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 */
86 /* Whether in multicast promiscuous mode when last changed */
90 /* An arbitrary search limit for the software hash table */
91 #define EFX_EF10_FILTER_SEARCH_LIMIT 200
93 static void efx_ef10_rx_free_indir_table(struct efx_nic *efx);
94 static void efx_ef10_filter_table_remove(struct efx_nic *efx);
96 static int efx_ef10_get_warm_boot_count(struct efx_nic *efx)
100 efx_readd(efx, ®, ER_DZ_BIU_MC_SFT_STATUS);
101 return EFX_DWORD_FIELD(reg, EFX_WORD_1) == 0xb007 ?
102 EFX_DWORD_FIELD(reg, EFX_WORD_0) : -EIO;
105 static unsigned int efx_ef10_mem_map_size(struct efx_nic *efx)
109 bar = efx->type->mem_bar;
110 return resource_size(&efx->pci_dev->resource[bar]);
113 static bool efx_ef10_is_vf(struct efx_nic *efx)
115 return efx->type->is_vf;
118 static int efx_ef10_get_pf_index(struct efx_nic *efx)
120 MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_FUNCTION_INFO_OUT_LEN);
121 struct efx_ef10_nic_data *nic_data = efx->nic_data;
125 rc = efx_mcdi_rpc(efx, MC_CMD_GET_FUNCTION_INFO, NULL, 0, outbuf,
126 sizeof(outbuf), &outlen);
129 if (outlen < sizeof(outbuf))
132 nic_data->pf_index = MCDI_DWORD(outbuf, GET_FUNCTION_INFO_OUT_PF);
136 #ifdef CONFIG_SFC_SRIOV
137 static int efx_ef10_get_vf_index(struct efx_nic *efx)
139 MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_FUNCTION_INFO_OUT_LEN);
140 struct efx_ef10_nic_data *nic_data = efx->nic_data;
144 rc = efx_mcdi_rpc(efx, MC_CMD_GET_FUNCTION_INFO, NULL, 0, outbuf,
145 sizeof(outbuf), &outlen);
148 if (outlen < sizeof(outbuf))
151 nic_data->vf_index = MCDI_DWORD(outbuf, GET_FUNCTION_INFO_OUT_VF);
156 static int efx_ef10_init_datapath_caps(struct efx_nic *efx)
158 MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_CAPABILITIES_OUT_LEN);
159 struct efx_ef10_nic_data *nic_data = efx->nic_data;
163 BUILD_BUG_ON(MC_CMD_GET_CAPABILITIES_IN_LEN != 0);
165 rc = efx_mcdi_rpc(efx, MC_CMD_GET_CAPABILITIES, NULL, 0,
166 outbuf, sizeof(outbuf), &outlen);
169 if (outlen < sizeof(outbuf)) {
170 netif_err(efx, drv, efx->net_dev,
171 "unable to read datapath firmware capabilities\n");
175 nic_data->datapath_caps =
176 MCDI_DWORD(outbuf, GET_CAPABILITIES_OUT_FLAGS1);
178 /* record the DPCPU firmware IDs to determine VEB vswitching support.
180 nic_data->rx_dpcpu_fw_id =
181 MCDI_WORD(outbuf, GET_CAPABILITIES_OUT_RX_DPCPU_FW_ID);
182 nic_data->tx_dpcpu_fw_id =
183 MCDI_WORD(outbuf, GET_CAPABILITIES_OUT_TX_DPCPU_FW_ID);
185 if (!(nic_data->datapath_caps &
186 (1 << MC_CMD_GET_CAPABILITIES_OUT_RX_PREFIX_LEN_14_LBN))) {
187 netif_err(efx, probe, efx->net_dev,
188 "current firmware does not support an RX prefix\n");
195 static int efx_ef10_get_sysclk_freq(struct efx_nic *efx)
197 MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_CLOCK_OUT_LEN);
200 rc = efx_mcdi_rpc(efx, MC_CMD_GET_CLOCK, NULL, 0,
201 outbuf, sizeof(outbuf), NULL);
204 rc = MCDI_DWORD(outbuf, GET_CLOCK_OUT_SYS_FREQ);
205 return rc > 0 ? rc : -ERANGE;
208 static int efx_ef10_get_mac_address_pf(struct efx_nic *efx, u8 *mac_address)
210 MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_MAC_ADDRESSES_OUT_LEN);
214 BUILD_BUG_ON(MC_CMD_GET_MAC_ADDRESSES_IN_LEN != 0);
216 rc = efx_mcdi_rpc(efx, MC_CMD_GET_MAC_ADDRESSES, NULL, 0,
217 outbuf, sizeof(outbuf), &outlen);
220 if (outlen < MC_CMD_GET_MAC_ADDRESSES_OUT_LEN)
223 ether_addr_copy(mac_address,
224 MCDI_PTR(outbuf, GET_MAC_ADDRESSES_OUT_MAC_ADDR_BASE));
228 static int efx_ef10_get_mac_address_vf(struct efx_nic *efx, u8 *mac_address)
230 MCDI_DECLARE_BUF(inbuf, MC_CMD_VPORT_GET_MAC_ADDRESSES_IN_LEN);
231 MCDI_DECLARE_BUF(outbuf, MC_CMD_VPORT_GET_MAC_ADDRESSES_OUT_LENMAX);
235 MCDI_SET_DWORD(inbuf, VPORT_GET_MAC_ADDRESSES_IN_VPORT_ID,
236 EVB_PORT_ID_ASSIGNED);
237 rc = efx_mcdi_rpc(efx, MC_CMD_VPORT_GET_MAC_ADDRESSES, inbuf,
238 sizeof(inbuf), outbuf, sizeof(outbuf), &outlen);
242 if (outlen < MC_CMD_VPORT_GET_MAC_ADDRESSES_OUT_LENMIN)
245 num_addrs = MCDI_DWORD(outbuf,
246 VPORT_GET_MAC_ADDRESSES_OUT_MACADDR_COUNT);
248 WARN_ON(num_addrs != 1);
250 ether_addr_copy(mac_address,
251 MCDI_PTR(outbuf, VPORT_GET_MAC_ADDRESSES_OUT_MACADDR));
256 static ssize_t efx_ef10_show_link_control_flag(struct device *dev,
257 struct device_attribute *attr,
260 struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));
262 return sprintf(buf, "%d\n",
263 ((efx->mcdi->fn_flags) &
264 (1 << MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_LINKCTRL))
268 static ssize_t efx_ef10_show_primary_flag(struct device *dev,
269 struct device_attribute *attr,
272 struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));
274 return sprintf(buf, "%d\n",
275 ((efx->mcdi->fn_flags) &
276 (1 << MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_PRIMARY))
280 static DEVICE_ATTR(link_control_flag, 0444, efx_ef10_show_link_control_flag,
282 static DEVICE_ATTR(primary_flag, 0444, efx_ef10_show_primary_flag, NULL);
284 static int efx_ef10_probe(struct efx_nic *efx)
286 struct efx_ef10_nic_data *nic_data;
287 struct net_device *net_dev = efx->net_dev;
290 /* We can have one VI for each 8K region. However, until we
291 * use TX option descriptors we need two TX queues per channel.
293 efx->max_channels = min_t(unsigned int,
295 efx_ef10_mem_map_size(efx) /
296 (EFX_VI_PAGE_SIZE * EFX_TXQ_TYPES));
297 efx->max_tx_channels = efx->max_channels;
298 if (WARN_ON(efx->max_channels == 0))
301 nic_data = kzalloc(sizeof(*nic_data), GFP_KERNEL);
304 efx->nic_data = nic_data;
306 /* we assume later that we can copy from this buffer in dwords */
307 BUILD_BUG_ON(MCDI_CTL_SDU_LEN_MAX_V2 % 4);
309 rc = efx_nic_alloc_buffer(efx, &nic_data->mcdi_buf,
310 8 + MCDI_CTL_SDU_LEN_MAX_V2, GFP_KERNEL);
314 /* Get the MC's warm boot count. In case it's rebooting right
315 * now, be prepared to retry.
319 rc = efx_ef10_get_warm_boot_count(efx);
326 nic_data->warm_boot_count = rc;
328 nic_data->rx_rss_context = EFX_EF10_RSS_CONTEXT_INVALID;
330 nic_data->vport_id = EVB_PORT_ID_ASSIGNED;
332 /* In case we're recovering from a crash (kexec), we want to
333 * cancel any outstanding request by the previous user of this
334 * function. We send a special message using the least
335 * significant bits of the 'high' (doorbell) register.
337 _efx_writed(efx, cpu_to_le32(1), ER_DZ_MC_DB_HWRD);
339 rc = efx_mcdi_init(efx);
343 /* Reset (most) configuration for this function */
344 rc = efx_mcdi_reset(efx, RESET_TYPE_ALL);
348 /* Enable event logging */
349 rc = efx_mcdi_log_ctrl(efx, true, false, 0);
353 rc = device_create_file(&efx->pci_dev->dev,
354 &dev_attr_link_control_flag);
358 rc = device_create_file(&efx->pci_dev->dev, &dev_attr_primary_flag);
362 rc = efx_ef10_get_pf_index(efx);
366 rc = efx_ef10_init_datapath_caps(efx);
370 efx->rx_packet_len_offset =
371 ES_DZ_RX_PREFIX_PKTLEN_OFST - ES_DZ_RX_PREFIX_SIZE;
373 rc = efx_mcdi_port_get_number(efx);
377 net_dev->dev_port = rc;
379 rc = efx->type->get_mac_address(efx, efx->net_dev->perm_addr);
383 rc = efx_ef10_get_sysclk_freq(efx);
386 efx->timer_quantum_ns = 1536000 / rc; /* 1536 cycles */
388 /* Check whether firmware supports bug 35388 workaround.
389 * First try to enable it, then if we get EPERM, just
390 * ask if it's already enabled
392 rc = efx_mcdi_set_workaround(efx, MC_CMD_WORKAROUND_BUG35388, true, NULL);
394 nic_data->workaround_35388 = true;
395 } else if (rc == -EPERM) {
396 unsigned int enabled;
398 rc = efx_mcdi_get_workarounds(efx, NULL, &enabled);
401 nic_data->workaround_35388 = enabled &
402 MC_CMD_GET_WORKAROUNDS_OUT_BUG35388;
403 } else if (rc != -ENOSYS && rc != -ENOENT) {
406 netif_dbg(efx, probe, efx->net_dev,
407 "workaround for bug 35388 is %sabled\n",
408 nic_data->workaround_35388 ? "en" : "dis");
410 rc = efx_mcdi_mon_probe(efx);
411 if (rc && rc != -EPERM)
414 efx_ptp_probe(efx, NULL);
416 #ifdef CONFIG_SFC_SRIOV
417 if ((efx->pci_dev->physfn) && (!efx->pci_dev->is_physfn)) {
418 struct pci_dev *pci_dev_pf = efx->pci_dev->physfn;
419 struct efx_nic *efx_pf = pci_get_drvdata(pci_dev_pf);
421 efx_pf->type->get_mac_address(efx_pf, nic_data->port_id);
424 ether_addr_copy(nic_data->port_id, efx->net_dev->perm_addr);
429 device_remove_file(&efx->pci_dev->dev, &dev_attr_primary_flag);
431 device_remove_file(&efx->pci_dev->dev, &dev_attr_link_control_flag);
435 efx_nic_free_buffer(efx, &nic_data->mcdi_buf);
438 efx->nic_data = NULL;
442 static int efx_ef10_free_vis(struct efx_nic *efx)
444 MCDI_DECLARE_BUF_ERR(outbuf);
446 int rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FREE_VIS, NULL, 0,
447 outbuf, sizeof(outbuf), &outlen);
449 /* -EALREADY means nothing to free, so ignore */
453 efx_mcdi_display_error(efx, MC_CMD_FREE_VIS, 0, outbuf, outlen,
460 static void efx_ef10_free_piobufs(struct efx_nic *efx)
462 struct efx_ef10_nic_data *nic_data = efx->nic_data;
463 MCDI_DECLARE_BUF(inbuf, MC_CMD_FREE_PIOBUF_IN_LEN);
467 BUILD_BUG_ON(MC_CMD_FREE_PIOBUF_OUT_LEN != 0);
469 for (i = 0; i < nic_data->n_piobufs; i++) {
470 MCDI_SET_DWORD(inbuf, FREE_PIOBUF_IN_PIOBUF_HANDLE,
471 nic_data->piobuf_handle[i]);
472 rc = efx_mcdi_rpc(efx, MC_CMD_FREE_PIOBUF, inbuf, sizeof(inbuf),
477 nic_data->n_piobufs = 0;
480 static int efx_ef10_alloc_piobufs(struct efx_nic *efx, unsigned int n)
482 struct efx_ef10_nic_data *nic_data = efx->nic_data;
483 MCDI_DECLARE_BUF(outbuf, MC_CMD_ALLOC_PIOBUF_OUT_LEN);
488 BUILD_BUG_ON(MC_CMD_ALLOC_PIOBUF_IN_LEN != 0);
490 for (i = 0; i < n; i++) {
491 rc = efx_mcdi_rpc_quiet(efx, MC_CMD_ALLOC_PIOBUF, NULL, 0,
492 outbuf, sizeof(outbuf), &outlen);
494 /* Don't display the MC error if we didn't have space
497 if (!(efx_ef10_is_vf(efx) && rc == -ENOSPC))
498 efx_mcdi_display_error(efx, MC_CMD_ALLOC_PIOBUF,
499 0, outbuf, outlen, rc);
502 if (outlen < MC_CMD_ALLOC_PIOBUF_OUT_LEN) {
506 nic_data->piobuf_handle[i] =
507 MCDI_DWORD(outbuf, ALLOC_PIOBUF_OUT_PIOBUF_HANDLE);
508 netif_dbg(efx, probe, efx->net_dev,
509 "allocated PIO buffer %u handle %x\n", i,
510 nic_data->piobuf_handle[i]);
513 nic_data->n_piobufs = i;
515 efx_ef10_free_piobufs(efx);
519 static int efx_ef10_link_piobufs(struct efx_nic *efx)
521 struct efx_ef10_nic_data *nic_data = efx->nic_data;
522 _MCDI_DECLARE_BUF(inbuf,
523 max(MC_CMD_LINK_PIOBUF_IN_LEN,
524 MC_CMD_UNLINK_PIOBUF_IN_LEN));
525 struct efx_channel *channel;
526 struct efx_tx_queue *tx_queue;
527 unsigned int offset, index;
530 BUILD_BUG_ON(MC_CMD_LINK_PIOBUF_OUT_LEN != 0);
531 BUILD_BUG_ON(MC_CMD_UNLINK_PIOBUF_OUT_LEN != 0);
533 memset(inbuf, 0, sizeof(inbuf));
535 /* Link a buffer to each VI in the write-combining mapping */
536 for (index = 0; index < nic_data->n_piobufs; ++index) {
537 MCDI_SET_DWORD(inbuf, LINK_PIOBUF_IN_PIOBUF_HANDLE,
538 nic_data->piobuf_handle[index]);
539 MCDI_SET_DWORD(inbuf, LINK_PIOBUF_IN_TXQ_INSTANCE,
540 nic_data->pio_write_vi_base + index);
541 rc = efx_mcdi_rpc(efx, MC_CMD_LINK_PIOBUF,
542 inbuf, MC_CMD_LINK_PIOBUF_IN_LEN,
545 netif_err(efx, drv, efx->net_dev,
546 "failed to link VI %u to PIO buffer %u (%d)\n",
547 nic_data->pio_write_vi_base + index, index,
551 netif_dbg(efx, probe, efx->net_dev,
552 "linked VI %u to PIO buffer %u\n",
553 nic_data->pio_write_vi_base + index, index);
556 /* Link a buffer to each TX queue */
557 efx_for_each_channel(channel, efx) {
558 efx_for_each_channel_tx_queue(tx_queue, channel) {
559 /* We assign the PIO buffers to queues in
560 * reverse order to allow for the following
563 offset = ((efx->tx_channel_offset + efx->n_tx_channels -
564 tx_queue->channel->channel - 1) *
566 index = offset / ER_DZ_TX_PIOBUF_SIZE;
567 offset = offset % ER_DZ_TX_PIOBUF_SIZE;
569 /* When the host page size is 4K, the first
570 * host page in the WC mapping may be within
571 * the same VI page as the last TX queue. We
572 * can only link one buffer to each VI.
574 if (tx_queue->queue == nic_data->pio_write_vi_base) {
578 MCDI_SET_DWORD(inbuf,
579 LINK_PIOBUF_IN_PIOBUF_HANDLE,
580 nic_data->piobuf_handle[index]);
581 MCDI_SET_DWORD(inbuf,
582 LINK_PIOBUF_IN_TXQ_INSTANCE,
584 rc = efx_mcdi_rpc(efx, MC_CMD_LINK_PIOBUF,
585 inbuf, MC_CMD_LINK_PIOBUF_IN_LEN,
590 /* This is non-fatal; the TX path just
591 * won't use PIO for this queue
593 netif_err(efx, drv, efx->net_dev,
594 "failed to link VI %u to PIO buffer %u (%d)\n",
595 tx_queue->queue, index, rc);
596 tx_queue->piobuf = NULL;
599 nic_data->pio_write_base +
600 index * EFX_VI_PAGE_SIZE + offset;
601 tx_queue->piobuf_offset = offset;
602 netif_dbg(efx, probe, efx->net_dev,
603 "linked VI %u to PIO buffer %u offset %x addr %p\n",
604 tx_queue->queue, index,
605 tx_queue->piobuf_offset,
615 MCDI_SET_DWORD(inbuf, UNLINK_PIOBUF_IN_TXQ_INSTANCE,
616 nic_data->pio_write_vi_base + index);
617 efx_mcdi_rpc(efx, MC_CMD_UNLINK_PIOBUF,
618 inbuf, MC_CMD_UNLINK_PIOBUF_IN_LEN,
624 static void efx_ef10_forget_old_piobufs(struct efx_nic *efx)
626 struct efx_channel *channel;
627 struct efx_tx_queue *tx_queue;
629 /* All our existing PIO buffers went away */
630 efx_for_each_channel(channel, efx)
631 efx_for_each_channel_tx_queue(tx_queue, channel)
632 tx_queue->piobuf = NULL;
635 #else /* !EFX_USE_PIO */
637 static int efx_ef10_alloc_piobufs(struct efx_nic *efx, unsigned int n)
639 return n == 0 ? 0 : -ENOBUFS;
642 static int efx_ef10_link_piobufs(struct efx_nic *efx)
647 static void efx_ef10_free_piobufs(struct efx_nic *efx)
651 static void efx_ef10_forget_old_piobufs(struct efx_nic *efx)
655 #endif /* EFX_USE_PIO */
657 static void efx_ef10_remove(struct efx_nic *efx)
659 struct efx_ef10_nic_data *nic_data = efx->nic_data;
662 #ifdef CONFIG_SFC_SRIOV
663 struct efx_ef10_nic_data *nic_data_pf;
664 struct pci_dev *pci_dev_pf;
665 struct efx_nic *efx_pf;
668 if (efx->pci_dev->is_virtfn) {
669 pci_dev_pf = efx->pci_dev->physfn;
671 efx_pf = pci_get_drvdata(pci_dev_pf);
672 nic_data_pf = efx_pf->nic_data;
673 vf = nic_data_pf->vf + nic_data->vf_index;
676 netif_info(efx, drv, efx->net_dev,
677 "Could not get the PF id from VF\n");
683 efx_mcdi_mon_remove(efx);
685 efx_ef10_rx_free_indir_table(efx);
687 if (nic_data->wc_membase)
688 iounmap(nic_data->wc_membase);
690 rc = efx_ef10_free_vis(efx);
693 if (!nic_data->must_restore_piobufs)
694 efx_ef10_free_piobufs(efx);
696 device_remove_file(&efx->pci_dev->dev, &dev_attr_primary_flag);
697 device_remove_file(&efx->pci_dev->dev, &dev_attr_link_control_flag);
700 efx_nic_free_buffer(efx, &nic_data->mcdi_buf);
704 static int efx_ef10_probe_pf(struct efx_nic *efx)
706 return efx_ef10_probe(efx);
709 int efx_ef10_vadaptor_alloc(struct efx_nic *efx, unsigned int port_id)
711 MCDI_DECLARE_BUF(inbuf, MC_CMD_VADAPTOR_ALLOC_IN_LEN);
713 MCDI_SET_DWORD(inbuf, VADAPTOR_ALLOC_IN_UPSTREAM_PORT_ID, port_id);
714 return efx_mcdi_rpc(efx, MC_CMD_VADAPTOR_ALLOC, inbuf, sizeof(inbuf),
718 int efx_ef10_vadaptor_free(struct efx_nic *efx, unsigned int port_id)
720 MCDI_DECLARE_BUF(inbuf, MC_CMD_VADAPTOR_FREE_IN_LEN);
722 MCDI_SET_DWORD(inbuf, VADAPTOR_FREE_IN_UPSTREAM_PORT_ID, port_id);
723 return efx_mcdi_rpc(efx, MC_CMD_VADAPTOR_FREE, inbuf, sizeof(inbuf),
727 int efx_ef10_vport_add_mac(struct efx_nic *efx,
728 unsigned int port_id, u8 *mac)
730 MCDI_DECLARE_BUF(inbuf, MC_CMD_VPORT_ADD_MAC_ADDRESS_IN_LEN);
732 MCDI_SET_DWORD(inbuf, VPORT_ADD_MAC_ADDRESS_IN_VPORT_ID, port_id);
733 ether_addr_copy(MCDI_PTR(inbuf, VPORT_ADD_MAC_ADDRESS_IN_MACADDR), mac);
735 return efx_mcdi_rpc(efx, MC_CMD_VPORT_ADD_MAC_ADDRESS, inbuf,
736 sizeof(inbuf), NULL, 0, NULL);
739 int efx_ef10_vport_del_mac(struct efx_nic *efx,
740 unsigned int port_id, u8 *mac)
742 MCDI_DECLARE_BUF(inbuf, MC_CMD_VPORT_DEL_MAC_ADDRESS_IN_LEN);
744 MCDI_SET_DWORD(inbuf, VPORT_DEL_MAC_ADDRESS_IN_VPORT_ID, port_id);
745 ether_addr_copy(MCDI_PTR(inbuf, VPORT_DEL_MAC_ADDRESS_IN_MACADDR), mac);
747 return efx_mcdi_rpc(efx, MC_CMD_VPORT_DEL_MAC_ADDRESS, inbuf,
748 sizeof(inbuf), NULL, 0, NULL);
751 #ifdef CONFIG_SFC_SRIOV
752 static int efx_ef10_probe_vf(struct efx_nic *efx)
755 struct pci_dev *pci_dev_pf;
757 /* If the parent PF has no VF data structure, it doesn't know about this
758 * VF so fail probe. The VF needs to be re-created. This can happen
759 * if the PF driver is unloaded while the VF is assigned to a guest.
