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
31enum {
32	EFX_EF10_TEST = 1,
33	EFX_EF10_REFILL,
34};
35
36/* VLAN list entry */
37struct efx_ef10_vlan {
38	struct list_head list;
39	u16 vid;
40};
41
42static int efx_ef10_set_udp_tnl_ports(struct efx_nic *efx, bool unloading);
43static const struct udp_tunnel_nic_info efx_ef10_udp_tunnels;
44
45static 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 */
58static 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 */
69static unsigned int efx_ef10_vf_mem_bar(struct efx_nic *efx)
70{
71	return 0;
72}
73
74static 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
82static bool efx_ef10_is_vf(struct efx_nic *efx)
83{
84	return efx->type->is_vf;
85}
86
87#ifdef CONFIG_SFC_SRIOV
88static 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
107static 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
183static 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
202static 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
215static 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
261static 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
290static 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
325static 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
345static 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
373static ssize_t efx_ef10_show_link_control_flag(struct device *dev,
374					       struct device_attribute *attr,
375					       char *buf)
376{
377	struct efx_nic *efx = dev_get_drvdata(dev);
378
379	return sprintf(buf, "%d\n",
380		       ((efx->mcdi->fn_flags) &
381			(1 << MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_LINKCTRL))
382		       ? 1 : 0);
383}
384
385static ssize_t efx_ef10_show_primary_flag(struct device *dev,
386					  struct device_attribute *attr,
387					  char *buf)
388{
389	struct efx_nic *efx = dev_get_drvdata(dev);
390
391	return sprintf(buf, "%d\n",
392		       ((efx->mcdi->fn_flags) &
393			(1 << MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_PRIMARY))
394		       ? 1 : 0);
395}
396
397static 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
412static 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
452done_unlock:
453	mutex_unlock(&nic_data->vlan_lock);
454	return 0;
455
456fail_filter_add_vlan:
457	list_del(&vlan->list);
458	kfree(vlan);
459fail_alloc:
460fail_exist:
461	mutex_unlock(&nic_data->vlan_lock);
462	return rc;
463}
464
465static 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
482static 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
511static 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
522static DEVICE_ATTR(link_control_flag, 0444, efx_ef10_show_link_control_flag,
523		   NULL);
524static DEVICE_ATTR(primary_flag, 0444, efx_ef10_show_primary_flag, NULL);
525
526static int efx_ef10_probe(struct efx_nic *efx)
527{
528	struct efx_ef10_nic_data *nic_data;
529	int i, rc;
530
531	nic_data = kzalloc(sizeof(*nic_data), GFP_KERNEL);
532	if (!nic_data)
533		return -ENOMEM;
534	efx->nic_data = nic_data;
535
536	/* we assume later that we can copy from this buffer in dwords */
537	BUILD_BUG_ON(MCDI_CTL_SDU_LEN_MAX_V2 % 4);
538
539	rc = efx_nic_alloc_buffer(efx, &nic_data->mcdi_buf,
540				  8 + MCDI_CTL_SDU_LEN_MAX_V2, GFP_KERNEL);
541	if (rc)
542		goto fail1;
543
544	/* Get the MC's warm boot count.  In case it's rebooting right
545	 * now, be prepared to retry.
546	 */
547	i = 0;
548	for (;;) {
549		rc = efx_ef10_get_warm_boot_count(efx);
550		if (rc >= 0)
551			break;
552		if (++i == 5)
553			goto fail2;
554		ssleep(1);
555	}
556	nic_data->warm_boot_count = rc;
557
558	/* In case we're recovering from a crash (kexec), we want to
559	 * cancel any outstanding request by the previous user of this
560	 * function.  We send a special message using the least
561	 * significant bits of the 'high' (doorbell) register.
562	 */
563	_efx_writed(efx, cpu_to_le32(1), ER_DZ_MC_DB_HWRD);
564
565	rc = efx_mcdi_init(efx);
566	if (rc)
567		goto fail2;
568
569	mutex_init(&nic_data->udp_tunnels_lock);
570	for (i = 0; i < ARRAY_SIZE(nic_data->udp_tunnels); ++i)
571		nic_data->udp_tunnels[i].type =
572			TUNNEL_ENCAP_UDP_PORT_ENTRY_INVALID;
573
574	/* Reset (most) configuration for this function */
575	rc = efx_mcdi_reset(efx, RESET_TYPE_ALL);
576	if (rc)
577		goto fail3;
578
579	/* Enable event logging */
580	rc = efx_mcdi_log_ctrl(efx, true, false, 0);
581	if (rc)
582		goto fail3;
583
584	rc = device_create_file(&efx->pci_dev->dev,
585				&dev_attr_link_control_flag);
586	if (rc)
587		goto fail3;
588
589	rc = device_create_file(&efx->pci_dev->dev, &dev_attr_primary_flag);
590	if (rc)
591		goto fail4;
592
593	rc = efx_get_pf_index(efx, &nic_data->pf_index);
594	if (rc)
595		goto fail5;
596
597	rc = efx_ef10_init_datapath_caps(efx);
598	if (rc < 0)
599		goto fail5;
600
601	efx_ef10_read_licensed_features(efx);
602
603	/* We can have one VI for each vi_stride-byte region.
604	 * However, until we use TX option descriptors we need up to four
605	 * TX queues per channel for different checksumming combinations.
606	 */
607	if (nic_data->datapath_caps &
608	    (1 << MC_CMD_GET_CAPABILITIES_OUT_VXLAN_NVGRE_LBN))
609		efx->tx_queues_per_channel = 4;
610	else
611		efx->tx_queues_per_channel = 2;
612	efx->max_vis = efx_ef10_mem_map_size(efx) / efx->vi_stride;
613	if (!efx->max_vis) {
614		netif_err(efx, drv, efx->net_dev, "error determining max VIs\n");
615		rc = -EIO;
616		goto fail5;
617	}
618	efx->max_channels = min_t(unsigned int, EFX_MAX_CHANNELS,
619				  efx->max_vis / efx->tx_queues_per_channel);
620	efx->max_tx_channels = efx->max_channels;
621	if (WARN_ON(efx->max_channels == 0)) {
622		rc = -EIO;
623		goto fail5;
624	}
625
626	efx->rx_packet_len_offset =
627		ES_DZ_RX_PREFIX_PKTLEN_OFST - ES_DZ_RX_PREFIX_SIZE;
628
629	if (nic_data->datapath_caps &
630	    (1 << MC_CMD_GET_CAPABILITIES_OUT_RX_INCLUDE_FCS_LBN))
631		efx->net_dev->hw_features |= NETIF_F_RXFCS;
632
633	rc = efx_mcdi_port_get_number(efx);
634	if (rc < 0)
635		goto fail5;
636	efx->port_num = rc;
637
638	rc = efx->type->get_mac_address(efx, efx->net_dev->perm_addr);
639	if (rc)
640		goto fail5;
641
642	rc = efx_ef10_get_timer_config(efx);
643	if (rc < 0)
644		goto fail5;
645
646	rc = efx_mcdi_mon_probe(efx);
647	if (rc && rc != -EPERM)
648		goto fail5;
649
650	efx_ptp_defer_probe_with_channel(efx);
651
652#ifdef CONFIG_SFC_SRIOV
653	if ((efx->pci_dev->physfn) && (!efx->pci_dev->is_physfn)) {
654		struct pci_dev *pci_dev_pf = efx->pci_dev->physfn;
655		struct efx_nic *efx_pf = pci_get_drvdata(pci_dev_pf);
656
657		efx_pf->type->get_mac_address(efx_pf, nic_data->port_id);
658	} else
659#endif
660		ether_addr_copy(nic_data->port_id, efx->net_dev->perm_addr);
661
662	INIT_LIST_HEAD(&nic_data->vlan_list);
663	mutex_init(&nic_data->vlan_lock);
664
665	/* Add unspecified VID to support VLAN filtering being disabled */
666	rc = efx_ef10_add_vlan(efx, EFX_FILTER_VID_UNSPEC);
667	if (rc)
668		goto fail_add_vid_unspec;
669
670	/* If VLAN filtering is enabled, we need VID 0 to get untagged
671	 * traffic.  It is added automatically if 8021q module is loaded,
672	 * but we can't rely on it since module may be not loaded.
673	 */
674	rc = efx_ef10_add_vlan(efx, 0);
675	if (rc)
676		goto fail_add_vid_0;
677
678	if (nic_data->datapath_caps &
679	    (1 << MC_CMD_GET_CAPABILITIES_OUT_VXLAN_NVGRE_LBN) &&
680	    efx->mcdi->fn_flags &
681	    (1 << MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_TRUSTED))
682		efx->net_dev->udp_tunnel_nic_info = &efx_ef10_udp_tunnels;
683
684	return 0;
685
686fail_add_vid_0:
687	efx_ef10_cleanup_vlans(efx);
688fail_add_vid_unspec:
689	mutex_destroy(&nic_data->vlan_lock);
690	efx_ptp_remove(efx);
691	efx_mcdi_mon_remove(efx);
692fail5:
693	device_remove_file(&efx->pci_dev->dev, &dev_attr_primary_flag);
694fail4:
695	device_remove_file(&efx->pci_dev->dev, &dev_attr_link_control_flag);
696fail3:
697	efx_mcdi_detach(efx);
698
699	mutex_lock(&nic_data->udp_tunnels_lock);
700	memset(nic_data->udp_tunnels, 0, sizeof(nic_data->udp_tunnels));
701	(void)efx_ef10_set_udp_tnl_ports(efx, true);
702	mutex_unlock(&nic_data->udp_tunnels_lock);
703	mutex_destroy(&nic_data->udp_tunnels_lock);
704
705	efx_mcdi_fini(efx);
706fail2:
707	efx_nic_free_buffer(efx, &nic_data->mcdi_buf);
708fail1:
709	kfree(nic_data);
710	efx->nic_data = NULL;
711	return rc;
712}
713
714#ifdef EFX_USE_PIO
715
716static void efx_ef10_free_piobufs(struct efx_nic *efx)
717{
718	struct efx_ef10_nic_data *nic_data = efx->nic_data;
719	MCDI_DECLARE_BUF(inbuf, MC_CMD_FREE_PIOBUF_IN_LEN);
720	unsigned int i;
721	int rc;
722
723	BUILD_BUG_ON(MC_CMD_FREE_PIOBUF_OUT_LEN != 0);
724
725	for (i = 0; i < nic_data->n_piobufs; i++) {
726		MCDI_SET_DWORD(inbuf, FREE_PIOBUF_IN_PIOBUF_HANDLE,
727			       nic_data->piobuf_handle[i]);
728		rc = efx_mcdi_rpc(efx, MC_CMD_FREE_PIOBUF, inbuf, sizeof(inbuf),
729				  NULL, 0, NULL);
730		WARN_ON(rc);
731	}
732
733	nic_data->n_piobufs = 0;
734}
735
736static int efx_ef10_alloc_piobufs(struct efx_nic *efx, unsigned int n)
737{
738	struct efx_ef10_nic_data *nic_data = efx->nic_data;
739	MCDI_DECLARE_BUF(outbuf, MC_CMD_ALLOC_PIOBUF_OUT_LEN);
740	unsigned int i;
741	size_t outlen;
742	int rc = 0;
743
744	BUILD_BUG_ON(MC_CMD_ALLOC_PIOBUF_IN_LEN != 0);
745
746	for (i = 0; i < n; i++) {
747		rc = efx_mcdi_rpc_quiet(efx, MC_CMD_ALLOC_PIOBUF, NULL, 0,
748					outbuf, sizeof(outbuf), &outlen);
749		if (rc) {
750			/* Don't display the MC error if we didn't have space
751			 * for a VF.
752			 */
753			if (!(efx_ef10_is_vf(efx) && rc == -ENOSPC))
754				efx_mcdi_display_error(efx, MC_CMD_ALLOC_PIOBUF,
755						       0, outbuf, outlen, rc);
756			break;
757		}
758		if (outlen < MC_CMD_ALLOC_PIOBUF_OUT_LEN) {
759			rc = -EIO;
760			break;
761		}
762		nic_data->piobuf_handle[i] =
763			MCDI_DWORD(outbuf, ALLOC_PIOBUF_OUT_PIOBUF_HANDLE);
764		netif_dbg(efx, probe, efx->net_dev,
765			  "allocated PIO buffer %u handle %x\n", i,
766			  nic_data->piobuf_handle[i]);
767	}
768
769	nic_data->n_piobufs = i;
770	if (rc)
771		efx_ef10_free_piobufs(efx);
772	return rc;
773}
774
775static int efx_ef10_link_piobufs(struct efx_nic *efx)
776{
777	struct efx_ef10_nic_data *nic_data = efx->nic_data;
778	MCDI_DECLARE_BUF(inbuf, MC_CMD_LINK_PIOBUF_IN_LEN);
779	struct efx_channel *channel;
780	struct efx_tx_queue *tx_queue;
781	unsigned int offset, index;
782	int rc;
783
784	BUILD_BUG_ON(MC_CMD_LINK_PIOBUF_OUT_LEN != 0);
785	BUILD_BUG_ON(MC_CMD_UNLINK_PIOBUF_OUT_LEN != 0);
786
787	/* Link a buffer to each VI in the write-combining mapping */
788	for (index = 0; index < nic_data->n_piobufs; ++index) {
789		MCDI_SET_DWORD(inbuf, LINK_PIOBUF_IN_PIOBUF_HANDLE,
790			       nic_data->piobuf_handle[index]);
791		MCDI_SET_DWORD(inbuf, LINK_PIOBUF_IN_TXQ_INSTANCE,
792			       nic_data->pio_write_vi_base + index);
793		rc = efx_mcdi_rpc(efx, MC_CMD_LINK_PIOBUF,
794				  inbuf, MC_CMD_LINK_PIOBUF_IN_LEN,
795				  NULL, 0, NULL);
796		if (rc) {
797			netif_err(efx, drv, efx->net_dev,
798				  "failed to link VI %u to PIO buffer %u (%d)\n",
799				  nic_data->pio_write_vi_base + index, index,
800				  rc);
801			goto fail;
802		}
803		netif_dbg(efx, probe, efx->net_dev,
804			  "linked VI %u to PIO buffer %u\n",
805			  nic_data->pio_write_vi_base + index, index);
806	}
807
808	/* Link a buffer to each TX queue */
809	efx_for_each_channel(channel, efx) {
810		/* Extra channels, even those with TXQs (PTP), do not require
811		 * PIO resources.
812		 */
813		if (!channel->type->want_pio ||
814		    channel->channel >= efx->xdp_channel_offset)
815			continue;
816
817		efx_for_each_channel_tx_queue(tx_queue, channel) {
818			/* We assign the PIO buffers to queues in
819			 * reverse order to allow for the following
820			 * special case.
821			 */
822			offset = ((efx->tx_channel_offset + efx->n_tx_channels -
823				   tx_queue->channel->channel - 1) *
824				  efx_piobuf_size);
825			index = offset / nic_data->piobuf_size;
826			offset = offset % nic_data->piobuf_size;
827
828			/* When the host page size is 4K, the first
829			 * host page in the WC mapping may be within
830			 * the same VI page as the last TX queue.  We
831			 * can only link one buffer to each VI.
832			 */
833			if (tx_queue->queue == nic_data->pio_write_vi_base) {
834				BUG_ON(index != 0);
835				rc = 0;
836			} else {
837				MCDI_SET_DWORD(inbuf,
838					       LINK_PIOBUF_IN_PIOBUF_HANDLE,
839					       nic_data->piobuf_handle[index]);
840				MCDI_SET_DWORD(inbuf,
841					       LINK_PIOBUF_IN_TXQ_INSTANCE,
842					       tx_queue->queue);
843				rc = efx_mcdi_rpc(efx, MC_CMD_LINK_PIOBUF,
844						  inbuf, MC_CMD_LINK_PIOBUF_IN_LEN,
845						  NULL, 0, NULL);
846			}
847
848			if (rc) {
849				/* This is non-fatal; the TX path just
850				 * won't use PIO for this queue
851				 */
852				netif_err(efx, drv, efx->net_dev,
853					  "failed to link VI %u to PIO buffer %u (%d)\n",
854					  tx_queue->queue, index, rc);
855				tx_queue->piobuf = NULL;
856			} else {
857				tx_queue->piobuf =
858					nic_data->pio_write_base +
859					index * efx->vi_stride + offset;
860				tx_queue->piobuf_offset = offset;
861				netif_dbg(efx, probe, efx->net_dev,
862					  "linked VI %u to PIO buffer %u offset %x addr %p\n",
863					  tx_queue->queue, index,
864					  tx_queue->piobuf_offset,
865					  tx_queue->piobuf);
866			}
867		}
868	}
869
870	return 0;
871
872fail:
873	/* inbuf was defined for MC_CMD_LINK_PIOBUF.  We can use the same
874	 * buffer for MC_CMD_UNLINK_PIOBUF because it's shorter.
875	 */
876	BUILD_BUG_ON(MC_CMD_LINK_PIOBUF_IN_LEN < MC_CMD_UNLINK_PIOBUF_IN_LEN);
877	while (index--) {
878		MCDI_SET_DWORD(inbuf, UNLINK_PIOBUF_IN_TXQ_INSTANCE,
879			       nic_data->pio_write_vi_base + index);
880		efx_mcdi_rpc(efx, MC_CMD_UNLINK_PIOBUF,
881			     inbuf, MC_CMD_UNLINK_PIOBUF_IN_LEN,
882			     NULL, 0, NULL);
883	}
884	return rc;
885}
886
887static void efx_ef10_forget_old_piobufs(struct efx_nic *efx)
888{
889	struct efx_channel *channel;
890	struct efx_tx_queue *tx_queue;
891
892	/* All our existing PIO buffers went away */
893	efx_for_each_channel(channel, efx)
894		efx_for_each_channel_tx_queue(tx_queue, channel)
895			tx_queue->piobuf = NULL;
896}
897
898#else /* !EFX_USE_PIO */
899
900static int efx_ef10_alloc_piobufs(struct efx_nic *efx, unsigned int n)
901{
902	return n == 0 ? 0 : -ENOBUFS;
903}
904
905static int efx_ef10_link_piobufs(struct efx_nic *efx)
906{
907	return 0;
908}
909
910static void efx_ef10_free_piobufs(struct efx_nic *efx)
911{
912}
913
914static void efx_ef10_forget_old_piobufs(struct efx_nic *efx)
915{
916}
917
918#endif /* EFX_USE_PIO */
919
920static void efx_ef10_remove(struct efx_nic *efx)
921{
922	struct efx_ef10_nic_data *nic_data = efx->nic_data;
923	int rc;
924
925#ifdef CONFIG_SFC_SRIOV
926	struct efx_ef10_nic_data *nic_data_pf;
927	struct pci_dev *pci_dev_pf;
928	struct efx_nic *efx_pf;
929	struct ef10_vf *vf;
930
931	if (efx->pci_dev->is_virtfn) {
932		pci_dev_pf = efx->pci_dev->physfn;
933		if (pci_dev_pf) {
934			efx_pf = pci_get_drvdata(pci_dev_pf);
935			nic_data_pf = efx_pf->nic_data;
936			vf = nic_data_pf->vf + nic_data->vf_index;
937			vf->efx = NULL;
938		} else
939			netif_info(efx, drv, efx->net_dev,
940				   "Could not get the PF id from VF\n");
941	}
942#endif
943
944	efx_ef10_cleanup_vlans(efx);
945	mutex_destroy(&nic_data->vlan_lock);
946
947	efx_ptp_remove(efx);
948
949	efx_mcdi_mon_remove(efx);
950
951	efx_mcdi_rx_free_indir_table(efx);
952
953	if (nic_data->wc_membase)
954		iounmap(nic_data->wc_membase);
955
956	rc = efx_mcdi_free_vis(efx);
957	WARN_ON(rc != 0);
958
959	if (!nic_data->must_restore_piobufs)
960		efx_ef10_free_piobufs(efx);
961
962	device_remove_file(&efx->pci_dev->dev, &dev_attr_primary_flag);
963	device_remove_file(&efx->pci_dev->dev, &dev_attr_link_control_flag);
964
965	efx_mcdi_detach(efx);
966
967	memset(nic_data->udp_tunnels, 0, sizeof(nic_data->udp_tunnels));
968	mutex_lock(&nic_data->udp_tunnels_lock);
969	(void)efx_ef10_set_udp_tnl_ports(efx, true);
970	mutex_unlock(&nic_data->udp_tunnels_lock);
971
972	mutex_destroy(&nic_data->udp_tunnels_lock);
973
974	efx_mcdi_fini(efx);
975	efx_nic_free_buffer(efx, &nic_data->mcdi_buf);
976	kfree(nic_data);
977}
978
979static int efx_ef10_probe_pf(struct efx_nic *efx)
980{
981	return efx_ef10_probe(efx);
982}
983
984int efx_ef10_vadaptor_query(struct efx_nic *efx, unsigned int port_id,
985			    u32 *port_flags, u32 *vadaptor_flags,
986			    unsigned int *vlan_tags)
987{
988	struct efx_ef10_nic_data *nic_data = efx->nic_data;
989	MCDI_DECLARE_BUF(inbuf, MC_CMD_VADAPTOR_QUERY_IN_LEN);
990	MCDI_DECLARE_BUF(outbuf, MC_CMD_VADAPTOR_QUERY_OUT_LEN);
991	size_t outlen;
992	int rc;
993
994	if (nic_data->datapath_caps &
995	    (1 << MC_CMD_GET_CAPABILITIES_OUT_VADAPTOR_QUERY_LBN)) {
996		MCDI_SET_DWORD(inbuf, VADAPTOR_QUERY_IN_UPSTREAM_PORT_ID,
997			       port_id);
998
999		rc = efx_mcdi_rpc(efx, MC_CMD_VADAPTOR_QUERY, inbuf, sizeof(inbuf),
1000				  outbuf, sizeof(outbuf), &outlen);
1001		if (rc)
1002			return rc;
1003
1004		if (outlen < sizeof(outbuf)) {
1005			rc = -EIO;
1006			return rc;
1007		}
1008	}
1009
1010	if (port_flags)
1011		*port_flags = MCDI_DWORD(outbuf, VADAPTOR_QUERY_OUT_PORT_FLAGS);
1012	if (vadaptor_flags)
1013		*vadaptor_flags =
1014			MCDI_DWORD(outbuf, VADAPTOR_QUERY_OUT_VADAPTOR_FLAGS);
1015	if (vlan_tags)
1016		*vlan_tags =
1017			MCDI_DWORD(outbuf,
1018				   VADAPTOR_QUERY_OUT_NUM_AVAILABLE_VLAN_TAGS);
1019
1020	return 0;
1021}
1022
1023int efx_ef10_vadaptor_alloc(struct efx_nic *efx, unsigned int port_id)
1024{
1025	MCDI_DECLARE_BUF(inbuf, MC_CMD_VADAPTOR_ALLOC_IN_LEN);
1026
1027	MCDI_SET_DWORD(inbuf, VADAPTOR_ALLOC_IN_UPSTREAM_PORT_ID, port_id);
1028	return efx_mcdi_rpc(efx, MC_CMD_VADAPTOR_ALLOC, inbuf, sizeof(inbuf),
1029			    NULL, 0, NULL);
1030}
1031
1032int efx_ef10_vadaptor_free(struct efx_nic *efx, unsigned int port_id)
1033{
1034	MCDI_DECLARE_BUF(inbuf, MC_CMD_VADAPTOR_FREE_IN_LEN);
1035
1036	MCDI_SET_DWORD(inbuf, VADAPTOR_FREE_IN_UPSTREAM_PORT_ID, port_id);
1037	return efx_mcdi_rpc(efx, MC_CMD_VADAPTOR_FREE, inbuf, sizeof(inbuf),
1038			    NULL, 0, NULL);
1039}
1040
1041int efx_ef10_vport_add_mac(struct efx_nic *efx,
1042			   unsigned int port_id, u8 *mac)
1043{
1044	MCDI_DECLARE_BUF(inbuf, MC_CMD_VPORT_ADD_MAC_ADDRESS_IN_LEN);
1045
1046	MCDI_SET_DWORD(inbuf, VPORT_ADD_MAC_ADDRESS_IN_VPORT_ID, port_id);
1047	ether_addr_copy(MCDI_PTR(inbuf, VPORT_ADD_MAC_ADDRESS_IN_MACADDR), mac);
1048
1049	return efx_mcdi_rpc(efx, MC_CMD_VPORT_ADD_MAC_ADDRESS, inbuf,
1050			    sizeof(inbuf), NULL, 0, NULL);
1051}
1052
1053int efx_ef10_vport_del_mac(struct efx_nic *efx,
1054			   unsigned int port_id, u8 *mac)
1055{
1056	MCDI_DECLARE_BUF(inbuf, MC_CMD_VPORT_DEL_MAC_ADDRESS_IN_LEN);
1057
1058	MCDI_SET_DWORD(inbuf, VPORT_DEL_MAC_ADDRESS_IN_VPORT_ID, port_id);
1059	ether_addr_copy(MCDI_PTR(inbuf, VPORT_DEL_MAC_ADDRESS_IN_MACADDR), mac);
1060
1061	return efx_mcdi_rpc(efx, MC_CMD_VPORT_DEL_MAC_ADDRESS, inbuf,
1062			    sizeof(inbuf), NULL, 0, NULL);
1063}
1064
1065#ifdef CONFIG_SFC_SRIOV
1066static int efx_ef10_probe_vf(struct efx_nic *efx)
1067{
1068	int rc;
1069	struct pci_dev *pci_dev_pf;
1070
1071	/* If the parent PF has no VF data structure, it doesn't know about this
1072	 * VF so fail probe.  The VF needs to be re-created.  This can happen
1073	 * if the PF driver is unloaded while the VF is assigned to a guest.
