sfc:On MCDI timeout, issue an FLR (and mark MCDI to fail-fast)

When an MCDI command times out (whether or not we find it
completed when we poll), call efx_mcdi_abandon(), which tells
all subsequent MCDI calls to fail-fast, and queues up an FLR.

Because an FLR doesn't lead to receiving any reboot even from
the MC (unlike most other types of reset), we have to call
efx_ef10_reset_mc_allocations.
In efx_start_all(), if a reset (of any kind) is pending, we
bail out.
Without this, attempts to reconfigure (e.g. change mtu) can
cause driver/mc state inconsistency if the first MCDI call
triggers an FLR.

For similar reasons, on EF10, in
efx_reset_down(method=RESET_TYPE_MCDI_TIMEOUT), set the number
of active queues to zero before calling efx_stop_all().
And, on farch, in efx_reset_up(method=RESET_TYPE_MCDI_TIMEOUT),
set active_queues and flushes pending & outstanding to zero.

efx_mcdi_mode_{poll,event}() should not take us out of fail-fast
 mode. Instead, this is done by efx_mcdi_reset() after the FLR
completes.

The new FLR reset_type RESET_TYPE_MCDI_TIMEOUT doesn't really
fit into the hierarchy of reset 'scopes' whereby efx_reset()
decides some resets subsume others.  Thus, it uses separate logic.

Also, fixed up some inconsistency around RESET_TYPE_MC_BIST,
which was in the wrong place in that hierarchy.

Signed-off-by: Shradha Shah <sshah@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

drivers/net/ethernet/sfc/ef10.c

@@ -738,8 +738,11 @@ static int efx_ef10_reset(struct efx_nic *efx, enum reset_type reset_type)
 	/* If it was a port reset, trigger reallocation of MC resources.
 	 * Note that on an MC reset nothing needs to be done now because we'll
 	 * detect the MC reset later and handle it then.
+	 * For an FLR, we never get an MC reset event, but the MC has reset all
+	 * resources assigned to us, so we have to trigger reallocation now.
 	 */
-	if (reset_type == RESET_TYPE_ALL && !rc)
+	if ((reset_type == RESET_TYPE_ALL ||
+	     reset_type == RESET_TYPE_MCDI_TIMEOUT) && !rc)
 		efx_ef10_reset_mc_allocations(efx);
 	return rc;
 }
@@ -2141,6 +2144,11 @@ static int efx_ef10_fini_dmaq(struct efx_nic *efx)
 	return 0;
 }
 
+static void efx_ef10_prepare_flr(struct efx_nic *efx)
+{
+	atomic_set(&efx->active_queues, 0);
+}
+
 static bool efx_ef10_filter_equal(const struct efx_filter_spec *left,
 				  const struct efx_filter_spec *right)
 {
@@ -3603,6 +3611,8 @@ const struct efx_nic_type efx_hunt_a0_nic_type = {
 	.probe_port = efx_mcdi_port_probe,
 	.remove_port = efx_mcdi_port_remove,
 	.fini_dmaq = efx_ef10_fini_dmaq,
+	.prepare_flr = efx_ef10_prepare_flr,
+	.finish_flr = efx_port_dummy_op_void,
 	.describe_stats = efx_ef10_describe_stats,
 	.update_stats = efx_ef10_update_stats,
 	.start_stats = efx_mcdi_mac_start_stats,

drivers/net/ethernet/sfc/efx.c

@@ -76,14 +76,15 @@ const char *const efx_reset_type_names[] = {
 	[RESET_TYPE_RECOVER_OR_ALL]     = "RECOVER_OR_ALL",
 	[RESET_TYPE_WORLD]              = "WORLD",
 	[RESET_TYPE_RECOVER_OR_DISABLE] = "RECOVER_OR_DISABLE",
+	[RESET_TYPE_MC_BIST]		= "MC_BIST",
 	[RESET_TYPE_DISABLE]            = "DISABLE",
 	[RESET_TYPE_TX_WATCHDOG]        = "TX_WATCHDOG",
 	[RESET_TYPE_INT_ERROR]          = "INT_ERROR",
 	[RESET_TYPE_RX_RECOVERY]        = "RX_RECOVERY",
 	[RESET_TYPE_DMA_ERROR]          = "DMA_ERROR",
 	[RESET_TYPE_TX_SKIP]            = "TX_SKIP",
 	[RESET_TYPE_MC_FAILURE]         = "MC_FAILURE",
-	[RESET_TYPE_MC_BIST]		= "MC_BIST",
+	[RESET_TYPE_MCDI_TIMEOUT]	= "MCDI_TIMEOUT (FLR)",
 };
 
