On Thu, Jun 6, 2024 at 6:16 PM Kalle Valo <[email protected]> wrote:
>
> Bartosz Golaszewski <[email protected]> writes:
>
> > On Thu, Jun 6, 2024 at 4:02 PM Kalle Valo <[email protected]> wrote:
> >
> >>
> >> Bartosz Golaszewski <[email protected]> writes:
> >>
> >> > On Thu, Jun 6, 2024 at 3:30 PM Kalle Valo <[email protected]> wrote:
> >> >
> >> >>
> >> >> Bartosz Golaszewski <[email protected]> writes:
> >> >>
> >> >> > From: Bartosz Golaszewski <[email protected]>
> >> >> >
> >> >> > Add a PCI compatible for the ATH11K module on QCA6390 and describe the
> >> >> > power inputs from the PMU that it consumes.
> >> >> >
> >> >> > Reviewed-by: Krzysztof Kozlowski <[email protected]>
> >> >> > Signed-off-by: Bartosz Golaszewski <[email protected]>
> >> >>
> >> >> [...]
> >> >>
> >> >> > +allOf:
> >> >> > + - if:
> >> >> > + properties:
> >> >> > + compatible:
> >> >> > + contains:
> >> >> > + const: pci17cb,1101
> >> >> > + then:
> >> >> > + required:
> >> >> > + - vddrfacmn-supply
> >> >> > + - vddaon-supply
> >> >> > + - vddwlcx-supply
> >> >> > + - vddwlmx-supply
> >> >> > + - vddrfa0p8-supply
> >> >> > + - vddrfa1p2-supply
> >> >> > + - vddrfa1p7-supply
> >> >> > + - vddpcie0p9-supply
> >> >> > + - vddpcie1p8-supply
> >> >>
> >> >> Not sure if we discussed this before, but based on this I understand
> >> >> that there can't be an DT entry for device pci17cb,1101 without all the
> >> >> supply properties? But there are QCA6390 devices with PCI id 17cb:1101
> >> >> which do not need these supplies and already work. For example, my Dell
> >> >> XPS 13 x86 laptop is one. Or anyone who manually installs QCA6390 board
> >> >> to their PCI slot and some of them might want to use DT, for example
> >> >> setting qcom,ath11k-calibration-variant.
> >> >>
> >> >> This is not a blocker for me, just making sure that we are not breaking
> >> >> any existing setups.
> >> >>
> >> >
> >> > If they are already powered up without the need for the PCI pwrctl
> >> > driver to do it, then they will work alright. Bindings don't affect
> >> > functionality.
> >>
> >> Sure, I'm not worried about functionality. I'm worried that if I
> >> there's, for example, an ARM based setup which uses DT and wants to use
> >> a similar QCA6390 board that I have, and set
> >> qcom,ath11k-calibration-variant in DT. In other words, I'm worried if
> >> you are looking at this only for Snapdragon family of boards?
> >>
> >
> > No, what I'm looking at is the entire QCA6390 package. That means WLAN
> > *and* Bluetooth *and* the PMU that manages power.
>
> I think we are just looking at this from different point of views. You
> are looking at a datasheet (most likely for a Snapdragon based system)
> and I'm looking what actual devices there are out in the field.
>
> > If you're using the QCA6390 on a device-tree system then you should
> > probably model at least the WLAN node and the PMU and the problem with
> > supplies is fixed.
>
> But why? If there are boards out there who don't need any of this why
> would they still need to model all this in DT?
>
Because this is what is there? The goal of the device tree is to
describe the hardware. The fact we didn't describe it before doesn't
make it correct.
> Based on the discussions I have heard only Snapdragon systems who
> require all this configuration you describe. Of course there can be
> other systems but I have not heard about those.
>
DT is not configuration, it is description of actual hardware. It
doesn't matter if Snapdragon systems are the only ones that actually
*require* this description to make WLAN/BT functional upstream. The
chipset would be the same on any PCIe board, it's just that the host
systems wouldn't need to take care with its power sequence. But for a
dynamic board like this, you don't need DT.
