00:20fdobridge: <airlied> I'll take solve sync over D32S8 🙂
00:20fdobridge: <airlied> I managed to run CTS here over it and sampler minmax and didn't see any problems, so yes just code review
01:41fdobridge: <gfxstrand> I merged that one already. Forgot to code review it, though. 😳
01:43fdobridge: <airlied> didn't have much code in it
01:44fdobridge: <airlied> your first review meant I just removed most of it 🙂
05:04fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> Synchronization?
05:12fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> I wonder if this is still needed with latest `nvk/main`: https://gitlab.freedesktop.org/nouveau/mesa/-/merge_requests/196
05:44fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> Also I should try to upstream some of my more recent changes if possible 🐸
05:59fdobridge: <airlied> I think that should be fixed already
06:02fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> https://gitlab.freedesktop.org/nouveau/mesa/-/merge_requests/218 🐸
06:14fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> So GSP Talos should work on `nvk/main`, right?
06:15fdobridge: <airlied> not sure I haven't retested it in the last few weeks
06:21fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> You should retest it at a lower resolution so the FPS is higher and the races are going to be fully exposed
06:21fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> You should probably retest it at a lower resolution so the FPS is higher and the races are going to be fully exposed (edited)
06:22fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> You should probably retest it at a lower resolution so the FPS is higher and the races are going to be more visible (edited)
06:23fdobridge: <🐧sima🐧> someone's celebrating early
07:14fdobridge: <airlied> looks like talos runs for me still with nvk/main on top of gsp
07:16fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> So no extra changes are needed for GSP support on NVK?
07:16fdobridge: <![NVK Whacker] Echo (she) 🇱🇹>
07:16fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> You should also try SuperTuxKart's Vulkan renderer 🐸
07:17fdobridge: <airlied> depends on which gsp branch you are using still
07:18fdobridge: <airlied> I think only my branch can handle nvk so far
07:18fdobridge: <airlied> ben is reworking my patches for better integration into his stuff
07:19fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> Do you mean `nouveau-gsprm-nonstall`?
07:20fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> I tried that and I got display output errors, missing temperature sensor and broken NVK (`fail ttm_validate` errors)
07:20fdobridge: <airlied> yup that branch
07:20fdobridge: <airlied> yeah no surprises in any of those
07:21fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> Interesting
07:21fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> Maybe TU116/TU117 paths are broken?
07:21fdobridge: <airlied> no more likely gsp display and temp sensor paths are bokrn
07:21fdobridge: <airlied> broken
07:22fdobridge: <airlied> looks like dota2 starts and seems to mostly render correct, dies eventually
07:23fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> I'm surprised that Talos works now 🐸
07:23fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> I'm surprised that Talos works 🐸 (edited)
07:23fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> I'm surprised that Talos even works 🐸 (edited)
07:26fdobridge: <airlied> I'm running on "nouvelles/airlied-testing-mctestface" but that shouldn't be much different to the nonstall branch
07:27fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> But that has the new uAPI which I can't really use (because it requires re-compiling the whole kernel)
07:29fdobridge: <airlied> doesn't nouveau-gsprm-nonstall not require recompiling the whole kernel?
07:30fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> I did a massive hack and made `nouveau` build as a DKMS module called `nouveau_gsp` 😅
07:31fdobridge: <airlied> okay might cause some of the ttm related problems
07:31fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> So `nouveau-gsp-nonstall` actually modifies stuff outside of `nouveau`?
07:32fdobridge: <airlied> no but the kernel you are running against might not be api/abi compatible
07:32fdobridge: <airlied> it's definitely not a tested combo, so I'd say it's likely the cause of some of the problems
07:32fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> I was running a 6.2 Arch Linux kernel (maybe 6.2.9?)
07:34fdobridge: <airlied> yeah as I said that's not an interesting case, and it's likely to explode
07:34fdobridge: <airlied> i.e. you get to keep both pieces
07:35fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> `vkcube` actually worked though (so maybe GSP is just in the very early phases?)
07:35fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> Ben's ada branch was even worse (vkcube just hanged I think)
07:36fdobridge: <airlied> please don't report any problems with a dkms thing, it's not interesting and wastes time
07:38fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> Maybe I'll try compiling the full kernel later (with some tricks like modprobe-db it might not actually be too bad)
07:38fdobridge: <![NVK Whacker] Echo (she) 🇱🇹>
07:38fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> And because I'm compiling the full kernel anyway I'll try your mctestface branch to see if that's better
07:47fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> SuperTuxKart (with Vulkan renderer) would be interesting case to test though (because I want to make sure that DKMS is the actual problem and not something else)
07:50fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> I'm currently trying to explode the driver with synchronization2 and memory_model tests 😅
07:55fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> I love these results /s
07:55fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> https://cdn.discordapp.com/attachments/1034184951790305330/1098154988246675466/Screenshot_20230419_105521.png
07:57fdobridge: <airlied> for proper timelines etc we probably do want to wait for the new uapi to be done
08:01fdobridge: <airlied> supertuxkart vulkan seems to race for me
08:04fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> What errors do you get in dmesg? I'm very curious
08:05fdobridge: <airlied> errors in dmesg for what?