761 pci_dev_pf = efx->pci_dev->physfn;
763 struct efx_nic *efx_pf = pci_get_drvdata(pci_dev_pf);
764 struct efx_ef10_nic_data *nic_data_pf = efx_pf->nic_data;
766 if (!nic_data_pf->vf) {
767 netif_info(efx, drv, efx->net_dev,
768 "The VF cannot link to its parent PF; "
769 "please destroy and re-create the VF\n");
774 rc = efx_ef10_probe(efx);
778 rc = efx_ef10_get_vf_index(efx);
782 if (efx->pci_dev->is_virtfn) {
783 if (efx->pci_dev->physfn) {
784 struct efx_nic *efx_pf =
785 pci_get_drvdata(efx->pci_dev->physfn);
786 struct efx_ef10_nic_data *nic_data_p = efx_pf->nic_data;
787 struct efx_ef10_nic_data *nic_data = efx->nic_data;
789 nic_data_p->vf[nic_data->vf_index].efx = efx;
790 nic_data_p->vf[nic_data->vf_index].pci_dev =
793 netif_info(efx, drv, efx->net_dev,
794 "Could not get the PF id from VF\n");
800 efx_ef10_remove(efx);
804 static int efx_ef10_probe_vf(struct efx_nic *efx __attribute__ ((unused)))
810 static int efx_ef10_alloc_vis(struct efx_nic *efx,
811 unsigned int min_vis, unsigned int max_vis)
813 MCDI_DECLARE_BUF(inbuf, MC_CMD_ALLOC_VIS_IN_LEN);
814 MCDI_DECLARE_BUF(outbuf, MC_CMD_ALLOC_VIS_OUT_LEN);
815 struct efx_ef10_nic_data *nic_data = efx->nic_data;
819 MCDI_SET_DWORD(inbuf, ALLOC_VIS_IN_MIN_VI_COUNT, min_vis);
820 MCDI_SET_DWORD(inbuf, ALLOC_VIS_IN_MAX_VI_COUNT, max_vis);
821 rc = efx_mcdi_rpc(efx, MC_CMD_ALLOC_VIS, inbuf, sizeof(inbuf),
822 outbuf, sizeof(outbuf), &outlen);
826 if (outlen < MC_CMD_ALLOC_VIS_OUT_LEN)
829 netif_dbg(efx, drv, efx->net_dev, "base VI is A0x%03x\n",
830 MCDI_DWORD(outbuf, ALLOC_VIS_OUT_VI_BASE));
832 nic_data->vi_base = MCDI_DWORD(outbuf, ALLOC_VIS_OUT_VI_BASE);
833 nic_data->n_allocated_vis = MCDI_DWORD(outbuf, ALLOC_VIS_OUT_VI_COUNT);
837 /* Note that the failure path of this function does not free
838 * resources, as this will be done by efx_ef10_remove().
840 static int efx_ef10_dimension_resources(struct efx_nic *efx)
842 struct efx_ef10_nic_data *nic_data = efx->nic_data;
843 unsigned int uc_mem_map_size, wc_mem_map_size;
844 unsigned int min_vis = max(EFX_TXQ_TYPES,
845 efx_separate_tx_channels ? 2 : 1);
846 unsigned int channel_vis, pio_write_vi_base, max_vis;
847 void __iomem *membase;
850 channel_vis = max(efx->n_channels, efx->n_tx_channels * EFX_TXQ_TYPES);
853 /* Try to allocate PIO buffers if wanted and if the full
854 * number of PIO buffers would be sufficient to allocate one
855 * copy-buffer per TX channel. Failure is non-fatal, as there
856 * are only a small number of PIO buffers shared between all
857 * functions of the controller.
859 if (efx_piobuf_size != 0 &&
860 ER_DZ_TX_PIOBUF_SIZE / efx_piobuf_size * EF10_TX_PIOBUF_COUNT >=
861 efx->n_tx_channels) {
862 unsigned int n_piobufs =
863 DIV_ROUND_UP(efx->n_tx_channels,
864 ER_DZ_TX_PIOBUF_SIZE / efx_piobuf_size);
866 rc = efx_ef10_alloc_piobufs(efx, n_piobufs);
868 netif_err(efx, probe, efx->net_dev,
869 "failed to allocate PIO buffers (%d)\n", rc);
871 netif_dbg(efx, probe, efx->net_dev,
872 "allocated %u PIO buffers\n", n_piobufs);
875 nic_data->n_piobufs = 0;
878 /* PIO buffers should be mapped with write-combining enabled,
879 * and we want to make single UC and WC mappings rather than
880 * several of each (in fact that's the only option if host
881 * page size is >4K). So we may allocate some extra VIs just
882 * for writing PIO buffers through.
884 * The UC mapping contains (channel_vis - 1) complete VIs and the
885 * first half of the next VI. Then the WC mapping begins with
886 * the second half of this last VI.
888 uc_mem_map_size = PAGE_ALIGN((channel_vis - 1) * EFX_VI_PAGE_SIZE +
890 if (nic_data->n_piobufs) {
891 /* pio_write_vi_base rounds down to give the number of complete
892 * VIs inside the UC mapping.
894 pio_write_vi_base = uc_mem_map_size / EFX_VI_PAGE_SIZE;
895 wc_mem_map_size = (PAGE_ALIGN((pio_write_vi_base +
896 nic_data->n_piobufs) *
899 max_vis = pio_write_vi_base + nic_data->n_piobufs;
901 pio_write_vi_base = 0;
903 max_vis = channel_vis;
906 /* In case the last attached driver failed to free VIs, do it now */
907 rc = efx_ef10_free_vis(efx);
911 rc = efx_ef10_alloc_vis(efx, min_vis, max_vis);
915 if (nic_data->n_allocated_vis < channel_vis) {
916 netif_info(efx, drv, efx->net_dev,
917 "Could not allocate enough VIs to satisfy RSS"
918 " requirements. Performance may not be optimal.\n");
919 /* We didn't get the VIs to populate our channels.
920 * We could keep what we got but then we'd have more
921 * interrupts than we need.
922 * Instead calculate new max_channels and restart
924 efx->max_channels = nic_data->n_allocated_vis;
925 efx->max_tx_channels =
926 nic_data->n_allocated_vis / EFX_TXQ_TYPES;
928 efx_ef10_free_vis(efx);
932 /* If we didn't get enough VIs to map all the PIO buffers, free the
935 if (nic_data->n_piobufs &&
936 nic_data->n_allocated_vis <
937 pio_write_vi_base + nic_data->n_piobufs) {
938 netif_dbg(efx, probe, efx->net_dev,
939 "%u VIs are not sufficient to map %u PIO buffers\n",
940 nic_data->n_allocated_vis, nic_data->n_piobufs);
941 efx_ef10_free_piobufs(efx);
944 /* Shrink the original UC mapping of the memory BAR */
945 membase = ioremap_nocache(efx->membase_phys, uc_mem_map_size);
947 netif_err(efx, probe, efx->net_dev,
948 "could not shrink memory BAR to %x\n",
952 iounmap(efx->membase);
953 efx->membase = membase;
955 /* Set up the WC mapping if needed */
956 if (wc_mem_map_size) {
957 nic_data->wc_membase = ioremap_wc(efx->membase_phys +
960 if (!nic_data->wc_membase) {
961 netif_err(efx, probe, efx->net_dev,
962 "could not allocate WC mapping of size %x\n",
966 nic_data->pio_write_vi_base = pio_write_vi_base;
967 nic_data->pio_write_base =
968 nic_data->wc_membase +
969 (pio_write_vi_base * EFX_VI_PAGE_SIZE + ER_DZ_TX_PIOBUF -
972 rc = efx_ef10_link_piobufs(efx);
974 efx_ef10_free_piobufs(efx);
977 netif_dbg(efx, probe, efx->net_dev,
978 "memory BAR at %pa (virtual %p+%x UC, %p+%x WC)\n",
979 &efx->membase_phys, efx->membase, uc_mem_map_size,
980 nic_data->wc_membase, wc_mem_map_size);
985 static int efx_ef10_init_nic(struct efx_nic *efx)
987 struct efx_ef10_nic_data *nic_data = efx->nic_data;
990 if (nic_data->must_check_datapath_caps) {
991 rc = efx_ef10_init_datapath_caps(efx);
994 nic_data->must_check_datapath_caps = false;
997 if (nic_data->must_realloc_vis) {
998 /* We cannot let the number of VIs change now */
999 rc = efx_ef10_alloc_vis(efx, nic_data->n_allocated_vis,
1000 nic_data->n_allocated_vis);
1003 nic_data->must_realloc_vis = false;
1006 if (nic_data->must_restore_piobufs && nic_data->n_piobufs) {
1007 rc = efx_ef10_alloc_piobufs(efx, nic_data->n_piobufs);
1009 rc = efx_ef10_link_piobufs(efx);
1011 efx_ef10_free_piobufs(efx);
1014 /* Log an error on failure, but this is non-fatal */
1016 netif_err(efx, drv, efx->net_dev,
1017 "failed to restore PIO buffers (%d)\n", rc);
1018 nic_data->must_restore_piobufs = false;
1021 /* don't fail init if RSS setup doesn't work */
1022 efx->type->rx_push_rss_config(efx, false, efx->rx_indir_table);
1027 static void efx_ef10_reset_mc_allocations(struct efx_nic *efx)
1029 struct efx_ef10_nic_data *nic_data = efx->nic_data;
1030 #ifdef CONFIG_SFC_SRIOV
1034 /* All our allocations have been reset */
1035 nic_data->must_realloc_vis = true;
1036 nic_data->must_restore_filters = true;
1037 nic_data->must_restore_piobufs = true;
1038 efx_ef10_forget_old_piobufs(efx);
1039 nic_data->rx_rss_context = EFX_EF10_RSS_CONTEXT_INVALID;
1041 /* Driver-created vswitches and vports must be re-created */
1042 nic_data->must_probe_vswitching = true;
1043 nic_data->vport_id = EVB_PORT_ID_ASSIGNED;
1044 #ifdef CONFIG_SFC_SRIOV
1046 for (i = 0; i < efx->vf_count; i++)
1047 nic_data->vf[i].vport_id = 0;
1051 static enum reset_type efx_ef10_map_reset_reason(enum reset_type reason)
1053 if (reason == RESET_TYPE_MC_FAILURE)
1054 return RESET_TYPE_DATAPATH;
1056 return efx_mcdi_map_reset_reason(reason);
1059 static int efx_ef10_map_reset_flags(u32 *flags)
1062 EF10_RESET_PORT = ((ETH_RESET_MAC | ETH_RESET_PHY) <<
1063 ETH_RESET_SHARED_SHIFT),
1064 EF10_RESET_MC = ((ETH_RESET_DMA | ETH_RESET_FILTER |
1065 ETH_RESET_OFFLOAD | ETH_RESET_MAC |
1066 ETH_RESET_PHY | ETH_RESET_MGMT) <<
1067 ETH_RESET_SHARED_SHIFT)
1070 /* We assume for now that our PCI function is permitted to
1074 if ((*flags & EF10_RESET_MC) == EF10_RESET_MC) {
1075 *flags &= ~EF10_RESET_MC;
1076 return RESET_TYPE_WORLD;
1079 if ((*flags & EF10_RESET_PORT) == EF10_RESET_PORT) {
1080 *flags &= ~EF10_RESET_PORT;
1081 return RESET_TYPE_ALL;
1084 /* no invisible reset implemented */
1089 static int efx_ef10_reset(struct efx_nic *efx, enum reset_type reset_type)
1091 int rc = efx_mcdi_reset(efx, reset_type);
1093 /* Unprivileged functions return -EPERM, but need to return success
1094 * here so that the datapath is brought back up.
1096 if (reset_type == RESET_TYPE_WORLD && rc == -EPERM)
1099 /* If it was a port reset, trigger reallocation of MC resources.
1100 * Note that on an MC reset nothing needs to be done now because we'll
1101 * detect the MC reset later and handle it then.
1102 * For an FLR, we never get an MC reset event, but the MC has reset all
1103 * resources assigned to us, so we have to trigger reallocation now.
1105 if ((reset_type == RESET_TYPE_ALL ||
1106 reset_type == RESET_TYPE_MCDI_TIMEOUT) && !rc)
1107 efx_ef10_reset_mc_allocations(efx);
1111 #define EF10_DMA_STAT(ext_name, mcdi_name) \
1112 [EF10_STAT_ ## ext_name] = \
1113 { #ext_name, 64, 8 * MC_CMD_MAC_ ## mcdi_name }
1114 #define EF10_DMA_INVIS_STAT(int_name, mcdi_name) \
1115 [EF10_STAT_ ## int_name] = \
1116 { NULL, 64, 8 * MC_CMD_MAC_ ## mcdi_name }
1117 #define EF10_OTHER_STAT(ext_name) \
1118 [EF10_STAT_ ## ext_name] = { #ext_name, 0, 0 }
1119 #define GENERIC_SW_STAT(ext_name) \
1120 [GENERIC_STAT_ ## ext_name] = { #ext_name, 0, 0 }
1122 static const struct efx_hw_stat_desc efx_ef10_stat_desc[EF10_STAT_COUNT] = {
1123 EF10_DMA_STAT(port_tx_bytes, TX_BYTES),
1124 EF10_DMA_STAT(port_tx_packets, TX_PKTS),
1125 EF10_DMA_STAT(port_tx_pause, TX_PAUSE_PKTS),
1126 EF10_DMA_STAT(port_tx_control, TX_CONTROL_PKTS),
1127 EF10_DMA_STAT(port_tx_unicast, TX_UNICAST_PKTS),
1128 EF10_DMA_STAT(port_tx_multicast, TX_MULTICAST_PKTS),
1129 EF10_DMA_STAT(port_tx_broadcast, TX_BROADCAST_PKTS),
1130 EF10_DMA_STAT(port_tx_lt64, TX_LT64_PKTS),
1131 EF10_DMA_STAT(port_tx_64, TX_64_PKTS),
1132 EF10_DMA_STAT(port_tx_65_to_127, TX_65_TO_127_PKTS),
1133 EF10_DMA_STAT(port_tx_128_to_255, TX_128_TO_255_PKTS),
1134 EF10_DMA_STAT(port_tx_256_to_511, TX_256_TO_511_PKTS),
1135 EF10_DMA_STAT(port_tx_512_to_1023, TX_512_TO_1023_PKTS),
1136 EF10_DMA_STAT(port_tx_1024_to_15xx, TX_1024_TO_15XX_PKTS),
1137 EF10_DMA_STAT(port_tx_15xx_to_jumbo, TX_15XX_TO_JUMBO_PKTS),
1138 EF10_DMA_STAT(port_rx_bytes, RX_BYTES),
1139 EF10_DMA_INVIS_STAT(port_rx_bytes_minus_good_bytes, RX_BAD_BYTES),
1140 EF10_OTHER_STAT(port_rx_good_bytes),
1141 EF10_OTHER_STAT(port_rx_bad_bytes),
1142 EF10_DMA_STAT(port_rx_packets, RX_PKTS),
1143 EF10_DMA_STAT(port_rx_good, RX_GOOD_PKTS),
1144 EF10_DMA_STAT(port_rx_bad, RX_BAD_FCS_PKTS),
1145 EF10_DMA_STAT(port_rx_pause, RX_PAUSE_PKTS),
1146 EF10_DMA_STAT(port_rx_control, RX_CONTROL_PKTS),
1147 EF10_DMA_STAT(port_rx_unicast, RX_UNICAST_PKTS),
1148 EF10_DMA_STAT(port_rx_multicast, RX_MULTICAST_PKTS),
1149 EF10_DMA_STAT(port_rx_broadcast, RX_BROADCAST_PKTS),
1150 EF10_DMA_STAT(port_rx_lt64, RX_UNDERSIZE_PKTS),
1151 EF10_DMA_STAT(port_rx_64, RX_64_PKTS),
1152 EF10_DMA_STAT(port_rx_65_to_127, RX_65_TO_127_PKTS),
1153 EF10_DMA_STAT(port_rx_128_to_255, RX_128_TO_255_PKTS),
1154 EF10_DMA_STAT(port_rx_256_to_511, RX_256_TO_511_PKTS),
1155 EF10_DMA_STAT(port_rx_512_to_1023, RX_512_TO_1023_PKTS),
1156 EF10_DMA_STAT(port_rx_1024_to_15xx, RX_1024_TO_15XX_PKTS),
1157 EF10_DMA_STAT(port_rx_15xx_to_jumbo, RX_15XX_TO_JUMBO_PKTS),
1158 EF10_DMA_STAT(port_rx_gtjumbo, RX_GTJUMBO_PKTS),
1159 EF10_DMA_STAT(port_rx_bad_gtjumbo, RX_JABBER_PKTS),
1160 EF10_DMA_STAT(port_rx_overflow, RX_OVERFLOW_PKTS),
1161 EF10_DMA_STAT(port_rx_align_error, RX_ALIGN_ERROR_PKTS),
1162 EF10_DMA_STAT(port_rx_length_error, RX_LENGTH_ERROR_PKTS),
1163 EF10_DMA_STAT(port_rx_nodesc_drops, RX_NODESC_DROPS),
1164 GENERIC_SW_STAT(rx_nodesc_trunc),
1165 GENERIC_SW_STAT(rx_noskb_drops),
1166 EF10_DMA_STAT(port_rx_pm_trunc_bb_overflow, PM_TRUNC_BB_OVERFLOW),
1167 EF10_DMA_STAT(port_rx_pm_discard_bb_overflow, PM_DISCARD_BB_OVERFLOW),
1168 EF10_DMA_STAT(port_rx_pm_trunc_vfifo_full, PM_TRUNC_VFIFO_FULL),
1169 EF10_DMA_STAT(port_rx_pm_discard_vfifo_full, PM_DISCARD_VFIFO_FULL),
1170 EF10_DMA_STAT(port_rx_pm_trunc_qbb, PM_TRUNC_QBB),
1171 EF10_DMA_STAT(port_rx_pm_discard_qbb, PM_DISCARD_QBB),
1172 EF10_DMA_STAT(port_rx_pm_discard_mapping, PM_DISCARD_MAPPING),
1173 EF10_DMA_STAT(port_rx_dp_q_disabled_packets, RXDP_Q_DISABLED_PKTS),
1174 EF10_DMA_STAT(port_rx_dp_di_dropped_packets, RXDP_DI_DROPPED_PKTS),
1175 EF10_DMA_STAT(port_rx_dp_streaming_packets, RXDP_STREAMING_PKTS),
1176 EF10_DMA_STAT(port_rx_dp_hlb_fetch, RXDP_HLB_FETCH_CONDITIONS),
1177 EF10_DMA_STAT(port_rx_dp_hlb_wait, RXDP_HLB_WAIT_CONDITIONS),
1178 EF10_DMA_STAT(rx_unicast, VADAPTER_RX_UNICAST_PACKETS),
1179 EF10_DMA_STAT(rx_unicast_bytes, VADAPTER_RX_UNICAST_BYTES),
1180 EF10_DMA_STAT(rx_multicast, VADAPTER_RX_MULTICAST_PACKETS),
1181 EF10_DMA_STAT(rx_multicast_bytes, VADAPTER_RX_MULTICAST_BYTES),
1182 EF10_DMA_STAT(rx_broadcast, VADAPTER_RX_BROADCAST_PACKETS),
1183 EF10_DMA_STAT(rx_broadcast_bytes, VADAPTER_RX_BROADCAST_BYTES),
1184 EF10_DMA_STAT(rx_bad, VADAPTER_RX_BAD_PACKETS),
1185 EF10_DMA_STAT(rx_bad_bytes, VADAPTER_RX_BAD_BYTES),
1186 EF10_DMA_STAT(rx_overflow, VADAPTER_RX_OVERFLOW),
1187 EF10_DMA_STAT(tx_unicast, VADAPTER_TX_UNICAST_PACKETS),
1188 EF10_DMA_STAT(tx_unicast_bytes, VADAPTER_TX_UNICAST_BYTES),
1189 EF10_DMA_STAT(tx_multicast, VADAPTER_TX_MULTICAST_PACKETS),
1190 EF10_DMA_STAT(tx_multicast_bytes, VADAPTER_TX_MULTICAST_BYTES),
1191 EF10_DMA_STAT(tx_broadcast, VADAPTER_TX_BROADCAST_PACKETS),
1192 EF10_DMA_STAT(tx_broadcast_bytes, VADAPTER_TX_BROADCAST_BYTES),
1193 EF10_DMA_STAT(tx_bad, VADAPTER_TX_BAD_PACKETS),
1194 EF10_DMA_STAT(tx_bad_bytes, VADAPTER_TX_BAD_BYTES),
1195 EF10_DMA_STAT(tx_overflow, VADAPTER_TX_OVERFLOW),
1198 #define HUNT_COMMON_STAT_MASK ((1ULL << EF10_STAT_port_tx_bytes) | \
1199 (1ULL << EF10_STAT_port_tx_packets) | \
1200 (1ULL << EF10_STAT_port_tx_pause) | \
1201 (1ULL << EF10_STAT_port_tx_unicast) | \
1202 (1ULL << EF10_STAT_port_tx_multicast) | \
1203 (1ULL << EF10_STAT_port_tx_broadcast) | \
1204 (1ULL << EF10_STAT_port_rx_bytes) | \
1206 EF10_STAT_port_rx_bytes_minus_good_bytes) | \
1207 (1ULL << EF10_STAT_port_rx_good_bytes) | \
1208 (1ULL << EF10_STAT_port_rx_bad_bytes) | \
1209 (1ULL << EF10_STAT_port_rx_packets) | \
1210 (1ULL << EF10_STAT_port_rx_good) | \
1211 (1ULL << EF10_STAT_port_rx_bad) | \
1212 (1ULL << EF10_STAT_port_rx_pause) | \
1213 (1ULL << EF10_STAT_port_rx_control) | \
1214 (1ULL << EF10_STAT_port_rx_unicast) | \
1215 (1ULL << EF10_STAT_port_rx_multicast) | \
1216 (1ULL << EF10_STAT_port_rx_broadcast) | \
1217 (1ULL << EF10_STAT_port_rx_lt64) | \
1218 (1ULL << EF10_STAT_port_rx_64) | \
1219 (1ULL << EF10_STAT_port_rx_65_to_127) | \
1220 (1ULL << EF10_STAT_port_rx_128_to_255) | \
1221 (1ULL << EF10_STAT_port_rx_256_to_511) | \
1222 (1ULL << EF10_STAT_port_rx_512_to_1023) |\
1223 (1ULL << EF10_STAT_port_rx_1024_to_15xx) |\
1224 (1ULL << EF10_STAT_port_rx_15xx_to_jumbo) |\
1225 (1ULL << EF10_STAT_port_rx_gtjumbo) | \
1226 (1ULL << EF10_STAT_port_rx_bad_gtjumbo) |\
1227 (1ULL << EF10_STAT_port_rx_overflow) | \
1228 (1ULL << EF10_STAT_port_rx_nodesc_drops) |\
1229 (1ULL << GENERIC_STAT_rx_nodesc_trunc) | \
1230 (1ULL << GENERIC_STAT_rx_noskb_drops))
1232 /* These statistics are only provided by the 10G MAC. For a 10G/40G
1233 * switchable port we do not expose these because they might not
1234 * include all the packets they should.
1236 #define HUNT_10G_ONLY_STAT_MASK ((1ULL << EF10_STAT_port_tx_control) | \
1237 (1ULL << EF10_STAT_port_tx_lt64) | \
1238 (1ULL << EF10_STAT_port_tx_64) | \
1239 (1ULL << EF10_STAT_port_tx_65_to_127) |\
1240 (1ULL << EF10_STAT_port_tx_128_to_255) |\
1241 (1ULL << EF10_STAT_port_tx_256_to_511) |\
1242 (1ULL << EF10_STAT_port_tx_512_to_1023) |\
1243 (1ULL << EF10_STAT_port_tx_1024_to_15xx) |\
1244 (1ULL << EF10_STAT_port_tx_15xx_to_jumbo))
1246 /* These statistics are only provided by the 40G MAC. For a 10G/40G
1247 * switchable port we do expose these because the errors will otherwise
1250 #define HUNT_40G_EXTRA_STAT_MASK ((1ULL << EF10_STAT_port_rx_align_error) |\
1251 (1ULL << EF10_STAT_port_rx_length_error))
1253 /* These statistics are only provided if the firmware supports the
1254 * capability PM_AND_RXDP_COUNTERS.
