1 /****************************************************************************
2 * Driver for Solarflare network controllers and boards
3 * Copyright 2005-2006 Fen Systems Ltd.
4 * Copyright 2006-2012 Solarflare Communications Inc.
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 as published
8 * by the Free Software Foundation, incorporated herein by reference.
11 #include <linux/netdevice.h>
12 #include <linux/module.h>
13 #include <linux/delay.h>
14 #include <linux/kernel_stat.h>
15 #include <linux/pci.h>
16 #include <linux/ethtool.h>
19 #include <linux/udp.h>
20 #include <linux/rtnetlink.h>
21 #include <linux/slab.h>
22 #include "net_driver.h"
26 #include "workarounds.h"
28 /* IRQ latency can be enormous because:
29 * - All IRQs may be disabled on a CPU for a *long* time by e.g. a
30 * slow serial console or an old IDE driver doing error recovery
31 * - The PREEMPT_RT patches mostly deal with this, but also allow a
32 * tasklet or normal task to be given higher priority than our IRQ
34 * Try to avoid blaming the hardware for this.
36 #define IRQ_TIMEOUT HZ
39 * Loopback test packet structure
41 * The self-test should stress every RSS vector, and unfortunately
42 * Falcon only performs RSS on TCP/UDP packets.
44 struct efx_loopback_payload {
52 /* Loopback test source MAC address */
53 static const u8 payload_source[ETH_ALEN] __aligned(2) = {
54 0x00, 0x0f, 0x53, 0x1b, 0x1b, 0x1b,
57 static const char payload_msg[] =
58 "Hello world! This is an Efx loopback test in progress!";
60 /* Interrupt mode names */
61 static const unsigned int efx_interrupt_mode_max = EFX_INT_MODE_MAX;
62 static const char *const efx_interrupt_mode_names[] = {
63 [EFX_INT_MODE_MSIX] = "MSI-X",
64 [EFX_INT_MODE_MSI] = "MSI",
65 [EFX_INT_MODE_LEGACY] = "legacy",
67 #define INT_MODE(efx) \
68 STRING_TABLE_LOOKUP(efx->interrupt_mode, efx_interrupt_mode)
71 * efx_loopback_state - persistent state during a loopback selftest
72 * @flush: Drop all packets in efx_loopback_rx_packet
73 * @packet_count: Number of packets being used in this test
74 * @skbs: An array of skbs transmitted
75 * @offload_csum: Checksums are being offloaded
76 * @rx_good: RX good packet count
77 * @rx_bad: RX bad packet count
78 * @payload: Payload used in tests
80 struct efx_loopback_state {
83 struct sk_buff **skbs;
87 struct efx_loopback_payload payload;
90 /* How long to wait for all the packets to arrive (in ms) */
91 #define LOOPBACK_TIMEOUT_MS 1000
93 /**************************************************************************
95 * MII, NVRAM and register tests
97 **************************************************************************/
99 static int efx_test_phy_alive(struct efx_nic *efx, struct efx_self_tests *tests)
103 if (efx->phy_op->test_alive) {
104 rc = efx->phy_op->test_alive(efx);
105 tests->phy_alive = rc ? -1 : 1;
111 static int efx_test_nvram(struct efx_nic *efx, struct efx_self_tests *tests)
115 if (efx->type->test_nvram) {
116 rc = efx->type->test_nvram(efx);
120 tests->nvram = rc ? -1 : 1;
126 /**************************************************************************
128 * Interrupt and event queue testing
130 **************************************************************************/
132 /* Test generation and receipt of interrupts */
133 static int efx_test_interrupts(struct efx_nic *efx,
134 struct efx_self_tests *tests)
136 unsigned long timeout, wait;
140 netif_dbg(efx, drv, efx->net_dev, "testing interrupts\n");
141 tests->interrupt = -1;
143 rc = efx_nic_irq_test_start(efx);
