On Tue, 13 Jul 2021 23:31:28 +0200 Pavel Machek pavel@denx.de wrote:
Hi!
commit 7cd04c863f9e1655d607705455e7714f24451984 upstream.
Allocating an IRQ is conditional to the IRQ existence, but freeing it was not. If no IRQ was allocate, the driver would still try to free IRQ 0. Add the missing checks.
This fixes the following trace when the driver is removed:
[ 100.667788] Trying to free already-free IRQ 0
AFAICT this will need more fixing, because IRQ == 0 is a valid IRQ. NO_IRQ (aka -1) should be safe to use for "no irq assigned".
I thought we had long put this to bed. IRQ == 0 is not a valid irq number. If there is an error in parsing DT (e.g. no irq specified) it returns 0 not NO_IRQ. Naturally irq chips can have a 0, but that's pre translation to virtual IRQ space.
Many years ago, ARM did have 0 as as valid IRQ, but that got cleaned up to make it easier to do generic code like this.
Jonathan
Best regards, Pavel
+++ b/drivers/iio/light/tcs3472.c @@ -531,7 +531,8 @@ static int tcs3472_probe(struct i2c_clie return 0; free_irq:
- free_irq(client->irq, indio_dev);
- if (client->irq)
free_irq(client->irq, indio_dev);buffer_cleanup: iio_triggered_buffer_cleanup(indio_dev); return ret;