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. STACK_OLD is
57 * used to mark and sweep stack-owned MAC filters.
59 #define EFX_EF10_FILTER_FLAG_BUSY 1UL
60 #define EFX_EF10_FILTER_FLAG_STACK_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_STACK_UC_MAX 32
67 #define EFX_EF10_FILTER_STACK_MC_MAX 256
71 } stack_uc_list[EFX_EF10_FILTER_STACK_UC_MAX],
72 stack_mc_list[EFX_EF10_FILTER_STACK_MC_MAX];
73 int stack_uc_count; /* negative for PROMISC */
74 int stack_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)
166 MCDI_PTR(outbuf, GET_MAC_ADDRESSES_OUT_MAC_ADDR_BASE), ETH_ALEN);
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 we need
176 * multiple 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 uc_mem_map_size = PAGE_ALIGN((min_vis - 1) * EFX_VI_PAGE_SIZE +
571 if (nic_data->n_piobufs) {
572 pio_write_vi_base = uc_mem_map_size / EFX_VI_PAGE_SIZE;
573 wc_mem_map_size = (PAGE_ALIGN((pio_write_vi_base +
574 nic_data->n_piobufs) *
577 max_vis = pio_write_vi_base + nic_data->n_piobufs;
579 pio_write_vi_base = 0;
584 /* In case the last attached driver failed to free VIs, do it now */
585 rc = efx_ef10_free_vis(efx);
589 rc = efx_ef10_alloc_vis(efx, min_vis, max_vis);
593 /* If we didn't get enough VIs to map all the PIO buffers, free the
596 if (nic_data->n_piobufs &&
597 nic_data->n_allocated_vis <
598 pio_write_vi_base + nic_data->n_piobufs) {
599 netif_dbg(efx, probe, efx->net_dev,
600 "%u VIs are not sufficient to map %u PIO buffers\n",
601 nic_data->n_allocated_vis, nic_data->n_piobufs);
602 efx_ef10_free_piobufs(efx);
605 /* Shrink the original UC mapping of the memory BAR */
606 membase = ioremap_nocache(efx->membase_phys, uc_mem_map_size);
608 netif_err(efx, probe, efx->net_dev,
609 "could not shrink memory BAR to %x\n",
613 iounmap(efx->membase);
614 efx->membase = membase;
616 /* Set up the WC mapping if needed */
617 if (wc_mem_map_size) {
618 nic_data->wc_membase = ioremap_wc(efx->membase_phys +
621 if (!nic_data->wc_membase) {
622 netif_err(efx, probe, efx->net_dev,
623 "could not allocate WC mapping of size %x\n",
627 nic_data->pio_write_vi_base = pio_write_vi_base;
628 nic_data->pio_write_base =
629 nic_data->wc_membase +
630 (pio_write_vi_base * EFX_VI_PAGE_SIZE + ER_DZ_TX_PIOBUF -
633 rc = efx_ef10_link_piobufs(efx);
635 efx_ef10_free_piobufs(efx);
638 netif_dbg(efx, probe, efx->net_dev,
639 "memory BAR at %pa (virtual %p+%x UC, %p+%x WC)\n",
640 &efx->membase_phys, efx->membase, uc_mem_map_size,
641 nic_data->wc_membase, wc_mem_map_size);
646 static int efx_ef10_init_nic(struct efx_nic *efx)
648 struct efx_ef10_nic_data *nic_data = efx->nic_data;
651 if (nic_data->must_check_datapath_caps) {
652 rc = efx_ef10_init_datapath_caps(efx);
655 nic_data->must_check_datapath_caps = false;
658 if (nic_data->must_realloc_vis) {
659 /* We cannot let the number of VIs change now */
660 rc = efx_ef10_alloc_vis(efx, nic_data->n_allocated_vis,
661 nic_data->n_allocated_vis);
664 nic_data->must_realloc_vis = false;
667 if (nic_data->must_restore_piobufs && nic_data->n_piobufs) {
668 rc = efx_ef10_alloc_piobufs(efx, nic_data->n_piobufs);
670 rc = efx_ef10_link_piobufs(efx);
672 efx_ef10_free_piobufs(efx);
675 /* Log an error on failure, but this is non-fatal */
677 netif_err(efx, drv, efx->net_dev,
678 "failed to restore PIO buffers (%d)\n", rc);
679 nic_data->must_restore_piobufs = false;
682 efx_ef10_rx_push_rss_config(efx);
686 static int efx_ef10_map_reset_flags(u32 *flags)
689 EF10_RESET_PORT = ((ETH_RESET_MAC | ETH_RESET_PHY) <<
690 ETH_RESET_SHARED_SHIFT),
691 EF10_RESET_MC = ((ETH_RESET_DMA | ETH_RESET_FILTER |
692 ETH_RESET_OFFLOAD | ETH_RESET_MAC |
693 ETH_RESET_PHY | ETH_RESET_MGMT) <<
694 ETH_RESET_SHARED_SHIFT)
697 /* We assume for now that our PCI function is permitted to
701 if ((*flags & EF10_RESET_MC) == EF10_RESET_MC) {
702 *flags &= ~EF10_RESET_MC;
703 return RESET_TYPE_WORLD;
706 if ((*flags & EF10_RESET_PORT) == EF10_RESET_PORT) {
707 *flags &= ~EF10_RESET_PORT;
708 return RESET_TYPE_ALL;
711 /* no invisible reset implemented */
716 #define EF10_DMA_STAT(ext_name, mcdi_name) \
717 [EF10_STAT_ ## ext_name] = \
718 { #ext_name, 64, 8 * MC_CMD_MAC_ ## mcdi_name }
719 #define EF10_DMA_INVIS_STAT(int_name, mcdi_name) \
720 [EF10_STAT_ ## int_name] = \
721 { NULL, 64, 8 * MC_CMD_MAC_ ## mcdi_name }
722 #define EF10_OTHER_STAT(ext_name) \
723 [EF10_STAT_ ## ext_name] = { #ext_name, 0, 0 }
725 static const struct efx_hw_stat_desc efx_ef10_stat_desc[EF10_STAT_COUNT] = {
726 EF10_DMA_STAT(tx_bytes, TX_BYTES),
727 EF10_DMA_STAT(tx_packets, TX_PKTS),
728 EF10_DMA_STAT(tx_pause, TX_PAUSE_PKTS),
729 EF10_DMA_STAT(tx_control, TX_CONTROL_PKTS),
730 EF10_DMA_STAT(tx_unicast, TX_UNICAST_PKTS),
731 EF10_DMA_STAT(tx_multicast, TX_MULTICAST_PKTS),
732 EF10_DMA_STAT(tx_broadcast, TX_BROADCAST_PKTS),
733 EF10_DMA_STAT(tx_lt64, TX_LT64_PKTS),
734 EF10_DMA_STAT(tx_64, TX_64_PKTS),
735 EF10_DMA_STAT(tx_65_to_127, TX_65_TO_127_PKTS),
736 EF10_DMA_STAT(tx_128_to_255, TX_128_TO_255_PKTS),
737 EF10_DMA_STAT(tx_256_to_511, TX_256_TO_511_PKTS),
738 EF10_DMA_STAT(tx_512_to_1023, TX_512_TO_1023_PKTS),
739 EF10_DMA_STAT(tx_1024_to_15xx, TX_1024_TO_15XX_PKTS),
740 EF10_DMA_STAT(tx_15xx_to_jumbo, TX_15XX_TO_JUMBO_PKTS),
741 EF10_DMA_STAT(rx_bytes, RX_BYTES),
742 EF10_DMA_INVIS_STAT(rx_bytes_minus_good_bytes, RX_BAD_BYTES),
743 EF10_OTHER_STAT(rx_good_bytes),
744 EF10_OTHER_STAT(rx_bad_bytes),
745 EF10_DMA_STAT(rx_packets, RX_PKTS),
746 EF10_DMA_STAT(rx_good, RX_GOOD_PKTS),
747 EF10_DMA_STAT(rx_bad, RX_BAD_FCS_PKTS),
748 EF10_DMA_STAT(rx_pause, RX_PAUSE_PKTS),
749 EF10_DMA_STAT(rx_control, RX_CONTROL_PKTS),
750 EF10_DMA_STAT(rx_unicast, RX_UNICAST_PKTS),
751 EF10_DMA_STAT(rx_multicast, RX_MULTICAST_PKTS),
752 EF10_DMA_STAT(rx_broadcast, RX_BROADCAST_PKTS),
753 EF10_DMA_STAT(rx_lt64, RX_UNDERSIZE_PKTS),
754 EF10_DMA_STAT(rx_64, RX_64_PKTS),
755 EF10_DMA_STAT(rx_65_to_127, RX_65_TO_127_PKTS),
756 EF10_DMA_STAT(rx_128_to_255, RX_128_TO_255_PKTS),
757 EF10_DMA_STAT(rx_256_to_511, RX_256_TO_511_PKTS),
758 EF10_DMA_STAT(rx_512_to_1023, RX_512_TO_1023_PKTS),
759 EF10_DMA_STAT(rx_1024_to_15xx, RX_1024_TO_15XX_PKTS),
760 EF10_DMA_STAT(rx_15xx_to_jumbo, RX_15XX_TO_JUMBO_PKTS),
761 EF10_DMA_STAT(rx_gtjumbo, RX_GTJUMBO_PKTS),
762 EF10_DMA_STAT(rx_bad_gtjumbo, RX_JABBER_PKTS),
763 EF10_DMA_STAT(rx_overflow, RX_OVERFLOW_PKTS),
764 EF10_DMA_STAT(rx_align_error, RX_ALIGN_ERROR_PKTS),
765 EF10_DMA_STAT(rx_length_error, RX_LENGTH_ERROR_PKTS),
766 EF10_DMA_STAT(rx_nodesc_drops, RX_NODESC_DROPS),
767 EF10_DMA_STAT(rx_pm_trunc_bb_overflow, PM_TRUNC_BB_OVERFLOW),
768 EF10_DMA_STAT(rx_pm_discard_bb_overflow, PM_DISCARD_BB_OVERFLOW),
769 EF10_DMA_STAT(rx_pm_trunc_vfifo_full, PM_TRUNC_VFIFO_FULL),
770 EF10_DMA_STAT(rx_pm_discard_vfifo_full, PM_DISCARD_VFIFO_FULL),
771 EF10_DMA_STAT(rx_pm_trunc_qbb, PM_TRUNC_QBB),
772 EF10_DMA_STAT(rx_pm_discard_qbb, PM_DISCARD_QBB),
773 EF10_DMA_STAT(rx_pm_discard_mapping, PM_DISCARD_MAPPING),
774 EF10_DMA_STAT(rx_dp_q_disabled_packets, RXDP_Q_DISABLED_PKTS),
775 EF10_DMA_STAT(rx_dp_di_dropped_packets, RXDP_DI_DROPPED_PKTS),
776 EF10_DMA_STAT(rx_dp_streaming_packets, RXDP_STREAMING_PKTS),
777 EF10_DMA_STAT(rx_dp_emerg_fetch, RXDP_EMERGENCY_FETCH_CONDITIONS),
778 EF10_DMA_STAT(rx_dp_emerg_wait, RXDP_EMERGENCY_WAIT_CONDITIONS),
781 #define HUNT_COMMON_STAT_MASK ((1ULL << EF10_STAT_tx_bytes) | \
782 (1ULL << EF10_STAT_tx_packets) | \
783 (1ULL << EF10_STAT_tx_pause) | \
784 (1ULL << EF10_STAT_tx_unicast) | \
785 (1ULL << EF10_STAT_tx_multicast) | \
786 (1ULL << EF10_STAT_tx_broadcast) | \
787 (1ULL << EF10_STAT_rx_bytes) | \
788 (1ULL << EF10_STAT_rx_bytes_minus_good_bytes) | \
789 (1ULL << EF10_STAT_rx_good_bytes) | \
790 (1ULL << EF10_STAT_rx_bad_bytes) | \
791 (1ULL << EF10_STAT_rx_packets) | \
792 (1ULL << EF10_STAT_rx_good) | \
793 (1ULL << EF10_STAT_rx_bad) | \
794 (1ULL << EF10_STAT_rx_pause) | \
795 (1ULL << EF10_STAT_rx_control) | \
796 (1ULL << EF10_STAT_rx_unicast) | \
797 (1ULL << EF10_STAT_rx_multicast) | \
798 (1ULL << EF10_STAT_rx_broadcast) | \
799 (1ULL << EF10_STAT_rx_lt64) | \
800 (1ULL << EF10_STAT_rx_64) | \
801 (1ULL << EF10_STAT_rx_65_to_127) | \
802 (1ULL << EF10_STAT_rx_128_to_255) | \
803 (1ULL << EF10_STAT_rx_256_to_511) | \
804 (1ULL << EF10_STAT_rx_512_to_1023) | \
805 (1ULL << EF10_STAT_rx_1024_to_15xx) | \
806 (1ULL << EF10_STAT_rx_15xx_to_jumbo) | \
807 (1ULL << EF10_STAT_rx_gtjumbo) | \
808 (1ULL << EF10_STAT_rx_bad_gtjumbo) | \
809 (1ULL << EF10_STAT_rx_overflow) | \
810 (1ULL << EF10_STAT_rx_nodesc_drops))
812 /* These statistics are only provided by the 10G MAC. For a 10G/40G
813 * switchable port we do not expose these because they might not
814 * include all the packets they should.
816 #define HUNT_10G_ONLY_STAT_MASK ((1ULL << EF10_STAT_tx_control) | \
817 (1ULL << EF10_STAT_tx_lt64) | \
818 (1ULL << EF10_STAT_tx_64) | \
819 (1ULL << EF10_STAT_tx_65_to_127) | \
820 (1ULL << EF10_STAT_tx_128_to_255) | \
821 (1ULL << EF10_STAT_tx_256_to_511) | \
822 (1ULL << EF10_STAT_tx_512_to_1023) | \
823 (1ULL << EF10_STAT_tx_1024_to_15xx) | \
824 (1ULL << EF10_STAT_tx_15xx_to_jumbo))
826 /* These statistics are only provided by the 40G MAC. For a 10G/40G
827 * switchable port we do expose these because the errors will otherwise
830 #define HUNT_40G_EXTRA_STAT_MASK ((1ULL << EF10_STAT_rx_align_error) | \
831 (1ULL << EF10_STAT_rx_length_error))
833 /* These statistics are only provided if the firmware supports the
834 * capability PM_AND_RXDP_COUNTERS.
