xref: /openbmc/linux/drivers/net/ethernet/sfc/ef10.c (revision 3bf90eca)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /****************************************************************************
3  * Driver for Solarflare network controllers and boards
4  * Copyright 2012-2013 Solarflare Communications Inc.
5  */
6 
7 #include "net_driver.h"
8 #include "rx_common.h"
9 #include "tx_common.h"
10 #include "ef10_regs.h"
11 #include "io.h"
12 #include "mcdi.h"
13 #include "mcdi_pcol.h"
14 #include "mcdi_port.h"
15 #include "mcdi_port_common.h"
16 #include "mcdi_functions.h"
17 #include "nic.h"
18 #include "mcdi_filters.h"
19 #include "workarounds.h"
20 #include "selftest.h"
21 #include "ef10_sriov.h"
22 #include <linux/in.h>
23 #include <linux/jhash.h>
24 #include <linux/wait.h>
25 #include <linux/workqueue.h>
26 #include <net/udp_tunnel.h>
27 
28 /* Hardware control for EF10 architecture including 'Huntington'. */
29 
30 #define EFX_EF10_DRVGEN_EV		7
31 enum {
32 	EFX_EF10_TEST = 1,
33 	EFX_EF10_REFILL,
34 };
35 
36 /* VLAN list entry */
37 struct efx_ef10_vlan {
38 	struct list_head list;
39 	u16 vid;
40 };
41 
42 static int efx_ef10_set_udp_tnl_ports(struct efx_nic *efx, bool unloading);
43 static const struct udp_tunnel_nic_info efx_ef10_udp_tunnels;
44 
45 static int efx_ef10_get_warm_boot_count(struct efx_nic *efx)
46 {
47 	efx_dword_t reg;
48 
49 	efx_readd(efx, &reg, ER_DZ_BIU_MC_SFT_STATUS);
50 	return EFX_DWORD_FIELD(reg, EFX_WORD_1) == 0xb007 ?
51 		EFX_DWORD_FIELD(reg, EFX_WORD_0) : -EIO;
52 }
53 
54 /* On all EF10s up to and including SFC9220 (Medford1), all PFs use BAR 0 for
55  * I/O space and BAR 2(&3) for memory.  On SFC9250 (Medford2), there is no I/O
56  * bar; PFs use BAR 0/1 for memory.
57  */
58 static unsigned int efx_ef10_pf_mem_bar(struct efx_nic *efx)
59 {
60 	switch (efx->pci_dev->device) {
61 	case 0x0b03: /* SFC9250 PF */
62 		return 0;
63 	default:
64 		return 2;
65 	}
66 }
67 
68 /* All VFs use BAR 0/1 for memory */
69 static unsigned int efx_ef10_vf_mem_bar(struct efx_nic *efx)
70 {
71 	return 0;
72 }
73 
74 static unsigned int efx_ef10_mem_map_size(struct efx_nic *efx)
75 {
76 	int bar;
77 
78 	bar = efx->type->mem_bar(efx);
79 	return resource_size(&efx->pci_dev->resource[bar]);
80 }
81 
82 static bool efx_ef10_is_vf(struct efx_nic *efx)
83 {
84 	return efx->type->is_vf;
85 }
86 
87 #ifdef CONFIG_SFC_SRIOV
88 static int efx_ef10_get_vf_index(struct efx_nic *efx)
89 {
90 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_FUNCTION_INFO_OUT_LEN);
91 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
92 	size_t outlen;
93 	int rc;
94 
95 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_FUNCTION_INFO, NULL, 0, outbuf,
96 			  sizeof(outbuf), &outlen);
97 	if (rc)
98 		return rc;
99 	if (outlen < sizeof(outbuf))
100 		return -EIO;
101 
102 	nic_data->vf_index = MCDI_DWORD(outbuf, GET_FUNCTION_INFO_OUT_VF);
103 	return 0;
104 }
105 #endif
106 
107 static int efx_ef10_init_datapath_caps(struct efx_nic *efx)
108 {
109 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_CAPABILITIES_V4_OUT_LEN);
110 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
111 	size_t outlen;
112 	int rc;
113 
114 	BUILD_BUG_ON(MC_CMD_GET_CAPABILITIES_IN_LEN != 0);
115 
116 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_CAPABILITIES, NULL, 0,
117 			  outbuf, sizeof(outbuf), &outlen);
118 	if (rc)
119 		return rc;
120 	if (outlen < MC_CMD_GET_CAPABILITIES_OUT_LEN) {
121 		netif_err(efx, drv, efx->net_dev,
122 			  "unable to read datapath firmware capabilities\n");
123 		return -EIO;
124 	}
125 
126 	nic_data->datapath_caps =
127 		MCDI_DWORD(outbuf, GET_CAPABILITIES_OUT_FLAGS1);
128 
129 	if (outlen >= MC_CMD_GET_CAPABILITIES_V2_OUT_LEN) {
130 		nic_data->datapath_caps2 = MCDI_DWORD(outbuf,
131 				GET_CAPABILITIES_V2_OUT_FLAGS2);
132 		nic_data->piobuf_size = MCDI_WORD(outbuf,
133 				GET_CAPABILITIES_V2_OUT_SIZE_PIO_BUFF);
134 	} else {
135 		nic_data->datapath_caps2 = 0;
136 		nic_data->piobuf_size = ER_DZ_TX_PIOBUF_SIZE;
137 	}
138 
139 	/* record the DPCPU firmware IDs to determine VEB vswitching support.
140 	 */
141 	nic_data->rx_dpcpu_fw_id =
142 		MCDI_WORD(outbuf, GET_CAPABILITIES_OUT_RX_DPCPU_FW_ID);
143 	nic_data->tx_dpcpu_fw_id =
144 		MCDI_WORD(outbuf, GET_CAPABILITIES_OUT_TX_DPCPU_FW_ID);
145 
146 	if (!(nic_data->datapath_caps &
147 	      (1 << MC_CMD_GET_CAPABILITIES_OUT_RX_PREFIX_LEN_14_LBN))) {
148 		netif_err(efx, probe, efx->net_dev,
149 			  "current firmware does not support an RX prefix\n");
150 		return -ENODEV;
151 	}
152 
153 	if (outlen >= MC_CMD_GET_CAPABILITIES_V3_OUT_LEN) {
154 		u8 vi_window_mode = MCDI_BYTE(outbuf,
155 				GET_CAPABILITIES_V3_OUT_VI_WINDOW_MODE);
156 
157 		rc = efx_mcdi_window_mode_to_stride(efx, vi_window_mode);
158 		if (rc)
159 			return rc;
160 	} else {
161 		/* keep default VI stride */
162 		netif_dbg(efx, probe, efx->net_dev,
163 			  "firmware did not report VI window mode, assuming vi_stride = %u\n",
164 			  efx->vi_stride);
165 	}
166 
167 	if (outlen >= MC_CMD_GET_CAPABILITIES_V4_OUT_LEN) {
168 		efx->num_mac_stats = MCDI_WORD(outbuf,
169 				GET_CAPABILITIES_V4_OUT_MAC_STATS_NUM_STATS);
170 		netif_dbg(efx, probe, efx->net_dev,
171 			  "firmware reports num_mac_stats = %u\n",
172 			  efx->num_mac_stats);
173 	} else {
174 		/* leave num_mac_stats as the default value, MC_CMD_MAC_NSTATS */
175 		netif_dbg(efx, probe, efx->net_dev,
176 			  "firmware did not report num_mac_stats, assuming %u\n",
177 			  efx->num_mac_stats);
178 	}
179 
180 	return 0;
181 }
182 
183 static void efx_ef10_read_licensed_features(struct efx_nic *efx)
184 {
185 	MCDI_DECLARE_BUF(inbuf, MC_CMD_LICENSING_V3_IN_LEN);
186 	MCDI_DECLARE_BUF(outbuf, MC_CMD_LICENSING_V3_OUT_LEN);
187 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
188 	size_t outlen;
189 	int rc;
190 
191 	MCDI_SET_DWORD(inbuf, LICENSING_V3_IN_OP,
192 		       MC_CMD_LICENSING_V3_IN_OP_REPORT_LICENSE);
193 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_LICENSING_V3, inbuf, sizeof(inbuf),
194 				outbuf, sizeof(outbuf), &outlen);
195 	if (rc || (outlen < MC_CMD_LICENSING_V3_OUT_LEN))
196 		return;
197 
198 	nic_data->licensed_features = MCDI_QWORD(outbuf,
199 					 LICENSING_V3_OUT_LICENSED_FEATURES);
200 }
201 
202 static int efx_ef10_get_sysclk_freq(struct efx_nic *efx)
203 {
204 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_CLOCK_OUT_LEN);
205 	int rc;
206 
207 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_CLOCK, NULL, 0,
208 			  outbuf, sizeof(outbuf), NULL);
209 	if (rc)
210 		return rc;
211 	rc = MCDI_DWORD(outbuf, GET_CLOCK_OUT_SYS_FREQ);
212 	return rc > 0 ? rc : -ERANGE;
213 }
214 
215 static int efx_ef10_get_timer_workarounds(struct efx_nic *efx)
216 {
217 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
218 	unsigned int implemented;
219 	unsigned int enabled;
220 	int rc;
221 
222 	nic_data->workaround_35388 = false;
223 	nic_data->workaround_61265 = false;
224 
225 	rc = efx_mcdi_get_workarounds(efx, &implemented, &enabled);
226 
227 	if (rc == -ENOSYS) {
228 		/* Firmware without GET_WORKAROUNDS - not a problem. */
229 		rc = 0;
230 	} else if (rc == 0) {
231 		/* Bug61265 workaround is always enabled if implemented. */
232 		if (enabled & MC_CMD_GET_WORKAROUNDS_OUT_BUG61265)
233 			nic_data->workaround_61265 = true;
234 
235 		if (enabled & MC_CMD_GET_WORKAROUNDS_OUT_BUG35388) {
236 			nic_data->workaround_35388 = true;
237 		} else if (implemented & MC_CMD_GET_WORKAROUNDS_OUT_BUG35388) {
238 			/* Workaround is implemented but not enabled.
239 			 * Try to enable it.
240 			 */
241 			rc = efx_mcdi_set_workaround(efx,
242 						     MC_CMD_WORKAROUND_BUG35388,
243 						     true, NULL);
244 			if (rc == 0)
245 				nic_data->workaround_35388 = true;
246 			/* If we failed to set the workaround just carry on. */
247 			rc = 0;
248 		}
249 	}
250 
251 	netif_dbg(efx, probe, efx->net_dev,
252 		  "workaround for bug 35388 is %sabled\n",
253 		  nic_data->workaround_35388 ? "en" : "dis");
254 	netif_dbg(efx, probe, efx->net_dev,
255 		  "workaround for bug 61265 is %sabled\n",
256 		  nic_data->workaround_61265 ? "en" : "dis");
257 
258 	return rc;
259 }
260 
261 static void efx_ef10_process_timer_config(struct efx_nic *efx,
262 					  const efx_dword_t *data)
263 {
264 	unsigned int max_count;
265 
266 	if (EFX_EF10_WORKAROUND_61265(efx)) {
267 		efx->timer_quantum_ns = MCDI_DWORD(data,
268 			GET_EVQ_TMR_PROPERTIES_OUT_MCDI_TMR_STEP_NS);
269 		efx->timer_max_ns = MCDI_DWORD(data,
270 			GET_EVQ_TMR_PROPERTIES_OUT_MCDI_TMR_MAX_NS);
271 	} else if (EFX_EF10_WORKAROUND_35388(efx)) {
272 		efx->timer_quantum_ns = MCDI_DWORD(data,
273 			GET_EVQ_TMR_PROPERTIES_OUT_BUG35388_TMR_NS_PER_COUNT);
274 		max_count = MCDI_DWORD(data,
275 			GET_EVQ_TMR_PROPERTIES_OUT_BUG35388_TMR_MAX_COUNT);
276 		efx->timer_max_ns = max_count * efx->timer_quantum_ns;
277 	} else {
278 		efx->timer_quantum_ns = MCDI_DWORD(data,
279 			GET_EVQ_TMR_PROPERTIES_OUT_TMR_REG_NS_PER_COUNT);
280 		max_count = MCDI_DWORD(data,
281 			GET_EVQ_TMR_PROPERTIES_OUT_TMR_REG_MAX_COUNT);
282 		efx->timer_max_ns = max_count * efx->timer_quantum_ns;
283 	}
284 
285 	netif_dbg(efx, probe, efx->net_dev,
286 		  "got timer properties from MC: quantum %u ns; max %u ns\n",
287 		  efx->timer_quantum_ns, efx->timer_max_ns);
288 }
289 
290 static int efx_ef10_get_timer_config(struct efx_nic *efx)
291 {
292 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_EVQ_TMR_PROPERTIES_OUT_LEN);
293 	int rc;
294 
295 	rc = efx_ef10_get_timer_workarounds(efx);
296 	if (rc)
297 		return rc;
298 
299 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_GET_EVQ_TMR_PROPERTIES, NULL, 0,
300 				outbuf, sizeof(outbuf), NULL);
301 
302 	if (rc == 0) {
303 		efx_ef10_process_timer_config(efx, outbuf);
304 	} else if (rc == -ENOSYS || rc == -EPERM) {
305 		/* Not available - fall back to Huntington defaults. */
306 		unsigned int quantum;
307 
308 		rc = efx_ef10_get_sysclk_freq(efx);
309 		if (rc < 0)
310 			return rc;
311 
312 		quantum = 1536000 / rc; /* 1536 cycles */
313 		efx->timer_quantum_ns = quantum;
314 		efx->timer_max_ns = efx->type->timer_period_max * quantum;
315 		rc = 0;
316 	} else {
317 		efx_mcdi_display_error(efx, MC_CMD_GET_EVQ_TMR_PROPERTIES,
318 				       MC_CMD_GET_EVQ_TMR_PROPERTIES_OUT_LEN,
319 				       NULL, 0, rc);
320 	}
321 
322 	return rc;
323 }
324 
325 static int efx_ef10_get_mac_address_pf(struct efx_nic *efx, u8 *mac_address)
326 {
327 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_MAC_ADDRESSES_OUT_LEN);
328 	size_t outlen;
329 	int rc;
330 
331 	BUILD_BUG_ON(MC_CMD_GET_MAC_ADDRESSES_IN_LEN != 0);
332 
333 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_MAC_ADDRESSES, NULL, 0,
334 			  outbuf, sizeof(outbuf), &outlen);
335 	if (rc)
336 		return rc;
337 	if (outlen < MC_CMD_GET_MAC_ADDRESSES_OUT_LEN)
338 		return -EIO;
339 
340 	ether_addr_copy(mac_address,
341 			MCDI_PTR(outbuf, GET_MAC_ADDRESSES_OUT_MAC_ADDR_BASE));
342 	return 0;
343 }
344 
345 static int efx_ef10_get_mac_address_vf(struct efx_nic *efx, u8 *mac_address)
346 {
347 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VPORT_GET_MAC_ADDRESSES_IN_LEN);
348 	MCDI_DECLARE_BUF(outbuf, MC_CMD_VPORT_GET_MAC_ADDRESSES_OUT_LENMAX);
349 	size_t outlen;
350 	int num_addrs, rc;
351 
352 	MCDI_SET_DWORD(inbuf, VPORT_GET_MAC_ADDRESSES_IN_VPORT_ID,
353 		       EVB_PORT_ID_ASSIGNED);
354 	rc = efx_mcdi_rpc(efx, MC_CMD_VPORT_GET_MAC_ADDRESSES, inbuf,
355 			  sizeof(inbuf), outbuf, sizeof(outbuf), &outlen);
356 
357 	if (rc)
358 		return rc;
359 	if (outlen < MC_CMD_VPORT_GET_MAC_ADDRESSES_OUT_LENMIN)
360 		return -EIO;
361 
362 	num_addrs = MCDI_DWORD(outbuf,
363 			       VPORT_GET_MAC_ADDRESSES_OUT_MACADDR_COUNT);
364 
365 	WARN_ON(num_addrs != 1);
366 
367 	ether_addr_copy(mac_address,
368 			MCDI_PTR(outbuf, VPORT_GET_MAC_ADDRESSES_OUT_MACADDR));
369 
370 	return 0;
371 }
372 
373 static ssize_t link_control_flag_show(struct device *dev,
374 				      struct device_attribute *attr,
375 				      char *buf)
376 {
377 	struct efx_nic *efx = dev_get_drvdata(dev);
378 
379 	return sprintf(buf, "%d\n",
380 		       ((efx->mcdi->fn_flags) &
381 			(1 << MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_LINKCTRL))
382 		       ? 1 : 0);
383 }
384 
385 static ssize_t primary_flag_show(struct device *dev,
386 				 struct device_attribute *attr,
387 				 char *buf)
388 {
389 	struct efx_nic *efx = dev_get_drvdata(dev);
390 
391 	return sprintf(buf, "%d\n",
392 		       ((efx->mcdi->fn_flags) &
393 			(1 << MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_PRIMARY))
394 		       ? 1 : 0);
395 }
396 
397 static struct efx_ef10_vlan *efx_ef10_find_vlan(struct efx_nic *efx, u16 vid)
398 {
399 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
400 	struct efx_ef10_vlan *vlan;
401 
402 	WARN_ON(!mutex_is_locked(&nic_data->vlan_lock));
403 
404 	list_for_each_entry(vlan, &nic_data->vlan_list, list) {
405 		if (vlan->vid == vid)
406 			return vlan;
407 	}
408 
409 	return NULL;
410 }
411 
412 static int efx_ef10_add_vlan(struct efx_nic *efx, u16 vid)
413 {
414 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
415 	struct efx_ef10_vlan *vlan;
416 	int rc;
417 
418 	mutex_lock(&nic_data->vlan_lock);
419 
420 	vlan = efx_ef10_find_vlan(efx, vid);
421 	if (vlan) {
422 		/* We add VID 0 on init. 8021q adds it on module init
423 		 * for all interfaces with VLAN filtring feature.
424 		 */
425 		if (vid == 0)
426 			goto done_unlock;
427 		netif_warn(efx, drv, efx->net_dev,
428 			   "VLAN %u already added\n", vid);
429 		rc = -EALREADY;
430 		goto fail_exist;
431 	}
432 
433 	rc = -ENOMEM;
434 	vlan = kzalloc(sizeof(*vlan), GFP_KERNEL);
435 	if (!vlan)
436 		goto fail_alloc;
437 
438 	vlan->vid = vid;
439 
440 	list_add_tail(&vlan->list, &nic_data->vlan_list);
441 
442 	if (efx->filter_state) {
443 		mutex_lock(&efx->mac_lock);
444 		down_write(&efx->filter_sem);
445 		rc = efx_mcdi_filter_add_vlan(efx, vlan->vid);
446 		up_write(&efx->filter_sem);
447 		mutex_unlock(&efx->mac_lock);
448 		if (rc)
449 			goto fail_filter_add_vlan;
450 	}
451 
452 done_unlock:
453 	mutex_unlock(&nic_data->vlan_lock);
454 	return 0;
455 
456 fail_filter_add_vlan:
457 	list_del(&vlan->list);
458 	kfree(vlan);
459 fail_alloc:
460 fail_exist:
461 	mutex_unlock(&nic_data->vlan_lock);
462 	return rc;
463 }
464 
465 static void efx_ef10_del_vlan_internal(struct efx_nic *efx,
466 				       struct efx_ef10_vlan *vlan)
467 {
468 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
469 
470 	WARN_ON(!mutex_is_locked(&nic_data->vlan_lock));
471 
472 	if (efx->filter_state) {
473 		down_write(&efx->filter_sem);
474 		efx_mcdi_filter_del_vlan(efx, vlan->vid);
475 		up_write(&efx->filter_sem);
476 	}
477 
478 	list_del(&vlan->list);
479 	kfree(vlan);
480 }
481 
482 static int efx_ef10_del_vlan(struct efx_nic *efx, u16 vid)
483 {
484 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
485 	struct efx_ef10_vlan *vlan;
486 	int rc = 0;
487 
488 	/* 8021q removes VID 0 on module unload for all interfaces
489 	 * with VLAN filtering feature. We need to keep it to receive
490 	 * untagged traffic.
491 	 */
492 	if (vid == 0)
493 		return 0;
494 
495 	mutex_lock(&nic_data->vlan_lock);
496 
497 	vlan = efx_ef10_find_vlan(efx, vid);
498 	if (!vlan) {
499 		netif_err(efx, drv, efx->net_dev,
500 			  "VLAN %u to be deleted not found\n", vid);
501 		rc = -ENOENT;
502 	} else {
503 		efx_ef10_del_vlan_internal(efx, vlan);
504 	}
505 
506 	mutex_unlock(&nic_data->vlan_lock);
507 
508 	return rc;
509 }
510 
511 static void efx_ef10_cleanup_vlans(struct efx_nic *efx)
512 {
513 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
514 	struct efx_ef10_vlan *vlan, *next_vlan;
515 
516 	mutex_lock(&nic_data->vlan_lock);
517 	list_for_each_entry_safe(vlan, next_vlan, &nic_data->vlan_list, list)
518 		efx_ef10_del_vlan_internal(efx, vlan);
519 	mutex_unlock(&nic_data->vlan_lock);
520 }
521 
522 static DEVICE_ATTR_RO(link_control_flag);
523 static DEVICE_ATTR_RO(primary_flag);
524 
525 static int efx_ef10_probe(struct efx_nic *efx)
526 {
527 	struct efx_ef10_nic_data *nic_data;
528 	int i, rc;
529 
530 	nic_data = kzalloc(sizeof(*nic_data), GFP_KERNEL);
531 	if (!nic_data)
532 		return -ENOMEM;
533 	efx->nic_data = nic_data;
534 
535 	/* we assume later that we can copy from this buffer in dwords */
536 	BUILD_BUG_ON(MCDI_CTL_SDU_LEN_MAX_V2 % 4);
537 
538 	rc = efx_nic_alloc_buffer(efx, &nic_data->mcdi_buf,
539 				  8 + MCDI_CTL_SDU_LEN_MAX_V2, GFP_KERNEL);
540 	if (rc)
541 		goto fail1;
542 
543 	/* Get the MC's warm boot count.  In case it's rebooting right
544 	 * now, be prepared to retry.
545 	 */
546 	i = 0;
547 	for (;;) {
548 		rc = efx_ef10_get_warm_boot_count(efx);
549 		if (rc >= 0)
550 			break;
551 		if (++i == 5)
552 			goto fail2;
553 		ssleep(1);
554 	}
555 	nic_data->warm_boot_count = rc;
556 
557 	/* In case we're recovering from a crash (kexec), we want to
558 	 * cancel any outstanding request by the previous user of this
559 	 * function.  We send a special message using the least
560 	 * significant bits of the 'high' (doorbell) register.
