On Fri, May 06, 2016 at 10:49:25PM +0100, Colin Pitrat wrote:
> This fixes the issue descirbe in bug 117531.
Can you include the oops log in the changelog as well?
> It's a regression introduced in linux 4.5 that causes a Oops at load of
> gpio_sch and prevents powering off the computer.
>
> The patch consist in reverting commit
> 737c8fccf1c5b2aae3c6d9a66dce17e35fc39b71
> (a.k.a 'gpio: sch: use gpiochip data pointer') that causes this regression.
> However, although it does work for me, I'm not sure of the impact of reverting
> only this part of the patch.
What happens here is that after the change we expect that
gpiochip_get_data() returns the correct pointer but that only happens
after devm_gpiochip_add_data() is called, which is the last call in
sch_gpio_probe(). Before that we call sch_gpio_reg_set() couple of times
if the device id is PCI_DEVICE_ID_INTEL_SCH_LPC and that causes NULL
pointer dereference.
I think better fix is to make sch_gpio_reg_get() and sch_gpio_reg_set()
take pointer to struct sch_gpio and use that directly like:
static void sch_gpio_reg_set(struct sch_gpio *sch, unsigned gpio,
unsigned reg, int val)
{
...
}
Neither function actually need gc for anything anyway.
