On Tue, Jun 11, 2019 at 8:39 PM Bjorn Andersson <bjorn.andersson@xxxxxxxxxx> wrote: > Last time we discussed this the _only_ offender was the loop issuing a > get_direction() on all descs towards the end of > gpiochip_add_data_with_key() I think that is still the only offender. We were a bit back and forth: adding that code, removing it and then adding it back again. In a way it is good that we detect it so users do not crash their kernels by asking for these GPIOs at runtime from userspace instead. It makes a lot of sense for us to ask for the initial direction for all pins, as they get registered as GPIOs, which by definition makes them available as such and we should be able to inspect them. "GPIOs" reserved by ACPI arguably are not GPIOs anymore since ACPI have dedicated them to a special purpose (no more "general purpose"). Yours, Linus Walleij
