Re: [patch] Latency Tracer, voluntary-preempt-2.6.8-rc4-O6

[Ingo Molnar]

* Florian Schmidt <mista.tapas@xxxxxxx> wrote:

> On Fri, 13 Aug 2004 12:54:06 +0200
> Ingo Molnar <mingo@xxxxxxx> wrote:
> 
> > > (ksoftirqd/0/2): 307 us critical section violates 250 us threshold.
> > >  => started at: <___do_softirq+0x20/0x90>
> > >  => ended at: <cond_resched_softirq+0x59/0x70>
> > 
> > this is too opaque - could you try -O7, enable tracing and save a
> > /proc/latency_trace instance of such a latency? It looks like some
> > sort of softirq latency - perhaps one particular driver's timer fn
> > causes it- we'll be able to tell more from the trace.
> 
> Hi, this looks like one of them:
> 
> mango:~# cat /proc/latency_trace 
> preemption latency trace v1.0
> -----------------------------
>  latency: 308 us, entries: 12 (12)
>  process: ksoftirqd/0/2, uid: 0
>  nice: -10, policy: 0, rt_priority: 0
> =======>
>  0.000ms (+0.000ms): run_timer_softirq (___do_softirq)
>  0.000ms (+0.000ms): sis900_timer (run_timer_softirq)
>  0.001ms (+0.000ms): mdio_read (sis900_timer)
>  0.002ms (+0.000ms): mdio_reset (mdio_read)
>  0.071ms (+0.069ms): mdio_idle (mdio_read)
>  0.151ms (+0.079ms): mdio_read (sis900_timer)
>  0.151ms (+0.000ms): mdio_reset (mdio_read)
>  0.220ms (+0.069ms): mdio_idle (mdio_read)
>  0.300ms (+0.079ms): __mod_timer (sis900_timer)
>  0.300ms (+0.000ms): internal_add_timer (__mod_timer)
>  0.300ms (+0.000ms): cond_resched_softirq (run_timer_softirq)
>  0.301ms (+0.000ms): check_preempt_timing (touch_preempt_timing)

indeed this seems to be a driver related timer: driver/net/sis900.c's
sis900_timer() function. This timer polls your network card for link
status once every second. The mdio_read() function does alot of in/out
instructions which are quite slow.

could you try the following and disable mdio_delay():

	#define mdio_delay() do { } while (0)

normally it should work just fine. Worst-case you'd get a non-working
network driver. Does this change reduce the latency?

	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/