sfc: Work-around flush timeout when flushes have completed

We sometimes hit a "failed to flush" timeout on some TX queues, but the flushes have completed and the flush completion events seem to go missing. In this case, we can check the TX_DESC_PTR_TBL register and drain the queues if the flushes had finished. [bwh: Minor fixes to coding style] Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>

sfc: Work-around flush timeout when flushes have completed
We sometimes hit a "failed to flush" timeout on some TX queues, but the flushes have completed and the flush completion events seem to go missing. In this case, we can check the TX_DESC_PTR_TBL register and drain the queues if the flushes had finished. [bwh: Minor fixes to coding style] Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
Daniel Pieczko · Ben Hutchings
1 parent 876be083b6
Showing 2 changed files with 53 additions and 4 deletions Side-by-side Diff
drivers/net/ethernet/sfc/net_driver.h
drivers/net/ethernet/sfc/nic.c
@@ -200,6 +200,7 @@
 	/* Members shared between paths and sometimes updated */
 	unsigned int empty_read_count ____cacheline_aligned_in_smp;
 #define EFX_EMPTY_COUNT_VALID 0x80000000
+	atomic_t flush_outstanding;
 };
  
 /**
@@ -73,6 +73,8 @@
 	_EFX_CHANNEL_MAGIC(_EFX_CHANNEL_MAGIC_TX_DRAIN,			\
 			   (_tx_queue)->queue)
  
+static void efx_magic_event(struct efx_channel *channel, u32 magic);
+
 /**************************************************************************
  *
  * Solarstorm hardware access
@@ -491,6 +493,9 @@
 	struct efx_nic *efx = tx_queue->efx;
 	efx_oword_t tx_flush_descq;
  
+	WARN_ON(atomic_read(&tx_queue->flush_outstanding));
+	atomic_set(&tx_queue->flush_outstanding, 1);
+
 	EFX_POPULATE_OWORD_2(tx_flush_descq,
 			     FRF_AZ_TX_FLUSH_DESCQ_CMD, 1,
 			     FRF_AZ_TX_FLUSH_DESCQ, tx_queue->queue);
@@ -666,6 +671,47 @@
 		 && atomic_read(&efx->rxq_flush_pending) > 0));
 }
  
+static bool efx_check_tx_flush_complete(struct efx_nic *efx)
+{
+	bool i = true;
+	efx_oword_t txd_ptr_tbl;
+	struct efx_channel *channel;
+	struct efx_tx_queue *tx_queue;
+
+	efx_for_each_channel(channel, efx) {
+		efx_for_each_channel_tx_queue(tx_queue, channel) {
+			efx_reado_table(efx, &txd_ptr_tbl,
+					FR_BZ_TX_DESC_PTR_TBL, tx_queue->queue);
+			if (EFX_OWORD_FIELD(txd_ptr_tbl,
+					    FRF_AZ_TX_DESCQ_FLUSH) ||
+			    EFX_OWORD_FIELD(txd_ptr_tbl,
+					    FRF_AZ_TX_DESCQ_EN)) {
+				netif_dbg(efx, hw, efx->net_dev,
+					  "flush did not complete on TXQ %d\n",
+					  tx_queue->queue);
+				i = false;
+			} else if (atomic_cmpxchg(&tx_queue->flush_outstanding,
+						  1, 0)) {
+				/* The flush is complete, but we didn't
+				 * receive a flush completion event
+				 */
+				netif_dbg(efx, hw, efx->net_dev,
+					  "flush complete on TXQ %d, so drain "
+					  "the queue\n", tx_queue->queue);
+				/* Don't need to increment drain_pending as it
+				 * has already been incremented for the queues
+				 * which did not drain
+				 */
+				efx_magic_event(channel,
+						EFX_CHANNEL_MAGIC_TX_DRAIN(
+							tx_queue));
+			}
+		}
+	}
+
+	return i;
+}
+
 /* Flush all the transmit queues, and continue flushing receive queues until
  * they're all flushed. Wait for the DRAIN events to be recieved so that there
  * are no more RX and TX events left on any channel. */
@@ -726,7 +772,8 @@
 					     timeout);
 	}
  
