = efx_ef10_filter_table_probe(efx);
3230 	up_write(&efx->filter_sem);
3231 	if (rc2)
3232 		goto reset_nic;
3233 
3234 	rc2 = efx_net_open(efx->net_dev);
3235 	if (rc2)
3236 		goto reset_nic;
3237 
3238 	efx_device_attach_if_not_resetting(efx);
3239 
3240 	return rc;
3241 
3242 reset_nic:
3243 	netif_err(efx, drv, efx->net_dev,
3244 		  "Failed to restore when changing MAC address - scheduling reset\n");
3245 	efx_schedule_reset(efx, RESET_TYPE_DATAPATH);
3246 
3247 	return rc ? rc : rc2;
3248 }
3249 
3250 static int efx_ef10_set_mac_address(struct efx_nic *efx)
3251 {
3252 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VADAPTOR_SET_MAC_IN_LEN);
3253 	bool was_enabled = efx->port_enabled;
3254 	int rc;
3255 
3256 	efx_device_detach_sync(efx);
3257 	efx_net_stop(efx->net_dev);
3258 
3259 	mutex_lock(&efx->mac_lock);
3260 	down_write(&efx->filter_sem);
3261 	efx_mcdi_filter_table_remove(efx);
3262 
3263 	ether_addr_copy(MCDI_PTR(inbuf, VADAPTOR_SET_MAC_IN_MACADDR),
3264 			efx->net_dev->dev_addr);
3265 	MCDI_SET_DWORD(inbuf, VADAPTOR_SET_MAC_IN_UPSTREAM_PORT_ID,
3266 		       efx->vport_id);
3267 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_VADAPTOR_SET_MAC, inbuf,
3268 				sizeof(inbuf), NULL, 0, NULL);
3269 
3270 	efx_ef10_filter_table_probe(efx);
3271 	up_write(&efx->filter_sem);
3272 	mutex_unlock(&efx->mac_lock);
3273 
3274 	if (was_enabled)
3275 		efx_net_open(efx->net_dev);
3276 	efx_device_attach_if_not_resetting(efx);
3277 
3278 #ifdef CONFIG_SFC_SRIOV
3279 	if (efx->pci_dev->is_virtfn && efx->pci_dev->physfn) {
3280 		struct efx_ef10_nic_data *nic_data = efx->nic_data;
3281 		struct pci_dev *pci_dev_pf = efx->pci_dev->physfn;
3282 
3283 		if (rc == -EPERM) {
3284 			struct efx_nic *efx_pf;
3285 
3286 			/* Switch to PF and change MAC address on vport */
3287 			efx_pf = pci_get_drvdata(pci_dev_pf);
3288 
3289 			rc = efx_ef10_sriov_set_vf_mac(efx_pf,
3290 						       nic_data->vf_index,
3291 						       efx->net_dev->dev_addr);
3292 		} else if (!rc) {
3293 			struct efx_nic *efx_pf = pci_get_drvdata(pci_dev_pf);
3294 			struct efx_ef10_nic_data *nic_data = efx_pf->nic_data;
3295 			unsigned int i;
3296 
3297 			/* MAC address successfully changed by VF (with MAC
3298 			 * spoofing) so update the parent PF if possible.
3299 			 */
3300 			for (i = 0; i < efx_pf->vf_count; ++i) {
3301 				struct ef10_vf *vf = nic_data->vf + i;
3302 
3303 				if (vf->efx == efx) {
3304 					ether_addr_copy(vf->mac,
3305 							efx->net_dev->dev_addr);
3306 					return 0;
3307 				}
3308 			}
3309 		}
3310 	} else
3311 #endif
3312 	if (rc == -EPERM) {
3313 		netif_err(efx, drv, efx->net_dev,
3314 			  "Cannot change MAC address; use sfboot to enable"
3315 			  " mac-spoofing on this interface\n");
3316 	} else if (rc == -ENOSYS && !efx_ef10_is_vf(efx)) {
3317 		/* If the active MCFW does not support MC_CMD_VADAPTOR_SET_MAC
3318 		 * fall-back to the method of changing the MAC address on the
3319 		 * vport.  This only applies to PFs because such versions of
3320 		 * MCFW do not support VFs.
3321 		 */
3322 		rc = efx_ef10_vport_set_mac_address(efx);
3323 	} else if (rc) {
3324 		efx_mcdi_display_error(efx, MC_CMD_VADAPTOR_SET_MAC,
3325 				       sizeof(inbuf), NULL, 0, rc);
3326 	}
3327 
3328 	return rc;
3329 }
3330 
3331 static int efx_ef10_mac_reconfigure(struct efx_nic *efx, bool mtu_only)
3332 {
3333 	WARN_ON(!mutex_is_locked(&efx->mac_lock));
3334 
3335 	efx_mcdi_filter_sync_rx_mode(efx);
3336 
3337 	if (mtu_only && efx_has_cap(efx, SET_MAC_ENHANCED))
3338 		return efx_mcdi_set_mtu(efx);
3339 	return efx_mcdi_set_mac(efx);
3340 }
3341 
3342 static int efx_ef10_start_bist(struct efx_nic *efx, u32 bist_type)
3343 {
3344 	MCDI_DECLARE_BUF(inbuf, MC_CMD_START_BIST_IN_LEN);
3345 
3346 	MCDI_SET_DWORD(inbuf, START_BIST_IN_TYPE, bist_type);
3347 	return efx_mcdi_rpc(efx, MC_CMD_START_BIST, inbuf, sizeof(inbuf),
3348 			    NULL, 0, NULL);
3349 }
3350 
3351 /* MC BISTs follow a different poll mechanism to phy BISTs.
3352  * The BIST is done in the poll handler on the MC, and the MCDI command
3353  * will block until the BIST is done.