1074	 */
1075	pci_dev_pf = efx->pci_dev->physfn;
1076	if (pci_dev_pf) {
1077		struct efx_nic *efx_pf = pci_get_drvdata(pci_dev_pf);
1078		struct efx_ef10_nic_data *nic_data_pf = efx_pf->nic_data;
1079
1080		if (!nic_data_pf->vf) {
1081			netif_info(efx, drv, efx->net_dev,
1082				   "The VF cannot link to its parent PF; "
1083				   "please destroy and re-create the VF\n");
1084			return -EBUSY;
1085		}
1086	}
1087
1088	rc = efx_ef10_probe(efx);
1089	if (rc)
1090		return rc;
1091
1092	rc = efx_ef10_get_vf_index(efx);
1093	if (rc)
1094		goto fail;
1095
1096	if (efx->pci_dev->is_virtfn) {
1097		if (efx->pci_dev->physfn) {
1098			struct efx_nic *efx_pf =
1099				pci_get_drvdata(efx->pci_dev->physfn);
1100			struct efx_ef10_nic_data *nic_data_p = efx_pf->nic_data;
1101			struct efx_ef10_nic_data *nic_data = efx->nic_data;
1102
1103			nic_data_p->vf[nic_data->vf_index].efx = efx;
1104			nic_data_p->vf[nic_data->vf_index].pci_dev =
1105				efx->pci_dev;
1106		} else
1107			netif_info(efx, drv, efx->net_dev,
1108				   "Could not get the PF id from VF\n");
1109	}
1110
1111	return 0;
1112
1113fail:
1114	efx_ef10_remove(efx);
1115	return rc;
1116}
1117#else
1118static int efx_ef10_probe_vf(struct efx_nic *efx __attribute__ ((unused)))
1119{
1120	return 0;
1121}
1122#endif
1123
1124static 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 */
1136static 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
1296static void efx_ef10_fini_nic(struct efx_nic *efx)
1297{
1298	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1299
1300	spin_lock_bh(&efx->stats_lock);
1301	kfree(nic_data->mc_stats);
1302	nic_data->mc_stats = NULL;
1303	spin_unlock_bh(&efx->stats_lock);
1304}
1305
1306static int efx_ef10_init_nic(struct efx_nic *efx)
1307{
1308	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1309	struct net_device *net_dev = efx->net_dev;
1310	netdev_features_t tun_feats, tso_feats;
1311	int rc;
1312
1313	if (nic_data->must_check_datapath_caps) {
1314		rc = efx_ef10_init_datapath_caps(efx);
1315		if (rc)
1316			return rc;
1317		nic_data->must_check_datapath_caps = false;
1318	}
1319
1320	if (efx->must_realloc_vis) {
1321		/* We cannot let the number of VIs change now */
1322		rc = efx_ef10_alloc_vis(efx, nic_data->n_allocated_vis,
1323					nic_data->n_allocated_vis);
1324		if (rc)
1325			return rc;
1326		efx->must_realloc_vis = false;
1327	}
1328
1329	nic_data->mc_stats = kmalloc(efx->num_mac_stats * sizeof(__le64),
1330				     GFP_KERNEL);
1331	if (!nic_data->mc_stats)
1332		return -ENOMEM;
1333
1334	if (nic_data->must_restore_piobufs && nic_data->n_piobufs) {
1335		rc = efx_ef10_alloc_piobufs(efx, nic_data->n_piobufs);
1336		if (rc == 0) {
1337			rc = efx_ef10_link_piobufs(efx);
1338			if (rc)
1339				efx_ef10_free_piobufs(efx);
1340		}
1341
1342		/* Log an error on failure, but this is non-fatal.
1343		 * Permission errors are less important - we've presumably
1344		 * had the PIO buffer licence removed.
1345		 */
1346		if (rc == -EPERM)
1347			netif_dbg(efx, drv, efx->net_dev,
1348				  "not permitted to restore PIO buffers\n");
1349		else if (rc)
1350			netif_err(efx, drv, efx->net_dev,
1351				  "failed to restore PIO buffers (%d)\n", rc);
1352		nic_data->must_restore_piobufs = false;
1353	}
1354
1355	/* encap features might change during reset if fw variant changed */
1356	if (efx_has_cap(efx, VXLAN_NVGRE) && !efx_ef10_is_vf(efx))
1357		net_dev->hw_enc_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
1358	else
1359		net_dev->hw_enc_features &= ~(NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
1360
1361	tun_feats = NETIF_F_GSO_UDP_TUNNEL | NETIF_F_GSO_GRE |
1362		    NETIF_F_GSO_UDP_TUNNEL_CSUM | NETIF_F_GSO_GRE_CSUM;
1363	tso_feats = NETIF_F_TSO | NETIF_F_TSO6;
1364
1365	if (efx_has_cap(efx, TX_TSO_V2_ENCAP)) {
1366		/* If this is first nic_init, or if it is a reset and a new fw
1367		 * variant has added new features, enable them by default.
1368		 * If the features are not new, maintain their current value.
1369		 */
1370		if (!(net_dev->hw_features & tun_feats))
1371			net_dev->features |= tun_feats;
1372		net_dev->hw_enc_features |= tun_feats | tso_feats;
1373		net_dev->hw_features |= tun_feats;
1374	} else {
1375		net_dev->hw_enc_features &= ~(tun_feats | tso_feats);
1376		net_dev->hw_features &= ~tun_feats;
1377		net_dev->features &= ~tun_feats;
1378	}
1379
1380	/* don't fail init if RSS setup doesn't work */
1381	rc = efx->type->rx_push_rss_config(efx, false,
1382					   efx->rss_context.rx_indir_table, NULL);
1383
1384	return 0;
1385}
1386
1387static void efx_ef10_table_reset_mc_allocations(struct efx_nic *efx)
1388{
1389	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1390#ifdef CONFIG_SFC_SRIOV
1391	unsigned int i;
1392#endif
1393
1394	/* All our allocations have been reset */
1395	efx->must_realloc_vis = true;
1396	efx_mcdi_filter_table_reset_mc_allocations(efx);
1397	nic_data->must_restore_piobufs = true;
1398	efx_ef10_forget_old_piobufs(efx);
1399	efx->rss_context.context_id = EFX_MCDI_RSS_CONTEXT_INVALID;
1400
1401	/* Driver-created vswitches and vports must be re-created */
1402	nic_data->must_probe_vswitching = true;
1403	efx->vport_id = EVB_PORT_ID_ASSIGNED;
1404#ifdef CONFIG_SFC_SRIOV
1405	if (nic_data->vf)
1406		for (i = 0; i < efx->vf_count; i++)
1407			nic_data->vf[i].vport_id = 0;
1408#endif
1409}
1410
1411static enum reset_type efx_ef10_map_reset_reason(enum reset_type reason)
1412{
1413	if (reason == RESET_TYPE_MC_FAILURE)
1414		return RESET_TYPE_DATAPATH;
1415
1416	return efx_mcdi_map_reset_reason(reason);
1417}
1418
1419static int efx_ef10_map_reset_flags(u32 *flags)
1420{
1421	enum {
1422		EF10_RESET_PORT = ((ETH_RESET_MAC | ETH_RESET_PHY) <<
1423				   ETH_RESET_SHARED_SHIFT),
1424		EF10_RESET_MC = ((ETH_RESET_DMA | ETH_RESET_FILTER |
1425				  ETH_RESET_OFFLOAD | ETH_RESET_MAC |
1426				  ETH_RESET_PHY | ETH_RESET_MGMT) <<
1427				 ETH_RESET_SHARED_SHIFT)
1428	};
1429
1430	/* We assume for now that our PCI function is permitted to
1431	 * reset everything.
1432	 */
1433
1434	if ((*flags & EF10_RESET_MC) == EF10_RESET_MC) {
1435		*flags &= ~EF10_RESET_MC;
1436		return RESET_TYPE_WORLD;
1437	}
1438
1439	if ((*flags & EF10_RESET_PORT) == EF10_RESET_PORT) {
1440		*flags &= ~EF10_RESET_PORT;
1441		return RESET_TYPE_ALL;
1442	}
1443
1444	/* no invisible reset implemented */
1445
1446	return -EINVAL;
1447}
1448
1449static int efx_ef10_reset(struct efx_nic *efx, enum reset_type reset_type)
1450{
1451	int rc = efx_mcdi_reset(efx, reset_type);
1452
1453	/* Unprivileged functions return -EPERM, but need to return success
1454	 * here so that the datapath is brought back up.
1455	 */
1456	if (reset_type == RESET_TYPE_WORLD && rc == -EPERM)
1457		rc = 0;
1458
1459	/* If it was a port reset, trigger reallocation of MC resources.
1460	 * Note that on an MC reset nothing needs to be done now because we'll
1461	 * detect the MC reset later and handle it then.
1462	 * For an FLR, we never get an MC reset event, but the MC has reset all
1463	 * resources assigned to us, so we have to trigger reallocation now.
1464	 */
1465	if ((reset_type == RESET_TYPE_ALL ||
1466	     reset_type == RESET_TYPE_MCDI_TIMEOUT) && !rc)
1467		efx_ef10_table_reset_mc_allocations(efx);
1468	return rc;
1469}
1470
1471#define EF10_DMA_STAT(ext_name, mcdi_name)			\
1472	[EF10_STAT_ ## ext_name] =				\
1473	{ #ext_name, 64, 8 * MC_CMD_MAC_ ## mcdi_name }
1474#define EF10_DMA_INVIS_STAT(int_name, mcdi_name)		\
1475	[EF10_STAT_ ## int_name] =				\
1476	{ NULL, 64, 8 * MC_CMD_MAC_ ## mcdi_name }
1477#define EF10_OTHER_STAT(ext_name)				\
1478	[EF10_STAT_ ## ext_name] = { #ext_name, 0, 0 }
1479
1480static const struct efx_hw_stat_desc efx_ef10_stat_desc[EF10_STAT_COUNT] = {
1481	EF10_DMA_STAT(port_tx_bytes, TX_BYTES),
1482	EF10_DMA_STAT(port_tx_packets, TX_PKTS),
1483	EF10_DMA_STAT(port_tx_pause, TX_PAUSE_PKTS),
1484	EF10_DMA_STAT(port_tx_control, TX_CONTROL_PKTS),
1485	EF10_DMA_STAT(port_tx_unicast, TX_UNICAST_PKTS),
1486	EF10_DMA_STAT(port_tx_multicast, TX_MULTICAST_PKTS),
1487	EF10_DMA_STAT(port_tx_broadcast, TX_BROADCAST_PKTS),
1488	EF10_DMA_STAT(port_tx_lt64, TX_LT64_PKTS),
1489	EF10_DMA_STAT(port_tx_64, TX_64_PKTS),
1490	EF10_DMA_STAT(port_tx_65_to_127, TX_65_TO_127_PKTS),
1491	EF10_DMA_STAT(port_tx_128_to_255, TX_128_TO_255_PKTS),
1492	EF10_DMA_STAT(port_tx_256_to_511, TX_256_TO_511_PKTS),
1493	EF10_DMA_STAT(port_tx_512_to_1023, TX_512_TO_1023_PKTS),
1494	EF10_DMA_STAT(port_tx_1024_to_15xx, TX_1024_TO_15XX_PKTS),
1495	EF10_DMA_STAT(port_tx_15xx_to_jumbo, TX_15XX_TO_JUMBO_PKTS),
1496	EF10_DMA_STAT(port_rx_bytes, RX_BYTES),
1497	EF10_DMA_INVIS_STAT(port_rx_bytes_minus_good_bytes, RX_BAD_BYTES),
1498	EF10_OTHER_STAT(port_rx_good_bytes),
1499	EF10_OTHER_STAT(port_rx_bad_bytes),
1500	EF10_DMA_STAT(port_rx_packets, RX_PKTS),
1501	EF10_DMA_STAT(port_rx_good, RX_GOOD_PKTS),
1502	EF10_DMA_STAT(port_rx_bad, RX_BAD_FCS_PKTS),
1503	EF10_DMA_STAT(port_rx_pause, RX_PAUSE_PKTS),
1504	EF10_DMA_STAT(port_rx_control, RX_CONTROL_PKTS),
1505	EF10_DMA_STAT(port_rx_unicast, RX_UNICAST_PKTS),
1506	EF10_DMA_STAT(port_rx_multicast, RX_MULTICAST_PKTS),
1507	EF10_DMA_STAT(port_rx_broadcast, RX_BROADCAST_PKTS),
1508	EF10_DMA_STAT(port_rx_lt64, RX_UNDERSIZE_PKTS),
1509	EF10_DMA_STAT(port_rx_64, RX_64_PKTS),
1510	EF10_DMA_STAT(port_rx_65_to_127, RX_65_TO_127_PKTS),
1511	EF10_DMA_STAT(port_rx_128_to_255, RX_128_TO_255_PKTS),
1512	EF10_DMA_STAT(port_rx_256_to_511, RX_256_TO_511_PKTS),
1513	EF10_DMA_STAT(port_rx_512_to_1023, RX_512_TO_1023_PKTS),
1514	EF10_DMA_STAT(port_rx_1024_to_15xx, RX_1024_TO_15XX_PKTS),
1515	EF10_DMA_STAT(port_rx_15xx_to_jumbo, RX_15XX_TO_JUMBO_PKTS),
1516	EF10_DMA_STAT(port_rx_gtjumbo, RX_GTJUMBO_PKTS),
1517	EF10_DMA_STAT(port_rx_bad_gtjumbo, RX_JABBER_PKTS),
1518	EF10_DMA_STAT(port_rx_overflow, RX_OVERFLOW_PKTS),
1519	EF10_DMA_STAT(port_rx_align_error, RX_ALIGN_ERROR_PKTS),
1520	EF10_DMA_STAT(port_rx_length_error, RX_LENGTH_ERROR_PKTS),
1521	EF10_DMA_STAT(port_rx_nodesc_drops, RX_NODESC_DROPS),
1522	EFX_GENERIC_SW_STAT(rx_nodesc_trunc),
1523	EFX_GENERIC_SW_STAT(rx_noskb_drops),
1524	EF10_DMA_STAT(port_rx_pm_trunc_bb_overflow, PM_TRUNC_BB_OVERFLOW),
1525	EF10_DMA_STAT(port_rx_pm_discard_bb_overflow, PM_DISCARD_BB_OVERFLOW),
1526	EF10_DMA_STAT(port_rx_pm_trunc_vfifo_full, PM_TRUNC_VFIFO_FULL),
1527	EF10_DMA_STAT(port_rx_pm_discard_vfifo_full, PM_DISCARD_VFIFO_FULL),
1528	EF10_DMA_STAT(port_rx_pm_trunc_qbb, PM_TRUNC_QBB),
1529	EF10_DMA_STAT(port_rx_pm_discard_qbb, PM_DISCARD_QBB),
1530	EF10_DMA_STAT(port_rx_pm_discard_mapping, PM_DISCARD_MAPPING),
1531	EF10_DMA_STAT(port_rx_dp_q_disabled_packets, RXDP_Q_DISABLED_PKTS),
1532	EF10_DMA_STAT(port_rx_dp_di_dropped_packets, RXDP_DI_DROPPED_PKTS),
1533	EF10_DMA_STAT(port_rx_dp_streaming_packets, RXDP_STREAMING_PKTS),
1534	EF10_DMA_STAT(port_rx_dp_hlb_fetch, RXDP_HLB_FETCH_CONDITIONS),
1535	EF10_DMA_STAT(port_rx_dp_hlb_wait, RXDP_HLB_WAIT_CONDITIONS),
1536	EF10_DMA_STAT(rx_unicast, VADAPTER_RX_UNICAST_PACKETS),
1537	EF10_DMA_STAT(rx_unicast_bytes, VADAPTER_RX_UNICAST_BYTES),
1538	EF10_DMA_STAT(rx_multicast, VADAPTER_RX_MULTICAST_PACKETS),
1539	EF10_DMA_STAT(rx_multicast_bytes, VADAPTER_RX_MULTICAST_BYTES),
1540	EF10_DMA_STAT(rx_broadcast, VADAPTER_RX_BROADCAST_PACKETS),
1541	EF10_DMA_STAT(rx_broadcast_bytes, VADAPTER_RX_BROADCAST_BYTES),
1542	EF10_DMA_STAT(rx_bad, VADAPTER_RX_BAD_PACKETS),
1543	EF10_DMA_STAT(rx_bad_bytes, VADAPTER_RX_BAD_BYTES),
1544	EF10_DMA_STAT(rx_overflow, VADAPTER_RX_OVERFLOW),
1545	EF10_DMA_STAT(tx_unicast, VADAPTER_TX_UNICAST_PACKETS),
1546	EF10_DMA_STAT(tx_unicast_bytes, VADAPTER_TX_UNICAST_BYTES),
1547	EF10_DMA_STAT(tx_multicast, VADAPTER_TX_MULTICAST_PACKETS),
1548	EF10_DMA_STAT(tx_multicast_bytes, VADAPTER_TX_MULTICAST_BYTES),
1549	EF10_DMA_STAT(tx_broadcast, VADAPTER_TX_BROADCAST_PACKETS),
1550	EF10_DMA_STAT(tx_broadcast_bytes, VADAPTER_TX_BROADCAST_BYTES),
1551	EF10_DMA_STAT(tx_bad, VADAPTER_TX_BAD_PACKETS),
1552	EF10_DMA_STAT(tx_bad_bytes, VADAPTER_TX_BAD_BYTES),
1553	EF10_DMA_STAT(tx_overflow, VADAPTER_TX_OVERFLOW),
1554	EF10_DMA_STAT(fec_uncorrected_errors, FEC_UNCORRECTED_ERRORS),
1555	EF10_DMA_STAT(fec_corrected_errors, FEC_CORRECTED_ERRORS),
1556	EF10_DMA_STAT(fec_corrected_symbols_lane0, FEC_CORRECTED_SYMBOLS_LANE0),
1557	EF10_DMA_STAT(fec_corrected_symbols_lane1, FEC_CORRECTED_SYMBOLS_LANE1),
1558	EF10_DMA_STAT(fec_corrected_symbols_lane2, FEC_CORRECTED_SYMBOLS_LANE2),
1559	EF10_DMA_STAT(fec_corrected_symbols_lane3, FEC_CORRECTED_SYMBOLS_LANE3),
1560	EF10_DMA_STAT(ctpio_vi_busy_fallback, CTPIO_VI_BUSY_FALLBACK),
1561	EF10_DMA_STAT(ctpio_long_write_success, CTPIO_LONG_WRITE_SUCCESS),
1562	EF10_DMA_STAT(ctpio_missing_dbell_fail, CTPIO_MISSING_DBELL_FAIL),
1563	EF10_DMA_STAT(ctpio_overflow_fail, CTPIO_OVERFLOW_FAIL),
1564	EF10_DMA_STAT(ctpio_underflow_fail, CTPIO_UNDERFLOW_FAIL),
1565	EF10_DMA_STAT(ctpio_timeout_fail, CTPIO_TIMEOUT_FAIL),
1566	EF10_DMA_STAT(ctpio_noncontig_wr_fail, CTPIO_NONCONTIG_WR_FAIL),
1567	EF10_DMA_STAT(ctpio_frm_clobber_fail, CTPIO_FRM_CLOBBER_FAIL),
1568	EF10_DMA_STAT(ctpio_invalid_wr_fail, CTPIO_INVALID_WR_FAIL),
1569	EF10_DMA_STAT(ctpio_vi_clobber_fallback, CTPIO_VI_CLOBBER_FALLBACK),
1570	EF10_DMA_STAT(ctpio_unqualified_fallback, CTPIO_UNQUALIFIED_FALLBACK),
1571	EF10_DMA_STAT(ctpio_runt_fallback, CTPIO_RUNT_FALLBACK),
1572	EF10_DMA_STAT(ctpio_success, CTPIO_SUCCESS),
1573	EF10_DMA_STAT(ctpio_fallback, CTPIO_FALLBACK),
1574	EF10_DMA_STAT(ctpio_poison, CTPIO_POISON),
1575	EF10_DMA_STAT(ctpio_erase, CTPIO_ERASE),
1576};
1577
1578#define HUNT_COMMON_STAT_MASK ((1ULL << EF10_STAT_port_tx_bytes) |	\
1579			       (1ULL << EF10_STAT_port_tx_packets) |	\
1580			       (1ULL << EF10_STAT_port_tx_pause) |	\
1581			       (1ULL << EF10_STAT_port_tx_unicast) |	\
1582			       (1ULL << EF10_STAT_port_tx_multicast) |	\
1583			       (1ULL << EF10_STAT_port_tx_broadcast) |	\
1584			       (1ULL << EF10_STAT_port_rx_bytes) |	\
1585			       (1ULL <<                                 \
1586				EF10_STAT_port_rx_bytes_minus_good_bytes) | \
1587			       (1ULL << EF10_STAT_port_rx_good_bytes) |	\
1588			       (1ULL << EF10_STAT_port_rx_bad_bytes) |	\
1589			       (1ULL << EF10_STAT_port_rx_packets) |	\
1590			       (1ULL << EF10_STAT_port_rx_good) |	\
1591			       (1ULL << EF10_STAT_port_rx_bad) |	\
1592			       (1ULL << EF10_STAT_port_rx_pause) |	\
1593			       (1ULL << EF10_STAT_port_rx_control) |	\
1594			       (1ULL << EF10_STAT_port_rx_unicast) |	\
1595			       (1ULL << EF10_STAT_port_rx_multicast) |	\
1596			       (1ULL << EF10_STAT_port_rx_broadcast) |	\
1597			       (1ULL << EF10_STAT_port_rx_lt64) |	\
1598			       (1ULL << EF10_STAT_port_rx_64) |		\
1599			       (1ULL << EF10_STAT_port_rx_65_to_127) |	\
1600			       (1ULL << EF10_STAT_port_rx_128_to_255) |	\
1601			       (1ULL << EF10_STAT_port_rx_256_to_511) |	\
1602			       (1ULL << EF10_STAT_port_rx_512_to_1023) |\
1603			       (1ULL << EF10_STAT_port_rx_1024_to_15xx) |\
1604			       (1ULL << EF10_STAT_port_rx_15xx_to_jumbo) |\
1605			       (1ULL << EF10_STAT_port_rx_gtjumbo) |	\
1606			       (1ULL << EF10_STAT_port_rx_bad_gtjumbo) |\
1607			       (1ULL << EF10_STAT_port_rx_overflow) |	\
1608			       (1ULL << EF10_STAT_port_rx_nodesc_drops) |\
1609			       (1ULL << GENERIC_STAT_rx_nodesc_trunc) |	\
1610			       (1ULL << GENERIC_STAT_rx_noskb_drops))
1611
1612/* On 7000 series NICs, these statistics are only provided by the 10G MAC.
