[PATCH net-next] virtio_net: handle non-napi callers to virtnet_poll_tx

[PATCH net-next] virtio_net: handle non-napi callers to virtnet_poll_tx

[Jonathan Lemon]

From: Jonathan Lemon <bsd@fb.com>

netcons will call napi_poll with a budget of 0, indicating
a non-napi caller (and also called with irqs disabled).  Call
free_old_xmit_skbs() with the is_napi parameter set correctly.

Found by this splat:

WARNING: CPU: 2 PID: 1368 at net/core/skbuff.c:650 skb_release_head_state+0x7f/0x90
RIP: 0010:skb_release_head_state+0x7f/0x90
Code: e7 7f 08 00 80 7b 7f 00 74 f3 48 8b bb d8 00 00 00 5b e9 d4 f2 ff ff 48 83 e7 fe e8 bb f5 01 00

RSP: 0000:ffffc900000e4d18 EFLAGS: 00010006
RAX: ffffffff818fc470 RBX: ffff888235a36d00 RCX: ffff88822df23000
RDX: 0000000000010000 RSI: 0000000000000001 RDI: 0000000000000000
RBP: ffff888235a36d00 R08: 0000000000000002 R09: 00000000ffffdd86
R10: 00000000ffffdd86 R11: 0000000000000000 R12: ffff8882367b4800
R13: 0000000000000001 R14: ffff8882367b4ac8 R15: ffff8882367b4ab8
FS:  00007fdc30f8a700(0000) GS:ffff888237d00000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 00007fdc300ea000 CR3: 000000022992f001 CR4: 0000000000360ee0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
Call Trace:
 <IRQ>
 skb_release_all+0xe/0x30
 napi_consume_skb+0x4c/0x110
 free_old_xmit_skbs+0x3e/0xa0 [virtio_net]
 virtnet_poll_tx+0x7e/0xe0 [virtio_net]
 netpoll_poll_dev+0xba/0x190
 netpoll_send_skb_on_dev+0x1e7/0x240
 netpoll_send_udp+0x2b3/0x3d0
 write_ext_msg+0x1be/0x1d0
 console_unlock+0x227/0x4b0
 vprintk_emit+0xe5/0x1b0
 printk+0x58/0x6f
 smp_apic_timer_interrupt+0x5e/0x120
 apic_timer_interrupt+0xf/0x20
 </IRQ>

Signed-off-by: Jonathan Lemon <jonathan.lemon@gmail.com>
---
 drivers/net/virtio_net.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
index 21b71148c532..59f65ac9e4c7 100644
--- a/drivers/net/virtio_net.c
+++ b/drivers/net/virtio_net.c
@@ -1518,7 +1518,7 @@ static int virtnet_poll_tx(struct napi_struct *napi, int budget)
 
 	txq = netdev_get_tx_queue(vi->dev, index);
 	__netif_tx_lock(txq, raw_smp_processor_id());
-	free_old_xmit_skbs(sq, true);
+	free_old_xmit_skbs(sq, budget != 0);
 	__netif_tx_unlock(txq);
 
 	virtqueue_napi_complete(napi, sq->vq, 0);
-- 
2.24.1

Re: [PATCH net-next] virtio_net: handle non-napi callers to virtnet_poll_tx

[Jakub Kicinski]

On Thu, 8 Oct 2020 11:34:36 -0700 Jonathan Lemon wrote:
> From: Jonathan Lemon <bsd@fb.com>
> 
> netcons will call napi_poll with a budget of 0, indicating
> a non-napi caller (and also called with irqs disabled).  Call
> free_old_xmit_skbs() with the is_napi parameter set correctly.

This is a fix, can we get a Fixes tag, please?

Re: [PATCH net-next] virtio_net: handle non-napi callers to virtnet_poll_tx

[Jonathan Lemon]

On Thu, Oct 08, 2020 at 12:01:02PM -0700, Jakub Kicinski wrote:
> On Thu, 8 Oct 2020 11:34:36 -0700 Jonathan Lemon wrote:
> > From: Jonathan Lemon <bsd@fb.com>
> > 
> > netcons will call napi_poll with a budget of 0, indicating
> > a non-napi caller (and also called with irqs disabled).  Call
> > free_old_xmit_skbs() with the is_napi parameter set correctly.
> 
> This is a fix, can we get a Fixes tag, please?

As best as I can tell:

Fixes: df133f3f9625 ("virtio_net: bulk free tx skbs")

Re: [PATCH net-next] virtio_net: handle non-napi callers to virtnet_poll_tx

[Jakub Kicinski]

On Thu, 8 Oct 2020 11:34:36 -0700 Jonathan Lemon wrote:
> diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
> index 21b71148c532..59f65ac9e4c7 100644
> --- a/drivers/net/virtio_net.c
> +++ b/drivers/net/virtio_net.c
> @@ -1518,7 +1518,7 @@ static int virtnet_poll_tx(struct napi_struct *napi, int budget)
>  
>  	txq = netdev_get_tx_queue(vi->dev, index);
>  	__netif_tx_lock(txq, raw_smp_processor_id());
> -	free_old_xmit_skbs(sq, true);
> +	free_old_xmit_skbs(sq, budget != 0);
>  	__netif_tx_unlock(txq);
>  
>  	virtqueue_napi_complete(napi, sq->vq, 0);

Looks like virtnet_poll_cleantx() needs the same treatment.