xref: /openbmc/linux/drivers/net/ethernet/sfc/ef10.c (revision 4ee812f6)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /****************************************************************************
3  * Driver for Solarflare network controllers and boards
4  * Copyright 2012-2013 Solarflare Communications Inc.
5  */
6 
7 #include "net_driver.h"
8 #include "ef10_regs.h"
9 #include "io.h"
10 #include "mcdi.h"
11 #include "mcdi_pcol.h"
12 #include "nic.h"
13 #include "workarounds.h"
14 #include "selftest.h"
15 #include "ef10_sriov.h"
16 #include <linux/in.h>
17 #include <linux/jhash.h>
18 #include <linux/wait.h>
19 #include <linux/workqueue.h>
20 
21 /* Hardware control for EF10 architecture including 'Huntington'. */
22 
23 #define EFX_EF10_DRVGEN_EV		7
24 enum {
25 	EFX_EF10_TEST = 1,
26 	EFX_EF10_REFILL,
27 };
28 /* The maximum size of a shared RSS context */
29 /* TODO: this should really be from the mcdi protocol export */
30 #define EFX_EF10_MAX_SHARED_RSS_CONTEXT_SIZE 64UL
31 
32 /* The filter table(s) are managed by firmware and we have write-only
33  * access.  When removing filters we must identify them to the
34  * firmware by a 64-bit handle, but this is too wide for Linux kernel
35  * interfaces (32-bit for RX NFC, 16-bit for RFS).  Also, we need to
36  * be able to tell in advance whether a requested insertion will
37  * replace an existing filter.  Therefore we maintain a software hash
38  * table, which should be at least as large as the hardware hash
39  * table.
40  *
41  * Huntington has a single 8K filter table shared between all filter
42  * types and both ports.
43  */
44 #define HUNT_FILTER_TBL_ROWS 8192
45 
46 #define EFX_EF10_FILTER_ID_INVALID 0xffff
47 
48 #define EFX_EF10_FILTER_DEV_UC_MAX	32
49 #define EFX_EF10_FILTER_DEV_MC_MAX	256
50 
51 /* VLAN list entry */
52 struct efx_ef10_vlan {
53 	struct list_head list;
54 	u16 vid;
55 };
56 
57 enum efx_ef10_default_filters {
58 	EFX_EF10_BCAST,
59 	EFX_EF10_UCDEF,
60 	EFX_EF10_MCDEF,
61 	EFX_EF10_VXLAN4_UCDEF,
62 	EFX_EF10_VXLAN4_MCDEF,
63 	EFX_EF10_VXLAN6_UCDEF,
64 	EFX_EF10_VXLAN6_MCDEF,
65 	EFX_EF10_NVGRE4_UCDEF,
66 	EFX_EF10_NVGRE4_MCDEF,
67 	EFX_EF10_NVGRE6_UCDEF,
68 	EFX_EF10_NVGRE6_MCDEF,
69 	EFX_EF10_GENEVE4_UCDEF,
70 	EFX_EF10_GENEVE4_MCDEF,
71 	EFX_EF10_GENEVE6_UCDEF,
72 	EFX_EF10_GENEVE6_MCDEF,
73 
74 	EFX_EF10_NUM_DEFAULT_FILTERS
75 };
76 
77 /* Per-VLAN filters information */
78 struct efx_ef10_filter_vlan {
79 	struct list_head list;
80 	u16 vid;
81 	u16 uc[EFX_EF10_FILTER_DEV_UC_MAX];
82 	u16 mc[EFX_EF10_FILTER_DEV_MC_MAX];
83 	u16 default_filters[EFX_EF10_NUM_DEFAULT_FILTERS];
84 };
85 
86 struct efx_ef10_dev_addr {
87 	u8 addr[ETH_ALEN];
88 };
89 
90 struct efx_ef10_filter_table {
91 /* The MCDI match masks supported by this fw & hw, in order of priority */
92 	u32 rx_match_mcdi_flags[
93 		MC_CMD_GET_PARSER_DISP_INFO_OUT_SUPPORTED_MATCHES_MAXNUM * 2];
94 	unsigned int rx_match_count;
95 
96 	struct rw_semaphore lock; /* Protects entries */
97 	struct {
98 		unsigned long spec;	/* pointer to spec plus flag bits */
99 /* AUTO_OLD is used to mark and sweep MAC filters for the device address lists. */
100 /* unused flag	1UL */
101 #define EFX_EF10_FILTER_FLAG_AUTO_OLD	2UL
102 #define EFX_EF10_FILTER_FLAGS		3UL
103 		u64 handle;		/* firmware handle */
104 	} *entry;
105 /* Shadow of net_device address lists, guarded by mac_lock */
106 	struct efx_ef10_dev_addr dev_uc_list[EFX_EF10_FILTER_DEV_UC_MAX];
107 	struct efx_ef10_dev_addr dev_mc_list[EFX_EF10_FILTER_DEV_MC_MAX];
108 	int dev_uc_count;
109 	int dev_mc_count;
110 	bool uc_promisc;
111 	bool mc_promisc;
112 /* Whether in multicast promiscuous mode when last changed */
113 	bool mc_promisc_last;
114 	bool mc_overflow; /* Too many MC addrs; should always imply mc_promisc */
115 	bool vlan_filter;
116 	struct list_head vlan_list;
117 };
118 
119 /* An arbitrary search limit for the software hash table */
120 #define EFX_EF10_FILTER_SEARCH_LIMIT 200
121 
122 static void efx_ef10_rx_free_indir_table(struct efx_nic *efx);
123 static void efx_ef10_filter_table_remove(struct efx_nic *efx);
124 static int efx_ef10_filter_add_vlan(struct efx_nic *efx, u16 vid);
125 static void efx_ef10_filter_del_vlan_internal(struct efx_nic *efx,
126 					      struct efx_ef10_filter_vlan *vlan);
127 static void efx_ef10_filter_del_vlan(struct efx_nic *efx, u16 vid);
128 static int efx_ef10_set_udp_tnl_ports(struct efx_nic *efx, bool unloading);
129 
130 static u32 efx_ef10_filter_get_unsafe_id(u32 filter_id)
131 {
132 	WARN_ON_ONCE(filter_id == EFX_EF10_FILTER_ID_INVALID);
133 	return filter_id & (HUNT_FILTER_TBL_ROWS - 1);
134 }
135 
136 static unsigned int efx_ef10_filter_get_unsafe_pri(u32 filter_id)
137 {
138 	return filter_id / (HUNT_FILTER_TBL_ROWS * 2);
139 }
140 
141 static u32 efx_ef10_make_filter_id(unsigned int pri, u16 idx)
142 {
143 	return pri * HUNT_FILTER_TBL_ROWS * 2 + idx;
144 }
145 
146 static int efx_ef10_get_warm_boot_count(struct efx_nic *efx)
147 {
148 	efx_dword_t reg;
149 
150 	efx_readd(efx, &reg, ER_DZ_BIU_MC_SFT_STATUS);
151 	return EFX_DWORD_FIELD(reg, EFX_WORD_1) == 0xb007 ?
152 		EFX_DWORD_FIELD(reg, EFX_WORD_0) : -EIO;
153 }
154 
155 /* On all EF10s up to and including SFC9220 (Medford1), all PFs use BAR 0 for
156  * I/O space and BAR 2(&3) for memory.  On SFC9250 (Medford2), there is no I/O
157  * bar; PFs use BAR 0/1 for memory.
158  */
159 static unsigned int efx_ef10_pf_mem_bar(struct efx_nic *efx)
160 {
161 	switch (efx->pci_dev->device) {
162 	case 0x0b03: /* SFC9250 PF */
163 		return 0;
164 	default:
165 		return 2;
166 	}
167 }
168 
169 /* All VFs use BAR 0/1 for memory */
170 static unsigned int efx_ef10_vf_mem_bar(struct efx_nic *efx)
171 {
172 	return 0;
173 }
174 
175 static unsigned int efx_ef10_mem_map_size(struct efx_nic *efx)
176 {
177 	int bar;
178 
179 	bar = efx->type->mem_bar(efx);
180 	return resource_size(&efx->pci_dev->resource[bar]);
181 }
182 
183 static bool efx_ef10_is_vf(struct efx_nic *efx)
184 {
185 	return efx->type->is_vf;
186 }
187 
188 static int efx_ef10_get_pf_index(struct efx_nic *efx)
189 {
190 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_FUNCTION_INFO_OUT_LEN);
191 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
192 	size_t outlen;
193 	int rc;
194 
195 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_FUNCTION_INFO, NULL, 0, outbuf,
196 			  sizeof(outbuf), &outlen);
197 	if (rc)
198 		return rc;
199 	if (outlen < sizeof(outbuf))
200 		return -EIO;
201 
202 	nic_data->pf_index = MCDI_DWORD(outbuf, GET_FUNCTION_INFO_OUT_PF);
203 	return 0;
204 }
205 
206 #ifdef CONFIG_SFC_SRIOV
207 static int efx_ef10_get_vf_index(struct efx_nic *efx)
208 {
209 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_FUNCTION_INFO_OUT_LEN);
210 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
211 	size_t outlen;
212 	int rc;
213 
214 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_FUNCTION_INFO, NULL, 0, outbuf,
215 			  sizeof(outbuf), &outlen);
216 	if (rc)
217 		return rc;
218 	if (outlen < sizeof(outbuf))
219 		return -EIO;
220 
221 	nic_data->vf_index = MCDI_DWORD(outbuf, GET_FUNCTION_INFO_OUT_VF);
222 	return 0;
223 }
224 #endif
225 
226 static int efx_ef10_init_datapath_caps(struct efx_nic *efx)
227 {
228 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_CAPABILITIES_V4_OUT_LEN);
229 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
230 	size_t outlen;
231 	int rc;
232 
233 	BUILD_BUG_ON(MC_CMD_GET_CAPABILITIES_IN_LEN != 0);
234 
235 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_CAPABILITIES, NULL, 0,
236 			  outbuf, sizeof(outbuf), &outlen);
237 	if (rc)
238 		return rc;
239 	if (outlen < MC_CMD_GET_CAPABILITIES_OUT_LEN) {
240 		netif_err(efx, drv, efx->net_dev,
241 			  "unable to read datapath firmware capabilities\n");
242 		return -EIO;
243 	}
244 
245 	nic_data->datapath_caps =
246 		MCDI_DWORD(outbuf, GET_CAPABILITIES_OUT_FLAGS1);
247 
248 	if (outlen >= MC_CMD_GET_CAPABILITIES_V2_OUT_LEN) {
249 		nic_data->datapath_caps2 = MCDI_DWORD(outbuf,
250 				GET_CAPABILITIES_V2_OUT_FLAGS2);
251 		nic_data->piobuf_size = MCDI_WORD(outbuf,
252 				GET_CAPABILITIES_V2_OUT_SIZE_PIO_BUFF);
253 	} else {
254 		nic_data->datapath_caps2 = 0;
255 		nic_data->piobuf_size = ER_DZ_TX_PIOBUF_SIZE;
256 	}
257 
258 	/* record the DPCPU firmware IDs to determine VEB vswitching support.
259 	 */
260 	nic_data->rx_dpcpu_fw_id =
261 		MCDI_WORD(outbuf, GET_CAPABILITIES_OUT_RX_DPCPU_FW_ID);
262 	nic_data->tx_dpcpu_fw_id =
263 		MCDI_WORD(outbuf, GET_CAPABILITIES_OUT_TX_DPCPU_FW_ID);
264 
265 	if (!(nic_data->datapath_caps &
266 	      (1 << MC_CMD_GET_CAPABILITIES_OUT_RX_PREFIX_LEN_14_LBN))) {
267 		netif_err(efx, probe, efx->net_dev,
268 			  "current firmware does not support an RX prefix\n");
269 		return -ENODEV;
270 	}
271 
272 	if (outlen >= MC_CMD_GET_CAPABILITIES_V3_OUT_LEN) {
273 		u8 vi_window_mode = MCDI_BYTE(outbuf,
274 				GET_CAPABILITIES_V3_OUT_VI_WINDOW_MODE);
275 
276 		switch (vi_window_mode) {
277 		case MC_CMD_GET_CAPABILITIES_V3_OUT_VI_WINDOW_MODE_8K:
278 			efx->vi_stride = 8192;
279 			break;
280 		case MC_CMD_GET_CAPABILITIES_V3_OUT_VI_WINDOW_MODE_16K:
281 			efx->vi_stride = 16384;
282 			break;
283 		case MC_CMD_GET_CAPABILITIES_V3_OUT_VI_WINDOW_MODE_64K:
284 			efx->vi_stride = 65536;
285 			break;
286 		default:
287 			netif_err(efx, probe, efx->net_dev,
288 				  "Unrecognised VI window mode %d\n",
289 				  vi_window_mode);
290 			return -EIO;
291 		}
292 		netif_dbg(efx, probe, efx->net_dev, "vi_stride = %u\n",
293 			  efx->vi_stride);
294 	} else {
295 		/* keep default VI stride */
296 		netif_dbg(efx, probe, efx->net_dev,
297 			  "firmware did not report VI window mode, assuming vi_stride = %u\n",
298 			  efx->vi_stride);
299 	}
300 
301 	if (outlen >= MC_CMD_GET_CAPABILITIES_V4_OUT_LEN) {
302 		efx->num_mac_stats = MCDI_WORD(outbuf,
303 				GET_CAPABILITIES_V4_OUT_MAC_STATS_NUM_STATS);
304 		netif_dbg(efx, probe, efx->net_dev,
305 			  "firmware reports num_mac_stats = %u\n",
306 			  efx->num_mac_stats);
307 	} else {
308 		/* leave num_mac_stats as the default value, MC_CMD_MAC_NSTATS */
309 		netif_dbg(efx, probe, efx->net_dev,
310 			  "firmware did not report num_mac_stats, assuming %u\n",
311 			  efx->num_mac_stats);
312 	}
313 
314 	return 0;
315 }
316 
317 static void efx_ef10_read_licensed_features(struct efx_nic *efx)
318 {
319 	MCDI_DECLARE_BUF(inbuf, MC_CMD_LICENSING_V3_IN_LEN);
320 	MCDI_DECLARE_BUF(outbuf, MC_CMD_LICENSING_V3_OUT_LEN);
321 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
322 	size_t outlen;
323 	int rc;
324 
325 	MCDI_SET_DWORD(inbuf, LICENSING_V3_IN_OP,
326 		       MC_CMD_LICENSING_V3_IN_OP_REPORT_LICENSE);
327 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_LICENSING_V3, inbuf, sizeof(inbuf),
328 				outbuf, sizeof(outbuf), &outlen);
329 	if (rc || (outlen < MC_CMD_LICENSING_V3_OUT_LEN))
330 		return;
331 
332 	nic_data->licensed_features = MCDI_QWORD(outbuf,
333 					 LICENSING_V3_OUT_LICENSED_FEATURES);
334 }
335 
336 static int efx_ef10_get_sysclk_freq(struct efx_nic *efx)
337 {
338 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_CLOCK_OUT_LEN);
339 	int rc;
340 
341 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_CLOCK, NULL, 0,
342 			  outbuf, sizeof(outbuf), NULL);
343 	if (rc)
344 		return rc;
345 	rc = MCDI_DWORD(outbuf, GET_CLOCK_OUT_SYS_FREQ);
346 	return rc > 0 ? rc : -ERANGE;
347 }
348 
349 static int efx_ef10_get_timer_workarounds(struct efx_nic *efx)
350 {
351 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
352 	unsigned int implemented;
353 	unsigned int enabled;
354 	int rc;
355 
356 	nic_data->workaround_35388 = false;
357 	nic_data->workaround_61265 = false;
358 
359 	rc = efx_mcdi_get_workarounds(efx, &implemented, &enabled);
360 
361 	if (rc == -ENOSYS) {
362 		/* Firmware without GET_WORKAROUNDS - not a problem. */
363 		rc = 0;
364 	} else if (rc == 0) {
365 		/* Bug61265 workaround is always enabled if implemented. */
366 		if (enabled & MC_CMD_GET_WORKAROUNDS_OUT_BUG61265)
367 			nic_data->workaround_61265 = true;
368 
369 		if (enabled & MC_CMD_GET_WORKAROUNDS_OUT_BUG35388) {
370 			nic_data->workaround_35388 = true;
371 		} else if (implemented & MC_CMD_GET_WORKAROUNDS_OUT_BUG35388) {
372 			/* Workaround is implemented but not enabled.
373 			 * Try to enable it.
374 			 */
375 			rc = efx_mcdi_set_workaround(efx,
376 						     MC_CMD_WORKAROUND_BUG35388,
377 						     true, NULL);
378 			if (rc == 0)
379 				nic_data->workaround_35388 = true;
380 			/* If we failed to set the workaround just carry on. */
381 			rc = 0;
382 		}
383 	}
384 
385 	netif_dbg(efx, probe, efx->net_dev,
386 		  "workaround for bug 35388 is %sabled\n",
387 		  nic_data->workaround_35388 ? "en" : "dis");
388 	netif_dbg(efx, probe, efx->net_dev,
389 		  "workaround for bug 61265 is %sabled\n",
390 		  nic_data->workaround_61265 ? "en" : "dis");
391 
392 	return rc;
393 }
394 
395 static void efx_ef10_process_timer_config(struct efx_nic *efx,
396 					  const efx_dword_t *data)
397 {
398 	unsigned int max_count;
399 
400 	if (EFX_EF10_WORKAROUND_61265(efx)) {
401 		efx->timer_quantum_ns = MCDI_DWORD(data,
402 			GET_EVQ_TMR_PROPERTIES_OUT_MCDI_TMR_STEP_NS);
403 		efx->timer_max_ns = MCDI_DWORD(data,
404 			GET_EVQ_TMR_PROPERTIES_OUT_MCDI_TMR_MAX_NS);
405 	} else if (EFX_EF10_WORKAROUND_35388(efx)) {
406 		efx->timer_quantum_ns = MCDI_DWORD(data,
407 			GET_EVQ_TMR_PROPERTIES_OUT_BUG35388_TMR_NS_PER_COUNT);
408 		max_count = MCDI_DWORD(data,
409 			GET_EVQ_TMR_PROPERTIES_OUT_BUG35388_TMR_MAX_COUNT);
410 		efx->timer_max_ns = max_count * efx->timer_quantum_ns;
411 	} else {
412 		efx->timer_quantum_ns = MCDI_DWORD(data,
413 			GET_EVQ_TMR_PROPERTIES_OUT_TMR_REG_NS_PER_COUNT);
414 		max_count = MCDI_DWORD(data,
415 			GET_EVQ_TMR_PROPERTIES_OUT_TMR_REG_MAX_COUNT);
416 		efx->timer_max_ns = max_count * efx->timer_quantum_ns;
417 	}
418 
419 	netif_dbg(efx, probe, efx->net_dev,
420 		  "got timer properties from MC: quantum %u ns; max %u ns\n",
421 		  efx->timer_quantum_ns, efx->timer_max_ns);
422 }
423 
424 static int efx_ef10_get_timer_config(struct efx_nic *efx)
425 {
426 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_EVQ_TMR_PROPERTIES_OUT_LEN);
427 	int rc;
428 
429 	rc = efx_ef10_get_timer_workarounds(efx);
430 	if (rc)
431 		return rc;
432 
433 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_GET_EVQ_TMR_PROPERTIES, NULL, 0,
434 				outbuf, sizeof(outbuf), NULL);
435 
436 	if (rc == 0) {
437 		efx_ef10_process_timer_config(efx, outbuf);
438 	} else if (rc == -ENOSYS || rc == -EPERM) {
439 		/* Not available - fall back to Huntington defaults. */
440 		unsigned int quantum;
441 
442 		rc = efx_ef10_get_sysclk_freq(efx);
443 		if (rc < 0)
444 			return rc;
445 
446 		quantum = 1536000 / rc; /* 1536 cycles */
447 		efx->timer_quantum_ns = quantum;
448 		efx->timer_max_ns = efx->type->timer_period_max * quantum;
449 		rc = 0;
450 	} else {
451 		efx_mcdi_display_error(efx, MC_CMD_GET_EVQ_TMR_PROPERTIES,
452 				       MC_CMD_GET_EVQ_TMR_PROPERTIES_OUT_LEN,
453 				       NULL, 0, rc);
454 	}
455 
456 	return rc;
457 }
458 
459 static int efx_ef10_get_mac_address_pf(struct efx_nic *efx, u8 *mac_address)
460 {
461 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_MAC_ADDRESSES_OUT_LEN);
462 	size_t outlen;
463 	int rc;
464 
465 	BUILD_BUG_ON(MC_CMD_GET_MAC_ADDRESSES_IN_LEN != 0);
466 
467 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_MAC_ADDRESSES, NULL, 0,
468 			  outbuf, sizeof(outbuf), &outlen);
469 	if (rc)
470 		return rc;
471 	if (outlen < MC_CMD_GET_MAC_ADDRESSES_OUT_LEN)
472 		return -EIO;
473 
474 	ether_addr_copy(mac_address,
475 			MCDI_PTR(outbuf, GET_MAC_ADDRESSES_OUT_MAC_ADDR_BASE));
476 	return 0;
477 }
478 
479 static int efx_ef10_get_mac_address_vf(struct efx_nic *efx, u8 *mac_address)
480 {
481 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VPORT_GET_MAC_ADDRESSES_IN_LEN);
482 	MCDI_DECLARE_BUF(outbuf, MC_CMD_VPORT_GET_MAC_ADDRESSES_OUT_LENMAX);
483 	size_t outlen;
484 	int num_addrs, rc;
485 
486 	MCDI_SET_DWORD(inbuf, VPORT_GET_MAC_ADDRESSES_IN_VPORT_ID,
487 		       EVB_PORT_ID_ASSIGNED);
488 	rc = efx_mcdi_rpc(efx, MC_CMD_VPORT_GET_MAC_ADDRESSES, inbuf,
489 			  sizeof(inbuf), outbuf, sizeof(outbuf), &outlen);
490 
491 	if (rc)
492 		return rc;
493 	if (outlen < MC_CMD_VPORT_GET_MAC_ADDRESSES_OUT_LENMIN)
494 		return -EIO;
495 
496 	num_addrs = MCDI_DWORD(outbuf,
497 			       VPORT_GET_MAC_ADDRESSES_OUT_MACADDR_COUNT);
498 
499 	WARN_ON(num_addrs != 1);
500 
501 	ether_addr_copy(mac_address,
502 			MCDI_PTR(outbuf, VPORT_GET_MAC_ADDRESSES_OUT_MACADDR));
503 
504 	return 0;
505 }
506 
507 static ssize_t efx_ef10_show_link_control_flag(struct device *dev,
508 					       struct device_attribute *attr,
509 					       char *buf)
510 {
511 	struct efx_nic *efx = dev_get_drvdata(dev);
512 
513 	return sprintf(buf, "%d\n",
514 		       ((efx->mcdi->fn_flags) &
515 			(1 << MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_LINKCTRL))
516 		       ? 1 : 0);
517 }
518 
519 static ssize_t efx_ef10_show_primary_flag(struct device *dev,
520 					  struct device_attribute *attr,
521 					  char *buf)
522 {
523 	struct efx_nic *efx = dev_get_drvdata(dev);
524 
525 	return sprintf(buf, "%d\n",
526 		       ((efx->mcdi->fn_flags) &
527 			(1 << MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_PRIMARY))
528 		       ? 1 : 0);
529 }
530 
531 static struct efx_ef10_vlan *efx_ef10_find_vlan(struct efx_nic *efx, u16 vid)
532 {
533 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
534 	struct efx_ef10_vlan *vlan;
535 
536 	WARN_ON(!mutex_is_locked(&nic_data->vlan_lock));
537 
538 	list_for_each_entry(vlan, &nic_data->vlan_list, list) {
539 		if (vlan->vid == vid)
540 			return vlan;
541 	}
542 
543 	return NULL;
544 }
545 
546 static int efx_ef10_add_vlan(struct efx_nic *efx, u16 vid)
547 {
548 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
549 	struct efx_ef10_vlan *vlan;
550 	int rc;
551 
552 	mutex_lock(&nic_data->vlan_lock);
553 
554 	vlan = efx_ef10_find_vlan(efx, vid);
555 	if (vlan) {
556 		/* We add VID 0 on init. 8021q adds it on module init
557 		 * for all interfaces with VLAN filtring feature.
558 		 */
559 		if (vid == 0)
560 			goto done_unlock;
561 		netif_warn(efx, drv, efx->net_dev,
562 			   "VLAN %u already added\n", vid);
563 		rc = -EALREADY;
564 		goto fail_exist;
565 	}
566 
567 	rc = -ENOMEM;
568 	vlan = kzalloc(sizeof(*vlan), GFP_KERNEL);
569 	if (!vlan)
570 		goto fail_alloc;
571 
572 	vlan->vid = vid;
573 
574 	list_add_tail(&vlan->list, &nic_data->vlan_list);
575 
576 	if (efx->filter_state) {
577 		mutex_lock(&efx->mac_lock);
578 		down_write(&efx->filter_sem);
579 		rc = efx_ef10_filter_add_vlan(efx, vlan->vid);
580 		up_write(&efx->filter_sem);
581 		mutex_unlock(&efx->mac_lock);
582 		if (rc)
583 			goto fail_filter_add_vlan;
584 	}
585 
586 done_unlock:
587 	mutex_unlock(&nic_data->vlan_lock);
588 	return 0;
589 
590 fail_filter_add_vlan:
591 	list_del(&vlan->list);
592 	kfree(vlan);
593 fail_alloc:
594 fail_exist:
595 	mutex_unlock(&nic_data->vlan_lock);
596 	return rc;
597 }
598 
599 static void efx_ef10_del_vlan_internal(struct efx_nic *efx,
600 				       struct efx_ef10_vlan *vlan)
601 {
602 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
603 
604 	WARN_ON(!mutex_is_locked(&nic_data->vlan_lock));
605 
606 	if (efx->filter_state) {
607 		down_write(&efx->filter_sem);
608 		efx_ef10_filter_del_vlan(efx, vlan->vid);
609 		up_write(&efx->filter_sem);
610 	}
611 
612 	list_del(&vlan->list);
613 	kfree(vlan);
614 }
615 
616 static int efx_ef10_del_vlan(struct efx_nic *efx, u16 vid)
617 {
618 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
619 	struct efx_ef10_vlan *vlan;
620 	int rc = 0;
621 
622 	/* 8021q removes VID 0 on module unload for all interfaces
623 	 * with VLAN filtering feature. We need to keep it to receive
624 	 * untagged traffic.
625 	 */
626 	if (vid == 0)
627 		return 0;
628 
629 	mutex_lock(&nic_data->vlan_lock);
630 
631 	vlan = efx_ef10_find_vlan(efx, vid);
632 	if (!vlan) {
633 		netif_err(efx, drv, efx->net_dev,
634 			  "VLAN %u to be deleted not found\n", vid);
635 		rc = -ENOENT;
636 	} else {
637 		efx_ef10_del_vlan_internal(efx, vlan);
638 	}
639 
640 	mutex_unlock(&nic_data->vlan_lock);
641 
642 	return rc;
643 }
644 
645 static void efx_ef10_cleanup_vlans(struct efx_nic *efx)
646 {
647 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
648 	struct efx_ef10_vlan *vlan, *next_vlan;
649 
650 	mutex_lock(&nic_data->vlan_lock);
651 	list_for_each_entry_safe(vlan, next_vlan, &nic_data->vlan_list, list)
652 		efx_ef10_del_vlan_internal(efx, vlan);
653 	mutex_unlock(&nic_data->vlan_lock);
654 }
655 
656 static DEVICE_ATTR(link_control_flag, 0444, efx_ef10_show_link_control_flag,
657 		   NULL);
658 static DEVICE_ATTR(primary_flag, 0444, efx_ef10_show_primary_flag, NULL);
659 
660 static int efx_ef10_probe(struct efx_nic *efx)
661 {
662 	struct efx_ef10_nic_data *nic_data;
663 	int i, rc;
664 
665 	nic_data = kzalloc(sizeof(*nic_data), GFP_KERNEL);
666 	if (!nic_data)
667 		return -ENOMEM;
668 	efx->nic_data = nic_data;
669 
670 	/* we assume later that we can copy from this buffer in dwords */
671 	BUILD_BUG_ON(MCDI_CTL_SDU_LEN_MAX_V2 % 4);
672 
673 	rc = efx_nic_alloc_buffer(efx, &nic_data->mcdi_buf,
674 				  8 + MCDI_CTL_SDU_LEN_MAX_V2, GFP_KERNEL);
675 	if (rc)
676 		goto fail1;
677 
678 	/* Get the MC's warm boot count.  In case it's rebooting right
679 	 * now, be prepared to retry.
680 	 */
681 	i = 0;
682 	for (;;) {
683 		rc = efx_ef10_get_warm_boot_count(efx);
684 		if (rc >= 0)
685 			break;
686 		if (++i == 5)
687 			goto fail2;
688 		ssleep(1);
689 	}
690 	nic_data->warm_boot_count = rc;
691 
692 	efx->rss_context.context_id = EFX_EF10_RSS_CONTEXT_INVALID;
693 
694 	nic_data->vport_id = EVB_PORT_ID_ASSIGNED;
695 
696 	/* In case we're recovering from a crash (kexec), we want to
697 	 * cancel any outstanding request by the previous user of this
698 	 * function.  We send a special message using the least
699 	 * significant bits of the 'high' (doorbell) register.
700 	 */
701 	_efx_writed(efx, cpu_to_le32(1), ER_DZ_MC_DB_HWRD);
702 
703 	rc = efx_mcdi_init(efx);
704 	if (rc)
705 		goto fail2;
706 
707 	mutex_init(&nic_data->udp_tunnels_lock);
708 
709 	/* Reset (most) configuration for this function */
710 	rc = efx_mcdi_reset(efx, RESET_TYPE_ALL);
711 	if (rc)
712 		goto fail3;
713 
714 	/* Enable event logging */
715 	rc = efx_mcdi_log_ctrl(efx, true, false, 0);
716 	if (rc)
717 		goto fail3;
718 
719 	rc = device_create_file(&efx->pci_dev->dev,
720 				&dev_attr_link_control_flag);
721 	if (rc)
722 		goto fail3;
723 
724 	rc = device_create_file(&efx->pci_dev->dev, &dev_attr_primary_flag);
725 	if (rc)
726 		goto fail4;
727 
728 	rc = efx_ef10_get_pf_index(efx);
729 	if (rc)
730 		goto fail5;
731 
732 	rc = efx_ef10_init_datapath_caps(efx);
733 	if (rc < 0)
734 		goto fail5;
735 
736 	efx_ef10_read_licensed_features(efx);
737 
738 	/* We can have one VI for each vi_stride-byte region.
739 	 * However, until we use TX option descriptors we need two TX queues
740 	 * per channel.
741 	 */
742 	efx->max_channels = min_t(unsigned int,
743 				  EFX_MAX_CHANNELS,
744 				  efx_ef10_mem_map_size(efx) /
745 				  (efx->vi_stride * EFX_TXQ_TYPES));
746 	efx->max_tx_channels = efx->max_channels;
747 	if (WARN_ON(efx->max_channels == 0)) {
748 		rc = -EIO;
749 		goto fail5;
750 	}
751 
752 	efx->rx_packet_len_offset =
753 		ES_DZ_RX_PREFIX_PKTLEN_OFST - ES_DZ_RX_PREFIX_SIZE;
754 
755 	if (nic_data->datapath_caps &
756 	    (1 << MC_CMD_GET_CAPABILITIES_OUT_RX_INCLUDE_FCS_LBN))
757 		efx->net_dev->hw_features |= NETIF_F_RXFCS;
758 
759 	rc = efx_mcdi_port_get_number(efx);
760 	if (rc < 0)
761 		goto fail5;
762 	efx->port_num = rc;
763 
764 	rc = efx->type->get_mac_address(efx, efx->net_dev->perm_addr);
765 	if (rc)
766 		goto fail5;
767 
768 	rc = efx_ef10_get_timer_config(efx);
769 	if (rc < 0)
770 		goto fail5;
771 
772 	rc = efx_mcdi_mon_probe(efx);
773 	if (rc && rc != -EPERM)
774 		goto fail5;
775 
776 	efx_ptp_defer_probe_with_channel(efx);
777 
778 #ifdef CONFIG_SFC_SRIOV
779 	if ((efx->pci_dev->physfn) && (!efx->pci_dev->is_physfn)) {
780 		struct pci_dev *pci_dev_pf = efx->pci_dev->physfn;
781 		struct efx_nic *efx_pf = pci_get_drvdata(pci_dev_pf);
782 
783 		efx_pf->type->get_mac_address(efx_pf, nic_data->port_id);
784 	} else
785 #endif
786 		ether_addr_copy(nic_data->port_id, efx->net_dev->perm_addr);
787 
788 	INIT_LIST_HEAD(&nic_data->vlan_list);
789 	mutex_init(&nic_data->vlan_lock);
790 
791 	/* Add unspecified VID to support VLAN filtering being disabled */
792 	rc = efx_ef10_add_vlan(efx, EFX_FILTER_VID_UNSPEC);
793 	if (rc)
794 		goto fail_add_vid_unspec;
795 
796 	/* If VLAN filtering is enabled, we need VID 0 to get untagged
797 	 * traffic.  It is added automatically if 8021q module is loaded,
798 	 * but we can't rely on it since module may be not loaded.
799 	 */
800 	rc = efx_ef10_add_vlan(efx, 0);
801 	if (rc)
802 		goto fail_add_vid_0;
803 
804 	return 0;
805 
806 fail_add_vid_0:
807 	efx_ef10_cleanup_vlans(efx);
808 fail_add_vid_unspec:
809 	mutex_destroy(&nic_data->vlan_lock);
810 	efx_ptp_remove(efx);
811 	efx_mcdi_mon_remove(efx);
812 fail5:
813 	device_remove_file(&efx->pci_dev->dev, &dev_attr_primary_flag);
814 fail4:
815 	device_remove_file(&efx->pci_dev->dev, &dev_attr_link_control_flag);
816 fail3:
817 	efx_mcdi_detach(efx);
818 
819 	mutex_lock(&nic_data->udp_tunnels_lock);
820 	memset(nic_data->udp_tunnels, 0, sizeof(nic_data->udp_tunnels));
821 	(void)efx_ef10_set_udp_tnl_ports(efx, true);
822 	mutex_unlock(&nic_data->udp_tunnels_lock);
823 	mutex_destroy(&nic_data->udp_tunnels_lock);
824 
825 	efx_mcdi_fini(efx);
826 fail2:
827 	efx_nic_free_buffer(efx, &nic_data->mcdi_buf);
828 fail1:
829 	kfree(nic_data);
830 	efx->nic_data = NULL;
831 	return rc;
832 }
833 
834 static int efx_ef10_free_vis(struct efx_nic *efx)
835 {
836 	MCDI_DECLARE_BUF_ERR(outbuf);
837 	size_t outlen;
838 	int rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FREE_VIS, NULL, 0,
839 				    outbuf, sizeof(outbuf), &outlen);
840 
841 	/* -EALREADY means nothing to free, so ignore */
842 	if (rc == -EALREADY)
843 		rc = 0;
844 	if (rc)
845 		efx_mcdi_display_error(efx, MC_CMD_FREE_VIS, 0, outbuf, outlen,
846 				       rc);
847 	return rc;
848 }
849 
850 #ifdef EFX_USE_PIO
851 
852 static void efx_ef10_free_piobufs(struct efx_nic *efx)
853 {
854 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
855 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FREE_PIOBUF_IN_LEN);
856 	unsigned int i;
857 	int rc;
858 
859 	BUILD_BUG_ON(MC_CMD_FREE_PIOBUF_OUT_LEN != 0);
860 
861 	for (i = 0; i < nic_data->n_piobufs; i++) {
862 		MCDI_SET_DWORD(inbuf, FREE_PIOBUF_IN_PIOBUF_HANDLE,
863 			       nic_data->piobuf_handle[i]);
864 		rc = efx_mcdi_rpc(efx, MC_CMD_FREE_PIOBUF, inbuf, sizeof(inbuf),
865 				  NULL, 0, NULL);
866 		WARN_ON(rc);
867 	}
868 
869 	nic_data->n_piobufs = 0;
870 }
871 
872 static int efx_ef10_alloc_piobufs(struct efx_nic *efx, unsigned int n)
873 {
874 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
875 	MCDI_DECLARE_BUF(outbuf, MC_CMD_ALLOC_PIOBUF_OUT_LEN);
876 	unsigned int i;
877 	size_t outlen;
878 	int rc = 0;
879 
880 	BUILD_BUG_ON(MC_CMD_ALLOC_PIOBUF_IN_LEN != 0);
881 
882 	for (i = 0; i < n; i++) {
883 		rc = efx_mcdi_rpc_quiet(efx, MC_CMD_ALLOC_PIOBUF, NULL, 0,
884 					outbuf, sizeof(outbuf), &outlen);
885 		if (rc) {
886 			/* Don't display the MC error if we didn't have space
887 			 * for a VF.
888 			 */
889 			if (!(efx_ef10_is_vf(efx) && rc == -ENOSPC))
890 				efx_mcdi_display_error(efx, MC_CMD_ALLOC_PIOBUF,
891 						       0, outbuf, outlen, rc);
892 			break;
893 		}
894 		if (outlen < MC_CMD_ALLOC_PIOBUF_OUT_LEN) {
895 			rc = -EIO;
896 			break;
897 		}
898 		nic_data->piobuf_handle[i] =
899 			MCDI_DWORD(outbuf, ALLOC_PIOBUF_OUT_PIOBUF_HANDLE);
900 		netif_dbg(efx, probe, efx->net_dev,
901 			  "allocated PIO buffer %u handle %x\n", i,
902 			  nic_data->piobuf_handle[i]);
903 	}
904 
905 	nic_data->n_piobufs = i;
906 	if (rc)
907 		efx_ef10_free_piobufs(efx);
908 	return rc;
909 }
910 
911 static int efx_ef10_link_piobufs(struct efx_nic *efx)
912 {
913 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
914 	MCDI_DECLARE_BUF(inbuf, MC_CMD_LINK_PIOBUF_IN_LEN);
915 	struct efx_channel *channel;
916 	struct efx_tx_queue *tx_queue;
917 	unsigned int offset, index;
918 	int rc;
919 
920 	BUILD_BUG_ON(MC_CMD_LINK_PIOBUF_OUT_LEN != 0);
921 	BUILD_BUG_ON(MC_CMD_UNLINK_PIOBUF_OUT_LEN != 0);
922 
923 	/* Link a buffer to each VI in the write-combining mapping */
924 	for (index = 0; index < nic_data->n_piobufs; ++index) {
925 		MCDI_SET_DWORD(inbuf, LINK_PIOBUF_IN_PIOBUF_HANDLE,
926 			       nic_data->piobuf_handle[index]);
927 		MCDI_SET_DWORD(inbuf, LINK_PIOBUF_IN_TXQ_INSTANCE,
928 			       nic_data->pio_write_vi_base + index);
929 		rc = efx_mcdi_rpc(efx, MC_CMD_LINK_PIOBUF,
930 				  inbuf, MC_CMD_LINK_PIOBUF_IN_LEN,
931 				  NULL, 0, NULL);
932 		if (rc) {
933 			netif_err(efx, drv, efx->net_dev,
934 				  "failed to link VI %u to PIO buffer %u (%d)\n",
935 				  nic_data->pio_write_vi_base + index, index,
936 				  rc);
937 			goto fail;
938 		}
939 		netif_dbg(efx, probe, efx->net_dev,
940 			  "linked VI %u to PIO buffer %u\n",
941 			  nic_data->pio_write_vi_base + index, index);
942 	}
943 
944 	/* Link a buffer to each TX queue */
945 	efx_for_each_channel(channel, efx) {
946 		/* Extra channels, even those with TXQs (PTP), do not require
947 		 * PIO resources.
948 		 */
949 		if (!channel->type->want_pio)
950 			continue;
951 		efx_for_each_channel_tx_queue(tx_queue, channel) {
952 			/* We assign the PIO buffers to queues in
953 			 * reverse order to allow for the following
954 			 * special case.
955 			 */
956 			offset = ((efx->tx_channel_offset + efx->n_tx_channels -
957 				   tx_queue->channel->channel - 1) *
958 				  efx_piobuf_size);
959 			index = offset / nic_data->piobuf_size;
960 			offset = offset % nic_data->piobuf_size;
961 
962 			/* When the host page size is 4K, the first
963 			 * host page in the WC mapping may be within
964 			 * the same VI page as the last TX queue.  We
965 			 * can only link one buffer to each VI.
966 			 */
967 			if (tx_queue->queue == nic_data->pio_write_vi_base) {
968 				BUG_ON(index != 0);
969 				rc = 0;
970 			} else {
971 				MCDI_SET_DWORD(inbuf,
972 					       LINK_PIOBUF_IN_PIOBUF_HANDLE,
973 					       nic_data->piobuf_handle[index]);
974 				MCDI_SET_DWORD(inbuf,
975 					       LINK_PIOBUF_IN_TXQ_INSTANCE,
976 					       tx_queue->queue);
977 				rc = efx_mcdi_rpc(efx, MC_CMD_LINK_PIOBUF,
978 						  inbuf, MC_CMD_LINK_PIOBUF_IN_LEN,
979 						  NULL, 0, NULL);
980 			}
981 
982 			if (rc) {
983 				/* This is non-fatal; the TX path just
984 				 * won't use PIO for this queue
985 				 */
986 				netif_err(efx, drv, efx->net_dev,
987 					  "failed to link VI %u to PIO buffer %u (%d)\n",
988 					  tx_queue->queue, index, rc);
989 				tx_queue->piobuf = NULL;
990 			} else {
991 				tx_queue->piobuf =
992 					nic_data->pio_write_base +
993 					index * efx->vi_stride + offset;
994 				tx_queue->piobuf_offset = offset;
995 				netif_dbg(efx, probe, efx->net_dev,
996 					  "linked VI %u to PIO buffer %u offset %x addr %p\n",
997 					  tx_queue->queue, index,
998 					  tx_queue->piobuf_offset,
999 					  tx_queue->piobuf);
1000 			}
1001 		}
1002 	}
1003 
1004 	return 0;
1005 
1006 fail:
1007 	/* inbuf was defined for MC_CMD_LINK_PIOBUF.  We can use the same
1008 	 * buffer for MC_CMD_UNLINK_PIOBUF because it's shorter.
