The allocation flags of anonymous transparent huge pages can be controlled
through the files in /sys/kernel/mm/transparent_hugepage/defrag, which can
help the system from getting bogged down in the page reclaim and compaction
code when many THPs are getting allocated simultaneously.
However, the gfp_mask for shmem THP allocations were not limited by those
configuration settings, and some workloads ended up with all CPUs stuck
on the LRU lock in the page reclaim code, trying to allocate dozens of
THPs simultaneously.
This patch applies the same configurated limitation of THPs to shmem
hugepage allocations, to prevent that from happening.
This way a THP defrag setting of "never" or "defer+madvise" will result
in quick allocation failures without direct reclaim when no 2MB free
pages are available.
With this patch applied, THP allocations for tmpfs will be a little
more aggressive than today for files mmapped with MADV_HUGEPAGE,
and a little less aggressive for files that are not mmapped or
mapped without that flag.
Signed-off-by: Rik van Riel <[email protected]>
---
include/linux/gfp.h | 2 ++
mm/huge_memory.c | 6 +++---
mm/shmem.c | 8 +++++---
3 files changed, 10 insertions(+), 6 deletions(-)
diff --git a/include/linux/gfp.h b/include/linux/gfp.h
index c603237e006c..c7615c9ba03c 100644
--- a/include/linux/gfp.h
+++ b/include/linux/gfp.h
@@ -614,6 +614,8 @@ bool gfp_pfmemalloc_allowed(gfp_t gfp_mask);
extern void pm_restrict_gfp_mask(void);
extern void pm_restore_gfp_mask(void);
+extern gfp_t vma_thp_gfp_mask(struct vm_area_struct *vma);
+
#ifdef CONFIG_PM_SLEEP
extern bool pm_suspended_storage(void);
#else
diff --git a/mm/huge_memory.c b/mm/huge_memory.c
index 9474dbc150ed..c5d03b2f2f2f 100644
--- a/mm/huge_memory.c
+++ b/mm/huge_memory.c
@@ -649,9 +649,9 @@ static vm_fault_t __do_huge_pmd_anonymous_page(struct vm_fault *vmf,
* available
* never: never stall for any thp allocation
*/
-static inline gfp_t alloc_hugepage_direct_gfpmask(struct vm_area_struct *vma)
+gfp_t vma_thp_gfp_mask(struct vm_area_struct *vma)
{
- const bool vma_madvised = !!(vma->vm_flags & VM_HUGEPAGE);
+ const bool vma_madvised = vma && (vma->vm_flags & VM_HUGEPAGE);
/* Always do synchronous compaction */
if (test_bit(TRANSPARENT_HUGEPAGE_DEFRAG_DIRECT_FLAG, &transparent_hugepage_flags))
@@ -744,7 +744,7 @@ vm_fault_t do_huge_pmd_anonymous_page(struct vm_fault *vmf)
pte_free(vma->vm_mm, pgtable);
return ret;
}
- gfp = alloc_hugepage_direct_gfpmask(vma);
+ gfp = vma_thp_gfp_mask(vma);
page = alloc_hugepage_vma(gfp, vma, haddr, HPAGE_PMD_ORDER);
if (unlikely(!page)) {
count_vm_event(THP_FAULT_FALLBACK);
diff --git a/mm/shmem.c b/mm/shmem.c
index 537c137698f8..6c3cb192a88d 100644
--- a/mm/shmem.c
+++ b/mm/shmem.c
@@ -1545,8 +1545,8 @@ static struct page *shmem_alloc_hugepage(gfp_t gfp,
return NULL;
shmem_pseudo_vma_init(&pvma, info, hindex);
- page = alloc_pages_vma(gfp | __GFP_COMP | __GFP_NORETRY | __GFP_NOWARN,
- HPAGE_PMD_ORDER, &pvma, 0, numa_node_id(), true);
+ page = alloc_pages_vma(gfp, HPAGE_PMD_ORDER, &pvma, 0, numa_node_id(),
+ true);
shmem_pseudo_vma_destroy(&pvma);
if (page)
prep_transhuge_page(page);
@@ -1802,6 +1802,7 @@ static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
struct page *page;
enum sgp_type sgp_huge = sgp;
pgoff_t hindex = index;
+ gfp_t huge_gfp;
int error;
int once = 0;
int alloced = 0;
@@ -1887,7 +1888,8 @@ static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
}
alloc_huge:
- page = shmem_alloc_and_acct_page(gfp, inode, index, true);
+ huge_gfp = vma_thp_gfp_mask(vma);
+ page = shmem_alloc_and_acct_page(huge_gfp, inode, index, true);
if (IS_ERR(page)) {
alloc_nohuge:
page = shmem_alloc_and_acct_page(gfp, inode,
--
2.25.4
On Thu 05-11-20 14:15:07, Rik van Riel wrote:
> The allocation flags of anonymous transparent huge pages can be controlled
> through the files in /sys/kernel/mm/transparent_hugepage/defrag, which can
> help the system from getting bogged down in the page reclaim and compaction
> code when many THPs are getting allocated simultaneously.
