On Fri, Mar 30, 2018 at 08:32:56AM +0200, Julia Lawall wrote: > --- a/drivers/gpio/gpiolib.c > +++ b/drivers/gpio/gpiolib.c > @@ -2774,8 +2773,7 @@ int gpiod_get_array_value_complex(bool r > trace_gpio_value(desc_to_gpio(desc), 1, value); > } > > - if (slowpath) > - kfree(slowpath); > + kfree(slowpath); > } > return 0; > } > @@ -3020,8 +3018,7 @@ int gpiod_set_array_value_complex(bool r > if (count != 0) > gpio_chip_set_multiple(chip, mask, bits); > > - if (slowpath) > - kfree(slowpath); > + kfree(slowpath); > } > return 0; > } The problem I see here is that kfree may not be in L1 cache, and in that case checking for non-NULL locally in this function should actually be cheaper. Note that kfree() need only be called in the slowpath, which is the *unlikely* case. Letting the branch predictor assume that kfree() is not called is the right thing to do here. The function is a hot path, on the Revolution Pi open source PLCs we're calling it every 250 usec to poll digital inputs and update digital outputs. Would "if (unlikely(slowpath))" be sufficient to make coccinelle happy? That's what I'd suggest then. Otherwise "if (unlikely(chip->ngpio > FASTPATH_NGPIO))" could be used, though that might be minimally slower due to the pointer chasing. > @@ -2758,8 +2758,7 @@ int gpiod_get_array_value_complex(bool r > > ret = gpio_chip_get_multiple(chip, mask, bits); > if (ret) { > - if (slowpath) > - kfree(slowpath); > + kfree(slowpath); > return ret; > } > This particular change on the other hand is fine because the kfree() is occurring in an error path, which we'll normally not enter anyway. Thanks, Lukas -- To unsubscribe from this list: send the line "unsubscribe linux-gpio" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html