From: Masami Hiramatsu (Google) mhiramat@kernel.org

Add poll syscall support on the `hist` file. The Waiter will be waken up when the histogram is updated with POLLIN.

Currently, there is no way to wait for a specific event in userspace. So user needs to peek the `trace` periodicaly, or wait on `trace_pipe`. But that is not good idea to peek the `trace` for the event randomely happens. And `trace_pipe` is not coming back until a page is filled with events.

This allows user to wait for a specific events on `hist` file. User can set a histogram trigger on the event which they want to monitor. And poll() on its `hist` file. Since this poll() returns POLLIN, the next poll() will return soon unless you do read() on hist file.

NOTE: To read the hist file again, you must set the file offset to 0, but just for monitoring the event, you may not need to read the histogram.

Signed-off-by: Masami Hiramatsu (Google) mhiramat@kernel.org Reviewed-by: Tom Zanussi zanussi@kernel.org --- include/linux/trace_events.h | 5 +++ kernel/trace/trace_events.c | 18 +++++++++ kernel/trace/trace_events_hist.c | 76 +++++++++++++++++++++++++++++++++++++- 3 files changed, 96 insertions(+), 3 deletions(-)

diff --git a/include/linux/trace_events.h b/include/linux/trace_events.h index 42bedcddd511..f3ec67d34097 100644 --- a/include/linux/trace_events.h +++ b/include/linux/trace_events.h @@ -663,6 +663,11 @@ struct trace_event_file { struct trace_subsystem_dir *system; struct list_head triggers;

+#ifdef CONFIG_HIST_TRIGGERS + struct irq_work hist_work; + wait_queue_head_t hist_wq; +#endif + /* * 32 bit flags: * bit 0: enabled diff --git a/kernel/trace/trace_events.c b/kernel/trace/trace_events.c index 7266ec2a4eea..0f077b32eea4 100644 --- a/kernel/trace/trace_events.c +++ b/kernel/trace/trace_events.c @@ -2972,6 +2972,20 @@ static bool event_in_systems(struct trace_event_call *call, return !*p || isspace(*p) || *p == ','; }

+#ifdef CONFIG_HIST_TRIGGERS +/* + * Wake up waiter on the hist_wq from irq_work because the hist trigger + * may happen in any context. + */ +static void hist_event_irq_work(struct irq_work *work) +{ + struct trace_event_file *event_file; + + event_file = container_of(work, struct trace_event_file, hist_work); + wake_up_all(&event_file->hist_wq); +} +#endif + static struct trace_event_file * trace_create_new_event(struct trace_event_call *call, struct trace_array *tr) @@ -3004,6 +3018,10 @@ trace_create_new_event(struct trace_event_call *call, INIT_LIST_HEAD(&file->triggers); list_add(&file->list, &tr->events); refcount_set(&file->ref, 1); +#ifdef CONFIG_HIST_TRIGGERS + init_irq_work(&file->hist_work, hist_event_irq_work); + init_waitqueue_head(&file->hist_wq); +#endif

return file; } diff --git a/kernel/trace/trace_events_hist.c b/kernel/trace/trace_events_hist.c index 5f9119eb7c67..d27b60f54f68 100644 --- a/kernel/trace/trace_events_hist.c +++ b/kernel/trace/trace_events_hist.c @@ -5314,6 +5314,9 @@ static void event_hist_trigger(struct event_trigger_data *data,

if (resolve_var_refs(hist_data, key, var_ref_vals, true)) hist_trigger_actions(hist_data, elt, buffer, rec, rbe, key, var_ref_vals); + + if (hist_data->event_file && wq_has_sleeper(&hist_data->event_file->hist_wq)) + irq_work_queue(&hist_data->event_file->hist_work); }

static void hist_trigger_stacktrace_print(struct seq_file *m, @@ -5593,15 +5596,36 @@ static void hist_trigger_show(struct seq_file *m, n_entries, (u64)atomic64_read(&hist_data->map->drops)); }

