Re: [RFC] Fix shared irq trigger-flags conflict when old irqaction uses IRQF_TRIGGER_NONE

From: Hans de Goede
Date: Mon Apr 10 2017 - 07:06:02 EST


HI,

On 10-04-17 12:13, Marc Zyngier wrote:
Hi Hans,

On 09/04/17 20:59, Hans de Goede wrote:
While writing a driver for the INT0002 ACPI device found on Intel
Bay and Cherry Trail devices I hit the following error:

"genirq: Flags mismatch irq 9. 00000084 (INT0002) vs. 00000080 (acpi)"

This is caused by drivers/acpi/osl.c first doing:

request_irq(irq, acpi_irq, IRQF_SHARED, "acpi", acpi_irq)

While the irqdata for the irq contains no trigger flags, resulting
in an irqaction with IRQF_TRIGGER_NONE.

And then the INT0002 driver I'm working on calling platform_get_irq
which does: irqd_set_trigger_type(irqd, r->flags & IORESOURCE_BITS);

And then request_irq(irq, ..., IRQF_SHARED, ...) on the irq returned
by platform_get_irq causes the error quoted above.

Arguably the genirq code should not hit the shared irq trigger-flags
mismatch code if the old irqaction has IRQF_TRIGGER_NONE as flags.

This patch is an attempt at fixing this, but I'm not sure it is the
right fix, hence it RFC status.

Signed-off-by: Hans de Goede <hdegoede@xxxxxxxxxx>
---
kernel/irq/manage.c | 7 ++++++-
1 file changed, 6 insertions(+), 1 deletion(-)

diff --git a/kernel/irq/manage.c b/kernel/irq/manage.c
index a4afe5c..24e5eef 100644
--- a/kernel/irq/manage.c
+++ b/kernel/irq/manage.c
@@ -1212,8 +1212,13 @@ __setup_irq(unsigned int irq, struct irq_desc *desc, struct irqaction *new)
* set the trigger type must match. Also all must
* agree on ONESHOT.
*/
+ unsigned int old_msk = old->flags & IRQF_TRIGGER_MASK;
+
+ if (!old_msk)
+ old_msk = irqd_get_trigger_type(&desc->irq_data);
+
if (!((old->flags & new->flags) & IRQF_SHARED) ||
- ((old->flags ^ new->flags) & IRQF_TRIGGER_MASK) ||
+ ((old_msk ^ new->flags) & IRQF_TRIGGER_MASK) ||
((old->flags ^ new->flags) & IRQF_ONESHOT))
goto mismatch;



I'm afraid you're just papering over the issue here, as you leave the
"old" descriptor in an inconsistent state w.r.t. the "new" descriptor.

My view is that the old irq_desc should be "upgraded" to the new trigger
configuration, because they are sharing a line and must have compatible
behaviours.

There is no old irq_desc only an old ircaction, but yes that can easily
be updated to reflect the mask from the irq_data, or maybe we should not use
the oldaction for the trigger_mask bits at all and instead use
the trigger_mask from the irq_data as done in:

https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/kernel/irq/manage.c?id=7ee7e87dfb158e79019ea1d5ea1b0e6f2bc93ee4

For a different code path ? Thomas?

NONE is effectively a wildcard, and the second interrupt
request should turn this wildcard into the real thing.
>
The opposite case also exists (request a LEVEL interrupt first, then a
NONE), and should be resolved the same way (NONE becomes LEVEL).

The 2nd order of requesting irqs will already work because of:

https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/kernel/irq/manage.c?id=4b357daed698c95d6b5eacc1c3c4afa206071ba2

Where the new_action trigger_mask gets filled with the triger_mask
from the irq_data, which is a further hint that looking at
irqd_get_trigger_type(&desc->irq_data) rather then at
(old->flags IRQF_TRIGGER_MASK) is probably the right fix.

Note btw that 4b357daed698 is (part of) what is breaking things for
my use-case, I request the irq with IRQF_TRIGGER_NONE but
4b357daed698 modifies that before comparing the new trigger flags
to the old.

Regards,

Hans