08:07fdobridge: <airlied> the ttm validate errors you see are due to the dkms thing I'd guess, I'm also only running nvk in offload to an i915 display
08:08fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> I'm also using PRIME
08:08fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> SuperTuxKart
08:09fdobridge: <airlied> supertuxkart runs fine, nothing in dmesg at all
08:10fdobridge: <Esdras Tarsis> Just compile with localmodconfig
08:11fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> I thought you mentioned race condition
08:11fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> I thought you mentioned a race condition (edited)
08:12fdobridge: <airlied> there was one ages ago, I though the ttm validate you were seeing might be it, but it wasn't it was the dkms that you didn't menton
09:33fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> Temp sensor is kind of unfortunate because I can't know how hot the GPU is getting at the full clocks
12:24fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> What is this?: `kernel: nouveau 0000:01:00.0: vfn: intr0: 00021000`
12:27fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> Also ttm_validate is failing with -ENOMEM 🤔
12:28fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> Maybe there are some memory leaks in the GSP code and my 4 GB of VRAM isn't enough? (I'll recompile the full kernel just to be sure)
12:30fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> If this happens with the full kernel then there's DEFINITELY something wrong here (or maybe my local NVK fork has some problematic changes?)
12:34fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> What I do know is that NFS Most Wanted 2012 ran out of VRAM once with non-GSP kernel (which never happened on the blob driver)
13:53fdobridge: <gfxstrand> I said "may" and assumed that airlied would interpret it as per RFC 2119
13:59fdobridge: <🐧sima🐧> I thought that trick only works in uppercase 😜
14:02fdobridge: <gfxstrand> @airlied On the upside, @sima and I **may** have solved synchronization. (edited)
14:02fdobridge: <gfxstrand> There you go. Fixed the bug.
14:02fdobridge: <gfxstrand> Vulkan variant of RFC 2119, so bold
14:38fdobridge: <🐧sima🐧> perfect
14:39fdobridge: <Mohamexiety> `VK_EXT_REQUIREMENT_LEVELS`
14:41fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> I'm finally compiling a full kernel
14:47fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> `drivers/gpu/drm/nouveau/nvkm/engine/disp/outp.c:262:1: warning: ‘nvkm_outp_init_route’ defined but not used [-Wunused-function]` 🤔
14:48fdobridge: <gfxstrand> Has anyone tried DiRT 4 with NVK?
14:50fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> I don't have the game or the disk space for it
14:50fdobridge: <![NVK Whacker] Echo (she) 🇱🇹>
14:50fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> Why did you mention this game specifically?
14:50fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> I don't have the game (or likely the disk space for it)
14:50fdobridge: <![NVK Whacker] Echo (she) 🇱🇹>
14:50fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> Why did you mention this game specifically? (edited)
14:52fdobridge: <gfxstrand> @Mr Fall🐧 Would you be able to get me a hack to enable power management on this laptop?
14:52fdobridge: <karolherbst🐧🦀> uhm.... maybe?
14:52fdobridge: <gfxstrand> It happens to be in the collabora account. Don't worry about it.
14:52fdobridge: <gfxstrand> And it's one I've played with before
14:54fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> Is it pre-Maxwell v2?
14:54fdobridge: <karolherbst🐧🦀> @gfxstrand something like this, just adjusted for a proper kernel tree https://github.com/karolherbst/nouveau/commit/241342504559eb76bba49fe6e3bd906dd9ea32df
14:54fdobridge: <karolherbst🐧🦀> I think
15:12fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> The kernel is done cooking 🐸
15:12fdobridge: <![NVK Whacker] Echo (she) 🇱🇹>
15:12fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> I just need to package the GSP blobs separately and THEN we can test for explosions (it's a -rc1 kernel so it might actually explode)
15:21fdobridge: <gfxstrand> If I'm reading this right, on my branch (old one from skeggsb), reclocking is always enabled. 🤔
15:26fdobridge: <karolherbst🐧🦀> not for GM20x
15:26fdobridge: <karolherbst🐧🦀> it needs the `clk` func pointers
15:29fdobridge: <gfxstrand> ok
15:29fdobridge: <gfxstrand> Once my build completes, I'll do a take 2 at rebasing the patch
15:32avoidr: Ok, I believe I have found the cause of the freeze I had two weeks ago or so. mpv with vdpau. Not modesetting DDX as I first believed.
15:33avoidr: Is this known, or can I help here?
15:34avoidr: I mean the video output freezes. Everything seems to work correctly, but my display won't react to anything until I reboot.
15:57karolherbst: not much. I'll look into making the system at least more reliable so that errors like that won't freeze the system, maybe I'll also figure out what's wrong with vdpau at the same time
16:01avoidr: That's good, thanks.
16:02fdobridge: <karolherbst🐧🦀> @airlied how is the `bo_wait` stuff progressing with GSP? Because I'm more or less planning to rely on that also for OpenGL to solve those screen freeze problems
16:06fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> Blobs have been packaged and installed
17:35fdobridge: <gfxstrand> What kernel flag do I need to set? nouveau.NvFanless doesn't work.