3354  */
3355 static int efx_ef10_poll_bist(struct efx_nic *efx)
3356 {
3357 	int rc;
3358 	MCDI_DECLARE_BUF(outbuf, MC_CMD_POLL_BIST_OUT_LEN);
3359 	size_t outlen;
3360 	u32 result;
3361 
3362 	rc = efx_mcdi_rpc(efx, MC_CMD_POLL_BIST, NULL, 0,
3363 			   outbuf, sizeof(outbuf), &outlen);
3364 	if (rc != 0)
3365 		return rc;
3366 
3367 	if (outlen < MC_CMD_POLL_BIST_OUT_LEN)
3368 		return -EIO;
3369 
3370 	result = MCDI_DWORD(outbuf, POLL_BIST_OUT_RESULT);
3371 	switch (result) {
3372 	case MC_CMD_POLL_BIST_PASSED:
3373 		netif_dbg(efx, hw, efx->net_dev, "BIST passed.\n");
3374 		return 0;
3375 	case MC_CMD_POLL_BIST_TIMEOUT:
3376 		netif_err(efx, hw, efx->net_dev, "BIST timed out\n");
3377 		return -EIO;
3378 	case MC_CMD_POLL_BIST_FAILED:
3379 		netif_err(efx, hw, efx->net_dev, "BIST failed.\n");
3380 		return -EIO;
3381 	default:
3382 		netif_err(efx, hw, efx->net_dev,
3383 			  "BIST returned unknown result %u", result);
3384 		return -EIO;
3385 	}
3386 }
3387 
3388 static int efx_ef10_run_bist(struct efx_nic *efx, u32 bist_type)
3389 {
3390 	int rc;
3391 
3392 	netif_dbg(efx, drv, efx->net_dev, "starting BIST type %u\n", bist_type);
3393 
3394 	rc = efx_ef10_start_bist(efx, bist_type);
3395 	if (rc != 0)
3396 		return rc;
3397 
3398 	return efx_ef10_poll_bist(efx);
3399 }
3400 
3401 static int
3402 efx_ef10_test_chip(struct efx_nic *efx, struct efx_self_tests *tests)
3403 {
3404 	int rc, rc2;
3405 
3406 	efx_reset_down(efx, RESET_TYPE_WORLD);
3407 
3408 	rc = efx_mcdi_rpc(efx, MC_CMD_ENABLE_OFFLINE_BIST,
3409 			  NULL, 0, NULL, 0, NULL);
3410 	if (rc != 0)
3411 		goto out;
3412 
3413 	tests->memory = efx_ef10_run_bist(efx, MC_CMD_MC_MEM_BIST) ? -1 : 1;
3414 	tests->registers = efx_ef10_run_bist(efx, MC_CMD_REG_BIST) ? -1 : 1;
3415 
3416 	rc = efx_mcdi_reset(efx, RESET_TYPE_WORLD);
3417 
3418 out:
3419 	if (rc == -EPERM)
3420 		rc = 0;
3421 	rc2 = efx_reset_up(efx, RESET_TYPE_WORLD, rc == 0);
3422 	return rc ? rc : rc2;
3423 }
3424 
3425 #ifdef CONFIG_SFC_MTD
3426 
3427 struct efx_ef10_nvram_type_info {
3428 	u16 type, type_mask;
3429 	u8 port;
3430 	const char *name;
3431 };
3432 
3433 static const struct efx_ef10_nvram_type_info efx_ef10_nvram_types[] = {
3434 	{ NVRAM_PARTITION_TYPE_MC_FIRMWARE,	   0,    0, "sfc_mcfw" },
3435 	{ NVRAM_PARTITION_TYPE_MC_FIRMWARE_BACKUP, 0,    0, "sfc_mcfw_backup" },
3436 	{ NVRAM_PARTITION_TYPE_EXPANSION_ROM,	   0,    0, "sfc_exp_rom" },
3437 	{ NVRAM_PARTITION_TYPE_STATIC_CONFIG,	   0,    0, "sfc_static_cfg" },
3438 	{ NVRAM_PARTITION_TYPE_DYNAMIC_CONFIG,	   0,    0, "sfc_dynamic_cfg" },
3439 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT0, 0,   0, "sfc_exp_rom_cfg" },
3440 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT1, 0,   1, "sfc_exp_rom_cfg" },
3441 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT2, 0,   2, "sfc_exp_rom_cfg" },
3442 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT3, 0,   3, "sfc_exp_rom_cfg" },
3443 	{ NVRAM_PARTITION_TYPE_LICENSE,		   0,    0, "sfc_license" },
3444 	{ NVRAM_PARTITION_TYPE_PHY_MIN,		   0xff, 0, "sfc_phy_fw" },
3445 	{ NVRAM_PARTITION_TYPE_MUM_FIRMWARE,	   0,    0, "sfc_mumfw" },
3446 	{ NVRAM_PARTITION_TYPE_EXPANSION_UEFI,	   0,    0, "sfc_uefi" },
3447 	{ NVRAM_PARTITION_TYPE_DYNCONFIG_DEFAULTS, 0,    0, "sfc_dynamic_cfg_dflt" },
3448 	{ NVRAM_PARTITION_TYPE_ROMCONFIG_DEFAULTS, 0,    0, "sfc_exp_rom_cfg_dflt" },
3449 	{ NVRAM_PARTITION_TYPE_STATUS,		   0,    0, "sfc_status" },
3450 	{ NVRAM_PARTITION_TYPE_BUNDLE,		   0,    0, "sfc_bundle" },
3451 	{ NVRAM_PARTITION_TYPE_BUNDLE_METADATA,	   0,    0, "sfc_bundle_metadata" },
3452 };
3453 #define EF10_NVRAM_PARTITION_COUNT	ARRAY_SIZE(efx_ef10_nvram_types)
3454 
3455 static int efx_ef10_mtd_probe_partition(struct efx_nic *efx,
3456 					struct efx_mcdi_mtd_partition *part,
3457 					unsigned int type,
3458 					unsigned long *found)
3459 {
3460 	MCDI_DECLARE_BUF(inbuf, MC_CMD_NVRAM_METADATA_IN_LEN);
3461 	MCDI_DECLARE_BUF(outbuf, MC_CMD_NVRAM_METADATA_OUT_LENMAX);
3462 	const struct efx_ef10_nvram_type_info *info;
3463 	size_t size, erase_size, outlen;
3464 	int type_idx = 0;
3465 	bool protected;
3466 	int rc;
3467 
3468 	for (type_idx = 0; ; type_idx++) {
3469 		if (type_idx == EF10_NVRAM_PARTITION_COUNT)
3470 			return -ENODEV;
3471 		info = efx_ef10_nvram_types + type_idx;
3472 		if ((type & ~info->type_mask) == info->type)
3473 			break;
3474 	}
3475 	if (info->port != efx_port_num(efx))
3476 		return -ENODEV;
3477 
3478 	rc = efx_mcdi_nvram_info(efx, type, &size, &erase_size, &protected);
3479 	if (rc)
3480 		return rc;
3481 	if (protected &&
3482 	    (type != NVRAM_PARTITION_TYPE_DYNCONFIG_DEFAULTS &&
3483 	     type != NVRAM_PARTITION_TYPE_ROMCONFIG_DEFAULTS))
3484 		/* Hide protected partitions that don't provide defaults. */
3485 		return -ENODEV;
3486 
3487 	if (protected)
3488 		/* Protected partitions are read only. */
3489 		erase_size = 0;
3490 
3491 	/* If we've already exposed a partition of this type, hide this
3492 	 * duplicate.  All operations on MTDs are keyed by the type anyway,
3493 	 * so we can't act on the duplicate.
