Re: [PATCH v2 12/15] ACPI: platform_profile: Make sure all profile handlers agree on profile

["Ilpo Järvinen" <ilpo.jarvinen@linux.intel.com>]

On Sun, 27 Oct 2024, Mario Limonciello wrote:

> If for any reason multiple profile handlers don't agree on the profile
> set for the system then the value shown in sysfs can be wrong.
> 
> Explicitly check that they match.
> 
> Tested-by: Matthew Schwartz <matthew.schwartz@linux.dev>
> Signed-off-by: Mario Limonciello <mario.limonciello@amd.com>
> ---
>  drivers/acpi/platform_profile.c | 61 ++++++++++++++++++++++++---------
>  1 file changed, 45 insertions(+), 16 deletions(-)
> 
> diff --git a/drivers/acpi/platform_profile.c b/drivers/acpi/platform_profile.c
> index db2ebd0393cf7..d22c4eb5f0c36 100644
> --- a/drivers/acpi/platform_profile.c
> +++ b/drivers/acpi/platform_profile.c
> @@ -51,6 +51,45 @@ static unsigned long platform_profile_get_choices(void)
>  	return seen;
>  }
>  
> +/* expected to be called under mutex */

Don't add comments like this but enforce it with a lockdep annotation.

"mutex" would have been too vague anyway :-).

> +static int platform_profile_get_active(enum platform_profile_option *profile)
> +{
> +	struct platform_profile_handler *handler;
> +	enum platform_profile_option active = PLATFORM_PROFILE_LAST;
> +	enum platform_profile_option active2 = PLATFORM_PROFILE_LAST;
> +	int err;
> +
> +	list_for_each_entry(handler, &platform_profile_handler_list, list) {
> +		if (active == PLATFORM_PROFILE_LAST)
> +			err = handler->profile_get(handler, &active);
> +		else
> +			err = handler->profile_get(handler, &active2);
> +		if (err) {
> +			pr_err("Failed to get profile for handler %s\n", handler->name);
> +			return err;
> +		}
> +
> +		if (WARN_ON(active == PLATFORM_PROFILE_LAST))
> +			return -EINVAL;
> +		if (active2 == PLATFORM_PROFILE_LAST)
> +			continue;
> +
> +		if (active != active2) {
> +			pr_warn("Profile handlers don't agree on current profile\n");
> +			return -EINVAL;
> +		}
> +		active = active2;

This looked very confusing (IMO). How about this:

	enum platform_profile_option active = PLATFORM_PROFILE_LAST;
	enum platform_profile_option val;
	...

		err = handler->profile_get(handler, &val);
		if (err) {
			pr_err(...);
			return err;
		}

		if (WARN_ON(val == PLATFORM_PROFILE_LAST))
			return -EINVAL;

		if (active != val && active != PLATFORM_PROFILE_LAST) {
			pr_warn("Profile handlers don't agree on current profile\n");
			return -EINVAL;
		}
		active = val;

> +	}
> +
> +	/* Check that profile is valid index */
> +	if (WARN_ON((active < 0) || (active >= ARRAY_SIZE(profile_names))))

What does that < 0 check do? Should it be checked right after reading 
profile_get()? Or perhaps check both of these right there?

> +		return -EIO;
> +
> +	*profile = active;
> +
> +	return 0;
> +}
> +
>  static ssize_t platform_profile_choices_show(struct device *dev,
>  					struct device_attribute *attr,
>  					char *buf)
> @@ -80,24 +119,14 @@ static ssize_t platform_profile_show(struct device *dev,
>  	enum platform_profile_option profile = PLATFORM_PROFILE_BALANCED;
>  	int err;
>  
> -	err = mutex_lock_interruptible(&profile_lock);
> -	if (err)
> -		return err;
> -
> -	if (!cur_profile) {
> -		mutex_unlock(&profile_lock);
> -		return -ENODEV;
> +	scoped_cond_guard(mutex_intr, return -ERESTARTSYS, &profile_lock) {

scoped_cond_guard() conversion should be made in the guard patch?

> +		if (!platform_profile_is_registered())
> +			return -ENODEV;
> +		err = platform_profile_get_active(&profile);
> +		if (err)
> +			return err;
>  	}
>  
> -	err = cur_profile->profile_get(cur_profile, &profile);
> -	mutex_unlock(&profile_lock);
> -	if (err)
> -		return err;
> -
> -	/* Check that profile is valid index */
> -	if (WARN_ON((profile < 0) || (profile >= ARRAY_SIZE(profile_names))))
> -		return -EIO;
> -
>  	return sysfs_emit(buf, "%s\n", profile_names[profile]);
>  }
>  
> 

-- 
 i.