Re: [Xen-devel] [PATCH 1/2] x86/VPMU: Disable VPMU when NMI watchdog is on

[Boris Ostrovsky <boris.ostrovsky@xxxxxxxxxx>]

On 01/28/2015 05:41 PM, Andrew Cooper wrote:
> On 28/01/2015 22:33, Boris Ostrovsky wrote:
> > On 01/28/2015 04:49 PM, Andrew Cooper wrote:
> > > On 28/01/2015 19:56, Boris Ostrovsky wrote:
> > > > NMI watchdog sets APIC_LVTPC register to generate an NMI when PMU
> > > > counter
> > > > overflow occurs. This may be overwritten by VPMU code later,
> > > > effectively
> > > > turning off the watchdog.
> > > >
> > > > We should disable VPMU when NMI watchdog is running.
> > > >
> > > > Signed-off-by: Boris Ostrovsky <boris.ostrovsky@xxxxxxxxxx>
> > > I completely agree with the aim, but this patch is clunky and you have
> > > missed the case where neither 'watchdog' nor 'vpmu' is specified on the
> > > command line, but the booleans have been tweaked (which is the XenServer
> > > way of choosing defaults while keeping the length of the command line
> > > down).
> > You mean binary patching? Or does XenServer change those flags before
> > compilation? Yes, I haven't thought about this.
> We patch the source before compiling.
>
> https://github.com/xenserver/xen-4.5.pg/blob/master/master/xen-tweak-cmdline-defaults.patch

How does this enable NMI watchdog? This patch sets watchdog_force but 
that alone doesn't enable anything. Or are there other changes that set 
opt_watchdog (which I was going to test as you suggested below).
-boris

> > > A more simple approach, which doesn't involve exposing opt_vpmu_enabled
> > > or changing any nmi code, would be to have a check in vpmu_initialise()
> > > which checks for opt_watchdog and opt_vpmu_enabled and bail.
_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel