Re: [PATCH v16] sys_membarrier(): system-wide memory barrier (generic, x86)

[Mathieu Desnoyers <mathieu.desnoyers@efficios.com>]

----- Original Message -----
> On Fri, Apr 17, 2015 at 11:06 AM, Mathieu Desnoyers
> <mathieu.desnoyers@efficios.com> wrote:
> > diff --git a/init/Kconfig b/init/Kconfig
> > index f5dbc6d..89bad6a 100644
> > --- a/init/Kconfig
> > +++ b/init/Kconfig
> > @@ -1559,6 +1559,19 @@ config PCI_QUIRKS
> >           bugs/quirks. Disable this only if your target machine is
> >           unaffected by PCI quirks.
> >
> > +config MEMBARRIER
> > +       bool "Enable membarrier() system call" if EXPERT
> > +       default y
> > +       depends on SMP
> > +       help
> > +         Enable the membarrier() system call that allows issuing memory
> > +         barriers across all running threads, which can be used to
> > distribute
> > +         the cost of user-space memory barriers asymmetrically by
> > transforming
> > +         pairs of memory barriers into pairs consisting of membarrier()
> > and a
> > +         compiler barrier.
> > +
> > +         If unsure, say Y.
> > +
> 
> I understand why this syscall makes sense on SMP only, but you are
> anyways checking num_online_cpus() and returning if it is only one. Is
> this limitation necessary then? How do !SMP systems handle this
> syscall? (I am guessing glibc wrapper?)

For !SMP, this system call is not implemented (returns -ENOSYS).
Userspace libs are expected to query sysconf(_SC_NPROCESSORS_CONF)
and check whether the system supports multiprocessor at all. If
only a single processor is supported by the kernel, then userspace
can skip the calls to sys_membarrier altogether, because they are
not even needed.

Do you think this kind of information belongs in a man page ?

Should we instead just implement the system call in !SMP, and
return 0 without any side-effect ? This would be a bit inefficient
to let userspace call a system call that has no effect whatsoever.

Thanks,

Mathieu

> ...
> 
> > +SYSCALL_DEFINE2(membarrier, int, cmd, int, flags)
> > +{
> > +       switch (cmd) {
> > +       case MEMBARRIER_CMD_QUERY:
> > +               return MEMBARRIER_CMD_BITMASK;
> > +       case MEMBARRIER_CMD_SHARED:
> > +               if (num_online_cpus() > 1)
> > +                       synchronize_sched();
> > +               return 0;
> > +       default:
> > +               return -EINVAL;
> > +       }
> > +}
> 
> 
> 
> --
> Pranith
> 

-- 
Mathieu Desnoyers
EfficiOS Inc.
http://www.efficios.com