Re: [PATCH] remove LOCK_SECTION from x86_64 spin_lock asm

[Ingo Molnar]

* Andi Kleen <ak@xxxxxxx> wrote:

> Something is mixed up here:
> 
> The whole problem only happens on kernels using frame pointer. I never
> saw it, simply because I don't use frame pointers.
>
> On a frame pointer less kernel profiling works just fine, and with
> this fix it should work the same on a FP kernel.

it only works on pointer less kernels because the spinlock profile 
unwinding is _conditional_ on an FP kernel right now:

 #if defined(CONFIG_SMP) && defined(CONFIG_FRAME_POINTER)
 unsigned long profile_pc(struct pt_regs *regs)
 {
 ...

on non-FP kernels you'll see all the overhead in the single spin_lock()
function, agreed?

> > in this respect - it might work if you can detect for sure at build time
> > whether there's any local variable. Tricks like this really tend to
> > haunt us later.
> 
> There are already lots of such assumptions in the kernel (e.g. WCHAN
> and others).  I don't think adding one more is a big issue.

wchan has only one assumption: that that all __sched section functions
have a valid frame pointer. This is not unrobust at all. (it is
nonperformant though on register-starved platforms and having proper
unwind would fix wchan too.) In fact ->real_pc could be used by __sched
functions as well, it would likely be cheaper (on x86) than having to
compile with -fno-omit-frame-pointers.

	Ingo
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/