On Fri, Oct 19, 2018 at 10:44:39PM +0000, Roman Gushchin wrote:
> On Fri, Oct 19, 2018 at 07:35:36PM +0200, Uladzislau Rezki (Sony) wrote:
> > Objective
> > ---------
> > Initiative of improving vmalloc allocator comes from getting many issues
> > related to allocation time, i.e. sometimes it is terribly slow. As a result
> > many workloads which are sensitive for long (more than 1 millisecond) preemption
> > off scenario are affected by that slowness(test cases like UI or audio, etc.).
> >
> > The problem is that, currently an allocation of the new VA area is done over
> > busy list iteration until a suitable hole is found between two busy areas.
> > Therefore each new allocation causes the list being grown. Due to long list
> > and different permissive parameters an allocation can take a long time on
> > embedded devices(milliseconds).
> ...
> > 3) This one is related to PCPU allocator(see pcpu_alloc_test()). In that
> > stress test case i see that SUnreclaim(/proc/meminfo) parameter gets increased,
> > i.e. there is a memory leek somewhere in percpu allocator. It sounds like
> > a memory that is allocated by pcpu_get_vm_areas() sometimes is not freed.
> > Resulting in memory leaking or "Kernel panic":
> >
>
> Can you, please, try the following patch:
> 6685b357363b ("percpu: stop leaking bitmap metadata blocks") ?
>
I have tested that patch. It fixes the leak for sure. Thank you for a
good point.
>
> BTW, with growing number of vmalloc users (per-cpu allocator and bpf stuff are
> big drivers), I find the patchset very interesting.
>
> Thanks!
Thank you!
--
Vlad Rezki
