Re: [PATCH] mm/shmem: Inline shmem_is_huge() for disabled transparent hugepages

[Peter Xu]

On Wed, Apr 10, 2024 at 06:40:55PM +0200, David Hildenbrand wrote:
> On 10.04.24 18:33, Peter Xu wrote:
> > On Wed, Apr 10, 2024 at 06:12:34PM +0200, David Hildenbrand wrote:
> > > On 10.04.24 18:07, Sumanth Korikkar wrote:
> > > > On Wed, Apr 10, 2024 at 05:51:28PM +0200, David Hildenbrand wrote:
> > > > > On 10.04.24 17:26, Sumanth Korikkar wrote:
> > > > > > On Wed, Apr 10, 2024 at 02:34:35PM +0200, David Hildenbrand wrote:
> > > > > > > On 09.04.24 17:54, Sumanth Korikkar wrote:
> > > > > > > > In order to  minimize code size (CONFIG_CC_OPTIMIZE_FOR_SIZE=y),
> > > > > > > > compiler might choose to make a regular function call (out-of-line) for
> > > > > > > > shmem_is_huge() instead of inlining it. When transparent hugepages are
> > > > > > > > disabled (CONFIG_TRANSPARENT_HUGEPAGE=n), it can cause compilation
> > > > > > > > error.
> > > > > > > > 
> > > > > > > > mm/shmem.c: In function ‘shmem_getattr’:
> > > > > > > > ./include/linux/huge_mm.h:383:27: note: in expansion of macro ‘BUILD_BUG’
> > > > > > > >       383 | #define HPAGE_PMD_SIZE ({ BUILD_BUG(); 0; })
> > > > > > > >           |                           ^~~~~~~~~
> > > > > > > > mm/shmem.c:1148:33: note: in expansion of macro ‘HPAGE_PMD_SIZE’
> > > > > > > >      1148 |                 stat->blksize = HPAGE_PMD_SIZE;
> > > > > > > > 
> > > > > > > > To prevent the possible error, always inline shmem_is_huge() when
> > > > > > > > transparent hugepages are disabled.
> > > > > > > > 
> > > > > > > 
> > > > > > > Do you know which commit introduced that?
> > > > > > Hi David,
> > > > > > 
> > > > > > Currently with CONFIG_CC_OPTIMIZE_FOR_SIZE=y and expirementing with
> > > > > > -fPIC kernel compiler option, I could see this error on s390.
> > > > > 
> > > > > Got it. I assume on Linus' tree, not mm/unstable?
> > > > 
> > > > It's not yet upstream.
> > > > > 
> > > > > > 
> > > > > > However, default kernel compiler options doesnt end up with the above
> > > > > > pattern right now.
> > > > > 
> > > > > Okay, just asking if this is related to recent HPAGE_PMD_SIZE changes:
> > > > > 
> > > > > commit c1a1e497a3d5711dbf8fa6d7432d6b83ec18c26f
> > > > > Author: Peter Xu <peterx@xxxxxxxxxx>
> > > > > Date:   Wed Mar 27 11:23:22 2024 -0400
> > > > > 
> > > > >       mm: make HPAGE_PXD_* macros even if !THP
> > > > > 
> > > > > Which is still in mm-unstable and not upstream.
> > > > 
> > > > Not related to this commit. I tried on master branch.
> > > 
> > > Thanks! Can you try with Peters patch? (ccing Peter)
> > > 
> > > If I am not wrong, that should also resolve the issue you are seeing.
> > 
> > David,
> > 
> > Do you mean this one?
> > 
> > https://lore.kernel.org/all/20240403013249.1418299-4-peterx@xxxxxxxxxx/
> > 
> 
> No, I meant:
> 
> https://lore.kernel.org/all/20240327152332.950956-4-peterx@xxxxxxxxxx/
> 
> which removes the "#define HPAGE_PMD_SIZE ({ BUILD_BUG(); 0; })" that we
> seem to trigger here.
> 
> 
> ... but it's been a long day, so maybe I'm all wrong :)

Ah..  So I thought it was one step further. :)

Then that shouldn't be the case; it didn't remove it but defined properly
with HPAGE_PMD_SHIFT:

+#define HPAGE_PMD_SIZE	((1UL) << HPAGE_PMD_SHIFT)

Now we even have that properly defined for HUGETLB_PAGE, while prior to
that we should hit this issue easier (even with !THP+HUGETLB_PAGE).

> 
> > That's indeed similar but that was for pud_pfn() not HPAGE_* stuff.
> > 
> > I just had a quick look, Sumanth's fix looks valid, and IIUC the goal is
> > also that we should keep these build checks around for the long term goal
> > (Jason definitely preferred that [1] too, which I agree).
> > 
> > I removed that build check there for pud_pfn just to avoid other build
> > fallouts for other archs as a temporary measure.  For this one if it's in
> > common code for a long time and if it's the single spot maybe it's nice to
> > have this patch as proposed, as it means it optimizes the if check too
> > besides fixing the build error.  After all referencing HPAGE_* with
> > !THP+!HUGETLB shouldn't happen logically.
> > 
> > [1] https://lore.kernel.org/r/20240404112404.GG1723999@xxxxxxxxxx
> > 
> > Thanks,
> > 
> 
> -- 
> Cheers,
> 
> David / dhildenb
> 

-- 
Peter Xu