836 #define HUNT_PM_AND_RXDP_STAT_MASK ( \
837 (1ULL << EF10_STAT_rx_pm_trunc_bb_overflow) | \
838 (1ULL << EF10_STAT_rx_pm_discard_bb_overflow) | \
839 (1ULL << EF10_STAT_rx_pm_trunc_vfifo_full) | \
840 (1ULL << EF10_STAT_rx_pm_discard_vfifo_full) | \
841 (1ULL << EF10_STAT_rx_pm_trunc_qbb) | \
842 (1ULL << EF10_STAT_rx_pm_discard_qbb) | \
843 (1ULL << EF10_STAT_rx_pm_discard_mapping) | \
844 (1ULL << EF10_STAT_rx_dp_q_disabled_packets) | \
845 (1ULL << EF10_STAT_rx_dp_di_dropped_packets) | \
846 (1ULL << EF10_STAT_rx_dp_streaming_packets) | \
847 (1ULL << EF10_STAT_rx_dp_emerg_fetch) | \
848 (1ULL << EF10_STAT_rx_dp_emerg_wait))
850 static u64 efx_ef10_raw_stat_mask(struct efx_nic *efx)
852 u64 raw_mask = HUNT_COMMON_STAT_MASK;
853 u32 port_caps = efx_mcdi_phy_get_caps(efx);
854 struct efx_ef10_nic_data *nic_data = efx->nic_data;
856 if (port_caps & (1 << MC_CMD_PHY_CAP_40000FDX_LBN))
857 raw_mask |= HUNT_40G_EXTRA_STAT_MASK;
859 raw_mask |= HUNT_10G_ONLY_STAT_MASK;
861 if (nic_data->datapath_caps &
862 (1 << MC_CMD_GET_CAPABILITIES_OUT_PM_AND_RXDP_COUNTERS_LBN))
863 raw_mask |= HUNT_PM_AND_RXDP_STAT_MASK;
868 static void efx_ef10_get_stat_mask(struct efx_nic *efx, unsigned long *mask)
870 u64 raw_mask = efx_ef10_raw_stat_mask(efx);
872 #if BITS_PER_LONG == 64
875 mask[0] = raw_mask & 0xffffffff;
876 mask[1] = raw_mask >> 32;
880 static size_t efx_ef10_describe_stats(struct efx_nic *efx, u8 *names)
882 DECLARE_BITMAP(mask, EF10_STAT_COUNT);
884 efx_ef10_get_stat_mask(efx, mask);
885 return efx_nic_describe_stats(efx_ef10_stat_desc, EF10_STAT_COUNT,
889 static int efx_ef10_try_update_nic_stats(struct efx_nic *efx)
891 struct efx_ef10_nic_data *nic_data = efx->nic_data;
892 DECLARE_BITMAP(mask, EF10_STAT_COUNT);
893 __le64 generation_start, generation_end;
894 u64 *stats = nic_data->stats;
897 efx_ef10_get_stat_mask(efx, mask);
899 dma_stats = efx->stats_buffer.addr;
900 nic_data = efx->nic_data;
902 generation_end = dma_stats[MC_CMD_MAC_GENERATION_END];
903 if (generation_end == EFX_MC_STATS_GENERATION_INVALID)
906 efx_nic_update_stats(efx_ef10_stat_desc, EF10_STAT_COUNT, mask,
907 stats, efx->stats_buffer.addr, false);
909 generation_start = dma_stats[MC_CMD_MAC_GENERATION_START];
910 if (generation_end != generation_start)
913 /* Update derived statistics */
914 efx_nic_fix_nodesc_drop_stat(efx, &stats[EF10_STAT_rx_nodesc_drops]);
915 stats[EF10_STAT_rx_good_bytes] =
916 stats[EF10_STAT_rx_bytes] -
917 stats[EF10_STAT_rx_bytes_minus_good_bytes];
918 efx_update_diff_stat(&stats[EF10_STAT_rx_bad_bytes],
919 stats[EF10_STAT_rx_bytes_minus_good_bytes]);
925 static size_t efx_ef10_update_stats(struct efx_nic *efx, u64 *full_stats,
926 struct rtnl_link_stats64 *core_stats)
928 DECLARE_BITMAP(mask, EF10_STAT_COUNT);
929 struct efx_ef10_nic_data *nic_data = efx->nic_data;
930 u64 *stats = nic_data->stats;
931 size_t stats_count = 0, index;
934 efx_ef10_get_stat_mask(efx, mask);
936 /* If we're unlucky enough to read statistics during the DMA, wait
937 * up to 10ms for it to finish (typically takes <500us)
939 for (retry = 0; retry < 100; ++retry) {
940 if (efx_ef10_try_update_nic_stats(efx) == 0)
946 for_each_set_bit(index, mask, EF10_STAT_COUNT) {
947 if (efx_ef10_stat_desc[index].name) {
948 *full_stats++ = stats[index];
955 core_stats->rx_packets = stats[EF10_STAT_rx_packets];
956 core_stats->tx_packets = stats[EF10_STAT_tx_packets];
957 core_stats->rx_bytes = stats[EF10_STAT_rx_bytes];
958 core_stats->tx_bytes = stats[EF10_STAT_tx_bytes];
959 core_stats->rx_dropped = stats[EF10_STAT_rx_nodesc_drops];
960 core_stats->multicast = stats[EF10_STAT_rx_multicast];
961 core_stats->rx_length_errors =
962 stats[EF10_STAT_rx_gtjumbo] +
963 stats[EF10_STAT_rx_length_error];
964 core_stats->rx_crc_errors = stats[EF10_STAT_rx_bad];
965 core_stats->rx_frame_errors = stats[EF10_STAT_rx_align_error];
966 core_stats->rx_fifo_errors = stats[EF10_STAT_rx_overflow];
967 core_stats->rx_errors = (core_stats->rx_length_errors +
968 core_stats->rx_crc_errors +
969 core_stats->rx_frame_errors);
975 static void efx_ef10_push_irq_moderation(struct efx_channel *channel)
977 struct efx_nic *efx = channel->efx;
978 unsigned int mode, value;
979 efx_dword_t timer_cmd;
981 if (channel->irq_moderation) {
983 value = channel->irq_moderation - 1;
989 if (EFX_EF10_WORKAROUND_35388(efx)) {
990 EFX_POPULATE_DWORD_3(timer_cmd, ERF_DD_EVQ_IND_TIMER_FLAGS,
991 EFE_DD_EVQ_IND_TIMER_FLAGS,
992 ERF_DD_EVQ_IND_TIMER_MODE, mode,
993 ERF_DD_EVQ_IND_TIMER_VAL, value);
994 efx_writed_page(efx, &timer_cmd, ER_DD_EVQ_INDIRECT,
997 EFX_POPULATE_DWORD_2(timer_cmd, ERF_DZ_TC_TIMER_MODE, mode,
998 ERF_DZ_TC_TIMER_VAL, value);
999 efx_writed_page(efx, &timer_cmd, ER_DZ_EVQ_TMR,
1004 static void efx_ef10_get_wol(struct efx_nic *efx, struct ethtool_wolinfo *wol)
1008 memset(&wol->sopass, 0, sizeof(wol->sopass));
1011 static int efx_ef10_set_wol(struct efx_nic *efx, u32 type)
1018 static void efx_ef10_mcdi_request(struct efx_nic *efx,
1019 const efx_dword_t *hdr, size_t hdr_len,
1020 const efx_dword_t *sdu, size_t sdu_len)
1022 struct efx_ef10_nic_data *nic_data = efx->nic_data;
1023 u8 *pdu = nic_data->mcdi_buf.addr;
1025 memcpy(pdu, hdr, hdr_len);
1026 memcpy(pdu + hdr_len, sdu, sdu_len);
1029 /* The hardware provides 'low' and 'high' (doorbell) registers
1030 * for passing the 64-bit address of an MCDI request to
1031 * firmware. However the dwords are swapped by firmware. The
1032 * least significant bits of the doorbell are then 0 for all
1033 * MCDI requests due to alignment.
1035 _efx_writed(efx, cpu_to_le32((u64)nic_data->mcdi_buf.dma_addr >> 32),
1037 _efx_writed(efx, cpu_to_le32((u32)nic_data->mcdi_buf.dma_addr),
1041 static bool efx_ef10_mcdi_poll_response(struct efx_nic *efx)
1043 struct efx_ef10_nic_data *nic_data = efx->nic_data;
1044 const efx_dword_t hdr = *(const efx_dword_t *)nic_data->mcdi_buf.addr;
1047 return EFX_DWORD_FIELD(hdr, MCDI_HEADER_RESPONSE);
1051 efx_ef10_mcdi_read_response(struct efx_nic *efx, efx_dword_t *outbuf,
1052 size_t offset, size_t outlen)
1054 struct efx_ef10_nic_data *nic_data = efx->nic_data;
1055 const u8 *pdu = nic_data->mcdi_buf.addr;
1057 memcpy(outbuf, pdu + offset, outlen);
1060 static int efx_ef10_mcdi_poll_reboot(struct efx_nic *efx)
1062 struct efx_ef10_nic_data *nic_data = efx->nic_data;
1065 rc = efx_ef10_get_warm_boot_count(efx);
1067 /* The firmware is presumably in the process of
1068 * rebooting. However, we are supposed to report each
1069 * reboot just once, so we must only do that once we
1070 * can read and store the updated warm boot count.
1075 if (rc == nic_data->warm_boot_count)
1078 nic_data->warm_boot_count = rc;
1080 /* All our allocations have been reset */
1081 nic_data->must_realloc_vis = true;
1082 nic_data->must_restore_filters = true;
1083 nic_data->must_restore_piobufs = true;
1084 nic_data->rx_rss_context = EFX_EF10_RSS_CONTEXT_INVALID;
1086 /* The datapath firmware might have been changed */
1087 nic_data->must_check_datapath_caps = true;
1089 /* MAC statistics have been cleared on the NIC; clear the local
1090 * statistic that we update with efx_update_diff_stat().
1092 nic_data->stats[EF10_STAT_rx_bad_bytes] = 0;
1097 /* Handle an MSI interrupt
1099 * Handle an MSI hardware interrupt. This routine schedules event
1100 * queue processing. No interrupt acknowledgement cycle is necessary.
1101 * Also, we never need to check that the interrupt is for us, since
1102 * MSI interrupts cannot be shared.
1104 static irqreturn_t efx_ef10_msi_interrupt(int irq, void *dev_id)
1106 struct efx_msi_context *context = dev_id;
1107 struct efx_nic *efx = context->efx;
1109 netif_vdbg(efx, intr, efx->net_dev,
1110 "IRQ %d on CPU %d\n", irq, raw_smp_processor_id());
1112 if (likely(ACCESS_ONCE(efx->irq_soft_enabled))) {
1113 /* Note test interrupts */
1114 if (context->index == efx->irq_level)
1115 efx->last_irq_cpu = raw_smp_processor_id();
1117 /* Schedule processing of the channel */
1118 efx_schedule_channel_irq(efx->channel[context->index]);
1124 static irqreturn_t efx_ef10_legacy_interrupt(int irq, void *dev_id)
1126 struct efx_nic *efx = dev_id;
1127 bool soft_enabled = ACCESS_ONCE(efx->irq_soft_enabled);
1128 struct efx_channel *channel;
1132 /* Read the ISR which also ACKs the interrupts */
1133 efx_readd(efx, ®, ER_DZ_BIU_INT_ISR);
1134 queues = EFX_DWORD_FIELD(reg, ERF_DZ_ISR_REG);
1139 if (likely(soft_enabled)) {
1140 /* Note test interrupts */
1141 if (queues & (1U << efx->irq_level))
1142 efx->last_irq_cpu = raw_smp_processor_id();
1144 efx_for_each_channel(channel, efx) {
1146 efx_schedule_channel_irq(channel);
1151 netif_vdbg(efx, intr, efx->net_dev,
1152 "IRQ %d on CPU %d status " EFX_DWORD_FMT "\n",
1153 irq, raw_smp_processor_id(), EFX_DWORD_VAL(reg));
1158 static void efx_ef10_irq_test_generate(struct efx_nic *efx)
1160 MCDI_DECLARE_BUF(inbuf, MC_CMD_TRIGGER_INTERRUPT_IN_LEN);
1162 BUILD_BUG_ON(MC_CMD_TRIGGER_INTERRUPT_OUT_LEN != 0);
1164 MCDI_SET_DWORD(inbuf, TRIGGER_INTERRUPT_IN_INTR_LEVEL, efx->irq_level);
1165 (void) efx_mcdi_rpc(efx, MC_CMD_TRIGGER_INTERRUPT,
1166 inbuf, sizeof(inbuf), NULL, 0, NULL);
1169 static int efx_ef10_tx_probe(struct efx_tx_queue *tx_queue)
1171 return efx_nic_alloc_buffer(tx_queue->efx, &tx_queue->txd.buf,
1172 (tx_queue->ptr_mask + 1) *
1173 sizeof(efx_qword_t),
1177 /* This writes to the TX_DESC_WPTR and also pushes data */
1178 static inline void efx_ef10_push_tx_desc(struct efx_tx_queue *tx_queue,
1179 const efx_qword_t *txd)
1181 unsigned int write_ptr;
1184 write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
1185 EFX_POPULATE_OWORD_1(reg, ERF_DZ_TX_DESC_WPTR, write_ptr);
1186 reg.qword[0] = *txd;
1187 efx_writeo_page(tx_queue->efx, ®,
1188 ER_DZ_TX_DESC_UPD, tx_queue->queue);
1191 static void efx_ef10_tx_init(struct efx_tx_queue *tx_queue)
1193 MCDI_DECLARE_BUF(inbuf, MC_CMD_INIT_TXQ_IN_LEN(EFX_MAX_DMAQ_SIZE * 8 /
1195 MCDI_DECLARE_BUF(outbuf, MC_CMD_INIT_TXQ_OUT_LEN);
1196 bool csum_offload = tx_queue->queue & EFX_TXQ_TYPE_OFFLOAD;
1197 size_t entries = tx_queue->txd.buf.len / EFX_BUF_SIZE;
1198 struct efx_channel *channel = tx_queue->channel;
1199 struct efx_nic *efx = tx_queue->efx;
1200 size_t inlen, outlen;
1201 dma_addr_t dma_addr;
1206 MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_SIZE, tx_queue->ptr_mask + 1);
1207 MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_TARGET_EVQ, channel->channel);
1208 MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_LABEL, tx_queue->queue);
1209 MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_INSTANCE, tx_queue->queue);
1210 MCDI_POPULATE_DWORD_2(inbuf, INIT_TXQ_IN_FLAGS,
1211 INIT_TXQ_IN_FLAG_IP_CSUM_DIS, !csum_offload,
1212 INIT_TXQ_IN_FLAG_TCP_CSUM_DIS, !csum_offload);
1213 MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_OWNER_ID, 0);
1214 MCDI_SET_DWORD(inbuf, INIT_TXQ_IN_PORT_ID, EVB_PORT_ID_ASSIGNED);
1216 dma_addr = tx_queue->txd.buf.dma_addr;
1218 netif_dbg(efx, hw, efx->net_dev, "pushing TXQ %d. %zu entries (%llx)\n",
1219 tx_queue->queue, entries, (u64)dma_addr);
1221 for (i = 0; i < entries; ++i) {
1222 MCDI_SET_ARRAY_QWORD(inbuf, INIT_TXQ_IN_DMA_ADDR, i, dma_addr);
1223 dma_addr += EFX_BUF_SIZE;
1226 inlen = MC_CMD_INIT_TXQ_IN_LEN(entries);
1228 rc = efx_mcdi_rpc(efx, MC_CMD_INIT_TXQ, inbuf, inlen,
1229 outbuf, sizeof(outbuf), &outlen);
1233 /* A previous user of this TX queue might have set us up the
1234 * bomb by writing a descriptor to the TX push collector but
1235 * not the doorbell. (Each collector belongs to a port, not a
1236 * queue or function, so cannot easily be reset.) We must
1237 * attempt to push a no-op descriptor in its place.
