similar to: [PATCH v2 1/2] virtio_net: fix error handling for mergeable buffers

Eric Dumazet noticed that if we encounter an error when processing a mergeable buffer, we don't dequeue all of the buffers from this packet, the result is almost sure to be loss of networking. Jason Wang noticed that we also leak a page and that we don't decrement the rq buf count, so we won't repost buffers (a resource leak). Fix both issues, and also make the logic a bit more

Eric Dumazet noticed that if we encounter an error when processing a mergeable buffer, we don't dequeue all of the buffers from this packet, the result is almost sure to be loss of networking. Jason Wang noticed that we also leak a page and that we don't decrement the rq buf count, so we won't repost buffers (a resource leak). Fix both issues, and also make the logic a bit more

Commit 2613af0ed18a ("virtio_net: migrate mergeable rx buffers to page frag allocators") changed the mergeable receive buffer size from PAGE_SIZE to MTU-size. However, the merge buffer size does not take into account the size of the virtio-net header. Consequently, packets that are MTU-size will take two buffers intead of one (to store the virtio-net header), substantially decreasing the

Commit 2613af0ed18a ("virtio_net: migrate mergeable rx buffers to page frag allocators") changed the mergeable receive buffer size from PAGE_SIZE to MTU-size. However, the merge buffer size does not take into account the size of the virtio-net header. Consequently, packets that are MTU-size will take two buffers intead of one (to store the virtio-net header), substantially decreasing the

On 11/13/2013 06:21 AM, Michael Dalton wrote: > Commit 2613af0ed18a ("virtio_net: migrate mergeable rx buffers to page frag > allocators") changed the mergeable receive buffer size from PAGE_SIZE to > MTU-size, introducing a single-stream regression for benchmarks with large > average packet size. There is no single optimal buffer size for all workloads. > For workloads

On 11/13/2013 06:21 AM, Michael Dalton wrote: > Commit 2613af0ed18a ("virtio_net: migrate mergeable rx buffers to page frag > allocators") changed the mergeable receive buffer size from PAGE_SIZE to > MTU-size, introducing a single-stream regression for benchmarks with large > average packet size. There is no single optimal buffer size for all workloads. > For workloads

When mergeable buffer were used, we only put the first page buf leave the rest of buffers in the virt queue. This will cause the driver could not get the correct head buffer any more. Fix this by dropping the rest of buffers for this packet. The bug was introduced by commit 9ab86bbcf8be755256f0a5e994e0b38af6b4d399 (virtio_net: Defer skb allocation in receive path). Cc: Rusty Russell <rusty at

When mergeable buffer were used, we only put the first page buf leave the rest of buffers in the virt queue. This will cause the driver could not get the correct head buffer any more. Fix this by dropping the rest of buffers for this packet. The bug was introduced by commit 9ab86bbcf8be755256f0a5e994e0b38af6b4d399 (virtio_net: Defer skb allocation in receive path). Cc: Rusty Russell <rusty at

On Wed, Nov 20, 2013 at 05:07:26PM +0800, Jason Wang wrote: > We need decrease the rq->num after we can get a buf through > virtqueue_get_buf() even if we could not allocate frag skb. Otherwise, the > refill routine won't be triggered under heavy memory stress since the driver may > still think there's enough room. > > This bug was introduced by commit

On Wed, Nov 20, 2013 at 05:07:26PM +0800, Jason Wang wrote: > We need decrease the rq->num after we can get a buf through > virtqueue_get_buf() even if we could not allocate frag skb. Otherwise, the > refill routine won't be triggered under heavy memory stress since the driver may > still think there's enough room. > > This bug was introduced by commit

The virtio_net driver's mergeable receive buffer allocator uses 4KB packet buffers. For MTU-sized traffic, SKB truesize is > 4KB but only ~1500 bytes of the buffer is used to store packet data, reducing the effective TCP window size substantially. This patch addresses the performance concerns with mergeable receive buffers by allocating MTU-sized packet buffers using page frag allocators.

The virtio_net driver's mergeable receive buffer allocator uses 4KB packet buffers. For MTU-sized traffic, SKB truesize is > 4KB but only ~1500 bytes of the buffer is used to store packet data, reducing the effective TCP window size substantially. This patch addresses the performance concerns with mergeable receive buffers by allocating MTU-sized packet buffers using page frag allocators.

On 11/28/2013 12:31 AM, Michael S. Tsirkin wrote: > Eric Dumazet noticed that if we encounter an error > when processing a mergeable buffer, we don't > dequeue all of the buffers from this packet, > the result is almost sure to be loss of networking. > > Jason Wang noticed that we also leak a page and that we don't decrement > the rq buf count, so we won't repost

skb_page_frag_refill currently permits only order-0 page allocs unless GFP_WAIT is used. Change skb_page_frag_refill to attempt higher-order page allocations whether or not GFP_WAIT is used. If memory cannot be allocated, the allocator will fall back to successively smaller page allocs (down to order-0 page allocs). This change brings skb_page_frag_refill in line with the existing page allocation

skb_page_frag_refill currently permits only order-0 page allocs unless GFP_WAIT is used. Change skb_page_frag_refill to attempt higher-order page allocations whether or not GFP_WAIT is used. If memory cannot be allocated, the allocator will fall back to successively smaller page allocs (down to order-0 page allocs). This change brings skb_page_frag_refill in line with the existing page allocation

----- ???? ----- > On Wed, Nov 20, 2013 at 05:07:25PM +0800, Jason Wang wrote: > > When mergeable buffer were used, we only put the first page buf leave the > > rest > > of buffers in the virt queue. This will cause the driver could not get the > > correct head buffer any more. Fix this by dropping the rest of buffers for > > this > > packet. > > >

----- ???? ----- > On Wed, Nov 20, 2013 at 05:07:25PM +0800, Jason Wang wrote: > > When mergeable buffer were used, we only put the first page buf leave the > > rest > > of buffers in the virt queue. This will cause the driver could not get the > > correct head buffer any more. Fix this by dropping the rest of buffers for > > this > > packet. > > >

Commit 2613af0ed18a ("virtio_net: migrate mergeable rx buffers to page frag allocators") changed the mergeable receive buffer size from PAGE_SIZE to MTU-size, introducing a single-stream regression for benchmarks with large average packet size. There is no single optimal buffer size for all workloads. For workloads with packet size <= MTU bytes, MTU + virtio-net header-sized buffers

The virtio-net driver currently uses netdev_alloc_frag() for GFP_ATOMIC mergeable rx buffer allocations. This commit migrates virtio-net to use per-receive queue page frags for GFP_ATOMIC allocation. This change unifies mergeable rx buffer memory allocation, which now will use skb_refill_frag() for both atomic and GFP-WAIT buffer allocations. To address fragmentation concerns, if after buffer

The purpose of this patch is to simplify the receive_mergeable(). Separate all the logic of XDP into a function. Signed-off-by: Xuan Zhuo <xuanzhuo at linux.alibaba.com> --- drivers/net/virtio_net.c | 128 +++++++++++++++++++++++---------------- 1 file changed, 76 insertions(+), 52 deletions(-) diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c index