Skip to contentSkip to conversation

Panic in reserve_memtype()

DISCUSSION Jack Steiner 2 participants 8 messages Feb 24, 2010 – Mar 1, 2010

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

We see an X86_64 regression that started a few days ago. The kernel is booted via EFI & panics in the pat.c code trying to deref a NULL pointer.

I didn't debug the problem but am suspicious of
x86, pat: Migrate to rbtree only backend for pat memtype management x86/pat
author Pallipadi, Venkatesh <[email protected]>
Wed, 10 Feb 2010 23:26:07 +0000 (15:26 -0800)
committer H. Peter Anvin <[email protected]>
Thu, 18 Feb 2010 23:41:36 +0000 (15:41 -0800)

Has anyone seen this? If not, I can debug further....

Problem is in the git://git.kernel.org/pub/scm/linux/kernel/git/x86/linux-2.6-x86.git tree.

Pid: 0, comm: swapper Not tainted 2.6.33-rc8-tip-medusa+ #2 /

RIP: 0010:[<ffffffff810304b0>]  [<ffffffff810304b0>] rbt_memtype_check_insert+0x1b2/0x232
RSP: 0000:ffffffff81601df8  EFLAGS: 00000256
RAX: 00000000000b0000 RBX: ffff88000f840100 RCX: 00000000000001c1
RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffff88000f8244c0
RBP: ffffffff81601e38 R08: 0000000000000001 R09: ffffffff8152cd79
R10: ffffffff8152cd79 R11: 0000000000018620 R12: ffff88000f8244c0
R13: 0000000000000010 R14: 0000000000000000 R15: 00000000fffffff4
FS:  0000000000000000(0000) GS:ffff880001c00000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 000000008005007b
CR2: 0000000000000000 CR3: 0000000001604000 CR4: 00000000000006b0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 0000000000000000 DR7: 0000000000000000
Process swapper (pid: 0, threadinfo ffffffff81600000, task ffffffff8160c020)
Stack:
 ffffffff81601e38 00000000000b0000 0000000000006000 ffff88000f840100
<0> ffff88000f8244c0 0000000000000000 0000000000000010 00000000fffffff4
<0> ffffffff81601e88 ffffffff8102edff 00000000000b0000 0000000000006000
Call Trace:
 [<ffffffff8102edff>] reserve_memtype+0x2ce/0x4c9
 [<ffffffff8102e0d0>] set_memory_uc+0x41/0x89
 [<ffffffff816b92be>] efi_enter_virtual_mode+0xc9/0x269
 [<ffffffff816aada0>] start_kernel+0x3b8/0x42b
 [<ffffffff816aa140>] ? early_idt_handler+0x0/0x71
 [<ffffffff816aa29e>] x86_64_start_reservations+0xa5/0xa9
 [<ffffffff816aa3ed>] x86_64_start_kernel+0x14b/0x15a

Source of the NULL pointer is:

	int set_memory_uc(unsigned long addr, int numpages)
	{
		...
		ret = reserve_memtype(__pa(addr), __pa(addr) + numpages * PAGE_SIZE,
                            _PAGE_CACHE_UC_MINUS, NULL);

--- jack

#1 lore raw
Quoted 12 lines · Jack Steiner

We see an X86_64 regression that started a few days ago. The kernel is booted via EFI & panics in the pat.c code trying to deref a NULL pointer.

I didn't debug the problem but am suspicious of
x86, pat: Migrate to rbtree only backend for pat memtype management x86/pat
author Pallipadi, Venkatesh <[email protected]>
Wed, 10 Feb 2010 23:26:07 +0000 (15:26 -0800)
committer H. Peter Anvin <[email protected]>
Thu, 18 Feb 2010 23:41:36 +0000 (15:41 -0800)

Quoted 2 lines

Has anyone seen this? If not, I can debug further....

Haven't seen this on my test systems here, but I haven't tested with EFI boot either.

I assume this is repeatable, and you always see this panic. I am looking at the code right now. Can you rollback this particular patch and see whether it goes away?

