e1000_xmit_frame is expected to stop the queue and dispatch frames to
hardware if there is not sufficient space for the next frame in the
buffer, but sometimes it failed to do so because the estimated maxmium
size of frame was wrong. As the consequence, the later invocation of
e1000_xmit_frame failed with NETDEV_TX_BUSY, and the frame in the buffer
remained forever, resulting in a watchdog failure.
This change fixes the estimated size by making it match with the
condition for NETDEV_TX_BUSY. Apparently, the old estimation failed to
account for the following lines which determines the space requirement
for not causing NETDEV_TX_BUSY:
```
/* reserve a descriptor for the offload context */
if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
count++;
count++;
count += DIV_ROUND_UP(len, adapter->tx_fifo_limit);
```
This issue was found when running http-stress02 test included in Linux
Test Project 20220930 on QEMU with the following commandline:
```
qemu-system-x86_64 -M q35,accel=kvm -m 8G -smp 8
-drive if=virtio,format=raw,file=root.img,file.locking=on
-device e1000e,netdev=netdev
-netdev tap,script=ifup,downscript=no,id=netdev
```
Fixes: bc7f75fa9788 ("[E1000E]: New pci-express e1000 driver (currently for ICH9 devices only)")
Signed-off-by: Akihiko Odaki <[email protected]>
---
drivers/net/ethernet/intel/e1000e/netdev.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/drivers/net/ethernet/intel/e1000e/netdev.c b/drivers/net/ethernet/intel/e1000e/netdev.c
index 49e926959ad3..55cf2f62bb30 100644
--- a/drivers/net/ethernet/intel/e1000e/netdev.c
+++ b/drivers/net/ethernet/intel/e1000e/netdev.c
@@ -5936,9 +5936,9 @@ static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
e1000_tx_queue(tx_ring, tx_flags, count);
/* Make sure there is space in the ring for the next send. */
e1000_maybe_stop_tx(tx_ring,
- (MAX_SKB_FRAGS *
+ ((MAX_SKB_FRAGS + 1) *
DIV_ROUND_UP(PAGE_SIZE,
- adapter->tx_fifo_limit) + 2));
+ adapter->tx_fifo_limit) + 4));
if (!netdev_xmit_more() ||
netif_xmit_stopped(netdev_get_tx_queue(netdev, 0))) {
--
2.37.3
> -----Original Message-----
> From: Intel-wired-lan <[email protected]> On Behalf Of
> Akihiko Odaki
> Sent: Friday, October 28, 2022 6:30 PM
> Cc: Paul Menzel <[email protected]>; [email protected];
> [email protected]; Yuri Benditovich
> <[email protected]>; Eric Dumazet <[email protected]>;
> [email protected]; Yan Vugenfirer <[email protected]>; Jakub
> Kicinski <[email protected]>; Paolo Abeni <[email protected]>; David S .
> Miller <[email protected]>
> Subject: [Intel-wired-lan] [PATCH v2] e1000e: Fix TX dispatch condition
>
> e1000_xmit_frame is expected to stop the queue and dispatch frames to
> hardware if there is not sufficient space for the next frame in the buffer, but
> sometimes it failed to do so because the estimated maxmium size of frame
> was wrong. As the consequence, the later invocation of e1000_xmit_frame
> failed with NETDEV_TX_BUSY, and the frame in the buffer remained forever,
> resulting in a watchdog failure.
>
> This change fixes the estimated size by making it match with the condition for
> NETDEV_TX_BUSY. Apparently, the old estimation failed to account for the
> following lines which determines the space requirement for not causing
> NETDEV_TX_BUSY:
> ```
> /* reserve a descriptor for the offload context */
> if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
> count++;
> count++;
>
> count += DIV_ROUND_UP(len, adapter->tx_fifo_limit);
> ```
>
> This issue was found when running http-stress02 test included in Linux Test
> Project 20220930 on QEMU with the following commandline:
> ```
> qemu-system-x86_64 -M q35,accel=kvm -m 8G -smp 8
> -drive if=virtio,format=raw,file=root.img,file.locking=on
> -device e1000e,netdev=netdev
> -netdev tap,script=ifup,downscript=no,id=netdev
> ```
>
> Fixes: bc7f75fa9788 ("[E1000E]: New pci-express e1000 driver (currently for
> ICH9 devices only)")
> Signed-off-by: Akihiko Odaki <[email protected]>
> ---
> drivers/net/ethernet/intel/e1000e/netdev.c | 4 ++--
> 1 file changed, 2 insertions(+), 2 deletions(-)
>
Tested-by: Gurucharan <[email protected]> (A Contingent worker at Intel)
On 10/28/2022 16:00, Akihiko Odaki wrote:
> e1000_xmit_frame is expected to stop the queue and dispatch frames to
> hardware if there is not sufficient space for the next frame in the
> buffer, but sometimes it failed to do so because the estimated maxmium
> size of frame was wrong. As the consequence, the later invocation of
> e1000_xmit_frame failed with NETDEV_TX_BUSY, and the frame in the buffer
> remained forever, resulting in a watchdog failure.
>
> This change fixes the estimated size by making it match with the
> condition for NETDEV_TX_BUSY. Apparently, the old estimation failed to
> account for the following lines which determines the space requirement
> for not causing NETDEV_TX_BUSY:
> ```
> /* reserve a descriptor for the offload context */
> if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
> count++;
> count++;
>
> count += DIV_ROUND_UP(len, adapter->tx_fifo_limit);
> ```
>
> This issue was found when running http-stress02 test included in Linux
> Test Project 20220930 on QEMU with the following commandline:
> ```
> qemu-system-x86_64 -M q35,accel=kvm -m 8G -smp 8
> -drive if=virtio,format=raw,file=root.img,file.locking=on
> -device e1000e,netdev=netdev
> -netdev tap,script=ifup,downscript=no,id=netdev
> ```
>
> Fixes: bc7f75fa9788 ("[E1000E]: New pci-express e1000 driver (currently for ICH9 devices only)")
> Signed-off-by: Akihiko Odaki <[email protected]>
> ---
> drivers/net/ethernet/intel/e1000e/netdev.c | 4 ++--
> 1 file changed, 2 insertions(+), 2 deletions(-)
Tested-by: Naama Meir <[email protected]>