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"
19 #include <linux/jhash.h>
20 #include <linux/wait.h>
21 #include <linux/workqueue.h>
23 /* Hardware control for EF10 architecture including 'Huntington'. */
25 #define EFX_EF10_DRVGEN_EV 7
31 /* The reserved RSS context value */
32 #define EFX_EF10_RSS_CONTEXT_INVALID 0xffffffff
34 /* The filter table(s) are managed by firmware and we have write-only
35 * access. When removing filters we must identify them to the
36 * firmware by a 64-bit handle, but this is too wide for Linux kernel
37 * interfaces (32-bit for RX NFC, 16-bit for RFS). Also, we need to
38 * be able to tell in advance whether a requested insertion will
39 * replace an existing filter. Therefore we maintain a software hash
40 * table, which should be at least as large as the hardware hash
43 * Huntington has a single 8K filter table shared between all filter
44 * types and both ports.
46 #define HUNT_FILTER_TBL_ROWS 8192
48 struct efx_ef10_filter_table {
49 /* The RX match field masks supported by this fw & hw, in order of priority */
50 enum efx_filter_match_flags rx_match_flags[
51 MC_CMD_GET_PARSER_DISP_INFO_OUT_SUPPORTED_MATCHES_MAXNUM];
52 unsigned int rx_match_count;
55 unsigned long spec; /* pointer to spec plus flag bits */
56 /* BUSY flag indicates that an update is in progress. AUTO_OLD is
57 * used to mark and sweep MAC filters for the device address lists.
59 #define EFX_EF10_FILTER_FLAG_BUSY 1UL
60 #define EFX_EF10_FILTER_FLAG_AUTO_OLD 2UL
61 #define EFX_EF10_FILTER_FLAGS 3UL
62 u64 handle; /* firmware handle */
64 wait_queue_head_t waitq;
65 /* Shadow of net_device address lists, guarded by mac_lock */
66 #define EFX_EF10_FILTER_DEV_UC_MAX 32
67 #define EFX_EF10_FILTER_DEV_MC_MAX 256
71 } dev_uc_list[EFX_EF10_FILTER_DEV_UC_MAX],
72 dev_mc_list[EFX_EF10_FILTER_DEV_MC_MAX];
73 int dev_uc_count; /* negative for PROMISC */
74 int dev_mc_count; /* negative for PROMISC/ALLMULTI */
77 /* An arbitrary search limit for the software hash table */
78 #define EFX_EF10_FILTER_SEARCH_LIMIT 200
80 static void efx_ef10_rx_push_rss_config(struct efx_nic *efx);
81 static void efx_ef10_rx_free_indir_table(struct efx_nic *efx);
82 static void efx_ef10_filter_table_remove(struct efx_nic *efx);
84 static int efx_ef10_get_warm_boot_count(struct efx_nic *efx)
88 efx_readd(efx, ®, ER_DZ_BIU_MC_SFT_STATUS);
89 return EFX_DWORD_FIELD(reg, EFX_WORD_1) == 0xb007 ?
90 EFX_DWORD_FIELD(reg, EFX_WORD_0) : -EIO;
93 static unsigned int efx_ef10_mem_map_size(struct efx_nic *efx)
95 return resource_size(&efx->pci_dev->resource[EFX_MEM_BAR]);
98 static int efx_ef10_init_datapath_caps(struct efx_nic *efx)
100 MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_CAPABILITIES_OUT_LEN);
101 struct efx_ef10_nic_data *nic_data = efx->nic_data;
105 BUILD_BUG_ON(MC_CMD_GET_CAPABILITIES_IN_LEN != 0);
107 rc = efx_mcdi_rpc(efx, MC_CMD_GET_CAPABILITIES, NULL, 0,
108 outbuf, sizeof(outbuf), &outlen);
111 if (outlen < sizeof(outbuf)) {
112 netif_err(efx, drv, efx->net_dev,
113 "unable to read datapath firmware capabilities\n");
117 nic_data->datapath_caps =
118 MCDI_DWORD(outbuf, GET_CAPABILITIES_OUT_FLAGS1);
120 if (!(nic_data->datapath_caps &
121 (1 << MC_CMD_GET_CAPABILITIES_OUT_TX_TSO_LBN))) {
122 netif_err(efx, drv, efx->net_dev,
123 "current firmware does not support TSO\n");
127 if (!(nic_data->datapath_caps &
128 (1 << MC_CMD_GET_CAPABILITIES_OUT_RX_PREFIX_LEN_14_LBN))) {
129 netif_err(efx, probe, efx->net_dev,
130 "current firmware does not support an RX prefix\n");
137 static int efx_ef10_get_sysclk_freq(struct efx_nic *efx)
139 MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_CLOCK_OUT_LEN);
142 rc = efx_mcdi_rpc(efx, MC_CMD_GET_CLOCK, NULL, 0,
143 outbuf, sizeof(outbuf), NULL);
146 rc = MCDI_DWORD(outbuf, GET_CLOCK_OUT_SYS_FREQ);
147 return rc > 0 ? rc : -ERANGE;
150 static int efx_ef10_get_mac_address(struct efx_nic *efx, u8 *mac_address)
152 MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_MAC_ADDRESSES_OUT_LEN);
156 BUILD_BUG_ON(MC_CMD_GET_MAC_ADDRESSES_IN_LEN != 0);
158 rc = efx_mcdi_rpc(efx, MC_CMD_GET_MAC_ADDRESSES, NULL, 0,
159 outbuf, sizeof(outbuf), &outlen);
162 if (outlen < MC_CMD_GET_MAC_ADDRESSES_OUT_LEN)
165 ether_addr_copy(mac_address,
166 MCDI_PTR(outbuf, GET_MAC_ADDRESSES_OUT_MAC_ADDR_BASE));
170 static int efx_ef10_probe(struct efx_nic *efx)
172 struct efx_ef10_nic_data *nic_data;
175 /* We can have one VI for each 8K region. However, until we
176 * use TX option descriptors we need two TX queues per channel.
181 resource_size(&efx->pci_dev->resource[EFX_MEM_BAR]) /
182 (EFX_VI_PAGE_SIZE * EFX_TXQ_TYPES));
183 BUG_ON(efx->max_channels == 0);
185 nic_data = kzalloc(sizeof(*nic_data), GFP_KERNEL);
188 efx->nic_data = nic_data;
190 rc = efx_nic_alloc_buffer(efx, &nic_data->mcdi_buf,
191 8 + MCDI_CTL_SDU_LEN_MAX_V2, GFP_KERNEL);
195 /* Get the MC's warm boot count. In case it's rebooting right
196 * now, be prepared to retry.
200 rc = efx_ef10_get_warm_boot_count(efx);
207 nic_data->warm_boot_count = rc;
209 nic_data->rx_rss_context = EFX_EF10_RSS_CONTEXT_INVALID;
211 /* In case we're recovering from a crash (kexec), we want to
212 * cancel any outstanding request by the previous user of this
213 * function. We send a special message using the least
214 * significant bits of the 'high' (doorbell) register.
216 _efx_writed(efx, cpu_to_le32(1), ER_DZ_MC_DB_HWRD);
218 rc = efx_mcdi_init(efx);
222 /* Reset (most) configuration for this function */
223 rc = efx_mcdi_reset(efx, RESET_TYPE_ALL);
227 /* Enable event logging */
228 rc = efx_mcdi_log_ctrl(efx, true, false, 0);
232 rc = efx_ef10_init_datapath_caps(efx);
236 efx->rx_packet_len_offset =
237 ES_DZ_RX_PREFIX_PKTLEN_OFST - ES_DZ_RX_PREFIX_SIZE;
239 rc = efx_mcdi_port_get_number(efx);
244 rc = efx_ef10_get_mac_address(efx, efx->net_dev->perm_addr);
248 rc = efx_ef10_get_sysclk_freq(efx);
251 efx->timer_quantum_ns = 1536000 / rc; /* 1536 cycles */
253 /* Check whether firmware supports bug 35388 workaround */
254 rc = efx_mcdi_set_workaround(efx, MC_CMD_WORKAROUND_BUG35388, true);
256 nic_data->workaround_35388 = true;
257 else if (rc != -ENOSYS && rc != -ENOENT)
259 netif_dbg(efx, probe, efx->net_dev,
260 "workaround for bug 35388 is %sabled\n",
261 nic_data->workaround_35388 ? "en" : "dis");
263 rc = efx_mcdi_mon_probe(efx);
267 efx_ptp_probe(efx, NULL);
274 efx_nic_free_buffer(efx, &nic_data->mcdi_buf);
277 efx->nic_data = NULL;
281 static int efx_ef10_free_vis(struct efx_nic *efx)
283 MCDI_DECLARE_BUF_OUT_OR_ERR(outbuf, 0);
285 int rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FREE_VIS, NULL, 0,
286 outbuf, sizeof(outbuf), &outlen);
288 /* -EALREADY means nothing to free, so ignore */
292 efx_mcdi_display_error(efx, MC_CMD_FREE_VIS, 0, outbuf, outlen,
299 static void efx_ef10_free_piobufs(struct efx_nic *efx)
301 struct efx_ef10_nic_data *nic_data = efx->nic_data;
302 MCDI_DECLARE_BUF(inbuf, MC_CMD_FREE_PIOBUF_IN_LEN);
306 BUILD_BUG_ON(MC_CMD_FREE_PIOBUF_OUT_LEN != 0);
308 for (i = 0; i < nic_data->n_piobufs; i++) {
309 MCDI_SET_DWORD(inbuf, FREE_PIOBUF_IN_PIOBUF_HANDLE,
310 nic_data->piobuf_handle[i]);
311 rc = efx_mcdi_rpc(efx, MC_CMD_FREE_PIOBUF, inbuf, sizeof(inbuf),
316 nic_data->n_piobufs = 0;
319 static int efx_ef10_alloc_piobufs(struct efx_nic *efx, unsigned int n)
321 struct efx_ef10_nic_data *nic_data = efx->nic_data;
322 MCDI_DECLARE_BUF(outbuf, MC_CMD_ALLOC_PIOBUF_OUT_LEN);
327 BUILD_BUG_ON(MC_CMD_ALLOC_PIOBUF_IN_LEN != 0);
329 for (i = 0; i < n; i++) {
330 rc = efx_mcdi_rpc(efx, MC_CMD_ALLOC_PIOBUF, NULL, 0,
331 outbuf, sizeof(outbuf), &outlen);
334 if (outlen < MC_CMD_ALLOC_PIOBUF_OUT_LEN) {
338 nic_data->piobuf_handle[i] =
339 MCDI_DWORD(outbuf, ALLOC_PIOBUF_OUT_PIOBUF_HANDLE);
340 netif_dbg(efx, probe, efx->net_dev,
341 "allocated PIO buffer %u handle %x\n", i,
342 nic_data->piobuf_handle[i]);
345 nic_data->n_piobufs = i;
347 efx_ef10_free_piobufs(efx);
351 static int efx_ef10_link_piobufs(struct efx_nic *efx)
353 struct efx_ef10_nic_data *nic_data = efx->nic_data;
354 MCDI_DECLARE_BUF(inbuf,
355 max(MC_CMD_LINK_PIOBUF_IN_LEN,
356 MC_CMD_UNLINK_PIOBUF_IN_LEN));
357 struct efx_channel *channel;
358 struct efx_tx_queue *tx_queue;
359 unsigned int offset, index;
362 BUILD_BUG_ON(MC_CMD_LINK_PIOBUF_OUT_LEN != 0);
363 BUILD_BUG_ON(MC_CMD_UNLINK_PIOBUF_OUT_LEN != 0);
365 /* Link a buffer to each VI in the write-combining mapping */
366 for (index = 0; index < nic_data->n_piobufs; ++index) {
367 MCDI_SET_DWORD(inbuf, LINK_PIOBUF_IN_PIOBUF_HANDLE,
368 nic_data->piobuf_handle[index]);
369 MCDI_SET_DWORD(inbuf, LINK_PIOBUF_IN_TXQ_INSTANCE,
370 nic_data->pio_write_vi_base + index);
371 rc = efx_mcdi_rpc(efx, MC_CMD_LINK_PIOBUF,
372 inbuf, MC_CMD_LINK_PIOBUF_IN_LEN,
375 netif_err(efx, drv, efx->net_dev,
376 "failed to link VI %u to PIO buffer %u (%d)\n",
377 nic_data->pio_write_vi_base + index, index,
381 netif_dbg(efx, probe, efx->net_dev,
382 "linked VI %u to PIO buffer %u\n",
383 nic_data->pio_write_vi_base + index, index);
386 /* Link a buffer to each TX queue */
387 efx_for_each_channel(channel, efx) {
388 efx_for_each_channel_tx_queue(tx_queue, channel) {
389 /* We assign the PIO buffers to queues in
390 * reverse order to allow for the following
393 offset = ((efx->tx_channel_offset + efx->n_tx_channels -
394 tx_queue->channel->channel - 1) *
396 index = offset / ER_DZ_TX_PIOBUF_SIZE;
397 offset = offset % ER_DZ_TX_PIOBUF_SIZE;
399 /* When the host page size is 4K, the first
400 * host page in the WC mapping may be within
401 * the same VI page as the last TX queue. We
402 * can only link one buffer to each VI.
404 if (tx_queue->queue == nic_data->pio_write_vi_base) {
408 MCDI_SET_DWORD(inbuf,
409 LINK_PIOBUF_IN_PIOBUF_HANDLE,
410 nic_data->piobuf_handle[index]);
411 MCDI_SET_DWORD(inbuf,
412 LINK_PIOBUF_IN_TXQ_INSTANCE,
414 rc = efx_mcdi_rpc(efx, MC_CMD_LINK_PIOBUF,
415 inbuf, MC_CMD_LINK_PIOBUF_IN_LEN,
420 /* This is non-fatal; the TX path just
421 * won't use PIO for this queue
423 netif_err(efx, drv, efx->net_dev,
424 "failed to link VI %u to PIO buffer %u (%d)\n",
425 tx_queue->queue, index, rc);
426 tx_queue->piobuf = NULL;
429 nic_data->pio_write_base +
430 index * EFX_VI_PAGE_SIZE + offset;
431 tx_queue->piobuf_offset = offset;
432 netif_dbg(efx, probe, efx->net_dev,
433 "linked VI %u to PIO buffer %u offset %x addr %p\n",
434 tx_queue->queue, index,
435 tx_queue->piobuf_offset,
445 MCDI_SET_DWORD(inbuf, UNLINK_PIOBUF_IN_TXQ_INSTANCE,
446 nic_data->pio_write_vi_base + index);
447 efx_mcdi_rpc(efx, MC_CMD_UNLINK_PIOBUF,
448 inbuf, MC_CMD_UNLINK_PIOBUF_IN_LEN,
454 #else /* !EFX_USE_PIO */
456 static int efx_ef10_alloc_piobufs(struct efx_nic *efx, unsigned int n)
458 return n == 0 ? 0 : -ENOBUFS;
461 static int efx_ef10_link_piobufs(struct efx_nic *efx)
466 static void efx_ef10_free_piobufs(struct efx_nic *efx)
470 #endif /* EFX_USE_PIO */
472 static void efx_ef10_remove(struct efx_nic *efx)
474 struct efx_ef10_nic_data *nic_data = efx->nic_data;
479 efx_mcdi_mon_remove(efx);
481 efx_ef10_rx_free_indir_table(efx);
483 if (nic_data->wc_membase)
484 iounmap(nic_data->wc_membase);
486 rc = efx_ef10_free_vis(efx);
489 if (!nic_data->must_restore_piobufs)
490 efx_ef10_free_piobufs(efx);
493 efx_nic_free_buffer(efx, &nic_data->mcdi_buf);
497 static int efx_ef10_alloc_vis(struct efx_nic *efx,
498 unsigned int min_vis, unsigned int max_vis)
500 MCDI_DECLARE_BUF(inbuf, MC_CMD_ALLOC_VIS_IN_LEN);
501 MCDI_DECLARE_BUF(outbuf, MC_CMD_ALLOC_VIS_OUT_LEN);
502 struct efx_ef10_nic_data *nic_data = efx->nic_data;
506 MCDI_SET_DWORD(inbuf, ALLOC_VIS_IN_MIN_VI_COUNT, min_vis);
507 MCDI_SET_DWORD(inbuf, ALLOC_VIS_IN_MAX_VI_COUNT, max_vis);
508 rc = efx_mcdi_rpc(efx, MC_CMD_ALLOC_VIS, inbuf, sizeof(inbuf),
509 outbuf, sizeof(outbuf), &outlen);
513 if (outlen < MC_CMD_ALLOC_VIS_OUT_LEN)
516 netif_dbg(efx, drv, efx->net_dev, "base VI is A0x%03x\n",
517 MCDI_DWORD(outbuf, ALLOC_VIS_OUT_VI_BASE));
519 nic_data->vi_base = MCDI_DWORD(outbuf, ALLOC_VIS_OUT_VI_BASE);
520 nic_data->n_allocated_vis = MCDI_DWORD(outbuf, ALLOC_VIS_OUT_VI_COUNT);
524 /* Note that the failure path of this function does not free
525 * resources, as this will be done by efx_ef10_remove().
527 static int efx_ef10_dimension_resources(struct efx_nic *efx)
529 struct efx_ef10_nic_data *nic_data = efx->nic_data;
530 unsigned int uc_mem_map_size, wc_mem_map_size;
531 unsigned int min_vis, pio_write_vi_base, max_vis;
532 void __iomem *membase;
535 min_vis = max(efx->n_channels, efx->n_tx_channels * EFX_TXQ_TYPES);
538 /* Try to allocate PIO buffers if wanted and if the full
539 * number of PIO buffers would be sufficient to allocate one
540 * copy-buffer per TX channel. Failure is non-fatal, as there
541 * are only a small number of PIO buffers shared between all
542 * functions of the controller.