561 	 */
562 	_efx_writed(efx, cpu_to_le32(1), ER_DZ_MC_DB_HWRD);
563 
564 	rc = efx_mcdi_init(efx);
565 	if (rc)
566 		goto fail2;
567 
568 	mutex_init(&nic_data->udp_tunnels_lock);
569 	for (i = 0; i < ARRAY_SIZE(nic_data->udp_tunnels); ++i)
570 		nic_data->udp_tunnels[i].type =
571 			TUNNEL_ENCAP_UDP_PORT_ENTRY_INVALID;
572 
573 	/* Reset (most) configuration for this function */
574 	rc = efx_mcdi_reset(efx, RESET_TYPE_ALL);
575 	if (rc)
576 		goto fail3;
577 
578 	/* Enable event logging */
579 	rc = efx_mcdi_log_ctrl(efx, true, false, 0);
580 	if (rc)
581 		goto fail3;
582 
583 	rc = device_create_file(&efx->pci_dev->dev,
584 				&dev_attr_link_control_flag);
585 	if (rc)
586 		goto fail3;
587 
588 	rc = device_create_file(&efx->pci_dev->dev, &dev_attr_primary_flag);
589 	if (rc)
590 		goto fail4;
591 
592 	rc = efx_get_pf_index(efx, &nic_data->pf_index);
593 	if (rc)
594 		goto fail5;
595 
596 	rc = efx_ef10_init_datapath_caps(efx);
597 	if (rc < 0)
598 		goto fail5;
599 
600 	efx_ef10_read_licensed_features(efx);
601 
602 	/* We can have one VI for each vi_stride-byte region.
603 	 * However, until we use TX option descriptors we need up to four
604 	 * TX queues per channel for different checksumming combinations.
605 	 */
606 	if (nic_data->datapath_caps &
607 	    (1 << MC_CMD_GET_CAPABILITIES_OUT_VXLAN_NVGRE_LBN))
608 		efx->tx_queues_per_channel = 4;
609 	else
610 		efx->tx_queues_per_channel = 2;
611 	efx->max_vis = efx_ef10_mem_map_size(efx) / efx->vi_stride;
612 	if (!efx->max_vis) {
613 		netif_err(efx, drv, efx->net_dev, "error determining max VIs\n");
614 		rc = -EIO;
615 		goto fail5;
616 	}
617 	efx->max_channels = min_t(unsigned int, EFX_MAX_CHANNELS,
618 				  efx->max_vis / efx->tx_queues_per_channel);
619 	efx->max_tx_channels = efx->max_channels;
620 	if (WARN_ON(efx->max_channels == 0)) {
621 		rc = -EIO;
622 		goto fail5;
623 	}
624 
625 	efx->rx_packet_len_offset =
626 		ES_DZ_RX_PREFIX_PKTLEN_OFST - ES_DZ_RX_PREFIX_SIZE;
627 
628 	if (nic_data->datapath_caps &
629 	    (1 << MC_CMD_GET_CAPABILITIES_OUT_RX_INCLUDE_FCS_LBN))
630 		efx->net_dev->hw_features |= NETIF_F_RXFCS;
631 
632 	rc = efx_mcdi_port_get_number(efx);
633 	if (rc < 0)
634 		goto fail5;
635 	efx->port_num = rc;
636 
637 	rc = efx->type->get_mac_address(efx, efx->net_dev->perm_addr);
638 	if (rc)
639 		goto fail5;
640 
641 	rc = efx_ef10_get_timer_config(efx);
642 	if (rc < 0)
643 		goto fail5;
644 
645 	rc = efx_mcdi_mon_probe(efx);
646 	if (rc && rc != -EPERM)
647 		goto fail5;
648 
649 	efx_ptp_defer_probe_with_channel(efx);
650 
651 #ifdef CONFIG_SFC_SRIOV
652 	if ((efx->pci_dev->physfn) && (!efx->pci_dev->is_physfn)) {
653 		struct pci_dev *pci_dev_pf = efx->pci_dev->physfn;
654 		struct efx_nic *efx_pf = pci_get_drvdata(pci_dev_pf);
655 
656 		efx_pf->type->get_mac_address(efx_pf, nic_data->port_id);
657 	} else
658 #endif
659 		ether_addr_copy(nic_data->port_id, efx->net_dev->perm_addr);
660 
661 	INIT_LIST_HEAD(&nic_data->vlan_list);
662 	mutex_init(&nic_data->vlan_lock);
663 
664 	/* Add unspecified VID to support VLAN filtering being disabled */
665 	rc = efx_ef10_add_vlan(efx, EFX_FILTER_VID_UNSPEC);
666 	if (rc)
667 		goto fail_add_vid_unspec;
668 
669 	/* If VLAN filtering is enabled, we need VID 0 to get untagged
670 	 * traffic.  It is added automatically if 8021q module is loaded,
671 	 * but we can't rely on it since module may be not loaded.
672 	 */
673 	rc = efx_ef10_add_vlan(efx, 0);
674 	if (rc)
675 		goto fail_add_vid_0;
676 
677 	if (nic_data->datapath_caps &
678 	    (1 << MC_CMD_GET_CAPABILITIES_OUT_VXLAN_NVGRE_LBN) &&
679 	    efx->mcdi->fn_flags &
680 	    (1 << MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_TRUSTED))
681 		efx->net_dev->udp_tunnel_nic_info = &efx_ef10_udp_tunnels;
682 
683 	return 0;
684 
685 fail_add_vid_0:
686 	efx_ef10_cleanup_vlans(efx);
687 fail_add_vid_unspec:
688 	mutex_destroy(&nic_data->vlan_lock);
689 	efx_ptp_remove(efx);
690 	efx_mcdi_mon_remove(efx);
691 fail5:
692 	device_remove_file(&efx->pci_dev->dev, &dev_attr_primary_flag);
693 fail4:
694 	device_remove_file(&efx->pci_dev->dev, &dev_attr_link_control_flag);
695 fail3:
696 	efx_mcdi_detach(efx);
697 
698 	mutex_lock(&nic_data->udp_tunnels_lock);
699 	memset(nic_data->udp_tunnels, 0, sizeof(nic_data->udp_tunnels));
700 	(void)efx_ef10_set_udp_tnl_ports(efx, true);
701 	mutex_unlock(&nic_data->udp_tunnels_lock);
702 	mutex_destroy(&nic_data->udp_tunnels_lock);
703 
704 	efx_mcdi_fini(efx);
705 fail2:
706 	efx_nic_free_buffer(efx, &nic_data->mcdi_buf);
707 fail1:
708 	kfree(nic_data);
709 	efx->nic_data = NULL;
710 	return rc;
711 }
712 
713 #ifdef EFX_USE_PIO
714 
715 static void efx_ef10_free_piobufs(struct efx_nic *efx)
716 {
717 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
718 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FREE_PIOBUF_IN_LEN);
719 	unsigned int i;
720 	int rc;
721 
722 	BUILD_BUG_ON(MC_CMD_FREE_PIOBUF_OUT_LEN != 0);
723 
724 	for (i = 0; i < nic_data->n_piobufs; i++) {
725 		MCDI_SET_DWORD(inbuf, FREE_PIOBUF_IN_PIOBUF_HANDLE,
726 			       nic_data->piobuf_handle[i]);
727 		rc = efx_mcdi_rpc(efx, MC_CMD_FREE_PIOBUF, inbuf, sizeof(inbuf),
728 				  NULL, 0, NULL);
729 		WARN_ON(rc);
730 	}
731 
732 	nic_data->n_piobufs = 0;
733 }
734 
735 static int efx_ef10_alloc_piobufs(struct efx_nic *efx, unsigned int n)
736 {
737 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
738 	MCDI_DECLARE_BUF(outbuf, MC_CMD_ALLOC_PIOBUF_OUT_LEN);
739 	unsigned int i;
740 	size_t outlen;
741 	int rc = 0;
742 
743 	BUILD_BUG_ON(MC_CMD_ALLOC_PIOBUF_IN_LEN != 0);
744 
745 	for (i = 0; i < n; i++) {
746 		rc = efx_mcdi_rpc_quiet(efx, MC_CMD_ALLOC_PIOBUF, NULL, 0,
747 					outbuf, sizeof(outbuf), &outlen);
748 		if (rc) {
749 			/* Don't display the MC error if we didn't have space
750 			 * for a VF.
751 			 */
752 			if (!(efx_ef10_is_vf(efx) && rc == -ENOSPC))
753 				efx_mcdi_display_error(efx, MC_CMD_ALLOC_PIOBUF,
754 						       0, outbuf, outlen, rc);
755 			break;
756 		}
757 		if (outlen < MC_CMD_ALLOC_PIOBUF_OUT_LEN) {
758 			rc = -EIO;
759 			break;
760 		}
761 		nic_data->piobuf_handle[i] =
762 			MCDI_DWORD(outbuf, ALLOC_PIOBUF_OUT_PIOBUF_HANDLE);
763 		netif_dbg(efx, probe, efx->net_dev,
764 			  "allocated PIO buffer %u handle %x\n", i,
765 			  nic_data->piobuf_handle[i]);
766 	}
767 
768 	nic_data->n_piobufs = i;
769 	if (rc)
770 		efx_ef10_free_piobufs(efx);
771 	return rc;
772 }
773 
774 static int efx_ef10_link_piobufs(struct efx_nic *efx)
775 {
776 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
777 	MCDI_DECLARE_BUF(inbuf, MC_CMD_LINK_PIOBUF_IN_LEN);
778 	struct efx_channel *channel;
779 	struct efx_tx_queue *tx_queue;
780 	unsigned int offset, index;
781 	int rc;
782 
783 	BUILD_BUG_ON(MC_CMD_LINK_PIOBUF_OUT_LEN != 0);
784 	BUILD_BUG_ON(MC_CMD_UNLINK_PIOBUF_OUT_LEN != 0);
785 
786 	/* Link a buffer to each VI in the write-combining mapping */
787 	for (index = 0; index < nic_data->n_piobufs; ++index) {
788 		MCDI_SET_DWORD(inbuf, LINK_PIOBUF_IN_PIOBUF_HANDLE,
789 			       nic_data->piobuf_handle[index]);
790 		MCDI_SET_DWORD(inbuf, LINK_PIOBUF_IN_TXQ_INSTANCE,
791 			       nic_data->pio_write_vi_base + index);
792 		rc = efx_mcdi_rpc(efx, MC_CMD_LINK_PIOBUF,
793 				  inbuf, MC_CMD_LINK_PIOBUF_IN_LEN,
794 				  NULL, 0, NULL);
795 		if (rc) {
796 			netif_err(efx, drv, efx->net_dev,
797 				  "failed to link VI %u to PIO buffer %u (%d)\n",
798 				  nic_data->pio_write_vi_base + index, index,
799 				  rc);
800 			goto fail;
801 		}
802 		netif_dbg(efx, probe, efx->net_dev,
803 			  "linked VI %u to PIO buffer %u\n",
804 			  nic_data->pio_write_vi_base + index, index);
805 	}
806 
807 	/* Link a buffer to each TX queue */
808 	efx_for_each_channel(channel, efx) {
809 		/* Extra channels, even those with TXQs (PTP), do not require
810 		 * PIO resources.
811 		 */
812 		if (!channel->type->want_pio ||
813 		    channel->channel >= efx->xdp_channel_offset)
814 			continue;
815 
816 		efx_for_each_channel_tx_queue(tx_queue, channel) {
817 			/* We assign the PIO buffers to queues in
818 			 * reverse order to allow for the following
819 			 * special case.
820 			 */
821 			offset = ((efx->tx_channel_offset + efx->n_tx_channels -
822 				   tx_queue->channel->channel - 1) *
823 				  efx_piobuf_size);
824 			index = offset / nic_data->piobuf_size;
825 			offset = offset % nic_data->piobuf_size;
826 
827 			/* When the host page size is 4K, the first
828 			 * host page in the WC mapping may be within
829 			 * the same VI page as the last TX queue.  We
830 			 * can only link one buffer to each VI.
831 			 */
832 			if (tx_queue->queue == nic_data->pio_write_vi_base) {
833 				BUG_ON(index != 0);
834 				rc = 0;
835 			} else {
836 				MCDI_SET_DWORD(inbuf,
837 					       LINK_PIOBUF_IN_PIOBUF_HANDLE,
838 					       nic_data->piobuf_handle[index]);
839 				MCDI_SET_DWORD(inbuf,
840 					       LINK_PIOBUF_IN_TXQ_INSTANCE,
841 					       tx_queue->queue);
842 				rc = efx_mcdi_rpc(efx, MC_CMD_LINK_PIOBUF,
843 						  inbuf, MC_CMD_LINK_PIOBUF_IN_LEN,
844 						  NULL, 0, NULL);
845 			}
846 
847 			if (rc) {
848 				/* This is non-fatal; the TX path just
849 				 * won't use PIO for this queue
850 				 */
851 				netif_err(efx, drv, efx->net_dev,
852 					  "failed to link VI %u to PIO buffer %u (%d)\n",
853 					  tx_queue->queue, index, rc);
854 				tx_queue->piobuf = NULL;
855 			} else {
856 				tx_queue->piobuf =
857 					nic_data->pio_write_base +
858 					index * efx->vi_stride + offset;
859 				tx_queue->piobuf_offset = offset;
860 				netif_dbg(efx, probe, efx->net_dev,
861 					  "linked VI %u to PIO buffer %u offset %x addr %p\n",
862 					  tx_queue->queue, index,
863 					  tx_queue->piobuf_offset,
864 					  tx_queue->piobuf);
865 			}
866 		}
867 	}
868 
869 	return 0;
870 
871 fail:
872 	/* inbuf was defined for MC_CMD_LINK_PIOBUF.  We can use the same
873 	 * buffer for MC_CMD_UNLINK_PIOBUF because it's shorter.
874 	 */
875 	BUILD_BUG_ON(MC_CMD_LINK_PIOBUF_IN_LEN < MC_CMD_UNLINK_PIOBUF_IN_LEN);
876 	while (index--) {
877 		MCDI_SET_DWORD(inbuf, UNLINK_PIOBUF_IN_TXQ_INSTANCE,
878 			       nic_data->pio_write_vi_base + index);
879 		efx_mcdi_rpc(efx, MC_CMD_UNLINK_PIOBUF,
880 			     inbuf, MC_CMD_UNLINK_PIOBUF_IN_LEN,
881 			     NULL, 0, NULL);
882 	}
883 	return rc;
884 }
885 
886 static void efx_ef10_forget_old_piobufs(struct efx_nic *efx)
887 {
888 	struct efx_channel *channel;
889 	struct efx_tx_queue *tx_queue;
890 
891 	/* All our existing PIO buffers went away */
892 	efx_for_each_channel(channel, efx)
893 		efx_for_each_channel_tx_queue(tx_queue, channel)
894 			tx_queue->piobuf = NULL;
895 }
896 
897 #else /* !EFX_USE_PIO */
898 
899 static int efx_ef10_alloc_piobufs(struct efx_nic *efx, unsigned int n)
900 {
901 	return n == 0 ? 0 : -ENOBUFS;
902 }
903 
904 static int efx_ef10_link_piobufs(struct efx_nic *efx)
905 {
906 	return 0;
907 }
908 
909 static void efx_ef10_free_piobufs(struct efx_nic *efx)
910 {
911 }
912 
913 static void efx_ef10_forget_old_piobufs(struct efx_nic *efx)
914 {
915 }
916 
917 #endif /* EFX_USE_PIO */
918 
919 static void efx_ef10_remove(struct efx_nic *efx)
920 {
921 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
922 	int rc;
923 
924 #ifdef CONFIG_SFC_SRIOV
925 	struct efx_ef10_nic_data *nic_data_pf;
926 	struct pci_dev *pci_dev_pf;
927 	struct efx_nic *efx_pf;
928 	struct ef10_vf *vf;
929 
930 	if (efx->pci_dev->is_virtfn) {
931 		pci_dev_pf = efx->pci_dev->physfn;
932 		if (pci_dev_pf) {
933 			efx_pf = pci_get_drvdata(pci_dev_pf);
934 			nic_data_pf = efx_pf->nic_data;
935 			vf = nic_data_pf->vf + nic_data->vf_index;
936 			vf->efx = NULL;
937 		} else
938 			netif_info(efx, drv, efx->net_dev,
939 				   "Could not get the PF id from VF\n");
940 	}
941 #endif
942 
943 	efx_ef10_cleanup_vlans(efx);
944 	mutex_destroy(&nic_data->vlan_lock);
945 
946 	efx_ptp_remove(efx);
947 
948 	efx_mcdi_mon_remove(efx);
949 
950 	efx_mcdi_rx_free_indir_table(efx);
951 
952 	if (nic_data->wc_membase)
953 		iounmap(nic_data->wc_membase);
954 
955 	rc = efx_mcdi_free_vis(efx);
956 	WARN_ON(rc != 0);
957 
958 	if (!nic_data->must_restore_piobufs)
959 		efx_ef10_free_piobufs(efx);
960 
961 	device_remove_file(&efx->pci_dev->dev, &dev_attr_primary_flag);
962 	device_remove_file(&efx->pci_dev->dev, &dev_attr_link_control_flag);
963 
964 	efx_mcdi_detach(efx);
965 
966 	memset(nic_data->udp_tunnels, 0, sizeof(nic_data->udp_tunnels));
967 	mutex_lock(&nic_data->udp_tunnels_lock);
968 	(void)efx_ef10_set_udp_tnl_ports(efx, true);
969 	mutex_unlock(&nic_data->udp_tunnels_lock);
970 
971 	mutex_destroy(&nic_data->udp_tunnels_lock);
972 
973 	efx_mcdi_fini(efx);
974 	efx_nic_free_buffer(efx, &nic_data->mcdi_buf);
975 	kfree(nic_data);
976 }
977 
978 static int efx_ef10_probe_pf(struct efx_nic *efx)
979 {
980 	return efx_ef10_probe(efx);
981 }
982 
983 int efx_ef10_vadaptor_query(struct efx_nic *efx, unsigned int port_id,
984 			    u32 *port_flags, u32 *vadaptor_flags,
985 			    unsigned int *vlan_tags)
986 {
987 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
988 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VADAPTOR_QUERY_IN_LEN);
989 	MCDI_DECLARE_BUF(outbuf, MC_CMD_VADAPTOR_QUERY_OUT_LEN);
990 	size_t outlen;
991 	int rc;
992 
993 	if (nic_data->datapath_caps &
994 	    (1 << MC_CMD_GET_CAPABILITIES_OUT_VADAPTOR_QUERY_LBN)) {
995 		MCDI_SET_DWORD(inbuf, VADAPTOR_QUERY_IN_UPSTREAM_PORT_ID,
996 			       port_id);
997 
998 		rc = efx_mcdi_rpc(efx, MC_CMD_VADAPTOR_QUERY, inbuf, sizeof(inbuf),
999 				  outbuf, sizeof(outbuf), &outlen);
1000 		if (rc)
1001 			return rc;
1002 
1003 		if (outlen < sizeof(outbuf)) {
1004 			rc = -EIO;
1005 			return rc;
1006 		}
1007 	}
1008 
1009 	if (port_flags)
1010 		*port_flags = MCDI_DWORD(outbuf, VADAPTOR_QUERY_OUT_PORT_FLAGS);
1011 	if (vadaptor_flags)
1012 		*vadaptor_flags =
1013 			MCDI_DWORD(outbuf, VADAPTOR_QUERY_OUT_VADAPTOR_FLAGS);
1014 	if (vlan_tags)
1015 		*vlan_tags =
1016 			MCDI_DWORD(outbuf,
1017 				   VADAPTOR_QUERY_OUT_NUM_AVAILABLE_VLAN_TAGS);
1018 
1019 	return 0;
1020 }
1021 
1022 int efx_ef10_vadaptor_alloc(struct efx_nic *efx, unsigned int port_id)
1023 {
1024 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VADAPTOR_ALLOC_IN_LEN);
1025 
1026 	MCDI_SET_DWORD(inbuf, VADAPTOR_ALLOC_IN_UPSTREAM_PORT_ID, port_id);
1027 	return efx_mcdi_rpc(efx, MC_CMD_VADAPTOR_ALLOC, inbuf, sizeof(inbuf),
1028 			    NULL, 0, NULL);
1029 }
1030 
1031 int efx_ef10_vadaptor_free(struct efx_nic *efx, unsigned int port_id)
1032 {
1033 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VADAPTOR_FREE_IN_LEN);
1034 
1035 	MCDI_SET_DWORD(inbuf, VADAPTOR_FREE_IN_UPSTREAM_PORT_ID, port_id);
1036 	return efx_mcdi_rpc(efx, MC_CMD_VADAPTOR_FREE, inbuf, sizeof(inbuf),
1037 			    NULL, 0, NULL);
1038 }
1039 
1040 int efx_ef10_vport_add_mac(struct efx_nic *efx,
1041 			   unsigned int port_id, const u8 *mac)
1042 {
1043 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VPORT_ADD_MAC_ADDRESS_IN_LEN);
1044 
1045 	MCDI_SET_DWORD(inbuf, VPORT_ADD_MAC_ADDRESS_IN_VPORT_ID, port_id);
1046 	ether_addr_copy(MCDI_PTR(inbuf, VPORT_ADD_MAC_ADDRESS_IN_MACADDR), mac);
1047 
1048 	return efx_mcdi_rpc(efx, MC_CMD_VPORT_ADD_MAC_ADDRESS, inbuf,
1049 			    sizeof(inbuf), NULL, 0, NULL);
1050 }
1051 
1052 int efx_ef10_vport_del_mac(struct efx_nic *efx,
1053 			   unsigned int port_id, const u8 *mac)
1054 {
1055 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VPORT_DEL_MAC_ADDRESS_IN_LEN);
1056 
1057 	MCDI_SET_DWORD(inbuf, VPORT_DEL_MAC_ADDRESS_IN_VPORT_ID, port_id);
1058 	ether_addr_copy(MCDI_PTR(inbuf, VPORT_DEL_MAC_ADDRESS_IN_MACADDR), mac);
1059 
1060 	return efx_mcdi_rpc(efx, MC_CMD_VPORT_DEL_MAC_ADDRESS, inbuf,
1061 			    sizeof(inbuf), NULL, 0, NULL);
1062 }
1063 
1064 #ifdef CONFIG_SFC_SRIOV
1065 static int efx_ef10_probe_vf(struct efx_nic *efx)
1066 {
1067 	int rc;
1068 	struct pci_dev *pci_dev_pf;
1069 
1070 	/* If the parent PF has no VF data structure, it doesn't know about this
1071 	 * VF so fail probe.  The VF needs to be re-created.  This can happen
1072 	 * if the PF driver was unloaded while any VF was assigned to a guest
1073 	 * (using Xen, only).
