[iwinfo PATCH] devices: add support for declaring compatible matched devices

[Andre Heider]

On 09/01/2023 20:17, Christian Marangi wrote:
> On Mon, Jan 09, 2023 at 08:14:36PM +0100, Andre Heider wrote:
>> On 09/01/2023 19:58, Christian Marangi wrote:
>>> On Mon, Jan 09, 2023 at 07:53:02PM +0100, Andre Heider wrote:
>>>> On 09/01/2023 19:46, Christian Marangi wrote:
>>>>> On Mon, Jan 09, 2023 at 07:44:34PM +0100, Andre Heider wrote:
>>>>>> On 09/01/2023 18:28, Christian Marangi wrote:
>>>>>>> From: Jo-Philipp Wich <jo at mein.io>
>>>>>>>
>>>>>>> Some device have embedded wifi card that are not connected with usb or
>>>>>>> internall with pci. Such device have fake device_id and only the
>>>>>>> vendor_id actually reflect something real but internally they don't have
>>>>>>> any id and are just matched by the node compatible binding in DT.
>>>>>>
>>>>>> Nice cleanup! But those fake entries in devices.txt can then be removed,
>>>>>> right? (Assuming all of those _are_ fake and not mapped to actual pci ids)
>>>>>>
>>>>>
>>>>> But they are dropped. Am I missing something? Everything with compatible
>>>>> doesn't have the id declared and internally they are all set to 0.
>>>>
>>>> The code that maps them to ids is dropped, the id entries mapping those to
>>>> the result values are not (which is why this is so ugly ;)
>>>>
>>>> For the first if() that's:
>>>>
>>>> diff --git a/devices.txt b/devices.txt
>>>> index e0663b8..040766c 100644
>>>> --- a/devices.txt
>>>> +++ b/devices.txt
>>>> @@ -141,7 +141,6 @@
>>>>    0x168c 0x002a 0x0777 0xe202   12      0  "Ubiquiti" "Bullet M2"
>>>>    0x168c 0x002a 0x0777 0xe805    5      0  "Ubiquiti" "Bullet M5"
>>>>    0x168c 0x002a 0x0777 0xe345    0      0  "Ubiquiti" "WispStation M5" /*
>>>> ToDo: confirm offset - Wrong! */
>>>> -0x168c 0x0029 0x168c 0x9130    0      0  "Atheros"  "AR9130"
>>>>    0x168c 0x0029 0x168c 0xa094    0      0  "Atheros"  "AR9220"
>>>>    0x168c 0x0029 0x168c 0xa095    0      0  "Atheros"  "AR9223"
>>>>    0x168c 0x002a 0x168c 0xa093    0      0  "Atheros"  "AR9280"
>>>
>>> Still I'm not following... the only place we did compatible matching was
>>> in nl80211_hardware_id_from_fdt. Stuff that was present in devices.txt
>>> is all matched from pci id. Unless they are faked by some driver iwinfo
>>> in theory is using provided pci id to do the match.
>>>
>>> If that is not the case then I have no idea how iwinfo is taking the id if
>>> the connected pci card is not providing them.
>>>
>>> Happy to fix if you can give me better example but I wasn't aware we had
>>> pci device with fake id. Am I missing a patch on a different place?
>>
>> Drivers are not faking anything, iwinfo is.
>>
>> Once upon a time it only supported pci based cards, using the four id
>> values. Later support for non-pci dt-compatible was hacked in. For that, it
>> maps those compatible strings to made up pci ids in code, and at the same
>> time added those fake ids to devices.txt.
>>
>> Like the recent commit here:
>> https://git.openwrt.org/?p=project/iwinfo.git;a=commitdiff;h=5914d7113ecf77de63eb21fc233684d1a1a52ca5
>>
>> This patch cleans that up, but it left out the device entries, which are
>> dead now.
>>
>> Hope that clears it up?
> 
> OHHHHHHH ok yes!
> I was aware that iwinfo was faking ids but totally forgot the entry were
> still there in devices.txt
> 
> Let me send v2 with the entry dropped... Wonder if you can take an extra
> check in case I miss some.

Sure.

BTW, not directly related, but mainline linux also supports something 
related to map dt wifi devices to pci ids so the desired driver binds:

$ git grep 'compatible = "pci'
...
target/linux/ath79/dts/ar7161_aruba_ap-105.dts- ath9k0: wifi at 0,11 { /* 
2.4 GHz */
target/linux/ath79/dts/ar7161_aruba_ap-105.dts:     compatible = 
"pci168c,0029";
...

iwinfo could evaluate those embedded pci ids to then properly match 
based on pci instead of dt-compatibles. I don't have a board using that 
feature though.

Cheers,
Andre