xref: /linux/drivers/net/ethernet/sfc/ef10.c (revision 0883c2c06fb5bcf5b9e008270827e63c09a88c1e)
1 /****************************************************************************
2  * Driver for Solarflare network controllers and boards
3  * Copyright 2012-2013 Solarflare Communications Inc.
4  *
5  * This program is free software; you can redistribute it and/or modify it
6  * under the terms of the GNU General Public License version 2 as published
7  * by the Free Software Foundation, incorporated herein by reference.
8  */
9 
10 #include "net_driver.h"
11 #include "ef10_regs.h"
12 #include "io.h"
13 #include "mcdi.h"
14 #include "mcdi_pcol.h"
15 #include "nic.h"
16 #include "workarounds.h"
17 #include "selftest.h"
18 #include "ef10_sriov.h"
19 #include <linux/in.h>
20 #include <linux/jhash.h>
21 #include <linux/wait.h>
22 #include <linux/workqueue.h>
23 
24 /* Hardware control for EF10 architecture including 'Huntington'. */
25 
26 #define EFX_EF10_DRVGEN_EV		7
27 enum {
28 	EFX_EF10_TEST = 1,
29 	EFX_EF10_REFILL,
30 };
31 
32 /* The reserved RSS context value */
33 #define EFX_EF10_RSS_CONTEXT_INVALID	0xffffffff
34 /* The maximum size of a shared RSS context */
35 /* TODO: this should really be from the mcdi protocol export */
36 #define EFX_EF10_MAX_SHARED_RSS_CONTEXT_SIZE 64UL
37 
38 /* The filter table(s) are managed by firmware and we have write-only
39  * access.  When removing filters we must identify them to the
40  * firmware by a 64-bit handle, but this is too wide for Linux kernel
41  * interfaces (32-bit for RX NFC, 16-bit for RFS).  Also, we need to
42  * be able to tell in advance whether a requested insertion will
43  * replace an existing filter.  Therefore we maintain a software hash
44  * table, which should be at least as large as the hardware hash
45  * table.
46  *
47  * Huntington has a single 8K filter table shared between all filter
48  * types and both ports.
49  */
50 #define HUNT_FILTER_TBL_ROWS 8192
51 
52 #define EFX_EF10_FILTER_ID_INVALID 0xffff
53 struct efx_ef10_dev_addr {
54 	u8 addr[ETH_ALEN];
55 	u16 id;
56 };
57 
58 struct efx_ef10_filter_table {
59 /* The RX match field masks supported by this fw & hw, in order of priority */
60 	enum efx_filter_match_flags rx_match_flags[
61 		MC_CMD_GET_PARSER_DISP_INFO_OUT_SUPPORTED_MATCHES_MAXNUM];
62 	unsigned int rx_match_count;
63 
64 	struct {
65 		unsigned long spec;	/* pointer to spec plus flag bits */
66 /* BUSY flag indicates that an update is in progress.  AUTO_OLD is
67  * used to mark and sweep MAC filters for the device address lists.
68  */
69 #define EFX_EF10_FILTER_FLAG_BUSY	1UL
70 #define EFX_EF10_FILTER_FLAG_AUTO_OLD	2UL
71 #define EFX_EF10_FILTER_FLAGS		3UL
72 		u64 handle;		/* firmware handle */
73 	} *entry;
74 	wait_queue_head_t waitq;
75 /* Shadow of net_device address lists, guarded by mac_lock */
76 #define EFX_EF10_FILTER_DEV_UC_MAX	32
77 #define EFX_EF10_FILTER_DEV_MC_MAX	256
78 	struct efx_ef10_dev_addr dev_uc_list[EFX_EF10_FILTER_DEV_UC_MAX];
79 	struct efx_ef10_dev_addr dev_mc_list[EFX_EF10_FILTER_DEV_MC_MAX];
80 	int dev_uc_count;
81 	int dev_mc_count;
82 /* Indices (like efx_ef10_dev_addr.id) for promisc/allmulti filters */
83 	u16 ucdef_id;
84 	u16 bcast_id;
85 	u16 mcdef_id;
86 };
87 
88 /* An arbitrary search limit for the software hash table */
89 #define EFX_EF10_FILTER_SEARCH_LIMIT 200
90 
91 static void efx_ef10_rx_free_indir_table(struct efx_nic *efx);
92 static void efx_ef10_filter_table_remove(struct efx_nic *efx);
93 
94 static int efx_ef10_get_warm_boot_count(struct efx_nic *efx)
95 {
96 	efx_dword_t reg;
97 
98 	efx_readd(efx, &reg, ER_DZ_BIU_MC_SFT_STATUS);
99 	return EFX_DWORD_FIELD(reg, EFX_WORD_1) == 0xb007 ?
100 		EFX_DWORD_FIELD(reg, EFX_WORD_0) : -EIO;
101 }
102 
103 static unsigned int efx_ef10_mem_map_size(struct efx_nic *efx)
104 {
105 	int bar;
106 
107 	bar = efx->type->mem_bar;
108 	return resource_size(&efx->pci_dev->resource[bar]);
109 }
110 
111 static bool efx_ef10_is_vf(struct efx_nic *efx)
112 {
113 	return efx->type->is_vf;
114 }
115 
116 static int efx_ef10_get_pf_index(struct efx_nic *efx)
117 {
118 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_FUNCTION_INFO_OUT_LEN);
119 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
120 	size_t outlen;
121 	int rc;
122 
123 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_FUNCTION_INFO, NULL, 0, outbuf,
124 			  sizeof(outbuf), &outlen);
125 	if (rc)
126 		return rc;
127 	if (outlen < sizeof(outbuf))
128 		return -EIO;
129 
130 	nic_data->pf_index = MCDI_DWORD(outbuf, GET_FUNCTION_INFO_OUT_PF);
131 	return 0;
132 }
133 
134 #ifdef CONFIG_SFC_SRIOV
135 static int efx_ef10_get_vf_index(struct efx_nic *efx)
136 {
137 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_FUNCTION_INFO_OUT_LEN);
138 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
139 	size_t outlen;
140 	int rc;
141 
142 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_FUNCTION_INFO, NULL, 0, outbuf,
143 			  sizeof(outbuf), &outlen);
144 	if (rc)
145 		return rc;
146 	if (outlen < sizeof(outbuf))
147 		return -EIO;
148 
149 	nic_data->vf_index = MCDI_DWORD(outbuf, GET_FUNCTION_INFO_OUT_VF);
150 	return 0;
151 }
152 #endif
153 
154 static int efx_ef10_init_datapath_caps(struct efx_nic *efx)
155 {
156 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_CAPABILITIES_OUT_LEN);
157 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
158 	size_t outlen;
159 	int rc;
160 
161 	BUILD_BUG_ON(MC_CMD_GET_CAPABILITIES_IN_LEN != 0);
162 
163 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_CAPABILITIES, NULL, 0,
164 			  outbuf, sizeof(outbuf), &outlen);
165 	if (rc)
166 		return rc;
167 	if (outlen < sizeof(outbuf)) {
168 		netif_err(efx, drv, efx->net_dev,
169 			  "unable to read datapath firmware capabilities\n");
170 		return -EIO;
171 	}
172 
173 	nic_data->datapath_caps =
174 		MCDI_DWORD(outbuf, GET_CAPABILITIES_OUT_FLAGS1);
175 
176 	/* record the DPCPU firmware IDs to determine VEB vswitching support.
177 	 */
178 	nic_data->rx_dpcpu_fw_id =
179 		MCDI_WORD(outbuf, GET_CAPABILITIES_OUT_RX_DPCPU_FW_ID);
180 	nic_data->tx_dpcpu_fw_id =
181 		MCDI_WORD(outbuf, GET_CAPABILITIES_OUT_TX_DPCPU_FW_ID);
182 
183 	if (!(nic_data->datapath_caps &
184 	      (1 << MC_CMD_GET_CAPABILITIES_OUT_RX_PREFIX_LEN_14_LBN))) {
185 		netif_err(efx, probe, efx->net_dev,
186 			  "current firmware does not support an RX prefix\n");
187 		return -ENODEV;
188 	}
189 
190 	return 0;
191 }
192 
193 static int efx_ef10_get_sysclk_freq(struct efx_nic *efx)
194 {
195 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_CLOCK_OUT_LEN);
196 	int rc;
197 
198 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_CLOCK, NULL, 0,
199 			  outbuf, sizeof(outbuf), NULL);
200 	if (rc)
201 		return rc;
202 	rc = MCDI_DWORD(outbuf, GET_CLOCK_OUT_SYS_FREQ);
203 	return rc > 0 ? rc : -ERANGE;
204 }
205 
206 static int efx_ef10_get_mac_address_pf(struct efx_nic *efx, u8 *mac_address)
207 {
208 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_MAC_ADDRESSES_OUT_LEN);
209 	size_t outlen;
210 	int rc;
211 
212 	BUILD_BUG_ON(MC_CMD_GET_MAC_ADDRESSES_IN_LEN != 0);
213 
214 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_MAC_ADDRESSES, NULL, 0,
215 			  outbuf, sizeof(outbuf), &outlen);
216 	if (rc)
217 		return rc;
218 	if (outlen < MC_CMD_GET_MAC_ADDRESSES_OUT_LEN)
219 		return -EIO;
220 
221 	ether_addr_copy(mac_address,
222 			MCDI_PTR(outbuf, GET_MAC_ADDRESSES_OUT_MAC_ADDR_BASE));
223 	return 0;
224 }
225 
226 static int efx_ef10_get_mac_address_vf(struct efx_nic *efx, u8 *mac_address)
227 {
228 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VPORT_GET_MAC_ADDRESSES_IN_LEN);
229 	MCDI_DECLARE_BUF(outbuf, MC_CMD_VPORT_GET_MAC_ADDRESSES_OUT_LENMAX);
230 	size_t outlen;
231 	int num_addrs, rc;
232 
233 	MCDI_SET_DWORD(inbuf, VPORT_GET_MAC_ADDRESSES_IN_VPORT_ID,
234 		       EVB_PORT_ID_ASSIGNED);
235 	rc = efx_mcdi_rpc(efx, MC_CMD_VPORT_GET_MAC_ADDRESSES, inbuf,
236 			  sizeof(inbuf), outbuf, sizeof(outbuf), &outlen);
237 
238 	if (rc)
239 		return rc;
240 	if (outlen < MC_CMD_VPORT_GET_MAC_ADDRESSES_OUT_LENMIN)
241 		return -EIO;
242 
243 	num_addrs = MCDI_DWORD(outbuf,
244 			       VPORT_GET_MAC_ADDRESSES_OUT_MACADDR_COUNT);
245 
246 	WARN_ON(num_addrs != 1);
247 
248 	ether_addr_copy(mac_address,
249 			MCDI_PTR(outbuf, VPORT_GET_MAC_ADDRESSES_OUT_MACADDR));
250 
251 	return 0;
252 }
253 
254 static ssize_t efx_ef10_show_link_control_flag(struct device *dev,
255 					       struct device_attribute *attr,
256 					       char *buf)
257 {
258 	struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));
259 
260 	return sprintf(buf, "%d\n",
261 		       ((efx->mcdi->fn_flags) &
262 			(1 << MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_LINKCTRL))
263 		       ? 1 : 0);
264 }
265 
266 static ssize_t efx_ef10_show_primary_flag(struct device *dev,
267 					  struct device_attribute *attr,
268 					  char *buf)
269 {
270 	struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));
271 
272 	return sprintf(buf, "%d\n",
273 		       ((efx->mcdi->fn_flags) &
274 			(1 << MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_PRIMARY))
275 		       ? 1 : 0);
276 }
277 
278 static DEVICE_ATTR(link_control_flag, 0444, efx_ef10_show_link_control_flag,
279 		   NULL);
280 static DEVICE_ATTR(primary_flag, 0444, efx_ef10_show_primary_flag, NULL);
281 
282 static int efx_ef10_probe(struct efx_nic *efx)
283 {
284 	struct efx_ef10_nic_data *nic_data;
285 	struct net_device *net_dev = efx->net_dev;
286 	int i, rc;
287 
288 	/* We can have one VI for each 8K region.  However, until we
289 	 * use TX option descriptors we need two TX queues per channel.
290 	 */
291 	efx->max_channels = min_t(unsigned int,
292 				  EFX_MAX_CHANNELS,
293 				  efx_ef10_mem_map_size(efx) /
294 				  (EFX_VI_PAGE_SIZE * EFX_TXQ_TYPES));
295 	efx->max_tx_channels = efx->max_channels;
296 	if (WARN_ON(efx->max_channels == 0))
297 		return -EIO;
298 
299 	nic_data = kzalloc(sizeof(*nic_data), GFP_KERNEL);
300 	if (!nic_data)
301 		return -ENOMEM;
302 	efx->nic_data = nic_data;
303 
304 	/* we assume later that we can copy from this buffer in dwords */
305 	BUILD_BUG_ON(MCDI_CTL_SDU_LEN_MAX_V2 % 4);
306 
307 	rc = efx_nic_alloc_buffer(efx, &nic_data->mcdi_buf,
308 				  8 + MCDI_CTL_SDU_LEN_MAX_V2, GFP_KERNEL);
309 	if (rc)
310 		goto fail1;
311 
312 	/* Get the MC's warm boot count.  In case it's rebooting right
313 	 * now, be prepared to retry.
314 	 */
315 	i = 0;
316 	for (;;) {
317 		rc = efx_ef10_get_warm_boot_count(efx);
318 		if (rc >= 0)
319 			break;
320 		if (++i == 5)
321 			goto fail2;
322 		ssleep(1);
323 	}
324 	nic_data->warm_boot_count = rc;
325 
326 	nic_data->rx_rss_context = EFX_EF10_RSS_CONTEXT_INVALID;
327 
328 	nic_data->vport_id = EVB_PORT_ID_ASSIGNED;
329 
330 	/* In case we're recovering from a crash (kexec), we want to
331 	 * cancel any outstanding request by the previous user of this
332 	 * function.  We send a special message using the least
333 	 * significant bits of the 'high' (doorbell) register.
334 	 */
335 	_efx_writed(efx, cpu_to_le32(1), ER_DZ_MC_DB_HWRD);
336 
337 	rc = efx_mcdi_init(efx);
338 	if (rc)
339 		goto fail2;
340 
341 	/* Reset (most) configuration for this function */
342 	rc = efx_mcdi_reset(efx, RESET_TYPE_ALL);
343 	if (rc)
344 		goto fail3;
345 
346 	/* Enable event logging */
347 	rc = efx_mcdi_log_ctrl(efx, true, false, 0);
348 	if (rc)
349 		goto fail3;
350 
351 	rc = device_create_file(&efx->pci_dev->dev,
352 				&dev_attr_link_control_flag);
353 	if (rc)
354 		goto fail3;
355 
356 	rc = device_create_file(&efx->pci_dev->dev, &dev_attr_primary_flag);
357 	if (rc)
358 		goto fail4;
359 
360 	rc = efx_ef10_get_pf_index(efx);
361 	if (rc)
362 		goto fail5;
363 
364 	rc = efx_ef10_init_datapath_caps(efx);
365 	if (rc < 0)
366 		goto fail5;
367 
368 	efx->rx_packet_len_offset =
369 		ES_DZ_RX_PREFIX_PKTLEN_OFST - ES_DZ_RX_PREFIX_SIZE;
370 
371 	rc = efx_mcdi_port_get_number(efx);
372 	if (rc < 0)
373 		goto fail5;
374 	efx->port_num = rc;
375 	net_dev->dev_port = rc;
376 
377 	rc = efx->type->get_mac_address(efx, efx->net_dev->perm_addr);
378 	if (rc)
379 		goto fail5;
380 
381 	rc = efx_ef10_get_sysclk_freq(efx);
382 	if (rc < 0)
383 		goto fail5;
384 	efx->timer_quantum_ns = 1536000 / rc; /* 1536 cycles */
385 
386 	/* Check whether firmware supports bug 35388 workaround.
387 	 * First try to enable it, then if we get EPERM, just
388 	 * ask if it's already enabled
389 	 */
390 	rc = efx_mcdi_set_workaround(efx, MC_CMD_WORKAROUND_BUG35388, true, NULL);
391 	if (rc == 0) {
392 		nic_data->workaround_35388 = true;
393 	} else if (rc == -EPERM) {
394 		unsigned int enabled;
395 
396 		rc = efx_mcdi_get_workarounds(efx, NULL, &enabled);
397 		if (rc)
398 			goto fail3;
399 		nic_data->workaround_35388 = enabled &
400 			MC_CMD_GET_WORKAROUNDS_OUT_BUG35388;
401 	} else if (rc != -ENOSYS && rc != -ENOENT) {
402 		goto fail5;
403 	}
404 	netif_dbg(efx, probe, efx->net_dev,
405 		  "workaround for bug 35388 is %sabled\n",
406 		  nic_data->workaround_35388 ? "en" : "dis");
407 
408 	rc = efx_mcdi_mon_probe(efx);
409 	if (rc && rc != -EPERM)
410 		goto fail5;
411 
412 	efx_ptp_probe(efx, NULL);
413 
414 #ifdef CONFIG_SFC_SRIOV
415 	if ((efx->pci_dev->physfn) && (!efx->pci_dev->is_physfn)) {
416 		struct pci_dev *pci_dev_pf = efx->pci_dev->physfn;
417 		struct efx_nic *efx_pf = pci_get_drvdata(pci_dev_pf);
418 
419 		efx_pf->type->get_mac_address(efx_pf, nic_data->port_id);
420 	} else
421 #endif
422 		ether_addr_copy(nic_data->port_id, efx->net_dev->perm_addr);
423 
424 	return 0;
425 
426 fail5:
427 	device_remove_file(&efx->pci_dev->dev, &dev_attr_primary_flag);
428 fail4:
429 	device_remove_file(&efx->pci_dev->dev, &dev_attr_link_control_flag);
430 fail3:
431 	efx_mcdi_fini(efx);
432 fail2:
433 	efx_nic_free_buffer(efx, &nic_data->mcdi_buf);
434 fail1:
435 	kfree(nic_data);
436 	efx->nic_data = NULL;
437 	return rc;
438 }
439 
440 static int efx_ef10_free_vis(struct efx_nic *efx)
441 {
442 	MCDI_DECLARE_BUF_ERR(outbuf);
443 	size_t outlen;
444 	int rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FREE_VIS, NULL, 0,
445 				    outbuf, sizeof(outbuf), &outlen);
446 
447 	/* -EALREADY means nothing to free, so ignore */
448 	if (rc == -EALREADY)
449 		rc = 0;
450 	if (rc)
451 		efx_mcdi_display_error(efx, MC_CMD_FREE_VIS, 0, outbuf, outlen,
452 				       rc);
453 	return rc;
454 }
455 
456 #ifdef EFX_USE_PIO
457 
458 static void efx_ef10_free_piobufs(struct efx_nic *efx)
459 {
460 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
461 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FREE_PIOBUF_IN_LEN);
462 	unsigned int i;
463 	int rc;
464 
465 	BUILD_BUG_ON(MC_CMD_FREE_PIOBUF_OUT_LEN != 0);
466 
467 	for (i = 0; i < nic_data->n_piobufs; i++) {
468 		MCDI_SET_DWORD(inbuf, FREE_PIOBUF_IN_PIOBUF_HANDLE,
469 			       nic_data->piobuf_handle[i]);
470 		rc = efx_mcdi_rpc(efx, MC_CMD_FREE_PIOBUF, inbuf, sizeof(inbuf),
471 				  NULL, 0, NULL);
472 		WARN_ON(rc);
473 	}
474 
475 	nic_data->n_piobufs = 0;
476 }
477 
478 static int efx_ef10_alloc_piobufs(struct efx_nic *efx, unsigned int n)
479 {
480 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
481 	MCDI_DECLARE_BUF(outbuf, MC_CMD_ALLOC_PIOBUF_OUT_LEN);
482 	unsigned int i;
483 	size_t outlen;
484 	int rc = 0;
485 
486 	BUILD_BUG_ON(MC_CMD_ALLOC_PIOBUF_IN_LEN != 0);
487 
488 	for (i = 0; i < n; i++) {
489 		rc = efx_mcdi_rpc_quiet(efx, MC_CMD_ALLOC_PIOBUF, NULL, 0,
490 					outbuf, sizeof(outbuf), &outlen);
491 		if (rc) {
492 			/* Don't display the MC error if we didn't have space
493 			 * for a VF.
494 			 */
495 			if (!(efx_ef10_is_vf(efx) && rc == -ENOSPC))
496 				efx_mcdi_display_error(efx, MC_CMD_ALLOC_PIOBUF,
497 						       0, outbuf, outlen, rc);
498 			break;
499 		}
500 		if (outlen < MC_CMD_ALLOC_PIOBUF_OUT_LEN) {
501 			rc = -EIO;
502 			break;
503 		}
504 		nic_data->piobuf_handle[i] =
505 			MCDI_DWORD(outbuf, ALLOC_PIOBUF_OUT_PIOBUF_HANDLE);
506 		netif_dbg(efx, probe, efx->net_dev,
507 			  "allocated PIO buffer %u handle %x\n", i,
508 			  nic_data->piobuf_handle[i]);
509 	}
510 
511 	nic_data->n_piobufs = i;
512 	if (rc)
513 		efx_ef10_free_piobufs(efx);
514 	return rc;
515 }
516 
517 static int efx_ef10_link_piobufs(struct efx_nic *efx)
518 {
519 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
520 	_MCDI_DECLARE_BUF(inbuf,
521 			  max(MC_CMD_LINK_PIOBUF_IN_LEN,
522 			      MC_CMD_UNLINK_PIOBUF_IN_LEN));
523 	struct efx_channel *channel;
524 	struct efx_tx_queue *tx_queue;
525 	unsigned int offset, index;
526 	int rc;
527 
528 	BUILD_BUG_ON(MC_CMD_LINK_PIOBUF_OUT_LEN != 0);
529 	BUILD_BUG_ON(MC_CMD_UNLINK_PIOBUF_OUT_LEN != 0);
530 
531 	memset(inbuf, 0, sizeof(inbuf));
532 
533 	/* Link a buffer to each VI in the write-combining mapping */
534 	for (index = 0; index < nic_data->n_piobufs; ++index) {
535 		MCDI_SET_DWORD(inbuf, LINK_PIOBUF_IN_PIOBUF_HANDLE,
536 			       nic_data->piobuf_handle[index]);
537 		MCDI_SET_DWORD(inbuf, LINK_PIOBUF_IN_TXQ_INSTANCE,
538 			       nic_data->pio_write_vi_base + index);
539 		rc = efx_mcdi_rpc(efx, MC_CMD_LINK_PIOBUF,
540 				  inbuf, MC_CMD_LINK_PIOBUF_IN_LEN,
541 				  NULL, 0, NULL);
542 		if (rc) {
543 			netif_err(efx, drv, efx->net_dev,
544 				  "failed to link VI %u to PIO buffer %u (%d)\n",
545 				  nic_data->pio_write_vi_base + index, index,
546 				  rc);
547 			goto fail;
548 		}
549 		netif_dbg(efx, probe, efx->net_dev,
550 			  "linked VI %u to PIO buffer %u\n",
551 			  nic_data->pio_write_vi_base + index, index);
552 	}
553 
554 	/* Link a buffer to each TX queue */
555 	efx_for_each_channel(channel, efx) {
556 		efx_for_each_channel_tx_queue(tx_queue, channel) {
557 			/* We assign the PIO buffers to queues in
558 			 * reverse order to allow for the following
559 			 * special case.
560 			 */
561 			offset = ((efx->tx_channel_offset + efx->n_tx_channels -
562 				   tx_queue->channel->channel - 1) *
563 				  efx_piobuf_size);
564 			index = offset / ER_DZ_TX_PIOBUF_SIZE;
565 			offset = offset % ER_DZ_TX_PIOBUF_SIZE;
566 
567 			/* When the host page size is 4K, the first
568 			 * host page in the WC mapping may be within
569 			 * the same VI page as the last TX queue.  We
570 			 * can only link one buffer to each VI.
571 			 */
572 			if (tx_queue->queue == nic_data->pio_write_vi_base) {
573 				BUG_ON(index != 0);
574 				rc = 0;
575 			} else {
576 				MCDI_SET_DWORD(inbuf,
577 					       LINK_PIOBUF_IN_PIOBUF_HANDLE,
578 					       nic_data->piobuf_handle[index]);
579 				MCDI_SET_DWORD(inbuf,
580 					       LINK_PIOBUF_IN_TXQ_INSTANCE,
581 					       tx_queue->queue);
582 				rc = efx_mcdi_rpc(efx, MC_CMD_LINK_PIOBUF,
583 						  inbuf, MC_CMD_LINK_PIOBUF_IN_LEN,
584 						  NULL, 0, NULL);
585 			}
586 
587 			if (rc) {
588 				/* This is non-fatal; the TX path just
589 				 * won't use PIO for this queue
590 				 */
591 				netif_err(efx, drv, efx->net_dev,
592 					  "failed to link VI %u to PIO buffer %u (%d)\n",
593 					  tx_queue->queue, index, rc);
594 				tx_queue->piobuf = NULL;
595 			} else {
596 				tx_queue->piobuf =
597 					nic_data->pio_write_base +
598 					index * EFX_VI_PAGE_SIZE + offset;
599 				tx_queue->piobuf_offset = offset;
600 				netif_dbg(efx, probe, efx->net_dev,
601 					  "linked VI %u to PIO buffer %u offset %x addr %p\n",
602 					  tx_queue->queue, index,
603 					  tx_queue->piobuf_offset,
604 					  tx_queue->piobuf);
605 			}
606 		}
607 	}
608 
609 	return 0;
610 
611 fail:
612 	while (index--) {
613 		MCDI_SET_DWORD(inbuf, UNLINK_PIOBUF_IN_TXQ_INSTANCE,
614 			       nic_data->pio_write_vi_base + index);
615 		efx_mcdi_rpc(efx, MC_CMD_UNLINK_PIOBUF,
616 			     inbuf, MC_CMD_UNLINK_PIOBUF_IN_LEN,
617 			     NULL, 0, NULL);
618 	}
619 	return rc;
620 }
621 
622 static void efx_ef10_forget_old_piobufs(struct efx_nic *efx)
623 {
624 	struct efx_channel *channel;
625 	struct efx_tx_queue *tx_queue;
626 
627 	/* All our existing PIO buffers went away */
628 	efx_for_each_channel(channel, efx)
629 		efx_for_each_channel_tx_queue(tx_queue, channel)
630 			tx_queue->piobuf = NULL;
631 }
632 
633 #else /* !EFX_USE_PIO */
634 
635 static int efx_ef10_alloc_piobufs(struct efx_nic *efx, unsigned int n)
636 {
637 	return n == 0 ? 0 : -ENOBUFS;
638 }
639 
640 static int efx_ef10_link_piobufs(struct efx_nic *efx)
641 {
642 	return 0;
643 }
644 
645 static void efx_ef10_free_piobufs(struct efx_nic *efx)
646 {
647 }
648 
649 static void efx_ef10_forget_old_piobufs(struct efx_nic *efx)
650 {
651 }
652 
653 #endif /* EFX_USE_PIO */
654 
655 static void efx_ef10_remove(struct efx_nic *efx)
656 {
657 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
658 	int rc;
659 
660 #ifdef CONFIG_SFC_SRIOV
661 	struct efx_ef10_nic_data *nic_data_pf;
662 	struct pci_dev *pci_dev_pf;
663 	struct efx_nic *efx_pf;
664 	struct ef10_vf *vf;
665 
666 	if (efx->pci_dev->is_virtfn) {
667 		pci_dev_pf = efx->pci_dev->physfn;
668 		if (pci_dev_pf) {
669 			efx_pf = pci_get_drvdata(pci_dev_pf);
670 			nic_data_pf = efx_pf->nic_data;
671 			vf = nic_data_pf->vf + nic_data->vf_index;
672 			vf->efx = NULL;
673 		} else
674 			netif_info(efx, drv, efx->net_dev,
675 				   "Could not get the PF id from VF\n");
676 	}
677 #endif
678 
679 	efx_ptp_remove(efx);
680 
681 	efx_mcdi_mon_remove(efx);
682 
683 	efx_ef10_rx_free_indir_table(efx);
684 
685 	if (nic_data->wc_membase)
686 		iounmap(nic_data->wc_membase);
687 
688 	rc = efx_ef10_free_vis(efx);
689 	WARN_ON(rc != 0);
690 
691 	if (!nic_data->must_restore_piobufs)
692 		efx_ef10_free_piobufs(efx);
693 
694 	device_remove_file(&efx->pci_dev->dev, &dev_attr_primary_flag);
695 	device_remove_file(&efx->pci_dev->dev, &dev_attr_link_control_flag);
696 
697 	efx_mcdi_fini(efx);
698 	efx_nic_free_buffer(efx, &nic_data->mcdi_buf);
699 	kfree(nic_data);
700 }
701 
702 static int efx_ef10_probe_pf(struct efx_nic *efx)
703 {
704 	return efx_ef10_probe(efx);
705 }
706 
707 int efx_ef10_vadaptor_alloc(struct efx_nic *efx, unsigned int port_id)
708 {
709 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VADAPTOR_ALLOC_IN_LEN);
710 
711 	MCDI_SET_DWORD(inbuf, VADAPTOR_ALLOC_IN_UPSTREAM_PORT_ID, port_id);
712 	return efx_mcdi_rpc(efx, MC_CMD_VADAPTOR_ALLOC, inbuf, sizeof(inbuf),
713 			    NULL, 0, NULL);
714 }
715 
716 int efx_ef10_vadaptor_free(struct efx_nic *efx, unsigned int port_id)
717 {
718 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VADAPTOR_FREE_IN_LEN);
719 
720 	MCDI_SET_DWORD(inbuf, VADAPTOR_FREE_IN_UPSTREAM_PORT_ID, port_id);
721 	return efx_mcdi_rpc(efx, MC_CMD_VADAPTOR_FREE, inbuf, sizeof(inbuf),
722 			    NULL, 0, NULL);
723 }
724 
725 int efx_ef10_vport_add_mac(struct efx_nic *efx,
726 			   unsigned int port_id, u8 *mac)
727 {
728 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VPORT_ADD_MAC_ADDRESS_IN_LEN);
729 
730 	MCDI_SET_DWORD(inbuf, VPORT_ADD_MAC_ADDRESS_IN_VPORT_ID, port_id);
731 	ether_addr_copy(MCDI_PTR(inbuf, VPORT_ADD_MAC_ADDRESS_IN_MACADDR), mac);
732 
733 	return efx_mcdi_rpc(efx, MC_CMD_VPORT_ADD_MAC_ADDRESS, inbuf,
734 			    sizeof(inbuf), NULL, 0, NULL);
735 }
736 
737 int efx_ef10_vport_del_mac(struct efx_nic *efx,
738 			   unsigned int port_id, u8 *mac)
739 {
740 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VPORT_DEL_MAC_ADDRESS_IN_LEN);
741 
742 	MCDI_SET_DWORD(inbuf, VPORT_DEL_MAC_ADDRESS_IN_VPORT_ID, port_id);
743 	ether_addr_copy(MCDI_PTR(inbuf, VPORT_DEL_MAC_ADDRESS_IN_MACADDR), mac);
744 
745 	return efx_mcdi_rpc(efx, MC_CMD_VPORT_DEL_MAC_ADDRESS, inbuf,
746 			    sizeof(inbuf), NULL, 0, NULL);
747 }
748 
749 #ifdef CONFIG_SFC_SRIOV
750 static int efx_ef10_probe_vf(struct efx_nic *efx)
751 {
752 	int rc;
753 	struct pci_dev *pci_dev_pf;
754 
755 	/* If the parent PF has no VF data structure, it doesn't know about this
756 	 * VF so fail probe.  The VF needs to be re-created.  This can happen
757 	 * if the PF driver is unloaded while the VF is assigned to a guest.