 /* Reset workqueue. If any NIC has a hardware failure then a reset will be
@@ -1739,7 +1740,8 @@ static void efx_start_all(struct efx_nic *efx)
 
 	/* Check that it is appropriate to restart the interface. All
 	 * of these flags are safe to read under just the rtnl lock */
-	if (efx->port_enabled || !netif_running(efx->net_dev))
+	if (efx->port_enabled || !netif_running(efx->net_dev) ||
+	    efx->reset_pending)
 		return;
 
 	efx_start_port(efx);
@@ -2334,6 +2336,9 @@ void efx_reset_down(struct efx_nic *efx, enum reset_type method)
 {
 	EFX_ASSERT_RESET_SERIALISED(efx);
 
+	if (method == RESET_TYPE_MCDI_TIMEOUT)
+		efx->type->prepare_flr(efx);
+
 	efx_stop_all(efx);
 	efx_disable_interrupts(efx);
 
@@ -2354,6 +2359,10 @@ int efx_reset_up(struct efx_nic *efx, enum reset_type method, bool ok)
 
 	EFX_ASSERT_RESET_SERIALISED(efx);
 
+	if (method == RESET_TYPE_MCDI_TIMEOUT)
+		efx->type->finish_flr(efx);
+
+	/* Ensure that SRAM is initialised even if we're disabling the device */
 	rc = efx->type->init(efx);
 	if (rc) {
 		netif_err(efx, drv, efx->net_dev, "failed to initialise NIC\n");
@@ -2417,7 +2426,10 @@ int efx_reset(struct efx_nic *efx, enum reset_type method)
 	/* Clear flags for the scopes we covered.  We assume the NIC and
 	 * driver are now quiescent so that there is no race here.
 	 */
-	efx->reset_pending &= -(1 << (method + 1));
+	if (method < RESET_TYPE_MAX_METHOD)
+		efx->reset_pending &= -(1 << (method + 1));
+	else /* it doesn't fit into the well-ordered scope hierarchy */
+		__clear_bit(method, &efx->reset_pending);
 
 	/* Reinitialise bus-mastering, which may have been turned off before
 	 * the reset was scheduled. This is still appropriate, even in the
@@ -2546,6 +2558,7 @@ void efx_schedule_reset(struct efx_nic *efx, enum reset_type type)
 	case RESET_TYPE_DISABLE:
 	case RESET_TYPE_RECOVER_OR_DISABLE:
 	case RESET_TYPE_MC_BIST:
+	case RESET_TYPE_MCDI_TIMEOUT:
 		method = type;
 		netif_dbg(efx, drv, efx->net_dev, "scheduling %s reset\n",
 			  RESET_TYPE(method));

