Christoph Lameter wrote on Tuesday, March 28, 2006 3:48 PM
> Could we simply define these smb_mb__*_clear_bit to be noops
> and then make the atomic bit ops to have full barriers? That would satisfy
> Nick's objections.
Oh, it also penalize all other 1,055 call site of clear_bit(), though I don't
know how many actually needs memory barrier. I suspect some need "lock"
barrier, some need "unlock" barrier, and of course some needs full fence.
Why not make unlock_buffer use test_and_clear_bit()? Utilizing it's implied
full memory fence and throw away the return value? OK, OK, this is obscured.
Then introduce clear_bit_memory_fence API or some sort.
- Ken
diff -Nurp linux-2.6.16/fs/buffer.c linux-2.6.16.ken/fs/buffer.c
--- linux-2.6.16/fs/buffer.c 2006-03-19 21:53:29.000000000 -0800
+++ linux-2.6.16.ken/fs/buffer.c 2006-03-28 17:20:02.000000000 -0800
@@ -78,8 +78,7 @@ EXPORT_SYMBOL(__lock_buffer);
void fastcall unlock_buffer(struct buffer_head *bh)
{
- clear_buffer_locked(bh);
- smp_mb__after_clear_bit();
+ test_clear_buffer_locked(bh);
wake_up_bit(&bh->b_state, BH_Lock);
}
-
: send the line "unsubscribe linux-ia64" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
|