758 	 */
759 	pci_dev_pf = efx->pci_dev->physfn;
760 	if (pci_dev_pf) {
761 		struct efx_nic *efx_pf = pci_get_drvdata(pci_dev_pf);
762 		struct efx_ef10_nic_data *nic_data_pf = efx_pf->nic_data;
763 
764 		if (!nic_data_pf->vf) {
765 			netif_info(efx, drv, efx->net_dev,
766 				   "The VF cannot link to its parent PF; "
767 				   "please destroy and re-create the VF\n");
768 			return -EBUSY;
769 		}
770 	}
771 
772 	rc = efx_ef10_probe(efx);
773 	if (rc)
774 		return rc;
775 
776 	rc = efx_ef10_get_vf_index(efx);
777 	if (rc)
778 		goto fail;
779 
780 	if (efx->pci_dev->is_virtfn) {
781 		if (efx->pci_dev->physfn) {
782 			struct efx_nic *efx_pf =
783 				pci_get_drvdata(efx->pci_dev->physfn);
784 			struct efx_ef10_nic_data *nic_data_p = efx_pf->nic_data;
785 			struct efx_ef10_nic_data *nic_data = efx->nic_data;
786 
787 			nic_data_p->vf[nic_data->vf_index].efx = efx;
788 			nic_data_p->vf[nic_data->vf_index].pci_dev =
789 				efx->pci_dev;
790 		} else
791 			netif_info(efx, drv, efx->net_dev,
792 				   "Could not get the PF id from VF\n");
793 	}
794 
795 	return 0;
796 
797 fail:
798 	efx_ef10_remove(efx);
799 	return rc;
800 }
801 #else
802 static int efx_ef10_probe_vf(struct efx_nic *efx __attribute__ ((unused)))
803 {
804 	return 0;
805 }
806 #endif
807 
808 static int efx_ef10_alloc_vis(struct efx_nic *efx,
809 			      unsigned int min_vis, unsigned int max_vis)
810 {
811 	MCDI_DECLARE_BUF(inbuf, MC_CMD_ALLOC_VIS_IN_LEN);
812 	MCDI_DECLARE_BUF(outbuf, MC_CMD_ALLOC_VIS_OUT_LEN);
813 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
814 	size_t outlen;
815 	int rc;
816 
817 	MCDI_SET_DWORD(inbuf, ALLOC_VIS_IN_MIN_VI_COUNT, min_vis);
818 	MCDI_SET_DWORD(inbuf, ALLOC_VIS_IN_MAX_VI_COUNT, max_vis);
819 	rc = efx_mcdi_rpc(efx, MC_CMD_ALLOC_VIS, inbuf, sizeof(inbuf),
820 			  outbuf, sizeof(outbuf), &outlen);
821 	if (rc != 0)
822 		return rc;
823 
824 	if (outlen < MC_CMD_ALLOC_VIS_OUT_LEN)
825 		return -EIO;
826 
827 	netif_dbg(efx, drv, efx->net_dev, "base VI is A0x%03x\n",
828 		  MCDI_DWORD(outbuf, ALLOC_VIS_OUT_VI_BASE));
829 
830 	nic_data->vi_base = MCDI_DWORD(outbuf, ALLOC_VIS_OUT_VI_BASE);
831 	nic_data->n_allocated_vis = MCDI_DWORD(outbuf, ALLOC_VIS_OUT_VI_COUNT);
832 	return 0;
833 }
834 
835 /* Note that the failure path of this function does not free
836  * resources, as this will be done by efx_ef10_remove().
837  */
838 static int efx_ef10_dimension_resources(struct efx_nic *efx)
839 {
840 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
841 	unsigned int uc_mem_map_size, wc_mem_map_size;
842 	unsigned int min_vis = max(EFX_TXQ_TYPES,
843 				   efx_separate_tx_channels ? 2 : 1);
844 	unsigned int channel_vis, pio_write_vi_base, max_vis;
845 	void __iomem *membase;
846 	int rc;
847 
848 	channel_vis = max(efx->n_channels, efx->n_tx_channels * EFX_TXQ_TYPES);
849 
850 #ifdef EFX_USE_PIO
851 	/* Try to allocate PIO buffers if wanted and if the full
852 	 * number of PIO buffers would be sufficient to allocate one
853 	 * copy-buffer per TX channel.  Failure is non-fatal, as there
854 	 * are only a small number of PIO buffers shared between all
855 	 * functions of the controller.
856 	 */
857 	if (efx_piobuf_size != 0 &&
858 	    ER_DZ_TX_PIOBUF_SIZE / efx_piobuf_size * EF10_TX_PIOBUF_COUNT >=
859 	    efx->n_tx_channels) {
860 		unsigned int n_piobufs =
861 			DIV_ROUND_UP(efx->n_tx_channels,
862 				     ER_DZ_TX_PIOBUF_SIZE / efx_piobuf_size);
863 
864 		rc = efx_ef10_alloc_piobufs(efx, n_piobufs);
865 		if (rc)
866 			netif_err(efx, probe, efx->net_dev,
867 				  "failed to allocate PIO buffers (%d)\n", rc);
868 		else
869 			netif_dbg(efx, probe, efx->net_dev,
870 				  "allocated %u PIO buffers\n", n_piobufs);
871 	}
872 #else
873 	nic_data->n_piobufs = 0;
874 #endif
875 
876 	/* PIO buffers should be mapped with write-combining enabled,
877 	 * and we want to make single UC and WC mappings rather than
878 	 * several of each (in fact that's the only option if host
879 	 * page size is >4K).  So we may allocate some extra VIs just
880 	 * for writing PIO buffers through.
881 	 *
882 	 * The UC mapping contains (channel_vis - 1) complete VIs and the
883 	 * first half of the next VI.  Then the WC mapping begins with
884 	 * the second half of this last VI.
885 	 */
886 	uc_mem_map_size = PAGE_ALIGN((channel_vis - 1) * EFX_VI_PAGE_SIZE +
887 				     ER_DZ_TX_PIOBUF);
888 	if (nic_data->n_piobufs) {
889 		/* pio_write_vi_base rounds down to give the number of complete
890 		 * VIs inside the UC mapping.
891 		 */
892 		pio_write_vi_base = uc_mem_map_size / EFX_VI_PAGE_SIZE;
893 		wc_mem_map_size = (PAGE_ALIGN((pio_write_vi_base +
894 					       nic_data->n_piobufs) *
895 					      EFX_VI_PAGE_SIZE) -
896 				   uc_mem_map_size);
897 		max_vis = pio_write_vi_base + nic_data->n_piobufs;
898 	} else {
899 		pio_write_vi_base = 0;
900 		wc_mem_map_size = 0;
901 		max_vis = channel_vis;
902 	}
903 
904 	/* In case the last attached driver failed to free VIs, do it now */
905 	rc = efx_ef10_free_vis(efx);
906 	if (rc != 0)
907 		return rc;
908 
909 	rc = efx_ef10_alloc_vis(efx, min_vis, max_vis);
910 	if (rc != 0)
911 		return rc;
912 
913 	if (nic_data->n_allocated_vis < channel_vis) {
914 		netif_info(efx, drv, efx->net_dev,
915 			   "Could not allocate enough VIs to satisfy RSS"
916 			   " requirements. Performance may not be optimal.\n");
917 		/* We didn't get the VIs to populate our channels.
918 		 * We could keep what we got but then we'd have more
919 		 * interrupts than we need.
920 		 * Instead calculate new max_channels and restart
921 		 */
922 		efx->max_channels = nic_data->n_allocated_vis;
923 		efx->max_tx_channels =
924 			nic_data->n_allocated_vis / EFX_TXQ_TYPES;
925 
926 		efx_ef10_free_vis(efx);
927 		return -EAGAIN;
928 	}
929 
930 	/* If we didn't get enough VIs to map all the PIO buffers, free the
931 	 * PIO buffers
932 	 */
933 	if (nic_data->n_piobufs &&
934 	    nic_data->n_allocated_vis <
935 	    pio_write_vi_base + nic_data->n_piobufs) {
936 		netif_dbg(efx, probe, efx->net_dev,
937 			  "%u VIs are not sufficient to map %u PIO buffers\n",
938 			  nic_data->n_allocated_vis, nic_data->n_piobufs);
939 		efx_ef10_free_piobufs(efx);
940 	}
941 
942 	/* Shrink the original UC mapping of the memory BAR */
943 	membase = ioremap_nocache(efx->membase_phys, uc_mem_map_size);
944 	if (!membase) {
945 		netif_err(efx, probe, efx->net_dev,
946 			  "could not shrink memory BAR to %x\n",
947 			  uc_mem_map_size);
948 		return -ENOMEM;
949 	}
950 	iounmap(efx->membase);
951 	efx->membase = membase;
952 
953 	/* Set up the WC mapping if needed */
954 	if (wc_mem_map_size) {
955 		nic_data->wc_membase = ioremap_wc(efx->membase_phys +
956 						  uc_mem_map_size,
957 						  wc_mem_map_size);
958 		if (!nic_data->wc_membase) {
959 			netif_err(efx, probe, efx->net_dev,
960 				  "could not allocate WC mapping of size %x\n",
961 				  wc_mem_map_size);
962 			return -ENOMEM;
963 		}
964 		nic_data->pio_write_vi_base = pio_write_vi_base;
965 		nic_data->pio_write_base =
966 			nic_data->wc_membase +
967 			(pio_write_vi_base * EFX_VI_PAGE_SIZE + ER_DZ_TX_PIOBUF -
968 			 uc_mem_map_size);
969 
970 		rc = efx_ef10_link_piobufs(efx);
971 		if (rc)
972 			efx_ef10_free_piobufs(efx);
973 	}
974 
975 	netif_dbg(efx, probe, efx->net_dev,
976 		  "memory BAR at %pa (virtual %p+%x UC, %p+%x WC)\n",
977 		  &efx->membase_phys, efx->membase, uc_mem_map_size,
978 		  nic_data->wc_membase, wc_mem_map_size);
979 
980 	return 0;
981 }
982 
983 static int efx_ef10_init_nic(struct efx_nic *efx)
984 {
985 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
986 	int rc;
987 
988 	if (nic_data->must_check_datapath_caps) {
989 		rc = efx_ef10_init_datapath_caps(efx);
990 		if (rc)
991 			return rc;
992 		nic_data->must_check_datapath_caps = false;
993 	}
994 
995 	if (nic_data->must_realloc_vis) {
996 		/* We cannot let the number of VIs change now */
997 		rc = efx_ef10_alloc_vis(efx, nic_data->n_allocated_vis,
998 					nic_data->n_allocated_vis);
999 		if (rc)
1000 			return rc;
1001 		nic_data->must_realloc_vis = false;
1002 	}
1003 
1004 	if (nic_data->must_restore_piobufs && nic_data->n_piobufs) {
1005 		rc = efx_ef10_alloc_piobufs(efx, nic_data->n_piobufs);
1006 		if (rc == 0) {
1007 			rc = efx_ef10_link_piobufs(efx);
1008 			if (rc)
1009 				efx_ef10_free_piobufs(efx);
1010 		}
1011 
1012 		/* Log an error on failure, but this is non-fatal */
1013 		if (rc)
1014 			netif_err(efx, drv, efx->net_dev,
1015 				  "failed to restore PIO buffers (%d)\n", rc);
1016 		nic_data->must_restore_piobufs = false;
1017 	}
1018 
1019 	/* don't fail init if RSS setup doesn't work */
1020 	efx->type->rx_push_rss_config(efx, false, efx->rx_indir_table);
1021 
1022 	return 0;
1023 }
1024 
1025 static void efx_ef10_reset_mc_allocations(struct efx_nic *efx)
1026 {
1027 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1028 #ifdef CONFIG_SFC_SRIOV
1029 	unsigned int i;
1030 #endif
1031 
1032 	/* All our allocations have been reset */
1033 	nic_data->must_realloc_vis = true;
1034 	nic_data->must_restore_filters = true;
1035 	nic_data->must_restore_piobufs = true;
1036 	efx_ef10_forget_old_piobufs(efx);
1037 	nic_data->rx_rss_context = EFX_EF10_RSS_CONTEXT_INVALID;
1038 
1039 	/* Driver-created vswitches and vports must be re-created */
1040 	nic_data->must_probe_vswitching = true;
1041 	nic_data->vport_id = EVB_PORT_ID_ASSIGNED;
1042 #ifdef CONFIG_SFC_SRIOV
1043 	if (nic_data->vf)
1044 		for (i = 0; i < efx->vf_count; i++)
1045 			nic_data->vf[i].vport_id = 0;
1046 #endif
1047 }
1048 
1049 static enum reset_type efx_ef10_map_reset_reason(enum reset_type reason)
1050 {
1051 	if (reason == RESET_TYPE_MC_FAILURE)
1052 		return RESET_TYPE_DATAPATH;
1053 
1054 	return efx_mcdi_map_reset_reason(reason);
1055 }
1056 
1057 static int efx_ef10_map_reset_flags(u32 *flags)
1058 {
1059 	enum {
1060 		EF10_RESET_PORT = ((ETH_RESET_MAC | ETH_RESET_PHY) <<
1061 				   ETH_RESET_SHARED_SHIFT),
1062 		EF10_RESET_MC = ((ETH_RESET_DMA | ETH_RESET_FILTER |
1063 				  ETH_RESET_OFFLOAD | ETH_RESET_MAC |
1064 				  ETH_RESET_PHY | ETH_RESET_MGMT) <<
1065 				 ETH_RESET_SHARED_SHIFT)
1066 	};
1067 
1068 	/* We assume for now that our PCI function is permitted to
1069 	 * reset everything.
1070 	 */
1071 
1072 	if ((*flags & EF10_RESET_MC) == EF10_RESET_MC) {
1073 		*flags &= ~EF10_RESET_MC;
1074 		return RESET_TYPE_WORLD;
1075 	}
1076 
1077 	if ((*flags & EF10_RESET_PORT) == EF10_RESET_PORT) {
1078 		*flags &= ~EF10_RESET_PORT;
1079 		return RESET_TYPE_ALL;
1080 	}
1081 
1082 	/* no invisible reset implemented */
1083 
1084 	return -EINVAL;
1085 }
1086 
1087 static int efx_ef10_reset(struct efx_nic *efx, enum reset_type reset_type)
1088 {
1089 	int rc = efx_mcdi_reset(efx, reset_type);
1090 
1091 	/* Unprivileged functions return -EPERM, but need to return success
1092 	 * here so that the datapath is brought back up.
1093 	 */
1094 	if (reset_type == RESET_TYPE_WORLD && rc == -EPERM)
1095 		rc = 0;
1096 
1097 	/* If it was a port reset, trigger reallocation of MC resources.
1098 	 * Note that on an MC reset nothing needs to be done now because we'll
1099 	 * detect the MC reset later and handle it then.
1100 	 * For an FLR, we never get an MC reset event, but the MC has reset all
1101 	 * resources assigned to us, so we have to trigger reallocation now.
1102 	 */
1103 	if ((reset_type == RESET_TYPE_ALL ||
1104 	     reset_type == RESET_TYPE_MCDI_TIMEOUT) && !rc)
1105 		efx_ef10_reset_mc_allocations(efx);
1106 	return rc;
1107 }
1108 
1109 #define EF10_DMA_STAT(ext_name, mcdi_name)			\
1110 	[EF10_STAT_ ## ext_name] =				\
1111 	{ #ext_name, 64, 8 * MC_CMD_MAC_ ## mcdi_name }
1112 #define EF10_DMA_INVIS_STAT(int_name, mcdi_name)		\
1113 	[EF10_STAT_ ## int_name] =				\
1114 	{ NULL, 64, 8 * MC_CMD_MAC_ ## mcdi_name }
1115 #define EF10_OTHER_STAT(ext_name)				\
1116 	[EF10_STAT_ ## ext_name] = { #ext_name, 0, 0 }
1117 #define GENERIC_SW_STAT(ext_name)				\
1118 	[GENERIC_STAT_ ## ext_name] = { #ext_name, 0, 0 }
1119 
1120 static const struct efx_hw_stat_desc efx_ef10_stat_desc[EF10_STAT_COUNT] = {
1121 	EF10_DMA_STAT(port_tx_bytes, TX_BYTES),
1122 	EF10_DMA_STAT(port_tx_packets, TX_PKTS),
1123 	EF10_DMA_STAT(port_tx_pause, TX_PAUSE_PKTS),
1124 	EF10_DMA_STAT(port_tx_control, TX_CONTROL_PKTS),
1125 	EF10_DMA_STAT(port_tx_unicast, TX_UNICAST_PKTS),
1126 	EF10_DMA_STAT(port_tx_multicast, TX_MULTICAST_PKTS),
1127 	EF10_DMA_STAT(port_tx_broadcast, TX_BROADCAST_PKTS),
1128 	EF10_DMA_STAT(port_tx_lt64, TX_LT64_PKTS),
1129 	EF10_DMA_STAT(port_tx_64, TX_64_PKTS),
1130 	EF10_DMA_STAT(port_tx_65_to_127, TX_65_TO_127_PKTS),
1131 	EF10_DMA_STAT(port_tx_128_to_255, TX_128_TO_255_PKTS),
1132 	EF10_DMA_STAT(port_tx_256_to_511, TX_256_TO_511_PKTS),
1133 	EF10_DMA_STAT(port_tx_512_to_1023, TX_512_TO_1023_PKTS),
1134 	EF10_DMA_STAT(port_tx_1024_to_15xx, TX_1024_TO_15XX_PKTS),
1135 	EF10_DMA_STAT(port_tx_15xx_to_jumbo, TX_15XX_TO_JUMBO_PKTS),
1136 	EF10_DMA_STAT(port_rx_bytes, RX_BYTES),
1137 	EF10_DMA_INVIS_STAT(port_rx_bytes_minus_good_bytes, RX_BAD_BYTES),
1138 	EF10_OTHER_STAT(port_rx_good_bytes),
1139 	EF10_OTHER_STAT(port_rx_bad_bytes),
1140 	EF10_DMA_STAT(port_rx_packets, RX_PKTS),
1141 	EF10_DMA_STAT(port_rx_good, RX_GOOD_PKTS),
1142 	EF10_DMA_STAT(port_rx_bad, RX_BAD_FCS_PKTS),
1143 	EF10_DMA_STAT(port_rx_pause, RX_PAUSE_PKTS),
1144 	EF10_DMA_STAT(port_rx_control, RX_CONTROL_PKTS),
1145 	EF10_DMA_STAT(port_rx_unicast, RX_UNICAST_PKTS),
1146 	EF10_DMA_STAT(port_rx_multicast, RX_MULTICAST_PKTS),
1147 	EF10_DMA_STAT(port_rx_broadcast, RX_BROADCAST_PKTS),
1148 	EF10_DMA_STAT(port_rx_lt64, RX_UNDERSIZE_PKTS),
1149 	EF10_DMA_STAT(port_rx_64, RX_64_PKTS),
1150 	EF10_DMA_STAT(port_rx_65_to_127, RX_65_TO_127_PKTS),
1151 	EF10_DMA_STAT(port_rx_128_to_255, RX_128_TO_255_PKTS),
1152 	EF10_DMA_STAT(port_rx_256_to_511, RX_256_TO_511_PKTS),
1153 	EF10_DMA_STAT(port_rx_512_to_1023, RX_512_TO_1023_PKTS),
1154 	EF10_DMA_STAT(port_rx_1024_to_15xx, RX_1024_TO_15XX_PKTS),
1155 	EF10_DMA_STAT(port_rx_15xx_to_jumbo, RX_15XX_TO_JUMBO_PKTS),
1156 	EF10_DMA_STAT(port_rx_gtjumbo, RX_GTJUMBO_PKTS),
1157 	EF10_DMA_STAT(port_rx_bad_gtjumbo, RX_JABBER_PKTS),
1158 	EF10_DMA_STAT(port_rx_overflow, RX_OVERFLOW_PKTS),
1159 	EF10_DMA_STAT(port_rx_align_error, RX_ALIGN_ERROR_PKTS),
1160 	EF10_DMA_STAT(port_rx_length_error, RX_LENGTH_ERROR_PKTS),
1161 	EF10_DMA_STAT(port_rx_nodesc_drops, RX_NODESC_DROPS),
1162 	GENERIC_SW_STAT(rx_nodesc_trunc),
1163 	GENERIC_SW_STAT(rx_noskb_drops),
1164 	EF10_DMA_STAT(port_rx_pm_trunc_bb_overflow, PM_TRUNC_BB_OVERFLOW),
1165 	EF10_DMA_STAT(port_rx_pm_discard_bb_overflow, PM_DISCARD_BB_OVERFLOW),
1166 	EF10_DMA_STAT(port_rx_pm_trunc_vfifo_full, PM_TRUNC_VFIFO_FULL),
1167 	EF10_DMA_STAT(port_rx_pm_discard_vfifo_full, PM_DISCARD_VFIFO_FULL),
1168 	EF10_DMA_STAT(port_rx_pm_trunc_qbb, PM_TRUNC_QBB),
1169 	EF10_DMA_STAT(port_rx_pm_discard_qbb, PM_DISCARD_QBB),
1170 	EF10_DMA_STAT(port_rx_pm_discard_mapping, PM_DISCARD_MAPPING),
1171 	EF10_DMA_STAT(port_rx_dp_q_disabled_packets, RXDP_Q_DISABLED_PKTS),
1172 	EF10_DMA_STAT(port_rx_dp_di_dropped_packets, RXDP_DI_DROPPED_PKTS),
1173 	EF10_DMA_STAT(port_rx_dp_streaming_packets, RXDP_STREAMING_PKTS),
1174 	EF10_DMA_STAT(port_rx_dp_hlb_fetch, RXDP_HLB_FETCH_CONDITIONS),
1175 	EF10_DMA_STAT(port_rx_dp_hlb_wait, RXDP_HLB_WAIT_CONDITIONS),
1176 	EF10_DMA_STAT(rx_unicast, VADAPTER_RX_UNICAST_PACKETS),
1177 	EF10_DMA_STAT(rx_unicast_bytes, VADAPTER_RX_UNICAST_BYTES),
1178 	EF10_DMA_STAT(rx_multicast, VADAPTER_RX_MULTICAST_PACKETS),
1179 	EF10_DMA_STAT(rx_multicast_bytes, VADAPTER_RX_MULTICAST_BYTES),
1180 	EF10_DMA_STAT(rx_broadcast, VADAPTER_RX_BROADCAST_PACKETS),
1181 	EF10_DMA_STAT(rx_broadcast_bytes, VADAPTER_RX_BROADCAST_BYTES),
1182 	EF10_DMA_STAT(rx_bad, VADAPTER_RX_BAD_PACKETS),
1183 	EF10_DMA_STAT(rx_bad_bytes, VADAPTER_RX_BAD_BYTES),
1184 	EF10_DMA_STAT(rx_overflow, VADAPTER_RX_OVERFLOW),
1185 	EF10_DMA_STAT(tx_unicast, VADAPTER_TX_UNICAST_PACKETS),
1186 	EF10_DMA_STAT(tx_unicast_bytes, VADAPTER_TX_UNICAST_BYTES),
1187 	EF10_DMA_STAT(tx_multicast, VADAPTER_TX_MULTICAST_PACKETS),
1188 	EF10_DMA_STAT(tx_multicast_bytes, VADAPTER_TX_MULTICAST_BYTES),
1189 	EF10_DMA_STAT(tx_broadcast, VADAPTER_TX_BROADCAST_PACKETS),
1190 	EF10_DMA_STAT(tx_broadcast_bytes, VADAPTER_TX_BROADCAST_BYTES),
1191 	EF10_DMA_STAT(tx_bad, VADAPTER_TX_BAD_PACKETS),
1192 	EF10_DMA_STAT(tx_bad_bytes, VADAPTER_TX_BAD_BYTES),
1193 	EF10_DMA_STAT(tx_overflow, VADAPTER_TX_OVERFLOW),
1194 };
1195 
1196 #define HUNT_COMMON_STAT_MASK ((1ULL << EF10_STAT_port_tx_bytes) |	\
1197 			       (1ULL << EF10_STAT_port_tx_packets) |	\
1198 			       (1ULL << EF10_STAT_port_tx_pause) |	\
1199 			       (1ULL << EF10_STAT_port_tx_unicast) |	\
1200 			       (1ULL << EF10_STAT_port_tx_multicast) |	\
1201 			       (1ULL << EF10_STAT_port_tx_broadcast) |	\
1202 			       (1ULL << EF10_STAT_port_rx_bytes) |	\
1203 			       (1ULL <<                                 \
1204 				EF10_STAT_port_rx_bytes_minus_good_bytes) | \
1205 			       (1ULL << EF10_STAT_port_rx_good_bytes) |	\
1206 			       (1ULL << EF10_STAT_port_rx_bad_bytes) |	\
1207 			       (1ULL << EF10_STAT_port_rx_packets) |	\
1208 			       (1ULL << EF10_STAT_port_rx_good) |	\
1209 			       (1ULL << EF10_STAT_port_rx_bad) |	\
1210 			       (1ULL << EF10_STAT_port_rx_pause) |	\
1211 			       (1ULL << EF10_STAT_port_rx_control) |	\
1212 			       (1ULL << EF10_STAT_port_rx_unicast) |	\
1213 			       (1ULL << EF10_STAT_port_rx_multicast) |	\
1214 			       (1ULL << EF10_STAT_port_rx_broadcast) |	\
1215 			       (1ULL << EF10_STAT_port_rx_lt64) |	\
1216 			       (1ULL << EF10_STAT_port_rx_64) |		\
1217 			       (1ULL << EF10_STAT_port_rx_65_to_127) |	\
1218 			       (1ULL << EF10_STAT_port_rx_128_to_255) |	\
1219 			       (1ULL << EF10_STAT_port_rx_256_to_511) |	\
1220 			       (1ULL << EF10_STAT_port_rx_512_to_1023) |\
1221 			       (1ULL << EF10_STAT_port_rx_1024_to_15xx) |\
1222 			       (1ULL << EF10_STAT_port_rx_15xx_to_jumbo) |\
1223 			       (1ULL << EF10_STAT_port_rx_gtjumbo) |	\
1224 			       (1ULL << EF10_STAT_port_rx_bad_gtjumbo) |\
1225 			       (1ULL << EF10_STAT_port_rx_overflow) |	\
1226 			       (1ULL << EF10_STAT_port_rx_nodesc_drops) |\
1227 			       (1ULL << GENERIC_STAT_rx_nodesc_trunc) |	\
1228 			       (1ULL << GENERIC_STAT_rx_noskb_drops))
1229 
1230 /* These statistics are only provided by the 10G MAC.  For a 10G/40G
1231  * switchable port we do not expose these because they might not
1232  * include all the packets they should.
