Re: iwn(4) hangs after r257133

Sure, flip on 'wlandebug +rate' (assuming you compiled with IEEE80211_DEBUG)


-a

On 9 November 2013 21:08, Brandon Gooch <jamesbrandongooch_at_gmail.com> wrote:
> On Sat, Nov 9, 2013 at 8:46 PM, Adrian Chadd <adrian_at_freebsd.org> wrote:
>> Hi!
>>
>> On 9 November 2013 18:29, Brandon Gooch <jamesbrandongooch_at_gmail.com> wrote:
>>> Turns out that not enabling MRR causes my Intel Ultimate N WiFi Link
>>> 5300 to hang after only a few moments of use.
>>
>> That's .. odd. Ok.
>>
>>> For now, I've just reverted only those aspects of r257133, enabling
>>> MRR and keeping the rate index lookup, which seems to do something on
>>> my hardware at least (I assume it's not the right thing based on
>>> Adrian's analysis, but it works never-the-less).
>>>
>>> Has anyone else hit this with Intel WiFi hardware?
>>>
>>> Also, what needs to be done to have MRR working properly?
>>
>> So, it could be a fall out of how utterly crap AMRR is at 11n rates.
>>
>> Please compile with IWN_DEBUG, then do this before you associate:
>>
>> sysctl dev.iwn.0.debug=0x1
>>
>> (that's IWN_DEBUG_XMIT in sys/dev/iwn/if_iwn_debug.h)
>>
>> You can do the same on a kernel with and without the MRR stuff enabled.
>>
>> I'd like to see what the actual rate selection looks like and what the
>> final rate selection is. We may have to patch the kernel to print out
>> the contents of 'rate' and 'tx->rate' in iwn_tx_data() to get that.
>
> I've attached the log output, both with and without MRR.
>
> The output looks very much the same within iwn_tx_data(); in
> iwn5000_tx_done() things are
> clearly different.
>
> If I'm reading the rate conversion bits correctly, I see anywhere from
> 6 Mbps to 60 Mbps, with the
> the non-MRR module "getting stuck" -- I really don't know what has
> happened when it does this.
>
> Is there any further debugging output that would be helpful?
>
> -Brandon
>
>> The MRR stuff is a bit special. I don't know how the link table works
>> in great depth yet. I know it's broken for 11n; it's plainly using the
>> wrong indexes now. That's why I disabled it. It shouldn't take that
>> much work to get it in the tree again; it'll just be fiddly. The easy
>> bit is populating the table. The hard bit is knowing which index to
>> set linkq to when transmitting a frame.
>>
>>
>>
>> -adrian