1 /****************************************************************************
2 * Driver for Solarflare Solarstorm network controllers and boards
3 * Copyright 2005-2006 Fen Systems Ltd.
4 * Copyright 2006-2010 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 unsigned char payload_source[ETH_ALEN] = {
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);
117 tests->nvram = rc ? -1 : 1;
123 /**************************************************************************
125 * Interrupt and event queue testing
127 **************************************************************************/
129 /* Test generation and receipt of interrupts */
130 static int efx_test_interrupts(struct efx_nic *efx,
131 struct efx_self_tests *tests)
133 unsigned long timeout, wait;
136 netif_dbg(efx, drv, efx->net_dev, "testing interrupts\n");
137 tests->interrupt = -1;
139 efx_nic_irq_test_start(efx);
140 timeout = jiffies + IRQ_TIMEOUT;
143 /* Wait for arrival of test interrupt. */
144 netif_dbg(efx, drv, efx->net_dev, "waiting for test interrupt\n");
146 schedule_timeout_uninterruptible(wait);
147 cpu = efx_nic_irq_test_irq_cpu(efx);
151 } while (time_before(jiffies, timeout));
153 netif_err(efx, drv, efx->net_dev, "timed out waiting for interrupt\n");
157 netif_dbg(efx, drv, efx->net_dev, "%s test interrupt seen on CPU%d\n",
159 tests->interrupt = 1;
163 /* Test generation and receipt of interrupting events */
164 static int efx_test_eventq_irq(struct efx_nic *efx,
165 struct efx_self_tests *tests)
167 struct efx_channel *channel;
168 unsigned int read_ptr[EFX_MAX_CHANNELS];
169 unsigned long napi_ran = 0, dma_pend = 0, int_pend = 0;
170 unsigned long timeout, wait;
172 BUILD_BUG_ON(EFX_MAX_CHANNELS > BITS_PER_LONG);
174 efx_for_each_channel(channel, efx) {
175 read_ptr[channel->channel] = channel->eventq_read_ptr;
176 set_bit(channel->channel, &dma_pend);
177 set_bit(channel->channel, &int_pend);
178 efx_nic_event_test_start(channel);
181 timeout = jiffies + IRQ_TIMEOUT;
184 /* Wait for arrival of interrupts. NAPI processing may or may
185 * not complete in time, but we can cope in any case.
188 schedule_timeout_uninterruptible(wait);
190 efx_for_each_channel(channel, efx) {
191 napi_disable(&channel->napi_str);
192 if (channel->eventq_read_ptr !=
193 read_ptr[channel->channel]) {
194 set_bit(channel->channel, &napi_ran);
195 clear_bit(channel->channel, &dma_pend);
196 clear_bit(channel->channel, &int_pend);
198 if (efx_nic_event_present(channel))
199 clear_bit(channel->channel, &dma_pend);
200 if (efx_nic_event_test_irq_cpu(channel) >= 0)
201 clear_bit(channel->channel, &int_pend);
203 napi_enable(&channel->napi_str);
204 efx_nic_eventq_read_ack(channel);
208 } while ((dma_pend || int_pend) && time_before(jiffies, timeout));
210 efx_for_each_channel(channel, efx) {
211 bool dma_seen = !test_bit(channel->channel, &dma_pend);
212 bool int_seen = !test_bit(channel->channel, &int_pend);
214 tests->eventq_dma[channel->channel] = dma_seen ? 1 : -1;
215 tests->eventq_int[channel->channel] = int_seen ? 1 : -1;
217 if (dma_seen && int_seen) {
218 netif_dbg(efx, drv, efx->net_dev,
219 "channel %d event queue passed (with%s NAPI)\n",
221 test_bit(channel->channel, &napi_ran) ?
224 /* Report failure and whether either interrupt or DMA
227 netif_err(efx, drv, efx->net_dev,
228 "channel %d timed out waiting for event queue\n",
231 netif_err(efx, drv, efx->net_dev,
232 "channel %d saw interrupt "
233 "during event queue test\n",
236 netif_err(efx, drv, efx->net_dev,
237 "channel %d event was generated, but "
238 "failed to trigger an interrupt\n",
243 return (dma_pend || int_pend) ? -ETIMEDOUT : 0;
246 static int efx_test_phy(struct efx_nic *efx, struct efx_self_tests *tests,
251 if (!efx->phy_op->run_tests)
254 mutex_lock(&efx->mac_lock);
255 rc = efx->phy_op->run_tests(efx, tests->phy_ext, flags);
256 mutex_unlock(&efx->mac_lock);
260 /**************************************************************************
263 * NB Only one loopback test can be executing concurrently.