drivers/net/ethernet/sfc/enum.h

@@ -143,29 +143,38 @@ enum efx_loopback_mode {
  * @RESET_TYPE_WORLD: Reset as much as possible
  * @RESET_TYPE_RECOVER_OR_DISABLE: Try to recover. Apply RESET_TYPE_DISABLE if
  * unsuccessful.
+ * @RESET_TYPE_MC_BIST: MC entering BIST mode.
  * @RESET_TYPE_DISABLE: Reset datapath, MAC and PHY; leave NIC disabled
  * @RESET_TYPE_TX_WATCHDOG: reset due to TX watchdog
  * @RESET_TYPE_INT_ERROR: reset due to internal error
  * @RESET_TYPE_RX_RECOVERY: reset to recover from RX datapath errors
  * @RESET_TYPE_DMA_ERROR: DMA error
  * @RESET_TYPE_TX_SKIP: hardware completed empty tx descriptors
  * @RESET_TYPE_MC_FAILURE: MC reboot/assertion
+ * @RESET_TYPE_MCDI_TIMEOUT: MCDI timeout.
  */
 enum reset_type {
-	RESET_TYPE_INVISIBLE = 0,
-	RESET_TYPE_RECOVER_OR_ALL = 1,
-	RESET_TYPE_ALL = 2,
-	RESET_TYPE_WORLD = 3,
-	RESET_TYPE_RECOVER_OR_DISABLE = 4,
-	RESET_TYPE_DISABLE = 5,
+	RESET_TYPE_INVISIBLE,
+	RESET_TYPE_RECOVER_OR_ALL,
+	RESET_TYPE_ALL,
+	RESET_TYPE_WORLD,
+	RESET_TYPE_RECOVER_OR_DISABLE,
+	RESET_TYPE_MC_BIST,
+	RESET_TYPE_DISABLE,
 	RESET_TYPE_MAX_METHOD,
 	RESET_TYPE_TX_WATCHDOG,
 	RESET_TYPE_INT_ERROR,
 	RESET_TYPE_RX_RECOVERY,
 	RESET_TYPE_DMA_ERROR,
 	RESET_TYPE_TX_SKIP,
 	RESET_TYPE_MC_FAILURE,
-	RESET_TYPE_MC_BIST,
+	/* RESET_TYPE_MCDI_TIMEOUT is actually a method, not just a reason, but
+	 * it doesn't fit the scope hierarchy (not well-ordered by inclusion).
+	 * We encode this by having its enum value be greater than
+	 * RESET_TYPE_MAX_METHOD. This also prevents issuing it with
+	 * efx_ioctl_reset.
+	 */
+	RESET_TYPE_MCDI_TIMEOUT,
 	RESET_TYPE_MAX,
 };
 

drivers/net/ethernet/sfc/falcon.c

@@ -2696,6 +2696,8 @@ const struct efx_nic_type falcon_a1_nic_type = {
 	.fini_dmaq = efx_farch_fini_dmaq,
 	.prepare_flush = falcon_prepare_flush,
 	.finish_flush = efx_port_dummy_op_void,
+	.prepare_flr = efx_port_dummy_op_void,
+	.finish_flr = efx_farch_finish_flr,
 	.describe_stats = falcon_describe_nic_stats,
 	.update_stats = falcon_update_nic_stats,
 	.start_stats = falcon_start_nic_stats,
@@ -2790,6 +2792,8 @@ const struct efx_nic_type falcon_b0_nic_type = {
 	.fini_dmaq = efx_farch_fini_dmaq,
 	.prepare_flush = falcon_prepare_flush,
 	.finish_flush = efx_port_dummy_op_void,
+	.prepare_flr = efx_port_dummy_op_void,
+	.finish_flr = efx_farch_finish_flr,
 	.describe_stats = falcon_describe_nic_stats,
 	.update_stats = falcon_update_nic_stats,
 	.start_stats = falcon_start_nic_stats,

drivers/net/ethernet/sfc/farch.c

@@ -741,6 +741,28 @@ int efx_farch_fini_dmaq(struct efx_nic *efx)
 	return rc;
 }
 
+/* Reset queue and flush accounting after FLR
+ *
+ * One possible cause of FLR recovery is that DMA may be failing (eg. if bus
+ * mastering was disabled), in which case we don't receive (RXQ) flush
+ * completion events.  This means that efx->rxq_flush_outstanding remained at 4
+ * after the FLR; also, efx->active_queues was non-zero (as no flush completion
+ * events were received, and we didn't go through efx_check_tx_flush_complete())
+ * If we don't fix this up, on the next call to efx_realloc_channels() we won't
+ * flush any RX queues because efx->rxq_flush_outstanding is at the limit of 4
+ * for batched flush requests; and the efx->active_queues gets messed up because
+ * we keep incrementing for the newly initialised queues, but it never went to
+ * zero previously.  Then we get a timeout every time we try to restart the
+ * queues, as it doesn't go back to zero when we should be flushing the queues.
+ */
+void efx_farch_finish_flr(struct efx_nic *efx)
+{
+	atomic_set(&efx->rxq_flush_pending, 0);
+	atomic_set(&efx->rxq_flush_outstanding, 0);
+	atomic_set(&efx->active_queues, 0);
+}
+
+
 /**************************************************************************
  *
  * Event queue processing

