Re: [PATCH -v2] x86, tsc: Limit CPU frequency calibration on AMD

[Andreas Herrmann <andreas.herrmann3@amd.com>]

On Wed, Aug 25, 2010 at 03:04:54PM +0200, Andreas Herrmann wrote:
> On Wed, Aug 25, 2010 at 03:06:53AM -0400, Borislav Petkov wrote:
> > From: "H. Peter Anvin" <hpa@zytor.com>
> > Date: Tue, Aug 24, 2010 at 06:33:07PM -0400
> > 
> > > Build failure:
> > > 
> > > /home/hpa/kernel/linux-2.6-tip.cpu/arch/x86/kernel/cpu/amd.c: In
> > > function ‘amd_calibrate_cpu’:
> > > /home/hpa/kernel/linux-2.6-tip.cpu/arch/x86/kernel/cpu/amd.c:397: error:
> > > implicit declaration of function ‘avail_to_resrv_perfctr_nmi_bit’
> > > /home/hpa/kernel/linux-2.6-tip.cpu/arch/x86/kernel/cpu/amd.c:409: error:
> > > implicit declaration of function ‘reserve_perfctr_nmi’
> > > /home/hpa/kernel/linux-2.6-tip.cpu/arch/x86/kernel/cpu/amd.c:410: error:
> > > implicit declaration of function ‘reserve_evntsel_nmi’
> > > /home/hpa/kernel/linux-2.6-tip.cpu/arch/x86/kernel/cpu/amd.c:429: error:
> > > implicit declaration of function ‘release_perfctr_nmi’
> > > /home/hpa/kernel/linux-2.6-tip.cpu/arch/x86/kernel/cpu/amd.c:430: error:
> > > implicit declaration of function ‘release_evntsel_nmi’
> > > 
> > > Reproducible by doing "make ARCH=i386 allnoconfig".
> > 
> > Sh*t, I can't catch a break with that Kconfig dependency stuff, can I?
> > 
> > This happens because perfctr-watchdog.c gets pulled in by
> > CONFIG_X86_LOCAL_APIC which is, _of course_, not selected in an
> > allnoconfig build. Fixing this would mean exporting all that perfcounter
> > reservation functionality for the allnoconfig case, which is of course
> > doable but I'm starting to question the need for recalibrating the TSC
> > at all:
> > 
> > I mean, in the 99% of the cases MSRC001_0015[24] should be set by the
> > BIOS and if not then the BIOS which does that is pretty b0rked anyway.
> > So I'm thinking of removing the recalibration code and simply warning
> > the user instead, for the 1% case.
> > 
> > Andreas, what do you think?
> 
> I opt for removing the recalibration code plus keeping a FIRMWARE_WARN
> for borked BIOSes (just in case that there are any old systems with
> the wrong setting).

... and checking the HWCR MSR and issuing the firmware warning should
only happen if not running on a hypervisor. (Validate whether CPU has
X86_FEATURE_HYPERVISOR bit set or not.)


Andreas

-- 
Operating | Advanced Micro Devices GmbH
  System  | Einsteinring 24, 85609 Dornach b. München, Germany
 Research | Geschäftsführer: Alberto Bozzo, Andrew Bowd
  Center  | Sitz: Dornach, Gemeinde Aschheim, Landkreis München
  (OSRC)  | Registergericht München, HRB Nr. 43632