On Wed 17-08-16 19:49:59, Tetsuo Handa wrote:
Olaf Hering wrote:
On Wed, Aug 17, Jiri Slaby wrote:
On 08/17/2016, 08:04 AM, Olaf Hering wrote:
It happend now the third time that the OOM killer kicks in with current TW and kernel 4.7. This never happend with my selfcompiled 4.5.
Any idea if FireFox just leaks memory and deserves to die, or if the kernel is broken?
There is a kernel memory leak, I very suspect. ARM and ppc kernel builds fail due to that. But I was unable to catch that beast yet :/.
Ok. I will try 4.7.1 which mentions some ipc leak, and also 4.6.4. Attached is meminfo, slabinfo and yet another OOM killer after I restarted firefox. It happens to work with another profile, which I will use now to get something done.
I don't suspect a kernel memory leak. The logic for declaring OOM has changed in Linux 4.7, and that change is causing premature OOM killer invocations in several environments.
http://lkml.kernel.org/r/201608120901.41463.a.miskiewicz@gmail.com http://lkml.kernel.org/r/20160801061625.GA11623@dhcp22.suse.cz
The former thread is an OOM report using ext4 filesystem. Please report your problem to linux-mm@kvack.org .
Yes this seems to be the case. Let's see if we can help compaction to not back off so early and try harder for !costly high order requests. Vlastimil has already a series which is sitting in the current mmotm http://lkml.kernel.org/r/20160810091226.6709-1-vbabka@suse.cz which might help and we might need more fixes. If all the efforts turn out insufficient we might need to revert to the original OOM detection behavior for !costly high order requests and rely solely on the reclaim progress. This would be really unfortunate and I believe that the decision based on the compaction feedback is really the way forward longterm. So let's try to fix compaction first. -- Michal Hocko SUSE Labs -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-kernel+owner@opensuse.org