1009 	 */
1010 	BUILD_BUG_ON(MC_CMD_LINK_PIOBUF_IN_LEN < MC_CMD_UNLINK_PIOBUF_IN_LEN);
1011 	while (index--) {
1012 		MCDI_SET_DWORD(inbuf, UNLINK_PIOBUF_IN_TXQ_INSTANCE,
1013 			       nic_data->pio_write_vi_base + index);
1014 		efx_mcdi_rpc(efx, MC_CMD_UNLINK_PIOBUF,
1015 			     inbuf, MC_CMD_UNLINK_PIOBUF_IN_LEN,
1016 			     NULL, 0, NULL);
1017 	}
1018 	return rc;
1019 }
1020 
1021 static void efx_ef10_forget_old_piobufs(struct efx_nic *efx)
1022 {
1023 	struct efx_channel *channel;
1024 	struct efx_tx_queue *tx_queue;
1025 
1026 	/* All our existing PIO buffers went away */
1027 	efx_for_each_channel(channel, efx)
1028 		efx_for_each_channel_tx_queue(tx_queue, channel)
1029 			tx_queue->piobuf = NULL;
1030 }
1031 
1032 #else /* !EFX_USE_PIO */
1033 
1034 static int efx_ef10_alloc_piobufs(struct efx_nic *efx, unsigned int n)
1035 {
1036 	return n == 0 ? 0 : -ENOBUFS;
1037 }
1038 
1039 static int efx_ef10_link_piobufs(struct efx_nic *efx)
1040 {
1041 	return 0;
1042 }
1043 
1044 static void efx_ef10_free_piobufs(struct efx_nic *efx)
1045 {
1046 }
1047 
1048 static void efx_ef10_forget_old_piobufs(struct efx_nic *efx)
1049 {
1050 }
1051 
1052 #endif /* EFX_USE_PIO */
1053 
1054 static void efx_ef10_remove(struct efx_nic *efx)
1055 {
1056 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1057 	int rc;
1058 
1059 #ifdef CONFIG_SFC_SRIOV
1060 	struct efx_ef10_nic_data *nic_data_pf;
1061 	struct pci_dev *pci_dev_pf;
1062 	struct efx_nic *efx_pf;
1063 	struct ef10_vf *vf;
1064 
1065 	if (efx->pci_dev->is_virtfn) {
1066 		pci_dev_pf = efx->pci_dev->physfn;
1067 		if (pci_dev_pf) {
1068 			efx_pf = pci_get_drvdata(pci_dev_pf);
1069 			nic_data_pf = efx_pf->nic_data;
1070 			vf = nic_data_pf->vf + nic_data->vf_index;
1071 			vf->efx = NULL;
1072 		} else
1073 			netif_info(efx, drv, efx->net_dev,
1074 				   "Could not get the PF id from VF\n");
1075 	}
1076 #endif
1077 
1078 	efx_ef10_cleanup_vlans(efx);
1079 	mutex_destroy(&nic_data->vlan_lock);
1080 
1081 	efx_ptp_remove(efx);
1082 
1083 	efx_mcdi_mon_remove(efx);
1084 
1085 	efx_ef10_rx_free_indir_table(efx);
1086 
1087 	if (nic_data->wc_membase)
1088 		iounmap(nic_data->wc_membase);
1089 
1090 	rc = efx_ef10_free_vis(efx);
1091 	WARN_ON(rc != 0);
1092 
1093 	if (!nic_data->must_restore_piobufs)
1094 		efx_ef10_free_piobufs(efx);
1095 
1096 	device_remove_file(&efx->pci_dev->dev, &dev_attr_primary_flag);
1097 	device_remove_file(&efx->pci_dev->dev, &dev_attr_link_control_flag);
1098 
1099 	efx_mcdi_detach(efx);
1100 
1101 	memset(nic_data->udp_tunnels, 0, sizeof(nic_data->udp_tunnels));
1102 	mutex_lock(&nic_data->udp_tunnels_lock);
1103 	(void)efx_ef10_set_udp_tnl_ports(efx, true);
1104 	mutex_unlock(&nic_data->udp_tunnels_lock);
1105 
1106 	mutex_destroy(&nic_data->udp_tunnels_lock);
1107 
1108 	efx_mcdi_fini(efx);
1109 	efx_nic_free_buffer(efx, &nic_data->mcdi_buf);
1110 	kfree(nic_data);
1111 }
1112 
1113 static int efx_ef10_probe_pf(struct efx_nic *efx)
1114 {
1115 	return efx_ef10_probe(efx);
1116 }
1117 
1118 int efx_ef10_vadaptor_query(struct efx_nic *efx, unsigned int port_id,
1119 			    u32 *port_flags, u32 *vadaptor_flags,
1120 			    unsigned int *vlan_tags)
1121 {
1122 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1123 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VADAPTOR_QUERY_IN_LEN);
1124 	MCDI_DECLARE_BUF(outbuf, MC_CMD_VADAPTOR_QUERY_OUT_LEN);
1125 	size_t outlen;
1126 	int rc;
1127 
1128 	if (nic_data->datapath_caps &
1129 	    (1 << MC_CMD_GET_CAPABILITIES_OUT_VADAPTOR_QUERY_LBN)) {
1130 		MCDI_SET_DWORD(inbuf, VADAPTOR_QUERY_IN_UPSTREAM_PORT_ID,
1131 			       port_id);
1132 
1133 		rc = efx_mcdi_rpc(efx, MC_CMD_VADAPTOR_QUERY, inbuf, sizeof(inbuf),
1134 				  outbuf, sizeof(outbuf), &outlen);
1135 		if (rc)
1136 			return rc;
1137 
1138 		if (outlen < sizeof(outbuf)) {
1139 			rc = -EIO;
1140 			return rc;
1141 		}
1142 	}
1143 
1144 	if (port_flags)
1145 		*port_flags = MCDI_DWORD(outbuf, VADAPTOR_QUERY_OUT_PORT_FLAGS);
1146 	if (vadaptor_flags)
1147 		*vadaptor_flags =
1148 			MCDI_DWORD(outbuf, VADAPTOR_QUERY_OUT_VADAPTOR_FLAGS);
1149 	if (vlan_tags)
1150 		*vlan_tags =
1151 			MCDI_DWORD(outbuf,
1152 				   VADAPTOR_QUERY_OUT_NUM_AVAILABLE_VLAN_TAGS);
1153 
1154 	return 0;
1155 }
1156 
1157 int efx_ef10_vadaptor_alloc(struct efx_nic *efx, unsigned int port_id)
1158 {
1159 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VADAPTOR_ALLOC_IN_LEN);
1160 
1161 	MCDI_SET_DWORD(inbuf, VADAPTOR_ALLOC_IN_UPSTREAM_PORT_ID, port_id);
1162 	return efx_mcdi_rpc(efx, MC_CMD_VADAPTOR_ALLOC, inbuf, sizeof(inbuf),
1163 			    NULL, 0, NULL);
1164 }
1165 
1166 int efx_ef10_vadaptor_free(struct efx_nic *efx, unsigned int port_id)
1167 {
1168 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VADAPTOR_FREE_IN_LEN);
1169 
1170 	MCDI_SET_DWORD(inbuf, VADAPTOR_FREE_IN_UPSTREAM_PORT_ID, port_id);
1171 	return efx_mcdi_rpc(efx, MC_CMD_VADAPTOR_FREE, inbuf, sizeof(inbuf),
1172 			    NULL, 0, NULL);
1173 }
1174 
1175 int efx_ef10_vport_add_mac(struct efx_nic *efx,
1176 			   unsigned int port_id, u8 *mac)
1177 {
1178 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VPORT_ADD_MAC_ADDRESS_IN_LEN);
1179 
1180 	MCDI_SET_DWORD(inbuf, VPORT_ADD_MAC_ADDRESS_IN_VPORT_ID, port_id);
1181 	ether_addr_copy(MCDI_PTR(inbuf, VPORT_ADD_MAC_ADDRESS_IN_MACADDR), mac);
1182 
1183 	return efx_mcdi_rpc(efx, MC_CMD_VPORT_ADD_MAC_ADDRESS, inbuf,
1184 			    sizeof(inbuf), NULL, 0, NULL);
1185 }
1186 
1187 int efx_ef10_vport_del_mac(struct efx_nic *efx,
1188 			   unsigned int port_id, u8 *mac)
1189 {
1190 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VPORT_DEL_MAC_ADDRESS_IN_LEN);
1191 
1192 	MCDI_SET_DWORD(inbuf, VPORT_DEL_MAC_ADDRESS_IN_VPORT_ID, port_id);
1193 	ether_addr_copy(MCDI_PTR(inbuf, VPORT_DEL_MAC_ADDRESS_IN_MACADDR), mac);
1194 
1195 	return efx_mcdi_rpc(efx, MC_CMD_VPORT_DEL_MAC_ADDRESS, inbuf,
1196 			    sizeof(inbuf), NULL, 0, NULL);
1197 }
1198 
1199 #ifdef CONFIG_SFC_SRIOV
1200 static int efx_ef10_probe_vf(struct efx_nic *efx)
1201 {
1202 	int rc;
1203 	struct pci_dev *pci_dev_pf;
1204 
1205 	/* If the parent PF has no VF data structure, it doesn't know about this
1206 	 * VF so fail probe.  The VF needs to be re-created.  This can happen
1207 	 * if the PF driver is unloaded while the VF is assigned to a guest.
1208 	 */
1209 	pci_dev_pf = efx->pci_dev->physfn;
1210 	if (pci_dev_pf) {
1211 		struct efx_nic *efx_pf = pci_get_drvdata(pci_dev_pf);
1212 		struct efx_ef10_nic_data *nic_data_pf = efx_pf->nic_data;
1213 
1214 		if (!nic_data_pf->vf) {
1215 			netif_info(efx, drv, efx->net_dev,
1216 				   "The VF cannot link to its parent PF; "
1217 				   "please destroy and re-create the VF\n");
1218 			return -EBUSY;
1219 		}
1220 	}
1221 
1222 	rc = efx_ef10_probe(efx);
1223 	if (rc)
1224 		return rc;
1225 
1226 	rc = efx_ef10_get_vf_index(efx);
1227 	if (rc)
1228 		goto fail;
1229 
1230 	if (efx->pci_dev->is_virtfn) {
1231 		if (efx->pci_dev->physfn) {
1232 			struct efx_nic *efx_pf =
1233 				pci_get_drvdata(efx->pci_dev->physfn);
1234 			struct efx_ef10_nic_data *nic_data_p = efx_pf->nic_data;
1235 			struct efx_ef10_nic_data *nic_data = efx->nic_data;
1236 
1237 			nic_data_p->vf[nic_data->vf_index].efx = efx;
1238 			nic_data_p->vf[nic_data->vf_index].pci_dev =
1239 				efx->pci_dev;
1240 		} else
1241 			netif_info(efx, drv, efx->net_dev,
1242 				   "Could not get the PF id from VF\n");
1243 	}
1244 
1245 	return 0;
1246 
1247 fail:
1248 	efx_ef10_remove(efx);
1249 	return rc;
1250 }
1251 #else
1252 static int efx_ef10_probe_vf(struct efx_nic *efx __attribute__ ((unused)))
1253 {
1254 	return 0;
1255 }
1256 #endif
1257 
1258 static int efx_ef10_alloc_vis(struct efx_nic *efx,
1259 			      unsigned int min_vis, unsigned int max_vis)
1260 {
1261 	MCDI_DECLARE_BUF(inbuf, MC_CMD_ALLOC_VIS_IN_LEN);
1262 	MCDI_DECLARE_BUF(outbuf, MC_CMD_ALLOC_VIS_OUT_LEN);
1263 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1264 	size_t outlen;
1265 	int rc;
1266 
1267 	MCDI_SET_DWORD(inbuf, ALLOC_VIS_IN_MIN_VI_COUNT, min_vis);
1268 	MCDI_SET_DWORD(inbuf, ALLOC_VIS_IN_MAX_VI_COUNT, max_vis);
1269 	rc = efx_mcdi_rpc(efx, MC_CMD_ALLOC_VIS, inbuf, sizeof(inbuf),
1270 			  outbuf, sizeof(outbuf), &outlen);
1271 	if (rc != 0)
1272 		return rc;
1273 
1274 	if (outlen < MC_CMD_ALLOC_VIS_OUT_LEN)
1275 		return -EIO;
1276 
1277 	netif_dbg(efx, drv, efx->net_dev, "base VI is A0x%03x\n",
1278 		  MCDI_DWORD(outbuf, ALLOC_VIS_OUT_VI_BASE));
1279 
1280 	nic_data->vi_base = MCDI_DWORD(outbuf, ALLOC_VIS_OUT_VI_BASE);
1281 	nic_data->n_allocated_vis = MCDI_DWORD(outbuf, ALLOC_VIS_OUT_VI_COUNT);
1282 	return 0;
1283 }
1284 
1285 /* Note that the failure path of this function does not free
1286  * resources, as this will be done by efx_ef10_remove().
1287  */
1288 static int efx_ef10_dimension_resources(struct efx_nic *efx)
1289 {
1290 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1291 	unsigned int uc_mem_map_size, wc_mem_map_size;
1292 	unsigned int min_vis = max(EFX_TXQ_TYPES,
1293 				   efx_separate_tx_channels ? 2 : 1);
1294 	unsigned int channel_vis, pio_write_vi_base, max_vis;
1295 	void __iomem *membase;
1296 	int rc;
1297 
1298 	channel_vis = max(efx->n_channels,
1299 			  (efx->n_tx_channels + efx->n_extra_tx_channels) *
1300 			  EFX_TXQ_TYPES);
1301 
1302 #ifdef EFX_USE_PIO
1303 	/* Try to allocate PIO buffers if wanted and if the full
1304 	 * number of PIO buffers would be sufficient to allocate one
1305 	 * copy-buffer per TX channel.  Failure is non-fatal, as there
1306 	 * are only a small number of PIO buffers shared between all
1307 	 * functions of the controller.
1308 	 */
1309 	if (efx_piobuf_size != 0 &&
1310 	    nic_data->piobuf_size / efx_piobuf_size * EF10_TX_PIOBUF_COUNT >=
1311 	    efx->n_tx_channels) {
1312 		unsigned int n_piobufs =
1313 			DIV_ROUND_UP(efx->n_tx_channels,
1314 				     nic_data->piobuf_size / efx_piobuf_size);
1315 
1316 		rc = efx_ef10_alloc_piobufs(efx, n_piobufs);
1317 		if (rc == -ENOSPC)
1318 			netif_dbg(efx, probe, efx->net_dev,
1319 				  "out of PIO buffers; cannot allocate more\n");
1320 		else if (rc == -EPERM)
1321 			netif_dbg(efx, probe, efx->net_dev,
1322 				  "not permitted to allocate PIO buffers\n");
1323 		else if (rc)
1324 			netif_err(efx, probe, efx->net_dev,
1325 				  "failed to allocate PIO buffers (%d)\n", rc);
1326 		else
1327 			netif_dbg(efx, probe, efx->net_dev,
1328 				  "allocated %u PIO buffers\n", n_piobufs);
1329 	}
1330 #else
1331 	nic_data->n_piobufs = 0;
1332 #endif
1333 
1334 	/* PIO buffers should be mapped with write-combining enabled,
1335 	 * and we want to make single UC and WC mappings rather than
1336 	 * several of each (in fact that's the only option if host
1337 	 * page size is >4K).  So we may allocate some extra VIs just
1338 	 * for writing PIO buffers through.
1339 	 *
1340 	 * The UC mapping contains (channel_vis - 1) complete VIs and the
1341 	 * first 4K of the next VI.  Then the WC mapping begins with
1342 	 * the remainder of this last VI.
1343 	 */
1344 	uc_mem_map_size = PAGE_ALIGN((channel_vis - 1) * efx->vi_stride +
1345 				     ER_DZ_TX_PIOBUF);
1346 	if (nic_data->n_piobufs) {
1347 		/* pio_write_vi_base rounds down to give the number of complete
1348 		 * VIs inside the UC mapping.
1349 		 */
1350 		pio_write_vi_base = uc_mem_map_size / efx->vi_stride;
1351 		wc_mem_map_size = (PAGE_ALIGN((pio_write_vi_base +
1352 					       nic_data->n_piobufs) *
1353 					      efx->vi_stride) -
1354 				   uc_mem_map_size);
1355 		max_vis = pio_write_vi_base + nic_data->n_piobufs;
1356 	} else {
1357 		pio_write_vi_base = 0;
1358 		wc_mem_map_size = 0;
1359 		max_vis = channel_vis;
1360 	}
1361 
1362 	/* In case the last attached driver failed to free VIs, do it now */
1363 	rc = efx_ef10_free_vis(efx);
1364 	if (rc != 0)
1365 		return rc;
1366 
1367 	rc = efx_ef10_alloc_vis(efx, min_vis, max_vis);
1368 	if (rc != 0)
1369 		return rc;
1370 
1371 	if (nic_data->n_allocated_vis < channel_vis) {
1372 		netif_info(efx, drv, efx->net_dev,
1373 			   "Could not allocate enough VIs to satisfy RSS"
1374 			   " requirements. Performance may not be optimal.\n");
1375 		/* We didn't get the VIs to populate our channels.
1376 		 * We could keep what we got but then we'd have more
1377 		 * interrupts than we need.
1378 		 * Instead calculate new max_channels and restart
1379 		 */
1380 		efx->max_channels = nic_data->n_allocated_vis;
1381 		efx->max_tx_channels =
1382 			nic_data->n_allocated_vis / EFX_TXQ_TYPES;
1383 
1384 		efx_ef10_free_vis(efx);
1385 		return -EAGAIN;
1386 	}
1387 
1388 	/* If we didn't get enough VIs to map all the PIO buffers, free the
1389 	 * PIO buffers
1390 	 */
1391 	if (nic_data->n_piobufs &&
1392 	    nic_data->n_allocated_vis <
1393 	    pio_write_vi_base + nic_data->n_piobufs) {
1394 		netif_dbg(efx, probe, efx->net_dev,
1395 			  "%u VIs are not sufficient to map %u PIO buffers\n",
1396 			  nic_data->n_allocated_vis, nic_data->n_piobufs);
1397 		efx_ef10_free_piobufs(efx);
1398 	}
1399 
1400 	/* Shrink the original UC mapping of the memory BAR */
1401 	membase = ioremap_nocache(efx->membase_phys, uc_mem_map_size);
1402 	if (!membase) {
1403 		netif_err(efx, probe, efx->net_dev,
1404 			  "could not shrink memory BAR to %x\n",
1405 			  uc_mem_map_size);
1406 		return -ENOMEM;
1407 	}
1408 	iounmap(efx->membase);
1409 	efx->membase = membase;
1410 
1411 	/* Set up the WC mapping if needed */
1412 	if (wc_mem_map_size) {
1413 		nic_data->wc_membase = ioremap_wc(efx->membase_phys +
1414 						  uc_mem_map_size,
1415 						  wc_mem_map_size);
1416 		if (!nic_data->wc_membase) {
1417 			netif_err(efx, probe, efx->net_dev,
1418 				  "could not allocate WC mapping of size %x\n",
1419 				  wc_mem_map_size);
1420 			return -ENOMEM;
1421 		}
1422 		nic_data->pio_write_vi_base = pio_write_vi_base;
1423 		nic_data->pio_write_base =
1424 			nic_data->wc_membase +
1425 			(pio_write_vi_base * efx->vi_stride + ER_DZ_TX_PIOBUF -
1426 			 uc_mem_map_size);
1427 
1428 		rc = efx_ef10_link_piobufs(efx);
1429 		if (rc)
1430 			efx_ef10_free_piobufs(efx);
1431 	}
1432 
1433 	netif_dbg(efx, probe, efx->net_dev,
1434 		  "memory BAR at %pa (virtual %p+%x UC, %p+%x WC)\n",
1435 		  &efx->membase_phys, efx->membase, uc_mem_map_size,
1436 		  nic_data->wc_membase, wc_mem_map_size);
1437 
1438 	return 0;
1439 }
1440 
1441 static int efx_ef10_init_nic(struct efx_nic *efx)
1442 {
1443 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1444 	int rc;
1445 
1446 	if (nic_data->must_check_datapath_caps) {
1447 		rc = efx_ef10_init_datapath_caps(efx);
1448 		if (rc)
1449 			return rc;
1450 		nic_data->must_check_datapath_caps = false;
1451 	}
1452 
1453 	if (nic_data->must_realloc_vis) {
1454 		/* We cannot let the number of VIs change now */
1455 		rc = efx_ef10_alloc_vis(efx, nic_data->n_allocated_vis,
1456 					nic_data->n_allocated_vis);
1457 		if (rc)
1458 			return rc;
1459 		nic_data->must_realloc_vis = false;
1460 	}
1461 
1462 	if (nic_data->must_restore_piobufs && nic_data->n_piobufs) {
1463 		rc = efx_ef10_alloc_piobufs(efx, nic_data->n_piobufs);
1464 		if (rc == 0) {
1465 			rc = efx_ef10_link_piobufs(efx);
1466 			if (rc)
1467 				efx_ef10_free_piobufs(efx);
1468 		}
1469 
1470 		/* Log an error on failure, but this is non-fatal.
1471 		 * Permission errors are less important - we've presumably
1472 		 * had the PIO buffer licence removed.
1473 		 */
1474 		if (rc == -EPERM)
1475 			netif_dbg(efx, drv, efx->net_dev,
1476 				  "not permitted to restore PIO buffers\n");
1477 		else if (rc)
1478 			netif_err(efx, drv, efx->net_dev,
1479 				  "failed to restore PIO buffers (%d)\n", rc);
1480 		nic_data->must_restore_piobufs = false;
1481 	}
1482 
1483 	/* don't fail init if RSS setup doesn't work */
1484 	rc = efx->type->rx_push_rss_config(efx, false,
1485 					   efx->rss_context.rx_indir_table, NULL);
1486 
1487 	return 0;
1488 }
1489 
1490 static void efx_ef10_reset_mc_allocations(struct efx_nic *efx)
1491 {
1492 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1493 #ifdef CONFIG_SFC_SRIOV
1494 	unsigned int i;
1495 #endif
1496 
1497 	/* All our allocations have been reset */
1498 	nic_data->must_realloc_vis = true;
1499 	nic_data->must_restore_rss_contexts = true;
1500 	nic_data->must_restore_filters = true;
1501 	nic_data->must_restore_piobufs = true;
1502 	efx_ef10_forget_old_piobufs(efx);
1503 	efx->rss_context.context_id = EFX_EF10_RSS_CONTEXT_INVALID;
1504 
1505 	/* Driver-created vswitches and vports must be re-created */
1506 	nic_data->must_probe_vswitching = true;
1507 	nic_data->vport_id = EVB_PORT_ID_ASSIGNED;
1508 #ifdef CONFIG_SFC_SRIOV
1509 	if (nic_data->vf)
1510 		for (i = 0; i < efx->vf_count; i++)
1511 			nic_data->vf[i].vport_id = 0;
1512 #endif
1513 }
1514 
1515 static enum reset_type efx_ef10_map_reset_reason(enum reset_type reason)
1516 {
1517 	if (reason == RESET_TYPE_MC_FAILURE)
1518 		return RESET_TYPE_DATAPATH;
1519 
1520 	return efx_mcdi_map_reset_reason(reason);
1521 }
1522 
1523 static int efx_ef10_map_reset_flags(u32 *flags)
1524 {
1525 	enum {
1526 		EF10_RESET_PORT = ((ETH_RESET_MAC | ETH_RESET_PHY) <<
1527 				   ETH_RESET_SHARED_SHIFT),
1528 		EF10_RESET_MC = ((ETH_RESET_DMA | ETH_RESET_FILTER |
1529 				  ETH_RESET_OFFLOAD | ETH_RESET_MAC |
1530 				  ETH_RESET_PHY | ETH_RESET_MGMT) <<
1531 				 ETH_RESET_SHARED_SHIFT)
1532 	};
1533 
1534 	/* We assume for now that our PCI function is permitted to
1535 	 * reset everything.
1536 	 */
1537 
1538 	if ((*flags & EF10_RESET_MC) == EF10_RESET_MC) {
1539 		*flags &= ~EF10_RESET_MC;
1540 		return RESET_TYPE_WORLD;
1541 	}
1542 
1543 	if ((*flags & EF10_RESET_PORT) == EF10_RESET_PORT) {
1544 		*flags &= ~EF10_RESET_PORT;
1545 		return RESET_TYPE_ALL;
1546 	}
1547 
1548 	/* no invisible reset implemented */
1549 
1550 	return -EINVAL;
1551 }
1552 
1553 static int efx_ef10_reset(struct efx_nic *efx, enum reset_type reset_type)
1554 {
1555 	int rc = efx_mcdi_reset(efx, reset_type);
1556 
1557 	/* Unprivileged functions return -EPERM, but need to return success
1558 	 * here so that the datapath is brought back up.
1559 	 */
1560 	if (reset_type == RESET_TYPE_WORLD && rc == -EPERM)
1561 		rc = 0;
1562 
1563 	/* If it was a port reset, trigger reallocation of MC resources.
1564 	 * Note that on an MC reset nothing needs to be done now because we'll
1565 	 * detect the MC reset later and handle it then.
1566 	 * For an FLR, we never get an MC reset event, but the MC has reset all
1567 	 * resources assigned to us, so we have to trigger reallocation now.
1568 	 */
1569 	if ((reset_type == RESET_TYPE_ALL ||
1570 	     reset_type == RESET_TYPE_MCDI_TIMEOUT) && !rc)
1571 		efx_ef10_reset_mc_allocations(efx);
1572 	return rc;
1573 }
1574 
1575 #define EF10_DMA_STAT(ext_name, mcdi_name)			\
1576 	[EF10_STAT_ ## ext_name] =				\
1577 	{ #ext_name, 64, 8 * MC_CMD_MAC_ ## mcdi_name }
1578 #define EF10_DMA_INVIS_STAT(int_name, mcdi_name)		\
1579 	[EF10_STAT_ ## int_name] =				\
1580 	{ NULL, 64, 8 * MC_CMD_MAC_ ## mcdi_name }
1581 #define EF10_OTHER_STAT(ext_name)				\
1582 	[EF10_STAT_ ## ext_name] = { #ext_name, 0, 0 }
1583 #define GENERIC_SW_STAT(ext_name)				\
1584 	[GENERIC_STAT_ ## ext_name] = { #ext_name, 0, 0 }
1585 
1586 static const struct efx_hw_stat_desc efx_ef10_stat_desc[EF10_STAT_COUNT] = {
1587 	EF10_DMA_STAT(port_tx_bytes, TX_BYTES),
1588 	EF10_DMA_STAT(port_tx_packets, TX_PKTS),
1589 	EF10_DMA_STAT(port_tx_pause, TX_PAUSE_PKTS),
1590 	EF10_DMA_STAT(port_tx_control, TX_CONTROL_PKTS),
1591 	EF10_DMA_STAT(port_tx_unicast, TX_UNICAST_PKTS),
1592 	EF10_DMA_STAT(port_tx_multicast, TX_MULTICAST_PKTS),
1593 	EF10_DMA_STAT(port_tx_broadcast, TX_BROADCAST_PKTS),
1594 	EF10_DMA_STAT(port_tx_lt64, TX_LT64_PKTS),
1595 	EF10_DMA_STAT(port_tx_64, TX_64_PKTS),
1596 	EF10_DMA_STAT(port_tx_65_to_127, TX_65_TO_127_PKTS),
1597 	EF10_DMA_STAT(port_tx_128_to_255, TX_128_TO_255_PKTS),
1598 	EF10_DMA_STAT(port_tx_256_to_511, TX_256_TO_511_PKTS),
1599 	EF10_DMA_STAT(port_tx_512_to_1023, TX_512_TO_1023_PKTS),
1600 	EF10_DMA_STAT(port_tx_1024_to_15xx, TX_1024_TO_15XX_PKTS),
1601 	EF10_DMA_STAT(port_tx_15xx_to_jumbo, TX_15XX_TO_JUMBO_PKTS),
1602 	EF10_DMA_STAT(port_rx_bytes, RX_BYTES),
1603 	EF10_DMA_INVIS_STAT(port_rx_bytes_minus_good_bytes, RX_BAD_BYTES),
1604 	EF10_OTHER_STAT(port_rx_good_bytes),
1605 	EF10_OTHER_STAT(port_rx_bad_bytes),
1606 	EF10_DMA_STAT(port_rx_packets, RX_PKTS),
1607 	EF10_DMA_STAT(port_rx_good, RX_GOOD_PKTS),
1608 	EF10_DMA_STAT(port_rx_bad, RX_BAD_FCS_PKTS),
1609 	EF10_DMA_STAT(port_rx_pause, RX_PAUSE_PKTS),
1610 	EF10_DMA_STAT(port_rx_control, RX_CONTROL_PKTS),
1611 	EF10_DMA_STAT(port_rx_unicast, RX_UNICAST_PKTS),
1612 	EF10_DMA_STAT(port_rx_multicast, RX_MULTICAST_PKTS),
1613 	EF10_DMA_STAT(port_rx_broadcast, RX_BROADCAST_PKTS),
1614 	EF10_DMA_STAT(port_rx_lt64, RX_UNDERSIZE_PKTS),
1615 	EF10_DMA_STAT(port_rx_64, RX_64_PKTS),
1616 	EF10_DMA_STAT(port_rx_65_to_127, RX_65_TO_127_PKTS),
1617 	EF10_DMA_STAT(port_rx_128_to_255, RX_128_TO_255_PKTS),
1618 	EF10_DMA_STAT(port_rx_256_to_511, RX_256_TO_511_PKTS),
1619 	EF10_DMA_STAT(port_rx_512_to_1023, RX_512_TO_1023_PKTS),
1620 	EF10_DMA_STAT(port_rx_1024_to_15xx, RX_1024_TO_15XX_PKTS),
1621 	EF10_DMA_STAT(port_rx_15xx_to_jumbo, RX_15XX_TO_JUMBO_PKTS),
1622 	EF10_DMA_STAT(port_rx_gtjumbo, RX_GTJUMBO_PKTS),
1623 	EF10_DMA_STAT(port_rx_bad_gtjumbo, RX_JABBER_PKTS),
1624 	EF10_DMA_STAT(port_rx_overflow, RX_OVERFLOW_PKTS),
1625 	EF10_DMA_STAT(port_rx_align_error, RX_ALIGN_ERROR_PKTS),
1626 	EF10_DMA_STAT(port_rx_length_error, RX_LENGTH_ERROR_PKTS),
1627 	EF10_DMA_STAT(port_rx_nodesc_drops, RX_NODESC_DROPS),
1628 	GENERIC_SW_STAT(rx_nodesc_trunc),
1629 	GENERIC_SW_STAT(rx_noskb_drops),
1630 	EF10_DMA_STAT(port_rx_pm_trunc_bb_overflow, PM_TRUNC_BB_OVERFLOW),
1631 	EF10_DMA_STAT(port_rx_pm_discard_bb_overflow, PM_DISCARD_BB_OVERFLOW),
1632 	EF10_DMA_STAT(port_rx_pm_trunc_vfifo_full, PM_TRUNC_VFIFO_FULL),
1633 	EF10_DMA_STAT(port_rx_pm_discard_vfifo_full, PM_DISCARD_VFIFO_FULL),
1634 	EF10_DMA_STAT(port_rx_pm_trunc_qbb, PM_TRUNC_QBB),
1635 	EF10_DMA_STAT(port_rx_pm_discard_qbb, PM_DISCARD_QBB),
1636 	EF10_DMA_STAT(port_rx_pm_discard_mapping, PM_DISCARD_MAPPING),
1637 	EF10_DMA_STAT(port_rx_dp_q_disabled_packets, RXDP_Q_DISABLED_PKTS),
1638 	EF10_DMA_STAT(port_rx_dp_di_dropped_packets, RXDP_DI_DROPPED_PKTS),
1639 	EF10_DMA_STAT(port_rx_dp_streaming_packets, RXDP_STREAMING_PKTS),
1640 	EF10_DMA_STAT(port_rx_dp_hlb_fetch, RXDP_HLB_FETCH_CONDITIONS),
1641 	EF10_DMA_STAT(port_rx_dp_hlb_wait, RXDP_HLB_WAIT_CONDITIONS),
1642 	EF10_DMA_STAT(rx_unicast, VADAPTER_RX_UNICAST_PACKETS),
1643 	EF10_DMA_STAT(rx_unicast_bytes, VADAPTER_RX_UNICAST_BYTES),
1644 	EF10_DMA_STAT(rx_multicast, VADAPTER_RX_MULTICAST_PACKETS),
1645 	EF10_DMA_STAT(rx_multicast_bytes, VADAPTER_RX_MULTICAST_BYTES),
1646 	EF10_DMA_STAT(rx_broadcast, VADAPTER_RX_BROADCAST_PACKETS),
1647 	EF10_DMA_STAT(rx_broadcast_bytes, VADAPTER_RX_BROADCAST_BYTES),
1648 	EF10_DMA_STAT(rx_bad, VADAPTER_RX_BAD_PACKETS),
1649 	EF10_DMA_STAT(rx_bad_bytes, VADAPTER_RX_BAD_BYTES),
1650 	EF10_DMA_STAT(rx_overflow, VADAPTER_RX_OVERFLOW),
1651 	EF10_DMA_STAT(tx_unicast, VADAPTER_TX_UNICAST_PACKETS),
1652 	EF10_DMA_STAT(tx_unicast_bytes, VADAPTER_TX_UNICAST_BYTES),
1653 	EF10_DMA_STAT(tx_multicast, VADAPTER_TX_MULTICAST_PACKETS),
1654 	EF10_DMA_STAT(tx_multicast_bytes, VADAPTER_TX_MULTICAST_BYTES),
1655 	EF10_DMA_STAT(tx_broadcast, VADAPTER_TX_BROADCAST_PACKETS),
1656 	EF10_DMA_STAT(tx_broadcast_bytes, VADAPTER_TX_BROADCAST_BYTES),
1657 	EF10_DMA_STAT(tx_bad, VADAPTER_TX_BAD_PACKETS),
1658 	EF10_DMA_STAT(tx_bad_bytes, VADAPTER_TX_BAD_BYTES),
1659 	EF10_DMA_STAT(tx_overflow, VADAPTER_TX_OVERFLOW),
1660 	EF10_DMA_STAT(fec_uncorrected_errors, FEC_UNCORRECTED_ERRORS),
1661 	EF10_DMA_STAT(fec_corrected_errors, FEC_CORRECTED_ERRORS),
1662 	EF10_DMA_STAT(fec_corrected_symbols_lane0, FEC_CORRECTED_SYMBOLS_LANE0),
1663 	EF10_DMA_STAT(fec_corrected_symbols_lane1, FEC_CORRECTED_SYMBOLS_LANE1),
1664 	EF10_DMA_STAT(fec_corrected_symbols_lane2, FEC_CORRECTED_SYMBOLS_LANE2),
1665 	EF10_DMA_STAT(fec_corrected_symbols_lane3, FEC_CORRECTED_SYMBOLS_LANE3),
1666 	EF10_DMA_STAT(ctpio_vi_busy_fallback, CTPIO_VI_BUSY_FALLBACK),
1667 	EF10_DMA_STAT(ctpio_long_write_success, CTPIO_LONG_WRITE_SUCCESS),
1668 	EF10_DMA_STAT(ctpio_missing_dbell_fail, CTPIO_MISSING_DBELL_FAIL),
1669 	EF10_DMA_STAT(ctpio_overflow_fail, CTPIO_OVERFLOW_FAIL),
1670 	EF10_DMA_STAT(ctpio_underflow_fail, CTPIO_UNDERFLOW_FAIL),
1671 	EF10_DMA_STAT(ctpio_timeout_fail, CTPIO_TIMEOUT_FAIL),
1672 	EF10_DMA_STAT(ctpio_noncontig_wr_fail, CTPIO_NONCONTIG_WR_FAIL),
1673 	EF10_DMA_STAT(ctpio_frm_clobber_fail, CTPIO_FRM_CLOBBER_FAIL),
1674 	EF10_DMA_STAT(ctpio_invalid_wr_fail, CTPIO_INVALID_WR_FAIL),
1675 	EF10_DMA_STAT(ctpio_vi_clobber_fallback, CTPIO_VI_CLOBBER_FALLBACK),
1676 	EF10_DMA_STAT(ctpio_unqualified_fallback, CTPIO_UNQUALIFIED_FALLBACK),
1677 	EF10_DMA_STAT(ctpio_runt_fallback, CTPIO_RUNT_FALLBACK),
1678 	EF10_DMA_STAT(ctpio_success, CTPIO_SUCCESS),
1679 	EF10_DMA_STAT(ctpio_fallback, CTPIO_FALLBACK),
1680 	EF10_DMA_STAT(ctpio_poison, CTPIO_POISON),
1681 	EF10_DMA_STAT(ctpio_erase, CTPIO_ERASE),
1682 };
1683 
1684 #define HUNT_COMMON_STAT_MASK ((1ULL << EF10_STAT_port_tx_bytes) |	\
1685 			       (1ULL << EF10_STAT_port_tx_packets) |	\
1686 			       (1ULL << EF10_STAT_port_tx_pause) |	\
1687 			       (1ULL << EF10_STAT_port_tx_unicast) |	\
1688 			       (1ULL << EF10_STAT_port_tx_multicast) |	\
1689 			       (1ULL << EF10_STAT_port_tx_broadcast) |	\
1690 			       (1ULL << EF10_STAT_port_rx_bytes) |	\
1691 			       (1ULL <<                                 \
1692 				EF10_STAT_port_rx_bytes_minus_good_bytes) | \
1693 			       (1ULL << EF10_STAT_port_rx_good_bytes) |	\
1694 			       (1ULL << EF10_STAT_port_rx_bad_bytes) |	\
1695 			       (1ULL << EF10_STAT_port_rx_packets) |	\
1696 			       (1ULL << EF10_STAT_port_rx_good) |	\
1697 			       (1ULL << EF10_STAT_port_rx_bad) |	\
1698 			       (1ULL << EF10_STAT_port_rx_pause) |	\
1699 			       (1ULL << EF10_STAT_port_rx_control) |	\
1700 			       (1ULL << EF10_STAT_port_rx_unicast) |	\
1701 			       (1ULL << EF10_STAT_port_rx_multicast) |	\
1702 			       (1ULL << EF10_STAT_port_rx_broadcast) |	\
1703 			       (1ULL << EF10_STAT_port_rx_lt64) |	\
1704 			       (1ULL << EF10_STAT_port_rx_64) |		\
1705 			       (1ULL << EF10_STAT_port_rx_65_to_127) |	\
1706 			       (1ULL << EF10_STAT_port_rx_128_to_255) |	\
1707 			       (1ULL << EF10_STAT_port_rx_256_to_511) |	\
1708 			       (1ULL << EF10_STAT_port_rx_512_to_1023) |\
1709 			       (1ULL << EF10_STAT_port_rx_1024_to_15xx) |\
1710 			       (1ULL << EF10_STAT_port_rx_15xx_to_jumbo) |\
1711 			       (1ULL << EF10_STAT_port_rx_gtjumbo) |	\
1712 			       (1ULL << EF10_STAT_port_rx_bad_gtjumbo) |\
1713 			       (1ULL << EF10_STAT_port_rx_overflow) |	\
1714 			       (1ULL << EF10_STAT_port_rx_nodesc_drops) |\
1715 			       (1ULL << GENERIC_STAT_rx_nodesc_trunc) |	\
1716 			       (1ULL << GENERIC_STAT_rx_noskb_drops))
1717 
1718 /* On 7000 series NICs, these statistics are only provided by the 10G MAC.
1719  * For a 10G/40G switchable port we do not expose these because they might
1720  * not include all the packets they should.
1721  * On 8000 series NICs these statistics are always provided.
1722  */
1723 #define HUNT_10G_ONLY_STAT_MASK ((1ULL << EF10_STAT_port_tx_control) |	\
1724 				 (1ULL << EF10_STAT_port_tx_lt64) |	\
1725 				 (1ULL << EF10_STAT_port_tx_64) |	\
1726 				 (1ULL << EF10_STAT_port_tx_65_to_127) |\
1727 				 (1ULL << EF10_STAT_port_tx_128_to_255) |\
1728 				 (1ULL << EF10_STAT_port_tx_256_to_511) |\
1729 				 (1ULL << EF10_STAT_port_tx_512_to_1023) |\
1730 				 (1ULL << EF10_STAT_port_tx_1024_to_15xx) |\
1731 				 (1ULL << EF10_STAT_port_tx_15xx_to_jumbo))
1732 
1733 /* These statistics are only provided by the 40G MAC.  For a 10G/40G
1734  * switchable port we do expose these because the errors will otherwise
1735  * be silent.
1736  */
1737 #define HUNT_40G_EXTRA_STAT_MASK ((1ULL << EF10_STAT_port_rx_align_error) |\
1738 				  (1ULL << EF10_STAT_port_rx_length_error))
1739 
1740 /* These statistics are only provided if the firmware supports the
1741  * capability PM_AND_RXDP_COUNTERS.
