Re: "double" hpet clocksource && hard freeze [bisected]

[Paolo Ornati <ornati@fastwebnet.it>]

On Tue, 28 Aug 2007 10:27:09 +0200
Clemens Ladisch <clemens@ladisch.de> wrote:

> Luck, Tony wrote:
> > [...] Given that the hang went away when you applied the earlier patch, I
> > conclude that the drivers/char/hpet.c code is the one that got selected when
> > you had two "hpet" entries ... and that there is something wrong with that
> > code that doesn't work right on x86_64.  
> 
> Apparently, the 'generic' code was just copied from ia64 and assumes that the
> timer is 64 bits.  This is not true with hardware from VIA (even on x86_64).

The hardware of this PC is Intel, not VIA.

> 
> This patch should make it work (although I'd prefer to set the mask  
> dynamically
> according to the hardware caps).
> 
> --- a/drivers/char/hpet.c	Tue Aug 28 09:42:22 2007
> +++ b/drivers/char/hpet.c	Tue Aug 28 10:16:54 2007
> @@ -73,7 +73,7 @@
>           .name           = "hpet",
>           .rating         = 250,
>           .read           = read_hpet,
> -        .mask           = CLOCKSOURCE_MASK(64),
> +        .mask           = CLOCKSOURCE_MASK(32),


Anyway, I've applied it (manually... whitespace/mime damage) to -rc4 and
it seems to work, no crash so far (I'm sure I'm testing it because
plain -rc4 doesn't have the "2hpet" fix and still manifest the problem).

-- 
	Paolo Ornati
	Linux 2.6.23-rc4-dirty on x86_64