Re: + work_on_cpu-rewrite-it-to-create-a-kernel-thread-on-demand.patch added to -mm tree

[Rusty Russell <rusty@rustcorp.com.au>]

On Friday 13 February 2009 07:08:36 Eric W. Biederman wrote:
> Andrew Morton <akpm@linux-foundation.org> writes:
> >
> > Series of four patches:
> >
> > - switch cstate.c frmo work_on_cpu to smp_call_function_single()
> >
> > - ditto acpi-cpufreq.c
> >
> > - ditto mce_amd_64.c
> >
> > The final work_on_cpu() caller is pci_call_probe().  I'd like to find a
> > way of removing that callsite as well, so we can finally remove this
> > turkey but for now, just fix the bugs in it:
> 
> As far as I can tell when we are doing the probes we are in a function
> that can sleep, so we should be able to just call set_cpus_allowed to
> remove the need for work_on_cpu in pci_call_probe.  Possibly with a
> save/restore of the allowed cpus.
> 
> Am I missing something?

Yes.  This is a questionable practice when it's a kernel thread (but not
really a problem), but a definite no-no on a real process.  Userspace is
allowed to change affinity on any process at any time; that's why we need
a real method to replace this meme.

(I'm also getting those cpumask's off the stack for core and x86 code, which
is why I'm hitting them all).

Thanks,
Rusty.