xref: /openbmc/linux/drivers/net/ethernet/sfc/ef10.c (revision 9dae47aba0a055f761176d9297371d5bb24289ec)
1 /****************************************************************************
2  * Driver for Solarflare network controllers and boards
3  * Copyright 2012-2013 Solarflare Communications Inc.
4  *
5  * This program is free software; you can redistribute it and/or modify it
6  * under the terms of the GNU General Public License version 2 as published
7  * by the Free Software Foundation, incorporated herein by reference.
8  */
9 
10 #include "net_driver.h"
11 #include "ef10_regs.h"
12 #include "io.h"
13 #include "mcdi.h"
14 #include "mcdi_pcol.h"
15 #include "nic.h"
16 #include "workarounds.h"
17 #include "selftest.h"
18 #include "ef10_sriov.h"
19 #include <linux/in.h>
20 #include <linux/jhash.h>
21 #include <linux/wait.h>
22 #include <linux/workqueue.h>
23 
24 /* Hardware control for EF10 architecture including 'Huntington'. */
25 
26 #define EFX_EF10_DRVGEN_EV		7
27 enum {
28 	EFX_EF10_TEST = 1,
29 	EFX_EF10_REFILL,
30 };
31 
32 /* The reserved RSS context value */
33 #define EFX_EF10_RSS_CONTEXT_INVALID	0xffffffff
34 /* The maximum size of a shared RSS context */
35 /* TODO: this should really be from the mcdi protocol export */
36 #define EFX_EF10_MAX_SHARED_RSS_CONTEXT_SIZE 64UL
37 
38 /* The filter table(s) are managed by firmware and we have write-only
39  * access.  When removing filters we must identify them to the
40  * firmware by a 64-bit handle, but this is too wide for Linux kernel
41  * interfaces (32-bit for RX NFC, 16-bit for RFS).  Also, we need to
42  * be able to tell in advance whether a requested insertion will
43  * replace an existing filter.  Therefore we maintain a software hash
44  * table, which should be at least as large as the hardware hash
45  * table.
46  *
47  * Huntington has a single 8K filter table shared between all filter
48  * types and both ports.
49  */
50 #define HUNT_FILTER_TBL_ROWS 8192
51 
52 #define EFX_EF10_FILTER_ID_INVALID 0xffff
53 
54 #define EFX_EF10_FILTER_DEV_UC_MAX	32
55 #define EFX_EF10_FILTER_DEV_MC_MAX	256
56 
57 /* VLAN list entry */
58 struct efx_ef10_vlan {
59 	struct list_head list;
60 	u16 vid;
61 };
62 
63 enum efx_ef10_default_filters {
64 	EFX_EF10_BCAST,
65 	EFX_EF10_UCDEF,
66 	EFX_EF10_MCDEF,
67 	EFX_EF10_VXLAN4_UCDEF,
68 	EFX_EF10_VXLAN4_MCDEF,
69 	EFX_EF10_VXLAN6_UCDEF,
70 	EFX_EF10_VXLAN6_MCDEF,
71 	EFX_EF10_NVGRE4_UCDEF,
72 	EFX_EF10_NVGRE4_MCDEF,
73 	EFX_EF10_NVGRE6_UCDEF,
74 	EFX_EF10_NVGRE6_MCDEF,
75 	EFX_EF10_GENEVE4_UCDEF,
76 	EFX_EF10_GENEVE4_MCDEF,
77 	EFX_EF10_GENEVE6_UCDEF,
78 	EFX_EF10_GENEVE6_MCDEF,
79 
80 	EFX_EF10_NUM_DEFAULT_FILTERS
81 };
82 
83 /* Per-VLAN filters information */
84 struct efx_ef10_filter_vlan {
85 	struct list_head list;
86 	u16 vid;
87 	u16 uc[EFX_EF10_FILTER_DEV_UC_MAX];
88 	u16 mc[EFX_EF10_FILTER_DEV_MC_MAX];
89 	u16 default_filters[EFX_EF10_NUM_DEFAULT_FILTERS];
90 };
91 
92 struct efx_ef10_dev_addr {
93 	u8 addr[ETH_ALEN];
94 };
95 
96 struct efx_ef10_filter_table {
97 /* The MCDI match masks supported by this fw & hw, in order of priority */
98 	u32 rx_match_mcdi_flags[
99 		MC_CMD_GET_PARSER_DISP_INFO_OUT_SUPPORTED_MATCHES_MAXNUM * 2];
100 	unsigned int rx_match_count;
101 
102 	struct {
103 		unsigned long spec;	/* pointer to spec plus flag bits */
104 /* BUSY flag indicates that an update is in progress.  AUTO_OLD is
105  * used to mark and sweep MAC filters for the device address lists.
106  */
107 #define EFX_EF10_FILTER_FLAG_BUSY	1UL
108 #define EFX_EF10_FILTER_FLAG_AUTO_OLD	2UL
109 #define EFX_EF10_FILTER_FLAGS		3UL
110 		u64 handle;		/* firmware handle */
111 	} *entry;
112 	wait_queue_head_t waitq;
113 /* Shadow of net_device address lists, guarded by mac_lock */
114 	struct efx_ef10_dev_addr dev_uc_list[EFX_EF10_FILTER_DEV_UC_MAX];
115 	struct efx_ef10_dev_addr dev_mc_list[EFX_EF10_FILTER_DEV_MC_MAX];
116 	int dev_uc_count;
117 	int dev_mc_count;
118 	bool uc_promisc;
119 	bool mc_promisc;
120 /* Whether in multicast promiscuous mode when last changed */
121 	bool mc_promisc_last;
122 	bool mc_overflow; /* Too many MC addrs; should always imply mc_promisc */
123 	bool vlan_filter;
124 	struct list_head vlan_list;
125 };
126 
127 /* An arbitrary search limit for the software hash table */
128 #define EFX_EF10_FILTER_SEARCH_LIMIT 200
129 
130 static void efx_ef10_rx_free_indir_table(struct efx_nic *efx);
131 static void efx_ef10_filter_table_remove(struct efx_nic *efx);
132 static int efx_ef10_filter_add_vlan(struct efx_nic *efx, u16 vid);
133 static void efx_ef10_filter_del_vlan_internal(struct efx_nic *efx,
134 					      struct efx_ef10_filter_vlan *vlan);
135 static void efx_ef10_filter_del_vlan(struct efx_nic *efx, u16 vid);
136 static int efx_ef10_set_udp_tnl_ports(struct efx_nic *efx, bool unloading);
137 
138 static u32 efx_ef10_filter_get_unsafe_id(u32 filter_id)
139 {
140 	WARN_ON_ONCE(filter_id == EFX_EF10_FILTER_ID_INVALID);
141 	return filter_id & (HUNT_FILTER_TBL_ROWS - 1);
142 }
143 
144 static unsigned int efx_ef10_filter_get_unsafe_pri(u32 filter_id)
145 {
146 	return filter_id / (HUNT_FILTER_TBL_ROWS * 2);
147 }
148 
149 static u32 efx_ef10_make_filter_id(unsigned int pri, u16 idx)
150 {
151 	return pri * HUNT_FILTER_TBL_ROWS * 2 + idx;
152 }
153 
154 static int efx_ef10_get_warm_boot_count(struct efx_nic *efx)
155 {
156 	efx_dword_t reg;
157 
158 	efx_readd(efx, &reg, ER_DZ_BIU_MC_SFT_STATUS);
159 	return EFX_DWORD_FIELD(reg, EFX_WORD_1) == 0xb007 ?
160 		EFX_DWORD_FIELD(reg, EFX_WORD_0) : -EIO;
161 }
162 
163 /* On all EF10s up to and including SFC9220 (Medford1), all PFs use BAR 0 for
164  * I/O space and BAR 2(&3) for memory.  On SFC9250 (Medford2), there is no I/O
165  * bar; PFs use BAR 0/1 for memory.
166  */
167 static unsigned int efx_ef10_pf_mem_bar(struct efx_nic *efx)
168 {
169 	switch (efx->pci_dev->device) {
170 	case 0x0b03: /* SFC9250 PF */
171 		return 0;
172 	default:
173 		return 2;
174 	}
175 }
176 
177 /* All VFs use BAR 0/1 for memory */
178 static unsigned int efx_ef10_vf_mem_bar(struct efx_nic *efx)
179 {
180 	return 0;
181 }
182 
183 static unsigned int efx_ef10_mem_map_size(struct efx_nic *efx)
184 {
185 	int bar;
186 
187 	bar = efx->type->mem_bar(efx);
188 	return resource_size(&efx->pci_dev->resource[bar]);
189 }
190 
191 static bool efx_ef10_is_vf(struct efx_nic *efx)
192 {
193 	return efx->type->is_vf;
194 }
195 
196 static int efx_ef10_get_pf_index(struct efx_nic *efx)
197 {
198 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_FUNCTION_INFO_OUT_LEN);
199 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
200 	size_t outlen;
201 	int rc;
202 
203 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_FUNCTION_INFO, NULL, 0, outbuf,
204 			  sizeof(outbuf), &outlen);
205 	if (rc)
206 		return rc;
207 	if (outlen < sizeof(outbuf))
208 		return -EIO;
209 
210 	nic_data->pf_index = MCDI_DWORD(outbuf, GET_FUNCTION_INFO_OUT_PF);
211 	return 0;
212 }
213 
214 #ifdef CONFIG_SFC_SRIOV
215 static int efx_ef10_get_vf_index(struct efx_nic *efx)
216 {
217 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_FUNCTION_INFO_OUT_LEN);
218 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
219 	size_t outlen;
220 	int rc;
221 
222 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_FUNCTION_INFO, NULL, 0, outbuf,
223 			  sizeof(outbuf), &outlen);
224 	if (rc)
225 		return rc;
226 	if (outlen < sizeof(outbuf))
227 		return -EIO;
228 
229 	nic_data->vf_index = MCDI_DWORD(outbuf, GET_FUNCTION_INFO_OUT_VF);
230 	return 0;
231 }
232 #endif
233 
234 static int efx_ef10_init_datapath_caps(struct efx_nic *efx)
235 {
236 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_CAPABILITIES_V4_OUT_LEN);
237 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
238 	size_t outlen;
239 	int rc;
240 
241 	BUILD_BUG_ON(MC_CMD_GET_CAPABILITIES_IN_LEN != 0);
242 
243 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_CAPABILITIES, NULL, 0,
244 			  outbuf, sizeof(outbuf), &outlen);
245 	if (rc)
246 		return rc;
247 	if (outlen < MC_CMD_GET_CAPABILITIES_OUT_LEN) {
248 		netif_err(efx, drv, efx->net_dev,
249 			  "unable to read datapath firmware capabilities\n");
250 		return -EIO;
251 	}
252 
253 	nic_data->datapath_caps =
254 		MCDI_DWORD(outbuf, GET_CAPABILITIES_OUT_FLAGS1);
255 
256 	if (outlen >= MC_CMD_GET_CAPABILITIES_V2_OUT_LEN) {
257 		nic_data->datapath_caps2 = MCDI_DWORD(outbuf,
258 				GET_CAPABILITIES_V2_OUT_FLAGS2);
259 		nic_data->piobuf_size = MCDI_WORD(outbuf,
260 				GET_CAPABILITIES_V2_OUT_SIZE_PIO_BUFF);
261 	} else {
262 		nic_data->datapath_caps2 = 0;
263 		nic_data->piobuf_size = ER_DZ_TX_PIOBUF_SIZE;
264 	}
265 
266 	/* record the DPCPU firmware IDs to determine VEB vswitching support.
267 	 */
268 	nic_data->rx_dpcpu_fw_id =
269 		MCDI_WORD(outbuf, GET_CAPABILITIES_OUT_RX_DPCPU_FW_ID);
270 	nic_data->tx_dpcpu_fw_id =
271 		MCDI_WORD(outbuf, GET_CAPABILITIES_OUT_TX_DPCPU_FW_ID);
272 
273 	if (!(nic_data->datapath_caps &
274 	      (1 << MC_CMD_GET_CAPABILITIES_OUT_RX_PREFIX_LEN_14_LBN))) {
275 		netif_err(efx, probe, efx->net_dev,
276 			  "current firmware does not support an RX prefix\n");
277 		return -ENODEV;
278 	}
279 
280 	if (outlen >= MC_CMD_GET_CAPABILITIES_V3_OUT_LEN) {
281 		u8 vi_window_mode = MCDI_BYTE(outbuf,
282 				GET_CAPABILITIES_V3_OUT_VI_WINDOW_MODE);
283 
284 		switch (vi_window_mode) {
285 		case MC_CMD_GET_CAPABILITIES_V3_OUT_VI_WINDOW_MODE_8K:
286 			efx->vi_stride = 8192;
287 			break;
288 		case MC_CMD_GET_CAPABILITIES_V3_OUT_VI_WINDOW_MODE_16K:
289 			efx->vi_stride = 16384;
290 			break;
291 		case MC_CMD_GET_CAPABILITIES_V3_OUT_VI_WINDOW_MODE_64K:
292 			efx->vi_stride = 65536;
293 			break;
294 		default:
295 			netif_err(efx, probe, efx->net_dev,
296 				  "Unrecognised VI window mode %d\n",
297 				  vi_window_mode);
298 			return -EIO;
299 		}
300 		netif_dbg(efx, probe, efx->net_dev, "vi_stride = %u\n",
301 			  efx->vi_stride);
302 	} else {
303 		/* keep default VI stride */
304 		netif_dbg(efx, probe, efx->net_dev,
305 			  "firmware did not report VI window mode, assuming vi_stride = %u\n",
306 			  efx->vi_stride);
307 	}
308 
309 	if (outlen >= MC_CMD_GET_CAPABILITIES_V4_OUT_LEN) {
310 		efx->num_mac_stats = MCDI_WORD(outbuf,
311 				GET_CAPABILITIES_V4_OUT_MAC_STATS_NUM_STATS);
312 		netif_dbg(efx, probe, efx->net_dev,
313 			  "firmware reports num_mac_stats = %u\n",
314 			  efx->num_mac_stats);
315 	} else {
316 		/* leave num_mac_stats as the default value, MC_CMD_MAC_NSTATS */
317 		netif_dbg(efx, probe, efx->net_dev,
318 			  "firmware did not report num_mac_stats, assuming %u\n",
319 			  efx->num_mac_stats);
320 	}
321 
322 	return 0;
323 }
324 
325 static int efx_ef10_get_sysclk_freq(struct efx_nic *efx)
326 {
327 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_CLOCK_OUT_LEN);
328 	int rc;
329 
330 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_CLOCK, NULL, 0,
331 			  outbuf, sizeof(outbuf), NULL);
332 	if (rc)
333 		return rc;
334 	rc = MCDI_DWORD(outbuf, GET_CLOCK_OUT_SYS_FREQ);
335 	return rc > 0 ? rc : -ERANGE;
336 }
337 
338 static int efx_ef10_get_timer_workarounds(struct efx_nic *efx)
339 {
340 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
341 	unsigned int implemented;
342 	unsigned int enabled;
343 	int rc;
344 
345 	nic_data->workaround_35388 = false;
346 	nic_data->workaround_61265 = false;
347 
348 	rc = efx_mcdi_get_workarounds(efx, &implemented, &enabled);
349 
350 	if (rc == -ENOSYS) {
351 		/* Firmware without GET_WORKAROUNDS - not a problem. */
352 		rc = 0;
353 	} else if (rc == 0) {
354 		/* Bug61265 workaround is always enabled if implemented. */
355 		if (enabled & MC_CMD_GET_WORKAROUNDS_OUT_BUG61265)
356 			nic_data->workaround_61265 = true;
357 
358 		if (enabled & MC_CMD_GET_WORKAROUNDS_OUT_BUG35388) {
359 			nic_data->workaround_35388 = true;
360 		} else if (implemented & MC_CMD_GET_WORKAROUNDS_OUT_BUG35388) {
361 			/* Workaround is implemented but not enabled.
362 			 * Try to enable it.
363 			 */
364 			rc = efx_mcdi_set_workaround(efx,
365 						     MC_CMD_WORKAROUND_BUG35388,
366 						     true, NULL);
367 			if (rc == 0)
368 				nic_data->workaround_35388 = true;
369 			/* If we failed to set the workaround just carry on. */
370 			rc = 0;
371 		}
372 	}
373 
374 	netif_dbg(efx, probe, efx->net_dev,
375 		  "workaround for bug 35388 is %sabled\n",
376 		  nic_data->workaround_35388 ? "en" : "dis");
377 	netif_dbg(efx, probe, efx->net_dev,
378 		  "workaround for bug 61265 is %sabled\n",
379 		  nic_data->workaround_61265 ? "en" : "dis");
380 
381 	return rc;
382 }
383 
384 static void efx_ef10_process_timer_config(struct efx_nic *efx,
385 					  const efx_dword_t *data)
386 {
387 	unsigned int max_count;
388 
389 	if (EFX_EF10_WORKAROUND_61265(efx)) {
390 		efx->timer_quantum_ns = MCDI_DWORD(data,
391 			GET_EVQ_TMR_PROPERTIES_OUT_MCDI_TMR_STEP_NS);
392 		efx->timer_max_ns = MCDI_DWORD(data,
393 			GET_EVQ_TMR_PROPERTIES_OUT_MCDI_TMR_MAX_NS);
394 	} else if (EFX_EF10_WORKAROUND_35388(efx)) {
395 		efx->timer_quantum_ns = MCDI_DWORD(data,
396 			GET_EVQ_TMR_PROPERTIES_OUT_BUG35388_TMR_NS_PER_COUNT);
397 		max_count = MCDI_DWORD(data,
398 			GET_EVQ_TMR_PROPERTIES_OUT_BUG35388_TMR_MAX_COUNT);
399 		efx->timer_max_ns = max_count * efx->timer_quantum_ns;
400 	} else {
401 		efx->timer_quantum_ns = MCDI_DWORD(data,
402 			GET_EVQ_TMR_PROPERTIES_OUT_TMR_REG_NS_PER_COUNT);
403 		max_count = MCDI_DWORD(data,
404 			GET_EVQ_TMR_PROPERTIES_OUT_TMR_REG_MAX_COUNT);
405 		efx->timer_max_ns = max_count * efx->timer_quantum_ns;
406 	}
407 
408 	netif_dbg(efx, probe, efx->net_dev,
409 		  "got timer properties from MC: quantum %u ns; max %u ns\n",
410 		  efx->timer_quantum_ns, efx->timer_max_ns);
411 }
412 
413 static int efx_ef10_get_timer_config(struct efx_nic *efx)
414 {
415 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_EVQ_TMR_PROPERTIES_OUT_LEN);
416 	int rc;
417 
418 	rc = efx_ef10_get_timer_workarounds(efx);
419 	if (rc)
420 		return rc;
421 
422 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_GET_EVQ_TMR_PROPERTIES, NULL, 0,
423 				outbuf, sizeof(outbuf), NULL);
424 
425 	if (rc == 0) {
426 		efx_ef10_process_timer_config(efx, outbuf);
427 	} else if (rc == -ENOSYS || rc == -EPERM) {
428 		/* Not available - fall back to Huntington defaults. */
429 		unsigned int quantum;
430 
431 		rc = efx_ef10_get_sysclk_freq(efx);
432 		if (rc < 0)
433 			return rc;
434 
435 		quantum = 1536000 / rc; /* 1536 cycles */
436 		efx->timer_quantum_ns = quantum;
437 		efx->timer_max_ns = efx->type->timer_period_max * quantum;
438 		rc = 0;
439 	} else {
440 		efx_mcdi_display_error(efx, MC_CMD_GET_EVQ_TMR_PROPERTIES,
441 				       MC_CMD_GET_EVQ_TMR_PROPERTIES_OUT_LEN,
442 				       NULL, 0, rc);
443 	}
444 
445 	return rc;
446 }
447 
448 static int efx_ef10_get_mac_address_pf(struct efx_nic *efx, u8 *mac_address)
449 {
450 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_MAC_ADDRESSES_OUT_LEN);
451 	size_t outlen;
452 	int rc;
453 
454 	BUILD_BUG_ON(MC_CMD_GET_MAC_ADDRESSES_IN_LEN != 0);
455 
456 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_MAC_ADDRESSES, NULL, 0,
457 			  outbuf, sizeof(outbuf), &outlen);
458 	if (rc)
459 		return rc;
460 	if (outlen < MC_CMD_GET_MAC_ADDRESSES_OUT_LEN)
461 		return -EIO;
462 
463 	ether_addr_copy(mac_address,
464 			MCDI_PTR(outbuf, GET_MAC_ADDRESSES_OUT_MAC_ADDR_BASE));
465 	return 0;
466 }
467 
468 static int efx_ef10_get_mac_address_vf(struct efx_nic *efx, u8 *mac_address)
469 {
470 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VPORT_GET_MAC_ADDRESSES_IN_LEN);
471 	MCDI_DECLARE_BUF(outbuf, MC_CMD_VPORT_GET_MAC_ADDRESSES_OUT_LENMAX);
472 	size_t outlen;
473 	int num_addrs, rc;
474 
475 	MCDI_SET_DWORD(inbuf, VPORT_GET_MAC_ADDRESSES_IN_VPORT_ID,
476 		       EVB_PORT_ID_ASSIGNED);
477 	rc = efx_mcdi_rpc(efx, MC_CMD_VPORT_GET_MAC_ADDRESSES, inbuf,
478 			  sizeof(inbuf), outbuf, sizeof(outbuf), &outlen);
479 
480 	if (rc)
481 		return rc;
482 	if (outlen < MC_CMD_VPORT_GET_MAC_ADDRESSES_OUT_LENMIN)
483 		return -EIO;
484 
485 	num_addrs = MCDI_DWORD(outbuf,
486 			       VPORT_GET_MAC_ADDRESSES_OUT_MACADDR_COUNT);
487 
488 	WARN_ON(num_addrs != 1);
489 
490 	ether_addr_copy(mac_address,
491 			MCDI_PTR(outbuf, VPORT_GET_MAC_ADDRESSES_OUT_MACADDR));
492 
493 	return 0;
494 }
495 
496 static ssize_t efx_ef10_show_link_control_flag(struct device *dev,
497 					       struct device_attribute *attr,
498 					       char *buf)
499 {
500 	struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));
501 
502 	return sprintf(buf, "%d\n",
503 		       ((efx->mcdi->fn_flags) &
504 			(1 << MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_LINKCTRL))
505 		       ? 1 : 0);
506 }
507 
508 static ssize_t efx_ef10_show_primary_flag(struct device *dev,
509 					  struct device_attribute *attr,
510 					  char *buf)
511 {
512 	struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));
513 
514 	return sprintf(buf, "%d\n",
515 		       ((efx->mcdi->fn_flags) &
516 			(1 << MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_PRIMARY))
517 		       ? 1 : 0);
518 }
519 
520 static struct efx_ef10_vlan *efx_ef10_find_vlan(struct efx_nic *efx, u16 vid)
521 {
522 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
523 	struct efx_ef10_vlan *vlan;
524 
525 	WARN_ON(!mutex_is_locked(&nic_data->vlan_lock));
526 
527 	list_for_each_entry(vlan, &nic_data->vlan_list, list) {
528 		if (vlan->vid == vid)
529 			return vlan;
530 	}
531 
532 	return NULL;
533 }
534 
535 static int efx_ef10_add_vlan(struct efx_nic *efx, u16 vid)
536 {
537 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
538 	struct efx_ef10_vlan *vlan;
539 	int rc;
540 
541 	mutex_lock(&nic_data->vlan_lock);
542 
543 	vlan = efx_ef10_find_vlan(efx, vid);
544 	if (vlan) {
545 		/* We add VID 0 on init. 8021q adds it on module init
546 		 * for all interfaces with VLAN filtring feature.
547 		 */
548 		if (vid == 0)
549 			goto done_unlock;
550 		netif_warn(efx, drv, efx->net_dev,
551 			   "VLAN %u already added\n", vid);
552 		rc = -EALREADY;
553 		goto fail_exist;
554 	}
555 
556 	rc = -ENOMEM;
557 	vlan = kzalloc(sizeof(*vlan), GFP_KERNEL);
558 	if (!vlan)
559 		goto fail_alloc;
560 
561 	vlan->vid = vid;
562 
563 	list_add_tail(&vlan->list, &nic_data->vlan_list);
564 
565 	if (efx->filter_state) {
566 		mutex_lock(&efx->mac_lock);
567 		down_write(&efx->filter_sem);
568 		rc = efx_ef10_filter_add_vlan(efx, vlan->vid);
569 		up_write(&efx->filter_sem);
570 		mutex_unlock(&efx->mac_lock);
571 		if (rc)
572 			goto fail_filter_add_vlan;
573 	}
574 
575 done_unlock:
576 	mutex_unlock(&nic_data->vlan_lock);
577 	return 0;
578 
579 fail_filter_add_vlan:
580 	list_del(&vlan->list);
581 	kfree(vlan);
582 fail_alloc:
583 fail_exist:
584 	mutex_unlock(&nic_data->vlan_lock);
585 	return rc;
586 }
587 
588 static void efx_ef10_del_vlan_internal(struct efx_nic *efx,
589 				       struct efx_ef10_vlan *vlan)
590 {
591 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
592 
593 	WARN_ON(!mutex_is_locked(&nic_data->vlan_lock));
594 
595 	if (efx->filter_state) {
596 		down_write(&efx->filter_sem);
597 		efx_ef10_filter_del_vlan(efx, vlan->vid);
598 		up_write(&efx->filter_sem);
599 	}
600 
601 	list_del(&vlan->list);
602 	kfree(vlan);
603 }
604 
605 static int efx_ef10_del_vlan(struct efx_nic *efx, u16 vid)
606 {
607 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
608 	struct efx_ef10_vlan *vlan;
609 	int rc = 0;
610 
611 	/* 8021q removes VID 0 on module unload for all interfaces
612 	 * with VLAN filtering feature. We need to keep it to receive
613 	 * untagged traffic.
614 	 */
615 	if (vid == 0)
616 		return 0;
617 
618 	mutex_lock(&nic_data->vlan_lock);
619 
620 	vlan = efx_ef10_find_vlan(efx, vid);
621 	if (!vlan) {
622 		netif_err(efx, drv, efx->net_dev,
623 			  "VLAN %u to be deleted not found\n", vid);
624 		rc = -ENOENT;
625 	} else {
626 		efx_ef10_del_vlan_internal(efx, vlan);
627 	}
628 
629 	mutex_unlock(&nic_data->vlan_lock);
630 
631 	return rc;
632 }
633 
634 static void efx_ef10_cleanup_vlans(struct efx_nic *efx)
635 {
636 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
637 	struct efx_ef10_vlan *vlan, *next_vlan;
638 
639 	mutex_lock(&nic_data->vlan_lock);
640 	list_for_each_entry_safe(vlan, next_vlan, &nic_data->vlan_list, list)
641 		efx_ef10_del_vlan_internal(efx, vlan);
642 	mutex_unlock(&nic_data->vlan_lock);
643 }
644 
645 static DEVICE_ATTR(link_control_flag, 0444, efx_ef10_show_link_control_flag,
646 		   NULL);
647 static DEVICE_ATTR(primary_flag, 0444, efx_ef10_show_primary_flag, NULL);
648 
649 static int efx_ef10_probe(struct efx_nic *efx)
650 {
651 	struct efx_ef10_nic_data *nic_data;
652 	int i, rc;
653 
654 	nic_data = kzalloc(sizeof(*nic_data), GFP_KERNEL);
655 	if (!nic_data)
656 		return -ENOMEM;
657 	efx->nic_data = nic_data;
658 
659 	/* we assume later that we can copy from this buffer in dwords */
660 	BUILD_BUG_ON(MCDI_CTL_SDU_LEN_MAX_V2 % 4);
661 
662 	rc = efx_nic_alloc_buffer(efx, &nic_data->mcdi_buf,
663 				  8 + MCDI_CTL_SDU_LEN_MAX_V2, GFP_KERNEL);
664 	if (rc)
665 		goto fail1;
666 
667 	/* Get the MC's warm boot count.  In case it's rebooting right
668 	 * now, be prepared to retry.
669 	 */
670 	i = 0;
671 	for (;;) {
672 		rc = efx_ef10_get_warm_boot_count(efx);
673 		if (rc >= 0)
674 			break;
675 		if (++i == 5)
676 			goto fail2;
677 		ssleep(1);
678 	}
679 	nic_data->warm_boot_count = rc;
680 
681 	nic_data->rx_rss_context = EFX_EF10_RSS_CONTEXT_INVALID;
682 
683 	nic_data->vport_id = EVB_PORT_ID_ASSIGNED;
684 
685 	/* In case we're recovering from a crash (kexec), we want to
686 	 * cancel any outstanding request by the previous user of this
687 	 * function.  We send a special message using the least
688 	 * significant bits of the 'high' (doorbell) register.
689 	 */
690 	_efx_writed(efx, cpu_to_le32(1), ER_DZ_MC_DB_HWRD);
691 
692 	rc = efx_mcdi_init(efx);
693 	if (rc)
694 		goto fail2;
695 
696 	mutex_init(&nic_data->udp_tunnels_lock);
697 
698 	/* Reset (most) configuration for this function */
699 	rc = efx_mcdi_reset(efx, RESET_TYPE_ALL);
700 	if (rc)
701 		goto fail3;
702 
703 	/* Enable event logging */
704 	rc = efx_mcdi_log_ctrl(efx, true, false, 0);
705 	if (rc)
706 		goto fail3;
707 
708 	rc = device_create_file(&efx->pci_dev->dev,
709 				&dev_attr_link_control_flag);
710 	if (rc)
711 		goto fail3;
712 
713 	rc = device_create_file(&efx->pci_dev->dev, &dev_attr_primary_flag);
714 	if (rc)
715 		goto fail4;
716 
717 	rc = efx_ef10_get_pf_index(efx);
718 	if (rc)
719 		goto fail5;
720 
721 	rc = efx_ef10_init_datapath_caps(efx);
722 	if (rc < 0)
723 		goto fail5;
724 
725 	/* We can have one VI for each vi_stride-byte region.
726 	 * However, until we use TX option descriptors we need two TX queues
727 	 * per channel.
728 	 */
729 	efx->max_channels = min_t(unsigned int,
730 				  EFX_MAX_CHANNELS,
731 				  efx_ef10_mem_map_size(efx) /
732 				  (efx->vi_stride * EFX_TXQ_TYPES));
733 	efx->max_tx_channels = efx->max_channels;
734 	if (WARN_ON(efx->max_channels == 0)) {
735 		rc = -EIO;
736 		goto fail5;
737 	}
738 
739 	efx->rx_packet_len_offset =
740 		ES_DZ_RX_PREFIX_PKTLEN_OFST - ES_DZ_RX_PREFIX_SIZE;
741 
742 	if (nic_data->datapath_caps &
743 	    (1 << MC_CMD_GET_CAPABILITIES_OUT_RX_INCLUDE_FCS_LBN))
744 		efx->net_dev->hw_features |= NETIF_F_RXFCS;
745 
746 	rc = efx_mcdi_port_get_number(efx);
747 	if (rc < 0)
748 		goto fail5;
749 	efx->port_num = rc;
750 
751 	rc = efx->type->get_mac_address(efx, efx->net_dev->perm_addr);
752 	if (rc)
753 		goto fail5;
754 
755 	rc = efx_ef10_get_timer_config(efx);
756 	if (rc < 0)
757 		goto fail5;
758 
759 	rc = efx_mcdi_mon_probe(efx);
760 	if (rc && rc != -EPERM)
761 		goto fail5;
762 
763 	rc = efx_ptp_probe(efx, NULL);
764 	/* Failure to probe PTP is not fatal.
765 	 * In the case of EPERM, efx_ptp_probe will print its own message (in
766 	 * efx_ptp_get_attributes()), so we don't need to.
767 	 */
768 	if (rc && rc != -EPERM)
769 		netif_warn(efx, drv, efx->net_dev,
770 			   "Failed to probe PTP, rc=%d\n", rc);
771 
772 #ifdef CONFIG_SFC_SRIOV
773 	if ((efx->pci_dev->physfn) && (!efx->pci_dev->is_physfn)) {
774 		struct pci_dev *pci_dev_pf = efx->pci_dev->physfn;
775 		struct efx_nic *efx_pf = pci_get_drvdata(pci_dev_pf);
776 
777 		efx_pf->type->get_mac_address(efx_pf, nic_data->port_id);
778 	} else
779 #endif
780 		ether_addr_copy(nic_data->port_id, efx->net_dev->perm_addr);
781 
782 	INIT_LIST_HEAD(&nic_data->vlan_list);
783 	mutex_init(&nic_data->vlan_lock);
784 
785 	/* Add unspecified VID to support VLAN filtering being disabled */
786 	rc = efx_ef10_add_vlan(efx, EFX_FILTER_VID_UNSPEC);
787 	if (rc)
788 		goto fail_add_vid_unspec;
789 
790 	/* If VLAN filtering is enabled, we need VID 0 to get untagged
791 	 * traffic.  It is added automatically if 8021q module is loaded,
792 	 * but we can't rely on it since module may be not loaded.
793 	 */
794 	rc = efx_ef10_add_vlan(efx, 0);
795 	if (rc)
796 		goto fail_add_vid_0;
797 
798 	return 0;
799 
800 fail_add_vid_0:
801 	efx_ef10_cleanup_vlans(efx);
802 fail_add_vid_unspec:
803 	mutex_destroy(&nic_data->vlan_lock);
804 	efx_ptp_remove(efx);
805 	efx_mcdi_mon_remove(efx);
806 fail5:
807 	device_remove_file(&efx->pci_dev->dev, &dev_attr_primary_flag);
808 fail4:
809 	device_remove_file(&efx->pci_dev->dev, &dev_attr_link_control_flag);
810 fail3:
811 	efx_mcdi_detach(efx);
812 
813 	mutex_lock(&nic_data->udp_tunnels_lock);
814 	memset(nic_data->udp_tunnels, 0, sizeof(nic_data->udp_tunnels));
815 	(void)efx_ef10_set_udp_tnl_ports(efx, true);
816 	mutex_unlock(&nic_data->udp_tunnels_lock);
817 	mutex_destroy(&nic_data->udp_tunnels_lock);
818 
819 	efx_mcdi_fini(efx);
820 fail2:
821 	efx_nic_free_buffer(efx, &nic_data->mcdi_buf);
822 fail1:
823 	kfree(nic_data);
824 	efx->nic_data = NULL;
825 	return rc;
826 }
827 
828 static int efx_ef10_free_vis(struct efx_nic *efx)
829 {
830 	MCDI_DECLARE_BUF_ERR(outbuf);
831 	size_t outlen;
832 	int rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FREE_VIS, NULL, 0,
833 				    outbuf, sizeof(outbuf), &outlen);
834 
835 	/* -EALREADY means nothing to free, so ignore */
836 	if (rc == -EALREADY)
837 		rc = 0;
838 	if (rc)
839 		efx_mcdi_display_error(efx, MC_CMD_FREE_VIS, 0, outbuf, outlen,
840 				       rc);
841 	return rc;
842 }
843 
844 #ifdef EFX_USE_PIO
845 
846 static void efx_ef10_free_piobufs(struct efx_nic *efx)
847 {
848 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
849 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FREE_PIOBUF_IN_LEN);
850 	unsigned int i;
851 	int rc;
852 
853 	BUILD_BUG_ON(MC_CMD_FREE_PIOBUF_OUT_LEN != 0);
854 
855 	for (i = 0; i < nic_data->n_piobufs; i++) {
856 		MCDI_SET_DWORD(inbuf, FREE_PIOBUF_IN_PIOBUF_HANDLE,
857 			       nic_data->piobuf_handle[i]);
858 		rc = efx_mcdi_rpc(efx, MC_CMD_FREE_PIOBUF, inbuf, sizeof(inbuf),
859 				  NULL, 0, NULL);
860 		WARN_ON(rc);
861 	}
862 
863 	nic_data->n_piobufs = 0;
864 }
865 
866 static int efx_ef10_alloc_piobufs(struct efx_nic *efx, unsigned int n)
867 {
868 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
869 	MCDI_DECLARE_BUF(outbuf, MC_CMD_ALLOC_PIOBUF_OUT_LEN);
870 	unsigned int i;
871 	size_t outlen;
872 	int rc = 0;
873 
874 	BUILD_BUG_ON(MC_CMD_ALLOC_PIOBUF_IN_LEN != 0);
875 
876 	for (i = 0; i < n; i++) {
877 		rc = efx_mcdi_rpc_quiet(efx, MC_CMD_ALLOC_PIOBUF, NULL, 0,
878 					outbuf, sizeof(outbuf), &outlen);
879 		if (rc) {
880 			/* Don't display the MC error if we didn't have space
881 			 * for a VF.
882 			 */
883 			if (!(efx_ef10_is_vf(efx) && rc == -ENOSPC))
884 				efx_mcdi_display_error(efx, MC_CMD_ALLOC_PIOBUF,
885 						       0, outbuf, outlen, rc);
886 			break;
887 		}
888 		if (outlen < MC_CMD_ALLOC_PIOBUF_OUT_LEN) {
889 			rc = -EIO;
890 			break;
891 		}
892 		nic_data->piobuf_handle[i] =
893 			MCDI_DWORD(outbuf, ALLOC_PIOBUF_OUT_PIOBUF_HANDLE);
894 		netif_dbg(efx, probe, efx->net_dev,
895 			  "allocated PIO buffer %u handle %x\n", i,
896 			  nic_data->piobuf_handle[i]);
897 	}
898 
899 	nic_data->n_piobufs = i;
900 	if (rc)
901 		efx_ef10_free_piobufs(efx);
902 	return rc;
903 }
904 
905 static int efx_ef10_link_piobufs(struct efx_nic *efx)
906 {
907 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
908 	MCDI_DECLARE_BUF(inbuf, MC_CMD_LINK_PIOBUF_IN_LEN);
909 	struct efx_channel *channel;
910 	struct efx_tx_queue *tx_queue;
911 	unsigned int offset, index;
912 	int rc;
913 
914 	BUILD_BUG_ON(MC_CMD_LINK_PIOBUF_OUT_LEN != 0);
915 	BUILD_BUG_ON(MC_CMD_UNLINK_PIOBUF_OUT_LEN != 0);
916 
917 	/* Link a buffer to each VI in the write-combining mapping */
918 	for (index = 0; index < nic_data->n_piobufs; ++index) {
919 		MCDI_SET_DWORD(inbuf, LINK_PIOBUF_IN_PIOBUF_HANDLE,
920 			       nic_data->piobuf_handle[index]);
921 		MCDI_SET_DWORD(inbuf, LINK_PIOBUF_IN_TXQ_INSTANCE,
922 			       nic_data->pio_write_vi_base + index);
923 		rc = efx_mcdi_rpc(efx, MC_CMD_LINK_PIOBUF,
924 				  inbuf, MC_CMD_LINK_PIOBUF_IN_LEN,
925 				  NULL, 0, NULL);
926 		if (rc) {
927 			netif_err(efx, drv, efx->net_dev,
928 				  "failed to link VI %u to PIO buffer %u (%d)\n",
929 				  nic_data->pio_write_vi_base + index, index,
930 				  rc);
931 			goto fail;
932 		}
933 		netif_dbg(efx, probe, efx->net_dev,
934 			  "linked VI %u to PIO buffer %u\n",
935 			  nic_data->pio_write_vi_base + index, index);
936 	}
937 
938 	/* Link a buffer to each TX queue */
939 	efx_for_each_channel(channel, efx) {
940 		efx_for_each_channel_tx_queue(tx_queue, channel) {
941 			/* We assign the PIO buffers to queues in
942 			 * reverse order to allow for the following
943 			 * special case.
944 			 */
945 			offset = ((efx->tx_channel_offset + efx->n_tx_channels -
946 				   tx_queue->channel->channel - 1) *
947 				  efx_piobuf_size);
948 			index = offset / nic_data->piobuf_size;
949 			offset = offset % nic_data->piobuf_size;
950 
951 			/* When the host page size is 4K, the first
952 			 * host page in the WC mapping may be within
953 			 * the same VI page as the last TX queue.  We
954 			 * can only link one buffer to each VI.
955 			 */
956 			if (tx_queue->queue == nic_data->pio_write_vi_base) {
957 				BUG_ON(index != 0);
958 				rc = 0;
959 			} else {
960 				MCDI_SET_DWORD(inbuf,
961 					       LINK_PIOBUF_IN_PIOBUF_HANDLE,
962 					       nic_data->piobuf_handle[index]);
963 				MCDI_SET_DWORD(inbuf,
964 					       LINK_PIOBUF_IN_TXQ_INSTANCE,
965 					       tx_queue->queue);
966 				rc = efx_mcdi_rpc(efx, MC_CMD_LINK_PIOBUF,
967 						  inbuf, MC_CMD_LINK_PIOBUF_IN_LEN,
968 						  NULL, 0, NULL);
969 			}
970 
971 			if (rc) {
972 				/* This is non-fatal; the TX path just
973 				 * won't use PIO for this queue
974 				 */
975 				netif_err(efx, drv, efx->net_dev,
976 					  "failed to link VI %u to PIO buffer %u (%d)\n",
977 					  tx_queue->queue, index, rc);
978 				tx_queue->piobuf = NULL;
979 			} else {
980 				tx_queue->piobuf =
981 					nic_data->pio_write_base +
982 					index * efx->vi_stride + offset;
983 				tx_queue->piobuf_offset = offset;
984 				netif_dbg(efx, probe, efx->net_dev,
985 					  "linked VI %u to PIO buffer %u offset %x addr %p\n",
986 					  tx_queue->queue, index,
987 					  tx_queue->piobuf_offset,
988 					  tx_queue->piobuf);
989 			}
990 		}
991 	}
992 
993 	return 0;
994 
995 fail:
996 	/* inbuf was defined for MC_CMD_LINK_PIOBUF.  We can use the same
997 	 * buffer for MC_CMD_UNLINK_PIOBUF because it's shorter.