>
> However, the gfp_mask for shmem THP allocations were not limited by those
> configuration settings, and some workloads ended up with all CPUs stuck
> on the LRU lock in the page reclaim code, trying to allocate dozens of
> THPs simultaneously.
>
> This patch applies the same configurated limitation of THPs to shmem
> hugepage allocations, to prevent that from happening.
I believe you should also exaplain why we want to control defrag by the
global knob while the enable logic is per mount.
> This way a THP defrag setting of "never" or "defer+madvise" will result
> in quick allocation failures without direct reclaim when no 2MB free
> pages are available.
>
> With this patch applied, THP allocations for tmpfs will be a little
> more aggressive than today for files mmapped with MADV_HUGEPAGE,
> and a little less aggressive for files that are not mmapped or
> mapped without that flag.
This begs some numbers. A little is rather bad unit of performance. I do
agree that unifying those makes sense in general though.
> Signed-off-by: Rik van Riel <[email protected]>
> ---
> include/linux/gfp.h | 2 ++
> mm/huge_memory.c | 6 +++---
> mm/shmem.c | 8 +++++---
> 3 files changed, 10 insertions(+), 6 deletions(-)
--
Michal Hocko
SUSE Labs
On Thu, 2020-11-12 at 11:52 +0100, Michal Hocko wrote:
> On Thu 05-11-20 14:15:07, Rik van Riel wrote:
> >
> > This patch applies the same configurated limitation of THPs to
> > shmem
> > hugepage allocations, to prevent that from happening.
>
> I believe you should also exaplain why we want to control defrag by
> the
> global knob while the enable logic is per mount.
I added that to the changelog for the next version of
the patches.
> > This way a THP defrag setting of "never" or "defer+madvise" will
> > result
> > in quick allocation failures without direct reclaim when no 2MB
> > free
> > pages are available.
> >
> > With this patch applied, THP allocations for tmpfs will be a little
> > more aggressive than today for files mmapped with MADV_HUGEPAGE,
> > and a little less aggressive for files that are not mmapped or
> > mapped without that flag.
>
> This begs some numbers. A little is rather bad unit of performance. I
> do
> agree that unifying those makes sense in general though.
The aggressiveness is in changes to the gfp_mask, eg by
adding __GFP_NORETRY. How that translates into THP
allocation success rates is entirely dependent on the
workload and on what else is in memory at the time.
I am not sure any
numbers I could gather will be
representative for anything but the workloads I am
testing.
However, I did find an issue in hugepage_vma_check
that prevents khugepaged from collapsing pages on
shmem filesystems mounted with huge=always or
huge=within_size when transparent_hugepage/enabled
is set to [madvise].
The next version of the series will have a third
patch, in order to fix that.
--
All Rights Reversed.
On Fri 13-11-20 22:44:20, Rik van Riel wrote:
> On Thu, 2020-11-12 at 11:52 +0100, Michal Hocko wrote:
> > On Thu 05-11-20 14:15:07, Rik van Riel wrote:
> > >
> > > This patch applies the same configurated limitation of THPs to
> > > shmem
> > > hugepage allocations, to prevent that from happening.
> >
> > I believe you should also exaplain why we want to control defrag by
> > the
> > global knob while the enable logic is per mount.
>
> I added that to the changelog for the next version of
> the patches.
>
> > > This way a THP defrag setting of "never" or "defer+madvise" will
> > > result
> > > in quick allocation failures without direct reclaim when no 2MB
> > > free
> > > pages are available.
> > >
> > > With this patch applied, THP allocations for tmpfs will be a little
> > > more aggressive than today for files mmapped with MADV_HUGEPAGE,
> > > and a little less aggressive for files that are not mmapped or
> > > mapped without that flag.
> >
> > This begs some numbers. A little is rather bad unit of performance. I
> > do
> > agree that unifying those makes sense in general though.
>
> The aggressiveness is in changes to the gfp_mask, eg by
> adding __GFP_NORETRY. How that translates into THP
> allocation success rates is entirely dependent on the
> workload and on what else is in memory at the time.
Yes and that is why I would argue about consistency with THP rather than
put claims that hard to back by numbers.
--
Michal Hocko
SUSE Labs