[PATCH v2 2/2] tracing/hwlat: Fix deadlock in cpuhp processing
From: Wei Li <hidden>
Date: 2024-11-21 02:06:14
Subsystem:
the rest, tracing, tracing os noise / latency tracers · Maintainers:
Linus Torvalds, Steven Rostedt, Masami Hiramatsu
One "hung task" error was reported during the test, and i figured out
the deadlock scenario is as follows:
T1 [BP] | T2 [AP] | T3 [hwlatd/1] | T4
work_for_cpu_fn() | cpuhp_thread_fun() | kthread_fn() | hwlat_hotplug_workfn()
_cpu_down() | stop_cpu_kthread() | | mutex_lock(&hwlat_data.lock)
cpus_write_lock() | kthread_stop(hwlatd/1) | mutex_lock(&hwlat_data.lock) |
__cpuhp_kick_ap() | wait_for_completion() | | cpus_read_lock()
It constitutes ABBA deadlock indirectly between "cpu_hotplug_lock" and
"hwlat_data.lock", remove the mutex_lock in kthread_fn() to fix this.
Fixes: ba998f7d9531 ("trace/hwlat: Support hotplug operations")
Signed-off-by: Wei Li <redacted>
---
kernel/trace/trace_hwlat.c | 15 +++++++++++----
1 file changed, 11 insertions(+), 4 deletions(-)
diff --git a/kernel/trace/trace_hwlat.c b/kernel/trace/trace_hwlat.c
index 3bd6071441ad..7d50b5ea999a 100644
--- a/kernel/trace/trace_hwlat.c
+++ b/kernel/trace/trace_hwlat.c@@ -108,6 +108,7 @@ static struct hwlat_data { u64 sample_window; /* total sampling window (on+off) */ u64 sample_width; /* active sampling portion of window */ + u64 sample_interval; /* sample_window - sample_width */ int thread_mode; /* thread mode */
@@ -370,10 +371,7 @@ static int kthread_fn(void *data) get_sample(); local_irq_enable(); - mutex_lock(&hwlat_data.lock); - interval = hwlat_data.sample_window - hwlat_data.sample_width; - mutex_unlock(&hwlat_data.lock); - + interval = READ_ONCE(hwlat_data.sample_interval); do_div(interval, USEC_PER_MSEC); /* modifies interval value */ /* Always sleep for at least 1ms */
@@ -729,6 +727,13 @@ static ssize_t hwlat_mode_write(struct file *filp, const char __user *ubuf, return ret; } +static void hwlat_width_window_write(u64 *val, u64 data) +{ + *val = data; + WRITE_ONCE(hwlat_data.sample_interval, + hwlat_data.sample_window - hwlat_data.sample_width); +} + /* * The width parameter is read/write using the generic trace_min_max_param * method. The *val is protected by the hwlat_data lock and is upper
@@ -737,6 +742,7 @@ static ssize_t hwlat_mode_write(struct file *filp, const char __user *ubuf, static struct trace_min_max_param hwlat_width = { .lock = &hwlat_data.lock, .val = &hwlat_data.sample_width, + .write = &hwlat_width_window_write, .max = &hwlat_data.sample_window, .min = NULL, };
@@ -749,6 +755,7 @@ static struct trace_min_max_param hwlat_width = { static struct trace_min_max_param hwlat_window = { .lock = &hwlat_data.lock, .val = &hwlat_data.sample_window, + .write = &hwlat_width_window_write, .max = NULL, .min = &hwlat_data.sample_width, };
--
2.25.1