lkml.org 
[lkml]   [2008]   [May]   [4]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [patch 00/37] Linux Kernel Markers instrumentation for sched-devel.git
* Peter Zijlstra (a.p.zijlstra@chello.nl) wrote:
> On Sat, 2008-04-26 at 16:11 -0400, Mathieu Desnoyers wrote:
> > * Peter Zijlstra (a.p.zijlstra@chello.nl) wrote:
> > > On Thu, 2008-04-24 at 11:03 -0400, Mathieu Desnoyers wrote:
> > > > Hi Ingo,
> > > >
> > > > Here is a rather large patchset applying kernel instrumentation to
> > > > sched-devel.git. It includes, mainly :
> > >
> > > I saw this land in sched-devel, how about this:
> > >
> >
> > I think the main reason why those markers are ugly is they have
> > information meant for use by a general purpose tracer (it will only take
> > the parameters before the ## signs).
> >
> > The other way around is to start specializing the general purpose tracer
> > to extract the information it needs from the task and rq pointers and
> > put that in the traces. Actually, that's the approach I use currently
> > use in LTTng, but Ingo seemed interested to have the union of parameters
> > needed by specialized and GP tracers, so this is what I did.
> >
> > The only thing I dislike with the approach of putting everything in a
> > separatered header is that it adds a layer of indirection when one try
> > to to quickly grep for trace_mark() in the kernel tree to see where the
> > conceptual tracing points are. Therefore, it might be interesting to
> > simply remove the parameters meant for the general purpose tracer and
> > let this GP tracer create specialized probes for these instrumentation
> > sites. It would therefore remove the ugliness without creating a
> > supplementary indirection layer.
>
> In part its the extra parameters, but to a large part its that darn
> string. I'm still not getting why we can't just live with trace marks
> like 'regular' functions much like the ones I used to wrap trace_mark().
>
> > > Index: linux-2.6-2/kernel/sched_trace.h
> > > ===================================================================
> > > --- /dev/null
> > > +++ linux-2.6-2/kernel/sched_trace.h
> > > @@ -0,0 +1,41 @@
> > > +#include <linux/marker.h>
> > > +
> > > +static inline void trace_kernel_sched_wait(struct task_struct *p)
> > > +{
> > > + trace_mark(kernel_sched_wait_task, "pid %d state %ld",
> > > + p->pid, p->state);
> > > +}
> > > +
> > > +static inline
> > > +void trace_kernel_sched_wakeup(struct rq *rq, struct task_struct *p)
> > > +{
> > > + trace_mark(kernel_sched_wakeup,
> > > + "pid %d state %ld ## rq %p task %p rq->curr %p",
> > > + p->pid, p->state, rq, p, rq->curr);
> > > +}
> > > +
> > > +static inline
> > > +void trace_kernel_sched_wakeup_new(struct rq *rq, struct task_struct *p)
> > > +{
> > > + trace_mark(kernel_sched_wakeup_new,
> > > + "pid %d state %ld ## rq %p task %p rq->curr %p",
> > > + p->pid, p->state, rq, p, rq->curr);
> > > +}
> > > +
> > > +static inline void trace_kernel_sched_switch(struct rq *rq,
> > > + struct task_struct *prev, struct task_struct *next)
> > > +{
> > > + trace_mark(kernel_sched_schedule,
> > > + "prev_pid %d next_pid %d prev_state %ld "
> > > + "## rq %p prev %p next %p",
> > > + prev->pid, next->pid, prev->state,
> > > + rq, prev, next);
> > > +}
> > > +
> > > +static inline void
> > > +trace_kernel_sched_migrate_task(struct task_struct *p, int src, int dst)
> > > +{
> > > + trace_mark(kernel_sched_migrate_task,
> > > + "pid %d state %ld dest_cpu %d",
> > > + p->pid, p->state, dst);
> > > +}
>
> One advantage of having these things close together would be that it
> stimulates consistency between the various trace points. Something that
> would be sorely needed with such a free form mechanism.
>

I was thinking of the possibility of per-subsystem list of
instrumentation sites, which would become more or less fixed, anyway,
which is what you seem to propose here. Given that it also cleans up the
scheduler code, I think it's interesting to do.

Do you have an updated version following Andrew's comments I could Ack ?
:)

Mathieu


> Peter
>

--
Mathieu Desnoyers
OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F BA06 3F25 A8FE 3BAE 9A68


\
 
 \ /
  Last update: 2008-05-04 17:11    [W:1.409 / U:0.076 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site