Re: [PATCH v4 3/5] nvmem: core: Add consumer API to get nvmem cell from node

[Stefan Agner]

On 2016-07-08 08:41, Srinivas Kandagatla wrote:
> On 07/07/16 14:48, maitysanchayan@xxxxxxxxx wrote:
>> Hello Srinivas,
>>
>> On 16-07-07 14:16:36, Srinivas Kandagatla wrote:
>>>
>>>
>>> On 07/07/16 13:33, maitysanchayan@xxxxxxxxx wrote:
>>>> Hello Srinivas,
>>>>
>>>> On 16-07-07 10:18:49, Srinivas Kandagatla wrote:
>>>>>
>>>>>
>>>>> On 07/07/16 07:39, Sanchayan Maity wrote:
>>>>>> From: Stefan Agner <stefan@xxxxxxxx>
>>>>>>
>>>>>> The existing NVMEM consumer API's do not allow to get a
>>>>>> NVMEM cell directly given a device tree node. This patch
>>>>>> adds a function to provide this functionality.
>>>>>>
>>>>>> Assuming the nvmem cell id name is known, this can be used
>>>>>> as follows
>>>>>>
>>>>>> struct device_node *cell_np;
>>>>>> struct nvmem_cell *foo_cell;
>>>>>>
>>>>>> cell_np = of_find_node_by_name(parent, "foo");
>>>>>> foo_cell = of_nvmem_cell_get_direct(cell_np);
>>>>>
>>>>> I don't see a real gain in adding this new api,
>>>>> This will encourage people to use non-standard nvmem bindings.
>>>>>
>>>>> why not just use standard nvmem bindings.. and use
>>>>>
>>>>> of_nvmem_cell_get(np, "foo");
>>>>>
>>>>> which should work in your case.
>>>>
>>>> It will not work in our case. I believe what you are referring to will
>>>> work if I were to pass the device node pointer which was a NVMEM consumer
>>>> containing the nvmem-cells property. In our case, we pass the device node
>>>> pointer pointing to /soc which is not a nvmem consumer. In this case it
>>>> will not have nvmem-cells property causing of_nvmem_cell_get to return
>>>> EINVAL when it calls of_parse_phandle with "nvmem-cells".
>>>
>>> I could not see any bindings/ dt patches or dt examples for this this
>>> series.. so Am guessing your node would look like:
>>>
>>> soc {
>>> 	cfg0: cfg0 {
>>> 		...
>>> 	};
>>> 	cfg1: cfg1 {
>>> 		...
>>> 	};
>>> };
>>>
>>> If this is not how it looks, can you share the details.
>>>
>>> What Am saying is that why not have:
>>>
>>> soc {
>>> 	nvmem-cells = <&cfg0>, <&cfg1>;
>>> 	nvmem-cell-names = "cfg0", "cfg1";
>>>
>>> 	cfg0: cfg0 {
>>> 		...
>>> 	};
>>>
>>> 	cfg1: cfg1 {
>>> 		...
>>> 	};
>>> };
>>>
>>>>
>>>> Our requirement is to be able to pass the soc node pointer and then
>>>> be able to get a nvmem cell by specifying it's name. So for our case
>>>
>>> Why?
>>
>> Sorry for not providing the background directly. The patches before this
>> series used that approach. In the previous discussions it has been pointed
>> out that it is not acceptable to have additional device tree bindings for
>> providing data that the driver wants at the SoC node level or to have bindings
>> just for the SoC bus driver alone since we aren't really describing the
>> hardware.
>>
> SOC driver seems to search for an arbitrary node by its name, which is
> not a binding and can break anytime in cases If the scope of nvmem
> provider is out of soc node or if the nvmem cells are not named as
> expected. That looks very fragile.

In that case, that just "won't happen" because the soc driver is a very
soc specific driver only used for this device tree. We it will always
bind to that high level soc node.

> 
> If the soc node is actual consumer of nvmem cells, I see no reason why
> we should not use proper nvmem bindings?

There is a reason: We don't describe the hardware with it...

The cfg0/cfg1 register which Sanchayan needs to read in the soc bus
driver are just two register with a unique ID of the SoC. In whatever
driver throughout the system we use that ID (e.g. in a random generator
for initialization) we never describe an actual hardware relation... Its
just software and how we use that unique ID. The device tree is ment to
describe hardware. Hence the NVMEM consumer binding is not suited for
such NVMEM cells...

By describing the NVMEM cells location in device tree (producer API, the
NVMEM cells are in hardware at that location, so using the device tree
for that part is fine) and hard coding the NVMEM cell we need in the
driver code we don't violate the device tree matra "describe the
hardware"...

Looking-up the nodes direcly is what Rob suggested here:
https://lkml.org/lkml/2016/5/23/573

> 
> Given the fact that the patch is potentially bypassing the nvmem
> bindings, am not happy to take it!

If you can provide a solution acceptable by the device tree folks and
works without this patch, I am happy to do it...

Btw, I am not entirely happy with the API name, but did not had a better
idea... And we we should probably add a note that the device tree
consumer binding is the preferred way to do it.

