diff mbox series

[v1] gpio: pca953x: fix IRQ storm on system wake up

Message ID 20250326173838.4617-1-francesco@dolcini.it
State New
Headers show
Series [v1] gpio: pca953x: fix IRQ storm on system wake up | expand

Commit Message

Francesco Dolcini March 26, 2025, 5:38 p.m. UTC 
From: Emanuele Ghidoli <emanuele.ghidoli@toradex.com>

If an input changes state during wake-up and is used as an interrupt
source, the IRQ handler reads the volatile input register to clear the
interrupt mask and deassert the IRQ line. However, the IRQ handler is
triggered before access to the register is granted, causing the read
operation to fail.

As a result, the IRQ handler enters a loop, repeatedly printing the
"failed reading register" message, until `pca953x_resume` is eventually
called, which restores the driver context and enables access to
registers.

Fix by using DEFINE_NOIRQ_DEV_PM_OPS which ensures that `pca953x_resume`
is called before the IRQ handler is called.

Fixes: b76574300504 ("gpio: pca953x: Restore registers after suspend/resume cycle")
Cc: stable@vger.kernel.org
Signed-off-by: Emanuele Ghidoli <emanuele.ghidoli@toradex.com>
Signed-off-by: Francesco Dolcini <francesco.dolcini@toradex.com>
---
 drivers/gpio/gpio-pca953x.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

Comments

Bartosz Golaszewski April 3, 2025, 1:56 p.m. UTC | #1 
On Thu, Apr 3, 2025 at 3:54 PM Andy Shevchenko
<andriy.shevchenko@intel.com> wrote:
>
> +Cc: Geert
>
> On Thu, Apr 03, 2025 at 02:07:05PM +0200, Bartosz Golaszewski wrote:
> > On Wed, 26 Mar 2025 18:38:38 +0100, Francesco Dolcini wrote:
>
> > > If an input changes state during wake-up and is used as an interrupt
> > > source, the IRQ handler reads the volatile input register to clear the
> > > interrupt mask and deassert the IRQ line. However, the IRQ handler is
> > > triggered before access to the register is granted, causing the read
> > > operation to fail.
> > >
> > > As a result, the IRQ handler enters a loop, repeatedly printing the
> > > "failed reading register" message, until `pca953x_resume` is eventually
> > > called, which restores the driver context and enables access to
> > > registers.
>
> [...]
>
> > Applied, thanks!
>
> Won't this regress as it happens the last time [1]?
>
> [1]: https://lore.kernel.org/linux-gpio/CAMuHMdVnKX23yi7ir1LVxfXAMeeWMFzM+cdgSSTNjpn1OnC2xw@mail.gmail.com/
>

Ah, good catch. I'm wondering what the right fix here is but don't
really have any ideas at the moment. Any hints are appreciated.

For now, I'm dropping it.

Bart
diff mbox series

Patch

diff --git a/drivers/gpio/gpio-pca953x.c b/drivers/gpio/gpio-pca953x.c
index d63c1030e6ac..d39bdc125cfc 100644
--- a/drivers/gpio/gpio-pca953x.c
+++ b/drivers/gpio/gpio-pca953x.c
@@ -1252,7 +1252,7 @@  static int pca953x_resume(struct device *dev)
 	return ret;
 }
 
-static DEFINE_SIMPLE_DEV_PM_OPS(pca953x_pm_ops, pca953x_suspend, pca953x_resume);
+static DEFINE_NOIRQ_DEV_PM_OPS(pca953x_pm_ops, pca953x_suspend, pca953x_resume);
 
 /* convenience to stop overlong match-table lines */
 #define OF_653X(__nrgpio, __int) ((void *)(__nrgpio | PCAL653X_TYPE | __int))