The pmd_huge() is used to validate if the pmd entry is mapped by a huge
page, also including the case of non-present (migration or hwpoisoned)
pmd entry on arm64 or x86 architectures. That means the pmd_pfn() can
not get the correct pfn number for the non-present pmd entry, which
will cause damon_get_page() to get an incorrect page struct (also
may be NULL by pfn_to_online_page()) to make the access statistics
incorrect.
Moreover it does not make sense that we still waste time to get the
page of the non-present entry, just treat it as not-accessed and skip it,
that keeps consistent with non-present pte level entry.
Thus adding a pmd entry present validation to fix above issues.
Signed-off-by: Baolin Wang <[email protected]>
Reviewed-by: SeongJae Park <[email protected]>
---
Changes from v1:
- Update the commit message to make it more clear.
- Add reviewed tag from SeongJae.
---
mm/damon/vaddr.c | 10 ++++++++++
1 file changed, 10 insertions(+)
diff --git a/mm/damon/vaddr.c b/mm/damon/vaddr.c
index 3c7b9d6..1d16c6c 100644
--- a/mm/damon/vaddr.c
+++ b/mm/damon/vaddr.c
@@ -304,6 +304,11 @@ static int damon_mkold_pmd_entry(pmd_t *pmd, unsigned long addr,
if (pmd_huge(*pmd)) {
ptl = pmd_lock(walk->mm, pmd);
+ if (!pmd_present(*pmd)) {
+ spin_unlock(ptl);
+ return 0;
+ }
+
if (pmd_huge(*pmd)) {
damon_pmdp_mkold(pmd, walk->mm, addr);
spin_unlock(ptl);
@@ -431,6 +436,11 @@ static int damon_young_pmd_entry(pmd_t *pmd, unsigned long addr,
#ifdef CONFIG_TRANSPARENT_HUGEPAGE
if (pmd_huge(*pmd)) {
ptl = pmd_lock(walk->mm, pmd);
+ if (!pmd_present(*pmd)) {
+ spin_unlock(ptl);
+ return 0;
+ }
+
if (!pmd_huge(*pmd)) {
spin_unlock(ptl);
goto regular_page;
--
1.8.3.1
The pmd_huge() is usually used to indicate if a pmd level hugetlb,
however a pmd mapped huge page can only be THP in damon_mkold_pmd_entry()
or damon_young_pmd_entry(), so replacing pmd_huge() with pmd_trans_huge()
in this case to make code more readable according to the discussion [1].
[1] https://lore.kernel.org/all/[email protected]/
Signed-off-by: Baolin Wang <[email protected]>
---
- New patch in v2.
---
mm/damon/vaddr.c | 8 ++++----
1 file changed, 4 insertions(+), 4 deletions(-)
diff --git a/mm/damon/vaddr.c b/mm/damon/vaddr.c
index 1d16c6c..cc04d46 100644
--- a/mm/damon/vaddr.c
+++ b/mm/damon/vaddr.c
@@ -302,14 +302,14 @@ static int damon_mkold_pmd_entry(pmd_t *pmd, unsigned long addr,
pte_t *pte;
spinlock_t *ptl;
- if (pmd_huge(*pmd)) {
+ if (pmd_trans_huge(*pmd)) {
ptl = pmd_lock(walk->mm, pmd);
if (!pmd_present(*pmd)) {
spin_unlock(ptl);
return 0;
}
- if (pmd_huge(*pmd)) {
+ if (pmd_trans_huge(*pmd)) {
damon_pmdp_mkold(pmd, walk->mm, addr);
spin_unlock(ptl);
return 0;
@@ -434,14 +434,14 @@ static int damon_young_pmd_entry(pmd_t *pmd, unsigned long addr,
struct damon_young_walk_private *priv = walk->private;
#ifdef CONFIG_TRANSPARENT_HUGEPAGE
- if (pmd_huge(*pmd)) {
+ if (pmd_trans_huge(*pmd)) {
ptl = pmd_lock(walk->mm, pmd);
if (!pmd_present(*pmd)) {
spin_unlock(ptl);
return 0;
}
- if (!pmd_huge(*pmd)) {
+ if (!pmd_trans_huge(*pmd)) {
spin_unlock(ptl);
goto regular_page;
}
--
1.8.3.1
> On Aug 18, 2022, at 15:37, Baolin Wang <[email protected]> wrote:
>
> The pmd_huge() is used to validate if the pmd entry is mapped by a huge
> page, also including the case of non-present (migration or hwpoisoned)
> pmd entry on arm64 or x86 architectures. That means the pmd_pfn() can
> not get the correct pfn number for the non-present pmd entry, which
> will cause damon_get_page() to get an incorrect page struct (also
> may be NULL by pfn_to_online_page()) to make the access statistics
> incorrect.
>
> Moreover it does not make sense that we still waste time to get the
> page of the non-present entry, just treat it as not-accessed and skip it,
> that keeps consistent with non-present pte level entry.
>
> Thus adding a pmd entry present validation to fix above issues.
>
> Signed-off-by: Baolin Wang <[email protected]>
> Reviewed-by: SeongJae Park <[email protected]>
Reviewed-by: Muchun Song <[email protected]>
Thanks.
> On Aug 18, 2022, at 15:37, Baolin Wang <[email protected]> wrote:
>
> The pmd_huge() is usually used to indicate if a pmd level hugetlb,
> however a pmd mapped huge page can only be THP in damon_mkold_pmd_entry()
> or damon_young_pmd_entry(), so replacing pmd_huge() with pmd_trans_huge()
> in this case to make code more readable according to the discussion [1].
