[PATCH v3 1/2] hwmon: (lm90) Stop work before releasing hwmon device

[PATCH v3 1/2] hwmon: (lm90) Stop work before releasing hwmon device

[Guenter Roeck]

Sashiko reports:

In lm90_probe(), the devm action to cancel the alert_work and report_work
(lm90_restore_conf) is registered in lm90_init_client() before
devm_hwmon_device_register_with_info() is called.

Because devm executes cleanup actions in reverse order during module
unbind or probe failure, the hwmon device is unregistered and freed first.

If lm90_alert_work() or lm90_report_alarms() runs in the window between
the hwmon device being freed and the delayed works being cancelled,
lm90_update_alarms() will dereference the freed data->hwmon_dev here.

Fix the problem by canceling the workers separately after registering
the hwmon device and before registering the interrupt handler. This ensures
that the workers are canceled after interrupts are disabled and before
the hwmon device is released. Add "shutdown" flag to indicate that device
shutdown is in progress to prevent workers from being re-armed.

Fixes: f6d0775119fb9 ("hwmon: (lm90) Rework alarm/status handling")
Signed-off-by: Guenter Roeck <linux@roeck-us.net>
---
v3: Hold hwmon subsystem lock while setting ->shutdown flag
v2: Add 'shutdown' flag to prevent workers from being re-armed

 drivers/hwmon/lm90.c | 24 ++++++++++++++++++++----
 1 file changed, 20 insertions(+), 4 deletions(-)

diff --git a/drivers/hwmon/lm90.c b/drivers/hwmon/lm90.c
index 3c10a5066b53..c4a9dafff81d 100644
--- a/drivers/hwmon/lm90.c
+++ b/drivers/hwmon/lm90.c
@@ -736,6 +736,7 @@ struct lm90_data {
 	struct hwmon_chip_info chip;
 	struct delayed_work alert_work;
 	struct work_struct report_work;
+	bool shutdown;		/* true if shutting down */
 	bool valid;		/* true if register values are valid */
 	bool alarms_valid;	/* true if status register values are valid */
 	unsigned long last_updated; /* in jiffies */
@@ -1154,6 +1155,9 @@ static void lm90_report_alarms(struct work_struct *work)
 
 static int lm90_update_alarms_locked(struct lm90_data *data, bool force)
 {
+	if (data->shutdown)
+		return 0;
+
 	if (force || !data->alarms_valid ||
 	    time_after(jiffies, data->alarms_updated + msecs_to_jiffies(data->update_interval))) {
 		struct i2c_client *client = data->client;
@@ -2584,15 +2588,23 @@ static void lm90_restore_conf(void *_data)
 	struct lm90_data *data = _data;
 	struct i2c_client *client = data->client;
 
-	cancel_delayed_work_sync(&data->alert_work);
-	cancel_work_sync(&data->report_work);
-
 	/* Restore initial configuration */
 	if (data->flags & LM90_HAVE_CONVRATE)
 		lm90_write_convrate(data, data->convrate_orig);
 	lm90_write_reg(client, LM90_REG_CONFIG1, data->config_orig);
 }
 
+static void lm90_stop_work(void *_data)
+{
+	struct lm90_data *data = _data;
+
+	hwmon_lock(data->hwmon_dev);
+	data->shutdown = true;
+	hwmon_unlock(data->hwmon_dev);
+	cancel_delayed_work_sync(&data->alert_work);
+	cancel_work_sync(&data->report_work);
+}
+
 static int lm90_init_client(struct i2c_client *client, struct lm90_data *data)
 {
 	struct device_node *np = client->dev.of_node;
@@ -2902,6 +2914,10 @@ static int lm90_probe(struct i2c_client *client)
 
 	data->hwmon_dev = hwmon_dev;
 
+	err = devm_add_action_or_reset(&client->dev, lm90_stop_work, data);
+	if (err)
+		return err;
+
 	if (client->irq) {
 		dev_dbg(dev, "IRQ: %d\n", client->irq);
 		err = devm_request_threaded_irq(dev, client->irq,
@@ -2930,7 +2946,7 @@ static void lm90_alert(struct i2c_client *client, enum i2c_alert_protocol type,
 		 */
 		struct lm90_data *data = i2c_get_clientdata(client);
 
-		if ((data->flags & LM90_HAVE_BROKEN_ALERT) &&
+		if (!data->shutdown && (data->flags & LM90_HAVE_BROKEN_ALERT) &&
 		    (data->current_alarms & data->alert_alarms)) {
 			if (!(data->config & 0x80)) {
 				dev_dbg(&client->dev, "Disabling ALERT#\n");
-- 
2.45.2

[PATCH v3 2/2] hwmon: (lm90) Add lock protection to lm90_alert

[Guenter Roeck]

Sashiko reports:

lm90_alert() executes in the smbus alert context and calls
lm90_update_confreg() to disable the hardware alert line, without
acquiring hwmon_lock.

Concurrently, sysfs write operations (such as lm90_write_convrate) hold
the hwmon_lock, temporarily modify data->config, and then restore it.

If an alert interrupt occurs concurrently with a sysfs write, the sysfs
path will overwrite the alert handler's modifications to data->config
and the hardware register.

This unintentionally re-enables the hardware alert line while the alarm is
still active, causing an interrupt storm.

Add the missing lock to lm90_alert() to solve the problem.

Fixes: 7a1d220ccb0cc ("hwmon: (lm90) Introduce function to update configuration register")
Signed-off-by: Guenter Roeck <linux@roeck-us.net>
---
v3: No change
v2: No change

 drivers/hwmon/lm90.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/drivers/hwmon/lm90.c b/drivers/hwmon/lm90.c
index c4a9dafff81d..1eeb608e5903 100644
--- a/drivers/hwmon/lm90.c
+++ b/drivers/hwmon/lm90.c
@@ -2946,6 +2946,7 @@ static void lm90_alert(struct i2c_client *client, enum i2c_alert_protocol type,
 		 */
 		struct lm90_data *data = i2c_get_clientdata(client);
 
+		hwmon_lock(data->hwmon_dev);
 		if (!data->shutdown && (data->flags & LM90_HAVE_BROKEN_ALERT) &&
 		    (data->current_alarms & data->alert_alarms)) {
 			if (!(data->config & 0x80)) {
@@ -2955,6 +2956,7 @@ static void lm90_alert(struct i2c_client *client, enum i2c_alert_protocol type,
 			schedule_delayed_work(&data->alert_work,
 				max_t(int, HZ, msecs_to_jiffies(data->update_interval)));
 		}
+		hwmon_unlock(data->hwmon_dev);
 	} else {
 		dev_dbg(&client->dev, "Everything OK\n");
 	}
-- 
2.45.2