544 if (efx_piobuf_size != 0 &&
545 ER_DZ_TX_PIOBUF_SIZE / efx_piobuf_size * EF10_TX_PIOBUF_COUNT >=
546 efx->n_tx_channels) {
547 unsigned int n_piobufs =
548 DIV_ROUND_UP(efx->n_tx_channels,
549 ER_DZ_TX_PIOBUF_SIZE / efx_piobuf_size);
551 rc = efx_ef10_alloc_piobufs(efx, n_piobufs);
553 netif_err(efx, probe, efx->net_dev,
554 "failed to allocate PIO buffers (%d)\n", rc);
556 netif_dbg(efx, probe, efx->net_dev,
557 "allocated %u PIO buffers\n", n_piobufs);
560 nic_data->n_piobufs = 0;
563 /* PIO buffers should be mapped with write-combining enabled,
564 * and we want to make single UC and WC mappings rather than
565 * several of each (in fact that's the only option if host
566 * page size is >4K). So we may allocate some extra VIs just
567 * for writing PIO buffers through.
569 * The UC mapping contains (min_vis - 1) complete VIs and the
570 * first half of the next VI. Then the WC mapping begins with
571 * the second half of this last VI.
573 uc_mem_map_size = PAGE_ALIGN((min_vis - 1) * EFX_VI_PAGE_SIZE +
575 if (nic_data->n_piobufs) {
576 /* pio_write_vi_base rounds down to give the number of complete
577 * VIs inside the UC mapping.
579 pio_write_vi_base = uc_mem_map_size / EFX_VI_PAGE_SIZE;
580 wc_mem_map_size = (PAGE_ALIGN((pio_write_vi_base +
581 nic_data->n_piobufs) *
584 max_vis = pio_write_vi_base + nic_data->n_piobufs;
586 pio_write_vi_base = 0;
591 /* In case the last attached driver failed to free VIs, do it now */
592 rc = efx_ef10_free_vis(efx);
596 rc = efx_ef10_alloc_vis(efx, min_vis, max_vis);
600 /* If we didn't get enough VIs to map all the PIO buffers, free the
603 if (nic_data->n_piobufs &&
604 nic_data->n_allocated_vis <
605 pio_write_vi_base + nic_data->n_piobufs) {
606 netif_dbg(efx, probe, efx->net_dev,
607 "%u VIs are not sufficient to map %u PIO buffers\n",
608 nic_data->n_allocated_vis, nic_data->n_piobufs);
609 efx_ef10_free_piobufs(efx);
612 /* Shrink the original UC mapping of the memory BAR */
613 membase = ioremap_nocache(efx->membase_phys, uc_mem_map_size);
615 netif_err(efx, probe, efx->net_dev,
616 "could not shrink memory BAR to %x\n",
620 iounmap(efx->membase);
621 efx->membase = membase;
623 /* Set up the WC mapping if needed */
624 if (wc_mem_map_size) {
625 nic_data->wc_membase = ioremap_wc(efx->membase_phys +
628 if (!nic_data->wc_membase) {
629 netif_err(efx, probe, efx->net_dev,
630 "could not allocate WC mapping of size %x\n",
634 nic_data->pio_write_vi_base = pio_write_vi_base;
635 nic_data->pio_write_base =
636 nic_data->wc_membase +
637 (pio_write_vi_base * EFX_VI_PAGE_SIZE + ER_DZ_TX_PIOBUF -
640 rc = efx_ef10_link_piobufs(efx);
642 efx_ef10_free_piobufs(efx);
645 netif_dbg(efx, probe, efx->net_dev,
646 "memory BAR at %pa (virtual %p+%x UC, %p+%x WC)\n",
647 &efx->membase_phys, efx->membase, uc_mem_map_size,
648 nic_data->wc_membase, wc_mem_map_size);
653 static int efx_ef10_init_nic(struct efx_nic *efx)
655 struct efx_ef10_nic_data *nic_data = efx->nic_data;
658 if (nic_data->must_check_datapath_caps) {
659 rc = efx_ef10_init_datapath_caps(efx);
662 nic_data->must_check_datapath_caps = false;
665 if (nic_data->must_realloc_vis) {
666 /* We cannot let the number of VIs change now */
667 rc = efx_ef10_alloc_vis(efx, nic_data->n_allocated_vis,
668 nic_data->n_allocated_vis);
671 nic_data->must_realloc_vis = false;
674 if (nic_data->must_restore_piobufs && nic_data->n_piobufs) {
675 rc = efx_ef10_alloc_piobufs(efx, nic_data->n_piobufs);
677 rc = efx_ef10_link_piobufs(efx);
679 efx_ef10_free_piobufs(efx);
682 /* Log an error on failure, but this is non-fatal */
684 netif_err(efx, drv, efx->net_dev,
685 "failed to restore PIO buffers (%d)\n", rc);
686 nic_data->must_restore_piobufs = false;
689 efx_ef10_rx_push_rss_config(efx);
693 static void efx_ef10_reset_mc_allocations(struct efx_nic *efx)
695 struct efx_ef10_nic_data *nic_data = efx->nic_data;
697 /* All our allocations have been reset */
698 nic_data->must_realloc_vis = true;
699 nic_data->must_restore_filters = true;
700 nic_data->must_restore_piobufs = true;
701 nic_data->rx_rss_context = EFX_EF10_RSS_CONTEXT_INVALID;
704 static int efx_ef10_map_reset_flags(u32 *flags)
707 EF10_RESET_PORT = ((ETH_RESET_MAC | ETH_RESET_PHY) <<
708 ETH_RESET_SHARED_SHIFT),
709 EF10_RESET_MC = ((ETH_RESET_DMA | ETH_RESET_FILTER |
710 ETH_RESET_OFFLOAD | ETH_RESET_MAC |
711 ETH_RESET_PHY | ETH_RESET_MGMT) <<
712 ETH_RESET_SHARED_SHIFT)
715 /* We assume for now that our PCI function is permitted to
719 if ((*flags & EF10_RESET_MC) == EF10_RESET_MC) {
720 *flags &= ~EF10_RESET_MC;
721 return RESET_TYPE_WORLD;
724 if ((*flags & EF10_RESET_PORT) == EF10_RESET_PORT) {
725 *flags &= ~EF10_RESET_PORT;
726 return RESET_TYPE_ALL;
729 /* no invisible reset implemented */
734 static int efx_ef10_reset(struct efx_nic *efx, enum reset_type reset_type)
736 int rc = efx_mcdi_reset(efx, reset_type);
738 /* If it was a port reset, trigger reallocation of MC resources.
739 * Note that on an MC reset nothing needs to be done now because we'll
740 * detect the MC reset later and handle it then.
741 * For an FLR, we never get an MC reset event, but the MC has reset all
742 * resources assigned to us, so we have to trigger reallocation now.
744 if ((reset_type == RESET_TYPE_ALL ||
745 reset_type == RESET_TYPE_MCDI_TIMEOUT) && !rc)
746 efx_ef10_reset_mc_allocations(efx);
750 #define EF10_DMA_STAT(ext_name, mcdi_name) \
751 [EF10_STAT_ ## ext_name] = \
752 { #ext_name, 64, 8 * MC_CMD_MAC_ ## mcdi_name }
753 #define EF10_DMA_INVIS_STAT(int_name, mcdi_name) \
754 [EF10_STAT_ ## int_name] = \
755 { NULL, 64, 8 * MC_CMD_MAC_ ## mcdi_name }
756 #define EF10_OTHER_STAT(ext_name) \
757 [EF10_STAT_ ## ext_name] = { #ext_name, 0, 0 }
758 #define GENERIC_SW_STAT(ext_name) \
759 [GENERIC_STAT_ ## ext_name] = { #ext_name, 0, 0 }
761 static const struct efx_hw_stat_desc efx_ef10_stat_desc[EF10_STAT_COUNT] = {
762 EF10_DMA_STAT(tx_bytes, TX_BYTES),
763 EF10_DMA_STAT(tx_packets, TX_PKTS),
764 EF10_DMA_STAT(tx_pause, TX_PAUSE_PKTS),
765 EF10_DMA_STAT(tx_control, TX_CONTROL_PKTS),
766 EF10_DMA_STAT(tx_unicast, TX_UNICAST_PKTS),
767 EF10_DMA_STAT(tx_multicast, TX_MULTICAST_PKTS),
768 EF10_DMA_STAT(tx_broadcast, TX_BROADCAST_PKTS),
769 EF10_DMA_STAT(tx_lt64, TX_LT64_PKTS),
770 EF10_DMA_STAT(tx_64, TX_64_PKTS),
771 EF10_DMA_STAT(tx_65_to_127, TX_65_TO_127_PKTS),
772 EF10_DMA_STAT(tx_128_to_255, TX_128_TO_255_PKTS),
773 EF10_DMA_STAT(tx_256_to_511, TX_256_TO_511_PKTS),
774 EF10_DMA_STAT(tx_512_to_1023, TX_512_TO_1023_PKTS),
775 EF10_DMA_STAT(tx_1024_to_15xx, TX_1024_TO_15XX_PKTS),
776 EF10_DMA_STAT(tx_15xx_to_jumbo, TX_15XX_TO_JUMBO_PKTS),
777 EF10_DMA_STAT(rx_bytes, RX_BYTES),
778 EF10_DMA_INVIS_STAT(rx_bytes_minus_good_bytes, RX_BAD_BYTES),
779 EF10_OTHER_STAT(rx_good_bytes),
780 EF10_OTHER_STAT(rx_bad_bytes),
781 EF10_DMA_STAT(rx_packets, RX_PKTS),
782 EF10_DMA_STAT(rx_good, RX_GOOD_PKTS),
783 EF10_DMA_STAT(rx_bad, RX_BAD_FCS_PKTS),
784 EF10_DMA_STAT(rx_pause, RX_PAUSE_PKTS),
785 EF10_DMA_STAT(rx_control, RX_CONTROL_PKTS),
786 EF10_DMA_STAT(rx_unicast, RX_UNICAST_PKTS),
787 EF10_DMA_STAT(rx_multicast, RX_MULTICAST_PKTS),
788 EF10_DMA_STAT(rx_broadcast, RX_BROADCAST_PKTS),
789 EF10_DMA_STAT(rx_lt64, RX_UNDERSIZE_PKTS),
790 EF10_DMA_STAT(rx_64, RX_64_PKTS),
791 EF10_DMA_STAT(rx_65_to_127, RX_65_TO_127_PKTS),
792 EF10_DMA_STAT(rx_128_to_255, RX_128_TO_255_PKTS),
793 EF10_DMA_STAT(rx_256_to_511, RX_256_TO_511_PKTS),
794 EF10_DMA_STAT(rx_512_to_1023, RX_512_TO_1023_PKTS),
795 EF10_DMA_STAT(rx_1024_to_15xx, RX_1024_TO_15XX_PKTS),
796 EF10_DMA_STAT(rx_15xx_to_jumbo, RX_15XX_TO_JUMBO_PKTS),
797 EF10_DMA_STAT(rx_gtjumbo, RX_GTJUMBO_PKTS),
798 EF10_DMA_STAT(rx_bad_gtjumbo, RX_JABBER_PKTS),
799 EF10_DMA_STAT(rx_overflow, RX_OVERFLOW_PKTS),
800 EF10_DMA_STAT(rx_align_error, RX_ALIGN_ERROR_PKTS),
801 EF10_DMA_STAT(rx_length_error, RX_LENGTH_ERROR_PKTS),
802 EF10_DMA_STAT(rx_nodesc_drops, RX_NODESC_DROPS),
803 GENERIC_SW_STAT(rx_nodesc_trunc),
804 GENERIC_SW_STAT(rx_noskb_drops),
805 EF10_DMA_STAT(rx_pm_trunc_bb_overflow, PM_TRUNC_BB_OVERFLOW),
806 EF10_DMA_STAT(rx_pm_discard_bb_overflow, PM_DISCARD_BB_OVERFLOW),
807 EF10_DMA_STAT(rx_pm_trunc_vfifo_full, PM_TRUNC_VFIFO_FULL),
808 EF10_DMA_STAT(rx_pm_discard_vfifo_full, PM_DISCARD_VFIFO_FULL),
809 EF10_DMA_STAT(rx_pm_trunc_qbb, PM_TRUNC_QBB),
810 EF10_DMA_STAT(rx_pm_discard_qbb, PM_DISCARD_QBB),
811 EF10_DMA_STAT(rx_pm_discard_mapping, PM_DISCARD_MAPPING),
812 EF10_DMA_STAT(rx_dp_q_disabled_packets, RXDP_Q_DISABLED_PKTS),
813 EF10_DMA_STAT(rx_dp_di_dropped_packets, RXDP_DI_DROPPED_PKTS),
814 EF10_DMA_STAT(rx_dp_streaming_packets, RXDP_STREAMING_PKTS),
815 EF10_DMA_STAT(rx_dp_hlb_fetch, RXDP_EMERGENCY_FETCH_CONDITIONS),
816 EF10_DMA_STAT(rx_dp_hlb_wait, RXDP_EMERGENCY_WAIT_CONDITIONS),
819 #define HUNT_COMMON_STAT_MASK ((1ULL << EF10_STAT_tx_bytes) | \
820 (1ULL << EF10_STAT_tx_packets) | \
821 (1ULL << EF10_STAT_tx_pause) | \
822 (1ULL << EF10_STAT_tx_unicast) | \
823 (1ULL << EF10_STAT_tx_multicast) | \
824 (1ULL << EF10_STAT_tx_broadcast) | \
825 (1ULL << EF10_STAT_rx_bytes) | \
826 (1ULL << EF10_STAT_rx_bytes_minus_good_bytes) | \
827 (1ULL << EF10_STAT_rx_good_bytes) | \
828 (1ULL << EF10_STAT_rx_bad_bytes) | \
829 (1ULL << EF10_STAT_rx_packets) | \
830 (1ULL << EF10_STAT_rx_good) | \
831 (1ULL << EF10_STAT_rx_bad) | \
832 (1ULL << EF10_STAT_rx_pause) | \
833 (1ULL << EF10_STAT_rx_control) | \
834 (1ULL << EF10_STAT_rx_unicast) | \
835 (1ULL << EF10_STAT_rx_multicast) | \
836 (1ULL << EF10_STAT_rx_broadcast) | \
837 (1ULL << EF10_STAT_rx_lt64) | \
838 (1ULL << EF10_STAT_rx_64) | \
839 (1ULL << EF10_STAT_rx_65_to_127) | \
840 (1ULL << EF10_STAT_rx_128_to_255) | \
841 (1ULL << EF10_STAT_rx_256_to_511) | \
842 (1ULL << EF10_STAT_rx_512_to_1023) | \
843 (1ULL << EF10_STAT_rx_1024_to_15xx) | \
844 (1ULL << EF10_STAT_rx_15xx_to_jumbo) | \
845 (1ULL << EF10_STAT_rx_gtjumbo) | \
846 (1ULL << EF10_STAT_rx_bad_gtjumbo) | \
847 (1ULL << EF10_STAT_rx_overflow) | \
848 (1ULL << EF10_STAT_rx_nodesc_drops) | \
849 (1ULL << GENERIC_STAT_rx_nodesc_trunc) | \
850 (1ULL << GENERIC_STAT_rx_noskb_drops))
852 /* These statistics are only provided by the 10G MAC. For a 10G/40G
853 * switchable port we do not expose these because they might not
854 * include all the packets they should.
856 #define HUNT_10G_ONLY_STAT_MASK ((1ULL << EF10_STAT_tx_control) | \
857 (1ULL << EF10_STAT_tx_lt64) | \
858 (1ULL << EF10_STAT_tx_64) | \
859 (1ULL << EF10_STAT_tx_65_to_127) | \
860 (1ULL << EF10_STAT_tx_128_to_255) | \
861 (1ULL << EF10_STAT_tx_256_to_511) | \
862 (1ULL << EF10_STAT_tx_512_to_1023) | \
863 (1ULL << EF10_STAT_tx_1024_to_15xx) | \
864 (1ULL << EF10_STAT_tx_15xx_to_jumbo))
866 /* These statistics are only provided by the 40G MAC. For a 10G/40G
867 * switchable port we do expose these because the errors will otherwise
870 #define HUNT_40G_EXTRA_STAT_MASK ((1ULL << EF10_STAT_rx_align_error) | \
871 (1ULL << EF10_STAT_rx_length_error))
873 /* These statistics are only provided if the firmware supports the
874 * capability PM_AND_RXDP_COUNTERS.
