Re: Fwd: HT and VHT testing for modern iwlwifi chipsets (notes about others)

On 03.03.2025 23:47, Bjoern A. Zeeb wrote:
> On Mon, 3 Mar 2025, Eirik Øverby wrote:
> 
>> I don't know why, but the mailing lists seem to swallow all my emails 
>> as of yesterday or so. Please ignore this if it turns out to be a 
>> duplicate.
> 
> I havenn't made it to the list yet.
> 
> Likely because of binary attachments if they don't make it through.

I seem to be unable to send to any mailing list right now, for reasons I 
cannot fathom. Have emailed postmaster@.


> It's a TLC_MNG_CONFIG_CMD in iwlwifi.  I'll have to check if I can
> figureit out;  I wonder if something went wrong when coming from 11a to
> 11g;  I assume you don't have a ifconfig -v wlan0 output from when this
> happend;  that along with your logs is very valuable (the full wpa log
> in this case not so much).

Attached full set of logs here, including a number of ifconfig -v 
outputs. The first one in that file is from before I try to 
disable/enable; the others are with a few seconds interval after doing 
so. As you can see, it does associate for a while, but then drops again. 
At this point I have to reboot.

Note that there has been *no* suspend/resume between last night and this 
morning. There was a suspend/resume cycle earlier in the day, but it did 
not cause the problem - so that seems unrelated.

> If you have a firmware crash normally service netif restart wlan0 should 
> fix it.
> If it doesn't you could try to do a service netif stop wlan; kldunload 
> if_iwlwifi
> which will automatically come back right away and if needed service
> netif start wlan0.

ifconfig wlan0 destroy
service wpa_supplicant stop
devctl reset pci0:170:0:0
devctl disable pci0:170:0:0
sudo kldunload if_iwlwifi
devctl enable pci0:170:0:0

This is what I do - I know the wpa_supplicant and devctl reset are 
superfluous. This helped yesterday, but not today.

> Once the firmware crashed it's done for the moment; the hw_scan is just
> a net80211 driven follow-up event.  we'll one day need to implement the
> hardware restart to "cover" these events but then all kinds of things
> will happen -- sometimes the user may not even notice this happens and
> we'll never know about them.

Hope these logs help; right now it seems like i have to reboot about 
once per day, which isn't nice :)

But what does one not do for speeeeeeed....? ;)

Take care,
/Eirik


> Lots of health,
> Bjoern
> 
> 
>> Take care,
>> /EIrik
>>
>> -------- Forwarded Message --------
>> Subject: Re: HT and VHT testing for modern iwlwifi chipsets (notes 
>> about others)
>> Date: Mon, 3 Mar 2025 20:15:26 +0100
>> From: ltning-freebsd-wireless@anduin.net
>> To: wireless@freebsd.org
>>
>> [Re-sending this since it did not show up on the list; compressed one 
>> of the attachments in case size was an issue]
>>
>> On 28.02.2025 22:35, Bjoern A. Zeeb wrote:
>>> I am sure there'll be plenty of rough edges to fix for me... but I
>>> wanted to get the code out at this point.
>>
>> Hi, perhaps this is one of the rough edges:
>> Attaching a couple logs - the wifi stopped working out of the blue 
>> today; the machine was idle but on (no suspend/resume). It took a bit 
>> of faffing around with devctl detach/attach and disable/enable to get 
>> it working again.
>>
>> The two logs are from
>>   - messages starting yesterday; error happened at 15:07 today
>>   - dmesg from an attempt to 'service netif restart'
>>
>> Last one is /var/log/debug.log with wpa_supplicant output.
>>
>> Is any of this useful? Anything else I should provide?
>>
>> The same symptom happens after suspend/resume: wlan0 is brought up, 
>> associates, never gets an IP, then disassociates and dmesg starts 
>> showing the "ERROR: lkpi_ic_scan_start: hw_scan returned -5" messages. 
>> This time I had to reboot the machine to recover, disable/enable does 
>> not help - just triggers the same sequence of events.
>>
>> When this happens it looks like it's able to scan and associate once - 
>> an ifconfig wlan0 scan shows the expected list of networks immediately 
>> after 'devctl enable'. Once it disassociates again, a scan returns an 
>> empty list.
>>
>> /Eirik
>