1742  */
1743 #define HUNT_PM_AND_RXDP_STAT_MASK (					\
1744 	(1ULL << EF10_STAT_port_rx_pm_trunc_bb_overflow) |		\
1745 	(1ULL << EF10_STAT_port_rx_pm_discard_bb_overflow) |		\
1746 	(1ULL << EF10_STAT_port_rx_pm_trunc_vfifo_full) |		\
1747 	(1ULL << EF10_STAT_port_rx_pm_discard_vfifo_full) |		\
1748 	(1ULL << EF10_STAT_port_rx_pm_trunc_qbb) |			\
1749 	(1ULL << EF10_STAT_port_rx_pm_discard_qbb) |			\
1750 	(1ULL << EF10_STAT_port_rx_pm_discard_mapping) |		\
1751 	(1ULL << EF10_STAT_port_rx_dp_q_disabled_packets) |		\
1752 	(1ULL << EF10_STAT_port_rx_dp_di_dropped_packets) |		\
1753 	(1ULL << EF10_STAT_port_rx_dp_streaming_packets) |		\
1754 	(1ULL << EF10_STAT_port_rx_dp_hlb_fetch) |			\
1755 	(1ULL << EF10_STAT_port_rx_dp_hlb_wait))
1756 
1757 /* These statistics are only provided if the NIC supports MC_CMD_MAC_STATS_V2,
1758  * indicated by returning a value >= MC_CMD_MAC_NSTATS_V2 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_FEC_STAT_MASK (						\
1763 	(1ULL << (EF10_STAT_fec_uncorrected_errors - 64)) |		\
1764 	(1ULL << (EF10_STAT_fec_corrected_errors - 64)) |		\
1765 	(1ULL << (EF10_STAT_fec_corrected_symbols_lane0 - 64)) |	\
1766 	(1ULL << (EF10_STAT_fec_corrected_symbols_lane1 - 64)) |	\
1767 	(1ULL << (EF10_STAT_fec_corrected_symbols_lane2 - 64)) |	\
1768 	(1ULL << (EF10_STAT_fec_corrected_symbols_lane3 - 64)))
1769 
1770 /* These statistics are only provided if the NIC supports MC_CMD_MAC_STATS_V3,
1771  * indicated by returning a value >= MC_CMD_MAC_NSTATS_V3 in
1772  * MC_CMD_GET_CAPABILITIES_V4_OUT_MAC_STATS_NUM_STATS.
1773  * These bits are in the second u64 of the raw mask.
1774  */
1775 #define EF10_CTPIO_STAT_MASK (						\
1776 	(1ULL << (EF10_STAT_ctpio_vi_busy_fallback - 64)) |		\
1777 	(1ULL << (EF10_STAT_ctpio_long_write_success - 64)) |		\
1778 	(1ULL << (EF10_STAT_ctpio_missing_dbell_fail - 64)) |		\
1779 	(1ULL << (EF10_STAT_ctpio_overflow_fail - 64)) |		\
1780 	(1ULL << (EF10_STAT_ctpio_underflow_fail - 64)) |		\
1781 	(1ULL << (EF10_STAT_ctpio_timeout_fail - 64)) |			\
1782 	(1ULL << (EF10_STAT_ctpio_noncontig_wr_fail - 64)) |		\
1783 	(1ULL << (EF10_STAT_ctpio_frm_clobber_fail - 64)) |		\
1784 	(1ULL << (EF10_STAT_ctpio_invalid_wr_fail - 64)) |		\
1785 	(1ULL << (EF10_STAT_ctpio_vi_clobber_fallback - 64)) |		\
1786 	(1ULL << (EF10_STAT_ctpio_unqualified_fallback - 64)) |		\
1787 	(1ULL << (EF10_STAT_ctpio_runt_fallback - 64)) |		\
1788 	(1ULL << (EF10_STAT_ctpio_success - 64)) |			\
1789 	(1ULL << (EF10_STAT_ctpio_fallback - 64)) |			\
1790 	(1ULL << (EF10_STAT_ctpio_poison - 64)) |			\
1791 	(1ULL << (EF10_STAT_ctpio_erase - 64)))
1792 
1793 static u64 efx_ef10_raw_stat_mask(struct efx_nic *efx)
1794 {
1795 	u64 raw_mask = HUNT_COMMON_STAT_MASK;
1796 	u32 port_caps = efx_mcdi_phy_get_caps(efx);
1797 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1798 
1799 	if (!(efx->mcdi->fn_flags &
1800 	      1 << MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_LINKCTRL))
1801 		return 0;
1802 
1803 	if (port_caps & (1 << MC_CMD_PHY_CAP_40000FDX_LBN)) {
1804 		raw_mask |= HUNT_40G_EXTRA_STAT_MASK;
1805 		/* 8000 series have everything even at 40G */
1806 		if (nic_data->datapath_caps2 &
1807 		    (1 << MC_CMD_GET_CAPABILITIES_V2_OUT_MAC_STATS_40G_TX_SIZE_BINS_LBN))
1808 			raw_mask |= HUNT_10G_ONLY_STAT_MASK;
1809 	} else {
1810 		raw_mask |= HUNT_10G_ONLY_STAT_MASK;
1811 	}
1812 
1813 	if (nic_data->datapath_caps &
1814 	    (1 << MC_CMD_GET_CAPABILITIES_OUT_PM_AND_RXDP_COUNTERS_LBN))
1815 		raw_mask |= HUNT_PM_AND_RXDP_STAT_MASK;
1816 
1817 	return raw_mask;
1818 }
1819 
1820 static void efx_ef10_get_stat_mask(struct efx_nic *efx, unsigned long *mask)
1821 {
1822 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1823 	u64 raw_mask[2];
1824 
1825 	raw_mask[0] = efx_ef10_raw_stat_mask(efx);
1826 
1827 	/* Only show vadaptor stats when EVB capability is present */
1828 	if (nic_data->datapath_caps &
1829 	    (1 << MC_CMD_GET_CAPABILITIES_OUT_EVB_LBN)) {
1830 		raw_mask[0] |= ~((1ULL << EF10_STAT_rx_unicast) - 1);
1831 		raw_mask[1] = (1ULL << (EF10_STAT_V1_COUNT - 64)) - 1;
1832 	} else {
1833 		raw_mask[1] = 0;
1834 	}
1835 	/* Only show FEC stats when NIC supports MC_CMD_MAC_STATS_V2 */
1836 	if (efx->num_mac_stats >= MC_CMD_MAC_NSTATS_V2)
1837 		raw_mask[1] |= EF10_FEC_STAT_MASK;
1838 
1839 	/* CTPIO stats appear in V3. Only show them on devices that actually
1840 	 * support CTPIO. Although this driver doesn't use CTPIO others might,
1841 	 * and we may be reporting the stats for the underlying port.
1842 	 */
1843 	if (efx->num_mac_stats >= MC_CMD_MAC_NSTATS_V3 &&
1844 	    (nic_data->datapath_caps2 &
1845 	     (1 << MC_CMD_GET_CAPABILITIES_V4_OUT_CTPIO_LBN)))
1846 		raw_mask[1] |= EF10_CTPIO_STAT_MASK;
1847 
1848 #if BITS_PER_LONG == 64
1849 	BUILD_BUG_ON(BITS_TO_LONGS(EF10_STAT_COUNT) != 2);
1850 	mask[0] = raw_mask[0];
1851 	mask[1] = raw_mask[1];
1852 #else
1853 	BUILD_BUG_ON(BITS_TO_LONGS(EF10_STAT_COUNT) != 3);
1854 	mask[0] = raw_mask[0] & 0xffffffff;
1855 	mask[1] = raw_mask[0] >> 32;
1856 	mask[2] = raw_mask[1] & 0xffffffff;
1857 #endif
1858 }
1859 
1860 static size_t efx_ef10_describe_stats(struct efx_nic *efx, u8 *names)
1861 {
1862 	DECLARE_BITMAP(mask, EF10_STAT_COUNT);
1863 
1864 	efx_ef10_get_stat_mask(efx, mask);
1865 	return efx_nic_describe_stats(efx_ef10_stat_desc, EF10_STAT_COUNT,
1866 				      mask, names);
1867 }
1868 
1869 static size_t efx_ef10_update_stats_common(struct efx_nic *efx, u64 *full_stats,
1870 					   struct rtnl_link_stats64 *core_stats)
1871 {
1872 	DECLARE_BITMAP(mask, EF10_STAT_COUNT);
1873 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1874 	u64 *stats = nic_data->stats;
1875 	size_t stats_count = 0, index;
1876 
1877 	efx_ef10_get_stat_mask(efx, mask);
1878 
1879 	if (full_stats) {
1880 		for_each_set_bit(index, mask, EF10_STAT_COUNT) {
1881 			if (efx_ef10_stat_desc[index].name) {
1882 				*full_stats++ = stats[index];
1883 				++stats_count;
1884 			}
1885 		}
1886 	}
1887 
1888 	if (!core_stats)
1889 		return stats_count;
1890 
1891 	if (nic_data->datapath_caps &
1892 			1 << MC_CMD_GET_CAPABILITIES_OUT_EVB_LBN) {
1893 		/* Use vadaptor stats. */
1894 		core_stats->rx_packets = stats[EF10_STAT_rx_unicast] +
1895 					 stats[EF10_STAT_rx_multicast] +
1896 					 stats[EF10_STAT_rx_broadcast];
1897 		core_stats->tx_packets = stats[EF10_STAT_tx_unicast] +
1898 					 stats[EF10_STAT_tx_multicast] +
1899 					 stats[EF10_STAT_tx_broadcast];
1900 		core_stats->rx_bytes = stats[EF10_STAT_rx_unicast_bytes] +
1901 				       stats[EF10_STAT_rx_multicast_bytes] +
1902 				       stats[EF10_STAT_rx_broadcast_bytes];
1903 		core_stats->tx_bytes = stats[EF10_STAT_tx_unicast_bytes] +
1904 				       stats[EF10_STAT_tx_multicast_bytes] +
1905 				       stats[EF10_STAT_tx_broadcast_bytes];
1906 		core_stats->rx_dropped = stats[GENERIC_STAT_rx_nodesc_trunc] +
1907 					 stats[GENERIC_STAT_rx_noskb_drops];
1908 		core_stats->multicast = stats[EF10_STAT_rx_multicast];
1909 		core_stats->rx_crc_errors = stats[EF10_STAT_rx_bad];
1910 		core_stats->rx_fifo_errors = stats[EF10_STAT_rx_overflow];
1911 		core_stats->rx_errors = core_stats->rx_crc_errors;
1912 		core_stats->tx_errors = stats[EF10_STAT_tx_bad];
1913 	} else {
1914 		/* Use port stats. */
1915 		core_stats->rx_packets = stats[EF10_STAT_port_rx_packets];
1916 		core_stats->tx_packets = stats[EF10_STAT_port_tx_packets];
1917 		core_stats->rx_bytes = stats[EF10_STAT_port_rx_bytes];
1918 		core_stats->tx_bytes = stats[EF10_STAT_port_tx_bytes];
1919 		core_stats->rx_dropped = stats[EF10_STAT_port_rx_nodesc_drops] +
1920 					 stats[GENERIC_STAT_rx_nodesc_trunc] +
1921 					 stats[GENERIC_STAT_rx_noskb_drops];
1922 		core_stats->multicast = stats[EF10_STAT_port_rx_multicast];
1923 		core_stats->rx_length_errors =
1924 				stats[EF10_STAT_port_rx_gtjumbo] +
1925 				stats[EF10_STAT_port_rx_length_error];
1926 		core_stats->rx_crc_errors = stats[EF10_STAT_port_rx_bad];
1927 		core_stats->rx_frame_errors =
1928 				stats[EF10_STAT_port_rx_align_error];
1929 		core_stats->rx_fifo_errors = stats[EF10_STAT_port_rx_overflow];
1930 		core_stats->rx_errors = (core_stats->rx_length_errors +
1931 					 core_stats->rx_crc_errors +
1932 					 core_stats->rx_frame_errors);
1933 	}
1934 
1935 	return stats_count;
1936 }
1937 
1938 static int efx_ef10_try_update_nic_stats_pf(struct efx_nic *efx)
1939 {
1940 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1941 	DECLARE_BITMAP(mask, EF10_STAT_COUNT);
1942 	__le64 generation_start, generation_end;
1943 	u64 *stats = nic_data->stats;
1944 	__le64 *dma_stats;
1945 
1946 	efx_ef10_get_stat_mask(efx, mask);
1947 
1948 	dma_stats = efx->stats_buffer.addr;
1949 
1950 	generation_end = dma_stats[efx->num_mac_stats - 1];
1951 	if (generation_end == EFX_MC_STATS_GENERATION_INVALID)
1952 		return 0;
1953 	rmb();
1954 	efx_nic_update_stats(efx_ef10_stat_desc, EF10_STAT_COUNT, mask,
1955 			     stats, efx->stats_buffer.addr, false);
1956 	rmb();
1957 	generation_start = dma_stats[MC_CMD_MAC_GENERATION_START];
1958 	if (generation_end != generation_start)
1959 		return -EAGAIN;
1960 
1961 	/* Update derived statistics */
1962 	efx_nic_fix_nodesc_drop_stat(efx,
1963 				     &stats[EF10_STAT_port_rx_nodesc_drops]);
1964 	stats[EF10_STAT_port_rx_good_bytes] =
1965 		stats[EF10_STAT_port_rx_bytes] -
1966 		stats[EF10_STAT_port_rx_bytes_minus_good_bytes];
1967 	efx_update_diff_stat(&stats[EF10_STAT_port_rx_bad_bytes],
1968 			     stats[EF10_STAT_port_rx_bytes_minus_good_bytes]);
1969 	efx_update_sw_stats(efx, stats);
1970 	return 0;
1971 }
1972 
1973 
1974 static size_t efx_ef10_update_stats_pf(struct efx_nic *efx, u64 *full_stats,
1975 				       struct rtnl_link_stats64 *core_stats)
1976 {
1977 	int retry;
1978 
1979 	/* If we're unlucky enough to read statistics during the DMA, wait
1980 	 * up to 10ms for it to finish (typically takes <500us)
1981 	 */
1982 	for (retry = 0; retry < 100; ++retry) {
1983 		if (efx_ef10_try_update_nic_stats_pf(efx) == 0)
1984 			break;
1985 		udelay(100);
1986 	}
1987 
1988 	return efx_ef10_update_stats_common(efx, full_stats, core_stats);
1989 }
1990 
1991 static int efx_ef10_try_update_nic_stats_vf(struct efx_nic *efx)
1992 {
1993 	MCDI_DECLARE_BUF(inbuf, MC_CMD_MAC_STATS_IN_LEN);
1994 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1995 	DECLARE_BITMAP(mask, EF10_STAT_COUNT);
1996 	__le64 generation_start, generation_end;
1997 	u64 *stats = nic_data->stats;
1998 	u32 dma_len = efx->num_mac_stats * sizeof(u64);
1999 	struct efx_buffer stats_buf;
2000 	__le64 *dma_stats;
2001 	int rc;
2002 
2003 	spin_unlock_bh(&efx->stats_lock);
2004 
2005 	if (in_interrupt()) {
2006 		/* If in atomic context, cannot update stats.  Just update the
2007 		 * software stats and return so the caller can continue.
2008 		 */
2009 		spin_lock_bh(&efx->stats_lock);
2010 		efx_update_sw_stats(efx, stats);
2011 		return 0;
2012 	}
2013 
2014 	efx_ef10_get_stat_mask(efx, mask);
2015 
2016 	rc = efx_nic_alloc_buffer(efx, &stats_buf, dma_len, GFP_ATOMIC);
2017 	if (rc) {
2018 		spin_lock_bh(&efx->stats_lock);
2019 		return rc;
2020 	}
2021 
2022 	dma_stats = stats_buf.addr;
2023 	dma_stats[efx->num_mac_stats - 1] = EFX_MC_STATS_GENERATION_INVALID;
2024 
2025 	MCDI_SET_QWORD(inbuf, MAC_STATS_IN_DMA_ADDR, stats_buf.dma_addr);
2026 	MCDI_POPULATE_DWORD_1(inbuf, MAC_STATS_IN_CMD,
2027 			      MAC_STATS_IN_DMA, 1);
2028 	MCDI_SET_DWORD(inbuf, MAC_STATS_IN_DMA_LEN, dma_len);
2029 	MCDI_SET_DWORD(inbuf, MAC_STATS_IN_PORT_ID, EVB_PORT_ID_ASSIGNED);
2030 
2031 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_MAC_STATS, inbuf, sizeof(inbuf),
2032 				NULL, 0, NULL);
2033 	spin_lock_bh(&efx->stats_lock);
2034 	if (rc) {
2035 		/* Expect ENOENT if DMA queues have not been set up */
2036 		if (rc != -ENOENT || atomic_read(&efx->active_queues))
2037 			efx_mcdi_display_error(efx, MC_CMD_MAC_STATS,
2038 					       sizeof(inbuf), NULL, 0, rc);
2039 		goto out;
2040 	}
2041 
2042 	generation_end = dma_stats[efx->num_mac_stats - 1];
2043 	if (generation_end == EFX_MC_STATS_GENERATION_INVALID) {
2044 		WARN_ON_ONCE(1);
2045 		goto out;
2046 	}
2047 	rmb();
2048 	efx_nic_update_stats(efx_ef10_stat_desc, EF10_STAT_COUNT, mask,
2049 			     stats, stats_buf.addr, false);
2050 	rmb();
2051 	generation_start = dma_stats[MC_CMD_MAC_GENERATION_START];
2052 	if (generation_end != generation_start) {
2053 		rc = -EAGAIN;
2054 		goto out;
2055 	}
2056 
2057 	efx_update_sw_stats(efx, stats);
2058 out:
2059 	efx_nic_free_buffer(efx, &stats_buf);
2060 	return rc;
2061 }
2062 
2063 static size_t efx_ef10_update_stats_vf(struct efx_nic *efx, u64 *full_stats,
2064 				       struct rtnl_link_stats64 *core_stats)
2065 {
2066 	if (efx_ef10_try_update_nic_stats_vf(efx))
2067 		return 0;
2068 
2069 	return efx_ef10_update_stats_common(efx, full_stats, core_stats);
2070 }
2071 
2072 static void efx_ef10_push_irq_moderation(struct efx_channel *channel)
2073 {
2074 	struct efx_nic *efx = channel->efx;
2075 	unsigned int mode, usecs;
2076 	efx_dword_t timer_cmd;
2077 
2078 	if (channel->irq_moderation_us) {
2079 		mode = 3;
2080 		usecs = channel->irq_moderation_us;
2081 	} else {
2082 		mode = 0;
2083 		usecs = 0;
2084 	}
2085 
2086 	if (EFX_EF10_WORKAROUND_61265(efx)) {
2087 		MCDI_DECLARE_BUF(inbuf, MC_CMD_SET_EVQ_TMR_IN_LEN);
2088 		unsigned int ns = usecs * 1000;
2089 
2090 		MCDI_SET_DWORD(inbuf, SET_EVQ_TMR_IN_INSTANCE,
2091 			       channel->channel);
2092 		MCDI_SET_DWORD(inbuf, SET_EVQ_TMR_IN_TMR_LOAD_REQ_NS, ns);
2093 		MCDI_SET_DWORD(inbuf, SET_EVQ_TMR_IN_TMR_RELOAD_REQ_NS, ns);
2094 		MCDI_SET_DWORD(inbuf, SET_EVQ_TMR_IN_TMR_MODE, mode);
2095 
2096 		efx_mcdi_rpc_async(efx, MC_CMD_SET_EVQ_TMR,
2097 				   inbuf, sizeof(inbuf), 0, NULL, 0);
2098 	} else if (EFX_EF10_WORKAROUND_35388(efx)) {
2099 		unsigned int ticks = efx_usecs_to_ticks(efx, usecs);
2100 
2101 		EFX_POPULATE_DWORD_3(timer_cmd, ERF_DD_EVQ_IND_TIMER_FLAGS,
2102 				     EFE_DD_EVQ_IND_TIMER_FLAGS,
2103 				     ERF_DD_EVQ_IND_TIMER_MODE, mode,
2104 				     ERF_DD_EVQ_IND_TIMER_VAL, ticks);
2105 		efx_writed_page(efx, &timer_cmd, ER_DD_EVQ_INDIRECT,
2106 				channel->channel);
2107 	} else {
2108 		unsigned int ticks = efx_usecs_to_ticks(efx, usecs);
2109 
2110 		EFX_POPULATE_DWORD_3(timer_cmd, ERF_DZ_TC_TIMER_MODE, mode,
2111 				     ERF_DZ_TC_TIMER_VAL, ticks,
2112 				     ERF_FZ_TC_TMR_REL_VAL, ticks);
2113 		efx_writed_page(efx, &timer_cmd, ER_DZ_EVQ_TMR,
2114 				channel->channel);
2115 	}
2116 }
2117 
2118 static void efx_ef10_get_wol_vf(struct efx_nic *efx,
2119 				struct ethtool_wolinfo *wol) {}
2120 
2121 static int efx_ef10_set_wol_vf(struct efx_nic *efx, u32 type)
2122 {
2123 	return -EOPNOTSUPP;
2124 }
2125 
2126 static void efx_ef10_get_wol(struct efx_nic *efx, struct ethtool_wolinfo *wol)
2127 {
2128 	wol->supported = 0;
2129 	wol->wolopts = 0;
2130 	memset(&wol->sopass, 0, sizeof(wol->sopass));
2131 }
2132 
2133 static int efx_ef10_set_wol(struct efx_nic *efx, u32 type)
2134 {
2135 	if (type != 0)
2136 		return -EINVAL;
2137 	return 0;
2138 }
2139 
2140 static void efx_ef10_mcdi_request(struct efx_nic *efx,
2141 				  const efx_dword_t *hdr, size_t hdr_len,
2142 				  const efx_dword_t *sdu, size_t sdu_len)
2143 {
2144 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2145 	u8 *pdu = nic_data->mcdi_buf.addr;
2146 
2147 	memcpy(pdu, hdr, hdr_len);
2148 	memcpy(pdu + hdr_len, sdu, sdu_len);
2149 	wmb();
2150 
2151 	/* The hardware provides 'low' and 'high' (doorbell) registers
2152 	 * for passing the 64-bit address of an MCDI request to
2153 	 * firmware.  However the dwords are swapped by firmware.  The
2154 	 * least significant bits of the doorbell are then 0 for all
2155 	 * MCDI requests due to alignment.
2156 	 */
2157 	_efx_writed(efx, cpu_to_le32((u64)nic_data->mcdi_buf.dma_addr >> 32),
2158 		    ER_DZ_MC_DB_LWRD);
2159 	_efx_writed(efx, cpu_to_le32((u32)nic_data->mcdi_buf.dma_addr),
2160 		    ER_DZ_MC_DB_HWRD);
2161 }
2162 
2163 static bool efx_ef10_mcdi_poll_response(struct efx_nic *efx)
2164 {
2165 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2166 	const efx_dword_t hdr = *(const efx_dword_t *)nic_data->mcdi_buf.addr;
2167 
2168 	rmb();
2169 	return EFX_DWORD_FIELD(hdr, MCDI_HEADER_RESPONSE);
2170 }
2171 
2172 static void
2173 efx_ef10_mcdi_read_response(struct efx_nic *efx, efx_dword_t *outbuf,
2174 			    size_t offset, size_t outlen)
2175 {
2176 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2177 	const u8 *pdu = nic_data->mcdi_buf.addr;
2178 
2179 	memcpy(outbuf, pdu + offset, outlen);
2180 }
2181 
2182 static void efx_ef10_mcdi_reboot_detected(struct efx_nic *efx)
2183 {
2184 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2185 
2186 	/* All our allocations have been reset */
2187 	efx_ef10_reset_mc_allocations(efx);
2188 
2189 	/* The datapath firmware might have been changed */
2190 	nic_data->must_check_datapath_caps = true;
2191 
2192 	/* MAC statistics have been cleared on the NIC; clear the local
2193 	 * statistic that we update with efx_update_diff_stat().
2194 	 */
2195 	nic_data->stats[EF10_STAT_port_rx_bad_bytes] = 0;
2196 }
2197 
2198 static int efx_ef10_mcdi_poll_reboot(struct efx_nic *efx)
2199 {
2200 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2201 	int rc;
2202 
2203 	rc = efx_ef10_get_warm_boot_count(efx);
2204 	if (rc < 0) {
2205 		/* The firmware is presumably in the process of
2206 		 * rebooting.  However, we are supposed to report each
2207 		 * reboot just once, so we must only do that once we
2208 		 * can read and store the updated warm boot count.
2209 		 */
2210 		return 0;
2211 	}
2212 
2213 	if (rc == nic_data->warm_boot_count)
2214 		return 0;
2215 
2216 	nic_data->warm_boot_count = rc;
2217 	efx_ef10_mcdi_reboot_detected(efx);
2218 
2219 	return -EIO;
2220 }
2221 
2222 /* Handle an MSI interrupt
2223  *
2224  * Handle an MSI hardware interrupt.  This routine schedules event
2225  * queue processing.  No interrupt acknowledgement cycle is necessary.
2226  * Also, we never need to check that the interrupt is for us, since
2227  * MSI interrupts cannot be shared.
2228  */
2229 static irqreturn_t efx_ef10_msi_interrupt(int irq, void *dev_id)
2230 {
2231 	struct efx_msi_context *context = dev_id;
2232 	struct efx_nic *efx = context->efx;
2233 
2234 	netif_vdbg(efx, intr, efx->net_dev,
2235 		   "IRQ %d on CPU %d\n", irq, raw_smp_processor_id());
2236 
2237 	if (likely(READ_ONCE(efx->irq_soft_enabled))) {
2238 		/* Note test interrupts */
2239 		if (context->index == efx->irq_level)
2240 			efx->last_irq_cpu = raw_smp_processor_id();
2241 
2242 		/* Schedule processing of the channel */
2243 		efx_schedule_channel_irq(efx->channel[context->index]);
2244 	}
2245 
2246 	return IRQ_HANDLED;
2247 }
2248 
2249 static irqreturn_t efx_ef10_legacy_interrupt(int irq, void *dev_id)
2250 {
2251 	struct efx_nic *efx = dev_id;
2252 	bool soft_enabled = READ_ONCE(efx->irq_soft_enabled);
2253 	struct efx_channel *channel;
2254 	efx_dword_t reg;
2255 	u32 queues;
2256 
2257 	/* Read the ISR which also ACKs the interrupts */
2258 	efx_readd(efx, &reg, ER_DZ_BIU_INT_ISR);
2259 	queues = EFX_DWORD_FIELD(reg, ERF_DZ_ISR_REG);
2260 
2261 	if (queues == 0)
2262 		return IRQ_NONE;
2263 
2264 	if (likely(soft_enabled)) {
2265 		/* Note test interrupts */
2266 		if (queues & (1U << efx->irq_level))
2267 			efx->last_irq_cpu = raw_smp_processor_id();
2268 
2269 		efx_for_each_channel(channel, efx) {
2270 			if (queues & 1)
2271 				efx_schedule_channel_irq(channel);
2272 			queues >>= 1;
2273 		}
2274 	}
2275 
2276 	netif_vdbg(efx, intr, efx->net_dev,
2277 		   "IRQ %d on CPU %d status " EFX_DWORD_FMT "\n",
2278 		   irq, raw_smp_processor_id(), EFX_DWORD_VAL(reg));
2279 
2280 	return IRQ_HANDLED;
2281 }
2282 
2283 static int efx_ef10_irq_test_generate(struct efx_nic *efx)
2284 {
2285 	MCDI_DECLARE_BUF(inbuf, MC_CMD_TRIGGER_INTERRUPT_IN_LEN);
2286 
2287 	if (efx_mcdi_set_workaround(efx, MC_CMD_WORKAROUND_BUG41750, true,
2288 				    NULL) == 0)
2289 		return -ENOTSUPP;
2290 
2291 	BUILD_BUG_ON(MC_CMD_TRIGGER_INTERRUPT_OUT_LEN != 0);
2292 
2293 	MCDI_SET_DWORD(inbuf, TRIGGER_INTERRUPT_IN_INTR_LEVEL, efx->irq_level);
2294 	return efx_mcdi_rpc(efx, MC_CMD_TRIGGER_INTERRUPT,
2295 			    inbuf, sizeof(inbuf), NULL, 0, NULL);
2296 }
2297 
2298 static int efx_ef10_tx_probe(struct efx_tx_queue *tx_queue)
2299 {
2300 	return efx_nic_alloc_buffer(tx_queue->efx, &tx_queue->txd.buf,
2301 				    (tx_queue->ptr_mask + 1) *
2302 				    sizeof(efx_qword_t),
2303 				    GFP_KERNEL);
2304 }
2305 
2306 /* This writes to the TX_DESC_WPTR and also pushes data */
2307 static inline void efx_ef10_push_tx_desc(struct efx_tx_queue *tx_queue,
2308 					 const efx_qword_t *txd)
2309 {
2310 	unsigned int write_ptr;
2311 	efx_oword_t reg;
2312 
2313 	write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
2314 	EFX_POPULATE_OWORD_1(reg, ERF_DZ_TX_DESC_WPTR, write_ptr);
2315 	reg.qword[0] = *txd;
2316 	efx_writeo_page(tx_queue->efx, &reg,
2317 			ER_DZ_TX_DESC_UPD, tx_queue->queue);
2318 }
2319 
2320 /* Add Firmware-Assisted TSO v2 option descriptors to a queue.
2321  */
2322 static int efx_ef10_tx_tso_desc(struct efx_tx_queue *tx_queue,
2323 				struct sk_buff *skb,
2324 				bool *data_mapped)
2325 {
2326 	struct efx_tx_buffer *buffer;
2327 	struct tcphdr *tcp;
2328 	struct iphdr *ip;
2329 
2330 	u16 ipv4_id;
2331 	u32 seqnum;
2332 	u32 mss;
2333 
2334 	EFX_WARN_ON_ONCE_PARANOID(tx_queue->tso_version != 2);
2335 
2336 	mss = skb_shinfo(skb)->gso_size;
2337 
2338 	if (unlikely(mss < 4)) {
2339 		WARN_ONCE(1, "MSS of %u is too small for TSO v2\n", mss);
2340 		return -EINVAL;
2341 	}
2342 
2343 	ip = ip_hdr(skb);
2344 	if (ip->version == 4) {
2345 		/* Modify IPv4 header if needed. */
2346 		ip->tot_len = 0;
2347 		ip->check = 0;
2348 		ipv4_id = ntohs(ip->id);
2349 	} else {
2350 		/* Modify IPv6 header if needed. */
2351 		struct ipv6hdr *ipv6 = ipv6_hdr(skb);
2352 
2353 		ipv6->payload_len = 0;
2354 		ipv4_id = 0;
2355 	}
2356 
2357 	tcp = tcp_hdr(skb);
2358 	seqnum = ntohl(tcp->seq);
2359 
2360 	buffer = efx_tx_queue_get_insert_buffer(tx_queue);
2361 
2362 	buffer->flags = EFX_TX_BUF_OPTION;
2363 	buffer->len = 0;
2364 	buffer->unmap_len = 0;
2365 	EFX_POPULATE_QWORD_5(buffer->option,
2366 			ESF_DZ_TX_DESC_IS_OPT, 1,
2367 			ESF_DZ_TX_OPTION_TYPE, ESE_DZ_TX_OPTION_DESC_TSO,
2368 			ESF_DZ_TX_TSO_OPTION_TYPE,
2369 			ESE_DZ_TX_TSO_OPTION_DESC_FATSO2A,
2370 			ESF_DZ_TX_TSO_IP_ID, ipv4_id,
2371 			ESF_DZ_TX_TSO_TCP_SEQNO, seqnum
2372 			);
2373 	++tx_queue->insert_count;
2374 
2375 	buffer = efx_tx_queue_get_insert_buffer(tx_queue);
2376 
2377 	buffer->flags = EFX_TX_BUF_OPTION;
2378 	buffer->len = 0;
2379 	buffer->unmap_len = 0;
2380 	EFX_POPULATE_QWORD_4(buffer->option,
2381 			ESF_DZ_TX_DESC_IS_OPT, 1,
2382 			ESF_DZ_TX_OPTION_TYPE, ESE_DZ_TX_OPTION_DESC_TSO,
2383 			ESF_DZ_TX_TSO_OPTION_TYPE,
2384 			ESE_DZ_TX_TSO_OPTION_DESC_FATSO2B,
2385 			ESF_DZ_TX_TSO_TCP_MSS, mss
2386 			);
2387 	++tx_queue->insert_count;
2388 
2389 	return 0;
2390 }
2391 
2392 static u32 efx_ef10_tso_versions(struct efx_nic *efx)
2393 {
2394 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2395 	u32 tso_versions = 0;
2396 
2397 	if (nic_data->datapath_caps &
2398 	    (1 << MC_CMD_GET_CAPABILITIES_OUT_TX_TSO_LBN))
2399 		tso_versions |= BIT(1);
2400 	if (nic_data->datapath_caps2 &
2401 	    (1 << MC_CMD_GET_CAPABILITIES_V2_OUT_TX_TSO_V2_LBN))
2402 		tso_versions |= BIT(2);
2403 	return tso_versions;
2404 }
2405 
2406 static void efx_ef10_tx_init(struct efx_tx_queue *tx_queue)
2407 {
2408 	MCDI_DECLARE_BUF(inbuf, MC_CMD_INIT_TXQ_IN_LEN(EFX_MAX_DMAQ_SIZE * 8 /
2409 						       EFX_BUF_SIZE));
2410 	bool csum_offload = tx_queue->queue & EFX_TXQ_TYPE_OFFLOAD;
2411 	size_t entries = tx_queue->txd.buf.len / EFX_BUF_SIZE;
2412 	struct efx_channel *channel = tx_queue->channel;
2413 	struct efx_nic *efx = tx_queue->efx;
2414 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2415 	bool tso_v2 = false;
2416 	size_t inlen;
2417 	dma_addr_t dma_addr;
2418 	efx_qword_t *txd;
2419 	int rc;
2420 	int i;
2421 	BUILD_BUG_ON(MC_CMD_INIT_TXQ_OUT_LEN != 0);
2422 
2423 	/* Only attempt to enable TX timestamping if we have the license for it,
2424 	 * otherwise TXQ init will fail
2425 	 */
2426 	if (!(nic_data->licensed_features &
2427 	      (1 << LICENSED_V3_FEATURES_TX_TIMESTAMPS_LBN))) {
2428 		tx_queue->timestamping = false;
2429 		/* Disable sync events on this channel. */
2430 		if (efx->type->ptp_set_ts_sync_events)
2431 			efx->type->ptp_set_ts_sync_events(efx, false, false);
2432 	}
2433 
2434 	/* TSOv2 is a limited resource that can only be configured on a limited
2435 	 * number of queues. TSO without checksum offload is not really a thing,
2436 	 * so we only enable it for those queues.
2437 	 * TSOv2 cannot be used with Hardware timestamping.
2438 	 */
2439 	if (csum_offload && (nic_data->datapath_caps2 &
2440 			(1 << MC_CMD_GET_CAPABILITIES_V2_OUT_TX_TSO_V2_LBN)) &&
2441 	    !tx_queue->timestamping) {
2442 		tso_v2 = true;
2443 		netif_dbg(efx, hw, efx->net_dev, "Using TSOv2 for channel %u\n",
2444 				channel->channel);
2445 	}
2446 
2447 	MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_SIZE, tx_queue->ptr_mask + 1);
2448 	MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_TARGET_EVQ, channel->channel);
2449 	MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_LABEL, tx_queue->queue);
2450 	MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_INSTANCE, tx_queue->queue);
2451 	MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_OWNER_ID, 0);
2452 	MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_PORT_ID, nic_data->vport_id);
2453 
2454 	dma_addr = tx_queue->txd.buf.dma_addr;
2455 
2456 	netif_dbg(efx, hw, efx->net_dev, "pushing TXQ %d. %zu entries (%llx)\n",
2457 		  tx_queue->queue, entries, (u64)dma_addr);
2458 
2459 	for (i = 0; i < entries; ++i) {
2460 		MCDI_SET_ARRAY_QWORD(inbuf, INIT_TXQ_IN_DMA_ADDR, i, dma_addr);
2461 		dma_addr += EFX_BUF_SIZE;
2462 	}
2463 
2464 	inlen = MC_CMD_INIT_TXQ_IN_LEN(entries);
2465 
2466 	do {
2467 		MCDI_POPULATE_DWORD_4(inbuf, INIT_TXQ_IN_FLAGS,
2468 				/* This flag was removed from mcdi_pcol.h for
2469 				 * the non-_EXT version of INIT_TXQ.  However,
2470 				 * firmware still honours it.
2471 				 */
2472 				INIT_TXQ_EXT_IN_FLAG_TSOV2_EN, tso_v2,
2473 				INIT_TXQ_IN_FLAG_IP_CSUM_DIS, !csum_offload,
2474 				INIT_TXQ_IN_FLAG_TCP_CSUM_DIS, !csum_offload,
2475 				INIT_TXQ_EXT_IN_FLAG_TIMESTAMP,
2476 						tx_queue->timestamping);
2477 
2478 		rc = efx_mcdi_rpc_quiet(efx, MC_CMD_INIT_TXQ, inbuf, inlen,
2479 					NULL, 0, NULL);
2480 		if (rc == -ENOSPC && tso_v2) {
2481 			/* Retry without TSOv2 if we're short on contexts. */
2482 			tso_v2 = false;
2483 			netif_warn(efx, probe, efx->net_dev,
2484 				   "TSOv2 context not available to segment in hardware. TCP performance may be reduced.\n");
2485 		} else if (rc) {
2486 			efx_mcdi_display_error(efx, MC_CMD_INIT_TXQ,
2487 					       MC_CMD_INIT_TXQ_EXT_IN_LEN,
2488 					       NULL, 0, rc);
2489 			goto fail;
2490 		}
2491 	} while (rc);
2492 
2493 	/* A previous user of this TX queue might have set us up the
2494 	 * bomb by writing a descriptor to the TX push collector but
2495 	 * not the doorbell.  (Each collector belongs to a port, not a
2496 	 * queue or function, so cannot easily be reset.)  We must
2497 	 * attempt to push a no-op descriptor in its place.
2498 	 */
2499 	tx_queue->buffer[0].flags = EFX_TX_BUF_OPTION;
2500 	tx_queue->insert_count = 1;
2501 	txd = efx_tx_desc(tx_queue, 0);
2502 	EFX_POPULATE_QWORD_5(*txd,
2503 			     ESF_DZ_TX_DESC_IS_OPT, true,
2504 			     ESF_DZ_TX_OPTION_TYPE,
2505 			     ESE_DZ_TX_OPTION_DESC_CRC_CSUM,
2506 			     ESF_DZ_TX_OPTION_UDP_TCP_CSUM, csum_offload,
2507 			     ESF_DZ_TX_OPTION_IP_CSUM, csum_offload,
2508 			     ESF_DZ_TX_TIMESTAMP, tx_queue->timestamping);
2509 	tx_queue->write_count = 1;
2510 
2511 	if (tso_v2) {
2512 		tx_queue->handle_tso = efx_ef10_tx_tso_desc;
2513 		tx_queue->tso_version = 2;
2514 	} else if (nic_data->datapath_caps &
2515 			(1 << MC_CMD_GET_CAPABILITIES_OUT_TX_TSO_LBN)) {
2516 		tx_queue->tso_version = 1;
2517 	}
2518 
2519 	wmb();
2520 	efx_ef10_push_tx_desc(tx_queue, txd);
2521 
2522 	return;
2523 
2524 fail:
2525 	netdev_WARN(efx->net_dev, "failed to initialise TXQ %d\n",
2526 		    tx_queue->queue);
2527 }
2528 
2529 static void efx_ef10_tx_fini(struct efx_tx_queue *tx_queue)
2530 {
2531 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FINI_TXQ_IN_LEN);
2532 	MCDI_DECLARE_BUF_ERR(outbuf);
2533 	struct efx_nic *efx = tx_queue->efx;
2534 	size_t outlen;
2535 	int rc;
2536 
2537 	MCDI_SET_DWORD(inbuf, FINI_TXQ_IN_INSTANCE,
2538 		       tx_queue->queue);
2539 
2540 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FINI_TXQ, inbuf, sizeof(inbuf),
2541 			  outbuf, sizeof(outbuf), &outlen);
2542 
2543 	if (rc && rc != -EALREADY)
2544 		goto fail;
2545 
2546 	return;
2547 
2548 fail:
2549 	efx_mcdi_display_error(efx, MC_CMD_FINI_TXQ, MC_CMD_FINI_TXQ_IN_LEN,
2550 			       outbuf, outlen, rc);
2551 }
2552 
2553 static void efx_ef10_tx_remove(struct efx_tx_queue *tx_queue)
2554 {
2555 	efx_nic_free_buffer(tx_queue->efx, &tx_queue->txd.buf);
2556 }
2557 
2558 /* This writes to the TX_DESC_WPTR; write pointer for TX descriptor ring */
2559 static inline void efx_ef10_notify_tx_desc(struct efx_tx_queue *tx_queue)
2560 {
2561 	unsigned int write_ptr;
2562 	efx_dword_t reg;
2563 
2564 	write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
2565 	EFX_POPULATE_DWORD_1(reg, ERF_DZ_TX_DESC_WPTR_DWORD, write_ptr);
2566 	efx_writed_page(tx_queue->efx, &reg,
2567 			ER_DZ_TX_DESC_UPD_DWORD, tx_queue->queue);
2568 }
2569 
2570 #define EFX_EF10_MAX_TX_DESCRIPTOR_LEN 0x3fff
2571 
2572 static unsigned int efx_ef10_tx_limit_len(struct efx_tx_queue *tx_queue,
2573 					  dma_addr_t dma_addr, unsigned int len)
2574 {
2575 	if (len > EFX_EF10_MAX_TX_DESCRIPTOR_LEN) {
2576 		/* If we need to break across multiple descriptors we should
2577 		 * stop at a page boundary. This assumes the length limit is
2578 		 * greater than the page size.