876 #define HUNT_PM_AND_RXDP_STAT_MASK ( \
877 (1ULL << EF10_STAT_rx_pm_trunc_bb_overflow) | \
878 (1ULL << EF10_STAT_rx_pm_discard_bb_overflow) | \
879 (1ULL << EF10_STAT_rx_pm_trunc_vfifo_full) | \
880 (1ULL << EF10_STAT_rx_pm_discard_vfifo_full) | \
881 (1ULL << EF10_STAT_rx_pm_trunc_qbb) | \
882 (1ULL << EF10_STAT_rx_pm_discard_qbb) | \
883 (1ULL << EF10_STAT_rx_pm_discard_mapping) | \
884 (1ULL << EF10_STAT_rx_dp_q_disabled_packets) | \
885 (1ULL << EF10_STAT_rx_dp_di_dropped_packets) | \
886 (1ULL << EF10_STAT_rx_dp_streaming_packets) | \
887 (1ULL << EF10_STAT_rx_dp_hlb_fetch) | \
888 (1ULL << EF10_STAT_rx_dp_hlb_wait))
890 static u64 efx_ef10_raw_stat_mask(struct efx_nic *efx)
892 u64 raw_mask = HUNT_COMMON_STAT_MASK;
893 u32 port_caps = efx_mcdi_phy_get_caps(efx);
894 struct efx_ef10_nic_data *nic_data = efx->nic_data;
896 if (port_caps & (1 << MC_CMD_PHY_CAP_40000FDX_LBN))
897 raw_mask |= HUNT_40G_EXTRA_STAT_MASK;
899 raw_mask |= HUNT_10G_ONLY_STAT_MASK;
901 if (nic_data->datapath_caps &
902 (1 << MC_CMD_GET_CAPABILITIES_OUT_PM_AND_RXDP_COUNTERS_LBN))
903 raw_mask |= HUNT_PM_AND_RXDP_STAT_MASK;
908 static void efx_ef10_get_stat_mask(struct efx_nic *efx, unsigned long *mask)
910 u64 raw_mask = efx_ef10_raw_stat_mask(efx);
912 #if BITS_PER_LONG == 64
915 mask[0] = raw_mask & 0xffffffff;
916 mask[1] = raw_mask >> 32;
920 static size_t efx_ef10_describe_stats(struct efx_nic *efx, u8 *names)
922 DECLARE_BITMAP(mask, EF10_STAT_COUNT);
924 efx_ef10_get_stat_mask(efx, mask);
925 return efx_nic_describe_stats(efx_ef10_stat_desc, EF10_STAT_COUNT,
929 static int efx_ef10_try_update_nic_stats(struct efx_nic *efx)
931 struct efx_ef10_nic_data *nic_data = efx->nic_data;
932 DECLARE_BITMAP(mask, EF10_STAT_COUNT);
933 __le64 generation_start, generation_end;
934 u64 *stats = nic_data->stats;
937 efx_ef10_get_stat_mask(efx, mask);
939 dma_stats = efx->stats_buffer.addr;
940 nic_data = efx->nic_data;
942 generation_end = dma_stats[MC_CMD_MAC_GENERATION_END];
943 if (generation_end == EFX_MC_STATS_GENERATION_INVALID)
946 efx_nic_update_stats(efx_ef10_stat_desc, EF10_STAT_COUNT, mask,
947 stats, efx->stats_buffer.addr, false);
949 generation_start = dma_stats[MC_CMD_MAC_GENERATION_START];
950 if (generation_end != generation_start)
953 /* Update derived statistics */
954 efx_nic_fix_nodesc_drop_stat(efx, &stats[EF10_STAT_rx_nodesc_drops]);
955 stats[EF10_STAT_rx_good_bytes] =
956 stats[EF10_STAT_rx_bytes] -
957 stats[EF10_STAT_rx_bytes_minus_good_bytes];
958 efx_update_diff_stat(&stats[EF10_STAT_rx_bad_bytes],
959 stats[EF10_STAT_rx_bytes_minus_good_bytes]);
960 efx_update_sw_stats(efx, stats);
965 static size_t efx_ef10_update_stats(struct efx_nic *efx, u64 *full_stats,
966 struct rtnl_link_stats64 *core_stats)
968 DECLARE_BITMAP(mask, EF10_STAT_COUNT);
969 struct efx_ef10_nic_data *nic_data = efx->nic_data;
970 u64 *stats = nic_data->stats;
971 size_t stats_count = 0, index;
974 efx_ef10_get_stat_mask(efx, mask);
976 /* If we're unlucky enough to read statistics during the DMA, wait
977 * up to 10ms for it to finish (typically takes <500us)
979 for (retry = 0; retry < 100; ++retry) {
980 if (efx_ef10_try_update_nic_stats(efx) == 0)
986 for_each_set_bit(index, mask, EF10_STAT_COUNT) {
987 if (efx_ef10_stat_desc[index].name) {
988 *full_stats++ = stats[index];
995 core_stats->rx_packets = stats[EF10_STAT_rx_packets];
996 core_stats->tx_packets = stats[EF10_STAT_tx_packets];
997 core_stats->rx_bytes = stats[EF10_STAT_rx_bytes];
998 core_stats->tx_bytes = stats[EF10_STAT_tx_bytes];
999 core_stats->rx_dropped = stats[EF10_STAT_rx_nodesc_drops] +
1000 stats[GENERIC_STAT_rx_nodesc_trunc] +
1001 stats[GENERIC_STAT_rx_noskb_drops];
1002 core_stats->multicast = stats[EF10_STAT_rx_multicast];
1003 core_stats->rx_length_errors =
1004 stats[EF10_STAT_rx_gtjumbo] +
1005 stats[EF10_STAT_rx_length_error];
1006 core_stats->rx_crc_errors = stats[EF10_STAT_rx_bad];
1007 core_stats->rx_frame_errors = stats[EF10_STAT_rx_align_error];
1008 core_stats->rx_fifo_errors = stats[EF10_STAT_rx_overflow];
1009 core_stats->rx_errors = (core_stats->rx_length_errors +
1010 core_stats->rx_crc_errors +
1011 core_stats->rx_frame_errors);
1017 static void efx_ef10_push_irq_moderation(struct efx_channel *channel)
1019 struct efx_nic *efx = channel->efx;
1020 unsigned int mode, value;
1021 efx_dword_t timer_cmd;
1023 if (channel->irq_moderation) {
1025 value = channel->irq_moderation - 1;
1031 if (EFX_EF10_WORKAROUND_35388(efx)) {
1032 EFX_POPULATE_DWORD_3(timer_cmd, ERF_DD_EVQ_IND_TIMER_FLAGS,
1033 EFE_DD_EVQ_IND_TIMER_FLAGS,
1034 ERF_DD_EVQ_IND_TIMER_MODE, mode,
1035 ERF_DD_EVQ_IND_TIMER_VAL, value);
1036 efx_writed_page(efx, &timer_cmd, ER_DD_EVQ_INDIRECT,
1039 EFX_POPULATE_DWORD_2(timer_cmd, ERF_DZ_TC_TIMER_MODE, mode,
1040 ERF_DZ_TC_TIMER_VAL, value);
1041 efx_writed_page(efx, &timer_cmd, ER_DZ_EVQ_TMR,
1046 static void efx_ef10_get_wol(struct efx_nic *efx, struct ethtool_wolinfo *wol)
1050 memset(&wol->sopass, 0, sizeof(wol->sopass));
1053 static int efx_ef10_set_wol(struct efx_nic *efx, u32 type)
1060 static void efx_ef10_mcdi_request(struct efx_nic *efx,
1061 const efx_dword_t *hdr, size_t hdr_len,
1062 const efx_dword_t *sdu, size_t sdu_len)
1064 struct efx_ef10_nic_data *nic_data = efx->nic_data;
1065 u8 *pdu = nic_data->mcdi_buf.addr;
1067 memcpy(pdu, hdr, hdr_len);
1068 memcpy(pdu + hdr_len, sdu, sdu_len);
1071 /* The hardware provides 'low' and 'high' (doorbell) registers
1072 * for passing the 64-bit address of an MCDI request to
1073 * firmware. However the dwords are swapped by firmware. The
1074 * least significant bits of the doorbell are then 0 for all
1075 * MCDI requests due to alignment.
1077 _efx_writed(efx, cpu_to_le32((u64)nic_data->mcdi_buf.dma_addr >> 32),
1079 _efx_writed(efx, cpu_to_le32((u32)nic_data->mcdi_buf.dma_addr),
1083 static bool efx_ef10_mcdi_poll_response(struct efx_nic *efx)
1085 struct efx_ef10_nic_data *nic_data = efx->nic_data;
1086 const efx_dword_t hdr = *(const efx_dword_t *)nic_data->mcdi_buf.addr;
1089 return EFX_DWORD_FIELD(hdr, MCDI_HEADER_RESPONSE);
1093 efx_ef10_mcdi_read_response(struct efx_nic *efx, efx_dword_t *outbuf,
1094 size_t offset, size_t outlen)
1096 struct efx_ef10_nic_data *nic_data = efx->nic_data;
1097 const u8 *pdu = nic_data->mcdi_buf.addr;
1099 memcpy(outbuf, pdu + offset, outlen);
1102 static int efx_ef10_mcdi_poll_reboot(struct efx_nic *efx)
1104 struct efx_ef10_nic_data *nic_data = efx->nic_data;
1107 rc = efx_ef10_get_warm_boot_count(efx);
1109 /* The firmware is presumably in the process of
1110 * rebooting. However, we are supposed to report each
1111 * reboot just once, so we must only do that once we
1112 * can read and store the updated warm boot count.
1117 if (rc == nic_data->warm_boot_count)
1120 nic_data->warm_boot_count = rc;
1122 /* All our allocations have been reset */
1123 efx_ef10_reset_mc_allocations(efx);
1125 /* The datapath firmware might have been changed */
1126 nic_data->must_check_datapath_caps = true;
1128 /* MAC statistics have been cleared on the NIC; clear the local
1129 * statistic that we update with efx_update_diff_stat().
1131 nic_data->stats[EF10_STAT_rx_bad_bytes] = 0;
1136 /* Handle an MSI interrupt
1138 * Handle an MSI hardware interrupt. This routine schedules event
1139 * queue processing. No interrupt acknowledgement cycle is necessary.
1140 * Also, we never need to check that the interrupt is for us, since
1141 * MSI interrupts cannot be shared.
1143 static irqreturn_t efx_ef10_msi_interrupt(int irq, void *dev_id)
1145 struct efx_msi_context *context = dev_id;
1146 struct efx_nic *efx = context->efx;
1148 netif_vdbg(efx, intr, efx->net_dev,
1149 "IRQ %d on CPU %d\n", irq, raw_smp_processor_id());
1151 if (likely(ACCESS_ONCE(efx->irq_soft_enabled))) {
1152 /* Note test interrupts */
1153 if (context->index == efx->irq_level)
1154 efx->last_irq_cpu = raw_smp_processor_id();
1156 /* Schedule processing of the channel */
1157 efx_schedule_channel_irq(efx->channel[context->index]);
1163 static irqreturn_t efx_ef10_legacy_interrupt(int irq, void *dev_id)
1165 struct efx_nic *efx = dev_id;
1166 bool soft_enabled = ACCESS_ONCE(efx->irq_soft_enabled);
1167 struct efx_channel *channel;
1171 /* Read the ISR which also ACKs the interrupts */
1172 efx_readd(efx, ®, ER_DZ_BIU_INT_ISR);
1173 queues = EFX_DWORD_FIELD(reg, ERF_DZ_ISR_REG);
1178 if (likely(soft_enabled)) {
1179 /* Note test interrupts */
1180 if (queues & (1U << efx->irq_level))
1181 efx->last_irq_cpu = raw_smp_processor_id();
1183 efx_for_each_channel(channel, efx) {
1185 efx_schedule_channel_irq(channel);
1190 netif_vdbg(efx, intr, efx->net_dev,
1191 "IRQ %d on CPU %d status " EFX_DWORD_FMT "\n",
1192 irq, raw_smp_processor_id(), EFX_DWORD_VAL(reg));
1197 static void efx_ef10_irq_test_generate(struct efx_nic *efx)
1199 MCDI_DECLARE_BUF(inbuf, MC_CMD_TRIGGER_INTERRUPT_IN_LEN);
1201 BUILD_BUG_ON(MC_CMD_TRIGGER_INTERRUPT_OUT_LEN != 0);
1203 MCDI_SET_DWORD(inbuf, TRIGGER_INTERRUPT_IN_INTR_LEVEL, efx->irq_level);
1204 (void) efx_mcdi_rpc(efx, MC_CMD_TRIGGER_INTERRUPT,
1205 inbuf, sizeof(inbuf), NULL, 0, NULL);
1208 static int efx_ef10_tx_probe(struct efx_tx_queue *tx_queue)
1210 return efx_nic_alloc_buffer(tx_queue->efx, &tx_queue->txd.buf,
1211 (tx_queue->ptr_mask + 1) *
1212 sizeof(efx_qword_t),
1216 /* This writes to the TX_DESC_WPTR and also pushes data */
1217 static inline void efx_ef10_push_tx_desc(struct efx_tx_queue *tx_queue,
1218 const efx_qword_t *txd)
1220 unsigned int write_ptr;
1223 write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
1224 EFX_POPULATE_OWORD_1(reg, ERF_DZ_TX_DESC_WPTR, write_ptr);
1225 reg.qword[0] = *txd;
1226 efx_writeo_page(tx_queue->efx, ®,
1227 ER_DZ_TX_DESC_UPD, tx_queue->queue);
1230 static void efx_ef10_tx_init(struct efx_tx_queue *tx_queue)
1232 MCDI_DECLARE_BUF(inbuf, MC_CMD_INIT_TXQ_IN_LEN(EFX_MAX_DMAQ_SIZE * 8 /
1234 MCDI_DECLARE_BUF(outbuf, MC_CMD_INIT_TXQ_OUT_LEN);
1235 bool csum_offload = tx_queue->queue & EFX_TXQ_TYPE_OFFLOAD;
1236 size_t entries = tx_queue->txd.buf.len / EFX_BUF_SIZE;
1237 struct efx_channel *channel = tx_queue->channel;
1238 struct efx_nic *efx = tx_queue->efx;
1239 size_t inlen, outlen;
1240 dma_addr_t dma_addr;
1245 MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_SIZE, tx_queue->ptr_mask + 1);
1246 MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_TARGET_EVQ, channel->channel);
1247 MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_LABEL, tx_queue->queue);
1248 MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_INSTANCE, tx_queue->queue);
1249 MCDI_POPULATE_DWORD_2(inbuf, INIT_TXQ_IN_FLAGS,
1250 INIT_TXQ_IN_FLAG_IP_CSUM_DIS, !csum_offload,
1251 INIT_TXQ_IN_FLAG_TCP_CSUM_DIS, !csum_offload);
1252 MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_OWNER_ID, 0);
1253 MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_PORT_ID, EVB_PORT_ID_ASSIGNED);
1255 dma_addr = tx_queue->txd.buf.dma_addr;
1257 netif_dbg(efx, hw, efx->net_dev, "pushing TXQ %d. %zu entries (%llx)\n",
1258 tx_queue->queue, entries, (u64)dma_addr);
1260 for (i = 0; i < entries; ++i) {
1261 MCDI_SET_ARRAY_QWORD(inbuf, INIT_TXQ_IN_DMA_ADDR, i, dma_addr);
1262 dma_addr += EFX_BUF_SIZE;
1265 inlen = MC_CMD_INIT_TXQ_IN_LEN(entries);
1267 rc = efx_mcdi_rpc(efx, MC_CMD_INIT_TXQ, inbuf, inlen,
1268 outbuf, sizeof(outbuf), &outlen);
1272 /* A previous user of this TX queue might have set us up the
1273 * bomb by writing a descriptor to the TX push collector but
1274 * not the doorbell. (Each collector belongs to a port, not a
1275 * queue or function, so cannot easily be reset.) We must
1276 * attempt to push a no-op descriptor in its place.
