The fwnode_irq_get() and the fwnode_irq_get_byname() may have returned
zero if mapping the IRQ fails. This contradicts the
fwnode_irq_get_byname() documentation. Furthermore, returning zero or
errno on error is unepected and can easily lead to problems
like:
int probe(foo)
{
...
ret = fwnode_irq_get_byname(...);
if (ret < 0)
return ret;
...
}
or
int probe(foo)
{
...
ret = fwnode_irq_get_byname(...);
if (ret <= 0)
return ret;
...
}
which are both likely to be wrong. First treats zero as successful call and
misses the IRQ mapping failure. Second returns zero from probe even though
it detects the IRQ mapping failure correvtly.
Here we change the fwnode_irq_get() and the fwnode_irq_get_byname() to
always return a negative errno upon failure.
I have audited following callers (v6.4-rc2):
fwnode_irq_get_byname():
drivers/i2c/i2c-smbus.c
drivers/iio/accel/adxl355_core.c
drivers/iio/accel/kionix-kx022a.c
drivers/iio/adc/ad4130.c
drivers/iio/adc/max11410.c
drivers/iio/addac/ad74115.c
drivers/iio/gyro/fxas21002c_core.c
drivers/iio/imu/adis16480.c
drivers/iio/imu/bmi160/bmi160_core.c
drivers/iio/imu/bmi160/bmi160_core.c
fwnode_irq_get():
drivers/gpio/gpio-dwapb.c
drivers/iio/chemical/scd30_serial.c
drivers/iio/proximity/mb1232.c
drivers/net/ethernet/marvell/mvpp2/mvpp2_main.c
drivers/net/mdio/fwnode_mdio.c
drivers/pinctrl/pinctrl-ingenic.c
drivers/pinctrl/pinctrl-microchip-sgpio.c
drivers/pinctrl/pinctrl-pistachio.c
and it seems to me these calls will be Ok after the change. The
i2c-smbus.c and kionix-kx022a.c will gain a functional change (bugfix?) as
after this patch the probe will return -EINVAL should the IRQ mapping fail.
The series will also adjust the return value check for zero to be omitted.
NOTES:
Changes are compile-tested only.
drivers/pinctrl/nuvoton/pinctrl-wpcm450.c
will also gain a functional change. The pinctrl-wpcm450.c change is easy
to see - after this series the device-tree mapping failures will be
handled as any other errors - probe will be aborted with -EINVAL. Other
feasible option could be treating other errors in IRQ getting same way
as the DT mapping failures - just silently skip the IRQ. Please see
comment in the respective patch.
drivers/iio/cdc/ad7150.c
will gain functional change as well. Here the logic is less
straightforward but boils down to the same question as with the
pinctrl-wpcm450.c. Should all the IRQ getting errors jump to same
'no-IRQ' branch as the DT mapping error, or should the DT mapping error
abort the probe with error same way as other IRQ getting failures do?
Revision history:
v4 => v5:
- Fix subject lines for mvpp2 and wpcm450
- drop unnecessary irqno assignment from mb1232
- add back the drivers/i2c/i2c-smbus.c change which was accidentally
dropped during v3 => v4 work
v3 => v4:
- Change also the fwnode_irq_get() as was suggested by Jonathan.
Changelog v2 => v3:
- rebase/resend/add kx022a fix.
