RE: [PATCH] perf cs-etm: Fix missing decoder for per-process trace

From: Besar Wicaksono
Date: Wed Sep 20 2023 - 12:48:17 EST

Next message: Nick Desaulniers: "Re: [PATCH v1 1/1] device property: Replace custom implementation of COUNT_ARGS()"
Previous message: Rob Herring: "Re: [PATCH 2/2] arm64: errata: Add Cortex-A520 speculative unprivileged load workaround"
In reply to: Besar Wicaksono: "RE: [PATCH] perf cs-etm: Fix missing decoder for per-process trace"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

Hi Mike,

Looks like I missed replying to your other comments on the code change.
I apologize for that. Please see more of my comments inline.

> -----Original Message-----
> From: Mike Leach <mike.leach@xxxxxxxxxx>
> Sent: Wednesday, September 20, 2023 3:46 AM
> To: Besar Wicaksono <bwicaksono@xxxxxxxxxx>
> Cc: james.clark@xxxxxxx; suzuki.poulose@xxxxxxx; linux-arm-
> kernel@xxxxxxxxxxxxxxxxxxx; linux-kernel@xxxxxxxxxxxxxxx;
> coresight@xxxxxxxxxxxxxxxx; linux-tegra@xxxxxxxxxxxxxxx; Thierry Reding
> <treding@xxxxxxxxxx>; Jonathan Hunter <jonathanh@xxxxxxxxxx>; Vikram
> Sethi <vsethi@xxxxxxxxxx>; Richard Wiley <rwiley@xxxxxxxxxx>; Yifei Wan
> <ywan@xxxxxxxxxx>
> Subject: Re: [PATCH] perf cs-etm: Fix missing decoder for per-process trace
>
> External email: Use caution opening links or attachments
>
>
> Hi,
>
> Can you provide a perf command line for both the record and decode
> phases that demonstrates the problem you are having?
>

perf record commands:
- perf record -e cs_etm//u -- taskset -c 3,4 ./test
- perf record -e cs_etm//u --per-thread -- taskset -c 3,4 ./test

Perf report command:
- perf report --stdio --dump

> Also note the cpu / trace protocol types
>

This is on Neoverse V2 with ETE source and TRBE sink.

> On Tue, 19 Sept 2023 at 23:47, Besar Wicaksono <bwicaksono@xxxxxxxxxx>
> wrote:
> >
> > The decoder creation for raw trace uses metadata from the first CPU.
> > On per-process/per-thread traces, the first CPU is CPU0. If CPU0 trace
> > is not enabled, its metadata will be marked unused and the decoder is
> > not created. Perf report dump skips the decoding part because the
> > decoder is missing.
> >
> > To fix this, use metadata of the CPU associated with sample object.
> >
> > Signed-off-by: Besar Wicaksono <bwicaksono@xxxxxxxxxx>
> > ---
> > tools/perf/util/cs-etm.c | 130 +++++++++++++++++++++++----------------
> > 1 file changed, 77 insertions(+), 53 deletions(-)
> >
> > diff --git a/tools/perf/util/cs-etm.c b/tools/perf/util/cs-etm.c
> > index 9729d006550d..3c3179a5eac6 100644
> > --- a/tools/perf/util/cs-etm.c
> > +++ b/tools/perf/util/cs-etm.c
> > @@ -640,71 +640,94 @@ static void cs_etm__packet_dump(const char
> *pkt_string)
> > fflush(stdout);
> > }
> >
> > -static void cs_etm__set_trace_param_etmv3(struct cs_etm_trace_params
> *t_params,
> > - struct cs_etm_auxtrace *etm, int idx,
> > - u32 etmidr)
> > +static void cs_etm__set_trace_param_etmv3(struct cs_etm_trace_params
> *t_param,
> > + u64 *metadata, u32 etmidr)
> > {
> > - u64 **metadata = etm->metadata;
> > -
> > - t_params[idx].protocol = cs_etm__get_v7_protocol_version(etmidr);
> > - t_params[idx].etmv3.reg_ctrl = metadata[idx][CS_ETM_ETMCR];
> > - t_params[idx].etmv3.reg_trc_id =
> metadata[idx][CS_ETM_ETMTRACEIDR];
> > + t_param->protocol = cs_etm__get_v7_protocol_version(etmidr);
> > + t_param->etmv3.reg_ctrl = metadata[CS_ETM_ETMCR];
> > + t_param->etmv3.reg_trc_id = metadata[CS_ETM_ETMTRACEIDR];
> > }
> >
>
> This is inherently incorrect - you are assuming that all trace devices
> on a given system are the same and have the same parameters. This does
> not have to be the case.
>

