From: Jérôme de Bretagne <jerome.debretagne@gmail.com>

For some devices, Wi-Fi is entirely hard blocked by default making
the Wi-Fi radio unusable, except if rfkill is disabled as expected
on those models.

Commit c6a7c0b09d5f ("wifi: ath12k: Add Support for enabling or
disabling specific features based on ACPI bitflag") added a way to
support features set via ACPI, including the DISABLE_RFKILL bit.

Add a disable-rfkill property to expose the DISABLE_RFKILL bit
equivalent for devices described by a Devicetree instead of ACPI.

Signed-off-by: Jérôme de Bretagne <jerome.debretagne@gmail.com>
---
 Documentation/devicetree/bindings/net/wireless/ieee80211.yaml | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/Documentation/devicetree/bindings/net/wireless/ieee80211.yaml b/Documentation/devicetree/bindings/net/wireless/ieee80211.yaml
index d89f7a3f88a71d45d6f4ab2ae909eae09cbcaf9a..c10a4675640be947cd0b5eaec2c7ff367fd93945 100644
--- a/Documentation/devicetree/bindings/net/wireless/ieee80211.yaml
+++ b/Documentation/devicetree/bindings/net/wireless/ieee80211.yaml
@@ -29,6 +29,12 @@ properties:
       different 5 GHz subbands. Using them incorrectly could not work or
       decrease performance noticeably
 
+  disable-rfkill:
+    type: boolean
+    description:
+      Disable rfkill for some devices on which Wi-Fi would be entirely hard
+      blocked by default otherwise
+
 additionalProperties: true
 
 examples:

-- 
2.47.3

On 20/12/2025 00:21, Jérôme de Bretagne via B4 Relay wrote:
> From: Jérôme de Bretagne <jerome.debretagne@gmail.com>
> 
> For some devices, Wi-Fi is entirely hard blocked by default making
> the Wi-Fi radio unusable, except if rfkill is disabled as expected
> on those models.
> 
> Commit c6a7c0b09d5f ("wifi: ath12k: Add Support for enabling or
> disabling specific features based on ACPI bitflag") added a way to
> support features set via ACPI, including the DISABLE_RFKILL bit.
> 
> Add a disable-rfkill property to expose the DISABLE_RFKILL bit
> equivalent for devices described by a Devicetree instead of ACPI.
> 
> Signed-off-by: Jérôme de Bretagne <jerome.debretagne@gmail.com>
> ---
>   Documentation/devicetree/bindings/net/wireless/ieee80211.yaml | 6 ++++++
>   1 file changed, 6 insertions(+)
> 
> diff --git a/Documentation/devicetree/bindings/net/wireless/ieee80211.yaml b/Documentation/devicetree/bindings/net/wireless/ieee80211.yaml
> index d89f7a3f88a71d45d6f4ab2ae909eae09cbcaf9a..c10a4675640be947cd0b5eaec2c7ff367fd93945 100644
> --- a/Documentation/devicetree/bindings/net/wireless/ieee80211.yaml
> +++ b/Documentation/devicetree/bindings/net/wireless/ieee80211.yaml
> @@ -29,6 +29,12 @@ properties:
>         different 5 GHz subbands. Using them incorrectly could not work or
>         decrease performance noticeably
> 
> +  disable-rfkill:
> +    type: boolean
> +    description:
> +      Disable rfkill for some devices on which Wi-Fi would be entirely hard
> +      blocked by default otherwise
> +
>   additionalProperties: true
> 
>   examples:
> 
> --
> 2.47.3
> 
> 
> 

Is this really a hardware description though ?

Its really more of a logical/functional description. It tells the 
runtime what todo, not what the hardware is.

You could also have a list of quirks in ath12k for this or have a 
user-space utility look for the appropriate platform device string name 
and disable rfkill.

I think this logic belongs in drivers/net/wireless/ath/ath12k/ 
triggering on a compat string.

Should be achievable.

---
bod

On Sat, Dec 20, 2025 at 06:04:00AM +0000, Bryan O'Donoghue wrote:
> On 20/12/2025 00:21, Jérôme de Bretagne via B4 Relay wrote:
> > From: Jérôme de Bretagne <jerome.debretagne@gmail.com>
> > 
> > For some devices, Wi-Fi is entirely hard blocked by default making
> > the Wi-Fi radio unusable, except if rfkill is disabled as expected
> > on those models.
> > 
> > Commit c6a7c0b09d5f ("wifi: ath12k: Add Support for enabling or
> > disabling specific features based on ACPI bitflag") added a way to
> > support features set via ACPI, including the DISABLE_RFKILL bit.
> > 
> > Add a disable-rfkill property to expose the DISABLE_RFKILL bit
> > equivalent for devices described by a Devicetree instead of ACPI.
> > 
> > Signed-off-by: Jérôme de Bretagne <jerome.debretagne@gmail.com>
> > ---
> >   Documentation/devicetree/bindings/net/wireless/ieee80211.yaml | 6 ++++++
> >   1 file changed, 6 insertions(+)
> > 
> > diff --git a/Documentation/devicetree/bindings/net/wireless/ieee80211.yaml b/Documentation/devicetree/bindings/net/wireless/ieee80211.yaml
> > index d89f7a3f88a71d45d6f4ab2ae909eae09cbcaf9a..c10a4675640be947cd0b5eaec2c7ff367fd93945 100644
> > --- a/Documentation/devicetree/bindings/net/wireless/ieee80211.yaml
> > +++ b/Documentation/devicetree/bindings/net/wireless/ieee80211.yaml
> > @@ -29,6 +29,12 @@ properties:
> >         different 5 GHz subbands. Using them incorrectly could not work or
> >         decrease performance noticeably
> > 
> > +  disable-rfkill:
> > +    type: boolean
> > +    description:
> > +      Disable rfkill for some devices on which Wi-Fi would be entirely hard
> > +      blocked by default otherwise
> > +
> >   additionalProperties: true
> > 
> >   examples:
> > 
> > --
> > 2.47.3
> > 
> > 
> > 
> 
> Is this really a hardware description though ?
> 
> Its really more of a logical/functional description. It tells the runtime
> what todo, not what the hardware is.
> 
> You could also have a list of quirks in ath12k for this or have a user-space
> utility look for the appropriate platform device string name and disable
> rfkill.
> 
> I think this logic belongs in drivers/net/wireless/ath/ath12k/ triggering on
> a compat string.

This is good point. Either this could be deducible from the compatible
or this should actually describe the hardware and whatever is there
wired/configured, not what OS should do.

Best regards,
Krzysztof

Le sam. 20 déc. 2025 à 10:12, Krzysztof Kozlowski <krzk@kernel.org> a écrit :
>
> On Sat, Dec 20, 2025 at 06:04:00AM +0000, Bryan O'Donoghue wrote:
> > On 20/12/2025 00:21, Jérôme de Bretagne via B4 Relay wrote:
> > > From: Jérôme de Bretagne <jerome.debretagne@gmail.com>
> > >
> > > For some devices, Wi-Fi is entirely hard blocked by default making
> > > the Wi-Fi radio unusable, except if rfkill is disabled as expected
> > > on those models.
> > >
> > > Commit c6a7c0b09d5f ("wifi: ath12k: Add Support for enabling or
> > > disabling specific features based on ACPI bitflag") added a way to
> > > support features set via ACPI, including the DISABLE_RFKILL bit.
> > >
> > > Add a disable-rfkill property to expose the DISABLE_RFKILL bit
> > > equivalent for devices described by a Devicetree instead of ACPI.
> > >
> > > Signed-off-by: Jérôme de Bretagne <jerome.debretagne@gmail.com>
> > > ---
> > >   Documentation/devicetree/bindings/net/wireless/ieee80211.yaml | 6 ++++++
> > >   1 file changed, 6 insertions(+)
> > >
> > > diff --git a/Documentation/devicetree/bindings/net/wireless/ieee80211.yaml b/Documentation/devicetree/bindings/net/wireless/ieee80211.yaml
> > > index d89f7a3f88a71d45d6f4ab2ae909eae09cbcaf9a..c10a4675640be947cd0b5eaec2c7ff367fd93945 100644
> > > --- a/Documentation/devicetree/bindings/net/wireless/ieee80211.yaml
> > > +++ b/Documentation/devicetree/bindings/net/wireless/ieee80211.yaml
> > > @@ -29,6 +29,12 @@ properties:
> > >         different 5 GHz subbands. Using them incorrectly could not work or
> > >         decrease performance noticeably
> > >
> > > +  disable-rfkill:
> > > +    type: boolean
> > > +    description:
> > > +      Disable rfkill for some devices on which Wi-Fi would be entirely hard
> > > +      blocked by default otherwise
> > > +
> > >   additionalProperties: true
> > >
> > >   examples:
> > >
> > > --
> > > 2.47.3
> > >
> > >
> > >
> >
> > Is this really a hardware description though ?
> >
> > Its really more of a logical/functional description. It tells the runtime
> > what todo, not what the hardware is.
> >
> > You could also have a list of quirks in ath12k for this or have a user-space
> > utility look for the appropriate platform device string name and disable
> > rfkill.
> >
> > I think this logic belongs in drivers/net/wireless/ath/ath12k/ triggering on
> > a compat string.
>
> This is good point. Either this could be deducible from the compatible

Thank you Bryan and Krzysztof for your feedback, I will drop the
disable-rfkill patches from this patchset in v6 then. I will work on
a separate patch using a list of quirks in ath12k as suggested.

