[PATCH 1/2] sfc: Remove requirement to set tx-usecs-irq for shared channels when modifying coalescing parameters

@@ -1556,7 +1556,7 @@ static void efx_remove_all(struct efx_nic *efx)
  *
  **************************************************************************/
 
-static unsigned irq_mod_ticks(int usecs, int resolution)
+unsigned efx_irq_mod_ticks(int usecs, int resolution)
 {
 	if (usecs <= 0)
 		return 0; /* cannot receive interrupts ahead of time :-) */

@@ -1570,8 +1570,8 @@ void efx_init_irq_moderation(struct efx_nic *efx, int tx_usecs, int rx_usecs,
 			     bool rx_adaptive)
 {
 	struct efx_channel *channel;
-	unsigned tx_ticks = irq_mod_ticks(tx_usecs, EFX_IRQ_MOD_RESOLUTION);
-	unsigned rx_ticks = irq_mod_ticks(rx_usecs, EFX_IRQ_MOD_RESOLUTION);
+	unsigned tx_ticks = efx_irq_mod_ticks(tx_usecs, EFX_IRQ_MOD_RESOLUTION);
+	unsigned rx_ticks = efx_irq_mod_ticks(rx_usecs, EFX_IRQ_MOD_RESOLUTION);
 
 	EFX_ASSERT_RESET_SERIALISED(efx);
 

@@ -113,6 +113,7 @@ extern int efx_reset_up(struct efx_nic *efx, enum reset_type method, bool ok);
 extern void efx_schedule_reset(struct efx_nic *efx, enum reset_type type);
 extern void efx_init_irq_moderation(struct efx_nic *efx, int tx_usecs,
 				    int rx_usecs, bool rx_adaptive);
+extern unsigned efx_irq_mod_ticks(int usecs, int resolution);
 
 /* Dummy PHY ops for PHY drivers */
 extern int efx_port_dummy_op_int(struct efx_nic *efx);

@@ -644,7 +644,9 @@ static int efx_ethtool_set_coalesce(struct net_device *net_dev,
 	efx_for_each_channel(channel, efx) {
 		if (efx_channel_has_rx_queue(channel) &&
 		    efx_channel_has_tx_queues(channel) &&
-		    tx_usecs) {
+		    tx_usecs &&
+		    efx_irq_mod_ticks(tx_usecs, EFX_IRQ_MOD_RESOLUTION) !=
+		    channel->irq_moderation) {
 			netif_err(efx, drv, efx->net_dev, "Channel is shared. "
 				  "Only RX coalescing may be set\n");
 			return -EOPNOTSUPP;

@@ -628,12 +628,12 @@ static int efx_ethtool_set_coalesce(struct net_device *net_dev,
 	unsigned tx_usecs, rx_usecs, adaptive;
 
 	if (coalesce->use_adaptive_tx_coalesce)
-		return -EOPNOTSUPP;
+		return -EINVAL;
 
 	if (coalesce->rx_coalesce_usecs || coalesce->tx_coalesce_usecs) {
 		netif_err(efx, drv, efx->net_dev, "invalid coalescing setting. "
 			  "Only rx/tx_coalesce_usecs_irq are supported\n");
-		return -EOPNOTSUPP;
+		return -EINVAL;
 	}
 
 	rx_usecs = coalesce->rx_coalesce_usecs_irq;

@@ -647,7 +647,7 @@ static int efx_ethtool_set_coalesce(struct net_device *net_dev,
 		    tx_usecs) {
 			netif_err(efx, drv, efx->net_dev, "Channel is shared. "
 				  "Only RX coalescing may be set\n");
-			return -EOPNOTSUPP;
+			return -EINVAL;
 		}
 	}
 

@@ -1556,18 +1556,18 @@ static void efx_remove_all(struct efx_nic *efx)
  *
  **************************************************************************/
 
-static unsigned irq_mod_ticks(int usecs, int resolution)
+static unsigned int irq_mod_ticks(unsigned int usecs, unsigned int resolution)
 {
-	if (usecs <= 0)
-		return 0; /* cannot receive interrupts ahead of time :-) */
+	if (usecs == 0)
+		return 0;
 	if (usecs < resolution)
 		return 1; /* never round down to 0 */
 	return usecs / resolution;
 }
 