998 	 */
999 	BUILD_BUG_ON(MC_CMD_LINK_PIOBUF_IN_LEN < MC_CMD_UNLINK_PIOBUF_IN_LEN);
1000 	while (index--) {
1001 		MCDI_SET_DWORD(inbuf, UNLINK_PIOBUF_IN_TXQ_INSTANCE,
1002 			       nic_data->pio_write_vi_base + index);
1003 		efx_mcdi_rpc(efx, MC_CMD_UNLINK_PIOBUF,
1004 			     inbuf, MC_CMD_UNLINK_PIOBUF_IN_LEN,
1005 			     NULL, 0, NULL);
1006 	}
1007 	return rc;
1008 }
1009 
1010 static void efx_ef10_forget_old_piobufs(struct efx_nic *efx)
1011 {
1012 	struct efx_channel *channel;
1013 	struct efx_tx_queue *tx_queue;
1014 
1015 	/* All our existing PIO buffers went away */
1016 	efx_for_each_channel(channel, efx)
1017 		efx_for_each_channel_tx_queue(tx_queue, channel)
1018 			tx_queue->piobuf = NULL;
1019 }
1020 
1021 #else /* !EFX_USE_PIO */
1022 
1023 static int efx_ef10_alloc_piobufs(struct efx_nic *efx, unsigned int n)
1024 {
1025 	return n == 0 ? 0 : -ENOBUFS;
1026 }
1027 
1028 static int efx_ef10_link_piobufs(struct efx_nic *efx)
1029 {
1030 	return 0;
1031 }
1032 
1033 static void efx_ef10_free_piobufs(struct efx_nic *efx)
1034 {
1035 }
1036 
1037 static void efx_ef10_forget_old_piobufs(struct efx_nic *efx)
1038 {
1039 }
1040 
1041 #endif /* EFX_USE_PIO */
1042 
1043 static void efx_ef10_remove(struct efx_nic *efx)
1044 {
1045 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1046 	int rc;
1047 
1048 #ifdef CONFIG_SFC_SRIOV
1049 	struct efx_ef10_nic_data *nic_data_pf;
1050 	struct pci_dev *pci_dev_pf;
1051 	struct efx_nic *efx_pf;
1052 	struct ef10_vf *vf;
1053 
1054 	if (efx->pci_dev->is_virtfn) {
1055 		pci_dev_pf = efx->pci_dev->physfn;
1056 		if (pci_dev_pf) {
1057 			efx_pf = pci_get_drvdata(pci_dev_pf);
1058 			nic_data_pf = efx_pf->nic_data;
1059 			vf = nic_data_pf->vf + nic_data->vf_index;
1060 			vf->efx = NULL;
1061 		} else
1062 			netif_info(efx, drv, efx->net_dev,
1063 				   "Could not get the PF id from VF\n");
1064 	}
1065 #endif
1066 
1067 	efx_ef10_cleanup_vlans(efx);
1068 	mutex_destroy(&nic_data->vlan_lock);
1069 
1070 	efx_ptp_remove(efx);
1071 
1072 	efx_mcdi_mon_remove(efx);
1073 
1074 	efx_ef10_rx_free_indir_table(efx);
1075 
1076 	if (nic_data->wc_membase)
1077 		iounmap(nic_data->wc_membase);
1078 
1079 	rc = efx_ef10_free_vis(efx);
1080 	WARN_ON(rc != 0);
1081 
1082 	if (!nic_data->must_restore_piobufs)
1083 		efx_ef10_free_piobufs(efx);
1084 
1085 	device_remove_file(&efx->pci_dev->dev, &dev_attr_primary_flag);
1086 	device_remove_file(&efx->pci_dev->dev, &dev_attr_link_control_flag);
1087 
1088 	efx_mcdi_detach(efx);
1089 
1090 	memset(nic_data->udp_tunnels, 0, sizeof(nic_data->udp_tunnels));
1091 	mutex_lock(&nic_data->udp_tunnels_lock);
1092 	(void)efx_ef10_set_udp_tnl_ports(efx, true);
1093 	mutex_unlock(&nic_data->udp_tunnels_lock);
1094 
1095 	mutex_destroy(&nic_data->udp_tunnels_lock);
1096 
1097 	efx_mcdi_fini(efx);
1098 	efx_nic_free_buffer(efx, &nic_data->mcdi_buf);
1099 	kfree(nic_data);
1100 }
1101 
1102 static int efx_ef10_probe_pf(struct efx_nic *efx)
1103 {
1104 	return efx_ef10_probe(efx);
1105 }
1106 
1107 int efx_ef10_vadaptor_query(struct efx_nic *efx, unsigned int port_id,
1108 			    u32 *port_flags, u32 *vadaptor_flags,
1109 			    unsigned int *vlan_tags)
1110 {
1111 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1112 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VADAPTOR_QUERY_IN_LEN);
1113 	MCDI_DECLARE_BUF(outbuf, MC_CMD_VADAPTOR_QUERY_OUT_LEN);
1114 	size_t outlen;
1115 	int rc;
1116 
1117 	if (nic_data->datapath_caps &
1118 	    (1 << MC_CMD_GET_CAPABILITIES_OUT_VADAPTOR_QUERY_LBN)) {
1119 		MCDI_SET_DWORD(inbuf, VADAPTOR_QUERY_IN_UPSTREAM_PORT_ID,
1120 			       port_id);
1121 
1122 		rc = efx_mcdi_rpc(efx, MC_CMD_VADAPTOR_QUERY, inbuf, sizeof(inbuf),
1123 				  outbuf, sizeof(outbuf), &outlen);
1124 		if (rc)
1125 			return rc;
1126 
1127 		if (outlen < sizeof(outbuf)) {
1128 			rc = -EIO;
1129 			return rc;
1130 		}
1131 	}
1132 
1133 	if (port_flags)
1134 		*port_flags = MCDI_DWORD(outbuf, VADAPTOR_QUERY_OUT_PORT_FLAGS);
1135 	if (vadaptor_flags)
1136 		*vadaptor_flags =
1137 			MCDI_DWORD(outbuf, VADAPTOR_QUERY_OUT_VADAPTOR_FLAGS);
1138 	if (vlan_tags)
1139 		*vlan_tags =
1140 			MCDI_DWORD(outbuf,
1141 				   VADAPTOR_QUERY_OUT_NUM_AVAILABLE_VLAN_TAGS);
1142 
1143 	return 0;
1144 }
1145 
1146 int efx_ef10_vadaptor_alloc(struct efx_nic *efx, unsigned int port_id)
1147 {
1148 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VADAPTOR_ALLOC_IN_LEN);
1149 
1150 	MCDI_SET_DWORD(inbuf, VADAPTOR_ALLOC_IN_UPSTREAM_PORT_ID, port_id);
1151 	return efx_mcdi_rpc(efx, MC_CMD_VADAPTOR_ALLOC, inbuf, sizeof(inbuf),
1152 			    NULL, 0, NULL);
1153 }
1154 
1155 int efx_ef10_vadaptor_free(struct efx_nic *efx, unsigned int port_id)
1156 {
1157 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VADAPTOR_FREE_IN_LEN);
1158 
1159 	MCDI_SET_DWORD(inbuf, VADAPTOR_FREE_IN_UPSTREAM_PORT_ID, port_id);
1160 	return efx_mcdi_rpc(efx, MC_CMD_VADAPTOR_FREE, inbuf, sizeof(inbuf),
1161 			    NULL, 0, NULL);
1162 }
1163 
1164 int efx_ef10_vport_add_mac(struct efx_nic *efx,
1165 			   unsigned int port_id, u8 *mac)
1166 {
1167 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VPORT_ADD_MAC_ADDRESS_IN_LEN);
1168 
1169 	MCDI_SET_DWORD(inbuf, VPORT_ADD_MAC_ADDRESS_IN_VPORT_ID, port_id);
1170 	ether_addr_copy(MCDI_PTR(inbuf, VPORT_ADD_MAC_ADDRESS_IN_MACADDR), mac);
1171 
1172 	return efx_mcdi_rpc(efx, MC_CMD_VPORT_ADD_MAC_ADDRESS, inbuf,
1173 			    sizeof(inbuf), NULL, 0, NULL);
1174 }
1175 
1176 int efx_ef10_vport_del_mac(struct efx_nic *efx,
1177 			   unsigned int port_id, u8 *mac)
1178 {
1179 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VPORT_DEL_MAC_ADDRESS_IN_LEN);
1180 
1181 	MCDI_SET_DWORD(inbuf, VPORT_DEL_MAC_ADDRESS_IN_VPORT_ID, port_id);
1182 	ether_addr_copy(MCDI_PTR(inbuf, VPORT_DEL_MAC_ADDRESS_IN_MACADDR), mac);
1183 
1184 	return efx_mcdi_rpc(efx, MC_CMD_VPORT_DEL_MAC_ADDRESS, inbuf,
1185 			    sizeof(inbuf), NULL, 0, NULL);
1186 }
1187 
1188 #ifdef CONFIG_SFC_SRIOV
1189 static int efx_ef10_probe_vf(struct efx_nic *efx)
1190 {
1191 	int rc;
1192 	struct pci_dev *pci_dev_pf;
1193 
1194 	/* If the parent PF has no VF data structure, it doesn't know about this
1195 	 * VF so fail probe.  The VF needs to be re-created.  This can happen
1196 	 * if the PF driver is unloaded while the VF is assigned to a guest.
1197 	 */
1198 	pci_dev_pf = efx->pci_dev->physfn;
1199 	if (pci_dev_pf) {
1200 		struct efx_nic *efx_pf = pci_get_drvdata(pci_dev_pf);
1201 		struct efx_ef10_nic_data *nic_data_pf = efx_pf->nic_data;
1202 
1203 		if (!nic_data_pf->vf) {
1204 			netif_info(efx, drv, efx->net_dev,
1205 				   "The VF cannot link to its parent PF; "
1206 				   "please destroy and re-create the VF\n");
1207 			return -EBUSY;
1208 		}
1209 	}
1210 
1211 	rc = efx_ef10_probe(efx);
1212 	if (rc)
1213 		return rc;
1214 
1215 	rc = efx_ef10_get_vf_index(efx);
1216 	if (rc)
1217 		goto fail;
1218 
1219 	if (efx->pci_dev->is_virtfn) {
1220 		if (efx->pci_dev->physfn) {
1221 			struct efx_nic *efx_pf =
1222 				pci_get_drvdata(efx->pci_dev->physfn);
1223 			struct efx_ef10_nic_data *nic_data_p = efx_pf->nic_data;
1224 			struct efx_ef10_nic_data *nic_data = efx->nic_data;
1225 
1226 			nic_data_p->vf[nic_data->vf_index].efx = efx;
1227 			nic_data_p->vf[nic_data->vf_index].pci_dev =
1228 				efx->pci_dev;
1229 		} else
1230 			netif_info(efx, drv, efx->net_dev,
1231 				   "Could not get the PF id from VF\n");
1232 	}
1233 
1234 	return 0;
1235 
1236 fail:
1237 	efx_ef10_remove(efx);
1238 	return rc;
1239 }
1240 #else
1241 static int efx_ef10_probe_vf(struct efx_nic *efx __attribute__ ((unused)))
1242 {
1243 	return 0;
1244 }
1245 #endif
1246 
1247 static int efx_ef10_alloc_vis(struct efx_nic *efx,
1248 			      unsigned int min_vis, unsigned int max_vis)
1249 {
1250 	MCDI_DECLARE_BUF(inbuf, MC_CMD_ALLOC_VIS_IN_LEN);
1251 	MCDI_DECLARE_BUF(outbuf, MC_CMD_ALLOC_VIS_OUT_LEN);
1252 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1253 	size_t outlen;
1254 	int rc;
1255 
1256 	MCDI_SET_DWORD(inbuf, ALLOC_VIS_IN_MIN_VI_COUNT, min_vis);
1257 	MCDI_SET_DWORD(inbuf, ALLOC_VIS_IN_MAX_VI_COUNT, max_vis);
1258 	rc = efx_mcdi_rpc(efx, MC_CMD_ALLOC_VIS, inbuf, sizeof(inbuf),
1259 			  outbuf, sizeof(outbuf), &outlen);
1260 	if (rc != 0)
1261 		return rc;
1262 
1263 	if (outlen < MC_CMD_ALLOC_VIS_OUT_LEN)
1264 		return -EIO;
1265 
1266 	netif_dbg(efx, drv, efx->net_dev, "base VI is A0x%03x\n",
1267 		  MCDI_DWORD(outbuf, ALLOC_VIS_OUT_VI_BASE));
1268 
1269 	nic_data->vi_base = MCDI_DWORD(outbuf, ALLOC_VIS_OUT_VI_BASE);
1270 	nic_data->n_allocated_vis = MCDI_DWORD(outbuf, ALLOC_VIS_OUT_VI_COUNT);
1271 	return 0;
1272 }
1273 
1274 /* Note that the failure path of this function does not free
1275  * resources, as this will be done by efx_ef10_remove().
1276  */
1277 static int efx_ef10_dimension_resources(struct efx_nic *efx)
1278 {
1279 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1280 	unsigned int uc_mem_map_size, wc_mem_map_size;
1281 	unsigned int min_vis = max(EFX_TXQ_TYPES,
1282 				   efx_separate_tx_channels ? 2 : 1);
1283 	unsigned int channel_vis, pio_write_vi_base, max_vis;
1284 	void __iomem *membase;
1285 	int rc;
1286 
1287 	channel_vis = max(efx->n_channels, efx->n_tx_channels * EFX_TXQ_TYPES);
1288 
1289 #ifdef EFX_USE_PIO
1290 	/* Try to allocate PIO buffers if wanted and if the full
1291 	 * number of PIO buffers would be sufficient to allocate one
1292 	 * copy-buffer per TX channel.  Failure is non-fatal, as there
1293 	 * are only a small number of PIO buffers shared between all
1294 	 * functions of the controller.
1295 	 */
1296 	if (efx_piobuf_size != 0 &&
1297 	    nic_data->piobuf_size / efx_piobuf_size * EF10_TX_PIOBUF_COUNT >=
1298 	    efx->n_tx_channels) {
1299 		unsigned int n_piobufs =
1300 			DIV_ROUND_UP(efx->n_tx_channels,
1301 				     nic_data->piobuf_size / efx_piobuf_size);
1302 
1303 		rc = efx_ef10_alloc_piobufs(efx, n_piobufs);
1304 		if (rc == -ENOSPC)
1305 			netif_dbg(efx, probe, efx->net_dev,
1306 				  "out of PIO buffers; cannot allocate more\n");
1307 		else if (rc == -EPERM)
1308 			netif_dbg(efx, probe, efx->net_dev,
1309 				  "not permitted to allocate PIO buffers\n");
1310 		else if (rc)
1311 			netif_err(efx, probe, efx->net_dev,
1312 				  "failed to allocate PIO buffers (%d)\n", rc);
1313 		else
1314 			netif_dbg(efx, probe, efx->net_dev,
1315 				  "allocated %u PIO buffers\n", n_piobufs);
1316 	}
1317 #else
1318 	nic_data->n_piobufs = 0;
1319 #endif
1320 
1321 	/* PIO buffers should be mapped with write-combining enabled,
1322 	 * and we want to make single UC and WC mappings rather than
1323 	 * several of each (in fact that's the only option if host
1324 	 * page size is >4K).  So we may allocate some extra VIs just
1325 	 * for writing PIO buffers through.
1326 	 *
1327 	 * The UC mapping contains (channel_vis - 1) complete VIs and the
1328 	 * first 4K of the next VI.  Then the WC mapping begins with
1329 	 * the remainder of this last VI.
1330 	 */
1331 	uc_mem_map_size = PAGE_ALIGN((channel_vis - 1) * efx->vi_stride +
1332 				     ER_DZ_TX_PIOBUF);
1333 	if (nic_data->n_piobufs) {
1334 		/* pio_write_vi_base rounds down to give the number of complete
1335 		 * VIs inside the UC mapping.
1336 		 */
1337 		pio_write_vi_base = uc_mem_map_size / efx->vi_stride;
1338 		wc_mem_map_size = (PAGE_ALIGN((pio_write_vi_base +
1339 					       nic_data->n_piobufs) *
1340 					      efx->vi_stride) -
1341 				   uc_mem_map_size);
1342 		max_vis = pio_write_vi_base + nic_data->n_piobufs;
1343 	} else {
1344 		pio_write_vi_base = 0;
1345 		wc_mem_map_size = 0;
1346 		max_vis = channel_vis;
1347 	}
1348 
1349 	/* In case the last attached driver failed to free VIs, do it now */
1350 	rc = efx_ef10_free_vis(efx);
1351 	if (rc != 0)
1352 		return rc;
1353 
1354 	rc = efx_ef10_alloc_vis(efx, min_vis, max_vis);
1355 	if (rc != 0)
1356 		return rc;
1357 
1358 	if (nic_data->n_allocated_vis < channel_vis) {
1359 		netif_info(efx, drv, efx->net_dev,
1360 			   "Could not allocate enough VIs to satisfy RSS"
1361 			   " requirements. Performance may not be optimal.\n");
1362 		/* We didn't get the VIs to populate our channels.
1363 		 * We could keep what we got but then we'd have more
1364 		 * interrupts than we need.
1365 		 * Instead calculate new max_channels and restart
1366 		 */
1367 		efx->max_channels = nic_data->n_allocated_vis;
1368 		efx->max_tx_channels =
1369 			nic_data->n_allocated_vis / EFX_TXQ_TYPES;
1370 
1371 		efx_ef10_free_vis(efx);
1372 		return -EAGAIN;
1373 	}
1374 
1375 	/* If we didn't get enough VIs to map all the PIO buffers, free the
1376 	 * PIO buffers
1377 	 */
1378 	if (nic_data->n_piobufs &&
1379 	    nic_data->n_allocated_vis <
1380 	    pio_write_vi_base + nic_data->n_piobufs) {
1381 		netif_dbg(efx, probe, efx->net_dev,
1382 			  "%u VIs are not sufficient to map %u PIO buffers\n",
1383 			  nic_data->n_allocated_vis, nic_data->n_piobufs);
1384 		efx_ef10_free_piobufs(efx);
1385 	}
1386 
1387 	/* Shrink the original UC mapping of the memory BAR */
1388 	membase = ioremap_nocache(efx->membase_phys, uc_mem_map_size);
1389 	if (!membase) {
1390 		netif_err(efx, probe, efx->net_dev,
1391 			  "could not shrink memory BAR to %x\n",
1392 			  uc_mem_map_size);
1393 		return -ENOMEM;
1394 	}
1395 	iounmap(efx->membase);
1396 	efx->membase = membase;
1397 
1398 	/* Set up the WC mapping if needed */
1399 	if (wc_mem_map_size) {
1400 		nic_data->wc_membase = ioremap_wc(efx->membase_phys +
1401 						  uc_mem_map_size,
1402 						  wc_mem_map_size);
1403 		if (!nic_data->wc_membase) {
1404 			netif_err(efx, probe, efx->net_dev,
1405 				  "could not allocate WC mapping of size %x\n",
1406 				  wc_mem_map_size);
1407 			return -ENOMEM;
1408 		}
1409 		nic_data->pio_write_vi_base = pio_write_vi_base;
1410 		nic_data->pio_write_base =
1411 			nic_data->wc_membase +
1412 			(pio_write_vi_base * efx->vi_stride + ER_DZ_TX_PIOBUF -
1413 			 uc_mem_map_size);
1414 
1415 		rc = efx_ef10_link_piobufs(efx);
1416 		if (rc)
1417 			efx_ef10_free_piobufs(efx);
1418 	}
1419 
1420 	netif_dbg(efx, probe, efx->net_dev,
1421 		  "memory BAR at %pa (virtual %p+%x UC, %p+%x WC)\n",
1422 		  &efx->membase_phys, efx->membase, uc_mem_map_size,
1423 		  nic_data->wc_membase, wc_mem_map_size);
1424 
1425 	return 0;
1426 }
1427 
1428 static int efx_ef10_init_nic(struct efx_nic *efx)
1429 {
1430 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1431 	int rc;
1432 
1433 	if (nic_data->must_check_datapath_caps) {
1434 		rc = efx_ef10_init_datapath_caps(efx);
1435 		if (rc)
1436 			return rc;
1437 		nic_data->must_check_datapath_caps = false;
1438 	}
1439 
1440 	if (nic_data->must_realloc_vis) {
1441 		/* We cannot let the number of VIs change now */
1442 		rc = efx_ef10_alloc_vis(efx, nic_data->n_allocated_vis,
1443 					nic_data->n_allocated_vis);
1444 		if (rc)
1445 			return rc;
1446 		nic_data->must_realloc_vis = false;
1447 	}
1448 
1449 	if (nic_data->must_restore_piobufs && nic_data->n_piobufs) {
1450 		rc = efx_ef10_alloc_piobufs(efx, nic_data->n_piobufs);
1451 		if (rc == 0) {
1452 			rc = efx_ef10_link_piobufs(efx);
1453 			if (rc)
1454 				efx_ef10_free_piobufs(efx);
1455 		}
1456 
1457 		/* Log an error on failure, but this is non-fatal.
1458 		 * Permission errors are less important - we've presumably
1459 		 * had the PIO buffer licence removed.
1460 		 */
1461 		if (rc == -EPERM)
1462 			netif_dbg(efx, drv, efx->net_dev,
1463 				  "not permitted to restore PIO buffers\n");
1464 		else if (rc)
1465 			netif_err(efx, drv, efx->net_dev,
1466 				  "failed to restore PIO buffers (%d)\n", rc);
1467 		nic_data->must_restore_piobufs = false;
1468 	}
1469 
1470 	/* don't fail init if RSS setup doesn't work */
1471 	rc = efx->type->rx_push_rss_config(efx, false, efx->rx_indir_table, NULL);
1472 	efx->rss_active = (rc == 0);
1473 
1474 	return 0;
1475 }
1476 
1477 static void efx_ef10_reset_mc_allocations(struct efx_nic *efx)
1478 {
1479 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1480 #ifdef CONFIG_SFC_SRIOV
1481 	unsigned int i;
1482 #endif
1483 
1484 	/* All our allocations have been reset */
1485 	nic_data->must_realloc_vis = true;
1486 	nic_data->must_restore_filters = true;
1487 	nic_data->must_restore_piobufs = true;
1488 	efx_ef10_forget_old_piobufs(efx);
1489 	nic_data->rx_rss_context = EFX_EF10_RSS_CONTEXT_INVALID;
1490 
1491 	/* Driver-created vswitches and vports must be re-created */
1492 	nic_data->must_probe_vswitching = true;
1493 	nic_data->vport_id = EVB_PORT_ID_ASSIGNED;
1494 #ifdef CONFIG_SFC_SRIOV
1495 	if (nic_data->vf)
1496 		for (i = 0; i < efx->vf_count; i++)
1497 			nic_data->vf[i].vport_id = 0;
1498 #endif
1499 }
1500 
1501 static enum reset_type efx_ef10_map_reset_reason(enum reset_type reason)
1502 {
1503 	if (reason == RESET_TYPE_MC_FAILURE)
1504 		return RESET_TYPE_DATAPATH;
1505 
1506 	return efx_mcdi_map_reset_reason(reason);
1507 }
1508 
1509 static int efx_ef10_map_reset_flags(u32 *flags)
1510 {
1511 	enum {
1512 		EF10_RESET_PORT = ((ETH_RESET_MAC | ETH_RESET_PHY) <<
1513 				   ETH_RESET_SHARED_SHIFT),
1514 		EF10_RESET_MC = ((ETH_RESET_DMA | ETH_RESET_FILTER |
1515 				  ETH_RESET_OFFLOAD | ETH_RESET_MAC |
1516 				  ETH_RESET_PHY | ETH_RESET_MGMT) <<
1517 				 ETH_RESET_SHARED_SHIFT)
1518 	};
1519 
1520 	/* We assume for now that our PCI function is permitted to
1521 	 * reset everything.
1522 	 */
1523 
1524 	if ((*flags & EF10_RESET_MC) == EF10_RESET_MC) {
1525 		*flags &= ~EF10_RESET_MC;
1526 		return RESET_TYPE_WORLD;
1527 	}
1528 
1529 	if ((*flags & EF10_RESET_PORT) == EF10_RESET_PORT) {
1530 		*flags &= ~EF10_RESET_PORT;
1531 		return RESET_TYPE_ALL;
1532 	}
1533 
1534 	/* no invisible reset implemented */
1535 
1536 	return -EINVAL;
1537 }
1538 
1539 static int efx_ef10_reset(struct efx_nic *efx, enum reset_type reset_type)
1540 {
1541 	int rc = efx_mcdi_reset(efx, reset_type);
1542 
1543 	/* Unprivileged functions return -EPERM, but need to return success
1544 	 * here so that the datapath is brought back up.
1545 	 */
1546 	if (reset_type == RESET_TYPE_WORLD && rc == -EPERM)
1547 		rc = 0;
1548 
1549 	/* If it was a port reset, trigger reallocation of MC resources.
1550 	 * Note that on an MC reset nothing needs to be done now because we'll
1551 	 * detect the MC reset later and handle it then.
1552 	 * For an FLR, we never get an MC reset event, but the MC has reset all
1553 	 * resources assigned to us, so we have to trigger reallocation now.
1554 	 */
1555 	if ((reset_type == RESET_TYPE_ALL ||
1556 	     reset_type == RESET_TYPE_MCDI_TIMEOUT) && !rc)
1557 		efx_ef10_reset_mc_allocations(efx);
1558 	return rc;
1559 }
1560 
1561 #define EF10_DMA_STAT(ext_name, mcdi_name)			\
1562 	[EF10_STAT_ ## ext_name] =				\
1563 	{ #ext_name, 64, 8 * MC_CMD_MAC_ ## mcdi_name }
1564 #define EF10_DMA_INVIS_STAT(int_name, mcdi_name)		\
1565 	[EF10_STAT_ ## int_name] =				\
1566 	{ NULL, 64, 8 * MC_CMD_MAC_ ## mcdi_name }
1567 #define EF10_OTHER_STAT(ext_name)				\
1568 	[EF10_STAT_ ## ext_name] = { #ext_name, 0, 0 }
1569 #define GENERIC_SW_STAT(ext_name)				\
1570 	[GENERIC_STAT_ ## ext_name] = { #ext_name, 0, 0 }
1571 
1572 static const struct efx_hw_stat_desc efx_ef10_stat_desc[EF10_STAT_COUNT] = {
1573 	EF10_DMA_STAT(port_tx_bytes, TX_BYTES),
1574 	EF10_DMA_STAT(port_tx_packets, TX_PKTS),
1575 	EF10_DMA_STAT(port_tx_pause, TX_PAUSE_PKTS),
1576 	EF10_DMA_STAT(port_tx_control, TX_CONTROL_PKTS),
1577 	EF10_DMA_STAT(port_tx_unicast, TX_UNICAST_PKTS),
1578 	EF10_DMA_STAT(port_tx_multicast, TX_MULTICAST_PKTS),
1579 	EF10_DMA_STAT(port_tx_broadcast, TX_BROADCAST_PKTS),
1580 	EF10_DMA_STAT(port_tx_lt64, TX_LT64_PKTS),
1581 	EF10_DMA_STAT(port_tx_64, TX_64_PKTS),
1582 	EF10_DMA_STAT(port_tx_65_to_127, TX_65_TO_127_PKTS),
1583 	EF10_DMA_STAT(port_tx_128_to_255, TX_128_TO_255_PKTS),
1584 	EF10_DMA_STAT(port_tx_256_to_511, TX_256_TO_511_PKTS),
1585 	EF10_DMA_STAT(port_tx_512_to_1023, TX_512_TO_1023_PKTS),
1586 	EF10_DMA_STAT(port_tx_1024_to_15xx, TX_1024_TO_15XX_PKTS),
1587 	EF10_DMA_STAT(port_tx_15xx_to_jumbo, TX_15XX_TO_JUMBO_PKTS),
1588 	EF10_DMA_STAT(port_rx_bytes, RX_BYTES),
1589 	EF10_DMA_INVIS_STAT(port_rx_bytes_minus_good_bytes, RX_BAD_BYTES),
1590 	EF10_OTHER_STAT(port_rx_good_bytes),
1591 	EF10_OTHER_STAT(port_rx_bad_bytes),
1592 	EF10_DMA_STAT(port_rx_packets, RX_PKTS),
1593 	EF10_DMA_STAT(port_rx_good, RX_GOOD_PKTS),
1594 	EF10_DMA_STAT(port_rx_bad, RX_BAD_FCS_PKTS),
1595 	EF10_DMA_STAT(port_rx_pause, RX_PAUSE_PKTS),
1596 	EF10_DMA_STAT(port_rx_control, RX_CONTROL_PKTS),
1597 	EF10_DMA_STAT(port_rx_unicast, RX_UNICAST_PKTS),
1598 	EF10_DMA_STAT(port_rx_multicast, RX_MULTICAST_PKTS),
1599 	EF10_DMA_STAT(port_rx_broadcast, RX_BROADCAST_PKTS),
1600 	EF10_DMA_STAT(port_rx_lt64, RX_UNDERSIZE_PKTS),
1601 	EF10_DMA_STAT(port_rx_64, RX_64_PKTS),
1602 	EF10_DMA_STAT(port_rx_65_to_127, RX_65_TO_127_PKTS),
1603 	EF10_DMA_STAT(port_rx_128_to_255, RX_128_TO_255_PKTS),
1604 	EF10_DMA_STAT(port_rx_256_to_511, RX_256_TO_511_PKTS),
1605 	EF10_DMA_STAT(port_rx_512_to_1023, RX_512_TO_1023_PKTS),
1606 	EF10_DMA_STAT(port_rx_1024_to_15xx, RX_1024_TO_15XX_PKTS),
1607 	EF10_DMA_STAT(port_rx_15xx_to_jumbo, RX_15XX_TO_JUMBO_PKTS),
1608 	EF10_DMA_STAT(port_rx_gtjumbo, RX_GTJUMBO_PKTS),
1609 	EF10_DMA_STAT(port_rx_bad_gtjumbo, RX_JABBER_PKTS),
1610 	EF10_DMA_STAT(port_rx_overflow, RX_OVERFLOW_PKTS),
1611 	EF10_DMA_STAT(port_rx_align_error, RX_ALIGN_ERROR_PKTS),
1612 	EF10_DMA_STAT(port_rx_length_error, RX_LENGTH_ERROR_PKTS),
1613 	EF10_DMA_STAT(port_rx_nodesc_drops, RX_NODESC_DROPS),
1614 	GENERIC_SW_STAT(rx_nodesc_trunc),
1615 	GENERIC_SW_STAT(rx_noskb_drops),
1616 	EF10_DMA_STAT(port_rx_pm_trunc_bb_overflow, PM_TRUNC_BB_OVERFLOW),
1617 	EF10_DMA_STAT(port_rx_pm_discard_bb_overflow, PM_DISCARD_BB_OVERFLOW),
1618 	EF10_DMA_STAT(port_rx_pm_trunc_vfifo_full, PM_TRUNC_VFIFO_FULL),
1619 	EF10_DMA_STAT(port_rx_pm_discard_vfifo_full, PM_DISCARD_VFIFO_FULL),
1620 	EF10_DMA_STAT(port_rx_pm_trunc_qbb, PM_TRUNC_QBB),
1621 	EF10_DMA_STAT(port_rx_pm_discard_qbb, PM_DISCARD_QBB),
1622 	EF10_DMA_STAT(port_rx_pm_discard_mapping, PM_DISCARD_MAPPING),
1623 	EF10_DMA_STAT(port_rx_dp_q_disabled_packets, RXDP_Q_DISABLED_PKTS),
1624 	EF10_DMA_STAT(port_rx_dp_di_dropped_packets, RXDP_DI_DROPPED_PKTS),
1625 	EF10_DMA_STAT(port_rx_dp_streaming_packets, RXDP_STREAMING_PKTS),
1626 	EF10_DMA_STAT(port_rx_dp_hlb_fetch, RXDP_HLB_FETCH_CONDITIONS),
1627 	EF10_DMA_STAT(port_rx_dp_hlb_wait, RXDP_HLB_WAIT_CONDITIONS),
1628 	EF10_DMA_STAT(rx_unicast, VADAPTER_RX_UNICAST_PACKETS),
1629 	EF10_DMA_STAT(rx_unicast_bytes, VADAPTER_RX_UNICAST_BYTES),
1630 	EF10_DMA_STAT(rx_multicast, VADAPTER_RX_MULTICAST_PACKETS),
1631 	EF10_DMA_STAT(rx_multicast_bytes, VADAPTER_RX_MULTICAST_BYTES),
1632 	EF10_DMA_STAT(rx_broadcast, VADAPTER_RX_BROADCAST_PACKETS),
1633 	EF10_DMA_STAT(rx_broadcast_bytes, VADAPTER_RX_BROADCAST_BYTES),
1634 	EF10_DMA_STAT(rx_bad, VADAPTER_RX_BAD_PACKETS),
1635 	EF10_DMA_STAT(rx_bad_bytes, VADAPTER_RX_BAD_BYTES),
1636 	EF10_DMA_STAT(rx_overflow, VADAPTER_RX_OVERFLOW),
1637 	EF10_DMA_STAT(tx_unicast, VADAPTER_TX_UNICAST_PACKETS),
1638 	EF10_DMA_STAT(tx_unicast_bytes, VADAPTER_TX_UNICAST_BYTES),
1639 	EF10_DMA_STAT(tx_multicast, VADAPTER_TX_MULTICAST_PACKETS),
1640 	EF10_DMA_STAT(tx_multicast_bytes, VADAPTER_TX_MULTICAST_BYTES),
1641 	EF10_DMA_STAT(tx_broadcast, VADAPTER_TX_BROADCAST_PACKETS),
1642 	EF10_DMA_STAT(tx_broadcast_bytes, VADAPTER_TX_BROADCAST_BYTES),
1643 	EF10_DMA_STAT(tx_bad, VADAPTER_TX_BAD_PACKETS),
1644 	EF10_DMA_STAT(tx_bad_bytes, VADAPTER_TX_BAD_BYTES),
1645 	EF10_DMA_STAT(tx_overflow, VADAPTER_TX_OVERFLOW),
1646 	EF10_DMA_STAT(fec_uncorrected_errors, FEC_UNCORRECTED_ERRORS),
1647 	EF10_DMA_STAT(fec_corrected_errors, FEC_CORRECTED_ERRORS),
1648 	EF10_DMA_STAT(fec_corrected_symbols_lane0, FEC_CORRECTED_SYMBOLS_LANE0),
1649 	EF10_DMA_STAT(fec_corrected_symbols_lane1, FEC_CORRECTED_SYMBOLS_LANE1),
1650 	EF10_DMA_STAT(fec_corrected_symbols_lane2, FEC_CORRECTED_SYMBOLS_LANE2),
1651 	EF10_DMA_STAT(fec_corrected_symbols_lane3, FEC_CORRECTED_SYMBOLS_LANE3),
1652 	EF10_DMA_STAT(ctpio_dmabuf_start, CTPIO_DMABUF_START),
1653 	EF10_DMA_STAT(ctpio_vi_busy_fallback, CTPIO_VI_BUSY_FALLBACK),
1654 	EF10_DMA_STAT(ctpio_long_write_success, CTPIO_LONG_WRITE_SUCCESS),
1655 	EF10_DMA_STAT(ctpio_missing_dbell_fail, CTPIO_MISSING_DBELL_FAIL),
1656 	EF10_DMA_STAT(ctpio_overflow_fail, CTPIO_OVERFLOW_FAIL),
1657 	EF10_DMA_STAT(ctpio_underflow_fail, CTPIO_UNDERFLOW_FAIL),
1658 	EF10_DMA_STAT(ctpio_timeout_fail, CTPIO_TIMEOUT_FAIL),
1659 	EF10_DMA_STAT(ctpio_noncontig_wr_fail, CTPIO_NONCONTIG_WR_FAIL),
1660 	EF10_DMA_STAT(ctpio_frm_clobber_fail, CTPIO_FRM_CLOBBER_FAIL),
1661 	EF10_DMA_STAT(ctpio_invalid_wr_fail, CTPIO_INVALID_WR_FAIL),
1662 	EF10_DMA_STAT(ctpio_vi_clobber_fallback, CTPIO_VI_CLOBBER_FALLBACK),
1663 	EF10_DMA_STAT(ctpio_unqualified_fallback, CTPIO_UNQUALIFIED_FALLBACK),
1664 	EF10_DMA_STAT(ctpio_runt_fallback, CTPIO_RUNT_FALLBACK),
1665 	EF10_DMA_STAT(ctpio_success, CTPIO_SUCCESS),
1666 	EF10_DMA_STAT(ctpio_fallback, CTPIO_FALLBACK),
1667 	EF10_DMA_STAT(ctpio_poison, CTPIO_POISON),
1668 	EF10_DMA_STAT(ctpio_erase, CTPIO_ERASE),
1669 };
1670 
1671 #define HUNT_COMMON_STAT_MASK ((1ULL << EF10_STAT_port_tx_bytes) |	\
1672 			       (1ULL << EF10_STAT_port_tx_packets) |	\
1673 			       (1ULL << EF10_STAT_port_tx_pause) |	\
1674 			       (1ULL << EF10_STAT_port_tx_unicast) |	\
1675 			       (1ULL << EF10_STAT_port_tx_multicast) |	\
1676 			       (1ULL << EF10_STAT_port_tx_broadcast) |	\
1677 			       (1ULL << EF10_STAT_port_rx_bytes) |	\
1678 			       (1ULL <<                                 \
1679 				EF10_STAT_port_rx_bytes_minus_good_bytes) | \
1680 			       (1ULL << EF10_STAT_port_rx_good_bytes) |	\
1681 			       (1ULL << EF10_STAT_port_rx_bad_bytes) |	\
1682 			       (1ULL << EF10_STAT_port_rx_packets) |	\
1683 			       (1ULL << EF10_STAT_port_rx_good) |	\
1684 			       (1ULL << EF10_STAT_port_rx_bad) |	\
1685 			       (1ULL << EF10_STAT_port_rx_pause) |	\
1686 			       (1ULL << EF10_STAT_port_rx_control) |	\
1687 			       (1ULL << EF10_STAT_port_rx_unicast) |	\
1688 			       (1ULL << EF10_STAT_port_rx_multicast) |	\
1689 			       (1ULL << EF10_STAT_port_rx_broadcast) |	\
1690 			       (1ULL << EF10_STAT_port_rx_lt64) |	\
1691 			       (1ULL << EF10_STAT_port_rx_64) |		\
1692 			       (1ULL << EF10_STAT_port_rx_65_to_127) |	\
1693 			       (1ULL << EF10_STAT_port_rx_128_to_255) |	\
1694 			       (1ULL << EF10_STAT_port_rx_256_to_511) |	\
1695 			       (1ULL << EF10_STAT_port_rx_512_to_1023) |\
1696 			       (1ULL << EF10_STAT_port_rx_1024_to_15xx) |\
1697 			       (1ULL << EF10_STAT_port_rx_15xx_to_jumbo) |\
1698 			       (1ULL << EF10_STAT_port_rx_gtjumbo) |	\
1699 			       (1ULL << EF10_STAT_port_rx_bad_gtjumbo) |\
1700 			       (1ULL << EF10_STAT_port_rx_overflow) |	\
1701 			       (1ULL << EF10_STAT_port_rx_nodesc_drops) |\
1702 			       (1ULL << GENERIC_STAT_rx_nodesc_trunc) |	\
1703 			       (1ULL << GENERIC_STAT_rx_noskb_drops))
1704 
1705 /* On 7000 series NICs, these statistics are only provided by the 10G MAC.
1706  * For a 10G/40G switchable port we do not expose these because they might
1707  * not include all the packets they should.
1708  * On 8000 series NICs these statistics are always provided.
1709  */
1710 #define HUNT_10G_ONLY_STAT_MASK ((1ULL << EF10_STAT_port_tx_control) |	\
1711 				 (1ULL << EF10_STAT_port_tx_lt64) |	\
1712 				 (1ULL << EF10_STAT_port_tx_64) |	\
1713 				 (1ULL << EF10_STAT_port_tx_65_to_127) |\
1714 				 (1ULL << EF10_STAT_port_tx_128_to_255) |\
1715 				 (1ULL << EF10_STAT_port_tx_256_to_511) |\
1716 				 (1ULL << EF10_STAT_port_tx_512_to_1023) |\
1717 				 (1ULL << EF10_STAT_port_tx_1024_to_15xx) |\
1718 				 (1ULL << EF10_STAT_port_tx_15xx_to_jumbo))
1719 
1720 /* These statistics are only provided by the 40G MAC.  For a 10G/40G
1721  * switchable port we do expose these because the errors will otherwise
1722  * be silent.
