Re: [PATCH] mm: page_alloc: document kmemleak's non-blockable __GFP_NOFAIL case

[Yang Shi]

On 7/15/19 6:06 AM, Matthew Wilcox wrote:
> On Sun, Jul 14, 2019 at 08:47:07PM -0700, Yang Shi wrote:
> > On 7/13/19 2:25 PM, Matthew Wilcox wrote:
> > > On Sat, Jul 13, 2019 at 04:49:04AM +0800, Yang Shi wrote:
> > > > When running ltp's oom test with kmemleak enabled, the below warning was
> > > > triggerred since kernel detects __GFP_NOFAIL & ~__GFP_DIRECT_RECLAIM is
> > > > passed in:
> > > There are lots of places where kmemleak will call kmalloc with
> > > __GFP_NOFAIL and ~__GFP_DIRECT_RECLAIM (including the XArray code, which
> > > is how I know about it).  It needs to be fixed to allow its internal
> > > allocations to fail and return failure of the original allocation as
> > > a consequence.
> > Do you mean kmemleak internal allocation? It would fail even though
> > __GFP_NOFAIL is passed in if GFP_NOWAIT is specified. Currently buddy
> > allocator will not retry if the allocation is non-blockable.
> Actually it sets off a warning.  Which is the right response from the
> core mm code because specifying __GFP_NOFAIL and __GFP_NOWAIT makes no
> sense.

Yes, this is what I meant. Kmemleak did a trick to fool fault-injection 
by passing in __GFP_NOFAIL, but it doesn't make sense for non-blockable 
allocation.