On 2/21/20 2:52 PM, Linus Walleij wrote:

On Tue, Feb 11, 2020 at 2:59 PM Uwe Kleine-König [off-list ref] wrote:

quoted

All the irq related callbacks are called with the (raw) spinlock
desc->lock being held. So the lock here must be raw as well. Also irqs
were already disabled by the caller for the irq chip callbacks, so the
non-irq variants of spin_lock must be used there.

Fixes: be8c8facc707 ("gpio: new driver to work with a 8x12 siox")
Signed-off-by: Uwe Kleine-König <redacted>

Patch applied. Is this a regression so I should put it in fixes?
I put it for v5.7 for now but I can easily change that.

I don't care much. AFAIK siox is only used at Eckelmann and we will have
to deploy the fix there anyhow. Conceptually the conversion to raw is
only relevant for RT (please correct me if I'm wrong), but too early
enablement of irqs probably can yield bad races.

Best regards
Uwe