> or this should actually describe the hardware and whatever is there
> wired/configured, not what OS should do.
>
> Best regards,
> Krzysztof

Best regards,
Jérôme

On Sat, Dec 20, 2025 at 10:12:14AM +0100, Krzysztof Kozlowski wrote:
> On Sat, Dec 20, 2025 at 06:04:00AM +0000, Bryan O'Donoghue wrote:
> > On 20/12/2025 00:21, Jérôme de Bretagne via B4 Relay wrote:
> > > From: Jérôme de Bretagne <jerome.debretagne@gmail.com>
> > > 
> > > For some devices, Wi-Fi is entirely hard blocked by default making
> > > the Wi-Fi radio unusable, except if rfkill is disabled as expected
> > > on those models.
> > > 
> > > Commit c6a7c0b09d5f ("wifi: ath12k: Add Support for enabling or
> > > disabling specific features based on ACPI bitflag") added a way to
> > > support features set via ACPI, including the DISABLE_RFKILL bit.
> > > 
> > > Add a disable-rfkill property to expose the DISABLE_RFKILL bit
> > > equivalent for devices described by a Devicetree instead of ACPI.
> > > 
> > > Signed-off-by: Jérôme de Bretagne <jerome.debretagne@gmail.com>
> > > ---
> > >   Documentation/devicetree/bindings/net/wireless/ieee80211.yaml | 6 ++++++
> > >   1 file changed, 6 insertions(+)
> > > 
> > > diff --git a/Documentation/devicetree/bindings/net/wireless/ieee80211.yaml b/Documentation/devicetree/bindings/net/wireless/ieee80211.yaml
> > > index d89f7a3f88a71d45d6f4ab2ae909eae09cbcaf9a..c10a4675640be947cd0b5eaec2c7ff367fd93945 100644
> > > --- a/Documentation/devicetree/bindings/net/wireless/ieee80211.yaml
> > > +++ b/Documentation/devicetree/bindings/net/wireless/ieee80211.yaml
> > > @@ -29,6 +29,12 @@ properties:
> > >         different 5 GHz subbands. Using them incorrectly could not work or
> > >         decrease performance noticeably
> > > 
> > > +  disable-rfkill:
> > > +    type: boolean
> > > +    description:
> > > +      Disable rfkill for some devices on which Wi-Fi would be entirely hard
> > > +      blocked by default otherwise
> > > +
> > >   additionalProperties: true
> > > 
> > >   examples:
> > > 
> > > --
> > > 2.47.3
> > > 
> > > 
> > > 
> > 
> > Is this really a hardware description though ?
> > 
> > Its really more of a logical/functional description. It tells the runtime
> > what todo, not what the hardware is.
> > 
> > You could also have a list of quirks in ath12k for this or have a user-space
> > utility look for the appropriate platform device string name and disable
> > rfkill.
> > 
> > I think this logic belongs in drivers/net/wireless/ath/ath12k/ triggering on
> > a compat string.
> 
> This is good point. Either this could be deducible from the compatible
> or this should actually describe the hardware and whatever is there
> wired/configured, not what OS should do.

One of the examples _might_ be broken-rfkill, e.g. if the chip expects
to have the actual rfkill control from the EC, but the board doesn't
provide one.

-- 
With best wishes
Dmitry

On 12/20/25 5:05 PM, Dmitry Baryshkov wrote:
> On Sat, Dec 20, 2025 at 10:12:14AM +0100, Krzysztof Kozlowski wrote:
>> On Sat, Dec 20, 2025 at 06:04:00AM +0000, Bryan O'Donoghue wrote:
>>> On 20/12/2025 00:21, Jérôme de Bretagne via B4 Relay wrote:
>>>> From: Jérôme de Bretagne <jerome.debretagne@gmail.com>
>>>>
>>>> For some devices, Wi-Fi is entirely hard blocked by default making
>>>> the Wi-Fi radio unusable, except if rfkill is disabled as expected
>>>> on those models.
>>>>
>>>> Commit c6a7c0b09d5f ("wifi: ath12k: Add Support for enabling or
>>>> disabling specific features based on ACPI bitflag") added a way to
>>>> support features set via ACPI, including the DISABLE_RFKILL bit.
>>>>
>>>> Add a disable-rfkill property to expose the DISABLE_RFKILL bit
>>>> equivalent for devices described by a Devicetree instead of ACPI.
>>>>
>>>> Signed-off-by: Jérôme de Bretagne <jerome.debretagne@gmail.com>
>>>> ---
>>>>   Documentation/devicetree/bindings/net/wireless/ieee80211.yaml | 6 ++++++
>>>>   1 file changed, 6 insertions(+)
>>>>
>>>> diff --git a/Documentation/devicetree/bindings/net/wireless/ieee80211.yaml b/Documentation/devicetree/bindings/net/wireless/ieee80211.yaml
>>>> index d89f7a3f88a71d45d6f4ab2ae909eae09cbcaf9a..c10a4675640be947cd0b5eaec2c7ff367fd93945 100644
>>>> --- a/Documentation/devicetree/bindings/net/wireless/ieee80211.yaml
>>>> +++ b/Documentation/devicetree/bindings/net/wireless/ieee80211.yaml
>>>> @@ -29,6 +29,12 @@ properties:
>>>>         different 5 GHz subbands. Using them incorrectly could not work or
>>>>         decrease performance noticeably
>>>>
>>>> +  disable-rfkill:
>>>> +    type: boolean
>>>> +    description:
>>>> +      Disable rfkill for some devices on which Wi-Fi would be entirely hard
>>>> +      blocked by default otherwise
>>>> +
>>>>   additionalProperties: true
>>>>
>>>>   examples:
>>>>
>>>> --
>>>> 2.47.3
>>>>
>>>>
>>>>
>>>
>>> Is this really a hardware description though ?
>>>
>>> Its really more of a logical/functional description. It tells the runtime
>>> what todo, not what the hardware is.
>>>
>>> You could also have a list of quirks in ath12k for this or have a user-space
>>> utility look for the appropriate platform device string name and disable
>>> rfkill.
>>>
>>> I think this logic belongs in drivers/net/wireless/ath/ath12k/ triggering on
>>> a compat string.
>>
>> This is good point. Either this could be deducible from the compatible
>> or this should actually describe the hardware and whatever is there
>> wired/configured, not what OS should do.
> 
> One of the examples _might_ be broken-rfkill, e.g. if the chip expects
> to have the actual rfkill control from the EC, but the board doesn't
> provide one.

Hm, I haven't thought about the EC being involved previously. Maybe
+Maximilian would have an idea whether this could be a factor that we
simply haven't implemented yet in the SAM driver..

