Re: CVS commit: src/sys/arch/x86/x86

[Jukka Ruohonen <jruohonen%iki.fi@localhost>]

On Mon, Aug 08, 2011 at 07:49:42AM -0400, Jared McNeill wrote:
> On Mon, 8 Aug 2011, Jukka Ruohonen wrote:
> >>Why?
> >
> >Because we (the operating system) know this better than the BIOS writer.
> >And because this flag is not reliable; numerous systems where tsc(9) is
> >"broken" miss this flag in my ACPI table collection, and vice versa.
> 
> This flag is new in ACPI v3.0. If the flag is missing, it does not change 
> behaviour from what the tsc driver did before. If the flag is present, the 
> tsc driver will take the safe approach and prefer the HPET or PM-Timer as 
> per ACPI spec. If you're really really really sure you know more about 
> the platform than the platform code, you can re-enable TSC timecounter at 
> runtime.

Yes, but when possible, we should always prefer actual, reliable, CPU
information instead of the BIOS. Note also that at least with Intel CPUs,
the TSC should be entirely safe with the current stock-NetBSD (i.e. no
automatic CPU power management), but now you've marked it unreliable for
many systems (cf. also [1]).

- Jukka.

[1] http://lkml.org/lkml/2005/11/4/173