xref: /openbmc/linux/drivers/net/ethernet/sfc/ef10.c (revision d2f4a190)
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 	struct net_device *net_dev = efx->net_dev;
1308 	netdev_features_t tun_feats, tso_feats;
1309 	int rc;
1310 
1311 	if (nic_data->must_check_datapath_caps) {
1312 		rc = efx_ef10_init_datapath_caps(efx);
1313 		if (rc)
1314 			return rc;
1315 		nic_data->must_check_datapath_caps = false;
1316 	}
1317 
1318 	if (efx->must_realloc_vis) {
1319 		/* We cannot let the number of VIs change now */
1320 		rc = efx_ef10_alloc_vis(efx, nic_data->n_allocated_vis,
1321 					nic_data->n_allocated_vis);
1322 		if (rc)
1323 			return rc;
1324 		efx->must_realloc_vis = false;
1325 	}
1326 
1327 	nic_data->mc_stats = kmalloc(efx->num_mac_stats * sizeof(__le64),
1328 				     GFP_KERNEL);
1329 	if (!nic_data->mc_stats)
1330 		return -ENOMEM;
1331 
1332 	if (nic_data->must_restore_piobufs && nic_data->n_piobufs) {
1333 		rc = efx_ef10_alloc_piobufs(efx, nic_data->n_piobufs);
1334 		if (rc == 0) {
1335 			rc = efx_ef10_link_piobufs(efx);
1336 			if (rc)
1337 				efx_ef10_free_piobufs(efx);
1338 		}
1339 
1340 		/* Log an error on failure, but this is non-fatal.
1341 		 * Permission errors are less important - we've presumably
1342 		 * had the PIO buffer licence removed.
1343 		 */
1344 		if (rc == -EPERM)
1345 			netif_dbg(efx, drv, efx->net_dev,
1346 				  "not permitted to restore PIO buffers\n");
1347 		else if (rc)
1348 			netif_err(efx, drv, efx->net_dev,
1349 				  "failed to restore PIO buffers (%d)\n", rc);
1350 		nic_data->must_restore_piobufs = false;
1351 	}
1352 
1353 	/* encap features might change during reset if fw variant changed */
1354 	if (efx_has_cap(efx, VXLAN_NVGRE) && !efx_ef10_is_vf(efx))
1355 		net_dev->hw_enc_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
1356 	else
1357 		net_dev->hw_enc_features &= ~(NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
1358 
1359 	tun_feats = NETIF_F_GSO_UDP_TUNNEL | NETIF_F_GSO_GRE |
1360 		    NETIF_F_GSO_UDP_TUNNEL_CSUM | NETIF_F_GSO_GRE_CSUM;
1361 	tso_feats = NETIF_F_TSO | NETIF_F_TSO6;
1362 
1363 	if (efx_has_cap(efx, TX_TSO_V2_ENCAP)) {
1364 		/* If this is first nic_init, or if it is a reset and a new fw
1365 		 * variant has added new features, enable them by default.
1366 		 * If the features are not new, maintain their current value.
1367 		 */
1368 		if (!(net_dev->hw_features & tun_feats))
1369 			net_dev->features |= tun_feats;
1370 		net_dev->hw_enc_features |= tun_feats | tso_feats;
1371 		net_dev->hw_features |= tun_feats;
1372 	} else {
1373 		net_dev->hw_enc_features &= ~(tun_feats | tso_feats);
1374 		net_dev->hw_features &= ~tun_feats;
1375 		net_dev->features &= ~tun_feats;
1376 	}
1377 
1378 	/* don't fail init if RSS setup doesn't work */
1379 	rc = efx->type->rx_push_rss_config(efx, false,
1380 					   efx->rss_context.rx_indir_table, NULL);
1381 
1382 	return 0;
1383 }
1384 
1385 static void efx_ef10_table_reset_mc_allocations(struct efx_nic *efx)
1386 {
1387 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1388 #ifdef CONFIG_SFC_SRIOV
1389 	unsigned int i;
1390 #endif
1391 
1392 	/* All our allocations have been reset */
1393 	efx->must_realloc_vis = true;
1394 	efx_mcdi_filter_table_reset_mc_allocations(efx);
1395 	nic_data->must_restore_piobufs = true;
1396 	efx_ef10_forget_old_piobufs(efx);
1397 	efx->rss_context.context_id = EFX_MCDI_RSS_CONTEXT_INVALID;
1398 
1399 	/* Driver-created vswitches and vports must be re-created */
1400 	nic_data->must_probe_vswitching = true;
1401 	efx->vport_id = EVB_PORT_ID_ASSIGNED;
1402 #ifdef CONFIG_SFC_SRIOV
1403 	if (nic_data->vf)
1404 		for (i = 0; i < efx->vf_count; i++)
1405 			nic_data->vf[i].vport_id = 0;
1406 #endif
1407 }
1408 
1409 static enum reset_type efx_ef10_map_reset_reason(enum reset_type reason)
1410 {
1411 	if (reason == RESET_TYPE_MC_FAILURE)
1412 		return RESET_TYPE_DATAPATH;
1413 
1414 	return efx_mcdi_map_reset_reason(reason);
1415 }
1416 
1417 static int efx_ef10_map_reset_flags(u32 *flags)
1418 {
1419 	enum {
1420 		EF10_RESET_PORT = ((ETH_RESET_MAC | ETH_RESET_PHY) <<
1421 				   ETH_RESET_SHARED_SHIFT),
1422 		EF10_RESET_MC = ((ETH_RESET_DMA | ETH_RESET_FILTER |
1423 				  ETH_RESET_OFFLOAD | ETH_RESET_MAC |
1424 				  ETH_RESET_PHY | ETH_RESET_MGMT) <<
1425 				 ETH_RESET_SHARED_SHIFT)
1426 	};
1427 
1428 	/* We assume for now that our PCI function is permitted to
1429 	 * reset everything.
1430 	 */
1431 
1432 	if ((*flags & EF10_RESET_MC) == EF10_RESET_MC) {
1433 		*flags &= ~EF10_RESET_MC;
1434 		return RESET_TYPE_WORLD;
1435 	}
1436 
1437 	if ((*flags & EF10_RESET_PORT) == EF10_RESET_PORT) {
1438 		*flags &= ~EF10_RESET_PORT;
1439 		return RESET_TYPE_ALL;
1440 	}
1441 
1442 	/* no invisible reset implemented */
1443 
1444 	return -EINVAL;
1445 }
1446 
1447 static int efx_ef10_reset(struct efx_nic *efx, enum reset_type reset_type)
1448 {
1449 	int rc = efx_mcdi_reset(efx, reset_type);
1450 
1451 	/* Unprivileged functions return -EPERM, but need to return success
1452 	 * here so that the datapath is brought back up.
1453 	 */
1454 	if (reset_type == RESET_TYPE_WORLD && rc == -EPERM)
1455 		rc = 0;
1456 
1457 	/* If it was a port reset, trigger reallocation of MC resources.
1458 	 * Note that on an MC reset nothing needs to be done now because we'll
1459 	 * detect the MC reset later and handle it then.
1460 	 * For an FLR, we never get an MC reset event, but the MC has reset all
1461 	 * resources assigned to us, so we have to trigger reallocation now.
1462 	 */
1463 	if ((reset_type == RESET_TYPE_ALL ||
1464 	     reset_type == RESET_TYPE_MCDI_TIMEOUT) && !rc)
1465 		efx_ef10_table_reset_mc_allocations(efx);
1466 	return rc;
1467 }
1468 
1469 #define EF10_DMA_STAT(ext_name, mcdi_name)			\
1470 	[EF10_STAT_ ## ext_name] =				\
1471 	{ #ext_name, 64, 8 * MC_CMD_MAC_ ## mcdi_name }
1472 #define EF10_DMA_INVIS_STAT(int_name, mcdi_name)		\
1473 	[EF10_STAT_ ## int_name] =				\
1474 	{ NULL, 64, 8 * MC_CMD_MAC_ ## mcdi_name }
1475 #define EF10_OTHER_STAT(ext_name)				\
1476 	[EF10_STAT_ ## ext_name] = { #ext_name, 0, 0 }
1477 
1478 static const struct efx_hw_stat_desc efx_ef10_stat_desc[EF10_STAT_COUNT] = {
1479 	EF10_DMA_STAT(port_tx_bytes, TX_BYTES),
1480 	EF10_DMA_STAT(port_tx_packets, TX_PKTS),
1481 	EF10_DMA_STAT(port_tx_pause, TX_PAUSE_PKTS),
1482 	EF10_DMA_STAT(port_tx_control, TX_CONTROL_PKTS),
1483 	EF10_DMA_STAT(port_tx_unicast, TX_UNICAST_PKTS),
1484 	EF10_DMA_STAT(port_tx_multicast, TX_MULTICAST_PKTS),
1485 	EF10_DMA_STAT(port_tx_broadcast, TX_BROADCAST_PKTS),
1486 	EF10_DMA_STAT(port_tx_lt64, TX_LT64_PKTS),
1487 	EF10_DMA_STAT(port_tx_64, TX_64_PKTS),
1488 	EF10_DMA_STAT(port_tx_65_to_127, TX_65_TO_127_PKTS),
1489 	EF10_DMA_STAT(port_tx_128_to_255, TX_128_TO_255_PKTS),
1490 	EF10_DMA_STAT(port_tx_256_to_511, TX_256_TO_511_PKTS),
1491 	EF10_DMA_STAT(port_tx_512_to_1023, TX_512_TO_1023_PKTS),
1492 	EF10_DMA_STAT(port_tx_1024_to_15xx, TX_1024_TO_15XX_PKTS),
1493 	EF10_DMA_STAT(port_tx_15xx_to_jumbo, TX_15XX_TO_JUMBO_PKTS),
1494 	EF10_DMA_STAT(port_rx_bytes, RX_BYTES),
1495 	EF10_DMA_INVIS_STAT(port_rx_bytes_minus_good_bytes, RX_BAD_BYTES),
1496 	EF10_OTHER_STAT(port_rx_good_bytes),
1497 	EF10_OTHER_STAT(port_rx_bad_bytes),
1498 	EF10_DMA_STAT(port_rx_packets, RX_PKTS),
1499 	EF10_DMA_STAT(port_rx_good, RX_GOOD_PKTS),
1500 	EF10_DMA_STAT(port_rx_bad, RX_BAD_FCS_PKTS),
1501 	EF10_DMA_STAT(port_rx_pause, RX_PAUSE_PKTS),
1502 	EF10_DMA_STAT(port_rx_control, RX_CONTROL_PKTS),
1503 	EF10_DMA_STAT(port_rx_unicast, RX_UNICAST_PKTS),
1504 	EF10_DMA_STAT(port_rx_multicast, RX_MULTICAST_PKTS),
1505 	EF10_DMA_STAT(port_rx_broadcast, RX_BROADCAST_PKTS),
1506 	EF10_DMA_STAT(port_rx_lt64, RX_UNDERSIZE_PKTS),
1507 	EF10_DMA_STAT(port_rx_64, RX_64_PKTS),
1508 	EF10_DMA_STAT(port_rx_65_to_127, RX_65_TO_127_PKTS),
1509 	EF10_DMA_STAT(port_rx_128_to_255, RX_128_TO_255_PKTS),
1510 	EF10_DMA_STAT(port_rx_256_to_511, RX_256_TO_511_PKTS),
1511 	EF10_DMA_STAT(port_rx_512_to_1023, RX_512_TO_1023_PKTS),
1512 	EF10_DMA_STAT(port_rx_1024_to_15xx, RX_1024_TO_15XX_PKTS),
1513 	EF10_DMA_STAT(port_rx_15xx_to_jumbo, RX_15XX_TO_JUMBO_PKTS),
1514 	EF10_DMA_STAT(port_rx_gtjumbo, RX_GTJUMBO_PKTS),
1515 	EF10_DMA_STAT(port_rx_bad_gtjumbo, RX_JABBER_PKTS),
1516 	EF10_DMA_STAT(port_rx_overflow, RX_OVERFLOW_PKTS),
1517 	EF10_DMA_STAT(port_rx_align_error, RX_ALIGN_ERROR_PKTS),
1518 	EF10_DMA_STAT(port_rx_length_error, RX_LENGTH_ERROR_PKTS),
1519 	EF10_DMA_STAT(port_rx_nodesc_drops, RX_NODESC_DROPS),
1520 	EFX_GENERIC_SW_STAT(rx_nodesc_trunc),
1521 	EFX_GENERIC_SW_STAT(rx_noskb_drops),
1522 	EF10_DMA_STAT(port_rx_pm_trunc_bb_overflow, PM_TRUNC_BB_OVERFLOW),
1523 	EF10_DMA_STAT(port_rx_pm_discard_bb_overflow, PM_DISCARD_BB_OVERFLOW),
1524 	EF10_DMA_STAT(port_rx_pm_trunc_vfifo_full, PM_TRUNC_VFIFO_FULL),
1525 	EF10_DMA_STAT(port_rx_pm_discard_vfifo_full, PM_DISCARD_VFIFO_FULL),
1526 	EF10_DMA_STAT(port_rx_pm_trunc_qbb, PM_TRUNC_QBB),
1527 	EF10_DMA_STAT(port_rx_pm_discard_qbb, PM_DISCARD_QBB),
1528 	EF10_DMA_STAT(port_rx_pm_discard_mapping, PM_DISCARD_MAPPING),
1529 	EF10_DMA_STAT(port_rx_dp_q_disabled_packets, RXDP_Q_DISABLED_PKTS),
1530 	EF10_DMA_STAT(port_rx_dp_di_dropped_packets, RXDP_DI_DROPPED_PKTS),
1531 	EF10_DMA_STAT(port_rx_dp_streaming_packets, RXDP_STREAMING_PKTS),
1532 	EF10_DMA_STAT(port_rx_dp_hlb_fetch, RXDP_HLB_FETCH_CONDITIONS),
1533 	EF10_DMA_STAT(port_rx_dp_hlb_wait, RXDP_HLB_WAIT_CONDITIONS),
1534 	EF10_DMA_STAT(rx_unicast, VADAPTER_RX_UNICAST_PACKETS),
1535 	EF10_DMA_STAT(rx_unicast_bytes, VADAPTER_RX_UNICAST_BYTES),
1536 	EF10_DMA_STAT(rx_multicast, VADAPTER_RX_MULTICAST_PACKETS),
1537 	EF10_DMA_STAT(rx_multicast_bytes, VADAPTER_RX_MULTICAST_BYTES),
1538 	EF10_DMA_STAT(rx_broadcast, VADAPTER_RX_BROADCAST_PACKETS),
1539 	EF10_DMA_STAT(rx_broadcast_bytes, VADAPTER_RX_BROADCAST_BYTES),
1540 	EF10_DMA_STAT(rx_bad, VADAPTER_RX_BAD_PACKETS),
1541 	EF10_DMA_STAT(rx_bad_bytes, VADAPTER_RX_BAD_BYTES),
1542 	EF10_DMA_STAT(rx_overflow, VADAPTER_RX_OVERFLOW),
1543 	EF10_DMA_STAT(tx_unicast, VADAPTER_TX_UNICAST_PACKETS),
1544 	EF10_DMA_STAT(tx_unicast_bytes, VADAPTER_TX_UNICAST_BYTES),
1545 	EF10_DMA_STAT(tx_multicast, VADAPTER_TX_MULTICAST_PACKETS),
1546 	EF10_DMA_STAT(tx_multicast_bytes, VADAPTER_TX_MULTICAST_BYTES),
1547 	EF10_DMA_STAT(tx_broadcast, VADAPTER_TX_BROADCAST_PACKETS),
1548 	EF10_DMA_STAT(tx_broadcast_bytes, VADAPTER_TX_BROADCAST_BYTES),
1549 	EF10_DMA_STAT(tx_bad, VADAPTER_TX_BAD_PACKETS),
1550 	EF10_DMA_STAT(tx_bad_bytes, VADAPTER_TX_BAD_BYTES),
1551 	EF10_DMA_STAT(tx_overflow, VADAPTER_TX_OVERFLOW),
1552 	EF10_DMA_STAT(fec_uncorrected_errors, FEC_UNCORRECTED_ERRORS),
1553 	EF10_DMA_STAT(fec_corrected_errors, FEC_CORRECTED_ERRORS),
1554 	EF10_DMA_STAT(fec_corrected_symbols_lane0, FEC_CORRECTED_SYMBOLS_LANE0),
1555 	EF10_DMA_STAT(fec_corrected_symbols_lane1, FEC_CORRECTED_SYMBOLS_LANE1),
1556 	EF10_DMA_STAT(fec_corrected_symbols_lane2, FEC_CORRECTED_SYMBOLS_LANE2),
1557 	EF10_DMA_STAT(fec_corrected_symbols_lane3, FEC_CORRECTED_SYMBOLS_LANE3),
1558 	EF10_DMA_STAT(ctpio_vi_busy_fallback, CTPIO_VI_BUSY_FALLBACK),
1559 	EF10_DMA_STAT(ctpio_long_write_success, CTPIO_LONG_WRITE_SUCCESS),
1560 	EF10_DMA_STAT(ctpio_missing_dbell_fail, CTPIO_MISSING_DBELL_FAIL),
1561 	EF10_DMA_STAT(ctpio_overflow_fail, CTPIO_OVERFLOW_FAIL),
1562 	EF10_DMA_STAT(ctpio_underflow_fail, CTPIO_UNDERFLOW_FAIL),
1563 	EF10_DMA_STAT(ctpio_timeout_fail, CTPIO_TIMEOUT_FAIL),
1564 	EF10_DMA_STAT(ctpio_noncontig_wr_fail, CTPIO_NONCONTIG_WR_FAIL),
1565 	EF10_DMA_STAT(ctpio_frm_clobber_fail, CTPIO_FRM_CLOBBER_FAIL),
1566 	EF10_DMA_STAT(ctpio_invalid_wr_fail, CTPIO_INVALID_WR_FAIL),
1567 	EF10_DMA_STAT(ctpio_vi_clobber_fallback, CTPIO_VI_CLOBBER_FALLBACK),
1568 	EF10_DMA_STAT(ctpio_unqualified_fallback, CTPIO_UNQUALIFIED_FALLBACK),
1569 	EF10_DMA_STAT(ctpio_runt_fallback, CTPIO_RUNT_FALLBACK),
1570 	EF10_DMA_STAT(ctpio_success, CTPIO_SUCCESS),
1571 	EF10_DMA_STAT(ctpio_fallback, CTPIO_FALLBACK),
1572 	EF10_DMA_STAT(ctpio_poison, CTPIO_POISON),
1573 	EF10_DMA_STAT(ctpio_erase, CTPIO_ERASE),
1574 };
1575 
1576 #define HUNT_COMMON_STAT_MASK ((1ULL << EF10_STAT_port_tx_bytes) |	\
1577 			       (1ULL << EF10_STAT_port_tx_packets) |	\
1578 			       (1ULL << EF10_STAT_port_tx_pause) |	\
1579 			       (1ULL << EF10_STAT_port_tx_unicast) |	\
1580 			       (1ULL << EF10_STAT_port_tx_multicast) |	\
1581 			       (1ULL << EF10_STAT_port_tx_broadcast) |	\
1582 			       (1ULL << EF10_STAT_port_rx_bytes) |	\
1583 			       (1ULL <<                                 \
1584 				EF10_STAT_port_rx_bytes_minus_good_bytes) | \
1585 			       (1ULL << EF10_STAT_port_rx_good_bytes) |	\
1586 			       (1ULL << EF10_STAT_port_rx_bad_bytes) |	\
1587 			       (1ULL << EF10_STAT_port_rx_packets) |	\
1588 			       (1ULL << EF10_STAT_port_rx_good) |	\
1589 			       (1ULL << EF10_STAT_port_rx_bad) |	\
1590 			       (1ULL << EF10_STAT_port_rx_pause) |	\
1591 			       (1ULL << EF10_STAT_port_rx_control) |	\
1592 			       (1ULL << EF10_STAT_port_rx_unicast) |	\
1593 			       (1ULL << EF10_STAT_port_rx_multicast) |	\
1594 			       (1ULL << EF10_STAT_port_rx_broadcast) |	\
1595 			       (1ULL << EF10_STAT_port_rx_lt64) |	\
1596 			       (1ULL << EF10_STAT_port_rx_64) |		\
1597 			       (1ULL << EF10_STAT_port_rx_65_to_127) |	\
1598 			       (1ULL << EF10_STAT_port_rx_128_to_255) |	\
1599 			       (1ULL << EF10_STAT_port_rx_256_to_511) |	\
1600 			       (1ULL << EF10_STAT_port_rx_512_to_1023) |\
1601 			       (1ULL << EF10_STAT_port_rx_1024_to_15xx) |\
1602 			       (1ULL << EF10_STAT_port_rx_15xx_to_jumbo) |\
1603 			       (1ULL << EF10_STAT_port_rx_gtjumbo) |	\
1604 			       (1ULL << EF10_STAT_port_rx_bad_gtjumbo) |\
1605 			       (1ULL << EF10_STAT_port_rx_overflow) |	\
1606 			       (1ULL << EF10_STAT_port_rx_nodesc_drops) |\
1607 			       (1ULL << GENERIC_STAT_rx_nodesc_trunc) |	\
1608 			       (1ULL << GENERIC_STAT_rx_noskb_drops))
1609 
1610 /* On 7000 series NICs, these statistics are only provided by the 10G MAC.