Konrad

+Cc Mani

Hi,

On 20-Dec-25 07:04, Bryan O'Donoghue wrote:
> On 20/12/2025 00:21, Jérôme de Bretagne via B4 Relay wrote:
>> From: Jérôme de Bretagne <jerome.debretagne@gmail.com>
>>
>> For some devices, Wi-Fi is entirely hard blocked by default making
>> the Wi-Fi radio unusable, except if rfkill is disabled as expected
>> on those models.
>>
>> Commit c6a7c0b09d5f ("wifi: ath12k: Add Support for enabling or
>> disabling specific features based on ACPI bitflag") added a way to
>> support features set via ACPI, including the DISABLE_RFKILL bit.
>>
>> Add a disable-rfkill property to expose the DISABLE_RFKILL bit
>> equivalent for devices described by a Devicetree instead of ACPI.
>>
>> Signed-off-by: Jérôme de Bretagne <jerome.debretagne@gmail.com>
>> ---
>>   Documentation/devicetree/bindings/net/wireless/ieee80211.yaml | 6 ++++++
>>   1 file changed, 6 insertions(+)
>>
>> diff --git a/Documentation/devicetree/bindings/net/wireless/ieee80211.yaml b/Documentation/devicetree/bindings/net/wireless/ieee80211.yaml
>> index d89f7a3f88a71d45d6f4ab2ae909eae09cbcaf9a..c10a4675640be947cd0b5eaec2c7ff367fd93945 100644
>> --- a/Documentation/devicetree/bindings/net/wireless/ieee80211.yaml
>> +++ b/Documentation/devicetree/bindings/net/wireless/ieee80211.yaml
>> @@ -29,6 +29,12 @@ properties:
>>         different 5 GHz subbands. Using them incorrectly could not work or
>>         decrease performance noticeably
>>
>> +  disable-rfkill:
>> +    type: boolean
>> +    description:
>> +      Disable rfkill for some devices on which Wi-Fi would be entirely hard
>> +      blocked by default otherwise
>> +
>>   additionalProperties: true
>>
>>   examples:
>>
>> -- 
>> 2.47.3
>>
>>
>>
> 
> Is this really a hardware description though ?

I would say yes it is. The wifi chip has an rfkill input pin and
things will be broken when that pin is hardwired to a fixed value
rather then being actually connected to a GPIO from say
the embedded controller.

So I think that we would need here is not a disable-rfkill property
but some way to indicate in the DT-node that the rfkill input pin
is not connected and thus should be ignored.

This (the rfkill input pin being not-connected) IMHO very much
is hw-description.

Also see the
"[PATCH 0/9] Add support for handling PCIe M.2 Key E connectors in devicetree"
series and then specifically:

https://lore.kernel.org/platform-driver-x86/20251112-pci-m2-e-v1-7-97413d6bf824@oss.qualcomm.com/

Which adds:

+  w_disable1-gpios:
+    description: GPIO controlled connection to W_DISABLE1# signal. This signal
+      is used by the system to disable WiFi radio in the M.2 card. Refer, PCI
+      Express M.2 Specification r4.0, sec 3.1.12.3 for more details.
+    maxItems: 1

What if there is no such GPIO, because the W_DISABLE1# signal is hardwired
in a specific implementation of the M.2 slot ?

In that case we will also need some way to propagate that info to the wifi
driver, having some sort of generic devicetree property for wifi-cards
which can be injected as a software-node property in the PCI-device being
instantiated for the WIFI card to let the driver no not to honor to
W_DISABLE1# signal will be useful here too and this is as hardware-description
as hardware-description can get.

So how about: "w_disable1-not-connected" + "w_disable2-not-connected" boolean
properties in a generic WIFI devicetree binding and also use that here?

> I think this logic belongs in drivers/net/wireless/ath/ath12k/ triggering on a compat string.

See above, I do not believe that abusing compat-strings for this is the way
to go.

Regards,

Hans

On 22/12/2025 11:23, Hans de Goede wrote:
> +Cc Mani
> 
> Hi,
> 
> On 20-Dec-25 07:04, Bryan O'Donoghue wrote:
>> On 20/12/2025 00:21, Jérôme de Bretagne via B4 Relay wrote:
>>> From: Jérôme de Bretagne <jerome.debretagne@gmail.com>
>>>
>>> For some devices, Wi-Fi is entirely hard blocked by default making
>>> the Wi-Fi radio unusable, except if rfkill is disabled as expected
>>> on those models.
>>>
>>> Commit c6a7c0b09d5f ("wifi: ath12k: Add Support for enabling or
>>> disabling specific features based on ACPI bitflag") added a way to
>>> support features set via ACPI, including the DISABLE_RFKILL bit.
>>>
>>> Add a disable-rfkill property to expose the DISABLE_RFKILL bit
>>> equivalent for devices described by a Devicetree instead of ACPI.
>>>
>>> Signed-off-by: Jérôme de Bretagne <jerome.debretagne@gmail.com>
>>> ---
>>>   Documentation/devicetree/bindings/net/wireless/ieee80211.yaml | 6 ++++++
>>>   1 file changed, 6 insertions(+)
>>>
>>> diff --git a/Documentation/devicetree/bindings/net/wireless/ieee80211.yaml b/Documentation/devicetree/bindings/net/wireless/ieee80211.yaml
>>> index d89f7a3f88a71d45d6f4ab2ae909eae09cbcaf9a..c10a4675640be947cd0b5eaec2c7ff367fd93945 100644
>>> --- a/Documentation/devicetree/bindings/net/wireless/ieee80211.yaml
>>> +++ b/Documentation/devicetree/bindings/net/wireless/ieee80211.yaml
>>> @@ -29,6 +29,12 @@ properties:
>>>         different 5 GHz subbands. Using them incorrectly could not work or
>>>         decrease performance noticeably
>>>
>>> +  disable-rfkill:
>>> +    type: boolean
>>> +    description:
>>> +      Disable rfkill for some devices on which Wi-Fi would be entirely hard
>>> +      blocked by default otherwise
>>> +
>>>   additionalProperties: true
>>>
>>>   examples:
>>>
>>> -- 
>>> 2.47.3
>>>
>>>
>>>
>>
>> Is this really a hardware description though ?
> 
> I would say yes it is. The wifi chip has an rfkill input pin and
> things will be broken when that pin is hardwired to a fixed value
> rather then being actually connected to a GPIO from say
> the embedded controller.

You still do not describe the hardware. Read my comment.

> 
> So I think that we would need here is not a disable-rfkill property
> but some way to indicate in the DT-node that the rfkill input pin
> is not connected and thus should be ignored.
> 
> This (the rfkill input pin being not-connected) IMHO very much
> is hw-description.
> 
> Also see the
> "[PATCH 0/9] Add support for handling PCIe M.2 Key E connectors in devicetree"
> series and then specifically:
> 
> https://lore.kernel.org/platform-driver-x86/20251112-pci-m2-e-v1-7-97413d6bf824@oss.qualcomm.com/
> 
> Which adds:
> 
> +  w_disable1-gpios:
> +    description: GPIO controlled connection to W_DISABLE1# signal. This signal
> +      is used by the system to disable WiFi radio in the M.2 card. Refer, PCI
> +      Express M.2 Specification r4.0, sec 3.1.12.3 for more details.
> +    maxItems: 1
> 
> What if there is no such GPIO, because the W_DISABLE1# signal is hardwired
> in a specific implementation of the M.2 slot ?
> 
> In that case we will also need some way to propagate that info to the wifi
> driver, having some sort of generic devicetree property for wifi-cards
> which can be injected as a software-node property in the PCI-device being
> instantiated for the WIFI card to let the driver no not to honor to
> W_DISABLE1# signal will be useful here too and this is as hardware-description
> as hardware-description can get.

Please do not use antipatterns as examples. Above patchset is obviously
wrong there, it's a terrible example but also completely different thing
(spec). This is not a spec here.


Best regards,
Krzysztof

On Mon, Dec 22, 2025 at 11:23:18AM +0100, Hans de Goede wrote:
> +Cc Mani
> 
> Hi,
> 
> On 20-Dec-25 07:04, Bryan O'Donoghue wrote:
> > On 20/12/2025 00:21, Jérôme de Bretagne via B4 Relay wrote:
> >> From: Jérôme de Bretagne <jerome.debretagne@gmail.com>
> >>
> >> For some devices, Wi-Fi is entirely hard blocked by default making
> >> the Wi-Fi radio unusable, except if rfkill is disabled as expected
> >> on those models.
> >>
> >> Commit c6a7c0b09d5f ("wifi: ath12k: Add Support for enabling or
> >> disabling specific features based on ACPI bitflag") added a way to
> >> support features set via ACPI, including the DISABLE_RFKILL bit.
> >>
> >> Add a disable-rfkill property to expose the DISABLE_RFKILL bit
> >> equivalent for devices described by a Devicetree instead of ACPI.
> >>
> >> Signed-off-by: Jérôme de Bretagne <jerome.debretagne@gmail.com>
> >> ---
> >>   Documentation/devicetree/bindings/net/wireless/ieee80211.yaml | 6 ++++++
> >>   1 file changed, 6 insertions(+)
> >>
> >> diff --git a/Documentation/devicetree/bindings/net/wireless/ieee80211.yaml b/Documentation/devicetree/bindings/net/wireless/ieee80211.yaml
> >> index d89f7a3f88a71d45d6f4ab2ae909eae09cbcaf9a..c10a4675640be947cd0b5eaec2c7ff367fd93945 100644
> >> --- a/Documentation/devicetree/bindings/net/wireless/ieee80211.yaml
> >> +++ b/Documentation/devicetree/bindings/net/wireless/ieee80211.yaml
> >> @@ -29,6 +29,12 @@ properties:
> >>         different 5 GHz subbands. Using them incorrectly could not work or
> >>         decrease performance noticeably
> >>
> >> +  disable-rfkill:
> >> +    type: boolean
> >> +    description:
> >> +      Disable rfkill for some devices on which Wi-Fi would be entirely hard
> >> +      blocked by default otherwise
> >> +
> >>   additionalProperties: true
> >>
> >>   examples:
> >>
> >> -- 
> >> 2.47.3
> >>
> >>
> >>
> > 
> > Is this really a hardware description though ?
> 
> I would say yes it is. The wifi chip has an rfkill input pin and
> things will be broken when that pin is hardwired to a fixed value
> rather then being actually connected to a GPIO from say
> the embedded controller.
> 

IIUC, even if the M.2 slot has the W_DISABLE1# signal routed from the host,
the device won't make use of it as there is no physical connection. So you want
the WLAN driver to change the state through SW?

> So I think that we would need here is not a disable-rfkill property
> but some way to indicate in the DT-node that the rfkill input pin
> is not connected and thus should be ignored.
> 
> This (the rfkill input pin being not-connected) IMHO very much
> is hw-description.
> 

Though we can argue this way, I would prefer to handle it in the driver. For
example, with my M.2 series, we will end up describing the M.2 slot:

	connector {
		compatible = "pcie-m2-e-connector";
		w-disable1-gpios = <&tlmm 117 GPIO_ACTIVE_LOW>;
		...
		ports {
			...
			endpoint@0 {
				reg = <0>;
				remote-endpoint = <&pcie4_port0_ep>;
			};
		};
	};

Then if we use a DT property to convey the RFKILL pin state of the device, we
would need to describe the endpoint device in DT and hardcode the state:

	&pcie4_port0 {
		...
		port {
			pcie4_port0_ep: endpoint {
				remote-endpoint = <&m2_e_pcie_ep>;
				disable-rfkill;
			};
		};
	};

So this will essentially make the M.2 device non-swappable unless you change the
DT since you've how hardcoded the device property in the binding. This is
something I try to avoid to make the M.2 slot really swappable.

For this reason, I would prefer to handle the RFKILL state in the WLAN driver
using the device specific compatible. This will be problematic only if multiple
cards of the same Device ID have different RFKILL state and the devices are not
distinguishable even with sub IDs.

FWIW, ACPI support added with commit c6a7c0b09d5f, suffers from the same
limitation.

- Mani

-- 
மணிவண்ணன் சதாசிவம்

Hi Mani,

On 22-Dec-25 12:45, Manivannan Sadhasivam wrote:
> On Mon, Dec 22, 2025 at 11:23:18AM +0100, Hans de Goede wrote:
>> +Cc Mani
>>
>> Hi,
>>
>> On 20-Dec-25 07:04, Bryan O'Donoghue wrote:
>>> On 20/12/2025 00:21, Jérôme de Bretagne via B4 Relay wrote:
>>>> From: Jérôme de Bretagne <jerome.debretagne@gmail.com>
>>>>
>>>> For some devices, Wi-Fi is entirely hard blocked by default making
>>>> the Wi-Fi radio unusable, except if rfkill is disabled as expected
>>>> on those models.
>>>>
>>>> Commit c6a7c0b09d5f ("wifi: ath12k: Add Support for enabling or
>>>> disabling specific features based on ACPI bitflag") added a way to
>>>> support features set via ACPI, including the DISABLE_RFKILL bit.
>>>>
>>>> Add a disable-rfkill property to expose the DISABLE_RFKILL bit
>>>> equivalent for devices described by a Devicetree instead of ACPI.
>>>>
>>>> Signed-off-by: Jérôme de Bretagne <jerome.debretagne@gmail.com>
>>>> ---
>>>>   Documentation/devicetree/bindings/net/wireless/ieee80211.yaml | 6 ++++++
>>>>   1 file changed, 6 insertions(+)
>>>>
>>>> diff --git a/Documentation/devicetree/bindings/net/wireless/ieee80211.yaml b/Documentation/devicetree/bindings/net/wireless/ieee80211.yaml
>>>> index d89f7a3f88a71d45d6f4ab2ae909eae09cbcaf9a..c10a4675640be947cd0b5eaec2c7ff367fd93945 100644
>>>> --- a/Documentation/devicetree/bindings/net/wireless/ieee80211.yaml
>>>> +++ b/Documentation/devicetree/bindings/net/wireless/ieee80211.yaml
>>>> @@ -29,6 +29,12 @@ properties:
>>>>         different 5 GHz subbands. Using them incorrectly could not work or
>>>>         decrease performance noticeably
>>>>
>>>> +  disable-rfkill:
>>>> +    type: boolean
>>>> +    description:
>>>> +      Disable rfkill for some devices on which Wi-Fi would be entirely hard
>>>> +      blocked by default otherwise
>>>> +
>>>>   additionalProperties: true
>>>>
>>>>   examples:
>>>>
>>>> -- 
>>>> 2.47.3
>>>>
>>>>
>>>>
>>>
>>> Is this really a hardware description though ?
>>
>> I would say yes it is. The wifi chip has an rfkill input pin and
>> things will be broken when that pin is hardwired to a fixed value
>> rather then being actually connected to a GPIO from say
>> the embedded controller.
>>
> 
> IIUC, even if the M.2 slot has the W_DISABLE1# signal routed from the host,
> the device won't make use of it as there is no physical connection. So you want
> the WLAN driver to change the state through SW?
> 
>> So I think that we would need here is not a disable-rfkill property
>> but some way to indicate in the DT-node that the rfkill input pin
>> is not connected and thus should be ignored.
>>
>> This (the rfkill input pin being not-connected) IMHO very much
>> is hw-description.
>>
> 
> Though we can argue this way, I would prefer to handle it in the driver. For
> example, with my M.2 series, we will end up describing the M.2 slot:
> 
> 	connector {
> 		compatible = "pcie-m2-e-connector";
> 		w-disable1-gpios = <&tlmm 117 GPIO_ACTIVE_LOW>;
> 		...
> 		ports {
> 			...
> 			endpoint@0 {
> 				reg = <0>;
> 				remote-endpoint = <&pcie4_port0_ep>;
> 			};
> 		};
> 	};
> 
> Then if we use a DT property to convey the RFKILL pin state of the device, we
> would need to describe the endpoint device in DT and hardcode the state:
> 
> 	&pcie4_port0 {
> 		...
> 		port {
> 			pcie4_port0_ep: endpoint {
> 				remote-endpoint = <&m2_e_pcie_ep>;
> 				disable-rfkill;
> 			};
> 		};
> 	};
> 
> So this will essentially make the M.2 device non-swappable unless you change the
> DT since you've how hardcoded the device property in the binding. This is
> something I try to avoid to make the M.2 slot really swappable.
> 
> For this reason, I would prefer to handle the RFKILL state in the WLAN driver
> using the device specific compatible. This will be problematic only if multiple
> cards of the same Device ID have different RFKILL state and the devices are not
> distinguishable even with sub IDs.

I think we're miscommunicating here. I'm not talking about the card having
a broken rfkill implementation, I'm talking about the M.2 slot on the mainboard
having e.g. W_DISABLE1# hardwired in such a way that cards would interpret it as
having to always disable their wifi radio which is very similar to what is
happening on the surface device. Except that on the Surface there is no M.2 slot,
the wifi is just soldered onto the mainboard I believe.

Based on experience I'm pretty sure we will encounter M.2 slots which such
a hardwired W_DISABLE1# signal sooner rather then later.

So my proposal is to come up with a generic way to solve the broken M.2 slot
case and then apply that to the Surface case.