> > But if you don't have the supplies, that's alright for downstream.
>
> What do you mean downstream in this context?
>
I mean: if you wanted to upstream the DT sources, then they should
include the supplies AND the PMU node. But if you just want to make
the WLAN run on some vendor kernel then you don't need to think about
it, it will work.
> >> Again, I don't see this as a blocker. I just want to understand how this
> >> should work for all types of devices there are out there.
> >>
> >> > But if you have a QCA6390 then you have its PMU too and the bindings
> >> > model the real-world hardware.
> >> >
> >> > IOW: your laptop should be alright but the supplies are really there
> >> > which warrants adding them to the bindings.
> >>
> >> Sorry, not following here. Can you clarify your comment "the supplies
> >> are really there"? You mean inside the PCI board? But that's not visible
> >> to the kernel in anyway, the PCI board just works after I plug it in.
> >> It's like a regular PCI device. So I don't understand why that should be
> >> visible in DT, but I can very well be missing something.
> >>
> >
> > I think you're thinking about some kind of detachable PCIe board with
> > this chipset on it.
>
> Exactly, a lot of WLAN boards are like this.
>
> > I refer to the QCA6390 chipset itself which is also more than just
> > PCI. The Bluetooth interface doesn't use PCI at all. On the boards I'm
> > working on, the chipset is just soldered to the main board.
>
> And I guess you are looking at Snapdragon boards only?
>
But what is your point?
> > If your detachable board "just works" then it must be wired in a way
> > that enables WLAN the moment it's plugged in but this doesn't happen
> > over PCI. The chipset has a power input and GPIOs to enable each
> > module.
>
> I don't know how the boards are implemented but it could be so. But from
> host system point of view it's just a regular PCI device.
>
And you don't need DT anyway for this type of devices.
> > Also: I doubt you need DT for your detachable board?
>
> Sure, I don't need DT but that's not my point. My point is why require
> these supplies for _all_ devices having PCI id 17cb:1101 (ie. QCA6390)
> then clearly there are such devices which don't need it? To me that's
> bad design and, if I'm understanding correctly, prevents use of
> qcom,ath11k-calibration-variant property. To me having the supplies
> optional in DT is more approriate.
>
We require them because *they are physically there*.
Bart
> --
> https://patchwork.kernel.org/project/linux-wireless/list/
>
> https://wireless.wiki.kernel.org/en/developers/documentation/submittingpatches
On 06/06/2024 20:08, Bartosz Golaszewski wrote:
> On Thu, Jun 6, 2024 at 6:16 PM Kalle Valo <[email protected]> wrote:
>>
>> Bartosz Golaszewski <[email protected]> writes:
>>
>>> On Thu, Jun 6, 2024 at 4:02 PM Kalle Valo <[email protected]> wrote:
>>>
>>>>
>>>> Bartosz Golaszewski <[email protected]> writes:
>>>>
>>>>> On Thu, Jun 6, 2024 at 3:30 PM Kalle Valo <[email protected]> wrote:
>>>>>
>>>>>>
>>>>>> Bartosz Golaszewski <[email protected]> writes:
>>>>>>
>>>>>>> From: Bartosz Golaszewski <[email protected]>
>>>>>>>
>>>>>>> Add a PCI compatible for the ATH11K module on QCA6390 and describe the
>>>>>>> power inputs from the PMU that it consumes.
>>>>>>>
>>>>>>> Reviewed-by: Krzysztof Kozlowski <[email protected]>
>>>>>>> Signed-off-by: Bartosz Golaszewski <[email protected]>
>>>>>>
>>>>>> [...]