1239 tx_queue->buffer[0].flags = EFX_TX_BUF_OPTION;
1240 tx_queue->insert_count = 1;
1241 txd = efx_tx_desc(tx_queue, 0);
1242 EFX_POPULATE_QWORD_4(*txd,
1243 ESF_DZ_TX_DESC_IS_OPT, true,
1244 ESF_DZ_TX_OPTION_TYPE,
1245 ESE_DZ_TX_OPTION_DESC_CRC_CSUM,
1246 ESF_DZ_TX_OPTION_UDP_TCP_CSUM, csum_offload,
1247 ESF_DZ_TX_OPTION_IP_CSUM, csum_offload);
1248 tx_queue->write_count = 1;
1250 efx_ef10_push_tx_desc(tx_queue, txd);
1255 netdev_WARN(efx->net_dev, "failed to initialise TXQ %d\n",
1259 static void efx_ef10_tx_fini(struct efx_tx_queue *tx_queue)
1261 MCDI_DECLARE_BUF(inbuf, MC_CMD_FINI_TXQ_IN_LEN);
1262 MCDI_DECLARE_BUF(outbuf, MC_CMD_FINI_TXQ_OUT_LEN);
1263 struct efx_nic *efx = tx_queue->efx;
1267 MCDI_SET_DWORD(inbuf, FINI_TXQ_IN_INSTANCE,
1270 rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FINI_TXQ, inbuf, sizeof(inbuf),
1271 outbuf, sizeof(outbuf), &outlen);
1273 if (rc && rc != -EALREADY)
1279 efx_mcdi_display_error(efx, MC_CMD_FINI_TXQ, MC_CMD_FINI_TXQ_IN_LEN,
1280 outbuf, outlen, rc);
1283 static void efx_ef10_tx_remove(struct efx_tx_queue *tx_queue)
1285 efx_nic_free_buffer(tx_queue->efx, &tx_queue->txd.buf);
1288 /* This writes to the TX_DESC_WPTR; write pointer for TX descriptor ring */
1289 static inline void efx_ef10_notify_tx_desc(struct efx_tx_queue *tx_queue)
1291 unsigned int write_ptr;
1294 write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
1295 EFX_POPULATE_DWORD_1(reg, ERF_DZ_TX_DESC_WPTR_DWORD, write_ptr);
1296 efx_writed_page(tx_queue->efx, ®,
1297 ER_DZ_TX_DESC_UPD_DWORD, tx_queue->queue);
1300 static void efx_ef10_tx_write(struct efx_tx_queue *tx_queue)
1302 unsigned int old_write_count = tx_queue->write_count;
1303 struct efx_tx_buffer *buffer;
1304 unsigned int write_ptr;
1307 BUG_ON(tx_queue->write_count == tx_queue->insert_count);
1310 write_ptr = tx_queue->write_count & tx_queue->ptr_mask;
1311 buffer = &tx_queue->buffer[write_ptr];
1312 txd = efx_tx_desc(tx_queue, write_ptr);
1313 ++tx_queue->write_count;
1315 /* Create TX descriptor ring entry */
1316 if (buffer->flags & EFX_TX_BUF_OPTION) {
1317 *txd = buffer->option;
1319 BUILD_BUG_ON(EFX_TX_BUF_CONT != 1);
1320 EFX_POPULATE_QWORD_3(
1323 buffer->flags & EFX_TX_BUF_CONT,
1324 ESF_DZ_TX_KER_BYTE_CNT, buffer->len,
1325 ESF_DZ_TX_KER_BUF_ADDR, buffer->dma_addr);
1327 } while (tx_queue->write_count != tx_queue->insert_count);
1329 wmb(); /* Ensure descriptors are written before they are fetched */
1331 if (efx_nic_may_push_tx_desc(tx_queue, old_write_count)) {
1332 txd = efx_tx_desc(tx_queue,
1333 old_write_count & tx_queue->ptr_mask);
1334 efx_ef10_push_tx_desc(tx_queue, txd);
1337 efx_ef10_notify_tx_desc(tx_queue);
1341 static int efx_ef10_alloc_rss_context(struct efx_nic *efx, u32 *context)
1343 MCDI_DECLARE_BUF(inbuf, MC_CMD_RSS_CONTEXT_ALLOC_IN_LEN);
1344 MCDI_DECLARE_BUF(outbuf, MC_CMD_RSS_CONTEXT_ALLOC_OUT_LEN);
1348 MCDI_SET_DWORD(inbuf, RSS_CONTEXT_ALLOC_IN_UPSTREAM_PORT_ID,
1349 EVB_PORT_ID_ASSIGNED);
1350 MCDI_SET_DWORD(inbuf, RSS_CONTEXT_ALLOC_IN_TYPE,
1351 MC_CMD_RSS_CONTEXT_ALLOC_IN_TYPE_EXCLUSIVE);
1352 MCDI_SET_DWORD(inbuf, RSS_CONTEXT_ALLOC_IN_NUM_QUEUES,
1355 rc = efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_ALLOC, inbuf, sizeof(inbuf),
1356 outbuf, sizeof(outbuf), &outlen);
1360 if (outlen < MC_CMD_RSS_CONTEXT_ALLOC_OUT_LEN)
1363 *context = MCDI_DWORD(outbuf, RSS_CONTEXT_ALLOC_OUT_RSS_CONTEXT_ID);
1368 static void efx_ef10_free_rss_context(struct efx_nic *efx, u32 context)
1370 MCDI_DECLARE_BUF(inbuf, MC_CMD_RSS_CONTEXT_FREE_IN_LEN);
1373 MCDI_SET_DWORD(inbuf, RSS_CONTEXT_FREE_IN_RSS_CONTEXT_ID,
1376 rc = efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_FREE, inbuf, sizeof(inbuf),
1381 static int efx_ef10_populate_rss_table(struct efx_nic *efx, u32 context)
1383 MCDI_DECLARE_BUF(tablebuf, MC_CMD_RSS_CONTEXT_SET_TABLE_IN_LEN);
1384 MCDI_DECLARE_BUF(keybuf, MC_CMD_RSS_CONTEXT_SET_KEY_IN_LEN);
1387 MCDI_SET_DWORD(tablebuf, RSS_CONTEXT_SET_TABLE_IN_RSS_CONTEXT_ID,
1389 BUILD_BUG_ON(ARRAY_SIZE(efx->rx_indir_table) !=
1390 MC_CMD_RSS_CONTEXT_SET_TABLE_IN_INDIRECTION_TABLE_LEN);
1392 for (i = 0; i < ARRAY_SIZE(efx->rx_indir_table); ++i)
1394 RSS_CONTEXT_SET_TABLE_IN_INDIRECTION_TABLE)[i] =
1395 (u8) efx->rx_indir_table[i];
1397 rc = efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_SET_TABLE, tablebuf,
1398 sizeof(tablebuf), NULL, 0, NULL);
1402 MCDI_SET_DWORD(keybuf, RSS_CONTEXT_SET_KEY_IN_RSS_CONTEXT_ID,
1404 BUILD_BUG_ON(ARRAY_SIZE(efx->rx_hash_key) !=
1405 MC_CMD_RSS_CONTEXT_SET_KEY_IN_TOEPLITZ_KEY_LEN);
1406 for (i = 0; i < ARRAY_SIZE(efx->rx_hash_key); ++i)
1407 MCDI_PTR(keybuf, RSS_CONTEXT_SET_KEY_IN_TOEPLITZ_KEY)[i] =
1408 efx->rx_hash_key[i];
1410 return efx_mcdi_rpc(efx, MC_CMD_RSS_CONTEXT_SET_KEY, keybuf,
1411 sizeof(keybuf), NULL, 0, NULL);
1414 static void efx_ef10_rx_free_indir_table(struct efx_nic *efx)
1416 struct efx_ef10_nic_data *nic_data = efx->nic_data;
1418 if (nic_data->rx_rss_context != EFX_EF10_RSS_CONTEXT_INVALID)
1419 efx_ef10_free_rss_context(efx, nic_data->rx_rss_context);
1420 nic_data->rx_rss_context = EFX_EF10_RSS_CONTEXT_INVALID;
1423 static void efx_ef10_rx_push_rss_config(struct efx_nic *efx)
1425 struct efx_ef10_nic_data *nic_data = efx->nic_data;
1428 netif_dbg(efx, drv, efx->net_dev, "pushing RSS config\n");
1430 if (nic_data->rx_rss_context == EFX_EF10_RSS_CONTEXT_INVALID) {
1431 rc = efx_ef10_alloc_rss_context(efx, &nic_data->rx_rss_context);
1436 rc = efx_ef10_populate_rss_table(efx, nic_data->rx_rss_context);
1443 netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
1446 static int efx_ef10_rx_probe(struct efx_rx_queue *rx_queue)
1448 return efx_nic_alloc_buffer(rx_queue->efx, &rx_queue->rxd.buf,
1449 (rx_queue->ptr_mask + 1) *
1450 sizeof(efx_qword_t),
1454 static void efx_ef10_rx_init(struct efx_rx_queue *rx_queue)
1456 MCDI_DECLARE_BUF(inbuf,
1457 MC_CMD_INIT_RXQ_IN_LEN(EFX_MAX_DMAQ_SIZE * 8 /
1459 MCDI_DECLARE_BUF(outbuf, MC_CMD_INIT_RXQ_OUT_LEN);
1460 struct efx_channel *channel = efx_rx_queue_channel(rx_queue);
1461 size_t entries = rx_queue->rxd.buf.len / EFX_BUF_SIZE;
1462 struct efx_nic *efx = rx_queue->efx;
1463 size_t inlen, outlen;
1464 dma_addr_t dma_addr;
1468 rx_queue->scatter_n = 0;
1469 rx_queue->scatter_len = 0;
1471 MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_SIZE, rx_queue->ptr_mask + 1);
1472 MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_TARGET_EVQ, channel->channel);
1473 MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_LABEL, efx_rx_queue_index(rx_queue));
1474 MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_INSTANCE,
1475 efx_rx_queue_index(rx_queue));
1476 MCDI_POPULATE_DWORD_2(inbuf, INIT_RXQ_IN_FLAGS,
1477 INIT_RXQ_IN_FLAG_PREFIX, 1,
1478 INIT_RXQ_IN_FLAG_TIMESTAMP, 1);
1479 MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_OWNER_ID, 0);
1480 MCDI_SET_DWORD(inbuf, INIT_RXQ_IN_PORT_ID, EVB_PORT_ID_ASSIGNED);
1482 dma_addr = rx_queue->rxd.buf.dma_addr;
1484 netif_dbg(efx, hw, efx->net_dev, "pushing RXQ %d. %zu entries (%llx)\n",
1485 efx_rx_queue_index(rx_queue), entries, (u64)dma_addr);
1487 for (i = 0; i < entries; ++i) {
1488 MCDI_SET_ARRAY_QWORD(inbuf, INIT_RXQ_IN_DMA_ADDR, i, dma_addr);
1489 dma_addr += EFX_BUF_SIZE;
1492 inlen = MC_CMD_INIT_RXQ_IN_LEN(entries);
1494 rc = efx_mcdi_rpc(efx, MC_CMD_INIT_RXQ, inbuf, inlen,
1495 outbuf, sizeof(outbuf), &outlen);
1497 netdev_WARN(efx->net_dev, "failed to initialise RXQ %d\n",
1498 efx_rx_queue_index(rx_queue));
1501 static void efx_ef10_rx_fini(struct efx_rx_queue *rx_queue)
1503 MCDI_DECLARE_BUF(inbuf, MC_CMD_FINI_RXQ_IN_LEN);
1504 MCDI_DECLARE_BUF(outbuf, MC_CMD_FINI_RXQ_OUT_LEN);
1505 struct efx_nic *efx = rx_queue->efx;
1509 MCDI_SET_DWORD(inbuf, FINI_RXQ_IN_INSTANCE,
1510 efx_rx_queue_index(rx_queue));
1512 rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FINI_RXQ, inbuf, sizeof(inbuf),
1513 outbuf, sizeof(outbuf), &outlen);
1515 if (rc && rc != -EALREADY)
1521 efx_mcdi_display_error(efx, MC_CMD_FINI_RXQ, MC_CMD_FINI_RXQ_IN_LEN,
1522 outbuf, outlen, rc);
1525 static void efx_ef10_rx_remove(struct efx_rx_queue *rx_queue)
1527 efx_nic_free_buffer(rx_queue->efx, &rx_queue->rxd.buf);
1530 /* This creates an entry in the RX descriptor queue */
1532 efx_ef10_build_rx_desc(struct efx_rx_queue *rx_queue, unsigned int index)
1534 struct efx_rx_buffer *rx_buf;
1537 rxd = efx_rx_desc(rx_queue, index);
1538 rx_buf = efx_rx_buffer(rx_queue, index);
1539 EFX_POPULATE_QWORD_2(*rxd,
1540 ESF_DZ_RX_KER_BYTE_CNT, rx_buf->len,
1541 ESF_DZ_RX_KER_BUF_ADDR, rx_buf->dma_addr);
1544 static void efx_ef10_rx_write(struct efx_rx_queue *rx_queue)
1546 struct efx_nic *efx = rx_queue->efx;
1547 unsigned int write_count;
1550 /* Firmware requires that RX_DESC_WPTR be a multiple of 8 */
1551 write_count = rx_queue->added_count & ~7;
1552 if (rx_queue->notified_count == write_count)
1556 efx_ef10_build_rx_desc(
1558 rx_queue->notified_count & rx_queue->ptr_mask);
1559 while (++rx_queue->notified_count != write_count);
1562 EFX_POPULATE_DWORD_1(reg, ERF_DZ_RX_DESC_WPTR,
1563 write_count & rx_queue->ptr_mask);
1564 efx_writed_page(efx, ®, ER_DZ_RX_DESC_UPD,
1565 efx_rx_queue_index(rx_queue));
1568 static efx_mcdi_async_completer efx_ef10_rx_defer_refill_complete;
1570 static void efx_ef10_rx_defer_refill(struct efx_rx_queue *rx_queue)
1572 struct efx_channel *channel = efx_rx_queue_channel(rx_queue);
1573 MCDI_DECLARE_BUF(inbuf, MC_CMD_DRIVER_EVENT_IN_LEN);
1576 EFX_POPULATE_QWORD_2(event,
1577 ESF_DZ_EV_CODE, EFX_EF10_DRVGEN_EV,
1578 ESF_DZ_EV_DATA, EFX_EF10_REFILL);
1580 MCDI_SET_DWORD(inbuf, DRIVER_EVENT_IN_EVQ, channel->channel);
1582 /* MCDI_SET_QWORD is not appropriate here since EFX_POPULATE_* has
1583 * already swapped the data to little-endian order.
