Re: [dm-devel] [RFC PATCH 2/2] mm, mempool: do not throttle PF_LESS_THROTTLE tasks

[Mikulas Patocka]

On Thu, 24 Nov 2016, Michal Hocko wrote:

> On Wed 23-11-16 16:11:59, Mikulas Patocka wrote:
> [...]
> > Hi Michal
> > 
> > So, here Google developers hit a stacktrace where a block device driver is 
> > being throttled in the memory management:
> > 
> > https://www.redhat.com/archives/dm-devel/2016-November/msg00158.html
> > 
> > dm-bufio layer is something like a buffer cache, used by block device 
> > drivers. Unlike the real buffer cache, dm-bufio guarantees forward 
> > progress even if there is no memory free.
> > 
> > dm-bufio does something similar like a mempool allocation, it tries an 
> > allocation with GFP_NOIO | __GFP_NORETRY | __GFP_NOMEMALLOC | __GFP_NOWARN 
> > (just like a mempool) and if it fails, it will reuse some existing buffer.
> > 
> > Here, they caught it being throttled in the memory management:
> > 
> >    Workqueue: kverityd verity_prefetch_io
> >    __switch_to+0x9c/0xa8
> >    __schedule+0x440/0x6d8
> >    schedule+0x94/0xb4
> >    schedule_timeout+0x204/0x27c
> >    schedule_timeout_uninterruptible+0x44/0x50
> >    wait_iff_congested+0x9c/0x1f0
> >    shrink_inactive_list+0x3a0/0x4cc
> >    shrink_lruvec+0x418/0x5cc
> >    shrink_zone+0x88/0x198
> >    try_to_free_pages+0x51c/0x588
> >    __alloc_pages_nodemask+0x648/0xa88
> >    __get_free_pages+0x34/0x7c
> >    alloc_buffer+0xa4/0x144
> >    __bufio_new+0x84/0x278
> >    dm_bufio_prefetch+0x9c/0x154
> >    verity_prefetch_io+0xe8/0x10c
> >    process_one_work+0x240/0x424
> >    worker_thread+0x2fc/0x424
> >    kthread+0x10c/0x114
> > 
> > Will you consider removing vm throttling for __GFP_NORETRY allocations?
> 
> As I've already said before I do not think that tweaking __GFP_NORETRY
> is the right approach is the right approach. The whole point of the flag
> is to not loop in the _allocator_ and it has nothing to do with the reclaim
> and the way how it is doing throttling.
> 
> On the other hand I perfectly understand your point and a lack of
> anything between GFP_NOWAIT and ___GFP_DIRECT_RECLAIM can be a bit
> frustrating. It would be nice to have sime middle ground - only a
> light reclaim involved and a quick back off if the memory is harder to
> reclaim. That is a hard thing to do, though because all the reclaimers
> (including slab shrinkers) would have to be aware of this concept to
> work properly.
> 
> I have read the report from the link above and I am really wondering why
> s@GFP_NOIO@GFP_NOWAIT@ is not the right way to go there. You have argued
> about a clean page cache would force buffer reuse. That might be true
> to some extent but is it a real problem?

The dm-bufio cache is limited by default to 2% of all memory. And the 
buffers are freed after 5 minutes of not being used.

It is unfair to reclaim the small dm-bufio cache (that was recently used) 
instead of the big page cache (that could be indefinitely old).

> Please note that even
> GFP_NOWAIT allocations will wake up kspwad which should clean up that

The mempool is also using GFP_NOIO allocations - so do you claim that it 
should not use GFP_NOIO too?

You should provide a clear API that the block device drivers should use to 
allocate memory - not to apply band aid to vm throttling problems as they 
are being discovered.

> clean page cache in the background. I would even expect kswapd being
> active at the time when NOWAIT requests hit the min watermark. If that
> is not the case then we should probably think about why kspwad is not
> proactive enough rather than tweaking __GFP_NORETRY semantic.
> 
> Thanks!
> -- 
> Michal Hocko
> SUSE Labs

Mikulas