Skip to contentSkip to conversation

lockdep fix incorrect percpu usage

PATCH In review Mathieu Desnoyers 4 participants 9 messages Mar 30, 2010 – Apr 22, 2010

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 lockdep fix incorrect percpu usage

Should use per_cpu_ptr() to obfuscate the per cpu pointers (RELOC_HIDE is needed for per cpu pointers).

git blame points to commit:

lockdep.c: commit 8e18257d29238311e82085152741f0c3aa18b74d

But it's really just moving the code around. But it's enough to say that the problems appeared before Jul 19 01:48:54 2007, which brings us back to 2.6.23.

So it should be applied to stable 2.6.23.x to 2.6.33.x (or whichever of these stable branches are still maintained) and to mainline 2.6.34-rc2.

This patch should be queued for the stable branch. (tested on 2.6.33.1 x86_64)

CC Randy Dunlap <[email protected]>
CC Eric Dumazet <[email protected]>
CC Rusty Russell <[email protected]>
CC Peter Zijlstra <[email protected]>
CC Tejun Heo <[email protected]>
CC Ingo Molnar <[email protected]>
CC Andrew Morton <[email protected]>
CC Linus Torvalds <[email protected]>
CC Greg Kroah-Hartman <[email protected]>
CC Steven Rostedt <[email protected]>
CC stable <[email protected]>
---
 kernel/lockdep.c |    6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
kernel/lockdep.c +3 -3
Index: linux-2.6-lttng/kernel/lockdep.c
===================================================================
@@ -609,9 +609,9 @@ static int static_obj(void *obj)
 	 * percpu var?
 	 */
 	for_each_possible_cpu(i) {
-		start = (unsigned long) &__per_cpu_start + per_cpu_offset(i);
-		end   = (unsigned long) &__per_cpu_start + PERCPU_ENOUGH_ROOM
-					+ per_cpu_offset(i);
+		start = (unsigned long) per_cpu_ptr(&__per_cpu_start, i);
+		end   = (unsigned long) per_cpu_ptr(&__per_cpu_start
+						    + PERCPU_ENOUGH_ROOM, i);
 
 		if ((addr >= start) && (addr < end))
 			return 1;
Signature · 4 lines
Mathieu Desnoyers
Operating System Efficiency R&D Consultant
EfficiOS Inc.
http://www.efficios.com
#1 lore raw
Quoted 5 lines · Mathieu Desnoyers

Should use per_cpu_ptr() to obfuscate the per cpu pointers (RELOC_HIDE is needed for per cpu pointers).

git blame points to commit:

Quoted 7 lines

lockdep.c: commit 8e18257d29238311e82085152741f0c3aa18b74d

But it's really just moving the code around. But it's enough to say that the problems appeared before Jul 19 01:48:54 2007, which brings us back to 2.6.23.

So it should be applied to stable 2.6.23.x to 2.6.33.x (or whichever of these stable branches are still maintained) and to mainline 2.6.34-rc2.

well, definately not to mainline, since that code is utterly busted in mainline due to recent per-cpu changes.

#2 lore raw
Quoted 8 lines · * Peter Zijlstra ([email protected]) wrote:
Quoted 7 lines · Mathieu Desnoyers

Should use per_cpu_ptr() to obfuscate the per cpu pointers (RELOC_HIDE is needed for per cpu pointers).

git blame points to commit:

lockdep.c: commit 8e18257d29238311e82085152741f0c3aa18b74d

Quoted 8 lines
Quoted 5 lines

But it's really just moving the code around. But it's enough to say that the problems appeared before Jul 19 01:48:54 2007, which brings us back to 2.6.23.

So it should be applied to stable 2.6.23.x to 2.6.33.x (or whichever of these stable branches are still maintained) and to mainline 2.6.34-rc2.

well, definately not to mainline, since that code is utterly busted in mainline due to recent per-cpu changes.

