08:22 <kusma> daniels: So a variation of this? http://web.archive.org/web/20220702055143/https://bash.org/?947444 :)

11:25 <daniels> haha

15:19 <kusma> zmike: I'm seeing crashes in RADV when running Q3A on top of Zink+RADV, due to what appears to be missing image memory attached to an image view...? Does that ring any bell to you?

15:21 <kusma> The code in RADV where this happens does consider some plane-stuff... I also notice that while iview->vk.format VK_FORMAT_B8G8R8A8_UNORM, iview->image->vk.format is 33751040, which seems a bit bonkers...

15:24 <kusma> Also other members of iview->image->vk seems... off. Like: extent = {width = 2473000960, height = 55826, depth = 0}, mip_levels = 0, array_layers = 1148846080, samples = 0

15:28 <kusma> Valgrind seems to have something to say about the matter: https://gitlab.freedesktop.org/-/snippets/7364

15:29 <kusma> That first one is by coincidence very close to where I see the null deref

15:29 <kusma> (and with "by coincidence" I mean probably not at all)

15:48 <kusma> Also, seems to be the same culprit as the problem I mention here: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20613#note_1732238

15:53 <kusma> increasing the refcount of cdt doesn't seem to help :/

16:08 <zmike> 🤔 haven't seen before but I can check when I get home if you give me a repro?

18:00 <kusma> Just run quake3

18:00 <zmike> I don't have quake3?

18:01 <kusma> I see... Seems to be the same as the doom3 issue, BTW

18:01 <zmike> so I saw

18:02 <kusma> I can try to capture the issue on Monday if needed

18:04 <zmike> I've got a doom trace I'll try, if that doesn't work then see if you can get a q3 trace that hits it

18:04 <kusma> I guess it's intentional that this is a weak reference? That object is usually recounted...

18:04 <zmike> yes it's just used for pointer comparisons

18:04 <kusma> But if it's just to keep track of changes, I think it's safe

18:05 <kusma> Yeah, exactly

18:05 <kusma> I guess we could miss an update if the object gets deleted and then realloced with the same pointer or some garbage like that...

18:06 <kusma> That seems extremely unlikely, given that the swapchains kinda live longer than ideal ATM, though

18:07 <zmike> dt doesn't get deleted though

18:08 <zmike> that's the outer object

18:08 <kusma> Aha

18:08 <zmike> hm maybe I didn't keep up with the kopper changes as well as I thought

18:08 <zmike> yea ok I think I see it

18:09 <kusma> Addrefing didn't help for me

18:10 <zmike> it wouldn't

18:12 <zmike> I think this should do it...

18:14 <zmike> kusma: I think https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20814 should do it

18:14 <zmike> though my doom testing was inconclusive

18:15 <kusma> Yeah, that looks likely. I'm out and about to have a few cold ones, but I can test tomorrow.

18:16 <zmike> have another cold one for me

18:16 <kusma> 🍺

18:16 <zmike> 👍

19:56 <zmike> these validation errors are actually insane

19:59 <zmike> ajax: can I get an ack on https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20814