1611  * For a 10G/40G switchable port we do not expose these because they might
1612  * not include all the packets they should.
1613  * On 8000 series NICs these statistics are always provided.
1614  */
1615 #define HUNT_10G_ONLY_STAT_MASK ((1ULL << EF10_STAT_port_tx_control) |	\
1616 				 (1ULL << EF10_STAT_port_tx_lt64) |	\
1617 				 (1ULL << EF10_STAT_port_tx_64) |	\
1618 				 (1ULL << EF10_STAT_port_tx_65_to_127) |\
1619 				 (1ULL << EF10_STAT_port_tx_128_to_255) |\
1620 				 (1ULL << EF10_STAT_port_tx_256_to_511) |\
1621 				 (1ULL << EF10_STAT_port_tx_512_to_1023) |\
1622 				 (1ULL << EF10_STAT_port_tx_1024_to_15xx) |\
1623 				 (1ULL << EF10_STAT_port_tx_15xx_to_jumbo))
1624 
1625 /* These statistics are only provided by the 40G MAC.  For a 10G/40G
1626  * switchable port we do expose these because the errors will otherwise
1627  * be silent.
1628  */
1629 #define HUNT_40G_EXTRA_STAT_MASK ((1ULL << EF10_STAT_port_rx_align_error) |\
1630 				  (1ULL << EF10_STAT_port_rx_length_error))
1631 
1632 /* These statistics are only provided if the firmware supports the
1633  * capability PM_AND_RXDP_COUNTERS.
1634  */
1635 #define HUNT_PM_AND_RXDP_STAT_MASK (					\
1636 	(1ULL << EF10_STAT_port_rx_pm_trunc_bb_overflow) |		\
1637 	(1ULL << EF10_STAT_port_rx_pm_discard_bb_overflow) |		\
1638 	(1ULL << EF10_STAT_port_rx_pm_trunc_vfifo_full) |		\
1639 	(1ULL << EF10_STAT_port_rx_pm_discard_vfifo_full) |		\
1640 	(1ULL << EF10_STAT_port_rx_pm_trunc_qbb) |			\
1641 	(1ULL << EF10_STAT_port_rx_pm_discard_qbb) |			\
1642 	(1ULL << EF10_STAT_port_rx_pm_discard_mapping) |		\
1643 	(1ULL << EF10_STAT_port_rx_dp_q_disabled_packets) |		\
1644 	(1ULL << EF10_STAT_port_rx_dp_di_dropped_packets) |		\
1645 	(1ULL << EF10_STAT_port_rx_dp_streaming_packets) |		\
1646 	(1ULL << EF10_STAT_port_rx_dp_hlb_fetch) |			\
1647 	(1ULL << EF10_STAT_port_rx_dp_hlb_wait))
1648 
1649 /* These statistics are only provided if the NIC supports MC_CMD_MAC_STATS_V2,
1650  * indicated by returning a value >= MC_CMD_MAC_NSTATS_V2 in
1651  * MC_CMD_GET_CAPABILITIES_V4_OUT_MAC_STATS_NUM_STATS.
1652  * These bits are in the second u64 of the raw mask.
1653  */
1654 #define EF10_FEC_STAT_MASK (						\
1655 	(1ULL << (EF10_STAT_fec_uncorrected_errors - 64)) |		\
1656 	(1ULL << (EF10_STAT_fec_corrected_errors - 64)) |		\
1657 	(1ULL << (EF10_STAT_fec_corrected_symbols_lane0 - 64)) |	\
1658 	(1ULL << (EF10_STAT_fec_corrected_symbols_lane1 - 64)) |	\
1659 	(1ULL << (EF10_STAT_fec_corrected_symbols_lane2 - 64)) |	\
1660 	(1ULL << (EF10_STAT_fec_corrected_symbols_lane3 - 64)))
1661 
1662 /* These statistics are only provided if the NIC supports MC_CMD_MAC_STATS_V3,
1663  * indicated by returning a value >= MC_CMD_MAC_NSTATS_V3 in
1664  * MC_CMD_GET_CAPABILITIES_V4_OUT_MAC_STATS_NUM_STATS.
1665  * These bits are in the second u64 of the raw mask.
1666  */
1667 #define EF10_CTPIO_STAT_MASK (						\
1668 	(1ULL << (EF10_STAT_ctpio_vi_busy_fallback - 64)) |		\
1669 	(1ULL << (EF10_STAT_ctpio_long_write_success - 64)) |		\
1670 	(1ULL << (EF10_STAT_ctpio_missing_dbell_fail - 64)) |		\
1671 	(1ULL << (EF10_STAT_ctpio_overflow_fail - 64)) |		\
1672 	(1ULL << (EF10_STAT_ctpio_underflow_fail - 64)) |		\
1673 	(1ULL << (EF10_STAT_ctpio_timeout_fail - 64)) |			\
1674 	(1ULL << (EF10_STAT_ctpio_noncontig_wr_fail - 64)) |		\
1675 	(1ULL << (EF10_STAT_ctpio_frm_clobber_fail - 64)) |		\
1676 	(1ULL << (EF10_STAT_ctpio_invalid_wr_fail - 64)) |		\
1677 	(1ULL << (EF10_STAT_ctpio_vi_clobber_fallback - 64)) |		\
1678 	(1ULL << (EF10_STAT_ctpio_unqualified_fallback - 64)) |		\
1679 	(1ULL << (EF10_STAT_ctpio_runt_fallback - 64)) |		\
1680 	(1ULL << (EF10_STAT_ctpio_success - 64)) |			\
1681 	(1ULL << (EF10_STAT_ctpio_fallback - 64)) |			\
1682 	(1ULL << (EF10_STAT_ctpio_poison - 64)) |			\
1683 	(1ULL << (EF10_STAT_ctpio_erase - 64)))
1684 
1685 static u64 efx_ef10_raw_stat_mask(struct efx_nic *efx)
1686 {
1687 	u64 raw_mask = HUNT_COMMON_STAT_MASK;
1688 	u32 port_caps = efx_mcdi_phy_get_caps(efx);
1689 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1690 
1691 	if (!(efx->mcdi->fn_flags &
1692 	      1 << MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_LINKCTRL))
1693 		return 0;
1694 
1695 	if (port_caps & (1 << MC_CMD_PHY_CAP_40000FDX_LBN)) {
1696 		raw_mask |= HUNT_40G_EXTRA_STAT_MASK;
1697 		/* 8000 series have everything even at 40G */
1698 		if (nic_data->datapath_caps2 &
1699 		    (1 << MC_CMD_GET_CAPABILITIES_V2_OUT_MAC_STATS_40G_TX_SIZE_BINS_LBN))
1700 			raw_mask |= HUNT_10G_ONLY_STAT_MASK;
1701 	} else {
1702 		raw_mask |= HUNT_10G_ONLY_STAT_MASK;
1703 	}
1704 
1705 	if (nic_data->datapath_caps &
1706 	    (1 << MC_CMD_GET_CAPABILITIES_OUT_PM_AND_RXDP_COUNTERS_LBN))
1707 		raw_mask |= HUNT_PM_AND_RXDP_STAT_MASK;
1708 
1709 	return raw_mask;
1710 }
1711 
1712 static void efx_ef10_get_stat_mask(struct efx_nic *efx, unsigned long *mask)
1713 {
1714 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1715 	u64 raw_mask[2];
1716 
1717 	raw_mask[0] = efx_ef10_raw_stat_mask(efx);
1718 
1719 	/* Only show vadaptor stats when EVB capability is present */
1720 	if (nic_data->datapath_caps &
1721 	    (1 << MC_CMD_GET_CAPABILITIES_OUT_EVB_LBN)) {
1722 		raw_mask[0] |= ~((1ULL << EF10_STAT_rx_unicast) - 1);
1723 		raw_mask[1] = (1ULL << (EF10_STAT_V1_COUNT - 64)) - 1;
1724 	} else {
1725 		raw_mask[1] = 0;
1726 	}
1727 	/* Only show FEC stats when NIC supports MC_CMD_MAC_STATS_V2 */
1728 	if (efx->num_mac_stats >= MC_CMD_MAC_NSTATS_V2)
1729 		raw_mask[1] |= EF10_FEC_STAT_MASK;
1730 
1731 	/* CTPIO stats appear in V3. Only show them on devices that actually
1732 	 * support CTPIO. Although this driver doesn't use CTPIO others might,
1733 	 * and we may be reporting the stats for the underlying port.
1734 	 */
1735 	if (efx->num_mac_stats >= MC_CMD_MAC_NSTATS_V3 &&
1736 	    (nic_data->datapath_caps2 &
1737 	     (1 << MC_CMD_GET_CAPABILITIES_V4_OUT_CTPIO_LBN)))
1738 		raw_mask[1] |= EF10_CTPIO_STAT_MASK;
1739 
1740 #if BITS_PER_LONG == 64
1741 	BUILD_BUG_ON(BITS_TO_LONGS(EF10_STAT_COUNT) != 2);
1742 	mask[0] = raw_mask[0];
1743 	mask[1] = raw_mask[1];
1744 #else
1745 	BUILD_BUG_ON(BITS_TO_LONGS(EF10_STAT_COUNT) != 3);
1746 	mask[0] = raw_mask[0] & 0xffffffff;
1747 	mask[1] = raw_mask[0] >> 32;
1748 	mask[2] = raw_mask[1] & 0xffffffff;
1749 #endif
1750 }
1751 
1752 static size_t efx_ef10_describe_stats(struct efx_nic *efx, u8 *names)
1753 {
1754 	DECLARE_BITMAP(mask, EF10_STAT_COUNT);
1755 
1756 	efx_ef10_get_stat_mask(efx, mask);
1757 	return efx_nic_describe_stats(efx_ef10_stat_desc, EF10_STAT_COUNT,
1758 				      mask, names);
1759 }
1760 
1761 static void efx_ef10_get_fec_stats(struct efx_nic *efx,
1762 				   struct ethtool_fec_stats *fec_stats)
1763 {
1764 	DECLARE_BITMAP(mask, EF10_STAT_COUNT);
1765 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1766 	u64 *stats = nic_data->stats;
1767 
1768 	efx_ef10_get_stat_mask(efx, mask);
1769 	if (test_bit(EF10_STAT_fec_corrected_errors, mask))
1770 		fec_stats->corrected_blocks.total =
1771 			stats[EF10_STAT_fec_corrected_errors];
1772 	if (test_bit(EF10_STAT_fec_uncorrected_errors, mask))
1773 		fec_stats->uncorrectable_blocks.total =
1774 			stats[EF10_STAT_fec_uncorrected_errors];
1775 }
1776 
1777 static size_t efx_ef10_update_stats_common(struct efx_nic *efx, u64 *full_stats,
1778 					   struct rtnl_link_stats64 *core_stats)
1779 {
1780 	DECLARE_BITMAP(mask, EF10_STAT_COUNT);
1781 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1782 	u64 *stats = nic_data->stats;
1783 	size_t stats_count = 0, index;
1784 
1785 	efx_ef10_get_stat_mask(efx, mask);
1786 
1787 	if (full_stats) {
1788 		for_each_set_bit(index, mask, EF10_STAT_COUNT) {
1789 			if (efx_ef10_stat_desc[index].name) {
1790 				*full_stats++ = stats[index];
1791 				++stats_count;
1792 			}
1793 		}
1794 	}
1795 
1796 	if (!core_stats)
1797 		return stats_count;
1798 
1799 	if (nic_data->datapath_caps &
1800 			1 << MC_CMD_GET_CAPABILITIES_OUT_EVB_LBN) {
1801 		/* Use vadaptor stats. */
1802 		core_stats->rx_packets = stats[EF10_STAT_rx_unicast] +
1803 					 stats[EF10_STAT_rx_multicast] +
1804 					 stats[EF10_STAT_rx_broadcast];
1805 		core_stats->tx_packets = stats[EF10_STAT_tx_unicast] +
1806 					 stats[EF10_STAT_tx_multicast] +
1807 					 stats[EF10_STAT_tx_broadcast];
1808 		core_stats->rx_bytes = stats[EF10_STAT_rx_unicast_bytes] +
1809 				       stats[EF10_STAT_rx_multicast_bytes] +
1810 				       stats[EF10_STAT_rx_broadcast_bytes];
1811 		core_stats->tx_bytes = stats[EF10_STAT_tx_unicast_bytes] +
1812 				       stats[EF10_STAT_tx_multicast_bytes] +
1813 				       stats[EF10_STAT_tx_broadcast_bytes];
1814 		core_stats->rx_dropped = stats[GENERIC_STAT_rx_nodesc_trunc] +
1815 					 stats[GENERIC_STAT_rx_noskb_drops];
1816 		core_stats->multicast = stats[EF10_STAT_rx_multicast];
1817 		core_stats->rx_crc_errors = stats[EF10_STAT_rx_bad];
1818 		core_stats->rx_fifo_errors = stats[EF10_STAT_rx_overflow];
1819 		core_stats->rx_errors = core_stats->rx_crc_errors;
1820 		core_stats->tx_errors = stats[EF10_STAT_tx_bad];
1821 	} else {
1822 		/* Use port stats. */
1823 		core_stats->rx_packets = stats[EF10_STAT_port_rx_packets];
1824 		core_stats->tx_packets = stats[EF10_STAT_port_tx_packets];
1825 		core_stats->rx_bytes = stats[EF10_STAT_port_rx_bytes];
1826 		core_stats->tx_bytes = stats[EF10_STAT_port_tx_bytes];
1827 		core_stats->rx_dropped = stats[EF10_STAT_port_rx_nodesc_drops] +
1828 					 stats[GENERIC_STAT_rx_nodesc_trunc] +
1829 					 stats[GENERIC_STAT_rx_noskb_drops];
1830 		core_stats->multicast = stats[EF10_STAT_port_rx_multicast];
1831 		core_stats->rx_length_errors =
1832 				stats[EF10_STAT_port_rx_gtjumbo] +
1833 				stats[EF10_STAT_port_rx_length_error];
1834 		core_stats->rx_crc_errors = stats[EF10_STAT_port_rx_bad];
1835 		core_stats->rx_frame_errors =
1836 				stats[EF10_STAT_port_rx_align_error];
1837 		core_stats->rx_fifo_errors = stats[EF10_STAT_port_rx_overflow];
1838 		core_stats->rx_errors = (core_stats->rx_length_errors +
1839 					 core_stats->rx_crc_errors +
1840 					 core_stats->rx_frame_errors);
1841 	}
1842 
1843 	return stats_count;
1844 }
1845 
1846 static size_t efx_ef10_update_stats_pf(struct efx_nic *efx, u64 *full_stats,
1847 				       struct rtnl_link_stats64 *core_stats)
1848 {
1849 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1850 	DECLARE_BITMAP(mask, EF10_STAT_COUNT);
1851 	u64 *stats = nic_data->stats;
1852 
1853 	efx_ef10_get_stat_mask(efx, mask);
1854 
1855 	efx_nic_copy_stats(efx, nic_data->mc_stats);
1856 	efx_nic_update_stats(efx_ef10_stat_desc, EF10_STAT_COUNT,
1857 			     mask, stats, nic_data->mc_stats, false);
1858 
1859 	/* Update derived statistics */
1860 	efx_nic_fix_nodesc_drop_stat(efx,
1861 				     &stats[EF10_STAT_port_rx_nodesc_drops]);
1862 	/* MC Firmware reads RX_BYTES and RX_GOOD_BYTES from the MAC.
1863 	 * It then calculates RX_BAD_BYTES and DMAs it to us with RX_BYTES.
1864 	 * We report these as port_rx_ stats. We are not given RX_GOOD_BYTES.
1865 	 * Here we calculate port_rx_good_bytes.
1866 	 */
1867 	stats[EF10_STAT_port_rx_good_bytes] =
1868 		stats[EF10_STAT_port_rx_bytes] -
1869 		stats[EF10_STAT_port_rx_bytes_minus_good_bytes];
1870 
1871 	/* The asynchronous reads used to calculate RX_BAD_BYTES in
1872 	 * MC Firmware are done such that we should not see an increase in
1873 	 * RX_BAD_BYTES when a good packet has arrived. Unfortunately this
1874 	 * does mean that the stat can decrease at times. Here we do not
1875 	 * update the stat unless it has increased or has gone to zero
1876 	 * (In the case of the NIC rebooting).
1877 	 * Please see Bug 33781 for a discussion of why things work this way.