Regards,

Hans

On Mon, Dec 22, 2025 at 01:41:48PM +0100, Hans de Goede wrote:
> Hi Mani,
> 
> On 22-Dec-25 12:45, Manivannan Sadhasivam wrote:
> > On Mon, Dec 22, 2025 at 11:23:18AM +0100, Hans de Goede wrote:
> >> +Cc Mani
> >>
> >> Hi,
> >>
> >> On 20-Dec-25 07:04, Bryan O'Donoghue wrote:
> >>> On 20/12/2025 00:21, Jérôme de Bretagne via B4 Relay wrote:
> >>>> From: Jérôme de Bretagne <jerome.debretagne@gmail.com>
> >>>>
> >>>> For some devices, Wi-Fi is entirely hard blocked by default making
> >>>> the Wi-Fi radio unusable, except if rfkill is disabled as expected
> >>>> on those models.
> >>>>
> >>>> Commit c6a7c0b09d5f ("wifi: ath12k: Add Support for enabling or
> >>>> disabling specific features based on ACPI bitflag") added a way to
> >>>> support features set via ACPI, including the DISABLE_RFKILL bit.
> >>>>
> >>>> Add a disable-rfkill property to expose the DISABLE_RFKILL bit
> >>>> equivalent for devices described by a Devicetree instead of ACPI.
> >>>>
> >>>> Signed-off-by: Jérôme de Bretagne <jerome.debretagne@gmail.com>
> >>>> ---
> >>>>   Documentation/devicetree/bindings/net/wireless/ieee80211.yaml | 6 ++++++
> >>>>   1 file changed, 6 insertions(+)
> >>>>
> >>>> diff --git a/Documentation/devicetree/bindings/net/wireless/ieee80211.yaml b/Documentation/devicetree/bindings/net/wireless/ieee80211.yaml
> >>>> index d89f7a3f88a71d45d6f4ab2ae909eae09cbcaf9a..c10a4675640be947cd0b5eaec2c7ff367fd93945 100644
> >>>> --- a/Documentation/devicetree/bindings/net/wireless/ieee80211.yaml
> >>>> +++ b/Documentation/devicetree/bindings/net/wireless/ieee80211.yaml
> >>>> @@ -29,6 +29,12 @@ properties:
> >>>>         different 5 GHz subbands. Using them incorrectly could not work or
> >>>>         decrease performance noticeably
> >>>>
> >>>> +  disable-rfkill:
> >>>> +    type: boolean
> >>>> +    description:
> >>>> +      Disable rfkill for some devices on which Wi-Fi would be entirely hard
> >>>> +      blocked by default otherwise
> >>>> +
> >>>>   additionalProperties: true
> >>>>
> >>>>   examples:
> >>>>
> >>>> -- 
> >>>> 2.47.3
> >>>>
> >>>>
> >>>>
> >>>
> >>> Is this really a hardware description though ?
> >>
> >> I would say yes it is. The wifi chip has an rfkill input pin and
> >> things will be broken when that pin is hardwired to a fixed value
> >> rather then being actually connected to a GPIO from say
> >> the embedded controller.
> >>
> > 
> > IIUC, even if the M.2 slot has the W_DISABLE1# signal routed from the host,
> > the device won't make use of it as there is no physical connection. So you want
> > the WLAN driver to change the state through SW?
> > 
> >> So I think that we would need here is not a disable-rfkill property
> >> but some way to indicate in the DT-node that the rfkill input pin
> >> is not connected and thus should be ignored.
> >>
> >> This (the rfkill input pin being not-connected) IMHO very much
> >> is hw-description.
> >>
> > 
> > Though we can argue this way, I would prefer to handle it in the driver. For
> > example, with my M.2 series, we will end up describing the M.2 slot:
> > 
> > 	connector {
> > 		compatible = "pcie-m2-e-connector";
> > 		w-disable1-gpios = <&tlmm 117 GPIO_ACTIVE_LOW>;
> > 		...
> > 		ports {
> > 			...
> > 			endpoint@0 {
> > 				reg = <0>;
> > 				remote-endpoint = <&pcie4_port0_ep>;
> > 			};
> > 		};
> > 	};
> > 
> > Then if we use a DT property to convey the RFKILL pin state of the device, we
> > would need to describe the endpoint device in DT and hardcode the state:
> > 
> > 	&pcie4_port0 {
> > 		...
> > 		port {
> > 			pcie4_port0_ep: endpoint {
> > 				remote-endpoint = <&m2_e_pcie_ep>;
> > 				disable-rfkill;
> > 			};
> > 		};
> > 	};
> > 
> > So this will essentially make the M.2 device non-swappable unless you change the
> > DT since you've how hardcoded the device property in the binding. This is
> > something I try to avoid to make the M.2 slot really swappable.
> > 
> > For this reason, I would prefer to handle the RFKILL state in the WLAN driver
> > using the device specific compatible. This will be problematic only if multiple
> > cards of the same Device ID have different RFKILL state and the devices are not
> > distinguishable even with sub IDs.
> 
> I think we're miscommunicating here. I'm not talking about the card having
> a broken rfkill implementation, I'm talking about the M.2 slot on the mainboard
> having e.g. W_DISABLE1# hardwired in such a way that cards would interpret it as
> having to always disable their wifi radio which is very similar to what is
> happening on the surface device. Except that on the Surface there is no M.2 slot,
> the wifi is just soldered onto the mainboard I believe.
> 

Ah, sorry for the confusion. I did misinterpret what you said.

> Based on experience I'm pretty sure we will encounter M.2 slots which such
> a hardwired W_DISABLE1# signal sooner rather then later.
> 

But it makes no sense IMO. Vendors will usually connect unimplemented W_DISABL1#
GPIOs to a pull-up resistor so that the radios are operational all the time. I
don't see how they would expect a WLAN or any device with a radio to be
connected to the slot if they hardwire the pin to low.

Are you sure that on the surface the pin is actually hardwired to low and not
connected to a GPIO that drivers the signal low?

It is just hard to believe that board designers can do a blunder like this.

- Mani

-- 
மணிவண்ணன் சதாசிவம்

Hi,

On 22-Dec-25 14:41, Manivannan Sadhasivam wrote:
> On Mon, Dec 22, 2025 at 01:41:48PM +0100, Hans de Goede wrote:
>> Hi Mani,
>>
>> On 22-Dec-25 12:45, Manivannan Sadhasivam wrote:
>>> On Mon, Dec 22, 2025 at 11:23:18AM +0100, Hans de Goede wrote:
>>>> +Cc Mani
>>>>
>>>> Hi,
>>>>
>>>> On 20-Dec-25 07:04, Bryan O'Donoghue wrote:
>>>>> On 20/12/2025 00:21, Jérôme de Bretagne via B4 Relay wrote:
>>>>>> From: Jérôme de Bretagne <jerome.debretagne@gmail.com>
>>>>>>
>>>>>> For some devices, Wi-Fi is entirely hard blocked by default making
>>>>>> the Wi-Fi radio unusable, except if rfkill is disabled as expected
>>>>>> on those models.
>>>>>>
>>>>>> Commit c6a7c0b09d5f ("wifi: ath12k: Add Support for enabling or
>>>>>> disabling specific features based on ACPI bitflag") added a way to
>>>>>> support features set via ACPI, including the DISABLE_RFKILL bit.
>>>>>>
>>>>>> Add a disable-rfkill property to expose the DISABLE_RFKILL bit
>>>>>> equivalent for devices described by a Devicetree instead of ACPI.
>>>>>>
>>>>>> Signed-off-by: Jérôme de Bretagne <jerome.debretagne@gmail.com>
>>>>>> ---
>>>>>>   Documentation/devicetree/bindings/net/wireless/ieee80211.yaml | 6 ++++++
>>>>>>   1 file changed, 6 insertions(+)
>>>>>>
>>>>>> diff --git a/Documentation/devicetree/bindings/net/wireless/ieee80211.yaml b/Documentation/devicetree/bindings/net/wireless/ieee80211.yaml
>>>>>> index d89f7a3f88a71d45d6f4ab2ae909eae09cbcaf9a..c10a4675640be947cd0b5eaec2c7ff367fd93945 100644
>>>>>> --- a/Documentation/devicetree/bindings/net/wireless/ieee80211.yaml
>>>>>> +++ b/Documentation/devicetree/bindings/net/wireless/ieee80211.yaml
>>>>>> @@ -29,6 +29,12 @@ properties:
>>>>>>         different 5 GHz subbands. Using them incorrectly could not work or
>>>>>>         decrease performance noticeably
>>>>>>
>>>>>> +  disable-rfkill:
>>>>>> +    type: boolean
>>>>>> +    description:
>>>>>> +      Disable rfkill for some devices on which Wi-Fi would be entirely hard
>>>>>> +      blocked by default otherwise
>>>>>> +
>>>>>>   additionalProperties: true
>>>>>>
>>>>>>   examples:
>>>>>>
>>>>>> -- 
>>>>>> 2.47.3
>>>>>>
>>>>>>
>>>>>>
>>>>>
>>>>> Is this really a hardware description though ?
>>>>
>>>> I would say yes it is. The wifi chip has an rfkill input pin and
>>>> things will be broken when that pin is hardwired to a fixed value
>>>> rather then being actually connected to a GPIO from say
>>>> the embedded controller.
>>>>
>>>
>>> IIUC, even if the M.2 slot has the W_DISABLE1# signal routed from the host,
>>> the device won't make use of it as there is no physical connection. So you want
>>> the WLAN driver to change the state through SW?
>>>
>>>> So I think that we would need here is not a disable-rfkill property
>>>> but some way to indicate in the DT-node that the rfkill input pin
>>>> is not connected and thus should be ignored.
>>>>
>>>> This (the rfkill input pin being not-connected) IMHO very much
>>>> is hw-description.
>>>>
>>>
>>> Though we can argue this way, I would prefer to handle it in the driver. For
>>> example, with my M.2 series, we will end up describing the M.2 slot:
>>>
>>> 	connector {
>>> 		compatible = "pcie-m2-e-connector";
>>> 		w-disable1-gpios = <&tlmm 117 GPIO_ACTIVE_LOW>;
>>> 		...
>>> 		ports {
>>> 			...
>>> 			endpoint@0 {
>>> 				reg = <0>;
>>> 				remote-endpoint = <&pcie4_port0_ep>;
>>> 			};
>>> 		};
>>> 	};
>>>
>>> Then if we use a DT property to convey the RFKILL pin state of the device, we
>>> would need to describe the endpoint device in DT and hardcode the state:
>>>
>>> 	&pcie4_port0 {
>>> 		...
>>> 		port {
>>> 			pcie4_port0_ep: endpoint {
>>> 				remote-endpoint = <&m2_e_pcie_ep>;
>>> 				disable-rfkill;
>>> 			};
>>> 		};
>>> 	};
>>>
>>> So this will essentially make the M.2 device non-swappable unless you change the
>>> DT since you've how hardcoded the device property in the binding. This is
>>> something I try to avoid to make the M.2 slot really swappable.
>>>
>>> For this reason, I would prefer to handle the RFKILL state in the WLAN driver
>>> using the device specific compatible. This will be problematic only if multiple
>>> cards of the same Device ID have different RFKILL state and the devices are not
>>> distinguishable even with sub IDs.
>>
>> I think we're miscommunicating here. I'm not talking about the card having
>> a broken rfkill implementation, I'm talking about the M.2 slot on the mainboard
>> having e.g. W_DISABLE1# hardwired in such a way that cards would interpret it as
>> having to always disable their wifi radio which is very similar to what is
>> happening on the surface device. Except that on the Surface there is no M.2 slot,
>> the wifi is just soldered onto the mainboard I believe.
>>
> 
> Ah, sorry for the confusion. I did misinterpret what you said.
> 
>> Based on experience I'm pretty sure we will encounter M.2 slots which such
>> a hardwired W_DISABLE1# signal sooner rather then later.
>>
> 
> But it makes no sense IMO. Vendors will usually connect unimplemented W_DISABL1#
> GPIOs to a pull-up resistor so that the radios are operational all the time. I
> don't see how they would expect a WLAN or any device with a radio to be
> connected to the slot if they hardwire the pin to low.

