On Tue, Mar 31, 2015 at 6:56 PM, Andrew Bresticker <abrestic@xxxxxxxxxxxx> wrote: > Hi Paul, > > On Tue, Mar 31, 2015 at 1:10 AM, Paul Bolle <pebolle@xxxxxxxxxx> wrote: >> The patch adds a mismatch between the Kconfig symbol (a bool) and the >> code (which suggests that a modular build is also possible). > > Nearly all of the pinctrl drivers (with the exception of qcom and > intel) are like this. They use a bool Kconfig symbol but they are > written so that they could be built as a module in the future. There are many aspects to this and I have no strong opinion, it doesn't really disturb me either way as both are familiar ways of writing drivers be it modular or not. I think it's more disturbing that bool drivers have .remove() functions, and as pointed out elsewhere this is because the module can still be bound/unbound from sysfs even if compiled into the kernel and then that code path will actually be executed. And as I remarked again, elsewhere, that can be overcome by adding the .suppress_bind_attrs = true to struct device_driver inside the platform driver. So if going through all the hazzle to remove anything compiled-out for non-modules, also take the step above removing .remove() calls and setting .suppress_bind_attrs. I will maybe do so sometime myself for any bool drivers... Yours, Linus Walleij