Thanks,
Venki

#2 lore raw
Quoted 18 lines · Venkatesh
Quoted 12 lines · Jack Steiner

We see an X86_64 regression that started a few days ago. The kernel is booted via EFI & panics in the pat.c code trying to deref a NULL pointer.

I didn't debug the problem but am suspicious of
x86, pat: Migrate to rbtree only backend for pat memtype management x86/pat
author Pallipadi, Venkatesh <[email protected]>
Wed, 10 Feb 2010 23:26:07 +0000 (15:26 -0800)
committer H. Peter Anvin <[email protected]>
Thu, 18 Feb 2010 23:41:36 +0000 (15:41 -0800)

Quoted 2 lines

Has anyone seen this? If not, I can debug further....

Quoted 6 lines

Haven't seen this on my test systems here, but I haven't tested with EFI boot either.

I assume this is repeatable, and you always see this panic. I am looking at the code right now. Can you rollback this particular patch and see whether it goes away?

The problem is very repeatible.

FWIW, we have a nightly regression test that builds/tests the x86 tree everynight at 1 AM. The failure started on the morning of Feb 22.

The build on Feb 21 (& all of Feb before then) passed w/o errors. I can't rule out other errors but I don't see anything else that changed.

The linux-next tree appears to have the same problem.

--- jack

#3 lore raw
Quoted 28 lines · Jack Steiner
Quoted 18 lines · Venkatesh
Quoted 12 lines · Jack Steiner

We see an X86_64 regression that started a few days ago. The kernel is booted via EFI & panics in the pat.c code trying to deref a NULL pointer.

I didn't debug the problem but am suspicious of
x86, pat: Migrate to rbtree only backend for pat memtype management x86/pat
author Pallipadi, Venkatesh <[email protected]>
Wed, 10 Feb 2010 23:26:07 +0000 (15:26 -0800)
committer H. Peter Anvin <[email protected]>
Thu, 18 Feb 2010 23:41:36 +0000 (15:41 -0800)

Quoted 2 lines

Has anyone seen this? If not, I can debug further....

Quoted 6 lines

Haven't seen this on my test systems here, but I haven't tested with EFI boot either.

I assume this is repeatable, and you always see this panic. I am looking at the code right now. Can you rollback this particular patch and see whether it goes away?

The problem is very repeatible.

Quoted 8 lines

FWIW, we have a nightly regression test that builds/tests the x86 tree everynight at 1 AM. The failure started on the morning of Feb 22.

The build on Feb 21 (& all of Feb before then) passed w/o errors. I can't rule out other errors but I don't see anything else that changed.

The linux-next tree appears to have the same problem.

I guess I found an obvious problem in the code. Can you check whether the below patch resolves the panic you are seeing.

Thanks,
Venki

new->type should only change when there is a valid ret_type. Otherwise requested type and return type should be same.

---
 arch/x86/mm/pat_rbtree.c |    4 +++-
 1 files changed, 3 insertions(+), 1 deletions(-)
arch/x86/mm/pat_rbtree.c +3 -1
index e4cd229..58b6de1 100644
@@ -223,7 +223,9 @@ int rbt_memtype_check_insert(struct memtype *new, unsigned long *ret_type)
 						new->type, ret_type);
 
 	if (!err) {
-		new->type = *ret_type;
+		if (ret_type)
+			new->type = *ret_type;
+
 		memtype_rb_insert(&memtype_rbroot, new);
 	}
 	return err;
Signature · 1 line
1.6.0.6
#4 lore raw
Quoted 6 lines

I guess I found an obvious problem in the code. Can you check whether the below patch resolves the panic you are seeing.

Thanks,
Venki

Works great!! Thanks...

Quoted 27 lines

new->type should only change when there is a valid ret_type. Otherwise requested type and return type should be same.

---
 arch/x86/mm/pat_rbtree.c |    4 +++-
 1 files changed, 3 insertions(+), 1 deletions(-)
arch/x86/mm/pat_rbtree.c +3 -1
index e4cd229..58b6de1 100644
@@ -223,7 +223,9 @@ int rbt_memtype_check_insert(struct memtype *new, unsigned long *ret_type)
 						new->type, ret_type);
 
 	if (!err) {
-		new->type = *ret_type;
+		if (ret_type)
+			new->type = *ret_type;
+
 		memtype_rb_insert(&memtype_rbroot, new);
 	}
 	return err;
Signature · 1 line
1.6.0.6
#5 lore raw
Quoted 9 lines · Jack Steiner
Quoted 6 lines

I guess I found an obvious problem in the code. Can you check whether the below patch resolves the panic you are seeing.

