Re: [PATCH] rpmsg: virtio: don't let virtio core to validate used length

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Hello Mickael, Jason,

On 11/23/21 7:15 AM, Michael S. Tsirkin wrote:
> On Mon, Nov 22, 2021 at 05:08:12PM +0100, Arnaud Pouliquen wrote:
>> For RX virtqueue, the used length is validated in all the three paths
>> (big, small and mergeable). For control vq, we never tries to use used
>> length. So this patch forbids the core to validate the used length.
> 
> Jason commented on this. This is copy paste from virtio net
> where the change was merely an optimization.
> 

Right, I did it too fast last night (European time) to share the regression as
soon as possible.
For that, I copied and pasted the first commit I found related to the problem.
Need to rework this.

>> Without patch the rpmsg client sample does not work.
> 
> Hmm that's not enough of a description. Could you please
> provide more detail? Does rpmsg device set used length to a
> value > dma read buffer size? what kind of error message
> do you get? what are the plans to fix the device?

Let's me explain the context.
I run the rpmsg client sample test to communicate with a remote processor
that runs a Zephyr FW designed to answer to the Linux kernel driver sample.

The Zephyr is relying on OpenAMP library to implement the RPMsg and VirtIO layers.

In TX direction (Linux to Zephyr) 8 buffers of 512 bytes are allocated.
The first 8 RPMsg sent are OK. But when virtio loop back the the TX buffer index
0 (so already used and free one time) the following error occurs in
virtqueue_get_buf_ctx_split[1]:
" virtio_rpmsg_bus virtio0: output:used len 28 is larger than in buflen 0"

I have investigated the problem further today. Here is my analysis

rpmsg_send_offchannel_raw
-> virtqueue_add_outbuf
   -> virtqueue_add
      -> virtqueue_add_split
	 Here we use the "out_sgs" (in_sgs == 0)
	 buflen is not incremented in loop [2]
	 We don't enter in loop [3] as "in_sgs == 0"
	 consequence is that vq->buflen[head] is set to 0 [4]

[1]
https://elixir.bootlin.com/linux/v5.16-rc2/source/drivers/virtio/virtio_ring.c#L799
[2]
https://elixir.bootlin.com/linux/v5.16-rc2/source/drivers/virtio/virtio_ring.c#L551
[3]
https://elixir.bootlin.com/linux/v5.16-rc2/source/drivers/virtio/virtio_ring.c#L567
[4]
https://elixir.bootlin.com/linux/v5.16-rc2/source/drivers/virtio/virtio_ring.c#L622


An alternative to fix the issue is to set buflen in loop 2, but I'm not enough
expert to ensure that this will not have any side effect...

@@ -559,10 +559,11 @@ static inline int virtqueue_add_split(struct virtqueue *_vq,
 			 * table since it use stream DMA mapping.
 			 */
 			i = virtqueue_add_desc_split(_vq, desc, i, addr, sg->length,
 						     VRING_DESC_F_NEXT,
 						     indirect);
+			buflen += sg->length;
 		}
 	}
 	for (; n < (out_sgs + in_sgs); n++) {

So can you tell me if you prefer me to send a V2 updating the commit message or
a new message to fix virtio_ring (or both)?

Thanks,
Arnaud

> 
>> Fixes: 939779f5152d ("virtio_ring: validate used buffer length")
>>
>> Signed-off-by: Arnaud Pouliquen <arnaud.pouliquen@xxxxxxxxxxx>
>> Cc: Jason Wang <jasowang@xxxxxxxxxx>
>> Cc: Michael S. Tsirkin <mst@xxxxxxxxxx>
>> ---
>> base-commit: fa55b7dcdc43c1aa1ba12bca9d2dd4318c2a0dbf
>> ---
>>  drivers/rpmsg/virtio_rpmsg_bus.c | 1 +
>>  1 file changed, 1 insertion(+)
>>
>> diff --git a/drivers/rpmsg/virtio_rpmsg_bus.c b/drivers/rpmsg/virtio_rpmsg_bus.c
>> index 9c112aa65040..5f73f19c2c38 100644
>> --- a/drivers/rpmsg/virtio_rpmsg_bus.c
>> +++ b/drivers/rpmsg/virtio_rpmsg_bus.c
>> @@ -1054,6 +1054,7 @@ static struct virtio_driver virtio_ipc_driver = {
>>  	.feature_table_size = ARRAY_SIZE(features),
>>  	.driver.name	= KBUILD_MODNAME,
>>  	.driver.owner	= THIS_MODULE,
>> +	.suppress_used_validation = true,
>>  	.id_table	= id_table,
>>  	.probe		= rpmsg_probe,
>>  	.remove		= rpmsg_remove,
>> -- 
>> 2.17.1
> 



[Index of Archives]     [Linux Sound]     [ALSA Users]     [ALSA Devel]     [Linux Audio Users]     [Linux Media]     [Kernel]     [Photo Sharing]     [Gimp]     [Yosemite News]     [Linux Media]

  Powered by Linux