xref: /linux/drivers/net/ethernet/sfc/ef10.c (revision 2c956a5ad4de7376ee792e888809edf2b2b39b86)
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 	if (in_interrupt()) {
1875 		/* If in atomic context, cannot update stats.  Just update the
1876 		 * software stats and return so the caller can continue.
1877 		 */
1878 		spin_lock_bh(&efx->stats_lock);
1879 		efx_update_sw_stats(efx, stats);
1880 		return 0;
1881 	}
1882 
1883 	efx_ef10_get_stat_mask(efx, mask);
1884 
1885 	rc = efx_nic_alloc_buffer(efx, &stats_buf, dma_len, GFP_ATOMIC);
1886 	if (rc) {
1887 		spin_lock_bh(&efx->stats_lock);
1888 		return rc;
1889 	}
1890 
1891 	dma_stats = stats_buf.addr;
1892 	dma_stats[efx->num_mac_stats - 1] = EFX_MC_STATS_GENERATION_INVALID;
1893 
1894 	MCDI_SET_QWORD(inbuf, MAC_STATS_IN_DMA_ADDR, stats_buf.dma_addr);
1895 	MCDI_POPULATE_DWORD_1(inbuf, MAC_STATS_IN_CMD,
1896 			      MAC_STATS_IN_DMA, 1);
1897 	MCDI_SET_DWORD(inbuf, MAC_STATS_IN_DMA_LEN, dma_len);
1898 	MCDI_SET_DWORD(inbuf, MAC_STATS_IN_PORT_ID, EVB_PORT_ID_ASSIGNED);
1899 
1900 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_MAC_STATS, inbuf, sizeof(inbuf),
1901 				NULL, 0, NULL);
1902 	spin_lock_bh(&efx->stats_lock);
1903 	if (rc) {
1904 		/* Expect ENOENT if DMA queues have not been set up */
1905 		if (rc != -ENOENT || atomic_read(&efx->active_queues))
1906 			efx_mcdi_display_error(efx, MC_CMD_MAC_STATS,
1907 					       sizeof(inbuf), NULL, 0, rc);
1908 		goto out;
1909 	}
1910 
1911 	generation_end = dma_stats[efx->num_mac_stats - 1];
1912 	if (generation_end == EFX_MC_STATS_GENERATION_INVALID) {
1913 		WARN_ON_ONCE(1);
1914 		goto out;
1915 	}
1916 	rmb();
1917 	efx_nic_update_stats(efx_ef10_stat_desc, EF10_STAT_COUNT, mask,
1918 			     stats, stats_buf.addr, false);
1919 	rmb();
1920 	generation_start = dma_stats[MC_CMD_MAC_GENERATION_START];
1921 	if (generation_end != generation_start) {
1922 		rc = -EAGAIN;
1923 		goto out;
1924 	}
1925 
1926 	efx_update_sw_stats(efx, stats);
1927 out:
1928 	efx_nic_free_buffer(efx, &stats_buf);
1929 	return rc;
1930 }
1931 
1932 static size_t efx_ef10_update_stats_vf(struct efx_nic *efx, u64 *full_stats,
1933 				       struct rtnl_link_stats64 *core_stats)
1934 {
1935 	if (efx_ef10_try_update_nic_stats_vf(efx))
1936 		return 0;
1937 
1938 	return efx_ef10_update_stats_common(efx, full_stats, core_stats);
1939 }
1940 
1941 static void efx_ef10_push_irq_moderation(struct efx_channel *channel)
1942 {
1943 	struct efx_nic *efx = channel->efx;
1944 	unsigned int mode, usecs;
1945 	efx_dword_t timer_cmd;
1946 
1947 	if (channel->irq_moderation_us) {
1948 		mode = 3;
1949 		usecs = channel->irq_moderation_us;
1950 	} else {
1951 		mode = 0;
1952 		usecs = 0;
1953 	}
1954 
1955 	if (EFX_EF10_WORKAROUND_61265(efx)) {
1956 		MCDI_DECLARE_BUF(inbuf, MC_CMD_SET_EVQ_TMR_IN_LEN);
1957 		unsigned int ns = usecs * 1000;
1958 
1959 		MCDI_SET_DWORD(inbuf, SET_EVQ_TMR_IN_INSTANCE,
1960 			       channel->channel);
1961 		MCDI_SET_DWORD(inbuf, SET_EVQ_TMR_IN_TMR_LOAD_REQ_NS, ns);
1962 		MCDI_SET_DWORD(inbuf, SET_EVQ_TMR_IN_TMR_RELOAD_REQ_NS, ns);
1963 		MCDI_SET_DWORD(inbuf, SET_EVQ_TMR_IN_TMR_MODE, mode);
1964 
1965 		efx_mcdi_rpc_async(efx, MC_CMD_SET_EVQ_TMR,
1966 				   inbuf, sizeof(inbuf), 0, NULL, 0);
1967 	} else if (EFX_EF10_WORKAROUND_35388(efx)) {
1968 		unsigned int ticks = efx_usecs_to_ticks(efx, usecs);
1969 
1970 		EFX_POPULATE_DWORD_3(timer_cmd, ERF_DD_EVQ_IND_TIMER_FLAGS,
1971 				     EFE_DD_EVQ_IND_TIMER_FLAGS,
1972 				     ERF_DD_EVQ_IND_TIMER_MODE, mode,
1973 				     ERF_DD_EVQ_IND_TIMER_VAL, ticks);
1974 		efx_writed_page(efx, &timer_cmd, ER_DD_EVQ_INDIRECT,
1975 				channel->channel);
1976 	} else {
1977 		unsigned int ticks = efx_usecs_to_ticks(efx, usecs);
1978 
1979 		EFX_POPULATE_DWORD_3(timer_cmd, ERF_DZ_TC_TIMER_MODE, mode,
1980 				     ERF_DZ_TC_TIMER_VAL, ticks,
1981 				     ERF_FZ_TC_TMR_REL_VAL, ticks);
1982 		efx_writed_page(efx, &timer_cmd, ER_DZ_EVQ_TMR,
1983 				channel->channel);
1984 	}
1985 }
1986 
1987 static void efx_ef10_get_wol_vf(struct efx_nic *efx,
1988 				struct ethtool_wolinfo *wol) {}
1989 
1990 static int efx_ef10_set_wol_vf(struct efx_nic *efx, u32 type)
1991 {
1992 	return -EOPNOTSUPP;
1993 }
1994 
1995 static void efx_ef10_get_wol(struct efx_nic *efx, struct ethtool_wolinfo *wol)
1996 {
1997 	wol->supported = 0;
1998 	wol->wolopts = 0;
1999 	memset(&wol->sopass, 0, sizeof(wol->sopass));
2000 }
2001 
2002 static int efx_ef10_set_wol(struct efx_nic *efx, u32 type)
2003 {
2004 	if (type != 0)
2005 		return -EINVAL;
2006 	return 0;
2007 }
2008 
2009 static void efx_ef10_mcdi_request(struct efx_nic *efx,
2010 				  const efx_dword_t *hdr, size_t hdr_len,
2011 				  const efx_dword_t *sdu, size_t sdu_len)
2012 {
2013 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2014 	u8 *pdu = nic_data->mcdi_buf.addr;
2015 
2016 	memcpy(pdu, hdr, hdr_len);
2017 	memcpy(pdu + hdr_len, sdu, sdu_len);
2018 	wmb();
2019 
2020 	/* The hardware provides 'low' and 'high' (doorbell) registers
2021 	 * for passing the 64-bit address of an MCDI request to
2022 	 * firmware.  However the dwords are swapped by firmware.  The
2023 	 * least significant bits of the doorbell are then 0 for all
2024 	 * MCDI requests due to alignment.
2025 	 */
2026 	_efx_writed(efx, cpu_to_le32((u64)nic_data->mcdi_buf.dma_addr >> 32),
2027 		    ER_DZ_MC_DB_LWRD);
2028 	_efx_writed(efx, cpu_to_le32((u32)nic_data->mcdi_buf.dma_addr),
2029 		    ER_DZ_MC_DB_HWRD);
2030 }
2031 
2032 static bool efx_ef10_mcdi_poll_response(struct efx_nic *efx)
2033 {
2034 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2035 	const efx_dword_t hdr = *(const efx_dword_t *)nic_data->mcdi_buf.addr;
2036 
2037 	rmb();
2038 	return EFX_DWORD_FIELD(hdr, MCDI_HEADER_RESPONSE);
2039 }
2040 
2041 static void
2042 efx_ef10_mcdi_read_response(struct efx_nic *efx, efx_dword_t *outbuf,
2043 			    size_t offset, size_t outlen)
2044 {
2045 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2046 	const u8 *pdu = nic_data->mcdi_buf.addr;
2047 
2048 	memcpy(outbuf, pdu + offset, outlen);
2049 }
2050 
2051 static void efx_ef10_mcdi_reboot_detected(struct efx_nic *efx)
2052 {
2053 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2054 
2055 	/* All our allocations have been reset */
2056 	efx_ef10_table_reset_mc_allocations(efx);
2057 
2058 	/* The datapath firmware might have been changed */
2059 	nic_data->must_check_datapath_caps = true;
2060 
2061 	/* MAC statistics have been cleared on the NIC; clear the local
2062 	 * statistic that we update with efx_update_diff_stat().
2063 	 */
2064 	nic_data->stats[EF10_STAT_port_rx_bad_bytes] = 0;
2065 }
2066 
2067 static int efx_ef10_mcdi_poll_reboot(struct efx_nic *efx)
2068 {
2069 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2070 	int rc;
2071 
2072 	rc = efx_ef10_get_warm_boot_count(efx);
2073 	if (rc < 0) {
2074 		/* The firmware is presumably in the process of
2075 		 * rebooting.  However, we are supposed to report each
2076 		 * reboot just once, so we must only do that once we
2077 		 * can read and store the updated warm boot count.
2078 		 */
2079 		return 0;
2080 	}
2081 
2082 	if (rc == nic_data->warm_boot_count)
2083 		return 0;
2084 
2085 	nic_data->warm_boot_count = rc;
2086 	efx_ef10_mcdi_reboot_detected(efx);
2087 
2088 	return -EIO;
2089 }
2090 
2091 /* Handle an MSI interrupt
2092  *
2093  * Handle an MSI hardware interrupt.  This routine schedules event
2094  * queue processing.  No interrupt acknowledgement cycle is necessary.
2095  * Also, we never need to check that the interrupt is for us, since
2096  * MSI interrupts cannot be shared.
2097  */
2098 static irqreturn_t efx_ef10_msi_interrupt(int irq, void *dev_id)
2099 {
2100 	struct efx_msi_context *context = dev_id;
2101 	struct efx_nic *efx = context->efx;
2102 
2103 	netif_vdbg(efx, intr, efx->net_dev,
2104 		   "IRQ %d on CPU %d\n", irq, raw_smp_processor_id());
2105 
2106 	if (likely(READ_ONCE(efx->irq_soft_enabled))) {
2107 		/* Note test interrupts */
2108 		if (context->index == efx->irq_level)
2109 			efx->last_irq_cpu = raw_smp_processor_id();
2110 
2111 		/* Schedule processing of the channel */
2112 		efx_schedule_channel_irq(efx->channel[context->index]);
2113 	}
2114 
2115 	return IRQ_HANDLED;
2116 }
2117 
2118 static irqreturn_t efx_ef10_legacy_interrupt(int irq, void *dev_id)
2119 {
2120 	struct efx_nic *efx = dev_id;
2121 	bool soft_enabled = READ_ONCE(efx->irq_soft_enabled);
2122 	struct efx_channel *channel;
2123 	efx_dword_t reg;
2124 	u32 queues;
2125 
2126 	/* Read the ISR which also ACKs the interrupts */
2127 	efx_readd(efx, &reg, ER_DZ_BIU_INT_ISR);
2128 	queues = EFX_DWORD_FIELD(reg, ERF_DZ_ISR_REG);
2129 
2130 	if (queues == 0)
2131 		return IRQ_NONE;
2132 
2133 	if (likely(soft_enabled)) {
2134 		/* Note test interrupts */
2135 		if (queues & (1U << efx->irq_level))
2136 			efx->last_irq_cpu = raw_smp_processor_id();
2137 
2138 		efx_for_each_channel(channel, efx) {
2139 			if (queues & 1)
2140 				efx_schedule_channel_irq(channel);
2141 			queues >>= 1;
2142 		}
2143 	}
2144 
2145 	netif_vdbg(efx, intr, efx->net_dev,
2146 		   "IRQ %d on CPU %d status " EFX_DWORD_FMT "\n",
2147 		   irq, raw_smp_processor_id(), EFX_DWORD_VAL(reg));
2148 
2149 	return IRQ_HANDLED;
2150 }
2151 
2152 static int efx_ef10_irq_test_generate(struct efx_nic *efx)
2153 {
2154 	MCDI_DECLARE_BUF(inbuf, MC_CMD_TRIGGER_INTERRUPT_IN_LEN);
2155 
2156 	if (efx_mcdi_set_workaround(efx, MC_CMD_WORKAROUND_BUG41750, true,
2157 				    NULL) == 0)
2158 		return -ENOTSUPP;
2159 
2160 	BUILD_BUG_ON(MC_CMD_TRIGGER_INTERRUPT_OUT_LEN != 0);
2161 
2162 	MCDI_SET_DWORD(inbuf, TRIGGER_INTERRUPT_IN_INTR_LEVEL, efx->irq_level);
2163 	return efx_mcdi_rpc(efx, MC_CMD_TRIGGER_INTERRUPT,
2164 			    inbuf, sizeof(inbuf), NULL, 0, NULL);
2165 }
2166 
2167 static int efx_ef10_tx_probe(struct efx_tx_queue *tx_queue)
2168 {
2169 	/* low two bits of label are what we want for type */
2170 	BUILD_BUG_ON((EFX_TXQ_TYPE_OUTER_CSUM | EFX_TXQ_TYPE_INNER_CSUM) != 3);
2171 	tx_queue->type = tx_queue->label & 3;
2172 	return efx_nic_alloc_buffer(tx_queue->efx, &tx_queue->txd.buf,
2173 				    (tx_queue->ptr_mask + 1) *
2174 				    sizeof(efx_qword_t),
2175 				    GFP_KERNEL);
2176 }
2177 
2178 /* This writes to the TX_DESC_WPTR and also pushes data */
2179 static inline void efx_ef10_push_tx_desc(struct efx_tx_queue *tx_queue,
2180 					 const efx_qword_t *txd)
2181 {
2182 	unsigned int write_ptr;
2183 	efx_oword_t reg;
2184 
2185 	write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
2186 	EFX_POPULATE_OWORD_1(reg, ERF_DZ_TX_DESC_WPTR, write_ptr);
2187 	reg.qword[0] = *txd;
2188 	efx_writeo_page(tx_queue->efx, &reg,
2189 			ER_DZ_TX_DESC_UPD, tx_queue->queue);
2190 }
2191 
2192 /* Add Firmware-Assisted TSO v2 option descriptors to a queue.