265 **************************************************************************/
267 /* Loopback test RX callback
268 * This is called for each received packet during loopback testing.
270 void efx_loopback_rx_packet(struct efx_nic *efx,
271 const char *buf_ptr, int pkt_len)
273 struct efx_loopback_state *state = efx->loopback_selftest;
274 struct efx_loopback_payload *received;
275 struct efx_loopback_payload *payload;
279 /* If we are just flushing, then drop the packet */
280 if ((state == NULL) || state->flush)
283 payload = &state->payload;
285 received = (struct efx_loopback_payload *) buf_ptr;
286 received->ip.saddr = payload->ip.saddr;
287 if (state->offload_csum)
288 received->ip.check = payload->ip.check;
290 /* Check that header exists */
291 if (pkt_len < sizeof(received->header)) {
292 netif_err(efx, drv, efx->net_dev,
293 "saw runt RX packet (length %d) in %s loopback "
294 "test\n", pkt_len, LOOPBACK_MODE(efx));
298 /* Check that the ethernet header exists */
299 if (memcmp(&received->header, &payload->header, ETH_HLEN) != 0) {
300 netif_err(efx, drv, efx->net_dev,
301 "saw non-loopback RX packet in %s loopback test\n",
306 /* Check packet length */
307 if (pkt_len != sizeof(*payload)) {
308 netif_err(efx, drv, efx->net_dev,
309 "saw incorrect RX packet length %d (wanted %d) in "
310 "%s loopback test\n", pkt_len, (int)sizeof(*payload),
315 /* Check that IP header matches */
316 if (memcmp(&received->ip, &payload->ip, sizeof(payload->ip)) != 0) {
317 netif_err(efx, drv, efx->net_dev,
318 "saw corrupted IP header in %s loopback test\n",
323 /* Check that msg and padding matches */
324 if (memcmp(&received->msg, &payload->msg, sizeof(received->msg)) != 0) {
325 netif_err(efx, drv, efx->net_dev,
326 "saw corrupted RX packet in %s loopback test\n",
331 /* Check that iteration matches */
332 if (received->iteration != payload->iteration) {
333 netif_err(efx, drv, efx->net_dev,
334 "saw RX packet from iteration %d (wanted %d) in "
335 "%s loopback test\n", ntohs(received->iteration),
336 ntohs(payload->iteration), LOOPBACK_MODE(efx));
340 /* Increase correct RX count */
341 netif_vdbg(efx, drv, efx->net_dev,
342 "got loopback RX in %s loopback test\n", LOOPBACK_MODE(efx));
344 atomic_inc(&state->rx_good);
349 if (atomic_read(&state->rx_bad) == 0) {
350 netif_err(efx, drv, efx->net_dev, "received packet:\n");
351 print_hex_dump(KERN_ERR, "", DUMP_PREFIX_OFFSET, 0x10, 1,
352 buf_ptr, pkt_len, 0);
353 netif_err(efx, drv, efx->net_dev, "expected packet:\n");
354 print_hex_dump(KERN_ERR, "", DUMP_PREFIX_OFFSET, 0x10, 1,
355 &state->payload, sizeof(state->payload), 0);
358 atomic_inc(&state->rx_bad);
361 /* Initialise an efx_selftest_state for a new iteration */
362 static void efx_iterate_state(struct efx_nic *efx)
364 struct efx_loopback_state *state = efx->loopback_selftest;
365 struct net_device *net_dev = efx->net_dev;
366 struct efx_loopback_payload *payload = &state->payload;
368 /* Initialise the layerII header */
369 memcpy(&payload->header.h_dest, net_dev->dev_addr, ETH_ALEN);
370 memcpy(&payload->header.h_source, &payload_source, ETH_ALEN);
371 payload->header.h_proto = htons(ETH_P_IP);
373 /* saddr set later and used as incrementing count */