144 if (rc == -ENOTSUPP) {
145 netif_dbg(efx, drv, efx->net_dev,
146 "direct interrupt testing not supported\n");
147 tests->interrupt = 0;
151 timeout = jiffies + IRQ_TIMEOUT;
154 /* Wait for arrival of test interrupt. */
155 netif_dbg(efx, drv, efx->net_dev, "waiting for test interrupt\n");
157 schedule_timeout_uninterruptible(wait);
158 cpu = efx_nic_irq_test_irq_cpu(efx);
162 } while (time_before(jiffies, timeout));
164 netif_err(efx, drv, efx->net_dev, "timed out waiting for interrupt\n");
168 netif_dbg(efx, drv, efx->net_dev, "%s test interrupt seen on CPU%d\n",
170 tests->interrupt = 1;
174 /* Test generation and receipt of interrupting events */
175 static int efx_test_eventq_irq(struct efx_nic *efx,
176 struct efx_self_tests *tests)
178 struct efx_channel *channel;
179 unsigned int read_ptr[EFX_MAX_CHANNELS];
180 unsigned long napi_ran = 0, dma_pend = 0, int_pend = 0;
181 unsigned long timeout, wait;
183 BUILD_BUG_ON(EFX_MAX_CHANNELS > BITS_PER_LONG);
185 efx_for_each_channel(channel, efx) {
186 read_ptr[channel->channel] = channel->eventq_read_ptr;
187 set_bit(channel->channel, &dma_pend);
188 set_bit(channel->channel, &int_pend);
189 efx_nic_event_test_start(channel);
192 timeout = jiffies + IRQ_TIMEOUT;
195 /* Wait for arrival of interrupts. NAPI processing may or may
196 * not complete in time, but we can cope in any case.
199 schedule_timeout_uninterruptible(wait);
201 efx_for_each_channel(channel, efx) {
202 efx_stop_eventq(channel);
203 if (channel->eventq_read_ptr !=
204 read_ptr[channel->channel]) {
205 set_bit(channel->channel, &napi_ran);
206 clear_bit(channel->channel, &dma_pend);
207 clear_bit(channel->channel, &int_pend);
209 if (efx_nic_event_present(channel))
210 clear_bit(channel->channel, &dma_pend);
211 if (efx_nic_event_test_irq_cpu(channel) >= 0)
212 clear_bit(channel->channel, &int_pend);
214 efx_start_eventq(channel);
218 } while ((dma_pend || int_pend) && time_before(jiffies, timeout));
220 efx_for_each_channel(channel, efx) {
221 bool dma_seen = !test_bit(channel->channel, &dma_pend);
222 bool int_seen = !test_bit(channel->channel, &int_pend);
224 tests->eventq_dma[channel->channel] = dma_seen ? 1 : -1;
225 tests->eventq_int[channel->channel] = int_seen ? 1 : -1;
227 if (dma_seen && int_seen) {
228 netif_dbg(efx, drv, efx->net_dev,
229 "channel %d event queue passed (with%s NAPI)\n",
231 test_bit(channel->channel, &napi_ran) ?
234 /* Report failure and whether either interrupt or DMA
237 netif_err(efx, drv, efx->net_dev,
238 "channel %d timed out waiting for event queue\n",
241 netif_err(efx, drv, efx->net_dev,
242 "channel %d saw interrupt "
243 "during event queue test\n",
246 netif_err(efx, drv, efx->net_dev,
247 "channel %d event was generated, but "
248 "failed to trigger an interrupt\n",
253 return (dma_pend || int_pend) ? -ETIMEDOUT : 0;
256 static int efx_test_phy(struct efx_nic *efx, struct efx_self_tests *tests,
261 if (!efx->phy_op->run_tests)
264 mutex_lock(&efx->mac_lock);
265 rc = efx->phy_op->run_tests(efx, tests->phy_ext, flags);
266 mutex_unlock(&efx->mac_lock);
270 netif_info(efx, drv, efx->net_dev,
271 "%s phy selftest\n", rc ? "Failed" : "Passed");
276 /**************************************************************************
279 * NB Only one loopback test can be executing concurrently.
281 **************************************************************************/
283 /* Loopback test RX callback
284 * This is called for each received packet during loopback testing.