2193  */
2194 int efx_ef10_tx_tso_desc(struct efx_tx_queue *tx_queue, struct sk_buff *skb,
2195 			 bool *data_mapped)
2196 {
2197 	struct efx_tx_buffer *buffer;
2198 	u16 inner_ipv4_id = 0;
2199 	u16 outer_ipv4_id = 0;
2200 	struct tcphdr *tcp;
2201 	struct iphdr *ip;
2202 	u16 ip_tot_len;
2203 	u32 seqnum;
2204 	u32 mss;
2205 
2206 	EFX_WARN_ON_ONCE_PARANOID(tx_queue->tso_version != 2);
2207 
2208 	mss = skb_shinfo(skb)->gso_size;
2209 
2210 	if (unlikely(mss < 4)) {
2211 		WARN_ONCE(1, "MSS of %u is too small for TSO v2\n", mss);
2212 		return -EINVAL;
2213 	}
2214 
2215 	if (skb->encapsulation) {
2216 		if (!tx_queue->tso_encap)
2217 			return -EINVAL;
2218 		ip = ip_hdr(skb);
2219 		if (ip->version == 4)
2220 			outer_ipv4_id = ntohs(ip->id);
2221 
2222 		ip = inner_ip_hdr(skb);
2223 		tcp = inner_tcp_hdr(skb);
2224 	} else {
2225 		ip = ip_hdr(skb);
2226 		tcp = tcp_hdr(skb);
2227 	}
2228 
2229 	/* 8000-series EF10 hardware requires that IP Total Length be
2230 	 * greater than or equal to the value it will have in each segment
2231 	 * (which is at most mss + 208 + TCP header length), but also less
2232 	 * than (0x10000 - inner_network_header).  Otherwise the TCP
2233 	 * checksum calculation will be broken for encapsulated packets.
2234 	 * We fill in ip->tot_len with 0xff30, which should satisfy the
2235 	 * first requirement unless the MSS is ridiculously large (which
2236 	 * should be impossible as the driver max MTU is 9216); it is
2237 	 * guaranteed to satisfy the second as we only attempt TSO if
2238 	 * inner_network_header <= 208.
2239 	 */
2240 	ip_tot_len = -EFX_TSO2_MAX_HDRLEN;
2241 	EFX_WARN_ON_ONCE_PARANOID(mss + EFX_TSO2_MAX_HDRLEN +
2242 				  (tcp->doff << 2u) > ip_tot_len);
2243 
2244 	if (ip->version == 4) {
2245 		ip->tot_len = htons(ip_tot_len);
2246 		ip->check = 0;
2247 		inner_ipv4_id = ntohs(ip->id);
2248 	} else {
2249 		((struct ipv6hdr *)ip)->payload_len = htons(ip_tot_len);
2250 	}
2251 
2252 	seqnum = ntohl(tcp->seq);
2253 
2254 	buffer = efx_tx_queue_get_insert_buffer(tx_queue);
2255 
2256 	buffer->flags = EFX_TX_BUF_OPTION;
2257 	buffer->len = 0;
2258 	buffer->unmap_len = 0;
2259 	EFX_POPULATE_QWORD_5(buffer->option,
2260 			ESF_DZ_TX_DESC_IS_OPT, 1,
2261 			ESF_DZ_TX_OPTION_TYPE, ESE_DZ_TX_OPTION_DESC_TSO,
2262 			ESF_DZ_TX_TSO_OPTION_TYPE,
2263 			ESE_DZ_TX_TSO_OPTION_DESC_FATSO2A,
2264 			ESF_DZ_TX_TSO_IP_ID, inner_ipv4_id,
2265 			ESF_DZ_TX_TSO_TCP_SEQNO, seqnum
2266 			);
2267 	++tx_queue->insert_count;
2268 
2269 	buffer = efx_tx_queue_get_insert_buffer(tx_queue);
2270 
2271 	buffer->flags = EFX_TX_BUF_OPTION;
2272 	buffer->len = 0;
2273 	buffer->unmap_len = 0;
2274 	EFX_POPULATE_QWORD_5(buffer->option,
2275 			ESF_DZ_TX_DESC_IS_OPT, 1,
2276 			ESF_DZ_TX_OPTION_TYPE, ESE_DZ_TX_OPTION_DESC_TSO,
2277 			ESF_DZ_TX_TSO_OPTION_TYPE,
2278 			ESE_DZ_TX_TSO_OPTION_DESC_FATSO2B,
2279 			ESF_DZ_TX_TSO_OUTER_IPID, outer_ipv4_id,
2280 			ESF_DZ_TX_TSO_TCP_MSS, mss
2281 			);
2282 	++tx_queue->insert_count;
2283 
2284 	return 0;
2285 }
2286 
2287 static u32 efx_ef10_tso_versions(struct efx_nic *efx)
2288 {
2289 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2290 	u32 tso_versions = 0;
2291 
2292 	if (nic_data->datapath_caps &
2293 	    (1 << MC_CMD_GET_CAPABILITIES_OUT_TX_TSO_LBN))
2294 		tso_versions |= BIT(1);
2295 	if (nic_data->datapath_caps2 &
2296 	    (1 << MC_CMD_GET_CAPABILITIES_V2_OUT_TX_TSO_V2_LBN))
2297 		tso_versions |= BIT(2);
2298 	return tso_versions;
2299 }
2300 
2301 static void efx_ef10_tx_init(struct efx_tx_queue *tx_queue)
2302 {
2303 	bool csum_offload = tx_queue->type & EFX_TXQ_TYPE_OUTER_CSUM;
2304 	bool inner_csum = tx_queue->type & EFX_TXQ_TYPE_INNER_CSUM;
2305 	struct efx_channel *channel = tx_queue->channel;
2306 	struct efx_nic *efx = tx_queue->efx;
2307 	struct efx_ef10_nic_data *nic_data;
2308 	efx_qword_t *txd;
2309 	int rc;
2310 
2311 	nic_data = efx->nic_data;
2312 
2313 	/* Only attempt to enable TX timestamping if we have the license for it,
2314 	 * otherwise TXQ init will fail
2315 	 */
2316 	if (!(nic_data->licensed_features &
2317 	      (1 << LICENSED_V3_FEATURES_TX_TIMESTAMPS_LBN))) {
2318 		tx_queue->timestamping = false;
2319 		/* Disable sync events on this channel. */
2320 		if (efx->type->ptp_set_ts_sync_events)
2321 			efx->type->ptp_set_ts_sync_events(efx, false, false);
2322 	}
2323 
2324 	/* TSOv2 is a limited resource that can only be configured on a limited
2325 	 * number of queues. TSO without checksum offload is not really a thing,
2326 	 * so we only enable it for those queues.
2327 	 * TSOv2 cannot be used with Hardware timestamping, and is never needed
2328 	 * for XDP tx.
2329 	 */
2330 	if (efx_has_cap(efx, TX_TSO_V2)) {
2331 		if ((csum_offload || inner_csum) &&
2332 		    !tx_queue->timestamping && !tx_queue->xdp_tx) {
2333 			tx_queue->tso_version = 2;
2334 			netif_dbg(efx, hw, efx->net_dev, "Using TSOv2 for channel %u\n",
2335 				  channel->channel);
2336 		}
2337 	} else if (efx_has_cap(efx, TX_TSO)) {
2338 		tx_queue->tso_version = 1;
2339 	}
2340 
2341 	rc = efx_mcdi_tx_init(tx_queue);
2342 	if (rc)
2343 		goto fail;
2344 
2345 	/* A previous user of this TX queue might have set us up the
2346 	 * bomb by writing a descriptor to the TX push collector but
2347 	 * not the doorbell.  (Each collector belongs to a port, not a
2348 	 * queue or function, so cannot easily be reset.)  We must
2349 	 * attempt to push a no-op descriptor in its place.
2350 	 */
2351 	tx_queue->buffer[0].flags = EFX_TX_BUF_OPTION;
2352 	tx_queue->insert_count = 1;
2353 	txd = efx_tx_desc(tx_queue, 0);
2354 	EFX_POPULATE_QWORD_7(*txd,
2355 			     ESF_DZ_TX_DESC_IS_OPT, true,
2356 			     ESF_DZ_TX_OPTION_TYPE,
2357 			     ESE_DZ_TX_OPTION_DESC_CRC_CSUM,
2358 			     ESF_DZ_TX_OPTION_UDP_TCP_CSUM, csum_offload,
2359 			     ESF_DZ_TX_OPTION_IP_CSUM, csum_offload && tx_queue->tso_version != 2,
2360 			     ESF_DZ_TX_OPTION_INNER_UDP_TCP_CSUM, inner_csum,
2361 			     ESF_DZ_TX_OPTION_INNER_IP_CSUM, inner_csum && tx_queue->tso_version != 2,
2362 			     ESF_DZ_TX_TIMESTAMP, tx_queue->timestamping);
2363 	tx_queue->write_count = 1;
2364 
2365 	if (tx_queue->tso_version == 2 && efx_has_cap(efx, TX_TSO_V2_ENCAP))
2366 		tx_queue->tso_encap = true;
2367 
2368 	wmb();
2369 	efx_ef10_push_tx_desc(tx_queue, txd);
2370 
2371 	return;
2372 
2373 fail:
2374 	netdev_WARN(efx->net_dev, "failed to initialise TXQ %d\n",
2375 		    tx_queue->queue);
2376 }
2377 
2378 /* This writes to the TX_DESC_WPTR; write pointer for TX descriptor ring */
2379 static inline void efx_ef10_notify_tx_desc(struct efx_tx_queue *tx_queue)
2380 {
2381 	unsigned int write_ptr;
2382 	efx_dword_t reg;
2383 
2384 	write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
2385 	EFX_POPULATE_DWORD_1(reg, ERF_DZ_TX_DESC_WPTR_DWORD, write_ptr);
2386 	efx_writed_page(tx_queue->efx, &reg,
2387 			ER_DZ_TX_DESC_UPD_DWORD, tx_queue->queue);
2388 }
2389 
2390 #define EFX_EF10_MAX_TX_DESCRIPTOR_LEN 0x3fff
2391 
2392 static unsigned int efx_ef10_tx_limit_len(struct efx_tx_queue *tx_queue,
2393 					  dma_addr_t dma_addr, unsigned int len)
2394 {
2395 	if (len > EFX_EF10_MAX_TX_DESCRIPTOR_LEN) {
2396 		/* If we need to break across multiple descriptors we should
2397 		 * stop at a page boundary. This assumes the length limit is
2398 		 * greater than the page size.
2399 		 */
2400 		dma_addr_t end = dma_addr + EFX_EF10_MAX_TX_DESCRIPTOR_LEN;
2401 
2402 		BUILD_BUG_ON(EFX_EF10_MAX_TX_DESCRIPTOR_LEN < EFX_PAGE_SIZE);
2403 		len = (end & (~(EFX_PAGE_SIZE - 1))) - dma_addr;
2404 	}
2405 
2406 	return len;
2407 }
2408 
2409 static void efx_ef10_tx_write(struct efx_tx_queue *tx_queue)
2410 {
2411 	unsigned int old_write_count = tx_queue->write_count;
2412 	struct efx_tx_buffer *buffer;
2413 	unsigned int write_ptr;
2414 	efx_qword_t *txd;
2415 
2416 	tx_queue->xmit_pending = false;
2417 	if (unlikely(tx_queue->write_count == tx_queue->insert_count))
2418 		return;
2419 
2420 	do {
2421 		write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
2422 		buffer = &tx_queue->buffer[write_ptr];
2423 		txd = efx_tx_desc(tx_queue, write_ptr);
2424 		++tx_queue->write_count;
2425 
2426 		/* Create TX descriptor ring entry */
2427 		if (buffer->flags & EFX_TX_BUF_OPTION) {
2428 			*txd = buffer->option;
2429 			if (EFX_QWORD_FIELD(*txd, ESF_DZ_TX_OPTION_TYPE) == 1)
2430 				/* PIO descriptor */
2431 				tx_queue->packet_write_count = tx_queue->write_count;
2432 		} else {
2433 			tx_queue->packet_write_count = tx_queue->write_count;
2434 			BUILD_BUG_ON(EFX_TX_BUF_CONT != 1);
2435 			EFX_POPULATE_QWORD_3(
2436 				*txd,
2437 				ESF_DZ_TX_KER_CONT,
2438 				buffer->flags & EFX_TX_BUF_CONT,
2439 				ESF_DZ_TX_KER_BYTE_CNT, buffer->len,
2440 				ESF_DZ_TX_KER_BUF_ADDR, buffer->dma_addr);
2441 		}
2442 	} while (tx_queue->write_count != tx_queue->insert_count);
2443 
2444 	wmb(); /* Ensure descriptors are written before they are fetched */
2445 
2446 	if (efx_nic_may_push_tx_desc(tx_queue, old_write_count)) {
2447 		txd = efx_tx_desc(tx_queue,
2448 				  old_write_count & tx_queue->ptr_mask);
2449 		efx_ef10_push_tx_desc(tx_queue, txd);
2450 		++tx_queue->pushes;
2451 	} else {
2452 		efx_ef10_notify_tx_desc(tx_queue);
2453 	}
2454 }
2455 
2456 static int efx_ef10_probe_multicast_chaining(struct efx_nic *efx)
2457 {
2458 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2459 	unsigned int enabled, implemented;
2460 	bool want_workaround_26807;
2461 	int rc;
2462 
2463 	rc = efx_mcdi_get_workarounds(efx, &implemented, &enabled);
2464 	if (rc == -ENOSYS) {
2465 		/* GET_WORKAROUNDS was implemented before this workaround,
2466 		 * thus it must be unavailable in this firmware.
2467 		 */
2468 		nic_data->workaround_26807 = false;
2469 		return 0;
2470 	}
2471 	if (rc)
2472 		return rc;
2473 	want_workaround_26807 =
2474 		implemented & MC_CMD_GET_WORKAROUNDS_OUT_BUG26807;
2475 	nic_data->workaround_26807 =
2476 		!!(enabled & MC_CMD_GET_WORKAROUNDS_OUT_BUG26807);
2477 
2478 	if (want_workaround_26807 && !nic_data->workaround_26807) {
2479 		unsigned int flags;
2480 
2481 		rc = efx_mcdi_set_workaround(efx,
2482 					     MC_CMD_WORKAROUND_BUG26807,
2483 					     true, &flags);
2484 		if (!rc) {
2485 			if (flags &
2486 			    1 << MC_CMD_WORKAROUND_EXT_OUT_FLR_DONE_LBN) {
2487 				netif_info(efx, drv, efx->net_dev,
2488 					   "other functions on NIC have been reset\n");
2489 
2490 				/* With MCFW v4.6.x and earlier, the
2491 				 * boot count will have incremented,
2492 				 * so re-read the warm_boot_count
2493 				 * value now to ensure this function
2494 				 * doesn't think it has changed next
2495 				 * time it checks.