Changelog v1 => v2:
- minor styling
---
Matti Vaittinen (8):
drivers: fwnode: fix fwnode_irq_get[_byname]()
iio: mb1232: relax return value check for IRQ get
net-next: mvpp2: relax return value check for IRQ get
pinctrl: wpcm450: relax return value check for IRQ get
pinctrl: ingenic: relax return value check for IRQ get
pinctrl: pistachio: relax return value check for IRQ get
iio: cdc: ad7150: Functional change
i2c: i2c-smbus: fwnode_irq_get_byname() return value fix
drivers/base/property.c | 12 +++++++++---
drivers/i2c/i2c-smbus.c | 2 +-
drivers/iio/cdc/ad7150.c | 3 +--
drivers/iio/proximity/mb1232.c | 7 ++-----
drivers/net/ethernet/marvell/mvpp2/mvpp2_main.c | 4 ++--
drivers/pinctrl/nuvoton/pinctrl-wpcm450.c | 2 --
drivers/pinctrl/pinctrl-ingenic.c | 2 --
drivers/pinctrl/pinctrl-pistachio.c | 6 ------
8 files changed, 15 insertions(+), 23 deletions(-)
base-commit: f1fcbaa18b28dec10281551dfe6ed3a3ed80e3d6
--
2.40.1
--
Matti Vaittinen, Linux device drivers
ROHM Semiconductors, Finland SWDC
Kiviharjunlenkki 1E
90220 OULU
FINLAND
~~~ "I don't think so," said Rene Descartes. Just then he vanished ~~~
Simon says - in Latin please.
~~~ "non cogito me" dixit Rene Descarte, deinde evanescavit ~~~
Thanks to Simon Glass for the translation =]
fwnode_irq_get[_byname]() were changed to not return 0 anymore.
Drop check for return value 0.
Signed-off-by: Matti Vaittinen <[email protected]>
Acked-by: Linus Walleij <[email protected]>
---
Please note, I took Linus' reply to v4 cover-letter as ack && added the
tag. Please let me know if this was not Ok.
The first patch of the series changes the fwnode_irq_get() so this depends
on the first patch of the series and should not be applied alone.
---
drivers/pinctrl/pinctrl-pistachio.c | 6 ------
1 file changed, 6 deletions(-)
diff --git a/drivers/pinctrl/pinctrl-pistachio.c b/drivers/pinctrl/pinctrl-pistachio.c
index 53408344927a..8c50e0091b32 100644
--- a/drivers/pinctrl/pinctrl-pistachio.c
+++ b/drivers/pinctrl/pinctrl-pistachio.c
@@ -1393,12 +1393,6 @@ static int pistachio_gpio_register(struct pistachio_pinctrl *pctl)
dev_err(pctl->dev, "Failed to retrieve IRQ for bank %u\n", i);
goto err;
}
- if (!ret) {
- fwnode_handle_put(child);
- dev_err(pctl->dev, "No IRQ for bank %u\n", i);
- ret = -EINVAL;
- goto err;
- }
irq = ret;
bank = &pctl->gpio_banks[i];
--
2.40.1
--
Matti Vaittinen, Linux device drivers
ROHM Semiconductors, Finland SWDC
Kiviharjunlenkki 1E
90220 OULU
FINLAND
~~~ "I don't think so," said Rene Descartes. Just then he vanished ~~~
Simon says - in Latin please.
~~~ "non cogito me" dixit Rene Descarte, deinde evanescavit ~~~
Thanks to Simon Glass for the translation =]
The fwnode_irq_get_byname() was changed to not return 0 upon failure so
return value check can be adjusted to reflect the change.
Reviewed-by: Sakari Ailus <[email protected]>
Reviewed-by: Andy Shevchenko <[email protected]>
Signed-off-by: Matti Vaittinen <[email protected]>
---
Revision history:
v4 => v5:
- Added back after this was accidentally dropped at v4.
Depends on the mentioned return value change which is in patch 1/2. The
return value change does also cause a functional change here. Eg. when
IRQ mapping fails, the fwnode_irq_get_byname() no longer returns zero.
This will cause also the probe here to return nonzero failure. I guess
this is desired behaviour - but I would appreciate any confirmation.
Please, see also previous discussion here:
https://lore.kernel.org/all/fbd52f5f5253b382b8d7b3e8046134de29f965b8.1666710197.git.mazziesaccount@gmail.com/
Another suggestion has been to drop the check altogether. I am slightly
reluctant on doing that unless it gets confirmed that is the "right
thing to do".
