Re: [PATCH] introduce sys_membarrier(): process-wide memory barrier (v10)

[Josh Triplett <josh@joshtriplett.org>]

On Mon, Apr 05, 2010 at 03:10:57PM -0400, Mathieu Desnoyers wrote:
> * Randy Dunlap (randy.dunlap@oracle.com) wrote:
> > On Mon, 5 Apr 2010 13:57:37 -0400 Mathieu Desnoyers wrote:
> > > +/*
> > > + * sys_membarrier - issue memory barrier on current process running threads
> > > + * @flags: One of these must be set:
> > > + *         MEMBARRIER_EXPEDITED
> > > + *             Adds some overhead, fast execution (few microseconds)
> > > + *         MEMBARRIER_DELAYED
> > > + *             Low overhead, but slow execution (few milliseconds)
> > > + *
> > > + *         MEMBARRIER_QUERY
> > > + *           This optional flag can be set to query if the kernel supports
> > > + *           a set of flags.
> > > + *
> > > + * return values: Returns -EINVAL if the flags are incorrect. Testing for kernel
> > > + * sys_membarrier support can be done by checking for -ENOSYS return value.
> > > + * Return values >= 0 indicate success. For a given set of flags on a given
> > > + * kernel, this system call will always return the same value. It is therefore
> > > + * correct to check the return value only once at library load, passing the
> > 
> > library load assumes caller is a library?  does the kernel care about that?
> 
> Nope, it doesn't. Will rephrase:
> 
>     ... It is therefore
>  * correct to check the return value only once during a process lifetime,
>  * passing the MEMBARRIER_QUERY flag in addition to only check if the flags are
>  * supported, without performing any synchronization.

Technically you can optimize even more than "process lifetime", since as
you said the results hold "For a given set of flags on a given kernel".
So you could check once and use the results as long as you remain on the
same running system.  (Craziness like live process migration and
checkpoint/restart aside. :) )

- Josh Triplett