2579 		 */
2580 		dma_addr_t end = dma_addr + EFX_EF10_MAX_TX_DESCRIPTOR_LEN;
2581 
2582 		BUILD_BUG_ON(EFX_EF10_MAX_TX_DESCRIPTOR_LEN < EFX_PAGE_SIZE);
2583 		len = (end & (~(EFX_PAGE_SIZE - 1))) - dma_addr;
2584 	}
2585 
2586 	return len;
2587 }
2588 
2589 static void efx_ef10_tx_write(struct efx_tx_queue *tx_queue)
2590 {
2591 	unsigned int old_write_count = tx_queue->write_count;
2592 	struct efx_tx_buffer *buffer;
2593 	unsigned int write_ptr;
2594 	efx_qword_t *txd;
2595 
2596 	tx_queue->xmit_more_available = false;
2597 	if (unlikely(tx_queue->write_count == tx_queue->insert_count))
2598 		return;
2599 
2600 	do {
2601 		write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
2602 		buffer = &tx_queue->buffer[write_ptr];
2603 		txd = efx_tx_desc(tx_queue, write_ptr);
2604 		++tx_queue->write_count;
2605 
2606 		/* Create TX descriptor ring entry */
2607 		if (buffer->flags & EFX_TX_BUF_OPTION) {
2608 			*txd = buffer->option;
2609 			if (EFX_QWORD_FIELD(*txd, ESF_DZ_TX_OPTION_TYPE) == 1)
2610 				/* PIO descriptor */
2611 				tx_queue->packet_write_count = tx_queue->write_count;
2612 		} else {
2613 			tx_queue->packet_write_count = tx_queue->write_count;
2614 			BUILD_BUG_ON(EFX_TX_BUF_CONT != 1);
2615 			EFX_POPULATE_QWORD_3(
2616 				*txd,
2617 				ESF_DZ_TX_KER_CONT,
2618 				buffer->flags & EFX_TX_BUF_CONT,
2619 				ESF_DZ_TX_KER_BYTE_CNT, buffer->len,
2620 				ESF_DZ_TX_KER_BUF_ADDR, buffer->dma_addr);
2621 		}
2622 	} while (tx_queue->write_count != tx_queue->insert_count);
2623 
2624 	wmb(); /* Ensure descriptors are written before they are fetched */
2625 
2626 	if (efx_nic_may_push_tx_desc(tx_queue, old_write_count)) {
2627 		txd = efx_tx_desc(tx_queue,
2628 				  old_write_count & tx_queue->ptr_mask);
2629 		efx_ef10_push_tx_desc(tx_queue, txd);
2630 		++tx_queue->pushes;
2631 	} else {
2632 		efx_ef10_notify_tx_desc(tx_queue);
2633 	}
2634 }
2635 
2636 #define RSS_MODE_HASH_ADDRS	(1 << RSS_MODE_HASH_SRC_ADDR_LBN |\
2637 				 1 << RSS_MODE_HASH_DST_ADDR_LBN)
2638 #define RSS_MODE_HASH_PORTS	(1 << RSS_MODE_HASH_SRC_PORT_LBN |\
2639 				 1 << RSS_MODE_HASH_DST_PORT_LBN)
2640 #define RSS_CONTEXT_FLAGS_DEFAULT	(1 << MC_CMD_RSS_CONTEXT_GET_FLAGS_OUT_TOEPLITZ_IPV4_EN_LBN |\
2641 					 1 << MC_CMD_RSS_CONTEXT_GET_FLAGS_OUT_TOEPLITZ_TCPV4_EN_LBN |\
2642 					 1 << MC_CMD_RSS_CONTEXT_GET_FLAGS_OUT_TOEPLITZ_IPV6_EN_LBN |\
2643 					 1 << MC_CMD_RSS_CONTEXT_GET_FLAGS_OUT_TOEPLITZ_TCPV6_EN_LBN |\
2644 					 (RSS_MODE_HASH_ADDRS | RSS_MODE_HASH_PORTS) << MC_CMD_RSS_CONTEXT_GET_FLAGS_OUT_TCP_IPV4_RSS_MODE_LBN |\
2645 					 RSS_MODE_HASH_ADDRS << MC_CMD_RSS_CONTEXT_GET_FLAGS_OUT_UDP_IPV4_RSS_MODE_LBN |\
2646 					 RSS_MODE_HASH_ADDRS << MC_CMD_RSS_CONTEXT_GET_FLAGS_OUT_OTHER_IPV4_RSS_MODE_LBN |\
2647 					 (RSS_MODE_HASH_ADDRS | RSS_MODE_HASH_PORTS) << MC_CMD_RSS_CONTEXT_GET_FLAGS_OUT_TCP_IPV6_RSS_MODE_LBN |\
2648 					 RSS_MODE_HASH_ADDRS << MC_CMD_RSS_CONTEXT_GET_FLAGS_OUT_UDP_IPV6_RSS_MODE_LBN |\
2649 					 RSS_MODE_HASH_ADDRS << MC_CMD_RSS_CONTEXT_GET_FLAGS_OUT_OTHER_IPV6_RSS_MODE_LBN)
2650 
2651 static int efx_ef10_get_rss_flags(struct efx_nic *efx, u32 context, u32 *flags)
2652 {
2653 	/* Firmware had a bug (sfc bug 61952) where it would not actually
2654 	 * fill in the flags field in the response to MC_CMD_RSS_CONTEXT_GET_FLAGS.
2655 	 * This meant that it would always contain whatever was previously
2656 	 * in the MCDI buffer.  Fortunately, all firmware versions with
2657 	 * this bug have the same default flags value for a newly-allocated
2658 	 * RSS context, and the only time we want to get the flags is just
2659 	 * after allocating.  Moreover, the response has a 32-bit hole
2660 	 * where the context ID would be in the request, so we can use an
2661 	 * overlength buffer in the request and pre-fill the flags field
2662 	 * with what we believe the default to be.  Thus if the firmware
2663 	 * has the bug, it will leave our pre-filled value in the flags
2664 	 * field of the response, and we will get the right answer.
2665 	 *
2666 	 * However, this does mean that this function should NOT be used if
2667 	 * the RSS context flags might not be their defaults - it is ONLY
2668 	 * reliably correct for a newly-allocated RSS context.
2669 	 */
2670 	MCDI_DECLARE_BUF(inbuf, MC_CMD_RSS_CONTEXT_GET_FLAGS_OUT_LEN);
2671 	MCDI_DECLARE_BUF(outbuf, MC_CMD_RSS_CONTEXT_GET_FLAGS_OUT_LEN);
2672 	size_t outlen;
2673 	int rc;
2674 
2675 	/* Check we have a hole for the context ID */
2676 	BUILD_BUG_ON(MC_CMD_RSS_CONTEXT_GET_FLAGS_IN_LEN != MC_CMD_RSS_CONTEXT_GET_FLAGS_OUT_FLAGS_OFST);
2677 	MCDI_SET_DWORD(inbuf, RSS_CONTEXT_GET_FLAGS_IN_RSS_CONTEXT_ID, context);
2678 	MCDI_SET_DWORD(inbuf, RSS_CONTEXT_GET_FLAGS_OUT_FLAGS,
2679 		       RSS_CONTEXT_FLAGS_DEFAULT);
2680 	rc = efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_GET_FLAGS, inbuf,
2681 			  sizeof(inbuf), outbuf, sizeof(outbuf), &outlen);
2682 	if (rc == 0) {
2683 		if (outlen < MC_CMD_RSS_CONTEXT_GET_FLAGS_OUT_LEN)
2684 			rc = -EIO;
2685 		else
2686 			*flags = MCDI_DWORD(outbuf, RSS_CONTEXT_GET_FLAGS_OUT_FLAGS);
2687 	}
2688 	return rc;
2689 }
2690 
2691 /* Attempt to enable 4-tuple UDP hashing on the specified RSS context.
2692  * If we fail, we just leave the RSS context at its default hash settings,
2693  * which is safe but may slightly reduce performance.
2694  * Defaults are 4-tuple for TCP and 2-tuple for UDP and other-IP, so we
2695  * just need to set the UDP ports flags (for both IP versions).
2696  */
2697 static void efx_ef10_set_rss_flags(struct efx_nic *efx,
2698 				   struct efx_rss_context *ctx)
2699 {
2700 	MCDI_DECLARE_BUF(inbuf, MC_CMD_RSS_CONTEXT_SET_FLAGS_IN_LEN);
2701 	u32 flags;
2702 
2703 	BUILD_BUG_ON(MC_CMD_RSS_CONTEXT_SET_FLAGS_OUT_LEN != 0);
2704 
2705 	if (efx_ef10_get_rss_flags(efx, ctx->context_id, &flags) != 0)
2706 		return;
2707 	MCDI_SET_DWORD(inbuf, RSS_CONTEXT_SET_FLAGS_IN_RSS_CONTEXT_ID,
2708 		       ctx->context_id);
2709 	flags |= RSS_MODE_HASH_PORTS << MC_CMD_RSS_CONTEXT_GET_FLAGS_OUT_UDP_IPV4_RSS_MODE_LBN;
2710 	flags |= RSS_MODE_HASH_PORTS << MC_CMD_RSS_CONTEXT_GET_FLAGS_OUT_UDP_IPV6_RSS_MODE_LBN;
2711 	MCDI_SET_DWORD(inbuf, RSS_CONTEXT_SET_FLAGS_IN_FLAGS, flags);
2712 	if (!efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_SET_FLAGS, inbuf, sizeof(inbuf),
2713 			  NULL, 0, NULL))
2714 		/* Succeeded, so UDP 4-tuple is now enabled */
2715 		ctx->rx_hash_udp_4tuple = true;
2716 }
2717 
2718 static int efx_ef10_alloc_rss_context(struct efx_nic *efx, bool exclusive,
2719 				      struct efx_rss_context *ctx,
2720 				      unsigned *context_size)
2721 {
2722 	MCDI_DECLARE_BUF(inbuf, MC_CMD_RSS_CONTEXT_ALLOC_IN_LEN);
2723 	MCDI_DECLARE_BUF(outbuf, MC_CMD_RSS_CONTEXT_ALLOC_OUT_LEN);
2724 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2725 	size_t outlen;
2726 	int rc;
2727 	u32 alloc_type = exclusive ?
2728 				MC_CMD_RSS_CONTEXT_ALLOC_IN_TYPE_EXCLUSIVE :
2729 				MC_CMD_RSS_CONTEXT_ALLOC_IN_TYPE_SHARED;
2730 	unsigned rss_spread = exclusive ?
2731 				efx->rss_spread :
2732 				min(rounddown_pow_of_two(efx->rss_spread),
2733 				    EFX_EF10_MAX_SHARED_RSS_CONTEXT_SIZE);
2734 
2735 	if (!exclusive && rss_spread == 1) {
2736 		ctx->context_id = EFX_EF10_RSS_CONTEXT_INVALID;
2737 		if (context_size)
2738 			*context_size = 1;
2739 		return 0;
2740 	}
2741 
2742 	if (nic_data->datapath_caps &
2743 	    1 << MC_CMD_GET_CAPABILITIES_OUT_RX_RSS_LIMITED_LBN)
2744 		return -EOPNOTSUPP;
2745 
2746 	MCDI_SET_DWORD(inbuf, RSS_CONTEXT_ALLOC_IN_UPSTREAM_PORT_ID,
2747 		       nic_data->vport_id);
2748 	MCDI_SET_DWORD(inbuf, RSS_CONTEXT_ALLOC_IN_TYPE, alloc_type);
2749 	MCDI_SET_DWORD(inbuf, RSS_CONTEXT_ALLOC_IN_NUM_QUEUES, rss_spread);
2750 
2751 	rc = efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_ALLOC, inbuf, sizeof(inbuf),
2752 		outbuf, sizeof(outbuf), &outlen);
2753 	if (rc != 0)
2754 		return rc;
2755 
2756 	if (outlen < MC_CMD_RSS_CONTEXT_ALLOC_OUT_LEN)
2757 		return -EIO;
2758 
2759 	ctx->context_id = MCDI_DWORD(outbuf, RSS_CONTEXT_ALLOC_OUT_RSS_CONTEXT_ID);
2760 
2761 	if (context_size)
2762 		*context_size = rss_spread;
2763 
2764 	if (nic_data->datapath_caps &
2765 	    1 << MC_CMD_GET_CAPABILITIES_OUT_ADDITIONAL_RSS_MODES_LBN)
2766 		efx_ef10_set_rss_flags(efx, ctx);
2767 
2768 	return 0;
2769 }
2770 
2771 static int efx_ef10_free_rss_context(struct efx_nic *efx, u32 context)
2772 {
2773 	MCDI_DECLARE_BUF(inbuf, MC_CMD_RSS_CONTEXT_FREE_IN_LEN);
2774 
2775 	MCDI_SET_DWORD(inbuf, RSS_CONTEXT_FREE_IN_RSS_CONTEXT_ID,
2776 		       context);
2777 	return efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_FREE, inbuf, sizeof(inbuf),
2778 			    NULL, 0, NULL);
2779 }
2780 
2781 static int efx_ef10_populate_rss_table(struct efx_nic *efx, u32 context,
2782 				       const u32 *rx_indir_table, const u8 *key)
2783 {
2784 	MCDI_DECLARE_BUF(tablebuf, MC_CMD_RSS_CONTEXT_SET_TABLE_IN_LEN);
2785 	MCDI_DECLARE_BUF(keybuf, MC_CMD_RSS_CONTEXT_SET_KEY_IN_LEN);
2786 	int i, rc;
2787 
2788 	MCDI_SET_DWORD(tablebuf, RSS_CONTEXT_SET_TABLE_IN_RSS_CONTEXT_ID,
2789 		       context);
2790 	BUILD_BUG_ON(ARRAY_SIZE(efx->rss_context.rx_indir_table) !=
2791 		     MC_CMD_RSS_CONTEXT_SET_TABLE_IN_INDIRECTION_TABLE_LEN);
2792 
2793 	/* This iterates over the length of efx->rss_context.rx_indir_table, but
2794 	 * copies bytes from rx_indir_table.  That's because the latter is a
2795 	 * pointer rather than an array, but should have the same length.
2796 	 * The efx->rss_context.rx_hash_key loop below is similar.
2797 	 */
2798 	for (i = 0; i < ARRAY_SIZE(efx->rss_context.rx_indir_table); ++i)
2799 		MCDI_PTR(tablebuf,
2800 			 RSS_CONTEXT_SET_TABLE_IN_INDIRECTION_TABLE)[i] =
2801 				(u8) rx_indir_table[i];
2802 
2803 	rc = efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_SET_TABLE, tablebuf,
2804 			  sizeof(tablebuf), NULL, 0, NULL);
2805 	if (rc != 0)
2806 		return rc;
2807 
2808 	MCDI_SET_DWORD(keybuf, RSS_CONTEXT_SET_KEY_IN_RSS_CONTEXT_ID,
2809 		       context);
2810 	BUILD_BUG_ON(ARRAY_SIZE(efx->rss_context.rx_hash_key) !=
2811 		     MC_CMD_RSS_CONTEXT_SET_KEY_IN_TOEPLITZ_KEY_LEN);
2812 	for (i = 0; i < ARRAY_SIZE(efx->rss_context.rx_hash_key); ++i)
2813 		MCDI_PTR(keybuf, RSS_CONTEXT_SET_KEY_IN_TOEPLITZ_KEY)[i] = key[i];
2814 
2815 	return efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_SET_KEY, keybuf,
2816 			    sizeof(keybuf), NULL, 0, NULL);
2817 }
2818 
2819 static void efx_ef10_rx_free_indir_table(struct efx_nic *efx)
2820 {
2821 	int rc;
2822 
2823 	if (efx->rss_context.context_id != EFX_EF10_RSS_CONTEXT_INVALID) {
2824 		rc = efx_ef10_free_rss_context(efx, efx->rss_context.context_id);
2825 		WARN_ON(rc != 0);
2826 	}
2827 	efx->rss_context.context_id = EFX_EF10_RSS_CONTEXT_INVALID;
2828 }
2829 
2830 static int efx_ef10_rx_push_shared_rss_config(struct efx_nic *efx,
2831 					      unsigned *context_size)
2832 {
2833 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2834 	int rc = efx_ef10_alloc_rss_context(efx, false, &efx->rss_context,
2835 					    context_size);
2836 
2837 	if (rc != 0)
2838 		return rc;
2839 
2840 	nic_data->rx_rss_context_exclusive = false;
2841 	efx_set_default_rx_indir_table(efx, &efx->rss_context);
2842 	return 0;
2843 }
2844 
2845 static int efx_ef10_rx_push_exclusive_rss_config(struct efx_nic *efx,
2846 						 const u32 *rx_indir_table,
2847 						 const u8 *key)
2848 {
2849 	u32 old_rx_rss_context = efx->rss_context.context_id;
2850 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2851 	int rc;
2852 
2853 	if (efx->rss_context.context_id == EFX_EF10_RSS_CONTEXT_INVALID ||
2854 	    !nic_data->rx_rss_context_exclusive) {
2855 		rc = efx_ef10_alloc_rss_context(efx, true, &efx->rss_context,
2856 						NULL);
2857 		if (rc == -EOPNOTSUPP)
2858 			return rc;
2859 		else if (rc != 0)
2860 			goto fail1;
2861 	}
2862 
2863 	rc = efx_ef10_populate_rss_table(efx, efx->rss_context.context_id,
2864 					 rx_indir_table, key);
2865 	if (rc != 0)
2866 		goto fail2;
2867 
2868 	if (efx->rss_context.context_id != old_rx_rss_context &&
2869 	    old_rx_rss_context != EFX_EF10_RSS_CONTEXT_INVALID)
2870 		WARN_ON(efx_ef10_free_rss_context(efx, old_rx_rss_context) != 0);
2871 	nic_data->rx_rss_context_exclusive = true;
2872 	if (rx_indir_table != efx->rss_context.rx_indir_table)
2873 		memcpy(efx->rss_context.rx_indir_table, rx_indir_table,
2874 		       sizeof(efx->rss_context.rx_indir_table));
2875 	if (key != efx->rss_context.rx_hash_key)
2876 		memcpy(efx->rss_context.rx_hash_key, key,
2877 		       efx->type->rx_hash_key_size);
2878 
2879 	return 0;
2880 
2881 fail2:
2882 	if (old_rx_rss_context != efx->rss_context.context_id) {
2883 		WARN_ON(efx_ef10_free_rss_context(efx, efx->rss_context.context_id) != 0);
2884 		efx->rss_context.context_id = old_rx_rss_context;
2885 	}
2886 fail1:
2887 	netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
2888 	return rc;
2889 }
2890 
2891 static int efx_ef10_rx_push_rss_context_config(struct efx_nic *efx,
2892 					       struct efx_rss_context *ctx,
2893 					       const u32 *rx_indir_table,
2894 					       const u8 *key)
2895 {
2896 	int rc;
2897 
2898 	WARN_ON(!mutex_is_locked(&efx->rss_lock));
2899 
2900 	if (ctx->context_id == EFX_EF10_RSS_CONTEXT_INVALID) {
2901 		rc = efx_ef10_alloc_rss_context(efx, true, ctx, NULL);
2902 		if (rc)
2903 			return rc;
2904 	}
2905 
2906 	if (!rx_indir_table) /* Delete this context */
2907 		return efx_ef10_free_rss_context(efx, ctx->context_id);
2908 
2909 	rc = efx_ef10_populate_rss_table(efx, ctx->context_id,
2910 					 rx_indir_table, key);
2911 	if (rc)
2912 		return rc;
2913 
2914 	memcpy(ctx->rx_indir_table, rx_indir_table,
2915 	       sizeof(efx->rss_context.rx_indir_table));
2916 	memcpy(ctx->rx_hash_key, key, efx->type->rx_hash_key_size);
2917 
2918 	return 0;
2919 }
2920 
2921 static int efx_ef10_rx_pull_rss_context_config(struct efx_nic *efx,
2922 					       struct efx_rss_context *ctx)
2923 {
2924 	MCDI_DECLARE_BUF(inbuf, MC_CMD_RSS_CONTEXT_GET_TABLE_IN_LEN);
2925 	MCDI_DECLARE_BUF(tablebuf, MC_CMD_RSS_CONTEXT_GET_TABLE_OUT_LEN);
2926 	MCDI_DECLARE_BUF(keybuf, MC_CMD_RSS_CONTEXT_GET_KEY_OUT_LEN);
2927 	size_t outlen;
2928 	int rc, i;
2929 
2930 	WARN_ON(!mutex_is_locked(&efx->rss_lock));
2931 
2932 	BUILD_BUG_ON(MC_CMD_RSS_CONTEXT_GET_TABLE_IN_LEN !=
2933 		     MC_CMD_RSS_CONTEXT_GET_KEY_IN_LEN);
2934 
2935 	if (ctx->context_id == EFX_EF10_RSS_CONTEXT_INVALID)
2936 		return -ENOENT;
2937 
2938 	MCDI_SET_DWORD(inbuf, RSS_CONTEXT_GET_TABLE_IN_RSS_CONTEXT_ID,
2939 		       ctx->context_id);
2940 	BUILD_BUG_ON(ARRAY_SIZE(ctx->rx_indir_table) !=
2941 		     MC_CMD_RSS_CONTEXT_GET_TABLE_OUT_INDIRECTION_TABLE_LEN);
2942 	rc = efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_GET_TABLE, inbuf, sizeof(inbuf),
2943 			  tablebuf, sizeof(tablebuf), &outlen);
2944 	if (rc != 0)
2945 		return rc;
2946 
2947 	if (WARN_ON(outlen != MC_CMD_RSS_CONTEXT_GET_TABLE_OUT_LEN))
2948 		return -EIO;
2949 
2950 	for (i = 0; i < ARRAY_SIZE(ctx->rx_indir_table); i++)
2951 		ctx->rx_indir_table[i] = MCDI_PTR(tablebuf,
2952 				RSS_CONTEXT_GET_TABLE_OUT_INDIRECTION_TABLE)[i];
2953 
2954 	MCDI_SET_DWORD(inbuf, RSS_CONTEXT_GET_KEY_IN_RSS_CONTEXT_ID,
2955 		       ctx->context_id);
2956 	BUILD_BUG_ON(ARRAY_SIZE(ctx->rx_hash_key) !=
2957 		     MC_CMD_RSS_CONTEXT_SET_KEY_IN_TOEPLITZ_KEY_LEN);
2958 	rc = efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_GET_KEY, inbuf, sizeof(inbuf),
2959 			  keybuf, sizeof(keybuf), &outlen);
2960 	if (rc != 0)
2961 		return rc;
2962 
2963 	if (WARN_ON(outlen != MC_CMD_RSS_CONTEXT_GET_KEY_OUT_LEN))
2964 		return -EIO;
2965 
2966 	for (i = 0; i < ARRAY_SIZE(ctx->rx_hash_key); ++i)
2967 		ctx->rx_hash_key[i] = MCDI_PTR(
2968 				keybuf, RSS_CONTEXT_GET_KEY_OUT_TOEPLITZ_KEY)[i];
2969 
2970 	return 0;
2971 }
2972 
2973 static int efx_ef10_rx_pull_rss_config(struct efx_nic *efx)
2974 {
2975 	int rc;
2976 
2977 	mutex_lock(&efx->rss_lock);
2978 	rc = efx_ef10_rx_pull_rss_context_config(efx, &efx->rss_context);
2979 	mutex_unlock(&efx->rss_lock);
2980 	return rc;
2981 }
2982 
2983 static void efx_ef10_rx_restore_rss_contexts(struct efx_nic *efx)
2984 {
2985 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2986 	struct efx_rss_context *ctx;
2987 	int rc;
2988 
2989 	WARN_ON(!mutex_is_locked(&efx->rss_lock));
2990 
2991 	if (!nic_data->must_restore_rss_contexts)
2992 		return;
2993 
2994 	list_for_each_entry(ctx, &efx->rss_context.list, list) {
2995 		/* previous NIC RSS context is gone */
2996 		ctx->context_id = EFX_EF10_RSS_CONTEXT_INVALID;
2997 		/* so try to allocate a new one */
2998 		rc = efx_ef10_rx_push_rss_context_config(efx, ctx,
2999 							 ctx->rx_indir_table,
3000 							 ctx->rx_hash_key);
3001 		if (rc)
3002 			netif_warn(efx, probe, efx->net_dev,
3003 				   "failed to restore RSS context %u, rc=%d"
3004 				   "; RSS filters may fail to be applied\n",
3005 				   ctx->user_id, rc);
3006 	}
3007 	nic_data->must_restore_rss_contexts = false;
3008 }
3009 
3010 static int efx_ef10_pf_rx_push_rss_config(struct efx_nic *efx, bool user,
3011 					  const u32 *rx_indir_table,
3012 					  const u8 *key)
3013 {
3014 	int rc;
3015 
3016 	if (efx->rss_spread == 1)
3017 		return 0;
3018 
3019 	if (!key)
3020 		key = efx->rss_context.rx_hash_key;
3021 
3022 	rc = efx_ef10_rx_push_exclusive_rss_config(efx, rx_indir_table, key);
3023 
3024 	if (rc == -ENOBUFS && !user) {
3025 		unsigned context_size;
3026 		bool mismatch = false;
3027 		size_t i;
3028 
3029 		for (i = 0;
3030 		     i < ARRAY_SIZE(efx->rss_context.rx_indir_table) && !mismatch;
3031 		     i++)
3032 			mismatch = rx_indir_table[i] !=
3033 				ethtool_rxfh_indir_default(i, efx->rss_spread);
3034 
3035 		rc = efx_ef10_rx_push_shared_rss_config(efx, &context_size);
3036 		if (rc == 0) {
3037 			if (context_size != efx->rss_spread)
3038 				netif_warn(efx, probe, efx->net_dev,
3039 					   "Could not allocate an exclusive RSS"
3040 					   " context; allocated a shared one of"
3041 					   " different size."
3042 					   " Wanted %u, got %u.\n",
3043 					   efx->rss_spread, context_size);
3044 			else if (mismatch)
3045 				netif_warn(efx, probe, efx->net_dev,
3046 					   "Could not allocate an exclusive RSS"
3047 					   " context; allocated a shared one but"
3048 					   " could not apply custom"
3049 					   " indirection.\n");
3050 			else
3051 				netif_info(efx, probe, efx->net_dev,
3052 					   "Could not allocate an exclusive RSS"
3053 					   " context; allocated a shared one.\n");
3054 		}
3055 	}
3056 	return rc;
3057 }
3058 
3059 static int efx_ef10_vf_rx_push_rss_config(struct efx_nic *efx, bool user,
3060 					  const u32 *rx_indir_table
3061 					  __attribute__ ((unused)),
3062 					  const u8 *key
3063 					  __attribute__ ((unused)))
3064 {
3065 	if (user)
3066 		return -EOPNOTSUPP;
3067 	if (efx->rss_context.context_id != EFX_EF10_RSS_CONTEXT_INVALID)
3068 		return 0;
3069 	return efx_ef10_rx_push_shared_rss_config(efx, NULL);
3070 }
3071 
3072 static int efx_ef10_rx_probe(struct efx_rx_queue *rx_queue)
3073 {
3074 	return efx_nic_alloc_buffer(rx_queue->efx, &rx_queue->rxd.buf,
3075 				    (rx_queue->ptr_mask + 1) *
3076 				    sizeof(efx_qword_t),
3077 				    GFP_KERNEL);
3078 }
3079 
3080 static void efx_ef10_rx_init(struct efx_rx_queue *rx_queue)
3081 {
3082 	MCDI_DECLARE_BUF(inbuf,
3083 			 MC_CMD_INIT_RXQ_IN_LEN(EFX_MAX_DMAQ_SIZE * 8 /
3084 						EFX_BUF_SIZE));
3085 	struct efx_channel *channel = efx_rx_queue_channel(rx_queue);
3086 	size_t entries = rx_queue->rxd.buf.len / EFX_BUF_SIZE;
3087 	struct efx_nic *efx = rx_queue->efx;
3088 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
3089 	size_t inlen;
3090 	dma_addr_t dma_addr;
3091 	int rc;
3092 	int i;
3093 	BUILD_BUG_ON(MC_CMD_INIT_RXQ_OUT_LEN != 0);
3094 
3095 	rx_queue->scatter_n = 0;
3096 	rx_queue->scatter_len = 0;
3097 
3098 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_SIZE, rx_queue->ptr_mask + 1);
3099 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_TARGET_EVQ, channel->channel);
3100 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_LABEL, efx_rx_queue_index(rx_queue));
3101 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_INSTANCE,
3102 		       efx_rx_queue_index(rx_queue));
3103 	MCDI_POPULATE_DWORD_2(inbuf, INIT_RXQ_IN_FLAGS,
3104 			      INIT_RXQ_IN_FLAG_PREFIX, 1,
3105 			      INIT_RXQ_IN_FLAG_TIMESTAMP, 1);
3106 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_OWNER_ID, 0);
3107 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_PORT_ID, nic_data->vport_id);
3108 
3109 	dma_addr = rx_queue->rxd.buf.dma_addr;
3110 
3111 	netif_dbg(efx, hw, efx->net_dev, "pushing RXQ %d. %zu entries (%llx)\n",
3112 		  efx_rx_queue_index(rx_queue), entries, (u64)dma_addr);
3113 
3114 	for (i = 0; i < entries; ++i) {
3115 		MCDI_SET_ARRAY_QWORD(inbuf, INIT_RXQ_IN_DMA_ADDR, i, dma_addr);
3116 		dma_addr += EFX_BUF_SIZE;
3117 	}
3118 
3119 	inlen = MC_CMD_INIT_RXQ_IN_LEN(entries);
3120 
3121 	rc = efx_mcdi_rpc(efx, MC_CMD_INIT_RXQ, inbuf, inlen,
3122 			  NULL, 0, NULL);
3123 	if (rc)
3124 		netdev_WARN(efx->net_dev, "failed to initialise RXQ %d\n",
3125 			    efx_rx_queue_index(rx_queue));
3126 }
3127 
3128 static void efx_ef10_rx_fini(struct efx_rx_queue *rx_queue)
3129 {
3130 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FINI_RXQ_IN_LEN);
3131 	MCDI_DECLARE_BUF_ERR(outbuf);
3132 	struct efx_nic *efx = rx_queue->efx;
3133 	size_t outlen;
3134 	int rc;
3135 
3136 	MCDI_SET_DWORD(inbuf, FINI_RXQ_IN_INSTANCE,
3137 		       efx_rx_queue_index(rx_queue));
3138 
3139 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FINI_RXQ, inbuf, sizeof(inbuf),
3140 			  outbuf, sizeof(outbuf), &outlen);
3141 
3142 	if (rc && rc != -EALREADY)
3143 		goto fail;
3144 
3145 	return;
3146 
3147 fail:
3148 	efx_mcdi_display_error(efx, MC_CMD_FINI_RXQ, MC_CMD_FINI_RXQ_IN_LEN,
3149 			       outbuf, outlen, rc);
3150 }
3151 
3152 static void efx_ef10_rx_remove(struct efx_rx_queue *rx_queue)
3153 {
3154 	efx_nic_free_buffer(rx_queue->efx, &rx_queue->rxd.buf);
3155 }
3156 
3157 /* This creates an entry in the RX descriptor queue */
3158 static inline void
3159 efx_ef10_build_rx_desc(struct efx_rx_queue *rx_queue, unsigned int index)
3160 {
3161 	struct efx_rx_buffer *rx_buf;
3162 	efx_qword_t *rxd;
3163 
3164 	rxd = efx_rx_desc(rx_queue, index);
3165 	rx_buf = efx_rx_buffer(rx_queue, index);
3166 	EFX_POPULATE_QWORD_2(*rxd,
3167 			     ESF_DZ_RX_KER_BYTE_CNT, rx_buf->len,
3168 			     ESF_DZ_RX_KER_BUF_ADDR, rx_buf->dma_addr);
3169 }
3170 
3171 static void efx_ef10_rx_write(struct efx_rx_queue *rx_queue)
3172 {
3173 	struct efx_nic *efx = rx_queue->efx;
3174 	unsigned int write_count;
3175 	efx_dword_t reg;
3176 
3177 	/* Firmware requires that RX_DESC_WPTR be a multiple of 8 */
3178 	write_count = rx_queue->added_count & ~7;
3179 	if (rx_queue->notified_count == write_count)
3180 		return;
3181 
3182 	do
3183 		efx_ef10_build_rx_desc(
3184 			rx_queue,
3185 			rx_queue->notified_count & rx_queue->ptr_mask);
3186 	while (++rx_queue->notified_count != write_count);
3187 
3188 	wmb();
3189 	EFX_POPULATE_DWORD_1(reg, ERF_DZ_RX_DESC_WPTR,
3190 			     write_count & rx_queue->ptr_mask);
3191 	efx_writed_page(efx, &reg, ER_DZ_RX_DESC_UPD,
3192 			efx_rx_queue_index(rx_queue));
3193 }
3194 
3195 static efx_mcdi_async_completer efx_ef10_rx_defer_refill_complete;
3196 
3197 static void efx_ef10_rx_defer_refill(struct efx_rx_queue *rx_queue)
3198 {
3199 	struct efx_channel *channel = efx_rx_queue_channel(rx_queue);
3200 	MCDI_DECLARE_BUF(inbuf, MC_CMD_DRIVER_EVENT_IN_LEN);
3201 	efx_qword_t event;
3202 
3203 	EFX_POPULATE_QWORD_2(event,
3204 			     ESF_DZ_EV_CODE, EFX_EF10_DRVGEN_EV,
3205 			     ESF_DZ_EV_DATA, EFX_EF10_REFILL);
3206 
3207 	MCDI_SET_DWORD(inbuf, DRIVER_EVENT_IN_EVQ, channel->channel);
3208 
3209 	/* MCDI_SET_QWORD is not appropriate here since EFX_POPULATE_* has
3210 	 * already swapped the data to little-endian order.
3211 	 */
3212 	memcpy(MCDI_PTR(inbuf, DRIVER_EVENT_IN_DATA), &event.u64[0],
3213 	       sizeof(efx_qword_t));
3214 
3215 	efx_mcdi_rpc_async(channel->efx, MC_CMD_DRIVER_EVENT,
3216 			   inbuf, sizeof(inbuf), 0,
3217 			   efx_ef10_rx_defer_refill_complete, 0);
3218 }
3219 
3220 static void
3221 efx_ef10_rx_defer_refill_complete(struct efx_nic *efx, unsigned long cookie,
3222 				  int rc, efx_dword_t *outbuf,
3223 				  size_t outlen_actual)
3224 {
3225 	/* nothing to do */
3226 }
3227 
3228 static int efx_ef10_ev_probe(struct efx_channel *channel)
3229 {
3230 	return efx_nic_alloc_buffer(channel->efx, &channel->eventq.buf,
3231 				    (channel->eventq_mask + 1) *
3232 				    sizeof(efx_qword_t),
3233 				    GFP_KERNEL);
3234 }
3235 
3236 static void efx_ef10_ev_fini(struct efx_channel *channel)
3237 {
3238 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FINI_EVQ_IN_LEN);
3239 	MCDI_DECLARE_BUF_ERR(outbuf);
3240 	struct efx_nic *efx = channel->efx;
3241 	size_t outlen;
3242 	int rc;
3243 
3244 	MCDI_SET_DWORD(inbuf, FINI_EVQ_IN_INSTANCE, channel->channel);
3245 
3246 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FINI_EVQ, inbuf, sizeof(inbuf),
3247 			  outbuf, sizeof(outbuf), &outlen);
3248 
3249 	if (rc && rc != -EALREADY)
3250 		goto fail;
3251 
3252 	return;
3253 
3254 fail:
3255 	efx_mcdi_display_error(efx, MC_CMD_FINI_EVQ, MC_CMD_FINI_EVQ_IN_LEN,
3256 			       outbuf, outlen, rc);
3257 }
3258 
3259 static int efx_ef10_ev_init(struct efx_channel *channel)
3260 {
3261 	MCDI_DECLARE_BUF(inbuf,
3262 			 MC_CMD_INIT_EVQ_V2_IN_LEN(EFX_MAX_EVQ_SIZE * 8 /
3263 						   EFX_BUF_SIZE));
3264 	MCDI_DECLARE_BUF(outbuf, MC_CMD_INIT_EVQ_V2_OUT_LEN);
3265 	size_t entries = channel->eventq.buf.len / EFX_BUF_SIZE;
3266 	struct efx_nic *efx = channel->efx;
3267 	struct efx_ef10_nic_data *nic_data;
3268 	size_t inlen, outlen;
3269 	unsigned int enabled, implemented;
3270 	dma_addr_t dma_addr;
3271 	int rc;
3272 	int i;
3273 
3274 	nic_data = efx->nic_data;
3275 
3276 	/* Fill event queue with all ones (i.e. empty events) */
3277 	memset(channel->eventq.buf.addr, 0xff, channel->eventq.buf.len);
3278 
3279 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_SIZE, channel->eventq_mask + 1);
3280 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_INSTANCE, channel->channel);
3281 	/* INIT_EVQ expects index in vector table, not absolute */
3282 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_IRQ_NUM, channel->channel);
3283 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_TMR_MODE,
3284 		       MC_CMD_INIT_EVQ_IN_TMR_MODE_DIS);
3285 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_TMR_LOAD, 0);
3286 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_TMR_RELOAD, 0);
3287 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_COUNT_MODE,
3288 		       MC_CMD_INIT_EVQ_IN_COUNT_MODE_DIS);
3289 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_COUNT_THRSHLD, 0);
3290 
3291 	if (nic_data->datapath_caps2 &
3292 	    1 << MC_CMD_GET_CAPABILITIES_V2_OUT_INIT_EVQ_V2_LBN) {
3293 		/* Use the new generic approach to specifying event queue
3294 		 * configuration, requesting lower latency or higher throughput.
3295 		 * The options that actually get used appear in the output.
3296 		 */
3297 		MCDI_POPULATE_DWORD_2(inbuf, INIT_EVQ_V2_IN_FLAGS,
3298 				      INIT_EVQ_V2_IN_FLAG_INTERRUPTING, 1,
3299 				      INIT_EVQ_V2_IN_FLAG_TYPE,
3300 				      MC_CMD_INIT_EVQ_V2_IN_FLAG_TYPE_AUTO);
3301 	} else {
3302 		bool cut_thru = !(nic_data->datapath_caps &
3303 			1 << MC_CMD_GET_CAPABILITIES_OUT_RX_BATCHING_LBN);
3304 
3305 		MCDI_POPULATE_DWORD_4(inbuf, INIT_EVQ_IN_FLAGS,
3306 				      INIT_EVQ_IN_FLAG_INTERRUPTING, 1,
3307 				      INIT_EVQ_IN_FLAG_RX_MERGE, 1,
3308 				      INIT_EVQ_IN_FLAG_TX_MERGE, 1,
3309 				      INIT_EVQ_IN_FLAG_CUT_THRU, cut_thru);
3310 	}
3311 
3312 	dma_addr = channel->eventq.buf.dma_addr;
3313 	for (i = 0; i < entries; ++i) {
3314 		MCDI_SET_ARRAY_QWORD(inbuf, INIT_EVQ_IN_DMA_ADDR, i, dma_addr);
3315 		dma_addr += EFX_BUF_SIZE;
3316 	}
3317 
3318 	inlen = MC_CMD_INIT_EVQ_IN_LEN(entries);
3319 
3320 	rc = efx_mcdi_rpc(efx, MC_CMD_INIT_EVQ, inbuf, inlen,
3321 			  outbuf, sizeof(outbuf), &outlen);
3322 
3323 	if (outlen >= MC_CMD_INIT_EVQ_V2_OUT_LEN)
3324 		netif_dbg(efx, drv, efx->net_dev,
3325 			  "Channel %d using event queue flags %08x\n",
3326 			  channel->channel,
3327 			  MCDI_DWORD(outbuf, INIT_EVQ_V2_OUT_FLAGS));
3328 
3329 	/* IRQ return is ignored */
3330 	if (channel->channel || rc)
3331 		return rc;
3332 
3333 	/* Successfully created event queue on channel 0 */
3334 	rc = efx_mcdi_get_workarounds(efx, &implemented, &enabled);
3335 	if (rc == -ENOSYS) {
3336 		/* GET_WORKAROUNDS was implemented before this workaround,
3337 		 * thus it must be unavailable in this firmware.
3338 		 */
3339 		nic_data->workaround_26807 = false;
3340 		rc = 0;
3341 	} else if (rc) {
3342 		goto fail;
3343 	} else {
3344 		nic_data->workaround_26807 =
3345 			!!(enabled & MC_CMD_GET_WORKAROUNDS_OUT_BUG26807);
3346 
3347 		if (implemented & MC_CMD_GET_WORKAROUNDS_OUT_BUG26807 &&
3348 		    !nic_data->workaround_26807) {
3349 			unsigned int flags;
3350 
3351 			rc = efx_mcdi_set_workaround(efx,
3352 						     MC_CMD_WORKAROUND_BUG26807,
3353 						     true, &flags);
3354 
3355 			if (!rc) {
3356 				if (flags &
3357 				    1 << MC_CMD_WORKAROUND_EXT_OUT_FLR_DONE_LBN) {
3358 					netif_info(efx, drv, efx->net_dev,
3359 						   "other functions on NIC have been reset\n");
3360 
3361 					/* With MCFW v4.6.x and earlier, the
3362 					 * boot count will have incremented,
3363 					 * so re-read the warm_boot_count
3364 					 * value now to ensure this function
3365 					 * doesn't think it has changed next
3366 					 * time it checks.