2496 				 */
2497 				rc = efx_ef10_get_warm_boot_count(efx);
2498 				if (rc >= 0) {
2499 					nic_data->warm_boot_count = rc;
2500 					rc = 0;
2501 				}
2502 			}
2503 			nic_data->workaround_26807 = true;
2504 		} else if (rc == -EPERM) {
2505 			rc = 0;
2506 		}
2507 	}
2508 	return rc;
2509 }
2510 
2511 static int efx_ef10_filter_table_probe(struct efx_nic *efx)
2512 {
2513 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2514 	int rc = efx_ef10_probe_multicast_chaining(efx);
2515 	struct efx_mcdi_filter_vlan *vlan;
2516 
2517 	if (rc)
2518 		return rc;
2519 	rc = efx_mcdi_filter_table_probe(efx, nic_data->workaround_26807);
2520 
2521 	if (rc)
2522 		return rc;
2523 
2524 	list_for_each_entry(vlan, &nic_data->vlan_list, list) {
2525 		rc = efx_mcdi_filter_add_vlan(efx, vlan->vid);
2526 		if (rc)
2527 			goto fail_add_vlan;
2528 	}
2529 	return 0;
2530 
2531 fail_add_vlan:
2532 	efx_mcdi_filter_table_remove(efx);
2533 	return rc;
2534 }
2535 
2536 /* This creates an entry in the RX descriptor queue */
2537 static inline void
2538 efx_ef10_build_rx_desc(struct efx_rx_queue *rx_queue, unsigned int index)
2539 {
2540 	struct efx_rx_buffer *rx_buf;
2541 	efx_qword_t *rxd;
2542 
2543 	rxd = efx_rx_desc(rx_queue, index);
2544 	rx_buf = efx_rx_buffer(rx_queue, index);
2545 	EFX_POPULATE_QWORD_2(*rxd,
2546 			     ESF_DZ_RX_KER_BYTE_CNT, rx_buf->len,
2547 			     ESF_DZ_RX_KER_BUF_ADDR, rx_buf->dma_addr);
2548 }
2549 
2550 static void efx_ef10_rx_write(struct efx_rx_queue *rx_queue)
2551 {
2552 	struct efx_nic *efx = rx_queue->efx;
2553 	unsigned int write_count;
2554 	efx_dword_t reg;
2555 
2556 	/* Firmware requires that RX_DESC_WPTR be a multiple of 8 */
2557 	write_count = rx_queue->added_count & ~7;
2558 	if (rx_queue->notified_count == write_count)
2559 		return;
2560 
2561 	do
2562 		efx_ef10_build_rx_desc(
2563 			rx_queue,
2564 			rx_queue->notified_count & rx_queue->ptr_mask);
2565 	while (++rx_queue->notified_count != write_count);
2566 
2567 	wmb();
2568 	EFX_POPULATE_DWORD_1(reg, ERF_DZ_RX_DESC_WPTR,
2569 			     write_count & rx_queue->ptr_mask);
2570 	efx_writed_page(efx, &reg, ER_DZ_RX_DESC_UPD,
2571 			efx_rx_queue_index(rx_queue));
2572 }
2573 
2574 static efx_mcdi_async_completer efx_ef10_rx_defer_refill_complete;
2575 
2576 static void efx_ef10_rx_defer_refill(struct efx_rx_queue *rx_queue)
2577 {
2578 	struct efx_channel *channel = efx_rx_queue_channel(rx_queue);
2579 	MCDI_DECLARE_BUF(inbuf, MC_CMD_DRIVER_EVENT_IN_LEN);
2580 	efx_qword_t event;
2581 
2582 	EFX_POPULATE_QWORD_2(event,
2583 			     ESF_DZ_EV_CODE, EFX_EF10_DRVGEN_EV,
2584 			     ESF_DZ_EV_DATA, EFX_EF10_REFILL);
2585 
2586 	MCDI_SET_DWORD(inbuf, DRIVER_EVENT_IN_EVQ, channel->channel);
2587 
2588 	/* MCDI_SET_QWORD is not appropriate here since EFX_POPULATE_* has
2589 	 * already swapped the data to little-endian order.
2590 	 */
2591 	memcpy(MCDI_PTR(inbuf, DRIVER_EVENT_IN_DATA), &event.u64[0],
2592 	       sizeof(efx_qword_t));
2593 
2594 	efx_mcdi_rpc_async(channel->efx, MC_CMD_DRIVER_EVENT,
2595 			   inbuf, sizeof(inbuf), 0,
2596 			   efx_ef10_rx_defer_refill_complete, 0);
2597 }
2598 
2599 static void
2600 efx_ef10_rx_defer_refill_complete(struct efx_nic *efx, unsigned long cookie,
2601 				  int rc, efx_dword_t *outbuf,
2602 				  size_t outlen_actual)
2603 {
2604 	/* nothing to do */
2605 }
2606 
2607 static int efx_ef10_ev_init(struct efx_channel *channel)
2608 {
2609 	struct efx_nic *efx = channel->efx;
2610 	struct efx_ef10_nic_data *nic_data;
2611 	bool use_v2, cut_thru;
2612 
2613 	nic_data = efx->nic_data;
2614 	use_v2 = nic_data->datapath_caps2 &
2615 			    1 << MC_CMD_GET_CAPABILITIES_V2_OUT_INIT_EVQ_V2_LBN;
2616 	cut_thru = !(nic_data->datapath_caps &
2617 			      1 << MC_CMD_GET_CAPABILITIES_OUT_RX_BATCHING_LBN);
2618 	return efx_mcdi_ev_init(channel, cut_thru, use_v2);
2619 }
2620 
2621 static void efx_ef10_handle_rx_wrong_queue(struct efx_rx_queue *rx_queue,
2622 					   unsigned int rx_queue_label)
2623 {
2624 	struct efx_nic *efx = rx_queue->efx;
2625 
2626 	netif_info(efx, hw, efx->net_dev,
2627 		   "rx event arrived on queue %d labeled as queue %u\n",
2628 		   efx_rx_queue_index(rx_queue), rx_queue_label);
2629 
2630 	efx_schedule_reset(efx, RESET_TYPE_DISABLE);
2631 }
2632 
2633 static void
2634 efx_ef10_handle_rx_bad_lbits(struct efx_rx_queue *rx_queue,
2635 			     unsigned int actual, unsigned int expected)
2636 {
2637 	unsigned int dropped = (actual - expected) & rx_queue->ptr_mask;
2638 	struct efx_nic *efx = rx_queue->efx;
2639 
2640 	netif_info(efx, hw, efx->net_dev,
2641 		   "dropped %d events (index=%d expected=%d)\n",
2642 		   dropped, actual, expected);
2643 
2644 	efx_schedule_reset(efx, RESET_TYPE_DISABLE);
2645 }
2646 
2647 /* partially received RX was aborted. clean up. */
2648 static void efx_ef10_handle_rx_abort(struct efx_rx_queue *rx_queue)
2649 {
2650 	unsigned int rx_desc_ptr;
2651 
2652 	netif_dbg(rx_queue->efx, hw, rx_queue->efx->net_dev,
2653 		  "scattered RX aborted (dropping %u buffers)\n",
2654 		  rx_queue->scatter_n);
2655 
2656 	rx_desc_ptr = rx_queue->removed_count & rx_queue->ptr_mask;
2657 
2658 	efx_rx_packet(rx_queue, rx_desc_ptr, rx_queue->scatter_n,
2659 		      0, EFX_RX_PKT_DISCARD);
2660 
2661 	rx_queue->removed_count += rx_queue->scatter_n;
2662 	rx_queue->scatter_n = 0;
2663 	rx_queue->scatter_len = 0;
2664 	++efx_rx_queue_channel(rx_queue)->n_rx_nodesc_trunc;
2665 }
2666 
2667 static u16 efx_ef10_handle_rx_event_errors(struct efx_channel *channel,
2668 					   unsigned int n_packets,
2669 					   unsigned int rx_encap_hdr,
2670 					   unsigned int rx_l3_class,
2671 					   unsigned int rx_l4_class,
2672 					   const efx_qword_t *event)
2673 {
2674 	struct efx_nic *efx = channel->efx;
2675 	bool handled = false;
2676 
2677 	if (EFX_QWORD_FIELD(*event, ESF_DZ_RX_ECRC_ERR)) {
2678 		if (!(efx->net_dev->features & NETIF_F_RXALL)) {
2679 			if (!efx->loopback_selftest)
2680 				channel->n_rx_eth_crc_err += n_packets;
2681 			return EFX_RX_PKT_DISCARD;
2682 		}
2683 		handled = true;
2684 	}
2685 	if (EFX_QWORD_FIELD(*event, ESF_DZ_RX_IPCKSUM_ERR)) {
2686 		if (unlikely(rx_encap_hdr != ESE_EZ_ENCAP_HDR_VXLAN &&
2687 			     rx_l3_class != ESE_DZ_L3_CLASS_IP4 &&
2688 			     rx_l3_class != ESE_DZ_L3_CLASS_IP4_FRAG &&
2689 			     rx_l3_class != ESE_DZ_L3_CLASS_IP6 &&
2690 			     rx_l3_class != ESE_DZ_L3_CLASS_IP6_FRAG))
2691 			netdev_WARN(efx->net_dev,
2692 				    "invalid class for RX_IPCKSUM_ERR: event="
2693 				    EFX_QWORD_FMT "\n",
2694 				    EFX_QWORD_VAL(*event));
2695 		if (!efx->loopback_selftest)
2696 			*(rx_encap_hdr ?
2697 			  &channel->n_rx_outer_ip_hdr_chksum_err :
2698 			  &channel->n_rx_ip_hdr_chksum_err) += n_packets;
2699 		return 0;
2700 	}
2701 	if (EFX_QWORD_FIELD(*event, ESF_DZ_RX_TCPUDP_CKSUM_ERR)) {
2702 		if (unlikely(rx_encap_hdr != ESE_EZ_ENCAP_HDR_VXLAN &&
2703 			     ((rx_l3_class != ESE_DZ_L3_CLASS_IP4 &&
2704 			       rx_l3_class != ESE_DZ_L3_CLASS_IP6) ||
2705 			      (rx_l4_class != ESE_FZ_L4_CLASS_TCP &&
2706 			       rx_l4_class != ESE_FZ_L4_CLASS_UDP))))
2707 			netdev_WARN(efx->net_dev,
2708 				    "invalid class for RX_TCPUDP_CKSUM_ERR: event="
2709 				    EFX_QWORD_FMT "\n",
2710 				    EFX_QWORD_VAL(*event));
2711 		if (!efx->loopback_selftest)
2712 			*(rx_encap_hdr ?
2713 			  &channel->n_rx_outer_tcp_udp_chksum_err :
2714 			  &channel->n_rx_tcp_udp_chksum_err) += n_packets;
2715 		return 0;
2716 	}
2717 	if (EFX_QWORD_FIELD(*event, ESF_EZ_RX_IP_INNER_CHKSUM_ERR)) {
2718 		if (unlikely(!rx_encap_hdr))
2719 			netdev_WARN(efx->net_dev,
2720 				    "invalid encapsulation type for RX_IP_INNER_CHKSUM_ERR: event="
2721 				    EFX_QWORD_FMT "\n",
2722 				    EFX_QWORD_VAL(*event));
2723 		else if (unlikely(rx_l3_class != ESE_DZ_L3_CLASS_IP4 &&
2724 				  rx_l3_class != ESE_DZ_L3_CLASS_IP4_FRAG &&
2725 				  rx_l3_class != ESE_DZ_L3_CLASS_IP6 &&
2726 				  rx_l3_class != ESE_DZ_L3_CLASS_IP6_FRAG))
2727 			netdev_WARN(efx->net_dev,
2728 				    "invalid class for RX_IP_INNER_CHKSUM_ERR: event="
2729 				    EFX_QWORD_FMT "\n",
2730 				    EFX_QWORD_VAL(*event));
2731 		if (!efx->loopback_selftest)
2732 			channel->n_rx_inner_ip_hdr_chksum_err += n_packets;
2733 		return 0;
2734 	}
2735 	if (EFX_QWORD_FIELD(*event, ESF_EZ_RX_TCP_UDP_INNER_CHKSUM_ERR)) {
2736 		if (unlikely(!rx_encap_hdr))
2737 			netdev_WARN(efx->net_dev,
2738 				    "invalid encapsulation type for RX_TCP_UDP_INNER_CHKSUM_ERR: event="
2739 				    EFX_QWORD_FMT "\n",
2740 				    EFX_QWORD_VAL(*event));
2741 		else if (unlikely((rx_l3_class != ESE_DZ_L3_CLASS_IP4 &&
2742 				   rx_l3_class != ESE_DZ_L3_CLASS_IP6) ||
2743 				  (rx_l4_class != ESE_FZ_L4_CLASS_TCP &&
2744 				   rx_l4_class != ESE_FZ_L4_CLASS_UDP)))
2745 			netdev_WARN(efx->net_dev,
2746 				    "invalid class for RX_TCP_UDP_INNER_CHKSUM_ERR: event="
2747 				    EFX_QWORD_FMT "\n",
2748 				    EFX_QWORD_VAL(*event));
2749 		if (!efx->loopback_selftest)
2750 			channel->n_rx_inner_tcp_udp_chksum_err += n_packets;
2751 		return 0;
2752 	}
2753 
2754 	WARN_ON(!handled); /* No error bits were recognised */
2755 	return 0;
2756 }
2757 
2758 static int efx_ef10_handle_rx_event(struct efx_channel *channel,
2759 				    const efx_qword_t *event)
2760 {
2761 	unsigned int rx_bytes, next_ptr_lbits, rx_queue_label;
2762 	unsigned int rx_l3_class, rx_l4_class, rx_encap_hdr;
2763 	unsigned int n_descs, n_packets, i;
2764 	struct efx_nic *efx = channel->efx;
2765 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2766 	struct efx_rx_queue *rx_queue;
2767 	efx_qword_t errors;
2768 	bool rx_cont;
2769 	u16 flags = 0;
2770 
2771 	if (unlikely(READ_ONCE(efx->reset_pending)))
2772 		return 0;
2773 
2774 	/* Basic packet information */
2775 	rx_bytes = EFX_QWORD_FIELD(*event, ESF_DZ_RX_BYTES);
2776 	next_ptr_lbits = EFX_QWORD_FIELD(*event, ESF_DZ_RX_DSC_PTR_LBITS);
2777 	rx_queue_label = EFX_QWORD_FIELD(*event, ESF_DZ_RX_QLABEL);
2778 	rx_l3_class = EFX_QWORD_FIELD(*event, ESF_DZ_RX_L3_CLASS);
2779 	rx_l4_class = EFX_QWORD_FIELD(*event, ESF_FZ_RX_L4_CLASS);
2780 	rx_cont = EFX_QWORD_FIELD(*event, ESF_DZ_RX_CONT);
2781 	rx_encap_hdr =
2782 		nic_data->datapath_caps &
2783 			(1 << MC_CMD_GET_CAPABILITIES_OUT_VXLAN_NVGRE_LBN) ?