374 payload->ip.daddr = htonl(INADDR_LOOPBACK);
376 payload->ip.check = (__force __sum16) htons(0xdead);
377 payload->ip.tot_len = htons(sizeof(*payload) - sizeof(struct ethhdr));
378 payload->ip.version = IPVERSION;
379 payload->ip.protocol = IPPROTO_UDP;
381 /* Initialise udp header */
382 payload->udp.source = 0;
383 payload->udp.len = htons(sizeof(*payload) - sizeof(struct ethhdr) -
384 sizeof(struct iphdr));
385 payload->udp.check = 0; /* checksum ignored */
387 /* Fill out payload */
388 payload->iteration = htons(ntohs(payload->iteration) + 1);
389 memcpy(&payload->msg, payload_msg, sizeof(payload_msg));
391 /* Fill out remaining state members */
392 atomic_set(&state->rx_good, 0);
393 atomic_set(&state->rx_bad, 0);
397 static int efx_begin_loopback(struct efx_tx_queue *tx_queue)
399 struct efx_nic *efx = tx_queue->efx;
400 struct efx_loopback_state *state = efx->loopback_selftest;
401 struct efx_loopback_payload *payload;
406 /* Transmit N copies of buffer */
407 for (i = 0; i < state->packet_count; i++) {
408 /* Allocate an skb, holding an extra reference for
409 * transmit completion counting */
410 skb = alloc_skb(sizeof(state->payload), GFP_KERNEL);
413 state->skbs[i] = skb;
416 /* Copy the payload in, incrementing the source address to
417 * exercise the rss vectors */
418 payload = ((struct efx_loopback_payload *)
419 skb_put(skb, sizeof(state->payload)));
420 memcpy(payload, &state->payload, sizeof(state->payload));
421 payload->ip.saddr = htonl(INADDR_LOOPBACK | (i << 2));
423 /* Ensure everything we've written is visible to the
424 * interrupt handler. */
427 netif_tx_lock_bh(efx->net_dev);
428 rc = efx_enqueue_skb(tx_queue, skb);
429 netif_tx_unlock_bh(efx->net_dev);
431 if (rc != NETDEV_TX_OK) {
432 netif_err(efx, drv, efx->net_dev,
433 "TX queue %d could not transmit packet %d of "
434 "%d in %s loopback test\n", tx_queue->queue,
435 i + 1, state->packet_count,
438 /* Defer cleaning up the other skbs for the caller */
447 static int efx_poll_loopback(struct efx_nic *efx)
449 struct efx_loopback_state *state = efx->loopback_selftest;
450 struct efx_channel *channel;
452 /* NAPI polling is not enabled, so process channels
454 efx_for_each_channel(channel, efx) {
455 if (channel->work_pending)
456 efx_process_channel_now(channel);
458 return atomic_read(&state->rx_good) == state->packet_count;
461 static int efx_end_loopback(struct efx_tx_queue *tx_queue,
462 struct efx_loopback_self_tests *lb_tests)
464 struct efx_nic *efx = tx_queue->efx;
465 struct efx_loopback_state *state = efx->loopback_selftest;
467 int tx_done = 0, rx_good, rx_bad;
470 netif_tx_lock_bh(efx->net_dev);
472 /* Count the number of tx completions, and decrement the refcnt. Any
473 * skbs not already completed will be free'd when the queue is flushed */
474 for (i = 0; i < state->packet_count; i++) {
475 skb = state->skbs[i];
476 if (skb && !skb_shared(skb))
481 netif_tx_unlock_bh(efx->net_dev);
483 /* Check TX completion and received packet counts */
484 rx_good = atomic_read(&state->rx_good);
485 rx_bad = atomic_read(&state->rx_bad);
486 if (tx_done != state->packet_count) {
487 /* Don't free the skbs; they will be picked up on TX
488 * overflow or channel teardown.