1074 	 */
1075 	pci_dev_pf = efx->pci_dev->physfn;
1076 	if (pci_dev_pf) {
1077 		struct efx_nic *efx_pf = pci_get_drvdata(pci_dev_pf);
1078 		struct efx_ef10_nic_data *nic_data_pf = efx_pf->nic_data;
1079 
1080 		if (!nic_data_pf->vf) {
1081 			netif_info(efx, drv, efx->net_dev,
1082 				   "The VF cannot link to its parent PF; "
1083 				   "please destroy and re-create the VF\n");
1084 			return -EBUSY;
1085 		}
1086 	}
1087 
1088 	rc = efx_ef10_probe(efx);
1089 	if (rc)
1090 		return rc;
1091 
1092 	rc = efx_ef10_get_vf_index(efx);
1093 	if (rc)
1094 		goto fail;
1095 
1096 	if (efx->pci_dev->is_virtfn) {
1097 		if (efx->pci_dev->physfn) {
1098 			struct efx_nic *efx_pf =
1099 				pci_get_drvdata(efx->pci_dev->physfn);
1100 			struct efx_ef10_nic_data *nic_data_p = efx_pf->nic_data;
1101 			struct efx_ef10_nic_data *nic_data = efx->nic_data;
1102 
1103 			nic_data_p->vf[nic_data->vf_index].efx = efx;
1104 			nic_data_p->vf[nic_data->vf_index].pci_dev =
1105 				efx->pci_dev;
1106 		} else
1107 			netif_info(efx, drv, efx->net_dev,
1108 				   "Could not get the PF id from VF\n");
1109 	}
1110 
1111 	return 0;
1112 
1113 fail:
1114 	efx_ef10_remove(efx);
1115 	return rc;
1116 }
1117 #else
1118 static int efx_ef10_probe_vf(struct efx_nic *efx __attribute__ ((unused)))
1119 {
1120 	return 0;
1121 }
1122 #endif
1123 
1124 static int efx_ef10_alloc_vis(struct efx_nic *efx,
1125 			      unsigned int min_vis, unsigned int max_vis)
1126 {
1127 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1128 
1129 	return efx_mcdi_alloc_vis(efx, min_vis, max_vis, &nic_data->vi_base,
1130 				  &nic_data->n_allocated_vis);
1131 }
1132 
1133 /* Note that the failure path of this function does not free
1134  * resources, as this will be done by efx_ef10_remove().
1135  */
1136 static int efx_ef10_dimension_resources(struct efx_nic *efx)
1137 {
1138 	unsigned int min_vis = max_t(unsigned int, efx->tx_queues_per_channel,
1139 				     efx_separate_tx_channels ? 2 : 1);
1140 	unsigned int channel_vis, pio_write_vi_base, max_vis;
1141 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1142 	unsigned int uc_mem_map_size, wc_mem_map_size;
1143 	void __iomem *membase;
1144 	int rc;
1145 
1146 	channel_vis = max(efx->n_channels,
1147 			  ((efx->n_tx_channels + efx->n_extra_tx_channels) *
1148 			   efx->tx_queues_per_channel) +
1149 			   efx->n_xdp_channels * efx->xdp_tx_per_channel);
1150 	if (efx->max_vis && efx->max_vis < channel_vis) {
1151 		netif_dbg(efx, drv, efx->net_dev,
1152 			  "Reducing channel VIs from %u to %u\n",
1153 			  channel_vis, efx->max_vis);
1154 		channel_vis = efx->max_vis;
1155 	}
1156 
1157 #ifdef EFX_USE_PIO
1158 	/* Try to allocate PIO buffers if wanted and if the full
1159 	 * number of PIO buffers would be sufficient to allocate one
1160 	 * copy-buffer per TX channel.  Failure is non-fatal, as there
1161 	 * are only a small number of PIO buffers shared between all
1162 	 * functions of the controller.
1163 	 */
1164 	if (efx_piobuf_size != 0 &&
1165 	    nic_data->piobuf_size / efx_piobuf_size * EF10_TX_PIOBUF_COUNT >=
1166 	    efx->n_tx_channels) {
1167 		unsigned int n_piobufs =
1168 			DIV_ROUND_UP(efx->n_tx_channels,
1169 				     nic_data->piobuf_size / efx_piobuf_size);
1170 
1171 		rc = efx_ef10_alloc_piobufs(efx, n_piobufs);
1172 		if (rc == -ENOSPC)
1173 			netif_dbg(efx, probe, efx->net_dev,
1174 				  "out of PIO buffers; cannot allocate more\n");
1175 		else if (rc == -EPERM)
1176 			netif_dbg(efx, probe, efx->net_dev,
1177 				  "not permitted to allocate PIO buffers\n");
1178 		else if (rc)
1179 			netif_err(efx, probe, efx->net_dev,
1180 				  "failed to allocate PIO buffers (%d)\n", rc);
1181 		else
1182 			netif_dbg(efx, probe, efx->net_dev,
1183 				  "allocated %u PIO buffers\n", n_piobufs);
1184 	}
1185 #else
1186 	nic_data->n_piobufs = 0;
1187 #endif
1188 
1189 	/* PIO buffers should be mapped with write-combining enabled,
1190 	 * and we want to make single UC and WC mappings rather than
1191 	 * several of each (in fact that's the only option if host
1192 	 * page size is >4K).  So we may allocate some extra VIs just
1193 	 * for writing PIO buffers through.
1194 	 *
1195 	 * The UC mapping contains (channel_vis - 1) complete VIs and the
1196 	 * first 4K of the next VI.  Then the WC mapping begins with
1197 	 * the remainder of this last VI.
1198 	 */
1199 	uc_mem_map_size = PAGE_ALIGN((channel_vis - 1) * efx->vi_stride +
1200 				     ER_DZ_TX_PIOBUF);
1201 	if (nic_data->n_piobufs) {
1202 		/* pio_write_vi_base rounds down to give the number of complete
1203 		 * VIs inside the UC mapping.
1204 		 */
1205 		pio_write_vi_base = uc_mem_map_size / efx->vi_stride;
1206 		wc_mem_map_size = (PAGE_ALIGN((pio_write_vi_base +
1207 					       nic_data->n_piobufs) *
1208 					      efx->vi_stride) -
1209 				   uc_mem_map_size);
1210 		max_vis = pio_write_vi_base + nic_data->n_piobufs;
1211 	} else {
1212 		pio_write_vi_base = 0;
1213 		wc_mem_map_size = 0;
1214 		max_vis = channel_vis;
1215 	}
1216 
1217 	/* In case the last attached driver failed to free VIs, do it now */
1218 	rc = efx_mcdi_free_vis(efx);
1219 	if (rc != 0)
1220 		return rc;
1221 
1222 	rc = efx_ef10_alloc_vis(efx, min_vis, max_vis);
1223 	if (rc != 0)
1224 		return rc;
1225 
1226 	if (nic_data->n_allocated_vis < channel_vis) {
1227 		netif_info(efx, drv, efx->net_dev,
1228 			   "Could not allocate enough VIs to satisfy RSS"
1229 			   " requirements. Performance may not be optimal.\n");
1230 		/* We didn't get the VIs to populate our channels.
1231 		 * We could keep what we got but then we'd have more
1232 		 * interrupts than we need.
1233 		 * Instead calculate new max_channels and restart
1234 		 */
1235 		efx->max_channels = nic_data->n_allocated_vis;
1236 		efx->max_tx_channels =
1237 			nic_data->n_allocated_vis / efx->tx_queues_per_channel;
1238 
1239 		efx_mcdi_free_vis(efx);
1240 		return -EAGAIN;
1241 	}
1242 
1243 	/* If we didn't get enough VIs to map all the PIO buffers, free the
1244 	 * PIO buffers
1245 	 */
1246 	if (nic_data->n_piobufs &&
1247 	    nic_data->n_allocated_vis <
1248 	    pio_write_vi_base + nic_data->n_piobufs) {
1249 		netif_dbg(efx, probe, efx->net_dev,
1250 			  "%u VIs are not sufficient to map %u PIO buffers\n",
1251 			  nic_data->n_allocated_vis, nic_data->n_piobufs);
1252 		efx_ef10_free_piobufs(efx);
1253 	}
1254 
1255 	/* Shrink the original UC mapping of the memory BAR */
1256 	membase = ioremap(efx->membase_phys, uc_mem_map_size);
1257 	if (!membase) {
1258 		netif_err(efx, probe, efx->net_dev,
1259 			  "could not shrink memory BAR to %x\n",
1260 			  uc_mem_map_size);
1261 		return -ENOMEM;
1262 	}
1263 	iounmap(efx->membase);
1264 	efx->membase = membase;
1265 
1266 	/* Set up the WC mapping if needed */
1267 	if (wc_mem_map_size) {
1268 		nic_data->wc_membase = ioremap_wc(efx->membase_phys +
1269 						  uc_mem_map_size,
1270 						  wc_mem_map_size);
1271 		if (!nic_data->wc_membase) {
1272 			netif_err(efx, probe, efx->net_dev,
1273 				  "could not allocate WC mapping of size %x\n",
1274 				  wc_mem_map_size);
1275 			return -ENOMEM;
1276 		}
1277 		nic_data->pio_write_vi_base = pio_write_vi_base;
1278 		nic_data->pio_write_base =
1279 			nic_data->wc_membase +
1280 			(pio_write_vi_base * efx->vi_stride + ER_DZ_TX_PIOBUF -
1281 			 uc_mem_map_size);
1282 
1283 		rc = efx_ef10_link_piobufs(efx);
1284 		if (rc)
1285 			efx_ef10_free_piobufs(efx);
1286 	}
1287 
1288 	netif_dbg(efx, probe, efx->net_dev,
1289 		  "memory BAR at %pa (virtual %p+%x UC, %p+%x WC)\n",
1290 		  &efx->membase_phys, efx->membase, uc_mem_map_size,
1291 		  nic_data->wc_membase, wc_mem_map_size);
1292 
1293 	return 0;
1294 }
1295 
1296 static void efx_ef10_fini_nic(struct efx_nic *efx)
1297 {
1298 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1299 
1300 	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 void efx_ef10_get_fec_stats(struct efx_nic *efx,
1751 				   struct ethtool_fec_stats *fec_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 
1757 	efx_ef10_get_stat_mask(efx, mask);
1758 	if (test_bit(EF10_STAT_fec_corrected_errors, mask))
1759 		fec_stats->corrected_blocks.total =
1760 			stats[EF10_STAT_fec_corrected_errors];
1761 	if (test_bit(EF10_STAT_fec_uncorrected_errors, mask))
1762 		fec_stats->uncorrectable_blocks.total =
1763 			stats[EF10_STAT_fec_uncorrected_errors];
1764 }
1765 
1766 static size_t efx_ef10_update_stats_common(struct efx_nic *efx, u64 *full_stats,
1767 					   struct rtnl_link_stats64 *core_stats)
1768 {
1769 	DECLARE_BITMAP(mask, EF10_STAT_COUNT);
1770 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1771 	u64 *stats = nic_data->stats;
1772 	size_t stats_count = 0, index;
1773 
1774 	efx_ef10_get_stat_mask(efx, mask);
1775 
1776 	if (full_stats) {
1777 		for_each_set_bit(index, mask, EF10_STAT_COUNT) {
1778 			if (efx_ef10_stat_desc[index].name) {
1779 				*full_stats++ = stats[index];
1780 				++stats_count;
1781 			}
1782 		}
1783 	}
1784 
1785 	if (!core_stats)
1786 		return stats_count;
1787 
1788 	if (nic_data->datapath_caps &
1789 			1 << MC_CMD_GET_CAPABILITIES_OUT_EVB_LBN) {
1790 		/* Use vadaptor stats. */
1791 		core_stats->rx_packets = stats[EF10_STAT_rx_unicast] +
1792 					 stats[EF10_STAT_rx_multicast] +
1793 					 stats[EF10_STAT_rx_broadcast];
1794 		core_stats->tx_packets = stats[EF10_STAT_tx_unicast] +
1795 					 stats[EF10_STAT_tx_multicast] +
1796 					 stats[EF10_STAT_tx_broadcast];
1797 		core_stats->rx_bytes = stats[EF10_STAT_rx_unicast_bytes] +
1798 				       stats[EF10_STAT_rx_multicast_bytes] +
1799 				       stats[EF10_STAT_rx_broadcast_bytes];
1800 		core_stats->tx_bytes = stats[EF10_STAT_tx_unicast_bytes] +
1801 				       stats[EF10_STAT_tx_multicast_bytes] +
1802 				       stats[EF10_STAT_tx_broadcast_bytes];
1803 		core_stats->rx_dropped = stats[GENERIC_STAT_rx_nodesc_trunc] +
1804 					 stats[GENERIC_STAT_rx_noskb_drops];
1805 		core_stats->multicast = stats[EF10_STAT_rx_multicast];
1806 		core_stats->rx_crc_errors = stats[EF10_STAT_rx_bad];
1807 		core_stats->rx_fifo_errors = stats[EF10_STAT_rx_overflow];
1808 		core_stats->rx_errors = core_stats->rx_crc_errors;
1809 		core_stats->tx_errors = stats[EF10_STAT_tx_bad];
1810 	} else {
1811 		/* Use port stats. */
1812 		core_stats->rx_packets = stats[EF10_STAT_port_rx_packets];
1813 		core_stats->tx_packets = stats[EF10_STAT_port_tx_packets];
1814 		core_stats->rx_bytes = stats[EF10_STAT_port_rx_bytes];
1815 		core_stats->tx_bytes = stats[EF10_STAT_port_tx_bytes];
1816 		core_stats->rx_dropped = stats[EF10_STAT_port_rx_nodesc_drops] +
1817 					 stats[GENERIC_STAT_rx_nodesc_trunc] +
1818 					 stats[GENERIC_STAT_rx_noskb_drops];
1819 		core_stats->multicast = stats[EF10_STAT_port_rx_multicast];
1820 		core_stats->rx_length_errors =
1821 				stats[EF10_STAT_port_rx_gtjumbo] +
1822 				stats[EF10_STAT_port_rx_length_error];
1823 		core_stats->rx_crc_errors = stats[EF10_STAT_port_rx_bad];
1824 		core_stats->rx_frame_errors =
1825 				stats[EF10_STAT_port_rx_align_error];
1826 		core_stats->rx_fifo_errors = stats[EF10_STAT_port_rx_overflow];
1827 		core_stats->rx_errors = (core_stats->rx_length_errors +
1828 					 core_stats->rx_crc_errors +
1829 					 core_stats->rx_frame_errors);
1830 	}
1831 
1832 	return stats_count;
1833 }
1834 
1835 static size_t efx_ef10_update_stats_pf(struct efx_nic *efx, u64 *full_stats,
1836 				       struct rtnl_link_stats64 *core_stats)
1837 {
1838 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1839 	DECLARE_BITMAP(mask, EF10_STAT_COUNT);
1840 	u64 *stats = nic_data->stats;
1841 
1842 	efx_ef10_get_stat_mask(efx, mask);
1843 
1844 	efx_nic_copy_stats(efx, nic_data->mc_stats);
1845 	efx_nic_update_stats(efx_ef10_stat_desc, EF10_STAT_COUNT,
1846 			     mask, stats, nic_data->mc_stats, false);
1847 
1848 	/* Update derived statistics */
1849 	efx_nic_fix_nodesc_drop_stat(efx,
1850 				     &stats[EF10_STAT_port_rx_nodesc_drops]);
1851 	/* MC Firmware reads RX_BYTES and RX_GOOD_BYTES from the MAC.
1852 	 * It then calculates RX_BAD_BYTES and DMAs it to us with RX_BYTES.
1853 	 * We report these as port_rx_ stats. We are not given RX_GOOD_BYTES.
1854 	 * Here we calculate port_rx_good_bytes.
1855 	 */
1856 	stats[EF10_STAT_port_rx_good_bytes] =
1857 		stats[EF10_STAT_port_rx_bytes] -
1858 		stats[EF10_STAT_port_rx_bytes_minus_good_bytes];
1859 
1860 	/* The asynchronous reads used to calculate RX_BAD_BYTES in
1861 	 * MC Firmware are done such that we should not see an increase in
1862 	 * RX_BAD_BYTES when a good packet has arrived. Unfortunately this
1863 	 * does mean that the stat can decrease at times. Here we do not
1864 	 * update the stat unless it has increased or has gone to zero
1865 	 * (In the case of the NIC rebooting).
1866 	 * Please see Bug 33781 for a discussion of why things work this way.
1867 	 */
1868 	efx_update_diff_stat(&stats[EF10_STAT_port_rx_bad_bytes],
1869 			     stats[EF10_STAT_port_rx_bytes_minus_good_bytes]);
1870 	efx_update_sw_stats(efx, stats);
1871 
1872 	return efx_ef10_update_stats_common(efx, full_stats, core_stats);
1873 }
1874 
1875 static int efx_ef10_try_update_nic_stats_vf(struct efx_nic *efx)
1876 	__must_hold(&efx->stats_lock)
1877 {
1878 	MCDI_DECLARE_BUF(inbuf, MC_CMD_MAC_STATS_IN_LEN);
1879 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1880 	DECLARE_BITMAP(mask, EF10_STAT_COUNT);
1881 	__le64 generation_start, generation_end;
1882 	u64 *stats = nic_data->stats;
1883 	u32 dma_len = efx->num_mac_stats * sizeof(u64);
1884 	struct efx_buffer stats_buf;
1885 	__le64 *dma_stats;
1886 	int rc;
1887 
1888 	spin_unlock_bh(&efx->stats_lock);
1889 
1890 	efx_ef10_get_stat_mask(efx, mask);
1891 
1892 	rc = efx_nic_alloc_buffer(efx, &stats_buf, dma_len, GFP_KERNEL);
1893 	if (rc) {
1894 		spin_lock_bh(&efx->stats_lock);
1895 		return rc;
1896 	}
1897 
1898 	dma_stats = stats_buf.addr;
1899 	dma_stats[efx->num_mac_stats - 1] = EFX_MC_STATS_GENERATION_INVALID;
1900 
1901 	MCDI_SET_QWORD(inbuf, MAC_STATS_IN_DMA_ADDR, stats_buf.dma_addr);
1902 	MCDI_POPULATE_DWORD_1(inbuf, MAC_STATS_IN_CMD,
1903 			      MAC_STATS_IN_DMA, 1);
1904 	MCDI_SET_DWORD(inbuf, MAC_STATS_IN_DMA_LEN, dma_len);
1905 	MCDI_SET_DWORD(inbuf, MAC_STATS_IN_PORT_ID, EVB_PORT_ID_ASSIGNED);
1906 
1907 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_MAC_STATS, inbuf, sizeof(inbuf),
1908 				NULL, 0, NULL);
1909 	spin_lock_bh(&efx->stats_lock);
1910 	if (rc) {
1911 		/* Expect ENOENT if DMA queues have not been set up */
1912 		if (rc != -ENOENT || atomic_read(&efx->active_queues))
1913 			efx_mcdi_display_error(efx, MC_CMD_MAC_STATS,
1914 					       sizeof(inbuf), NULL, 0, rc);
1915 		goto out;
1916 	}
1917 
1918 	generation_end = dma_stats[efx->num_mac_stats - 1];
1919 	if (generation_end == EFX_MC_STATS_GENERATION_INVALID) {
1920 		WARN_ON_ONCE(1);
1921 		goto out;
1922 	}
1923 	rmb();
1924 	efx_nic_update_stats(efx_ef10_stat_desc, EF10_STAT_COUNT, mask,
1925 			     stats, stats_buf.addr, false);
1926 	rmb();
1927 	generation_start = dma_stats[MC_CMD_MAC_GENERATION_START];
1928 	if (generation_end != generation_start) {
1929 		rc = -EAGAIN;
1930 		goto out;
1931 	}
1932 
1933 	efx_update_sw_stats(efx, stats);
1934 out:
1935 	/* releasing a DMA coherent buffer with BH disabled can panic */
1936 	spin_unlock_bh(&efx->stats_lock);
1937 	efx_nic_free_buffer(efx, &stats_buf);
1938 	spin_lock_bh(&efx->stats_lock);
1939 	return rc;
1940 }
1941 
1942 static size_t efx_ef10_update_stats_vf(struct efx_nic *efx, u64 *full_stats,
1943 				       struct rtnl_link_stats64 *core_stats)
1944 {
1945 	if (efx_ef10_try_update_nic_stats_vf(efx))
1946 		return 0;
1947 
1948 	return efx_ef10_update_stats_common(efx, full_stats, core_stats);
1949 }
1950 
1951 static size_t efx_ef10_update_stats_atomic_vf(struct efx_nic *efx, u64 *full_stats,
1952 					      struct rtnl_link_stats64 *core_stats)
1953 {
1954 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1955 
1956 	/* In atomic context, cannot update HW stats.  Just update the
1957 	 * software stats and return so the caller can continue.