The intention here is the t_params index and metadata index could be different.
I was letting cs_etm__init_trace_params to provide the correct t_param and
metadata pointer. Maybe a different way is just by adding separate index argument
for the metadata ? For example:

@@ -641,14 +641,14 @@ static void cs_etm__packet_dump(const char *pkt_string)
}

static void cs_etm__set_trace_param_etmv3(struct cs_etm_trace_params *t_params,
- struct cs_etm_auxtrace *etm, int idx,
- u32 etmidr)
+ struct cs_etm_auxtrace *etm, int t_idx,
+ int m_idx, u32 etmidr)
{
u64 **metadata = etm->metadata;

- t_params[idx].protocol = cs_etm__get_v7_protocol_version(etmidr);
- t_params[idx].etmv3.reg_ctrl = metadata[idx][CS_ETM_ETMCR];
- t_params[idx].etmv3.reg_trc_id = metadata[idx][CS_ETM_ETMTRACEIDR];
+ t_params[t_idx].protocol = cs_etm__get_v7_protocol_version(etmidr);
+ t_params[t_idx].etmv3.reg_ctrl = metadata[m_idx][CS_ETM_ETMCR];
+ t_params[t_idx].etmv3.reg_trc_id = metadata[m_idx][CS_ETM_ETMTRACEIDR];
}

What do you think ?

