[PATCH] powerpc: thp: Fix crash on mremap

Subsystems: memory management, memory management - thp (transparent huge page), the rest

STALE4594d

7 messages, 4 authors, 2014-01-14 · open the first message on its own page

[PATCH] powerpc: thp: Fix crash on mremap

From: Aneesh Kumar K.V <hidden>
Date: 2014-01-01 09:54:09 

From: "Aneesh Kumar K.V" <redacted>

This patch fix the below crash

NIP [c00000000004cee4] .__hash_page_thp+0x2a4/0x440
LR [c0000000000439ac] .hash_page+0x18c/0x5e0
...
Call Trace:
[c000000736103c40] [00001ffffb000000] 0x1ffffb000000(unreliable)
[437908.479693] [c000000736103d50] [c0000000000439ac] .hash_page+0x18c/0x5e0
[437908.479699] [c000000736103e30] [c00000000000924c] .do_hash_page+0x4c/0x58

On ppc64 we use the pgtable for storing the hpte slot information and
store address to the pgtable at a constant offset (PTRS_PER_PMD) from
pmd. On mremap, when we switch the pmd, we need to withdraw and deposit
the pgtable again, so that we find the pgtable at PTRS_PER_PMD offset
from new pmd.

We also want to move the withdraw and deposit before the set_pmd so
that, when page fault find the pmd as trans huge we can be sure that
pgtable can be located at the offset.

Signed-off-by: Aneesh Kumar K.V <redacted>
---
NOTE:
For other archs we would just be removing the pgtable from the list and adding it back.
I didn't find an easy way to make it not do that without lots of #ifdef around. Any
suggestion around that is welcome.

 mm/huge_memory.c | 21 ++++++++++-----------
 1 file changed, 10 insertions(+), 11 deletions(-)

diff --git a/mm/huge_memory.c b/mm/huge_memory.c
index 7de1bf85f683..eb2e60d9ba45 100644
--- a/mm/huge_memory.c
+++ b/mm/huge_memory.c
@@ -1500,24 +1500,23 @@ int move_huge_pmd(struct vm_area_struct *vma, struct vm_area_struct *new_vma,
 	 */
 	ret = __pmd_trans_huge_lock(old_pmd, vma, &old_ptl);
 	if (ret == 1) {
+		pgtable_t pgtable;
+
 		new_ptl = pmd_lockptr(mm, new_pmd);
 		if (new_ptl != old_ptl)
 			spin_lock_nested(new_ptl, SINGLE_DEPTH_NESTING);
 		pmd = pmdp_get_and_clear(mm, old_addr, old_pmd);
 		VM_BUG_ON(!pmd_none(*new_pmd));
+		/*
+		 * Archs like ppc64 use pgtable to store per pmd
+		 * specific information. So when we switch the pmd,
+		 * we should also withdraw and deposit the pgtable
+		 */
+		pgtable = pgtable_trans_huge_withdraw(mm, old_pmd);
+		pgtable_trans_huge_deposit(mm, new_pmd, pgtable);
 		set_pmd_at(mm, new_addr, new_pmd, pmd_mksoft_dirty(pmd));
-		if (new_ptl != old_ptl) {
-			pgtable_t pgtable;
-
-			/*
-			 * Move preallocated PTE page table if new_pmd is on
-			 * different PMD page table.
-			 */
-			pgtable = pgtable_trans_huge_withdraw(mm, old_pmd);
-			pgtable_trans_huge_deposit(mm, new_pmd, pgtable);
-
+		if (new_ptl != old_ptl)
 			spin_unlock(new_ptl);
-		}
 		spin_unlock(old_ptl);
 	}
 out:
-- 
1.8.3.2

Re: [PATCH] powerpc: thp: Fix crash on mremap

From: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Date: 2014-01-01 10:29:14 

On Wed, 2014-01-01 at 15:23 +0530, Aneesh Kumar K.V wrote:
From: "Aneesh Kumar K.V" <redacted>

