linux-kernel - Re: [PATCH 06/10] coresight: perf: traceid: Add perf notifiers for trace ID

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite for Android: free password hash cracker in your pocket

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <CAJ9a7Vjb0G4fSB6U33JruVWcBzpP459rYq+5CzMqxRcWeiJb0w@mail.gmail.com>
Date:   Wed, 6 Apr 2022 20:38:08 +0100
From:   Mike Leach <mike.leach@...aro.org>
To:     Mathieu Poirier <mathieu.poirier@...aro.org>
Cc:     suzuki.poulose@....com, coresight@...ts.linaro.org,
        linux-arm-kernel@...ts.infradead.org, linux-kernel@...r.kernel.org,
        peterz@...radead.org, mingo@...hat.com, acme@...nel.org,
        linux-perf-users@...r.kernel.org, leo.yan@...aro.org
Subject: Re: [PATCH 06/10] coresight: perf: traceid: Add perf notifiers for
 trace ID

Hi Mathieu,

On Wed, 6 Apr 2022 at 18:11, Mathieu Poirier <mathieu.poirier@...aro.org> wrote:
>
> On Tue, Mar 08, 2022 at 08:49:56PM +0000, Mike Leach wrote:
> > Adds in notifier calls to the trace ID allocator that perf
> > events are starting and stopping.
> >
> > This ensures that Trace IDs associated with CPUs remain the same
> > throughout the perf session, and are only release when all perf
> > sessions are complete.
> >
> > Signed-off-by: Mike Leach <mike.leach@...aro.org>
> > ---
> >  drivers/hwtracing/coresight/coresight-etm-perf.c | 16 +++++++++++++++-
> >  1 file changed, 15 insertions(+), 1 deletion(-)
> >
> > diff --git a/drivers/hwtracing/coresight/coresight-etm-perf.c b/drivers/hwtracing/coresight/coresight-etm-perf.c
> > index c039b6ae206f..008f9dac429d 100644
> > --- a/drivers/hwtracing/coresight/coresight-etm-perf.c
> > +++ b/drivers/hwtracing/coresight/coresight-etm-perf.c
> > @@ -22,6 +22,7 @@
> >  #include "coresight-etm-perf.h"
> >  #include "coresight-priv.h"
> >  #include "coresight-syscfg.h"
> > +#include "coresight-trace-id.h"
> >
> >  static struct pmu etm_pmu;
> >  static bool etm_perf_up;
> > @@ -223,11 +224,21 @@ static void free_event_data(struct work_struct *work)
> >               struct list_head **ppath;
> >
> >               ppath = etm_event_cpu_path_ptr(event_data, cpu);
> > -             if (!(IS_ERR_OR_NULL(*ppath)))
> > +             if (!(IS_ERR_OR_NULL(*ppath))) {
> >                       coresight_release_path(*ppath);
> > +                     /*
> > +                      * perf may have read a trace id for a cpu, but never actually
> > +                      * executed code on that cpu - which means the trace id would
> > +                      * not release on disable. Re-release here to be sure.
> > +                      */
> > +                     coresight_trace_id_put_cpu_id(cpu, coresight_get_trace_id_map());
>
> A CPU gets a traceID in event_etm_start() when the event is installed for
> running.  Do you see a scenario where etm_free_aux() is called without
> previously calling event_etm_stop()?
>

To ensure that IDs are obtained in a timely manner, they assign on
read. Therefore when cs_etm.c::cs_etm_info_fill() is called,
potentially the ID will be read for all CPUs - and dumped into the
AUXINFO data at the top of the perf.data file.
However, a --per-thread execution may actually only start the event on
a subset of cpus, hence we ensure that all cpus are released.
This was proven during testing when I ran both --per-thread and cpu
wide tests with logging monitoring the ID assignments.

I have programmed this deliberately defensively - on the basis that
the timings and orderings of the code/callbacks in todays perf may not
necessarily be the same in tomorrows. perf.

In future we may be able to use Suzuki's idea of embedding the ID into
an alternative packet in the perf.data file. but I think that should
be a separate change as it affects decode in a big way.

Regards

Mike


> > +             }
> >               *ppath = NULL;
> >       }
> >
> > +     /* mark perf event as done for trace id allocator */
> > +     coresight_trace_id_perf_stop();
> > +
> >       free_percpu(event_data->path);
> >       kfree(event_data);
> >  }
> > @@ -314,6 +325,9 @@ static void *etm_setup_aux(struct perf_event *event, void **pages,
> >               sink = user_sink = coresight_get_sink_by_id(id);
> >       }
> >
> > +     /* tell the trace ID allocator that a perf event is starting up */
> > +     coresight_trace_id_perf_start();
> > +
> >       /* check if user wants a coresight configuration selected */
> >       cfg_hash = (u32)((event->attr.config2 & GENMASK_ULL(63, 32)) >> 32);
> >       if (cfg_hash) {
> > --
> > 2.17.1
> >



-- 
Mike Leach
Principal Engineer, ARM Ltd.
Manchester Design Centre. UK