1256 #define HUNT_PM_AND_RXDP_STAT_MASK ( \
1257 (1ULL << EF10_STAT_port_rx_pm_trunc_bb_overflow) | \
1258 (1ULL << EF10_STAT_port_rx_pm_discard_bb_overflow) | \
1259 (1ULL << EF10_STAT_port_rx_pm_trunc_vfifo_full) | \
1260 (1ULL << EF10_STAT_port_rx_pm_discard_vfifo_full) | \
1261 (1ULL << EF10_STAT_port_rx_pm_trunc_qbb) | \
1262 (1ULL << EF10_STAT_port_rx_pm_discard_qbb) | \
1263 (1ULL << EF10_STAT_port_rx_pm_discard_mapping) | \
1264 (1ULL << EF10_STAT_port_rx_dp_q_disabled_packets) | \
1265 (1ULL << EF10_STAT_port_rx_dp_di_dropped_packets) | \
1266 (1ULL << EF10_STAT_port_rx_dp_streaming_packets) | \
1267 (1ULL << EF10_STAT_port_rx_dp_hlb_fetch) | \
1268 (1ULL << EF10_STAT_port_rx_dp_hlb_wait))
1270 static u64 efx_ef10_raw_stat_mask(struct efx_nic *efx)
1272 u64 raw_mask = HUNT_COMMON_STAT_MASK;
1273 u32 port_caps = efx_mcdi_phy_get_caps(efx);
1274 struct efx_ef10_nic_data *nic_data = efx->nic_data;
1276 if (!(efx->mcdi->fn_flags &
1277 1 << MC_CMD_DRV_ATTACH_EXT_OUT_FLAG_LINKCTRL))
1280 if (port_caps & (1 << MC_CMD_PHY_CAP_40000FDX_LBN))
1281 raw_mask |= HUNT_40G_EXTRA_STAT_MASK;
1283 raw_mask |= HUNT_10G_ONLY_STAT_MASK;
1285 if (nic_data->datapath_caps &
1286 (1 << MC_CMD_GET_CAPABILITIES_OUT_PM_AND_RXDP_COUNTERS_LBN))
1287 raw_mask |= HUNT_PM_AND_RXDP_STAT_MASK;
1292 static void efx_ef10_get_stat_mask(struct efx_nic *efx, unsigned long *mask)
1294 struct efx_ef10_nic_data *nic_data = efx->nic_data;
1297 raw_mask[0] = efx_ef10_raw_stat_mask(efx);
1299 /* Only show vadaptor stats when EVB capability is present */
1300 if (nic_data->datapath_caps &
1301 (1 << MC_CMD_GET_CAPABILITIES_OUT_EVB_LBN)) {
1302 raw_mask[0] |= ~((1ULL << EF10_STAT_rx_unicast) - 1);
1303 raw_mask[1] = (1ULL << (EF10_STAT_COUNT - 63)) - 1;
1308 #if BITS_PER_LONG == 64
1309 mask[0] = raw_mask[0];
1310 mask[1] = raw_mask[1];
1312 mask[0] = raw_mask[0] & 0xffffffff;
1313 mask[1] = raw_mask[0] >> 32;
1314 mask[2] = raw_mask[1] & 0xffffffff;
1315 mask[3] = raw_mask[1] >> 32;
1319 static size_t efx_ef10_describe_stats(struct efx_nic *efx, u8 *names)
1321 DECLARE_BITMAP(mask, EF10_STAT_COUNT);
1323 efx_ef10_get_stat_mask(efx, mask);
1324 return efx_nic_describe_stats(efx_ef10_stat_desc, EF10_STAT_COUNT,
1328 static size_t efx_ef10_update_stats_common(struct efx_nic *efx, u64 *full_stats,
1329 struct rtnl_link_stats64 *core_stats)
1331 DECLARE_BITMAP(mask, EF10_STAT_COUNT);
1332 struct efx_ef10_nic_data *nic_data = efx->nic_data;
1333 u64 *stats = nic_data->stats;
1334 size_t stats_count = 0, index;
1336 efx_ef10_get_stat_mask(efx, mask);
1339 for_each_set_bit(index, mask, EF10_STAT_COUNT) {
1340 if (efx_ef10_stat_desc[index].name) {
1341 *full_stats++ = stats[index];
1350 if (nic_data->datapath_caps &
1351 1 << MC_CMD_GET_CAPABILITIES_OUT_EVB_LBN) {
1352 /* Use vadaptor stats. */
1353 core_stats->rx_packets = stats[EF10_STAT_rx_unicast] +
1354 stats[EF10_STAT_rx_multicast] +
1355 stats[EF10_STAT_rx_broadcast];
1356 core_stats->tx_packets = stats[EF10_STAT_tx_unicast] +
1357 stats[EF10_STAT_tx_multicast] +
1358 stats[EF10_STAT_tx_broadcast];
1359 core_stats->rx_bytes = stats[EF10_STAT_rx_unicast_bytes] +
1360 stats[EF10_STAT_rx_multicast_bytes] +
1361 stats[EF10_STAT_rx_broadcast_bytes];
1362 core_stats->tx_bytes = stats[EF10_STAT_tx_unicast_bytes] +
1363 stats[EF10_STAT_tx_multicast_bytes] +
1364 stats[EF10_STAT_tx_broadcast_bytes];
1365 core_stats->rx_dropped = stats[GENERIC_STAT_rx_nodesc_trunc] +
1366 stats[GENERIC_STAT_rx_noskb_drops];
1367 core_stats->multicast = stats[EF10_STAT_rx_multicast];
1368 core_stats->rx_crc_errors = stats[EF10_STAT_rx_bad];
1369 core_stats->rx_fifo_errors = stats[EF10_STAT_rx_overflow];
1370 core_stats->rx_errors = core_stats->rx_crc_errors;
1371 core_stats->tx_errors = stats[EF10_STAT_tx_bad];
1373 /* Use port stats. */
1374 core_stats->rx_packets = stats[EF10_STAT_port_rx_packets];
1375 core_stats->tx_packets = stats[EF10_STAT_port_tx_packets];
1376 core_stats->rx_bytes = stats[EF10_STAT_port_rx_bytes];
1377 core_stats->tx_bytes = stats[EF10_STAT_port_tx_bytes];
1378 core_stats->rx_dropped = stats[EF10_STAT_port_rx_nodesc_drops] +
1379 stats[GENERIC_STAT_rx_nodesc_trunc] +
1380 stats[GENERIC_STAT_rx_noskb_drops];
1381 core_stats->multicast = stats[EF10_STAT_port_rx_multicast];
1382 core_stats->rx_length_errors =
1383 stats[EF10_STAT_port_rx_gtjumbo] +
1384 stats[EF10_STAT_port_rx_length_error];
1385 core_stats->rx_crc_errors = stats[EF10_STAT_port_rx_bad];
1386 core_stats->rx_frame_errors =
1387 stats[EF10_STAT_port_rx_align_error];
1388 core_stats->rx_fifo_errors = stats[EF10_STAT_port_rx_overflow];
1389 core_stats->rx_errors = (core_stats->rx_length_errors +
1390 core_stats->rx_crc_errors +
1391 core_stats->rx_frame_errors);
1397 static int efx_ef10_try_update_nic_stats_pf(struct efx_nic *efx)
1399 struct efx_ef10_nic_data *nic_data = efx->nic_data;
1400 DECLARE_BITMAP(mask, EF10_STAT_COUNT);
1401 __le64 generation_start, generation_end;
1402 u64 *stats = nic_data->stats;
1405 efx_ef10_get_stat_mask(efx, mask);
1407 dma_stats = efx->stats_buffer.addr;
1408 nic_data = efx->nic_data;
1410 generation_end = dma_stats[MC_CMD_MAC_GENERATION_END];
1411 if (generation_end == EFX_MC_STATS_GENERATION_INVALID)
1414 efx_nic_update_stats(efx_ef10_stat_desc, EF10_STAT_COUNT, mask,
1415 stats, efx->stats_buffer.addr, false);
1417 generation_start = dma_stats[MC_CMD_MAC_GENERATION_START];
1418 if (generation_end != generation_start)
1421 /* Update derived statistics */
1422 efx_nic_fix_nodesc_drop_stat(efx,
1423 &stats[EF10_STAT_port_rx_nodesc_drops]);
1424 stats[EF10_STAT_port_rx_good_bytes] =
1425 stats[EF10_STAT_port_rx_bytes] -
1426 stats[EF10_STAT_port_rx_bytes_minus_good_bytes];
1427 efx_update_diff_stat(&stats[EF10_STAT_port_rx_bad_bytes],
1428 stats[EF10_STAT_port_rx_bytes_minus_good_bytes]);
1429 efx_update_sw_stats(efx, stats);
1434 static size_t efx_ef10_update_stats_pf(struct efx_nic *efx, u64 *full_stats,
1435 struct rtnl_link_stats64 *core_stats)
1439 /* If we're unlucky enough to read statistics during the DMA, wait
1440 * up to 10ms for it to finish (typically takes <500us)
1442 for (retry = 0; retry < 100; ++retry) {
1443 if (efx_ef10_try_update_nic_stats_pf(efx) == 0)
1448 return efx_ef10_update_stats_common(efx, full_stats, core_stats);
1451 static int efx_ef10_try_update_nic_stats_vf(struct efx_nic *efx)
1453 MCDI_DECLARE_BUF(inbuf, MC_CMD_MAC_STATS_IN_LEN);
1454 struct efx_ef10_nic_data *nic_data = efx->nic_data;
1455 DECLARE_BITMAP(mask, EF10_STAT_COUNT);
1456 __le64 generation_start, generation_end;
1457 u64 *stats = nic_data->stats;
1458 u32 dma_len = MC_CMD_MAC_NSTATS * sizeof(u64);
1459 struct efx_buffer stats_buf;
1463 spin_unlock_bh(&efx->stats_lock);
1465 if (in_interrupt()) {
1466 /* If in atomic context, cannot update stats. Just update the
1467 * software stats and return so the caller can continue.
1469 spin_lock_bh(&efx->stats_lock);
1470 efx_update_sw_stats(efx, stats);
1474 efx_ef10_get_stat_mask(efx, mask);
1476 rc = efx_nic_alloc_buffer(efx, &stats_buf, dma_len, GFP_ATOMIC);
1478 spin_lock_bh(&efx->stats_lock);
1482 dma_stats = stats_buf.addr;
1483 dma_stats[MC_CMD_MAC_GENERATION_END] = EFX_MC_STATS_GENERATION_INVALID;
1485 MCDI_SET_QWORD(inbuf, MAC_STATS_IN_DMA_ADDR, stats_buf.dma_addr);
1486 MCDI_POPULATE_DWORD_1(inbuf, MAC_STATS_IN_CMD,
1487 MAC_STATS_IN_DMA, 1);
1488 MCDI_SET_DWORD(inbuf, MAC_STATS_IN_DMA_LEN, dma_len);
1489 MCDI_SET_DWORD(inbuf, MAC_STATS_IN_PORT_ID, EVB_PORT_ID_ASSIGNED);
1491 rc = efx_mcdi_rpc_quiet(efx, MC_CMD_MAC_STATS, inbuf, sizeof(inbuf),
1493 spin_lock_bh(&efx->stats_lock);
1495 /* Expect ENOENT if DMA queues have not been set up */
1496 if (rc != -ENOENT || atomic_read(&efx->active_queues))
1497 efx_mcdi_display_error(efx, MC_CMD_MAC_STATS,
1498 sizeof(inbuf), NULL, 0, rc);
1502 generation_end = dma_stats[MC_CMD_MAC_GENERATION_END];
1503 if (generation_end == EFX_MC_STATS_GENERATION_INVALID) {
1508 efx_nic_update_stats(efx_ef10_stat_desc, EF10_STAT_COUNT, mask,
1509 stats, stats_buf.addr, false);
1511 generation_start = dma_stats[MC_CMD_MAC_GENERATION_START];
1512 if (generation_end != generation_start) {
1517 efx_update_sw_stats(efx, stats);
1519 efx_nic_free_buffer(efx, &stats_buf);
1523 static size_t efx_ef10_update_stats_vf(struct efx_nic *efx, u64 *full_stats,
1524 struct rtnl_link_stats64 *core_stats)
1526 if (efx_ef10_try_update_nic_stats_vf(efx))
1529 return efx_ef10_update_stats_common(efx, full_stats, core_stats);
1532 static void efx_ef10_push_irq_moderation(struct efx_channel *channel)
1534 struct efx_nic *efx = channel->efx;
1535 unsigned int mode, value;
1536 efx_dword_t timer_cmd;
1538 if (channel->irq_moderation) {
1540 value = channel->irq_moderation - 1;
1546 if (EFX_EF10_WORKAROUND_35388(efx)) {
1547 EFX_POPULATE_DWORD_3(timer_cmd, ERF_DD_EVQ_IND_TIMER_FLAGS,
1548 EFE_DD_EVQ_IND_TIMER_FLAGS,
1549 ERF_DD_EVQ_IND_TIMER_MODE, mode,
1550 ERF_DD_EVQ_IND_TIMER_VAL, value);
1551 efx_writed_page(efx, &timer_cmd, ER_DD_EVQ_INDIRECT,
1554 EFX_POPULATE_DWORD_2(timer_cmd, ERF_DZ_TC_TIMER_MODE, mode,
1555 ERF_DZ_TC_TIMER_VAL, value);
1556 efx_writed_page(efx, &timer_cmd, ER_DZ_EVQ_TMR,
1561 static void efx_ef10_get_wol_vf(struct efx_nic *efx,
1562 struct ethtool_wolinfo *wol) {}
1564 static int efx_ef10_set_wol_vf(struct efx_nic *efx, u32 type)
1569 static void efx_ef10_get_wol(struct efx_nic *efx, struct ethtool_wolinfo *wol)
1573 memset(&wol->sopass, 0, sizeof(wol->sopass));
1576 static int efx_ef10_set_wol(struct efx_nic *efx, u32 type)
1583 static void efx_ef10_mcdi_request(struct efx_nic *efx,
1584 const efx_dword_t *hdr, size_t hdr_len,
1585 const efx_dword_t *sdu, size_t sdu_len)
1587 struct efx_ef10_nic_data *nic_data = efx->nic_data;
1588 u8 *pdu = nic_data->mcdi_buf.addr;
1590 memcpy(pdu, hdr, hdr_len);
1591 memcpy(pdu + hdr_len, sdu, sdu_len);
1594 /* The hardware provides 'low' and 'high' (doorbell) registers
1595 * for passing the 64-bit address of an MCDI request to
1596 * firmware. However the dwords are swapped by firmware. The
1597 * least significant bits of the doorbell are then 0 for all
1598 * MCDI requests due to alignment.
1600 _efx_writed(efx, cpu_to_le32((u64)nic_data->mcdi_buf.dma_addr >> 32),
1602 _efx_writed(efx, cpu_to_le32((u32)nic_data->mcdi_buf.dma_addr),
1606 static bool efx_ef10_mcdi_poll_response(struct efx_nic *efx)
1608 struct efx_ef10_nic_data *nic_data = efx->nic_data;
1609 const efx_dword_t hdr = *(const efx_dword_t *)nic_data->mcdi_buf.addr;
1612 return EFX_DWORD_FIELD(hdr, MCDI_HEADER_RESPONSE);
1616 efx_ef10_mcdi_read_response(struct efx_nic *efx, efx_dword_t *outbuf,
1617 size_t offset, size_t outlen)
1619 struct efx_ef10_nic_data *nic_data = efx->nic_data;
1620 const u8 *pdu = nic_data->mcdi_buf.addr;
1622 memcpy(outbuf, pdu + offset, outlen);
1625 static void efx_ef10_mcdi_reboot_detected(struct efx_nic *efx)
1627 struct efx_ef10_nic_data *nic_data = efx->nic_data;
1629 /* All our allocations have been reset */
1630 efx_ef10_reset_mc_allocations(efx);
1632 /* The datapath firmware might have been changed */
1633 nic_data->must_check_datapath_caps = true;
1635 /* MAC statistics have been cleared on the NIC; clear the local
1636 * statistic that we update with efx_update_diff_stat().
1638 nic_data->stats[EF10_STAT_port_rx_bad_bytes] = 0;
1641 static int efx_ef10_mcdi_poll_reboot(struct efx_nic *efx)
1643 struct efx_ef10_nic_data *nic_data = efx->nic_data;
1646 rc = efx_ef10_get_warm_boot_count(efx);
1648 /* The firmware is presumably in the process of
1649 * rebooting. However, we are supposed to report each
1650 * reboot just once, so we must only do that once we
1651 * can read and store the updated warm boot count.
1656 if (rc == nic_data->warm_boot_count)
1659 nic_data->warm_boot_count = rc;
1660 efx_ef10_mcdi_reboot_detected(efx);
1665 /* Handle an MSI interrupt
1667 * Handle an MSI hardware interrupt. This routine schedules event
1668 * queue processing. No interrupt acknowledgement cycle is necessary.
1669 * Also, we never need to check that the interrupt is for us, since
1670 * MSI interrupts cannot be shared.
1672 static irqreturn_t efx_ef10_msi_interrupt(int irq, void *dev_id)
1674 struct efx_msi_context *context = dev_id;
1675 struct efx_nic *efx = context->efx;
1677 netif_vdbg(efx, intr, efx->net_dev,
1678 "IRQ %d on CPU %d\n", irq, raw_smp_processor_id());
1680 if (likely(ACCESS_ONCE(efx->irq_soft_enabled))) {
1681 /* Note test interrupts */
1682 if (context->index == efx->irq_level)
1683 efx->last_irq_cpu = raw_smp_processor_id();
1685 /* Schedule processing of the channel */
1686 efx_schedule_channel_irq(efx->channel[context->index]);
1692 static irqreturn_t efx_ef10_legacy_interrupt(int irq, void *dev_id)
1694 struct efx_nic *efx = dev_id;
1695 bool soft_enabled = ACCESS_ONCE(efx->irq_soft_enabled);
1696 struct efx_channel *channel;
1700 /* Read the ISR which also ACKs the interrupts */
1701 efx_readd(efx, ®, ER_DZ_BIU_INT_ISR);
1702 queues = EFX_DWORD_FIELD(reg, ERF_DZ_ISR_REG);
1707 if (likely(soft_enabled)) {
1708 /* Note test interrupts */
1709 if (queues & (1U << efx->irq_level))
1710 efx->last_irq_cpu = raw_smp_processor_id();
1712 efx_for_each_channel(channel, efx) {
1714 efx_schedule_channel_irq(channel);
1719 netif_vdbg(efx, intr, efx->net_dev,
1720 "IRQ %d on CPU %d status " EFX_DWORD_FMT "\n",
1721 irq, raw_smp_processor_id(), EFX_DWORD_VAL(reg));
1726 static void efx_ef10_irq_test_generate(struct efx_nic *efx)
1728 MCDI_DECLARE_BUF(inbuf, MC_CMD_TRIGGER_INTERRUPT_IN_LEN);
1730 BUILD_BUG_ON(MC_CMD_TRIGGER_INTERRUPT_OUT_LEN != 0);
1732 MCDI_SET_DWORD(inbuf, TRIGGER_INTERRUPT_IN_INTR_LEVEL, efx->irq_level);
1733 (void) efx_mcdi_rpc(efx, MC_CMD_TRIGGER_INTERRUPT,
1734 inbuf, sizeof(inbuf), NULL, 0, NULL);
1737 static int efx_ef10_tx_probe(struct efx_tx_queue *tx_queue)
1739 return efx_nic_alloc_buffer(tx_queue->efx, &tx_queue->txd.buf,
1740 (tx_queue->ptr_mask + 1) *
1741 sizeof(efx_qword_t),
1745 /* This writes to the TX_DESC_WPTR and also pushes data */
1746 static inline void efx_ef10_push_tx_desc(struct efx_tx_queue *tx_queue,
1747 const efx_qword_t *txd)
1749 unsigned int write_ptr;
1752 write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
1753 EFX_POPULATE_OWORD_1(reg, ERF_DZ_TX_DESC_WPTR, write_ptr);
1754 reg.qword[0] = *txd;
1755 efx_writeo_page(tx_queue->efx, ®,
1756 ER_DZ_TX_DESC_UPD, tx_queue->queue);
1759 static void efx_ef10_tx_init(struct efx_tx_queue *tx_queue)
1761 MCDI_DECLARE_BUF(inbuf, MC_CMD_INIT_TXQ_IN_LEN(EFX_MAX_DMAQ_SIZE * 8 /
1763 bool csum_offload = tx_queue->queue & EFX_TXQ_TYPE_OFFLOAD;
1764 size_t entries = tx_queue->txd.buf.len / EFX_BUF_SIZE;
1765 struct efx_channel *channel = tx_queue->channel;
1766 struct efx_nic *efx = tx_queue->efx;
1767 struct efx_ef10_nic_data *nic_data = efx->nic_data;
1769 dma_addr_t dma_addr;
1773 BUILD_BUG_ON(MC_CMD_INIT_TXQ_OUT_LEN != 0);
1775 MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_SIZE, tx_queue->ptr_mask + 1);
1776 MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_TARGET_EVQ, channel->channel);
1777 MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_LABEL, tx_queue->queue);
1778 MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_INSTANCE, tx_queue->queue);
1779 MCDI_POPULATE_DWORD_2(inbuf, INIT_TXQ_IN_FLAGS,
1780 INIT_TXQ_IN_FLAG_IP_CSUM_DIS, !csum_offload,
1781 INIT_TXQ_IN_FLAG_TCP_CSUM_DIS, !csum_offload);
1782 MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_OWNER_ID, 0);
1783 MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_PORT_ID, nic_data->vport_id);
1785 dma_addr = tx_queue->txd.buf.dma_addr;
1787 netif_dbg(efx, hw, efx->net_dev, "pushing TXQ %d. %zu entries (%llx)\n",
1788 tx_queue->queue, entries, (u64)dma_addr);
1790 for (i = 0; i < entries; ++i) {
1791 MCDI_SET_ARRAY_QWORD(inbuf, INIT_TXQ_IN_DMA_ADDR, i, dma_addr);
1792 dma_addr += EFX_BUF_SIZE;
1795 inlen = MC_CMD_INIT_TXQ_IN_LEN(entries);
1797 rc = efx_mcdi_rpc(efx, MC_CMD_INIT_TXQ, inbuf, inlen,
1802 /* A previous user of this TX queue might have set us up the
1803 * bomb by writing a descriptor to the TX push collector but
1804 * not the doorbell. (Each collector belongs to a port, not a
1805 * queue or function, so cannot easily be reset.) We must
1806 * attempt to push a no-op descriptor in its place.