2784 		EFX_QWORD_FIELD(*event, ESF_EZ_RX_ENCAP_HDR) :
2785 		ESE_EZ_ENCAP_HDR_NONE;
2786 
2787 	if (EFX_QWORD_FIELD(*event, ESF_DZ_RX_DROP_EVENT))
2788 		netdev_WARN(efx->net_dev, "saw RX_DROP_EVENT: event="
2789 			    EFX_QWORD_FMT "\n",
2790 			    EFX_QWORD_VAL(*event));
2791 
2792 	rx_queue = efx_channel_get_rx_queue(channel);
2793 
2794 	if (unlikely(rx_queue_label != efx_rx_queue_index(rx_queue)))
2795 		efx_ef10_handle_rx_wrong_queue(rx_queue, rx_queue_label);
2796 
2797 	n_descs = ((next_ptr_lbits - rx_queue->removed_count) &
2798 		   ((1 << ESF_DZ_RX_DSC_PTR_LBITS_WIDTH) - 1));
2799 
2800 	if (n_descs != rx_queue->scatter_n + 1) {
2801 		struct efx_ef10_nic_data *nic_data = efx->nic_data;
2802 
2803 		/* detect rx abort */
2804 		if (unlikely(n_descs == rx_queue->scatter_n)) {
2805 			if (rx_queue->scatter_n == 0 || rx_bytes != 0)
2806 				netdev_WARN(efx->net_dev,
2807 					    "invalid RX abort: scatter_n=%u event="
2808 					    EFX_QWORD_FMT "\n",
2809 					    rx_queue->scatter_n,
2810 					    EFX_QWORD_VAL(*event));
2811 			efx_ef10_handle_rx_abort(rx_queue);
2812 			return 0;
2813 		}
2814 
2815 		/* Check that RX completion merging is valid, i.e.
2816 		 * the current firmware supports it and this is a
2817 		 * non-scattered packet.
2818 		 */
2819 		if (!(nic_data->datapath_caps &
2820 		      (1 << MC_CMD_GET_CAPABILITIES_OUT_RX_BATCHING_LBN)) ||
2821 		    rx_queue->scatter_n != 0 || rx_cont) {
2822 			efx_ef10_handle_rx_bad_lbits(
2823 				rx_queue, next_ptr_lbits,
2824 				(rx_queue->removed_count +
2825 				 rx_queue->scatter_n + 1) &
2826 				((1 << ESF_DZ_RX_DSC_PTR_LBITS_WIDTH) - 1));
2827 			return 0;
2828 		}
2829 
2830 		/* Merged completion for multiple non-scattered packets */
2831 		rx_queue->scatter_n = 1;
2832 		rx_queue->scatter_len = 0;
2833 		n_packets = n_descs;
2834 		++channel->n_rx_merge_events;
2835 		channel->n_rx_merge_packets += n_packets;
2836 		flags |= EFX_RX_PKT_PREFIX_LEN;
2837 	} else {
2838 		++rx_queue->scatter_n;
2839 		rx_queue->scatter_len += rx_bytes;
2840 		if (rx_cont)
2841 			return 0;
2842 		n_packets = 1;
2843 	}
2844 
2845 	EFX_POPULATE_QWORD_5(errors, ESF_DZ_RX_ECRC_ERR, 1,
2846 				     ESF_DZ_RX_IPCKSUM_ERR, 1,
2847 				     ESF_DZ_RX_TCPUDP_CKSUM_ERR, 1,
2848 				     ESF_EZ_RX_IP_INNER_CHKSUM_ERR, 1,
2849 				     ESF_EZ_RX_TCP_UDP_INNER_CHKSUM_ERR, 1);
2850 	EFX_AND_QWORD(errors, *event, errors);
2851 	if (unlikely(!EFX_QWORD_IS_ZERO(errors))) {
2852 		flags |= efx_ef10_handle_rx_event_errors(channel, n_packets,
2853 							 rx_encap_hdr,
2854 							 rx_l3_class, rx_l4_class,
2855 							 event);
2856 	} else {
2857 		bool tcpudp = rx_l4_class == ESE_FZ_L4_CLASS_TCP ||
2858 			      rx_l4_class == ESE_FZ_L4_CLASS_UDP;
2859 
2860 		switch (rx_encap_hdr) {
2861 		case ESE_EZ_ENCAP_HDR_VXLAN: /* VxLAN or GENEVE */
2862 			flags |= EFX_RX_PKT_CSUMMED; /* outer UDP csum */
2863 			if (tcpudp)
2864 				flags |= EFX_RX_PKT_CSUM_LEVEL; /* inner L4 */
2865 			break;
2866 		case ESE_EZ_ENCAP_HDR_GRE:
2867 		case ESE_EZ_ENCAP_HDR_NONE:
2868 			if (tcpudp)
2869 				flags |= EFX_RX_PKT_CSUMMED;
2870 			break;
2871 		default:
2872 			netdev_WARN(efx->net_dev,
2873 				    "unknown encapsulation type: event="
2874 				    EFX_QWORD_FMT "\n",
2875 				    EFX_QWORD_VAL(*event));
2876 		}
2877 	}
2878 
2879 	if (rx_l4_class == ESE_FZ_L4_CLASS_TCP)
2880 		flags |= EFX_RX_PKT_TCP;
2881 
2882 	channel->irq_mod_score += 2 * n_packets;
2883 
2884 	/* Handle received packet(s) */
2885 	for (i = 0; i < n_packets; i++) {
2886 		efx_rx_packet(rx_queue,
2887 			      rx_queue->removed_count & rx_queue->ptr_mask,
2888 			      rx_queue->scatter_n, rx_queue->scatter_len,
2889 			      flags);
2890 		rx_queue->removed_count += rx_queue->scatter_n;
2891 	}
2892 
2893 	rx_queue->scatter_n = 0;
2894 	rx_queue->scatter_len = 0;
2895 
2896 	return n_packets;
2897 }
2898 
2899 static u32 efx_ef10_extract_event_ts(efx_qword_t *event)
2900 {
2901 	u32 tstamp;
2902 
2903 	tstamp = EFX_QWORD_FIELD(*event, TX_TIMESTAMP_EVENT_TSTAMP_DATA_HI);
2904 	tstamp <<= 16;
2905 	tstamp |= EFX_QWORD_FIELD(*event, TX_TIMESTAMP_EVENT_TSTAMP_DATA_LO);
2906 
2907 	return tstamp;
2908 }
2909 
2910 static void
2911 efx_ef10_handle_tx_event(struct efx_channel *channel, efx_qword_t *event)
2912 {
2913 	struct efx_nic *efx = channel->efx;
2914 	struct efx_tx_queue *tx_queue;
2915 	unsigned int tx_ev_desc_ptr;
2916 	unsigned int tx_ev_q_label;
2917 	unsigned int tx_ev_type;
2918 	u64 ts_part;
2919 
2920 	if (unlikely(READ_ONCE(efx->reset_pending)))
2921 		return;
2922 
2923 	if (unlikely(EFX_QWORD_FIELD(*event, ESF_DZ_TX_DROP_EVENT)))
2924 		return;
2925 
2926 	/* Get the transmit queue */
2927 	tx_ev_q_label = EFX_QWORD_FIELD(*event, ESF_DZ_TX_QLABEL);
2928 	tx_queue = efx_channel_get_tx_queue(channel,
2929 					    tx_ev_q_label % EFX_MAX_TXQ_PER_CHANNEL);
2930 
2931 	if (!tx_queue->timestamping) {
2932 		/* Transmit completion */
2933 		tx_ev_desc_ptr = EFX_QWORD_FIELD(*event, ESF_DZ_TX_DESCR_INDX);
2934 		efx_xmit_done(tx_queue, tx_ev_desc_ptr & tx_queue->ptr_mask);
2935 		return;
2936 	}
2937 
2938 	/* Transmit timestamps are only available for 8XXX series. They result
2939 	 * in up to three events per packet. These occur in order, and are:
2940 	 *  - the normal completion event (may be omitted)
2941 	 *  - the low part of the timestamp
2942 	 *  - the high part of the timestamp
2943 	 *
2944 	 * It's possible for multiple completion events to appear before the
2945 	 * corresponding timestamps. So we can for example get:
2946 	 *  COMP N
2947 	 *  COMP N+1
2948 	 *  TS_LO N
2949 	 *  TS_HI N
2950 	 *  TS_LO N+1
2951 	 *  TS_HI N+1
2952 	 *
2953 	 * In addition it's also possible for the adjacent completions to be
2954 	 * merged, so we may not see COMP N above. As such, the completion
2955 	 * events are not very useful here.
2956 	 *
2957 	 * Each part of the timestamp is itself split across two 16 bit
2958 	 * fields in the event.
2959 	 */
2960 	tx_ev_type = EFX_QWORD_FIELD(*event, ESF_EZ_TX_SOFT1);
2961 
2962 	switch (tx_ev_type) {
2963 	case TX_TIMESTAMP_EVENT_TX_EV_COMPLETION:
2964 		/* Ignore this event - see above. */
2965 		break;
2966 
2967 	case TX_TIMESTAMP_EVENT_TX_EV_TSTAMP_LO:
2968 		ts_part = efx_ef10_extract_event_ts(event);
2969 		tx_queue->completed_timestamp_minor = ts_part;
2970 		break;
2971 
2972 	case TX_TIMESTAMP_EVENT_TX_EV_TSTAMP_HI:
2973 		ts_part = efx_ef10_extract_event_ts(event);
2974 		tx_queue->completed_timestamp_major = ts_part;
2975 
2976 		efx_xmit_done_single(tx_queue);
2977 		break;
2978 
2979 	default:
2980 		netif_err(efx, hw, efx->net_dev,
2981 			  "channel %d unknown tx event type %d (data "
2982 			  EFX_QWORD_FMT ")\n",
2983 			  channel->channel, tx_ev_type,
2984 			  EFX_QWORD_VAL(*event));
2985 		break;
2986 	}
2987 }
2988 
2989 static void
2990 efx_ef10_handle_driver_event(struct efx_channel *channel, efx_qword_t *event)
2991 {
2992 	struct efx_nic *efx = channel->efx;
2993 	int subcode;
2994 
2995 	subcode = EFX_QWORD_FIELD(*event, ESF_DZ_DRV_SUB_CODE);
2996 
2997 	switch (subcode) {
2998 	case ESE_DZ_DRV_TIMER_EV:
2999 	case ESE_DZ_DRV_WAKE_UP_EV:
3000 		break;
3001 	case ESE_DZ_DRV_START_UP_EV:
3002 		/* event queue init complete. ok. */
3003 		break;
3004 	default:
3005 		netif_err(efx, hw, efx->net_dev,
3006 			  "channel %d unknown driver event type %d"
3007 			  " (data " EFX_QWORD_FMT ")\n",
3008 			  channel->channel, subcode,
3009 			  EFX_QWORD_VAL(*event));
3010 
3011 	}
3012 }
3013 
3014 static void efx_ef10_handle_driver_generated_event(struct efx_channel *channel,
3015 						   efx_qword_t *event)
3016 {
3017 	struct efx_nic *efx = channel->efx;
3018 	u32 subcode;
3019 
3020 	subcode = EFX_QWORD_FIELD(*event, EFX_DWORD_0);
3021 
3022 	switch (subcode) {
3023 	case EFX_EF10_TEST:
3024 		channel->event_test_cpu = raw_smp_processor_id();
3025 		break;
3026 	case EFX_EF10_REFILL:
3027 		/* The queue must be empty, so we won't receive any rx
3028 		 * events, so efx_process_channel() won't refill the
3029 		 * queue. Refill it here
3030 		 */
3031 		efx_fast_push_rx_descriptors(&channel->rx_queue, true);
3032 		break;
3033 	default:
3034 		netif_err(efx, hw, efx->net_dev,
3035 			  "channel %d unknown driver event type %u"
3036 			  " (data " EFX_QWORD_FMT ")\n",
3037 			  channel->channel, (unsigned) subcode,
3038 			  EFX_QWORD_VAL(*event));
3039 	}
3040 }
3041 
3042 static int efx_ef10_ev_process(struct efx_channel *channel, int quota)
3043 {
3044 	struct efx_nic *efx = channel->efx;
3045 	efx_qword_t event, *p_event;
3046 	unsigned int read_ptr;
3047 	int ev_code;
3048 	int spent = 0;
3049 
3050 	if (quota <= 0)
3051 		return spent;
3052 
3053 	read_ptr = channel->eventq_read_ptr;
3054 
3055 	for (;;) {
3056 		p_event = efx_event(channel, read_ptr);
3057 		event = *p_event;
3058 
3059 		if (!efx_event_present(&event))
3060 			break;
3061 
3062 		EFX_SET_QWORD(*p_event);
3063 
3064 		++read_ptr;
3065 
3066 		ev_code = EFX_QWORD_FIELD(event, ESF_DZ_EV_CODE);
3067 
3068 		netif_vdbg(efx, drv, efx->net_dev,
3069 			   "processing event on %d " EFX_QWORD_FMT "\n",
3070 			   channel->channel, EFX_QWORD_VAL(event));
3071 
3072 		switch (ev_code) {
3073 		case ESE_DZ_EV_CODE_MCDI_EV:
3074 			efx_mcdi_process_event(channel, &event);
3075 			break;
3076 		case ESE_DZ_EV_CODE_RX_EV:
3077 			spent += efx_ef10_handle_rx_event(channel, &event);
3078 			if (spent >= quota) {
3079 				/* XXX can we split a merged event to
3080 				 * avoid going over-quota?