---
drivers/i2c/i2c-smbus.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/drivers/i2c/i2c-smbus.c b/drivers/i2c/i2c-smbus.c
index 138c3f5e0093..893fe7cd3e41 100644
--- a/drivers/i2c/i2c-smbus.c
+++ b/drivers/i2c/i2c-smbus.c
@@ -129,7 +129,7 @@ static int smbalert_probe(struct i2c_client *ara)
} else {
irq = fwnode_irq_get_byname(dev_fwnode(adapter->dev.parent),
"smbus_alert");
- if (irq <= 0)
+ if (irq < 0)
return irq;
}
--
2.40.1
--
Matti Vaittinen, Linux device drivers
ROHM Semiconductors, Finland SWDC
Kiviharjunlenkki 1E
90220 OULU
FINLAND
~~~ "I don't think so," said Rene Descartes. Just then he vanished ~~~
Simon says - in Latin please.
~~~ "non cogito me" dixit Rene Descarte, deinde evanescavit ~~~
Thanks to Simon Glass for the translation =]
fwnode_irq_get[_byname]() were changed to not return 0 anymore.
Drop check for return value 0.
Signed-off-by: Matti Vaittinen <[email protected]>
---
Revision history:
v4 = v5:
Fix the subject, mb1232 => mvpp2
The first patch of the series changes the fwnode_irq_get() so this depends
on the first patch of the series and should not be applied alone.
---
drivers/net/ethernet/marvell/mvpp2/mvpp2_main.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/drivers/net/ethernet/marvell/mvpp2/mvpp2_main.c b/drivers/net/ethernet/marvell/mvpp2/mvpp2_main.c
index adc953611913..5b987af306a5 100644
--- a/drivers/net/ethernet/marvell/mvpp2/mvpp2_main.c
+++ b/drivers/net/ethernet/marvell/mvpp2/mvpp2_main.c
@@ -5833,7 +5833,7 @@ static int mvpp2_multi_queue_vectors_init(struct mvpp2_port *port,
v->irq = of_irq_get_byname(port_node, irqname);
else
v->irq = fwnode_irq_get(port->fwnode, i);
- if (v->irq <= 0) {
+ if (v->irq < 0) {
ret = -EINVAL;
goto err;
}
@@ -6764,7 +6764,7 @@ static int mvpp2_port_probe(struct platform_device *pdev,
err = -EPROBE_DEFER;
goto err_deinit_qvecs;
}
- if (port->port_irq <= 0)
+ if (port->port_irq < 0)
/* the link irq is optional */
port->port_irq = 0;
--
2.40.1
--
Matti Vaittinen, Linux device drivers
ROHM Semiconductors, Finland SWDC
Kiviharjunlenkki 1E
90220 OULU
FINLAND
~~~ "I don't think so," said Rene Descartes. Just then he vanished ~~~
Simon says - in Latin please.
~~~ "non cogito me" dixit Rene Descarte, deinde evanescavit ~~~
Thanks to Simon Glass for the translation =]
fwnode_irq_get[_byname]() were changed to not return 0 anymore. The
special error case where device-tree based IRQ mapping fails can't no
longer be reliably detected from this return value. This yields a
functional change in the driver where the mapping failure is treated as
an error.
The mapping failure can occur for example when the device-tree IRQ
information translation call-back(s) (xlate) fail, IRQ domain is not
found, IRQ type conflicts, etc. In most cases this indicates an error in
the device-tree and special handling is not really required.
One more thing to note is that ACPI APIs do not return zero for any
failures so this special handling did only apply on device-tree based
systems.
Drop the special handling for DT mapping failures as these can no longer
be separated from other errors at driver side.
Signed-off-by: Matti Vaittinen <[email protected]>
---
Please note that I don't have the hardware to test this change.
Furthermore, testing this type of device-tree error cases is not
trivial, as the question we probably dive in is "what happens with the
existing users who have errors in the device-tree". Answering to this
question is not simple.