1613 * For a 10G/40G switchable port we do not expose these because they might
1614 * not include all the packets they should.
1615 * On 8000 series NICs these statistics are always provided.
1616 */
1617#define HUNT_10G_ONLY_STAT_MASK ((1ULL << EF10_STAT_port_tx_control) |	\
1618				 (1ULL << EF10_STAT_port_tx_lt64) |	\
1619				 (1ULL << EF10_STAT_port_tx_64) |	\
1620				 (1ULL << EF10_STAT_port_tx_65_to_127) |\
1621				 (1ULL << EF10_STAT_port_tx_128_to_255) |\
1622				 (1ULL << EF10_STAT_port_tx_256_to_511) |\
1623				 (1ULL << EF10_STAT_port_tx_512_to_1023) |\
1624				 (1ULL << EF10_STAT_port_tx_1024_to_15xx) |\
1625				 (1ULL << EF10_STAT_port_tx_15xx_to_jumbo))
1626
1627/* These statistics are only provided by the 40G MAC.  For a 10G/40G
1628 * switchable port we do expose these because the errors will otherwise
1629 * be silent.
1630 */
1631#define HUNT_40G_EXTRA_STAT_MASK ((1ULL << EF10_STAT_port_rx_align_error) |\
1632				  (1ULL << EF10_STAT_port_rx_length_error))
1633
1634/* These statistics are only provided if the firmware supports the
1635 * capability PM_AND_RXDP_COUNTERS.
1636 */
1637#define HUNT_PM_AND_RXDP_STAT_MASK (					\
1638	(1ULL << EF10_STAT_port_rx_pm_trunc_bb_overflow) |		\
1639	(1ULL << EF10_STAT_port_rx_pm_discard_bb_overflow) |		\
1640	(1ULL << EF10_STAT_port_rx_pm_trunc_vfifo_full) |		\
1641	(1ULL << EF10_STAT_port_rx_pm_discard_vfifo_full) |		\
1642	(1ULL << EF10_STAT_port_rx_pm_trunc_qbb) |			\
1643	(1ULL << EF10_STAT_port_rx_pm_discard_qbb) |			\
1644	(1ULL << EF10_STAT_port_rx_pm_discard_mapping) |		\
1645	(1ULL << EF10_STAT_port_rx_dp_q_disabled_packets) |		\
1646	(1ULL << EF10_STAT_port_rx_dp_di_dropped_packets) |		\
1647	(1ULL << EF10_STAT_port_rx_dp_streaming_packets) |		\
1648	(1ULL << EF10_STAT_port_rx_dp_hlb_fetch) |			\
1649	(1ULL << EF10_STAT_port_rx_dp_hlb_wait))
1650
1651/* These statistics are only provided if the NIC supports MC_CMD_MAC_STATS_V2,
1652 * indicated by returning a value >= MC_CMD_MAC_NSTATS_V2 in
1653 * MC_CMD_GET_CAPABILITIES_V4_OUT_MAC_STATS_NUM_STATS.
1654 * These bits are in the second u64 of the raw mask.
1655 */
1656#define EF10_FEC_STAT_MASK (						\
1657	(1ULL << (EF10_STAT_fec_uncorrected_errors - 64)) |		\
1658	(1ULL << (EF10_STAT_fec_corrected_errors - 64)) |		\
1659	(1ULL << (EF10_STAT_fec_corrected_symbols_lane0 - 64)) |	\
1660	(1ULL << (EF10_STAT_fec_corrected_symbols_lane1 - 64)) |	\
1661	(1ULL << (EF10_STAT_fec_corrected_symbols_lane2 - 64)) |	\
1662	(1ULL << (EF10_STAT_fec_corrected_symbols_lane3 - 64)))
1663
1664/* These statistics are only provided if the NIC supports MC_CMD_MAC_STATS_V3,
1665 * indicated by returning a value >= MC_CMD_MAC_NSTATS_V3 in
1666 * MC_CMD_GET_CAPABILITIES_V4_OUT_MAC_STATS_NUM_STATS.
1667 * These bits are in the second u64 of the raw mask.
1668 */
1669#define EF10_CTPIO_STAT_MASK (						\
1670	(1ULL << (EF10_STAT_ctpio_vi_busy_fallback - 64)) |		\
1671	(1ULL << (EF10_STAT_ctpio_long_write_success - 64)) |		\
1672	(1ULL << (EF10_STAT_ctpio_missing_dbell_fail - 64)) |		\
1673	(1ULL << (EF10_STAT_ctpio_overflow_fail - 64)) |		\
1674	(1ULL << (EF10_STAT_ctpio_underflow_fail - 64)) |		\
1675	(1ULL << (EF10_STAT_ctpio_timeout_fail - 64)) |			\
1676	(1ULL << (EF10_STAT_ctpio_noncontig_wr_fail - 64)) |		\
1677	(1ULL << (EF10_STAT_ctpio_frm_clobber_fail - 64)) |		\
1678	(1ULL << (EF10_STAT_ctpio_invalid_wr_fail - 64)) |		\
1679	(1ULL << (EF10_STAT_ctpio_vi_clobber_fallback - 64)) |		\
1680	(1ULL << (EF10_STAT_ctpio_unqualified_fallback - 64)) |		\
1681	(1ULL << (EF10_STAT_ctpio_runt_fallback - 64)) |		\
1682	(1ULL << (EF10_STAT_ctpio_success - 64)) |			\
1683	(1ULL << (EF10_STAT_ctpio_fallback - 64)) |			\
1684	(1ULL << (EF10_STAT_ctpio_poison - 64)) |			\
1685	(1ULL << (EF10_STAT_ctpio_erase - 64)))
1686
1687static u64 efx_ef10_raw_stat_mask(struct efx_nic *efx)
1688{
1689	u64 raw_mask = HUNT_COMMON_STAT_MASK;
1690	u32 port_caps = efx_mcdi_phy_get_caps(efx);
1691	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1692
1693	if (!(efx->mcdi->fn_flags &
1694	      1 << MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_LINKCTRL))
1695		return 0;
1696
1697	if (port_caps & (1 << MC_CMD_PHY_CAP_40000FDX_LBN)) {
1698		raw_mask |= HUNT_40G_EXTRA_STAT_MASK;
1699		/* 8000 series have everything even at 40G */
1700		if (nic_data->datapath_caps2 &
1701		    (1 << MC_CMD_GET_CAPABILITIES_V2_OUT_MAC_STATS_40G_TX_SIZE_BINS_LBN))
1702			raw_mask |= HUNT_10G_ONLY_STAT_MASK;
1703	} else {
1704		raw_mask |= HUNT_10G_ONLY_STAT_MASK;
1705	}
1706
1707	if (nic_data->datapath_caps &
1708	    (1 << MC_CMD_GET_CAPABILITIES_OUT_PM_AND_RXDP_COUNTERS_LBN))
1709		raw_mask |= HUNT_PM_AND_RXDP_STAT_MASK;
1710
1711	return raw_mask;
1712}
1713
1714static void efx_ef10_get_stat_mask(struct efx_nic *efx, unsigned long *mask)
1715{
1716	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1717	u64 raw_mask[2];
1718
1719	raw_mask[0] = efx_ef10_raw_stat_mask(efx);
1720
1721	/* Only show vadaptor stats when EVB capability is present */
1722	if (nic_data->datapath_caps &
1723	    (1 << MC_CMD_GET_CAPABILITIES_OUT_EVB_LBN)) {
1724		raw_mask[0] |= ~((1ULL << EF10_STAT_rx_unicast) - 1);
1725		raw_mask[1] = (1ULL << (EF10_STAT_V1_COUNT - 64)) - 1;
1726	} else {
1727		raw_mask[1] = 0;
1728	}
1729	/* Only show FEC stats when NIC supports MC_CMD_MAC_STATS_V2 */
1730	if (efx->num_mac_stats >= MC_CMD_MAC_NSTATS_V2)
1731		raw_mask[1] |= EF10_FEC_STAT_MASK;
1732
1733	/* CTPIO stats appear in V3. Only show them on devices that actually
1734	 * support CTPIO. Although this driver doesn't use CTPIO others might,
1735	 * and we may be reporting the stats for the underlying port.
1736	 */
1737	if (efx->num_mac_stats >= MC_CMD_MAC_NSTATS_V3 &&
1738	    (nic_data->datapath_caps2 &
1739	     (1 << MC_CMD_GET_CAPABILITIES_V4_OUT_CTPIO_LBN)))
1740		raw_mask[1] |= EF10_CTPIO_STAT_MASK;
1741
1742#if BITS_PER_LONG == 64
1743	BUILD_BUG_ON(BITS_TO_LONGS(EF10_STAT_COUNT) != 2);
1744	mask[0] = raw_mask[0];
1745	mask[1] = raw_mask[1];
1746#else
1747	BUILD_BUG_ON(BITS_TO_LONGS(EF10_STAT_COUNT) != 3);
1748	mask[0] = raw_mask[0] & 0xffffffff;
1749	mask[1] = raw_mask[0] >> 32;
1750	mask[2] = raw_mask[1] & 0xffffffff;
1751#endif
1752}
1753
1754static size_t efx_ef10_describe_stats(struct efx_nic *efx, u8 *names)
1755{
1756	DECLARE_BITMAP(mask, EF10_STAT_COUNT);
1757
1758	efx_ef10_get_stat_mask(efx, mask);
1759	return efx_nic_describe_stats(efx_ef10_stat_desc, EF10_STAT_COUNT,
1760				      mask, names);
1761}
1762
1763static size_t efx_ef10_update_stats_common(struct efx_nic *efx, u64 *full_stats,
1764					   struct rtnl_link_stats64 *core_stats)
1765{
1766	DECLARE_BITMAP(mask, EF10_STAT_COUNT);
1767	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1768	u64 *stats = nic_data->stats;
1769	size_t stats_count = 0, index;
1770
1771	efx_ef10_get_stat_mask(efx, mask);
1772
1773	if (full_stats) {
1774		for_each_set_bit(index, mask, EF10_STAT_COUNT) {
1775			if (efx_ef10_stat_desc[index].name) {
1776				*full_stats++ = stats[index];
1777				++stats_count;
1778			}
1779		}
1780	}
1781
1782	if (!core_stats)
1783		return stats_count;
1784
1785	if (nic_data->datapath_caps &
1786			1 << MC_CMD_GET_CAPABILITIES_OUT_EVB_LBN) {
1787		/* Use vadaptor stats. */
1788		core_stats->rx_packets = stats[EF10_STAT_rx_unicast] +
1789					 stats[EF10_STAT_rx_multicast] +
1790					 stats[EF10_STAT_rx_broadcast];
1791		core_stats->tx_packets = stats[EF10_STAT_tx_unicast] +
1792					 stats[EF10_STAT_tx_multicast] +
1793					 stats[EF10_STAT_tx_broadcast];
1794		core_stats->rx_bytes = stats[EF10_STAT_rx_unicast_bytes] +
1795				       stats[EF10_STAT_rx_multicast_bytes] +
1796				       stats[EF10_STAT_rx_broadcast_bytes];
1797		core_stats->tx_bytes = stats[EF10_STAT_tx_unicast_bytes] +
1798				       stats[EF10_STAT_tx_multicast_bytes] +
1799				       stats[EF10_STAT_tx_broadcast_bytes];
1800		core_stats->rx_dropped = stats[GENERIC_STAT_rx_nodesc_trunc] +
1801					 stats[GENERIC_STAT_rx_noskb_drops];
1802		core_stats->multicast = stats[EF10_STAT_rx_multicast];
1803		core_stats->rx_crc_errors = stats[EF10_STAT_rx_bad];
1804		core_stats->rx_fifo_errors = stats[EF10_STAT_rx_overflow];
1805		core_stats->rx_errors = core_stats->rx_crc_errors;
1806		core_stats->tx_errors = stats[EF10_STAT_tx_bad];
1807	} else {
1808		/* Use port stats. */
1809		core_stats->rx_packets = stats[EF10_STAT_port_rx_packets];
1810		core_stats->tx_packets = stats[EF10_STAT_port_tx_packets];
1811		core_stats->rx_bytes = stats[EF10_STAT_port_rx_bytes];
1812		core_stats->tx_bytes = stats[EF10_STAT_port_tx_bytes];
1813		core_stats->rx_dropped = stats[EF10_STAT_port_rx_nodesc_drops] +
1814					 stats[GENERIC_STAT_rx_nodesc_trunc] +
1815					 stats[GENERIC_STAT_rx_noskb_drops];
1816		core_stats->multicast = stats[EF10_STAT_port_rx_multicast];
1817		core_stats->rx_length_errors =
1818				stats[EF10_STAT_port_rx_gtjumbo] +
1819				stats[EF10_STAT_port_rx_length_error];
1820		core_stats->rx_crc_errors = stats[EF10_STAT_port_rx_bad];
1821		core_stats->rx_frame_errors =
1822				stats[EF10_STAT_port_rx_align_error];
1823		core_stats->rx_fifo_errors = stats[EF10_STAT_port_rx_overflow];
1824		core_stats->rx_errors = (core_stats->rx_length_errors +
1825					 core_stats->rx_crc_errors +
1826					 core_stats->rx_frame_errors);
1827	}
1828
1829	return stats_count;
1830}
1831
1832static size_t efx_ef10_update_stats_pf(struct efx_nic *efx, u64 *full_stats,
1833				       struct rtnl_link_stats64 *core_stats)
1834{
1835	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1836	DECLARE_BITMAP(mask, EF10_STAT_COUNT);
1837	u64 *stats = nic_data->stats;
1838
1839	efx_ef10_get_stat_mask(efx, mask);
1840
1841	/* If NIC was fini'd (probably resetting), then we can't read
1842	 * updated stats right now.
1843	 */
1844	if (nic_data->mc_stats) {
1845		efx_nic_copy_stats(efx, nic_data->mc_stats);
1846		efx_nic_update_stats(efx_ef10_stat_desc, EF10_STAT_COUNT,
1847				     mask, stats, nic_data->mc_stats, false);
1848	}
1849
1850	/* Update derived statistics */
1851	efx_nic_fix_nodesc_drop_stat(efx,
1852				     &stats[EF10_STAT_port_rx_nodesc_drops]);
1853	/* MC Firmware reads RX_BYTES and RX_GOOD_BYTES from the MAC.
1854	 * It then calculates RX_BAD_BYTES and DMAs it to us with RX_BYTES.
1855	 * We report these as port_rx_ stats. We are not given RX_GOOD_BYTES.
1856	 * Here we calculate port_rx_good_bytes.
1857	 */
1858	stats[EF10_STAT_port_rx_good_bytes] =
1859		stats[EF10_STAT_port_rx_bytes] -
1860		stats[EF10_STAT_port_rx_bytes_minus_good_bytes];
1861
1862	/* The asynchronous reads used to calculate RX_BAD_BYTES in
1863	 * MC Firmware are done such that we should not see an increase in
1864	 * RX_BAD_BYTES when a good packet has arrived. Unfortunately this
1865	 * does mean that the stat can decrease at times. Here we do not
1866	 * update the stat unless it has increased or has gone to zero
1867	 * (In the case of the NIC rebooting).
1868	 * Please see Bug 33781 for a discussion of why things work this way.
1869	 */
1870	efx_update_diff_stat(&stats[EF10_STAT_port_rx_bad_bytes],
1871			     stats[EF10_STAT_port_rx_bytes_minus_good_bytes]);
1872	efx_update_sw_stats(efx, stats);
1873
1874	return efx_ef10_update_stats_common(efx, full_stats, core_stats);
1875}
1876
1877static int efx_ef10_try_update_nic_stats_vf(struct efx_nic *efx)
1878	__must_hold(&efx->stats_lock)
1879{
1880	MCDI_DECLARE_BUF(inbuf, MC_CMD_MAC_STATS_IN_LEN);
1881	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1882	DECLARE_BITMAP(mask, EF10_STAT_COUNT);
1883	__le64 generation_start, generation_end;
1884	u64 *stats = nic_data->stats;
1885	u32 dma_len = efx->num_mac_stats * sizeof(u64);
1886	struct efx_buffer stats_buf;
1887	__le64 *dma_stats;
1888	int rc;
1889
1890	spin_unlock_bh(&efx->stats_lock);
1891
1892	efx_ef10_get_stat_mask(efx, mask);
1893
1894	rc = efx_nic_alloc_buffer(efx, &stats_buf, dma_len, GFP_KERNEL);
1895	if (rc) {
1896		spin_lock_bh(&efx->stats_lock);
1897		return rc;
1898	}
1899
1900	dma_stats = stats_buf.addr;
1901	dma_stats[efx->num_mac_stats - 1] = EFX_MC_STATS_GENERATION_INVALID;
1902
1903	MCDI_SET_QWORD(inbuf, MAC_STATS_IN_DMA_ADDR, stats_buf.dma_addr);
1904	MCDI_POPULATE_DWORD_1(inbuf, MAC_STATS_IN_CMD,
1905			      MAC_STATS_IN_DMA, 1);
1906	MCDI_SET_DWORD(inbuf, MAC_STATS_IN_DMA_LEN, dma_len);
1907	MCDI_SET_DWORD(inbuf, MAC_STATS_IN_PORT_ID, EVB_PORT_ID_ASSIGNED);
1908
1909	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_MAC_STATS, inbuf, sizeof(inbuf),
1910				NULL, 0, NULL);
1911	spin_lock_bh(&efx->stats_lock);
1912	if (rc) {
1913		/* Expect ENOENT if DMA queues have not been set up */
1914		if (rc != -ENOENT || atomic_read(&efx->active_queues))
1915			efx_mcdi_display_error(efx, MC_CMD_MAC_STATS,
1916					       sizeof(inbuf), NULL, 0, rc);
1917		goto out;
1918	}
1919
1920	generation_end = dma_stats[efx->num_mac_stats - 1];
1921	if (generation_end == EFX_MC_STATS_GENERATION_INVALID) {
1922		WARN_ON_ONCE(1);
1923		goto out;
1924	}
1925	rmb();
1926	efx_nic_update_stats(efx_ef10_stat_desc, EF10_STAT_COUNT, mask,
1927			     stats, stats_buf.addr, false);
1928	rmb();
1929	generation_start = dma_stats[MC_CMD_MAC_GENERATION_START];
1930	if (generation_end != generation_start) {
1931		rc = -EAGAIN;
1932		goto out;
1933	}
1934
1935	efx_update_sw_stats(efx, stats);
1936out:
1937	/* releasing a DMA coherent buffer with BH disabled can panic */
1938	spin_unlock_bh(&efx->stats_lock);
1939	efx_nic_free_buffer(efx, &stats_buf);
1940	spin_lock_bh(&efx->stats_lock);
1941	return rc;
1942}
1943
1944static size_t efx_ef10_update_stats_vf(struct efx_nic *efx, u64 *full_stats,
1945				       struct rtnl_link_stats64 *core_stats)
1946{
1947	if (efx_ef10_try_update_nic_stats_vf(efx))
1948		return 0;
1949
1950	return efx_ef10_update_stats_common(efx, full_stats, core_stats);
1951}
1952
1953static size_t efx_ef10_update_stats_atomic_vf(struct efx_nic *efx, u64 *full_stats,
1954					      struct rtnl_link_stats64 *core_stats)
1955{
1956	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1957
1958	/* In atomic context, cannot update HW stats.  Just update the
1959	 * software stats and return so the caller can continue.
