From: Arnaud Fiorini <arnaud.fiorini@polymtl.ca>
To: Thomas Monjalon <thomas@monjalon.net>,
Harry van Haaren <harry.van.haaren@intel.com>,
Jerin Jacob <jerinj@marvell.com>,
Sunil Kumar Kori <skori@marvell.com>
Cc: dev@dpdk.org
Subject: Re: [PATCH v3] eal: add tracepoints to track lcores and services
Date: Mon, 5 Jun 2023 10:48:36 -0400 [thread overview]
Message-ID: <3df7820c-b87e-6a9c-96bf-f37c8f87e649@polymtl.ca> (raw)
In-Reply-To: <20230512143031.1753859-1-arnaud.fiorini@polymtl.ca>
On 2023-05-12 10:30 a.m., Arnaud Fiorini wrote:
> The tracepoints added are used to track lcore role and status,
> as well as service mapping and service runstates. These
> tracepoints are then used in analyses in Trace Compass.
>
> Signed-off-by: Arnaud Fiorini <arnaud.fiorini@polymtl.ca>
> ---
> .mailmap | 1 +
> doc/guides/prog_guide/service_cores.rst | 14 ++++++
> lib/eal/common/eal_common_thread.c | 4 ++
> lib/eal/common/eal_common_trace_points.c | 21 +++++++++
> lib/eal/common/rte_service.c | 17 ++++++-
> lib/eal/include/eal_trace_internal.h | 60 ++++++++++++++++++++++++
> 6 files changed, 116 insertions(+), 1 deletion(-)
>
> diff --git a/.mailmap b/.mailmap
> index 0859104404..6453cb2aa5 100644
> --- a/.mailmap
> +++ b/.mailmap
> @@ -120,6 +120,7 @@ Archana Muniganti <marchana@marvell.com> <muniganti.archana@caviumnetworks.com>
> Archit Pandey <architpandeynitk@gmail.com>
> Arkadiusz Kubalewski <arkadiusz.kubalewski@intel.com>
> Arkadiusz Kusztal <arkadiuszx.kusztal@intel.com>
> +Arnaud Fiorini <arnaud.fiorini@polymtl.ca>
> Arnon Warshavsky <arnon@qwilt.com>
> Arshdeep Kaur <arshdeep.kaur@intel.com>
> Artem V. Andreev <artem.andreev@oktetlabs.ru>
> diff --git a/doc/guides/prog_guide/service_cores.rst b/doc/guides/prog_guide/service_cores.rst
> index 270b875783..3b9a947e4c 100644
> --- a/doc/guides/prog_guide/service_cores.rst
> +++ b/doc/guides/prog_guide/service_cores.rst
> @@ -52,3 +52,17 @@ The service core library is capable of collecting runtime statistics like number
> of calls to a specific service, and number of cycles used by the service. The
> cycle count collection is dynamically configurable, allowing any application to
> profile the services running on the system at any time.
> +
> +Service Core Tracing
> +~~~~~~~~~~~~~~~~~~~~
> +
> +The service core library is instrumented with tracepoints using the DPDK Trace
> +Library. These tracepoints allow you to track the service and logical cores
> +state. To activate tracing when launching a DPDK program it is necessary to use the
> +``--trace`` option to specify a regular expression to select which tracepoints
> +to enable. Here is an example if you want to only specify service core tracing::
> +
> + ./dpdk/examples/service_cores/build/service_cores --trace="lib.eal.thread*" --trace="lib.eal.service*"
> +
> +See the `DPDK Trace Library <https://doc.dpdk.org/guides/prog_guide/trace_lib.html>`_
> +for details.