3367 					 */
3368 					rc = efx_ef10_get_warm_boot_count(efx);
3369 					if (rc >= 0) {
3370 						nic_data->warm_boot_count = rc;
3371 						rc = 0;
3372 					}
3373 				}
3374 				nic_data->workaround_26807 = true;
3375 			} else if (rc == -EPERM) {
3376 				rc = 0;
3377 			}
3378 		}
3379 	}
3380 
3381 	if (!rc)
3382 		return 0;
3383 
3384 fail:
3385 	efx_ef10_ev_fini(channel);
3386 	return rc;
3387 }
3388 
3389 static void efx_ef10_ev_remove(struct efx_channel *channel)
3390 {
3391 	efx_nic_free_buffer(channel->efx, &channel->eventq.buf);
3392 }
3393 
3394 static void efx_ef10_handle_rx_wrong_queue(struct efx_rx_queue *rx_queue,
3395 					   unsigned int rx_queue_label)
3396 {
3397 	struct efx_nic *efx = rx_queue->efx;
3398 
3399 	netif_info(efx, hw, efx->net_dev,
3400 		   "rx event arrived on queue %d labeled as queue %u\n",
3401 		   efx_rx_queue_index(rx_queue), rx_queue_label);
3402 
3403 	efx_schedule_reset(efx, RESET_TYPE_DISABLE);
3404 }
3405 
3406 static void
3407 efx_ef10_handle_rx_bad_lbits(struct efx_rx_queue *rx_queue,
3408 			     unsigned int actual, unsigned int expected)
3409 {
3410 	unsigned int dropped = (actual - expected) & rx_queue->ptr_mask;
3411 	struct efx_nic *efx = rx_queue->efx;
3412 
3413 	netif_info(efx, hw, efx->net_dev,
3414 		   "dropped %d events (index=%d expected=%d)\n",
3415 		   dropped, actual, expected);
3416 
3417 	efx_schedule_reset(efx, RESET_TYPE_DISABLE);
3418 }
3419 
3420 /* partially received RX was aborted. clean up. */
3421 static void efx_ef10_handle_rx_abort(struct efx_rx_queue *rx_queue)
3422 {
3423 	unsigned int rx_desc_ptr;
3424 
3425 	netif_dbg(rx_queue->efx, hw, rx_queue->efx->net_dev,
3426 		  "scattered RX aborted (dropping %u buffers)\n",
3427 		  rx_queue->scatter_n);
3428 
3429 	rx_desc_ptr = rx_queue->removed_count & rx_queue->ptr_mask;
3430 
3431 	efx_rx_packet(rx_queue, rx_desc_ptr, rx_queue->scatter_n,
3432 		      0, EFX_RX_PKT_DISCARD);
3433 
3434 	rx_queue->removed_count += rx_queue->scatter_n;
3435 	rx_queue->scatter_n = 0;
3436 	rx_queue->scatter_len = 0;
3437 	++efx_rx_queue_channel(rx_queue)->n_rx_nodesc_trunc;
3438 }
3439 
3440 static u16 efx_ef10_handle_rx_event_errors(struct efx_channel *channel,
3441 					   unsigned int n_packets,
3442 					   unsigned int rx_encap_hdr,
3443 					   unsigned int rx_l3_class,
3444 					   unsigned int rx_l4_class,
3445 					   const efx_qword_t *event)
3446 {
3447 	struct efx_nic *efx = channel->efx;
3448 	bool handled = false;
3449 
3450 	if (EFX_QWORD_FIELD(*event, ESF_DZ_RX_ECRC_ERR)) {
3451 		if (!(efx->net_dev->features & NETIF_F_RXALL)) {
3452 			if (!efx->loopback_selftest)
3453 				channel->n_rx_eth_crc_err += n_packets;
3454 			return EFX_RX_PKT_DISCARD;
3455 		}
3456 		handled = true;
3457 	}
3458 	if (EFX_QWORD_FIELD(*event, ESF_DZ_RX_IPCKSUM_ERR)) {
3459 		if (unlikely(rx_encap_hdr != ESE_EZ_ENCAP_HDR_VXLAN &&
3460 			     rx_l3_class != ESE_DZ_L3_CLASS_IP4 &&
3461 			     rx_l3_class != ESE_DZ_L3_CLASS_IP4_FRAG &&
3462 			     rx_l3_class != ESE_DZ_L3_CLASS_IP6 &&
3463 			     rx_l3_class != ESE_DZ_L3_CLASS_IP6_FRAG))
3464 			netdev_WARN(efx->net_dev,
3465 				    "invalid class for RX_IPCKSUM_ERR: event="
3466 				    EFX_QWORD_FMT "\n",
3467 				    EFX_QWORD_VAL(*event));
3468 		if (!efx->loopback_selftest)
3469 			*(rx_encap_hdr ?
3470 			  &channel->n_rx_outer_ip_hdr_chksum_err :
3471 			  &channel->n_rx_ip_hdr_chksum_err) += n_packets;
3472 		return 0;
3473 	}
3474 	if (EFX_QWORD_FIELD(*event, ESF_DZ_RX_TCPUDP_CKSUM_ERR)) {
3475 		if (unlikely(rx_encap_hdr != ESE_EZ_ENCAP_HDR_VXLAN &&
3476 			     ((rx_l3_class != ESE_DZ_L3_CLASS_IP4 &&
3477 			       rx_l3_class != ESE_DZ_L3_CLASS_IP6) ||
3478 			      (rx_l4_class != ESE_FZ_L4_CLASS_TCP &&
3479 			       rx_l4_class != ESE_FZ_L4_CLASS_UDP))))
3480 			netdev_WARN(efx->net_dev,
3481 				    "invalid class for RX_TCPUDP_CKSUM_ERR: event="
3482 				    EFX_QWORD_FMT "\n",
3483 				    EFX_QWORD_VAL(*event));
3484 		if (!efx->loopback_selftest)
3485 			*(rx_encap_hdr ?
3486 			  &channel->n_rx_outer_tcp_udp_chksum_err :
3487 			  &channel->n_rx_tcp_udp_chksum_err) += n_packets;
3488 		return 0;
3489 	}
3490 	if (EFX_QWORD_FIELD(*event, ESF_EZ_RX_IP_INNER_CHKSUM_ERR)) {
3491 		if (unlikely(!rx_encap_hdr))
3492 			netdev_WARN(efx->net_dev,
3493 				    "invalid encapsulation type for RX_IP_INNER_CHKSUM_ERR: event="
3494 				    EFX_QWORD_FMT "\n",
3495 				    EFX_QWORD_VAL(*event));
3496 		else if (unlikely(rx_l3_class != ESE_DZ_L3_CLASS_IP4 &&
3497 				  rx_l3_class != ESE_DZ_L3_CLASS_IP4_FRAG &&
3498 				  rx_l3_class != ESE_DZ_L3_CLASS_IP6 &&
3499 				  rx_l3_class != ESE_DZ_L3_CLASS_IP6_FRAG))
3500 			netdev_WARN(efx->net_dev,
3501 				    "invalid class for RX_IP_INNER_CHKSUM_ERR: event="
3502 				    EFX_QWORD_FMT "\n",
3503 				    EFX_QWORD_VAL(*event));
3504 		if (!efx->loopback_selftest)
3505 			channel->n_rx_inner_ip_hdr_chksum_err += n_packets;
3506 		return 0;
3507 	}
3508 	if (EFX_QWORD_FIELD(*event, ESF_EZ_RX_TCP_UDP_INNER_CHKSUM_ERR)) {
3509 		if (unlikely(!rx_encap_hdr))
3510 			netdev_WARN(efx->net_dev,
3511 				    "invalid encapsulation type for RX_TCP_UDP_INNER_CHKSUM_ERR: event="
3512 				    EFX_QWORD_FMT "\n",
3513 				    EFX_QWORD_VAL(*event));
3514 		else if (unlikely((rx_l3_class != ESE_DZ_L3_CLASS_IP4 &&
3515 				   rx_l3_class != ESE_DZ_L3_CLASS_IP6) ||
3516 				  (rx_l4_class != ESE_FZ_L4_CLASS_TCP &&
3517 				   rx_l4_class != ESE_FZ_L4_CLASS_UDP)))
3518 			netdev_WARN(efx->net_dev,
3519 				    "invalid class for RX_TCP_UDP_INNER_CHKSUM_ERR: event="
3520 				    EFX_QWORD_FMT "\n",
3521 				    EFX_QWORD_VAL(*event));
3522 		if (!efx->loopback_selftest)
3523 			channel->n_rx_inner_tcp_udp_chksum_err += n_packets;
3524 		return 0;
3525 	}
3526 
3527 	WARN_ON(!handled); /* No error bits were recognised */
3528 	return 0;
3529 }
3530 
3531 static int efx_ef10_handle_rx_event(struct efx_channel *channel,
3532 				    const efx_qword_t *event)
3533 {
3534 	unsigned int rx_bytes, next_ptr_lbits, rx_queue_label;
3535 	unsigned int rx_l3_class, rx_l4_class, rx_encap_hdr;
3536 	unsigned int n_descs, n_packets, i;
3537 	struct efx_nic *efx = channel->efx;
3538 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
3539 	struct efx_rx_queue *rx_queue;
3540 	efx_qword_t errors;
3541 	bool rx_cont;
3542 	u16 flags = 0;
3543 
3544 	if (unlikely(READ_ONCE(efx->reset_pending)))
3545 		return 0;
3546 
3547 	/* Basic packet information */
3548 	rx_bytes = EFX_QWORD_FIELD(*event, ESF_DZ_RX_BYTES);
3549 	next_ptr_lbits = EFX_QWORD_FIELD(*event, ESF_DZ_RX_DSC_PTR_LBITS);
3550 	rx_queue_label = EFX_QWORD_FIELD(*event, ESF_DZ_RX_QLABEL);
3551 	rx_l3_class = EFX_QWORD_FIELD(*event, ESF_DZ_RX_L3_CLASS);
3552 	rx_l4_class = EFX_QWORD_FIELD(*event, ESF_FZ_RX_L4_CLASS);
3553 	rx_cont = EFX_QWORD_FIELD(*event, ESF_DZ_RX_CONT);
3554 	rx_encap_hdr =
3555 		nic_data->datapath_caps &
3556 			(1 << MC_CMD_GET_CAPABILITIES_OUT_VXLAN_NVGRE_LBN) ?
3557 		EFX_QWORD_FIELD(*event, ESF_EZ_RX_ENCAP_HDR) :
3558 		ESE_EZ_ENCAP_HDR_NONE;
3559 
3560 	if (EFX_QWORD_FIELD(*event, ESF_DZ_RX_DROP_EVENT))
3561 		netdev_WARN(efx->net_dev, "saw RX_DROP_EVENT: event="
3562 			    EFX_QWORD_FMT "\n",
3563 			    EFX_QWORD_VAL(*event));
3564 
3565 	rx_queue = efx_channel_get_rx_queue(channel);
3566 
3567 	if (unlikely(rx_queue_label != efx_rx_queue_index(rx_queue)))
3568 		efx_ef10_handle_rx_wrong_queue(rx_queue, rx_queue_label);
3569 
3570 	n_descs = ((next_ptr_lbits - rx_queue->removed_count) &
3571 		   ((1 << ESF_DZ_RX_DSC_PTR_LBITS_WIDTH) - 1));
3572 
3573 	if (n_descs != rx_queue->scatter_n + 1) {
3574 		struct efx_ef10_nic_data *nic_data = efx->nic_data;
3575 
3576 		/* detect rx abort */
3577 		if (unlikely(n_descs == rx_queue->scatter_n)) {
3578 			if (rx_queue->scatter_n == 0 || rx_bytes != 0)
3579 				netdev_WARN(efx->net_dev,
3580 					    "invalid RX abort: scatter_n=%u event="
3581 					    EFX_QWORD_FMT "\n",
3582 					    rx_queue->scatter_n,
3583 					    EFX_QWORD_VAL(*event));
3584 			efx_ef10_handle_rx_abort(rx_queue);
3585 			return 0;
3586 		}
3587 
3588 		/* Check that RX completion merging is valid, i.e.
3589 		 * the current firmware supports it and this is a
3590 		 * non-scattered packet.
3591 		 */
3592 		if (!(nic_data->datapath_caps &
3593 		      (1 << MC_CMD_GET_CAPABILITIES_OUT_RX_BATCHING_LBN)) ||
3594 		    rx_queue->scatter_n != 0 || rx_cont) {
3595 			efx_ef10_handle_rx_bad_lbits(
3596 				rx_queue, next_ptr_lbits,
3597 				(rx_queue->removed_count +
3598 				 rx_queue->scatter_n + 1) &
3599 				((1 << ESF_DZ_RX_DSC_PTR_LBITS_WIDTH) - 1));
3600 			return 0;
3601 		}
3602 
3603 		/* Merged completion for multiple non-scattered packets */
3604 		rx_queue->scatter_n = 1;
3605 		rx_queue->scatter_len = 0;
3606 		n_packets = n_descs;
3607 		++channel->n_rx_merge_events;
3608 		channel->n_rx_merge_packets += n_packets;
3609 		flags |= EFX_RX_PKT_PREFIX_LEN;
3610 	} else {
3611 		++rx_queue->scatter_n;
3612 		rx_queue->scatter_len += rx_bytes;
3613 		if (rx_cont)
3614 			return 0;
3615 		n_packets = 1;
3616 	}
3617 
3618 	EFX_POPULATE_QWORD_5(errors, ESF_DZ_RX_ECRC_ERR, 1,
3619 				     ESF_DZ_RX_IPCKSUM_ERR, 1,
3620 				     ESF_DZ_RX_TCPUDP_CKSUM_ERR, 1,
3621 				     ESF_EZ_RX_IP_INNER_CHKSUM_ERR, 1,
3622 				     ESF_EZ_RX_TCP_UDP_INNER_CHKSUM_ERR, 1);
3623 	EFX_AND_QWORD(errors, *event, errors);
3624 	if (unlikely(!EFX_QWORD_IS_ZERO(errors))) {
3625 		flags |= efx_ef10_handle_rx_event_errors(channel, n_packets,
3626 							 rx_encap_hdr,
3627 							 rx_l3_class, rx_l4_class,
3628 							 event);
3629 	} else {
3630 		bool tcpudp = rx_l4_class == ESE_FZ_L4_CLASS_TCP ||
3631 			      rx_l4_class == ESE_FZ_L4_CLASS_UDP;
3632 
3633 		switch (rx_encap_hdr) {
3634 		case ESE_EZ_ENCAP_HDR_VXLAN: /* VxLAN or GENEVE */
3635 			flags |= EFX_RX_PKT_CSUMMED; /* outer UDP csum */
3636 			if (tcpudp)
3637 				flags |= EFX_RX_PKT_CSUM_LEVEL; /* inner L4 */
3638 			break;
3639 		case ESE_EZ_ENCAP_HDR_GRE:
3640 		case ESE_EZ_ENCAP_HDR_NONE:
3641 			if (tcpudp)
3642 				flags |= EFX_RX_PKT_CSUMMED;
3643 			break;
3644 		default:
3645 			netdev_WARN(efx->net_dev,
3646 				    "unknown encapsulation type: event="
3647 				    EFX_QWORD_FMT "\n",
3648 				    EFX_QWORD_VAL(*event));
3649 		}
3650 	}
3651 
3652 	if (rx_l4_class == ESE_FZ_L4_CLASS_TCP)
3653 		flags |= EFX_RX_PKT_TCP;
3654 
3655 	channel->irq_mod_score += 2 * n_packets;
3656 
3657 	/* Handle received packet(s) */
3658 	for (i = 0; i < n_packets; i++) {
3659 		efx_rx_packet(rx_queue,
3660 			      rx_queue->removed_count & rx_queue->ptr_mask,
3661 			      rx_queue->scatter_n, rx_queue->scatter_len,
3662 			      flags);
3663 		rx_queue->removed_count += rx_queue->scatter_n;
3664 	}
3665 
3666 	rx_queue->scatter_n = 0;
3667 	rx_queue->scatter_len = 0;
3668 
3669 	return n_packets;
3670 }
3671 
3672 static u32 efx_ef10_extract_event_ts(efx_qword_t *event)
3673 {
3674 	u32 tstamp;
3675 
3676 	tstamp = EFX_QWORD_FIELD(*event, TX_TIMESTAMP_EVENT_TSTAMP_DATA_HI);
3677 	tstamp <<= 16;
3678 	tstamp |= EFX_QWORD_FIELD(*event, TX_TIMESTAMP_EVENT_TSTAMP_DATA_LO);
3679 
3680 	return tstamp;
3681 }
3682 
3683 static void
3684 efx_ef10_handle_tx_event(struct efx_channel *channel, efx_qword_t *event)
3685 {
3686 	struct efx_nic *efx = channel->efx;
3687 	struct efx_tx_queue *tx_queue;
3688 	unsigned int tx_ev_desc_ptr;
3689 	unsigned int tx_ev_q_label;
3690 	unsigned int tx_ev_type;
3691 	u64 ts_part;
3692 
3693 	if (unlikely(READ_ONCE(efx->reset_pending)))
3694 		return;
3695 
3696 	if (unlikely(EFX_QWORD_FIELD(*event, ESF_DZ_TX_DROP_EVENT)))
3697 		return;
3698 
3699 	/* Get the transmit queue */
3700 	tx_ev_q_label = EFX_QWORD_FIELD(*event, ESF_DZ_TX_QLABEL);
3701 	tx_queue = efx_channel_get_tx_queue(channel,
3702 					    tx_ev_q_label % EFX_TXQ_TYPES);
3703 
3704 	if (!tx_queue->timestamping) {
3705 		/* Transmit completion */
3706 		tx_ev_desc_ptr = EFX_QWORD_FIELD(*event, ESF_DZ_TX_DESCR_INDX);
3707 		efx_xmit_done(tx_queue, tx_ev_desc_ptr & tx_queue->ptr_mask);
3708 		return;
3709 	}
3710 
3711 	/* Transmit timestamps are only available for 8XXX series. They result
3712 	 * in three events per packet. These occur in order, and are:
3713 	 *  - the normal completion event
3714 	 *  - the low part of the timestamp
3715 	 *  - the high part of the timestamp
3716 	 *
3717 	 * Each part of the timestamp is itself split across two 16 bit
3718 	 * fields in the event.
3719 	 */
3720 	tx_ev_type = EFX_QWORD_FIELD(*event, ESF_EZ_TX_SOFT1);
3721 
3722 	switch (tx_ev_type) {
3723 	case TX_TIMESTAMP_EVENT_TX_EV_COMPLETION:
3724 		/* In case of Queue flush or FLR, we might have received
3725 		 * the previous TX completion event but not the Timestamp
3726 		 * events.
3727 		 */
3728 		if (tx_queue->completed_desc_ptr != tx_queue->ptr_mask)
3729 			efx_xmit_done(tx_queue, tx_queue->completed_desc_ptr);
3730 
3731 		tx_ev_desc_ptr = EFX_QWORD_FIELD(*event,
3732 						 ESF_DZ_TX_DESCR_INDX);
3733 		tx_queue->completed_desc_ptr =
3734 					tx_ev_desc_ptr & tx_queue->ptr_mask;
3735 		break;
3736 
3737 	case TX_TIMESTAMP_EVENT_TX_EV_TSTAMP_LO:
3738 		ts_part = efx_ef10_extract_event_ts(event);
3739 		tx_queue->completed_timestamp_minor = ts_part;
3740 		break;
3741 
3742 	case TX_TIMESTAMP_EVENT_TX_EV_TSTAMP_HI:
3743 		ts_part = efx_ef10_extract_event_ts(event);
3744 		tx_queue->completed_timestamp_major = ts_part;
3745 
3746 		efx_xmit_done(tx_queue, tx_queue->completed_desc_ptr);
3747 		tx_queue->completed_desc_ptr = tx_queue->ptr_mask;
3748 		break;
3749 
3750 	default:
3751 		netif_err(efx, hw, efx->net_dev,
3752 			  "channel %d unknown tx event type %d (data "
3753 			  EFX_QWORD_FMT ")\n",
3754 			  channel->channel, tx_ev_type,
3755 			  EFX_QWORD_VAL(*event));
3756 		break;
3757 	}
3758 }
3759 
3760 static void
3761 efx_ef10_handle_driver_event(struct efx_channel *channel, efx_qword_t *event)
3762 {
3763 	struct efx_nic *efx = channel->efx;
3764 	int subcode;
3765 
3766 	subcode = EFX_QWORD_FIELD(*event, ESF_DZ_DRV_SUB_CODE);
3767 
3768 	switch (subcode) {
3769 	case ESE_DZ_DRV_TIMER_EV:
3770 	case ESE_DZ_DRV_WAKE_UP_EV:
3771 		break;
3772 	case ESE_DZ_DRV_START_UP_EV:
3773 		/* event queue init complete. ok. */
3774 		break;
3775 	default:
3776 		netif_err(efx, hw, efx->net_dev,
3777 			  "channel %d unknown driver event type %d"
3778 			  " (data " EFX_QWORD_FMT ")\n",
3779 			  channel->channel, subcode,
3780 			  EFX_QWORD_VAL(*event));
3781 
3782 	}
3783 }
3784 
3785 static void efx_ef10_handle_driver_generated_event(struct efx_channel *channel,
3786 						   efx_qword_t *event)
3787 {
3788 	struct efx_nic *efx = channel->efx;
3789 	u32 subcode;
3790 
3791 	subcode = EFX_QWORD_FIELD(*event, EFX_DWORD_0);
3792 
3793 	switch (subcode) {
3794 	case EFX_EF10_TEST:
3795 		channel->event_test_cpu = raw_smp_processor_id();
3796 		break;
3797 	case EFX_EF10_REFILL:
3798 		/* The queue must be empty, so we won't receive any rx
3799 		 * events, so efx_process_channel() won't refill the
3800 		 * queue. Refill it here
3801 		 */
3802 		efx_fast_push_rx_descriptors(&channel->rx_queue, true);
3803 		break;
3804 	default:
3805 		netif_err(efx, hw, efx->net_dev,
3806 			  "channel %d unknown driver event type %u"
3807 			  " (data " EFX_QWORD_FMT ")\n",
3808 			  channel->channel, (unsigned) subcode,
3809 			  EFX_QWORD_VAL(*event));
3810 	}
3811 }
3812 
3813 static int efx_ef10_ev_process(struct efx_channel *channel, int quota)
3814 {
3815 	struct efx_nic *efx = channel->efx;
3816 	efx_qword_t event, *p_event;
3817 	unsigned int read_ptr;
3818 	int ev_code;
3819 	int spent = 0;
3820 
3821 	if (quota <= 0)
3822 		return spent;
3823 
3824 	read_ptr = channel->eventq_read_ptr;
3825 
3826 	for (;;) {
3827 		p_event = efx_event(channel, read_ptr);
3828 		event = *p_event;
3829 
3830 		if (!efx_event_present(&event))
3831 			break;
3832 
3833 		EFX_SET_QWORD(*p_event);
3834 
3835 		++read_ptr;
3836 
3837 		ev_code = EFX_QWORD_FIELD(event, ESF_DZ_EV_CODE);
3838 
3839 		netif_vdbg(efx, drv, efx->net_dev,
3840 			   "processing event on %d " EFX_QWORD_FMT "\n",
3841 			   channel->channel, EFX_QWORD_VAL(event));
3842 
3843 		switch (ev_code) {
3844 		case ESE_DZ_EV_CODE_MCDI_EV:
3845 			efx_mcdi_process_event(channel, &event);
3846 			break;
3847 		case ESE_DZ_EV_CODE_RX_EV:
3848 			spent += efx_ef10_handle_rx_event(channel, &event);
3849 			if (spent >= quota) {
3850 				/* XXX can we split a merged event to
3851 				 * avoid going over-quota?
3852 				 */
3853 				spent = quota;
3854 				goto out;
3855 			}
3856 			break;
3857 		case ESE_DZ_EV_CODE_TX_EV:
3858 			efx_ef10_handle_tx_event(channel, &event);
3859 			break;
3860 		case ESE_DZ_EV_CODE_DRIVER_EV:
3861 			efx_ef10_handle_driver_event(channel, &event);
3862 			if (++spent == quota)
3863 				goto out;
3864 			break;
3865 		case EFX_EF10_DRVGEN_EV:
3866 			efx_ef10_handle_driver_generated_event(channel, &event);
3867 			break;
3868 		default:
3869 			netif_err(efx, hw, efx->net_dev,
3870 				  "channel %d unknown event type %d"
3871 				  " (data " EFX_QWORD_FMT ")\n",
3872 				  channel->channel, ev_code,
3873 				  EFX_QWORD_VAL(event));
3874 		}
3875 	}
3876 
3877 out:
3878 	channel->eventq_read_ptr = read_ptr;
3879 	return spent;
3880 }
3881 
3882 static void efx_ef10_ev_read_ack(struct efx_channel *channel)
3883 {
3884 	struct efx_nic *efx = channel->efx;
3885 	efx_dword_t rptr;
3886 
3887 	if (EFX_EF10_WORKAROUND_35388(efx)) {
3888 		BUILD_BUG_ON(EFX_MIN_EVQ_SIZE <
3889 			     (1 << ERF_DD_EVQ_IND_RPTR_WIDTH));
3890 		BUILD_BUG_ON(EFX_MAX_EVQ_SIZE >
3891 			     (1 << 2 * ERF_DD_EVQ_IND_RPTR_WIDTH));
3892 
3893 		EFX_POPULATE_DWORD_2(rptr, ERF_DD_EVQ_IND_RPTR_FLAGS,
3894 				     EFE_DD_EVQ_IND_RPTR_FLAGS_HIGH,
3895 				     ERF_DD_EVQ_IND_RPTR,
3896 				     (channel->eventq_read_ptr &
3897 				      channel->eventq_mask) >>
3898 				     ERF_DD_EVQ_IND_RPTR_WIDTH);
3899 		efx_writed_page(efx, &rptr, ER_DD_EVQ_INDIRECT,
3900 				channel->channel);
3901 		EFX_POPULATE_DWORD_2(rptr, ERF_DD_EVQ_IND_RPTR_FLAGS,
3902 				     EFE_DD_EVQ_IND_RPTR_FLAGS_LOW,
3903 				     ERF_DD_EVQ_IND_RPTR,
3904 				     channel->eventq_read_ptr &
3905 				     ((1 << ERF_DD_EVQ_IND_RPTR_WIDTH) - 1));
3906 		efx_writed_page(efx, &rptr, ER_DD_EVQ_INDIRECT,
3907 				channel->channel);
3908 	} else {
3909 		EFX_POPULATE_DWORD_1(rptr, ERF_DZ_EVQ_RPTR,
3910 				     channel->eventq_read_ptr &
3911 				     channel->eventq_mask);
3912 		efx_writed_page(efx, &rptr, ER_DZ_EVQ_RPTR, channel->channel);
3913 	}
3914 }
3915 
3916 static void efx_ef10_ev_test_generate(struct efx_channel *channel)
3917 {
3918 	MCDI_DECLARE_BUF(inbuf, MC_CMD_DRIVER_EVENT_IN_LEN);
3919 	struct efx_nic *efx = channel->efx;
3920 	efx_qword_t event;
3921 	int rc;
3922 
3923 	EFX_POPULATE_QWORD_2(event,
3924 			     ESF_DZ_EV_CODE, EFX_EF10_DRVGEN_EV,
3925 			     ESF_DZ_EV_DATA, EFX_EF10_TEST);
3926 
3927 	MCDI_SET_DWORD(inbuf, DRIVER_EVENT_IN_EVQ, channel->channel);
3928 
3929 	/* MCDI_SET_QWORD is not appropriate here since EFX_POPULATE_* has
3930 	 * already swapped the data to little-endian order.
3931 	 */
3932 	memcpy(MCDI_PTR(inbuf, DRIVER_EVENT_IN_DATA), &event.u64[0],
3933 	       sizeof(efx_qword_t));
3934 
3935 	rc = efx_mcdi_rpc(efx, MC_CMD_DRIVER_EVENT, inbuf, sizeof(inbuf),
3936 			  NULL, 0, NULL);
3937 	if (rc != 0)
3938 		goto fail;
3939 
3940 	return;
3941 
3942 fail:
3943 	WARN_ON(true);
3944 	netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
3945 }
3946 
3947 void efx_ef10_handle_drain_event(struct efx_nic *efx)
3948 {
3949 	if (atomic_dec_and_test(&efx->active_queues))
3950 		wake_up(&efx->flush_wq);
3951 
3952 	WARN_ON(atomic_read(&efx->active_queues) < 0);
3953 }
3954 
3955 static int efx_ef10_fini_dmaq(struct efx_nic *efx)
3956 {
3957 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
3958 	struct efx_channel *channel;
3959 	struct efx_tx_queue *tx_queue;
3960 	struct efx_rx_queue *rx_queue;
3961 	int pending;
3962 
3963 	/* If the MC has just rebooted, the TX/RX queues will have already been
3964 	 * torn down, but efx->active_queues needs to be set to zero.
3965 	 */
3966 	if (nic_data->must_realloc_vis) {
3967 		atomic_set(&efx->active_queues, 0);
3968 		return 0;
3969 	}
3970 
3971 	/* Do not attempt to write to the NIC during EEH recovery */
3972 	if (efx->state != STATE_RECOVERY) {
3973 		efx_for_each_channel(channel, efx) {
3974 			efx_for_each_channel_rx_queue(rx_queue, channel)
3975 				efx_ef10_rx_fini(rx_queue);
3976 			efx_for_each_channel_tx_queue(tx_queue, channel)
3977 				efx_ef10_tx_fini(tx_queue);
3978 		}
3979 
3980 		wait_event_timeout(efx->flush_wq,
3981 				   atomic_read(&efx->active_queues) == 0,
3982 				   msecs_to_jiffies(EFX_MAX_FLUSH_TIME));
3983 		pending = atomic_read(&efx->active_queues);
3984 		if (pending) {
3985 			netif_err(efx, hw, efx->net_dev, "failed to flush %d queues\n",
3986 				  pending);
3987 			return -ETIMEDOUT;
3988 		}
3989 	}
3990 
3991 	return 0;
3992 }
3993 
3994 static void efx_ef10_prepare_flr(struct efx_nic *efx)
3995 {
3996 	atomic_set(&efx->active_queues, 0);
3997 }
3998 
3999 /* Decide whether a filter should be exclusive or else should allow
4000  * delivery to additional recipients.  Currently we decide that
4001  * filters for specific local unicast MAC and IP addresses are
4002  * exclusive.
4003  */
4004 static bool efx_ef10_filter_is_exclusive(const struct efx_filter_spec *spec)
4005 {
4006 	if (spec->match_flags & EFX_FILTER_MATCH_LOC_MAC &&
4007 	    !is_multicast_ether_addr(spec->loc_mac))
4008 		return true;
4009 
4010 	if ((spec->match_flags &
4011 	     (EFX_FILTER_MATCH_ETHER_TYPE | EFX_FILTER_MATCH_LOC_HOST)) ==
4012 	    (EFX_FILTER_MATCH_ETHER_TYPE | EFX_FILTER_MATCH_LOC_HOST)) {
4013 		if (spec->ether_type == htons(ETH_P_IP) &&
4014 		    !ipv4_is_multicast(spec->loc_host[0]))
4015 			return true;
4016 		if (spec->ether_type == htons(ETH_P_IPV6) &&
4017 		    ((const u8 *)spec->loc_host)[0] != 0xff)
4018 			return true;
4019 	}
4020 
4021 	return false;
4022 }
4023 
4024 static struct efx_filter_spec *
4025 efx_ef10_filter_entry_spec(const struct efx_ef10_filter_table *table,
4026 			   unsigned int filter_idx)
4027 {
4028 	return (struct efx_filter_spec *)(table->entry[filter_idx].spec &
4029 					  ~EFX_EF10_FILTER_FLAGS);
4030 }
4031 
4032 static unsigned int
4033 efx_ef10_filter_entry_flags(const struct efx_ef10_filter_table *table,
4034 			   unsigned int filter_idx)
4035 {
4036 	return table->entry[filter_idx].spec & EFX_EF10_FILTER_FLAGS;
4037 }
4038 
4039 static void
4040 efx_ef10_filter_set_entry(struct efx_ef10_filter_table *table,
4041 			  unsigned int filter_idx,
4042 			  const struct efx_filter_spec *spec,
4043 			  unsigned int flags)
4044 {
4045 	table->entry[filter_idx].spec =	(unsigned long)spec | flags;
4046 }
4047 
4048 static void
4049 efx_ef10_filter_push_prep_set_match_fields(struct efx_nic *efx,
4050 					   const struct efx_filter_spec *spec,
4051 					   efx_dword_t *inbuf)
4052 {
4053 	enum efx_encap_type encap_type = efx_filter_get_encap_type(spec);
4054 	u32 match_fields = 0, uc_match, mc_match;
4055 
4056 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
4057 		       efx_ef10_filter_is_exclusive(spec) ?
4058 		       MC_CMD_FILTER_OP_IN_OP_INSERT :
4059 		       MC_CMD_FILTER_OP_IN_OP_SUBSCRIBE);
4060 
4061 	/* Convert match flags and values.  Unlike almost
4062 	 * everything else in MCDI, these fields are in
4063 	 * network byte order.
4064 	 */
4065 #define COPY_VALUE(value, mcdi_field)					     \
4066 	do {							     \
4067 		match_fields |=					     \
4068 			1 << MC_CMD_FILTER_OP_IN_MATCH_ ##	     \
4069 			mcdi_field ## _LBN;			     \
4070 		BUILD_BUG_ON(					     \
4071 			MC_CMD_FILTER_OP_IN_ ## mcdi_field ## _LEN < \
4072 			sizeof(value));				     \
4073 		memcpy(MCDI_PTR(inbuf, FILTER_OP_IN_ ##	mcdi_field), \
4074 		       &value, sizeof(value));			     \
4075 	} while (0)
4076 #define COPY_FIELD(gen_flag, gen_field, mcdi_field)			     \
4077 	if (spec->match_flags & EFX_FILTER_MATCH_ ## gen_flag) {     \
4078 		COPY_VALUE(spec->gen_field, mcdi_field);	     \
4079 	}
4080 	/* Handle encap filters first.  They will always be mismatch
4081 	 * (unknown UC or MC) filters
4082 	 */
4083 	if (encap_type) {
4084 		/* ether_type and outer_ip_proto need to be variables
4085 		 * because COPY_VALUE wants to memcpy them
4086 		 */
4087 		__be16 ether_type =
4088 			htons(encap_type & EFX_ENCAP_FLAG_IPV6 ?
4089 			      ETH_P_IPV6 : ETH_P_IP);
4090 		u8 vni_type = MC_CMD_FILTER_OP_EXT_IN_VNI_TYPE_GENEVE;
4091 		u8 outer_ip_proto;
4092 
4093 		switch (encap_type & EFX_ENCAP_TYPES_MASK) {
4094 		case EFX_ENCAP_TYPE_VXLAN:
4095 			vni_type = MC_CMD_FILTER_OP_EXT_IN_VNI_TYPE_VXLAN;
4096 			/* fallthrough */
4097 		case EFX_ENCAP_TYPE_GENEVE:
4098 			COPY_VALUE(ether_type, ETHER_TYPE);
4099 			outer_ip_proto = IPPROTO_UDP;
4100 			COPY_VALUE(outer_ip_proto, IP_PROTO);
4101 			/* We always need to set the type field, even
4102 			 * though we're not matching on the TNI.
4103 			 */
4104 			MCDI_POPULATE_DWORD_1(inbuf,
4105 				FILTER_OP_EXT_IN_VNI_OR_VSID,
4106 				FILTER_OP_EXT_IN_VNI_TYPE,
4107 				vni_type);
4108 			break;
4109 		case EFX_ENCAP_TYPE_NVGRE:
4110 			COPY_VALUE(ether_type, ETHER_TYPE);
4111 			outer_ip_proto = IPPROTO_GRE;
4112 			COPY_VALUE(outer_ip_proto, IP_PROTO);
4113 			break;
4114 		default:
4115 			WARN_ON(1);
4116 		}
4117 
4118 		uc_match = MC_CMD_FILTER_OP_EXT_IN_MATCH_IFRM_UNKNOWN_UCAST_DST_LBN;
4119 		mc_match = MC_CMD_FILTER_OP_EXT_IN_MATCH_IFRM_UNKNOWN_MCAST_DST_LBN;
4120 	} else {
4121 		uc_match = MC_CMD_FILTER_OP_EXT_IN_MATCH_UNKNOWN_UCAST_DST_LBN;
4122 		mc_match = MC_CMD_FILTER_OP_EXT_IN_MATCH_UNKNOWN_MCAST_DST_LBN;
4123 	}
4124 
4125 	if (spec->match_flags & EFX_FILTER_MATCH_LOC_MAC_IG)
4126 		match_fields |=
4127 			is_multicast_ether_addr(spec->loc_mac) ?
4128 			1 << mc_match :
4129 			1 << uc_match;
4130 	COPY_FIELD(REM_HOST, rem_host, SRC_IP);
4131 	COPY_FIELD(LOC_HOST, loc_host, DST_IP);
4132 	COPY_FIELD(REM_MAC, rem_mac, SRC_MAC);
4133 	COPY_FIELD(REM_PORT, rem_port, SRC_PORT);
4134 	COPY_FIELD(LOC_MAC, loc_mac, DST_MAC);
4135 	COPY_FIELD(LOC_PORT, loc_port, DST_PORT);
4136 	COPY_FIELD(ETHER_TYPE, ether_type, ETHER_TYPE);
4137 	COPY_FIELD(INNER_VID, inner_vid, INNER_VLAN);
4138 	COPY_FIELD(OUTER_VID, outer_vid, OUTER_VLAN);
4139 	COPY_FIELD(IP_PROTO, ip_proto, IP_PROTO);
4140 #undef COPY_FIELD
4141 #undef COPY_VALUE
4142 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_MATCH_FIELDS,
4143 		       match_fields);
4144 }
4145 
4146 static void efx_ef10_filter_push_prep(struct efx_nic *efx,
4147 				      const struct efx_filter_spec *spec,
4148 				      efx_dword_t *inbuf, u64 handle,
4149 				      struct efx_rss_context *ctx,
4150 				      bool replacing)
4151 {
4152 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
4153 	u32 flags = spec->flags;
4154 
4155 	memset(inbuf, 0, MC_CMD_FILTER_OP_EXT_IN_LEN);
4156 
4157 	/* If RSS filter, caller better have given us an RSS context */
4158 	if (flags & EFX_FILTER_FLAG_RX_RSS) {
4159 		/* We don't have the ability to return an error, so we'll just
4160 		 * log a warning and disable RSS for the filter.
4161 		 */
4162 		if (WARN_ON_ONCE(!ctx))
4163 			flags &= ~EFX_FILTER_FLAG_RX_RSS;
4164 		else if (WARN_ON_ONCE(ctx->context_id == EFX_EF10_RSS_CONTEXT_INVALID))
4165 			flags &= ~EFX_FILTER_FLAG_RX_RSS;
4166 	}
4167 
4168 	if (replacing) {
4169 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
4170 			       MC_CMD_FILTER_OP_IN_OP_REPLACE);
4171 		MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE, handle);
4172 	} else {
4173 		efx_ef10_filter_push_prep_set_match_fields(efx, spec, inbuf);
4174 	}
4175 
4176 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_PORT_ID, nic_data->vport_id);
4177 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_DEST,
4178 		       spec->dmaq_id == EFX_FILTER_RX_DMAQ_ID_DROP ?
4179 		       MC_CMD_FILTER_OP_IN_RX_DEST_DROP :
4180 		       MC_CMD_FILTER_OP_IN_RX_DEST_HOST);
4181 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_TX_DOMAIN, 0);
4182 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_TX_DEST,
4183 		       MC_CMD_FILTER_OP_IN_TX_DEST_DEFAULT);
4184 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_QUEUE,
4185 		       spec->dmaq_id == EFX_FILTER_RX_DMAQ_ID_DROP ?
4186 		       0 : spec->dmaq_id);
4187 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_MODE,
4188 		       (flags & EFX_FILTER_FLAG_RX_RSS) ?
