[Beignet] Unrecoverable system lockup when allocating too much memory

[Rebecca N. Palmer]

On 10/11/15 02:37, Zou, Nanhai wrote:
> looks like something related to drm driver bo management.
>
> dose
> export bo_reuse=0
> help?
No; if anything, it makes it fill the memory _faster_.

(All of this is in 1.1.1; I haven't tried it in master, at least not 
recently.)

>
> Thanks
> Zou Nanhai
>
>
>> -----Original Message-----
>> From: Beignet [mailto:beignet-bounces at lists.freedesktop.org] On Behalf Of
>> Rebecca N. Palmer
>> Sent: Saturday, November 07, 2015 6:06 AM
>> To: beignet at lists.freedesktop.org
>> Subject: Re: [Beignet] Unrecoverable system lockup when allocating too much
>> memory
>>
>>> (The example in
>>> https://bugs.launchpad.net/ubuntu/+source/pyopencl/+bug/1354086 no
>>> longer hangs, so the "rapidly allocating and freeing pyopencl objects
>>> doesn't actually free the memory" aspect has evidently been fixed, but
>>> keeping too many objects for the available memory still does hang.
>>
>> On further investigation, it isn't that simple:
>>
>> -Create and keep (large_array_test.py): With swap, creates the objects in
>> swap (becoming slow but not actually hanging), then errors out on trying to do
>> arithmetic on them.  Without swap, errors out in object creation, but not
>> before the kernel's OOM killer terminates a few pieces of the desktop
>> (according to the log, it thinks my test is only using ~50MB, so it (like
>> gnome-system-monitor) evidently can't see GPU memory use).
>> -Rapid create-then-free (arraybug_test.py) using the result (naive
>> pyopencl.clmath): Doesn't hang, but may segfault on exit.  (Kernel log:
>> python3[8857]: segfault at 20 ip 00007fa5caac3c84 sp 00007ffca782b2a0 error
>> 4 in libpthread-2.21.so[7fa5caaba000+18000] ; I can't be more specific because
>> it doesn't crash under gdb).
>> -Rapid create-then-free throwing away the result (this would be a bug in a real
>> program, but probably shouldn't hang the entire system): With swap, starts
>> using it, hence becoming slow but not actually hanging (though I haven't taken
>> it to the point of actually filling the swap).
>> Without swap, hangs (at least screen output does: desktop clock frozen, mouse
>> pointer minimally responsive), with disk activity (!);
>> Alt+SysRq+s,u,b restarts, but usually doesn't write the log).