08:02 daniels: FWIW GitLab is going to be down for around half an hour, probably in 45-60min
08:02 daniels: need to do some unpleasant database mangling to restore all of whot's data
12:21 hanetzer: oof.
12:24 udovdh: hello
12:25 udovdh: [drm:amdgpu_job_timedout [amdgpu]] *ERROR* ring gfx timeout,(etc) still happens on 5.6.x
12:25 udovdh: https://pastebin.com/wFzQz145
12:25 udovdh: what can I try to solve this?
12:25 udovdh: 5.6.13 actually
12:26 udovdh: any ideas
12:30 udovdh: ?
12:38 daniels: gitlab is back btw
12:39 udovdh: In the log at https://pastebin.com/wFzQz145 I was not at the keyboard....!
12:42 udovdh: this is Fedora 31 on AMD Ryzen 5 3400G with Radeon Vega Graphics w/ git mesa, etc
17:32 jekstrand: HdkR: There are ICL power management issues on recent kernels. Kayden knows the details
17:35 robclark: anholt, MrCooper: I noticed something kinda strange.. I get two pipelines for the same commit, but one has additional radv jobs.. https://gitlab.freedesktop.org/robclark/mesa/pipelines/147854 vs https://gitlab.freedesktop.org/robclark/mesa/pipelines/147855 .. I guess not really a problem, I just start the pipeline with fewer jobs, but seemed strange so I thought I'd point it out in case it was unexpected
17:37 Kayden: HdkR: I haven't seen that one. I'm using 5.6.10-arch1-1 currently. /sys/class/drm/card0/gt_{cur,act}_freq_mhz will show the kernel requested current (cur) and actual firmware-accepted current (act) frequency
17:38 Kayden: In the past I've heard of some suspend/resume issues
17:39 Kayden: there was a patch where i915 started interpreting various IRQs as "clock down" and there was a core platform IRQ bug for icelake that sent bogus ones periodically, causing it to clock down all the time
17:39 Kayden: curro knows the details there, I don't recall the exact patches
17:39 Kayden: (one's from Chris, one's from Srinivas, iirc)
17:40 ickle: no
17:41 ickle: the bug was that it didn't respond to the register write, so it never changed
17:43 ickle: the fun with idling is that we idle within an evaluation period, and so it never completed an ei and sent interrupts before we went to sleep, so it would always stay at the boost frequency
18:13 jekstrand: Kayden: Have you tried suspend/resume on your laptop recently? Mine can't wake up anymore. :-(
19:13 Kayden: jekstrand: I haven't tried suspend/resume in years
19:14 Kayden: I still run IRC on my laptop without a bouncer, so I just leave it on continually
19:37 MrCooper: robclark: expected, been like that for a long time :) one pipeline for the personal branch, the other for the MR
19:38 robclark: right.. I was more just curious about why they have different # of jobs