Pins which are considered "unused" are also often hardwired
to ground. If the m.2 slot is tested with a wifi-module where
the W_DISABLE1# signal is not used on the wifi-module I can
easily see this happen. I've seen a lot crazier / buggy stuff
happen.

> Are you sure that on the surface the pin is actually hardwired to low and not
> connected to a GPIO that drivers the signal low?

I don't know what is the exact problem on the Surface. I just
expect to see this more often, we've certainly seen lots of
issues like this on x86 laptops. Things end up looking like
the hard rfkill is activated all the time (and we often don't know
if this is a fw issue, or an actually hardwired problem).

Just an example from the top of my head the Broadcom windows
drivers use different BT fw files for the same wifi/bt combo
chip depending on the vend:prod id pair. One of the things which
is different is that some fw files invert the BT rfkill signal
because it is wired wrong (or there is an EC fw bug) and this
is then worked around in the bt fw.

As we see a growing proliferation of arm64 laptops I fully
expect all the fun from having a gazillion different designs
with a time to market rush behind them result in similar issues
on arm64.

Note I'm not saying we must tackle this today, we can wait
till we see the first case in the real world I guess.

I just thought that based on my experience this is more or
less bound to happen, we could comeup with a solution for
this now and then this solution could also nicely serve
the Surface case which started this thread.

But we can also delay tackling this and come up with some
bespoke solution for the Surface case, like as suggested
maybe a special compatible string ?

> It is just hard to believe that board designers can do a blunder like this.

Heh, you won't believe all the crap happening on cheap
x86 devices.

