linux-kernel - Re: radeon blink during probe regression (was Re: Linux 2.6.35-rc2)

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:	Tue, 8 Jun 2010 13:09:41 +1000
From:	Dave Airlie <airlied@...il.com>
To:	Alex Deucher <alexdeucher@...il.com>
Cc:	Torsten Kaiser <just.for.lkml@...glemail.com>,
	Linus Torvalds <torvalds@...ux-foundation.org>,
	Linux Kernel Mailing List <linux-kernel@...r.kernel.org>
Subject: Re: radeon blink during probe regression (was Re: Linux 2.6.35-rc2)

On Tue, Jun 8, 2010 at 11:50 AM, Alex Deucher <alexdeucher@...il.com> wrote:
> On Mon, Jun 7, 2010 at 8:49 PM, Dave Airlie <airlied@...il.com> wrote:
>> (changing subject
>>
>> On Tue, Jun 8, 2010 at 3:28 AM, Torsten Kaiser
>> <just.for.lkml@...glemail.com> wrote:
>>> On Mon, Jun 7, 2010 at 8:52 AM, Dave Airlie <airlied@...il.com> wrote:
>>>> On Mon, Jun 7, 2010 at 4:32 PM, Alex Deucher <alexdeucher@...il.com> wrote:
>>>>> On Mon, Jun 7, 2010 at 2:22 AM, Dave Airlie <airlied@...il.com> wrote:
>>>>>> On Mon, Jun 7, 2010 at 4:09 PM, Torsten Kaiser
>>>>>> <just.for.lkml@...glemail.com> wrote:
>>>>>>> The switchoff intervall is ~10 seconds, not 1..2 as I guessed in the first mail.
>>>>>>> Any idea if there is something in the KMS code that triggers at this intervall?
>>>>>>
>>>>>> The new mode probing code happens every 10 seconds, though we
>>>>>> shouldn't be turning anything on or off with it.
>>>>>>
>>>>>> So I suspect the DAC detection table might be doing bad things on your
>>>>>> hardware, we have had a problem where the dac detect table on one DAC
>>>>>> would turn the other one off which clearly wasn't what we wanted to
>>>>>> see.
>>>>>> ,
>>>>>
>>>>> The x300 is a non-atom card, so it uses the legacy load detection
>>>>> routines which do mess with some of the crtc regs.
>>>>>
>>>>
>>>> Okay I see it now, I'm amazed that I was seeing this issue last week
>>>> and blaming it on something complete different and only seeing it on
>>>> one card.
>>>>
>>>> I expect its the same problem I'll try and track down a proper fix for
>>>> it tomorrow.
>>>
>>> It really seems to be this polling code.
>>> After the fix to drivers/char/vt.c the 2.6.35-rc2 kernel now works for
>>> me without crashing.
>>>
>>> I also changed #define DRM_OUTPUT_POLL_PERIOD (10*HZ) to
>>> #define DRM_OUTPUT_POLL_PERIOD (3*HZ) and now the flickering before X
>>> starts is at a three second interval. That should prove, that the
>>> polling from drm_crtc_helper.c is the cause.
>>>
>>> After X starts the LCD is stable, but the secondary CRT still flickers
>>> every 3 seconds.
>>>
>>> My hardware is an X300-PCIe card with a VGA-, a DVI-I- and a video output.
>>> I have a 1280x1024 LCD attached to the DVI-I and an old CRT to the
>>> VGA. The video output is unused.
>>> The CRT is normally switched off, but its still detected correctly.
>>>
>>> The kernel output from 2.6.34 and 2.6.35-rc2 is here:
>>> http://lkml.org/lkml/2010/6/6/126
>>>
>>> xrandr says:
>>> Screen 0: minimum 320 x 200, current 2080 x 1024, maximum 4096 x 4096
>>> DVI-0 connected 1280x1024+0+0 (normal left inverted right x axis y
>>> axis) 338mm x 270mm
>>>   1280x1024      60.0*+   75.0
>>>   1280x960       60.0
>>>   1152x864       75.0
>>>   1024x768       75.1     70.1     60.0
>>>   832x624        74.6
>>>   800x600        72.2     75.0     60.3     56.2
>>>   640x480        72.8     75.0     66.7     60.0
>>>   720x400        70.1
>>>   640x400        70.0
>>> VGA-0 connected 800x600+1280+150 (normal left inverted right x axis y
>>> axis) 280mm x 210mm
>>>   800x600        72.2*+   75.0     60.3     56.2
>>>   1280x1024      60.0
>>>   1024x768       75.1     70.1     60.0
>>>   640x480        72.8     75.0     60.0
>>>   720x400        87.8     70.1
>>> S-video disconnected (normal left inverted right x axis y axis)
>>>
>>> If you have something for me to try, just send it. :-)
>>>
>>> Torsten
>>>
>>
>> Okay here is a patch that seems to work on my test system, but I want
>> to give it a bit more playing around with before I sent it onwards.
>>
>> Alex, seem sane to you?
>
> Seems sane.
>

While discussing this with Alex we realised this code should never
have been reached so it showed a bug higher up.

Looks like we were defining i2c lines for tv-out which doesn't have
any, the attached patch makes sure the tv-out on legacy radeon hw
doesn't get an i2c line attach so doesn't trigger the polling on it.

Torsten can you test this patch instead of the previous one, though I
think I'll send both to Linus for safety sakes.

Dave.

Download attachment "0001-drm-radeon-reset-i2c-valid-to-avoid-incorrect-tv-out.patch" of type "application/mbox" (1496 bytes)