Re: [PATCH 3/9] rt2x00: Align RT chipset definitions with vendor driver.

[Gertjan van Wingerde <gwingerde@gmail.com>]

On 04/09/10 14:32, Helmut Schaa wrote:
> Am Freitag 09 April 2010 schrieb Felix Fietkau:
>> On 2010-04-09 2:23 PM, Helmut Schaa wrote:
>>> Am Freitag 09 April 2010 schrieb Felix Fietkau:
>>>> On 2010-04-09 1:32 PM, Helmut Schaa wrote:
>>>>> Am Freitag 09 April 2010 schrieb Felix Fietkau:
>>>>>> On 2010-04-09 7:10 AM, Gertjan van Wingerde wrote:
>>>>>>> On 04/09/10 00:28, Felix Fietkau wrote:
>>>>>>>> On 2010-04-08 11:50 PM, Gertjan van Wingerde wrote:
>>>>>>>>> Only include definitions for RT chipsets that are also used inside the
>>>>>>>>> Ralink vendor drivers.
>>>>>>>>>
>>>>>>>>> Signed-off-by: Gertjan van Wingerde <gwingerde@gmail.com>
>>>>>>>>> ---
>>>>>>>>>  drivers/net/wireless/rt2x00/rt2800lib.c |   13 -------------
>>>>>>>>>  drivers/net/wireless/rt2x00/rt2x00.h    |    7 +++----
>>>>>>>>>  2 files changed, 3 insertions(+), 17 deletions(-)
>>>>>>>>>
>>>>>>>>> diff --git a/drivers/net/wireless/rt2x00/rt2800lib.c b/drivers/net/wireless/rt2x00/rt2800lib.c
>>>>>>>>> index 394c8e4..4bc7e09 100644
>>>>>>>>> --- a/drivers/net/wireless/rt2x00/rt2800lib.c
>>>>>>>>> +++ b/drivers/net/wireless/rt2x00/rt2800lib.c
>>>>>>>>> @@ -1209,10 +1209,7 @@ int rt2800_init_registers(struct rt2x00_dev *rt2x00dev)
>>>>>>>>>  	rt2x00_set_field32(&reg, MAX_LEN_CFG_MAX_MPDU, AGGREGATION_SIZE);
>>>>>>>>>  	if ((rt2x00_rt(rt2x00dev, RT2872) &&
>>>>>>>>>  	     (rt2x00_rev(rt2x00dev) >= RT2880E_VERSION)) ||
>>>>>>>>> -	    rt2x00_rt(rt2x00dev, RT2880) ||
>>>>>>>>>  	    rt2x00_rt(rt2x00dev, RT2883) ||
>>>>>>>>> -	    rt2x00_rt(rt2x00dev, RT2890) ||
>>>>>>>>> -	    rt2x00_rt(rt2x00dev, RT3052) ||
>>>>>>>>>  	    (rt2x00_rt(rt2x00dev, RT3070) &&
>>>>>>>>>  	     (rt2x00_rev(rt2x00dev) < RT3070_VERSION)))
>>>>>>>>>  		rt2x00_set_field32(&reg, MAX_LEN_CFG_MAX_PSDU, 2);
>>>>>>>>> @@ -1511,12 +1508,6 @@ int rt2800_init_bbp(struct rt2x00_dev *rt2x00dev)
>>>>>>>>>  		rt2800_bbp_write(rt2x00dev, 105, 0x05);
>>>>>>>>>  	}
>>>>>>>>>  
>>>>>>>>> -	if (rt2x00_rt(rt2x00dev, RT3052)) {
>>>>>>>>> -		rt2800_bbp_write(rt2x00dev, 31, 0x08);
>>>>>>>>> -		rt2800_bbp_write(rt2x00dev, 78, 0x0e);
>>>>>>>>> -		rt2800_bbp_write(rt2x00dev, 80, 0x08);
>>>>>>>>> -	}
>>>>>>>>> -
>>>>>>>> Why are you removing support for RT3052? IMHO those writes were
>>>>>>>> necessary, last time I tested the rt2800pci code on the RT3052 WiSoC.
>>>>>>>
>>>>>>> That is because I have not been able to find them in any of the Ralink vendor drivers.
>>>>>>> Actually, none of the Ralink vendor drivers mention an RT chipset that identifies itself
>>>>>>> as a RT3052. The only mentioning Ive seen is RT305x devices that identify themselves as
>>>>>>> RT2872 devices, but even for them I haven't found these BBP initializations.
>>>>>>> That's why I removed this part.
>>>>>>>
>>>>>>> I have no problem re-instating this if I can find some evidence that these devices
>>>>>>> actually exist.
>>>>>> These chipsets won't show up in STA-only drivers, because they belong to
>>>>>> embedded APs. If you download GPL sources for devices such as ASUS
>>>>>> RT-N15 you will find ifdefs for CONFIG_RALINK_RT3052 and the above
>>>>>> values in BBPRegTable in the driver sources.
>>>>>> I have a few devices based on RT3052, which is why I added this code.
>>>>>> At some point I even had basic Rx/Tx working on it, but haven't tested
>>>>>> in a while.
>>>>>
>>>>> I also couldn't find any evidence of the existence of an 3052 _rt_ chipset.
>>>>> However, the ralink drivers defines a 3052 _rf_ chip:
>>>>>
>>>>> #define RFIC_3052                   9       // 2.4G/5G 2T2R
>>>> RT3052 is the name of the whole WiSoC chip, not just the MAC or RF part
>>>> of it. Since wifi is integrated in the SoC, I don't think there is a
>>>> separate name for just the wifi part.
>>>
>>> There is. I have one 3052 and one 3050 board (basically a 3052 but only
>>> 1T1R), and both identify themselves as RT2872 with different rf "chips"
>>> (of course, there are no additional chips). Hence, the check for RT3052
>>> was never true on both platforms.
>> Back when I tested it, I forced the chip to identify itself as RT3052 by
>> taking the id from the platform device.
> 
> Ah, understood.
> 
>>>>> I don't have such an rf chip in my devices but I don't think the BPP
>>>>> register setup should depend on the actual rf chip. So, if the register
>>>>> setup is really needed we should maybe check for rt2x00_is_soc instead
>>>>> of removing the code?
>>>> Yes, but rt2x00_is_soc() is not enough, since RT2880 is also SoC, but
>>>> slightly different compared to RT3052. I've only tested RT3052 myself.
>>>
>>> Right. So, from what I've seen so far it seems like all 305x boards identify
>>> themselves as rt2872. Hence, we can just replace 3052 with 2872 in the above
>>> check and leave the code as is.
>> OK, if that doesn't conflict with any PCI based stuff...
> 
> According to Gertjan there seems to be no none-SOC adapter which identifies
> itself like that. So I guess that should be safe.
> 

(looks like you guys had some fun discussing things while I was away earning my money ;-))

Anyway, Helmut is right here. The correct check to make is to see if the chip
has identified itself as an RT2872.
However, based on the Ralink source-code I saw, there is more that we need to do, i.e. in
the RF CSR initialization space. Also, a lot of the default BBP CSR values are different
from the other RT chipsets.

I'll cook up a patch to properly add RT2872 support to the driver.

---
Gertjan