Re: [PATCH] clk: Free struct clk allocated during clk_hw_register()

[Rajendra Nayak <rnayak@codeaurora.org>]

On 11/08/2016 03:34 PM, Rajendra Nayak wrote:
> 
> 
> On 11/08/2016 03:06 PM, Geert Uytterhoeven wrote:
>> Hi Rajendra,
>>
>> On Tue, Nov 8, 2016 at 9:23 AM, Rajendra Nayak <rnayak@codeaurora.org> wrote:
>>> With clk_hw_register() API we hide the struct clk from the caller
>>> and return an int error code instead, so the caller (clk provider)
>>> is not expected to use hw->clk on return.
>>
>> That's correct, in case of failure.
> 
> sorry, maybe the commit text needs to be reworded. I meant 'clk_hw_register() returns
> an int (not a struct clk pointer), 0 on success or an error code in case of a failure.
> 
>>
>>> Free the memory, and mark hw->clk as NULL before returning.
>>>
>>> Signed-off-by: Rajendra Nayak <rnayak@codeaurora.org>
>>> ---
>>>  drivers/clk/clk.c | 10 +++++++++-
>>>  1 file changed, 9 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/drivers/clk/clk.c b/drivers/clk/clk.c
>>> index 0fb39fe..f81e4aa 100644
>>> --- a/drivers/clk/clk.c
>>> +++ b/drivers/clk/clk.c
>>> @@ -2628,7 +2628,15 @@ struct clk *clk_register(struct device *dev, struct clk_hw *hw)
>>>   */
>>>  int clk_hw_register(struct device *dev, struct clk_hw *hw)
>>>  {
>>> -       return PTR_ERR_OR_ZERO(clk_register(dev, hw));
>>> +       struct clk *c;
>>> +
>>> +       c = clk_register(dev, hw);
>>> +       if (IS_ERR(c))
>>> +               return PTR_ERR(c);
>>> +
>>> +       __clk_free_clk(c);
>>> +       hw->clk = NULL;
>>
>> This is the success path, not the failure path (on failure, clk_register()
>> has already freed the struct clk).
>> Why do you free the struct clk in case of success?
>>
>> What am I missing?
> 
> so with 'per-user' clks, I thought we now have one struct clk per user, allocated
> when the user does a clk_get() and freed with a clk_put(), so we shouldn't ideally
> need one during clk registration?
> The one allocated in clk_register() is for legacy users who need to get a struct clk *
> back. For users of clk_hw_register() this should not be needed, no?

Looking through this a little more, I don't think we can get rid of the 'struct clk'
allocations at registration time as yet.
It seems to be used by clk_get_parent() at least, which does not yet do a 
__clk_create_clk() and relies on hw->clk

-- 
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
hosted by The Linux Foundation