1958 	 */
1959 	efx_update_sw_stats(efx, nic_data->stats);
1960 	return efx_ef10_update_stats_common(efx, full_stats, core_stats);
1961 }
1962 
1963 static void efx_ef10_push_irq_moderation(struct efx_channel *channel)
1964 {
1965 	struct efx_nic *efx = channel->efx;
1966 	unsigned int mode, usecs;
1967 	efx_dword_t timer_cmd;
1968 
1969 	if (channel->irq_moderation_us) {
1970 		mode = 3;
1971 		usecs = channel->irq_moderation_us;
1972 	} else {
1973 		mode = 0;
1974 		usecs = 0;
1975 	}
1976 
1977 	if (EFX_EF10_WORKAROUND_61265(efx)) {
1978 		MCDI_DECLARE_BUF(inbuf, MC_CMD_SET_EVQ_TMR_IN_LEN);
1979 		unsigned int ns = usecs * 1000;
1980 
1981 		MCDI_SET_DWORD(inbuf, SET_EVQ_TMR_IN_INSTANCE,
1982 			       channel->channel);
1983 		MCDI_SET_DWORD(inbuf, SET_EVQ_TMR_IN_TMR_LOAD_REQ_NS, ns);
1984 		MCDI_SET_DWORD(inbuf, SET_EVQ_TMR_IN_TMR_RELOAD_REQ_NS, ns);
1985 		MCDI_SET_DWORD(inbuf, SET_EVQ_TMR_IN_TMR_MODE, mode);
1986 
1987 		efx_mcdi_rpc_async(efx, MC_CMD_SET_EVQ_TMR,
1988 				   inbuf, sizeof(inbuf), 0, NULL, 0);
1989 	} else if (EFX_EF10_WORKAROUND_35388(efx)) {
1990 		unsigned int ticks = efx_usecs_to_ticks(efx, usecs);
1991 
1992 		EFX_POPULATE_DWORD_3(timer_cmd, ERF_DD_EVQ_IND_TIMER_FLAGS,
1993 				     EFE_DD_EVQ_IND_TIMER_FLAGS,
1994 				     ERF_DD_EVQ_IND_TIMER_MODE, mode,
1995 				     ERF_DD_EVQ_IND_TIMER_VAL, ticks);
1996 		efx_writed_page(efx, &timer_cmd, ER_DD_EVQ_INDIRECT,
1997 				channel->channel);
1998 	} else {
1999 		unsigned int ticks = efx_usecs_to_ticks(efx, usecs);
2000 
2001 		EFX_POPULATE_DWORD_3(timer_cmd, ERF_DZ_TC_TIMER_MODE, mode,
2002 				     ERF_DZ_TC_TIMER_VAL, ticks,
2003 				     ERF_FZ_TC_TMR_REL_VAL, ticks);
2004 		efx_writed_page(efx, &timer_cmd, ER_DZ_EVQ_TMR,
2005 				channel->channel);
2006 	}
2007 }
2008 
2009 static void efx_ef10_get_wol_vf(struct efx_nic *efx,
2010 				struct ethtool_wolinfo *wol) {}
2011 
2012 static int efx_ef10_set_wol_vf(struct efx_nic *efx, u32 type)
2013 {
2014 	return -EOPNOTSUPP;
2015 }
2016 
2017 static void efx_ef10_get_wol(struct efx_nic *efx, struct ethtool_wolinfo *wol)
2018 {
2019 	wol->supported = 0;
2020 	wol->wolopts = 0;
2021 	memset(&wol->sopass, 0, sizeof(wol->sopass));
2022 }
2023 
2024 static int efx_ef10_set_wol(struct efx_nic *efx, u32 type)
2025 {
2026 	if (type != 0)
2027 		return -EINVAL;
2028 	return 0;
2029 }
2030 
2031 static void efx_ef10_mcdi_request(struct efx_nic *efx,
2032 				  const efx_dword_t *hdr, size_t hdr_len,
2033 				  const efx_dword_t *sdu, size_t sdu_len)
2034 {
2035 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2036 	u8 *pdu = nic_data->mcdi_buf.addr;
2037 
2038 	memcpy(pdu, hdr, hdr_len);
2039 	memcpy(pdu + hdr_len, sdu, sdu_len);
2040 	wmb();
2041 
2042 	/* The hardware provides 'low' and 'high' (doorbell) registers
2043 	 * for passing the 64-bit address of an MCDI request to
2044 	 * firmware.  However the dwords are swapped by firmware.  The
2045 	 * least significant bits of the doorbell are then 0 for all
2046 	 * MCDI requests due to alignment.
2047 	 */
2048 	_efx_writed(efx, cpu_to_le32((u64)nic_data->mcdi_buf.dma_addr >> 32),
2049 		    ER_DZ_MC_DB_LWRD);
2050 	_efx_writed(efx, cpu_to_le32((u32)nic_data->mcdi_buf.dma_addr),
2051 		    ER_DZ_MC_DB_HWRD);
2052 }
2053 
2054 static bool efx_ef10_mcdi_poll_response(struct efx_nic *efx)
2055 {
2056 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2057 	const efx_dword_t hdr = *(const efx_dword_t *)nic_data->mcdi_buf.addr;
2058 
2059 	rmb();
2060 	return EFX_DWORD_FIELD(hdr, MCDI_HEADER_RESPONSE);
2061 }
2062 
2063 static void
2064 efx_ef10_mcdi_read_response(struct efx_nic *efx, efx_dword_t *outbuf,
2065 			    size_t offset, size_t outlen)
2066 {
2067 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2068 	const u8 *pdu = nic_data->mcdi_buf.addr;
2069 
2070 	memcpy(outbuf, pdu + offset, outlen);
2071 }
2072 
2073 static void efx_ef10_mcdi_reboot_detected(struct efx_nic *efx)
2074 {
2075 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2076 
2077 	/* All our allocations have been reset */
2078 	efx_ef10_table_reset_mc_allocations(efx);
2079 
2080 	/* The datapath firmware might have been changed */
2081 	nic_data->must_check_datapath_caps = true;
2082 
2083 	/* MAC statistics have been cleared on the NIC; clear the local
2084 	 * statistic that we update with efx_update_diff_stat().
2085 	 */
2086 	nic_data->stats[EF10_STAT_port_rx_bad_bytes] = 0;
2087 }
2088 
2089 static int efx_ef10_mcdi_poll_reboot(struct efx_nic *efx)
2090 {
2091 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2092 	int rc;
2093 
2094 	rc = efx_ef10_get_warm_boot_count(efx);
2095 	if (rc < 0) {
2096 		/* The firmware is presumably in the process of
2097 		 * rebooting.  However, we are supposed to report each
2098 		 * reboot just once, so we must only do that once we
2099 		 * can read and store the updated warm boot count.
2100 		 */
2101 		return 0;
2102 	}
2103 
2104 	if (rc == nic_data->warm_boot_count)
2105 		return 0;
2106 
2107 	nic_data->warm_boot_count = rc;
2108 	efx_ef10_mcdi_reboot_detected(efx);
2109 
2110 	return -EIO;
2111 }
2112 
2113 /* Handle an MSI interrupt
2114  *
2115  * Handle an MSI hardware interrupt.  This routine schedules event
2116  * queue processing.  No interrupt acknowledgement cycle is necessary.
2117  * Also, we never need to check that the interrupt is for us, since
2118  * MSI interrupts cannot be shared.
2119  */
2120 static irqreturn_t efx_ef10_msi_interrupt(int irq, void *dev_id)
2121 {
2122 	struct efx_msi_context *context = dev_id;
2123 	struct efx_nic *efx = context->efx;
2124 
2125 	netif_vdbg(efx, intr, efx->net_dev,
2126 		   "IRQ %d on CPU %d\n", irq, raw_smp_processor_id());
2127 
2128 	if (likely(READ_ONCE(efx->irq_soft_enabled))) {
2129 		/* Note test interrupts */
2130 		if (context->index == efx->irq_level)
2131 			efx->last_irq_cpu = raw_smp_processor_id();
2132 
2133 		/* Schedule processing of the channel */
2134 		efx_schedule_channel_irq(efx->channel[context->index]);
2135 	}
2136 
2137 	return IRQ_HANDLED;
2138 }
2139 
2140 static irqreturn_t efx_ef10_legacy_interrupt(int irq, void *dev_id)
2141 {
2142 	struct efx_nic *efx = dev_id;
2143 	bool soft_enabled = READ_ONCE(efx->irq_soft_enabled);
2144 	struct efx_channel *channel;
2145 	efx_dword_t reg;
2146 	u32 queues;
2147 
2148 	/* Read the ISR which also ACKs the interrupts */
2149 	efx_readd(efx, &reg, ER_DZ_BIU_INT_ISR);
2150 	queues = EFX_DWORD_FIELD(reg, ERF_DZ_ISR_REG);
2151 
2152 	if (queues == 0)
2153 		return IRQ_NONE;
2154 
2155 	if (likely(soft_enabled)) {
2156 		/* Note test interrupts */
2157 		if (queues & (1U << efx->irq_level))
2158 			efx->last_irq_cpu = raw_smp_processor_id();
2159 
2160 		efx_for_each_channel(channel, efx) {
2161 			if (queues & 1)
2162 				efx_schedule_channel_irq(channel);
2163 			queues >>= 1;
2164 		}
2165 	}
2166 
2167 	netif_vdbg(efx, intr, efx->net_dev,
2168 		   "IRQ %d on CPU %d status " EFX_DWORD_FMT "\n",
2169 		   irq, raw_smp_processor_id(), EFX_DWORD_VAL(reg));
2170 
2171 	return IRQ_HANDLED;
2172 }
2173 
2174 static int efx_ef10_irq_test_generate(struct efx_nic *efx)
2175 {
2176 	MCDI_DECLARE_BUF(inbuf, MC_CMD_TRIGGER_INTERRUPT_IN_LEN);
2177 
2178 	if (efx_mcdi_set_workaround(efx, MC_CMD_WORKAROUND_BUG41750, true,
2179 				    NULL) == 0)
2180 		return -ENOTSUPP;
2181 
2182 	BUILD_BUG_ON(MC_CMD_TRIGGER_INTERRUPT_OUT_LEN != 0);
2183 
2184 	MCDI_SET_DWORD(inbuf, TRIGGER_INTERRUPT_IN_INTR_LEVEL, efx->irq_level);
2185 	return efx_mcdi_rpc(efx, MC_CMD_TRIGGER_INTERRUPT,
2186 			    inbuf, sizeof(inbuf), NULL, 0, NULL);
2187 }
2188 
2189 static int efx_ef10_tx_probe(struct efx_tx_queue *tx_queue)
2190 {
2191 	/* low two bits of label are what we want for type */
2192 	BUILD_BUG_ON((EFX_TXQ_TYPE_OUTER_CSUM | EFX_TXQ_TYPE_INNER_CSUM) != 3);
2193 	tx_queue->type = tx_queue->label & 3;
2194 	return efx_nic_alloc_buffer(tx_queue->efx, &tx_queue->txd.buf,
2195 				    (tx_queue->ptr_mask + 1) *
2196 				    sizeof(efx_qword_t),
2197 				    GFP_KERNEL);
2198 }
2199 
2200 /* This writes to the TX_DESC_WPTR and also pushes data */
2201 static inline void efx_ef10_push_tx_desc(struct efx_tx_queue *tx_queue,
2202 					 const efx_qword_t *txd)
2203 {
2204 	unsigned int write_ptr;
2205 	efx_oword_t reg;
2206 
2207 	write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
2208 	EFX_POPULATE_OWORD_1(reg, ERF_DZ_TX_DESC_WPTR, write_ptr);
2209 	reg.qword[0] = *txd;
2210 	efx_writeo_page(tx_queue->efx, &reg,
2211 			ER_DZ_TX_DESC_UPD, tx_queue->queue);
2212 }
2213 
2214 /* Add Firmware-Assisted TSO v2 option descriptors to a queue.
2215  */
2216 int efx_ef10_tx_tso_desc(struct efx_tx_queue *tx_queue, struct sk_buff *skb,
2217 			 bool *data_mapped)
2218 {
2219 	struct efx_tx_buffer *buffer;
2220 	u16 inner_ipv4_id = 0;
2221 	u16 outer_ipv4_id = 0;
2222 	struct tcphdr *tcp;
2223 	struct iphdr *ip;
2224 	u16 ip_tot_len;
2225 	u32 seqnum;
2226 	u32 mss;
2227 
2228 	EFX_WARN_ON_ONCE_PARANOID(tx_queue->tso_version != 2);
2229 
2230 	mss = skb_shinfo(skb)->gso_size;
2231 
2232 	if (unlikely(mss < 4)) {
2233 		WARN_ONCE(1, "MSS of %u is too small for TSO v2\n", mss);
2234 		return -EINVAL;
2235 	}
2236 
2237 	if (skb->encapsulation) {
2238 		if (!tx_queue->tso_encap)
2239 			return -EINVAL;
2240 		ip = ip_hdr(skb);
2241 		if (ip->version == 4)
2242 			outer_ipv4_id = ntohs(ip->id);
2243 
2244 		ip = inner_ip_hdr(skb);
2245 		tcp = inner_tcp_hdr(skb);
2246 	} else {
2247 		ip = ip_hdr(skb);
2248 		tcp = tcp_hdr(skb);
2249 	}
2250 
2251 	/* 8000-series EF10 hardware requires that IP Total Length be
2252 	 * greater than or equal to the value it will have in each segment
2253 	 * (which is at most mss + 208 + TCP header length), but also less
2254 	 * than (0x10000 - inner_network_header).  Otherwise the TCP
2255 	 * checksum calculation will be broken for encapsulated packets.
2256 	 * We fill in ip->tot_len with 0xff30, which should satisfy the
2257 	 * first requirement unless the MSS is ridiculously large (which
2258 	 * should be impossible as the driver max MTU is 9216); it is
2259 	 * guaranteed to satisfy the second as we only attempt TSO if
2260 	 * inner_network_header <= 208.
2261 	 */
2262 	ip_tot_len = 0x10000 - EFX_TSO2_MAX_HDRLEN;
2263 	EFX_WARN_ON_ONCE_PARANOID(mss + EFX_TSO2_MAX_HDRLEN +
2264 				  (tcp->doff << 2u) > ip_tot_len);
2265 
2266 	if (ip->version == 4) {
2267 		ip->tot_len = htons(ip_tot_len);
2268 		ip->check = 0;
2269 		inner_ipv4_id = ntohs(ip->id);
2270 	} else {
2271 		((struct ipv6hdr *)ip)->payload_len = htons(ip_tot_len);
2272 	}
2273 
2274 	seqnum = ntohl(tcp->seq);
2275 
2276 	buffer = efx_tx_queue_get_insert_buffer(tx_queue);
2277 
2278 	buffer->flags = EFX_TX_BUF_OPTION;
2279 	buffer->len = 0;
2280 	buffer->unmap_len = 0;
2281 	EFX_POPULATE_QWORD_5(buffer->option,
2282 			ESF_DZ_TX_DESC_IS_OPT, 1,
2283 			ESF_DZ_TX_OPTION_TYPE, ESE_DZ_TX_OPTION_DESC_TSO,
2284 			ESF_DZ_TX_TSO_OPTION_TYPE,
2285 			ESE_DZ_TX_TSO_OPTION_DESC_FATSO2A,
2286 			ESF_DZ_TX_TSO_IP_ID, inner_ipv4_id,
2287 			ESF_DZ_TX_TSO_TCP_SEQNO, seqnum
2288 			);
2289 	++tx_queue->insert_count;
2290 
2291 	buffer = efx_tx_queue_get_insert_buffer(tx_queue);
2292 
2293 	buffer->flags = EFX_TX_BUF_OPTION;
2294 	buffer->len = 0;
2295 	buffer->unmap_len = 0;
2296 	EFX_POPULATE_QWORD_5(buffer->option,
2297 			ESF_DZ_TX_DESC_IS_OPT, 1,
2298 			ESF_DZ_TX_OPTION_TYPE, ESE_DZ_TX_OPTION_DESC_TSO,
2299 			ESF_DZ_TX_TSO_OPTION_TYPE,
2300 			ESE_DZ_TX_TSO_OPTION_DESC_FATSO2B,
2301 			ESF_DZ_TX_TSO_OUTER_IPID, outer_ipv4_id,
2302 			ESF_DZ_TX_TSO_TCP_MSS, mss
2303 			);
2304 	++tx_queue->insert_count;
2305 
2306 	return 0;
2307 }
2308 
2309 static u32 efx_ef10_tso_versions(struct efx_nic *efx)
2310 {
2311 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2312 	u32 tso_versions = 0;
2313 
2314 	if (nic_data->datapath_caps &
2315 	    (1 << MC_CMD_GET_CAPABILITIES_OUT_TX_TSO_LBN))
2316 		tso_versions |= BIT(1);
2317 	if (nic_data->datapath_caps2 &
2318 	    (1 << MC_CMD_GET_CAPABILITIES_V2_OUT_TX_TSO_V2_LBN))
2319 		tso_versions |= BIT(2);
2320 	return tso_versions;
2321 }
2322 
2323 static void efx_ef10_tx_init(struct efx_tx_queue *tx_queue)
2324 {
2325 	bool csum_offload = tx_queue->type & EFX_TXQ_TYPE_OUTER_CSUM;
2326 	bool inner_csum = tx_queue->type & EFX_TXQ_TYPE_INNER_CSUM;
2327 	struct efx_channel *channel = tx_queue->channel;
2328 	struct efx_nic *efx = tx_queue->efx;
2329 	struct efx_ef10_nic_data *nic_data;
2330 	efx_qword_t *txd;
2331 	int rc;
2332 
2333 	nic_data = efx->nic_data;
2334 
2335 	/* Only attempt to enable TX timestamping if we have the license for it,
2336 	 * otherwise TXQ init will fail
2337 	 */
2338 	if (!(nic_data->licensed_features &
2339 	      (1 << LICENSED_V3_FEATURES_TX_TIMESTAMPS_LBN))) {
2340 		tx_queue->timestamping = false;
2341 		/* Disable sync events on this channel. */
2342 		if (efx->type->ptp_set_ts_sync_events)
2343 			efx->type->ptp_set_ts_sync_events(efx, false, false);
2344 	}
2345 
2346 	/* TSOv2 is a limited resource that can only be configured on a limited
2347 	 * number of queues. TSO without checksum offload is not really a thing,
2348 	 * so we only enable it for those queues.
2349 	 * TSOv2 cannot be used with Hardware timestamping, and is never needed
2350 	 * for XDP tx.
2351 	 */
2352 	if (efx_has_cap(efx, TX_TSO_V2)) {
2353 		if ((csum_offload || inner_csum) &&
2354 		    !tx_queue->timestamping && !tx_queue->xdp_tx) {
2355 			tx_queue->tso_version = 2;
2356 			netif_dbg(efx, hw, efx->net_dev, "Using TSOv2 for channel %u\n",
2357 				  channel->channel);
2358 		}
2359 	} else if (efx_has_cap(efx, TX_TSO)) {
2360 		tx_queue->tso_version = 1;
2361 	}
2362 
2363 	rc = efx_mcdi_tx_init(tx_queue);
2364 	if (rc)
2365 		goto fail;
2366 
2367 	/* A previous user of this TX queue might have set us up the
2368 	 * bomb by writing a descriptor to the TX push collector but
2369 	 * not the doorbell.  (Each collector belongs to a port, not a
2370 	 * queue or function, so cannot easily be reset.)  We must
2371 	 * attempt to push a no-op descriptor in its place.
2372 	 */
2373 	tx_queue->buffer[0].flags = EFX_TX_BUF_OPTION;
2374 	tx_queue->insert_count = 1;
2375 	txd = efx_tx_desc(tx_queue, 0);
2376 	EFX_POPULATE_QWORD_7(*txd,
2377 			     ESF_DZ_TX_DESC_IS_OPT, true,
2378 			     ESF_DZ_TX_OPTION_TYPE,
2379 			     ESE_DZ_TX_OPTION_DESC_CRC_CSUM,
2380 			     ESF_DZ_TX_OPTION_UDP_TCP_CSUM, csum_offload,
2381 			     ESF_DZ_TX_OPTION_IP_CSUM, csum_offload && tx_queue->tso_version != 2,
2382 			     ESF_DZ_TX_OPTION_INNER_UDP_TCP_CSUM, inner_csum,
2383 			     ESF_DZ_TX_OPTION_INNER_IP_CSUM, inner_csum && tx_queue->tso_version != 2,
2384 			     ESF_DZ_TX_TIMESTAMP, tx_queue->timestamping);
2385 	tx_queue->write_count = 1;
2386 
2387 	if (tx_queue->tso_version == 2 && efx_has_cap(efx, TX_TSO_V2_ENCAP))
2388 		tx_queue->tso_encap = true;
2389 
2390 	wmb();
2391 	efx_ef10_push_tx_desc(tx_queue, txd);
2392 
2393 	return;
2394 
2395 fail:
2396 	netdev_WARN(efx->net_dev, "failed to initialise TXQ %d\n",
2397 		    tx_queue->queue);
2398 }
2399 
2400 /* This writes to the TX_DESC_WPTR; write pointer for TX descriptor ring */
2401 static inline void efx_ef10_notify_tx_desc(struct efx_tx_queue *tx_queue)
2402 {
2403 	unsigned int write_ptr;
2404 	efx_dword_t reg;
2405 
2406 	write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
2407 	EFX_POPULATE_DWORD_1(reg, ERF_DZ_TX_DESC_WPTR_DWORD, write_ptr);
2408 	efx_writed_page(tx_queue->efx, &reg,
2409 			ER_DZ_TX_DESC_UPD_DWORD, tx_queue->queue);
2410 }
2411 
2412 #define EFX_EF10_MAX_TX_DESCRIPTOR_LEN 0x3fff
2413 
2414 static unsigned int efx_ef10_tx_limit_len(struct efx_tx_queue *tx_queue,
2415 					  dma_addr_t dma_addr, unsigned int len)
2416 {
2417 	if (len > EFX_EF10_MAX_TX_DESCRIPTOR_LEN) {
2418 		/* If we need to break across multiple descriptors we should
2419 		 * stop at a page boundary. This assumes the length limit is
2420 		 * greater than the page size.
2421 		 */
2422 		dma_addr_t end = dma_addr + EFX_EF10_MAX_TX_DESCRIPTOR_LEN;
2423 
2424 		BUILD_BUG_ON(EFX_EF10_MAX_TX_DESCRIPTOR_LEN < EFX_PAGE_SIZE);
2425 		len = (end & (~(EFX_PAGE_SIZE - 1))) - dma_addr;
2426 	}
2427 
2428 	return len;
2429 }
2430 
2431 static void efx_ef10_tx_write(struct efx_tx_queue *tx_queue)
2432 {
2433 	unsigned int old_write_count = tx_queue->write_count;
2434 	struct efx_tx_buffer *buffer;
2435 	unsigned int write_ptr;
2436 	efx_qword_t *txd;
2437 
2438 	tx_queue->xmit_pending = false;
2439 	if (unlikely(tx_queue->write_count == tx_queue->insert_count))
2440 		return;
2441 
2442 	do {
2443 		write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
2444 		buffer = &tx_queue->buffer[write_ptr];
2445 		txd = efx_tx_desc(tx_queue, write_ptr);
2446 		++tx_queue->write_count;
2447 
2448 		/* Create TX descriptor ring entry */
2449 		if (buffer->flags & EFX_TX_BUF_OPTION) {
2450 			*txd = buffer->option;
2451 			if (EFX_QWORD_FIELD(*txd, ESF_DZ_TX_OPTION_TYPE) == 1)
2452 				/* PIO descriptor */
2453 				tx_queue->packet_write_count = tx_queue->write_count;
2454 		} else {
2455 			tx_queue->packet_write_count = tx_queue->write_count;
2456 			BUILD_BUG_ON(EFX_TX_BUF_CONT != 1);
2457 			EFX_POPULATE_QWORD_3(
2458 				*txd,
2459 				ESF_DZ_TX_KER_CONT,
2460 				buffer->flags & EFX_TX_BUF_CONT,
2461 				ESF_DZ_TX_KER_BYTE_CNT, buffer->len,
2462 				ESF_DZ_TX_KER_BUF_ADDR, buffer->dma_addr);
2463 		}
2464 	} while (tx_queue->write_count != tx_queue->insert_count);
2465 
2466 	wmb(); /* Ensure descriptors are written before they are fetched */
2467 
2468 	if (efx_nic_may_push_tx_desc(tx_queue, old_write_count)) {
2469 		txd = efx_tx_desc(tx_queue,
2470 				  old_write_count & tx_queue->ptr_mask);
2471 		efx_ef10_push_tx_desc(tx_queue, txd);
2472 		++tx_queue->pushes;
2473 	} else {
2474 		efx_ef10_notify_tx_desc(tx_queue);
2475 	}
2476 }
2477 
2478 static int efx_ef10_probe_multicast_chaining(struct efx_nic *efx)
2479 {
2480 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2481 	unsigned int enabled, implemented;
2482 	bool want_workaround_26807;
2483 	int rc;
2484 
2485 	rc = efx_mcdi_get_workarounds(efx, &implemented, &enabled);
2486 	if (rc == -ENOSYS) {
2487 		/* GET_WORKAROUNDS was implemented before this workaround,
2488 		 * thus it must be unavailable in this firmware.