1878 	 */
1879 	efx_update_diff_stat(&stats[EF10_STAT_port_rx_bad_bytes],
1880 			     stats[EF10_STAT_port_rx_bytes_minus_good_bytes]);
1881 	efx_update_sw_stats(efx, stats);
1882 
1883 	return efx_ef10_update_stats_common(efx, full_stats, core_stats);
1884 }
1885 
1886 static int efx_ef10_try_update_nic_stats_vf(struct efx_nic *efx)
1887 	__must_hold(&efx->stats_lock)
1888 {
1889 	MCDI_DECLARE_BUF(inbuf, MC_CMD_MAC_STATS_IN_LEN);
1890 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1891 	DECLARE_BITMAP(mask, EF10_STAT_COUNT);
1892 	__le64 generation_start, generation_end;
1893 	u64 *stats = nic_data->stats;
1894 	u32 dma_len = efx->num_mac_stats * sizeof(u64);
1895 	struct efx_buffer stats_buf;
1896 	__le64 *dma_stats;
1897 	int rc;
1898 
1899 	spin_unlock_bh(&efx->stats_lock);
1900 
1901 	efx_ef10_get_stat_mask(efx, mask);
1902 
1903 	rc = efx_nic_alloc_buffer(efx, &stats_buf, dma_len, GFP_KERNEL);
1904 	if (rc) {
1905 		spin_lock_bh(&efx->stats_lock);
1906 		return rc;
1907 	}
1908 
1909 	dma_stats = stats_buf.addr;
1910 	dma_stats[efx->num_mac_stats - 1] = EFX_MC_STATS_GENERATION_INVALID;
1911 
1912 	MCDI_SET_QWORD(inbuf, MAC_STATS_IN_DMA_ADDR, stats_buf.dma_addr);
1913 	MCDI_POPULATE_DWORD_1(inbuf, MAC_STATS_IN_CMD,
1914 			      MAC_STATS_IN_DMA, 1);
1915 	MCDI_SET_DWORD(inbuf, MAC_STATS_IN_DMA_LEN, dma_len);
1916 	MCDI_SET_DWORD(inbuf, MAC_STATS_IN_PORT_ID, EVB_PORT_ID_ASSIGNED);
1917 
1918 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_MAC_STATS, inbuf, sizeof(inbuf),
1919 				NULL, 0, NULL);
1920 	spin_lock_bh(&efx->stats_lock);
1921 	if (rc) {
1922 		/* Expect ENOENT if DMA queues have not been set up */
1923 		if (rc != -ENOENT || atomic_read(&efx->active_queues))
1924 			efx_mcdi_display_error(efx, MC_CMD_MAC_STATS,
1925 					       sizeof(inbuf), NULL, 0, rc);
1926 		goto out;
1927 	}
1928 
1929 	generation_end = dma_stats[efx->num_mac_stats - 1];
1930 	if (generation_end == EFX_MC_STATS_GENERATION_INVALID) {
1931 		WARN_ON_ONCE(1);
1932 		goto out;
1933 	}
1934 	rmb();
1935 	efx_nic_update_stats(efx_ef10_stat_desc, EF10_STAT_COUNT, mask,
1936 			     stats, stats_buf.addr, false);
1937 	rmb();
1938 	generation_start = dma_stats[MC_CMD_MAC_GENERATION_START];
1939 	if (generation_end != generation_start) {
1940 		rc = -EAGAIN;
1941 		goto out;
1942 	}
1943 
1944 	efx_update_sw_stats(efx, stats);
1945 out:
1946 	/* releasing a DMA coherent buffer with BH disabled can panic */
1947 	spin_unlock_bh(&efx->stats_lock);
1948 	efx_nic_free_buffer(efx, &stats_buf);
1949 	spin_lock_bh(&efx->stats_lock);
1950 	return rc;
1951 }
1952 
1953 static size_t efx_ef10_update_stats_vf(struct efx_nic *efx, u64 *full_stats,
1954 				       struct rtnl_link_stats64 *core_stats)
1955 {
1956 	if (efx_ef10_try_update_nic_stats_vf(efx))
1957 		return 0;
1958 
1959 	return efx_ef10_update_stats_common(efx, full_stats, core_stats);
1960 }
1961 
1962 static size_t efx_ef10_update_stats_atomic_vf(struct efx_nic *efx, u64 *full_stats,
1963 					      struct rtnl_link_stats64 *core_stats)
1964 {
1965 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1966 
1967 	/* In atomic context, cannot update HW stats.  Just update the
1968 	 * software stats and return so the caller can continue.
1969 	 */
1970 	efx_update_sw_stats(efx, nic_data->stats);
1971 	return efx_ef10_update_stats_common(efx, full_stats, core_stats);
1972 }
1973 
1974 static void efx_ef10_push_irq_moderation(struct efx_channel *channel)
1975 {
1976 	struct efx_nic *efx = channel->efx;
1977 	unsigned int mode, usecs;
1978 	efx_dword_t timer_cmd;
1979 
1980 	if (channel->irq_moderation_us) {
1981 		mode = 3;
1982 		usecs = channel->irq_moderation_us;
1983 	} else {
1984 		mode = 0;
1985 		usecs = 0;
1986 	}
1987 
1988 	if (EFX_EF10_WORKAROUND_61265(efx)) {
1989 		MCDI_DECLARE_BUF(inbuf, MC_CMD_SET_EVQ_TMR_IN_LEN);
1990 		unsigned int ns = usecs * 1000;
1991 
1992 		MCDI_SET_DWORD(inbuf, SET_EVQ_TMR_IN_INSTANCE,
1993 			       channel->channel);
1994 		MCDI_SET_DWORD(inbuf, SET_EVQ_TMR_IN_TMR_LOAD_REQ_NS, ns);
1995 		MCDI_SET_DWORD(inbuf, SET_EVQ_TMR_IN_TMR_RELOAD_REQ_NS, ns);
1996 		MCDI_SET_DWORD(inbuf, SET_EVQ_TMR_IN_TMR_MODE, mode);
1997 
1998 		efx_mcdi_rpc_async(efx, MC_CMD_SET_EVQ_TMR,
1999 				   inbuf, sizeof(inbuf), 0, NULL, 0);
2000 	} else if (EFX_EF10_WORKAROUND_35388(efx)) {
2001 		unsigned int ticks = efx_usecs_to_ticks(efx, usecs);
2002 
2003 		EFX_POPULATE_DWORD_3(timer_cmd, ERF_DD_EVQ_IND_TIMER_FLAGS,
2004 				     EFE_DD_EVQ_IND_TIMER_FLAGS,
2005 				     ERF_DD_EVQ_IND_TIMER_MODE, mode,
2006 				     ERF_DD_EVQ_IND_TIMER_VAL, ticks);
2007 		efx_writed_page(efx, &timer_cmd, ER_DD_EVQ_INDIRECT,
2008 				channel->channel);
2009 	} else {
2010 		unsigned int ticks = efx_usecs_to_ticks(efx, usecs);
2011 
2012 		EFX_POPULATE_DWORD_3(timer_cmd, ERF_DZ_TC_TIMER_MODE, mode,
2013 				     ERF_DZ_TC_TIMER_VAL, ticks,
2014 				     ERF_FZ_TC_TMR_REL_VAL, ticks);
2015 		efx_writed_page(efx, &timer_cmd, ER_DZ_EVQ_TMR,
2016 				channel->channel);
2017 	}
2018 }
2019 
2020 static void efx_ef10_get_wol_vf(struct efx_nic *efx,
2021 				struct ethtool_wolinfo *wol) {}
2022 
2023 static int efx_ef10_set_wol_vf(struct efx_nic *efx, u32 type)
2024 {
2025 	return -EOPNOTSUPP;
2026 }
2027 
2028 static void efx_ef10_get_wol(struct efx_nic *efx, struct ethtool_wolinfo *wol)
2029 {
2030 	wol->supported = 0;
2031 	wol->wolopts = 0;
2032 	memset(&wol->sopass, 0, sizeof(wol->sopass));
2033 }
2034 
2035 static int efx_ef10_set_wol(struct efx_nic *efx, u32 type)
2036 {
2037 	if (type != 0)
2038 		return -EINVAL;
2039 	return 0;
2040 }
2041 
2042 static void efx_ef10_mcdi_request(struct efx_nic *efx,
2043 				  const efx_dword_t *hdr, size_t hdr_len,
2044 				  const efx_dword_t *sdu, size_t sdu_len)
2045 {
2046 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2047 	u8 *pdu = nic_data->mcdi_buf.addr;
2048 
2049 	memcpy(pdu, hdr, hdr_len);
2050 	memcpy(pdu + hdr_len, sdu, sdu_len);
2051 	wmb();
2052 
2053 	/* The hardware provides 'low' and 'high' (doorbell) registers
2054 	 * for passing the 64-bit address of an MCDI request to
2055 	 * firmware.  However the dwords are swapped by firmware.  The
2056 	 * least significant bits of the doorbell are then 0 for all
2057 	 * MCDI requests due to alignment.
2058 	 */
2059 	_efx_writed(efx, cpu_to_le32((u64)nic_data->mcdi_buf.dma_addr >> 32),
2060 		    ER_DZ_MC_DB_LWRD);
2061 	_efx_writed(efx, cpu_to_le32((u32)nic_data->mcdi_buf.dma_addr),
2062 		    ER_DZ_MC_DB_HWRD);
2063 }
2064 
2065 static bool efx_ef10_mcdi_poll_response(struct efx_nic *efx)
2066 {
2067 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2068 	const efx_dword_t hdr = *(const efx_dword_t *)nic_data->mcdi_buf.addr;
2069 
2070 	rmb();
2071 	return EFX_DWORD_FIELD(hdr, MCDI_HEADER_RESPONSE);
2072 }
2073 
2074 static void
2075 efx_ef10_mcdi_read_response(struct efx_nic *efx, efx_dword_t *outbuf,
2076 			    size_t offset, size_t outlen)
2077 {
2078 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2079 	const u8 *pdu = nic_data->mcdi_buf.addr;
2080 
2081 	memcpy(outbuf, pdu + offset, outlen);
2082 }
2083 
2084 static void efx_ef10_mcdi_reboot_detected(struct efx_nic *efx)
2085 {
2086 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2087 
2088 	/* All our allocations have been reset */
2089 	efx_ef10_table_reset_mc_allocations(efx);
2090 
2091 	/* The datapath firmware might have been changed */
2092 	nic_data->must_check_datapath_caps = true;
2093 
2094 	/* MAC statistics have been cleared on the NIC; clear the local
2095 	 * statistic that we update with efx_update_diff_stat().
2096 	 */
2097 	nic_data->stats[EF10_STAT_port_rx_bad_bytes] = 0;
2098 }
2099 
2100 static int efx_ef10_mcdi_poll_reboot(struct efx_nic *efx)
2101 {
2102 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2103 	int rc;
2104 
2105 	rc = efx_ef10_get_warm_boot_count(efx);
2106 	if (rc < 0) {
2107 		/* The firmware is presumably in the process of
2108 		 * rebooting.  However, we are supposed to report each
2109 		 * reboot just once, so we must only do that once we
2110 		 * can read and store the updated warm boot count.
2111 		 */
2112 		return 0;
2113 	}
2114 
2115 	if (rc == nic_data->warm_boot_count)
2116 		return 0;
2117 
2118 	nic_data->warm_boot_count = rc;
2119 	efx_ef10_mcdi_reboot_detected(efx);
2120 
2121 	return -EIO;
2122 }
2123 
2124 /* Handle an MSI interrupt
2125  *
2126  * Handle an MSI hardware interrupt.  This routine schedules event
2127  * queue processing.  No interrupt acknowledgement cycle is necessary.
2128  * Also, we never need to check that the interrupt is for us, since
2129  * MSI interrupts cannot be shared.
2130  */
2131 static irqreturn_t efx_ef10_msi_interrupt(int irq, void *dev_id)
2132 {
2133 	struct efx_msi_context *context = dev_id;
2134 	struct efx_nic *efx = context->efx;
2135 
2136 	netif_vdbg(efx, intr, efx->net_dev,
2137 		   "IRQ %d on CPU %d\n", irq, raw_smp_processor_id());
2138 
2139 	if (likely(READ_ONCE(efx->irq_soft_enabled))) {
2140 		/* Note test interrupts */
2141 		if (context->index == efx->irq_level)
2142 			efx->last_irq_cpu = raw_smp_processor_id();
2143 
2144 		/* Schedule processing of the channel */
2145 		efx_schedule_channel_irq(efx->channel[context->index]);
2146 	}
2147 
2148 	return IRQ_HANDLED;
2149 }
2150 
2151 static irqreturn_t efx_ef10_legacy_interrupt(int irq, void *dev_id)
2152 {
2153 	struct efx_nic *efx = dev_id;
2154 	bool soft_enabled = READ_ONCE(efx->irq_soft_enabled);
2155 	struct efx_channel *channel;
2156 	efx_dword_t reg;
2157 	u32 queues;
2158 
2159 	/* Read the ISR which also ACKs the interrupts */
2160 	efx_readd(efx, &reg, ER_DZ_BIU_INT_ISR);
2161 	queues = EFX_DWORD_FIELD(reg, ERF_DZ_ISR_REG);
2162 
2163 	if (queues == 0)
2164 		return IRQ_NONE;
2165 
2166 	if (likely(soft_enabled)) {
2167 		/* Note test interrupts */
2168 		if (queues & (1U << efx->irq_level))
2169 			efx->last_irq_cpu = raw_smp_processor_id();
2170 
2171 		efx_for_each_channel(channel, efx) {
2172 			if (queues & 1)
2173 				efx_schedule_channel_irq(channel);
2174 			queues >>= 1;
2175 		}
2176 	}
2177 
2178 	netif_vdbg(efx, intr, efx->net_dev,
2179 		   "IRQ %d on CPU %d status " EFX_DWORD_FMT "\n",
2180 		   irq, raw_smp_processor_id(), EFX_DWORD_VAL(reg));
2181 
2182 	return IRQ_HANDLED;
2183 }
2184 
2185 static int efx_ef10_irq_test_generate(struct efx_nic *efx)
2186 {
2187 	MCDI_DECLARE_BUF(inbuf, MC_CMD_TRIGGER_INTERRUPT_IN_LEN);
2188 
2189 	if (efx_mcdi_set_workaround(efx, MC_CMD_WORKAROUND_BUG41750, true,
2190 				    NULL) == 0)
2191 		return -ENOTSUPP;
2192 
2193 	BUILD_BUG_ON(MC_CMD_TRIGGER_INTERRUPT_OUT_LEN != 0);
2194 
2195 	MCDI_SET_DWORD(inbuf, TRIGGER_INTERRUPT_IN_INTR_LEVEL, efx->irq_level);
2196 	return efx_mcdi_rpc(efx, MC_CMD_TRIGGER_INTERRUPT,
2197 			    inbuf, sizeof(inbuf), NULL, 0, NULL);
2198 }
2199 
2200 static int efx_ef10_tx_probe(struct efx_tx_queue *tx_queue)
2201 {
2202 	/* low two bits of label are what we want for type */
2203 	BUILD_BUG_ON((EFX_TXQ_TYPE_OUTER_CSUM | EFX_TXQ_TYPE_INNER_CSUM) != 3);
2204 	tx_queue->type = tx_queue->label & 3;
2205 	return efx_nic_alloc_buffer(tx_queue->efx, &tx_queue->txd.buf,
2206 				    (tx_queue->ptr_mask + 1) *
2207 				    sizeof(efx_qword_t),
2208 				    GFP_KERNEL);
2209 }
2210 
2211 /* This writes to the TX_DESC_WPTR and also pushes data */
2212 static inline void efx_ef10_push_tx_desc(struct efx_tx_queue *tx_queue,
2213 					 const efx_qword_t *txd)
2214 {
2215 	unsigned int write_ptr;
2216 	efx_oword_t reg;
2217 
2218 	write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
2219 	EFX_POPULATE_OWORD_1(reg, ERF_DZ_TX_DESC_WPTR, write_ptr);
2220 	reg.qword[0] = *txd;
2221 	efx_writeo_page(tx_queue->efx, &reg,
2222 			ER_DZ_TX_DESC_UPD, tx_queue->queue);
2223 }
2224 
2225 /* Add Firmware-Assisted TSO v2 option descriptors to a queue.
2226  */
2227 int efx_ef10_tx_tso_desc(struct efx_tx_queue *tx_queue, struct sk_buff *skb,
2228 			 bool *data_mapped)
2229 {
2230 	struct efx_tx_buffer *buffer;
2231 	u16 inner_ipv4_id = 0;
2232 	u16 outer_ipv4_id = 0;
2233 	struct tcphdr *tcp;
2234 	struct iphdr *ip;
2235 	u16 ip_tot_len;
2236 	u32 seqnum;
2237 	u32 mss;
2238 
2239 	EFX_WARN_ON_ONCE_PARANOID(tx_queue->tso_version != 2);
2240 
2241 	mss = skb_shinfo(skb)->gso_size;
2242 
2243 	if (unlikely(mss < 4)) {
2244 		WARN_ONCE(1, "MSS of %u is too small for TSO v2\n", mss);
2245 		return -EINVAL;
2246 	}
2247 
2248 	if (skb->encapsulation) {
2249 		if (!tx_queue->tso_encap)
2250 			return -EINVAL;
2251 		ip = ip_hdr(skb);
2252 		if (ip->version == 4)
2253 			outer_ipv4_id = ntohs(ip->id);
2254 
2255 		ip = inner_ip_hdr(skb);
2256 		tcp = inner_tcp_hdr(skb);
2257 	} else {
2258 		ip = ip_hdr(skb);
2259 		tcp = tcp_hdr(skb);
2260 	}
2261 
2262 	/* 8000-series EF10 hardware requires that IP Total Length be
2263 	 * greater than or equal to the value it will have in each segment
2264 	 * (which is at most mss + 208 + TCP header length), but also less
2265 	 * than (0x10000 - inner_network_header).  Otherwise the TCP
2266 	 * checksum calculation will be broken for encapsulated packets.
2267 	 * We fill in ip->tot_len with 0xff30, which should satisfy the
2268 	 * first requirement unless the MSS is ridiculously large (which
2269 	 * should be impossible as the driver max MTU is 9216); it is
2270 	 * guaranteed to satisfy the second as we only attempt TSO if
2271 	 * inner_network_header <= 208.
2272 	 */
2273 	ip_tot_len = 0x10000 - EFX_TSO2_MAX_HDRLEN;
2274 	EFX_WARN_ON_ONCE_PARANOID(mss + EFX_TSO2_MAX_HDRLEN +
2275 				  (tcp->doff << 2u) > ip_tot_len);
2276 
2277 	if (ip->version == 4) {
2278 		ip->tot_len = htons(ip_tot_len);
2279 		ip->check = 0;
2280 		inner_ipv4_id = ntohs(ip->id);
2281 	} else {
2282 		((struct ipv6hdr *)ip)->payload_len = htons(ip_tot_len);
2283 	}
2284 
2285 	seqnum = ntohl(tcp->seq);
2286 
2287 	buffer = efx_tx_queue_get_insert_buffer(tx_queue);
2288 
2289 	buffer->flags = EFX_TX_BUF_OPTION;
2290 	buffer->len = 0;
2291 	buffer->unmap_len = 0;
2292 	EFX_POPULATE_QWORD_5(buffer->option,
2293 			ESF_DZ_TX_DESC_IS_OPT, 1,
2294 			ESF_DZ_TX_OPTION_TYPE, ESE_DZ_TX_OPTION_DESC_TSO,
2295 			ESF_DZ_TX_TSO_OPTION_TYPE,
2296 			ESE_DZ_TX_TSO_OPTION_DESC_FATSO2A,
2297 			ESF_DZ_TX_TSO_IP_ID, inner_ipv4_id,
2298 			ESF_DZ_TX_TSO_TCP_SEQNO, seqnum
2299 			);
2300 	++tx_queue->insert_count;
2301 
2302 	buffer = efx_tx_queue_get_insert_buffer(tx_queue);
2303 
2304 	buffer->flags = EFX_TX_BUF_OPTION;
2305 	buffer->len = 0;
2306 	buffer->unmap_len = 0;
2307 	EFX_POPULATE_QWORD_5(buffer->option,
2308 			ESF_DZ_TX_DESC_IS_OPT, 1,
2309 			ESF_DZ_TX_OPTION_TYPE, ESE_DZ_TX_OPTION_DESC_TSO,
2310 			ESF_DZ_TX_TSO_OPTION_TYPE,
2311 			ESE_DZ_TX_TSO_OPTION_DESC_FATSO2B,
2312 			ESF_DZ_TX_TSO_OUTER_IPID, outer_ipv4_id,
2313 			ESF_DZ_TX_TSO_TCP_MSS, mss
2314 			);
2315 	++tx_queue->insert_count;
2316 
2317 	return 0;
2318 }
2319 
2320 static u32 efx_ef10_tso_versions(struct efx_nic *efx)
2321 {
2322 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2323 	u32 tso_versions = 0;
2324 
2325 	if (nic_data->datapath_caps &
2326 	    (1 << MC_CMD_GET_CAPABILITIES_OUT_TX_TSO_LBN))
2327 		tso_versions |= BIT(1);
2328 	if (nic_data->datapath_caps2 &
2329 	    (1 << MC_CMD_GET_CAPABILITIES_V2_OUT_TX_TSO_V2_LBN))
2330 		tso_versions |= BIT(2);
2331 	return tso_versions;
2332 }
2333 
2334 static void efx_ef10_tx_init(struct efx_tx_queue *tx_queue)
2335 {
2336 	bool csum_offload = tx_queue->type & EFX_TXQ_TYPE_OUTER_CSUM;
2337 	bool inner_csum = tx_queue->type & EFX_TXQ_TYPE_INNER_CSUM;
2338 	struct efx_channel *channel = tx_queue->channel;
2339 	struct efx_nic *efx = tx_queue->efx;
2340 	struct efx_ef10_nic_data *nic_data;
2341 	efx_qword_t *txd;
2342 	int rc;
2343 
2344 	nic_data = efx->nic_data;
2345 
2346 	/* Only attempt to enable TX timestamping if we have the license for it,
2347 	 * otherwise TXQ init will fail
2348 	 */
2349 	if (!(nic_data->licensed_features &
2350 	      (1 << LICENSED_V3_FEATURES_TX_TIMESTAMPS_LBN))) {
2351 		tx_queue->timestamping = false;
2352 		/* Disable sync events on this channel. */
2353 		if (efx->type->ptp_set_ts_sync_events)
2354 			efx->type->ptp_set_ts_sync_events(efx, false, false);
2355 	}
2356 
2357 	/* TSOv2 is a limited resource that can only be configured on a limited
2358 	 * number of queues. TSO without checksum offload is not really a thing,
2359 	 * so we only enable it for those queues.