3494 	 */
3495 	if (__test_and_set_bit(type_idx, found))
3496 		return -EEXIST;
3497 
3498 	part->nvram_type = type;
3499 
3500 	MCDI_SET_DWORD(inbuf, NVRAM_METADATA_IN_TYPE, type);
3501 	rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_METADATA, inbuf, sizeof(inbuf),
3502 			  outbuf, sizeof(outbuf), &outlen);
3503 	if (rc)
3504 		return rc;
3505 	if (outlen < MC_CMD_NVRAM_METADATA_OUT_LENMIN)
3506 		return -EIO;
3507 	if (MCDI_DWORD(outbuf, NVRAM_METADATA_OUT_FLAGS) &
3508 	    (1 << MC_CMD_NVRAM_METADATA_OUT_SUBTYPE_VALID_LBN))
3509 		part->fw_subtype = MCDI_DWORD(outbuf,
3510 					      NVRAM_METADATA_OUT_SUBTYPE);
3511 
3512 	part->common.dev_type_name = "EF10 NVRAM manager";
3513 	part->common.type_name = info->name;
3514 
3515 	part->common.mtd.type = MTD_NORFLASH;
3516 	part->common.mtd.flags = MTD_CAP_NORFLASH;
3517 	part->common.mtd.size = size;
3518 	part->common.mtd.erasesize = erase_size;
3519 	/* sfc_status is read-only */
3520 	if (!erase_size)
3521 		part->common.mtd.flags |= MTD_NO_ERASE;
3522 
3523 	return 0;
3524 }
3525 
3526 static int efx_ef10_mtd_probe(struct efx_nic *efx)
3527 {
3528 	MCDI_DECLARE_BUF(outbuf, MC_CMD_NVRAM_PARTITIONS_OUT_LENMAX);
3529 	DECLARE_BITMAP(found, EF10_NVRAM_PARTITION_COUNT) = { 0 };
3530 	struct efx_mcdi_mtd_partition *parts;
3531 	size_t outlen, n_parts_total, i, n_parts;
3532 	unsigned int type;
3533 	int rc;
3534 
3535 	ASSERT_RTNL();
3536 
3537 	BUILD_BUG_ON(MC_CMD_NVRAM_PARTITIONS_IN_LEN != 0);
3538 	rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_PARTITIONS, NULL, 0,
3539 			  outbuf, sizeof(outbuf), &outlen);
3540 	if (rc)
3541 		return rc;
3542 	if (outlen < MC_CMD_NVRAM_PARTITIONS_OUT_LENMIN)
3543 		return -EIO;
3544 
3545 	n_parts_total = MCDI_DWORD(outbuf, NVRAM_PARTITIONS_OUT_NUM_PARTITIONS);
3546 	if (n_parts_total >
3547 	    MCDI_VAR_ARRAY_LEN(outlen, NVRAM_PARTITIONS_OUT_TYPE_ID))
3548 		return -EIO;
3549 
3550 	parts = kcalloc(n_parts_total, sizeof(*parts), GFP_KERNEL);
3551 	if (!parts)
3552 		return -ENOMEM;
3553 
3554 	n_parts = 0;
3555 	for (i = 0; i < n_parts_total; i++) {
3556 		type = MCDI_ARRAY_DWORD(outbuf, NVRAM_PARTITIONS_OUT_TYPE_ID,
3557 					i);
3558 		rc = efx_ef10_mtd_probe_partition(efx, &parts[n_parts], type,
3559 						  found);
3560 		if (rc == -EEXIST || rc == -ENODEV)
3561 			continue;
3562 		if (rc)
3563 			goto fail;
3564 		n_parts++;
3565 	}
3566 
3567 	rc = efx_mtd_add(efx, &parts[0].common, n_parts, sizeof(*parts));
3568 fail:
3569 	if (rc)
3570 		kfree(parts);
3571 	return rc;
3572 }
3573 
3574 #endif /* CONFIG_SFC_MTD */
3575 
3576 static void efx_ef10_ptp_write_host_time(struct efx_nic *efx, u32 host_time)
3577 {
3578 	_efx_writed(efx, cpu_to_le32(host_time), ER_DZ_MC_DB_LWRD);
3579 }
3580 
3581 static void efx_ef10_ptp_write_host_time_vf(struct efx_nic *efx,
3582 					    u32 host_time) {}
3583 
3584 static int efx_ef10_rx_enable_timestamping(struct efx_channel *channel,
3585 					   bool temp)
3586 {
3587 	MCDI_DECLARE_BUF(inbuf, MC_CMD_PTP_IN_TIME_EVENT_SUBSCRIBE_LEN);
3588 	int rc;
3589 
3590 	if (channel->sync_events_state == SYNC_EVENTS_REQUESTED ||
3591 	    channel->sync_events_state == SYNC_EVENTS_VALID ||
3592 	    (temp && channel->sync_events_state == SYNC_EVENTS_DISABLED))
3593 		return 0;
3594 	channel->sync_events_state = SYNC_EVENTS_REQUESTED;
3595 
3596 	MCDI_SET_DWORD(inbuf, PTP_IN_OP, MC_CMD_PTP_OP_TIME_EVENT_SUBSCRIBE);
3597 	MCDI_SET_DWORD(inbuf, PTP_IN_PERIPH_ID, 0);
3598 	MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_SUBSCRIBE_QUEUE,
3599 		       channel->channel);
3600 
3601 	rc = efx_mcdi_rpc(channel->efx, MC_CMD_PTP,
3602 			  inbuf, sizeof(inbuf), NULL, 0, NULL);
3603 
3604 	if (rc != 0)
3605 		channel->sync_events_state = temp ? SYNC_EVENTS_QUIESCENT :
3606 						    SYNC_EVENTS_DISABLED;
3607 
3608 	return rc;
3609 }
3610 
3611 static int efx_ef10_rx_disable_timestamping(struct efx_channel *channel,
3612 					    bool temp)
3613 {
3614 	MCDI_DECLARE_BUF(inbuf, MC_CMD_PTP_IN_TIME_EVENT_UNSUBSCRIBE_LEN);
3615 	int rc;
3616 
3617 	if (channel->sync_events_state == SYNC_EVENTS_DISABLED ||
3618 	    (temp && channel->sync_events_state == SYNC_EVENTS_QUIESCENT))
3619 		return 0;
3620 	if (channel->sync_events_state == SYNC_EVENTS_QUIESCENT) {
3621 		channel->sync_events_state = SYNC_EVENTS_DISABLED;
3622 		return 0;
3623 	}
3624 	channel->sync_events_state = temp ? SYNC_EVENTS_QUIESCENT :
3625 					    SYNC_EVENTS_DISABLED;
3626 
3627 	MCDI_SET_DWORD(inbuf, PTP_IN_OP, MC_CMD_PTP_OP_TIME_EVENT_UNSUBSCRIBE);
3628 	MCDI_SET_DWORD(inbuf, PTP_IN_PERIPH_ID, 0);
3629 	MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_UNSUBSCRIBE_CONTROL,
3630 		       MC_CMD_PTP_IN_TIME_EVENT_UNSUBSCRIBE_SINGLE);
3631 	MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_UNSUBSCRIBE_QUEUE,
3632 		       channel->channel);
3633 
3634 	rc = efx_mcdi_rpc(channel->efx, MC_CMD_PTP,
3635 			  inbuf, sizeof(inbuf), NULL, 0, NULL);
3636 
3637 	return rc;
3638 }
3639 
3640 static int efx_ef10_ptp_set_ts_sync_events(struct efx_nic *efx, bool en,
3641 					   bool temp)
3642 {
3643 	int (*set)(struct efx_channel *channel, bool temp);
3644 	struct efx_channel *channel;
3645 
3646 	set = en ?
