01:32fdobridge: <gfxstrand> I like that my CTS runs are back down to 50 minutes.
02:25fdobridge: <gfxstrand> Okay, let's run Maxwell and see if my machine survives.
02:34fdobridge: <gfxstrand> Yeah, it's not going well...
02:34fdobridge: <gfxstrand> ```
02:34fdobridge: <gfxstrand> [209201.708804] Call Trace:
02:34fdobridge: <gfxstrand> [209201.708808] <IRQ>
02:34fdobridge: <gfxstrand> [209201.708811] ? nvkm_engn_cgrp_get+0x80/0x90 [nouveau]
02:34fdobridge: <gfxstrand> [209201.709204] ? __warn+0x81/0x130
02:34fdobridge: <gfxstrand> [209201.709213] ? nvkm_engn_cgrp_get+0x80/0x90 [nouveau]
02:34fdobridge: <gfxstrand> [209201.709606] ? report_bug+0x171/0x1a0
02:34fdobridge: <gfxstrand> [209201.709614] ? handle_bug+0x3c/0x80
02:34fdobridge: <gfxstrand> [209201.709618] ? exc_invalid_op+0x17/0x70
02:34fdobridge: <gfxstrand> [209201.709621] ? asm_exc_invalid_op+0x1a/0x20
02:34fdobridge: <gfxstrand> [209201.709630] ? nvkm_engn_cgrp_get+0x80/0x90 [nouveau]
02:34fdobridge: <gfxstrand> [209201.710020] ? nvkm_engn_cgrp_get+0x75/0x90 [nouveau]
02:34fdobridge: <gfxstrand> [209201.710409] nvkm_runl_rc_engn+0x3f/0xc0 [nouveau]
02:35fdobridge: <gfxstrand> [209201.710798] gf100_fifo_mmu_fault_recover+0x32c/0x340 [nouveau]
02:35fdobridge: <gfxstrand> [209201.711187] gm107_fifo_intr_mmu_fault_unit+0xf2/0x120 [nouveau]
02:35fdobridge: <gfxstrand> [209201.711574] gf100_fifo_intr_mmu_fault+0x51/0xb0 [nouveau]
02:35fdobridge: <gfxstrand> [209201.711961] gk104_fifo_intr+0x2b8/0x3a0 [nouveau]
02:35fdobridge: <gfxstrand> [209201.712345] nvkm_intr+0x129/0x240 [nouveau]
02:35fdobridge: <gfxstrand> [209201.712642] __handle_irq_event_percpu+0x47/0x1a0
02:35fdobridge: <gfxstrand> [209201.712651] handle_irq_event+0x38/0x80
02:35fdobridge: <gfxstrand> [209201.712657] handle_edge_irq+0x8b/0x230
02:35fdobridge: <gfxstrand> [209201.712663] __common_interrupt+0x3c/0xa0
02:35fdobridge: <gfxstrand> [209201.712669] common_interrupt+0x81/0xa0
02:35fdobridge: <gfxstrand> [209201.712674] </IRQ>
02:35fdobridge: <gfxstrand> [209201.712675] <TASK>
02:35fdobridge: <gfxstrand> [209201.712677] asm_common_interrupt+0x26/0x40
02:35fdobridge: <gfxstrand> ```
02:38fdobridge: <airlied> Looks like a racy mmu fault recover, just fix the mmu fault :-p
02:39fdobridge: <esdrastarsis> I saw the Rust project that you are helping to develop, why did you choose Rust?
02:41fdobridge: <airlied> Just experimenting with things, was there another language to write rust in?
02:45fdobridge: <airlied> You can write kernel drivers in C or rust and we know how to write them in C, so it doesn't leave a lot of scope to pick another language
02:49fdobridge: <esdrastarsis> I asked because I wanted to know if Rust's security features help a lot in developing DRM drivers
02:50fdobridge: <airlied> Not really just help the same as in any driver, better lifetime and lack of use after frees
02:52fdobridge: <airlied> Don't think it helps DRM drivers in any special way
03:17fdobridge: <gfxstrand> Yeah, I started again andmore of the same. I'm going to give up for now.
03:28fdobridge: <airlied> at least we can probably avoid blaming gsp for that 😛
03:28fdobridge: <airlied> is there some more info prior to that call trace?
