Re: [PATCH] ipc/sem.c: fix update sem_otime when calling sem_op in semaphore initialization

[Jia He <jiakernel@gmail.com>]

 

On Sun, 22 Sep 2013 12:00:21 +0200 from bitbucket@online.de wrote:
> On Sun, 2013-09-22 at 17:34 +0800, Jia He wrote: 
>> Thanks for the comments, but pls add my email as "from jiakernel@gmail.com"
>> if you have a better implementation.U know, it is my first kernel patch, maybe
>> will give me a brilliant memory in the future :)
> You can have the blame if you like :)
>
>> Anyway, your implementation looks not correct to me. Because from "man semop"
>> sem_otime will record the last sem operation time of semop. If you change the
>> otime in semget(), it changes the meanings in stardard, doesn't it?
> A Linux kernel doing a semop in 1970 would be a kinda neat trick :)
I will try to make it more clear
comes to my test case again:

process_a(server)       process_b(client)
semget()                                  <-seems you choose to set it here
---------------  <1>  --------------------
semctl(SETVAL)
semop()
                        semget()
                        setctl(IP_STAT)
                        for(;;) {
                          check until sem_otime > 0
                        }
And assume that schedule() happenes at <1>, then sem_otime will >0
in process_b's for(;;), but at that time, the process_a's semctl() 
hasn't been called yet.

>
> -Mike
>
> .
>