1278 tx_queue->buffer[0].flags = EFX_TX_BUF_OPTION;
1279 tx_queue->insert_count = 1;
1280 txd = efx_tx_desc(tx_queue, 0);
1281 EFX_POPULATE_QWORD_4(*txd,
1282 ESF_DZ_TX_DESC_IS_OPT, true,
1283 ESF_DZ_TX_OPTION_TYPE,
1284 ESE_DZ_TX_OPTION_DESC_CRC_CSUM,
1285 ESF_DZ_TX_OPTION_UDP_TCP_CSUM, csum_offload,
1286 ESF_DZ_TX_OPTION_IP_CSUM, csum_offload);
1287 tx_queue->write_count = 1;
1289 efx_ef10_push_tx_desc(tx_queue, txd);
1294 netdev_WARN(efx->net_dev, "failed to initialise TXQ %d\n",
1298 static void efx_ef10_tx_fini(struct efx_tx_queue *tx_queue)
1300 MCDI_DECLARE_BUF(inbuf, MC_CMD_FINI_TXQ_IN_LEN);
1301 MCDI_DECLARE_BUF(outbuf, MC_CMD_FINI_TXQ_OUT_LEN);
1302 struct efx_nic *efx = tx_queue->efx;
1306 MCDI_SET_DWORD(inbuf, FINI_TXQ_IN_INSTANCE,
1309 rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FINI_TXQ, inbuf, sizeof(inbuf),
1310 outbuf, sizeof(outbuf), &outlen);
1312 if (rc && rc != -EALREADY)
1318 efx_mcdi_display_error(efx, MC_CMD_FINI_TXQ, MC_CMD_FINI_TXQ_IN_LEN,
1319 outbuf, outlen, rc);
1322 static void efx_ef10_tx_remove(struct efx_tx_queue *tx_queue)
1324 efx_nic_free_buffer(tx_queue->efx, &tx_queue->txd.buf);
1327 /* This writes to the TX_DESC_WPTR; write pointer for TX descriptor ring */
1328 static inline void efx_ef10_notify_tx_desc(struct efx_tx_queue *tx_queue)
1330 unsigned int write_ptr;
1333 write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
1334 EFX_POPULATE_DWORD_1(reg, ERF_DZ_TX_DESC_WPTR_DWORD, write_ptr);
1335 efx_writed_page(tx_queue->efx, ®,
1336 ER_DZ_TX_DESC_UPD_DWORD, tx_queue->queue);
1339 static void efx_ef10_tx_write(struct efx_tx_queue *tx_queue)
1341 unsigned int old_write_count = tx_queue->write_count;
1342 struct efx_tx_buffer *buffer;
1343 unsigned int write_ptr;
1346 BUG_ON(tx_queue->write_count == tx_queue->insert_count);
1349 write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
1350 buffer = &tx_queue->buffer[write_ptr];
1351 txd = efx_tx_desc(tx_queue, write_ptr);
1352 ++tx_queue->write_count;
1354 /* Create TX descriptor ring entry */
1355 if (buffer->flags & EFX_TX_BUF_OPTION) {
1356 *txd = buffer->option;
1358 BUILD_BUG_ON(EFX_TX_BUF_CONT != 1);
1359 EFX_POPULATE_QWORD_3(
1362 buffer->flags & EFX_TX_BUF_CONT,
1363 ESF_DZ_TX_KER_BYTE_CNT, buffer->len,
1364 ESF_DZ_TX_KER_BUF_ADDR, buffer->dma_addr);
1366 } while (tx_queue->write_count != tx_queue->insert_count);
1368 wmb(); /* Ensure descriptors are written before they are fetched */
1370 if (efx_nic_may_push_tx_desc(tx_queue, old_write_count)) {
1371 txd = efx_tx_desc(tx_queue,
1372 old_write_count & tx_queue->ptr_mask);
1373 efx_ef10_push_tx_desc(tx_queue, txd);
1376 efx_ef10_notify_tx_desc(tx_queue);
1380 static int efx_ef10_alloc_rss_context(struct efx_nic *efx, u32 *context)
1382 MCDI_DECLARE_BUF(inbuf, MC_CMD_RSS_CONTEXT_ALLOC_IN_LEN);
1383 MCDI_DECLARE_BUF(outbuf, MC_CMD_RSS_CONTEXT_ALLOC_OUT_LEN);
1387 MCDI_SET_DWORD(inbuf, RSS_CONTEXT_ALLOC_IN_UPSTREAM_PORT_ID,
1388 EVB_PORT_ID_ASSIGNED);
1389 MCDI_SET_DWORD(inbuf, RSS_CONTEXT_ALLOC_IN_TYPE,
1390 MC_CMD_RSS_CONTEXT_ALLOC_IN_TYPE_EXCLUSIVE);
1391 MCDI_SET_DWORD(inbuf, RSS_CONTEXT_ALLOC_IN_NUM_QUEUES,
1394 rc = efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_ALLOC, inbuf, sizeof(inbuf),
1395 outbuf, sizeof(outbuf), &outlen);
1399 if (outlen < MC_CMD_RSS_CONTEXT_ALLOC_OUT_LEN)
1402 *context = MCDI_DWORD(outbuf, RSS_CONTEXT_ALLOC_OUT_RSS_CONTEXT_ID);
1407 static void efx_ef10_free_rss_context(struct efx_nic *efx, u32 context)
1409 MCDI_DECLARE_BUF(inbuf, MC_CMD_RSS_CONTEXT_FREE_IN_LEN);
1412 MCDI_SET_DWORD(inbuf, RSS_CONTEXT_FREE_IN_RSS_CONTEXT_ID,
1415 rc = efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_FREE, inbuf, sizeof(inbuf),
1420 static int efx_ef10_populate_rss_table(struct efx_nic *efx, u32 context)
1422 MCDI_DECLARE_BUF(tablebuf, MC_CMD_RSS_CONTEXT_SET_TABLE_IN_LEN);
1423 MCDI_DECLARE_BUF(keybuf, MC_CMD_RSS_CONTEXT_SET_KEY_IN_LEN);
1426 MCDI_SET_DWORD(tablebuf, RSS_CONTEXT_SET_TABLE_IN_RSS_CONTEXT_ID,
1428 BUILD_BUG_ON(ARRAY_SIZE(efx->rx_indir_table) !=
1429 MC_CMD_RSS_CONTEXT_SET_TABLE_IN_INDIRECTION_TABLE_LEN);
1431 for (i = 0; i < ARRAY_SIZE(efx->rx_indir_table); ++i)
1433 RSS_CONTEXT_SET_TABLE_IN_INDIRECTION_TABLE)[i] =
1434 (u8) efx->rx_indir_table[i];
1436 rc = efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_SET_TABLE, tablebuf,
1437 sizeof(tablebuf), NULL, 0, NULL);
1441 MCDI_SET_DWORD(keybuf, RSS_CONTEXT_SET_KEY_IN_RSS_CONTEXT_ID,
1443 BUILD_BUG_ON(ARRAY_SIZE(efx->rx_hash_key) !=
1444 MC_CMD_RSS_CONTEXT_SET_KEY_IN_TOEPLITZ_KEY_LEN);
1445 for (i = 0; i < ARRAY_SIZE(efx->rx_hash_key); ++i)
1446 MCDI_PTR(keybuf, RSS_CONTEXT_SET_KEY_IN_TOEPLITZ_KEY)[i] =
1447 efx->rx_hash_key[i];
1449 return efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_SET_KEY, keybuf,
1450 sizeof(keybuf), NULL, 0, NULL);
1453 static void efx_ef10_rx_free_indir_table(struct efx_nic *efx)
1455 struct efx_ef10_nic_data *nic_data = efx->nic_data;
1457 if (nic_data->rx_rss_context != EFX_EF10_RSS_CONTEXT_INVALID)
1458 efx_ef10_free_rss_context(efx, nic_data->rx_rss_context);
1459 nic_data->rx_rss_context = EFX_EF10_RSS_CONTEXT_INVALID;
1462 static void efx_ef10_rx_push_rss_config(struct efx_nic *efx)
1464 struct efx_ef10_nic_data *nic_data = efx->nic_data;
1467 netif_dbg(efx, drv, efx->net_dev, "pushing RSS config\n");
1469 if (nic_data->rx_rss_context == EFX_EF10_RSS_CONTEXT_INVALID) {
1470 rc = efx_ef10_alloc_rss_context(efx, &nic_data->rx_rss_context);
1475 rc = efx_ef10_populate_rss_table(efx, nic_data->rx_rss_context);
1482 netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
1485 static int efx_ef10_rx_probe(struct efx_rx_queue *rx_queue)
1487 return efx_nic_alloc_buffer(rx_queue->efx, &rx_queue->rxd.buf,
1488 (rx_queue->ptr_mask + 1) *
1489 sizeof(efx_qword_t),
1493 static void efx_ef10_rx_init(struct efx_rx_queue *rx_queue)
1495 MCDI_DECLARE_BUF(inbuf,
1496 MC_CMD_INIT_RXQ_IN_LEN(EFX_MAX_DMAQ_SIZE * 8 /
1498 MCDI_DECLARE_BUF(outbuf, MC_CMD_INIT_RXQ_OUT_LEN);
1499 struct efx_channel *channel = efx_rx_queue_channel(rx_queue);
1500 size_t entries = rx_queue->rxd.buf.len / EFX_BUF_SIZE;
1501 struct efx_nic *efx = rx_queue->efx;
1502 size_t inlen, outlen;
1503 dma_addr_t dma_addr;
1507 rx_queue->scatter_n = 0;
1508 rx_queue->scatter_len = 0;
1510 MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_SIZE, rx_queue->ptr_mask + 1);
1511 MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_TARGET_EVQ, channel->channel);
1512 MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_LABEL, efx_rx_queue_index(rx_queue));
1513 MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_INSTANCE,
1514 efx_rx_queue_index(rx_queue));
1515 MCDI_POPULATE_DWORD_2(inbuf, INIT_RXQ_IN_FLAGS,
1516 INIT_RXQ_IN_FLAG_PREFIX, 1,
1517 INIT_RXQ_IN_FLAG_TIMESTAMP, 1);
1518 MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_OWNER_ID, 0);
1519 MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_PORT_ID, EVB_PORT_ID_ASSIGNED);
1521 dma_addr = rx_queue->rxd.buf.dma_addr;
1523 netif_dbg(efx, hw, efx->net_dev, "pushing RXQ %d. %zu entries (%llx)\n",
1524 efx_rx_queue_index(rx_queue), entries, (u64)dma_addr);
1526 for (i = 0; i < entries; ++i) {
1527 MCDI_SET_ARRAY_QWORD(inbuf, INIT_RXQ_IN_DMA_ADDR, i, dma_addr);
1528 dma_addr += EFX_BUF_SIZE;
1531 inlen = MC_CMD_INIT_RXQ_IN_LEN(entries);
1533 rc = efx_mcdi_rpc(efx, MC_CMD_INIT_RXQ, inbuf, inlen,
1534 outbuf, sizeof(outbuf), &outlen);
1536 netdev_WARN(efx->net_dev, "failed to initialise RXQ %d\n",
1537 efx_rx_queue_index(rx_queue));
1540 static void efx_ef10_rx_fini(struct efx_rx_queue *rx_queue)
1542 MCDI_DECLARE_BUF(inbuf, MC_CMD_FINI_RXQ_IN_LEN);
1543 MCDI_DECLARE_BUF(outbuf, MC_CMD_FINI_RXQ_OUT_LEN);
1544 struct efx_nic *efx = rx_queue->efx;
1548 MCDI_SET_DWORD(inbuf, FINI_RXQ_IN_INSTANCE,
1549 efx_rx_queue_index(rx_queue));
1551 rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FINI_RXQ, inbuf, sizeof(inbuf),
1552 outbuf, sizeof(outbuf), &outlen);
1554 if (rc && rc != -EALREADY)
1560 efx_mcdi_display_error(efx, MC_CMD_FINI_RXQ, MC_CMD_FINI_RXQ_IN_LEN,
1561 outbuf, outlen, rc);
1564 static void efx_ef10_rx_remove(struct efx_rx_queue *rx_queue)
1566 efx_nic_free_buffer(rx_queue->efx, &rx_queue->rxd.buf);
1569 /* This creates an entry in the RX descriptor queue */
1571 efx_ef10_build_rx_desc(struct efx_rx_queue *rx_queue, unsigned int index)
1573 struct efx_rx_buffer *rx_buf;
1576 rxd = efx_rx_desc(rx_queue, index);
1577 rx_buf = efx_rx_buffer(rx_queue, index);
1578 EFX_POPULATE_QWORD_2(*rxd,
1579 ESF_DZ_RX_KER_BYTE_CNT, rx_buf->len,
1580 ESF_DZ_RX_KER_BUF_ADDR, rx_buf->dma_addr);
1583 static void efx_ef10_rx_write(struct efx_rx_queue *rx_queue)
1585 struct efx_nic *efx = rx_queue->efx;
1586 unsigned int write_count;
1589 /* Firmware requires that RX_DESC_WPTR be a multiple of 8 */
1590 write_count = rx_queue->added_count & ~7;
1591 if (rx_queue->notified_count == write_count)
1595 efx_ef10_build_rx_desc(
1597 rx_queue->notified_count & rx_queue->ptr_mask);
1598 while (++rx_queue->notified_count != write_count);
1601 EFX_POPULATE_DWORD_1(reg, ERF_DZ_RX_DESC_WPTR,
1602 write_count & rx_queue->ptr_mask);
1603 efx_writed_page(efx, ®, ER_DZ_RX_DESC_UPD,
1604 efx_rx_queue_index(rx_queue));
1607 static efx_mcdi_async_completer efx_ef10_rx_defer_refill_complete;
1609 static void efx_ef10_rx_defer_refill(struct efx_rx_queue *rx_queue)
1611 struct efx_channel *channel = efx_rx_queue_channel(rx_queue);
1612 MCDI_DECLARE_BUF(inbuf, MC_CMD_DRIVER_EVENT_IN_LEN);
1615 EFX_POPULATE_QWORD_2(event,
1616 ESF_DZ_EV_CODE, EFX_EF10_DRVGEN_EV,
1617 ESF_DZ_EV_DATA, EFX_EF10_REFILL);
1619 MCDI_SET_DWORD(inbuf, DRIVER_EVENT_IN_EVQ, channel->channel);
1621 /* MCDI_SET_QWORD is not appropriate here since EFX_POPULATE_* has
1622 * already swapped the data to little-endian order.
1624 memcpy(MCDI_PTR(inbuf, DRIVER_EVENT_IN_DATA), &event.u64[0],
1625 sizeof(efx_qword_t));
1627 efx_mcdi_rpc_async(channel->efx, MC_CMD_DRIVER_EVENT,
1628 inbuf, sizeof(inbuf), 0,
1629 efx_ef10_rx_defer_refill_complete, 0);
1633 efx_ef10_rx_defer_refill_complete(struct efx_nic *efx, unsigned long cookie,
1634 int rc, efx_dword_t *outbuf,
1635 size_t outlen_actual)
1640 static int efx_ef10_ev_probe(struct efx_channel *channel)
1642 return efx_nic_alloc_buffer(channel->efx, &channel->eventq.buf,
1643 (channel->eventq_mask + 1) *
1644 sizeof(efx_qword_t),
1648 static int efx_ef10_ev_init(struct efx_channel *channel)
1650 MCDI_DECLARE_BUF(inbuf,
1651 MC_CMD_INIT_EVQ_IN_LEN(EFX_MAX_EVQ_SIZE * 8 /
1653 MCDI_DECLARE_BUF(outbuf, MC_CMD_INIT_EVQ_OUT_LEN);
1654 size_t entries = channel->eventq.buf.len / EFX_BUF_SIZE;
1655 struct efx_nic *efx = channel->efx;
1656 struct efx_ef10_nic_data *nic_data;
1657 bool supports_rx_merge;
1658 size_t inlen, outlen;
1659 dma_addr_t dma_addr;
1663 nic_data = efx->nic_data;
1665 !!(nic_data->datapath_caps &
1666 1 << MC_CMD_GET_CAPABILITIES_OUT_RX_BATCHING_LBN);
1668 /* Fill event queue with all ones (i.e. empty events) */
1669 memset(channel->eventq.buf.addr, 0xff, channel->eventq.buf.len);
1671 MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_SIZE, channel->eventq_mask + 1);
1672 MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_INSTANCE, channel->channel);
1673 /* INIT_EVQ expects index in vector table, not absolute */
1674 MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_IRQ_NUM, channel->channel);
1675 MCDI_POPULATE_DWORD_4(inbuf, INIT_EVQ_IN_FLAGS,
1676 INIT_EVQ_IN_FLAG_INTERRUPTING, 1,
1677 INIT_EVQ_IN_FLAG_RX_MERGE, 1,
1678 INIT_EVQ_IN_FLAG_TX_MERGE, 1,
1679 INIT_EVQ_IN_FLAG_CUT_THRU, !supports_rx_merge);
1680 MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_TMR_MODE,
1681 MC_CMD_INIT_EVQ_IN_TMR_MODE_DIS);
1682 MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_TMR_LOAD, 0);
1683 MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_TMR_RELOAD, 0);
1684 MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_COUNT_MODE,
1685 MC_CMD_INIT_EVQ_IN_COUNT_MODE_DIS);
1686 MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_COUNT_THRSHLD, 0);
1688 dma_addr = channel->eventq.buf.dma_addr;
1689 for (i = 0; i < entries; ++i) {
1690 MCDI_SET_ARRAY_QWORD(inbuf, INIT_EVQ_IN_DMA_ADDR, i, dma_addr);
1691 dma_addr += EFX_BUF_SIZE;
1694 inlen = MC_CMD_INIT_EVQ_IN_LEN(entries);
1696 rc = efx_mcdi_rpc(efx, MC_CMD_INIT_EVQ, inbuf, inlen,
1697 outbuf, sizeof(outbuf), &outlen);
1698 /* IRQ return is ignored */
1702 static void efx_ef10_ev_fini(struct efx_channel *channel)
1704 MCDI_DECLARE_BUF(inbuf, MC_CMD_FINI_EVQ_IN_LEN);
1705 MCDI_DECLARE_BUF(outbuf, MC_CMD_FINI_EVQ_OUT_LEN);
1706 struct efx_nic *efx = channel->efx;
1710 MCDI_SET_DWORD(inbuf, FINI_EVQ_IN_INSTANCE, channel->channel);
1712 rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FINI_EVQ, inbuf, sizeof(inbuf),
1713 outbuf, sizeof(outbuf), &outlen);
1715 if (rc && rc != -EALREADY)
1721 efx_mcdi_display_error(efx, MC_CMD_FINI_EVQ, MC_CMD_FINI_EVQ_IN_LEN,
1722 outbuf, outlen, rc);
1725 static void efx_ef10_ev_remove(struct efx_channel *channel)
1727 efx_nic_free_buffer(channel->efx, &channel->eventq.buf);
1730 static void efx_ef10_handle_rx_wrong_queue(struct efx_rx_queue *rx_queue,
1731 unsigned int rx_queue_label)
1733 struct efx_nic *efx = rx_queue->efx;
1735 netif_info(efx, hw, efx->net_dev,
1736 "rx event arrived on queue %d labeled as queue %u\n",
1737 efx_rx_queue_index(rx_queue), rx_queue_label);
1739 efx_schedule_reset(efx, RESET_TYPE_DISABLE);
1743 efx_ef10_handle_rx_bad_lbits(struct efx_rx_queue *rx_queue,
1744 unsigned int actual, unsigned int expected)
1746 unsigned int dropped = (actual - expected) & rx_queue->ptr_mask;
1747 struct efx_nic *efx = rx_queue->efx;
1749 netif_info(efx, hw, efx->net_dev,
1750 "dropped %d events (index=%d expected=%d)\n",
1751 dropped, actual, expected);
1753 efx_schedule_reset(efx, RESET_TYPE_DISABLE);
1756 /* partially received RX was aborted. clean up. */
1757 static void efx_ef10_handle_rx_abort(struct efx_rx_queue *rx_queue)
1759 unsigned int rx_desc_ptr;
1761 netif_dbg(rx_queue->efx, hw, rx_queue->efx->net_dev,
1762 "scattered RX aborted (dropping %u buffers)\n",
1763 rx_queue->scatter_n);
1765 rx_desc_ptr = rx_queue->removed_count & rx_queue->ptr_mask;
1767 efx_rx_packet(rx_queue, rx_desc_ptr, rx_queue->scatter_n,
1768 0, EFX_RX_PKT_DISCARD);
1770 rx_queue->removed_count += rx_queue->scatter_n;
1771 rx_queue->scatter_n = 0;
1772 rx_queue->scatter_len = 0;
1773 ++efx_rx_queue_channel(rx_queue)->n_rx_nodesc_trunc;
1776 static int efx_ef10_handle_rx_event(struct efx_channel *channel,
1777 const efx_qword_t *event)
1779 unsigned int rx_bytes, next_ptr_lbits, rx_queue_label, rx_l4_class;
1780 unsigned int n_descs, n_packets, i;
1781 struct efx_nic *efx = channel->efx;
1782 struct efx_rx_queue *rx_queue;
1786 if (unlikely(ACCESS_ONCE(efx->reset_pending)))
1789 /* Basic packet information */
1790 rx_bytes = EFX_QWORD_FIELD(*event, ESF_DZ_RX_BYTES);
1791 next_ptr_lbits = EFX_QWORD_FIELD(*event, ESF_DZ_RX_DSC_PTR_LBITS);
1792 rx_queue_label = EFX_QWORD_FIELD(*event, ESF_DZ_RX_QLABEL);
1793 rx_l4_class = EFX_QWORD_FIELD(*event, ESF_DZ_RX_L4_CLASS);
1794 rx_cont = EFX_QWORD_FIELD(*event, ESF_DZ_RX_CONT);
1796 if (EFX_QWORD_FIELD(*event, ESF_DZ_RX_DROP_EVENT))
1797 netdev_WARN(efx->net_dev, "saw RX_DROP_EVENT: event="
1799 EFX_QWORD_VAL(*event));
1801 rx_queue = efx_channel_get_rx_queue(channel);
1803 if (unlikely(rx_queue_label != efx_rx_queue_index(rx_queue)))
1804 efx_ef10_handle_rx_wrong_queue(rx_queue, rx_queue_label);
1806 n_descs = ((next_ptr_lbits - rx_queue->removed_count) &
1807 ((1 << ESF_DZ_RX_DSC_PTR_LBITS_WIDTH) - 1));
1809 if (n_descs != rx_queue->scatter_n + 1) {
1810 struct efx_ef10_nic_data *nic_data = efx->nic_data;
1812 /* detect rx abort */
1813 if (unlikely(n_descs == rx_queue->scatter_n)) {
1814 if (rx_queue->scatter_n == 0 || rx_bytes != 0)
1815 netdev_WARN(efx->net_dev,
1816 "invalid RX abort: scatter_n=%u event="
1818 rx_queue->scatter_n,
1819 EFX_QWORD_VAL(*event));
1820 efx_ef10_handle_rx_abort(rx_queue);
1824 /* Check that RX completion merging is valid, i.e.
