If an error occurs after the k3_dsp_rproc_request_mbox() call,
mbox_free_channel() must be called, as already done in the remove function.

Instead of adding an error handling path in the probe and changing all
error handling in the function, add a new devm_add_action_or_reset() and
simplify the .remove() function.

Fixes: ea1d6fb5b571 ("remoteproc: k3-dsp: Acquire mailbox handle during probe routine")
Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
Reviewed-by: Andrew Davis <afd@ti.com>
---
Compile tested only

Change in v2:
  - fix the subject (cut'n'paste issue)   [Andrew Davis]
  - add R-b tag
  
v1: https://lore.kernel.org/all/9485e127a00419c76cf13dbccf4874af395ef6ba.1725653543.git.christophe.jaillet@wanadoo.fr/
---
 drivers/remoteproc/ti_k3_dsp_remoteproc.c | 13 +++++++++++--
 1 file changed, 11 insertions(+), 2 deletions(-)

diff --git a/drivers/remoteproc/ti_k3_dsp_remoteproc.c b/drivers/remoteproc/ti_k3_dsp_remoteproc.c
index 8be3f631c192..f29780de37a5 100644
--- a/drivers/remoteproc/ti_k3_dsp_remoteproc.c
+++ b/drivers/remoteproc/ti_k3_dsp_remoteproc.c
@@ -610,6 +610,13 @@ static void k3_dsp_release_tsp(void *data)
 	ti_sci_proc_release(tsp);
 }
 
+static void k3_dsp_free_channel(void *data)
+{
+	struct k3_dsp_rproc *kproc = data;
+
+	mbox_free_channel(kproc->mbox);
+}
+
 static int k3_dsp_rproc_probe(struct platform_device *pdev)
 {
 	struct device *dev = &pdev->dev;
@@ -649,6 +656,10 @@ static int k3_dsp_rproc_probe(struct platform_device *pdev)
 	if (ret)
 		return ret;
 
+	ret = devm_add_action_or_reset(dev, k3_dsp_free_channel, rproc);
+	if (ret)
+		return ret;
+
 	kproc->ti_sci = devm_ti_sci_get_by_phandle(dev, "ti,sci");
 	if (IS_ERR(kproc->ti_sci))
 		return dev_err_probe(dev, PTR_ERR(kproc->ti_sci),
@@ -741,8 +752,6 @@ static void k3_dsp_rproc_remove(struct platform_device *pdev)
 		if (ret)
 			dev_err(dev, "failed to detach proc (%pe)\n", ERR_PTR(ret));
 	}
-
-	mbox_free_channel(kproc->mbox);
 }
 
 static const struct k3_dsp_mem_data c66_mems[] = {
-- 
2.46.0

On Sat, Sep 07, 2024 at 08:33:36AM +0200, Christophe JAILLET wrote:
> If an error occurs after the k3_dsp_rproc_request_mbox() call,
> mbox_free_channel() must be called, as already done in the remove function.
> 
> Instead of adding an error handling path in the probe and changing all
> error handling in the function, add a new devm_add_action_or_reset() and
> simplify the .remove() function.
> 
> Fixes: ea1d6fb5b571 ("remoteproc: k3-dsp: Acquire mailbox handle during probe routine")
> Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
> Reviewed-by: Andrew Davis <afd@ti.com>
> ---
> Compile tested only
> 
> Change in v2:
>   - fix the subject (cut'n'paste issue)   [Andrew Davis]
>   - add R-b tag
>   
> v1: https://lore.kernel.org/all/9485e127a00419c76cf13dbccf4874af395ef6ba.1725653543.git.christophe.jaillet@wanadoo.fr/
> ---
>  drivers/remoteproc/ti_k3_dsp_remoteproc.c | 13 +++++++++++--
>  1 file changed, 11 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/remoteproc/ti_k3_dsp_remoteproc.c b/drivers/remoteproc/ti_k3_dsp_remoteproc.c
> index 8be3f631c192..f29780de37a5 100644
> --- a/drivers/remoteproc/ti_k3_dsp_remoteproc.c
> +++ b/drivers/remoteproc/ti_k3_dsp_remoteproc.c
> @@ -610,6 +610,13 @@ static void k3_dsp_release_tsp(void *data)
>  	ti_sci_proc_release(tsp);
>  }
>  
> +static void k3_dsp_free_channel(void *data)
> +{
> +	struct k3_dsp_rproc *kproc = data;

How did the struct rproc from devm_add_action_or_reset() got turned into a
struct k3_dsp_rproc?

> +
> +	mbox_free_channel(kproc->mbox);
> +}
> +
>  static int k3_dsp_rproc_probe(struct platform_device *pdev)
>  {
>  	struct device *dev = &pdev->dev;
> @@ -649,6 +656,10 @@ static int k3_dsp_rproc_probe(struct platform_device *pdev)
>  	if (ret)
>  		return ret;
>  
> +	ret = devm_add_action_or_reset(dev, k3_dsp_free_channel, rproc);
> +	if (ret)
> +		return ret;
> +
>  	kproc->ti_sci = devm_ti_sci_get_by_phandle(dev, "ti,sci");
>  	if (IS_ERR(kproc->ti_sci))
>  		return dev_err_probe(dev, PTR_ERR(kproc->ti_sci),
> @@ -741,8 +752,6 @@ static void k3_dsp_rproc_remove(struct platform_device *pdev)
>  		if (ret)
>  			dev_err(dev, "failed to detach proc (%pe)\n", ERR_PTR(ret));
>  	}
> -
> -	mbox_free_channel(kproc->mbox);
>  }
>  
>  static const struct k3_dsp_mem_data c66_mems[] = {
> -- 
> 2.46.0
>

Le 10/09/2024 à 17:29, Mathieu Poirier a écrit :

>> +static void k3_dsp_free_channel(void *data)
>> +{
>> +	struct k3_dsp_rproc *kproc = data;
> 
> How did the struct rproc from devm_add_action_or_reset() got turned into a
> struct k3_dsp_rproc?
> 

Well, Linux is a wonderful system, that is able to make wonderful thinks!

In this particular case, if it is not the correct explanation, it is 
likely an unfortunate typo :(
Sorry about that, and thanks for the careful review.
I'll send a v3.


But, looking at it, and trying to see if similar issues may exist, the 
following naive script spots 2 similar issues. 1 of them looks valid.


I'll try to improve it (Julia in copy, if she wants to give it a look 
as-well :))

CJ



@devm@
expression RET, DEV;
identifier FCT, VAR;
type T;
@@
	T VAR;
	...
(
	RET = devm_add_action_or_reset(DEV, FCT, VAR);
|
	devm_add_action_or_reset(DEV, FCT, VAR);
)

@depends on devm@
identifier FCT = devm.FCT, x, VAR;
type T1 = devm.T, T2;
@@
void FCT(void *x)
{
	...
(
	T1 VAR = x;
|
*	T2 VAR = x;
)
	...
}