>>>>>>
>>>>>>> +allOf:
>>>>>>> + - if:
>>>>>>> + properties:
>>>>>>> + compatible:
>>>>>>> + contains:
>>>>>>> + const: pci17cb,1101
>>>>>>> + then:
>>>>>>> + required:
>>>>>>> + - vddrfacmn-supply
>>>>>>> + - vddaon-supply
>>>>>>> + - vddwlcx-supply
>>>>>>> + - vddwlmx-supply
>>>>>>> + - vddrfa0p8-supply
>>>>>>> + - vddrfa1p2-supply
>>>>>>> + - vddrfa1p7-supply
>>>>>>> + - vddpcie0p9-supply
>>>>>>> + - vddpcie1p8-supply
>>>>>>
>>>>>> Not sure if we discussed this before, but based on this I understand
>>>>>> that there can't be an DT entry for device pci17cb,1101 without all the
>>>>>> supply properties? But there are QCA6390 devices with PCI id 17cb:1101
>>>>>> which do not need these supplies and already work. For example, my Dell
>>>>>> XPS 13 x86 laptop is one. Or anyone who manually installs QCA6390 board
>>>>>> to their PCI slot and some of them might want to use DT, for example
>>>>>> setting qcom,ath11k-calibration-variant.
>>>>>>
>>>>>> This is not a blocker for me, just making sure that we are not breaking
>>>>>> any existing setups.
>>>>>>
>>>>>
>>>>> If they are already powered up without the need for the PCI pwrctl
>>>>> driver to do it, then they will work alright. Bindings don't affect
>>>>> functionality.
>>>>
>>>> Sure, I'm not worried about functionality. I'm worried that if I
>>>> there's, for example, an ARM based setup which uses DT and wants to use
>>>> a similar QCA6390 board that I have, and set
>>>> qcom,ath11k-calibration-variant in DT. In other words, I'm worried if
>>>> you are looking at this only for Snapdragon family of boards?
>>>>
>>>
>>> No, what I'm looking at is the entire QCA6390 package. That means WLAN
>>> *and* Bluetooth *and* the PMU that manages power.
>>
>> I think we are just looking at this from different point of views. You
>> are looking at a datasheet (most likely for a Snapdragon based system)
>> and I'm looking what actual devices there are out in the field.
>>
>>> If you're using the QCA6390 on a device-tree system then you should
>>> probably model at least the WLAN node and the PMU and the problem with
>>> supplies is fixed.
>>
>> But why? If there are boards out there who don't need any of this why
>> would they still need to model all this in DT?
>>
>
> Because this is what is there? The goal of the device tree is to
> describe the hardware. The fact we didn't describe it before doesn't
> make it correct.
Correct.
Kalle,
All of the devices out there need these supplies, but they are sometimes
provided by generic PCI supply and on-board regulators. Basically your
PCI adapter is not the same as QCA6390 chip on Snapdragon board.
>
>> Based on the discussions I have heard only Snapdragon systems who
>> require all this configuration you describe. Of course there can be
>> other systems but I have not heard about those.
>>
>
> DT is not configuration, it is description of actual hardware. It
> doesn't matter if Snapdragon systems are the only ones that actually
> *require* this description to make WLAN/BT functional upstream. The
> chipset would be the same on any PCIe board, it's just that the host
> systems wouldn't need to take care with its power sequence. But for a
> dynamic board like this, you don't need DT.
>
Correct.
...
>
>>> If your detachable board "just works" then it must be wired in a way
>>> that enables WLAN the moment it's plugged in but this doesn't happen
>>> over PCI. The chipset has a power input and GPIOs to enable each
>>> module.
>>
>> I don't know how the boards are implemented but it could be so. But from
>> host system point of view it's just a regular PCI device.
>>
>
> And you don't need DT anyway for this type of devices.
Detechable board, like PCI adapter, derives these supplies from generic
PCI whatever-3.3v through additional regulators. All these supplies are
there - on the board.
>
>>> Also: I doubt you need DT for your detachable board?
>>
>> Sure, I don't need DT but that's not my point. My point is why require
>> these supplies for _all_ devices having PCI id 17cb:1101 (ie. QCA6390)
>> then clearly there are such devices which don't need it? To me that's
>> bad design and, if I'm understanding correctly, prevents use of
>> qcom,ath11k-calibration-variant property. To me having the supplies
>> optional in DT is more approriate.