This patch fix the below crash

NIP [c00000000004cee4] .__hash_page_thp+0x2a4/0x440
LR [c0000000000439ac] .hash_page+0x18c/0x5e0
...
Call Trace:
[c000000736103c40] [00001ffffb000000] 0x1ffffb000000(unreliable)
[437908.479693] [c000000736103d50] [c0000000000439ac] .hash_page+0x18c/0x5e0
[437908.479699] [c000000736103e30] [c00000000000924c] .do_hash_page+0x4c/0x58

On ppc64 we use the pgtable for storing the hpte slot information and
store address to the pgtable at a constant offset (PTRS_PER_PMD) from
pmd. On mremap, when we switch the pmd, we need to withdraw and deposit
the pgtable again, so that we find the pgtable at PTRS_PER_PMD offset
from new pmd.

We also want to move the withdraw and deposit before the set_pmd so
that, when page fault find the pmd as trans huge we can be sure that
pgtable can be located at the offset.

Signed-off-by: Aneesh Kumar K.V <redacted>
---
NOTE:
For other archs we would just be removing the pgtable from the list and adding it back.
I didn't find an easy way to make it not do that without lots of #ifdef around. Any
suggestion around that is welcome.
What about

-		if (new_ptl != old_ptl) {
+               if (new_ptl != old_ptl || ARCH_THP_MOVE_PMD_ALWAYS_WITHDRAW) {

Or something similar ?

Cheers,
Ben.
quoted hunk
 mm/huge_memory.c | 21 ++++++++++-----------
 1 file changed, 10 insertions(+), 11 deletions(-)

diff --git a/mm/huge_memory.c b/mm/huge_memory.c
index 7de1bf85f683..eb2e60d9ba45 100644
--- a/mm/huge_memory.c
+++ b/mm/huge_memory.c
@@ -1500,24 +1500,23 @@ int move_huge_pmd(struct vm_area_struct *vma, struct vm_area_struct *new_vma,
 	 */
 	ret = __pmd_trans_huge_lock(old_pmd, vma, &old_ptl);
 	if (ret == 1) {
+		pgtable_t pgtable;
+
 		new_ptl = pmd_lockptr(mm, new_pmd);
 		if (new_ptl != old_ptl)
 			spin_lock_nested(new_ptl, SINGLE_DEPTH_NESTING);
 		pmd = pmdp_get_and_clear(mm, old_addr, old_pmd);
 		VM_BUG_ON(!pmd_none(*new_pmd));
+		/*
+		 * Archs like ppc64 use pgtable to store per pmd
+		 * specific information. So when we switch the pmd,
+		 * we should also withdraw and deposit the pgtable
+		 */
+		pgtable = pgtable_trans_huge_withdraw(mm, old_pmd);
+		pgtable_trans_huge_deposit(mm, new_pmd, pgtable);
 		set_pmd_at(mm, new_addr, new_pmd, pmd_mksoft_dirty(pmd));
-		if (new_ptl != old_ptl) {
-			pgtable_t pgtable;
-
-			/*
-			 * Move preallocated PTE page table if new_pmd is on
-			 * different PMD page table.
-			 */
-			pgtable = pgtable_trans_huge_withdraw(mm, old_pmd);
-			pgtable_trans_huge_deposit(mm, new_pmd, pgtable);
-
+		if (new_ptl != old_ptl)
 			spin_unlock(new_ptl);
-		}
 		spin_unlock(old_ptl);
 	}
 out:

Re: [PATCH] powerpc: thp: Fix crash on mremap

From: Kirill A. Shutemov <hidden>
Date: 2014-01-02 02:19:59 

On Wed, Jan 01, 2014 at 09:29:05PM +1100, Benjamin Herrenschmidt wrote:
On Wed, 2014-01-01 at 15:23 +0530, Aneesh Kumar K.V wrote:
quoted
From: "Aneesh Kumar K.V" <redacted>

