On Wed, May 29, 2019 at 07:43:51AM +0200, Dmitry Vyukov wrote:
>
> If fqdir->dead read/write are concurrent, then this still needs to be
> READ_ONCE/WRITE_ONCE. Ordering is orthogonal to atomicity.
No they do not. READ_ONCE/WRITE_ONCE are basically a more fine-tuned
version of barrier(). In this case we already have an implicit
barrier() call due to the memory barrier semantics so this is simply
unnecessary.
It's the same reason you don't need READ_ONCE/WRITE_ONCE when you do:
CPU1 CPU2
---- ----
spin_lock
shared_var = 1 spin_lock
spin_unlock if (shared_var == 1)
...
spin_unlock
Cheers,
--
Email: Herbert Xu <herb...@gondor.apana.org.au>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt
