Hi all,
On Tue Mar 4, 2025 at 7:49 AM -05, Mario Limonciello wrote:
>
>
> On 3/4/25 02:38, Antheas Kapenekakis wrote:
>> On Tue, 4 Mar 2025 at 07:48, Mario Limonciello <superm1@xxxxxxxxxx> wrote:
>>>
>>> From: Mario Limonciello <mario.limonciello@xxxxxxx>
>>>
>>> When two drivers don't support all the same profiles the legacy interface
>>> only exports the common profiles.
>>>
>>> This causes problems for cases where one driver uses low-power but another
>>> uses quiet because the result is that neither is exported to sysfs.
>>>
>>> If one platform profile handler supports quiet and the other
>>> supports low power treat them as the same for the purpose of
>>> the sysfs interface.
>>>
>>> Fixes: 688834743d67 ("ACPI: platform_profile: Allow multiple handlers")
>>> Reported-by: Antheas Kapenekakis <lkml@xxxxxxxxxxx>
>>> Closes: https://lore.kernel.org/platform-driver-x86/e64b771e-3255-42ad-9257-5b8fc6c24ac9@xxxxxx/T/#mc068042dd29df36c16c8af92664860fc4763974b
>>> Signed-off-by: Mario Limonciello <mario.limonciello@xxxxxxx>
>>> ---
>>> drivers/acpi/platform_profile.c | 38 ++++++++++++++++++++++++++++++---
>>> 1 file changed, 35 insertions(+), 3 deletions(-)
>>>
>>> diff --git a/drivers/acpi/platform_profile.c b/drivers/acpi/platform_profile.c
>>> index 2ad53cc6aae53..d9a7cc5891734 100644
>>> --- a/drivers/acpi/platform_profile.c
>>> +++ b/drivers/acpi/platform_profile.c
>>> @@ -73,8 +73,20 @@ static int _store_class_profile(struct device *dev, void *data)
>>>
>>> lockdep_assert_held(&profile_lock);
>>> handler = to_pprof_handler(dev);
>>> - if (!test_bit(*bit, handler->choices))
>>> - return -EOPNOTSUPP;
>>> + if (!test_bit(*bit, handler->choices)) {
>>> + switch (*bit) {
>>> + case PLATFORM_PROFILE_QUIET:
>>> + *bit = PLATFORM_PROFILE_LOW_POWER;
>>> + break;
>>> + case PLATFORM_PROFILE_LOW_POWER:
>>> + *bit = PLATFORM_PROFILE_QUIET;
>>> + break;
>>> + default:
>>> + return -EOPNOTSUPP;
>>> + }
>>> + if (!test_bit(*bit, handler->choices))
>>> + return -EOPNOTSUPP;
>>> + }
>>>
>>> return handler->ops->profile_set(dev, *bit);
>>> }
>>> @@ -252,8 +264,16 @@ static int _aggregate_choices(struct device *dev, void *data)
>>> handler = to_pprof_handler(dev);
>>> if (test_bit(PLATFORM_PROFILE_LAST, aggregate))
>>> bitmap_copy(aggregate, handler->choices, PLATFORM_PROFILE_LAST);
>>> - else
>>> + else {
>>> + /* treat quiet and low power the same for aggregation purposes */
>>> + if (test_bit(PLATFORM_PROFILE_QUIET, handler->choices) &&
>>> + test_bit(PLATFORM_PROFILE_LOW_POWER, aggregate))
>>> + set_bit(PLATFORM_PROFILE_QUIET, aggregate);
>>> + else if (test_bit(PLATFORM_PROFILE_LOW_POWER, handler->choices) &&
>>> + test_bit(PLATFORM_PROFILE_QUIET, aggregate))
>>> + set_bit(PLATFORM_PROFILE_LOW_POWER, aggregate);
>>> bitmap_and(aggregate, handler->choices, aggregate, PLATFORM_PROFILE_LAST);
>>> + }
>>
>> So you end up showing both? If that's the case, isn't it equivalent to
>> just make amd-pmf show both quiet and low-power?
>>
>> I guess it is not ideal for framework devices. But if asus devices end
>> up showing both, then it should be ok for framework devices to show
>> both.
>>
>> I like the behavior of the V1 personally.
>
> No; this doesn't cause it to show both. It only causes one to show up.
> I confirmed it with a contrived situation on my laptop that forced
> multiple profile handlers that supported a mix.
>
>
> # cat /sys/firmware/acpi/platform_profile*
> low-power
> low-power balanced performance
>
> # cat /sys/class/platform-profile/platform-profile-*/profile
> quiet
> quiet
> quiet
> quiet
> quiet
> quiet
> quiet
> quiet
> quiet
> quiet
> quiet
> quiet
> quiet
> quiet
> quiet
> quiet
> quiet
> quiet
> quiet
> quiet
> quiet
> quiet
> quiet
> quiet
> low-power
>
>>
>>> return 0;
>>> }
>>> @@ -305,6 +325,13 @@ static int _aggregate_profiles(struct device *dev, void *data)
>>> if (err)
>>> return err;
>>>
>>> + /* treat low-power and quiet as the same */
>>> + if ((*profile == PLATFORM_PROFILE_LOW_POWER &&
>>> + val == PLATFORM_PROFILE_QUIET) ||
>>> + (*profile == PLATFORM_PROFILE_QUIET &&
>>> + val == PLATFORM_PROFILE_LOW_POWER))
>>> + *profile = val;
>>> +
>>> if (*profile != PLATFORM_PROFILE_LAST && *profile != val)
>>> *profile = PLATFORM_PROFILE_CUSTOM;
>>> else
>>> @@ -531,6 +558,11 @@ struct device *platform_profile_register(struct device *dev, const char *name,
>>> dev_err(dev, "Failed to register platform_profile class device with empty choices\n");
>>> return ERR_PTR(-EINVAL);
>>> }
>>> + if (test_bit(PLATFORM_PROFILE_QUIET, pprof->choices) &&
>>> + test_bit(PLATFORM_PROFILE_LOW_POWER, pprof->choices)) {
>>> + dev_err(dev, "Failed to register platform_profile class device with both quiet and low-power\n");
>>> + return ERR_PTR(-EINVAL);
>>> + }
>>
>> Can you avoid failing here? It caused a lot of issues in the past (the
>> WMI driver bails). a dev_err should be enough. Since you do not fail
>> maybe it can be increased to dev_crit.
>>
>> There is at least one driver that implements both currently, and a fix
>> would have to precede this patch.
>
> Oh, acer-wmi? Kurt; can you please comment? Are both simultaneous?
There are a few laptops supported by alienware-wmi that definitely have
both (including mine). The acer-wmi and the samsung-galaxybook drivers
also probe for available choices dynamically, so some of those devices
may be affected by this too.
So yes, we shouldn't fail registration here.
Anyway, I like this approach more than v1. What do you think about
constraining this fix to the legacy interface?
--
~ Kurt
>
>>
>>>
>>> guard(mutex)(&profile_lock);
>>>
>>> --
>>> 2.43.0
>>>