-	if (atomic_read(&efx->drain_pending)) {
+	if (atomic_read(&efx->drain_pending) &&
+	    !efx_check_tx_flush_complete(efx)) {
 		netif_err(efx, hw, efx->net_dev, "failed to flush %d queues "
 			  "(rx %d+%d)\n", atomic_read(&efx->drain_pending),
 			  atomic_read(&efx->rxq_flush_outstanding),
@@ -1018,9 +1065,10 @@
 	if (qid < EFX_TXQ_TYPES * efx->n_tx_channels) {
 		tx_queue = efx_get_tx_queue(efx, qid / EFX_TXQ_TYPES,
 					    qid % EFX_TXQ_TYPES);
-
-		efx_magic_event(tx_queue->channel,
-				EFX_CHANNEL_MAGIC_TX_DRAIN(tx_queue));
+		if (atomic_cmpxchg(&tx_queue->flush_outstanding, 1, 0)) {
+			efx_magic_event(tx_queue->channel,
+					EFX_CHANNEL_MAGIC_TX_DRAIN(tx_queue));
+		}
 	}
 }
...	...	@@ -200,6 +200,7 @@
200	200	/* Members shared between paths and sometimes updated */
201	201	unsigned int empty_read_count ____cacheline_aligned_in_smp;
202	202	#define EFX_EMPTY_COUNT_VALID 0x80000000
	203	+ atomic_t flush_outstanding;
203	204	};
204	205
205	206	/**
...	...	@@ -73,6 +73,8 @@
73	73	_EFX_CHANNEL_MAGIC(_EFX_CHANNEL_MAGIC_TX_DRAIN, \
74	74	(_tx_queue)->queue)
75	75
	76	+static void efx_magic_event(struct efx_channel *channel, u32 magic);
	77	+
76	78	/**************************************************************************
77	79	*
78	80	* Solarstorm hardware access
...	...	@@ -491,6 +493,9 @@
491	493	struct efx_nic *efx = tx_queue->efx;
492	494	efx_oword_t tx_flush_descq;
493	495
	496	+ WARN_ON(atomic_read(&tx_queue->flush_outstanding));
	497	+ atomic_set(&tx_queue->flush_outstanding, 1);
	498	+
494	499	EFX_POPULATE_OWORD_2(tx_flush_descq,
495	500	FRF_AZ_TX_FLUSH_DESCQ_CMD, 1,
496	501	FRF_AZ_TX_FLUSH_DESCQ, tx_queue->queue);
...	...	@@ -666,6 +671,47 @@
666	671	&& atomic_read(&efx->rxq_flush_pending) > 0));
667	672	}
668	673
	674	+static bool efx_check_tx_flush_complete(struct efx_nic *efx)
	675	+{
	676	+ bool i = true;
	677	+ efx_oword_t txd_ptr_tbl;
	678	+ struct efx_channel *channel;
	679	+ struct efx_tx_queue *tx_queue;
	680	+
	681	+ efx_for_each_channel(channel, efx) {
	682	+ efx_for_each_channel_tx_queue(tx_queue, channel) {
	683	+ efx_reado_table(efx, &txd_ptr_tbl,
	684	+ FR_BZ_TX_DESC_PTR_TBL, tx_queue->queue);
	685	+ if (EFX_OWORD_FIELD(txd_ptr_tbl,
	686	+ FRF_AZ_TX_DESCQ_FLUSH) \|\|
	687	+ EFX_OWORD_FIELD(txd_ptr_tbl,
	688	+ FRF_AZ_TX_DESCQ_EN)) {
	689	+ netif_dbg(efx, hw, efx->net_dev,
	690	+ "flush did not complete on TXQ %d\n",
	691	+ tx_queue->queue);
	692	+ i = false;
	693	+ } else if (atomic_cmpxchg(&tx_queue->flush_outstanding,
	694	+ 1, 0)) {
	695	+ /* The flush is complete, but we didn't
	696	+ * receive a flush completion event
	697	+ */
	698	+ netif_dbg(efx, hw, efx->net_dev,
	699	+ "flush complete on TXQ %d, so drain "
	700	+ "the queue\n", tx_queue->queue);
	701	+ /* Don't need to increment drain_pending as it
	702	+ * has already been incremented for the queues
	703	+ * which did not drain
	704	+ */
	705	+ efx_magic_event(channel,
	706	+ EFX_CHANNEL_MAGIC_TX_DRAIN(
	707	+ tx_queue));
	708	+ }
	709	+ }
	710	+ }
	711	+
	712	+ return i;
	713	+}
	714	+
669	715	/* Flush all the transmit queues, and continue flushing receive queues until
670	716	* they're all flushed. Wait for the DRAIN events to be recieved so that there
671	717	* are no more RX and TX events left on any channel. */
...	...	@@ -726,7 +772,8 @@
726	772	timeout);
727	773	}
728	774
729		- if (atomic_read(&efx->drain_pending)) {
	775	+ if (atomic_read(&efx->drain_pending) &&
	776	+ !efx_check_tx_flush_complete(efx)) {
730	777	netif_err(efx, hw, efx->net_dev, "failed to flush %d queues "
731	778	"(rx %d+%d)\n", atomic_read(&efx->drain_pending),
732	779	atomic_read(&efx->rxq_flush_outstanding),
...	...	@@ -1018,9 +1065,10 @@
1018	1065	if (qid < EFX_TXQ_TYPES * efx->n_tx_channels) {
1019	1066	tx_queue = efx_get_tx_queue(efx, qid / EFX_TXQ_TYPES,
1020	1067	qid % EFX_TXQ_TYPES);
1021		-
1022		- efx_magic_event(tx_queue->channel,
1023		- EFX_CHANNEL_MAGIC_TX_DRAIN(tx_queue));
	1068	+ if (atomic_cmpxchg(&tx_queue->flush_outstanding, 1, 0)) {
	1069	+ efx_magic_event(tx_queue->channel,
	1070	+ EFX_CHANNEL_MAGIC_TX_DRAIN(tx_queue));
	1071	+ }
1024	1072	}
1025	1073	}
1026	1074