Hi, On 7/6/23 14:56, Andy Shevchenko wrote: > On Thu, Jul 06, 2023 at 02:47:36PM +0200, Hans de Goede wrote: >> On 7/6/23 14:42, Andy Shevchenko wrote: >>> On Thu, Jul 06, 2023 at 02:31:14PM +0200, Hans de Goede wrote: >>>> On 7/6/23 12:15, Andy Shevchenko wrote: >>>>> On Wed, Jul 05, 2023 at 11:30:10PM +0200, Hans de Goede wrote: >>> >>>>>> + vcm_type = kstrdup(obj->string.pointer, GFP_KERNEL); >>>>> >>>>> Where is the counterpart kfree()? >>>> >>>> The vcm-type is stored in one of the generated sw-nodes and the ipu-bridge >>>> code only creates those once and them leaves them in memory, even on >>>> a rmmod. So this is deliberately leaked just like that the ipu_bridge >>>> struct which contains all the swnode-s is deliberately leaked by >>>> ipu-bridge.c >>> >>> Should we worry about those leakages? >> >> No this is by design because removing the swnodes while e.g. a sensor >> driver might still be bound to the i2c-client is trouble-some and >> the callers of ipu_bridge_init check if it has already run and then >> skip calling it. >> >> So after a rmmod + modprobe of the atomisp / ipu3-cio2 driver >> ipu_bridge_init() will not get called a second time. Instead >> the old swnodes (1) which are already set as secondary fwnodes for >> the sensor and bridge devices are re-used. > > But this will be actual leak if we hot unplug/plug back the device, right? > (I think we can do that in some [debug?] cases). No, the kstrdup() is called from the parse_sensor_fwnode() callback passed to ipu_bridge_init(), so it will only happen once even on unbind + re-bind or rmmod + modprobe. Regards, Hans
