Re: [PATCH v2 1/1] vfio-ccw: allow non-prefetch ORBs

[Jared Rossi]

On 2020-05-14 11:20, Cornelia Huck wrote:
> On Tue, 12 May 2020 14:15:35 -0400
> Jared Rossi <address@hidden> wrote:
>
> > Remove the explicit prefetch check when using vfio-ccw devices.
> > This check does not trigger in practice as all Linux channel programs
> > are intended to use prefetch.
> >
> > It is no longer required to force the PFCH flag when using vfio-ccw
> > devices.
>
> That's not quite true: Only kernels that include the currently-queued
> patch do not require it. Maybe
>
> "Newer Linux kernel versions do not require to force the PFCH flag with
> vfio-ccw devices anymore."
>
> ?

This is a good point and your proposed message is reasonable.

> > Signed-off-by: Jared Rossi <address@hidden>
> > ---
> >  hw/vfio/ccw.c | 13 +++----------
> >  1 file changed, 3 insertions(+), 10 deletions(-)
> >
> > diff --git a/hw/vfio/ccw.c b/hw/vfio/ccw.c
> > index 50cc2ec75c..e649377b68 100644
> > --- a/hw/vfio/ccw.c
> > +++ b/hw/vfio/ccw.c
> > @@ -74,16 +74,9 @@ static IOInstEnding 
> > vfio_ccw_handle_request(SubchDev *sch)
> >      struct ccw_io_region *region = vcdev->io_region;
> >      int ret;
> >
> > -    if (!(sch->orb.ctrl0 & ORB_CTRL0_MASK_PFCH)) {
> > -        if (!(vcdev->force_orb_pfch)) {
> > -            warn_once_pfch(vcdev, sch, "requires PFCH flag set");
> > -            sch_gen_unit_exception(sch);
> > -            css_inject_io_interrupt(sch);
> > -            return IOINST_CC_EXPECTED;
> > -        } else {
> > -            sch->orb.ctrl0 |= ORB_CTRL0_MASK_PFCH;
> > -            warn_once_pfch(vcdev, sch, "PFCH flag forced");
> > -        }
> > +    if (!(sch->orb.ctrl0 & ORB_CTRL0_MASK_PFCH) && 
> > vcdev->force_orb_pfch) {
> > +        sch->orb.ctrl0 |= ORB_CTRL0_MASK_PFCH;
> > +        warn_once_pfch(vcdev, sch, "PFCH flag forced");
> >      }
> >
> >      QEMU_BUILD_BUG_ON(sizeof(region->orb_area) != sizeof(ORB));
>
> Let me spell out what happens:
> - PFCH bit set -> no change
> - PFCH bit not set, but force_orb_pfch set -> no change
> - neither PFCH bit nor force_orb_pfch set:
>   - older kernels: QEMU makes the request, the kernel rejects it, guest
>     gets a unit exception (same result for the guest as before, only a
>     different code flow)
>   - newer kernels: QEMU makes the request, the kernel forwards the
>     request (logging a rate-limited warning); the result depends on
>     whether the guest actually tries to rewrite the channel program or
>     not

This is correct, but I think it is worth noting that while the exception
is the same in the case of new QEMU + old kernel, the logging is 
different.
The old kernel code did not issue any warning if a non-prefetch ORB was
rejected, it simply raised the exception. In reality, the old kernel 
code
path was not accessible because QEMU would always reject ORBs before 
then
with the "requires PFCH flag set" message.  The new QEMU code does not
issue a warning in this case.

I considered keeping a warning for the non-prefetch path, but it seemed
excessive to me, since it causes a redundant warning when used with the
new kernel code (which I expect to be the case normally). Do you think
some sort of warning should still be issued by QEMU in this case, even
if it is redundant with the kernel warning?

> I think that is what we want, and I think I'll queue this patch with
> the tweaked commit message, but I'd like a second opinion.
>
> (We should also deprecate force_orb_pfch in the future.)