Thanks,
Venki

Works great!! Thanks...

Venki

Were you planning to push this fix upstream? Both the x86 & linux-next tree currently fail to boot on EFI-enabled systems because of this bug.

Thanks...

--- jack

Quoted 29 lines
Quoted 29 lines

new->type should only change when there is a valid ret_type. Otherwise requested type and return type should be same.

---
 arch/x86/mm/pat_rbtree.c |    4 +++-
 1 files changed, 3 insertions(+), 1 deletions(-)
arch/x86/mm/pat_rbtree.c +3 -1
index e4cd229..58b6de1 100644
@@ -223,7 +223,9 @@ int rbt_memtype_check_insert(struct memtype *new, unsigned long *ret_type)
 						new->type, ret_type);
 
 	if (!err) {
-		new->type = *ret_type;
+		if (ret_type)
+			new->type = *ret_type;
+
 		memtype_rb_insert(&memtype_rbroot, new);
 	}
 	return err;
Signature · 1 line
1.6.0.6
#6 lore raw

Adding Peter.

Can you push this patch along.

Thanks
Venki

Quoted 51 lines · Jack Steiner
Quoted 9 lines · Jack Steiner
Quoted 6 lines

I guess I found an obvious problem in the code. Can you check whether the below patch resolves the panic you are seeing.

Thanks,
Venki

Works great!! Thanks...

Venki

Were you planning to push this fix upstream? Both the x86 & linux-next tree currently fail to boot on EFI-enabled systems because of this bug.

Thanks...

--- jack

Quoted 29 lines
Quoted 29 lines

new->type should only change when there is a valid ret_type. Otherwise requested type and return type should be same.

---
 arch/x86/mm/pat_rbtree.c |    4 +++-
 1 files changed, 3 insertions(+), 1 deletions(-)
arch/x86/mm/pat_rbtree.c +3 -1
index e4cd229..58b6de1 100644
@@ -223,7 +223,9 @@ int rbt_memtype_check_insert(struct memtype *new, unsigned long *ret_type)
 						new->type, ret_type);
 
 	if (!err) {
-		new->type = *ret_type;
+		if (ret_type)
+			new->type = *ret_type;
+
 		memtype_rb_insert(&memtype_rbroot, new);
 	}
 	return err;
Signature · 1 line
1.6.0.6
#7 lore raw

[tip:x86/pat] x86, pat: In rbt_memtype_check_insert(), update new->type only if valid

In reply to Pallipadi, Venkatesh
Commit-ID:  4daa2a8093ecd1148270a1fc64e99f072b8c2901
Gitweb:     http://git.kernel.org/tip/4daa2a8093ecd1148270a1fc64e99f072b8c2901
Author:     Pallipadi, Venkatesh <[email protected]>
AuthorDate: Wed, 24 Feb 2010 13:43:55 -0800
Committer:  H. Peter Anvin <[email protected]>
CommitDate: Mon, 1 Mar 2010 14:28:48 -0800

x86, pat: In rbt_memtype_check_insert(), update new->type only if valid

new->type should only change when there is a valid ret_type. Otherwise the requested type and return type should be same.

LKML-Reference: <[email protected]>

---
 arch/x86/mm/pat_rbtree.c |    4 +++-
 1 files changed, 3 insertions(+), 1 deletions(-)
arch/x86/mm/pat_rbtree.c +3 -1
index 9063f40..07de4cb 100644
@@ -223,7 +223,9 @@ int rbt_memtype_check_insert(struct memtype *new, unsigned long *ret_type)
 						new->type, ret_type);
 
 	if (!err) {
-		new->type = *ret_type;
+		if (ret_type)
+			new->type = *ret_type;
+
 		memtype_rb_insert(&memtype_rbroot, new);
 	}
 	return err;