3647 	      efx_ef10_rx_enable_timestamping :
3648 	      efx_ef10_rx_disable_timestamping;
3649 
3650 	channel = efx_ptp_channel(efx);
3651 	if (channel) {
3652 		int rc = set(channel, temp);
3653 		if (en && rc != 0) {
3654 			efx_ef10_ptp_set_ts_sync_events(efx, false, temp);
3655 			return rc;
3656 		}
3657 	}
3658 
3659 	return 0;
3660 }
3661 
3662 static int efx_ef10_ptp_set_ts_config_vf(struct efx_nic *efx,
3663 					 struct hwtstamp_config *init)
3664 {
3665 	return -EOPNOTSUPP;
3666 }
3667 
3668 static int efx_ef10_ptp_set_ts_config(struct efx_nic *efx,
3669 				      struct hwtstamp_config *init)
3670 {
3671 	int rc;
3672 
3673 	switch (init->rx_filter) {
3674 	case HWTSTAMP_FILTER_NONE:
3675 		efx_ef10_ptp_set_ts_sync_events(efx, false, false);
3676 		/* if TX timestamping is still requested then leave PTP on */
3677 		return efx_ptp_change_mode(efx,
3678 					   init->tx_type != HWTSTAMP_TX_OFF, 0);
3679 	case HWTSTAMP_FILTER_ALL:
3680 	case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
3681 	case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
3682 	case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
3683 	case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
3684 	case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
3685 	case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
3686 	case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
3687 	case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
3688 	case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
3689 	case HWTSTAMP_FILTER_PTP_V2_EVENT:
3690 	case HWTSTAMP_FILTER_PTP_V2_SYNC:
3691 	case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
3692 	case HWTSTAMP_FILTER_NTP_ALL:
3693 		init->rx_filter = HWTSTAMP_FILTER_ALL;
3694 		rc = efx_ptp_change_mode(efx, true, 0);
3695 		if (!rc)
3696 			rc = efx_ef10_ptp_set_ts_sync_events(efx, true, false);
3697 		if (rc)
3698 			efx_ptp_change_mode(efx, false, 0);
3699 		return rc;
3700 	default:
3701 		return -ERANGE;
3702 	}
3703 }
3704 
3705 static int efx_ef10_get_phys_port_id(struct efx_nic *efx,
3706 				     struct netdev_phys_item_id *ppid)
3707 {
3708 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
3709 
3710 	if (!is_valid_ether_addr(nic_data->port_id))
3711 		return -EOPNOTSUPP;
3712 
3713 	ppid->id_len = ETH_ALEN;
3714 	memcpy(ppid->id, nic_data->port_id, ppid->id_len);
3715 
3716 	return 0;
3717 }
3718 
3719 static int efx_ef10_vlan_rx_add_vid(struct efx_nic *efx, __be16 proto, u16 vid)
3720 {
3721 	if (proto != htons(ETH_P_8021Q))
3722 		return -EINVAL;
3723 
3724 	return efx_ef10_add_vlan(efx, vid);
3725 }
3726 
3727 static int efx_ef10_vlan_rx_kill_vid(struct efx_nic *efx, __be16 proto, u16 vid)
3728 {
3729 	if (proto != htons(ETH_P_8021Q))
3730 		return -EINVAL;
3731 
3732 	return efx_ef10_del_vlan(efx, vid);
3733 }
3734 
3735 /* We rely on the MCDI wiping out our TX rings if it made any changes to the
3736  * ports table, ensuring that any TSO descriptors that were made on a now-
3737  * removed tunnel port will be blown away and won't break things when we try
3738  * to transmit them using the new ports table.
3739  */
3740 static int efx_ef10_set_udp_tnl_ports(struct efx_nic *efx, bool unloading)
3741 {
3742 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
3743 	MCDI_DECLARE_BUF(inbuf, MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_IN_LENMAX);
3744 	MCDI_DECLARE_BUF(outbuf, MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_OUT_LEN);
3745 	bool will_reset = false;
3746 	size_t num_entries = 0;
3747 	size_t inlen, outlen;
3748 	size_t i;
3749 	int rc;
3750 	efx_dword_t flags_and_num_entries;
3751 
3752 	WARN_ON(!mutex_is_locked(&nic_data->udp_tunnels_lock));
3753 
3754 	nic_data->udp_tunnels_dirty = false;
3755 
3756 	if (!(nic_data->datapath_caps &
3757 	    (1 << MC_CMD_GET_CAPABILITIES_OUT_VXLAN_NVGRE_LBN))) {
3758 		efx_device_attach_if_not_resetting(efx);
3759 		return 0;
3760 	}
3761 
3762 	BUILD_BUG_ON(ARRAY_SIZE(nic_data->udp_tunnels) >
3763 		     MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_IN_ENTRIES_MAXNUM);
3764 
3765 	for (i = 0; i < ARRAY_SIZE(nic_data->udp_tunnels); ++i) {
3766 		if (nic_data->udp_tunnels[i].type !=
3767 		    TUNNEL_ENCAP_UDP_PORT_ENTRY_INVALID) {
3768 			efx_dword_t entry;
3769 
3770 			EFX_POPULATE_DWORD_2(entry,
3771 				TUNNEL_ENCAP_UDP_PORT_ENTRY_UDP_PORT,
3772 					ntohs(nic_data->udp_tunnels[i].port),
3773 				TUNNEL_ENCAP_UDP_PORT_ENTRY_PROTOCOL,
3774 					nic_data->udp_tunnels[i].type);
3775 			*_MCDI_ARRAY_DWORD(inbuf,
3776 				SET_TUNNEL_ENCAP_UDP_PORTS_IN_ENTRIES,
3777 				num_entries++) = entry;
3778 		}
3779 	}
3780 
3781 	BUILD_BUG_ON((MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_IN_NUM_ENTRIES_OFST -
3782 		      MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_IN_FLAGS_OFST) * 8 !=
3783 		     EFX_WORD_1_LBN);
3784 	BUILD_BUG_ON(MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_IN_NUM_ENTRIES_LEN * 8 !=
3785 		     EFX_WORD_1_WIDTH);
3786 	EFX_POPULATE_DWORD_2(flags_and_num_entries,
3787 			     MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_IN_UNLOADING,
3788 				!!unloading,
3789 			     EFX_WORD_1, num_entries);
3790 	*_MCDI_DWORD(inbuf, SET_TUNNEL_ENCAP_UDP_PORTS_IN_FLAGS) =
3791 		flags_and_num_entries;
3792 
3793 	inlen = MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_IN_LEN(num_entries);
3794 
3795 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS,
3796 				inbuf, inlen, outbuf, sizeof(outbuf), &outlen);
3797 	if (rc == -EIO) {
3798 		/* Most likely the MC rebooted due to another function also
3799 		 * setting its tunnel port list. Mark the tunnel port list as
3800 		 * dirty, so it will be pushed upon coming up from the reboot.
