Re: [PATCH] x86_64: Display HPET timer option

[Erwin Rol <mailinglists@erwinrol.com>]

Hey Andi,

I looked a bit better at what happens (by adding a bunch of printk's in
drivers/char/hpet.c) and it seems that read_counter(&hpet->hpet_mc);
always returns -1 and that causes the following loop in 
static unsigned long hpet_calibrate(struct hpets *hpetp)
to endlessly loop while having local_irq_save(flags).

do {
	m = read_counter(&hpet->hpet_mc);
        write_counter(t + m + hpetp->hp_delta, &timer->hpet_compare);
} while (i++, (m - start) < count);

count is 0xe9 and start and (m - start) is 0. 

What could be the reason that read_counter(&hpet->hpet_mc) is and stays
-1 ?


- Erwin 

On Thu, 2005-12-01 at 22:18 +0100, Andi Kleen wrote: 
> On Thu, Dec 01, 2005 at 10:06:37PM +0100, Erwin Rol wrote:
> > On Thu, 2005-12-01 at 21:43 +0100, Andi Kleen wrote:
> > > On Thu, Dec 01, 2005 at 12:30:03PM -0800, Zwane Mwaikambo wrote:
> > > > On Thu, 1 Dec 2005, Linus Torvalds wrote:
> > > > 
> > > > > On Thu, 1 Dec 2005, Zwane Mwaikambo wrote:
> > > > > >
> > > > > > Currently the HPET timer option isn't visible in menuconfig.
> > > > > 
> > > > > Do you want it to?
> > > > > 
> > > > > Why would you ever compile it out?
> > > > 
> > > > For timer testing purposes i sometimes would like not to use the HPET. 
> > > > Would a runtime switch be preferred?
> > > 
> > > nohpet already exists.
> > > 
> > 
> > And luckily it does cause without "nohpet" i can't boot my shuttle
> > ST20G5, the NMI watchdog kills it because ti hangs when initializing the
> > hpet. If the nmi watchdog is off it just hangs for ever. 
> 
> Can you give details on the machine? lspci, dmidecode, acpidmp output,
> boot log from the hang case?
> 
> Then perhaps it can be blacklisted or the failure otherwise avoided.
> 
> [Looks like I finally need to add DMI decode support to x86-64 too :-/]
> 
> -Andi
> 
>