Re: [PATCH 10/17] 2.6.17.1 perfmon2 patch for review: PMU context switch

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Stephane Eranian <[email protected]> writes:

> Andi,
> 
> Thanks for your feedback. I will make the changes you
> requested.
> 
> About the context switch code, what about I do the following
> in __switch_to():
> 
> __kprobes struct task_struct *
> __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
> {
>         struct thread_struct *prev = &prev_p->thread,
>                                  *next = &next_p->thread;
>         int cpu = smp_processor_id();
>         struct tss_struct *tss = &per_cpu(init_tss, cpu);
> 
>         if (unlikely(__get_cpu_var(pmu_ctx) || next_p->pfm_context))
>                 __pfm_ctxswout(prev_p, next_p);
> 
>         /*
>          * Reload esp0, LDT and the page table pointer:
>          */
>         tss->rsp0 = next->rsp0;
> 
> There is now a single hook and a conditional branch.
> this is similar to what you have with the debug registers.

It's still more than there was before. Also __get_cpu_var 
is quite a lot of instructions.

I would suggest you borrow some bits in one of the process
or thread info flags and then do a single test

if (unlikely(thr->flags & (DEBUG|PERFMON)) != 0) { 
        if (flags & DEBUG)
                ... do debug ...
        if (flags & PERFMON)
                ... do perfmon ...
}

[which you're at it you can probably add ioports in there too -
improving existing code is always a good thing]

Ideally flags is in some cache line that is already 
touched during context switch. If not you might need
to change the layout.

It's ok to put the do_perfmon stuff into a separate noinline
function because that will disturb the register allocation
in the caller less.

I would suggest doing this in separate preparing patches that
first just do it for existing facilities.

-Andi

P.S.: My comments probably apply to the i386 versions too
although I haven't read them.
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

[Index of Archives]     [Kernel Newbies]     [Netfilter]     [Bugtraq]     [Photo]     [Stuff]     [Gimp]     [Yosemite News]     [MIPS Linux]     [ARM Linux]     [Linux Security]     [Linux RAID]     [Video 4 Linux]     [Linux for the blind]     [Linux Resources]
  Powered by Linux