| Message ID | 20180313194141.12913-1-doshir@vmware.com |
|---|---|
| State | Changes Requested, archived |
| Delegated to: | David Miller |
| Headers | show |
| Series | [net,1/2] vmxnet3: avoid xmit reset due to a race in vmxnet3 | expand |
From: Ronak Doshi <doshir@vmware.com> Date: Tue, 13 Mar 2018 12:41:41 -0700 > 'Commit 45dac1d6ea04 ("vmxnet3: Changes for vmxnet3 adapter version 2 > (fwd)")' introduced a flag "lro" in structure vmxnet3_adapter which is > used to indicate whether LRO is enabled or not. However, the patch > did not set the flag and hence it was never exercised. > > So, when LRO is enabled, it resulted in poor TCP performance due to > delayed acks. This issue is seen with packets which are larger than > the mss getting a delayed ack rather than an immediate ack, thus > resulting in high latency. > > This patch removes the lro flag and directly uses device features > against NETIF_F_LRO to check if lro is enabled. > > Reported-by: Rachel Lunnon <rachel_lunnon@stormagic.com> > Signed-off-by: Ronak Doshi <doshir@vmware.com> > Acked-by: Shrikrishna Khare <skhare@vmware.com> Please add an appropriate "Fixes: " tag for the commit which introduced this problem. This helps with -stable submissions and bug fix triage.
diff --git a/drivers/net/vmxnet3/vmxnet3_drv.c b/drivers/net/vmxnet3/vmxnet3_drv.c index 052eef2f729f..86c4d6e4dfaa 100644 --- a/drivers/net/vmxnet3/vmxnet3_drv.c +++ b/drivers/net/vmxnet3/vmxnet3_drv.c @@ -1473,7 +1473,8 @@ vmxnet3_rq_rx_complete(struct vmxnet3_rx_queue *rq, vmxnet3_rx_csum(adapter, skb, (union Vmxnet3_GenericDesc *)rcd); skb->protocol = eth_type_trans(skb, adapter->netdev); - if (!rcd->tcp || !adapter->lro) + if (!rcd->tcp || + !(adapter->netdev->features & NETIF_F_LRO)) goto not_lro; if (segCnt != 0 && mss != 0) { diff --git a/drivers/net/vmxnet3/vmxnet3_int.h b/drivers/net/vmxnet3/vmxnet3_int.h index b94fdfd0b6f1..99387a4a20a8 100644 --- a/drivers/net/vmxnet3/vmxnet3_int.h +++ b/drivers/net/vmxnet3/vmxnet3_int.h @@ -69,10 +69,10 @@ /* * Version numbers */ -#define VMXNET3_DRIVER_VERSION_STRING "1.4.12.0-k" +#define VMXNET3_DRIVER_VERSION_STRING "1.4.13.0-k" /* a 32-bit int, each byte encode a verion number in VMXNET3_DRIVER_VERSION */ -#define VMXNET3_DRIVER_VERSION_NUM 0x01040c00 +#define VMXNET3_DRIVER_VERSION_NUM 0x01040d00 #if defined(CONFIG_PCI_MSI) /* RSS only makes sense if MSI-X is supported. */ @@ -343,7 +343,6 @@ struct vmxnet3_adapter { u8 version; bool rxcsum; - bool lro; #ifdef VMXNET3_RSS struct UPT1_RSSConf *rss_conf;
'Commit 45dac1d6ea04 ("vmxnet3: Changes for vmxnet3 adapter version 2 (fwd)")' introduced a flag "lro" in structure vmxnet3_adapter which is used to indicate whether LRO is enabled or not. However, the patch did not set the flag and hence it was never exercised. So, when LRO is enabled, it resulted in poor TCP performance due to delayed acks. This issue is seen with packets which are larger than the mss getting a delayed ack rather than an immediate ack, thus resulting in high latency. This patch removes the lro flag and directly uses device features against NETIF_F_LRO to check if lro is enabled. Reported-by: Rachel Lunnon <rachel_lunnon@stormagic.com> Signed-off-by: Ronak Doshi <doshir@vmware.com> Acked-by: Shrikrishna Khare <skhare@vmware.com> --- drivers/net/vmxnet3/vmxnet3_drv.c | 3 ++- drivers/net/vmxnet3/vmxnet3_int.h | 5 ++--- 2 files changed, 4 insertions(+), 4 deletions(-)