Re: PR 297155 and PR 297184

> On Sep 10, 2026, at 4:03 AM, Bjoern A. Zeeb <bz@FreeBSD.org> wrote:
> 
> On Mon, 31 Aug 2026, Bjoern A. Zeeb wrote:
> 
> Hi,
> 
>> On Sun, 30 Aug 2026, Kim Shrier wrote:
>> 
>> Hi,
>> 
>>> I am in the process of implementing HOSTAP mode in the
>>> LinuxKPI 802.11 compat layer.
>> 
>> Great.
>> 
>> Are you doing this "clean room"?
>> 
>> Do you have the code anywhere?
> 
> I would still be curious about this.
> Feel free to reply privately as well.

I did use claude to outline how to develop
the missing pieces to have HOSTAP mode
working with the hardware I had laying around.

I also had claude generate some of the code.  I am
not pleased with the current results and the code
seems fragile.  I want to overhaul all of it before
I turn it loose.  In fact, in my efforts to clean it up,
it has broken again.  My laptop can connect but
my phone can’t.

Right now all the code lives only on my test machine.

> 
> 
>>>  I discovered the two bugs
>>> mentioned in the subject while working on this code.
>>> On ath10k/QCA6174 hw3.2 I now have a WPA2-PSK AP
>>> that associates real clients (including an iPhone) at HT rates,
>>> rekeys the GTK, and keeps power-saving clients associated
>>> across sleep; it's 2.4GHz-only so far, VHT is plumbed but
>>> untested, and I'm chasing one panic that looks like an upstream
>>> ath10k tx-pause locking issue rather than something in my code.
>> 
>> 
>> At least in STA mode it currently suffers from a (*set_bitrate_mask)
>> issue as well with LinuxKPI.
>> 
>> I think I also still have small a cleanup patch not comitted to
>> FreeBSD yet but I can go and look. Yes:
>> % git diff --stat main sys/contrib/dev/athk/ath10k/
>> sys/contrib/dev/athk/ath10k/pci.c | 10 ----------
>> 1 file changed, 10 deletions(-)
> 
> I pushed that cleanup.

Thanks.

> 
> 
>>> I have patches submitted with the PRs.  I don’t want to be an
> 
> I beleive both are fixed now and merged to stable/15.
> 
> Let us know if there's anything else.

I may have identified 1 or 2 other bugs in the adjacent code
but I have not proved that to my satisfaction yet.

> 
> 
> /bz

Kim
--
C++ is an off-by-one error