Cornelia Huck
2015-May-12 13:14 UTC
[Qemu-devel] [PATCH RFC 1/7] virtio: relax feature check
On Wed, 06 May 2015 14:07:37 +0200 Greg Kurz <gkurz at linux.vnet.ibm.com> wrote:> Unlike with add and clear, there is no valid reason to abort when checking > for a feature. It makes more sense to return false (i.e. the feature bit > isn't set). This is exactly what __virtio_has_feature() does if fbit >= 32. > > This allows to introduce code that is aware about new 64-bit features like > VIRTIO_F_VERSION_1, even if they are still not implemented. > > Signed-off-by: Greg Kurz <gkurz at linux.vnet.ibm.com> > --- > include/hw/virtio/virtio.h | 1 - > 1 file changed, 1 deletion(-) > > diff --git a/include/hw/virtio/virtio.h b/include/hw/virtio/virtio.h > index d95f8b6..6ef70f1 100644 > --- a/include/hw/virtio/virtio.h > +++ b/include/hw/virtio/virtio.h > @@ -233,7 +233,6 @@ static inline void virtio_clear_feature(uint32_t *features, unsigned int fbit) > > static inline bool __virtio_has_feature(uint32_t features, unsigned int fbit) > { > - assert(fbit < 32); > return !!(features & (1 << fbit)); > } > > >I must say I'm not very comfortable with knowingly passing out-of-rage values to this function. Can we perhaps apply at least the feature-bit-size extending patches prior to your patchset, if the remainder of the virtio-1 patchset still takes some time?
Michael S. Tsirkin
2015-May-12 13:34 UTC
[Qemu-devel] [PATCH RFC 1/7] virtio: relax feature check
On Tue, May 12, 2015 at 03:14:53PM +0200, Cornelia Huck wrote:> On Wed, 06 May 2015 14:07:37 +0200 > Greg Kurz <gkurz at linux.vnet.ibm.com> wrote: > > > Unlike with add and clear, there is no valid reason to abort when checking > > for a feature. It makes more sense to return false (i.e. the feature bit > > isn't set). This is exactly what __virtio_has_feature() does if fbit >= 32. > > > > This allows to introduce code that is aware about new 64-bit features like > > VIRTIO_F_VERSION_1, even if they are still not implemented. > > > > Signed-off-by: Greg Kurz <gkurz at linux.vnet.ibm.com> > > --- > > include/hw/virtio/virtio.h | 1 - > > 1 file changed, 1 deletion(-) > > > > diff --git a/include/hw/virtio/virtio.h b/include/hw/virtio/virtio.h > > index d95f8b6..6ef70f1 100644 > > --- a/include/hw/virtio/virtio.h > > +++ b/include/hw/virtio/virtio.h > > @@ -233,7 +233,6 @@ static inline void virtio_clear_feature(uint32_t *features, unsigned int fbit) > > > > static inline bool __virtio_has_feature(uint32_t features, unsigned int fbit) > > { > > - assert(fbit < 32); > > return !!(features & (1 << fbit)); > > } > > > > > > > > I must say I'm not very comfortable with knowingly passing out-of-rage > values to this function. > > Can we perhaps apply at least the feature-bit-size extending patches > prior to your patchset, if the remainder of the virtio-1 patchset still > takes some time?So the feature-bit-size extending patches currently don't support migration correctly, that's why they are not merged. What I think we need to do for this is move host_features out from transports into core virtio device. Then we can simply check host features >31 and skip migrating low guest features is none set. Thoughts? Any takers? -- MST
Cornelia Huck
2015-May-12 13:44 UTC
[Qemu-devel] [PATCH RFC 1/7] virtio: relax feature check
On Tue, 12 May 2015 15:34:47 +0200 "Michael S. Tsirkin" <mst at redhat.com> wrote:> On Tue, May 12, 2015 at 03:14:53PM +0200, Cornelia Huck wrote: > > On Wed, 06 May 2015 14:07:37 +0200 > > Greg Kurz <gkurz at linux.vnet.ibm.com> wrote: > > > > > Unlike with add and clear, there is no valid reason to abort when checking > > > for a feature. It makes more sense to return false (i.e. the feature bit > > > isn't set). This is exactly what __virtio_has_feature() does if fbit >= 32. > > > > > > This allows to introduce code that is aware about new 64-bit features like > > > VIRTIO_F_VERSION_1, even if they are still not implemented. > > > > > > Signed-off-by: Greg Kurz <gkurz at linux.vnet.ibm.com> > > > --- > > > include/hw/virtio/virtio.h | 1 - > > > 1 file changed, 1 deletion(-) > > > > > > diff --git a/include/hw/virtio/virtio.h b/include/hw/virtio/virtio.h > > > index d95f8b6..6ef70f1 100644 > > > --- a/include/hw/virtio/virtio.h > > > +++ b/include/hw/virtio/virtio.h > > > @@ -233,7 +233,6 @@ static inline void virtio_clear_feature(uint32_t *features, unsigned int fbit) > > > > > > static inline bool __virtio_has_feature(uint32_t features, unsigned int fbit) > > > { > > > - assert(fbit < 32); > > > return !!