[RFC PATCH 6/8] PM / Domains: Remove a provider by referencing the data pointer

[rafael@kernel.org (Rafael J. Wysocki)]

On Thu, Aug 11, 2016 at 6:39 PM, Jon Hunter <jonathanh@nvidia.com> wrote:
>
> On 05/08/16 12:55, Ulf Hansson wrote:
>> On 21 June 2016 at 15:47, Jon Hunter <jonathanh@nvidia.com> wrote:
>>>
>>> On 15/06/16 15:38, Ulf Hansson wrote:
>>>> On 4 March 2016 at 12:23, Jon Hunter <jonathanh@nvidia.com> wrote:
>>>>> To remove a PM domain from the system, it is necessary to ensure
>>>>> that any PM domain providers associated with the PM domain have
>>>>> been removed. Otherwise it could be possible to obtain a pointer
>>>>> to a PM domain structure that has been removed.
>>>>>
>>>>> PM domains now have a reference to the pointer for the PM domain
>>>>> provider's data variable. Add a function so that a PM domain can
>>>>> remove a PM domain provider by referencing the data pointer.
>>
>>
>>>>>
>>>>> Signed-off-by: Jon Hunter <jonathanh@nvidia.com>
>>>>> ---
>>>>>  drivers/base/power/domain.c | 24 ++++++++++++++++++++++++
>>>>>  include/linux/pm_domain.h   |  2 ++
>>>>>  2 files changed, 26 insertions(+)
>>>>>
>>>>> diff --git a/drivers/base/power/domain.c b/drivers/base/power/domain.c
>>>>> index 72055fef6256..438885f2455f 100644
>>>>> --- a/drivers/base/power/domain.c
>>>>> +++ b/drivers/base/power/domain.c
>>>>> @@ -1738,6 +1738,30 @@ void of_genpd_del_provider(struct device_node *np)
>>>>>  EXPORT_SYMBOL_GPL(of_genpd_del_provider);
>>>>>
>>>>>  /**
>>>>> + * of_genpd_del_provider_by_data() - Remove a registered PM domain provider
>>>>> + * @data: Pointer to the data associated with the PM domain provider
>>>>> + *
>>>>> + * Look up a PM domain provider based upon a pointer to it's data and
>>>>> + * remove the PM domain provider from the list of providers.
>>>>> + */
>>>>> +void of_genpd_del_provider_by_data(void *data)
>>>>> +{
>>>>> +       struct of_genpd_provider *c, *cp;
>>>>> +
>>>>> +       mutex_lock(&of_genpd_mutex);
>>>>> +       list_for_each_entry_safe(cp, c, &of_genpd_providers, link) {
>>>>> +               if (cp->data == data) {
>>>>> +                       list_del(&cp->link);
>>>>> +                       of_node_put(cp->node);
>>>>> +                       kfree(cp);
>>>>> +                       break;
>>>>> +               }
>>>>> +       }
>>>>> +       mutex_unlock(&of_genpd_mutex);
>>>>> +}
>>>>> +EXPORT_SYMBOL_GPL(of_genpd_del_provider_by_data);
>>>>> +
>>>>> +/**
>>>>>   * of_genpd_get_from_provider() - Look-up PM domain
>>>>>   * @genpdspec: OF phandle args to use for look-up
>>>>>   *
>>>>> diff --git a/include/linux/pm_domain.h b/include/linux/pm_domain.h
>>>>> index bed84413546f..7b7921a65cb0 100644
>>>>> --- a/include/linux/pm_domain.h
>>>>> +++ b/include/linux/pm_domain.h
>>>>> @@ -199,6 +199,7 @@ int of_genpd_add_provider_simple(struct device_node *np,
>>>>>  int of_genpd_add_provider_onecell(struct device_node *np,
>>>>>                                   struct genpd_onecell_data *data);
>>>>>  void of_genpd_del_provider(struct device_node *np);
>>>>
>>>> There's currently only one user of of_genpd_del_provider().
>>>>
>>>> Could this patch just convert that user to the new API, so we don't
>>>> need to keep both the legacy and new one?
>>>>
>>>> I guess we could then just stick to the name "of_genpd_del_provider()".
>>>
>>> I had a look at this and to do that we would end up with
>>> of_genpd_del_provider(struct device_node *np, void *data) where the user
>>> should only pass one of the arguments. It seems a bit odd. However,
>>> unless I have forgotten something, I wonder if we should just make
>>> of_genpd_del_provider_by_name() a local function and not export this at
>>> all? It seems more natural for users to delete a provider by the
>>> device_node than by name rather than the data argument.
>>>
>>> The only problem I see with making of_genpd_del_provider_by_name() local
>>> is that I need to add a prototype for the function at the top of the
>>> domain.c source file so that it builds because __pm_genpd_remove() is
>>> defined above it. Yes I could move __pm_genpd_remove() to the bottom of
>>> the file but then it is not located next to pm_genpd_init() which seems odd.
>>>
>>> Let me know what you think.
>>
>> Sorry for delay! I have now looked into this in more detail.
>
> No problem. Thanks!
>
>> When an genpd provider is added today, it's supposed to get a
>> corresponding *unique* OF device node associated with it, right!?
>>
>> If we store this OF device node from the provider in the struct
>> generic_pm_domain, instead of the "provider_data pointer", we wouldn't
>> need to the add of_genpd_del_provider_by_data() at all. Because we can
>> use the currently available of_genpd_del_provider(), right!?
>>
>> Or what am I missing? :-)

Please don't store device_node pointers in generic data structures at
least in the code that I maintain (some other people may not care).

Store struct fwnode_handle pointers instead if you have to.

Thanks,
Rafael