1960	 */
1961	efx_update_sw_stats(efx, nic_data->stats);
1962	return efx_ef10_update_stats_common(efx, full_stats, core_stats);
1963}
1964
1965static void efx_ef10_push_irq_moderation(struct efx_channel *channel)
1966{
1967	struct efx_nic *efx = channel->efx;
1968	unsigned int mode, usecs;
1969	efx_dword_t timer_cmd;
1970
1971	if (channel->irq_moderation_us) {
1972		mode = 3;
1973		usecs = channel->irq_moderation_us;
1974	} else {
1975		mode = 0;
1976		usecs = 0;
1977	}
1978
1979	if (EFX_EF10_WORKAROUND_61265(efx)) {
1980		MCDI_DECLARE_BUF(inbuf, MC_CMD_SET_EVQ_TMR_IN_LEN);
1981		unsigned int ns = usecs * 1000;
1982
1983		MCDI_SET_DWORD(inbuf, SET_EVQ_TMR_IN_INSTANCE,
1984			       channel->channel);
1985		MCDI_SET_DWORD(inbuf, SET_EVQ_TMR_IN_TMR_LOAD_REQ_NS, ns);
1986		MCDI_SET_DWORD(inbuf, SET_EVQ_TMR_IN_TMR_RELOAD_REQ_NS, ns);
1987		MCDI_SET_DWORD(inbuf, SET_EVQ_TMR_IN_TMR_MODE, mode);
1988
1989		efx_mcdi_rpc_async(efx, MC_CMD_SET_EVQ_TMR,
1990				   inbuf, sizeof(inbuf), 0, NULL, 0);
1991	} else if (EFX_EF10_WORKAROUND_35388(efx)) {
1992		unsigned int ticks = efx_usecs_to_ticks(efx, usecs);
1993
1994		EFX_POPULATE_DWORD_3(timer_cmd, ERF_DD_EVQ_IND_TIMER_FLAGS,
1995				     EFE_DD_EVQ_IND_TIMER_FLAGS,
1996				     ERF_DD_EVQ_IND_TIMER_MODE, mode,
1997				     ERF_DD_EVQ_IND_TIMER_VAL, ticks);
1998		efx_writed_page(efx, &timer_cmd, ER_DD_EVQ_INDIRECT,
1999				channel->channel);
2000	} else {
2001		unsigned int ticks = efx_usecs_to_ticks(efx, usecs);
2002
2003		EFX_POPULATE_DWORD_3(timer_cmd, ERF_DZ_TC_TIMER_MODE, mode,
2004				     ERF_DZ_TC_TIMER_VAL, ticks,
2005				     ERF_FZ_TC_TMR_REL_VAL, ticks);
2006		efx_writed_page(efx, &timer_cmd, ER_DZ_EVQ_TMR,
2007				channel->channel);
2008	}
2009}
2010
2011static void efx_ef10_get_wol_vf(struct efx_nic *efx,
2012				struct ethtool_wolinfo *wol) {}
2013
2014static int efx_ef10_set_wol_vf(struct efx_nic *efx, u32 type)
2015{
2016	return -EOPNOTSUPP;
2017}
2018
2019static void efx_ef10_get_wol(struct efx_nic *efx, struct ethtool_wolinfo *wol)
2020{
2021	wol->supported = 0;
2022	wol->wolopts = 0;
2023	memset(&wol->sopass, 0, sizeof(wol->sopass));
2024}
2025
2026static int efx_ef10_set_wol(struct efx_nic *efx, u32 type)
2027{
2028	if (type != 0)
2029		return -EINVAL;
2030	return 0;
2031}
2032
2033static void efx_ef10_mcdi_request(struct efx_nic *efx,
2034				  const efx_dword_t *hdr, size_t hdr_len,
2035				  const efx_dword_t *sdu, size_t sdu_len)
2036{
2037	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2038	u8 *pdu = nic_data->mcdi_buf.addr;
2039
2040	memcpy(pdu, hdr, hdr_len);
2041	memcpy(pdu + hdr_len, sdu, sdu_len);
2042	wmb();
2043
2044	/* The hardware provides 'low' and 'high' (doorbell) registers
2045	 * for passing the 64-bit address of an MCDI request to
2046	 * firmware.  However the dwords are swapped by firmware.  The
2047	 * least significant bits of the doorbell are then 0 for all
2048	 * MCDI requests due to alignment.
2049	 */
2050	_efx_writed(efx, cpu_to_le32((u64)nic_data->mcdi_buf.dma_addr >> 32),
2051		    ER_DZ_MC_DB_LWRD);
2052	_efx_writed(efx, cpu_to_le32((u32)nic_data->mcdi_buf.dma_addr),
2053		    ER_DZ_MC_DB_HWRD);
2054}
2055
2056static bool efx_ef10_mcdi_poll_response(struct efx_nic *efx)
2057{
2058	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2059	const efx_dword_t hdr = *(const efx_dword_t *)nic_data->mcdi_buf.addr;
2060
2061	rmb();
2062	return EFX_DWORD_FIELD(hdr, MCDI_HEADER_RESPONSE);
2063}
2064
2065static void
2066efx_ef10_mcdi_read_response(struct efx_nic *efx, efx_dword_t *outbuf,
2067			    size_t offset, size_t outlen)
2068{
2069	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2070	const u8 *pdu = nic_data->mcdi_buf.addr;
2071
2072	memcpy(outbuf, pdu + offset, outlen);
2073}
2074
2075static void efx_ef10_mcdi_reboot_detected(struct efx_nic *efx)
2076{
2077	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2078
2079	/* All our allocations have been reset */
2080	efx_ef10_table_reset_mc_allocations(efx);
2081
2082	/* The datapath firmware might have been changed */
2083	nic_data->must_check_datapath_caps = true;
2084
2085	/* MAC statistics have been cleared on the NIC; clear the local
2086	 * statistic that we update with efx_update_diff_stat().
2087	 */
2088	nic_data->stats[EF10_STAT_port_rx_bad_bytes] = 0;
2089}
2090
2091static int efx_ef10_mcdi_poll_reboot(struct efx_nic *efx)
2092{
2093	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2094	int rc;
2095
2096	rc = efx_ef10_get_warm_boot_count(efx);
2097	if (rc < 0) {
2098		/* The firmware is presumably in the process of
2099		 * rebooting.  However, we are supposed to report each
2100		 * reboot just once, so we must only do that once we
2101		 * can read and store the updated warm boot count.
2102		 */
2103		return 0;
2104	}
2105
2106	if (rc == nic_data->warm_boot_count)
2107		return 0;
2108
2109	nic_data->warm_boot_count = rc;
2110	efx_ef10_mcdi_reboot_detected(efx);
2111
2112	return -EIO;
2113}
2114
2115/* Handle an MSI interrupt
2116 *
2117 * Handle an MSI hardware interrupt.  This routine schedules event
2118 * queue processing.  No interrupt acknowledgement cycle is necessary.
2119 * Also, we never need to check that the interrupt is for us, since
2120 * MSI interrupts cannot be shared.
2121 */
2122static irqreturn_t efx_ef10_msi_interrupt(int irq, void *dev_id)
2123{
2124	struct efx_msi_context *context = dev_id;
2125	struct efx_nic *efx = context->efx;
2126
2127	netif_vdbg(efx, intr, efx->net_dev,
2128		   "IRQ %d on CPU %d\n", irq, raw_smp_processor_id());
2129
2130	if (likely(READ_ONCE(efx->irq_soft_enabled))) {
2131		/* Note test interrupts */
2132		if (context->index == efx->irq_level)
2133			efx->last_irq_cpu = raw_smp_processor_id();
2134
2135		/* Schedule processing of the channel */
2136		efx_schedule_channel_irq(efx->channel[context->index]);
2137	}
2138
2139	return IRQ_HANDLED;
2140}
2141
2142static irqreturn_t efx_ef10_legacy_interrupt(int irq, void *dev_id)
2143{
2144	struct efx_nic *efx = dev_id;
2145	bool soft_enabled = READ_ONCE(efx->irq_soft_enabled);
2146	struct efx_channel *channel;
2147	efx_dword_t reg;
2148	u32 queues;
2149
2150	/* Read the ISR which also ACKs the interrupts */
2151	efx_readd(efx, &reg, ER_DZ_BIU_INT_ISR);
2152	queues = EFX_DWORD_FIELD(reg, ERF_DZ_ISR_REG);
2153
2154	if (queues == 0)
2155		return IRQ_NONE;
2156
2157	if (likely(soft_enabled)) {
2158		/* Note test interrupts */
2159		if (queues & (1U << efx->irq_level))
2160			efx->last_irq_cpu = raw_smp_processor_id();
2161
2162		efx_for_each_channel(channel, efx) {
2163			if (queues & 1)
2164				efx_schedule_channel_irq(channel);
2165			queues >>= 1;
2166		}
2167	}
2168
2169	netif_vdbg(efx, intr, efx->net_dev,
2170		   "IRQ %d on CPU %d status " EFX_DWORD_FMT "\n",
2171		   irq, raw_smp_processor_id(), EFX_DWORD_VAL(reg));
2172
2173	return IRQ_HANDLED;
2174}
2175
2176static int efx_ef10_irq_test_generate(struct efx_nic *efx)
2177{
2178	MCDI_DECLARE_BUF(inbuf, MC_CMD_TRIGGER_INTERRUPT_IN_LEN);
2179
2180	if (efx_mcdi_set_workaround(efx, MC_CMD_WORKAROUND_BUG41750, true,
2181				    NULL) == 0)
2182		return -ENOTSUPP;
2183
2184	BUILD_BUG_ON(MC_CMD_TRIGGER_INTERRUPT_OUT_LEN != 0);
2185
2186	MCDI_SET_DWORD(inbuf, TRIGGER_INTERRUPT_IN_INTR_LEVEL, efx->irq_level);
2187	return efx_mcdi_rpc(efx, MC_CMD_TRIGGER_INTERRUPT,
2188			    inbuf, sizeof(inbuf), NULL, 0, NULL);
2189}
2190
2191static int efx_ef10_tx_probe(struct efx_tx_queue *tx_queue)
2192{
2193	/* low two bits of label are what we want for type */
2194	BUILD_BUG_ON((EFX_TXQ_TYPE_OUTER_CSUM | EFX_TXQ_TYPE_INNER_CSUM) != 3);
2195	tx_queue->type = tx_queue->label & 3;
2196	return efx_nic_alloc_buffer(tx_queue->efx, &tx_queue->txd.buf,
2197				    (tx_queue->ptr_mask + 1) *
2198				    sizeof(efx_qword_t),
2199				    GFP_KERNEL);
2200}
2201
2202/* This writes to the TX_DESC_WPTR and also pushes data */
2203static inline void efx_ef10_push_tx_desc(struct efx_tx_queue *tx_queue,
2204					 const efx_qword_t *txd)
2205{
2206	unsigned int write_ptr;
2207	efx_oword_t reg;
2208
2209	write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
2210	EFX_POPULATE_OWORD_1(reg, ERF_DZ_TX_DESC_WPTR, write_ptr);
2211	reg.qword[0] = *txd;
2212	efx_writeo_page(tx_queue->efx, &reg,
2213			ER_DZ_TX_DESC_UPD, tx_queue->queue);
2214}
2215
2216/* Add Firmware-Assisted TSO v2 option descriptors to a queue.
2217 */
2218int efx_ef10_tx_tso_desc(struct efx_tx_queue *tx_queue, struct sk_buff *skb,
2219			 bool *data_mapped)
2220{
2221	struct efx_tx_buffer *buffer;
2222	u16 inner_ipv4_id = 0;
2223	u16 outer_ipv4_id = 0;
2224	struct tcphdr *tcp;
2225	struct iphdr *ip;
2226	u16 ip_tot_len;
2227	u32 seqnum;
2228	u32 mss;
2229
2230	EFX_WARN_ON_ONCE_PARANOID(tx_queue->tso_version != 2);
2231
2232	mss = skb_shinfo(skb)->gso_size;
2233
2234	if (unlikely(mss < 4)) {
2235		WARN_ONCE(1, "MSS of %u is too small for TSO v2\n", mss);
2236		return -EINVAL;
2237	}
2238
2239	if (skb->encapsulation) {
2240		if (!tx_queue->tso_encap)
2241			return -EINVAL;
2242		ip = ip_hdr(skb);
2243		if (ip->version == 4)
2244			outer_ipv4_id = ntohs(ip->id);
2245
2246		ip = inner_ip_hdr(skb);
2247		tcp = inner_tcp_hdr(skb);
2248	} else {
2249		ip = ip_hdr(skb);
2250		tcp = tcp_hdr(skb);
2251	}
2252
2253	/* 8000-series EF10 hardware requires that IP Total Length be
2254	 * greater than or equal to the value it will have in each segment
2255	 * (which is at most mss + 208 + TCP header length), but also less
2256	 * than (0x10000 - inner_network_header).  Otherwise the TCP
2257	 * checksum calculation will be broken for encapsulated packets.
2258	 * We fill in ip->tot_len with 0xff30, which should satisfy the
2259	 * first requirement unless the MSS is ridiculously large (which
2260	 * should be impossible as the driver max MTU is 9216); it is
2261	 * guaranteed to satisfy the second as we only attempt TSO if
2262	 * inner_network_header <= 208.
2263	 */
2264	ip_tot_len = 0x10000 - EFX_TSO2_MAX_HDRLEN;
2265	EFX_WARN_ON_ONCE_PARANOID(mss + EFX_TSO2_MAX_HDRLEN +
2266				  (tcp->doff << 2u) > ip_tot_len);
2267
2268	if (ip->version == 4) {
2269		ip->tot_len = htons(ip_tot_len);
2270		ip->check = 0;
2271		inner_ipv4_id = ntohs(ip->id);
2272	} else {
2273		((struct ipv6hdr *)ip)->payload_len = htons(ip_tot_len);
2274	}
2275
2276	seqnum = ntohl(tcp->seq);
2277
2278	buffer = efx_tx_queue_get_insert_buffer(tx_queue);
2279
2280	buffer->flags = EFX_TX_BUF_OPTION;
2281	buffer->len = 0;
2282	buffer->unmap_len = 0;
2283	EFX_POPULATE_QWORD_5(buffer->option,
2284			ESF_DZ_TX_DESC_IS_OPT, 1,
2285			ESF_DZ_TX_OPTION_TYPE, ESE_DZ_TX_OPTION_DESC_TSO,
2286			ESF_DZ_TX_TSO_OPTION_TYPE,
2287			ESE_DZ_TX_TSO_OPTION_DESC_FATSO2A,
2288			ESF_DZ_TX_TSO_IP_ID, inner_ipv4_id,
2289			ESF_DZ_TX_TSO_TCP_SEQNO, seqnum
2290			);
2291	++tx_queue->insert_count;
2292
2293	buffer = efx_tx_queue_get_insert_buffer(tx_queue);
2294
2295	buffer->flags = EFX_TX_BUF_OPTION;
2296	buffer->len = 0;
2297	buffer->unmap_len = 0;
2298	EFX_POPULATE_QWORD_5(buffer->option,
2299			ESF_DZ_TX_DESC_IS_OPT, 1,
2300			ESF_DZ_TX_OPTION_TYPE, ESE_DZ_TX_OPTION_DESC_TSO,
2301			ESF_DZ_TX_TSO_OPTION_TYPE,
2302			ESE_DZ_TX_TSO_OPTION_DESC_FATSO2B,
2303			ESF_DZ_TX_TSO_OUTER_IPID, outer_ipv4_id,
2304			ESF_DZ_TX_TSO_TCP_MSS, mss
2305			);
2306	++tx_queue->insert_count;
2307
2308	return 0;
2309}
2310
2311static u32 efx_ef10_tso_versions(struct efx_nic *efx)
2312{
2313	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2314	u32 tso_versions = 0;
2315
2316	if (nic_data->datapath_caps &
2317	    (1 << MC_CMD_GET_CAPABILITIES_OUT_TX_TSO_LBN))
2318		tso_versions |= BIT(1);
2319	if (nic_data->datapath_caps2 &
2320	    (1 << MC_CMD_GET_CAPABILITIES_V2_OUT_TX_TSO_V2_LBN))
2321		tso_versions |= BIT(2);
2322	return tso_versions;
2323}
2324
2325static void efx_ef10_tx_init(struct efx_tx_queue *tx_queue)
2326{
2327	bool csum_offload = tx_queue->type & EFX_TXQ_TYPE_OUTER_CSUM;
2328	bool inner_csum = tx_queue->type & EFX_TXQ_TYPE_INNER_CSUM;
2329	struct efx_channel *channel = tx_queue->channel;
2330	struct efx_nic *efx = tx_queue->efx;
2331	struct efx_ef10_nic_data *nic_data;
2332	efx_qword_t *txd;
2333	int rc;
2334
2335	nic_data = efx->nic_data;
2336
2337	/* Only attempt to enable TX timestamping if we have the license for it,
2338	 * otherwise TXQ init will fail
2339	 */
2340	if (!(nic_data->licensed_features &
2341	      (1 << LICENSED_V3_FEATURES_TX_TIMESTAMPS_LBN))) {
2342		tx_queue->timestamping = false;
2343		/* Disable sync events on this channel. */
2344		if (efx->type->ptp_set_ts_sync_events)
2345			efx->type->ptp_set_ts_sync_events(efx, false, false);
2346	}
2347
2348	/* TSOv2 is a limited resource that can only be configured on a limited
2349	 * number of queues. TSO without checksum offload is not really a thing,
2350	 * so we only enable it for those queues.
2351	 * TSOv2 cannot be used with Hardware timestamping, and is never needed
2352	 * for XDP tx.
2353	 */
2354	if (efx_has_cap(efx, TX_TSO_V2)) {
2355		if ((csum_offload || inner_csum) &&
2356		    !tx_queue->timestamping && !tx_queue->xdp_tx) {
2357			tx_queue->tso_version = 2;
2358			netif_dbg(efx, hw, efx->net_dev, "Using TSOv2 for channel %u\n",
2359				  channel->channel);
2360		}
2361	} else if (efx_has_cap(efx, TX_TSO)) {
2362		tx_queue->tso_version = 1;
2363	}
2364
2365	rc = efx_mcdi_tx_init(tx_queue);
2366	if (rc)
2367		goto fail;
2368
2369	/* A previous user of this TX queue might have set us up the
2370	 * bomb by writing a descriptor to the TX push collector but
2371	 * not the doorbell.  (Each collector belongs to a port, not a
2372	 * queue or function, so cannot easily be reset.)  We must
2373	 * attempt to push a no-op descriptor in its place.
2374	 */
2375	tx_queue->buffer[0].flags = EFX_TX_BUF_OPTION;
2376	tx_queue->insert_count = 1;
2377	txd = efx_tx_desc(tx_queue, 0);
2378	EFX_POPULATE_QWORD_7(*txd,
2379			     ESF_DZ_TX_DESC_IS_OPT, true,
2380			     ESF_DZ_TX_OPTION_TYPE,
2381			     ESE_DZ_TX_OPTION_DESC_CRC_CSUM,
2382			     ESF_DZ_TX_OPTION_UDP_TCP_CSUM, csum_offload,
2383			     ESF_DZ_TX_OPTION_IP_CSUM, csum_offload && tx_queue->tso_version != 2,
2384			     ESF_DZ_TX_OPTION_INNER_UDP_TCP_CSUM, inner_csum,
2385			     ESF_DZ_TX_OPTION_INNER_IP_CSUM, inner_csum && tx_queue->tso_version != 2,
2386			     ESF_DZ_TX_TIMESTAMP, tx_queue->timestamping);
2387	tx_queue->write_count = 1;
2388
2389	if (tx_queue->tso_version == 2 && efx_has_cap(efx, TX_TSO_V2_ENCAP))
2390		tx_queue->tso_encap = true;
2391
2392	wmb();
2393	efx_ef10_push_tx_desc(tx_queue, txd);
2394
2395	return;
2396
2397fail:
2398	netdev_WARN(efx->net_dev, "failed to initialise TXQ %d\n",
2399		    tx_queue->queue);
2400}
2401
2402/* This writes to the TX_DESC_WPTR; write pointer for TX descriptor ring */
2403static inline void efx_ef10_notify_tx_desc(struct efx_tx_queue *tx_queue)
2404{
2405	unsigned int write_ptr;
2406	efx_dword_t reg;
2407
2408	write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
2409	EFX_POPULATE_DWORD_1(reg, ERF_DZ_TX_DESC_WPTR_DWORD, write_ptr);
2410	efx_writed_page(tx_queue->efx, &reg,
2411			ER_DZ_TX_DESC_UPD_DWORD, tx_queue->queue);
2412}
2413
2414#define EFX_EF10_MAX_TX_DESCRIPTOR_LEN 0x3fff
2415
2416static unsigned int efx_ef10_tx_limit_len(struct efx_tx_queue *tx_queue,
2417					  dma_addr_t dma_addr, unsigned int len)
2418{
2419	if (len > EFX_EF10_MAX_TX_DESCRIPTOR_LEN) {
2420		/* If we need to break across multiple descriptors we should
2421		 * stop at a page boundary. This assumes the length limit is
2422		 * greater than the page size.