2360 	 * TSOv2 cannot be used with Hardware timestamping, and is never needed
2361 	 * for XDP tx.
2362 	 */
2363 	if (efx_has_cap(efx, TX_TSO_V2)) {
2364 		if ((csum_offload || inner_csum) &&
2365 		    !tx_queue->timestamping && !tx_queue->xdp_tx) {
2366 			tx_queue->tso_version = 2;
2367 			netif_dbg(efx, hw, efx->net_dev, "Using TSOv2 for channel %u\n",
2368 				  channel->channel);
2369 		}
2370 	} else if (efx_has_cap(efx, TX_TSO)) {
2371 		tx_queue->tso_version = 1;
2372 	}
2373 
2374 	rc = efx_mcdi_tx_init(tx_queue);
2375 	if (rc)
2376 		goto fail;
2377 
2378 	/* A previous user of this TX queue might have set us up the
2379 	 * bomb by writing a descriptor to the TX push collector but
2380 	 * not the doorbell.  (Each collector belongs to a port, not a
2381 	 * queue or function, so cannot easily be reset.)  We must
2382 	 * attempt to push a no-op descriptor in its place.
2383 	 */
2384 	tx_queue->buffer[0].flags = EFX_TX_BUF_OPTION;
2385 	tx_queue->insert_count = 1;
2386 	txd = efx_tx_desc(tx_queue, 0);
2387 	EFX_POPULATE_QWORD_7(*txd,
2388 			     ESF_DZ_TX_DESC_IS_OPT, true,
2389 			     ESF_DZ_TX_OPTION_TYPE,
2390 			     ESE_DZ_TX_OPTION_DESC_CRC_CSUM,
2391 			     ESF_DZ_TX_OPTION_UDP_TCP_CSUM, csum_offload,
2392 			     ESF_DZ_TX_OPTION_IP_CSUM, csum_offload && tx_queue->tso_version != 2,
2393 			     ESF_DZ_TX_OPTION_INNER_UDP_TCP_CSUM, inner_csum,
2394 			     ESF_DZ_TX_OPTION_INNER_IP_CSUM, inner_csum && tx_queue->tso_version != 2,
2395 			     ESF_DZ_TX_TIMESTAMP, tx_queue->timestamping);
2396 	tx_queue->write_count = 1;
2397 
2398 	if (tx_queue->tso_version == 2 && efx_has_cap(efx, TX_TSO_V2_ENCAP))
2399 		tx_queue->tso_encap = true;
2400 
2401 	wmb();
2402 	efx_ef10_push_tx_desc(tx_queue, txd);
2403 
2404 	return;
2405 
2406 fail:
2407 	netdev_WARN(efx->net_dev, "failed to initialise TXQ %d\n",
2408 		    tx_queue->queue);
2409 }
2410 
2411 /* This writes to the TX_DESC_WPTR; write pointer for TX descriptor ring */
2412 static inline void efx_ef10_notify_tx_desc(struct efx_tx_queue *tx_queue)
2413 {
2414 	unsigned int write_ptr;
2415 	efx_dword_t reg;
2416 
2417 	write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
2418 	EFX_POPULATE_DWORD_1(reg, ERF_DZ_TX_DESC_WPTR_DWORD, write_ptr);
2419 	efx_writed_page(tx_queue->efx, &reg,
2420 			ER_DZ_TX_DESC_UPD_DWORD, tx_queue->queue);
2421 }
2422 
2423 #define EFX_EF10_MAX_TX_DESCRIPTOR_LEN 0x3fff
2424 
2425 static unsigned int efx_ef10_tx_limit_len(struct efx_tx_queue *tx_queue,
2426 					  dma_addr_t dma_addr, unsigned int len)
2427 {
2428 	if (len > EFX_EF10_MAX_TX_DESCRIPTOR_LEN) {
2429 		/* If we need to break across multiple descriptors we should
2430 		 * stop at a page boundary. This assumes the length limit is
2431 		 * greater than the page size.
2432 		 */
2433 		dma_addr_t end = dma_addr + EFX_EF10_MAX_TX_DESCRIPTOR_LEN;
2434 
2435 		BUILD_BUG_ON(EFX_EF10_MAX_TX_DESCRIPTOR_LEN < EFX_PAGE_SIZE);
2436 		len = (end & (~(EFX_PAGE_SIZE - 1))) - dma_addr;
2437 	}
2438 
2439 	return len;
2440 }
2441 
2442 static void efx_ef10_tx_write(struct efx_tx_queue *tx_queue)
2443 {
2444 	unsigned int old_write_count = tx_queue->write_count;
2445 	struct efx_tx_buffer *buffer;
2446 	unsigned int write_ptr;
2447 	efx_qword_t *txd;
2448 
2449 	tx_queue->xmit_pending = false;
2450 	if (unlikely(tx_queue->write_count == tx_queue->insert_count))
2451 		return;
2452 
2453 	do {
2454 		write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
2455 		buffer = &tx_queue->buffer[write_ptr];
2456 		txd = efx_tx_desc(tx_queue, write_ptr);
2457 		++tx_queue->write_count;
2458 
2459 		/* Create TX descriptor ring entry */
2460 		if (buffer->flags & EFX_TX_BUF_OPTION) {
2461 			*txd = buffer->option;
2462 			if (EFX_QWORD_FIELD(*txd, ESF_DZ_TX_OPTION_TYPE) == 1)
2463 				/* PIO descriptor */
2464 				tx_queue->packet_write_count = tx_queue->write_count;
2465 		} else {
2466 			tx_queue->packet_write_count = tx_queue->write_count;
2467 			BUILD_BUG_ON(EFX_TX_BUF_CONT != 1);
2468 			EFX_POPULATE_QWORD_3(
2469 				*txd,
2470 				ESF_DZ_TX_KER_CONT,
2471 				buffer->flags & EFX_TX_BUF_CONT,
2472 				ESF_DZ_TX_KER_BYTE_CNT, buffer->len,
2473 				ESF_DZ_TX_KER_BUF_ADDR, buffer->dma_addr);
2474 		}
2475 	} while (tx_queue->write_count != tx_queue->insert_count);
2476 
2477 	wmb(); /* Ensure descriptors are written before they are fetched */
2478 
2479 	if (efx_nic_may_push_tx_desc(tx_queue, old_write_count)) {
2480 		txd = efx_tx_desc(tx_queue,
2481 				  old_write_count & tx_queue->ptr_mask);
2482 		efx_ef10_push_tx_desc(tx_queue, txd);
2483 		++tx_queue->pushes;
2484 	} else {
2485 		efx_ef10_notify_tx_desc(tx_queue);
2486 	}
2487 }
2488 
2489 static int efx_ef10_probe_multicast_chaining(struct efx_nic *efx)
2490 {
2491 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2492 	unsigned int enabled, implemented;
2493 	bool want_workaround_26807;
2494 	int rc;
2495 
2496 	rc = efx_mcdi_get_workarounds(efx, &implemented, &enabled);
2497 	if (rc == -ENOSYS) {
2498 		/* GET_WORKAROUNDS was implemented before this workaround,
2499 		 * thus it must be unavailable in this firmware.
2500 		 */
2501 		nic_data->workaround_26807 = false;
2502 		return 0;
2503 	}
2504 	if (rc)
2505 		return rc;
2506 	want_workaround_26807 =
2507 		implemented & MC_CMD_GET_WORKAROUNDS_OUT_BUG26807;
2508 	nic_data->workaround_26807 =
2509 		!!(enabled & MC_CMD_GET_WORKAROUNDS_OUT_BUG26807);
2510 
2511 	if (want_workaround_26807 && !nic_data->workaround_26807) {
2512 		unsigned int flags;
2513 
2514 		rc = efx_mcdi_set_workaround(efx,
2515 					     MC_CMD_WORKAROUND_BUG26807,
2516 					     true, &flags);
2517 		if (!rc) {
2518 			if (flags &
2519 			    1 << MC_CMD_WORKAROUND_EXT_OUT_FLR_DONE_LBN) {
2520 				netif_info(efx, drv, efx->net_dev,
2521 					   "other functions on NIC have been reset\n");
2522 
2523 				/* With MCFW v4.6.x and earlier, the
2524 				 * boot count will have incremented,
2525 				 * so re-read the warm_boot_count
2526 				 * value now to ensure this function
2527 				 * doesn't think it has changed next
2528 				 * time it checks.
2529 				 */
2530 				rc = efx_ef10_get_warm_boot_count(efx);
2531 				if (rc >= 0) {
2532 					nic_data->warm_boot_count = rc;
2533 					rc = 0;
2534 				}
2535 			}
2536 			nic_data->workaround_26807 = true;
2537 		} else if (rc == -EPERM) {
2538 			rc = 0;
2539 		}
2540 	}
2541 	return rc;
2542 }
2543 
2544 static int efx_ef10_filter_table_probe(struct efx_nic *efx)
2545 {
2546 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2547 	int rc = efx_ef10_probe_multicast_chaining(efx);
2548 	struct efx_mcdi_filter_vlan *vlan;
2549 
2550 	if (rc)
2551 		return rc;
2552 	down_write(&efx->filter_sem);
2553 	rc = efx_mcdi_filter_table_probe(efx, nic_data->workaround_26807);
2554 
2555 	if (rc)
2556 		goto out_unlock;
2557 
2558 	list_for_each_entry(vlan, &nic_data->vlan_list, list) {
2559 		rc = efx_mcdi_filter_add_vlan(efx, vlan->vid);
2560 		if (rc)
2561 			goto fail_add_vlan;
2562 	}
2563 	goto out_unlock;
2564 
2565 fail_add_vlan:
2566 	efx_mcdi_filter_table_remove(efx);
2567 out_unlock:
2568 	up_write(&efx->filter_sem);
2569 	return rc;
2570 }
2571 
2572 static void efx_ef10_filter_table_remove(struct efx_nic *efx)
2573 {
2574 	down_write(&efx->filter_sem);
2575 	efx_mcdi_filter_table_remove(efx);
2576 	up_write(&efx->filter_sem);
2577 }
2578 
2579 /* This creates an entry in the RX descriptor queue */
2580 static inline void
2581 efx_ef10_build_rx_desc(struct efx_rx_queue *rx_queue, unsigned int index)
2582 {
2583 	struct efx_rx_buffer *rx_buf;
2584 	efx_qword_t *rxd;
2585 
2586 	rxd = efx_rx_desc(rx_queue, index);
2587 	rx_buf = efx_rx_buffer(rx_queue, index);
2588 	EFX_POPULATE_QWORD_2(*rxd,
2589 			     ESF_DZ_RX_KER_BYTE_CNT, rx_buf->len,
2590 			     ESF_DZ_RX_KER_BUF_ADDR, rx_buf->dma_addr);
2591 }
2592 
2593 static void efx_ef10_rx_write(struct efx_rx_queue *rx_queue)
2594 {
2595 	struct efx_nic *efx = rx_queue->efx;
2596 	unsigned int write_count;
2597 	efx_dword_t reg;
2598 
2599 	/* Firmware requires that RX_DESC_WPTR be a multiple of 8 */
2600 	write_count = rx_queue->added_count & ~7;
2601 	if (rx_queue->notified_count == write_count)
2602 		return;
2603 
2604 	do
2605 		efx_ef10_build_rx_desc(
2606 			rx_queue,
2607 			rx_queue->notified_count & rx_queue->ptr_mask);
2608 	while (++rx_queue->notified_count != write_count);
2609 
2610 	wmb();
2611 	EFX_POPULATE_DWORD_1(reg, ERF_DZ_RX_DESC_WPTR,
2612 			     write_count & rx_queue->ptr_mask);
2613 	efx_writed_page(efx, &reg, ER_DZ_RX_DESC_UPD,
2614 			efx_rx_queue_index(rx_queue));
2615 }
2616 
2617 static efx_mcdi_async_completer efx_ef10_rx_defer_refill_complete;
2618 
2619 static void efx_ef10_rx_defer_refill(struct efx_rx_queue *rx_queue)
2620 {
2621 	struct efx_channel *channel = efx_rx_queue_channel(rx_queue);
2622 	MCDI_DECLARE_BUF(inbuf, MC_CMD_DRIVER_EVENT_IN_LEN);
2623 	efx_qword_t event;
2624 
2625 	EFX_POPULATE_QWORD_2(event,
2626 			     ESF_DZ_EV_CODE, EFX_EF10_DRVGEN_EV,
2627 			     ESF_DZ_EV_DATA, EFX_EF10_REFILL);
2628 
2629 	MCDI_SET_DWORD(inbuf, DRIVER_EVENT_IN_EVQ, channel->channel);
2630 
2631 	/* MCDI_SET_QWORD is not appropriate here since EFX_POPULATE_* has
2632 	 * already swapped the data to little-endian order.
2633 	 */
2634 	memcpy(MCDI_PTR(inbuf, DRIVER_EVENT_IN_DATA), &event.u64[0],
2635 	       sizeof(efx_qword_t));
2636 
2637 	efx_mcdi_rpc_async(channel->efx, MC_CMD_DRIVER_EVENT,
2638 			   inbuf, sizeof(inbuf), 0,
2639 			   efx_ef10_rx_defer_refill_complete, 0);
2640 }
2641 
2642 static void
2643 efx_ef10_rx_defer_refill_complete(struct efx_nic *efx, unsigned long cookie,
2644 				  int rc, efx_dword_t *outbuf,
2645 				  size_t outlen_actual)
2646 {
2647 	/* nothing to do */
2648 }
2649 
2650 static int efx_ef10_ev_init(struct efx_channel *channel)
2651 {
2652 	struct efx_nic *efx = channel->efx;
2653 	struct efx_ef10_nic_data *nic_data;
2654 	bool use_v2, cut_thru;
2655 
2656 	nic_data = efx->nic_data;
2657 	use_v2 = nic_data->datapath_caps2 &
2658 			    1 << MC_CMD_GET_CAPABILITIES_V2_OUT_INIT_EVQ_V2_LBN;
2659 	cut_thru = !(nic_data->datapath_caps &
2660 			      1 << MC_CMD_GET_CAPABILITIES_OUT_RX_BATCHING_LBN);
2661 	return efx_mcdi_ev_init(channel, cut_thru, use_v2);
2662 }
2663 
2664 static void efx_ef10_handle_rx_wrong_queue(struct efx_rx_queue *rx_queue,
2665 					   unsigned int rx_queue_label)
2666 {
2667 	struct efx_nic *efx = rx_queue->efx;
2668 
2669 	netif_info(efx, hw, efx->net_dev,
2670 		   "rx event arrived on queue %d labeled as queue %u\n",
2671 		   efx_rx_queue_index(rx_queue), rx_queue_label);
2672 
2673 	efx_schedule_reset(efx, RESET_TYPE_DISABLE);
2674 }
2675 
2676 static void
2677 efx_ef10_handle_rx_bad_lbits(struct efx_rx_queue *rx_queue,
2678 			     unsigned int actual, unsigned int expected)
2679 {
2680 	unsigned int dropped = (actual - expected) & rx_queue->ptr_mask;
2681 	struct efx_nic *efx = rx_queue->efx;
2682 
2683 	netif_info(efx, hw, efx->net_dev,
2684 		   "dropped %d events (index=%d expected=%d)\n",
2685 		   dropped, actual, expected);
2686 
2687 	efx_schedule_reset(efx, RESET_TYPE_DISABLE);
2688 }
2689 
2690 /* partially received RX was aborted. clean up. */
2691 static void efx_ef10_handle_rx_abort(struct efx_rx_queue *rx_queue)
2692 {
2693 	unsigned int rx_desc_ptr;
2694 
2695 	netif_dbg(rx_queue->efx, hw, rx_queue->efx->net_dev,
2696 		  "scattered RX aborted (dropping %u buffers)\n",
2697 		  rx_queue->scatter_n);
2698 
2699 	rx_desc_ptr = rx_queue->removed_count & rx_queue->ptr_mask;
2700 
2701 	efx_rx_packet(rx_queue, rx_desc_ptr, rx_queue->scatter_n,
2702 		      0, EFX_RX_PKT_DISCARD);
2703 
2704 	rx_queue->removed_count += rx_queue->scatter_n;
2705 	rx_queue->scatter_n = 0;
2706 	rx_queue->scatter_len = 0;
2707 	++efx_rx_queue_channel(rx_queue)->n_rx_nodesc_trunc;
2708 }
2709 
2710 static u16 efx_ef10_handle_rx_event_errors(struct efx_channel *channel,
2711 					   unsigned int n_packets,
2712 					   unsigned int rx_encap_hdr,
2713 					   unsigned int rx_l3_class,
2714 					   unsigned int rx_l4_class,
2715 					   const efx_qword_t *event)
2716 {
2717 	struct efx_nic *efx = channel->efx;
2718 	bool handled = false;
2719 
2720 	if (EFX_QWORD_FIELD(*event, ESF_DZ_RX_ECRC_ERR)) {
2721 		if (!(efx->net_dev->features & NETIF_F_RXALL)) {
2722 			if (!efx->loopback_selftest)
2723 				channel->n_rx_eth_crc_err += n_packets;
2724 			return EFX_RX_PKT_DISCARD;
2725 		}
2726 		handled = true;
2727 	}
2728 	if (EFX_QWORD_FIELD(*event, ESF_DZ_RX_IPCKSUM_ERR)) {
2729 		if (unlikely(rx_encap_hdr != ESE_EZ_ENCAP_HDR_VXLAN &&
2730 			     rx_l3_class != ESE_DZ_L3_CLASS_IP4 &&
2731 			     rx_l3_class != ESE_DZ_L3_CLASS_IP4_FRAG &&
2732 			     rx_l3_class != ESE_DZ_L3_CLASS_IP6 &&
2733 			     rx_l3_class != ESE_DZ_L3_CLASS_IP6_FRAG))
2734 			netdev_WARN(efx->net_dev,
2735 				    "invalid class for RX_IPCKSUM_ERR: event="
2736 				    EFX_QWORD_FMT "\n",
2737 				    EFX_QWORD_VAL(*event));
2738 		if (!efx->loopback_selftest)
2739 			*(rx_encap_hdr ?