Regards,

Hans

On Mon, Dec 22, 2025 at 03:22:55PM +0100, Hans de Goede wrote:
> Hi,
> 
> On 22-Dec-25 14:41, Manivannan Sadhasivam wrote:
> > On Mon, Dec 22, 2025 at 01:41:48PM +0100, Hans de Goede wrote:
> >> Hi Mani,
> >>
> >> On 22-Dec-25 12:45, Manivannan Sadhasivam wrote:
> >>> On Mon, Dec 22, 2025 at 11:23:18AM +0100, Hans de Goede wrote:
> >>>> +Cc Mani
> >>>>
> >>>> Hi,
> >>>>
> >>>> On 20-Dec-25 07:04, Bryan O'Donoghue wrote:
> >>>>> On 20/12/2025 00:21, Jérôme de Bretagne via B4 Relay wrote:
> >>>>>> From: Jérôme de Bretagne <jerome.debretagne@gmail.com>
> >>>>>>
> >>>>>> For some devices, Wi-Fi is entirely hard blocked by default making
> >>>>>> the Wi-Fi radio unusable, except if rfkill is disabled as expected
> >>>>>> on those models.
> >>>>>>
> >>>>>> Commit c6a7c0b09d5f ("wifi: ath12k: Add Support for enabling or
> >>>>>> disabling specific features based on ACPI bitflag") added a way to
> >>>>>> support features set via ACPI, including the DISABLE_RFKILL bit.
> >>>>>>
> >>>>>> Add a disable-rfkill property to expose the DISABLE_RFKILL bit
> >>>>>> equivalent for devices described by a Devicetree instead of ACPI.
> >>>>>>
> >>>>>> Signed-off-by: Jérôme de Bretagne <jerome.debretagne@gmail.com>
> >>>>>> ---
> >>>>>>   Documentation/devicetree/bindings/net/wireless/ieee80211.yaml | 6 ++++++
> >>>>>>   1 file changed, 6 insertions(+)
> >>>>>>
> >>>>>> diff --git a/Documentation/devicetree/bindings/net/wireless/ieee80211.yaml b/Documentation/devicetree/bindings/net/wireless/ieee80211.yaml
> >>>>>> index d89f7a3f88a71d45d6f4ab2ae909eae09cbcaf9a..c10a4675640be947cd0b5eaec2c7ff367fd93945 100644
> >>>>>> --- a/Documentation/devicetree/bindings/net/wireless/ieee80211.yaml
> >>>>>> +++ b/Documentation/devicetree/bindings/net/wireless/ieee80211.yaml
> >>>>>> @@ -29,6 +29,12 @@ properties:
> >>>>>>         different 5 GHz subbands. Using them incorrectly could not work or
> >>>>>>         decrease performance noticeably
> >>>>>>
> >>>>>> +  disable-rfkill:
> >>>>>> +    type: boolean
> >>>>>> +    description:
> >>>>>> +      Disable rfkill for some devices on which Wi-Fi would be entirely hard
> >>>>>> +      blocked by default otherwise
> >>>>>> +
> >>>>>>   additionalProperties: true
> >>>>>>
> >>>>>>   examples:
> >>>>>>
> >>>>>> -- 
> >>>>>> 2.47.3
> >>>>>>
> >>>>>>
> >>>>>>
> >>>>>
> >>>>> Is this really a hardware description though ?
> >>>>
> >>>> I would say yes it is. The wifi chip has an rfkill input pin and
> >>>> things will be broken when that pin is hardwired to a fixed value
> >>>> rather then being actually connected to a GPIO from say
> >>>> the embedded controller.
> >>>>
> >>>
> >>> IIUC, even if the M.2 slot has the W_DISABLE1# signal routed from the host,
> >>> the device won't make use of it as there is no physical connection. So you want
> >>> the WLAN driver to change the state through SW?
> >>>
> >>>> So I think that we would need here is not a disable-rfkill property
> >>>> but some way to indicate in the DT-node that the rfkill input pin
> >>>> is not connected and thus should be ignored.
> >>>>
> >>>> This (the rfkill input pin being not-connected) IMHO very much
> >>>> is hw-description.
> >>>>
> >>>
> >>> Though we can argue this way, I would prefer to handle it in the driver. For
> >>> example, with my M.2 series, we will end up describing the M.2 slot:
> >>>
> >>> 	connector {
> >>> 		compatible = "pcie-m2-e-connector";
> >>> 		w-disable1-gpios = <&tlmm 117 GPIO_ACTIVE_LOW>;
> >>> 		...
> >>> 		ports {
> >>> 			...
> >>> 			endpoint@0 {
> >>> 				reg = <0>;
> >>> 				remote-endpoint = <&pcie4_port0_ep>;
> >>> 			};
> >>> 		};
> >>> 	};
> >>>
> >>> Then if we use a DT property to convey the RFKILL pin state of the device, we
> >>> would need to describe the endpoint device in DT and hardcode the state:
> >>>
> >>> 	&pcie4_port0 {
> >>> 		...
> >>> 		port {
> >>> 			pcie4_port0_ep: endpoint {
> >>> 				remote-endpoint = <&m2_e_pcie_ep>;
> >>> 				disable-rfkill;
> >>> 			};
> >>> 		};
> >>> 	};
> >>>
> >>> So this will essentially make the M.2 device non-swappable unless you change the
> >>> DT since you've how hardcoded the device property in the binding. This is
> >>> something I try to avoid to make the M.2 slot really swappable.
> >>>
> >>> For this reason, I would prefer to handle the RFKILL state in the WLAN driver
> >>> using the device specific compatible. This will be problematic only if multiple
> >>> cards of the same Device ID have different RFKILL state and the devices are not
> >>> distinguishable even with sub IDs.
> >>
> >> I think we're miscommunicating here. I'm not talking about the card having
> >> a broken rfkill implementation, I'm talking about the M.2 slot on the mainboard
> >> having e.g. W_DISABLE1# hardwired in such a way that cards would interpret it as
> >> having to always disable their wifi radio which is very similar to what is
> >> happening on the surface device. Except that on the Surface there is no M.2 slot,
> >> the wifi is just soldered onto the mainboard I believe.
> >>
> > 
> > Ah, sorry for the confusion. I did misinterpret what you said.
> > 
> >> Based on experience I'm pretty sure we will encounter M.2 slots which such
> >> a hardwired W_DISABLE1# signal sooner rather then later.
> >>
> > 
> > But it makes no sense IMO. Vendors will usually connect unimplemented W_DISABL1#
> > GPIOs to a pull-up resistor so that the radios are operational all the time. I
> > don't see how they would expect a WLAN or any device with a radio to be
> > connected to the slot if they hardwire the pin to low.
> 
> Pins which are considered "unused" are also often hardwired
> to ground. If the m.2 slot is tested with a wifi-module where
> the W_DISABLE1# signal is not used on the wifi-module I can
> easily see this happen. I've seen a lot crazier / buggy stuff
> happen.
> 
> > Are you sure that on the surface the pin is actually hardwired to low and not
> > connected to a GPIO that drivers the signal low?
> 
> I don't know what is the exact problem on the Surface. I just
> expect to see this more often, we've certainly seen lots of
> issues like this on x86 laptops. Things end up looking like
> the hard rfkill is activated all the time (and we often don't know
> if this is a fw issue, or an actually hardwired problem).
> 
> Just an example from the top of my head the Broadcom windows
> drivers use different BT fw files for the same wifi/bt combo
> chip depending on the vend:prod id pair. One of the things which
> is different is that some fw files invert the BT rfkill signal
> because it is wired wrong (or there is an EC fw bug) and this
> is then worked around in the bt fw.
> 
> As we see a growing proliferation of arm64 laptops I fully
> expect all the fun from having a gazillion different designs
> with a time to market rush behind them result in similar issues
> on arm64.
> 
> Note I'm not saying we must tackle this today, we can wait
> till we see the first case in the real world I guess.
> 
> I just thought that based on my experience this is more or
> less bound to happen, we could comeup with a solution for
> this now and then this solution could also nicely serve
> the Surface case which started this thread.
> 
> But we can also delay tackling this and come up with some
> bespoke solution for the Surface case, like as suggested
> maybe a special compatible string ?
> 

Hmm. If we want to go with the DT property, I'd use something like
'broken-rfkill' or 'broken-w-disable1' or similar in the connector node, not in
the device node. This will convey the fact that the RFKILL switch is broken
in the connector or the hardware topology is not known.

But we do not have the connector binding merged yet. Until then, I'd suggest to
keep the hack in the WLAN driver by using the platform compatible and Device ID:

diff --git a/drivers/net/wireless/ath/ath12k/core.c b/drivers/net/wireless/ath/ath12k/core.c
index cc352eef1939..481778eb2c95 100644
--- a/drivers/net/wireless/ath/ath12k/core.c
+++ b/drivers/net/wireless/ath/ath12k/core.c
@@ -77,6 +77,16 @@ static int ath12k_core_rfkill_config(struct ath12k_base *ab)
        if (ath12k_acpi_get_disable_rfkill(ab))
                return 0;
 
+       /*
+        * TODO: On Microsoft Surface Pro 11, OS is not able to control the
+        * RFKILL switch. So keep the RFKILL disabled until the OS learns about
+        * it. Ideally, this info should come from the connector node of the
+        * board DT file. But since the connector DT node is not available,
+        * implement the hack in the driver.
+        */
+       if (of_machine_is_compatible("microsoft,denali") && (ab->id.device == 0x1107))
+               return 0;
+
        for (i = 0; i < ab->num_radios; i++) {
                ar = ab->pdevs[i].ar;


Once the connector binding gets merged, hopefully we can add the DT property and
use it in the driver.

> > It is just hard to believe that board designers can do a blunder like this.
> 
> Heh, you won't believe all the crap happening on cheap
> x86 devices.
> 

Coming from the DT world, I thought the ACPI world is superior, but this reminds
of the fact "No world is superior to another" :)