1585 memcpy(MCDI_PTR(inbuf, DRIVER_EVENT_IN_DATA), &event.u64[0],
1586 sizeof(efx_qword_t));
1588 efx_mcdi_rpc_async(channel->efx, MC_CMD_DRIVER_EVENT,
1589 inbuf, sizeof(inbuf), 0,
1590 efx_ef10_rx_defer_refill_complete, 0);
1594 efx_ef10_rx_defer_refill_complete(struct efx_nic *efx, unsigned long cookie,
1595 int rc, efx_dword_t *outbuf,
1596 size_t outlen_actual)
1601 static int efx_ef10_ev_probe(struct efx_channel *channel)
1603 return efx_nic_alloc_buffer(channel->efx, &channel->eventq.buf,
1604 (channel->eventq_mask + 1) *
1605 sizeof(efx_qword_t),
1609 static int efx_ef10_ev_init(struct efx_channel *channel)
1611 MCDI_DECLARE_BUF(inbuf,
1612 MC_CMD_INIT_EVQ_IN_LEN(EFX_MAX_EVQ_SIZE * 8 /
1614 MCDI_DECLARE_BUF(outbuf, MC_CMD_INIT_EVQ_OUT_LEN);
1615 size_t entries = channel->eventq.buf.len / EFX_BUF_SIZE;
1616 struct efx_nic *efx = channel->efx;
1617 struct efx_ef10_nic_data *nic_data;
1618 bool supports_rx_merge;
1619 size_t inlen, outlen;
1620 dma_addr_t dma_addr;
1624 nic_data = efx->nic_data;
1626 !!(nic_data->datapath_caps &
1627 1 << MC_CMD_GET_CAPABILITIES_OUT_RX_BATCHING_LBN);
1629 /* Fill event queue with all ones (i.e. empty events) */
1630 memset(channel->eventq.buf.addr, 0xff, channel->eventq.buf.len);
1632 MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_SIZE, channel->eventq_mask + 1);
1633 MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_INSTANCE, channel->channel);
1634 /* INIT_EVQ expects index in vector table, not absolute */
1635 MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_IRQ_NUM, channel->channel);
1636 MCDI_POPULATE_DWORD_4(inbuf, INIT_EVQ_IN_FLAGS,
1637 INIT_EVQ_IN_FLAG_INTERRUPTING, 1,
1638 INIT_EVQ_IN_FLAG_RX_MERGE, 1,
1639 INIT_EVQ_IN_FLAG_TX_MERGE, 1,
1640 INIT_EVQ_IN_FLAG_CUT_THRU, !supports_rx_merge);
1641 MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_TMR_MODE,
1642 MC_CMD_INIT_EVQ_IN_TMR_MODE_DIS);
1643 MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_TMR_LOAD, 0);
1644 MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_TMR_RELOAD, 0);
1645 MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_COUNT_MODE,
1646 MC_CMD_INIT_EVQ_IN_COUNT_MODE_DIS);
1647 MCDI_SET_DWORD(inbuf, INIT_EVQ_IN_COUNT_THRSHLD, 0);
1649 dma_addr = channel->eventq.buf.dma_addr;
1650 for (i = 0; i < entries; ++i) {
1651 MCDI_SET_ARRAY_QWORD(inbuf, INIT_EVQ_IN_DMA_ADDR, i, dma_addr);
1652 dma_addr += EFX_BUF_SIZE;
1655 inlen = MC_CMD_INIT_EVQ_IN_LEN(entries);
1657 rc = efx_mcdi_rpc(efx, MC_CMD_INIT_EVQ, inbuf, inlen,
1658 outbuf, sizeof(outbuf), &outlen);
1659 /* IRQ return is ignored */
1663 static void efx_ef10_ev_fini(struct efx_channel *channel)
1665 MCDI_DECLARE_BUF(inbuf, MC_CMD_FINI_EVQ_IN_LEN);
1666 MCDI_DECLARE_BUF(outbuf, MC_CMD_FINI_EVQ_OUT_LEN);
1667 struct efx_nic *efx = channel->efx;
1671 MCDI_SET_DWORD(inbuf, FINI_EVQ_IN_INSTANCE, channel->channel);
1673 rc = efx_mcdi_rpc_quiet(efx, MC_CMD_FINI_EVQ, inbuf, sizeof(inbuf),
1674 outbuf, sizeof(outbuf), &outlen);
1676 if (rc && rc != -EALREADY)
1682 efx_mcdi_display_error(efx, MC_CMD_FINI_EVQ, MC_CMD_FINI_EVQ_IN_LEN,
1683 outbuf, outlen, rc);
1686 static void efx_ef10_ev_remove(struct efx_channel *channel)
1688 efx_nic_free_buffer(channel->efx, &channel->eventq.buf);
1691 static void efx_ef10_handle_rx_wrong_queue(struct efx_rx_queue *rx_queue,
1692 unsigned int rx_queue_label)
1694 struct efx_nic *efx = rx_queue->efx;
1696 netif_info(efx, hw, efx->net_dev,
1697 "rx event arrived on queue %d labeled as queue %u\n",
1698 efx_rx_queue_index(rx_queue), rx_queue_label);
1700 efx_schedule_reset(efx, RESET_TYPE_DISABLE);
1704 efx_ef10_handle_rx_bad_lbits(struct efx_rx_queue *rx_queue,
1705 unsigned int actual, unsigned int expected)
1707 unsigned int dropped = (actual - expected) & rx_queue->ptr_mask;
1708 struct efx_nic *efx = rx_queue->efx;
1710 netif_info(efx, hw, efx->net_dev,
1711 "dropped %d events (index=%d expected=%d)\n",
1712 dropped, actual, expected);
1714 efx_schedule_reset(efx, RESET_TYPE_DISABLE);
1717 /* partially received RX was aborted. clean up. */
1718 static void efx_ef10_handle_rx_abort(struct efx_rx_queue *rx_queue)
1720 unsigned int rx_desc_ptr;
1722 netif_dbg(rx_queue->efx, hw, rx_queue->efx->net_dev,
1723 "scattered RX aborted (dropping %u buffers)\n",
1724 rx_queue->scatter_n);
1726 rx_desc_ptr = rx_queue->removed_count & rx_queue->ptr_mask;
1728 efx_rx_packet(rx_queue, rx_desc_ptr, rx_queue->scatter_n,
1729 0, EFX_RX_PKT_DISCARD);
1731 rx_queue->removed_count += rx_queue->scatter_n;
1732 rx_queue->scatter_n = 0;
1733 rx_queue->scatter_len = 0;
1734 ++efx_rx_queue_channel(rx_queue)->n_rx_nodesc_trunc;
1737 static int efx_ef10_handle_rx_event(struct efx_channel *channel,
1738 const efx_qword_t *event)
1740 unsigned int rx_bytes, next_ptr_lbits, rx_queue_label, rx_l4_class;
1741 unsigned int n_descs, n_packets, i;
1742 struct efx_nic *efx = channel->efx;
1743 struct efx_rx_queue *rx_queue;
1747 if (unlikely(ACCESS_ONCE(efx->reset_pending)))
1750 /* Basic packet information */
1751 rx_bytes = EFX_QWORD_FIELD(*event, ESF_DZ_RX_BYTES);
1752 next_ptr_lbits = EFX_QWORD_FIELD(*event, ESF_DZ_RX_DSC_PTR_LBITS);
1753 rx_queue_label = EFX_QWORD_FIELD(*event, ESF_DZ_RX_QLABEL);
1754 rx_l4_class = EFX_QWORD_FIELD(*event, ESF_DZ_RX_L4_CLASS);
1755 rx_cont = EFX_QWORD_FIELD(*event, ESF_DZ_RX_CONT);
1757 if (EFX_QWORD_FIELD(*event, ESF_DZ_RX_DROP_EVENT))
1758 netdev_WARN(efx->net_dev, "saw RX_DROP_EVENT: event="
1760 EFX_QWORD_VAL(*event));
1762 rx_queue = efx_channel_get_rx_queue(channel);
1764 if (unlikely(rx_queue_label != efx_rx_queue_index(rx_queue)))
1765 efx_ef10_handle_rx_wrong_queue(rx_queue, rx_queue_label);
1767 n_descs = ((next_ptr_lbits - rx_queue->removed_count) &
1768 ((1 << ESF_DZ_RX_DSC_PTR_LBITS_WIDTH) - 1));
1770 if (n_descs != rx_queue->scatter_n + 1) {
1771 struct efx_ef10_nic_data *nic_data = efx->nic_data;
1773 /* detect rx abort */
1774 if (unlikely(n_descs == rx_queue->scatter_n)) {
1775 if (rx_queue->scatter_n == 0 || rx_bytes != 0)
1776 netdev_WARN(efx->net_dev,
1777 "invalid RX abort: scatter_n=%u event="
1779 rx_queue->scatter_n,
1780 EFX_QWORD_VAL(*event));
1781 efx_ef10_handle_rx_abort(rx_queue);
1785 /* Check that RX completion merging is valid, i.e.
1786 * the current firmware supports it and this is a
1787 * non-scattered packet.
1789 if (!(nic_data->datapath_caps &
1790 (1 << MC_CMD_GET_CAPABILITIES_OUT_RX_BATCHING_LBN)) ||
1791 rx_queue->scatter_n != 0 || rx_cont) {
1792 efx_ef10_handle_rx_bad_lbits(
1793 rx_queue, next_ptr_lbits,
1794 (rx_queue->removed_count +
1795 rx_queue->scatter_n + 1) &
1796 ((1 << ESF_DZ_RX_DSC_PTR_LBITS_WIDTH) - 1));
1800 /* Merged completion for multiple non-scattered packets */
1801 rx_queue->scatter_n = 1;
1802 rx_queue->scatter_len = 0;
1803 n_packets = n_descs;
1804 ++channel->n_rx_merge_events;
1805 channel->n_rx_merge_packets += n_packets;
1806 flags |= EFX_RX_PKT_PREFIX_LEN;
1808 ++rx_queue->scatter_n;
1809 rx_queue->scatter_len += rx_bytes;
1815 if (unlikely(EFX_QWORD_FIELD(*event, ESF_DZ_RX_ECRC_ERR)))
1816 flags |= EFX_RX_PKT_DISCARD;
1818 if (unlikely(EFX_QWORD_FIELD(*event, ESF_DZ_RX_IPCKSUM_ERR))) {
1819 channel->n_rx_ip_hdr_chksum_err += n_packets;
1820 } else if (unlikely(EFX_QWORD_FIELD(*event,
1821 ESF_DZ_RX_TCPUDP_CKSUM_ERR))) {
1822 channel->n_rx_tcp_udp_chksum_err += n_packets;
1823 } else if (rx_l4_class == ESE_DZ_L4_CLASS_TCP ||
1824 rx_l4_class == ESE_DZ_L4_CLASS_UDP) {
1825 flags |= EFX_RX_PKT_CSUMMED;
1828 if (rx_l4_class == ESE_DZ_L4_CLASS_TCP)
1829 flags |= EFX_RX_PKT_TCP;
1831 channel->irq_mod_score += 2 * n_packets;
1833 /* Handle received packet(s) */
1834 for (i = 0; i < n_packets; i++) {
1835 efx_rx_packet(rx_queue,
1836 rx_queue->removed_count & rx_queue->ptr_mask,
1837 rx_queue->scatter_n, rx_queue->scatter_len,
1839 rx_queue->removed_count += rx_queue->scatter_n;
1842 rx_queue->scatter_n = 0;
1843 rx_queue->scatter_len = 0;
1849 efx_ef10_handle_tx_event(struct efx_channel *channel, efx_qword_t *event)
1851 struct efx_nic *efx = channel->efx;
1852 struct efx_tx_queue *tx_queue;
1853 unsigned int tx_ev_desc_ptr;
1854 unsigned int tx_ev_q_label;
1857 if (unlikely(ACCESS_ONCE(efx->reset_pending)))
1860 if (unlikely(EFX_QWORD_FIELD(*event, ESF_DZ_TX_DROP_EVENT)))
1863 /* Transmit completion */
1864 tx_ev_desc_ptr = EFX_QWORD_FIELD(*event, ESF_DZ_TX_DESCR_INDX);
1865 tx_ev_q_label = EFX_QWORD_FIELD(*event, ESF_DZ_TX_QLABEL);
1866 tx_queue = efx_channel_get_tx_queue(channel,
1867 tx_ev_q_label % EFX_TXQ_TYPES);
1868 tx_descs = ((tx_ev_desc_ptr + 1 - tx_queue->read_count) &
1869 tx_queue->ptr_mask);
1870 efx_xmit_done(tx_queue, tx_ev_desc_ptr & tx_queue->ptr_mask);
1876 efx_ef10_handle_driver_event(struct efx_channel *channel, efx_qword_t *event)
1878 struct efx_nic *efx = channel->efx;
1881 subcode = EFX_QWORD_FIELD(*event, ESF_DZ_DRV_SUB_CODE);
1884 case ESE_DZ_DRV_TIMER_EV:
1885 case ESE_DZ_DRV_WAKE_UP_EV:
1887 case ESE_DZ_DRV_START_UP_EV:
1888 /* event queue init complete. ok. */
1891 netif_err(efx, hw, efx->net_dev,
1892 "channel %d unknown driver event type %d"
1893 " (data " EFX_QWORD_FMT ")\n",
1894 channel->channel, subcode,
1895 EFX_QWORD_VAL(*event));
1900 static void efx_ef10_handle_driver_generated_event(struct efx_channel *channel,
1903 struct efx_nic *efx = channel->efx;
1906 subcode = EFX_QWORD_FIELD(*event, EFX_DWORD_0);
1910 channel->event_test_cpu = raw_smp_processor_id();
1912 case EFX_EF10_REFILL:
1913 /* The queue must be empty, so we won't receive any rx
1914 * events, so efx_process_channel() won't refill the
1915 * queue. Refill it here
1917 efx_fast_push_rx_descriptors(&channel->rx_queue, true);
1920 netif_err(efx, hw, efx->net_dev,
1921 "channel %d unknown driver event type %u"
1922 " (data " EFX_QWORD_FMT ")\n",
1923 channel->channel, (unsigned) subcode,
1924 EFX_QWORD_VAL(*event));
1928 static int efx_ef10_ev_process(struct efx_channel *channel, int quota)
1930 struct efx_nic *efx = channel->efx;
1931 efx_qword_t event, *p_event;
1932 unsigned int read_ptr;
1937 read_ptr = channel->eventq_read_ptr;
1940 p_event = efx_event(channel, read_ptr);
1943 if (!efx_event_present(&event))
1946 EFX_SET_QWORD(*p_event);
1950 ev_code = EFX_QWORD_FIELD(event, ESF_DZ_EV_CODE);
1952 netif_vdbg(efx, drv, efx->net_dev,
1953 "processing event on %d " EFX_QWORD_FMT "\n",
1954 channel->channel, EFX_QWORD_VAL(event));
1957 case ESE_DZ_EV_CODE_MCDI_EV:
1958 efx_mcdi_process_event(channel, &event);
1960 case ESE_DZ_EV_CODE_RX_EV:
1961 spent += efx_ef10_handle_rx_event(channel, &event);
1962 if (spent >= quota) {
1963 /* XXX can we split a merged event to
1964 * avoid going over-quota?
