1 /****************************************************************************
2 * Driver for Solarflare network controllers and boards
3 * Copyright 2012-2013 Solarflare Communications Inc.
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.
10 #include "net_driver.h"
11 #include "ef10_regs.h"
14 #include "mcdi_pcol.h"
16 #include "workarounds.h"
18 #include "ef10_sriov.h"
20 #include <linux/jhash.h>
21 #include <linux/wait.h>
22 #include <linux/workqueue.h>
24 /* Hardware control for EF10 architecture including 'Huntington'. */
26 #define EFX_EF10_DRVGEN_EV 7
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
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
47 * Huntington has a single 8K filter table shared between all filter
48 * types and both ports.
50 #define HUNT_FILTER_TBL_ROWS 8192
52 #define EFX_EF10_FILTER_ID_INVALID 0xffff
53 struct efx_ef10_dev_addr {
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;
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.
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 */
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];
82 /* Indices (like efx_ef10_dev_addr.id) for promisc/allmulti filters */
88 /* An arbitrary search limit for the software hash table */
89 #define EFX_EF10_FILTER_SEARCH_LIMIT 200
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);
94 static int efx_ef10_get_warm_boot_count(struct efx_nic *efx)
98 efx_readd(efx, ®, 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;
103 static unsigned int efx_ef10_mem_map_size(struct efx_nic *efx)
107 bar = efx->type->mem_bar;
108 return resource_size(&efx->pci_dev->resource[bar]);
111 static bool efx_ef10_is_vf(struct efx_nic *efx)
113 return efx->type->is_vf;
116 static int efx_ef10_get_pf_index(struct efx_nic *efx)
118 MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_FUNCTION_INFO_OUT_LEN);
119 struct efx_ef10_nic_data *nic_data = efx->nic_data;
123 rc = efx_mcdi_rpc(efx, MC_CMD_GET_FUNCTION_INFO, NULL, 0, outbuf,
124 sizeof(outbuf), &outlen);
127 if (outlen < sizeof(outbuf))
130 nic_data->pf_index = MCDI_DWORD(outbuf, GET_FUNCTION_INFO_OUT_PF);
134 #ifdef CONFIG_SFC_SRIOV
135 static int efx_ef10_get_vf_index(struct efx_nic *efx)
137 MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_FUNCTION_INFO_OUT_LEN);
138 struct efx_ef10_nic_data *nic_data = efx->nic_data;
142 rc = efx_mcdi_rpc(efx, MC_CMD_GET_FUNCTION_INFO, NULL, 0, outbuf,
143 sizeof(outbuf), &outlen);
146 if (outlen < sizeof(outbuf))
149 nic_data->vf_index = MCDI_DWORD(outbuf, GET_FUNCTION_INFO_OUT_VF);
154 static int efx_ef10_init_datapath_caps(struct efx_nic *efx)
156 MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_CAPABILITIES_OUT_LEN);
157 struct efx_ef10_nic_data *nic_data = efx->nic_data;
161 BUILD_BUG_ON(MC_CMD_GET_CAPABILITIES_IN_LEN != 0);
163 rc = efx_mcdi_rpc(efx, MC_CMD_GET_CAPABILITIES, NULL, 0,
164 outbuf, sizeof(outbuf), &outlen);
167 if (outlen < sizeof(outbuf)) {
168 netif_err(efx, drv, efx->net_dev,
169 "unable to read datapath firmware capabilities\n");
173 nic_data->datapath_caps =
174 MCDI_DWORD(outbuf, GET_CAPABILITIES_OUT_FLAGS1);
176 /* record the DPCPU firmware IDs to determine VEB vswitching support.
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);
183 if (!(nic_data->datapath_caps &
184 (1 << MC_CMD_GET_CAPABILITIES_OUT_TX_TSO_LBN))) {
185 netif_err(efx, drv, efx->net_dev,
186 "current firmware does not support TSO\n");
190 if (!(nic_data->datapath_caps &
191 (1 << MC_CMD_GET_CAPABILITIES_OUT_RX_PREFIX_LEN_14_LBN))) {
192 netif_err(efx, probe, efx->net_dev,
193 "current firmware does not support an RX prefix\n");
200 static int efx_ef10_get_sysclk_freq(struct efx_nic *efx)
202 MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_CLOCK_OUT_LEN);
205 rc = efx_mcdi_rpc(efx, MC_CMD_GET_CLOCK, NULL, 0,
206 outbuf, sizeof(outbuf), NULL);
209 rc = MCDI_DWORD(outbuf, GET_CLOCK_OUT_SYS_FREQ);
210 return rc > 0 ? rc : -ERANGE;
213 static int efx_ef10_get_mac_address_pf(struct efx_nic *efx, u8 *mac_address)
215 MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_MAC_ADDRESSES_OUT_LEN);
219 BUILD_BUG_ON(MC_CMD_GET_MAC_ADDRESSES_IN_LEN != 0);
221 rc = efx_mcdi_rpc(efx, MC_CMD_GET_MAC_ADDRESSES, NULL, 0,
222 outbuf, sizeof(outbuf), &outlen);
225 if (outlen < MC_CMD_GET_MAC_ADDRESSES_OUT_LEN)
228 ether_addr_copy(mac_address,
229 MCDI_PTR(outbuf, GET_MAC_ADDRESSES_OUT_MAC_ADDR_BASE));
233 static int efx_ef10_get_mac_address_vf(struct efx_nic *efx, u8 *mac_address)
235 MCDI_DECLARE_BUF(inbuf, MC_CMD_VPORT_GET_MAC_ADDRESSES_IN_LEN);
236 MCDI_DECLARE_BUF(outbuf, MC_CMD_VPORT_GET_MAC_ADDRESSES_OUT_LENMAX);
240 MCDI_SET_DWORD(inbuf, VPORT_GET_MAC_ADDRESSES_IN_VPORT_ID,
241 EVB_PORT_ID_ASSIGNED);
242 rc = efx_mcdi_rpc(efx, MC_CMD_VPORT_GET_MAC_ADDRESSES, inbuf,
243 sizeof(inbuf), outbuf, sizeof(outbuf), &outlen);
247 if (outlen < MC_CMD_VPORT_GET_MAC_ADDRESSES_OUT_LENMIN)
250 num_addrs = MCDI_DWORD(outbuf,
251 VPORT_GET_MAC_ADDRESSES_OUT_MACADDR_COUNT);
253 WARN_ON(num_addrs != 1);
255 ether_addr_copy(mac_address,
256 MCDI_PTR(outbuf, VPORT_GET_MAC_ADDRESSES_OUT_MACADDR));
261 static ssize_t efx_ef10_show_link_control_flag(struct device *dev,
262 struct device_attribute *attr,
265 struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));
267 return sprintf(buf, "%d\n",
268 ((efx->mcdi->fn_flags) &
269 (1 << MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_LINKCTRL))
273 static ssize_t efx_ef10_show_primary_flag(struct device *dev,
274 struct device_attribute *attr,
277 struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));
279 return sprintf(buf, "%d\n",
280 ((efx->mcdi->fn_flags) &
281 (1 << MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_PRIMARY))
285 static DEVICE_ATTR(link_control_flag, 0444, efx_ef10_show_link_control_flag,
287 static DEVICE_ATTR(primary_flag, 0444, efx_ef10_show_primary_flag, NULL);
289 static int efx_ef10_probe(struct efx_nic *efx)
291 struct efx_ef10_nic_data *nic_data;
292 struct net_device *net_dev = efx->net_dev;
295 /* We can have one VI for each 8K region. However, until we
296 * use TX option descriptors we need two TX queues per channel.
298 efx->max_channels = min_t(unsigned int,
300 efx_ef10_mem_map_size(efx) /
301 (EFX_VI_PAGE_SIZE * EFX_TXQ_TYPES));
302 efx->max_tx_channels = efx->max_channels;
303 if (WARN_ON(efx->max_channels == 0))
306 nic_data = kzalloc(sizeof(*nic_data), GFP_KERNEL);
309 efx->nic_data = nic_data;
311 /* we assume later that we can copy from this buffer in dwords */
312 BUILD_BUG_ON(MCDI_CTL_SDU_LEN_MAX_V2 % 4);
314 rc = efx_nic_alloc_buffer(efx, &nic_data->mcdi_buf,
315 8 + MCDI_CTL_SDU_LEN_MAX_V2, GFP_KERNEL);
319 /* Get the MC's warm boot count. In case it's rebooting right
320 * now, be prepared to retry.
324 rc = efx_ef10_get_warm_boot_count(efx);
331 nic_data->warm_boot_count = rc;
333 nic_data->rx_rss_context = EFX_EF10_RSS_CONTEXT_INVALID;
335 nic_data->vport_id = EVB_PORT_ID_ASSIGNED;
337 /* In case we're recovering from a crash (kexec), we want to
338 * cancel any outstanding request by the previous user of this
339 * function. We send a special message using the least
340 * significant bits of the 'high' (doorbell) register.
342 _efx_writed(efx, cpu_to_le32(1), ER_DZ_MC_DB_HWRD);
344 rc = efx_mcdi_init(efx);
348 /* Reset (most) configuration for this function */
349 rc = efx_mcdi_reset(efx, RESET_TYPE_ALL);
353 /* Enable event logging */
354 rc = efx_mcdi_log_ctrl(efx, true, false, 0);
358 rc = device_create_file(&efx->pci_dev->dev,
359 &dev_attr_link_control_flag);
363 rc = device_create_file(&efx->pci_dev->dev, &dev_attr_primary_flag);
367 rc = efx_ef10_get_pf_index(efx);
371 rc = efx_ef10_init_datapath_caps(efx);
375 efx->rx_packet_len_offset =
376 ES_DZ_RX_PREFIX_PKTLEN_OFST - ES_DZ_RX_PREFIX_SIZE;
378 rc = efx_mcdi_port_get_number(efx);
382 net_dev->dev_port = rc;
384 rc = efx->type->get_mac_address(efx, efx->net_dev->perm_addr);
388 rc = efx_ef10_get_sysclk_freq(efx);
391 efx->timer_quantum_ns = 1536000 / rc; /* 1536 cycles */
393 /* Check whether firmware supports bug 35388 workaround.
394 * First try to enable it, then if we get EPERM, just
395 * ask if it's already enabled
397 rc = efx_mcdi_set_workaround(efx, MC_CMD_WORKAROUND_BUG35388, true, NULL);
399 nic_data->workaround_35388 = true;
400 } else if (rc == -EPERM) {
401 unsigned int enabled;
403 rc = efx_mcdi_get_workarounds(efx, NULL, &enabled);
406 nic_data->workaround_35388 = enabled &
407 MC_CMD_GET_WORKAROUNDS_OUT_BUG35388;
408 } else if (rc != -ENOSYS && rc != -ENOENT) {
411 netif_dbg(efx, probe, efx->net_dev,
412 "workaround for bug 35388 is %sabled\n",
413 nic_data->workaround_35388 ? "en" : "dis");
415 rc = efx_mcdi_mon_probe(efx);
416 if (rc && rc != -EPERM)
419 efx_ptp_probe(efx, NULL);
421 #ifdef CONFIG_SFC_SRIOV
422 if ((efx->pci_dev->physfn) && (!efx->pci_dev->is_physfn)) {
423 struct pci_dev *pci_dev_pf = efx->pci_dev->physfn;
424 struct efx_nic *efx_pf = pci_get_drvdata(pci_dev_pf);
426 efx_pf->type->get_mac_address(efx_pf, nic_data->port_id);
429 ether_addr_copy(nic_data->port_id, efx->net_dev->perm_addr);
434 device_remove_file(&efx->pci_dev->dev, &dev_attr_primary_flag);
436 device_remove_file(&efx->pci_dev->dev, &dev_attr_link_control_flag);
440 efx_nic_free_buffer(efx, &nic_data->mcdi_buf);
443 efx->nic_data = NULL;
447 static int efx_ef10_free_vis(struct efx_nic *efx)
449 MCDI_DECLARE_BUF_ERR(outbuf);
451 int rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FREE_VIS, NULL, 0,
452 outbuf, sizeof(outbuf), &outlen);
454 /* -EALREADY means nothing to free, so ignore */
458 efx_mcdi_display_error(efx, MC_CMD_FREE_VIS, 0, outbuf, outlen,
465 static void efx_ef10_free_piobufs(struct efx_nic *efx)
467 struct efx_ef10_nic_data *nic_data = efx->nic_data;
468 MCDI_DECLARE_BUF(inbuf, MC_CMD_FREE_PIOBUF_IN_LEN);
472 BUILD_BUG_ON(MC_CMD_FREE_PIOBUF_OUT_LEN != 0);
474 for (i = 0; i < nic_data->n_piobufs; i++) {
475 MCDI_SET_DWORD(inbuf, FREE_PIOBUF_IN_PIOBUF_HANDLE,
476 nic_data->piobuf_handle[i]);
477 rc = efx_mcdi_rpc(efx, MC_CMD_FREE_PIOBUF, inbuf, sizeof(inbuf),
482 nic_data->n_piobufs = 0;
485 static int efx_ef10_alloc_piobufs(struct efx_nic *efx, unsigned int n)
487 struct efx_ef10_nic_data *nic_data = efx->nic_data;
488 MCDI_DECLARE_BUF(outbuf, MC_CMD_ALLOC_PIOBUF_OUT_LEN);
493 BUILD_BUG_ON(MC_CMD_ALLOC_PIOBUF_IN_LEN != 0);
495 for (i = 0; i < n; i++) {
496 rc = efx_mcdi_rpc(efx, MC_CMD_ALLOC_PIOBUF, NULL, 0,
497 outbuf, sizeof(outbuf), &outlen);
500 if (outlen < MC_CMD_ALLOC_PIOBUF_OUT_LEN) {
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]);
511 nic_data->n_piobufs = i;
513 efx_ef10_free_piobufs(efx);
517 static int efx_ef10_link_piobufs(struct efx_nic *efx)
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;
528 BUILD_BUG_ON(MC_CMD_LINK_PIOBUF_OUT_LEN != 0);
529 BUILD_BUG_ON(MC_CMD_UNLINK_PIOBUF_OUT_LEN != 0);
531 memset(inbuf, 0, sizeof(inbuf));
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,
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,
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);
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
561 offset = ((efx->tx_channel_offset + efx->n_tx_channels -
562 tx_queue->channel->channel - 1) *
564 index = offset / ER_DZ_TX_PIOBUF_SIZE;
565 offset = offset % ER_DZ_TX_PIOBUF_SIZE;
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.
572 if (tx_queue->queue == nic_data->pio_write_vi_base) {
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,
582 rc = efx_mcdi_rpc(efx, MC_CMD_LINK_PIOBUF,
583 inbuf, MC_CMD_LINK_PIOBUF_IN_LEN,
588 /* This is non-fatal; the TX path just
589 * won't use PIO for this queue
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;
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,
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,
622 static void efx_ef10_forget_old_piobufs(struct efx_nic *efx)
624 struct efx_channel *channel;
625 struct efx_tx_queue *tx_queue;
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;
633 #else /* !EFX_USE_PIO */
635 static int efx_ef10_alloc_piobufs(struct efx_nic *efx, unsigned int n)
637 return n == 0 ? 0 : -ENOBUFS;
640 static int efx_ef10_link_piobufs(struct efx_nic *efx)
645 static void efx_ef10_free_piobufs(struct efx_nic *efx)
649 static void efx_ef10_forget_old_piobufs(struct efx_nic *efx)
653 #endif /* EFX_USE_PIO */
655 static void efx_ef10_remove(struct efx_nic *efx)
657 struct efx_ef10_nic_data *nic_data = efx->nic_data;
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;
666 if (efx->pci_dev->is_virtfn) {
667 pci_dev_pf = efx->pci_dev->physfn;
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;
674 netif_info(efx, drv, efx->net_dev,
675 "Could not get the PF id from VF\n");
681 efx_mcdi_mon_remove(efx);
683 efx_ef10_rx_free_indir_table(efx);
685 if (nic_data->wc_membase)
686 iounmap(nic_data->wc_membase);
688 rc = efx_ef10_free_vis(efx);
691 if (!nic_data->must_restore_piobufs)
692 efx_ef10_free_piobufs(efx);
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);
698 efx_nic_free_buffer(efx, &nic_data->mcdi_buf);
702 static int efx_ef10_probe_pf(struct efx_nic *efx)
704 return efx_ef10_probe(efx);
707 int efx_ef10_vadaptor_alloc(struct efx_nic *efx, unsigned int port_id)
709 MCDI_DECLARE_BUF(inbuf, MC_CMD_VADAPTOR_ALLOC_IN_LEN);
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),
716 int efx_ef10_vadaptor_free(struct efx_nic *efx, unsigned int port_id)
718 MCDI_DECLARE_BUF(inbuf, MC_CMD_VADAPTOR_FREE_IN_LEN);
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),
725 int efx_ef10_vport_add_mac(struct efx_nic *efx,
726 unsigned int port_id, u8 *mac)
728 MCDI_DECLARE_BUF(inbuf, MC_CMD_VPORT_ADD_MAC_ADDRESS_IN_LEN);
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);
733 return efx_mcdi_rpc(efx, MC_CMD_VPORT_ADD_MAC_ADDRESS, inbuf,
734 sizeof(inbuf), NULL, 0, NULL);
737 int efx_ef10_vport_del_mac(struct efx_nic *efx,
738 unsigned int port_id, u8 *mac)
740 MCDI_DECLARE_BUF(inbuf, MC_CMD_VPORT_DEL_MAC_ADDRESS_IN_LEN);
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);
745 return efx_mcdi_rpc(efx, MC_CMD_VPORT_DEL_MAC_ADDRESS, inbuf,
746 sizeof(inbuf), NULL, 0, NULL);
749 #ifdef CONFIG_SFC_SRIOV
750 static int efx_ef10_probe_vf(struct efx_nic *efx)
753 struct pci_dev *pci_dev_pf;
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.
759 pci_dev_pf = efx->pci_dev->physfn;
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;
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");
772 rc = efx_ef10_probe(efx);
776 rc = efx_ef10_get_vf_index(efx);
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;
787 nic_data_p->vf[nic_data->vf_index].efx = efx;
788 nic_data_p->vf[nic_data->vf_index].pci_dev =
791 netif_info(efx, drv, efx->net_dev,
792 "Could not get the PF id from VF\n");
798 efx_ef10_remove(efx);
802 static int efx_ef10_probe_vf(struct efx_nic *efx __attribute__ ((unused)))
808 static int efx_ef10_alloc_vis(struct efx_nic *efx,
809 unsigned int min_vis, unsigned int max_vis)
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;
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);
824 if (outlen < MC_CMD_ALLOC_VIS_OUT_LEN)
827 netif_dbg(efx, drv, efx->net_dev, "base VI is A0x%03x\n",
828 MCDI_DWORD(outbuf, ALLOC_VIS_OUT_VI_BASE));
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);
835 /* Note that the failure path of this function does not free
836 * resources, as this will be done by efx_ef10_remove().