1825 * the current firmware supports it and this is a
1826 * non-scattered packet.
1828 if (!(nic_data->datapath_caps &
1829 (1 << MC_CMD_GET_CAPABILITIES_OUT_RX_BATCHING_LBN)) ||
1830 rx_queue->scatter_n != 0 || rx_cont) {
1831 efx_ef10_handle_rx_bad_lbits(
1832 rx_queue, next_ptr_lbits,
1833 (rx_queue->removed_count +
1834 rx_queue->scatter_n + 1) &
1835 ((1 << ESF_DZ_RX_DSC_PTR_LBITS_WIDTH) - 1));
1839 /* Merged completion for multiple non-scattered packets */
1840 rx_queue->scatter_n = 1;
1841 rx_queue->scatter_len = 0;
1842 n_packets = n_descs;
1843 ++channel->n_rx_merge_events;
1844 channel->n_rx_merge_packets += n_packets;
1845 flags |= EFX_RX_PKT_PREFIX_LEN;
1847 ++rx_queue->scatter_n;
1848 rx_queue->scatter_len += rx_bytes;
1854 if (unlikely(EFX_QWORD_FIELD(*event, ESF_DZ_RX_ECRC_ERR)))
1855 flags |= EFX_RX_PKT_DISCARD;
1857 if (unlikely(EFX_QWORD_FIELD(*event, ESF_DZ_RX_IPCKSUM_ERR))) {
1858 channel->n_rx_ip_hdr_chksum_err += n_packets;
1859 } else if (unlikely(EFX_QWORD_FIELD(*event,
1860 ESF_DZ_RX_TCPUDP_CKSUM_ERR))) {
1861 channel->n_rx_tcp_udp_chksum_err += n_packets;
1862 } else if (rx_l4_class == ESE_DZ_L4_CLASS_TCP ||
1863 rx_l4_class == ESE_DZ_L4_CLASS_UDP) {
1864 flags |= EFX_RX_PKT_CSUMMED;
1867 if (rx_l4_class == ESE_DZ_L4_CLASS_TCP)
1868 flags |= EFX_RX_PKT_TCP;
1870 channel->irq_mod_score += 2 * n_packets;
1872 /* Handle received packet(s) */
1873 for (i = 0; i < n_packets; i++) {
1874 efx_rx_packet(rx_queue,
1875 rx_queue->removed_count & rx_queue->ptr_mask,
1876 rx_queue->scatter_n, rx_queue->scatter_len,
1878 rx_queue->removed_count += rx_queue->scatter_n;
1881 rx_queue->scatter_n = 0;
1882 rx_queue->scatter_len = 0;
1888 efx_ef10_handle_tx_event(struct efx_channel *channel, efx_qword_t *event)
1890 struct efx_nic *efx = channel->efx;
1891 struct efx_tx_queue *tx_queue;
1892 unsigned int tx_ev_desc_ptr;
1893 unsigned int tx_ev_q_label;
1896 if (unlikely(ACCESS_ONCE(efx->reset_pending)))
1899 if (unlikely(EFX_QWORD_FIELD(*event, ESF_DZ_TX_DROP_EVENT)))
1902 /* Transmit completion */
1903 tx_ev_desc_ptr = EFX_QWORD_FIELD(*event, ESF_DZ_TX_DESCR_INDX);
1904 tx_ev_q_label = EFX_QWORD_FIELD(*event, ESF_DZ_TX_QLABEL);
1905 tx_queue = efx_channel_get_tx_queue(channel,
1906 tx_ev_q_label % EFX_TXQ_TYPES);
1907 tx_descs = ((tx_ev_desc_ptr + 1 - tx_queue->read_count) &
1908 tx_queue->ptr_mask);
1909 efx_xmit_done(tx_queue, tx_ev_desc_ptr & tx_queue->ptr_mask);
1915 efx_ef10_handle_driver_event(struct efx_channel *channel, efx_qword_t *event)
1917 struct efx_nic *efx = channel->efx;
1920 subcode = EFX_QWORD_FIELD(*event, ESF_DZ_DRV_SUB_CODE);
1923 case ESE_DZ_DRV_TIMER_EV:
1924 case ESE_DZ_DRV_WAKE_UP_EV:
1926 case ESE_DZ_DRV_START_UP_EV:
1927 /* event queue init complete. ok. */
1930 netif_err(efx, hw, efx->net_dev,
1931 "channel %d unknown driver event type %d"
1932 " (data " EFX_QWORD_FMT ")\n",
1933 channel->channel, subcode,
1934 EFX_QWORD_VAL(*event));
1939 static void efx_ef10_handle_driver_generated_event(struct efx_channel *channel,
1942 struct efx_nic *efx = channel->efx;
1945 subcode = EFX_QWORD_FIELD(*event, EFX_DWORD_0);
1949 channel->event_test_cpu = raw_smp_processor_id();
1951 case EFX_EF10_REFILL:
1952 /* The queue must be empty, so we won't receive any rx
1953 * events, so efx_process_channel() won't refill the
1954 * queue. Refill it here
1956 efx_fast_push_rx_descriptors(&channel->rx_queue, true);
1959 netif_err(efx, hw, efx->net_dev,
1960 "channel %d unknown driver event type %u"
1961 " (data " EFX_QWORD_FMT ")\n",
1962 channel->channel, (unsigned) subcode,
1963 EFX_QWORD_VAL(*event));
1967 static int efx_ef10_ev_process(struct efx_channel *channel, int quota)
1969 struct efx_nic *efx = channel->efx;
1970 efx_qword_t event, *p_event;
1971 unsigned int read_ptr;
1979 read_ptr = channel->eventq_read_ptr;
1982 p_event = efx_event(channel, read_ptr);
1985 if (!efx_event_present(&event))
1988 EFX_SET_QWORD(*p_event);
1992 ev_code = EFX_QWORD_FIELD(event, ESF_DZ_EV_CODE);
1994 netif_vdbg(efx, drv, efx->net_dev,
1995 "processing event on %d " EFX_QWORD_FMT "\n",
1996 channel->channel, EFX_QWORD_VAL(event));
1999 case ESE_DZ_EV_CODE_MCDI_EV:
2000 efx_mcdi_process_event(channel, &event);
2002 case ESE_DZ_EV_CODE_RX_EV:
2003 spent += efx_ef10_handle_rx_event(channel, &event);
2004 if (spent >= quota) {
2005 /* XXX can we split a merged event to
2006 * avoid going over-quota?
2012 case ESE_DZ_EV_CODE_TX_EV:
2013 tx_descs += efx_ef10_handle_tx_event(channel, &event);
2014 if (tx_descs > efx->txq_entries) {
2017 } else if (++spent == quota) {
2021 case ESE_DZ_EV_CODE_DRIVER_EV:
2022 efx_ef10_handle_driver_event(channel, &event);
2023 if (++spent == quota)
2026 case EFX_EF10_DRVGEN_EV:
2027 efx_ef10_handle_driver_generated_event(channel, &event);
2030 netif_err(efx, hw, efx->net_dev,
2031 "channel %d unknown event type %d"
2032 " (data " EFX_QWORD_FMT ")\n",
2033 channel->channel, ev_code,
2034 EFX_QWORD_VAL(event));
2039 channel->eventq_read_ptr = read_ptr;
2043 static void efx_ef10_ev_read_ack(struct efx_channel *channel)
2045 struct efx_nic *efx = channel->efx;
2048 if (EFX_EF10_WORKAROUND_35388(efx)) {
2049 BUILD_BUG_ON(EFX_MIN_EVQ_SIZE <
2050 (1 << ERF_DD_EVQ_IND_RPTR_WIDTH));
2051 BUILD_BUG_ON(EFX_MAX_EVQ_SIZE >
2052 (1 << 2 * ERF_DD_EVQ_IND_RPTR_WIDTH));
2054 EFX_POPULATE_DWORD_2(rptr, ERF_DD_EVQ_IND_RPTR_FLAGS,
2055 EFE_DD_EVQ_IND_RPTR_FLAGS_HIGH,
2056 ERF_DD_EVQ_IND_RPTR,
2057 (channel->eventq_read_ptr &
2058 channel->eventq_mask) >>
2059 ERF_DD_EVQ_IND_RPTR_WIDTH);
2060 efx_writed_page(efx, &rptr, ER_DD_EVQ_INDIRECT,
2062 EFX_POPULATE_DWORD_2(rptr, ERF_DD_EVQ_IND_RPTR_FLAGS,
2063 EFE_DD_EVQ_IND_RPTR_FLAGS_LOW,
2064 ERF_DD_EVQ_IND_RPTR,
2065 channel->eventq_read_ptr &
2066 ((1 << ERF_DD_EVQ_IND_RPTR_WIDTH) - 1));
2067 efx_writed_page(efx, &rptr, ER_DD_EVQ_INDIRECT,
2070 EFX_POPULATE_DWORD_1(rptr, ERF_DZ_EVQ_RPTR,
2071 channel->eventq_read_ptr &
2072 channel->eventq_mask);
2073 efx_writed_page(efx, &rptr, ER_DZ_EVQ_RPTR, channel->channel);
2077 static void efx_ef10_ev_test_generate(struct efx_channel *channel)
2079 MCDI_DECLARE_BUF(inbuf, MC_CMD_DRIVER_EVENT_IN_LEN);
2080 struct efx_nic *efx = channel->efx;
2084 EFX_POPULATE_QWORD_2(event,
2085 ESF_DZ_EV_CODE, EFX_EF10_DRVGEN_EV,
2086 ESF_DZ_EV_DATA, EFX_EF10_TEST);
2088 MCDI_SET_DWORD(inbuf, DRIVER_EVENT_IN_EVQ, channel->channel);
2090 /* MCDI_SET_QWORD is not appropriate here since EFX_POPULATE_* has
2091 * already swapped the data to little-endian order.
2093 memcpy(MCDI_PTR(inbuf, DRIVER_EVENT_IN_DATA), &event.u64[0],
2094 sizeof(efx_qword_t));
2096 rc = efx_mcdi_rpc(efx, MC_CMD_DRIVER_EVENT, inbuf, sizeof(inbuf),
2105 netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
2108 void efx_ef10_handle_drain_event(struct efx_nic *efx)
2110 if (atomic_dec_and_test(&efx->active_queues))
2111 wake_up(&efx->flush_wq);
2113 WARN_ON(atomic_read(&efx->active_queues) < 0);
2116 static int efx_ef10_fini_dmaq(struct efx_nic *efx)
2118 struct efx_ef10_nic_data *nic_data = efx->nic_data;
2119 struct efx_channel *channel;
2120 struct efx_tx_queue *tx_queue;
2121 struct efx_rx_queue *rx_queue;
2124 /* If the MC has just rebooted, the TX/RX queues will have already been
2125 * torn down, but efx->active_queues needs to be set to zero.
2127 if (nic_data->must_realloc_vis) {
2128 atomic_set(&efx->active_queues, 0);
2132 /* Do not attempt to write to the NIC during EEH recovery */
2133 if (efx->state != STATE_RECOVERY) {
2134 efx_for_each_channel(channel, efx) {
2135 efx_for_each_channel_rx_queue(rx_queue, channel)
2136 efx_ef10_rx_fini(rx_queue);
2137 efx_for_each_channel_tx_queue(tx_queue, channel)
2138 efx_ef10_tx_fini(tx_queue);
2141 wait_event_timeout(efx->flush_wq,
2142 atomic_read(&efx->active_queues) == 0,
2143 msecs_to_jiffies(EFX_MAX_FLUSH_TIME));
2144 pending = atomic_read(&efx->active_queues);
2146 netif_err(efx, hw, efx->net_dev, "failed to flush %d queues\n",
2155 static void efx_ef10_prepare_flr(struct efx_nic *efx)
2157 atomic_set(&efx->active_queues, 0);
2160 static bool efx_ef10_filter_equal(const struct efx_filter_spec *left,
2161 const struct efx_filter_spec *right)
2163 if ((left->match_flags ^ right->match_flags) |
2164 ((left->flags ^ right->flags) &
2165 (EFX_FILTER_FLAG_RX | EFX_FILTER_FLAG_TX)))
2168 return memcmp(&left->outer_vid, &right->outer_vid,
2169 sizeof(struct efx_filter_spec) -
2170 offsetof(struct efx_filter_spec, outer_vid)) == 0;
2173 static unsigned int efx_ef10_filter_hash(const struct efx_filter_spec *spec)
2175 BUILD_BUG_ON(offsetof(struct efx_filter_spec, outer_vid) & 3);
2176 return jhash2((const u32 *)&spec->outer_vid,
2177 (sizeof(struct efx_filter_spec) -
2178 offsetof(struct efx_filter_spec, outer_vid)) / 4,
2180 /* XXX should we randomise the initval? */
2183 /* Decide whether a filter should be exclusive or else should allow
2184 * delivery to additional recipients. Currently we decide that
2185 * filters for specific local unicast MAC and IP addresses are
2188 static bool efx_ef10_filter_is_exclusive(const struct efx_filter_spec *spec)
2190 if (spec->match_flags & EFX_FILTER_MATCH_LOC_MAC &&
2191 !is_multicast_ether_addr(spec->loc_mac))
2194 if ((spec->match_flags &
2195 (EFX_FILTER_MATCH_ETHER_TYPE | EFX_FILTER_MATCH_LOC_HOST)) ==
2196 (EFX_FILTER_MATCH_ETHER_TYPE | EFX_FILTER_MATCH_LOC_HOST)) {
2197 if (spec->ether_type == htons(ETH_P_IP) &&
2198 !ipv4_is_multicast(spec->loc_host[0]))
2200 if (spec->ether_type == htons(ETH_P_IPV6) &&
2201 ((const u8 *)spec->loc_host)[0] != 0xff)
2208 static struct efx_filter_spec *
2209 efx_ef10_filter_entry_spec(const struct efx_ef10_filter_table *table,
2210 unsigned int filter_idx)
2212 return (struct efx_filter_spec *)(table->entry[filter_idx].spec &
2213 ~EFX_EF10_FILTER_FLAGS);
2217 efx_ef10_filter_entry_flags(const struct efx_ef10_filter_table *table,
2218 unsigned int filter_idx)
2220 return table->entry[filter_idx].spec & EFX_EF10_FILTER_FLAGS;
2224 efx_ef10_filter_set_entry(struct efx_ef10_filter_table *table,
2225 unsigned int filter_idx,
2226 const struct efx_filter_spec *spec,
2229 table->entry[filter_idx].spec = (unsigned long)spec | flags;
2232 static void efx_ef10_filter_push_prep(struct efx_nic *efx,
2233 const struct efx_filter_spec *spec,
2234 efx_dword_t *inbuf, u64 handle,
2237 struct efx_ef10_nic_data *nic_data = efx->nic_data;
2239 memset(inbuf, 0, MC_CMD_FILTER_OP_IN_LEN);
2242 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
2243 MC_CMD_FILTER_OP_IN_OP_REPLACE);
2244 MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE, handle);
2246 u32 match_fields = 0;
2248 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
2249 efx_ef10_filter_is_exclusive(spec) ?
2250 MC_CMD_FILTER_OP_IN_OP_INSERT :
2251 MC_CMD_FILTER_OP_IN_OP_SUBSCRIBE);
2253 /* Convert match flags and values. Unlike almost
2254 * everything else in MCDI, these fields are in
2255 * network byte order.
