Re: [PATCH] x86/PCI: Convert force_disable_hpet() to standard quirk

[Bjorn Helgaas <helgaas@kernel.org>]

On Wed, Nov 25, 2020 at 01:46:23PM +0100, Thomas Gleixner wrote:
> On Thu, Nov 19 2020 at 12:19, Bjorn Helgaas wrote:
> > 62187910b0fc ("x86/intel: Add quirk to disable HPET for the Baytrail
> > platform") implemented force_disable_hpet() as a special early quirk.
> > These run before the PCI core is initialized and depend on the
> > x86/pci/early.c accessors that use I/O ports 0xcf8 and 0xcfc.
> >
> > But force_disable_hpet() doesn't need to be one of these special early
> > quirks.  It merely sets "boot_hpet_disable", which is tested by
> > is_hpet_capable(), which is only used by hpet_enable() and hpet_disable().
> > hpet_enable() is an fs_initcall(), so it runs after the PCI core is
> > initialized.
> 
> hpet_enable() is not an fs_initcall(). hpet_late_init() is and that
> invokes hpet_enable() only for the case that ACPI did not advertise it
> and the force_hpet quirk provided a base address.
> 
> But hpet_enable() is also invoked via:
> 
>  start_kernel()
>    late_time_init()
>      x86_late_time_init()
>        hpet_time_init()
> 
> which is way before the PCI core is available and we really don't want
> to set it up there if it's known to be broken :)

Wow, I really blew that, don't know how I missed that path.  Thanks
for catching this!  I'll drop this patch.

> Now the more interesting question is why this needs to be a PCI quirk in
> the first place. Can't we just disable the HPET based on family/model
> quirks?

You mean like a CPUID check or something?  I'm all in favor of doing
something that doesn't depend on PCI.

> e0748539e3d5 ("x86/intel: Disable HPET on Intel Ice Lake platforms")
> f8edbde885bb ("x86/intel: Disable HPET on Intel Coffee Lake H platforms")
> fc5db58539b4 ("x86/quirks: Disable HPET on Intel Coffe Lake platforms")
> 62187910b0fc ("x86/intel: Add quirk to disable HPET for the Baytrail platform")
> 
> I might be missing something here, but in general on anything modern
> HPET is mostly useless.
> 
> Thanks,
> 
>         tglx
>