00:23anholt_: hmm. u-boot jumping into my zImage stops at "Starting kernel..." on the tk1, even with "earlycon=uart8250,mmio32,0x70006000 earlyprintk"
00:24anholt_: the uncompressed Image is 26MB, so I suspected that was too big for the fixed addresses, but moving fdt to 0x88... address to get it out of the way didn't help.
00:29karolherbst: I had the exact same issue once
00:30karolherbst: but my solution was to put it back into my shelf, and when I was ready to look into it, it magically worked again
00:31karolherbst: anholt_: ohh.. do you have console set?
00:31karolherbst: I remember, that was the fix
00:31karolherbst: although earlycon and console or more or less the same, no?
05:51anholt_: karolherbst: yeah, console=ttyS0,115200n8 as well, but not getting any earlycon either
06:59airlied: TimurTabi: so nouveau pmu.c code actually parses the older version of the falcon ucode format tables
07:00airlied: which you may already know, or not!
13:04karolherbst: okay.. gnome 42 runs on nvc0 and nv50 without issues with my MT fixes :)
13:04karolherbst: nv30 seems to have some issues :(
13:05karolherbst: it's broken without my patches, but differently
13:07karolherbst: ehh.. it's the fencing commit this time, interesting
13:07karolherbst: ahh yeah.. cause I didn't fix nv30 there
13:22karolherbst: glennk: btw, your mesa repo is private, not sure if this is on purpose
15:11Wally: Sorry, whats svm?\
15:54karolherbst: why after switching my nv43 with a different one does gnome-42 look perfect?
15:54karolherbst: granted... the other nv43 also got seriously hot
15:55karolherbst: maybe something with GDDR3 vs DDR1
16:59anholt_: both my nv4xs were completely unstable, I wonder if the hw is just actually end-of-life at this point.
17:00karolherbst: anholt_: yeah.... the one showed a lot of graphical artifacts, but the fencing patch in my MR made it even worse
17:00karolherbst: the other is totally fine
17:00karolherbst: just slow
17:01karolherbst: maybe I should check with all my nv4x GPUs
17:01karolherbst: I only used those without fans, because... the ones with fans are super loud
17:05karolherbst: ehh rebooted the wrong machine :(
17:15karolherbst: G73 also okay
17:15karolherbst: yeah.. I suspect after 15 years broken hardware isn't that unlikely :D
17:18karolherbst: "[ 12.238753] nouveau 0000:01:00.0: gpio: GPU is missing power, check its power cables. Boot with nouveau.config=NvPowerChecks=0 to disable." :)
17:18karolherbst: I am still surprised I get some output, but I think it's the fb one
17:18karolherbst: we might want to rephrase the message a little
17:20karolherbst: nv42 is fine as well
17:24karolherbst: something is up with my nv35, oh well
18:16karolherbst: Lyude: I am wondering.... ist the HDMI 2.1 stuff in nvidias driver?
20:52TimurTabi: The more I work with VBIOS, the less I understand why nvbios_addr exists.
21:00karolherbst: nice.... android simulator works without a single issue now :)
21:00karolherbst: or at least I think
21:00karolherbst: chrome doesn't do anything but not sure if that's y fault or not
21:00karolherbst: TimurTabi: mhh?
21:02TimurTabi: I just don't understand why we need to treat the second image as being adjacent to the first.
21:02karolherbst: because the offsets inside the vbios assume it's adjacent
21:02karolherbst: like the BIT table pointers at the start of the vbios
21:03karolherbst: so they'd e.g point to 0x15324 or something, but that's in the middle of frts
21:04karolherbst: yeah, so your idea of doing a copy into host mem/vram would make it obsolete
21:04TimurTabi: Ok, I think I get it now
21:06TimurTabi: Ok, now I see what RM is doing. It's actually doing the same thing, just in a messier way (as usual)
21:06karolherbst: cool :)
21:07TimurTabi: I feel sorry for anyone who wants to port OpenRM to Nouveau.