838 static int efx_ef10_dimension_resources(struct efx_nic *efx)
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;
848 channel_vis = max(efx->n_channels, efx->n_tx_channels * EFX_TXQ_TYPES);
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.
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);
864 rc = efx_ef10_alloc_piobufs(efx, n_piobufs);
866 netif_err(efx, probe, efx->net_dev,
867 "failed to allocate PIO buffers (%d)\n", rc);
869 netif_dbg(efx, probe, efx->net_dev,
870 "allocated %u PIO buffers\n", n_piobufs);
873 nic_data->n_piobufs = 0;
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.
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.
886 uc_mem_map_size = PAGE_ALIGN((channel_vis - 1) * EFX_VI_PAGE_SIZE +
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.
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) *
897 max_vis = pio_write_vi_base + nic_data->n_piobufs;
899 pio_write_vi_base = 0;
901 max_vis = channel_vis;
904 /* In case the last attached driver failed to free VIs, do it now */
905 rc = efx_ef10_free_vis(efx);
909 rc = efx_ef10_alloc_vis(efx, min_vis, max_vis);
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
922 efx->max_channels = nic_data->n_allocated_vis;
923 efx->max_tx_channels =
924 nic_data->n_allocated_vis / EFX_TXQ_TYPES;
926 efx_ef10_free_vis(efx);
930 /* If we didn't get enough VIs to map all the PIO buffers, free the
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);
942 /* Shrink the original UC mapping of the memory BAR */
943 membase = ioremap_nocache(efx->membase_phys, uc_mem_map_size);
945 netif_err(efx, probe, efx->net_dev,
946 "could not shrink memory BAR to %x\n",
950 iounmap(efx->membase);
951 efx->membase = membase;
953 /* Set up the WC mapping if needed */
954 if (wc_mem_map_size) {
955 nic_data->wc_membase = ioremap_wc(efx->membase_phys +
958 if (!nic_data->wc_membase) {
959 netif_err(efx, probe, efx->net_dev,
960 "could not allocate WC mapping of size %x\n",
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 -
970 rc = efx_ef10_link_piobufs(efx);
972 efx_ef10_free_piobufs(efx);
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);
983 static int efx_ef10_init_nic(struct efx_nic *efx)
985 struct efx_ef10_nic_data *nic_data = efx->nic_data;
988 if (nic_data->must_check_datapath_caps) {
989 rc = efx_ef10_init_datapath_caps(efx);
992 nic_data->must_check_datapath_caps = false;
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);
1001 nic_data->must_realloc_vis = false;
1004 if (nic_data->must_restore_piobufs && nic_data->n_piobufs) {
1005 rc = efx_ef10_alloc_piobufs(efx, nic_data->n_piobufs);
1007 rc = efx_ef10_link_piobufs(efx);
1009 efx_ef10_free_piobufs(efx);
1012 /* Log an error on failure, but this is non-fatal */
1014 netif_err(efx, drv, efx->net_dev,
1015 "failed to restore PIO buffers (%d)\n", rc);
1016 nic_data->must_restore_piobufs = false;
1019 /* don't fail init if RSS setup doesn't work */
1020 efx->type->rx_push_rss_config(efx, false, efx->rx_indir_table);
1025 static void efx_ef10_reset_mc_allocations(struct efx_nic *efx)
1027 struct efx_ef10_nic_data *nic_data = efx->nic_data;
1028 #ifdef CONFIG_SFC_SRIOV
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;
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
1044 for (i = 0; i < efx->vf_count; i++)
1045 nic_data->vf[i].vport_id = 0;
1049 static enum reset_type efx_ef10_map_reset_reason(enum reset_type reason)
1051 if (reason == RESET_TYPE_MC_FAILURE)
1052 return RESET_TYPE_DATAPATH;
1054 return efx_mcdi_map_reset_reason(reason);
1057 static int efx_ef10_map_reset_flags(u32 *flags)
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)
1068 /* We assume for now that our PCI function is permitted to
1072 if ((*flags & EF10_RESET_MC) == EF10_RESET_MC) {
1073 *flags &= ~EF10_RESET_MC;
1074 return RESET_TYPE_WORLD;
1077 if ((*flags & EF10_RESET_PORT) == EF10_RESET_PORT) {
1078 *flags &= ~EF10_RESET_PORT;
1079 return RESET_TYPE_ALL;
1082 /* no invisible reset implemented */
1087 static int efx_ef10_reset(struct efx_nic *efx, enum reset_type reset_type)
1089 int rc = efx_mcdi_reset(efx, reset_type);
1091 /* Unprivileged functions return -EPERM, but need to return success
1092 * here so that the datapath is brought back up.
1094 if (reset_type == RESET_TYPE_WORLD && rc == -EPERM)
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.
1103 if ((reset_type == RESET_TYPE_ALL ||
1104 reset_type == RESET_TYPE_MCDI_TIMEOUT) && !rc)
1105 efx_ef10_reset_mc_allocations(efx);
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 }
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),
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) | \
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))
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.
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))
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
1248 #define HUNT_40G_EXTRA_STAT_MASK ((1ULL << EF10_STAT_port_rx_align_error) |\
1249 (1ULL << EF10_STAT_port_rx_length_error))
1251 /* These statistics are only provided if the firmware supports the
1252 * capability PM_AND_RXDP_COUNTERS.
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))
1268 static u64 efx_ef10_raw_stat_mask(struct efx_nic *efx)
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;
1274 if (!(efx->mcdi->fn_flags &
1275 1 << MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_LINKCTRL))
1278 if (port_caps & (1 << MC_CMD_PHY_CAP_40000FDX_LBN))
1279 raw_mask |= HUNT_40G_EXTRA_STAT_MASK;
1281 raw_mask |= HUNT_10G_ONLY_STAT_MASK;
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;
1290 static void efx_ef10_get_stat_mask(struct efx_nic *efx, unsigned long *mask)
1292 struct efx_ef10_nic_data *nic_data = efx->nic_data;
1295 raw_mask[0] = efx_ef10_raw_stat_mask(efx);
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;
1306 #if BITS_PER_LONG == 64
1307 mask[0] = raw_mask[0];
1308 mask[1] = raw_mask[1];
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;
1317 static size_t efx_ef10_describe_stats(struct efx_nic *efx, u8 *names)
1319 DECLARE_BITMAP(mask, EF10_STAT_COUNT);
1321 efx_ef10_get_stat_mask(efx, mask);
1322 return efx_nic_describe_stats(efx_ef10_stat_desc, EF10_STAT_COUNT,
1326 static size_t efx_ef10_update_stats_common(struct efx_nic *efx, u64 *full_stats,
1327 struct rtnl_link_stats64 *core_stats)
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;
1334 efx_ef10_get_stat_mask(efx, mask);
1337 for_each_set_bit(index, mask, EF10_STAT_COUNT) {
1338 if (efx_ef10_stat_desc[index].name) {
1339 *full_stats++ = stats[index];
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];
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);
1395 static int efx_ef10_try_update_nic_stats_pf(struct efx_nic *efx)
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;
1403 efx_ef10_get_stat_mask(efx, mask);
1405 dma_stats = efx->stats_buffer.addr;
1406 nic_data = efx->nic_data;
1408 generation_end = dma_stats[MC_CMD_MAC_GENERATION_END];
1409 if (generation_end == EFX_MC_STATS_GENERATION_INVALID)
1412 efx_nic_update_stats(efx_ef10_stat_desc, EF10_STAT_COUNT, mask,
1413 stats, efx->stats_buffer.addr, false);
1415 generation_start = dma_stats[MC_CMD_MAC_GENERATION_START];
1416 if (generation_end != generation_start)
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);
1432 static size_t efx_ef10_update_stats_pf(struct efx_nic *efx, u64 *full_stats,
1433 struct rtnl_link_stats64 *core_stats)
1437 /* If we're unlucky enough to read statistics during the DMA, wait
1438 * up to 10ms for it to finish (typically takes <500us)
1440 for (retry = 0; retry < 100; ++retry) {
1441 if (efx_ef10_try_update_nic_stats_pf(efx) == 0)
1446 return efx_ef10_update_stats_common(efx, full_stats, core_stats);
1449 static int efx_ef10_try_update_nic_stats_vf(struct efx_nic *efx)
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;
1461 spin_unlock_bh(&efx->stats_lock);
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.
1467 spin_lock_bh(&efx->stats_lock);
1468 efx_update_sw_stats(efx, stats);
1472 efx_ef10_get_stat_mask(efx, mask);
1474 rc = efx_nic_alloc_buffer(efx, &stats_buf, dma_len, GFP_ATOMIC);
1476 spin_lock_bh(&efx->stats_lock);
1480 dma_stats = stats_buf.addr;
1481 dma_stats[MC_CMD_MAC_GENERATION_END] = EFX_MC_STATS_GENERATION_INVALID;
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);
1489 rc = efx_mcdi_rpc_quiet(efx, MC_CMD_MAC_STATS, inbuf, sizeof(inbuf),
1491 spin_lock_bh(&efx->stats_lock);
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);
1500 generation_end = dma_stats[MC_CMD_MAC_GENERATION_END];
1501 if (generation_end == EFX_MC_STATS_GENERATION_INVALID) {
1506 efx_nic_update_stats(efx_ef10_stat_desc, EF10_STAT_COUNT, mask,
1507 stats, stats_buf.addr, false);
1509 generation_start = dma_stats[MC_CMD_MAC_GENERATION_START];
1510 if (generation_end != generation_start) {
1515 efx_update_sw_stats(efx, stats);
1517 efx_nic_free_buffer(efx, &stats_buf);
1521 static size_t efx_ef10_update_stats_vf(struct efx_nic *efx, u64 *full_stats,
1522 struct rtnl_link_stats64 *core_stats)
1524 if (efx_ef10_try_update_nic_stats_vf(efx))
1527 return efx_ef10_update_stats_common(efx, full_stats, core_stats);
1530 static void efx_ef10_push_irq_moderation(struct efx_channel *channel)
1532 struct efx_nic *efx = channel->efx;
1533 unsigned int mode, value;
1534 efx_dword_t timer_cmd;
1536 if (channel->irq_moderation) {
1538 value = channel->irq_moderation - 1;
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,
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,
1559 static void efx_ef10_get_wol_vf(struct efx_nic *efx,
1560 struct ethtool_wolinfo *wol) {}
1562 static int efx_ef10_set_wol_vf(struct efx_nic *efx, u32 type)
1567 static void efx_ef10_get_wol(struct efx_nic *efx, struct ethtool_wolinfo *wol)
1571 memset(&wol->sopass, 0, sizeof(wol->sopass));
1574 static int efx_ef10_set_wol(struct efx_nic *efx, u32 type)
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)
1585 struct efx_ef10_nic_data *nic_data = efx->nic_data;
1586 u8 *pdu = nic_data->mcdi_buf.addr;
1588 memcpy(pdu, hdr, hdr_len);
1589 memcpy(pdu + hdr_len, sdu, sdu_len);
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.
1598 _efx_writed(efx, cpu_to_le32((u64)nic_data->mcdi_buf.dma_addr >> 32),
1600 _efx_writed(efx, cpu_to_le32((u32)nic_data->mcdi_buf.dma_addr),
1604 static bool efx_ef10_mcdi_poll_response(struct efx_nic *efx)
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;
1610 return EFX_DWORD_FIELD(hdr, MCDI_HEADER_RESPONSE);
1614 efx_ef10_mcdi_read_response(struct efx_nic *efx, efx_dword_t *outbuf,
1615 size_t offset, size_t outlen)
1617 struct efx_ef10_nic_data *nic_data = efx->nic_data;
1618 const u8 *pdu = nic_data->mcdi_buf.addr;
1620 memcpy(outbuf, pdu + offset, outlen);
1623 static void efx_ef10_mcdi_reboot_detected(struct efx_nic *efx)
1625 struct efx_ef10_nic_data *nic_data = efx->nic_data;
1627 /* All our allocations have been reset */
1628 efx_ef10_reset_mc_allocations(efx);
1630 /* The datapath firmware might have been changed */
1631 nic_data->must_check_datapath_caps = true;
1633 /* MAC statistics have been cleared on the NIC; clear the local
1634 * statistic that we update with efx_update_diff_stat().
1636 nic_data->stats[EF10_STAT_port_rx_bad_bytes] = 0;
1639 static int efx_ef10_mcdi_poll_reboot(struct efx_nic *efx)
1641 struct efx_ef10_nic_data *nic_data = efx->nic_data;
1644 rc = efx_ef10_get_warm_boot_count(efx);
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.
1654 if (rc == nic_data->warm_boot_count)
1657 nic_data->warm_boot_count = rc;
1658 efx_ef10_mcdi_reboot_detected(efx);
1663 /* Handle an MSI interrupt
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.
1670 static irqreturn_t efx_ef10_msi_interrupt(int irq, void *dev_id)
1672 struct efx_msi_context *context = dev_id;
1673 struct efx_nic *efx = context->efx;
1675 netif_vdbg(efx, intr, efx->net_dev,
1676 "IRQ %d on CPU %d\n", irq, raw_smp_processor_id());
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();
1683 /* Schedule processing of the channel */
1684 efx_schedule_channel_irq(efx->channel[context->index]);
1690 static irqreturn_t efx_ef10_legacy_interrupt(int irq, void *dev_id)
1692 struct efx_nic *efx = dev_id;
1693 bool soft_enabled = ACCESS_ONCE(efx->irq_soft_enabled);
1694 struct efx_channel *channel;
1698 /* Read the ISR which also ACKs the interrupts */
1699 efx_readd(efx, ®, ER_DZ_BIU_INT_ISR);
1700 queues = EFX_DWORD_FIELD(reg, ERF_DZ_ISR_REG);
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();
1710 efx_for_each_channel(channel, efx) {
1712 efx_schedule_channel_irq(channel);
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));
1724 static void efx_ef10_irq_test_generate(struct efx_nic *efx)
1726 MCDI_DECLARE_BUF(inbuf, MC_CMD_TRIGGER_INTERRUPT_IN_LEN);
1728 BUILD_BUG_ON(MC_CMD_TRIGGER_INTERRUPT_OUT_LEN != 0);
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);
1735 static int efx_ef10_tx_probe(struct efx_tx_queue *tx_queue)
1737 return efx_nic_alloc_buffer(tx_queue->efx, &tx_queue->txd.buf,
1738 (tx_queue->ptr_mask + 1) *
1739 sizeof(efx_qword_t),
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)
1747 unsigned int write_ptr;
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, ®,
1754 ER_DZ_TX_DESC_UPD, tx_queue->queue);
1757 static void efx_ef10_tx_init(struct efx_tx_queue *tx_queue)
1759 MCDI_DECLARE_BUF(inbuf, MC_CMD_INIT_TXQ_IN_LEN(EFX_MAX_DMAQ_SIZE * 8 /
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;
1767 dma_addr_t dma_addr;
1771 BUILD_BUG_ON(MC_CMD_INIT_TXQ_OUT_LEN != 0);
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);
1783 dma_addr = tx_queue->txd.buf.dma_addr;
1785 netif_dbg(efx, hw, efx->net_dev, "pushing TXQ %d. %zu entries (%llx)\n",
1786 tx_queue->queue, entries, (u64)dma_addr);
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;
1793 inlen = MC_CMD_INIT_TXQ_IN_LEN(entries);
1795 rc = efx_mcdi_rpc(efx, MC_CMD_INIT_TXQ, inbuf, inlen,
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.
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;
1817 efx_ef10_push_tx_desc(tx_queue, txd);
1822 netdev_WARN(efx->net_dev, "failed to initialise TXQ %d\n",
1826 static void efx_ef10_tx_fini(struct efx_tx_queue *tx_queue)
1828 MCDI_DECLARE_BUF(inbuf, MC_CMD_FINI_TXQ_IN_LEN);
1829 MCDI_DECLARE_BUF_ERR(outbuf);
1830 struct efx_nic *efx = tx_queue->efx;
1834 MCDI_SET_DWORD(inbuf, FINI_TXQ_IN_INSTANCE,
1837 rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FINI_TXQ, inbuf, sizeof(inbuf),
1838 outbuf, sizeof(outbuf), &outlen);
1840 if (rc && rc != -EALREADY)
1846 efx_mcdi_display_error(efx, MC_CMD_FINI_TXQ, MC_CMD_FINI_TXQ_IN_LEN,
1847 outbuf, outlen, rc);
1850 static void efx_ef10_tx_remove(struct efx_tx_queue *tx_queue)
1852 efx_nic_free_buffer(tx_queue->efx, &tx_queue->txd.buf);
1855 /* This writes to the TX_DESC_WPTR; write pointer for TX descriptor ring */
1856 static inline void efx_ef10_notify_tx_desc(struct efx_tx_queue *tx_queue)
1858 unsigned int write_ptr;
1861 write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
1862 EFX_POPULATE_DWORD_1(reg, ERF_DZ_TX_DESC_WPTR_DWORD, write_ptr);
1863 efx_writed_page(tx_queue->efx, ®,
1864 ER_DZ_TX_DESC_UPD_DWORD, tx_queue->queue);
1867 static void efx_ef10_tx_write(struct efx_tx_queue *tx_queue)
1869 unsigned int old_write_count = tx_queue->write_count;
1870 struct efx_tx_buffer *buffer;
1871 unsigned int write_ptr;
1874 tx_queue->xmit_more_available = false;
1875 if (unlikely(tx_queue->write_count == tx_queue->insert_count))
1879 write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
1880 buffer = &tx_queue->buffer[write_ptr];
1881 txd = efx_tx_desc(tx_queue, write_ptr);
1882 ++tx_queue->write_count;
1884 /* Create TX descriptor ring entry */
1885 if (buffer->flags & EFX_TX_BUF_OPTION) {
1886 *txd = buffer->option;
1888 BUILD_BUG_ON(EFX_TX_BUF_CONT != 1);
1889 EFX_POPULATE_QWORD_3(
1892 buffer->flags & EFX_TX_BUF_CONT,
1893 ESF_DZ_TX_KER_BYTE_CNT, buffer->len,
1894 ESF_DZ_TX_KER_BUF_ADDR, buffer->dma_addr);
1896 } while (tx_queue->write_count != tx_queue->insert_count);
1898 wmb(); /* Ensure descriptors are written before they are fetched */
1900 if (efx_nic_may_push_tx_desc(tx_queue, old_write_count)) {
1901 txd = efx_tx_desc(tx_queue,
1902 old_write_count & tx_queue->ptr_mask);
1903 efx_ef10_push_tx_desc(tx_queue, txd);
1906 efx_ef10_notify_tx_desc(tx_queue);
1910 static int efx_ef10_alloc_rss_context(struct efx_nic *efx, u32 *context,
1911 bool exclusive, unsigned *context_size)
1913 MCDI_DECLARE_BUF(inbuf, MC_CMD_RSS_CONTEXT_ALLOC_IN_LEN);
1914 MCDI_DECLARE_BUF(outbuf, MC_CMD_RSS_CONTEXT_ALLOC_OUT_LEN);
1915 struct efx_ef10_nic_data *nic_data = efx->nic_data;
1918 u32 alloc_type = exclusive ?
