On Tue, 9 Aug 2011 11:04:12 -0700, Guenter Roeck wrote: > On Tue, 2011-08-09 at 11:10 -0400, Julia Lawall wrote: > > From: Julia Lawall <julia@xxxxxxx> > > > > rs_resp is dynamically allocated in this function, so it should be freed > > before exiting in every case. This collects the kfree and the return at > > the end of the function. > > > > Signed-off-by: Julia Lawall <julia@xxxxxxx> > > Good catch. Applied. Real bug and good fix, however... This repeated kzalloc seems very bad from performance and memory fragmentation perspectives. Given that calls to aem_read_sensor() are serialized with &data->lock, it should be possible to allocate the buffer once at initialization time and always reuse it. I see that the buffer has variable size, but as far as I can see the maximum is known in advance (sizeof struct aem_read_sensor_resp + 8) so that should be a problem. It also seems to me that the error paths in functions aem_init_aem1_inst() and aem_init_aem2_inst() are incorrect. I'll send two patches for testing and review shortly. -- Jean Delvare _______________________________________________ lm-sensors mailing list lm-sensors@xxxxxxxxxxxxxx http://lists.lm-sensors.org/mailman/listinfo/lm-sensors