3801 		 */
3802 		nic_data->udp_tunnels_dirty = true;
3803 		return 0;
3804 	}
3805 
3806 	if (rc) {
3807 		/* expected not available on unprivileged functions */
3808 		if (rc != -EPERM)
3809 			netif_warn(efx, drv, efx->net_dev,
3810 				   "Unable to set UDP tunnel ports; rc=%d.\n", rc);
3811 	} else if (MCDI_DWORD(outbuf, SET_TUNNEL_ENCAP_UDP_PORTS_OUT_FLAGS) &
3812 		   (1 << MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_OUT_RESETTING_LBN)) {
3813 		netif_info(efx, drv, efx->net_dev,
3814 			   "Rebooting MC due to UDP tunnel port list change\n");
3815 		will_reset = true;
3816 		if (unloading)
3817 			/* Delay for the MC reset to complete. This will make
3818 			 * unloading other functions a bit smoother. This is a
3819 			 * race, but the other unload will work whichever way
3820 			 * it goes, this just avoids an unnecessary error
3821 			 * message.
3822 			 */
3823 			msleep(100);
3824 	}
3825 	if (!will_reset && !unloading) {
3826 		/* The caller will have detached, relying on the MC reset to
3827 		 * trigger a re-attach.  Since there won't be an MC reset, we
3828 		 * have to do the attach ourselves.
3829 		 */
3830 		efx_device_attach_if_not_resetting(efx);
3831 	}
3832 
3833 	return rc;
3834 }
3835 
3836 static int efx_ef10_udp_tnl_push_ports(struct efx_nic *efx)
3837 {
3838 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
3839 	int rc = 0;
3840 
3841 	mutex_lock(&nic_data->udp_tunnels_lock);
3842 	if (nic_data->udp_tunnels_dirty) {
3843 		/* Make sure all TX are stopped while we modify the table, else
3844 		 * we might race against an efx_features_check().
3845 		 */
3846 		efx_device_detach_sync(efx);
3847 		rc = efx_ef10_set_udp_tnl_ports(efx, false);
3848 	}
3849 	mutex_unlock(&nic_data->udp_tunnels_lock);
3850 	return rc;
3851 }
3852 
3853 static int efx_ef10_udp_tnl_set_port(struct net_device *dev,
3854 				     unsigned int table, unsigned int entry,
3855 				     struct udp_tunnel_info *ti)
3856 {
3857 	struct efx_nic *efx = netdev_priv(dev);
3858 	struct efx_ef10_nic_data *nic_data;
3859 	int efx_tunnel_type, rc;
3860 
3861 	if (ti->type == UDP_TUNNEL_TYPE_VXLAN)
3862 		efx_tunnel_type = TUNNEL_ENCAP_UDP_PORT_ENTRY_VXLAN;
3863 	else
3864 		efx_tunnel_type = TUNNEL_ENCAP_UDP_PORT_ENTRY_GENEVE;
3865 
3866 	nic_data = efx->nic_data;
3867 	if (!(nic_data->datapath_caps &
3868 	      (1 << MC_CMD_GET_CAPABILITIES_OUT_VXLAN_NVGRE_LBN)))
3869 		return -EOPNOTSUPP;
3870 
3871 	mutex_lock(&nic_data->udp_tunnels_lock);
3872 	/* Make sure all TX are stopped while we add to the table, else we
3873 	 * might race against an efx_features_check().
3874 	 */
3875 	efx_device_detach_sync(efx);
3876 	nic_data->udp_tunnels[entry].type = efx_tunnel_type;
3877 	nic_data->udp_tunnels[entry].port = ti->port;
3878 	rc = efx_ef10_set_udp_tnl_ports(efx, false);
3879 	mutex_unlock(&nic_data->udp_tunnels_lock);
3880 
3881 	return rc;
3882 }
3883 
3884 /* Called under the TX lock with the TX queue running, hence no-one can be
3885  * in the middle of updating the UDP tunnels table.  However, they could
3886  * have tried and failed the MCDI, in which case they'll have set the dirty
3887  * flag before dropping their locks.
3888  */
3889 static bool efx_ef10_udp_tnl_has_port(struct efx_nic *efx, __be16 port)
3890 {
3891 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
3892 	size_t i;
3893 
3894 	if (!(nic_data->datapath_caps &
3895 	      (1 << MC_CMD_GET_CAPABILITIES_OUT_VXLAN_NVGRE_LBN)))
3896 		return false;
3897 
3898 	if (nic_data->udp_tunnels_dirty)
3899 		/* SW table may not match HW state, so just assume we can't
3900 		 * use any UDP tunnel offloads.
3901 		 */
3902 		return false;
3903 
3904 	for (i = 0; i < ARRAY_SIZE(nic_data->udp_tunnels); ++i)
3905 		if (nic_data->udp_tunnels[i].type !=
3906 		    TUNNEL_ENCAP_UDP_PORT_ENTRY_INVALID &&
3907 		    nic_data->udp_tunnels[i].port == port)
3908 			return true;
3909 
3910 	return false;
3911 }
3912 
3913 static int efx_ef10_udp_tnl_unset_port(struct net_device *dev,
3914 				       unsigned int table, unsigned int entry,
3915 				       struct udp_tunnel_info *ti)
3916 {
3917 	struct efx_nic *efx = netdev_priv(dev);
3918 	struct efx_ef10_nic_data *nic_data;
3919 	int rc;
3920 
3921 	nic_data = efx->nic_data;
3922 
3923 	mutex_lock(&nic_data->udp_tunnels_lock);
3924 	/* Make sure all TX are stopped while we remove from the table, else we
3925 	 * might race against an efx_features_check().
