12:00:55 <jki> #startmeeting CIP IRC weekly meeting
12:00:55 <collab-meetbot> Meeting started Thu Aug 20 12:00:55 2026 UTC and is due to finish in 60 minutes.  The chair is jki. Information about MeetBot at http://wiki.debian.org/MeetBot.
12:00:55 <collab-meetbot> Useful Commands: #action #agreed #help #info #idea #link #topic #startvote.
12:00:55 <collab-meetbot> The meeting name has been set to 'cip_irc_weekly_meeting'
12:01:00 <jki> #topic AI review
12:01:28 <jki> nothing on the list - but I notice that there if a namespace collision now :D
12:01:39 <jki> "AI review"...
12:01:49 <jki> anyway
12:01:53 <uli> :)
12:01:54 <jki> 5
12:01:56 <jki> 4
12:01:59 <jki> 3
12:02:01 <jki> 2
12:02:03 <jki> 1
12:02:05 <jki> #topic Kernel maintenance updates
12:02:19 <uli> i pushed 4.4
12:02:20 <pave1> AI should really be called "LLM".
12:02:33 <pave1> I did reviews, 6.12.103 and 104.
12:02:40 <masami> This week reported 854 new CVEs and 43 updated CVEs.
12:02:59 <jki> normal numbers again, nice
12:03:42 <iwamatsu85> hello
12:04:11 <iwamatsu85> I did not work on this week.
12:04:48 <iwamatsu85> beacause I was on vacation.
12:05:06 <jki> hope you enjoyed
12:05:14 <iwamatsu__> :-)
12:05:37 <jki> anything else to add?
12:05:59 <jki> 5
12:06:01 <jki> 4
12:06:03 <jki> 3
12:06:05 <jki> 2
12:06:07 <jki> 1
12:06:09 <jki> #topic Kernel release status
12:06:21 <jki> 6.1-rt was late - when I looked
12:06:21 <pave1> rt did suitable 6.1-rt, so ... I was lucky this time.
12:06:24 <jki> now we are fine
12:06:29 <jki> perfect
12:06:57 <jki> next to release would be 6.12
12:07:22 <jki> anything else on this?
12:07:24 <pave1> and I guess 4.4-rt.
12:07:40 <pave1> No need to wait when -cip is available.
12:07:56 <jki> 4.4-rt is only due in september
12:08:04 <jki> is it out of sync?
12:08:10 <pave1> aha, my notes say Aug 30, I'll double check.
12:08:32 <pave1> sorry, likely my mistake.
12:08:58 <jki> all fine
12:09:07 <jki> then let's move on
12:09:09 <jki> 5
12:09:11 <jki> 4
12:09:13 <jki> 3
12:09:15 <jki> 2
12:09:17 <jki> 1
12:09:18 <jki> #topic Kernel testing
12:09:49 <patersonc> Hello
12:10:02 <patersonc> The Ti LAVA lab is currently offline for maintenance
12:10:14 <patersonc> Not sure I have any other news
12:10:46 <jki> lava master had issues last Friday - did it auto-recover?
12:11:09 <patersonc> Yes, leter on Friday. I couldn't find a cause tbh
12:11:48 <jki> could be worse
12:12:12 <jki> ok, other topics for testing?
12:12:26 <jki> 5
12:12:28 <jki> 4
12:12:30 <jki> 3
12:12:33 <jki> 2
12:12:33 <iwamatsu__> https://lava.ciplatform.org/scheduler/device/qemu-prg-01 appears to be a configuration issue. Could you please check it?
12:13:12 <jki> will forward this notice, yes
12:13:41 <iwamatsu__> Thanks
12:13:42 <jki> BTW, "Physical access    Nobuhiro Iwamatsu" - not quite correct
12:13:51 <jki> I've seen that with more devices / labs
12:14:00 <jki> patersonc: who can fix that?
12:14:43 <patersonc> That worker needs to have `use_docker: true` set. Can only be done by the lab owner
12:14:48 <patersonc> We've had this issue before
12:15:20 <jki> No, I was referring to that description entry
12:15:54 <jki> but that other hint is also useful, forwarded as well
12:16:22 <patersonc> Oh I see - not sure how some boards have been assigned to Iwamatsu-san. I can fix this in the admin interface
12:16:32 <jki> TIA!
12:17:06 <jki> then move on
12:17:09 <jki> 3
12:17:11 <jki> 2
12:17:13 <jki> 1
12:17:15 <jki> #topic AOB
12:17:27 <jki> - LLM review tool news
12:17:35 <uli> the sltr scripts needed a bit of a fix for the new server setup
12:17:40 <uli> pushed it this morning and sent an email
12:18:14 <pave1> I got enough setup to talk to LLM via web interface.
12:18:27 <pave1> Well, it is a LLM. :-)
12:18:47 <jki> then it's probably answering ;)
12:19:06 <uli> michael set the default to qwen 3.8. i haven't tried that at all yet.
12:19:19 <pave1> Yeah. I have seen better models and I have seen worse models. Evaluating them is hard.
12:20:00 <uli> i have a set of known-good and known-bad patches on top of 4.19 that i use for evaluation
12:20:23 <pave1> That's a way to go.
12:20:27 <uli> i wonder how/where i can put them up so they won't be confused with actual patches
12:20:43 <uli> because they need to look realistic i cannot change authors or commit messages etc.
12:20:46 <pave1> Are we evaluating for "will it break things" or "does it match stable kernel criteria"?
12:20:58 <uli> for the default prompt it's the former
12:21:19 <uli> but you can put whatever you want in it, of course
12:21:36 <uli> i'm using it, among other things, to check if i accidentally pasted my passwords or keys in the commit message :)
12:21:58 <pave1> Yeah, LLMs are versatile.
12:22:06 <jki> put those tests in a separate kernel tree, or just separate, clearly named branches?
12:22:22 <pave1> jki: We don't want llm to find that tree :-)
12:22:46 <pave1> I was thinking "highlight me patches where description says it is not for stable".
12:23:16 <pave1> And perhaps patches having stable-dep-of: are useful as a test set, they don't usually match the published stable kernel rules.
12:23:23 <jki> you cannot prevent that future trainings will find today's public work
12:23:40 <jki> and it may even be useful
12:24:05 <jki> but during the inference, that may not be the case - and it could be avoided by working offline, against a local checkout
12:24:48 <uli> what does a patch description say when it's _not_ for stable? i only know "Cc: stable", which is the opposite.
12:26:00 <pave1> uli: People usually say something like "this does not fix a bug so don't put it into stable". There's
12:26:14 <pave1> no common tag, and that's why LLM could be useful.
12:27:19 <uli> write something like "analyze the commit message and determine if the commit is intended and suitable for inclusion in the stable kernel tree." and then add some criteria.
12:27:22 <uli> should work
12:28:08 <pave1> ok.
12:28:16 <jki> ... and then review and fine-tune the criteria - or let the LLM look into past decisions
12:29:38 <jki> good - more on this?
12:30:10 <jki> ...or other AOBs?
12:30:44 <jki> 5
12:30:46 <jki> 4
12:30:48 <jki> 3
12:30:49 <jki> 2
12:30:51 <jki> 1
12:30:53 <jki> #endmeeting