[Qemu-devel] virtio: Question on VRING_DESC_F_WRITE flag

[Qemu-devel] virtio: Question on VRING_DESC_F_WRITE flag

[Subhashini Venkataraman]

[-- Attachment #1: Type: text/plain, Size: 516 bytes --]

Hi,

I have a question on VRING_DESC_F_WRITE flag.  As per the specifications, if this flag is set, then the corresponding buffer pointed by the descriptor is Write only, otherwise Read Only.  I am trying to understand why there is no READ_AND_WRITE option, i.e. a buffer could be Read by device emulation and written back into.

Is there any issue or concern because of which it was not considered?

If this is not the right mailing list to raise this question,  any pointers would help.

Thanks,
Subha


[-- Attachment #2: Type: text/html, Size: 2546 bytes --]

Re: [Qemu-devel] virtio: Question on VRING_DESC_F_WRITE flag

[Stefan Hajnoczi]

[-- Attachment #1: Type: text/plain, Size: 892 bytes --]

On Wed, May 13, 2015 at 07:16:13PM +0000, Subhashini Venkataraman wrote:
> I have a question on VRING_DESC_F_WRITE flag.  As per the specifications, if this flag is set, then the corresponding buffer pointed by the descriptor is Write only, otherwise Read Only.  I am trying to understand why there is no READ_AND_WRITE option, i.e. a buffer could be Read by device emulation and written back into.

All virtio devices are designed so that descriptors are read-only or
write-only.  A typical request looks like this:

 [header (RO)][data (RO)][status (WO)]

or

 [header (RO)][data (WO)][status (WO)]

Not sure what the original requirement for this design was, but the
read-only/write-only approach means that buffers could be sent over a
transport (like a pipe or TCP socket).  If a buffer was read-write then
it would be involve sending back a modified copy, which is unnecessary.

Stefan

[-- Attachment #2: Type: application/pgp-signature, Size: 473 bytes --]