3926 	 */
3927 	efx_device_detach_sync(efx);
3928 	nic_data->udp_tunnels[entry].type = TUNNEL_ENCAP_UDP_PORT_ENTRY_INVALID;
3929 	nic_data->udp_tunnels[entry].port = 0;
3930 	rc = efx_ef10_set_udp_tnl_ports(efx, false);
3931 	mutex_unlock(&nic_data->udp_tunnels_lock);
3932 
3933 	return rc;
3934 }
3935 
3936 static const struct udp_tunnel_nic_info efx_ef10_udp_tunnels = {
3937 	.set_port	= efx_ef10_udp_tnl_set_port,
3938 	.unset_port	= efx_ef10_udp_tnl_unset_port,
3939 	.flags          = UDP_TUNNEL_NIC_INFO_MAY_SLEEP,
3940 	.tables         = {
3941 		{
3942 			.n_entries = 16,
3943 			.tunnel_types = UDP_TUNNEL_TYPE_VXLAN |
3944 					UDP_TUNNEL_TYPE_GENEVE,
3945 		},
3946 	},
3947 };
3948 
3949 /* EF10 may have multiple datapath firmware variants within a
3950  * single version.  Report which variants are running.
3951  */
3952 static size_t efx_ef10_print_additional_fwver(struct efx_nic *efx, char *buf,
3953 					      size_t len)
3954 {
3955 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
3956 
3957 	return scnprintf(buf, len, " rx%x tx%x",
3958 			 nic_data->rx_dpcpu_fw_id,
3959 			 nic_data->tx_dpcpu_fw_id);
3960 }
3961 
3962 static unsigned int ef10_check_caps(const struct efx_nic *efx,
3963 				    u8 flag,
3964 				    u32 offset)
3965 {
3966 	const struct efx_ef10_nic_data *nic_data = efx->nic_data;
3967 
3968 	switch (offset) {
3969 	case(MC_CMD_GET_CAPABILITIES_V4_OUT_FLAGS1_OFST):
3970 		return nic_data->datapath_caps & BIT_ULL(flag);
3971 	case(MC_CMD_GET_CAPABILITIES_V4_OUT_FLAGS2_OFST):
3972 		return nic_data->datapath_caps2 & BIT_ULL(flag);
3973 	default:
3974 		return 0;
3975 	}
3976 }
3977 
3978 #define EF10_OFFLOAD_FEATURES		\
3979 	(NETIF_F_IP_CSUM |		\
3980 	 NETIF_F_HW_VLAN_CTAG_FILTER |	\
3981 	 NETIF_F_IPV6_CSUM |		\
3982 	 NETIF_F_RXHASH |		\
3983 	 NETIF_F_NTUPLE)
3984 
3985 const struct efx_nic_type efx_hunt_a0_vf_nic_type = {
3986 	.is_vf = true,
3987 	.mem_bar = efx_ef10_vf_mem_bar,
3988 	.mem_map_size = efx_ef10_mem_map_size,
3989 	.probe = efx_ef10_probe_vf,
3990 	.remove = efx_ef10_remove,
3991 	.dimension_resources = efx_ef10_dimension_resources,
3992 	.init = efx_ef10_init_nic,
3993 	.fini = efx_ef10_fini_nic,
3994 	.map_reset_reason = efx_ef10_map_reset_reason,
3995 	.map_reset_flags = efx_ef10_map_reset_flags,
3996 	.reset = efx_ef10_reset,
3997 	.probe_port = efx_mcdi_port_probe,
3998 	.remove_port = efx_mcdi_port_remove,
3999 	.fini_dmaq = efx_fini_dmaq,
4000 	.prepare_flr = efx_ef10_prepare_flr,
4001 	.finish_flr = efx_port_dummy_op_void,
4002 	.describe_stats = efx_ef10_describe_stats,
4003 	.update_stats = efx_ef10_update_stats_vf,
4004 	.update_stats_atomic = efx_ef10_update_stats_atomic_vf,
4005 	.start_stats = efx_port_dummy_op_void,
4006 	.pull_stats = efx_port_dummy_op_void,
4007 	.stop_stats = efx_port_dummy_op_void,
4008 	.push_irq_moderation = efx_ef10_push_irq_moderation,
4009 	.reconfigure_mac = efx_ef10_mac_reconfigure,
4010 	.check_mac_fault = efx_mcdi_mac_check_fault,
4011 	.reconfigure_port = efx_mcdi_port_reconfigure,
4012 	.get_wol = efx_ef10_get_wol_vf,
4013 	.set_wol = efx_ef10_set_wol_vf,
4014 	.resume_wol = efx_port_dummy_op_void,
4015 	.mcdi_request = efx_ef10_mcdi_request,
4016 	.mcdi_poll_response = efx_ef10_mcdi_poll_response,
4017 	.mcdi_read_response = efx_ef10_mcdi_read_response,
4018 	.mcdi_poll_reboot = efx_ef10_mcdi_poll_reboot,
4019 	.mcdi_reboot_detected = efx_ef10_mcdi_reboot_detected,
4020 	.irq_enable_master = efx_port_dummy_op_void,
4021 	.irq_test_generate = efx_ef10_irq_test_generate,
4022 	.irq_disable_non_ev = efx_port_dummy_op_void,
4023 	.irq_handle_msi = efx_ef10_msi_interrupt,
4024 	.irq_handle_legacy = efx_ef10_legacy_interrupt,
4025 	.tx_probe = efx_ef10_tx_probe,
4026 	.tx_init = efx_ef10_tx_init,
4027 	.tx_remove = efx_mcdi_tx_remove,
4028 	.tx_write = efx_ef10_tx_write,
4029 	.tx_limit_len = efx_ef10_tx_limit_len,
4030 	.tx_enqueue = __efx_enqueue_skb,
4031 	.rx_push_rss_config = efx_mcdi_vf_rx_push_rss_config,
4032 	.rx_pull_rss_config = efx_mcdi_rx_pull_rss_config,
4033 	.rx_probe = efx_mcdi_rx_probe,
4034 	.rx_init = efx_mcdi_rx_init,
4035 	.rx_remove = efx_mcdi_rx_remove,
4036 	.rx_write = efx_ef10_rx_write,
4037 	.rx_defer_refill = efx_ef10_rx_defer_refill,
4038 	.rx_packet = __efx_rx_packet,
4039 	.ev_probe = efx_mcdi_ev_probe,
4040 	.ev_init = efx_ef10_ev_init,
