On 6/22/18 9:19 AM, Kirill A. Shutemov wrote:
On Fri, Jun 22, 2018 at 04:04:12PM +0000, yang.shi@xxxxxxxxxxxxxxxxx wrote:
On 6/22/18 12:59 AM, Kirill A. Shutemov wrote:
On Thu, Jun 21, 2018 at 11:15:48PM +0000, yang.shi@xxxxxxxxxxxxxxxxx wrote:
When merging anonymous page vma, if the size of vma can fit in at least
one hugepage, the mm will be registered for khugepaged for collapsing
THP in the future.
But, it skips shmem vma. Doing so for shmem too, but not file-private
mapping, when merging vma in order to increase the odd to collapse
hugepage by khugepaged.
Signed-off-by: Yang Shi <yang.shi@xxxxxxxxxxxxxxxxx>
Cc: Hugh Dickins <hughd@xxxxxxxxxx>
Cc: Kirill A. Shutemov <kirill.shutemov@xxxxxxxxxxxxxxx>
Cc: Vlastimil Babka <vbabka@xxxxxxx>
---
v1 --> 2:
* Exclude file-private mapping per Kirill's comment
mm/khugepaged.c | 8 ++++++--
1 file changed, 6 insertions(+), 2 deletions(-)
diff --git a/mm/khugepaged.c b/mm/khugepaged.c
index d7b2a4b..9b0ec30 100644
--- a/mm/khugepaged.c
+++ b/mm/khugepaged.c
@@ -440,8 +440,12 @@ int khugepaged_enter_vma_merge(struct vm_area_struct *vma,
* page fault if needed.
*/
return 0;
- if (vma->vm_ops || (vm_flags & VM_NO_KHUGEPAGED))
- /* khugepaged not yet working on file or special mappings */
+ if ((vma->vm_ops && (!shmem_file(vma->vm_file) || vma->anon_vma)) ||
+ (vm_flags & VM_NO_KHUGEPAGED))
+ /*
+ * khugepaged not yet working on non-shmem file or special
+ * mappings. And, file-private shmem THP is not supported.
+ */
return 0;
My point was that vma->anon_vma check above this one should not prevent
collapse for shmem.
Looking into this more, I think we should just replace all these checks
with hugepage_vma_check() call.
I got a little bit confused here. I thought the condition to *not* collapse
file-private shmem mapping should be:
shmem_file(vma->vm_file) && vma->anon_vma
Is this right?
No, if shmem_file(vma->vm_file) is true, vma->anon_vma doesn't matter.
We don't care about anon_vma in such VMA as we don't touch file-private
pages.
Thanks, I misunderstood that. hugepage_vma_check() sounds reasonable.
Will fix in v3 soon.
Yang
