Backlight doesn't turn on after resume

Binary package hint: xserver-xorg-video-intel

On my laptop (a ThinkPad T43), suspend to RAM works reliably except for one flaw: when the system comes back up, the display backlight is turned off. I can just barely see the gnome-screensaver unlock dialog, and at this point, I can switch to a text console (which turns the backlight back on) and then back to X, and begin using the system.

I'm unsure about whether this is a kernel or an X issue; please reassign appropriately.

Confirmed on my X61. There was an upstream commit that should have fixed this, and I once tried the git master version which worked fine. Maybe some commit is missing still.

Bug 134391 indicates this may be a kernel issue, and has some possible workarounds which I haven't tried yet.

Also confirmed on an R61 here.

According to freedesktop.org bug 13709 at http://bugs.freedesktop.org/show_bug.cgi?id=13709 the issue lies with the kernel.

Hi Guys,

From the upstream kernel bugzilla report: http://bugzilla.kernel.org/show_bug.cgi?id=9827

Len posted this comment "Please try the "s2ram" application, as it has a number of workarounds for this issue built into it.
If s2ram doesn't know about this laptop, you can try the workarounds manually. They are listed in: Documentation/power/video.txt"

Care to test if any of the workarounds work for you? Here is a link to Documentation/power/video.txt in case you need it:

http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=blob;f=Documentation/power/video.txt;h=2b358498d095b59d8ddf950b1c1ec109c2c6e7d8;hb=HEAD

Thanks.

Continually setting this bug to "incomplete" seems like a useless buck-passing maneuver to me. The fix is released upstream. It's in freedesktop.org git master in mesa/drm, and it's in Linux 2.6.25-rc1 and 2.6.25-rc2. Why not just import the fix instead of suggesting silly workarounds?

Hmm, an interesting read. I've tried these fixes already though.
Unfortunately none of them worked for me.

Was nice to learn a little more about them though.

Hi Jeffrey,

There are reasons for flipping the status - please refer to https://wiki.ubuntu.com/Bugs/Status . Flipping the status to Incomplete, at least to me, does not mean we're just setting your bug aside or "a useless buck-passing maneuver". It is quite the contrary - we regularly monitor Incomplete bugs hoping that we have received useful information to help get the bug resolved. I'm glad you mentioned this was resolved in upstream 2.6.25-rc1 and 2.6.25-rc2. Care to provide a reference to the patch or a git commit id as that would be much more useful and helpful to the kernel team. I'm sure they would strongly consider pulling the patch into the Hardy kernel to help fix this bug. Thanks.

Sorry, I didn't realize that "Incomplete" should now be read as "the reporter hasn't personally fixed the problem." It used to mean that the bug report was lacking in some detail needed to verify that the problem exists. That's obviously nonsense in the case where the bug is already verified and fixed upstream, so I must have been mistaken.

If I were on the Ubuntu xorg package team, I would be interested in the following commits from the mesa/drm master branch at git.freedesktop.org:

d63b57749f097b36df04c6beff9b35a1dd859523
"On many chipsets, the checks for DPLL enable or VGA mode will prevent the
pipeconf regs from being restored, which could result in a blank display or X
failing to come back after resume. So restore them unconditionally along with
actually restoring pipe B's palette correctly."

01f6afcfea9d315ad1473045da141bfc95bcb7e6
"In hibernate, we may end up calling the VGA save regs function twice, so we
need to make sure it's idempotent. That means leaving ARX in index mode after
the first save operation. Fixes hibernate on 965."

On Sat, Feb 16, 2008 at 04:29:16AM -0000, Jeffrey Baker wrote:
> Sorry, I didn't realize that "Incomplete" should now be read as "the
> reporter hasn't personally fixed the problem." It used to mean that the
> bug report was lacking in some detail needed to verify that the problem
> exists. That's obviously nonsense in the case where the bug is already
> verified and fixed upstream, so I must have been mistaken.

You wrote in the upstream bug that the problem was fixed for you with only
an update to mesa, which would seem to imply that this doesn't need fixing
in the kernel (despite your comment in
https://bugs.edge.launchpad.net/ubuntu/+source/linux/+bug/189260/comments/6).

That would have been a much more useful piece of information to post in this
bug, rather than bickering about bug status values.

> If I were on the Ubuntu xorg package team, I would be interested in the
> following commits from the mesa/drm master branch at
> git.freedesktop.org:

I will move this bug over to mesa so that the right folks see it now,
thanks.

--
 - mdz

Matt: the mesa/drm bits are in the kernel, so the commits should be applied there.

On Sat, Feb 16, 2008 at 09:37:18AM -0000, Timo Aaltonen wrote:
> Matt: the mesa/drm bits are in the kernel, so the commits should be
> applied there.

