[PATCH net-next-2.6 1/2] ethtool: Add support for control of RX flow hash indirection

[PATCH net-next-2.6 1/2] ethtool: Add support for control of RX flow hash indirection

[Ben Hutchings]

Many NICs use an indirection table to map an RX flow hash value to one
of an arbitrary number of queues (not necessarily a power of 2).  It
can be useful to remove some queues from this indirection table so
that they are only used for flows that are specifically filtered
there.  It may also be useful to weight the mapping to account for
user processes with the same CPU-affinity as the RX interrupts.

Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
---
 include/linux/ethtool.h |   15 +++++++++
 net/core/ethtool.c      |   80 +++++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 95 insertions(+), 0 deletions(-)

diff --git a/include/linux/ethtool.h b/include/linux/ethtool.h
index 2c8af09..3b089d8 100644
--- a/include/linux/ethtool.h
+++ b/include/linux/ethtool.h
@@ -384,6 +384,15 @@ struct ethtool_rxnfc {
 	__u32				rule_locs[0];
 };
 
+struct ethtool_rxfh_indir {
+	__u32	cmd;
+	/* On entry, this is the array size of the user buffer.  On
+	 * return from ETHTOOL_GRXFHINDIR, this is the array size of
+	 * the hardware indirection table. */
+	__u32	size;
+	__u32	ring_index[0];	/* ring/queue index for each hash value */
+};
+
 struct ethtool_rx_ntuple_flow_spec {
 	__u32		 flow_type;
 	union {
@@ -576,6 +585,10 @@ struct ethtool_ops {
 	int	(*set_rx_ntuple)(struct net_device *,
 				 struct ethtool_rx_ntuple *);
 	int	(*get_rx_ntuple)(struct net_device *, u32 stringset, void *);
+	int	(*get_rxfh_indir)(struct net_device *,
+				  struct ethtool_rxfh_indir *);
+	int	(*set_rxfh_indir)(struct net_device *,
+				  const struct ethtool_rxfh_indir *);
 };
 #endif /* __KERNEL__ */
 
@@ -637,6 +650,8 @@ struct ethtool_ops {
 #define ETHTOOL_SRXNTUPLE	0x00000035 /* Add an n-tuple filter to device */
 #define ETHTOOL_GRXNTUPLE	0x00000036 /* Get n-tuple filters from device */
 #define ETHTOOL_GSSET_INFO	0x00000037 /* Get string set info */
+#define ETHTOOL_GRXFHINDIR	0x00000038 /* Get RX flow hash indir'n table */
+#define ETHTOOL_SRXFHINDIR	0x00000039 /* Set RX flow hash indir'n table */
 
 /* compatibility with older code */
 #define SPARC_ETH_GSET		ETHTOOL_GSET
diff --git a/net/core/ethtool.c b/net/core/ethtool.c
index a0f4964..d978096 100644
--- a/net/core/ethtool.c
+++ b/net/core/ethtool.c
@@ -376,6 +376,80 @@ err_out:
 	return ret;
 }
 