1233  */
1234 #define HUNT_10G_ONLY_STAT_MASK ((1ULL << EF10_STAT_port_tx_control) |	\
1235 				 (1ULL << EF10_STAT_port_tx_lt64) |	\
1236 				 (1ULL << EF10_STAT_port_tx_64) |	\
1237 				 (1ULL << EF10_STAT_port_tx_65_to_127) |\
1238 				 (1ULL << EF10_STAT_port_tx_128_to_255) |\
1239 				 (1ULL << EF10_STAT_port_tx_256_to_511) |\
1240 				 (1ULL << EF10_STAT_port_tx_512_to_1023) |\
1241 				 (1ULL << EF10_STAT_port_tx_1024_to_15xx) |\
1242 				 (1ULL << EF10_STAT_port_tx_15xx_to_jumbo))
1243 
1244 /* These statistics are only provided by the 40G MAC.  For a 10G/40G
1245  * switchable port we do expose these because the errors will otherwise
1246  * be silent.
1247  */
1248 #define HUNT_40G_EXTRA_STAT_MASK ((1ULL << EF10_STAT_port_rx_align_error) |\
1249 				  (1ULL << EF10_STAT_port_rx_length_error))
1250 
1251 /* These statistics are only provided if the firmware supports the
1252  * capability PM_AND_RXDP_COUNTERS.
1253  */
1254 #define HUNT_PM_AND_RXDP_STAT_MASK (					\
1255 	(1ULL << EF10_STAT_port_rx_pm_trunc_bb_overflow) |		\
1256 	(1ULL << EF10_STAT_port_rx_pm_discard_bb_overflow) |		\
1257 	(1ULL << EF10_STAT_port_rx_pm_trunc_vfifo_full) |		\
1258 	(1ULL << EF10_STAT_port_rx_pm_discard_vfifo_full) |		\
1259 	(1ULL << EF10_STAT_port_rx_pm_trunc_qbb) |			\
1260 	(1ULL << EF10_STAT_port_rx_pm_discard_qbb) |			\
1261 	(1ULL << EF10_STAT_port_rx_pm_discard_mapping) |		\
1262 	(1ULL << EF10_STAT_port_rx_dp_q_disabled_packets) |		\
1263 	(1ULL << EF10_STAT_port_rx_dp_di_dropped_packets) |		\
1264 	(1ULL << EF10_STAT_port_rx_dp_streaming_packets) |		\
1265 	(1ULL << EF10_STAT_port_rx_dp_hlb_fetch) |			\
1266 	(1ULL << EF10_STAT_port_rx_dp_hlb_wait))
1267 
1268 static u64 efx_ef10_raw_stat_mask(struct efx_nic *efx)
1269 {
1270 	u64 raw_mask = HUNT_COMMON_STAT_MASK;
1271 	u32 port_caps = efx_mcdi_phy_get_caps(efx);
1272 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1273 
1274 	if (!(efx->mcdi->fn_flags &
1275 	      1 << MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_LINKCTRL))
1276 		return 0;
1277 
1278 	if (port_caps & (1 << MC_CMD_PHY_CAP_40000FDX_LBN))
1279 		raw_mask |= HUNT_40G_EXTRA_STAT_MASK;
1280 	else
1281 		raw_mask |= HUNT_10G_ONLY_STAT_MASK;
1282 
1283 	if (nic_data->datapath_caps &
1284 	    (1 << MC_CMD_GET_CAPABILITIES_OUT_PM_AND_RXDP_COUNTERS_LBN))
1285 		raw_mask |= HUNT_PM_AND_RXDP_STAT_MASK;
1286 
1287 	return raw_mask;
1288 }
1289 
1290 static void efx_ef10_get_stat_mask(struct efx_nic *efx, unsigned long *mask)
1291 {
1292 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1293 	u64 raw_mask[2];
1294 
1295 	raw_mask[0] = efx_ef10_raw_stat_mask(efx);
1296 
1297 	/* Only show vadaptor stats when EVB capability is present */
1298 	if (nic_data->datapath_caps &
1299 	    (1 << MC_CMD_GET_CAPABILITIES_OUT_EVB_LBN)) {
1300 		raw_mask[0] |= ~((1ULL << EF10_STAT_rx_unicast) - 1);
1301 		raw_mask[1] = (1ULL << (EF10_STAT_COUNT - 63)) - 1;
1302 	} else {
1303 		raw_mask[1] = 0;
1304 	}
1305 
1306 #if BITS_PER_LONG == 64
1307 	mask[0] = raw_mask[0];
1308 	mask[1] = raw_mask[1];
1309 #else
1310 	mask[0] = raw_mask[0] & 0xffffffff;
1311 	mask[1] = raw_mask[0] >> 32;
1312 	mask[2] = raw_mask[1] & 0xffffffff;
1313 	mask[3] = raw_mask[1] >> 32;
1314 #endif
1315 }
1316 
1317 static size_t efx_ef10_describe_stats(struct efx_nic *efx, u8 *names)
1318 {
1319 	DECLARE_BITMAP(mask, EF10_STAT_COUNT);
1320 
1321 	efx_ef10_get_stat_mask(efx, mask);
1322 	return efx_nic_describe_stats(efx_ef10_stat_desc, EF10_STAT_COUNT,
1323 				      mask, names);
1324 }
1325 
1326 static size_t efx_ef10_update_stats_common(struct efx_nic *efx, u64 *full_stats,
1327 					   struct rtnl_link_stats64 *core_stats)
1328 {
1329 	DECLARE_BITMAP(mask, EF10_STAT_COUNT);
1330 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1331 	u64 *stats = nic_data->stats;
1332 	size_t stats_count = 0, index;
1333 
1334 	efx_ef10_get_stat_mask(efx, mask);
1335 
1336 	if (full_stats) {
1337 		for_each_set_bit(index, mask, EF10_STAT_COUNT) {
1338 			if (efx_ef10_stat_desc[index].name) {
1339 				*full_stats++ = stats[index];
1340 				++stats_count;
1341 			}
1342 		}
1343 	}
1344 
1345 	if (!core_stats)
1346 		return stats_count;
1347 
1348 	if (nic_data->datapath_caps &
1349 			1 << MC_CMD_GET_CAPABILITIES_OUT_EVB_LBN) {
1350 		/* Use vadaptor stats. */
1351 		core_stats->rx_packets = stats[EF10_STAT_rx_unicast] +
1352 					 stats[EF10_STAT_rx_multicast] +
1353 					 stats[EF10_STAT_rx_broadcast];
1354 		core_stats->tx_packets = stats[EF10_STAT_tx_unicast] +
1355 					 stats[EF10_STAT_tx_multicast] +
1356 					 stats[EF10_STAT_tx_broadcast];
1357 		core_stats->rx_bytes = stats[EF10_STAT_rx_unicast_bytes] +
1358 				       stats[EF10_STAT_rx_multicast_bytes] +
1359 				       stats[EF10_STAT_rx_broadcast_bytes];
1360 		core_stats->tx_bytes = stats[EF10_STAT_tx_unicast_bytes] +
1361 				       stats[EF10_STAT_tx_multicast_bytes] +
1362 				       stats[EF10_STAT_tx_broadcast_bytes];
1363 		core_stats->rx_dropped = stats[GENERIC_STAT_rx_nodesc_trunc] +
1364 					 stats[GENERIC_STAT_rx_noskb_drops];
1365 		core_stats->multicast = stats[EF10_STAT_rx_multicast];
1366 		core_stats->rx_crc_errors = stats[EF10_STAT_rx_bad];
1367 		core_stats->rx_fifo_errors = stats[EF10_STAT_rx_overflow];
1368 		core_stats->rx_errors = core_stats->rx_crc_errors;
1369 		core_stats->tx_errors = stats[EF10_STAT_tx_bad];
1370 	} else {
1371 		/* Use port stats. */
1372 		core_stats->rx_packets = stats[EF10_STAT_port_rx_packets];
1373 		core_stats->tx_packets = stats[EF10_STAT_port_tx_packets];
1374 		core_stats->rx_bytes = stats[EF10_STAT_port_rx_bytes];
1375 		core_stats->tx_bytes = stats[EF10_STAT_port_tx_bytes];
1376 		core_stats->rx_dropped = stats[EF10_STAT_port_rx_nodesc_drops] +
1377 					 stats[GENERIC_STAT_rx_nodesc_trunc] +
1378 					 stats[GENERIC_STAT_rx_noskb_drops];
1379 		core_stats->multicast = stats[EF10_STAT_port_rx_multicast];
1380 		core_stats->rx_length_errors =
1381 				stats[EF10_STAT_port_rx_gtjumbo] +
1382 				stats[EF10_STAT_port_rx_length_error];
1383 		core_stats->rx_crc_errors = stats[EF10_STAT_port_rx_bad];
1384 		core_stats->rx_frame_errors =
1385 				stats[EF10_STAT_port_rx_align_error];
1386 		core_stats->rx_fifo_errors = stats[EF10_STAT_port_rx_overflow];
1387 		core_stats->rx_errors = (core_stats->rx_length_errors +
1388 					 core_stats->rx_crc_errors +
1389 					 core_stats->rx_frame_errors);
1390 	}
1391 
1392 	return stats_count;
1393 }
1394 
1395 static int efx_ef10_try_update_nic_stats_pf(struct efx_nic *efx)
1396 {
1397 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1398 	DECLARE_BITMAP(mask, EF10_STAT_COUNT);
1399 	__le64 generation_start, generation_end;
1400 	u64 *stats = nic_data->stats;
1401 	__le64 *dma_stats;
1402 
1403 	efx_ef10_get_stat_mask(efx, mask);
1404 
1405 	dma_stats = efx->stats_buffer.addr;
1406 	nic_data = efx->nic_data;
1407 
1408 	generation_end = dma_stats[MC_CMD_MAC_GENERATION_END];
1409 	if (generation_end == EFX_MC_STATS_GENERATION_INVALID)
1410 		return 0;
1411 	rmb();
1412 	efx_nic_update_stats(efx_ef10_stat_desc, EF10_STAT_COUNT, mask,
1413 			     stats, efx->stats_buffer.addr, false);
1414 	rmb();
1415 	generation_start = dma_stats[MC_CMD_MAC_GENERATION_START];
1416 	if (generation_end != generation_start)
1417 		return -EAGAIN;
1418 
1419 	/* Update derived statistics */
1420 	efx_nic_fix_nodesc_drop_stat(efx,
1421 				     &stats[EF10_STAT_port_rx_nodesc_drops]);
1422 	stats[EF10_STAT_port_rx_good_bytes] =
1423 		stats[EF10_STAT_port_rx_bytes] -
1424 		stats[EF10_STAT_port_rx_bytes_minus_good_bytes];
1425 	efx_update_diff_stat(&stats[EF10_STAT_port_rx_bad_bytes],
1426 			     stats[EF10_STAT_port_rx_bytes_minus_good_bytes]);
1427 	efx_update_sw_stats(efx, stats);
1428 	return 0;
1429 }
1430 
1431 
1432 static size_t efx_ef10_update_stats_pf(struct efx_nic *efx, u64 *full_stats,
1433 				       struct rtnl_link_stats64 *core_stats)
1434 {
1435 	int retry;
1436 
1437 	/* If we're unlucky enough to read statistics during the DMA, wait
1438 	 * up to 10ms for it to finish (typically takes <500us)
1439 	 */
1440 	for (retry = 0; retry < 100; ++retry) {
1441 		if (efx_ef10_try_update_nic_stats_pf(efx) == 0)
1442 			break;
1443 		udelay(100);
1444 	}
1445 
1446 	return efx_ef10_update_stats_common(efx, full_stats, core_stats);
1447 }
1448 
1449 static int efx_ef10_try_update_nic_stats_vf(struct efx_nic *efx)
1450 {
1451 	MCDI_DECLARE_BUF(inbuf, MC_CMD_MAC_STATS_IN_LEN);
1452 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1453 	DECLARE_BITMAP(mask, EF10_STAT_COUNT);
1454 	__le64 generation_start, generation_end;
1455 	u64 *stats = nic_data->stats;
1456 	u32 dma_len = MC_CMD_MAC_NSTATS * sizeof(u64);
1457 	struct efx_buffer stats_buf;
1458 	__le64 *dma_stats;
1459 	int rc;
1460 
1461 	spin_unlock_bh(&efx->stats_lock);
1462 
1463 	if (in_interrupt()) {
1464 		/* If in atomic context, cannot update stats.  Just update the
1465 		 * software stats and return so the caller can continue.
1466 		 */
1467 		spin_lock_bh(&efx->stats_lock);
1468 		efx_update_sw_stats(efx, stats);
1469 		return 0;
1470 	}
1471 
1472 	efx_ef10_get_stat_mask(efx, mask);
1473 
1474 	rc = efx_nic_alloc_buffer(efx, &stats_buf, dma_len, GFP_ATOMIC);
1475 	if (rc) {
1476 		spin_lock_bh(&efx->stats_lock);
1477 		return rc;
1478 	}
1479 
1480 	dma_stats = stats_buf.addr;
1481 	dma_stats[MC_CMD_MAC_GENERATION_END] = EFX_MC_STATS_GENERATION_INVALID;
1482 
1483 	MCDI_SET_QWORD(inbuf, MAC_STATS_IN_DMA_ADDR, stats_buf.dma_addr);
1484 	MCDI_POPULATE_DWORD_1(inbuf, MAC_STATS_IN_CMD,
1485 			      MAC_STATS_IN_DMA, 1);
1486 	MCDI_SET_DWORD(inbuf, MAC_STATS_IN_DMA_LEN, dma_len);
1487 	MCDI_SET_DWORD(inbuf, MAC_STATS_IN_PORT_ID, EVB_PORT_ID_ASSIGNED);
1488 
1489 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_MAC_STATS, inbuf, sizeof(inbuf),
1490 				NULL, 0, NULL);
1491 	spin_lock_bh(&efx->stats_lock);
1492 	if (rc) {
1493 		/* Expect ENOENT if DMA queues have not been set up */
1494 		if (rc != -ENOENT || atomic_read(&efx->active_queues))
1495 			efx_mcdi_display_error(efx, MC_CMD_MAC_STATS,
1496 					       sizeof(inbuf), NULL, 0, rc);
1497 		goto out;
1498 	}
1499 
1500 	generation_end = dma_stats[MC_CMD_MAC_GENERATION_END];
1501 	if (generation_end == EFX_MC_STATS_GENERATION_INVALID) {
1502 		WARN_ON_ONCE(1);
1503 		goto out;
1504 	}
1505 	rmb();
1506 	efx_nic_update_stats(efx_ef10_stat_desc, EF10_STAT_COUNT, mask,
1507 			     stats, stats_buf.addr, false);
1508 	rmb();
1509 	generation_start = dma_stats[MC_CMD_MAC_GENERATION_START];
1510 	if (generation_end != generation_start) {
1511 		rc = -EAGAIN;
1512 		goto out;
1513 	}
1514 
1515 	efx_update_sw_stats(efx, stats);
1516 out:
1517 	efx_nic_free_buffer(efx, &stats_buf);
1518 	return rc;
1519 }
1520 
1521 static size_t efx_ef10_update_stats_vf(struct efx_nic *efx, u64 *full_stats,
1522 				       struct rtnl_link_stats64 *core_stats)
1523 {
1524 	if (efx_ef10_try_update_nic_stats_vf(efx))
1525 		return 0;
1526 
1527 	return efx_ef10_update_stats_common(efx, full_stats, core_stats);
1528 }
1529 
1530 static void efx_ef10_push_irq_moderation(struct efx_channel *channel)
1531 {
1532 	struct efx_nic *efx = channel->efx;
1533 	unsigned int mode, value;
1534 	efx_dword_t timer_cmd;
1535 
1536 	if (channel->irq_moderation) {
1537 		mode = 3;
1538 		value = channel->irq_moderation - 1;
1539 	} else {
1540 		mode = 0;
1541 		value = 0;
1542 	}
1543 
1544 	if (EFX_EF10_WORKAROUND_35388(efx)) {
1545 		EFX_POPULATE_DWORD_3(timer_cmd, ERF_DD_EVQ_IND_TIMER_FLAGS,
1546 				     EFE_DD_EVQ_IND_TIMER_FLAGS,
1547 				     ERF_DD_EVQ_IND_TIMER_MODE, mode,
1548 				     ERF_DD_EVQ_IND_TIMER_VAL, value);
1549 		efx_writed_page(efx, &timer_cmd, ER_DD_EVQ_INDIRECT,
1550 				channel->channel);
1551 	} else {
1552 		EFX_POPULATE_DWORD_2(timer_cmd, ERF_DZ_TC_TIMER_MODE, mode,
1553 				     ERF_DZ_TC_TIMER_VAL, value);
1554 		efx_writed_page(efx, &timer_cmd, ER_DZ_EVQ_TMR,
1555 				channel->channel);
1556 	}
1557 }
1558 
1559 static void efx_ef10_get_wol_vf(struct efx_nic *efx,
1560 				struct ethtool_wolinfo *wol) {}
1561 
1562 static int efx_ef10_set_wol_vf(struct efx_nic *efx, u32 type)
1563 {
1564 	return -EOPNOTSUPP;
1565 }
1566 
1567 static void efx_ef10_get_wol(struct efx_nic *efx, struct ethtool_wolinfo *wol)
1568 {
1569 	wol->supported = 0;
1570 	wol->wolopts = 0;
1571 	memset(&wol->sopass, 0, sizeof(wol->sopass));
1572 }
1573 
1574 static int efx_ef10_set_wol(struct efx_nic *efx, u32 type)
1575 {
1576 	if (type != 0)
1577 		return -EINVAL;
1578 	return 0;
1579 }
1580 
1581 static void efx_ef10_mcdi_request(struct efx_nic *efx,
1582 				  const efx_dword_t *hdr, size_t hdr_len,
1583 				  const efx_dword_t *sdu, size_t sdu_len)
1584 {
1585 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1586 	u8 *pdu = nic_data->mcdi_buf.addr;
1587 
1588 	memcpy(pdu, hdr, hdr_len);
1589 	memcpy(pdu + hdr_len, sdu, sdu_len);
1590 	wmb();
1591 
1592 	/* The hardware provides 'low' and 'high' (doorbell) registers
1593 	 * for passing the 64-bit address of an MCDI request to
1594 	 * firmware.  However the dwords are swapped by firmware.  The
1595 	 * least significant bits of the doorbell are then 0 for all
1596 	 * MCDI requests due to alignment.
1597 	 */
1598 	_efx_writed(efx, cpu_to_le32((u64)nic_data->mcdi_buf.dma_addr >> 32),
1599 		    ER_DZ_MC_DB_LWRD);
1600 	_efx_writed(efx, cpu_to_le32((u32)nic_data->mcdi_buf.dma_addr),
1601 		    ER_DZ_MC_DB_HWRD);
1602 }
1603 
1604 static bool efx_ef10_mcdi_poll_response(struct efx_nic *efx)
1605 {
1606 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1607 	const efx_dword_t hdr = *(const efx_dword_t *)nic_data->mcdi_buf.addr;
1608 
1609 	rmb();
1610 	return EFX_DWORD_FIELD(hdr, MCDI_HEADER_RESPONSE);
1611 }
1612 
1613 static void
1614 efx_ef10_mcdi_read_response(struct efx_nic *efx, efx_dword_t *outbuf,
1615 			    size_t offset, size_t outlen)
1616 {
1617 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1618 	const u8 *pdu = nic_data->mcdi_buf.addr;
1619 
1620 	memcpy(outbuf, pdu + offset, outlen);
1621 }
1622 
1623 static void efx_ef10_mcdi_reboot_detected(struct efx_nic *efx)
1624 {
1625 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1626 
1627 	/* All our allocations have been reset */
1628 	efx_ef10_reset_mc_allocations(efx);
1629 
1630 	/* The datapath firmware might have been changed */
1631 	nic_data->must_check_datapath_caps = true;
1632 
1633 	/* MAC statistics have been cleared on the NIC; clear the local
1634 	 * statistic that we update with efx_update_diff_stat().
1635 	 */
1636 	nic_data->stats[EF10_STAT_port_rx_bad_bytes] = 0;
1637 }
1638 
1639 static int efx_ef10_mcdi_poll_reboot(struct efx_nic *efx)
1640 {
1641 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1642 	int rc;
1643 
1644 	rc = efx_ef10_get_warm_boot_count(efx);
1645 	if (rc < 0) {
1646 		/* The firmware is presumably in the process of
1647 		 * rebooting.  However, we are supposed to report each
1648 		 * reboot just once, so we must only do that once we
1649 		 * can read and store the updated warm boot count.
1650 		 */
1651 		return 0;
1652 	}
1653 
1654 	if (rc == nic_data->warm_boot_count)
1655 		return 0;
1656 
1657 	nic_data->warm_boot_count = rc;
1658 	efx_ef10_mcdi_reboot_detected(efx);
1659 
1660 	return -EIO;
1661 }
1662 
1663 /* Handle an MSI interrupt
1664  *
1665  * Handle an MSI hardware interrupt.  This routine schedules event
1666  * queue processing.  No interrupt acknowledgement cycle is necessary.
1667  * Also, we never need to check that the interrupt is for us, since
1668  * MSI interrupts cannot be shared.
1669  */
1670 static irqreturn_t efx_ef10_msi_interrupt(int irq, void *dev_id)
1671 {
1672 	struct efx_msi_context *context = dev_id;
1673 	struct efx_nic *efx = context->efx;
1674 
1675 	netif_vdbg(efx, intr, efx->net_dev,
1676 		   "IRQ %d on CPU %d\n", irq, raw_smp_processor_id());
1677 
1678 	if (likely(ACCESS_ONCE(efx->irq_soft_enabled))) {
1679 		/* Note test interrupts */
1680 		if (context->index == efx->irq_level)
1681 			efx->last_irq_cpu = raw_smp_processor_id();
1682 
1683 		/* Schedule processing of the channel */
1684 		efx_schedule_channel_irq(efx->channel[context->index]);
1685 	}
1686 
1687 	return IRQ_HANDLED;
1688 }
1689 
1690 static irqreturn_t efx_ef10_legacy_interrupt(int irq, void *dev_id)
1691 {
1692 	struct efx_nic *efx = dev_id;
1693 	bool soft_enabled = ACCESS_ONCE(efx->irq_soft_enabled);
1694 	struct efx_channel *channel;
1695 	efx_dword_t reg;
1696 	u32 queues;
1697 
1698 	/* Read the ISR which also ACKs the interrupts */
1699 	efx_readd(efx, &reg, ER_DZ_BIU_INT_ISR);
1700 	queues = EFX_DWORD_FIELD(reg, ERF_DZ_ISR_REG);
1701 
1702 	if (queues == 0)
1703 		return IRQ_NONE;
1704 
1705 	if (likely(soft_enabled)) {
1706 		/* Note test interrupts */
1707 		if (queues & (1U << efx->irq_level))
1708 			efx->last_irq_cpu = raw_smp_processor_id();
1709 
1710 		efx_for_each_channel(channel, efx) {
1711 			if (queues & 1)
1712 				efx_schedule_channel_irq(channel);
1713 			queues >>= 1;
1714 		}
1715 	}
1716 
1717 	netif_vdbg(efx, intr, efx->net_dev,
1718 		   "IRQ %d on CPU %d status " EFX_DWORD_FMT "\n",
1719 		   irq, raw_smp_processor_id(), EFX_DWORD_VAL(reg));
1720 
1721 	return IRQ_HANDLED;
1722 }
1723 
1724 static void efx_ef10_irq_test_generate(struct efx_nic *efx)
1725 {
1726 	MCDI_DECLARE_BUF(inbuf, MC_CMD_TRIGGER_INTERRUPT_IN_LEN);
1727 
1728 	BUILD_BUG_ON(MC_CMD_TRIGGER_INTERRUPT_OUT_LEN != 0);
1729 
1730 	MCDI_SET_DWORD(inbuf, TRIGGER_INTERRUPT_IN_INTR_LEVEL, efx->irq_level);
1731 	(void) efx_mcdi_rpc(efx, MC_CMD_TRIGGER_INTERRUPT,
1732 			    inbuf, sizeof(inbuf), NULL, 0, NULL);
1733 }
1734 
1735 static int efx_ef10_tx_probe(struct efx_tx_queue *tx_queue)
1736 {
1737 	return efx_nic_alloc_buffer(tx_queue->efx, &tx_queue->txd.buf,
1738 				    (tx_queue->ptr_mask + 1) *
1739 				    sizeof(efx_qword_t),
1740 				    GFP_KERNEL);
1741 }
1742 
1743 /* This writes to the TX_DESC_WPTR and also pushes data */
1744 static inline void efx_ef10_push_tx_desc(struct efx_tx_queue *tx_queue,
1745 					 const efx_qword_t *txd)
1746 {
1747 	unsigned int write_ptr;
1748 	efx_oword_t reg;
1749 
1750 	write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
1751 	EFX_POPULATE_OWORD_1(reg, ERF_DZ_TX_DESC_WPTR, write_ptr);
1752 	reg.qword[0] = *txd;
1753 	efx_writeo_page(tx_queue->efx, &reg,
1754 			ER_DZ_TX_DESC_UPD, tx_queue->queue);
1755 }
1756 
1757 static void efx_ef10_tx_init(struct efx_tx_queue *tx_queue)
1758 {
1759 	MCDI_DECLARE_BUF(inbuf, MC_CMD_INIT_TXQ_IN_LEN(EFX_MAX_DMAQ_SIZE * 8 /
1760 						       EFX_BUF_SIZE));
1761 	bool csum_offload = tx_queue->queue & EFX_TXQ_TYPE_OFFLOAD;
1762 	size_t entries = tx_queue->txd.buf.len / EFX_BUF_SIZE;
1763 	struct efx_channel *channel = tx_queue->channel;
1764 	struct efx_nic *efx = tx_queue->efx;
1765 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1766 	size_t inlen;
1767 	dma_addr_t dma_addr;
1768 	efx_qword_t *txd;
1769 	int rc;
1770 	int i;
1771 	BUILD_BUG_ON(MC_CMD_INIT_TXQ_OUT_LEN != 0);
1772 
1773 	MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_SIZE, tx_queue->ptr_mask + 1);
1774 	MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_TARGET_EVQ, channel->channel);
1775 	MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_LABEL, tx_queue->queue);
1776 	MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_INSTANCE, tx_queue->queue);
1777 	MCDI_POPULATE_DWORD_2(inbuf, INIT_TXQ_IN_FLAGS,
1778 			      INIT_TXQ_IN_FLAG_IP_CSUM_DIS, !csum_offload,
1779 			      INIT_TXQ_IN_FLAG_TCP_CSUM_DIS, !csum_offload);
1780 	MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_OWNER_ID, 0);
1781 	MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_PORT_ID, nic_data->vport_id);
1782 
1783 	dma_addr = tx_queue->txd.buf.dma_addr;
1784 
1785 	netif_dbg(efx, hw, efx->net_dev, "pushing TXQ %d. %zu entries (%llx)\n",
1786 		  tx_queue->queue, entries, (u64)dma_addr);
1787 
1788 	for (i = 0; i < entries; ++i) {
1789 		MCDI_SET_ARRAY_QWORD(inbuf, INIT_TXQ_IN_DMA_ADDR, i, dma_addr);
1790 		dma_addr += EFX_BUF_SIZE;
1791 	}
1792 
1793 	inlen = MC_CMD_INIT_TXQ_IN_LEN(entries);
1794 
1795 	rc = efx_mcdi_rpc(efx, MC_CMD_INIT_TXQ, inbuf, inlen,
1796 			  NULL, 0, NULL);
1797 	if (rc)
1798 		goto fail;
1799 
1800 	/* A previous user of this TX queue might have set us up the
1801 	 * bomb by writing a descriptor to the TX push collector but
1802 	 * not the doorbell.  (Each collector belongs to a port, not a
1803 	 * queue or function, so cannot easily be reset.)  We must
1804 	 * attempt to push a no-op descriptor in its place.