--
Stefan


> 
> thanks,
> srini
> 
>> For the discussion,
>> https://lkml.org/lkml/2016/5/23/573
>> https://lkml.org/lkml/2016/5/2/71
>>
>> Regards,
>> Sanchayan.
>>
>>
>>>
>>>> ocotp node has cfg0 and cfg1 which we want but we cannot use existing
>>>> nvmem consumer API since that requires having the nvmem consumer properties
>>>> in the node we are binding to viz. is a direct nvmem consumer.
>>>>
>>>> Regards,
>>>> Sanchayan.
>>>>
>>>>>
>>>>> thanks,
>>>>> srini
>>>>>>
>>>>>> Parent node can also be the of_node of the main SoC device
>>>>>> node.
>>>>>>
>>>>>> Signed-off-by: Sanchayan Maity <maitysanchayan@xxxxxxxxx>
>>>>>> ---
>>>>>>     drivers/nvmem/core.c           | 44 +++++++++++++++++++++++++++++-------------
>>>>>>     include/linux/nvmem-consumer.h |  1 +
>>>>>>     2 files changed, 32 insertions(+), 13 deletions(-)
>>>>>>
>>>>>> diff --git a/drivers/nvmem/core.c b/drivers/nvmem/core.c
>>>>>> index 965911d..470abee 100644
>>>>>> --- a/drivers/nvmem/core.c
>>>>>> +++ b/drivers/nvmem/core.c
>>>>>> @@ -743,29 +743,21 @@ static struct nvmem_cell *nvmem_cell_get_from_list(const char *cell_id)
>>>>>>
>>>>>>     #if IS_ENABLED(CONFIG_NVMEM) && IS_ENABLED(CONFIG_OF)
>>>>>>     /**
>>>>>> - * of_nvmem_cell_get() - Get a nvmem cell from given device node and cell id
>>>>>> + * of_nvmem_cell_get_direct() - Get a nvmem cell from given device node
>>>>>>      *
>>>>>> - * @dev node: Device tree node that uses the nvmem cell
>>>>>> - * @id: nvmem cell name from nvmem-cell-names property.
>>>>>> + * @dev node: Device tree node that uses nvmem cell
>>>>>>      *
>>>>>>      * Return: Will be an ERR_PTR() on error or a valid pointer
>>>>>>      * to a struct nvmem_cell.  The nvmem_cell will be freed by the
>>>>>>      * nvmem_cell_put().
>>>>>>      */
>>>>>> -struct nvmem_cell *of_nvmem_cell_get(struct device_node *np,
>>>>>> -					    const char *name)
>>>>>> +struct nvmem_cell *of_nvmem_cell_get_direct(struct device_node *cell_np)
>>>>>>     {
>>>>>> -	struct device_node *cell_np, *nvmem_np;
>>>>>> +	struct device_node *nvmem_np;
>>>>>>     	struct nvmem_cell *cell;
>>>>>>     	struct nvmem_device *nvmem;
>>>>>>     	const __be32 *addr;
>>>>>> -	int rval, len, index;
>>>>>> -
>>>>>> -	index = of_property_match_string(np, "nvmem-cell-names", name);
>>>>>> -
>>>>>> -	cell_np = of_parse_phandle(np, "nvmem-cells", index);
>>>>>> -	if (!cell_np)
>>>>>> -		return ERR_PTR(-EINVAL);
>>>>>> +	int rval, len;
>>>>>>
>>>>>>     	nvmem_np = of_get_next_parent(cell_np);
>>>>>>     	if (!nvmem_np)
>>>>>> @@ -824,6 +816,32 @@ err_mem:
>>>>>>
>>>>>>     	return ERR_PTR(rval);
>>>>>>     }
>>>>>> +EXPORT_SYMBOL_GPL(of_nvmem_cell_get_direct);
>>>>>> +
>>>>>> +/**
>>>>>> + * of_nvmem_cell_get() - Get a nvmem cell from given device node and cell id
>>>>>> + *
>>>>>> + * @dev node: Device tree node that uses the nvmem cell
>>>>>> + * @id: nvmem cell name from nvmem-cell-names property.
>>>>>> + *
>>>>>> + * Return: Will be an ERR_PTR() on error or a valid pointer
>>>>>> + * to a struct nvmem_cell.  The nvmem_cell will be freed by the
>>>>>> + * nvmem_cell_put().
>>>>>> + */
>>>>>> +struct nvmem_cell *of_nvmem_cell_get(struct device_node *np,
>>>>>> +					    const char *name)
>>>>>> +{
>>>>>> +	struct device_node *cell_np;
>>>>>> +	int index;
>>>>>> +
>>>>>> +	index = of_property_match_string(np, "nvmem-cell-names", name);
>>>>>> +
>>>>>> +	cell_np = of_parse_phandle(np, "nvmem-cells", index);
>>>>>> +	if (!cell_np)
>>>>>> +		return ERR_PTR(-EINVAL);
>>>>>> +
>>>>>> +	return of_nvmem_cell_get_direct(cell_np);
>>>>>> +}
>>>>>>     EXPORT_SYMBOL_GPL(of_nvmem_cell_get);
>>>>>>     #endif
>>>>>>
>>>>>> diff --git a/include/linux/nvmem-consumer.h b/include/linux/nvmem-consumer.h
>>>>>> index 9bb77d3..bf879fc 100644
>>>>>> --- a/include/linux/nvmem-consumer.h
>>>>>> +++ b/include/linux/nvmem-consumer.h
>>>>>> @@ -136,6 +136,7 @@ static inline int nvmem_device_write(struct nvmem_device *nvmem,
>>>>>>     #endif /* CONFIG_NVMEM */
>>>>>>
>>>>>>     #if IS_ENABLED(CONFIG_NVMEM) && IS_ENABLED(CONFIG_OF)
>>>>>> +struct nvmem_cell *of_nvmem_cell_get_direct(struct device_node *cell_np);
>>>>>>     struct nvmem_cell *of_nvmem_cell_get(struct device_node *np,
>>>>>>     				     const char *name);
>>>>>>     struct nvmem_device *of_nvmem_device_get(struct device_node *np,
>>>>>>