| 28 Nov 2025 |
 K900 | Yeah I was hoping for something smaller maybe | 18:37:46 |
 Yureka (she/her) | (also earlier, but at the latest 5-10min) | 18:37:52 |
| Yureka (she/her) | I don't think anyone has another repro yet | 18:38:14 |
| K900 | Can you try running it with GALLIUM_REFCNT_LOG=1? | 18:38:16 |
| K900 | Can you try running it with GALLIUM_REFCNT_LOG=refcnt.log? | 18:38:36 |
 dramforever | so i can reproduce it but don't get the file anywhere with this... | 23:51:17 |
| 29 Nov 2025 |
 ghpzin | It needs debug build (in case it was not one), do not think that env var does anything otherwise:
https://gitlab.freedesktop.org/mesa/mesa/-/blob/mesa-25.3.0/src/gallium/auxiliary/util/u_debug_refcnt.c#L27 | 00:57:22 |
 vcunat | Now I added a gcc15 jobset/eval, as the queue had basically just darwin stuff. | 08:59:53 |
| vcunat | (I'll micromanage if need be, so that it won't slow down e.g. 25.11 builds significantly.) | 09:00:28 |
| dramforever | oh no, oh no i can count the frames now... and it stopped crashing, perhaps because it's not rendering many frames to begin with? | 13:01:59 |
| dramforever | actually i'm going to try running the debug build without refcnt log | 13:04:29 |
| dramforever | would be interesting if it stops crashing | 13:04:43 |
| dramforever | ooh it crashes with just the debug build, time to keep refcnt logging | 13:07:38 |
| dramforever | i got the log but it's like 3 gigs | 13:18:15 |
| dramforever | do you have anything specific you're looking for? | 13:18:52 |
| K900 | Not particularly | 13:19:31 |
| K900 | Refcounts going negative possibly | 13:19:39 |
| dramforever | oh hey
<pipe_martian_resource<229>> 0x7fff47038d00 26137 Create
<pipe_martian_resource<229>> 0x7fff47038d00 26137 AddRef 3857049062
| 13:23:07 |
| dramforever | is this ... bad? | 13:23:11 |
| dramforever | oh that's 0xe5 | 13:24:07 |
| dramforever | yeah that's pretty bad | 13:24:13 |
| K900 | Uh oh | 13:24:27 |
| K900 | Yeah that looks bad | 13:24:30 |
| K900 | "pipemartianresource" is also kinda weird | 13:24:54 |
| K900 | pipe_martian_resource is also kinda weird | 13:24:58 |
| K900 | https://elixir.bootlin.com/mesa/mesa-25.3.0/source/src/gallium/auxiliary/util/u_debug_describe.c#L70 | 13:25:12 |
| K900 | That's a "we don't know what this is" type | 13:25:17 |
| dramforever | i think that makes sense if the entire struct has been memset(0xe5)'d | 13:25:36 |
| K900 | Oh yeah | 13:25:46 |
| dramforever | including the refcnt 3857049062 which is 0xe5e5e5e6 | 13:26:01 |
