Revert "gpio: mpc8xxx: change the gpio interrupt flags."
authorRasmus Villemoes <linux@rasmusvillemoes.dk>
Fri, 2 Jul 2021 13:37:12 +0000 (15:37 +0200)
committerBartosz Golaszewski <bgolaszewski@baylibre.com>
Sun, 18 Jul 2021 12:52:30 +0000 (14:52 +0200)
This reverts commit 3d5bfbd9716318b1ca5c38488aa69f64d38a9aa5.

When booting with threadirqs, it causes a splat

  WARNING: CPU: 0 PID: 29 at kernel/irq/handle.c:159 __handle_irq_event_percpu+0x1ec/0x27c
  irq 66 handler irq_default_primary_handler+0x0/0x1c enabled interrupts

That splat later went away with commit 81e2073c175b ("genirq: Disable
interrupts for force threaded handlers"), which got backported to
-stable. However, when running an -rt kernel, the splat still
exists. Moreover, quoting Thomas Gleixner [1]

  But 3d5bfbd97163 ("gpio: mpc8xxx: change the gpio interrupt flags.")
  has nothing to do with that:

      "Delete the interrupt IRQF_NO_THREAD flags in order to gpio interrupts
       can be threaded to allow high-priority processes to preempt."

  This changelog is blatantly wrong. In mainline forced irq threads
  have always been invoked with softirqs disabled, which obviously
  makes them non-preemptible.

So the patch didn't even do what its commit log said.

[1] https://lore.kernel.org/lkml/871r8zey88.ffs@nanos.tec.linutronix.de/

Cc: stable@vger.kernel.org # v5.9+
Signed-off-by: Rasmus Villemoes <linux@rasmusvillemoes.dk>
Signed-off-by: Bartosz Golaszewski <bgolaszewski@baylibre.com>
drivers/gpio/gpio-mpc8xxx.c

index 4b9157a..50b321a 100644 (file)
@@ -405,7 +405,7 @@ static int mpc8xxx_probe(struct platform_device *pdev)
 
        ret = devm_request_irq(&pdev->dev, mpc8xxx_gc->irqn,
                               mpc8xxx_gpio_irq_cascade,
-                              IRQF_SHARED, "gpio-cascade",
+                              IRQF_NO_THREAD | IRQF_SHARED, "gpio-cascade",
                               mpc8xxx_gc);
        if (ret) {
                dev_err(&pdev->dev,