3081 				 */
3082 				spent = quota;
3083 				goto out;
3084 			}
3085 			break;
3086 		case ESE_DZ_EV_CODE_TX_EV:
3087 			efx_ef10_handle_tx_event(channel, &event);
3088 			break;
3089 		case ESE_DZ_EV_CODE_DRIVER_EV:
3090 			efx_ef10_handle_driver_event(channel, &event);
3091 			if (++spent == quota)
3092 				goto out;
3093 			break;
3094 		case EFX_EF10_DRVGEN_EV:
3095 			efx_ef10_handle_driver_generated_event(channel, &event);
3096 			break;
3097 		default:
3098 			netif_err(efx, hw, efx->net_dev,
3099 				  "channel %d unknown event type %d"
3100 				  " (data " EFX_QWORD_FMT ")\n",
3101 				  channel->channel, ev_code,
3102 				  EFX_QWORD_VAL(event));
3103 		}
3104 	}
3105 
3106 out:
3107 	channel->eventq_read_ptr = read_ptr;
3108 	return spent;
3109 }
3110 
3111 static void efx_ef10_ev_read_ack(struct efx_channel *channel)
3112 {
3113 	struct efx_nic *efx = channel->efx;
3114 	efx_dword_t rptr;
3115 
3116 	if (EFX_EF10_WORKAROUND_35388(efx)) {
3117 		BUILD_BUG_ON(EFX_MIN_EVQ_SIZE <
3118 			     (1 << ERF_DD_EVQ_IND_RPTR_WIDTH));
3119 		BUILD_BUG_ON(EFX_MAX_EVQ_SIZE >
3120 			     (1 << 2 * ERF_DD_EVQ_IND_RPTR_WIDTH));
3121 
3122 		EFX_POPULATE_DWORD_2(rptr, ERF_DD_EVQ_IND_RPTR_FLAGS,
3123 				     EFE_DD_EVQ_IND_RPTR_FLAGS_HIGH,
3124 				     ERF_DD_EVQ_IND_RPTR,
3125 				     (channel->eventq_read_ptr &
3126 				      channel->eventq_mask) >>
3127 				     ERF_DD_EVQ_IND_RPTR_WIDTH);
3128 		efx_writed_page(efx, &rptr, ER_DD_EVQ_INDIRECT,
3129 				channel->channel);
3130 		EFX_POPULATE_DWORD_2(rptr, ERF_DD_EVQ_IND_RPTR_FLAGS,
3131 				     EFE_DD_EVQ_IND_RPTR_FLAGS_LOW,
3132 				     ERF_DD_EVQ_IND_RPTR,
3133 				     channel->eventq_read_ptr &
3134 				     ((1 << ERF_DD_EVQ_IND_RPTR_WIDTH) - 1));
3135 		efx_writed_page(efx, &rptr, ER_DD_EVQ_INDIRECT,
3136 				channel->channel);
3137 	} else {
3138 		EFX_POPULATE_DWORD_1(rptr, ERF_DZ_EVQ_RPTR,
3139 				     channel->eventq_read_ptr &
3140 				     channel->eventq_mask);
3141 		efx_writed_page(efx, &rptr, ER_DZ_EVQ_RPTR, channel->channel);
3142 	}
3143 }
3144 
3145 static void efx_ef10_ev_test_generate(struct efx_channel *channel)
3146 {
3147 	MCDI_DECLARE_BUF(inbuf, MC_CMD_DRIVER_EVENT_IN_LEN);
3148 	struct efx_nic *efx = channel->efx;
3149 	efx_qword_t event;
3150 	int rc;
3151 
3152 	EFX_POPULATE_QWORD_2(event,
3153 			     ESF_DZ_EV_CODE, EFX_EF10_DRVGEN_EV,
3154 			     ESF_DZ_EV_DATA, EFX_EF10_TEST);
3155 
3156 	MCDI_SET_DWORD(inbuf, DRIVER_EVENT_IN_EVQ, channel->channel);
3157 
3158 	/* MCDI_SET_QWORD is not appropriate here since EFX_POPULATE_* has
3159 	 * already swapped the data to little-endian order.
3160 	 */
3161 	memcpy(MCDI_PTR(inbuf, DRIVER_EVENT_IN_DATA), &event.u64[0],
3162 	       sizeof(efx_qword_t));
3163 
3164 	rc = efx_mcdi_rpc(efx, MC_CMD_DRIVER_EVENT, inbuf, sizeof(inbuf),
3165 			  NULL, 0, NULL);
3166 	if (rc != 0)
3167 		goto fail;
3168 
3169 	return;
3170 
3171 fail:
3172 	WARN_ON(true);
3173 	netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
3174 }
3175 
3176 static void efx_ef10_prepare_flr(struct efx_nic *efx)
3177 {
3178 	atomic_set(&efx->active_queues, 0);
3179 }
3180 
3181 static int efx_ef10_vport_set_mac_address(struct efx_nic *efx)
3182 {
3183 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
3184 	u8 mac_old[ETH_ALEN];
3185 	int rc, rc2;
3186 
3187 	/* Only reconfigure a PF-created vport */
3188 	if (is_zero_ether_addr(nic_data->vport_mac))
3189 		return 0;
3190 
3191 	efx_device_detach_sync(efx);
3192 	efx_net_stop(efx->net_dev);
3193 	down_write(&efx->filter_sem);
3194 	efx_mcdi_filter_table_remove(efx);
3195 	up_write(&efx->filter_sem);
3196 
3197 	rc = efx_ef10_vadaptor_free(efx, efx->vport_id);
3198 	if (rc)
3199 		goto restore_filters;
3200 
3201 	ether_addr_copy(mac_old, nic_data->vport_mac);
3202 	rc = efx_ef10_vport_del_mac(efx, efx->vport_id,
3203 				    nic_data->vport_mac);
3204 	if (rc)
3205 		goto restore_vadaptor;
3206 
3207 	rc = efx_ef10_vport_add_mac(efx, efx->vport_id,
3208 				    efx->net_dev->dev_addr);
3209 	if (!rc) {
3210 		ether_addr_copy(nic_data->vport_mac, efx->net_dev->dev_addr);
3211 	} else {
3212 		rc2 = efx_ef10_vport_add_mac(efx, efx->vport_id, mac_old);
3213 		if (rc2) {
3214 			/* Failed to add original MAC, so clear vport_mac */
3215 			eth_zero_addr(nic_data->vport_mac);
3216 			goto reset_nic;
3217 		}
3218 	}
3219 
3220 restore_vadaptor:
3221 	rc2 = efx_ef10_vadaptor_alloc(efx, efx->vport_id);
3222 	if (rc2)
3223 		goto reset_nic;
3224 restore_filters:
3225 	down_write(&efx->filter_sem);
3226 	rc2 = efx_ef10_filter_table_probe(efx);
3227 	up_write(&efx->filter_sem);
3228 	if (rc2)
3229 		goto reset_nic;
3230 
3231 	rc2 = efx_net_open(efx->net_dev);
3232 	if (rc2)
3233 		goto reset_nic;
3234 
3235 	efx_device_attach_if_not_resetting(efx);
3236 
3237 	return rc;
3238 
3239 reset_nic:
3240 	netif_err(efx, drv, efx->net_dev,
3241 		  "Failed to restore when changing MAC address - scheduling reset\n");
3242 	efx_schedule_reset(efx, RESET_TYPE_DATAPATH);
3243 
3244 	return rc ? rc : rc2;
3245 }
3246 
3247 static int efx_ef10_set_mac_address(struct efx_nic *efx)
3248 {
3249 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VADAPTOR_SET_MAC_IN_LEN);
3250 	bool was_enabled = efx->port_enabled;
3251 	int rc;
3252 
3253 	efx_device_detach_sync(efx);
3254 	efx_net_stop(efx->net_dev);
3255 
3256 	mutex_lock(&efx->mac_lock);
3257 	down_write(&efx->filter_sem);
3258 	efx_mcdi_filter_table_remove(efx);
3259 
3260 	ether_addr_copy(MCDI_PTR(inbuf, VADAPTOR_SET_MAC_IN_MACADDR),
3261 			efx->net_dev->dev_addr);
3262 	MCDI_SET_DWORD(inbuf, VADAPTOR_SET_MAC_IN_UPSTREAM_PORT_ID,
3263 		       efx->vport_id);
3264 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_VADAPTOR_SET_MAC, inbuf,
3265 				sizeof(inbuf), NULL, 0, NULL);
3266 
3267 	efx_ef10_filter_table_probe(efx);
3268 	up_write(&efx->filter_sem);
3269 	mutex_unlock(&efx->mac_lock);
3270 
3271 	if (was_enabled)
3272 		efx_net_open(efx->net_dev);
3273 	efx_device_attach_if_not_resetting(efx);
3274 
3275 #ifdef CONFIG_SFC_SRIOV
3276 	if (efx->pci_dev->is_virtfn && efx->pci_dev->physfn) {
3277 		struct efx_ef10_nic_data *nic_data = efx->nic_data;
3278 		struct pci_dev *pci_dev_pf = efx->pci_dev->physfn;
3279 
3280 		if (rc == -EPERM) {
3281 			struct efx_nic *efx_pf;
3282 
3283 			/* Switch to PF and change MAC address on vport */
3284 			efx_pf = pci_get_drvdata(pci_dev_pf);
3285 
3286 			rc = efx_ef10_sriov_set_vf_mac(efx_pf,
3287 						       nic_data->vf_index,
3288 						       efx->net_dev->dev_addr);
3289 		} else if (!rc) {
3290 			struct efx_nic *efx_pf = pci_get_drvdata(pci_dev_pf);
3291 			struct efx_ef10_nic_data *nic_data = efx_pf->nic_data;
3292 			unsigned int i;
3293 
3294 			/* MAC address successfully changed by VF (with MAC
3295 			 * spoofing) so update the parent PF if possible.
3296 			 */
3297 			for (i = 0; i < efx_pf->vf_count; ++i) {
3298 				struct ef10_vf *vf = nic_data->vf + i;
3299 
3300 				if (vf->efx == efx) {
3301 					ether_addr_copy(vf->mac,
3302 							efx->net_dev->dev_addr);
3303 					return 0;
3304 				}
3305 			}
3306 		}
3307 	} else
3308 #endif
3309 	if (rc == -EPERM) {
3310 		netif_err(efx, drv, efx->net_dev,
3311 			  "Cannot change MAC address; use sfboot to enable"
3312 			  " mac-spoofing on this interface\n");
3313 	} else if (rc == -ENOSYS && !efx_ef10_is_vf(efx)) {
3314 		/* If the active MCFW does not support MC_CMD_VADAPTOR_SET_MAC
3315 		 * fall-back to the method of changing the MAC address on the
3316 		 * vport.  This only applies to PFs because such versions of
3317 		 * MCFW do not support VFs.
3318 		 */
3319 		rc = efx_ef10_vport_set_mac_address(efx);
3320 	} else if (rc) {
3321 		efx_mcdi_display_error(efx, MC_CMD_VADAPTOR_SET_MAC,
3322 				       sizeof(inbuf), NULL, 0, rc);
3323 	}
3324 
3325 	return rc;
3326 }
3327 
3328 static int efx_ef10_mac_reconfigure(struct efx_nic *efx, bool mtu_only)
3329 {
3330 	WARN_ON(!mutex_is_locked(&efx->mac_lock));
3331 
3332 	efx_mcdi_filter_sync_rx_mode(efx);
3333 
3334 	if (mtu_only && efx_has_cap(efx, SET_MAC_ENHANCED))
3335 		return efx_mcdi_set_mtu(efx);
3336 	return efx_mcdi_set_mac(efx);
3337 }
3338 
3339 static int efx_ef10_start_bist(struct efx_nic *efx, u32 bist_type)
3340 {
3341 	MCDI_DECLARE_BUF(inbuf, MC_CMD_START_BIST_IN_LEN);
3342 
3343 	MCDI_SET_DWORD(inbuf, START_BIST_IN_TYPE, bist_type);
3344 	return efx_mcdi_rpc(efx, MC_CMD_START_BIST, inbuf, sizeof(inbuf),
3345 			    NULL, 0, NULL);
3346 }
3347 
3348 /* MC BISTs follow a different poll mechanism to phy BISTs.
3349  * The BIST is done in the poll handler on the MC, and the MCDI command
3350  * will block until the BIST is done.
