Re: [PATCH v2 6/7] nvmem: split struct nvmem_device into refcounted and provider-owned data

[Johan Hovold]

On Mon, Feb 23, 2026 at 11:57:07AM +0100, Bartosz Golaszewski wrote:
> Data owned by the nvmem provider must not be part of the
> reference-counted struct nvmem_device.

It can be as long as it's not referenced after deregistration so this
claim is too strong.

It can be handled the way you are doing it here, but you could also just
use a boolean flag to indicate that no further callbacks should be made.

> Ahead of protecting that data
> with SRCU, move it into a separate structure the address of which is
> stored in nvmem_device.
 
> +/*
> + * Holds data owned by the provider of the nvmem implementation. This goes
> + * away immediately the moment nvmem_unregister() is called.
> + */
> +struct nvmem_impl {
> +	nvmem_reg_read_t	reg_read;
> +	nvmem_reg_write_t	reg_write;
> +	void			*priv;

And priv is just a pointer that is never used by nvmem core directly.

> +};

How about just wrapping the callbacks and calling the struct nvmem_ops
to make it more clear how this is used?

That is, that nvmem core guarantees that there will be no further
callbacks after deregistration.

>  struct nvmem_device {
>  	struct module		*owner;
>  	struct device		dev;
> +	struct nvmem_impl	*impl;
>  	int			stride;
>  	int			word_size;
>  	int			id;
> @@ -26,11 +37,8 @@ struct nvmem_device {
>  				   struct nvmem_cell_info *cell);
>  	const struct nvmem_keepout *keepout;
>  	unsigned int		nkeepout;
> -	nvmem_reg_read_t	reg_read;
> -	nvmem_reg_write_t	reg_write;
>  	struct gpio_desc	*wp_gpio;
>  	struct nvmem_layout	*layout;
> -	void *priv;
>  	bool			sysfs_cells_populated;
>  };

Johan