03:32fdobridge: <gfxstrand> Probably but the machine died so IDK
03:33fdobridge: <gfxstrand> If I get more info, is someone going to care? Should I file a bug?
03:33fdobridge: <gfxstrand> It would help a lot with improving Maxwell if it could survive a CTS run.
03:42fdobridge: <airlied> I'd file a bug if you can get a bit more info, but fixing it probably involves finding what the crash is, NULL ptr or whatever and trying to see what is racing
05:42fdobridge: <gfxstrand> @asdqueerfromeu This should fix your dmesg problem: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26441
05:42fdobridge: <gfxstrand> @asdqueerfromeu Did you ever file a bug about that? If so, could you either paste it here or tag the MR with it?
05:43fdobridge: <gfxstrand> If not, that's fine. We'll just consider this a code clean-up
07:19fdobridge: <!DodoNVK (she) 🇱🇹> Do you mean the GR class error?
10:02simplymaster: That you should understand that I have these methods and routines, cause I designed and engineered those all , including the most efficient ones. If you did not understand the method I can elaborate, the equations underneath, but it's simple
10:16fdobridge: <!DodoNVK (she) 🇱🇹> NFS Most Wanted 2012 starts hitting 60 FPS at 1080p now with max settings and no AO/AA (I'm not sure if it's because of MR !26408 though)
10:16fdobridge: <!DodoNVK (she) 🇱🇹> https://cdn.discordapp.com/attachments/1034184951790305330/1180090045928058943/Screenshot_20231201_121318.png?ex=657c277f&is=6569b27f&hm=155846c4a2f74e63426b7259a0f191bcad1dde2134fec2f4124f2e3784b725f6&
10:19fdobridge: <!DodoNVK (she) 🇱🇹> The game was hitting 7 FPS at 800x600 lowest settings back in April without GSP (so that's quite the improvement): https://discord.com/channels/1033216351990456371/1034184951790305330/1091715986215948381
10:44simplymaster: Lance Armstrong went to have a brain surgery and also genitals to control and eliminate the spread cancer spread, and 7times later Tour the France winner then merely best in us asked why should he trust his life to surgeon? Shapiro MD replied, Lance what you are in cycling, I am several times better in brain and cancer surgery :), and I am not the under dog either, what you are I am several times more.
10:46simplymaster: The one who designs hardware circuits and all your algorithms is not the weak outsider.
10:59simplymaster: It's not like you beat me.
11:00fdobridge: <!DodoNVK (she) 🇱🇹> The weird graphical flashes are gone (but the game still stutters/freezes a lot so GPL will definitely be important) :triangle_nvk:
11:00fdobridge: <!DodoNVK (she) 🇱🇹> https://cdn.discordapp.com/attachments/1034184951790305330/1180101021276569630/Screenshot_20231201_125732.png?ex=657c31b8&is=6569bcb8&hm=1e31e2b7409de0db4c3be8e5e649a4c030b679c6a01b9741545dc8b2101eb958&
11:01fdobridge: <!DodoNVK (she) 🇱🇹> I'm not sure about the weird text flickering though at the bottom (I'll have to check RADV too)
11:01simplymaster: Let's just fix the performance and maintain the SW chiplets.
11:03simplymaster: If you contact me, with a wish to do business or with a possible way I can offer you service of mental kind and technical, I do not refuse perhaps
11:04Ermine: karolherbst: requesting banhammer
11:08simplymaster: We are also marked as number one in the world towards most precious semi-metals not phosphor , our country is extremely rich
11:09simplymaster: Our men are very strong too
11:14fdobridge: <!DodoNVK (she) 🇱🇹> And my system kind of did a GPU reset (screen flashing a few times) before being weirdly dead (no SSH but I can still see disk activity from the LED)
11:14simplymaster: Why I should quit having relations with my own neighbor? Our own people, brotherhood Russians, they handle things so well!
11:16simplymaster: I understand Muslims, but Russians are so good
11:24simplymaster: We were born together we age together it's our close ethnic nation, we age together we love us both.
11:25simplymaster: We do not talk about Africans ukranian and Russian people is our own
11:28simplymaster: What eu and NATO requests or does is not so possible to be maintained, cause Russians should be our mutual ally.
