Hi ManfredNo. The original goal was an optimization:
On Sun, 22 Sep 2013 12:42:05 +0200 from manfred@xxxxxxxxxxxxxxxx wrote:Hi all,Not all perform_atomic_semop() can cover the points of do_smart_update()
On 09/22/2013 10:26 AM, Mike Galbraith wrote:On Sun, 2013-09-22 at 10:17 +0200, Mike Galbraith wrote:Good catch:On Sun, 2013-09-22 at 10:11 +0800, Jia He wrote:In commit 0a2b9d4c,the update of semaphore's sem_otime(last semop time)
was removed because he wanted to move setting sem->sem_otime to one
place. But after that, the initial semop() will not set the otime
because its sem_op value is 0(in semtimedop,will not change
otime if alter == 1).
the error case:
process_a(server) process_b(client)
semget()
semctl(SETVAL)
semop()
semget()
setctl(IP_STAT)
for(;;) { <--not successful here
check until sem_otime > 0
}
Since commit 0a2b9d4c, wait-for-zero semops do not update sem_otime anymore.
Let's reverse that part of my commit and move the update of sem_otime back
into perform_atomic_semop().
Jia: If perform_atomic_semop() updates sem_otime, then the update in
do_smart_update() is not necessary anymore.
Thus the whole logic with passing arround "semop_completed" can be removed, too.
Are you interested in writing that patch?
after I move the parts of updating otime.
Eg. in semctl_setval/exit_sem/etc. That is, it seems I need to write some
other codes to update sem_otime outside perform_atomic_semop(). That
still violate your original goal---let one function do all the update otime
things.
IMO, what if just remove the condition alter in sys_semtimedop:do_smart_update() can be expensive - thus it shouldn't be called if we didn't change anything.
- if (alter && error == 0)
+ if (error == 0)
do_smart_update(sma, sops, nsops, 1, &tasks);
In old codes, it would set the otime even when sem_op == 0