1970 case ESE_DZ_EV_CODE_TX_EV:
1971 tx_descs += efx_ef10_handle_tx_event(channel, &event);
1972 if (tx_descs > efx->txq_entries) {
1975 } else if (++spent == quota) {
1979 case ESE_DZ_EV_CODE_DRIVER_EV:
1980 efx_ef10_handle_driver_event(channel, &event);
1981 if (++spent == quota)
1984 case EFX_EF10_DRVGEN_EV:
1985 efx_ef10_handle_driver_generated_event(channel, &event);
1988 netif_err(efx, hw, efx->net_dev,
1989 "channel %d unknown event type %d"
1990 " (data " EFX_QWORD_FMT ")\n",
1991 channel->channel, ev_code,
1992 EFX_QWORD_VAL(event));
1997 channel->eventq_read_ptr = read_ptr;
2001 static void efx_ef10_ev_read_ack(struct efx_channel *channel)
2003 struct efx_nic *efx = channel->efx;
2006 if (EFX_EF10_WORKAROUND_35388(efx)) {
2007 BUILD_BUG_ON(EFX_MIN_EVQ_SIZE <
2008 (1 << ERF_DD_EVQ_IND_RPTR_WIDTH));
2009 BUILD_BUG_ON(EFX_MAX_EVQ_SIZE >
2010 (1 << 2 * ERF_DD_EVQ_IND_RPTR_WIDTH));
2012 EFX_POPULATE_DWORD_2(rptr, ERF_DD_EVQ_IND_RPTR_FLAGS,
2013 EFE_DD_EVQ_IND_RPTR_FLAGS_HIGH,
2014 ERF_DD_EVQ_IND_RPTR,
2015 (channel->eventq_read_ptr &
2016 channel->eventq_mask) >>
2017 ERF_DD_EVQ_IND_RPTR_WIDTH);
2018 efx_writed_page(efx, &rptr, ER_DD_EVQ_INDIRECT,
2020 EFX_POPULATE_DWORD_2(rptr, ERF_DD_EVQ_IND_RPTR_FLAGS,
2021 EFE_DD_EVQ_IND_RPTR_FLAGS_LOW,
2022 ERF_DD_EVQ_IND_RPTR,
2023 channel->eventq_read_ptr &
2024 ((1 << ERF_DD_EVQ_IND_RPTR_WIDTH) - 1));
2025 efx_writed_page(efx, &rptr, ER_DD_EVQ_INDIRECT,
2028 EFX_POPULATE_DWORD_1(rptr, ERF_DZ_EVQ_RPTR,
2029 channel->eventq_read_ptr &
2030 channel->eventq_mask);
2031 efx_writed_page(efx, &rptr, ER_DZ_EVQ_RPTR, channel->channel);
2035 static void efx_ef10_ev_test_generate(struct efx_channel *channel)
2037 MCDI_DECLARE_BUF(inbuf, MC_CMD_DRIVER_EVENT_IN_LEN);
2038 struct efx_nic *efx = channel->efx;
2042 EFX_POPULATE_QWORD_2(event,
2043 ESF_DZ_EV_CODE, EFX_EF10_DRVGEN_EV,
2044 ESF_DZ_EV_DATA, EFX_EF10_TEST);
2046 MCDI_SET_DWORD(inbuf, DRIVER_EVENT_IN_EVQ, channel->channel);
2048 /* MCDI_SET_QWORD is not appropriate here since EFX_POPULATE_* has
2049 * already swapped the data to little-endian order.
2051 memcpy(MCDI_PTR(inbuf, DRIVER_EVENT_IN_DATA), &event.u64[0],
2052 sizeof(efx_qword_t));
2054 rc = efx_mcdi_rpc(efx, MC_CMD_DRIVER_EVENT, inbuf, sizeof(inbuf),
2063 netif_err(efx, hw, efx->net_dev, "%s: failed rc=%d\n", __func__, rc);
2066 void efx_ef10_handle_drain_event(struct efx_nic *efx)
2068 if (atomic_dec_and_test(&efx->active_queues))
2069 wake_up(&efx->flush_wq);
2071 WARN_ON(atomic_read(&efx->active_queues) < 0);
2074 static int efx_ef10_fini_dmaq(struct efx_nic *efx)
2076 struct efx_ef10_nic_data *nic_data = efx->nic_data;
2077 struct efx_channel *channel;
2078 struct efx_tx_queue *tx_queue;
2079 struct efx_rx_queue *rx_queue;
2082 /* If the MC has just rebooted, the TX/RX queues will have already been
2083 * torn down, but efx->active_queues needs to be set to zero.
2085 if (nic_data->must_realloc_vis) {
2086 atomic_set(&efx->active_queues, 0);
2090 /* Do not attempt to write to the NIC during EEH recovery */
2091 if (efx->state != STATE_RECOVERY) {
2092 efx_for_each_channel(channel, efx) {
2093 efx_for_each_channel_rx_queue(rx_queue, channel)
2094 efx_ef10_rx_fini(rx_queue);
2095 efx_for_each_channel_tx_queue(tx_queue, channel)
2096 efx_ef10_tx_fini(tx_queue);
2099 wait_event_timeout(efx->flush_wq,
2100 atomic_read(&efx->active_queues) == 0,
2101 msecs_to_jiffies(EFX_MAX_FLUSH_TIME));
2102 pending = atomic_read(&efx->active_queues);
2104 netif_err(efx, hw, efx->net_dev, "failed to flush %d queues\n",
2113 static bool efx_ef10_filter_equal(const struct efx_filter_spec *left,
2114 const struct efx_filter_spec *right)
2116 if ((left->match_flags ^ right->match_flags) |
2117 ((left->flags ^ right->flags) &
2118 (EFX_FILTER_FLAG_RX | EFX_FILTER_FLAG_TX)))
2121 return memcmp(&left->outer_vid, &right->outer_vid,
2122 sizeof(struct efx_filter_spec) -
2123 offsetof(struct efx_filter_spec, outer_vid)) == 0;
2126 static unsigned int efx_ef10_filter_hash(const struct efx_filter_spec *spec)
2128 BUILD_BUG_ON(offsetof(struct efx_filter_spec, outer_vid) & 3);
2129 return jhash2((const u32 *)&spec->outer_vid,
2130 (sizeof(struct efx_filter_spec) -
2131 offsetof(struct efx_filter_spec, outer_vid)) / 4,
2133 /* XXX should we randomise the initval? */
2136 /* Decide whether a filter should be exclusive or else should allow
2137 * delivery to additional recipients. Currently we decide that
2138 * filters for specific local unicast MAC and IP addresses are
2141 static bool efx_ef10_filter_is_exclusive(const struct efx_filter_spec *spec)
2143 if (spec->match_flags & EFX_FILTER_MATCH_LOC_MAC &&
2144 !is_multicast_ether_addr(spec->loc_mac))
2147 if ((spec->match_flags &
2148 (EFX_FILTER_MATCH_ETHER_TYPE | EFX_FILTER_MATCH_LOC_HOST)) ==
2149 (EFX_FILTER_MATCH_ETHER_TYPE | EFX_FILTER_MATCH_LOC_HOST)) {
2150 if (spec->ether_type == htons(ETH_P_IP) &&
2151 !ipv4_is_multicast(spec->loc_host[0]))
2153 if (spec->ether_type == htons(ETH_P_IPV6) &&
2154 ((const u8 *)spec->loc_host)[0] != 0xff)
2161 static struct efx_filter_spec *
2162 efx_ef10_filter_entry_spec(const struct efx_ef10_filter_table *table,
2163 unsigned int filter_idx)
2165 return (struct efx_filter_spec *)(table->entry[filter_idx].spec &
2166 ~EFX_EF10_FILTER_FLAGS);
2170 efx_ef10_filter_entry_flags(const struct efx_ef10_filter_table *table,
2171 unsigned int filter_idx)
2173 return table->entry[filter_idx].spec & EFX_EF10_FILTER_FLAGS;
2177 efx_ef10_filter_set_entry(struct efx_ef10_filter_table *table,
2178 unsigned int filter_idx,
2179 const struct efx_filter_spec *spec,
2182 table->entry[filter_idx].spec = (unsigned long)spec | flags;
2185 static void efx_ef10_filter_push_prep(struct efx_nic *efx,
2186 const struct efx_filter_spec *spec,
2187 efx_dword_t *inbuf, u64 handle,
2190 struct efx_ef10_nic_data *nic_data = efx->nic_data;
2192 memset(inbuf, 0, MC_CMD_FILTER_OP_IN_LEN);
2195 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
2196 MC_CMD_FILTER_OP_IN_OP_REPLACE);
2197 MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE, handle);
2199 u32 match_fields = 0;
2201 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
2202 efx_ef10_filter_is_exclusive(spec) ?
2203 MC_CMD_FILTER_OP_IN_OP_INSERT :
2204 MC_CMD_FILTER_OP_IN_OP_SUBSCRIBE);
2206 /* Convert match flags and values. Unlike almost
2207 * everything else in MCDI, these fields are in
2208 * network byte order.
2210 if (spec->match_flags & EFX_FILTER_MATCH_LOC_MAC_IG)
2212 is_multicast_ether_addr(spec->loc_mac) ?
2213 1 << MC_CMD_FILTER_OP_IN_MATCH_UNKNOWN_MCAST_DST_LBN :
2214 1 << MC_CMD_FILTER_OP_IN_MATCH_UNKNOWN_UCAST_DST_LBN;
2215 #define COPY_FIELD(gen_flag, gen_field, mcdi_field) \
2216 if (spec->match_flags & EFX_FILTER_MATCH_ ## gen_flag) { \
2218 1 << MC_CMD_FILTER_OP_IN_MATCH_ ## \
2219 mcdi_field ## _LBN; \
2221 MC_CMD_FILTER_OP_IN_ ## mcdi_field ## _LEN < \
2222 sizeof(spec->gen_field)); \
2223 memcpy(MCDI_PTR(inbuf, FILTER_OP_IN_ ## mcdi_field), \
2224 &spec->gen_field, sizeof(spec->gen_field)); \
2226 COPY_FIELD(REM_HOST, rem_host, SRC_IP);
2227 COPY_FIELD(LOC_HOST, loc_host, DST_IP);
2228 COPY_FIELD(REM_MAC, rem_mac, SRC_MAC);
2229 COPY_FIELD(REM_PORT, rem_port, SRC_PORT);
2230 COPY_FIELD(LOC_MAC, loc_mac, DST_MAC);
2231 COPY_FIELD(LOC_PORT, loc_port, DST_PORT);
2232 COPY_FIELD(ETHER_TYPE, ether_type, ETHER_TYPE);
2233 COPY_FIELD(INNER_VID, inner_vid, INNER_VLAN);
2234 COPY_FIELD(OUTER_VID, outer_vid, OUTER_VLAN);
2235 COPY_FIELD(IP_PROTO, ip_proto, IP_PROTO);
2237 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_MATCH_FIELDS,
2241 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_PORT_ID, EVB_PORT_ID_ASSIGNED);
2242 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_DEST,
2243 spec->dmaq_id == EFX_FILTER_RX_DMAQ_ID_DROP ?
2244 MC_CMD_FILTER_OP_IN_RX_DEST_DROP :
2245 MC_CMD_FILTER_OP_IN_RX_DEST_HOST);
2246 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_TX_DEST,
2247 MC_CMD_FILTER_OP_IN_TX_DEST_DEFAULT);
2248 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_QUEUE, spec->dmaq_id);
2249 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_MODE,
2250 (spec->flags & EFX_FILTER_FLAG_RX_RSS) ?
2251 MC_CMD_FILTER_OP_IN_RX_MODE_RSS :
2252 MC_CMD_FILTER_OP_IN_RX_MODE_SIMPLE);
2253 if (spec->flags & EFX_FILTER_FLAG_RX_RSS)
2254 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_RX_CONTEXT,
2255 spec->rss_context !=
2256 EFX_FILTER_RSS_CONTEXT_DEFAULT ?
2257 spec->rss_context : nic_data->rx_rss_context);
2260 static int efx_ef10_filter_push(struct efx_nic *efx,
2261 const struct efx_filter_spec *spec,
2262 u64 *handle, bool replacing)
2264 MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_IN_LEN);
2265 MCDI_DECLARE_BUF(outbuf, MC_CMD_FILTER_OP_OUT_LEN);
2268 efx_ef10_filter_push_prep(efx, spec, inbuf, *handle, replacing);
2269 rc = efx_mcdi_rpc(efx, MC_CMD_FILTER_OP, inbuf, sizeof(inbuf),
2270 outbuf, sizeof(outbuf), NULL);
2272 *handle = MCDI_QWORD(outbuf, FILTER_OP_OUT_HANDLE);
2274 rc = -EBUSY; /* to match efx_farch_filter_insert() */
2278 static int efx_ef10_filter_rx_match_pri(struct efx_ef10_filter_table *table,
2279 enum efx_filter_match_flags match_flags)
2281 unsigned int match_pri;
2284 match_pri < table->rx_match_count;
2286 if (table->rx_match_flags[match_pri] == match_flags)
2289 return -EPROTONOSUPPORT;
2292 static s32 efx_ef10_filter_insert(struct efx_nic *efx,
2293 struct efx_filter_spec *spec,
2296 struct efx_ef10_filter_table *table = efx->filter_state;
2297 DECLARE_BITMAP(mc_rem_map, EFX_EF10_FILTER_SEARCH_LIMIT);
2298 struct efx_filter_spec *saved_spec;
2299 unsigned int match_pri, hash;
2300 unsigned int priv_flags;
2301 bool replacing = false;
2307 /* For now, only support RX filters */
2308 if ((spec->flags & (EFX_FILTER_FLAG_RX | EFX_FILTER_FLAG_TX)) !=
2312 rc = efx_ef10_filter_rx_match_pri(table, spec->match_flags);
2317 hash = efx_ef10_filter_hash(spec);
2318 is_mc_recip = efx_filter_is_mc_recipient(spec);
2320 bitmap_zero(mc_rem_map, EFX_EF10_FILTER_SEARCH_LIMIT);
2322 /* Find any existing filters with the same match tuple or
2323 * else a free slot to insert at. If any of them are busy,
2324 * we have to wait and retry.
2327 unsigned int depth = 1;
2330 spin_lock_bh(&efx->filter_lock);
2333 i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
2334 saved_spec = efx_ef10_filter_entry_spec(table, i);
2339 } else if (efx_ef10_filter_equal(spec, saved_spec)) {
2340 if (table->entry[i].spec &
2341 EFX_EF10_FILTER_FLAG_BUSY)
2343 if (spec->priority < saved_spec->priority &&
2344 spec->priority != EFX_FILTER_PRI_AUTO) {
2349 /* This is the only one */
2350 if (spec->priority ==
2351 saved_spec->priority &&
2358 } else if (spec->priority >
2359 saved_spec->priority ||
2361 saved_spec->priority &&
2366 __set_bit(depth, mc_rem_map);
2370 /* Once we reach the maximum search depth, use
2371 * the first suitable slot or return -EBUSY if
2374 if (depth == EFX_EF10_FILTER_SEARCH_LIMIT) {
2375 if (ins_index < 0) {
2385 prepare_to_wait(&table->waitq, &wait, TASK_UNINTERRUPTIBLE);
2386 spin_unlock_bh(&efx->filter_lock);
2391 /* Create a software table entry if necessary, and mark it
2392 * busy. We might yet fail to insert, but any attempt to
2393 * insert a conflicting filter while we're waiting for the
2394 * firmware must find the busy entry.
