Linux Virtualization - Jul 2013 - [PATCH] virtio-spec: add field for scsi command size

On Tue, Jul 02, 2013 at 03:34:09PM +0930, Rusty Russell
wrote:
> Paolo Bonzini <pbonzini at redhat.com> writes:
> > Il 01/07/2013 01:47, Rusty Russell ha scritto:
> >> > > 
> >> > > Mainly because I'm not sure that *all* devices are
now safe.  Are they?
> >> >
> >> > virtio-scsi's implementation in QEMU is not safe (been
delaying that for
> >> > too long, sorry), but the spec is safe.
> >> 
> >> Then if we added a transport feature, we couldn't use it :(
> >
> > Transport feature bits are still negotiated per device though.
> > virtio-scsi devices in QEMU would not negotiate that feature.
> 
> That's a good point; I tend to think of them as tied to the transport
> but there's nothing specifying that, nor any implementation requiring
> it.
> 
> OK, so VIRTIO_F_ANY_LAYOUT it is?
> 
> Cheers,
> Rusty.
> 
> Message Framing
>  
> The original intent of the specification was that message framing 
> (the particular layout of descriptors) be independent of the 
> contents of the buffers. For example, a network transmit buffer 
> consists of a 12 byte header followed by the network packet. This 
> could be most simply placed in the descriptor table as a 12 byte 
> output descriptor followed by a 1514 byte output descriptor, but 
> it could also consist of a single 1526 byte output descriptor in 
> the case where the header and packet are adjacent, or even three 
> or more descriptors (possibly with loss of efficiency in that 
> case).
> 
> Regrettably, initial driver implementations used simple layouts
> and devices came to rely on it, despite this specification 
> wording. It is thus recommended that drivers be conservative in 
> their assumptions, unless specific device features indicate that 
> general layout is permitted using VIRTIO_F_ANY_LAYOUT. In
> addition, some implementations may have large-but-reasonable
> restrictions on total descriptor size (such as based on IOV_MAX
> in the host OS). This has not been a problem in practice: little
> sympathy will be given to drivers which create unreasonably-sized
> descriptors such as dividing a network packet into 1500
> single-byte descriptors!
That's fine with me too.
So which bit are we using for this?
I'd like to rebase to latest bits and merge the optimization for 3.11.

-- 
MST

On Thu, Jul 04, 2013 at 10:49:42AM +0300, Michael S. Tsirkin
wrote:
> On Tue, Jul 02, 2013 at 03:34:09PM +0930, Rusty Russell wrote:
> > Paolo Bonzini <pbonzini at redhat.com> writes:
> > > Il 01/07/2013 01:47, Rusty Russell ha scritto:
> > >> > > 
> > >> > > Mainly because I'm not sure that *all* devices
are now safe.  Are they?
> > >> >
> > >> > virtio-scsi's implementation in QEMU is not safe
(been delaying that for
> > >> > too long, sorry), but the spec is safe.
> > >> 
> > >> Then if we added a transport feature, we couldn't use it
:(
> > >
> > > Transport feature bits are still negotiated per device though.
> > > virtio-scsi devices in QEMU would not negotiate that feature.
> > 
> > That's a good point; I tend to think of them as tied to the
transport
> > but there's nothing specifying that, nor any implementation
requiring
> > it.
> > 
> > OK, so VIRTIO_F_ANY_LAYOUT it is?
> > 
> > Cheers,
> > Rusty.
> > 
> > Message Framing
> >  
> > The original intent of the specification was that message framing 
> > (the particular layout of descriptors) be independent of the 
> > contents of the buffers. For example, a network transmit buffer 
> > consists of a 12 byte header followed by the network packet. This 
> > could be most simply placed in the descriptor table as a 12 byte 
> > output descriptor followed by a 1514 byte output descriptor, but 
> > it could also consist of a single 1526 byte output descriptor in 
> > the case where the header and packet are adjacent, or even three 
> > or more descriptors (possibly with loss of efficiency in that 
> > case).
> > 
> > Regrettably, initial driver implementations used simple layouts
> > and devices came to rely on it, despite this specification 
> > wording. It is thus recommended that drivers be conservative in 
> > their assumptions, unless specific device features indicate that 
> > general layout is permitted using VIRTIO_F_ANY_LAYOUT. In
> > addition, some implementations may have large-but-reasonable
> > restrictions on total descriptor size (such as based on IOV_MAX
> > in the host OS). This has not been a problem in practice: little
> > sympathy will be given to drivers which create unreasonably-sized
> > descriptors such as dividing a network packet into 1500
> > single-byte descriptors!
> 
> That's fine with me too.
> So which bit are we using for this?
> I'd like to rebase to latest bits and merge the optimization for 3.11.

Rusty, could you please tell me which feature bit do you
prefer for ANY_LAYOUT?
It would be sad to miss another release of both qemu and kernel with
this obvious optimization for the only reason we can't settle on a bit
to use to signal it.

> -- 
> MST

"Michael S. Tsirkin" <mst at redhat.com>
writes:
> On Thu, Jul 04, 2013 at 10:49:42AM +0300, Michael S. Tsirkin wrote:
>> > case).
>> > 
>> > Regrettably, initial driver implementations used simple layouts
>> > and devices came to rely on it, despite this specification 
>> > wording. It is thus recommended that drivers be conservative in 
>> > their assumptions, unless specific device features indicate that 
>> > general layout is permitted using VIRTIO_F_ANY_LAYOUT. In
>> > addition, some implementations may have large-but-reasonable
>> > restrictions on total descriptor size (such as based on IOV_MAX
>> > in the host OS). This has not been a problem in practice: little
>> > sympathy will be given to drivers which create unreasonably-sized
>> > descriptors such as dividing a network packet into 1500
>> > single-byte descriptors!
>> 
>> That's fine with me too.
>> So which bit are we using for this?
>> I'd like to rebase to latest bits and merge the optimization for
3.11.
>
>
> Rusty, could you please tell me which feature bit do you
> prefer for ANY_LAYOUT?
> It would be sad to miss another release of both qemu and kernel with
> this obvious optimization for the only reason we can't settle on a bit
> to use to signal it.
Let's use bit 30.  Here's the kernel patch:

virtio: VIRTIO_F_ANY_LAYOUT feature

Also known as the "no really, I read the spec" bit.

Signed-off-by: Rusty Russell <rusty at rustcorp.com.au>

diff --git a/include/uapi/linux/virtio_config.h
b/include/uapi/linux/virtio_config.h
index b7cda39..4b5da48 100644
--- a/include/uapi/linux/virtio_config.h
+++ b/include/uapi/linux/virtio_config.h
@@ -51,4 +51,7 @@
  * suppressed them? */
 #define VIRTIO_F_NOTIFY_ON_EMPTY	24
 
+/* Can the device handle any descriptor layout? */
+#define VIRTIO_F_ANY_LAYOUT		30
+
 #endif /* _UAPI_LINUX_VIRTIO_CONFIG_H */