Re: WARNING: smp_call_function_single() and smp_call_function_mask()

[Arjan van de Ven <arjan@infradead.org>]

On Sun, 02 Dec 2007 09:43:39 +0100
Tomas Carnecky <tom@dbservice.com> wrote:

> WARNING: at arch/x86/kernel/smp_64.c:427 smp_call_function_single()
> WARNING: at arch/x86/kernel/smp_64.c:397 smp_call_function_mask()
> 
> dmesg and config attached.
> 
> I'm getting about three of each at boot. I'm running:
> commit e1cca7e8d484390169777b423a7fe46c7021fec1
> Date:   Thu Nov 29 16:25:29 2007 -0800
> which is the latest git as of yesterday plus a one (unrelated) debug
> statement patch in usb uhci.
> 
> There was a similar bug report after 2.6.23-rc8-mm was released.
> Though there seems to be a fundamental problem with how people use
> smp_call_function*() [1]. And this can just as well be another
> incarnation of it.
> 
> Is that easy enough to fix or do I need to bisect (it didn't happen in
> 2.6.24-rc3)?
> 

this appears to be a bug in the acpi code, to be exact in
processor_throttling.c file, function
acpi_processor_set_throttling_ptc(); it disables interrupts and then
appears to do a cross-cpu IPI to set the state. Well... we can't do
that due to deadlock reasons (you can't do IPI's with interrupts off or
you can get a very nice deadlock with the cpu that you IPI trying to
do the same thing to you).