Re: [PATCH 2/5] nvmem: expose nvmem cells as cdev

[Sascha Hauer <s.hauer@pengutronix.de>]

On Fri, Mar 22, 2024 at 05:45:56PM +0100, Marco Felsch wrote:
> Expose the nvmem cells via cdevs which is our equivalent to the Linux
> sysfs exposure. This allows the easier user queries for board code and
> shell. Keep the Linux function name scheme for
> nvmem_populate_sysfs_cells() to reduce the diff for nvmem_register()
> function.
> 
> Signed-off-by: Marco Felsch <m.felsch@pengutronix.de>
> ---
>  drivers/nvmem/core.c | 109 +++++++++++++++++++++++++++++++++++++++++++
>  1 file changed, 109 insertions(+)
> 
> diff --git a/drivers/nvmem/core.c b/drivers/nvmem/core.c
> index 657025daddb3..b4a29e4b67f3 100644
> --- a/drivers/nvmem/core.c
> +++ b/drivers/nvmem/core.c
> @@ -44,6 +44,8 @@ struct nvmem_cell_entry {
>  	struct device_node	*np;
>  	struct nvmem_device	*nvmem;
>  	struct list_head	node;
> +
> +	struct cdev		cdev;
>  };
>  
>  struct nvmem_cell {
> @@ -144,6 +146,107 @@ static struct nvmem_device *of_nvmem_find(struct device_node *nvmem_np)
>  	return NULL;
>  }
>  
> +static struct nvmem_cell *nvmem_create_cell(struct nvmem_cell_entry *entry,
> +					    const char *id, int index);
> +
> +static ssize_t nvmem_cell_cdev_read(struct cdev *cdev, void *buf, size_t count,
> +				    loff_t offset, unsigned long flags)
> +{
> +	struct nvmem_cell_entry *entry;
> +	struct nvmem_cell *cell = NULL;
> +	size_t cell_sz, read_len;
> +	void *content;
> +
> +	entry = container_of(cdev, struct nvmem_cell_entry, cdev);
> +	cell = nvmem_create_cell(entry, entry->name, 0);
> +	if (IS_ERR(cell))
> +		return PTR_ERR(cell);
> +
> +	if (!cell)
> +		return -EINVAL;

>From looking at the implementation of nvmem_create_cell() I'd say this
can't happen.

> +
> +	content = nvmem_cell_read(cell, &cell_sz);
> +	if (IS_ERR(content)) {
> +		read_len = PTR_ERR(content);
> +		goto destroy_cell;
> +	}
> +
> +	read_len = min_t(unsigned int, cell_sz - offset, count);
> +	memcpy(buf, content + offset, read_len);
> +	kfree(content);
> +
> +destroy_cell:
> +	kfree_const(cell->id);
> +	kfree(cell);
> +
> +	return read_len;
> +}
> +
> +static ssize_t nvmem_cell_cdev_write(struct cdev *cdev, const void *buf, size_t count,
> +				     loff_t offset, unsigned long flags)
> +{
> +	struct nvmem_cell_entry *entry;
> +	struct nvmem_cell *cell;
> +	int ret;
> +
> +	entry = container_of(cdev, struct nvmem_cell_entry, cdev);
> +
> +	if (!entry->nvmem->reg_write)
> +		return -EPERM;
> +
> +	if (offset >= entry->bytes)
> +		return -EFBIG;
> +
> +	if (offset + count > entry->bytes)
> +		count = entry->bytes - offset;
> +
> +	cell = nvmem_create_cell(entry, entry->name, 0);
> +	if (IS_ERR(cell))
> +		return PTR_ERR(cell);
> +
> +	if (!cell)
> +		return -EINVAL;
> +
> +	ret = nvmem_cell_write(cell, buf, count);
> +
> +	kfree_const(cell->id);
> +	kfree(cell);
> +
> +	return ret;
> +}
> +
> +static struct cdev_operations nvmem_cell_chrdev_ops = {
> +	.read  = nvmem_cell_cdev_read,
> +	.write = nvmem_cell_cdev_write,
> +};
> +
> +static int nvmem_populate_sysfs_cells(struct nvmem_device *nvmem)
> +{
> +	struct device *dev = &nvmem->dev;
> +	struct nvmem_cell_entry *entry;
> +
> +	if (list_empty(&nvmem->cells))
> +		return 0;

This is unnecessary.

> +
> +	list_for_each_entry(entry, &nvmem->cells, node) {
> +		struct cdev *cdev;
> +		int ret;
> +
> +		cdev = &entry->cdev;
> +		cdev->name = xasprintf("%s.%s", dev_name(dev),
> +				       kbasename(entry->name));
> +		cdev->ops = &nvmem_cell_chrdev_ops;
> +		cdev->dev = dev;
> +		cdev->size = entry->bytes;
> +
> +		ret = devfs_create(cdev);
> +		if (ret)
> +			return ret;
> +	}

Can't we just register a cdev when the cell is actually created? Why do
we iterate over all cells instead?

I am looking at the corresponding kernel code and I wonder how
u-boot-env is supposed to work. In u_boot_env_probe() first
nvmem_register() is called and nvmem_add_one_cell() for each variable
afterwards. nvmem_populate_sysfs_cells() is called during
nvmem_register(), so how are the variables added later are supposed to
get a sysfs entry?

> +
> +	return 0;
> +}
> +
>  static void nvmem_cell_entry_add(struct nvmem_cell_entry *cell)
>  {
>  	list_add_tail(&cell->node, &cell->nvmem->cells);
> @@ -337,6 +440,12 @@ struct nvmem_device *nvmem_register(const struct nvmem_config *config)
>  		}
>  	}
>  
> +	rval = nvmem_populate_sysfs_cells(nvmem);
> +	if (rval) {
> +		kfree(nvmem);

It's fine returning an error without cleaning up properly, but freeing
the memory on an half registered device is leading to memory
corruptions which must be fixed. We have the same in barebox master
already:

>        rval = register_device(&nvmem->dev);
>        if (rval) {
>                kfree(nvmem);
>                return ERR_PTR(rval);
>        }
>
>        if (!config->cdev) {
>                rval = nvmem_register_cdev(nvmem, config->name);
>                if (rval) {
>                        kfree(nvmem);

Either we unregister the previously registered device before freeing the
memory or we keep the allocation, but freeing the memory without
unregistering the device is wrong.

>                        return ERR_PTR(rval);
>                }
>        }

Sascha


-- 
Pengutronix e.K.                           |                             |
Steuerwalder Str. 21                       | http://www.pengutronix.de/  |
31137 Hildesheim, Germany                  | Phone: +49-5121-206917-0    |
Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-5555 |