1808 tx_queue->buffer[0].flags = EFX_TX_BUF_OPTION;
1809 tx_queue->insert_count = 1;
1810 txd = efx_tx_desc(tx_queue, 0);
1811 EFX_POPULATE_QWORD_4(*txd,
1812 ESF_DZ_TX_DESC_IS_OPT, true,
1813 ESF_DZ_TX_OPTION_TYPE,
1814 ESE_DZ_TX_OPTION_DESC_CRC_CSUM,
1815 ESF_DZ_TX_OPTION_UDP_TCP_CSUM, csum_offload,
1816 ESF_DZ_TX_OPTION_IP_CSUM, csum_offload);
1817 tx_queue->write_count = 1;
1819 if (nic_data->datapath_caps &
1820 (1 << MC_CMD_GET_CAPABILITIES_OUT_TX_TSO_LBN)) {
1821 tx_queue->tso_version = 1;
1825 efx_ef10_push_tx_desc(tx_queue, txd);
1830 netdev_WARN(efx->net_dev, "failed to initialise TXQ %d\n",
1834 static void efx_ef10_tx_fini(struct efx_tx_queue *tx_queue)
1836 MCDI_DECLARE_BUF(inbuf, MC_CMD_FINI_TXQ_IN_LEN);
1837 MCDI_DECLARE_BUF_ERR(outbuf);
1838 struct efx_nic *efx = tx_queue->efx;
1842 MCDI_SET_DWORD(inbuf, FINI_TXQ_IN_INSTANCE,
1845 rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FINI_TXQ, inbuf, sizeof(inbuf),
1846 outbuf, sizeof(outbuf), &outlen);
1848 if (rc && rc != -EALREADY)
1854 efx_mcdi_display_error(efx, MC_CMD_FINI_TXQ, MC_CMD_FINI_TXQ_IN_LEN,
1855 outbuf, outlen, rc);
1858 static void efx_ef10_tx_remove(struct efx_tx_queue *tx_queue)
1860 efx_nic_free_buffer(tx_queue->efx, &tx_queue->txd.buf);
1863 /* This writes to the TX_DESC_WPTR; write pointer for TX descriptor ring */
1864 static inline void efx_ef10_notify_tx_desc(struct efx_tx_queue *tx_queue)
1866 unsigned int write_ptr;
1869 write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
1870 EFX_POPULATE_DWORD_1(reg, ERF_DZ_TX_DESC_WPTR_DWORD, write_ptr);
1871 efx_writed_page(tx_queue->efx, ®,
1872 ER_DZ_TX_DESC_UPD_DWORD, tx_queue->queue);
1875 static void efx_ef10_tx_write(struct efx_tx_queue *tx_queue)
1877 unsigned int old_write_count = tx_queue->write_count;
1878 struct efx_tx_buffer *buffer;
1879 unsigned int write_ptr;
1882 tx_queue->xmit_more_available = false;
1883 if (unlikely(tx_queue->write_count == tx_queue->insert_count))
1887 write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
1888 buffer = &tx_queue->buffer[write_ptr];
1889 txd = efx_tx_desc(tx_queue, write_ptr);
1890 ++tx_queue->write_count;
1892 /* Create TX descriptor ring entry */
1893 if (buffer->flags & EFX_TX_BUF_OPTION) {
1894 *txd = buffer->option;
1896 BUILD_BUG_ON(EFX_TX_BUF_CONT != 1);
1897 EFX_POPULATE_QWORD_3(
1900 buffer->flags & EFX_TX_BUF_CONT,
1901 ESF_DZ_TX_KER_BYTE_CNT, buffer->len,
1902 ESF_DZ_TX_KER_BUF_ADDR, buffer->dma_addr);
1904 } while (tx_queue->write_count != tx_queue->insert_count);
1906 wmb(); /* Ensure descriptors are written before they are fetched */
1908 if (efx_nic_may_push_tx_desc(tx_queue, old_write_count)) {
1909 txd = efx_tx_desc(tx_queue,
1910 old_write_count & tx_queue->ptr_mask);
1911 efx_ef10_push_tx_desc(tx_queue, txd);
1914 efx_ef10_notify_tx_desc(tx_queue);
1918 static int efx_ef10_alloc_rss_context(struct efx_nic *efx, u32 *context,
1919 bool exclusive, unsigned *context_size)
1921 MCDI_DECLARE_BUF(inbuf, MC_CMD_RSS_CONTEXT_ALLOC_IN_LEN);
1922 MCDI_DECLARE_BUF(outbuf, MC_CMD_RSS_CONTEXT_ALLOC_OUT_LEN);
1923 struct efx_ef10_nic_data *nic_data = efx->nic_data;
1926 u32 alloc_type = exclusive ?
1927 MC_CMD_RSS_CONTEXT_ALLOC_IN_TYPE_EXCLUSIVE :
1928 MC_CMD_RSS_CONTEXT_ALLOC_IN_TYPE_SHARED;
1929 unsigned rss_spread = exclusive ?
1931 min(rounddown_pow_of_two(efx->rss_spread),
1932 EFX_EF10_MAX_SHARED_RSS_CONTEXT_SIZE);
1934 if (!exclusive && rss_spread == 1) {
1935 *context = EFX_EF10_RSS_CONTEXT_INVALID;
1941 if (nic_data->datapath_caps &
1942 1 << MC_CMD_GET_CAPABILITIES_OUT_RX_RSS_LIMITED_LBN)
1945 MCDI_SET_DWORD(inbuf, RSS_CONTEXT_ALLOC_IN_UPSTREAM_PORT_ID,
1946 nic_data->vport_id);
1947 MCDI_SET_DWORD(inbuf, RSS_CONTEXT_ALLOC_IN_TYPE, alloc_type);
1948 MCDI_SET_DWORD(inbuf, RSS_CONTEXT_ALLOC_IN_NUM_QUEUES, rss_spread);
1950 rc = efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_ALLOC, inbuf, sizeof(inbuf),
1951 outbuf, sizeof(outbuf), &outlen);
1955 if (outlen < MC_CMD_RSS_CONTEXT_ALLOC_OUT_LEN)
1958 *context = MCDI_DWORD(outbuf, RSS_CONTEXT_ALLOC_OUT_RSS_CONTEXT_ID);
1961 *context_size = rss_spread;
1966 static void efx_ef10_free_rss_context(struct efx_nic *efx, u32 context)
1968 MCDI_DECLARE_BUF(inbuf, MC_CMD_RSS_CONTEXT_FREE_IN_LEN);
1971 MCDI_SET_DWORD(inbuf, RSS_CONTEXT_FREE_IN_RSS_CONTEXT_ID,
1974 rc = efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_FREE, inbuf, sizeof(inbuf),
1979 static int efx_ef10_populate_rss_table(struct efx_nic *efx, u32 context,
1980 const u32 *rx_indir_table)
1982 MCDI_DECLARE_BUF(tablebuf, MC_CMD_RSS_CONTEXT_SET_TABLE_IN_LEN);
1983 MCDI_DECLARE_BUF(keybuf, MC_CMD_RSS_CONTEXT_SET_KEY_IN_LEN);
1986 MCDI_SET_DWORD(tablebuf, RSS_CONTEXT_SET_TABLE_IN_RSS_CONTEXT_ID,
1988 BUILD_BUG_ON(ARRAY_SIZE(efx->rx_indir_table) !=
1989 MC_CMD_RSS_CONTEXT_SET_TABLE_IN_INDIRECTION_TABLE_LEN);
1991 for (i = 0; i < ARRAY_SIZE(efx->rx_indir_table); ++i)
1993 RSS_CONTEXT_SET_TABLE_IN_INDIRECTION_TABLE)[i] =
1994 (u8) rx_indir_table[i];
1996 rc = efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_SET_TABLE, tablebuf,
1997 sizeof(tablebuf), NULL, 0, NULL);
2001 MCDI_SET_DWORD(keybuf, RSS_CONTEXT_SET_KEY_IN_RSS_CONTEXT_ID,
2003 BUILD_BUG_ON(ARRAY_SIZE(efx->rx_hash_key) !=
2004 MC_CMD_RSS_CONTEXT_SET_KEY_IN_TOEPLITZ_KEY_LEN);
2005 for (i = 0; i < ARRAY_SIZE(efx->rx_hash_key); ++i)
2006 MCDI_PTR(keybuf, RSS_CONTEXT_SET_KEY_IN_TOEPLITZ_KEY)[i] =
2007 efx->rx_hash_key[i];
2009 return efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_SET_KEY, keybuf,
2010 sizeof(keybuf), NULL, 0, NULL);
2013 static void efx_ef10_rx_free_indir_table(struct efx_nic *efx)
2015 struct efx_ef10_nic_data *nic_data = efx->nic_data;
2017 if (nic_data->rx_rss_context != EFX_EF10_RSS_CONTEXT_INVALID)
2018 efx_ef10_free_rss_context(efx, nic_data->rx_rss_context);
2019 nic_data->rx_rss_context = EFX_EF10_RSS_CONTEXT_INVALID;
2022 static int efx_ef10_rx_push_shared_rss_config(struct efx_nic *efx,
2023 unsigned *context_size)
2025 u32 new_rx_rss_context;
2026 struct efx_ef10_nic_data *nic_data = efx->nic_data;
2027 int rc = efx_ef10_alloc_rss_context(efx, &new_rx_rss_context,
2028 false, context_size);
2033 nic_data->rx_rss_context = new_rx_rss_context;
2034 nic_data->rx_rss_context_exclusive = false;
2035 efx_set_default_rx_indir_table(efx);
2039 static int efx_ef10_rx_push_exclusive_rss_config(struct efx_nic *efx,
2040 const u32 *rx_indir_table)
2042 struct efx_ef10_nic_data *nic_data = efx->nic_data;
2044 u32 new_rx_rss_context;
2046 if (nic_data->rx_rss_context == EFX_EF10_RSS_CONTEXT_INVALID ||
2047 !nic_data->rx_rss_context_exclusive) {
2048 rc = efx_ef10_alloc_rss_context(efx, &new_rx_rss_context,
2050 if (rc == -EOPNOTSUPP)
2055 new_rx_rss_context = nic_data->rx_rss_context;
2058 rc = efx_ef10_populate_rss_table(efx, new_rx_rss_context,
2063 if (nic_data->rx_rss_context != new_rx_rss_context)
2064 efx_ef10_rx_free_indir_table(efx);
2065 nic_data->rx_rss_context = new_rx_rss_context;
2066 nic_data->rx_rss_context_exclusive = true;
2067 if (rx_indir_table != efx->rx_indir_table)
2068 memcpy(efx->rx_indir_table, rx_indir_table,
2069 sizeof(efx->rx_indir_table));
2073 if (new_rx_rss_context != nic_data->rx_rss_context)
2074 efx_ef10_free_rss_context(efx, new_rx_rss_context);
2076 netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
2080 static int efx_ef10_pf_rx_push_rss_config(struct efx_nic *efx, bool user,
2081 const u32 *rx_indir_table)
2085 if (efx->rss_spread == 1)
2088 rc = efx_ef10_rx_push_exclusive_rss_config(efx, rx_indir_table);
2090 if (rc == -ENOBUFS && !user) {
2091 unsigned context_size;
2092 bool mismatch = false;
2095 for (i = 0; i < ARRAY_SIZE(efx->rx_indir_table) && !mismatch;
2097 mismatch = rx_indir_table[i] !=
2098 ethtool_rxfh_indir_default(i, efx->rss_spread);
2100 rc = efx_ef10_rx_push_shared_rss_config(efx, &context_size);
2102 if (context_size != efx->rss_spread)
2103 netif_warn(efx, probe, efx->net_dev,
2104 "Could not allocate an exclusive RSS"
2105 " context; allocated a shared one of"
2107 " Wanted %u, got %u.\n",
2108 efx->rss_spread, context_size);
2110 netif_warn(efx, probe, efx->net_dev,
2111 "Could not allocate an exclusive RSS"
2112 " context; allocated a shared one but"
2113 " could not apply custom"
2116 netif_info(efx, probe, efx->net_dev,
2117 "Could not allocate an exclusive RSS"
2118 " context; allocated a shared one.\n");
2124 static int efx_ef10_vf_rx_push_rss_config(struct efx_nic *efx, bool user,
2125 const u32 *rx_indir_table
2126 __attribute__ ((unused)))
2128 struct efx_ef10_nic_data *nic_data = efx->nic_data;
2132 if (nic_data->rx_rss_context != EFX_EF10_RSS_CONTEXT_INVALID)
2134 return efx_ef10_rx_push_shared_rss_config(efx, NULL);
2137 static int efx_ef10_rx_probe(struct efx_rx_queue *rx_queue)
2139 return efx_nic_alloc_buffer(rx_queue->efx, &rx_queue->rxd.buf,
2140 (rx_queue->ptr_mask + 1) *
2141 sizeof(efx_qword_t),
2145 static void efx_ef10_rx_init(struct efx_rx_queue *rx_queue)
2147 MCDI_DECLARE_BUF(inbuf,
2148 MC_CMD_INIT_RXQ_IN_LEN(EFX_MAX_DMAQ_SIZE * 8 /
2150 struct efx_channel *channel = efx_rx_queue_channel(rx_queue);
2151 size_t entries = rx_queue->rxd.buf.len / EFX_BUF_SIZE;
2152 struct efx_nic *efx = rx_queue->efx;
2153 struct efx_ef10_nic_data *nic_data = efx->nic_data;
2155 dma_addr_t dma_addr;
2158 BUILD_BUG_ON(MC_CMD_INIT_RXQ_OUT_LEN != 0);
2160 rx_queue->scatter_n = 0;
2161 rx_queue->scatter_len = 0;
2163 MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_SIZE, rx_queue->ptr_mask + 1);
2164 MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_TARGET_EVQ, channel->channel);
2165 MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_LABEL, efx_rx_queue_index(rx_queue));
2166 MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_INSTANCE,
2167 efx_rx_queue_index(rx_queue));
2168 MCDI_POPULATE_DWORD_2(inbuf, INIT_RXQ_IN_FLAGS,
2169 INIT_RXQ_IN_FLAG_PREFIX, 1,
2170 INIT_RXQ_IN_FLAG_TIMESTAMP, 1);
2171 MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_OWNER_ID, 0);
2172 MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_PORT_ID, nic_data->vport_id);
2174 dma_addr = rx_queue->rxd.buf.dma_addr;
2176 netif_dbg(efx, hw, efx->net_dev, "pushing RXQ %d. %zu entries (%llx)\n",
2177 efx_rx_queue_index(rx_queue), entries, (u64)dma_addr);
2179 for (i = 0; i < entries; ++i) {
2180 MCDI_SET_ARRAY_QWORD(inbuf, INIT_RXQ_IN_DMA_ADDR, i, dma_addr);
2181 dma_addr += EFX_BUF_SIZE;
2184 inlen = MC_CMD_INIT_RXQ_IN_LEN(entries);
2186 rc = efx_mcdi_rpc(efx, MC_CMD_INIT_RXQ, inbuf, inlen,
2189 netdev_WARN(efx->net_dev, "failed to initialise RXQ %d\n",
2190 efx_rx_queue_index(rx_queue));
2193 static void efx_ef10_rx_fini(struct efx_rx_queue *rx_queue)
2195 MCDI_DECLARE_BUF(inbuf, MC_CMD_FINI_RXQ_IN_LEN);
2196 MCDI_DECLARE_BUF_ERR(outbuf);
2197 struct efx_nic *efx = rx_queue->efx;
2201 MCDI_SET_DWORD(inbuf, FINI_RXQ_IN_INSTANCE,
2202 efx_rx_queue_index(rx_queue));
2204 rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FINI_RXQ, inbuf, sizeof(inbuf),
2205 outbuf, sizeof(outbuf), &outlen);
2207 if (rc && rc != -EALREADY)
2213 efx_mcdi_display_error(efx, MC_CMD_FINI_RXQ, MC_CMD_FINI_RXQ_IN_LEN,
2214 outbuf, outlen, rc);
2217 static void efx_ef10_rx_remove(struct efx_rx_queue *rx_queue)
2219 efx_nic_free_buffer(rx_queue->efx, &rx_queue->rxd.buf);
2222 /* This creates an entry in the RX descriptor queue */
2224 efx_ef10_build_rx_desc(struct efx_rx_queue *rx_queue, unsigned int index)
2226 struct efx_rx_buffer *rx_buf;
2229 rxd = efx_rx_desc(rx_queue, index);
2230 rx_buf = efx_rx_buffer(rx_queue, index);
2231 EFX_POPULATE_QWORD_2(*rxd,
2232 ESF_DZ_RX_KER_BYTE_CNT, rx_buf->len,
2233 ESF_DZ_RX_KER_BUF_ADDR, rx_buf->dma_addr);
2236 static void efx_ef10_rx_write(struct efx_rx_queue *rx_queue)
2238 struct efx_nic *efx = rx_queue->efx;
2239 unsigned int write_count;
2242 /* Firmware requires that RX_DESC_WPTR be a multiple of 8 */
2243 write_count = rx_queue->added_count & ~7;
2244 if (rx_queue->notified_count == write_count)
2248 efx_ef10_build_rx_desc(
2250 rx_queue->notified_count & rx_queue->ptr_mask);
2251 while (++rx_queue->notified_count != write_count);
2254 EFX_POPULATE_DWORD_1(reg, ERF_DZ_RX_DESC_WPTR,
2255 write_count & rx_queue->ptr_mask);
2256 efx_writed_page(efx, ®, ER_DZ_RX_DESC_UPD,
2257 efx_rx_queue_index(rx_queue));
2260 static efx_mcdi_async_completer efx_ef10_rx_defer_refill_complete;
2262 static void efx_ef10_rx_defer_refill(struct efx_rx_queue *rx_queue)
2264 struct efx_channel *channel = efx_rx_queue_channel(rx_queue);
2265 MCDI_DECLARE_BUF(inbuf, MC_CMD_DRIVER_EVENT_IN_LEN);
2268 EFX_POPULATE_QWORD_2(event,
2269 ESF_DZ_EV_CODE, EFX_EF10_DRVGEN_EV,
2270 ESF_DZ_EV_DATA, EFX_EF10_REFILL);
2272 MCDI_SET_DWORD(inbuf, DRIVER_EVENT_IN_EVQ, channel->channel);
2274 /* MCDI_SET_QWORD is not appropriate here since EFX_POPULATE_* has
2275 * already swapped the data to little-endian order.
2277 memcpy(MCDI_PTR(inbuf, DRIVER_EVENT_IN_DATA), &event.u64[0],
2278 sizeof(efx_qword_t));
2280 efx_mcdi_rpc_async(channel->efx, MC_CMD_DRIVER_EVENT,
2281 inbuf, sizeof(inbuf), 0,
2282 efx_ef10_rx_defer_refill_complete, 0);
2286 efx_ef10_rx_defer_refill_complete(struct efx_nic *efx, unsigned long cookie,
2287 int rc, efx_dword_t *outbuf,
2288 size_t outlen_actual)
2293 static int efx_ef10_ev_probe(struct efx_channel *channel)
2295 return efx_nic_alloc_buffer(channel->efx, &channel->eventq.buf,
2296 (channel->eventq_mask + 1) *
2297 sizeof(efx_qword_t),
2301 static void efx_ef10_ev_fini(struct efx_channel *channel)
2303 MCDI_DECLARE_BUF(inbuf, MC_CMD_FINI_EVQ_IN_LEN);
2304 MCDI_DECLARE_BUF_ERR(outbuf);
2305 struct efx_nic *efx = channel->efx;
2309 MCDI_SET_DWORD(inbuf, FINI_EVQ_IN_INSTANCE, channel->channel);
2311 rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FINI_EVQ, inbuf, sizeof(inbuf),
2312 outbuf, sizeof(outbuf), &outlen);
2314 if (rc && rc != -EALREADY)
2320 efx_mcdi_display_error(efx, MC_CMD_FINI_EVQ, MC_CMD_FINI_EVQ_IN_LEN,
2321 outbuf, outlen, rc);
2324 static int efx_ef10_ev_init(struct efx_channel *channel)
2326 MCDI_DECLARE_BUF(inbuf,
2327 MC_CMD_INIT_EVQ_IN_LEN(EFX_MAX_EVQ_SIZE * 8 /
2329 MCDI_DECLARE_BUF(outbuf, MC_CMD_INIT_EVQ_OUT_LEN);
2330 size_t entries = channel->eventq.buf.len / EFX_BUF_SIZE;
2331 struct efx_nic *efx = channel->efx;
2332 struct efx_ef10_nic_data *nic_data;
2333 bool supports_rx_merge;
2334 size_t inlen, outlen;
2335 unsigned int enabled, implemented;
2336 dma_addr_t dma_addr;
2340 nic_data = efx->nic_data;
2342 !!(nic_data->datapath_caps &
2343 1 << MC_CMD_GET_CAPABILITIES_OUT_RX_BATCHING_LBN);
2345 /* Fill event queue with all ones (i.e. empty events) */
2346 memset(channel->eventq.buf.addr, 0xff, channel->eventq.buf.len);
2348 MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_SIZE, channel->eventq_mask + 1);
2349 MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_INSTANCE, channel->channel);
2350 /* INIT_EVQ expects index in vector table, not absolute */
2351 MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_IRQ_NUM, channel->channel);
2352 MCDI_POPULATE_DWORD_4(inbuf, INIT_EVQ_IN_FLAGS,
2353 INIT_EVQ_IN_FLAG_INTERRUPTING, 1,
2354 INIT_EVQ_IN_FLAG_RX_MERGE, 1,
2355 INIT_EVQ_IN_FLAG_TX_MERGE, 1,
2356 INIT_EVQ_IN_FLAG_CUT_THRU, !supports_rx_merge);
2357 MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_TMR_MODE,
2358 MC_CMD_INIT_EVQ_IN_TMR_MODE_DIS);
2359 MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_TMR_LOAD, 0);
2360 MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_TMR_RELOAD, 0);
2361 MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_COUNT_MODE,
2362 MC_CMD_INIT_EVQ_IN_COUNT_MODE_DIS);
2363 MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_COUNT_THRSHLD, 0);
2365 dma_addr = channel->eventq.buf.dma_addr;
2366 for (i = 0; i < entries; ++i) {
2367 MCDI_SET_ARRAY_QWORD(inbuf, INIT_EVQ_IN_DMA_ADDR, i, dma_addr);
2368 dma_addr += EFX_BUF_SIZE;
2371 inlen = MC_CMD_INIT_EVQ_IN_LEN(entries);
2373 rc = efx_mcdi_rpc(efx, MC_CMD_INIT_EVQ, inbuf, inlen,
2374 outbuf, sizeof(outbuf), &outlen);
2375 /* IRQ return is ignored */
2376 if (channel->channel || rc)
2379 /* Successfully created event queue on channel 0 */
2380 rc = efx_mcdi_get_workarounds(efx, &implemented, &enabled);
2381 if (rc == -ENOSYS) {
2382 /* GET_WORKAROUNDS was implemented before the bug26807
2383 * workaround, thus the latter must be unavailable in this fw
2385 nic_data->workaround_26807 = false;
2390 nic_data->workaround_26807 =
2391 !!(enabled & MC_CMD_GET_WORKAROUNDS_OUT_BUG26807);
2393 if (implemented & MC_CMD_GET_WORKAROUNDS_OUT_BUG26807 &&
2394 !nic_data->workaround_26807) {
2397 rc = efx_mcdi_set_workaround(efx,
2398 MC_CMD_WORKAROUND_BUG26807,
2403 1 << MC_CMD_WORKAROUND_EXT_OUT_FLR_DONE_LBN) {
2404 netif_info(efx, drv, efx->net_dev,
2405 "other functions on NIC have been reset\n");
2407 /* With MCFW v4.6.x and earlier, the
2408 * boot count will have incremented,
2409 * so re-read the warm_boot_count
2410 * value now to ensure this function
2411 * doesn't think it has changed next
2414 rc = efx_ef10_get_warm_boot_count(efx);
2416 nic_data->warm_boot_count = rc;
2420 nic_data->workaround_26807 = true;
2421 } else if (rc == -EPERM) {
2431 efx_ef10_ev_fini(channel);
2435 static void efx_ef10_ev_remove(struct efx_channel *channel)
2437 efx_nic_free_buffer(channel->efx, &channel->eventq.buf);
2440 static void efx_ef10_handle_rx_wrong_queue(struct efx_rx_queue *rx_queue,
2441 unsigned int rx_queue_label)
2443 struct efx_nic *efx = rx_queue->efx;
2445 netif_info(efx, hw, efx->net_dev,
2446 "rx event arrived on queue %d labeled as queue %u\n",
2447 efx_rx_queue_index(rx_queue), rx_queue_label);
2449 efx_schedule_reset(efx, RESET_TYPE_DISABLE);
2453 efx_ef10_handle_rx_bad_lbits(struct efx_rx_queue *rx_queue,
2454 unsigned int actual, unsigned int expected)
2456 unsigned int dropped = (actual - expected) & rx_queue->ptr_mask;
2457 struct efx_nic *efx = rx_queue->efx;
2459 netif_info(efx, hw, efx->net_dev,
2460 "dropped %d events (index=%d expected=%d)\n",
2461 dropped, actual, expected);
2463 efx_schedule_reset(efx, RESET_TYPE_DISABLE);
2466 /* partially received RX was aborted. clean up. */
2467 static void efx_ef10_handle_rx_abort(struct efx_rx_queue *rx_queue)
2469 unsigned int rx_desc_ptr;
2471 netif_dbg(rx_queue->efx, hw, rx_queue->efx->net_dev,
2472 "scattered RX aborted (dropping %u buffers)\n",
2473 rx_queue->scatter_n);
2475 rx_desc_ptr = rx_queue->removed_count & rx_queue->ptr_mask;
2477 efx_rx_packet(rx_queue, rx_desc_ptr, rx_queue->scatter_n,
2478 0, EFX_RX_PKT_DISCARD);
2480 rx_queue->removed_count += rx_queue->scatter_n;
2481 rx_queue->scatter_n = 0;
2482 rx_queue->scatter_len = 0;
2483 ++efx_rx_queue_channel(rx_queue)->n_rx_nodesc_trunc;
2486 static int efx_ef10_handle_rx_event(struct efx_channel *channel,
2487 const efx_qword_t *event)
2489 unsigned int rx_bytes, next_ptr_lbits, rx_queue_label, rx_l4_class;
2490 unsigned int n_descs, n_packets, i;
2491 struct efx_nic *efx = channel->efx;
2492 struct efx_rx_queue *rx_queue;
2496 if (unlikely(ACCESS_ONCE(efx->reset_pending)))
2499 /* Basic packet information */
2500 rx_bytes = EFX_QWORD_FIELD(*event, ESF_DZ_RX_BYTES);
2501 next_ptr_lbits = EFX_QWORD_FIELD(*event, ESF_DZ_RX_DSC_PTR_LBITS);
2502 rx_queue_label = EFX_QWORD_FIELD(*event, ESF_DZ_RX_QLABEL);
2503 rx_l4_class = EFX_QWORD_FIELD(*event, ESF_DZ_RX_L4_CLASS);
2504 rx_cont = EFX_QWORD_FIELD(*event, ESF_DZ_RX_CONT);
2506 if (EFX_QWORD_FIELD(*event, ESF_DZ_RX_DROP_EVENT))
2507 netdev_WARN(efx->net_dev, "saw RX_DROP_EVENT: event="
2509 EFX_QWORD_VAL(*event));
2511 rx_queue = efx_channel_get_rx_queue(channel);
2513 if (unlikely(rx_queue_label != efx_rx_queue_index(rx_queue)))
2514 efx_ef10_handle_rx_wrong_queue(rx_queue, rx_queue_label);
2516 n_descs = ((next_ptr_lbits - rx_queue->removed_count) &
2517 ((1 << ESF_DZ_RX_DSC_PTR_LBITS_WIDTH) - 1));
2519 if (n_descs != rx_queue->scatter_n + 1) {
2520 struct efx_ef10_nic_data *nic_data = efx->nic_data;
2522 /* detect rx abort */
2523 if (unlikely(n_descs == rx_queue->scatter_n)) {
2524 if (rx_queue->scatter_n == 0 || rx_bytes != 0)
2525 netdev_WARN(efx->net_dev,
2526 "invalid RX abort: scatter_n=%u event="
2528 rx_queue->scatter_n,
2529 EFX_QWORD_VAL(*event));
2530 efx_ef10_handle_rx_abort(rx_queue);
2534 /* Check that RX completion merging is valid, i.e.