>
> [1] https://lore.kernel.org/all/[email protected]/
>
> Signed-off-by: Baolin Wang <[email protected]>
Reviewed-by: Muchun Song <[email protected]>
Thanks.
Hi Baolin,
On Thu, 18 Aug 2022 15:37:44 +0800 Baolin Wang <[email protected]> wrote:
> The pmd_huge() is usually used to indicate if a pmd level hugetlb,
> however a pmd mapped huge page can only be THP in damon_mkold_pmd_entry()
> or damon_young_pmd_entry(), so replacing pmd_huge() with pmd_trans_huge()
> in this case to make code more readable according to the discussion [1].
Thanks to Baolin and Muchun for the discussion and this patch!
>
> [1] https://lore.kernel.org/all/[email protected]/
>
> Signed-off-by: Baolin Wang <[email protected]>
Reviewed-by: SeongJae Park <[email protected]>
Thanks,
SJ
[...]
On Thu, 18 Aug 2022 15:37:43 +0800 Baolin Wang <[email protected]> wrote:
> The pmd_huge() is used to validate if the pmd entry is mapped by a huge
> page, also including the case of non-present (migration or hwpoisoned)
> pmd entry on arm64 or x86 architectures. That means the pmd_pfn() can
> not get the correct pfn number for the non-present pmd entry, which
> will cause damon_get_page() to get an incorrect page struct (also
> may be NULL by pfn_to_online_page()) to make the access statistics
> incorrect.
>
> Moreover it does not make sense that we still waste time to get the
> page of the non-present entry, just treat it as not-accessed and skip it,
> that keeps consistent with non-present pte level entry.
>
> Thus adding a pmd entry present validation to fix above issues.
>
Do we have a Fixes: for this?
What are the user-visible runtime effects of the bug? "make the access
statistics incorrect" is rather vague.
Do we feel that a cc:stable is warranted?
On 8/21/2022 5:17 AM, Andrew Morton wrote:
> On Thu, 18 Aug 2022 15:37:43 +0800 Baolin Wang <[email protected]> wrote:
>
>> The pmd_huge() is used to validate if the pmd entry is mapped by a huge
>> page, also including the case of non-present (migration or hwpoisoned)
>> pmd entry on arm64 or x86 architectures. That means the pmd_pfn() can
>> not get the correct pfn number for the non-present pmd entry, which
>> will cause damon_get_page() to get an incorrect page struct (also
>> may be NULL by pfn_to_online_page()) to make the access statistics
>> incorrect.
>>
>> Moreover it does not make sense that we still waste time to get the
>> page of the non-present entry, just treat it as not-accessed and skip it,
>> that keeps consistent with non-present pte level entry.
>>
>> Thus adding a pmd entry present validation to fix above issues.
>>
>
> Do we have a Fixes: for this?
OK, should be:
Fixes: 3f49584b262c ("mm/damon: implement primitives for the virtual
memory address spaces")
> What are the user-visible runtime effects of the bug? "make the access
> statistics incorrect" is rather vague.
"access statistics incorrect" means that the DAMON may make incorrect
decision according to the incorrect statistics, for example, DAMON may
can not reclaim cold page in time due to this cold page was regarded as
accessed mistakenly if DAMOS_PAGEOUT operation is specified.
> Do we feel that a cc:stable is warranted?
Though this is not a regular case, I think this patch is suitable to be
backported to cover this unusual case. So please help to add a stable
tag when you apply this patch, or please let me know if you want a new
version with adding Fixes and stable tags. Thanks.
On Sun, 21 Aug 2022 13:22:42 +0800 Baolin Wang <[email protected]> wrote:
>
>
> On 8/21/2022 5:17 AM, Andrew Morton wrote:
> > On Thu, 18 Aug 2022 15:37:43 +0800 Baolin Wang <[email protected]> wrote:
> >
> >> The pmd_huge() is used to validate if the pmd entry is mapped by a huge
> >> page, also including the case of non-present (migration or hwpoisoned)
> >> pmd entry on arm64 or x86 architectures. That means the pmd_pfn() can
> >> not get the correct pfn number for the non-present pmd entry, which
> >> will cause damon_get_page() to get an incorrect page struct (also
> >> may be NULL by pfn_to_online_page()) to make the access statistics
> >> incorrect.
> >>
> >> Moreover it does not make sense that we still waste time to get the
> >> page of the non-present entry, just treat it as not-accessed and skip it,
> >> that keeps consistent with non-present pte level entry.
> >>
> >> Thus adding a pmd entry present validation to fix above issues.
> >>
> >
> > Do we have a Fixes: for this?
>
> OK, should be:
> Fixes: 3f49584b262c ("mm/damon: implement primitives for the virtual
> memory address spaces")
>
> > What are the user-visible runtime effects of the bug? "make the access
> > statistics incorrect" is rather vague.
>
> "access statistics incorrect" means that the DAMON may make incorrect
> decision according to the incorrect statistics, for example, DAMON may
> can not reclaim cold page in time due to this cold page was regarded as
> accessed mistakenly if DAMOS_PAGEOUT operation is specified.
>
> > Do we feel that a cc:stable is warranted?
>
> Though this is not a regular case, I think this patch is suitable to be
> backported to cover this unusual case. So please help to add a stable
> tag when you apply this patch, or please let me know if you want a new
> version with adding Fixes and stable tags. Thanks.
Thanks, I took care of all that.