drivers/net/ethernet/sfc/mcdi.c

@@ -52,12 +52,7 @@ static void efx_mcdi_timeout_async(unsigned long context);
 static int efx_mcdi_drv_attach(struct efx_nic *efx, bool driver_operating,
 			       bool *was_attached_out);
 static bool efx_mcdi_poll_once(struct efx_nic *efx);
-
-static inline struct efx_mcdi_iface *efx_mcdi(struct efx_nic *efx)
-{
-	EFX_BUG_ON_PARANOID(!efx->mcdi);
-	return &efx->mcdi->iface;
-}
+static void efx_mcdi_abandon(struct efx_nic *efx);
 
 int efx_mcdi_init(struct efx_nic *efx)
 {
@@ -558,6 +553,8 @@ static int _efx_mcdi_rpc_finish(struct efx_nic *efx, unsigned cmd, size_t inlen,
 			rc = 0;
 		}
 
+		efx_mcdi_abandon(efx);
+
 		/* Close the race with efx_mcdi_ev_cpl() executing just too late
 		 * and completing a request we've just cancelled, by ensuring
 		 * that the seqno check therein fails.
@@ -672,6 +669,9 @@ int efx_mcdi_rpc_start(struct efx_nic *efx, unsigned cmd,
 	if (efx->mc_bist_for_other_fn)
 		return -ENETDOWN;
 
+	if (mcdi->mode == MCDI_MODE_FAIL)
+		return -ENETDOWN;
+
 	efx_mcdi_acquire_sync(mcdi);
 	efx_mcdi_send_request(efx, cmd, inbuf, inlen);
 	return 0;
@@ -812,7 +812,11 @@ void efx_mcdi_mode_poll(struct efx_nic *efx)
 		return;
 
 	mcdi = efx_mcdi(efx);
-	if (mcdi->mode == MCDI_MODE_POLL)
+	/* If already in polling mode, nothing to do.
+	 * If in fail-fast state, don't switch to polled completion.
+	 * FLR recovery will do that later.
+	 */
+	if (mcdi->mode == MCDI_MODE_POLL || mcdi->mode == MCDI_MODE_FAIL)
 		return;
 
 	/* We can switch from event completion to polled completion, because
@@ -841,8 +845,8 @@ void efx_mcdi_flush_async(struct efx_nic *efx)
 
 	mcdi = efx_mcdi(efx);
 
-	/* We must be in polling mode so no more requests can be queued */
-	BUG_ON(mcdi->mode != MCDI_MODE_POLL);
+	/* We must be in poll or fail mode so no more requests can be queued */
+	BUG_ON(mcdi->mode == MCDI_MODE_EVENTS);
 
 	del_timer_sync(&mcdi->async_timer);
 
@@ -875,8 +879,11 @@ void efx_mcdi_mode_event(struct efx_nic *efx)
 		return;
 
 	mcdi = efx_mcdi(efx);
-
-	if (mcdi->mode == MCDI_MODE_EVENTS)
+	/* If already in event completion mode, nothing to do.
+	 * If in fail-fast state, don't switch to event completion.  FLR
+	 * recovery will do that later.
+	 */
+	if (mcdi->mode == MCDI_MODE_EVENTS || mcdi->mode == MCDI_MODE_FAIL)
 		return;
 
 	/* We can't switch from polled to event completion in the middle of a
@@ -966,6 +973,19 @@ static void efx_mcdi_ev_bist(struct efx_nic *efx)
 	spin_unlock(&mcdi->iface_lock);
 }
 
+/* MCDI timeouts seen, so make all MCDI calls fail-fast and issue an FLR to try
+ * to recover.
+ */
+static void efx_mcdi_abandon(struct efx_nic *efx)
+{
+	struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
+
+	if (xchg(&mcdi->mode, MCDI_MODE_FAIL) == MCDI_MODE_FAIL)
+		return; /* it had already been done */
+	netif_dbg(efx, hw, efx->net_dev, "MCDI is timing out; trying to recover\n");
+	efx_schedule_reset(efx, RESET_TYPE_MCDI_TIMEOUT);
+}
+
 /* Called from  falcon_process_eventq for MCDI events */
 void efx_mcdi_process_event(struct efx_channel *channel,
 			    efx_qword_t *event)
@@ -1512,6 +1532,19 @@ int efx_mcdi_reset(struct efx_nic *efx, enum reset_type method)
 {
 	int rc;
 
