| Message ID | 20230810031557.135557-1-yin31149@gmail.com (mailing list archive) |
|---|---|
| State | Changes Requested |
| Delegated to: | Netdev Maintainers |
| Headers | show |
| Series | virtio-net: Zero max_tx_vq field for VIRTIO_NET_CTRL_MQ_HASH_CONFIG case | expand |
On Thu, Aug 10, 2023 at 11:16 AM Hawkins Jiawei <yin31149@gmail.com> wrote: > > Kernel uses `struct virtio_net_ctrl_rss` to save command-specific-data > for both the VIRTIO_NET_CTRL_MQ_HASH_CONFIG and > VIRTIO_NET_CTRL_MQ_RSS_CONFIG commands. This is tricky. > > According to the VirtIO standard, "Field reserved MUST contain zeroes. > It is defined to make the structure to match the layout of > virtio_net_rss_config structure, defined in 5.1.6.5.7.". > > Yet for the VIRTIO_NET_CTRL_MQ_HASH_CONFIG command case, the `max_tx_vq` > field in struct virtio_net_ctrl_rss, which corresponds to the > `reserved` field in struct virtio_net_hash_config, is not zeroed, > thereby violating the VirtIO standard. > > This patch solves this problem by zeroing this field in > virtnet_init_default_rss(). > > Signed-off-by: Hawkins Jiawei <yin31149@gmail.com> > --- Acked-by: Jason Wang <jasowang@redhat.com> Thanks
On Thu, Aug 10, 2023 at 5:16 AM Hawkins Jiawei <yin31149@gmail.com> wrote: > > Kernel uses `struct virtio_net_ctrl_rss` to save command-specific-data > for both the VIRTIO_NET_CTRL_MQ_HASH_CONFIG and > VIRTIO_NET_CTRL_MQ_RSS_CONFIG commands. > > According to the VirtIO standard, "Field reserved MUST contain zeroes. > It is defined to make the structure to match the layout of > virtio_net_rss_config structure, defined in 5.1.6.5.7.". > > Yet for the VIRTIO_NET_CTRL_MQ_HASH_CONFIG command case, the `max_tx_vq` > field in struct virtio_net_ctrl_rss, which corresponds to the > `reserved` field in struct virtio_net_hash_config, is not zeroed, > thereby violating the VirtIO standard. > > This patch solves this problem by zeroing this field in > virtnet_init_default_rss(). > > Signed-off-by: Hawkins Jiawei <yin31149@gmail.com> Acked-by: Eugenio Pérez <eperezma@redhat.com> > --- > > TestStep > ======== > 1. Boot QEMU with one virtio-net-pci net device with `mq` and `hash` > feature on, command line like: > -netdev tap,vhost=off,... > -device virtio-net-pci,mq=on,hash=on,... > > 2. Trigger VIRTIO_NET_CTRL_MQ_HASH_CONFIG command in guest, command > line like: > ethtool -K eth0 rxhash on > > Without this patch, in virtnet_commit_rss_command(), we can see the > `max_tx_vq` field is 1 in gdb like below: > > pwndbg> p vi->ctrl->rss > $1 = { > hash_types = 63, > indirection_table_mask = 0, > unclassified_queue = 0, > indirection_table = {0 <repeats 128 times>}, > max_tx_vq = 1, > hash_key_length = 40 '(', > ... > } > > With this patch, in virtnet_commit_rss_command(), we can see the > `max_tx_vq` field is 0 in gdb like below: > > pwndbg> p vi->ctrl->rss > $1 = { > hash_types = 63, > indirection_table_mask = 0, > unclassified_queue = 0, > indirection_table = {0 <repeats 128 times>}, > max_tx_vq = 0, > hash_key_length = 40 '(', > ... > } > > 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 1270c8d23463..8db38634ae82 100644 > --- a/drivers/net/virtio_net.c > +++ b/drivers/net/virtio_net.c > @@ -2761,7 +2761,7 @@ static void virtnet_init_default_rss(struct virtnet_info *vi) > vi->ctrl->rss.indirection_table[i] = indir_val; > } > > - vi->ctrl->rss.max_tx_vq = vi->curr_queue_pairs; > + vi->ctrl->rss.max_tx_vq = vi->has_rss ? vi->curr_queue_pairs : 0; > vi->ctrl->rss.hash_key_length = vi->rss_key_size; > > netdev_rss_key_fill(vi->ctrl->rss.key, vi->rss_key_size); > -- > 2.34.1 >