1723  */
1724 #define HUNT_40G_EXTRA_STAT_MASK ((1ULL << EF10_STAT_port_rx_align_error) |\
1725 				  (1ULL << EF10_STAT_port_rx_length_error))
1726 
1727 /* These statistics are only provided if the firmware supports the
1728  * capability PM_AND_RXDP_COUNTERS.
1729  */
1730 #define HUNT_PM_AND_RXDP_STAT_MASK (					\
1731 	(1ULL << EF10_STAT_port_rx_pm_trunc_bb_overflow) |		\
1732 	(1ULL << EF10_STAT_port_rx_pm_discard_bb_overflow) |		\
1733 	(1ULL << EF10_STAT_port_rx_pm_trunc_vfifo_full) |		\
1734 	(1ULL << EF10_STAT_port_rx_pm_discard_vfifo_full) |		\
1735 	(1ULL << EF10_STAT_port_rx_pm_trunc_qbb) |			\
1736 	(1ULL << EF10_STAT_port_rx_pm_discard_qbb) |			\
1737 	(1ULL << EF10_STAT_port_rx_pm_discard_mapping) |		\
1738 	(1ULL << EF10_STAT_port_rx_dp_q_disabled_packets) |		\
1739 	(1ULL << EF10_STAT_port_rx_dp_di_dropped_packets) |		\
1740 	(1ULL << EF10_STAT_port_rx_dp_streaming_packets) |		\
1741 	(1ULL << EF10_STAT_port_rx_dp_hlb_fetch) |			\
1742 	(1ULL << EF10_STAT_port_rx_dp_hlb_wait))
1743 
1744 /* These statistics are only provided if the NIC supports MC_CMD_MAC_STATS_V2,
1745  * indicated by returning a value >= MC_CMD_MAC_NSTATS_V2 in
1746  * MC_CMD_GET_CAPABILITIES_V4_OUT_MAC_STATS_NUM_STATS.
1747  * These bits are in the second u64 of the raw mask.
1748  */
1749 #define EF10_FEC_STAT_MASK (						\
1750 	(1ULL << (EF10_STAT_fec_uncorrected_errors - 64)) |		\
1751 	(1ULL << (EF10_STAT_fec_corrected_errors - 64)) |		\
1752 	(1ULL << (EF10_STAT_fec_corrected_symbols_lane0 - 64)) |	\
1753 	(1ULL << (EF10_STAT_fec_corrected_symbols_lane1 - 64)) |	\
1754 	(1ULL << (EF10_STAT_fec_corrected_symbols_lane2 - 64)) |	\
1755 	(1ULL << (EF10_STAT_fec_corrected_symbols_lane3 - 64)))
1756 
1757 /* These statistics are only provided if the NIC supports MC_CMD_MAC_STATS_V3,
1758  * indicated by returning a value >= MC_CMD_MAC_NSTATS_V3 in
1759  * MC_CMD_GET_CAPABILITIES_V4_OUT_MAC_STATS_NUM_STATS.
1760  * These bits are in the second u64 of the raw mask.
1761  */
1762 #define EF10_CTPIO_STAT_MASK (						\
1763 	(1ULL << (EF10_STAT_ctpio_dmabuf_start - 64)) |			\
1764 	(1ULL << (EF10_STAT_ctpio_vi_busy_fallback - 64)) |		\
1765 	(1ULL << (EF10_STAT_ctpio_long_write_success - 64)) |		\
1766 	(1ULL << (EF10_STAT_ctpio_missing_dbell_fail - 64)) |		\
1767 	(1ULL << (EF10_STAT_ctpio_overflow_fail - 64)) |		\
1768 	(1ULL << (EF10_STAT_ctpio_underflow_fail - 64)) |		\
1769 	(1ULL << (EF10_STAT_ctpio_timeout_fail - 64)) |			\
1770 	(1ULL << (EF10_STAT_ctpio_noncontig_wr_fail - 64)) |		\
1771 	(1ULL << (EF10_STAT_ctpio_frm_clobber_fail - 64)) |		\
1772 	(1ULL << (EF10_STAT_ctpio_invalid_wr_fail - 64)) |		\
1773 	(1ULL << (EF10_STAT_ctpio_vi_clobber_fallback - 64)) |		\
1774 	(1ULL << (EF10_STAT_ctpio_unqualified_fallback - 64)) |		\
1775 	(1ULL << (EF10_STAT_ctpio_runt_fallback - 64)) |		\
1776 	(1ULL << (EF10_STAT_ctpio_success - 64)) |			\
1777 	(1ULL << (EF10_STAT_ctpio_fallback - 64)) |			\
1778 	(1ULL << (EF10_STAT_ctpio_poison - 64)) |			\
1779 	(1ULL << (EF10_STAT_ctpio_erase - 64)))
1780 
1781 static u64 efx_ef10_raw_stat_mask(struct efx_nic *efx)
1782 {
1783 	u64 raw_mask = HUNT_COMMON_STAT_MASK;
1784 	u32 port_caps = efx_mcdi_phy_get_caps(efx);
1785 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1786 
1787 	if (!(efx->mcdi->fn_flags &
1788 	      1 << MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_LINKCTRL))
1789 		return 0;
1790 
1791 	if (port_caps & (1 << MC_CMD_PHY_CAP_40000FDX_LBN)) {
1792 		raw_mask |= HUNT_40G_EXTRA_STAT_MASK;
1793 		/* 8000 series have everything even at 40G */
1794 		if (nic_data->datapath_caps2 &
1795 		    (1 << MC_CMD_GET_CAPABILITIES_V2_OUT_MAC_STATS_40G_TX_SIZE_BINS_LBN))
1796 			raw_mask |= HUNT_10G_ONLY_STAT_MASK;
1797 	} else {
1798 		raw_mask |= HUNT_10G_ONLY_STAT_MASK;
1799 	}
1800 
1801 	if (nic_data->datapath_caps &
1802 	    (1 << MC_CMD_GET_CAPABILITIES_OUT_PM_AND_RXDP_COUNTERS_LBN))
1803 		raw_mask |= HUNT_PM_AND_RXDP_STAT_MASK;
1804 
1805 	return raw_mask;
1806 }
1807 
1808 static void efx_ef10_get_stat_mask(struct efx_nic *efx, unsigned long *mask)
1809 {
1810 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1811 	u64 raw_mask[2];
1812 
1813 	raw_mask[0] = efx_ef10_raw_stat_mask(efx);
1814 
1815 	/* Only show vadaptor stats when EVB capability is present */
1816 	if (nic_data->datapath_caps &
1817 	    (1 << MC_CMD_GET_CAPABILITIES_OUT_EVB_LBN)) {
1818 		raw_mask[0] |= ~((1ULL << EF10_STAT_rx_unicast) - 1);
1819 		raw_mask[1] = (1ULL << (EF10_STAT_V1_COUNT - 64)) - 1;
1820 	} else {
1821 		raw_mask[1] = 0;
1822 	}
1823 	/* Only show FEC stats when NIC supports MC_CMD_MAC_STATS_V2 */
1824 	if (efx->num_mac_stats >= MC_CMD_MAC_NSTATS_V2)
1825 		raw_mask[1] |= EF10_FEC_STAT_MASK;
1826 
1827 	/* CTPIO stats appear in V3. Only show them on devices that actually
1828 	 * support CTPIO. Although this driver doesn't use CTPIO others might,
1829 	 * and we may be reporting the stats for the underlying port.
1830 	 */
1831 	if (efx->num_mac_stats >= MC_CMD_MAC_NSTATS_V3 &&
1832 	    (nic_data->datapath_caps2 &
1833 	     (1 << MC_CMD_GET_CAPABILITIES_V4_OUT_CTPIO_LBN)))
1834 		raw_mask[1] |= EF10_CTPIO_STAT_MASK;
1835 
1836 #if BITS_PER_LONG == 64
1837 	BUILD_BUG_ON(BITS_TO_LONGS(EF10_STAT_COUNT) != 2);
1838 	mask[0] = raw_mask[0];
1839 	mask[1] = raw_mask[1];
1840 #else
1841 	BUILD_BUG_ON(BITS_TO_LONGS(EF10_STAT_COUNT) != 3);
1842 	mask[0] = raw_mask[0] & 0xffffffff;
1843 	mask[1] = raw_mask[0] >> 32;
1844 	mask[2] = raw_mask[1] & 0xffffffff;
1845 #endif
1846 }
1847 
1848 static size_t efx_ef10_describe_stats(struct efx_nic *efx, u8 *names)
1849 {
1850 	DECLARE_BITMAP(mask, EF10_STAT_COUNT);
1851 
1852 	efx_ef10_get_stat_mask(efx, mask);
1853 	return efx_nic_describe_stats(efx_ef10_stat_desc, EF10_STAT_COUNT,
1854 				      mask, names);
1855 }
1856 
1857 static size_t efx_ef10_update_stats_common(struct efx_nic *efx, u64 *full_stats,
1858 					   struct rtnl_link_stats64 *core_stats)
1859 {
1860 	DECLARE_BITMAP(mask, EF10_STAT_COUNT);
1861 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1862 	u64 *stats = nic_data->stats;
1863 	size_t stats_count = 0, index;
1864 
1865 	efx_ef10_get_stat_mask(efx, mask);
1866 
1867 	if (full_stats) {
1868 		for_each_set_bit(index, mask, EF10_STAT_COUNT) {
1869 			if (efx_ef10_stat_desc[index].name) {
1870 				*full_stats++ = stats[index];
1871 				++stats_count;
1872 			}
1873 		}
1874 	}
1875 
1876 	if (!core_stats)
1877 		return stats_count;
1878 
1879 	if (nic_data->datapath_caps &
1880 			1 << MC_CMD_GET_CAPABILITIES_OUT_EVB_LBN) {
1881 		/* Use vadaptor stats. */
1882 		core_stats->rx_packets = stats[EF10_STAT_rx_unicast] +
1883 					 stats[EF10_STAT_rx_multicast] +
1884 					 stats[EF10_STAT_rx_broadcast];
1885 		core_stats->tx_packets = stats[EF10_STAT_tx_unicast] +
1886 					 stats[EF10_STAT_tx_multicast] +
1887 					 stats[EF10_STAT_tx_broadcast];
1888 		core_stats->rx_bytes = stats[EF10_STAT_rx_unicast_bytes] +
1889 				       stats[EF10_STAT_rx_multicast_bytes] +
1890 				       stats[EF10_STAT_rx_broadcast_bytes];
1891 		core_stats->tx_bytes = stats[EF10_STAT_tx_unicast_bytes] +
1892 				       stats[EF10_STAT_tx_multicast_bytes] +
1893 				       stats[EF10_STAT_tx_broadcast_bytes];
1894 		core_stats->rx_dropped = stats[GENERIC_STAT_rx_nodesc_trunc] +
1895 					 stats[GENERIC_STAT_rx_noskb_drops];
1896 		core_stats->multicast = stats[EF10_STAT_rx_multicast];
1897 		core_stats->rx_crc_errors = stats[EF10_STAT_rx_bad];
1898 		core_stats->rx_fifo_errors = stats[EF10_STAT_rx_overflow];
1899 		core_stats->rx_errors = core_stats->rx_crc_errors;
1900 		core_stats->tx_errors = stats[EF10_STAT_tx_bad];
1901 	} else {
1902 		/* Use port stats. */
1903 		core_stats->rx_packets = stats[EF10_STAT_port_rx_packets];
1904 		core_stats->tx_packets = stats[EF10_STAT_port_tx_packets];
1905 		core_stats->rx_bytes = stats[EF10_STAT_port_rx_bytes];
1906 		core_stats->tx_bytes = stats[EF10_STAT_port_tx_bytes];
1907 		core_stats->rx_dropped = stats[EF10_STAT_port_rx_nodesc_drops] +
1908 					 stats[GENERIC_STAT_rx_nodesc_trunc] +
1909 					 stats[GENERIC_STAT_rx_noskb_drops];
1910 		core_stats->multicast = stats[EF10_STAT_port_rx_multicast];
1911 		core_stats->rx_length_errors =
1912 				stats[EF10_STAT_port_rx_gtjumbo] +
1913 				stats[EF10_STAT_port_rx_length_error];
1914 		core_stats->rx_crc_errors = stats[EF10_STAT_port_rx_bad];
1915 		core_stats->rx_frame_errors =
1916 				stats[EF10_STAT_port_rx_align_error];
1917 		core_stats->rx_fifo_errors = stats[EF10_STAT_port_rx_overflow];
1918 		core_stats->rx_errors = (core_stats->rx_length_errors +
1919 					 core_stats->rx_crc_errors +
1920 					 core_stats->rx_frame_errors);
1921 	}
1922 
1923 	return stats_count;
1924 }
1925 
1926 static int efx_ef10_try_update_nic_stats_pf(struct efx_nic *efx)
1927 {
1928 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1929 	DECLARE_BITMAP(mask, EF10_STAT_COUNT);
1930 	__le64 generation_start, generation_end;
1931 	u64 *stats = nic_data->stats;
1932 	__le64 *dma_stats;
1933 
1934 	efx_ef10_get_stat_mask(efx, mask);
1935 
1936 	dma_stats = efx->stats_buffer.addr;
1937 
1938 	generation_end = dma_stats[efx->num_mac_stats - 1];
1939 	if (generation_end == EFX_MC_STATS_GENERATION_INVALID)
1940 		return 0;
1941 	rmb();
1942 	efx_nic_update_stats(efx_ef10_stat_desc, EF10_STAT_COUNT, mask,
1943 			     stats, efx->stats_buffer.addr, false);
1944 	rmb();
1945 	generation_start = dma_stats[MC_CMD_MAC_GENERATION_START];
1946 	if (generation_end != generation_start)
1947 		return -EAGAIN;
1948 
1949 	/* Update derived statistics */
1950 	efx_nic_fix_nodesc_drop_stat(efx,
1951 				     &stats[EF10_STAT_port_rx_nodesc_drops]);
1952 	stats[EF10_STAT_port_rx_good_bytes] =
1953 		stats[EF10_STAT_port_rx_bytes] -
1954 		stats[EF10_STAT_port_rx_bytes_minus_good_bytes];
1955 	efx_update_diff_stat(&stats[EF10_STAT_port_rx_bad_bytes],
1956 			     stats[EF10_STAT_port_rx_bytes_minus_good_bytes]);
1957 	efx_update_sw_stats(efx, stats);
1958 	return 0;
1959 }
1960 
1961 
1962 static size_t efx_ef10_update_stats_pf(struct efx_nic *efx, u64 *full_stats,
1963 				       struct rtnl_link_stats64 *core_stats)
1964 {
1965 	int retry;
1966 
1967 	/* If we're unlucky enough to read statistics during the DMA, wait
1968 	 * up to 10ms for it to finish (typically takes <500us)
1969 	 */
1970 	for (retry = 0; retry < 100; ++retry) {
1971 		if (efx_ef10_try_update_nic_stats_pf(efx) == 0)
1972 			break;
1973 		udelay(100);
1974 	}
1975 
1976 	return efx_ef10_update_stats_common(efx, full_stats, core_stats);
1977 }
1978 
1979 static int efx_ef10_try_update_nic_stats_vf(struct efx_nic *efx)
1980 {
1981 	MCDI_DECLARE_BUF(inbuf, MC_CMD_MAC_STATS_IN_LEN);
1982 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1983 	DECLARE_BITMAP(mask, EF10_STAT_COUNT);
1984 	__le64 generation_start, generation_end;
1985 	u64 *stats = nic_data->stats;
1986 	u32 dma_len = efx->num_mac_stats * sizeof(u64);
1987 	struct efx_buffer stats_buf;
1988 	__le64 *dma_stats;
1989 	int rc;
1990 
1991 	spin_unlock_bh(&efx->stats_lock);
1992 
1993 	if (in_interrupt()) {
1994 		/* If in atomic context, cannot update stats.  Just update the
1995 		 * software stats and return so the caller can continue.
1996 		 */
1997 		spin_lock_bh(&efx->stats_lock);
1998 		efx_update_sw_stats(efx, stats);
1999 		return 0;
2000 	}
2001 
2002 	efx_ef10_get_stat_mask(efx, mask);
2003 
2004 	rc = efx_nic_alloc_buffer(efx, &stats_buf, dma_len, GFP_ATOMIC);
2005 	if (rc) {
2006 		spin_lock_bh(&efx->stats_lock);
2007 		return rc;
2008 	}
2009 
2010 	dma_stats = stats_buf.addr;
2011 	dma_stats[efx->num_mac_stats - 1] = EFX_MC_STATS_GENERATION_INVALID;
2012 
2013 	MCDI_SET_QWORD(inbuf, MAC_STATS_IN_DMA_ADDR, stats_buf.dma_addr);
2014 	MCDI_POPULATE_DWORD_1(inbuf, MAC_STATS_IN_CMD,
2015 			      MAC_STATS_IN_DMA, 1);
2016 	MCDI_SET_DWORD(inbuf, MAC_STATS_IN_DMA_LEN, dma_len);
2017 	MCDI_SET_DWORD(inbuf, MAC_STATS_IN_PORT_ID, EVB_PORT_ID_ASSIGNED);
2018 
2019 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_MAC_STATS, inbuf, sizeof(inbuf),
2020 				NULL, 0, NULL);
2021 	spin_lock_bh(&efx->stats_lock);
2022 	if (rc) {
2023 		/* Expect ENOENT if DMA queues have not been set up */
2024 		if (rc != -ENOENT || atomic_read(&efx->active_queues))
2025 			efx_mcdi_display_error(efx, MC_CMD_MAC_STATS,
2026 					       sizeof(inbuf), NULL, 0, rc);
2027 		goto out;
2028 	}
2029 
2030 	generation_end = dma_stats[efx->num_mac_stats - 1];
2031 	if (generation_end == EFX_MC_STATS_GENERATION_INVALID) {
2032 		WARN_ON_ONCE(1);
2033 		goto out;
2034 	}
2035 	rmb();
2036 	efx_nic_update_stats(efx_ef10_stat_desc, EF10_STAT_COUNT, mask,
2037 			     stats, stats_buf.addr, false);
2038 	rmb();
2039 	generation_start = dma_stats[MC_CMD_MAC_GENERATION_START];
2040 	if (generation_end != generation_start) {
2041 		rc = -EAGAIN;
2042 		goto out;
2043 	}
2044 
2045 	efx_update_sw_stats(efx, stats);
2046 out:
2047 	efx_nic_free_buffer(efx, &stats_buf);
2048 	return rc;
2049 }
2050 
2051 static size_t efx_ef10_update_stats_vf(struct efx_nic *efx, u64 *full_stats,
2052 				       struct rtnl_link_stats64 *core_stats)
2053 {
2054 	if (efx_ef10_try_update_nic_stats_vf(efx))
2055 		return 0;
2056 
2057 	return efx_ef10_update_stats_common(efx, full_stats, core_stats);
2058 }
2059 
2060 static void efx_ef10_push_irq_moderation(struct efx_channel *channel)
2061 {
2062 	struct efx_nic *efx = channel->efx;
2063 	unsigned int mode, usecs;
2064 	efx_dword_t timer_cmd;
2065 
2066 	if (channel->irq_moderation_us) {
2067 		mode = 3;
2068 		usecs = channel->irq_moderation_us;
2069 	} else {
2070 		mode = 0;
2071 		usecs = 0;
2072 	}
2073 
2074 	if (EFX_EF10_WORKAROUND_61265(efx)) {
2075 		MCDI_DECLARE_BUF(inbuf, MC_CMD_SET_EVQ_TMR_IN_LEN);
2076 		unsigned int ns = usecs * 1000;
2077 
2078 		MCDI_SET_DWORD(inbuf, SET_EVQ_TMR_IN_INSTANCE,
2079 			       channel->channel);
2080 		MCDI_SET_DWORD(inbuf, SET_EVQ_TMR_IN_TMR_LOAD_REQ_NS, ns);
2081 		MCDI_SET_DWORD(inbuf, SET_EVQ_TMR_IN_TMR_RELOAD_REQ_NS, ns);
2082 		MCDI_SET_DWORD(inbuf, SET_EVQ_TMR_IN_TMR_MODE, mode);
2083 
2084 		efx_mcdi_rpc_async(efx, MC_CMD_SET_EVQ_TMR,
2085 				   inbuf, sizeof(inbuf), 0, NULL, 0);
2086 	} else if (EFX_EF10_WORKAROUND_35388(efx)) {
2087 		unsigned int ticks = efx_usecs_to_ticks(efx, usecs);
2088 
2089 		EFX_POPULATE_DWORD_3(timer_cmd, ERF_DD_EVQ_IND_TIMER_FLAGS,
2090 				     EFE_DD_EVQ_IND_TIMER_FLAGS,
2091 				     ERF_DD_EVQ_IND_TIMER_MODE, mode,
2092 				     ERF_DD_EVQ_IND_TIMER_VAL, ticks);
2093 		efx_writed_page(efx, &timer_cmd, ER_DD_EVQ_INDIRECT,
2094 				channel->channel);
2095 	} else {
2096 		unsigned int ticks = efx_usecs_to_ticks(efx, usecs);
2097 
2098 		EFX_POPULATE_DWORD_3(timer_cmd, ERF_DZ_TC_TIMER_MODE, mode,
2099 				     ERF_DZ_TC_TIMER_VAL, ticks,
2100 				     ERF_FZ_TC_TMR_REL_VAL, ticks);
2101 		efx_writed_page(efx, &timer_cmd, ER_DZ_EVQ_TMR,
2102 				channel->channel);
2103 	}
2104 }
2105 
2106 static void efx_ef10_get_wol_vf(struct efx_nic *efx,
2107 				struct ethtool_wolinfo *wol) {}
2108 
2109 static int efx_ef10_set_wol_vf(struct efx_nic *efx, u32 type)
2110 {
2111 	return -EOPNOTSUPP;
2112 }
2113 
2114 static void efx_ef10_get_wol(struct efx_nic *efx, struct ethtool_wolinfo *wol)
2115 {
2116 	wol->supported = 0;
2117 	wol->wolopts = 0;
2118 	memset(&wol->sopass, 0, sizeof(wol->sopass));
2119 }
2120 
2121 static int efx_ef10_set_wol(struct efx_nic *efx, u32 type)
2122 {
2123 	if (type != 0)
2124 		return -EINVAL;
2125 	return 0;
2126 }
2127 
2128 static void efx_ef10_mcdi_request(struct efx_nic *efx,
2129 				  const efx_dword_t *hdr, size_t hdr_len,
2130 				  const efx_dword_t *sdu, size_t sdu_len)
2131 {
2132 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2133 	u8 *pdu = nic_data->mcdi_buf.addr;
2134 
2135 	memcpy(pdu, hdr, hdr_len);
2136 	memcpy(pdu + hdr_len, sdu, sdu_len);
2137 	wmb();
2138 
2139 	/* The hardware provides 'low' and 'high' (doorbell) registers
2140 	 * for passing the 64-bit address of an MCDI request to
2141 	 * firmware.  However the dwords are swapped by firmware.  The
2142 	 * least significant bits of the doorbell are then 0 for all
2143 	 * MCDI requests due to alignment.
2144 	 */
2145 	_efx_writed(efx, cpu_to_le32((u64)nic_data->mcdi_buf.dma_addr >> 32),
2146 		    ER_DZ_MC_DB_LWRD);
2147 	_efx_writed(efx, cpu_to_le32((u32)nic_data->mcdi_buf.dma_addr),
2148 		    ER_DZ_MC_DB_HWRD);
2149 }
2150 
2151 static bool efx_ef10_mcdi_poll_response(struct efx_nic *efx)
2152 {
2153 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2154 	const efx_dword_t hdr = *(const efx_dword_t *)nic_data->mcdi_buf.addr;
2155 
2156 	rmb();
2157 	return EFX_DWORD_FIELD(hdr, MCDI_HEADER_RESPONSE);
2158 }
2159 
2160 static void
2161 efx_ef10_mcdi_read_response(struct efx_nic *efx, efx_dword_t *outbuf,
2162 			    size_t offset, size_t outlen)
2163 {
2164 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2165 	const u8 *pdu = nic_data->mcdi_buf.addr;
2166 
2167 	memcpy(outbuf, pdu + offset, outlen);
2168 }
2169 
2170 static void efx_ef10_mcdi_reboot_detected(struct efx_nic *efx)
2171 {
2172 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2173 
2174 	/* All our allocations have been reset */
2175 	efx_ef10_reset_mc_allocations(efx);
2176 
2177 	/* The datapath firmware might have been changed */
2178 	nic_data->must_check_datapath_caps = true;
2179 
2180 	/* MAC statistics have been cleared on the NIC; clear the local
2181 	 * statistic that we update with efx_update_diff_stat().
2182 	 */
2183 	nic_data->stats[EF10_STAT_port_rx_bad_bytes] = 0;
2184 }
2185 
2186 static int efx_ef10_mcdi_poll_reboot(struct efx_nic *efx)
2187 {
2188 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2189 	int rc;
2190 
2191 	rc = efx_ef10_get_warm_boot_count(efx);
2192 	if (rc < 0) {
2193 		/* The firmware is presumably in the process of
2194 		 * rebooting.  However, we are supposed to report each
2195 		 * reboot just once, so we must only do that once we
2196 		 * can read and store the updated warm boot count.
2197 		 */
2198 		return 0;
2199 	}
2200 
2201 	if (rc == nic_data->warm_boot_count)
2202 		return 0;
2203 
2204 	nic_data->warm_boot_count = rc;
2205 	efx_ef10_mcdi_reboot_detected(efx);
2206 
2207 	return -EIO;
2208 }
2209 
2210 /* Handle an MSI interrupt
2211  *
2212  * Handle an MSI hardware interrupt.  This routine schedules event
2213  * queue processing.  No interrupt acknowledgement cycle is necessary.
2214  * Also, we never need to check that the interrupt is for us, since
2215  * MSI interrupts cannot be shared.
2216  */
2217 static irqreturn_t efx_ef10_msi_interrupt(int irq, void *dev_id)
2218 {
2219 	struct efx_msi_context *context = dev_id;
2220 	struct efx_nic *efx = context->efx;
2221 
2222 	netif_vdbg(efx, intr, efx->net_dev,
2223 		   "IRQ %d on CPU %d\n", irq, raw_smp_processor_id());
2224 
2225 	if (likely(READ_ONCE(efx->irq_soft_enabled))) {
2226 		/* Note test interrupts */
2227 		if (context->index == efx->irq_level)
2228 			efx->last_irq_cpu = raw_smp_processor_id();
2229 
2230 		/* Schedule processing of the channel */
2231 		efx_schedule_channel_irq(efx->channel[context->index]);
2232 	}
2233 
2234 	return IRQ_HANDLED;
2235 }
2236 
2237 static irqreturn_t efx_ef10_legacy_interrupt(int irq, void *dev_id)
2238 {
2239 	struct efx_nic *efx = dev_id;
2240 	bool soft_enabled = READ_ONCE(efx->irq_soft_enabled);
2241 	struct efx_channel *channel;
2242 	efx_dword_t reg;
2243 	u32 queues;
2244 
2245 	/* Read the ISR which also ACKs the interrupts */
2246 	efx_readd(efx, &reg, ER_DZ_BIU_INT_ISR);
2247 	queues = EFX_DWORD_FIELD(reg, ERF_DZ_ISR_REG);
2248 
2249 	if (queues == 0)
2250 		return IRQ_NONE;
2251 
2252 	if (likely(soft_enabled)) {
2253 		/* Note test interrupts */
2254 		if (queues & (1U << efx->irq_level))
2255 			efx->last_irq_cpu = raw_smp_processor_id();
2256 
2257 		efx_for_each_channel(channel, efx) {
2258 			if (queues & 1)
2259 				efx_schedule_channel_irq(channel);
2260 			queues >>= 1;
2261 		}
2262 	}
2263 
2264 	netif_vdbg(efx, intr, efx->net_dev,
2265 		   "IRQ %d on CPU %d status " EFX_DWORD_FMT "\n",
2266 		   irq, raw_smp_processor_id(), EFX_DWORD_VAL(reg));
2267 
2268 	return IRQ_HANDLED;
2269 }
2270 
2271 static int efx_ef10_irq_test_generate(struct efx_nic *efx)
2272 {
2273 	MCDI_DECLARE_BUF(inbuf, MC_CMD_TRIGGER_INTERRUPT_IN_LEN);
2274 
2275 	if (efx_mcdi_set_workaround(efx, MC_CMD_WORKAROUND_BUG41750, true,
2276 				    NULL) == 0)
2277 		return -ENOTSUPP;
2278 
2279 	BUILD_BUG_ON(MC_CMD_TRIGGER_INTERRUPT_OUT_LEN != 0);
2280 
2281 	MCDI_SET_DWORD(inbuf, TRIGGER_INTERRUPT_IN_INTR_LEVEL, efx->irq_level);
2282 	return efx_mcdi_rpc(efx, MC_CMD_TRIGGER_INTERRUPT,
2283 			    inbuf, sizeof(inbuf), NULL, 0, NULL);
2284 }
2285 
2286 static int efx_ef10_tx_probe(struct efx_tx_queue *tx_queue)
2287 {
2288 	return efx_nic_alloc_buffer(tx_queue->efx, &tx_queue->txd.buf,
2289 				    (tx_queue->ptr_mask + 1) *
2290 				    sizeof(efx_qword_t),
2291 				    GFP_KERNEL);
2292 }
2293 
2294 /* This writes to the TX_DESC_WPTR and also pushes data */
2295 static inline void efx_ef10_push_tx_desc(struct efx_tx_queue *tx_queue,
2296 					 const efx_qword_t *txd)
2297 {
2298 	unsigned int write_ptr;
2299 	efx_oword_t reg;
2300 
2301 	write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
2302 	EFX_POPULATE_OWORD_1(reg, ERF_DZ_TX_DESC_WPTR, write_ptr);
2303 	reg.qword[0] = *txd;
2304 	efx_writeo_page(tx_queue->efx, &reg,
2305 			ER_DZ_TX_DESC_UPD, tx_queue->queue);
2306 }
2307 
2308 /* Add Firmware-Assisted TSO v2 option descriptors to a queue.
2309  */
2310 static int efx_ef10_tx_tso_desc(struct efx_tx_queue *tx_queue,
2311 				struct sk_buff *skb,
2312 				bool *data_mapped)
2313 {
2314 	struct efx_tx_buffer *buffer;
2315 	struct tcphdr *tcp;
2316 	struct iphdr *ip;
2317 
2318 	u16 ipv4_id;
2319 	u32 seqnum;
2320 	u32 mss;
2321 
2322 	EFX_WARN_ON_ONCE_PARANOID(tx_queue->tso_version != 2);
2323 
2324 	mss = skb_shinfo(skb)->gso_size;
2325 
2326 	if (unlikely(mss < 4)) {
2327 		WARN_ONCE(1, "MSS of %u is too small for TSO v2\n", mss);
2328 		return -EINVAL;
2329 	}
2330 
2331 	ip = ip_hdr(skb);
2332 	if (ip->version == 4) {
2333 		/* Modify IPv4 header if needed. */
2334 		ip->tot_len = 0;
2335 		ip->check = 0;
2336 		ipv4_id = ntohs(ip->id);
2337 	} else {
2338 		/* Modify IPv6 header if needed. */
2339 		struct ipv6hdr *ipv6 = ipv6_hdr(skb);
2340 
2341 		ipv6->payload_len = 0;
2342 		ipv4_id = 0;
2343 	}
2344 
2345 	tcp = tcp_hdr(skb);
2346 	seqnum = ntohl(tcp->seq);
2347 
2348 	buffer = efx_tx_queue_get_insert_buffer(tx_queue);
2349 
2350 	buffer->flags = EFX_TX_BUF_OPTION;
2351 	buffer->len = 0;
2352 	buffer->unmap_len = 0;
2353 	EFX_POPULATE_QWORD_5(buffer->option,
2354 			ESF_DZ_TX_DESC_IS_OPT, 1,
2355 			ESF_DZ_TX_OPTION_TYPE, ESE_DZ_TX_OPTION_DESC_TSO,
2356 			ESF_DZ_TX_TSO_OPTION_TYPE,
2357 			ESE_DZ_TX_TSO_OPTION_DESC_FATSO2A,
2358 			ESF_DZ_TX_TSO_IP_ID, ipv4_id,
2359 			ESF_DZ_TX_TSO_TCP_SEQNO, seqnum
2360 			);
2361 	++tx_queue->insert_count;
2362 
2363 	buffer = efx_tx_queue_get_insert_buffer(tx_queue);
2364 
2365 	buffer->flags = EFX_TX_BUF_OPTION;
2366 	buffer->len = 0;
2367 	buffer->unmap_len = 0;
2368 	EFX_POPULATE_QWORD_4(buffer->option,
2369 			ESF_DZ_TX_DESC_IS_OPT, 1,
2370 			ESF_DZ_TX_OPTION_TYPE, ESE_DZ_TX_OPTION_DESC_TSO,
2371 			ESF_DZ_TX_TSO_OPTION_TYPE,
2372 			ESE_DZ_TX_TSO_OPTION_DESC_FATSO2B,
2373 			ESF_DZ_TX_TSO_TCP_MSS, mss
2374 			);
2375 	++tx_queue->insert_count;
2376 
2377 	return 0;
2378 }
2379 
2380 static u32 efx_ef10_tso_versions(struct efx_nic *efx)
2381 {
2382 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2383 	u32 tso_versions = 0;
2384 
2385 	if (nic_data->datapath_caps &
2386 	    (1 << MC_CMD_GET_CAPABILITIES_OUT_TX_TSO_LBN))
2387 		tso_versions |= BIT(1);
2388 	if (nic_data->datapath_caps2 &
2389 	    (1 << MC_CMD_GET_CAPABILITIES_V2_OUT_TX_TSO_V2_LBN))
2390 		tso_versions |= BIT(2);
2391 	return tso_versions;
2392 }
2393 
2394 static void efx_ef10_tx_init(struct efx_tx_queue *tx_queue)
2395 {
2396 	MCDI_DECLARE_BUF(inbuf, MC_CMD_INIT_TXQ_IN_LEN(EFX_MAX_DMAQ_SIZE * 8 /
2397 						       EFX_BUF_SIZE));
2398 	bool csum_offload = tx_queue->queue & EFX_TXQ_TYPE_OFFLOAD;
2399 	size_t entries = tx_queue->txd.buf.len / EFX_BUF_SIZE;
2400 	struct efx_channel *channel = tx_queue->channel;
2401 	struct efx_nic *efx = tx_queue->efx;
2402 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2403 	bool tso_v2 = false;
2404 	size_t inlen;
2405 	dma_addr_t dma_addr;
2406 	efx_qword_t *txd;
2407 	int rc;
2408 	int i;
2409 	BUILD_BUG_ON(MC_CMD_INIT_TXQ_OUT_LEN != 0);
2410 
2411 	/* TSOv2 is a limited resource that can only be configured on a limited
2412 	 * number of queues. TSO without checksum offload is not really a thing,
2413 	 * so we only enable it for those queues.
2414 	 */
2415 	if (csum_offload && (nic_data->datapath_caps2 &
2416 			(1 << MC_CMD_GET_CAPABILITIES_V2_OUT_TX_TSO_V2_LBN))) {
2417 		tso_v2 = true;
2418 		netif_dbg(efx, hw, efx->net_dev, "Using TSOv2 for channel %u\n",
2419 				channel->channel);
2420 	}
2421 
2422 	MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_SIZE, tx_queue->ptr_mask + 1);
2423 	MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_TARGET_EVQ, channel->channel);
2424 	MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_LABEL, tx_queue->queue);
2425 	MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_INSTANCE, tx_queue->queue);
2426 	MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_OWNER_ID, 0);
2427 	MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_PORT_ID, nic_data->vport_id);
2428 
2429 	dma_addr = tx_queue->txd.buf.dma_addr;
2430 
2431 	netif_dbg(efx, hw, efx->net_dev, "pushing TXQ %d. %zu entries (%llx)\n",
2432 		  tx_queue->queue, entries, (u64)dma_addr);
2433 
2434 	for (i = 0; i < entries; ++i) {
2435 		MCDI_SET_ARRAY_QWORD(inbuf, INIT_TXQ_IN_DMA_ADDR, i, dma_addr);
2436 		dma_addr += EFX_BUF_SIZE;
2437 	}
2438 
2439 	inlen = MC_CMD_INIT_TXQ_IN_LEN(entries);
2440 
2441 	do {
2442 		MCDI_POPULATE_DWORD_3(inbuf, INIT_TXQ_IN_FLAGS,
2443 				/* This flag was removed from mcdi_pcol.h for
2444 				 * the non-_EXT version of INIT_TXQ.  However,
2445 				 * firmware still honours it.
2446 				 */
2447 				INIT_TXQ_EXT_IN_FLAG_TSOV2_EN, tso_v2,
2448 				INIT_TXQ_IN_FLAG_IP_CSUM_DIS, !csum_offload,
2449 				INIT_TXQ_IN_FLAG_TCP_CSUM_DIS, !csum_offload);
2450 
2451 		rc = efx_mcdi_rpc_quiet(efx, MC_CMD_INIT_TXQ, inbuf, inlen,
2452 					NULL, 0, NULL);
2453 		if (rc == -ENOSPC && tso_v2) {
2454 			/* Retry without TSOv2 if we're short on contexts. */
2455 			tso_v2 = false;
2456 			netif_warn(efx, probe, efx->net_dev,
2457 				   "TSOv2 context not available to segment in hardware. TCP performance may be reduced.\n");
2458 		} else if (rc) {
2459 			efx_mcdi_display_error(efx, MC_CMD_INIT_TXQ,
2460 					       MC_CMD_INIT_TXQ_EXT_IN_LEN,
2461 					       NULL, 0, rc);
2462 			goto fail;
2463 		}
2464 	} while (rc);
2465 
2466 	/* A previous user of this TX queue might have set us up the
2467 	 * bomb by writing a descriptor to the TX push collector but
2468 	 * not the doorbell.  (Each collector belongs to a port, not a
2469 	 * queue or function, so cannot easily be reset.)  We must
2470 	 * attempt to push a no-op descriptor in its place.
2471 	 */
2472 	tx_queue->buffer[0].flags = EFX_TX_BUF_OPTION;
2473 	tx_queue->insert_count = 1;
2474 	txd = efx_tx_desc(tx_queue, 0);
2475 	EFX_POPULATE_QWORD_4(*txd,
2476 			     ESF_DZ_TX_DESC_IS_OPT, true,
2477 			     ESF_DZ_TX_OPTION_TYPE,
2478 			     ESE_DZ_TX_OPTION_DESC_CRC_CSUM,
2479 			     ESF_DZ_TX_OPTION_UDP_TCP_CSUM, csum_offload,
2480 			     ESF_DZ_TX_OPTION_IP_CSUM, csum_offload);
2481 	tx_queue->write_count = 1;
2482 
2483 	if (tso_v2) {
2484 		tx_queue->handle_tso = efx_ef10_tx_tso_desc;
2485 		tx_queue->tso_version = 2;
2486 	} else if (nic_data->datapath_caps &
2487 			(1 << MC_CMD_GET_CAPABILITIES_OUT_TX_TSO_LBN)) {
2488 		tx_queue->tso_version = 1;
2489 	}
2490 
2491 	wmb();
2492 	efx_ef10_push_tx_desc(tx_queue, txd);
2493 
2494 	return;
2495 
2496 fail:
2497 	netdev_WARN(efx->net_dev, "failed to initialise TXQ %d\n",
2498 		    tx_queue->queue);
2499 }
2500 
2501 static void efx_ef10_tx_fini(struct efx_tx_queue *tx_queue)
2502 {
2503 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FINI_TXQ_IN_LEN);
2504 	MCDI_DECLARE_BUF_ERR(outbuf);
2505 	struct efx_nic *efx = tx_queue->efx;
2506 	size_t outlen;
2507 	int rc;
2508 
2509 	MCDI_SET_DWORD(inbuf, FINI_TXQ_IN_INSTANCE,
2510 		       tx_queue->queue);
2511 
2512 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FINI_TXQ, inbuf, sizeof(inbuf),
2513 			  outbuf, sizeof(outbuf), &outlen);
2514 
2515 	if (rc && rc != -EALREADY)
2516 		goto fail;
2517 
2518 	return;
2519 
2520 fail:
2521 	efx_mcdi_display_error(efx, MC_CMD_FINI_TXQ, MC_CMD_FINI_TXQ_IN_LEN,
2522 			       outbuf, outlen, rc);
2523 }
2524 
2525 static void efx_ef10_tx_remove(struct efx_tx_queue *tx_queue)
2526 {
2527 	efx_nic_free_buffer(tx_queue->efx, &tx_queue->txd.buf);
2528 }
2529 
2530 /* This writes to the TX_DESC_WPTR; write pointer for TX descriptor ring */
2531 static inline void efx_ef10_notify_tx_desc(struct efx_tx_queue *tx_queue)
2532 {
2533 	unsigned int write_ptr;
2534 	efx_dword_t reg;
2535 
2536 	write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
2537 	EFX_POPULATE_DWORD_1(reg, ERF_DZ_TX_DESC_WPTR_DWORD, write_ptr);
2538 	efx_writed_page(tx_queue->efx, &reg,
2539 			ER_DZ_TX_DESC_UPD_DWORD, tx_queue->queue);
2540 }
2541 
2542 #define EFX_EF10_MAX_TX_DESCRIPTOR_LEN 0x3fff
2543 
2544 static unsigned int efx_ef10_tx_limit_len(struct efx_tx_queue *tx_queue,
2545 					  dma_addr_t dma_addr, unsigned int len)
2546 {
2547 	if (len > EFX_EF10_MAX_TX_DESCRIPTOR_LEN) {
2548 		/* If we need to break across multiple descriptors we should
2549 		 * stop at a page boundary. This assumes the length limit is
2550 		 * greater than the page size.
