On Tue, Jun 26, 2012 at 10:34:00PM +0200, Andi Kleen wrote:
> > How is the compiler meant to optimise away "cond" if it's a function
> > call?
>
> Inlines can be optimized away. These tests are usually inlines.
>
> > What did I miss? If nothing, then I will revert this particular change
> > and Rafael will need to be sure his patch is not using VM_BUG_ON to call
> > a function with side-effects.
>
> Do you have an example where the code is actually different,
> or are you just speculating?
>
> FWIW for my config both generates the same code:
>
> size vmlinux-andi-vmbug vmlinux-vmbug-nothing
> text data bss dec hex filename
> 11809704 1457352 1159168 14426224 dc2070 vmlinux-andi-vmbug
> 11809704 1457352 1159168 14426224 dc2070 vmlinux-vmbug-nothing
>
They are the same size because CONFIG_DEBUG_VM and !CONFIG_DEBUG_VM
generate the same code! I applied the patch below again 3.4 and got the
following results
text data bss dec hex filename
6918617 1795640 2260992 10975249 a77811 vmlinux-default-no-vmdebug
6916633 1795640 2260992 10973265 a77051 vmlinux-patched-no-vmdebug
That's almost 2K of text!
I see now that in 3.5 this was already spotted and fixed by Konstantin
in commit [02602a18: bug: completely remove code generated by disabled
VM_BUG_ON()]. That patch restores the rule that VM_BUG_ON() cannot call
anything with side-effects. So Rafael, watch your use of VM_BUG_ON or
you'll find that the your patches work in 3.4 and leak in 3.5.
diff --git a/include/linux/mmdebug.h b/include/linux/mmdebug.h
index c04ecfe..ee24ef8 100644
--- a/include/linux/mmdebug.h
+++ b/include/linux/mmdebug.h
@@ -4,7 +4,7 @@
#ifdef CONFIG_DEBUG_VM
#define VM_BUG_ON(cond) BUG_ON(cond)
#else
-#define VM_BUG_ON(cond) do { (void)(cond); } while (0)
+#define VM_BUG_ON(cond) do { } while (0)
#endif
#ifdef CONFIG_DEBUG_VIRTUAL
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxx. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>