+	/* If MCDI is down, we can't handle_assertion */
+	if (method == RESET_TYPE_MCDI_TIMEOUT) {
+		rc = pci_reset_function(efx->pci_dev);
+		if (rc)
+			return rc;
+		/* Re-enable polled MCDI completion */
+		if (efx->mcdi) {
+			struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
+			mcdi->mode = MCDI_MODE_POLL;
+		}
+		return 0;
+	}
+
 	/* Recover from a failed assertion pre-reset */
 	rc = efx_mcdi_handle_assertion(efx);
 	if (rc)

drivers/net/ethernet/sfc/mcdi.h

@@ -28,9 +28,16 @@ enum efx_mcdi_state {
 	MCDI_STATE_COMPLETED,
 };
 
+/**
+ * enum efx_mcdi_mode - MCDI transaction mode
+ * @MCDI_MODE_POLL: poll for MCDI completion, until timeout
+ * @MCDI_MODE_EVENTS: wait for an mcdi_event.  On timeout, poll once
+ * @MCDI_MODE_FAIL: we think MCDI is dead, so fail-fast all calls
+ */
 enum efx_mcdi_mode {
 	MCDI_MODE_POLL,
 	MCDI_MODE_EVENTS,
+	MCDI_MODE_FAIL,
 };
 
 /**
@@ -104,6 +111,12 @@ struct efx_mcdi_data {
 	u32 fn_flags;
 };
 
+static inline struct efx_mcdi_iface *efx_mcdi(struct efx_nic *efx)
+{
+	EFX_BUG_ON_PARANOID(!efx->mcdi);
+	return &efx->mcdi->iface;
+}
+
 #ifdef CONFIG_SFC_MCDI_MON
 static inline struct efx_mcdi_mon *efx_mcdi_mon(struct efx_nic *efx)
 {

drivers/net/ethernet/sfc/net_driver.h

@@ -972,6 +972,8 @@ struct efx_mtd_partition {
  *	(for Falcon architecture)
  * @finish_flush: Clean up after flushing the DMA queues (for Falcon
  *	architecture)
+ * @prepare_flr: Prepare for an FLR
+ * @finish_flr: Clean up after an FLR
  * @describe_stats: Describe statistics for ethtool
  * @update_stats: Update statistics not provided by event handling.
  *	Either argument may be %NULL.
@@ -1100,6 +1102,8 @@ struct efx_nic_type {
 	int (*fini_dmaq)(struct efx_nic *efx);
 	void (*prepare_flush)(struct efx_nic *efx);
 	void (*finish_flush)(struct efx_nic *efx);
+	void (*prepare_flr)(struct efx_nic *efx);
+	void (*finish_flr)(struct efx_nic *efx);
 	size_t (*describe_stats)(struct efx_nic *efx, u8 *names);
 	size_t (*update_stats)(struct efx_nic *efx, u64 *full_stats,
 			       struct rtnl_link_stats64 *core_stats);

drivers/net/ethernet/sfc/nic.h

@@ -757,6 +757,7 @@ static inline int efx_nic_irq_test_irq_cpu(struct efx_nic *efx)
 int efx_nic_flush_queues(struct efx_nic *efx);
 void siena_prepare_flush(struct efx_nic *efx);
 int efx_farch_fini_dmaq(struct efx_nic *efx);
+void efx_farch_finish_flr(struct efx_nic *efx);
 void siena_finish_flush(struct efx_nic *efx);
 void falcon_start_nic_stats(struct efx_nic *efx);
 void falcon_stop_nic_stats(struct efx_nic *efx);

drivers/net/ethernet/sfc/siena.c

@@ -921,6 +921,8 @@ const struct efx_nic_type siena_a0_nic_type = {
 	.fini_dmaq = efx_farch_fini_dmaq,
 	.prepare_flush = siena_prepare_flush,
 	.finish_flush = siena_finish_flush,
+	.prepare_flr = efx_port_dummy_op_void,
+	.finish_flr = efx_farch_finish_flr,
 	.describe_stats = siena_describe_nic_stats,
 	.update_stats = siena_update_nic_stats,
 	.start_stats = efx_mcdi_mac_start_stats,