On Sun, 2013-05-12 at 23:36 -0400, Theodore Ts'o wrote:
> On Sun, May 12, 2013 at 08:11:59PM -0700, Tony Luck wrote:
> >
> > My best guess as to why this commit causes problems is that there are places
> > where updates to individual fields in this structure used to be independent
> > because they were to whole words. Now we have bitfileds there are races
> > between access to different fields in the same word.
>
> Yeah, except we access the fields while holding a lock.... wait a
> minute. We're using bit_spinlocks().... and am I missing something?
>
> Where are the barrier statements to prevent the CPU or the compiler
> from reordering statements around bit_spin_lock()? But if that's the
> problem, I would have expected lots of other things to be broken.
Those use test_and_set_bit(), which per Paul McMemory-Wizard...
ATOMIC OPERATIONS
-----------------
Whilst they are technically interprocessor interaction considerations, atomic
operations are noted specially as some of them imply full memory barriers and
some don't, but they're very heavily relied on as a group throughout the
kernel.
Any atomic operation that modifies some state in memory and returns information
about the state (old or new) implies an SMP-conditional general memory barrier
(smp_mb()) on each side of the actual operation (with the exception of
explicit lock operations, described later). These include:
xchg();
cmpxchg();
atomic_xchg();
atomic_cmpxchg();
atomic_inc_return();
atomic_dec_return();
atomic_add_return();
atomic_sub_return();
atomic_inc_and_test();
atomic_dec_and_test();
atomic_sub_and_test();
atomic_add_negative();
atomic_add_unless(); /* when succeeds (returns 1) */
test_and_set_bit();
test_and_clear_bit();
test_and_change_bit();
--
To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
