Re: [PATCH 1/4] tracing: move __DO_TRACE out of line

[Jeremy Fitzhardinge]

Ingo Molnar wrote:
> * Jeremy Fitzhardinge <jeremy@xxxxxxxx> wrote:
>
>   
> > Taking __do_trace_sched_switch out of lines inserts this into the 
> > hot path (6 instructions, 31 bytes):
> >
> >        cmpl    $0, __tracepoint_sched_switch+8(%rip)   #, __tracepoint_sched_switch.state
> >        je      .L1748  #,
> >        movq    -136(%rbp), %rdx        # next,
> >        movq    -144(%rbp), %rsi        # prev,
> >        movq    %rbx, %rdi      # rq,
> >        call    __do_trace_sched_switch #
> > .L1748:
> >     
>
> Hm, why isnt this off-line in the function? It's marked unlikely(), 
> isnt it?
>   

Yes, its unlikely().  I don't know why it doesn't move it; I've never 
seen unlikely() do anything useful.

> also, did you investigate the effect on the _instrumented_ function 
> itself? (i.e. the non-tracing related bits) A function call clobbers 
> various registers and creates pressure on gcc to shuffle registers 
> around.
>   

Well, there's a function call in either case, so I don't think it makes 
much difference.

   J
--
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/