2013-04-26 13:57:16

by Axel Lin

[permalink] [raw]
Subject: [PATCH] regulator: ab3100: Fix regulator register error handling

Ensure to unregister all regulators before return error in probe().

The regulator register order depends on the regulator ID pass to
ab3100_regulator_register() function. Thus we need to scan ab3100_regulator_desc
and find the index of successfully registered regulators, or alternatively just
call ab3100_regulators_remove() to unregister all registered regulators.

Since current code uses a static ab3100_regulators table, explicitly set
reg->rdev = NULL after regulator_unregister() call to ensure calling
ab3100_regulators_remove() in the unwind path always work.

Also move ab3100_regulators_remove() to avoid forward declaration.

Signed-off-by: Axel Lin <[email protected]>
---
drivers/regulator/ab3100.c | 33 +++++++++++++++++++--------------
1 file changed, 19 insertions(+), 14 deletions(-)

diff --git a/drivers/regulator/ab3100.c b/drivers/regulator/ab3100.c
index be1e6ad..3be9e46 100644
--- a/drivers/regulator/ab3100.c
+++ b/drivers/regulator/ab3100.c
@@ -609,6 +609,19 @@ static const u8 ab3100_reg_initvals[] = {
LDO_D_SETTING,
};

+static int ab3100_regulators_remove(struct platform_device *pdev)
+{
+ int i;
+
+ for (i = 0; i < AB3100_NUM_REGULATORS; i++) {
+ struct ab3100_regulator *reg = &ab3100_regulators[i];
+
+ regulator_unregister(reg->rdev);
+ reg->rdev = NULL;
+ }
+ return 0;
+}
+
static int
ab3100_regulator_of_probe(struct platform_device *pdev, struct device_node *np)
{
@@ -635,8 +648,10 @@ ab3100_regulator_of_probe(struct platform_device *pdev, struct device_node *np)
pdev, NULL, ab3100_regulator_matches[i].init_data,
ab3100_regulator_matches[i].of_node,
(int) ab3100_regulator_matches[i].driver_data);
- if (err)
+ if (err) {
+ ab3100_regulators_remove(pdev);
return err;
+ }
}

return 0;
@@ -695,25 +710,15 @@ static int ab3100_regulators_probe(struct platform_device *pdev)

err = ab3100_regulator_register(pdev, plfdata, NULL, NULL,
desc->id);
- if (err)
+ if (err) {
+ ab3100_regulators_remove(pdev);
return err;
+ }
}

return 0;
}

-static int ab3100_regulators_remove(struct platform_device *pdev)
-{
- int i;
-
- for (i = 0; i < AB3100_NUM_REGULATORS; i++) {
- struct ab3100_regulator *reg = &ab3100_regulators[i];
-
- regulator_unregister(reg->rdev);
- }
- return 0;
-}
-
static struct platform_driver ab3100_regulators_driver = {
.driver = {
.name = "ab3100-regulators",
--
1.7.10.4



2013-04-26 15:33:57

by Mark Brown

[permalink] [raw]
Subject: Re: [PATCH] regulator: ab3100: Fix regulator register error handling

On Fri, Apr 26, 2013 at 09:57:09PM +0800, Axel Lin wrote:
> Ensure to unregister all regulators before return error in probe().
>
> The regulator register order depends on the regulator ID pass to
> ab3100_regulator_register() function. Thus we need to scan ab3100_regulator_desc
> and find the index of successfully registered regulators, or alternatively just
> call ab3100_regulators_remove() to unregister all registered regulators.

Applied, thanks.


Attachments:
(No filename) (456.00 B)
signature.asc (836.00 B)
Digital signature
Download all attachments

2013-04-29 12:53:05

by Linus Walleij

[permalink] [raw]
Subject: Re: [PATCH] regulator: ab3100: Fix regulator register error handling

On Fri, Apr 26, 2013 at 3:57 PM, Axel Lin <[email protected]> wrote:

> Ensure to unregister all regulators before return error in probe().
>
> The regulator register order depends on the regulator ID pass to
> ab3100_regulator_register() function. Thus we need to scan ab3100_regulator_desc
> and find the index of successfully registered regulators, or alternatively just
> call ab3100_regulators_remove() to unregister all registered regulators.
>
> Since current code uses a static ab3100_regulators table, explicitly set
> reg->rdev = NULL after regulator_unregister() call to ensure calling
> ab3100_regulators_remove() in the unwind path always work.
>
> Also move ab3100_regulators_remove() to avoid forward declaration.
>
> Signed-off-by: Axel Lin <[email protected]>

Excellent patch Axel, thanks!
Reviewed-by: Linus Walleij <[email protected]>

Yours,
Linus Walleij