diff mbox series

vhost: disable host TSO for linear buffers without extbuf

Message ID 20191017150019.19191-1-i.maximets@ovn.org (mailing list archive)
State Accepted, archived
Delegated to: Maxime Coquelin
Headers
Series vhost: disable host TSO for linear buffers without extbuf |

Checks

Context Check Description
ci/checkpatch success coding style OK
ci/Intel-compilation success Compilation OK
ci/iol-intel-Performance success Performance Testing PASS
ci/iol-compilation success Compile Testing PASS
ci/iol-mellanox-Performance success Performance Testing PASS
ci/travis-robot warning Travis build: failed

Commit Message

Ilya Maximets Oct. 17, 2019, 3 p.m. UTC 
  If linear buffers requested and external buffers are not, vhost
will not be able to receive any buffer that doesn't fit in a
single mbuf.  Moreover, if such a buffer will appear in a vring
it will never be dequeued and the whole vring will become dead
breaking the network connection.

Disable segmentation offloading from the host side to avoid
having such a big buffers.

Cc: Flavio Leitner <fbl@sysclose.org>

Fixes: 5005bcda7123 ("vhost: add support for large buffers")
Signed-off-by: Ilya Maximets <i.maximets@ovn.org>
---

There is still an assumption that users are sane enough to have
MTU sized mbufs in a memory pool and that guest will not change
MTU to higher values.

We, probably, still need to have a check on dequeue path and
drop oversized buffers in case of linear buffers to avoid stuck
of the virtqueue.  Or simply drop support of '+linear -extbuf'
case.

Note: Only compile tested due to lack of HW.

 lib/librte_vhost/socket.c | 18 ++++++++++++++++++
 1 file changed, 18 insertions(+)

Comments

Maxime Coquelin Oct. 24, 2019, 9:38 a.m. UTC | #1 
On 10/17/19 5:00 PM, Ilya Maximets wrote:
> If linear buffers requested and external buffers are not, vhost
> will not be able to receive any buffer that doesn't fit in a
> single mbuf.  Moreover, if such a buffer will appear in a vring
> it will never be dequeued and the whole vring will become dead
> breaking the network connection.
> 
> Disable segmentation offloading from the host side to avoid
> having such a big buffers.
> 
> Cc: Flavio Leitner <fbl@sysclose.org>
> 
> Fixes: 5005bcda7123 ("vhost: add support for large buffers")
> Signed-off-by: Ilya Maximets <i.maximets@ovn.org>
> ---
> 
> There is still an assumption that users are sane enough to have
> MTU sized mbufs in a memory pool and that guest will not change
> MTU to higher values.
> 
> We, probably, still need to have a check on dequeue path and
> drop oversized buffers in case of linear buffers to avoid stuck
> of the virtqueue.  Or simply drop support of '+linear -extbuf'
> case.
> 
> Note: Only compile tested due to lack of HW.
> 
>  lib/librte_vhost/socket.c | 18 ++++++++++++++++++
>  1 file changed, 18 insertions(+)
> 


Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>

Thanks,
Maxime
Maxime Coquelin Oct. 24, 2019, 10:18 a.m. UTC | #2 
On 10/24/19 11:38 AM, Maxime Coquelin wrote:
> 
> 
> On 10/17/19 5:00 PM, Ilya Maximets wrote:
>> If linear buffers requested and external buffers are not, vhost
>> will not be able to receive any buffer that doesn't fit in a
>> single mbuf.  Moreover, if such a buffer will appear in a vring
>> it will never be dequeued and the whole vring will become dead
>> breaking the network connection.
>>
>> Disable segmentation offloading from the host side to avoid
>> having such a big buffers.
>>
>> Cc: Flavio Leitner <fbl@sysclose.org>
>>
>> Fixes: 5005bcda7123 ("vhost: add support for large buffers")
>> Signed-off-by: Ilya Maximets <i.maximets@ovn.org>
>> ---
>>
>> There is still an assumption that users are sane enough to have
>> MTU sized mbufs in a memory pool and that guest will not change
>> MTU to higher values.
>>
>> We, probably, still need to have a check on dequeue path and
>> drop oversized buffers in case of linear buffers to avoid stuck
>> of the virtqueue.  Or simply drop support of '+linear -extbuf'
>> case.
>>
>> Note: Only compile tested due to lack of HW.
>>
>>  lib/librte_vhost/socket.c | 18 ++++++++++++++++++
>>  1 file changed, 18 insertions(+)
>>
> 
> 
> Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
> 
> Thanks,
> Maxime
> 


Applied to dpdk-next-virtio/master.

Thanks,
Maxime
diff mbox series

Patch

diff --git a/lib/librte_vhost/socket.c b/lib/librte_vhost/socket.c
index 2d3d20804..1036eaad4 100644
--- a/lib/librte_vhost/socket.c
+++ b/lib/librte_vhost/socket.c
@@ -933,6 +933,24 @@  rte_vhost_driver_register(const char *path, uint64_t flags)
 			~(1ULL << VHOST_USER_PROTOCOL_F_PAGEFAULT);
 	}
 
+	/*
+	 * We'll not be able to receive a buffer from guest in linear mode
+	 * without external buffer if it will not fit in a single mbuf, which is
+	 * likely if segmentation offloading enabled.
+	 */
+	if (vsocket->linearbuf && !vsocket->extbuf) {
+		uint64_t seg_offload_features =
+				(1ULL << VIRTIO_NET_F_HOST_TSO4) |
+				(1ULL << VIRTIO_NET_F_HOST_TSO6) |
+				(1ULL << VIRTIO_NET_F_HOST_UFO);
+
+		RTE_LOG(INFO, VHOST_CONFIG,
+			"Linear buffers requested without external buffers, "
+			"disabling host segmentation offloading support\n");
+		vsocket->supported_features &= ~seg_offload_features;
+		vsocket->features &= ~seg_offload_features;
+	}
+
 	if (!(flags & RTE_VHOST_USER_IOMMU_SUPPORT)) {
 		vsocket->supported_features &= ~(1ULL << VIRTIO_F_IOMMU_PLATFORM);
 		vsocket->features &= ~(1ULL << VIRTIO_F_IOMMU_PLATFORM);