>
> > -static void cs_etm__set_trace_param_etmv4(struct cs_etm_trace_params
> *t_params,
> > - struct cs_etm_auxtrace *etm, int idx)
> > +static void cs_etm__set_trace_param_etmv4(struct cs_etm_trace_params
> *t_param,
> > + u64 *metadata)
> > {
> > - u64 **metadata = etm->metadata;
> > + t_param->protocol = CS_ETM_PROTO_ETMV4i;
> > + t_param->etmv4.reg_idr0 = metadata[CS_ETMV4_TRCIDR0];
> > + t_param->etmv4.reg_idr1 = metadata[CS_ETMV4_TRCIDR1];
> > + t_param->etmv4.reg_idr2 = metadata[CS_ETMV4_TRCIDR2];
> > + t_param->etmv4.reg_idr8 = metadata[CS_ETMV4_TRCIDR8];
> > + t_param->etmv4.reg_configr = metadata[CS_ETMV4_TRCCONFIGR];
> > + t_param->etmv4.reg_traceidr = metadata[CS_ETMV4_TRCTRACEIDR];
> > +}
> >
> > - t_params[idx].protocol = CS_ETM_PROTO_ETMV4i;
> > - t_params[idx].etmv4.reg_idr0 = metadata[idx][CS_ETMV4_TRCIDR0];
> > - t_params[idx].etmv4.reg_idr1 = metadata[idx][CS_ETMV4_TRCIDR1];
> > - t_params[idx].etmv4.reg_idr2 = metadata[idx][CS_ETMV4_TRCIDR2];
> > - t_params[idx].etmv4.reg_idr8 = metadata[idx][CS_ETMV4_TRCIDR8];
> > - t_params[idx].etmv4.reg_configr =
> metadata[idx][CS_ETMV4_TRCCONFIGR];
> > - t_params[idx].etmv4.reg_traceidr =
> metadata[idx][CS_ETMV4_TRCTRACEIDR];
> > +static void cs_etm__set_trace_param_ete(struct cs_etm_trace_params
> *t_param,
> > + u64 *metadata)
> > +{
> > + t_param->protocol = CS_ETM_PROTO_ETE;
> > + t_param->ete.reg_idr0 = metadata[CS_ETE_TRCIDR0];
> > + t_param->ete.reg_idr1 = metadata[CS_ETE_TRCIDR1];
> > + t_param->ete.reg_idr2 = metadata[CS_ETE_TRCIDR2];
> > + t_param->ete.reg_idr8 = metadata[CS_ETE_TRCIDR8];
> > + t_param->ete.reg_configr = metadata[CS_ETE_TRCCONFIGR];
> > + t_param->ete.reg_traceidr = metadata[CS_ETE_TRCTRACEIDR];
> > + t_param->ete.reg_devarch = metadata[CS_ETE_TRCDEVARCH];
> > }
> >
>
> as above for ETE & ETMv4
>
> > -static void cs_etm__set_trace_param_ete(struct cs_etm_trace_params
> *t_params,
> > - struct cs_etm_auxtrace *etm, int idx)
> > +static int cs_etm__set_trace_param(struct cs_etm_trace_params *t_param,
> > + u64 *metadata)
> > {
> > - u64 **metadata = etm->metadata;
> > + u32 etmidr;
> > + u64 architecture = metadata[CS_ETM_MAGIC];
> > +
> > + switch (architecture) {
> > + case __perf_cs_etmv3_magic:
> > + etmidr = metadata[CS_ETM_ETMIDR];
> > + cs_etm__set_trace_param_etmv3(t_param, metadata, etmidr);
> > + break;
> > + case __perf_cs_etmv4_magic:
> > + cs_etm__set_trace_param_etmv4(t_param, metadata);
> > + break;
> > + case __perf_cs_ete_magic:
> > + cs_etm__set_trace_param_ete(t_param, metadata);
> > + break;
> > + default:
> > + return -EINVAL;
> > + }
> >
> > - t_params[idx].protocol = CS_ETM_PROTO_ETE;
> > - t_params[idx].ete.reg_idr0 = metadata[idx][CS_ETE_TRCIDR0];
> > - t_params[idx].ete.reg_idr1 = metadata[idx][CS_ETE_TRCIDR1];
> > - t_params[idx].ete.reg_idr2 = metadata[idx][CS_ETE_TRCIDR2];
> > - t_params[idx].ete.reg_idr8 = metadata[idx][CS_ETE_TRCIDR8];
> > - t_params[idx].ete.reg_configr = metadata[idx][CS_ETE_TRCCONFIGR];
> > - t_params[idx].ete.reg_traceidr = metadata[idx][CS_ETE_TRCTRACEIDR];
> > - t_params[idx].ete.reg_devarch = metadata[idx][CS_ETE_TRCDEVARCH];
> > + return 0;
> > }
> >
> > static int cs_etm__init_trace_params(struct cs_etm_trace_params
> *t_params,
> > struct cs_etm_auxtrace *etm,
> > + bool formatted,
> > + int sample_cpu,
> > int decoders)
> > {
> > - int i;
> > - u32 etmidr;
> > - u64 architecture;
> > -
> > - for (i = 0; i < decoders; i++) {
> > - architecture = etm->metadata[i][CS_ETM_MAGIC];
> > + int i, ret;
> > + struct cs_etm_trace_params *t_param;
> > + u64 *metadata;
> >
> > - switch (architecture) {
> > - case __perf_cs_etmv3_magic:
> > - etmidr = etm->metadata[i][CS_ETM_ETMIDR];
> > - cs_etm__set_trace_param_etmv3(t_params, etm, i, etmidr);
> > - break;
> > - case __perf_cs_etmv4_magic:
> > - cs_etm__set_trace_param_etmv4(t_params, etm, i);
> > - break;
> > - case __perf_cs_ete_magic:
> > - cs_etm__set_trace_param_ete(t_params, etm, i);
> > - break;
> > - default:
> > + if (!formatted) {
> > + /*
> > + * There is only one decoder when unformatted. Use metadata of
> > + * sample AUX cpu.
> > + */
>
>
> Unformatted ETE data should not only have a single decoder - there
> must be a decoder per CPU - as per all trace decode flows. The key
> difference with ETE is that it usually has its own trace sink (TRBE),
> so no formatting is needed, but per cpu decoder is still required.
>

The change is in the context of queue allocation. My understanding is there is
one CPU per queue when unformatted, which I think is aligned with your statement.
I could mention one decoder per CPU to clarify more.

