Re: [PATCH net-next v2] dt-bindings: net: ath11k: document the inputs of the ath11k on WCN6855

[Bartosz Golaszewski]

On Mon, Sep 9, 2024 at 10:19 AM Bartosz Golaszewski <brgl@xxxxxxxx> wrote:
>
> On Fri, Sep 6, 2024 at 8:38 PM Jeff Johnson <quic_jjohnson@xxxxxxxxxxx> wrote:
> >
> > On 9/6/2024 12:44 AM, Bartosz Golaszewski wrote:
> > > For upstream - if you're using the WCN6855, you must specify the
> > > inputs for the WLAN module so it's only fair they be described as
> > > "required". For out-of-tree DTS I couldn't care less.
> > >
> > > You are not correct saying that "M.2 boards don't need these" because
> > > as a matter of fact: the WLAN module on your M.2 card takes these
> > > inputs from the PMU inside the WCN6855 package.
> >
> > Let me start by saying that DT is one area where I'm a newbie, so I hope I can
> > get some education.
> >
> > I'd like to start with an observation: I've used both WCN6855 with ath11k and
> > WCN7850 with ath12k on an x86 laptop without any device tree, so from that
> > perspective none of the device tree stuff is "required" -- these modules "just
> > work".
> >
>
> Yes. This is what I refer to as "fully dynamic" M.2 cards, where the
> card typically has an on-board PMIC that handles the power-up of the
> device, respecting all timings etc. No custom pins are used. You don't
> need device-tree. DT bindings don't concern this case. Even it this
> was an ARM, DT-based platform, you wouldn't need the DT entry.
>
> > However I also realize that when these are installed on Qualcomm ARM platforms
> > that there are GPIO pins that control things like XO clock, WLAN enable &
> > Bluetooth enable, as well as voltage regulators, and the device is
> > non-functional without those configured, so the device tree items are required
> > in that environment.
> >
> > So just from that perspective saying something is "required" is confusing when
> > there are platforms where it isn't required. And perhaps that is what is
> > confusing Kalle as well?
> >
>
> The properties are required IF you have a DT representation. Because
> if you're modeling the physical package, this is what it looks like.
> The one on your "fully dynamic" M.2 card is the same - it also has the
> same internal inputs and outputs but you're not modeling the external
> package in the first place so you don't need to care about them. But
> if you do represent the chipset and not as a black box WCN6855 in its
> entirety but its WLAN, BT and PMU modules separately then it warrants
> making the true inputs of the WLAN module mandatory in the schema.
>
> Please let me know if this is enough of an explanation.
>
> Bart

One more thing to add. You guys worry about existing device-trees,
don't, they will continue to work fine. This change only mandates that
upstream DT sources model the inputs of the WLAN module and it's only
verified by make_dtbs. This doesn't make the code require them.
Whatever works for you now, will keep on working.

Bart