On 31.03.2021 11:18, Nicolas Ferre wrote:
> On 09/02/2021 at 12:01, Claudiu Beznea wrote:
>> Free resources on exit path (failure path of probe and remove).
>
> I'm not sure we can use this driver as a module anyway.
>
> Otherwise, it looks fine, but isn't it possible to use devm_of_iomap(),
> even in loop, and avoid having to deal with exit path?
For:
reset->rstc_base = of_iomap(pdev->dev.of_node, 0);
it should work.
For the maps in the loop I have to double check. Basically, the struct
resource object to pass to devm_of_iomap() is needed and for this the
pointer to a struct platform_device object corresponding to the node we
look for in the loop is needed. So, I think this cannot be done this way.
>
>> Reported-by: kernel test robot <l...@intel.com>
>> Reported-by: Dan Carpenter <dan.carpen...@oracle.com>
>> Signed-off-by: Claudiu Beznea <claudiu.bez...@microchip.com>
>> ---
>> drivers/power/reset/at91-reset.c | 25 ++++++++++++++++++++-----
>> 1 file changed, 20 insertions(+), 5 deletions(-)
>>
>> diff --git a/drivers/power/reset/at91-reset.c
>> b/drivers/power/reset/at91-reset.c
>> index 3ff9d93a5226..2ff7833153b6 100644
>> --- a/drivers/power/reset/at91-reset.c
>> +++ b/drivers/power/reset/at91-reset.c
>> @@ -206,7 +206,8 @@ static int __init at91_reset_probe(struct
>> platform_device *pdev)
>> if (!reset->ramc_base[idx]) {
>> dev_err(&pdev->dev, "Could not map ram controller
>> address\n");
>> of_node_put(np);
>> - return -ENODEV;
>> + ret = -ENODEV;
>> + goto unmap;
>> }
>> idx++;
>> }
>> @@ -218,13 +219,15 @@ static int __init at91_reset_probe(struct
>> platform_device *pdev)
>> reset->args = (u32)match->data;
>> reset->sclk = devm_clk_get(&pdev->dev, NULL);
>> - if (IS_ERR(reset->sclk))
>> - return PTR_ERR(reset->sclk);
>> + if (IS_ERR(reset->sclk)) {
>> + ret = PTR_ERR(reset->sclk);
>> + goto unmap;
>> + }
>> ret = clk_prepare_enable(reset->sclk);
>> if (ret) {
>> dev_err(&pdev->dev, "Could not enable slow clock\n");
>> - return ret;
>> + goto unmap;
>> }
>> platform_set_drvdata(pdev, reset);
>> @@ -239,21 +242,33 @@ static int __init at91_reset_probe(struct
>> platform_device *pdev)
>> ret = register_restart_handler(&reset->nb);
>> if (ret) {
>> clk_disable_unprepare(reset->sclk);
>> - return ret;
>> + goto unmap;
>> }
>> at91_reset_status(pdev, reset->rstc_base);
>> return 0;
>> +
>> +unmap:
>> + iounmap(reset->rstc_base);
>> + for (idx = 0; idx < ARRAY_SIZE(reset->ramc_base); idx++)
>> + iounmap(reset->ramc_base[idx]);
>
> But if we keep this loop, I have the feeling that some kind of
> "of_node_put()" is needed as well.
No! In the loop:
for_each_matching_node_and_match(np, at91_ramc_of_match, &match) {
reset->ramc_lpr = (u32)match->data;
reset->ramc_base[idx] = of_iomap(np, 0);
if (!reset->ramc_base[idx]) {
dev_err(&pdev->dev, "Could not map ram controller address\n");
of_node_put(np);
ret = -ENODEV;
goto unmap;
}
idx++;
}
the of_node_put() is needed only if the loop is interrupted as the macro:
for_each_matching_node_and_match() is defined as follows:
#define for_each_matching_node_and_match(dn, matches, match) \
for (dn = of_find_matching_node_and_match(NULL, matches, match); \
dn; dn = of_find_matching_node_and_match(dn, matches, match))
and of_find_matching_node_and_match() will return a np with refcount
incremented but at the next loop step the of_find_matching_node_and_match()
will be called with the same np pointer and the np refcount will be
decremented.
struct device_node *of_find_matching_node_and_match(
struct device_node *from,
const struct of_device_id *matches,
const struct of_device_id **match)
{
// ...
of_node_put(from);
// ...
}
>
>> +
>> + return ret;
>> }
>> static int __exit at91_reset_remove(struct platform_device *pdev)
>> {
>> struct at91_reset *reset = platform_get_drvdata(pdev);
>> + int idx;
>> unregister_restart_handler(&reset->nb);
>> clk_disable_unprepare(reset->sclk);
>> + iounmap(reset->rstc_base);
>> + for (idx = 0; idx < ARRAY_SIZE(reset->ramc_base); idx++)
>> + iounmap(reset->ramc_base[idx]);
>
> Ditto
>
>> +
>> return 0;
>> }
>>
>
>
