On Mon, 25 Mar 2019 15:15:41 -0700
Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> wrote:
> On Mon, 25 Mar 2019 16:13:09 +0800 Yue Hu <zbestahu@xxxxxxxxx> wrote:
>
> > From: Yue Hu <huyue2@xxxxxxxxxx>
> >
> > A previous commit f022d8cb7ec7 ("mm: cma: Don't crash on allocation
> > if CMA area can't be activated") fixes the crash issue when activation
> > fails via setting cma->count as 0, same logic exists if bitmap
> > allocation fails.
> >
> > --- a/mm/cma.c
> > +++ b/mm/cma.c
> > @@ -106,8 +106,10 @@ static int __init cma_activate_area(struct cma *cma)
> >
> > cma->bitmap = kzalloc(bitmap_size, GFP_KERNEL);
> >
> > - if (!cma->bitmap)
> > + if (!cma->bitmap) {
> > + cma->count = 0;
> > return -ENOMEM;
> > + }
> >
> > WARN_ON_ONCE(!pfn_valid(pfn));
> > zone = page_zone(pfn_to_page(pfn));
>
> I'm unsure whether this is needed.
>
> kmalloc() within __init code is generally considered to be a "can't
> fail".
>
> If this was the only issue then I guess I'd take the patch if only for
> documentation/clarity purposes. But cma_areas[] is in bss and is
> guaranteed to be all-zeroes, so I suspect this bug is a can't-happen.
However, firstly cma->count will be assigned to size >> PAGE_SHIFT in
cma_init_reserved_mem().
> And we could revert f022d8cb7ec7 if we could be bothered (I can't).
>
>
