Messages in this thread Patch in this message | ![/](/images/icornerl.gif) | | Date | Mon, 27 Mar 2006 19:59:02 -0800 (PST) | From | Christoph Lameter <> | Subject | Fix unlock_buffer() to work the same way as bit_unlock() |
| |
Currently unlock_buffer() contains a smb_mb__after_clear_bit() which is weird because bit_spin_unlock() uses smb_mb__before_clear_bit():
From include/linux/bit_spinlock.h:
static inline void bit_spin_unlock(int bitnum, unsigned long *addr) { smp_mb__before_clear_bit(); clear_bit(bitnum, addr); preempt_enable(); __release(bitlock); }
For most architectures there is no difference because both smp_mb__after_clear_bit() and smp_mb__before_clear_bit() are both memory barriers and clear_buffer_locked() is an atomic operation. However, they differ under IA64.
Note that this potential race has never been seen under IA64. It was discovered by inspection by Zoltan Menyhart <Zoltan.Menyhart@free.fr>.
Regardless if this is a true race or not, I think the unlock sequence needs to be the same for bit locks and unlock_buffer(). Maybe unlock_buffer and lock_buffer better use bit spinlock operations?
Change unlock_buffer() to work the same way as bit_spin_unlock.
Signed-off-by: Christoph Lameter <clameter@sgi.com>
Index: linux-2.6/fs/buffer.c =================================================================== --- linux-2.6.orig/fs/buffer.c 2006-03-27 14:09:54.000000000 -0800 +++ linux-2.6/fs/buffer.c 2006-03-27 19:40:32.000000000 -0800 @@ -78,8 +78,8 @@ EXPORT_SYMBOL(__lock_buffer); void fastcall unlock_buffer(struct buffer_head *bh) { + smp_mb__before_clear_bit(); clear_buffer_locked(bh); - smp_mb__after_clear_bit(); wake_up_bit(&bh->b_state, BH_Lock); } - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
| ![\](/images/icornerr.gif) |