Re: [PATCH v2] platform/x86: Add driver for Apple gmux device

On Thu, Mar 1, 2012 at 3:53 PM, Seth Forshee [off-list ref] wrote:

On Thu, Mar 01, 2012 at 10:19:58AM +0100, Corentin Chary wrote:

quoted

On Wed, Feb 29, 2012 at 11:56 PM, Seth Forshee
[off-list ref] wrote:

quoted

On Wed, Feb 29, 2012 at 04:32:28PM -0600, Grant Likely wrote:

quoted

On Wed, Feb 29, 2012 at 4:08 PM, Seth Forshee
[off-list ref] wrote:

quoted

On Wed, Feb 29, 2012 at 03:23:20PM -0600, Grant Likely wrote:

quoted

On Wed, Feb 29, 2012 at 1:50 PM, Seth Forshee
[off-list ref] wrote:

quoted

On Wed, Feb 29, 2012 at 12:43:23PM -0600, Grant Likely wrote:

quoted

On Wed, Feb 29, 2012 at 11:53 AM, Seth Forshee
[off-list ref] wrote:

quoted

On Wed, Feb 29, 2012 at 11:46:39AM -0600, Grant Likely wrote:

quoted

On Wed, Feb 22, 2012 at 8:37 AM, Seth Forshee
[off-list ref] wrote:

quoted

Apple laptops with hybrid graphics have a device named gmux that
controls the muxing of the LVDS panel between the GPUs as well as screen
brightness. This driver adds support for the gmux device. Only backlight
control is supported initially.

Signed-off-by: Seth Forshee <redacted>

Works for me.

Tested-by: Grant Likely <redacted>

Now I just need to figure out how to get the desktop backlight widget
to use gmux_backlight instead of acpi_video0...

The easy way is to pass acpi_backlight=vendor to the kernel, then you
won't have acpi_vidoe0.

That did it, thanks.  I'm assume something is in the works to set it
up automatically?

Not that I'm aware of. A number machines have this problem, that the
standard ACPI backlight interfaces are implemented but don't work. This
generally isn't detectable in software; with the Apples at least
everything looks like it's working except that the brightness doesn't
change (but not all Apple laptops are affected, so qurking based on
manufacturer wouldn't work). All we're left with is DMI quirking, which
isn't practical. Maybe we could add something so a platform driver can
tell acpi_video that it knows the ACPI backlight doesn't work, but I
think on some platforms that still is going to be based off of DMI
information.

blacklisting based on specific product name (ie. MacBookPro8,*) or
machine model is probably the best.  It wouldn't be the first
blacklist in the linux kernel.

I think the blacklist would have to be against specific product names.
For example, the MacBook Pro 8,1 has a working acpi_video backlight and
no gmux_backlight, the 8,2 has both but only gmux_backlight works, and I
suspect the 8,3 is the same as the 8,2.

I have the 8,3, and my testing confirms that.

quoted

We'd probably end up with an
entry in the blacklist for every single model whose acpi_video backlight
doesn't work, adding entries for each new generation of MacBooks.

And if we start blacklisting Macs we'd have start doing it for other
machines too, I guess. From what I've seen, open-ended blacklists like
this get nacked pretty consistently nowadays.

An alternative would be to blacklist or disable acpi0_backlight when
the apple-gmux driver loads.  I don't know how acceptable that is, but
I also don't have much sympathy for nacking blacklists if there isn't
a viable alternative.

Yes, that's one idea I was thinking about. For all the machines I've
been able to get tested, if the gmux is present then it can control the
backlight and acpi_video cannot, so that approach is reasonable for
Macs.

There are quite a few machines in this situation though, and whatever
solution is arrived at should be flexible enough to work beyond just
Macs. I'll try to find some time soon to explore this further and see if
I can come up with something.

Old Samsung laptops have the same issue, and I ended up patching
drivers/acpi/video_detect.c (check "[PATCH] ACPI / Video: blacklist
some samsung laptops").
Doing it in the vendor module is complicated, since, it will be loaded
after the acpi video module most of the time, and that means addding
acpi_backlight, then removing it, which will probably confuse
usespace.
Patching drivers/acpi/video_detect.c seems safer.

I certainly agree that it's the easiest solution, so maybe it is worth
trying first and seeing what the response is. But last I saw the patch
you're referring to hadn't been merged or even commented on yet.

I'm still waiting for comments, but Matthew seems to be very busy
currently, so I'll probably repost it directly to linux-acpi in some
days.
Ccing Len and linux-acpi.

And I can't help but note that you included as justification for the patch
that the acpi_video backlight works on newer Samsungs so the list
wouldn't grow ;)

Beyond Samsung and Apple I know that there are also Toshibas that suffer
from this, and judging by the list from older versions of g-s-d [1] I'd
say there may well be others.

Seth

[1] http://git.gnome.org/browse/gnome-settings-daemon/tree/plugins/power/gsd-backlight-helper.c?id=GNOME_SETTINGS_DAEMON_3_2_2#n50

Anyway, like said before, there are already ton of blacklist in the
kernel, so why not ?

-- 
Corentin Chary
http://xf.iksaif.net