286 void efx_loopback_rx_packet(struct efx_nic *efx,
287 const char *buf_ptr, int pkt_len)
289 struct efx_loopback_state *state = efx->loopback_selftest;
290 struct efx_loopback_payload *received;
291 struct efx_loopback_payload *payload;
295 /* If we are just flushing, then drop the packet */
296 if ((state == NULL) || state->flush)
299 payload = &state->payload;
301 received = (struct efx_loopback_payload *) buf_ptr;
302 received->ip.saddr = payload->ip.saddr;
303 if (state->offload_csum)
304 received->ip.check = payload->ip.check;
306 /* Check that header exists */
307 if (pkt_len < sizeof(received->header)) {
308 netif_err(efx, drv, efx->net_dev,
309 "saw runt RX packet (length %d) in %s loopback "
310 "test\n", pkt_len, LOOPBACK_MODE(efx));
314 /* Check that the ethernet header exists */
315 if (memcmp(&received->header, &payload->header, ETH_HLEN) != 0) {
316 netif_err(efx, drv, efx->net_dev,
317 "saw non-loopback RX packet in %s loopback test\n",
322 /* Check packet length */
323 if (pkt_len != sizeof(*payload)) {
324 netif_err(efx, drv, efx->net_dev,
325 "saw incorrect RX packet length %d (wanted %d) in "
326 "%s loopback test\n", pkt_len, (int)sizeof(*payload),
331 /* Check that IP header matches */
332 if (memcmp(&received->ip, &payload->ip, sizeof(payload->ip)) != 0) {
333 netif_err(efx, drv, efx->net_dev,
334 "saw corrupted IP header in %s loopback test\n",
339 /* Check that msg and padding matches */
340 if (memcmp(&received->msg, &payload->msg, sizeof(received->msg)) != 0) {
341 netif_err(efx, drv, efx->net_dev,
342 "saw corrupted RX packet in %s loopback test\n",
347 /* Check that iteration matches */
348 if (received->iteration != payload->iteration) {
349 netif_err(efx, drv, efx->net_dev,
350 "saw RX packet from iteration %d (wanted %d) in "
351 "%s loopback test\n", ntohs(received->iteration),
352 ntohs(payload->iteration), LOOPBACK_MODE(efx));
356 /* Increase correct RX count */
357 netif_vdbg(efx, drv, efx->net_dev,
358 "got loopback RX in %s loopback test\n", LOOPBACK_MODE(efx));
360 atomic_inc(&state->rx_good);
365 if (atomic_read(&state->rx_bad) == 0) {
366 netif_err(efx, drv, efx->net_dev, "received packet:\n");
367 print_hex_dump(KERN_ERR, "", DUMP_PREFIX_OFFSET, 0x10, 1,
368 buf_ptr, pkt_len, 0);
369 netif_err(efx, drv, efx->net_dev, "expected packet:\n");
370 print_hex_dump(KERN_ERR, "", DUMP_PREFIX_OFFSET, 0x10, 1,
371 &state->payload, sizeof(state->payload), 0);
374 atomic_inc(&state->rx_bad);
377 /* Initialise an efx_selftest_state for a new iteration */
378 static void efx_iterate_state(struct efx_nic *efx)
380 struct efx_loopback_state *state = efx->loopback_selftest;
381 struct net_device *net_dev = efx->net_dev;
382 struct efx_loopback_payload *payload = &state->payload;
384 /* Initialise the layerII header */
385 ether_addr_copy((u8 *)&payload->header.h_dest, net_dev->dev_addr);
386 ether_addr_copy((u8 *)&payload->header.h_source, payload_source);
387 payload->header.h_proto = htons(ETH_P_IP);