1919 MC_CMD_RSS_CONTEXT_ALLOC_IN_TYPE_EXCLUSIVE :
1920 MC_CMD_RSS_CONTEXT_ALLOC_IN_TYPE_SHARED;
1921 unsigned rss_spread = exclusive ?
1923 min(rounddown_pow_of_two(efx->rss_spread),
1924 EFX_EF10_MAX_SHARED_RSS_CONTEXT_SIZE);
1926 if (!exclusive && rss_spread == 1) {
1927 *context = EFX_EF10_RSS_CONTEXT_INVALID;
1933 MCDI_SET_DWORD(inbuf, RSS_CONTEXT_ALLOC_IN_UPSTREAM_PORT_ID,
1934 nic_data->vport_id);
1935 MCDI_SET_DWORD(inbuf, RSS_CONTEXT_ALLOC_IN_TYPE, alloc_type);
1936 MCDI_SET_DWORD(inbuf, RSS_CONTEXT_ALLOC_IN_NUM_QUEUES, rss_spread);
1938 rc = efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_ALLOC, inbuf, sizeof(inbuf),
1939 outbuf, sizeof(outbuf), &outlen);
1943 if (outlen < MC_CMD_RSS_CONTEXT_ALLOC_OUT_LEN)
1946 *context = MCDI_DWORD(outbuf, RSS_CONTEXT_ALLOC_OUT_RSS_CONTEXT_ID);
1949 *context_size = rss_spread;
1954 static void efx_ef10_free_rss_context(struct efx_nic *efx, u32 context)
1956 MCDI_DECLARE_BUF(inbuf, MC_CMD_RSS_CONTEXT_FREE_IN_LEN);
1959 MCDI_SET_DWORD(inbuf, RSS_CONTEXT_FREE_IN_RSS_CONTEXT_ID,
1962 rc = efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_FREE, inbuf, sizeof(inbuf),
1967 static int efx_ef10_populate_rss_table(struct efx_nic *efx, u32 context,
1968 const u32 *rx_indir_table)
1970 MCDI_DECLARE_BUF(tablebuf, MC_CMD_RSS_CONTEXT_SET_TABLE_IN_LEN);
1971 MCDI_DECLARE_BUF(keybuf, MC_CMD_RSS_CONTEXT_SET_KEY_IN_LEN);
1974 MCDI_SET_DWORD(tablebuf, RSS_CONTEXT_SET_TABLE_IN_RSS_CONTEXT_ID,
1976 BUILD_BUG_ON(ARRAY_SIZE(efx->rx_indir_table) !=
1977 MC_CMD_RSS_CONTEXT_SET_TABLE_IN_INDIRECTION_TABLE_LEN);
1979 for (i = 0; i < ARRAY_SIZE(efx->rx_indir_table); ++i)
1981 RSS_CONTEXT_SET_TABLE_IN_INDIRECTION_TABLE)[i] =
1982 (u8) rx_indir_table[i];
1984 rc = efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_SET_TABLE, tablebuf,
1985 sizeof(tablebuf), NULL, 0, NULL);
1989 MCDI_SET_DWORD(keybuf, RSS_CONTEXT_SET_KEY_IN_RSS_CONTEXT_ID,
1991 BUILD_BUG_ON(ARRAY_SIZE(efx->rx_hash_key) !=
1992 MC_CMD_RSS_CONTEXT_SET_KEY_IN_TOEPLITZ_KEY_LEN);
1993 for (i = 0; i < ARRAY_SIZE(efx->rx_hash_key); ++i)
1994 MCDI_PTR(keybuf, RSS_CONTEXT_SET_KEY_IN_TOEPLITZ_KEY)[i] =
1995 efx->rx_hash_key[i];
1997 return efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_SET_KEY, keybuf,
1998 sizeof(keybuf), NULL, 0, NULL);
2001 static void efx_ef10_rx_free_indir_table(struct efx_nic *efx)
2003 struct efx_ef10_nic_data *nic_data = efx->nic_data;
2005 if (nic_data->rx_rss_context != EFX_EF10_RSS_CONTEXT_INVALID)
2006 efx_ef10_free_rss_context(efx, nic_data->rx_rss_context);
2007 nic_data->rx_rss_context = EFX_EF10_RSS_CONTEXT_INVALID;
2010 static int efx_ef10_rx_push_shared_rss_config(struct efx_nic *efx,
2011 unsigned *context_size)
2013 u32 new_rx_rss_context;
2014 struct efx_ef10_nic_data *nic_data = efx->nic_data;
2015 int rc = efx_ef10_alloc_rss_context(efx, &new_rx_rss_context,
2016 false, context_size);
2021 nic_data->rx_rss_context = new_rx_rss_context;
2022 nic_data->rx_rss_context_exclusive = false;
2023 efx_set_default_rx_indir_table(efx);
2027 static int efx_ef10_rx_push_exclusive_rss_config(struct efx_nic *efx,
2028 const u32 *rx_indir_table)
2030 struct efx_ef10_nic_data *nic_data = efx->nic_data;
2032 u32 new_rx_rss_context;
2034 if (nic_data->rx_rss_context == EFX_EF10_RSS_CONTEXT_INVALID ||
2035 !nic_data->rx_rss_context_exclusive) {
2036 rc = efx_ef10_alloc_rss_context(efx, &new_rx_rss_context,
2038 if (rc == -EOPNOTSUPP)
2043 new_rx_rss_context = nic_data->rx_rss_context;
2046 rc = efx_ef10_populate_rss_table(efx, new_rx_rss_context,
2051 if (nic_data->rx_rss_context != new_rx_rss_context)
2052 efx_ef10_rx_free_indir_table(efx);
2053 nic_data->rx_rss_context = new_rx_rss_context;
2054 nic_data->rx_rss_context_exclusive = true;
2055 if (rx_indir_table != efx->rx_indir_table)
2056 memcpy(efx->rx_indir_table, rx_indir_table,
2057 sizeof(efx->rx_indir_table));
2061 if (new_rx_rss_context != nic_data->rx_rss_context)
2062 efx_ef10_free_rss_context(efx, new_rx_rss_context);
2064 netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
2068 static int efx_ef10_pf_rx_push_rss_config(struct efx_nic *efx, bool user,
2069 const u32 *rx_indir_table)
2073 if (efx->rss_spread == 1)
2076 rc = efx_ef10_rx_push_exclusive_rss_config(efx, rx_indir_table);
2078 if (rc == -ENOBUFS && !user) {
2079 unsigned context_size;
2080 bool mismatch = false;
2083 for (i = 0; i < ARRAY_SIZE(efx->rx_indir_table) && !mismatch;
2085 mismatch = rx_indir_table[i] !=
2086 ethtool_rxfh_indir_default(i, efx->rss_spread);
2088 rc = efx_ef10_rx_push_shared_rss_config(efx, &context_size);
2090 if (context_size != efx->rss_spread)
2091 netif_warn(efx, probe, efx->net_dev,
2092 "Could not allocate an exclusive RSS"
2093 " context; allocated a shared one of"
2095 " Wanted %u, got %u.\n",
2096 efx->rss_spread, context_size);
2098 netif_warn(efx, probe, efx->net_dev,
2099 "Could not allocate an exclusive RSS"
2100 " context; allocated a shared one but"
2101 " could not apply custom"
2104 netif_info(efx, probe, efx->net_dev,
2105 "Could not allocate an exclusive RSS"
2106 " context; allocated a shared one.\n");
2112 static int efx_ef10_vf_rx_push_rss_config(struct efx_nic *efx, bool user,
2113 const u32 *rx_indir_table
2114 __attribute__ ((unused)))
2116 struct efx_ef10_nic_data *nic_data = efx->nic_data;
2120 if (nic_data->rx_rss_context != EFX_EF10_RSS_CONTEXT_INVALID)
2122 return efx_ef10_rx_push_shared_rss_config(efx, NULL);
2125 static int efx_ef10_rx_probe(struct efx_rx_queue *rx_queue)
2127 return efx_nic_alloc_buffer(rx_queue->efx, &rx_queue->rxd.buf,
2128 (rx_queue->ptr_mask + 1) *
2129 sizeof(efx_qword_t),
2133 static void efx_ef10_rx_init(struct efx_rx_queue *rx_queue)
2135 MCDI_DECLARE_BUF(inbuf,
2136 MC_CMD_INIT_RXQ_IN_LEN(EFX_MAX_DMAQ_SIZE * 8 /
2138 struct efx_channel *channel = efx_rx_queue_channel(rx_queue);
2139 size_t entries = rx_queue->rxd.buf.len / EFX_BUF_SIZE;
2140 struct efx_nic *efx = rx_queue->efx;
2141 struct efx_ef10_nic_data *nic_data = efx->nic_data;
2143 dma_addr_t dma_addr;
2146 BUILD_BUG_ON(MC_CMD_INIT_RXQ_OUT_LEN != 0);
2148 rx_queue->scatter_n = 0;
2149 rx_queue->scatter_len = 0;
2151 MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_SIZE, rx_queue->ptr_mask + 1);
2152 MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_TARGET_EVQ, channel->channel);
2153 MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_LABEL, efx_rx_queue_index(rx_queue));
2154 MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_INSTANCE,
2155 efx_rx_queue_index(rx_queue));
2156 MCDI_POPULATE_DWORD_2(inbuf, INIT_RXQ_IN_FLAGS,
2157 INIT_RXQ_IN_FLAG_PREFIX, 1,
2158 INIT_RXQ_IN_FLAG_TIMESTAMP, 1);
2159 MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_OWNER_ID, 0);
2160 MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_PORT_ID, nic_data->vport_id);
2162 dma_addr = rx_queue->rxd.buf.dma_addr;
2164 netif_dbg(efx, hw, efx->net_dev, "pushing RXQ %d. %zu entries (%llx)\n",
2165 efx_rx_queue_index(rx_queue), entries, (u64)dma_addr);
2167 for (i = 0; i < entries; ++i) {
2168 MCDI_SET_ARRAY_QWORD(inbuf, INIT_RXQ_IN_DMA_ADDR, i, dma_addr);
2169 dma_addr += EFX_BUF_SIZE;
2172 inlen = MC_CMD_INIT_RXQ_IN_LEN(entries);
2174 rc = efx_mcdi_rpc(efx, MC_CMD_INIT_RXQ, inbuf, inlen,
2177 netdev_WARN(efx->net_dev, "failed to initialise RXQ %d\n",
2178 efx_rx_queue_index(rx_queue));
2181 static void efx_ef10_rx_fini(struct efx_rx_queue *rx_queue)
2183 MCDI_DECLARE_BUF(inbuf, MC_CMD_FINI_RXQ_IN_LEN);
2184 MCDI_DECLARE_BUF_ERR(outbuf);
2185 struct efx_nic *efx = rx_queue->efx;
2189 MCDI_SET_DWORD(inbuf, FINI_RXQ_IN_INSTANCE,
2190 efx_rx_queue_index(rx_queue));
2192 rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FINI_RXQ, inbuf, sizeof(inbuf),
2193 outbuf, sizeof(outbuf), &outlen);
2195 if (rc && rc != -EALREADY)
2201 efx_mcdi_display_error(efx, MC_CMD_FINI_RXQ, MC_CMD_FINI_RXQ_IN_LEN,
2202 outbuf, outlen, rc);
2205 static void efx_ef10_rx_remove(struct efx_rx_queue *rx_queue)
2207 efx_nic_free_buffer(rx_queue->efx, &rx_queue->rxd.buf);
2210 /* This creates an entry in the RX descriptor queue */
2212 efx_ef10_build_rx_desc(struct efx_rx_queue *rx_queue, unsigned int index)
2214 struct efx_rx_buffer *rx_buf;
2217 rxd = efx_rx_desc(rx_queue, index);
2218 rx_buf = efx_rx_buffer(rx_queue, index);
2219 EFX_POPULATE_QWORD_2(*rxd,
2220 ESF_DZ_RX_KER_BYTE_CNT, rx_buf->len,
2221 ESF_DZ_RX_KER_BUF_ADDR, rx_buf->dma_addr);
2224 static void efx_ef10_rx_write(struct efx_rx_queue *rx_queue)
2226 struct efx_nic *efx = rx_queue->efx;
2227 unsigned int write_count;
2230 /* Firmware requires that RX_DESC_WPTR be a multiple of 8 */
2231 write_count = rx_queue->added_count & ~7;
2232 if (rx_queue->notified_count == write_count)
2236 efx_ef10_build_rx_desc(
2238 rx_queue->notified_count & rx_queue->ptr_mask);
2239 while (++rx_queue->notified_count != write_count);
2242 EFX_POPULATE_DWORD_1(reg, ERF_DZ_RX_DESC_WPTR,
2243 write_count & rx_queue->ptr_mask);
2244 efx_writed_page(efx, ®, ER_DZ_RX_DESC_UPD,
2245 efx_rx_queue_index(rx_queue));
2248 static efx_mcdi_async_completer efx_ef10_rx_defer_refill_complete;
2250 static void efx_ef10_rx_defer_refill(struct efx_rx_queue *rx_queue)
2252 struct efx_channel *channel = efx_rx_queue_channel(rx_queue);
2253 MCDI_DECLARE_BUF(inbuf, MC_CMD_DRIVER_EVENT_IN_LEN);
2256 EFX_POPULATE_QWORD_2(event,
2257 ESF_DZ_EV_CODE, EFX_EF10_DRVGEN_EV,
2258 ESF_DZ_EV_DATA, EFX_EF10_REFILL);
2260 MCDI_SET_DWORD(inbuf, DRIVER_EVENT_IN_EVQ, channel->channel);
2262 /* MCDI_SET_QWORD is not appropriate here since EFX_POPULATE_* has
2263 * already swapped the data to little-endian order.