Oh, you're quite right, my apologies.

There should now be enough information in the bug for the kernel team to
address it.

--
 - mdz

Hi,

I have to say that for a few days now the "backlight off/blank screen" problem in Hardy doesn't appear after resuming from suspend (I think it was after the last kernel change). Now there is something different going on...

I've always had a problem where the "blank screen after x minutes" setting in the power management wizard would blank the screen but keep the backlight on. Now, it goes completely blank (with backlight off) but it doesn't come up when I move the mouse or hit a key. As it was the case when the backlight remained off after suspend, I just have to close and open the lid for the screen to come back up.

Toshiba Tecra M2 - daily updated Hardy - nvidia nForce FX

cheers

note that the drm commits that fix this are in 2.6.25-rc1.

I had a look at the proposed patches from freedesktop.org and the Hardy kernel. The problem is that those patches are against the post 2.6.24 code. The driver code, as well as a lot of other drm code has changed substantially after the 2.6.24 kernel, so this would mean a huge effort to pick everything that is required and at the same time make sure nothing else breaks. So I think this should wait for later.

Do you really want to ship Hardy in a state where no laptop with Intel graphics can suspend and resume? I can see your point about rebasing to post-2.6.24 code, but in this case why not roll back to a previous, known-working kernel?

I don't think rolling the kernel back would help anything. The fix for 2.6.25 changes suspend/resume support of the kernel that wasn't there before. But I could be wrong there.
However, if I read the initial issue correctly, suspend/resume works but the backlight is not turned on. If manually switching to text mode helps, might this be done automatically? The hibernate package seems to have an option (probably in conjunction with the vbetool) that could help:

SwitchToTextModeOnResume <boolean>
Use this to, on resume, switch to a text console and back.

On Tue, Feb 19, 2008 at 09:53:14PM -0000, Stefan Bader wrote:
> I don't think rolling the kernel back would help anything. The fix for 2.6.25 changes suspend/resume support of the kernel that wasn't there before. But I could be wrong there.
> However, if I read the initial issue correctly, suspend/resume works but the backlight is not turned on. If manually switching to text mode helps, might this be done automatically? The hibernate package seems to have an option (probably in conjunction with the vbetool) that could help:
>
> SwitchToTextModeOnResume <boolean>
> Use this to, on resume, switch to a text console and back.

We don't use the hibernate package in Ubuntu. acpi-support provides similar
functionality (and various other bits) and is installed by default in Ubuntu.

--
 - mdz

Switch to a text console and back does switch the backlight on. In gutsy it happened automatically, for the Thinkpad X61 at least, by using the fix in http://www.thinkwiki.org/wiki/Problem_with_display_remaining_black_after_resume:

  By setting the following in /etc/default/acpi-support the display comes back on X61s using Intel chipset:

    DOUBLE_CONSOLE_SWITCH=true

This does not work after upgrading to hardy.

Sigh, reopening. Maybe I was a bit vague last time, but it's not like we need to pull the whole set:

commit ba8bbcf6ff4650712f64c0ef61139c73898e2165
Author: Jesse Barnes <email address hidden>
Date: Thu Nov 22 14:14:14 2007 +1000

    i915: add suspend/resume support

    Add suspend/resume support to the i915 driver. Moves some of the
    initialization into the driver load routine, and fixes up places where we
    assumed no dev_private existed in some of the cleanup paths. This allows
    us to suspend/resume properly even if X isn't running.

    Signed-off-by: Dave Airlie <email address hidden>

although jbarnes says that there are others that are useful:

23:22 < tjaalton> jbarnes: was the commit "i915: add suspend/resume support the one that should fix backlight
                              not turning on after resume?
23:36 < jbarnes> tjaalton: yah that's the one
23:36 < jbarnes> tjaalton: there are a few subsequent fixes in the drm tree in the past week or so that also help

Subject: Fix vblank enable/disable callbacks
X-Git-Url: http://gitweb.freedesktop.org/?p=mesa/drm.git;a=commitdiff;h=79d69285202b55f269aa88a6bcda257257c9dee3

Subject: i915: save/restore interrupt state
X-Git-Url: http://gitweb.freedesktop.org/?p=mesa/drm.git;a=commitdiff;h=8b6c96dedd4ba5dfbfec6a7c831d566e31d28781

Subject: Fix saveGR array size
X-Git-Url: http://gitweb.freedesktop.org/?p=mesa/drm.git;a=commitdiff;h=6f19473191ae543fcc199d252c5865c0734d38ad

Subject: Restore pipeconf regs unconditionally
X-Git-Url: http://gitweb.freedesktop.org/?p=mesa/drm.git;a=commitdiff;h=d63b57749f097b36df04c6beff9b35a1dd859523

Subject: [915]: more registers for S3 (DSPCLK_GATE_D, CACHE_MODE_0, MI_ARB_STATE)
X-Git-Url: http://gitweb.freedesktop.org/?p=mesa/drm.git;a=commitdiff;h=5d8c754bc2c720d70bbdeca6b294660105717a62

those will get in 2.6.25-rc3.