- Mani

-- 
மணிவண்ணன் சதாசிவம்

Le mar. 23 déc. 2025 à 07:31, Manivannan Sadhasivam <mani@kernel.org> a écrit :
>
> On Mon, Dec 22, 2025 at 03:22:55PM +0100, Hans de Goede wrote:
> > Hi,
> >
> > On 22-Dec-25 14:41, Manivannan Sadhasivam wrote:
> > > On Mon, Dec 22, 2025 at 01:41:48PM +0100, Hans de Goede wrote:
> > >> Hi Mani,
> > >>
> > >> On 22-Dec-25 12:45, Manivannan Sadhasivam wrote:
> > >>> On Mon, Dec 22, 2025 at 11:23:18AM +0100, Hans de Goede wrote:
> > >>>> +Cc Mani
> > >>>>
> > >>>> Hi,
> > >>>>
> > >>>> On 20-Dec-25 07:04, Bryan O'Donoghue wrote:
> > >>>>> On 20/12/2025 00:21, Jérôme de Bretagne via B4 Relay wrote:
> > >>>>>> From: Jérôme de Bretagne <jerome.debretagne@gmail.com>
> > >>>>>>
> > >>>>>> For some devices, Wi-Fi is entirely hard blocked by default making
> > >>>>>> the Wi-Fi radio unusable, except if rfkill is disabled as expected
> > >>>>>> on those models.
> > >>>>>>
> > >>>>>> Commit c6a7c0b09d5f ("wifi: ath12k: Add Support for enabling or
> > >>>>>> disabling specific features based on ACPI bitflag") added a way to
> > >>>>>> support features set via ACPI, including the DISABLE_RFKILL bit.
> > >>>>>>
> > >>>>>> Add a disable-rfkill property to expose the DISABLE_RFKILL bit
> > >>>>>> equivalent for devices described by a Devicetree instead of ACPI.
> > >>>>>>
> > >>>>>> Signed-off-by: Jérôme de Bretagne <jerome.debretagne@gmail.com>
> > >>>>>> ---
> > >>>>>>   Documentation/devicetree/bindings/net/wireless/ieee80211.yaml | 6 ++++++
> > >>>>>>   1 file changed, 6 insertions(+)
> > >>>>>>
> > >>>>>> diff --git a/Documentation/devicetree/bindings/net/wireless/ieee80211.yaml b/Documentation/devicetree/bindings/net/wireless/ieee80211.yaml
> > >>>>>> index d89f7a3f88a71d45d6f4ab2ae909eae09cbcaf9a..c10a4675640be947cd0b5eaec2c7ff367fd93945 100644
> > >>>>>> --- a/Documentation/devicetree/bindings/net/wireless/ieee80211.yaml
> > >>>>>> +++ b/Documentation/devicetree/bindings/net/wireless/ieee80211.yaml
> > >>>>>> @@ -29,6 +29,12 @@ properties:
> > >>>>>>         different 5 GHz subbands. Using them incorrectly could not work or
> > >>>>>>         decrease performance noticeably
> > >>>>>>
> > >>>>>> +  disable-rfkill:
> > >>>>>> +    type: boolean
> > >>>>>> +    description:
> > >>>>>> +      Disable rfkill for some devices on which Wi-Fi would be entirely hard
> > >>>>>> +      blocked by default otherwise
> > >>>>>> +
> > >>>>>>   additionalProperties: true
> > >>>>>>
> > >>>>>>   examples:
> > >>>>>>
> > >>>>>> --
> > >>>>>> 2.47.3
> > >>>>>>
> > >>>>>>
> > >>>>>>
> > >>>>>
> > >>>>> Is this really a hardware description though ?
> > >>>>
> > >>>> I would say yes it is. The wifi chip has an rfkill input pin and
> > >>>> things will be broken when that pin is hardwired to a fixed value
> > >>>> rather then being actually connected to a GPIO from say
> > >>>> the embedded controller.
> > >>>>
> > >>>
> > >>> IIUC, even if the M.2 slot has the W_DISABLE1# signal routed from the host,
> > >>> the device won't make use of it as there is no physical connection. So you want
> > >>> the WLAN driver to change the state through SW?
> > >>>
> > >>>> So I think that we would need here is not a disable-rfkill property
> > >>>> but some way to indicate in the DT-node that the rfkill input pin
> > >>>> is not connected and thus should be ignored.
> > >>>>
> > >>>> This (the rfkill input pin being not-connected) IMHO very much
> > >>>> is hw-description.
> > >>>>
> > >>>
> > >>> Though we can argue this way, I would prefer to handle it in the driver. For
> > >>> example, with my M.2 series, we will end up describing the M.2 slot:
> > >>>
> > >>>   connector {
> > >>>           compatible = "pcie-m2-e-connector";
> > >>>           w-disable1-gpios = <&tlmm 117 GPIO_ACTIVE_LOW>;
> > >>>           ...
> > >>>           ports {
> > >>>                   ...
> > >>>                   endpoint@0 {
> > >>>                           reg = <0>;
> > >>>                           remote-endpoint = <&pcie4_port0_ep>;
> > >>>                   };
> > >>>           };
> > >>>   };
> > >>>
> > >>> Then if we use a DT property to convey the RFKILL pin state of the device, we
> > >>> would need to describe the endpoint device in DT and hardcode the state:
> > >>>
> > >>>   &pcie4_port0 {
> > >>>           ...
> > >>>           port {
> > >>>                   pcie4_port0_ep: endpoint {
> > >>>                           remote-endpoint = <&m2_e_pcie_ep>;
> > >>>                           disable-rfkill;
> > >>>                   };
> > >>>           };
> > >>>   };
> > >>>
> > >>> So this will essentially make the M.2 device non-swappable unless you change the
> > >>> DT since you've how hardcoded the device property in the binding. This is
> > >>> something I try to avoid to make the M.2 slot really swappable.
> > >>>
> > >>> For this reason, I would prefer to handle the RFKILL state in the WLAN driver
> > >>> using the device specific compatible. This will be problematic only if multiple
> > >>> cards of the same Device ID have different RFKILL state and the devices are not
> > >>> distinguishable even with sub IDs.
> > >>
> > >> I think we're miscommunicating here. I'm not talking about the card having
> > >> a broken rfkill implementation, I'm talking about the M.2 slot on the mainboard
> > >> having e.g. W_DISABLE1# hardwired in such a way that cards would interpret it as
> > >> having to always disable their wifi radio which is very similar to what is
> > >> happening on the surface device. Except that on the Surface there is no M.2 slot,
> > >> the wifi is just soldered onto the mainboard I believe.
> > >>
> > >
> > > Ah, sorry for the confusion. I did misinterpret what you said.
> > >
> > >> Based on experience I'm pretty sure we will encounter M.2 slots which such
> > >> a hardwired W_DISABLE1# signal sooner rather then later.
> > >>
> > >
> > > But it makes no sense IMO. Vendors will usually connect unimplemented W_DISABL1#
> > > GPIOs to a pull-up resistor so that the radios are operational all the time. I
> > > don't see how they would expect a WLAN or any device with a radio to be
> > > connected to the slot if they hardwire the pin to low.
> >
> > Pins which are considered "unused" are also often hardwired
> > to ground. If the m.2 slot is tested with a wifi-module where
> > the W_DISABLE1# signal is not used on the wifi-module I can
> > easily see this happen. I've seen a lot crazier / buggy stuff
> > happen.
> >
> > > Are you sure that on the surface the pin is actually hardwired to low and not
> > > connected to a GPIO that drivers the signal low?
> >
> > I don't know what is the exact problem on the Surface. I just
> > expect to see this more often, we've certainly seen lots of
> > issues like this on x86 laptops. Things end up looking like
> > the hard rfkill is activated all the time (and we often don't know
> > if this is a fw issue, or an actually hardwired problem).
> >
> > Just an example from the top of my head the Broadcom windows
> > drivers use different BT fw files for the same wifi/bt combo
> > chip depending on the vend:prod id pair. One of the things which
> > is different is that some fw files invert the BT rfkill signal
> > because it is wired wrong (or there is an EC fw bug) and this
> > is then worked around in the bt fw.
> >
> > As we see a growing proliferation of arm64 laptops I fully
> > expect all the fun from having a gazillion different designs
> > with a time to market rush behind them result in similar issues
> > on arm64.
> >
> > Note I'm not saying we must tackle this today, we can wait
> > till we see the first case in the real world I guess.
> >
> > I just thought that based on my experience this is more or
> > less bound to happen, we could comeup with a solution for
> > this now and then this solution could also nicely serve
> > the Surface case which started this thread.
> >
> > But we can also delay tackling this and come up with some
> > bespoke solution for the Surface case, like as suggested
> > maybe a special compatible string ?
> >
>
> Hmm. If we want to go with the DT property, I'd use something like
> 'broken-rfkill' or 'broken-w-disable1' or similar in the connector node, not in
> the device node. This will convey the fact that the RFKILL switch is broken
> in the connector or the hardware topology is not known.
>
> But we do not have the connector binding merged yet. Until then, I'd suggest to
> keep the hack in the WLAN driver by using the platform compatible and Device ID:

If that suggestion is accepted, please keep in mind that this
issue is also impacting the Surface Laptop 7 family, with the
"microsoft,romulus13" and "microsoft,romulus15" compatibles.

The SL7 rfkill issue is mentioned here:
   https://github.com/bryce-hoehn/linux-surface-laptop-7
or here:
   https://github.com/linux-surface/linux-surface/issues/1590
for reference.

> diff --git a/drivers/net/wireless/ath/ath12k/core.c b/drivers/net/wireless/ath/ath12k/core.c
> index cc352eef1939..481778eb2c95 100644
> --- a/drivers/net/wireless/ath/ath12k/core.c
> +++ b/drivers/net/wireless/ath/ath12k/core.c
> @@ -77,6 +77,16 @@ static int ath12k_core_rfkill_config(struct ath12k_base *ab)
>         if (ath12k_acpi_get_disable_rfkill(ab))
>                 return 0;
>
> +       /*
> +        * TODO: On Microsoft Surface Pro 11, OS is not able to control the
> +        * RFKILL switch. So keep the RFKILL disabled until the OS learns about
> +        * it. Ideally, this info should come from the connector node of the
> +        * board DT file. But since the connector DT node is not available,
> +        * implement the hack in the driver.
> +        */
> +       if (of_machine_is_compatible("microsoft,denali") && (ab->id.device == 0x1107))
> +               return 0;
> +
>         for (i = 0; i < ab->num_radios; i++) {
>                 ar = ab->pdevs[i].ar;
>
>
> Once the connector binding gets merged, hopefully we can add the DT property and
> use it in the driver.
>
> > > It is just hard to believe that board designers can do a blunder like this.
> >
> > Heh, you won't believe all the crap happening on cheap
> > x86 devices.
> >
>
> Coming from the DT world, I thought the ACPI world is superior, but this reminds
> of the fact "No world is superior to another" :)
>
> - Mani
>
> --
> மணிவண்ணன் சதாசிவம்

Thank you,
Jérôme