2423		 */
2424		dma_addr_t end = dma_addr + EFX_EF10_MAX_TX_DESCRIPTOR_LEN;
2425
2426		BUILD_BUG_ON(EFX_EF10_MAX_TX_DESCRIPTOR_LEN < EFX_PAGE_SIZE);
2427		len = (end & (~(EFX_PAGE_SIZE - 1))) - dma_addr;
2428	}
2429
2430	return len;
2431}
2432
2433static void efx_ef10_tx_write(struct efx_tx_queue *tx_queue)
2434{
2435	unsigned int old_write_count = tx_queue->write_count;
2436	struct efx_tx_buffer *buffer;
2437	unsigned int write_ptr;
2438	efx_qword_t *txd;
2439
2440	tx_queue->xmit_pending = false;
2441	if (unlikely(tx_queue->write_count == tx_queue->insert_count))
2442		return;
2443
2444	do {
2445		write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
2446		buffer = &tx_queue->buffer[write_ptr];
2447		txd = efx_tx_desc(tx_queue, write_ptr);
2448		++tx_queue->write_count;
2449
2450		/* Create TX descriptor ring entry */
2451		if (buffer->flags & EFX_TX_BUF_OPTION) {
2452			*txd = buffer->option;
2453			if (EFX_QWORD_FIELD(*txd, ESF_DZ_TX_OPTION_TYPE) == 1)
2454				/* PIO descriptor */
2455				tx_queue->packet_write_count = tx_queue->write_count;
2456		} else {
2457			tx_queue->packet_write_count = tx_queue->write_count;
2458			BUILD_BUG_ON(EFX_TX_BUF_CONT != 1);
2459			EFX_POPULATE_QWORD_3(
2460				*txd,
2461				ESF_DZ_TX_KER_CONT,
2462				buffer->flags & EFX_TX_BUF_CONT,
2463				ESF_DZ_TX_KER_BYTE_CNT, buffer->len,
2464				ESF_DZ_TX_KER_BUF_ADDR, buffer->dma_addr);
2465		}
2466	} while (tx_queue->write_count != tx_queue->insert_count);
2467
2468	wmb(); /* Ensure descriptors are written before they are fetched */
2469
2470	if (efx_nic_may_push_tx_desc(tx_queue, old_write_count)) {
2471		txd = efx_tx_desc(tx_queue,
2472				  old_write_count & tx_queue->ptr_mask);
2473		efx_ef10_push_tx_desc(tx_queue, txd);
2474		++tx_queue->pushes;
2475	} else {
2476		efx_ef10_notify_tx_desc(tx_queue);
2477	}
2478}
2479
2480static int efx_ef10_probe_multicast_chaining(struct efx_nic *efx)
2481{
2482	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2483	unsigned int enabled, implemented;
2484	bool want_workaround_26807;
2485	int rc;
2486
2487	rc = efx_mcdi_get_workarounds(efx, &implemented, &enabled);
2488	if (rc == -ENOSYS) {
2489		/* GET_WORKAROUNDS was implemented before this workaround,
2490		 * thus it must be unavailable in this firmware.
2491		 */
2492		nic_data->workaround_26807 = false;
2493		return 0;
2494	}
2495	if (rc)
2496		return rc;
2497	want_workaround_26807 =
2498		implemented & MC_CMD_GET_WORKAROUNDS_OUT_BUG26807;
2499	nic_data->workaround_26807 =
2500		!!(enabled & MC_CMD_GET_WORKAROUNDS_OUT_BUG26807);
2501
2502	if (want_workaround_26807 && !nic_data->workaround_26807) {
2503		unsigned int flags;
2504
2505		rc = efx_mcdi_set_workaround(efx,
2506					     MC_CMD_WORKAROUND_BUG26807,
2507					     true, &flags);
2508		if (!rc) {
2509			if (flags &
2510			    1 << MC_CMD_WORKAROUND_EXT_OUT_FLR_DONE_LBN) {
2511				netif_info(efx, drv, efx->net_dev,
2512					   "other functions on NIC have been reset\n");
2513
2514				/* With MCFW v4.6.x and earlier, the
2515				 * boot count will have incremented,
2516				 * so re-read the warm_boot_count
2517				 * value now to ensure this function
2518				 * doesn't think it has changed next
2519				 * time it checks.
2520				 */
2521				rc = efx_ef10_get_warm_boot_count(efx);
2522				if (rc >= 0) {
2523					nic_data->warm_boot_count = rc;
2524					rc = 0;
2525				}
2526			}
2527			nic_data->workaround_26807 = true;
2528		} else if (rc == -EPERM) {
2529			rc = 0;
2530		}
2531	}
2532	return rc;
2533}
2534
2535static int efx_ef10_filter_table_probe(struct efx_nic *efx)
2536{
2537	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2538	int rc = efx_ef10_probe_multicast_chaining(efx);
2539	struct efx_mcdi_filter_vlan *vlan;
2540
2541	if (rc)
2542		return rc;
2543	rc = efx_mcdi_filter_table_probe(efx, nic_data->workaround_26807);
2544
2545	if (rc)
2546		return rc;
2547
2548	list_for_each_entry(vlan, &nic_data->vlan_list, list) {
2549		rc = efx_mcdi_filter_add_vlan(efx, vlan->vid);
2550		if (rc)
2551			goto fail_add_vlan;
2552	}
2553	return 0;
2554
2555fail_add_vlan:
2556	efx_mcdi_filter_table_remove(efx);
2557	return rc;
2558}
2559
2560/* This creates an entry in the RX descriptor queue */
2561static inline void
2562efx_ef10_build_rx_desc(struct efx_rx_queue *rx_queue, unsigned int index)
2563{
2564	struct efx_rx_buffer *rx_buf;
2565	efx_qword_t *rxd;
2566
2567	rxd = efx_rx_desc(rx_queue, index);
2568	rx_buf = efx_rx_buffer(rx_queue, index);
2569	EFX_POPULATE_QWORD_2(*rxd,
2570			     ESF_DZ_RX_KER_BYTE_CNT, rx_buf->len,
2571			     ESF_DZ_RX_KER_BUF_ADDR, rx_buf->dma_addr);
2572}
2573
2574static void efx_ef10_rx_write(struct efx_rx_queue *rx_queue)
2575{
2576	struct efx_nic *efx = rx_queue->efx;
2577	unsigned int write_count;
2578	efx_dword_t reg;
2579
2580	/* Firmware requires that RX_DESC_WPTR be a multiple of 8 */
2581	write_count = rx_queue->added_count & ~7;
2582	if (rx_queue->notified_count == write_count)
2583		return;
2584
2585	do
2586		efx_ef10_build_rx_desc(
2587			rx_queue,
2588			rx_queue->notified_count & rx_queue->ptr_mask);
2589	while (++rx_queue->notified_count != write_count);
2590
2591	wmb();
2592	EFX_POPULATE_DWORD_1(reg, ERF_DZ_RX_DESC_WPTR,
2593			     write_count & rx_queue->ptr_mask);
2594	efx_writed_page(efx, &reg, ER_DZ_RX_DESC_UPD,
2595			efx_rx_queue_index(rx_queue));
2596}
2597
2598static efx_mcdi_async_completer efx_ef10_rx_defer_refill_complete;
2599
2600static void efx_ef10_rx_defer_refill(struct efx_rx_queue *rx_queue)
2601{
2602	struct efx_channel *channel = efx_rx_queue_channel(rx_queue);
2603	MCDI_DECLARE_BUF(inbuf, MC_CMD_DRIVER_EVENT_IN_LEN);
2604	efx_qword_t event;
2605
2606	EFX_POPULATE_QWORD_2(event,
2607			     ESF_DZ_EV_CODE, EFX_EF10_DRVGEN_EV,
2608			     ESF_DZ_EV_DATA, EFX_EF10_REFILL);
2609
2610	MCDI_SET_DWORD(inbuf, DRIVER_EVENT_IN_EVQ, channel->channel);
2611
2612	/* MCDI_SET_QWORD is not appropriate here since EFX_POPULATE_* has
2613	 * already swapped the data to little-endian order.
2614	 */
2615	memcpy(MCDI_PTR(inbuf, DRIVER_EVENT_IN_DATA), &event.u64[0],
2616	       sizeof(efx_qword_t));
2617
2618	efx_mcdi_rpc_async(channel->efx, MC_CMD_DRIVER_EVENT,
2619			   inbuf, sizeof(inbuf), 0,
2620			   efx_ef10_rx_defer_refill_complete, 0);
2621}
2622
2623static void
2624efx_ef10_rx_defer_refill_complete(struct efx_nic *efx, unsigned long cookie,
2625				  int rc, efx_dword_t *outbuf,
2626				  size_t outlen_actual)
2627{
2628	/* nothing to do */
2629}
2630
2631static int efx_ef10_ev_init(struct efx_channel *channel)
2632{
2633	struct efx_nic *efx = channel->efx;
2634	struct efx_ef10_nic_data *nic_data;
2635	bool use_v2, cut_thru;
2636
2637	nic_data = efx->nic_data;
2638	use_v2 = nic_data->datapath_caps2 &
2639			    1 << MC_CMD_GET_CAPABILITIES_V2_OUT_INIT_EVQ_V2_LBN;
2640	cut_thru = !(nic_data->datapath_caps &
2641			      1 << MC_CMD_GET_CAPABILITIES_OUT_RX_BATCHING_LBN);
2642	return efx_mcdi_ev_init(channel, cut_thru, use_v2);
2643}
2644
2645static void efx_ef10_handle_rx_wrong_queue(struct efx_rx_queue *rx_queue,
2646					   unsigned int rx_queue_label)
2647{
2648	struct efx_nic *efx = rx_queue->efx;
2649
2650	netif_info(efx, hw, efx->net_dev,
2651		   "rx event arrived on queue %d labeled as queue %u\n",
2652		   efx_rx_queue_index(rx_queue), rx_queue_label);
2653
2654	efx_schedule_reset(efx, RESET_TYPE_DISABLE);
2655}
2656
2657static void
2658efx_ef10_handle_rx_bad_lbits(struct efx_rx_queue *rx_queue,
2659			     unsigned int actual, unsigned int expected)
2660{
2661	unsigned int dropped = (actual - expected) & rx_queue->ptr_mask;
2662	struct efx_nic *efx = rx_queue->efx;
2663
2664	netif_info(efx, hw, efx->net_dev,
2665		   "dropped %d events (index=%d expected=%d)\n",
2666		   dropped, actual, expected);
2667
2668	efx_schedule_reset(efx, RESET_TYPE_DISABLE);
2669}
2670
2671/* partially received RX was aborted. clean up. */
2672static void efx_ef10_handle_rx_abort(struct efx_rx_queue *rx_queue)
2673{
2674	unsigned int rx_desc_ptr;
2675
2676	netif_dbg(rx_queue->efx, hw, rx_queue->efx->net_dev,
2677		  "scattered RX aborted (dropping %u buffers)\n",
2678		  rx_queue->scatter_n);
2679
2680	rx_desc_ptr = rx_queue->removed_count & rx_queue->ptr_mask;
2681
2682	efx_rx_packet(rx_queue, rx_desc_ptr, rx_queue->scatter_n,
2683		      0, EFX_RX_PKT_DISCARD);
2684
2685	rx_queue->removed_count += rx_queue->scatter_n;
2686	rx_queue->scatter_n = 0;
2687	rx_queue->scatter_len = 0;
2688	++efx_rx_queue_channel(rx_queue)->n_rx_nodesc_trunc;
2689}
2690
2691static u16 efx_ef10_handle_rx_event_errors(struct efx_channel *channel,
2692					   unsigned int n_packets,
2693					   unsigned int rx_encap_hdr,
2694					   unsigned int rx_l3_class,
2695					   unsigned int rx_l4_class,
2696					   const efx_qword_t *event)
2697{
2698	struct efx_nic *efx = channel->efx;
2699	bool handled = false;
2700
2701	if (EFX_QWORD_FIELD(*event, ESF_DZ_RX_ECRC_ERR)) {
2702		if (!(efx->net_dev->features & NETIF_F_RXALL)) {
2703			if (!efx->loopback_selftest)
2704				channel->n_rx_eth_crc_err += n_packets;
2705			return EFX_RX_PKT_DISCARD;
2706		}
2707		handled = true;
2708	}
2709	if (EFX_QWORD_FIELD(*event, ESF_DZ_RX_IPCKSUM_ERR)) {
2710		if (unlikely(rx_encap_hdr != ESE_EZ_ENCAP_HDR_VXLAN &&
2711			     rx_l3_class != ESE_DZ_L3_CLASS_IP4 &&
2712			     rx_l3_class != ESE_DZ_L3_CLASS_IP4_FRAG &&
2713			     rx_l3_class != ESE_DZ_L3_CLASS_IP6 &&
2714			     rx_l3_class != ESE_DZ_L3_CLASS_IP6_FRAG))
2715			netdev_WARN(efx->net_dev,
2716				    "invalid class for RX_IPCKSUM_ERR: event="
2717				    EFX_QWORD_FMT "\n",
2718				    EFX_QWORD_VAL(*event));
2719		if (!efx->loopback_selftest)
2720			*(rx_encap_hdr ?
2721			  &channel->n_rx_outer_ip_hdr_chksum_err :
2722			  &channel->n_rx_ip_hdr_chksum_err) += n_packets;
2723		return 0;
2724	}
2725	if (EFX_QWORD_FIELD(*event, ESF_DZ_RX_TCPUDP_CKSUM_ERR)) {
2726		if (unlikely(rx_encap_hdr != ESE_EZ_ENCAP_HDR_VXLAN &&
2727			     ((rx_l3_class != ESE_DZ_L3_CLASS_IP4 &&
2728			       rx_l3_class != ESE_DZ_L3_CLASS_IP6) ||
2729			      (rx_l4_class != ESE_FZ_L4_CLASS_TCP &&
2730			       rx_l4_class != ESE_FZ_L4_CLASS_UDP))))
2731			netdev_WARN(efx->net_dev,
2732				    "invalid class for RX_TCPUDP_CKSUM_ERR: event="
2733				    EFX_QWORD_FMT "\n",
2734				    EFX_QWORD_VAL(*event));
2735		if (!efx->loopback_selftest)
2736			*(rx_encap_hdr ?
2737			  &channel->n_rx_outer_tcp_udp_chksum_err :
2738			  &channel->n_rx_tcp_udp_chksum_err) += n_packets;
2739		return 0;
2740	}
2741	if (EFX_QWORD_FIELD(*event, ESF_EZ_RX_IP_INNER_CHKSUM_ERR)) {
2742		if (unlikely(!rx_encap_hdr))
2743			netdev_WARN(efx->net_dev,
2744				    "invalid encapsulation type for RX_IP_INNER_CHKSUM_ERR: event="
2745				    EFX_QWORD_FMT "\n",
2746				    EFX_QWORD_VAL(*event));
2747		else if (unlikely(rx_l3_class != ESE_DZ_L3_CLASS_IP4 &&
2748				  rx_l3_class != ESE_DZ_L3_CLASS_IP4_FRAG &&
2749				  rx_l3_class != ESE_DZ_L3_CLASS_IP6 &&
2750				  rx_l3_class != ESE_DZ_L3_CLASS_IP6_FRAG))
2751			netdev_WARN(efx->net_dev,
2752				    "invalid class for RX_IP_INNER_CHKSUM_ERR: event="
2753				    EFX_QWORD_FMT "\n",
2754				    EFX_QWORD_VAL(*event));
2755		if (!efx->loopback_selftest)
2756			channel->n_rx_inner_ip_hdr_chksum_err += n_packets;
2757		return 0;
2758	}
2759	if (EFX_QWORD_FIELD(*event, ESF_EZ_RX_TCP_UDP_INNER_CHKSUM_ERR)) {
2760		if (unlikely(!rx_encap_hdr))
2761			netdev_WARN(efx->net_dev,
2762				    "invalid encapsulation type for RX_TCP_UDP_INNER_CHKSUM_ERR: event="
2763				    EFX_QWORD_FMT "\n",
2764				    EFX_QWORD_VAL(*event));
2765		else if (unlikely((rx_l3_class != ESE_DZ_L3_CLASS_IP4 &&
2766				   rx_l3_class != ESE_DZ_L3_CLASS_IP6) ||
2767				  (rx_l4_class != ESE_FZ_L4_CLASS_TCP &&
2768				   rx_l4_class != ESE_FZ_L4_CLASS_UDP)))
2769			netdev_WARN(efx->net_dev,
2770				    "invalid class for RX_TCP_UDP_INNER_CHKSUM_ERR: event="
2771				    EFX_QWORD_FMT "\n",
2772				    EFX_QWORD_VAL(*event));
2773		if (!efx->loopback_selftest)
2774			channel->n_rx_inner_tcp_udp_chksum_err += n_packets;
2775		return 0;
2776	}
2777
2778	WARN_ON(!handled); /* No error bits were recognised */
2779	return 0;
2780}
2781
2782static int efx_ef10_handle_rx_event(struct efx_channel *channel,
2783				    const efx_qword_t *event)
2784{
2785	unsigned int rx_bytes, next_ptr_lbits, rx_queue_label;
2786	unsigned int rx_l3_class, rx_l4_class, rx_encap_hdr;
2787	unsigned int n_descs, n_packets, i;
2788	struct efx_nic *efx = channel->efx;
2789	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2790	struct efx_rx_queue *rx_queue;
2791	efx_qword_t errors;
2792	bool rx_cont;
2793	u16 flags = 0;
2794
2795	if (unlikely(READ_ONCE(efx->reset_pending)))
2796		return 0;
2797
2798	/* Basic packet information */
2799	rx_bytes = EFX_QWORD_FIELD(*event, ESF_DZ_RX_BYTES);
2800	next_ptr_lbits = EFX_QWORD_FIELD(*event, ESF_DZ_RX_DSC_PTR_LBITS);
2801	rx_queue_label = EFX_QWORD_FIELD(*event, ESF_DZ_RX_QLABEL);
2802	rx_l3_class = EFX_QWORD_FIELD(*event, ESF_DZ_RX_L3_CLASS);
2803	rx_l4_class = EFX_QWORD_FIELD(*event, ESF_FZ_RX_L4_CLASS);
2804	rx_cont = EFX_QWORD_FIELD(*event, ESF_DZ_RX_CONT);
2805	rx_encap_hdr =
2806		nic_data->datapath_caps &
2807			(1 << MC_CMD_GET_CAPABILITIES_OUT_VXLAN_NVGRE_LBN) ?
2808		EFX_QWORD_FIELD(*event, ESF_EZ_RX_ENCAP_HDR) :
2809		ESE_EZ_ENCAP_HDR_NONE;
2810
2811	if (EFX_QWORD_FIELD(*event, ESF_DZ_RX_DROP_EVENT))
2812		netdev_WARN(efx->net_dev, "saw RX_DROP_EVENT: event="
2813			    EFX_QWORD_FMT "\n",
2814			    EFX_QWORD_VAL(*event));
2815
2816	rx_queue = efx_channel_get_rx_queue(channel);
2817
2818	if (unlikely(rx_queue_label != efx_rx_queue_index(rx_queue)))
2819		efx_ef10_handle_rx_wrong_queue(rx_queue, rx_queue_label);
2820
2821	n_descs = ((next_ptr_lbits - rx_queue->removed_count) &
2822		   ((1 << ESF_DZ_RX_DSC_PTR_LBITS_WIDTH) - 1));
2823
2824	if (n_descs != rx_queue->scatter_n + 1) {
2825		struct efx_ef10_nic_data *nic_data = efx->nic_data;
2826
2827		/* detect rx abort */
2828		if (unlikely(n_descs == rx_queue->scatter_n)) {
2829			if (rx_queue->scatter_n == 0 || rx_bytes != 0)
2830				netdev_WARN(efx->net_dev,
2831					    "invalid RX abort: scatter_n=%u event="
2832					    EFX_QWORD_FMT "\n",
2833					    rx_queue->scatter_n,
2834					    EFX_QWORD_VAL(*event));
2835			efx_ef10_handle_rx_abort(rx_queue);
2836			return 0;
2837		}
2838
2839		/* Check that RX completion merging is valid, i.e.
2840		 * the current firmware supports it and this is a
2841		 * non-scattered packet.