2535 * the current firmware supports it and this is a
2536 * non-scattered packet.
2538 if (!(nic_data->datapath_caps &
2539 (1 << MC_CMD_GET_CAPABILITIES_OUT_RX_BATCHING_LBN)) ||
2540 rx_queue->scatter_n != 0 || rx_cont) {
2541 efx_ef10_handle_rx_bad_lbits(
2542 rx_queue, next_ptr_lbits,
2543 (rx_queue->removed_count +
2544 rx_queue->scatter_n + 1) &
2545 ((1 << ESF_DZ_RX_DSC_PTR_LBITS_WIDTH) - 1));
2549 /* Merged completion for multiple non-scattered packets */
2550 rx_queue->scatter_n = 1;
2551 rx_queue->scatter_len = 0;
2552 n_packets = n_descs;
2553 ++channel->n_rx_merge_events;
2554 channel->n_rx_merge_packets += n_packets;
2555 flags |= EFX_RX_PKT_PREFIX_LEN;
2557 ++rx_queue->scatter_n;
2558 rx_queue->scatter_len += rx_bytes;
2564 if (unlikely(EFX_QWORD_FIELD(*event, ESF_DZ_RX_ECRC_ERR)))
2565 flags |= EFX_RX_PKT_DISCARD;
2567 if (unlikely(EFX_QWORD_FIELD(*event, ESF_DZ_RX_IPCKSUM_ERR))) {
2568 channel->n_rx_ip_hdr_chksum_err += n_packets;
2569 } else if (unlikely(EFX_QWORD_FIELD(*event,
2570 ESF_DZ_RX_TCPUDP_CKSUM_ERR))) {
2571 channel->n_rx_tcp_udp_chksum_err += n_packets;
2572 } else if (rx_l4_class == ESE_DZ_L4_CLASS_TCP ||
2573 rx_l4_class == ESE_DZ_L4_CLASS_UDP) {
2574 flags |= EFX_RX_PKT_CSUMMED;
2577 if (rx_l4_class == ESE_DZ_L4_CLASS_TCP)
2578 flags |= EFX_RX_PKT_TCP;
2580 channel->irq_mod_score += 2 * n_packets;
2582 /* Handle received packet(s) */
2583 for (i = 0; i < n_packets; i++) {
2584 efx_rx_packet(rx_queue,
2585 rx_queue->removed_count & rx_queue->ptr_mask,
2586 rx_queue->scatter_n, rx_queue->scatter_len,
2588 rx_queue->removed_count += rx_queue->scatter_n;
2591 rx_queue->scatter_n = 0;
2592 rx_queue->scatter_len = 0;
2598 efx_ef10_handle_tx_event(struct efx_channel *channel, efx_qword_t *event)
2600 struct efx_nic *efx = channel->efx;
2601 struct efx_tx_queue *tx_queue;
2602 unsigned int tx_ev_desc_ptr;
2603 unsigned int tx_ev_q_label;
2606 if (unlikely(ACCESS_ONCE(efx->reset_pending)))
2609 if (unlikely(EFX_QWORD_FIELD(*event, ESF_DZ_TX_DROP_EVENT)))
2612 /* Transmit completion */
2613 tx_ev_desc_ptr = EFX_QWORD_FIELD(*event, ESF_DZ_TX_DESCR_INDX);
2614 tx_ev_q_label = EFX_QWORD_FIELD(*event, ESF_DZ_TX_QLABEL);
2615 tx_queue = efx_channel_get_tx_queue(channel,
2616 tx_ev_q_label % EFX_TXQ_TYPES);
2617 tx_descs = ((tx_ev_desc_ptr + 1 - tx_queue->read_count) &
2618 tx_queue->ptr_mask);
2619 efx_xmit_done(tx_queue, tx_ev_desc_ptr & tx_queue->ptr_mask);
2625 efx_ef10_handle_driver_event(struct efx_channel *channel, efx_qword_t *event)
2627 struct efx_nic *efx = channel->efx;
2630 subcode = EFX_QWORD_FIELD(*event, ESF_DZ_DRV_SUB_CODE);
2633 case ESE_DZ_DRV_TIMER_EV:
2634 case ESE_DZ_DRV_WAKE_UP_EV:
2636 case ESE_DZ_DRV_START_UP_EV:
2637 /* event queue init complete. ok. */
2640 netif_err(efx, hw, efx->net_dev,
2641 "channel %d unknown driver event type %d"
2642 " (data " EFX_QWORD_FMT ")\n",
2643 channel->channel, subcode,
2644 EFX_QWORD_VAL(*event));
2649 static void efx_ef10_handle_driver_generated_event(struct efx_channel *channel,
2652 struct efx_nic *efx = channel->efx;
2655 subcode = EFX_QWORD_FIELD(*event, EFX_DWORD_0);
2659 channel->event_test_cpu = raw_smp_processor_id();
2661 case EFX_EF10_REFILL:
2662 /* The queue must be empty, so we won't receive any rx
2663 * events, so efx_process_channel() won't refill the
2664 * queue. Refill it here
2666 efx_fast_push_rx_descriptors(&channel->rx_queue, true);
2669 netif_err(efx, hw, efx->net_dev,
2670 "channel %d unknown driver event type %u"
2671 " (data " EFX_QWORD_FMT ")\n",
2672 channel->channel, (unsigned) subcode,
2673 EFX_QWORD_VAL(*event));
2677 static int efx_ef10_ev_process(struct efx_channel *channel, int quota)
2679 struct efx_nic *efx = channel->efx;
2680 efx_qword_t event, *p_event;
2681 unsigned int read_ptr;
2689 read_ptr = channel->eventq_read_ptr;
2692 p_event = efx_event(channel, read_ptr);
2695 if (!efx_event_present(&event))
2698 EFX_SET_QWORD(*p_event);
2702 ev_code = EFX_QWORD_FIELD(event, ESF_DZ_EV_CODE);
2704 netif_vdbg(efx, drv, efx->net_dev,
2705 "processing event on %d " EFX_QWORD_FMT "\n",
2706 channel->channel, EFX_QWORD_VAL(event));
2709 case ESE_DZ_EV_CODE_MCDI_EV:
2710 efx_mcdi_process_event(channel, &event);
2712 case ESE_DZ_EV_CODE_RX_EV:
2713 spent += efx_ef10_handle_rx_event(channel, &event);
2714 if (spent >= quota) {
2715 /* XXX can we split a merged event to
2716 * avoid going over-quota?
2722 case ESE_DZ_EV_CODE_TX_EV:
2723 tx_descs += efx_ef10_handle_tx_event(channel, &event);
2724 if (tx_descs > efx->txq_entries) {
2727 } else if (++spent == quota) {
2731 case ESE_DZ_EV_CODE_DRIVER_EV:
2732 efx_ef10_handle_driver_event(channel, &event);
2733 if (++spent == quota)
2736 case EFX_EF10_DRVGEN_EV:
2737 efx_ef10_handle_driver_generated_event(channel, &event);
2740 netif_err(efx, hw, efx->net_dev,
2741 "channel %d unknown event type %d"
2742 " (data " EFX_QWORD_FMT ")\n",
2743 channel->channel, ev_code,
2744 EFX_QWORD_VAL(event));
2749 channel->eventq_read_ptr = read_ptr;
2753 static void efx_ef10_ev_read_ack(struct efx_channel *channel)
2755 struct efx_nic *efx = channel->efx;
2758 if (EFX_EF10_WORKAROUND_35388(efx)) {
2759 BUILD_BUG_ON(EFX_MIN_EVQ_SIZE <
2760 (1 << ERF_DD_EVQ_IND_RPTR_WIDTH));
2761 BUILD_BUG_ON(EFX_MAX_EVQ_SIZE >
2762 (1 << 2 * ERF_DD_EVQ_IND_RPTR_WIDTH));
2764 EFX_POPULATE_DWORD_2(rptr, ERF_DD_EVQ_IND_RPTR_FLAGS,
2765 EFE_DD_EVQ_IND_RPTR_FLAGS_HIGH,
2766 ERF_DD_EVQ_IND_RPTR,
2767 (channel->eventq_read_ptr &
2768 channel->eventq_mask) >>
2769 ERF_DD_EVQ_IND_RPTR_WIDTH);
2770 efx_writed_page(efx, &rptr, ER_DD_EVQ_INDIRECT,
2772 EFX_POPULATE_DWORD_2(rptr, ERF_DD_EVQ_IND_RPTR_FLAGS,
2773 EFE_DD_EVQ_IND_RPTR_FLAGS_LOW,
2774 ERF_DD_EVQ_IND_RPTR,
2775 channel->eventq_read_ptr &
2776 ((1 << ERF_DD_EVQ_IND_RPTR_WIDTH) - 1));
2777 efx_writed_page(efx, &rptr, ER_DD_EVQ_INDIRECT,
2780 EFX_POPULATE_DWORD_1(rptr, ERF_DZ_EVQ_RPTR,
2781 channel->eventq_read_ptr &
2782 channel->eventq_mask);
2783 efx_writed_page(efx, &rptr, ER_DZ_EVQ_RPTR, channel->channel);
2787 static void efx_ef10_ev_test_generate(struct efx_channel *channel)
2789 MCDI_DECLARE_BUF(inbuf, MC_CMD_DRIVER_EVENT_IN_LEN);
2790 struct efx_nic *efx = channel->efx;
2794 EFX_POPULATE_QWORD_2(event,
2795 ESF_DZ_EV_CODE, EFX_EF10_DRVGEN_EV,
2796 ESF_DZ_EV_DATA, EFX_EF10_TEST);
2798 MCDI_SET_DWORD(inbuf, DRIVER_EVENT_IN_EVQ, channel->channel);
2800 /* MCDI_SET_QWORD is not appropriate here since EFX_POPULATE_* has
2801 * already swapped the data to little-endian order.
2803 memcpy(MCDI_PTR(inbuf, DRIVER_EVENT_IN_DATA), &event.u64[0],
2804 sizeof(efx_qword_t));
2806 rc = efx_mcdi_rpc(efx, MC_CMD_DRIVER_EVENT, inbuf, sizeof(inbuf),
2815 netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
2818 void efx_ef10_handle_drain_event(struct efx_nic *efx)
2820 if (atomic_dec_and_test(&efx->active_queues))
2821 wake_up(&efx->flush_wq);
2823 WARN_ON(atomic_read(&efx->active_queues) < 0);
2826 static int efx_ef10_fini_dmaq(struct efx_nic *efx)
2828 struct efx_ef10_nic_data *nic_data = efx->nic_data;
2829 struct efx_channel *channel;
2830 struct efx_tx_queue *tx_queue;
2831 struct efx_rx_queue *rx_queue;
2834 /* If the MC has just rebooted, the TX/RX queues will have already been
2835 * torn down, but efx->active_queues needs to be set to zero.
2837 if (nic_data->must_realloc_vis) {
2838 atomic_set(&efx->active_queues, 0);
2842 /* Do not attempt to write to the NIC during EEH recovery */
2843 if (efx->state != STATE_RECOVERY) {
2844 efx_for_each_channel(channel, efx) {
2845 efx_for_each_channel_rx_queue(rx_queue, channel)
2846 efx_ef10_rx_fini(rx_queue);
2847 efx_for_each_channel_tx_queue(tx_queue, channel)
2848 efx_ef10_tx_fini(tx_queue);
2851 wait_event_timeout(efx->flush_wq,
2852 atomic_read(&efx->active_queues) == 0,
2853 msecs_to_jiffies(EFX_MAX_FLUSH_TIME));
2854 pending = atomic_read(&efx->active_queues);
2856 netif_err(efx, hw, efx->net_dev, "failed to flush %d queues\n",
2865 static void efx_ef10_prepare_flr(struct efx_nic *efx)
2867 atomic_set(&efx->active_queues, 0);
2870 static bool efx_ef10_filter_equal(const struct efx_filter_spec *left,
2871 const struct efx_filter_spec *right)
2873 if ((left->match_flags ^ right->match_flags) |
2874 ((left->flags ^ right->flags) &
2875 (EFX_FILTER_FLAG_RX | EFX_FILTER_FLAG_TX)))
2878 return memcmp(&left->outer_vid, &right->outer_vid,
2879 sizeof(struct efx_filter_spec) -
2880 offsetof(struct efx_filter_spec, outer_vid)) == 0;
2883 static unsigned int efx_ef10_filter_hash(const struct efx_filter_spec *spec)
2885 BUILD_BUG_ON(offsetof(struct efx_filter_spec, outer_vid) & 3);
2886 return jhash2((const u32 *)&spec->outer_vid,
2887 (sizeof(struct efx_filter_spec) -
2888 offsetof(struct efx_filter_spec, outer_vid)) / 4,
2890 /* XXX should we randomise the initval? */
2893 /* Decide whether a filter should be exclusive or else should allow
2894 * delivery to additional recipients. Currently we decide that
2895 * filters for specific local unicast MAC and IP addresses are
2898 static bool efx_ef10_filter_is_exclusive(const struct efx_filter_spec *spec)
2900 if (spec->match_flags & EFX_FILTER_MATCH_LOC_MAC &&
2901 !is_multicast_ether_addr(spec->loc_mac))
2904 if ((spec->match_flags &
2905 (EFX_FILTER_MATCH_ETHER_TYPE | EFX_FILTER_MATCH_LOC_HOST)) ==
2906 (EFX_FILTER_MATCH_ETHER_TYPE | EFX_FILTER_MATCH_LOC_HOST)) {
2907 if (spec->ether_type == htons(ETH_P_IP) &&
2908 !ipv4_is_multicast(spec->loc_host[0]))
2910 if (spec->ether_type == htons(ETH_P_IPV6) &&
2911 ((const u8 *)spec->loc_host)[0] != 0xff)
2918 static struct efx_filter_spec *
2919 efx_ef10_filter_entry_spec(const struct efx_ef10_filter_table *table,
2920 unsigned int filter_idx)
2922 return (struct efx_filter_spec *)(table->entry[filter_idx].spec &
2923 ~EFX_EF10_FILTER_FLAGS);
2927 efx_ef10_filter_entry_flags(const struct efx_ef10_filter_table *table,
2928 unsigned int filter_idx)
2930 return table->entry[filter_idx].spec & EFX_EF10_FILTER_FLAGS;
2934 efx_ef10_filter_set_entry(struct efx_ef10_filter_table *table,
2935 unsigned int filter_idx,
2936 const struct efx_filter_spec *spec,
2939 table->entry[filter_idx].spec = (unsigned long)spec | flags;
2942 static void efx_ef10_filter_push_prep(struct efx_nic *efx,
2943 const struct efx_filter_spec *spec,
2944 efx_dword_t *inbuf, u64 handle,
2947 struct efx_ef10_nic_data *nic_data = efx->nic_data;
2948 u32 flags = spec->flags;
2950 memset(inbuf, 0, MC_CMD_FILTER_OP_IN_LEN);
2952 /* Remove RSS flag if we don't have an RSS context. */
2953 if (flags & EFX_FILTER_FLAG_RX_RSS &&
2954 spec->rss_context == EFX_FILTER_RSS_CONTEXT_DEFAULT &&
2955 nic_data->rx_rss_context == EFX_EF10_RSS_CONTEXT_INVALID)
2956 flags &= ~EFX_FILTER_FLAG_RX_RSS;
2959 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
2960 MC_CMD_FILTER_OP_IN_OP_REPLACE);
2961 MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE, handle);
2963 u32 match_fields = 0;
2965 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
2966 efx_ef10_filter_is_exclusive(spec) ?
2967 MC_CMD_FILTER_OP_IN_OP_INSERT :
2968 MC_CMD_FILTER_OP_IN_OP_SUBSCRIBE);
2970 /* Convert match flags and values. Unlike almost
2971 * everything else in MCDI, these fields are in
2972 * network byte order.
2974 if (spec->match_flags & EFX_FILTER_MATCH_LOC_MAC_IG)
2976 is_multicast_ether_addr(spec->loc_mac) ?
2977 1 << MC_CMD_FILTER_OP_IN_MATCH_UNKNOWN_MCAST_DST_LBN :
2978 1 << MC_CMD_FILTER_OP_IN_MATCH_UNKNOWN_UCAST_DST_LBN;
2979 #define COPY_FIELD(gen_flag, gen_field, mcdi_field) \
2980 if (spec->match_flags & EFX_FILTER_MATCH_ ## gen_flag) { \
2982 1 << MC_CMD_FILTER_OP_IN_MATCH_ ## \
2983 mcdi_field ## _LBN; \
2985 MC_CMD_FILTER_OP_IN_ ## mcdi_field ## _LEN < \
2986 sizeof(spec->gen_field)); \
2987 memcpy(MCDI_PTR(inbuf, FILTER_OP_IN_ ## mcdi_field), \
2988 &spec->gen_field, sizeof(spec->gen_field)); \
2990 COPY_FIELD(REM_HOST, rem_host, SRC_IP);
2991 COPY_FIELD(LOC_HOST, loc_host, DST_IP);
2992 COPY_FIELD(REM_MAC, rem_mac, SRC_MAC);
2993 COPY_FIELD(REM_PORT, rem_port, SRC_PORT);
2994 COPY_FIELD(LOC_MAC, loc_mac, DST_MAC);
2995 COPY_FIELD(LOC_PORT, loc_port, DST_PORT);
2996 COPY_FIELD(ETHER_TYPE, ether_type, ETHER_TYPE);
2997 COPY_FIELD(INNER_VID, inner_vid, INNER_VLAN);
2998 COPY_FIELD(OUTER_VID, outer_vid, OUTER_VLAN);
2999 COPY_FIELD(IP_PROTO, ip_proto, IP_PROTO);
3001 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_MATCH_FIELDS,
3005 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_PORT_ID, nic_data->vport_id);
3006 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_DEST,
3007 spec->dmaq_id == EFX_FILTER_RX_DMAQ_ID_DROP ?
3008 MC_CMD_FILTER_OP_IN_RX_DEST_DROP :
3009 MC_CMD_FILTER_OP_IN_RX_DEST_HOST);
3010 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_TX_DOMAIN, 0);
3011 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_TX_DEST,
3012 MC_CMD_FILTER_OP_IN_TX_DEST_DEFAULT);
3013 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_QUEUE,
3014 spec->dmaq_id == EFX_FILTER_RX_DMAQ_ID_DROP ?
3016 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_MODE,
3017 (flags & EFX_FILTER_FLAG_RX_RSS) ?
3018 MC_CMD_FILTER_OP_IN_RX_MODE_RSS :
3019 MC_CMD_FILTER_OP_IN_RX_MODE_SIMPLE);
3020 if (flags & EFX_FILTER_FLAG_RX_RSS)
3021 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_CONTEXT,
3022 spec->rss_context !=
3023 EFX_FILTER_RSS_CONTEXT_DEFAULT ?
3024 spec->rss_context : nic_data->rx_rss_context);
3027 static int efx_ef10_filter_push(struct efx_nic *efx,
3028 const struct efx_filter_spec *spec,
3029 u64 *handle, bool replacing)
3031 MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_IN_LEN);
3032 MCDI_DECLARE_BUF(outbuf, MC_CMD_FILTER_OP_OUT_LEN);
3035 efx_ef10_filter_push_prep(efx, spec, inbuf, *handle, replacing);
3036 rc = efx_mcdi_rpc(efx, MC_CMD_FILTER_OP, inbuf, sizeof(inbuf),
3037 outbuf, sizeof(outbuf), NULL);
3039 *handle = MCDI_QWORD(outbuf, FILTER_OP_OUT_HANDLE);
3041 rc = -EBUSY; /* to match efx_farch_filter_insert() */
3045 static int efx_ef10_filter_rx_match_pri(struct efx_ef10_filter_table *table,
3046 enum efx_filter_match_flags match_flags)
3048 unsigned int match_pri;
3051 match_pri < table->rx_match_count;
3053 if (table->rx_match_flags[match_pri] == match_flags)
3056 return -EPROTONOSUPPORT;
3059 static s32 efx_ef10_filter_insert(struct efx_nic *efx,
3060 struct efx_filter_spec *spec,
3063 struct efx_ef10_filter_table *table = efx->filter_state;
3064 DECLARE_BITMAP(mc_rem_map, EFX_EF10_FILTER_SEARCH_LIMIT);
3065 struct efx_filter_spec *saved_spec;
3066 unsigned int match_pri, hash;
3067 unsigned int priv_flags;
3068 bool replacing = false;
3074 /* For now, only support RX filters */
3075 if ((spec->flags & (EFX_FILTER_FLAG_RX | EFX_FILTER_FLAG_TX)) !=
3079 rc = efx_ef10_filter_rx_match_pri(table, spec->match_flags);
3084 hash = efx_ef10_filter_hash(spec);
3085 is_mc_recip = efx_filter_is_mc_recipient(spec);
3087 bitmap_zero(mc_rem_map, EFX_EF10_FILTER_SEARCH_LIMIT);
3089 /* Find any existing filters with the same match tuple or
3090 * else a free slot to insert at. If any of them are busy,
3091 * we have to wait and retry.