389 /* saddr set later and used as incrementing count */
390 payload->ip.daddr = htonl(INADDR_LOOPBACK);
392 payload->ip.check = (__force __sum16) htons(0xdead);
393 payload->ip.tot_len = htons(sizeof(*payload) - sizeof(struct ethhdr));
394 payload->ip.version = IPVERSION;
395 payload->ip.protocol = IPPROTO_UDP;
397 /* Initialise udp header */
398 payload->udp.source = 0;
399 payload->udp.len = htons(sizeof(*payload) - sizeof(struct ethhdr) -
400 sizeof(struct iphdr));
401 payload->udp.check = 0; /* checksum ignored */
403 /* Fill out payload */
404 payload->iteration = htons(ntohs(payload->iteration) + 1);
405 memcpy(&payload->msg, payload_msg, sizeof(payload_msg));
407 /* Fill out remaining state members */
408 atomic_set(&state->rx_good, 0);
409 atomic_set(&state->rx_bad, 0);
413 static int efx_begin_loopback(struct efx_tx_queue *tx_queue)
415 struct efx_nic *efx = tx_queue->efx;
416 struct efx_loopback_state *state = efx->loopback_selftest;
417 struct efx_loopback_payload *payload;
422 /* Transmit N copies of buffer */
423 for (i = 0; i < state->packet_count; i++) {
424 /* Allocate an skb, holding an extra reference for
425 * transmit completion counting */
426 skb = alloc_skb(sizeof(state->payload), GFP_KERNEL);
429 state->skbs[i] = skb;
432 /* Copy the payload in, incrementing the source address to
433 * exercise the rss vectors */
434 payload = skb_put(skb, sizeof(state->payload));
435 memcpy(payload, &state->payload, sizeof(state->payload));
436 payload->ip.saddr = htonl(INADDR_LOOPBACK | (i << 2));
438 /* Ensure everything we've written is visible to the
439 * interrupt handler. */
442 netif_tx_lock_bh(efx->net_dev);
443 rc = efx_enqueue_skb(tx_queue, skb);
444 netif_tx_unlock_bh(efx->net_dev);
446 if (rc != NETDEV_TX_OK) {
447 netif_err(efx, drv, efx->net_dev,
448 "TX queue %d could not transmit packet %d of "
449 "%d in %s loopback test\n", tx_queue->queue,
450 i + 1, state->packet_count,
453 /* Defer cleaning up the other skbs for the caller */
462 static int efx_poll_loopback(struct efx_nic *efx)
464 struct efx_loopback_state *state = efx->loopback_selftest;
466 return atomic_read(&state->rx_good) == state->packet_count;
469 static int efx_end_loopback(struct efx_tx_queue *tx_queue,
470 struct efx_loopback_self_tests *lb_tests)
472 struct efx_nic *efx = tx_queue->efx;
473 struct efx_loopback_state *state = efx->loopback_selftest;
475 int tx_done = 0, rx_good, rx_bad;
478 netif_tx_lock_bh(efx->net_dev);
480 /* Count the number of tx completions, and decrement the refcnt. Any
481 * skbs not already completed will be free'd when the queue is flushed */
482 for (i = 0; i < state->packet_count; i++) {
483 skb = state->skbs[i];
484 if (skb && !skb_shared(skb))
489 netif_tx_unlock_bh(efx->net_dev);
491 /* Check TX completion and received packet counts */
492 rx_good = atomic_read(&state->rx_good);
493 rx_bad = atomic_read(&state->rx_bad);
494 if (tx_done != state->packet_count) {
495 /* Don't free the skbs; they will be picked up on TX
496 * overflow or channel teardown.
