CHROMIUM: verity: use atomic_set instead of atomic_cmpxchg in read_completed

CHROMIUM: verity: use atomic_set instead of atomic_cmpxchg in read_completed

When moving from the PENDING to READY state it is sufficient to use
an atomic_set since we know the previous state MUST have been
PENDING. In addition, there is a BUG_ON which verifies this.

BUG=9752
TEST=Ran tests in verity.git. Ran platform_DMVerityCorruption on H/W.

Change-Id: I3e7a2a8cdd909f85e232b77bc8190a7550461688
Signed-off-by: Mandeep Singh Baines <msb@chromium.org>;

R=wad@chromium.org,taysom@chromium.org,ups@chromium.org

Committed: http://chrome-svn/viewvc/chromeos?view=rev&revision=a191aac

[Will Drewry, 2011-03-15 16:26:12]

This just seems to introduce a race in the BUG detection :)

But that's probably okay.  Are we really getting a speed bump from switching to
cmpxchg?

[Will Drewry, 2011-03-15 16:29:36]

Actually, the only concern is if someone else updates that entry with an ERROR,
but in theory that shouldn't be possible.  So,

LGTM

On 2011/03/15 16:26:12, Will Drewry wrote:
> This just seems to introduce a race in the BUG detection :)
> 
> But that's probably okay.  Are we really getting a speed bump from switching
to
> cmpxchg?

[Mandeep Singh Baines, 2011-03-15 17:45:38]

On Tue, Mar 15, 2011 at 9:26 AM,  <wad@chromium.org> wrote:
> This just seems to introduce a race in the BUG detection :)
>
> But that's probably okay.  Are we really getting a speed bump from switching
> to
> cmpxchg?
>

On x86, cmpxchg is a bus locking instruction while atomic_read and
atomic_write are simple loads and stores.

> http://codereview.chromium.org/6677048/
>

[Will Drewry, 2011-03-15 18:50:46]

On Tue, Mar 15, 2011 at 12:45 PM, Mandeep Singh Baines <msb@chromium.org> wrote:
> On Tue, Mar 15, 2011 at 9:26 AM,  <wad@chromium.org> wrote:
>> This just seems to introduce a race in the BUG detection :)
>>
>> But that's probably okay.  Are we really getting a speed bump from switching
>> to
>> cmpxchg?
>>
>
> On x86, cmpxchg is a bus locking instruction while atomic_read and
> atomic_write are simple loads and stores.

Yup - that wasn't the question.  The question is if there was a
tangible benefit :)  I had been erring on the side of over-detecting
errors in case of cascading bugs, but for the sake of simplifying the
code and making it more direct.  It makes sense to clean it up all the
way to match expectations.  At least then the code won't send
confusing messages.  Perhaps just a comment over the "BUG_ON" which
says that "atomic_read" followed by a set is race-prone, but for bug
detection alone, seems practical enough.  Or just submit as is.

[Paul T, 2011-03-15 20:45:40]

LGTM