2551 		 */
2552 		dma_addr_t end = dma_addr + EFX_EF10_MAX_TX_DESCRIPTOR_LEN;
2553 
2554 		BUILD_BUG_ON(EFX_EF10_MAX_TX_DESCRIPTOR_LEN < EFX_PAGE_SIZE);
2555 		len = (end & (~(EFX_PAGE_SIZE - 1))) - dma_addr;
2556 	}
2557 
2558 	return len;
2559 }
2560 
2561 static void efx_ef10_tx_write(struct efx_tx_queue *tx_queue)
2562 {
2563 	unsigned int old_write_count = tx_queue->write_count;
2564 	struct efx_tx_buffer *buffer;
2565 	unsigned int write_ptr;
2566 	efx_qword_t *txd;
2567 
2568 	tx_queue->xmit_more_available = false;
2569 	if (unlikely(tx_queue->write_count == tx_queue->insert_count))
2570 		return;
2571 
2572 	do {
2573 		write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
2574 		buffer = &tx_queue->buffer[write_ptr];
2575 		txd = efx_tx_desc(tx_queue, write_ptr);
2576 		++tx_queue->write_count;
2577 
2578 		/* Create TX descriptor ring entry */
2579 		if (buffer->flags & EFX_TX_BUF_OPTION) {
2580 			*txd = buffer->option;
2581 			if (EFX_QWORD_FIELD(*txd, ESF_DZ_TX_OPTION_TYPE) == 1)
2582 				/* PIO descriptor */
2583 				tx_queue->packet_write_count = tx_queue->write_count;
2584 		} else {
2585 			tx_queue->packet_write_count = tx_queue->write_count;
2586 			BUILD_BUG_ON(EFX_TX_BUF_CONT != 1);
2587 			EFX_POPULATE_QWORD_3(
2588 				*txd,
2589 				ESF_DZ_TX_KER_CONT,
2590 				buffer->flags & EFX_TX_BUF_CONT,
2591 				ESF_DZ_TX_KER_BYTE_CNT, buffer->len,
2592 				ESF_DZ_TX_KER_BUF_ADDR, buffer->dma_addr);
2593 		}
2594 	} while (tx_queue->write_count != tx_queue->insert_count);
2595 
2596 	wmb(); /* Ensure descriptors are written before they are fetched */
2597 
2598 	if (efx_nic_may_push_tx_desc(tx_queue, old_write_count)) {
2599 		txd = efx_tx_desc(tx_queue,
2600 				  old_write_count & tx_queue->ptr_mask);
2601 		efx_ef10_push_tx_desc(tx_queue, txd);
2602 		++tx_queue->pushes;
2603 	} else {
2604 		efx_ef10_notify_tx_desc(tx_queue);
2605 	}
2606 }
2607 
2608 #define RSS_MODE_HASH_ADDRS	(1 << RSS_MODE_HASH_SRC_ADDR_LBN |\
2609 				 1 << RSS_MODE_HASH_DST_ADDR_LBN)
2610 #define RSS_MODE_HASH_PORTS	(1 << RSS_MODE_HASH_SRC_PORT_LBN |\
2611 				 1 << RSS_MODE_HASH_DST_PORT_LBN)
2612 #define RSS_CONTEXT_FLAGS_DEFAULT	(1 << MC_CMD_RSS_CONTEXT_GET_FLAGS_OUT_TOEPLITZ_IPV4_EN_LBN |\
2613 					 1 << MC_CMD_RSS_CONTEXT_GET_FLAGS_OUT_TOEPLITZ_TCPV4_EN_LBN |\
2614 					 1 << MC_CMD_RSS_CONTEXT_GET_FLAGS_OUT_TOEPLITZ_IPV6_EN_LBN |\
2615 					 1 << MC_CMD_RSS_CONTEXT_GET_FLAGS_OUT_TOEPLITZ_TCPV6_EN_LBN |\
2616 					 (RSS_MODE_HASH_ADDRS | RSS_MODE_HASH_PORTS) << MC_CMD_RSS_CONTEXT_GET_FLAGS_OUT_TCP_IPV4_RSS_MODE_LBN |\
2617 					 RSS_MODE_HASH_ADDRS << MC_CMD_RSS_CONTEXT_GET_FLAGS_OUT_UDP_IPV4_RSS_MODE_LBN |\
2618 					 RSS_MODE_HASH_ADDRS << MC_CMD_RSS_CONTEXT_GET_FLAGS_OUT_OTHER_IPV4_RSS_MODE_LBN |\
2619 					 (RSS_MODE_HASH_ADDRS | RSS_MODE_HASH_PORTS) << MC_CMD_RSS_CONTEXT_GET_FLAGS_OUT_TCP_IPV6_RSS_MODE_LBN |\
2620 					 RSS_MODE_HASH_ADDRS << MC_CMD_RSS_CONTEXT_GET_FLAGS_OUT_UDP_IPV6_RSS_MODE_LBN |\
2621 					 RSS_MODE_HASH_ADDRS << MC_CMD_RSS_CONTEXT_GET_FLAGS_OUT_OTHER_IPV6_RSS_MODE_LBN)
2622 
2623 static int efx_ef10_get_rss_flags(struct efx_nic *efx, u32 context, u32 *flags)
2624 {
2625 	/* Firmware had a bug (sfc bug 61952) where it would not actually
2626 	 * fill in the flags field in the response to MC_CMD_RSS_CONTEXT_GET_FLAGS.
2627 	 * This meant that it would always contain whatever was previously
2628 	 * in the MCDI buffer.  Fortunately, all firmware versions with
2629 	 * this bug have the same default flags value for a newly-allocated
2630 	 * RSS context, and the only time we want to get the flags is just
2631 	 * after allocating.  Moreover, the response has a 32-bit hole
2632 	 * where the context ID would be in the request, so we can use an
2633 	 * overlength buffer in the request and pre-fill the flags field
2634 	 * with what we believe the default to be.  Thus if the firmware
2635 	 * has the bug, it will leave our pre-filled value in the flags
2636 	 * field of the response, and we will get the right answer.
2637 	 *
2638 	 * However, this does mean that this function should NOT be used if
2639 	 * the RSS context flags might not be their defaults - it is ONLY
2640 	 * reliably correct for a newly-allocated RSS context.
2641 	 */
2642 	MCDI_DECLARE_BUF(inbuf, MC_CMD_RSS_CONTEXT_GET_FLAGS_OUT_LEN);
2643 	MCDI_DECLARE_BUF(outbuf, MC_CMD_RSS_CONTEXT_GET_FLAGS_OUT_LEN);
2644 	size_t outlen;
2645 	int rc;
2646 
2647 	/* Check we have a hole for the context ID */
2648 	BUILD_BUG_ON(MC_CMD_RSS_CONTEXT_GET_FLAGS_IN_LEN != MC_CMD_RSS_CONTEXT_GET_FLAGS_OUT_FLAGS_OFST);
2649 	MCDI_SET_DWORD(inbuf, RSS_CONTEXT_GET_FLAGS_IN_RSS_CONTEXT_ID, context);
2650 	MCDI_SET_DWORD(inbuf, RSS_CONTEXT_GET_FLAGS_OUT_FLAGS,
2651 		       RSS_CONTEXT_FLAGS_DEFAULT);
2652 	rc = efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_GET_FLAGS, inbuf,
2653 			  sizeof(inbuf), outbuf, sizeof(outbuf), &outlen);
2654 	if (rc == 0) {
2655 		if (outlen < MC_CMD_RSS_CONTEXT_GET_FLAGS_OUT_LEN)
2656 			rc = -EIO;
2657 		else
2658 			*flags = MCDI_DWORD(outbuf, RSS_CONTEXT_GET_FLAGS_OUT_FLAGS);
2659 	}
2660 	return rc;
2661 }
2662 
2663 /* Attempt to enable 4-tuple UDP hashing on the specified RSS context.
2664  * If we fail, we just leave the RSS context at its default hash settings,
2665  * which is safe but may slightly reduce performance.
2666  * Defaults are 4-tuple for TCP and 2-tuple for UDP and other-IP, so we
2667  * just need to set the UDP ports flags (for both IP versions).
2668  */
2669 static void efx_ef10_set_rss_flags(struct efx_nic *efx, u32 context)
2670 {
2671 	MCDI_DECLARE_BUF(inbuf, MC_CMD_RSS_CONTEXT_SET_FLAGS_IN_LEN);
2672 	u32 flags;
2673 
2674 	BUILD_BUG_ON(MC_CMD_RSS_CONTEXT_SET_FLAGS_OUT_LEN != 0);
2675 
2676 	if (efx_ef10_get_rss_flags(efx, context, &flags) != 0)
2677 		return;
2678 	MCDI_SET_DWORD(inbuf, RSS_CONTEXT_SET_FLAGS_IN_RSS_CONTEXT_ID, context);
2679 	flags |= RSS_MODE_HASH_PORTS << MC_CMD_RSS_CONTEXT_GET_FLAGS_OUT_UDP_IPV4_RSS_MODE_LBN;
2680 	flags |= RSS_MODE_HASH_PORTS << MC_CMD_RSS_CONTEXT_GET_FLAGS_OUT_UDP_IPV6_RSS_MODE_LBN;
2681 	MCDI_SET_DWORD(inbuf, RSS_CONTEXT_SET_FLAGS_IN_FLAGS, flags);
2682 	if (!efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_SET_FLAGS, inbuf, sizeof(inbuf),
2683 			  NULL, 0, NULL))
2684 		/* Succeeded, so UDP 4-tuple is now enabled */
2685 		efx->rx_hash_udp_4tuple = true;
2686 }
2687 
2688 static int efx_ef10_alloc_rss_context(struct efx_nic *efx, u32 *context,
2689 				      bool exclusive, unsigned *context_size)
2690 {
2691 	MCDI_DECLARE_BUF(inbuf, MC_CMD_RSS_CONTEXT_ALLOC_IN_LEN);
2692 	MCDI_DECLARE_BUF(outbuf, MC_CMD_RSS_CONTEXT_ALLOC_OUT_LEN);
2693 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2694 	size_t outlen;
2695 	int rc;
2696 	u32 alloc_type = exclusive ?
2697 				MC_CMD_RSS_CONTEXT_ALLOC_IN_TYPE_EXCLUSIVE :
2698 				MC_CMD_RSS_CONTEXT_ALLOC_IN_TYPE_SHARED;
2699 	unsigned rss_spread = exclusive ?
2700 				efx->rss_spread :
2701 				min(rounddown_pow_of_two(efx->rss_spread),
2702 				    EFX_EF10_MAX_SHARED_RSS_CONTEXT_SIZE);
2703 
2704 	if (!exclusive && rss_spread == 1) {
2705 		*context = EFX_EF10_RSS_CONTEXT_INVALID;
2706 		if (context_size)
2707 			*context_size = 1;
2708 		return 0;
2709 	}
2710 
2711 	if (nic_data->datapath_caps &
2712 	    1 << MC_CMD_GET_CAPABILITIES_OUT_RX_RSS_LIMITED_LBN)
2713 		return -EOPNOTSUPP;
2714 
2715 	MCDI_SET_DWORD(inbuf, RSS_CONTEXT_ALLOC_IN_UPSTREAM_PORT_ID,
2716 		       nic_data->vport_id);
2717 	MCDI_SET_DWORD(inbuf, RSS_CONTEXT_ALLOC_IN_TYPE, alloc_type);
2718 	MCDI_SET_DWORD(inbuf, RSS_CONTEXT_ALLOC_IN_NUM_QUEUES, rss_spread);
2719 
2720 	rc = efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_ALLOC, inbuf, sizeof(inbuf),
2721 		outbuf, sizeof(outbuf), &outlen);
2722 	if (rc != 0)
2723 		return rc;
2724 
2725 	if (outlen < MC_CMD_RSS_CONTEXT_ALLOC_OUT_LEN)
2726 		return -EIO;
2727 
2728 	*context = MCDI_DWORD(outbuf, RSS_CONTEXT_ALLOC_OUT_RSS_CONTEXT_ID);
2729 
2730 	if (context_size)
2731 		*context_size = rss_spread;
2732 
2733 	if (nic_data->datapath_caps &
2734 	    1 << MC_CMD_GET_CAPABILITIES_OUT_ADDITIONAL_RSS_MODES_LBN)
2735 		efx_ef10_set_rss_flags(efx, *context);
2736 
2737 	return 0;
2738 }
2739 
2740 static void efx_ef10_free_rss_context(struct efx_nic *efx, u32 context)
2741 {
2742 	MCDI_DECLARE_BUF(inbuf, MC_CMD_RSS_CONTEXT_FREE_IN_LEN);
2743 	int rc;
2744 
2745 	MCDI_SET_DWORD(inbuf, RSS_CONTEXT_FREE_IN_RSS_CONTEXT_ID,
2746 		       context);
2747 
2748 	rc = efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_FREE, inbuf, sizeof(inbuf),
2749 			    NULL, 0, NULL);
2750 	WARN_ON(rc != 0);
2751 }
2752 
2753 static int efx_ef10_populate_rss_table(struct efx_nic *efx, u32 context,
2754 				       const u32 *rx_indir_table, const u8 *key)
2755 {
2756 	MCDI_DECLARE_BUF(tablebuf, MC_CMD_RSS_CONTEXT_SET_TABLE_IN_LEN);
2757 	MCDI_DECLARE_BUF(keybuf, MC_CMD_RSS_CONTEXT_SET_KEY_IN_LEN);
2758 	int i, rc;
2759 
2760 	MCDI_SET_DWORD(tablebuf, RSS_CONTEXT_SET_TABLE_IN_RSS_CONTEXT_ID,
2761 		       context);
2762 	BUILD_BUG_ON(ARRAY_SIZE(efx->rx_indir_table) !=
2763 		     MC_CMD_RSS_CONTEXT_SET_TABLE_IN_INDIRECTION_TABLE_LEN);
2764 
2765 	/* This iterates over the length of efx->rx_indir_table, but copies
2766 	 * bytes from rx_indir_table.  That's because the latter is a pointer
2767 	 * rather than an array, but should have the same length.
2768 	 * The efx->rx_hash_key loop below is similar.
2769 	 */
2770 	for (i = 0; i < ARRAY_SIZE(efx->rx_indir_table); ++i)
2771 		MCDI_PTR(tablebuf,
2772 			 RSS_CONTEXT_SET_TABLE_IN_INDIRECTION_TABLE)[i] =
2773 				(u8) rx_indir_table[i];
2774 
2775 	rc = efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_SET_TABLE, tablebuf,
2776 			  sizeof(tablebuf), NULL, 0, NULL);
2777 	if (rc != 0)
2778 		return rc;
2779 
2780 	MCDI_SET_DWORD(keybuf, RSS_CONTEXT_SET_KEY_IN_RSS_CONTEXT_ID,
2781 		       context);
2782 	BUILD_BUG_ON(ARRAY_SIZE(efx->rx_hash_key) !=
2783 		     MC_CMD_RSS_CONTEXT_SET_KEY_IN_TOEPLITZ_KEY_LEN);
2784 	for (i = 0; i < ARRAY_SIZE(efx->rx_hash_key); ++i)
2785 		MCDI_PTR(keybuf, RSS_CONTEXT_SET_KEY_IN_TOEPLITZ_KEY)[i] = key[i];
2786 
2787 	return efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_SET_KEY, keybuf,
2788 			    sizeof(keybuf), NULL, 0, NULL);
2789 }
2790 
2791 static void efx_ef10_rx_free_indir_table(struct efx_nic *efx)
2792 {
2793 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2794 
2795 	if (nic_data->rx_rss_context != EFX_EF10_RSS_CONTEXT_INVALID)
2796 		efx_ef10_free_rss_context(efx, nic_data->rx_rss_context);
2797 	nic_data->rx_rss_context = EFX_EF10_RSS_CONTEXT_INVALID;
2798 }
2799 
2800 static int efx_ef10_rx_push_shared_rss_config(struct efx_nic *efx,
2801 					      unsigned *context_size)
2802 {
2803 	u32 new_rx_rss_context;
2804 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2805 	int rc = efx_ef10_alloc_rss_context(efx, &new_rx_rss_context,
2806 					    false, context_size);
2807 
2808 	if (rc != 0)
2809 		return rc;
2810 
2811 	nic_data->rx_rss_context = new_rx_rss_context;
2812 	nic_data->rx_rss_context_exclusive = false;
2813 	efx_set_default_rx_indir_table(efx);
2814 	return 0;
2815 }
2816 
2817 static int efx_ef10_rx_push_exclusive_rss_config(struct efx_nic *efx,
2818 						 const u32 *rx_indir_table,
2819 						 const u8 *key)
2820 {
2821 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2822 	int rc;
2823 	u32 new_rx_rss_context;
2824 
2825 	if (nic_data->rx_rss_context == EFX_EF10_RSS_CONTEXT_INVALID ||
2826 	    !nic_data->rx_rss_context_exclusive) {
2827 		rc = efx_ef10_alloc_rss_context(efx, &new_rx_rss_context,
2828 						true, NULL);
2829 		if (rc == -EOPNOTSUPP)
2830 			return rc;
2831 		else if (rc != 0)
2832 			goto fail1;
2833 	} else {
2834 		new_rx_rss_context = nic_data->rx_rss_context;
2835 	}
2836 
2837 	rc = efx_ef10_populate_rss_table(efx, new_rx_rss_context,
2838 					 rx_indir_table, key);
2839 	if (rc != 0)
2840 		goto fail2;
2841 
2842 	if (nic_data->rx_rss_context != new_rx_rss_context)
2843 		efx_ef10_rx_free_indir_table(efx);
2844 	nic_data->rx_rss_context = new_rx_rss_context;
2845 	nic_data->rx_rss_context_exclusive = true;
2846 	if (rx_indir_table != efx->rx_indir_table)
2847 		memcpy(efx->rx_indir_table, rx_indir_table,
2848 		       sizeof(efx->rx_indir_table));
2849 	if (key != efx->rx_hash_key)
2850 		memcpy(efx->rx_hash_key, key, efx->type->rx_hash_key_size);
2851 
2852 	return 0;
2853 
2854 fail2:
2855 	if (new_rx_rss_context != nic_data->rx_rss_context)
2856 		efx_ef10_free_rss_context(efx, new_rx_rss_context);
2857 fail1:
2858 	netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
2859 	return rc;
2860 }
2861 
2862 static int efx_ef10_rx_pull_rss_config(struct efx_nic *efx)
2863 {
2864 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2865 	MCDI_DECLARE_BUF(inbuf, MC_CMD_RSS_CONTEXT_GET_TABLE_IN_LEN);
2866 	MCDI_DECLARE_BUF(tablebuf, MC_CMD_RSS_CONTEXT_GET_TABLE_OUT_LEN);
2867 	MCDI_DECLARE_BUF(keybuf, MC_CMD_RSS_CONTEXT_GET_KEY_OUT_LEN);
2868 	size_t outlen;
2869 	int rc, i;
2870 
2871 	BUILD_BUG_ON(MC_CMD_RSS_CONTEXT_GET_TABLE_IN_LEN !=
2872 		     MC_CMD_RSS_CONTEXT_GET_KEY_IN_LEN);
2873 
2874 	if (nic_data->rx_rss_context == EFX_EF10_RSS_CONTEXT_INVALID)
2875 		return -ENOENT;
2876 
2877 	MCDI_SET_DWORD(inbuf, RSS_CONTEXT_GET_TABLE_IN_RSS_CONTEXT_ID,
2878 		       nic_data->rx_rss_context);
2879 	BUILD_BUG_ON(ARRAY_SIZE(efx->rx_indir_table) !=
2880 		     MC_CMD_RSS_CONTEXT_GET_TABLE_OUT_INDIRECTION_TABLE_LEN);
2881 	rc = efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_GET_TABLE, inbuf, sizeof(inbuf),
2882 			  tablebuf, sizeof(tablebuf), &outlen);
2883 	if (rc != 0)
2884 		return rc;
2885 
2886 	if (WARN_ON(outlen != MC_CMD_RSS_CONTEXT_GET_TABLE_OUT_LEN))
2887 		return -EIO;
2888 
2889 	for (i = 0; i < ARRAY_SIZE(efx->rx_indir_table); i++)
2890 		efx->rx_indir_table[i] = MCDI_PTR(tablebuf,
2891 				RSS_CONTEXT_GET_TABLE_OUT_INDIRECTION_TABLE)[i];
2892 
2893 	MCDI_SET_DWORD(inbuf, RSS_CONTEXT_GET_KEY_IN_RSS_CONTEXT_ID,
2894 		       nic_data->rx_rss_context);
2895 	BUILD_BUG_ON(ARRAY_SIZE(efx->rx_hash_key) !=
2896 		     MC_CMD_RSS_CONTEXT_SET_KEY_IN_TOEPLITZ_KEY_LEN);
2897 	rc = efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_GET_KEY, inbuf, sizeof(inbuf),
2898 			  keybuf, sizeof(keybuf), &outlen);
2899 	if (rc != 0)
2900 		return rc;
2901 
2902 	if (WARN_ON(outlen != MC_CMD_RSS_CONTEXT_GET_KEY_OUT_LEN))
2903 		return -EIO;
2904 
2905 	for (i = 0; i < ARRAY_SIZE(efx->rx_hash_key); ++i)
2906 		efx->rx_hash_key[i] = MCDI_PTR(
2907 				keybuf, RSS_CONTEXT_GET_KEY_OUT_TOEPLITZ_KEY)[i];
2908 
2909 	return 0;
2910 }
2911 
2912 static int efx_ef10_pf_rx_push_rss_config(struct efx_nic *efx, bool user,
2913 					  const u32 *rx_indir_table,
2914 					  const u8 *key)
2915 {
2916 	int rc;
2917 
2918 	if (efx->rss_spread == 1)
2919 		return 0;
2920 
2921 	if (!key)
2922 		key = efx->rx_hash_key;
2923 
2924 	rc = efx_ef10_rx_push_exclusive_rss_config(efx, rx_indir_table, key);
2925 
2926 	if (rc == -ENOBUFS && !user) {
2927 		unsigned context_size;
2928 		bool mismatch = false;
2929 		size_t i;
2930 
2931 		for (i = 0; i < ARRAY_SIZE(efx->rx_indir_table) && !mismatch;
2932 		     i++)
2933 			mismatch = rx_indir_table[i] !=
2934 				ethtool_rxfh_indir_default(i, efx->rss_spread);
2935 
2936 		rc = efx_ef10_rx_push_shared_rss_config(efx, &context_size);
2937 		if (rc == 0) {
2938 			if (context_size != efx->rss_spread)
2939 				netif_warn(efx, probe, efx->net_dev,
2940 					   "Could not allocate an exclusive RSS"
2941 					   " context; allocated a shared one of"
2942 					   " different size."
2943 					   " Wanted %u, got %u.\n",
2944 					   efx->rss_spread, context_size);
2945 			else if (mismatch)
2946 				netif_warn(efx, probe, efx->net_dev,
2947 					   "Could not allocate an exclusive RSS"
2948 					   " context; allocated a shared one but"
2949 					   " could not apply custom"
2950 					   " indirection.\n");
2951 			else
2952 				netif_info(efx, probe, efx->net_dev,
2953 					   "Could not allocate an exclusive RSS"
2954 					   " context; allocated a shared one.\n");
2955 		}
2956 	}
2957 	return rc;
2958 }
2959 
2960 static int efx_ef10_vf_rx_push_rss_config(struct efx_nic *efx, bool user,
2961 					  const u32 *rx_indir_table
2962 					  __attribute__ ((unused)),
2963 					  const u8 *key
2964 					  __attribute__ ((unused)))
2965 {
2966 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2967 
2968 	if (user)
2969 		return -EOPNOTSUPP;
2970 	if (nic_data->rx_rss_context != EFX_EF10_RSS_CONTEXT_INVALID)
2971 		return 0;
2972 	return efx_ef10_rx_push_shared_rss_config(efx, NULL);
2973 }
2974 
2975 static int efx_ef10_rx_probe(struct efx_rx_queue *rx_queue)
2976 {
2977 	return efx_nic_alloc_buffer(rx_queue->efx, &rx_queue->rxd.buf,
2978 				    (rx_queue->ptr_mask + 1) *
2979 				    sizeof(efx_qword_t),
2980 				    GFP_KERNEL);
2981 }
2982 
2983 static void efx_ef10_rx_init(struct efx_rx_queue *rx_queue)
2984 {
2985 	MCDI_DECLARE_BUF(inbuf,
2986 			 MC_CMD_INIT_RXQ_IN_LEN(EFX_MAX_DMAQ_SIZE * 8 /
2987 						EFX_BUF_SIZE));
2988 	struct efx_channel *channel = efx_rx_queue_channel(rx_queue);
2989 	size_t entries = rx_queue->rxd.buf.len / EFX_BUF_SIZE;
2990 	struct efx_nic *efx = rx_queue->efx;
2991 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2992 	size_t inlen;
2993 	dma_addr_t dma_addr;
2994 	int rc;
2995 	int i;
2996 	BUILD_BUG_ON(MC_CMD_INIT_RXQ_OUT_LEN != 0);
2997 
2998 	rx_queue->scatter_n = 0;
2999 	rx_queue->scatter_len = 0;
3000 
3001 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_SIZE, rx_queue->ptr_mask + 1);
3002 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_TARGET_EVQ, channel->channel);
3003 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_LABEL, efx_rx_queue_index(rx_queue));
3004 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_INSTANCE,
3005 		       efx_rx_queue_index(rx_queue));
3006 	MCDI_POPULATE_DWORD_2(inbuf, INIT_RXQ_IN_FLAGS,
3007 			      INIT_RXQ_IN_FLAG_PREFIX, 1,
3008 			      INIT_RXQ_IN_FLAG_TIMESTAMP, 1);
3009 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_OWNER_ID, 0);
3010 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_PORT_ID, nic_data->vport_id);
3011 
3012 	dma_addr = rx_queue->rxd.buf.dma_addr;
3013 
3014 	netif_dbg(efx, hw, efx->net_dev, "pushing RXQ %d. %zu entries (%llx)\n",
3015 		  efx_rx_queue_index(rx_queue), entries, (u64)dma_addr);
3016 
3017 	for (i = 0; i < entries; ++i) {
3018 		MCDI_SET_ARRAY_QWORD(inbuf, INIT_RXQ_IN_DMA_ADDR, i, dma_addr);
3019 		dma_addr += EFX_BUF_SIZE;
3020 	}
3021 
3022 	inlen = MC_CMD_INIT_RXQ_IN_LEN(entries);
3023 
3024 	rc = efx_mcdi_rpc(efx, MC_CMD_INIT_RXQ, inbuf, inlen,
3025 			  NULL, 0, NULL);
3026 	if (rc)
3027 		netdev_WARN(efx->net_dev, "failed to initialise RXQ %d\n",
3028 			    efx_rx_queue_index(rx_queue));
3029 }
3030 
3031 static void efx_ef10_rx_fini(struct efx_rx_queue *rx_queue)
3032 {
3033 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FINI_RXQ_IN_LEN);
3034 	MCDI_DECLARE_BUF_ERR(outbuf);
3035 	struct efx_nic *efx = rx_queue->efx;
3036 	size_t outlen;
3037 	int rc;
3038 
3039 	MCDI_SET_DWORD(inbuf, FINI_RXQ_IN_INSTANCE,
3040 		       efx_rx_queue_index(rx_queue));
3041 
3042 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FINI_RXQ, inbuf, sizeof(inbuf),
3043 			  outbuf, sizeof(outbuf), &outlen);
3044 
3045 	if (rc && rc != -EALREADY)
3046 		goto fail;
3047 
3048 	return;
3049 
3050 fail:
3051 	efx_mcdi_display_error(efx, MC_CMD_FINI_RXQ, MC_CMD_FINI_RXQ_IN_LEN,
3052 			       outbuf, outlen, rc);
3053 }
3054 
3055 static void efx_ef10_rx_remove(struct efx_rx_queue *rx_queue)
3056 {
3057 	efx_nic_free_buffer(rx_queue->efx, &rx_queue->rxd.buf);
3058 }
3059 
3060 /* This creates an entry in the RX descriptor queue */
3061 static inline void
3062 efx_ef10_build_rx_desc(struct efx_rx_queue *rx_queue, unsigned int index)
3063 {
3064 	struct efx_rx_buffer *rx_buf;
3065 	efx_qword_t *rxd;
3066 
3067 	rxd = efx_rx_desc(rx_queue, index);
3068 	rx_buf = efx_rx_buffer(rx_queue, index);
3069 	EFX_POPULATE_QWORD_2(*rxd,
3070 			     ESF_DZ_RX_KER_BYTE_CNT, rx_buf->len,
3071 			     ESF_DZ_RX_KER_BUF_ADDR, rx_buf->dma_addr);
3072 }
3073 
3074 static void efx_ef10_rx_write(struct efx_rx_queue *rx_queue)
3075 {
3076 	struct efx_nic *efx = rx_queue->efx;
3077 	unsigned int write_count;
3078 	efx_dword_t reg;
3079 
3080 	/* Firmware requires that RX_DESC_WPTR be a multiple of 8 */
3081 	write_count = rx_queue->added_count & ~7;
3082 	if (rx_queue->notified_count == write_count)
3083 		return;
3084 
3085 	do
3086 		efx_ef10_build_rx_desc(
3087 			rx_queue,
3088 			rx_queue->notified_count & rx_queue->ptr_mask);
3089 	while (++rx_queue->notified_count != write_count);
3090 
3091 	wmb();
3092 	EFX_POPULATE_DWORD_1(reg, ERF_DZ_RX_DESC_WPTR,
3093 			     write_count & rx_queue->ptr_mask);
3094 	efx_writed_page(efx, &reg, ER_DZ_RX_DESC_UPD,
3095 			efx_rx_queue_index(rx_queue));
3096 }
3097 
3098 static efx_mcdi_async_completer efx_ef10_rx_defer_refill_complete;
3099 
3100 static void efx_ef10_rx_defer_refill(struct efx_rx_queue *rx_queue)
3101 {
3102 	struct efx_channel *channel = efx_rx_queue_channel(rx_queue);
3103 	MCDI_DECLARE_BUF(inbuf, MC_CMD_DRIVER_EVENT_IN_LEN);
3104 	efx_qword_t event;
3105 
3106 	EFX_POPULATE_QWORD_2(event,
3107 			     ESF_DZ_EV_CODE, EFX_EF10_DRVGEN_EV,
3108 			     ESF_DZ_EV_DATA, EFX_EF10_REFILL);
3109 
3110 	MCDI_SET_DWORD(inbuf, DRIVER_EVENT_IN_EVQ, channel->channel);
3111 
3112 	/* MCDI_SET_QWORD is not appropriate here since EFX_POPULATE_* has
3113 	 * already swapped the data to little-endian order.
3114 	 */
3115 	memcpy(MCDI_PTR(inbuf, DRIVER_EVENT_IN_DATA), &event.u64[0],
3116 	       sizeof(efx_qword_t));
3117 
3118 	efx_mcdi_rpc_async(channel->efx, MC_CMD_DRIVER_EVENT,
3119 			   inbuf, sizeof(inbuf), 0,
3120 			   efx_ef10_rx_defer_refill_complete, 0);
3121 }
3122 
3123 static void
3124 efx_ef10_rx_defer_refill_complete(struct efx_nic *efx, unsigned long cookie,
3125 				  int rc, efx_dword_t *outbuf,
3126 				  size_t outlen_actual)
3127 {
3128 	/* nothing to do */
3129 }
3130 
3131 static int efx_ef10_ev_probe(struct efx_channel *channel)
3132 {
3133 	return efx_nic_alloc_buffer(channel->efx, &channel->eventq.buf,
3134 				    (channel->eventq_mask + 1) *
3135 				    sizeof(efx_qword_t),
3136 				    GFP_KERNEL);
3137 }
3138 
3139 static void efx_ef10_ev_fini(struct efx_channel *channel)
3140 {
3141 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FINI_EVQ_IN_LEN);
3142 	MCDI_DECLARE_BUF_ERR(outbuf);
3143 	struct efx_nic *efx = channel->efx;
3144 	size_t outlen;
3145 	int rc;
3146 
3147 	MCDI_SET_DWORD(inbuf, FINI_EVQ_IN_INSTANCE, channel->channel);
3148 
3149 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FINI_EVQ, inbuf, sizeof(inbuf),
3150 			  outbuf, sizeof(outbuf), &outlen);
3151 
3152 	if (rc && rc != -EALREADY)
3153 		goto fail;
3154 
3155 	return;
3156 
3157 fail:
3158 	efx_mcdi_display_error(efx, MC_CMD_FINI_EVQ, MC_CMD_FINI_EVQ_IN_LEN,
3159 			       outbuf, outlen, rc);
3160 }
3161 
3162 static int efx_ef10_ev_init(struct efx_channel *channel)
3163 {
3164 	MCDI_DECLARE_BUF(inbuf,
3165 			 MC_CMD_INIT_EVQ_V2_IN_LEN(EFX_MAX_EVQ_SIZE * 8 /
3166 						   EFX_BUF_SIZE));
3167 	MCDI_DECLARE_BUF(outbuf, MC_CMD_INIT_EVQ_V2_OUT_LEN);
3168 	size_t entries = channel->eventq.buf.len / EFX_BUF_SIZE;
3169 	struct efx_nic *efx = channel->efx;
3170 	struct efx_ef10_nic_data *nic_data;
3171 	size_t inlen, outlen;
3172 	unsigned int enabled, implemented;
3173 	dma_addr_t dma_addr;
3174 	int rc;
3175 	int i;
3176 
3177 	nic_data = efx->nic_data;
3178 
3179 	/* Fill event queue with all ones (i.e. empty events) */
3180 	memset(channel->eventq.buf.addr, 0xff, channel->eventq.buf.len);
3181 
3182 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_SIZE, channel->eventq_mask + 1);
3183 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_INSTANCE, channel->channel);
3184 	/* INIT_EVQ expects index in vector table, not absolute */
3185 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_IRQ_NUM, channel->channel);
3186 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_TMR_MODE,
3187 		       MC_CMD_INIT_EVQ_IN_TMR_MODE_DIS);
3188 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_TMR_LOAD, 0);
3189 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_TMR_RELOAD, 0);
3190 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_COUNT_MODE,
3191 		       MC_CMD_INIT_EVQ_IN_COUNT_MODE_DIS);
3192 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_COUNT_THRSHLD, 0);
3193 
3194 	if (nic_data->datapath_caps2 &
3195 	    1 << MC_CMD_GET_CAPABILITIES_V2_OUT_INIT_EVQ_V2_LBN) {
3196 		/* Use the new generic approach to specifying event queue
3197 		 * configuration, requesting lower latency or higher throughput.
3198 		 * The options that actually get used appear in the output.
3199 		 */
3200 		MCDI_POPULATE_DWORD_2(inbuf, INIT_EVQ_V2_IN_FLAGS,
3201 				      INIT_EVQ_V2_IN_FLAG_INTERRUPTING, 1,
3202 				      INIT_EVQ_V2_IN_FLAG_TYPE,
3203 				      MC_CMD_INIT_EVQ_V2_IN_FLAG_TYPE_AUTO);
3204 	} else {
3205 		bool cut_thru = !(nic_data->datapath_caps &
3206 			1 << MC_CMD_GET_CAPABILITIES_OUT_RX_BATCHING_LBN);
3207 
3208 		MCDI_POPULATE_DWORD_4(inbuf, INIT_EVQ_IN_FLAGS,
3209 				      INIT_EVQ_IN_FLAG_INTERRUPTING, 1,
3210 				      INIT_EVQ_IN_FLAG_RX_MERGE, 1,
3211 				      INIT_EVQ_IN_FLAG_TX_MERGE, 1,
3212 				      INIT_EVQ_IN_FLAG_CUT_THRU, cut_thru);
3213 	}
3214 
3215 	dma_addr = channel->eventq.buf.dma_addr;
3216 	for (i = 0; i < entries; ++i) {
3217 		MCDI_SET_ARRAY_QWORD(inbuf, INIT_EVQ_IN_DMA_ADDR, i, dma_addr);
3218 		dma_addr += EFX_BUF_SIZE;
3219 	}
3220 
3221 	inlen = MC_CMD_INIT_EVQ_IN_LEN(entries);
3222 
3223 	rc = efx_mcdi_rpc(efx, MC_CMD_INIT_EVQ, inbuf, inlen,
3224 			  outbuf, sizeof(outbuf), &outlen);
3225 
3226 	if (outlen >= MC_CMD_INIT_EVQ_V2_OUT_LEN)
3227 		netif_dbg(efx, drv, efx->net_dev,
3228 			  "Channel %d using event queue flags %08x\n",
3229 			  channel->channel,
3230 			  MCDI_DWORD(outbuf, INIT_EVQ_V2_OUT_FLAGS));
3231 
3232 	/* IRQ return is ignored */
3233 	if (channel->channel || rc)
3234 		return rc;
3235 
3236 	/* Successfully created event queue on channel 0 */
3237 	rc = efx_mcdi_get_workarounds(efx, &implemented, &enabled);
3238 	if (rc == -ENOSYS) {
3239 		/* GET_WORKAROUNDS was implemented before this workaround,
3240 		 * thus it must be unavailable in this firmware.
3241 		 */
3242 		nic_data->workaround_26807 = false;
3243 		rc = 0;
3244 	} else if (rc) {
3245 		goto fail;
3246 	} else {
3247 		nic_data->workaround_26807 =
3248 			!!(enabled & MC_CMD_GET_WORKAROUNDS_OUT_BUG26807);
3249 
3250 		if (implemented & MC_CMD_GET_WORKAROUNDS_OUT_BUG26807 &&
3251 		    !nic_data->workaround_26807) {
3252 			unsigned int flags;
3253 
3254 			rc = efx_mcdi_set_workaround(efx,
3255 						     MC_CMD_WORKAROUND_BUG26807,
3256 						     true, &flags);
3257 
3258 			if (!rc) {
3259 				if (flags &
3260 				    1 << MC_CMD_WORKAROUND_EXT_OUT_FLR_DONE_LBN) {
3261 					netif_info(efx, drv, efx->net_dev,
3262 						   "other functions on NIC have been reset\n");
3263 
3264 					/* With MCFW v4.6.x and earlier, the
3265 					 * boot count will have incremented,
3266 					 * so re-read the warm_boot_count
3267 					 * value now to ensure this function
3268 					 * doesn't think it has changed next
3269 					 * time it checks.
3270 					 */
3271 					rc = efx_ef10_get_warm_boot_count(efx);
3272 					if (rc >= 0) {
3273 						nic_data->warm_boot_count = rc;
3274 						rc = 0;
3275 					}
3276 				}
3277 				nic_data->workaround_26807 = true;
3278 			} else if (rc == -EPERM) {
3279 				rc = 0;
3280 			}
3281 		}
3282 	}
3283 
3284 	if (!rc)
3285 		return 0;
3286 
3287 fail:
3288 	efx_ef10_ev_fini(channel);
3289 	return rc;
3290 }
3291 
3292 static void efx_ef10_ev_remove(struct efx_channel *channel)
3293 {
3294 	efx_nic_free_buffer(channel->efx, &channel->eventq.buf);
3295 }
3296 
3297 static void efx_ef10_handle_rx_wrong_queue(struct efx_rx_queue *rx_queue,
3298 					   unsigned int rx_queue_label)
3299 {
3300 	struct efx_nic *efx = rx_queue->efx;
3301 
3302 	netif_info(efx, hw, efx->net_dev,
3303 		   "rx event arrived on queue %d labeled as queue %u\n",
3304 		   efx_rx_queue_index(rx_queue), rx_queue_label);
3305 
3306 	efx_schedule_reset(efx, RESET_TYPE_DISABLE);
3307 }
3308 
3309 static void
3310 efx_ef10_handle_rx_bad_lbits(struct efx_rx_queue *rx_queue,
3311 			     unsigned int actual, unsigned int expected)
3312 {
3313 	unsigned int dropped = (actual - expected) & rx_queue->ptr_mask;
3314 	struct efx_nic *efx = rx_queue->efx;
3315 
3316 	netif_info(efx, hw, efx->net_dev,
3317 		   "dropped %d events (index=%d expected=%d)\n",
3318 		   dropped, actual, expected);
3319 
3320 	efx_schedule_reset(efx, RESET_TYPE_DISABLE);
3321 }
3322 
3323 /* partially received RX was aborted. clean up. */
3324 static void efx_ef10_handle_rx_abort(struct efx_rx_queue *rx_queue)
3325 {
3326 	unsigned int rx_desc_ptr;
3327 
3328 	netif_dbg(rx_queue->efx, hw, rx_queue->efx->net_dev,
3329 		  "scattered RX aborted (dropping %u buffers)\n",
3330 		  rx_queue->scatter_n);
3331 
3332 	rx_desc_ptr = rx_queue->removed_count & rx_queue->ptr_mask;
3333 
3334 	efx_rx_packet(rx_queue, rx_desc_ptr, rx_queue->scatter_n,
3335 		      0, EFX_RX_PKT_DISCARD);
3336 
3337 	rx_queue->removed_count += rx_queue->scatter_n;
3338 	rx_queue->scatter_n = 0;
3339 	rx_queue->scatter_len = 0;
3340 	++efx_rx_queue_channel(rx_queue)->n_rx_nodesc_trunc;
3341 }
3342 
3343 static u16 efx_ef10_handle_rx_event_errors(struct efx_channel *channel,
3344 					   unsigned int n_packets,
3345 					   unsigned int rx_encap_hdr,
3346 					   unsigned int rx_l3_class,
3347 					   unsigned int rx_l4_class,
3348 					   const efx_qword_t *event)
3349 {
3350 	struct efx_nic *efx = channel->efx;
3351 	bool handled = false;
3352 
3353 	if (EFX_QWORD_FIELD(*event, ESF_DZ_RX_ECRC_ERR)) {
3354 		if (!(efx->net_dev->features & NETIF_F_RXALL)) {
3355 			if (!efx->loopback_selftest)
3356 				channel->n_rx_eth_crc_err += n_packets;
3357 			return EFX_RX_PKT_DISCARD;
3358 		}
3359 		handled = true;
3360 	}
3361 	if (EFX_QWORD_FIELD(*event, ESF_DZ_RX_IPCKSUM_ERR)) {
3362 		if (unlikely(rx_encap_hdr != ESE_EZ_ENCAP_HDR_VXLAN &&
3363 			     rx_l3_class != ESE_DZ_L3_CLASS_IP4 &&
3364 			     rx_l3_class != ESE_DZ_L3_CLASS_IP4_FRAG &&
3365 			     rx_l3_class != ESE_DZ_L3_CLASS_IP6 &&
3366 			     rx_l3_class != ESE_DZ_L3_CLASS_IP6_FRAG))
3367 			netdev_WARN(efx->net_dev,
3368 				    "invalid class for RX_IPCKSUM_ERR: event="
3369 				    EFX_QWORD_FMT "\n",
3370 				    EFX_QWORD_VAL(*event));
3371 		if (!efx->loopback_selftest)
3372 			*(rx_encap_hdr ?
