On Wed, Jan 08, 2020 at 01:26:09PM +0200, Andy Shevchenko wrote: > On Wed, Jan 08, 2020 at 04:46:02PM +0800, Yingjoe Chen wrote: > > On Wed, 2020-01-08 at 14:29 +0800, fengping yu wrote: > > > > + tasklet_init(&keypad->tasklet, kpd_keymap_handler, > > > + (unsigned long)keypad); > > > + > > > + writew((u16)(keypad->key_debounce & KPD_DEBOUNCE_MASK), > > > + keypad->base + KP_DEBOUNCE); > > > > You use a 13 bits mask and set it directly to KP_DEBOUNCE register. Are > > you sure the debounce unit is ms? > > > > > + > > > + /* register IRQ */ > > > + err = request_irq(keypad->irqnr, kpd_irq_handler, IRQF_TRIGGER_NONE, > > > + KPD_NAME, keypad); > > > > please consider using devm_request_irq, otherwise you have to free it in > > _remove function. > > No, you may not use devm_*_irq() when tasklets are in use. There is a nasty > race condition. > > Actually the rule of thumb is to NOT use devm_*_irq() unless you exactly know > what you are doing. > > P.S. Why simple not to switch to threaded IRQ handler and drop tasklet? In such > case devm_*_irq() is fine. Actually, we are simply reading iomem data and forward it to input subsystem, there is no need to use threaded interrupt nor the tasklet. Thanks. -- Dmitry