2489 		 */
2490 		nic_data->workaround_26807 = false;
2491 		return 0;
2492 	}
2493 	if (rc)
2494 		return rc;
2495 	want_workaround_26807 =
2496 		implemented & MC_CMD_GET_WORKAROUNDS_OUT_BUG26807;
2497 	nic_data->workaround_26807 =
2498 		!!(enabled & MC_CMD_GET_WORKAROUNDS_OUT_BUG26807);
2499 
2500 	if (want_workaround_26807 && !nic_data->workaround_26807) {
2501 		unsigned int flags;
2502 
2503 		rc = efx_mcdi_set_workaround(efx,
2504 					     MC_CMD_WORKAROUND_BUG26807,
2505 					     true, &flags);
2506 		if (!rc) {
2507 			if (flags &
2508 			    1 << MC_CMD_WORKAROUND_EXT_OUT_FLR_DONE_LBN) {
2509 				netif_info(efx, drv, efx->net_dev,
2510 					   "other functions on NIC have been reset\n");
2511 
2512 				/* With MCFW v4.6.x and earlier, the
2513 				 * boot count will have incremented,
2514 				 * so re-read the warm_boot_count
2515 				 * value now to ensure this function
2516 				 * doesn't think it has changed next
2517 				 * time it checks.
2518 				 */
2519 				rc = efx_ef10_get_warm_boot_count(efx);
2520 				if (rc >= 0) {
2521 					nic_data->warm_boot_count = rc;
2522 					rc = 0;
2523 				}
2524 			}
2525 			nic_data->workaround_26807 = true;
2526 		} else if (rc == -EPERM) {
2527 			rc = 0;
2528 		}
2529 	}
2530 	return rc;
2531 }
2532 
2533 static int efx_ef10_filter_table_probe(struct efx_nic *efx)
2534 {
2535 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2536 	int rc = efx_ef10_probe_multicast_chaining(efx);
2537 	struct efx_mcdi_filter_vlan *vlan;
2538 
2539 	if (rc)
2540 		return rc;
2541 	down_write(&efx->filter_sem);
2542 	rc = efx_mcdi_filter_table_probe(efx, nic_data->workaround_26807);
2543 
2544 	if (rc)
2545 		goto out_unlock;
2546 
2547 	list_for_each_entry(vlan, &nic_data->vlan_list, list) {
2548 		rc = efx_mcdi_filter_add_vlan(efx, vlan->vid);
2549 		if (rc)
2550 			goto fail_add_vlan;
2551 	}
2552 	goto out_unlock;
2553 
2554 fail_add_vlan:
2555 	efx_mcdi_filter_table_remove(efx);
2556 out_unlock:
2557 	up_write(&efx->filter_sem);
2558 	return rc;
2559 }
2560 
2561 static void efx_ef10_filter_table_remove(struct efx_nic *efx)
2562 {
2563 	down_write(&efx->filter_sem);
2564 	efx_mcdi_filter_table_remove(efx);
2565 	up_write(&efx->filter_sem);
2566 }
2567 
2568 /* This creates an entry in the RX descriptor queue */
2569 static inline void
2570 efx_ef10_build_rx_desc(struct efx_rx_queue *rx_queue, unsigned int index)
2571 {
2572 	struct efx_rx_buffer *rx_buf;
2573 	efx_qword_t *rxd;
2574 
2575 	rxd = efx_rx_desc(rx_queue, index);
2576 	rx_buf = efx_rx_buffer(rx_queue, index);
2577 	EFX_POPULATE_QWORD_2(*rxd,
2578 			     ESF_DZ_RX_KER_BYTE_CNT, rx_buf->len,
2579 			     ESF_DZ_RX_KER_BUF_ADDR, rx_buf->dma_addr);
2580 }
2581 
2582 static void efx_ef10_rx_write(struct efx_rx_queue *rx_queue)
2583 {
2584 	struct efx_nic *efx = rx_queue->efx;
2585 	unsigned int write_count;
2586 	efx_dword_t reg;
2587 
2588 	/* Firmware requires that RX_DESC_WPTR be a multiple of 8 */
2589 	write_count = rx_queue->added_count & ~7;
2590 	if (rx_queue->notified_count == write_count)
2591 		return;
2592 
2593 	do
2594 		efx_ef10_build_rx_desc(
2595 			rx_queue,
2596 			rx_queue->notified_count & rx_queue->ptr_mask);
2597 	while (++rx_queue->notified_count != write_count);
2598 
2599 	wmb();
2600 	EFX_POPULATE_DWORD_1(reg, ERF_DZ_RX_DESC_WPTR,
2601 			     write_count & rx_queue->ptr_mask);
2602 	efx_writed_page(efx, &reg, ER_DZ_RX_DESC_UPD,
2603 			efx_rx_queue_index(rx_queue));
2604 }
2605 
2606 static efx_mcdi_async_completer efx_ef10_rx_defer_refill_complete;
2607 
2608 static void efx_ef10_rx_defer_refill(struct efx_rx_queue *rx_queue)
2609 {
2610 	struct efx_channel *channel = efx_rx_queue_channel(rx_queue);
2611 	MCDI_DECLARE_BUF(inbuf, MC_CMD_DRIVER_EVENT_IN_LEN);
2612 	efx_qword_t event;
2613 
2614 	EFX_POPULATE_QWORD_2(event,
2615 			     ESF_DZ_EV_CODE, EFX_EF10_DRVGEN_EV,
2616 			     ESF_DZ_EV_DATA, EFX_EF10_REFILL);
2617 
2618 	MCDI_SET_DWORD(inbuf, DRIVER_EVENT_IN_EVQ, channel->channel);
2619 
2620 	/* MCDI_SET_QWORD is not appropriate here since EFX_POPULATE_* has
2621 	 * already swapped the data to little-endian order.
2622 	 */
2623 	memcpy(MCDI_PTR(inbuf, DRIVER_EVENT_IN_DATA), &event.u64[0],
2624 	       sizeof(efx_qword_t));
2625 
2626 	efx_mcdi_rpc_async(channel->efx, MC_CMD_DRIVER_EVENT,
2627 			   inbuf, sizeof(inbuf), 0,
2628 			   efx_ef10_rx_defer_refill_complete, 0);
2629 }
2630 
2631 static void
2632 efx_ef10_rx_defer_refill_complete(struct efx_nic *efx, unsigned long cookie,
2633 				  int rc, efx_dword_t *outbuf,
2634 				  size_t outlen_actual)
2635 {
2636 	/* nothing to do */
2637 }
2638 
2639 static int efx_ef10_ev_init(struct efx_channel *channel)
2640 {
2641 	struct efx_nic *efx = channel->efx;
2642 	struct efx_ef10_nic_data *nic_data;
2643 	bool use_v2, cut_thru;
2644 
2645 	nic_data = efx->nic_data;
2646 	use_v2 = nic_data->datapath_caps2 &
2647 			    1 << MC_CMD_GET_CAPABILITIES_V2_OUT_INIT_EVQ_V2_LBN;
2648 	cut_thru = !(nic_data->datapath_caps &
2649 			      1 << MC_CMD_GET_CAPABILITIES_OUT_RX_BATCHING_LBN);
2650 	return efx_mcdi_ev_init(channel, cut_thru, use_v2);
2651 }
2652 
2653 static void efx_ef10_handle_rx_wrong_queue(struct efx_rx_queue *rx_queue,
2654 					   unsigned int rx_queue_label)
2655 {
2656 	struct efx_nic *efx = rx_queue->efx;
2657 
2658 	netif_info(efx, hw, efx->net_dev,
2659 		   "rx event arrived on queue %d labeled as queue %u\n",
2660 		   efx_rx_queue_index(rx_queue), rx_queue_label);
2661 
2662 	efx_schedule_reset(efx, RESET_TYPE_DISABLE);
2663 }
2664 
2665 static void
2666 efx_ef10_handle_rx_bad_lbits(struct efx_rx_queue *rx_queue,
2667 			     unsigned int actual, unsigned int expected)
2668 {
2669 	unsigned int dropped = (actual - expected) & rx_queue->ptr_mask;
2670 	struct efx_nic *efx = rx_queue->efx;
2671 
2672 	netif_info(efx, hw, efx->net_dev,
2673 		   "dropped %d events (index=%d expected=%d)\n",
2674 		   dropped, actual, expected);
2675 
2676 	efx_schedule_reset(efx, RESET_TYPE_DISABLE);
2677 }
2678 
2679 /* partially received RX was aborted. clean up. */
2680 static void efx_ef10_handle_rx_abort(struct efx_rx_queue *rx_queue)
2681 {
2682 	unsigned int rx_desc_ptr;
2683 
2684 	netif_dbg(rx_queue->efx, hw, rx_queue->efx->net_dev,
2685 		  "scattered RX aborted (dropping %u buffers)\n",
2686 		  rx_queue->scatter_n);
2687 
2688 	rx_desc_ptr = rx_queue->removed_count & rx_queue->ptr_mask;
2689 
2690 	efx_rx_packet(rx_queue, rx_desc_ptr, rx_queue->scatter_n,
2691 		      0, EFX_RX_PKT_DISCARD);
2692 
2693 	rx_queue->removed_count += rx_queue->scatter_n;
2694 	rx_queue->scatter_n = 0;
2695 	rx_queue->scatter_len = 0;
2696 	++efx_rx_queue_channel(rx_queue)->n_rx_nodesc_trunc;
2697 }
2698 
2699 static u16 efx_ef10_handle_rx_event_errors(struct efx_channel *channel,
2700 					   unsigned int n_packets,
2701 					   unsigned int rx_encap_hdr,
2702 					   unsigned int rx_l3_class,
2703 					   unsigned int rx_l4_class,
2704 					   const efx_qword_t *event)
2705 {
2706 	struct efx_nic *efx = channel->efx;
2707 	bool handled = false;
2708 
2709 	if (EFX_QWORD_FIELD(*event, ESF_DZ_RX_ECRC_ERR)) {
2710 		if (!(efx->net_dev->features & NETIF_F_RXALL)) {
2711 			if (!efx->loopback_selftest)
2712 				channel->n_rx_eth_crc_err += n_packets;
2713 			return EFX_RX_PKT_DISCARD;
2714 		}
2715 		handled = true;
2716 	}
2717 	if (EFX_QWORD_FIELD(*event, ESF_DZ_RX_IPCKSUM_ERR)) {
2718 		if (unlikely(rx_encap_hdr != ESE_EZ_ENCAP_HDR_VXLAN &&
2719 			     rx_l3_class != ESE_DZ_L3_CLASS_IP4 &&
2720 			     rx_l3_class != ESE_DZ_L3_CLASS_IP4_FRAG &&
2721 			     rx_l3_class != ESE_DZ_L3_CLASS_IP6 &&
2722 			     rx_l3_class != ESE_DZ_L3_CLASS_IP6_FRAG))
2723 			netdev_WARN(efx->net_dev,
2724 				    "invalid class for RX_IPCKSUM_ERR: event="
2725 				    EFX_QWORD_FMT "\n",
2726 				    EFX_QWORD_VAL(*event));
2727 		if (!efx->loopback_selftest)
2728 			*(rx_encap_hdr ?
2729 			  &channel->n_rx_outer_ip_hdr_chksum_err :
2730 			  &channel->n_rx_ip_hdr_chksum_err) += n_packets;
2731 		return 0;
2732 	}
2733 	if (EFX_QWORD_FIELD(*event, ESF_DZ_RX_TCPUDP_CKSUM_ERR)) {
2734 		if (unlikely(rx_encap_hdr != ESE_EZ_ENCAP_HDR_VXLAN &&
2735 			     ((rx_l3_class != ESE_DZ_L3_CLASS_IP4 &&
2736 			       rx_l3_class != ESE_DZ_L3_CLASS_IP6) ||
2737 			      (rx_l4_class != ESE_FZ_L4_CLASS_TCP &&
2738 			       rx_l4_class != ESE_FZ_L4_CLASS_UDP))))
2739 			netdev_WARN(efx->net_dev,
2740 				    "invalid class for RX_TCPUDP_CKSUM_ERR: event="
2741 				    EFX_QWORD_FMT "\n",
2742 				    EFX_QWORD_VAL(*event));
2743 		if (!efx->loopback_selftest)
2744 			*(rx_encap_hdr ?
2745 			  &channel->n_rx_outer_tcp_udp_chksum_err :
2746 			  &channel->n_rx_tcp_udp_chksum_err) += n_packets;
2747 		return 0;
2748 	}
2749 	if (EFX_QWORD_FIELD(*event, ESF_EZ_RX_IP_INNER_CHKSUM_ERR)) {
2750 		if (unlikely(!rx_encap_hdr))
2751 			netdev_WARN(efx->net_dev,
2752 				    "invalid encapsulation type for RX_IP_INNER_CHKSUM_ERR: event="
2753 				    EFX_QWORD_FMT "\n",
2754 				    EFX_QWORD_VAL(*event));
2755 		else if (unlikely(rx_l3_class != ESE_DZ_L3_CLASS_IP4 &&
2756 				  rx_l3_class != ESE_DZ_L3_CLASS_IP4_FRAG &&
2757 				  rx_l3_class != ESE_DZ_L3_CLASS_IP6 &&
2758 				  rx_l3_class != ESE_DZ_L3_CLASS_IP6_FRAG))
2759 			netdev_WARN(efx->net_dev,
2760 				    "invalid class for RX_IP_INNER_CHKSUM_ERR: event="
2761 				    EFX_QWORD_FMT "\n",
2762 				    EFX_QWORD_VAL(*event));
2763 		if (!efx->loopback_selftest)
2764 			channel->n_rx_inner_ip_hdr_chksum_err += n_packets;
2765 		return 0;
2766 	}
2767 	if (EFX_QWORD_FIELD(*event, ESF_EZ_RX_TCP_UDP_INNER_CHKSUM_ERR)) {
2768 		if (unlikely(!rx_encap_hdr))
2769 			netdev_WARN(efx->net_dev,
2770 				    "invalid encapsulation type for RX_TCP_UDP_INNER_CHKSUM_ERR: event="
2771 				    EFX_QWORD_FMT "\n",
2772 				    EFX_QWORD_VAL(*event));
2773 		else if (unlikely((rx_l3_class != ESE_DZ_L3_CLASS_IP4 &&
2774 				   rx_l3_class != ESE_DZ_L3_CLASS_IP6) ||
2775 				  (rx_l4_class != ESE_FZ_L4_CLASS_TCP &&
2776 				   rx_l4_class != ESE_FZ_L4_CLASS_UDP)))
2777 			netdev_WARN(efx->net_dev,
2778 				    "invalid class for RX_TCP_UDP_INNER_CHKSUM_ERR: event="
2779 				    EFX_QWORD_FMT "\n",
2780 				    EFX_QWORD_VAL(*event));
2781 		if (!efx->loopback_selftest)
2782 			channel->n_rx_inner_tcp_udp_chksum_err += n_packets;
2783 		return 0;
2784 	}
2785 
2786 	WARN_ON(!handled); /* No error bits were recognised */
2787 	return 0;
2788 }
2789 
2790 static int efx_ef10_handle_rx_event(struct efx_channel *channel,
2791 				    const efx_qword_t *event)
2792 {
2793 	unsigned int rx_bytes, next_ptr_lbits, rx_queue_label;
2794 	unsigned int rx_l3_class, rx_l4_class, rx_encap_hdr;
2795 	unsigned int n_descs, n_packets, i;
2796 	struct efx_nic *efx = channel->efx;
2797 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2798 	struct efx_rx_queue *rx_queue;
2799 	efx_qword_t errors;
2800 	bool rx_cont;
2801 	u16 flags = 0;
2802 
2803 	if (unlikely(READ_ONCE(efx->reset_pending)))
2804 		return 0;
2805 
2806 	/* Basic packet information */
2807 	rx_bytes = EFX_QWORD_FIELD(*event, ESF_DZ_RX_BYTES);
2808 	next_ptr_lbits = EFX_QWORD_FIELD(*event, ESF_DZ_RX_DSC_PTR_LBITS);
2809 	rx_queue_label = EFX_QWORD_FIELD(*event, ESF_DZ_RX_QLABEL);
2810 	rx_l3_class = EFX_QWORD_FIELD(*event, ESF_DZ_RX_L3_CLASS);
2811 	rx_l4_class = EFX_QWORD_FIELD(*event, ESF_FZ_RX_L4_CLASS);
2812 	rx_cont = EFX_QWORD_FIELD(*event, ESF_DZ_RX_CONT);
2813 	rx_encap_hdr =
2814 		nic_data->datapath_caps &
2815 			(1 << MC_CMD_GET_CAPABILITIES_OUT_VXLAN_NVGRE_LBN) ?
2816 		EFX_QWORD_FIELD(*event, ESF_EZ_RX_ENCAP_HDR) :
2817 		ESE_EZ_ENCAP_HDR_NONE;
2818 
2819 	if (EFX_QWORD_FIELD(*event, ESF_DZ_RX_DROP_EVENT))
2820 		netdev_WARN(efx->net_dev, "saw RX_DROP_EVENT: event="
2821 			    EFX_QWORD_FMT "\n",
2822 			    EFX_QWORD_VAL(*event));
2823 
2824 	rx_queue = efx_channel_get_rx_queue(channel);
2825 
2826 	if (unlikely(rx_queue_label != efx_rx_queue_index(rx_queue)))
2827 		efx_ef10_handle_rx_wrong_queue(rx_queue, rx_queue_label);
2828 
2829 	n_descs = ((next_ptr_lbits - rx_queue->removed_count) &
2830 		   ((1 << ESF_DZ_RX_DSC_PTR_LBITS_WIDTH) - 1));
2831 
2832 	if (n_descs != rx_queue->scatter_n + 1) {
2833 		struct efx_ef10_nic_data *nic_data = efx->nic_data;
2834 
2835 		/* detect rx abort */
2836 		if (unlikely(n_descs == rx_queue->scatter_n)) {
2837 			if (rx_queue->scatter_n == 0 || rx_bytes != 0)
2838 				netdev_WARN(efx->net_dev,
2839 					    "invalid RX abort: scatter_n=%u event="
2840 					    EFX_QWORD_FMT "\n",
2841 					    rx_queue->scatter_n,
2842 					    EFX_QWORD_VAL(*event));
2843 			efx_ef10_handle_rx_abort(rx_queue);
2844 			return 0;
2845 		}
2846 
2847 		/* Check that RX completion merging is valid, i.e.
2848 		 * the current firmware supports it and this is a
2849 		 * non-scattered packet.
