# \[PATCH 1/1\] virtio/vsock: Make vsock virtio packet buff size configurable

- Subsystems: NETWORKING \[GENERAL\], VIRTIO AND VHOST VSOCK DRIVER, VM SOCKETS (AF\_VSOCK), THE REST, VIRTIO CORE
- Maintainers: "David S. Miller", Eric Dumazet, Jakub Kicinski, Paolo Abeni, Stefan Hajnoczi, Stefano Garzarella, Linus Torvalds, "Michael S. Tsirkin", Jason Wang, Eugenio Pérez
- List: netdev
- 4 messages, 3 authors
- Page: https://ratatoskr.run/netdev/2021/07/1453484/t

## \[PATCH 1/1\] virtio/vsock: Make vsock virtio packet buff size configurable

- From: Lee Jones \<hidden\>
- Date: 2021-07-21 14:30:23 
- Message: https://ratatoskr.run/netdev/2021/07/1453484.md

```text
From: Ram Muthiah <redacted>

After a virtual device has been running for some time, the SLAB
sustains ever increasing fragmentation. Contributing to this
fragmentation are the virtio packet buffer allocations which
are a drain on 64Kb compound pages. Eventually these can't be
allocated due to fragmentation.

To enable successful allocations for this packet buffer, the
packet buffer's size needs to be reduced.

In order to enable a reduction without impacting current users,
this variable is being exposed as a command line parameter.

Cc: "Michael S. Tsirkin" <mst@redhat.com>
Cc: Jason Wang <redacted>
Cc: Stefan Hajnoczi <stefanha@redhat.com>
Cc: Stefano Garzarella <sgarzare@redhat.com>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Jakub Kicinski <kuba@kernel.org>
Cc: <redacted>
Cc: <redacted>
Cc: <redacted>
Signed-off-by: Ram Muthiah <redacted>
Signed-off-by: Lee Jones <redacted>
---
 include/linux/virtio_vsock.h            | 4 +++-
 net/vmw_vsock/virtio_transport_common.c | 4 ++++
 2 files changed, 7 insertions(+), 1 deletion(-)

diff --git a/include/linux/virtio_vsock.h b/include/linux/virtio_vsock.h
index 35d7eedb5e8e4..8c77d60a74d34 100644
--- a/include/linux/virtio_vsock.h
+++ b/include/linux/virtio_vsock.h
@@ -7,9 +7,11 @@
 #include <net/sock.h>
 #include <net/af_vsock.h>
 
+extern uint virtio_transport_max_vsock_pkt_buf_size;
+
 #define VIRTIO_VSOCK_DEFAULT_RX_BUF_SIZE	(1024 * 4)
 #define VIRTIO_VSOCK_MAX_BUF_SIZE		0xFFFFFFFFUL
-#define VIRTIO_VSOCK_MAX_PKT_BUF_SIZE		(1024 * 64)
+#define VIRTIO_VSOCK_MAX_PKT_BUF_SIZE		virtio_transport_max_vsock_pkt_buf_size
 
 enum {
 	VSOCK_VQ_RX     = 0, /* for host to guest data */
diff --git a/net/vmw_vsock/virtio_transport_common.c b/net/vmw_vsock/virtio_transport_common.c
index 169ba8b72a630..d0d913afec8b6 100644
--- a/net/vmw_vsock/virtio_transport_common.c
+++ b/net/vmw_vsock/virtio_transport_common.c
@@ -26,6 +26,10 @@
 /* Threshold for detecting small packets to copy */
 #define GOOD_COPY_LEN  128
 
+uint virtio_transport_max_vsock_pkt_buf_size = 1024 * 64;
+module_param(virtio_transport_max_vsock_pkt_buf_size, uint, 0444);
+EXPORT_SYMBOL_GPL(virtio_transport_max_vsock_pkt_buf_size);
+
 static const struct virtio_transport *
 virtio_transport_get_ops(struct vsock_sock *vsk)
 {
-- 
2.32.0.402.g57bb445576-goog

