Re: ftraced and suspend to ram

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

On Friday, 22 of August 2008, Steven Rostedt wrote:
> 
> On Fri, 22 Aug 2008, Rafael J. Wysocki wrote:
> > > > tracing. Certainly all the assembly functions.
> > > 
> > > I'm looking into that now too. Are the functions in arch/x86/power/cpu*.c 
> > > the suspend to ram code?
> > 
> > They contain code executed during suspend to RAM, but such code is also:
> > - in all files under arch/x86/kernel/acpi/
> > - in main.c, console.c under kernel/power
> > - in all files under drivers/acpi/sleep
> > - in drivers/acpi/hardware/hwsleep.c
> > 
> > Generally, ACPI is heavily involved and I'm not the right person to ask which
> > of the ACPI functions should get the 'notrace' thing.  Also, I'm not sure about
> > the device drivers' ->suspend() and ->resume() callbacks, especially for
> > sysdevs and ->suspend_late(), ->resume_early() for platform devices and PCI.
> > 
> > Well, how exactly suspend to RAM is broken by ftrace?
> > 
> 
> I know that the smp_processor_id may be defined in the %fs register, but 
> if ftrace is called before the %fs is set up, it may crash because it 
> uses smp_processor_id.

The wake-up code is the most interesting for you, then.  It's located in
arch/x86/kernel/acpi/ , but on x86-64 it also uses trampoline code in
trampoline_64.S (this is assembly, though).

On x86-64, wakeup_long64 in wakeup_64.S is where we get from the real-mode
wake-up code, but under arch/x86/kernel/acpi/realmode there are some C files
containing functions executed in real mode.  Everything in there and everything
referred to from there should be 'notrace' IMO.

Thanks,
Rafael