498 netif_err(efx, drv, efx->net_dev,
499 "TX queue %d saw only %d out of an expected %d "
500 "TX completion events in %s loopback test\n",
501 tx_queue->queue, tx_done, state->packet_count,
504 /* Allow to fall through so we see the RX errors as well */
507 /* We may always be up to a flush away from our desired packet total */
508 if (rx_good != state->packet_count) {
509 netif_dbg(efx, drv, efx->net_dev,
510 "TX queue %d saw only %d out of an expected %d "
511 "received packets in %s loopback test\n",
512 tx_queue->queue, rx_good, state->packet_count,
518 /* Update loopback test structure */
519 lb_tests->tx_sent[tx_queue->queue] += state->packet_count;
520 lb_tests->tx_done[tx_queue->queue] += tx_done;
521 lb_tests->rx_good += rx_good;
522 lb_tests->rx_bad += rx_bad;
528 efx_test_loopback(struct efx_tx_queue *tx_queue,
529 struct efx_loopback_self_tests *lb_tests)
531 struct efx_nic *efx = tx_queue->efx;
532 struct efx_loopback_state *state = efx->loopback_selftest;
533 int i, begin_rc, end_rc;
535 for (i = 0; i < 3; i++) {
536 /* Determine how many packets to send */
537 state->packet_count = efx->txq_entries / 3;
538 state->packet_count = min(1 << (i << 2), state->packet_count);
539 state->skbs = kcalloc(state->packet_count,
540 sizeof(state->skbs[0]), GFP_KERNEL);
543 state->flush = false;
545 netif_dbg(efx, drv, efx->net_dev,
546 "TX queue %d testing %s loopback with %d packets\n",
547 tx_queue->queue, LOOPBACK_MODE(efx),
548 state->packet_count);
550 efx_iterate_state(efx);
551 begin_rc = efx_begin_loopback(tx_queue);
553 /* This will normally complete very quickly, but be
554 * prepared to wait much longer. */
556 if (!efx_poll_loopback(efx)) {
557 msleep(LOOPBACK_TIMEOUT_MS);
558 efx_poll_loopback(efx);
561 end_rc = efx_end_loopback(tx_queue, lb_tests);
564 if (begin_rc || end_rc) {
565 /* Wait a while to ensure there are no packets
566 * floating around after a failure. */
567 schedule_timeout_uninterruptible(HZ / 10);
568 return begin_rc ? begin_rc : end_rc;
572 netif_dbg(efx, drv, efx->net_dev,
573 "TX queue %d passed %s loopback test with a burst length "
574 "of %d packets\n", tx_queue->queue, LOOPBACK_MODE(efx),
575 state->packet_count);
580 /* Wait for link up. On Falcon, we would prefer to rely on efx_monitor, but
581 * any contention on the mac lock (via e.g. efx_mac_mcast_work) causes it
582 * to delay and retry. Therefore, it's safer to just poll directly. Wait
583 * for link up and any faults to dissipate. */
584 static int efx_wait_for_link(struct efx_nic *efx)
586 struct efx_link_state *link_state = &efx->link_state;
587 int count, link_up_count = 0;
590 for (count = 0; count < 40; count++) {
591 schedule_timeout_uninterruptible(HZ / 10);
593 if (efx->type->monitor != NULL) {
594 mutex_lock(&efx->mac_lock);
595 efx->type->monitor(efx);
596 mutex_unlock(&efx->mac_lock);
599 mutex_lock(&efx->mac_lock);
600 link_up = link_state->up;
602 link_up = !efx->type->check_mac_fault(efx);
603 mutex_unlock(&efx->mac_lock);
606 if (++link_up_count == 2)
616 static int efx_test_loopbacks(struct efx_nic *efx, struct efx_self_tests *tests,
617 unsigned int loopback_modes)
619 enum efx_loopback_mode mode;
620 struct efx_loopback_state *state;
621 struct efx_channel *channel =
622 efx_get_channel(efx, efx->tx_channel_offset);
623 struct efx_tx_queue *tx_queue;
626 /* Set the port loopback_selftest member. From this point on
627 * all received packets will be dropped. Mark the state as
628 * "flushing" so all inflight packets are dropped */
629 state = kzalloc(sizeof(*state), GFP_KERNEL);
632 BUG_ON(efx->loopback_selftest);
634 efx->loopback_selftest = state;