3373 			  &channel->n_rx_outer_ip_hdr_chksum_err :
3374 			  &channel->n_rx_ip_hdr_chksum_err) += n_packets;
3375 		return 0;
3376 	}
3377 	if (EFX_QWORD_FIELD(*event, ESF_DZ_RX_TCPUDP_CKSUM_ERR)) {
3378 		if (unlikely(rx_encap_hdr != ESE_EZ_ENCAP_HDR_VXLAN &&
3379 			     ((rx_l3_class != ESE_DZ_L3_CLASS_IP4 &&
3380 			       rx_l3_class != ESE_DZ_L3_CLASS_IP6) ||
3381 			      (rx_l4_class != ESE_FZ_L4_CLASS_TCP &&
3382 			       rx_l4_class != ESE_FZ_L4_CLASS_UDP))))
3383 			netdev_WARN(efx->net_dev,
3384 				    "invalid class for RX_TCPUDP_CKSUM_ERR: event="
3385 				    EFX_QWORD_FMT "\n",
3386 				    EFX_QWORD_VAL(*event));
3387 		if (!efx->loopback_selftest)
3388 			*(rx_encap_hdr ?
3389 			  &channel->n_rx_outer_tcp_udp_chksum_err :
3390 			  &channel->n_rx_tcp_udp_chksum_err) += n_packets;
3391 		return 0;
3392 	}
3393 	if (EFX_QWORD_FIELD(*event, ESF_EZ_RX_IP_INNER_CHKSUM_ERR)) {
3394 		if (unlikely(!rx_encap_hdr))
3395 			netdev_WARN(efx->net_dev,
3396 				    "invalid encapsulation type for RX_IP_INNER_CHKSUM_ERR: event="
3397 				    EFX_QWORD_FMT "\n",
3398 				    EFX_QWORD_VAL(*event));
3399 		else if (unlikely(rx_l3_class != ESE_DZ_L3_CLASS_IP4 &&
3400 				  rx_l3_class != ESE_DZ_L3_CLASS_IP4_FRAG &&
3401 				  rx_l3_class != ESE_DZ_L3_CLASS_IP6 &&
3402 				  rx_l3_class != ESE_DZ_L3_CLASS_IP6_FRAG))
3403 			netdev_WARN(efx->net_dev,
3404 				    "invalid class for RX_IP_INNER_CHKSUM_ERR: event="
3405 				    EFX_QWORD_FMT "\n",
3406 				    EFX_QWORD_VAL(*event));
3407 		if (!efx->loopback_selftest)
3408 			channel->n_rx_inner_ip_hdr_chksum_err += n_packets;
3409 		return 0;
3410 	}
3411 	if (EFX_QWORD_FIELD(*event, ESF_EZ_RX_TCP_UDP_INNER_CHKSUM_ERR)) {
3412 		if (unlikely(!rx_encap_hdr))
3413 			netdev_WARN(efx->net_dev,
3414 				    "invalid encapsulation type for RX_TCP_UDP_INNER_CHKSUM_ERR: event="
3415 				    EFX_QWORD_FMT "\n",
3416 				    EFX_QWORD_VAL(*event));
3417 		else if (unlikely((rx_l3_class != ESE_DZ_L3_CLASS_IP4 &&
3418 				   rx_l3_class != ESE_DZ_L3_CLASS_IP6) ||
3419 				  (rx_l4_class != ESE_FZ_L4_CLASS_TCP &&
3420 				   rx_l4_class != ESE_FZ_L4_CLASS_UDP)))
3421 			netdev_WARN(efx->net_dev,
3422 				    "invalid class for RX_TCP_UDP_INNER_CHKSUM_ERR: event="
3423 				    EFX_QWORD_FMT "\n",
3424 				    EFX_QWORD_VAL(*event));
3425 		if (!efx->loopback_selftest)
3426 			channel->n_rx_inner_tcp_udp_chksum_err += n_packets;
3427 		return 0;
3428 	}
3429 
3430 	WARN_ON(!handled); /* No error bits were recognised */
3431 	return 0;
3432 }
3433 
3434 static int efx_ef10_handle_rx_event(struct efx_channel *channel,
3435 				    const efx_qword_t *event)
3436 {
3437 	unsigned int rx_bytes, next_ptr_lbits, rx_queue_label;
3438 	unsigned int rx_l3_class, rx_l4_class, rx_encap_hdr;
3439 	unsigned int n_descs, n_packets, i;
3440 	struct efx_nic *efx = channel->efx;
3441 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
3442 	struct efx_rx_queue *rx_queue;
3443 	efx_qword_t errors;
3444 	bool rx_cont;
3445 	u16 flags = 0;
3446 
3447 	if (unlikely(READ_ONCE(efx->reset_pending)))
3448 		return 0;
3449 
3450 	/* Basic packet information */
3451 	rx_bytes = EFX_QWORD_FIELD(*event, ESF_DZ_RX_BYTES);
3452 	next_ptr_lbits = EFX_QWORD_FIELD(*event, ESF_DZ_RX_DSC_PTR_LBITS);
3453 	rx_queue_label = EFX_QWORD_FIELD(*event, ESF_DZ_RX_QLABEL);
3454 	rx_l3_class = EFX_QWORD_FIELD(*event, ESF_DZ_RX_L3_CLASS);
3455 	rx_l4_class = EFX_QWORD_FIELD(*event, ESF_FZ_RX_L4_CLASS);
3456 	rx_cont = EFX_QWORD_FIELD(*event, ESF_DZ_RX_CONT);
3457 	rx_encap_hdr =
3458 		nic_data->datapath_caps &
3459 			(1 << MC_CMD_GET_CAPABILITIES_OUT_VXLAN_NVGRE_LBN) ?
3460 		EFX_QWORD_FIELD(*event, ESF_EZ_RX_ENCAP_HDR) :
3461 		ESE_EZ_ENCAP_HDR_NONE;
3462 
3463 	if (EFX_QWORD_FIELD(*event, ESF_DZ_RX_DROP_EVENT))
3464 		netdev_WARN(efx->net_dev, "saw RX_DROP_EVENT: event="
3465 			    EFX_QWORD_FMT "\n",
3466 			    EFX_QWORD_VAL(*event));
3467 
3468 	rx_queue = efx_channel_get_rx_queue(channel);
3469 
3470 	if (unlikely(rx_queue_label != efx_rx_queue_index(rx_queue)))
3471 		efx_ef10_handle_rx_wrong_queue(rx_queue, rx_queue_label);
3472 
3473 	n_descs = ((next_ptr_lbits - rx_queue->removed_count) &
3474 		   ((1 << ESF_DZ_RX_DSC_PTR_LBITS_WIDTH) - 1));
3475 
3476 	if (n_descs != rx_queue->scatter_n + 1) {
3477 		struct efx_ef10_nic_data *nic_data = efx->nic_data;
3478 
3479 		/* detect rx abort */
3480 		if (unlikely(n_descs == rx_queue->scatter_n)) {
3481 			if (rx_queue->scatter_n == 0 || rx_bytes != 0)
3482 				netdev_WARN(efx->net_dev,
3483 					    "invalid RX abort: scatter_n=%u event="
3484 					    EFX_QWORD_FMT "\n",
3485 					    rx_queue->scatter_n,
3486 					    EFX_QWORD_VAL(*event));
3487 			efx_ef10_handle_rx_abort(rx_queue);
3488 			return 0;
3489 		}
3490 
3491 		/* Check that RX completion merging is valid, i.e.
3492 		 * the current firmware supports it and this is a
3493 		 * non-scattered packet.
3494 		 */
3495 		if (!(nic_data->datapath_caps &
3496 		      (1 << MC_CMD_GET_CAPABILITIES_OUT_RX_BATCHING_LBN)) ||
3497 		    rx_queue->scatter_n != 0 || rx_cont) {
3498 			efx_ef10_handle_rx_bad_lbits(
3499 				rx_queue, next_ptr_lbits,
3500 				(rx_queue->removed_count +
3501 				 rx_queue->scatter_n + 1) &
3502 				((1 << ESF_DZ_RX_DSC_PTR_LBITS_WIDTH) - 1));
3503 			return 0;
3504 		}
3505 
3506 		/* Merged completion for multiple non-scattered packets */
3507 		rx_queue->scatter_n = 1;
3508 		rx_queue->scatter_len = 0;
3509 		n_packets = n_descs;
3510 		++channel->n_rx_merge_events;
3511 		channel->n_rx_merge_packets += n_packets;
3512 		flags |= EFX_RX_PKT_PREFIX_LEN;
3513 	} else {
3514 		++rx_queue->scatter_n;
3515 		rx_queue->scatter_len += rx_bytes;
3516 		if (rx_cont)
3517 			return 0;
3518 		n_packets = 1;
3519 	}
3520 
3521 	EFX_POPULATE_QWORD_5(errors, ESF_DZ_RX_ECRC_ERR, 1,
3522 				     ESF_DZ_RX_IPCKSUM_ERR, 1,
3523 				     ESF_DZ_RX_TCPUDP_CKSUM_ERR, 1,
3524 				     ESF_EZ_RX_IP_INNER_CHKSUM_ERR, 1,
3525 				     ESF_EZ_RX_TCP_UDP_INNER_CHKSUM_ERR, 1);
3526 	EFX_AND_QWORD(errors, *event, errors);
3527 	if (unlikely(!EFX_QWORD_IS_ZERO(errors))) {
3528 		flags |= efx_ef10_handle_rx_event_errors(channel, n_packets,
3529 							 rx_encap_hdr,
3530 							 rx_l3_class, rx_l4_class,
3531 							 event);
3532 	} else {
3533 		bool tcpudp = rx_l4_class == ESE_FZ_L4_CLASS_TCP ||
3534 			      rx_l4_class == ESE_FZ_L4_CLASS_UDP;
3535 
3536 		switch (rx_encap_hdr) {
3537 		case ESE_EZ_ENCAP_HDR_VXLAN: /* VxLAN or GENEVE */
3538 			flags |= EFX_RX_PKT_CSUMMED; /* outer UDP csum */
3539 			if (tcpudp)
3540 				flags |= EFX_RX_PKT_CSUM_LEVEL; /* inner L4 */
3541 			break;
3542 		case ESE_EZ_ENCAP_HDR_GRE:
3543 		case ESE_EZ_ENCAP_HDR_NONE:
3544 			if (tcpudp)
3545 				flags |= EFX_RX_PKT_CSUMMED;
3546 			break;
3547 		default:
3548 			netdev_WARN(efx->net_dev,
3549 				    "unknown encapsulation type: event="
3550 				    EFX_QWORD_FMT "\n",
3551 				    EFX_QWORD_VAL(*event));
3552 		}
3553 	}
3554 
3555 	if (rx_l4_class == ESE_FZ_L4_CLASS_TCP)
3556 		flags |= EFX_RX_PKT_TCP;
3557 
3558 	channel->irq_mod_score += 2 * n_packets;
3559 
3560 	/* Handle received packet(s) */
3561 	for (i = 0; i < n_packets; i++) {
3562 		efx_rx_packet(rx_queue,
3563 			      rx_queue->removed_count & rx_queue->ptr_mask,
3564 			      rx_queue->scatter_n, rx_queue->scatter_len,
3565 			      flags);
3566 		rx_queue->removed_count += rx_queue->scatter_n;
3567 	}
3568 
3569 	rx_queue->scatter_n = 0;
3570 	rx_queue->scatter_len = 0;
3571 
3572 	return n_packets;
3573 }
3574 
3575 static int
3576 efx_ef10_handle_tx_event(struct efx_channel *channel, efx_qword_t *event)
3577 {
3578 	struct efx_nic *efx = channel->efx;
3579 	struct efx_tx_queue *tx_queue;
3580 	unsigned int tx_ev_desc_ptr;
3581 	unsigned int tx_ev_q_label;
3582 	int tx_descs = 0;
3583 
3584 	if (unlikely(READ_ONCE(efx->reset_pending)))
3585 		return 0;
3586 
3587 	if (unlikely(EFX_QWORD_FIELD(*event, ESF_DZ_TX_DROP_EVENT)))
3588 		return 0;
3589 
3590 	/* Transmit completion */
3591 	tx_ev_desc_ptr = EFX_QWORD_FIELD(*event, ESF_DZ_TX_DESCR_INDX);
3592 	tx_ev_q_label = EFX_QWORD_FIELD(*event, ESF_DZ_TX_QLABEL);
3593 	tx_queue = efx_channel_get_tx_queue(channel,
3594 					    tx_ev_q_label % EFX_TXQ_TYPES);
3595 	tx_descs = ((tx_ev_desc_ptr + 1 - tx_queue->read_count) &
3596 		    tx_queue->ptr_mask);
3597 	efx_xmit_done(tx_queue, tx_ev_desc_ptr & tx_queue->ptr_mask);
3598 
3599 	return tx_descs;
3600 }
3601 
3602 static void
3603 efx_ef10_handle_driver_event(struct efx_channel *channel, efx_qword_t *event)
3604 {
3605 	struct efx_nic *efx = channel->efx;
3606 	int subcode;
3607 
3608 	subcode = EFX_QWORD_FIELD(*event, ESF_DZ_DRV_SUB_CODE);
3609 
3610 	switch (subcode) {
3611 	case ESE_DZ_DRV_TIMER_EV:
3612 	case ESE_DZ_DRV_WAKE_UP_EV:
3613 		break;
3614 	case ESE_DZ_DRV_START_UP_EV:
3615 		/* event queue init complete. ok. */
3616 		break;
3617 	default:
3618 		netif_err(efx, hw, efx->net_dev,
3619 			  "channel %d unknown driver event type %d"
3620 			  " (data " EFX_QWORD_FMT ")\n",
3621 			  channel->channel, subcode,
3622 			  EFX_QWORD_VAL(*event));
3623 
3624 	}
3625 }
3626 
3627 static void efx_ef10_handle_driver_generated_event(struct efx_channel *channel,
3628 						   efx_qword_t *event)
3629 {
3630 	struct efx_nic *efx = channel->efx;
3631 	u32 subcode;
3632 
3633 	subcode = EFX_QWORD_FIELD(*event, EFX_DWORD_0);
3634 
3635 	switch (subcode) {
3636 	case EFX_EF10_TEST:
3637 		channel->event_test_cpu = raw_smp_processor_id();
3638 		break;
3639 	case EFX_EF10_REFILL:
3640 		/* The queue must be empty, so we won't receive any rx
3641 		 * events, so efx_process_channel() won't refill the
3642 		 * queue. Refill it here
3643 		 */
3644 		efx_fast_push_rx_descriptors(&channel->rx_queue, true);
3645 		break;
3646 	default:
3647 		netif_err(efx, hw, efx->net_dev,
3648 			  "channel %d unknown driver event type %u"
3649 			  " (data " EFX_QWORD_FMT ")\n",
3650 			  channel->channel, (unsigned) subcode,
3651 			  EFX_QWORD_VAL(*event));
3652 	}
3653 }
3654 
3655 static int efx_ef10_ev_process(struct efx_channel *channel, int quota)
3656 {
3657 	struct efx_nic *efx = channel->efx;
3658 	efx_qword_t event, *p_event;
3659 	unsigned int read_ptr;
3660 	int ev_code;
3661 	int tx_descs = 0;
3662 	int spent = 0;
3663 
3664 	if (quota <= 0)
3665 		return spent;
3666 
3667 	read_ptr = channel->eventq_read_ptr;
3668 
3669 	for (;;) {
3670 		p_event = efx_event(channel, read_ptr);
3671 		event = *p_event;
3672 
3673 		if (!efx_event_present(&event))
3674 			break;
3675 
3676 		EFX_SET_QWORD(*p_event);
3677 
3678 		++read_ptr;
3679 
3680 		ev_code = EFX_QWORD_FIELD(event, ESF_DZ_EV_CODE);
3681 
3682 		netif_vdbg(efx, drv, efx->net_dev,
3683 			   "processing event on %d " EFX_QWORD_FMT "\n",
3684 			   channel->channel, EFX_QWORD_VAL(event));
3685 
3686 		switch (ev_code) {
3687 		case ESE_DZ_EV_CODE_MCDI_EV:
3688 			efx_mcdi_process_event(channel, &event);
3689 			break;
3690 		case ESE_DZ_EV_CODE_RX_EV:
3691 			spent += efx_ef10_handle_rx_event(channel, &event);
3692 			if (spent >= quota) {
3693 				/* XXX can we split a merged event to
3694 				 * avoid going over-quota?
3695 				 */
3696 				spent = quota;
3697 				goto out;
3698 			}
3699 			break;
3700 		case ESE_DZ_EV_CODE_TX_EV:
3701 			tx_descs += efx_ef10_handle_tx_event(channel, &event);
3702 			if (tx_descs > efx->txq_entries) {
3703 				spent = quota;
3704 				goto out;
3705 			} else if (++spent == quota) {
3706 				goto out;
3707 			}
3708 			break;
3709 		case ESE_DZ_EV_CODE_DRIVER_EV:
3710 			efx_ef10_handle_driver_event(channel, &event);
3711 			if (++spent == quota)
3712 				goto out;
3713 			break;
3714 		case EFX_EF10_DRVGEN_EV:
3715 			efx_ef10_handle_driver_generated_event(channel, &event);
3716 			break;
3717 		default:
3718 			netif_err(efx, hw, efx->net_dev,
3719 				  "channel %d unknown event type %d"
3720 				  " (data " EFX_QWORD_FMT ")\n",
3721 				  channel->channel, ev_code,
3722 				  EFX_QWORD_VAL(event));
3723 		}
3724 	}
3725 
3726 out:
3727 	channel->eventq_read_ptr = read_ptr;
3728 	return spent;
3729 }
3730 
3731 static void efx_ef10_ev_read_ack(struct efx_channel *channel)
3732 {
3733 	struct efx_nic *efx = channel->efx;
3734 	efx_dword_t rptr;
3735 
3736 	if (EFX_EF10_WORKAROUND_35388(efx)) {
3737 		BUILD_BUG_ON(EFX_MIN_EVQ_SIZE <
3738 			     (1 << ERF_DD_EVQ_IND_RPTR_WIDTH));
3739 		BUILD_BUG_ON(EFX_MAX_EVQ_SIZE >
3740 			     (1 << 2 * ERF_DD_EVQ_IND_RPTR_WIDTH));
3741 
3742 		EFX_POPULATE_DWORD_2(rptr, ERF_DD_EVQ_IND_RPTR_FLAGS,
3743 				     EFE_DD_EVQ_IND_RPTR_FLAGS_HIGH,
3744 				     ERF_DD_EVQ_IND_RPTR,
3745 				     (channel->eventq_read_ptr &
3746 				      channel->eventq_mask) >>
3747 				     ERF_DD_EVQ_IND_RPTR_WIDTH);
3748 		efx_writed_page(efx, &rptr, ER_DD_EVQ_INDIRECT,
3749 				channel->channel);
3750 		EFX_POPULATE_DWORD_2(rptr, ERF_DD_EVQ_IND_RPTR_FLAGS,
3751 				     EFE_DD_EVQ_IND_RPTR_FLAGS_LOW,
3752 				     ERF_DD_EVQ_IND_RPTR,
3753 				     channel->eventq_read_ptr &
3754 				     ((1 << ERF_DD_EVQ_IND_RPTR_WIDTH) - 1));
3755 		efx_writed_page(efx, &rptr, ER_DD_EVQ_INDIRECT,
3756 				channel->channel);
3757 	} else {
3758 		EFX_POPULATE_DWORD_1(rptr, ERF_DZ_EVQ_RPTR,
3759 				     channel->eventq_read_ptr &
3760 				     channel->eventq_mask);
3761 		efx_writed_page(efx, &rptr, ER_DZ_EVQ_RPTR, channel->channel);
3762 	}
3763 }
3764 
3765 static void efx_ef10_ev_test_generate(struct efx_channel *channel)
3766 {
3767 	MCDI_DECLARE_BUF(inbuf, MC_CMD_DRIVER_EVENT_IN_LEN);
3768 	struct efx_nic *efx = channel->efx;
3769 	efx_qword_t event;
3770 	int rc;
3771 
3772 	EFX_POPULATE_QWORD_2(event,
3773 			     ESF_DZ_EV_CODE, EFX_EF10_DRVGEN_EV,
3774 			     ESF_DZ_EV_DATA, EFX_EF10_TEST);
3775 
3776 	MCDI_SET_DWORD(inbuf, DRIVER_EVENT_IN_EVQ, channel->channel);
3777 
3778 	/* MCDI_SET_QWORD is not appropriate here since EFX_POPULATE_* has
3779 	 * already swapped the data to little-endian order.
3780 	 */
3781 	memcpy(MCDI_PTR(inbuf, DRIVER_EVENT_IN_DATA), &event.u64[0],
3782 	       sizeof(efx_qword_t));
3783 
3784 	rc = efx_mcdi_rpc(efx, MC_CMD_DRIVER_EVENT, inbuf, sizeof(inbuf),
3785 			  NULL, 0, NULL);
3786 	if (rc != 0)
3787 		goto fail;
3788 
3789 	return;
3790 
3791 fail:
3792 	WARN_ON(true);
3793 	netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
3794 }
3795 
3796 void efx_ef10_handle_drain_event(struct efx_nic *efx)
3797 {
3798 	if (atomic_dec_and_test(&efx->active_queues))
3799 		wake_up(&efx->flush_wq);
3800 
3801 	WARN_ON(atomic_read(&efx->active_queues) < 0);
3802 }
3803 
3804 static int efx_ef10_fini_dmaq(struct efx_nic *efx)
3805 {
3806 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
3807 	struct efx_channel *channel;
3808 	struct efx_tx_queue *tx_queue;
3809 	struct efx_rx_queue *rx_queue;
3810 	int pending;
3811 
3812 	/* If the MC has just rebooted, the TX/RX queues will have already been
3813 	 * torn down, but efx->active_queues needs to be set to zero.
3814 	 */
3815 	if (nic_data->must_realloc_vis) {
3816 		atomic_set(&efx->active_queues, 0);
3817 		return 0;
3818 	}
3819 
3820 	/* Do not attempt to write to the NIC during EEH recovery */
3821 	if (efx->state != STATE_RECOVERY) {
3822 		efx_for_each_channel(channel, efx) {
3823 			efx_for_each_channel_rx_queue(rx_queue, channel)
3824 				efx_ef10_rx_fini(rx_queue);
3825 			efx_for_each_channel_tx_queue(tx_queue, channel)
3826 				efx_ef10_tx_fini(tx_queue);
3827 		}
3828 
3829 		wait_event_timeout(efx->flush_wq,
3830 				   atomic_read(&efx->active_queues) == 0,
3831 				   msecs_to_jiffies(EFX_MAX_FLUSH_TIME));
3832 		pending = atomic_read(&efx->active_queues);
3833 		if (pending) {
3834 			netif_err(efx, hw, efx->net_dev, "failed to flush %d queues\n",
3835 				  pending);
3836 			return -ETIMEDOUT;
3837 		}
3838 	}
3839 
3840 	return 0;
3841 }
3842 
3843 static void efx_ef10_prepare_flr(struct efx_nic *efx)
3844 {
3845 	atomic_set(&efx->active_queues, 0);
3846 }
3847 
3848 static bool efx_ef10_filter_equal(const struct efx_filter_spec *left,
3849 				  const struct efx_filter_spec *right)
3850 {
3851 	if ((left->match_flags ^ right->match_flags) |
3852 	    ((left->flags ^ right->flags) &
3853 	     (EFX_FILTER_FLAG_RX | EFX_FILTER_FLAG_TX)))
3854 		return false;
3855 
3856 	return memcmp(&left->outer_vid, &right->outer_vid,
3857 		      sizeof(struct efx_filter_spec) -
3858 		      offsetof(struct efx_filter_spec, outer_vid)) == 0;
3859 }
3860 
3861 static unsigned int efx_ef10_filter_hash(const struct efx_filter_spec *spec)
3862 {
3863 	BUILD_BUG_ON(offsetof(struct efx_filter_spec, outer_vid) & 3);
3864 	return jhash2((const u32 *)&spec->outer_vid,
3865 		      (sizeof(struct efx_filter_spec) -
3866 		       offsetof(struct efx_filter_spec, outer_vid)) / 4,
3867 		      0);
3868 	/* XXX should we randomise the initval? */
3869 }
3870 
3871 /* Decide whether a filter should be exclusive or else should allow
3872  * delivery to additional recipients.  Currently we decide that
3873  * filters for specific local unicast MAC and IP addresses are
3874  * exclusive.
3875  */
3876 static bool efx_ef10_filter_is_exclusive(const struct efx_filter_spec *spec)
3877 {
3878 	if (spec->match_flags & EFX_FILTER_MATCH_LOC_MAC &&
3879 	    !is_multicast_ether_addr(spec->loc_mac))
3880 		return true;
3881 
3882 	if ((spec->match_flags &
3883 	     (EFX_FILTER_MATCH_ETHER_TYPE | EFX_FILTER_MATCH_LOC_HOST)) ==
3884 	    (EFX_FILTER_MATCH_ETHER_TYPE | EFX_FILTER_MATCH_LOC_HOST)) {
3885 		if (spec->ether_type == htons(ETH_P_IP) &&
3886 		    !ipv4_is_multicast(spec->loc_host[0]))
3887 			return true;
3888 		if (spec->ether_type == htons(ETH_P_IPV6) &&
3889 		    ((const u8 *)spec->loc_host)[0] != 0xff)
3890 			return true;
3891 	}
3892 
3893 	return false;
3894 }
3895 
3896 static struct efx_filter_spec *
3897 efx_ef10_filter_entry_spec(const struct efx_ef10_filter_table *table,
3898 			   unsigned int filter_idx)
3899 {
3900 	return (struct efx_filter_spec *)(table->entry[filter_idx].spec &
3901 					  ~EFX_EF10_FILTER_FLAGS);
3902 }
3903 
3904 static unsigned int
3905 efx_ef10_filter_entry_flags(const struct efx_ef10_filter_table *table,
3906 			   unsigned int filter_idx)
3907 {
3908 	return table->entry[filter_idx].spec & EFX_EF10_FILTER_FLAGS;
3909 }
3910 
3911 static void
3912 efx_ef10_filter_set_entry(struct efx_ef10_filter_table *table,
3913 			  unsigned int filter_idx,
3914 			  const struct efx_filter_spec *spec,
3915 			  unsigned int flags)
3916 {
3917 	table->entry[filter_idx].spec =	(unsigned long)spec | flags;
3918 }
3919 
3920 static void
3921 efx_ef10_filter_push_prep_set_match_fields(struct efx_nic *efx,
3922 					   const struct efx_filter_spec *spec,
3923 					   efx_dword_t *inbuf)
3924 {
3925 	enum efx_encap_type encap_type = efx_filter_get_encap_type(spec);
3926 	u32 match_fields = 0, uc_match, mc_match;
3927 
3928 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
3929 		       efx_ef10_filter_is_exclusive(spec) ?
3930 		       MC_CMD_FILTER_OP_IN_OP_INSERT :
3931 		       MC_CMD_FILTER_OP_IN_OP_SUBSCRIBE);
3932 
3933 	/* Convert match flags and values.  Unlike almost
3934 	 * everything else in MCDI, these fields are in
3935 	 * network byte order.
3936 	 */
3937 #define COPY_VALUE(value, mcdi_field)					     \
3938 	do {							     \
3939 		match_fields |=					     \
3940 			1 << MC_CMD_FILTER_OP_IN_MATCH_ ##	     \
3941 			mcdi_field ## _LBN;			     \
3942 		BUILD_BUG_ON(					     \
3943 			MC_CMD_FILTER_OP_IN_ ## mcdi_field ## _LEN < \
3944 			sizeof(value));				     \
3945 		memcpy(MCDI_PTR(inbuf, FILTER_OP_IN_ ##	mcdi_field), \
3946 		       &value, sizeof(value));			     \
3947 	} while (0)
3948 #define COPY_FIELD(gen_flag, gen_field, mcdi_field)			     \
3949 	if (spec->match_flags & EFX_FILTER_MATCH_ ## gen_flag) {     \
3950 		COPY_VALUE(spec->gen_field, mcdi_field);	     \
3951 	}
3952 	/* Handle encap filters first.  They will always be mismatch
3953 	 * (unknown UC or MC) filters
3954 	 */
3955 	if (encap_type) {
3956 		/* ether_type and outer_ip_proto need to be variables
3957 		 * because COPY_VALUE wants to memcpy them
3958 		 */
3959 		__be16 ether_type =
3960 			htons(encap_type & EFX_ENCAP_FLAG_IPV6 ?
3961 			      ETH_P_IPV6 : ETH_P_IP);
3962 		u8 vni_type = MC_CMD_FILTER_OP_EXT_IN_VNI_TYPE_GENEVE;
3963 		u8 outer_ip_proto;
3964 
3965 		switch (encap_type & EFX_ENCAP_TYPES_MASK) {
3966 		case EFX_ENCAP_TYPE_VXLAN:
3967 			vni_type = MC_CMD_FILTER_OP_EXT_IN_VNI_TYPE_VXLAN;
3968 			/* fallthrough */
3969 		case EFX_ENCAP_TYPE_GENEVE:
3970 			COPY_VALUE(ether_type, ETHER_TYPE);
3971 			outer_ip_proto = IPPROTO_UDP;
3972 			COPY_VALUE(outer_ip_proto, IP_PROTO);
3973 			/* We always need to set the type field, even
3974 			 * though we're not matching on the TNI.
3975 			 */
3976 			MCDI_POPULATE_DWORD_1(inbuf,
3977 				FILTER_OP_EXT_IN_VNI_OR_VSID,
3978 				FILTER_OP_EXT_IN_VNI_TYPE,
3979 				vni_type);
3980 			break;
3981 		case EFX_ENCAP_TYPE_NVGRE:
3982 			COPY_VALUE(ether_type, ETHER_TYPE);
3983 			outer_ip_proto = IPPROTO_GRE;
3984 			COPY_VALUE(outer_ip_proto, IP_PROTO);
3985 			break;
3986 		default:
3987 			WARN_ON(1);
3988 		}
3989 
3990 		uc_match = MC_CMD_FILTER_OP_EXT_IN_MATCH_IFRM_UNKNOWN_UCAST_DST_LBN;
3991 		mc_match = MC_CMD_FILTER_OP_EXT_IN_MATCH_IFRM_UNKNOWN_MCAST_DST_LBN;
3992 	} else {
3993 		uc_match = MC_CMD_FILTER_OP_EXT_IN_MATCH_UNKNOWN_UCAST_DST_LBN;
3994 		mc_match = MC_CMD_FILTER_OP_EXT_IN_MATCH_UNKNOWN_MCAST_DST_LBN;
3995 	}
3996 
3997 	if (spec->match_flags & EFX_FILTER_MATCH_LOC_MAC_IG)
3998 		match_fields |=
3999 			is_multicast_ether_addr(spec->loc_mac) ?
4000 			1 << mc_match :
4001 			1 << uc_match;
4002 	COPY_FIELD(REM_HOST, rem_host, SRC_IP);
4003 	COPY_FIELD(LOC_HOST, loc_host, DST_IP);
4004 	COPY_FIELD(REM_MAC, rem_mac, SRC_MAC);
4005 	COPY_FIELD(REM_PORT, rem_port, SRC_PORT);
4006 	COPY_FIELD(LOC_MAC, loc_mac, DST_MAC);
4007 	COPY_FIELD(LOC_PORT, loc_port, DST_PORT);
4008 	COPY_FIELD(ETHER_TYPE, ether_type, ETHER_TYPE);
4009 	COPY_FIELD(INNER_VID, inner_vid, INNER_VLAN);
4010 	COPY_FIELD(OUTER_VID, outer_vid, OUTER_VLAN);
4011 	COPY_FIELD(IP_PROTO, ip_proto, IP_PROTO);
4012 #undef COPY_FIELD
4013 #undef COPY_VALUE
4014 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_MATCH_FIELDS,
4015 		       match_fields);
4016 }
4017 
4018 static void efx_ef10_filter_push_prep(struct efx_nic *efx,
4019 				      const struct efx_filter_spec *spec,
4020 				      efx_dword_t *inbuf, u64 handle,
4021 				      bool replacing)
4022 {
4023 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
4024 	u32 flags = spec->flags;
4025 
4026 	memset(inbuf, 0, MC_CMD_FILTER_OP_EXT_IN_LEN);
4027 
4028 	/* Remove RSS flag if we don't have an RSS context. */
4029 	if (flags & EFX_FILTER_FLAG_RX_RSS &&
4030 	    spec->rss_context == EFX_FILTER_RSS_CONTEXT_DEFAULT &&
4031 	    nic_data->rx_rss_context == EFX_EF10_RSS_CONTEXT_INVALID)
4032 		flags &= ~EFX_FILTER_FLAG_RX_RSS;
4033 
4034 	if (replacing) {
4035 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
4036 			       MC_CMD_FILTER_OP_IN_OP_REPLACE);
4037 		MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE, handle);
4038 	} else {
4039 		efx_ef10_filter_push_prep_set_match_fields(efx, spec, inbuf);
4040 	}
4041 
4042 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_PORT_ID, nic_data->vport_id);
4043 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_DEST,
4044 		       spec->dmaq_id == EFX_FILTER_RX_DMAQ_ID_DROP ?
4045 		       MC_CMD_FILTER_OP_IN_RX_DEST_DROP :
4046 		       MC_CMD_FILTER_OP_IN_RX_DEST_HOST);
4047 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_TX_DOMAIN, 0);
4048 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_TX_DEST,
4049 		       MC_CMD_FILTER_OP_IN_TX_DEST_DEFAULT);
4050 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_QUEUE,
4051 		       spec->dmaq_id == EFX_FILTER_RX_DMAQ_ID_DROP ?
4052 		       0 : spec->dmaq_id);
4053 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_MODE,
4054 		       (flags & EFX_FILTER_FLAG_RX_RSS) ?
4055 		       MC_CMD_FILTER_OP_IN_RX_MODE_RSS :
4056 		       MC_CMD_FILTER_OP_IN_RX_MODE_SIMPLE);
4057 	if (flags & EFX_FILTER_FLAG_RX_RSS)
4058 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_CONTEXT,
4059 			       spec->rss_context !=
4060 			       EFX_FILTER_RSS_CONTEXT_DEFAULT ?
4061 			       spec->rss_context : nic_data->rx_rss_context);
4062 }
4063 
4064 static int efx_ef10_filter_push(struct efx_nic *efx,
4065 				const struct efx_filter_spec *spec,
4066 				u64 *handle, bool replacing)
4067 {
4068 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_EXT_IN_LEN);
4069 	MCDI_DECLARE_BUF(outbuf, MC_CMD_FILTER_OP_EXT_OUT_LEN);
4070 	int rc;
4071 
4072 	efx_ef10_filter_push_prep(efx, spec, inbuf, *handle, replacing);
4073 	rc = efx_mcdi_rpc(efx, MC_CMD_FILTER_OP, inbuf, sizeof(inbuf),
4074 			  outbuf, sizeof(outbuf), NULL);
4075 	if (rc == 0)
4076 		*handle = MCDI_QWORD(outbuf, FILTER_OP_OUT_HANDLE);
4077 	if (rc == -ENOSPC)
4078 		rc = -EBUSY; /* to match efx_farch_filter_insert() */
4079 	return rc;
4080 }
4081 
4082 static u32 efx_ef10_filter_mcdi_flags_from_spec(const struct efx_filter_spec *spec)
4083 {
4084 	enum efx_encap_type encap_type = efx_filter_get_encap_type(spec);
4085 	unsigned int match_flags = spec->match_flags;
4086 	unsigned int uc_match, mc_match;
4087 	u32 mcdi_flags = 0;
4088 
4089 #define MAP_FILTER_TO_MCDI_FLAG(gen_flag, mcdi_field, encap) {		\
4090 		unsigned int  old_match_flags = match_flags;		\
4091 		match_flags &= ~EFX_FILTER_MATCH_ ## gen_flag;		\
4092 		if (match_flags != old_match_flags)			\
4093 			mcdi_flags |=					\
4094 				(1 << ((encap) ?			\
4095 				       MC_CMD_FILTER_OP_EXT_IN_MATCH_IFRM_ ## \
4096 				       mcdi_field ## _LBN :		\
4097 				       MC_CMD_FILTER_OP_EXT_IN_MATCH_ ##\
4098 				       mcdi_field ## _LBN));		\
4099 	}
4100 	/* inner or outer based on encap type */
4101 	MAP_FILTER_TO_MCDI_FLAG(REM_HOST, SRC_IP, encap_type);
4102 	MAP_FILTER_TO_MCDI_FLAG(LOC_HOST, DST_IP, encap_type);
4103 	MAP_FILTER_TO_MCDI_FLAG(REM_MAC, SRC_MAC, encap_type);
4104 	MAP_FILTER_TO_MCDI_FLAG(REM_PORT, SRC_PORT, encap_type);
4105 	MAP_FILTER_TO_MCDI_FLAG(LOC_MAC, DST_MAC, encap_type);
4106 	MAP_FILTER_TO_MCDI_FLAG(LOC_PORT, DST_PORT, encap_type);
4107 	MAP_FILTER_TO_MCDI_FLAG(ETHER_TYPE, ETHER_TYPE, encap_type);
4108 	MAP_FILTER_TO_MCDI_FLAG(IP_PROTO, IP_PROTO, encap_type);
4109 	/* always outer */
4110 	MAP_FILTER_TO_MCDI_FLAG(INNER_VID, INNER_VLAN, false);
4111 	MAP_FILTER_TO_MCDI_FLAG(OUTER_VID, OUTER_VLAN, false);
4112 #undef MAP_FILTER_TO_MCDI_FLAG
4113 
4114 	/* special handling for encap type, and mismatch */
4115 	if (encap_type) {
4116 		match_flags &= ~EFX_FILTER_MATCH_ENCAP_TYPE;
4117 		mcdi_flags |=
4118 			(1 << MC_CMD_FILTER_OP_EXT_IN_MATCH_ETHER_TYPE_LBN);
4119 		mcdi_flags |= (1 << MC_CMD_FILTER_OP_EXT_IN_MATCH_IP_PROTO_LBN);
4120 
4121 		uc_match = MC_CMD_FILTER_OP_EXT_IN_MATCH_IFRM_UNKNOWN_UCAST_DST_LBN;
4122 		mc_match = MC_CMD_FILTER_OP_EXT_IN_MATCH_IFRM_UNKNOWN_MCAST_DST_LBN;
4123 	} else {
4124 		uc_match = MC_CMD_FILTER_OP_EXT_IN_MATCH_UNKNOWN_UCAST_DST_LBN;
4125 		mc_match = MC_CMD_FILTER_OP_EXT_IN_MATCH_UNKNOWN_MCAST_DST_LBN;
4126 	}
4127 
4128 	if (match_flags & EFX_FILTER_MATCH_LOC_MAC_IG) {
4129 		match_flags &= ~EFX_FILTER_MATCH_LOC_MAC_IG;
4130 		mcdi_flags |=
4131 			is_multicast_ether_addr(spec->loc_mac) ?