1805 	 */
1806 	tx_queue->buffer[0].flags = EFX_TX_BUF_OPTION;
1807 	tx_queue->insert_count = 1;
1808 	txd = efx_tx_desc(tx_queue, 0);
1809 	EFX_POPULATE_QWORD_4(*txd,
1810 			     ESF_DZ_TX_DESC_IS_OPT, true,
1811 			     ESF_DZ_TX_OPTION_TYPE,
1812 			     ESE_DZ_TX_OPTION_DESC_CRC_CSUM,
1813 			     ESF_DZ_TX_OPTION_UDP_TCP_CSUM, csum_offload,
1814 			     ESF_DZ_TX_OPTION_IP_CSUM, csum_offload);
1815 	tx_queue->write_count = 1;
1816 
1817 	if (nic_data->datapath_caps &
1818 	    (1 << MC_CMD_GET_CAPABILITIES_OUT_TX_TSO_LBN)) {
1819 		tx_queue->tso_version = 1;
1820 	}
1821 
1822 	wmb();
1823 	efx_ef10_push_tx_desc(tx_queue, txd);
1824 
1825 	return;
1826 
1827 fail:
1828 	netdev_WARN(efx->net_dev, "failed to initialise TXQ %d\n",
1829 		    tx_queue->queue);
1830 }
1831 
1832 static void efx_ef10_tx_fini(struct efx_tx_queue *tx_queue)
1833 {
1834 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FINI_TXQ_IN_LEN);
1835 	MCDI_DECLARE_BUF_ERR(outbuf);
1836 	struct efx_nic *efx = tx_queue->efx;
1837 	size_t outlen;
1838 	int rc;
1839 
1840 	MCDI_SET_DWORD(inbuf, FINI_TXQ_IN_INSTANCE,
1841 		       tx_queue->queue);
1842 
1843 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FINI_TXQ, inbuf, sizeof(inbuf),
1844 			  outbuf, sizeof(outbuf), &outlen);
1845 
1846 	if (rc && rc != -EALREADY)
1847 		goto fail;
1848 
1849 	return;
1850 
1851 fail:
1852 	efx_mcdi_display_error(efx, MC_CMD_FINI_TXQ, MC_CMD_FINI_TXQ_IN_LEN,
1853 			       outbuf, outlen, rc);
1854 }
1855 
1856 static void efx_ef10_tx_remove(struct efx_tx_queue *tx_queue)
1857 {
1858 	efx_nic_free_buffer(tx_queue->efx, &tx_queue->txd.buf);
1859 }
1860 
1861 /* This writes to the TX_DESC_WPTR; write pointer for TX descriptor ring */
1862 static inline void efx_ef10_notify_tx_desc(struct efx_tx_queue *tx_queue)
1863 {
1864 	unsigned int write_ptr;
1865 	efx_dword_t reg;
1866 
1867 	write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
1868 	EFX_POPULATE_DWORD_1(reg, ERF_DZ_TX_DESC_WPTR_DWORD, write_ptr);
1869 	efx_writed_page(tx_queue->efx, &reg,
1870 			ER_DZ_TX_DESC_UPD_DWORD, tx_queue->queue);
1871 }
1872 
1873 static void efx_ef10_tx_write(struct efx_tx_queue *tx_queue)
1874 {
1875 	unsigned int old_write_count = tx_queue->write_count;
1876 	struct efx_tx_buffer *buffer;
1877 	unsigned int write_ptr;
1878 	efx_qword_t *txd;
1879 
1880 	tx_queue->xmit_more_available = false;
1881 	if (unlikely(tx_queue->write_count == tx_queue->insert_count))
1882 		return;
1883 
1884 	do {
1885 		write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
1886 		buffer = &tx_queue->buffer[write_ptr];
1887 		txd = efx_tx_desc(tx_queue, write_ptr);
1888 		++tx_queue->write_count;
1889 
1890 		/* Create TX descriptor ring entry */
1891 		if (buffer->flags & EFX_TX_BUF_OPTION) {
1892 			*txd = buffer->option;
1893 		} else {
1894 			BUILD_BUG_ON(EFX_TX_BUF_CONT != 1);
1895 			EFX_POPULATE_QWORD_3(
1896 				*txd,
1897 				ESF_DZ_TX_KER_CONT,
1898 				buffer->flags & EFX_TX_BUF_CONT,
1899 				ESF_DZ_TX_KER_BYTE_CNT, buffer->len,
1900 				ESF_DZ_TX_KER_BUF_ADDR, buffer->dma_addr);
1901 		}
1902 	} while (tx_queue->write_count != tx_queue->insert_count);
1903 
1904 	wmb(); /* Ensure descriptors are written before they are fetched */
1905 
1906 	if (efx_nic_may_push_tx_desc(tx_queue, old_write_count)) {
1907 		txd = efx_tx_desc(tx_queue,
1908 				  old_write_count & tx_queue->ptr_mask);
1909 		efx_ef10_push_tx_desc(tx_queue, txd);
1910 		++tx_queue->pushes;
1911 	} else {
1912 		efx_ef10_notify_tx_desc(tx_queue);
1913 	}
1914 }
1915 
1916 static int efx_ef10_alloc_rss_context(struct efx_nic *efx, u32 *context,
1917 				      bool exclusive, unsigned *context_size)
1918 {
1919 	MCDI_DECLARE_BUF(inbuf, MC_CMD_RSS_CONTEXT_ALLOC_IN_LEN);
1920 	MCDI_DECLARE_BUF(outbuf, MC_CMD_RSS_CONTEXT_ALLOC_OUT_LEN);
1921 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
1922 	size_t outlen;
1923 	int rc;
1924 	u32 alloc_type = exclusive ?
1925 				MC_CMD_RSS_CONTEXT_ALLOC_IN_TYPE_EXCLUSIVE :
1926 				MC_CMD_RSS_CONTEXT_ALLOC_IN_TYPE_SHARED;
1927 	unsigned rss_spread = exclusive ?
1928 				efx->rss_spread :
1929 				min(rounddown_pow_of_two(efx->rss_spread),
1930 				    EFX_EF10_MAX_SHARED_RSS_CONTEXT_SIZE);
1931 
1932 	if (!exclusive && rss_spread == 1) {
1933 		*context = EFX_EF10_RSS_CONTEXT_INVALID;
1934 		if (context_size)
1935 			*context_size = 1;
1936 		return 0;
1937 	}
1938 
1939 	if (nic_data->datapath_caps &
1940 	    1 << MC_CMD_GET_CAPABILITIES_OUT_RX_RSS_LIMITED_LBN)
1941 		return -EOPNOTSUPP;
1942 
1943 	MCDI_SET_DWORD(inbuf, RSS_CONTEXT_ALLOC_IN_UPSTREAM_PORT_ID,
1944 		       nic_data->vport_id);
1945 	MCDI_SET_DWORD(inbuf, RSS_CONTEXT_ALLOC_IN_TYPE, alloc_type);
1946 	MCDI_SET_DWORD(inbuf, RSS_CONTEXT_ALLOC_IN_NUM_QUEUES, rss_spread);
1947 
1948 	rc = efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_ALLOC, inbuf, sizeof(inbuf),
1949 		outbuf, sizeof(outbuf), &outlen);
1950 	if (rc != 0)
1951 		return rc;
1952 
1953 	if (outlen < MC_CMD_RSS_CONTEXT_ALLOC_OUT_LEN)
1954 		return -EIO;
1955 
1956 	*context = MCDI_DWORD(outbuf, RSS_CONTEXT_ALLOC_OUT_RSS_CONTEXT_ID);
1957 
1958 	if (context_size)
1959 		*context_size = rss_spread;
1960 
1961 	return 0;
1962 }
1963 
1964 static void efx_ef10_free_rss_context(struct efx_nic *efx, u32 context)
1965 {
1966 	MCDI_DECLARE_BUF(inbuf, MC_CMD_RSS_CONTEXT_FREE_IN_LEN);
1967 	int rc;
1968 
1969 	MCDI_SET_DWORD(inbuf, RSS_CONTEXT_FREE_IN_RSS_CONTEXT_ID,
1970 		       context);
1971 
1972 	rc = efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_FREE, inbuf, sizeof(inbuf),
1973 			    NULL, 0, NULL);
1974 	WARN_ON(rc != 0);
1975 }
1976 
1977 static int efx_ef10_populate_rss_table(struct efx_nic *efx, u32 context,
1978 				       const u32 *rx_indir_table)
1979 {
1980 	MCDI_DECLARE_BUF(tablebuf, MC_CMD_RSS_CONTEXT_SET_TABLE_IN_LEN);
1981 	MCDI_DECLARE_BUF(keybuf, MC_CMD_RSS_CONTEXT_SET_KEY_IN_LEN);
1982 	int i, rc;
1983 
1984 	MCDI_SET_DWORD(tablebuf, RSS_CONTEXT_SET_TABLE_IN_RSS_CONTEXT_ID,
1985 		       context);
1986 	BUILD_BUG_ON(ARRAY_SIZE(efx->rx_indir_table) !=
1987 		     MC_CMD_RSS_CONTEXT_SET_TABLE_IN_INDIRECTION_TABLE_LEN);
1988 
1989 	for (i = 0; i < ARRAY_SIZE(efx->rx_indir_table); ++i)
1990 		MCDI_PTR(tablebuf,
1991 			 RSS_CONTEXT_SET_TABLE_IN_INDIRECTION_TABLE)[i] =
1992 				(u8) rx_indir_table[i];
1993 
1994 	rc = efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_SET_TABLE, tablebuf,
1995 			  sizeof(tablebuf), NULL, 0, NULL);
1996 	if (rc != 0)
1997 		return rc;
1998 
1999 	MCDI_SET_DWORD(keybuf, RSS_CONTEXT_SET_KEY_IN_RSS_CONTEXT_ID,
2000 		       context);
2001 	BUILD_BUG_ON(ARRAY_SIZE(efx->rx_hash_key) !=
2002 		     MC_CMD_RSS_CONTEXT_SET_KEY_IN_TOEPLITZ_KEY_LEN);
2003 	for (i = 0; i < ARRAY_SIZE(efx->rx_hash_key); ++i)
2004 		MCDI_PTR(keybuf, RSS_CONTEXT_SET_KEY_IN_TOEPLITZ_KEY)[i] =
2005 			efx->rx_hash_key[i];
2006 
2007 	return efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_SET_KEY, keybuf,
2008 			    sizeof(keybuf), NULL, 0, NULL);
2009 }
2010 
2011 static void efx_ef10_rx_free_indir_table(struct efx_nic *efx)
2012 {
2013 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2014 
2015 	if (nic_data->rx_rss_context != EFX_EF10_RSS_CONTEXT_INVALID)
2016 		efx_ef10_free_rss_context(efx, nic_data->rx_rss_context);
2017 	nic_data->rx_rss_context = EFX_EF10_RSS_CONTEXT_INVALID;
2018 }
2019 
2020 static int efx_ef10_rx_push_shared_rss_config(struct efx_nic *efx,
2021 					      unsigned *context_size)
2022 {
2023 	u32 new_rx_rss_context;
2024 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2025 	int rc = efx_ef10_alloc_rss_context(efx, &new_rx_rss_context,
2026 					    false, context_size);
2027 
2028 	if (rc != 0)
2029 		return rc;
2030 
2031 	nic_data->rx_rss_context = new_rx_rss_context;
2032 	nic_data->rx_rss_context_exclusive = false;
2033 	efx_set_default_rx_indir_table(efx);
2034 	return 0;
2035 }
2036 
2037 static int efx_ef10_rx_push_exclusive_rss_config(struct efx_nic *efx,
2038 						 const u32 *rx_indir_table)
2039 {
2040 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2041 	int rc;
2042 	u32 new_rx_rss_context;
2043 
2044 	if (nic_data->rx_rss_context == EFX_EF10_RSS_CONTEXT_INVALID ||
2045 	    !nic_data->rx_rss_context_exclusive) {
2046 		rc = efx_ef10_alloc_rss_context(efx, &new_rx_rss_context,
2047 						true, NULL);
2048 		if (rc == -EOPNOTSUPP)
2049 			return rc;
2050 		else if (rc != 0)
2051 			goto fail1;
2052 	} else {
2053 		new_rx_rss_context = nic_data->rx_rss_context;
2054 	}
2055 
2056 	rc = efx_ef10_populate_rss_table(efx, new_rx_rss_context,
2057 					 rx_indir_table);
2058 	if (rc != 0)
2059 		goto fail2;
2060 
2061 	if (nic_data->rx_rss_context != new_rx_rss_context)
2062 		efx_ef10_rx_free_indir_table(efx);
2063 	nic_data->rx_rss_context = new_rx_rss_context;
2064 	nic_data->rx_rss_context_exclusive = true;
2065 	if (rx_indir_table != efx->rx_indir_table)
2066 		memcpy(efx->rx_indir_table, rx_indir_table,
2067 		       sizeof(efx->rx_indir_table));
2068 	return 0;
2069 
2070 fail2:
2071 	if (new_rx_rss_context != nic_data->rx_rss_context)
2072 		efx_ef10_free_rss_context(efx, new_rx_rss_context);
2073 fail1:
2074 	netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
2075 	return rc;
2076 }
2077 
2078 static int efx_ef10_pf_rx_push_rss_config(struct efx_nic *efx, bool user,
2079 					  const u32 *rx_indir_table)
2080 {
2081 	int rc;
2082 
2083 	if (efx->rss_spread == 1)
2084 		return 0;
2085 
2086 	rc = efx_ef10_rx_push_exclusive_rss_config(efx, rx_indir_table);
2087 
2088 	if (rc == -ENOBUFS && !user) {
2089 		unsigned context_size;
2090 		bool mismatch = false;
2091 		size_t i;
2092 
2093 		for (i = 0; i < ARRAY_SIZE(efx->rx_indir_table) && !mismatch;
2094 		     i++)
2095 			mismatch = rx_indir_table[i] !=
2096 				ethtool_rxfh_indir_default(i, efx->rss_spread);
2097 
2098 		rc = efx_ef10_rx_push_shared_rss_config(efx, &context_size);
2099 		if (rc == 0) {
2100 			if (context_size != efx->rss_spread)
2101 				netif_warn(efx, probe, efx->net_dev,
2102 					   "Could not allocate an exclusive RSS"
2103 					   " context; allocated a shared one of"
2104 					   " different size."
2105 					   " Wanted %u, got %u.\n",
2106 					   efx->rss_spread, context_size);
2107 			else if (mismatch)
2108 				netif_warn(efx, probe, efx->net_dev,
2109 					   "Could not allocate an exclusive RSS"
2110 					   " context; allocated a shared one but"
2111 					   " could not apply custom"
2112 					   " indirection.\n");
2113 			else
2114 				netif_info(efx, probe, efx->net_dev,
2115 					   "Could not allocate an exclusive RSS"
2116 					   " context; allocated a shared one.\n");
2117 		}
2118 	}
2119 	return rc;
2120 }
2121 
2122 static int efx_ef10_vf_rx_push_rss_config(struct efx_nic *efx, bool user,
2123 					  const u32 *rx_indir_table
2124 					  __attribute__ ((unused)))
2125 {
2126 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2127 
2128 	if (user)
2129 		return -EOPNOTSUPP;
2130 	if (nic_data->rx_rss_context != EFX_EF10_RSS_CONTEXT_INVALID)
2131 		return 0;
2132 	return efx_ef10_rx_push_shared_rss_config(efx, NULL);
2133 }
2134 
2135 static int efx_ef10_rx_probe(struct efx_rx_queue *rx_queue)
2136 {
2137 	return efx_nic_alloc_buffer(rx_queue->efx, &rx_queue->rxd.buf,
2138 				    (rx_queue->ptr_mask + 1) *
2139 				    sizeof(efx_qword_t),
2140 				    GFP_KERNEL);
2141 }
2142 
2143 static void efx_ef10_rx_init(struct efx_rx_queue *rx_queue)
2144 {
2145 	MCDI_DECLARE_BUF(inbuf,
2146 			 MC_CMD_INIT_RXQ_IN_LEN(EFX_MAX_DMAQ_SIZE * 8 /
2147 						EFX_BUF_SIZE));
2148 	struct efx_channel *channel = efx_rx_queue_channel(rx_queue);
2149 	size_t entries = rx_queue->rxd.buf.len / EFX_BUF_SIZE;
2150 	struct efx_nic *efx = rx_queue->efx;
2151 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2152 	size_t inlen;
2153 	dma_addr_t dma_addr;
2154 	int rc;
2155 	int i;
2156 	BUILD_BUG_ON(MC_CMD_INIT_RXQ_OUT_LEN != 0);
2157 
2158 	rx_queue->scatter_n = 0;
2159 	rx_queue->scatter_len = 0;
2160 
2161 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_SIZE, rx_queue->ptr_mask + 1);
2162 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_TARGET_EVQ, channel->channel);
2163 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_LABEL, efx_rx_queue_index(rx_queue));
2164 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_INSTANCE,
2165 		       efx_rx_queue_index(rx_queue));
2166 	MCDI_POPULATE_DWORD_2(inbuf, INIT_RXQ_IN_FLAGS,
2167 			      INIT_RXQ_IN_FLAG_PREFIX, 1,
2168 			      INIT_RXQ_IN_FLAG_TIMESTAMP, 1);
2169 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_OWNER_ID, 0);
2170 	MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_PORT_ID, nic_data->vport_id);
2171 
2172 	dma_addr = rx_queue->rxd.buf.dma_addr;
2173 
2174 	netif_dbg(efx, hw, efx->net_dev, "pushing RXQ %d. %zu entries (%llx)\n",
2175 		  efx_rx_queue_index(rx_queue), entries, (u64)dma_addr);
2176 
2177 	for (i = 0; i < entries; ++i) {
2178 		MCDI_SET_ARRAY_QWORD(inbuf, INIT_RXQ_IN_DMA_ADDR, i, dma_addr);
2179 		dma_addr += EFX_BUF_SIZE;
2180 	}
2181 
2182 	inlen = MC_CMD_INIT_RXQ_IN_LEN(entries);
2183 
2184 	rc = efx_mcdi_rpc(efx, MC_CMD_INIT_RXQ, inbuf, inlen,
2185 			  NULL, 0, NULL);
2186 	if (rc)
2187 		netdev_WARN(efx->net_dev, "failed to initialise RXQ %d\n",
2188 			    efx_rx_queue_index(rx_queue));
2189 }
2190 
2191 static void efx_ef10_rx_fini(struct efx_rx_queue *rx_queue)
2192 {
2193 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FINI_RXQ_IN_LEN);
2194 	MCDI_DECLARE_BUF_ERR(outbuf);
2195 	struct efx_nic *efx = rx_queue->efx;
2196 	size_t outlen;
2197 	int rc;
2198 
2199 	MCDI_SET_DWORD(inbuf, FINI_RXQ_IN_INSTANCE,
2200 		       efx_rx_queue_index(rx_queue));
2201 
2202 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FINI_RXQ, inbuf, sizeof(inbuf),
2203 			  outbuf, sizeof(outbuf), &outlen);
2204 
2205 	if (rc && rc != -EALREADY)
2206 		goto fail;
2207 
2208 	return;
2209 
2210 fail:
2211 	efx_mcdi_display_error(efx, MC_CMD_FINI_RXQ, MC_CMD_FINI_RXQ_IN_LEN,
2212 			       outbuf, outlen, rc);
2213 }
2214 
2215 static void efx_ef10_rx_remove(struct efx_rx_queue *rx_queue)
2216 {
2217 	efx_nic_free_buffer(rx_queue->efx, &rx_queue->rxd.buf);
2218 }
2219 
2220 /* This creates an entry in the RX descriptor queue */
2221 static inline void
2222 efx_ef10_build_rx_desc(struct efx_rx_queue *rx_queue, unsigned int index)
2223 {
2224 	struct efx_rx_buffer *rx_buf;
2225 	efx_qword_t *rxd;
2226 
2227 	rxd = efx_rx_desc(rx_queue, index);
2228 	rx_buf = efx_rx_buffer(rx_queue, index);
2229 	EFX_POPULATE_QWORD_2(*rxd,
2230 			     ESF_DZ_RX_KER_BYTE_CNT, rx_buf->len,
2231 			     ESF_DZ_RX_KER_BUF_ADDR, rx_buf->dma_addr);
2232 }
2233 
2234 static void efx_ef10_rx_write(struct efx_rx_queue *rx_queue)
2235 {
2236 	struct efx_nic *efx = rx_queue->efx;
2237 	unsigned int write_count;
2238 	efx_dword_t reg;
2239 
2240 	/* Firmware requires that RX_DESC_WPTR be a multiple of 8 */
2241 	write_count = rx_queue->added_count & ~7;
2242 	if (rx_queue->notified_count == write_count)
2243 		return;
2244 
2245 	do
2246 		efx_ef10_build_rx_desc(
2247 			rx_queue,
2248 			rx_queue->notified_count & rx_queue->ptr_mask);
2249 	while (++rx_queue->notified_count != write_count);
2250 
2251 	wmb();
2252 	EFX_POPULATE_DWORD_1(reg, ERF_DZ_RX_DESC_WPTR,
2253 			     write_count & rx_queue->ptr_mask);
2254 	efx_writed_page(efx, &reg, ER_DZ_RX_DESC_UPD,
2255 			efx_rx_queue_index(rx_queue));
2256 }
2257 
2258 static efx_mcdi_async_completer efx_ef10_rx_defer_refill_complete;
2259 
2260 static void efx_ef10_rx_defer_refill(struct efx_rx_queue *rx_queue)
2261 {
2262 	struct efx_channel *channel = efx_rx_queue_channel(rx_queue);
2263 	MCDI_DECLARE_BUF(inbuf, MC_CMD_DRIVER_EVENT_IN_LEN);
2264 	efx_qword_t event;
2265 
2266 	EFX_POPULATE_QWORD_2(event,
2267 			     ESF_DZ_EV_CODE, EFX_EF10_DRVGEN_EV,
2268 			     ESF_DZ_EV_DATA, EFX_EF10_REFILL);
2269 
2270 	MCDI_SET_DWORD(inbuf, DRIVER_EVENT_IN_EVQ, channel->channel);
2271 
2272 	/* MCDI_SET_QWORD is not appropriate here since EFX_POPULATE_* has
2273 	 * already swapped the data to little-endian order.
2274 	 */
2275 	memcpy(MCDI_PTR(inbuf, DRIVER_EVENT_IN_DATA), &event.u64[0],
2276 	       sizeof(efx_qword_t));
2277 
2278 	efx_mcdi_rpc_async(channel->efx, MC_CMD_DRIVER_EVENT,
2279 			   inbuf, sizeof(inbuf), 0,
2280 			   efx_ef10_rx_defer_refill_complete, 0);
2281 }
2282 
2283 static void
2284 efx_ef10_rx_defer_refill_complete(struct efx_nic *efx, unsigned long cookie,
2285 				  int rc, efx_dword_t *outbuf,
2286 				  size_t outlen_actual)
2287 {
2288 	/* nothing to do */
2289 }
2290 
2291 static int efx_ef10_ev_probe(struct efx_channel *channel)
2292 {
2293 	return efx_nic_alloc_buffer(channel->efx, &channel->eventq.buf,
2294 				    (channel->eventq_mask + 1) *
2295 				    sizeof(efx_qword_t),
2296 				    GFP_KERNEL);
2297 }
2298 
2299 static void efx_ef10_ev_fini(struct efx_channel *channel)
2300 {
2301 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FINI_EVQ_IN_LEN);
2302 	MCDI_DECLARE_BUF_ERR(outbuf);
2303 	struct efx_nic *efx = channel->efx;
2304 	size_t outlen;
2305 	int rc;
2306 
2307 	MCDI_SET_DWORD(inbuf, FINI_EVQ_IN_INSTANCE, channel->channel);
2308 
2309 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FINI_EVQ, inbuf, sizeof(inbuf),
2310 			  outbuf, sizeof(outbuf), &outlen);
2311 
2312 	if (rc && rc != -EALREADY)
2313 		goto fail;
2314 
2315 	return;
2316 
2317 fail:
2318 	efx_mcdi_display_error(efx, MC_CMD_FINI_EVQ, MC_CMD_FINI_EVQ_IN_LEN,
2319 			       outbuf, outlen, rc);
2320 }
2321 
2322 static int efx_ef10_ev_init(struct efx_channel *channel)
2323 {
2324 	MCDI_DECLARE_BUF(inbuf,
2325 			 MC_CMD_INIT_EVQ_IN_LEN(EFX_MAX_EVQ_SIZE * 8 /
2326 						EFX_BUF_SIZE));
2327 	MCDI_DECLARE_BUF(outbuf, MC_CMD_INIT_EVQ_OUT_LEN);
2328 	size_t entries = channel->eventq.buf.len / EFX_BUF_SIZE;
2329 	struct efx_nic *efx = channel->efx;
2330 	struct efx_ef10_nic_data *nic_data;
2331 	bool supports_rx_merge;
2332 	size_t inlen, outlen;
2333 	unsigned int enabled, implemented;
2334 	dma_addr_t dma_addr;
2335 	int rc;
2336 	int i;
2337 
2338 	nic_data = efx->nic_data;
2339 	supports_rx_merge =
2340 		!!(nic_data->datapath_caps &
2341 		   1 << MC_CMD_GET_CAPABILITIES_OUT_RX_BATCHING_LBN);
2342 
2343 	/* Fill event queue with all ones (i.e. empty events) */
2344 	memset(channel->eventq.buf.addr, 0xff, channel->eventq.buf.len);
2345 
2346 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_SIZE, channel->eventq_mask + 1);
2347 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_INSTANCE, channel->channel);
2348 	/* INIT_EVQ expects index in vector table, not absolute */
2349 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_IRQ_NUM, channel->channel);
2350 	MCDI_POPULATE_DWORD_4(inbuf, INIT_EVQ_IN_FLAGS,
2351 			      INIT_EVQ_IN_FLAG_INTERRUPTING, 1,
2352 			      INIT_EVQ_IN_FLAG_RX_MERGE, 1,
2353 			      INIT_EVQ_IN_FLAG_TX_MERGE, 1,
2354 			      INIT_EVQ_IN_FLAG_CUT_THRU, !supports_rx_merge);
2355 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_TMR_MODE,
2356 		       MC_CMD_INIT_EVQ_IN_TMR_MODE_DIS);
2357 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_TMR_LOAD, 0);
2358 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_TMR_RELOAD, 0);
2359 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_COUNT_MODE,
2360 		       MC_CMD_INIT_EVQ_IN_COUNT_MODE_DIS);
2361 	MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_COUNT_THRSHLD, 0);
2362 
2363 	dma_addr = channel->eventq.buf.dma_addr;
2364 	for (i = 0; i < entries; ++i) {
2365 		MCDI_SET_ARRAY_QWORD(inbuf, INIT_EVQ_IN_DMA_ADDR, i, dma_addr);
2366 		dma_addr += EFX_BUF_SIZE;
2367 	}
2368 
2369 	inlen = MC_CMD_INIT_EVQ_IN_LEN(entries);
2370 
2371 	rc = efx_mcdi_rpc(efx, MC_CMD_INIT_EVQ, inbuf, inlen,
2372 			  outbuf, sizeof(outbuf), &outlen);
2373 	/* IRQ return is ignored */
2374 	if (channel->channel || rc)
2375 		return rc;
2376 
2377 	/* Successfully created event queue on channel 0 */
2378 	rc = efx_mcdi_get_workarounds(efx, &implemented, &enabled);
2379 	if (rc == -ENOSYS) {
2380 		/* GET_WORKAROUNDS was implemented before the bug26807
2381 		 * workaround, thus the latter must be unavailable in this fw
2382 		 */
2383 		nic_data->workaround_26807 = false;
2384 		rc = 0;
2385 	} else if (rc) {
2386 		goto fail;
2387 	} else {
2388 		nic_data->workaround_26807 =
2389 			!!(enabled & MC_CMD_GET_WORKAROUNDS_OUT_BUG26807);
2390 
2391 		if (implemented & MC_CMD_GET_WORKAROUNDS_OUT_BUG26807 &&
2392 		    !nic_data->workaround_26807) {
2393 			unsigned int flags;
2394 
2395 			rc = efx_mcdi_set_workaround(efx,
2396 						     MC_CMD_WORKAROUND_BUG26807,
2397 						     true, &flags);
2398 
2399 			if (!rc) {
2400 				if (flags &
2401 				    1 << MC_CMD_WORKAROUND_EXT_OUT_FLR_DONE_LBN) {
2402 					netif_info(efx, drv, efx->net_dev,
2403 						   "other functions on NIC have been reset\n");
2404 
2405 					/* With MCFW v4.6.x and earlier, the
2406 					 * boot count will have incremented,
2407 					 * so re-read the warm_boot_count
2408 					 * value now to ensure this function
2409 					 * doesn't think it has changed next
2410 					 * time it checks.
