| Message ID | alpine.DEB.2.21.2002172139310.152060@chino.kir.corp.google.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | mm, thp: track fallbacks due to failed memcg charges separately | expand |
On Mon, Feb 17, 2020 at 09:41:40PM -0800, David Rientjes wrote: > The thp_fault_fallback stat in either /proc/vmstat is incremented if > either the hugepage allocation fails through the page allocator or the > hugepage charge fails through mem cgroup. > > This patch leaves this field untouched but adds a new field, > thp_fault_fallback_charge, which is incremented only when the mem cgroup > charge fails. > > This distinguishes between faults that want to be backed by hugepages but > fail due to fragmentation (or low memory conditions) and those that fail > due to mem cgroup limits. That can be used to determine the impact of > fragmentation on the system by excluding faults that failed due to memcg > usage. > > Signed-off-by: David Rientjes <rientjes@google.com> The patch looks good to me, but I noticed that we miss THP_FAULT_FALLBACK (and THP_FAULT_FALLBACK_CHARGE) accounting in shmem_getpage_gfp(). Could you fix this while you are there?
On Mon, Feb 17, 2020 at 09:41:40PM -0800, David Rientjes wrote: > The thp_fault_fallback stat in either /proc/vmstat is incremented if Nit: ^ "either" here looks wrong > either the hugepage allocation fails through the page allocator or the > hugepage charge fails through mem cgroup. > > This patch leaves this field untouched but adds a new field, > thp_fault_fallback_charge, which is incremented only when the mem cgroup > charge fails. > > This distinguishes between faults that want to be backed by hugepages but > fail due to fragmentation (or low memory conditions) and those that fail > due to mem cgroup limits. That can be used to determine the impact of > fragmentation on the system by excluding faults that failed due to memcg > usage. > > Signed-off-by: David Rientjes <rientjes@google.com> > --- > Documentation/admin-guide/mm/transhuge.rst | 5 +++++ > include/linux/vm_event_item.h | 1 + > mm/huge_memory.c | 2 ++ > mm/vmstat.c | 1 + > 4 files changed, 9 insertions(+) > > diff --git a/Documentation/admin-guide/mm/transhuge.rst b/Documentation/admin-guide/mm/transhuge.rst > --- a/Documentation/admin-guide/mm/transhuge.rst > +++ b/Documentation/admin-guide/mm/transhuge.rst > @@ -310,6 +310,11 @@ thp_fault_fallback > is incremented if a page fault fails to allocate > a huge page and instead falls back to using small pages. > > +thp_fault_fallback_charge > + is incremented if a page fault fails to charge a huge page and > + instead falls back to using small pages even through the ^ though > + allocation was successful. > + > thp_collapse_alloc_failed > is incremented if khugepaged found a range > of pages that should be collapsed into one huge page but failed > diff --git a/include/linux/vm_event_item.h b/include/linux/vm_event_item.h > --- a/include/linux/vm_event_item.h > +++ b/include/linux/vm_event_item.h > @@ -73,6 +73,7 @@ enum vm_event_item { PGPGIN, PGPGOUT, PSWPIN, PSWPOUT, > #ifdef CONFIG_TRANSPARENT_HUGEPAGE > THP_FAULT_ALLOC, > THP_FAULT_FALLBACK, > + THP_FAULT_FALLBACK_CHARGE, > THP_COLLAPSE_ALLOC, > THP_COLLAPSE_ALLOC_FAILED, > THP_FILE_ALLOC, > diff --git a/mm/huge_memory.c b/mm/huge_memory.c > --- a/mm/huge_memory.c > +++ b/mm/huge_memory.c > @@ -597,6 +597,7 @@ static vm_fault_t __do_huge_pmd_anonymous_page(struct vm_fault *vmf, > if (mem_cgroup_try_charge_delay(page, vma->vm_mm, gfp, &memcg, true)) { > put_page(page); > count_vm_event(THP_FAULT_FALLBACK); > + count_vm_event(THP_FAULT_FALLBACK_CHARGE); > return VM_FAULT_FALLBACK; > } > > @@ -1406,6 +1407,7 @@ vm_fault_t do_huge_pmd_wp_page(struct vm_fault *vmf, pmd_t orig_pmd) > put_page(page); > ret |= VM_FAULT_FALLBACK; > count_vm_event(THP_FAULT_FALLBACK); > + count_vm_event(THP_FAULT_FALLBACK_CHARGE); > goto out; > } > > diff --git a/mm/vmstat.c b/mm/vmstat.c > --- a/mm/vmstat.c > +++ b/mm/vmstat.c > @@ -1254,6 +1254,7 @@ const char * const vmstat_text[] = { > #ifdef CONFIG_TRANSPARENT_HUGEPAGE > "thp_fault_alloc", > "thp_fault_fallback", > + "thp_fault_fallback_charge", > "thp_collapse_alloc", > "thp_collapse_alloc_failed", > "thp_file_alloc",
On Tue, 18 Feb 2020, Kirill A. Shutemov wrote: > On Mon, Feb 17, 2020 at 09:41:40PM -0800, David Rientjes wrote: > > The thp_fault_fallback stat in either /proc/vmstat is incremented if > > either the hugepage allocation fails through the page allocator or the > > hugepage charge fails through mem cgroup. > > > > This patch leaves this field untouched but adds a new field, > > thp_fault_fallback_charge, which is incremented only when the mem cgroup > > charge fails. > > > > This distinguishes between faults that want to be backed by hugepages but > > fail due to fragmentation (or low memory conditions) and those that fail > > due to mem cgroup limits. That can be used to determine the impact of > > fragmentation on the system by excluding faults that failed due to memcg > > usage. > > > > Signed-off-by: David Rientjes <rientjes@google.com> > > The patch looks good to me, but I noticed that we miss THP_FAULT_FALLBACK > (and THP_FAULT_FALLBACK_CHARGE) accounting in shmem_getpage_gfp(). > > Could you fix this while you are there? Sure, I'll add it as a predecessor and also count THP_FAULT_ALLOC for consistency.
