Re: [git pull] PCI changes

[Matthew Garrett <mjg59@srcf.ucam.org>]

On Fri, Jan 13, 2012 at 03:14:21PM -0800, Linus Torvalds wrote:
> On Fri, Jan 13, 2012 at 1:24 PM, Matthew Garrett <mjg59@srcf.ucam.org> wrote:
> >>
> >> So why is acpi_pci_root_add() special? Cna you explain that part to me?
> >
> > pcie_no_aspm() means "Do not permit ASPM to be enabled" - it doesn't
> > alter the existing state. pcie_clear_aspm() does that.
> 
> I *know*.
> 
> So look again.
> 
> Why is acpi_pci_root_add() special?
> 
> Because dammit, it does exactly that pcie_clear_aspm() for the
> ACPI_FADT_NO_ASPM case.

I'm sorry, I thought you were referring to the status quo rather than 
the patch - I see what you mean now. The intent here is to clear ASPM 
state if the FADT bit has been set *and* we've got _OSC control, whereas 
previously we'd clear ASPM state any time the FADT bit was set. There 
are some machines which (a) set the FADT bit, (b) hand over _OSC 
control, (c) program the device and (d) crash under certain 
circumstances if you then try to use it. My assumption here is that the 
FADT bit *does* mean "clear ASPM state", but only if we've been granted 
_OSC control. That's the behaviour described in the changelog.

-- 
Matthew Garrett | mjg59@srcf.ucam.org