netdev.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
* [PATCBN net-next] mlx4_en: fix skb truesize underestimation
@ 2011-10-18 10:13 Eric Dumazet
  2011-10-19 23:53 ` David Miller
  2011-10-20  4:49 ` [PATCH V2 " Eric Dumazet
  0 siblings, 2 replies; 4+ messages in thread
From: Eric Dumazet @ 2011-10-18 10:13 UTC (permalink / raw)
  To: David Miller; +Cc: netdev, Yevgeny Petrilin

skb->truesize must account for allocated memory, not the used part of
it. Doing this work is important to avoid unexpected OOM situations.

Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
CC: Yevgeny Petrilin <yevgenyp@mellanox.co.il>
---
Note: To improve performance, this driver should avoid all the
get_page() done for every frag since it dirties page->_count and
conflict with previous skbs consumers (on a different cpu maybe)

Only at alloc_pages() time should we add to page->count the
(number_of_frags on this page) - 1

 drivers/net/ethernet/mellanox/mlx4/en_rx.c |   11 +++++------
 1 file changed, 5 insertions(+), 6 deletions(-)

diff --git a/drivers/net/ethernet/mellanox/mlx4/en_rx.c b/drivers/net/ethernet/mellanox/mlx4/en_rx.c
index 37cc9e5..c9aaf20 100644
--- a/drivers/net/ethernet/mellanox/mlx4/en_rx.c
+++ b/drivers/net/ethernet/mellanox/mlx4/en_rx.c
@@ -404,10 +404,11 @@ void mlx4_en_deactivate_rx_ring(struct mlx4_en_priv *priv,
 static int mlx4_en_complete_rx_desc(struct mlx4_en_priv *priv,
 				    struct mlx4_en_rx_desc *rx_desc,
 				    struct skb_frag_struct *skb_frags,
-				    struct skb_frag_struct *skb_frags_rx,
+				    struct sk_buff *skb,
 				    struct mlx4_en_rx_alloc *page_alloc,
 				    int length)
 {
+	struct skb_frag_struct *skb_frags_rx = skb_shinfo(skb)->frags;
 	struct mlx4_en_dev *mdev = priv->mdev;
 	struct mlx4_en_frag_info *frag_info;
 	int nr;
@@ -423,6 +424,7 @@ static int mlx4_en_complete_rx_desc(struct mlx4_en_priv *priv,
 		skb_frags_rx[nr].page = skb_frags[nr].page;
 		skb_frags_rx[nr].size = skb_frags[nr].size;
 		skb_frags_rx[nr].page_offset = skb_frags[nr].page_offset;
+		skb->truesize += frag_info->frag_stride;
 		dma = be64_to_cpu(rx_desc->data[nr].addr);
 
 		/* Allocate a replacement page */
@@ -470,7 +472,6 @@ static struct sk_buff *mlx4_en_rx_skb(struct mlx4_en_priv *priv,
 	skb->dev = priv->dev;
 	skb_reserve(skb, NET_IP_ALIGN);
 	skb->len = length;
-	skb->truesize = length + sizeof(struct sk_buff);
 
 	/* Get pointer to first fragment so we could copy the headers into the
 	 * (linear part of the) skb */
@@ -490,8 +491,7 @@ static struct sk_buff *mlx4_en_rx_skb(struct mlx4_en_priv *priv,
 
 		/* Move relevant fragments to skb */
 		used_frags = mlx4_en_complete_rx_desc(priv, rx_desc, skb_frags,
-						      skb_shinfo(skb)->frags,
-						      page_alloc, length);
+						      skb, page_alloc, length);
 		if (unlikely(!used_frags)) {
 			kfree_skb(skb);
 			return NULL;
@@ -600,7 +600,7 @@ int mlx4_en_process_rx_cq(struct net_device *dev, struct mlx4_en_cq *cq, int bud
 
 					nr = mlx4_en_complete_rx_desc(
 						priv, rx_desc,
-						skb_frags, skb_shinfo(gro_skb)->frags,
+						skb_frags, gro_skb,
 						ring->page_alloc, length);
 					if (!nr)
 						goto next;
@@ -608,7 +608,6 @@ int mlx4_en_process_rx_cq(struct net_device *dev, struct mlx4_en_cq *cq, int bud
 					skb_shinfo(gro_skb)->nr_frags = nr;
 					gro_skb->len = length;
 					gro_skb->data_len = length;
-					gro_skb->truesize += length;
 					gro_skb->ip_summed = CHECKSUM_UNNECESSARY;
 
 					if (cqe->vlan_my_qpn &

^ permalink raw reply related	[flat|nested] 4+ messages in thread

* Re: [PATCBN net-next] mlx4_en: fix skb truesize underestimation
  2011-10-18 10:13 [PATCBN net-next] mlx4_en: fix skb truesize underestimation Eric Dumazet
@ 2011-10-19 23:53 ` David Miller
  2011-10-20  4:49 ` [PATCH V2 " Eric Dumazet
  1 sibling, 0 replies; 4+ messages in thread
From: David Miller @ 2011-10-19 23:53 UTC (permalink / raw)
  To: eric.dumazet; +Cc: netdev, yevgenyp

From: Eric Dumazet <eric.dumazet@gmail.com>
Date: Tue, 18 Oct 2011 12:13:02 +0200

> skb->truesize must account for allocated memory, not the used part of
> it. Doing this work is important to avoid unexpected OOM situations.
> 
> Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
> CC: Yevgeny Petrilin <yevgenyp@mellanox.co.il>

Eric please respin this against current net-next, it doesn't apply
after the mlx4 series I applied today.

Thanks!

