@@ -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,
};
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 <liwei391@huawei.com> --- kernel/trace/trace_hwlat.c | 15 +++++++++++---- 1 file changed, 11 insertions(+), 4 deletions(-)