2257 if (spec->match_flags & EFX_FILTER_MATCH_LOC_MAC_IG)
2259 is_multicast_ether_addr(spec->loc_mac) ?
2260 1 << MC_CMD_FILTER_OP_IN_MATCH_UNKNOWN_MCAST_DST_LBN :
2261 1 << MC_CMD_FILTER_OP_IN_MATCH_UNKNOWN_UCAST_DST_LBN;
2262 #define COPY_FIELD(gen_flag, gen_field, mcdi_field) \
2263 if (spec->match_flags & EFX_FILTER_MATCH_ ## gen_flag) { \
2265 1 << MC_CMD_FILTER_OP_IN_MATCH_ ## \
2266 mcdi_field ## _LBN; \
2268 MC_CMD_FILTER_OP_IN_ ## mcdi_field ## _LEN < \
2269 sizeof(spec->gen_field)); \
2270 memcpy(MCDI_PTR(inbuf, FILTER_OP_IN_ ## mcdi_field), \
2271 &spec->gen_field, sizeof(spec->gen_field)); \
2273 COPY_FIELD(REM_HOST, rem_host, SRC_IP);
2274 COPY_FIELD(LOC_HOST, loc_host, DST_IP);
2275 COPY_FIELD(REM_MAC, rem_mac, SRC_MAC);
2276 COPY_FIELD(REM_PORT, rem_port, SRC_PORT);
2277 COPY_FIELD(LOC_MAC, loc_mac, DST_MAC);
2278 COPY_FIELD(LOC_PORT, loc_port, DST_PORT);
2279 COPY_FIELD(ETHER_TYPE, ether_type, ETHER_TYPE);
2280 COPY_FIELD(INNER_VID, inner_vid, INNER_VLAN);
2281 COPY_FIELD(OUTER_VID, outer_vid, OUTER_VLAN);
2282 COPY_FIELD(IP_PROTO, ip_proto, IP_PROTO);
2284 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_MATCH_FIELDS,
2288 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_PORT_ID, EVB_PORT_ID_ASSIGNED);
2289 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_DEST,
2290 spec->dmaq_id == EFX_FILTER_RX_DMAQ_ID_DROP ?
2291 MC_CMD_FILTER_OP_IN_RX_DEST_DROP :
2292 MC_CMD_FILTER_OP_IN_RX_DEST_HOST);
2293 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_TX_DEST,
2294 MC_CMD_FILTER_OP_IN_TX_DEST_DEFAULT);
2295 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_QUEUE,
2296 spec->dmaq_id == EFX_FILTER_RX_DMAQ_ID_DROP ?
2298 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_MODE,
2299 (spec->flags & EFX_FILTER_FLAG_RX_RSS) ?
2300 MC_CMD_FILTER_OP_IN_RX_MODE_RSS :
2301 MC_CMD_FILTER_OP_IN_RX_MODE_SIMPLE);
2302 if (spec->flags & EFX_FILTER_FLAG_RX_RSS)
2303 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_CONTEXT,
2304 spec->rss_context !=
2305 EFX_FILTER_RSS_CONTEXT_DEFAULT ?
2306 spec->rss_context : nic_data->rx_rss_context);
2309 static int efx_ef10_filter_push(struct efx_nic *efx,
2310 const struct efx_filter_spec *spec,
2311 u64 *handle, bool replacing)
2313 MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_IN_LEN);
2314 MCDI_DECLARE_BUF(outbuf, MC_CMD_FILTER_OP_OUT_LEN);
2317 efx_ef10_filter_push_prep(efx, spec, inbuf, *handle, replacing);
2318 rc = efx_mcdi_rpc(efx, MC_CMD_FILTER_OP, inbuf, sizeof(inbuf),
2319 outbuf, sizeof(outbuf), NULL);
2321 *handle = MCDI_QWORD(outbuf, FILTER_OP_OUT_HANDLE);
2323 rc = -EBUSY; /* to match efx_farch_filter_insert() */
2327 static int efx_ef10_filter_rx_match_pri(struct efx_ef10_filter_table *table,
2328 enum efx_filter_match_flags match_flags)
2330 unsigned int match_pri;
2333 match_pri < table->rx_match_count;
2335 if (table->rx_match_flags[match_pri] == match_flags)
2338 return -EPROTONOSUPPORT;
2341 static s32 efx_ef10_filter_insert(struct efx_nic *efx,
2342 struct efx_filter_spec *spec,
2345 struct efx_ef10_filter_table *table = efx->filter_state;
2346 DECLARE_BITMAP(mc_rem_map, EFX_EF10_FILTER_SEARCH_LIMIT);
2347 struct efx_filter_spec *saved_spec;
2348 unsigned int match_pri, hash;
2349 unsigned int priv_flags;
2350 bool replacing = false;
2356 /* For now, only support RX filters */
2357 if ((spec->flags & (EFX_FILTER_FLAG_RX | EFX_FILTER_FLAG_TX)) !=
2361 rc = efx_ef10_filter_rx_match_pri(table, spec->match_flags);
2366 hash = efx_ef10_filter_hash(spec);
2367 is_mc_recip = efx_filter_is_mc_recipient(spec);
2369 bitmap_zero(mc_rem_map, EFX_EF10_FILTER_SEARCH_LIMIT);
2371 /* Find any existing filters with the same match tuple or
2372 * else a free slot to insert at. If any of them are busy,
2373 * we have to wait and retry.
2376 unsigned int depth = 1;
2379 spin_lock_bh(&efx->filter_lock);
2382 i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
2383 saved_spec = efx_ef10_filter_entry_spec(table, i);
2388 } else if (efx_ef10_filter_equal(spec, saved_spec)) {
2389 if (table->entry[i].spec &
2390 EFX_EF10_FILTER_FLAG_BUSY)
2392 if (spec->priority < saved_spec->priority &&
2393 spec->priority != EFX_FILTER_PRI_AUTO) {
2398 /* This is the only one */
2399 if (spec->priority ==
2400 saved_spec->priority &&
2407 } else if (spec->priority >
2408 saved_spec->priority ||
2410 saved_spec->priority &&
2415 __set_bit(depth, mc_rem_map);
2419 /* Once we reach the maximum search depth, use
2420 * the first suitable slot or return -EBUSY if
2423 if (depth == EFX_EF10_FILTER_SEARCH_LIMIT) {
2424 if (ins_index < 0) {
2434 prepare_to_wait(&table->waitq, &wait, TASK_UNINTERRUPTIBLE);
2435 spin_unlock_bh(&efx->filter_lock);
2440 /* Create a software table entry if necessary, and mark it
2441 * busy. We might yet fail to insert, but any attempt to
2442 * insert a conflicting filter while we're waiting for the
2443 * firmware must find the busy entry.
2445 saved_spec = efx_ef10_filter_entry_spec(table, ins_index);
2447 if (spec->priority == EFX_FILTER_PRI_AUTO &&
2448 saved_spec->priority >= EFX_FILTER_PRI_AUTO) {
2449 /* Just make sure it won't be removed */
2450 if (saved_spec->priority > EFX_FILTER_PRI_AUTO)
2451 saved_spec->flags |= EFX_FILTER_FLAG_RX_OVER_AUTO;
2452 table->entry[ins_index].spec &=
2453 ~EFX_EF10_FILTER_FLAG_AUTO_OLD;
2458 priv_flags = efx_ef10_filter_entry_flags(table, ins_index);
2460 saved_spec = kmalloc(sizeof(*spec), GFP_ATOMIC);
2465 *saved_spec = *spec;
2468 efx_ef10_filter_set_entry(table, ins_index, saved_spec,
2469 priv_flags | EFX_EF10_FILTER_FLAG_BUSY);
2471 /* Mark lower-priority multicast recipients busy prior to removal */
2473 unsigned int depth, i;
2475 for (depth = 0; depth < EFX_EF10_FILTER_SEARCH_LIMIT; depth++) {
2476 i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
2477 if (test_bit(depth, mc_rem_map))
2478 table->entry[i].spec |=
2479 EFX_EF10_FILTER_FLAG_BUSY;
2483 spin_unlock_bh(&efx->filter_lock);
2485 rc = efx_ef10_filter_push(efx, spec, &table->entry[ins_index].handle,
2488 /* Finalise the software table entry */
2489 spin_lock_bh(&efx->filter_lock);
2492 /* Update the fields that may differ */
2493 if (saved_spec->priority == EFX_FILTER_PRI_AUTO)
2494 saved_spec->flags |=
2495 EFX_FILTER_FLAG_RX_OVER_AUTO;
2496 saved_spec->priority = spec->priority;
2497 saved_spec->flags &= EFX_FILTER_FLAG_RX_OVER_AUTO;
2498 saved_spec->flags |= spec->flags;
2499 saved_spec->rss_context = spec->rss_context;
2500 saved_spec->dmaq_id = spec->dmaq_id;
2502 } else if (!replacing) {
2506 efx_ef10_filter_set_entry(table, ins_index, saved_spec, priv_flags);
2508 /* Remove and finalise entries for lower-priority multicast
2512 MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_IN_LEN);
2513 unsigned int depth, i;
2515 memset(inbuf, 0, sizeof(inbuf));
2517 for (depth = 0; depth < EFX_EF10_FILTER_SEARCH_LIMIT; depth++) {
2518 if (!test_bit(depth, mc_rem_map))
2521 i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
2522 saved_spec = efx_ef10_filter_entry_spec(table, i);
2523 priv_flags = efx_ef10_filter_entry_flags(table, i);
2526 spin_unlock_bh(&efx->filter_lock);
2527 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
2528 MC_CMD_FILTER_OP_IN_OP_UNSUBSCRIBE);
2529 MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
2530 table->entry[i].handle);
2531 rc = efx_mcdi_rpc(efx, MC_CMD_FILTER_OP,
2532 inbuf, sizeof(inbuf),
2534 spin_lock_bh(&efx->filter_lock);
2542 priv_flags &= ~EFX_EF10_FILTER_FLAG_BUSY;
2544 efx_ef10_filter_set_entry(table, i, saved_spec,
2549 /* If successful, return the inserted filter ID */
2551 rc = match_pri * HUNT_FILTER_TBL_ROWS + ins_index;
2553 wake_up_all(&table->waitq);
2555 spin_unlock_bh(&efx->filter_lock);
2556 finish_wait(&table->waitq, &wait);
2560 static void efx_ef10_filter_update_rx_scatter(struct efx_nic *efx)
2562 /* no need to do anything here on EF10 */
2566 * If !by_index, remove by ID
2567 * If by_index, remove by index
2568 * Filter ID may come from userland and must be range-checked.
2570 static int efx_ef10_filter_remove_internal(struct efx_nic *efx,
2571 unsigned int priority_mask,
2572 u32 filter_id, bool by_index)
2574 unsigned int filter_idx = filter_id % HUNT_FILTER_TBL_ROWS;
2575 struct efx_ef10_filter_table *table = efx->filter_state;
2576 MCDI_DECLARE_BUF(inbuf,
2577 MC_CMD_FILTER_OP_IN_HANDLE_OFST +
2578 MC_CMD_FILTER_OP_IN_HANDLE_LEN);
2579 struct efx_filter_spec *spec;
2583 /* Find the software table entry and mark it busy. Don't
2584 * remove it yet; any attempt to update while we're waiting
2585 * for the firmware must find the busy entry.
2588 spin_lock_bh(&efx->filter_lock);
2589 if (!(table->entry[filter_idx].spec &
2590 EFX_EF10_FILTER_FLAG_BUSY))
2592 prepare_to_wait(&table->waitq, &wait, TASK_UNINTERRUPTIBLE);
2593 spin_unlock_bh(&efx->filter_lock);
2597 spec = efx_ef10_filter_entry_spec(table, filter_idx);
2600 efx_ef10_filter_rx_match_pri(table, spec->match_flags) !=
2601 filter_id / HUNT_FILTER_TBL_ROWS)) {
2606 if (spec->flags & EFX_FILTER_FLAG_RX_OVER_AUTO &&
2607 priority_mask == (1U << EFX_FILTER_PRI_AUTO)) {
2608 /* Just remove flags */
2609 spec->flags &= ~EFX_FILTER_FLAG_RX_OVER_AUTO;
2610 table->entry[filter_idx].spec &= ~EFX_EF10_FILTER_FLAG_AUTO_OLD;
2615 if (!(priority_mask & (1U << spec->priority))) {
2620 table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_BUSY;
2621 spin_unlock_bh(&efx->filter_lock);
2623 if (spec->flags & EFX_FILTER_FLAG_RX_OVER_AUTO) {
2624 /* Reset to an automatic filter */
2626 struct efx_filter_spec new_spec = *spec;
2628 new_spec.priority = EFX_FILTER_PRI_AUTO;
2629 new_spec.flags = (EFX_FILTER_FLAG_RX |
2630 EFX_FILTER_FLAG_RX_RSS);
2631 new_spec.dmaq_id = 0;
2632 new_spec.rss_context = EFX_FILTER_RSS_CONTEXT_DEFAULT;
2633 rc = efx_ef10_filter_push(efx, &new_spec,
2634 &table->entry[filter_idx].handle,
2637 spin_lock_bh(&efx->filter_lock);
2641 /* Really remove the filter */
2643 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
2644 efx_ef10_filter_is_exclusive(spec) ?
2645 MC_CMD_FILTER_OP_IN_OP_REMOVE :
2646 MC_CMD_FILTER_OP_IN_OP_UNSUBSCRIBE);
2647 MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
2648 table->entry[filter_idx].handle);
2649 rc = efx_mcdi_rpc(efx, MC_CMD_FILTER_OP,
2650 inbuf, sizeof(inbuf), NULL, 0, NULL);
2652 spin_lock_bh(&efx->filter_lock);
2655 efx_ef10_filter_set_entry(table, filter_idx, NULL, 0);
2659 table->entry[filter_idx].spec &= ~EFX_EF10_FILTER_FLAG_BUSY;
2660 wake_up_all(&table->waitq);
2662 spin_unlock_bh(&efx->filter_lock);
2663 finish_wait(&table->waitq, &wait);
2667 static int efx_ef10_filter_remove_safe(struct efx_nic *efx,
2668 enum efx_filter_priority priority,
2671 return efx_ef10_filter_remove_internal(efx, 1U << priority,
2675 static int efx_ef10_filter_get_safe(struct efx_nic *efx,
2676 enum efx_filter_priority priority,
2677 u32 filter_id, struct efx_filter_spec *spec)
2679 unsigned int filter_idx = filter_id % HUNT_FILTER_TBL_ROWS;
2680 struct efx_ef10_filter_table *table = efx->filter_state;
2681 const struct efx_filter_spec *saved_spec;
2684 spin_lock_bh(&efx->filter_lock);
2685 saved_spec = efx_ef10_filter_entry_spec(table, filter_idx);
2686 if (saved_spec && saved_spec->priority == priority &&
2687 efx_ef10_filter_rx_match_pri(table, saved_spec->match_flags) ==
2688 filter_id / HUNT_FILTER_TBL_ROWS) {
2689 *spec = *saved_spec;
2694 spin_unlock_bh(&efx->filter_lock);
2698 static int efx_ef10_filter_clear_rx(struct efx_nic *efx,
2699 enum efx_filter_priority priority)
2701 unsigned int priority_mask;
2705 priority_mask = (((1U << (priority + 1)) - 1) &
2706 ~(1U << EFX_FILTER_PRI_AUTO));
2708 for (i = 0; i < HUNT_FILTER_TBL_ROWS; i++) {
2709 rc = efx_ef10_filter_remove_internal(efx, priority_mask,
2711 if (rc && rc != -ENOENT)
2718 static u32 efx_ef10_filter_count_rx_used(struct efx_nic *efx,
2719 enum efx_filter_priority priority)
2721 struct efx_ef10_filter_table *table = efx->filter_state;
2722 unsigned int filter_idx;
2725 spin_lock_bh(&efx->filter_lock);
2726 for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
2727 if (table->entry[filter_idx].spec &&
2728 efx_ef10_filter_entry_spec(table, filter_idx)->priority ==
2732 spin_unlock_bh(&efx->filter_lock);
2736 static u32 efx_ef10_filter_get_rx_id_limit(struct efx_nic *efx)
2738 struct efx_ef10_filter_table *table = efx->filter_state;
2740 return table->rx_match_count * HUNT_FILTER_TBL_ROWS;
2743 static s32 efx_ef10_filter_get_rx_ids(struct efx_nic *efx,
2744 enum efx_filter_priority priority,
2747 struct efx_ef10_filter_table *table = efx->filter_state;
2748 struct efx_filter_spec *spec;
2749 unsigned int filter_idx;
2752 spin_lock_bh(&efx->filter_lock);
2753 for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
2754 spec = efx_ef10_filter_entry_spec(table, filter_idx);
2755 if (spec && spec->priority == priority) {
2756 if (count == size) {
2760 buf[count++] = (efx_ef10_filter_rx_match_pri(
2761 table, spec->match_flags) *
2762 HUNT_FILTER_TBL_ROWS +
2766 spin_unlock_bh(&efx->filter_lock);
2770 #ifdef CONFIG_RFS_ACCEL
2772 static efx_mcdi_async_completer efx_ef10_filter_rfs_insert_complete;
2774 static s32 efx_ef10_filter_rfs_insert(struct efx_nic *efx,
2775 struct efx_filter_spec *spec)
2777 struct efx_ef10_filter_table *table = efx->filter_state;
2778 MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_IN_LEN);
2779 struct efx_filter_spec *saved_spec;
2780 unsigned int hash, i, depth = 1;
2781 bool replacing = false;
2786 /* Must be an RX filter without RSS and not for a multicast
2787 * destination address (RFS only works for connected sockets).
2788 * These restrictions allow us to pass only a tiny amount of
2789 * data through to the completion function.