How recent ? I'm based on

commit f57d4e859a8acd63f878cd0534ec4b990b1710dc
Merge: 0528faa... eed6351...
Author: Ingo Molnar <[email protected]>
Date: Mon Mar 29 18:56:00 2010 +0200

from -tip and I see the problem there, both in module.c and lockdep.c.

Thanks,

Mathieu

Signature · 4 lines
Mathieu Desnoyers
Operating System Efficiency R&D Consultant
EfficiOS Inc.
http://www.efficios.com
#3 lore raw
Quoted 18 lines · Mathieu Desnoyers
Quoted 8 lines · * Peter Zijlstra ([email protected]) wrote:
Quoted 7 lines · Mathieu Desnoyers

Should use per_cpu_ptr() to obfuscate the per cpu pointers (RELOC_HIDE is needed for per cpu pointers).

git blame points to commit:

lockdep.c: commit 8e18257d29238311e82085152741f0c3aa18b74d

Quoted 8 lines
Quoted 5 lines

But it's really just moving the code around. But it's enough to say that the problems appeared before Jul 19 01:48:54 2007, which brings us back to 2.6.23.

So it should be applied to stable 2.6.23.x to 2.6.33.x (or whichever of these stable branches are still maintained) and to mainline 2.6.34-rc2.

well, definately not to mainline, since that code is utterly busted in mainline due to recent per-cpu changes.

Quoted 8 lines

How recent ? I'm based on

commit f57d4e859a8acd63f878cd0534ec4b990b1710dc
Merge: 0528faa... eed6351...
Author: Ingo Molnar <[email protected]>
Date: Mon Mar 29 18:56:00 2010 +0200

from -tip and I see the problem there, both in module.c and lockdep.c.

Yeah, its basically been busted since the early merge window period, hopefully Tejun's patches will make it in soon:

http://lkml.org/lkml/2010/3/10/79

#4 lore raw
Quoted 26 lines · * Peter Zijlstra ([email protected]) wrote:
Quoted 25 lines · Mathieu Desnoyers
Quoted 8 lines · * Peter Zijlstra ([email protected]) wrote:
Quoted 7 lines · Mathieu Desnoyers

Should use per_cpu_ptr() to obfuscate the per cpu pointers (RELOC_HIDE is needed for per cpu pointers).

git blame points to commit:

lockdep.c: commit 8e18257d29238311e82085152741f0c3aa18b74d

Quoted 8 lines
Quoted 5 lines

But it's really just moving the code around. But it's enough to say that the problems appeared before Jul 19 01:48:54 2007, which brings us back to 2.6.23.

So it should be applied to stable 2.6.23.x to 2.6.33.x (or whichever of these stable branches are still maintained) and to mainline 2.6.34-rc2.

well, definately not to mainline, since that code is utterly busted in mainline due to recent per-cpu changes.

How recent ? I'm based on

commit f57d4e859a8acd63f878cd0534ec4b990b1710dc
Merge: 0528faa... eed6351...
Author: Ingo Molnar <[email protected]>
Date: Mon Mar 29 18:56:00 2010 +0200

Quoted 6 lines
Quoted 1 line

from -tip and I see the problem there, both in module.c and lockdep.c.

Yeah, its basically been busted since the early merge window period, hopefully Tejun's patches will make it in soon:

http://lkml.org/lkml/2010/3/10/79

I see. These patches are "on their way" to mainline, so it's better not to create conflicts. So the lockdep patch should only be applied to -stable, but separate module.c patch should apply to both -stable and mainline. Am I correct ?

Thanks,

Mathieu

Quoted 3 lines
Signature · 4 lines
Mathieu Desnoyers
Operating System Efficiency R&D Consultant
EfficiOS Inc.
http://www.efficios.com
#5 lore raw
Quoted 4 lines · Mathieu Desnoyers

I see. These patches are "on their way" to mainline, so it's better not to create conflicts. So the lockdep patch should only be applied to -stable, but separate module.c patch should apply to both -stable and mainline. Am I correct ?