Pasting the documentation under cs_etm__alloc_queue:
/*
* Each queue can only contain data from one CPU when unformatted, so only one decoder is
* needed.
*/

Thanks,
Besar

> Regards
>
> Mike
>
>
> > + t_param = t_params;
> > + metadata = get_cpu_data(etm, sample_cpu);
> > + if (!metadata) {
> > + pr_err("CS_ETM: invalid sample CPU: %d\n", sample_cpu);
> > return -EINVAL;
> > }
> > +
> > + return cs_etm__set_trace_param(t_param, metadata);
> > + }
> > +
> > + for (i = 0; i < decoders; i++) {
> > + t_param = &t_params[i];
> > + metadata = etm->metadata[i];
> > + ret = cs_etm__set_trace_param(t_param, metadata);
> > + if (ret)
> > + return ret;
> > }
> >
> > return 0;
> > @@ -1016,7 +1039,7 @@ static u32 cs_etm__mem_access(struct
> cs_etm_queue *etmq, u8 trace_chan_id,
> > }
> >
> > static struct cs_etm_queue *cs_etm__alloc_queue(struct cs_etm_auxtrace
> *etm,
> > - bool formatted)
> > + bool formatted, int sample_cpu)
> > {
> > struct cs_etm_decoder_params d_params;
> > struct cs_etm_trace_params *t_params = NULL;
> > @@ -1041,7 +1064,7 @@ static struct cs_etm_queue
> *cs_etm__alloc_queue(struct cs_etm_auxtrace *etm,
> > if (!t_params)
> > goto out_free;
> >
> > - if (cs_etm__init_trace_params(t_params, etm, decoders))
> > + if (cs_etm__init_trace_params(t_params, etm, formatted, sample_cpu,
> decoders))
> > goto out_free;
> >
> > /* Set decoder parameters to decode trace packets */
> > @@ -1081,14 +1104,15 @@ static struct cs_etm_queue
> *cs_etm__alloc_queue(struct cs_etm_auxtrace *etm,
> > static int cs_etm__setup_queue(struct cs_etm_auxtrace *etm,
> > struct auxtrace_queue *queue,
> > unsigned int queue_nr,
> > - bool formatted)
> > + bool formatted,
> > + int sample_cpu)
> > {
> > struct cs_etm_queue *etmq = queue->priv;
> >
> > if (list_empty(&queue->head) || etmq)
> > return 0;
> >
> > - etmq = cs_etm__alloc_queue(etm, formatted);
> > + etmq = cs_etm__alloc_queue(etm, formatted, sample_cpu);
> >
> > if (!etmq)
> > return -ENOMEM;
> > @@ -2816,7 +2840,7 @@ static int
> cs_etm__process_auxtrace_event(struct perf_session *session,
> > * formatted in piped mode (true).
> > */
> > err = cs_etm__setup_queue(etm, &etm->queues.queue_array[idx],
> > - idx, true);
> > + idx, true, -1);
> > if (err)
> > return err;
> >
> > @@ -3022,7 +3046,7 @@ static int cs_etm__queue_aux_fragment(struct
> perf_session *session, off_t file_o
> > idx = auxtrace_event->idx;
> > formatted = !(aux_event->flags &
> PERF_AUX_FLAG_CORESIGHT_FORMAT_RAW);
> > return cs_etm__setup_queue(etm, &etm-
> >queues.queue_array[idx],
> > - idx, formatted);
> > + idx, formatted, sample->cpu);
> > }
> >
> > /* Wasn't inside this buffer, but there were no parse errors. 1 == 'not
> found' */
> > --
> > 2.34.1
> >
>
>
> --
> Mike Leach
> Principal Engineer, ARM Ltd.
> Manchester Design Centre. UK

Next message: Nick Desaulniers: "Re: [PATCH v1 1/1] device property: Replace custom implementation of COUNT_ARGS()"
Previous message: Rob Herring: "Re: [PATCH 2/2] arm64: errata: Add Cortex-A520 speculative unprivileged load workaround"
In reply to: Besar Wicaksono: "RE: [PATCH] perf cs-etm: Fix missing decoder for per-process trace"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]