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 --- Changes in v6: - Use a global poll irq_work and wait_queue. --- include/linux/trace_events.h | 14 +++++++ kernel/trace/trace_events.c | 14 +++++++ kernel/trace/trace_events_hist.c | 75 ++++++++++++++++++++++++++++++++++++-- 3 files changed, 100 insertions(+), 3 deletions(-)

diff --git a/include/linux/trace_events.h b/include/linux/trace_events.h index 42bedcddd511..46c771a61f2a 100644 --- a/include/linux/trace_events.h +++ b/include/linux/trace_events.h @@ -685,6 +685,20 @@ struct trace_event_file { atomic_t tm_ref; /* trigger-mode reference counter */ };

+#ifdef CONFIG_HIST_TRIGGERS +extern struct irq_work hist_poll_work; +extern wait_queue_head_t hist_poll_wq; + +static inline void hist_poll_wakeup(void) +{ + if (wq_has_sleeper(&hist_poll_wq)) + irq_work_queue(&hist_poll_work); +} + +#define hist_poll_wait(file, wait) \ + poll_wait(file, &hist_poll_wq, wait) +#endif + #define __TRACE_EVENT_FLAGS(name, value) \ static int __init trace_init_flags_##name(void) \ { \ diff --git a/kernel/trace/trace_events.c b/kernel/trace/trace_events.c index 7266ec2a4eea..0a7cb30043ef 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_poll_wq from irq_work because the hist trigger + * may happen in any context. + */ +static void hist_poll_event_irq_work(struct irq_work *work) +{ + wake_up_all(&hist_poll_wq); +} + +DEFINE_IRQ_WORK(hist_poll_work, hist_poll_event_irq_work); +DECLARE_WAIT_QUEUE_HEAD(hist_poll_wq); +#endif + static struct trace_event_file * trace_create_new_event(struct trace_event_call *call, struct trace_array *tr) diff --git a/kernel/trace/trace_events_hist.c b/kernel/trace/trace_events_hist.c index 5f9119eb7c67..107eaa0f40f1 100644 --- a/kernel/trace/trace_events_hist.c +++ b/kernel/trace/trace_events_hist.c @@ -5314,6 +5314,8 @@ 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); + + hist_poll_wakeup(); }

static void hist_trigger_stacktrace_print(struct seq_file *m, @@ -5593,15 +5595,36 @@ 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 +5634,7 @@ 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 +5642,69 @@ 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; + } + + hist_poll_wait(file, 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

Since POLLIN will not be flashed until read the hist file, user needs to repeat read() and poll() on hist for monitoring the event continuously. But the read() is somewhat redundant only for monitoring events.

This add POLLPRI poll event on hist, this event returns when a histogram is updated after open(), poll() or read(). Thus it is possible to wait next event without read().

Signed-off-by: Masami Hiramatsu (Google) mhiramat@kernel.org Reviewed-by: Tom Zanussi zanussi@kernel.org --- kernel/trace/trace_events_hist.c | 29 +++++++++++++++++++++++++++-- 1 file changed, 27 insertions(+), 2 deletions(-)

diff --git a/kernel/trace/trace_events_hist.c b/kernel/trace/trace_events_hist.c index 107eaa0f40f1..8819a8cc4d53 100644 --- a/kernel/trace/trace_events_hist.c +++ b/kernel/trace/trace_events_hist.c @@ -5598,6 +5598,7 @@ static void hist_trigger_show(struct seq_file *m, struct hist_file_data { struct file *file; u64 last_read; + u64 last_act; };

static u64 get_hist_hit_count(struct trace_event_file *event_file) @@ -5635,6 +5636,11 @@ static int hist_show(struct seq_file *m, void *v) hist_trigger_show(m, data, n++); } hist_file->last_read = get_hist_hit_count(event_file); + /* + * Update last_act too so that poll()/POLLPRI can wait for the next + * event after any syscall on hist file. + */ + hist_file->last_act = hist_file->last_read;

out_unlock: mutex_unlock(&event_mutex); @@ -5648,6 +5654,7 @@ static __poll_t event_hist_poll(struct file *file, struct poll_table_struct *wai struct seq_file *m = file->private_data; struct hist_file_data *hist_file = m->private; __poll_t ret = 0; + u64 cnt;

mutex_lock(&event_mutex);

@@ -5659,8 +5666,13 @@ static __poll_t event_hist_poll(struct file *file, struct poll_table_struct *wai

hist_poll_wait(file, wait);

- if (hist_file->last_read != get_hist_hit_count(event_file)) - ret = EPOLLIN | EPOLLRDNORM; + cnt = get_hist_hit_count(event_file); + if (hist_file->last_read != cnt) + ret |= EPOLLIN | EPOLLRDNORM; + if (hist_file->last_act != cnt) { + hist_file->last_act = cnt; + ret |= EPOLLPRI; + }

out_unlock: mutex_unlock(&event_mutex); @@ -5679,6 +5691,7 @@ static int event_hist_release(struct inode *inode, struct file *file)

static int event_hist_open(struct inode *inode, struct file *file) { + struct trace_event_file *event_file; struct hist_file_data *hist_file; int ret;

@@ -5689,13 +5702,25 @@ static int event_hist_open(struct inode *inode, struct file *file) hist_file = kzalloc(sizeof(*hist_file), GFP_KERNEL); if (!hist_file) return -ENOMEM; + + mutex_lock(&event_mutex); + event_file = event_file_data(file); + if (!event_file) { + ret = -ENODEV; + goto out_unlock; + } + hist_file->file = file; + hist_file->last_act = get_hist_hit_count(event_file);

/* Clear private_data to avoid warning in single_open() */ file->private_data = NULL; ret = single_open(file, hist_show, hist_file); + +out_unlock: if (ret) kfree(hist_file); + mutex_unlock(&event_mutex); return ret; }

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 Reviewed-by: Shuah Khan skhan@linuxfoundation.org --- tools/testing/selftests/ftrace/Makefile | 2 + tools/testing/selftests/ftrace/poll.c | 74 ++++++++++++++++++++ .../ftrace/test.d/trigger/trigger-hist-poll.tc | 74 ++++++++++++++++++++ 3 files changed, 150 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..53258f7515e7 --- /dev/null +++ b/tools/testing/selftests/ftrace/poll.c @@ -0,0 +1,74 @@ +// 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 [-I|-P] [-t timeout] FILE + */ +int main(int argc, char *argv[]) +{ + struct pollfd pfd = {.events = POLLIN}; + char buf[BUFSIZE]; + int timeout = -1; + int ret, opt; + + while ((opt = getopt(argc, argv, "IPt:")) != -1) { + switch (opt) { + case 'I': + pfd.events = POLLIN; + break; + case 'P': + pfd.events = POLLPRI; + break; + case 't': + timeout = atoi(optarg); + break; + default: + fprintf(stderr, "Usage: %s [-I|-P] [-t timeout] FILE\n", + argv[0]); + return -1; + } + } + if (optind >= argc) { + fprintf(stderr, "Error: Polling file is not specified\n"); + return -1; + } + + pfd.fd = open(argv[optind], O_RDONLY); + if (pfd.fd < 0) { + fprintf(stderr, "failed to open %s", argv[optind]); + perror("open"); + return -1; + } + + /* 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 (ret == 0), exit code is 1 */ + 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..cbd01a71ecad --- /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} -I -t 1000 ${EVENT}/hist +ret=$? +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} -I -t 4000 ${EVENT}/hist +echo 0 > tracing_on + +if [ -d /proc/${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} -P -t 4000 ${EVENT}/hist +echo 0 > tracing_on + +if [ -d /proc/${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