2396 saved_spec = efx_ef10_filter_entry_spec(table, ins_index);
2398 if (spec->priority == EFX_FILTER_PRI_AUTO &&
2399 saved_spec->priority >= EFX_FILTER_PRI_AUTO) {
2400 /* Just make sure it won't be removed */
2401 if (saved_spec->priority > EFX_FILTER_PRI_AUTO)
2402 saved_spec->flags |= EFX_FILTER_FLAG_RX_OVER_AUTO;
2403 table->entry[ins_index].spec &=
2404 ~EFX_EF10_FILTER_FLAG_STACK_OLD;
2409 priv_flags = efx_ef10_filter_entry_flags(table, ins_index);
2411 saved_spec = kmalloc(sizeof(*spec), GFP_ATOMIC);
2416 *saved_spec = *spec;
2419 efx_ef10_filter_set_entry(table, ins_index, saved_spec,
2420 priv_flags | EFX_EF10_FILTER_FLAG_BUSY);
2422 /* Mark lower-priority multicast recipients busy prior to removal */
2424 unsigned int depth, i;
2426 for (depth = 0; depth < EFX_EF10_FILTER_SEARCH_LIMIT; depth++) {
2427 i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
2428 if (test_bit(depth, mc_rem_map))
2429 table->entry[i].spec |=
2430 EFX_EF10_FILTER_FLAG_BUSY;
2434 spin_unlock_bh(&efx->filter_lock);
2436 rc = efx_ef10_filter_push(efx, spec, &table->entry[ins_index].handle,
2439 /* Finalise the software table entry */
2440 spin_lock_bh(&efx->filter_lock);
2443 /* Update the fields that may differ */
2444 if (saved_spec->priority == EFX_FILTER_PRI_AUTO)
2445 saved_spec->flags |=
2446 EFX_FILTER_FLAG_RX_OVER_AUTO;
2447 saved_spec->priority = spec->priority;
2448 saved_spec->flags &= EFX_FILTER_FLAG_RX_OVER_AUTO;
2449 saved_spec->flags |= spec->flags;
2450 saved_spec->rss_context = spec->rss_context;
2451 saved_spec->dmaq_id = spec->dmaq_id;
2453 } else if (!replacing) {
2457 efx_ef10_filter_set_entry(table, ins_index, saved_spec, priv_flags);
2459 /* Remove and finalise entries for lower-priority multicast
2463 MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_IN_LEN);
2464 unsigned int depth, i;
2466 memset(inbuf, 0, sizeof(inbuf));
2468 for (depth = 0; depth < EFX_EF10_FILTER_SEARCH_LIMIT; depth++) {
2469 if (!test_bit(depth, mc_rem_map))
2472 i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
2473 saved_spec = efx_ef10_filter_entry_spec(table, i);
2474 priv_flags = efx_ef10_filter_entry_flags(table, i);
2477 spin_unlock_bh(&efx->filter_lock);
2478 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
2479 MC_CMD_FILTER_OP_IN_OP_UNSUBSCRIBE);
2480 MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
2481 table->entry[i].handle);
2482 rc = efx_mcdi_rpc(efx, MC_CMD_FILTER_OP,
2483 inbuf, sizeof(inbuf),
2485 spin_lock_bh(&efx->filter_lock);
2493 priv_flags &= ~EFX_EF10_FILTER_FLAG_BUSY;
2495 efx_ef10_filter_set_entry(table, i, saved_spec,
2500 /* If successful, return the inserted filter ID */
2502 rc = match_pri * HUNT_FILTER_TBL_ROWS + ins_index;
2504 wake_up_all(&table->waitq);
2506 spin_unlock_bh(&efx->filter_lock);
2507 finish_wait(&table->waitq, &wait);
2511 static void efx_ef10_filter_update_rx_scatter(struct efx_nic *efx)
2513 /* no need to do anything here on EF10 */
2517 * If !stack_requested, remove by ID
2518 * If stack_requested, remove by index
2519 * Filter ID may come from userland and must be range-checked.
2521 static int efx_ef10_filter_remove_internal(struct efx_nic *efx,
2522 enum efx_filter_priority priority,
2523 u32 filter_id, bool stack_requested)
2525 unsigned int filter_idx = filter_id % HUNT_FILTER_TBL_ROWS;
2526 struct efx_ef10_filter_table *table = efx->filter_state;
2527 MCDI_DECLARE_BUF(inbuf,
2528 MC_CMD_FILTER_OP_IN_HANDLE_OFST +
2529 MC_CMD_FILTER_OP_IN_HANDLE_LEN);
2530 struct efx_filter_spec *spec;
2534 /* Find the software table entry and mark it busy. Don't
2535 * remove it yet; any attempt to update while we're waiting
2536 * for the firmware must find the busy entry.
2539 spin_lock_bh(&efx->filter_lock);
2540 if (!(table->entry[filter_idx].spec &
2541 EFX_EF10_FILTER_FLAG_BUSY))
2543 prepare_to_wait(&table->waitq, &wait, TASK_UNINTERRUPTIBLE);
2544 spin_unlock_bh(&efx->filter_lock);
2548 spec = efx_ef10_filter_entry_spec(table, filter_idx);
2550 (!stack_requested &&
2551 efx_ef10_filter_rx_match_pri(table, spec->match_flags) !=
2552 filter_id / HUNT_FILTER_TBL_ROWS)) {
2557 if (spec->flags & EFX_FILTER_FLAG_RX_OVER_AUTO &&
2558 priority == EFX_FILTER_PRI_AUTO) {
2559 /* Just remove flags */
2560 spec->flags &= ~EFX_FILTER_FLAG_RX_OVER_AUTO;
2561 table->entry[filter_idx].spec &= ~EFX_EF10_FILTER_FLAG_STACK_OLD;
2566 if (spec->priority > priority) {
2571 table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_BUSY;
2572 spin_unlock_bh(&efx->filter_lock);
2574 if (spec->flags & EFX_FILTER_FLAG_RX_OVER_AUTO) {
2575 /* Reset steering of a stack-owned filter */
2577 struct efx_filter_spec new_spec = *spec;
2579 new_spec.priority = EFX_FILTER_PRI_AUTO;
2580 new_spec.flags = (EFX_FILTER_FLAG_RX |
2581 EFX_FILTER_FLAG_RX_RSS);
2582 new_spec.dmaq_id = 0;
2583 new_spec.rss_context = EFX_FILTER_RSS_CONTEXT_DEFAULT;
2584 rc = efx_ef10_filter_push(efx, &new_spec,
2585 &table->entry[filter_idx].handle,
2588 spin_lock_bh(&efx->filter_lock);
2592 /* Really remove the filter */
2594 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
2595 efx_ef10_filter_is_exclusive(spec) ?
2596 MC_CMD_FILTER_OP_IN_OP_REMOVE :
2597 MC_CMD_FILTER_OP_IN_OP_UNSUBSCRIBE);
2598 MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
2599 table->entry[filter_idx].handle);
2600 rc = efx_mcdi_rpc(efx, MC_CMD_FILTER_OP,
2601 inbuf, sizeof(inbuf), NULL, 0, NULL);
2603 spin_lock_bh(&efx->filter_lock);
2606 efx_ef10_filter_set_entry(table, filter_idx, NULL, 0);
2610 table->entry[filter_idx].spec &= ~EFX_EF10_FILTER_FLAG_BUSY;
2611 wake_up_all(&table->waitq);
2613 spin_unlock_bh(&efx->filter_lock);
2614 finish_wait(&table->waitq, &wait);
2618 static int efx_ef10_filter_remove_safe(struct efx_nic *efx,
2619 enum efx_filter_priority priority,
2622 return efx_ef10_filter_remove_internal(efx, priority, filter_id, false);
2625 static int efx_ef10_filter_get_safe(struct efx_nic *efx,
2626 enum efx_filter_priority priority,
2627 u32 filter_id, struct efx_filter_spec *spec)
2629 unsigned int filter_idx = filter_id % HUNT_FILTER_TBL_ROWS;
2630 struct efx_ef10_filter_table *table = efx->filter_state;
2631 const struct efx_filter_spec *saved_spec;
2634 spin_lock_bh(&efx->filter_lock);
2635 saved_spec = efx_ef10_filter_entry_spec(table, filter_idx);
2636 if (saved_spec && saved_spec->priority == priority &&
2637 efx_ef10_filter_rx_match_pri(table, saved_spec->match_flags) ==
2638 filter_id / HUNT_FILTER_TBL_ROWS) {
2639 *spec = *saved_spec;
2644 spin_unlock_bh(&efx->filter_lock);
2648 static void efx_ef10_filter_clear_rx(struct efx_nic *efx,
2649 enum efx_filter_priority priority)
2654 static u32 efx_ef10_filter_count_rx_used(struct efx_nic *efx,
2655 enum efx_filter_priority priority)
2657 struct efx_ef10_filter_table *table = efx->filter_state;
2658 unsigned int filter_idx;
2661 spin_lock_bh(&efx->filter_lock);
2662 for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
2663 if (table->entry[filter_idx].spec &&
2664 efx_ef10_filter_entry_spec(table, filter_idx)->priority ==
2668 spin_unlock_bh(&efx->filter_lock);
2672 static u32 efx_ef10_filter_get_rx_id_limit(struct efx_nic *efx)
2674 struct efx_ef10_filter_table *table = efx->filter_state;
2676 return table->rx_match_count * HUNT_FILTER_TBL_ROWS;
2679 static s32 efx_ef10_filter_get_rx_ids(struct efx_nic *efx,
2680 enum efx_filter_priority priority,
2683 struct efx_ef10_filter_table *table = efx->filter_state;
2684 struct efx_filter_spec *spec;
2685 unsigned int filter_idx;
2688 spin_lock_bh(&efx->filter_lock);
2689 for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
2690 spec = efx_ef10_filter_entry_spec(table, filter_idx);
2691 if (spec && spec->priority == priority) {
2692 if (count == size) {
2696 buf[count++] = (efx_ef10_filter_rx_match_pri(
2697 table, spec->match_flags) *
2698 HUNT_FILTER_TBL_ROWS +
2702 spin_unlock_bh(&efx->filter_lock);
2706 #ifdef CONFIG_RFS_ACCEL
2708 static efx_mcdi_async_completer efx_ef10_filter_rfs_insert_complete;
2710 static s32 efx_ef10_filter_rfs_insert(struct efx_nic *efx,
2711 struct efx_filter_spec *spec)
2713 struct efx_ef10_filter_table *table = efx->filter_state;
2714 MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_IN_LEN);
2715 struct efx_filter_spec *saved_spec;
2716 unsigned int hash, i, depth = 1;
2717 bool replacing = false;
2722 /* Must be an RX filter without RSS and not for a multicast
2723 * destination address (RFS only works for connected sockets).
2724 * These restrictions allow us to pass only a tiny amount of
2725 * data through to the completion function.
2727 EFX_WARN_ON_PARANOID(spec->flags !=
2728 (EFX_FILTER_FLAG_RX | EFX_FILTER_FLAG_RX_SCATTER));
2729 EFX_WARN_ON_PARANOID(spec->priority != EFX_FILTER_PRI_HINT);
2730 EFX_WARN_ON_PARANOID(efx_filter_is_mc_recipient(spec));
2732 hash = efx_ef10_filter_hash(spec);
2734 spin_lock_bh(&efx->filter_lock);
2736 /* Find any existing filter with the same match tuple or else
2737 * a free slot to insert at. If an existing filter is busy,
2738 * we have to give up.
2741 i = (hash + depth) & (HUNT_FILTER_TBL_ROWS - 1);
2742 saved_spec = efx_ef10_filter_entry_spec(table, i);
2747 } else if (efx_ef10_filter_equal(spec, saved_spec)) {
2748 if (table->entry[i].spec & EFX_EF10_FILTER_FLAG_BUSY) {
2752 if (spec->priority < saved_spec->priority) {
2760 /* Once we reach the maximum search depth, use the
2761 * first suitable slot or return -EBUSY if there was
2764 if (depth == EFX_EF10_FILTER_SEARCH_LIMIT) {
2765 if (ins_index < 0) {
2775 /* Create a software table entry if necessary, and mark it
2776 * busy. We might yet fail to insert, but any attempt to
2777 * insert a conflicting filter while we're waiting for the
2778 * firmware must find the busy entry.