2740 			  &channel->n_rx_outer_ip_hdr_chksum_err :
2741 			  &channel->n_rx_ip_hdr_chksum_err) += n_packets;
2742 		return 0;
2743 	}
2744 	if (EFX_QWORD_FIELD(*event, ESF_DZ_RX_TCPUDP_CKSUM_ERR)) {
2745 		if (unlikely(rx_encap_hdr != ESE_EZ_ENCAP_HDR_VXLAN &&
2746 			     ((rx_l3_class != ESE_DZ_L3_CLASS_IP4 &&
2747 			       rx_l3_class != ESE_DZ_L3_CLASS_IP6) ||
2748 			      (rx_l4_class != ESE_FZ_L4_CLASS_TCP &&
2749 			       rx_l4_class != ESE_FZ_L4_CLASS_UDP))))
2750 			netdev_WARN(efx->net_dev,
2751 				    "invalid class for RX_TCPUDP_CKSUM_ERR: event="
2752 				    EFX_QWORD_FMT "\n",
2753 				    EFX_QWORD_VAL(*event));
2754 		if (!efx->loopback_selftest)
2755 			*(rx_encap_hdr ?
2756 			  &channel->n_rx_outer_tcp_udp_chksum_err :
2757 			  &channel->n_rx_tcp_udp_chksum_err) += n_packets;
2758 		return 0;
2759 	}
2760 	if (EFX_QWORD_FIELD(*event, ESF_EZ_RX_IP_INNER_CHKSUM_ERR)) {
2761 		if (unlikely(!rx_encap_hdr))
2762 			netdev_WARN(efx->net_dev,
2763 				    "invalid encapsulation type for RX_IP_INNER_CHKSUM_ERR: event="
2764 				    EFX_QWORD_FMT "\n",
2765 				    EFX_QWORD_VAL(*event));
2766 		else if (unlikely(rx_l3_class != ESE_DZ_L3_CLASS_IP4 &&
2767 				  rx_l3_class != ESE_DZ_L3_CLASS_IP4_FRAG &&
2768 				  rx_l3_class != ESE_DZ_L3_CLASS_IP6 &&
2769 				  rx_l3_class != ESE_DZ_L3_CLASS_IP6_FRAG))
2770 			netdev_WARN(efx->net_dev,
2771 				    "invalid class for RX_IP_INNER_CHKSUM_ERR: event="
2772 				    EFX_QWORD_FMT "\n",
2773 				    EFX_QWORD_VAL(*event));
2774 		if (!efx->loopback_selftest)
2775 			channel->n_rx_inner_ip_hdr_chksum_err += n_packets;
2776 		return 0;
2777 	}
2778 	if (EFX_QWORD_FIELD(*event, ESF_EZ_RX_TCP_UDP_INNER_CHKSUM_ERR)) {
2779 		if (unlikely(!rx_encap_hdr))
2780 			netdev_WARN(efx->net_dev,
2781 				    "invalid encapsulation type for RX_TCP_UDP_INNER_CHKSUM_ERR: event="
2782 				    EFX_QWORD_FMT "\n",
2783 				    EFX_QWORD_VAL(*event));
2784 		else if (unlikely((rx_l3_class != ESE_DZ_L3_CLASS_IP4 &&
2785 				   rx_l3_class != ESE_DZ_L3_CLASS_IP6) ||
2786 				  (rx_l4_class != ESE_FZ_L4_CLASS_TCP &&
2787 				   rx_l4_class != ESE_FZ_L4_CLASS_UDP)))
2788 			netdev_WARN(efx->net_dev,
2789 				    "invalid class for RX_TCP_UDP_INNER_CHKSUM_ERR: event="
2790 				    EFX_QWORD_FMT "\n",
2791 				    EFX_QWORD_VAL(*event));
2792 		if (!efx->loopback_selftest)
2793 			channel->n_rx_inner_tcp_udp_chksum_err += n_packets;
2794 		return 0;
2795 	}
2796 
2797 	WARN_ON(!handled); /* No error bits were recognised */
2798 	return 0;
2799 }
2800 
2801 static int efx_ef10_handle_rx_event(struct efx_channel *channel,
2802 				    const efx_qword_t *event)
2803 {
2804 	unsigned int rx_bytes, next_ptr_lbits, rx_queue_label;
2805 	unsigned int rx_l3_class, rx_l4_class, rx_encap_hdr;
2806 	unsigned int n_descs, n_packets, i;
2807 	struct efx_nic *efx = channel->efx;
2808 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2809 	struct efx_rx_queue *rx_queue;
2810 	efx_qword_t errors;
2811 	bool rx_cont;
2812 	u16 flags = 0;
2813 
2814 	if (unlikely(READ_ONCE(efx->reset_pending)))
2815 		return 0;
2816 
2817 	/* Basic packet information */
2818 	rx_bytes = EFX_QWORD_FIELD(*event, ESF_DZ_RX_BYTES);
2819 	next_ptr_lbits = EFX_QWORD_FIELD(*event, ESF_DZ_RX_DSC_PTR_LBITS);
2820 	rx_queue_label = EFX_QWORD_FIELD(*event, ESF_DZ_RX_QLABEL);
2821 	rx_l3_class = EFX_QWORD_FIELD(*event, ESF_DZ_RX_L3_CLASS);
2822 	rx_l4_class = EFX_QWORD_FIELD(*event, ESF_FZ_RX_L4_CLASS);
2823 	rx_cont = EFX_QWORD_FIELD(*event, ESF_DZ_RX_CONT);
2824 	rx_encap_hdr =
2825 		nic_data->datapath_caps &
2826 			(1 << MC_CMD_GET_CAPABILITIES_OUT_VXLAN_NVGRE_LBN) ?
2827 		EFX_QWORD_FIELD(*event, ESF_EZ_RX_ENCAP_HDR) :
2828 		ESE_EZ_ENCAP_HDR_NONE;
2829 
2830 	if (EFX_QWORD_FIELD(*event, ESF_DZ_RX_DROP_EVENT))
2831 		netdev_WARN(efx->net_dev, "saw RX_DROP_EVENT: event="
2832 			    EFX_QWORD_FMT "\n",
2833 			    EFX_QWORD_VAL(*event));
2834 
2835 	rx_queue = efx_channel_get_rx_queue(channel);
2836 
2837 	if (unlikely(rx_queue_label != efx_rx_queue_index(rx_queue)))
2838 		efx_ef10_handle_rx_wrong_queue(rx_queue, rx_queue_label);
2839 
2840 	n_descs = ((next_ptr_lbits - rx_queue->removed_count) &
2841 		   ((1 << ESF_DZ_RX_DSC_PTR_LBITS_WIDTH) - 1));
2842 
2843 	if (n_descs != rx_queue->scatter_n + 1) {
2844 		struct efx_ef10_nic_data *nic_data = efx->nic_data;
2845 
2846 		/* detect rx abort */
2847 		if (unlikely(n_descs == rx_queue->scatter_n)) {
2848 			if (rx_queue->scatter_n == 0 || rx_bytes != 0)
2849 				netdev_WARN(efx->net_dev,
2850 					    "invalid RX abort: scatter_n=%u event="
2851 					    EFX_QWORD_FMT "\n",
2852 					    rx_queue->scatter_n,
2853 					    EFX_QWORD_VAL(*event));
2854 			efx_ef10_handle_rx_abort(rx_queue);
2855 			return 0;
2856 		}
2857 
2858 		/* Check that RX completion merging is valid, i.e.
2859 		 * the current firmware supports it and this is a
2860 		 * non-scattered packet.
2861 		 */
2862 		if (!(nic_data->datapath_caps &
2863 		      (1 << MC_CMD_GET_CAPABILITIES_OUT_RX_BATCHING_LBN)) ||
2864 		    rx_queue->scatter_n != 0 || rx_cont) {
2865 			efx_ef10_handle_rx_bad_lbits(
2866 				rx_queue, next_ptr_lbits,
2867 				(rx_queue->removed_count +
2868 				 rx_queue->scatter_n + 1) &
2869 				((1 << ESF_DZ_RX_DSC_PTR_LBITS_WIDTH) - 1));
2870 			return 0;
2871 		}
2872 
2873 		/* Merged completion for multiple non-scattered packets */
2874 		rx_queue->scatter_n = 1;
2875 		rx_queue->scatter_len = 0;
2876 		n_packets = n_descs;
2877 		++channel->n_rx_merge_events;
2878 		channel->n_rx_merge_packets += n_packets;
2879 		flags |= EFX_RX_PKT_PREFIX_LEN;
2880 	} else {
2881 		++rx_queue->scatter_n;
2882 		rx_queue->scatter_len += rx_bytes;
2883 		if (rx_cont)
2884 			return 0;
2885 		n_packets = 1;
2886 	}
2887 
2888 	EFX_POPULATE_QWORD_5(errors, ESF_DZ_RX_ECRC_ERR, 1,
2889 				     ESF_DZ_RX_IPCKSUM_ERR, 1,
2890 				     ESF_DZ_RX_TCPUDP_CKSUM_ERR, 1,
2891 				     ESF_EZ_RX_IP_INNER_CHKSUM_ERR, 1,
2892 				     ESF_EZ_RX_TCP_UDP_INNER_CHKSUM_ERR, 1);
2893 	EFX_AND_QWORD(errors, *event, errors);
2894 	if (unlikely(!EFX_QWORD_IS_ZERO(errors))) {
2895 		flags |= efx_ef10_handle_rx_event_errors(channel, n_packets,
2896 							 rx_encap_hdr,
2897 							 rx_l3_class, rx_l4_class,
2898 							 event);
2899 	} else {
2900 		bool tcpudp = rx_l4_class == ESE_FZ_L4_CLASS_TCP ||
2901 			      rx_l4_class == ESE_FZ_L4_CLASS_UDP;
2902 
2903 		switch (rx_encap_hdr) {
2904 		case ESE_EZ_ENCAP_HDR_VXLAN: /* VxLAN or GENEVE */
2905 			flags |= EFX_RX_PKT_CSUMMED; /* outer UDP csum */
2906 			if (tcpudp)
2907 				flags |= EFX_RX_PKT_CSUM_LEVEL; /* inner L4 */
2908 			break;
2909 		case ESE_EZ_ENCAP_HDR_GRE:
2910 		case ESE_EZ_ENCAP_HDR_NONE:
2911 			if (tcpudp)
2912 				flags |= EFX_RX_PKT_CSUMMED;
2913 			break;
2914 		default:
2915 			netdev_WARN(efx->net_dev,
2916 				    "unknown encapsulation type: event="
2917 				    EFX_QWORD_FMT "\n",
2918 				    EFX_QWORD_VAL(*event));
2919 		}
2920 	}
2921 
2922 	if (rx_l4_class == ESE_FZ_L4_CLASS_TCP)
2923 		flags |= EFX_RX_PKT_TCP;
2924 
2925 	channel->irq_mod_score += 2 * n_packets;
2926 
2927 	/* Handle received packet(s) */
2928 	for (i = 0; i < n_packets; i++) {
2929 		efx_rx_packet(rx_queue,
2930 			      rx_queue->removed_count & rx_queue->ptr_mask,
2931 			      rx_queue->scatter_n, rx_queue->scatter_len,
2932 			      flags);
2933 		rx_queue->removed_count += rx_queue->scatter_n;
2934 	}
2935 
2936 	rx_queue->scatter_n = 0;
2937 	rx_queue->scatter_len = 0;
2938 
2939 	return n_packets;
2940 }
2941 
2942 static u32 efx_ef10_extract_event_ts(efx_qword_t *event)
2943 {
2944 	u32 tstamp;
2945 
2946 	tstamp = EFX_QWORD_FIELD(*event, TX_TIMESTAMP_EVENT_TSTAMP_DATA_HI);
2947 	tstamp <<= 16;
2948 	tstamp |= EFX_QWORD_FIELD(*event, TX_TIMESTAMP_EVENT_TSTAMP_DATA_LO);
2949 
2950 	return tstamp;
2951 }
2952 
2953 static void
2954 efx_ef10_handle_tx_event(struct efx_channel *channel, efx_qword_t *event)
2955 {
2956 	struct efx_nic *efx = channel->efx;
2957 	struct efx_tx_queue *tx_queue;
2958 	unsigned int tx_ev_desc_ptr;
2959 	unsigned int tx_ev_q_label;
2960 	unsigned int tx_ev_type;
2961 	u64 ts_part;
2962 
2963 	if (unlikely(READ_ONCE(efx->reset_pending)))
2964 		return;
2965 
2966 	if (unlikely(EFX_QWORD_FIELD(*event, ESF_DZ_TX_DROP_EVENT)))
2967 		return;
2968 
2969 	/* Get the transmit queue */
2970 	tx_ev_q_label = EFX_QWORD_FIELD(*event, ESF_DZ_TX_QLABEL);
2971 	tx_queue = channel->tx_queue + (tx_ev_q_label % EFX_MAX_TXQ_PER_CHANNEL);
2972 
2973 	if (!tx_queue->timestamping) {
2974 		/* Transmit completion */
2975 		tx_ev_desc_ptr = EFX_QWORD_FIELD(*event, ESF_DZ_TX_DESCR_INDX);
2976 		efx_xmit_done(tx_queue, tx_ev_desc_ptr & tx_queue->ptr_mask);
2977 		return;
2978 	}
2979 
2980 	/* Transmit timestamps are only available for 8XXX series. They result
2981 	 * in up to three events per packet. These occur in order, and are:
2982 	 *  - the normal completion event (may be omitted)
2983 	 *  - the low part of the timestamp
2984 	 *  - the high part of the timestamp
2985 	 *
2986 	 * It's possible for multiple completion events to appear before the
2987 	 * corresponding timestamps. So we can for example get:
2988 	 *  COMP N
2989 	 *  COMP N+1
2990 	 *  TS_LO N
2991 	 *  TS_HI N
2992 	 *  TS_LO N+1
2993 	 *  TS_HI N+1
2994 	 *
2995 	 * In addition it's also possible for the adjacent completions to be
2996 	 * merged, so we may not see COMP N above. As such, the completion
2997 	 * events are not very useful here.
2998 	 *
2999 	 * Each part of the timestamp is itself split across two 16 bit
3000 	 * fields in the event.
3001 	 */
3002 	tx_ev_type = EFX_QWORD_FIELD(*event, ESF_EZ_TX_SOFT1);
3003 
3004 	switch (tx_ev_type) {
3005 	case TX_TIMESTAMP_EVENT_TX_EV_COMPLETION:
3006 		/* Ignore this event - see above. */
3007 		break;
3008 
3009 	case TX_TIMESTAMP_EVENT_TX_EV_TSTAMP_LO:
3010 		ts_part = efx_ef10_extract_event_ts(event);
3011 		tx_queue->completed_timestamp_minor = ts_part;
3012 		break;
3013 
3014 	case TX_TIMESTAMP_EVENT_TX_EV_TSTAMP_HI:
3015 		ts_part = efx_ef10_extract_event_ts(event);
3016 		tx_queue->completed_timestamp_major = ts_part;
3017 
3018 		efx_xmit_done_single(tx_queue);
3019 		break;
3020 
3021 	default:
3022 		netif_err(efx, hw, efx->net_dev,
3023 			  "channel %d unknown tx event type %d (data "
3024 			  EFX_QWORD_FMT ")\n",
3025 			  channel->channel, tx_ev_type,
3026 			  EFX_QWORD_VAL(*event));
3027 		break;
3028 	}
3029 }
3030 
3031 static void
3032 efx_ef10_handle_driver_event(struct efx_channel *channel, efx_qword_t *event)
3033 {
3034 	struct efx_nic *efx = channel->efx;
3035 	int subcode;
3036 
3037 	subcode = EFX_QWORD_FIELD(*event, ESF_DZ_DRV_SUB_CODE);
3038 
3039 	switch (subcode) {
3040 	case ESE_DZ_DRV_TIMER_EV:
3041 	case ESE_DZ_DRV_WAKE_UP_EV:
3042 		break;
3043 	case ESE_DZ_DRV_START_UP_EV:
3044 		/* event queue init complete. ok. */
3045 		break;
3046 	default:
3047 		netif_err(efx, hw, efx->net_dev,
3048 			  "channel %d unknown driver event type %d"
3049 			  " (data " EFX_QWORD_FMT ")\n",
3050 			  channel->channel, subcode,
3051 			  EFX_QWORD_VAL(*event));
3052 
3053 	}
3054 }
3055 
3056 static void efx_ef10_handle_driver_generated_event(struct efx_channel *channel,
3057 						   efx_qword_t *event)
3058 {
3059 	struct efx_nic *efx = channel->efx;
3060 	u32 subcode;
3061 
3062 	subcode = EFX_QWORD_FIELD(*event, EFX_DWORD_0);
3063 
3064 	switch (subcode) {
3065 	case EFX_EF10_TEST:
3066 		channel->event_test_cpu = raw_smp_processor_id();
3067 		break;
3068 	case EFX_EF10_REFILL:
3069 		/* The queue must be empty, so we won't receive any rx
3070 		 * events, so efx_process_channel() won't refill the
3071 		 * queue. Refill it here
3072 		 */
3073 		efx_fast_push_rx_descriptors(&channel->rx_queue, true);
3074 		break;
3075 	default:
3076 		netif_err(efx, hw, efx->net_dev,
3077 			  "channel %d unknown driver event type %u"
3078 			  " (data " EFX_QWORD_FMT ")\n",
3079 			  channel->channel, (unsigned) subcode,
3080 			  EFX_QWORD_VAL(*event));
3081 	}
3082 }
3083 
3084 static int efx_ef10_ev_process(struct efx_channel *channel, int quota)
3085 {
3086 	struct efx_nic *efx = channel->efx;
3087 	efx_qword_t event, *p_event;
3088 	unsigned int read_ptr;
3089 	int ev_code;
3090 	int spent = 0;
3091 
3092 	if (quota <= 0)
3093 		return spent;
3094 
3095 	read_ptr = channel->eventq_read_ptr;
3096 
3097 	for (;;) {
3098 		p_event = efx_event(channel, read_ptr);
3099 		event = *p_event;
3100 
3101 		if (!efx_event_present(&event))
3102 			break;
3103 
3104 		EFX_SET_QWORD(*p_event);
3105 
3106 		++read_ptr;
3107 
3108 		ev_code = EFX_QWORD_FIELD(event, ESF_DZ_EV_CODE);
3109 
3110 		netif_vdbg(efx, drv, efx->net_dev,
3111 			   "processing event on %d " EFX_QWORD_FMT "\n",
3112 			   channel->channel, EFX_QWORD_VAL(event));
3113 
3114 		switch (ev_code) {
3115 		case ESE_DZ_EV_CODE_MCDI_EV:
3116 			efx_mcdi_process_event(channel, &event);
3117 			break;
3118 		case ESE_DZ_EV_CODE_RX_EV:
3119 			spent += efx_ef10_handle_rx_event(channel, &event);
3120 			if (spent >= quota) {
3121 				/* XXX can we split a merged event to
3122 				 * avoid going over-quota?
3123 				 */
3124 				spent = quota;
3125 				goto out;
3126 			}
3127 			break;
3128 		case ESE_DZ_EV_CODE_TX_EV:
3129 			efx_ef10_handle_tx_event(channel, &event);
3130 			break;
3131 		case ESE_DZ_EV_CODE_DRIVER_EV:
3132 			efx_ef10_handle_driver_event(channel, &event);
3133 			if (++spent == quota)
3134 				goto out;
3135 			break;
3136 		case EFX_EF10_DRVGEN_EV:
3137 			efx_ef10_handle_driver_generated_event(channel, &event);
3138 			break;
3139 		default:
3140 			netif_err(efx, hw, efx->net_dev,
3141 				  "channel %d unknown event type %d"
3142 				  " (data " EFX_QWORD_FMT ")\n",
3143 				  channel->channel, ev_code,
3144 				  EFX_QWORD_VAL(event));
3145 		}
3146 	}
3147 
3148 out:
3149 	channel->eventq_read_ptr = read_ptr;
3150 	return spent;
3151 }
3152 
3153 static void efx_ef10_ev_read_ack(struct efx_channel *channel)
3154 {
3155 	struct efx_nic *efx = channel->efx;
3156 	efx_dword_t rptr;
3157 
3158 	if (EFX_EF10_WORKAROUND_35388(efx)) {
3159 		BUILD_BUG_ON(EFX_MIN_EVQ_SIZE <
3160 			     (1 << ERF_DD_EVQ_IND_RPTR_WIDTH));
3161 		BUILD_BUG_ON(EFX_MAX_EVQ_SIZE >
3162 			     (1 << 2 * ERF_DD_EVQ_IND_RPTR_WIDTH));
3163 
3164 		EFX_POPULATE_DWORD_2(rptr, ERF_DD_EVQ_IND_RPTR_FLAGS,
3165 				     EFE_DD_EVQ_IND_RPTR_FLAGS_HIGH,
3166 				     ERF_DD_EVQ_IND_RPTR,
3167 				     (channel->eventq_read_ptr &
3168 				      channel->eventq_mask) >>
3169 				     ERF_DD_EVQ_IND_RPTR_WIDTH);
3170 		efx_writed_page(efx, &rptr, ER_DD_EVQ_INDIRECT,
3171 				channel->channel);
3172 		EFX_POPULATE_DWORD_2(rptr, ERF_DD_EVQ_IND_RPTR_FLAGS,
3173 				     EFE_DD_EVQ_IND_RPTR_FLAGS_LOW,
3174 				     ERF_DD_EVQ_IND_RPTR,
3175 				     channel->eventq_read_ptr &
3176 				     ((1 << ERF_DD_EVQ_IND_RPTR_WIDTH) - 1));
3177 		efx_writed_page(efx, &rptr, ER_DD_EVQ_INDIRECT,
3178 				channel->channel);
3179 	} else {
3180 		EFX_POPULATE_DWORD_1(rptr, ERF_DZ_EVQ_RPTR,
3181 				     channel->eventq_read_ptr &
3182 				     channel->eventq_mask);
3183 		efx_writed_page(efx, &rptr, ER_DZ_EVQ_RPTR, channel->channel);
3184 	}
3185 }
3186 
3187 static void efx_ef10_ev_test_generate(struct efx_channel *channel)
3188 {
3189 	MCDI_DECLARE_BUF(inbuf, MC_CMD_DRIVER_EVENT_IN_LEN);
3190 	struct efx_nic *efx = channel->efx;
3191 	efx_qword_t event;
3192 	int rc;
3193 
3194 	EFX_POPULATE_QWORD_2(event,
3195 			     ESF_DZ_EV_CODE, EFX_EF10_DRVGEN_EV,
3196 			     ESF_DZ_EV_DATA, EFX_EF10_TEST);
3197 
3198 	MCDI_SET_DWORD(inbuf, DRIVER_EVENT_IN_EVQ, channel->channel);
3199 
3200 	/* MCDI_SET_QWORD is not appropriate here since EFX_POPULATE_* has
3201 	 * already swapped the data to little-endian order.
3202 	 */
3203 	memcpy(MCDI_PTR(inbuf, DRIVER_EVENT_IN_DATA), &event.u64[0],
3204 	       sizeof(efx_qword_t));
3205 
3206 	rc = efx_mcdi_rpc(efx, MC_CMD_DRIVER_EVENT, inbuf, sizeof(inbuf),
3207 			  NULL, 0, NULL);
3208 	if (rc != 0)
3209 		goto fail;
3210 
3211 	return;
3212 
3213 fail:
3214 	WARN_ON(true);
3215 	netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
3216 }
3217 
3218 static void efx_ef10_prepare_flr(struct efx_nic *efx)
3219 {
3220 	atomic_set(&efx->active_queues, 0);
3221 }
3222 
3223 static int efx_ef10_vport_set_mac_address(struct efx_nic *efx)
3224 {
3225 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
3226 	u8 mac_old[ETH_ALEN];
3227 	int rc, rc2;
3228 
3229 	/* Only reconfigure a PF-created vport */
3230 	if (is_zero_ether_addr(nic_data->vport_mac))
3231 		return 0;
3232 
3233 	efx_device_detach_sync(efx);
3234 	efx_net_stop(efx->net_dev);
3235 	efx_ef10_filter_table_remove(efx);
3236 
3237 	rc = efx_ef10_vadaptor_free(efx, efx->vport_id);
3238 	if (rc)
3239 		goto restore_filters;
3240 
3241 	ether_addr_copy(mac_old, nic_data->vport_mac);
3242 	rc = efx_ef10_vport_del_mac(efx, efx->vport_id,
3243 				    nic_data->vport_mac);
3244 	if (rc)
3245 		goto restore_vadaptor;
3246 
3247 	rc = efx_ef10_vport_add_mac(efx, efx->vport_id,
3248 				    efx->net_dev->dev_addr);
3249 	if (!rc) {
3250 		ether_addr_copy(nic_data->vport_mac, efx->net_dev->dev_addr);
3251 	} else {
3252 		rc2 = efx_ef10_vport_add_mac(efx, efx->vport_id, mac_old);
3253 		if (rc2) {
3254 			/* Failed to add original MAC, so clear vport_mac */
3255 			eth_zero_addr(nic_data->vport_mac);
3256 			goto reset_nic;
3257 		}
3258 	}
3259 
3260 restore_vadaptor:
3261 	rc2 = efx_ef10_vadaptor_alloc(efx, efx->vport_id);
3262 	if (rc2)
3263 		goto reset_nic;
3264 restore_filters:
3265 	rc2 = efx_ef10_filter_table_probe(efx);
3266 	if (rc2)
3267 		goto reset_nic;
3268 
3269 	rc2 = efx_net_open(efx->net_dev);
3270 	if (rc2)
3271 		goto reset_nic;
3272 
3273 	efx_device_attach_if_not_resetting(efx);
3274 
3275 	return rc;
3276 
3277 reset_nic:
3278 	netif_err(efx, drv, efx->net_dev,
3279 		  "Failed to restore when changing MAC address - scheduling reset\n");
3280 	efx_schedule_reset(efx, RESET_TYPE_DATAPATH);
3281 
3282 	return rc ? rc : rc2;
3283 }
3284 
3285 static int efx_ef10_set_mac_address(struct efx_nic *efx)
3286 {
3287 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VADAPTOR_SET_MAC_IN_LEN);
3288 	bool was_enabled = efx->port_enabled;
3289 	int rc;
3290 
3291 #ifdef CONFIG_SFC_SRIOV
3292 	/* If this function is a VF and we have access to the parent PF,
3293 	 * then use the PF control path to attempt to change the VF MAC address.
3294 	 */
3295 	if (efx->pci_dev->is_virtfn && efx->pci_dev->physfn) {
3296 		struct efx_nic *efx_pf = pci_get_drvdata(efx->pci_dev->physfn);
3297 		struct efx_ef10_nic_data *nic_data = efx->nic_data;
3298 		u8 mac[ETH_ALEN];
3299 
3300 		/* net_dev->dev_addr can be zeroed by efx_net_stop in
3301 		 * efx_ef10_sriov_set_vf_mac, so pass in a copy.
3302 		 */
3303 		ether_addr_copy(mac, efx->net_dev->dev_addr);
3304 
3305 		rc = efx_ef10_sriov_set_vf_mac(efx_pf, nic_data->vf_index, mac);
3306 		if (!rc)
3307 			return 0;
3308 
3309 		netif_dbg(efx, drv, efx->net_dev,
3310 			  "Updating VF mac via PF failed (%d), setting directly\n",
3311 			  rc);
3312 	}
3313 #endif
3314 
3315 	efx_device_detach_sync(efx);
3316 	efx_net_stop(efx->net_dev);
3317 
3318 	mutex_lock(&efx->mac_lock);
3319 	efx_ef10_filter_table_remove(efx);
3320 
3321 	ether_addr_copy(MCDI_PTR(inbuf, VADAPTOR_SET_MAC_IN_MACADDR),
3322 			efx->net_dev->dev_addr);
3323 	MCDI_SET_DWORD(inbuf, VADAPTOR_SET_MAC_IN_UPSTREAM_PORT_ID,
3324 		       efx->vport_id);
3325 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_VADAPTOR_SET_MAC, inbuf,
3326 				sizeof(inbuf), NULL, 0, NULL);
3327 
3328 	efx_ef10_filter_table_probe(efx);
3329 	mutex_unlock(&efx->mac_lock);
3330 
3331 	if (was_enabled)
3332 		efx_net_open(efx->net_dev);
3333 	efx_device_attach_if_not_resetting(efx);
3334 
3335 	if (rc == -EPERM) {
3336 		netif_err(efx, drv, efx->net_dev,
3337 			  "Cannot change MAC address; use sfboot to enable"
3338 			  " mac-spoofing on this interface\n");
3339 	} else if (rc == -ENOSYS && !efx_ef10_is_vf(efx)) {
3340 		/* If the active MCFW does not support MC_CMD_VADAPTOR_SET_MAC
3341 		 * fall-back to the method of changing the MAC address on the
3342 		 * vport.  This only applies to PFs because such versions of
3343 		 * MCFW do not support VFs.
3344 		 */
3345 		rc = efx_ef10_vport_set_mac_address(efx);
3346 	} else if (rc) {
3347 		efx_mcdi_display_error(efx, MC_CMD_VADAPTOR_SET_MAC,
3348 				       sizeof(inbuf), NULL, 0, rc);
3349 	}
3350 
3351 	return rc;
3352 }
3353 
3354 static int efx_ef10_mac_reconfigure(struct efx_nic *efx, bool mtu_only)
3355 {
3356 	WARN_ON(!mutex_is_locked(&efx->mac_lock));
3357 
3358 	efx_mcdi_filter_sync_rx_mode(efx);
3359 
3360 	if (mtu_only && efx_has_cap(efx, SET_MAC_ENHANCED))
3361 		return efx_mcdi_set_mtu(efx);
3362 	return efx_mcdi_set_mac(efx);
3363 }
3364 
3365 static int efx_ef10_start_bist(struct efx_nic *efx, u32 bist_type)
3366 {
3367 	MCDI_DECLARE_BUF(inbuf, MC_CMD_START_BIST_IN_LEN);
3368 
3369 	MCDI_SET_DWORD(inbuf, START_BIST_IN_TYPE, bist_type);
3370 	return efx_mcdi_rpc(efx, MC_CMD_START_BIST, inbuf, sizeof(inbuf),
3371 			    NULL, 0, NULL);
3372 }
3373 
3374 /* MC BISTs follow a different poll mechanism to phy BISTs.
3375  * The BIST is done in the poll handler on the MC, and the MCDI command
3376  * will block until the BIST is done.
3377  */
3378 static int efx_ef10_poll_bist(struct efx_nic *efx)
3379 {
3380 	int rc;
3381 	MCDI_DECLARE_BUF(outbuf, MC_CMD_POLL_BIST_OUT_LEN);
3382 	size_t outlen;
3383 	u32 result;
3384 
3385 	rc = efx_mcdi_rpc(efx, MC_CMD_POLL_BIST, NULL, 0,
3386 			   outbuf, sizeof(outbuf), &outlen);
3387 	if (rc != 0)
3388 		return rc;
3389 
3390 	if (outlen < MC_CMD_POLL_BIST_OUT_LEN)
3391 		return -EIO;
3392 
3393 	result = MCDI_DWORD(outbuf, POLL_BIST_OUT_RESULT);
3394 	switch (result) {
3395 	case MC_CMD_POLL_BIST_PASSED:
3396 		netif_dbg(efx, hw, efx->net_dev, "BIST passed.\n");
3397 		return 0;
3398 	case MC_CMD_POLL_BIST_TIMEOUT:
3399 		netif_err(efx, hw, efx->net_dev, "BIST timed out\n");
3400 		return -EIO;
3401 	case MC_CMD_POLL_BIST_FAILED:
3402 		netif_err(efx, hw, efx->net_dev, "BIST failed.\n");
3403 		return -EIO;
3404 	default:
3405 		netif_err(efx, hw, efx->net_dev,
3406 			  "BIST returned unknown result %u", result);
3407 		return -EIO;
3408 	}
3409 }
3410 
3411 static int efx_ef10_run_bist(struct efx_nic *efx, u32 bist_type)
3412 {
3413 	int rc;
3414 
3415 	netif_dbg(efx, drv, efx->net_dev, "starting BIST type %u\n", bist_type);
3416 
3417 	rc = efx_ef10_start_bist(efx, bist_type);
3418 	if (rc != 0)
3419 		return rc;
3420 
3421 	return efx_ef10_poll_bist(efx);
3422 }
3423 
3424 static int
3425 efx_ef10_test_chip(struct efx_nic *efx, struct efx_self_tests *tests)
3426 {
3427 	int rc, rc2;
3428 
3429 	efx_reset_down(efx, RESET_TYPE_WORLD);
3430 
3431 	rc = efx_mcdi_rpc(efx, MC_CMD_ENABLE_OFFLINE_BIST,
3432 			  NULL, 0, NULL, 0, NULL);
3433 	if (rc != 0)
3434 		goto out;
3435 
3436 	tests->memory = efx_ef10_run_bist(efx, MC_CMD_MC_MEM_BIST) ? -1 : 1;
3437 	tests->registers = efx_ef10_run_bist(efx, MC_CMD_REG_BIST) ? -1 : 1;
3438 
3439 	rc = efx_mcdi_reset(efx, RESET_TYPE_WORLD);
3440 
3441 out:
3442 	if (rc == -EPERM)
3443 		rc = 0;
3444 	rc2 = efx_reset_up(efx, RESET_TYPE_WORLD, rc == 0);
3445 	return rc ? rc : rc2;
3446 }
3447 
3448 #ifdef CONFIG_SFC_MTD
3449 
3450 struct efx_ef10_nvram_type_info {
3451 	u16 type, type_mask;
3452 	u8 port;
3453 	const char *name;
3454 };
3455 
3456 static const struct efx_ef10_nvram_type_info efx_ef10_nvram_types[] = {
3457 	{ NVRAM_PARTITION_TYPE_MC_FIRMWARE,	   0,    0, "sfc_mcfw" },
3458 	{ NVRAM_PARTITION_TYPE_MC_FIRMWARE_BACKUP, 0,    0, "sfc_mcfw_backup" },
3459 	{ NVRAM_PARTITION_TYPE_EXPANSION_ROM,	   0,    0, "sfc_exp_rom" },
3460 	{ NVRAM_PARTITION_TYPE_STATIC_CONFIG,	   0,    0, "sfc_static_cfg" },
3461 	{ NVRAM_PARTITION_TYPE_DYNAMIC_CONFIG,	   0,    0, "sfc_dynamic_cfg" },
3462 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT0, 0,   0, "sfc_exp_rom_cfg" },
3463 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT1, 0,   1, "sfc_exp_rom_cfg" },
3464 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT2, 0,   2, "sfc_exp_rom_cfg" },
3465 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT3, 0,   3, "sfc_exp_rom_cfg" },
3466 	{ NVRAM_PARTITION_TYPE_LICENSE,		   0,    0, "sfc_license" },
3467 	{ NVRAM_PARTITION_TYPE_PHY_MIN,		   0xff, 0, "sfc_phy_fw" },
3468 	{ NVRAM_PARTITION_TYPE_MUM_FIRMWARE,	   0,    0, "sfc_mumfw" },
3469 	{ NVRAM_PARTITION_TYPE_EXPANSION_UEFI,	   0,    0, "sfc_uefi" },
3470 	{ NVRAM_PARTITION_TYPE_DYNCONFIG_DEFAULTS, 0,    0, "sfc_dynamic_cfg_dflt" },
3471 	{ NVRAM_PARTITION_TYPE_ROMCONFIG_DEFAULTS, 0,    0, "sfc_exp_rom_cfg_dflt" },
3472 	{ NVRAM_PARTITION_TYPE_STATUS,		   0,    0, "sfc_status" },
3473 	{ NVRAM_PARTITION_TYPE_BUNDLE,		   0,    0, "sfc_bundle" },
3474 	{ NVRAM_PARTITION_TYPE_BUNDLE_METADATA,	   0,    0, "sfc_bundle_metadata" },
3475 };
3476 #define EF10_NVRAM_PARTITION_COUNT	ARRAY_SIZE(efx_ef10_nvram_types)
3477 
3478 static int efx_ef10_mtd_probe_partition(struct efx_nic *efx,
3479 					struct efx_mcdi_mtd_partition *part,
3480 					unsigned int type,
3481 					unsigned long *found)
3482 {
3483 	MCDI_DECLARE_BUF(inbuf, MC_CMD_NVRAM_METADATA_IN_LEN);
3484 	MCDI_DECLARE_BUF(outbuf, MC_CMD_NVRAM_METADATA_OUT_LENMAX);
3485 	const struct efx_ef10_nvram_type_info *info;
3486 	size_t size, erase_size, outlen;
3487 	int type_idx = 0;
3488 	bool protected;
3489 	int rc;
3490 
3491 	for (type_idx = 0; ; type_idx++) {
3492 		if (type_idx == EF10_NVRAM_PARTITION_COUNT)
3493 			return -ENODEV;
3494 		info = efx_ef10_nvram_types + type_idx;
3495 		if ((type & ~info->type_mask) == info->type)
3496 			break;
3497 	}
3498 	if (info->port != efx_port_num(efx))
3499 		return -ENODEV;
3500 
3501 	rc = efx_mcdi_nvram_info(efx, type, &size, &erase_size, &protected);
3502 	if (rc)
3503 		return rc;
3504 	if (protected &&
3505 	    (type != NVRAM_PARTITION_TYPE_DYNCONFIG_DEFAULTS &&
3506 	     type != NVRAM_PARTITION_TYPE_ROMCONFIG_DEFAULTS))
3507 		/* Hide protected partitions that don't provide defaults. */
3508 		return -ENODEV;
3509 
3510 	if (protected)
3511 		/* Protected partitions are read only. */
3512 		erase_size = 0;
3513 
3514 	/* If we've already exposed a partition of this type, hide this
3515 	 * duplicate.  All operations on MTDs are keyed by the type anyway,
3516 	 * so we can't act on the duplicate.