4132 			1 << mc_match :
4133 			1 << uc_match;
4134 	}
4135 
4136 	/* Did we map them all? */
4137 	WARN_ON_ONCE(match_flags);
4138 
4139 	return mcdi_flags;
4140 }
4141 
4142 static int efx_ef10_filter_pri(struct efx_ef10_filter_table *table,
4143 			       const struct efx_filter_spec *spec)
4144 {
4145 	u32 mcdi_flags = efx_ef10_filter_mcdi_flags_from_spec(spec);
4146 	unsigned int match_pri;
4147 
4148 	for (match_pri = 0;
4149 	     match_pri < table->rx_match_count;
4150 	     match_pri++)
4151 		if (table->rx_match_mcdi_flags[match_pri] == mcdi_flags)
4152 			return match_pri;
4153 
4154 	return -EPROTONOSUPPORT;
4155 }
4156 
4157 static s32 efx_ef10_filter_insert(struct efx_nic *efx,
4158 				  struct efx_filter_spec *spec,
4159 				  bool replace_equal)
4160 {
4161 	struct efx_ef10_filter_table *table = efx->filter_state;
4162 	DECLARE_BITMAP(mc_rem_map, EFX_EF10_FILTER_SEARCH_LIMIT);
4163 	struct efx_filter_spec *saved_spec;
4164 	unsigned int match_pri, hash;
4165 	unsigned int priv_flags;
4166 	bool replacing = false;
4167 	int ins_index = -1;
4168 	DEFINE_WAIT(wait);
4169 	bool is_mc_recip;
4170 	s32 rc;
4171 
4172 	/* For now, only support RX filters */
4173 	if ((spec->flags & (EFX_FILTER_FLAG_RX | EFX_FILTER_FLAG_TX)) !=
4174 	    EFX_FILTER_FLAG_RX)
4175 		return -EINVAL;
4176 
4177 	rc = efx_ef10_filter_pri(table, spec);
4178 	if (rc < 0)
4179 		return rc;
4180 	match_pri = rc;
4181 
4182 	hash = efx_ef10_filter_hash(spec);
4183 	is_mc_recip = efx_filter_is_mc_recipient(spec);
4184 	if (is_mc_recip)
4185 		bitmap_zero(mc_rem_map, EFX_EF10_FILTER_SEARCH_LIMIT);
4186 
4187 	/* Find any existing filters with the same match tuple or
4188 	 * else a free slot to insert at.  If any of them are busy,
4189 	 * we have to wait and retry.
4190 	 */
4191 	for (;;) {
4192 		unsigned int depth = 1;
4193 		unsigned int i;
4194 
4195 		spin_lock_bh(&efx->filter_lock);
4196 
4197 		for (;;) {
4198 			i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
4199 			saved_spec = efx_ef10_filter_entry_spec(table, i);
4200 
4201 			if (!saved_spec) {
4202 				if (ins_index < 0)
4203 					ins_index = i;
4204 			} else if (efx_ef10_filter_equal(spec, saved_spec)) {
4205 				if (table->entry[i].spec &
4206 				    EFX_EF10_FILTER_FLAG_BUSY)
4207 					break;
4208 				if (spec->priority < saved_spec->priority &&
4209 				    spec->priority != EFX_FILTER_PRI_AUTO) {
4210 					rc = -EPERM;
4211 					goto out_unlock;
4212 				}
4213 				if (!is_mc_recip) {
4214 					/* This is the only one */
4215 					if (spec->priority ==
4216 					    saved_spec->priority &&
4217 					    !replace_equal) {
4218 						rc = -EEXIST;
4219 						goto out_unlock;
4220 					}
4221 					ins_index = i;
4222 					goto found;
4223 				} else if (spec->priority >
4224 					   saved_spec->priority ||
4225 					   (spec->priority ==
4226 					    saved_spec->priority &&
4227 					    replace_equal)) {
4228 					if (ins_index < 0)
4229 						ins_index = i;
4230 					else
4231 						__set_bit(depth, mc_rem_map);
4232 				}
4233 			}
4234 
4235 			/* Once we reach the maximum search depth, use
4236 			 * the first suitable slot or return -EBUSY if
4237 			 * there was none
4238 			 */
4239 			if (depth == EFX_EF10_FILTER_SEARCH_LIMIT) {
4240 				if (ins_index < 0) {
4241 					rc = -EBUSY;
4242 					goto out_unlock;
4243 				}
4244 				goto found;
4245 			}
4246 
4247 			++depth;
4248 		}
4249 
4250 		prepare_to_wait(&table->waitq, &wait, TASK_UNINTERRUPTIBLE);
4251 		spin_unlock_bh(&efx->filter_lock);
4252 		schedule();
4253 	}
4254 
4255 found:
4256 	/* Create a software table entry if necessary, and mark it
4257 	 * busy.  We might yet fail to insert, but any attempt to
4258 	 * insert a conflicting filter while we're waiting for the
4259 	 * firmware must find the busy entry.
4260 	 */
4261 	saved_spec = efx_ef10_filter_entry_spec(table, ins_index);
4262 	if (saved_spec) {
4263 		if (spec->priority == EFX_FILTER_PRI_AUTO &&
4264 		    saved_spec->priority >= EFX_FILTER_PRI_AUTO) {
4265 			/* Just make sure it won't be removed */
4266 			if (saved_spec->priority > EFX_FILTER_PRI_AUTO)
4267 				saved_spec->flags |= EFX_FILTER_FLAG_RX_OVER_AUTO;
4268 			table->entry[ins_index].spec &=
4269 				~EFX_EF10_FILTER_FLAG_AUTO_OLD;
4270 			rc = ins_index;
4271 			goto out_unlock;
4272 		}
4273 		replacing = true;
4274 		priv_flags = efx_ef10_filter_entry_flags(table, ins_index);
4275 	} else {
4276 		saved_spec = kmalloc(sizeof(*spec), GFP_ATOMIC);
4277 		if (!saved_spec) {
4278 			rc = -ENOMEM;
4279 			goto out_unlock;
4280 		}
4281 		*saved_spec = *spec;
4282 		priv_flags = 0;
4283 	}
4284 	efx_ef10_filter_set_entry(table, ins_index, saved_spec,
4285 				  priv_flags | EFX_EF10_FILTER_FLAG_BUSY);
4286 
4287 	/* Mark lower-priority multicast recipients busy prior to removal */
4288 	if (is_mc_recip) {
4289 		unsigned int depth, i;
4290 
4291 		for (depth = 0; depth < EFX_EF10_FILTER_SEARCH_LIMIT; depth++) {
4292 			i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
4293 			if (test_bit(depth, mc_rem_map))
4294 				table->entry[i].spec |=
4295 					EFX_EF10_FILTER_FLAG_BUSY;
4296 		}
4297 	}
4298 
4299 	spin_unlock_bh(&efx->filter_lock);
4300 
4301 	rc = efx_ef10_filter_push(efx, spec, &table->entry[ins_index].handle,
4302 				  replacing);
4303 
4304 	/* Finalise the software table entry */
4305 	spin_lock_bh(&efx->filter_lock);
4306 	if (rc == 0) {
4307 		if (replacing) {
4308 			/* Update the fields that may differ */
4309 			if (saved_spec->priority == EFX_FILTER_PRI_AUTO)
4310 				saved_spec->flags |=
4311 					EFX_FILTER_FLAG_RX_OVER_AUTO;
4312 			saved_spec->priority = spec->priority;
4313 			saved_spec->flags &= EFX_FILTER_FLAG_RX_OVER_AUTO;
4314 			saved_spec->flags |= spec->flags;
4315 			saved_spec->rss_context = spec->rss_context;
4316 			saved_spec->dmaq_id = spec->dmaq_id;
4317 		}
4318 	} else if (!replacing) {
4319 		kfree(saved_spec);
4320 		saved_spec = NULL;
4321 	}
4322 	efx_ef10_filter_set_entry(table, ins_index, saved_spec, priv_flags);
4323 
4324 	/* Remove and finalise entries for lower-priority multicast
4325 	 * recipients
4326 	 */
4327 	if (is_mc_recip) {
4328 		MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_EXT_IN_LEN);
4329 		unsigned int depth, i;
4330 
4331 		memset(inbuf, 0, sizeof(inbuf));
4332 
4333 		for (depth = 0; depth < EFX_EF10_FILTER_SEARCH_LIMIT; depth++) {
4334 			if (!test_bit(depth, mc_rem_map))
4335 				continue;
4336 
4337 			i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
4338 			saved_spec = efx_ef10_filter_entry_spec(table, i);
4339 			priv_flags = efx_ef10_filter_entry_flags(table, i);
4340 
4341 			if (rc == 0) {
4342 				spin_unlock_bh(&efx->filter_lock);
4343 				MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
4344 					       MC_CMD_FILTER_OP_IN_OP_UNSUBSCRIBE);
4345 				MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
4346 					       table->entry[i].handle);
4347 				rc = efx_mcdi_rpc(efx, MC_CMD_FILTER_OP,
4348 						  inbuf, sizeof(inbuf),
4349 						  NULL, 0, NULL);
4350 				spin_lock_bh(&efx->filter_lock);
4351 			}
4352 
4353 			if (rc == 0) {
4354 				kfree(saved_spec);
4355 				saved_spec = NULL;
4356 				priv_flags = 0;
4357 			} else {
4358 				priv_flags &= ~EFX_EF10_FILTER_FLAG_BUSY;
4359 			}
4360 			efx_ef10_filter_set_entry(table, i, saved_spec,
4361 						  priv_flags);
4362 		}
4363 	}
4364 
4365 	/* If successful, return the inserted filter ID */
4366 	if (rc == 0)
4367 		rc = efx_ef10_make_filter_id(match_pri, ins_index);
4368 
4369 	wake_up_all(&table->waitq);
4370 out_unlock:
4371 	spin_unlock_bh(&efx->filter_lock);
4372 	finish_wait(&table->waitq, &wait);
4373 	return rc;
4374 }
4375 
4376 static void efx_ef10_filter_update_rx_scatter(struct efx_nic *efx)
4377 {
4378 	/* no need to do anything here on EF10 */
4379 }
4380 
4381 /* Remove a filter.
4382  * If !by_index, remove by ID
4383  * If by_index, remove by index
4384  * Filter ID may come from userland and must be range-checked.
4385  */
4386 static int efx_ef10_filter_remove_internal(struct efx_nic *efx,
4387 					   unsigned int priority_mask,
4388 					   u32 filter_id, bool by_index)
4389 {
4390 	unsigned int filter_idx = efx_ef10_filter_get_unsafe_id(filter_id);
4391 	struct efx_ef10_filter_table *table = efx->filter_state;
4392 	MCDI_DECLARE_BUF(inbuf,
4393 			 MC_CMD_FILTER_OP_IN_HANDLE_OFST +
4394 			 MC_CMD_FILTER_OP_IN_HANDLE_LEN);
4395 	struct efx_filter_spec *spec;
4396 	DEFINE_WAIT(wait);
4397 	int rc;
4398 
4399 	/* Find the software table entry and mark it busy.  Don't
4400 	 * remove it yet; any attempt to update while we're waiting
4401 	 * for the firmware must find the busy entry.
4402 	 */
4403 	for (;;) {
4404 		spin_lock_bh(&efx->filter_lock);
4405 		if (!(table->entry[filter_idx].spec &
4406 		      EFX_EF10_FILTER_FLAG_BUSY))
4407 			break;
4408 		prepare_to_wait(&table->waitq, &wait, TASK_UNINTERRUPTIBLE);
4409 		spin_unlock_bh(&efx->filter_lock);
4410 		schedule();
4411 	}
4412 
4413 	spec = efx_ef10_filter_entry_spec(table, filter_idx);
4414 	if (!spec ||
4415 	    (!by_index &&
4416 	     efx_ef10_filter_pri(table, spec) !=
4417 	     efx_ef10_filter_get_unsafe_pri(filter_id))) {
4418 		rc = -ENOENT;
4419 		goto out_unlock;
4420 	}
4421 
4422 	if (spec->flags & EFX_FILTER_FLAG_RX_OVER_AUTO &&
4423 	    priority_mask == (1U << EFX_FILTER_PRI_AUTO)) {
4424 		/* Just remove flags */
4425 		spec->flags &= ~EFX_FILTER_FLAG_RX_OVER_AUTO;
4426 		table->entry[filter_idx].spec &= ~EFX_EF10_FILTER_FLAG_AUTO_OLD;
4427 		rc = 0;
4428 		goto out_unlock;
4429 	}
4430 
4431 	if (!(priority_mask & (1U << spec->priority))) {
4432 		rc = -ENOENT;
4433 		goto out_unlock;
4434 	}
4435 
4436 	table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_BUSY;
4437 	spin_unlock_bh(&efx->filter_lock);
4438 
4439 	if (spec->flags & EFX_FILTER_FLAG_RX_OVER_AUTO) {
4440 		/* Reset to an automatic filter */
4441 
4442 		struct efx_filter_spec new_spec = *spec;
4443 
4444 		new_spec.priority = EFX_FILTER_PRI_AUTO;
4445 		new_spec.flags = (EFX_FILTER_FLAG_RX |
4446 				  (efx_rss_enabled(efx) ?
4447 				   EFX_FILTER_FLAG_RX_RSS : 0));
4448 		new_spec.dmaq_id = 0;
4449 		new_spec.rss_context = EFX_FILTER_RSS_CONTEXT_DEFAULT;
4450 		rc = efx_ef10_filter_push(efx, &new_spec,
4451 					  &table->entry[filter_idx].handle,
4452 					  true);
4453 
4454 		spin_lock_bh(&efx->filter_lock);
4455 		if (rc == 0)
4456 			*spec = new_spec;
4457 	} else {
4458 		/* Really remove the filter */
4459 
4460 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
4461 			       efx_ef10_filter_is_exclusive(spec) ?
4462 			       MC_CMD_FILTER_OP_IN_OP_REMOVE :
4463 			       MC_CMD_FILTER_OP_IN_OP_UNSUBSCRIBE);
4464 		MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
4465 			       table->entry[filter_idx].handle);
4466 		rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FILTER_OP,
4467 					inbuf, sizeof(inbuf), NULL, 0, NULL);
4468 
4469 		spin_lock_bh(&efx->filter_lock);
4470 		if ((rc == 0) || (rc == -ENOENT)) {
4471 			/* Filter removed OK or didn't actually exist */
4472 			kfree(spec);
4473 			efx_ef10_filter_set_entry(table, filter_idx, NULL, 0);
4474 		} else {
4475 			efx_mcdi_display_error(efx, MC_CMD_FILTER_OP,
4476 					       MC_CMD_FILTER_OP_EXT_IN_LEN,
4477 					       NULL, 0, rc);
4478 		}
4479 	}
4480 
4481 	table->entry[filter_idx].spec &= ~EFX_EF10_FILTER_FLAG_BUSY;
4482 	wake_up_all(&table->waitq);
4483 out_unlock:
4484 	spin_unlock_bh(&efx->filter_lock);
4485 	finish_wait(&table->waitq, &wait);
4486 	return rc;
4487 }
4488 
4489 static int efx_ef10_filter_remove_safe(struct efx_nic *efx,
4490 				       enum efx_filter_priority priority,
4491 				       u32 filter_id)
4492 {
4493 	return efx_ef10_filter_remove_internal(efx, 1U << priority,
4494 					       filter_id, false);
4495 }
4496 
4497 static void efx_ef10_filter_remove_unsafe(struct efx_nic *efx,
4498 					  enum efx_filter_priority priority,
4499 					  u32 filter_id)
4500 {
4501 	if (filter_id == EFX_EF10_FILTER_ID_INVALID)
4502 		return;
4503 	efx_ef10_filter_remove_internal(efx, 1U << priority, filter_id, true);
4504 }
4505 
4506 static int efx_ef10_filter_get_safe(struct efx_nic *efx,
4507 				    enum efx_filter_priority priority,
4508 				    u32 filter_id, struct efx_filter_spec *spec)
4509 {
4510 	unsigned int filter_idx = efx_ef10_filter_get_unsafe_id(filter_id);
4511 	struct efx_ef10_filter_table *table = efx->filter_state;
4512 	const struct efx_filter_spec *saved_spec;
4513 	int rc;
4514 
4515 	spin_lock_bh(&efx->filter_lock);
4516 	saved_spec = efx_ef10_filter_entry_spec(table, filter_idx);
4517 	if (saved_spec && saved_spec->priority == priority &&
4518 	    efx_ef10_filter_pri(table, saved_spec) ==
4519 	    efx_ef10_filter_get_unsafe_pri(filter_id)) {
4520 		*spec = *saved_spec;
4521 		rc = 0;
4522 	} else {
4523 		rc = -ENOENT;
4524 	}
4525 	spin_unlock_bh(&efx->filter_lock);
4526 	return rc;
4527 }
4528 
4529 static int efx_ef10_filter_clear_rx(struct efx_nic *efx,
4530 				     enum efx_filter_priority priority)
4531 {
4532 	unsigned int priority_mask;
4533 	unsigned int i;
4534 	int rc;
4535 
4536 	priority_mask = (((1U << (priority + 1)) - 1) &
4537 			 ~(1U << EFX_FILTER_PRI_AUTO));
4538 
4539 	for (i = 0; i < HUNT_FILTER_TBL_ROWS; i++) {
4540 		rc = efx_ef10_filter_remove_internal(efx, priority_mask,
4541 						     i, true);
4542 		if (rc && rc != -ENOENT)
4543 			return rc;
4544 	}
4545 
4546 	return 0;
4547 }
4548 
4549 static u32 efx_ef10_filter_count_rx_used(struct efx_nic *efx,
4550 					 enum efx_filter_priority priority)
4551 {
4552 	struct efx_ef10_filter_table *table = efx->filter_state;
4553 	unsigned int filter_idx;
4554 	s32 count = 0;
4555 
4556 	spin_lock_bh(&efx->filter_lock);
4557 	for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
4558 		if (table->entry[filter_idx].spec &&
4559 		    efx_ef10_filter_entry_spec(table, filter_idx)->priority ==
4560 		    priority)
4561 			++count;
4562 	}
4563 	spin_unlock_bh(&efx->filter_lock);
4564 	return count;
4565 }
4566 
4567 static u32 efx_ef10_filter_get_rx_id_limit(struct efx_nic *efx)
4568 {
4569 	struct efx_ef10_filter_table *table = efx->filter_state;
4570 
4571 	return table->rx_match_count * HUNT_FILTER_TBL_ROWS * 2;
4572 }
4573 
4574 static s32 efx_ef10_filter_get_rx_ids(struct efx_nic *efx,
4575 				      enum efx_filter_priority priority,
4576 				      u32 *buf, u32 size)
4577 {
4578 	struct efx_ef10_filter_table *table = efx->filter_state;
4579 	struct efx_filter_spec *spec;
4580 	unsigned int filter_idx;
4581 	s32 count = 0;
4582 
4583 	spin_lock_bh(&efx->filter_lock);
4584 	for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
4585 		spec = efx_ef10_filter_entry_spec(table, filter_idx);
4586 		if (spec && spec->priority == priority) {
4587 			if (count == size) {
4588 				count = -EMSGSIZE;
4589 				break;
4590 			}
4591 			buf[count++] =
4592 				efx_ef10_make_filter_id(
4593 					efx_ef10_filter_pri(table, spec),
4594 					filter_idx);
4595 		}
4596 	}
4597 	spin_unlock_bh(&efx->filter_lock);
4598 	return count;
4599 }
4600 
4601 #ifdef CONFIG_RFS_ACCEL
4602 
4603 static efx_mcdi_async_completer efx_ef10_filter_rfs_insert_complete;
4604 
4605 static s32 efx_ef10_filter_rfs_insert(struct efx_nic *efx,
4606 				      struct efx_filter_spec *spec)
4607 {
4608 	struct efx_ef10_filter_table *table = efx->filter_state;
4609 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_EXT_IN_LEN);
4610 	struct efx_filter_spec *saved_spec;
4611 	unsigned int hash, i, depth = 1;
4612 	bool replacing = false;
4613 	int ins_index = -1;
4614 	u64 cookie;
4615 	s32 rc;
4616 
4617 	/* Must be an RX filter without RSS and not for a multicast
4618 	 * destination address (RFS only works for connected sockets).
4619 	 * These restrictions allow us to pass only a tiny amount of
4620 	 * data through to the completion function.
4621 	 */
4622 	EFX_WARN_ON_PARANOID(spec->flags !=
4623 			     (EFX_FILTER_FLAG_RX | EFX_FILTER_FLAG_RX_SCATTER));
4624 	EFX_WARN_ON_PARANOID(spec->priority != EFX_FILTER_PRI_HINT);
4625 	EFX_WARN_ON_PARANOID(efx_filter_is_mc_recipient(spec));
4626 
4627 	hash = efx_ef10_filter_hash(spec);
4628 
4629 	spin_lock_bh(&efx->filter_lock);
4630 
4631 	/* Find any existing filter with the same match tuple or else
4632 	 * a free slot to insert at.  If an existing filter is busy,
4633 	 * we have to give up.
4634 	 */
4635 	for (;;) {
4636 		i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
4637 		saved_spec = efx_ef10_filter_entry_spec(table, i);
4638 
4639 		if (!saved_spec) {
4640 			if (ins_index < 0)
4641 				ins_index = i;
4642 		} else if (efx_ef10_filter_equal(spec, saved_spec)) {
4643 			if (table->entry[i].spec & EFX_EF10_FILTER_FLAG_BUSY) {
4644 				rc = -EBUSY;
4645 				goto fail_unlock;
4646 			}
4647 			if (spec->priority < saved_spec->priority) {
4648 				rc = -EPERM;
4649 				goto fail_unlock;
4650 			}
4651 			ins_index = i;
4652 			break;
4653 		}
4654 
4655 		/* Once we reach the maximum search depth, use the
4656 		 * first suitable slot or return -EBUSY if there was
4657 		 * none
4658 		 */
4659 		if (depth == EFX_EF10_FILTER_SEARCH_LIMIT) {
4660 			if (ins_index < 0) {
4661 				rc = -EBUSY;
4662 				goto fail_unlock;
4663 			}
4664 			break;
4665 		}
4666 
4667 		++depth;
4668 	}
4669 
4670 	/* Create a software table entry if necessary, and mark it
4671 	 * busy.  We might yet fail to insert, but any attempt to
4672 	 * insert a conflicting filter while we're waiting for the
4673 	 * firmware must find the busy entry.
4674 	 */
4675 	saved_spec = efx_ef10_filter_entry_spec(table, ins_index);
4676 	if (saved_spec) {
4677 		replacing = true;
4678 	} else {
4679 		saved_spec = kmalloc(sizeof(*spec), GFP_ATOMIC);
4680 		if (!saved_spec) {
4681 			rc = -ENOMEM;
4682 			goto fail_unlock;
4683 		}
4684 		*saved_spec = *spec;
4685 	}
4686 	efx_ef10_filter_set_entry(table, ins_index, saved_spec,
4687 				  EFX_EF10_FILTER_FLAG_BUSY);
4688 
4689 	spin_unlock_bh(&efx->filter_lock);
4690 
4691 	/* Pack up the variables needed on completion */
4692 	cookie = replacing << 31 | ins_index << 16 | spec->dmaq_id;
4693 
4694 	efx_ef10_filter_push_prep(efx, spec, inbuf,
4695 				  table->entry[ins_index].handle, replacing);
4696 	efx_mcdi_rpc_async(efx, MC_CMD_FILTER_OP, inbuf, sizeof(inbuf),
4697 			   MC_CMD_FILTER_OP_OUT_LEN,
4698 			   efx_ef10_filter_rfs_insert_complete, cookie);
4699 
4700 	return ins_index;
4701 
4702 fail_unlock:
4703 	spin_unlock_bh(&efx->filter_lock);
4704 	return rc;
4705 }
4706 
4707 static void
4708 efx_ef10_filter_rfs_insert_complete(struct efx_nic *efx, unsigned long cookie,
4709 				    int rc, efx_dword_t *outbuf,
4710 				    size_t outlen_actual)
4711 {
4712 	struct efx_ef10_filter_table *table = efx->filter_state;
4713 	unsigned int ins_index, dmaq_id;
4714 	struct efx_filter_spec *spec;
4715 	bool replacing;
4716 
4717 	/* Unpack the cookie */
4718 	replacing = cookie >> 31;
4719 	ins_index = (cookie >> 16) & (HUNT_FILTER_TBL_ROWS - 1);
4720 	dmaq_id = cookie & 0xffff;
4721 
4722 	spin_lock_bh(&efx->filter_lock);
4723 	spec = efx_ef10_filter_entry_spec(table, ins_index);
4724 	if (rc == 0) {
4725 		table->entry[ins_index].handle =
4726 			MCDI_QWORD(outbuf, FILTER_OP_OUT_HANDLE);
4727 		if (replacing)
4728 			spec->dmaq_id = dmaq_id;
4729 	} else if (!replacing) {
4730 		kfree(spec);
4731 		spec = NULL;
4732 	}
4733 	efx_ef10_filter_set_entry(table, ins_index, spec, 0);
4734 	spin_unlock_bh(&efx->filter_lock);
4735 
4736 	wake_up_all(&table->waitq);
4737 }
4738 
4739 static void
4740 efx_ef10_filter_rfs_expire_complete(struct efx_nic *efx,
4741 				    unsigned long filter_idx,
4742 				    int rc, efx_dword_t *outbuf,
4743 				    size_t outlen_actual);
4744 
4745 static bool efx_ef10_filter_rfs_expire_one(struct efx_nic *efx, u32 flow_id,
4746 					   unsigned int filter_idx)
4747 {
4748 	struct efx_ef10_filter_table *table = efx->filter_state;
4749 	struct efx_filter_spec *spec =
4750 		efx_ef10_filter_entry_spec(table, filter_idx);
4751 	MCDI_DECLARE_BUF(inbuf,
4752 			 MC_CMD_FILTER_OP_IN_HANDLE_OFST +
4753 			 MC_CMD_FILTER_OP_IN_HANDLE_LEN);
4754 
4755 	if (!spec ||
4756 	    (table->entry[filter_idx].spec & EFX_EF10_FILTER_FLAG_BUSY) ||
4757 	    spec->priority != EFX_FILTER_PRI_HINT ||
4758 	    !rps_may_expire_flow(efx->net_dev, spec->dmaq_id,
4759 				 flow_id, filter_idx))
4760 		return false;
4761 
4762 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
4763 		       MC_CMD_FILTER_OP_IN_OP_REMOVE);
4764 	MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
4765 		       table->entry[filter_idx].handle);
4766 	if (efx_mcdi_rpc_async(efx, MC_CMD_FILTER_OP, inbuf, sizeof(inbuf), 0,
4767 			       efx_ef10_filter_rfs_expire_complete, filter_idx))
4768 		return false;
4769 
4770 	table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_BUSY;
4771 	return true;
4772 }
4773 
4774 static void
4775 efx_ef10_filter_rfs_expire_complete(struct efx_nic *efx,
4776 				    unsigned long filter_idx,
4777 				    int rc, efx_dword_t *outbuf,
4778 				    size_t outlen_actual)
4779 {
4780 	struct efx_ef10_filter_table *table = efx->filter_state;
4781 	struct efx_filter_spec *spec =
4782 		efx_ef10_filter_entry_spec(table, filter_idx);
4783 
4784 	spin_lock_bh(&efx->filter_lock);
4785 	if (rc == 0) {
4786 		kfree(spec);
4787 		efx_ef10_filter_set_entry(table, filter_idx, NULL, 0);
4788 	}
4789 	table->entry[filter_idx].spec &= ~EFX_EF10_FILTER_FLAG_BUSY;
4790 	wake_up_all(&table->waitq);
4791 	spin_unlock_bh(&efx->filter_lock);
4792 }
4793 
4794 #endif /* CONFIG_RFS_ACCEL */
4795 
4796 static int efx_ef10_filter_match_flags_from_mcdi(bool encap, u32 mcdi_flags)
4797 {
4798 	int match_flags = 0;
4799 
4800 #define MAP_FLAG(gen_flag, mcdi_field) do {				\
4801 		u32 old_mcdi_flags = mcdi_flags;			\
4802 		mcdi_flags &= ~(1 << MC_CMD_FILTER_OP_EXT_IN_MATCH_ ##	\
4803 				     mcdi_field ## _LBN);		\
4804 		if (mcdi_flags != old_mcdi_flags)			\
4805 			match_flags |= EFX_FILTER_MATCH_ ## gen_flag;	\
4806 	} while (0)
4807 
4808 	if (encap) {
4809 		/* encap filters must specify encap type */
4810 		match_flags |= EFX_FILTER_MATCH_ENCAP_TYPE;
4811 		/* and imply ethertype and ip proto */
4812 		mcdi_flags &=
4813 			~(1 << MC_CMD_FILTER_OP_EXT_IN_MATCH_IP_PROTO_LBN);
4814 		mcdi_flags &=
4815 			~(1 << MC_CMD_FILTER_OP_EXT_IN_MATCH_ETHER_TYPE_LBN);
4816 		/* VLAN tags refer to the outer packet */
4817 		MAP_FLAG(INNER_VID, INNER_VLAN);
4818 		MAP_FLAG(OUTER_VID, OUTER_VLAN);
4819 		/* everything else refers to the inner packet */
4820 		MAP_FLAG(LOC_MAC_IG, IFRM_UNKNOWN_UCAST_DST);
4821 		MAP_FLAG(LOC_MAC_IG, IFRM_UNKNOWN_MCAST_DST);
4822 		MAP_FLAG(REM_HOST, IFRM_SRC_IP);
4823 		MAP_FLAG(LOC_HOST, IFRM_DST_IP);
4824 		MAP_FLAG(REM_MAC, IFRM_SRC_MAC);
4825 		MAP_FLAG(REM_PORT, IFRM_SRC_PORT);
4826 		MAP_FLAG(LOC_MAC, IFRM_DST_MAC);
4827 		MAP_FLAG(LOC_PORT, IFRM_DST_PORT);
4828 		MAP_FLAG(ETHER_TYPE, IFRM_ETHER_TYPE);
4829 		MAP_FLAG(IP_PROTO, IFRM_IP_PROTO);
4830 	} else {
4831 		MAP_FLAG(LOC_MAC_IG, UNKNOWN_UCAST_DST);
4832 		MAP_FLAG(LOC_MAC_IG, UNKNOWN_MCAST_DST);
4833 		MAP_FLAG(REM_HOST, SRC_IP);
4834 		MAP_FLAG(LOC_HOST, DST_IP);
4835 		MAP_FLAG(REM_MAC, SRC_MAC);
4836 		MAP_FLAG(REM_PORT, SRC_PORT);
4837 		MAP_FLAG(LOC_MAC, DST_MAC);
4838 		MAP_FLAG(LOC_PORT, DST_PORT);
4839 		MAP_FLAG(ETHER_TYPE, ETHER_TYPE);
4840 		MAP_FLAG(INNER_VID, INNER_VLAN);
4841 		MAP_FLAG(OUTER_VID, OUTER_VLAN);
4842 		MAP_FLAG(IP_PROTO, IP_PROTO);
4843 	}
4844 #undef MAP_FLAG
4845 
4846 	/* Did we map them all? */
4847 	if (mcdi_flags)
4848 		return -EINVAL;
4849 
4850 	return match_flags;
4851 }
4852 
4853 static void efx_ef10_filter_cleanup_vlans(struct efx_nic *efx)
4854 {
4855 	struct efx_ef10_filter_table *table = efx->filter_state;
4856 	struct efx_ef10_filter_vlan *vlan, *next_vlan;
4857 
4858 	/* See comment in efx_ef10_filter_table_remove() */
4859 	if (!efx_rwsem_assert_write_locked(&efx->filter_sem))
4860 		return;
4861 
4862 	if (!table)
4863 		return;
4864 
4865 	list_for_each_entry_safe(vlan, next_vlan, &table->vlan_list, list)
4866 		efx_ef10_filter_del_vlan_internal(efx, vlan);
4867 }
4868 
4869 static bool efx_ef10_filter_match_supported(struct efx_ef10_filter_table *table,
4870 					    bool encap,
4871 					    enum efx_filter_match_flags match_flags)
4872 {
4873 	unsigned int match_pri;
4874 	int mf;
4875 
4876 	for (match_pri = 0;
4877 	     match_pri < table->rx_match_count;
4878 	     match_pri++) {
4879 		mf = efx_ef10_filter_match_flags_from_mcdi(encap,
4880 				table->rx_match_mcdi_flags[match_pri]);
4881 		if (mf == match_flags)
4882 			return true;
4883 	}
4884 
4885 	return false;
4886 }
4887 
4888 static int
4889 efx_ef10_filter_table_probe_matches(struct efx_nic *efx,
4890 				    struct efx_ef10_filter_table *table,
4891 				    bool encap)
4892 {
4893 	MCDI_DECLARE_BUF(inbuf, MC_CMD_GET_PARSER_DISP_INFO_IN_LEN);
4894 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_PARSER_DISP_INFO_OUT_LENMAX);
4895 	unsigned int pd_match_pri, pd_match_count;
4896 	size_t outlen;
4897 	int rc;
4898 
4899 	/* Find out which RX filter types are supported, and their priorities */
4900 	MCDI_SET_DWORD(inbuf, GET_PARSER_DISP_INFO_IN_OP,
4901 		       encap ?
4902 		       MC_CMD_GET_PARSER_DISP_INFO_IN_OP_GET_SUPPORTED_ENCAP_RX_MATCHES :
4903 		       MC_CMD_GET_PARSER_DISP_INFO_IN_OP_GET_SUPPORTED_RX_MATCHES);
4904 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_PARSER_DISP_INFO,
4905 			  inbuf, sizeof(inbuf), outbuf, sizeof(outbuf),
4906 			  &outlen);
4907 	if (rc)
4908 		return rc;
4909 
4910 	pd_match_count = MCDI_VAR_ARRAY_LEN(
4911 		outlen, GET_PARSER_DISP_INFO_OUT_SUPPORTED_MATCHES);
4912 
4913 	for (pd_match_pri = 0; pd_match_pri < pd_match_count; pd_match_pri++) {
4914 		u32 mcdi_flags =
4915 			MCDI_ARRAY_DWORD(
4916 				outbuf,
4917 				GET_PARSER_DISP_INFO_OUT_SUPPORTED_MATCHES,
4918 				pd_match_pri);
4919 		rc = efx_ef10_filter_match_flags_from_mcdi(encap, mcdi_flags);
4920 		if (rc < 0) {
4921 			netif_dbg(efx, probe, efx->net_dev,
4922 				  "%s: fw flags %#x pri %u not supported in driver\n",
4923 				  __func__, mcdi_flags, pd_match_pri);
4924 		} else {
4925 			netif_dbg(efx, probe, efx->net_dev,
4926 				  "%s: fw flags %#x pri %u supported as driver flags %#x pri %u\n",
4927 				  __func__, mcdi_flags, pd_match_pri,
4928 				  rc, table->rx_match_count);
4929 			table->rx_match_mcdi_flags[table->rx_match_count] = mcdi_flags;
4930 			table->rx_match_count++;
4931 		}
4932 	}
4933 
4934 	return 0;
4935 }
4936 
4937 static int efx_ef10_filter_table_probe(struct efx_nic *efx)
4938 {
4939 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
4940 	struct net_device *net_dev = efx->net_dev;
4941 	struct efx_ef10_filter_table *table;
4942 	struct efx_ef10_vlan *vlan;
4943 	int rc;
4944 
4945 	if (!efx_rwsem_assert_write_locked(&efx->filter_sem))
4946 		return -EINVAL;
4947 
4948 	if (efx->filter_state) /* already probed */
4949 		return 0;
4950 
4951 	table = kzalloc(sizeof(*table), GFP_KERNEL);
4952 	if (!table)
4953 		return -ENOMEM;
4954 
4955 	table->rx_match_count = 0;
4956 	rc = efx_ef10_filter_table_probe_matches(efx, table, false);
4957 	if (rc)
4958 		goto fail;
4959 	if (nic_data->datapath_caps &
4960 		   (1 << MC_CMD_GET_CAPABILITIES_OUT_VXLAN_NVGRE_LBN))
4961 		rc = efx_ef10_filter_table_probe_matches(efx, table, true);
4962 	if (rc)
4963 		goto fail;
4964 	if ((efx_supported_features(efx) & NETIF_F_HW_VLAN_CTAG_FILTER) &&
4965 	    !(efx_ef10_filter_match_supported(table, false,
4966 		(EFX_FILTER_MATCH_OUTER_VID | EFX_FILTER_MATCH_LOC_MAC)) &&
4967 	      efx_ef10_filter_match_supported(table, false,
4968 		(EFX_FILTER_MATCH_OUTER_VID | EFX_FILTER_MATCH_LOC_MAC_IG)))) {
4969 		netif_info(efx, probe, net_dev,
4970 			   "VLAN filters are not supported in this firmware variant\n");
4971 		net_dev->features &= ~NETIF_F_HW_VLAN_CTAG_FILTER;
4972 		efx->fixed_features &= ~NETIF_F_HW_VLAN_CTAG_FILTER;
4973 		net_dev->hw_features &= ~NETIF_F_HW_VLAN_CTAG_FILTER;
4974 	}
4975 
4976 	table->entry = vzalloc(HUNT_FILTER_TBL_ROWS * sizeof(*table->entry));
4977 	if (!table->entry) {
4978 		rc = -ENOMEM;
4979 		goto fail;
4980 	}
4981 
4982 	table->mc_promisc_last = false;
4983 	table->vlan_filter =
4984 		!!(efx->net_dev->features & NETIF_F_HW_VLAN_CTAG_FILTER);
4985 	INIT_LIST_HEAD(&table->vlan_list);
4986 
4987 	efx->filter_state = table;
4988 	init_waitqueue_head(&table->waitq);
4989 
4990 	list_for_each_entry(vlan, &nic_data->vlan_list, list) {
4991 		rc = efx_ef10_filter_add_vlan(efx, vlan->vid);
4992 		if (rc)
4993 			goto fail_add_vlan;
4994 	}
4995 
4996 	return 0;
4997 
4998 fail_add_vlan:
4999 	efx_ef10_filter_cleanup_vlans(efx);
5000 	efx->filter_state = NULL;
5001 fail:
5002 	kfree(table);
5003 	return rc;
5004 }
5005 
5006 /* Caller must hold efx->filter_sem for read if race against
5007  * efx_ef10_filter_table_remove() is possible
5008  */
5009 static void efx_ef10_filter_table_restore(struct efx_nic *efx)
5010 {
5011 	struct efx_ef10_filter_table *table = efx->filter_state;
5012 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
5013 	unsigned int invalid_filters = 0, failed = 0;
5014 	struct efx_ef10_filter_vlan *vlan;
5015 	struct efx_filter_spec *spec;
5016 	unsigned int filter_idx;
5017 	u32 mcdi_flags;
5018 	int match_pri;
5019 	int rc, i;
5020 
5021 	WARN_ON(!rwsem_is_locked(&efx->filter_sem));
5022 
5023 	if (!nic_data->must_restore_filters)
5024 		return;
5025 
5026 	if (!table)
5027 		return;
5028 
5029 	spin_lock_bh(&efx->filter_lock);
5030 
5031 	for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
5032 		spec = efx_ef10_filter_entry_spec(table, filter_idx);
5033 		if (!spec)
5034 			continue;
5035 
5036 		mcdi_flags = efx_ef10_filter_mcdi_flags_from_spec(spec);
5037 		match_pri = 0;
5038 		while (match_pri < table->rx_match_count &&
5039 		       table->rx_match_mcdi_flags[match_pri] != mcdi_flags)
5040 			++match_pri;
5041 		if (match_pri >= table->rx_match_count) {
5042 			invalid_filters++;
5043 			goto not_restored;
5044 		}
5045 		if (spec->rss_context != EFX_FILTER_RSS_CONTEXT_DEFAULT &&
5046 		    spec->rss_context != nic_data->rx_rss_context)
5047 			netif_warn(efx, drv, efx->net_dev,
5048 				   "Warning: unable to restore a filter with specific RSS context.\n");
5049 
5050 		table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_BUSY;
5051 		spin_unlock_bh(&efx->filter_lock);
5052 
5053 		rc = efx_ef10_filter_push(efx, spec,
5054 					  &table->entry[filter_idx].handle,
5055 					  false);
5056 		if (rc)
5057 			failed++;
5058 		spin_lock_bh(&efx->filter_lock);
5059 
5060 		if (rc) {
5061 not_restored:
5062 			list_for_each_entry(vlan, &table->vlan_list, list)
5063 				for (i = 0; i < EFX_EF10_NUM_DEFAULT_FILTERS; ++i)
5064 					if (vlan->default_filters[i] == filter_idx)
5065 						vlan->default_filters[i] =
5066 							EFX_EF10_FILTER_ID_INVALID;
5067 
5068 			kfree(spec);
5069 			efx_ef10_filter_set_entry(table, filter_idx, NULL, 0);
5070 		} else {
5071 			table->entry[filter_idx].spec &=
5072 				~EFX_EF10_FILTER_FLAG_BUSY;
5073 		}
5074 	}
5075 
5076 	spin_unlock_bh(&efx->filter_lock);
5077 
5078 	/* This can happen validly if the MC's capabilities have changed, so
5079 	 * is not an error.