2411 					 */
2412 					rc = efx_ef10_get_warm_boot_count(efx);
2413 					if (rc >= 0) {
2414 						nic_data->warm_boot_count = rc;
2415 						rc = 0;
2416 					}
2417 				}
2418 				nic_data->workaround_26807 = true;
2419 			} else if (rc == -EPERM) {
2420 				rc = 0;
2421 			}
2422 		}
2423 	}
2424 
2425 	if (!rc)
2426 		return 0;
2427 
2428 fail:
2429 	efx_ef10_ev_fini(channel);
2430 	return rc;
2431 }
2432 
2433 static void efx_ef10_ev_remove(struct efx_channel *channel)
2434 {
2435 	efx_nic_free_buffer(channel->efx, &channel->eventq.buf);
2436 }
2437 
2438 static void efx_ef10_handle_rx_wrong_queue(struct efx_rx_queue *rx_queue,
2439 					   unsigned int rx_queue_label)
2440 {
2441 	struct efx_nic *efx = rx_queue->efx;
2442 
2443 	netif_info(efx, hw, efx->net_dev,
2444 		   "rx event arrived on queue %d labeled as queue %u\n",
2445 		   efx_rx_queue_index(rx_queue), rx_queue_label);
2446 
2447 	efx_schedule_reset(efx, RESET_TYPE_DISABLE);
2448 }
2449 
2450 static void
2451 efx_ef10_handle_rx_bad_lbits(struct efx_rx_queue *rx_queue,
2452 			     unsigned int actual, unsigned int expected)
2453 {
2454 	unsigned int dropped = (actual - expected) & rx_queue->ptr_mask;
2455 	struct efx_nic *efx = rx_queue->efx;
2456 
2457 	netif_info(efx, hw, efx->net_dev,
2458 		   "dropped %d events (index=%d expected=%d)\n",
2459 		   dropped, actual, expected);
2460 
2461 	efx_schedule_reset(efx, RESET_TYPE_DISABLE);
2462 }
2463 
2464 /* partially received RX was aborted. clean up. */
2465 static void efx_ef10_handle_rx_abort(struct efx_rx_queue *rx_queue)
2466 {
2467 	unsigned int rx_desc_ptr;
2468 
2469 	netif_dbg(rx_queue->efx, hw, rx_queue->efx->net_dev,
2470 		  "scattered RX aborted (dropping %u buffers)\n",
2471 		  rx_queue->scatter_n);
2472 
2473 	rx_desc_ptr = rx_queue->removed_count & rx_queue->ptr_mask;
2474 
2475 	efx_rx_packet(rx_queue, rx_desc_ptr, rx_queue->scatter_n,
2476 		      0, EFX_RX_PKT_DISCARD);
2477 
2478 	rx_queue->removed_count += rx_queue->scatter_n;
2479 	rx_queue->scatter_n = 0;
2480 	rx_queue->scatter_len = 0;
2481 	++efx_rx_queue_channel(rx_queue)->n_rx_nodesc_trunc;
2482 }
2483 
2484 static int efx_ef10_handle_rx_event(struct efx_channel *channel,
2485 				    const efx_qword_t *event)
2486 {
2487 	unsigned int rx_bytes, next_ptr_lbits, rx_queue_label, rx_l4_class;
2488 	unsigned int n_descs, n_packets, i;
2489 	struct efx_nic *efx = channel->efx;
2490 	struct efx_rx_queue *rx_queue;
2491 	bool rx_cont;
2492 	u16 flags = 0;
2493 
2494 	if (unlikely(ACCESS_ONCE(efx->reset_pending)))
2495 		return 0;
2496 
2497 	/* Basic packet information */
2498 	rx_bytes = EFX_QWORD_FIELD(*event, ESF_DZ_RX_BYTES);
2499 	next_ptr_lbits = EFX_QWORD_FIELD(*event, ESF_DZ_RX_DSC_PTR_LBITS);
2500 	rx_queue_label = EFX_QWORD_FIELD(*event, ESF_DZ_RX_QLABEL);
2501 	rx_l4_class = EFX_QWORD_FIELD(*event, ESF_DZ_RX_L4_CLASS);
2502 	rx_cont = EFX_QWORD_FIELD(*event, ESF_DZ_RX_CONT);
2503 
2504 	if (EFX_QWORD_FIELD(*event, ESF_DZ_RX_DROP_EVENT))
2505 		netdev_WARN(efx->net_dev, "saw RX_DROP_EVENT: event="
2506 			    EFX_QWORD_FMT "\n",
2507 			    EFX_QWORD_VAL(*event));
2508 
2509 	rx_queue = efx_channel_get_rx_queue(channel);
2510 
2511 	if (unlikely(rx_queue_label != efx_rx_queue_index(rx_queue)))
2512 		efx_ef10_handle_rx_wrong_queue(rx_queue, rx_queue_label);
2513 
2514 	n_descs = ((next_ptr_lbits - rx_queue->removed_count) &
2515 		   ((1 << ESF_DZ_RX_DSC_PTR_LBITS_WIDTH) - 1));
2516 
2517 	if (n_descs != rx_queue->scatter_n + 1) {
2518 		struct efx_ef10_nic_data *nic_data = efx->nic_data;
2519 
2520 		/* detect rx abort */
2521 		if (unlikely(n_descs == rx_queue->scatter_n)) {
2522 			if (rx_queue->scatter_n == 0 || rx_bytes != 0)
2523 				netdev_WARN(efx->net_dev,
2524 					    "invalid RX abort: scatter_n=%u event="
2525 					    EFX_QWORD_FMT "\n",
2526 					    rx_queue->scatter_n,
2527 					    EFX_QWORD_VAL(*event));
2528 			efx_ef10_handle_rx_abort(rx_queue);
2529 			return 0;
2530 		}
2531 
2532 		/* Check that RX completion merging is valid, i.e.
2533 		 * the current firmware supports it and this is a
2534 		 * non-scattered packet.
2535 		 */
2536 		if (!(nic_data->datapath_caps &
2537 		      (1 << MC_CMD_GET_CAPABILITIES_OUT_RX_BATCHING_LBN)) ||
2538 		    rx_queue->scatter_n != 0 || rx_cont) {
2539 			efx_ef10_handle_rx_bad_lbits(
2540 				rx_queue, next_ptr_lbits,
2541 				(rx_queue->removed_count +
2542 				 rx_queue->scatter_n + 1) &
2543 				((1 << ESF_DZ_RX_DSC_PTR_LBITS_WIDTH) - 1));
2544 			return 0;
2545 		}
2546 
2547 		/* Merged completion for multiple non-scattered packets */
2548 		rx_queue->scatter_n = 1;
2549 		rx_queue->scatter_len = 0;
2550 		n_packets = n_descs;
2551 		++channel->n_rx_merge_events;
2552 		channel->n_rx_merge_packets += n_packets;
2553 		flags |= EFX_RX_PKT_PREFIX_LEN;
2554 	} else {
2555 		++rx_queue->scatter_n;
2556 		rx_queue->scatter_len += rx_bytes;
2557 		if (rx_cont)
2558 			return 0;
2559 		n_packets = 1;
2560 	}
2561 
2562 	if (unlikely(EFX_QWORD_FIELD(*event, ESF_DZ_RX_ECRC_ERR)))
2563 		flags |= EFX_RX_PKT_DISCARD;
2564 
2565 	if (unlikely(EFX_QWORD_FIELD(*event, ESF_DZ_RX_IPCKSUM_ERR))) {
2566 		channel->n_rx_ip_hdr_chksum_err += n_packets;
2567 	} else if (unlikely(EFX_QWORD_FIELD(*event,
2568 					    ESF_DZ_RX_TCPUDP_CKSUM_ERR))) {
2569 		channel->n_rx_tcp_udp_chksum_err += n_packets;
2570 	} else if (rx_l4_class == ESE_DZ_L4_CLASS_TCP ||
2571 		   rx_l4_class == ESE_DZ_L4_CLASS_UDP) {
2572 		flags |= EFX_RX_PKT_CSUMMED;
2573 	}
2574 
2575 	if (rx_l4_class == ESE_DZ_L4_CLASS_TCP)
2576 		flags |= EFX_RX_PKT_TCP;
2577 
2578 	channel->irq_mod_score += 2 * n_packets;
2579 
2580 	/* Handle received packet(s) */
2581 	for (i = 0; i < n_packets; i++) {
2582 		efx_rx_packet(rx_queue,
2583 			      rx_queue->removed_count & rx_queue->ptr_mask,
2584 			      rx_queue->scatter_n, rx_queue->scatter_len,
2585 			      flags);
2586 		rx_queue->removed_count += rx_queue->scatter_n;
2587 	}
2588 
2589 	rx_queue->scatter_n = 0;
2590 	rx_queue->scatter_len = 0;
2591 
2592 	return n_packets;
2593 }
2594 
2595 static int
2596 efx_ef10_handle_tx_event(struct efx_channel *channel, efx_qword_t *event)
2597 {
2598 	struct efx_nic *efx = channel->efx;
2599 	struct efx_tx_queue *tx_queue;
2600 	unsigned int tx_ev_desc_ptr;
2601 	unsigned int tx_ev_q_label;
2602 	int tx_descs = 0;
2603 
2604 	if (unlikely(ACCESS_ONCE(efx->reset_pending)))
2605 		return 0;
2606 
2607 	if (unlikely(EFX_QWORD_FIELD(*event, ESF_DZ_TX_DROP_EVENT)))
2608 		return 0;
2609 
2610 	/* Transmit completion */
2611 	tx_ev_desc_ptr = EFX_QWORD_FIELD(*event, ESF_DZ_TX_DESCR_INDX);
2612 	tx_ev_q_label = EFX_QWORD_FIELD(*event, ESF_DZ_TX_QLABEL);
2613 	tx_queue = efx_channel_get_tx_queue(channel,
2614 					    tx_ev_q_label % EFX_TXQ_TYPES);
2615 	tx_descs = ((tx_ev_desc_ptr + 1 - tx_queue->read_count) &
2616 		    tx_queue->ptr_mask);
2617 	efx_xmit_done(tx_queue, tx_ev_desc_ptr & tx_queue->ptr_mask);
2618 
2619 	return tx_descs;
2620 }
2621 
2622 static void
2623 efx_ef10_handle_driver_event(struct efx_channel *channel, efx_qword_t *event)
2624 {
2625 	struct efx_nic *efx = channel->efx;
2626 	int subcode;
2627 
2628 	subcode = EFX_QWORD_FIELD(*event, ESF_DZ_DRV_SUB_CODE);
2629 
2630 	switch (subcode) {
2631 	case ESE_DZ_DRV_TIMER_EV:
2632 	case ESE_DZ_DRV_WAKE_UP_EV:
2633 		break;
2634 	case ESE_DZ_DRV_START_UP_EV:
2635 		/* event queue init complete. ok. */
2636 		break;
2637 	default:
2638 		netif_err(efx, hw, efx->net_dev,
2639 			  "channel %d unknown driver event type %d"
2640 			  " (data " EFX_QWORD_FMT ")\n",
2641 			  channel->channel, subcode,
2642 			  EFX_QWORD_VAL(*event));
2643 
2644 	}
2645 }
2646 
2647 static void efx_ef10_handle_driver_generated_event(struct efx_channel *channel,
2648 						   efx_qword_t *event)
2649 {
2650 	struct efx_nic *efx = channel->efx;
2651 	u32 subcode;
2652 
2653 	subcode = EFX_QWORD_FIELD(*event, EFX_DWORD_0);
2654 
2655 	switch (subcode) {
2656 	case EFX_EF10_TEST:
2657 		channel->event_test_cpu = raw_smp_processor_id();
2658 		break;
2659 	case EFX_EF10_REFILL:
2660 		/* The queue must be empty, so we won't receive any rx
2661 		 * events, so efx_process_channel() won't refill the
2662 		 * queue. Refill it here
2663 		 */
2664 		efx_fast_push_rx_descriptors(&channel->rx_queue, true);
2665 		break;
2666 	default:
2667 		netif_err(efx, hw, efx->net_dev,
2668 			  "channel %d unknown driver event type %u"
2669 			  " (data " EFX_QWORD_FMT ")\n",
2670 			  channel->channel, (unsigned) subcode,
2671 			  EFX_QWORD_VAL(*event));
2672 	}
2673 }
2674 
2675 static int efx_ef10_ev_process(struct efx_channel *channel, int quota)
2676 {
2677 	struct efx_nic *efx = channel->efx;
2678 	efx_qword_t event, *p_event;
2679 	unsigned int read_ptr;
2680 	int ev_code;
2681 	int tx_descs = 0;
2682 	int spent = 0;
2683 
2684 	if (quota <= 0)
2685 		return spent;
2686 
2687 	read_ptr = channel->eventq_read_ptr;
2688 
2689 	for (;;) {
2690 		p_event = efx_event(channel, read_ptr);
2691 		event = *p_event;
2692 
2693 		if (!efx_event_present(&event))
2694 			break;
2695 
2696 		EFX_SET_QWORD(*p_event);
2697 
2698 		++read_ptr;
2699 
2700 		ev_code = EFX_QWORD_FIELD(event, ESF_DZ_EV_CODE);
2701 
2702 		netif_vdbg(efx, drv, efx->net_dev,
2703 			   "processing event on %d " EFX_QWORD_FMT "\n",
2704 			   channel->channel, EFX_QWORD_VAL(event));
2705 
2706 		switch (ev_code) {
2707 		case ESE_DZ_EV_CODE_MCDI_EV:
2708 			efx_mcdi_process_event(channel, &event);
2709 			break;
2710 		case ESE_DZ_EV_CODE_RX_EV:
2711 			spent += efx_ef10_handle_rx_event(channel, &event);
2712 			if (spent >= quota) {
2713 				/* XXX can we split a merged event to
2714 				 * avoid going over-quota?
2715 				 */
2716 				spent = quota;
2717 				goto out;
2718 			}
2719 			break;
2720 		case ESE_DZ_EV_CODE_TX_EV:
2721 			tx_descs += efx_ef10_handle_tx_event(channel, &event);
2722 			if (tx_descs > efx->txq_entries) {
2723 				spent = quota;
2724 				goto out;
2725 			} else if (++spent == quota) {
2726 				goto out;
2727 			}
2728 			break;
2729 		case ESE_DZ_EV_CODE_DRIVER_EV:
2730 			efx_ef10_handle_driver_event(channel, &event);
2731 			if (++spent == quota)
2732 				goto out;
2733 			break;
2734 		case EFX_EF10_DRVGEN_EV:
2735 			efx_ef10_handle_driver_generated_event(channel, &event);
2736 			break;
2737 		default:
2738 			netif_err(efx, hw, efx->net_dev,
2739 				  "channel %d unknown event type %d"
2740 				  " (data " EFX_QWORD_FMT ")\n",
2741 				  channel->channel, ev_code,
2742 				  EFX_QWORD_VAL(event));
2743 		}
2744 	}
2745 
2746 out:
2747 	channel->eventq_read_ptr = read_ptr;
2748 	return spent;
2749 }
2750 
2751 static void efx_ef10_ev_read_ack(struct efx_channel *channel)
2752 {
2753 	struct efx_nic *efx = channel->efx;
2754 	efx_dword_t rptr;
2755 
2756 	if (EFX_EF10_WORKAROUND_35388(efx)) {
2757 		BUILD_BUG_ON(EFX_MIN_EVQ_SIZE <
2758 			     (1 << ERF_DD_EVQ_IND_RPTR_WIDTH));
2759 		BUILD_BUG_ON(EFX_MAX_EVQ_SIZE >
2760 			     (1 << 2 * ERF_DD_EVQ_IND_RPTR_WIDTH));
2761 
2762 		EFX_POPULATE_DWORD_2(rptr, ERF_DD_EVQ_IND_RPTR_FLAGS,
2763 				     EFE_DD_EVQ_IND_RPTR_FLAGS_HIGH,
2764 				     ERF_DD_EVQ_IND_RPTR,
2765 				     (channel->eventq_read_ptr &
2766 				      channel->eventq_mask) >>
2767 				     ERF_DD_EVQ_IND_RPTR_WIDTH);
2768 		efx_writed_page(efx, &rptr, ER_DD_EVQ_INDIRECT,
2769 				channel->channel);
2770 		EFX_POPULATE_DWORD_2(rptr, ERF_DD_EVQ_IND_RPTR_FLAGS,
2771 				     EFE_DD_EVQ_IND_RPTR_FLAGS_LOW,
2772 				     ERF_DD_EVQ_IND_RPTR,
2773 				     channel->eventq_read_ptr &
2774 				     ((1 << ERF_DD_EVQ_IND_RPTR_WIDTH) - 1));
2775 		efx_writed_page(efx, &rptr, ER_DD_EVQ_INDIRECT,
2776 				channel->channel);
2777 	} else {
2778 		EFX_POPULATE_DWORD_1(rptr, ERF_DZ_EVQ_RPTR,
2779 				     channel->eventq_read_ptr &
2780 				     channel->eventq_mask);
2781 		efx_writed_page(efx, &rptr, ER_DZ_EVQ_RPTR, channel->channel);
2782 	}
2783 }
2784 
2785 static void efx_ef10_ev_test_generate(struct efx_channel *channel)
2786 {
2787 	MCDI_DECLARE_BUF(inbuf, MC_CMD_DRIVER_EVENT_IN_LEN);
2788 	struct efx_nic *efx = channel->efx;
2789 	efx_qword_t event;
2790 	int rc;
2791 
2792 	EFX_POPULATE_QWORD_2(event,
2793 			     ESF_DZ_EV_CODE, EFX_EF10_DRVGEN_EV,
2794 			     ESF_DZ_EV_DATA, EFX_EF10_TEST);
2795 
2796 	MCDI_SET_DWORD(inbuf, DRIVER_EVENT_IN_EVQ, channel->channel);
2797 
2798 	/* MCDI_SET_QWORD is not appropriate here since EFX_POPULATE_* has
2799 	 * already swapped the data to little-endian order.
2800 	 */
2801 	memcpy(MCDI_PTR(inbuf, DRIVER_EVENT_IN_DATA), &event.u64[0],
2802 	       sizeof(efx_qword_t));
2803 
2804 	rc = efx_mcdi_rpc(efx, MC_CMD_DRIVER_EVENT, inbuf, sizeof(inbuf),
2805 			  NULL, 0, NULL);
2806 	if (rc != 0)
2807 		goto fail;
2808 
2809 	return;
2810 
2811 fail:
2812 	WARN_ON(true);
2813 	netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
2814 }
2815 
2816 void efx_ef10_handle_drain_event(struct efx_nic *efx)
2817 {
2818 	if (atomic_dec_and_test(&efx->active_queues))
2819 		wake_up(&efx->flush_wq);
2820 
2821 	WARN_ON(atomic_read(&efx->active_queues) < 0);
2822 }
2823 
2824 static int efx_ef10_fini_dmaq(struct efx_nic *efx)
2825 {
2826 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2827 	struct efx_channel *channel;
2828 	struct efx_tx_queue *tx_queue;
2829 	struct efx_rx_queue *rx_queue;
2830 	int pending;
2831 
2832 	/* If the MC has just rebooted, the TX/RX queues will have already been
2833 	 * torn down, but efx->active_queues needs to be set to zero.
2834 	 */
2835 	if (nic_data->must_realloc_vis) {
2836 		atomic_set(&efx->active_queues, 0);
2837 		return 0;
2838 	}
2839 
2840 	/* Do not attempt to write to the NIC during EEH recovery */
2841 	if (efx->state != STATE_RECOVERY) {
2842 		efx_for_each_channel(channel, efx) {
2843 			efx_for_each_channel_rx_queue(rx_queue, channel)
2844 				efx_ef10_rx_fini(rx_queue);
2845 			efx_for_each_channel_tx_queue(tx_queue, channel)
2846 				efx_ef10_tx_fini(tx_queue);
2847 		}
2848 
2849 		wait_event_timeout(efx->flush_wq,
2850 				   atomic_read(&efx->active_queues) == 0,
2851 				   msecs_to_jiffies(EFX_MAX_FLUSH_TIME));
2852 		pending = atomic_read(&efx->active_queues);
2853 		if (pending) {
2854 			netif_err(efx, hw, efx->net_dev, "failed to flush %d queues\n",
2855 				  pending);
2856 			return -ETIMEDOUT;
2857 		}
2858 	}
2859 
2860 	return 0;
2861 }
2862 
2863 static void efx_ef10_prepare_flr(struct efx_nic *efx)
2864 {
2865 	atomic_set(&efx->active_queues, 0);
2866 }
2867 
2868 static bool efx_ef10_filter_equal(const struct efx_filter_spec *left,
2869 				  const struct efx_filter_spec *right)
2870 {
2871 	if ((left->match_flags ^ right->match_flags) |
2872 	    ((left->flags ^ right->flags) &
2873 	     (EFX_FILTER_FLAG_RX | EFX_FILTER_FLAG_TX)))
2874 		return false;
2875 
2876 	return memcmp(&left->outer_vid, &right->outer_vid,
2877 		      sizeof(struct efx_filter_spec) -
2878 		      offsetof(struct efx_filter_spec, outer_vid)) == 0;
2879 }
2880 
2881 static unsigned int efx_ef10_filter_hash(const struct efx_filter_spec *spec)
2882 {
2883 	BUILD_BUG_ON(offsetof(struct efx_filter_spec, outer_vid) & 3);
2884 	return jhash2((const u32 *)&spec->outer_vid,
2885 		      (sizeof(struct efx_filter_spec) -
2886 		       offsetof(struct efx_filter_spec, outer_vid)) / 4,
2887 		      0);
2888 	/* XXX should we randomise the initval? */
2889 }
2890 
2891 /* Decide whether a filter should be exclusive or else should allow
2892  * delivery to additional recipients.  Currently we decide that
2893  * filters for specific local unicast MAC and IP addresses are
2894  * exclusive.
2895  */
2896 static bool efx_ef10_filter_is_exclusive(const struct efx_filter_spec *spec)
2897 {
2898 	if (spec->match_flags & EFX_FILTER_MATCH_LOC_MAC &&
2899 	    !is_multicast_ether_addr(spec->loc_mac))
2900 		return true;
2901 
2902 	if ((spec->match_flags &
2903 	     (EFX_FILTER_MATCH_ETHER_TYPE | EFX_FILTER_MATCH_LOC_HOST)) ==
2904 	    (EFX_FILTER_MATCH_ETHER_TYPE | EFX_FILTER_MATCH_LOC_HOST)) {
2905 		if (spec->ether_type == htons(ETH_P_IP) &&
2906 		    !ipv4_is_multicast(spec->loc_host[0]))
2907 			return true;
2908 		if (spec->ether_type == htons(ETH_P_IPV6) &&
2909 		    ((const u8 *)spec->loc_host)[0] != 0xff)
2910 			return true;
2911 	}
2912 
2913 	return false;
2914 }
2915 
2916 static struct efx_filter_spec *
2917 efx_ef10_filter_entry_spec(const struct efx_ef10_filter_table *table,
2918 			   unsigned int filter_idx)
2919 {
2920 	return (struct efx_filter_spec *)(table->entry[filter_idx].spec &
2921 					  ~EFX_EF10_FILTER_FLAGS);
2922 }
2923 
2924 static unsigned int
2925 efx_ef10_filter_entry_flags(const struct efx_ef10_filter_table *table,
2926 			   unsigned int filter_idx)
2927 {
2928 	return table->entry[filter_idx].spec & EFX_EF10_FILTER_FLAGS;
2929 }
2930 
2931 static void
2932 efx_ef10_filter_set_entry(struct efx_ef10_filter_table *table,
2933 			  unsigned int filter_idx,
2934 			  const struct efx_filter_spec *spec,
2935 			  unsigned int flags)
2936 {
2937 	table->entry[filter_idx].spec =	(unsigned long)spec | flags;
2938 }
2939 
2940 static void efx_ef10_filter_push_prep(struct efx_nic *efx,
2941 				      const struct efx_filter_spec *spec,
2942 				      efx_dword_t *inbuf, u64 handle,
2943 				      bool replacing)
2944 {
2945 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
2946 	u32 flags = spec->flags;
2947 
2948 	memset(inbuf, 0, MC_CMD_FILTER_OP_IN_LEN);
2949 
2950 	/* Remove RSS flag if we don't have an RSS context. */
2951 	if (flags & EFX_FILTER_FLAG_RX_RSS &&
2952 	    spec->rss_context == EFX_FILTER_RSS_CONTEXT_DEFAULT &&
2953 	    nic_data->rx_rss_context == EFX_EF10_RSS_CONTEXT_INVALID)
2954 		flags &= ~EFX_FILTER_FLAG_RX_RSS;
2955 
2956 	if (replacing) {
2957 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
2958 			       MC_CMD_FILTER_OP_IN_OP_REPLACE);
2959 		MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE, handle);
2960 	} else {
2961 		u32 match_fields = 0;
2962 
2963 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
2964 			       efx_ef10_filter_is_exclusive(spec) ?
2965 			       MC_CMD_FILTER_OP_IN_OP_INSERT :
2966 			       MC_CMD_FILTER_OP_IN_OP_SUBSCRIBE);
2967 
2968 		/* Convert match flags and values.  Unlike almost
2969 		 * everything else in MCDI, these fields are in
2970 		 * network byte order.
2971 		 */
2972 		if (spec->match_flags & EFX_FILTER_MATCH_LOC_MAC_IG)
2973 			match_fields |=
2974 				is_multicast_ether_addr(spec->loc_mac) ?
2975 				1 << MC_CMD_FILTER_OP_IN_MATCH_UNKNOWN_MCAST_DST_LBN :
2976 				1 << MC_CMD_FILTER_OP_IN_MATCH_UNKNOWN_UCAST_DST_LBN;
2977 #define COPY_FIELD(gen_flag, gen_field, mcdi_field)			     \
2978 		if (spec->match_flags & EFX_FILTER_MATCH_ ## gen_flag) {     \
2979 			match_fields |=					     \
2980 				1 << MC_CMD_FILTER_OP_IN_MATCH_ ##	     \
2981 				mcdi_field ## _LBN;			     \
2982 			BUILD_BUG_ON(					     \
2983 				MC_CMD_FILTER_OP_IN_ ## mcdi_field ## _LEN < \
2984 				sizeof(spec->gen_field));		     \
2985 			memcpy(MCDI_PTR(inbuf, FILTER_OP_IN_ ##	mcdi_field), \
2986 			       &spec->gen_field, sizeof(spec->gen_field));   \
2987 		}
2988 		COPY_FIELD(REM_HOST, rem_host, SRC_IP);
2989 		COPY_FIELD(LOC_HOST, loc_host, DST_IP);
2990 		COPY_FIELD(REM_MAC, rem_mac, SRC_MAC);
2991 		COPY_FIELD(REM_PORT, rem_port, SRC_PORT);
2992 		COPY_FIELD(LOC_MAC, loc_mac, DST_MAC);
2993 		COPY_FIELD(LOC_PORT, loc_port, DST_PORT);
2994 		COPY_FIELD(ETHER_TYPE, ether_type, ETHER_TYPE);
2995 		COPY_FIELD(INNER_VID, inner_vid, INNER_VLAN);
2996 		COPY_FIELD(OUTER_VID, outer_vid, OUTER_VLAN);
2997 		COPY_FIELD(IP_PROTO, ip_proto, IP_PROTO);
2998 #undef COPY_FIELD
2999 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_MATCH_FIELDS,
3000 			       match_fields);
3001 	}
3002 
3003 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_PORT_ID, nic_data->vport_id);
3004 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_DEST,
3005 		       spec->dmaq_id == EFX_FILTER_RX_DMAQ_ID_DROP ?
3006 		       MC_CMD_FILTER_OP_IN_RX_DEST_DROP :
3007 		       MC_CMD_FILTER_OP_IN_RX_DEST_HOST);
3008 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_TX_DOMAIN, 0);
3009 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_TX_DEST,
3010 		       MC_CMD_FILTER_OP_IN_TX_DEST_DEFAULT);
3011 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_QUEUE,
3012 		       spec->dmaq_id == EFX_FILTER_RX_DMAQ_ID_DROP ?
3013 		       0 : spec->dmaq_id);
3014 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_MODE,
3015 		       (flags & EFX_FILTER_FLAG_RX_RSS) ?
3016 		       MC_CMD_FILTER_OP_IN_RX_MODE_RSS :
3017 		       MC_CMD_FILTER_OP_IN_RX_MODE_SIMPLE);
3018 	if (flags & EFX_FILTER_FLAG_RX_RSS)
3019 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_CONTEXT,
3020 			       spec->rss_context !=
3021 			       EFX_FILTER_RSS_CONTEXT_DEFAULT ?
3022 			       spec->rss_context : nic_data->rx_rss_context);
3023 }
3024 
3025 static int efx_ef10_filter_push(struct efx_nic *efx,
3026 				const struct efx_filter_spec *spec,
3027 				u64 *handle, bool replacing)
3028 {
3029 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_IN_LEN);
3030 	MCDI_DECLARE_BUF(outbuf, MC_CMD_FILTER_OP_OUT_LEN);
3031 	int rc;
3032 
3033 	efx_ef10_filter_push_prep(efx, spec, inbuf, *handle, replacing);
3034 	rc = efx_mcdi_rpc(efx, MC_CMD_FILTER_OP, inbuf, sizeof(inbuf),
3035 			  outbuf, sizeof(outbuf), NULL);
3036 	if (rc == 0)
3037 		*handle = MCDI_QWORD(outbuf, FILTER_OP_OUT_HANDLE);
3038 	if (rc == -ENOSPC)
3039 		rc = -EBUSY; /* to match efx_farch_filter_insert() */
3040 	return rc;
3041 }
3042 
3043 static int efx_ef10_filter_rx_match_pri(struct efx_ef10_filter_table *table,
3044 					enum efx_filter_match_flags match_flags)
3045 {
3046 	unsigned int match_pri;
3047 
3048 	for (match_pri = 0;
3049 	     match_pri < table->rx_match_count;
3050 	     match_pri++)
3051 		if (table->rx_match_flags[match_pri] == match_flags)
3052 			return match_pri;
3053 
3054 	return -EPROTONOSUPPORT;
3055 }
3056 
3057 static s32 efx_ef10_filter_insert(struct efx_nic *efx,
3058 				  struct efx_filter_spec *spec,
3059 				  bool replace_equal)
3060 {
3061 	struct efx_ef10_filter_table *table = efx->filter_state;
3062 	DECLARE_BITMAP(mc_rem_map, EFX_EF10_FILTER_SEARCH_LIMIT);
3063 	struct efx_filter_spec *saved_spec;
3064 	unsigned int match_pri, hash;
3065 	unsigned int priv_flags;
3066 	bool replacing = false;
3067 	int ins_index = -1;
3068 	DEFINE_WAIT(wait);
3069 	bool is_mc_recip;
3070 	s32 rc;
3071 
3072 	/* For now, only support RX filters */
3073 	if ((spec->flags & (EFX_FILTER_FLAG_RX | EFX_FILTER_FLAG_TX)) !=
3074 	    EFX_FILTER_FLAG_RX)
3075 		return -EINVAL;
3076 
3077 	rc = efx_ef10_filter_rx_match_pri(table, spec->match_flags);
3078 	if (rc < 0)
3079 		return rc;
3080 	match_pri = rc;
3081 
3082 	hash = efx_ef10_filter_hash(spec);
3083 	is_mc_recip = efx_filter_is_mc_recipient(spec);
3084 	if (is_mc_recip)
3085 		bitmap_zero(mc_rem_map, EFX_EF10_FILTER_SEARCH_LIMIT);
3086 
3087 	/* Find any existing filters with the same match tuple or
3088 	 * else a free slot to insert at.  If any of them are busy,
3089 	 * we have to wait and retry.
