Re: [3/3] hwmon: iio_hwmon: defer probe when no channel is found

From: Guenter Roeck
Date: Sun Jul 03 2016 - 11:49:02 EST


On 07/03/2016 03:47 AM, Jonathan Cameron wrote:
On 30/06/16 15:51, Guenter Roeck wrote:
On 06/30/2016 06:59 AM, Jonathan Cameron wrote:


On 30 June 2016 04:47:25 BST, Guenter Roeck <linux@xxxxxxxxxxxx> wrote:
On Tue, Jun 28, 2016 at 10:18:17AM +0200, Quentin Schulz wrote:
iio_channel_get_all returns -ENODEV when it cannot find either
phandles and
properties in the Device Tree or channels whose consumer_dev_name
matches
iio_hwmon in iio_map_list. The iio_map_list is filled in by iio
drivers
which might be probed after iio_hwmon.

It is better to defer the probe of iio_hwmon if such error is
returned by
iio_channel_get_all in order to let a chance to iio drivers to expose
channels in iio_map_list.

Signed-off-by: Quentin Schulz <quentin.schulz@xxxxxxxxxxxxxxxxxx>
---
drivers/hwmon/iio_hwmon.c | 5 ++++-
1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/drivers/hwmon/iio_hwmon.c b/drivers/hwmon/iio_hwmon.c
index b550ba5..c0da4d9 100644
--- a/drivers/hwmon/iio_hwmon.c
+++ b/drivers/hwmon/iio_hwmon.c
@@ -73,8 +73,11 @@ static int iio_hwmon_probe(struct platform_device
*pdev)
name = dev->of_node->name;

channels = iio_channel_get_all(dev);
- if (IS_ERR(channels))
+ if (IS_ERR(channels)) {
+ if (PTR_ERR(channels) == -ENODEV)
+ return -EPROBE_DEFER;

The problem, as I see it, is with iio, which should return
-EPROBE_DEFER
in this situation.
Agreed. New fangled stuff this deferred probing :)

We can not convert -ENODEV to -EPROBE_DEFER without risking that the
channels are _really_ not there, which would result in endless
"deferred"
messages.
Hmm not entirely sure how we prevent that happening wherever it is done..


Outch. Better at the source, though. I didn't look at the iio code recently,
but can you detect the defer situation at least with devicetree ?

For non-devicetree situations, the only option I can think of would be
to replace the module initcall with a later initcall. That should solve
the problem if both iio_hwmon and and underlying drivers are built
into the kernel. If iio_hwmon is modular, the only real option I can
see is to make sure that all drivers it needs are loaded first.

Does this make sense ?
I think we need to look in a couple of directions. Firstly, investigate doing
something similar to gpio and basically move the setup of maps much earlier.
This will replace drivers presenting their own maps

The other direction is to get userspace (i.e. configfs) setup of these maps
working so for cases where it's a bit less hardware defined (i.e. using an
accelerometer as an input device) we can do once we know all devices relevant
are present and instantiate new instances on the fly.

Anyhow, neither is trivial unfortunately.


On a higher level, I think it would be better if the iio-hwmon bridge was tied
to chips (and thus to the chip driver), and not be an independent binding.
No idea if/how we can do that, though.

Guenter