Skip to contentSkip to conversation

CONFIG_SLAB_DEBUG and NUMA API

DISCUSSION Andi Kleen 2 participants 2 messages Jul 6, 2004 – Jul 7, 2004 2 messages not in the archive

lore
Share & tools
lore

Andi Kleen

author
#2 lore raw
In reply to a message not in the archive
Quoted 8 lines · Andrew Morton
Quoted 5 lines · Andi Kleen

I tested 2.6.7-mm6 with NUMA on with CONFIG_SLAB_DEBUG and I didn't see any oopses. Do you have a recipe to reproduce them?

Still happens here. Booting SLES9.1 with the attached config.

[...]

Here's a patch. The problem was that the kernel exit would allocate memory to send exit signals after the local mempolicy was already freed, but not zeroed. When the allocator tried to grab more memory it would fall over.

-Andi

-------------------------------------------------------------

Move the memory policy freeing to later in exit to make sure the last memory allocations don't use an uninitialized policy

kernel/exit.c renamed from kernel/exit.c-o +4 -3
diff -u linux-2.6.7-mm6/kernel/exit.c-o linux-2.6.7-mm6/kernel/exit.c
@@ -828,9 +828,6 @@
 	__exit_fs(tsk);
 	exit_namespace(tsk);
 	exit_thread();
-#ifdef CONFIG_NUMA
-	mpol_free(tsk->mempolicy);
-#endif
 
 	if (tsk->signal->leader)
 		disassociate_ctty(1);

@@ -841,6 +838,10 @@
 
 	tsk->exit_code = code;
 	exit_notify(tsk);
+#ifdef CONFIG_NUMA
+	mpol_free(tsk->mempolicy);
+	tsk->mempolicy = NULL;
+#endif
 	schedule();
 	BUG();
 	/* Avoid "noreturn function does return".  */
#3 lore raw
Quoted 14 lines · Andi Kleen
Quoted 5 lines · Andrew Morton
Quoted 2 lines · Andi Kleen

I tested 2.6.7-mm6 with NUMA on with CONFIG_SLAB_DEBUG and I didn't see any oopses. Do you have a recipe to reproduce them?

Still happens here. Booting SLES9.1 with the attached config.

[...]

Here's a patch. The problem was that the kernel exit would allocate memory to send exit signals after the local mempolicy was already freed, but not zeroed. When the allocator tried to grab more memory it would fall over.

Quoted 6 lines

-Andi

-------------------------------------------------------------

Move the memory policy freeing to later in exit to make sure the last memory allocations don't use an uninitialized policy

It fixed the problems that I had reported on both 2.6.7-mm6 and 2.6.7-bk18. Thanks a lot,

rjw

Signature · 5 lines
Rafael J. Wysocki
----------------------------
For a successful technology, reality must take precedence over public 
relations, for nature cannot be fooled.
					-- Richard P. Feynman