potential problem with qdisk backend

potential problem with qdisk backend

[Juergen Gross]

Looking at the qdisk backend implementation I wondered whether
blkif_get_x86_32_req() is really correct, especially for the
BLKIF_OP_DISCARD case. The Linux kernel based blk backend seems to
distinguish 32- and 64-bit layouts of blkif_request_discard while
qemu treats them to be the same.

Am I completely wrong here?


Juergen

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel

Re: potential problem with qdisk backend

[Juergen Gross]

On 06/05/16 13:41, Juergen Gross wrote:
> Looking at the qdisk backend implementation I wondered whether
> blkif_get_x86_32_req() is really correct, especially for the
> BLKIF_OP_DISCARD case. The Linux kernel based blk backend seems to
> distinguish 32- and 64-bit layouts of blkif_request_discard while
> qemu treats them to be the same.
> 
> Am I completely wrong here?

Adding some maintainers.


Juergen


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel

Re: potential problem with qdisk backend

[Stefano Stabellini]

On Wed, 11 May 2016, Juergen Gross wrote:
> On 06/05/16 13:41, Juergen Gross wrote:
> > Looking at the qdisk backend implementation I wondered whether
> > blkif_get_x86_32_req() is really correct, especially for the
> > BLKIF_OP_DISCARD case. The Linux kernel based blk backend seems to
> > distinguish 32- and 64-bit layouts of blkif_request_discard while
> > qemu treats them to be the same.
> > 
> > Am I completely wrong here?
> 
> Adding some maintainers.
 
Yes, it seems to me that the internal alignment of the fields in struct
blkif_request_discard (as declared by xen/include/public/io/blkif.h)
differs between x86_32 and x86_64. The two different alignments are both
specified explicitly in Linux, by using __packed__ and explicit
padding.

On the other hand QEMU uses Xen's declaration of struct
blkif_request_discard without paying attention to the word size in the
guest. As a consequence it will work correctly only if the guest bitness
and the dom0 bitness match.

The discard struct should have two implementations in QEMU, like
blkif_x86_32_request and blkif_x86_64_request.

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel

Re: potential problem with qdisk backend

[Jan Beulich]

>>> On 06.05.16 at 13:41, <JGross@suse.com> wrote:
> Looking at the qdisk backend implementation I wondered whether
> blkif_get_x86_32_req() is really correct, especially for the
> BLKIF_OP_DISCARD case. The Linux kernel based blk backend seems to
> distinguish 32- and 64-bit layouts of blkif_request_discard while
> qemu treats them to be the same.

I agree, this can't work well.

Jan


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel