When an interrupt controller uses a function such as handle_level_irq()
as an interrupt handler and the controller implements the irq_disable()
callback, the following scenario will appear in the i2c-hid driver in
the sleep scenario:
in the sleep flow, while the user is still triggering the i2c-hid
interrupt, we get the following function call:
handle_level_irq()
-> mask_ack_irq()
-> mask_irq()
i2c_hid_core_suspend()
-> disable_irq()
-> __irq_disable()
-> irq_state_set_disabled()
-> irq_state_set_masked()
irq_thread_fn()
-> irq_finalize_oneshot()
-> if (!desc->threads_oneshot && !irqd_irq_disabled() &&
irqd_irq_masked())
unmask_threaded_irq()
-> unmask_irq()
That is, when __irq_disable() is called between mask_irq() and
irq_finalize_oneshot(), the code in irq_finalize_oneshot() will cause
the !irqd_irq_disabled() fails to enter the unmask_irq() branch, which
causes mask_irq/unmask_irq to be called unpaired and the i2c-hid
interrupt to be masked.
Since mask_irq/unmask_irq and irq_disabled() belong to two different
hardware registers or policies, the !irqd_irq_disabled() assertion may
not be used to determine whether unmask_irq() needs to be called.
Cc: stable@xxxxxxxxxxxxxxx
Signed-off-by: xiongxin <xiongxin@xxxxxxxxxx>
Signed-off-by: Riwen Lu <luriwen@xxxxxxxxxx>
diff --git a/kernel/irq/manage.c b/kernel/irq/manage.c
index 1782f90cd8c6..9160fc9170b3 100644
--- a/kernel/irq/manage.c
+++ b/kernel/irq/manage.c
@@ -1120,8 +1120,7 @@ static void irq_finalize_oneshot(struct irq_desc *desc,
desc->threads_oneshot &= ~action->thread_mask;
- if (!desc->threads_oneshot && !irqd_irq_disabled(&desc->irq_data) &&
- irqd_irq_masked(&desc->irq_data))
+ if (!desc->threads_oneshot && irqd_irq_masked(&desc->irq_data))
unmask_threaded_irq(desc);
out_unlock:
--
2.34.1