On Thu, Aug 10, 2023 at 11:15:57AM +0800, Hawkins Jiawei wrote: > Kernel uses `struct virtio_net_ctrl_rss` to save command-specific-data > for both the VIRTIO_NET_CTRL_MQ_HASH_CONFIG and > VIRTIO_NET_CTRL_MQ_RSS_CONFIG commands. > > According to the VirtIO standard, "Field reserved MUST contain zeroes. > It is defined to make the structure to match the layout of > virtio_net_rss_config structure, defined in 5.1.6.5.7.". > > Yet for the VIRTIO_NET_CTRL_MQ_HASH_CONFIG command case, the `max_tx_vq` > field in struct virtio_net_ctrl_rss, which corresponds to the > `reserved` field in struct virtio_net_hash_config, is not zeroed, > thereby violating the VirtIO standard. > > This patch solves this problem by zeroing this field in > virtnet_init_default_rss(). > > Signed-off-by: Hawkins Jiawei <yin31149@gmail.com> > --- > > TestStep > ======== > 1. Boot QEMU with one virtio-net-pci net device with `mq` and `hash` > feature on, command line like: > -netdev tap,vhost=off,... > -device virtio-net-pci,mq=on,hash=on,... > > 2. Trigger VIRTIO_NET_CTRL_MQ_HASH_CONFIG command in guest, command > line like: > ethtool -K eth0 rxhash on > > Without this patch, in virtnet_commit_rss_command(), we can see the > `max_tx_vq` field is 1 in gdb like below: > > pwndbg> p vi->ctrl->rss > $1 = { > hash_types = 63, > indirection_table_mask = 0, > unclassified_queue = 0, > indirection_table = {0 <repeats 128 times>}, > max_tx_vq = 1, > hash_key_length = 40 '(', > ... > } > > With this patch, in virtnet_commit_rss_command(), we can see the > `max_tx_vq` field is 0 in gdb like below: > > pwndbg> p vi->ctrl->rss > $1 = { > hash_types = 63, > indirection_table_mask = 0, > unclassified_queue = 0, > indirection_table = {0 <repeats 128 times>}, > max_tx_vq = 0, > hash_key_length = 40 '(', > ... > } > > drivers/net/virtio_net.c | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) Fixes tag pls? > diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c > index 1270c8d23463..8db38634ae82 100644 > --- a/drivers/net/virtio_net.c > +++ b/drivers/net/virtio_net.c > @@ -2761,7 +2761,7 @@ static void virtnet_init_default_rss(struct virtnet_info *vi) > vi->ctrl->rss.indirection_table[i] = indir_val; > } > > - vi->ctrl->rss.max_tx_vq = vi->curr_queue_pairs; > + vi->ctrl->rss.max_tx_vq = vi->has_rss ? vi->curr_queue_pairs : 0; > vi->ctrl->rss.hash_key_length = vi->rss_key_size; > > netdev_rss_key_fill(vi->ctrl->rss.key, vi->rss_key_size); > -- > 2.34.1
On Thu, Aug 10, 2023 at 11:15:57AM +0800, Hawkins Jiawei wrote: > Kernel uses `struct virtio_net_ctrl_rss` to save command-specific-data > for both the VIRTIO_NET_CTRL_MQ_HASH_CONFIG and > VIRTIO_NET_CTRL_MQ_RSS_CONFIG commands. > > According to the VirtIO standard, "Field reserved MUST contain zeroes. > It is defined to make the structure to match the layout of > virtio_net_rss_config structure, defined in 5.1.6.5.7.". > > Yet for the VIRTIO_NET_CTRL_MQ_HASH_CONFIG command case, the `max_tx_vq` > field in struct virtio_net_ctrl_rss, which corresponds to the > `reserved` field in struct virtio_net_hash_config, is not zeroed, > thereby violating the VirtIO standard. > > This patch solves this problem by zeroing this field in > virtnet_init_default_rss(). > > Signed-off-by: Hawkins Jiawei <yin31149@gmail.com> Fixes: c7114b1249fa ("drivers/net/virtio_net: Added basic RSS support.") Cc: Andrew Melnychenko <andrew@daynix.com> Acked-by: Michael S. Tsirkin <mst@redhat.com> And this is stable material I believe. > --- > > TestStep > ======== > 1. Boot QEMU with one virtio-net-pci net device with `mq` and `hash` > feature on, command line like: > -netdev tap,vhost=off,... > -device virtio-net-pci,mq=on,hash=on,... > > 2. Trigger VIRTIO_NET_CTRL_MQ_HASH_CONFIG command in guest, command > line like: > ethtool -K eth0 rxhash on > > Without this patch, in virtnet_commit_rss_command(), we can see the > `max_tx_vq` field is 1 in gdb like below: > > pwndbg> p vi->ctrl->rss > $1 = { > hash_types = 63, > indirection_table_mask = 0, > unclassified_queue = 0, > indirection_table = {0 <repeats 128 times>}, > max_tx_vq = 1, > hash_key_length = 40 '(', > ... > } > > With this patch, in virtnet_commit_rss_command(), we can see the > `max_tx_vq` field is 0 in gdb like below: > > pwndbg> p vi->ctrl->rss > $1 = { > hash_types = 63, > indirection_table_mask = 0, > unclassified_queue = 0, > indirection_table = {0 <repeats 128 times>}, > max_tx_vq = 0, > hash_key_length = 40 '(', > ... > } > > 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 1270c8d23463..8db38634ae82 100644 > --- a/drivers/net/virtio_net.c > +++ b/drivers/net/virtio_net.c > @@ -2761,7 +2761,7 @@ static void virtnet_init_default_rss(struct virtnet_info *vi) > vi->ctrl->rss.indirection_table[i] = indir_val; > } > > - vi->ctrl->rss.max_tx_vq = vi->curr_queue_pairs; > + vi->ctrl->rss.max_tx_vq = vi->has_rss ? vi->curr_queue_pairs : 0; > vi->ctrl->rss.hash_key_length = vi->rss_key_size; > > netdev_rss_key_fill(vi->ctrl->rss.key, vi->rss_key_size); > -- > 2.34.1
On 2023/8/10 16:51, Michael S. Tsirkin wrote: > On Thu, Aug 10, 2023 at 11:15:57AM +0800, Hawkins Jiawei wrote: >> Kernel uses `struct virtio_net_ctrl_rss` to save command-specific-data >> for both the VIRTIO_NET_CTRL_MQ_HASH_CONFIG and >> VIRTIO_NET_CTRL_MQ_RSS_CONFIG commands. >> >> According to the VirtIO standard, "Field reserved MUST contain zeroes. >> It is defined to make the structure to match the layout of >> virtio_net_rss_config structure, defined in 5.1.6.5.7.". >> >> Yet for the VIRTIO_NET_CTRL_MQ_HASH_CONFIG command case, the `max_tx_vq` >> field in struct virtio_net_ctrl_rss, which corresponds to the >> `reserved` field in struct virtio_net_hash_config, is not zeroed, >> thereby violating the VirtIO standard. >> >> This patch solves this problem by zeroing this field in >> virtnet_init_default_rss(). >> >> Signed-off-by: Hawkins Jiawei <yin31149@gmail.com> > > > > Fixes: c7114b1249fa ("drivers/net/virtio_net: Added basic RSS support.") > Cc: Andrew Melnychenko <andrew@daynix.com> > Acked-by: Michael S. Tsirkin <mst@redhat.com> > > And this is stable material I believe. Hi Michael, Thank you for the reminder, I will send the v2 patch with all these tags included. Thanks! > > > >> --- >> >> TestStep >> ======== >> 1. Boot QEMU with one virtio-net-pci net device with `mq` and `hash` >> feature on, command line like: >> -netdev tap,vhost=off,... >> -device virtio-net-pci,mq=on,hash=on,... >> >> 2. Trigger VIRTIO_NET_CTRL_MQ_HASH_CONFIG command in guest, command >> line like: >> ethtool -K eth0 rxhash on >> >> Without this patch, in virtnet_commit_rss_command(), we can