Re: [PATCH] mm: hugetlb: Only prep and add allocated folios for non-gigantic pages

[Nathan Chancellor <nathan@kernel.org>]

On Thu, Oct 19, 2023 at 01:33:05PM +0900, Sergey Senozhatsky wrote:
> On (23/10/18 15:20), Mike Kravetz wrote:
> > > I think you need to initialize h, otherwise what value is passed to
> > > prep_and_add_bootmem_folios if the loop is not run because the list is
> > > empty.  The compiler sees `h` is only given a value in the loop, so
> > > the loop must be run.  That's obviously hazardous, but the compiler
> > > assumes there's no UB. At least that's my limited understanding
> > > looking at the IR diff Nathan got me in
> > > https://github.com/ClangBuiltLinux/linux/issues/1946.
> > 
> > Thanks for looking closer at this Nick and Nathan!
> > 
> > I think you are saying the compiler is running the loop because it wants
> > to initialize h before passing the value to another function.  It does
> > this even if the explicit loop entry condition is false.  Is that correct?
> 
> The loop is getting promoted to "infinite" loop, there is no
> &pos->member != (head) condition check in the generated code
> at all (at least on my machine).
> 
> I wish we could at least get the "possibly uninitialized variable"
> warning from the compiler in this case, which we'd translate to
> "hold my beer, I'm going to try one thing".

GCC would warn about this under -Wmaybe-uninitialized but it has been
disabled in a normal build for the past three years, see commit
78a5255ffb6a ("Stop the ad-hoc games with -Wno-maybe-initialized").

  In function 'gather_bootmem_prealloc',
      inlined from 'hugetlb_init' at mm/hugetlb.c:4299:2:
  mm/hugetlb.c:3203:9: warning: 'h' may be used uninitialized [-Wmaybe-uninitialized]
   3203 |         prep_and_add_allocated_folios(h, &folio_list);
        |         ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  mm/hugetlb.c: In function 'hugetlb_init':
  mm/hugetlb.c:3166:24: note: 'h' was declared here
   3166 |         struct hstate *h, *prev_h = NULL;
        |                        ^

Clang's -Wconditional-uninitialized would have flagged it too but it
suffers from the same problems as -Wmaybe-uninitialized.

  mm/hugetlb.c:3203:32: warning: variable 'h' may be uninitialized when used here [-Wconditional-uninitialized]
   3203 |         prep_and_add_allocated_folios(h, &folio_list);
        |                                       ^
  mm/hugetlb.c:3166:18: note: initialize the variable 'h' to silence this warning
   3166 |         struct hstate *h, *prev_h = NULL;
        |                         ^
        |                          = NULL

I know clang has some handling for loops in -Wsometimes-uninitialized, I
wonder why that does not trigger here...

Cheers,
Nathan