Re: [PATCH resend] mm/page_alloc: fix comment is __get_free_pages

From: Michal Hocko
Date: Sat Dec 16 2017 - 06:52:35 EST

On Fri 15-12-17 12:57:35, Andrew Morton wrote:
> On Fri, 15 Dec 2017 10:36:18 +0100 Michal Hocko <mhocko@xxxxxxxxxx> wrote:
> > > >
> > > > So do we care and I will resend the patch in that case or I just drop
> > > > this from my patch queue?
> > >
> > > Well.. I still think that silently accepting bad input would be bad
> > > practice. If we can just delete the assertion and have such a caller
> > > reliably blow up later on then that's good enough.
> >
> > The point is that if the caller checks for the failed allocation then
> > the result is a memory leak.
> That's if page_address(highmem page) returns NULL. I'm not sure what
> it returns, really - so many different implementations across so many
> different architectures.

I am not sure I follow. We only do care for HIGHMEM, right? And that one
returns NULL unless the high mem page is not kmaped.

> Oh well, it would have been nice to remove that VM_BUG_ON(). Why not
> just leave the code as it is now?

BUGing on a bogus usage is not popular anymore. Also checking for
something nobody actually does is a bit pointless. I will not insist

Michal Hocko