2780 saved_spec = efx_ef10_filter_entry_spec(table, ins_index);
2784 saved_spec = kmalloc(sizeof(*spec), GFP_ATOMIC);
2789 *saved_spec = *spec;
2791 efx_ef10_filter_set_entry(table, ins_index, saved_spec,
2792 EFX_EF10_FILTER_FLAG_BUSY);
2794 spin_unlock_bh(&efx->filter_lock);
2796 /* Pack up the variables needed on completion */
2797 cookie = replacing << 31 | ins_index << 16 | spec->dmaq_id;
2799 efx_ef10_filter_push_prep(efx, spec, inbuf,
2800 table->entry[ins_index].handle, replacing);
2801 efx_mcdi_rpc_async(efx, MC_CMD_FILTER_OP, inbuf, sizeof(inbuf),
2802 MC_CMD_FILTER_OP_OUT_LEN,
2803 efx_ef10_filter_rfs_insert_complete, cookie);
2808 spin_unlock_bh(&efx->filter_lock);
2813 efx_ef10_filter_rfs_insert_complete(struct efx_nic *efx, unsigned long cookie,
2814 int rc, efx_dword_t *outbuf,
2815 size_t outlen_actual)
2817 struct efx_ef10_filter_table *table = efx->filter_state;
2818 unsigned int ins_index, dmaq_id;
2819 struct efx_filter_spec *spec;
2822 /* Unpack the cookie */
2823 replacing = cookie >> 31;
2824 ins_index = (cookie >> 16) & (HUNT_FILTER_TBL_ROWS - 1);
2825 dmaq_id = cookie & 0xffff;
2827 spin_lock_bh(&efx->filter_lock);
2828 spec = efx_ef10_filter_entry_spec(table, ins_index);
2830 table->entry[ins_index].handle =
2831 MCDI_QWORD(outbuf, FILTER_OP_OUT_HANDLE);
2833 spec->dmaq_id = dmaq_id;
2834 } else if (!replacing) {
2838 efx_ef10_filter_set_entry(table, ins_index, spec, 0);
2839 spin_unlock_bh(&efx->filter_lock);
2841 wake_up_all(&table->waitq);
2845 efx_ef10_filter_rfs_expire_complete(struct efx_nic *efx,
2846 unsigned long filter_idx,
2847 int rc, efx_dword_t *outbuf,
2848 size_t outlen_actual);
2850 static bool efx_ef10_filter_rfs_expire_one(struct efx_nic *efx, u32 flow_id,
2851 unsigned int filter_idx)
2853 struct efx_ef10_filter_table *table = efx->filter_state;
2854 struct efx_filter_spec *spec =
2855 efx_ef10_filter_entry_spec(table, filter_idx);
2856 MCDI_DECLARE_BUF(inbuf,
2857 MC_CMD_FILTER_OP_IN_HANDLE_OFST +
2858 MC_CMD_FILTER_OP_IN_HANDLE_LEN);
2861 (table->entry[filter_idx].spec & EFX_EF10_FILTER_FLAG_BUSY) ||
2862 spec->priority != EFX_FILTER_PRI_HINT ||
2863 !rps_may_expire_flow(efx->net_dev, spec->dmaq_id,
2864 flow_id, filter_idx))
2867 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
2868 MC_CMD_FILTER_OP_IN_OP_REMOVE);
2869 MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
2870 table->entry[filter_idx].handle);
2871 if (efx_mcdi_rpc_async(efx, MC_CMD_FILTER_OP, inbuf, sizeof(inbuf), 0,
2872 efx_ef10_filter_rfs_expire_complete, filter_idx))
2875 table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_BUSY;
2880 efx_ef10_filter_rfs_expire_complete(struct efx_nic *efx,
2881 unsigned long filter_idx,
2882 int rc, efx_dword_t *outbuf,
2883 size_t outlen_actual)
2885 struct efx_ef10_filter_table *table = efx->filter_state;
2886 struct efx_filter_spec *spec =
2887 efx_ef10_filter_entry_spec(table, filter_idx);
2889 spin_lock_bh(&efx->filter_lock);
2892 efx_ef10_filter_set_entry(table, filter_idx, NULL, 0);
2894 table->entry[filter_idx].spec &= ~EFX_EF10_FILTER_FLAG_BUSY;
2895 wake_up_all(&table->waitq);
2896 spin_unlock_bh(&efx->filter_lock);
2899 #endif /* CONFIG_RFS_ACCEL */
2901 static int efx_ef10_filter_match_flags_from_mcdi(u32 mcdi_flags)
2903 int match_flags = 0;
2905 #define MAP_FLAG(gen_flag, mcdi_field) { \
2906 u32 old_mcdi_flags = mcdi_flags; \
2907 mcdi_flags &= ~(1 << MC_CMD_FILTER_OP_IN_MATCH_ ## \
2908 mcdi_field ## _LBN); \
2909 if (mcdi_flags != old_mcdi_flags) \
2910 match_flags |= EFX_FILTER_MATCH_ ## gen_flag; \
2912 MAP_FLAG(LOC_MAC_IG, UNKNOWN_UCAST_DST);
2913 MAP_FLAG(LOC_MAC_IG, UNKNOWN_MCAST_DST);
2914 MAP_FLAG(REM_HOST, SRC_IP);
2915 MAP_FLAG(LOC_HOST, DST_IP);
2916 MAP_FLAG(REM_MAC, SRC_MAC);
2917 MAP_FLAG(REM_PORT, SRC_PORT);
2918 MAP_FLAG(LOC_MAC, DST_MAC);
2919 MAP_FLAG(LOC_PORT, DST_PORT);
2920 MAP_FLAG(ETHER_TYPE, ETHER_TYPE);
2921 MAP_FLAG(INNER_VID, INNER_VLAN);
2922 MAP_FLAG(OUTER_VID, OUTER_VLAN);
2923 MAP_FLAG(IP_PROTO, IP_PROTO);
2926 /* Did we map them all? */
2933 static int efx_ef10_filter_table_probe(struct efx_nic *efx)
2935 MCDI_DECLARE_BUF(inbuf, MC_CMD_GET_PARSER_DISP_INFO_IN_LEN);
2936 MCDI_DECLARE_BUF(outbuf, MC_CMD_GET_PARSER_DISP_INFO_OUT_LENMAX);
2937 unsigned int pd_match_pri, pd_match_count;
2938 struct efx_ef10_filter_table *table;
2942 table = kzalloc(sizeof(*table), GFP_KERNEL);
2946 /* Find out which RX filter types are supported, and their priorities */
2947 MCDI_SET_DWORD(inbuf, GET_PARSER_DISP_INFO_IN_OP,
2948 MC_CMD_GET_PARSER_DISP_INFO_IN_OP_GET_SUPPORTED_RX_MATCHES);
2949 rc = efx_mcdi_rpc(efx, MC_CMD_GET_PARSER_DISP_INFO,
2950 inbuf, sizeof(inbuf), outbuf, sizeof(outbuf),
2954 pd_match_count = MCDI_VAR_ARRAY_LEN(
2955 outlen, GET_PARSER_DISP_INFO_OUT_SUPPORTED_MATCHES);
2956 table->rx_match_count = 0;
2958 for (pd_match_pri = 0; pd_match_pri < pd_match_count; pd_match_pri++) {
2962 GET_PARSER_DISP_INFO_OUT_SUPPORTED_MATCHES,
2964 rc = efx_ef10_filter_match_flags_from_mcdi(mcdi_flags);
2966 netif_dbg(efx, probe, efx->net_dev,
2967 "%s: fw flags %#x pri %u not supported in driver\n",
2968 __func__, mcdi_flags, pd_match_pri);
2970 netif_dbg(efx, probe, efx->net_dev,
2971 "%s: fw flags %#x pri %u supported as driver flags %#x pri %u\n",
2972 __func__, mcdi_flags, pd_match_pri,
2973 rc, table->rx_match_count);
2974 table->rx_match_flags[table->rx_match_count++] = rc;
2978 table->entry = vzalloc(HUNT_FILTER_TBL_ROWS * sizeof(*table->entry));
2979 if (!table->entry) {
2984 efx->filter_state = table;
2985 init_waitqueue_head(&table->waitq);
2993 static void efx_ef10_filter_table_restore(struct efx_nic *efx)
2995 struct efx_ef10_filter_table *table = efx->filter_state;
2996 struct efx_ef10_nic_data *nic_data = efx->nic_data;
2997 struct efx_filter_spec *spec;
2998 unsigned int filter_idx;
2999 bool failed = false;
3002 if (!nic_data->must_restore_filters)
3005 spin_lock_bh(&efx->filter_lock);
3007 for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
3008 spec = efx_ef10_filter_entry_spec(table, filter_idx);
3012 table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_BUSY;
3013 spin_unlock_bh(&efx->filter_lock);
3015 rc = efx_ef10_filter_push(efx, spec,
3016 &table->entry[filter_idx].handle,
3021 spin_lock_bh(&efx->filter_lock);
3024 efx_ef10_filter_set_entry(table, filter_idx, NULL, 0);
3026 table->entry[filter_idx].spec &=
3027 ~EFX_EF10_FILTER_FLAG_BUSY;
3031 spin_unlock_bh(&efx->filter_lock);
3034 netif_err(efx, hw, efx->net_dev,
3035 "unable to restore all filters\n");
3037 nic_data->must_restore_filters = false;
3040 static void efx_ef10_filter_table_remove(struct efx_nic *efx)
3042 struct efx_ef10_filter_table *table = efx->filter_state;
3043 MCDI_DECLARE_BUF(inbuf, MC_CMD_FILTER_OP_IN_LEN);
3044 struct efx_filter_spec *spec;
3045 unsigned int filter_idx;
3048 for (filter_idx = 0; filter_idx < HUNT_FILTER_TBL_ROWS; filter_idx++) {
3049 spec = efx_ef10_filter_entry_spec(table, filter_idx);
3053 MCDI_SET_DWORD(inbuf, FILTER_OP_IN_OP,
3054 efx_ef10_filter_is_exclusive(spec) ?
3055 MC_CMD_FILTER_OP_IN_OP_REMOVE :
3056 MC_CMD_FILTER_OP_IN_OP_UNSUBSCRIBE);
3057 MCDI_SET_QWORD(inbuf, FILTER_OP_IN_HANDLE,
3058 table->entry[filter_idx].handle);
3059 rc = efx_mcdi_rpc(efx, MC_CMD_FILTER_OP, inbuf, sizeof(inbuf),
3062 netdev_WARN(efx->net_dev,
3063 "filter_idx=%#x handle=%#llx\n",
3065 table->entry[filter_idx].handle);
3069 vfree(table->entry);
3073 static void efx_ef10_filter_sync_rx_mode(struct efx_nic *efx)
3075 struct efx_ef10_filter_table *table = efx->filter_state;
3076 struct net_device *net_dev = efx->net_dev;
3077 struct efx_filter_spec spec;
3078 bool remove_failed = false;
3079 struct netdev_hw_addr *uc;
3080 struct netdev_hw_addr *mc;
3081 unsigned int filter_idx;
3084 if (!efx_dev_registered(efx))
3087 /* Mark old filters that may need to be removed */
3088 spin_lock_bh(&efx->filter_lock);
3089 n = table->stack_uc_count < 0 ? 1 : table->stack_uc_count;
3090 for (i = 0; i < n; i++) {
3091 filter_idx = table->stack_uc_list[i].id % HUNT_FILTER_TBL_ROWS;
3092 table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_STACK_OLD;
3094 n = table->stack_mc_count < 0 ? 1 : table->stack_mc_count;
3095 for (i = 0; i < n; i++) {
3096 filter_idx = table->stack_mc_list[i].id % HUNT_FILTER_TBL_ROWS;
3097 table->entry[filter_idx].spec |= EFX_EF10_FILTER_FLAG_STACK_OLD;
3099 spin_unlock_bh(&efx->filter_lock);
3101 /* Copy/convert the address lists; add the primary station
3102 * address and broadcast address
3104 netif_addr_lock_bh(net_dev);
3105 if (net_dev->flags & IFF_PROMISC ||
3106 netdev_uc_count(net_dev) >= EFX_EF10_FILTER_STACK_UC_MAX) {
3107 table->stack_uc_count = -1;
3109 table->stack_uc_count = 1 + netdev_uc_count(net_dev);
3110 memcpy(table->stack_uc_list[0].addr, net_dev->dev_addr,
3113 netdev_for_each_uc_addr(uc, net_dev) {
3114 memcpy(table->stack_uc_list[i].addr,
3115 uc->addr, ETH_ALEN);
3119 if (net_dev->flags & (IFF_PROMISC | IFF_ALLMULTI) ||
3120 netdev_mc_count(net_dev) >= EFX_EF10_FILTER_STACK_MC_MAX) {
3121 table->stack_mc_count = -1;
3123 table->stack_mc_count = 1 + netdev_mc_count(net_dev);
3124 eth_broadcast_addr(table->stack_mc_list[0].addr);
3126 netdev_for_each_mc_addr(mc, net_dev) {
3127 memcpy(table->stack_mc_list[i].addr,
3128 mc->addr, ETH_ALEN);
3132 netif_addr_unlock_bh(net_dev);
3134 /* Insert/renew unicast filters */
3135 if (table->stack_uc_count >= 0) {
3136 for (i = 0; i < table->stack_uc_count; i++) {
3137 efx_filter_init_rx(&spec, EFX_FILTER_PRI_AUTO,
3138 EFX_FILTER_FLAG_RX_RSS,
3140 efx_filter_set_eth_local(&spec, EFX_FILTER_VID_UNSPEC,
3141 table->stack_uc_list[i].addr);
3142 rc = efx_ef10_filter_insert(efx, &spec, true);
3144 /* Fall back to unicast-promisc */
3146 efx_ef10_filter_remove_safe(
3147 efx, EFX_FILTER_PRI_AUTO,
3148 table->stack_uc_list[i].id);
3149 table->stack_uc_count = -1;
3152 table->stack_uc_list[i].id = rc;
3155 if (table->stack_uc_count < 0) {
3156 efx_filter_init_rx(&spec, EFX_FILTER_PRI_AUTO,
3157 EFX_FILTER_FLAG_RX_RSS,
3159 efx_filter_set_uc_def(&spec);
3160 rc = efx_ef10_filter_insert(efx, &spec, true);
3163 table->stack_uc_count = 0;
3165 table->stack_uc_list[0].id = rc;
3169 /* Insert/renew multicast filters */
3170 if (table->stack_mc_count >= 0) {
3171 for (i = 0; i < table->stack_mc_count; i++) {
3172 efx_filter_init_rx(&spec, EFX_FILTER_PRI_AUTO,
3173 EFX_FILTER_FLAG_RX_RSS,
3175 efx_filter_set_eth_local(&spec, EFX_FILTER_VID_UNSPEC,
3176 table->stack_mc_list[i].addr);
3177 rc = efx_ef10_filter_insert(efx, &spec, true);
3179 /* Fall back to multicast-promisc */
3181 efx_ef10_filter_remove_safe(
3182 efx, EFX_FILTER_PRI_AUTO,
3183 table->stack_mc_list[i].id);
3184 table->stack_mc_count = -1;
3187 table->stack_mc_list[i].id = rc;
3190 if (table->stack_mc_count < 0) {
3191 efx_filter_init_rx(&spec, EFX_FILTER_PRI_AUTO,
3192 EFX_FILTER_FLAG_RX_RSS,
3194 efx_filter_set_mc_def(&spec);
3195 rc = efx_ef10_filter_insert(efx, &spec, true);
3198 table->stack_mc_count = 0;
3200 table->stack_mc_list[0].id = rc;
3204 /* Remove filters that weren't renewed. Since nothing else
3205 * changes the STACK_OLD flag or removes these filters, we
3206 * don't need to hold the filter_lock while scanning for
3209 for (i = 0; i < HUNT_FILTER_TBL_ROWS; i++) {
3210 if (ACCESS_ONCE(table->entry[i].spec) &
3211 EFX_EF10_FILTER_FLAG_STACK_OLD) {
3212 if (efx_ef10_filter_remove_internal(
3213 efx, EFX_FILTER_PRI_AUTO, i, true) < 0)
3214 remove_failed = true;
3217 WARN_ON(remove_failed);
3220 static int efx_ef10_mac_reconfigure(struct efx_nic *efx)
3222 efx_ef10_filter_sync_rx_mode(efx);
3224 return efx_mcdi_set_mac(efx);
3227 static int efx_ef10_start_bist(struct efx_nic *efx, u32 bist_type)
3229 MCDI_DECLARE_BUF(inbuf, MC_CMD_START_BIST_IN_LEN);
3231 MCDI_SET_DWORD(inbuf, START_BIST_IN_TYPE, bist_type);
3232 return efx_mcdi_rpc(efx, MC_CMD_START_BIST, inbuf, sizeof(inbuf),
3236 /* MC BISTs follow a different poll mechanism to phy BISTs.
3237 * The BIST is done in the poll handler on the MC, and the MCDI command
3238 * will block until the BIST is done.
