Re: [patch 3/9] no PF_MEMALLOC tinkering

[Nick Piggin <nickpiggin@yahoo.com.au>]

Andrew Morton wrote:
> Nick Piggin <nickpiggin@yahoo.com.au> wrote:
> 
>>PF_MEMALLOC is really not a tool for tinkering. It is pretty specifically
>> used to prevent recursion into page reclaim, and to prevent low memory
>> deadlocks.
>>
>> The mm/swap_state.c code was the only legitimate tinkerer. Its concern
>> was addressed by the previous patch.
> 
> 
> What previous patch?  radix tree allocation doesn't use mempools, so this
> patch will cause add_to_swap() to oom the machine with radix tree node
> allocations.
> 

Sorry, just assumed they did fromt that comment.

> Now if we were to add __GFP_NOMEMALLOC in add_to_swap() then things would
> work as we want them to.
> 

That would be good.

> 
> The dm_crypt change looks OK.
> 
> 
> The code in mpage.c is saying "if we failed to allocate a correctly-sized
> bvec and if we're doing pageout then try to allocate a smaller-sized bvec
> instead".
> 
> It's probably fairly useless, but afaict there's nothing in any of the
> other patches here which makes it redundant.
> 

Well, I didn't like it because it uses mempools anyway, so it
might as well just wait for its allocation.

My motivation is to remove all external users of PF_MEMALLOC,
really.

But it looks like in this case, the code is dead anyway, because
mempool_alloc will never return NULL if it is passed __GFP_WAIT.

-- 
SUSE Labs, Novell Inc.