Ok, I will take a look at those. Hopefully there isn't too much change outside the driver itself.

>
> > SwitchToTextModeOnResume <boolean>
> > Use this to, on resume, switch to a text console and back.
>
> We don't use the hibernate package in Ubuntu. acpi-support provides
> similar
> functionality (and various other bits) and is installed by default in
> Ubuntu.
>

I've tried this anyway. It doesn't work for some reason.

I pulled together most of the described patches. Only the one called "Fix vblank enable/disable callbacks" I left out since the dependencies are quite complicated. I placed a test kernel at:

http://people.ubuntu.com/~smb

Could someone please give that a try and report the results? Thanks.

Works a treat here.

Thanks very much for all your hard work.

Will this be added to the 'production' kernel?

Yes it will. This might be in alpha5 but I am not sure here. But for alpha6 it will be in.

This bug was fixed in the package linux - 2.6.24-10.16

---------------
linux (2.6.24-10.16) hardy; urgency=low

  [Alessio Igor Bogani]

  * rt: Synchronized with upstream (2.6.24.2-rt2)
  * rt: Updated configuration files

  [Eric Piel]

  * SAUCE: ACPI: Allow custom DSDT tables to be loaded from initramfs
    Amit Kucheria consolidated the DSDT patch with another fix that
    ifdefs symbols required when BLK_DEV_INITR is disabled.

  [Stefan Bader]

  * Add Optiarc DVD drive to audio quirks list.
    - LP: #186664
  * Update drm and i915 drm driver to fix suspend issues.
    - LP: #189260

  [Tim Gardner]

  * Fix FTBS without BLK_DEV_INITRD
    - LP: #193507
  * 64 bit CPA cache attribute bug
    - LP: #193736
  * Implemented default EDD control

  [Upstream Kernel Changes]

  * bonding: fix NULL pointer deref in startup processing
  * dlm: bind connections from known local address when using TCP
  * dlm: proper prototypes
  * dlm: don't print common non-errors
  * dlm: use dlm prefix on alloc and free functions
  * dlm: close othercons
  * dlm: align midcomms message buffer
  * dlm: swap bytes for rcom lock reply
  * dlm: use fixed errno values in messages
  * dlm: clear ast_type when removing from astqueue
  * dlm: recover locks waiting for overlap replies
  * dlm: another call to confirm_master in receive_request_reply
  * dlm: reject messages from non-members
  * dlm: validate messages before processing
  * dlm: reject normal unlock when lock is waiting for lookup
  * dlm: limit dir lookup loop
  * dlm: fix possible use-after-free
  * dlm: change error message to debug
  * dlm: keep cached master rsbs during recovery
  * dlm: Sanity check namelen before copying it
  * dlm: clean ups
  * dlm: static initialization improvements
  * dlm: use proper C for dlm/requestqueue stuff (and fix alignment bug)
  * dlm: dlm_process_incoming_buffer() fixes
  * dlm: do not byteswap rcom_lock
  * dlm: do not byteswap rcom_config
  * dlm: use proper type for ->ls_recover_buf
  * dlm: missing length check in check_config()
  * dlm: validate data in dlm_recover_directory()
  * dlm: verify that places expecting rcom_lock have packet long enough
  * dlm: receive_rcom_lock_args() overflow check
  * dlm: make find_rsb() fail gracefully when namelen is too large
  * dlm: fix overflows when copying from ->m_extra to lvb
  * dlm: fix dlm_dir_lookup() handling of too long names
  * dlm: dlm/user.c input validation fixes
  * dlm: proper types for asts and basts
  * dlm: eliminate astparam type casting
  * dlm: add __init and __exit marks to init and exit functions
  * virtio: Use PCI revision field to indicate virtio PCI ABI version

 -- Tim Gardner <email address hidden> Tue, 19 Feb 2008 09:57:18 -0700

Big thanks. It works now. But i found some new issues.
1. It seems that screen could not get blank for powersaving.
2. I get back problem with hd clicking. And it`s much more important than suspend to ram problem, so i think there is some where in kernel power-managment policy for hdd....

Sorry. It was my fault. All works now.

Hi,

on my Sony VAIO VGN-B1VP this is still an issue. Switching to VT does not help, the backlight stays off.