Re: [PATCH v6 11/22] ACPI: platform_profile: Add name attribute to class interface

[Armin Wolf <W_Armin@gmx.de>]

Am 18.11.24 um 20:43 schrieb Armin Wolf:

> Am 09.11.24 um 05:41 schrieb Mario Limonciello:
>
>> The name attribute shows the name of the associated platform profile
>> handler.
>>
>> Tested-by: Mark Pearson <mpearson-lenovo@squebb.ca>
>> Signed-off-by: Mario Limonciello <mario.limonciello@amd.com>
>> ---
>>   drivers/acpi/platform_profile.c | 27 +++++++++++++++++++++++++++
>>   1 file changed, 27 insertions(+)
>>
>> diff --git a/drivers/acpi/platform_profile.c
>> b/drivers/acpi/platform_profile.c
>> index ef6af2c655524..4e2eda18f7f5f 100644
>> --- a/drivers/acpi/platform_profile.c
>> +++ b/drivers/acpi/platform_profile.c
>> @@ -25,8 +25,35 @@ static_assert(ARRAY_SIZE(profile_names) ==
>> PLATFORM_PROFILE_LAST);
>>
>>   static DEFINE_IDA(platform_profile_ida);
>>
>> +/**
>> + * name_show - Show the name of the profile handler
>> + * @dev: The device
>> + * @attr: The attribute
>> + * @buf: The buffer to write to
>> + * Return: The number of bytes written
>> + */
>> +static ssize_t name_show(struct device *dev,
>> +             struct device_attribute *attr,
>> +             char *buf)
>> +{
>> +    struct platform_profile_handler *handler = dev_get_drvdata(dev);
>> +
>> +    scoped_cond_guard(mutex_intr, return -ERESTARTSYS, &profile_lock) {
>> +        return sysfs_emit(buf, "%s\n", handler->name);
>> +    }
>> +    return -ERESTARTSYS;
>
> I still have a bad feeling about the locking inside the class
> attributes...
>
> Can we assume that no sysfs accesses occur after unregistering the
> class device?
>
> Even if this is not the case then the locking fails to protect the
> platform_profile_handler here.
> If the device is unregistered right after dev_get_drvdata() was
> called, then we would sill operate
> on possibly stale data once we take the profile_lock.
>
> Does someone have any clue how sysfs attributes act during removal?
>
I think i found the answer to my questions inside this patch series:
https://lore.kernel.org/linux-kernel/1390951311-15325-1-git-send-email-tj@kernel.org

It says that:

	kernfs / sysfs implement the "sever" semantic for userland accesses.
	When a node is removed, no further userland operations are allowed and
	the in-flight ones are drained before removal is finished.  This makes
	policing post-mortem userland accesses trivial for its users.

In this case taking the profile_lock when reading/writing class attributes seems to be unnecessary.
Please remove the unnecessary locking inside the class attributes.

Thanks,
Armin Wolf

> Thanks,
> Armin Wolf
>
>> +}
>> +
>> +static DEVICE_ATTR_RO(name);
>> +static struct attribute *profile_attrs[] = {
>> +    &dev_attr_name.attr,
>> +    NULL
>> +};
>> +ATTRIBUTE_GROUPS(profile);
>> +
>>   static const struct class platform_profile_class = {
>>       .name = "platform-profile",
>> +    .dev_groups = profile_groups,
>>   };
>>
>>   static ssize_t platform_profile_choices_show(struct device *dev,
>