3351  */
3352 static int efx_ef10_poll_bist(struct efx_nic *efx)
3353 {
3354 	int rc;
3355 	MCDI_DECLARE_BUF(outbuf, MC_CMD_POLL_BIST_OUT_LEN);
3356 	size_t outlen;
3357 	u32 result;
3358 
3359 	rc = efx_mcdi_rpc(efx, MC_CMD_POLL_BIST, NULL, 0,
3360 			   outbuf, sizeof(outbuf), &outlen);
3361 	if (rc != 0)
3362 		return rc;
3363 
3364 	if (outlen < MC_CMD_POLL_BIST_OUT_LEN)
3365 		return -EIO;
3366 
3367 	result = MCDI_DWORD(outbuf, POLL_BIST_OUT_RESULT);
3368 	switch (result) {
3369 	case MC_CMD_POLL_BIST_PASSED:
3370 		netif_dbg(efx, hw, efx->net_dev, "BIST passed.\n");
3371 		return 0;
3372 	case MC_CMD_POLL_BIST_TIMEOUT:
3373 		netif_err(efx, hw, efx->net_dev, "BIST timed out\n");
3374 		return -EIO;
3375 	case MC_CMD_POLL_BIST_FAILED:
3376 		netif_err(efx, hw, efx->net_dev, "BIST failed.\n");
3377 		return -EIO;
3378 	default:
3379 		netif_err(efx, hw, efx->net_dev,
3380 			  "BIST returned unknown result %u", result);
3381 		return -EIO;
3382 	}
3383 }
3384 
3385 static int efx_ef10_run_bist(struct efx_nic *efx, u32 bist_type)
3386 {
3387 	int rc;
3388 
3389 	netif_dbg(efx, drv, efx->net_dev, "starting BIST type %u\n", bist_type);
3390 
3391 	rc = efx_ef10_start_bist(efx, bist_type);
3392 	if (rc != 0)
3393 		return rc;
3394 
3395 	return efx_ef10_poll_bist(efx);
3396 }
3397 
3398 static int
3399 efx_ef10_test_chip(struct efx_nic *efx, struct efx_self_tests *tests)
3400 {
3401 	int rc, rc2;
3402 
3403 	efx_reset_down(efx, RESET_TYPE_WORLD);
3404 
3405 	rc = efx_mcdi_rpc(efx, MC_CMD_ENABLE_OFFLINE_BIST,
3406 			  NULL, 0, NULL, 0, NULL);
3407 	if (rc != 0)
3408 		goto out;
3409 
3410 	tests->memory = efx_ef10_run_bist(efx, MC_CMD_MC_MEM_BIST) ? -1 : 1;
3411 	tests->registers = efx_ef10_run_bist(efx, MC_CMD_REG_BIST) ? -1 : 1;
3412 
3413 	rc = efx_mcdi_reset(efx, RESET_TYPE_WORLD);
3414 
3415 out:
3416 	if (rc == -EPERM)
3417 		rc = 0;
3418 	rc2 = efx_reset_up(efx, RESET_TYPE_WORLD, rc == 0);
3419 	return rc ? rc : rc2;
3420 }
3421 
3422 #ifdef CONFIG_SFC_MTD
3423 
3424 struct efx_ef10_nvram_type_info {
3425 	u16 type, type_mask;
3426 	u8 port;
3427 	const char *name;
3428 };
3429 
3430 static const struct efx_ef10_nvram_type_info efx_ef10_nvram_types[] = {
3431 	{ NVRAM_PARTITION_TYPE_MC_FIRMWARE,	   0,    0, "sfc_mcfw" },
3432 	{ NVRAM_PARTITION_TYPE_MC_FIRMWARE_BACKUP, 0,    0, "sfc_mcfw_backup" },
3433 	{ NVRAM_PARTITION_TYPE_EXPANSION_ROM,	   0,    0, "sfc_exp_rom" },
3434 	{ NVRAM_PARTITION_TYPE_STATIC_CONFIG,	   0,    0, "sfc_static_cfg" },
3435 	{ NVRAM_PARTITION_TYPE_DYNAMIC_CONFIG,	   0,    0, "sfc_dynamic_cfg" },
3436 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT0, 0,   0, "sfc_exp_rom_cfg" },
3437 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT1, 0,   1, "sfc_exp_rom_cfg" },
3438 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT2, 0,   2, "sfc_exp_rom_cfg" },
3439 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT3, 0,   3, "sfc_exp_rom_cfg" },
3440 	{ NVRAM_PARTITION_TYPE_LICENSE,		   0,    0, "sfc_license" },
3441 	{ NVRAM_PARTITION_TYPE_PHY_MIN,		   0xff, 0, "sfc_phy_fw" },
3442 	{ NVRAM_PARTITION_TYPE_MUM_FIRMWARE,	   0,    0, "sfc_mumfw" },
3443 	{ NVRAM_PARTITION_TYPE_EXPANSION_UEFI,	   0,    0, "sfc_uefi" },
3444 	{ NVRAM_PARTITION_TYPE_DYNCONFIG_DEFAULTS, 0,    0, "sfc_dynamic_cfg_dflt" },
3445 	{ NVRAM_PARTITION_TYPE_ROMCONFIG_DEFAULTS, 0,    0, "sfc_exp_rom_cfg_dflt" },
3446 	{ NVRAM_PARTITION_TYPE_STATUS,		   0,    0, "sfc_status" },
3447 	{ NVRAM_PARTITION_TYPE_BUNDLE,		   0,    0, "sfc_bundle" },
3448 	{ NVRAM_PARTITION_TYPE_BUNDLE_METADATA,	   0,    0, "sfc_bundle_metadata" },
3449 };
3450 #define EF10_NVRAM_PARTITION_COUNT	ARRAY_SIZE(efx_ef10_nvram_types)
3451 
3452 static int efx_ef10_mtd_probe_partition(struct efx_nic *efx,
3453 					struct efx_mcdi_mtd_partition *part,
3454 					unsigned int type,
3455 					unsigned long *found)
3456 {
3457 	MCDI_DECLARE_BUF(inbuf, MC_CMD_NVRAM_METADATA_IN_LEN);
3458 	MCDI_DECLARE_BUF(outbuf, MC_CMD_NVRAM_METADATA_OUT_LENMAX);
3459 	const struct efx_ef10_nvram_type_info *info;
3460 	size_t size, erase_size, outlen;
3461 	int type_idx = 0;
3462 	bool protected;
3463 	int rc;
3464 
3465 	for (type_idx = 0; ; type_idx++) {
3466 		if (type_idx == EF10_NVRAM_PARTITION_COUNT)
3467 			return -ENODEV;
3468 		info = efx_ef10_nvram_types + type_idx;
3469 		if ((type & ~info->type_mask) == info->type)
3470 			break;
3471 	}
3472 	if (info->port != efx_port_num(efx))
3473 		return -ENODEV;
3474 
3475 	rc = efx_mcdi_nvram_info(efx, type, &size, &erase_size, &protected);
3476 	if (rc)
3477 		return rc;
3478 	if (protected &&
3479 	    (type != NVRAM_PARTITION_TYPE_DYNCONFIG_DEFAULTS &&
3480 	     type != NVRAM_PARTITION_TYPE_ROMCONFIG_DEFAULTS))
3481 		/* Hide protected partitions that don't provide defaults. */
3482 		return -ENODEV;
3483 
3484 	if (protected)
3485 		/* Protected partitions are read only. */
3486 		erase_size = 0;
3487 
3488 	/* If we've already exposed a partition of this type, hide this
3489 	 * duplicate.  All operations on MTDs are keyed by the type anyway,
3490 	 * so we can't act on the duplicate.
3491 	 */
3492 	if (__test_and_set_bit(type_idx, found))
3493 		return -EEXIST;
3494 
3495 	part->nvram_type = type;
3496 
3497 	MCDI_SET_DWORD(inbuf, NVRAM_METADATA_IN_TYPE, type);
3498 	rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_METADATA, inbuf, sizeof(inbuf),
3499 			  outbuf, sizeof(outbuf), &outlen);
3500 	if (rc)
3501 		return rc;
3502 	if (outlen < MC_CMD_NVRAM_METADATA_OUT_LENMIN)
3503 		return -EIO;
3504 	if (MCDI_DWORD(outbuf, NVRAM_METADATA_OUT_FLAGS) &
3505 	    (1 << MC_CMD_NVRAM_METADATA_OUT_SUBTYPE_VALID_LBN))
3506 		part->fw_subtype = MCDI_DWORD(outbuf,
3507 					      NVRAM_METADATA_OUT_SUBTYPE);
3508 
3509 	part->common.dev_type_name = "EF10 NVRAM manager";
3510 	part->common.type_name = info->name;
3511 
3512 	part->common.mtd.type = MTD_NORFLASH;
3513 	part->common.mtd.flags = MTD_CAP_NORFLASH;
3514 	part->common.mtd.size = size;
3515 	part->common.mtd.erasesize = erase_size;
3516 	/* sfc_status is read-only */
3517 	if (!erase_size)
3518 		part->common.mtd.flags |= MTD_NO_ERASE;
3519 
3520 	return 0;
3521 }
3522 
3523 static int efx_ef10_mtd_probe(struct efx_nic *efx)
3524 {
3525 	MCDI_DECLARE_BUF(outbuf, MC_CMD_NVRAM_PARTITIONS_OUT_LENMAX);
3526 	DECLARE_BITMAP(found, EF10_NVRAM_PARTITION_COUNT) = { 0 };
3527 	struct efx_mcdi_mtd_partition *parts;
3528 	size_t outlen, n_parts_total, i, n_parts;
3529 	unsigned int type;
3530 	int rc;
3531 
3532 	ASSERT_RTNL();
3533 
3534 	BUILD_BUG_ON(MC_CMD_NVRAM_PARTITIONS_IN_LEN != 0);
3535 	rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_PARTITIONS, NULL, 0,
3536 			  outbuf, sizeof(outbuf), &outlen);
3537 	if (rc)
3538 		return rc;
3539 	if (outlen < MC_CMD_NVRAM_PARTITIONS_OUT_LENMIN)
3540 		return -EIO;
3541 
3542 	n_parts_total = MCDI_DWORD(outbuf, NVRAM_PARTITIONS_OUT_NUM_PARTITIONS);
3543 	if (n_parts_total >
3544 	    MCDI_VAR_ARRAY_LEN(outlen, NVRAM_PARTITIONS_OUT_TYPE_ID))
3545 		return -EIO;
3546 
3547 	parts = kcalloc(n_parts_total, sizeof(*parts), GFP_KERNEL);
3548 	if (!parts)
3549 		return -ENOMEM;
3550 
3551 	n_parts = 0;
3552 	for (i = 0; i < n_parts_total; i++) {
3553 		type = MCDI_ARRAY_DWORD(outbuf, NVRAM_PARTITIONS_OUT_TYPE_ID,
3554 					i);
3555 		rc = efx_ef10_mtd_probe_partition(efx, &parts[n_parts], type,
3556 						  found);
3557 		if (rc == -EEXIST || rc == -ENODEV)
3558 			continue;
3559 		if (rc)
3560 			goto fail;
3561 		n_parts++;
3562 	}
3563 
3564 	rc = efx_mtd_add(efx, &parts[0].common, n_parts, sizeof(*parts));
3565 fail:
3566 	if (rc)
3567 		kfree(parts);
3568 	return rc;
3569 }
3570 
3571 #endif /* CONFIG_SFC_MTD */
3572 
3573 static void efx_ef10_ptp_write_host_time(struct efx_nic *efx, u32 host_time)
3574 {
3575 	_efx_writed(efx, cpu_to_le32(host_time), ER_DZ_MC_DB_LWRD);
3576 }
3577 
3578 static void efx_ef10_ptp_write_host_time_vf(struct efx_nic *efx,
3579 					    u32 host_time) {}
3580 
3581 static int efx_ef10_rx_enable_timestamping(struct efx_channel *channel,
3582 					   bool temp)
3583 {
3584 	MCDI_DECLARE_BUF(inbuf, MC_CMD_PTP_IN_TIME_EVENT_SUBSCRIBE_LEN);
3585 	int rc;
3586 
3587 	if (channel->sync_events_state == SYNC_EVENTS_REQUESTED ||
3588 	    channel->sync_events_state == SYNC_EVENTS_VALID ||
3589 	    (temp && channel->sync_events_state == SYNC_EVENTS_DISABLED))
3590 		return 0;
3591 	channel->sync_events_state = SYNC_EVENTS_REQUESTED;
3592 
3593 	MCDI_SET_DWORD(inbuf, PTP_IN_OP, MC_CMD_PTP_OP_TIME_EVENT_SUBSCRIBE);
3594 	MCDI_SET_DWORD(inbuf, PTP_IN_PERIPH_ID, 0);
3595 	MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_SUBSCRIBE_QUEUE,
3596 		       channel->channel);
3597 
3598 	rc = efx_mcdi_rpc(channel->efx, MC_CMD_PTP,
3599 			  inbuf, sizeof(inbuf), NULL, 0, NULL);
3600 
3601 	if (rc != 0)
3602 		channel->sync_events_state = temp ? SYNC_EVENTS_QUIESCENT :
3603 						    SYNC_EVENTS_DISABLED;
3604 
3605 	return rc;
3606 }
3607 
3608 static int efx_ef10_rx_disable_timestamping(struct efx_channel *channel,
3609 					    bool temp)
3610 {
3611 	MCDI_DECLARE_BUF(inbuf, MC_CMD_PTP_IN_TIME_EVENT_UNSUBSCRIBE_LEN);
3612 	int rc;
3613 
3614 	if (channel->sync_events_state == SYNC_EVENTS_DISABLED ||
3615 	    (temp && channel->sync_events_state == SYNC_EVENTS_QUIESCENT))
3616 		return 0;
3617 	if (channel->sync_events_state == SYNC_EVENTS_QUIESCENT) {
3618 		channel->sync_events_state = SYNC_EVENTS_DISABLED;
3619 		return 0;
3620 	}
3621 	channel->sync_events_state = temp ? SYNC_EVENTS_QUIESCENT :
3622 					    SYNC_EVENTS_DISABLED;
3623 
3624 	MCDI_SET_DWORD(inbuf, PTP_IN_OP, MC_CMD_PTP_OP_TIME_EVENT_UNSUBSCRIBE);
3625 	MCDI_SET_DWORD(inbuf, PTP_IN_PERIPH_ID, 0);
3626 	MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_UNSUBSCRIBE_CONTROL,
3627 		       MC_CMD_PTP_IN_TIME_EVENT_UNSUBSCRIBE_SINGLE);
3628 	MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_UNSUBSCRIBE_QUEUE,
3629 		       channel->channel);
3630 
3631 	rc = efx_mcdi_rpc(channel->efx, MC_CMD_PTP,
3632 			  inbuf, sizeof(inbuf), NULL, 0, NULL);
3633 
3634 	return rc;
3635 }
3636 
3637 static int efx_ef10_ptp_set_ts_sync_events(struct efx_nic *efx, bool en,
3638 					   bool temp)
3639 {
3640 	int (*set)(struct efx_channel *channel, bool temp);
3641 	struct efx_channel *channel;
3642 
3643 	set = en ?
3644 	      efx_ef10_rx_enable_timestamping :
3645 	      efx_ef10_rx_disable_timestamping;
3646 
3647 	channel = efx_ptp_channel(efx);
3648 	if (channel) {
3649 		int rc = set(channel, temp);
3650 		if (en && rc != 0) {
3651 			efx_ef10_ptp_set_ts_sync_events(efx, false, temp);
3652 			return rc;
3653 		}
3654 	}
3655 
3656 	return 0;
3657 }
3658 
3659 static int efx_ef10_ptp_set_ts_config_vf(struct efx_nic *efx,
3660 					 struct hwtstamp_config *init)
3661 {
3662 	return -EOPNOTSUPP;
3663 }
3664 
3665 static int efx_ef10_ptp_set_ts_config(struct efx_nic *efx,
3666 				      struct hwtstamp_config *init)
3667 {
3668 	int rc;
3669 
3670 	switch (init->rx_filter) {
3671 	case HWTSTAMP_FILTER_NONE:
3672 		efx_ef10_ptp_set_ts_sync_events(efx, false, false);
3673 		/* if TX timestamping is still requested then leave PTP on */
3674 		return efx_ptp_change_mode(efx,
3675 					   init->tx_type != HWTSTAMP_TX_OFF, 0);
3676 	case HWTSTAMP_FILTER_ALL:
3677 	case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
3678 	case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
3679 	case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
3680 	case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
3681 	case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
3682 	case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
3683 	case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
3684 	case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
3685 	case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
3686 	case HWTSTAMP_FILTER_PTP_V2_EVENT:
3687 	case HWTSTAMP_FILTER_PTP_V2_SYNC:
3688 	case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
3689 	case HWTSTAMP_FILTER_NTP_ALL:
3690 		init->rx_filter = HWTSTAMP_FILTER_ALL;
3691 		rc = efx_ptp_change_mode(efx, true, 0);
3692 		if (!rc)
3693 			rc = efx_ef10_ptp_set_ts_sync_events(efx, true, false);
3694 		if (rc)
3695 			efx_ptp_change_mode(efx, false, 0);
3696 		return rc;
3697 	default:
3698 		return -ERANGE;
3699 	}
3700 }
3701 
3702 static int efx_ef10_get_phys_port_id(struct efx_nic *efx,
3703 				     struct netdev_phys_item_id *ppid)
3704 {
3705 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
3706 
3707 	if (!is_valid_ether_addr(nic_data->port_id))
3708 		return -EOPNOTSUPP;
3709 
3710 	ppid->id_len = ETH_ALEN;
3711 	memcpy(ppid->id, nic_data->port_id, ppid->id_len);
3712 
3713 	return 0;
3714 }
3715 
3716 static int efx_ef10_vlan_rx_add_vid(struct efx_nic *efx, __be16 proto, u16 vid)
3717 {
3718 	if (proto != htons(ETH_P_8021Q))
3719 		return -EINVAL;
3720 
3721 	return efx_ef10_add_vlan(efx, vid);
3722 }
3723 
3724 static int efx_ef10_vlan_rx_kill_vid(struct efx_nic *efx, __be16 proto, u16 vid)
3725 {
3726 	if (proto != htons(ETH_P_8021Q))
3727 		return -EINVAL;
3728 
3729 	return efx_ef10_del_vlan(efx, vid);
3730 }
3731 
3732 /* We rely on the MCDI wiping out our TX rings if it made any changes to the
3733  * ports table, ensuring that any TSO descriptors that were made on a now-
3734  * removed tunnel port will be blown away and won't break things when we try
3735  * to transmit them using the new ports table.
