16:40 mhenning[d]: phomes_[d]: Okay, I manged to get before/after renderdoc captures for the raw_access_chains MR. Could you either share your script or run it on the data I've collected?
16:54 phomes_[d]: Oh I thought it was not possible with raw access chains. I am on the phone now but will share the script when I get online later. I have a few other scripts so I thought I would create a repo in gitlab. Then I got stuck thinking of a good name for it 🙂
16:55 mhenning[d]: oh yeah, naming things is the hard part 🙂
16:57 mhenning[d]: I patched my local renderdoc to allow raw_access_chains, which is enough to get some timing info out of it
17:54 Siggi: Blegh, I was hoping I could pass the K2100M through to a KVM to play with it, but it seems the BIOS or hardware doesn't support that. Maybe I'm missing something - anyone done this on a mid-2011 iMac (21.5")
17:59 f_: Siggi: your iMac is LVDS right?
18:00 f_: in my case my display is internally using eDP. And yes it does also happen in fbcon
18:00 Siggi: mine is eDP
18:01 f_: it also only, only happens on the internal display. Any displays connected over VGA or even DisplayPort do not exhibit this issue
18:01 f_: if there are any experiments that would help to try, ping me
18:02 Siggi: I wish I knew how to dump the crtc registers. This smacks of a timing issue.
18:02 Siggi: I was hoping running under a K(VM) would allow me to at least observe the memory reads/writes to the card during initialization
18:03 f_: also, the eDP is 1920x1080, the external DP monitor is 3440x1440 and the VGA one is 1680x1050. Can barely handle 3440x1440 but I don't have these flickering issues (only sometimes feels like slightly have less fps)
18:03 Siggi: but alas, it looks like my hardware is not going to cooperate
18:03 f_: Siggi: why? Can't you add/enable debug printf's to your kernel?
18:03 f_: would be much easier than figuring out kvm
18:04 Siggi: I could, but that won't tell me what the nvidia driver is writing to the HW
18:05 Siggi: Ashahi linux apparently reversed the M1 hardware, including the GPU (to an extent) by interposing on the HW memory traffic
18:05 Siggi: (I did the same thing back in the 90s for the sound chip in my Mac IIcx, by rewriting the paging tables and tracing the memory traffic)
18:06 Siggi: (there was a lot of stuttering :)
18:07 Siggi: I'm working off the hypothesis that either the nvidia driver has a workaround for a FW problem in these cards, or else there's a misunderstanding in communicating to the FW
18:08 f_: 19:04 <Siggi> I could, but that won't tell me what the nvidia driver is writing to the HW
18:08 f_: it would
18:08 Siggi: f_: ok? how?
18:08 f_: ah unless you mean the proprietary driver
18:08 Siggi: I can instrument the nouveau driver
18:08 Siggi: ... yeah
18:09 f_: for some reason thought you meant nouveau..
18:09 Siggi: the nvidia driver is unfortunately ... gotcha
18:10 Siggi: yeah, the nvidia driver is able to init the hardware properly for my LCD panel, so I was hoping to either dump the crtc registers for both and dump them
18:11 Siggi: or else see how nvidia (and nouveau) are initializing the HW and see what sticks out