4041 	.ev_fini = efx_mcdi_ev_fini,
4042 	.ev_remove = efx_mcdi_ev_remove,
4043 	.ev_process = efx_ef10_ev_process,
4044 	.ev_read_ack = efx_ef10_ev_read_ack,
4045 	.ev_test_generate = efx_ef10_ev_test_generate,
4046 	.filter_table_probe = efx_ef10_filter_table_probe,
4047 	.filter_table_restore = efx_mcdi_filter_table_restore,
4048 	.filter_table_remove = efx_mcdi_filter_table_remove,
4049 	.filter_update_rx_scatter = efx_mcdi_update_rx_scatter,
4050 	.filter_insert = efx_mcdi_filter_insert,
4051 	.filter_remove_safe = efx_mcdi_filter_remove_safe,
4052 	.filter_get_safe = efx_mcdi_filter_get_safe,
4053 	.filter_clear_rx = efx_mcdi_filter_clear_rx,
4054 	.filter_count_rx_used = efx_mcdi_filter_count_rx_used,
4055 	.filter_get_rx_id_limit = efx_mcdi_filter_get_rx_id_limit,
4056 	.filter_get_rx_ids = efx_mcdi_filter_get_rx_ids,
4057 #ifdef CONFIG_RFS_ACCEL
4058 	.filter_rfs_expire_one = efx_mcdi_filter_rfs_expire_one,
4059 #endif
4060 #ifdef CONFIG_SFC_MTD
4061 	.mtd_probe = efx_port_dummy_op_int,
4062 #endif
4063 	.ptp_write_host_time = efx_ef10_ptp_write_host_time_vf,
4064 	.ptp_set_ts_config = efx_ef10_ptp_set_ts_config_vf,
4065 	.vlan_rx_add_vid = efx_ef10_vlan_rx_add_vid,
4066 	.vlan_rx_kill_vid = efx_ef10_vlan_rx_kill_vid,
4067 #ifdef CONFIG_SFC_SRIOV
4068 	.vswitching_probe = efx_ef10_vswitching_probe_vf,
4069 	.vswitching_restore = efx_ef10_vswitching_restore_vf,
4070 	.vswitching_remove = efx_ef10_vswitching_remove_vf,
4071 #endif
4072 	.get_mac_address = efx_ef10_get_mac_address_vf,
4073 	.set_mac_address = efx_ef10_set_mac_address,
4074 
4075 	.get_phys_port_id = efx_ef10_get_phys_port_id,
4076 	.revision = EFX_REV_HUNT_A0,
4077 	.max_dma_mask = DMA_BIT_MASK(ESF_DZ_TX_KER_BUF_ADDR_WIDTH),
4078 	.rx_prefix_size = ES_DZ_RX_PREFIX_SIZE,
4079 	.rx_hash_offset = ES_DZ_RX_PREFIX_HASH_OFST,
4080 	.rx_ts_offset = ES_DZ_RX_PREFIX_TSTAMP_OFST,
4081 	.can_rx_scatter = true,
4082 	.always_rx_scatter = true,
4083 	.min_interrupt_mode = EFX_INT_MODE_MSIX,
4084 	.timer_period_max = 1 << ERF_DD_EVQ_IND_TIMER_VAL_WIDTH,
4085 	.offload_features = EF10_OFFLOAD_FEATURES,
4086 	.mcdi_max_ver = 2,
4087 	.max_rx_ip_filters = EFX_MCDI_FILTER_TBL_ROWS,
4088 	.hwtstamp_filters = 1 << HWTSTAMP_FILTER_NONE |
4089 			    1 << HWTSTAMP_FILTER_ALL,
4090 	.rx_hash_key_size = 40,
4091 	.check_caps = ef10_check_caps,
4092 	.print_additional_fwver = efx_ef10_print_additional_fwver,
4093 	.sensor_event = efx_mcdi_sensor_event,
4094 };
4095 
4096 const struct efx_nic_type efx_hunt_a0_nic_type = {
4097 	.is_vf = false,
4098 	.mem_bar = efx_ef10_pf_mem_bar,
4099 	.mem_map_size = efx_ef10_mem_map_size,
4100 	.probe = efx_ef10_probe_pf,
4101 	.remove = efx_ef10_remove,
4102 	.dimension_resources = efx_ef10_dimension_resources,
4103 	.init = efx_ef10_init_nic,
4104 	.fini = efx_ef10_fini_nic,
4105 	.map_reset_reason = efx_ef10_map_reset_reason,
4106 	.map_reset_flags = efx_ef10_map_reset_flags,
4107 	.reset = efx_ef10_reset,
4108 	.probe_port = efx_mcdi_port_probe,
4109 	.remove_port = efx_mcdi_port_remove,
4110 	.fini_dmaq = efx_fini_dmaq,
4111 	.prepare_flr = efx_ef10_prepare_flr,
4112 	.finish_flr = efx_port_dummy_op_void,
4113 	.describe_stats = efx_ef10_describe_stats,
4114 	.update_stats = efx_ef10_update_stats_pf,
4115 	.start_stats = efx_mcdi_mac_start_stats,
4116 	.pull_stats = efx_mcdi_mac_pull_stats,
4117 	.stop_stats = efx_mcdi_mac_stop_stats,
4118 	.push_irq_moderation = efx_ef10_push_irq_moderation,
4119 	.reconfigure_mac = efx_ef10_mac_reconfigure,
4120 	.check_mac_fault = efx_mcdi_mac_check_fault,
4121 	.reconfigure_port = efx_mcdi_port_reconfigure,
4122 	.get_wol = efx_ef10_get_wol,
4123 	.set_wol = efx_ef10_set_wol,
4124 	.resume_wol = efx_port_dummy_op_void,
4125 	.test_chip = efx_ef10_test_chip,
4126 	.test_nvram = efx_mcdi_nvram_test_all,
4127 	.mcdi_request = efx_ef10_mcdi_request,
4128 	.mcdi_poll_response = efx_ef10_mcdi_poll_response,
4129 	.mcdi_read_response = efx_ef10_mcdi_read_response,
4130 	.mcdi_poll_reboot = efx_ef10_mcdi_poll_reboot,
4131 	.mcdi_reboot_detected = efx_ef10_mcdi_reboot_detected,
4132 	.irq_enable_master = efx_port_dummy_op_void,
4133 	.irq_test_generate = efx_ef10_irq_test_generate,
4134 	.irq_disable_non_ev = efx_port_dummy_op_void,
4135 	.irq_handle_msi = efx_ef10_msi_interrupt,
4136 	.irq_handle_legacy = efx_ef10_legacy_interrupt,