636 /* Test all supported loopback modes */
637 for (mode = LOOPBACK_NONE; mode <= LOOPBACK_TEST_MAX; mode++) {
638 if (!(loopback_modes & (1 << mode)))
641 /* Move the port into the specified loopback mode. */
643 mutex_lock(&efx->mac_lock);
644 efx->loopback_mode = mode;
645 rc = __efx_reconfigure_port(efx);
646 mutex_unlock(&efx->mac_lock);
648 netif_err(efx, drv, efx->net_dev,
649 "unable to move into %s loopback\n",
654 rc = efx_wait_for_link(efx);
656 netif_err(efx, drv, efx->net_dev,
657 "loopback %s never came up\n",
662 /* Test all enabled types of TX queue */
663 efx_for_each_channel_tx_queue(tx_queue, channel) {
664 state->offload_csum = (tx_queue->queue &
665 EFX_TXQ_TYPE_OFFLOAD);
666 rc = efx_test_loopback(tx_queue,
667 &tests->loopback[mode]);
674 /* Remove the flush. The caller will remove the loopback setting */
676 efx->loopback_selftest = NULL;
686 /**************************************************************************
690 *************************************************************************/
692 int efx_selftest(struct efx_nic *efx, struct efx_self_tests *tests,
695 enum efx_loopback_mode loopback_mode = efx->loopback_mode;
696 int phy_mode = efx->phy_mode;
697 int rc_test = 0, rc_reset, rc;
699 efx_selftest_async_cancel(efx);
701 /* Online (i.e. non-disruptive) testing
702 * This checks interrupt generation, event delivery and PHY presence. */
704 rc = efx_test_phy_alive(efx, tests);
708 rc = efx_test_nvram(efx, tests);
712 rc = efx_test_interrupts(efx, tests);
716 rc = efx_test_eventq_irq(efx, tests);
723 if (!(flags & ETH_TEST_FL_OFFLINE))
724 return efx_test_phy(efx, tests, flags);
726 /* Offline (i.e. disruptive) testing
727 * This checks MAC and PHY loopback on the specified port. */
729 /* Detach the device so the kernel doesn't transmit during the
730 * loopback test and the watchdog timeout doesn't fire.
732 efx_device_detach_sync(efx);
734 if (efx->type->test_chip) {
735 rc_reset = efx->type->test_chip(efx, tests);
737 netif_err(efx, hw, efx->net_dev,
738 "Unable to recover from chip test\n");
739 efx_schedule_reset(efx, RESET_TYPE_DISABLE);
743 if ((tests->memory < 0 || tests->registers < 0) && !rc_test)
747 /* Ensure that the phy is powered and out of loopback
748 * for the bist and loopback tests */
749 mutex_lock(&efx->mac_lock);
750 efx->phy_mode &= ~PHY_MODE_LOW_POWER;
751 efx->loopback_mode = LOOPBACK_NONE;
752 __efx_reconfigure_port(efx);
753 mutex_unlock(&efx->mac_lock);
755 rc = efx_test_phy(efx, tests, flags);
759 rc = efx_test_loopbacks(efx, tests, efx->loopback_modes);
763 /* restore the PHY to the previous state */
764 mutex_lock(&efx->mac_lock);
765 efx->phy_mode = phy_mode;
766 efx->loopback_mode = loopback_mode;
767 __efx_reconfigure_port(efx);
768 mutex_unlock(&efx->mac_lock);
770 efx_device_attach_if_not_resetting(efx);
775 void efx_selftest_async_start(struct efx_nic *efx)
777 struct efx_channel *channel;
779 efx_for_each_channel(channel, efx)
780 efx_nic_event_test_start(channel);
781 schedule_delayed_work(&efx->selftest_work, IRQ_TIMEOUT);
784 void efx_selftest_async_cancel(struct efx_nic *efx)
786 cancel_delayed_work_sync(&efx->selftest_work);
789 void efx_selftest_async_work(struct work_struct *data)
791 struct efx_nic *efx = container_of(data, struct efx_nic,
793 struct efx_channel *channel;
796 efx_for_each_channel(channel, efx) {
797 cpu = efx_nic_event_test_irq_cpu(channel);
799 netif_err(efx, ifup, efx->net_dev,
800 "channel %d failed to trigger an interrupt\n",
803 netif_dbg(efx, ifup, efx->net_dev,
804 "channel %d triggered interrupt on CPU %d\n",
805 channel->channel, cpu);