Bug#689268: Intel HD 4000 (Ivy Bridge) graphics freeze

[Per Foreby <per@foreby.se>]

On 2012-10-20 00:39, Per Foreby wrote:

> I noticed something strange with allocation of GPU RAM. In the old BIOS,
> the default was 64 MB, but in the new bios, "Auto" was default. So I set
> it explicitly to 64 MB. However, this is what the OS reports:
>
> # lspci -vv
> ...
> 00:02.0 VGA compatible controller: Intel Corporation Xeon E3-1200 v2/3rd Gen
> Core processor Graphics Controller (rev 09) (prog-if 00 [VGA controller])
> ...
>           Region 0: Memory at f7800000 (64-bit, non-prefetchable) [size=4M]
>           Region 2: Memory at e0000000 (64-bit, prefetchable) [size=256M]
> ...
>
> # dmesg | grep aperture
> [    0.000000] Checking aperture...
> [    1.644376] agpgart-intel 0000:00:00.0: AGP aperture is 256M @ 0xe0000000
>
>
> # grep -i mem /var/log/Xorg.0.log
> [    18.557] (--) PCI:*(0:0:2:0) 8086:0162:1043:84ca rev 9, Mem @
> 0xf7800000/4194304, 0xe0000000/268435456, I/O @ 0x0000f000/64
>
> (268435456 is 256*1024*1024.)

I changed the "iGPU memory" setting to 64 MB once more, and rebooted 
with the 3.5.5 kernel to see how much memory was reported. No change 
from 3.2.0 - everything still says 256 MB.

Could this be the problem? As Ingo reported, his problem disappeared 
with 256 MB GPU memory, and for me it took much longer for the next 
crash to occur, so maybe it not entirely correct when using 256 MB, but 
enough to make the freeze very rare?

And for my success (for 11 days) with the 3.5.5 kernel, maybe memory 
allocation works differently so it never (or more seldom) tries to 
access GPU memory that doesn't exist. Haven't tried 3.5.5 with 64 MB though.


Back on 3.2.0 with 64 MB now to test this theory.

/Per