2842		 */
2843		if (!(nic_data->datapath_caps &
2844		      (1 << MC_CMD_GET_CAPABILITIES_OUT_RX_BATCHING_LBN)) ||
2845		    rx_queue->scatter_n != 0 || rx_cont) {
2846			efx_ef10_handle_rx_bad_lbits(
2847				rx_queue, next_ptr_lbits,
2848				(rx_queue->removed_count +
2849				 rx_queue->scatter_n + 1) &
2850				((1 << ESF_DZ_RX_DSC_PTR_LBITS_WIDTH) - 1));
2851			return 0;
2852		}
2853
2854		/* Merged completion for multiple non-scattered packets */
2855		rx_queue->scatter_n = 1;
2856		rx_queue->scatter_len = 0;
2857		n_packets = n_descs;
2858		++channel->n_rx_merge_events;
2859		channel->n_rx_merge_packets += n_packets;
2860		flags |= EFX_RX_PKT_PREFIX_LEN;
2861	} else {
2862		++rx_queue->scatter_n;
2863		rx_queue->scatter_len += rx_bytes;
2864		if (rx_cont)
2865			return 0;
2866		n_packets = 1;
2867	}
2868
2869	EFX_POPULATE_QWORD_5(errors, ESF_DZ_RX_ECRC_ERR, 1,
2870				     ESF_DZ_RX_IPCKSUM_ERR, 1,
2871				     ESF_DZ_RX_TCPUDP_CKSUM_ERR, 1,
2872				     ESF_EZ_RX_IP_INNER_CHKSUM_ERR, 1,
2873				     ESF_EZ_RX_TCP_UDP_INNER_CHKSUM_ERR, 1);
2874	EFX_AND_QWORD(errors, *event, errors);
2875	if (unlikely(!EFX_QWORD_IS_ZERO(errors))) {
2876		flags |= efx_ef10_handle_rx_event_errors(channel, n_packets,
2877							 rx_encap_hdr,
2878							 rx_l3_class, rx_l4_class,
2879							 event);
2880	} else {
2881		bool tcpudp = rx_l4_class == ESE_FZ_L4_CLASS_TCP ||
2882			      rx_l4_class == ESE_FZ_L4_CLASS_UDP;
2883
2884		switch (rx_encap_hdr) {
2885		case ESE_EZ_ENCAP_HDR_VXLAN: /* VxLAN or GENEVE */
2886			flags |= EFX_RX_PKT_CSUMMED; /* outer UDP csum */
2887			if (tcpudp)
2888				flags |= EFX_RX_PKT_CSUM_LEVEL; /* inner L4 */
2889			break;
2890		case ESE_EZ_ENCAP_HDR_GRE:
2891		case ESE_EZ_ENCAP_HDR_NONE:
2892			if (tcpudp)
2893				flags |= EFX_RX_PKT_CSUMMED;
2894			break;
2895		default:
2896			netdev_WARN(efx->net_dev,
2897				    "unknown encapsulation type: event="
2898				    EFX_QWORD_FMT "\n",
2899				    EFX_QWORD_VAL(*event));
2900		}
2901	}
2902
2903	if (rx_l4_class == ESE_FZ_L4_CLASS_TCP)
2904		flags |= EFX_RX_PKT_TCP;
2905
2906	channel->irq_mod_score += 2 * n_packets;
2907
2908	/* Handle received packet(s) */
2909	for (i = 0; i < n_packets; i++) {
2910		efx_rx_packet(rx_queue,
2911			      rx_queue->removed_count & rx_queue->ptr_mask,
2912			      rx_queue->scatter_n, rx_queue->scatter_len,
2913			      flags);
2914		rx_queue->removed_count += rx_queue->scatter_n;
2915	}
2916
2917	rx_queue->scatter_n = 0;
2918	rx_queue->scatter_len = 0;
2919
2920	return n_packets;
2921}
2922
2923static u32 efx_ef10_extract_event_ts(efx_qword_t *event)
2924{
2925	u32 tstamp;
2926
2927	tstamp = EFX_QWORD_FIELD(*event, TX_TIMESTAMP_EVENT_TSTAMP_DATA_HI);
2928	tstamp <<= 16;
2929	tstamp |= EFX_QWORD_FIELD(*event, TX_TIMESTAMP_EVENT_TSTAMP_DATA_LO);
2930
2931	return tstamp;
2932}
2933
2934static void
2935efx_ef10_handle_tx_event(struct efx_channel *channel, efx_qword_t *event)
2936{
2937	struct efx_nic *efx = channel->efx;
2938	struct efx_tx_queue *tx_queue;
2939	unsigned int tx_ev_desc_ptr;
2940	unsigned int tx_ev_q_label;
2941	unsigned int tx_ev_type;
2942	u64 ts_part;
2943
2944	if (unlikely(READ_ONCE(efx->reset_pending)))
2945		return;
2946
2947	if (unlikely(EFX_QWORD_FIELD(*event, ESF_DZ_TX_DROP_EVENT)))
2948		return;
2949
2950	/* Get the transmit queue */
2951	tx_ev_q_label = EFX_QWORD_FIELD(*event, ESF_DZ_TX_QLABEL);
2952	tx_queue = channel->tx_queue + (tx_ev_q_label % EFX_MAX_TXQ_PER_CHANNEL);
2953
2954	if (!tx_queue->timestamping) {
2955		/* Transmit completion */
2956		tx_ev_desc_ptr = EFX_QWORD_FIELD(*event, ESF_DZ_TX_DESCR_INDX);
2957		efx_xmit_done(tx_queue, tx_ev_desc_ptr & tx_queue->ptr_mask);
2958		return;
2959	}
2960
2961	/* Transmit timestamps are only available for 8XXX series. They result
2962	 * in up to three events per packet. These occur in order, and are:
2963	 *  - the normal completion event (may be omitted)
2964	 *  - the low part of the timestamp
2965	 *  - the high part of the timestamp
2966	 *
2967	 * It's possible for multiple completion events to appear before the
2968	 * corresponding timestamps. So we can for example get:
2969	 *  COMP N
2970	 *  COMP N+1
2971	 *  TS_LO N
2972	 *  TS_HI N
2973	 *  TS_LO N+1
2974	 *  TS_HI N+1
2975	 *
2976	 * In addition it's also possible for the adjacent completions to be
2977	 * merged, so we may not see COMP N above. As such, the completion
2978	 * events are not very useful here.
2979	 *
2980	 * Each part of the timestamp is itself split across two 16 bit
2981	 * fields in the event.
2982	 */
2983	tx_ev_type = EFX_QWORD_FIELD(*event, ESF_EZ_TX_SOFT1);
2984
2985	switch (tx_ev_type) {
2986	case TX_TIMESTAMP_EVENT_TX_EV_COMPLETION:
2987		/* Ignore this event - see above. */
2988		break;
2989
2990	case TX_TIMESTAMP_EVENT_TX_EV_TSTAMP_LO:
2991		ts_part = efx_ef10_extract_event_ts(event);
2992		tx_queue->completed_timestamp_minor = ts_part;
2993		break;
2994
2995	case TX_TIMESTAMP_EVENT_TX_EV_TSTAMP_HI:
2996		ts_part = efx_ef10_extract_event_ts(event);
2997		tx_queue->completed_timestamp_major = ts_part;
2998
2999		efx_xmit_done_single(tx_queue);
3000		break;
3001
3002	default:
3003		netif_err(efx, hw, efx->net_dev,
3004			  "channel %d unknown tx event type %d (data "
3005			  EFX_QWORD_FMT ")\n",
3006			  channel->channel, tx_ev_type,
3007			  EFX_QWORD_VAL(*event));
3008		break;
3009	}
3010}
3011
3012static void
3013efx_ef10_handle_driver_event(struct efx_channel *channel, efx_qword_t *event)
3014{
3015	struct efx_nic *efx = channel->efx;
3016	int subcode;
3017
3018	subcode = EFX_QWORD_FIELD(*event, ESF_DZ_DRV_SUB_CODE);
3019
3020	switch (subcode) {
3021	case ESE_DZ_DRV_TIMER_EV:
3022	case ESE_DZ_DRV_WAKE_UP_EV:
3023		break;
3024	case ESE_DZ_DRV_START_UP_EV:
3025		/* event queue init complete. ok. */
3026		break;
3027	default:
3028		netif_err(efx, hw, efx->net_dev,
3029			  "channel %d unknown driver event type %d"
3030			  " (data " EFX_QWORD_FMT ")\n",
3031			  channel->channel, subcode,
3032			  EFX_QWORD_VAL(*event));
3033
3034	}
3035}
3036
3037static void efx_ef10_handle_driver_generated_event(struct efx_channel *channel,
3038						   efx_qword_t *event)
3039{
3040	struct efx_nic *efx = channel->efx;
3041	u32 subcode;
3042
3043	subcode = EFX_QWORD_FIELD(*event, EFX_DWORD_0);
3044
3045	switch (subcode) {
3046	case EFX_EF10_TEST:
3047		channel->event_test_cpu = raw_smp_processor_id();
3048		break;
3049	case EFX_EF10_REFILL:
3050		/* The queue must be empty, so we won't receive any rx
3051		 * events, so efx_process_channel() won't refill the
3052		 * queue. Refill it here
3053		 */
3054		efx_fast_push_rx_descriptors(&channel->rx_queue, true);
3055		break;
3056	default:
3057		netif_err(efx, hw, efx->net_dev,
3058			  "channel %d unknown driver event type %u"
3059			  " (data " EFX_QWORD_FMT ")\n",
3060			  channel->channel, (unsigned) subcode,
3061			  EFX_QWORD_VAL(*event));
3062	}
3063}
3064
3065static int efx_ef10_ev_process(struct efx_channel *channel, int quota)
3066{
3067	struct efx_nic *efx = channel->efx;
3068	efx_qword_t event, *p_event;
3069	unsigned int read_ptr;
3070	int ev_code;
3071	int spent = 0;
3072
3073	if (quota <= 0)
3074		return spent;
3075
3076	read_ptr = channel->eventq_read_ptr;
3077
3078	for (;;) {
3079		p_event = efx_event(channel, read_ptr);
3080		event = *p_event;
3081
3082		if (!efx_event_present(&event))
3083			break;
3084
3085		EFX_SET_QWORD(*p_event);
3086
3087		++read_ptr;
3088
3089		ev_code = EFX_QWORD_FIELD(event, ESF_DZ_EV_CODE);
3090
3091		netif_vdbg(efx, drv, efx->net_dev,
3092			   "processing event on %d " EFX_QWORD_FMT "\n",
3093			   channel->channel, EFX_QWORD_VAL(event));
3094
3095		switch (ev_code) {
3096		case ESE_DZ_EV_CODE_MCDI_EV:
3097			efx_mcdi_process_event(channel, &event);
3098			break;
3099		case ESE_DZ_EV_CODE_RX_EV:
3100			spent += efx_ef10_handle_rx_event(channel, &event);
3101			if (spent >= quota) {
3102				/* XXX can we split a merged event to
3103				 * avoid going over-quota?
3104				 */
3105				spent = quota;
3106				goto out;
3107			}
3108			break;
3109		case ESE_DZ_EV_CODE_TX_EV:
3110			efx_ef10_handle_tx_event(channel, &event);
3111			break;
3112		case ESE_DZ_EV_CODE_DRIVER_EV:
3113			efx_ef10_handle_driver_event(channel, &event);
3114			if (++spent == quota)
3115				goto out;
3116			break;
3117		case EFX_EF10_DRVGEN_EV:
3118			efx_ef10_handle_driver_generated_event(channel, &event);
3119			break;
3120		default:
3121			netif_err(efx, hw, efx->net_dev,
3122				  "channel %d unknown event type %d"
3123				  " (data " EFX_QWORD_FMT ")\n",
3124				  channel->channel, ev_code,
3125				  EFX_QWORD_VAL(event));
3126		}
3127	}
3128
3129out:
3130	channel->eventq_read_ptr = read_ptr;
3131	return spent;
3132}
3133
3134static void efx_ef10_ev_read_ack(struct efx_channel *channel)
3135{
3136	struct efx_nic *efx = channel->efx;
3137	efx_dword_t rptr;
3138
3139	if (EFX_EF10_WORKAROUND_35388(efx)) {
3140		BUILD_BUG_ON(EFX_MIN_EVQ_SIZE <
3141			     (1 << ERF_DD_EVQ_IND_RPTR_WIDTH));
3142		BUILD_BUG_ON(EFX_MAX_EVQ_SIZE >
3143			     (1 << 2 * ERF_DD_EVQ_IND_RPTR_WIDTH));
3144
3145		EFX_POPULATE_DWORD_2(rptr, ERF_DD_EVQ_IND_RPTR_FLAGS,
3146				     EFE_DD_EVQ_IND_RPTR_FLAGS_HIGH,
3147				     ERF_DD_EVQ_IND_RPTR,
3148				     (channel->eventq_read_ptr &
3149				      channel->eventq_mask) >>
3150				     ERF_DD_EVQ_IND_RPTR_WIDTH);
3151		efx_writed_page(efx, &rptr, ER_DD_EVQ_INDIRECT,
3152				channel->channel);
3153		EFX_POPULATE_DWORD_2(rptr, ERF_DD_EVQ_IND_RPTR_FLAGS,
3154				     EFE_DD_EVQ_IND_RPTR_FLAGS_LOW,
3155				     ERF_DD_EVQ_IND_RPTR,
3156				     channel->eventq_read_ptr &
3157				     ((1 << ERF_DD_EVQ_IND_RPTR_WIDTH) - 1));
3158		efx_writed_page(efx, &rptr, ER_DD_EVQ_INDIRECT,
3159				channel->channel);
3160	} else {
3161		EFX_POPULATE_DWORD_1(rptr, ERF_DZ_EVQ_RPTR,
3162				     channel->eventq_read_ptr &
3163				     channel->eventq_mask);
3164		efx_writed_page(efx, &rptr, ER_DZ_EVQ_RPTR, channel->channel);
3165	}
3166}
3167
3168static void efx_ef10_ev_test_generate(struct efx_channel *channel)
3169{
3170	MCDI_DECLARE_BUF(inbuf, MC_CMD_DRIVER_EVENT_IN_LEN);
3171	struct efx_nic *efx = channel->efx;
3172	efx_qword_t event;
3173	int rc;
3174
3175	EFX_POPULATE_QWORD_2(event,
3176			     ESF_DZ_EV_CODE, EFX_EF10_DRVGEN_EV,
3177			     ESF_DZ_EV_DATA, EFX_EF10_TEST);
3178
3179	MCDI_SET_DWORD(inbuf, DRIVER_EVENT_IN_EVQ, channel->channel);
3180
3181	/* MCDI_SET_QWORD is not appropriate here since EFX_POPULATE_* has
3182	 * already swapped the data to little-endian order.
3183	 */
3184	memcpy(MCDI_PTR(inbuf, DRIVER_EVENT_IN_DATA), &event.u64[0],
3185	       sizeof(efx_qword_t));
3186
3187	rc = efx_mcdi_rpc(efx, MC_CMD_DRIVER_EVENT, inbuf, sizeof(inbuf),
3188			  NULL, 0, NULL);
3189	if (rc != 0)
3190		goto fail;
3191
3192	return;
3193
3194fail:
3195	WARN_ON(true);
3196	netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
3197}
3198
3199static void efx_ef10_prepare_flr(struct efx_nic *efx)
3200{
3201	atomic_set(&efx->active_queues, 0);
3202}
3203
3204static int efx_ef10_vport_set_mac_address(struct efx_nic *efx)
3205{
3206	struct efx_ef10_nic_data *nic_data = efx->nic_data;
3207	u8 mac_old[ETH_ALEN];
3208	int rc, rc2;
3209
3210	/* Only reconfigure a PF-created vport */
3211	if (is_zero_ether_addr(nic_data->vport_mac))
3212		return 0;
3213
3214	efx_device_detach_sync(efx);
3215	efx_net_stop(efx->net_dev);
3216	down_write(&efx->filter_sem);
3217	efx_mcdi_filter_table_remove(efx);
3218	up_write(&efx->filter_sem);
3219
3220	rc = efx_ef10_vadaptor_free(efx, efx->vport_id);
3221	if (rc)
3222		goto restore_filters;
3223
3224	ether_addr_copy(mac_old, nic_data->vport_mac);
3225	rc = efx_ef10_vport_del_mac(efx, efx->vport_id,
3226				    nic_data->vport_mac);
3227	if (rc)
3228		goto restore_vadaptor;
3229
3230	rc = efx_ef10_vport_add_mac(efx, efx->vport_id,
3231				    efx->net_dev->dev_addr);
3232	if (!rc) {
3233		ether_addr_copy(nic_data->vport_mac, efx->net_dev->dev_addr);
3234	} else {
3235		rc2 = efx_ef10_vport_add_mac(efx, efx->vport_id, mac_old);
3236		if (rc2) {
3237			/* Failed to add original MAC, so clear vport_mac */
3238			eth_zero_addr(nic_data->vport_mac);
3239			goto reset_nic;
3240		}
3241	}
3242
3243restore_vadaptor:
3244	rc2 = efx_ef10_vadaptor_alloc(efx, efx->vport_id);
3245	if (rc2)
3246		goto reset_nic;
3247restore_filters:
3248	down_write(&efx->filter_sem);
3249	rc2 = efx_ef10_filter_table_probe(efx);
3250	up_write(&efx->filter_sem);
3251	if (rc2)
3252		goto reset_nic;
3253
3254	rc2 = efx_net_open(efx->net_dev);
3255	if (rc2)
3256		goto reset_nic;
3257
3258	efx_device_attach_if_not_resetting(efx);
3259
3260	return rc;
3261
3262reset_nic:
3263	netif_err(efx, drv, efx->net_dev,
3264		  "Failed to restore when changing MAC address - scheduling reset\n");
3265	efx_schedule_reset(efx, RESET_TYPE_DATAPATH);
3266
3267	return rc ? rc : rc2;
3268}
3269
3270static int efx_ef10_set_mac_address(struct efx_nic *efx)
3271{
3272	MCDI_DECLARE_BUF(inbuf, MC_CMD_VADAPTOR_SET_MAC_IN_LEN);
3273	bool was_enabled = efx->port_enabled;
3274	int rc;
3275
3276#ifdef CONFIG_SFC_SRIOV
3277	/* If this function is a VF and we have access to the parent PF,
3278	 * then use the PF control path to attempt to change the VF MAC address.
3279	 */
3280	if (efx->pci_dev->is_virtfn && efx->pci_dev->physfn) {
3281		struct efx_nic *efx_pf = pci_get_drvdata(efx->pci_dev->physfn);
3282		struct efx_ef10_nic_data *nic_data = efx->nic_data;
3283		u8 mac[ETH_ALEN];
3284
3285		/* net_dev->dev_addr can be zeroed by efx_net_stop in
3286		 * efx_ef10_sriov_set_vf_mac, so pass in a copy.
3287		 */
3288		ether_addr_copy(mac, efx->net_dev->dev_addr);
3289
3290		rc = efx_ef10_sriov_set_vf_mac(efx_pf, nic_data->vf_index, mac);
3291		if (!rc)
3292			return 0;
3293
3294		netif_dbg(efx, drv, efx->net_dev,
3295			  "Updating VF mac via PF failed (%d), setting directly\n",
3296			  rc);
3297	}
3298#endif
3299
3300	efx_device_detach_sync(efx);
3301	efx_net_stop(efx->net_dev);
3302
3303	mutex_lock(&efx->mac_lock);
3304	down_write(&efx->filter_sem);
3305	efx_mcdi_filter_table_remove(efx);
3306
3307	ether_addr_copy(MCDI_PTR(inbuf, VADAPTOR_SET_MAC_IN_MACADDR),
3308			efx->net_dev->dev_addr);
3309	MCDI_SET_DWORD(inbuf, VADAPTOR_SET_MAC_IN_UPSTREAM_PORT_ID,
3310		       efx->vport_id);
3311	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_VADAPTOR_SET_MAC, inbuf,
3312				sizeof(inbuf), NULL, 0, NULL);
3313
3314	efx_ef10_filter_table_probe(efx);
3315	up_write(&efx->filter_sem);
3316	mutex_unlock(&efx->mac_lock);
3317
3318	if (was_enabled)
3319		efx_net_open(efx->net_dev);
3320	efx_device_attach_if_not_resetting(efx);
3321
3322	if (rc == -EPERM) {
3323		netif_err(efx, drv, efx->net_dev,
3324			  "Cannot change MAC address; use sfboot to enable"
3325			  " mac-spoofing on this interface\n");
3326	} else if (rc == -ENOSYS && !efx_ef10_is_vf(efx)) {
3327		/* If the active MCFW does not support MC_CMD_VADAPTOR_SET_MAC
3328		 * fall-back to the method of changing the MAC address on the
3329		 * vport.  This only applies to PFs because such versions of
3330		 * MCFW do not support VFs.
3331		 */
3332		rc = efx_ef10_vport_set_mac_address(efx);
3333	} else if (rc) {
3334		efx_mcdi_display_error(efx, MC_CMD_VADAPTOR_SET_MAC,
3335				       sizeof(inbuf), NULL, 0, rc);
3336	}
3337
3338	return rc;
3339}
3340
3341static int efx_ef10_mac_reconfigure(struct efx_nic *efx, bool mtu_only)
3342{
3343	WARN_ON(!mutex_is_locked(&efx->mac_lock));
3344
3345	efx_mcdi_filter_sync_rx_mode(efx);
3346
3347	if (mtu_only && efx_has_cap(efx, SET_MAC_ENHANCED))
3348		return efx_mcdi_set_mtu(efx);
3349	return efx_mcdi_set_mac(efx);
3350}
3351
3352static int efx_ef10_start_bist(struct efx_nic *efx, u32 bist_type)
3353{
3354	MCDI_DECLARE_BUF(inbuf, MC_CMD_START_BIST_IN_LEN);
3355
3356	MCDI_SET_DWORD(inbuf, START_BIST_IN_TYPE, bist_type);
3357	return efx_mcdi_rpc(efx, MC_CMD_START_BIST, inbuf, sizeof(inbuf),
3358			    NULL, 0, NULL);
3359}
3360
3361/* MC BISTs follow a different poll mechanism to phy BISTs.
3362 * The BIST is done in the poll handler on the MC, and the MCDI command
3363 * will block until the BIST is done.
