Re: [BUG] hp-wmi-sensors: probe of 8F1F6435-9F42-42C8-BADC-0E9424F20C9A failed with error -22

[James Seo <james@equiv.tech>]

On Fri, Oct 27, 2023 at 05:07:09PM +0200, Lukasz Stelmach wrote:
> Hi,

Hi Łukasz, thanks for the report.

> I've got HP EliteDesk 800 G6 Tower PC running Linux 6.1 from Debian 12.
> I managed to build the hp-wmi-sensors out of tree. When I loaded it I
> got EINVAL.
> 
>     hp-wmi-sensors: probe of 8F1F6435-9F42-42C8-BADC-0E9424F20C9A failed with error -22
> 
> I managed to track it down. And it happens in check_wobj() called from
> check_platform_events_wobj() because in the for loop when prop==0 the
> type is ACPI_TYPE_BUFFER instead of ACPI_TYPE_STRING. When I bypass this
> particular check like this
> 
>     if (prop == 0 && type == ACPI_TYPE_BUFFER)
>             continue;
> 
> everything else works like charm and I can read senosrs via sysfs.

I'm surprised that your bypass is working! That check is intended to make it
safe to assume HPBIOS_BIOSNumericSensor.Name is a ACPI_TYPE_STRING. Which the
driver continues to do, of course.

Maybe .Name is being erroneously reported as an ACPI_TYPE_BUFFER on just that
model, but is in reality still a string? Dealing with that string in e.g.
hp_wmi_strdup() would still work because .string.pointer and .buffer.pointer
seem to end up at the same offset in a union acpi_object, at least for now.

> I'd like to perpare a proper patch, but I've got no idea how to do properly
> work this quirk around. What are your suggestions?

Could you share output from acpidump and sensors before we go any further?

Thanks.

-James Seo

> Kind regards,
> -- 
> Łukasz Stelmach
> Samsung R&D Institute Poland
> Samsung Electronics