```

## Re: \[PATCH 1/1\] virtio/vsock: Make vsock virtio packet buff size configurable

- From: Stefano Garzarella \<sgarzare@redhat.com\>
- Date: 2021-07-22 12:55:32 
- Message: https://ratatoskr.run/netdev/2021/07/1454509.md

```text
On Wed, Jul 21, 2021 at 03:30:00PM +0100, Lee Jones wrote:
>From: Ram Muthiah <redacted>
>
>After a virtual device has been running for some time, the SLAB
>sustains ever increasing fragmentation. Contributing to this
>fragmentation are the virtio packet buffer allocations which
>are a drain on 64Kb compound pages. Eventually these can't be
>allocated due to fragmentation.
>
>To enable successful allocations for this packet buffer, the
>packet buffer's size needs to be reduced.
>
>In order to enable a reduction without impacting current users,
>this variable is being exposed as a command line parameter.
>
>Cc: "Michael S. Tsirkin" <mst@redhat.com>
>Cc: Jason Wang <redacted>
>Cc: Stefan Hajnoczi <stefanha@redhat.com>
>Cc: Stefano Garzarella <sgarzare@redhat.com>
>Cc: "David S. Miller" <davem@davemloft.net>
>Cc: Jakub Kicinski <kuba@kernel.org>
>Cc: <redacted>
>Cc: <redacted>
>Cc: <redacted>
>Signed-off-by: Ram Muthiah <redacted>
>Signed-off-by: Lee Jones <redacted>
>---
> include/linux/virtio_vsock.h            | 4 +++-
> net/vmw_vsock/virtio_transport_common.c | 4 ++++
> 2 files changed, 7 insertions(+), 1 deletion(-)
>
>diff --git a/include/linux/virtio_vsock.h b/include/linux/virtio_vsock.h
>index 35d7eedb5e8e4..8c77d60a74d34 100644
>--- a/include/linux/virtio_vsock.h
>+++ b/include/linux/virtio_vsock.h
>@@ -7,9 +7,11 @@
> #include <net/sock.h>
> #include <net/af_vsock.h>
>
>+extern uint virtio_transport_max_vsock_pkt_buf_size;
>+
> #define VIRTIO_VSOCK_DEFAULT_RX_BUF_SIZE	(1024 * 4)
> #define VIRTIO_VSOCK_MAX_BUF_SIZE		0xFFFFFFFFUL
>-#define VIRTIO_VSOCK_MAX_PKT_BUF_SIZE		(1024 * 64)
>+#define VIRTIO_VSOCK_MAX_PKT_BUF_SIZE		virtio_transport_max_vsock_pkt_buf_size
>
> enum {
> 	VSOCK_VQ_RX     = 0, /* for host to guest data */
>diff --git a/net/vmw_vsock/virtio_transport_common.c b/net/vmw_vsock/virtio_transport_common.c
>index 169ba8b72a630..d0d913afec8b6 100644
>--- a/net/vmw_vsock/virtio_transport_common.c
>+++ b/net/vmw_vsock/virtio_transport_common.c
>@@ -26,6 +26,10 @@
> /* Threshold for detecting small packets to copy */
> #define GOOD_COPY_LEN  128
>
>+uint virtio_transport_max_vsock_pkt_buf_size = 1024 * 64;
>+module_param(virtio_transport_max_vsock_pkt_buf_size, uint, 0444);
>+EXPORT_SYMBOL_GPL(virtio_transport_max_vsock_pkt_buf_size);
>+

Maybe better to add an entry under sysfs similar to what Jiang proposed here:
https://lists.linuxfoundation.org/pipermail/virtualization/2021-June/054769.html

Thanks,
Stefano

```

## Re: \[PATCH 1/1\] virtio/vsock: Make vsock virtio packet buff size configurable

- From: Carlos Llamas \<cmllamas@google.com\>
- Date: 2022-12-09 19:48:14 
- Message: https://ratatoskr.run/netdev/2022/12/2036188.md

```text
On Thu, Jul 22, 2021 at 02:55:19PM +0200, Stefano Garzarella wrote:
> > 
> > +uint virtio_transport_max_vsock_pkt_buf_size = 1024 * 64;
> > +module_param(virtio_transport_max_vsock_pkt_buf_size, uint, 0444);
> > +EXPORT_SYMBOL_GPL(virtio_transport_max_vsock_pkt_buf_size);
> > +

I'm interested on this functionality, so I could take this on.

> 
> Maybe better to add an entry under sysfs similar to what Jiang proposed
> here:
> https://lists.linuxfoundation.org/pipermail/virtualization/2021-June/054769.html

Having a look at Jiang's RFC patch it seems the proposed sysfs node
hangs off from the main kernel object e.g. /sys/kernel. So I wonder if
there is a more appropriate parent for this knob?

Also, I noticed that Ram's patch here is using read-only permissions for
the module parameter and switching to sysfs would mean opening this knob
up to be dynamically configured? I'd need to be careful here.

--
Carlos Llamas
```

## Re: \[PATCH 1/1\] virtio/vsock: Make vsock virtio packet buff size configurable

- From: Stefano Garzarella \<sgarzare@redhat.com\>
- Date: 2022-12-12 13:11:49 
- Message: https://ratatoskr.run/netdev/2022/12/2037329.md

```text
On Fri, Dec 09, 2022 at 07:48:02PM +0000, Carlos Llamas wrote:
>On Thu, Jul 22, 2021 at 02:55:19PM +0200, Stefano Garzarella wrote:
>> >
>> > +uint virtio_transport_max_vsock_pkt_buf_size = 1024 * 64;
>> > +module_param(virtio_transport_max_vsock_pkt_buf_size, uint, 0444);
>> > +EXPORT_SYMBOL_GPL(virtio_transport_max_vsock_pkt_buf_size);
>> > +
>
>I'm interested on this functionality, so I could take this on.

Great!
We are changing the packet handling using sk_buff [1], so I think it's better to rebase on that work that should be merged in net-next after the current merge window will close.

>
>>
>> Maybe better to add an entry under sysfs similar to what Jiang proposed
>> here:
>> https://lists.linuxfoundation.org/pipermail/virtualization/2021-June/054769.html
>
>Having a look at Jiang's RFC patch it seems the proposed sysfs node
>hangs off from the main kernel object e.g. /sys/kernel. So I wonder if
>there is a more appropriate parent for this knob?

Agree, what about /sys/devices ?
I would take a closer look at what is recommend in this case.

>
>Also, I noticed that Ram's patch here is using read-only permissions for
>the module parameter and switching to sysfs would mean opening this knob
>up to be dynamically configured? I'd need to be careful here.
>

True, but even if it's changed while we're running, I don't think it's a big problem.

Maybe the problem here would be the allocation of RX buffers made during the probe. Could this be a good reason to use a module parameter?

Thanks,
Stefano

[1] https://lore.kernel.org/lkml/20221202173520.10428-1-bobby.eshleman@bytedance.com/

```
