Re: [Qemu-devel] [PATCH v7 1/8] virtio: endian-ambivalent targets using legacy virtio

[Michael S. Tsirkin]

On Mon, Apr 14, 2014 at 02:38:14PM +0200, Alexander Graf wrote:
> 
> On 14.04.14 14:34, Michael S. Tsirkin wrote:
> >On Mon, Apr 14, 2014 at 02:31:06PM +0200, Alexander Graf wrote:
> >>On 14.04.14 14:28, Michael S. Tsirkin wrote:
> >>>On Mon, Apr 14, 2014 at 02:22:36PM +0200, Alexander Graf wrote:
> >>>>Could we try to poison any non-virtio, non-endian-specific memory
> >>>>accessors when this file is included? That way we don't fall into
> >>>>pitfalls where we end up running stl_p in a tiny corner case
> >>>>somewhere still.
> >>>>
> >>>>
> >>>>Alex
> >>>Not sure about all users but it looks like the only file including this
> >>>is virtio.c right?
> >>>I'm guessing that's safe there.
> >>any virtio device implementations, since they need to communicate
> >>with the guest :).
> >In that case how can we poison regular memory accesses?
> >Devices normally do more than virtio related things.
> 
> Device should never do anything based on the guest endianness
> (because they don't know), so stx_p and ldx_p should always be
> poisoned.
> 
> 
> Alex

Makes sense.