(features & (1 << fbit)); > > > } > > > > > > > > > > > > > I must say I'm not very comfortable with knowingly passing out-of-rage > > values to this function. > > > > Can we perhaps apply at least the feature-bit-size extending patches > > prior to your patchset, if the remainder of the virtio-1 patchset still > > takes some time? > > So the feature-bit-size extending patches currently don't support > migration correctly, that's why they are not merged. > > What I think we need to do for this is move host_features out > from transports into core virtio device. > > Then we can simply check host features >31 and skip > migrating low guest features is none set. > > Thoughts? Any takers? >After we move host_features, put them into an optional vmstate subsection? I think with the recent patchsets, most of the interesting stuff is already not handled by the transport anymore. There's only VIRTIO_F_NOTIFY_ON_EMPTY and VIRTIO_F_BAD_FEATURE left (set by pci and ccw).
Peter Maydell
2015-May-12 13:49 UTC
[Qemu-devel] [PATCH RFC 1/7] virtio: relax feature check
On 12 May 2015 at 14:14, Cornelia Huck <cornelia.huck at de.ibm.com> wrote:> On Wed, 06 May 2015 14:07:37 +0200 > Greg Kurz <gkurz at linux.vnet.ibm.com> wrote: >> @@ -233,7 +233,6 @@ static inline void virtio_clear_feature(uint32_t *features, unsigned int fbit) >> >> static inline bool __virtio_has_feature(uint32_t features, unsigned int fbit) >> { >> - assert(fbit < 32); >> return !!(features & (1 << fbit)); >> } >> >> >> > > I must say I'm not very comfortable with knowingly passing out-of-rage > values to this function.It would invoke C undefined behaviour, so clearly a bug if we did pass an out-of-range value here. You'd need to at least do if (fbit >= 32) { return false; } if you want to make it valid. -- PMM
On Tue, 12 May 2015 15:14:53 +0200 Cornelia Huck <cornelia.huck at de.ibm.com> wrote:> On Wed, 06 May 2015 14:07:37 +0200 > Greg Kurz <gkurz at linux.vnet.ibm.com> wrote: > > > Unlike with add and clear, there is no valid reason to abort when checking > > for a feature. It makes more sense to return false (i.e. the feature bit > > isn't set). This is exactly what __virtio_has_feature() does if fbit >= 32. > > > > This allows to introduce code that is aware about new 64-bit features like > > VIRTIO_F_VERSION_1, even if they are still not implemented. > > > > Signed-off-by: Greg Kurz <gkurz at linux.vnet.ibm.com> > > --- > > include/hw/virtio/virtio.h | 1 - > > 1 file changed, 1 deletion(-) > > > > diff --git a/include/hw/virtio/virtio.h b/include/hw/virtio/virtio.h > > index d95f8b6..6ef70f1 100644 > > --- a/include/hw/virtio/virtio.h > > +++ b/include/hw/virtio/virtio.h > > @@ -233,7 +233,6 @@ static inline void virtio_clear_feature(uint32_t *features, unsigned int fbit) > > > > static inline bool __virtio_has_feature(uint32_t features, unsigned int fbit) > > { > > - assert(fbit < 32); > > return !!(features & (1 << fbit)); > > } > > > > > > > > I must say I'm not very comfortable with knowingly passing out-of-rage > values to this function. >I take that as a valid reason then :)> Can we perhaps apply at least the feature-bit-size extending patches > prior to your patchset, if the remainder of the virtio-1 patchset still > takes some time?Hmm... if I remember well, it still lacks migration support. -- Greg
Seemingly Similar Threads
- [Qemu-devel] [PATCH RFC 1/7] virtio: relax feature check
- [Qemu-devel] [PATCH RFC 1/7] virtio: relax feature check
- [Qemu-devel] [PATCH RFC 1/7] virtio: relax feature check
- [Qemu-devel] [PATCH RFC 1/7] virtio: relax feature check
- [Qemu-devel] [PATCH RFC 1/7] virtio: relax feature check