2791 EFX_WARN_ON_PARANOID(spec->flags !=
2792 (EFX_FILTER_FLAG_RX | EFX_FILTER_FLAG_RX_SCATTER));
2793 EFX_WARN_ON_PARANOID(spec->priority != EFX_FILTER_PRI_HINT);
2794 EFX_WARN_ON_PARANOID(efx_filter_is_mc_recipient(spec));
2796 hash = efx_ef10_filter_hash(spec);
2798 spin_lock_bh(&efx->filter_lock);
2800 /* Find any existing filter with the same match tuple or else
2801 * a free slot to insert at. If an existing filter is busy,
2802 * we have to give up.
2805 i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
2806 saved_spec = efx_ef10_filter_entry_spec(table, i);
2811 } else if (efx_ef10_filter_equal(spec, saved_spec)) {
2812 if (table->entry[i].spec & EFX_EF10_FILTER_FLAG_BUSY) {
2816 if (spec->priority < saved_spec->priority) {
2824 /* Once we reach the maximum search depth, use the
2825 * first suitable slot or return -EBUSY if there was
2828 if (depth == EFX_EF10_FILTER_SEARCH_LIMIT) {
2829 if (ins_index < 0) {
2839 /* Create a software table entry if necessary, and mark it
2840 * busy. We might yet fail to insert, but any attempt to
2841 * insert a conflicting filter while we're waiting for the
2842 * firmware must find the busy entry.
2844 saved_spec = efx_ef10_filter_entry_spec(table, ins_index);
2848 saved_spec = kmalloc(sizeof(*spec), GFP_ATOMIC);
2853 *saved_spec = *spec;
2855 efx_ef10_filter_set_entry(table, ins_index, saved_spec,
2856 EFX_EF10_FILTER_FLAG_BUSY);
2858 spin_unlock_bh(&efx->filter_lock);
2860 /* Pack up the variables needed on completion */
2861 cookie = replacing << 31 | ins_index << 16 | spec->dmaq_id;
2863 efx_ef10_filter_push_prep(efx, spec, inbuf,
2864 table->entry[ins_index].handle, replacing);
2865 efx_mcdi_rpc_async(efx, MC_CMD_FILTER_OP, inbuf, sizeof(inbuf),
2866 MC_CMD_FILTER_OP_OUT_LEN,
2867 efx_ef10_filter_rfs_insert_complete, cookie);
2872 spin_unlock_bh(&efx->filter_lock);
2877 efx_ef10_filter_rfs_insert_complete(struct efx_nic *efx, unsigned long cookie,
2878 int rc, efx_dword_t *outbuf,
2879 size_t outlen_actual)
2881 struct efx_ef10_filter_table *table = efx->filter_state;
2882 unsigned int ins_index, dmaq_id;
2883 struct efx_filter_spec *spec;
2886 /* Unpack the cookie */
2887 replacing = cookie >> 31;
2888 ins_index = (cookie >> 16) & (HUNT_FILTER_TBL_ROWS - 1);
2889 dmaq_id = cookie & 0xffff;
2891 spin_lock_bh(&efx->filter_lock);
2892 spec = efx_ef10_filter_entry_spec(table, ins_index);
2894 table->entry[ins_index].handle =
2895 MCDI_QWORD(outbuf, FILTER_OP_OUT_HANDLE);
2897 spec->dmaq_id = dmaq_id;
2898 } else if (!replacing) {
2902 efx_ef10_filter_set_entry(table, ins_index, spec, 0);
2903 spin_unlock_bh(&efx->filter_lock);
2905 wake_up_all(&table->waitq);
2909 efx_ef10_filter_rfs_expire_complete(struct efx_nic *efx,
2910 unsigned long filter_idx,
2911 int rc, efx_dword_t *outbuf,
2912 size_t outlen_actual);
2914 static bool efx_ef10_filter_rfs_expire_one(struct efx_nic *efx, u32 flow_id,
2915 unsigned int filter_idx)
2917 struct efx_ef10_filter_table *table = efx->filter_state;
2918 struct efx_filter_spec *spec =
2919 efx_ef10_filter_entry_spec(table, filter_idx);
2920 MCDI_DECLARE_BUF(inbuf,
2921 MC_CMD_FILTER_OP_IN_HANDLE_OFST +
2922 MC_CMD_FILTER_OP_IN_HANDLE_LEN);
2925 (table->entry[filter_idx].spec & EFX_EF10_FILTER_FLAG_BUSY) ||
2926 spec->priority != EFX_FILTER_PRI_HINT ||
2927 !rps_may_expire_flow(efx->net_dev, spec->dmaq_id,
2928 flow_id, filter_idx))
2931 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
2932 MC_CMD_FILTER_OP_IN_OP_REMOVE);
2933 MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
2934 table->entry[filter_idx].handle);
2935 if (efx_mcdi_rpc_async(efx, MC_CMD_FILTER_OP, inbuf, sizeof(inbuf), 0,
2936 efx_ef10_filter_rfs_expire_complete, filter_idx))
2939 table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_BUSY;
2944 efx_ef10_filter_rfs_expire_complete(struct efx_nic *efx,
2945 unsigned long filter_idx,
2946 int rc, efx_dword_t *outbuf,
2947 size_t outlen_actual)
2949 struct efx_ef10_filter_table *table = efx->filter_state;
2950 struct efx_filter_spec *spec =
2951 efx_ef10_filter_entry_spec(table, filter_idx);
2953 spin_lock_bh(&efx->filter_lock);
2956 efx_ef10_filter_set_entry(table, filter_idx, NULL, 0);
2958 table->entry[filter_idx].spec &= ~EFX_EF10_FILTER_FLAG_BUSY;
2959 wake_up_all(&table->waitq);
2960 spin_unlock_bh(&efx->filter_lock);
2963 #endif /* CONFIG_RFS_ACCEL */
2965 static int efx_ef10_filter_match_flags_from_mcdi(u32 mcdi_flags)
2967 int match_flags = 0;
2969 #define MAP_FLAG(gen_flag, mcdi_field) { \
2970 u32 old_mcdi_flags = mcdi_flags; \
2971 mcdi_flags &= ~(1 << MC_CMD_FILTER_OP_IN_MATCH_ ## \
2972 mcdi_field ## _LBN); \
2973 if (mcdi_flags != old_mcdi_flags) \
2974 match_flags |= EFX_FILTER_MATCH_ ## gen_flag; \
2976 MAP_FLAG(LOC_MAC_IG, UNKNOWN_UCAST_DST);
2977 MAP_FLAG(LOC_MAC_IG, UNKNOWN_MCAST_DST);
2978 MAP_FLAG(REM_HOST, SRC_IP);
2979 MAP_FLAG(LOC_HOST, DST_IP);
2980 MAP_FLAG(REM_MAC, SRC_MAC);
2981 MAP_FLAG(REM_PORT, SRC_PORT);
2982 MAP_FLAG(LOC_MAC, DST_MAC);
2983 MAP_FLAG(LOC_PORT, DST_PORT);
2984 MAP_FLAG(ETHER_TYPE, ETHER_TYPE);
2985 MAP_FLAG(INNER_VID, INNER_VLAN);
2986 MAP_FLAG(OUTER_VID, OUTER_VLAN);
2987 MAP_FLAG(IP_PROTO, IP_PROTO);
2990 /* Did we map them all? */
2997 static int efx_ef10_filter_table_probe(struct efx_nic *efx)
2999 MCDI_DECLARE_BUF(inbuf, MC_CMD_GET_PARSER_DISP_INFO_IN_LEN);
3000 MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_PARSER_DISP_INFO_OUT_LENMAX);
3001 unsigned int pd_match_pri, pd_match_count;
3002 struct efx_ef10_filter_table *table;
3006 table = kzalloc(sizeof(*table), GFP_KERNEL);
3010 /* Find out which RX filter types are supported, and their priorities */
3011 MCDI_SET_DWORD(inbuf, GET_PARSER_DISP_INFO_IN_OP,
3012 MC_CMD_GET_PARSER_DISP_INFO_IN_OP_GET_SUPPORTED_RX_MATCHES);
3013 rc = efx_mcdi_rpc(efx, MC_CMD_GET_PARSER_DISP_INFO,
3014 inbuf, sizeof(inbuf), outbuf, sizeof(outbuf),
3018 pd_match_count = MCDI_VAR_ARRAY_LEN(
3019 outlen, GET_PARSER_DISP_INFO_OUT_SUPPORTED_MATCHES);
3020 table->rx_match_count = 0;
3022 for (pd_match_pri = 0; pd_match_pri < pd_match_count; pd_match_pri++) {
3026 GET_PARSER_DISP_INFO_OUT_SUPPORTED_MATCHES,
3028 rc = efx_ef10_filter_match_flags_from_mcdi(mcdi_flags);
3030 netif_dbg(efx, probe, efx->net_dev,
3031 "%s: fw flags %#x pri %u not supported in driver\n",
3032 __func__, mcdi_flags, pd_match_pri);
3034 netif_dbg(efx, probe, efx->net_dev,
3035 "%s: fw flags %#x pri %u supported as driver flags %#x pri %u\n",
3036 __func__, mcdi_flags, pd_match_pri,
3037 rc, table->rx_match_count);
3038 table->rx_match_flags[table->rx_match_count++] = rc;
3042 table->entry = vzalloc(HUNT_FILTER_TBL_ROWS * sizeof(*table->entry));
3043 if (!table->entry) {
3048 efx->filter_state = table;
3049 init_waitqueue_head(&table->waitq);
3057 static void efx_ef10_filter_table_restore(struct efx_nic *efx)
3059 struct efx_ef10_filter_table *table = efx->filter_state;
3060 struct efx_ef10_nic_data *nic_data = efx->nic_data;
3061 struct efx_filter_spec *spec;
3062 unsigned int filter_idx;
3063 bool failed = false;
3066 if (!nic_data->must_restore_filters)
3069 spin_lock_bh(&efx->filter_lock);
3071 for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
3072 spec = efx_ef10_filter_entry_spec(table, filter_idx);
3076 table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_BUSY;
3077 spin_unlock_bh(&efx->filter_lock);
3079 rc = efx_ef10_filter_push(efx, spec,
3080 &table->entry[filter_idx].handle,
3085 spin_lock_bh(&efx->filter_lock);
3088 efx_ef10_filter_set_entry(table, filter_idx, NULL, 0);
3090 table->entry[filter_idx].spec &=
3091 ~EFX_EF10_FILTER_FLAG_BUSY;
3095 spin_unlock_bh(&efx->filter_lock);
3098 netif_err(efx, hw, efx->net_dev,
3099 "unable to restore all filters\n");
3101 nic_data->must_restore_filters = false;
3104 static void efx_ef10_filter_table_remove(struct efx_nic *efx)
3106 struct efx_ef10_filter_table *table = efx->filter_state;
3107 MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_IN_LEN);
3108 struct efx_filter_spec *spec;
3109 unsigned int filter_idx;
3112 for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
3113 spec = efx_ef10_filter_entry_spec(table, filter_idx);
3117 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
3118 efx_ef10_filter_is_exclusive(spec) ?
3119 MC_CMD_FILTER_OP_IN_OP_REMOVE :
3120 MC_CMD_FILTER_OP_IN_OP_UNSUBSCRIBE);
3121 MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
3122 table->entry[filter_idx].handle);
3123 rc = efx_mcdi_rpc(efx, MC_CMD_FILTER_OP, inbuf, sizeof(inbuf),
3126 netdev_WARN(efx->net_dev,
3127 "filter_idx=%#x handle=%#llx\n",
3129 table->entry[filter_idx].handle);
3133 vfree(table->entry);
3137 static void efx_ef10_filter_sync_rx_mode(struct efx_nic *efx)
3139 struct efx_ef10_filter_table *table = efx->filter_state;
3140 struct net_device *net_dev = efx->net_dev;
3141 struct efx_filter_spec spec;
3142 bool remove_failed = false;
3143 struct netdev_hw_addr *uc;
3144 struct netdev_hw_addr *mc;
3145 unsigned int filter_idx;
3148 if (!efx_dev_registered(efx))
3151 /* Mark old filters that may need to be removed */
3152 spin_lock_bh(&efx->filter_lock);
3153 n = table->dev_uc_count < 0 ? 1 : table->dev_uc_count;
3154 for (i = 0; i < n; i++) {
3155 filter_idx = table->dev_uc_list[i].id % HUNT_FILTER_TBL_ROWS;
3156 table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_AUTO_OLD;
3158 n = table->dev_mc_count < 0 ? 1 : table->dev_mc_count;
3159 for (i = 0; i < n; i++) {
3160 filter_idx = table->dev_mc_list[i].id % HUNT_FILTER_TBL_ROWS;
3161 table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_AUTO_OLD;
3163 spin_unlock_bh(&efx->filter_lock);
3165 /* Copy/convert the address lists; add the primary station
3166 * address and broadcast address
3168 netif_addr_lock_bh(net_dev);
3169 if (net_dev->flags & IFF_PROMISC ||
3170 netdev_uc_count(net_dev) >= EFX_EF10_FILTER_DEV_UC_MAX) {
3171 table->dev_uc_count = -1;
3173 table->dev_uc_count = 1 + netdev_uc_count(net_dev);
3174 ether_addr_copy(table->dev_uc_list[0].addr, net_dev->dev_addr);
3176 netdev_for_each_uc_addr(uc, net_dev) {
3177 ether_addr_copy(table->dev_uc_list[i].addr, uc->addr);
3181 if (net_dev->flags & (IFF_PROMISC | IFF_ALLMULTI) ||
3182 netdev_mc_count(net_dev) >= EFX_EF10_FILTER_DEV_MC_MAX) {
3183 table->dev_mc_count = -1;
3185 table->dev_mc_count = 1 + netdev_mc_count(net_dev);
3186 eth_broadcast_addr(table->dev_mc_list[0].addr);
3188 netdev_for_each_mc_addr(mc, net_dev) {
3189 ether_addr_copy(table->dev_mc_list[i].addr, mc->addr);
3193 netif_addr_unlock_bh(net_dev);
3195 /* Insert/renew unicast filters */
3196 if (table->dev_uc_count >= 0) {
3197 for (i = 0; i < table->dev_uc_count; i++) {
3198 efx_filter_init_rx(&spec, EFX_FILTER_PRI_AUTO,
3199 EFX_FILTER_FLAG_RX_RSS,
3201 efx_filter_set_eth_local(&spec, EFX_FILTER_VID_UNSPEC,
3202 table->dev_uc_list[i].addr);
3203 rc = efx_ef10_filter_insert(efx, &spec, true);
3205 /* Fall back to unicast-promisc */
3207 efx_ef10_filter_remove_safe(
3208 efx, EFX_FILTER_PRI_AUTO,
3209 table->dev_uc_list[i].id);
3210 table->dev_uc_count = -1;
3213 table->dev_uc_list[i].id = rc;
3216 if (table->dev_uc_count < 0) {
3217 efx_filter_init_rx(&spec, EFX_FILTER_PRI_AUTO,
3218 EFX_FILTER_FLAG_RX_RSS,
3220 efx_filter_set_uc_def(&spec);
3221 rc = efx_ef10_filter_insert(efx, &spec, true);
3224 table->dev_uc_count = 0;
3226 table->dev_uc_list[0].id = rc;
3230 /* Insert/renew multicast filters */
3231 if (table->dev_mc_count >= 0) {
3232 for (i = 0; i < table->dev_mc_count; i++) {
3233 efx_filter_init_rx(&spec, EFX_FILTER_PRI_AUTO,
3234 EFX_FILTER_FLAG_RX_RSS,
3236 efx_filter_set_eth_local(&spec, EFX_FILTER_VID_UNSPEC,
3237 table->dev_mc_list[i].addr);
3238 rc = efx_ef10_filter_insert(efx, &spec, true);
3240 /* Fall back to multicast-promisc */
3242 efx_ef10_filter_remove_safe(
3243 efx, EFX_FILTER_PRI_AUTO,
3244 table->dev_mc_list[i].id);
3245 table->dev_mc_count = -1;
3248 table->dev_mc_list[i].id = rc;
3251 if (table->dev_mc_count < 0) {
3252 efx_filter_init_rx(&spec, EFX_FILTER_PRI_AUTO,
3253 EFX_FILTER_FLAG_RX_RSS,
3255 efx_filter_set_mc_def(&spec);
3256 rc = efx_ef10_filter_insert(efx, &spec, true);
3259 table->dev_mc_count = 0;
3261 table->dev_mc_list[0].id = rc;
3265 /* Remove filters that weren't renewed. Since nothing else
3266 * changes the AUTO_OLD flag or removes these filters, we
3267 * don't need to hold the filter_lock while scanning for
3270 for (i = 0; i < HUNT_FILTER_TBL_ROWS; i++) {
3271 if (ACCESS_ONCE(table->entry[i].spec) &
3272 EFX_EF10_FILTER_FLAG_AUTO_OLD) {
3273 if (efx_ef10_filter_remove_internal(
3274 efx, 1U << EFX_FILTER_PRI_AUTO,
3276 remove_failed = true;
3279 WARN_ON(remove_failed);
3282 static int efx_ef10_mac_reconfigure(struct efx_nic *efx)
3284 efx_ef10_filter_sync_rx_mode(efx);
3286 return efx_mcdi_set_mac(efx);
3289 static int efx_ef10_start_bist(struct efx_nic *efx, u32 bist_type)
3291 MCDI_DECLARE_BUF(inbuf, MC_CMD_START_BIST_IN_LEN);
3293 MCDI_SET_DWORD(inbuf, START_BIST_IN_TYPE, bist_type);
3294 return efx_mcdi_rpc(efx, MC_CMD_START_BIST, inbuf, sizeof(inbuf),
3298 /* MC BISTs follow a different poll mechanism to phy BISTs.
3299 * The BIST is done in the poll handler on the MC, and the MCDI command
3300 * will block until the BIST is done.