>>
>
> We require them because *they are physically there*.
I understand that for all known DT QCA6390 hardware, the supplies should
be provided thus they should be required. If in the future we have
different design or we represent some pluggable PCI card, then:
1. Probably that PCI card does not need power sequencing, thus no DT
description,
2. If still needs power sequencing, you can always amend bindings and
un-require the supplies.
Best regards,
Krzysztof
On Fri, Jun 7, 2024 at 8:40 AM Krzysztof Kozlowski <[email protected]> wrote:
>
> Kalle,
> All of the devices out there need these supplies, but they are sometimes
> provided by generic PCI supply and on-board regulators. Basically your
> PCI adapter is not the same as QCA6390 chip on Snapdragon board.
>
>
> >
> >> Based on the discussions I have heard only Snapdragon systems who
> >> require all this configuration you describe. Of course there can be
> >> other systems but I have not heard about those.
> >>
> >
> > DT is not configuration, it is description of actual hardware. It
> > doesn't matter if Snapdragon systems are the only ones that actually
> > *require* this description to make WLAN/BT functional upstream. The
> > chipset would be the same on any PCIe board, it's just that the host
> > systems wouldn't need to take care with its power sequence. But for a
> > dynamic board like this, you don't need DT.
> >
>
> Correct.
>
> ...
>
> >
> >>> If your detachable board "just works" then it must be wired in a way
> >>> that enables WLAN the moment it's plugged in but this doesn't happen
> >>> over PCI. The chipset has a power input and GPIOs to enable each
> >>> module.
> >>
> >> I don't know how the boards are implemented but it could be so. But from
> >> host system point of view it's just a regular PCI device.
> >>
> >
> > And you don't need DT anyway for this type of devices.
>
> Detechable board, like PCI adapter, derives these supplies from generic
> PCI whatever-3.3v through additional regulators. All these supplies are
> there - on the board.
>
> >
> >>> Also: I doubt you need DT for your detachable board?
> >>
> >> Sure, I don't need DT but that's not my point. My point is why require
> >> these supplies for _all_ devices having PCI id 17cb:1101 (ie. QCA6390)
> >> then clearly there are such devices which don't need it? To me that's
> >> bad design and, if I'm understanding correctly, prevents use of
> >> qcom,ath11k-calibration-variant property. To me having the supplies
> >> optional in DT is more approriate.
> >>
> >
> > We require them because *they are physically there*.
>
> I understand that for all known DT QCA6390 hardware, the supplies should
> be provided thus they should be required. If in the future we have
> different design or we represent some pluggable PCI card, then:
> 1. Probably that PCI card does not need power sequencing, thus no DT
> description,
> 2. If still needs power sequencing, you can always amend bindings and
> un-require the supplies.
>
>
> Best regards,
> Krzysztof
>
Kalle, does the above answer your questions? Are these bindings good to go?
Bart
Bartosz Golaszewski <[email protected]> writes:
> On Thu, Jun 6, 2024 at 6:16 PM Kalle Valo <[email protected]> wrote:
>
>>
>> Bartosz Golaszewski <[email protected]> writes:
>>
>> > On Thu, Jun 6, 2024 at 4:02 PM Kalle Valo <[email protected]> wrote:
>> >
>> >> Sure, I'm not worried about functionality. I'm worried that if I
>> >> there's, for example, an ARM based setup which uses DT and wants to use
>> >> a similar QCA6390 board that I have, and set
>> >> qcom,ath11k-calibration-variant in DT. In other words, I'm worried if
>> >> you are looking at this only for Snapdragon family of boards?
>> >>
>> >
>> > No, what I'm looking at is the entire QCA6390 package. That means WLAN
>> > *and* Bluetooth *and* the PMU that manages power.
>>
>> I think we are just looking at this from different point of views. You
>> are looking at a datasheet (most likely for a Snapdragon based system)
>> and I'm looking what actual devices there are out in the field.