11:28simplymaster: Cause they are in the same ethnic group
11:29simplymaster: I love those people
11:32simplymaster: That we should remember that Russia is a white race who is minority in world, they are very beautiful too like others or even more though, but that's minor it's bad that we do wars in between us
14:05fdobridge: <gfxstrand> Yes
14:06fdobridge: <gfxstrand> Yes
14:07fdobridge: <!DodoNVK (she) 🇱🇹> I just submitted a comment about that
14:07fdobridge: <gfxstrand> Okay, cool
14:07fdobridge: <!DodoNVK (she) 🇱🇹> I'm not sure if that's nouveau-related though (I think @tiredchiku did encounter this too)
14:07fdobridge: <gfxstrand> That gives us a breadcrumb at least
14:08fdobridge: <Sid> I did
14:25fdobridge: <!DodoNVK (she) 🇱🇹> To clarify things, that NVK copy engine MR fixes the weird GR class errors I had on NVK (the nouveau OpenGL driver obviously has no change)
14:46Tom^: regarding dgpu laptops, power management like the D3 one nvidia blob does. power it off when nothing is using it, is that something thats either being worked on or is it already working? or is that bits in the GSP thingie
14:48fdobridge: <!DodoNVK (she) 🇱🇹> D3 support already exists (I was able to make the GPU go into D3cold mode)
14:48fdobridge: <gfxstrand> Cool. Yeah, I never really trusted the 2D engine. With that MR, we don't use it at all anymore.
14:49Tom^: oh nice, just gotta wait for 6.7 to land and il fire the testing up then on this 3060 heh
14:49fdobridge: <!DodoNVK (she) 🇱🇹> What's wrong with it?
14:54fdobridge: <gfxstrand> It's got piles of restrictions we don't fully understand. It's also probably slower than the copy engine. We mostly used it because FillBuffer was one of the first things we ever wired up and we didn't know the copy engine could do that.
14:55fdobridge: <gfxstrand> Yesterday, I took one look at a blob trace and it was obvious. 🤦🏻♀️
16:45fdobridge: <gfxstrand> I like seeing all these 60 FPS numbers. 😁
16:45fdobridge: <gfxstrand> I'm legit looking forward to the next Phoronix benchmark run. I expect things will look WAY better than the pre-GSP run he did earlier this year.
17:03fdobridge: <!DodoNVK (she) 🇱🇹> This was with an unlocked framerate (so it can go higher and actually up to 600 FPS)
19:12fdobridge: <gfxstrand> Nice!
19:13fdobridge: <gfxstrand> @karolherbst is there a flag I can set on ldl/stl which will make them noop on OOB instead of faulting?
19:13fdobridge: <karolherbst🐧🦀> yes and no
19:14fdobridge: <karolherbst🐧🦀> ehh wait LDL?
19:14fdobridge: <karolherbst🐧🦀> mhhh
19:14fdobridge: <karolherbst🐧🦀> not sure
21:08fdobridge: <gfxstrand> Ugh... I have no idea how instruction deps are supposed to work with these dumb BMOV instructions
21:10fdobridge: <gfxstrand> It's clear that the barrier file is internally scoreboarded. That's fine.
21:11fdobridge: <gfxstrand> But some barrier instructions require `.yld`, most seem to ignore (or require 0) for other dep stuff.
21:12fdobridge: <gfxstrand> But `bmov R B` appears to be a fixed-latency instruction?!?
21:15fdobridge: <gfxstrand> Requiring `.yld` makes sense if it's internally scoreboarded. Yield is really the only way the HW has to bail out if it's waiting on something.
21:15fdobridge: <gfxstrand> Especially requiring `.yld` on `bsync` is pretty much a "no duh"
21:15fdobridge: <gfxstrand> I'm less sure about `bmov` and `break`
22:22fdobridge: <karolherbst🐧🦀> nvidia puts `.yld` on almost everything
22:22fdobridge: <karolherbst🐧🦀> I think it means something "this block/thread is free to be scheduled away from" or something
22:23fdobridge: <karolherbst🐧🦀> and if you don't put it, it can't or something
22:25fdobridge: <karolherbst🐧🦀> yield is also required to ensure forward progress
22:28fdobridge: <karolherbst🐧🦀> nothing is internally scoreboarded
22:29fdobridge: <karolherbst🐧🦀> the main difference is just if they are fixed or variable latency
22:31fdobridge: <karolherbst🐧🦀> the yield flag really just means that control can be moved over to a different set of threads
22:31fdobridge: <karolherbst🐧🦀> which makes sense if the thread blocks on bsync