3736  */
3737 static int efx_ef10_set_udp_tnl_ports(struct efx_nic *efx, bool unloading)
3738 {
3739 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
3740 	MCDI_DECLARE_BUF(inbuf, MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_IN_LENMAX);
3741 	MCDI_DECLARE_BUF(outbuf, MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_OUT_LEN);
3742 	bool will_reset = false;
3743 	size_t num_entries = 0;
3744 	size_t inlen, outlen;
3745 	size_t i;
3746 	int rc;
3747 	efx_dword_t flags_and_num_entries;
3748 
3749 	WARN_ON(!mutex_is_locked(&nic_data->udp_tunnels_lock));
3750 
3751 	nic_data->udp_tunnels_dirty = false;
3752 
3753 	if (!(nic_data->datapath_caps &
3754 	    (1 << MC_CMD_GET_CAPABILITIES_OUT_VXLAN_NVGRE_LBN))) {
3755 		efx_device_attach_if_not_resetting(efx);
3756 		return 0;
3757 	}
3758 
3759 	BUILD_BUG_ON(ARRAY_SIZE(nic_data->udp_tunnels) >
3760 		     MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_IN_ENTRIES_MAXNUM);
3761 
3762 	for (i = 0; i < ARRAY_SIZE(nic_data->udp_tunnels); ++i) {
3763 		if (nic_data->udp_tunnels[i].type !=
3764 		    TUNNEL_ENCAP_UDP_PORT_ENTRY_INVALID) {
3765 			efx_dword_t entry;
3766 
3767 			EFX_POPULATE_DWORD_2(entry,
3768 				TUNNEL_ENCAP_UDP_PORT_ENTRY_UDP_PORT,
3769 					ntohs(nic_data->udp_tunnels[i].port),
3770 				TUNNEL_ENCAP_UDP_PORT_ENTRY_PROTOCOL,
3771 					nic_data->udp_tunnels[i].type);
3772 			*_MCDI_ARRAY_DWORD(inbuf,
3773 				SET_TUNNEL_ENCAP_UDP_PORTS_IN_ENTRIES,
3774 				num_entries++) = entry;
3775 		}
3776 	}
3777 
3778 	BUILD_BUG_ON((MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_IN_NUM_ENTRIES_OFST -
3779 		      MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_IN_FLAGS_OFST) * 8 !=
3780 		     EFX_WORD_1_LBN);
3781 	BUILD_BUG_ON(MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_IN_NUM_ENTRIES_LEN * 8 !=
3782 		     EFX_WORD_1_WIDTH);
3783 	EFX_POPULATE_DWORD_2(flags_and_num_entries,
3784 			     MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_IN_UNLOADING,
3785 				!!unloading,
3786 			     EFX_WORD_1, num_entries);
3787 	*_MCDI_DWORD(inbuf, SET_TUNNEL_ENCAP_UDP_PORTS_IN_FLAGS) =
3788 		flags_and_num_entries;
3789 
3790 	inlen = MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_IN_LEN(num_entries);
3791 
3792 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS,
3793 				inbuf, inlen, outbuf, sizeof(outbuf), &outlen);
3794 	if (rc == -EIO) {
3795 		/* Most likely the MC rebooted due to another function also
3796 		 * setting its tunnel port list. Mark the tunnel port list as
3797 		 * dirty, so it will be pushed upon coming up from the reboot.
3798 		 */
3799 		nic_data->udp_tunnels_dirty = true;
3800 		return 0;
3801 	}
3802 
3803 	if (rc) {
3804 		/* expected not available on unprivileged functions */
3805 		if (rc != -EPERM)
3806 			netif_warn(efx, drv, efx->net_dev,
3807 				   "Unable to set UDP tunnel ports; rc=%d.\n", rc);
3808 	} else if (MCDI_DWORD(outbuf, SET_TUNNEL_ENCAP_UDP_PORTS_OUT_FLAGS) &
3809 		   (1 << MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_OUT_RESETTING_LBN)) {
3810 		netif_info(efx, drv, efx->net_dev,
3811 			   "Rebooting MC due to UDP tunnel port list change\n");
3812 		will_reset = true;
3813 		if (unloading)
3814 			/* Delay for the MC reset to complete. This will make
3815 			 * unloading other functions a bit smoother. This is a
3816 			 * race, but the other unload will work whichever way
3817 			 * it goes, this just avoids an unnecessary error
3818 			 * message.
3819 			 */
3820 			msleep(100);
3821 	}
3822 	if (!will_reset && !unloading) {
3823 		/* The caller will have detached, relying on the MC reset to
3824 		 * trigger a re-attach.  Since there won't be an MC reset, we
3825 		 * have to do the attach ourselves.
3826 		 */
3827 		efx_device_attach_if_not_resetting(efx);
3828 	}
3829 
3830 	return rc;
3831 }
3832 
3833 static int efx_ef10_udp_tnl_push_ports(struct efx_nic *efx)
3834 {
3835 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
3836 	int rc = 0;
3837 
3838 	mutex_lock(&nic_data->udp_tunnels_lock);
3839 	if (nic_data->udp_tunnels_dirty) {
3840 		/* Make sure all TX are stopped while we modify the table, else
3841 		 * we might race against an efx_features_check().
3842 		 */
3843 		efx_device_detach_sync(efx);
3844 		rc = efx_ef10_set_udp_tnl_ports(efx, false);
3845 	}
3846 	mutex_unlock(&nic_data->udp_tunnels_lock);
3847 	return rc;
3848 }
3849 
3850 static int efx_ef10_udp_tnl_set_port(struct net_device *dev,
3851 				     unsigned int table, unsigned int entry,
3852 				     struct udp_tunnel_info *ti)
3853 {
3854 	struct efx_nic *efx = netdev_priv(dev);
3855 	struct efx_ef10_nic_data *nic_data;
3856 	int efx_tunnel_type, rc;
3857 
3858 	if (ti->type == UDP_TUNNEL_TYPE_VXLAN)
3859 		efx_tunnel_type = TUNNEL_ENCAP_UDP_PORT_ENTRY_VXLAN;
3860 	else
3861 		efx_tunnel_type = TUNNEL_ENCAP_UDP_PORT_ENTRY_GENEVE;
3862 
3863 	nic_data = efx->nic_data;
3864 	if (!(nic_data->datapath_caps &
3865 	      (1 << MC_CMD_GET_CAPABILITIES_OUT_VXLAN_NVGRE_LBN)))
3866 		return -EOPNOTSUPP;
3867 
3868 	mutex_lock(&nic_data->udp_tunnels_lock);
3869 	/* Make sure all TX are stopped while we add to the table, else we
3870 	 * might race against an efx_features_check().
3871 	 */
3872 	efx_device_detach_sync(efx);
3873 	nic_data->udp_tunnels[entry].type = efx_tunnel_type;
3874 	nic_data->udp_tunnels[entry].port = ti->port;
3875 	rc = efx_ef10_set_udp_tnl_ports(efx, false);
3876 	mutex_unlock(&nic_data->udp_tunnels_lock);
3877 
3878 	return rc;
3879 }
3880 
3881 /* Called under the TX lock with the TX queue running, hence no-one can be
3882  * in the middle of updating the UDP tunnels table.  However, they could
3883  * have tried and failed the MCDI, in which case they'll have set the dirty
3884  * flag before dropping their locks.
3885  */
3886 static bool efx_ef10_udp_tnl_has_port(struct efx_nic *efx, __be16 port)
3887 {
3888 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
3889 	size_t i;
3890 
3891 	if (!(nic_data->datapath_caps &
3892 	      (1 << MC_CMD_GET_CAPABILITIES_OUT_VXLAN_NVGRE_LBN)))
3893 		return false;
3894 
3895 	if (nic_data->udp_tunnels_dirty)
3896 		/* SW table may not match HW state, so just assume we can't
3897 		 * use any UDP tunnel offloads.
3898 		 */
3899 		return false;
3900 
3901 	for (i = 0; i < ARRAY_SIZE(nic_data->udp_tunnels); ++i)
3902 		if (nic_data->udp_tunnels[i].type !=
3903 		    TUNNEL_ENCAP_UDP_PORT_ENTRY_INVALID &&
3904 		    nic_data->udp_tunnels[i].port == port)
3905 			return true;
3906 
3907 	return false;
3908 }
3909 
3910 static int efx_ef10_udp_tnl_unset_port(struct net_device *dev,
3911 				       unsigned int table, unsigned int entry,
3912 				       struct udp_tunnel_info *ti)
3913 {
3914 	struct efx_nic *efx = netdev_priv(dev);
3915 	struct efx_ef10_nic_data *nic_data;
3916 	int rc;
3917 
3918 	nic_data = efx->nic_data;
3919 
3920 	mutex_lock(&nic_data->udp_tunnels_lock);
3921 	/* Make sure all TX are stopped while we remove from the table, else we
3922 	 * might race against an efx_features_check().
3923 	 */
3924 	efx_device_detach_sync(efx);
3925 	nic_data->udp_tunnels[entry].type = TUNNEL_ENCAP_UDP_PORT_ENTRY_INVALID;
3926 	nic_data->udp_tunnels[entry].port = 0;
3927 	rc = efx_ef10_set_udp_tnl_ports(efx, false);
3928 	mutex_unlock(&nic_data->udp_tunnels_lock);
3929 
3930 	return rc;
3931 }
3932 
3933 static const struct udp_tunnel_nic_info efx_ef10_udp_tunnels = {
3934 	.set_port	= efx_ef10_udp_tnl_set_port,
3935 	.unset_port	= efx_ef10_udp_tnl_unset_port,
3936 	.flags          = UDP_TUNNEL_NIC_INFO_MAY_SLEEP,
3937 	.tables         = {
3938 		{
3939 			.n_entries = 16,
3940 			.tunnel_types = UDP_TUNNEL_TYPE_VXLAN |
3941 					UDP_TUNNEL_TYPE_GENEVE,
3942 		},
3943 	},
3944 };
3945 
3946 /* EF10 may have multiple datapath firmware variants within a
3947  * single version.  Report which variants are running.