+struct hist_file_data { + struct file *file; + u64 last_read; +}; + +static u64 get_hist_hit_count(struct trace_event_file *event_file) +{ + struct hist_trigger_data *hist_data; + struct event_trigger_data *data; + u64 ret = 0; + + list_for_each_entry(data, &event_file->triggers, list) { + if (data->cmd_ops->trigger_type == ETT_EVENT_HIST) { + hist_data = data->private_data; + ret += atomic64_read(&hist_data->map->hits); + } + } + return ret; +} + static int hist_show(struct seq_file *m, void *v) { + struct hist_file_data *hist_file = m->private; struct event_trigger_data *data; struct trace_event_file *event_file; int n = 0, ret = 0;

mutex_lock(&event_mutex);

- event_file = event_file_file(m->private); + event_file = event_file_file(hist_file->file); if (unlikely(!event_file)) { ret = -ENODEV; goto out_unlock; @@ -5611,6 +5635,7 @@ static int hist_show(struct seq_file *m, void *v) if (data->cmd_ops->trigger_type == ETT_EVENT_HIST) hist_trigger_show(m, data, n++); } + hist_file->last_read = get_hist_hit_count(event_file);

out_unlock: mutex_unlock(&event_mutex); @@ -5618,24 +5643,69 @@ static int hist_show(struct seq_file *m, void *v) return ret; }

+static __poll_t event_hist_poll(struct file *file, struct poll_table_struct *wait) +{ + struct trace_event_file *event_file; + struct seq_file *m = file->private_data; + struct hist_file_data *hist_file = m->private; + __poll_t ret = 0; + + mutex_lock(&event_mutex); + + event_file = event_file_data(file); + if (!event_file) { + ret = EPOLLERR; + goto out_unlock; + } + + poll_wait(file, &event_file->hist_wq, wait); + + if (hist_file->last_read != get_hist_hit_count(event_file)) + ret = EPOLLIN | EPOLLRDNORM; + +out_unlock: + mutex_unlock(&event_mutex); + + return ret; +} + +static int event_hist_release(struct inode *inode, struct file *file) +{ + struct seq_file *m = file->private_data; + struct hist_file_data *hist_file = m->private; + + kfree(hist_file); + return tracing_single_release_file_tr(inode, file); +} + static int event_hist_open(struct inode *inode, struct file *file) { + struct hist_file_data *hist_file; int ret;

ret = tracing_open_file_tr(inode, file); if (ret) return ret;

+ hist_file = kzalloc(sizeof(*hist_file), GFP_KERNEL); + if (!hist_file) + return -ENOMEM; + hist_file->file = file; + /* Clear private_data to avoid warning in single_open() */ file->private_data = NULL; - return single_open(file, hist_show, file); + ret = single_open(file, hist_show, hist_file); + if (ret) + kfree(hist_file); + return ret; }

const struct file_operations event_hist_fops = { .open = event_hist_open, .read = seq_read, .llseek = seq_lseek, - .release = tracing_single_release_file_tr, + .release = event_hist_release, + .poll = event_hist_poll, };

#ifdef CONFIG_HIST_TRIGGERS_DEBUG

From: Masami Hiramatsu (Google) mhiramat@kernel.org

Add a testcase for poll() on hist file. This introduces a helper binary to the ftracetest, because there is no good way to reliably execute poll() on hist file.

Signed-off-by: Masami Hiramatsu (Google) mhiramat@kernel.org --- Changes in v2: - Update poll command to support both of POLLIN and POLLPRI, and timeout. - Identify unsupported stable kernel if poll-in returns soon. - Test both of POLLIN and POLLPRI. --- tools/testing/selftests/ftrace/Makefile | 2 + tools/testing/selftests/ftrace/poll.c | 62 +++++++++++++++++ .../ftrace/test.d/trigger/trigger-hist-poll.tc | 74 ++++++++++++++++++++ 3 files changed, 138 insertions(+) create mode 100644 tools/testing/selftests/ftrace/poll.c create mode 100644 tools/testing/selftests/ftrace/test.d/trigger/trigger-hist-poll.tc

diff --git a/tools/testing/selftests/ftrace/Makefile b/tools/testing/selftests/ftrace/Makefile index a1e955d2de4c..49d96bb16355 100644 --- a/tools/testing/selftests/ftrace/Makefile +++ b/tools/testing/selftests/ftrace/Makefile @@ -6,4 +6,6 @@ TEST_PROGS := ftracetest-ktap TEST_FILES := test.d settings EXTRA_CLEAN := $(OUTPUT)/logs/*

