[CRIU] [PATCH 5/5] x86: atomic -- Adopt code from linux kernel

[Cyrill Gorcunov]

On Fri, Feb 15, 2013 at 05:54:01PM +0400, Pavel Emelyanov wrote:
> On 02/15/2013 05:48 PM, Cyrill Gorcunov wrote:
> > On Fri, Feb 15, 2013 at 05:28:42PM +0400, Pavel Emelyanov wrote:
> >> On 02/06/2013 12:04 AM, Cyrill Gorcunov wrote:
> >>>
> >>> This allow us to eliminate "asm/types.h" inclusion as well,
> >>> together with simplifying the code.
> >>>
> >>> Signed-off-by: Cyrill Gorcunov <gorcunov at openvz.org>
> >>> ---
> >>>  arch/x86/include/asm/atomic.h | 92 +++++++++++++++++++------------------------
> >>>  include/lock.h                |  3 +-
> >>>  2 files changed, 43 insertions(+), 52 deletions(-)
> >>>
> >>
> >>
> >>> @@ -119,7 +119,8 @@ static void inline mutex_lock(mutex_t *m)
> >>>  	u32 c;
> >>>  	int ret;
> >>>  
> >>> -	while ((c = atomic_inc(&m->raw))) {
> >>> +	while ((c = atomic_read(&m->raw))) {
> >>> +		atomic_inc(&m->raw);
> >>
> >> This is not equal change. It was while (atomic_get_and_inc) now its while (atomic_get) atomic_inc.
> > 
> > Which actually doesn't change much. Look, atomic_inc was incrementing value and
> > returning its previous meaning, now we do the same but this looks as "two"
> > operations from C code, but the cycle below is constructed to be lockless so you will
> > not get any harm in such transition. I rather feel sorry that I wrote it in
> > "lock xchg" manner earlier.
> 
> Cyrill,
> 
> if (atomic_get_and_inc())
> 
> and
> 
> if (atomic_get())
> 	atomic_inc();
> 
> _do_ _work_ _differently_.

I'll update the patch, thanks for review!