3517 	 */
3518 	if (__test_and_set_bit(type_idx, found))
3519 		return -EEXIST;
3520 
3521 	part->nvram_type = type;
3522 
3523 	MCDI_SET_DWORD(inbuf, NVRAM_METADATA_IN_TYPE, type);
3524 	rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_METADATA, inbuf, sizeof(inbuf),
3525 			  outbuf, sizeof(outbuf), &outlen);
3526 	if (rc)
3527 		return rc;
3528 	if (outlen < MC_CMD_NVRAM_METADATA_OUT_LENMIN)
3529 		return -EIO;
3530 	if (MCDI_DWORD(outbuf, NVRAM_METADATA_OUT_FLAGS) &
3531 	    (1 << MC_CMD_NVRAM_METADATA_OUT_SUBTYPE_VALID_LBN))
3532 		part->fw_subtype = MCDI_DWORD(outbuf,
3533 					      NVRAM_METADATA_OUT_SUBTYPE);
3534 
3535 	part->common.dev_type_name = "EF10 NVRAM manager";
3536 	part->common.type_name = info->name;
3537 
3538 	part->common.mtd.type = MTD_NORFLASH;
3539 	part->common.mtd.flags = MTD_CAP_NORFLASH;
3540 	part->common.mtd.size = size;
3541 	part->common.mtd.erasesize = erase_size;
3542 	/* sfc_status is read-only */
3543 	if (!erase_size)
3544 		part->common.mtd.flags |= MTD_NO_ERASE;
3545 
3546 	return 0;
3547 }
3548 
3549 static int efx_ef10_mtd_probe(struct efx_nic *efx)
3550 {
3551 	MCDI_DECLARE_BUF(outbuf, MC_CMD_NVRAM_PARTITIONS_OUT_LENMAX);
3552 	DECLARE_BITMAP(found, EF10_NVRAM_PARTITION_COUNT) = { 0 };
3553 	struct efx_mcdi_mtd_partition *parts;
3554 	size_t outlen, n_parts_total, i, n_parts;
3555 	unsigned int type;
3556 	int rc;
3557 
3558 	ASSERT_RTNL();
3559 
3560 	BUILD_BUG_ON(MC_CMD_NVRAM_PARTITIONS_IN_LEN != 0);
3561 	rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_PARTITIONS, NULL, 0,
3562 			  outbuf, sizeof(outbuf), &outlen);
3563 	if (rc)
3564 		return rc;
3565 	if (outlen < MC_CMD_NVRAM_PARTITIONS_OUT_LENMIN)
3566 		return -EIO;
3567 
3568 	n_parts_total = MCDI_DWORD(outbuf, NVRAM_PARTITIONS_OUT_NUM_PARTITIONS);
3569 	if (n_parts_total >
3570 	    MCDI_VAR_ARRAY_LEN(outlen, NVRAM_PARTITIONS_OUT_TYPE_ID))
3571 		return -EIO;
3572 
3573 	parts = kcalloc(n_parts_total, sizeof(*parts), GFP_KERNEL);
3574 	if (!parts)
3575 		return -ENOMEM;
3576 
3577 	n_parts = 0;
3578 	for (i = 0; i < n_parts_total; i++) {
3579 		type = MCDI_ARRAY_DWORD(outbuf, NVRAM_PARTITIONS_OUT_TYPE_ID,
3580 					i);
3581 		rc = efx_ef10_mtd_probe_partition(efx, &parts[n_parts], type,
3582 						  found);
3583 		if (rc == -EEXIST || rc == -ENODEV)
3584 			continue;
3585 		if (rc)
3586 			goto fail;
3587 		n_parts++;
3588 	}
3589 
3590 	if (!n_parts) {
3591 		kfree(parts);
3592 		return 0;
3593 	}
3594 
3595 	rc = efx_mtd_add(efx, &parts[0].common, n_parts, sizeof(*parts));
3596 fail:
3597 	if (rc)
3598 		kfree(parts);
3599 	return rc;
3600 }
3601 
3602 #endif /* CONFIG_SFC_MTD */
3603 
3604 static void efx_ef10_ptp_write_host_time(struct efx_nic *efx, u32 host_time)
3605 {
3606 	_efx_writed(efx, cpu_to_le32(host_time), ER_DZ_MC_DB_LWRD);
3607 }
3608 
3609 static void efx_ef10_ptp_write_host_time_vf(struct efx_nic *efx,
3610 					    u32 host_time) {}
3611 
3612 static int efx_ef10_rx_enable_timestamping(struct efx_channel *channel,
3613 					   bool temp)
3614 {
3615 	MCDI_DECLARE_BUF(inbuf, MC_CMD_PTP_IN_TIME_EVENT_SUBSCRIBE_LEN);
3616 	int rc;
3617 
3618 	if (channel->sync_events_state == SYNC_EVENTS_REQUESTED ||
3619 	    channel->sync_events_state == SYNC_EVENTS_VALID ||
3620 	    (temp && channel->sync_events_state == SYNC_EVENTS_DISABLED))
3621 		return 0;
3622 	channel->sync_events_state = SYNC_EVENTS_REQUESTED;
3623 
3624 	MCDI_SET_DWORD(inbuf, PTP_IN_OP, MC_CMD_PTP_OP_TIME_EVENT_SUBSCRIBE);
3625 	MCDI_SET_DWORD(inbuf, PTP_IN_PERIPH_ID, 0);
3626 	MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_SUBSCRIBE_QUEUE,
3627 		       channel->channel);
3628 
3629 	rc = efx_mcdi_rpc(channel->efx, MC_CMD_PTP,
3630 			  inbuf, sizeof(inbuf), NULL, 0, NULL);
3631 
3632 	if (rc != 0)
3633 		channel->sync_events_state = temp ? SYNC_EVENTS_QUIESCENT :
3634 						    SYNC_EVENTS_DISABLED;
3635 
3636 	return rc;
3637 }
3638 
3639 static int efx_ef10_rx_disable_timestamping(struct efx_channel *channel,
3640 					    bool temp)
3641 {
3642 	MCDI_DECLARE_BUF(inbuf, MC_CMD_PTP_IN_TIME_EVENT_UNSUBSCRIBE_LEN);
3643 	int rc;
3644 
3645 	if (channel->sync_events_state == SYNC_EVENTS_DISABLED ||
3646 	    (temp && channel->sync_events_state == SYNC_EVENTS_QUIESCENT))
3647 		return 0;
3648 	if (channel->sync_events_state == SYNC_EVENTS_QUIESCENT) {
3649 		channel->sync_events_state = SYNC_EVENTS_DISABLED;
3650 		return 0;
3651 	}
3652 	channel->sync_events_state = temp ? SYNC_EVENTS_QUIESCENT :
3653 					    SYNC_EVENTS_DISABLED;
3654 
3655 	MCDI_SET_DWORD(inbuf, PTP_IN_OP, MC_CMD_PTP_OP_TIME_EVENT_UNSUBSCRIBE);
3656 	MCDI_SET_DWORD(inbuf, PTP_IN_PERIPH_ID, 0);
3657 	MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_UNSUBSCRIBE_CONTROL,
3658 		       MC_CMD_PTP_IN_TIME_EVENT_UNSUBSCRIBE_SINGLE);
3659 	MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_UNSUBSCRIBE_QUEUE,
3660 		       channel->channel);
3661 
3662 	rc = efx_mcdi_rpc(channel->efx, MC_CMD_PTP,
3663 			  inbuf, sizeof(inbuf), NULL, 0, NULL);
3664 
3665 	return rc;
3666 }
3667 
3668 static int efx_ef10_ptp_set_ts_sync_events(struct efx_nic *efx, bool en,
3669 					   bool temp)
3670 {
3671 	int (*set)(struct efx_channel *channel, bool temp);
3672 	struct efx_channel *channel;
3673 
3674 	set = en ?
3675 	      efx_ef10_rx_enable_timestamping :
3676 	      efx_ef10_rx_disable_timestamping;
3677 
3678 	channel = efx_ptp_channel(efx);
3679 	if (channel) {
3680 		int rc = set(channel, temp);
3681 		if (en && rc != 0) {
3682 			efx_ef10_ptp_set_ts_sync_events(efx, false, temp);
3683 			return rc;
3684 		}
3685 	}
3686 
3687 	return 0;
3688 }
3689 
3690 static int efx_ef10_ptp_set_ts_config_vf(struct efx_nic *efx,
3691 					 struct hwtstamp_config *init)
3692 {
3693 	return -EOPNOTSUPP;
3694 }
3695 
3696 static int efx_ef10_ptp_set_ts_config(struct efx_nic *efx,
3697 				      struct hwtstamp_config *init)
3698 {
3699 	int rc;
3700 
3701 	switch (init->rx_filter) {
3702 	case HWTSTAMP_FILTER_NONE:
3703 		efx_ef10_ptp_set_ts_sync_events(efx, false, false);
3704 		/* if TX timestamping is still requested then leave PTP on */
3705 		return efx_ptp_change_mode(efx,
3706 					   init->tx_type != HWTSTAMP_TX_OFF, 0);
3707 	case HWTSTAMP_FILTER_ALL:
3708 	case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
3709 	case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
3710 	case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
3711 	case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
3712 	case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
3713 	case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
3714 	case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
3715 	case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
3716 	case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
3717 	case HWTSTAMP_FILTER_PTP_V2_EVENT:
3718 	case HWTSTAMP_FILTER_PTP_V2_SYNC:
3719 	case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
3720 	case HWTSTAMP_FILTER_NTP_ALL:
3721 		init->rx_filter = HWTSTAMP_FILTER_ALL;
3722 		rc = efx_ptp_change_mode(efx, true, 0);
3723 		if (!rc)
3724 			rc = efx_ef10_ptp_set_ts_sync_events(efx, true, false);
3725 		if (rc)
3726 			efx_ptp_change_mode(efx, false, 0);
3727 		return rc;
3728 	default:
3729 		return -ERANGE;
3730 	}
3731 }
3732 
3733 static int efx_ef10_get_phys_port_id(struct efx_nic *efx,
3734 				     struct netdev_phys_item_id *ppid)
3735 {
3736 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
3737 
3738 	if (!is_valid_ether_addr(nic_data->port_id))
3739 		return -EOPNOTSUPP;
3740 
3741 	ppid->id_len = ETH_ALEN;
3742 	memcpy(ppid->id, nic_data->port_id, ppid->id_len);
3743 
3744 	return 0;
3745 }
3746 
3747 static int efx_ef10_vlan_rx_add_vid(struct efx_nic *efx, __be16 proto, u16 vid)
3748 {
3749 	if (proto != htons(ETH_P_8021Q))
3750 		return -EINVAL;
3751 
3752 	return efx_ef10_add_vlan(efx, vid);
3753 }
3754 
3755 static int efx_ef10_vlan_rx_kill_vid(struct efx_nic *efx, __be16 proto, u16 vid)
3756 {
3757 	if (proto != htons(ETH_P_8021Q))
3758 		return -EINVAL;
3759 
3760 	return efx_ef10_del_vlan(efx, vid);
3761 }
3762 
3763 /* We rely on the MCDI wiping out our TX rings if it made any changes to the
3764  * ports table, ensuring that any TSO descriptors that were made on a now-
3765  * removed tunnel port will be blown away and won't break things when we try
3766  * to transmit them using the new ports table.
3767  */
3768 static int efx_ef10_set_udp_tnl_ports(struct efx_nic *efx, bool unloading)
3769 {
3770 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
3771 	MCDI_DECLARE_BUF(inbuf, MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_IN_LENMAX);
3772 	MCDI_DECLARE_BUF(outbuf, MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_OUT_LEN);
3773 	bool will_reset = false;
3774 	size_t num_entries = 0;
3775 	size_t inlen, outlen;
3776 	size_t i;
3777 	int rc;
3778 	efx_dword_t flags_and_num_entries;
3779 
3780 	WARN_ON(!mutex_is_locked(&nic_data->udp_tunnels_lock));
3781 
3782 	nic_data->udp_tunnels_dirty = false;
3783 
3784 	if (!(nic_data->datapath_caps &
3785 	    (1 << MC_CMD_GET_CAPABILITIES_OUT_VXLAN_NVGRE_LBN))) {
3786 		efx_device_attach_if_not_resetting(efx);
3787 		return 0;
3788 	}
3789 
3790 	BUILD_BUG_ON(ARRAY_SIZE(nic_data->udp_tunnels) >
3791 		     MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_IN_ENTRIES_MAXNUM);
3792 
3793 	for (i = 0; i < ARRAY_SIZE(nic_data->udp_tunnels); ++i) {
3794 		if (nic_data->udp_tunnels[i].type !=
3795 		    TUNNEL_ENCAP_UDP_PORT_ENTRY_INVALID) {
3796 			efx_dword_t entry;
3797 
3798 			EFX_POPULATE_DWORD_2(entry,
3799 				TUNNEL_ENCAP_UDP_PORT_ENTRY_UDP_PORT,
3800 					ntohs(nic_data->udp_tunnels[i].port),
3801 				TUNNEL_ENCAP_UDP_PORT_ENTRY_PROTOCOL,
3802 					nic_data->udp_tunnels[i].type);
3803 			*_MCDI_ARRAY_DWORD(inbuf,
3804 				SET_TUNNEL_ENCAP_UDP_PORTS_IN_ENTRIES,
3805 				num_entries++) = entry;
3806 		}
3807 	}
3808 
3809 	BUILD_BUG_ON((MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_IN_NUM_ENTRIES_OFST -
3810 		      MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_IN_FLAGS_OFST) * 8 !=
3811 		     EFX_WORD_1_LBN);
3812 	BUILD_BUG_ON(MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_IN_NUM_ENTRIES_LEN * 8 !=
3813 		     EFX_WORD_1_WIDTH);
3814 	EFX_POPULATE_DWORD_2(flags_and_num_entries,
3815 			     MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_IN_UNLOADING,
3816 				!!unloading,
3817 			     EFX_WORD_1, num_entries);
3818 	*_MCDI_DWORD(inbuf, SET_TUNNEL_ENCAP_UDP_PORTS_IN_FLAGS) =
3819 		flags_and_num_entries;
3820 
3821 	inlen = MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_IN_LEN(num_entries);
3822 
3823 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS,
3824 				inbuf, inlen, outbuf, sizeof(outbuf), &outlen);
3825 	if (rc == -EIO) {
3826 		/* Most likely the MC rebooted due to another function also
3827 		 * setting its tunnel port list. Mark the tunnel port list as
3828 		 * dirty, so it will be pushed upon coming up from the reboot.
3829 		 */
3830 		nic_data->udp_tunnels_dirty = true;
3831 		return 0;
3832 	}
3833 
3834 	if (rc) {
3835 		/* expected not available on unprivileged functions */
3836 		if (rc != -EPERM)
3837 			netif_warn(efx, drv, efx->net_dev,
3838 				   "Unable to set UDP tunnel ports; rc=%d.\n", rc);
3839 	} else if (MCDI_DWORD(outbuf, SET_TUNNEL_ENCAP_UDP_PORTS_OUT_FLAGS) &
3840 		   (1 << MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_OUT_RESETTING_LBN)) {
3841 		netif_info(efx, drv, efx->net_dev,
3842 			   "Rebooting MC due to UDP tunnel port list change\n");
3843 		will_reset = true;
3844 		if (unloading)
3845 			/* Delay for the MC reset to complete. This will make
3846 			 * unloading other functions a bit smoother. This is a
3847 			 * race, but the other unload will work whichever way
3848 			 * it goes, this just avoids an unnecessary error
3849 			 * message.
3850 			 */
3851 			msleep(100);
3852 	}
3853 	if (!will_reset && !unloading) {
3854 		/* The caller will have detached, relying on the MC reset to
3855 		 * trigger a re-attach.  Since there won't be an MC reset, we
3856 		 * have to do the attach ourselves.
3857 		 */
3858 		efx_device_attach_if_not_resetting(efx);
3859 	}
3860 
3861 	return rc;
3862 }
3863 
3864 static int efx_ef10_udp_tnl_push_ports(struct efx_nic *efx)
3865 {
3866 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
3867 	int rc = 0;
3868 
3869 	mutex_lock(&nic_data->udp_tunnels_lock);
3870 	if (nic_data->udp_tunnels_dirty) {
3871 		/* Make sure all TX are stopped while we modify the table, else
3872 		 * we might race against an efx_features_check().
3873 		 */
3874 		efx_device_detach_sync(efx);
3875 		rc = efx_ef10_set_udp_tnl_ports(efx, false);
3876 	}
3877 	mutex_unlock(&nic_data->udp_tunnels_lock);
3878 	return rc;
3879 }
3880 
3881 static int efx_ef10_udp_tnl_set_port(struct net_device *dev,
3882 				     unsigned int table, unsigned int entry,
3883 				     struct udp_tunnel_info *ti)
3884 {
3885 	struct efx_nic *efx = efx_netdev_priv(dev);
3886 	struct efx_ef10_nic_data *nic_data;
3887 	int efx_tunnel_type, rc;
3888 
3889 	if (ti->type == UDP_TUNNEL_TYPE_VXLAN)
3890 		efx_tunnel_type = TUNNEL_ENCAP_UDP_PORT_ENTRY_VXLAN;
3891 	else
3892 		efx_tunnel_type = TUNNEL_ENCAP_UDP_PORT_ENTRY_GENEVE;
3893 
3894 	nic_data = efx->nic_data;
3895 	if (!(nic_data->datapath_caps &
3896 	      (1 << MC_CMD_GET_CAPABILITIES_OUT_VXLAN_NVGRE_LBN)))
3897 		return -EOPNOTSUPP;
3898 
3899 	mutex_lock(&nic_data->udp_tunnels_lock);
3900 	/* Make sure all TX are stopped while we add to the table, else we
3901 	 * might race against an efx_features_check().
3902 	 */
3903 	efx_device_detach_sync(efx);
3904 	nic_data->udp_tunnels[entry].type = efx_tunnel_type;
3905 	nic_data->udp_tunnels[entry].port = ti->port;
3906 	rc = efx_ef10_set_udp_tnl_ports(efx, false);
3907 	mutex_unlock(&nic_data->udp_tunnels_lock);
3908 
3909 	return rc;
3910 }
3911 
3912 /* Called under the TX lock with the TX queue running, hence no-one can be
3913  * in the middle of updating the UDP tunnels table.  However, they could
3914  * have tried and failed the MCDI, in which case they'll have set the dirty
3915  * flag before dropping their locks.
3916  */
3917 static bool efx_ef10_udp_tnl_has_port(struct efx_nic *efx, __be16 port)
3918 {
3919 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
3920 	size_t i;
3921 
3922 	if (!(nic_data->datapath_caps &
3923 	      (1 << MC_CMD_GET_CAPABILITIES_OUT_VXLAN_NVGRE_LBN)))
3924 		return false;
3925 
3926 	if (nic_data->udp_tunnels_dirty)
3927 		/* SW table may not match HW state, so just assume we can't
3928 		 * use any UDP tunnel offloads.
3929 		 */
3930 		return false;
3931 
3932 	for (i = 0; i < ARRAY_SIZE(nic_data->udp_tunnels); ++i)
3933 		if (nic_data->udp_tunnels[i].type !=
3934 		    TUNNEL_ENCAP_UDP_PORT_ENTRY_INVALID &&
3935 		    nic_data->udp_tunnels[i].port == port)
3936 			return true;
3937 
3938 	return false;
3939 }
3940 
3941 static int efx_ef10_udp_tnl_unset_port(struct net_device *dev,
3942 				       unsigned int table, unsigned int entry,
3943 				       struct udp_tunnel_info *ti)
3944 {
3945 	struct efx_nic *efx = efx_netdev_priv(dev);
3946 	struct efx_ef10_nic_data *nic_data;
3947 	int rc;
3948 
3949 	nic_data = efx->nic_data;
3950 
3951 	mutex_lock(&nic_data->udp_tunnels_lock);
3952 	/* Make sure all TX are stopped while we remove from the table, else we
3953 	 * might race against an efx_features_check().
3954 	 */
3955 	efx_device_detach_sync(efx);
3956 	nic_data->udp_tunnels[entry].type = TUNNEL_ENCAP_UDP_PORT_ENTRY_INVALID;
3957 	nic_data->udp_tunnels[entry].port = 0;
3958 	rc = efx_ef10_set_udp_tnl_ports(efx, false);
3959 	mutex_unlock(&nic_data->udp_tunnels_lock);
3960 
3961 	return rc;
3962 }
3963 
3964 static const struct udp_tunnel_nic_info efx_ef10_udp_tunnels = {
3965 	.set_port	= efx_ef10_udp_tnl_set_port,
3966 	.unset_port	= efx_ef10_udp_tnl_unset_port,
3967 	.flags          = UDP_TUNNEL_NIC_INFO_MAY_SLEEP,
3968 	.tables         = {
3969 		{
3970 			.n_entries = 16,
3971 			.tunnel_types = UDP_TUNNEL_TYPE_VXLAN |
3972 					UDP_TUNNEL_TYPE_GENEVE,
3973 		},
3974 	},
3975 };
3976 
3977 /* EF10 may have multiple datapath firmware variants within a
3978  * single version.  Report which variants are running.