I'll push proper fixes to mainline in a few days. For -stable, yeah, probably.

Thanks.

Signature · 1 line
tejun
#6 lore raw
Quoted 8 lines · Tejun Heo
Quoted 4 lines · Mathieu Desnoyers

I see. These patches are "on their way" to mainline, so it's better not to create conflicts. So the lockdep patch should only be applied to -stable, but separate module.c patch should apply to both -stable and mainline. Am I correct ?

I'll push proper fixes to mainline in a few days. For -stable, yeah, probably.

Ok, did a patch ever end up in Linus's tree for this that I can pull into the -stable releases?

thanks,

greg k-h

#7 lore raw
Quoted 10 lines · * Greg KH ([email protected]) wrote:
Quoted 8 lines · Tejun Heo
Quoted 4 lines · Mathieu Desnoyers

I see. These patches are "on their way" to mainline, so it's better not to create conflicts. So the lockdep patch should only be applied to -stable, but separate module.c patch should apply to both -stable and mainline. Am I correct ?

I'll push proper fixes to mainline in a few days. For -stable, yeah, probably.

Quoted 6 lines

Ok, did a patch ever end up in Linus's tree for this that I can pull into the -stable releases?

thanks,

greg k-h

Hi Greg,

Here is the updated patch, stating which mainline commit from Tejun fixes it by refactoring the code. I'll leave the decision to pick this targeted fix or Tejun refactoring into -stable up to you.

lockdep fix incorrect percpu usage

Should use per_cpu_ptr() to obfuscate the per cpu pointers (RELOC_HIDE is needed for per cpu pointers).

git blame points to commit:

lockdep.c: commit 8e18257d29238311e82085152741f0c3aa18b74d

But it's really just moving the code around. But it's enough to say that the problems appeared before Jul 19 01:48:54 2007, which brings us back to 2.6.23.

It should be applied to stable 2.6.23.x to 2.6.33.x (or whichever of these stable branches are still maintained).

The mainline kernel as of 2.6.34-rc5 is not affected by this problem because commit 10fad5e46f6c7bdfb01b1a012380a38e3c6ab346 fixed it by refactoring.

(tested on 2.6.33.1 x86_64)

CC Randy Dunlap <[email protected]>
CC Eric Dumazet <[email protected]>
CC Rusty Russell <[email protected]>
CC Peter Zijlstra <[email protected]>
CC Tejun Heo <[email protected]>
CC Ingo Molnar <[email protected]>
CC Andrew Morton <[email protected]>
CC Linus Torvalds <[email protected]>
CC Greg Kroah-Hartman <[email protected]>
CC Steven Rostedt <[email protected]>
CC stable <[email protected]>
---
 kernel/lockdep.c |    6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
kernel/lockdep.c +3 -3
Index: linux.trees.git/kernel/lockdep.c
===================================================================
@@ -600,9 +600,9 @@ static int static_obj(void *obj)
 	 * percpu var?
 	 */
 	for_each_possible_cpu(i) {
-		start = (unsigned long) &__per_cpu_start + per_cpu_offset(i);
-		end   = (unsigned long) &__per_cpu_start + PERCPU_ENOUGH_ROOM
-					+ per_cpu_offset(i);
+		start = (unsigned long) per_cpu_ptr(&__per_cpu_start, i);
+		end   = (unsigned long) per_cpu_ptr(&__per_cpu_start
+						    + PERCPU_ENOUGH_ROOM, i);
 
 		if ((addr >= start) && (addr < end))
 			return 1;
Signature · 4 lines
Mathieu Desnoyers
Operating System Efficiency R&D Consultant
EfficiOS Inc.
http://www.efficios.com
#8 lore raw
Quoted 3 lines · Mathieu Desnoyers

Here is the updated patch, stating which mainline commit from Tejun fixes it by refactoring the code. I'll leave the decision to pick this targeted fix or Tejun refactoring into -stable up to you.

Ditto, please take this one.

Thanks.

Signature · 1 line
tejun