Linux Virtualization - Jul 2019 - [PATCH] vhost: disable metadata prefetch optimization

This seems to cause guest and host memory corruption.
Disable for now until we get a better handle on that.

Signed-off-by: Michael S. Tsirkin <mst at redhat.com>
---

I put this in linux-next, we'll re-enable if we can fix
the outstanding issues in a short order.

 drivers/vhost/vhost.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/vhost/vhost.h b/drivers/vhost/vhost.h
index 819296332913..42a8c2a13ab1 100644
--- a/drivers/vhost/vhost.h
+++ b/drivers/vhost/vhost.h
@@ -96,7 +96,7 @@ struct vhost_uaddr {
 };
 
 #if defined(CONFIG_MMU_NOTIFIER) && ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
== 0
-#define VHOST_ARCH_CAN_ACCEL_UACCESS 1
+#define VHOST_ARCH_CAN_ACCEL_UACCESS 0
 #else
 #define VHOST_ARCH_CAN_ACCEL_UACCESS 0
 #endif
-- 
MST

On 2019/7/26 ??7:51, Michael S. Tsirkin wrote:
> This seems to cause guest and host memory corruption.
> Disable for now until we get a better handle on that.
>
> Signed-off-by: Michael S. Tsirkin <mst at redhat.com>
> ---
>
> I put this in linux-next, we'll re-enable if we can fix
> the outstanding issues in a short order.

Btw, is this more suitable to e.g revert the 
842aa64eddacd23adc6ecdbc69cb2030bec47122 and let syzbot fuzz more on the 
current code?

I think we won't accept that patch eventually, so I suspect what syzbot 
reports today is a false positives.

Thanks

>
>   drivers/vhost/vhost.h | 2 +-
>   1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/drivers/vhost/vhost.h b/drivers/vhost/vhost.h
> index 819296332913..42a8c2a13ab1 100644
> --- a/drivers/vhost/vhost.h
> +++ b/drivers/vhost/vhost.h
> @@ -96,7 +96,7 @@ struct vhost_uaddr {
>   };
>   
>   #if defined(CONFIG_MMU_NOTIFIER) &&
ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 0
> -#define VHOST_ARCH_CAN_ACCEL_UACCESS 1
> +#define VHOST_ARCH_CAN_ACCEL_UACCESS 0
>   #else
>   #define VHOST_ARCH_CAN_ACCEL_UACCESS 0
>   #endif

On Fri, Jul 26, 2019 at 07:57:25PM +0800, Jason Wang
wrote:
> 
> On 2019/7/26 ??7:51, Michael S. Tsirkin wrote:
> > This seems to cause guest and host memory corruption.
> > Disable for now until we get a better handle on that.
> > 
> > Signed-off-by: Michael S. Tsirkin <mst at redhat.com>
> > ---
> > 
> > I put this in linux-next, we'll re-enable if we can fix
> > the outstanding issues in a short order.
> 
> 
> Btw, is this more suitable to e.g revert the
> 842aa64eddacd23adc6ecdbc69cb2030bec47122
Yes I did that too.
> and let syzbot fuzz more on the
> current code?
Current metadata direct access code is known to corrupt guest and host
memory - I don't feel we need more fuzzing.
> 
> I think we won't accept that patch eventually, so I suspect what syzbot
> reports today is a false positives.
Today's reports are real, it's a bug in my patch. But I reverted it -
the below is an easier way to make sure at least linux-next is stable
for everyone.
> 
> Thanks
> 
> 
> > 
> >   drivers/vhost/vhost.h | 2 +-
> >   1 file changed, 1 insertion(+), 1 deletion(-)
> > 
> > diff --git a/drivers/vhost/vhost.h b/drivers/vhost/vhost.h
> > index 819296332913..42a8c2a13ab1 100644
> > --- a/drivers/vhost/vhost.h
> > +++ b/drivers/vhost/vhost.h
> > @@ -96,7 +96,7 @@ struct vhost_uaddr {
> >   };
> >   #if defined(CONFIG_MMU_NOTIFIER) &&
ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 0
> > -#define VHOST_ARCH_CAN_ACCEL_UACCESS 1
> > +#define VHOST_ARCH_CAN_ACCEL_UACCESS 0
> >   #else
> >   #define VHOST_ARCH_CAN_ACCEL_UACCESS 0
> >   #endif