2265 memcpy(MCDI_PTR(inbuf, DRIVER_EVENT_IN_DATA), &event.u64[0],
2266 sizeof(efx_qword_t));
2268 efx_mcdi_rpc_async(channel->efx, MC_CMD_DRIVER_EVENT,
2269 inbuf, sizeof(inbuf), 0,
2270 efx_ef10_rx_defer_refill_complete, 0);
2274 efx_ef10_rx_defer_refill_complete(struct efx_nic *efx, unsigned long cookie,
2275 int rc, efx_dword_t *outbuf,
2276 size_t outlen_actual)
2281 static int efx_ef10_ev_probe(struct efx_channel *channel)
2283 return efx_nic_alloc_buffer(channel->efx, &channel->eventq.buf,
2284 (channel->eventq_mask + 1) *
2285 sizeof(efx_qword_t),
2289 static void efx_ef10_ev_fini(struct efx_channel *channel)
2291 MCDI_DECLARE_BUF(inbuf, MC_CMD_FINI_EVQ_IN_LEN);
2292 MCDI_DECLARE_BUF_ERR(outbuf);
2293 struct efx_nic *efx = channel->efx;
2297 MCDI_SET_DWORD(inbuf, FINI_EVQ_IN_INSTANCE, channel->channel);
2299 rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FINI_EVQ, inbuf, sizeof(inbuf),
2300 outbuf, sizeof(outbuf), &outlen);
2302 if (rc && rc != -EALREADY)
2308 efx_mcdi_display_error(efx, MC_CMD_FINI_EVQ, MC_CMD_FINI_EVQ_IN_LEN,
2309 outbuf, outlen, rc);
2312 static int efx_ef10_ev_init(struct efx_channel *channel)
2314 MCDI_DECLARE_BUF(inbuf,
2315 MC_CMD_INIT_EVQ_IN_LEN(EFX_MAX_EVQ_SIZE * 8 /
2317 MCDI_DECLARE_BUF(outbuf, MC_CMD_INIT_EVQ_OUT_LEN);
2318 size_t entries = channel->eventq.buf.len / EFX_BUF_SIZE;
2319 struct efx_nic *efx = channel->efx;
2320 struct efx_ef10_nic_data *nic_data;
2321 bool supports_rx_merge;
2322 size_t inlen, outlen;
2323 unsigned int enabled, implemented;
2324 dma_addr_t dma_addr;
2328 nic_data = efx->nic_data;
2330 !!(nic_data->datapath_caps &
2331 1 << MC_CMD_GET_CAPABILITIES_OUT_RX_BATCHING_LBN);
2333 /* Fill event queue with all ones (i.e. empty events) */
2334 memset(channel->eventq.buf.addr, 0xff, channel->eventq.buf.len);
2336 MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_SIZE, channel->eventq_mask + 1);
2337 MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_INSTANCE, channel->channel);
2338 /* INIT_EVQ expects index in vector table, not absolute */
2339 MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_IRQ_NUM, channel->channel);
2340 MCDI_POPULATE_DWORD_4(inbuf, INIT_EVQ_IN_FLAGS,
2341 INIT_EVQ_IN_FLAG_INTERRUPTING, 1,
2342 INIT_EVQ_IN_FLAG_RX_MERGE, 1,
2343 INIT_EVQ_IN_FLAG_TX_MERGE, 1,
2344 INIT_EVQ_IN_FLAG_CUT_THRU, !supports_rx_merge);
2345 MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_TMR_MODE,
2346 MC_CMD_INIT_EVQ_IN_TMR_MODE_DIS);
2347 MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_TMR_LOAD, 0);
2348 MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_TMR_RELOAD, 0);
2349 MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_COUNT_MODE,
2350 MC_CMD_INIT_EVQ_IN_COUNT_MODE_DIS);
2351 MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_COUNT_THRSHLD, 0);
2353 dma_addr = channel->eventq.buf.dma_addr;
2354 for (i = 0; i < entries; ++i) {
2355 MCDI_SET_ARRAY_QWORD(inbuf, INIT_EVQ_IN_DMA_ADDR, i, dma_addr);
2356 dma_addr += EFX_BUF_SIZE;
2359 inlen = MC_CMD_INIT_EVQ_IN_LEN(entries);
2361 rc = efx_mcdi_rpc(efx, MC_CMD_INIT_EVQ, inbuf, inlen,
2362 outbuf, sizeof(outbuf), &outlen);
2363 /* IRQ return is ignored */
2364 if (channel->channel || rc)
2367 /* Successfully created event queue on channel 0 */
2368 rc = efx_mcdi_get_workarounds(efx, &implemented, &enabled);
2369 if (rc == -ENOSYS) {
2370 /* GET_WORKAROUNDS was implemented before the bug26807
2371 * workaround, thus the latter must be unavailable in this fw
2373 nic_data->workaround_26807 = false;
2378 nic_data->workaround_26807 =
2379 !!(enabled & MC_CMD_GET_WORKAROUNDS_OUT_BUG26807);
2381 if (implemented & MC_CMD_GET_WORKAROUNDS_OUT_BUG26807 &&
2382 !nic_data->workaround_26807) {
2385 rc = efx_mcdi_set_workaround(efx,
2386 MC_CMD_WORKAROUND_BUG26807,
2391 1 << MC_CMD_WORKAROUND_EXT_OUT_FLR_DONE_LBN) {
2392 netif_info(efx, drv, efx->net_dev,
2393 "other functions on NIC have been reset\n");
2394 /* MC's boot count has incremented */
2395 ++nic_data->warm_boot_count;
2397 nic_data->workaround_26807 = true;
2398 } else if (rc == -EPERM) {
2408 efx_ef10_ev_fini(channel);
2412 static void efx_ef10_ev_remove(struct efx_channel *channel)
2414 efx_nic_free_buffer(channel->efx, &channel->eventq.buf);
2417 static void efx_ef10_handle_rx_wrong_queue(struct efx_rx_queue *rx_queue,
2418 unsigned int rx_queue_label)
2420 struct efx_nic *efx = rx_queue->efx;
2422 netif_info(efx, hw, efx->net_dev,
2423 "rx event arrived on queue %d labeled as queue %u\n",
2424 efx_rx_queue_index(rx_queue), rx_queue_label);
2426 efx_schedule_reset(efx, RESET_TYPE_DISABLE);
2430 efx_ef10_handle_rx_bad_lbits(struct efx_rx_queue *rx_queue,
2431 unsigned int actual, unsigned int expected)
2433 unsigned int dropped = (actual - expected) & rx_queue->ptr_mask;
2434 struct efx_nic *efx = rx_queue->efx;
2436 netif_info(efx, hw, efx->net_dev,
2437 "dropped %d events (index=%d expected=%d)\n",
2438 dropped, actual, expected);
2440 efx_schedule_reset(efx, RESET_TYPE_DISABLE);
2443 /* partially received RX was aborted. clean up. */
2444 static void efx_ef10_handle_rx_abort(struct efx_rx_queue *rx_queue)
2446 unsigned int rx_desc_ptr;
2448 netif_dbg(rx_queue->efx, hw, rx_queue->efx->net_dev,
2449 "scattered RX aborted (dropping %u buffers)\n",
2450 rx_queue->scatter_n);
2452 rx_desc_ptr = rx_queue->removed_count & rx_queue->ptr_mask;
2454 efx_rx_packet(rx_queue, rx_desc_ptr, rx_queue->scatter_n,
2455 0, EFX_RX_PKT_DISCARD);
2457 rx_queue->removed_count += rx_queue->scatter_n;
2458 rx_queue->scatter_n = 0;
2459 rx_queue->scatter_len = 0;
2460 ++efx_rx_queue_channel(rx_queue)->n_rx_nodesc_trunc;
2463 static int efx_ef10_handle_rx_event(struct efx_channel *channel,
2464 const efx_qword_t *event)
2466 unsigned int rx_bytes, next_ptr_lbits, rx_queue_label, rx_l4_class;
2467 unsigned int n_descs, n_packets, i;
2468 struct efx_nic *efx = channel->efx;
2469 struct efx_rx_queue *rx_queue;
2473 if (unlikely(ACCESS_ONCE(efx->reset_pending)))
2476 /* Basic packet information */
2477 rx_bytes = EFX_QWORD_FIELD(*event, ESF_DZ_RX_BYTES);
2478 next_ptr_lbits = EFX_QWORD_FIELD(*event, ESF_DZ_RX_DSC_PTR_LBITS);
2479 rx_queue_label = EFX_QWORD_FIELD(*event, ESF_DZ_RX_QLABEL);
2480 rx_l4_class = EFX_QWORD_FIELD(*event, ESF_DZ_RX_L4_CLASS);
2481 rx_cont = EFX_QWORD_FIELD(*event, ESF_DZ_RX_CONT);
2483 if (EFX_QWORD_FIELD(*event, ESF_DZ_RX_DROP_EVENT))
2484 netdev_WARN(efx->net_dev, "saw RX_DROP_EVENT: event="
2486 EFX_QWORD_VAL(*event));
2488 rx_queue = efx_channel_get_rx_queue(channel);
2490 if (unlikely(rx_queue_label != efx_rx_queue_index(rx_queue)))
2491 efx_ef10_handle_rx_wrong_queue(rx_queue, rx_queue_label);
2493 n_descs = ((next_ptr_lbits - rx_queue->removed_count) &
2494 ((1 << ESF_DZ_RX_DSC_PTR_LBITS_WIDTH) - 1));
2496 if (n_descs != rx_queue->scatter_n + 1) {
2497 struct efx_ef10_nic_data *nic_data = efx->nic_data;
2499 /* detect rx abort */
2500 if (unlikely(n_descs == rx_queue->scatter_n)) {
2501 if (rx_queue->scatter_n == 0 || rx_bytes != 0)
2502 netdev_WARN(efx->net_dev,
2503 "invalid RX abort: scatter_n=%u event="
2505 rx_queue->scatter_n,
2506 EFX_QWORD_VAL(*event));
2507 efx_ef10_handle_rx_abort(rx_queue);
2511 /* Check that RX completion merging is valid, i.e.
2512 * the current firmware supports it and this is a
2513 * non-scattered packet.
2515 if (!(nic_data->datapath_caps &
2516 (1 << MC_CMD_GET_CAPABILITIES_OUT_RX_BATCHING_LBN)) ||
2517 rx_queue->scatter_n != 0 || rx_cont) {
2518 efx_ef10_handle_rx_bad_lbits(
2519 rx_queue, next_ptr_lbits,
2520 (rx_queue->removed_count +
2521 rx_queue->scatter_n + 1) &
2522 ((1 << ESF_DZ_RX_DSC_PTR_LBITS_WIDTH) - 1));
2526 /* Merged completion for multiple non-scattered packets */
2527 rx_queue->scatter_n = 1;
2528 rx_queue->scatter_len = 0;
2529 n_packets = n_descs;
2530 ++channel->n_rx_merge_events;
2531 channel->n_rx_merge_packets += n_packets;
2532 flags |= EFX_RX_PKT_PREFIX_LEN;
2534 ++rx_queue->scatter_n;
2535 rx_queue->scatter_len += rx_bytes;
2541 if (unlikely(EFX_QWORD_FIELD(*event, ESF_DZ_RX_ECRC_ERR)))
2542 flags |= EFX_RX_PKT_DISCARD;
2544 if (unlikely(EFX_QWORD_FIELD(*event, ESF_DZ_RX_IPCKSUM_ERR))) {
2545 channel->n_rx_ip_hdr_chksum_err += n_packets;
2546 } else if (unlikely(EFX_QWORD_FIELD(*event,
2547 ESF_DZ_RX_TCPUDP_CKSUM_ERR))) {
2548 channel->n_rx_tcp_udp_chksum_err += n_packets;
2549 } else if (rx_l4_class == ESE_DZ_L4_CLASS_TCP ||
2550 rx_l4_class == ESE_DZ_L4_CLASS_UDP) {
2551 flags |= EFX_RX_PKT_CSUMMED;
2554 if (rx_l4_class == ESE_DZ_L4_CLASS_TCP)
2555 flags |= EFX_RX_PKT_TCP;
2557 channel->irq_mod_score += 2 * n_packets;
2559 /* Handle received packet(s) */
2560 for (i = 0; i < n_packets; i++) {
2561 efx_rx_packet(rx_queue,
2562 rx_queue->removed_count & rx_queue->ptr_mask,
2563 rx_queue->scatter_n, rx_queue->scatter_len,
2565 rx_queue->removed_count += rx_queue->scatter_n;
2568 rx_queue->scatter_n = 0;
2569 rx_queue->scatter_len = 0;
2575 efx_ef10_handle_tx_event(struct efx_channel *channel, efx_qword_t *event)
2577 struct efx_nic *efx = channel->efx;
2578 struct efx_tx_queue *tx_queue;
2579 unsigned int tx_ev_desc_ptr;
2580 unsigned int tx_ev_q_label;
2583 if (unlikely(ACCESS_ONCE(efx->reset_pending)))
2586 if (unlikely(EFX_QWORD_FIELD(*event, ESF_DZ_TX_DROP_EVENT)))
2589 /* Transmit completion */
2590 tx_ev_desc_ptr = EFX_QWORD_FIELD(*event, ESF_DZ_TX_DESCR_INDX);
2591 tx_ev_q_label = EFX_QWORD_FIELD(*event, ESF_DZ_TX_QLABEL);
2592 tx_queue = efx_channel_get_tx_queue(channel,
2593 tx_ev_q_label % EFX_TXQ_TYPES);
2594 tx_descs = ((tx_ev_desc_ptr + 1 - tx_queue->read_count) &
2595 tx_queue->ptr_mask);
2596 efx_xmit_done(tx_queue, tx_ev_desc_ptr & tx_queue->ptr_mask);
2602 efx_ef10_handle_driver_event(struct efx_channel *channel, efx_qword_t *event)
2604 struct efx_nic *efx = channel->efx;
2607 subcode = EFX_QWORD_FIELD(*event, ESF_DZ_DRV_SUB_CODE);
2610 case ESE_DZ_DRV_TIMER_EV:
2611 case ESE_DZ_DRV_WAKE_UP_EV:
2613 case ESE_DZ_DRV_START_UP_EV:
2614 /* event queue init complete. ok. */
2617 netif_err(efx, hw, efx->net_dev,
2618 "channel %d unknown driver event type %d"
2619 " (data " EFX_QWORD_FMT ")\n",
2620 channel->channel, subcode,
2621 EFX_QWORD_VAL(*event));
2626 static void efx_ef10_handle_driver_generated_event(struct efx_channel *channel,
2629 struct efx_nic *efx = channel->efx;
2632 subcode = EFX_QWORD_FIELD(*event, EFX_DWORD_0);
2636 channel->event_test_cpu = raw_smp_processor_id();
2638 case EFX_EF10_REFILL:
2639 /* The queue must be empty, so we won't receive any rx
2640 * events, so efx_process_channel() won't refill the
2641 * queue. Refill it here
2643 efx_fast_push_rx_descriptors(&channel->rx_queue, true);
2646 netif_err(efx, hw, efx->net_dev,
2647 "channel %d unknown driver event type %u"
2648 " (data " EFX_QWORD_FMT ")\n",
2649 channel->channel, (unsigned) subcode,
2650 EFX_QWORD_VAL(*event));
2654 static int efx_ef10_ev_process(struct efx_channel *channel, int quota)
2656 struct efx_nic *efx = channel->efx;
2657 efx_qword_t event, *p_event;
2658 unsigned int read_ptr;
2666 read_ptr = channel->eventq_read_ptr;
2669 p_event = efx_event(channel, read_ptr);
2672 if (!efx_event_present(&event))
2675 EFX_SET_QWORD(*p_event);
2679 ev_code = EFX_QWORD_FIELD(event, ESF_DZ_EV_CODE);
2681 netif_vdbg(efx, drv, efx->net_dev,
2682 "processing event on %d " EFX_QWORD_FMT "\n",
2683 channel->channel, EFX_QWORD_VAL(event));
2686 case ESE_DZ_EV_CODE_MCDI_EV:
2687 efx_mcdi_process_event(channel, &event);
2689 case ESE_DZ_EV_CODE_RX_EV:
2690 spent += efx_ef10_handle_rx_event(channel, &event);
2691 if (spent >= quota) {
2692 /* XXX can we split a merged event to
2693 * avoid going over-quota?
2699 case ESE_DZ_EV_CODE_TX_EV:
2700 tx_descs += efx_ef10_handle_tx_event(channel, &event);
2701 if (tx_descs > efx->txq_entries) {
2704 } else if (++spent == quota) {
2708 case ESE_DZ_EV_CODE_DRIVER_EV:
2709 efx_ef10_handle_driver_event(channel, &event);
2710 if (++spent == quota)
2713 case EFX_EF10_DRVGEN_EV:
2714 efx_ef10_handle_driver_generated_event(channel, &event);
2717 netif_err(efx, hw, efx->net_dev,
2718 "channel %d unknown event type %d"
2719 " (data " EFX_QWORD_FMT ")\n",
2720 channel->channel, ev_code,
2721 EFX_QWORD_VAL(event));
2726 channel->eventq_read_ptr = read_ptr;
2730 static void efx_ef10_ev_read_ack(struct efx_channel *channel)
2732 struct efx_nic *efx = channel->efx;
2735 if (EFX_EF10_WORKAROUND_35388(efx)) {
2736 BUILD_BUG_ON(EFX_MIN_EVQ_SIZE <
2737 (1 << ERF_DD_EVQ_IND_RPTR_WIDTH));
2738 BUILD_BUG_ON(EFX_MAX_EVQ_SIZE >
2739 (1 << 2 * ERF_DD_EVQ_IND_RPTR_WIDTH));
2741 EFX_POPULATE_DWORD_2(rptr, ERF_DD_EVQ_IND_RPTR_FLAGS,
2742 EFE_DD_EVQ_IND_RPTR_FLAGS_HIGH,
2743 ERF_DD_EVQ_IND_RPTR,
2744 (channel->eventq_read_ptr &
2745 channel->eventq_mask) >>
2746 ERF_DD_EVQ_IND_RPTR_WIDTH);
2747 efx_writed_page(efx, &rptr, ER_DD_EVQ_INDIRECT,
2749 EFX_POPULATE_DWORD_2(rptr, ERF_DD_EVQ_IND_RPTR_FLAGS,
2750 EFE_DD_EVQ_IND_RPTR_FLAGS_LOW,
2751 ERF_DD_EVQ_IND_RPTR,
2752 channel->eventq_read_ptr &
2753 ((1 << ERF_DD_EVQ_IND_RPTR_WIDTH) - 1));
2754 efx_writed_page(efx, &rptr, ER_DD_EVQ_INDIRECT,
2757 EFX_POPULATE_DWORD_1(rptr, ERF_DZ_EVQ_RPTR,
2758 channel->eventq_read_ptr &
2759 channel->eventq_mask);
2760 efx_writed_page(efx, &rptr, ER_DZ_EVQ_RPTR, channel->channel);
2764 static void efx_ef10_ev_test_generate(struct efx_channel *channel)
2766 MCDI_DECLARE_BUF(inbuf, MC_CMD_DRIVER_EVENT_IN_LEN);
2767 struct efx_nic *efx = channel->efx;
2771 EFX_POPULATE_QWORD_2(event,
2772 ESF_DZ_EV_CODE, EFX_EF10_DRVGEN_EV,
2773 ESF_DZ_EV_DATA, EFX_EF10_TEST);
2775 MCDI_SET_DWORD(inbuf, DRIVER_EVENT_IN_EVQ, channel->channel);
2777 /* MCDI_SET_QWORD is not appropriate here since EFX_POPULATE_* has
2778 * already swapped the data to little-endian order.
2780 memcpy(MCDI_PTR(inbuf, DRIVER_EVENT_IN_DATA), &event.u64[0],
2781 sizeof(efx_qword_t));
2783 rc = efx_mcdi_rpc(efx, MC_CMD_DRIVER_EVENT, inbuf, sizeof(inbuf),
2792 netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
2795 void efx_ef10_handle_drain_event(struct efx_nic *efx)
2797 if (atomic_dec_and_test(&efx->active_queues))
2798 wake_up(&efx->flush_wq);
2800 WARN_ON(atomic_read(&efx->active_queues) < 0);
2803 static int efx_ef10_fini_dmaq(struct efx_nic *efx)
2805 struct efx_ef10_nic_data *nic_data = efx->nic_data;
2806 struct efx_channel *channel;
2807 struct efx_tx_queue *tx_queue;
2808 struct efx_rx_queue *rx_queue;
2811 /* If the MC has just rebooted, the TX/RX queues will have already been
2812 * torn down, but efx->active_queues needs to be set to zero.
2814 if (nic_data->must_realloc_vis) {
2815 atomic_set(&efx->active_queues, 0);
2819 /* Do not attempt to write to the NIC during EEH recovery */
2820 if (efx->state != STATE_RECOVERY) {
2821 efx_for_each_channel(channel, efx) {
2822 efx_for_each_channel_rx_queue(rx_queue, channel)
2823 efx_ef10_rx_fini(rx_queue);
2824 efx_for_each_channel_tx_queue(tx_queue, channel)
2825 efx_ef10_tx_fini(tx_queue);
2828 wait_event_timeout(efx->flush_wq,
2829 atomic_read(&efx->active_queues) == 0,
2830 msecs_to_jiffies(EFX_MAX_FLUSH_TIME));
2831 pending = atomic_read(&efx->active_queues);
2833 netif_err(efx, hw, efx->net_dev, "failed to flush %d queues\n",
2842 static void efx_ef10_prepare_flr(struct efx_nic *efx)
2844 atomic_set(&efx->active_queues, 0);
2847 static bool efx_ef10_filter_equal(const struct efx_filter_spec *left,
2848 const struct efx_filter_spec *right)
2850 if ((left->match_flags ^ right->match_flags) |
2851 ((left->flags ^ right->flags) &
2852 (EFX_FILTER_FLAG_RX | EFX_FILTER_FLAG_TX)))
2855 return memcmp(&left->outer_vid, &right->outer_vid,
2856 sizeof(struct efx_filter_spec) -
2857 offsetof(struct efx_filter_spec, outer_vid)) == 0;
2860 static unsigned int efx_ef10_filter_hash(const struct efx_filter_spec *spec)
2862 BUILD_BUG_ON(offsetof(struct efx_filter_spec, outer_vid) & 3);
2863 return jhash2((const u32 *)&spec->outer_vid,
2864 (sizeof(struct efx_filter_spec) -
2865 offsetof(struct efx_filter_spec, outer_vid)) / 4,
2867 /* XXX should we randomise the initval? */
2870 /* Decide whether a filter should be exclusive or else should allow
2871 * delivery to additional recipients. Currently we decide that
2872 * filters for specific local unicast MAC and IP addresses are
2875 static bool efx_ef10_filter_is_exclusive(const struct efx_filter_spec *spec)
2877 if (spec->match_flags & EFX_FILTER_MATCH_LOC_MAC &&
2878 !is_multicast_ether_addr(spec->loc_mac))
2881 if ((spec->match_flags &
2882 (EFX_FILTER_MATCH_ETHER_TYPE | EFX_FILTER_MATCH_LOC_HOST)) ==
2883 (EFX_FILTER_MATCH_ETHER_TYPE | EFX_FILTER_MATCH_LOC_HOST)) {
2884 if (spec->ether_type == htons(ETH_P_IP) &&
2885 !ipv4_is_multicast(spec->loc_host[0]))
2887 if (spec->ether_type == htons(ETH_P_IPV6) &&
2888 ((const u8 *)spec->loc_host)[0] != 0xff)
2895 static struct efx_filter_spec *
2896 efx_ef10_filter_entry_spec(const struct efx_ef10_filter_table *table,
2897 unsigned int filter_idx)
2899 return (struct efx_filter_spec *)(table->entry[filter_idx].spec &
2900 ~EFX_EF10_FILTER_FLAGS);
2904 efx_ef10_filter_entry_flags(const struct efx_ef10_filter_table *table,
2905 unsigned int filter_idx)
2907 return table->entry[filter_idx].spec & EFX_EF10_FILTER_FLAGS;
2911 efx_ef10_filter_set_entry(struct efx_ef10_filter_table *table,
2912 unsigned int filter_idx,
2913 const struct efx_filter_spec *spec,
2916 table->entry[filter_idx].spec = (unsigned long)spec | flags;
2919 static void efx_ef10_filter_push_prep(struct efx_nic *efx,
2920 const struct efx_filter_spec *spec,
2921 efx_dword_t *inbuf, u64 handle,
2924 struct efx_ef10_nic_data *nic_data = efx->nic_data;
2926 memset(inbuf, 0, MC_CMD_FILTER_OP_IN_LEN);
2929 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
2930 MC_CMD_FILTER_OP_IN_OP_REPLACE);
2931 MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE, handle);
2933 u32 match_fields = 0;
2935 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
2936 efx_ef10_filter_is_exclusive(spec) ?