4189 		       MC_CMD_FILTER_OP_IN_RX_MODE_RSS :
4190 		       MC_CMD_FILTER_OP_IN_RX_MODE_SIMPLE);
4191 	if (flags & EFX_FILTER_FLAG_RX_RSS)
4192 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_CONTEXT, ctx->context_id);
4193 }
4194 
4195 static int efx_ef10_filter_push(struct efx_nic *efx,
4196 				const struct efx_filter_spec *spec, u64 *handle,
4197 				struct efx_rss_context *ctx, bool replacing)
4198 {
4199 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_EXT_IN_LEN);
4200 	MCDI_DECLARE_BUF(outbuf, MC_CMD_FILTER_OP_EXT_OUT_LEN);
4201 	int rc;
4202 
4203 	efx_ef10_filter_push_prep(efx, spec, inbuf, *handle, ctx, replacing);
4204 	rc = efx_mcdi_rpc(efx, MC_CMD_FILTER_OP, inbuf, sizeof(inbuf),
4205 			  outbuf, sizeof(outbuf), NULL);
4206 	if (rc == 0)
4207 		*handle = MCDI_QWORD(outbuf, FILTER_OP_OUT_HANDLE);
4208 	if (rc == -ENOSPC)
4209 		rc = -EBUSY; /* to match efx_farch_filter_insert() */
4210 	return rc;
4211 }
4212 
4213 static u32 efx_ef10_filter_mcdi_flags_from_spec(const struct efx_filter_spec *spec)
4214 {
4215 	enum efx_encap_type encap_type = efx_filter_get_encap_type(spec);
4216 	unsigned int match_flags = spec->match_flags;
4217 	unsigned int uc_match, mc_match;
4218 	u32 mcdi_flags = 0;
4219 
4220 #define MAP_FILTER_TO_MCDI_FLAG(gen_flag, mcdi_field, encap) {		\
4221 		unsigned int  old_match_flags = match_flags;		\
4222 		match_flags &= ~EFX_FILTER_MATCH_ ## gen_flag;		\
4223 		if (match_flags != old_match_flags)			\
4224 			mcdi_flags |=					\
4225 				(1 << ((encap) ?			\
4226 				       MC_CMD_FILTER_OP_EXT_IN_MATCH_IFRM_ ## \
4227 				       mcdi_field ## _LBN :		\
4228 				       MC_CMD_FILTER_OP_EXT_IN_MATCH_ ##\
4229 				       mcdi_field ## _LBN));		\
4230 	}
4231 	/* inner or outer based on encap type */
4232 	MAP_FILTER_TO_MCDI_FLAG(REM_HOST, SRC_IP, encap_type);
4233 	MAP_FILTER_TO_MCDI_FLAG(LOC_HOST, DST_IP, encap_type);
4234 	MAP_FILTER_TO_MCDI_FLAG(REM_MAC, SRC_MAC, encap_type);
4235 	MAP_FILTER_TO_MCDI_FLAG(REM_PORT, SRC_PORT, encap_type);
4236 	MAP_FILTER_TO_MCDI_FLAG(LOC_MAC, DST_MAC, encap_type);
4237 	MAP_FILTER_TO_MCDI_FLAG(LOC_PORT, DST_PORT, encap_type);
4238 	MAP_FILTER_TO_MCDI_FLAG(ETHER_TYPE, ETHER_TYPE, encap_type);
4239 	MAP_FILTER_TO_MCDI_FLAG(IP_PROTO, IP_PROTO, encap_type);
4240 	/* always outer */
4241 	MAP_FILTER_TO_MCDI_FLAG(INNER_VID, INNER_VLAN, false);
4242 	MAP_FILTER_TO_MCDI_FLAG(OUTER_VID, OUTER_VLAN, false);
4243 #undef MAP_FILTER_TO_MCDI_FLAG
4244 
4245 	/* special handling for encap type, and mismatch */
4246 	if (encap_type) {
4247 		match_flags &= ~EFX_FILTER_MATCH_ENCAP_TYPE;
4248 		mcdi_flags |=
4249 			(1 << MC_CMD_FILTER_OP_EXT_IN_MATCH_ETHER_TYPE_LBN);
4250 		mcdi_flags |= (1 << MC_CMD_FILTER_OP_EXT_IN_MATCH_IP_PROTO_LBN);
4251 
4252 		uc_match = MC_CMD_FILTER_OP_EXT_IN_MATCH_IFRM_UNKNOWN_UCAST_DST_LBN;
4253 		mc_match = MC_CMD_FILTER_OP_EXT_IN_MATCH_IFRM_UNKNOWN_MCAST_DST_LBN;
4254 	} else {
4255 		uc_match = MC_CMD_FILTER_OP_EXT_IN_MATCH_UNKNOWN_UCAST_DST_LBN;
4256 		mc_match = MC_CMD_FILTER_OP_EXT_IN_MATCH_UNKNOWN_MCAST_DST_LBN;
4257 	}
4258 
4259 	if (match_flags & EFX_FILTER_MATCH_LOC_MAC_IG) {
4260 		match_flags &= ~EFX_FILTER_MATCH_LOC_MAC_IG;
4261 		mcdi_flags |=
4262 			is_multicast_ether_addr(spec->loc_mac) ?
4263 			1 << mc_match :
4264 			1 << uc_match;
4265 	}
4266 
4267 	/* Did we map them all? */
4268 	WARN_ON_ONCE(match_flags);
4269 
4270 	return mcdi_flags;
4271 }
4272 
4273 static int efx_ef10_filter_pri(struct efx_ef10_filter_table *table,
4274 			       const struct efx_filter_spec *spec)
4275 {
4276 	u32 mcdi_flags = efx_ef10_filter_mcdi_flags_from_spec(spec);
4277 	unsigned int match_pri;
4278 
4279 	for (match_pri = 0;
4280 	     match_pri < table->rx_match_count;
4281 	     match_pri++)
4282 		if (table->rx_match_mcdi_flags[match_pri] == mcdi_flags)
4283 			return match_pri;
4284 
4285 	return -EPROTONOSUPPORT;
4286 }
4287 
4288 static s32 efx_ef10_filter_insert_locked(struct efx_nic *efx,
4289 					 struct efx_filter_spec *spec,
4290 					 bool replace_equal)
4291 {
4292 	DECLARE_BITMAP(mc_rem_map, EFX_EF10_FILTER_SEARCH_LIMIT);
4293 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
4294 	struct efx_ef10_filter_table *table;
4295 	struct efx_filter_spec *saved_spec;
4296 	struct efx_rss_context *ctx = NULL;
4297 	unsigned int match_pri, hash;
4298 	unsigned int priv_flags;
4299 	bool rss_locked = false;
4300 	bool replacing = false;
4301 	unsigned int depth, i;
4302 	int ins_index = -1;
4303 	DEFINE_WAIT(wait);
4304 	bool is_mc_recip;
4305 	s32 rc;
4306 
4307 	WARN_ON(!rwsem_is_locked(&efx->filter_sem));
4308 	table = efx->filter_state;
4309 	down_write(&table->lock);
4310 
4311 	/* For now, only support RX filters */
4312 	if ((spec->flags & (EFX_FILTER_FLAG_RX | EFX_FILTER_FLAG_TX)) !=
4313 	    EFX_FILTER_FLAG_RX) {
4314 		rc = -EINVAL;
4315 		goto out_unlock;
4316 	}
4317 
4318 	rc = efx_ef10_filter_pri(table, spec);
4319 	if (rc < 0)
4320 		goto out_unlock;
4321 	match_pri = rc;
4322 
4323 	hash = efx_filter_spec_hash(spec);
4324 	is_mc_recip = efx_filter_is_mc_recipient(spec);
4325 	if (is_mc_recip)
4326 		bitmap_zero(mc_rem_map, EFX_EF10_FILTER_SEARCH_LIMIT);
4327 
4328 	if (spec->flags & EFX_FILTER_FLAG_RX_RSS) {
4329 		mutex_lock(&efx->rss_lock);
4330 		rss_locked = true;
4331 		if (spec->rss_context)
4332 			ctx = efx_find_rss_context_entry(efx, spec->rss_context);
4333 		else
4334 			ctx = &efx->rss_context;
4335 		if (!ctx) {
4336 			rc = -ENOENT;
4337 			goto out_unlock;
4338 		}
4339 		if (ctx->context_id == EFX_EF10_RSS_CONTEXT_INVALID) {
4340 			rc = -EOPNOTSUPP;
4341 			goto out_unlock;
4342 		}
4343 	}
4344 
4345 	/* Find any existing filters with the same match tuple or
4346 	 * else a free slot to insert at.
4347 	 */
4348 	for (depth = 1; depth < EFX_EF10_FILTER_SEARCH_LIMIT; depth++) {
4349 		i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
4350 		saved_spec = efx_ef10_filter_entry_spec(table, i);
4351 
4352 		if (!saved_spec) {
4353 			if (ins_index < 0)
4354 				ins_index = i;
4355 		} else if (efx_filter_spec_equal(spec, saved_spec)) {
4356 			if (spec->priority < saved_spec->priority &&
4357 			    spec->priority != EFX_FILTER_PRI_AUTO) {
4358 				rc = -EPERM;
4359 				goto out_unlock;
4360 			}
4361 			if (!is_mc_recip) {
4362 				/* This is the only one */
4363 				if (spec->priority ==
4364 				    saved_spec->priority &&
4365 				    !replace_equal) {
4366 					rc = -EEXIST;
4367 					goto out_unlock;
4368 				}
4369 				ins_index = i;
4370 				break;
4371 			} else if (spec->priority >
4372 				   saved_spec->priority ||
4373 				   (spec->priority ==
4374 				    saved_spec->priority &&
4375 				    replace_equal)) {
4376 				if (ins_index < 0)
4377 					ins_index = i;
4378 				else
4379 					__set_bit(depth, mc_rem_map);
4380 			}
4381 		}
4382 	}
4383 
4384 	/* Once we reach the maximum search depth, use the first suitable
4385 	 * slot, or return -EBUSY if there was none
4386 	 */
4387 	if (ins_index < 0) {
4388 		rc = -EBUSY;
4389 		goto out_unlock;
4390 	}
4391 
4392 	/* Create a software table entry if necessary. */
4393 	saved_spec = efx_ef10_filter_entry_spec(table, ins_index);
4394 	if (saved_spec) {
4395 		if (spec->priority == EFX_FILTER_PRI_AUTO &&
4396 		    saved_spec->priority >= EFX_FILTER_PRI_AUTO) {
4397 			/* Just make sure it won't be removed */
4398 			if (saved_spec->priority > EFX_FILTER_PRI_AUTO)
4399 				saved_spec->flags |= EFX_FILTER_FLAG_RX_OVER_AUTO;
4400 			table->entry[ins_index].spec &=
4401 				~EFX_EF10_FILTER_FLAG_AUTO_OLD;
4402 			rc = ins_index;
4403 			goto out_unlock;
4404 		}
4405 		replacing = true;
4406 		priv_flags = efx_ef10_filter_entry_flags(table, ins_index);
4407 	} else {
4408 		saved_spec = kmalloc(sizeof(*spec), GFP_ATOMIC);
4409 		if (!saved_spec) {
4410 			rc = -ENOMEM;
4411 			goto out_unlock;
4412 		}
4413 		*saved_spec = *spec;
4414 		priv_flags = 0;
4415 	}
4416 	efx_ef10_filter_set_entry(table, ins_index, saved_spec, priv_flags);
4417 
4418 	/* Actually insert the filter on the HW */
4419 	rc = efx_ef10_filter_push(efx, spec, &table->entry[ins_index].handle,
4420 				  ctx, replacing);
4421 
4422 	if (rc == -EINVAL && nic_data->must_realloc_vis)
4423 		/* The MC rebooted under us, causing it to reject our filter
4424 		 * insertion as pointing to an invalid VI (spec->dmaq_id).
4425 		 */
4426 		rc = -EAGAIN;
4427 
4428 	/* Finalise the software table entry */
4429 	if (rc == 0) {
4430 		if (replacing) {
4431 			/* Update the fields that may differ */
4432 			if (saved_spec->priority == EFX_FILTER_PRI_AUTO)
4433 				saved_spec->flags |=
4434 					EFX_FILTER_FLAG_RX_OVER_AUTO;
4435 			saved_spec->priority = spec->priority;
4436 			saved_spec->flags &= EFX_FILTER_FLAG_RX_OVER_AUTO;
4437 			saved_spec->flags |= spec->flags;
4438 			saved_spec->rss_context = spec->rss_context;
4439 			saved_spec->dmaq_id = spec->dmaq_id;
4440 		}
4441 	} else if (!replacing) {
4442 		kfree(saved_spec);
4443 		saved_spec = NULL;
4444 	} else {
4445 		/* We failed to replace, so the old filter is still present.
4446 		 * Roll back the software table to reflect this.  In fact the
4447 		 * efx_ef10_filter_set_entry() call below will do the right
4448 		 * thing, so nothing extra is needed here.
4449 		 */
4450 	}
4451 	efx_ef10_filter_set_entry(table, ins_index, saved_spec, priv_flags);
4452 
4453 	/* Remove and finalise entries for lower-priority multicast
4454 	 * recipients
4455 	 */
4456 	if (is_mc_recip) {
4457 		MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_EXT_IN_LEN);
4458 		unsigned int depth, i;
4459 
4460 		memset(inbuf, 0, sizeof(inbuf));
4461 
4462 		for (depth = 0; depth < EFX_EF10_FILTER_SEARCH_LIMIT; depth++) {
4463 			if (!test_bit(depth, mc_rem_map))
4464 				continue;
4465 
4466 			i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
4467 			saved_spec = efx_ef10_filter_entry_spec(table, i);
4468 			priv_flags = efx_ef10_filter_entry_flags(table, i);
4469 
4470 			if (rc == 0) {
4471 				MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
4472 					       MC_CMD_FILTER_OP_IN_OP_UNSUBSCRIBE);
4473 				MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
4474 					       table->entry[i].handle);
4475 				rc = efx_mcdi_rpc(efx, MC_CMD_FILTER_OP,
4476 						  inbuf, sizeof(inbuf),
4477 						  NULL, 0, NULL);
4478 			}
4479 
4480 			if (rc == 0) {
4481 				kfree(saved_spec);
4482 				saved_spec = NULL;
4483 				priv_flags = 0;
4484 			}
4485 			efx_ef10_filter_set_entry(table, i, saved_spec,
4486 						  priv_flags);
4487 		}
4488 	}
4489 
4490 	/* If successful, return the inserted filter ID */
4491 	if (rc == 0)
4492 		rc = efx_ef10_make_filter_id(match_pri, ins_index);
4493 
4494 out_unlock:
4495 	if (rss_locked)
4496 		mutex_unlock(&efx->rss_lock);
4497 	up_write(&table->lock);
4498 	return rc;
4499 }
4500 
4501 static s32 efx_ef10_filter_insert(struct efx_nic *efx,
4502 				  struct efx_filter_spec *spec,
4503 				  bool replace_equal)
4504 {
4505 	s32 ret;
4506 
4507 	down_read(&efx->filter_sem);
4508 	ret = efx_ef10_filter_insert_locked(efx, spec, replace_equal);
4509 	up_read(&efx->filter_sem);
4510 
4511 	return ret;
4512 }
4513 
4514 static void efx_ef10_filter_update_rx_scatter(struct efx_nic *efx)
4515 {
4516 	/* no need to do anything here on EF10 */
4517 }
4518 
4519 /* Remove a filter.
4520  * If !by_index, remove by ID
4521  * If by_index, remove by index
4522  * Filter ID may come from userland and must be range-checked.
4523  * Caller must hold efx->filter_sem for read, and efx->filter_state->lock
4524  * for write.
4525  */
4526 static int efx_ef10_filter_remove_internal(struct efx_nic *efx,
4527 					   unsigned int priority_mask,
4528 					   u32 filter_id, bool by_index)
4529 {
4530 	unsigned int filter_idx = efx_ef10_filter_get_unsafe_id(filter_id);
4531 	struct efx_ef10_filter_table *table = efx->filter_state;
4532 	MCDI_DECLARE_BUF(inbuf,
4533 			 MC_CMD_FILTER_OP_IN_HANDLE_OFST +
4534 			 MC_CMD_FILTER_OP_IN_HANDLE_LEN);
4535 	struct efx_filter_spec *spec;
4536 	DEFINE_WAIT(wait);
4537 	int rc;
4538 
4539 	spec = efx_ef10_filter_entry_spec(table, filter_idx);
4540 	if (!spec ||
4541 	    (!by_index &&
4542 	     efx_ef10_filter_pri(table, spec) !=
4543 	     efx_ef10_filter_get_unsafe_pri(filter_id)))
4544 		return -ENOENT;
4545 
4546 	if (spec->flags & EFX_FILTER_FLAG_RX_OVER_AUTO &&
4547 	    priority_mask == (1U << EFX_FILTER_PRI_AUTO)) {
4548 		/* Just remove flags */
4549 		spec->flags &= ~EFX_FILTER_FLAG_RX_OVER_AUTO;
4550 		table->entry[filter_idx].spec &= ~EFX_EF10_FILTER_FLAG_AUTO_OLD;
4551 		return 0;
4552 	}
4553 
4554 	if (!(priority_mask & (1U << spec->priority)))
4555 		return -ENOENT;
4556 
4557 	if (spec->flags & EFX_FILTER_FLAG_RX_OVER_AUTO) {
4558 		/* Reset to an automatic filter */
4559 
4560 		struct efx_filter_spec new_spec = *spec;
4561 
4562 		new_spec.priority = EFX_FILTER_PRI_AUTO;
4563 		new_spec.flags = (EFX_FILTER_FLAG_RX |
4564 				  (efx_rss_active(&efx->rss_context) ?
4565 				   EFX_FILTER_FLAG_RX_RSS : 0));
4566 		new_spec.dmaq_id = 0;
4567 		new_spec.rss_context = 0;
4568 		rc = efx_ef10_filter_push(efx, &new_spec,
4569 					  &table->entry[filter_idx].handle,
4570 					  &efx->rss_context,
4571 					  true);
4572 
4573 		if (rc == 0)
4574 			*spec = new_spec;
4575 	} else {
4576 		/* Really remove the filter */
4577 
4578 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
4579 			       efx_ef10_filter_is_exclusive(spec) ?
4580 			       MC_CMD_FILTER_OP_IN_OP_REMOVE :
4581 			       MC_CMD_FILTER_OP_IN_OP_UNSUBSCRIBE);
4582 		MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
4583 			       table->entry[filter_idx].handle);
4584 		rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FILTER_OP,
4585 					inbuf, sizeof(inbuf), NULL, 0, NULL);
4586 
4587 		if ((rc == 0) || (rc == -ENOENT)) {
4588 			/* Filter removed OK or didn't actually exist */
4589 			kfree(spec);
4590 			efx_ef10_filter_set_entry(table, filter_idx, NULL, 0);
4591 		} else {
4592 			efx_mcdi_display_error(efx, MC_CMD_FILTER_OP,
4593 					       MC_CMD_FILTER_OP_EXT_IN_LEN,
4594 					       NULL, 0, rc);
4595 		}
4596 	}
4597 
4598 	return rc;
4599 }
4600 
4601 static int efx_ef10_filter_remove_safe(struct efx_nic *efx,
4602 				       enum efx_filter_priority priority,
4603 				       u32 filter_id)
4604 {
4605 	struct efx_ef10_filter_table *table;
4606 	int rc;
4607 
4608 	down_read(&efx->filter_sem);
4609 	table = efx->filter_state;
4610 	down_write(&table->lock);
4611 	rc = efx_ef10_filter_remove_internal(efx, 1U << priority, filter_id,
4612 					     false);
4613 	up_write(&table->lock);
4614 	up_read(&efx->filter_sem);
4615 	return rc;
4616 }
4617 
4618 /* Caller must hold efx->filter_sem for read */
4619 static void efx_ef10_filter_remove_unsafe(struct efx_nic *efx,
4620 					  enum efx_filter_priority priority,
4621 					  u32 filter_id)
4622 {
4623 	struct efx_ef10_filter_table *table = efx->filter_state;
4624 
4625 	if (filter_id == EFX_EF10_FILTER_ID_INVALID)
4626 		return;
4627 
4628 	down_write(&table->lock);
4629 	efx_ef10_filter_remove_internal(efx, 1U << priority, filter_id,
4630 					true);
4631 	up_write(&table->lock);
4632 }
4633 
4634 static int efx_ef10_filter_get_safe(struct efx_nic *efx,
4635 				    enum efx_filter_priority priority,
4636 				    u32 filter_id, struct efx_filter_spec *spec)
4637 {
4638 	unsigned int filter_idx = efx_ef10_filter_get_unsafe_id(filter_id);
4639 	const struct efx_filter_spec *saved_spec;
4640 	struct efx_ef10_filter_table *table;
4641 	int rc;
4642 
4643 	down_read(&efx->filter_sem);
4644 	table = efx->filter_state;
4645 	down_read(&table->lock);
4646 	saved_spec = efx_ef10_filter_entry_spec(table, filter_idx);
4647 	if (saved_spec && saved_spec->priority == priority &&
4648 	    efx_ef10_filter_pri(table, saved_spec) ==
4649 	    efx_ef10_filter_get_unsafe_pri(filter_id)) {
4650 		*spec = *saved_spec;
4651 		rc = 0;
4652 	} else {
4653 		rc = -ENOENT;
4654 	}
4655 	up_read(&table->lock);
4656 	up_read(&efx->filter_sem);
4657 	return rc;
4658 }
4659 
4660 static int efx_ef10_filter_clear_rx(struct efx_nic *efx,
4661 				    enum efx_filter_priority priority)
4662 {
4663 	struct efx_ef10_filter_table *table;
4664 	unsigned int priority_mask;
4665 	unsigned int i;
4666 	int rc;
4667 
4668 	priority_mask = (((1U << (priority + 1)) - 1) &
4669 			 ~(1U << EFX_FILTER_PRI_AUTO));
4670 
4671 	down_read(&efx->filter_sem);
4672 	table = efx->filter_state;
4673 	down_write(&table->lock);
4674 	for (i = 0; i < HUNT_FILTER_TBL_ROWS; i++) {
4675 		rc = efx_ef10_filter_remove_internal(efx, priority_mask,
4676 						     i, true);
4677 		if (rc && rc != -ENOENT)
4678 			break;
4679 		rc = 0;
4680 	}
4681 
4682 	up_write(&table->lock);
4683 	up_read(&efx->filter_sem);
4684 	return rc;
4685 }
4686 
4687 static u32 efx_ef10_filter_count_rx_used(struct efx_nic *efx,
4688 					 enum efx_filter_priority priority)
4689 {
4690 	struct efx_ef10_filter_table *table;
4691 	unsigned int filter_idx;
4692 	s32 count = 0;
4693 
4694 	down_read(&efx->filter_sem);
4695 	table = efx->filter_state;
4696 	down_read(&table->lock);
4697 	for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
4698 		if (table->entry[filter_idx].spec &&
4699 		    efx_ef10_filter_entry_spec(table, filter_idx)->priority ==
4700 		    priority)
4701 			++count;
4702 	}
4703 	up_read(&table->lock);
4704 	up_read(&efx->filter_sem);
4705 	return count;
4706 }
4707 
4708 static u32 efx_ef10_filter_get_rx_id_limit(struct efx_nic *efx)
4709 {
4710 	struct efx_ef10_filter_table *table = efx->filter_state;
4711 
4712 	return table->rx_match_count * HUNT_FILTER_TBL_ROWS * 2;
4713 }
4714 
4715 static s32 efx_ef10_filter_get_rx_ids(struct efx_nic *efx,
4716 				      enum efx_filter_priority priority,
4717 				      u32 *buf, u32 size)
4718 {
4719 	struct efx_ef10_filter_table *table;
4720 	struct efx_filter_spec *spec;
4721 	unsigned int filter_idx;
4722 	s32 count = 0;
4723 
4724 	down_read(&efx->filter_sem);
4725 	table = efx->filter_state;
4726 	down_read(&table->lock);
4727 
4728 	for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
4729 		spec = efx_ef10_filter_entry_spec(table, filter_idx);
4730 		if (spec && spec->priority == priority) {
4731 			if (count == size) {
4732 				count = -EMSGSIZE;
4733 				break;
4734 			}
4735 			buf[count++] =
4736 				efx_ef10_make_filter_id(
4737 					efx_ef10_filter_pri(table, spec),
4738 					filter_idx);
4739 		}
4740 	}
4741 	up_read(&table->lock);
4742 	up_read(&efx->filter_sem);
4743 	return count;
4744 }
4745 
4746 #ifdef CONFIG_RFS_ACCEL
4747 
4748 static bool efx_ef10_filter_rfs_expire_one(struct efx_nic *efx, u32 flow_id,
4749 					   unsigned int filter_idx)
4750 {
4751 	struct efx_filter_spec *spec, saved_spec;
4752 	struct efx_ef10_filter_table *table;
4753 	struct efx_arfs_rule *rule = NULL;
4754 	bool ret = true, force = false;
4755 	u16 arfs_id;
4756 
4757 	down_read(&efx->filter_sem);
4758 	table = efx->filter_state;
4759 	down_write(&table->lock);
4760 	spec = efx_ef10_filter_entry_spec(table, filter_idx);
4761 
4762 	if (!spec || spec->priority != EFX_FILTER_PRI_HINT)
4763 		goto out_unlock;
4764 
4765 	spin_lock_bh(&efx->rps_hash_lock);
4766 	if (!efx->rps_hash_table) {
4767 		/* In the absence of the table, we always return 0 to ARFS. */
4768 		arfs_id = 0;
4769 	} else {
4770 		rule = efx_rps_hash_find(efx, spec);
4771 		if (!rule)
4772 			/* ARFS table doesn't know of this filter, so remove it */
4773 			goto expire;
4774 		arfs_id = rule->arfs_id;
4775 		ret = efx_rps_check_rule(rule, filter_idx, &force);
4776 		if (force)
4777 			goto expire;
4778 		if (!ret) {
4779 			spin_unlock_bh(&efx->rps_hash_lock);
4780 			goto out_unlock;
4781 		}
4782 	}
4783 	if (!rps_may_expire_flow(efx->net_dev, spec->dmaq_id, flow_id, arfs_id))
4784 		ret = false;
4785 	else if (rule)
4786 		rule->filter_id = EFX_ARFS_FILTER_ID_REMOVING;
4787 expire:
4788 	saved_spec = *spec; /* remove operation will kfree spec */
4789 	spin_unlock_bh(&efx->rps_hash_lock);
4790 	/* At this point (since we dropped the lock), another thread might queue
4791 	 * up a fresh insertion request (but the actual insertion will be held
4792 	 * up by our possession of the filter table lock).  In that case, it
4793 	 * will set rule->filter_id to EFX_ARFS_FILTER_ID_PENDING, meaning that
4794 	 * the rule is not removed by efx_rps_hash_del() below.
4795 	 */
4796 	if (ret)
4797 		ret = efx_ef10_filter_remove_internal(efx, 1U << spec->priority,
4798 						      filter_idx, true) == 0;
4799 	/* While we can't safely dereference rule (we dropped the lock), we can
4800 	 * still test it for NULL.
4801 	 */
4802 	if (ret && rule) {
4803 		/* Expiring, so remove entry from ARFS table */
4804 		spin_lock_bh(&efx->rps_hash_lock);
4805 		efx_rps_hash_del(efx, &saved_spec);
4806 		spin_unlock_bh(&efx->rps_hash_lock);
4807 	}
4808 out_unlock:
4809 	up_write(&table->lock);
4810 	up_read(&efx->filter_sem);
4811 	return ret;
4812 }
4813 
4814 #endif /* CONFIG_RFS_ACCEL */
4815 
4816 static int efx_ef10_filter_match_flags_from_mcdi(bool encap, u32 mcdi_flags)
4817 {
4818 	int match_flags = 0;
4819 
4820 #define MAP_FLAG(gen_flag, mcdi_field) do {				\
4821 		u32 old_mcdi_flags = mcdi_flags;			\
4822 		mcdi_flags &= ~(1 << MC_CMD_FILTER_OP_EXT_IN_MATCH_ ##	\
4823 				     mcdi_field ## _LBN);		\
4824 		if (mcdi_flags != old_mcdi_flags)			\
4825 			match_flags |= EFX_FILTER_MATCH_ ## gen_flag;	\
4826 	} while (0)
4827 
4828 	if (encap) {
4829 		/* encap filters must specify encap type */
4830 		match_flags |= EFX_FILTER_MATCH_ENCAP_TYPE;
4831 		/* and imply ethertype and ip proto */
4832 		mcdi_flags &=
4833 			~(1 << MC_CMD_FILTER_OP_EXT_IN_MATCH_IP_PROTO_LBN);
4834 		mcdi_flags &=
4835 			~(1 << MC_CMD_FILTER_OP_EXT_IN_MATCH_ETHER_TYPE_LBN);
4836 		/* VLAN tags refer to the outer packet */
4837 		MAP_FLAG(INNER_VID, INNER_VLAN);
4838 		MAP_FLAG(OUTER_VID, OUTER_VLAN);
4839 		/* everything else refers to the inner packet */
4840 		MAP_FLAG(LOC_MAC_IG, IFRM_UNKNOWN_UCAST_DST);
4841 		MAP_FLAG(LOC_MAC_IG, IFRM_UNKNOWN_MCAST_DST);
4842 		MAP_FLAG(REM_HOST, IFRM_SRC_IP);
4843 		MAP_FLAG(LOC_HOST, IFRM_DST_IP);
4844 		MAP_FLAG(REM_MAC, IFRM_SRC_MAC);
4845 		MAP_FLAG(REM_PORT, IFRM_SRC_PORT);
4846 		MAP_FLAG(LOC_MAC, IFRM_DST_MAC);
4847 		MAP_FLAG(LOC_PORT, IFRM_DST_PORT);
4848 		MAP_FLAG(ETHER_TYPE, IFRM_ETHER_TYPE);
4849 		MAP_FLAG(IP_PROTO, IFRM_IP_PROTO);
4850 	} else {
4851 		MAP_FLAG(LOC_MAC_IG, UNKNOWN_UCAST_DST);
4852 		MAP_FLAG(LOC_MAC_IG, UNKNOWN_MCAST_DST);
4853 		MAP_FLAG(REM_HOST, SRC_IP);
4854 		MAP_FLAG(LOC_HOST, DST_IP);
4855 		MAP_FLAG(REM_MAC, SRC_MAC);
4856 		MAP_FLAG(REM_PORT, SRC_PORT);
4857 		MAP_FLAG(LOC_MAC, DST_MAC);
4858 		MAP_FLAG(LOC_PORT, DST_PORT);
4859 		MAP_FLAG(ETHER_TYPE, ETHER_TYPE);
4860 		MAP_FLAG(INNER_VID, INNER_VLAN);
4861 		MAP_FLAG(OUTER_VID, OUTER_VLAN);
4862 		MAP_FLAG(IP_PROTO, IP_PROTO);
4863 	}
4864 #undef MAP_FLAG
4865 
4866 	/* Did we map them all? */
4867 	if (mcdi_flags)
4868 		return -EINVAL;
4869 
4870 	return match_flags;
4871 }
4872 
4873 static void efx_ef10_filter_cleanup_vlans(struct efx_nic *efx)
4874 {
4875 	struct efx_ef10_filter_table *table = efx->filter_state;
4876 	struct efx_ef10_filter_vlan *vlan, *next_vlan;
4877 
4878 	/* See comment in efx_ef10_filter_table_remove() */
4879 	if (!efx_rwsem_assert_write_locked(&efx->filter_sem))
4880 		return;
4881 
4882 	if (!table)
4883 		return;
4884 
4885 	list_for_each_entry_safe(vlan, next_vlan, &table->vlan_list, list)
4886 		efx_ef10_filter_del_vlan_internal(efx, vlan);
4887 }
4888 
4889 static bool efx_ef10_filter_match_supported(struct efx_ef10_filter_table *table,
4890 					    bool encap,
4891 					    enum efx_filter_match_flags match_flags)
4892 {
4893 	unsigned int match_pri;
4894 	int mf;
4895 
4896 	for (match_pri = 0;
4897 	     match_pri < table->rx_match_count;
4898 	     match_pri++) {
4899 		mf = efx_ef10_filter_match_flags_from_mcdi(encap,
4900 				table->rx_match_mcdi_flags[match_pri]);
4901 		if (mf == match_flags)
4902 			return true;
4903 	}
4904 
4905 	return false;
4906 }
4907 
4908 static int
4909 efx_ef10_filter_table_probe_matches(struct efx_nic *efx,
4910 				    struct efx_ef10_filter_table *table,
4911 				    bool encap)
4912 {
4913 	MCDI_DECLARE_BUF(inbuf, MC_CMD_GET_PARSER_DISP_INFO_IN_LEN);
4914 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_PARSER_DISP_INFO_OUT_LENMAX);
4915 	unsigned int pd_match_pri, pd_match_count;
4916 	size_t outlen;
4917 	int rc;
4918 
4919 	/* Find out which RX filter types are supported, and their priorities */
4920 	MCDI_SET_DWORD(inbuf, GET_PARSER_DISP_INFO_IN_OP,
4921 		       encap ?
4922 		       MC_CMD_GET_PARSER_DISP_INFO_IN_OP_GET_SUPPORTED_ENCAP_RX_MATCHES :
4923 		       MC_CMD_GET_PARSER_DISP_INFO_IN_OP_GET_SUPPORTED_RX_MATCHES);
4924 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_PARSER_DISP_INFO,
4925 			  inbuf, sizeof(inbuf), outbuf, sizeof(outbuf),
4926 			  &outlen);
4927 	if (rc)
4928 		return rc;
4929 
4930 	pd_match_count = MCDI_VAR_ARRAY_LEN(
4931 		outlen, GET_PARSER_DISP_INFO_OUT_SUPPORTED_MATCHES);
4932 
4933 	for (pd_match_pri = 0; pd_match_pri < pd_match_count; pd_match_pri++) {
4934 		u32 mcdi_flags =
4935 			MCDI_ARRAY_DWORD(
4936 				outbuf,
4937 				GET_PARSER_DISP_INFO_OUT_SUPPORTED_MATCHES,
4938 				pd_match_pri);
4939 		rc = efx_ef10_filter_match_flags_from_mcdi(encap, mcdi_flags);
4940 		if (rc < 0) {
4941 			netif_dbg(efx, probe, efx->net_dev,
4942 				  "%s: fw flags %#x pri %u not supported in driver\n",
4943 				  __func__, mcdi_flags, pd_match_pri);
4944 		} else {
4945 			netif_dbg(efx, probe, efx->net_dev,
4946 				  "%s: fw flags %#x pri %u supported as driver flags %#x pri %u\n",
4947 				  __func__, mcdi_flags, pd_match_pri,
4948 				  rc, table->rx_match_count);
4949 			table->rx_match_mcdi_flags[table->rx_match_count] = mcdi_flags;
4950 			table->rx_match_count++;
4951 		}
4952 	}
4953 
4954 	return 0;
4955 }
4956 
4957 static int efx_ef10_filter_table_probe(struct efx_nic *efx)
4958 {
4959 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
4960 	struct net_device *net_dev = efx->net_dev;
4961 	struct efx_ef10_filter_table *table;
4962 	struct efx_ef10_vlan *vlan;
4963 	int rc;
4964 
4965 	if (!efx_rwsem_assert_write_locked(&efx->filter_sem))
4966 		return -EINVAL;
4967 
4968 	if (efx->filter_state) /* already probed */
4969 		return 0;
4970 
4971 	table = kzalloc(sizeof(*table), GFP_KERNEL);
4972 	if (!table)
4973 		return -ENOMEM;
4974 
4975 	table->rx_match_count = 0;
4976 	rc = efx_ef10_filter_table_probe_matches(efx, table, false);
4977 	if (rc)
4978 		goto fail;
4979 	if (nic_data->datapath_caps &
4980 		   (1 << MC_CMD_GET_CAPABILITIES_OUT_VXLAN_NVGRE_LBN))
4981 		rc = efx_ef10_filter_table_probe_matches(efx, table, true);
4982 	if (rc)
4983 		goto fail;
4984 	if ((efx_supported_features(efx) & NETIF_F_HW_VLAN_CTAG_FILTER) &&
4985 	    !(efx_ef10_filter_match_supported(table, false,
4986 		(EFX_FILTER_MATCH_OUTER_VID | EFX_FILTER_MATCH_LOC_MAC)) &&
4987 	      efx_ef10_filter_match_supported(table, false,
4988 		(EFX_FILTER_MATCH_OUTER_VID | EFX_FILTER_MATCH_LOC_MAC_IG)))) {
4989 		netif_info(efx, probe, net_dev,
4990 			   "VLAN filters are not supported in this firmware variant\n");
4991 		net_dev->features &= ~NETIF_F_HW_VLAN_CTAG_FILTER;
4992 		efx->fixed_features &= ~NETIF_F_HW_VLAN_CTAG_FILTER;
4993 		net_dev->hw_features &= ~NETIF_F_HW_VLAN_CTAG_FILTER;
4994 	}
4995 
4996 	table->entry = vzalloc(array_size(HUNT_FILTER_TBL_ROWS,
4997 					  sizeof(*table->entry)));
4998 	if (!table->entry) {
4999 		rc = -ENOMEM;
5000 		goto fail;
5001 	}
5002 
5003 	table->mc_promisc_last = false;
5004 	table->vlan_filter =
5005 		!!(efx->net_dev->features & NETIF_F_HW_VLAN_CTAG_FILTER);
5006 	INIT_LIST_HEAD(&table->vlan_list);
5007 	init_rwsem(&table->lock);
5008 
5009 	efx->filter_state = table;
5010 
5011 	list_for_each_entry(vlan, &nic_data->vlan_list, list) {
5012 		rc = efx_ef10_filter_add_vlan(efx, vlan->vid);
5013 		if (rc)
5014 			goto fail_add_vlan;
5015 	}
5016 
5017 	return 0;
5018 
5019 fail_add_vlan:
5020 	efx_ef10_filter_cleanup_vlans(efx);
5021 	efx->filter_state = NULL;
5022 fail:
5023 	kfree(table);
5024 	return rc;
5025 }
5026 
5027 /* Caller must hold efx->filter_sem for read if race against
5028  * efx_ef10_filter_table_remove() is possible
5029  */
5030 static void efx_ef10_filter_table_restore(struct efx_nic *efx)
5031 {
5032 	struct efx_ef10_filter_table *table = efx->filter_state;
5033 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
5034 	unsigned int invalid_filters = 0, failed = 0;
5035 	struct efx_ef10_filter_vlan *vlan;
5036 	struct efx_filter_spec *spec;
5037 	struct efx_rss_context *ctx;
5038 	unsigned int filter_idx;
5039 	u32 mcdi_flags;
5040 	int match_pri;
5041 	int rc, i;
5042 
5043 	WARN_ON(!rwsem_is_locked(&efx->filter_sem));
5044 
5045 	if (!nic_data->must_restore_filters)
5046 		return;
5047 
5048 	if (!table)
5049 		return;
5050 
5051 	down_write(&table->lock);
5052 	mutex_lock(&efx->rss_lock);
5053 
5054 	for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
5055 		spec = efx_ef10_filter_entry_spec(table, filter_idx);
5056 		if (!spec)
5057 			continue;
5058 
5059 		mcdi_flags = efx_ef10_filter_mcdi_flags_from_spec(spec);
5060 		match_pri = 0;
5061 		while (match_pri < table->rx_match_count &&
5062 		       table->rx_match_mcdi_flags[match_pri] != mcdi_flags)
5063 			++match_pri;
5064 		if (match_pri >= table->rx_match_count) {
5065 			invalid_filters++;
5066 			goto not_restored;
5067 		}
5068 		if (spec->rss_context)
5069 			ctx = efx_find_rss_context_entry(efx, spec->rss_context);
5070 		else
5071 			ctx = &efx->rss_context;
5072 		if (spec->flags & EFX_FILTER_FLAG_RX_RSS) {
5073 			if (!ctx) {
5074 				netif_warn(efx, drv, efx->net_dev,
5075 					   "Warning: unable to restore a filter with nonexistent RSS context %u.\n",
5076 					   spec->rss_context);
5077 				invalid_filters++;
5078 				goto not_restored;
5079 			}
5080 			if (ctx->context_id == EFX_EF10_RSS_CONTEXT_INVALID) {
5081 				netif_warn(efx, drv, efx->net_dev,
5082 					   "Warning: unable to restore a filter with RSS context %u as it was not created.\n",
5083 					   spec->rss_context);
5084 				invalid_filters++;
5085 				goto not_restored;
5086 			}
5087 		}
5088 
5089 		rc = efx_ef10_filter_push(efx, spec,
5090 					  &table->entry[filter_idx].handle,
5091 					  ctx, false);
5092 		if (rc)
5093 			failed++;
5094 
5095 		if (rc) {
5096 not_restored:
5097 			list_for_each_entry(vlan, &table->vlan_list, list)
5098 				for (i = 0; i < EFX_EF10_NUM_DEFAULT_FILTERS; ++i)
5099 					if (vlan->default_filters[i] == filter_idx)
5100 						vlan->default_filters[i] =
5101 							EFX_EF10_FILTER_ID_INVALID;
5102 
5103 			kfree(spec);
5104 			efx_ef10_filter_set_entry(table, filter_idx, NULL, 0);
5105 		}
5106 	}
5107 
5108 	mutex_unlock(&efx->rss_lock);
5109 	up_write(&table->lock);
5110 
5111 	/* This can happen validly if the MC's capabilities have changed, so
5112 	 * is not an error.
5113 	 */
5114 	if (invalid_filters)
5115 		netif_dbg(efx, drv, efx->net_dev,
5116 			  "Did not restore %u filters that are now unsupported.\n",
5117 			  invalid_filters);
5118 
5119 	if (failed)
5120 		netif_err(efx, hw, efx->net_dev,
5121 			  "unable to restore %u filters\n", failed);
5122 	else
5123 		nic_data->must_restore_filters = false;
5124 }
5125 
5126 static void efx_ef10_filter_table_remove(struct efx_nic *efx)
5127 {
5128 	struct efx_ef10_filter_table *table = efx->filter_state;
5129 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_EXT_IN_LEN);
5130 	struct efx_filter_spec *spec;
5131 	unsigned int filter_idx;
5132 	int rc;
5133 
5134 	efx_ef10_filter_cleanup_vlans(efx);
5135 	efx->filter_state = NULL;
5136 	/* If we were called without locking, then it's not safe to free
5137 	 * the table as others might be using it.  So we just WARN, leak
5138 	 * the memory, and potentially get an inconsistent filter table
5139 	 * state.
5140 	 * This should never actually happen.
5141 	 */
5142 	if (!efx_rwsem_assert_write_locked(&efx->filter_sem))
5143 		return;
5144 
5145 	if (!table)
5146 		return;
5147 
5148 	for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
5149 		spec = efx_ef10_filter_entry_spec(table, filter_idx);
5150 		if (!spec)
5151 			continue;
5152 
5153 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
5154 			       efx_ef10_filter_is_exclusive(spec) ?