+static noinline_for_stack int ethtool_get_rxfh_indir(struct net_device *dev,
+						     void __user *useraddr)
+{
+	struct ethtool_rxfh_indir *indir;
+	u32 table_size;
+	size_t full_size;
+	int ret;
+
+	if (!dev->ethtool_ops->get_rxfh_indir)
+		return -EOPNOTSUPP;
+
+	if (copy_from_user(&table_size,
+			   useraddr + offsetof(struct ethtool_rxfh_indir, size),
+			   sizeof(table_size)))
+		return -EFAULT;
+
+	if (table_size >
+	    (KMALLOC_MAX_SIZE - sizeof(*indir)) / sizeof(*indir->ring_index))
+		return -ENOMEM;
+	full_size = sizeof(*indir) + sizeof(*indir->ring_index) * table_size;
+	indir = kmalloc(full_size, GFP_USER);
+	if (!indir)
+		return -ENOMEM;
+
+	indir->cmd = ETHTOOL_GRXFHINDIR;
+	indir->size = table_size;
+	ret = dev->ethtool_ops->get_rxfh_indir(dev, indir);
+	if (ret)
+		goto out;
+
+	if (copy_to_user(useraddr, indir, full_size))
+		ret = -EFAULT;
+
+out:
+	kfree(indir);
+	return ret;
+}
+
+static noinline_for_stack int ethtool_set_rxfh_indir(struct net_device *dev,
+						     void __user *useraddr)
+{
+	struct ethtool_rxfh_indir *indir;
+	u32 table_size;
+	size_t full_size;
+	int ret;
+
+	if (!dev->ethtool_ops->set_rxfh_indir)
+		return -EOPNOTSUPP;
+
+	if (copy_from_user(&table_size,
+			   useraddr + offsetof(struct ethtool_rxfh_indir, size),
+			   sizeof(table_size)))
+		return -EFAULT;
+
+	if (table_size >
+	    (KMALLOC_MAX_SIZE - sizeof(*indir)) / sizeof(*indir->ring_index))
+		return -ENOMEM;
+	full_size = sizeof(*indir) + sizeof(*indir->ring_index) * table_size;
+	indir = kmalloc(full_size, GFP_USER);
+	if (!indir)
+		return -ENOMEM;
+
+	if (copy_from_user(indir, useraddr, full_size)) {
+		ret = -EFAULT;
+		goto out;
+	}
+
+	ret = dev->ethtool_ops->set_rxfh_indir(dev, indir);
+
+out:
+	kfree(indir);
+	return ret;
+}
+
 static void __rx_ntuple_filter_add(struct ethtool_rx_ntuple_list *list,
 			struct ethtool_rx_ntuple_flow_spec *spec,
 			struct ethtool_rx_ntuple_flow_spec_container *fsc)
@@ -1544,6 +1618,12 @@ int dev_ethtool(struct net *net, struct ifreq *ifr)
 	case ETHTOOL_GSSET_INFO:
 		rc = ethtool_get_sset_info(dev, useraddr);
 		break;
+	case ETHTOOL_GRXFHINDIR:
+		rc = ethtool_get_rxfh_indir(dev, useraddr);
+		break;
+	case ETHTOOL_SRXFHINDIR:
+		rc = ethtool_set_rxfh_indir(dev, useraddr);
+		break;
 	default:
 		rc = -EOPNOTSUPP;
 	}
-- 
1.6.2.5


-- 
Ben Hutchings, Senior Software Engineer, Solarflare Communications
Not speaking for my employer; that's the marketing department's job.
They asked us to note that Solarflare product names are trademarked.

[PATCH net-next-2.6 2/2] sfc: Add support for RX flow hash control

[Ben Hutchings]

Allow ethtool to query the number of RX rings, the fields used in RX
flow hashing and the hash indirection table.

Allow ethtool to update the RX flow hash indirection table.

Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>
---
 drivers/net/sfc/efx.c        |    3 +
 drivers/net/sfc/ethtool.c    |   90 ++++++++++++++++++++++++++++++++++++++++++
 drivers/net/sfc/net_driver.h |    2 +
 drivers/net/sfc/nic.c        |   19 ++++-----
 drivers/net/sfc/nic.h        |    1 +
 5 files changed, 105 insertions(+), 10 deletions(-)

diff --git a/drivers/net/sfc/efx.c b/drivers/net/sfc/efx.c
index 2a90bf9..35b3f29 100644
--- a/drivers/net/sfc/efx.c
+++ b/drivers/net/sfc/efx.c
@@ -1121,6 +1121,7 @@ static void efx_set_channels(struct efx_nic *efx)
 
 static int efx_probe_nic(struct efx_nic *efx)
 {
+	size_t i;
 	int rc;
 
 	netif_dbg(efx, probe, efx->net_dev, "creating NIC\n");
@@ -1136,6 +1137,8 @@ static int efx_probe_nic(struct efx_nic *efx)
 
 	if (efx->n_channels > 1)
 		get_random_bytes(&efx->rx_hash_key, sizeof(efx->rx_hash_key));
+	for (i = 0; i < ARRAY_SIZE(efx->rx_indir_table); i++)
+		efx->rx_indir_table[i] = i % efx->n_rx_channels;
 
 	efx_set_channels(efx);
 	efx->net_dev->real_num_tx_queues = efx->n_tx_channels;
diff --git a/drivers/net/sfc/ethtool.c b/drivers/net/sfc/ethtool.c
index 7693cfb..ab2510d 100644
--- a/drivers/net/sfc/ethtool.c
+++ b/drivers/net/sfc/ethtool.c
@@ -871,6 +871,93 @@ extern int efx_ethtool_reset(struct net_device *net_dev, u32 *flags)
 	return efx_reset(efx, method);
 }
 