3094 unsigned int depth = 1;
3097 spin_lock_bh(&efx->filter_lock);
3100 i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
3101 saved_spec = efx_ef10_filter_entry_spec(table, i);
3106 } else if (efx_ef10_filter_equal(spec, saved_spec)) {
3107 if (table->entry[i].spec &
3108 EFX_EF10_FILTER_FLAG_BUSY)
3110 if (spec->priority < saved_spec->priority &&
3111 spec->priority != EFX_FILTER_PRI_AUTO) {
3116 /* This is the only one */
3117 if (spec->priority ==
3118 saved_spec->priority &&
3125 } else if (spec->priority >
3126 saved_spec->priority ||
3128 saved_spec->priority &&
3133 __set_bit(depth, mc_rem_map);
3137 /* Once we reach the maximum search depth, use
3138 * the first suitable slot or return -EBUSY if
3141 if (depth == EFX_EF10_FILTER_SEARCH_LIMIT) {
3142 if (ins_index < 0) {
3152 prepare_to_wait(&table->waitq, &wait, TASK_UNINTERRUPTIBLE);
3153 spin_unlock_bh(&efx->filter_lock);
3158 /* Create a software table entry if necessary, and mark it
3159 * busy. We might yet fail to insert, but any attempt to
3160 * insert a conflicting filter while we're waiting for the
3161 * firmware must find the busy entry.
3163 saved_spec = efx_ef10_filter_entry_spec(table, ins_index);
3165 if (spec->priority == EFX_FILTER_PRI_AUTO &&
3166 saved_spec->priority >= EFX_FILTER_PRI_AUTO) {
3167 /* Just make sure it won't be removed */
3168 if (saved_spec->priority > EFX_FILTER_PRI_AUTO)
3169 saved_spec->flags |= EFX_FILTER_FLAG_RX_OVER_AUTO;
3170 table->entry[ins_index].spec &=
3171 ~EFX_EF10_FILTER_FLAG_AUTO_OLD;
3176 priv_flags = efx_ef10_filter_entry_flags(table, ins_index);
3178 saved_spec = kmalloc(sizeof(*spec), GFP_ATOMIC);
3183 *saved_spec = *spec;
3186 efx_ef10_filter_set_entry(table, ins_index, saved_spec,
3187 priv_flags | EFX_EF10_FILTER_FLAG_BUSY);
3189 /* Mark lower-priority multicast recipients busy prior to removal */
3191 unsigned int depth, i;
3193 for (depth = 0; depth < EFX_EF10_FILTER_SEARCH_LIMIT; depth++) {
3194 i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
3195 if (test_bit(depth, mc_rem_map))
3196 table->entry[i].spec |=
3197 EFX_EF10_FILTER_FLAG_BUSY;
3201 spin_unlock_bh(&efx->filter_lock);
3203 rc = efx_ef10_filter_push(efx, spec, &table->entry[ins_index].handle,
3206 /* Finalise the software table entry */
3207 spin_lock_bh(&efx->filter_lock);
3210 /* Update the fields that may differ */
3211 if (saved_spec->priority == EFX_FILTER_PRI_AUTO)
3212 saved_spec->flags |=
3213 EFX_FILTER_FLAG_RX_OVER_AUTO;
3214 saved_spec->priority = spec->priority;
3215 saved_spec->flags &= EFX_FILTER_FLAG_RX_OVER_AUTO;
3216 saved_spec->flags |= spec->flags;
3217 saved_spec->rss_context = spec->rss_context;
3218 saved_spec->dmaq_id = spec->dmaq_id;
3220 } else if (!replacing) {
3224 efx_ef10_filter_set_entry(table, ins_index, saved_spec, priv_flags);
3226 /* Remove and finalise entries for lower-priority multicast
3230 MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_IN_LEN);
3231 unsigned int depth, i;
3233 memset(inbuf, 0, sizeof(inbuf));
3235 for (depth = 0; depth < EFX_EF10_FILTER_SEARCH_LIMIT; depth++) {
3236 if (!test_bit(depth, mc_rem_map))
3239 i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
3240 saved_spec = efx_ef10_filter_entry_spec(table, i);
3241 priv_flags = efx_ef10_filter_entry_flags(table, i);
3244 spin_unlock_bh(&efx->filter_lock);
3245 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
3246 MC_CMD_FILTER_OP_IN_OP_UNSUBSCRIBE);
3247 MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
3248 table->entry[i].handle);
3249 rc = efx_mcdi_rpc(efx, MC_CMD_FILTER_OP,
3250 inbuf, sizeof(inbuf),
3252 spin_lock_bh(&efx->filter_lock);
3260 priv_flags &= ~EFX_EF10_FILTER_FLAG_BUSY;
3262 efx_ef10_filter_set_entry(table, i, saved_spec,
3267 /* If successful, return the inserted filter ID */
3269 rc = match_pri * HUNT_FILTER_TBL_ROWS + ins_index;
3271 wake_up_all(&table->waitq);
3273 spin_unlock_bh(&efx->filter_lock);
3274 finish_wait(&table->waitq, &wait);
3278 static void efx_ef10_filter_update_rx_scatter(struct efx_nic *efx)
3280 /* no need to do anything here on EF10 */
3284 * If !by_index, remove by ID
3285 * If by_index, remove by index
3286 * Filter ID may come from userland and must be range-checked.
3288 static int efx_ef10_filter_remove_internal(struct efx_nic *efx,
3289 unsigned int priority_mask,
3290 u32 filter_id, bool by_index)
3292 unsigned int filter_idx = filter_id % HUNT_FILTER_TBL_ROWS;
3293 struct efx_ef10_filter_table *table = efx->filter_state;
3294 MCDI_DECLARE_BUF(inbuf,
3295 MC_CMD_FILTER_OP_IN_HANDLE_OFST +
3296 MC_CMD_FILTER_OP_IN_HANDLE_LEN);
3297 struct efx_filter_spec *spec;
3301 /* Find the software table entry and mark it busy. Don't
3302 * remove it yet; any attempt to update while we're waiting
3303 * for the firmware must find the busy entry.
3306 spin_lock_bh(&efx->filter_lock);
3307 if (!(table->entry[filter_idx].spec &
3308 EFX_EF10_FILTER_FLAG_BUSY))
3310 prepare_to_wait(&table->waitq, &wait, TASK_UNINTERRUPTIBLE);
3311 spin_unlock_bh(&efx->filter_lock);
3315 spec = efx_ef10_filter_entry_spec(table, filter_idx);
3318 efx_ef10_filter_rx_match_pri(table, spec->match_flags) !=
3319 filter_id / HUNT_FILTER_TBL_ROWS)) {
3324 if (spec->flags & EFX_FILTER_FLAG_RX_OVER_AUTO &&
3325 priority_mask == (1U << EFX_FILTER_PRI_AUTO)) {
3326 /* Just remove flags */
3327 spec->flags &= ~EFX_FILTER_FLAG_RX_OVER_AUTO;
3328 table->entry[filter_idx].spec &= ~EFX_EF10_FILTER_FLAG_AUTO_OLD;
3333 if (!(priority_mask & (1U << spec->priority))) {
3338 table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_BUSY;
3339 spin_unlock_bh(&efx->filter_lock);
3341 if (spec->flags & EFX_FILTER_FLAG_RX_OVER_AUTO) {
3342 /* Reset to an automatic filter */
3344 struct efx_filter_spec new_spec = *spec;
3346 new_spec.priority = EFX_FILTER_PRI_AUTO;
3347 new_spec.flags = (EFX_FILTER_FLAG_RX |
3348 (efx_rss_enabled(efx) ?
3349 EFX_FILTER_FLAG_RX_RSS : 0));
3350 new_spec.dmaq_id = 0;
3351 new_spec.rss_context = EFX_FILTER_RSS_CONTEXT_DEFAULT;
3352 rc = efx_ef10_filter_push(efx, &new_spec,
3353 &table->entry[filter_idx].handle,
3356 spin_lock_bh(&efx->filter_lock);
3360 /* Really remove the filter */
3362 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
3363 efx_ef10_filter_is_exclusive(spec) ?
3364 MC_CMD_FILTER_OP_IN_OP_REMOVE :
3365 MC_CMD_FILTER_OP_IN_OP_UNSUBSCRIBE);
3366 MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
3367 table->entry[filter_idx].handle);
3368 rc = efx_mcdi_rpc(efx, MC_CMD_FILTER_OP,
3369 inbuf, sizeof(inbuf), NULL, 0, NULL);
3371 spin_lock_bh(&efx->filter_lock);
3374 efx_ef10_filter_set_entry(table, filter_idx, NULL, 0);
3378 table->entry[filter_idx].spec &= ~EFX_EF10_FILTER_FLAG_BUSY;
3379 wake_up_all(&table->waitq);
3381 spin_unlock_bh(&efx->filter_lock);
3382 finish_wait(&table->waitq, &wait);
3386 static int efx_ef10_filter_remove_safe(struct efx_nic *efx,
3387 enum efx_filter_priority priority,
3390 return efx_ef10_filter_remove_internal(efx, 1U << priority,
3394 static u32 efx_ef10_filter_get_unsafe_id(struct efx_nic *efx, u32 filter_id)
3396 return filter_id % HUNT_FILTER_TBL_ROWS;
3399 static int efx_ef10_filter_remove_unsafe(struct efx_nic *efx,
3400 enum efx_filter_priority priority,
3403 return efx_ef10_filter_remove_internal(efx, 1U << priority,
3407 static int efx_ef10_filter_get_safe(struct efx_nic *efx,
3408 enum efx_filter_priority priority,
3409 u32 filter_id, struct efx_filter_spec *spec)
3411 unsigned int filter_idx = filter_id % HUNT_FILTER_TBL_ROWS;
3412 struct efx_ef10_filter_table *table = efx->filter_state;
3413 const struct efx_filter_spec *saved_spec;
3416 spin_lock_bh(&efx->filter_lock);
3417 saved_spec = efx_ef10_filter_entry_spec(table, filter_idx);
3418 if (saved_spec && saved_spec->priority == priority &&
3419 efx_ef10_filter_rx_match_pri(table, saved_spec->match_flags) ==
3420 filter_id / HUNT_FILTER_TBL_ROWS) {
3421 *spec = *saved_spec;
3426 spin_unlock_bh(&efx->filter_lock);
3430 static int efx_ef10_filter_clear_rx(struct efx_nic *efx,
3431 enum efx_filter_priority priority)
3433 unsigned int priority_mask;
3437 priority_mask = (((1U << (priority + 1)) - 1) &
3438 ~(1U << EFX_FILTER_PRI_AUTO));
3440 for (i = 0; i < HUNT_FILTER_TBL_ROWS; i++) {
3441 rc = efx_ef10_filter_remove_internal(efx, priority_mask,
3443 if (rc && rc != -ENOENT)
3450 static u32 efx_ef10_filter_count_rx_used(struct efx_nic *efx,
3451 enum efx_filter_priority priority)
3453 struct efx_ef10_filter_table *table = efx->filter_state;
3454 unsigned int filter_idx;
3457 spin_lock_bh(&efx->filter_lock);
3458 for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
3459 if (table->entry[filter_idx].spec &&
3460 efx_ef10_filter_entry_spec(table, filter_idx)->priority ==
3464 spin_unlock_bh(&efx->filter_lock);
3468 static u32 efx_ef10_filter_get_rx_id_limit(struct efx_nic *efx)
3470 struct efx_ef10_filter_table *table = efx->filter_state;
3472 return table->rx_match_count * HUNT_FILTER_TBL_ROWS;
3475 static s32 efx_ef10_filter_get_rx_ids(struct efx_nic *efx,
3476 enum efx_filter_priority priority,
3479 struct efx_ef10_filter_table *table = efx->filter_state;
3480 struct efx_filter_spec *spec;
3481 unsigned int filter_idx;
3484 spin_lock_bh(&efx->filter_lock);
3485 for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
3486 spec = efx_ef10_filter_entry_spec(table, filter_idx);
3487 if (spec && spec->priority == priority) {
3488 if (count == size) {
3492 buf[count++] = (efx_ef10_filter_rx_match_pri(
3493 table, spec->match_flags) *
3494 HUNT_FILTER_TBL_ROWS +
3498 spin_unlock_bh(&efx->filter_lock);
3502 #ifdef CONFIG_RFS_ACCEL
3504 static efx_mcdi_async_completer efx_ef10_filter_rfs_insert_complete;
3506 static s32 efx_ef10_filter_rfs_insert(struct efx_nic *efx,
3507 struct efx_filter_spec *spec)
3509 struct efx_ef10_filter_table *table = efx->filter_state;
3510 MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_IN_LEN);
3511 struct efx_filter_spec *saved_spec;
3512 unsigned int hash, i, depth = 1;
3513 bool replacing = false;
3518 /* Must be an RX filter without RSS and not for a multicast
3519 * destination address (RFS only works for connected sockets).
3520 * These restrictions allow us to pass only a tiny amount of
3521 * data through to the completion function.
3523 EFX_WARN_ON_PARANOID(spec->flags !=
3524 (EFX_FILTER_FLAG_RX | EFX_FILTER_FLAG_RX_SCATTER));
3525 EFX_WARN_ON_PARANOID(spec->priority != EFX_FILTER_PRI_HINT);
3526 EFX_WARN_ON_PARANOID(efx_filter_is_mc_recipient(spec));
3528 hash = efx_ef10_filter_hash(spec);
3530 spin_lock_bh(&efx->filter_lock);
3532 /* Find any existing filter with the same match tuple or else
3533 * a free slot to insert at. If an existing filter is busy,
3534 * we have to give up.
3537 i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
3538 saved_spec = efx_ef10_filter_entry_spec(table, i);
3543 } else if (efx_ef10_filter_equal(spec, saved_spec)) {
3544 if (table->entry[i].spec & EFX_EF10_FILTER_FLAG_BUSY) {
3548 if (spec->priority < saved_spec->priority) {
3556 /* Once we reach the maximum search depth, use the
3557 * first suitable slot or return -EBUSY if there was
3560 if (depth == EFX_EF10_FILTER_SEARCH_LIMIT) {
3561 if (ins_index < 0) {
3571 /* Create a software table entry if necessary, and mark it
3572 * busy. We might yet fail to insert, but any attempt to
3573 * insert a conflicting filter while we're waiting for the
3574 * firmware must find the busy entry.
3576 saved_spec = efx_ef10_filter_entry_spec(table, ins_index);
3580 saved_spec = kmalloc(sizeof(*spec), GFP_ATOMIC);
3585 *saved_spec = *spec;
3587 efx_ef10_filter_set_entry(table, ins_index, saved_spec,
3588 EFX_EF10_FILTER_FLAG_BUSY);
3590 spin_unlock_bh(&efx->filter_lock);
3592 /* Pack up the variables needed on completion */
3593 cookie = replacing << 31 | ins_index << 16 | spec->dmaq_id;
3595 efx_ef10_filter_push_prep(efx, spec, inbuf,
3596 table->entry[ins_index].handle, replacing);
3597 efx_mcdi_rpc_async(efx, MC_CMD_FILTER_OP, inbuf, sizeof(inbuf),
3598 MC_CMD_FILTER_OP_OUT_LEN,
3599 efx_ef10_filter_rfs_insert_complete, cookie);
3604 spin_unlock_bh(&efx->filter_lock);
3609 efx_ef10_filter_rfs_insert_complete(struct efx_nic *efx, unsigned long cookie,
3610 int rc, efx_dword_t *outbuf,
3611 size_t outlen_actual)
3613 struct efx_ef10_filter_table *table = efx->filter_state;
3614 unsigned int ins_index, dmaq_id;
3615 struct efx_filter_spec *spec;
3618 /* Unpack the cookie */
3619 replacing = cookie >> 31;
3620 ins_index = (cookie >> 16) & (HUNT_FILTER_TBL_ROWS - 1);
3621 dmaq_id = cookie & 0xffff;
3623 spin_lock_bh(&efx->filter_lock);
3624 spec = efx_ef10_filter_entry_spec(table, ins_index);
3626 table->entry[ins_index].handle =
3627 MCDI_QWORD(outbuf, FILTER_OP_OUT_HANDLE);
3629 spec->dmaq_id = dmaq_id;
3630 } else if (!replacing) {
3634 efx_ef10_filter_set_entry(table, ins_index, spec, 0);
3635 spin_unlock_bh(&efx->filter_lock);
3637 wake_up_all(&table->waitq);
3641 efx_ef10_filter_rfs_expire_complete(struct efx_nic *efx,
3642 unsigned long filter_idx,
3643 int rc, efx_dword_t *outbuf,
3644 size_t outlen_actual);
3646 static bool efx_ef10_filter_rfs_expire_one(struct efx_nic *efx, u32 flow_id,
3647 unsigned int filter_idx)
3649 struct efx_ef10_filter_table *table = efx->filter_state;
3650 struct efx_filter_spec *spec =
3651 efx_ef10_filter_entry_spec(table, filter_idx);
3652 MCDI_DECLARE_BUF(inbuf,
3653 MC_CMD_FILTER_OP_IN_HANDLE_OFST +
3654 MC_CMD_FILTER_OP_IN_HANDLE_LEN);
3657 (table->entry[filter_idx].spec & EFX_EF10_FILTER_FLAG_BUSY) ||
3658 spec->priority != EFX_FILTER_PRI_HINT ||
3659 !rps_may_expire_flow(efx->net_dev, spec->dmaq_id,
3660 flow_id, filter_idx))
3663 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
3664 MC_CMD_FILTER_OP_IN_OP_REMOVE);
3665 MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
3666 table->entry[filter_idx].handle);
3667 if (efx_mcdi_rpc_async(efx, MC_CMD_FILTER_OP, inbuf, sizeof(inbuf), 0,
3668 efx_ef10_filter_rfs_expire_complete, filter_idx))
3671 table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_BUSY;
3676 efx_ef10_filter_rfs_expire_complete(struct efx_nic *efx,
3677 unsigned long filter_idx,
3678 int rc, efx_dword_t *outbuf,
3679 size_t outlen_actual)
3681 struct efx_ef10_filter_table *table = efx->filter_state;
3682 struct efx_filter_spec *spec =
3683 efx_ef10_filter_entry_spec(table, filter_idx);
3685 spin_lock_bh(&efx->filter_lock);
3688 efx_ef10_filter_set_entry(table, filter_idx, NULL, 0);
3690 table->entry[filter_idx].spec &= ~EFX_EF10_FILTER_FLAG_BUSY;
3691 wake_up_all(&table->waitq);
3692 spin_unlock_bh(&efx->filter_lock);
3695 #endif /* CONFIG_RFS_ACCEL */
3697 static int efx_ef10_filter_match_flags_from_mcdi(u32 mcdi_flags)
3699 int match_flags = 0;
3701 #define MAP_FLAG(gen_flag, mcdi_field) { \
3702 u32 old_mcdi_flags = mcdi_flags; \
3703 mcdi_flags &= ~(1 << MC_CMD_FILTER_OP_IN_MATCH_ ## \
3704 mcdi_field ## _LBN); \
3705 if (mcdi_flags != old_mcdi_flags) \
3706 match_flags |= EFX_FILTER_MATCH_ ## gen_flag; \
3708 MAP_FLAG(LOC_MAC_IG, UNKNOWN_UCAST_DST);
3709 MAP_FLAG(LOC_MAC_IG, UNKNOWN_MCAST_DST);
3710 MAP_FLAG(REM_HOST, SRC_IP);
3711 MAP_FLAG(LOC_HOST, DST_IP);
3712 MAP_FLAG(REM_MAC, SRC_MAC);
3713 MAP_FLAG(REM_PORT, SRC_PORT);
3714 MAP_FLAG(LOC_MAC, DST_MAC);
3715 MAP_FLAG(LOC_PORT, DST_PORT);
3716 MAP_FLAG(ETHER_TYPE, ETHER_TYPE);
3717 MAP_FLAG(INNER_VID, INNER_VLAN);
3718 MAP_FLAG(OUTER_VID, OUTER_VLAN);
3719 MAP_FLAG(IP_PROTO, IP_PROTO);
3722 /* Did we map them all? */
3729 static int efx_ef10_filter_table_probe(struct efx_nic *efx)
3731 MCDI_DECLARE_BUF(inbuf, MC_CMD_GET_PARSER_DISP_INFO_IN_LEN);
3732 MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_PARSER_DISP_INFO_OUT_LENMAX);
3733 unsigned int pd_match_pri, pd_match_count;
3734 struct efx_ef10_filter_table *table;
3739 if (!efx_rwsem_assert_write_locked(&efx->filter_sem))
3742 if (efx->filter_state) /* already probed */
3745 table = kzalloc(sizeof(*table), GFP_KERNEL);
3749 /* Find out which RX filter types are supported, and their priorities */
3750 MCDI_SET_DWORD(inbuf, GET_PARSER_DISP_INFO_IN_OP,
3751 MC_CMD_GET_PARSER_DISP_INFO_IN_OP_GET_SUPPORTED_RX_MATCHES);
3752 rc = efx_mcdi_rpc(efx, MC_CMD_GET_PARSER_DISP_INFO,
3753 inbuf, sizeof(inbuf), outbuf, sizeof(outbuf),
3757 pd_match_count = MCDI_VAR_ARRAY_LEN(
3758 outlen, GET_PARSER_DISP_INFO_OUT_SUPPORTED_MATCHES);
3759 table->rx_match_count = 0;
3761 for (pd_match_pri = 0; pd_match_pri < pd_match_count; pd_match_pri++) {
3765 GET_PARSER_DISP_INFO_OUT_SUPPORTED_MATCHES,
3767 rc = efx_ef10_filter_match_flags_from_mcdi(mcdi_flags);
3769 netif_dbg(efx, probe, efx->net_dev,
3770 "%s: fw flags %#x pri %u not supported in driver\n",
3771 __func__, mcdi_flags, pd_match_pri);
3773 netif_dbg(efx, probe, efx->net_dev,
3774 "%s: fw flags %#x pri %u supported as driver flags %#x pri %u\n",
3775 __func__, mcdi_flags, pd_match_pri,
3776 rc, table->rx_match_count);
3777 table->rx_match_flags[table->rx_match_count++] = rc;
3781 table->entry = vzalloc(HUNT_FILTER_TBL_ROWS * sizeof(*table->entry));
3782 if (!table->entry) {
3787 for (i = 0; i < ARRAY_SIZE(table->dev_uc_list); i++)
3788 table->dev_uc_list[i].id = EFX_EF10_FILTER_ID_INVALID;
3789 for (i = 0; i < ARRAY_SIZE(table->dev_mc_list); i++)
3790 table->dev_mc_list[i].id = EFX_EF10_FILTER_ID_INVALID;
3791 table->ucdef_id = EFX_EF10_FILTER_ID_INVALID;
3792 table->bcast_id = EFX_EF10_FILTER_ID_INVALID;
3793 table->mcdef_id = EFX_EF10_FILTER_ID_INVALID;
3794 table->mc_promisc_last = false;
3796 efx->filter_state = table;
3797 init_waitqueue_head(&table->waitq);
3805 /* Caller must hold efx->filter_sem for read if race against
3806 * efx_ef10_filter_table_remove() is possible
3808 static void efx_ef10_filter_table_restore(struct efx_nic *efx)
3810 struct efx_ef10_filter_table *table = efx->filter_state;
3811 struct efx_ef10_nic_data *nic_data = efx->nic_data;
3812 struct efx_filter_spec *spec;
3813 unsigned int filter_idx;
3814 bool failed = false;
3817 WARN_ON(!rwsem_is_locked(&efx->filter_sem));
3819 if (!nic_data->must_restore_filters)
3825 spin_lock_bh(&efx->filter_lock);
3827 for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
3828 spec = efx_ef10_filter_entry_spec(table, filter_idx);
3832 table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_BUSY;
3833 spin_unlock_bh(&efx->filter_lock);
3835 rc = efx_ef10_filter_push(efx, spec,
3836 &table->entry[filter_idx].handle,
3841 spin_lock_bh(&efx->filter_lock);
3844 efx_ef10_filter_set_entry(table, filter_idx, NULL, 0);
3846 table->entry[filter_idx].spec &=
3847 ~EFX_EF10_FILTER_FLAG_BUSY;
3851 spin_unlock_bh(&efx->filter_lock);
3854 netif_err(efx, hw, efx->net_dev,
3855 "unable to restore all filters\n");
3857 nic_data->must_restore_filters = false;
3860 static void efx_ef10_filter_table_remove(struct efx_nic *efx)
3862 struct efx_ef10_filter_table *table = efx->filter_state;
3863 MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_IN_LEN);
3864 struct efx_filter_spec *spec;
3865 unsigned int filter_idx;
3868 efx->filter_state = NULL;
3869 /* If we were called without locking, then it's not safe to free
3870 * the table as others might be using it. So we just WARN, leak
3871 * the memory, and potentially get an inconsistent filter table
3873 * This should never actually happen.