5155 			       MC_CMD_FILTER_OP_IN_OP_REMOVE :
5156 			       MC_CMD_FILTER_OP_IN_OP_UNSUBSCRIBE);
5157 		MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
5158 			       table->entry[filter_idx].handle);
5159 		rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FILTER_OP, inbuf,
5160 					sizeof(inbuf), NULL, 0, NULL);
5161 		if (rc)
5162 			netif_info(efx, drv, efx->net_dev,
5163 				   "%s: filter %04x remove failed\n",
5164 				   __func__, filter_idx);
5165 		kfree(spec);
5166 	}
5167 
5168 	vfree(table->entry);
5169 	kfree(table);
5170 }
5171 
5172 static void efx_ef10_filter_mark_one_old(struct efx_nic *efx, uint16_t *id)
5173 {
5174 	struct efx_ef10_filter_table *table = efx->filter_state;
5175 	unsigned int filter_idx;
5176 
5177 	efx_rwsem_assert_write_locked(&table->lock);
5178 
5179 	if (*id != EFX_EF10_FILTER_ID_INVALID) {
5180 		filter_idx = efx_ef10_filter_get_unsafe_id(*id);
5181 		if (!table->entry[filter_idx].spec)
5182 			netif_dbg(efx, drv, efx->net_dev,
5183 				  "marked null spec old %04x:%04x\n", *id,
5184 				  filter_idx);
5185 		table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_AUTO_OLD;
5186 		*id = EFX_EF10_FILTER_ID_INVALID;
5187 	}
5188 }
5189 
5190 /* Mark old per-VLAN filters that may need to be removed */
5191 static void _efx_ef10_filter_vlan_mark_old(struct efx_nic *efx,
5192 					   struct efx_ef10_filter_vlan *vlan)
5193 {
5194 	struct efx_ef10_filter_table *table = efx->filter_state;
5195 	unsigned int i;
5196 
5197 	for (i = 0; i < table->dev_uc_count; i++)
5198 		efx_ef10_filter_mark_one_old(efx, &vlan->uc[i]);
5199 	for (i = 0; i < table->dev_mc_count; i++)
5200 		efx_ef10_filter_mark_one_old(efx, &vlan->mc[i]);
5201 	for (i = 0; i < EFX_EF10_NUM_DEFAULT_FILTERS; i++)
5202 		efx_ef10_filter_mark_one_old(efx, &vlan->default_filters[i]);
5203 }
5204 
5205 /* Mark old filters that may need to be removed.
5206  * Caller must hold efx->filter_sem for read if race against
5207  * efx_ef10_filter_table_remove() is possible
5208  */
5209 static void efx_ef10_filter_mark_old(struct efx_nic *efx)
5210 {
5211 	struct efx_ef10_filter_table *table = efx->filter_state;
5212 	struct efx_ef10_filter_vlan *vlan;
5213 
5214 	down_write(&table->lock);
5215 	list_for_each_entry(vlan, &table->vlan_list, list)
5216 		_efx_ef10_filter_vlan_mark_old(efx, vlan);
5217 	up_write(&table->lock);
5218 }
5219 
5220 static void efx_ef10_filter_uc_addr_list(struct efx_nic *efx)
5221 {
5222 	struct efx_ef10_filter_table *table = efx->filter_state;
5223 	struct net_device *net_dev = efx->net_dev;
5224 	struct netdev_hw_addr *uc;
5225 	unsigned int i;
5226 
5227 	table->uc_promisc = !!(net_dev->flags & IFF_PROMISC);
5228 	ether_addr_copy(table->dev_uc_list[0].addr, net_dev->dev_addr);
5229 	i = 1;
5230 	netdev_for_each_uc_addr(uc, net_dev) {
5231 		if (i >= EFX_EF10_FILTER_DEV_UC_MAX) {
5232 			table->uc_promisc = true;
5233 			break;
5234 		}
5235 		ether_addr_copy(table->dev_uc_list[i].addr, uc->addr);
5236 		i++;
5237 	}
5238 
5239 	table->dev_uc_count = i;
5240 }
5241 
5242 static void efx_ef10_filter_mc_addr_list(struct efx_nic *efx)
5243 {
5244 	struct efx_ef10_filter_table *table = efx->filter_state;
5245 	struct net_device *net_dev = efx->net_dev;
5246 	struct netdev_hw_addr *mc;
5247 	unsigned int i;
5248 
5249 	table->mc_overflow = false;
5250 	table->mc_promisc = !!(net_dev->flags & (IFF_PROMISC | IFF_ALLMULTI));
5251 
5252 	i = 0;
5253 	netdev_for_each_mc_addr(mc, net_dev) {
5254 		if (i >= EFX_EF10_FILTER_DEV_MC_MAX) {
5255 			table->mc_promisc = true;
5256 			table->mc_overflow = true;
5257 			break;
5258 		}
5259 		ether_addr_copy(table->dev_mc_list[i].addr, mc->addr);
5260 		i++;
5261 	}
5262 
5263 	table->dev_mc_count = i;
5264 }
5265 
5266 static int efx_ef10_filter_insert_addr_list(struct efx_nic *efx,
5267 					    struct efx_ef10_filter_vlan *vlan,
5268 					    bool multicast, bool rollback)
5269 {
5270 	struct efx_ef10_filter_table *table = efx->filter_state;
5271 	struct efx_ef10_dev_addr *addr_list;
5272 	enum efx_filter_flags filter_flags;
5273 	struct efx_filter_spec spec;
5274 	u8 baddr[ETH_ALEN];
5275 	unsigned int i, j;
5276 	int addr_count;
5277 	u16 *ids;
5278 	int rc;
5279 
5280 	if (multicast) {
5281 		addr_list = table->dev_mc_list;
5282 		addr_count = table->dev_mc_count;
5283 		ids = vlan->mc;
5284 	} else {
5285 		addr_list = table->dev_uc_list;
5286 		addr_count = table->dev_uc_count;
5287 		ids = vlan->uc;
5288 	}
5289 
5290 	filter_flags = efx_rss_active(&efx->rss_context) ? EFX_FILTER_FLAG_RX_RSS : 0;
5291 
5292 	/* Insert/renew filters */
5293 	for (i = 0; i < addr_count; i++) {
5294 		EFX_WARN_ON_PARANOID(ids[i] != EFX_EF10_FILTER_ID_INVALID);
5295 		efx_filter_init_rx(&spec, EFX_FILTER_PRI_AUTO, filter_flags, 0);
5296 		efx_filter_set_eth_local(&spec, vlan->vid, addr_list[i].addr);
5297 		rc = efx_ef10_filter_insert_locked(efx, &spec, true);
5298 		if (rc < 0) {
5299 			if (rollback) {
5300 				netif_info(efx, drv, efx->net_dev,
5301 					   "efx_ef10_filter_insert failed rc=%d\n",
5302 					   rc);
5303 				/* Fall back to promiscuous */
5304 				for (j = 0; j < i; j++) {
5305 					efx_ef10_filter_remove_unsafe(
5306 						efx, EFX_FILTER_PRI_AUTO,
5307 						ids[j]);
5308 					ids[j] = EFX_EF10_FILTER_ID_INVALID;
5309 				}
5310 				return rc;
5311 			} else {
5312 				/* keep invalid ID, and carry on */
5313 			}
5314 		} else {
5315 			ids[i] = efx_ef10_filter_get_unsafe_id(rc);
5316 		}
5317 	}
5318 
5319 	if (multicast && rollback) {
5320 		/* Also need an Ethernet broadcast filter */
5321 		EFX_WARN_ON_PARANOID(vlan->default_filters[EFX_EF10_BCAST] !=
5322 				     EFX_EF10_FILTER_ID_INVALID);
5323 		efx_filter_init_rx(&spec, EFX_FILTER_PRI_AUTO, filter_flags, 0);
5324 		eth_broadcast_addr(baddr);
5325 		efx_filter_set_eth_local(&spec, vlan->vid, baddr);
5326 		rc = efx_ef10_filter_insert_locked(efx, &spec, true);
5327 		if (rc < 0) {
5328 			netif_warn(efx, drv, efx->net_dev,
5329 				   "Broadcast filter insert failed rc=%d\n", rc);
5330 			/* Fall back to promiscuous */
5331 			for (j = 0; j < i; j++) {
5332 				efx_ef10_filter_remove_unsafe(
5333 					efx, EFX_FILTER_PRI_AUTO,
5334 					ids[j]);
5335 				ids[j] = EFX_EF10_FILTER_ID_INVALID;
5336 			}
5337 			return rc;
5338 		} else {
5339 			vlan->default_filters[EFX_EF10_BCAST] =
5340 				efx_ef10_filter_get_unsafe_id(rc);
5341 		}
5342 	}
5343 
5344 	return 0;
5345 }
5346 
5347 static int efx_ef10_filter_insert_def(struct efx_nic *efx,
5348 				      struct efx_ef10_filter_vlan *vlan,
5349 				      enum efx_encap_type encap_type,
5350 				      bool multicast, bool rollback)
5351 {
5352 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
5353 	enum efx_filter_flags filter_flags;
5354 	struct efx_filter_spec spec;
5355 	u8 baddr[ETH_ALEN];
5356 	int rc;
5357 	u16 *id;
5358 
5359 	filter_flags = efx_rss_active(&efx->rss_context) ? EFX_FILTER_FLAG_RX_RSS : 0;
5360 
5361 	efx_filter_init_rx(&spec, EFX_FILTER_PRI_AUTO, filter_flags, 0);
5362 
5363 	if (multicast)
5364 		efx_filter_set_mc_def(&spec);
5365 	else
5366 		efx_filter_set_uc_def(&spec);
5367 
5368 	if (encap_type) {
5369 		if (nic_data->datapath_caps &
5370 		    (1 << MC_CMD_GET_CAPABILITIES_OUT_VXLAN_NVGRE_LBN))
5371 			efx_filter_set_encap_type(&spec, encap_type);
5372 		else
5373 			/* don't insert encap filters on non-supporting
5374 			 * platforms. ID will be left as INVALID.
5375 			 */
5376 			return 0;
5377 	}
5378 
5379 	if (vlan->vid != EFX_FILTER_VID_UNSPEC)
5380 		efx_filter_set_eth_local(&spec, vlan->vid, NULL);
5381 
5382 	rc = efx_ef10_filter_insert_locked(efx, &spec, true);
5383 	if (rc < 0) {
5384 		const char *um = multicast ? "Multicast" : "Unicast";
5385 		const char *encap_name = "";
5386 		const char *encap_ipv = "";
5387 
5388 		if ((encap_type & EFX_ENCAP_TYPES_MASK) ==
5389 		    EFX_ENCAP_TYPE_VXLAN)
5390 			encap_name = "VXLAN ";
5391 		else if ((encap_type & EFX_ENCAP_TYPES_MASK) ==
5392 			 EFX_ENCAP_TYPE_NVGRE)
5393 			encap_name = "NVGRE ";
5394 		else if ((encap_type & EFX_ENCAP_TYPES_MASK) ==
5395 			 EFX_ENCAP_TYPE_GENEVE)
5396 			encap_name = "GENEVE ";
5397 		if (encap_type & EFX_ENCAP_FLAG_IPV6)
5398 			encap_ipv = "IPv6 ";
5399 		else if (encap_type)
5400 			encap_ipv = "IPv4 ";
5401 
5402 		/* unprivileged functions can't insert mismatch filters
5403 		 * for encapsulated or unicast traffic, so downgrade
5404 		 * those warnings to debug.
5405 		 */
5406 		netif_cond_dbg(efx, drv, efx->net_dev,
5407 			       rc == -EPERM && (encap_type || !multicast), warn,
5408 			       "%s%s%s mismatch filter insert failed rc=%d\n",
5409 			       encap_name, encap_ipv, um, rc);
5410 	} else if (multicast) {
5411 		/* mapping from encap types to default filter IDs (multicast) */
5412 		static enum efx_ef10_default_filters map[] = {
5413 			[EFX_ENCAP_TYPE_NONE] = EFX_EF10_MCDEF,
5414 			[EFX_ENCAP_TYPE_VXLAN] = EFX_EF10_VXLAN4_MCDEF,
5415 			[EFX_ENCAP_TYPE_NVGRE] = EFX_EF10_NVGRE4_MCDEF,
5416 			[EFX_ENCAP_TYPE_GENEVE] = EFX_EF10_GENEVE4_MCDEF,
5417 			[EFX_ENCAP_TYPE_VXLAN | EFX_ENCAP_FLAG_IPV6] =
5418 				EFX_EF10_VXLAN6_MCDEF,
5419 			[EFX_ENCAP_TYPE_NVGRE | EFX_ENCAP_FLAG_IPV6] =
5420 				EFX_EF10_NVGRE6_MCDEF,
5421 			[EFX_ENCAP_TYPE_GENEVE | EFX_ENCAP_FLAG_IPV6] =
5422 				EFX_EF10_GENEVE6_MCDEF,
5423 		};
5424 
5425 		/* quick bounds check (BCAST result impossible) */
5426 		BUILD_BUG_ON(EFX_EF10_BCAST != 0);
5427 		if (encap_type >= ARRAY_SIZE(map) || map[encap_type] == 0) {
5428 			WARN_ON(1);
5429 			return -EINVAL;
5430 		}
5431 		/* then follow map */
5432 		id = &vlan->default_filters[map[encap_type]];
5433 
5434 		EFX_WARN_ON_PARANOID(*id != EFX_EF10_FILTER_ID_INVALID);
5435 		*id = efx_ef10_filter_get_unsafe_id(rc);
5436 		if (!nic_data->workaround_26807 && !encap_type) {
5437 			/* Also need an Ethernet broadcast filter */
5438 			efx_filter_init_rx(&spec, EFX_FILTER_PRI_AUTO,
5439 					   filter_flags, 0);
5440 			eth_broadcast_addr(baddr);
5441 			efx_filter_set_eth_local(&spec, vlan->vid, baddr);
5442 			rc = efx_ef10_filter_insert_locked(efx, &spec, true);
5443 			if (rc < 0) {
5444 				netif_warn(efx, drv, efx->net_dev,
5445 					   "Broadcast filter insert failed rc=%d\n",
5446 					   rc);
5447 				if (rollback) {
5448 					/* Roll back the mc_def filter */
5449 					efx_ef10_filter_remove_unsafe(
5450 							efx, EFX_FILTER_PRI_AUTO,
5451 							*id);
5452 					*id = EFX_EF10_FILTER_ID_INVALID;
5453 					return rc;
5454 				}
5455 			} else {
5456 				EFX_WARN_ON_PARANOID(
5457 					vlan->default_filters[EFX_EF10_BCAST] !=
5458 					EFX_EF10_FILTER_ID_INVALID);
5459 				vlan->default_filters[EFX_EF10_BCAST] =
5460 					efx_ef10_filter_get_unsafe_id(rc);
5461 			}
5462 		}
5463 		rc = 0;
5464 	} else {
5465 		/* mapping from encap types to default filter IDs (unicast) */
5466 		static enum efx_ef10_default_filters map[] = {
5467 			[EFX_ENCAP_TYPE_NONE] = EFX_EF10_UCDEF,
5468 			[EFX_ENCAP_TYPE_VXLAN] = EFX_EF10_VXLAN4_UCDEF,
5469 			[EFX_ENCAP_TYPE_NVGRE] = EFX_EF10_NVGRE4_UCDEF,
5470 			[EFX_ENCAP_TYPE_GENEVE] = EFX_EF10_GENEVE4_UCDEF,
5471 			[EFX_ENCAP_TYPE_VXLAN | EFX_ENCAP_FLAG_IPV6] =
5472 				EFX_EF10_VXLAN6_UCDEF,
5473 			[EFX_ENCAP_TYPE_NVGRE | EFX_ENCAP_FLAG_IPV6] =
5474 				EFX_EF10_NVGRE6_UCDEF,
5475 			[EFX_ENCAP_TYPE_GENEVE | EFX_ENCAP_FLAG_IPV6] =
5476 				EFX_EF10_GENEVE6_UCDEF,
5477 		};
5478 
5479 		/* quick bounds check (BCAST result impossible) */
5480 		BUILD_BUG_ON(EFX_EF10_BCAST != 0);
5481 		if (encap_type >= ARRAY_SIZE(map) || map[encap_type] == 0) {
5482 			WARN_ON(1);
5483 			return -EINVAL;
5484 		}
5485 		/* then follow map */
5486 		id = &vlan->default_filters[map[encap_type]];
5487 		EFX_WARN_ON_PARANOID(*id != EFX_EF10_FILTER_ID_INVALID);
5488 		*id = rc;
5489 		rc = 0;
5490 	}
5491 	return rc;
5492 }
5493 
5494 /* Remove filters that weren't renewed. */
5495 static void efx_ef10_filter_remove_old(struct efx_nic *efx)
5496 {
5497 	struct efx_ef10_filter_table *table = efx->filter_state;
5498 	int remove_failed = 0;
5499 	int remove_noent = 0;
5500 	int rc;
5501 	int i;
5502 
5503 	down_write(&table->lock);
5504 	for (i = 0; i < HUNT_FILTER_TBL_ROWS; i++) {
5505 		if (READ_ONCE(table->entry[i].spec) &
5506 		    EFX_EF10_FILTER_FLAG_AUTO_OLD) {
5507 			rc = efx_ef10_filter_remove_internal(efx,
5508 					1U << EFX_FILTER_PRI_AUTO, i, true);
5509 			if (rc == -ENOENT)
5510 				remove_noent++;
5511 			else if (rc)
5512 				remove_failed++;
5513 		}
5514 	}
5515 	up_write(&table->lock);
5516 
5517 	if (remove_failed)
5518 		netif_info(efx, drv, efx->net_dev,
5519 			   "%s: failed to remove %d filters\n",
5520 			   __func__, remove_failed);
5521 	if (remove_noent)
5522 		netif_info(efx, drv, efx->net_dev,
5523 			   "%s: failed to remove %d non-existent filters\n",
5524 			   __func__, remove_noent);
5525 }
5526 
5527 static int efx_ef10_vport_set_mac_address(struct efx_nic *efx)
5528 {
5529 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
5530 	u8 mac_old[ETH_ALEN];
5531 	int rc, rc2;
5532 
5533 	/* Only reconfigure a PF-created vport */
5534 	if (is_zero_ether_addr(nic_data->vport_mac))
5535 		return 0;
5536 
5537 	efx_device_detach_sync(efx);
5538 	efx_net_stop(efx->net_dev);
5539 	down_write(&efx->filter_sem);
5540 	efx_ef10_filter_table_remove(efx);
5541 	up_write(&efx->filter_sem);
5542 
5543 	rc = efx_ef10_vadaptor_free(efx, nic_data->vport_id);
5544 	if (rc)
5545 		goto restore_filters;
5546 
5547 	ether_addr_copy(mac_old, nic_data->vport_mac);
5548 	rc = efx_ef10_vport_del_mac(efx, nic_data->vport_id,
5549 				    nic_data->vport_mac);
5550 	if (rc)
5551 		goto restore_vadaptor;
5552 
5553 	rc = efx_ef10_vport_add_mac(efx, nic_data->vport_id,
5554 				    efx->net_dev->dev_addr);
5555 	if (!rc) {
5556 		ether_addr_copy(nic_data->vport_mac, efx->net_dev->dev_addr);
5557 	} else {
5558 		rc2 = efx_ef10_vport_add_mac(efx, nic_data->vport_id, mac_old);
5559 		if (rc2) {
5560 			/* Failed to add original MAC, so clear vport_mac */
5561 			eth_zero_addr(nic_data->vport_mac);
5562 			goto reset_nic;
5563 		}
5564 	}
5565 
5566 restore_vadaptor:
5567 	rc2 = efx_ef10_vadaptor_alloc(efx, nic_data->vport_id);
5568 	if (rc2)
5569 		goto reset_nic;
5570 restore_filters:
5571 	down_write(&efx->filter_sem);
5572 	rc2 = efx_ef10_filter_table_probe(efx);
5573 	up_write(&efx->filter_sem);
5574 	if (rc2)
5575 		goto reset_nic;
5576 
5577 	rc2 = efx_net_open(efx->net_dev);
5578 	if (rc2)
5579 		goto reset_nic;
5580 
5581 	efx_device_attach_if_not_resetting(efx);
5582 
5583 	return rc;
5584 
5585 reset_nic:
5586 	netif_err(efx, drv, efx->net_dev,
5587 		  "Failed to restore when changing MAC address - scheduling reset\n");
5588 	efx_schedule_reset(efx, RESET_TYPE_DATAPATH);
5589 
5590 	return rc ? rc : rc2;
5591 }
5592 
5593 /* Caller must hold efx->filter_sem for read if race against
5594  * efx_ef10_filter_table_remove() is possible
5595  */
5596 static void efx_ef10_filter_vlan_sync_rx_mode(struct efx_nic *efx,
5597 					      struct efx_ef10_filter_vlan *vlan)
5598 {
5599 	struct efx_ef10_filter_table *table = efx->filter_state;
5600 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
5601 
5602 	/* Do not install unspecified VID if VLAN filtering is enabled.
5603 	 * Do not install all specified VIDs if VLAN filtering is disabled.
5604 	 */
5605 	if ((vlan->vid == EFX_FILTER_VID_UNSPEC) == table->vlan_filter)
5606 		return;
5607 
5608 	/* Insert/renew unicast filters */
5609 	if (table->uc_promisc) {
5610 		efx_ef10_filter_insert_def(efx, vlan, EFX_ENCAP_TYPE_NONE,
5611 					   false, false);
5612 		efx_ef10_filter_insert_addr_list(efx, vlan, false, false);
5613 	} else {
5614 		/* If any of the filters failed to insert, fall back to
5615 		 * promiscuous mode - add in the uc_def filter.  But keep
5616 		 * our individual unicast filters.
5617 		 */
5618 		if (efx_ef10_filter_insert_addr_list(efx, vlan, false, false))
5619 			efx_ef10_filter_insert_def(efx, vlan,
5620 						   EFX_ENCAP_TYPE_NONE,
5621 						   false, false);
5622 	}
5623 	efx_ef10_filter_insert_def(efx, vlan, EFX_ENCAP_TYPE_VXLAN,
5624 				   false, false);
5625 	efx_ef10_filter_insert_def(efx, vlan, EFX_ENCAP_TYPE_VXLAN |
5626 					      EFX_ENCAP_FLAG_IPV6,
5627 				   false, false);
5628 	efx_ef10_filter_insert_def(efx, vlan, EFX_ENCAP_TYPE_NVGRE,
5629 				   false, false);
5630 	efx_ef10_filter_insert_def(efx, vlan, EFX_ENCAP_TYPE_NVGRE |
5631 					      EFX_ENCAP_FLAG_IPV6,
5632 				   false, false);
5633 	efx_ef10_filter_insert_def(efx, vlan, EFX_ENCAP_TYPE_GENEVE,
5634 				   false, false);
5635 	efx_ef10_filter_insert_def(efx, vlan, EFX_ENCAP_TYPE_GENEVE |
5636 					      EFX_ENCAP_FLAG_IPV6,
5637 				   false, false);
5638 
5639 	/* Insert/renew multicast filters */
5640 	/* If changing promiscuous state with cascaded multicast filters, remove
5641 	 * old filters first, so that packets are dropped rather than duplicated
5642 	 */
5643 	if (nic_data->workaround_26807 &&
5644 	    table->mc_promisc_last != table->mc_promisc)
5645 		efx_ef10_filter_remove_old(efx);
5646 	if (table->mc_promisc) {
5647 		if (nic_data->workaround_26807) {
5648 			/* If we failed to insert promiscuous filters, rollback
5649 			 * and fall back to individual multicast filters
5650 			 */
5651 			if (efx_ef10_filter_insert_def(efx, vlan,
5652 						       EFX_ENCAP_TYPE_NONE,
5653 						       true, true)) {
5654 				/* Changing promisc state, so remove old filters */
5655 				efx_ef10_filter_remove_old(efx);
5656 				efx_ef10_filter_insert_addr_list(efx, vlan,
5657 								 true, false);
5658 			}
5659 		} else {
5660 			/* If we failed to insert promiscuous filters, don't
5661 			 * rollback.  Regardless, also insert the mc_list,
5662 			 * unless it's incomplete due to overflow
5663 			 */
5664 			efx_ef10_filter_insert_def(efx, vlan,
5665 						   EFX_ENCAP_TYPE_NONE,
5666 						   true, false);
5667 			if (!table->mc_overflow)
5668 				efx_ef10_filter_insert_addr_list(efx, vlan,
5669 								 true, false);
5670 		}
5671 	} else {
5672 		/* If any filters failed to insert, rollback and fall back to
5673 		 * promiscuous mode - mc_def filter and maybe broadcast.  If
5674 		 * that fails, roll back again and insert as many of our
5675 		 * individual multicast filters as we can.
5676 		 */
5677 		if (efx_ef10_filter_insert_addr_list(efx, vlan, true, true)) {
5678 			/* Changing promisc state, so remove old filters */
5679 			if (nic_data->workaround_26807)
5680 				efx_ef10_filter_remove_old(efx);
5681 			if (efx_ef10_filter_insert_def(efx, vlan,
5682 						       EFX_ENCAP_TYPE_NONE,
5683 						       true, true))
5684 				efx_ef10_filter_insert_addr_list(efx, vlan,
5685 								 true, false);
5686 		}
5687 	}
5688 	efx_ef10_filter_insert_def(efx, vlan, EFX_ENCAP_TYPE_VXLAN,
5689 				   true, false);
5690 	efx_ef10_filter_insert_def(efx, vlan, EFX_ENCAP_TYPE_VXLAN |
5691 					      EFX_ENCAP_FLAG_IPV6,
5692 				   true, false);
5693 	efx_ef10_filter_insert_def(efx, vlan, EFX_ENCAP_TYPE_NVGRE,
5694 				   true, false);
5695 	efx_ef10_filter_insert_def(efx, vlan, EFX_ENCAP_TYPE_NVGRE |
5696 					      EFX_ENCAP_FLAG_IPV6,
5697 				   true, false);
5698 	efx_ef10_filter_insert_def(efx, vlan, EFX_ENCAP_TYPE_GENEVE,
5699 				   true, false);
5700 	efx_ef10_filter_insert_def(efx, vlan, EFX_ENCAP_TYPE_GENEVE |
5701 					      EFX_ENCAP_FLAG_IPV6,
5702 				   true, false);
5703 }
5704 
5705 /* Caller must hold efx->filter_sem for read if race against
5706  * efx_ef10_filter_table_remove() is possible
5707  */
5708 static void efx_ef10_filter_sync_rx_mode(struct efx_nic *efx)
5709 {
5710 	struct efx_ef10_filter_table *table = efx->filter_state;
5711 	struct net_device *net_dev = efx->net_dev;
5712 	struct efx_ef10_filter_vlan *vlan;
5713 	bool vlan_filter;
5714 
5715 	if (!efx_dev_registered(efx))
5716 		return;
5717 
5718 	if (!table)
5719 		return;
5720 
5721 	efx_ef10_filter_mark_old(efx);
5722 
5723 	/* Copy/convert the address lists; add the primary station
5724 	 * address and broadcast address
5725 	 */
5726 	netif_addr_lock_bh(net_dev);
5727 	efx_ef10_filter_uc_addr_list(efx);
5728 	efx_ef10_filter_mc_addr_list(efx);
5729 	netif_addr_unlock_bh(net_dev);
5730 
5731 	/* If VLAN filtering changes, all old filters are finally removed.
5732 	 * Do it in advance to avoid conflicts for unicast untagged and
5733 	 * VLAN 0 tagged filters.
5734 	 */
5735 	vlan_filter = !!(net_dev->features & NETIF_F_HW_VLAN_CTAG_FILTER);
5736 	if (table->vlan_filter != vlan_filter) {
5737 		table->vlan_filter = vlan_filter;
5738 		efx_ef10_filter_remove_old(efx);
5739 	}
5740 
5741 	list_for_each_entry(vlan, &table->vlan_list, list)
5742 		efx_ef10_filter_vlan_sync_rx_mode(efx, vlan);
5743 
5744 	efx_ef10_filter_remove_old(efx);
5745 	table->mc_promisc_last = table->mc_promisc;
5746 }
5747 
5748 static struct efx_ef10_filter_vlan *efx_ef10_filter_find_vlan(struct efx_nic *efx, u16 vid)
5749 {
5750 	struct efx_ef10_filter_table *table = efx->filter_state;
5751 	struct efx_ef10_filter_vlan *vlan;
5752 
5753 	WARN_ON(!rwsem_is_locked(&efx->filter_sem));
5754 
5755 	list_for_each_entry(vlan, &table->vlan_list, list) {
5756 		if (vlan->vid == vid)
5757 			return vlan;
5758 	}
5759 
5760 	return NULL;
5761 }
5762 
5763 static int efx_ef10_filter_add_vlan(struct efx_nic *efx, u16 vid)
5764 {
5765 	struct efx_ef10_filter_table *table = efx->filter_state;
5766 	struct efx_ef10_filter_vlan *vlan;
5767 	unsigned int i;
5768 
5769 	if (!efx_rwsem_assert_write_locked(&efx->filter_sem))
5770 		return -EINVAL;
5771 
5772 	vlan = efx_ef10_filter_find_vlan(efx, vid);
5773 	if (WARN_ON(vlan)) {
5774 		netif_err(efx, drv, efx->net_dev,
5775 			  "VLAN %u already added\n", vid);
5776 		return -EALREADY;
5777 	}
5778 
5779 	vlan = kzalloc(sizeof(*vlan), GFP_KERNEL);
5780 	if (!vlan)
5781 		return -ENOMEM;
5782 
5783 	vlan->vid = vid;
5784 
5785 	for (i = 0; i < ARRAY_SIZE(vlan->uc); i++)
5786 		vlan->uc[i] = EFX_EF10_FILTER_ID_INVALID;
5787 	for (i = 0; i < ARRAY_SIZE(vlan->mc); i++)
5788 		vlan->mc[i] = EFX_EF10_FILTER_ID_INVALID;
5789 	for (i = 0; i < EFX_EF10_NUM_DEFAULT_FILTERS; i++)
5790 		vlan->default_filters[i] = EFX_EF10_FILTER_ID_INVALID;
5791 
5792 	list_add_tail(&vlan->list, &table->vlan_list);
5793 
5794 	if (efx_dev_registered(efx))
5795 		efx_ef10_filter_vlan_sync_rx_mode(efx, vlan);
5796 
5797 	return 0;
5798 }
5799 
5800 static void efx_ef10_filter_del_vlan_internal(struct efx_nic *efx,
5801 					      struct efx_ef10_filter_vlan *vlan)
5802 {
5803 	unsigned int i;
5804 
5805 	/* See comment in efx_ef10_filter_table_remove() */
5806 	if (!efx_rwsem_assert_write_locked(&efx->filter_sem))
5807 		return;
5808 
5809 	list_del(&vlan->list);
5810 
5811 	for (i = 0; i < ARRAY_SIZE(vlan->uc); i++)
5812 		efx_ef10_filter_remove_unsafe(efx, EFX_FILTER_PRI_AUTO,
5813 					      vlan->uc[i]);
5814 	for (i = 0; i < ARRAY_SIZE(vlan->mc); i++)
5815 		efx_ef10_filter_remove_unsafe(efx, EFX_FILTER_PRI_AUTO,
5816 					      vlan->mc[i]);
5817 	for (i = 0; i < EFX_EF10_NUM_DEFAULT_FILTERS; i++)
5818 		if (vlan->default_filters[i] != EFX_EF10_FILTER_ID_INVALID)
5819 			efx_ef10_filter_remove_unsafe(efx, EFX_FILTER_PRI_AUTO,
5820 						      vlan->default_filters[i]);
5821 
5822 	kfree(vlan);
5823 }
5824 
5825 static void efx_ef10_filter_del_vlan(struct efx_nic *efx, u16 vid)
5826 {
5827 	struct efx_ef10_filter_vlan *vlan;
5828 
5829 	/* See comment in efx_ef10_filter_table_remove() */
5830 	if (!efx_rwsem_assert_write_locked(&efx->filter_sem))
5831 		return;
5832 
5833 	vlan = efx_ef10_filter_find_vlan(efx, vid);
5834 	if (!vlan) {
5835 		netif_err(efx, drv, efx->net_dev,
5836 			  "VLAN %u not found in filter state\n", vid);
5837 		return;
5838 	}
5839 
5840 	efx_ef10_filter_del_vlan_internal(efx, vlan);
5841 }
5842 
5843 static int efx_ef10_set_mac_address(struct efx_nic *efx)
5844 {
5845 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VADAPTOR_SET_MAC_IN_LEN);
5846 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
5847 	bool was_enabled = efx->port_enabled;
5848 	int rc;
5849 
5850 	efx_device_detach_sync(efx);
5851 	efx_net_stop(efx->net_dev);
5852 
5853 	mutex_lock(&efx->mac_lock);
5854 	down_write(&efx->filter_sem);
5855 	efx_ef10_filter_table_remove(efx);
5856 
5857 	ether_addr_copy(MCDI_PTR(inbuf, VADAPTOR_SET_MAC_IN_MACADDR),
5858 			efx->net_dev->dev_addr);
5859 	MCDI_SET_DWORD(inbuf, VADAPTOR_SET_MAC_IN_UPSTREAM_PORT_ID,
5860 		       nic_data->vport_id);
5861 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_VADAPTOR_SET_MAC, inbuf,
5862 				sizeof(inbuf), NULL, 0, NULL);
5863 
5864 	efx_ef10_filter_table_probe(efx);
5865 	up_write(&efx->filter_sem);
5866 	mutex_unlock(&efx->mac_lock);
5867 
5868 	if (was_enabled)
5869 		efx_net_open(efx->net_dev);
5870 	efx_device_attach_if_not_resetting(efx);
5871 
5872 #ifdef CONFIG_SFC_SRIOV
5873 	if (efx->pci_dev->is_virtfn && efx->pci_dev->physfn) {
5874 		struct pci_dev *pci_dev_pf = efx->pci_dev->physfn;
5875 
5876 		if (rc == -EPERM) {
5877 			struct efx_nic *efx_pf;
5878 
5879 			/* Switch to PF and change MAC address on vport */
5880 			efx_pf = pci_get_drvdata(pci_dev_pf);
5881 
5882 			rc = efx_ef10_sriov_set_vf_mac(efx_pf,
5883 						       nic_data->vf_index,
5884 						       efx->net_dev->dev_addr);
5885 		} else if (!rc) {
5886 			struct efx_nic *efx_pf = pci_get_drvdata(pci_dev_pf);
5887 			struct efx_ef10_nic_data *nic_data = efx_pf->nic_data;
5888 			unsigned int i;
5889 
5890 			/* MAC address successfully changed by VF (with MAC
5891 			 * spoofing) so update the parent PF if possible.
5892 			 */
5893 			for (i = 0; i < efx_pf->vf_count; ++i) {
5894 				struct ef10_vf *vf = nic_data->vf + i;
5895 
5896 				if (vf->efx == efx) {
5897 					ether_addr_copy(vf->mac,
5898 							efx->net_dev->dev_addr);
5899 					return 0;
5900 				}
5901 			}
5902 		}
5903 	} else
5904 #endif
5905 	if (rc == -EPERM) {
5906 		netif_err(efx, drv, efx->net_dev,
5907 			  "Cannot change MAC address; use sfboot to enable"
5908 			  " mac-spoofing on this interface\n");
5909 	} else if (rc == -ENOSYS && !efx_ef10_is_vf(efx)) {
5910 		/* If the active MCFW does not support MC_CMD_VADAPTOR_SET_MAC
5911 		 * fall-back to the method of changing the MAC address on the
5912 		 * vport.  This only applies to PFs because such versions of
5913 		 * MCFW do not support VFs.
5914 		 */
5915 		rc = efx_ef10_vport_set_mac_address(efx);
5916 	} else if (rc) {
5917 		efx_mcdi_display_error(efx, MC_CMD_VADAPTOR_SET_MAC,
5918 				       sizeof(inbuf), NULL, 0, rc);
5919 	}
5920 
5921 	return rc;
5922 }
5923 
5924 static int efx_ef10_mac_reconfigure(struct efx_nic *efx)
5925 {
5926 	efx_ef10_filter_sync_rx_mode(efx);
5927 
5928 	return efx_mcdi_set_mac(efx);
5929 }
5930 
5931 static int efx_ef10_mac_reconfigure_vf(struct efx_nic *efx)
5932 {
5933 	efx_ef10_filter_sync_rx_mode(efx);
5934 
5935 	return 0;
5936 }
5937 
5938 static int efx_ef10_start_bist(struct efx_nic *efx, u32 bist_type)
5939 {
5940 	MCDI_DECLARE_BUF(inbuf, MC_CMD_START_BIST_IN_LEN);
5941 
5942 	MCDI_SET_DWORD(inbuf, START_BIST_IN_TYPE, bist_type);
5943 	return efx_mcdi_rpc(efx, MC_CMD_START_BIST, inbuf, sizeof(inbuf),
5944 			    NULL, 0, NULL);
5945 }
5946 
5947 /* MC BISTs follow a different poll mechanism to phy BISTs.
5948  * The BIST is done in the poll handler on the MC, and the MCDI command
5949  * will block until the BIST is done.
5950  */
5951 static int efx_ef10_poll_bist(struct efx_nic *efx)
5952 {
5953 	int rc;
5954 	MCDI_DECLARE_BUF(outbuf, MC_CMD_POLL_BIST_OUT_LEN);
5955 	size_t outlen;
5956 	u32 result;
5957 
5958 	rc = efx_mcdi_rpc(efx, MC_CMD_POLL_BIST, NULL, 0,
5959 			   outbuf, sizeof(outbuf), &outlen);
5960 	if (rc != 0)
5961 		return rc;
5962 
5963 	if (outlen < MC_CMD_POLL_BIST_OUT_LEN)
5964 		return -EIO;
5965 
5966 	result = MCDI_DWORD(outbuf, POLL_BIST_OUT_RESULT);
5967 	switch (result) {
5968 	case MC_CMD_POLL_BIST_PASSED:
5969 		netif_dbg(efx, hw, efx->net_dev, "BIST passed.\n");
5970 		return 0;
5971 	case MC_CMD_POLL_BIST_TIMEOUT:
5972 		netif_err(efx, hw, efx->net_dev, "BIST timed out\n");
5973 		return -EIO;
5974 	case MC_CMD_POLL_BIST_FAILED:
5975 		netif_err(efx, hw, efx->net_dev, "BIST failed.\n");
5976 		return -EIO;
5977 	default:
5978 		netif_err(efx, hw, efx->net_dev,
5979 			  "BIST returned unknown result %u", result);
5980 		return -EIO;
5981 	}
5982 }
5983 
5984 static int efx_ef10_run_bist(struct efx_nic *efx, u32 bist_type)
5985 {
5986 	int rc;
5987 
5988 	netif_dbg(efx, drv, efx->net_dev, "starting BIST type %u\n", bist_type);
5989 
5990 	rc = efx_ef10_start_bist(efx, bist_type);
5991 	if (rc != 0)
5992 		return rc;
5993 
5994 	return efx_ef10_poll_bist(efx);
5995 }
5996 
5997 static int
5998 efx_ef10_test_chip(struct efx_nic *efx, struct efx_self_tests *tests)
5999 {
6000 	int rc, rc2;
6001 
6002 	efx_reset_down(efx, RESET_TYPE_WORLD);
6003 
6004 	rc = efx_mcdi_rpc(efx, MC_CMD_ENABLE_OFFLINE_BIST,
6005 			  NULL, 0, NULL, 0, NULL);
6006 	if (rc != 0)
6007 		goto out;
6008 
6009 	tests->memory = efx_ef10_run_bist(efx, MC_CMD_MC_MEM_BIST) ? -1 : 1;
6010 	tests->registers = efx_ef10_run_bist(efx, MC_CMD_REG_BIST) ? -1 : 1;
6011 
6012 	rc = efx_mcdi_reset(efx, RESET_TYPE_WORLD);
6013 
6014 out:
6015 	if (rc == -EPERM)
6016 		rc = 0;
6017 	rc2 = efx_reset_up(efx, RESET_TYPE_WORLD, rc == 0);
6018 	return rc ? rc : rc2;
6019 }
6020 
6021 #ifdef CONFIG_SFC_MTD
6022 
6023 struct efx_ef10_nvram_type_info {
6024 	u16 type, type_mask;
6025 	u8 port;
6026 	const char *name;
6027 };
6028 
6029 static const struct efx_ef10_nvram_type_info efx_ef10_nvram_types[] = {
6030 	{ NVRAM_PARTITION_TYPE_MC_FIRMWARE,	   0,    0, "sfc_mcfw" },
6031 	{ NVRAM_PARTITION_TYPE_MC_FIRMWARE_BACKUP, 0,    0, "sfc_mcfw_backup" },
6032 	{ NVRAM_PARTITION_TYPE_EXPANSION_ROM,	   0,    0, "sfc_exp_rom" },
6033 	{ NVRAM_PARTITION_TYPE_STATIC_CONFIG,	   0,    0, "sfc_static_cfg" },
6034 	{ NVRAM_PARTITION_TYPE_DYNAMIC_CONFIG,	   0,    0, "sfc_dynamic_cfg" },
6035 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT0, 0,   0, "sfc_exp_rom_cfg" },
6036 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT1, 0,   1, "sfc_exp_rom_cfg" },
6037 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT2, 0,   2, "sfc_exp_rom_cfg" },
6038 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT3, 0,   3, "sfc_exp_rom_cfg" },
6039 	{ NVRAM_PARTITION_TYPE_LICENSE,		   0,    0, "sfc_license" },
6040 	{ NVRAM_PARTITION_TYPE_PHY_MIN,		   0xff, 0, "sfc_phy_fw" },
6041 	{ NVRAM_PARTITION_TYPE_MUM_FIRMWARE,	   0,    0, "sfc_mumfw" },
6042 	{ NVRAM_PARTITION_TYPE_EXPANSION_UEFI,	   0,    0, "sfc_uefi" },
6043 	{ NVRAM_PARTITION_TYPE_DYNCONFIG_DEFAULTS, 0,    0, "sfc_dynamic_cfg_dflt" },
6044 	{ NVRAM_PARTITION_TYPE_ROMCONFIG_DEFAULTS, 0,    0, "sfc_exp_rom_cfg_dflt" },
6045 	{ NVRAM_PARTITION_TYPE_STATUS,		   0,    0, "sfc_status" },
6046 	{ NVRAM_PARTITION_TYPE_BUNDLE,		   0,    0, "sfc_bundle" },
6047 	{ NVRAM_PARTITION_TYPE_BUNDLE_METADATA,	   0,    0, "sfc_bundle_metadata" },
6048 };
6049 #define EF10_NVRAM_PARTITION_COUNT	ARRAY_SIZE(efx_ef10_nvram_types)
6050 
6051 static int efx_ef10_mtd_probe_partition(struct efx_nic *efx,
6052 					struct efx_mcdi_mtd_partition *part,
6053 					unsigned int type,
6054 					unsigned long *found)
6055 {
6056 	MCDI_DECLARE_BUF(inbuf, MC_CMD_NVRAM_METADATA_IN_LEN);
6057 	MCDI_DECLARE_BUF(outbuf, MC_CMD_NVRAM_METADATA_OUT_LENMAX);
6058 	const struct efx_ef10_nvram_type_info *info;
6059 	size_t size, erase_size, outlen;
6060 	int type_idx = 0;
6061 	bool protected;
6062 	int rc;
6063 
6064 	for (type_idx = 0; ; type_idx++) {
6065 		if (type_idx == EF10_NVRAM_PARTITION_COUNT)
6066 			return -ENODEV;
6067 		info = efx_ef10_nvram_types + type_idx;
6068 		if ((type & ~info->type_mask) == info->type)
6069 			break;
6070 	}
6071 	if (info->port != efx_port_num(efx))
6072 		return -ENODEV;
6073 
6074 	rc = efx_mcdi_nvram_info(efx, type, &size, &erase_size, &protected);
6075 	if (rc)
6076 		return rc;
6077 	if (protected &&
6078 	    (type != NVRAM_PARTITION_TYPE_DYNCONFIG_DEFAULTS &&
6079 	     type != NVRAM_PARTITION_TYPE_ROMCONFIG_DEFAULTS))
6080 		/* Hide protected partitions that don't provide defaults. */
6081 		return -ENODEV;
6082 
6083 	if (protected)
6084 		/* Protected partitions are read only. */
6085 		erase_size = 0;
6086 
6087 	/* If we've already exposed a partition of this type, hide this
6088 	 * duplicate.  All operations on MTDs are keyed by the type anyway,
6089 	 * so we can't act on the duplicate.
