IIUC PVMW checks if the vma is possibly huge PMD mapped by
transparent_hugepage_active() and "pvmw->nr_pages >= HPAGE_PMD_NR".
Actually pvmw->nr_pages is returned by compound_nr() or
folio_nr_pages(), so the page should be THP as long as "pvmw->nr_pages
>= HPAGE_PMD_NR". And it is guaranteed THP is allocated for valid VMA
in the first place. But it may be not PMD mapped if the VMA is file
VMA and it is not properly aligned. The transhuge_vma_suitable()
is used to do such check, so replace transparent_hugepage_active() to
it, which is too heavy and overkilling.
Fixes: 2aff7a4755be ("mm: Convert page_vma_mapped_walk to work on PFNs")
Signed-off-by: Yang Shi <[email protected]>
---
mm/page_vma_mapped.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/mm/page_vma_mapped.c b/mm/page_vma_mapped.c
index c10f839fc410..2634565be175 100644
--- a/mm/page_vma_mapped.c
+++ b/mm/page_vma_mapped.c
@@ -237,13 +237,14 @@ bool page_vma_mapped_walk(struct page_vma_mapped_walk *pvmw)
spin_unlock(pvmw->ptl);
pvmw->ptl = NULL;
} else if (!pmd_present(pmde)) {
+ unsigned long haddr = pvmw->address & HPAGE_PMD_MASK;
/*
* If PVMW_SYNC, take and drop THP pmd lock so that we
* cannot return prematurely, while zap_huge_pmd() has
* cleared *pmd but not decremented compound_mapcount().
*/
if ((pvmw->flags & PVMW_SYNC) &&
- transparent_hugepage_active(vma) &&
+ transhuge_vma_suitable(vma, haddr) &&
(pvmw->nr_pages >= HPAGE_PMD_NR)) {
spinlock_t *ptl = pmd_lock(mm, pvmw->pmd);
--
2.26.3
On Thu, May 12, 2022 at 10:45:51AM -0700, Yang Shi wrote:
> IIUC PVMW checks if the vma is possibly huge PMD mapped by
> transparent_hugepage_active() and "pvmw->nr_pages >= HPAGE_PMD_NR".
>
> Actually pvmw->nr_pages is returned by compound_nr() or
> folio_nr_pages(), so the page should be THP as long as "pvmw->nr_pages
> >= HPAGE_PMD_NR". And it is guaranteed THP is allocated for valid VMA
> in the first place. But it may be not PMD mapped if the VMA is file
> VMA and it is not properly aligned. The transhuge_vma_suitable()
> is used to do such check, so replace transparent_hugepage_active() to
> it, which is too heavy and overkilling.
>
> Fixes: 2aff7a4755be ("mm: Convert page_vma_mapped_walk to work on PFNs")
I think Fixes is a bit much. There's no bug being fixed here. This is
just an optimisation. Is it an important optimisation? We could put a
bool into page_vma_mapped_walk() so we only have to ask the page whether
it's PMD-mappable once per walk rather than for each VMA.
On Thu, 12 May 2022 10:45:51 -0700 Yang Shi <[email protected]> wrote:
> IIUC PVMW checks if the vma is possibly huge PMD mapped by
> transparent_hugepage_active() and "pvmw->nr_pages >= HPAGE_PMD_NR".
>
> Actually pvmw->nr_pages is returned by compound_nr() or
> folio_nr_pages(), so the page should be THP as long as "pvmw->nr_pages
> >= HPAGE_PMD_NR". And it is guaranteed THP is allocated for valid VMA
> in the first place. But it may be not PMD mapped if the VMA is file
> VMA and it is not properly aligned. The transhuge_vma_suitable()
> is used to do such check, so replace transparent_hugepage_active() to
> it, which is too heavy and overkilling.
>
> ...
>
> --- a/mm/page_vma_mapped.c
> +++ b/mm/page_vma_mapped.c
> @@ -237,13 +237,14 @@ bool page_vma_mapped_walk(struct page_vma_mapped_walk *pvmw)
> spin_unlock(pvmw->ptl);
> pvmw->ptl = NULL;
> } else if (!pmd_present(pmde)) {
> + unsigned long haddr = pvmw->address & HPAGE_PMD_MASK;
This hits
#define HPAGE_PMD_MASK ({ BUILD_BUG(); 0; })
when CONFIG_TRANSPARENT_HUGEPAGE=n (x86_64 allnoconfig).
On Thu, May 12, 2022 at 8:30 PM Matthew Wilcox <[email protected]> wrote:
>
> On Thu, May 12, 2022 at 10:45:51AM -0700, Yang Shi wrote:
> > IIUC PVMW checks if the vma is possibly huge PMD mapped by
> > transparent_hugepage_active() and "pvmw->nr_pages >= HPAGE_PMD_NR".
> >
> > Actually pvmw->nr_pages is returned by compound_nr() or
> > folio_nr_pages(), so the page should be THP as long as "pvmw->nr_pages
> > >= HPAGE_PMD_NR". And it is guaranteed THP is allocated for valid VMA
> > in the first place. But it may be not PMD mapped if the VMA is file
> > VMA and it is not properly aligned. The transhuge_vma_suitable()
> > is used to do such check, so replace transparent_hugepage_active() to
> > it, which is too heavy and overkilling.
> >
> > Fixes: 2aff7a4755be ("mm: Convert page_vma_mapped_walk to work on PFNs")
>
> I think Fixes is a bit much. There's no bug being fixed here. This is
> just an optimisation. Is it an important optimisation? We could put a
Yeah, it is just an optimization, will remove the fix tag.
I'm trying to do some cleanup for all the transhuge_page_* checks
suggested by Vlastimil. I should be able to kill
transparent_hugepage_active() by replacing it with
transhuge_vma_suitable() here.
> bool into page_vma_mapped_walk() so we only have to ask the page whether
> it's PMD-mappable once per walk rather than for each VMA.
The page may be PMD-mappable for one VMA, but not for the other VMA.
>