RE: x86-64 with nvidia MCP51 chipset: kernel does not find HPET

[Lee Revell <rlrevell@joe-job.com>]

On Mon, 2006-10-30 at 11:07 -0600, Langsdorf, Mark wrote:
> > I have a 6 month old x86-64 machine with nvidia MCP51 chipset on which
> > the kernel does not detect the HPET.  According to HPET maintainer
> > Clemens Ladisch, this machine certainly has one, but it cannot be
> > enabled for lack of hardware documentation.
> > 
> > Is there anything I can do to help debug this?
> 
> If the hardware is not providing the HPET description in ACPI,
> there's little you can do, and most vendors do not provide
> the HPET description.
> 
> Do you know if there's an entry for HPET in the ACPI?

I'm not exactly an ACPI expert, but I do not think there is an entry for
HPET in the ACPI, as the check in arch/x86_64/kernel/io_apic.c fails:

 358                         /*
 359                          * All timer overrides on Nvidia are
 360                          * wrong unless HPET is enabled.
 361                          */
 362                         nvidia_hpet_detected = 0;
 363                         acpi_table_parse(ACPI_HPET,
 364                                         nvidia_hpet_check);
 365                         if (nvidia_hpet_detected == 0) {
 366                                 acpi_skip_timer_override = 1;
 367                                 printk(KERN_INFO "Nvidia board "
 368                                     "detected. Ignoring ACPI "
 369                                     "timer override.\n");
 370                         }

But, with some help from anonymous sources, I have been able to find the
HPET and make it work using a userspace driver that pokes registers by
mmap'ing /dev/mem.  So we just need a way to tell the kernel it's there.
Presumably this would require a PCI quirk.

Is this likely to be worth the trouble?

Lee