3979  */
3980 static size_t efx_ef10_print_additional_fwver(struct efx_nic *efx, char *buf,
3981 					      size_t len)
3982 {
3983 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
3984 
3985 	return scnprintf(buf, len, " rx%x tx%x",
3986 			 nic_data->rx_dpcpu_fw_id,
3987 			 nic_data->tx_dpcpu_fw_id);
3988 }
3989 
3990 static unsigned int ef10_check_caps(const struct efx_nic *efx,
3991 				    u8 flag,
3992 				    u32 offset)
3993 {
3994 	const struct efx_ef10_nic_data *nic_data = efx->nic_data;
3995 
3996 	switch (offset) {
3997 	case(MC_CMD_GET_CAPABILITIES_V4_OUT_FLAGS1_OFST):
3998 		return nic_data->datapath_caps & BIT_ULL(flag);
3999 	case(MC_CMD_GET_CAPABILITIES_V4_OUT_FLAGS2_OFST):
4000 		return nic_data->datapath_caps2 & BIT_ULL(flag);
4001 	default:
4002 		return 0;
4003 	}
4004 }
4005 
4006 static unsigned int efx_ef10_recycle_ring_size(const struct efx_nic *efx)
4007 {
4008 	unsigned int ret = EFX_RECYCLE_RING_SIZE_10G;
4009 
4010 	/* There is no difference between PFs and VFs. The side is based on
4011 	 * the maximum link speed of a given NIC.
4012 	 */
4013 	switch (efx->pci_dev->device & 0xfff) {
4014 	case 0x0903:	/* Farmingdale can do up to 10G */
4015 		break;
4016 	case 0x0923:	/* Greenport can do up to 40G */
4017 	case 0x0a03:	/* Medford can do up to 40G */
4018 		ret *= 4;
4019 		break;
4020 	default:	/* Medford2 can do up to 100G */
4021 		ret *= 10;
4022 	}
4023 
4024 	if (IS_ENABLED(CONFIG_PPC64))
4025 		ret *= 4;
4026 
4027 	return ret;
4028 }
4029 
4030 #define EF10_OFFLOAD_FEATURES		\
4031 	(NETIF_F_IP_CSUM |		\
4032 	 NETIF_F_HW_VLAN_CTAG_FILTER |	\
4033 	 NETIF_F_IPV6_CSUM |		\
4034 	 NETIF_F_RXHASH |		\
4035 	 NETIF_F_NTUPLE |		\
4036 	 NETIF_F_SG |			\
4037 	 NETIF_F_RXCSUM |		\
4038 	 NETIF_F_RXALL)
4039 
4040 const struct efx_nic_type efx_hunt_a0_vf_nic_type = {
4041 	.is_vf = true,
4042 	.mem_bar = efx_ef10_vf_mem_bar,
4043 	.mem_map_size = efx_ef10_mem_map_size,
4044 	.probe = efx_ef10_probe_vf,
4045 	.remove = efx_ef10_remove,
4046 	.dimension_resources = efx_ef10_dimension_resources,
4047 	.init = efx_ef10_init_nic,
4048 	.fini = efx_ef10_fini_nic,
4049 	.map_reset_reason = efx_ef10_map_reset_reason,
4050 	.map_reset_flags = efx_ef10_map_reset_flags,
4051 	.reset = efx_ef10_reset,
4052 	.probe_port = efx_mcdi_port_probe,
4053 	.remove_port = efx_mcdi_port_remove,
4054 	.fini_dmaq = efx_fini_dmaq,
4055 	.prepare_flr = efx_ef10_prepare_flr,
4056 	.finish_flr = efx_port_dummy_op_void,
4057 	.describe_stats = efx_ef10_describe_stats,
4058 	.update_stats = efx_ef10_update_stats_vf,
4059 	.update_stats_atomic = efx_ef10_update_stats_atomic_vf,
4060 	.start_stats = efx_port_dummy_op_void,
4061 	.pull_stats = efx_port_dummy_op_void,
4062 	.stop_stats = efx_port_dummy_op_void,
4063 	.push_irq_moderation = efx_ef10_push_irq_moderation,
4064 	.reconfigure_mac = efx_ef10_mac_reconfigure,
4065 	.check_mac_fault = efx_mcdi_mac_check_fault,
4066 	.reconfigure_port = efx_mcdi_port_reconfigure,
4067 	.get_wol = efx_ef10_get_wol_vf,
4068 	.set_wol = efx_ef10_set_wol_vf,
4069 	.resume_wol = efx_port_dummy_op_void,
4070 	.mcdi_request = efx_ef10_mcdi_request,
4071 	.mcdi_poll_response = efx_ef10_mcdi_poll_response,
4072 	.mcdi_read_response = efx_ef10_mcdi_read_response,
4073 	.mcdi_poll_reboot = efx_ef10_mcdi_poll_reboot,
4074 	.mcdi_reboot_detected = efx_ef10_mcdi_reboot_detected,
4075 	.irq_enable_master = efx_port_dummy_op_void,
4076 	.irq_test_generate = efx_ef10_irq_test_generate,
4077 	.irq_disable_non_ev = efx_port_dummy_op_void,
4078 	.irq_handle_msi = efx_ef10_msi_interrupt,
4079 	.irq_handle_legacy = efx_ef10_legacy_interrupt,
4080 	.tx_probe = efx_ef10_tx_probe,
4081 	.tx_init = efx_ef10_tx_init,
4082 	.tx_remove = efx_mcdi_tx_remove,
4083 	.tx_write = efx_ef10_tx_write,
4084 	.tx_limit_len = efx_ef10_tx_limit_len,
4085 	.tx_enqueue = __efx_enqueue_skb,
4086 	.rx_push_rss_config = efx_mcdi_vf_rx_push_rss_config,
4087 	.rx_pull_rss_config = efx_mcdi_rx_pull_rss_config,
4088 	.rx_probe = efx_mcdi_rx_probe,
4089 	.rx_init = efx_mcdi_rx_init,
4090 	.rx_remove = efx_mcdi_rx_remove,
4091 	.rx_write = efx_ef10_rx_write,
4092 	.rx_defer_refill = efx_ef10_rx_defer_refill,
4093 	.rx_packet = __efx_rx_packet,
4094 	.ev_probe = efx_mcdi_ev_probe,
4095 	.ev_init = efx_ef10_ev_init,
4096 	.ev_fini = efx_mcdi_ev_fini,
4097 	.ev_remove = efx_mcdi_ev_remove,
4098 	.ev_process = efx_ef10_ev_process,
4099 	.ev_read_ack = efx_ef10_ev_read_ack,
4100 	.ev_test_generate = efx_ef10_ev_test_generate,
4101 	.filter_table_probe = efx_ef10_filter_table_probe,
4102 	.filter_table_restore = efx_mcdi_filter_table_restore,
4103 	.filter_table_remove = efx_ef10_filter_table_remove,
4104 	.filter_update_rx_scatter = efx_mcdi_update_rx_scatter,
4105 	.filter_insert = efx_mcdi_filter_insert,
4106 	.filter_remove_safe = efx_mcdi_filter_remove_safe,
4107 	.filter_get_safe = efx_mcdi_filter_get_safe,
4108 	.filter_clear_rx = efx_mcdi_filter_clear_rx,
4109 	.filter_count_rx_used = efx_mcdi_filter_count_rx_used,
4110 	.filter_get_rx_id_limit = efx_mcdi_filter_get_rx_id_limit,
4111 	.filter_get_rx_ids = efx_mcdi_filter_get_rx_ids,
4112 #ifdef CONFIG_RFS_ACCEL
4113 	.filter_rfs_expire_one = efx_mcdi_filter_rfs_expire_one,
4114 #endif
4115 #ifdef CONFIG_SFC_MTD
4116 	.mtd_probe = efx_port_dummy_op_int,
4117 #endif
4118 	.ptp_write_host_time = efx_ef10_ptp_write_host_time_vf,
4119 	.ptp_set_ts_config = efx_ef10_ptp_set_ts_config_vf,
4120 	.vlan_rx_add_vid = efx_ef10_vlan_rx_add_vid,
4121 	.vlan_rx_kill_vid = efx_ef10_vlan_rx_kill_vid,
4122 #ifdef CONFIG_SFC_SRIOV
4123 	.vswitching_probe = efx_ef10_vswitching_probe_vf,
4124 	.vswitching_restore = efx_ef10_vswitching_restore_vf,
4125 	.vswitching_remove = efx_ef10_vswitching_remove_vf,
4126 #endif
4127 	.get_mac_address = efx_ef10_get_mac_address_vf,
4128 	.set_mac_address = efx_ef10_set_mac_address,
4129 
4130 	.get_phys_port_id = efx_ef10_get_phys_port_id,
4131 	.revision = EFX_REV_HUNT_A0,
4132 	.max_dma_mask = DMA_BIT_MASK(ESF_DZ_TX_KER_BUF_ADDR_WIDTH),
4133 	.rx_prefix_size = ES_DZ_RX_PREFIX_SIZE,
4134 	.rx_hash_offset = ES_DZ_RX_PREFIX_HASH_OFST,
4135 	.rx_ts_offset = ES_DZ_RX_PREFIX_TSTAMP_OFST,
4136 	.can_rx_scatter = true,
4137 	.always_rx_scatter = true,
4138 	.min_interrupt_mode = EFX_INT_MODE_MSIX,
4139 	.timer_period_max = 1 << ERF_DD_EVQ_IND_TIMER_VAL_WIDTH,
4140 	.offload_features = EF10_OFFLOAD_FEATURES,
4141 	.mcdi_max_ver = 2,
4142 	.max_rx_ip_filters = EFX_MCDI_FILTER_TBL_ROWS,
4143 	.hwtstamp_filters = 1 << HWTSTAMP_FILTER_NONE |
4144 			    1 << HWTSTAMP_FILTER_ALL,
4145 	.rx_hash_key_size = 40,
4146 	.check_caps = ef10_check_caps,
4147 	.print_additional_fwver = efx_ef10_print_additional_fwver,
4148 	.sensor_event = efx_mcdi_sensor_event,
4149 	.rx_recycle_ring_size = efx_ef10_recycle_ring_size,
4150 };
4151 
4152 const struct efx_nic_type efx_hunt_a0_nic_type = {
4153 	.is_vf = false,
4154 	.mem_bar = efx_ef10_pf_mem_bar,
4155 	.mem_map_size = efx_ef10_mem_map_size,
4156 	.probe = efx_ef10_probe_pf,
4157 	.remove = efx_ef10_remove,
4158 	.dimension_resources = efx_ef10_dimension_resources,
4159 	.init = efx_ef10_init_nic,
4160 	.fini = efx_ef10_fini_nic,
4161 	.map_reset_reason = efx_ef10_map_reset_reason,
4162 	.map_reset_flags = efx_ef10_map_reset_flags,
4163 	.reset = efx_ef10_reset,
4164 	.probe_port = efx_mcdi_port_probe,
4165 	.remove_port = efx_mcdi_port_remove,
4166 	.fini_dmaq = efx_fini_dmaq,
4167 	.prepare_flr = efx_ef10_prepare_flr,
4168 	.finish_flr = efx_port_dummy_op_void,
4169 	.describe_stats = efx_ef10_describe_stats,
4170 	.update_stats = efx_ef10_update_stats_pf,
4171 	.start_stats = efx_mcdi_mac_start_stats,
4172 	.pull_stats = efx_mcdi_mac_pull_stats,
4173 	.stop_stats = efx_mcdi_mac_stop_stats,
4174 	.push_irq_moderation = efx_ef10_push_irq_moderation,
4175 	.reconfigure_mac = efx_ef10_mac_reconfigure,
4176 	.check_mac_fault = efx_mcdi_mac_check_fault,
4177 	.reconfigure_port = efx_mcdi_port_reconfigure,
4178 	.get_wol = efx_ef10_get_wol,
4179 	.set_wol = efx_ef10_set_wol,
4180 	.resume_wol = efx_port_dummy_op_void,
4181 	.get_fec_stats = efx_ef10_get_fec_stats,
4182 	.test_chip = efx_ef10_test_chip,
4183 	.test_nvram = efx_mcdi_nvram_test_all,
4184 	.mcdi_request = efx_ef10_mcdi_request,
4185 	.mcdi_poll_response = efx_ef10_mcdi_poll_response,
4186 	.mcdi_read_response = efx_ef10_mcdi_read_response,
4187 	.mcdi_poll_reboot = efx_ef10_mcdi_poll_reboot,
4188 	.mcdi_reboot_detected = efx_ef10_mcdi_reboot_detected,
4189 	.irq_enable_master = efx_port_dummy_op_void,
4190 	.irq_test_generate = efx_ef10_irq_test_generate,
4191 	.irq_disable_non_ev = efx_port_dummy_op_void,
4192 	.irq_handle_msi = efx_ef10_msi_interrupt,
4193 	.irq_handle_legacy = efx_ef10_legacy_interrupt,
4194 	.tx_probe = efx_ef10_tx_probe,
4195 	.tx_init = efx_ef10_tx_init,
4196 	.tx_remove = efx_mcdi_tx_remove,
4197 	.tx_write = efx_ef10_tx_write,
4198 	.tx_limit_len = efx_ef10_tx_limit_len,
4199 	.tx_enqueue = __efx_enqueue_skb,
4200 	.rx_push_rss_config = efx_mcdi_pf_rx_push_rss_config,
4201 	.rx_pull_rss_config = efx_mcdi_rx_pull_rss_config,
4202 	.rx_push_rss_context_config = efx_mcdi_rx_push_rss_context_config,
4203 	.rx_pull_rss_context_config = efx_mcdi_rx_pull_rss_context_config,
4204 	.rx_restore_rss_contexts = efx_mcdi_rx_restore_rss_contexts,
4205 	.rx_probe = efx_mcdi_rx_probe,
4206 	.rx_init = efx_mcdi_rx_init,
4207 	.rx_remove = efx_mcdi_rx_remove,
4208 	.rx_write = efx_ef10_rx_write,
4209 	.rx_defer_refill = efx_ef10_rx_defer_refill,
4210 	.rx_packet = __efx_rx_packet,
4211 	.ev_probe = efx_mcdi_ev_probe,
4212 	.ev_init = efx_ef10_ev_init,
4213 	.ev_fini = efx_mcdi_ev_fini,
4214 	.ev_remove = efx_mcdi_ev_remove,
4215 	.ev_process = efx_ef10_ev_process,
4216 	.ev_read_ack = efx_ef10_ev_read_ack,
4217 	.ev_test_generate = efx_ef10_ev_test_generate,
4218 	.filter_table_probe = efx_ef10_filter_table_probe,
4219 	.filter_table_restore = efx_mcdi_filter_table_restore,
4220 	.filter_table_remove = efx_ef10_filter_table_remove,
4221 	.filter_update_rx_scatter = efx_mcdi_update_rx_scatter,
4222 	.filter_insert = efx_mcdi_filter_insert,
4223 	.filter_remove_safe = efx_mcdi_filter_remove_safe,
4224 	.filter_get_safe = efx_mcdi_filter_get_safe,
4225 	.filter_clear_rx = efx_mcdi_filter_clear_rx,
4226 	.filter_count_rx_used = efx_mcdi_filter_count_rx_used,
4227 	.filter_get_rx_id_limit = efx_mcdi_filter_get_rx_id_limit,
4228 	.filter_get_rx_ids = efx_mcdi_filter_get_rx_ids,
4229 #ifdef CONFIG_RFS_ACCEL
4230 	.filter_rfs_expire_one = efx_mcdi_filter_rfs_expire_one,
4231 #endif
4232 #ifdef CONFIG_SFC_MTD
4233 	.mtd_probe = efx_ef10_mtd_probe,
4234 	.mtd_rename = efx_mcdi_mtd_rename,
4235 	.mtd_read = efx_mcdi_mtd_read,
4236 	.mtd_erase = efx_mcdi_mtd_erase,
4237 	.mtd_write = efx_mcdi_mtd_write,
4238 	.mtd_sync = efx_mcdi_mtd_sync,
4239 #endif
4240 	.ptp_write_host_time = efx_ef10_ptp_write_host_time,
4241 	.ptp_set_ts_sync_events = efx_ef10_ptp_set_ts_sync_events,
4242 	.ptp_set_ts_config = efx_ef10_ptp_set_ts_config,
4243 	.vlan_rx_add_vid = efx_ef10_vlan_rx_add_vid,
4244 	.vlan_rx_kill_vid = efx_ef10_vlan_rx_kill_vid,
4245 	.udp_tnl_push_ports = efx_ef10_udp_tnl_push_ports,
4246 	.udp_tnl_has_port = efx_ef10_udp_tnl_has_port,
4247 #ifdef CONFIG_SFC_SRIOV
4248 	.sriov_configure = efx_ef10_sriov_configure,
4249 	.sriov_init = efx_ef10_sriov_init,
4250 	.sriov_fini = efx_ef10_sriov_fini,
4251 	.sriov_wanted = efx_ef10_sriov_wanted,
4252 	.sriov_reset = efx_ef10_sriov_reset,
4253 	.sriov_flr = efx_ef10_sriov_flr,
4254 	.sriov_set_vf_mac = efx_ef10_sriov_set_vf_mac,
4255 	.sriov_set_vf_vlan = efx_ef10_sriov_set_vf_vlan,
4256 	.sriov_set_vf_spoofchk = efx_ef10_sriov_set_vf_spoofchk,
4257 	.sriov_get_vf_config = efx_ef10_sriov_get_vf_config,
4258 	.sriov_set_vf_link_state = efx_ef10_sriov_set_vf_link_state,
4259 	.vswitching_probe = efx_ef10_vswitching_probe_pf,
4260 	.vswitching_restore = efx_ef10_vswitching_restore_pf,
4261 	.vswitching_remove = efx_ef10_vswitching_remove_pf,
4262 #endif
4263 	.get_mac_address = efx_ef10_get_mac_address_pf,
4264 	.set_mac_address = efx_ef10_set_mac_address,
4265 	.tso_versions = efx_ef10_tso_versions,
4266 
4267 	.get_phys_port_id = efx_ef10_get_phys_port_id,
4268 	.revision = EFX_REV_HUNT_A0,
4269 	.max_dma_mask = DMA_BIT_MASK(ESF_DZ_TX_KER_BUF_ADDR_WIDTH),
4270 	.rx_prefix_size = ES_DZ_RX_PREFIX_SIZE,
4271 	.rx_hash_offset = ES_DZ_RX_PREFIX_HASH_OFST,
4272 	.rx_ts_offset = ES_DZ_RX_PREFIX_TSTAMP_OFST,
4273 	.can_rx_scatter = true,
4274 	.always_rx_scatter = true,
4275 	.option_descriptors = true,
4276 	.min_interrupt_mode = EFX_INT_MODE_LEGACY,
4277 	.timer_period_max = 1 << ERF_DD_EVQ_IND_TIMER_VAL_WIDTH,
4278 	.offload_features = EF10_OFFLOAD_FEATURES,
4279 	.mcdi_max_ver = 2,
4280 	.max_rx_ip_filters = EFX_MCDI_FILTER_TBL_ROWS,
4281 	.hwtstamp_filters = 1 << HWTSTAMP_FILTER_NONE |
4282 			    1 << HWTSTAMP_FILTER_ALL,
4283 	.rx_hash_key_size = 40,
4284 	.check_caps = ef10_check_caps,
4285 	.print_additional_fwver = efx_ef10_print_additional_fwver,
4286 	.sensor_event = efx_mcdi_sensor_event,
4287 	.rx_recycle_ring_size = efx_ef10_recycle_ring_size,
4288 };
4289