2850 		 */
2851 		if (!(nic_data->datapath_caps &
2852 		      (1 << MC_CMD_GET_CAPABILITIES_OUT_RX_BATCHING_LBN)) ||
2853 		    rx_queue->scatter_n != 0 || rx_cont) {
2854 			efx_ef10_handle_rx_bad_lbits(
2855 				rx_queue, next_ptr_lbits,
2856 				(rx_queue->removed_count +
2857 				 rx_queue->scatter_n + 1) &
2858 				((1 << ESF_DZ_RX_DSC_PTR_LBITS_WIDTH) - 1));
2859 			return 0;
2860 		}
2861 
2862 		/* Merged completion for multiple non-scattered packets */
2863 		rx_queue->scatter_n = 1;
2864 		rx_queue->scatter_len = 0;
2865 		n_packets = n_descs;
2866 		++channel->n_rx_merge_events;
2867 		channel->n_rx_merge_packets += n_packets;
2868 		flags |= EFX_RX_PKT_PREFIX_LEN;
2869 	} else {
2870 		++rx_queue->scatter_n;
2871 		rx_queue->scatter_len += rx_bytes;
2872 		if (rx_cont)
2873 			return 0;
2874 		n_packets = 1;
2875 	}
2876 
2877 	EFX_POPULATE_QWORD_5(errors, ESF_DZ_RX_ECRC_ERR, 1,
2878 				     ESF_DZ_RX_IPCKSUM_ERR, 1,
2879 				     ESF_DZ_RX_TCPUDP_CKSUM_ERR, 1,
2880 				     ESF_EZ_RX_IP_INNER_CHKSUM_ERR, 1,
2881 				     ESF_EZ_RX_TCP_UDP_INNER_CHKSUM_ERR, 1);
2882 	EFX_AND_QWORD(errors, *event, errors);
2883 	if (unlikely(!EFX_QWORD_IS_ZERO(errors))) {
2884 		flags |= efx_ef10_handle_rx_event_errors(channel, n_packets,
2885 							 rx_encap_hdr,
2886 							 rx_l3_class, rx_l4_class,
2887 							 event);
2888 	} else {
2889 		bool tcpudp = rx_l4_class == ESE_FZ_L4_CLASS_TCP ||
2890 			      rx_l4_class == ESE_FZ_L4_CLASS_UDP;
2891 
2892 		switch (rx_encap_hdr) {
2893 		case ESE_EZ_ENCAP_HDR_VXLAN: /* VxLAN or GENEVE */
2894 			flags |= EFX_RX_PKT_CSUMMED; /* outer UDP csum */
2895 			if (tcpudp)
2896 				flags |= EFX_RX_PKT_CSUM_LEVEL; /* inner L4 */
2897 			break;
2898 		case ESE_EZ_ENCAP_HDR_GRE:
2899 		case ESE_EZ_ENCAP_HDR_NONE:
2900 			if (tcpudp)
2901 				flags |= EFX_RX_PKT_CSUMMED;
2902 			break;
2903 		default:
2904 			netdev_WARN(efx->net_dev,
2905 				    "unknown encapsulation type: event="
2906 				    EFX_QWORD_FMT "\n",
2907 				    EFX_QWORD_VAL(*event));
2908 		}
2909 	}
2910 
2911 	if (rx_l4_class == ESE_FZ_L4_CLASS_TCP)
2912 		flags |= EFX_RX_PKT_TCP;
2913 
2914 	channel->irq_mod_score += 2 * n_packets;
2915 
2916 	/* Handle received packet(s) */
2917 	for (i = 0; i < n_packets; i++) {
2918 		efx_rx_packet(rx_queue,
2919 			      rx_queue->removed_count & rx_queue->ptr_mask,
2920 			      rx_queue->scatter_n, rx_queue->scatter_len,
2921 			      flags);
2922 		rx_queue->removed_count += rx_queue->scatter_n;
2923 	}
2924 
2925 	rx_queue->scatter_n = 0;
2926 	rx_queue->scatter_len = 0;
2927 
2928 	return n_packets;
2929 }
2930 
2931 static u32 efx_ef10_extract_event_ts(efx_qword_t *event)
2932 {
2933 	u32 tstamp;
2934 
2935 	tstamp = EFX_QWORD_FIELD(*event, TX_TIMESTAMP_EVENT_TSTAMP_DATA_HI);
2936 	tstamp <<= 16;
2937 	tstamp |= EFX_QWORD_FIELD(*event, TX_TIMESTAMP_EVENT_TSTAMP_DATA_LO);
2938 
2939 	return tstamp;
2940 }
2941 
2942 static void
2943 efx_ef10_handle_tx_event(struct efx_channel *channel, efx_qword_t *event)
2944 {
2945 	struct efx_nic *efx = channel->efx;
2946 	struct efx_tx_queue *tx_queue;
2947 	unsigned int tx_ev_desc_ptr;
2948 	unsigned int tx_ev_q_label;
2949 	unsigned int tx_ev_type;
2950 	u64 ts_part;
2951 
2952 	if (unlikely(READ_ONCE(efx->reset_pending)))
2953 		return;
2954 
2955 	if (unlikely(EFX_QWORD_FIELD(*event, ESF_DZ_TX_DROP_EVENT)))
2956 		return;
2957 
2958 	/* Get the transmit queue */
2959 	tx_ev_q_label = EFX_QWORD_FIELD(*event, ESF_DZ_TX_QLABEL);
2960 	tx_queue = channel->tx_queue + (tx_ev_q_label % EFX_MAX_TXQ_PER_CHANNEL);
2961 
2962 	if (!tx_queue->timestamping) {
2963 		/* Transmit completion */
2964 		tx_ev_desc_ptr = EFX_QWORD_FIELD(*event, ESF_DZ_TX_DESCR_INDX);
2965 		efx_xmit_done(tx_queue, tx_ev_desc_ptr & tx_queue->ptr_mask);
2966 		return;
2967 	}
2968 
2969 	/* Transmit timestamps are only available for 8XXX series. They result
2970 	 * in up to three events per packet. These occur in order, and are:
2971 	 *  - the normal completion event (may be omitted)
2972 	 *  - the low part of the timestamp
2973 	 *  - the high part of the timestamp
2974 	 *
2975 	 * It's possible for multiple completion events to appear before the
2976 	 * corresponding timestamps. So we can for example get:
2977 	 *  COMP N
2978 	 *  COMP N+1
2979 	 *  TS_LO N
2980 	 *  TS_HI N
2981 	 *  TS_LO N+1
2982 	 *  TS_HI N+1
2983 	 *
2984 	 * In addition it's also possible for the adjacent completions to be
2985 	 * merged, so we may not see COMP N above. As such, the completion
2986 	 * events are not very useful here.
2987 	 *
2988 	 * Each part of the timestamp is itself split across two 16 bit
2989 	 * fields in the event.
2990 	 */
2991 	tx_ev_type = EFX_QWORD_FIELD(*event, ESF_EZ_TX_SOFT1);
2992 
2993 	switch (tx_ev_type) {
2994 	case TX_TIMESTAMP_EVENT_TX_EV_COMPLETION:
2995 		/* Ignore this event - see above. */
2996 		break;
2997 
2998 	case TX_TIMESTAMP_EVENT_TX_EV_TSTAMP_LO:
2999 		ts_part = efx_ef10_extract_event_ts(event);
3000 		tx_queue->completed_timestamp_minor = ts_part;
3001 		break;
3002 
3003 	case TX_TIMESTAMP_EVENT_TX_EV_TSTAMP_HI:
3004 		ts_part = efx_ef10_extract_event_ts(event);
3005 		tx_queue->completed_timestamp_major = ts_part;
3006 
3007 		efx_xmit_done_single(tx_queue);
3008 		break;
3009 
3010 	default:
3011 		netif_err(efx, hw, efx->net_dev,
3012 			  "channel %d unknown tx event type %d (data "
3013 			  EFX_QWORD_FMT ")\n",
3014 			  channel->channel, tx_ev_type,
3015 			  EFX_QWORD_VAL(*event));
3016 		break;
3017 	}
3018 }
3019 
3020 static void
3021 efx_ef10_handle_driver_event(struct efx_channel *channel, efx_qword_t *event)
3022 {
3023 	struct efx_nic *efx = channel->efx;
3024 	int subcode;
3025 
3026 	subcode = EFX_QWORD_FIELD(*event, ESF_DZ_DRV_SUB_CODE);
3027 
3028 	switch (subcode) {
3029 	case ESE_DZ_DRV_TIMER_EV:
3030 	case ESE_DZ_DRV_WAKE_UP_EV:
3031 		break;
3032 	case ESE_DZ_DRV_START_UP_EV:
3033 		/* event queue init complete. ok. */
3034 		break;
3035 	default:
3036 		netif_err(efx, hw, efx->net_dev,
3037 			  "channel %d unknown driver event type %d"
3038 			  " (data " EFX_QWORD_FMT ")\n",
3039 			  channel->channel, subcode,
3040 			  EFX_QWORD_VAL(*event));
3041 
3042 	}
3043 }
3044 
3045 static void efx_ef10_handle_driver_generated_event(struct efx_channel *channel,
3046 						   efx_qword_t *event)
3047 {
3048 	struct efx_nic *efx = channel->efx;
3049 	u32 subcode;
3050 
3051 	subcode = EFX_QWORD_FIELD(*event, EFX_DWORD_0);
3052 
3053 	switch (subcode) {
3054 	case EFX_EF10_TEST:
3055 		channel->event_test_cpu = raw_smp_processor_id();
3056 		break;
3057 	case EFX_EF10_REFILL:
3058 		/* The queue must be empty, so we won't receive any rx
3059 		 * events, so efx_process_channel() won't refill the
3060 		 * queue. Refill it here
3061 		 */
3062 		efx_fast_push_rx_descriptors(&channel->rx_queue, true);
3063 		break;
3064 	default:
3065 		netif_err(efx, hw, efx->net_dev,
3066 			  "channel %d unknown driver event type %u"
3067 			  " (data " EFX_QWORD_FMT ")\n",
3068 			  channel->channel, (unsigned) subcode,
3069 			  EFX_QWORD_VAL(*event));
3070 	}
3071 }
3072 
3073 static int efx_ef10_ev_process(struct efx_channel *channel, int quota)
3074 {
3075 	struct efx_nic *efx = channel->efx;
3076 	efx_qword_t event, *p_event;
3077 	unsigned int read_ptr;
3078 	int ev_code;
3079 	int spent = 0;
3080 
3081 	if (quota <= 0)
3082 		return spent;
3083 
3084 	read_ptr = channel->eventq_read_ptr;
3085 
3086 	for (;;) {
3087 		p_event = efx_event(channel, read_ptr);
3088 		event = *p_event;
3089 
3090 		if (!efx_event_present(&event))
3091 			break;
3092 
3093 		EFX_SET_QWORD(*p_event);
3094 
3095 		++read_ptr;
3096 
3097 		ev_code = EFX_QWORD_FIELD(event, ESF_DZ_EV_CODE);
3098 
3099 		netif_vdbg(efx, drv, efx->net_dev,
3100 			   "processing event on %d " EFX_QWORD_FMT "\n",
3101 			   channel->channel, EFX_QWORD_VAL(event));
3102 
3103 		switch (ev_code) {
3104 		case ESE_DZ_EV_CODE_MCDI_EV:
3105 			efx_mcdi_process_event(channel, &event);
3106 			break;
3107 		case ESE_DZ_EV_CODE_RX_EV:
3108 			spent += efx_ef10_handle_rx_event(channel, &event);
3109 			if (spent >= quota) {
3110 				/* XXX can we split a merged event to
3111 				 * avoid going over-quota?
3112 				 */
3113 				spent = quota;
3114 				goto out;
3115 			}
3116 			break;
3117 		case ESE_DZ_EV_CODE_TX_EV:
3118 			efx_ef10_handle_tx_event(channel, &event);
3119 			break;
3120 		case ESE_DZ_EV_CODE_DRIVER_EV:
3121 			efx_ef10_handle_driver_event(channel, &event);
3122 			if (++spent == quota)
3123 				goto out;
3124 			break;
3125 		case EFX_EF10_DRVGEN_EV:
3126 			efx_ef10_handle_driver_generated_event(channel, &event);
3127 			break;
3128 		default:
3129 			netif_err(efx, hw, efx->net_dev,
3130 				  "channel %d unknown event type %d"
3131 				  " (data " EFX_QWORD_FMT ")\n",
3132 				  channel->channel, ev_code,
3133 				  EFX_QWORD_VAL(event));
3134 		}
3135 	}
3136 
3137 out:
3138 	channel->eventq_read_ptr = read_ptr;
3139 	return spent;
3140 }
3141 
3142 static void efx_ef10_ev_read_ack(struct efx_channel *channel)
3143 {
3144 	struct efx_nic *efx = channel->efx;
3145 	efx_dword_t rptr;
3146 
3147 	if (EFX_EF10_WORKAROUND_35388(efx)) {
3148 		BUILD_BUG_ON(EFX_MIN_EVQ_SIZE <
3149 			     (1 << ERF_DD_EVQ_IND_RPTR_WIDTH));
3150 		BUILD_BUG_ON(EFX_MAX_EVQ_SIZE >
3151 			     (1 << 2 * ERF_DD_EVQ_IND_RPTR_WIDTH));
3152 
3153 		EFX_POPULATE_DWORD_2(rptr, ERF_DD_EVQ_IND_RPTR_FLAGS,
3154 				     EFE_DD_EVQ_IND_RPTR_FLAGS_HIGH,
3155 				     ERF_DD_EVQ_IND_RPTR,
3156 				     (channel->eventq_read_ptr &
3157 				      channel->eventq_mask) >>
3158 				     ERF_DD_EVQ_IND_RPTR_WIDTH);
3159 		efx_writed_page(efx, &rptr, ER_DD_EVQ_INDIRECT,
3160 				channel->channel);
3161 		EFX_POPULATE_DWORD_2(rptr, ERF_DD_EVQ_IND_RPTR_FLAGS,
3162 				     EFE_DD_EVQ_IND_RPTR_FLAGS_LOW,
3163 				     ERF_DD_EVQ_IND_RPTR,
3164 				     channel->eventq_read_ptr &
3165 				     ((1 << ERF_DD_EVQ_IND_RPTR_WIDTH) - 1));
3166 		efx_writed_page(efx, &rptr, ER_DD_EVQ_INDIRECT,
3167 				channel->channel);
3168 	} else {
3169 		EFX_POPULATE_DWORD_1(rptr, ERF_DZ_EVQ_RPTR,
3170 				     channel->eventq_read_ptr &
3171 				     channel->eventq_mask);
3172 		efx_writed_page(efx, &rptr, ER_DZ_EVQ_RPTR, channel->channel);
3173 	}
3174 }
3175 
3176 static void efx_ef10_ev_test_generate(struct efx_channel *channel)
3177 {
3178 	MCDI_DECLARE_BUF(inbuf, MC_CMD_DRIVER_EVENT_IN_LEN);
3179 	struct efx_nic *efx = channel->efx;
3180 	efx_qword_t event;
3181 	int rc;
3182 
3183 	EFX_POPULATE_QWORD_2(event,
3184 			     ESF_DZ_EV_CODE, EFX_EF10_DRVGEN_EV,
3185 			     ESF_DZ_EV_DATA, EFX_EF10_TEST);
3186 
3187 	MCDI_SET_DWORD(inbuf, DRIVER_EVENT_IN_EVQ, channel->channel);
3188 
3189 	/* MCDI_SET_QWORD is not appropriate here since EFX_POPULATE_* has
3190 	 * already swapped the data to little-endian order.
3191 	 */
3192 	memcpy(MCDI_PTR(inbuf, DRIVER_EVENT_IN_DATA), &event.u64[0],
3193 	       sizeof(efx_qword_t));
3194 
3195 	rc = efx_mcdi_rpc(efx, MC_CMD_DRIVER_EVENT, inbuf, sizeof(inbuf),
3196 			  NULL, 0, NULL);
3197 	if (rc != 0)
3198 		goto fail;
3199 
3200 	return;
3201 
3202 fail:
3203 	WARN_ON(true);
3204 	netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
3205 }
3206 
3207 static void efx_ef10_prepare_flr(struct efx_nic *efx)
3208 {
3209 	atomic_set(&efx->active_queues, 0);
3210 }
3211 
3212 static int efx_ef10_vport_set_mac_address(struct efx_nic *efx)
3213 {
3214 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
3215 	u8 mac_old[ETH_ALEN];
3216 	int rc, rc2;
3217 
3218 	/* Only reconfigure a PF-created vport */
3219 	if (is_zero_ether_addr(nic_data->vport_mac))
3220 		return 0;
3221 
3222 	efx_device_detach_sync(efx);
3223 	efx_net_stop(efx->net_dev);
3224 	efx_ef10_filter_table_remove(efx);
3225 
3226 	rc = efx_ef10_vadaptor_free(efx, efx->vport_id);
3227 	if (rc)
3228 		goto restore_filters;
3229 
3230 	ether_addr_copy(mac_old, nic_data->vport_mac);
3231 	rc = efx_ef10_vport_del_mac(efx, efx->vport_id,
3232 				    nic_data->vport_mac);
3233 	if (rc)
3234 		goto restore_vadaptor;
3235 
3236 	rc = efx_ef10_vport_add_mac(efx, efx->vport_id,
3237 				    efx->net_dev->dev_addr);
3238 	if (!rc) {
3239 		ether_addr_copy(nic_data->vport_mac, efx->net_dev->dev_addr);
3240 	} else {
3241 		rc2 = efx_ef10_vport_add_mac(efx, efx->vport_id, mac_old);
3242 		if (rc2) {
3243 			/* Failed to add original MAC, so clear vport_mac */
3244 			eth_zero_addr(nic_data->vport_mac);
3245 			goto reset_nic;
3246 		}
3247 	}
3248 
3249 restore_vadaptor:
3250 	rc2 = efx_ef10_vadaptor_alloc(efx, efx->vport_id);
3251 	if (rc2)
3252 		goto reset_nic;
3253 restore_filters:
3254 	rc2 = efx_ef10_filter_table_probe(efx);
3255 	if (rc2)
3256 		goto reset_nic;
3257 
3258 	rc2 = efx_net_open(efx->net_dev);
3259 	if (rc2)
3260 		goto reset_nic;
3261 
3262 	efx_device_attach_if_not_resetting(efx);
3263 
3264 	return rc;
3265 
3266 reset_nic:
3267 	netif_err(efx, drv, efx->net_dev,
3268 		  "Failed to restore when changing MAC address - scheduling reset\n");
3269 	efx_schedule_reset(efx, RESET_TYPE_DATAPATH);
3270 
3271 	return rc ? rc : rc2;
3272 }
3273 
3274 static int efx_ef10_set_mac_address(struct efx_nic *efx)
3275 {
3276 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VADAPTOR_SET_MAC_IN_LEN);
3277 	bool was_enabled = efx->port_enabled;
3278 	int rc;
3279 
3280 #ifdef CONFIG_SFC_SRIOV
3281 	/* If this function is a VF and we have access to the parent PF,
3282 	 * then use the PF control path to attempt to change the VF MAC address.
3283 	 */
3284 	if (efx->pci_dev->is_virtfn && efx->pci_dev->physfn) {
3285 		struct efx_nic *efx_pf = pci_get_drvdata(efx->pci_dev->physfn);
3286 		struct efx_ef10_nic_data *nic_data = efx->nic_data;
3287 		u8 mac[ETH_ALEN];
3288 
3289 		/* net_dev->dev_addr can be zeroed by efx_net_stop in
3290 		 * efx_ef10_sriov_set_vf_mac, so pass in a copy.
3291 		 */
3292 		ether_addr_copy(mac, efx->net_dev->dev_addr);
3293 
3294 		rc = efx_ef10_sriov_set_vf_mac(efx_pf, nic_data->vf_index, mac);
3295 		if (!rc)
3296 			return 0;
3297 
3298 		netif_dbg(efx, drv, efx->net_dev,
3299 			  "Updating VF mac via PF failed (%d), setting directly\n",
3300 			  rc);
3301 	}
3302 #endif
3303 
3304 	efx_device_detach_sync(efx);
3305 	efx_net_stop(efx->net_dev);
3306 
3307 	mutex_lock(&efx->mac_lock);
3308 	efx_ef10_filter_table_remove(efx);
3309 
3310 	ether_addr_copy(MCDI_PTR(inbuf, VADAPTOR_SET_MAC_IN_MACADDR),
3311 			efx->net_dev->dev_addr);
3312 	MCDI_SET_DWORD(inbuf, VADAPTOR_SET_MAC_IN_UPSTREAM_PORT_ID,
3313 		       efx->vport_id);
3314 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_VADAPTOR_SET_MAC, inbuf,
3315 				sizeof(inbuf), NULL, 0, NULL);
3316 
3317 	efx_ef10_filter_table_probe(efx);
3318 	mutex_unlock(&efx->mac_lock);
3319 
3320 	if (was_enabled)
3321 		efx_net_open(efx->net_dev);
3322 	efx_device_attach_if_not_resetting(efx);
3323 
3324 	if (rc == -EPERM) {
3325 		netif_err(efx, drv, efx->net_dev,
3326 			  "Cannot change MAC address; use sfboot to enable"
3327 			  " mac-spoofing on this interface\n");
3328 	} else if (rc == -ENOSYS && !efx_ef10_is_vf(efx)) {
3329 		/* If the active MCFW does not support MC_CMD_VADAPTOR_SET_MAC
3330 		 * fall-back to the method of changing the MAC address on the
3331 		 * vport.  This only applies to PFs because such versions of
3332 		 * MCFW do not support VFs.
3333 		 */
3334 		rc = efx_ef10_vport_set_mac_address(efx);
3335 	} else if (rc) {
3336 		efx_mcdi_display_error(efx, MC_CMD_VADAPTOR_SET_MAC,
3337 				       sizeof(inbuf), NULL, 0, rc);
3338 	}
3339 
3340 	return rc;
3341 }
3342 
3343 static int efx_ef10_mac_reconfigure(struct efx_nic *efx, bool mtu_only)
3344 {
3345 	WARN_ON(!mutex_is_locked(&efx->mac_lock));
3346 
3347 	efx_mcdi_filter_sync_rx_mode(efx);
3348 
3349 	if (mtu_only && efx_has_cap(efx, SET_MAC_ENHANCED))
3350 		return efx_mcdi_set_mtu(efx);
3351 	return efx_mcdi_set_mac(efx);
3352 }
3353 
3354 static int efx_ef10_start_bist(struct efx_nic *efx, u32 bist_type)
3355 {
3356 	MCDI_DECLARE_BUF(inbuf, MC_CMD_START_BIST_IN_LEN);
3357 
3358 	MCDI_SET_DWORD(inbuf, START_BIST_IN_TYPE, bist_type);
3359 	return efx_mcdi_rpc(efx, MC_CMD_START_BIST, inbuf, sizeof(inbuf),
3360 			    NULL, 0, NULL);
3361 }
3362 
3363 /* MC BISTs follow a different poll mechanism to phy BISTs.
3364  * The BIST is done in the poll handler on the MC, and the MCDI command
3365  * will block until the BIST is done.