2937 MC_CMD_FILTER_OP_IN_OP_INSERT :
2938 MC_CMD_FILTER_OP_IN_OP_SUBSCRIBE);
2940 /* Convert match flags and values. Unlike almost
2941 * everything else in MCDI, these fields are in
2942 * network byte order.
2944 if (spec->match_flags & EFX_FILTER_MATCH_LOC_MAC_IG)
2946 is_multicast_ether_addr(spec->loc_mac) ?
2947 1 << MC_CMD_FILTER_OP_IN_MATCH_UNKNOWN_MCAST_DST_LBN :
2948 1 << MC_CMD_FILTER_OP_IN_MATCH_UNKNOWN_UCAST_DST_LBN;
2949 #define COPY_FIELD(gen_flag, gen_field, mcdi_field) \
2950 if (spec->match_flags & EFX_FILTER_MATCH_ ## gen_flag) { \
2952 1 << MC_CMD_FILTER_OP_IN_MATCH_ ## \
2953 mcdi_field ## _LBN; \
2955 MC_CMD_FILTER_OP_IN_ ## mcdi_field ## _LEN < \
2956 sizeof(spec->gen_field)); \
2957 memcpy(MCDI_PTR(inbuf, FILTER_OP_IN_ ## mcdi_field), \
2958 &spec->gen_field, sizeof(spec->gen_field)); \
2960 COPY_FIELD(REM_HOST, rem_host, SRC_IP);
2961 COPY_FIELD(LOC_HOST, loc_host, DST_IP);
2962 COPY_FIELD(REM_MAC, rem_mac, SRC_MAC);
2963 COPY_FIELD(REM_PORT, rem_port, SRC_PORT);
2964 COPY_FIELD(LOC_MAC, loc_mac, DST_MAC);
2965 COPY_FIELD(LOC_PORT, loc_port, DST_PORT);
2966 COPY_FIELD(ETHER_TYPE, ether_type, ETHER_TYPE);
2967 COPY_FIELD(INNER_VID, inner_vid, INNER_VLAN);
2968 COPY_FIELD(OUTER_VID, outer_vid, OUTER_VLAN);
2969 COPY_FIELD(IP_PROTO, ip_proto, IP_PROTO);
2971 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_MATCH_FIELDS,
2975 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_PORT_ID, nic_data->vport_id);
2976 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_DEST,
2977 spec->dmaq_id == EFX_FILTER_RX_DMAQ_ID_DROP ?
2978 MC_CMD_FILTER_OP_IN_RX_DEST_DROP :
2979 MC_CMD_FILTER_OP_IN_RX_DEST_HOST);
2980 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_TX_DOMAIN, 0);
2981 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_TX_DEST,
2982 MC_CMD_FILTER_OP_IN_TX_DEST_DEFAULT);
2983 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_QUEUE,
2984 spec->dmaq_id == EFX_FILTER_RX_DMAQ_ID_DROP ?
2986 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_MODE,
2987 (spec->flags & EFX_FILTER_FLAG_RX_RSS) ?
2988 MC_CMD_FILTER_OP_IN_RX_MODE_RSS :
2989 MC_CMD_FILTER_OP_IN_RX_MODE_SIMPLE);
2990 if (spec->flags & EFX_FILTER_FLAG_RX_RSS)
2991 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_CONTEXT,
2992 spec->rss_context !=
2993 EFX_FILTER_RSS_CONTEXT_DEFAULT ?
2994 spec->rss_context : nic_data->rx_rss_context);
2997 static int efx_ef10_filter_push(struct efx_nic *efx,
2998 const struct efx_filter_spec *spec,
2999 u64 *handle, bool replacing)
3001 MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_IN_LEN);
3002 MCDI_DECLARE_BUF(outbuf, MC_CMD_FILTER_OP_OUT_LEN);
3005 efx_ef10_filter_push_prep(efx, spec, inbuf, *handle, replacing);
3006 rc = efx_mcdi_rpc(efx, MC_CMD_FILTER_OP, inbuf, sizeof(inbuf),
3007 outbuf, sizeof(outbuf), NULL);
3009 *handle = MCDI_QWORD(outbuf, FILTER_OP_OUT_HANDLE);
3011 rc = -EBUSY; /* to match efx_farch_filter_insert() */
3015 static int efx_ef10_filter_rx_match_pri(struct efx_ef10_filter_table *table,
3016 enum efx_filter_match_flags match_flags)
3018 unsigned int match_pri;
3021 match_pri < table->rx_match_count;
3023 if (table->rx_match_flags[match_pri] == match_flags)
3026 return -EPROTONOSUPPORT;
3029 static s32 efx_ef10_filter_insert(struct efx_nic *efx,
3030 struct efx_filter_spec *spec,
3033 struct efx_ef10_filter_table *table = efx->filter_state;
3034 DECLARE_BITMAP(mc_rem_map, EFX_EF10_FILTER_SEARCH_LIMIT);
3035 struct efx_filter_spec *saved_spec;
3036 unsigned int match_pri, hash;
3037 unsigned int priv_flags;
3038 bool replacing = false;
3044 /* For now, only support RX filters */
3045 if ((spec->flags & (EFX_FILTER_FLAG_RX | EFX_FILTER_FLAG_TX)) !=
3049 rc = efx_ef10_filter_rx_match_pri(table, spec->match_flags);
3054 hash = efx_ef10_filter_hash(spec);
3055 is_mc_recip = efx_filter_is_mc_recipient(spec);
3057 bitmap_zero(mc_rem_map, EFX_EF10_FILTER_SEARCH_LIMIT);
3059 /* Find any existing filters with the same match tuple or
3060 * else a free slot to insert at. If any of them are busy,
3061 * we have to wait and retry.
3064 unsigned int depth = 1;
3067 spin_lock_bh(&efx->filter_lock);
3070 i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
3071 saved_spec = efx_ef10_filter_entry_spec(table, i);
3076 } else if (efx_ef10_filter_equal(spec, saved_spec)) {
3077 if (table->entry[i].spec &
3078 EFX_EF10_FILTER_FLAG_BUSY)
3080 if (spec->priority < saved_spec->priority &&
3081 spec->priority != EFX_FILTER_PRI_AUTO) {
3086 /* This is the only one */
3087 if (spec->priority ==
3088 saved_spec->priority &&
3095 } else if (spec->priority >
3096 saved_spec->priority ||
3098 saved_spec->priority &&
3103 __set_bit(depth, mc_rem_map);
3107 /* Once we reach the maximum search depth, use
3108 * the first suitable slot or return -EBUSY if
3111 if (depth == EFX_EF10_FILTER_SEARCH_LIMIT) {
3112 if (ins_index < 0) {
3122 prepare_to_wait(&table->waitq, &wait, TASK_UNINTERRUPTIBLE);
3123 spin_unlock_bh(&efx->filter_lock);
3128 /* Create a software table entry if necessary, and mark it
3129 * busy. We might yet fail to insert, but any attempt to
3130 * insert a conflicting filter while we're waiting for the
3131 * firmware must find the busy entry.
3133 saved_spec = efx_ef10_filter_entry_spec(table, ins_index);
3135 if (spec->priority == EFX_FILTER_PRI_AUTO &&
3136 saved_spec->priority >= EFX_FILTER_PRI_AUTO) {
3137 /* Just make sure it won't be removed */
3138 if (saved_spec->priority > EFX_FILTER_PRI_AUTO)
3139 saved_spec->flags |= EFX_FILTER_FLAG_RX_OVER_AUTO;
3140 table->entry[ins_index].spec &=
3141 ~EFX_EF10_FILTER_FLAG_AUTO_OLD;
3146 priv_flags = efx_ef10_filter_entry_flags(table, ins_index);
3148 saved_spec = kmalloc(sizeof(*spec), GFP_ATOMIC);
3153 *saved_spec = *spec;
3156 efx_ef10_filter_set_entry(table, ins_index, saved_spec,
3157 priv_flags | EFX_EF10_FILTER_FLAG_BUSY);
3159 /* Mark lower-priority multicast recipients busy prior to removal */
3161 unsigned int depth, i;
3163 for (depth = 0; depth < EFX_EF10_FILTER_SEARCH_LIMIT; depth++) {
3164 i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
3165 if (test_bit(depth, mc_rem_map))
3166 table->entry[i].spec |=
3167 EFX_EF10_FILTER_FLAG_BUSY;
3171 spin_unlock_bh(&efx->filter_lock);
3173 rc = efx_ef10_filter_push(efx, spec, &table->entry[ins_index].handle,
3176 /* Finalise the software table entry */
3177 spin_lock_bh(&efx->filter_lock);
3180 /* Update the fields that may differ */
3181 if (saved_spec->priority == EFX_FILTER_PRI_AUTO)
3182 saved_spec->flags |=
3183 EFX_FILTER_FLAG_RX_OVER_AUTO;
3184 saved_spec->priority = spec->priority;
3185 saved_spec->flags &= EFX_FILTER_FLAG_RX_OVER_AUTO;
3186 saved_spec->flags |= spec->flags;
3187 saved_spec->rss_context = spec->rss_context;
3188 saved_spec->dmaq_id = spec->dmaq_id;
3190 } else if (!replacing) {
3194 efx_ef10_filter_set_entry(table, ins_index, saved_spec, priv_flags);
3196 /* Remove and finalise entries for lower-priority multicast
3200 MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_IN_LEN);
3201 unsigned int depth, i;
3203 memset(inbuf, 0, sizeof(inbuf));
3205 for (depth = 0; depth < EFX_EF10_FILTER_SEARCH_LIMIT; depth++) {
3206 if (!test_bit(depth, mc_rem_map))
3209 i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
3210 saved_spec = efx_ef10_filter_entry_spec(table, i);
3211 priv_flags = efx_ef10_filter_entry_flags(table, i);
3214 spin_unlock_bh(&efx->filter_lock);
3215 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
3216 MC_CMD_FILTER_OP_IN_OP_UNSUBSCRIBE);
3217 MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
3218 table->entry[i].handle);
3219 rc = efx_mcdi_rpc(efx, MC_CMD_FILTER_OP,
3220 inbuf, sizeof(inbuf),
3222 spin_lock_bh(&efx->filter_lock);
3230 priv_flags &= ~EFX_EF10_FILTER_FLAG_BUSY;
3232 efx_ef10_filter_set_entry(table, i, saved_spec,
3237 /* If successful, return the inserted filter ID */
3239 rc = match_pri * HUNT_FILTER_TBL_ROWS + ins_index;
3241 wake_up_all(&table->waitq);
3243 spin_unlock_bh(&efx->filter_lock);
3244 finish_wait(&table->waitq, &wait);
3248 static void efx_ef10_filter_update_rx_scatter(struct efx_nic *efx)
3250 /* no need to do anything here on EF10 */
3254 * If !by_index, remove by ID
3255 * If by_index, remove by index
3256 * Filter ID may come from userland and must be range-checked.
3258 static int efx_ef10_filter_remove_internal(struct efx_nic *efx,
3259 unsigned int priority_mask,
3260 u32 filter_id, bool by_index)
3262 unsigned int filter_idx = filter_id % HUNT_FILTER_TBL_ROWS;
3263 struct efx_ef10_filter_table *table = efx->filter_state;
3264 MCDI_DECLARE_BUF(inbuf,
3265 MC_CMD_FILTER_OP_IN_HANDLE_OFST +
3266 MC_CMD_FILTER_OP_IN_HANDLE_LEN);
3267 struct efx_filter_spec *spec;
3271 /* Find the software table entry and mark it busy. Don't
3272 * remove it yet; any attempt to update while we're waiting
3273 * for the firmware must find the busy entry.
3276 spin_lock_bh(&efx->filter_lock);
3277 if (!(table->entry[filter_idx].spec &
3278 EFX_EF10_FILTER_FLAG_BUSY))
3280 prepare_to_wait(&table->waitq, &wait, TASK_UNINTERRUPTIBLE);
3281 spin_unlock_bh(&efx->filter_lock);
3285 spec = efx_ef10_filter_entry_spec(table, filter_idx);
3288 efx_ef10_filter_rx_match_pri(table, spec->match_flags) !=
3289 filter_id / HUNT_FILTER_TBL_ROWS)) {
3294 if (spec->flags & EFX_FILTER_FLAG_RX_OVER_AUTO &&
3295 priority_mask == (1U << EFX_FILTER_PRI_AUTO)) {
3296 /* Just remove flags */
3297 spec->flags &= ~EFX_FILTER_FLAG_RX_OVER_AUTO;
3298 table->entry[filter_idx].spec &= ~EFX_EF10_FILTER_FLAG_AUTO_OLD;
3303 if (!(priority_mask & (1U << spec->priority))) {
3308 table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_BUSY;
3309 spin_unlock_bh(&efx->filter_lock);
3311 if (spec->flags & EFX_FILTER_FLAG_RX_OVER_AUTO) {
3312 /* Reset to an automatic filter */
3314 struct efx_filter_spec new_spec = *spec;
3316 new_spec.priority = EFX_FILTER_PRI_AUTO;
3317 new_spec.flags = (EFX_FILTER_FLAG_RX |
3318 (efx_rss_enabled(efx) ?
3319 EFX_FILTER_FLAG_RX_RSS : 0));
3320 new_spec.dmaq_id = 0;
3321 new_spec.rss_context = EFX_FILTER_RSS_CONTEXT_DEFAULT;
3322 rc = efx_ef10_filter_push(efx, &new_spec,
3323 &table->entry[filter_idx].handle,
3326 spin_lock_bh(&efx->filter_lock);
3330 /* Really remove the filter */
3332 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
3333 efx_ef10_filter_is_exclusive(spec) ?
3334 MC_CMD_FILTER_OP_IN_OP_REMOVE :
3335 MC_CMD_FILTER_OP_IN_OP_UNSUBSCRIBE);
3336 MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
3337 table->entry[filter_idx].handle);
3338 rc = efx_mcdi_rpc(efx, MC_CMD_FILTER_OP,
3339 inbuf, sizeof(inbuf), NULL, 0, NULL);
3341 spin_lock_bh(&efx->filter_lock);
3344 efx_ef10_filter_set_entry(table, filter_idx, NULL, 0);
3348 table->entry[filter_idx].spec &= ~EFX_EF10_FILTER_FLAG_BUSY;
3349 wake_up_all(&table->waitq);
3351 spin_unlock_bh(&efx->filter_lock);
3352 finish_wait(&table->waitq, &wait);
3356 static int efx_ef10_filter_remove_safe(struct efx_nic *efx,
3357 enum efx_filter_priority priority,
3360 return efx_ef10_filter_remove_internal(efx, 1U << priority,
3364 static u32 efx_ef10_filter_get_unsafe_id(struct efx_nic *efx, u32 filter_id)
3366 return filter_id % HUNT_FILTER_TBL_ROWS;
3369 static int efx_ef10_filter_remove_unsafe(struct efx_nic *efx,
3370 enum efx_filter_priority priority,
3373 return efx_ef10_filter_remove_internal(efx, 1U << priority,
3377 static int efx_ef10_filter_get_safe(struct efx_nic *efx,
3378 enum efx_filter_priority priority,
3379 u32 filter_id, struct efx_filter_spec *spec)
3381 unsigned int filter_idx = filter_id % HUNT_FILTER_TBL_ROWS;
3382 struct efx_ef10_filter_table *table = efx->filter_state;
3383 const struct efx_filter_spec *saved_spec;
3386 spin_lock_bh(&efx->filter_lock);
3387 saved_spec = efx_ef10_filter_entry_spec(table, filter_idx);
3388 if (saved_spec && saved_spec->priority == priority &&
3389 efx_ef10_filter_rx_match_pri(table, saved_spec->match_flags) ==
3390 filter_id / HUNT_FILTER_TBL_ROWS) {
3391 *spec = *saved_spec;
3396 spin_unlock_bh(&efx->filter_lock);
3400 static int efx_ef10_filter_clear_rx(struct efx_nic *efx,
3401 enum efx_filter_priority priority)
3403 unsigned int priority_mask;
3407 priority_mask = (((1U << (priority + 1)) - 1) &
3408 ~(1U << EFX_FILTER_PRI_AUTO));
3410 for (i = 0; i < HUNT_FILTER_TBL_ROWS; i++) {
3411 rc = efx_ef10_filter_remove_internal(efx, priority_mask,
3413 if (rc && rc != -ENOENT)
3420 static u32 efx_ef10_filter_count_rx_used(struct efx_nic *efx,
3421 enum efx_filter_priority priority)
3423 struct efx_ef10_filter_table *table = efx->filter_state;
3424 unsigned int filter_idx;
3427 spin_lock_bh(&efx->filter_lock);
3428 for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
3429 if (table->entry[filter_idx].spec &&
3430 efx_ef10_filter_entry_spec(table, filter_idx)->priority ==
3434 spin_unlock_bh(&efx->filter_lock);
3438 static u32 efx_ef10_filter_get_rx_id_limit(struct efx_nic *efx)
3440 struct efx_ef10_filter_table *table = efx->filter_state;
3442 return table->rx_match_count * HUNT_FILTER_TBL_ROWS;
3445 static s32 efx_ef10_filter_get_rx_ids(struct efx_nic *efx,
3446 enum efx_filter_priority priority,
3449 struct efx_ef10_filter_table *table = efx->filter_state;
3450 struct efx_filter_spec *spec;
3451 unsigned int filter_idx;
3454 spin_lock_bh(&efx->filter_lock);
3455 for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
3456 spec = efx_ef10_filter_entry_spec(table, filter_idx);
3457 if (spec && spec->priority == priority) {
3458 if (count == size) {
3462 buf[count++] = (efx_ef10_filter_rx_match_pri(
3463 table, spec->match_flags) *
3464 HUNT_FILTER_TBL_ROWS +
3468 spin_unlock_bh(&efx->filter_lock);
3472 #ifdef CONFIG_RFS_ACCEL
3474 static efx_mcdi_async_completer efx_ef10_filter_rfs_insert_complete;
3476 static s32 efx_ef10_filter_rfs_insert(struct efx_nic *efx,
3477 struct efx_filter_spec *spec)
3479 struct efx_ef10_filter_table *table = efx->filter_state;
3480 MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_IN_LEN);
3481 struct efx_filter_spec *saved_spec;
3482 unsigned int hash, i, depth = 1;
3483 bool replacing = false;
3488 /* Must be an RX filter without RSS and not for a multicast
3489 * destination address (RFS only works for connected sockets).