3240 static int efx_ef10_poll_bist(struct efx_nic *efx)
3243 MCDI_DECLARE_BUF(outbuf, MC_CMD_POLL_BIST_OUT_LEN);
3247 rc = efx_mcdi_rpc(efx, MC_CMD_POLL_BIST, NULL, 0,
3248 outbuf, sizeof(outbuf), &outlen);
3252 if (outlen < MC_CMD_POLL_BIST_OUT_LEN)
3255 result = MCDI_DWORD(outbuf, POLL_BIST_OUT_RESULT);
3257 case MC_CMD_POLL_BIST_PASSED:
3258 netif_dbg(efx, hw, efx->net_dev, "BIST passed.\n");
3260 case MC_CMD_POLL_BIST_TIMEOUT:
3261 netif_err(efx, hw, efx->net_dev, "BIST timed out\n");
3263 case MC_CMD_POLL_BIST_FAILED:
3264 netif_err(efx, hw, efx->net_dev, "BIST failed.\n");
3267 netif_err(efx, hw, efx->net_dev,
3268 "BIST returned unknown result %u", result);
3273 static int efx_ef10_run_bist(struct efx_nic *efx, u32 bist_type)
3277 netif_dbg(efx, drv, efx->net_dev, "starting BIST type %u\n", bist_type);
3279 rc = efx_ef10_start_bist(efx, bist_type);
3283 return efx_ef10_poll_bist(efx);
3287 efx_ef10_test_chip(struct efx_nic *efx, struct efx_self_tests *tests)
3291 efx_reset_down(efx, RESET_TYPE_WORLD);
3293 rc = efx_mcdi_rpc(efx, MC_CMD_ENABLE_OFFLINE_BIST,
3294 NULL, 0, NULL, 0, NULL);
3298 tests->memory = efx_ef10_run_bist(efx, MC_CMD_MC_MEM_BIST) ? -1 : 1;
3299 tests->registers = efx_ef10_run_bist(efx, MC_CMD_REG_BIST) ? -1 : 1;
3301 rc = efx_mcdi_reset(efx, RESET_TYPE_WORLD);
3304 rc2 = efx_reset_up(efx, RESET_TYPE_WORLD, rc == 0);
3305 return rc ? rc : rc2;
3308 #ifdef CONFIG_SFC_MTD
3310 struct efx_ef10_nvram_type_info {
3311 u16 type, type_mask;
3316 static const struct efx_ef10_nvram_type_info efx_ef10_nvram_types[] = {
3317 { NVRAM_PARTITION_TYPE_MC_FIRMWARE, 0, 0, "sfc_mcfw" },
3318 { NVRAM_PARTITION_TYPE_MC_FIRMWARE_BACKUP, 0, 0, "sfc_mcfw_backup" },
3319 { NVRAM_PARTITION_TYPE_EXPANSION_ROM, 0, 0, "sfc_exp_rom" },
3320 { NVRAM_PARTITION_TYPE_STATIC_CONFIG, 0, 0, "sfc_static_cfg" },
3321 { NVRAM_PARTITION_TYPE_DYNAMIC_CONFIG, 0, 0, "sfc_dynamic_cfg" },
3322 { NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT0, 0, 0, "sfc_exp_rom_cfg" },
3323 { NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT1, 0, 1, "sfc_exp_rom_cfg" },
3324 { NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT2, 0, 2, "sfc_exp_rom_cfg" },
3325 { NVRAM_PARTITION_TYPE_EXPROM_CONFIG_PORT3, 0, 3, "sfc_exp_rom_cfg" },
3326 { NVRAM_PARTITION_TYPE_LICENSE, 0, 0, "sfc_license" },
3327 { NVRAM_PARTITION_TYPE_PHY_MIN, 0xff, 0, "sfc_phy_fw" },
3330 static int efx_ef10_mtd_probe_partition(struct efx_nic *efx,
3331 struct efx_mcdi_mtd_partition *part,
3334 MCDI_DECLARE_BUF(inbuf, MC_CMD_NVRAM_METADATA_IN_LEN);
3335 MCDI_DECLARE_BUF(outbuf, MC_CMD_NVRAM_METADATA_OUT_LENMAX);
3336 const struct efx_ef10_nvram_type_info *info;
3337 size_t size, erase_size, outlen;
3341 for (info = efx_ef10_nvram_types; ; info++) {
3343 efx_ef10_nvram_types + ARRAY_SIZE(efx_ef10_nvram_types))
3345 if ((type & ~info->type_mask) == info->type)
3348 if (info->port != efx_port_num(efx))
3351 rc = efx_mcdi_nvram_info(efx, type, &size, &erase_size, &protected);
3355 return -ENODEV; /* hide it */
3357 part->nvram_type = type;
3359 MCDI_SET_DWORD(inbuf, NVRAM_METADATA_IN_TYPE, type);
3360 rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_METADATA, inbuf, sizeof(inbuf),
3361 outbuf, sizeof(outbuf), &outlen);
3364 if (outlen < MC_CMD_NVRAM_METADATA_OUT_LENMIN)
3366 if (MCDI_DWORD(outbuf, NVRAM_METADATA_OUT_FLAGS) &
3367 (1 << MC_CMD_NVRAM_METADATA_OUT_SUBTYPE_VALID_LBN))
3368 part->fw_subtype = MCDI_DWORD(outbuf,
3369 NVRAM_METADATA_OUT_SUBTYPE);
3371 part->common.dev_type_name = "EF10 NVRAM manager";
3372 part->common.type_name = info->name;
3374 part->common.mtd.type = MTD_NORFLASH;
3375 part->common.mtd.flags = MTD_CAP_NORFLASH;
3376 part->common.mtd.size = size;
3377 part->common.mtd.erasesize = erase_size;
3382 static int efx_ef10_mtd_probe(struct efx_nic *efx)
3384 MCDI_DECLARE_BUF(outbuf, MC_CMD_NVRAM_PARTITIONS_OUT_LENMAX);
3385 struct efx_mcdi_mtd_partition *parts;
3386 size_t outlen, n_parts_total, i, n_parts;
3392 BUILD_BUG_ON(MC_CMD_NVRAM_PARTITIONS_IN_LEN != 0);
3393 rc = efx_mcdi_rpc(efx, MC_CMD_NVRAM_PARTITIONS, NULL, 0,
3394 outbuf, sizeof(outbuf), &outlen);
3397 if (outlen < MC_CMD_NVRAM_PARTITIONS_OUT_LENMIN)
3400 n_parts_total = MCDI_DWORD(outbuf, NVRAM_PARTITIONS_OUT_NUM_PARTITIONS);
3402 MCDI_VAR_ARRAY_LEN(outlen, NVRAM_PARTITIONS_OUT_TYPE_ID))
3405 parts = kcalloc(n_parts_total, sizeof(*parts), GFP_KERNEL);
3410 for (i = 0; i < n_parts_total; i++) {
3411 type = MCDI_ARRAY_DWORD(outbuf, NVRAM_PARTITIONS_OUT_TYPE_ID,
3413 rc = efx_ef10_mtd_probe_partition(efx, &parts[n_parts], type);
3416 else if (rc != -ENODEV)
3420 rc = efx_mtd_add(efx, &parts[0].common, n_parts, sizeof(*parts));
3427 #endif /* CONFIG_SFC_MTD */
3429 static void efx_ef10_ptp_write_host_time(struct efx_nic *efx, u32 host_time)
3431 _efx_writed(efx, cpu_to_le32(host_time), ER_DZ_MC_DB_LWRD);
3434 static int efx_ef10_rx_enable_timestamping(struct efx_channel *channel,
3437 MCDI_DECLARE_BUF(inbuf, MC_CMD_PTP_IN_TIME_EVENT_SUBSCRIBE_LEN);
3440 if (channel->sync_events_state == SYNC_EVENTS_REQUESTED ||
3441 channel->sync_events_state == SYNC_EVENTS_VALID ||
3442 (temp && channel->sync_events_state == SYNC_EVENTS_DISABLED))
3444 channel->sync_events_state = SYNC_EVENTS_REQUESTED;
3446 MCDI_SET_DWORD(inbuf, PTP_IN_OP, MC_CMD_PTP_OP_TIME_EVENT_SUBSCRIBE);
3447 MCDI_SET_DWORD(inbuf, PTP_IN_PERIPH_ID, 0);
3448 MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_SUBSCRIBE_QUEUE,
3451 rc = efx_mcdi_rpc(channel->efx, MC_CMD_PTP,
3452 inbuf, sizeof(inbuf), NULL, 0, NULL);
3455 channel->sync_events_state = temp ? SYNC_EVENTS_QUIESCENT :
3456 SYNC_EVENTS_DISABLED;
3461 static int efx_ef10_rx_disable_timestamping(struct efx_channel *channel,
3464 MCDI_DECLARE_BUF(inbuf, MC_CMD_PTP_IN_TIME_EVENT_UNSUBSCRIBE_LEN);
3467 if (channel->sync_events_state == SYNC_EVENTS_DISABLED ||
3468 (temp && channel->sync_events_state == SYNC_EVENTS_QUIESCENT))
3470 if (channel->sync_events_state == SYNC_EVENTS_QUIESCENT) {
3471 channel->sync_events_state = SYNC_EVENTS_DISABLED;
3474 channel->sync_events_state = temp ? SYNC_EVENTS_QUIESCENT :
3475 SYNC_EVENTS_DISABLED;
3477 MCDI_SET_DWORD(inbuf, PTP_IN_OP, MC_CMD_PTP_OP_TIME_EVENT_UNSUBSCRIBE);
3478 MCDI_SET_DWORD(inbuf, PTP_IN_PERIPH_ID, 0);
3479 MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_UNSUBSCRIBE_CONTROL,
3480 MC_CMD_PTP_IN_TIME_EVENT_UNSUBSCRIBE_SINGLE);
3481 MCDI_SET_DWORD(inbuf, PTP_IN_TIME_EVENT_UNSUBSCRIBE_QUEUE,
3484 rc = efx_mcdi_rpc(channel->efx, MC_CMD_PTP,
3485 inbuf, sizeof(inbuf), NULL, 0, NULL);
3490 static int efx_ef10_ptp_set_ts_sync_events(struct efx_nic *efx, bool en,
3493 int (*set)(struct efx_channel *channel, bool temp);
3494 struct efx_channel *channel;
3497 efx_ef10_rx_enable_timestamping :
3498 efx_ef10_rx_disable_timestamping;
3500 efx_for_each_channel(channel, efx) {
3501 int rc = set(channel, temp);
3502 if (en && rc != 0) {
3503 efx_ef10_ptp_set_ts_sync_events(efx, false, temp);
3511 static int efx_ef10_ptp_set_ts_config(struct efx_nic *efx,
3512 struct hwtstamp_config *init)
3516 switch (init->rx_filter) {
3517 case HWTSTAMP_FILTER_NONE:
3518 efx_ef10_ptp_set_ts_sync_events(efx, false, false);
3519 /* if TX timestamping is still requested then leave PTP on */
3520 return efx_ptp_change_mode(efx,
3521 init->tx_type != HWTSTAMP_TX_OFF, 0);
3522 case HWTSTAMP_FILTER_ALL:
3523 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
3524 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
3525 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
3526 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
3527 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
3528 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
3529 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
3530 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
3531 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
3532 case HWTSTAMP_FILTER_PTP_V2_EVENT:
3533 case HWTSTAMP_FILTER_PTP_V2_SYNC:
3534 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
3535 init->rx_filter = HWTSTAMP_FILTER_ALL;
3536 rc = efx_ptp_change_mode(efx, true, 0);
3538 rc = efx_ef10_ptp_set_ts_sync_events(efx, true, false);
3540 efx_ptp_change_mode(efx, false, 0);
3547 const struct efx_nic_type efx_hunt_a0_nic_type = {
3548 .mem_map_size = efx_ef10_mem_map_size,
3549 .probe = efx_ef10_probe,
3550 .remove = efx_ef10_remove,
3551 .dimension_resources = efx_ef10_dimension_resources,
3552 .init = efx_ef10_init_nic,
3553 .fini = efx_port_dummy_op_void,
3554 .map_reset_reason = efx_mcdi_map_reset_reason,
3555 .map_reset_flags = efx_ef10_map_reset_flags,
3556 .reset = efx_mcdi_reset,
3557 .probe_port = efx_mcdi_port_probe,
3558 .remove_port = efx_mcdi_port_remove,
3559 .fini_dmaq = efx_ef10_fini_dmaq,
3560 .describe_stats = efx_ef10_describe_stats,
3561 .update_stats = efx_ef10_update_stats,
3562 .start_stats = efx_mcdi_mac_start_stats,
3563 .pull_stats = efx_mcdi_mac_pull_stats,
3564 .stop_stats = efx_mcdi_mac_stop_stats,
3565 .set_id_led = efx_mcdi_set_id_led,
3566 .push_irq_moderation = efx_ef10_push_irq_moderation,
3567 .reconfigure_mac = efx_ef10_mac_reconfigure,
3568 .check_mac_fault = efx_mcdi_mac_check_fault,
3569 .reconfigure_port = efx_mcdi_port_reconfigure,
3570 .get_wol = efx_ef10_get_wol,
3571 .set_wol = efx_ef10_set_wol,
3572 .resume_wol = efx_port_dummy_op_void,
3573 .test_chip = efx_ef10_test_chip,
3574 .test_nvram = efx_mcdi_nvram_test_all,
3575 .mcdi_request = efx_ef10_mcdi_request,
3576 .mcdi_poll_response = efx_ef10_mcdi_poll_response,
3577 .mcdi_read_response = efx_ef10_mcdi_read_response,
3578 .mcdi_poll_reboot = efx_ef10_mcdi_poll_reboot,
3579 .irq_enable_master = efx_port_dummy_op_void,
3580 .irq_test_generate = efx_ef10_irq_test_generate,
3581 .irq_disable_non_ev = efx_port_dummy_op_void,
3582 .irq_handle_msi = efx_ef10_msi_interrupt,
3583 .irq_handle_legacy = efx_ef10_legacy_interrupt,
3584 .tx_probe = efx_ef10_tx_probe,
3585 .tx_init = efx_ef10_tx_init,
3586 .tx_remove = efx_ef10_tx_remove,
3587 .tx_write = efx_ef10_tx_write,
3588 .rx_push_rss_config = efx_ef10_rx_push_rss_config,
3589 .rx_probe = efx_ef10_rx_probe,
3590 .rx_init = efx_ef10_rx_init,
3591 .rx_remove = efx_ef10_rx_remove,
3592 .rx_write = efx_ef10_rx_write,
3593 .rx_defer_refill = efx_ef10_rx_defer_refill,
3594 .ev_probe = efx_ef10_ev_probe,
3595 .ev_init = efx_ef10_ev_init,
3596 .ev_fini = efx_ef10_ev_fini,
3597 .ev_remove = efx_ef10_ev_remove,
3598 .ev_process = efx_ef10_ev_process,
3599 .ev_read_ack = efx_ef10_ev_read_ack,
3600 .ev_test_generate = efx_ef10_ev_test_generate,
3601 .filter_table_probe = efx_ef10_filter_table_probe,
3602 .filter_table_restore = efx_ef10_filter_table_restore,
3603 .filter_table_remove = efx_ef10_filter_table_remove,
3604 .filter_update_rx_scatter = efx_ef10_filter_update_rx_scatter,
3605 .filter_insert = efx_ef10_filter_insert,
3606 .filter_remove_safe = efx_ef10_filter_remove_safe,
3607 .filter_get_safe = efx_ef10_filter_get_safe,
3608 .filter_clear_rx = efx_ef10_filter_clear_rx,
3609 .filter_count_rx_used = efx_ef10_filter_count_rx_used,
3610 .filter_get_rx_id_limit = efx_ef10_filter_get_rx_id_limit,
3611 .filter_get_rx_ids = efx_ef10_filter_get_rx_ids,
3612 #ifdef CONFIG_RFS_ACCEL
3613 .filter_rfs_insert = efx_ef10_filter_rfs_insert,
3614 .filter_rfs_expire_one = efx_ef10_filter_rfs_expire_one,
3616 #ifdef CONFIG_SFC_MTD
3617 .mtd_probe = efx_ef10_mtd_probe,
3618 .mtd_rename = efx_mcdi_mtd_rename,
3619 .mtd_read = efx_mcdi_mtd_read,
3620 .mtd_erase = efx_mcdi_mtd_erase,
3621 .mtd_write = efx_mcdi_mtd_write,
3622 .mtd_sync = efx_mcdi_mtd_sync,
3624 .ptp_write_host_time = efx_ef10_ptp_write_host_time,
3625 .ptp_set_ts_sync_events = efx_ef10_ptp_set_ts_sync_events,
3626 .ptp_set_ts_config = efx_ef10_ptp_set_ts_config,
3628 .revision = EFX_REV_HUNT_A0,
3629 .max_dma_mask = DMA_BIT_MASK(ESF_DZ_TX_KER_BUF_ADDR_WIDTH),
3630 .rx_prefix_size = ES_DZ_RX_PREFIX_SIZE,
3631 .rx_hash_offset = ES_DZ_RX_PREFIX_HASH_OFST,
3632 .rx_ts_offset = ES_DZ_RX_PREFIX_TSTAMP_OFST,
3633 .can_rx_scatter = true,
3634 .always_rx_scatter = true,
3635 .max_interrupt_mode = EFX_INT_MODE_MSIX,
3636 .timer_period_max = 1 << ERF_DD_EVQ_IND_TIMER_VAL_WIDTH,
3637 .offload_features = (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
3638 NETIF_F_RXHASH | NETIF_F_NTUPLE),
3640 .max_rx_ip_filters = HUNT_FILTER_TBL_ROWS,
3641 .hwtstamp_filters = 1 << HWTSTAMP_FILTER_NONE |
3642 1 << HWTSTAMP_FILTER_ALL,