3875 if (!efx_rwsem_assert_write_locked(&efx->filter_sem))
3881 for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
3882 spec = efx_ef10_filter_entry_spec(table, filter_idx);
3886 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
3887 efx_ef10_filter_is_exclusive(spec) ?
3888 MC_CMD_FILTER_OP_IN_OP_REMOVE :
3889 MC_CMD_FILTER_OP_IN_OP_UNSUBSCRIBE);
3890 MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
3891 table->entry[filter_idx].handle);
3892 rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FILTER_OP, inbuf,
3893 sizeof(inbuf), NULL, 0, NULL);
3895 netif_info(efx, drv, efx->net_dev,
3896 "%s: filter %04x remove failed\n",
3897 __func__, filter_idx);
3901 vfree(table->entry);
3905 static void efx_ef10_filter_mark_one_old(struct efx_nic *efx, uint16_t *id)
3907 struct efx_ef10_filter_table *table = efx->filter_state;
3908 unsigned int filter_idx;
3910 if (*id != EFX_EF10_FILTER_ID_INVALID) {
3911 filter_idx = efx_ef10_filter_get_unsafe_id(efx, *id);
3912 if (!table->entry[filter_idx].spec)
3913 netif_dbg(efx, drv, efx->net_dev,
3914 "marked null spec old %04x:%04x\n", *id,
3916 table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_AUTO_OLD;
3917 *id = EFX_EF10_FILTER_ID_INVALID;
3921 static void efx_ef10_filter_mark_old(struct efx_nic *efx)
3923 struct efx_ef10_filter_table *table = efx->filter_state;
3929 /* Mark old filters that may need to be removed */
3930 spin_lock_bh(&efx->filter_lock);
3931 for (i = 0; i < table->dev_uc_count; i++)
3932 efx_ef10_filter_mark_one_old(efx, &table->dev_uc_list[i].id);
3933 for (i = 0; i < table->dev_mc_count; i++)
3934 efx_ef10_filter_mark_one_old(efx, &table->dev_mc_list[i].id);
3935 efx_ef10_filter_mark_one_old(efx, &table->ucdef_id);
3936 efx_ef10_filter_mark_one_old(efx, &table->bcast_id);
3937 efx_ef10_filter_mark_one_old(efx, &table->mcdef_id);
3938 spin_unlock_bh(&efx->filter_lock);
3941 static void efx_ef10_filter_uc_addr_list(struct efx_nic *efx, bool *promisc)
3943 struct efx_ef10_filter_table *table = efx->filter_state;
3944 struct net_device *net_dev = efx->net_dev;
3945 struct netdev_hw_addr *uc;
3949 addr_count = netdev_uc_count(net_dev);
3950 if (net_dev->flags & IFF_PROMISC)
3952 table->dev_uc_count = 1 + addr_count;
3953 ether_addr_copy(table->dev_uc_list[0].addr, net_dev->dev_addr);
3955 netdev_for_each_uc_addr(uc, net_dev) {
3956 if (i >= EFX_EF10_FILTER_DEV_UC_MAX) {
3960 ether_addr_copy(table->dev_uc_list[i].addr, uc->addr);
3965 static void efx_ef10_filter_mc_addr_list(struct efx_nic *efx, bool *promisc)
3967 struct efx_ef10_filter_table *table = efx->filter_state;
3968 struct net_device *net_dev = efx->net_dev;
3969 struct netdev_hw_addr *mc;
3970 unsigned int i, addr_count;
3972 if (net_dev->flags & (IFF_PROMISC | IFF_ALLMULTI))
3975 addr_count = netdev_mc_count(net_dev);
3977 netdev_for_each_mc_addr(mc, net_dev) {
3978 if (i >= EFX_EF10_FILTER_DEV_MC_MAX) {
3982 ether_addr_copy(table->dev_mc_list[i].addr, mc->addr);
3986 table->dev_mc_count = i;
3989 static int efx_ef10_filter_insert_addr_list(struct efx_nic *efx,
3990 bool multicast, bool rollback)
3992 struct efx_ef10_filter_table *table = efx->filter_state;
3993 struct efx_ef10_dev_addr *addr_list;
3994 enum efx_filter_flags filter_flags;
3995 struct efx_filter_spec spec;
4002 addr_list = table->dev_mc_list;
4003 addr_count = table->dev_mc_count;
4005 addr_list = table->dev_uc_list;
4006 addr_count = table->dev_uc_count;
4009 filter_flags = efx_rss_enabled(efx) ? EFX_FILTER_FLAG_RX_RSS : 0;
4011 /* Insert/renew filters */
4012 for (i = 0; i < addr_count; i++) {
4013 efx_filter_init_rx(&spec, EFX_FILTER_PRI_AUTO, filter_flags, 0);
4014 efx_filter_set_eth_local(&spec, EFX_FILTER_VID_UNSPEC,
4016 rc = efx_ef10_filter_insert(efx, &spec, true);
4019 netif_info(efx, drv, efx->net_dev,
4020 "efx_ef10_filter_insert failed rc=%d\n",
4022 /* Fall back to promiscuous */
4023 for (j = 0; j < i; j++) {
4024 if (addr_list[j].id == EFX_EF10_FILTER_ID_INVALID)
4026 efx_ef10_filter_remove_unsafe(
4027 efx, EFX_FILTER_PRI_AUTO,
4029 addr_list[j].id = EFX_EF10_FILTER_ID_INVALID;
4033 /* mark as not inserted, and carry on */
4034 rc = EFX_EF10_FILTER_ID_INVALID;
4037 addr_list[i].id = efx_ef10_filter_get_unsafe_id(efx, rc);
4040 if (multicast && rollback) {
4041 /* Also need an Ethernet broadcast filter */
4042 efx_filter_init_rx(&spec, EFX_FILTER_PRI_AUTO, filter_flags, 0);
4043 eth_broadcast_addr(baddr);
4044 efx_filter_set_eth_local(&spec, EFX_FILTER_VID_UNSPEC, baddr);
4045 rc = efx_ef10_filter_insert(efx, &spec, true);
4047 netif_warn(efx, drv, efx->net_dev,
4048 "Broadcast filter insert failed rc=%d\n", rc);
4049 /* Fall back to promiscuous */
4050 for (j = 0; j < i; j++) {
4051 if (addr_list[j].id == EFX_EF10_FILTER_ID_INVALID)
4053 efx_ef10_filter_remove_unsafe(
4054 efx, EFX_FILTER_PRI_AUTO,
4056 addr_list[j].id = EFX_EF10_FILTER_ID_INVALID;
4060 EFX_WARN_ON_PARANOID(table->bcast_id !=
4061 EFX_EF10_FILTER_ID_INVALID);
4062 table->bcast_id = efx_ef10_filter_get_unsafe_id(efx, rc);
4069 static int efx_ef10_filter_insert_def(struct efx_nic *efx, bool multicast,
4072 struct efx_ef10_filter_table *table = efx->filter_state;
4073 struct efx_ef10_nic_data *nic_data = efx->nic_data;
4074 enum efx_filter_flags filter_flags;
4075 struct efx_filter_spec spec;
4079 filter_flags = efx_rss_enabled(efx) ? EFX_FILTER_FLAG_RX_RSS : 0;
4081 efx_filter_init_rx(&spec, EFX_FILTER_PRI_AUTO, filter_flags, 0);
4084 efx_filter_set_mc_def(&spec);
4086 efx_filter_set_uc_def(&spec);
4088 rc = efx_ef10_filter_insert(efx, &spec, true);
4090 netif_printk(efx, drv, rc == -EPERM ? KERN_DEBUG : KERN_WARNING,
4092 "%scast mismatch filter insert failed rc=%d\n",
4093 multicast ? "Multi" : "Uni", rc);
4094 } else if (multicast) {
4095 EFX_WARN_ON_PARANOID(table->mcdef_id !=
4096 EFX_EF10_FILTER_ID_INVALID);
4097 table->mcdef_id = efx_ef10_filter_get_unsafe_id(efx, rc);
4098 if (!nic_data->workaround_26807) {
4099 /* Also need an Ethernet broadcast filter */
4100 efx_filter_init_rx(&spec, EFX_FILTER_PRI_AUTO,
4102 eth_broadcast_addr(baddr);
4103 efx_filter_set_eth_local(&spec, EFX_FILTER_VID_UNSPEC,
4105 rc = efx_ef10_filter_insert(efx, &spec, true);
4107 netif_warn(efx, drv, efx->net_dev,
4108 "Broadcast filter insert failed rc=%d\n",
4111 /* Roll back the mc_def filter */
4112 efx_ef10_filter_remove_unsafe(
4113 efx, EFX_FILTER_PRI_AUTO,
4115 table->mcdef_id = EFX_EF10_FILTER_ID_INVALID;
4119 EFX_WARN_ON_PARANOID(table->bcast_id !=
4120 EFX_EF10_FILTER_ID_INVALID);
4121 table->bcast_id = efx_ef10_filter_get_unsafe_id(efx, rc);
4126 EFX_WARN_ON_PARANOID(table->ucdef_id != EFX_EF10_FILTER_ID_INVALID);
4127 table->ucdef_id = rc;
4133 /* Remove filters that weren't renewed. Since nothing else changes the AUTO_OLD
4134 * flag or removes these filters, we don't need to hold the filter_lock while
4135 * scanning for these filters.
4137 static void efx_ef10_filter_remove_old(struct efx_nic *efx)
4139 struct efx_ef10_filter_table *table = efx->filter_state;
4140 int remove_failed = 0;
4141 int remove_noent = 0;
4145 for (i = 0; i < HUNT_FILTER_TBL_ROWS; i++) {
4146 if (ACCESS_ONCE(table->entry[i].spec) &
4147 EFX_EF10_FILTER_FLAG_AUTO_OLD) {
4148 rc = efx_ef10_filter_remove_internal(efx,
4149 1U << EFX_FILTER_PRI_AUTO, i, true);
4158 netif_info(efx, drv, efx->net_dev,
4159 "%s: failed to remove %d filters\n",
4160 __func__, remove_failed);
4162 netif_info(efx, drv, efx->net_dev,
4163 "%s: failed to remove %d non-existent filters\n",
4164 __func__, remove_noent);
4167 static int efx_ef10_vport_set_mac_address(struct efx_nic *efx)
4169 struct efx_ef10_nic_data *nic_data = efx->nic_data;
4170 u8 mac_old[ETH_ALEN];
4173 /* Only reconfigure a PF-created vport */
4174 if (is_zero_ether_addr(nic_data->vport_mac))
4177 efx_device_detach_sync(efx);
4178 efx_net_stop(efx->net_dev);
4179 down_write(&efx->filter_sem);
4180 efx_ef10_filter_table_remove(efx);
4181 up_write(&efx->filter_sem);
4183 rc = efx_ef10_vadaptor_free(efx, nic_data->vport_id);
4185 goto restore_filters;
4187 ether_addr_copy(mac_old, nic_data->vport_mac);
4188 rc = efx_ef10_vport_del_mac(efx, nic_data->vport_id,
4189 nic_data->vport_mac);
4191 goto restore_vadaptor;
4193 rc = efx_ef10_vport_add_mac(efx, nic_data->vport_id,
4194 efx->net_dev->dev_addr);
4196 ether_addr_copy(nic_data->vport_mac, efx->net_dev->dev_addr);
4198 rc2 = efx_ef10_vport_add_mac(efx, nic_data->vport_id, mac_old);
4200 /* Failed to add original MAC, so clear vport_mac */
4201 eth_zero_addr(nic_data->vport_mac);
4207 rc2 = efx_ef10_vadaptor_alloc(efx, nic_data->vport_id);
4211 down_write(&efx->filter_sem);
4212 rc2 = efx_ef10_filter_table_probe(efx);
4213 up_write(&efx->filter_sem);
4217 rc2 = efx_net_open(efx->net_dev);
4221 netif_device_attach(efx->net_dev);
4226 netif_err(efx, drv, efx->net_dev,
4227 "Failed to restore when changing MAC address - scheduling reset\n");
4228 efx_schedule_reset(efx, RESET_TYPE_DATAPATH);
4230 return rc ? rc : rc2;
4233 /* Caller must hold efx->filter_sem for read if race against
4234 * efx_ef10_filter_table_remove() is possible
4236 static void efx_ef10_filter_sync_rx_mode(struct efx_nic *efx)
4238 struct efx_ef10_filter_table *table = efx->filter_state;
4239 struct efx_ef10_nic_data *nic_data = efx->nic_data;
4240 struct net_device *net_dev = efx->net_dev;
4241 bool uc_promisc = false, mc_promisc = false;
4243 if (!efx_dev_registered(efx))
4249 efx_ef10_filter_mark_old(efx);
4251 /* Copy/convert the address lists; add the primary station
4252 * address and broadcast address
4254 netif_addr_lock_bh(net_dev);
4255 efx_ef10_filter_uc_addr_list(efx, &uc_promisc);
4256 efx_ef10_filter_mc_addr_list(efx, &mc_promisc);
4257 netif_addr_unlock_bh(net_dev);
4259 /* Insert/renew unicast filters */
4261 efx_ef10_filter_insert_def(efx, false, false);
4262 efx_ef10_filter_insert_addr_list(efx, false, false);
4264 /* If any of the filters failed to insert, fall back to
4265 * promiscuous mode - add in the uc_def filter. But keep
4266 * our individual unicast filters.
4268 if (efx_ef10_filter_insert_addr_list(efx, false, false))
4269 efx_ef10_filter_insert_def(efx, false, false);
4272 /* Insert/renew multicast filters */
4273 /* If changing promiscuous state with cascaded multicast filters, remove
4274 * old filters first, so that packets are dropped rather than duplicated
4276 if (nic_data->workaround_26807 && table->mc_promisc_last != mc_promisc)
4277 efx_ef10_filter_remove_old(efx);
4279 if (nic_data->workaround_26807) {
4280 /* If we failed to insert promiscuous filters, rollback
4281 * and fall back to individual multicast filters
4283 if (efx_ef10_filter_insert_def(efx, true, true)) {
4284 /* Changing promisc state, so remove old filters */
4285 efx_ef10_filter_remove_old(efx);
4286 efx_ef10_filter_insert_addr_list(efx, true, false);
4289 /* If we failed to insert promiscuous filters, don't
4290 * rollback. Regardless, also insert the mc_list
4292 efx_ef10_filter_insert_def(efx, true, false);
4293 efx_ef10_filter_insert_addr_list(efx, true, false);
4296 /* If any filters failed to insert, rollback and fall back to
4297 * promiscuous mode - mc_def filter and maybe broadcast. If
4298 * that fails, roll back again and insert as many of our
4299 * individual multicast filters as we can.
4301 if (efx_ef10_filter_insert_addr_list(efx, true, true)) {
4302 /* Changing promisc state, so remove old filters */
4303 if (nic_data->workaround_26807)
4304 efx_ef10_filter_remove_old(efx);
4305 if (efx_ef10_filter_insert_def(efx, true, true))
4306 efx_ef10_filter_insert_addr_list(efx, true, false);
4310 efx_ef10_filter_remove_old(efx);
4311 table->mc_promisc_last = mc_promisc;
4314 static int efx_ef10_set_mac_address(struct efx_nic *efx)
4316 MCDI_DECLARE_BUF(inbuf, MC_CMD_VADAPTOR_SET_MAC_IN_LEN);
4317 struct efx_ef10_nic_data *nic_data = efx->nic_data;
4318 bool was_enabled = efx->port_enabled;
4321 efx_device_detach_sync(efx);
4322 efx_net_stop(efx->net_dev);
4323 down_write(&efx->filter_sem);
4324 efx_ef10_filter_table_remove(efx);
4326 ether_addr_copy(MCDI_PTR(inbuf, VADAPTOR_SET_MAC_IN_MACADDR),
4327 efx->net_dev->dev_addr);
4328 MCDI_SET_DWORD(inbuf, VADAPTOR_SET_MAC_IN_UPSTREAM_PORT_ID,
4329 nic_data->vport_id);
4330 rc = efx_mcdi_rpc_quiet(efx, MC_CMD_VADAPTOR_SET_MAC, inbuf,
4331 sizeof(inbuf), NULL, 0, NULL);
4333 efx_ef10_filter_table_probe(efx);
4334 up_write(&efx->filter_sem);
4336 efx_net_open(efx->net_dev);
4337 netif_device_attach(efx->net_dev);
4339 #ifdef CONFIG_SFC_SRIOV
4340 if (efx->pci_dev->is_virtfn && efx->pci_dev->physfn) {
4341 struct pci_dev *pci_dev_pf = efx->pci_dev->physfn;
4344 struct efx_nic *efx_pf;
4346 /* Switch to PF and change MAC address on vport */
4347 efx_pf = pci_get_drvdata(pci_dev_pf);
4349 rc = efx_ef10_sriov_set_vf_mac(efx_pf,
4351 efx->net_dev->dev_addr);
4353 struct efx_nic *efx_pf = pci_get_drvdata(pci_dev_pf);
4354 struct efx_ef10_nic_data *nic_data = efx_pf->nic_data;
4357 /* MAC address successfully changed by VF (with MAC
4358 * spoofing) so update the parent PF if possible.
4360 for (i = 0; i < efx_pf->vf_count; ++i) {
4361 struct ef10_vf *vf = nic_data->vf + i;
4363 if (vf->efx == efx) {
4364 ether_addr_copy(vf->mac,
4365 efx->net_dev->dev_addr);
4373 netif_err(efx, drv, efx->net_dev,
4374 "Cannot change MAC address; use sfboot to enable"
4375 " mac-spoofing on this interface\n");
4376 } else if (rc == -ENOSYS && !efx_ef10_is_vf(efx)) {
4377 /* If the active MCFW does not support MC_CMD_VADAPTOR_SET_MAC
4378 * fall-back to the method of changing the MAC address on the
4379 * vport. This only applies to PFs because such versions of
4380 * MCFW do not support VFs.
4382 rc = efx_ef10_vport_set_mac_address(efx);
4384 efx_mcdi_display_error(efx, MC_CMD_VADAPTOR_SET_MAC,
4385 sizeof(inbuf), NULL, 0, rc);
4391 static int efx_ef10_mac_reconfigure(struct efx_nic *efx)
4393 efx_ef10_filter_sync_rx_mode(efx);
4395 return efx_mcdi_set_mac(efx);
4398 static int efx_ef10_mac_reconfigure_vf(struct efx_nic *efx)
4400 efx_ef10_filter_sync_rx_mode(efx);
4405 static int efx_ef10_start_bist(struct efx_nic *efx, u32 bist_type)
4407 MCDI_DECLARE_BUF(inbuf, MC_CMD_START_BIST_IN_LEN);
4409 MCDI_SET_DWORD(inbuf, START_BIST_IN_TYPE, bist_type);
4410 return efx_mcdi_rpc(efx, MC_CMD_START_BIST, inbuf, sizeof(inbuf),
4414 /* MC BISTs follow a different poll mechanism to phy BISTs.
4415 * The BIST is done in the poll handler on the MC, and the MCDI command
4416 * will block until the BIST is done.
