semaphores and noatomic flag

semaphores and noatomic flag

[Mihai Moise]

I write this to discuss the reasons why the semop system call should
have an IPC_NOATOMIC flag.

Suppose we have two processes, called client and server, which
communicate through a shared memory segment and two semaphores, and need
to synchonize their activities so that they don't operate simultaneously
except at startup.

The server would do,

down(smephore 0)

to wait for a message from the client. When the client needs the server
to execute, it would,

up(semaphore 0)		/* wake up server */
down(semaphore 1)	/* put itself to sleep */

after the server has completed its portion of the task, it would,

up(semaphore 1)		/* wake up client */
down(semaphore 0)	/* put iself to sleep */

The problem is that the two system calls make the whole process twice as
slow as it needs to be, and they are both needed because the semop
system call is implemented in an atomic manner. If the semop system call
had an IPC_NOATOMIC flag, then the each process would only have to do
one call,

semop(up semaphore 0 & down semaphore 1, IPC_NOATOMIC)

which would be interpreted in the kernel as the sequence of two system
calls I have written previously.

I want to know what other people think about this idea.

Mihai

RE: semaphores and noatomic flag

[Anil Kumar]

Will it not be a very specialized case rather than being general call type?

-----Original Message-----
From: linux-kernel-owner@vger.kernel.org
[mailto:linux-kernel-owner@vger.kernel.org]On Behalf Of Mihai Moise
Sent: Monday, June 04, 2001 11:05 PM
To: linux-kernel@vger.kernel.org
Subject: semaphores and noatomic flag


I write this to discuss the reasons why the semop system call should
have an IPC_NOATOMIC flag.

Suppose we have two processes, called client and server, which
communicate through a shared memory segment and two semaphores, and need
to synchonize their activities so that they don't operate simultaneously
except at startup.

The server would do,

down(smephore 0)

to wait for a message from the client. When the client needs the server
to execute, it would,

up(semaphore 0)		/* wake up server */
down(semaphore 1)	/* put itself to sleep */

after the server has completed its portion of the task, it would,

up(semaphore 1)		/* wake up client */
down(semaphore 0)	/* put iself to sleep */

The problem is that the two system calls make the whole process twice as
slow as it needs to be, and they are both needed because the semop
system call is implemented in an atomic manner. If the semop system call
had an IPC_NOATOMIC flag, then the each process would only have to do
one call,

semop(up semaphore 0 & down semaphore 1, IPC_NOATOMIC)

which would be interpreted in the kernel as the sequence of two system
calls I have written previously.

I want to know what other people think about this idea.

Mihai
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Re: semaphores and noatomic flag

[Daniel Phillips]

> -----Original Message-----
> From: linux-kernel-owner@vger.kernel.org
> [mailto:linux-kernel-owner@vger.kernel.org]On Behalf Of Mihai Moise
> 
> up(semaphore 1)		/* wake up client */
> down(semaphore 0)	/* put iself to sleep */
>
> The problem is that the two system calls make the whole process twice
> as slow as it needs to be, and they are both needed because the semop
> system call is implemented in an atomic manner. If the semop system
> call had an IPC_NOATOMIC flag, then the each process would only have
> to do one call,
>
> semop(up semaphore 0 & down semaphore 1, IPC_NOATOMIC)
>
> which would be interpreted in the kernel as the sequence of two
> system calls I have written previously.

On Tuesday 05 June 2001 08:28, Anil Kumar wrote:
> Will it not be a very specialized case rather than being general call
> type?

The concept is general and useful, though I don't think its expression 
here is correct.  A similar feature used in dynix was described earlier 
in some detail by  Paul Cassella:

    http://marc.theaimsgroup.com/?l=linux-kernel&m=97742705300697&w=2
    (Re: [RFC] Semaphores used for daemon wakeup)

The idea is to atomically release one serializer and grab another.  We 
have quite a few flavors of lock so square that and you have the number 
of primitives you'd need for a complete set.

To sell the idea you'd have to come up with a few places where the new 
primitives would make the kernel run better, then you'd have to 
implement it or find someone with the necessary scheduler hacking 
skills to do it for you.  The last part is the tricky one. ;-)

--
We had 

Re: semaphores and noatomic flag

[davide.rossetti]

whould it be possible to use pthread semaphore/mutex/cond_var on
shared-via-mmap() chunks of memory instead ?

regards

-- 
+------------------------------------------------------------------+
|Rossetti Davide   INFN - Sezione Roma I - gruppo V, prog. APEmille|
|                  web    : http://apegate.roma1.infn.it/~rossetti |
|    """""         E-mail : davide.rossetti@roma1.infn.it          |
|    |o o|         phone  : (+39)-06-49914412                      |
|--o00O-O00o--     fax    : (+39)-06-49914423   (+39)-06-4957697   |
|                  address: Dipartimento di Fisica (V.E.)          |
|                           Universita' di Roma "La Sapienza"      |
|                           P.le Aldo Moro,5 I - 00185 Roma - Italy|
|  gnupg pub. key: http://apegate.roma1.infn.it/~rossetti/gnupg.txt|
|								   |
|"Outside of a dog,a book is a man's best friend. Inside of a dog, |
| it's too dark to read." - Groucho Marx                           |
+------------------------------------------------------------------+