> diff --git a/lib/eal/common/eal_common_thread.c b/lib/eal/common/eal_common_thread.c
> index 079a385630..25dbdd68e3 100644
> --- a/lib/eal/common/eal_common_thread.c
> +++ b/lib/eal/common/eal_common_thread.c
> @@ -205,6 +205,8 @@ eal_thread_loop(void *arg)
> __ATOMIC_ACQUIRE)) == NULL)
> rte_pause();
>
> + rte_eal_trace_thread_lcore_running(lcore_id, f);
> +
> /* call the function and store the return value */
> fct_arg = lcore_config[lcore_id].arg;
> ret = f(fct_arg);
> @@ -219,6 +221,8 @@ eal_thread_loop(void *arg)
> */
> __atomic_store_n(&lcore_config[lcore_id].state, WAIT,
> __ATOMIC_RELEASE);
> +
> + rte_eal_trace_thread_lcore_stopped(lcore_id);
> }
>
> /* never reached */
> diff --git a/lib/eal/common/eal_common_trace_points.c b/lib/eal/common/eal_common_trace_points.c
> index 3f5bf5c55c..0f1240ea3a 100644
> --- a/lib/eal/common/eal_common_trace_points.c
> +++ b/lib/eal/common/eal_common_trace_points.c
> @@ -70,6 +70,27 @@ RTE_TRACE_POINT_REGISTER(rte_eal_trace_thread_remote_launch,
> lib.eal.thread.remote.launch)
> RTE_TRACE_POINT_REGISTER(rte_eal_trace_thread_lcore_ready,
> lib.eal.thread.lcore.ready)
> +RTE_TRACE_POINT_REGISTER(rte_eal_trace_thread_lcore_running,
> + lib.eal.thread.lcore.running)
> +RTE_TRACE_POINT_REGISTER(rte_eal_trace_thread_lcore_stopped,
> + lib.eal.thread.lcore.stopped)
> +
> +RTE_TRACE_POINT_REGISTER(rte_eal_trace_service_map_lcore,
> + lib.eal.service.map.lcore)
> +RTE_TRACE_POINT_REGISTER(rte_eal_trace_service_lcore_state_change,
> + lib.eal.service.lcore.state.change)
> +RTE_TRACE_POINT_REGISTER(rte_eal_trace_service_lcore_start,
> + lib.eal.service.lcore.start)
> +RTE_TRACE_POINT_REGISTER(rte_eal_trace_service_lcore_stop,
> + lib.eal.service.lcore.stop)
> +RTE_TRACE_POINT_REGISTER(rte_eal_trace_service_run_begin,
> + lib.eal.service.run.begin)
> +RTE_TRACE_POINT_REGISTER(rte_eal_trace_service_runstate_set,
> + lib.eal.service.run.state.set)
> +RTE_TRACE_POINT_REGISTER(rte_eal_trace_service_run_end,
> + lib.eal.service.run.end)
> +RTE_TRACE_POINT_REGISTER(rte_eal_trace_service_component_register,
> + lib.eal.service.component.register)
>
> RTE_TRACE_POINT_REGISTER(rte_eal_trace_intr_callback_register,
> lib.eal.intr.register)
> diff --git a/lib/eal/common/rte_service.c b/lib/eal/common/rte_service.c
> index 7ab48f2be8..94e872a08a 100644
> --- a/lib/eal/common/rte_service.c
> +++ b/lib/eal/common/rte_service.c
> @@ -9,6 +9,7 @@
> #include <rte_service.h>
> #include <rte_service_component.h>
>
> +#include <eal_trace_internal.h>
> #include <rte_lcore.h>
> #include <rte_branch_prediction.h>
> #include <rte_common.h>
> @@ -16,6 +17,7 @@
> #include <rte_atomic.h>
> #include <rte_malloc.h>
> #include <rte_spinlock.h>
> +#include <rte_trace_point.h>
>
> #include "eal_private.h"
>
> @@ -276,6 +278,8 @@ rte_service_component_register(const struct rte_service_spec *spec,
> if (id_ptr)
> *id_ptr = free_slot;
>
> + rte_eal_trace_service_component_register(free_slot, spec->name);
> +
> return 0;
> }
>
> @@ -336,6 +340,7 @@ rte_service_runstate_set(uint32_t id, uint32_t runstate)
> __atomic_store_n(&s->app_runstate, RUNSTATE_STOPPED,
> __ATOMIC_RELEASE);
>
> + rte_eal_trace_service_runstate_set(id, runstate);
> return 0;
> }
>
> @@ -368,6 +373,7 @@ static inline void
> service_runner_do_callback(struct rte_service_spec_impl *s,
> struct core_state *cs, uint32_t service_idx)
> {
> + rte_eal_trace_service_run_begin(service_idx, rte_lcore_id());
> void *userdata = s->spec.callback_userdata;
>
> if (service_stats_enabled(s)) {
> @@ -395,8 +401,10 @@ service_runner_do_callback(struct rte_service_spec_impl *s,
>
> __atomic_store_n(&service_stats->calls,
> service_stats->calls + 1, __ATOMIC_RELAXED);
> - } else
> + } else {
> s->spec.callback(userdata);
> + }
> + rte_eal_trace_service_run_end(service_idx, rte_lcore_id());
> }
>
>
> @@ -658,6 +666,7 @@ int32_t
> rte_service_map_lcore_set(uint32_t id, uint32_t lcore, uint32_t enabled)
> {
> uint32_t on = enabled > 0;
> + rte_eal_trace_service_map_lcore(id, lcore, enabled);
> return service_update(id, lcore, &on, 0);
> }
>
> @@ -683,6 +692,8 @@ set_lcore_state(uint32_t lcore, int32_t state)