^ permalink raw reply	[flat|nested] 4+ messages in thread

* [PATCH V2 net-next] mlx4_en: fix skb truesize underestimation
  2011-10-18 10:13 [PATCBN net-next] mlx4_en: fix skb truesize underestimation Eric Dumazet
  2011-10-19 23:53 ` David Miller
@ 2011-10-20  4:49 ` Eric Dumazet
  2011-10-20  8:55   ` David Miller
  1 sibling, 1 reply; 4+ messages in thread
From: Eric Dumazet @ 2011-10-20  4:49 UTC (permalink / raw)
  To: David Miller; +Cc: netdev, Yevgeny Petrilin

skb->truesize must account for allocated memory, not the used part of
it. Doing this work is important to avoid unexpected OOM situations.

Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
CC: Yevgeny Petrilin <yevgenyp@mellanox.co.il>
---
V2: respin after recent patches

 drivers/net/ethernet/mellanox/mlx4/en_rx.c |   11 +++++------
 1 file changed, 5 insertions(+), 6 deletions(-)

diff --git a/drivers/net/ethernet/mellanox/mlx4/en_rx.c b/drivers/net/ethernet/mellanox/mlx4/en_rx.c
index 9b18d85..9aec8b8 100644
--- a/drivers/net/ethernet/mellanox/mlx4/en_rx.c
+++ b/drivers/net/ethernet/mellanox/mlx4/en_rx.c
@@ -404,10 +404,11 @@ void mlx4_en_deactivate_rx_ring(struct mlx4_en_priv *priv,
 static int mlx4_en_complete_rx_desc(struct mlx4_en_priv *priv,
 				    struct mlx4_en_rx_desc *rx_desc,
 				    struct skb_frag_struct *skb_frags,
-				    struct skb_frag_struct *skb_frags_rx,
+				    struct sk_buff *skb,
 				    struct mlx4_en_rx_alloc *page_alloc,
 				    int length)
 {
+	struct skb_frag_struct *skb_frags_rx = skb_shinfo(skb)->frags;
 	struct mlx4_en_dev *mdev = priv->mdev;
 	struct mlx4_en_frag_info *frag_info;
 	int nr;
@@ -423,6 +424,7 @@ static int mlx4_en_complete_rx_desc(struct mlx4_en_priv *priv,
 		skb_frags_rx[nr].page = skb_frags[nr].page;
 		skb_frag_size_set(&skb_frags_rx[nr], skb_frag_size(&skb_frags[nr]));
 		skb_frags_rx[nr].page_offset = skb_frags[nr].page_offset;
+		skb->truesize += frag_info->frag_stride;
 		dma = be64_to_cpu(rx_desc->data[nr].addr);
 
 		/* Allocate a replacement page */
@@ -470,7 +472,6 @@ static struct sk_buff *mlx4_en_rx_skb(struct mlx4_en_priv *priv,
 	skb->dev = priv->dev;
 	skb_reserve(skb, NET_IP_ALIGN);
 	skb->len = length;
-	skb->truesize = length + sizeof(struct sk_buff);
 
 	/* Get pointer to first fragment so we could copy the headers into the
 	 * (linear part of the) skb */
@@ -490,8 +491,7 @@ static struct sk_buff *mlx4_en_rx_skb(struct mlx4_en_priv *priv,
 
 		/* Move relevant fragments to skb */
 		used_frags = mlx4_en_complete_rx_desc(priv, rx_desc, skb_frags,
-						      skb_shinfo(skb)->frags,
-						      page_alloc, length);
+						      skb, page_alloc, length);
 		if (unlikely(!used_frags)) {
 			kfree_skb(skb);
 			return NULL;
@@ -600,7 +600,7 @@ int mlx4_en_process_rx_cq(struct net_device *dev, struct mlx4_en_cq *cq, int bud
 
 					nr = mlx4_en_complete_rx_desc(
 						priv, rx_desc,
-						skb_frags, skb_shinfo(gro_skb)->frags,
+						skb_frags, gro_skb,
 						ring->page_alloc, length);
 					if (!nr)
 						goto next;
@@ -608,7 +608,6 @@ int mlx4_en_process_rx_cq(struct net_device *dev, struct mlx4_en_cq *cq, int bud
 					skb_shinfo(gro_skb)->nr_frags = nr;
 					gro_skb->len = length;
 					gro_skb->data_len = length;
-					gro_skb->truesize += length;
 					gro_skb->ip_summed = CHECKSUM_UNNECESSARY;
 
 					if (cqe->vlan_my_qpn &

^ permalink raw reply related	[flat|nested] 4+ messages in thread

* Re: [PATCH V2 net-next] mlx4_en: fix skb truesize underestimation
  2011-10-20  4:49 ` [PATCH V2 " Eric Dumazet
@ 2011-10-20  8:55   ` David Miller
  0 siblings, 0 replies; 4+ messages in thread
From: David Miller @ 2011-10-20  8:55 UTC (permalink / raw)
  To: eric.dumazet; +Cc: netdev, yevgenyp

From: Eric Dumazet <eric.dumazet@gmail.com>
Date: Thu, 20 Oct 2011 06:49:52 +0200

> skb->truesize must account for allocated memory, not the used part of
> it. Doing this work is important to avoid unexpected OOM situations.
> 
> Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
> CC: Yevgeny Petrilin <yevgenyp@mellanox.co.il>
> ---
> V2: respin after recent patches

Applied, thanks Eric.

^ permalink raw reply	[flat|nested] 4+ messages in thread

end of thread, other threads:[~2011-10-20  8:56 UTC | newest]

Thread overview: 4+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2011-10-18 10:13 [PATCBN net-next] mlx4_en: fix skb truesize underestimation Eric Dumazet
2011-10-19 23:53 ` David Miller
2011-10-20  4:49 ` [PATCH V2 " Eric Dumazet
2011-10-20  8:55   ` David Miller

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).