3090 	 */
3091 	for (;;) {
3092 		unsigned int depth = 1;
3093 		unsigned int i;
3094 
3095 		spin_lock_bh(&efx->filter_lock);
3096 
3097 		for (;;) {
3098 			i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
3099 			saved_spec = efx_ef10_filter_entry_spec(table, i);
3100 
3101 			if (!saved_spec) {
3102 				if (ins_index < 0)
3103 					ins_index = i;
3104 			} else if (efx_ef10_filter_equal(spec, saved_spec)) {
3105 				if (table->entry[i].spec &
3106 				    EFX_EF10_FILTER_FLAG_BUSY)
3107 					break;
3108 				if (spec->priority < saved_spec->priority &&
3109 				    spec->priority != EFX_FILTER_PRI_AUTO) {
3110 					rc = -EPERM;
3111 					goto out_unlock;
3112 				}
3113 				if (!is_mc_recip) {
3114 					/* This is the only one */
3115 					if (spec->priority ==
3116 					    saved_spec->priority &&
3117 					    !replace_equal) {
3118 						rc = -EEXIST;
3119 						goto out_unlock;
3120 					}
3121 					ins_index = i;
3122 					goto found;
3123 				} else if (spec->priority >
3124 					   saved_spec->priority ||
3125 					   (spec->priority ==
3126 					    saved_spec->priority &&
3127 					    replace_equal)) {
3128 					if (ins_index < 0)
3129 						ins_index = i;
3130 					else
3131 						__set_bit(depth, mc_rem_map);
3132 				}
3133 			}
3134 
3135 			/* Once we reach the maximum search depth, use
3136 			 * the first suitable slot or return -EBUSY if
3137 			 * there was none
3138 			 */
3139 			if (depth == EFX_EF10_FILTER_SEARCH_LIMIT) {
3140 				if (ins_index < 0) {
3141 					rc = -EBUSY;
3142 					goto out_unlock;
3143 				}
3144 				goto found;
3145 			}
3146 
3147 			++depth;
3148 		}
3149 
3150 		prepare_to_wait(&table->waitq, &wait, TASK_UNINTERRUPTIBLE);
3151 		spin_unlock_bh(&efx->filter_lock);
3152 		schedule();
3153 	}
3154 
3155 found:
3156 	/* Create a software table entry if necessary, and mark it
3157 	 * busy.  We might yet fail to insert, but any attempt to
3158 	 * insert a conflicting filter while we're waiting for the
3159 	 * firmware must find the busy entry.
3160 	 */
3161 	saved_spec = efx_ef10_filter_entry_spec(table, ins_index);
3162 	if (saved_spec) {
3163 		if (spec->priority == EFX_FILTER_PRI_AUTO &&
3164 		    saved_spec->priority >= EFX_FILTER_PRI_AUTO) {
3165 			/* Just make sure it won't be removed */
3166 			if (saved_spec->priority > EFX_FILTER_PRI_AUTO)
3167 				saved_spec->flags |= EFX_FILTER_FLAG_RX_OVER_AUTO;
3168 			table->entry[ins_index].spec &=
3169 				~EFX_EF10_FILTER_FLAG_AUTO_OLD;
3170 			rc = ins_index;
3171 			goto out_unlock;
3172 		}
3173 		replacing = true;
3174 		priv_flags = efx_ef10_filter_entry_flags(table, ins_index);
3175 	} else {
3176 		saved_spec = kmalloc(sizeof(*spec), GFP_ATOMIC);
3177 		if (!saved_spec) {
3178 			rc = -ENOMEM;
3179 			goto out_unlock;
3180 		}
3181 		*saved_spec = *spec;
3182 		priv_flags = 0;
3183 	}
3184 	efx_ef10_filter_set_entry(table, ins_index, saved_spec,
3185 				  priv_flags | EFX_EF10_FILTER_FLAG_BUSY);
3186 
3187 	/* Mark lower-priority multicast recipients busy prior to removal */
3188 	if (is_mc_recip) {
3189 		unsigned int depth, i;
3190 
3191 		for (depth = 0; depth < EFX_EF10_FILTER_SEARCH_LIMIT; depth++) {
3192 			i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
3193 			if (test_bit(depth, mc_rem_map))
3194 				table->entry[i].spec |=
3195 					EFX_EF10_FILTER_FLAG_BUSY;
3196 		}
3197 	}
3198 
3199 	spin_unlock_bh(&efx->filter_lock);
3200 
3201 	rc = efx_ef10_filter_push(efx, spec, &table->entry[ins_index].handle,
3202 				  replacing);
3203 
3204 	/* Finalise the software table entry */
3205 	spin_lock_bh(&efx->filter_lock);
3206 	if (rc == 0) {
3207 		if (replacing) {
3208 			/* Update the fields that may differ */
3209 			if (saved_spec->priority == EFX_FILTER_PRI_AUTO)
3210 				saved_spec->flags |=
3211 					EFX_FILTER_FLAG_RX_OVER_AUTO;
3212 			saved_spec->priority = spec->priority;
3213 			saved_spec->flags &= EFX_FILTER_FLAG_RX_OVER_AUTO;
3214 			saved_spec->flags |= spec->flags;
3215 			saved_spec->rss_context = spec->rss_context;
3216 			saved_spec->dmaq_id = spec->dmaq_id;
3217 		}
3218 	} else if (!replacing) {
3219 		kfree(saved_spec);
3220 		saved_spec = NULL;
3221 	}
3222 	efx_ef10_filter_set_entry(table, ins_index, saved_spec, priv_flags);
3223 
3224 	/* Remove and finalise entries for lower-priority multicast
3225 	 * recipients
3226 	 */
3227 	if (is_mc_recip) {
3228 		MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_IN_LEN);
3229 		unsigned int depth, i;
3230 
3231 		memset(inbuf, 0, sizeof(inbuf));
3232 
3233 		for (depth = 0; depth < EFX_EF10_FILTER_SEARCH_LIMIT; depth++) {
3234 			if (!test_bit(depth, mc_rem_map))
3235 				continue;
3236 
3237 			i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
3238 			saved_spec = efx_ef10_filter_entry_spec(table, i);
3239 			priv_flags = efx_ef10_filter_entry_flags(table, i);
3240 
3241 			if (rc == 0) {
3242 				spin_unlock_bh(&efx->filter_lock);
3243 				MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
3244 					       MC_CMD_FILTER_OP_IN_OP_UNSUBSCRIBE);
3245 				MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
3246 					       table->entry[i].handle);
3247 				rc = efx_mcdi_rpc(efx, MC_CMD_FILTER_OP,
3248 						  inbuf, sizeof(inbuf),
3249 						  NULL, 0, NULL);
3250 				spin_lock_bh(&efx->filter_lock);
3251 			}
3252 
3253 			if (rc == 0) {
3254 				kfree(saved_spec);
3255 				saved_spec = NULL;
3256 				priv_flags = 0;
3257 			} else {
3258 				priv_flags &= ~EFX_EF10_FILTER_FLAG_BUSY;
3259 			}
3260 			efx_ef10_filter_set_entry(table, i, saved_spec,
3261 						  priv_flags);
3262 		}
3263 	}
3264 
3265 	/* If successful, return the inserted filter ID */
3266 	if (rc == 0)
3267 		rc = match_pri * HUNT_FILTER_TBL_ROWS + ins_index;
3268 
3269 	wake_up_all(&table->waitq);
3270 out_unlock:
3271 	spin_unlock_bh(&efx->filter_lock);
3272 	finish_wait(&table->waitq, &wait);
3273 	return rc;
3274 }
3275 
3276 static void efx_ef10_filter_update_rx_scatter(struct efx_nic *efx)
3277 {
3278 	/* no need to do anything here on EF10 */
3279 }
3280 
3281 /* Remove a filter.
3282  * If !by_index, remove by ID
3283  * If by_index, remove by index
3284  * Filter ID may come from userland and must be range-checked.
3285  */
3286 static int efx_ef10_filter_remove_internal(struct efx_nic *efx,
3287 					   unsigned int priority_mask,
3288 					   u32 filter_id, bool by_index)
3289 {
3290 	unsigned int filter_idx = filter_id % HUNT_FILTER_TBL_ROWS;
3291 	struct efx_ef10_filter_table *table = efx->filter_state;
3292 	MCDI_DECLARE_BUF(inbuf,
3293 			 MC_CMD_FILTER_OP_IN_HANDLE_OFST +
3294 			 MC_CMD_FILTER_OP_IN_HANDLE_LEN);
3295 	struct efx_filter_spec *spec;
3296 	DEFINE_WAIT(wait);
3297 	int rc;
3298 
3299 	/* Find the software table entry and mark it busy.  Don't
3300 	 * remove it yet; any attempt to update while we're waiting
3301 	 * for the firmware must find the busy entry.
3302 	 */
3303 	for (;;) {
3304 		spin_lock_bh(&efx->filter_lock);
3305 		if (!(table->entry[filter_idx].spec &
3306 		      EFX_EF10_FILTER_FLAG_BUSY))
3307 			break;
3308 		prepare_to_wait(&table->waitq, &wait, TASK_UNINTERRUPTIBLE);
3309 		spin_unlock_bh(&efx->filter_lock);
3310 		schedule();
3311 	}
3312 
3313 	spec = efx_ef10_filter_entry_spec(table, filter_idx);
3314 	if (!spec ||
3315 	    (!by_index &&
3316 	     efx_ef10_filter_rx_match_pri(table, spec->match_flags) !=
3317 	     filter_id / HUNT_FILTER_TBL_ROWS)) {
3318 		rc = -ENOENT;
3319 		goto out_unlock;
3320 	}
3321 
3322 	if (spec->flags & EFX_FILTER_FLAG_RX_OVER_AUTO &&
3323 	    priority_mask == (1U << EFX_FILTER_PRI_AUTO)) {
3324 		/* Just remove flags */
3325 		spec->flags &= ~EFX_FILTER_FLAG_RX_OVER_AUTO;
3326 		table->entry[filter_idx].spec &= ~EFX_EF10_FILTER_FLAG_AUTO_OLD;
3327 		rc = 0;
3328 		goto out_unlock;
3329 	}
3330 
3331 	if (!(priority_mask & (1U << spec->priority))) {
3332 		rc = -ENOENT;
3333 		goto out_unlock;
3334 	}
3335 
3336 	table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_BUSY;
3337 	spin_unlock_bh(&efx->filter_lock);
3338 
3339 	if (spec->flags & EFX_FILTER_FLAG_RX_OVER_AUTO) {
3340 		/* Reset to an automatic filter */
3341 
3342 		struct efx_filter_spec new_spec = *spec;
3343 
3344 		new_spec.priority = EFX_FILTER_PRI_AUTO;
3345 		new_spec.flags = (EFX_FILTER_FLAG_RX |
3346 				  (efx_rss_enabled(efx) ?
3347 				   EFX_FILTER_FLAG_RX_RSS : 0));
3348 		new_spec.dmaq_id = 0;
3349 		new_spec.rss_context = EFX_FILTER_RSS_CONTEXT_DEFAULT;
3350 		rc = efx_ef10_filter_push(efx, &new_spec,
3351 					  &table->entry[filter_idx].handle,
3352 					  true);
3353 
3354 		spin_lock_bh(&efx->filter_lock);
3355 		if (rc == 0)
3356 			*spec = new_spec;
3357 	} else {
3358 		/* Really remove the filter */
3359 
3360 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
3361 			       efx_ef10_filter_is_exclusive(spec) ?
3362 			       MC_CMD_FILTER_OP_IN_OP_REMOVE :
3363 			       MC_CMD_FILTER_OP_IN_OP_UNSUBSCRIBE);
3364 		MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
3365 			       table->entry[filter_idx].handle);
3366 		rc = efx_mcdi_rpc(efx, MC_CMD_FILTER_OP,
3367 				  inbuf, sizeof(inbuf), NULL, 0, NULL);
3368 
3369 		spin_lock_bh(&efx->filter_lock);
3370 		if (rc == 0) {
3371 			kfree(spec);
3372 			efx_ef10_filter_set_entry(table, filter_idx, NULL, 0);
3373 		}
3374 	}
3375 
3376 	table->entry[filter_idx].spec &= ~EFX_EF10_FILTER_FLAG_BUSY;
3377 	wake_up_all(&table->waitq);
3378 out_unlock:
3379 	spin_unlock_bh(&efx->filter_lock);
3380 	finish_wait(&table->waitq, &wait);
3381 	return rc;
3382 }
3383 
3384 static int efx_ef10_filter_remove_safe(struct efx_nic *efx,
3385 				       enum efx_filter_priority priority,
3386 				       u32 filter_id)
3387 {
3388 	return efx_ef10_filter_remove_internal(efx, 1U << priority,
3389 					       filter_id, false);
3390 }
3391 
3392 static u32 efx_ef10_filter_get_unsafe_id(struct efx_nic *efx, u32 filter_id)
3393 {
3394 	return filter_id % HUNT_FILTER_TBL_ROWS;
3395 }
3396 
3397 static int efx_ef10_filter_remove_unsafe(struct efx_nic *efx,
3398 					 enum efx_filter_priority priority,
3399 					 u32 filter_id)
3400 {
3401 	return efx_ef10_filter_remove_internal(efx, 1U << priority,
3402 					       filter_id, true);
3403 }
3404 
3405 static int efx_ef10_filter_get_safe(struct efx_nic *efx,
3406 				    enum efx_filter_priority priority,
3407 				    u32 filter_id, struct efx_filter_spec *spec)
3408 {
3409 	unsigned int filter_idx = filter_id % HUNT_FILTER_TBL_ROWS;
3410 	struct efx_ef10_filter_table *table = efx->filter_state;
3411 	const struct efx_filter_spec *saved_spec;
3412 	int rc;
3413 
3414 	spin_lock_bh(&efx->filter_lock);
3415 	saved_spec = efx_ef10_filter_entry_spec(table, filter_idx);
3416 	if (saved_spec && saved_spec->priority == priority &&
3417 	    efx_ef10_filter_rx_match_pri(table, saved_spec->match_flags) ==
3418 	    filter_id / HUNT_FILTER_TBL_ROWS) {
3419 		*spec = *saved_spec;
3420 		rc = 0;
3421 	} else {
3422 		rc = -ENOENT;
3423 	}
3424 	spin_unlock_bh(&efx->filter_lock);
3425 	return rc;
3426 }
3427 
3428 static int efx_ef10_filter_clear_rx(struct efx_nic *efx,
3429 				     enum efx_filter_priority priority)
3430 {
3431 	unsigned int priority_mask;
3432 	unsigned int i;
3433 	int rc;
3434 
3435 	priority_mask = (((1U << (priority + 1)) - 1) &
3436 			 ~(1U << EFX_FILTER_PRI_AUTO));
3437 
3438 	for (i = 0; i < HUNT_FILTER_TBL_ROWS; i++) {
3439 		rc = efx_ef10_filter_remove_internal(efx, priority_mask,
3440 						     i, true);
3441 		if (rc && rc != -ENOENT)
3442 			return rc;
3443 	}
3444 
3445 	return 0;
3446 }
3447 
3448 static u32 efx_ef10_filter_count_rx_used(struct efx_nic *efx,
3449 					 enum efx_filter_priority priority)
3450 {
3451 	struct efx_ef10_filter_table *table = efx->filter_state;
3452 	unsigned int filter_idx;
3453 	s32 count = 0;
3454 
3455 	spin_lock_bh(&efx->filter_lock);
3456 	for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
3457 		if (table->entry[filter_idx].spec &&
3458 		    efx_ef10_filter_entry_spec(table, filter_idx)->priority ==
3459 		    priority)
3460 			++count;
3461 	}
3462 	spin_unlock_bh(&efx->filter_lock);
3463 	return count;
3464 }
3465 
3466 static u32 efx_ef10_filter_get_rx_id_limit(struct efx_nic *efx)
3467 {
3468 	struct efx_ef10_filter_table *table = efx->filter_state;
3469 
3470 	return table->rx_match_count * HUNT_FILTER_TBL_ROWS;
3471 }
3472 
3473 static s32 efx_ef10_filter_get_rx_ids(struct efx_nic *efx,
3474 				      enum efx_filter_priority priority,
3475 				      u32 *buf, u32 size)
3476 {
3477 	struct efx_ef10_filter_table *table = efx->filter_state;
3478 	struct efx_filter_spec *spec;
3479 	unsigned int filter_idx;
3480 	s32 count = 0;
3481 
3482 	spin_lock_bh(&efx->filter_lock);
3483 	for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
3484 		spec = efx_ef10_filter_entry_spec(table, filter_idx);
3485 		if (spec && spec->priority == priority) {
3486 			if (count == size) {
3487 				count = -EMSGSIZE;
3488 				break;
3489 			}
3490 			buf[count++] = (efx_ef10_filter_rx_match_pri(
3491 						table, spec->match_flags) *
3492 					HUNT_FILTER_TBL_ROWS +
3493 					filter_idx);
3494 		}
3495 	}
3496 	spin_unlock_bh(&efx->filter_lock);
3497 	return count;
3498 }
3499 
3500 #ifdef CONFIG_RFS_ACCEL
3501 
3502 static efx_mcdi_async_completer efx_ef10_filter_rfs_insert_complete;
3503 
3504 static s32 efx_ef10_filter_rfs_insert(struct efx_nic *efx,
3505 				      struct efx_filter_spec *spec)
3506 {
3507 	struct efx_ef10_filter_table *table = efx->filter_state;
3508 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_IN_LEN);
3509 	struct efx_filter_spec *saved_spec;
3510 	unsigned int hash, i, depth = 1;
3511 	bool replacing = false;
3512 	int ins_index = -1;
3513 	u64 cookie;
3514 	s32 rc;
3515 
3516 	/* Must be an RX filter without RSS and not for a multicast
3517 	 * destination address (RFS only works for connected sockets).
3518 	 * These restrictions allow us to pass only a tiny amount of
3519 	 * data through to the completion function.
3520 	 */
3521 	EFX_WARN_ON_PARANOID(spec->flags !=
3522 			     (EFX_FILTER_FLAG_RX | EFX_FILTER_FLAG_RX_SCATTER));
3523 	EFX_WARN_ON_PARANOID(spec->priority != EFX_FILTER_PRI_HINT);
3524 	EFX_WARN_ON_PARANOID(efx_filter_is_mc_recipient(spec));
3525 
3526 	hash = efx_ef10_filter_hash(spec);
3527 
3528 	spin_lock_bh(&efx->filter_lock);
3529 
3530 	/* Find any existing filter with the same match tuple or else
3531 	 * a free slot to insert at.  If an existing filter is busy,
3532 	 * we have to give up.
3533 	 */
3534 	for (;;) {
3535 		i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
3536 		saved_spec = efx_ef10_filter_entry_spec(table, i);
3537 
3538 		if (!saved_spec) {
3539 			if (ins_index < 0)
3540 				ins_index = i;
3541 		} else if (efx_ef10_filter_equal(spec, saved_spec)) {
3542 			if (table->entry[i].spec & EFX_EF10_FILTER_FLAG_BUSY) {
3543 				rc = -EBUSY;
3544 				goto fail_unlock;
3545 			}
3546 			if (spec->priority < saved_spec->priority) {
3547 				rc = -EPERM;
3548 				goto fail_unlock;
3549 			}
3550 			ins_index = i;
3551 			break;
3552 		}
3553 
3554 		/* Once we reach the maximum search depth, use the
3555 		 * first suitable slot or return -EBUSY if there was
3556 		 * none
3557 		 */
3558 		if (depth == EFX_EF10_FILTER_SEARCH_LIMIT) {
3559 			if (ins_index < 0) {
3560 				rc = -EBUSY;
3561 				goto fail_unlock;
3562 			}
3563 			break;
3564 		}
3565 
3566 		++depth;
3567 	}
3568 
3569 	/* Create a software table entry if necessary, and mark it
3570 	 * busy.  We might yet fail to insert, but any attempt to
3571 	 * insert a conflicting filter while we're waiting for the
3572 	 * firmware must find the busy entry.
3573 	 */
3574 	saved_spec = efx_ef10_filter_entry_spec(table, ins_index);
3575 	if (saved_spec) {
3576 		replacing = true;
3577 	} else {
3578 		saved_spec = kmalloc(sizeof(*spec), GFP_ATOMIC);
3579 		if (!saved_spec) {
3580 			rc = -ENOMEM;
3581 			goto fail_unlock;
3582 		}
3583 		*saved_spec = *spec;
3584 	}
3585 	efx_ef10_filter_set_entry(table, ins_index, saved_spec,
3586 				  EFX_EF10_FILTER_FLAG_BUSY);
3587 
3588 	spin_unlock_bh(&efx->filter_lock);
3589 
3590 	/* Pack up the variables needed on completion */
3591 	cookie = replacing << 31 | ins_index << 16 | spec->dmaq_id;
3592 
3593 	efx_ef10_filter_push_prep(efx, spec, inbuf,
3594 				  table->entry[ins_index].handle, replacing);
3595 	efx_mcdi_rpc_async(efx, MC_CMD_FILTER_OP, inbuf, sizeof(inbuf),
3596 			   MC_CMD_FILTER_OP_OUT_LEN,
3597 			   efx_ef10_filter_rfs_insert_complete, cookie);
3598 
3599 	return ins_index;
3600 
3601 fail_unlock:
3602 	spin_unlock_bh(&efx->filter_lock);
3603 	return rc;
3604 }
3605 
3606 static void
3607 efx_ef10_filter_rfs_insert_complete(struct efx_nic *efx, unsigned long cookie,
3608 				    int rc, efx_dword_t *outbuf,
3609 				    size_t outlen_actual)
3610 {
3611 	struct efx_ef10_filter_table *table = efx->filter_state;
3612 	unsigned int ins_index, dmaq_id;
3613 	struct efx_filter_spec *spec;
3614 	bool replacing;
3615 
3616 	/* Unpack the cookie */
3617 	replacing = cookie >> 31;
3618 	ins_index = (cookie >> 16) & (HUNT_FILTER_TBL_ROWS - 1);
3619 	dmaq_id = cookie & 0xffff;
3620 
3621 	spin_lock_bh(&efx->filter_lock);
3622 	spec = efx_ef10_filter_entry_spec(table, ins_index);
3623 	if (rc == 0) {
3624 		table->entry[ins_index].handle =
3625 			MCDI_QWORD(outbuf, FILTER_OP_OUT_HANDLE);
3626 		if (replacing)
3627 			spec->dmaq_id = dmaq_id;
3628 	} else if (!replacing) {
3629 		kfree(spec);
3630 		spec = NULL;
3631 	}
3632 	efx_ef10_filter_set_entry(table, ins_index, spec, 0);
3633 	spin_unlock_bh(&efx->filter_lock);
3634 
3635 	wake_up_all(&table->waitq);
3636 }
3637 
3638 static void
3639 efx_ef10_filter_rfs_expire_complete(struct efx_nic *efx,
3640 				    unsigned long filter_idx,
3641 				    int rc, efx_dword_t *outbuf,
3642 				    size_t outlen_actual);
3643 
3644 static bool efx_ef10_filter_rfs_expire_one(struct efx_nic *efx, u32 flow_id,
3645 					   unsigned int filter_idx)
3646 {
3647 	struct efx_ef10_filter_table *table = efx->filter_state;
3648 	struct efx_filter_spec *spec =
3649 		efx_ef10_filter_entry_spec(table, filter_idx);
3650 	MCDI_DECLARE_BUF(inbuf,
3651 			 MC_CMD_FILTER_OP_IN_HANDLE_OFST +
3652 			 MC_CMD_FILTER_OP_IN_HANDLE_LEN);
3653 
3654 	if (!spec ||
3655 	    (table->entry[filter_idx].spec & EFX_EF10_FILTER_FLAG_BUSY) ||
3656 	    spec->priority != EFX_FILTER_PRI_HINT ||
3657 	    !rps_may_expire_flow(efx->net_dev, spec->dmaq_id,
3658 				 flow_id, filter_idx))
3659 		return false;
3660 
3661 	MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
3662 		       MC_CMD_FILTER_OP_IN_OP_REMOVE);
3663 	MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
3664 		       table->entry[filter_idx].handle);
3665 	if (efx_mcdi_rpc_async(efx, MC_CMD_FILTER_OP, inbuf, sizeof(inbuf), 0,
3666 			       efx_ef10_filter_rfs_expire_complete, filter_idx))
3667 		return false;
3668 
3669 	table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_BUSY;
3670 	return true;
3671 }
3672 
3673 static void
3674 efx_ef10_filter_rfs_expire_complete(struct efx_nic *efx,
3675 				    unsigned long filter_idx,
3676 				    int rc, efx_dword_t *outbuf,
3677 				    size_t outlen_actual)
3678 {
3679 	struct efx_ef10_filter_table *table = efx->filter_state;
3680 	struct efx_filter_spec *spec =
3681 		efx_ef10_filter_entry_spec(table, filter_idx);
3682 
3683 	spin_lock_bh(&efx->filter_lock);
3684 	if (rc == 0) {
3685 		kfree(spec);
3686 		efx_ef10_filter_set_entry(table, filter_idx, NULL, 0);
3687 	}
3688 	table->entry[filter_idx].spec &= ~EFX_EF10_FILTER_FLAG_BUSY;
3689 	wake_up_all(&table->waitq);
3690 	spin_unlock_bh(&efx->filter_lock);
3691 }
3692 
3693 #endif /* CONFIG_RFS_ACCEL */
3694 
3695 static int efx_ef10_filter_match_flags_from_mcdi(u32 mcdi_flags)
3696 {
3697 	int match_flags = 0;
3698 
3699 #define MAP_FLAG(gen_flag, mcdi_field) {				\
3700 		u32 old_mcdi_flags = mcdi_flags;			\
3701 		mcdi_flags &= ~(1 << MC_CMD_FILTER_OP_IN_MATCH_ ##	\
3702 				mcdi_field ## _LBN);			\
3703 		if (mcdi_flags != old_mcdi_flags)			\
3704 			match_flags |= EFX_FILTER_MATCH_ ## gen_flag;	\
3705 	}
3706 	MAP_FLAG(LOC_MAC_IG, UNKNOWN_UCAST_DST);
3707 	MAP_FLAG(LOC_MAC_IG, UNKNOWN_MCAST_DST);
3708 	MAP_FLAG(REM_HOST, SRC_IP);
3709 	MAP_FLAG(LOC_HOST, DST_IP);
3710 	MAP_FLAG(REM_MAC, SRC_MAC);
3711 	MAP_FLAG(REM_PORT, SRC_PORT);
3712 	MAP_FLAG(LOC_MAC, DST_MAC);
3713 	MAP_FLAG(LOC_PORT, DST_PORT);
3714 	MAP_FLAG(ETHER_TYPE, ETHER_TYPE);
3715 	MAP_FLAG(INNER_VID, INNER_VLAN);
3716 	MAP_FLAG(OUTER_VID, OUTER_VLAN);
3717 	MAP_FLAG(IP_PROTO, IP_PROTO);
3718 #undef MAP_FLAG
3719 
3720 	/* Did we map them all? */
3721 	if (mcdi_flags)
3722 		return -EINVAL;
3723 
3724 	return match_flags;
3725 }
3726 
3727 static int efx_ef10_filter_table_probe(struct efx_nic *efx)
3728 {
3729 	MCDI_DECLARE_BUF(inbuf, MC_CMD_GET_PARSER_DISP_INFO_IN_LEN);
3730 	MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_PARSER_DISP_INFO_OUT_LENMAX);
3731 	unsigned int pd_match_pri, pd_match_count;
3732 	struct efx_ef10_filter_table *table;
3733 	size_t outlen;
3734 	int rc;
3735 
3736 	table = kzalloc(sizeof(*table), GFP_KERNEL);
3737 	if (!table)
3738 		return -ENOMEM;
3739 
3740 	/* Find out which RX filter types are supported, and their priorities */
3741 	MCDI_SET_DWORD(inbuf, GET_PARSER_DISP_INFO_IN_OP,
3742 		       MC_CMD_GET_PARSER_DISP_INFO_IN_OP_GET_SUPPORTED_RX_MATCHES);
3743 	rc = efx_mcdi_rpc(efx, MC_CMD_GET_PARSER_DISP_INFO,
3744 			  inbuf, sizeof(inbuf), outbuf, sizeof(outbuf),
3745 			  &outlen);
3746 	if (rc)
3747 		goto fail;
3748 	pd_match_count = MCDI_VAR_ARRAY_LEN(
3749 		outlen, GET_PARSER_DISP_INFO_OUT_SUPPORTED_MATCHES);
3750 	table->rx_match_count = 0;
3751 
3752 	for (pd_match_pri = 0; pd_match_pri < pd_match_count; pd_match_pri++) {
3753 		u32 mcdi_flags =
3754 			MCDI_ARRAY_DWORD(
3755 				outbuf,
3756 				GET_PARSER_DISP_INFO_OUT_SUPPORTED_MATCHES,
3757 				pd_match_pri);
3758 		rc = efx_ef10_filter_match_flags_from_mcdi(mcdi_flags);
3759 		if (rc < 0) {
3760 			netif_dbg(efx, probe, efx->net_dev,
3761 				  "%s: fw flags %#x pri %u not supported in driver\n",
3762 				  __func__, mcdi_flags, pd_match_pri);
3763 		} else {
3764 			netif_dbg(efx, probe, efx->net_dev,
3765 				  "%s: fw flags %#x pri %u supported as driver flags %#x pri %u\n",
3766 				  __func__, mcdi_flags, pd_match_pri,
3767 				  rc, table->rx_match_count);
3768 			table->rx_match_flags[table->rx_match_count++] = rc;
3769 		}
3770 	}
3771 
3772 	table->entry = vzalloc(HUNT_FILTER_TBL_ROWS * sizeof(*table->entry));
3773 	if (!table->entry) {
3774 		rc = -ENOMEM;
3775 		goto fail;
3776 	}
3777 
3778 	table->ucdef_id = EFX_EF10_FILTER_ID_INVALID;
3779 	table->bcast_id = EFX_EF10_FILTER_ID_INVALID;
3780 	table->mcdef_id = EFX_EF10_FILTER_ID_INVALID;
3781 
3782 	efx->filter_state = table;
3783 	init_waitqueue_head(&table->waitq);
3784 	return 0;
3785 
3786 fail:
3787 	kfree(table);
3788 	return rc;
3789 }
3790 
3791 /* Caller must hold efx->filter_sem for read if race against
3792  * efx_ef10_filter_table_remove() is possible
3793  */
3794 static void efx_ef10_filter_table_restore(struct efx_nic *efx)
3795 {
3796 	struct efx_ef10_filter_table *table = efx->filter_state;
3797 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
3798 	struct efx_filter_spec *spec;
3799 	unsigned int filter_idx;
3800 	bool failed = false;
3801 	int rc;
3802 
3803 	WARN_ON(!rwsem_is_locked(&efx->filter_sem));
3804 
3805 	if (!nic_data->must_restore_filters)
3806 		return;
3807 
3808 	if (!table)
3809 		return;
3810 
3811 	spin_lock_bh(&efx->filter_lock);
3812 
3813 	for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
3814 		spec = efx_ef10_filter_entry_spec(table, filter_idx);
3815 		if (!spec)
3816 			continue;
3817 
3818 		table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_BUSY;
3819 		spin_unlock_bh(&efx->filter_lock);
3820 
3821 		rc = efx_ef10_filter_push(efx, spec,
3822 					  &table->entry[filter_idx].handle,
3823 					  false);
3824 		if (rc)
3825 			failed = true;
3826 
3827 		spin_lock_bh(&efx->filter_lock);
3828 		if (rc) {
3829 			kfree(spec);
3830 			efx_ef10_filter_set_entry(table, filter_idx, NULL, 0);
3831 		} else {
3832 			table->entry[filter_idx].spec &=
3833 				~EFX_EF10_FILTER_FLAG_BUSY;
3834 		}
3835 	}
3836 
3837 	spin_unlock_bh(&efx->filter_lock);
3838 
3839 	if (failed)
3840 		netif_err(efx, hw, efx->net_dev,
3841 			  "unable to restore all filters\n");
3842 	else
3843 		nic_data->must_restore_filters = false;
3844 }
3845 
3846 /* Caller must hold efx->filter_sem for write */
3847 static void efx_ef10_filter_table_remove(struct efx_nic *efx)
3848 {
3849 	struct efx_ef10_filter_table *table = efx->filter_state;
3850 	MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_IN_LEN);
3851 	struct efx_filter_spec *spec;
3852 	unsigned int filter_idx;
3853 	int rc;
3854 
3855 	efx->filter_state = NULL;
3856 	if (!table)
3857 		return;
3858 
3859 	for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
3860 		spec = efx_ef10_filter_entry_spec(table, filter_idx);
3861 		if (!spec)
3862 			continue;
3863 
3864 		MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
3865 			       efx_ef10_filter_is_exclusive(spec) ?
3866 			       MC_CMD_FILTER_OP_IN_OP_REMOVE :
3867 			       MC_CMD_FILTER_OP_IN_OP_UNSUBSCRIBE);
3868 		MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
3869 			       table->entry[filter_idx].handle);
3870 		rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FILTER_OP, inbuf,
3871 					sizeof(inbuf), NULL, 0, NULL);
3872 		if (rc)
3873 			netif_info(efx, drv, efx->net_dev,
3874 				   "%s: filter %04x remove failed\n",
3875 				   __func__, filter_idx);
3876 		kfree(spec);
3877 	}
3878 
3879 	vfree(table->entry);
3880 	kfree(table);
3881 }
3882 
3883 #define EFX_EF10_FILTER_DO_MARK_OLD(id) \
3884 	if (id != EFX_EF10_FILTER_ID_INVALID) { \
3885 		filter_idx = efx_ef10_filter_get_unsafe_id(efx, id); \
3886 		if (!table->entry[filter_idx].spec) \
3887 			netif_dbg(efx, drv, efx->net_dev, \
3888 				  "%s: marked null spec old %04x:%04x\n", \
3889 				  __func__, id, filter_idx); \
3890 		table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_AUTO_OLD;\
3891 	}
3892 static void efx_ef10_filter_mark_old(struct efx_nic *efx)
3893 {
3894 	struct efx_ef10_filter_table *table = efx->filter_state;
3895 	unsigned int filter_idx, i;
3896 
3897 	if (!table)
3898 		return;
3899 
3900 	/* Mark old filters that may need to be removed */
3901 	spin_lock_bh(&efx->filter_lock);
3902 	for (i = 0; i < table->dev_uc_count; i++)
3903 		EFX_EF10_FILTER_DO_MARK_OLD(table->dev_uc_list[i].id);
3904 	for (i = 0; i < table->dev_mc_count; i++)
3905 		EFX_EF10_FILTER_DO_MARK_OLD(table->dev_mc_list[i].id);
3906 	EFX_EF10_FILTER_DO_MARK_OLD(table->ucdef_id);
3907 	EFX_EF10_FILTER_DO_MARK_OLD(table->bcast_id);
3908 	EFX_EF10_FILTER_DO_MARK_OLD(table->mcdef_id);
3909 	spin_unlock_bh(&efx->filter_lock);
3910 }
3911 #undef EFX_EF10_FILTER_DO_MARK_OLD
3912 
3913 static void efx_ef10_filter_uc_addr_list(struct efx_nic *efx, bool *promisc)
3914 {
3915 	struct efx_ef10_filter_table *table = efx->filter_state;
3916 	struct net_device *net_dev = efx->net_dev;
3917 	struct netdev_hw_addr *uc;
3918 	int addr_count;
3919 	unsigned int i;
3920 
3921 	table->ucdef_id = EFX_EF10_FILTER_ID_INVALID;
3922 	addr_count = netdev_uc_count(net_dev);
3923 	if (net_dev->flags & IFF_PROMISC)
3924 		*promisc = true;
3925 	table->dev_uc_count = 1 + addr_count;
3926 	ether_addr_copy(table->dev_uc_list[0].addr, net_dev->dev_addr);
3927 	i = 1;
3928 	netdev_for_each_uc_addr(uc, net_dev) {
3929 		if (i >= EFX_EF10_FILTER_DEV_UC_MAX) {
3930 			*promisc = true;
3931 			break;
3932 		}
3933 		ether_addr_copy(table->dev_uc_list[i].addr, uc->addr);
3934 		table->dev_uc_list[i].id = EFX_EF10_FILTER_ID_INVALID;
3935 		i++;
3936 	}
3937 }
3938 
3939 static void efx_ef10_filter_mc_addr_list(struct efx_nic *efx, bool *promisc)
3940 {
3941 	struct efx_ef10_filter_table *table = efx->filter_state;
3942 	struct net_device *net_dev = efx->net_dev;
3943 	struct netdev_hw_addr *mc;
3944 	unsigned int i, addr_count;
3945 
3946 	table->mcdef_id = EFX_EF10_FILTER_ID_INVALID;
3947 	table->bcast_id = EFX_EF10_FILTER_ID_INVALID;
3948 	if (net_dev->flags & (IFF_PROMISC | IFF_ALLMULTI))
3949 		*promisc = true;
3950 
3951 	addr_count = netdev_mc_count(net_dev);
3952 	i = 0;
3953 	netdev_for_each_mc_addr(mc, net_dev) {
3954 		if (i >= EFX_EF10_FILTER_DEV_MC_MAX) {
3955 			*promisc = true;
3956 			break;
3957 		}
3958 		ether_addr_copy(table->dev_mc_list[i].addr, mc->addr);
3959 		table->dev_mc_list[i].id = EFX_EF10_FILTER_ID_INVALID;
3960 		i++;
3961 	}
3962 
3963 	table->dev_mc_count = i;
3964 }
3965 
3966 static int efx_ef10_filter_insert_addr_list(struct efx_nic *efx,
3967 					     bool multicast, bool rollback)
3968 {
3969 	struct efx_ef10_filter_table *table = efx->filter_state;
3970 	struct efx_ef10_dev_addr *addr_list;
3971 	enum efx_filter_flags filter_flags;
3972 	struct efx_filter_spec spec;
3973 	u8 baddr[ETH_ALEN];
3974 	unsigned int i, j;
3975 	int addr_count;
3976 	int rc;
3977 
3978 	if (multicast) {
3979 		addr_list = table->dev_mc_list;
3980 		addr_count = table->dev_mc_count;
3981 	} else {
3982 		addr_list = table->dev_uc_list;
3983 		addr_count = table->dev_uc_count;
3984 	}
3985 
3986 	filter_flags = efx_rss_enabled(efx) ? EFX_FILTER_FLAG_RX_RSS : 0;
3987 
3988 	/* Insert/renew filters */
3989 	for (i = 0; i < addr_count; i++) {
3990 		efx_filter_init_rx(&spec, EFX_FILTER_PRI_AUTO, filter_flags, 0);
3991 		efx_filter_set_eth_local(&spec, EFX_FILTER_VID_UNSPEC,
3992 					 addr_list[i].addr);
3993 		rc = efx_ef10_filter_insert(efx, &spec, true);
3994 		if (rc < 0) {
3995 			if (rollback) {
3996 				netif_info(efx, drv, efx->net_dev,
3997 					   "efx_ef10_filter_insert failed rc=%d\n",
3998 					   rc);
3999 				/* Fall back to promiscuous */
4000 				for (j = 0; j < i; j++) {
4001 					if (addr_list[j].id == EFX_EF10_FILTER_ID_INVALID)
4002 						continue;
4003 					efx_ef10_filter_remove_unsafe(
4004 						efx, EFX_FILTER_PRI_AUTO,
4005 						addr_list[j].id);
4006 					addr_list[j].id = EFX_EF10_FILTER_ID_INVALID;
4007 				}
4008 				return rc;
4009 			} else {
4010 				/* mark as not inserted, and carry on */
4011 				rc = EFX_EF10_FILTER_ID_INVALID;
4012 			}
4013 		}
4014 		addr_list[i].id = efx_ef10_filter_get_unsafe_id(efx, rc);
4015 	}
4016 
4017 	if (multicast && rollback) {
4018 		/* Also need an Ethernet broadcast filter */
4019 		efx_filter_init_rx(&spec, EFX_FILTER_PRI_AUTO, filter_flags, 0);
4020 		eth_broadcast_addr(baddr);
4021 		efx_filter_set_eth_local(&spec, EFX_FILTER_VID_UNSPEC, baddr);
4022 		rc = efx_ef10_filter_insert(efx, &spec, true);
4023 		if (rc < 0) {
4024 			netif_warn(efx, drv, efx->net_dev,
4025 				   "Broadcast filter insert failed rc=%d\n", rc);
4026 			/* Fall back to promiscuous */
4027 			for (j = 0; j < i; j++) {
4028 				if (addr_list[j].id == EFX_EF10_FILTER_ID_INVALID)
4029 					continue;
4030 				efx_ef10_filter_remove_unsafe(
4031 					efx, EFX_FILTER_PRI_AUTO,
4032 					addr_list[j].id);
4033 				addr_list[j].id = EFX_EF10_FILTER_ID_INVALID;
4034 			}
4035 			return rc;
4036 		} else {
4037 			table->bcast_id = efx_ef10_filter_get_unsafe_id(efx, rc);
4038 		}
4039 	}
4040 
4041 	return 0;
4042 }
4043 
4044 static int efx_ef10_filter_insert_def(struct efx_nic *efx, bool multicast,
4045 				      bool rollback)
4046 {
4047 	struct efx_ef10_filter_table *table = efx->filter_state;
4048 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
4049 	enum efx_filter_flags filter_flags;
4050 	struct efx_filter_spec spec;
4051 	u8 baddr[ETH_ALEN];
4052 	int rc;
4053 
4054 	filter_flags = efx_rss_enabled(efx) ? EFX_FILTER_FLAG_RX_RSS : 0;
4055 
4056 	efx_filter_init_rx(&spec, EFX_FILTER_PRI_AUTO, filter_flags, 0);
4057 
4058 	if (multicast)
4059 		efx_filter_set_mc_def(&spec);
4060 	else
4061 		efx_filter_set_uc_def(&spec);
4062 
4063 	rc = efx_ef10_filter_insert(efx, &spec, true);
4064 	if (rc < 0) {
4065 		netif_printk(efx, drv, rc == -EPERM ? KERN_DEBUG : KERN_WARNING,
4066 			     efx->net_dev,
4067 			     "%scast mismatch filter insert failed rc=%d\n",
4068 			     multicast ? "Multi" : "Uni", rc);
4069 	} else if (multicast) {
4070 		table->mcdef_id = efx_ef10_filter_get_unsafe_id(efx, rc);
4071 		if (!nic_data->workaround_26807) {
4072 			/* Also need an Ethernet broadcast filter */
4073 			efx_filter_init_rx(&spec, EFX_FILTER_PRI_AUTO,
4074 					   filter_flags, 0);
4075 			eth_broadcast_addr(baddr);
4076 			efx_filter_set_eth_local(&spec, EFX_FILTER_VID_UNSPEC,
4077 						 baddr);
4078 			rc = efx_ef10_filter_insert(efx, &spec, true);
4079 			if (rc < 0) {
4080 				netif_warn(efx, drv, efx->net_dev,
4081 					   "Broadcast filter insert failed rc=%d\n",
4082 					   rc);
4083 				if (rollback) {
4084 					/* Roll back the mc_def filter */
4085 					efx_ef10_filter_remove_unsafe(
4086 							efx, EFX_FILTER_PRI_AUTO,
4087 							table->mcdef_id);
4088 					table->mcdef_id = EFX_EF10_FILTER_ID_INVALID;
4089 					return rc;
4090 				}
4091 			} else {
4092 				table->bcast_id = efx_ef10_filter_get_unsafe_id(efx, rc);
4093 			}
4094 		}
4095 		rc = 0;
4096 	} else {
4097 		table->ucdef_id = rc;
4098 		rc = 0;
4099 	}
4100 	return rc;
4101 }
4102 
4103 /* Remove filters that weren't renewed.  Since nothing else changes the AUTO_OLD
4104  * flag or removes these filters, we don't need to hold the filter_lock while
4105  * scanning for these filters.
4106  */
4107 static void efx_ef10_filter_remove_old(struct efx_nic *efx)
4108 {
4109 	struct efx_ef10_filter_table *table = efx->filter_state;
4110 	int remove_failed = 0;
4111 	int remove_noent = 0;
4112 	int rc;
4113 	int i;
4114 
4115 	for (i = 0; i < HUNT_FILTER_TBL_ROWS; i++) {
4116 		if (ACCESS_ONCE(table->entry[i].spec) &
4117 		    EFX_EF10_FILTER_FLAG_AUTO_OLD) {
4118 			rc = efx_ef10_filter_remove_internal(efx,
4119 					1U << EFX_FILTER_PRI_AUTO, i, true);
4120 			if (rc == -ENOENT)
4121 				remove_noent++;
4122 			else if (rc)
4123 				remove_failed++;
4124 		}
4125 	}
4126 
4127 	if (remove_failed)
4128 		netif_info(efx, drv, efx->net_dev,
4129 			   "%s: failed to remove %d filters\n",
4130 			   __func__, remove_failed);
4131 	if (remove_noent)
4132 		netif_info(efx, drv, efx->net_dev,
4133 			   "%s: failed to remove %d non-existent filters\n",
4134 			   __func__, remove_noent);
4135 }
4136 
4137 static int efx_ef10_vport_set_mac_address(struct efx_nic *efx)
4138 {
4139 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
4140 	u8 mac_old[ETH_ALEN];
4141 	int rc, rc2;
4142 
4143 	/* Only reconfigure a PF-created vport */
4144 	if (is_zero_ether_addr(nic_data->vport_mac))
4145 		return 0;
4146 
4147 	efx_device_detach_sync(efx);
4148 	efx_net_stop(efx->net_dev);
4149 	down_write(&efx->filter_sem);
4150 	efx_ef10_filter_table_remove(efx);
4151 	up_write(&efx->filter_sem);
4152 
4153 	rc = efx_ef10_vadaptor_free(efx, nic_data->vport_id);
4154 	if (rc)
4155 		goto restore_filters;
4156 
4157 	ether_addr_copy(mac_old, nic_data->vport_mac);
4158 	rc = efx_ef10_vport_del_mac(efx, nic_data->vport_id,
4159 				    nic_data->vport_mac);
4160 	if (rc)
4161 		goto restore_vadaptor;
4162 
4163 	rc = efx_ef10_vport_add_mac(efx, nic_data->vport_id,
4164 				    efx->net_dev->dev_addr);
4165 	if (!rc) {
4166 		ether_addr_copy(nic_data->vport_mac, efx->net_dev->dev_addr);
4167 	} else {
4168 		rc2 = efx_ef10_vport_add_mac(efx, nic_data->vport_id, mac_old);
4169 		if (rc2) {
4170 			/* Failed to add original MAC, so clear vport_mac */
4171 			eth_zero_addr(nic_data->vport_mac);
4172 			goto reset_nic;
4173 		}
4174 	}
4175 
4176 restore_vadaptor:
4177 	rc2 = efx_ef10_vadaptor_alloc(efx, nic_data->vport_id);
4178 	if (rc2)
4179 		goto reset_nic;
4180 restore_filters:
4181 	down_write(&efx->filter_sem);
4182 	rc2 = efx_ef10_filter_table_probe(efx);
4183 	up_write(&efx->filter_sem);
4184 	if (rc2)
4185 		goto reset_nic;
4186 
4187 	rc2 = efx_net_open(efx->net_dev);
4188 	if (rc2)
4189 		goto reset_nic;
4190 
4191 	netif_device_attach(efx->net_dev);
4192 
4193 	return rc;
4194 
4195 reset_nic:
4196 	netif_err(efx, drv, efx->net_dev,
4197 		  "Failed to restore when changing MAC address - scheduling reset\n");
4198 	efx_schedule_reset(efx, RESET_TYPE_DATAPATH);
4199 
4200 	return rc ? rc : rc2;
4201 }
4202 
4203 /* Caller must hold efx->filter_sem for read if race against
4204  * efx_ef10_filter_table_remove() is possible
4205  */
4206 static void efx_ef10_filter_sync_rx_mode(struct efx_nic *efx)
4207 {
4208 	struct efx_ef10_filter_table *table = efx->filter_state;
4209 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
4210 	struct net_device *net_dev = efx->net_dev;
4211 	bool uc_promisc = false, mc_promisc = false;
4212 
4213 	if (!efx_dev_registered(efx))
4214 		return;
4215 
4216 	if (!table)
4217 		return;
4218 
4219 	efx_ef10_filter_mark_old(efx);
4220 
4221 	/* Copy/convert the address lists; add the primary station
4222 	 * address and broadcast address
4223 	 */
4224 	netif_addr_lock_bh(net_dev);
4225 	efx_ef10_filter_uc_addr_list(efx, &uc_promisc);
4226 	efx_ef10_filter_mc_addr_list(efx, &mc_promisc);
4227 	netif_addr_unlock_bh(net_dev);
4228 
4229 	/* Insert/renew unicast filters */
4230 	if (uc_promisc) {
4231 		efx_ef10_filter_insert_def(efx, false, false);
4232 		efx_ef10_filter_insert_addr_list(efx, false, false);
4233 	} else {
4234 		/* If any of the filters failed to insert, fall back to
4235 		 * promiscuous mode - add in the uc_def filter.  But keep
4236 		 * our individual unicast filters.
4237 		 */
4238 		if (efx_ef10_filter_insert_addr_list(efx, false, false))
4239 			efx_ef10_filter_insert_def(efx, false, false);
4240 	}
4241 
4242 	/* Insert/renew multicast filters */
4243 	/* If changing promiscuous state with cascaded multicast filters, remove
4244 	 * old filters first, so that packets are dropped rather than duplicated
4245 	 */
4246 	if (nic_data->workaround_26807 && efx->mc_promisc != mc_promisc)
4247 		efx_ef10_filter_remove_old(efx);
4248 	if (mc_promisc) {
4249 		if (nic_data->workaround_26807) {
4250 			/* If we failed to insert promiscuous filters, rollback
4251 			 * and fall back to individual multicast filters
4252 			 */
4253 			if (efx_ef10_filter_insert_def(efx, true, true)) {
4254 				/* Changing promisc state, so remove old filters */
4255 				efx_ef10_filter_remove_old(efx);
4256 				efx_ef10_filter_insert_addr_list(efx, true, false);
4257 			}
4258 		} else {
4259 			/* If we failed to insert promiscuous filters, don't
4260 			 * rollback.  Regardless, also insert the mc_list
4261 			 */
4262 			efx_ef10_filter_insert_def(efx, true, false);
4263 			efx_ef10_filter_insert_addr_list(efx, true, false);
4264 		}
4265 	} else {
4266 		/* If any filters failed to insert, rollback and fall back to
4267 		 * promiscuous mode - mc_def filter and maybe broadcast.  If
4268 		 * that fails, roll back again and insert as many of our
4269 		 * individual multicast filters as we can.
4270 		 */
4271 		if (efx_ef10_filter_insert_addr_list(efx, true, true)) {
4272 			/* Changing promisc state, so remove old filters */
4273 			if (nic_data->workaround_26807)
4274 				efx_ef10_filter_remove_old(efx);
4275 			if (efx_ef10_filter_insert_def(efx, true, true))
4276 				efx_ef10_filter_insert_addr_list(efx, true, false);
4277 		}
4278 	}
4279 
4280 	efx_ef10_filter_remove_old(efx);
4281 	efx->mc_promisc = mc_promisc;
4282 }
4283 
4284 static int efx_ef10_set_mac_address(struct efx_nic *efx)
4285 {
4286 	MCDI_DECLARE_BUF(inbuf, MC_CMD_VADAPTOR_SET_MAC_IN_LEN);
4287 	struct efx_ef10_nic_data *nic_data = efx->nic_data;
4288 	bool was_enabled = efx->port_enabled;
4289 	int rc;
4290 
4291 	efx_device_detach_sync(efx);
4292 	efx_net_stop(efx->net_dev);
4293 	down_write(&efx->filter_sem);
4294 	efx_ef10_filter_table_remove(efx);
4295 
4296 	ether_addr_copy(MCDI_PTR(inbuf, VADAPTOR_SET_MAC_IN_MACADDR),
4297 			efx->net_dev->dev_addr);
4298 	MCDI_SET_DWORD(inbuf, VADAPTOR_SET_MAC_IN_UPSTREAM_PORT_ID,
4299 		       nic_data->vport_id);
4300 	rc = efx_mcdi_rpc_quiet(efx, MC_CMD_VADAPTOR_SET_MAC, inbuf,
4301 				sizeof(inbuf), NULL, 0, NULL);
4302 
4303 	efx_ef10_filter_table_probe(efx);
4304 	up_write(&efx->filter_sem);
4305 	if (was_enabled)
4306 		efx_net_open(efx->net_dev);
4307 	netif_device_attach(efx->net_dev);
4308 
4309 #ifdef CONFIG_SFC_SRIOV
4310 	if (efx->pci_dev->is_virtfn && efx->pci_dev->physfn) {
4311 		struct pci_dev *pci_dev_pf = efx->pci_dev->physfn;
4312 
4313 		if (rc == -EPERM) {
4314 			struct efx_nic *efx_pf;
4315 
4316 			/* Switch to PF and change MAC address on vport */
4317 			efx_pf = pci_get_drvdata(pci_dev_pf);
4318 
4319 			rc = efx_ef10_sriov_set_vf_mac(efx_pf,
4320 						       nic_data->vf_index,
4321 						       efx->net_dev->dev_addr);
4322 		} else if (!rc) {
4323 			struct efx_nic *efx_pf = pci_get_drvdata(pci_dev_pf);
4324 			struct efx_ef10_nic_data *nic_data = efx_pf->nic_data;
4325 			unsigned int i;
4326 
4327 			/* MAC address successfully changed by VF (with MAC
4328 			 * spoofing) so update the parent PF if possible.
4329 			 */
4330 			for (i = 0; i < efx_pf->vf_count; ++i) {
4331 				struct ef10_vf *vf = nic_data->vf + i;
4332 
4333 				if (vf->efx == efx) {
4334 					ether_addr_copy(vf->mac,
4335 							efx->net_dev->dev_addr);
4336 					return 0;
4337 				}
4338 			}
4339 		}
4340 	} else
4341 #endif
4342 	if (rc == -EPERM) {
4343 		netif_err(efx, drv, efx->net_dev,
4344 			  "Cannot change MAC address; use sfboot to enable"
4345 			  " mac-spoofing on this interface\n");
4346 	} else if (rc == -ENOSYS && !efx_ef10_is_vf(efx)) {
4347 		/* If the active MCFW does not support MC_CMD_VADAPTOR_SET_MAC
4348 		 * fall-back to the method of changing the MAC address on the
4349 		 * vport.  This only applies to PFs because such versions of
4350 		 * MCFW do not support VFs.
4351 		 */
4352 		rc = efx_ef10_vport_set_mac_address(efx);
4353 	} else {
4354 		efx_mcdi_display_error(efx, MC_CMD_VADAPTOR_SET_MAC,
4355 				       sizeof(inbuf), NULL, 0, rc);
4356 	}
4357 
4358 	return rc;
4359 }
4360 
4361 static int efx_ef10_mac_reconfigure(struct efx_nic *efx)
4362 {
4363 	efx_ef10_filter_sync_rx_mode(efx);
4364 
4365 	return efx_mcdi_set_mac(efx);
4366 }
4367 
4368 static int efx_ef10_mac_reconfigure_vf(struct efx_nic *efx)
4369 {
4370 	efx_ef10_filter_sync_rx_mode(efx);
4371 
4372 	return 0;
4373 }
4374 
4375 static int efx_ef10_start_bist(struct efx_nic *efx, u32 bist_type)
4376 {
4377 	MCDI_DECLARE_BUF(inbuf, MC_CMD_START_BIST_IN_LEN);
4378 
4379 	MCDI_SET_DWORD(inbuf, START_BIST_IN_TYPE, bist_type);
4380 	return efx_mcdi_rpc(efx, MC_CMD_START_BIST, inbuf, sizeof(inbuf),
4381 			    NULL, 0, NULL);
4382 }
4383 
4384 /* MC BISTs follow a different poll mechanism to phy BISTs.
4385  * The BIST is done in the poll handler on the MC, and the MCDI command
4386  * will block until the BIST is done.
