Re: [PATCH v2] Introduce support for Systems Management Driver over WMI for Dell Systems

[Hans de Goede]

Hi,

On 9/17/20 7:22 AM, Bharathi, Divya wrote:

<snip>

> > > > > +
> > > > > +/**
> > > > > + * exit_enum_attributes() - Clear all attribute data
> > > > > + * @kset: The kset to free
> > > > > + *
> > > > > + * Clears all data allocated for this group of attributes  **/ void
> > > > > +exit_enum_attributes(struct kset *kset) {
> > > > > +	struct kobject *pos, *next;
> > > > > +
> > > > > +	mutex_lock(&kset_mutex);
> > > > > +	list_for_each_entry_safe(pos, next, &kset->list, entry) {
> > > > > +		sysfs_remove_group(pos, &enumeration_attr_group);
> > > > > +	}
> > > > > +	mutex_unlock(&kset_mutex);
> > > > > +	mutex_lock(&enum_mutex);
> > > > > +	kfree(enumeration_data);
> > > > > +	mutex_unlock(&enum_mutex);
> > > > > +}
> > > >
> > > > Since there is now only 1 kset for the main dir, you are now calling
> > > > sysfs_remove_group 4 times (for all the different times) on each entry
> > > > in the attributes dir. I guess this may fail silently, but it still is
> > > > not good. So this needs to be fixed.
> > > >
> > > > The remarks to this file also apply to the:
> > > >
> > > > drivers/platform/x86/dell-wmi-int-attributes.c
> > > > drivers/platform/x86/dell-wmi-string-attributes.c
> > > >
> > > > files.
>
> Since we maintained 4 different attribute groups under 1 kset, each time
> respective attribute group will be removed. And once all groups are
> removed, kset is deleted.

sysfs_remove_group() just does a kernfs_remove_by_name() for each
attribute in the group.

Since the integer_, enumeration_ and string_ attr_group-s all
have e.g. a current_value attribute that means that current_value
will be removed 3 times and for the 2nd and 3th call
kernfs_remove_by_name() will fail with -ENOENT.

Currently neither kernfs_remove_by_name() nor sysfs_remove_group() print
an error message for this, but still it is not very clean.

Instead why not do this:

int populate_enum_data(union acpi_object *enumeration_obj, int instance_id,
                        struct kobject *attr_name_kobj)
{
        int retval = sysfs_create_group(attr_name_kobj, &enumeration_attr_group);
        int i, next_obj;

        if (retval)
                goto out;

        mutex_lock(&wmi_priv.mutex);
	enumeration_data[instance_id].attr_name_kobj = attr_name_kobj;
	/* ^^^^^^^^^^^^^^^^ This line is new ^^^^^^^^^^^^^^^^^^^^^^^^*/
	...


void exit_enum_attributes(void)
{
        int i;

        for (i = 0; i < enumeration_instances_count; i++) {
		if (enumeration_data[instance_id].attr_name_kobj)
                        sysfs_remove_group(enumeration_data[instance_id].attr_name_kobj, &enumeration_attr_group);
	}

        kfree(enumeration_data);
}


That makes the teardown mirror the setup much more closely and as such is
a cleaner solution IMHO.

Regards,

Hans