3366  */
3367 static int efx_ef10_poll_bist(struct efx_nic *efx)
3368 {
3369 	int rc;
3370 	MCDI_DECLARE_BUF(outbuf, MC_CMD_POLL_BIST_OUT_LEN);
3371 	size_t outlen;
3372 	u32 result;
3373 
3374 	rc = efx_mcdi_rpc(efx, MC_CMD_POLL_BIST, NULL, 0,
3375 			   outbuf, sizeof(outbuf), &outlen);
3376 	if (rc != 0)
3377 		return rc;
3378 
3379 	if (outlen < MC_CMD_POLL_BIST_OUT_LEN)
3380 		return -EIO;
3381 
3382 	result = MCDI_DWORD(outbuf, POLL_BIST_OUT_RESULT);
3383 	switch (result) {
3384 	case MC_CMD_POLL_BIST_PASSED:
3385 		netif_dbg(efx, hw, efx->net_dev, "BIST passed.\n");
3386 		return 0;
3387 	case MC_CMD_POLL_BIST_TIMEOUT:
3388 		netif_err(efx, hw, efx->net_dev, "BIST timed out\n");
3389 		return -EIO;
3390 	case MC_CMD_POLL_BIST_FAILED:
3391 		netif_err(efx, hw, efx->net_dev, "BIST failed.\n");
3392 		return -EIO;
3393 	default:
3394 		netif_err(efx, hw, efx->net_dev,
3395 			  "BIST returned unknown result %u", result);
3396 		return -EIO;
3397 	}
3398 }
3399 
3400 static int efx_ef10_run_bist(struct efx_nic *efx, u32 bist_type)
3401 {
3402 	int rc;
3403 
3404 	netif_dbg(efx, drv, efx->net_dev, "starting BIST type %u\n", bist_type);
3405 
3406 	rc = efx_ef10_start_bist(efx, bist_type);
3407 	if (rc != 0)
3408 		return rc;
3409 
3410 	return efx_ef10_poll_bist(efx);
3411 }
3412 
3413 static int
3414 efx_ef10_test_chip(struct efx_nic *efx, struct efx_self_tests *tests)
3415 {
3416 	int rc, rc2;
3417 
3418 	efx_reset_down(efx, RESET_TYPE_WORLD);
3419 
3420 	rc = efx_mcdi_rpc(efx, MC_CMD_ENABLE_OFFLINE_BIST,
3421 			  NULL, 0, NULL, 0, NULL);
3422 	if (rc != 0)
3423 		goto out;
3424 
3425 	tests->memory = efx_ef10_run_bist(efx, MC_CMD_MC_MEM_BIST) ? -1 : 1;
3426 	tests->registers = efx_ef10_run_bist(efx, MC_CMD_REG_BIST) ? -1 : 1;
3427 
3428 	rc = efx_mcdi_reset(efx, RESET_TYPE_WORLD);
3429 
3430 out:
3431 	if (rc == -EPERM)
3432 		rc = 0;
3433 	rc2 = efx_reset_up(efx, RESET_TYPE_WORLD, rc == 0);
3434 	return rc ? rc : rc2;
3435 }
3436 
3437 #ifdef CONFIG_SFC_MTD
3438 
3439 struct efx_ef10_nvram_type_info {
3440 	u16 type, type_mask;
3441 	u8 port;
3442 	const char *name;
3443 };
3444 
3445 static const struct efx_ef10_nvram_type_info efx_ef10_nvram_types[] = {
3446 	{ NVRAM_PARTITION_TYPE_MC_FIRMWARE,	   0,    0, "sfc_mcfw" },
3447 	{ NVRAM_PARTITION_TYPE_MC_FIRMWARE_BACKUP, 0,    0, "sfc_mcfw_backup" },
3448 	{ NVRAM_PARTITION_TYPE_EXPANSION_ROM,	   0,    0, "sfc_exp_rom" },
3449 	{ NVRAM_PARTITION_TYPE_STATIC_CONFIG,	   0,    0, "sfc_static_cfg" },
3450 	{ NVRAM_PARTITION_TYPE_DYNAMIC_CONFIG,	   0,    0, "sfc_dynamic_cfg" },
3451 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT0, 0,   0, "sfc_exp_rom_cfg" },
3452 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT1, 0,   1, "sfc_exp_rom_cfg" },
3453 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT2, 0,   2, "sfc_exp_rom_cfg" },
3454 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT3, 0,   3, "sfc_exp_rom_cfg" },
3455 	{ NVRAM_PARTITION_TYPE_LICENSE,		   0,    0, "sfc_license" },
3456 	{ NVRAM_PARTITION_TYPE_PHY_MIN,		   0xff, 0, "sfc_phy_fw" },
3457 	{ NVRAM_PARTITION_TYPE_MUM_FIRMWARE,	   0,    0, "sfc_mumfw" },
3458 	{ NVRAM_PARTITION_TYPE_EXPANSION_UEFI,	   0,    0, "sfc_uefi" },
3459 	{ NVRAM_PARTITION_TYPE_DYNCONFIG_DEFAULTS, 0,    0, "sfc_dynamic_cfg_dflt" },
3460 	{ NVRAM_PARTITION_TYPE_ROMCONFIG_DEFAULTS, 0,    0, "sfc_exp_rom_cfg_dflt" },
3461 	{ NVRAM_PARTITION_TYPE_STATUS,		   0,    0, "sfc_status" },
3462 	{ NVRAM_PARTITION_TYPE_BUNDLE,		   0,    0, "sfc_bundle" },
3463 	{ NVRAM_PARTITION_TYPE_BUNDLE_METADATA,	   0,    0, "sfc_bundle_metadata" },
3464 };
3465 #define EF10_NVRAM_PARTITION_COUNT	ARRAY_SIZE(efx_ef10_nvram_types)
3466 
3467 static int efx_ef10_mtd_probe_partition(struct efx_nic *efx,
3468 					struct efx_mcdi_mtd_partition *part,
3469 					unsigned int type,
3470 					unsigned long *found)
3471 {
3472 	MCDI_DECLARE_BUF(inbuf, MC_CMD_NVRAM_METADATA_IN_LEN);
3473 	MCDI_DECLARE_BUF(outbuf, MC_CMD_NVRAM_METADATA_OUT_LENMAX);
3474 	const struct efx_ef10_nvram_type_info *info;
3475 	size_t size, erase_size, outlen;
3476 	int type_idx = 0;
3477 	bool protected;
3478 	int rc;
3479 
3480 	for (type_idx = 0; ; type_idx++) {
3481 		if (type_idx == EF10_NVRAM_PARTITION_COUNT)
3482 			return -ENODEV;
3483 		info = efx_ef10_nvram_types + type_idx;
3484 		if ((type & ~info->type_mask) == info->type)
3485 			break;
3486 	}
3487 	if (info->port != efx_port_num(efx))
3488 		return -ENODEV;
3489 
3490 	rc = efx_mcdi_nvram_info(efx, type, &size, &erase_size, &protected);
3491 	if (rc)
3492 		return rc;
3493 	if (protected &&
3494 	    (type != NVRAM_PARTITION_TYPE_DYNCONFIG_DEFAULTS &&
3495 	     type != NVRAM_PARTITION_TYPE_ROMCONFIG_DEFAULTS))
3496 		/* Hide protected partitions that don't provide defaults. */
3497 		return -ENODEV;
3498 
3499 	if (protected)
3500 		/* Protected partitions are read only. */
3501 		erase_size = 0;
3502 
3503 	/* If we've already exposed a partition of this type, hide this
3504 	 * duplicate.  All operations on MTDs are keyed by the type anyway,
3505 	 * so we can't act on the duplicate.
3506 	 */
3507 	if (__test_and_set_bit(type_idx, found))
3508 		return -EEXIST;
3509 
3510 	part->nvram_type = type;
3511 
3512 	MCDI_SET_DWORD(inbuf, NVRAM_METADATA_IN_TYPE, type);
3513 	rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_METADATA, inbuf, sizeof(inbuf),
3514 			  outbuf, sizeof(outbuf), &outlen);
3515 	if (rc)
3516 		return rc;
3517 	if (outlen < MC_CMD_NVRAM_METADATA_OUT_LENMIN)
3518 		return -EIO;
3519 	if (MCDI_DWORD(outbuf, NVRAM_METADATA_OUT_FLAGS) &
3520 	    (1 << MC_CMD_NVRAM_METADATA_OUT_SUBTYPE_VALID_LBN))
3521 		part->fw_subtype = MCDI_DWORD(outbuf,
3522 					      NVRAM_METADATA_OUT_SUBTYPE);
3523 
3524 	part->common.dev_type_name = "EF10 NVRAM manager";
3525 	part->common.type_name = info->name;
3526 
3527 	part->common.mtd.type = MTD_NORFLASH;
3528 	part->common.mtd.flags = MTD_CAP_NORFLASH;
3529 	part->common.mtd.size = size;
3530 	part->common.mtd.erasesize = erase_size;
3531 	/* sfc_status is read-only */
3532 	if (!erase_size)
3533 		part->common.mtd.flags |= MTD_NO_ERASE;
3534 
3535 	return 0;
3536 }
3537 
3538 static int efx_ef10_mtd_probe(struct efx_nic *efx)
3539 {
3540 	MCDI_DECLARE_BUF(outbuf, MC_CMD_NVRAM_PARTITIONS_OUT_LENMAX);
3541 	DECLARE_BITMAP(found, EF10_NVRAM_PARTITION_COUNT) = { 0 };
3542 	struct efx_mcdi_mtd_partition *parts;
3543 	size_t outlen, n_parts_total, i, n_parts;
3544 	unsigned int type;
3545 	int rc;
3546 
3547 	ASSERT_RTNL();
3548 
3549 	BUILD_BUG_ON(MC_CMD_NVRAM_PARTITIONS_IN_LEN != 0);
3550 	rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_PARTITIONS, NULL, 0,
3551 			  outbuf, sizeof(outbuf), &outlen);
3552 	if (rc)
3553 		return rc;
3554 	if (outlen < MC_CMD_NVRAM_PARTITIONS_OUT_LENMIN)
3555 		return -EIO;
3556 
3557 	n_parts_total = MCDI_DWORD(outbuf, NVRAM_PARTITIONS_OUT_NUM_PARTITIONS);
3558 	if (n_parts_total >
3559 	    MCDI_VAR_ARRAY_LEN(outlen, NVRAM_PARTITIONS_OUT_TYPE_ID))
3560 		return -EIO;
3561 
3562 	parts = kcalloc(n_parts_total, sizeof(*parts), GFP_KERNEL);
3563 	if (!parts)
3564 		return -ENOMEM;
3565 
3566 	n_parts = 0;
3567 	for (i = 0; i < n_parts_total; i++) {
3568 		type = MCDI_ARRAY_DWORD(outbuf, NVRAM_PARTITIONS_OUT_TYPE_ID,
3569 					i);
3570 		rc = efx_ef10_mtd_probe_partition(efx, &parts[n_parts], type,
3571 						  found);
3572 		if (rc == -EEXIST || rc == -ENODEV)
3573 			continue;
3574 		if (rc)
3575 			goto fail;
3576 		n_parts++;
3577 	}
3578 
3579 	if (!n_parts) {
3580 		kfree(parts);
3581 		return 0;
3582 	}
3583 
3584 	rc = efx_mtd_add(efx, &parts[0].common, n_parts, sizeof(*parts));
3585 fail:
3586 	if (rc)
3587 		kfree(parts);
3588 	return rc;
3589 }
3590 
3591 #endif /* CONFIG_SFC_MTD */
3592 
3593 static void efx_ef10_ptp_write_host_time(struct efx_nic *efx, u32 host_time)
3594 {
3595 	_efx_writed(efx, cpu_to_le32(host_time), ER_DZ_MC_DB_LWRD);
3596 }
3597 
3598 static void efx_ef10_ptp_write_host_time_vf(struct efx_nic *efx,
3599 					    u32 host_time) {}
3600 
3601 static int efx_ef10_rx_enable_timestamping(struct efx_channel *channel,
3602 					   bool temp)
3603 {
3604 	MCDI_DECLARE_BUF(inbuf, MC_CMD_PTP_IN_TIME_EVENT_SUBSCRIBE_LEN);
3605 	int rc;
3606 
3607 	if (channel->sync_events_state == SYNC_EVENTS_REQUESTED ||
3608 	    channel->sync_events_state == SYNC_EVENTS_VALID ||
3609 	    (temp && channel->sync_events_state == SYNC_EVENTS_DISABLED))
3610 		return 0;
3611 	channel->sync_events_state = SYNC_EVENTS_REQUESTED;
3612 
3613 	MCDI_SET_DWORD(inbuf, PTP_IN_OP, MC_CMD_PTP_OP_TIME_EVENT_SUBSCRIBE);
3614 	MCDI_SET_DWORD(inbuf, PTP_IN_PERIPH_ID, 0);
3615 	MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_SUBSCRIBE_QUEUE,
3616 		       channel->channel);
3617 
3618 	rc = efx_mcdi_rpc(channel->efx, MC_CMD_PTP,
3619 			  inbuf, sizeof(inbuf), NULL, 0, NULL);
3620 
3621 	if (rc != 0)
3622 		channel->sync_events_state = temp ? SYNC_EVENTS_QUIESCENT :
3623 						    SYNC_EVENTS_DISABLED;
3624 
3625 	return rc;
3626 }
3627 
3628 static int efx_ef10_rx_disable_timestamping(struct efx_channel *channel,
3629 					    bool temp)
3630 {
3631 	MCDI_DECLARE_BUF(inbuf, MC_CMD_PTP_IN_TIME_EVENT_UNSUBSCRIBE_LEN);
3632 	int rc;
3633 
3634 	if (channel->sync_events_state == SYNC_EVENTS_DISABLED ||
3635 	    (temp && channel->sync_events_state == SYNC_EVENTS_QUIESCENT))
3636 		return 0;
3637 	if (channel->sync_events_state == SYNC_EVENTS_QUIESCENT) {
3638 		channel->sync_events_state = SYNC_EVENTS_DISABLED;
3639 		return 0;
3640 	}
3641 	channel->sync_events_state = temp ? SYNC_EVENTS_QUIESCENT :
3642 					    SYNC_EVENTS_DISABLED;
3643 
3644 	MCDI_SET_DWORD(inbuf, PTP_IN_OP, MC_CMD_PTP_OP_TIME_EVENT_UNSUBSCRIBE);
3645 	MCDI_SET_DWORD(inbuf, PTP_IN_PERIPH_ID, 0);
3646 	MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_UNSUBSCRIBE_CONTROL,
3647 		       MC_CMD_PTP_IN_TIME_EVENT_UNSUBSCRIBE_SINGLE);
3648 	MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_UNSUBSCRIBE_QUEUE,
3649 		       channel->channel);
3650 
3651 	rc = efx_mcdi_rpc(channel->efx, MC_CMD_PTP,
3652 			  inbuf, sizeof(inbuf), NULL, 0, NULL);
3653 
3654 	return rc;
3655 }
3656 
3657 static int efx_ef10_ptp_set_ts_sync_events(struct efx_nic *efx, bool en,
3658 					   bool temp)
3659 {
3660 	int (*set)(struct efx_channel *channel, bool temp);
3661 	struct efx_channel *channel;
3662 
3663 	set = en ?
3664 	      efx_ef10_rx_enable_timestamping :
3665 	      efx_ef10_rx_disable_timestamping;
3666 
3667 	channel = efx_ptp_channel(efx);
3668 	if (channel) {
3669 		int rc = set(channel, temp);
3670 		if (en && rc != 0) {
3671 			efx_ef10_ptp_set_ts_sync_events(efx, false, temp);
3672 			return rc;
3673 		}
3674 	}
3675 
3676 	return 0;
3677 }
3678 
3679 static int efx_ef10_ptp_set_ts_config_vf(struct efx_nic *efx,
3680 					 struct hwtstamp_config *init)
3681 {
3682 	return -EOPNOTSUPP;
3683 }
3684 
3685 static int efx_ef10_ptp_set_ts_config(struct efx_nic *efx,
3686 				      struct hwtstamp_config *init)
3687 {
3688 	int rc;
3689 
3690 	switch (init->rx_filter) {
3691 	case HWTSTAMP_FILTER_NONE:
3692 		efx_ef10_ptp_set_ts_sync_events(efx, false, false);
3693 		/* if TX timestamping is still requested then leave PTP on */
3694 		return efx_ptp_change_mode(efx,
3695 					   init->tx_type != HWTSTAMP_TX_OFF, 0);
3696 	case HWTSTAMP_FILTER_ALL:
3697 	case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
3698 	case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
3699 	case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
3700 	case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
3701 	case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
3702 	case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
3703 	case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
3704 	case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
3705 	case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
3706 	case HWTSTAMP_FILTER_PTP_V2_EVENT:
3707 	case HWTSTAMP_FILTER_PTP_V2_SYNC:
3708 	case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
3709 	case HWTSTAMP_FILTER_NTP_ALL:
3710 		init->rx_filter = HWTSTAMP_FILTER_ALL;
3711 		rc = efx_ptp_change_mode(efx, true, 0);
3712 		if (!rc)
3713 			rc = efx_ef10_ptp_set_ts_sync_events(efx, true, false);
3714 		if (rc)
3715 			efx_ptp_change_mode(efx, false, 0);
3716 		return rc;
3717 	default:
3718 		return -ERANGE;
3719 	}
3720 }
3721 
3722 static int efx_ef10_get_phys_port_id(struct efx_nic *efx,
3723 				     struct netdev_phys_item_id *ppid)
3724 {
3725 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
3726 
3727 	if (!is_valid_ether_addr(nic_data->port_id))
3728 		return -EOPNOTSUPP;
3729 
3730 	ppid->id_len = ETH_ALEN;
3731 	memcpy(ppid->id, nic_data->port_id, ppid->id_len);
3732 
3733 	return 0;
3734 }
3735 
3736 static int efx_ef10_vlan_rx_add_vid(struct efx_nic *efx, __be16 proto, u16 vid)
3737 {
3738 	if (proto != htons(ETH_P_8021Q))
3739 		return -EINVAL;
3740 
3741 	return efx_ef10_add_vlan(efx, vid);
3742 }
3743 
3744 static int efx_ef10_vlan_rx_kill_vid(struct efx_nic *efx, __be16 proto, u16 vid)
3745 {
3746 	if (proto != htons(ETH_P_8021Q))
3747 		return -EINVAL;
3748 
3749 	return efx_ef10_del_vlan(efx, vid);
3750 }
3751 
3752 /* We rely on the MCDI wiping out our TX rings if it made any changes to the
3753  * ports table, ensuring that any TSO descriptors that were made on a now-
3754  * removed tunnel port will be blown away and won't break things when we try
3755  * to transmit them using the new ports table.
3756  */
3757 static int efx_ef10_set_udp_tnl_ports(struct efx_nic *efx, bool unloading)
3758 {
3759 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
3760 	MCDI_DECLARE_BUF(inbuf, MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_IN_LENMAX);
3761 	MCDI_DECLARE_BUF(outbuf, MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_OUT_LEN);
3762 	bool will_reset = false;
3763 	size_t num_entries = 0;
3764 	size_t inlen, outlen;
3765 	size_t i;
3766 	int rc;
3767 	efx_dword_t flags_and_num_entries;
3768 
3769 	WARN_ON(!mutex_is_locked(&nic_data->udp_tunnels_lock));
3770 
3771 	nic_data->udp_tunnels_dirty = false;
3772 
3773 	if (!(nic_data->datapath_caps &
3774 	    (1 << MC_CMD_GET_CAPABILITIES_OUT_VXLAN_NVGRE_LBN))) {
3775 		efx_device_attach_if_not_resetting(efx);
3776 		return 0;
3777 	}
3778 
3779 	BUILD_BUG_ON(ARRAY_SIZE(nic_data->udp_tunnels) >
3780 		     MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_IN_ENTRIES_MAXNUM);
3781 
3782 	for (i = 0; i < ARRAY_SIZE(nic_data->udp_tunnels); ++i) {
3783 		if (nic_data->udp_tunnels[i].type !=
3784 		    TUNNEL_ENCAP_UDP_PORT_ENTRY_INVALID) {
3785 			efx_dword_t entry;
3786 
3787 			EFX_POPULATE_DWORD_2(entry,
3788 				TUNNEL_ENCAP_UDP_PORT_ENTRY_UDP_PORT,
3789 					ntohs(nic_data->udp_tunnels[i].port),
3790 				TUNNEL_ENCAP_UDP_PORT_ENTRY_PROTOCOL,
3791 					nic_data->udp_tunnels[i].type);
3792 			*_MCDI_ARRAY_DWORD(inbuf,
3793 				SET_TUNNEL_ENCAP_UDP_PORTS_IN_ENTRIES,
3794 				num_entries++) = entry;
3795 		}
3796 	}
3797 
3798 	BUILD_BUG_ON((MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_IN_NUM_ENTRIES_OFST -
3799 		      MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_IN_FLAGS_OFST) * 8 !=
3800 		     EFX_WORD_1_LBN);
3801 	BUILD_BUG_ON(MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_IN_NUM_ENTRIES_LEN * 8 !=
3802 		     EFX_WORD_1_WIDTH);
3803 	EFX_POPULATE_DWORD_2(flags_and_num_entries,
3804 			     MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_IN_UNLOADING,
3805 				!!unloading,
3806 			     EFX_WORD_1, num_entries);
3807 	*_MCDI_DWORD(inbuf, SET_TUNNEL_ENCAP_UDP_PORTS_IN_FLAGS) =
3808 		flags_and_num_entries;
3809 
3810 	inlen = MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_IN_LEN(num_entries);
3811 
3812 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS,
3813 				inbuf, inlen, outbuf, sizeof(outbuf), &outlen);
3814 	if (rc == -EIO) {
3815 		/* Most likely the MC rebooted due to another function also
3816 		 * setting its tunnel port list. Mark the tunnel port list as
3817 		 * dirty, so it will be pushed upon coming up from the reboot.
3818 		 */
3819 		nic_data->udp_tunnels_dirty = true;
3820 		return 0;
3821 	}
3822 
3823 	if (rc) {
3824 		/* expected not available on unprivileged functions */
3825 		if (rc != -EPERM)
3826 			netif_warn(efx, drv, efx->net_dev,
3827 				   "Unable to set UDP tunnel ports; rc=%d.\n", rc);
3828 	} else if (MCDI_DWORD(outbuf, SET_TUNNEL_ENCAP_UDP_PORTS_OUT_FLAGS) &
3829 		   (1 << MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_OUT_RESETTING_LBN)) {
3830 		netif_info(efx, drv, efx->net_dev,
3831 			   "Rebooting MC due to UDP tunnel port list change\n");
3832 		will_reset = true;
3833 		if (unloading)
3834 			/* Delay for the MC reset to complete. This will make
3835 			 * unloading other functions a bit smoother. This is a
3836 			 * race, but the other unload will work whichever way
3837 			 * it goes, this just avoids an unnecessary error
3838 			 * message.
3839 			 */
3840 			msleep(100);
3841 	}
3842 	if (!will_reset && !unloading) {
3843 		/* The caller will have detached, relying on the MC reset to
3844 		 * trigger a re-attach.  Since there won't be an MC reset, we
3845 		 * have to do the attach ourselves.
3846 		 */
3847 		efx_device_attach_if_not_resetting(efx);
3848 	}
3849 
3850 	return rc;
3851 }
3852 
3853 static int efx_ef10_udp_tnl_push_ports(struct efx_nic *efx)
3854 {
3855 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
3856 	int rc = 0;
3857 
3858 	mutex_lock(&nic_data->udp_tunnels_lock);
3859 	if (nic_data->udp_tunnels_dirty) {
3860 		/* Make sure all TX are stopped while we modify the table, else
3861 		 * we might race against an efx_features_check().
3862 		 */
3863 		efx_device_detach_sync(efx);
3864 		rc = efx_ef10_set_udp_tnl_ports(efx, false);
3865 	}
3866 	mutex_unlock(&nic_data->udp_tunnels_lock);
3867 	return rc;
3868 }
3869 
3870 static int efx_ef10_udp_tnl_set_port(struct net_device *dev,
3871 				     unsigned int table, unsigned int entry,
3872 				     struct udp_tunnel_info *ti)
3873 {
3874 	struct efx_nic *efx = efx_netdev_priv(dev);
3875 	struct efx_ef10_nic_data *nic_data;
3876 	int efx_tunnel_type, rc;
3877 
3878 	if (ti->type == UDP_TUNNEL_TYPE_VXLAN)
3879 		efx_tunnel_type = TUNNEL_ENCAP_UDP_PORT_ENTRY_VXLAN;
3880 	else
3881 		efx_tunnel_type = TUNNEL_ENCAP_UDP_PORT_ENTRY_GENEVE;
3882 
3883 	nic_data = efx->nic_data;
3884 	if (!(nic_data->datapath_caps &
3885 	      (1 << MC_CMD_GET_CAPABILITIES_OUT_VXLAN_NVGRE_LBN)))
3886 		return -EOPNOTSUPP;
3887 
3888 	mutex_lock(&nic_data->udp_tunnels_lock);
3889 	/* Make sure all TX are stopped while we add to the table, else we
3890 	 * might race against an efx_features_check().