4418 static int efx_ef10_poll_bist(struct efx_nic *efx)
4421 MCDI_DECLARE_BUF(outbuf, MC_CMD_POLL_BIST_OUT_LEN);
4425 rc = efx_mcdi_rpc(efx, MC_CMD_POLL_BIST, NULL, 0,
4426 outbuf, sizeof(outbuf), &outlen);
4430 if (outlen < MC_CMD_POLL_BIST_OUT_LEN)
4433 result = MCDI_DWORD(outbuf, POLL_BIST_OUT_RESULT);
4435 case MC_CMD_POLL_BIST_PASSED:
4436 netif_dbg(efx, hw, efx->net_dev, "BIST passed.\n");
4438 case MC_CMD_POLL_BIST_TIMEOUT:
4439 netif_err(efx, hw, efx->net_dev, "BIST timed out\n");
4441 case MC_CMD_POLL_BIST_FAILED:
4442 netif_err(efx, hw, efx->net_dev, "BIST failed.\n");
4445 netif_err(efx, hw, efx->net_dev,
4446 "BIST returned unknown result %u", result);
4451 static int efx_ef10_run_bist(struct efx_nic *efx, u32 bist_type)
4455 netif_dbg(efx, drv, efx->net_dev, "starting BIST type %u\n", bist_type);
4457 rc = efx_ef10_start_bist(efx, bist_type);
4461 return efx_ef10_poll_bist(efx);
4465 efx_ef10_test_chip(struct efx_nic *efx, struct efx_self_tests *tests)
4469 efx_reset_down(efx, RESET_TYPE_WORLD);
4471 rc = efx_mcdi_rpc(efx, MC_CMD_ENABLE_OFFLINE_BIST,
4472 NULL, 0, NULL, 0, NULL);
4476 tests->memory = efx_ef10_run_bist(efx, MC_CMD_MC_MEM_BIST) ? -1 : 1;
4477 tests->registers = efx_ef10_run_bist(efx, MC_CMD_REG_BIST) ? -1 : 1;
4479 rc = efx_mcdi_reset(efx, RESET_TYPE_WORLD);
4484 rc2 = efx_reset_up(efx, RESET_TYPE_WORLD, rc == 0);
4485 return rc ? rc : rc2;
4488 #ifdef CONFIG_SFC_MTD
4490 struct efx_ef10_nvram_type_info {
4491 u16 type, type_mask;
4496 static const struct efx_ef10_nvram_type_info efx_ef10_nvram_types[] = {
4497 { NVRAM_PARTITION_TYPE_MC_FIRMWARE, 0, 0, "sfc_mcfw" },
4498 { NVRAM_PARTITION_TYPE_MC_FIRMWARE_BACKUP, 0, 0, "sfc_mcfw_backup" },
4499 { NVRAM_PARTITION_TYPE_EXPANSION_ROM, 0, 0, "sfc_exp_rom" },
4500 { NVRAM_PARTITION_TYPE_STATIC_CONFIG, 0, 0, "sfc_static_cfg" },
4501 { NVRAM_PARTITION_TYPE_DYNAMIC_CONFIG, 0, 0, "sfc_dynamic_cfg" },
4502 { NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT0, 0, 0, "sfc_exp_rom_cfg" },
4503 { NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT1, 0, 1, "sfc_exp_rom_cfg" },
4504 { NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT2, 0, 2, "sfc_exp_rom_cfg" },
4505 { NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT3, 0, 3, "sfc_exp_rom_cfg" },
4506 { NVRAM_PARTITION_TYPE_LICENSE, 0, 0, "sfc_license" },
4507 { NVRAM_PARTITION_TYPE_PHY_MIN, 0xff, 0, "sfc_phy_fw" },
4510 static int efx_ef10_mtd_probe_partition(struct efx_nic *efx,
4511 struct efx_mcdi_mtd_partition *part,
4514 MCDI_DECLARE_BUF(inbuf, MC_CMD_NVRAM_METADATA_IN_LEN);
4515 MCDI_DECLARE_BUF(outbuf, MC_CMD_NVRAM_METADATA_OUT_LENMAX);
4516 const struct efx_ef10_nvram_type_info *info;
4517 size_t size, erase_size, outlen;
4521 for (info = efx_ef10_nvram_types; ; info++) {
4523 efx_ef10_nvram_types + ARRAY_SIZE(efx_ef10_nvram_types))
4525 if ((type & ~info->type_mask) == info->type)
4528 if (info->port != efx_port_num(efx))
4531 rc = efx_mcdi_nvram_info(efx, type, &size, &erase_size, &protected);
4535 return -ENODEV; /* hide it */
4537 part->nvram_type = type;
4539 MCDI_SET_DWORD(inbuf, NVRAM_METADATA_IN_TYPE, type);
4540 rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_METADATA, inbuf, sizeof(inbuf),
4541 outbuf, sizeof(outbuf), &outlen);
4544 if (outlen < MC_CMD_NVRAM_METADATA_OUT_LENMIN)
4546 if (MCDI_DWORD(outbuf, NVRAM_METADATA_OUT_FLAGS) &
4547 (1 << MC_CMD_NVRAM_METADATA_OUT_SUBTYPE_VALID_LBN))
4548 part->fw_subtype = MCDI_DWORD(outbuf,
4549 NVRAM_METADATA_OUT_SUBTYPE);
4551 part->common.dev_type_name = "EF10 NVRAM manager";
4552 part->common.type_name = info->name;
4554 part->common.mtd.type = MTD_NORFLASH;
4555 part->common.mtd.flags = MTD_CAP_NORFLASH;
4556 part->common.mtd.size = size;
4557 part->common.mtd.erasesize = erase_size;
4562 static int efx_ef10_mtd_probe(struct efx_nic *efx)
4564 MCDI_DECLARE_BUF(outbuf, MC_CMD_NVRAM_PARTITIONS_OUT_LENMAX);
4565 struct efx_mcdi_mtd_partition *parts;
4566 size_t outlen, n_parts_total, i, n_parts;
4572 BUILD_BUG_ON(MC_CMD_NVRAM_PARTITIONS_IN_LEN != 0);
4573 rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_PARTITIONS, NULL, 0,
4574 outbuf, sizeof(outbuf), &outlen);
4577 if (outlen < MC_CMD_NVRAM_PARTITIONS_OUT_LENMIN)
4580 n_parts_total = MCDI_DWORD(outbuf, NVRAM_PARTITIONS_OUT_NUM_PARTITIONS);
4582 MCDI_VAR_ARRAY_LEN(outlen, NVRAM_PARTITIONS_OUT_TYPE_ID))
4585 parts = kcalloc(n_parts_total, sizeof(*parts), GFP_KERNEL);
4590 for (i = 0; i < n_parts_total; i++) {
4591 type = MCDI_ARRAY_DWORD(outbuf, NVRAM_PARTITIONS_OUT_TYPE_ID,
4593 rc = efx_ef10_mtd_probe_partition(efx, &parts[n_parts], type);
4596 else if (rc != -ENODEV)
4600 rc = efx_mtd_add(efx, &parts[0].common, n_parts, sizeof(*parts));
4607 #endif /* CONFIG_SFC_MTD */
4609 static void efx_ef10_ptp_write_host_time(struct efx_nic *efx, u32 host_time)
4611 _efx_writed(efx, cpu_to_le32(host_time), ER_DZ_MC_DB_LWRD);
4614 static void efx_ef10_ptp_write_host_time_vf(struct efx_nic *efx,
4617 static int efx_ef10_rx_enable_timestamping(struct efx_channel *channel,
4620 MCDI_DECLARE_BUF(inbuf, MC_CMD_PTP_IN_TIME_EVENT_SUBSCRIBE_LEN);
4623 if (channel->sync_events_state == SYNC_EVENTS_REQUESTED ||
4624 channel->sync_events_state == SYNC_EVENTS_VALID ||
4625 (temp && channel->sync_events_state == SYNC_EVENTS_DISABLED))
4627 channel->sync_events_state = SYNC_EVENTS_REQUESTED;
4629 MCDI_SET_DWORD(inbuf, PTP_IN_OP, MC_CMD_PTP_OP_TIME_EVENT_SUBSCRIBE);
4630 MCDI_SET_DWORD(inbuf, PTP_IN_PERIPH_ID, 0);
4631 MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_SUBSCRIBE_QUEUE,
4634 rc = efx_mcdi_rpc(channel->efx, MC_CMD_PTP,
4635 inbuf, sizeof(inbuf), NULL, 0, NULL);
4638 channel->sync_events_state = temp ? SYNC_EVENTS_QUIESCENT :
4639 SYNC_EVENTS_DISABLED;
4644 static int efx_ef10_rx_disable_timestamping(struct efx_channel *channel,
4647 MCDI_DECLARE_BUF(inbuf, MC_CMD_PTP_IN_TIME_EVENT_UNSUBSCRIBE_LEN);
4650 if (channel->sync_events_state == SYNC_EVENTS_DISABLED ||
4651 (temp && channel->sync_events_state == SYNC_EVENTS_QUIESCENT))
4653 if (channel->sync_events_state == SYNC_EVENTS_QUIESCENT) {
4654 channel->sync_events_state = SYNC_EVENTS_DISABLED;
4657 channel->sync_events_state = temp ? SYNC_EVENTS_QUIESCENT :
4658 SYNC_EVENTS_DISABLED;
4660 MCDI_SET_DWORD(inbuf, PTP_IN_OP, MC_CMD_PTP_OP_TIME_EVENT_UNSUBSCRIBE);
4661 MCDI_SET_DWORD(inbuf, PTP_IN_PERIPH_ID, 0);
4662 MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_UNSUBSCRIBE_CONTROL,
4663 MC_CMD_PTP_IN_TIME_EVENT_UNSUBSCRIBE_SINGLE);
4664 MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_UNSUBSCRIBE_QUEUE,
4667 rc = efx_mcdi_rpc(channel->efx, MC_CMD_PTP,
4668 inbuf, sizeof(inbuf), NULL, 0, NULL);
4673 static int efx_ef10_ptp_set_ts_sync_events(struct efx_nic *efx, bool en,
4676 int (*set)(struct efx_channel *channel, bool temp);
4677 struct efx_channel *channel;
4680 efx_ef10_rx_enable_timestamping :
4681 efx_ef10_rx_disable_timestamping;
4683 efx_for_each_channel(channel, efx) {
4684 int rc = set(channel, temp);
4685 if (en && rc != 0) {
4686 efx_ef10_ptp_set_ts_sync_events(efx, false, temp);
4694 static int efx_ef10_ptp_set_ts_config_vf(struct efx_nic *efx,
4695 struct hwtstamp_config *init)
4700 static int efx_ef10_ptp_set_ts_config(struct efx_nic *efx,
4701 struct hwtstamp_config *init)
4705 switch (init->rx_filter) {
4706 case HWTSTAMP_FILTER_NONE:
4707 efx_ef10_ptp_set_ts_sync_events(efx, false, false);
4708 /* if TX timestamping is still requested then leave PTP on */
4709 return efx_ptp_change_mode(efx,
4710 init->tx_type != HWTSTAMP_TX_OFF, 0);
4711 case HWTSTAMP_FILTER_ALL:
4712 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
4713 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
4714 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
4715 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
4716 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
4717 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
4718 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
4719 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
4720 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
4721 case HWTSTAMP_FILTER_PTP_V2_EVENT:
4722 case HWTSTAMP_FILTER_PTP_V2_SYNC:
4723 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
4724 init->rx_filter = HWTSTAMP_FILTER_ALL;
4725 rc = efx_ptp_change_mode(efx, true, 0);
4727 rc = efx_ef10_ptp_set_ts_sync_events(efx, true, false);
4729 efx_ptp_change_mode(efx, false, 0);
4736 #define EF10_OFFLOAD_FEATURES \
4737 (NETIF_F_IP_CSUM | \
4738 NETIF_F_IPV6_CSUM | \
4742 const struct efx_nic_type efx_hunt_a0_vf_nic_type = {
4744 .mem_bar = EFX_MEM_VF_BAR,
4745 .mem_map_size = efx_ef10_mem_map_size,
4746 .probe = efx_ef10_probe_vf,
4747 .remove = efx_ef10_remove,
4748 .dimension_resources = efx_ef10_dimension_resources,
4749 .init = efx_ef10_init_nic,
4750 .fini = efx_port_dummy_op_void,
4751 .map_reset_reason = efx_ef10_map_reset_reason,
4752 .map_reset_flags = efx_ef10_map_reset_flags,
4753 .reset = efx_ef10_reset,
4754 .probe_port = efx_mcdi_port_probe,
4755 .remove_port = efx_mcdi_port_remove,
4756 .fini_dmaq = efx_ef10_fini_dmaq,
4757 .prepare_flr = efx_ef10_prepare_flr,
4758 .finish_flr = efx_port_dummy_op_void,
4759 .describe_stats = efx_ef10_describe_stats,
4760 .update_stats = efx_ef10_update_stats_vf,
4761 .start_stats = efx_port_dummy_op_void,
4762 .pull_stats = efx_port_dummy_op_void,
4763 .stop_stats = efx_port_dummy_op_void,
4764 .set_id_led = efx_mcdi_set_id_led,
4765 .push_irq_moderation = efx_ef10_push_irq_moderation,
4766 .reconfigure_mac = efx_ef10_mac_reconfigure_vf,
4767 .check_mac_fault = efx_mcdi_mac_check_fault,
4768 .reconfigure_port = efx_mcdi_port_reconfigure,
4769 .get_wol = efx_ef10_get_wol_vf,
4770 .set_wol = efx_ef10_set_wol_vf,
4771 .resume_wol = efx_port_dummy_op_void,
4772 .mcdi_request = efx_ef10_mcdi_request,
4773 .mcdi_poll_response = efx_ef10_mcdi_poll_response,
4774 .mcdi_read_response = efx_ef10_mcdi_read_response,
4775 .mcdi_poll_reboot = efx_ef10_mcdi_poll_reboot,
4776 .mcdi_reboot_detected = efx_ef10_mcdi_reboot_detected,
4777 .irq_enable_master = efx_port_dummy_op_void,
4778 .irq_test_generate = efx_ef10_irq_test_generate,
4779 .irq_disable_non_ev = efx_port_dummy_op_void,
4780 .irq_handle_msi = efx_ef10_msi_interrupt,
4781 .irq_handle_legacy = efx_ef10_legacy_interrupt,
4782 .tx_probe = efx_ef10_tx_probe,
4783 .tx_init = efx_ef10_tx_init,
4784 .tx_remove = efx_ef10_tx_remove,
4785 .tx_write = efx_ef10_tx_write,
4786 .rx_push_rss_config = efx_ef10_vf_rx_push_rss_config,
4787 .rx_probe = efx_ef10_rx_probe,
4788 .rx_init = efx_ef10_rx_init,
4789 .rx_remove = efx_ef10_rx_remove,
4790 .rx_write = efx_ef10_rx_write,
4791 .rx_defer_refill = efx_ef10_rx_defer_refill,
4792 .ev_probe = efx_ef10_ev_probe,
4793 .ev_init = efx_ef10_ev_init,
4794 .ev_fini = efx_ef10_ev_fini,
4795 .ev_remove = efx_ef10_ev_remove,
4796 .ev_process = efx_ef10_ev_process,
4797 .ev_read_ack = efx_ef10_ev_read_ack,
4798 .ev_test_generate = efx_ef10_ev_test_generate,
4799 .filter_table_probe = efx_ef10_filter_table_probe,
4800 .filter_table_restore = efx_ef10_filter_table_restore,
4801 .filter_table_remove = efx_ef10_filter_table_remove,
4802 .filter_update_rx_scatter = efx_ef10_filter_update_rx_scatter,
4803 .filter_insert = efx_ef10_filter_insert,
4804 .filter_remove_safe = efx_ef10_filter_remove_safe,
4805 .filter_get_safe = efx_ef10_filter_get_safe,
4806 .filter_clear_rx = efx_ef10_filter_clear_rx,
4807 .filter_count_rx_used = efx_ef10_filter_count_rx_used,
4808 .filter_get_rx_id_limit = efx_ef10_filter_get_rx_id_limit,
4809 .filter_get_rx_ids = efx_ef10_filter_get_rx_ids,
4810 #ifdef CONFIG_RFS_ACCEL
4811 .filter_rfs_insert = efx_ef10_filter_rfs_insert,
4812 .filter_rfs_expire_one = efx_ef10_filter_rfs_expire_one,
4814 #ifdef CONFIG_SFC_MTD
4815 .mtd_probe = efx_port_dummy_op_int,
4817 .ptp_write_host_time = efx_ef10_ptp_write_host_time_vf,
4818 .ptp_set_ts_config = efx_ef10_ptp_set_ts_config_vf,
4819 #ifdef CONFIG_SFC_SRIOV
4820 .vswitching_probe = efx_ef10_vswitching_probe_vf,
4821 .vswitching_restore = efx_ef10_vswitching_restore_vf,
4822 .vswitching_remove = efx_ef10_vswitching_remove_vf,
4823 .sriov_get_phys_port_id = efx_ef10_sriov_get_phys_port_id,
4825 .get_mac_address = efx_ef10_get_mac_address_vf,
4826 .set_mac_address = efx_ef10_set_mac_address,
4828 .revision = EFX_REV_HUNT_A0,
4829 .max_dma_mask = DMA_BIT_MASK(ESF_DZ_TX_KER_BUF_ADDR_WIDTH),
4830 .rx_prefix_size = ES_DZ_RX_PREFIX_SIZE,
4831 .rx_hash_offset = ES_DZ_RX_PREFIX_HASH_OFST,
4832 .rx_ts_offset = ES_DZ_RX_PREFIX_TSTAMP_OFST,
4833 .can_rx_scatter = true,
4834 .always_rx_scatter = true,
4835 .max_interrupt_mode = EFX_INT_MODE_MSIX,
4836 .timer_period_max = 1 << ERF_DD_EVQ_IND_TIMER_VAL_WIDTH,
4837 .offload_features = EF10_OFFLOAD_FEATURES,
4839 .max_rx_ip_filters = HUNT_FILTER_TBL_ROWS,
4840 .hwtstamp_filters = 1 << HWTSTAMP_FILTER_NONE |
4841 1 << HWTSTAMP_FILTER_ALL,
4844 const struct efx_nic_type efx_hunt_a0_nic_type = {
4846 .mem_bar = EFX_MEM_BAR,
4847 .mem_map_size = efx_ef10_mem_map_size,
4848 .probe = efx_ef10_probe_pf,
4849 .remove = efx_ef10_remove,
4850 .dimension_resources = efx_ef10_dimension_resources,
4851 .init = efx_ef10_init_nic,
4852 .fini = efx_port_dummy_op_void,
4853 .map_reset_reason = efx_ef10_map_reset_reason,
4854 .map_reset_flags = efx_ef10_map_reset_flags,
4855 .reset = efx_ef10_reset,
4856 .probe_port = efx_mcdi_port_probe,
4857 .remove_port = efx_mcdi_port_remove,
4858 .fini_dmaq = efx_ef10_fini_dmaq,
4859 .prepare_flr = efx_ef10_prepare_flr,
4860 .finish_flr = efx_port_dummy_op_void,
4861 .describe_stats = efx_ef10_describe_stats,
4862 .update_stats = efx_ef10_update_stats_pf,
4863 .start_stats = efx_mcdi_mac_start_stats,
4864 .pull_stats = efx_mcdi_mac_pull_stats,
4865 .stop_stats = efx_mcdi_mac_stop_stats,
4866 .set_id_led = efx_mcdi_set_id_led,
4867 .push_irq_moderation = efx_ef10_push_irq_moderation,
4868 .reconfigure_mac = efx_ef10_mac_reconfigure,
4869 .check_mac_fault = efx_mcdi_mac_check_fault,
4870 .reconfigure_port = efx_mcdi_port_reconfigure,
4871 .get_wol = efx_ef10_get_wol,
4872 .set_wol = efx_ef10_set_wol,
4873 .resume_wol = efx_port_dummy_op_void,
4874 .test_chip = efx_ef10_test_chip,
4875 .test_nvram = efx_mcdi_nvram_test_all,
4876 .mcdi_request = efx_ef10_mcdi_request,
4877 .mcdi_poll_response = efx_ef10_mcdi_poll_response,
4878 .mcdi_read_response = efx_ef10_mcdi_read_response,
4879 .mcdi_poll_reboot = efx_ef10_mcdi_poll_reboot,
4880 .mcdi_reboot_detected = efx_ef10_mcdi_reboot_detected,
4881 .irq_enable_master = efx_port_dummy_op_void,
4882 .irq_test_generate = efx_ef10_irq_test_generate,
4883 .irq_disable_non_ev = efx_port_dummy_op_void,
4884 .irq_handle_msi = efx_ef10_msi_interrupt,
4885 .irq_handle_legacy = efx_ef10_legacy_interrupt,
4886 .tx_probe = efx_ef10_tx_probe,
4887 .tx_init = efx_ef10_tx_init,
4888 .tx_remove = efx_ef10_tx_remove,
4889 .tx_write = efx_ef10_tx_write,
4890 .rx_push_rss_config = efx_ef10_pf_rx_push_rss_config,
4891 .rx_probe = efx_ef10_rx_probe,
4892 .rx_init = efx_ef10_rx_init,
4893 .rx_remove = efx_ef10_rx_remove,
4894 .rx_write = efx_ef10_rx_write,
4895 .rx_defer_refill = efx_ef10_rx_defer_refill,
4896 .ev_probe = efx_ef10_ev_probe,
4897 .ev_init = efx_ef10_ev_init,
4898 .ev_fini = efx_ef10_ev_fini,
4899 .ev_remove = efx_ef10_ev_remove,
4900 .ev_process = efx_ef10_ev_process,
4901 .ev_read_ack = efx_ef10_ev_read_ack,
4902 .ev_test_generate = efx_ef10_ev_test_generate,
4903 .filter_table_probe = efx_ef10_filter_table_probe,
4904 .filter_table_restore = efx_ef10_filter_table_restore,
4905 .filter_table_remove = efx_ef10_filter_table_remove,
4906 .filter_update_rx_scatter = efx_ef10_filter_update_rx_scatter,
4907 .filter_insert = efx_ef10_filter_insert,
4908 .filter_remove_safe = efx_ef10_filter_remove_safe,
4909 .filter_get_safe = efx_ef10_filter_get_safe,
4910 .filter_clear_rx = efx_ef10_filter_clear_rx,
4911 .filter_count_rx_used = efx_ef10_filter_count_rx_used,
4912 .filter_get_rx_id_limit = efx_ef10_filter_get_rx_id_limit,
4913 .filter_get_rx_ids = efx_ef10_filter_get_rx_ids,
4914 #ifdef CONFIG_RFS_ACCEL
4915 .filter_rfs_insert = efx_ef10_filter_rfs_insert,
4916 .filter_rfs_expire_one = efx_ef10_filter_rfs_expire_one,
4918 #ifdef CONFIG_SFC_MTD
4919 .mtd_probe = efx_ef10_mtd_probe,
4920 .mtd_rename = efx_mcdi_mtd_rename,
4921 .mtd_read = efx_mcdi_mtd_read,
4922 .mtd_erase = efx_mcdi_mtd_erase,
4923 .mtd_write = efx_mcdi_mtd_write,
4924 .mtd_sync = efx_mcdi_mtd_sync,
4926 .ptp_write_host_time = efx_ef10_ptp_write_host_time,
4927 .ptp_set_ts_sync_events = efx_ef10_ptp_set_ts_sync_events,
4928 .ptp_set_ts_config = efx_ef10_ptp_set_ts_config,
4929 #ifdef CONFIG_SFC_SRIOV
4930 .sriov_configure = efx_ef10_sriov_configure,
4931 .sriov_init = efx_ef10_sriov_init,
4932 .sriov_fini = efx_ef10_sriov_fini,
4933 .sriov_wanted = efx_ef10_sriov_wanted,
4934 .sriov_reset = efx_ef10_sriov_reset,
4935 .sriov_flr = efx_ef10_sriov_flr,
4936 .sriov_set_vf_mac = efx_ef10_sriov_set_vf_mac,
4937 .sriov_set_vf_vlan = efx_ef10_sriov_set_vf_vlan,
4938 .sriov_set_vf_spoofchk = efx_ef10_sriov_set_vf_spoofchk,
4939 .sriov_get_vf_config = efx_ef10_sriov_get_vf_config,
4940 .sriov_set_vf_link_state = efx_ef10_sriov_set_vf_link_state,
4941 .vswitching_probe = efx_ef10_vswitching_probe_pf,
4942 .vswitching_restore = efx_ef10_vswitching_restore_pf,
4943 .vswitching_remove = efx_ef10_vswitching_remove_pf,
4945 .get_mac_address = efx_ef10_get_mac_address_pf,
4946 .set_mac_address = efx_ef10_set_mac_address,
4948 .revision = EFX_REV_HUNT_A0,
4949 .max_dma_mask = DMA_BIT_MASK(ESF_DZ_TX_KER_BUF_ADDR_WIDTH),
4950 .rx_prefix_size = ES_DZ_RX_PREFIX_SIZE,
4951 .rx_hash_offset = ES_DZ_RX_PREFIX_HASH_OFST,
4952 .rx_ts_offset = ES_DZ_RX_PREFIX_TSTAMP_OFST,
4953 .can_rx_scatter = true,
4954 .always_rx_scatter = true,
4955 .max_interrupt_mode = EFX_INT_MODE_MSIX,
4956 .timer_period_max = 1 << ERF_DD_EVQ_IND_TIMER_VAL_WIDTH,
4957 .offload_features = EF10_OFFLOAD_FEATURES,
4959 .max_rx_ip_filters = HUNT_FILTER_TBL_ROWS,
4960 .hwtstamp_filters = 1 << HWTSTAMP_FILTER_NONE |
4961 1 << HWTSTAMP_FILTER_ALL,