On Sun, Apr 7, 2019 at 4:07 AM Jonathan Cameron <jic23@xxxxxxxxxx> wrote: > > On Wed, 3 Apr 2019 00:03:22 -0700 > Andrey Smirnov <andrew.smirnov@xxxxxxxxx> wrote: > > > Use devm_iio_device_register() and drop explicit call to > > iio_device_unregister(). > > > > Signed-off-by: Andrey Smirnov <andrew.smirnov@xxxxxxxxx> > > Cc: Jonathan Cameron <jic23@xxxxxxxxxx> > > Cc: Hartmut Knaack <knaack.h@xxxxxx> > > Cc: Lars-Peter Clausen <lars@xxxxxxxxxx> > > Cc: Peter Meerwald-Stadler <pmeerw@xxxxxxxxxx> > > Cc: Chris Healy <cphealy@xxxxxxxxx> > > Cc: linux-iio@xxxxxxxxxxxxxxx > > Cc: linux-kernel@xxxxxxxxxxxxxxx > No to this one. The thing to think about is the resulting order > of the unwinding that happens in remove. > > Do take a look at the code flow, but in short what happens is: > > 1. driver.remove() > 2. Devm release functions run in the opposite order to they were > called during setup. > > The upshot of the change you just made here is that we turn the power > off to the device before we remove the userspace interfaces, giving > potentially interesting failures.. > > There are two options to avoid this: > > 1. Make everything use devm_ calls (often using devm_add_action_or_reset > for the ones that don't have their own versions). > > 2. Stop using devm managed functions at the first non devm_ call that needs > unwinding. From that point onwards in probe / remove you have to do everything > manually. > Yeah, missed the ordering proble, sorry about that. I'll re-order changes such that this conversion happens last to avoid said problem in v2. Thanks, Andrey Smirnov