3490 * These restrictions allow us to pass only a tiny amount of
3491 * data through to the completion function.
3493 EFX_WARN_ON_PARANOID(spec->flags !=
3494 (EFX_FILTER_FLAG_RX | EFX_FILTER_FLAG_RX_SCATTER));
3495 EFX_WARN_ON_PARANOID(spec->priority != EFX_FILTER_PRI_HINT);
3496 EFX_WARN_ON_PARANOID(efx_filter_is_mc_recipient(spec));
3498 hash = efx_ef10_filter_hash(spec);
3500 spin_lock_bh(&efx->filter_lock);
3502 /* Find any existing filter with the same match tuple or else
3503 * a free slot to insert at. If an existing filter is busy,
3504 * we have to give up.
3507 i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
3508 saved_spec = efx_ef10_filter_entry_spec(table, i);
3513 } else if (efx_ef10_filter_equal(spec, saved_spec)) {
3514 if (table->entry[i].spec & EFX_EF10_FILTER_FLAG_BUSY) {
3518 if (spec->priority < saved_spec->priority) {
3526 /* Once we reach the maximum search depth, use the
3527 * first suitable slot or return -EBUSY if there was
3530 if (depth == EFX_EF10_FILTER_SEARCH_LIMIT) {
3531 if (ins_index < 0) {
3541 /* Create a software table entry if necessary, and mark it
3542 * busy. We might yet fail to insert, but any attempt to
3543 * insert a conflicting filter while we're waiting for the
3544 * firmware must find the busy entry.
3546 saved_spec = efx_ef10_filter_entry_spec(table, ins_index);
3550 saved_spec = kmalloc(sizeof(*spec), GFP_ATOMIC);
3555 *saved_spec = *spec;
3557 efx_ef10_filter_set_entry(table, ins_index, saved_spec,
3558 EFX_EF10_FILTER_FLAG_BUSY);
3560 spin_unlock_bh(&efx->filter_lock);
3562 /* Pack up the variables needed on completion */
3563 cookie = replacing << 31 | ins_index << 16 | spec->dmaq_id;
3565 efx_ef10_filter_push_prep(efx, spec, inbuf,
3566 table->entry[ins_index].handle, replacing);
3567 efx_mcdi_rpc_async(efx, MC_CMD_FILTER_OP, inbuf, sizeof(inbuf),
3568 MC_CMD_FILTER_OP_OUT_LEN,
3569 efx_ef10_filter_rfs_insert_complete, cookie);
3574 spin_unlock_bh(&efx->filter_lock);
3579 efx_ef10_filter_rfs_insert_complete(struct efx_nic *efx, unsigned long cookie,
3580 int rc, efx_dword_t *outbuf,
3581 size_t outlen_actual)
3583 struct efx_ef10_filter_table *table = efx->filter_state;
3584 unsigned int ins_index, dmaq_id;
3585 struct efx_filter_spec *spec;
3588 /* Unpack the cookie */
3589 replacing = cookie >> 31;
3590 ins_index = (cookie >> 16) & (HUNT_FILTER_TBL_ROWS - 1);
3591 dmaq_id = cookie & 0xffff;
3593 spin_lock_bh(&efx->filter_lock);
3594 spec = efx_ef10_filter_entry_spec(table, ins_index);
3596 table->entry[ins_index].handle =
3597 MCDI_QWORD(outbuf, FILTER_OP_OUT_HANDLE);
3599 spec->dmaq_id = dmaq_id;
3600 } else if (!replacing) {
3604 efx_ef10_filter_set_entry(table, ins_index, spec, 0);
3605 spin_unlock_bh(&efx->filter_lock);
3607 wake_up_all(&table->waitq);
3611 efx_ef10_filter_rfs_expire_complete(struct efx_nic *efx,
3612 unsigned long filter_idx,
3613 int rc, efx_dword_t *outbuf,
3614 size_t outlen_actual);
3616 static bool efx_ef10_filter_rfs_expire_one(struct efx_nic *efx, u32 flow_id,
3617 unsigned int filter_idx)
3619 struct efx_ef10_filter_table *table = efx->filter_state;
3620 struct efx_filter_spec *spec =
3621 efx_ef10_filter_entry_spec(table, filter_idx);
3622 MCDI_DECLARE_BUF(inbuf,
3623 MC_CMD_FILTER_OP_IN_HANDLE_OFST +
3624 MC_CMD_FILTER_OP_IN_HANDLE_LEN);
3627 (table->entry[filter_idx].spec & EFX_EF10_FILTER_FLAG_BUSY) ||
3628 spec->priority != EFX_FILTER_PRI_HINT ||
3629 !rps_may_expire_flow(efx->net_dev, spec->dmaq_id,
3630 flow_id, filter_idx))
3633 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
3634 MC_CMD_FILTER_OP_IN_OP_REMOVE);
3635 MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
3636 table->entry[filter_idx].handle);
3637 if (efx_mcdi_rpc_async(efx, MC_CMD_FILTER_OP, inbuf, sizeof(inbuf), 0,
3638 efx_ef10_filter_rfs_expire_complete, filter_idx))
3641 table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_BUSY;
3646 efx_ef10_filter_rfs_expire_complete(struct efx_nic *efx,
3647 unsigned long filter_idx,
3648 int rc, efx_dword_t *outbuf,
3649 size_t outlen_actual)
3651 struct efx_ef10_filter_table *table = efx->filter_state;
3652 struct efx_filter_spec *spec =
3653 efx_ef10_filter_entry_spec(table, filter_idx);
3655 spin_lock_bh(&efx->filter_lock);
3658 efx_ef10_filter_set_entry(table, filter_idx, NULL, 0);
3660 table->entry[filter_idx].spec &= ~EFX_EF10_FILTER_FLAG_BUSY;
3661 wake_up_all(&table->waitq);
3662 spin_unlock_bh(&efx->filter_lock);
3665 #endif /* CONFIG_RFS_ACCEL */
3667 static int efx_ef10_filter_match_flags_from_mcdi(u32 mcdi_flags)
3669 int match_flags = 0;
3671 #define MAP_FLAG(gen_flag, mcdi_field) { \
3672 u32 old_mcdi_flags = mcdi_flags; \
3673 mcdi_flags &= ~(1 << MC_CMD_FILTER_OP_IN_MATCH_ ## \
3674 mcdi_field ## _LBN); \
3675 if (mcdi_flags != old_mcdi_flags) \
3676 match_flags |= EFX_FILTER_MATCH_ ## gen_flag; \
3678 MAP_FLAG(LOC_MAC_IG, UNKNOWN_UCAST_DST);
3679 MAP_FLAG(LOC_MAC_IG, UNKNOWN_MCAST_DST);
3680 MAP_FLAG(REM_HOST, SRC_IP);
3681 MAP_FLAG(LOC_HOST, DST_IP);
3682 MAP_FLAG(REM_MAC, SRC_MAC);
3683 MAP_FLAG(REM_PORT, SRC_PORT);
3684 MAP_FLAG(LOC_MAC, DST_MAC);
3685 MAP_FLAG(LOC_PORT, DST_PORT);
3686 MAP_FLAG(ETHER_TYPE, ETHER_TYPE);
3687 MAP_FLAG(INNER_VID, INNER_VLAN);
3688 MAP_FLAG(OUTER_VID, OUTER_VLAN);
3689 MAP_FLAG(IP_PROTO, IP_PROTO);
3692 /* Did we map them all? */
3699 static int efx_ef10_filter_table_probe(struct efx_nic *efx)
3701 MCDI_DECLARE_BUF(inbuf, MC_CMD_GET_PARSER_DISP_INFO_IN_LEN);
3702 MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_PARSER_DISP_INFO_OUT_LENMAX);
3703 unsigned int pd_match_pri, pd_match_count;
3704 struct efx_ef10_filter_table *table;
3708 table = kzalloc(sizeof(*table), GFP_KERNEL);
3712 /* Find out which RX filter types are supported, and their priorities */
3713 MCDI_SET_DWORD(inbuf, GET_PARSER_DISP_INFO_IN_OP,
3714 MC_CMD_GET_PARSER_DISP_INFO_IN_OP_GET_SUPPORTED_RX_MATCHES);
3715 rc = efx_mcdi_rpc(efx, MC_CMD_GET_PARSER_DISP_INFO,
3716 inbuf, sizeof(inbuf), outbuf, sizeof(outbuf),
3720 pd_match_count = MCDI_VAR_ARRAY_LEN(
3721 outlen, GET_PARSER_DISP_INFO_OUT_SUPPORTED_MATCHES);
3722 table->rx_match_count = 0;
3724 for (pd_match_pri = 0; pd_match_pri < pd_match_count; pd_match_pri++) {
3728 GET_PARSER_DISP_INFO_OUT_SUPPORTED_MATCHES,
3730 rc = efx_ef10_filter_match_flags_from_mcdi(mcdi_flags);
3732 netif_dbg(efx, probe, efx->net_dev,
3733 "%s: fw flags %#x pri %u not supported in driver\n",
3734 __func__, mcdi_flags, pd_match_pri);
3736 netif_dbg(efx, probe, efx->net_dev,
3737 "%s: fw flags %#x pri %u supported as driver flags %#x pri %u\n",
3738 __func__, mcdi_flags, pd_match_pri,
3739 rc, table->rx_match_count);
3740 table->rx_match_flags[table->rx_match_count++] = rc;
3744 table->entry = vzalloc(HUNT_FILTER_TBL_ROWS * sizeof(*table->entry));
3745 if (!table->entry) {
3750 table->ucdef_id = EFX_EF10_FILTER_ID_INVALID;
3751 table->bcast_id = EFX_EF10_FILTER_ID_INVALID;
3752 table->mcdef_id = EFX_EF10_FILTER_ID_INVALID;
3754 efx->filter_state = table;
3755 init_waitqueue_head(&table->waitq);
3763 /* Caller must hold efx->filter_sem for read if race against
3764 * efx_ef10_filter_table_remove() is possible
3766 static void efx_ef10_filter_table_restore(struct efx_nic *efx)
3768 struct efx_ef10_filter_table *table = efx->filter_state;
3769 struct efx_ef10_nic_data *nic_data = efx->nic_data;
3770 struct efx_filter_spec *spec;
3771 unsigned int filter_idx;
3772 bool failed = false;
3775 WARN_ON(!rwsem_is_locked(&efx->filter_sem));
3777 if (!nic_data->must_restore_filters)
3783 spin_lock_bh(&efx->filter_lock);
3785 for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
3786 spec = efx_ef10_filter_entry_spec(table, filter_idx);
3790 table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_BUSY;
3791 spin_unlock_bh(&efx->filter_lock);
3793 rc = efx_ef10_filter_push(efx, spec,
3794 &table->entry[filter_idx].handle,
3799 spin_lock_bh(&efx->filter_lock);
3802 efx_ef10_filter_set_entry(table, filter_idx, NULL, 0);
3804 table->entry[filter_idx].spec &=
3805 ~EFX_EF10_FILTER_FLAG_BUSY;
3809 spin_unlock_bh(&efx->filter_lock);
3812 netif_err(efx, hw, efx->net_dev,
3813 "unable to restore all filters\n");
3815 nic_data->must_restore_filters = false;
3818 /* Caller must hold efx->filter_sem for write */
3819 static void efx_ef10_filter_table_remove(struct efx_nic *efx)
3821 struct efx_ef10_filter_table *table = efx->filter_state;
3822 MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_IN_LEN);
3823 struct efx_filter_spec *spec;
3824 unsigned int filter_idx;
3827 efx->filter_state = NULL;
3831 for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
3832 spec = efx_ef10_filter_entry_spec(table, filter_idx);
3836 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
3837 efx_ef10_filter_is_exclusive(spec) ?
3838 MC_CMD_FILTER_OP_IN_OP_REMOVE :
3839 MC_CMD_FILTER_OP_IN_OP_UNSUBSCRIBE);
3840 MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
3841 table->entry[filter_idx].handle);
3842 rc = efx_mcdi_rpc(efx, MC_CMD_FILTER_OP, inbuf, sizeof(inbuf),
3845 netdev_WARN(efx->net_dev,
3846 "filter_idx=%#x handle=%#llx\n",
3848 table->entry[filter_idx].handle);
3852 vfree(table->entry);
3856 #define EFX_EF10_FILTER_DO_MARK_OLD(id) \
3857 if (id != EFX_EF10_FILTER_ID_INVALID) { \
3858 filter_idx = efx_ef10_filter_get_unsafe_id(efx, id); \
3859 WARN_ON(!table->entry[filter_idx].spec); \
3860 table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_AUTO_OLD; \
3862 static void efx_ef10_filter_mark_old(struct efx_nic *efx)
3864 struct efx_ef10_filter_table *table = efx->filter_state;
3865 unsigned int filter_idx, i;
3870 /* Mark old filters that may need to be removed */
3871 spin_lock_bh(&efx->filter_lock);
3872 for (i = 0; i < table->dev_uc_count; i++)
3873 EFX_EF10_FILTER_DO_MARK_OLD(table->dev_uc_list[i].id);
3874 for (i = 0; i < table->dev_mc_count; i++)
3875 EFX_EF10_FILTER_DO_MARK_OLD(table->dev_mc_list[i].id);
3876 EFX_EF10_FILTER_DO_MARK_OLD(table->ucdef_id);
3877 EFX_EF10_FILTER_DO_MARK_OLD(table->bcast_id);
3878 EFX_EF10_FILTER_DO_MARK_OLD(table->mcdef_id);
3879 spin_unlock_bh(&efx->filter_lock);
3881 #undef EFX_EF10_FILTER_DO_MARK_OLD
3883 static void efx_ef10_filter_uc_addr_list(struct efx_nic *efx, bool *promisc)
3885 struct efx_ef10_filter_table *table = efx->filter_state;
3886 struct net_device *net_dev = efx->net_dev;
3887 struct netdev_hw_addr *uc;
3891 table->ucdef_id = EFX_EF10_FILTER_ID_INVALID;
3892 addr_count = netdev_uc_count(net_dev);
3893 if (net_dev->flags & IFF_PROMISC)
3895 table->dev_uc_count = 1 + addr_count;
3896 ether_addr_copy(table->dev_uc_list[0].addr, net_dev->dev_addr);
3898 netdev_for_each_uc_addr(uc, net_dev) {
3899 if (i >= EFX_EF10_FILTER_DEV_UC_MAX) {
3903 ether_addr_copy(table->dev_uc_list[i].addr, uc->addr);
3904 table->dev_uc_list[i].id = EFX_EF10_FILTER_ID_INVALID;
3909 static void efx_ef10_filter_mc_addr_list(struct efx_nic *efx, bool *promisc)
3911 struct efx_ef10_filter_table *table = efx->filter_state;
3912 struct net_device *net_dev = efx->net_dev;
3913 struct netdev_hw_addr *mc;
3914 unsigned int i, addr_count;
3916 table->mcdef_id = EFX_EF10_FILTER_ID_INVALID;
3917 table->bcast_id = EFX_EF10_FILTER_ID_INVALID;
3918 if (net_dev->flags & (IFF_PROMISC | IFF_ALLMULTI))
3921 addr_count = netdev_mc_count(net_dev);
3923 netdev_for_each_mc_addr(mc, net_dev) {
3924 if (i >= EFX_EF10_FILTER_DEV_MC_MAX) {
3928 ether_addr_copy(table->dev_mc_list[i].addr, mc->addr);
3929 table->dev_mc_list[i].id = EFX_EF10_FILTER_ID_INVALID;
3933 table->dev_mc_count = i;
3936 static int efx_ef10_filter_insert_addr_list(struct efx_nic *efx,
3937 bool multicast, bool rollback)
3939 struct efx_ef10_filter_table *table = efx->filter_state;
3940 struct efx_ef10_dev_addr *addr_list;
3941 enum efx_filter_flags filter_flags;
3942 struct efx_filter_spec spec;
3949 addr_list = table->dev_mc_list;
3950 addr_count = table->dev_mc_count;
3952 addr_list = table->dev_uc_list;
3953 addr_count = table->dev_uc_count;
3956 filter_flags = efx_rss_enabled(efx) ? EFX_FILTER_FLAG_RX_RSS : 0;
3958 /* Insert/renew filters */
3959 for (i = 0; i < addr_count; i++) {
3960 efx_filter_init_rx(&spec, EFX_FILTER_PRI_AUTO, filter_flags, 0);
3961 efx_filter_set_eth_local(&spec, EFX_FILTER_VID_UNSPEC,
3963 rc = efx_ef10_filter_insert(efx, &spec, true);
3966 netif_info(efx, drv, efx->net_dev,
3967 "efx_ef10_filter_insert failed rc=%d\n",
3969 /* Fall back to promiscuous */
3970 for (j = 0; j < i; j++) {
3971 if (addr_list[j].id == EFX_EF10_FILTER_ID_INVALID)
3973 efx_ef10_filter_remove_unsafe(
3974 efx, EFX_FILTER_PRI_AUTO,
3976 addr_list[j].id = EFX_EF10_FILTER_ID_INVALID;
3980 /* mark as not inserted, and carry on */
3981 rc = EFX_EF10_FILTER_ID_INVALID;
3984 addr_list[i].id = efx_ef10_filter_get_unsafe_id(efx, rc);
3987 if (multicast && rollback) {
3988 /* Also need an Ethernet broadcast filter */
3989 efx_filter_init_rx(&spec, EFX_FILTER_PRI_AUTO, filter_flags, 0);
3990 eth_broadcast_addr(baddr);
3991 efx_filter_set_eth_local(&spec, EFX_FILTER_VID_UNSPEC, baddr);
3992 rc = efx_ef10_filter_insert(efx, &spec, true);
3994 netif_warn(efx, drv, efx->net_dev,
3995 "Broadcast filter insert failed rc=%d\n", rc);
3996 /* Fall back to promiscuous */
3997 for (j = 0; j < i; j++) {
3998 if (addr_list[j].id == EFX_EF10_FILTER_ID_INVALID)
4000 efx_ef10_filter_remove_unsafe(
4001 efx, EFX_FILTER_PRI_AUTO,
4003 addr_list[j].id = EFX_EF10_FILTER_ID_INVALID;
4007 table->bcast_id = efx_ef10_filter_get_unsafe_id(efx, rc);
4014 static int efx_ef10_filter_insert_def(struct efx_nic *efx, bool multicast,
4017 struct efx_ef10_filter_table *table = efx->filter_state;
4018 struct efx_ef10_nic_data *nic_data = efx->nic_data;
4019 enum efx_filter_flags filter_flags;
4020 struct efx_filter_spec spec;
4024 filter_flags = efx_rss_enabled(efx) ? EFX_FILTER_FLAG_RX_RSS : 0;
4026 efx_filter_init_rx(&spec, EFX_FILTER_PRI_AUTO, filter_flags, 0);
4029 efx_filter_set_mc_def(&spec);
4031 efx_filter_set_uc_def(&spec);
4033 rc = efx_ef10_filter_insert(efx, &spec, true);
4035 netif_warn(efx, drv, efx->net_dev,
4036 "%scast mismatch filter insert failed rc=%d\n",
4037 multicast ? "Multi" : "Uni", rc);
4038 } else if (multicast) {
4039 table->mcdef_id = efx_ef10_filter_get_unsafe_id(efx, rc);
4040 if (!nic_data->workaround_26807) {
4041 /* Also need an Ethernet broadcast filter */
4042 efx_filter_init_rx(&spec, EFX_FILTER_PRI_AUTO,
4044 eth_broadcast_addr(baddr);
4045 efx_filter_set_eth_local(&spec, EFX_FILTER_VID_UNSPEC,
4047 rc = efx_ef10_filter_insert(efx, &spec, true);
4049 netif_warn(efx, drv, efx->net_dev,
4050 "Broadcast filter insert failed rc=%d\n",
4053 /* Roll back the mc_def filter */
4054 efx_ef10_filter_remove_unsafe(
4055 efx, EFX_FILTER_PRI_AUTO,
4057 table->mcdef_id = EFX_EF10_FILTER_ID_INVALID;
4061 table->bcast_id = efx_ef10_filter_get_unsafe_id(efx, rc);
4066 table->ucdef_id = rc;
4072 /* Remove filters that weren't renewed. Since nothing else changes the AUTO_OLD
4073 * flag or removes these filters, we don't need to hold the filter_lock while
4074 * scanning for these filters.