3302 static int efx_ef10_poll_bist(struct efx_nic *efx)
3305 MCDI_DECLARE_BUF(outbuf, MC_CMD_POLL_BIST_OUT_LEN);
3309 rc = efx_mcdi_rpc(efx, MC_CMD_POLL_BIST, NULL, 0,
3310 outbuf, sizeof(outbuf), &outlen);
3314 if (outlen < MC_CMD_POLL_BIST_OUT_LEN)
3317 result = MCDI_DWORD(outbuf, POLL_BIST_OUT_RESULT);
3319 case MC_CMD_POLL_BIST_PASSED:
3320 netif_dbg(efx, hw, efx->net_dev, "BIST passed.\n");
3322 case MC_CMD_POLL_BIST_TIMEOUT:
3323 netif_err(efx, hw, efx->net_dev, "BIST timed out\n");
3325 case MC_CMD_POLL_BIST_FAILED:
3326 netif_err(efx, hw, efx->net_dev, "BIST failed.\n");
3329 netif_err(efx, hw, efx->net_dev,
3330 "BIST returned unknown result %u", result);
3335 static int efx_ef10_run_bist(struct efx_nic *efx, u32 bist_type)
3339 netif_dbg(efx, drv, efx->net_dev, "starting BIST type %u\n", bist_type);
3341 rc = efx_ef10_start_bist(efx, bist_type);
3345 return efx_ef10_poll_bist(efx);
3349 efx_ef10_test_chip(struct efx_nic *efx, struct efx_self_tests *tests)
3353 efx_reset_down(efx, RESET_TYPE_WORLD);
3355 rc = efx_mcdi_rpc(efx, MC_CMD_ENABLE_OFFLINE_BIST,
3356 NULL, 0, NULL, 0, NULL);
3360 tests->memory = efx_ef10_run_bist(efx, MC_CMD_MC_MEM_BIST) ? -1 : 1;
3361 tests->registers = efx_ef10_run_bist(efx, MC_CMD_REG_BIST) ? -1 : 1;
3363 rc = efx_mcdi_reset(efx, RESET_TYPE_WORLD);
3366 rc2 = efx_reset_up(efx, RESET_TYPE_WORLD, rc == 0);
3367 return rc ? rc : rc2;
3370 #ifdef CONFIG_SFC_MTD
3372 struct efx_ef10_nvram_type_info {
3373 u16 type, type_mask;
3378 static const struct efx_ef10_nvram_type_info efx_ef10_nvram_types[] = {
3379 { NVRAM_PARTITION_TYPE_MC_FIRMWARE, 0, 0, "sfc_mcfw" },
3380 { NVRAM_PARTITION_TYPE_MC_FIRMWARE_BACKUP, 0, 0, "sfc_mcfw_backup" },
3381 { NVRAM_PARTITION_TYPE_EXPANSION_ROM, 0, 0, "sfc_exp_rom" },
3382 { NVRAM_PARTITION_TYPE_STATIC_CONFIG, 0, 0, "sfc_static_cfg" },
3383 { NVRAM_PARTITION_TYPE_DYNAMIC_CONFIG, 0, 0, "sfc_dynamic_cfg" },
3384 { NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT0, 0, 0, "sfc_exp_rom_cfg" },
3385 { NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT1, 0, 1, "sfc_exp_rom_cfg" },
3386 { NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT2, 0, 2, "sfc_exp_rom_cfg" },
3387 { NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT3, 0, 3, "sfc_exp_rom_cfg" },
3388 { NVRAM_PARTITION_TYPE_LICENSE, 0, 0, "sfc_license" },
3389 { NVRAM_PARTITION_TYPE_PHY_MIN, 0xff, 0, "sfc_phy_fw" },
3392 static int efx_ef10_mtd_probe_partition(struct efx_nic *efx,
3393 struct efx_mcdi_mtd_partition *part,
3396 MCDI_DECLARE_BUF(inbuf, MC_CMD_NVRAM_METADATA_IN_LEN);
3397 MCDI_DECLARE_BUF(outbuf, MC_CMD_NVRAM_METADATA_OUT_LENMAX);
3398 const struct efx_ef10_nvram_type_info *info;
3399 size_t size, erase_size, outlen;
3403 for (info = efx_ef10_nvram_types; ; info++) {
3405 efx_ef10_nvram_types + ARRAY_SIZE(efx_ef10_nvram_types))
3407 if ((type & ~info->type_mask) == info->type)
3410 if (info->port != efx_port_num(efx))
3413 rc = efx_mcdi_nvram_info(efx, type, &size, &erase_size, &protected);
3417 return -ENODEV; /* hide it */
3419 part->nvram_type = type;
3421 MCDI_SET_DWORD(inbuf, NVRAM_METADATA_IN_TYPE, type);
3422 rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_METADATA, inbuf, sizeof(inbuf),
3423 outbuf, sizeof(outbuf), &outlen);
3426 if (outlen < MC_CMD_NVRAM_METADATA_OUT_LENMIN)
3428 if (MCDI_DWORD(outbuf, NVRAM_METADATA_OUT_FLAGS) &
3429 (1 << MC_CMD_NVRAM_METADATA_OUT_SUBTYPE_VALID_LBN))
3430 part->fw_subtype = MCDI_DWORD(outbuf,
3431 NVRAM_METADATA_OUT_SUBTYPE);
3433 part->common.dev_type_name = "EF10 NVRAM manager";
3434 part->common.type_name = info->name;
3436 part->common.mtd.type = MTD_NORFLASH;
3437 part->common.mtd.flags = MTD_CAP_NORFLASH;
3438 part->common.mtd.size = size;
3439 part->common.mtd.erasesize = erase_size;
3444 static int efx_ef10_mtd_probe(struct efx_nic *efx)
3446 MCDI_DECLARE_BUF(outbuf, MC_CMD_NVRAM_PARTITIONS_OUT_LENMAX);
3447 struct efx_mcdi_mtd_partition *parts;
3448 size_t outlen, n_parts_total, i, n_parts;
3454 BUILD_BUG_ON(MC_CMD_NVRAM_PARTITIONS_IN_LEN != 0);
3455 rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_PARTITIONS, NULL, 0,
3456 outbuf, sizeof(outbuf), &outlen);
3459 if (outlen < MC_CMD_NVRAM_PARTITIONS_OUT_LENMIN)
3462 n_parts_total = MCDI_DWORD(outbuf, NVRAM_PARTITIONS_OUT_NUM_PARTITIONS);
3464 MCDI_VAR_ARRAY_LEN(outlen, NVRAM_PARTITIONS_OUT_TYPE_ID))
3467 parts = kcalloc(n_parts_total, sizeof(*parts), GFP_KERNEL);
3472 for (i = 0; i < n_parts_total; i++) {
3473 type = MCDI_ARRAY_DWORD(outbuf, NVRAM_PARTITIONS_OUT_TYPE_ID,
3475 rc = efx_ef10_mtd_probe_partition(efx, &parts[n_parts], type);
3478 else if (rc != -ENODEV)
3482 rc = efx_mtd_add(efx, &parts[0].common, n_parts, sizeof(*parts));
3489 #endif /* CONFIG_SFC_MTD */
3491 static void efx_ef10_ptp_write_host_time(struct efx_nic *efx, u32 host_time)
3493 _efx_writed(efx, cpu_to_le32(host_time), ER_DZ_MC_DB_LWRD);
3496 static int efx_ef10_rx_enable_timestamping(struct efx_channel *channel,
3499 MCDI_DECLARE_BUF(inbuf, MC_CMD_PTP_IN_TIME_EVENT_SUBSCRIBE_LEN);
3502 if (channel->sync_events_state == SYNC_EVENTS_REQUESTED ||
3503 channel->sync_events_state == SYNC_EVENTS_VALID ||
3504 (temp && channel->sync_events_state == SYNC_EVENTS_DISABLED))
3506 channel->sync_events_state = SYNC_EVENTS_REQUESTED;
3508 MCDI_SET_DWORD(inbuf, PTP_IN_OP, MC_CMD_PTP_OP_TIME_EVENT_SUBSCRIBE);
3509 MCDI_SET_DWORD(inbuf, PTP_IN_PERIPH_ID, 0);
3510 MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_SUBSCRIBE_QUEUE,
3513 rc = efx_mcdi_rpc(channel->efx, MC_CMD_PTP,
3514 inbuf, sizeof(inbuf), NULL, 0, NULL);
3517 channel->sync_events_state = temp ? SYNC_EVENTS_QUIESCENT :
3518 SYNC_EVENTS_DISABLED;
3523 static int efx_ef10_rx_disable_timestamping(struct efx_channel *channel,
3526 MCDI_DECLARE_BUF(inbuf, MC_CMD_PTP_IN_TIME_EVENT_UNSUBSCRIBE_LEN);
3529 if (channel->sync_events_state == SYNC_EVENTS_DISABLED ||
3530 (temp && channel->sync_events_state == SYNC_EVENTS_QUIESCENT))
3532 if (channel->sync_events_state == SYNC_EVENTS_QUIESCENT) {
3533 channel->sync_events_state = SYNC_EVENTS_DISABLED;
3536 channel->sync_events_state = temp ? SYNC_EVENTS_QUIESCENT :
3537 SYNC_EVENTS_DISABLED;
3539 MCDI_SET_DWORD(inbuf, PTP_IN_OP, MC_CMD_PTP_OP_TIME_EVENT_UNSUBSCRIBE);
3540 MCDI_SET_DWORD(inbuf, PTP_IN_PERIPH_ID, 0);
3541 MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_UNSUBSCRIBE_CONTROL,
3542 MC_CMD_PTP_IN_TIME_EVENT_UNSUBSCRIBE_SINGLE);
3543 MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_UNSUBSCRIBE_QUEUE,
3546 rc = efx_mcdi_rpc(channel->efx, MC_CMD_PTP,
3547 inbuf, sizeof(inbuf), NULL, 0, NULL);
3552 static int efx_ef10_ptp_set_ts_sync_events(struct efx_nic *efx, bool en,
3555 int (*set)(struct efx_channel *channel, bool temp);
3556 struct efx_channel *channel;
3559 efx_ef10_rx_enable_timestamping :
3560 efx_ef10_rx_disable_timestamping;
3562 efx_for_each_channel(channel, efx) {
3563 int rc = set(channel, temp);
3564 if (en && rc != 0) {
3565 efx_ef10_ptp_set_ts_sync_events(efx, false, temp);
3573 static int efx_ef10_ptp_set_ts_config(struct efx_nic *efx,
3574 struct hwtstamp_config *init)
3578 switch (init->rx_filter) {
3579 case HWTSTAMP_FILTER_NONE:
3580 efx_ef10_ptp_set_ts_sync_events(efx, false, false);
3581 /* if TX timestamping is still requested then leave PTP on */
3582 return efx_ptp_change_mode(efx,
3583 init->tx_type != HWTSTAMP_TX_OFF, 0);
3584 case HWTSTAMP_FILTER_ALL:
3585 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
3586 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
3587 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
3588 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
3589 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
3590 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
3591 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
3592 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
3593 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
3594 case HWTSTAMP_FILTER_PTP_V2_EVENT:
3595 case HWTSTAMP_FILTER_PTP_V2_SYNC:
3596 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
3597 init->rx_filter = HWTSTAMP_FILTER_ALL;
3598 rc = efx_ptp_change_mode(efx, true, 0);
3600 rc = efx_ef10_ptp_set_ts_sync_events(efx, true, false);
3602 efx_ptp_change_mode(efx, false, 0);
3609 const struct efx_nic_type efx_hunt_a0_nic_type = {
3610 .mem_map_size = efx_ef10_mem_map_size,
3611 .probe = efx_ef10_probe,
3612 .remove = efx_ef10_remove,
3613 .dimension_resources = efx_ef10_dimension_resources,
3614 .init = efx_ef10_init_nic,
3615 .fini = efx_port_dummy_op_void,
3616 .map_reset_reason = efx_mcdi_map_reset_reason,
3617 .map_reset_flags = efx_ef10_map_reset_flags,
3618 .reset = efx_ef10_reset,
3619 .probe_port = efx_mcdi_port_probe,
3620 .remove_port = efx_mcdi_port_remove,
3621 .fini_dmaq = efx_ef10_fini_dmaq,
3622 .prepare_flr = efx_ef10_prepare_flr,
3623 .finish_flr = efx_port_dummy_op_void,
3624 .describe_stats = efx_ef10_describe_stats,
3625 .update_stats = efx_ef10_update_stats,
3626 .start_stats = efx_mcdi_mac_start_stats,
3627 .pull_stats = efx_mcdi_mac_pull_stats,
3628 .stop_stats = efx_mcdi_mac_stop_stats,
3629 .set_id_led = efx_mcdi_set_id_led,
3630 .push_irq_moderation = efx_ef10_push_irq_moderation,
3631 .reconfigure_mac = efx_ef10_mac_reconfigure,
3632 .check_mac_fault = efx_mcdi_mac_check_fault,
3633 .reconfigure_port = efx_mcdi_port_reconfigure,
3634 .get_wol = efx_ef10_get_wol,
3635 .set_wol = efx_ef10_set_wol,
3636 .resume_wol = efx_port_dummy_op_void,
3637 .test_chip = efx_ef10_test_chip,
3638 .test_nvram = efx_mcdi_nvram_test_all,
3639 .mcdi_request = efx_ef10_mcdi_request,
3640 .mcdi_poll_response = efx_ef10_mcdi_poll_response,
3641 .mcdi_read_response = efx_ef10_mcdi_read_response,
3642 .mcdi_poll_reboot = efx_ef10_mcdi_poll_reboot,
3643 .irq_enable_master = efx_port_dummy_op_void,
3644 .irq_test_generate = efx_ef10_irq_test_generate,
3645 .irq_disable_non_ev = efx_port_dummy_op_void,
3646 .irq_handle_msi = efx_ef10_msi_interrupt,
3647 .irq_handle_legacy = efx_ef10_legacy_interrupt,
3648 .tx_probe = efx_ef10_tx_probe,
3649 .tx_init = efx_ef10_tx_init,
3650 .tx_remove = efx_ef10_tx_remove,
3651 .tx_write = efx_ef10_tx_write,
3652 .rx_push_rss_config = efx_ef10_rx_push_rss_config,
3653 .rx_probe = efx_ef10_rx_probe,
3654 .rx_init = efx_ef10_rx_init,
3655 .rx_remove = efx_ef10_rx_remove,
3656 .rx_write = efx_ef10_rx_write,
3657 .rx_defer_refill = efx_ef10_rx_defer_refill,
3658 .ev_probe = efx_ef10_ev_probe,
3659 .ev_init = efx_ef10_ev_init,
3660 .ev_fini = efx_ef10_ev_fini,
3661 .ev_remove = efx_ef10_ev_remove,
3662 .ev_process = efx_ef10_ev_process,
3663 .ev_read_ack = efx_ef10_ev_read_ack,
3664 .ev_test_generate = efx_ef10_ev_test_generate,
3665 .filter_table_probe = efx_ef10_filter_table_probe,
3666 .filter_table_restore = efx_ef10_filter_table_restore,
3667 .filter_table_remove = efx_ef10_filter_table_remove,
3668 .filter_update_rx_scatter = efx_ef10_filter_update_rx_scatter,
3669 .filter_insert = efx_ef10_filter_insert,
3670 .filter_remove_safe = efx_ef10_filter_remove_safe,
3671 .filter_get_safe = efx_ef10_filter_get_safe,
3672 .filter_clear_rx = efx_ef10_filter_clear_rx,
3673 .filter_count_rx_used = efx_ef10_filter_count_rx_used,
3674 .filter_get_rx_id_limit = efx_ef10_filter_get_rx_id_limit,
3675 .filter_get_rx_ids = efx_ef10_filter_get_rx_ids,
3676 #ifdef CONFIG_RFS_ACCEL
3677 .filter_rfs_insert = efx_ef10_filter_rfs_insert,
3678 .filter_rfs_expire_one = efx_ef10_filter_rfs_expire_one,
3680 #ifdef CONFIG_SFC_MTD
3681 .mtd_probe = efx_ef10_mtd_probe,
3682 .mtd_rename = efx_mcdi_mtd_rename,
3683 .mtd_read = efx_mcdi_mtd_read,
3684 .mtd_erase = efx_mcdi_mtd_erase,
3685 .mtd_write = efx_mcdi_mtd_write,
3686 .mtd_sync = efx_mcdi_mtd_sync,
3688 .ptp_write_host_time = efx_ef10_ptp_write_host_time,
3689 .ptp_set_ts_sync_events = efx_ef10_ptp_set_ts_sync_events,
3690 .ptp_set_ts_config = efx_ef10_ptp_set_ts_config,
3692 .revision = EFX_REV_HUNT_A0,
3693 .max_dma_mask = DMA_BIT_MASK(ESF_DZ_TX_KER_BUF_ADDR_WIDTH),
3694 .rx_prefix_size = ES_DZ_RX_PREFIX_SIZE,
3695 .rx_hash_offset = ES_DZ_RX_PREFIX_HASH_OFST,
3696 .rx_ts_offset = ES_DZ_RX_PREFIX_TSTAMP_OFST,
3697 .can_rx_scatter = true,
3698 .always_rx_scatter = true,
3699 .max_interrupt_mode = EFX_INT_MODE_MSIX,
3700 .timer_period_max = 1 << ERF_DD_EVQ_IND_TIMER_VAL_WIDTH,
3701 .offload_features = (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
3702 NETIF_F_RXHASH | NETIF_F_NTUPLE),
3704 .max_rx_ip_filters = HUNT_FILTER_TBL_ROWS,
3705 .hwtstamp_filters = 1 << HWTSTAMP_FILTER_NONE |
3706 1 << HWTSTAMP_FILTER_ALL,