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

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: virtualization@lists.linux-foundation.org
Cc: kvm@vger.kernel.org
Cc: netdev@vger.kernel.org
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

On Wed, Jul 21, 2021 at 03:30:00PM +0100, Lee Jones wrote:

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

On Thu, Jul 22, 2021 at 02:55:19PM +0200, Stefano Garzarella wrote:

quoted

+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

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:

quoted

quoted

+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.

quoted

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/