diff --git a/Documentation/admin-guide/mm/transhuge.rst b/Documentation/admin-guide/mm/transhuge.rst --- a/Documentation/admin-guide/mm/transhuge.rst +++ b/Documentation/admin-guide/mm/transhuge.rst @@ -310,6 +310,11 @@ thp_fault_fallback is incremented if a page fault fails to allocate a huge page and instead falls back to using small pages. +thp_fault_fallback_charge + is incremented if a page fault fails to charge a huge page and + instead falls back to using small pages even through the + allocation was successful. + thp_collapse_alloc_failed is incremented if khugepaged found a range of pages that should be collapsed into one huge page but failed diff --git a/include/linux/vm_event_item.h b/include/linux/vm_event_item.h --- a/include/linux/vm_event_item.h +++ b/include/linux/vm_event_item.h @@ -73,6 +73,7 @@ enum vm_event_item { PGPGIN, PGPGOUT, PSWPIN, PSWPOUT, #ifdef CONFIG_TRANSPARENT_HUGEPAGE THP_FAULT_ALLOC, THP_FAULT_FALLBACK, + THP_FAULT_FALLBACK_CHARGE, THP_COLLAPSE_ALLOC, THP_COLLAPSE_ALLOC_FAILED, THP_FILE_ALLOC, diff --git a/mm/huge_memory.c b/mm/huge_memory.c --- a/mm/huge_memory.c +++ b/mm/huge_memory.c @@ -597,6 +597,7 @@ static vm_fault_t __do_huge_pmd_anonymous_page(struct vm_fault *vmf, if (mem_cgroup_try_charge_delay(page, vma->vm_mm, gfp, &memcg, true)) { put_page(page); count_vm_event(THP_FAULT_FALLBACK); + count_vm_event(THP_FAULT_FALLBACK_CHARGE); return VM_FAULT_FALLBACK; } @@ -1406,6 +1407,7 @@ vm_fault_t do_huge_pmd_wp_page(struct vm_fault *vmf, pmd_t orig_pmd) put_page(page); ret |= VM_FAULT_FALLBACK; count_vm_event(THP_FAULT_FALLBACK); + count_vm_event(THP_FAULT_FALLBACK_CHARGE); goto out; } diff --git a/mm/vmstat.c b/mm/vmstat.c --- a/mm/vmstat.c +++ b/mm/vmstat.c @@ -1254,6 +1254,7 @@ const char * const vmstat_text[] = { #ifdef CONFIG_TRANSPARENT_HUGEPAGE "thp_fault_alloc", "thp_fault_fallback", + "thp_fault_fallback_charge", "thp_collapse_alloc", "thp_collapse_alloc_failed", "thp_file_alloc",
The thp_fault_fallback stat in either /proc/vmstat is incremented if either the hugepage allocation fails through the page allocator or the hugepage charge fails through mem cgroup. This patch leaves this field untouched but adds a new field, thp_fault_fallback_charge, which is incremented only when the mem cgroup charge fails. This distinguishes between faults that want to be backed by hugepages but fail due to fragmentation (or low memory conditions) and those that fail due to mem cgroup limits. That can be used to determine the impact of fragmentation on the system by excluding faults that failed due to memcg usage. Signed-off-by: David Rientjes <rientjes@google.com> --- Documentation/admin-guide/mm/transhuge.rst | 5 +++++ include/linux/vm_event_item.h | 1 + mm/huge_memory.c | 2 ++ mm/vmstat.c | 1 + 4 files changed, 9 insertions(+)