>
> /* update per-lcore optimized state tracking */
> lcore_states[lcore].is_service_core = (state == ROLE_SERVICE);
> +
> + rte_eal_trace_service_lcore_state_change(lcore, state);
> }
>
> int32_t
> @@ -780,6 +791,8 @@ rte_service_lcore_start(uint32_t lcore)
> */
> __atomic_store_n(&cs->runstate, RUNSTATE_RUNNING, __ATOMIC_RELEASE);
>
> + rte_eal_trace_service_lcore_start(lcore);
> +
> int ret = rte_eal_remote_launch(service_runner_func, 0, lcore);
> /* returns -EBUSY if the core is already launched, 0 on success */
> return ret;
> @@ -824,6 +837,8 @@ rte_service_lcore_stop(uint32_t lcore)
> __atomic_store_n(&lcore_states[lcore].runstate, RUNSTATE_STOPPED,
> __ATOMIC_RELEASE);
>
> + rte_eal_trace_service_lcore_stop(lcore);
> +
> return 0;
> }
>
> diff --git a/lib/eal/include/eal_trace_internal.h b/lib/eal/include/eal_trace_internal.h
> index 57d6de2535..09c354717f 100644
> --- a/lib/eal/include/eal_trace_internal.h
> +++ b/lib/eal/include/eal_trace_internal.h
> @@ -174,6 +174,66 @@ RTE_TRACE_POINT(
> rte_trace_point_emit_u32(lcore_id);
> rte_trace_point_emit_string(cpuset);
> )
> +RTE_TRACE_POINT(
> + rte_eal_trace_thread_lcore_running,
> + RTE_TRACE_POINT_ARGS(unsigned int lcore_id, void *f),
> + rte_trace_point_emit_u32(lcore_id);
> + rte_trace_point_emit_ptr(f);
> +)
> +RTE_TRACE_POINT(
> + rte_eal_trace_thread_lcore_stopped,
> + RTE_TRACE_POINT_ARGS(unsigned int lcore_id),
> + rte_trace_point_emit_u32(lcore_id);
> +)
> +
> +/* Service */
> +RTE_TRACE_POINT(
> + rte_eal_trace_service_map_lcore,
> + RTE_TRACE_POINT_ARGS(unsigned int id, unsigned int lcore_id, unsigned int enabled),
> + rte_trace_point_emit_u32(id);
> + rte_trace_point_emit_u32(lcore_id);
> + rte_trace_point_emit_u32(enabled);
> +)
> +RTE_TRACE_POINT(
> + rte_eal_trace_service_lcore_state_change,
> + RTE_TRACE_POINT_ARGS(unsigned int lcore_id, int lcore_state),
> + rte_trace_point_emit_u32(lcore_id);
> + rte_trace_point_emit_i32(lcore_state);
> +)
> +RTE_TRACE_POINT(
> + rte_eal_trace_service_lcore_start,
> + RTE_TRACE_POINT_ARGS(unsigned int lcore_id),
> + rte_trace_point_emit_u32(lcore_id);
> +)
> +RTE_TRACE_POINT(
> + rte_eal_trace_service_lcore_stop,
> + RTE_TRACE_POINT_ARGS(unsigned int lcore_id),
> + rte_trace_point_emit_u32(lcore_id);
> +)
> +RTE_TRACE_POINT(
> + rte_eal_trace_service_run_begin,
> + RTE_TRACE_POINT_ARGS(unsigned int id, unsigned int lcore_id),
> + rte_trace_point_emit_u32(id);
> + rte_trace_point_emit_u32(lcore_id);
> +)
> +RTE_TRACE_POINT(
> + rte_eal_trace_service_runstate_set,
> + RTE_TRACE_POINT_ARGS(unsigned int id, unsigned int run_state),
> + rte_trace_point_emit_u32(id);
> + rte_trace_point_emit_u32(run_state);
> +)
> +RTE_TRACE_POINT(
> + rte_eal_trace_service_run_end,
> + RTE_TRACE_POINT_ARGS(unsigned int id, unsigned int lcore_id),
> + rte_trace_point_emit_u32(id);
> + rte_trace_point_emit_u32(lcore_id);
> +)
> +RTE_TRACE_POINT(
> + rte_eal_trace_service_component_register,
> + RTE_TRACE_POINT_ARGS(int id, const char *service_name),
> + rte_trace_point_emit_i32(id);
> + rte_trace_point_emit_string(service_name);
> +)
>
> #ifdef __cplusplus
> }
Hi,
I added documentation, removed tracepoint duplication and remove the
unnecessary use of RTE_TRACE_POINT_FP.
Are there any other changes needed for this patch ?
next prev parent reply other threads:[~2023-06-05 14:48 UTC|newest]
Thread overview: 9+ messages / expand[flat|nested] mbox.gz Atom feed top
2023-05-11 20:16 [PATCH v2] " Arnaud Fiorini
2023-05-11 20:50 ` Stephen Hemminger
2023-05-11 21:01 ` Arnaud Fiorini
2023-05-11 22:18 ` Stephen Hemminger
2023-05-12 14:30 ` [PATCH v3] " Arnaud Fiorini
2023-06-05 14:48 ` Arnaud Fiorini [this message]
2023-06-13 8:38 ` David Marchand
2023-06-13 10:00 ` Van Haaren, Harry
2023-06-21 8:25 ` David Marchand
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=3df7820c-b87e-6a9c-96bf-f37c8f87e649@polymtl.ca \
--to=arnaud.fiorini@polymtl.ca \
--cc=dev@dpdk.org \
--cc=harry.van.haaren@intel.com \
--cc=jerinj@marvell.com \
--cc=skori@marvell.com \
--cc=thomas@monjalon.net \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).