5080 	 */
5081 	if (invalid_filters)
5082 		netif_dbg(efx, drv, efx->net_dev,
5083 			  "Did not restore %u filters that are now unsupported.\n",
5084 			  invalid_filters);
5085 
5086 	if (failed)
5087 		netif_err(efx, hw, efx->net_dev,
5088 			  "unable to restore %u filters\n", failed);
5089 	else
5090 		nic_data->must_restore_filters = false;
5091 }
5092 
5093 static void efx_ef10_filter_table_remove(struct efx_nic *efx)
5094 {
5095 	struct efx_ef10_filter_table *table = efx->filter_state;
5096 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_EXT_IN_LEN);
5097 	struct efx_filter_spec *spec;
5098 	unsigned int filter_idx;
5099 	int rc;
5100 
5101 	efx_ef10_filter_cleanup_vlans(efx);
5102 	efx->filter_state = NULL;
5103 	/* If we were called without locking, then it's not safe to free
5104 	 * the table as others might be using it.  So we just WARN, leak
5105 	 * the memory, and potentially get an inconsistent filter table
5106 	 * state.
5107 	 * This should never actually happen.
5108 	 */
5109 	if (!efx_rwsem_assert_write_locked(&efx->filter_sem))
5110 		return;
5111 
5112 	if (!table)
5113 		return;
5114 
5115 	for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
5116 		spec = efx_ef10_filter_entry_spec(table, filter_idx);
5117 		if (!spec)
5118 			continue;
5119 
5120 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
5121 			       efx_ef10_filter_is_exclusive(spec) ?
5122 			       MC_CMD_FILTER_OP_IN_OP_REMOVE :
5123 			       MC_CMD_FILTER_OP_IN_OP_UNSUBSCRIBE);
5124 		MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
5125 			       table->entry[filter_idx].handle);
5126 		rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FILTER_OP, inbuf,
5127 					sizeof(inbuf), NULL, 0, NULL);
5128 		if (rc)
5129 			netif_info(efx, drv, efx->net_dev,
5130 				   "%s: filter %04x remove failed\n",
5131 				   __func__, filter_idx);
5132 		kfree(spec);
5133 	}
5134 
5135 	vfree(table->entry);
5136 	kfree(table);
5137 }
5138 
5139 static void efx_ef10_filter_mark_one_old(struct efx_nic *efx, uint16_t *id)
5140 {
5141 	struct efx_ef10_filter_table *table = efx->filter_state;
5142 	unsigned int filter_idx;
5143 
5144 	if (*id != EFX_EF10_FILTER_ID_INVALID) {
5145 		filter_idx = efx_ef10_filter_get_unsafe_id(*id);
5146 		if (!table->entry[filter_idx].spec)
5147 			netif_dbg(efx, drv, efx->net_dev,
5148 				  "marked null spec old %04x:%04x\n", *id,
5149 				  filter_idx);
5150 		table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_AUTO_OLD;
5151 		*id = EFX_EF10_FILTER_ID_INVALID;
5152 	}
5153 }
5154 
5155 /* Mark old per-VLAN filters that may need to be removed */
5156 static void _efx_ef10_filter_vlan_mark_old(struct efx_nic *efx,
5157 					   struct efx_ef10_filter_vlan *vlan)
5158 {
5159 	struct efx_ef10_filter_table *table = efx->filter_state;
5160 	unsigned int i;
5161 
5162 	for (i = 0; i < table->dev_uc_count; i++)
5163 		efx_ef10_filter_mark_one_old(efx, &vlan->uc[i]);
5164 	for (i = 0; i < table->dev_mc_count; i++)
5165 		efx_ef10_filter_mark_one_old(efx, &vlan->mc[i]);
5166 	for (i = 0; i < EFX_EF10_NUM_DEFAULT_FILTERS; i++)
5167 		efx_ef10_filter_mark_one_old(efx, &vlan->default_filters[i]);
5168 }
5169 
5170 /* Mark old filters that may need to be removed.
5171  * Caller must hold efx->filter_sem for read if race against
5172  * efx_ef10_filter_table_remove() is possible
5173  */
5174 static void efx_ef10_filter_mark_old(struct efx_nic *efx)
5175 {
5176 	struct efx_ef10_filter_table *table = efx->filter_state;
5177 	struct efx_ef10_filter_vlan *vlan;
5178 
5179 	spin_lock_bh(&efx->filter_lock);
5180 	list_for_each_entry(vlan, &table->vlan_list, list)
5181 		_efx_ef10_filter_vlan_mark_old(efx, vlan);
5182 	spin_unlock_bh(&efx->filter_lock);
5183 }
5184 
5185 static void efx_ef10_filter_uc_addr_list(struct efx_nic *efx)
5186 {
5187 	struct efx_ef10_filter_table *table = efx->filter_state;
5188 	struct net_device *net_dev = efx->net_dev;
5189 	struct netdev_hw_addr *uc;
5190 	unsigned int i;
5191 
5192 	table->uc_promisc = !!(net_dev->flags & IFF_PROMISC);
5193 	ether_addr_copy(table->dev_uc_list[0].addr, net_dev->dev_addr);
5194 	i = 1;
5195 	netdev_for_each_uc_addr(uc, net_dev) {
5196 		if (i >= EFX_EF10_FILTER_DEV_UC_MAX) {
5197 			table->uc_promisc = true;
5198 			break;
5199 		}
5200 		ether_addr_copy(table->dev_uc_list[i].addr, uc->addr);
5201 		i++;
5202 	}
5203 
5204 	table->dev_uc_count = i;
5205 }
5206 
5207 static void efx_ef10_filter_mc_addr_list(struct efx_nic *efx)
5208 {
5209 	struct efx_ef10_filter_table *table = efx->filter_state;
5210 	struct net_device *net_dev = efx->net_dev;
5211 	struct netdev_hw_addr *mc;
5212 	unsigned int i;
5213 
5214 	table->mc_overflow = false;
5215 	table->mc_promisc = !!(net_dev->flags & (IFF_PROMISC | IFF_ALLMULTI));
5216 
5217 	i = 0;
5218 	netdev_for_each_mc_addr(mc, net_dev) {
5219 		if (i >= EFX_EF10_FILTER_DEV_MC_MAX) {
5220 			table->mc_promisc = true;
5221 			table->mc_overflow = true;
5222 			break;
5223 		}
5224 		ether_addr_copy(table->dev_mc_list[i].addr, mc->addr);
5225 		i++;
5226 	}
5227 
5228 	table->dev_mc_count = i;
5229 }
5230 
5231 static int efx_ef10_filter_insert_addr_list(struct efx_nic *efx,
5232 					    struct efx_ef10_filter_vlan *vlan,
5233 					    bool multicast, bool rollback)
5234 {
5235 	struct efx_ef10_filter_table *table = efx->filter_state;
5236 	struct efx_ef10_dev_addr *addr_list;
5237 	enum efx_filter_flags filter_flags;
5238 	struct efx_filter_spec spec;
5239 	u8 baddr[ETH_ALEN];
5240 	unsigned int i, j;
5241 	int addr_count;
5242 	u16 *ids;
5243 	int rc;
5244 
5245 	if (multicast) {
5246 		addr_list = table->dev_mc_list;
5247 		addr_count = table->dev_mc_count;
5248 		ids = vlan->mc;
5249 	} else {
5250 		addr_list = table->dev_uc_list;
5251 		addr_count = table->dev_uc_count;
5252 		ids = vlan->uc;
5253 	}
5254 
5255 	filter_flags = efx_rss_enabled(efx) ? EFX_FILTER_FLAG_RX_RSS : 0;
5256 
5257 	/* Insert/renew filters */
5258 	for (i = 0; i < addr_count; i++) {
5259 		EFX_WARN_ON_PARANOID(ids[i] != EFX_EF10_FILTER_ID_INVALID);
5260 		efx_filter_init_rx(&spec, EFX_FILTER_PRI_AUTO, filter_flags, 0);
5261 		efx_filter_set_eth_local(&spec, vlan->vid, addr_list[i].addr);
5262 		rc = efx_ef10_filter_insert(efx, &spec, true);
5263 		if (rc < 0) {
5264 			if (rollback) {
5265 				netif_info(efx, drv, efx->net_dev,
5266 					   "efx_ef10_filter_insert failed rc=%d\n",
5267 					   rc);
5268 				/* Fall back to promiscuous */
5269 				for (j = 0; j < i; j++) {
5270 					efx_ef10_filter_remove_unsafe(
5271 						efx, EFX_FILTER_PRI_AUTO,
5272 						ids[j]);
5273 					ids[j] = EFX_EF10_FILTER_ID_INVALID;
5274 				}
5275 				return rc;
5276 			} else {
5277 				/* keep invalid ID, and carry on */
5278 			}
5279 		} else {
5280 			ids[i] = efx_ef10_filter_get_unsafe_id(rc);
5281 		}
5282 	}
5283 
5284 	if (multicast && rollback) {
5285 		/* Also need an Ethernet broadcast filter */
5286 		EFX_WARN_ON_PARANOID(vlan->default_filters[EFX_EF10_BCAST] !=
5287 				     EFX_EF10_FILTER_ID_INVALID);
5288 		efx_filter_init_rx(&spec, EFX_FILTER_PRI_AUTO, filter_flags, 0);
5289 		eth_broadcast_addr(baddr);
5290 		efx_filter_set_eth_local(&spec, vlan->vid, baddr);
5291 		rc = efx_ef10_filter_insert(efx, &spec, true);
5292 		if (rc < 0) {
5293 			netif_warn(efx, drv, efx->net_dev,
5294 				   "Broadcast filter insert failed rc=%d\n", rc);
5295 			/* Fall back to promiscuous */
5296 			for (j = 0; j < i; j++) {
5297 				efx_ef10_filter_remove_unsafe(
5298 					efx, EFX_FILTER_PRI_AUTO,
5299 					ids[j]);
5300 				ids[j] = EFX_EF10_FILTER_ID_INVALID;
5301 			}
5302 			return rc;
5303 		} else {
5304 			vlan->default_filters[EFX_EF10_BCAST] =
5305 				efx_ef10_filter_get_unsafe_id(rc);
5306 		}
5307 	}
5308 
5309 	return 0;
5310 }
5311 
5312 static int efx_ef10_filter_insert_def(struct efx_nic *efx,
5313 				      struct efx_ef10_filter_vlan *vlan,
5314 				      enum efx_encap_type encap_type,
5315 				      bool multicast, bool rollback)
5316 {
5317 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
5318 	enum efx_filter_flags filter_flags;
5319 	struct efx_filter_spec spec;
5320 	u8 baddr[ETH_ALEN];
5321 	int rc;
5322 	u16 *id;
5323 
5324 	filter_flags = efx_rss_enabled(efx) ? EFX_FILTER_FLAG_RX_RSS : 0;
5325 
5326 	efx_filter_init_rx(&spec, EFX_FILTER_PRI_AUTO, filter_flags, 0);
5327 
5328 	if (multicast)
5329 		efx_filter_set_mc_def(&spec);
5330 	else
5331 		efx_filter_set_uc_def(&spec);
5332 
5333 	if (encap_type) {
5334 		if (nic_data->datapath_caps &
5335 		    (1 << MC_CMD_GET_CAPABILITIES_OUT_VXLAN_NVGRE_LBN))
5336 			efx_filter_set_encap_type(&spec, encap_type);
5337 		else
5338 			/* don't insert encap filters on non-supporting
5339 			 * platforms. ID will be left as INVALID.
5340 			 */
5341 			return 0;
5342 	}
5343 
5344 	if (vlan->vid != EFX_FILTER_VID_UNSPEC)
5345 		efx_filter_set_eth_local(&spec, vlan->vid, NULL);
5346 
5347 	rc = efx_ef10_filter_insert(efx, &spec, true);
5348 	if (rc < 0) {
5349 		const char *um = multicast ? "Multicast" : "Unicast";
5350 		const char *encap_name = "";
5351 		const char *encap_ipv = "";
5352 
5353 		if ((encap_type & EFX_ENCAP_TYPES_MASK) ==
5354 		    EFX_ENCAP_TYPE_VXLAN)
5355 			encap_name = "VXLAN ";
5356 		else if ((encap_type & EFX_ENCAP_TYPES_MASK) ==
5357 			 EFX_ENCAP_TYPE_NVGRE)
5358 			encap_name = "NVGRE ";
5359 		else if ((encap_type & EFX_ENCAP_TYPES_MASK) ==
5360 			 EFX_ENCAP_TYPE_GENEVE)
5361 			encap_name = "GENEVE ";
5362 		if (encap_type & EFX_ENCAP_FLAG_IPV6)
5363 			encap_ipv = "IPv6 ";
5364 		else if (encap_type)
5365 			encap_ipv = "IPv4 ";
5366 
5367 		/* unprivileged functions can't insert mismatch filters
5368 		 * for encapsulated or unicast traffic, so downgrade
5369 		 * those warnings to debug.
5370 		 */
5371 		netif_cond_dbg(efx, drv, efx->net_dev,
5372 			       rc == -EPERM && (encap_type || !multicast), warn,
5373 			       "%s%s%s mismatch filter insert failed rc=%d\n",
5374 			       encap_name, encap_ipv, um, rc);
5375 	} else if (multicast) {
5376 		/* mapping from encap types to default filter IDs (multicast) */
5377 		static enum efx_ef10_default_filters map[] = {
5378 			[EFX_ENCAP_TYPE_NONE] = EFX_EF10_MCDEF,
5379 			[EFX_ENCAP_TYPE_VXLAN] = EFX_EF10_VXLAN4_MCDEF,
5380 			[EFX_ENCAP_TYPE_NVGRE] = EFX_EF10_NVGRE4_MCDEF,
5381 			[EFX_ENCAP_TYPE_GENEVE] = EFX_EF10_GENEVE4_MCDEF,
5382 			[EFX_ENCAP_TYPE_VXLAN | EFX_ENCAP_FLAG_IPV6] =
5383 				EFX_EF10_VXLAN6_MCDEF,
5384 			[EFX_ENCAP_TYPE_NVGRE | EFX_ENCAP_FLAG_IPV6] =
5385 				EFX_EF10_NVGRE6_MCDEF,
5386 			[EFX_ENCAP_TYPE_GENEVE | EFX_ENCAP_FLAG_IPV6] =
5387 				EFX_EF10_GENEVE6_MCDEF,
5388 		};
5389 
5390 		/* quick bounds check (BCAST result impossible) */
5391 		BUILD_BUG_ON(EFX_EF10_BCAST != 0);
5392 		if (encap_type >= ARRAY_SIZE(map) || map[encap_type] == 0) {
5393 			WARN_ON(1);
5394 			return -EINVAL;
5395 		}
5396 		/* then follow map */
5397 		id = &vlan->default_filters[map[encap_type]];
5398 
5399 		EFX_WARN_ON_PARANOID(*id != EFX_EF10_FILTER_ID_INVALID);
5400 		*id = efx_ef10_filter_get_unsafe_id(rc);
5401 		if (!nic_data->workaround_26807 && !encap_type) {
5402 			/* Also need an Ethernet broadcast filter */
5403 			efx_filter_init_rx(&spec, EFX_FILTER_PRI_AUTO,
5404 					   filter_flags, 0);
5405 			eth_broadcast_addr(baddr);
5406 			efx_filter_set_eth_local(&spec, vlan->vid, baddr);
5407 			rc = efx_ef10_filter_insert(efx, &spec, true);
5408 			if (rc < 0) {
5409 				netif_warn(efx, drv, efx->net_dev,
5410 					   "Broadcast filter insert failed rc=%d\n",
5411 					   rc);
5412 				if (rollback) {
5413 					/* Roll back the mc_def filter */
5414 					efx_ef10_filter_remove_unsafe(
5415 							efx, EFX_FILTER_PRI_AUTO,
5416 							*id);
5417 					*id = EFX_EF10_FILTER_ID_INVALID;
5418 					return rc;
5419 				}
5420 			} else {
5421 				EFX_WARN_ON_PARANOID(
5422 					vlan->default_filters[EFX_EF10_BCAST] !=
5423 					EFX_EF10_FILTER_ID_INVALID);
5424 				vlan->default_filters[EFX_EF10_BCAST] =
5425 					efx_ef10_filter_get_unsafe_id(rc);
5426 			}
5427 		}
5428 		rc = 0;
5429 	} else {
5430 		/* mapping from encap types to default filter IDs (unicast) */
5431 		static enum efx_ef10_default_filters map[] = {
5432 			[EFX_ENCAP_TYPE_NONE] = EFX_EF10_UCDEF,
5433 			[EFX_ENCAP_TYPE_VXLAN] = EFX_EF10_VXLAN4_UCDEF,
5434 			[EFX_ENCAP_TYPE_NVGRE] = EFX_EF10_NVGRE4_UCDEF,
5435 			[EFX_ENCAP_TYPE_GENEVE] = EFX_EF10_GENEVE4_UCDEF,
5436 			[EFX_ENCAP_TYPE_VXLAN | EFX_ENCAP_FLAG_IPV6] =
5437 				EFX_EF10_VXLAN6_UCDEF,
5438 			[EFX_ENCAP_TYPE_NVGRE | EFX_ENCAP_FLAG_IPV6] =
5439 				EFX_EF10_NVGRE6_UCDEF,
5440 			[EFX_ENCAP_TYPE_GENEVE | EFX_ENCAP_FLAG_IPV6] =
5441 				EFX_EF10_GENEVE6_UCDEF,
5442 		};
5443 
5444 		/* quick bounds check (BCAST result impossible) */
5445 		BUILD_BUG_ON(EFX_EF10_BCAST != 0);
5446 		if (encap_type >= ARRAY_SIZE(map) || map[encap_type] == 0) {
5447 			WARN_ON(1);
5448 			return -EINVAL;
5449 		}
5450 		/* then follow map */
5451 		id = &vlan->default_filters[map[encap_type]];
5452 		EFX_WARN_ON_PARANOID(*id != EFX_EF10_FILTER_ID_INVALID);
5453 		*id = rc;
5454 		rc = 0;
5455 	}
5456 	return rc;
5457 }
5458 
5459 /* Remove filters that weren't renewed.  Since nothing else changes the AUTO_OLD
5460  * flag or removes these filters, we don't need to hold the filter_lock while
5461  * scanning for these filters.
5462  */
5463 static void efx_ef10_filter_remove_old(struct efx_nic *efx)
5464 {
5465 	struct efx_ef10_filter_table *table = efx->filter_state;
5466 	int remove_failed = 0;
5467 	int remove_noent = 0;
5468 	int rc;
5469 	int i;
5470 
5471 	for (i = 0; i < HUNT_FILTER_TBL_ROWS; i++) {
5472 		if (READ_ONCE(table->entry[i].spec) &
5473 		    EFX_EF10_FILTER_FLAG_AUTO_OLD) {
5474 			rc = efx_ef10_filter_remove_internal(efx,
5475 					1U << EFX_FILTER_PRI_AUTO, i, true);
5476 			if (rc == -ENOENT)
5477 				remove_noent++;
5478 			else if (rc)
5479 				remove_failed++;
5480 		}
5481 	}
5482 
5483 	if (remove_failed)
5484 		netif_info(efx, drv, efx->net_dev,
5485 			   "%s: failed to remove %d filters\n",
5486 			   __func__, remove_failed);
5487 	if (remove_noent)
5488 		netif_info(efx, drv, efx->net_dev,
5489 			   "%s: failed to remove %d non-existent filters\n",
5490 			   __func__, remove_noent);
5491 }
5492 
5493 static int efx_ef10_vport_set_mac_address(struct efx_nic *efx)
5494 {
5495 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
5496 	u8 mac_old[ETH_ALEN];
5497 	int rc, rc2;
5498 
5499 	/* Only reconfigure a PF-created vport */
5500 	if (is_zero_ether_addr(nic_data->vport_mac))
5501 		return 0;
5502 
5503 	efx_device_detach_sync(efx);
5504 	efx_net_stop(efx->net_dev);
5505 	down_write(&efx->filter_sem);
5506 	efx_ef10_filter_table_remove(efx);
5507 	up_write(&efx->filter_sem);
5508 
5509 	rc = efx_ef10_vadaptor_free(efx, nic_data->vport_id);
5510 	if (rc)
5511 		goto restore_filters;
5512 
5513 	ether_addr_copy(mac_old, nic_data->vport_mac);
5514 	rc = efx_ef10_vport_del_mac(efx, nic_data->vport_id,
5515 				    nic_data->vport_mac);
5516 	if (rc)
5517 		goto restore_vadaptor;
5518 
5519 	rc = efx_ef10_vport_add_mac(efx, nic_data->vport_id,
5520 				    efx->net_dev->dev_addr);
5521 	if (!rc) {
5522 		ether_addr_copy(nic_data->vport_mac, efx->net_dev->dev_addr);
5523 	} else {
5524 		rc2 = efx_ef10_vport_add_mac(efx, nic_data->vport_id, mac_old);
5525 		if (rc2) {
5526 			/* Failed to add original MAC, so clear vport_mac */
5527 			eth_zero_addr(nic_data->vport_mac);
5528 			goto reset_nic;
5529 		}
5530 	}
5531 
5532 restore_vadaptor:
5533 	rc2 = efx_ef10_vadaptor_alloc(efx, nic_data->vport_id);
5534 	if (rc2)
5535 		goto reset_nic;
5536 restore_filters:
5537 	down_write(&efx->filter_sem);
5538 	rc2 = efx_ef10_filter_table_probe(efx);
5539 	up_write(&efx->filter_sem);
5540 	if (rc2)
5541 		goto reset_nic;
5542 
5543 	rc2 = efx_net_open(efx->net_dev);
5544 	if (rc2)
5545 		goto reset_nic;
5546 
5547 	efx_device_attach_if_not_resetting(efx);
5548 
5549 	return rc;
5550 
5551 reset_nic:
5552 	netif_err(efx, drv, efx->net_dev,
5553 		  "Failed to restore when changing MAC address - scheduling reset\n");
5554 	efx_schedule_reset(efx, RESET_TYPE_DATAPATH);
5555 
5556 	return rc ? rc : rc2;
5557 }
5558 
5559 /* Caller must hold efx->filter_sem for read if race against
5560  * efx_ef10_filter_table_remove() is possible
5561  */
5562 static void efx_ef10_filter_vlan_sync_rx_mode(struct efx_nic *efx,
5563 					      struct efx_ef10_filter_vlan *vlan)
5564 {
5565 	struct efx_ef10_filter_table *table = efx->filter_state;
5566 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
5567 
5568 	/* Do not install unspecified VID if VLAN filtering is enabled.
5569 	 * Do not install all specified VIDs if VLAN filtering is disabled.
5570 	 */
5571 	if ((vlan->vid == EFX_FILTER_VID_UNSPEC) == table->vlan_filter)
5572 		return;
5573 
5574 	/* Insert/renew unicast filters */
5575 	if (table->uc_promisc) {
5576 		efx_ef10_filter_insert_def(efx, vlan, EFX_ENCAP_TYPE_NONE,
5577 					   false, false);
5578 		efx_ef10_filter_insert_addr_list(efx, vlan, false, false);
5579 	} else {
5580 		/* If any of the filters failed to insert, fall back to
5581 		 * promiscuous mode - add in the uc_def filter.  But keep
5582 		 * our individual unicast filters.
5583 		 */
5584 		if (efx_ef10_filter_insert_addr_list(efx, vlan, false, false))
5585 			efx_ef10_filter_insert_def(efx, vlan,
5586 						   EFX_ENCAP_TYPE_NONE,
5587 						   false, false);
5588 	}
5589 	efx_ef10_filter_insert_def(efx, vlan, EFX_ENCAP_TYPE_VXLAN,
5590 				   false, false);
5591 	efx_ef10_filter_insert_def(efx, vlan, EFX_ENCAP_TYPE_VXLAN |
5592 					      EFX_ENCAP_FLAG_IPV6,
5593 				   false, false);
5594 	efx_ef10_filter_insert_def(efx, vlan, EFX_ENCAP_TYPE_NVGRE,
5595 				   false, false);
5596 	efx_ef10_filter_insert_def(efx, vlan, EFX_ENCAP_TYPE_NVGRE |
5597 					      EFX_ENCAP_FLAG_IPV6,
5598 				   false, false);
5599 	efx_ef10_filter_insert_def(efx, vlan, EFX_ENCAP_TYPE_GENEVE,
5600 				   false, false);
5601 	efx_ef10_filter_insert_def(efx, vlan, EFX_ENCAP_TYPE_GENEVE |
5602 					      EFX_ENCAP_FLAG_IPV6,
5603 				   false, false);
5604 
5605 	/* Insert/renew multicast filters */
5606 	/* If changing promiscuous state with cascaded multicast filters, remove
5607 	 * old filters first, so that packets are dropped rather than duplicated
5608 	 */
5609 	if (nic_data->workaround_26807 &&
5610 	    table->mc_promisc_last != table->mc_promisc)
5611 		efx_ef10_filter_remove_old(efx);
5612 	if (table->mc_promisc) {
5613 		if (nic_data->workaround_26807) {
5614 			/* If we failed to insert promiscuous filters, rollback
5615 			 * and fall back to individual multicast filters
5616 			 */
5617 			if (efx_ef10_filter_insert_def(efx, vlan,
5618 						       EFX_ENCAP_TYPE_NONE,
5619 						       true, true)) {
5620 				/* Changing promisc state, so remove old filters */
5621 				efx_ef10_filter_remove_old(efx);
5622 				efx_ef10_filter_insert_addr_list(efx, vlan,
5623 								 true, false);
5624 			}
5625 		} else {
5626 			/* If we failed to insert promiscuous filters, don't
5627 			 * rollback.  Regardless, also insert the mc_list,
5628 			 * unless it's incomplete due to overflow
5629 			 */
5630 			efx_ef10_filter_insert_def(efx, vlan,
5631 						   EFX_ENCAP_TYPE_NONE,
5632 						   true, false);
5633 			if (!table->mc_overflow)
5634 				efx_ef10_filter_insert_addr_list(efx, vlan,
5635 								 true, false);
5636 		}
5637 	} else {
5638 		/* If any filters failed to insert, rollback and fall back to
5639 		 * promiscuous mode - mc_def filter and maybe broadcast.  If
5640 		 * that fails, roll back again and insert as many of our
5641 		 * individual multicast filters as we can.
5642 		 */
5643 		if (efx_ef10_filter_insert_addr_list(efx, vlan, true, true)) {
5644 			/* Changing promisc state, so remove old filters */
5645 			if (nic_data->workaround_26807)
5646 				efx_ef10_filter_remove_old(efx);
5647 			if (efx_ef10_filter_insert_def(efx, vlan,
5648 						       EFX_ENCAP_TYPE_NONE,
5649 						       true, true))
5650 				efx_ef10_filter_insert_addr_list(efx, vlan,
5651 								 true, false);
5652 		}
5653 	}
5654 	efx_ef10_filter_insert_def(efx, vlan, EFX_ENCAP_TYPE_VXLAN,
5655 				   true, false);
5656 	efx_ef10_filter_insert_def(efx, vlan, EFX_ENCAP_TYPE_VXLAN |
5657 					      EFX_ENCAP_FLAG_IPV6,
5658 				   true, false);
5659 	efx_ef10_filter_insert_def(efx, vlan, EFX_ENCAP_TYPE_NVGRE,
5660 				   true, false);
5661 	efx_ef10_filter_insert_def(efx, vlan, EFX_ENCAP_TYPE_NVGRE |
5662 					      EFX_ENCAP_FLAG_IPV6,
5663 				   true, false);
5664 	efx_ef10_filter_insert_def(efx, vlan, EFX_ENCAP_TYPE_GENEVE,
5665 				   true, false);
5666 	efx_ef10_filter_insert_def(efx, vlan, EFX_ENCAP_TYPE_GENEVE |
5667 					      EFX_ENCAP_FLAG_IPV6,
5668 				   true, false);
5669 }
5670 
5671 /* Caller must hold efx->filter_sem for read if race against
5672  * efx_ef10_filter_table_remove() is possible
5673  */
5674 static void efx_ef10_filter_sync_rx_mode(struct efx_nic *efx)
5675 {
5676 	struct efx_ef10_filter_table *table = efx->filter_state;
5677 	struct net_device *net_dev = efx->net_dev;
5678 	struct efx_ef10_filter_vlan *vlan;
5679 	bool vlan_filter;
5680 
5681 	if (!efx_dev_registered(efx))
5682 		return;
5683 
5684 	if (!table)
5685 		return;
5686 
5687 	efx_ef10_filter_mark_old(efx);
5688 
5689 	/* Copy/convert the address lists; add the primary station
5690 	 * address and broadcast address
5691 	 */
5692 	netif_addr_lock_bh(net_dev);
5693 	efx_ef10_filter_uc_addr_list(efx);
5694 	efx_ef10_filter_mc_addr_list(efx);
5695 	netif_addr_unlock_bh(net_dev);
5696 
5697 	/* If VLAN filtering changes, all old filters are finally removed.
5698 	 * Do it in advance to avoid conflicts for unicast untagged and
5699 	 * VLAN 0 tagged filters.
5700 	 */
5701 	vlan_filter = !!(net_dev->features & NETIF_F_HW_VLAN_CTAG_FILTER);
5702 	if (table->vlan_filter != vlan_filter) {
5703 		table->vlan_filter = vlan_filter;
5704 		efx_ef10_filter_remove_old(efx);
5705 	}
5706 
5707 	list_for_each_entry(vlan, &table->vlan_list, list)
5708 		efx_ef10_filter_vlan_sync_rx_mode(efx, vlan);
5709 
5710 	efx_ef10_filter_remove_old(efx);
5711 	table->mc_promisc_last = table->mc_promisc;
5712 }
5713 
5714 static struct efx_ef10_filter_vlan *efx_ef10_filter_find_vlan(struct efx_nic *efx, u16 vid)
5715 {
5716 	struct efx_ef10_filter_table *table = efx->filter_state;
5717 	struct efx_ef10_filter_vlan *vlan;
5718 
5719 	WARN_ON(!rwsem_is_locked(&efx->filter_sem));
5720 
5721 	list_for_each_entry(vlan, &table->vlan_list, list) {
5722 		if (vlan->vid == vid)
5723 			return vlan;
5724 	}
5725 
5726 	return NULL;
5727 }
5728 
5729 static int efx_ef10_filter_add_vlan(struct efx_nic *efx, u16 vid)
5730 {
5731 	struct efx_ef10_filter_table *table = efx->filter_state;
5732 	struct efx_ef10_filter_vlan *vlan;
5733 	unsigned int i;
5734 
5735 	if (!efx_rwsem_assert_write_locked(&efx->filter_sem))
5736 		return -EINVAL;
5737 
5738 	vlan = efx_ef10_filter_find_vlan(efx, vid);
5739 	if (WARN_ON(vlan)) {
5740 		netif_err(efx, drv, efx->net_dev,
5741 			  "VLAN %u already added\n", vid);
5742 		return -EALREADY;
5743 	}
5744 
5745 	vlan = kzalloc(sizeof(*vlan), GFP_KERNEL);
5746 	if (!vlan)
5747 		return -ENOMEM;
5748 
5749 	vlan->vid = vid;
5750 
5751 	for (i = 0; i < ARRAY_SIZE(vlan->uc); i++)
5752 		vlan->uc[i] = EFX_EF10_FILTER_ID_INVALID;
5753 	for (i = 0; i < ARRAY_SIZE(vlan->mc); i++)
5754 		vlan->mc[i] = EFX_EF10_FILTER_ID_INVALID;
5755 	for (i = 0; i < EFX_EF10_NUM_DEFAULT_FILTERS; i++)
5756 		vlan->default_filters[i] = EFX_EF10_FILTER_ID_INVALID;
5757 
5758 	list_add_tail(&vlan->list, &table->vlan_list);
5759 
5760 	if (efx_dev_registered(efx))
5761 		efx_ef10_filter_vlan_sync_rx_mode(efx, vlan);
5762 
5763 	return 0;
5764 }
5765 
5766 static void efx_ef10_filter_del_vlan_internal(struct efx_nic *efx,
5767 					      struct efx_ef10_filter_vlan *vlan)
5768 {
5769 	unsigned int i;
5770 
5771 	/* See comment in efx_ef10_filter_table_remove() */
5772 	if (!efx_rwsem_assert_write_locked(&efx->filter_sem))
5773 		return;
5774 
5775 	list_del(&vlan->list);
5776 
5777 	for (i = 0; i < ARRAY_SIZE(vlan->uc); i++)
5778 		efx_ef10_filter_remove_unsafe(efx, EFX_FILTER_PRI_AUTO,
5779 					      vlan->uc[i]);
5780 	for (i = 0; i < ARRAY_SIZE(vlan->mc); i++)
5781 		efx_ef10_filter_remove_unsafe(efx, EFX_FILTER_PRI_AUTO,
5782 					      vlan->mc[i]);
5783 	for (i = 0; i < EFX_EF10_NUM_DEFAULT_FILTERS; i++)
5784 		if (vlan->default_filters[i] != EFX_EF10_FILTER_ID_INVALID)
5785 			efx_ef10_filter_remove_unsafe(efx, EFX_FILTER_PRI_AUTO,
5786 						      vlan->default_filters[i]);
5787 
5788 	kfree(vlan);
5789 }
5790 
5791 static void efx_ef10_filter_del_vlan(struct efx_nic *efx, u16 vid)
5792 {
5793 	struct efx_ef10_filter_vlan *vlan;
5794 
5795 	/* See comment in efx_ef10_filter_table_remove() */
5796 	if (!efx_rwsem_assert_write_locked(&efx->filter_sem))
5797 		return;
5798 
5799 	vlan = efx_ef10_filter_find_vlan(efx, vid);
5800 	if (!vlan) {
5801 		netif_err(efx, drv, efx->net_dev,
5802 			  "VLAN %u not found in filter state\n", vid);
5803 		return;
5804 	}
5805 
5806 	efx_ef10_filter_del_vlan_internal(efx, vlan);
5807 }
5808 
5809 static int efx_ef10_set_mac_address(struct efx_nic *efx)
5810 {
5811 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VADAPTOR_SET_MAC_IN_LEN);
5812 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
5813 	bool was_enabled = efx->port_enabled;
5814 	int rc;
5815 
5816 	efx_device_detach_sync(efx);
5817 	efx_net_stop(efx->net_dev);
5818 
5819 	mutex_lock(&efx->mac_lock);
5820 	down_write(&efx->filter_sem);
5821 	efx_ef10_filter_table_remove(efx);
5822 
5823 	ether_addr_copy(MCDI_PTR(inbuf, VADAPTOR_SET_MAC_IN_MACADDR),
5824 			efx->net_dev->dev_addr);
5825 	MCDI_SET_DWORD(inbuf, VADAPTOR_SET_MAC_IN_UPSTREAM_PORT_ID,
5826 		       nic_data->vport_id);
5827 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_VADAPTOR_SET_MAC, inbuf,
5828 				sizeof(inbuf), NULL, 0, NULL);
5829 
5830 	efx_ef10_filter_table_probe(efx);
5831 	up_write(&efx->filter_sem);
5832 	mutex_unlock(&efx->mac_lock);
5833 
5834 	if (was_enabled)
5835 		efx_net_open(efx->net_dev);
5836 	efx_device_attach_if_not_resetting(efx);
5837 
5838 #ifdef CONFIG_SFC_SRIOV
5839 	if (efx->pci_dev->is_virtfn && efx->pci_dev->physfn) {
5840 		struct pci_dev *pci_dev_pf = efx->pci_dev->physfn;
5841 
5842 		if (rc == -EPERM) {
5843 			struct efx_nic *efx_pf;
5844 
5845 			/* Switch to PF and change MAC address on vport */
5846 			efx_pf = pci_get_drvdata(pci_dev_pf);
5847 
5848 			rc = efx_ef10_sriov_set_vf_mac(efx_pf,
5849 						       nic_data->vf_index,
5850 						       efx->net_dev->dev_addr);
5851 		} else if (!rc) {
5852 			struct efx_nic *efx_pf = pci_get_drvdata(pci_dev_pf);
5853 			struct efx_ef10_nic_data *nic_data = efx_pf->nic_data;
5854 			unsigned int i;
5855 
5856 			/* MAC address successfully changed by VF (with MAC
5857 			 * spoofing) so update the parent PF if possible.
5858 			 */
5859 			for (i = 0; i < efx_pf->vf_count; ++i) {
5860 				struct ef10_vf *vf = nic_data->vf + i;
5861 
5862 				if (vf->efx == efx) {
5863 					ether_addr_copy(vf->mac,
5864 							efx->net_dev->dev_addr);
5865 					return 0;
5866 				}
5867 			}
5868 		}
5869 	} else
5870 #endif
5871 	if (rc == -EPERM) {
5872 		netif_err(efx, drv, efx->net_dev,
5873 			  "Cannot change MAC address; use sfboot to enable"
5874 			  " mac-spoofing on this interface\n");
5875 	} else if (rc == -ENOSYS && !efx_ef10_is_vf(efx)) {
5876 		/* If the active MCFW does not support MC_CMD_VADAPTOR_SET_MAC
5877 		 * fall-back to the method of changing the MAC address on the
5878 		 * vport.  This only applies to PFs because such versions of
5879 		 * MCFW do not support VFs.
5880 		 */
5881 		rc = efx_ef10_vport_set_mac_address(efx);
5882 	} else if (rc) {
5883 		efx_mcdi_display_error(efx, MC_CMD_VADAPTOR_SET_MAC,
5884 				       sizeof(inbuf), NULL, 0, rc);
5885 	}
5886 
5887 	return rc;
5888 }
5889 
5890 static int efx_ef10_mac_reconfigure(struct efx_nic *efx)
5891 {
5892 	efx_ef10_filter_sync_rx_mode(efx);
5893 
5894 	return efx_mcdi_set_mac(efx);
5895 }
5896 
5897 static int efx_ef10_mac_reconfigure_vf(struct efx_nic *efx)
5898 {
5899 	efx_ef10_filter_sync_rx_mode(efx);
5900 
5901 	return 0;
5902 }
5903 
5904 static int efx_ef10_start_bist(struct efx_nic *efx, u32 bist_type)
5905 {
5906 	MCDI_DECLARE_BUF(inbuf, MC_CMD_START_BIST_IN_LEN);
5907 
5908 	MCDI_SET_DWORD(inbuf, START_BIST_IN_TYPE, bist_type);
5909 	return efx_mcdi_rpc(efx, MC_CMD_START_BIST, inbuf, sizeof(inbuf),
5910 			    NULL, 0, NULL);
5911 }
5912 
5913 /* MC BISTs follow a different poll mechanism to phy BISTs.
5914  * The BIST is done in the poll handler on the MC, and the MCDI command
5915  * will block until the BIST is done.
