Re: [RFC PATCH] mm/vmalloc: fix vmalloc which may return null if called with __GFP_NOFAIL

[Gao Xiang]

Hi,

On 2024/5/8 22:43, Hailong Liu wrote:
> On Wed, 08. May 21:41, Gao Xiang wrote:
> > +Cc Michal,
> >
> > On 2024/5/8 20:58, hailong.liu@xxxxxxxx wrote:
> > > From: "Hailong.Liu" <hailong.liu@xxxxxxxx>
> > >
> > > Commit a421ef303008 ("mm: allow !GFP_KERNEL allocations for kvmalloc")
> > > includes support for __GFP_NOFAIL, but it presents a conflict with
> > > commit dd544141b9eb ("vmalloc: back off when the current task is
> > > OOM-killed"). A possible scenario is as belows:
> > >
> > > process-a
> > > kvcalloc(n, m, GFP_KERNEL | __GFP_NOFAIL)
> > >       __vmalloc_node_range()
> > > 	__vmalloc_area_node()
> > > 	    vm_area_alloc_pages()
> > >               --> oom-killer send SIGKILL to process-a
> > >               if (fatal_signal_pending(current)) break;
> > > --> return NULL;
> > >
> > > to fix this, do not check fatal_signal_pending() in vm_area_alloc_pages()
> > > if __GFP_NOFAIL set.
> > >
> > > Reported-by: Oven <liyangouwen1@xxxxxxxx>
> > > Signed-off-by: Hailong.Liu <hailong.liu@xxxxxxxx>
> >
> > Why taging this as RFC here?  It seems a corner-case fix of
> > commit a421ef303008
> >
> > Thanks,
> > Gao Xiang
>
> Hi Gao Xiang:
>
> RFC here to wait for a better way to handle this case :).
> IMO, if vmalloc support __GFP_NOFAIL it should not return
> null even system is deadlock on memory.

The starting point is that kmalloc doesn't support __GFP_NOFAIL
if order > 1 (even for very short temporary uses), see:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/mm/page_alloc.c?h=v6.8#n2896

but it is possible if we have such page pointer array (since two
(order-1) pages can only keep 1024 8-byte entries, it can happen
if compression ratios are high), and kvmalloc(__GFP_NOFAIL) has
already been supported for almost two years, it will fallback to
order-0 allocation as described in commit e9c3cda4d86e
("mm, vmalloc: fix high order __GFP_NOFAIL allocations").

With my limited understanding, I'm not sure why it can cause
deadlock here since it will fallback to order-0 allocation then,
and such allocation is just for short temporary uses again
because kmalloc doesn't support order > 1 short memory
allocation strictly.

Thanks,
Gao Xiang

> --
>
> Best Regards,
> Hailong.