I did this patch with minimal code changes - but a question is if we
should really jump into the else branch below on all IRQ getting errors?
} else {
indio_dev->info = &ad7150_info_no_irq;
switch (id->driver_data) {
case AD7150:
indio_dev->channels = ad7150_channels_no_irq;
indio_dev->num_channels =
ARRAY_SIZE(ad7150_channels_no_irq);
break;
case AD7151:
indio_dev->channels = ad7151_channels_no_irq;
indio_dev->num_channels =
ARRAY_SIZE(ad7151_channels_no_irq);
break;
default:
return -EINVAL;
}
Why do we have special handling for !chip->interrupts[0] while other
errors on getting the fwnode_irq_get(dev_fwnode(&client->dev), 0); will
abort the probe?
The first patch of the series changes the fwnode_irq_get() so this depends
on the first patch of the series and should not be applied alone.
---
drivers/iio/cdc/ad7150.c | 3 +--
1 file changed, 1 insertion(+), 2 deletions(-)
diff --git a/drivers/iio/cdc/ad7150.c b/drivers/iio/cdc/ad7150.c
index 79aeb0aaea67..d7ba50b9780d 100644
--- a/drivers/iio/cdc/ad7150.c
+++ b/drivers/iio/cdc/ad7150.c
@@ -567,8 +567,7 @@ static int ad7150_probe(struct i2c_client *client)
if (chip->interrupts[1] < 0)
return chip->interrupts[1];
}
- if (chip->interrupts[0] &&
- (id->driver_data == AD7151 || chip->interrupts[1])) {
+ if (id->driver_data == AD7151 || chip->interrupts[1]) {
irq_set_status_flags(chip->interrupts[0], IRQ_NOAUTOEN);
ret = devm_request_threaded_irq(&client->dev,
chip->interrupts[0],
--
2.40.1
--
Matti Vaittinen, Linux device drivers
ROHM Semiconductors, Finland SWDC
Kiviharjunlenkki 1E
90220 OULU
FINLAND
~~~ "I don't think so," said Rene Descartes. Just then he vanished ~~~
Simon says - in Latin please.
~~~ "non cogito me" dixit Rene Descarte, deinde evanescavit ~~~
Thanks to Simon Glass for the translation =]
On Fri, 19 May 2023 14:04:32 +0300
Matti Vaittinen <[email protected]> wrote:
> fwnode_irq_get[_byname]() were changed to not return 0 anymore. The
> special error case where device-tree based IRQ mapping fails can't no
> longer be reliably detected from this return value. This yields a
> functional change in the driver where the mapping failure is treated as
> an error.
>
> The mapping failure can occur for example when the device-tree IRQ
> information translation call-back(s) (xlate) fail, IRQ domain is not
> found, IRQ type conflicts, etc. In most cases this indicates an error in
> the device-tree and special handling is not really required.
>
> One more thing to note is that ACPI APIs do not return zero for any
> failures so this special handling did only apply on device-tree based
> systems.
>
> Drop the special handling for DT mapping failures as these can no longer
> be separated from other errors at driver side.
>
> Signed-off-by: Matti Vaittinen <[email protected]>
>
> ---
>
> Please note that I don't have the hardware to test this change.
> Furthermore, testing this type of device-tree error cases is not
> trivial, as the question we probably dive in is "what happens with the
> existing users who have errors in the device-tree". Answering to this
> question is not simple.
>
> I did this patch with minimal code changes - but a question is if we
> should really jump into the else branch below on all IRQ getting errors?
>
> } else {
> indio_dev->info = &ad7150_info_no_irq;
> switch (id->driver_data) {
> case AD7150:
> indio_dev->channels = ad7150_channels_no_irq;
> indio_dev->num_channels =
> ARRAY_SIZE(ad7150_channels_no_irq);
> break;
> case AD7151:
> indio_dev->channels = ad7151_channels_no_irq;
> indio_dev->num_channels =
> ARRAY_SIZE(ad7151_channels_no_irq);
> break;
> default:
> return -EINVAL;
> }
>
> Why do we have special handling for !chip->interrupts[0] while other
> errors on getting the fwnode_irq_get(dev_fwnode(&client->dev), 0); will
> abort the probe?