4076 static void efx_ef10_filter_remove_old(struct efx_nic *efx)
4078 struct efx_ef10_filter_table *table = efx->filter_state;
4079 bool remove_failed = false;
4082 for (i = 0; i < HUNT_FILTER_TBL_ROWS; i++) {
4083 if (ACCESS_ONCE(table->entry[i].spec) &
4084 EFX_EF10_FILTER_FLAG_AUTO_OLD) {
4085 if (efx_ef10_filter_remove_internal(
4086 efx, 1U << EFX_FILTER_PRI_AUTO,
4088 remove_failed = true;
4091 WARN_ON(remove_failed);
4094 static int efx_ef10_vport_set_mac_address(struct efx_nic *efx)
4096 struct efx_ef10_nic_data *nic_data = efx->nic_data;
4097 u8 mac_old[ETH_ALEN];
4100 /* Only reconfigure a PF-created vport */
4101 if (is_zero_ether_addr(nic_data->vport_mac))
4104 efx_device_detach_sync(efx);
4105 efx_net_stop(efx->net_dev);
4106 down_write(&efx->filter_sem);
4107 efx_ef10_filter_table_remove(efx);
4108 up_write(&efx->filter_sem);
4110 rc = efx_ef10_vadaptor_free(efx, nic_data->vport_id);
4112 goto restore_filters;
4114 ether_addr_copy(mac_old, nic_data->vport_mac);
4115 rc = efx_ef10_vport_del_mac(efx, nic_data->vport_id,
4116 nic_data->vport_mac);
4118 goto restore_vadaptor;
4120 rc = efx_ef10_vport_add_mac(efx, nic_data->vport_id,
4121 efx->net_dev->dev_addr);
4123 ether_addr_copy(nic_data->vport_mac, efx->net_dev->dev_addr);
4125 rc2 = efx_ef10_vport_add_mac(efx, nic_data->vport_id, mac_old);
4127 /* Failed to add original MAC, so clear vport_mac */
4128 eth_zero_addr(nic_data->vport_mac);
4134 rc2 = efx_ef10_vadaptor_alloc(efx, nic_data->vport_id);
4138 down_write(&efx->filter_sem);
4139 rc2 = efx_ef10_filter_table_probe(efx);
4140 up_write(&efx->filter_sem);
4144 rc2 = efx_net_open(efx->net_dev);
4148 netif_device_attach(efx->net_dev);
4153 netif_err(efx, drv, efx->net_dev,
4154 "Failed to restore when changing MAC address - scheduling reset\n");
4155 efx_schedule_reset(efx, RESET_TYPE_DATAPATH);
4157 return rc ? rc : rc2;
4160 /* Caller must hold efx->filter_sem for read if race against
4161 * efx_ef10_filter_table_remove() is possible
4163 static void efx_ef10_filter_sync_rx_mode(struct efx_nic *efx)
4165 struct efx_ef10_filter_table *table = efx->filter_state;
4166 struct efx_ef10_nic_data *nic_data = efx->nic_data;
4167 struct net_device *net_dev = efx->net_dev;
4168 bool uc_promisc = false, mc_promisc = false;
4170 if (!efx_dev_registered(efx))
4176 efx_ef10_filter_mark_old(efx);
4178 /* Copy/convert the address lists; add the primary station
4179 * address and broadcast address
4181 netif_addr_lock_bh(net_dev);
4182 efx_ef10_filter_uc_addr_list(efx, &uc_promisc);
4183 efx_ef10_filter_mc_addr_list(efx, &mc_promisc);
4184 netif_addr_unlock_bh(net_dev);
4186 /* Insert/renew unicast filters */
4188 efx_ef10_filter_insert_def(efx, false, false);
4189 efx_ef10_filter_insert_addr_list(efx, false, false);
4191 /* If any of the filters failed to insert, fall back to
4192 * promiscuous mode - add in the uc_def filter. But keep
4193 * our individual unicast filters.
4195 if (efx_ef10_filter_insert_addr_list(efx, false, false))
4196 efx_ef10_filter_insert_def(efx, false, false);
4199 /* Insert/renew multicast filters */
4200 /* If changing promiscuous state with cascaded multicast filters, remove
4201 * old filters first, so that packets are dropped rather than duplicated
4203 if (nic_data->workaround_26807 && efx->mc_promisc != mc_promisc)
4204 efx_ef10_filter_remove_old(efx);
4206 if (nic_data->workaround_26807) {
4207 /* If we failed to insert promiscuous filters, rollback
4208 * and fall back to individual multicast filters
4210 if (efx_ef10_filter_insert_def(efx, true, true)) {
4211 /* Changing promisc state, so remove old filters */
4212 efx_ef10_filter_remove_old(efx);
4213 efx_ef10_filter_insert_addr_list(efx, true, false);
4216 /* If we failed to insert promiscuous filters, don't
4217 * rollback. Regardless, also insert the mc_list
4219 efx_ef10_filter_insert_def(efx, true, false);
4220 efx_ef10_filter_insert_addr_list(efx, true, false);
4223 /* If any filters failed to insert, rollback and fall back to
4224 * promiscuous mode - mc_def filter and maybe broadcast. If
4225 * that fails, roll back again and insert as many of our
4226 * individual multicast filters as we can.
4228 if (efx_ef10_filter_insert_addr_list(efx, true, true)) {
4229 /* Changing promisc state, so remove old filters */
4230 if (nic_data->workaround_26807)
4231 efx_ef10_filter_remove_old(efx);
4232 if (efx_ef10_filter_insert_def(efx, true, true))
4233 efx_ef10_filter_insert_addr_list(efx, true, false);
4237 efx_ef10_filter_remove_old(efx);
4238 efx->mc_promisc = mc_promisc;
4241 static int efx_ef10_set_mac_address(struct efx_nic *efx)
4243 MCDI_DECLARE_BUF(inbuf, MC_CMD_VADAPTOR_SET_MAC_IN_LEN);
4244 struct efx_ef10_nic_data *nic_data = efx->nic_data;
4245 bool was_enabled = efx->port_enabled;
4248 efx_device_detach_sync(efx);
4249 efx_net_stop(efx->net_dev);
4250 down_write(&efx->filter_sem);
4251 efx_ef10_filter_table_remove(efx);
4253 ether_addr_copy(MCDI_PTR(inbuf, VADAPTOR_SET_MAC_IN_MACADDR),
4254 efx->net_dev->dev_addr);
4255 MCDI_SET_DWORD(inbuf, VADAPTOR_SET_MAC_IN_UPSTREAM_PORT_ID,
4256 nic_data->vport_id);
4257 rc = efx_mcdi_rpc_quiet(efx, MC_CMD_VADAPTOR_SET_MAC, inbuf,
4258 sizeof(inbuf), NULL, 0, NULL);
4260 efx_ef10_filter_table_probe(efx);
4261 up_write(&efx->filter_sem);
4263 efx_net_open(efx->net_dev);
4264 netif_device_attach(efx->net_dev);
4266 #ifdef CONFIG_SFC_SRIOV
4267 if (efx->pci_dev->is_virtfn && efx->pci_dev->physfn) {
4268 struct pci_dev *pci_dev_pf = efx->pci_dev->physfn;
4271 struct efx_nic *efx_pf;
4273 /* Switch to PF and change MAC address on vport */
4274 efx_pf = pci_get_drvdata(pci_dev_pf);
4276 rc = efx_ef10_sriov_set_vf_mac(efx_pf,
4278 efx->net_dev->dev_addr);
4280 struct efx_nic *efx_pf = pci_get_drvdata(pci_dev_pf);
4281 struct efx_ef10_nic_data *nic_data = efx_pf->nic_data;
4284 /* MAC address successfully changed by VF (with MAC
4285 * spoofing) so update the parent PF if possible.
4287 for (i = 0; i < efx_pf->vf_count; ++i) {
4288 struct ef10_vf *vf = nic_data->vf + i;
4290 if (vf->efx == efx) {
4291 ether_addr_copy(vf->mac,
4292 efx->net_dev->dev_addr);
4300 netif_err(efx, drv, efx->net_dev,
4301 "Cannot change MAC address; use sfboot to enable"
4302 " mac-spoofing on this interface\n");
4303 } else if (rc == -ENOSYS && !efx_ef10_is_vf(efx)) {
4304 /* If the active MCFW does not support MC_CMD_VADAPTOR_SET_MAC
4305 * fall-back to the method of changing the MAC address on the
4306 * vport. This only applies to PFs because such versions of
4307 * MCFW do not support VFs.
4309 rc = efx_ef10_vport_set_mac_address(efx);
4311 efx_mcdi_display_error(efx, MC_CMD_VADAPTOR_SET_MAC,
4312 sizeof(inbuf), NULL, 0, rc);
4318 static int efx_ef10_mac_reconfigure(struct efx_nic *efx)
4320 efx_ef10_filter_sync_rx_mode(efx);
4322 return efx_mcdi_set_mac(efx);
4325 static int efx_ef10_mac_reconfigure_vf(struct efx_nic *efx)
4327 efx_ef10_filter_sync_rx_mode(efx);
4332 static int efx_ef10_start_bist(struct efx_nic *efx, u32 bist_type)
4334 MCDI_DECLARE_BUF(inbuf, MC_CMD_START_BIST_IN_LEN);
4336 MCDI_SET_DWORD(inbuf, START_BIST_IN_TYPE, bist_type);
4337 return efx_mcdi_rpc(efx, MC_CMD_START_BIST, inbuf, sizeof(inbuf),
4341 /* MC BISTs follow a different poll mechanism to phy BISTs.
4342 * The BIST is done in the poll handler on the MC, and the MCDI command
4343 * will block until the BIST is done.
4345 static int efx_ef10_poll_bist(struct efx_nic *efx)
4348 MCDI_DECLARE_BUF(outbuf, MC_CMD_POLL_BIST_OUT_LEN);
4352 rc = efx_mcdi_rpc(efx, MC_CMD_POLL_BIST, NULL, 0,
4353 outbuf, sizeof(outbuf), &outlen);
4357 if (outlen < MC_CMD_POLL_BIST_OUT_LEN)
4360 result = MCDI_DWORD(outbuf, POLL_BIST_OUT_RESULT);
4362 case MC_CMD_POLL_BIST_PASSED:
4363 netif_dbg(efx, hw, efx->net_dev, "BIST passed.\n");
4365 case MC_CMD_POLL_BIST_TIMEOUT:
4366 netif_err(efx, hw, efx->net_dev, "BIST timed out\n");
4368 case MC_CMD_POLL_BIST_FAILED:
4369 netif_err(efx, hw, efx->net_dev, "BIST failed.\n");
4372 netif_err(efx, hw, efx->net_dev,
4373 "BIST returned unknown result %u", result);
4378 static int efx_ef10_run_bist(struct efx_nic *efx, u32 bist_type)
4382 netif_dbg(efx, drv, efx->net_dev, "starting BIST type %u\n", bist_type);
4384 rc = efx_ef10_start_bist(efx, bist_type);
4388 return efx_ef10_poll_bist(efx);
4392 efx_ef10_test_chip(struct efx_nic *efx, struct efx_self_tests *tests)
4396 efx_reset_down(efx, RESET_TYPE_WORLD);
4398 rc = efx_mcdi_rpc(efx, MC_CMD_ENABLE_OFFLINE_BIST,
4399 NULL, 0, NULL, 0, NULL);
4403 tests->memory = efx_ef10_run_bist(efx, MC_CMD_MC_MEM_BIST) ? -1 : 1;
4404 tests->registers = efx_ef10_run_bist(efx, MC_CMD_REG_BIST) ? -1 : 1;
4406 rc = efx_mcdi_reset(efx, RESET_TYPE_WORLD);
4411 rc2 = efx_reset_up(efx, RESET_TYPE_WORLD, rc == 0);
4412 return rc ? rc : rc2;
4415 #ifdef CONFIG_SFC_MTD
4417 struct efx_ef10_nvram_type_info {
4418 u16 type, type_mask;
4423 static const struct efx_ef10_nvram_type_info efx_ef10_nvram_types[] = {
4424 { NVRAM_PARTITION_TYPE_MC_FIRMWARE, 0, 0, "sfc_mcfw" },
4425 { NVRAM_PARTITION_TYPE_MC_FIRMWARE_BACKUP, 0, 0, "sfc_mcfw_backup" },
4426 { NVRAM_PARTITION_TYPE_EXPANSION_ROM, 0, 0, "sfc_exp_rom" },
4427 { NVRAM_PARTITION_TYPE_STATIC_CONFIG, 0, 0, "sfc_static_cfg" },
4428 { NVRAM_PARTITION_TYPE_DYNAMIC_CONFIG, 0, 0, "sfc_dynamic_cfg" },
4429 { NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT0, 0, 0, "sfc_exp_rom_cfg" },
4430 { NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT1, 0, 1, "sfc_exp_rom_cfg" },
4431 { NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT2, 0, 2, "sfc_exp_rom_cfg" },
4432 { NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT3, 0, 3, "sfc_exp_rom_cfg" },
4433 { NVRAM_PARTITION_TYPE_LICENSE, 0, 0, "sfc_license" },
4434 { NVRAM_PARTITION_TYPE_PHY_MIN, 0xff, 0, "sfc_phy_fw" },
4437 static int efx_ef10_mtd_probe_partition(struct efx_nic *efx,
4438 struct efx_mcdi_mtd_partition *part,
4441 MCDI_DECLARE_BUF(inbuf, MC_CMD_NVRAM_METADATA_IN_LEN);
4442 MCDI_DECLARE_BUF(outbuf, MC_CMD_NVRAM_METADATA_OUT_LENMAX);
4443 const struct efx_ef10_nvram_type_info *info;
4444 size_t size, erase_size, outlen;
4448 for (info = efx_ef10_nvram_types; ; info++) {
4450 efx_ef10_nvram_types + ARRAY_SIZE(efx_ef10_nvram_types))
4452 if ((type & ~info->type_mask) == info->type)
4455 if (info->port != efx_port_num(efx))
4458 rc = efx_mcdi_nvram_info(efx, type, &size, &erase_size, &protected);
4462 return -ENODEV; /* hide it */
4464 part->nvram_type = type;
4466 MCDI_SET_DWORD(inbuf, NVRAM_METADATA_IN_TYPE, type);
4467 rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_METADATA, inbuf, sizeof(inbuf),
4468 outbuf, sizeof(outbuf), &outlen);
4471 if (outlen < MC_CMD_NVRAM_METADATA_OUT_LENMIN)
4473 if (MCDI_DWORD(outbuf, NVRAM_METADATA_OUT_FLAGS) &
4474 (1 << MC_CMD_NVRAM_METADATA_OUT_SUBTYPE_VALID_LBN))
4475 part->fw_subtype = MCDI_DWORD(outbuf,
4476 NVRAM_METADATA_OUT_SUBTYPE);
4478 part->common.dev_type_name = "EF10 NVRAM manager";
4479 part->common.type_name = info->name;
4481 part->common.mtd.type = MTD_NORFLASH;
4482 part->common.mtd.flags = MTD_CAP_NORFLASH;
4483 part->common.mtd.size = size;
4484 part->common.mtd.erasesize = erase_size;
4489 static int efx_ef10_mtd_probe(struct efx_nic *efx)
4491 MCDI_DECLARE_BUF(outbuf, MC_CMD_NVRAM_PARTITIONS_OUT_LENMAX);
4492 struct efx_mcdi_mtd_partition *parts;
4493 size_t outlen, n_parts_total, i, n_parts;
4499 BUILD_BUG_ON(MC_CMD_NVRAM_PARTITIONS_IN_LEN != 0);
4500 rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_PARTITIONS, NULL, 0,
4501 outbuf, sizeof(outbuf), &outlen);
4504 if (outlen < MC_CMD_NVRAM_PARTITIONS_OUT_LENMIN)
4507 n_parts_total = MCDI_DWORD(outbuf, NVRAM_PARTITIONS_OUT_NUM_PARTITIONS);
4509 MCDI_VAR_ARRAY_LEN(outlen, NVRAM_PARTITIONS_OUT_TYPE_ID))
4512 parts = kcalloc(n_parts_total, sizeof(*parts), GFP_KERNEL);
4517 for (i = 0; i < n_parts_total; i++) {
4518 type = MCDI_ARRAY_DWORD(outbuf, NVRAM_PARTITIONS_OUT_TYPE_ID,
4520 rc = efx_ef10_mtd_probe_partition(efx, &parts[n_parts], type);
4523 else if (rc != -ENODEV)
4527 rc = efx_mtd_add(efx, &parts[0].common, n_parts, sizeof(*parts));
4534 #endif /* CONFIG_SFC_MTD */
4536 static void efx_ef10_ptp_write_host_time(struct efx_nic *efx, u32 host_time)
4538 _efx_writed(efx, cpu_to_le32(host_time), ER_DZ_MC_DB_LWRD);
4541 static void efx_ef10_ptp_write_host_time_vf(struct efx_nic *efx,
4544 static int efx_ef10_rx_enable_timestamping(struct efx_channel *channel,
4547 MCDI_DECLARE_BUF(inbuf, MC_CMD_PTP_IN_TIME_EVENT_SUBSCRIBE_LEN);
4550 if (channel->sync_events_state == SYNC_EVENTS_REQUESTED ||
4551 channel->sync_events_state == SYNC_EVENTS_VALID ||
4552 (temp && channel->sync_events_state == SYNC_EVENTS_DISABLED))
4554 channel->sync_events_state = SYNC_EVENTS_REQUESTED;