3364 */
3365static int efx_ef10_poll_bist(struct efx_nic *efx)
3366{
3367	int rc;
3368	MCDI_DECLARE_BUF(outbuf, MC_CMD_POLL_BIST_OUT_LEN);
3369	size_t outlen;
3370	u32 result;
3371
3372	rc = efx_mcdi_rpc(efx, MC_CMD_POLL_BIST, NULL, 0,
3373			   outbuf, sizeof(outbuf), &outlen);
3374	if (rc != 0)
3375		return rc;
3376
3377	if (outlen < MC_CMD_POLL_BIST_OUT_LEN)
3378		return -EIO;
3379
3380	result = MCDI_DWORD(outbuf, POLL_BIST_OUT_RESULT);
3381	switch (result) {
3382	case MC_CMD_POLL_BIST_PASSED:
3383		netif_dbg(efx, hw, efx->net_dev, "BIST passed.\n");
3384		return 0;
3385	case MC_CMD_POLL_BIST_TIMEOUT:
3386		netif_err(efx, hw, efx->net_dev, "BIST timed out\n");
3387		return -EIO;
3388	case MC_CMD_POLL_BIST_FAILED:
3389		netif_err(efx, hw, efx->net_dev, "BIST failed.\n");
3390		return -EIO;
3391	default:
3392		netif_err(efx, hw, efx->net_dev,
3393			  "BIST returned unknown result %u", result);
3394		return -EIO;
3395	}
3396}
3397
3398static int efx_ef10_run_bist(struct efx_nic *efx, u32 bist_type)
3399{
3400	int rc;
3401
3402	netif_dbg(efx, drv, efx->net_dev, "starting BIST type %u\n", bist_type);
3403
3404	rc = efx_ef10_start_bist(efx, bist_type);
3405	if (rc != 0)
3406		return rc;
3407
3408	return efx_ef10_poll_bist(efx);
3409}
3410
3411static int
3412efx_ef10_test_chip(struct efx_nic *efx, struct efx_self_tests *tests)
3413{
3414	int rc, rc2;
3415
3416	efx_reset_down(efx, RESET_TYPE_WORLD);
3417
3418	rc = efx_mcdi_rpc(efx, MC_CMD_ENABLE_OFFLINE_BIST,
3419			  NULL, 0, NULL, 0, NULL);
3420	if (rc != 0)
3421		goto out;
3422
3423	tests->memory = efx_ef10_run_bist(efx, MC_CMD_MC_MEM_BIST) ? -1 : 1;
3424	tests->registers = efx_ef10_run_bist(efx, MC_CMD_REG_BIST) ? -1 : 1;
3425
3426	rc = efx_mcdi_reset(efx, RESET_TYPE_WORLD);
3427
3428out:
3429	if (rc == -EPERM)
3430		rc = 0;
3431	rc2 = efx_reset_up(efx, RESET_TYPE_WORLD, rc == 0);
3432	return rc ? rc : rc2;
3433}
3434
3435#ifdef CONFIG_SFC_MTD
3436
3437struct efx_ef10_nvram_type_info {
3438	u16 type, type_mask;
3439	u8 port;
3440	const char *name;
3441};
3442
3443static const struct efx_ef10_nvram_type_info efx_ef10_nvram_types[] = {
3444	{ NVRAM_PARTITION_TYPE_MC_FIRMWARE,	   0,    0, "sfc_mcfw" },
3445	{ NVRAM_PARTITION_TYPE_MC_FIRMWARE_BACKUP, 0,    0, "sfc_mcfw_backup" },
3446	{ NVRAM_PARTITION_TYPE_EXPANSION_ROM,	   0,    0, "sfc_exp_rom" },
3447	{ NVRAM_PARTITION_TYPE_STATIC_CONFIG,	   0,    0, "sfc_static_cfg" },
3448	{ NVRAM_PARTITION_TYPE_DYNAMIC_CONFIG,	   0,    0, "sfc_dynamic_cfg" },
3449	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT0, 0,   0, "sfc_exp_rom_cfg" },
3450	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT1, 0,   1, "sfc_exp_rom_cfg" },
3451	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT2, 0,   2, "sfc_exp_rom_cfg" },
3452	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT3, 0,   3, "sfc_exp_rom_cfg" },
3453	{ NVRAM_PARTITION_TYPE_LICENSE,		   0,    0, "sfc_license" },
3454	{ NVRAM_PARTITION_TYPE_PHY_MIN,		   0xff, 0, "sfc_phy_fw" },
3455	{ NVRAM_PARTITION_TYPE_MUM_FIRMWARE,	   0,    0, "sfc_mumfw" },
3456	{ NVRAM_PARTITION_TYPE_EXPANSION_UEFI,	   0,    0, "sfc_uefi" },
3457	{ NVRAM_PARTITION_TYPE_DYNCONFIG_DEFAULTS, 0,    0, "sfc_dynamic_cfg_dflt" },
3458	{ NVRAM_PARTITION_TYPE_ROMCONFIG_DEFAULTS, 0,    0, "sfc_exp_rom_cfg_dflt" },
3459	{ NVRAM_PARTITION_TYPE_STATUS,		   0,    0, "sfc_status" },
3460	{ NVRAM_PARTITION_TYPE_BUNDLE,		   0,    0, "sfc_bundle" },
3461	{ NVRAM_PARTITION_TYPE_BUNDLE_METADATA,	   0,    0, "sfc_bundle_metadata" },
3462};
3463#define EF10_NVRAM_PARTITION_COUNT	ARRAY_SIZE(efx_ef10_nvram_types)
3464
3465static int efx_ef10_mtd_probe_partition(struct efx_nic *efx,
3466					struct efx_mcdi_mtd_partition *part,
3467					unsigned int type,
3468					unsigned long *found)
3469{
3470	MCDI_DECLARE_BUF(inbuf, MC_CMD_NVRAM_METADATA_IN_LEN);
3471	MCDI_DECLARE_BUF(outbuf, MC_CMD_NVRAM_METADATA_OUT_LENMAX);
3472	const struct efx_ef10_nvram_type_info *info;
3473	size_t size, erase_size, outlen;
3474	int type_idx = 0;
3475	bool protected;
3476	int rc;
3477
3478	for (type_idx = 0; ; type_idx++) {
3479		if (type_idx == EF10_NVRAM_PARTITION_COUNT)
3480			return -ENODEV;
3481		info = efx_ef10_nvram_types + type_idx;
3482		if ((type & ~info->type_mask) == info->type)
3483			break;
3484	}
3485	if (info->port != efx_port_num(efx))
3486		return -ENODEV;
3487
3488	rc = efx_mcdi_nvram_info(efx, type, &size, &erase_size, &protected);
3489	if (rc)
3490		return rc;
3491	if (protected &&
3492	    (type != NVRAM_PARTITION_TYPE_DYNCONFIG_DEFAULTS &&
3493	     type != NVRAM_PARTITION_TYPE_ROMCONFIG_DEFAULTS))
3494		/* Hide protected partitions that don't provide defaults. */
3495		return -ENODEV;
3496
3497	if (protected)
3498		/* Protected partitions are read only. */
3499		erase_size = 0;
3500
3501	/* If we've already exposed a partition of this type, hide this
3502	 * duplicate.  All operations on MTDs are keyed by the type anyway,
3503	 * so we can't act on the duplicate.
3504	 */
3505	if (__test_and_set_bit(type_idx, found))
3506		return -EEXIST;
3507
3508	part->nvram_type = type;
3509
3510	MCDI_SET_DWORD(inbuf, NVRAM_METADATA_IN_TYPE, type);
3511	rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_METADATA, inbuf, sizeof(inbuf),
3512			  outbuf, sizeof(outbuf), &outlen);
3513	if (rc)
3514		return rc;
3515	if (outlen < MC_CMD_NVRAM_METADATA_OUT_LENMIN)
3516		return -EIO;
3517	if (MCDI_DWORD(outbuf, NVRAM_METADATA_OUT_FLAGS) &
3518	    (1 << MC_CMD_NVRAM_METADATA_OUT_SUBTYPE_VALID_LBN))
3519		part->fw_subtype = MCDI_DWORD(outbuf,
3520					      NVRAM_METADATA_OUT_SUBTYPE);
3521
3522	part->common.dev_type_name = "EF10 NVRAM manager";
3523	part->common.type_name = info->name;
3524
3525	part->common.mtd.type = MTD_NORFLASH;
3526	part->common.mtd.flags = MTD_CAP_NORFLASH;
3527	part->common.mtd.size = size;
3528	part->common.mtd.erasesize = erase_size;
3529	/* sfc_status is read-only */
3530	if (!erase_size)
3531		part->common.mtd.flags |= MTD_NO_ERASE;
3532
3533	return 0;
3534}
3535
3536static int efx_ef10_mtd_probe(struct efx_nic *efx)
3537{
3538	MCDI_DECLARE_BUF(outbuf, MC_CMD_NVRAM_PARTITIONS_OUT_LENMAX);
3539	DECLARE_BITMAP(found, EF10_NVRAM_PARTITION_COUNT) = { 0 };
3540	struct efx_mcdi_mtd_partition *parts;
3541	size_t outlen, n_parts_total, i, n_parts;
3542	unsigned int type;
3543	int rc;
3544
3545	ASSERT_RTNL();
3546
3547	BUILD_BUG_ON(MC_CMD_NVRAM_PARTITIONS_IN_LEN != 0);
3548	rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_PARTITIONS, NULL, 0,
3549			  outbuf, sizeof(outbuf), &outlen);
3550	if (rc)
3551		return rc;
3552	if (outlen < MC_CMD_NVRAM_PARTITIONS_OUT_LENMIN)
3553		return -EIO;
3554
3555	n_parts_total = MCDI_DWORD(outbuf, NVRAM_PARTITIONS_OUT_NUM_PARTITIONS);
3556	if (n_parts_total >
3557	    MCDI_VAR_ARRAY_LEN(outlen, NVRAM_PARTITIONS_OUT_TYPE_ID))
3558		return -EIO;
3559
3560	parts = kcalloc(n_parts_total, sizeof(*parts), GFP_KERNEL);
3561	if (!parts)
3562		return -ENOMEM;
3563
3564	n_parts = 0;
3565	for (i = 0; i < n_parts_total; i++) {
3566		type = MCDI_ARRAY_DWORD(outbuf, NVRAM_PARTITIONS_OUT_TYPE_ID,
3567					i);
3568		rc = efx_ef10_mtd_probe_partition(efx, &parts[n_parts], type,
3569						  found);
3570		if (rc == -EEXIST || rc == -ENODEV)
3571			continue;
3572		if (rc)
3573			goto fail;
3574		n_parts++;
3575	}
3576
3577	if (!n_parts) {
3578		kfree(parts);
3579		return 0;
3580	}
3581
3582	rc = efx_mtd_add(efx, &parts[0].common, n_parts, sizeof(*parts));
3583fail:
3584	if (rc)
3585		kfree(parts);
3586	return rc;
3587}
3588
3589#endif /* CONFIG_SFC_MTD */
3590
3591static void efx_ef10_ptp_write_host_time(struct efx_nic *efx, u32 host_time)
3592{
3593	_efx_writed(efx, cpu_to_le32(host_time), ER_DZ_MC_DB_LWRD);
3594}
3595
3596static void efx_ef10_ptp_write_host_time_vf(struct efx_nic *efx,
3597					    u32 host_time) {}
3598
3599static int efx_ef10_rx_enable_timestamping(struct efx_channel *channel,
3600					   bool temp)
3601{
3602	MCDI_DECLARE_BUF(inbuf, MC_CMD_PTP_IN_TIME_EVENT_SUBSCRIBE_LEN);
3603	int rc;
3604
3605	if (channel->sync_events_state == SYNC_EVENTS_REQUESTED ||
3606	    channel->sync_events_state == SYNC_EVENTS_VALID ||
3607	    (temp && channel->sync_events_state == SYNC_EVENTS_DISABLED))
3608		return 0;
3609	channel->sync_events_state = SYNC_EVENTS_REQUESTED;
3610
3611	MCDI_SET_DWORD(inbuf, PTP_IN_OP, MC_CMD_PTP_OP_TIME_EVENT_SUBSCRIBE);
3612	MCDI_SET_DWORD(inbuf, PTP_IN_PERIPH_ID, 0);
3613	MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_SUBSCRIBE_QUEUE,
3614		       channel->channel);
3615
3616	rc = efx_mcdi_rpc(channel->efx, MC_CMD_PTP,
3617			  inbuf, sizeof(inbuf), NULL, 0, NULL);
3618
3619	if (rc != 0)
3620		channel->sync_events_state = temp ? SYNC_EVENTS_QUIESCENT :
3621						    SYNC_EVENTS_DISABLED;
3622
3623	return rc;
3624}
3625
3626static int efx_ef10_rx_disable_timestamping(struct efx_channel *channel,
3627					    bool temp)
3628{
3629	MCDI_DECLARE_BUF(inbuf, MC_CMD_PTP_IN_TIME_EVENT_UNSUBSCRIBE_LEN);
3630	int rc;
3631
3632	if (channel->sync_events_state == SYNC_EVENTS_DISABLED ||
3633	    (temp && channel->sync_events_state == SYNC_EVENTS_QUIESCENT))
3634		return 0;
3635	if (channel->sync_events_state == SYNC_EVENTS_QUIESCENT) {
3636		channel->sync_events_state = SYNC_EVENTS_DISABLED;
3637		return 0;
3638	}
3639	channel->sync_events_state = temp ? SYNC_EVENTS_QUIESCENT :
3640					    SYNC_EVENTS_DISABLED;
3641
3642	MCDI_SET_DWORD(inbuf, PTP_IN_OP, MC_CMD_PTP_OP_TIME_EVENT_UNSUBSCRIBE);
3643	MCDI_SET_DWORD(inbuf, PTP_IN_PERIPH_ID, 0);
3644	MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_UNSUBSCRIBE_CONTROL,
3645		       MC_CMD_PTP_IN_TIME_EVENT_UNSUBSCRIBE_SINGLE);
3646	MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_UNSUBSCRIBE_QUEUE,
3647		       channel->channel);
3648
3649	rc = efx_mcdi_rpc(channel->efx, MC_CMD_PTP,
3650			  inbuf, sizeof(inbuf), NULL, 0, NULL);
3651
3652	return rc;
3653}
3654
3655static int efx_ef10_ptp_set_ts_sync_events(struct efx_nic *efx, bool en,
3656					   bool temp)
3657{
3658	int (*set)(struct efx_channel *channel, bool temp);
3659	struct efx_channel *channel;
3660
3661	set = en ?
3662	      efx_ef10_rx_enable_timestamping :
3663	      efx_ef10_rx_disable_timestamping;
3664
3665	channel = efx_ptp_channel(efx);
3666	if (channel) {
3667		int rc = set(channel, temp);
3668		if (en && rc != 0) {
3669			efx_ef10_ptp_set_ts_sync_events(efx, false, temp);
3670			return rc;
3671		}
3672	}
3673
3674	return 0;
3675}
3676
3677static int efx_ef10_ptp_set_ts_config_vf(struct efx_nic *efx,
3678					 struct hwtstamp_config *init)
3679{
3680	return -EOPNOTSUPP;
3681}
3682
3683static int efx_ef10_ptp_set_ts_config(struct efx_nic *efx,
3684				      struct hwtstamp_config *init)
3685{
3686	int rc;
3687
3688	switch (init->rx_filter) {
3689	case HWTSTAMP_FILTER_NONE:
3690		efx_ef10_ptp_set_ts_sync_events(efx, false, false);
3691		/* if TX timestamping is still requested then leave PTP on */
3692		return efx_ptp_change_mode(efx,
3693					   init->tx_type != HWTSTAMP_TX_OFF, 0);
3694	case HWTSTAMP_FILTER_ALL:
3695	case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
3696	case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
3697	case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
3698	case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
3699	case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
3700	case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
3701	case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
3702	case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
3703	case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
3704	case HWTSTAMP_FILTER_PTP_V2_EVENT:
3705	case HWTSTAMP_FILTER_PTP_V2_SYNC:
3706	case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
3707	case HWTSTAMP_FILTER_NTP_ALL:
3708		init->rx_filter = HWTSTAMP_FILTER_ALL;
3709		rc = efx_ptp_change_mode(efx, true, 0);
3710		if (!rc)
3711			rc = efx_ef10_ptp_set_ts_sync_events(efx, true, false);
3712		if (rc)
3713			efx_ptp_change_mode(efx, false, 0);
3714		return rc;
3715	default:
3716		return -ERANGE;
3717	}
3718}
3719
3720static int efx_ef10_get_phys_port_id(struct efx_nic *efx,
3721				     struct netdev_phys_item_id *ppid)
3722{
3723	struct efx_ef10_nic_data *nic_data = efx->nic_data;
3724
3725	if (!is_valid_ether_addr(nic_data->port_id))
3726		return -EOPNOTSUPP;
3727
3728	ppid->id_len = ETH_ALEN;
3729	memcpy(ppid->id, nic_data->port_id, ppid->id_len);
3730
3731	return 0;
3732}
3733
3734static int efx_ef10_vlan_rx_add_vid(struct efx_nic *efx, __be16 proto, u16 vid)
3735{
3736	if (proto != htons(ETH_P_8021Q))
3737		return -EINVAL;
3738
3739	return efx_ef10_add_vlan(efx, vid);
3740}
3741
3742static int efx_ef10_vlan_rx_kill_vid(struct efx_nic *efx, __be16 proto, u16 vid)
3743{
3744	if (proto != htons(ETH_P_8021Q))
3745		return -EINVAL;
3746
3747	return efx_ef10_del_vlan(efx, vid);
3748}
3749
3750/* We rely on the MCDI wiping out our TX rings if it made any changes to the
3751 * ports table, ensuring that any TSO descriptors that were made on a now-
3752 * removed tunnel port will be blown away and won't break things when we try
3753 * to transmit them using the new ports table.
3754 */
3755static int efx_ef10_set_udp_tnl_ports(struct efx_nic *efx, bool unloading)
3756{
3757	struct efx_ef10_nic_data *nic_data = efx->nic_data;
3758	MCDI_DECLARE_BUF(inbuf, MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_IN_LENMAX);
3759	MCDI_DECLARE_BUF(outbuf, MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_OUT_LEN);
3760	bool will_reset = false;
3761	size_t num_entries = 0;
3762	size_t inlen, outlen;
3763	size_t i;
3764	int rc;
3765	efx_dword_t flags_and_num_entries;
3766
3767	WARN_ON(!mutex_is_locked(&nic_data->udp_tunnels_lock));
3768
3769	nic_data->udp_tunnels_dirty = false;
3770
3771	if (!(nic_data->datapath_caps &
3772	    (1 << MC_CMD_GET_CAPABILITIES_OUT_VXLAN_NVGRE_LBN))) {
3773		efx_device_attach_if_not_resetting(efx);
3774		return 0;
3775	}
3776
3777	BUILD_BUG_ON(ARRAY_SIZE(nic_data->udp_tunnels) >
3778		     MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_IN_ENTRIES_MAXNUM);
3779
3780	for (i = 0; i < ARRAY_SIZE(nic_data->udp_tunnels); ++i) {
3781		if (nic_data->udp_tunnels[i].type !=
3782		    TUNNEL_ENCAP_UDP_PORT_ENTRY_INVALID) {
3783			efx_dword_t entry;
3784
3785			EFX_POPULATE_DWORD_2(entry,
3786				TUNNEL_ENCAP_UDP_PORT_ENTRY_UDP_PORT,
3787					ntohs(nic_data->udp_tunnels[i].port),
3788				TUNNEL_ENCAP_UDP_PORT_ENTRY_PROTOCOL,
3789					nic_data->udp_tunnels[i].type);
3790			*_MCDI_ARRAY_DWORD(inbuf,
3791				SET_TUNNEL_ENCAP_UDP_PORTS_IN_ENTRIES,
3792				num_entries++) = entry;
3793		}
3794	}
3795
3796	BUILD_BUG_ON((MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_IN_NUM_ENTRIES_OFST -
3797		      MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_IN_FLAGS_OFST) * 8 !=
3798		     EFX_WORD_1_LBN);
3799	BUILD_BUG_ON(MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_IN_NUM_ENTRIES_LEN * 8 !=
3800		     EFX_WORD_1_WIDTH);
3801	EFX_POPULATE_DWORD_2(flags_and_num_entries,
3802			     MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_IN_UNLOADING,
3803				!!unloading,
3804			     EFX_WORD_1, num_entries);
3805	*_MCDI_DWORD(inbuf, SET_TUNNEL_ENCAP_UDP_PORTS_IN_FLAGS) =
3806		flags_and_num_entries;
3807
3808	inlen = MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_IN_LEN(num_entries);
3809
3810	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS,
3811				inbuf, inlen, outbuf, sizeof(outbuf), &outlen);
3812	if (rc == -EIO) {
3813		/* Most likely the MC rebooted due to another function also
3814		 * setting its tunnel port list. Mark the tunnel port list as
3815		 * dirty, so it will be pushed upon coming up from the reboot.
3816		 */
3817		nic_data->udp_tunnels_dirty = true;
3818		return 0;
3819	}
3820
3821	if (rc) {
3822		/* expected not available on unprivileged functions */
3823		if (rc != -EPERM)
3824			netif_warn(efx, drv, efx->net_dev,
3825				   "Unable to set UDP tunnel ports; rc=%d.\n", rc);
3826	} else if (MCDI_DWORD(outbuf, SET_TUNNEL_ENCAP_UDP_PORTS_OUT_FLAGS) &
3827		   (1 << MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_OUT_RESETTING_LBN)) {
3828		netif_info(efx, drv, efx->net_dev,
3829			   "Rebooting MC due to UDP tunnel port list change\n");
3830		will_reset = true;
3831		if (unloading)
3832			/* Delay for the MC reset to complete. This will make
3833			 * unloading other functions a bit smoother. This is a
3834			 * race, but the other unload will work whichever way
3835			 * it goes, this just avoids an unnecessary error
3836			 * message.
3837			 */
3838			msleep(100);
3839	}
3840	if (!will_reset && !unloading) {
3841		/* The caller will have detached, relying on the MC reset to
3842		 * trigger a re-attach.  Since there won't be an MC reset, we
3843		 * have to do the attach ourselves.
3844		 */
3845		efx_device_attach_if_not_resetting(efx);
3846	}
3847
3848	return rc;
3849}
3850
3851static int efx_ef10_udp_tnl_push_ports(struct efx_nic *efx)
3852{
3853	struct efx_ef10_nic_data *nic_data = efx->nic_data;
3854	int rc = 0;
3855
3856	mutex_lock(&nic_data->udp_tunnels_lock);
3857	if (nic_data->udp_tunnels_dirty) {
3858		/* Make sure all TX are stopped while we modify the table, else
3859		 * we might race against an efx_features_check().
