[netdevice] Retain and report detailed error breakdowns

netdev_rx_err() and netdev_tx_complete_err() get passed the error
code, but currently use it only in debug messages.

Retain error numbers and frequencey counts for up to
NETDEV_MAX_UNIQUE_ERRORS (4) different errors for each of TX and RX.
This allows the "ifstat" command to report the reasons for TX/RX
errors in most cases, even in non-debug builds.

src/include/gpxe/netdevice.h

@@ -193,16 +193,25 @@ struct net_device_operations {
 	void ( * irq ) ( struct net_device *netdev, int enable );
 };
 
+/** Network device error */
+struct net_device_error {
+	/** Error status code */
+	int rc;
+	/** Error count */
+	unsigned int count;
+};
+
+/** Maximum number of unique errors that we will keep track of */
+#define NETDEV_MAX_UNIQUE_ERRORS 4
+
 /** Network device statistics */
 struct net_device_stats {
-	/** Count of successfully completed transmissions */
-	unsigned int tx_ok;
-	/** Count of transmission errors */
-	unsigned int tx_err;
-	/** Count of successfully received packets */
-	unsigned int rx_ok;
-	/** Count of reception errors */
-	unsigned int rx_err;
+	/** Count of successful completions */
+	unsigned int good;
+	/** Count of error completions */
+	unsigned int bad;
+	/** Error breakdowns */
+	struct net_device_error errors[NETDEV_MAX_UNIQUE_ERRORS];
 };
 
 /**
@@ -250,8 +259,10 @@ struct net_device {
 	struct list_head tx_queue;
 	/** RX packet queue */
 	struct list_head rx_queue;
-	/** Device statistics */
-	struct net_device_stats stats;
+	/** TX statistics */
+	struct net_device_stats tx_stats;
+	/** RX statistics */
+	struct net_device_stats rx_stats;
 
 	/** Configuration settings applicable to this device */
 	struct simple_settings settings;

src/interface/efi/efi_snp.c

@@ -317,14 +317,14 @@ efi_snp_statistics ( EFI_SIMPLE_NETWORK_PROTOCOL *snp, BOOLEAN reset,
 
 	/* Gather statistics */
 	memset ( &stats_buf, 0, sizeof ( stats_buf ) );
-	stats_buf.TxGoodFrames = snpdev->netdev->stats.tx_ok;
-	stats_buf.TxDroppedFrames = snpdev->netdev->stats.tx_err;
-	stats_buf.TxTotalFrames = ( snpdev->netdev->stats.tx_ok +
-				    snpdev->netdev->stats.tx_err );
-	stats_buf.RxGoodFrames = snpdev->netdev->stats.rx_ok;
-	stats_buf.RxDroppedFrames = snpdev->netdev->stats.rx_err;
-	stats_buf.RxTotalFrames = ( snpdev->netdev->stats.rx_ok +
-				    snpdev->netdev->stats.rx_err );
+	stats_buf.TxGoodFrames = snpdev->netdev->tx_stats.good;
+	stats_buf.TxDroppedFrames = snpdev->netdev->tx_stats.bad;
+	stats_buf.TxTotalFrames = ( snpdev->netdev->tx_stats.good +
+				    snpdev->netdev->tx_stats.bad );
+	stats_buf.RxGoodFrames = snpdev->netdev->rx_stats.good;
+	stats_buf.RxDroppedFrames = snpdev->netdev->rx_stats.bad;
+	stats_buf.RxTotalFrames = ( snpdev->netdev->rx_stats.good +
+				    snpdev->netdev->rx_stats.bad );
 	if ( *stats_len > sizeof ( stats_buf ) )
 		*stats_len = sizeof ( stats_buf );
 	if ( stats )
@@ -332,8 +332,10 @@ efi_snp_statistics ( EFI_SIMPLE_NETWORK_PROTOCOL *snp, BOOLEAN reset,
 
 	/* Reset statistics if requested to do so */
 	if ( reset ) {
-		memset ( &snpdev->netdev->stats, 0,
-			 sizeof ( snpdev->netdev->stats ) );
+		memset ( &snpdev->netdev->tx_stats, 0,
+			 sizeof ( snpdev->netdev->tx_stats ) );
+		memset ( &snpdev->netdev->rx_stats, 0,
+			 sizeof ( snpdev->netdev->rx_stats ) );
 	}
 
 	return 0;

src/interface/pxe/pxe_undi.c

@@ -392,10 +392,10 @@ PXENV_EXIT_t pxenv_undi_get_statistics ( struct s_PXENV_UNDI_GET_STATISTICS
 					 *undi_get_statistics ) {
 	DBG ( "PXENV_UNDI_GET_STATISTICS" );
 
-	undi_get_statistics->XmtGoodFrames = pxe_netdev->stats.tx_ok;
-	undi_get_statistics->RcvGoodFrames = pxe_netdev->stats.rx_ok;
-	undi_get_statistics->RcvCRCErrors = pxe_netdev->stats.rx_err;
-	undi_get_statistics->RcvResourceErrors = pxe_netdev->stats.rx_err;
+	undi_get_statistics->XmtGoodFrames = pxe_netdev->tx_stats.good;
+	undi_get_statistics->RcvGoodFrames = pxe_netdev->rx_stats.good;
+	undi_get_statistics->RcvCRCErrors = pxe_netdev->rx_stats.bad;
+	undi_get_statistics->RcvResourceErrors = pxe_netdev->rx_stats.bad;
 
 	undi_get_statistics->Status = PXENV_STATUS_SUCCESS;
 	return PXENV_EXIT_SUCCESS;
@@ -409,7 +409,8 @@ PXENV_EXIT_t pxenv_undi_clear_statistics ( struct s_PXENV_UNDI_CLEAR_STATISTICS
 					   *undi_clear_statistics ) {
 	DBG ( "PXENV_UNDI_CLEAR_STATISTICS" );
 