+static int
+efx_ethtool_get_rxnfc(struct net_device *net_dev,
+		      struct ethtool_rxnfc *info, void *rules __always_unused)
+{
+	struct efx_nic *efx = netdev_priv(net_dev);
+	
+	switch (info->cmd) {
+	case ETHTOOL_GRXRINGS:
+		info->data = efx->n_rx_channels;
+		return 0;
+
+	case ETHTOOL_GRXFH: {
+		unsigned min_revision = 0;
+
+		info->data = 0;
+		switch (info->flow_type) {
+		case TCP_V4_FLOW:
+			info->data |= RXH_L4_B_0_1 | RXH_L4_B_2_3;
+			/* fall through */
+		case UDP_V4_FLOW:
+		case SCTP_V4_FLOW:
+		case AH_ESP_V4_FLOW:
+		case IPV4_FLOW:
+			info->data |= RXH_IP_SRC | RXH_IP_DST;
+			min_revision = EFX_REV_FALCON_B0;
+			break;
+		case TCP_V6_FLOW:
+			info->data |= RXH_L4_B_0_1 | RXH_L4_B_2_3;
+			/* fall through */
+		case UDP_V6_FLOW:
+		case SCTP_V6_FLOW:
+		case AH_ESP_V6_FLOW:
+		case IPV6_FLOW:
+			info->data |= RXH_IP_SRC | RXH_IP_DST;
+			min_revision = EFX_REV_SIENA_A0;
+			break;
+		default:
+			break;
+		}
+		if (efx_nic_rev(efx) < min_revision)
+			info->data = 0;
+		return 0;
+	}
+
+	default:
+		return -EOPNOTSUPP;
+	}
+}
+
+static int efx_ethtool_get_rxfh_indir(struct net_device *net_dev,
+				      struct ethtool_rxfh_indir *indir)
+{
+	struct efx_nic *efx = netdev_priv(net_dev);
+	size_t copy_size =
+		min_t(size_t, indir->size, ARRAY_SIZE(efx->rx_indir_table));
+
+	if (efx_nic_rev(efx) < EFX_REV_FALCON_B0)
+		return -EOPNOTSUPP;
+
+	indir->size = ARRAY_SIZE(efx->rx_indir_table);
+	memcpy(indir->ring_index, efx->rx_indir_table,
+	       copy_size * sizeof(indir->ring_index[0]));
+	return 0;
+}
+
+static int efx_ethtool_set_rxfh_indir(struct net_device *net_dev,
+				      const struct ethtool_rxfh_indir *indir)
+{
+	struct efx_nic *efx = netdev_priv(net_dev);
+	size_t i;
+
+	if (efx_nic_rev(efx) < EFX_REV_FALCON_B0)
+		return -EOPNOTSUPP;
+
+	/* Validate size and indices */
+	if (indir->size != ARRAY_SIZE(efx->rx_indir_table))
+		return -EINVAL;
+	for (i = 0; i < ARRAY_SIZE(efx->rx_indir_table); i++)
+		if (indir->ring_index[i] >= efx->n_rx_channels)
+			return -EINVAL;
+
+	memcpy(efx->rx_indir_table, indir->ring_index,
+	       sizeof(efx->rx_indir_table));
+	efx_nic_push_rx_indir_table(efx);
+	return 0;
+}
+
 const struct ethtool_ops efx_ethtool_ops = {
 	.get_settings		= efx_ethtool_get_settings,
 	.set_settings		= efx_ethtool_set_settings,
@@ -908,4 +995,7 @@ const struct ethtool_ops efx_ethtool_ops = {
 	.get_wol                = efx_ethtool_get_wol,
 	.set_wol                = efx_ethtool_set_wol,
 	.reset			= efx_ethtool_reset,
+	.get_rxnfc		= efx_ethtool_get_rxnfc,
+	.get_rxfh_indir		= efx_ethtool_get_rxfh_indir,
+	.set_rxfh_indir		= efx_ethtool_set_rxfh_indir,
 };
diff --git a/drivers/net/sfc/net_driver.h b/drivers/net/sfc/net_driver.h
index ba272a4..64e7caa 100644
--- a/drivers/net/sfc/net_driver.h
+++ b/drivers/net/sfc/net_driver.h
@@ -647,6 +647,7 @@ union efx_multicast_hash {
  * @n_tx_channels: Number of channels used for TX
  * @rx_buffer_len: RX buffer length
  * @rx_buffer_order: Order (log2) of number of pages for each RX buffer
+ * @rx_indir_table: Indirection table for RSS
  * @int_error_count: Number of internal errors seen recently
  * @int_error_expire: Time at which error count will be expired
  * @irq_status: Interrupt status buffer
@@ -734,6 +735,7 @@ struct efx_nic {
 	unsigned int rx_buffer_len;
 	unsigned int rx_buffer_order;
 	u8 rx_hash_key[40];
+	u32 rx_indir_table[128];
 