5916  */
5917 static int efx_ef10_poll_bist(struct efx_nic *efx)
5918 {
5919 	int rc;
5920 	MCDI_DECLARE_BUF(outbuf, MC_CMD_POLL_BIST_OUT_LEN);
5921 	size_t outlen;
5922 	u32 result;
5923 
5924 	rc = efx_mcdi_rpc(efx, MC_CMD_POLL_BIST, NULL, 0,
5925 			   outbuf, sizeof(outbuf), &outlen);
5926 	if (rc != 0)
5927 		return rc;
5928 
5929 	if (outlen < MC_CMD_POLL_BIST_OUT_LEN)
5930 		return -EIO;
5931 
5932 	result = MCDI_DWORD(outbuf, POLL_BIST_OUT_RESULT);
5933 	switch (result) {
5934 	case MC_CMD_POLL_BIST_PASSED:
5935 		netif_dbg(efx, hw, efx->net_dev, "BIST passed.\n");
5936 		return 0;
5937 	case MC_CMD_POLL_BIST_TIMEOUT:
5938 		netif_err(efx, hw, efx->net_dev, "BIST timed out\n");
5939 		return -EIO;
5940 	case MC_CMD_POLL_BIST_FAILED:
5941 		netif_err(efx, hw, efx->net_dev, "BIST failed.\n");
5942 		return -EIO;
5943 	default:
5944 		netif_err(efx, hw, efx->net_dev,
5945 			  "BIST returned unknown result %u", result);
5946 		return -EIO;
5947 	}
5948 }
5949 
5950 static int efx_ef10_run_bist(struct efx_nic *efx, u32 bist_type)
5951 {
5952 	int rc;
5953 
5954 	netif_dbg(efx, drv, efx->net_dev, "starting BIST type %u\n", bist_type);
5955 
5956 	rc = efx_ef10_start_bist(efx, bist_type);
5957 	if (rc != 0)
5958 		return rc;
5959 
5960 	return efx_ef10_poll_bist(efx);
5961 }
5962 
5963 static int
5964 efx_ef10_test_chip(struct efx_nic *efx, struct efx_self_tests *tests)
5965 {
5966 	int rc, rc2;
5967 
5968 	efx_reset_down(efx, RESET_TYPE_WORLD);
5969 
5970 	rc = efx_mcdi_rpc(efx, MC_CMD_ENABLE_OFFLINE_BIST,
5971 			  NULL, 0, NULL, 0, NULL);
5972 	if (rc != 0)
5973 		goto out;
5974 
5975 	tests->memory = efx_ef10_run_bist(efx, MC_CMD_MC_MEM_BIST) ? -1 : 1;
5976 	tests->registers = efx_ef10_run_bist(efx, MC_CMD_REG_BIST) ? -1 : 1;
5977 
5978 	rc = efx_mcdi_reset(efx, RESET_TYPE_WORLD);
5979 
5980 out:
5981 	if (rc == -EPERM)
5982 		rc = 0;
5983 	rc2 = efx_reset_up(efx, RESET_TYPE_WORLD, rc == 0);
5984 	return rc ? rc : rc2;
5985 }
5986 
5987 #ifdef CONFIG_SFC_MTD
5988 
5989 struct efx_ef10_nvram_type_info {
5990 	u16 type, type_mask;
5991 	u8 port;
5992 	const char *name;
5993 };
5994 
5995 static const struct efx_ef10_nvram_type_info efx_ef10_nvram_types[] = {
5996 	{ NVRAM_PARTITION_TYPE_MC_FIRMWARE,	   0,    0, "sfc_mcfw" },
5997 	{ NVRAM_PARTITION_TYPE_MC_FIRMWARE_BACKUP, 0,    0, "sfc_mcfw_backup" },
5998 	{ NVRAM_PARTITION_TYPE_EXPANSION_ROM,	   0,    0, "sfc_exp_rom" },
5999 	{ NVRAM_PARTITION_TYPE_STATIC_CONFIG,	   0,    0, "sfc_static_cfg" },
6000 	{ NVRAM_PARTITION_TYPE_DYNAMIC_CONFIG,	   0,    0, "sfc_dynamic_cfg" },
6001 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT0, 0,   0, "sfc_exp_rom_cfg" },
6002 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT1, 0,   1, "sfc_exp_rom_cfg" },
6003 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT2, 0,   2, "sfc_exp_rom_cfg" },
6004 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT3, 0,   3, "sfc_exp_rom_cfg" },
6005 	{ NVRAM_PARTITION_TYPE_LICENSE,		   0,    0, "sfc_license" },
6006 	{ NVRAM_PARTITION_TYPE_PHY_MIN,		   0xff, 0, "sfc_phy_fw" },
6007 };
6008 
6009 static int efx_ef10_mtd_probe_partition(struct efx_nic *efx,
6010 					struct efx_mcdi_mtd_partition *part,
6011 					unsigned int type)
6012 {
6013 	MCDI_DECLARE_BUF(inbuf, MC_CMD_NVRAM_METADATA_IN_LEN);
6014 	MCDI_DECLARE_BUF(outbuf, MC_CMD_NVRAM_METADATA_OUT_LENMAX);
6015 	const struct efx_ef10_nvram_type_info *info;
6016 	size_t size, erase_size, outlen;
6017 	bool protected;
6018 	int rc;
6019 
6020 	for (info = efx_ef10_nvram_types; ; info++) {
6021 		if (info ==
6022 		    efx_ef10_nvram_types + ARRAY_SIZE(efx_ef10_nvram_types))
6023 			return -ENODEV;
6024 		if ((type & ~info->type_mask) == info->type)
6025 			break;
6026 	}
6027 	if (info->port != efx_port_num(efx))
6028 		return -ENODEV;
6029 
6030 	rc = efx_mcdi_nvram_info(efx, type, &size, &erase_size, &protected);
6031 	if (rc)
6032 		return rc;
6033 	if (protected)
6034 		return -ENODEV; /* hide it */
6035 
6036 	part->nvram_type = type;
6037 
6038 	MCDI_SET_DWORD(inbuf, NVRAM_METADATA_IN_TYPE, type);
6039 	rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_METADATA, inbuf, sizeof(inbuf),
6040 			  outbuf, sizeof(outbuf), &outlen);
6041 	if (rc)
6042 		return rc;
6043 	if (outlen < MC_CMD_NVRAM_METADATA_OUT_LENMIN)
6044 		return -EIO;
6045 	if (MCDI_DWORD(outbuf, NVRAM_METADATA_OUT_FLAGS) &
6046 	    (1 << MC_CMD_NVRAM_METADATA_OUT_SUBTYPE_VALID_LBN))
6047 		part->fw_subtype = MCDI_DWORD(outbuf,
6048 					      NVRAM_METADATA_OUT_SUBTYPE);
6049 
6050 	part->common.dev_type_name = "EF10 NVRAM manager";
6051 	part->common.type_name = info->name;
6052 
6053 	part->common.mtd.type = MTD_NORFLASH;
6054 	part->common.mtd.flags = MTD_CAP_NORFLASH;
6055 	part->common.mtd.size = size;
6056 	part->common.mtd.erasesize = erase_size;
6057 
6058 	return 0;
6059 }
6060 
6061 static int efx_ef10_mtd_probe(struct efx_nic *efx)
6062 {
6063 	MCDI_DECLARE_BUF(outbuf, MC_CMD_NVRAM_PARTITIONS_OUT_LENMAX);
6064 	struct efx_mcdi_mtd_partition *parts;
6065 	size_t outlen, n_parts_total, i, n_parts;
6066 	unsigned int type;
6067 	int rc;
6068 
6069 	ASSERT_RTNL();
6070 
6071 	BUILD_BUG_ON(MC_CMD_NVRAM_PARTITIONS_IN_LEN != 0);
6072 	rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_PARTITIONS, NULL, 0,
6073 			  outbuf, sizeof(outbuf), &outlen);
6074 	if (rc)
6075 		return rc;
6076 	if (outlen < MC_CMD_NVRAM_PARTITIONS_OUT_LENMIN)
6077 		return -EIO;
6078 
6079 	n_parts_total = MCDI_DWORD(outbuf, NVRAM_PARTITIONS_OUT_NUM_PARTITIONS);
6080 	if (n_parts_total >
6081 	    MCDI_VAR_ARRAY_LEN(outlen, NVRAM_PARTITIONS_OUT_TYPE_ID))
6082 		return -EIO;
6083 
6084 	parts = kcalloc(n_parts_total, sizeof(*parts), GFP_KERNEL);
6085 	if (!parts)
6086 		return -ENOMEM;
6087 
6088 	n_parts = 0;
6089 	for (i = 0; i < n_parts_total; i++) {
6090 		type = MCDI_ARRAY_DWORD(outbuf, NVRAM_PARTITIONS_OUT_TYPE_ID,
6091 					i);
6092 		rc = efx_ef10_mtd_probe_partition(efx, &parts[n_parts], type);
6093 		if (rc == 0)
6094 			n_parts++;
6095 		else if (rc != -ENODEV)
6096 			goto fail;
6097 	}
6098 
6099 	rc = efx_mtd_add(efx, &parts[0].common, n_parts, sizeof(*parts));
6100 fail:
6101 	if (rc)
6102 		kfree(parts);
6103 	return rc;
6104 }
6105 
6106 #endif /* CONFIG_SFC_MTD */
6107 
6108 static void efx_ef10_ptp_write_host_time(struct efx_nic *efx, u32 host_time)
6109 {
6110 	_efx_writed(efx, cpu_to_le32(host_time), ER_DZ_MC_DB_LWRD);
6111 }
6112 
6113 static void efx_ef10_ptp_write_host_time_vf(struct efx_nic *efx,
6114 					    u32 host_time) {}
6115 
6116 static int efx_ef10_rx_enable_timestamping(struct efx_channel *channel,
6117 					   bool temp)
6118 {
6119 	MCDI_DECLARE_BUF(inbuf, MC_CMD_PTP_IN_TIME_EVENT_SUBSCRIBE_LEN);
6120 	int rc;
6121 
6122 	if (channel->sync_events_state == SYNC_EVENTS_REQUESTED ||
6123 	    channel->sync_events_state == SYNC_EVENTS_VALID ||
6124 	    (temp && channel->sync_events_state == SYNC_EVENTS_DISABLED))
6125 		return 0;
6126 	channel->sync_events_state = SYNC_EVENTS_REQUESTED;
6127 
6128 	MCDI_SET_DWORD(inbuf, PTP_IN_OP, MC_CMD_PTP_OP_TIME_EVENT_SUBSCRIBE);
6129 	MCDI_SET_DWORD(inbuf, PTP_IN_PERIPH_ID, 0);
6130 	MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_SUBSCRIBE_QUEUE,
6131 		       channel->channel);
6132 
6133 	rc = efx_mcdi_rpc(channel->efx, MC_CMD_PTP,
6134 			  inbuf, sizeof(inbuf), NULL, 0, NULL);
6135 
6136 	if (rc != 0)
6137 		channel->sync_events_state = temp ? SYNC_EVENTS_QUIESCENT :
6138 						    SYNC_EVENTS_DISABLED;
6139 
6140 	return rc;
6141 }
6142 
6143 static int efx_ef10_rx_disable_timestamping(struct efx_channel *channel,
6144 					    bool temp)
6145 {
6146 	MCDI_DECLARE_BUF(inbuf, MC_CMD_PTP_IN_TIME_EVENT_UNSUBSCRIBE_LEN);
6147 	int rc;
6148 
6149 	if (channel->sync_events_state == SYNC_EVENTS_DISABLED ||
6150 	    (temp && channel->sync_events_state == SYNC_EVENTS_QUIESCENT))
6151 		return 0;
6152 	if (channel->sync_events_state == SYNC_EVENTS_QUIESCENT) {
6153 		channel->sync_events_state = SYNC_EVENTS_DISABLED;
6154 		return 0;
6155 	}
6156 	channel->sync_events_state = temp ? SYNC_EVENTS_QUIESCENT :
6157 					    SYNC_EVENTS_DISABLED;
6158 
6159 	MCDI_SET_DWORD(inbuf, PTP_IN_OP, MC_CMD_PTP_OP_TIME_EVENT_UNSUBSCRIBE);
6160 	MCDI_SET_DWORD(inbuf, PTP_IN_PERIPH_ID, 0);
6161 	MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_UNSUBSCRIBE_CONTROL,
6162 		       MC_CMD_PTP_IN_TIME_EVENT_UNSUBSCRIBE_SINGLE);
6163 	MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_UNSUBSCRIBE_QUEUE,
6164 		       channel->channel);
6165 
6166 	rc = efx_mcdi_rpc(channel->efx, MC_CMD_PTP,
6167 			  inbuf, sizeof(inbuf), NULL, 0, NULL);
6168 
6169 	return rc;
6170 }
6171 
6172 static int efx_ef10_ptp_set_ts_sync_events(struct efx_nic *efx, bool en,
6173 					   bool temp)
6174 {
6175 	int (*set)(struct efx_channel *channel, bool temp);
6176 	struct efx_channel *channel;
6177 
6178 	set = en ?
6179 	      efx_ef10_rx_enable_timestamping :
6180 	      efx_ef10_rx_disable_timestamping;
6181 
6182 	efx_for_each_channel(channel, efx) {
6183 		int rc = set(channel, temp);
6184 		if (en && rc != 0) {
6185 			efx_ef10_ptp_set_ts_sync_events(efx, false, temp);
6186 			return rc;
6187 		}
6188 	}
6189 
6190 	return 0;
6191 }
6192 
6193 static int efx_ef10_ptp_set_ts_config_vf(struct efx_nic *efx,
6194 					 struct hwtstamp_config *init)
6195 {
6196 	return -EOPNOTSUPP;
6197 }
6198 
6199 static int efx_ef10_ptp_set_ts_config(struct efx_nic *efx,
6200 				      struct hwtstamp_config *init)
6201 {
6202 	int rc;
6203 
6204 	switch (init->rx_filter) {
6205 	case HWTSTAMP_FILTER_NONE:
6206 		efx_ef10_ptp_set_ts_sync_events(efx, false, false);
6207 		/* if TX timestamping is still requested then leave PTP on */
6208 		return efx_ptp_change_mode(efx,
6209 					   init->tx_type != HWTSTAMP_TX_OFF, 0);
6210 	case HWTSTAMP_FILTER_ALL:
6211 	case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
6212 	case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
6213 	case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
6214 	case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
6215 	case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
6216 	case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
6217 	case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
6218 	case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
6219 	case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
6220 	case HWTSTAMP_FILTER_PTP_V2_EVENT:
6221 	case HWTSTAMP_FILTER_PTP_V2_SYNC:
6222 	case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
6223 	case HWTSTAMP_FILTER_NTP_ALL:
6224 		init->rx_filter = HWTSTAMP_FILTER_ALL;
6225 		rc = efx_ptp_change_mode(efx, true, 0);
6226 		if (!rc)
6227 			rc = efx_ef10_ptp_set_ts_sync_events(efx, true, false);
6228 		if (rc)
6229 			efx_ptp_change_mode(efx, false, 0);
6230 		return rc;
6231 	default:
6232 		return -ERANGE;
6233 	}
6234 }
6235 
6236 static int efx_ef10_get_phys_port_id(struct efx_nic *efx,
6237 				     struct netdev_phys_item_id *ppid)
6238 {
6239 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
6240 
6241 	if (!is_valid_ether_addr(nic_data->port_id))
6242 		return -EOPNOTSUPP;
6243 
6244 	ppid->id_len = ETH_ALEN;
6245 	memcpy(ppid->id, nic_data->port_id, ppid->id_len);
6246 
6247 	return 0;
6248 }
6249 
6250 static int efx_ef10_vlan_rx_add_vid(struct efx_nic *efx, __be16 proto, u16 vid)
6251 {
6252 	if (proto != htons(ETH_P_8021Q))
6253 		return -EINVAL;
6254 
6255 	return efx_ef10_add_vlan(efx, vid);
6256 }
6257 
6258 static int efx_ef10_vlan_rx_kill_vid(struct efx_nic *efx, __be16 proto, u16 vid)
6259 {
6260 	if (proto != htons(ETH_P_8021Q))
6261 		return -EINVAL;
6262 
6263 	return efx_ef10_del_vlan(efx, vid);
6264 }
6265 
6266 /* We rely on the MCDI wiping out our TX rings if it made any changes to the
6267  * ports table, ensuring that any TSO descriptors that were made on a now-
6268  * removed tunnel port will be blown away and won't break things when we try
6269  * to transmit them using the new ports table.
6270  */
6271 static int efx_ef10_set_udp_tnl_ports(struct efx_nic *efx, bool unloading)
6272 {
6273 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
6274 	MCDI_DECLARE_BUF(inbuf, MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_IN_LENMAX);
6275 	MCDI_DECLARE_BUF(outbuf, MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_OUT_LEN);
6276 	bool will_reset = false;
6277 	size_t num_entries = 0;
6278 	size_t inlen, outlen;
6279 	size_t i;
6280 	int rc;
6281 	efx_dword_t flags_and_num_entries;
6282 
6283 	WARN_ON(!mutex_is_locked(&nic_data->udp_tunnels_lock));
6284 
6285 	nic_data->udp_tunnels_dirty = false;
6286 
6287 	if (!(nic_data->datapath_caps &
6288 	    (1 << MC_CMD_GET_CAPABILITIES_OUT_VXLAN_NVGRE_LBN))) {
6289 		efx_device_attach_if_not_resetting(efx);
6290 		return 0;
6291 	}
6292 
6293 	BUILD_BUG_ON(ARRAY_SIZE(nic_data->udp_tunnels) >
6294 		     MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_IN_ENTRIES_MAXNUM);
6295 
6296 	for (i = 0; i < ARRAY_SIZE(nic_data->udp_tunnels); ++i) {
6297 		if (nic_data->udp_tunnels[i].count &&
6298 		    nic_data->udp_tunnels[i].port) {
6299 			efx_dword_t entry;
6300 
6301 			EFX_POPULATE_DWORD_2(entry,
6302 				TUNNEL_ENCAP_UDP_PORT_ENTRY_UDP_PORT,
6303 					ntohs(nic_data->udp_tunnels[i].port),
6304 				TUNNEL_ENCAP_UDP_PORT_ENTRY_PROTOCOL,
6305 					nic_data->udp_tunnels[i].type);
6306 			*_MCDI_ARRAY_DWORD(inbuf,
6307 				SET_TUNNEL_ENCAP_UDP_PORTS_IN_ENTRIES,
6308 				num_entries++) = entry;
6309 		}
6310 	}
6311 
6312 	BUILD_BUG_ON((MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_IN_NUM_ENTRIES_OFST -
6313 		      MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_IN_FLAGS_OFST) * 8 !=
6314 		     EFX_WORD_1_LBN);
6315 	BUILD_BUG_ON(MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_IN_NUM_ENTRIES_LEN * 8 !=
6316 		     EFX_WORD_1_WIDTH);
6317 	EFX_POPULATE_DWORD_2(flags_and_num_entries,
6318 			     MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_IN_UNLOADING,
6319 				!!unloading,
6320 			     EFX_WORD_1, num_entries);
6321 	*_MCDI_DWORD(inbuf, SET_TUNNEL_ENCAP_UDP_PORTS_IN_FLAGS) =
6322 		flags_and_num_entries;
6323 
6324 	inlen = MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_IN_LEN(num_entries);
6325 
6326 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS,
6327 				inbuf, inlen, outbuf, sizeof(outbuf), &outlen);
6328 	if (rc == -EIO) {
6329 		/* Most likely the MC rebooted due to another function also
6330 		 * setting its tunnel port list. Mark the tunnel port list as
6331 		 * dirty, so it will be pushed upon coming up from the reboot.
6332 		 */
6333 		nic_data->udp_tunnels_dirty = true;
6334 		return 0;
6335 	}
6336 
6337 	if (rc) {
6338 		/* expected not available on unprivileged functions */
6339 		if (rc != -EPERM)
6340 			netif_warn(efx, drv, efx->net_dev,
6341 				   "Unable to set UDP tunnel ports; rc=%d.\n", rc);
6342 	} else if (MCDI_DWORD(outbuf, SET_TUNNEL_ENCAP_UDP_PORTS_OUT_FLAGS) &
6343 		   (1 << MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_OUT_RESETTING_LBN)) {
6344 		netif_info(efx, drv, efx->net_dev,
6345 			   "Rebooting MC due to UDP tunnel port list change\n");
6346 		will_reset = true;
6347 		if (unloading)
6348 			/* Delay for the MC reset to complete. This will make
6349 			 * unloading other functions a bit smoother. This is a
6350 			 * race, but the other unload will work whichever way
6351 			 * it goes, this just avoids an unnecessary error
6352 			 * message.
6353 			 */
6354 			msleep(100);
6355 	}
6356 	if (!will_reset && !unloading) {
6357 		/* The caller will have detached, relying on the MC reset to
6358 		 * trigger a re-attach.  Since there won't be an MC reset, we
6359 		 * have to do the attach ourselves.
6360 		 */
6361 		efx_device_attach_if_not_resetting(efx);
6362 	}
6363 
6364 	return rc;
6365 }
6366 
6367 static int efx_ef10_udp_tnl_push_ports(struct efx_nic *efx)
6368 {
6369 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
6370 	int rc = 0;
6371 
6372 	mutex_lock(&nic_data->udp_tunnels_lock);
6373 	if (nic_data->udp_tunnels_dirty) {
6374 		/* Make sure all TX are stopped while we modify the table, else
6375 		 * we might race against an efx_features_check().
6376 		 */
6377 		efx_device_detach_sync(efx);
6378 		rc = efx_ef10_set_udp_tnl_ports(efx, false);
6379 	}
6380 	mutex_unlock(&nic_data->udp_tunnels_lock);
6381 	return rc;
6382 }
6383 
6384 static struct efx_udp_tunnel *__efx_ef10_udp_tnl_lookup_port(struct efx_nic *efx,
6385 							     __be16 port)
6386 {
6387 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
6388 	size_t i;
6389 
6390 	for (i = 0; i < ARRAY_SIZE(nic_data->udp_tunnels); ++i) {
6391 		if (!nic_data->udp_tunnels[i].count)
6392 			continue;
6393 		if (nic_data->udp_tunnels[i].port == port)
6394 			return &nic_data->udp_tunnels[i];
6395 	}
6396 	return NULL;
6397 }
6398 
6399 static int efx_ef10_udp_tnl_add_port(struct efx_nic *efx,
6400 				     struct efx_udp_tunnel tnl)
6401 {
6402 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
6403 	struct efx_udp_tunnel *match;
6404 	char typebuf[8];
6405 	size_t i;
6406 	int rc;
6407 
6408 	if (!(nic_data->datapath_caps &
6409 	      (1 << MC_CMD_GET_CAPABILITIES_OUT_VXLAN_NVGRE_LBN)))
6410 		return 0;
6411 
6412 	efx_get_udp_tunnel_type_name(tnl.type, typebuf, sizeof(typebuf));
6413 	netif_dbg(efx, drv, efx->net_dev, "Adding UDP tunnel (%s) port %d\n",
6414 		  typebuf, ntohs(tnl.port));
6415 
6416 	mutex_lock(&nic_data->udp_tunnels_lock);
6417 	/* Make sure all TX are stopped while we add to the table, else we
6418 	 * might race against an efx_features_check().
6419 	 */
6420 	efx_device_detach_sync(efx);
6421 
6422 	match = __efx_ef10_udp_tnl_lookup_port(efx, tnl.port);
6423 	if (match != NULL) {
6424 		if (match->type == tnl.type) {
6425 			netif_dbg(efx, drv, efx->net_dev,
6426 				  "Referencing existing tunnel entry\n");
6427 			match->count++;
6428 			/* No need to cause an MCDI update */
6429 			rc = 0;
6430 			goto unlock_out;
6431 		}
6432 		efx_get_udp_tunnel_type_name(match->type,
6433 					     typebuf, sizeof(typebuf));
6434 		netif_dbg(efx, drv, efx->net_dev,
6435 			  "UDP port %d is already in use by %s\n",
6436 			  ntohs(tnl.port), typebuf);
6437 		rc = -EEXIST;
6438 		goto unlock_out;
6439 	}
6440 
6441 	for (i = 0; i < ARRAY_SIZE(nic_data->udp_tunnels); ++i)
6442 		if (!nic_data->udp_tunnels[i].count) {
6443 			nic_data->udp_tunnels[i] = tnl;
6444 			nic_data->udp_tunnels[i].count = 1;
6445 			rc = efx_ef10_set_udp_tnl_ports(efx, false);
6446 			goto unlock_out;
6447 		}
6448 
6449 	netif_dbg(efx, drv, efx->net_dev,
6450 		  "Unable to add UDP tunnel (%s) port %d; insufficient resources.\n",
6451 		  typebuf, ntohs(tnl.port));
6452 
6453 	rc = -ENOMEM;
6454 
6455 unlock_out:
6456 	mutex_unlock(&nic_data->udp_tunnels_lock);
6457 	return rc;
6458 }
6459 
6460 /* Called under the TX lock with the TX queue running, hence no-one can be
6461  * in the middle of updating the UDP tunnels table.  However, they could
6462  * have tried and failed the MCDI, in which case they'll have set the dirty
6463  * flag before dropping their locks.
6464  */
6465 static bool efx_ef10_udp_tnl_has_port(struct efx_nic *efx, __be16 port)
6466 {
6467 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
6468 
6469 	if (!(nic_data->datapath_caps &
6470 	      (1 << MC_CMD_GET_CAPABILITIES_OUT_VXLAN_NVGRE_LBN)))
6471 		return false;
6472 
6473 	if (nic_data->udp_tunnels_dirty)
6474 		/* SW table may not match HW state, so just assume we can't
6475 		 * use any UDP tunnel offloads.
6476 		 */
6477 		return false;
6478 
6479 	return __efx_ef10_udp_tnl_lookup_port(efx, port) != NULL;
6480 }
6481 
6482 static int efx_ef10_udp_tnl_del_port(struct efx_nic *efx,
6483 				     struct efx_udp_tunnel tnl)
6484 {
6485 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
6486 	struct efx_udp_tunnel *match;
6487 	char typebuf[8];
6488 	int rc;
6489 
6490 	if (!(nic_data->datapath_caps &
6491 	      (1 << MC_CMD_GET_CAPABILITIES_OUT_VXLAN_NVGRE_LBN)))
6492 		return 0;
6493 
6494 	efx_get_udp_tunnel_type_name(tnl.type, typebuf, sizeof(typebuf));
6495 	netif_dbg(efx, drv, efx->net_dev, "Removing UDP tunnel (%s) port %d\n",
6496 		  typebuf, ntohs(tnl.port));
6497 
6498 	mutex_lock(&nic_data->udp_tunnels_lock);
6499 	/* Make sure all TX are stopped while we remove from the table, else we
6500 	 * might race against an efx_features_check().
6501 	 */
6502 	efx_device_detach_sync(efx);
6503 
6504 	match = __efx_ef10_udp_tnl_lookup_port(efx, tnl.port);
6505 	if (match != NULL) {
6506 		if (match->type == tnl.type) {
6507 			if (--match->count) {
6508 				/* Port is still in use, so nothing to do */
6509 				netif_dbg(efx, drv, efx->net_dev,
6510 					  "UDP tunnel port %d remains active\n",
6511 					  ntohs(tnl.port));
6512 				rc = 0;
6513 				goto out_unlock;
6514 			}
6515 			rc = efx_ef10_set_udp_tnl_ports(efx, false);
6516 			goto out_unlock;
6517 		}
6518 		efx_get_udp_tunnel_type_name(match->type,
6519 					     typebuf, sizeof(typebuf));
6520 		netif_warn(efx, drv, efx->net_dev,
6521 			   "UDP port %d is actually in use by %s, not removing\n",
6522 			   ntohs(tnl.port), typebuf);
6523 	}
6524 	rc = -ENOENT;
6525 
6526 out_unlock:
6527 	mutex_unlock(&nic_data->udp_tunnels_lock);
6528 	return rc;
6529 }
6530 
6531 #define EF10_OFFLOAD_FEATURES		\
6532 	(NETIF_F_IP_CSUM |		\
6533 	 NETIF_F_HW_VLAN_CTAG_FILTER |	\
6534 	 NETIF_F_IPV6_CSUM |		\
6535 	 NETIF_F_RXHASH |		\
6536 	 NETIF_F_NTUPLE)
6537 
6538 const struct efx_nic_type efx_hunt_a0_vf_nic_type = {
6539 	.is_vf = true,
6540 	.mem_bar = efx_ef10_vf_mem_bar,
6541 	.mem_map_size = efx_ef10_mem_map_size,
6542 	.probe = efx_ef10_probe_vf,
6543 	.remove = efx_ef10_remove,
6544 	.dimension_resources = efx_ef10_dimension_resources,
6545 	.init = efx_ef10_init_nic,
6546 	.fini = efx_port_dummy_op_void,
6547 	.map_reset_reason = efx_ef10_map_reset_reason,
6548 	.map_reset_flags = efx_ef10_map_reset_flags,
6549 	.reset = efx_ef10_reset,
6550 	.probe_port = efx_mcdi_port_probe,
6551 	.remove_port = efx_mcdi_port_remove,
6552 	.fini_dmaq = efx_ef10_fini_dmaq,
6553 	.prepare_flr = efx_ef10_prepare_flr,
6554 	.finish_flr = efx_port_dummy_op_void,
6555 	.describe_stats = efx_ef10_describe_stats,
6556 	.update_stats = efx_ef10_update_stats_vf,
6557 	.start_stats = efx_port_dummy_op_void,
6558 	.pull_stats = efx_port_dummy_op_void,
6559 	.stop_stats = efx_port_dummy_op_void,
6560 	.set_id_led = efx_mcdi_set_id_led,
6561 	.push_irq_moderation = efx_ef10_push_irq_moderation,
6562 	.reconfigure_mac = efx_ef10_mac_reconfigure_vf,
6563 	.check_mac_fault = efx_mcdi_mac_check_fault,
6564 	.reconfigure_port = efx_mcdi_port_reconfigure,
6565 	.get_wol = efx_ef10_get_wol_vf,
6566 	.set_wol = efx_ef10_set_wol_vf,
6567 	.resume_wol = efx_port_dummy_op_void,
6568 	.mcdi_request = efx_ef10_mcdi_request,
6569 	.mcdi_poll_response = efx_ef10_mcdi_poll_response,
6570 	.mcdi_read_response = efx_ef10_mcdi_read_response,
6571 	.mcdi_poll_reboot = efx_ef10_mcdi_poll_reboot,
6572 	.mcdi_reboot_detected = efx_ef10_mcdi_reboot_detected,
6573 	.irq_enable_master = efx_port_dummy_op_void,
6574 	.irq_test_generate = efx_ef10_irq_test_generate,
6575 	.irq_disable_non_ev = efx_port_dummy_op_void,
6576 	.irq_handle_msi = efx_ef10_msi_interrupt,
6577 	.irq_handle_legacy = efx_ef10_legacy_interrupt,
6578 	.tx_probe = efx_ef10_tx_probe,
6579 	.tx_init = efx_ef10_tx_init,
6580 	.tx_remove = efx_ef10_tx_remove,
6581 	.tx_write = efx_ef10_tx_write,
6582 	.tx_limit_len = efx_ef10_tx_limit_len,
6583 	.rx_push_rss_config = efx_ef10_vf_rx_push_rss_config,
6584 	.rx_pull_rss_config = efx_ef10_rx_pull_rss_config,
6585 	.rx_probe = efx_ef10_rx_probe,
6586 	.rx_init = efx_ef10_rx_init,
6587 	.rx_remove = efx_ef10_rx_remove,
6588 	.rx_write = efx_ef10_rx_write,
6589 	.rx_defer_refill = efx_ef10_rx_defer_refill,
6590 	.ev_probe = efx_ef10_ev_probe,
6591 	.ev_init = efx_ef10_ev_init,
6592 	.ev_fini = efx_ef10_ev_fini,
6593 	.ev_remove = efx_ef10_ev_remove,
6594 	.ev_process = efx_ef10_ev_process,
6595 	.ev_read_ack = efx_ef10_ev_read_ack,
6596 	.ev_test_generate = efx_ef10_ev_test_generate,
6597 	.filter_table_probe = efx_ef10_filter_table_probe,
6598 	.filter_table_restore = efx_ef10_filter_table_restore,
6599 	.filter_table_remove = efx_ef10_filter_table_remove,
6600 	.filter_update_rx_scatter = efx_ef10_filter_update_rx_scatter,
6601 	.filter_insert = efx_ef10_filter_insert,
6602 	.filter_remove_safe = efx_ef10_filter_remove_safe,
6603 	.filter_get_safe = efx_ef10_filter_get_safe,
6604 	.filter_clear_rx = efx_ef10_filter_clear_rx,
6605 	.filter_count_rx_used = efx_ef10_filter_count_rx_used,
6606 	.filter_get_rx_id_limit = efx_ef10_filter_get_rx_id_limit,
6607 	.filter_get_rx_ids = efx_ef10_filter_get_rx_ids,
6608 #ifdef CONFIG_RFS_ACCEL
6609 	.filter_rfs_insert = efx_ef10_filter_rfs_insert,
6610 	.filter_rfs_expire_one = efx_ef10_filter_rfs_expire_one,
6611 #endif
6612 #ifdef CONFIG_SFC_MTD
6613 	.mtd_probe = efx_port_dummy_op_int,
6614 #endif
6615 	.ptp_write_host_time = efx_ef10_ptp_write_host_time_vf,
6616 	.ptp_set_ts_config = efx_ef10_ptp_set_ts_config_vf,
6617 	.vlan_rx_add_vid = efx_ef10_vlan_rx_add_vid,
6618 	.vlan_rx_kill_vid = efx_ef10_vlan_rx_kill_vid,
6619 #ifdef CONFIG_SFC_SRIOV
6620 	.vswitching_probe = efx_ef10_vswitching_probe_vf,
6621 	.vswitching_restore = efx_ef10_vswitching_restore_vf,
6622 	.vswitching_remove = efx_ef10_vswitching_remove_vf,
6623 #endif
6624 	.get_mac_address = efx_ef10_get_mac_address_vf,
6625 	.set_mac_address = efx_ef10_set_mac_address,
6626 
6627 	.get_phys_port_id = efx_ef10_get_phys_port_id,
6628 	.revision = EFX_REV_HUNT_A0,
6629 	.max_dma_mask = DMA_BIT_MASK(ESF_DZ_TX_KER_BUF_ADDR_WIDTH),
6630 	.rx_prefix_size = ES_DZ_RX_PREFIX_SIZE,
6631 	.rx_hash_offset = ES_DZ_RX_PREFIX_HASH_OFST,
6632 	.rx_ts_offset = ES_DZ_RX_PREFIX_TSTAMP_OFST,
6633 	.can_rx_scatter = true,
6634 	.always_rx_scatter = true,
6635 	.min_interrupt_mode = EFX_INT_MODE_MSIX,
6636 	.max_interrupt_mode = EFX_INT_MODE_MSIX,
6637 	.timer_period_max = 1 << ERF_DD_EVQ_IND_TIMER_VAL_WIDTH,
6638 	.offload_features = EF10_OFFLOAD_FEATURES,
6639 	.mcdi_max_ver = 2,
6640 	.max_rx_ip_filters = HUNT_FILTER_TBL_ROWS,
6641 	.hwtstamp_filters = 1 << HWTSTAMP_FILTER_NONE |
6642 			    1 << HWTSTAMP_FILTER_ALL,
6643 	.rx_hash_key_size = 40,
6644 };
6645 
6646 const struct efx_nic_type efx_hunt_a0_nic_type = {
6647 	.is_vf = false,
6648 	.mem_bar = efx_ef10_pf_mem_bar,
6649 	.mem_map_size = efx_ef10_mem_map_size,
6650 	.probe = efx_ef10_probe_pf,
6651 	.remove = efx_ef10_remove,
6652 	.dimension_resources = efx_ef10_dimension_resources,
6653 	.init = efx_ef10_init_nic,
6654 	.fini = efx_port_dummy_op_void,
6655 	.map_reset_reason = efx_ef10_map_reset_reason,
6656 	.map_reset_flags = efx_ef10_map_reset_flags,
6657 	.reset = efx_ef10_reset,
6658 	.probe_port = efx_mcdi_port_probe,
6659 	.remove_port = efx_mcdi_port_remove,
6660 	.fini_dmaq = efx_ef10_fini_dmaq,
6661 	.prepare_flr = efx_ef10_prepare_flr,
6662 	.finish_flr = efx_port_dummy_op_void,
6663 	.describe_stats = efx_ef10_describe_stats,
6664 	.update_stats = efx_ef10_update_stats_pf,
6665 	.start_stats = efx_mcdi_mac_start_stats,
6666 	.pull_stats = efx_mcdi_mac_pull_stats,
6667 	.stop_stats = efx_mcdi_mac_stop_stats,
6668 	.set_id_led = efx_mcdi_set_id_led,
6669 	.push_irq_moderation = efx_ef10_push_irq_moderation,
6670 	.reconfigure_mac = efx_ef10_mac_reconfigure,
6671 	.check_mac_fault = efx_mcdi_mac_check_fault,
6672 	.reconfigure_port = efx_mcdi_port_reconfigure,
6673 	.get_wol = efx_ef10_get_wol,
6674 	.set_wol = efx_ef10_set_wol,
6675 	.resume_wol = efx_port_dummy_op_void,
6676 	.test_chip = efx_ef10_test_chip,
6677 	.test_nvram = efx_mcdi_nvram_test_all,
6678 	.mcdi_request = efx_ef10_mcdi_request,
6679 	.mcdi_poll_response = efx_ef10_mcdi_poll_response,
6680 	.mcdi_read_response = efx_ef10_mcdi_read_response,
6681 	.mcdi_poll_reboot = efx_ef10_mcdi_poll_reboot,
6682 	.mcdi_reboot_detected = efx_ef10_mcdi_reboot_detected,
6683 	.irq_enable_master = efx_port_dummy_op_void,
6684 	.irq_test_generate = efx_ef10_irq_test_generate,
6685 	.irq_disable_non_ev = efx_port_dummy_op_void,
6686 	.irq_handle_msi = efx_ef10_msi_interrupt,
6687 	.irq_handle_legacy = efx_ef10_legacy_interrupt,
6688 	.tx_probe = efx_ef10_tx_probe,
6689 	.tx_init = efx_ef10_tx_init,
6690 	.tx_remove = efx_ef10_tx_remove,
6691 	.tx_write = efx_ef10_tx_write,
6692 	.tx_limit_len = efx_ef10_tx_limit_len,
6693 	.rx_push_rss_config = efx_ef10_pf_rx_push_rss_config,
6694 	.rx_pull_rss_config = efx_ef10_rx_pull_rss_config,
6695 	.rx_probe = efx_ef10_rx_probe,
6696 	.rx_init = efx_ef10_rx_init,
6697 	.rx_remove = efx_ef10_rx_remove,
6698 	.rx_write = efx_ef10_rx_write,
6699 	.rx_defer_refill = efx_ef10_rx_defer_refill,
6700 	.ev_probe = efx_ef10_ev_probe,
6701 	.ev_init = efx_ef10_ev_init,
6702 	.ev_fini = efx_ef10_ev_fini,
6703 	.ev_remove = efx_ef10_ev_remove,
6704 	.ev_process = efx_ef10_ev_process,
6705 	.ev_read_ack = efx_ef10_ev_read_ack,
6706 	.ev_test_generate = efx_ef10_ev_test_generate,
6707 	.filter_table_probe = efx_ef10_filter_table_probe,
6708 	.filter_table_restore = efx_ef10_filter_table_restore,
6709 	.filter_table_remove = efx_ef10_filter_table_remove,
6710 	.filter_update_rx_scatter = efx_ef10_filter_update_rx_scatter,
6711 	.filter_insert = efx_ef10_filter_insert,
6712 	.filter_remove_safe = efx_ef10_filter_remove_safe,
6713 	.filter_get_safe = efx_ef10_filter_get_safe,
6714 	.filter_clear_rx = efx_ef10_filter_clear_rx,
6715 	.filter_count_rx_used = efx_ef10_filter_count_rx_used,
6716 	.filter_get_rx_id_limit = efx_ef10_filter_get_rx_id_limit,
6717 	.filter_get_rx_ids = efx_ef10_filter_get_rx_ids,
6718 #ifdef CONFIG_RFS_ACCEL
6719 	.filter_rfs_insert = efx_ef10_filter_rfs_insert,
6720 	.filter_rfs_expire_one = efx_ef10_filter_rfs_expire_one,
6721 #endif
6722 #ifdef CONFIG_SFC_MTD
6723 	.mtd_probe = efx_ef10_mtd_probe,
6724 	.mtd_rename = efx_mcdi_mtd_rename,
6725 	.mtd_read = efx_mcdi_mtd_read,
6726 	.mtd_erase = efx_mcdi_mtd_erase,
6727 	.mtd_write = efx_mcdi_mtd_write,
6728 	.mtd_sync = efx_mcdi_mtd_sync,
6729 #endif
6730 	.ptp_write_host_time = efx_ef10_ptp_write_host_time,
6731 	.ptp_set_ts_sync_events = efx_ef10_ptp_set_ts_sync_events,
6732 	.ptp_set_ts_config = efx_ef10_ptp_set_ts_config,
6733 	.vlan_rx_add_vid = efx_ef10_vlan_rx_add_vid,
6734 	.vlan_rx_kill_vid = efx_ef10_vlan_rx_kill_vid,
6735 	.udp_tnl_push_ports = efx_ef10_udp_tnl_push_ports,
6736 	.udp_tnl_add_port = efx_ef10_udp_tnl_add_port,
6737 	.udp_tnl_has_port = efx_ef10_udp_tnl_has_port,
6738 	.udp_tnl_del_port = efx_ef10_udp_tnl_del_port,
6739 #ifdef CONFIG_SFC_SRIOV
6740 	.sriov_configure = efx_ef10_sriov_configure,
6741 	.sriov_init = efx_ef10_sriov_init,
6742 	.sriov_fini = efx_ef10_sriov_fini,
6743 	.sriov_wanted = efx_ef10_sriov_wanted,
6744 	.sriov_reset = efx_ef10_sriov_reset,
6745 	.sriov_flr = efx_ef10_sriov_flr,
6746 	.sriov_set_vf_mac = efx_ef10_sriov_set_vf_mac,
6747 	.sriov_set_vf_vlan = efx_ef10_sriov_set_vf_vlan,
6748 	.sriov_set_vf_spoofchk = efx_ef10_sriov_set_vf_spoofchk,
6749 	.sriov_get_vf_config = efx_ef10_sriov_get_vf_config,
6750 	.sriov_set_vf_link_state = efx_ef10_sriov_set_vf_link_state,
6751 	.vswitching_probe = efx_ef10_vswitching_probe_pf,
6752 	.vswitching_restore = efx_ef10_vswitching_restore_pf,
6753 	.vswitching_remove = efx_ef10_vswitching_remove_pf,
6754 #endif
6755 	.get_mac_address = efx_ef10_get_mac_address_pf,
6756 	.set_mac_address = efx_ef10_set_mac_address,
6757 	.tso_versions = efx_ef10_tso_versions,
6758 
6759 	.get_phys_port_id = efx_ef10_get_phys_port_id,
6760 	.revision = EFX_REV_HUNT_A0,
6761 	.max_dma_mask = DMA_BIT_MASK(ESF_DZ_TX_KER_BUF_ADDR_WIDTH),
6762 	.rx_prefix_size = ES_DZ_RX_PREFIX_SIZE,
6763 	.rx_hash_offset = ES_DZ_RX_PREFIX_HASH_OFST,
6764 	.rx_ts_offset = ES_DZ_RX_PREFIX_TSTAMP_OFST,
6765 	.can_rx_scatter = true,
6766 	.always_rx_scatter = true,
6767 	.option_descriptors = true,
6768 	.min_interrupt_mode = EFX_INT_MODE_LEGACY,
6769 	.max_interrupt_mode = EFX_INT_MODE_MSIX,
6770 	.timer_period_max = 1 << ERF_DD_EVQ_IND_TIMER_VAL_WIDTH,
6771 	.offload_features = EF10_OFFLOAD_FEATURES,
6772 	.mcdi_max_ver = 2,
6773 	.max_rx_ip_filters = HUNT_FILTER_TBL_ROWS,
6774 	.hwtstamp_filters = 1 << HWTSTAMP_FILTER_NONE |
6775 			    1 << HWTSTAMP_FILTER_ALL,
6776 	.rx_hash_key_size = 40,
6777 };
6778