This patch fix the below crash

NIP [c00000000004cee4] .__hash_page_thp+0x2a4/0x440
LR [c0000000000439ac] .hash_page+0x18c/0x5e0
...
Call Trace:
[c000000736103c40] [00001ffffb000000] 0x1ffffb000000(unreliable)
[437908.479693] [c000000736103d50] [c0000000000439ac] .hash_page+0x18c/0x5e0
[437908.479699] [c000000736103e30] [c00000000000924c] .do_hash_page+0x4c/0x58

On ppc64 we use the pgtable for storing the hpte slot information and
store address to the pgtable at a constant offset (PTRS_PER_PMD) from
pmd. On mremap, when we switch the pmd, we need to withdraw and deposit
the pgtable again, so that we find the pgtable at PTRS_PER_PMD offset
from new pmd.

We also want to move the withdraw and deposit before the set_pmd so
that, when page fault find the pmd as trans huge we can be sure that
pgtable can be located at the offset.

Signed-off-by: Aneesh Kumar K.V <redacted>
---
NOTE:
For other archs we would just be removing the pgtable from the list and adding it back.
I didn't find an easy way to make it not do that without lots of #ifdef around. Any
suggestion around that is welcome.
What about

-		if (new_ptl != old_ptl) {
+               if (new_ptl != old_ptl || ARCH_THP_MOVE_PMD_ALWAYS_WITHDRAW) {

Or something similar ?
Looks sane to me. Or something with IS_ENABLED(), if needed.

Cheers,
Ben.
quoted
 mm/huge_memory.c | 21 ++++++++++-----------
 1 file changed, 10 insertions(+), 11 deletions(-)

diff --git a/mm/huge_memory.c b/mm/huge_memory.c
index 7de1bf85f683..eb2e60d9ba45 100644
--- a/mm/huge_memory.c
+++ b/mm/huge_memory.c
@@ -1500,24 +1500,23 @@ int move_huge_pmd(struct vm_area_struct *vma, struct vm_area_struct *new_vma,
 	 */
 	ret = __pmd_trans_huge_lock(old_pmd, vma, &old_ptl);
 	if (ret == 1) {
+		pgtable_t pgtable;
+
 		new_ptl = pmd_lockptr(mm, new_pmd);
 		if (new_ptl != old_ptl)
 			spin_lock_nested(new_ptl, SINGLE_DEPTH_NESTING);
 		pmd = pmdp_get_and_clear(mm, old_addr, old_pmd);
 		VM_BUG_ON(!pmd_none(*new_pmd));
+		/*
+		 * Archs like ppc64 use pgtable to store per pmd
+		 * specific information. So when we switch the pmd,
+		 * we should also withdraw and deposit the pgtable
+		 */
+		pgtable = pgtable_trans_huge_withdraw(mm, old_pmd);
+		pgtable_trans_huge_deposit(mm, new_pmd, pgtable);
 		set_pmd_at(mm, new_addr, new_pmd, pmd_mksoft_dirty(pmd));
-		if (new_ptl != old_ptl) {
-			pgtable_t pgtable;
-
-			/*
-			 * Move preallocated PTE page table if new_pmd is on
-			 * different PMD page table.
-			 */
Please don't lose the comment.
quoted
-			pgtable = pgtable_trans_huge_withdraw(mm, old_pmd);
-			pgtable_trans_huge_deposit(mm, new_pmd, pgtable);
-
+		if (new_ptl != old_ptl)
 			spin_unlock(new_ptl);
-		}
 		spin_unlock(old_ptl);
 	}
 out:

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>
-- 
 Kirill A. Shutemov

Re: [PATCH] powerpc: thp: Fix crash on mremap

From: Andrew Morton <akpm@linux-foundation.org>
Date: 2014-01-13 22:17:51 

On Thu, 2 Jan 2014 04:19:51 +0200 "Kirill A. Shutemov" [off-list ref] wrote:

On Wed, Jan 01, 2014 at 09:29:05PM +1100, Benjamin Herrenschmidt wrote:
quoted
On Wed, 2014-01-01 at 15:23 +0530, Aneesh Kumar K.V wrote:
quoted
From: "Aneesh Kumar K.V" <redacted>