21:07karolherbst: well I already found useful bits
21:07karolherbst: we have this messy workaround for runtime pm, because apparently we have to hack the root port to make it work
21:08karolherbst: and rm apparently has a list of root ports where some stuff works or not :(
21:08karolherbst: and that's already messy enough :D
21:10karolherbst: restarted chrome and now it works, cool
21:20karolherbst: mhhh weird.. somehow new chrome tabs are the only thing not really working.. that's odd
22:09karolherbst: chrome works now as well after diabling vulkan and uninstalling the nvidia userspace stuff :)
22:09TimurTabi: So I think I was wrong about FRTS earlier. Maybe there really is nothing between the two images.
22:09karolherbst: well.. there obviously is something, but what is it? :P
22:27airlied: TimurTabi: I started parsing the v2/3 'p' bit tables but I'm not see what I'd expect yet
22:28airlied: 0x70 in open gpu driver is 'p'
22:28karolherbst: sounds about right
22:29karolherbst:spent too much time on that stuff
22:30karolherbst: which nv50 GPUs can we actually reclock btw?
22:31karolherbst: not sure the android emulator is any fun on base clocks
22:32imirkin: karolherbst: out of the box, iirc g94+
22:32karolherbst: okay.. gt218 with ddr2?
22:32karolherbst: I think GDDR3 is unsupported
22:32imirkin: GT215 + GDDR5 is unsupported
22:33imirkin: everything else should be fine
22:33karolherbst: then gt200 + GDDR3 sounds better than passively cooled gt218/g94 with DDR2 :)
22:33imirkin: by which you mean you enjoy the jet noise the g200 produces?
22:34karolherbst: the android emulator was already slow on the gp108 :(
22:34karolherbst: not terrible, but not smooth either
22:34imirkin: in terms of perf, even those old gpu's could get pretty strong results
22:34imirkin: feature set is limited, but the stuff they do is plenty fast, within reason
22:34karolherbst: not those
22:35karolherbst: it's a gt 210 :)
22:35imirkin: yeah, that's not going to be breaking any speed records
22:35imirkin: but that was true the day it was released
22:35imirkin: and it hasn't sped up since :)
22:35karolherbst: 44 gflops on max :D
22:36karolherbst: but yeah.. gnome was already slow
22:36imirkin: well, at the time gnome was plenty fast
22:37karolherbst: I am sure it's fast enough on the gt200
22:37karolherbst: and the fans aren't terrible
22:37imirkin: but then someone decided that render to render everything "gpu-accelerated"
22:37karolherbst: actually.. they sound quite smooth
22:37imirkin: aka "gpu-decelerated"
22:37karolherbst: not "crackling" like really old fans sound sometimes
22:38karolherbst: GPU not detected :(
22:40karolherbst: that's not fair
22:47karolherbst: how the fuck does this make any sense
22:47karolherbst: sure.. it works on the non PEGP port
22:49karolherbst: [ 78.936091] nouveau 0000:05:00.0: fifo: CACHE_ERROR - ch 3 [Xwayland] subc 6 mthd 027c data 00000000
22:50karolherbst: imirkin: do you know what that was all about?
22:54karolherbst: mhh "[ 372.782705] nouveau 0000:05:00.0: gr: TRAP_MP_EXEC - TP 2 MP 2: 00000010 [INVALID_OPCODE] at 07e8e0 warp 1, opcode 000003c0 00000000"
22:55karolherbst: okay cool
22:55karolherbst: with the gt200 clocked to max everything is super smooth :)
22:55karolherbst: just wondering where those bugs are comming from
22:55karolherbst: I hope I didn't break anything
23:08karolherbst: I am glad that I see the same issues with and without my MT fixes
23:08karolherbst: but also not glad I see them at all
23:14imirkin: karolherbst: yeah, "random error that happens on nv50"
23:14imirkin: i think we do fifo ctx switching slightly wrong
23:14karolherbst: well.. it's seems that the gt200 is pretty unstable sadly :(
23:14karolherbst: wouldn't surprise me
23:21karolherbst: fun.. using TGSI the wrong rendering simply looks different
23:23karolherbst: yeah.. something is completely bonkers with that gpu