2007-08-17 19:34:55

by Oleg Nesterov

Subject: [PATCH,RESEND] documentation: atomic_add_unless() doesn't imply mb() on failure

(the explicit ack/nack from maintainers is wanted).

A "typical" implementation of atomic_add_unless() can return 0 immediately
after the first atomic_read() (before doing cmpxchg). In that case it doesn't
provide any barrier semantics. See include/asm-ia64/atomic.h as an example.

We should either change the implementation, or fix the docs.

Signed-off-by: Oleg Nesterov <[email protected]>

Documentation/atomic_ops.txt | 3 ++-
Documentation/memory-barriers.txt | 2 +-
2 files changed, 3 insertions(+), 2 deletions(-)

diff -puN Documentation/atomic_ops.txt~doc-atomic_add_unless-doesnt-imply-mb-on-failure Documentation/atomic_ops.txt
--- a/Documentation/atomic_ops.txt~doc-atomic_add_unless-doesnt-imply-mb-on-failure
+++ a/Documentation/atomic_ops.txt
@@ -137,7 +137,8 @@ If the atomic value v is not equal to u,
returns non zero. If v is equal to u then it returns zero. This is done as
an atomic operation.

-atomic_add_unless requires explicit memory barriers around the operation.
+atomic_add_unless requires explicit memory barriers around the operation
+unless it fails (returns 0).

atomic_inc_not_zero, equivalent to atomic_add_unless(v, 1, 0)

diff -puN Documentation/memory-barriers.txt~doc-atomic_add_unless-doesnt-imply-mb-on-failure Documentation/memory-barriers.txt
--- a/Documentation/memory-barriers.txt~doc-atomic_add_unless-doesnt-imply-mb-on-failure
+++ a/Documentation/memory-barriers.txt
@@ -1492,7 +1492,7 @@ about the state (old or new) implies an
atomic_dec_and_test();
atomic_sub_and_test();
atomic_add_negative();
- atomic_add_unless();
+ atomic_add_unless(); /* when succeeds (returns 1) */
test_and_set_bit();
test_and_clear_bit();
test_and_change_bit();
_

2007-08-23 06:50:26

by Nick Piggin

Subject: Re: [PATCH,RESEND] documentation: atomic_add_unless() doesn't imply mb() on failure

Oleg Nesterov wrote:
> (the explicit ack/nack from maintainers is wanted).
>
> A "typical" implementation of atomic_add_unless() can return 0 immediately
> after the first atomic_read() (before doing cmpxchg). In that case it doesn't
> provide any barrier semantics. See include/asm-ia64/atomic.h as an example.
>
> We should either change the implementation, or fix the docs.

Did this end up getting merged? If not, it should, thanks.

Acked-by: Nick Piggin <[email protected]>

>
> Signed-off-by: Oleg Nesterov <[email protected]>
>
> Documentation/atomic_ops.txt | 3 ++-
> Documentation/memory-barriers.txt | 2 +-
> 2 files changed, 3 insertions(+), 2 deletions(-)
>
> diff -puN Documentation/atomic_ops.txt~doc-atomic_add_unless-doesnt-imply-mb-on-failure Documentation/atomic_ops.txt
> --- a/Documentation/atomic_ops.txt~doc-atomic_add_unless-doesnt-imply-mb-on-failure
> +++ a/Documentation/atomic_ops.txt
> @@ -137,7 +137,8 @@ If the atomic value v is not equal to u,
> returns non zero. If v is equal to u then it returns zero. This is done as
> an atomic operation.
>
> -atomic_add_unless requires explicit memory barriers around the operation.
> +atomic_add_unless requires explicit memory barriers around the operation
> +unless it fails (returns 0).
>
> atomic_inc_not_zero, equivalent to atomic_add_unless(v, 1, 0)
>
> diff -puN Documentation/memory-barriers.txt~doc-atomic_add_unless-doesnt-imply-mb-on-failure Documentation/memory-barriers.txt
> --- a/Documentation/memory-barriers.txt~doc-atomic_add_unless-doesnt-imply-mb-on-failure
> +++ a/Documentation/memory-barriers.txt
> @@ -1492,7 +1492,7 @@ about the state (old or new) implies an
> atomic_dec_and_test();
> atomic_sub_and_test();
> atomic_add_negative();
> - atomic_add_unless();
> + atomic_add_unless(); /* when succeeds (returns 1) */
> test_and_set_bit();
> test_and_clear_bit();
> test_and_change_bit();
> _
>
>


--
SUSE Labs, Novell Inc.