+TEST_GEN_PROGS = poll + include ../lib.mk diff --git a/tools/testing/selftests/ftrace/poll.c b/tools/testing/selftests/ftrace/poll.c new file mode 100644 index 000000000000..8003a59fe042 --- /dev/null +++ b/tools/testing/selftests/ftrace/poll.c @@ -0,0 +1,62 @@ +// SPDX-License-Identifier: GPL-2.0 +/* + * Simple poll on a file. + * + * Copyright (c) 2024 Google LLC. + */ + +#include <errno.h> +#include <fcntl.h> +#include <poll.h> +#include <stdio.h> +#include <stdlib.h> +#include <string.h> +#include <unistd.h> + +#define BUFSIZE 4096 + +/* + * Usage: + * poll <in|pri> <FILE> [timeout] + */ +int main(int argc, char *argv[]) +{ + struct pollfd pfd; + char buf[BUFSIZE]; + int timeout = -1; + int ret; + + if (argc < 3) + return -1; + + if (!strcmp(argv[1], "in")) + pfd.events = POLLIN; + else if (!strcmp(argv[1], "pri")) + pfd.events = POLLPRI; + + pfd.fd = open(argv[2], O_RDONLY); + if (pfd.fd < 0) { + perror("open"); + return -1; + } + + if (argc == 4) + timeout = atoi(argv[3]); + + /* Reset poll by read if POLLIN is specified. */ + if (pfd.events & POLLIN) + do {} while (read(pfd.fd, buf, BUFSIZE) == BUFSIZE); + + ret = poll(&pfd, 1, timeout); + if (ret < 0 && errno != EINTR) { + perror("poll"); + return -1; + } + close(pfd.fd); + + /* If timeout happned, return code is 0 */ + if (ret == 0) + return 1; + + return 0; +} diff --git a/tools/testing/selftests/ftrace/test.d/trigger/trigger-hist-poll.tc b/tools/testing/selftests/ftrace/test.d/trigger/trigger-hist-poll.tc new file mode 100644 index 000000000000..53bea74e2234 --- /dev/null +++ b/tools/testing/selftests/ftrace/test.d/trigger/trigger-hist-poll.tc @@ -0,0 +1,74 @@ +#!/bin/sh +# SPDX-License-Identifier: GPL-2.0 +# description: event trigger - test poll wait on histogram +# requires: set_event events/sched/sched_process_free/trigger events/sched/sched_process_free/hist +# flags: instance + +POLL=${FTRACETEST_ROOT}/poll + +if [ ! -x ${POLL} ]; then + echo "poll program is not compiled!" + exit_unresolved +fi + +EVENT=events/sched/sched_process_free/ + +# Check poll ops is supported. Before implementing poll on hist file, it +# returns soon with POLLIN | POLLOUT, but not POLLPRI. + +# This must wait >1 sec and return 1 (timeout). +set +e +${POLL} in ${EVENT}/hist 1000 +ret=$? +set -e +if [ ${ret} != 1 ]; then + echo "poll on hist file is not supported" + exit_unsupported +fi + +# Test POLLIN +echo > trace +echo "hist:key=comm" > ${EVENT}/trigger +echo 1 > ${EVENT}/enable + +# This sleep command will exit after 2 seconds. +sleep 2 & +BGPID=$! +# if timeout happens, poll returns 1. +${POLL} in ${EVENT}/hist 4000 +echo 0 > tracing_on + +if [ -d /proc/${BGPID} ]; then + echo "poll exits too soon" + kill -KILL ${BGPID} ||: + exit_fail +fi + +if ! grep -qw "sleep" trace; then + echo "poll exits before event happens" + exit_fail +fi + +# Test POLLPRI +echo > trace +echo 1 > tracing_on + +# This sleep command will exit after 2 seconds. +sleep 2 & +BGPID=$! +# if timeout happens, poll returns 1. +${POLL} pri ${EVENT}/hist 4000 +echo 0 > tracing_on + +if [ -d /proc/${BGPID} ]; then + echo "poll exits too soon" + kill -KILL ${BGPID} ||: + exit_fail +fi + +if ! grep -qw "sleep" trace; then + echo "poll exits before event happens" + exit_fail +fi + +exit_pass