4387  */
4388 static int efx_ef10_poll_bist(struct efx_nic *efx)
4389 {
4390 	int rc;
4391 	MCDI_DECLARE_BUF(outbuf, MC_CMD_POLL_BIST_OUT_LEN);
4392 	size_t outlen;
4393 	u32 result;
4394 
4395 	rc = efx_mcdi_rpc(efx, MC_CMD_POLL_BIST, NULL, 0,
4396 			   outbuf, sizeof(outbuf), &outlen);
4397 	if (rc != 0)
4398 		return rc;
4399 
4400 	if (outlen < MC_CMD_POLL_BIST_OUT_LEN)
4401 		return -EIO;
4402 
4403 	result = MCDI_DWORD(outbuf, POLL_BIST_OUT_RESULT);
4404 	switch (result) {
4405 	case MC_CMD_POLL_BIST_PASSED:
4406 		netif_dbg(efx, hw, efx->net_dev, "BIST passed.\n");
4407 		return 0;
4408 	case MC_CMD_POLL_BIST_TIMEOUT:
4409 		netif_err(efx, hw, efx->net_dev, "BIST timed out\n");
4410 		return -EIO;
4411 	case MC_CMD_POLL_BIST_FAILED:
4412 		netif_err(efx, hw, efx->net_dev, "BIST failed.\n");
4413 		return -EIO;
4414 	default:
4415 		netif_err(efx, hw, efx->net_dev,
4416 			  "BIST returned unknown result %u", result);
4417 		return -EIO;
4418 	}
4419 }
4420 
4421 static int efx_ef10_run_bist(struct efx_nic *efx, u32 bist_type)
4422 {
4423 	int rc;
4424 
4425 	netif_dbg(efx, drv, efx->net_dev, "starting BIST type %u\n", bist_type);
4426 
4427 	rc = efx_ef10_start_bist(efx, bist_type);
4428 	if (rc != 0)
4429 		return rc;
4430 
4431 	return efx_ef10_poll_bist(efx);
4432 }
4433 
4434 static int
4435 efx_ef10_test_chip(struct efx_nic *efx, struct efx_self_tests *tests)
4436 {
4437 	int rc, rc2;
4438 
4439 	efx_reset_down(efx, RESET_TYPE_WORLD);
4440 
4441 	rc = efx_mcdi_rpc(efx, MC_CMD_ENABLE_OFFLINE_BIST,
4442 			  NULL, 0, NULL, 0, NULL);
4443 	if (rc != 0)
4444 		goto out;
4445 
4446 	tests->memory = efx_ef10_run_bist(efx, MC_CMD_MC_MEM_BIST) ? -1 : 1;
4447 	tests->registers = efx_ef10_run_bist(efx, MC_CMD_REG_BIST) ? -1 : 1;
4448 
4449 	rc = efx_mcdi_reset(efx, RESET_TYPE_WORLD);
4450 
4451 out:
4452 	if (rc == -EPERM)
4453 		rc = 0;
4454 	rc2 = efx_reset_up(efx, RESET_TYPE_WORLD, rc == 0);
4455 	return rc ? rc : rc2;
4456 }
4457 
4458 #ifdef CONFIG_SFC_MTD
4459 
4460 struct efx_ef10_nvram_type_info {
4461 	u16 type, type_mask;
4462 	u8 port;
4463 	const char *name;
4464 };
4465 
4466 static const struct efx_ef10_nvram_type_info efx_ef10_nvram_types[] = {
4467 	{ NVRAM_PARTITION_TYPE_MC_FIRMWARE,	   0,    0, "sfc_mcfw" },
4468 	{ NVRAM_PARTITION_TYPE_MC_FIRMWARE_BACKUP, 0,    0, "sfc_mcfw_backup" },
4469 	{ NVRAM_PARTITION_TYPE_EXPANSION_ROM,	   0,    0, "sfc_exp_rom" },
4470 	{ NVRAM_PARTITION_TYPE_STATIC_CONFIG,	   0,    0, "sfc_static_cfg" },
4471 	{ NVRAM_PARTITION_TYPE_DYNAMIC_CONFIG,	   0,    0, "sfc_dynamic_cfg" },
4472 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT0, 0,   0, "sfc_exp_rom_cfg" },
4473 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT1, 0,   1, "sfc_exp_rom_cfg" },
4474 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT2, 0,   2, "sfc_exp_rom_cfg" },
4475 	{ NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT3, 0,   3, "sfc_exp_rom_cfg" },
4476 	{ NVRAM_PARTITION_TYPE_LICENSE,		   0,    0, "sfc_license" },
4477 	{ NVRAM_PARTITION_TYPE_PHY_MIN,		   0xff, 0, "sfc_phy_fw" },
4478 };
4479 
4480 static int efx_ef10_mtd_probe_partition(struct efx_nic *efx,
4481 					struct efx_mcdi_mtd_partition *part,
4482 					unsigned int type)
4483 {
4484 	MCDI_DECLARE_BUF(inbuf, MC_CMD_NVRAM_METADATA_IN_LEN);
4485 	MCDI_DECLARE_BUF(outbuf, MC_CMD_NVRAM_METADATA_OUT_LENMAX);
4486 	const struct efx_ef10_nvram_type_info *info;
4487 	size_t size, erase_size, outlen;
4488 	bool protected;
4489 	int rc;
4490 
4491 	for (info = efx_ef10_nvram_types; ; info++) {
4492 		if (info ==
4493 		    efx_ef10_nvram_types + ARRAY_SIZE(efx_ef10_nvram_types))
4494 			return -ENODEV;
4495 		if ((type & ~info->type_mask) == info->type)
4496 			break;
4497 	}
4498 	if (info->port != efx_port_num(efx))
4499 		return -ENODEV;
4500 
4501 	rc = efx_mcdi_nvram_info(efx, type, &size, &erase_size, &protected);
4502 	if (rc)
4503 		return rc;
4504 	if (protected)
4505 		return -ENODEV; /* hide it */
4506 
4507 	part->nvram_type = type;
4508 
4509 	MCDI_SET_DWORD(inbuf, NVRAM_METADATA_IN_TYPE, type);
4510 	rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_METADATA, inbuf, sizeof(inbuf),
4511 			  outbuf, sizeof(outbuf), &outlen);
4512 	if (rc)
4513 		return rc;
4514 	if (outlen < MC_CMD_NVRAM_METADATA_OUT_LENMIN)
4515 		return -EIO;
4516 	if (MCDI_DWORD(outbuf, NVRAM_METADATA_OUT_FLAGS) &
4517 	    (1 << MC_CMD_NVRAM_METADATA_OUT_SUBTYPE_VALID_LBN))
4518 		part->fw_subtype = MCDI_DWORD(outbuf,
4519 					      NVRAM_METADATA_OUT_SUBTYPE);
4520 
4521 	part->common.dev_type_name = "EF10 NVRAM manager";
4522 	part->common.type_name = info->name;
4523 
4524 	part->common.mtd.type = MTD_NORFLASH;
4525 	part->common.mtd.flags = MTD_CAP_NORFLASH;
4526 	part->common.mtd.size = size;
4527 	part->common.mtd.erasesize = erase_size;
4528 
4529 	return 0;
4530 }
4531 
4532 static int efx_ef10_mtd_probe(struct efx_nic *efx)
4533 {
4534 	MCDI_DECLARE_BUF(outbuf, MC_CMD_NVRAM_PARTITIONS_OUT_LENMAX);
4535 	struct efx_mcdi_mtd_partition *parts;
4536 	size_t outlen, n_parts_total, i, n_parts;
4537 	unsigned int type;
4538 	int rc;
4539 
4540 	ASSERT_RTNL();
4541 
4542 	BUILD_BUG_ON(MC_CMD_NVRAM_PARTITIONS_IN_LEN != 0);
4543 	rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_PARTITIONS, NULL, 0,
4544 			  outbuf, sizeof(outbuf), &outlen);
4545 	if (rc)
4546 		return rc;
4547 	if (outlen < MC_CMD_NVRAM_PARTITIONS_OUT_LENMIN)
4548 		return -EIO;
4549 
4550 	n_parts_total = MCDI_DWORD(outbuf, NVRAM_PARTITIONS_OUT_NUM_PARTITIONS);
4551 	if (n_parts_total >
4552 	    MCDI_VAR_ARRAY_LEN(outlen, NVRAM_PARTITIONS_OUT_TYPE_ID))
4553 		return -EIO;
4554 
4555 	parts = kcalloc(n_parts_total, sizeof(*parts), GFP_KERNEL);
4556 	if (!parts)
4557 		return -ENOMEM;
4558 
4559 	n_parts = 0;
4560 	for (i = 0; i < n_parts_total; i++) {
4561 		type = MCDI_ARRAY_DWORD(outbuf, NVRAM_PARTITIONS_OUT_TYPE_ID,
4562 					i);
4563 		rc = efx_ef10_mtd_probe_partition(efx, &parts[n_parts], type);
4564 		if (rc == 0)
4565 			n_parts++;
4566 		else if (rc != -ENODEV)
4567 			goto fail;
4568 	}
4569 
4570 	rc = efx_mtd_add(efx, &parts[0].common, n_parts, sizeof(*parts));
4571 fail:
4572 	if (rc)
4573 		kfree(parts);
4574 	return rc;
4575 }
4576 
4577 #endif /* CONFIG_SFC_MTD */
4578 
4579 static void efx_ef10_ptp_write_host_time(struct efx_nic *efx, u32 host_time)
4580 {
4581 	_efx_writed(efx, cpu_to_le32(host_time), ER_DZ_MC_DB_LWRD);
4582 }
4583 
4584 static void efx_ef10_ptp_write_host_time_vf(struct efx_nic *efx,
4585 					    u32 host_time) {}
4586 
4587 static int efx_ef10_rx_enable_timestamping(struct efx_channel *channel,
4588 					   bool temp)
4589 {
4590 	MCDI_DECLARE_BUF(inbuf, MC_CMD_PTP_IN_TIME_EVENT_SUBSCRIBE_LEN);
4591 	int rc;
4592 
4593 	if (channel->sync_events_state == SYNC_EVENTS_REQUESTED ||
4594 	    channel->sync_events_state == SYNC_EVENTS_VALID ||
4595 	    (temp && channel->sync_events_state == SYNC_EVENTS_DISABLED))
4596 		return 0;
4597 	channel->sync_events_state = SYNC_EVENTS_REQUESTED;
4598 
4599 	MCDI_SET_DWORD(inbuf, PTP_IN_OP, MC_CMD_PTP_OP_TIME_EVENT_SUBSCRIBE);
4600 	MCDI_SET_DWORD(inbuf, PTP_IN_PERIPH_ID, 0);
4601 	MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_SUBSCRIBE_QUEUE,
4602 		       channel->channel);
4603 
4604 	rc = efx_mcdi_rpc(channel->efx, MC_CMD_PTP,
4605 			  inbuf, sizeof(inbuf), NULL, 0, NULL);
4606 
4607 	if (rc != 0)
4608 		channel->sync_events_state = temp ? SYNC_EVENTS_QUIESCENT :
4609 						    SYNC_EVENTS_DISABLED;
4610 
4611 	return rc;
4612 }
4613 
4614 static int efx_ef10_rx_disable_timestamping(struct efx_channel *channel,
4615 					    bool temp)
4616 {
4617 	MCDI_DECLARE_BUF(inbuf, MC_CMD_PTP_IN_TIME_EVENT_UNSUBSCRIBE_LEN);
4618 	int rc;
4619 
4620 	if (channel->sync_events_state == SYNC_EVENTS_DISABLED ||
4621 	    (temp && channel->sync_events_state == SYNC_EVENTS_QUIESCENT))
4622 		return 0;
4623 	if (channel->sync_events_state == SYNC_EVENTS_QUIESCENT) {
4624 		channel->sync_events_state = SYNC_EVENTS_DISABLED;
4625 		return 0;
4626 	}
4627 	channel->sync_events_state = temp ? SYNC_EVENTS_QUIESCENT :
4628 					    SYNC_EVENTS_DISABLED;
4629 
4630 	MCDI_SET_DWORD(inbuf, PTP_IN_OP, MC_CMD_PTP_OP_TIME_EVENT_UNSUBSCRIBE);
4631 	MCDI_SET_DWORD(inbuf, PTP_IN_PERIPH_ID, 0);
4632 	MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_UNSUBSCRIBE_CONTROL,
4633 		       MC_CMD_PTP_IN_TIME_EVENT_UNSUBSCRIBE_SINGLE);
4634 	MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_UNSUBSCRIBE_QUEUE,
4635 		       channel->channel);
4636 
4637 	rc = efx_mcdi_rpc(channel->efx, MC_CMD_PTP,
4638 			  inbuf, sizeof(inbuf), NULL, 0, NULL);
4639 
4640 	return rc;
4641 }
4642 
4643 static int efx_ef10_ptp_set_ts_sync_events(struct efx_nic *efx, bool en,
4644 					   bool temp)
4645 {
4646 	int (*set)(struct efx_channel *channel, bool temp);
4647 	struct efx_channel *channel;
4648 
4649 	set = en ?
4650 	      efx_ef10_rx_enable_timestamping :
4651 	      efx_ef10_rx_disable_timestamping;
4652 
4653 	efx_for_each_channel(channel, efx) {
4654 		int rc = set(channel, temp);
4655 		if (en && rc != 0) {
4656 			efx_ef10_ptp_set_ts_sync_events(efx, false, temp);
4657 			return rc;
4658 		}
4659 	}
4660 
4661 	return 0;
4662 }
4663 
4664 static int efx_ef10_ptp_set_ts_config_vf(struct efx_nic *efx,
4665 					 struct hwtstamp_config *init)
4666 {
4667 	return -EOPNOTSUPP;
4668 }
4669 
4670 static int efx_ef10_ptp_set_ts_config(struct efx_nic *efx,
4671 				      struct hwtstamp_config *init)
4672 {
4673 	int rc;
4674 
4675 	switch (init->rx_filter) {
4676 	case HWTSTAMP_FILTER_NONE:
4677 		efx_ef10_ptp_set_ts_sync_events(efx, false, false);
4678 		/* if TX timestamping is still requested then leave PTP on */
4679 		return efx_ptp_change_mode(efx,
4680 					   init->tx_type != HWTSTAMP_TX_OFF, 0);
4681 	case HWTSTAMP_FILTER_ALL:
4682 	case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
4683 	case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
4684 	case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
4685 	case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
4686 	case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
4687 	case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
4688 	case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
4689 	case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
4690 	case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
4691 	case HWTSTAMP_FILTER_PTP_V2_EVENT:
4692 	case HWTSTAMP_FILTER_PTP_V2_SYNC:
4693 	case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
4694 		init->rx_filter = HWTSTAMP_FILTER_ALL;
4695 		rc = efx_ptp_change_mode(efx, true, 0);
4696 		if (!rc)
4697 			rc = efx_ef10_ptp_set_ts_sync_events(efx, true, false);
4698 		if (rc)
4699 			efx_ptp_change_mode(efx, false, 0);
4700 		return rc;
4701 	default:
4702 		return -ERANGE;
4703 	}
4704 }
4705 
4706 const struct efx_nic_type efx_hunt_a0_vf_nic_type = {
4707 	.is_vf = true,
4708 	.mem_bar = EFX_MEM_VF_BAR,
4709 	.mem_map_size = efx_ef10_mem_map_size,
4710 	.probe = efx_ef10_probe_vf,
4711 	.remove = efx_ef10_remove,
4712 	.dimension_resources = efx_ef10_dimension_resources,
4713 	.init = efx_ef10_init_nic,
4714 	.fini = efx_port_dummy_op_void,
4715 	.map_reset_reason = efx_ef10_map_reset_reason,
4716 	.map_reset_flags = efx_ef10_map_reset_flags,
4717 	.reset = efx_ef10_reset,
4718 	.probe_port = efx_mcdi_port_probe,
4719 	.remove_port = efx_mcdi_port_remove,
4720 	.fini_dmaq = efx_ef10_fini_dmaq,
4721 	.prepare_flr = efx_ef10_prepare_flr,
4722 	.finish_flr = efx_port_dummy_op_void,
4723 	.describe_stats = efx_ef10_describe_stats,
4724 	.update_stats = efx_ef10_update_stats_vf,
4725 	.start_stats = efx_port_dummy_op_void,
4726 	.pull_stats = efx_port_dummy_op_void,
4727 	.stop_stats = efx_port_dummy_op_void,
4728 	.set_id_led = efx_mcdi_set_id_led,
4729 	.push_irq_moderation = efx_ef10_push_irq_moderation,
4730 	.reconfigure_mac = efx_ef10_mac_reconfigure_vf,
4731 	.check_mac_fault = efx_mcdi_mac_check_fault,
4732 	.reconfigure_port = efx_mcdi_port_reconfigure,
4733 	.get_wol = efx_ef10_get_wol_vf,
4734 	.set_wol = efx_ef10_set_wol_vf,
4735 	.resume_wol = efx_port_dummy_op_void,
4736 	.mcdi_request = efx_ef10_mcdi_request,
4737 	.mcdi_poll_response = efx_ef10_mcdi_poll_response,
4738 	.mcdi_read_response = efx_ef10_mcdi_read_response,
4739 	.mcdi_poll_reboot = efx_ef10_mcdi_poll_reboot,
4740 	.mcdi_reboot_detected = efx_ef10_mcdi_reboot_detected,
4741 	.irq_enable_master = efx_port_dummy_op_void,
4742 	.irq_test_generate = efx_ef10_irq_test_generate,
4743 	.irq_disable_non_ev = efx_port_dummy_op_void,
4744 	.irq_handle_msi = efx_ef10_msi_interrupt,
4745 	.irq_handle_legacy = efx_ef10_legacy_interrupt,
4746 	.tx_probe = efx_ef10_tx_probe,
4747 	.tx_init = efx_ef10_tx_init,
4748 	.tx_remove = efx_ef10_tx_remove,
4749 	.tx_write = efx_ef10_tx_write,
4750 	.rx_push_rss_config = efx_ef10_vf_rx_push_rss_config,
4751 	.rx_probe = efx_ef10_rx_probe,
4752 	.rx_init = efx_ef10_rx_init,
4753 	.rx_remove = efx_ef10_rx_remove,
4754 	.rx_write = efx_ef10_rx_write,
4755 	.rx_defer_refill = efx_ef10_rx_defer_refill,
4756 	.ev_probe = efx_ef10_ev_probe,
4757 	.ev_init = efx_ef10_ev_init,
4758 	.ev_fini = efx_ef10_ev_fini,
4759 	.ev_remove = efx_ef10_ev_remove,
4760 	.ev_process = efx_ef10_ev_process,
4761 	.ev_read_ack = efx_ef10_ev_read_ack,
4762 	.ev_test_generate = efx_ef10_ev_test_generate,
4763 	.filter_table_probe = efx_ef10_filter_table_probe,
4764 	.filter_table_restore = efx_ef10_filter_table_restore,
4765 	.filter_table_remove = efx_ef10_filter_table_remove,
4766 	.filter_update_rx_scatter = efx_ef10_filter_update_rx_scatter,
4767 	.filter_insert = efx_ef10_filter_insert,
4768 	.filter_remove_safe = efx_ef10_filter_remove_safe,
4769 	.filter_get_safe = efx_ef10_filter_get_safe,
4770 	.filter_clear_rx = efx_ef10_filter_clear_rx,
4771 	.filter_count_rx_used = efx_ef10_filter_count_rx_used,
4772 	.filter_get_rx_id_limit = efx_ef10_filter_get_rx_id_limit,
4773 	.filter_get_rx_ids = efx_ef10_filter_get_rx_ids,
4774 #ifdef CONFIG_RFS_ACCEL
4775 	.filter_rfs_insert = efx_ef10_filter_rfs_insert,
4776 	.filter_rfs_expire_one = efx_ef10_filter_rfs_expire_one,
4777 #endif
4778 #ifdef CONFIG_SFC_MTD
4779 	.mtd_probe = efx_port_dummy_op_int,
4780 #endif
4781 	.ptp_write_host_time = efx_ef10_ptp_write_host_time_vf,
4782 	.ptp_set_ts_config = efx_ef10_ptp_set_ts_config_vf,
4783 #ifdef CONFIG_SFC_SRIOV
4784 	.vswitching_probe = efx_ef10_vswitching_probe_vf,
4785 	.vswitching_restore = efx_ef10_vswitching_restore_vf,
4786 	.vswitching_remove = efx_ef10_vswitching_remove_vf,
4787 	.sriov_get_phys_port_id = efx_ef10_sriov_get_phys_port_id,
4788 #endif
4789 	.get_mac_address = efx_ef10_get_mac_address_vf,
4790 	.set_mac_address = efx_ef10_set_mac_address,
4791 
4792 	.revision = EFX_REV_HUNT_A0,
4793 	.max_dma_mask = DMA_BIT_MASK(ESF_DZ_TX_KER_BUF_ADDR_WIDTH),
4794 	.rx_prefix_size = ES_DZ_RX_PREFIX_SIZE,
4795 	.rx_hash_offset = ES_DZ_RX_PREFIX_HASH_OFST,
4796 	.rx_ts_offset = ES_DZ_RX_PREFIX_TSTAMP_OFST,
4797 	.can_rx_scatter = true,
4798 	.always_rx_scatter = true,
4799 	.max_interrupt_mode = EFX_INT_MODE_MSIX,
4800 	.timer_period_max = 1 << ERF_DD_EVQ_IND_TIMER_VAL_WIDTH,
4801 	.offload_features = (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
4802 			     NETIF_F_RXHASH | NETIF_F_NTUPLE),
4803 	.mcdi_max_ver = 2,
4804 	.max_rx_ip_filters = HUNT_FILTER_TBL_ROWS,
4805 	.hwtstamp_filters = 1 << HWTSTAMP_FILTER_NONE |
4806 			    1 << HWTSTAMP_FILTER_ALL,
4807 };
4808 
4809 const struct efx_nic_type efx_hunt_a0_nic_type = {
4810 	.is_vf = false,
4811 	.mem_bar = EFX_MEM_BAR,
4812 	.mem_map_size = efx_ef10_mem_map_size,
4813 	.probe = efx_ef10_probe_pf,
4814 	.remove = efx_ef10_remove,
4815 	.dimension_resources = efx_ef10_dimension_resources,
4816 	.init = efx_ef10_init_nic,
4817 	.fini = efx_port_dummy_op_void,
4818 	.map_reset_reason = efx_ef10_map_reset_reason,
4819 	.map_reset_flags = efx_ef10_map_reset_flags,
4820 	.reset = efx_ef10_reset,
4821 	.probe_port = efx_mcdi_port_probe,
4822 	.remove_port = efx_mcdi_port_remove,
4823 	.fini_dmaq = efx_ef10_fini_dmaq,
4824 	.prepare_flr = efx_ef10_prepare_flr,
4825 	.finish_flr = efx_port_dummy_op_void,
4826 	.describe_stats = efx_ef10_describe_stats,
4827 	.update_stats = efx_ef10_update_stats_pf,
4828 	.start_stats = efx_mcdi_mac_start_stats,
4829 	.pull_stats = efx_mcdi_mac_pull_stats,
4830 	.stop_stats = efx_mcdi_mac_stop_stats,
4831 	.set_id_led = efx_mcdi_set_id_led,
4832 	.push_irq_moderation = efx_ef10_push_irq_moderation,
4833 	.reconfigure_mac = efx_ef10_mac_reconfigure,
4834 	.check_mac_fault = efx_mcdi_mac_check_fault,
4835 	.reconfigure_port = efx_mcdi_port_reconfigure,
4836 	.get_wol = efx_ef10_get_wol,
4837 	.set_wol = efx_ef10_set_wol,
4838 	.resume_wol = efx_port_dummy_op_void,
4839 	.test_chip = efx_ef10_test_chip,
4840 	.test_nvram = efx_mcdi_nvram_test_all,
4841 	.mcdi_request = efx_ef10_mcdi_request,
4842 	.mcdi_poll_response = efx_ef10_mcdi_poll_response,
4843 	.mcdi_read_response = efx_ef10_mcdi_read_response,
4844 	.mcdi_poll_reboot = efx_ef10_mcdi_poll_reboot,
4845 	.mcdi_reboot_detected = efx_ef10_mcdi_reboot_detected,
4846 	.irq_enable_master = efx_port_dummy_op_void,
4847 	.irq_test_generate = efx_ef10_irq_test_generate,
4848 	.irq_disable_non_ev = efx_port_dummy_op_void,
4849 	.irq_handle_msi = efx_ef10_msi_interrupt,
4850 	.irq_handle_legacy = efx_ef10_legacy_interrupt,
4851 	.tx_probe = efx_ef10_tx_probe,
4852 	.tx_init = efx_ef10_tx_init,
4853 	.tx_remove = efx_ef10_tx_remove,
4854 	.tx_write = efx_ef10_tx_write,
4855 	.rx_push_rss_config = efx_ef10_pf_rx_push_rss_config,
4856 	.rx_probe = efx_ef10_rx_probe,
4857 	.rx_init = efx_ef10_rx_init,
4858 	.rx_remove = efx_ef10_rx_remove,
4859 	.rx_write = efx_ef10_rx_write,
4860 	.rx_defer_refill = efx_ef10_rx_defer_refill,
4861 	.ev_probe = efx_ef10_ev_probe,
4862 	.ev_init = efx_ef10_ev_init,
4863 	.ev_fini = efx_ef10_ev_fini,
4864 	.ev_remove = efx_ef10_ev_remove,
4865 	.ev_process = efx_ef10_ev_process,
4866 	.ev_read_ack = efx_ef10_ev_read_ack,
4867 	.ev_test_generate = efx_ef10_ev_test_generate,
4868 	.filter_table_probe = efx_ef10_filter_table_probe,
4869 	.filter_table_restore = efx_ef10_filter_table_restore,
4870 	.filter_table_remove = efx_ef10_filter_table_remove,
4871 	.filter_update_rx_scatter = efx_ef10_filter_update_rx_scatter,
4872 	.filter_insert = efx_ef10_filter_insert,
4873 	.filter_remove_safe = efx_ef10_filter_remove_safe,
4874 	.filter_get_safe = efx_ef10_filter_get_safe,
4875 	.filter_clear_rx = efx_ef10_filter_clear_rx,
4876 	.filter_count_rx_used = efx_ef10_filter_count_rx_used,
4877 	.filter_get_rx_id_limit = efx_ef10_filter_get_rx_id_limit,
4878 	.filter_get_rx_ids = efx_ef10_filter_get_rx_ids,
4879 #ifdef CONFIG_RFS_ACCEL
4880 	.filter_rfs_insert = efx_ef10_filter_rfs_insert,
4881 	.filter_rfs_expire_one = efx_ef10_filter_rfs_expire_one,
4882 #endif
4883 #ifdef CONFIG_SFC_MTD
4884 	.mtd_probe = efx_ef10_mtd_probe,
4885 	.mtd_rename = efx_mcdi_mtd_rename,
4886 	.mtd_read = efx_mcdi_mtd_read,
4887 	.mtd_erase = efx_mcdi_mtd_erase,
4888 	.mtd_write = efx_mcdi_mtd_write,
4889 	.mtd_sync = efx_mcdi_mtd_sync,
4890 #endif
4891 	.ptp_write_host_time = efx_ef10_ptp_write_host_time,
4892 	.ptp_set_ts_sync_events = efx_ef10_ptp_set_ts_sync_events,
4893 	.ptp_set_ts_config = efx_ef10_ptp_set_ts_config,
4894 #ifdef CONFIG_SFC_SRIOV
4895 	.sriov_configure = efx_ef10_sriov_configure,
4896 	.sriov_init = efx_ef10_sriov_init,
4897 	.sriov_fini = efx_ef10_sriov_fini,
4898 	.sriov_wanted = efx_ef10_sriov_wanted,
4899 	.sriov_reset = efx_ef10_sriov_reset,
4900 	.sriov_flr = efx_ef10_sriov_flr,
4901 	.sriov_set_vf_mac = efx_ef10_sriov_set_vf_mac,
4902 	.sriov_set_vf_vlan = efx_ef10_sriov_set_vf_vlan,
4903 	.sriov_set_vf_spoofchk = efx_ef10_sriov_set_vf_spoofchk,
4904 	.sriov_get_vf_config = efx_ef10_sriov_get_vf_config,
4905 	.sriov_set_vf_link_state = efx_ef10_sriov_set_vf_link_state,
4906 	.vswitching_probe = efx_ef10_vswitching_probe_pf,
4907 	.vswitching_restore = efx_ef10_vswitching_restore_pf,
4908 	.vswitching_remove = efx_ef10_vswitching_remove_pf,
4909 #endif
4910 	.get_mac_address = efx_ef10_get_mac_address_pf,
4911 	.set_mac_address = efx_ef10_set_mac_address,
4912 
4913 	.revision = EFX_REV_HUNT_A0,
4914 	.max_dma_mask = DMA_BIT_MASK(ESF_DZ_TX_KER_BUF_ADDR_WIDTH),
4915 	.rx_prefix_size = ES_DZ_RX_PREFIX_SIZE,
4916 	.rx_hash_offset = ES_DZ_RX_PREFIX_HASH_OFST,
4917 	.rx_ts_offset = ES_DZ_RX_PREFIX_TSTAMP_OFST,
4918 	.can_rx_scatter = true,
4919 	.always_rx_scatter = true,
4920 	.max_interrupt_mode = EFX_INT_MODE_MSIX,
4921 	.timer_period_max = 1 << ERF_DD_EVQ_IND_TIMER_VAL_WIDTH,
4922 	.offload_features = (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
4923 			     NETIF_F_RXHASH | NETIF_F_NTUPLE),
4924 	.mcdi_max_ver = 2,
4925 	.max_rx_ip_filters = HUNT_FILTER_TBL_ROWS,
4926 	.hwtstamp_filters = 1 << HWTSTAMP_FILTER_NONE |
4927 			    1 << HWTSTAMP_FILTER_ALL,
4928 };
4929