>>
>> > If you're using the QCA6390 on a device-tree system then you should
>> > probably model at least the WLAN node and the PMU and the problem with
>> > supplies is fixed.
>>
>> But why? If there are boards out there who don't need any of this why
>> would they still need to model all this in DT?
>>
>
> Because this is what is there? The goal of the device tree is to
> describe the hardware. The fact we didn't describe it before doesn't
> make it correct.
>
>> Based on the discussions I have heard only Snapdragon systems who
>> require all this configuration you describe. Of course there can be
>> other systems but I have not heard about those.
>>
>
> DT is not configuration, it is description of actual hardware. It
> doesn't matter if Snapdragon systems are the only ones that actually
> *require* this description to make WLAN/BT functional upstream. The
> chipset would be the same on any PCIe board, it's just that the host
> systems wouldn't need to take care with its power sequence. But for a
> dynamic board like this, you don't need DT.
>
>> > But if you don't have the supplies, that's alright for downstream.
>>
>> What do you mean downstream in this context?
>>
>
> I mean: if you wanted to upstream the DT sources, then they should
> include the supplies AND the PMU node. But if you just want to make
> the WLAN run on some vendor kernel then you don't need to think about
> it, it will work.
>
>> >> Again, I don't see this as a blocker. I just want to understand how this
>> >> should work for all types of devices there are out there.
>> >>
>> >> > But if you have a QCA6390 then you have its PMU too and the bindings
>> >> > model the real-world hardware.
>> >> >
>> >> > IOW: your laptop should be alright but the supplies are really there
>> >> > which warrants adding them to the bindings.
>> >>
>> >> Sorry, not following here. Can you clarify your comment "the supplies
>> >> are really there"? You mean inside the PCI board? But that's not visible
>> >> to the kernel in anyway, the PCI board just works after I plug it in.
>> >> It's like a regular PCI device. So I don't understand why that should be
>> >> visible in DT, but I can very well be missing something.
>> >>
>> >
>> > I think you're thinking about some kind of detachable PCIe board with
>> > this chipset on it.
>>
>> Exactly, a lot of WLAN boards are like this.
>>
>> > I refer to the QCA6390 chipset itself which is also more than just
>> > PCI. The Bluetooth interface doesn't use PCI at all. On the boards I'm
>> > working on, the chipset is just soldered to the main board.
>>
>> And I guess you are looking at Snapdragon boards only?
>>
>
> But what is your point?
My point (again) is that to me it look likes that you are looking this
only for Snapdragon type of devices and ignoring the rest. I am looking
at this to support _all_ type of devices and I want to make sure that we
don't have any artificial restrictions to use ath11k or ath12k devices
in upstream Linux.
I could not find a public example of a QCA6390 M.2 board like I have, but
here's one for QCA2066:
https://compex.com.sg/shop/wifi-module/wlt206h-wifi6-ble5-1-11ax-qca2062-qca2065/
QCA2066 is a mobile chipset supported by ath11k, similarly like QCA6390.
It's just newer and different features, and with a different PCI id. In
the past using these kind of M.2 boards for Wi-Fi has been quite common
but don't know how commit it is nowadays.
>> > If your detachable board "just works" then it must be wired in a way
>> > that enables WLAN the moment it's plugged in but this doesn't happen
>> > over PCI. The chipset has a power input and GPIOs to enable each
>> > module.
>>
>> I don't know how the boards are implemented but it could be so. But from
>> host system point of view it's just a regular PCI device.
>>
>
> And you don't need DT anyway for this type of devices.
I wish we wouldn't need to use DT for such M.2 boards, but we do need to
use qcom,ath11k-calibration-variant in some cases when the device (or
the firmware) doesn't provide unique enough identifier to choose the
correct board file automatically. I already mentioned the property in my
earlier emails.
I hope this clears up what I'm trying to say.
--
https://patchwork.kernel.org/project/linux-wireless/list/
https://wireless.wiki.kernel.org/en/developers/documentation/submittingpatches