3860		 */
3861		efx_device_detach_sync(efx);
3862		rc = efx_ef10_set_udp_tnl_ports(efx, false);
3863	}
3864	mutex_unlock(&nic_data->udp_tunnels_lock);
3865	return rc;
3866}
3867
3868static int efx_ef10_udp_tnl_set_port(struct net_device *dev,
3869				     unsigned int table, unsigned int entry,
3870				     struct udp_tunnel_info *ti)
3871{
3872	struct efx_nic *efx = netdev_priv(dev);
3873	struct efx_ef10_nic_data *nic_data;
3874	int efx_tunnel_type, rc;
3875
3876	if (ti->type == UDP_TUNNEL_TYPE_VXLAN)
3877		efx_tunnel_type = TUNNEL_ENCAP_UDP_PORT_ENTRY_VXLAN;
3878	else
3879		efx_tunnel_type = TUNNEL_ENCAP_UDP_PORT_ENTRY_GENEVE;
3880
3881	nic_data = efx->nic_data;
3882	if (!(nic_data->datapath_caps &
3883	      (1 << MC_CMD_GET_CAPABILITIES_OUT_VXLAN_NVGRE_LBN)))
3884		return -EOPNOTSUPP;
3885
3886	mutex_lock(&nic_data->udp_tunnels_lock);
3887	/* Make sure all TX are stopped while we add to the table, else we
3888	 * might race against an efx_features_check().
3889	 */
3890	efx_device_detach_sync(efx);
3891	nic_data->udp_tunnels[entry].type = efx_tunnel_type;
3892	nic_data->udp_tunnels[entry].port = ti->port;
3893	rc = efx_ef10_set_udp_tnl_ports(efx, false);
3894	mutex_unlock(&nic_data->udp_tunnels_lock);
3895
3896	return rc;
3897}
3898
3899/* Called under the TX lock with the TX queue running, hence no-one can be
3900 * in the middle of updating the UDP tunnels table.  However, they could
3901 * have tried and failed the MCDI, in which case they'll have set the dirty
3902 * flag before dropping their locks.
3903 */
3904static bool efx_ef10_udp_tnl_has_port(struct efx_nic *efx, __be16 port)
3905{
3906	struct efx_ef10_nic_data *nic_data = efx->nic_data;
3907	size_t i;
3908
3909	if (!(nic_data->datapath_caps &
3910	      (1 << MC_CMD_GET_CAPABILITIES_OUT_VXLAN_NVGRE_LBN)))
3911		return false;
3912
3913	if (nic_data->udp_tunnels_dirty)
3914		/* SW table may not match HW state, so just assume we can't
3915		 * use any UDP tunnel offloads.
3916		 */
3917		return false;
3918
3919	for (i = 0; i < ARRAY_SIZE(nic_data->udp_tunnels); ++i)
3920		if (nic_data->udp_tunnels[i].type !=
3921		    TUNNEL_ENCAP_UDP_PORT_ENTRY_INVALID &&
3922		    nic_data->udp_tunnels[i].port == port)
3923			return true;
3924
3925	return false;
3926}
3927
3928static int efx_ef10_udp_tnl_unset_port(struct net_device *dev,
3929				       unsigned int table, unsigned int entry,
3930				       struct udp_tunnel_info *ti)
3931{
3932	struct efx_nic *efx = netdev_priv(dev);
3933	struct efx_ef10_nic_data *nic_data;
3934	int rc;
3935
3936	nic_data = efx->nic_data;
3937
3938	mutex_lock(&nic_data->udp_tunnels_lock);
3939	/* Make sure all TX are stopped while we remove from the table, else we
3940	 * might race against an efx_features_check().
3941	 */
3942	efx_device_detach_sync(efx);
3943	nic_data->udp_tunnels[entry].type = TUNNEL_ENCAP_UDP_PORT_ENTRY_INVALID;
3944	nic_data->udp_tunnels[entry].port = 0;
3945	rc = efx_ef10_set_udp_tnl_ports(efx, false);
3946	mutex_unlock(&nic_data->udp_tunnels_lock);
3947
3948	return rc;
3949}
3950
3951static const struct udp_tunnel_nic_info efx_ef10_udp_tunnels = {
3952	.set_port	= efx_ef10_udp_tnl_set_port,
3953	.unset_port	= efx_ef10_udp_tnl_unset_port,
3954	.flags          = UDP_TUNNEL_NIC_INFO_MAY_SLEEP,
3955	.tables         = {
3956		{
3957			.n_entries = 16,
3958			.tunnel_types = UDP_TUNNEL_TYPE_VXLAN |
3959					UDP_TUNNEL_TYPE_GENEVE,
3960		},
3961	},
3962};
3963
3964/* EF10 may have multiple datapath firmware variants within a
3965 * single version.  Report which variants are running.
3966 */
3967static size_t efx_ef10_print_additional_fwver(struct efx_nic *efx, char *buf,
3968					      size_t len)
3969{
3970	struct efx_ef10_nic_data *nic_data = efx->nic_data;
3971
3972	return scnprintf(buf, len, " rx%x tx%x",
3973			 nic_data->rx_dpcpu_fw_id,
3974			 nic_data->tx_dpcpu_fw_id);
3975}
3976
3977static unsigned int ef10_check_caps(const struct efx_nic *efx,
3978				    u8 flag,
3979				    u32 offset)
3980{
3981	const struct efx_ef10_nic_data *nic_data = efx->nic_data;
3982
3983	switch (offset) {
3984	case(MC_CMD_GET_CAPABILITIES_V4_OUT_FLAGS1_OFST):
3985		return nic_data->datapath_caps & BIT_ULL(flag);
3986	case(MC_CMD_GET_CAPABILITIES_V4_OUT_FLAGS2_OFST):
3987		return nic_data->datapath_caps2 & BIT_ULL(flag);
3988	default:
3989		return 0;
3990	}
3991}
3992
3993#define EF10_OFFLOAD_FEATURES		\
3994	(NETIF_F_IP_CSUM |		\
3995	 NETIF_F_HW_VLAN_CTAG_FILTER |	\
3996	 NETIF_F_IPV6_CSUM |		\
3997	 NETIF_F_RXHASH |		\
3998	 NETIF_F_NTUPLE |		\
3999	 NETIF_F_SG |			\
4000	 NETIF_F_RXCSUM |		\
4001	 NETIF_F_RXALL)
4002
4003const struct efx_nic_type efx_hunt_a0_vf_nic_type = {
4004	.is_vf = true,
4005	.mem_bar = efx_ef10_vf_mem_bar,
4006	.mem_map_size = efx_ef10_mem_map_size,
4007	.probe = efx_ef10_probe_vf,
4008	.remove = efx_ef10_remove,
4009	.dimension_resources = efx_ef10_dimension_resources,
4010	.init = efx_ef10_init_nic,
4011	.fini = efx_ef10_fini_nic,
4012	.map_reset_reason = efx_ef10_map_reset_reason,
4013	.map_reset_flags = efx_ef10_map_reset_flags,
4014	.reset = efx_ef10_reset,
4015	.probe_port = efx_mcdi_port_probe,
4016	.remove_port = efx_mcdi_port_remove,
4017	.fini_dmaq = efx_fini_dmaq,
4018	.prepare_flr = efx_ef10_prepare_flr,
4019	.finish_flr = efx_port_dummy_op_void,
4020	.describe_stats = efx_ef10_describe_stats,
4021	.update_stats = efx_ef10_update_stats_vf,
4022	.update_stats_atomic = efx_ef10_update_stats_atomic_vf,
4023	.start_stats = efx_port_dummy_op_void,
4024	.pull_stats = efx_port_dummy_op_void,
4025	.stop_stats = efx_port_dummy_op_void,
4026	.push_irq_moderation = efx_ef10_push_irq_moderation,
4027	.reconfigure_mac = efx_ef10_mac_reconfigure,
4028	.check_mac_fault = efx_mcdi_mac_check_fault,
4029	.reconfigure_port = efx_mcdi_port_reconfigure,
4030	.get_wol = efx_ef10_get_wol_vf,
4031	.set_wol = efx_ef10_set_wol_vf,
4032	.resume_wol = efx_port_dummy_op_void,
4033	.mcdi_request = efx_ef10_mcdi_request,
4034	.mcdi_poll_response = efx_ef10_mcdi_poll_response,
4035	.mcdi_read_response = efx_ef10_mcdi_read_response,
4036	.mcdi_poll_reboot = efx_ef10_mcdi_poll_reboot,
4037	.mcdi_reboot_detected = efx_ef10_mcdi_reboot_detected,
4038	.irq_enable_master = efx_port_dummy_op_void,
4039	.irq_test_generate = efx_ef10_irq_test_generate,
4040	.irq_disable_non_ev = efx_port_dummy_op_void,
4041	.irq_handle_msi = efx_ef10_msi_interrupt,
4042	.irq_handle_legacy = efx_ef10_legacy_interrupt,
4043	.tx_probe = efx_ef10_tx_probe,
4044	.tx_init = efx_ef10_tx_init,
4045	.tx_remove = efx_mcdi_tx_remove,
4046	.tx_write = efx_ef10_tx_write,
4047	.tx_limit_len = efx_ef10_tx_limit_len,
4048	.tx_enqueue = __efx_enqueue_skb,
4049	.rx_push_rss_config = efx_mcdi_vf_rx_push_rss_config,
4050	.rx_pull_rss_config = efx_mcdi_rx_pull_rss_config,
4051	.rx_probe = efx_mcdi_rx_probe,
4052	.rx_init = efx_mcdi_rx_init,
4053	.rx_remove = efx_mcdi_rx_remove,
4054	.rx_write = efx_ef10_rx_write,
4055	.rx_defer_refill = efx_ef10_rx_defer_refill,
4056	.rx_packet = __efx_rx_packet,
4057	.ev_probe = efx_mcdi_ev_probe,
4058	.ev_init = efx_ef10_ev_init,
4059	.ev_fini = efx_mcdi_ev_fini,
4060	.ev_remove = efx_mcdi_ev_remove,
4061	.ev_process = efx_ef10_ev_process,
4062	.ev_read_ack = efx_ef10_ev_read_ack,
4063	.ev_test_generate = efx_ef10_ev_test_generate,
4064	.filter_table_probe = efx_ef10_filter_table_probe,
4065	.filter_table_restore = efx_mcdi_filter_table_restore,
4066	.filter_table_remove = efx_mcdi_filter_table_remove,
4067	.filter_update_rx_scatter = efx_mcdi_update_rx_scatter,
4068	.filter_insert = efx_mcdi_filter_insert,
4069	.filter_remove_safe = efx_mcdi_filter_remove_safe,
4070	.filter_get_safe = efx_mcdi_filter_get_safe,
4071	.filter_clear_rx = efx_mcdi_filter_clear_rx,
4072	.filter_count_rx_used = efx_mcdi_filter_count_rx_used,
4073	.filter_get_rx_id_limit = efx_mcdi_filter_get_rx_id_limit,
4074	.filter_get_rx_ids = efx_mcdi_filter_get_rx_ids,
4075#ifdef CONFIG_RFS_ACCEL
4076	.filter_rfs_expire_one = efx_mcdi_filter_rfs_expire_one,
4077#endif
4078#ifdef CONFIG_SFC_MTD
4079	.mtd_probe = efx_port_dummy_op_int,
4080#endif
4081	.ptp_write_host_time = efx_ef10_ptp_write_host_time_vf,
4082	.ptp_set_ts_config = efx_ef10_ptp_set_ts_config_vf,
4083	.vlan_rx_add_vid = efx_ef10_vlan_rx_add_vid,
4084	.vlan_rx_kill_vid = efx_ef10_vlan_rx_kill_vid,
4085#ifdef CONFIG_SFC_SRIOV
4086	.vswitching_probe = efx_ef10_vswitching_probe_vf,
4087	.vswitching_restore = efx_ef10_vswitching_restore_vf,
4088	.vswitching_remove = efx_ef10_vswitching_remove_vf,
4089#endif
4090	.get_mac_address = efx_ef10_get_mac_address_vf,
4091	.set_mac_address = efx_ef10_set_mac_address,
4092
4093	.get_phys_port_id = efx_ef10_get_phys_port_id,
4094	.revision = EFX_REV_HUNT_A0,
4095	.max_dma_mask = DMA_BIT_MASK(ESF_DZ_TX_KER_BUF_ADDR_WIDTH),
4096	.rx_prefix_size = ES_DZ_RX_PREFIX_SIZE,
4097	.rx_hash_offset = ES_DZ_RX_PREFIX_HASH_OFST,
4098	.rx_ts_offset = ES_DZ_RX_PREFIX_TSTAMP_OFST,
4099	.can_rx_scatter = true,
4100	.always_rx_scatter = true,
4101	.min_interrupt_mode = EFX_INT_MODE_MSIX,
4102	.timer_period_max = 1 << ERF_DD_EVQ_IND_TIMER_VAL_WIDTH,
4103	.offload_features = EF10_OFFLOAD_FEATURES,
4104	.mcdi_max_ver = 2,
4105	.max_rx_ip_filters = EFX_MCDI_FILTER_TBL_ROWS,
4106	.hwtstamp_filters = 1 << HWTSTAMP_FILTER_NONE |
4107			    1 << HWTSTAMP_FILTER_ALL,
4108	.rx_hash_key_size = 40,
4109	.check_caps = ef10_check_caps,
4110	.print_additional_fwver = efx_ef10_print_additional_fwver,
4111	.sensor_event = efx_mcdi_sensor_event,
4112};
4113
4114const struct efx_nic_type efx_hunt_a0_nic_type = {
4115	.is_vf = false,
4116	.mem_bar = efx_ef10_pf_mem_bar,
4117	.mem_map_size = efx_ef10_mem_map_size,
4118	.probe = efx_ef10_probe_pf,
4119	.remove = efx_ef10_remove,
4120	.dimension_resources = efx_ef10_dimension_resources,
4121	.init = efx_ef10_init_nic,
4122	.fini = efx_ef10_fini_nic,
4123	.map_reset_reason = efx_ef10_map_reset_reason,
4124	.map_reset_flags = efx_ef10_map_reset_flags,
4125	.reset = efx_ef10_reset,
4126	.probe_port = efx_mcdi_port_probe,
4127	.remove_port = efx_mcdi_port_remove,
4128	.fini_dmaq = efx_fini_dmaq,
4129	.prepare_flr = efx_ef10_prepare_flr,
4130	.finish_flr = efx_port_dummy_op_void,
4131	.describe_stats = efx_ef10_describe_stats,
4132	.update_stats = efx_ef10_update_stats_pf,
4133	.start_stats = efx_mcdi_mac_start_stats,
4134	.pull_stats = efx_mcdi_mac_pull_stats,
4135	.stop_stats = efx_mcdi_mac_stop_stats,
4136	.push_irq_moderation = efx_ef10_push_irq_moderation,
4137	.reconfigure_mac = efx_ef10_mac_reconfigure,
4138	.check_mac_fault = efx_mcdi_mac_check_fault,
4139	.reconfigure_port = efx_mcdi_port_reconfigure,
4140	.get_wol = efx_ef10_get_wol,
4141	.set_wol = efx_ef10_set_wol,
4142	.resume_wol = efx_port_dummy_op_void,
4143	.test_chip = efx_ef10_test_chip,
4144	.test_nvram = efx_mcdi_nvram_test_all,
4145	.mcdi_request = efx_ef10_mcdi_request,
4146	.mcdi_poll_response = efx_ef10_mcdi_poll_response,
4147	.mcdi_read_response = efx_ef10_mcdi_read_response,
4148	.mcdi_poll_reboot = efx_ef10_mcdi_poll_reboot,
4149	.mcdi_reboot_detected = efx_ef10_mcdi_reboot_detected,
4150	.irq_enable_master = efx_port_dummy_op_void,
4151	.irq_test_generate = efx_ef10_irq_test_generate,
4152	.irq_disable_non_ev = efx_port_dummy_op_void,
4153	.irq_handle_msi = efx_ef10_msi_interrupt,
4154	.irq_handle_legacy = efx_ef10_legacy_interrupt,
4155	.tx_probe = efx_ef10_tx_probe,
4156	.tx_init = efx_ef10_tx_init,
4157	.tx_remove = efx_mcdi_tx_remove,
4158	.tx_write = efx_ef10_tx_write,
4159	.tx_limit_len = efx_ef10_tx_limit_len,
4160	.tx_enqueue = __efx_enqueue_skb,
4161	.rx_push_rss_config = efx_mcdi_pf_rx_push_rss_config,
4162	.rx_pull_rss_config = efx_mcdi_rx_pull_rss_config,
4163	.rx_push_rss_context_config = efx_mcdi_rx_push_rss_context_config,
4164	.rx_pull_rss_context_config = efx_mcdi_rx_pull_rss_context_config,
4165	.rx_restore_rss_contexts = efx_mcdi_rx_restore_rss_contexts,
4166	.rx_probe = efx_mcdi_rx_probe,
4167	.rx_init = efx_mcdi_rx_init,
4168	.rx_remove = efx_mcdi_rx_remove,
4169	.rx_write = efx_ef10_rx_write,
4170	.rx_defer_refill = efx_ef10_rx_defer_refill,
4171	.rx_packet = __efx_rx_packet,
4172	.ev_probe = efx_mcdi_ev_probe,
4173	.ev_init = efx_ef10_ev_init,
4174	.ev_fini = efx_mcdi_ev_fini,
4175	.ev_remove = efx_mcdi_ev_remove,
4176	.ev_process = efx_ef10_ev_process,
4177	.ev_read_ack = efx_ef10_ev_read_ack,
4178	.ev_test_generate = efx_ef10_ev_test_generate,
4179	.filter_table_probe = efx_ef10_filter_table_probe,
4180	.filter_table_restore = efx_mcdi_filter_table_restore,
4181	.filter_table_remove = efx_mcdi_filter_table_remove,
4182	.filter_update_rx_scatter = efx_mcdi_update_rx_scatter,
4183	.filter_insert = efx_mcdi_filter_insert,
4184	.filter_remove_safe = efx_mcdi_filter_remove_safe,
4185	.filter_get_safe = efx_mcdi_filter_get_safe,
4186	.filter_clear_rx = efx_mcdi_filter_clear_rx,
4187	.filter_count_rx_used = efx_mcdi_filter_count_rx_used,
4188	.filter_get_rx_id_limit = efx_mcdi_filter_get_rx_id_limit,
4189	.filter_get_rx_ids = efx_mcdi_filter_get_rx_ids,
4190#ifdef CONFIG_RFS_ACCEL
4191	.filter_rfs_expire_one = efx_mcdi_filter_rfs_expire_one,
4192#endif
4193#ifdef CONFIG_SFC_MTD
4194	.mtd_probe = efx_ef10_mtd_probe,
4195	.mtd_rename = efx_mcdi_mtd_rename,
4196	.mtd_read = efx_mcdi_mtd_read,
4197	.mtd_erase = efx_mcdi_mtd_erase,
4198	.mtd_write = efx_mcdi_mtd_write,
4199	.mtd_sync = efx_mcdi_mtd_sync,
4200#endif
4201	.ptp_write_host_time = efx_ef10_ptp_write_host_time,
4202	.ptp_set_ts_sync_events = efx_ef10_ptp_set_ts_sync_events,
4203	.ptp_set_ts_config = efx_ef10_ptp_set_ts_config,
4204	.vlan_rx_add_vid = efx_ef10_vlan_rx_add_vid,
4205	.vlan_rx_kill_vid = efx_ef10_vlan_rx_kill_vid,
4206	.udp_tnl_push_ports = efx_ef10_udp_tnl_push_ports,
4207	.udp_tnl_has_port = efx_ef10_udp_tnl_has_port,
4208#ifdef CONFIG_SFC_SRIOV
4209	.sriov_configure = efx_ef10_sriov_configure,
4210	.sriov_init = efx_ef10_sriov_init,
4211	.sriov_fini = efx_ef10_sriov_fini,
4212	.sriov_wanted = efx_ef10_sriov_wanted,
4213	.sriov_reset = efx_ef10_sriov_reset,
4214	.sriov_flr = efx_ef10_sriov_flr,
4215	.sriov_set_vf_mac = efx_ef10_sriov_set_vf_mac,
4216	.sriov_set_vf_vlan = efx_ef10_sriov_set_vf_vlan,
4217	.sriov_set_vf_spoofchk = efx_ef10_sriov_set_vf_spoofchk,
4218	.sriov_get_vf_config = efx_ef10_sriov_get_vf_config,
4219	.sriov_set_vf_link_state = efx_ef10_sriov_set_vf_link_state,
4220	.vswitching_probe = efx_ef10_vswitching_probe_pf,
4221	.vswitching_restore = efx_ef10_vswitching_restore_pf,
4222	.vswitching_remove = efx_ef10_vswitching_remove_pf,
4223#endif
4224	.get_mac_address = efx_ef10_get_mac_address_pf,
4225	.set_mac_address = efx_ef10_set_mac_address,
4226	.tso_versions = efx_ef10_tso_versions,
4227
4228	.get_phys_port_id = efx_ef10_get_phys_port_id,
4229	.revision = EFX_REV_HUNT_A0,
4230	.max_dma_mask = DMA_BIT_MASK(ESF_DZ_TX_KER_BUF_ADDR_WIDTH),
4231	.rx_prefix_size = ES_DZ_RX_PREFIX_SIZE,
4232	.rx_hash_offset = ES_DZ_RX_PREFIX_HASH_OFST,
4233	.rx_ts_offset = ES_DZ_RX_PREFIX_TSTAMP_OFST,
4234	.can_rx_scatter = true,
4235	.always_rx_scatter = true,
4236	.option_descriptors = true,
4237	.min_interrupt_mode = EFX_INT_MODE_LEGACY,
4238	.timer_period_max = 1 << ERF_DD_EVQ_IND_TIMER_VAL_WIDTH,
4239	.offload_features = EF10_OFFLOAD_FEATURES,
4240	.mcdi_max_ver = 2,
4241	.max_rx_ip_filters = EFX_MCDI_FILTER_TBL_ROWS,
4242	.hwtstamp_filters = 1 << HWTSTAMP_FILTER_NONE |
4243			    1 << HWTSTAMP_FILTER_ALL,
4244	.rx_hash_key_size = 40,
4245	.check_caps = ef10_check_caps,
4246	.print_additional_fwver = efx_ef10_print_additional_fwver,
4247	.sensor_event = efx_mcdi_sensor_event,
4248};
4249