3891 	 */
3892 	efx_device_detach_sync(efx);
3893 	nic_data->udp_tunnels[entry].type = efx_tunnel_type;
3894 	nic_data->udp_tunnels[entry].port = ti->port;
3895 	rc = efx_ef10_set_udp_tnl_ports(efx, false);
3896 	mutex_unlock(&nic_data->udp_tunnels_lock);
3897 
3898 	return rc;
3899 }
3900 
3901 /* Called under the TX lock with the TX queue running, hence no-one can be
3902  * in the middle of updating the UDP tunnels table.  However, they could
3903  * have tried and failed the MCDI, in which case they'll have set the dirty
3904  * flag before dropping their locks.
3905  */
3906 static bool efx_ef10_udp_tnl_has_port(struct efx_nic *efx, __be16 port)
3907 {
3908 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
3909 	size_t i;
3910 
3911 	if (!(nic_data->datapath_caps &
3912 	      (1 << MC_CMD_GET_CAPABILITIES_OUT_VXLAN_NVGRE_LBN)))
3913 		return false;
3914 
3915 	if (nic_data->udp_tunnels_dirty)
3916 		/* SW table may not match HW state, so just assume we can't
3917 		 * use any UDP tunnel offloads.
3918 		 */
3919 		return false;
3920 
3921 	for (i = 0; i < ARRAY_SIZE(nic_data->udp_tunnels); ++i)
3922 		if (nic_data->udp_tunnels[i].type !=
3923 		    TUNNEL_ENCAP_UDP_PORT_ENTRY_INVALID &&
3924 		    nic_data->udp_tunnels[i].port == port)
3925 			return true;
3926 
3927 	return false;
3928 }
3929 
3930 static int efx_ef10_udp_tnl_unset_port(struct net_device *dev,
3931 				       unsigned int table, unsigned int entry,
3932 				       struct udp_tunnel_info *ti)
3933 {
3934 	struct efx_nic *efx = efx_netdev_priv(dev);
3935 	struct efx_ef10_nic_data *nic_data;
3936 	int rc;
3937 
3938 	nic_data = efx->nic_data;
3939 
3940 	mutex_lock(&nic_data->udp_tunnels_lock);
3941 	/* Make sure all TX are stopped while we remove from the table, else we
3942 	 * might race against an efx_features_check().
3943 	 */
3944 	efx_device_detach_sync(efx);
3945 	nic_data->udp_tunnels[entry].type = TUNNEL_ENCAP_UDP_PORT_ENTRY_INVALID;
3946 	nic_data->udp_tunnels[entry].port = 0;
3947 	rc = efx_ef10_set_udp_tnl_ports(efx, false);
3948 	mutex_unlock(&nic_data->udp_tunnels_lock);
3949 
3950 	return rc;
3951 }
3952 
3953 static const struct udp_tunnel_nic_info efx_ef10_udp_tunnels = {
3954 	.set_port	= efx_ef10_udp_tnl_set_port,
3955 	.unset_port	= efx_ef10_udp_tnl_unset_port,
3956 	.flags          = UDP_TUNNEL_NIC_INFO_MAY_SLEEP,
3957 	.tables         = {
3958 		{
3959 			.n_entries = 16,
3960 			.tunnel_types = UDP_TUNNEL_TYPE_VXLAN |
3961 					UDP_TUNNEL_TYPE_GENEVE,
3962 		},
3963 	},
3964 };
3965 
3966 /* EF10 may have multiple datapath firmware variants within a
3967  * single version.  Report which variants are running.
3968  */
3969 static size_t efx_ef10_print_additional_fwver(struct efx_nic *efx, char *buf,
3970 					      size_t len)
3971 {
3972 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
3973 
3974 	return scnprintf(buf, len, " rx%x tx%x",
3975 			 nic_data->rx_dpcpu_fw_id,
3976 			 nic_data->tx_dpcpu_fw_id);
3977 }
3978 
3979 static unsigned int ef10_check_caps(const struct efx_nic *efx,
3980 				    u8 flag,
3981 				    u32 offset)
3982 {
3983 	const struct efx_ef10_nic_data *nic_data = efx->nic_data;
3984 
3985 	switch (offset) {
3986 	case(MC_CMD_GET_CAPABILITIES_V4_OUT_FLAGS1_OFST):
3987 		return nic_data->datapath_caps & BIT_ULL(flag);
3988 	case(MC_CMD_GET_CAPABILITIES_V4_OUT_FLAGS2_OFST):
3989 		return nic_data->datapath_caps2 & BIT_ULL(flag);
3990 	default:
3991 		return 0;
3992 	}
3993 }
3994 
3995 static unsigned int efx_ef10_recycle_ring_size(const struct efx_nic *efx)
3996 {
3997 	unsigned int ret = EFX_RECYCLE_RING_SIZE_10G;
3998 
3999 	/* There is no difference between PFs and VFs. The side is based on
4000 	 * the maximum link speed of a given NIC.
4001 	 */
4002 	switch (efx->pci_dev->device & 0xfff) {
4003 	case 0x0903:	/* Farmingdale can do up to 10G */
4004 		break;
4005 	case 0x0923:	/* Greenport can do up to 40G */
4006 	case 0x0a03:	/* Medford can do up to 40G */
4007 		ret *= 4;
4008 		break;
4009 	default:	/* Medford2 can do up to 100G */
4010 		ret *= 10;
4011 	}
4012 
4013 	if (IS_ENABLED(CONFIG_PPC64))
4014 		ret *= 4;
4015 
4016 	return ret;
4017 }
4018 
4019 #define EF10_OFFLOAD_FEATURES		\
4020 	(NETIF_F_IP_CSUM |		\
4021 	 NETIF_F_HW_VLAN_CTAG_FILTER |	\
4022 	 NETIF_F_IPV6_CSUM |		\
4023 	 NETIF_F_RXHASH |		\
4024 	 NETIF_F_NTUPLE)
4025 
4026 const struct efx_nic_type efx_hunt_a0_vf_nic_type = {
4027 	.is_vf = true,
4028 	.mem_bar = efx_ef10_vf_mem_bar,
4029 	.mem_map_size = efx_ef10_mem_map_size,
4030 	.probe = efx_ef10_probe_vf,
4031 	.remove = efx_ef10_remove,
4032 	.dimension_resources = efx_ef10_dimension_resources,
4033 	.init = efx_ef10_init_nic,
4034 	.fini = efx_ef10_fini_nic,
4035 	.map_reset_reason = efx_ef10_map_reset_reason,
4036 	.map_reset_flags = efx_ef10_map_reset_flags,
4037 	.reset = efx_ef10_reset,
4038 	.probe_port = efx_mcdi_port_probe,
4039 	.remove_port = efx_mcdi_port_remove,
4040 	.fini_dmaq = efx_fini_dmaq,
4041 	.prepare_flr = efx_ef10_prepare_flr,
4042 	.finish_flr = efx_port_dummy_op_void,
4043 	.describe_stats = efx_ef10_describe_stats,
4044 	.update_stats = efx_ef10_update_stats_vf,
4045 	.update_stats_atomic = efx_ef10_update_stats_atomic_vf,
4046 	.start_stats = efx_port_dummy_op_void,
4047 	.pull_stats = efx_port_dummy_op_void,
4048 	.stop_stats = efx_port_dummy_op_void,
4049 	.push_irq_moderation = efx_ef10_push_irq_moderation,
4050 	.reconfigure_mac = efx_ef10_mac_reconfigure,
4051 	.check_mac_fault = efx_mcdi_mac_check_fault,
4052 	.reconfigure_port = efx_mcdi_port_reconfigure,
4053 	.get_wol = efx_ef10_get_wol_vf,
4054 	.set_wol = efx_ef10_set_wol_vf,
4055 	.resume_wol = efx_port_dummy_op_void,
4056 	.mcdi_request = efx_ef10_mcdi_request,
4057 	.mcdi_poll_response = efx_ef10_mcdi_poll_response,
4058 	.mcdi_read_response = efx_ef10_mcdi_read_response,
4059 	.mcdi_poll_reboot = efx_ef10_mcdi_poll_reboot,
4060 	.mcdi_reboot_detected = efx_ef10_mcdi_reboot_detected,
4061 	.irq_enable_master = efx_port_dummy_op_void,
4062 	.irq_test_generate = efx_ef10_irq_test_generate,
4063 	.irq_disable_non_ev = efx_port_dummy_op_void,
4064 	.irq_handle_msi = efx_ef10_msi_interrupt,
4065 	.irq_handle_legacy = efx_ef10_legacy_interrupt,
4066 	.tx_probe = efx_ef10_tx_probe,
4067 	.tx_init = efx_ef10_tx_init,
4068 	.tx_remove = efx_mcdi_tx_remove,
4069 	.tx_write = efx_ef10_tx_write,
4070 	.tx_limit_len = efx_ef10_tx_limit_len,
4071 	.tx_enqueue = __efx_enqueue_skb,
4072 	.rx_push_rss_config = efx_mcdi_vf_rx_push_rss_config,
4073 	.rx_pull_rss_config = efx_mcdi_rx_pull_rss_config,
4074 	.rx_probe = efx_mcdi_rx_probe,
4075 	.rx_init = efx_mcdi_rx_init,
4076 	.rx_remove = efx_mcdi_rx_remove,
4077 	.rx_write = efx_ef10_rx_write,
4078 	.rx_defer_refill = efx_ef10_rx_defer_refill,
4079 	.rx_packet = __efx_rx_packet,
4080 	.ev_probe = efx_mcdi_ev_probe,
4081 	.ev_init = efx_ef10_ev_init,
4082 	.ev_fini = efx_mcdi_ev_fini,
4083 	.ev_remove = efx_mcdi_ev_remove,
4084 	.ev_process = efx_ef10_ev_process,
4085 	.ev_read_ack = efx_ef10_ev_read_ack,
4086 	.ev_test_generate = efx_ef10_ev_test_generate,
4087 	.filter_table_probe = efx_ef10_filter_table_probe,
4088 	.filter_table_restore = efx_mcdi_filter_table_restore,
4089 	.filter_table_remove = efx_ef10_filter_table_remove,
4090 	.filter_update_rx_scatter = efx_mcdi_update_rx_scatter,
4091 	.filter_insert = efx_mcdi_filter_insert,
4092 	.filter_remove_safe = efx_mcdi_filter_remove_safe,
4093 	.filter_get_safe = efx_mcdi_filter_get_safe,
4094 	.filter_clear_rx = efx_mcdi_filter_clear_rx,
4095 	.filter_count_rx_used = efx_mcdi_filter_count_rx_used,
4096 	.filter_get_rx_id_limit = efx_mcdi_filter_get_rx_id_limit,
4097 	.filter_get_rx_ids = efx_mcdi_filter_get_rx_ids,
4098 #ifdef CONFIG_RFS_ACCEL
4099 	.filter_rfs_expire_one = efx_mcdi_filter_rfs_expire_one,
4100 #endif
4101 #ifdef CONFIG_SFC_MTD
4102 	.mtd_probe = efx_port_dummy_op_int,
4103 #endif
4104 	.ptp_write_host_time = efx_ef10_ptp_write_host_time_vf,
4105 	.ptp_set_ts_config = efx_ef10_ptp_set_ts_config_vf,
4106 	.vlan_rx_add_vid = efx_ef10_vlan_rx_add_vid,
4107 	.vlan_rx_kill_vid = efx_ef10_vlan_rx_kill_vid,
4108 #ifdef CONFIG_SFC_SRIOV
4109 	.vswitching_probe = efx_ef10_vswitching_probe_vf,
4110 	.vswitching_restore = efx_ef10_vswitching_restore_vf,
4111 	.vswitching_remove = efx_ef10_vswitching_remove_vf,
4112 #endif
4113 	.get_mac_address = efx_ef10_get_mac_address_vf,
4114 	.set_mac_address = efx_ef10_set_mac_address,
4115 
4116 	.get_phys_port_id = efx_ef10_get_phys_port_id,
4117 	.revision = EFX_REV_HUNT_A0,
4118 	.max_dma_mask = DMA_BIT_MASK(ESF_DZ_TX_KER_BUF_ADDR_WIDTH),
4119 	.rx_prefix_size = ES_DZ_RX_PREFIX_SIZE,
4120 	.rx_hash_offset = ES_DZ_RX_PREFIX_HASH_OFST,
4121 	.rx_ts_offset = ES_DZ_RX_PREFIX_TSTAMP_OFST,
4122 	.can_rx_scatter = true,
4123 	.always_rx_scatter = true,
4124 	.min_interrupt_mode = EFX_INT_MODE_MSIX,
4125 	.timer_period_max = 1 << ERF_DD_EVQ_IND_TIMER_VAL_WIDTH,
4126 	.offload_features = EF10_OFFLOAD_FEATURES,
4127 	.mcdi_max_ver = 2,
4128 	.max_rx_ip_filters = EFX_MCDI_FILTER_TBL_ROWS,
4129 	.hwtstamp_filters = 1 << HWTSTAMP_FILTER_NONE |
4130 			    1 << HWTSTAMP_FILTER_ALL,
4131 	.rx_hash_key_size = 40,
4132 	.check_caps = ef10_check_caps,
4133 	.print_additional_fwver = efx_ef10_print_additional_fwver,
4134 	.sensor_event = efx_mcdi_sensor_event,
4135 	.rx_recycle_ring_size = efx_ef10_recycle_ring_size,
4136 };
4137 
4138 const struct efx_nic_type efx_hunt_a0_nic_type = {
4139 	.is_vf = false,
4140 	.mem_bar = efx_ef10_pf_mem_bar,
4141 	.mem_map_size = efx_ef10_mem_map_size,
4142 	.probe = efx_ef10_probe_pf,
4143 	.remove = efx_ef10_remove,
4144 	.dimension_resources = efx_ef10_dimension_resources,
4145 	.init = efx_ef10_init_nic,
4146 	.fini = efx_ef10_fini_nic,
4147 	.map_reset_reason = efx_ef10_map_reset_reason,
4148 	.map_reset_flags = efx_ef10_map_reset_flags,
4149 	.reset = efx_ef10_reset,
4150 	.probe_port = efx_mcdi_port_probe,
4151 	.remove_port = efx_mcdi_port_remove,
4152 	.fini_dmaq = efx_fini_dmaq,
4153 	.prepare_flr = efx_ef10_prepare_flr,
4154 	.finish_flr = efx_port_dummy_op_void,
4155 	.describe_stats = efx_ef10_describe_stats,
4156 	.update_stats = efx_ef10_update_stats_pf,
4157 	.start_stats = efx_mcdi_mac_start_stats,
4158 	.pull_stats = efx_mcdi_mac_pull_stats,
4159 	.stop_stats = efx_mcdi_mac_stop_stats,
4160 	.push_irq_moderation = efx_ef10_push_irq_moderation,
4161 	.reconfigure_mac = efx_ef10_mac_reconfigure,
4162 	.check_mac_fault = efx_mcdi_mac_check_fault,
4163 	.reconfigure_port = efx_mcdi_port_reconfigure,
4164 	.get_wol = efx_ef10_get_wol,
4165 	.set_wol = efx_ef10_set_wol,
4166 	.resume_wol = efx_port_dummy_op_void,
4167 	.get_fec_stats = efx_ef10_get_fec_stats,
4168 	.test_chip = efx_ef10_test_chip,
4169 	.test_nvram = efx_mcdi_nvram_test_all,
4170 	.mcdi_request = efx_ef10_mcdi_request,
4171 	.mcdi_poll_response = efx_ef10_mcdi_poll_response,
4172 	.mcdi_read_response = efx_ef10_mcdi_read_response,
4173 	.mcdi_poll_reboot = efx_ef10_mcdi_poll_reboot,
4174 	.mcdi_reboot_detected = efx_ef10_mcdi_reboot_detected,
4175 	.irq_enable_master = efx_port_dummy_op_void,
4176 	.irq_test_generate = efx_ef10_irq_test_generate,
4177 	.irq_disable_non_ev = efx_port_dummy_op_void,
4178 	.irq_handle_msi = efx_ef10_msi_interrupt,
4179 	.irq_handle_legacy = efx_ef10_legacy_interrupt,
4180 	.tx_probe = efx_ef10_tx_probe,
4181 	.tx_init = efx_ef10_tx_init,
4182 	.tx_remove = efx_mcdi_tx_remove,
4183 	.tx_write = efx_ef10_tx_write,
4184 	.tx_limit_len = efx_ef10_tx_limit_len,
4185 	.tx_enqueue = __efx_enqueue_skb,
4186 	.rx_push_rss_config = efx_mcdi_pf_rx_push_rss_config,
4187 	.rx_pull_rss_config = efx_mcdi_rx_pull_rss_config,
4188 	.rx_push_rss_context_config = efx_mcdi_rx_push_rss_context_config,
4189 	.rx_pull_rss_context_config = efx_mcdi_rx_pull_rss_context_config,
4190 	.rx_restore_rss_contexts = efx_mcdi_rx_restore_rss_contexts,
4191 	.rx_probe = efx_mcdi_rx_probe,
4192 	.rx_init = efx_mcdi_rx_init,
4193 	.rx_remove = efx_mcdi_rx_remove,
4194 	.rx_write = efx_ef10_rx_write,
4195 	.rx_defer_refill = efx_ef10_rx_defer_refill,
4196 	.rx_packet = __efx_rx_packet,
4197 	.ev_probe = efx_mcdi_ev_probe,
4198 	.ev_init = efx_ef10_ev_init,
4199 	.ev_fini = efx_mcdi_ev_fini,
4200 	.ev_remove = efx_mcdi_ev_remove,
4201 	.ev_process = efx_ef10_ev_process,
4202 	.ev_read_ack = efx_ef10_ev_read_ack,
4203 	.ev_test_generate = efx_ef10_ev_test_generate,
4204 	.filter_table_probe = efx_ef10_filter_table_probe,
4205 	.filter_table_restore = efx_mcdi_filter_table_restore,
4206 	.filter_table_remove = efx_ef10_filter_table_remove,
4207 	.filter_update_rx_scatter = efx_mcdi_update_rx_scatter,
4208 	.filter_insert = efx_mcdi_filter_insert,
4209 	.filter_remove_safe = efx_mcdi_filter_remove_safe,
4210 	.filter_get_safe = efx_mcdi_filter_get_safe,
4211 	.filter_clear_rx = efx_mcdi_filter_clear_rx,
4212 	.filter_count_rx_used = efx_mcdi_filter_count_rx_used,
4213 	.filter_get_rx_id_limit = efx_mcdi_filter_get_rx_id_limit,
4214 	.filter_get_rx_ids = efx_mcdi_filter_get_rx_ids,
4215 #ifdef CONFIG_RFS_ACCEL
4216 	.filter_rfs_expire_one = efx_mcdi_filter_rfs_expire_one,
4217 #endif
4218 #ifdef CONFIG_SFC_MTD
4219 	.mtd_probe = efx_ef10_mtd_probe,
4220 	.mtd_rename = efx_mcdi_mtd_rename,
4221 	.mtd_read = efx_mcdi_mtd_read,
4222 	.mtd_erase = efx_mcdi_mtd_erase,
4223 	.mtd_write = efx_mcdi_mtd_write,
4224 	.mtd_sync = efx_mcdi_mtd_sync,
4225 #endif
4226 	.ptp_write_host_time = efx_ef10_ptp_write_host_time,
4227 	.ptp_set_ts_sync_events = efx_ef10_ptp_set_ts_sync_events,
4228 	.ptp_set_ts_config = efx_ef10_ptp_set_ts_config,
4229 	.vlan_rx_add_vid = efx_ef10_vlan_rx_add_vid,
4230 	.vlan_rx_kill_vid = efx_ef10_vlan_rx_kill_vid,
4231 	.udp_tnl_push_ports = efx_ef10_udp_tnl_push_ports,
4232 	.udp_tnl_has_port = efx_ef10_udp_tnl_has_port,
4233 #ifdef CONFIG_SFC_SRIOV
4234 	.sriov_configure = efx_ef10_sriov_configure,
4235 	.sriov_init = efx_ef10_sriov_init,
4236 	.sriov_fini = efx_ef10_sriov_fini,
4237 	.sriov_wanted = efx_ef10_sriov_wanted,
4238 	.sriov_reset = efx_ef10_sriov_reset,
4239 	.sriov_flr = efx_ef10_sriov_flr,
4240 	.sriov_set_vf_mac = efx_ef10_sriov_set_vf_mac,
4241 	.sriov_set_vf_vlan = efx_ef10_sriov_set_vf_vlan,
4242 	.sriov_set_vf_spoofchk = efx_ef10_sriov_set_vf_spoofchk,
4243 	.sriov_get_vf_config = efx_ef10_sriov_get_vf_config,
4244 	.sriov_set_vf_link_state = efx_ef10_sriov_set_vf_link_state,
4245 	.vswitching_probe = efx_ef10_vswitching_probe_pf,
4246 	.vswitching_restore = efx_ef10_vswitching_restore_pf,
4247 	.vswitching_remove = efx_ef10_vswitching_remove_pf,
4248 #endif
4249 	.get_mac_address = efx_ef10_get_mac_address_pf,
4250 	.set_mac_address = efx_ef10_set_mac_address,
4251 	.tso_versions = efx_ef10_tso_versions,
4252 
4253 	.get_phys_port_id = efx_ef10_get_phys_port_id,
4254 	.revision = EFX_REV_HUNT_A0,
4255 	.max_dma_mask = DMA_BIT_MASK(ESF_DZ_TX_KER_BUF_ADDR_WIDTH),
4256 	.rx_prefix_size = ES_DZ_RX_PREFIX_SIZE,
4257 	.rx_hash_offset = ES_DZ_RX_PREFIX_HASH_OFST,
4258 	.rx_ts_offset = ES_DZ_RX_PREFIX_TSTAMP_OFST,
4259 	.can_rx_scatter = true,
4260 	.always_rx_scatter = true,
4261 	.option_descriptors = true,
4262 	.min_interrupt_mode = EFX_INT_MODE_LEGACY,
4263 	.timer_period_max = 1 << ERF_DD_EVQ_IND_TIMER_VAL_WIDTH,
4264 	.offload_features = EF10_OFFLOAD_FEATURES,
4265 	.mcdi_max_ver = 2,
4266 	.max_rx_ip_filters = EFX_MCDI_FILTER_TBL_ROWS,
4267 	.hwtstamp_filters = 1 << HWTSTAMP_FILTER_NONE |
4268 			    1 << HWTSTAMP_FILTER_ALL,
4269 	.rx_hash_key_size = 40,
4270 	.check_caps = ef10_check_caps,
4271 	.print_additional_fwver = efx_ef10_print_additional_fwver,
4272 	.sensor_event = efx_mcdi_sensor_event,
4273 	.rx_recycle_ring_size = efx_ef10_recycle_ring_size,
4274 };
4275