RE: [PATCH V4 1/2] perf ignore LBR and extra_regs.

[Liang, Kan]

> > >
> > > On Wed, Jul 09, 2014 at 02:32:28PM +0000, Liang, Kan wrote:
> > > >
> > > >
> > > > > On Tue, Jul 08, 2014 at 09:49:40AM -0700, kan.liang@xxxxxxxxx wrote:
> > > > > > +/*
> > > > > > + * Under certain circumstances, access certain MSR may cause #GP.
> > > > > > + * The function tests if the input MSR can be safely accessed.
> > > > > > + */
> > > > > > +static inline bool check_msr(unsigned long msr) {
> > > > > > +	u64 value;
> > > > > > +
> > > > > > +	if (rdmsrl_safe(msr, &value) < 0)
> > > > > > +		return false;
> > > > > > +	if (wrmsrl_safe(msr, value) < 0)
> > > > > > +		return false;
> > > > > > +	return true;
> > > > > > +}
> > > > >
> > > > > What does this thing return after patch 2? does the write still
> > > > > fault or will KVM silently take writes too?
> > > >
> > > > If applying patch 2, the function will return true. The KVM just
> > > > simply ignore
> > > the reads/writes.
> > >
> > > OK, then that's broken too. We want a function to return false for
> > > any malfunctioning MSR, ignoring writes and returning 0s is not
> > > proper functioning.
> >
> > The patch 2 is to handle the case that the administrator can only
> > patch the host. Don't need to force user to upgrade their guest to fix
> > the crash.  And ignore the annoying "unhandled...." KVM messages
> 
> Sure; but what I meant was, check_msr() is broken when ran on such a
> kernel. You need to fix check_msr() to return failure on these 'ignored'
> MSRs, after all they don't function as expected, they're effectively broken.

The function is designed to check if the MSRs can be safely accessed (no #GP). It cannot guarantee the correctness of the MSRs.
If KVM applied patch 2 and guest applied patch 1, from the guest's perspective, the MSRs can be accessed (no #GP triggered). So return true is expected. It should not be a broken.
The only unexpected thing for guest is that the counting/sampling result for LBR/extra reg is always 0. But the patch is a short term fix to stop things from crashing. I think it should be acceptable.


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/