6090 	 */
6091 	if (__test_and_set_bit(type_idx, found))
6092 		return -EEXIST;
6093 
6094 	part->nvram_type = type;
6095 
6096 	MCDI_SET_DWORD(inbuf, NVRAM_METADATA_IN_TYPE, type);
6097 	rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_METADATA, inbuf, sizeof(inbuf),
6098 			  outbuf, sizeof(outbuf), &outlen);
6099 	if (rc)
6100 		return rc;
6101 	if (outlen < MC_CMD_NVRAM_METADATA_OUT_LENMIN)
6102 		return -EIO;
6103 	if (MCDI_DWORD(outbuf, NVRAM_METADATA_OUT_FLAGS) &
6104 	    (1 << MC_CMD_NVRAM_METADATA_OUT_SUBTYPE_VALID_LBN))
6105 		part->fw_subtype = MCDI_DWORD(outbuf,
6106 					      NVRAM_METADATA_OUT_SUBTYPE);
6107 
6108 	part->common.dev_type_name = "EF10 NVRAM manager";
6109 	part->common.type_name = info->name;
6110 
6111 	part->common.mtd.type = MTD_NORFLASH;
6112 	part->common.mtd.flags = MTD_CAP_NORFLASH;
6113 	part->common.mtd.size = size;
6114 	part->common.mtd.erasesize = erase_size;
6115 	/* sfc_status is read-only */
6116 	if (!erase_size)
6117 		part->common.mtd.flags |= MTD_NO_ERASE;
6118 
6119 	return 0;
6120 }
6121 
6122 static int efx_ef10_mtd_probe(struct efx_nic *efx)
6123 {
6124 	MCDI_DECLARE_BUF(outbuf, MC_CMD_NVRAM_PARTITIONS_OUT_LENMAX);
6125 	DECLARE_BITMAP(found, EF10_NVRAM_PARTITION_COUNT) = { 0 };
6126 	struct efx_mcdi_mtd_partition *parts;
6127 	size_t outlen, n_parts_total, i, n_parts;
6128 	unsigned int type;
6129 	int rc;
6130 
6131 	ASSERT_RTNL();
6132 
6133 	BUILD_BUG_ON(MC_CMD_NVRAM_PARTITIONS_IN_LEN != 0);
6134 	rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_PARTITIONS, NULL, 0,
6135 			  outbuf, sizeof(outbuf), &outlen);
6136 	if (rc)
6137 		return rc;
6138 	if (outlen < MC_CMD_NVRAM_PARTITIONS_OUT_LENMIN)
6139 		return -EIO;
6140 
6141 	n_parts_total = MCDI_DWORD(outbuf, NVRAM_PARTITIONS_OUT_NUM_PARTITIONS);
6142 	if (n_parts_total >
6143 	    MCDI_VAR_ARRAY_LEN(outlen, NVRAM_PARTITIONS_OUT_TYPE_ID))
6144 		return -EIO;
6145 
6146 	parts = kcalloc(n_parts_total, sizeof(*parts), GFP_KERNEL);
6147 	if (!parts)
6148 		return -ENOMEM;
6149 
6150 	n_parts = 0;
6151 	for (i = 0; i < n_parts_total; i++) {
6152 		type = MCDI_ARRAY_DWORD(outbuf, NVRAM_PARTITIONS_OUT_TYPE_ID,
6153 					i);
6154 		rc = efx_ef10_mtd_probe_partition(efx, &parts[n_parts], type,
6155 						  found);
6156 		if (rc == -EEXIST || rc == -ENODEV)
6157 			continue;
6158 		if (rc)
6159 			goto fail;
6160 		n_parts++;
6161 	}
6162 
6163 	rc = efx_mtd_add(efx, &parts[0].common, n_parts, sizeof(*parts));
6164 fail:
6165 	if (rc)
6166 		kfree(parts);
6167 	return rc;
6168 }
6169 
6170 #endif /* CONFIG_SFC_MTD */
6171 
6172 static void efx_ef10_ptp_write_host_time(struct efx_nic *efx, u32 host_time)
6173 {
6174 	_efx_writed(efx, cpu_to_le32(host_time), ER_DZ_MC_DB_LWRD);
6175 }
6176 
6177 static void efx_ef10_ptp_write_host_time_vf(struct efx_nic *efx,
6178 					    u32 host_time) {}
6179 
6180 static int efx_ef10_rx_enable_timestamping(struct efx_channel *channel,
6181 					   bool temp)
6182 {
6183 	MCDI_DECLARE_BUF(inbuf, MC_CMD_PTP_IN_TIME_EVENT_SUBSCRIBE_LEN);
6184 	int rc;
6185 
6186 	if (channel->sync_events_state == SYNC_EVENTS_REQUESTED ||
6187 	    channel->sync_events_state == SYNC_EVENTS_VALID ||
6188 	    (temp && channel->sync_events_state == SYNC_EVENTS_DISABLED))
6189 		return 0;
6190 	channel->sync_events_state = SYNC_EVENTS_REQUESTED;
6191 
6192 	MCDI_SET_DWORD(inbuf, PTP_IN_OP, MC_CMD_PTP_OP_TIME_EVENT_SUBSCRIBE);
6193 	MCDI_SET_DWORD(inbuf, PTP_IN_PERIPH_ID, 0);
6194 	MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_SUBSCRIBE_QUEUE,
6195 		       channel->channel);
6196 
6197 	rc = efx_mcdi_rpc(channel->efx, MC_CMD_PTP,
6198 			  inbuf, sizeof(inbuf), NULL, 0, NULL);
6199 
6200 	if (rc != 0)
6201 		channel->sync_events_state = temp ? SYNC_EVENTS_QUIESCENT :
6202 						    SYNC_EVENTS_DISABLED;
6203 
6204 	return rc;
6205 }
6206 
6207 static int efx_ef10_rx_disable_timestamping(struct efx_channel *channel,
6208 					    bool temp)
6209 {
6210 	MCDI_DECLARE_BUF(inbuf, MC_CMD_PTP_IN_TIME_EVENT_UNSUBSCRIBE_LEN);
6211 	int rc;
6212 
6213 	if (channel->sync_events_state == SYNC_EVENTS_DISABLED ||
6214 	    (temp && channel->sync_events_state == SYNC_EVENTS_QUIESCENT))
6215 		return 0;
6216 	if (channel->sync_events_state == SYNC_EVENTS_QUIESCENT) {
6217 		channel->sync_events_state = SYNC_EVENTS_DISABLED;
6218 		return 0;
6219 	}
6220 	channel->sync_events_state = temp ? SYNC_EVENTS_QUIESCENT :
6221 					    SYNC_EVENTS_DISABLED;
6222 
6223 	MCDI_SET_DWORD(inbuf, PTP_IN_OP, MC_CMD_PTP_OP_TIME_EVENT_UNSUBSCRIBE);
6224 	MCDI_SET_DWORD(inbuf, PTP_IN_PERIPH_ID, 0);
6225 	MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_UNSUBSCRIBE_CONTROL,
6226 		       MC_CMD_PTP_IN_TIME_EVENT_UNSUBSCRIBE_SINGLE);
6227 	MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_UNSUBSCRIBE_QUEUE,
6228 		       channel->channel);
6229 
6230 	rc = efx_mcdi_rpc(channel->efx, MC_CMD_PTP,
6231 			  inbuf, sizeof(inbuf), NULL, 0, NULL);
6232 
6233 	return rc;
6234 }
6235 
6236 static int efx_ef10_ptp_set_ts_sync_events(struct efx_nic *efx, bool en,
6237 					   bool temp)
6238 {
6239 	int (*set)(struct efx_channel *channel, bool temp);
6240 	struct efx_channel *channel;
6241 
6242 	set = en ?
6243 	      efx_ef10_rx_enable_timestamping :
6244 	      efx_ef10_rx_disable_timestamping;
6245 
6246 	channel = efx_ptp_channel(efx);
6247 	if (channel) {
6248 		int rc = set(channel, temp);
6249 		if (en && rc != 0) {
6250 			efx_ef10_ptp_set_ts_sync_events(efx, false, temp);
6251 			return rc;
6252 		}
6253 	}
6254 
6255 	return 0;
6256 }
6257 
6258 static int efx_ef10_ptp_set_ts_config_vf(struct efx_nic *efx,
6259 					 struct hwtstamp_config *init)
6260 {
6261 	return -EOPNOTSUPP;
6262 }
6263 
6264 static int efx_ef10_ptp_set_ts_config(struct efx_nic *efx,
6265 				      struct hwtstamp_config *init)
6266 {
6267 	int rc;
6268 
6269 	switch (init->rx_filter) {
6270 	case HWTSTAMP_FILTER_NONE:
6271 		efx_ef10_ptp_set_ts_sync_events(efx, false, false);
6272 		/* if TX timestamping is still requested then leave PTP on */
6273 		return efx_ptp_change_mode(efx,
6274 					   init->tx_type != HWTSTAMP_TX_OFF, 0);
6275 	case HWTSTAMP_FILTER_ALL:
6276 	case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
6277 	case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
6278 	case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
6279 	case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
6280 	case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
6281 	case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
6282 	case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
6283 	case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
6284 	case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
6285 	case HWTSTAMP_FILTER_PTP_V2_EVENT:
6286 	case HWTSTAMP_FILTER_PTP_V2_SYNC:
6287 	case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
6288 	case HWTSTAMP_FILTER_NTP_ALL:
6289 		init->rx_filter = HWTSTAMP_FILTER_ALL;
6290 		rc = efx_ptp_change_mode(efx, true, 0);
6291 		if (!rc)
6292 			rc = efx_ef10_ptp_set_ts_sync_events(efx, true, false);
6293 		if (rc)
6294 			efx_ptp_change_mode(efx, false, 0);
6295 		return rc;
6296 	default:
6297 		return -ERANGE;
6298 	}
6299 }
6300 
6301 static int efx_ef10_get_phys_port_id(struct efx_nic *efx,
6302 				     struct netdev_phys_item_id *ppid)
6303 {
6304 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
6305 
6306 	if (!is_valid_ether_addr(nic_data->port_id))
6307 		return -EOPNOTSUPP;
6308 
6309 	ppid->id_len = ETH_ALEN;
6310 	memcpy(ppid->id, nic_data->port_id, ppid->id_len);
6311 
6312 	return 0;
6313 }
6314 
6315 static int efx_ef10_vlan_rx_add_vid(struct efx_nic *efx, __be16 proto, u16 vid)
6316 {
6317 	if (proto != htons(ETH_P_8021Q))
6318 		return -EINVAL;
6319 
6320 	return efx_ef10_add_vlan(efx, vid);
6321 }
6322 
6323 static int efx_ef10_vlan_rx_kill_vid(struct efx_nic *efx, __be16 proto, u16 vid)
6324 {
6325 	if (proto != htons(ETH_P_8021Q))
6326 		return -EINVAL;
6327 
6328 	return efx_ef10_del_vlan(efx, vid);
6329 }
6330 
6331 /* We rely on the MCDI wiping out our TX rings if it made any changes to the
6332  * ports table, ensuring that any TSO descriptors that were made on a now-
6333  * removed tunnel port will be blown away and won't break things when we try
6334  * to transmit them using the new ports table.
6335  */
6336 static int efx_ef10_set_udp_tnl_ports(struct efx_nic *efx, bool unloading)
6337 {
6338 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
6339 	MCDI_DECLARE_BUF(inbuf, MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_IN_LENMAX);
6340 	MCDI_DECLARE_BUF(outbuf, MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_OUT_LEN);
6341 	bool will_reset = false;
6342 	size_t num_entries = 0;
6343 	size_t inlen, outlen;
6344 	size_t i;
6345 	int rc;
6346 	efx_dword_t flags_and_num_entries;
6347 
6348 	WARN_ON(!mutex_is_locked(&nic_data->udp_tunnels_lock));
6349 
6350 	nic_data->udp_tunnels_dirty = false;
6351 
6352 	if (!(nic_data->datapath_caps &
6353 	    (1 << MC_CMD_GET_CAPABILITIES_OUT_VXLAN_NVGRE_LBN))) {
6354 		efx_device_attach_if_not_resetting(efx);
6355 		return 0;
6356 	}
6357 
6358 	BUILD_BUG_ON(ARRAY_SIZE(nic_data->udp_tunnels) >
6359 		     MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_IN_ENTRIES_MAXNUM);
6360 
6361 	for (i = 0; i < ARRAY_SIZE(nic_data->udp_tunnels); ++i) {
6362 		if (nic_data->udp_tunnels[i].count &&
6363 		    nic_data->udp_tunnels[i].port) {
6364 			efx_dword_t entry;
6365 
6366 			EFX_POPULATE_DWORD_2(entry,
6367 				TUNNEL_ENCAP_UDP_PORT_ENTRY_UDP_PORT,
6368 					ntohs(nic_data->udp_tunnels[i].port),
6369 				TUNNEL_ENCAP_UDP_PORT_ENTRY_PROTOCOL,
6370 					nic_data->udp_tunnels[i].type);
6371 			*_MCDI_ARRAY_DWORD(inbuf,
6372 				SET_TUNNEL_ENCAP_UDP_PORTS_IN_ENTRIES,
6373 				num_entries++) = entry;
6374 		}
6375 	}
6376 
6377 	BUILD_BUG_ON((MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_IN_NUM_ENTRIES_OFST -
6378 		      MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_IN_FLAGS_OFST) * 8 !=
6379 		     EFX_WORD_1_LBN);
6380 	BUILD_BUG_ON(MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_IN_NUM_ENTRIES_LEN * 8 !=
6381 		     EFX_WORD_1_WIDTH);
6382 	EFX_POPULATE_DWORD_2(flags_and_num_entries,
6383 			     MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_IN_UNLOADING,
6384 				!!unloading,
6385 			     EFX_WORD_1, num_entries);
6386 	*_MCDI_DWORD(inbuf, SET_TUNNEL_ENCAP_UDP_PORTS_IN_FLAGS) =
6387 		flags_and_num_entries;
6388 
6389 	inlen = MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_IN_LEN(num_entries);
6390 
6391 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS,
6392 				inbuf, inlen, outbuf, sizeof(outbuf), &outlen);
6393 	if (rc == -EIO) {
6394 		/* Most likely the MC rebooted due to another function also
6395 		 * setting its tunnel port list. Mark the tunnel port list as
6396 		 * dirty, so it will be pushed upon coming up from the reboot.
6397 		 */
6398 		nic_data->udp_tunnels_dirty = true;
6399 		return 0;
6400 	}
6401 
6402 	if (rc) {
6403 		/* expected not available on unprivileged functions */
6404 		if (rc != -EPERM)
6405 			netif_warn(efx, drv, efx->net_dev,
6406 				   "Unable to set UDP tunnel ports; rc=%d.\n", rc);
6407 	} else if (MCDI_DWORD(outbuf, SET_TUNNEL_ENCAP_UDP_PORTS_OUT_FLAGS) &
6408 		   (1 << MC_CMD_SET_TUNNEL_ENCAP_UDP_PORTS_OUT_RESETTING_LBN)) {
6409 		netif_info(efx, drv, efx->net_dev,
6410 			   "Rebooting MC due to UDP tunnel port list change\n");
6411 		will_reset = true;
6412 		if (unloading)
6413 			/* Delay for the MC reset to complete. This will make
6414 			 * unloading other functions a bit smoother. This is a
6415 			 * race, but the other unload will work whichever way
6416 			 * it goes, this just avoids an unnecessary error
6417 			 * message.
6418 			 */
6419 			msleep(100);
6420 	}
6421 	if (!will_reset && !unloading) {
6422 		/* The caller will have detached, relying on the MC reset to
6423 		 * trigger a re-attach.  Since there won't be an MC reset, we
6424 		 * have to do the attach ourselves.
6425 		 */
6426 		efx_device_attach_if_not_resetting(efx);
6427 	}
6428 
6429 	return rc;
6430 }
6431 
6432 static int efx_ef10_udp_tnl_push_ports(struct efx_nic *efx)
6433 {
6434 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
6435 	int rc = 0;
6436 
6437 	mutex_lock(&nic_data->udp_tunnels_lock);
6438 	if (nic_data->udp_tunnels_dirty) {
6439 		/* Make sure all TX are stopped while we modify the table, else
6440 		 * we might race against an efx_features_check().
6441 		 */
6442 		efx_device_detach_sync(efx);
6443 		rc = efx_ef10_set_udp_tnl_ports(efx, false);
6444 	}
6445 	mutex_unlock(&nic_data->udp_tunnels_lock);
6446 	return rc;
6447 }
6448 
6449 static struct efx_udp_tunnel *__efx_ef10_udp_tnl_lookup_port(struct efx_nic *efx,
6450 							     __be16 port)
6451 {
6452 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
6453 	size_t i;
6454 
6455 	for (i = 0; i < ARRAY_SIZE(nic_data->udp_tunnels); ++i) {
6456 		if (!nic_data->udp_tunnels[i].count)
6457 			continue;
6458 		if (nic_data->udp_tunnels[i].port == port)
6459 			return &nic_data->udp_tunnels[i];
6460 	}
6461 	return NULL;
6462 }
6463 
6464 static int efx_ef10_udp_tnl_add_port(struct efx_nic *efx,
6465 				     struct efx_udp_tunnel tnl)
6466 {
6467 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
6468 	struct efx_udp_tunnel *match;
6469 	char typebuf[8];
6470 	size_t i;
6471 	int rc;
6472 
6473 	if (!(nic_data->datapath_caps &
6474 	      (1 << MC_CMD_GET_CAPABILITIES_OUT_VXLAN_NVGRE_LBN)))
6475 		return 0;
6476 
6477 	efx_get_udp_tunnel_type_name(tnl.type, typebuf, sizeof(typebuf));
6478 	netif_dbg(efx, drv, efx->net_dev, "Adding UDP tunnel (%s) port %d\n",
6479 		  typebuf, ntohs(tnl.port));
6480 
6481 	mutex_lock(&nic_data->udp_tunnels_lock);
6482 	/* Make sure all TX are stopped while we add to the table, else we
6483 	 * might race against an efx_features_check().
6484 	 */
6485 	efx_device_detach_sync(efx);
6486 
6487 	match = __efx_ef10_udp_tnl_lookup_port(efx, tnl.port);
6488 	if (match != NULL) {
6489 		if (match->type == tnl.type) {
6490 			netif_dbg(efx, drv, efx->net_dev,
6491 				  "Referencing existing tunnel entry\n");
6492 			match->count++;
6493 			/* No need to cause an MCDI update */
6494 			rc = 0;
6495 			goto unlock_out;
6496 		}
6497 		efx_get_udp_tunnel_type_name(match->type,
6498 					     typebuf, sizeof(typebuf));
6499 		netif_dbg(efx, drv, efx->net_dev,
6500 			  "UDP port %d is already in use by %s\n",
6501 			  ntohs(tnl.port), typebuf);
6502 		rc = -EEXIST;
6503 		goto unlock_out;
6504 	}
6505 
6506 	for (i = 0; i < ARRAY_SIZE(nic_data->udp_tunnels); ++i)
6507 		if (!nic_data->udp_tunnels[i].count) {
6508 			nic_data->udp_tunnels[i] = tnl;
6509 			nic_data->udp_tunnels[i].count = 1;
6510 			rc = efx_ef10_set_udp_tnl_ports(efx, false);
6511 			goto unlock_out;
6512 		}
6513 
6514 	netif_dbg(efx, drv, efx->net_dev,
6515 		  "Unable to add UDP tunnel (%s) port %d; insufficient resources.\n",
6516 		  typebuf, ntohs(tnl.port));
6517 
6518 	rc = -ENOMEM;
6519 
6520 unlock_out:
6521 	mutex_unlock(&nic_data->udp_tunnels_lock);
6522 	return rc;
6523 }
6524 
6525 /* Called under the TX lock with the TX queue running, hence no-one can be
6526  * in the middle of updating the UDP tunnels table.  However, they could
6527  * have tried and failed the MCDI, in which case they'll have set the dirty
6528  * flag before dropping their locks.
6529  */
6530 static bool efx_ef10_udp_tnl_has_port(struct efx_nic *efx, __be16 port)
6531 {
6532 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
6533 
6534 	if (!(nic_data->datapath_caps &
6535 	      (1 << MC_CMD_GET_CAPABILITIES_OUT_VXLAN_NVGRE_LBN)))
6536 		return false;
6537 
6538 	if (nic_data->udp_tunnels_dirty)
6539 		/* SW table may not match HW state, so just assume we can't
6540 		 * use any UDP tunnel offloads.
6541 		 */
6542 		return false;
6543 
6544 	return __efx_ef10_udp_tnl_lookup_port(efx, port) != NULL;
6545 }
6546 
6547 static int efx_ef10_udp_tnl_del_port(struct efx_nic *efx,
6548 				     struct efx_udp_tunnel tnl)
6549 {
6550 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
6551 	struct efx_udp_tunnel *match;
6552 	char typebuf[8];
6553 	int rc;
6554 
6555 	if (!(nic_data->datapath_caps &
6556 	      (1 << MC_CMD_GET_CAPABILITIES_OUT_VXLAN_NVGRE_LBN)))
6557 		return 0;
6558 
6559 	efx_get_udp_tunnel_type_name(tnl.type, typebuf, sizeof(typebuf));
6560 	netif_dbg(efx, drv, efx->net_dev, "Removing UDP tunnel (%s) port %d\n",
6561 		  typebuf, ntohs(tnl.port));
6562 
6563 	mutex_lock(&nic_data->udp_tunnels_lock);
6564 	/* Make sure all TX are stopped while we remove from the table, else we
6565 	 * might race against an efx_features_check().
6566 	 */
6567 	efx_device_detach_sync(efx);
6568 
6569 	match = __efx_ef10_udp_tnl_lookup_port(efx, tnl.port);
6570 	if (match != NULL) {
6571 		if (match->type == tnl.type) {
6572 			if (--match->count) {
6573 				/* Port is still in use, so nothing to do */
6574 				netif_dbg(efx, drv, efx->net_dev,
6575 					  "UDP tunnel port %d remains active\n",
6576 					  ntohs(tnl.port));
6577 				rc = 0;
6578 				goto out_unlock;
6579 			}
6580 			rc = efx_ef10_set_udp_tnl_ports(efx, false);
6581 			goto out_unlock;
6582 		}
6583 		efx_get_udp_tunnel_type_name(match->type,
6584 					     typebuf, sizeof(typebuf));
6585 		netif_warn(efx, drv, efx->net_dev,
6586 			   "UDP port %d is actually in use by %s, not removing\n",
6587 			   ntohs(tnl.port), typebuf);
6588 	}
6589 	rc = -ENOENT;
6590 
6591 out_unlock:
6592 	mutex_unlock(&nic_data->udp_tunnels_lock);
6593 	return rc;
6594 }
6595 
6596 #define EF10_OFFLOAD_FEATURES		\
6597 	(NETIF_F_IP_CSUM |		\
6598 	 NETIF_F_HW_VLAN_CTAG_FILTER |	\
6599 	 NETIF_F_IPV6_CSUM |		\
6600 	 NETIF_F_RXHASH |		\
6601 	 NETIF_F_NTUPLE)
6602 
6603 const struct efx_nic_type efx_hunt_a0_vf_nic_type = {
6604 	.is_vf = true,
6605 	.mem_bar = efx_ef10_vf_mem_bar,
6606 	.mem_map_size = efx_ef10_mem_map_size,
6607 	.probe = efx_ef10_probe_vf,
6608 	.remove = efx_ef10_remove,
6609 	.dimension_resources = efx_ef10_dimension_resources,
6610 	.init = efx_ef10_init_nic,
6611 	.fini = efx_port_dummy_op_void,
6612 	.map_reset_reason = efx_ef10_map_reset_reason,
6613 	.map_reset_flags = efx_ef10_map_reset_flags,
6614 	.reset = efx_ef10_reset,
6615 	.probe_port = efx_mcdi_port_probe,
6616 	.remove_port = efx_mcdi_port_remove,
6617 	.fini_dmaq = efx_ef10_fini_dmaq,
6618 	.prepare_flr = efx_ef10_prepare_flr,
6619 	.finish_flr = efx_port_dummy_op_void,
6620 	.describe_stats = efx_ef10_describe_stats,
6621 	.update_stats = efx_ef10_update_stats_vf,
6622 	.start_stats = efx_port_dummy_op_void,
6623 	.pull_stats = efx_port_dummy_op_void,
6624 	.stop_stats = efx_port_dummy_op_void,
6625 	.set_id_led = efx_mcdi_set_id_led,
6626 	.push_irq_moderation = efx_ef10_push_irq_moderation,
6627 	.reconfigure_mac = efx_ef10_mac_reconfigure_vf,
6628 	.check_mac_fault = efx_mcdi_mac_check_fault,
6629 	.reconfigure_port = efx_mcdi_port_reconfigure,
6630 	.get_wol = efx_ef10_get_wol_vf,
6631 	.set_wol = efx_ef10_set_wol_vf,
6632 	.resume_wol = efx_port_dummy_op_void,
6633 	.mcdi_request = efx_ef10_mcdi_request,
6634 	.mcdi_poll_response = efx_ef10_mcdi_poll_response,
6635 	.mcdi_read_response = efx_ef10_mcdi_read_response,
6636 	.mcdi_poll_reboot = efx_ef10_mcdi_poll_reboot,
6637 	.mcdi_reboot_detected = efx_ef10_mcdi_reboot_detected,
6638 	.irq_enable_master = efx_port_dummy_op_void,
6639 	.irq_test_generate = efx_ef10_irq_test_generate,
6640 	.irq_disable_non_ev = efx_port_dummy_op_void,
6641 	.irq_handle_msi = efx_ef10_msi_interrupt,
6642 	.irq_handle_legacy = efx_ef10_legacy_interrupt,
6643 	.tx_probe = efx_ef10_tx_probe,
6644 	.tx_init = efx_ef10_tx_init,
6645 	.tx_remove = efx_ef10_tx_remove,
6646 	.tx_write = efx_ef10_tx_write,
6647 	.tx_limit_len = efx_ef10_tx_limit_len,
6648 	.rx_push_rss_config = efx_ef10_vf_rx_push_rss_config,
6649 	.rx_pull_rss_config = efx_ef10_rx_pull_rss_config,
6650 	.rx_probe = efx_ef10_rx_probe,
6651 	.rx_init = efx_ef10_rx_init,
6652 	.rx_remove = efx_ef10_rx_remove,
6653 	.rx_write = efx_ef10_rx_write,
6654 	.rx_defer_refill = efx_ef10_rx_defer_refill,
6655 	.ev_probe = efx_ef10_ev_probe,
6656 	.ev_init = efx_ef10_ev_init,
6657 	.ev_fini = efx_ef10_ev_fini,
6658 	.ev_remove = efx_ef10_ev_remove,
6659 	.ev_process = efx_ef10_ev_process,
6660 	.ev_read_ack = efx_ef10_ev_read_ack,
6661 	.ev_test_generate = efx_ef10_ev_test_generate,
6662 	.filter_table_probe = efx_ef10_filter_table_probe,
6663 	.filter_table_restore = efx_ef10_filter_table_restore,
6664 	.filter_table_remove = efx_ef10_filter_table_remove,
6665 	.filter_update_rx_scatter = efx_ef10_filter_update_rx_scatter,
6666 	.filter_insert = efx_ef10_filter_insert,
6667 	.filter_remove_safe = efx_ef10_filter_remove_safe,
6668 	.filter_get_safe = efx_ef10_filter_get_safe,
6669 	.filter_clear_rx = efx_ef10_filter_clear_rx,
6670 	.filter_count_rx_used = efx_ef10_filter_count_rx_used,
6671 	.filter_get_rx_id_limit = efx_ef10_filter_get_rx_id_limit,
6672 	.filter_get_rx_ids = efx_ef10_filter_get_rx_ids,
6673 #ifdef CONFIG_RFS_ACCEL
6674 	.filter_rfs_expire_one = efx_ef10_filter_rfs_expire_one,
6675 #endif
6676 #ifdef CONFIG_SFC_MTD
6677 	.mtd_probe = efx_port_dummy_op_int,
6678 #endif
6679 	.ptp_write_host_time = efx_ef10_ptp_write_host_time_vf,
6680 	.ptp_set_ts_config = efx_ef10_ptp_set_ts_config_vf,
6681 	.vlan_rx_add_vid = efx_ef10_vlan_rx_add_vid,
6682 	.vlan_rx_kill_vid = efx_ef10_vlan_rx_kill_vid,
6683 #ifdef CONFIG_SFC_SRIOV
6684 	.vswitching_probe = efx_ef10_vswitching_probe_vf,
6685 	.vswitching_restore = efx_ef10_vswitching_restore_vf,
6686 	.vswitching_remove = efx_ef10_vswitching_remove_vf,
6687 #endif
6688 	.get_mac_address = efx_ef10_get_mac_address_vf,
6689 	.set_mac_address = efx_ef10_set_mac_address,
6690 
6691 	.get_phys_port_id = efx_ef10_get_phys_port_id,
6692 	.revision = EFX_REV_HUNT_A0,
6693 	.max_dma_mask = DMA_BIT_MASK(ESF_DZ_TX_KER_BUF_ADDR_WIDTH),
6694 	.rx_prefix_size = ES_DZ_RX_PREFIX_SIZE,
6695 	.rx_hash_offset = ES_DZ_RX_PREFIX_HASH_OFST,
6696 	.rx_ts_offset = ES_DZ_RX_PREFIX_TSTAMP_OFST,
6697 	.can_rx_scatter = true,
6698 	.always_rx_scatter = true,
6699 	.min_interrupt_mode = EFX_INT_MODE_MSIX,
6700 	.max_interrupt_mode = EFX_INT_MODE_MSIX,
6701 	.timer_period_max = 1 << ERF_DD_EVQ_IND_TIMER_VAL_WIDTH,
6702 	.offload_features = EF10_OFFLOAD_FEATURES,
6703 	.mcdi_max_ver = 2,
6704 	.max_rx_ip_filters = HUNT_FILTER_TBL_ROWS,
6705 	.hwtstamp_filters = 1 << HWTSTAMP_FILTER_NONE |
6706 			    1 << HWTSTAMP_FILTER_ALL,
6707 	.rx_hash_key_size = 40,
6708 };
6709 
6710 const struct efx_nic_type efx_hunt_a0_nic_type = {
6711 	.is_vf = false,
6712 	.mem_bar = efx_ef10_pf_mem_bar,
6713 	.mem_map_size = efx_ef10_mem_map_size,
6714 	.probe = efx_ef10_probe_pf,
6715 	.remove = efx_ef10_remove,
6716 	.dimension_resources = efx_ef10_dimension_resources,
6717 	.init = efx_ef10_init_nic,
6718 	.fini = efx_port_dummy_op_void,
6719 	.map_reset_reason = efx_ef10_map_reset_reason,
6720 	.map_reset_flags = efx_ef10_map_reset_flags,
6721 	.reset = efx_ef10_reset,
6722 	.probe_port = efx_mcdi_port_probe,
6723 	.remove_port = efx_mcdi_port_remove,
6724 	.fini_dmaq = efx_ef10_fini_dmaq,
6725 	.prepare_flr = efx_ef10_prepare_flr,
6726 	.finish_flr = efx_port_dummy_op_void,
6727 	.describe_stats = efx_ef10_describe_stats,
6728 	.update_stats = efx_ef10_update_stats_pf,
6729 	.start_stats = efx_mcdi_mac_start_stats,
6730 	.pull_stats = efx_mcdi_mac_pull_stats,
6731 	.stop_stats = efx_mcdi_mac_stop_stats,
6732 	.set_id_led = efx_mcdi_set_id_led,
6733 	.push_irq_moderation = efx_ef10_push_irq_moderation,
6734 	.reconfigure_mac = efx_ef10_mac_reconfigure,
6735 	.check_mac_fault = efx_mcdi_mac_check_fault,
6736 	.reconfigure_port = efx_mcdi_port_reconfigure,
6737 	.get_wol = efx_ef10_get_wol,
6738 	.set_wol = efx_ef10_set_wol,
6739 	.resume_wol = efx_port_dummy_op_void,
6740 	.test_chip = efx_ef10_test_chip,
6741 	.test_nvram = efx_mcdi_nvram_test_all,
6742 	.mcdi_request = efx_ef10_mcdi_request,
6743 	.mcdi_poll_response = efx_ef10_mcdi_poll_response,
6744 	.mcdi_read_response = efx_ef10_mcdi_read_response,
6745 	.mcdi_poll_reboot = efx_ef10_mcdi_poll_reboot,
6746 	.mcdi_reboot_detected = efx_ef10_mcdi_reboot_detected,
6747 	.irq_enable_master = efx_port_dummy_op_void,
6748 	.irq_test_generate = efx_ef10_irq_test_generate,
6749 	.irq_disable_non_ev = efx_port_dummy_op_void,
6750 	.irq_handle_msi = efx_ef10_msi_interrupt,
6751 	.irq_handle_legacy = efx_ef10_legacy_interrupt,
6752 	.tx_probe = efx_ef10_tx_probe,
6753 	.tx_init = efx_ef10_tx_init,
6754 	.tx_remove = efx_ef10_tx_remove,
6755 	.tx_write = efx_ef10_tx_write,
6756 	.tx_limit_len = efx_ef10_tx_limit_len,
6757 	.rx_push_rss_config = efx_ef10_pf_rx_push_rss_config,
6758 	.rx_pull_rss_config = efx_ef10_rx_pull_rss_config,
6759 	.rx_push_rss_context_config = efx_ef10_rx_push_rss_context_config,
6760 	.rx_pull_rss_context_config = efx_ef10_rx_pull_rss_context_config,
6761 	.rx_restore_rss_contexts = efx_ef10_rx_restore_rss_contexts,
6762 	.rx_probe = efx_ef10_rx_probe,
6763 	.rx_init = efx_ef10_rx_init,
6764 	.rx_remove = efx_ef10_rx_remove,
6765 	.rx_write = efx_ef10_rx_write,
6766 	.rx_defer_refill = efx_ef10_rx_defer_refill,
6767 	.ev_probe = efx_ef10_ev_probe,
6768 	.ev_init = efx_ef10_ev_init,
6769 	.ev_fini = efx_ef10_ev_fini,
6770 	.ev_remove = efx_ef10_ev_remove,
6771 	.ev_process = efx_ef10_ev_process,
6772 	.ev_read_ack = efx_ef10_ev_read_ack,
6773 	.ev_test_generate = efx_ef10_ev_test_generate,
6774 	.filter_table_probe = efx_ef10_filter_table_probe,
6775 	.filter_table_restore = efx_ef10_filter_table_restore,
6776 	.filter_table_remove = efx_ef10_filter_table_remove,
6777 	.filter_update_rx_scatter = efx_ef10_filter_update_rx_scatter,
6778 	.filter_insert = efx_ef10_filter_insert,
6779 	.filter_remove_safe = efx_ef10_filter_remove_safe,
6780 	.filter_get_safe = efx_ef10_filter_get_safe,
6781 	.filter_clear_rx = efx_ef10_filter_clear_rx,
6782 	.filter_count_rx_used = efx_ef10_filter_count_rx_used,
6783 	.filter_get_rx_id_limit = efx_ef10_filter_get_rx_id_limit,
6784 	.filter_get_rx_ids = efx_ef10_filter_get_rx_ids,
6785 #ifdef CONFIG_RFS_ACCEL
6786 	.filter_rfs_expire_one = efx_ef10_filter_rfs_expire_one,
6787 #endif
6788 #ifdef CONFIG_SFC_MTD
6789 	.mtd_probe = efx_ef10_mtd_probe,
6790 	.mtd_rename = efx_mcdi_mtd_rename,
6791 	.mtd_read = efx_mcdi_mtd_read,
6792 	.mtd_erase = efx_mcdi_mtd_erase,
6793 	.mtd_write = efx_mcdi_mtd_write,
6794 	.mtd_sync = efx_mcdi_mtd_sync,
6795 #endif
6796 	.ptp_write_host_time = efx_ef10_ptp_write_host_time,
6797 	.ptp_set_ts_sync_events = efx_ef10_ptp_set_ts_sync_events,
6798 	.ptp_set_ts_config = efx_ef10_ptp_set_ts_config,
6799 	.vlan_rx_add_vid = efx_ef10_vlan_rx_add_vid,
6800 	.vlan_rx_kill_vid = efx_ef10_vlan_rx_kill_vid,
6801 	.udp_tnl_push_ports = efx_ef10_udp_tnl_push_ports,
6802 	.udp_tnl_add_port = efx_ef10_udp_tnl_add_port,
6803 	.udp_tnl_has_port = efx_ef10_udp_tnl_has_port,
6804 	.udp_tnl_del_port = efx_ef10_udp_tnl_del_port,
6805 #ifdef CONFIG_SFC_SRIOV
6806 	.sriov_configure = efx_ef10_sriov_configure,
6807 	.sriov_init = efx_ef10_sriov_init,
6808 	.sriov_fini = efx_ef10_sriov_fini,
6809 	.sriov_wanted = efx_ef10_sriov_wanted,
6810 	.sriov_reset = efx_ef10_sriov_reset,
6811 	.sriov_flr = efx_ef10_sriov_flr,
6812 	.sriov_set_vf_mac = efx_ef10_sriov_set_vf_mac,
6813 	.sriov_set_vf_vlan = efx_ef10_sriov_set_vf_vlan,
6814 	.sriov_set_vf_spoofchk = efx_ef10_sriov_set_vf_spoofchk,
6815 	.sriov_get_vf_config = efx_ef10_sriov_get_vf_config,
6816 	.sriov_set_vf_link_state = efx_ef10_sriov_set_vf_link_state,
6817 	.vswitching_probe = efx_ef10_vswitching_probe_pf,
6818 	.vswitching_restore = efx_ef10_vswitching_restore_pf,
6819 	.vswitching_remove = efx_ef10_vswitching_remove_pf,
6820 #endif
6821 	.get_mac_address = efx_ef10_get_mac_address_pf,
6822 	.set_mac_address = efx_ef10_set_mac_address,
6823 	.tso_versions = efx_ef10_tso_versions,
6824 
6825 	.get_phys_port_id = efx_ef10_get_phys_port_id,
6826 	.revision = EFX_REV_HUNT_A0,
6827 	.max_dma_mask = DMA_BIT_MASK(ESF_DZ_TX_KER_BUF_ADDR_WIDTH),
6828 	.rx_prefix_size = ES_DZ_RX_PREFIX_SIZE,
6829 	.rx_hash_offset = ES_DZ_RX_PREFIX_HASH_OFST,
6830 	.rx_ts_offset = ES_DZ_RX_PREFIX_TSTAMP_OFST,
6831 	.can_rx_scatter = true,
6832 	.always_rx_scatter = true,
6833 	.option_descriptors = true,
6834 	.min_interrupt_mode = EFX_INT_MODE_LEGACY,
6835 	.max_interrupt_mode = EFX_INT_MODE_MSIX,
6836 	.timer_period_max = 1 << ERF_DD_EVQ_IND_TIMER_VAL_WIDTH,
6837 	.offload_features = EF10_OFFLOAD_FEATURES,
6838 	.mcdi_max_ver = 2,
6839 	.max_rx_ip_filters = HUNT_FILTER_TBL_ROWS,
6840 	.hwtstamp_filters = 1 << HWTSTAMP_FILTER_NONE |
6841 			    1 << HWTSTAMP_FILTER_ALL,
6842 	.rx_hash_key_size = 40,
6843 };
6844