[Pm-utils] Skipping pm-utils chvt on suspend and resume under kernel modesetting

[Dan Nicholson]

On Fri, Mar 13, 2009 at 7:14 AM, Kristian Høgsberg <krh at redhat.com> wrote:
> On Thu, 2009-03-12 at 19:22 -0700, Dan Nicholson wrote:
>> On Thu, Mar 12, 2009 at 4:41 PM, Victor Lowther
>> <victor.lowther at gmail.com> wrote:
>> > On Mar 12, 2009, at 4:58 PM, Kristian Høgsberg <krh at redhat.com> wrote:
>> >
>> >> Hi Victor,
>> >>
>> >> I'm attaching a patch that I've just added to pm-utils in Fedora
>> >> Rawhide.  With a kernel modesetting capable drm driver and DRI2, we no
>> >> longer need to chvt away from X to suspend and resume correctly.
>> >> We'll
>> >> keep it in rawhide and see how it goes, but I've tested it on a
>> >> bunch of
>> >> intel machines and it works.  Dave Airlie is testing this with radeon
>> >> modesetting and it should work there too.
>> >
>> > No problem - I have been waiting for KMS to hit mainline before
>> > patching pm-utils myself.
>>
>> Speaking of which, right now smart_kernel_intel() doesn't disable the
>> S3 hooks. I think with KMS, we shouldn't need to run any video hooks.
>> So, perhaps something like this should be added:
>>
>> local kms=`cat /sys/module/i915/parameters/modeset 2>/dev/null`
>> [ "$kms" = 1 ] && remove_parameters --quirk-s3-mode --quirk-s3-bios
>
> I'm using
>
> kms_active()
> {
>        test -d /sys/class/drm/card0/device/graphics/fb0
> }
>
> to test for modesetting and I think it's a little better for a couple of
> reasons: you could pass modeset=1 to the module and still not get
> modesetting, whereas testing for the fb0 directory there only returns
> true if modesetting actually kicked in.  Second, it's chipset
> independent.
>
> Not sure what the s3 hooks do, but it sounds plausible that we won't
> need them with a KMS driver.

They set /proc/sys/kernel/acpi_video_flags. They get set to 0 by
pm-utils, then added to if you have --quirk-s3-mode and/or
--quirk-s3-bios. I don't know what that actually does in the kernel.

> Here's an update version of the patch that moves the kms detection in
> pm/functions.in and also disables the s3 hooks.

Nice, thanks. Since we have an driver-independent way to check for
kms, probably there should be a new function in 98-smart-kernel-video:

smart_kernel_kms()
{
    kms_active || return 1
    remove_parameters ...
}

Then any driver that gets kms support will just get the video quirks
removed. That would probably want to get called first in
smart_kernel_video(), or at least before smart_kernel_intel().

--
Dan