3948  */
3949 static size_t efx_ef10_print_additional_fwver(struct efx_nic *efx, char *buf,
3950 					      size_t len)
3951 {
3952 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
3953 
3954 	return scnprintf(buf, len, " rx%x tx%x",
3955 			 nic_data->rx_dpcpu_fw_id,
3956 			 nic_data->tx_dpcpu_fw_id);
3957 }
3958 
3959 static unsigned int ef10_check_caps(const struct efx_nic *efx,
3960 				    u8 flag,
3961 				    u32 offset)
3962 {
3963 	const struct efx_ef10_nic_data *nic_data = efx->nic_data;
3964 
3965 	switch (offset) {
3966 	case(MC_CMD_GET_CAPABILITIES_V4_OUT_FLAGS1_OFST):
3967 		return nic_data->datapath_caps & BIT_ULL(flag);
3968 	case(MC_CMD_GET_CAPABILITIES_V4_OUT_FLAGS2_OFST):
3969 		return nic_data->datapath_caps2 & BIT_ULL(flag);
3970 	default:
3971 		return 0;
3972 	}
3973 }
3974 
3975 #define EF10_OFFLOAD_FEATURES		\
3976 	(NETIF_F_IP_CSUM |		\
3977 	 NETIF_F_HW_VLAN_CTAG_FILTER |	\
3978 	 NETIF_F_IPV6_CSUM |		\
3979 	 NETIF_F_RXHASH |		\
3980 	 NETIF_F_NTUPLE)
3981 
3982 const struct efx_nic_type efx_hunt_a0_vf_nic_type = {
3983 	.is_vf = true,
3984 	.mem_bar = efx_ef10_vf_mem_bar,
3985 	.mem_map_size = efx_ef10_mem_map_size,
3986 	.probe = efx_ef10_probe_vf,
3987 	.remove = efx_ef10_remove,
3988 	.dimension_resources = efx_ef10_dimension_resources,
3989 	.init = efx_ef10_init_nic,
3990 	.fini = efx_ef10_fini_nic,
3991 	.map_reset_reason = efx_ef10_map_reset_reason,
3992 	.map_reset_flags = efx_ef10_map_reset_flags,
3993 	.reset = efx_ef10_reset,
3994 	.probe_port = efx_mcdi_port_probe,
3995 	.remove_port = efx_mcdi_port_remove,
3996 	.fini_dmaq = efx_fini_dmaq,
3997 	.prepare_flr = efx_ef10_prepare_flr,
3998 	.finish_flr = efx_port_dummy_op_void,
3999 	.describe_stats = efx_ef10_describe_stats,
4000 	.update_stats = efx_ef10_update_stats_vf,
4001 	.start_stats = efx_port_dummy_op_void,
4002 	.pull_stats = efx_port_dummy_op_void,
4003 	.stop_stats = efx_port_dummy_op_void,
4004 	.push_irq_moderation = efx_ef10_push_irq_moderation,
4005 	.reconfigure_mac = efx_ef10_mac_reconfigure,
4006 	.check_mac_fault = efx_mcdi_mac_check_fault,
4007 	.reconfigure_port = efx_mcdi_port_reconfigure,
4008 	.get_wol = efx_ef10_get_wol_vf,
4009 	.set_wol = efx_ef10_set_wol_vf,
4010 	.resume_wol = efx_port_dummy_op_void,
4011 	.mcdi_request = efx_ef10_mcdi_request,
4012 	.mcdi_poll_response = efx_ef10_mcdi_poll_response,
4013 	.mcdi_read_response = efx_ef10_mcdi_read_response,
4014 	.mcdi_poll_reboot = efx_ef10_mcdi_poll_reboot,
4015 	.mcdi_reboot_detected = efx_ef10_mcdi_reboot_detected,
4016 	.irq_enable_master = efx_port_dummy_op_void,
4017 	.irq_test_generate = efx_ef10_irq_test_generate,
4018 	.irq_disable_non_ev = efx_port_dummy_op_void,
4019 	.irq_handle_msi = efx_ef10_msi_interrupt,
4020 	.irq_handle_legacy = efx_ef10_legacy_interrupt,
4021 	.tx_probe = efx_ef10_tx_probe,
4022 	.tx_init = efx_ef10_tx_init,
4023 	.tx_remove = efx_mcdi_tx_remove,
4024 	.tx_write = efx_ef10_tx_write,
4025 	.tx_limit_len = efx_ef10_tx_limit_len,
4026 	.tx_enqueue = __efx_enqueue_skb,
4027 	.rx_push_rss_config = efx_mcdi_vf_rx_push_rss_config,
4028 	.rx_pull_rss_config = efx_mcdi_rx_pull_rss_config,
4029 	.rx_probe = efx_mcdi_rx_probe,
4030 	.rx_init = efx_mcdi_rx_init,
4031 	.rx_remove = efx_mcdi_rx_remove,
4032 	.rx_write = efx_ef10_rx_write,
4033 	.rx_defer_refill = efx_ef10_rx_defer_refill,
4034 	.rx_packet = __efx_rx_packet,
4035 	.ev_probe = efx_mcdi_ev_probe,
4036 	.ev_init = efx_ef10_ev_init,
4037 	.ev_fini = efx_mcdi_ev_fini,
4038 	.ev_remove = efx_mcdi_ev_remove,
4039 	.ev_process = efx_ef10_ev_process,
4040 	.ev_read_ack = efx_ef10_ev_read_ack,
4041 	.ev_test_generate = efx_ef10_ev_test_generate,
4042 	.filter_table_probe = efx_ef10_filter_table_probe,
4043 	.filter_table_restore = efx_mcdi_filter_table_restore,
4044 	.filter_table_remove = efx_mcdi_filter_table_remove,
4045 	.filter_update_rx_scatter = efx_mcdi_update_rx_scatter,
4046 	.filter_insert = efx_mcdi_filter_insert,
4047 	.filter_remove_safe = efx_mcdi_filter_remove_safe,
4048 	.filter_get_safe = efx_mcdi_filter_get_safe,
4049 	.filter_clear_rx = efx_mcdi_filter_clear_rx,
4050 	.filter_count_rx_used = efx_mcdi_filter_count_rx_used,
4051 	.filter_get_rx_id_limit = efx_mcdi_filter_get_rx_id_limit,
4052 	.filter_get_rx_ids = efx_mcdi_filter_get_rx_ids,
4053 #ifdef CONFIG_RFS_ACCEL
4054 	.filter_rfs_expire_one = efx_mcdi_filter_rfs_expire_one,
4055 #endif
4056 #ifdef CONFIG_SFC_MTD
4057 	.mtd_probe = efx_port_dummy_op_int,
4058 #endif
4059 	.ptp_write_host_time = efx_ef10_ptp_write_host_time_vf,
4060 	.ptp_set_ts_config = efx_ef10_ptp_set_ts_config_vf,
4061 	.vlan_rx_add_vid = efx_ef10_vlan_rx_add_vid,
4062 	.vlan_rx_kill_vid = efx_ef10_vlan_rx_kill_vid,
4063 #ifdef CONFIG_SFC_SRIOV
4064 	.vswitching_probe = efx_ef10_vswitching_probe_vf,
4065 	.vswitching_restore = efx_ef10_vswitching_restore_vf,
4066 	.vswitching_remove = efx_ef10_vswitching_remove_vf,
4067 #endif
4068 	.get_mac_address = efx_ef10_get_mac_address_vf,
4069 	.set_mac_address = efx_ef10_set_mac_address,
4070 
4071 	.get_phys_port_id = efx_ef10_get_phys_port_id,
4072 	.revision = EFX_REV_HUNT_A0,
4073 	.max_dma_mask = DMA_BIT_MASK(ESF_DZ_TX_KER_BUF_ADDR_WIDTH),
4074 	.rx_prefix_size = ES_DZ_RX_PREFIX_SIZE,
4075 	.rx_hash_offset = ES_DZ_RX_PREFIX_HASH_OFST,
4076 	.rx_ts_offset = ES_DZ_RX_PREFIX_TSTAMP_OFST,
4077 	.can_rx_scatter = true,
4078 	.always_rx_scatter = true,
4079 	.min_interrupt_mode = EFX_INT_MODE_MSIX,
4080 	.timer_period_max = 1 << ERF_DD_EVQ_IND_TIMER_VAL_WIDTH,
4081 	.offload_features = EF10_OFFLOAD_FEATURES,
4082 	.mcdi_max_ver = 2,
4083 	.max_rx_ip_filters = EFX_MCDI_FILTER_TBL_ROWS,
4084 	.hwtstamp_filters = 1 << HWTSTAMP_FILTER_NONE |
4085 			    1 << HWTSTAMP_FILTER_ALL,
4086 	.rx_hash_key_size = 40,
4087 	.check_caps = ef10_check_caps,
4088 	.print_additional_fwver = efx_ef10_print_additional_fwver,
4089 	.sensor_event = efx_mcdi_sensor_event,
4090 };
4091 
4092 const struct efx_nic_type efx_hunt_a0_nic_type = {
4093 	.is_vf = false,
4094 	.mem_bar = efx_ef10_pf_mem_bar,
4095 	.mem_map_size = efx_ef10_mem_map_size,
4096 	.probe = efx_ef10_probe_pf,
4097 	.remove = efx_ef10_remove,
4098 	.dimension_resources = efx_ef10_dimension_resources,
4099 	.init = efx_ef10_init_nic,
4100 	.fini = efx_ef10_fini_nic,
4101 	.map_reset_reason = efx_ef10_map_reset_reason,
4102 	.map_reset_flags = efx_ef10_map_reset_flags,
4103 	.reset = efx_ef10_reset,
4104 	.probe_port = efx_mcdi_port_probe,
4105 	.remove_port = efx_mcdi_port_remove,
4106 	.fini_dmaq = efx_fini_dmaq,
4107 	.prepare_flr = efx_ef10_prepare_flr,
4108 	.finish_flr = efx_port_dummy_op_void,
4109 	.describe_stats = efx_ef10_describe_stats,
4110 	.update_stats = efx_ef10_update_stats_pf,
4111 	.start_stats = efx_mcdi_mac_start_stats,
4112 	.pull_stats = efx_mcdi_mac_pull_stats,
4113 	.stop_stats = efx_mcdi_mac_stop_stats,
4114 	.push_irq_moderation = efx_ef10_push_irq_moderation,
4115 	.reconfigure_mac = efx_ef10_mac_reconfigure,
4116 	.check_mac_fault = efx_mcdi_mac_check_fault,
4117 	.reconfigure_port = efx_mcdi_port_reconfigure,
4118 	.get_wol = efx_ef10_get_wol,
4119 	.set_wol = efx_ef10_set_wol,
4120 	.resume_wol = efx_port_dummy_op_void,
4121 	.test_chip = efx_ef10_test_chip,
4122 	.test_nvram = efx_mcdi_nvram_test_all,
4123 	.mcdi_request = efx_ef10_mcdi_request,
4124 	.mcdi_poll_response = efx_ef10_mcdi_poll_response,
4125 	.mcdi_read_response = efx_ef10_mcdi_read_response,
4126 	.mcdi_poll_reboot = efx_ef10_mcdi_poll_reboot,
4127 	.mcdi_reboot_detected = efx_ef10_mcdi_reboot_detected,
4128 	.irq_enable_master = efx_port_dummy_op_void,
4129 	.irq_test_generate = efx_ef10_irq_test_generate,
4130 	.irq_disable_non_ev = efx_port_dummy_op_void,
4131 	.irq_handle_msi = efx_ef10_msi_interrupt,
4132 	.irq_handle_legacy = efx_ef10_legacy_interrupt,
4133 	.tx_probe = efx_ef10_tx_probe,
4134 	.tx_init = efx_ef10_tx_init,
4135 	.tx_remove = efx_mcdi_tx_remove,
4136 	.tx_write = efx_ef10_tx_write,
4137 	.tx_limit_len = efx_ef10_tx_limit_len,
4138 	.tx_enqueue = __efx_enqueue_skb,
4139 	.rx_push_rss_config = efx_mcdi_pf_rx_push_rss_config,
4140 	.rx_pull_rss_config = efx_mcdi_rx_pull_rss_config,
4141 	.rx_push_rss_context_config = efx_mcdi_rx_push_rss_context_config,
4142 	.rx_pull_rss_context_config = efx_mcdi_rx_pull_rss_context_config,
4143 	.rx_restore_rss_contexts = efx_mcdi_rx_restore_rss_contexts,
4144 	.rx_probe = efx_mcdi_rx_probe,
4145 	.rx_init = efx_mcdi_rx_init,
4146 	.rx_remove = efx_mcdi_rx_remove,
4147 	.rx_write = efx_ef10_rx_write,
4148 	.rx_defer_refill = efx_ef10_rx_defer_refill,
4149 	.rx_packet = __efx_rx_packet,
4150 	.ev_probe = efx_mcdi_ev_probe,
4151 	.ev_init = efx_ef10_ev_init,
4152 	.ev_fini = efx_mcdi_ev_fini,
4153 	.ev_remove = efx_mcdi_ev_remove,
4154 	.ev_process = efx_ef10_ev_process,
4155 	.ev_read_ack = efx_ef10_ev_read_ack,
4156 	.ev_test_generate = efx_ef10_ev_test_generate,
4157 	.filter_table_probe = efx_ef10_filter_table_probe,
4158 	.filter_table_restore = efx_mcdi_filter_table_restore,
4159 	.filter_table_remove = efx_mcdi_filter_table_remove,
4160 	.filter_update_rx_scatter = efx_mcdi_update_rx_scatter,
4161 	.filter_insert = efx_mcdi_filter_insert,
4162 	.filter_remove_safe = efx_mcdi_filter_remove_safe,
4163 	.filter_get_safe = efx_mcdi_filter_get_safe,
4164 	.filter_clear_rx = efx_mcdi_filter_clear_rx,
4165 	.filter_count_rx_used = efx_mcdi_filter_count_rx_used,
4166 	.filter_get_rx_id_limit = efx_mcdi_filter_get_rx_id_limit,
4167 	.filter_get_rx_ids = efx_mcdi_filter_get_rx_ids,
4168 #ifdef CONFIG_RFS_ACCEL
4169 	.filter_rfs_expire_one = efx_mcdi_filter_rfs_expire_one,
4170 #endif
4171 #ifdef CONFIG_SFC_MTD
4172 	.mtd_probe = efx_ef10_mtd_probe,
4173 	.mtd_rename = efx_mcdi_mtd_rename,
4174 	.mtd_read = efx_mcdi_mtd_read,
4175 	.mtd_erase = efx_mcdi_mtd_erase,
4176 	.mtd_write = efx_mcdi_mtd_write,
4177 	.mtd_sync = efx_mcdi_mtd_sync,
4178 #endif
4179 	.ptp_write_host_time = efx_ef10_ptp_write_host_time,
4180 	.ptp_set_ts_sync_events = efx_ef10_ptp_set_ts_sync_events,
4181 	.ptp_set_ts_config = efx_ef10_ptp_set_ts_config,
4182 	.vlan_rx_add_vid = efx_ef10_vlan_rx_add_vid,
4183 	.vlan_rx_kill_vid = efx_ef10_vlan_rx_kill_vid,
4184 	.udp_tnl_push_ports = efx_ef10_udp_tnl_push_ports,
4185 	.udp_tnl_has_port = efx_ef10_udp_tnl_has_port,
4186 #ifdef CONFIG_SFC_SRIOV
4187 	.sriov_configure = efx_ef10_sriov_configure,
4188 	.sriov_init = efx_ef10_sriov_init,
4189 	.sriov_fini = efx_ef10_sriov_fini,
4190 	.sriov_wanted = efx_ef10_sriov_wanted,
4191 	.sriov_reset = efx_ef10_sriov_reset,
4192 	.sriov_flr = efx_ef10_sriov_flr,
4193 	.sriov_set_vf_mac = efx_ef10_sriov_set_vf_mac,
4194 	.sriov_set_vf_vlan = efx_ef10_sriov_set_vf_vlan,
4195 	.sriov_set_vf_spoofchk = efx_ef10_sriov_set_vf_spoofchk,
4196 	.sriov_get_vf_config = efx_ef10_sriov_get_vf_config,
4197 	.sriov_set_vf_link_state = efx_ef10_sriov_set_vf_link_state,
4198 	.vswitching_probe = efx_ef10_vswitching_probe_pf,
4199 	.vswitching_restore = efx_ef10_vswitching_restore_pf,
4200 	.vswitching_remove = efx_ef10_vswitching_remove_pf,
4201 #endif
4202 	.get_mac_address = efx_ef10_get_mac_address_pf,
4203 	.set_mac_address = efx_ef10_set_mac_address,
4204 	.tso_versions = efx_ef10_tso_versions,
4205 
4206 	.get_phys_port_id = efx_ef10_get_phys_port_id,
4207 	.revision = EFX_REV_HUNT_A0,
4208 	.max_dma_mask = DMA_BIT_MASK(ESF_DZ_TX_KER_BUF_ADDR_WIDTH),
4209 	.rx_prefix_size = ES_DZ_RX_PREFIX_SIZE,
4210 	.rx_hash_offset = ES_DZ_RX_PREFIX_HASH_OFST,
4211 	.rx_ts_offset = ES_DZ_RX_PREFIX_TSTAMP_OFST,
4212 	.can_rx_scatter = true,
4213 	.always_rx_scatter = true,
4214 	.option_descriptors = true,
4215 	.min_interrupt_mode = EFX_INT_MODE_LEGACY,
4216 	.timer_period_max = 1 << ERF_DD_EVQ_IND_TIMER_VAL_WIDTH,
4217 	.offload_features = EF10_OFFLOAD_FEATURES,
4218 	.mcdi_max_ver = 2,
4219 	.max_rx_ip_filters = EFX_MCDI_FILTER_TBL_ROWS,
4220 	.hwtstamp_filters = 1 << HWTSTAMP_FILTER_NONE |
4221 			    1 << HWTSTAMP_FILTER_ALL,
4222 	.rx_hash_key_size = 40,
4223 	.check_caps = ef10_check_caps,
4224 	.print_additional_fwver = efx_ef10_print_additional_fwver,
4225 	.sensor_event = efx_mcdi_sensor_event,
4226 };
4227