From: Hongzhan Chen <hongzhan.c...@intel.com> For cobalt thread, there is special struct param_ex data stored in data record, we need to parse and print its content out correctly to hint user.
Signed-off-by: Hongzhan Chen <hongzhan.c...@intel.com> [Jan: rework installation] Signed-off-by: Jan Kiszka <jan.kis...@siemens.com> --- configure.ac | 1 + tracing/Makefile.am | 7 +- tracing/README | 84 ++++++++++ tracing/libtraceevent/Makefile.am | 19 +++ .../plugin_xenomai_schedparams.c | 158 ++++++++++++++++++ 5 files changed, 268 insertions(+), 1 deletion(-) create mode 100644 tracing/README create mode 100644 tracing/libtraceevent/Makefile.am create mode 100644 tracing/libtraceevent/plugin_xenomai_schedparams.c diff --git a/configure.ac b/configure.ac index 4d11a6e421..0194537935 100644 --- a/configure.ac +++ b/configure.ac @@ -977,6 +977,7 @@ AC_CONFIG_FILES([ \ scripts/xeno \ tracing/Makefile \ tracing/kernelshark/Makefile \ + tracing/libtraceevent/Makefile \ lib/Makefile \ lib/boilerplate/Makefile \ lib/boilerplate/init/Makefile \ diff --git a/tracing/Makefile.am b/tracing/Makefile.am index 885a9497f7..da7d979f9a 100644 --- a/tracing/Makefile.am +++ b/tracing/Makefile.am @@ -4,5 +4,10 @@ if XENO_KSHARK_PLUGIN SUBDIRS += kernelshark endif +if XENO_LIBTRACEEVENT_PLUGIN +SUBDIRS += libtraceevent +endif + DIST_SUBDIRS = \ - kernelshark + kernelshark \ + libtraceevent diff --git a/tracing/README b/tracing/README new file mode 100644 index 0000000000..298797052e --- /dev/null +++ b/tracing/README @@ -0,0 +1,84 @@ + +What is it? +============= + + The modules under this folder is plugins implemented for Xenomai tracing. + kernelshark subdir includes plugin for KernelShark. + libtraceevent subduir includes plugin for libtraceevent. + +Dependencies +================================ + + For KernelShark plugin: + Building standlone plugin for KernelShark depends on + following patchset after kernelshark-v2.1.0: + kernel-shark: Install missing headers (commit 5419186f4bbad68aa16849882a8f12fa9adb22c5) + kernel-shark: Add KsPluginsGUI.hpp/.cpp (commit 59b5763c7c52b703e3b8e05be801f7c85365c9d3) + kernel-shark: Load 'ctrl' interface for user plugins (commit e35970770b71f0cc849870512a806dff96bf19e1) + + If you cannot confirm that your local installed KernelShark include these patches, + please refer to https://kernelshark.org/build.html to download updated code + and build & install again. + +How to build and install plugin? +==================================== + + goto your xenomai folder + ./scripts/bootstrap + ./configure --enable-xxx-plugin + cd tracing/xxx/ + make + sudo make install + + Note: Please replace xxx with your module name that you want to build and install plugin. + +How to use plugin? +==================================== + + For KernelShark plugin: + + kernelshark -p your_path_of_libplugin_xenomai_cobalt_switch_events.so your_path_of_trace.dat + + For libtraceevent plugin: + + While libtracevent and its plugins is installed correctly, the plugins + would be loaded automatically by other module such as libtracecmd after + corresponding tracing tool such as KernelShark or trace-cmd run. + +What the difference is after plug works? +=============================================== + + For KernelShark plugin: + 1. For those trace log which does not include cobalt_switch_context event, you may + not expect that there is any difference after Xenomai plugin is loaded. + + 2. But when there is cobalt_switch_context event in trace log, the plugin would + analysis cobalt_switch_context event and visualize the OOB state. + + 3. How to check OOB state with cobalt blue hollow box? + + 3.1. Please check if there is cobalt_switch_context events existing in the + log using search boxes in the list area of kernelshark GUI. + + 3.2. When there is cobalt_switch_context events existing in the log, please goto + task plots dialog to choose tasks that you may want to check its OOB state by + clicking menu Plots->Tasks and then hit "Apply". + + 3.3. Selected tasks would be added to the bottom of the graph area as task plots. + You may zoom in to check if there is cobalt blue hollow box showing in + the corresponding task bar/plot if the time span is really large and + there is too much events. + + 3.4. When you double click the hollow box, there is two vertical lines + that marked as "Mark A" and "Mark B" which you may use to measure time of + running in OOB state between two cobalt_switch_context events. + + Note: + Please refer to [1] about introduction of search boxes or task plot or "Mark A" & + "Mark B" on Graph Control Area for more detailed info. + + For libtraceevent plugin: + + It would try to show sched_param_ex param correctly. + +[1]: https://kernelshark.org/Documentation.html diff --git a/tracing/libtraceevent/Makefile.am b/tracing/libtraceevent/Makefile.am new file mode 100644 index 0000000000..542d71cb6f --- /dev/null +++ b/tracing/libtraceevent/Makefile.am @@ -0,0 +1,19 @@ +libsub_LTLIBRARIES = plugin_xenomai_schedparams.la + +plugin_xenomai_schedparams_la_SOURCES = \ + plugin_xenomai_schedparams.c + +plugin_xenomai_schedparams_la_CPPFLAGS = \ + $(LIBTRACEEVENT_CFLAGS) \ + -I$(top_srcdir)/include + +plugin_xenomai_schedparams_la_LDFLAGS = \ + $(LIBTRACEEVENT_LIBS) \ + -module -avoid-version -nostartfiles + +AM_LIBTOOLFLAGS = --silent --tag=disable-static + +libsubdir := $(libdir)/traceevent/plugins + +install-data-hook: + $(RM) $(DESTDIR)$(libsubdir)/$(libsub_LTLIBRARIES) diff --git a/tracing/libtraceevent/plugin_xenomai_schedparams.c b/tracing/libtraceevent/plugin_xenomai_schedparams.c new file mode 100644 index 0000000000..01234fe5f0 --- /dev/null +++ b/tracing/libtraceevent/plugin_xenomai_schedparams.c @@ -0,0 +1,158 @@ +// SPDX-License-Identifier: LGPL-2.1 +/* + * Copyright (C) 2021 Intel Inc, Hongzhan Chen <hongzhan.c...@intel.com> + */ +#include <stdio.h> +#include <stdlib.h> +#include <string.h> +#include <linux/sched.h> +#include <cobalt/uapi/sched.h> +#include <sched.h> +#include "event-parse.h" +#include "trace-seq.h" + +static void write_policy(struct trace_seq *p, int policy) +{ + trace_seq_printf(p, "policy="); + + switch (policy) { + case SCHED_QUOTA: + trace_seq_printf(p, "quota "); + break; + case SCHED_TP: + trace_seq_printf(p, "tp "); + break; + case SCHED_NORMAL: + trace_seq_printf(p, "normal "); + break; + case SCHED_SPORADIC: + trace_seq_printf(p, "sporadic "); + break; + case SCHED_RR: + trace_seq_printf(p, "rr "); + break; + case SCHED_FIFO: + trace_seq_printf(p, "fifo "); + break; + case SCHED_COBALT: + trace_seq_printf(p, "cobalt "); + break; + case SCHED_WEAK: + trace_seq_printf(p, "weak "); + break; + default: + trace_seq_printf(p, "unknown "); + break; + } +} + +/* save param */ +static void write_param(struct tep_format_field *field, + struct tep_record *record, + struct trace_seq *p, int policy) +{ + int offset; + struct sched_param_ex *params; + + offset = field->offset; + + if (!strncmp(field->type, "__data_loc", 10)) { + unsigned long long v; + + if (tep_read_number_field(field, record->data, &v)) { + trace_seq_printf(p, "invalid_data_loc"); + return; + } + offset = v & 0xffff; + + } + + params = (struct sched_param_ex *)((char *)record->data + offset); + + trace_seq_printf(p, "param: { "); + + switch (policy) { + case SCHED_QUOTA: + trace_seq_printf(p, "priority=%d, group=%d", + params->sched_priority, + params->sched_quota_group); + break; + case SCHED_TP: + trace_seq_printf(p, "priority=%d, partition=%d", + params->sched_priority, + params->sched_tp_partition); + break; + case SCHED_NORMAL: + break; + case SCHED_SPORADIC: + trace_seq_printf(p, "priority=%d, low_priority=%d, ", + params->sched_priority, + params->sched_ss_low_priority); + + trace_seq_printf(p, "budget=(%ld.%09ld), period=(%ld.%09ld), ", + params->sched_ss_init_budget.tv_sec, + params->sched_ss_init_budget.tv_nsec); + + trace_seq_printf(p, "maxrepl=%d", + params->sched_ss_max_repl); + break; + case SCHED_RR: + case SCHED_FIFO: + case SCHED_COBALT: + case SCHED_WEAK: + default: + trace_seq_printf(p, "priority=%d", params->sched_priority); + break; + } + trace_seq_printf(p, " }"); + trace_seq_putc(p, '\0'); + +} + +static int cobalt_schedparam_handler(struct trace_seq *s, + struct tep_record *record, + struct tep_event *event, void *context) +{ + struct tep_format_field *field; + unsigned long long val; + + if (tep_get_field_val(s, event, "pth", record, &val, 1)) + return trace_seq_putc(s, '!'); + trace_seq_puts(s, "pth: "); + trace_seq_printf(s, "0x%08llx ", val); + + if (tep_get_field_val(s, event, "policy", record, &val, 1) == 0) + write_policy(s, val); + + field = tep_find_field(event, "param_ex"); + if (field) + write_param(field, record, s, val); + + return 0; +} + +int TEP_PLUGIN_LOADER(struct tep_handle *tep) +{ + tep_register_event_handler(tep, -1, "cobalt_posix", "cobalt_pthread_setschedparam", + cobalt_schedparam_handler, NULL); + + tep_register_event_handler(tep, -1, "cobalt_posix", "cobalt_pthread_getschedparam", + cobalt_schedparam_handler, NULL); + + tep_register_event_handler(tep, -1, "cobalt_posix", "cobalt_pthread_create", + cobalt_schedparam_handler, NULL); + + return 0; +} + +void TEP_PLUGIN_UNLOADER(struct tep_handle *tep) +{ + tep_unregister_event_handler(tep, -1, "cobalt_posix", "cobalt_pthread_setschedparam", + cobalt_schedparam_handler, NULL); + + tep_unregister_event_handler(tep, -1, "cobalt_posix", "cobalt_pthread_getschedparam", + cobalt_schedparam_handler, NULL); + + tep_unregister_event_handler(tep, -1, "cobalt_posix", "cobalt_pthread_create", + cobalt_schedparam_handler, NULL); +} -- 2.34.1