On Freedom boards, GPIO12 (WPS button) pad leakage lets the pin
linger near the switching threshold during cold boot, so the mainline
gpio-keys edge IRQ (IRQ36) storms at hundreds of thousands to millions
of edges per second, all pinned to CPU0.
CPU0 gets wedged in hardirq context, stops reporting RCU quiescent states,
and the resulting rcu_preempt stall blocks any driver bring-up that waits on synchronize_rcu(),
stalling i2c-mux/expander/LED probe and locking the board up.
Move the WPS button to a separate gpio-keys-polled node so it no longer requests an edge IRQ,
decoupling it from the pad leakage entirely.
Fixes: PCF-2655
Signed-off-by: Tim Hsu <tim.hsu@wnc.com.tw>
Signed-off-by: Petr Štetiar <petr.stetiar@prplfoundation.org> [rebased, removed PCI fix]
6d3667850a