see the >> `max_tx_vq` field is 1 in gdb like below: >> >> pwndbg> p vi->ctrl->rss >> $1 = { >> hash_types = 63, >> indirection_table_mask = 0, >> unclassified_queue = 0, >> indirection_table = {0 <repeats 128 times>}, >> max_tx_vq = 1, >> hash_key_length = 40 '(', >> ... >> } >> >> With this patch, in virtnet_commit_rss_command(), we can see the >> `max_tx_vq` field is 0 in gdb like below: >> >> pwndbg> p vi->ctrl->rss >> $1 = { >> hash_types = 63, >> indirection_table_mask = 0, >> unclassified_queue = 0, >> indirection_table = {0 <repeats 128 times>}, >> max_tx_vq = 0, >> hash_key_length = 40 '(', >> ... >> } >> >> 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 1270c8d23463..8db38634ae82 100644 >> --- a/drivers/net/virtio_net.c >> +++ b/drivers/net/virtio_net.c >> @@ -2761,7 +2761,7 @@ static void virtnet_init_default_rss(struct virtnet_info *vi) >> vi->ctrl->rss.indirection_table[i] = indir_val; >> } >> >> - vi->ctrl->rss.max_tx_vq = vi->curr_queue_pairs; >> + vi->ctrl->rss.max_tx_vq = vi->has_rss ? vi->curr_queue_pairs : 0; >> vi->ctrl->rss.hash_key_length = vi->rss_key_size; >> >> netdev_rss_key_fill(vi->ctrl->rss.key, vi->rss_key_size); >> -- >> 2.34.1 >
diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c index 1270c8d23463..8db38634ae82 100644 --- a/drivers/net/virtio_net.c +++ b/drivers/net/virtio_net.c @@ -2761,7 +2761,7 @@ static void virtnet_init_default_rss(struct virtnet_info *vi) vi->ctrl->rss.indirection_table[i] = indir_val; } - vi->ctrl->rss.max_tx_vq = vi->curr_queue_pairs; + vi->ctrl->rss.max_tx_vq = vi->has_rss ? vi->curr_queue_pairs : 0; vi->ctrl->rss.hash_key_length = vi->rss_key_size; netdev_rss_key_fill(vi->ctrl->rss.key, vi->rss_key_size);
Kernel uses `struct virtio_net_ctrl_rss` to save command-specific-data for both the VIRTIO_NET_CTRL_MQ_HASH_CONFIG and VIRTIO_NET_CTRL_MQ_RSS_CONFIG commands. According to the VirtIO standard, "Field reserved MUST contain zeroes. It is defined to make the structure to match the layout of virtio_net_rss_config structure, defined in 5.1.6.5.7.". Yet for the VIRTIO_NET_CTRL_MQ_HASH_CONFIG command case, the `max_tx_vq` field in struct virtio_net_ctrl_rss, which corresponds to the `reserved` field in struct virtio_net_hash_config, is not zeroed, thereby violating the VirtIO standard. This patch solves this problem by zeroing this field in virtnet_init_default_rss(). Signed-off-by: Hawkins Jiawei <yin31149@gmail.com> --- TestStep ======== 1. Boot QEMU with one virtio-net-pci net device with `mq` and `hash` feature on, command line like: -netdev tap,vhost=off,... -device virtio-net-pci,mq=on,hash=on,... 2. Trigger VIRTIO_NET_CTRL_MQ_HASH_CONFIG command in guest, command line like: ethtool -K eth0 rxhash on Without this patch, in virtnet_commit_rss_command(), we can see the `max_tx_vq` field is 1 in gdb like below: pwndbg> p vi->ctrl->rss $1 = { hash_types = 63, indirection_table_mask = 0, unclassified_queue = 0, indirection_table = {0 <repeats 128 times>}, max_tx_vq = 1, hash_key_length = 40 '(', ... } With this patch, in virtnet_commit_rss_command(), we can see the `max_tx_vq` field is 0 in gdb like below: pwndbg> p vi->ctrl->rss $1 = { hash_types = 63, indirection_table_mask = 0, unclassified_queue = 0, indirection_table = {0 <repeats 128 times>}, max_tx_vq = 0, hash_key_length = 40 '(', ... } drivers/net/virtio_net.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)