2015-12-01 09:06:26

by Sudip Mukherjee

Subject: [PATCH] clk: qcom: common: check for failure

We were not checking the return from devm_add_action() which can fail.

Signed-off-by: Sudip Mukherjee <[email protected]>
---
drivers/clk/qcom/common.c | 13 ++++++++++---
1 file changed, 10 insertions(+), 3 deletions(-)

diff --git a/drivers/clk/qcom/common.c b/drivers/clk/qcom/common.c
index c112eba..3541a9a 100644
--- a/drivers/clk/qcom/common.c
+++ b/drivers/clk/qcom/common.c
@@ -213,7 +213,10 @@ int qcom_cc_really_probe(struct platform_device *pdev,
if (ret)
return ret;

- devm_add_action(dev, qcom_cc_del_clk_provider, pdev->dev.of_node);
+ ret = devm_add_action(dev, qcom_cc_del_clk_provider,
+ pdev->dev.of_node);
+ if (ret)
+ return ret;

reset = &cc->reset;
reset->rcdev.of_node = dev->of_node;
@@ -236,8 +239,12 @@ int qcom_cc_really_probe(struct platform_device *pdev,
return ret;
}

- devm_add_action(dev, qcom_cc_gdsc_unregister, dev);
-
+ ret = devm_add_action(dev, qcom_cc_gdsc_unregister, dev);
+ if (ret) {
+ if (desc->gdscs && desc->num_gdscs)
+ gdsc_unregister(dev);
+ return ret;
+ }

return 0;
}
--
1.9.1

2015-12-03 07:39:20

by Stephen Boyd

Subject: Re: [PATCH] clk: qcom: common: check for failure

On 12/01, Sudip Mukherjee wrote:
> We were not checking the return from devm_add_action() which can fail.
>
> Signed-off-by: Sudip Mukherjee <[email protected]>
> ---
> drivers/clk/qcom/common.c | 13 ++++++++++---
> 1 file changed, 10 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/clk/qcom/common.c b/drivers/clk/qcom/common.c
> index c112eba..3541a9a 100644
> --- a/drivers/clk/qcom/common.c
> +++ b/drivers/clk/qcom/common.c
> @@ -213,7 +213,10 @@ int qcom_cc_really_probe(struct platform_device *pdev,
> if (ret)
> return ret;
>
> - devm_add_action(dev, qcom_cc_del_clk_provider, pdev->dev.of_node);
> + ret = devm_add_action(dev, qcom_cc_del_clk_provider,
> + pdev->dev.of_node);
> + if (ret)
> + return ret;

So now we don't remove the clk provider on allocation failure?
Confused.

>
> reset = &cc->reset;
> reset->rcdev.of_node = dev->of_node;
> @@ -236,8 +239,12 @@ int qcom_cc_really_probe(struct platform_device *pdev,
> return ret;
> }
>
> - devm_add_action(dev, qcom_cc_gdsc_unregister, dev);
> -
> + ret = devm_add_action(dev, qcom_cc_gdsc_unregister, dev);
> + if (ret) {
> + if (desc->gdscs && desc->num_gdscs)
> + gdsc_unregister(dev);
> + return ret;
> + }
>
> return 0;
> }

You seem to have missed the reset devm action. Why?

Also, I wonder if we could have devm_add_action() or some other
new devm_add_action() wrapper that tried to add the action, and
if it failed it ran the action right there and returned the
-ENOMEM? So then we can just have:

ret = devm_add_action_or_do_it(dev, qcom_cc_gdsc_unregister, dev)
if (ret)
return ret;

and we're assured that on the failure path we'll have already
called qcom_cc_gdsc_unregister.

--
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
a Linux Foundation Collaborative Project

2015-12-03 08:49:58

by Sudip Mukherjee

Subject: Re: [PATCH] clk: qcom: common: check for failure

On Wed, Dec 02, 2015 at 11:39:17PM -0800, Stephen Boyd wrote:
> On 12/01, Sudip Mukherjee wrote:
> > We were not checking the return from devm_add_action() which can fail.
> >
> > Signed-off-by: Sudip Mukherjee <[email protected]>
> > ---
> > drivers/clk/qcom/common.c | 13 ++++++++++---
> > 1 file changed, 10 insertions(+), 3 deletions(-)
> >
> > diff --git a/drivers/clk/qcom/common.c b/drivers/clk/qcom/common.c
> > index c112eba..3541a9a 100644
> > --- a/drivers/clk/qcom/common.c
> > +++ b/drivers/clk/qcom/common.c
> > @@ -213,7 +213,10 @@ int qcom_cc_really_probe(struct platform_device *pdev,
> > if (ret)
> > return ret;
> >
> > - devm_add_action(dev, qcom_cc_del_clk_provider, pdev->dev.of_node);
> > + ret = devm_add_action(dev, qcom_cc_del_clk_provider,
> > + pdev->dev.of_node);
> > + if (ret)
> > + return ret;
>
> So now we don't remove the clk provider on allocation failure?
> Confused.
>
> >
> > reset = &cc->reset;
> > reset->rcdev.of_node = dev->of_node;
> > @@ -236,8 +239,12 @@ int qcom_cc_really_probe(struct platform_device *pdev,
> > return ret;
> > }
> >
> > - devm_add_action(dev, qcom_cc_gdsc_unregister, dev);
> > -
> > + ret = devm_add_action(dev, qcom_cc_gdsc_unregister, dev);
> > + if (ret) {
> > + if (desc->gdscs && desc->num_gdscs)
> > + gdsc_unregister(dev);
> > + return ret;
> > + }
> >
> > return 0;
> > }
>
> You seem to have missed the reset devm action. Why?

I have messed up pretty bad in this patch. :(
>
> Also, I wonder if we could have devm_add_action() or some other
> new devm_add_action() wrapper that tried to add the action, and
> if it failed it ran the action right there and returned the
> -ENOMEM? So then we can just have:
>
> ret = devm_add_action_or_do_it(dev, qcom_cc_gdsc_unregister, dev)
> if (ret)
> return ret;
>
> and we're assured that on the failure path we'll have already
> called qcom_cc_gdsc_unregister.

I am on it, will send you a patch for this.

regards
sudip