Re: [PATCH v2] backlight: pwm: reject legacy pwm request for device defined in dt

On Mon, 12 Oct 2015 16:54:39 +0300
Vladimir Zapolskiy [off-list ref] wrote:

Hi Boris,

On 12.10.2015 16:30, Boris Brezillon wrote:

quoted

Hi Vladimir,

On Mon, 12 Oct 2015 15:16:44 +0200
Nicolas Ferre [off-list ref] wrote:

quoted

Le 12/10/2015 14:29, Vladimir Zapolskiy a écrit :

quoted

Platform PWM backlight data provided by board's device tree should be
complete enough to successfully request a pwm device using pwm_get()
API. This change fixes a bug, when an arbitrary (first found) PWM is
connected to a "pwm-backlight" compatible device, when explicit PWM
device reference is not given.

Documentation/devicetree/bindings/video/backlight/pwm-backlight.txt
already describes "pwms" as a required property, instead of blind
selection of a potentially wrong PWM reject legacy PWM device
registration request, leave legacy API only for non-dt cases.

Based on initial implementation done by Dmitry Eremin-Solenikov.

Reported-by: Dmitry Eremin-Solenikov <redacted>
Signed-off-by: Vladimir Zapolskiy <redacted>
Acked-by: Thierry Reding <redacted>
Acked-by: Lee Jones <redacted>

It seems good to me:
Acked-by: Nicolas Ferre <redacted>

(Adding some people to the Cc: list).

quoted

---
The change is based on lee-backlight/for-backlight-next

Changes from v1 to v2:
* rebased on top of Nicolas' commit
    68feaca0b13 ("backlight: pwm: Handle EPROBE_DEFER while requesting the PWM")

Links to previous discussions of the change:
* https://patchwork.ozlabs.org/patch/483993/
* https://patchwork.ozlabs.org/patch/398849/

 drivers/video/backlight/pwm_bl.c | 19 +++++++++----------
 1 file changed, 9 insertions(+), 10 deletions(-)

diff --git a/drivers/video/backlight/pwm_bl.c b/drivers/video/backlight/pwm_bl.c
index eff379b..ae3c6b6 100644
--- a/drivers/video/backlight/pwm_bl.c
+++ b/drivers/video/backlight/pwm_bl.c

@@ -271,19 +271,18 @@ static int pwm_backlight_probe(struct platform_device *pdev)
 	}
 
 	pb->pwm = devm_pwm_get(&pdev->dev, NULL);
-	if (IS_ERR(pb->pwm)) {
-		ret = PTR_ERR(pb->pwm);
-		if (ret = -EPROBE_DEFER)
-			goto err_alloc;
-
+	if (IS_ERR(pb->pwm) && PTR_ERR(pb->pwm) != -EPROBE_DEFER
+	    && !pdev->dev.of_node) {
 		dev_err(&pdev->dev, "unable to request PWM, trying legacy API\n");
 		pb->legacy = true;
 		pb->pwm = pwm_request(data->pwm_id, "pwm-backlight");
-		if (IS_ERR(pb->pwm)) {
-			dev_err(&pdev->dev, "unable to request legacy PWM\n");
-			ret = PTR_ERR(pb->pwm);
-			goto err_alloc;
-		}
+	}
+
+	if (IS_ERR(pb->pwm)) {
+		ret = PTR_ERR(pb->pwm);
+		if (ret != -EPROBE_DEFER)
+			dev_err(&pdev->dev, "unable to request PWM\n");
+		goto err_alloc;
 	}
 
 	dev_dbg(&pdev->dev, "got pwm for backlight\n");

I still think it would be cleaner to do what Thierry proposed here [1].
IMO, embedding the complexity of different error cases depending on the
way PWM devices were defined (OF, pdata, ...) is rather risky and
make the code even more complicated.

please correct me if I'm wrong, I suppose Thierry's change fixes
Nicolas' commit 68feaca0b13 only, and the intention of my change is to
fix an absolutely unrelated problem, see the commit message.

So, since still there is a remained chance of getting -EPROBE_DEFER from
pwm_get(), e.g. from of_pwm_get() or failed pwmchip_find_by_name() or
pwm->chip->ops->request() I don't see how Thierry's change alone may
help me to overcome the problem I'm trying to solve here.

The only valid case where EPROBE_DEFER should be returned is when we
have a device that is not ready to be used yet (but we're sure that we
have this device declared, using either the PWM lookup table or the DT
definition in the PWM subsystem case).
Thierry's patch makes sure that EPROBE_DEFER is not returned when the
PWM device definition is not found using in the PWM lookup tables or
the DT definition, and in this case the pwm_bl code will fallback to
the legacy PWM API, which AFAICT is what you're trying to solve.

Best Regards,

Boris
-- 
Boris Brezillon, Free Electrons
Embedded Linux and Kernel engineering
http://free-electrons.com