Re: [PATCH 2/2] lockdep: check that no locks held at freeze time

[Ingo Molnar <mingo-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org>]

* Pavel Machek <pavel-+ZI9xUNit7I@public.gmane.org> wrote:

> Hi!
> 
> > >> >> --- a/kernel/exit.c
> > >> >> +++ b/kernel/exit.c
> > >> >> @@ -835,7 +835,7 @@ void do_exit(long code)
> > >> >>       /*
> > >> >>        * Make sure we are holding no locks:
> > >> >>        */
> > >> >> -     debug_check_no_locks_held(tsk);
> > >> >> +     debug_check_no_locks_held();
> > >> >
> > >> > Is task guaranteed == current?
> > >>
> > >> Yes, the first line of do_exit is:
> > >>         struct task_struct *tsk = current;
> > >
> > > Aha, I understand it now.
> > >
> > > Accessing current is slower than local variable. So your "new" code
> > > will work but will be slower. Please revert this part.
> > 
> > Using current instead of passing in tsk was done at Andrew Morton's
> > suggestion, and makes no difference from the freezer's perspective
> > since it would have to use current to get the task to pass in, so I'm
> > going to leave it as is.
> 
> Well, current is:
> 
> static inline struct thread_info *current_thread_info(void)
> {
>         register unsigned long sp asm ("sp");
>         return (struct thread_info *)(sp & ~(THREAD_SIZE - 1));
> }

That's the old 32-bit x86 trick to compute 'current' from the kernel stack 
pointer.

It can be done better - for example on platforms with optimized percpu 
variables (x86-64) it looks like this:

static inline struct thread_info *current_thread_info(void)
{
        struct thread_info *ti;
        ti = (void *)(this_cpu_read_stable(kernel_stack) +
                      KERNEL_STACK_OFFSET - THREAD_SIZE);
        return ti;
}

Which turns the computation of 'current' into a single instruction. For 
example, to access current->pid [which fields is at offset 0x2a4], we get:

    3ad0:       65 48 8b 04 25 00 00    mov    %gs:0x0,%rax
    3ad7:       00 00
    3ad9:       8b 80 a4 02 00 00       mov    0x2a4(%rax),%eax

I also agree with the removal of the 'tsk' parameter because the function 
itself internally assumes that tsk == current.

[ We could perhaps rename the function to 
  debug_check_no_locks_held_curr(), to make it clear it operates on the 
  current task. ]

Thanks,

	Ingo
--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html