490 netif_err(efx, drv, efx->net_dev,
491 "TX queue %d saw only %d out of an expected %d "
492 "TX completion events in %s loopback test\n",
493 tx_queue->queue, tx_done, state->packet_count,
496 /* Allow to fall through so we see the RX errors as well */
499 /* We may always be up to a flush away from our desired packet total */
500 if (rx_good != state->packet_count) {
501 netif_dbg(efx, drv, efx->net_dev,
502 "TX queue %d saw only %d out of an expected %d "
503 "received packets in %s loopback test\n",
504 tx_queue->queue, rx_good, state->packet_count,
510 /* Update loopback test structure */
511 lb_tests->tx_sent[tx_queue->queue] += state->packet_count;
512 lb_tests->tx_done[tx_queue->queue] += tx_done;
513 lb_tests->rx_good += rx_good;
514 lb_tests->rx_bad += rx_bad;
520 efx_test_loopback(struct efx_tx_queue *tx_queue,
521 struct efx_loopback_self_tests *lb_tests)
523 struct efx_nic *efx = tx_queue->efx;
524 struct efx_loopback_state *state = efx->loopback_selftest;
525 int i, begin_rc, end_rc;
527 for (i = 0; i < 3; i++) {
528 /* Determine how many packets to send */
529 state->packet_count = efx->txq_entries / 3;
530 state->packet_count = min(1 << (i << 2), state->packet_count);
531 state->skbs = kcalloc(state->packet_count,
532 sizeof(state->skbs[0]), GFP_KERNEL);
535 state->flush = false;
537 netif_dbg(efx, drv, efx->net_dev,
538 "TX queue %d testing %s loopback with %d packets\n",
539 tx_queue->queue, LOOPBACK_MODE(efx),
540 state->packet_count);
542 efx_iterate_state(efx);
543 begin_rc = efx_begin_loopback(tx_queue);
545 /* This will normally complete very quickly, but be
546 * prepared to wait much longer. */
548 if (!efx_poll_loopback(efx)) {
549 msleep(LOOPBACK_TIMEOUT_MS);
550 efx_poll_loopback(efx);
553 end_rc = efx_end_loopback(tx_queue, lb_tests);
556 if (begin_rc || end_rc) {
557 /* Wait a while to ensure there are no packets
558 * floating around after a failure. */
559 schedule_timeout_uninterruptible(HZ / 10);
560 return begin_rc ? begin_rc : end_rc;
564 netif_dbg(efx, drv, efx->net_dev,
565 "TX queue %d passed %s loopback test with a burst length "
566 "of %d packets\n", tx_queue->queue, LOOPBACK_MODE(efx),
567 state->packet_count);
572 /* Wait for link up. On Falcon, we would prefer to rely on efx_monitor, but
573 * any contention on the mac lock (via e.g. efx_mac_mcast_work) causes it
574 * to delay and retry. Therefore, it's safer to just poll directly. Wait
575 * for link up and any faults to dissipate. */
576 static int efx_wait_for_link(struct efx_nic *efx)
578 struct efx_link_state *link_state = &efx->link_state;
579 int count, link_up_count = 0;
582 for (count = 0; count < 40; count++) {
583 schedule_timeout_uninterruptible(HZ / 10);
585 if (efx->type->monitor != NULL) {
586 mutex_lock(&efx->mac_lock);
587 efx->type->monitor(efx);
588 mutex_unlock(&efx->mac_lock);
590 struct efx_channel *channel = efx_get_channel(efx, 0);
591 if (channel->work_pending)
592 efx_process_channel_now(channel);
595 mutex_lock(&efx->mac_lock);
596 link_up = link_state->up;
598 link_up = !efx->type->check_mac_fault(efx);
599 mutex_unlock(&efx->mac_lock);
602 if (++link_up_count == 2)
612 static int efx_test_loopbacks(struct efx_nic *efx, struct efx_self_tests *tests,
613 unsigned int loopback_modes)
615 enum efx_loopback_mode mode;
616 struct efx_loopback_state *state;
617 struct efx_channel *channel =
618 efx_get_channel(efx, efx->tx_channel_offset);
619 struct efx_tx_queue *tx_queue;
622 /* Set the port loopback_selftest member. From this point on
623 * all received packets will be dropped. Mark the state as
624 * "flushing" so all inflight packets are dropped */
625 state = kzalloc(sizeof(*state), GFP_KERNEL);
