FreeBSD-8.0 802.11n support with ath/mwl

[Sam Leffler]

Jim Pingle wrote:
> On 2/28/2010 10:16 PM, Rui Paulo wrote:
>> On 1 Mar 2010, at 03:05, Jim Pingle wrote:
>>
>>> On 2/28/2010 9:41 PM, Rui Paulo wrote:
>>>> On 1 Mar 2010, at 02:26, Jim Pingle wrote:
>>>>> Ah, I wasn't aware of wlandebug(8). However, it doesn't seem to operate
>>>>> on this mwl(4) card. It sets the value of the sysctl net.wlan.0.debug
>>>>> and that doesn't show up on my system. Another system with a ral(4) card
>>>>> does have that sysctl. Judging by the information in the wlandebug(8)
>>>>> man page it appears as though this may be a side effect of mwl doing
>>>>> much of the work in firmware.
>>>> wlandebug takes an -i argument. I seem to recall you created your wlan interface named "mwl_wlan0", so you need to type wlandebug -i mwl_wlan0.
>>> I saw that, but that is hardcoded to expect wlan<x> (wlan0, wlan1, etc)
>>> for an interface name. Having seen that, I recompiled wlandebug without
>>> the hardcoded interface name check and it didn't work either, but it did
>>> toss an error for the sysctl it was trying to tweak.
>> The whole system was designed for the interfaces to start with "wlan" and be named "wlan<something>".
> 
> It does certainly seem to lean that way. Personally I prefer to keep the
> hardware name in there, but I wasn't aware that would cause other
> issues. Especially when VAPs come into play, I'd rather have, for
> example, mwl0_wlan1, mwl0_wlan2, ath0_wlan0, etc, so I can better tie
> what goes where. But that's more of a bikeshed of personal preference. :-)

Not following the wlanX naming convention will cause confusion for 
things like rc config files (I believe).  Definitely any tool I touched 
expects vap's to be named wlanX.

sysctl net.wlan.X.%parent gives the parent ifnet of a vap; I've 
considered many times including this in the ifconfig status.  Feel free 
to offer a patch that does this.

> 
>>> That made me look deeper at the code and see it was really just setting
>>> the debug sysctl based on flags that wlandebug was aware of. Handy, but
>>> the same thing could be done by hand with sysctl and some bitwise math
>>> in a pinch, assuming the interface has the right oids. (Which mine
>>> doesn't, for some reason...)
>> The purpose of wlandebug is to not do any math by hand.
> 
> Indeed, You're 110% right on that. I was just trying to work around the
> other issues I was seeing to get to the root of the issue, which seems
> to be the missing sysctl oid.
> 
> I need to run some more tests and straighten my antenna issues out, but
> I'll report what I find back to the list in a few days.

mwl 11n support broke sometime near the last firmware update and I never 
fixed it.  I know in particular there are issues with AMPDU and seq#'s 
but possible other things too.  The fw is rather finicky in how state is 
managed and it's likely the host is not in sync causing problems w/ the 
ampdu support and rate control algorithm that both operate in the fw. 
You should be able to get >100 Mb/s througput on an HT40 channel but I 
think I was seeing more like 35-40.  Turning off ampdu is usually 
helpful to stabilize operation.

	Sam