4556 MCDI_SET_DWORD(inbuf, PTP_IN_OP, MC_CMD_PTP_OP_TIME_EVENT_SUBSCRIBE);
4557 MCDI_SET_DWORD(inbuf, PTP_IN_PERIPH_ID, 0);
4558 MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_SUBSCRIBE_QUEUE,
4561 rc = efx_mcdi_rpc(channel->efx, MC_CMD_PTP,
4562 inbuf, sizeof(inbuf), NULL, 0, NULL);
4565 channel->sync_events_state = temp ? SYNC_EVENTS_QUIESCENT :
4566 SYNC_EVENTS_DISABLED;
4571 static int efx_ef10_rx_disable_timestamping(struct efx_channel *channel,
4574 MCDI_DECLARE_BUF(inbuf, MC_CMD_PTP_IN_TIME_EVENT_UNSUBSCRIBE_LEN);
4577 if (channel->sync_events_state == SYNC_EVENTS_DISABLED ||
4578 (temp && channel->sync_events_state == SYNC_EVENTS_QUIESCENT))
4580 if (channel->sync_events_state == SYNC_EVENTS_QUIESCENT) {
4581 channel->sync_events_state = SYNC_EVENTS_DISABLED;
4584 channel->sync_events_state = temp ? SYNC_EVENTS_QUIESCENT :
4585 SYNC_EVENTS_DISABLED;
4587 MCDI_SET_DWORD(inbuf, PTP_IN_OP, MC_CMD_PTP_OP_TIME_EVENT_UNSUBSCRIBE);
4588 MCDI_SET_DWORD(inbuf, PTP_IN_PERIPH_ID, 0);
4589 MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_UNSUBSCRIBE_CONTROL,
4590 MC_CMD_PTP_IN_TIME_EVENT_UNSUBSCRIBE_SINGLE);
4591 MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_UNSUBSCRIBE_QUEUE,
4594 rc = efx_mcdi_rpc(channel->efx, MC_CMD_PTP,
4595 inbuf, sizeof(inbuf), NULL, 0, NULL);
4600 static int efx_ef10_ptp_set_ts_sync_events(struct efx_nic *efx, bool en,
4603 int (*set)(struct efx_channel *channel, bool temp);
4604 struct efx_channel *channel;
4607 efx_ef10_rx_enable_timestamping :
4608 efx_ef10_rx_disable_timestamping;
4610 efx_for_each_channel(channel, efx) {
4611 int rc = set(channel, temp);
4612 if (en && rc != 0) {
4613 efx_ef10_ptp_set_ts_sync_events(efx, false, temp);
4621 static int efx_ef10_ptp_set_ts_config_vf(struct efx_nic *efx,
4622 struct hwtstamp_config *init)
4627 static int efx_ef10_ptp_set_ts_config(struct efx_nic *efx,
4628 struct hwtstamp_config *init)
4632 switch (init->rx_filter) {
4633 case HWTSTAMP_FILTER_NONE:
4634 efx_ef10_ptp_set_ts_sync_events(efx, false, false);
4635 /* if TX timestamping is still requested then leave PTP on */
4636 return efx_ptp_change_mode(efx,
4637 init->tx_type != HWTSTAMP_TX_OFF, 0);
4638 case HWTSTAMP_FILTER_ALL:
4639 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
4640 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
4641 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
4642 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
4643 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
4644 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
4645 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
4646 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
4647 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
4648 case HWTSTAMP_FILTER_PTP_V2_EVENT:
4649 case HWTSTAMP_FILTER_PTP_V2_SYNC:
4650 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
4651 init->rx_filter = HWTSTAMP_FILTER_ALL;
4652 rc = efx_ptp_change_mode(efx, true, 0);
4654 rc = efx_ef10_ptp_set_ts_sync_events(efx, true, false);
4656 efx_ptp_change_mode(efx, false, 0);
4663 const struct efx_nic_type efx_hunt_a0_vf_nic_type = {
4665 .mem_bar = EFX_MEM_VF_BAR,
4666 .mem_map_size = efx_ef10_mem_map_size,
4667 .probe = efx_ef10_probe_vf,
4668 .remove = efx_ef10_remove,
4669 .dimension_resources = efx_ef10_dimension_resources,
4670 .init = efx_ef10_init_nic,
4671 .fini = efx_port_dummy_op_void,
4672 .map_reset_reason = efx_ef10_map_reset_reason,
4673 .map_reset_flags = efx_ef10_map_reset_flags,
4674 .reset = efx_ef10_reset,
4675 .probe_port = efx_mcdi_port_probe,
4676 .remove_port = efx_mcdi_port_remove,
4677 .fini_dmaq = efx_ef10_fini_dmaq,
4678 .prepare_flr = efx_ef10_prepare_flr,
4679 .finish_flr = efx_port_dummy_op_void,
4680 .describe_stats = efx_ef10_describe_stats,
4681 .update_stats = efx_ef10_update_stats_vf,
4682 .start_stats = efx_port_dummy_op_void,
4683 .pull_stats = efx_port_dummy_op_void,
4684 .stop_stats = efx_port_dummy_op_void,
4685 .set_id_led = efx_mcdi_set_id_led,
4686 .push_irq_moderation = efx_ef10_push_irq_moderation,
4687 .reconfigure_mac = efx_ef10_mac_reconfigure_vf,
4688 .check_mac_fault = efx_mcdi_mac_check_fault,
4689 .reconfigure_port = efx_mcdi_port_reconfigure,
4690 .get_wol = efx_ef10_get_wol_vf,
4691 .set_wol = efx_ef10_set_wol_vf,
4692 .resume_wol = efx_port_dummy_op_void,
4693 .mcdi_request = efx_ef10_mcdi_request,
4694 .mcdi_poll_response = efx_ef10_mcdi_poll_response,
4695 .mcdi_read_response = efx_ef10_mcdi_read_response,
4696 .mcdi_poll_reboot = efx_ef10_mcdi_poll_reboot,
4697 .mcdi_reboot_detected = efx_ef10_mcdi_reboot_detected,
4698 .irq_enable_master = efx_port_dummy_op_void,
4699 .irq_test_generate = efx_ef10_irq_test_generate,
4700 .irq_disable_non_ev = efx_port_dummy_op_void,
4701 .irq_handle_msi = efx_ef10_msi_interrupt,
4702 .irq_handle_legacy = efx_ef10_legacy_interrupt,
4703 .tx_probe = efx_ef10_tx_probe,
4704 .tx_init = efx_ef10_tx_init,
4705 .tx_remove = efx_ef10_tx_remove,
4706 .tx_write = efx_ef10_tx_write,
4707 .rx_push_rss_config = efx_ef10_vf_rx_push_rss_config,
4708 .rx_probe = efx_ef10_rx_probe,
4709 .rx_init = efx_ef10_rx_init,
4710 .rx_remove = efx_ef10_rx_remove,
4711 .rx_write = efx_ef10_rx_write,
4712 .rx_defer_refill = efx_ef10_rx_defer_refill,
4713 .ev_probe = efx_ef10_ev_probe,
4714 .ev_init = efx_ef10_ev_init,
4715 .ev_fini = efx_ef10_ev_fini,
4716 .ev_remove = efx_ef10_ev_remove,
4717 .ev_process = efx_ef10_ev_process,
4718 .ev_read_ack = efx_ef10_ev_read_ack,
4719 .ev_test_generate = efx_ef10_ev_test_generate,
4720 .filter_table_probe = efx_ef10_filter_table_probe,
4721 .filter_table_restore = efx_ef10_filter_table_restore,
4722 .filter_table_remove = efx_ef10_filter_table_remove,
4723 .filter_update_rx_scatter = efx_ef10_filter_update_rx_scatter,
4724 .filter_insert = efx_ef10_filter_insert,
4725 .filter_remove_safe = efx_ef10_filter_remove_safe,
4726 .filter_get_safe = efx_ef10_filter_get_safe,
4727 .filter_clear_rx = efx_ef10_filter_clear_rx,
4728 .filter_count_rx_used = efx_ef10_filter_count_rx_used,
4729 .filter_get_rx_id_limit = efx_ef10_filter_get_rx_id_limit,
4730 .filter_get_rx_ids = efx_ef10_filter_get_rx_ids,
4731 #ifdef CONFIG_RFS_ACCEL
4732 .filter_rfs_insert = efx_ef10_filter_rfs_insert,
4733 .filter_rfs_expire_one = efx_ef10_filter_rfs_expire_one,
4735 #ifdef CONFIG_SFC_MTD
4736 .mtd_probe = efx_port_dummy_op_int,
4738 .ptp_write_host_time = efx_ef10_ptp_write_host_time_vf,
4739 .ptp_set_ts_config = efx_ef10_ptp_set_ts_config_vf,
4740 #ifdef CONFIG_SFC_SRIOV
4741 .vswitching_probe = efx_ef10_vswitching_probe_vf,
4742 .vswitching_restore = efx_ef10_vswitching_restore_vf,
4743 .vswitching_remove = efx_ef10_vswitching_remove_vf,
4744 .sriov_get_phys_port_id = efx_ef10_sriov_get_phys_port_id,
4746 .get_mac_address = efx_ef10_get_mac_address_vf,
4747 .set_mac_address = efx_ef10_set_mac_address,
4749 .revision = EFX_REV_HUNT_A0,
4750 .max_dma_mask = DMA_BIT_MASK(ESF_DZ_TX_KER_BUF_ADDR_WIDTH),
4751 .rx_prefix_size = ES_DZ_RX_PREFIX_SIZE,
4752 .rx_hash_offset = ES_DZ_RX_PREFIX_HASH_OFST,
4753 .rx_ts_offset = ES_DZ_RX_PREFIX_TSTAMP_OFST,
4754 .can_rx_scatter = true,
4755 .always_rx_scatter = true,
4756 .max_interrupt_mode = EFX_INT_MODE_MSIX,
4757 .timer_period_max = 1 << ERF_DD_EVQ_IND_TIMER_VAL_WIDTH,
4758 .offload_features = (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
4759 NETIF_F_RXHASH | NETIF_F_NTUPLE),
4761 .max_rx_ip_filters = HUNT_FILTER_TBL_ROWS,
4762 .hwtstamp_filters = 1 << HWTSTAMP_FILTER_NONE |
4763 1 << HWTSTAMP_FILTER_ALL,
4766 const struct efx_nic_type efx_hunt_a0_nic_type = {
4768 .mem_bar = EFX_MEM_BAR,
4769 .mem_map_size = efx_ef10_mem_map_size,
4770 .probe = efx_ef10_probe_pf,
4771 .remove = efx_ef10_remove,
4772 .dimension_resources = efx_ef10_dimension_resources,
4773 .init = efx_ef10_init_nic,
4774 .fini = efx_port_dummy_op_void,
4775 .map_reset_reason = efx_ef10_map_reset_reason,
4776 .map_reset_flags = efx_ef10_map_reset_flags,
4777 .reset = efx_ef10_reset,
4778 .probe_port = efx_mcdi_port_probe,
4779 .remove_port = efx_mcdi_port_remove,
4780 .fini_dmaq = efx_ef10_fini_dmaq,
4781 .prepare_flr = efx_ef10_prepare_flr,
4782 .finish_flr = efx_port_dummy_op_void,
4783 .describe_stats = efx_ef10_describe_stats,
4784 .update_stats = efx_ef10_update_stats_pf,
4785 .start_stats = efx_mcdi_mac_start_stats,
4786 .pull_stats = efx_mcdi_mac_pull_stats,
4787 .stop_stats = efx_mcdi_mac_stop_stats,
4788 .set_id_led = efx_mcdi_set_id_led,
4789 .push_irq_moderation = efx_ef10_push_irq_moderation,
4790 .reconfigure_mac = efx_ef10_mac_reconfigure,
4791 .check_mac_fault = efx_mcdi_mac_check_fault,
4792 .reconfigure_port = efx_mcdi_port_reconfigure,
4793 .get_wol = efx_ef10_get_wol,
4794 .set_wol = efx_ef10_set_wol,
4795 .resume_wol = efx_port_dummy_op_void,
4796 .test_chip = efx_ef10_test_chip,
4797 .test_nvram = efx_mcdi_nvram_test_all,
4798 .mcdi_request = efx_ef10_mcdi_request,
4799 .mcdi_poll_response = efx_ef10_mcdi_poll_response,
4800 .mcdi_read_response = efx_ef10_mcdi_read_response,
4801 .mcdi_poll_reboot = efx_ef10_mcdi_poll_reboot,
4802 .mcdi_reboot_detected = efx_ef10_mcdi_reboot_detected,
4803 .irq_enable_master = efx_port_dummy_op_void,
4804 .irq_test_generate = efx_ef10_irq_test_generate,
4805 .irq_disable_non_ev = efx_port_dummy_op_void,
4806 .irq_handle_msi = efx_ef10_msi_interrupt,
4807 .irq_handle_legacy = efx_ef10_legacy_interrupt,
4808 .tx_probe = efx_ef10_tx_probe,
4809 .tx_init = efx_ef10_tx_init,
4810 .tx_remove = efx_ef10_tx_remove,
4811 .tx_write = efx_ef10_tx_write,
4812 .rx_push_rss_config = efx_ef10_pf_rx_push_rss_config,
4813 .rx_probe = efx_ef10_rx_probe,
4814 .rx_init = efx_ef10_rx_init,
4815 .rx_remove = efx_ef10_rx_remove,
4816 .rx_write = efx_ef10_rx_write,
4817 .rx_defer_refill = efx_ef10_rx_defer_refill,
4818 .ev_probe = efx_ef10_ev_probe,
4819 .ev_init = efx_ef10_ev_init,
4820 .ev_fini = efx_ef10_ev_fini,
4821 .ev_remove = efx_ef10_ev_remove,
4822 .ev_process = efx_ef10_ev_process,
4823 .ev_read_ack = efx_ef10_ev_read_ack,
4824 .ev_test_generate = efx_ef10_ev_test_generate,
4825 .filter_table_probe = efx_ef10_filter_table_probe,
4826 .filter_table_restore = efx_ef10_filter_table_restore,
4827 .filter_table_remove = efx_ef10_filter_table_remove,
4828 .filter_update_rx_scatter = efx_ef10_filter_update_rx_scatter,
4829 .filter_insert = efx_ef10_filter_insert,
4830 .filter_remove_safe = efx_ef10_filter_remove_safe,
4831 .filter_get_safe = efx_ef10_filter_get_safe,
4832 .filter_clear_rx = efx_ef10_filter_clear_rx,
4833 .filter_count_rx_used = efx_ef10_filter_count_rx_used,
4834 .filter_get_rx_id_limit = efx_ef10_filter_get_rx_id_limit,
4835 .filter_get_rx_ids = efx_ef10_filter_get_rx_ids,
4836 #ifdef CONFIG_RFS_ACCEL
4837 .filter_rfs_insert = efx_ef10_filter_rfs_insert,
4838 .filter_rfs_expire_one = efx_ef10_filter_rfs_expire_one,
4840 #ifdef CONFIG_SFC_MTD
4841 .mtd_probe = efx_ef10_mtd_probe,
4842 .mtd_rename = efx_mcdi_mtd_rename,
4843 .mtd_read = efx_mcdi_mtd_read,
4844 .mtd_erase = efx_mcdi_mtd_erase,
4845 .mtd_write = efx_mcdi_mtd_write,
4846 .mtd_sync = efx_mcdi_mtd_sync,
4848 .ptp_write_host_time = efx_ef10_ptp_write_host_time,
4849 .ptp_set_ts_sync_events = efx_ef10_ptp_set_ts_sync_events,
4850 .ptp_set_ts_config = efx_ef10_ptp_set_ts_config,
4851 #ifdef CONFIG_SFC_SRIOV
4852 .sriov_configure = efx_ef10_sriov_configure,
4853 .sriov_init = efx_ef10_sriov_init,
4854 .sriov_fini = efx_ef10_sriov_fini,
4855 .sriov_wanted = efx_ef10_sriov_wanted,
4856 .sriov_reset = efx_ef10_sriov_reset,
4857 .sriov_flr = efx_ef10_sriov_flr,
4858 .sriov_set_vf_mac = efx_ef10_sriov_set_vf_mac,
4859 .sriov_set_vf_vlan = efx_ef10_sriov_set_vf_vlan,
4860 .sriov_set_vf_spoofchk = efx_ef10_sriov_set_vf_spoofchk,
4861 .sriov_get_vf_config = efx_ef10_sriov_get_vf_config,
4862 .sriov_set_vf_link_state = efx_ef10_sriov_set_vf_link_state,
4863 .vswitching_probe = efx_ef10_vswitching_probe_pf,
4864 .vswitching_restore = efx_ef10_vswitching_restore_pf,
4865 .vswitching_remove = efx_ef10_vswitching_remove_pf,
4867 .get_mac_address = efx_ef10_get_mac_address_pf,
4868 .set_mac_address = efx_ef10_set_mac_address,
4870 .revision = EFX_REV_HUNT_A0,
4871 .max_dma_mask = DMA_BIT_MASK(ESF_DZ_TX_KER_BUF_ADDR_WIDTH),
4872 .rx_prefix_size = ES_DZ_RX_PREFIX_SIZE,
4873 .rx_hash_offset = ES_DZ_RX_PREFIX_HASH_OFST,
4874 .rx_ts_offset = ES_DZ_RX_PREFIX_TSTAMP_OFST,
4875 .can_rx_scatter = true,
4876 .always_rx_scatter = true,
4877 .max_interrupt_mode = EFX_INT_MODE_MSIX,
4878 .timer_period_max = 1 << ERF_DD_EVQ_IND_TIMER_VAL_WIDTH,
4879 .offload_features = (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
4880 NETIF_F_RXHASH | NETIF_F_NTUPLE),
4882 .max_rx_ip_filters = HUNT_FILTER_TBL_ROWS,
4883 .hwtstamp_filters = 1 << HWTSTAMP_FILTER_NONE |
4884 1 << HWTSTAMP_FILTER_ALL,