 /* Set interrupt moderation parameters */
-void efx_init_irq_moderation(struct efx_nic *efx, int tx_usecs, int rx_usecs,
-			     bool rx_adaptive)
+void efx_init_irq_moderation(struct efx_nic *efx, unsigned int tx_usecs,
+			     unsigned int rx_usecs, bool rx_adaptive)
 {
 	struct efx_channel *channel;
 	unsigned tx_ticks = irq_mod_ticks(tx_usecs, EFX_IRQ_MOD_RESOLUTION);

@@ -111,8 +111,8 @@ extern int efx_reset_up(struct efx_nic *efx, enum reset_type method, bool ok);
 
 /* Global */
 extern void efx_schedule_reset(struct efx_nic *efx, enum reset_type type);
-extern void efx_init_irq_moderation(struct efx_nic *efx, int tx_usecs,
-				    int rx_usecs, bool rx_adaptive);
+extern void efx_init_irq_moderation(struct efx_nic *efx, unsigned int tx_usecs,
+				    unsigned int rx_usecs, bool rx_adaptive);
 
 /* Dummy PHY ops for PHY drivers */
 extern int efx_port_dummy_op_int(struct efx_nic *efx);

@@ -625,7 +625,8 @@ static int efx_ethtool_set_coalesce(struct net_device *net_dev,
 {
 	struct efx_nic *efx = netdev_priv(net_dev);
 	struct efx_channel *channel;
-	unsigned tx_usecs, rx_usecs, adaptive;
+	unsigned int tx_usecs, rx_usecs;
+	bool adaptive;
 
 	if (coalesce->use_adaptive_tx_coalesce)
 		return -EINVAL;

@@ -1585,6 +1585,24 @@ void efx_init_irq_moderation(struct efx_nic *efx, unsigned int tx_usecs,
 	}
 }
 
+void efx_get_irq_moderation(struct efx_nic *efx, unsigned int *tx_usecs,
+			    unsigned int *rx_usecs, bool *rx_adaptive)
+{
+	*rx_adaptive = efx->irq_rx_adaptive;
+	*rx_usecs = efx->irq_rx_moderation * EFX_IRQ_MOD_RESOLUTION;
+
+	/* If channels are shared between RX and TX, so is IRQ
+	 * moderation.  Otherwise, IRQ moderation is the same for all
+	 * TX channels and is not adaptive.
+	 */
+	if (efx->tx_channel_offset == 0)
+		*tx_usecs = *rx_usecs;
+	else
+		*tx_usecs =
+			efx->channel[efx->tx_channel_offset]->irq_moderation *
+			EFX_IRQ_MOD_RESOLUTION;
+}
+
 /**************************************************************************
  *
  * Hardware monitor

@@ -113,6 +113,8 @@ extern int efx_reset_up(struct efx_nic *efx, enum reset_type method, bool ok);
 extern void efx_schedule_reset(struct efx_nic *efx, enum reset_type type);
 extern void efx_init_irq_moderation(struct efx_nic *efx, unsigned int tx_usecs,
 				    unsigned int rx_usecs, bool rx_adaptive);
+extern void efx_get_irq_moderation(struct efx_nic *efx, unsigned int *tx_usecs,
+				   unsigned int *rx_usecs, bool *rx_adaptive);
 
 /* Dummy PHY ops for PHY drivers */
 extern int efx_port_dummy_op_int(struct efx_nic *efx);

@@ -586,40 +586,37 @@ static int efx_ethtool_nway_reset(struct net_device *net_dev)
 	return mdio45_nway_restart(&efx->mdio);
 }
 
+/*
+ * Each channel has a single IRQ and moderation timer, started by any
+ * completion (or other event).  Unless the module parameter
+ * separate_tx_channels is set, IRQs and moderation are therefore
+ * shared between RX and TX completions.  In this case, when RX IRQ
+ * moderation is explicitly changed then TX IRQ moderation is
+ * automatically changed too, but otherwise we fail if the two values
+ * are requested to be different.
+ *
+ * We implement adaptive IRQ moderation, but use a different algorithm
+ * from that assumed in the definition of struct ethtool_coalesce.
+ * Therefore we do not use any of the adaptive moderation parameters
+ * in it.
+ */
+
 static int efx_ethtool_get_coalesce(struct net_device *net_dev,
 				    struct ethtool_coalesce *coalesce)
 {
 	struct efx_nic *efx = netdev_priv(net_dev);
-	struct efx_channel *channel;
-
-	memset(coalesce, 0, sizeof(*coalesce));
-
-	/* Find lowest IRQ moderation across all used TX queues */
-	coalesce->tx_coalesce_usecs_irq = ~((u32) 0);
-	efx_for_each_channel(channel, efx) {
-		if (!efx_channel_has_tx_queues(channel))
-			continue;
-		if (channel->irq_moderation < coalesce->tx_coalesce_usecs_irq) {
-			if (channel->channel < efx->n_rx_channels)
-				coalesce->tx_coalesce_usecs_irq =
-					channel->irq_moderation;
-			else
-				coalesce->tx_coalesce_usecs_irq = 0;
-		}
-	}
+	unsigned int tx_usecs, rx_usecs;
+	bool rx_adaptive;
 