4137 	.tx_probe = efx_ef10_tx_probe,
4138 	.tx_init = efx_ef10_tx_init,
4139 	.tx_remove = efx_mcdi_tx_remove,
4140 	.tx_write = efx_ef10_tx_write,
4141 	.tx_limit_len = efx_ef10_tx_limit_len,
4142 	.tx_enqueue = __efx_enqueue_skb,
4143 	.rx_push_rss_config = efx_mcdi_pf_rx_push_rss_config,
4144 	.rx_pull_rss_config = efx_mcdi_rx_pull_rss_config,
4145 	.rx_push_rss_context_config = efx_mcdi_rx_push_rss_context_config,
4146 	.rx_pull_rss_context_config = efx_mcdi_rx_pull_rss_context_config,
4147 	.rx_restore_rss_contexts = efx_mcdi_rx_restore_rss_contexts,
4148 	.rx_probe = efx_mcdi_rx_probe,
4149 	.rx_init = efx_mcdi_rx_init,
4150 	.rx_remove = efx_mcdi_rx_remove,
4151 	.rx_write = efx_ef10_rx_write,
4152 	.rx_defer_refill = efx_ef10_rx_defer_refill,
4153 	.rx_packet = __efx_rx_packet,
4154 	.ev_probe = efx_mcdi_ev_probe,
4155 	.ev_init = efx_ef10_ev_init,
4156 	.ev_fini = efx_mcdi_ev_fini,
4157 	.ev_remove = efx_mcdi_ev_remove,
4158 	.ev_process = efx_ef10_ev_process,
4159 	.ev_read_ack = efx_ef10_ev_read_ack,
4160 	.ev_test_generate = efx_ef10_ev_test_generate,
4161 	.filter_table_probe = efx_ef10_filter_table_probe,
4162 	.filter_table_restore = efx_mcdi_filter_table_restore,
4163 	.filter_table_remove = efx_mcdi_filter_table_remove,
4164 	.filter_update_rx_scatter = efx_mcdi_update_rx_scatter,
4165 	.filter_insert = efx_mcdi_filter_insert,
4166 	.filter_remove_safe = efx_mcdi_filter_remove_safe,
4167 	.filter_get_safe = efx_mcdi_filter_get_safe,
4168 	.filter_clear_rx = efx_mcdi_filter_clear_rx,
4169 	.filter_count_rx_used = efx_mcdi_filter_count_rx_used,
4170 	.filter_get_rx_id_limit = efx_mcdi_filter_get_rx_id_limit,
4171 	.filter_get_rx_ids = efx_mcdi_filter_get_rx_ids,
4172 #ifdef CONFIG_RFS_ACCEL
4173 	.filter_rfs_expire_one = efx_mcdi_filter_rfs_expire_one,
4174 #endif
4175 #ifdef CONFIG_SFC_MTD
4176 	.mtd_probe = efx_ef10_mtd_probe,
4177 	.mtd_rename = efx_mcdi_mtd_rename,
4178 	.mtd_read = efx_mcdi_mtd_read,
4179 	.mtd_erase = efx_mcdi_mtd_erase,
4180 	.mtd_write = efx_mcdi_mtd_write,
4181 	.mtd_sync = efx_mcdi_mtd_sync,
4182 #endif
4183 	.ptp_write_host_time = efx_ef10_ptp_write_host_time,
4184 	.ptp_set_ts_sync_events = efx_ef10_ptp_set_ts_sync_events,
4185 	.ptp_set_ts_config = efx_ef10_ptp_set_ts_config,
4186 	.vlan_rx_add_vid = efx_ef10_vlan_rx_add_vid,
4187 	.vlan_rx_kill_vid = efx_ef10_vlan_rx_kill_vid,
4188 	.udp_tnl_push_ports = efx_ef10_udp_tnl_push_ports,
4189 	.udp_tnl_has_port = efx_ef10_udp_tnl_has_port,
4190 #ifdef CONFIG_SFC_SRIOV
4191 	.sriov_configure = efx_ef10_sriov_configure,
4192 	.sriov_init = efx_ef10_sriov_init,
4193 	.sriov_fini = efx_ef10_sriov_fini,
4194 	.sriov_wanted = efx_ef10_sriov_wanted,
4195 	.sriov_reset = efx_ef10_sriov_reset,
4196 	.sriov_flr = efx_ef10_sriov_flr,
4197 	.sriov_set_vf_mac = efx_ef10_sriov_set_vf_mac,
4198 	.sriov_set_vf_vlan = efx_ef10_sriov_set_vf_vlan,
4199 	.sriov_set_vf_spoofchk = efx_ef10_sriov_set_vf_spoofchk,
4200 	.sriov_get_vf_config = efx_ef10_sriov_get_vf_config,
4201 	.sriov_set_vf_link_state = efx_ef10_sriov_set_vf_link_state,
4202 	.vswitching_probe = efx_ef10_vswitching_probe_pf,
4203 	.vswitching_restore = efx_ef10_vswitching_restore_pf,
4204 	.vswitching_remove = efx_ef10_vswitching_remove_pf,
4205 #endif
4206 	.get_mac_address = efx_ef10_get_mac_address_pf,
4207 	.set_mac_address = efx_ef10_set_mac_address,
4208 	.tso_versions = efx_ef10_tso_versions,
4209 
4210 	.get_phys_port_id = efx_ef10_get_phys_port_id,
4211 	.revision = EFX_REV_HUNT_A0,
4212 	.max_dma_mask = DMA_BIT_MASK(ESF_DZ_TX_KER_BUF_ADDR_WIDTH),
4213 	.rx_prefix_size = ES_DZ_RX_PREFIX_SIZE,
4214 	.rx_hash_offset = ES_DZ_RX_PREFIX_HASH_OFST,
4215 	.rx_ts_offset = ES_DZ_RX_PREFIX_TSTAMP_OFST,
4216 	.can_rx_scatter = true,
4217 	.always_rx_scatter = true,
4218 	.option_descriptors = true,
4219 	.min_interrupt_mode = EFX_INT_MODE_LEGACY,
4220 	.timer_period_max = 1 << ERF_DD_EVQ_IND_TIMER_VAL_WIDTH,
4221 	.offload_features = EF10_OFFLOAD_FEATURES,
4222 	.mcdi_max_ver = 2,
4223 	.max_rx_ip_filters = EFX_MCDI_FILTER_TBL_ROWS,
4224 	.hwtstamp_filters = 1 << HWTSTAMP_FILTER_NONE |
4225 			    1 << HWTSTAMP_FILTER_ALL,
4226 	.rx_hash_key_size = 40,
4227 	.check_caps = ef10_check_caps,
4228 	.print_additional_fwver = efx_ef10_print_additional_fwver,
4229 	.sensor_event = efx_mcdi_sensor_event,
4230 };
4231