Re: [PATCH v3] mm/page_owner: use kvmalloc instead of kmalloc

[Michal Hocko]

On Wed 31-10-18 18:19:42, Miles Chen wrote:
> On Wed, 2018-10-31 at 11:15 +0100, Michal Hocko wrote:
> > On Wed 31-10-18 16:47:17, Miles Chen wrote:
> > > On Tue, 2018-10-30 at 09:15 +0100, Michal Hocko wrote:
> > > > On Tue 30-10-18 14:55:51, Miles Chen wrote:
> > > > [...]
> > > > > It's a real problem when using page_owner.
> > > > > I found this issue recently: I'm not able to read page_owner information
> > > > > during a overnight test. (error: read failed: Out of memory). I replace
> > > > > kmalloc() with vmalloc() and it worked well.
> > > > 
> > > > Is this with trimming the allocation to a single page and doing shorter
> > > > than requested reads?
> > > 
> > > 
> > > I printed out the allocate count on my device the request count is <=
> > > 4096. So I tested this scenario by trimming the count to from 4096 to
> > > 1024 bytes and it works fine. 
> > > 
> > > count = count > 1024? 1024: count;
> > > 
> > > It tested it on both 32bit and 64bit kernel.
> > 
> > Are you saying that you see OOMs for 4k size?
> > 
> yes, because kmalloc only use normal memor, not highmem + normal memory
> I think that's why vmalloc() works.

Can I see an OOM report please? I am especially interested that 1k
doesn't cause the problem because there shouldn't be that much of a
difference between the two. Larger allocations could be a result of
memory fragmentation but 1k vs. 4k to make a difference really seems
unexpected.

-- 
Michal Hocko
SUSE Labs