-	coalesce->use_adaptive_rx_coalesce = efx->irq_rx_adaptive;
-	coalesce->rx_coalesce_usecs_irq = efx->irq_rx_moderation;
+	efx_get_irq_moderation(efx, &tx_usecs, &rx_usecs, &rx_adaptive);
 
-	coalesce->tx_coalesce_usecs_irq *= EFX_IRQ_MOD_RESOLUTION;
-	coalesce->rx_coalesce_usecs_irq *= EFX_IRQ_MOD_RESOLUTION;
+	coalesce->tx_coalesce_usecs_irq = tx_usecs;
+	coalesce->rx_coalesce_usecs_irq = rx_usecs;
+	coalesce->use_adaptive_rx_coalesce = rx_adaptive;
 
 	return 0;
 }
 
-/* Set coalescing parameters
- * The difficulties occur for shared channels
- */
 static int efx_ethtool_set_coalesce(struct net_device *net_dev,
 				    struct ethtool_coalesce *coalesce)
 {

@@ -637,20 +634,22 @@ static int efx_ethtool_set_coalesce(struct net_device *net_dev,
 		return -EINVAL;
 	}
 
+	efx_get_irq_moderation(efx, &tx_usecs, &rx_usecs, &adaptive);
+
 	rx_usecs = coalesce->rx_coalesce_usecs_irq;
-	tx_usecs = coalesce->tx_coalesce_usecs_irq;
 	adaptive = coalesce->use_adaptive_rx_coalesce;
 
-	/* If the channel is shared only allow RX parameters to be set */
-	efx_for_each_channel(channel, efx) {
-		if (efx_channel_has_rx_queue(channel) &&
-		    efx_channel_has_tx_queues(channel) &&
-		    tx_usecs) {
-			netif_err(efx, drv, efx->net_dev, "Channel is shared. "
-				  "Only RX coalescing may be set\n");
-			return -EINVAL;
-		}
+	/* If channels are shared, TX IRQ moderation can be quietly
+	 * overridden unless it is changed from its old value.
+	 */
+	if (efx->tx_channel_offset == 0 &&
+	    coalesce->tx_coalesce_usecs_irq != tx_usecs &&
+	    coalesce->tx_coalesce_usecs_irq != rx_usecs) {
+		netif_err(efx, drv, efx->net_dev, "Channels are shared. "
+			  "RX and TX IRQ moderation must be equal\n");
+		return -EINVAL;
 	}
+	tx_usecs = coalesce->tx_coalesce_usecs_irq;
 
 	efx_init_irq_moderation(efx, tx_usecs, rx_usecs, adaptive);
 	efx_for_each_channel(channel, efx)

@@ -1357,7 +1357,8 @@ static int efx_probe_nic(struct efx_nic *efx)
 	netif_set_real_num_rx_queues(efx->net_dev, efx->n_rx_channels);
 
 	/* Initialise the interrupt moderation settings */
-	efx_init_irq_moderation(efx, tx_irq_mod_usec, rx_irq_mod_usec, true);
+	efx_init_irq_moderation(efx, tx_irq_mod_usec, rx_irq_mod_usec, true,
+				true);
 
 	return 0;
 

@@ -1566,8 +1567,9 @@ static unsigned int irq_mod_ticks(unsigned int usecs, unsigned int resolution)
 }
 
 /* Set interrupt moderation parameters */
-void efx_init_irq_moderation(struct efx_nic *efx, unsigned int tx_usecs,
-			     unsigned int rx_usecs, bool rx_adaptive)
+int efx_init_irq_moderation(struct efx_nic *efx, unsigned int tx_usecs,
+			    unsigned int rx_usecs, bool rx_adaptive,
+			    bool rx_may_override_tx)
 {
 	struct efx_channel *channel;
 	unsigned tx_ticks = irq_mod_ticks(tx_usecs, EFX_IRQ_MOD_RESOLUTION);

@@ -1575,6 +1577,16 @@ void efx_init_irq_moderation(struct efx_nic *efx, unsigned int tx_usecs,
 
 	EFX_ASSERT_RESET_SERIALISED(efx);
 
+	if (tx_ticks > EFX_IRQ_MOD_MAX || rx_ticks > EFX_IRQ_MOD_MAX)
+		return -EINVAL;
+
+	if (tx_ticks != rx_ticks && efx->tx_channel_offset == 0 &&
+	    !rx_may_override_tx) {
+		netif_err(efx, drv, efx->net_dev, "Channels are shared. "
+			  "RX and TX IRQ moderation must be equal\n");
+		return -EINVAL;
+	}
+
 	efx->irq_rx_adaptive = rx_adaptive;
 	efx->irq_rx_moderation = rx_ticks;
 	efx_for_each_channel(channel, efx) {

@@ -1583,6 +1595,8 @@ void efx_init_irq_moderation(struct efx_nic *efx, unsigned int tx_usecs,
 		else if (efx_channel_has_tx_queues(channel))
 			channel->irq_moderation = tx_ticks;
 	}
+
+	return 0;
 }
 
 void efx_get_irq_moderation(struct efx_nic *efx, unsigned int *tx_usecs,

@@ -111,8 +111,9 @@ extern int efx_reset_up(struct efx_nic *efx, enum reset_type method, bool ok);
 
 /* Global */
 extern void efx_schedule_reset(struct efx_nic *efx, enum reset_type type);
-extern void efx_init_irq_moderation(struct efx_nic *efx, unsigned int tx_usecs,
-				    unsigned int rx_usecs, bool rx_adaptive);
+extern int efx_init_irq_moderation(struct efx_nic *efx, unsigned int tx_usecs,
+				   unsigned int rx_usecs, bool rx_adaptive,
+				   bool rx_may_override_tx);
 extern void efx_get_irq_moderation(struct efx_nic *efx, unsigned int *tx_usecs,
 				   unsigned int *rx_usecs, bool *rx_adaptive);
 

@@ -623,7 +623,8 @@ static int efx_ethtool_set_coalesce(struct net_device *net_dev,
 	struct efx_nic *efx = netdev_priv(net_dev);
 	struct efx_channel *channel;
 	unsigned int tx_usecs, rx_usecs;
-	bool adaptive;
+	bool adaptive, rx_may_override_tx;
+	int rc;
 
 	if (coalesce->use_adaptive_tx_coalesce)
 		return -EINVAL;

@@ -642,16 +643,14 @@ static int efx_ethtool_set_coalesce(struct net_device *net_dev,
 	/* If channels are shared, TX IRQ moderation can be quietly
 	 * overridden unless it is changed from its old value.
 	 */
-	if (efx->tx_channel_offset == 0 &&
-	    coalesce->tx_coalesce_usecs_irq != tx_usecs &&
-	    coalesce->tx_coalesce_usecs_irq != rx_usecs) {
-		netif_err(efx, drv, efx->net_dev, "Channels are shared. "
-			  "RX and TX IRQ moderation must be equal\n");
-		return -EINVAL;
-	}
+	rx_may_override_tx = coalesce->tx_coalesce_usecs_irq == tx_usecs;
 	tx_usecs = coalesce->tx_coalesce_usecs_irq;
 
-	efx_init_irq_moderation(efx, tx_usecs, rx_usecs, adaptive);
+	rc = efx_init_irq_moderation(efx, tx_usecs, rx_usecs, adaptive,
+				     rx_may_override_tx);
+	if (rc != 0)
+		return rc;
+
 	efx_for_each_channel(channel, efx)
 		efx->type->push_irq_moderation(channel);
 

@@ -104,6 +104,8 @@ static void falcon_push_irq_moderation(struct efx_channel *channel)
 	efx_dword_t timer_cmd;
 	struct efx_nic *efx = channel->efx;
 
+	BUILD_BUG_ON(EFX_IRQ_MOD_MAX > (1 << FRF_AB_TC_TIMER_VAL_WIDTH));
+
 	/* Set timer register */
 	if (channel->irq_moderation) {
 		EFX_POPULATE_DWORD_2(timer_cmd,

@@ -204,7 +204,8 @@ extern irqreturn_t efx_nic_fatal_interrupt(struct efx_nic *efx);
 extern irqreturn_t falcon_legacy_interrupt_a1(int irq, void *dev_id);
 extern void falcon_irq_ack_a1(struct efx_nic *efx);
 
