Re: System slow down from udev

["Rafael J. Wysocki" <rjw@sisk.pl>]

On Thursday, May 30, 2013 10:49:07 AM Yinghai Lu wrote:
> On Thu, May 30, 2013 at 10:42 AM, Yinghai Lu <yinghai@kernel.org> wrote:
> > On Thu, May 30, 2013 at 9:36 AM, Rafael J. Wysocki <rjw@sisk.pl> wrote:
> >> On Thursday, May 30, 2013 04:45:35 PM Rafael J. Wysocki wrote:
> >>> On Thursday, May 30, 2013 04:34:51 PM Rafael J. Wysocki wrote:
> >> [...]
> >>> >
> >>> > I'll try to dig deeper locally in the meantime.
> >>>
> >>> Actually, I think I know what the problem is, but I need some more time to
> >>> debug it.  Fortunately, I'm able to see some symptoms. :-)
> >>
> >> OK, one fix is appended, but there still is one more bug I need to investigate.
> >>
> >> Please check if this one helps, though.
> >
> > yes, that fix the problem.
> >
> >
> >>
> >> Thanks,
> >> Rafael
> >>
> >>
> >> ---
> >>  drivers/acpi/acpi_processor.c   |    5 +++++
> >>  drivers/acpi/processor_driver.c |    3 ---
> >>  2 files changed, 5 insertions(+), 3 deletions(-)
> >>
> >> Index: linux-pm/drivers/acpi/processor_driver.c
> >> ===================================================================
> >> --- linux-pm.orig/drivers/acpi/processor_driver.c
> >> +++ linux-pm/drivers/acpi/processor_driver.c
> >> @@ -78,9 +78,6 @@ static struct device_driver acpi_process
> >>         .remove = acpi_processor_stop,
> >>  };
> >>
> >> -DEFINE_PER_CPU(struct acpi_processor *, processors);
> >> -EXPORT_PER_CPU_SYMBOL(processors);
> >> -
> >>  static void acpi_processor_notify(acpi_handle handle, u32 event, void *data)
> >>  {
> >>         struct acpi_device *device = data;
> >> Index: linux-pm/drivers/acpi/acpi_processor.c
> >> ===================================================================
> >> --- linux-pm.orig/drivers/acpi/acpi_processor.c
> >> +++ linux-pm/drivers/acpi/acpi_processor.c
> >> @@ -29,6 +29,9 @@
> >>
> >>  ACPI_MODULE_NAME("processor");
> >>
> >> +DEFINE_PER_CPU(struct acpi_processor *, processors);
> >> +EXPORT_PER_CPU_SYMBOL(processors);
> >> +
> >>  /* --------------------------------------------------------------------------
> >>                                  Errata Handling
> >>     -------------------------------------------------------------------------- */
> >> @@ -387,6 +390,7 @@ static int __cpuinit acpi_processor_add(
> >>          * checks.
> >>          */
> >>         per_cpu(processor_device_array, pr->id) = device;
> >> +       per_cpu(processors, pr->id) = pr;
> >>
> >>         dev = get_cpu_device(pr->id);
> >>         result = acpi_bind_one(dev, pr->handle);
> >> @@ -406,6 +410,7 @@ static int __cpuinit acpi_processor_add(
> >>   err:
> >>         free_cpumask_var(pr->throttling.shared_cpu_map);
> >>         device->driver_data = NULL;
> >> +       per_cpu(processors, pr->id) = NULL;
> >>   err_free_pr:
> >>         kfree(pr);
> >>         return result;
> >>
> >
> > do you need to remove
> >     per_cpu(processors, pr->id) = NULL;
> > from processor_driver.c::acpi_processor_stop ?
> 
> also can you kill per_cpu processors ?

Generally, yes, but not right now.

> You should always get it from
> acpi_device from per_cpu(processor_device_array, cpu)
> then driver_data from it.

That's correct, but that needs to be a separate patch and we'll need to track
down all of the users.

Thanks,
Rafael


-- 
I speak only for myself.
Rafael J. Wysocki, Intel Open Source Technology Center.