Messages in this thread | | | Subject | Re: [PATCH 0/6] ipc/sem.c: Fix semctl(,,{GETNCNT,GETZCNT}) | From | Davidlohr Bueso <> | Date | Sun, 11 May 2014 19:56:46 -0700 |
| |
On Sat, 2014-05-10 at 12:03 +0200, Manfred Spraul wrote: > Hi all, > > According to the man page of semop(), semzcnt or semncnt are increased > exactly for the operation that couldn't proceed. > > The Linux implementation always tried to be clever and to increase the counters > for all operations that might be the reason why a task sleeps.
... and I hate the fact that we do so on demand, instead of modifying the values when semop is called. This makes the whole semctl calls less accurate, and in fact its mentioned in the code.
> The following patches fix that and make the code conform to the > documentation. > > The series got fairly long, because I also noticed that semzcnt was calculated > incorrectly. > > What do you think?
I'm still going through the changes, sems make my brain hurt. But conceptually they do make sense... and hey, if semctl(GETNCNT,GETZCNT) calls are currently incomplete, then yeah, we should fix it.
> I ran a few test cases, and the semncnt and semzcnt counts now match > the expectation. > > Is anyone aware of an application that uses GETNCNT or GETZCNT?
Given how Oracle uses sysv semaphores I wouldn't be surprised if they make use of these, specially GETNCNT, for something like "get the amount of waiters" as opposed to "are there waiters"... but I'm just speculating here.
I did find that LTP does some calls to GETZNCT, GETNCNT, and these patches do not break those tests. However, they are pretty bogus since they always test for zero. That reminds me, it might be worthwhile adding some more tests in the selftests/ipc dir, we only have some trivial msgq program, for the rest I pretty much rely on LTP for correctness runs.
Thanks, Davidlohr
| |