This patch fix the below crash

NIP [c00000000004cee4] .__hash_page_thp+0x2a4/0x440
LR [c0000000000439ac] .hash_page+0x18c/0x5e0
...
Call Trace:
[c000000736103c40] [00001ffffb000000] 0x1ffffb000000(unreliable)
[437908.479693] [c000000736103d50] [c0000000000439ac] .hash_page+0x18c/0x5e0
[437908.479699] [c000000736103e30] [c00000000000924c] .do_hash_page+0x4c/0x58

On ppc64 we use the pgtable for storing the hpte slot information and
store address to the pgtable at a constant offset (PTRS_PER_PMD) from
pmd. On mremap, when we switch the pmd, we need to withdraw and deposit
the pgtable again, so that we find the pgtable at PTRS_PER_PMD offset
from new pmd.

We also want to move the withdraw and deposit before the set_pmd so
that, when page fault find the pmd as trans huge we can be sure that
pgtable can be located at the offset.
Did this get fixed?

Re: [PATCH] powerpc: thp: Fix crash on mremap

From: Kirill A. Shutemov <hidden>
Date: 2014-01-13 22:30:36 

On Mon, Jan 13, 2014 at 02:17:48PM -0800, Andrew Morton wrote:
On Thu, 2 Jan 2014 04:19:51 +0200 "Kirill A. Shutemov" [off-list ref] wrote:
quoted
On Wed, Jan 01, 2014 at 09:29:05PM +1100, Benjamin Herrenschmidt wrote:
quoted
On Wed, 2014-01-01 at 15:23 +0530, Aneesh Kumar K.V wrote:
quoted
From: "Aneesh Kumar K.V" <redacted>

This patch fix the below crash

NIP [c00000000004cee4] .__hash_page_thp+0x2a4/0x440
LR [c0000000000439ac] .hash_page+0x18c/0x5e0
...
Call Trace:
[c000000736103c40] [00001ffffb000000] 0x1ffffb000000(unreliable)
[437908.479693] [c000000736103d50] [c0000000000439ac] .hash_page+0x18c/0x5e0
[437908.479699] [c000000736103e30] [c00000000000924c] .do_hash_page+0x4c/0x58

On ppc64 we use the pgtable for storing the hpte slot information and
store address to the pgtable at a constant offset (PTRS_PER_PMD) from
pmd. On mremap, when we switch the pmd, we need to withdraw and deposit
the pgtable again, so that we find the pgtable at PTRS_PER_PMD offset
from new pmd.

We also want to move the withdraw and deposit before the set_pmd so
that, when page fault find the pmd as trans huge we can be sure that
pgtable can be located at the offset.
Did this get fixed?
New version: http://thread.gmane.org/gmane.linux.kernel.mm/111809

-- 
 Kirill A. Shutemov

Re: [PATCH] powerpc: thp: Fix crash on mremap

From: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Date: 2014-01-14 04:13:48 

On Mon, 2014-01-13 at 14:17 -0800, Andrew Morton wrote:

Did this get fixed?
Any chance you can Ack the patch on that thread ?

http://thread.gmane.org/gmane.linux.kernel.mm/111809

So I can put it in powerpc -next with a CC stable ? Or if you tell me
tat Kirill Ack is sufficient then I'll go for it.

Cheers,
Ben.

Re: [PATCH] powerpc: thp: Fix crash on mremap

From: Andrew Morton <akpm@linux-foundation.org>
Date: 2014-01-14 04:30:19 

On Tue, 14 Jan 2014 15:13:30 +1100 Benjamin Herrenschmidt [off-list ref] wrote:

On Mon, 2014-01-13 at 14:17 -0800, Andrew Morton wrote:
quoted
Did this get fixed?
Any chance you can Ack the patch on that thread ?

http://thread.gmane.org/gmane.linux.kernel.mm/111809

So I can put it in powerpc -next with a CC stable ? Or if you tell me
tat Kirill Ack is sufficient then I'll go for it.
yup, it looks OK to me from a non-ppc perspective.  Please proceed as
described.
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help