[PATCH v2] pinctrl/pinconfig: add debug interface

[swarren@wwwdotorg.org (Stephen Warren)]

On 04/03/2013 01:47 PM, Linus Walleij wrote:
> From: Laurent Meunier <laurent.meunier@st.com>
> 
> This update adds a debugfs interface to modify a pin configuration
> for a given state in the pinctrl map. This allows to modify the
> configuration for a non-active state, typically sleep state.
> This configuration is not applied right away, but only when the state
> will be entered.
> 
> This solution is mandated for us by HW validation: in order
> to test and verify several pin configurations during sleep without
> recompiling the software.
> 
> diff --git a/drivers/pinctrl/pinconf.c b/drivers/pinctrl/pinconf.c

> +enum pinconf_dbg_request_type {
> +	PINCONF_DBG_REQ_TYPE_INVALID,
> +	PINCONF_DBG_REQ_TYPE_MODIFY,
> +};

I'm not sure why that really needs an enum yet, since only one operation
is supported.

> +struct dbg_cfg {
> +	enum pinconf_dbg_request_type req_type;
> +	enum pinctrl_map_type map_type;
> +	char dev_name[MAX_NAME_LEN+1];
> +	char state_name[MAX_NAME_LEN+1];
> +	char pin_name[MAX_NAME_LEN+1];
> +	char config[MAX_NAME_LEN+1];
> +};

Many of those fields are only used internally to
pinconf_dbg_config_write(). Can't they be stored on the stack there
rather than globally.

> +/**
> + * pinconf_dbg_config_print() - display the pinctrl config from the pinctrl
> + * map, of a pin/state pair based on pinname and state that have been
> + * selected with the debugfs entries pinconf-name and pinconf-state
> + * @s: contains the 32bits config to be written
> + * @d: not used
> + */

This comment seems stale; I don't believe there are any pinconf-name and
pinconf-state files; aren't they part of the data written to the one
debugfs file each time?

> +static int pinconf_dbg_config_print(struct seq_file *s, void *d)

> +		if (strncmp(map->dev_name, dbg->dev_name, MAX_NAME_LEN) != 0)

Wouldn't it be simpler to always ensure dbg->dev_name was
NULL-terminated, and just use !strcmp() here? Same for dbg->state_name
below. Same comment for the other function below.

> +	mutex_unlock(&pinctrl_mutex);

Don't you need the lock held to call confops->xxx() below, to make sure
that the driver isn't unregistered between finding it, and calling the
confops function?

> +	if (!found) {
> +		seq_printf(s, "No config found for dev/state/pin, expected:\n");
> +		seq_printf(s, "modify config_pins devname " \
> +			   "state pinname value\n");
> +	}

Hmmm. At least including the parsed values that were being searched for
might be useful?

> +		confops->pin_config_config_dbg_show(pctldev,
> +						    s, config);

I think that function name is wrong; two "config_" in a row. Does this
compile?

> +/**
> + * pinconf_dbg_config_write() - overwrite the pinctrl config in the pinctrl
> + * map, of a pin/state pair based on pinname and state that have been
> + * selected with the debugfs entries pinconf-name and pinconf-state

Similar stale comment.

> + * @user_buf: contains 'modify configs_pin devicename state pinname newvalue'

It might be useful to say which parts of that are literal strings and
which are variables/values to be changed?

> +static int pinconf_dbg_config_write(struct file *file,
> +	const char __user *user_buf, size_t count, loff_t *ppos)

> +	/* Get arg: 'modify' */
> +	if (!strncmp(b, "modify ", strlen("modify "))) {
> +		dbg->req_type = PINCONF_DBG_REQ_TYPE_MODIFY;
> +		b += strlen("modify ");
> +	} else {
> +		return -EINVAL;
> +	}

There's rather a lot of duplication of strings and strlen calls there.
Why not:

a) Start using strsep() for the very first fields too, so everything is
consistent.

b) Put the error-path first.

In other words:

	token = strsep(&b, " ");
	if (!token)
		return -EINVAL;
	if (!strcmp(token, "modify"))
		return -EINVAL;

> +	/* Get arg type: "config_pin" type supported so far */
> +	if (!strncmp(b, "config_pins ", strlen("config_pins "))) {
> +		dbg->map_type = PIN_MAP_TYPE_CONFIGS_PIN;
> +		b += strlen("config_pins ");
> +	} else {
> +		return -EINVAL;
> +	}

That could be transformed the same way.

> +	/* get arg 'device_name' */
> +	token = strsep(&b, " ");
> +	if (token == NULL)
> +		return -EINVAL;
> +	if (strlen(token) < MAX_NAME_LEN)
> +		sscanf(token, "%s", dbg->dev_name);
> +	else
> +		return -EINVAL;

It might make sense to put the error-handling first, and why sscanf not
strcpy? So:

	token = strsep(&b, " ");
	if (token == NULL)
		return -EINVAL;
	if (strlen(token) >= MAX_NAME_LEN)
		return -EINVAL;
	strcpy(dbg->dev_name, token);

(or even just use strncpy followed by explicit NULL termination and
truncate text that's too long, or kstrdup() it to avoid limits?)

> +	if (confops && confops->pin_config_dbg_parse) {
> +		for (i = 0; i < configs->num_configs; i++) {
> +			confops->pin_config_dbg_parse(pctldev,
> +						    dbg->config,
> +						    &configs->configs[i]);

I assume that "parse" function is intended to actually write the new
result into configs->configs[i]. That's more than parsing.
s/dbg_parse/dbg_modify/ perhaps to make this more explicit, and also
mention this in the kerneldoc for the confops structure in the header?