628 BUG_ON(efx->loopback_selftest);
630 efx->loopback_selftest = state;
632 /* Test all supported loopback modes */
633 for (mode = LOOPBACK_NONE; mode <= LOOPBACK_TEST_MAX; mode++) {
634 if (!(loopback_modes & (1 << mode)))
637 /* Move the port into the specified loopback mode. */
639 mutex_lock(&efx->mac_lock);
640 efx->loopback_mode = mode;
641 rc = __efx_reconfigure_port(efx);
642 mutex_unlock(&efx->mac_lock);
644 netif_err(efx, drv, efx->net_dev,
645 "unable to move into %s loopback\n",
650 rc = efx_wait_for_link(efx);
652 netif_err(efx, drv, efx->net_dev,
653 "loopback %s never came up\n",
658 /* Test all enabled types of TX queue */
659 efx_for_each_channel_tx_queue(tx_queue, channel) {
660 state->offload_csum = (tx_queue->queue &
661 EFX_TXQ_TYPE_OFFLOAD);
662 rc = efx_test_loopback(tx_queue,
663 &tests->loopback[mode]);
670 /* Remove the flush. The caller will remove the loopback setting */
672 efx->loopback_selftest = NULL;
679 /**************************************************************************
683 *************************************************************************/
685 int efx_selftest(struct efx_nic *efx, struct efx_self_tests *tests,
688 enum efx_loopback_mode loopback_mode = efx->loopback_mode;
689 int phy_mode = efx->phy_mode;
690 int rc_test = 0, rc_reset, rc;
692 efx_selftest_async_cancel(efx);
694 /* Online (i.e. non-disruptive) testing
695 * This checks interrupt generation, event delivery and PHY presence. */
697 rc = efx_test_phy_alive(efx, tests);
701 rc = efx_test_nvram(efx, tests);
705 rc = efx_test_interrupts(efx, tests);
709 rc = efx_test_eventq_irq(efx, tests);
716 if (!(flags & ETH_TEST_FL_OFFLINE))
717 return efx_test_phy(efx, tests, flags);
719 /* Offline (i.e. disruptive) testing
720 * This checks MAC and PHY loopback on the specified port. */
722 /* Detach the device so the kernel doesn't transmit during the
723 * loopback test and the watchdog timeout doesn't fire.
725 efx_device_detach_sync(efx);
727 if (efx->type->test_chip) {
728 rc_reset = efx->type->test_chip(efx, tests);
730 netif_err(efx, hw, efx->net_dev,
731 "Unable to recover from chip test\n");
732 efx_schedule_reset(efx, RESET_TYPE_DISABLE);
736 if ((tests->registers < 0) && !rc_test)
740 /* Ensure that the phy is powered and out of loopback
741 * for the bist and loopback tests */
742 mutex_lock(&efx->mac_lock);
743 efx->phy_mode &= ~PHY_MODE_LOW_POWER;
744 efx->loopback_mode = LOOPBACK_NONE;
745 __efx_reconfigure_port(efx);
746 mutex_unlock(&efx->mac_lock);
748 rc = efx_test_phy(efx, tests, flags);
752 rc = efx_test_loopbacks(efx, tests, efx->loopback_modes);
756 /* restore the PHY to the previous state */
757 mutex_lock(&efx->mac_lock);
758 efx->phy_mode = phy_mode;
759 efx->loopback_mode = loopback_mode;
760 __efx_reconfigure_port(efx);
761 mutex_unlock(&efx->mac_lock);
763 netif_device_attach(efx->net_dev);
768 void efx_selftest_async_start(struct efx_nic *efx)
770 struct efx_channel *channel;
772 efx_for_each_channel(channel, efx)
773 efx_nic_event_test_start(channel);
774 schedule_delayed_work(&efx->selftest_work, IRQ_TIMEOUT);
777 void efx_selftest_async_cancel(struct efx_nic *efx)
779 cancel_delayed_work_sync(&efx->selftest_work);
782 void efx_selftest_async_work(struct work_struct *data)
784 struct efx_nic *efx = container_of(data, struct efx_nic,
786 struct efx_channel *channel;
789 efx_for_each_channel(channel, efx) {
790 cpu = efx_nic_event_test_irq_cpu(channel);
792 netif_err(efx, ifup, efx->net_dev,
793 "channel %d failed to trigger an interrupt\n",
796 netif_dbg(efx, ifup, efx->net_dev,
797 "channel %d triggered interrupt on CPU %d\n",
798 channel->channel, cpu);