-	memset ( &pxe_netdev->stats, 0, sizeof ( pxe_netdev->stats ) );
+	memset ( &pxe_netdev->tx_stats, 0, sizeof ( pxe_netdev->tx_stats ) );
+	memset ( &pxe_netdev->rx_stats, 0, sizeof ( pxe_netdev->rx_stats ) );
 
 	undi_clear_statistics->Status = PXENV_STATUS_SUCCESS;
 	return PXENV_EXIT_SUCCESS;

src/net/netdevice.c

@@ -45,6 +45,45 @@ static struct net_protocol net_protocols_end[0]
 /** List of network devices */
 struct list_head net_devices = LIST_HEAD_INIT ( net_devices );
 
+/**
+ * Record network device statistic
+ *
+ * @v stats		Network device statistics
+ * @v rc		Status code
+ */
+static void netdev_record_stat ( struct net_device_stats *stats, int rc ) {
+	struct net_device_error *error;
+	struct net_device_error *least_common_error;
+	unsigned int i;
+
+	/* If this is not an error, just update the good counter */
+	if ( rc == 0 ) {
+		stats->good++;
+		return;
+	}
+
+	/* Update the bad counter */
+	stats->bad++;
+
+	/* Locate the appropriate error record */
+	least_common_error = &stats->errors[0];
+	for ( i = 0 ; i < ( sizeof ( stats->errors ) /
+			    sizeof ( stats->errors[0] ) ) ; i++ ) {
+		error = &stats->errors[i];
+		/* Update matching record, if found */
+		if ( error->rc == rc ) {
+			error->count++;
+			return;
+		}
+		if ( error->count < least_common_error->count )
+			least_common_error = error;
+	}
+
+	/* Overwrite the least common error record */
+	least_common_error->rc = rc;
+	least_common_error->count = 1;
+}
+
 /**
  * Transmit raw packet via network device
  *
@@ -91,12 +130,11 @@ void netdev_tx_complete_err ( struct net_device *netdev,
 			      struct io_buffer *iobuf, int rc ) {
 
 	/* Update statistics counter */
+	netdev_record_stat ( &netdev->tx_stats, rc );
 	if ( rc == 0 ) {
-		netdev->stats.tx_ok++;
 		DBGC ( netdev, "NETDEV %p transmission %p complete\n",
 		       netdev, iobuf );
 	} else {
-		netdev->stats.tx_err++;
 		DBGC ( netdev, "NETDEV %p transmission %p failed: %s\n",
 		       netdev, iobuf, strerror ( rc ) );
 	}
@@ -158,7 +196,7 @@ void netdev_rx ( struct net_device *netdev, struct io_buffer *iobuf ) {
 	list_add_tail ( &iobuf->list, &netdev->rx_queue );
 
 	/* Update statistics counter */
-	netdev->stats.rx_ok++;
+	netdev_record_stat ( &netdev->rx_stats, 0 );
 }
 
 /**
@@ -183,7 +221,7 @@ void netdev_rx_err ( struct net_device *netdev,
 	free_iob ( iobuf );
 
 	/* Update statistics counter */
-	netdev->stats.rx_err++;
+	netdev_record_stat ( &netdev->rx_stats, rc );
 }
 
 /**

src/usr/ifmgmt.c

@@ -58,6 +58,25 @@ void ifclose ( struct net_device *netdev ) {
 	netdev_close ( netdev );
 }
 
+/**
+ * Print network device error breakdown
+ *
+ * @v stats		Network device statistics
+ * @v prefix		Message prefix
+ */
+static void ifstat_errors ( struct net_device_stats *stats,
+			    const char *prefix ) {
+	unsigned int i;
+
+	for ( i = 0 ; i < ( sizeof ( stats->errors ) /
+			    sizeof ( stats->errors[0] ) ) ; i++ ) {
+		if ( stats->errors[i].count )
+			printf ( "  [%s: %d x \"%s\"]\n", prefix,
+				 stats->errors[i].count,
+				 strerror ( stats->errors[i].rc ) );
+	}
+}
+
 /**
  * Print status of network device
  *
@@ -69,8 +88,10 @@ void ifstat ( struct net_device *netdev ) {
 		 netdev->name, netdev_hwaddr ( netdev ), netdev->dev->name,
 		 ( ( netdev->state & NETDEV_OPEN ) ? "open" : "closed" ),
 		 ( netdev_link_ok ( netdev ) ? "up" : "down" ),
-		 netdev->stats.tx_ok, netdev->stats.tx_err,
-		 netdev->stats.rx_ok, netdev->stats.rx_err );
+		 netdev->tx_stats.good, netdev->tx_stats.bad,
+		 netdev->rx_stats.good, netdev->rx_stats.bad );
+	ifstat_errors ( &netdev->tx_stats, "TXE" );
+	ifstat_errors ( &netdev->rx_stats, "RXE" );
 }
 
 /**