Re: [Xen-devel] [PATCH] public/io/blkif.h: make the comments on "sectors" self-consistent

[Paul Durrant <Paul.Durrant@xxxxxxxxxx>]

> -----Original Message-----
> From: Anthony PERARD [mailto:anthony.perard@xxxxxxxxxx]
> Sent: 21 March 2019 12:22
> To: Paul Durrant <Paul.Durrant@xxxxxxxxxx>
> Cc: xen-devel@xxxxxxxxxxxxxxxxxxxx; Konrad Rzeszutek Wilk 
> <konrad.wilk@xxxxxxxxxx>
> Subject: Re: [Xen-devel] [PATCH] public/io/blkif.h: make the comments on 
> "sectors" self-consistent
> 
> On Wed, Mar 20, 2019 at 12:52:28PM +0000, Paul Durrant wrote:
> > Currently the comment at line #267 claims that the value should be
> > expressed in number logical sectors, whereas the comment at line #613
> > states that the value should be expressed strictly in units of 512 bytes.
> >
> > Signed-off-by: Paul Durrant <paul.durrant@xxxxxxxxxx>
> > ---
> > Cc: Konrad Rzeszutek Wilk <konrad.wilk@xxxxxxxxxx>
> >
> > Looking at xen-blkfront in Linux, I'm also not convinced that it would
> > function correctly is sector-size != 512 anyway so I wonder whether this
> > patch should go further and define that sector-size is strictly 512.
> > ---
> >  xen/include/public/io/blkif.h | 3 +--
> >  1 file changed, 1 insertion(+), 2 deletions(-)
> >
> > diff --git a/xen/include/public/io/blkif.h b/xen/include/public/io/blkif.h
> > index 15a71e3fea..d7c904d9dc 100644
> > --- a/xen/include/public/io/blkif.h
> > +++ b/xen/include/public/io/blkif.h
> > @@ -264,8 +264,7 @@
> >   * sectors
> >   *      Values:         <uint64_t>
> >   *
> > - *      The size of the backend device, expressed in units of its logical
> > - *      sector size ("sector-size").
> > + *      The size of the backend device, expressed in units of 512 bytes.
> >   *
> >   
> > *****************************************************************************
> >   *                            Frontend XenBus Nodes
> 
> But OVMF's frontend, minios' frontend, FreeBSD's frontend all do
> sector-size * sectors to figure out the size of the media.
> But looks like for at least OVMF, IO requests are handled with a sectors
> size of 512.
> 
> I think FreeBSD's backend also set "sectors" based on "sector-size", but
> on the other hand, "sector-size" is always set to 512.
> I think it the same for the old qemu (before Paul's refactoring).
> 
> I think I would be fine with the patch going further and have
> "sector-size" always 512, as some implementation are backed with
> this assumption (Linux, which I haven't checked).
> 
> (I don't want to have to patch OVMF because the protocol changed.)

The problem we face is backends pointing at disks that don't do 512 byte 
logical block emulation. There has to be some hope for dealing with them. If we 
say that the current state of affairs is that everything is broken (which I 
think it is as far as Linux blkfront/blkback are concerned) so we enforce 
sector-size == 512 in the protocol then the only hope I can see is for 
frontends to introduce a flag to say 'no emulation' to prompt the frontend to 
use physical-sector-size as logical sector-size... or something like that.

  Paul

> 
> --
> Anthony PERARD

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/xen-devel