-#define EFX_IRQ_MOD_RESOLUTION 5
+#define EFX_IRQ_MOD_RESOLUTION	5
+#define EFX_IRQ_MOD_MAX		0x1000
 
 /* Global Resources */
 extern int efx_nic_flush_queues(struct efx_nic *efx);

@@ -36,6 +36,8 @@ static void siena_push_irq_moderation(struct efx_channel *channel)
 {
 	efx_dword_t timer_cmd;
 
+	BUILD_BUG_ON(EFX_IRQ_MOD_MAX > (1 << FRF_CZ_TC_TIMER_VAL_WIDTH));
+
 	if (channel->irq_moderation)
 		EFX_POPULATE_DWORD_2(timer_cmd,
 				     FRF_CZ_TC_TIMER_MODE,

@@ -595,6 +595,20 @@ static int efx_ethtool_nway_reset(struct net_device *net_dev)
  * automatically changed too, but otherwise we fail if the two values
  * are requested to be different.
  *
+ * The hardware does not support a limit on the number of completions
+ * before an IRQ, so we do not use the max_frames fields.  We should
+ * report and require that max_frames == (usecs != 0), but this would
+ * invalidate existing user documentation.
+ *
+ * The hardware does not have distinct settings for interrupt
+ * moderation while the previous IRQ is being handled, so we should
+ * not use the 'irq' fields.  However, an earlier developer
+ * misunderstood the meaning of the 'irq' fields and the driver did
+ * not support the standard fields.  To avoid invalidating existing
+ * user documentation, we report and accept changes through either the
+ * standard or 'irq' fields.  If both are changed at the same time, we
+ * prefer the standard field.
+ *
  * We implement adaptive IRQ moderation, but use a different algorithm
  * from that assumed in the definition of struct ethtool_coalesce.
  * Therefore we do not use any of the adaptive moderation parameters

@@ -610,7 +624,9 @@ static int efx_ethtool_get_coalesce(struct net_device *net_dev,
 
 	efx_get_irq_moderation(efx, &tx_usecs, &rx_usecs, &rx_adaptive);
 
+	coalesce->tx_coalesce_usecs = tx_usecs;
 	coalesce->tx_coalesce_usecs_irq = tx_usecs;
+	coalesce->rx_coalesce_usecs = rx_usecs;
 	coalesce->rx_coalesce_usecs_irq = rx_usecs;
 	coalesce->use_adaptive_rx_coalesce = rx_adaptive;
 

@@ -629,22 +645,24 @@ static int efx_ethtool_set_coalesce(struct net_device *net_dev,
 	if (coalesce->use_adaptive_tx_coalesce)
 		return -EINVAL;
 
-	if (coalesce->rx_coalesce_usecs || coalesce->tx_coalesce_usecs) {
-		netif_err(efx, drv, efx->net_dev, "invalid coalescing setting. "
-			  "Only rx/tx_coalesce_usecs_irq are supported\n");
-		return -EINVAL;
-	}
-
 	efx_get_irq_moderation(efx, &tx_usecs, &rx_usecs, &adaptive);
 
-	rx_usecs = coalesce->rx_coalesce_usecs_irq;
+	if (coalesce->rx_coalesce_usecs != rx_usecs)
+		rx_usecs = coalesce->rx_coalesce_usecs;
+	else
+		rx_usecs = coalesce->rx_coalesce_usecs_irq;
+
 	adaptive = coalesce->use_adaptive_rx_coalesce;
 
 	/* If channels are shared, TX IRQ moderation can be quietly
 	 * overridden unless it is changed from its old value.
 	 */
-	rx_may_override_tx = coalesce->tx_coalesce_usecs_irq == tx_usecs;
-	tx_usecs = coalesce->tx_coalesce_usecs_irq;
+	rx_may_override_tx = (coalesce->tx_coalesce_usecs == tx_usecs &&
+			      coalesce->tx_coalesce_usecs_irq == tx_usecs);
+	if (coalesce->tx_coalesce_usecs != tx_usecs)
+		tx_usecs = coalesce->tx_coalesce_usecs;
+	else
+		tx_usecs = coalesce->tx_coalesce_usecs_irq;
 
 	rc = efx_init_irq_moderation(efx, tx_usecs, rx_usecs, adaptive,
 				     rx_may_override_tx);