Re: [PATCH 2/3] work_on_cpu: Use our own workqueue.

[Andrew Morton <akpm@linux-foundation.org>]

On Wed, 4 Feb 2009 13:14:31 +1030 Rusty Russell <rusty@rustcorp.com.au> wrote:

> On Tuesday 03 February 2009 14:36:49 Andrew Morton wrote:
> > But as you say, rdmsr_on_cpu() is easy to do, using
> > smp_call_function_single().  Then we can easily convert all other
> > work_on_cpu() callers to smp_call_function_single() and zap
> > work_on_cpu().  The best outcome, methinks.
> 
> grepping for set_cpus_allowed reveals the following users in the
> first third (when I got tired of it):
> 
> File                         Function            Context      Calls
> powerpc/kernel/smp.c         smp_cpus_done       kernel_init smp_ops->setup_cpu()
> x86/.../speedstep-ich.c      _speedstep_get      cpufreq     speedstep_get_processor_frequency()
> x86/.../powernow-k8.c        check_supported_cpu cpufreq     cpuid_*()
> x86/.../powernow-k8.c        powernowk8_target   cpufreq     mutex_lock, etc
> x86/.../powernow-k8.c        powernowk8_cpu_init boot/hotplg rdmsr,wrmsr
> x86/.../powernow-k8.c        powernowk8_get      cpufreq     rdmsr
> x86/.../speedstep-centrino.c get_cur_freq        cpufreq     rdmsr
> x86/.../speedstep-centrino.c centrino_target     cpufreq     rdmsr,wrmsr
> x86/kernel/reboot.c          nativ..._shutdown   sys_reboot  misc
> x86/.../microcode_core.c     do_microcode_update sys_write   microcode_ops->request_microcode_user
> x86/.../microcode_core.c     reload_store        sys_write   mutex_lock, etc
> 
> Some of these (reboot) we don't care what they do to cpumask.  Some are
> poor style, but not critical (boot time).  Some can use smp_call_function.
> 
> But there are some which should *not* be frobbing their cpus_allowed, and
> really do seem to want a user context.
> 
> I think you're right though: smp_call_function_single (or neat wrappers)
> where possible, work_on_cpu which can fail for the others, and we'll just
> have to plumb in the error returns.

I bet a lot of those can use plain old schedule_work_on().

Could it be that after converting things to
smp_call_function_single()/schedule_work_on(), we don't need
work_on_cpu()?