Re: [PATCH v4 2/2] KVM: s390: use cookies for ioeventfd

[Gleb Natapov <gleb@redhat.com>]

On Thu, Jul 04, 2013 at 08:54:32AM +0200, Paolo Bonzini wrote:
> Il 03/07/2013 18:33, Cornelia Huck ha scritto:
> > On Wed, 03 Jul 2013 17:30:40 +0200
> > Paolo Bonzini <pbonzini@redhat.com> wrote:
> > 
> >> Il 03/07/2013 16:30, Cornelia Huck ha scritto:
> >>> +	/*
> >>> +	 * Return cookie in gpr 2, but don't overwrite the register if the
> >>> +	 * diagnose will be handled by userspace.
> >>> +	 */
> >>> +	if (ret != -EOPNOTSUPP)
> >>> +		vcpu->run->s.regs.gprs[2] = ret;
> >>
> >> I think this should now be "if (ret >= 0)".
> > 
> > Hm, we don't want to kill gpr 2's old contents if userspace will do
> > something, which means -EOPNOTSUPP.
> 
> In the end kvm_io_bus_write_cookie only returns -EOPNOTSUPP if there is
> an error, so it works.  But if this were to change, the code would
> break.  That's why I suggested testing "ret >= 0" rather than "ret !=
> -EOPNOTSUPP".  But in the end it is the same.
> 
> >>
> >>>  	/* kvm_io_bus_write returns -EOPNOTSUPP if it found no match. */
> >>
> >> The comment is now obsolete.
> > 
> > s/kvm_io_bus_write/kvm_io_bus_write_cookie/ ? Otherwise, this is still
> > true.
> 
> True but somewhat misplaced, it is basically saying the same thing as
> the "Return cookie in gpr 2" comment just above.
> 
> Anyhow, these are very small details.  I changed kvm_io_bus_write to
> kvm_io_bus_write_cookie in the comment and applied the patches to kvm-queue.
> 
1/2 broke x86. QEMU prints "Invalid read from memory region kvm-pic" and
guest hangs. Looks like IO is forwarded to userspace instead of in kernel device
for some reason. Un-applying for now.

--
			Gleb.