Re: [PATCH v3 4/6] mm: pvmw: add support for walking devmap pages

[Muchun Song <songmuchun@bytedance.com>]

On Tue, Mar 1, 2022 at 5:26 AM Andrew Morton <akpm@linux-foundation.org> wrote:
>
> On Mon, 28 Feb 2022 14:35:34 +0800 Muchun Song <songmuchun@bytedance.com> wrote:
>
> > The devmap pages can not use page_vma_mapped_walk() to check if a huge
> > devmap page is mapped into a vma.  Add support for walking huge devmap
> > pages so that DAX can use it in the next patch.
> >
>
> x86_64 allnoconfig:
>
> In file included from <command-line>:
> In function 'check_pmd',
>     inlined from 'page_vma_mapped_walk' at mm/page_vma_mapped.c:219:10:
> ././include/linux/compiler_types.h:347:45: error: call to '__compiletime_assert_232' declared with attribute error: BUILD_BUG failed
>   347 |         _compiletime_assert(condition, msg, __compiletime_assert_, __COUNTER__)
>       |                                             ^
> ././include/linux/compiler_types.h:328:25: note: in definition of macro '__compiletime_assert'
>   328 |                         prefix ## suffix();                             \
>       |                         ^~~~~~
> ././include/linux/compiler_types.h:347:9: note: in expansion of macro '_compiletime_assert'
>   347 |         _compiletime_assert(condition, msg, __compiletime_assert_, __COUNTER__)
>       |         ^~~~~~~~~~~~~~~~~~~
> ./include/linux/build_bug.h:39:37: note: in expansion of macro 'compiletime_assert'
>    39 | #define BUILD_BUG_ON_MSG(cond, msg) compiletime_assert(!(cond), msg)
>       |                                     ^~~~~~~~~~~~~~~~~~
> ./include/linux/build_bug.h:59:21: note: in expansion of macro 'BUILD_BUG_ON_MSG'
>    59 | #define BUILD_BUG() BUILD_BUG_ON_MSG(1, "BUILD_BUG failed")
>       |                     ^~~~~~~~~~~~~~~~
> ./include/linux/huge_mm.h:307:28: note: in expansion of macro 'BUILD_BUG'
>   307 | #define HPAGE_PMD_SHIFT ({ BUILD_BUG(); 0; })
>       |                            ^~~~~~~~~
> ./include/linux/huge_mm.h:104:26: note: in expansion of macro 'HPAGE_PMD_SHIFT'
>   104 | #define HPAGE_PMD_ORDER (HPAGE_PMD_SHIFT-PAGE_SHIFT)
>       |                          ^~~~~~~~~~~~~~~
> ./include/linux/huge_mm.h:105:26: note: in expansion of macro 'HPAGE_PMD_ORDER'
>   105 | #define HPAGE_PMD_NR (1<<HPAGE_PMD_ORDER)
>       |                          ^~~~~~~~~~~~~~~
> mm/page_vma_mapped.c:113:20: note: in expansion of macro 'HPAGE_PMD_NR'
>   113 |         if ((pfn + HPAGE_PMD_NR - 1) < pvmw->pfn)
>       |                    ^~~~~~~~~~~~
> make[1]: *** [scripts/Makefile.build:288: mm/page_vma_mapped.o] Error 1
> make: *** [Makefile:1971: mm] Error 2
>
>
> because check_pmd() uses HPAGE_PMD_NR and
>
> #else /* CONFIG_TRANSPARENT_HUGEPAGE */
> #define HPAGE_PMD_SHIFT ({ BUILD_BUG(); 0; })
>
> I don't immediately see why this patch triggers it...

Maybe the reason is as follows.

The first check_pmd() is wrapped inside `if (pmd_trans_huge(pmde))`
block, since pmd_trans_huge() just returns 0, check_pmd() will be
optimized out.  There is a `if (!thp_migration_supported()) return;` block
before the second check_pmd(), however, thp_migration_supported()
returns 0 on riscv. So the second check_pmd() can be optimized out as
well.  I think I should replace `pmd_leaf` with `pmd_trans_huge() ||
pmd_devmap()`
to fix it.

Thanks.