Gut feeling is that this was a rework of board file code where 0 meant not
provided. We should look to do the same here. I'm not sure we have a consistent
return for no irq though across the various fw types.
The driver looks like it should support either no interrupts or all the
ones for a given device.
Currrently it definitely doesn't handle the no irqs provided case right.
Its not elegant, but if we have to have all failures to get irqs result
in carrying on without them then that's better than now.
Jonathan
>
> The first patch of the series changes the fwnode_irq_get() so this depends
> on the first patch of the series and should not be applied alone.
> ---
> drivers/iio/cdc/ad7150.c | 3 +--
> 1 file changed, 1 insertion(+), 2 deletions(-)
>
> diff --git a/drivers/iio/cdc/ad7150.c b/drivers/iio/cdc/ad7150.c
> index 79aeb0aaea67..d7ba50b9780d 100644
> --- a/drivers/iio/cdc/ad7150.c
> +++ b/drivers/iio/cdc/ad7150.c
> @@ -567,8 +567,7 @@ static int ad7150_probe(struct i2c_client *client)
> if (chip->interrupts[1] < 0)
> return chip->interrupts[1];
> }
> - if (chip->interrupts[0] &&
> - (id->driver_data == AD7151 || chip->interrupts[1])) {
> + if (id->driver_data == AD7151 || chip->interrupts[1]) {
> irq_set_status_flags(chip->interrupts[0], IRQ_NOAUTOEN);
> ret = devm_request_threaded_irq(&client->dev,
> chip->interrupts[0],
Fri, May 19, 2023 at 02:01:47PM +0300, Matti Vaittinen kirjoitti:
> fwnode_irq_get[_byname]() were changed to not return 0 anymore.
>
> Drop check for return value 0.
...
> - if (v->irq <= 0) {
> + if (v->irq < 0) {
> ret = -EINVAL;
ret = v->irq;
?
> goto err;
> }
--
With Best Regards,
Andy Shevchenko
Hi Andy,
On 5/21/23 20:19, [email protected] wrote:
> Fri, May 19, 2023 at 02:01:47PM +0300, Matti Vaittinen kirjoitti:
>> fwnode_irq_get[_byname]() were changed to not return 0 anymore.
>>
>> Drop check for return value 0.
>
> ...
>
>> - if (v->irq <= 0) {
>> + if (v->irq < 0) {
>> ret = -EINVAL;
>
> ret = v->irq;
>
> ?
For me that seems to be correct, yes. This, however, would be a
functional change and in my opinion it should be done separately from
this API change.
>
>> goto err;
>> }
>
--
Matti Vaittinen
Linux kernel developer at ROHM Semiconductors
Oulu Finland
~~ When things go utterly wrong vim users can always type :help! ~~
On Mon, 22 May 2023 08:15:01 +0300
Matti Vaittinen <[email protected]> wrote:
> Hi Andy,
>
> On 5/21/23 20:19, [email protected] wrote:
> > Fri, May 19, 2023 at 02:01:47PM +0300, Matti Vaittinen kirjoitti:
> >> fwnode_irq_get[_byname]() were changed to not return 0 anymore.
> >>
> >> Drop check for return value 0.
> >
> > ...
> >
> >> - if (v->irq <= 0) {
> >> + if (v->irq < 0) {
> >> ret = -EINVAL;
> >
> > ret = v->irq;
> >
> > ?
>
> For me that seems to be correct, yes. This, however, would be a
> functional change and in my opinion it should be done separately from
> this API change.
Ah. I commented on this as well in v6. Roll us that separate patch
and I expect we'll both be happy ;)
Jonathan
>
> >
> >> goto err;
> >> }
> >
>