Is my Radeon HD 6970M dying? Hangs & init problems

[Alex Deucher]

On Mon, Feb 9, 2015 at 6:06 PM, Rafał Miłecki <zajec5 at gmail.com> wrote:
> My notebook Samsung NP700G7A-S01PL was working stable for more than 2 years.
> I was using 3.11, 3.17, 3.18, 3.19 (since rc1) and many more successfully.
> First hang has happened on 2015-02-08 (23:30) with 3.19-rc5 I was
> using for 3 weeks.
>
> So what I'm seeing are two possibly related problems:
>
> 1) Random hangs
> I don't have to be doing anything unusual. A single display, no UVD,
> just writing some code in kate. And then it randomly happens. My
> screen goes all white or green vertical lines or blue vertical lines.
> I can't use/access my machine, sound goes into a loop (last second).
> Sometimes it happens after hours, sometimes 30 minutes, sometimes few
> minutes. So far I got 5-7 hangs like this.
>
> 2) Init problems
> Unfortunately rebooting does not always help. Even cold boot (removing
> power & battery, keeping power button pressed for few seconds) isn't
> helpful.
> a) First I get UVD init errors:
> *ERROR* UVD not responding, trying to reset the VCPU!!!
> b) Then machine hangs after displaying "pitch is 7680"
> I've tracked it to be somewhere near register_framebuffer
> (see attached bad.txt)
>
> As long as I don't use radeon (booting with "nomodeset") it works stable.
>
> I tested my RAM with MemTest86 (one pass, took 1 hour), no errors, CPU
> temperature didn't exceed 70 degrees.
>
> This evening as the last hope I installed fglrx. It hangs my machine
> as well with following messages:
> [   36.472526] console [netcon0] enabled
> [   36.473106] netconsole: network logging started
> [   48.192215] fglrx_pci 0000:01:00.0: irq 56 for MSI/MSI-X
> [   48.192726] <6>[fglrx] Firegl kernel thread PID: 1481
> [   48.192833] <6>[fglrx] Firegl kernel thread PID: 1482
> [   48.192954] <6>[fglrx] Firegl kernel thread PID: 1483
> [   48.193077] <6>[fglrx] IRQ 56 Enabled
> [   48.240118] <6>[fglrx] Reserved FB block: Shared offset:0, size:1000000
> [   48.240122] <6>[fglrx] Reserved FB block: Unshared offset:3fab4000, size:4000
> [   48.240124] <6>[fglrx] Reserved FB block: Unshared offset:3fab8000,
> size:548000
> [   48.240126] <6>[fglrx] Reserved FB block: Unshared offset:7fff3000, size:d000
> However if I drop fglrx.ko and just use Xorg driver fglrx_drv.so it
> works stable.
>
> Any ideas? Is GPU on my motherboard just dying? :|

If it just started at some point in time regardless of the sw stack
you are running, then I suspect I hw problem.  If you can run an older
known-good stack and it works fine, then it's probably a sw problem.

If you disable the fglrx kernel module, you are basically running a
display only driver (no accel, etc.).

Alex