 	unsigned int_error_count;
 	unsigned long int_error_expire;
diff --git a/drivers/net/sfc/nic.c b/drivers/net/sfc/nic.c
index 3083657..f595d92 100644
--- a/drivers/net/sfc/nic.c
+++ b/drivers/net/sfc/nic.c
@@ -1484,22 +1484,21 @@ static irqreturn_t efx_msi_interrupt(int irq, void *dev_id)
 /* Setup RSS indirection table.
  * This maps from the hash value of the packet to RXQ
  */
-static void efx_setup_rss_indir_table(struct efx_nic *efx)
+void efx_nic_push_rx_indir_table(struct efx_nic *efx)
 {
-	int i = 0;
-	unsigned long offset;
+	size_t i = 0;
 	efx_dword_t dword;
 
 	if (efx_nic_rev(efx) < EFX_REV_FALCON_B0)
 		return;
 
-	for (offset = FR_BZ_RX_INDIRECTION_TBL;
-	     offset < FR_BZ_RX_INDIRECTION_TBL + 0x800;
-	     offset += 0x10) {
+	BUILD_BUG_ON(ARRAY_SIZE(efx->rx_indir_table) !=
+		     FR_BZ_RX_INDIRECTION_TBL_ROWS);
+
+	for (i = 0; i < FR_BZ_RX_INDIRECTION_TBL_ROWS; i++) {
 		EFX_POPULATE_DWORD_1(dword, FRF_BZ_IT_QUEUE,
-				     i % efx->n_rx_channels);
-		efx_writed(efx, &dword, offset);
-		i++;
+				     efx->rx_indir_table[i]);
+		efx_writed_table(efx, &dword, FR_BZ_RX_INDIRECTION_TBL, i);
 	}
 }
 
@@ -1634,7 +1633,7 @@ void efx_nic_init_common(struct efx_nic *efx)
 	EFX_INVERT_OWORD(temp);
 	efx_writeo(efx, &temp, FR_AZ_FATAL_INTR_KER);
 
-	efx_setup_rss_indir_table(efx);
+	efx_nic_push_rx_indir_table(efx);
 
 	/* Disable the ugly timer-based TX DMA backoff and allow TX DMA to be
 	 * controlled by the RX FIFO fill level. Set arbitration to one pkt/Q.
diff --git a/drivers/net/sfc/nic.h b/drivers/net/sfc/nic.h
index a39822d..0438dc9 100644
--- a/drivers/net/sfc/nic.h
+++ b/drivers/net/sfc/nic.h
@@ -207,6 +207,7 @@ extern void falcon_stop_nic_stats(struct efx_nic *efx);
 extern void falcon_setup_xaui(struct efx_nic *efx);
 extern int falcon_reset_xaui(struct efx_nic *efx);
 extern void efx_nic_init_common(struct efx_nic *efx);
+extern void efx_nic_push_rx_indir_table(struct efx_nic *efx);
 
 int efx_nic_alloc_buffer(struct efx_nic *efx, struct efx_buffer *buffer,
 			 unsigned int len);
-- 
1.6.2.5

-- 
Ben Hutchings, Senior Software Engineer, Solarflare Communications
Not speaking for my employer; that's the marketing department's job.
They asked us to note that Solarflare product names are trademarked.

Re: [PATCH net-next-2.6 1/2] ethtool: Add support for control of RX flow hash indirection

[David Miller]

From: Ben Hutchings <bhutchings@solarflare.com>
Date: Wed, 30 Jun 2010 16:05:23 +0100

> Many NICs use an indirection table to map an RX flow hash value to one
> of an arbitrary number of queues (not necessarily a power of 2).  It
> can be useful to remove some queues from this indirection table so
> that they are only used for flows that are specifically filtered
> there.  It may also be useful to weight the mapping to account for
> user processes with the same CPU-affinity as the RX interrupts.
> 
> Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>

Looks good, applied, thanks Ben.

Re: [PATCH net-next-2.6 2/2] sfc: Add support for RX flow hash control

[David Miller]

From: Ben Hutchings <bhutchings@solarflare.com>
Date: Wed, 30 Jun 2010 16:06:28 +0100

> Allow ethtool to query the number of RX rings, the fields used in RX
> flow hashing and the hash indirection table.
> 
> Allow ethtool to update the RX flow hash indirection table.
> 
> Signed-off-by: Ben Hutchings <bhutchings@solarflare.com>

Applied.