On Thursday 25 April 2024 14:27:32 Lyndon Sanche wrote:
> On Thu, Apr 25, 2024, at 2:12 PM, Pali Rohár wrote:
> > On Thursday 25 April 2024 11:27:57 Lyndon Sanche wrote:
> >> +int thermal_init(void)
> >> +{
> >> + int ret;
> >> + int supported_modes;
> >> +
> >> + ret = thermal_get_supported_modes(&supported_modes);
> >> +
> >> + if (ret != 0 || supported_modes == 0)
> >> + return -ENXIO;
> >> +
> >> + thermal_handler = kzalloc(sizeof(*thermal_handler), GFP_KERNEL);
> >> + if (!thermal_handler)
> >> + return -ENOMEM;
> >> + thermal_handler->profile_get = thermal_platform_profile_get;
> >> + thermal_handler->profile_set = thermal_platform_profile_set;
> >> +
> >> + if ((supported_modes >> DELL_QUIET) & 1)
> >> + set_bit(PLATFORM_PROFILE_QUIET, thermal_handler->choices);
> >> + if ((supported_modes >> DELL_COOL_BOTTOM) & 1)
> >> + set_bit(PLATFORM_PROFILE_COOL, thermal_handler->choices);
> >> + if ((supported_modes >> DELL_BALANCED) & 1)
> >> + set_bit(PLATFORM_PROFILE_BALANCED, thermal_handler->choices);
> >> + if ((supported_modes >> DELL_PERFORMANCE) & 1)
> >> + set_bit(PLATFORM_PROFILE_PERFORMANCE, thermal_handler->choices);
> >> +
> >> + platform_profile_register(thermal_handler);
> >> +
> >> + return 0;
> >> +}
> >> +
> >> +void thermal_cleanup(void)
> >> +{
> >> + platform_profile_remove();
> >> + kfree(thermal_handler);
> >> +}
> >> +
> >> static struct led_classdev mute_led_cdev = {
> >> .name = "platform::mute",
> >> .max_brightness = 1,
> >> @@ -2266,6 +2480,11 @@ static int __init dell_init(void)
> >> mute_led_registered = true;
> >> }
> >>
> >> + // Do not fail module if thermal modes not supported,
> >> + // just skip
> >> + if (thermal_init() != 0)
> >> + pr_warn("Unable to setup platform_profile, skipping");
> >
> > I think that -ENOMEM error should be failure of the loading the driver.
> > It does not make sense to continue of memory allocation failed.
> >
> > On the other hand when the thermal modes are not support (e.g. old
> > Latitude models) then there should not be a warning message. It is
> > expected that on systems without thermal modes the setup fails.
>
> Thank you for your feedback.
>
> I agree with your suggestion. -ENOMEM would indicate something bigger is amiss. I can add a check:
>
> If -ENOMEM, fail driver.
> If anything other error, skip, but do not show a message.
>
> Lyndon
Maybe you can simplify it more. thermal_init() could return 0 also for
the case when thermal modes are not supported. And dell_init() then can
unconditionally fail when thermal_init() returns non-zero value. It has
benefit that in case thermal_init() is extended in future for some other
fatal error, it would not be required to update also caller to handle
another error (and not just ENOMEM).