17:40fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> `nouveau.config=NvFanless=1` is the accurate option
17:41fdobridge: <gfxstrand> Thanks!
17:46fdobridge: <gfxstrand> Ok, now to figure out why it's still only giving my 40 gears FPS. 🫤
17:47fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> gears? glxgears?
17:48fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> I wonder if you're trying to get reclocking stuff working on the weird Maxwell v2/Pascal gap
17:52fdobridge: <gfxstrand> 980m
17:53fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> So Maxwell v2 👀
17:53fdobridge: <![NVK Whacker] Echo (she) 🇱🇹>
17:53fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> I'm curious how you're going to break through the signed firmware barrier
17:54fdobridge: <Mohamexiety> yeah that shouldn't have reclocking from what the feature matrix says
17:54fdobridge: <gfxstrand> Uh... What? Ugh. @Mr Fall🐧 implied it should be possible but maybe badly?
17:56fdobridge: <gfxstrand> Where "badly" meant it might burn up if your fans don't work.
17:57fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> I'm getting https://www.youtube.com/watch?v=NxNUK3U73SI flashbacks
18:03fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> It's time to reboot into the experimental kernel (let's hope it boots)
18:03fdobridge: <gfxstrand> Yeah, it's doing exactly nothing. 😓
18:04fdobridge: <gfxstrand> Ugh...
18:10fdobridge: <gfxstrand> Looking at the matrix, reclocking doesn't look reliable unless you go all the way back to Kepler. 😓
18:11fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> Let's ACTUALLY reboot now 🐸
18:17fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> I booted the full-fat kernel and I STILL get the same `fail ttm_validate` error
18:23fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> @ airlied Could you try `nouveau-gsprm-nonstall` instead to see if that's not stable?
18:23fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> @ airlied Could you try `nouveau-gsprm-nonstall` instead to see if that's more unstable? (edited)
18:27fdobridge: <marysaka> You could do a ROP in the falcon secure rom technically and get back control in HS
18:27fdobridge: <marysaka> tho it's --x and I haven't tried to setup that on dGPU, only did that with tegra...
18:27fdobridge: <marysaka> really need to find some time to mess with it again :AkkoDerp:
18:29fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> I'll try the `mctestface` branch myself (luckily 6.3-rc1 didn't explode on my system) 😅
18:40fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> `drivers/gpu/drm/nouveau/nvif/outp.c:335:1: warning: the frame size of 2064 bytes is larger than 2048 bytes [-Wframe-larger-than=]` 🤔
18:48fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> That's not a good sign: `drivers/gpu/drm/nouveau/nouveau_dmem.c:681:43: error: ‘chunk’ undeclared (first use in this function)`
18:50fdobridge: <airlied> Oh I should push a local fixup patch
18:50fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> I did a quick hack to fix that error (we'll see if nouveau works with it)
19:13fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> Is this correct?
19:13fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> https://cdn.discordapp.com/attachments/1034184951790305330/1098325547022557388/nouveau-fix.patch
19:17fdobridge: <airlied> Yes
19:18fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> Unfortunately I get the same error 😢
19:19fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> https://pastebin.com/jaQKbQtk
19:20fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> Maybe you're using the new uAPI PR?
19:22fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> Let's try unmodified NVK then
19:26fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> (and it STILL breaks)
19:27fdobridge: <airlied> Maybe you are running out of vram then
19:27fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> Then how does the game work without GSP?
19:28fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> (I have 4 GB of total VRAM)
19:28fdobridge: <airlied> Gsp uses a bunch of vram
19:29fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> Maybe I should try the nvidia-open driver with GSP and see if I still get issues 🐸
19:30fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> If that works fine then something's really wrong in nouveau/NVK code (or the GSP is just a bit too out-of-date on the nouveau driver)
19:31fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> So then I should try nvidia-open 530.30.02 to kill 2 birds with one stone
19:33fdobridge: <karolherbst🐧🦀> uhh.. how nasty of GSP
19:34fdobridge: <karolherbst🐧🦀> Can we do anything about it?
19:35fdobridge: <airlied> Lots of things are really wrong, it's a development project, we don't even allocate memory properly until we have new APIs sorted out
19:37fdobridge: <karolherbst🐧🦀> the entire point was more like "let's start a vulkan driver before we have the new uAPI, because people become impatient, and then things got out of hand 😛
19:37fdobridge: <karolherbst🐧🦀> initially there was no plan doing actual graphics until the new uAPI is finalized
19:44fdobridge: <airlied> No it needs vram areas to run gsp code and data in
19:46fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> Let's see how `nvidia-open` handles my games 🐸
19:49fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> Rebooting now to test `nvidia-open` 🐸
19:56fdobridge: <karolherbst🐧🦀> oh sure, but is it actually that much? or just a couple of MB?
19:56fdobridge: <karolherbst🐧🦀> well..if a couple of MBs are like 200 then that's still.. kinda okayish? dunno
19:56fdobridge: <karolherbst🐧🦀> do we know how much GSP alone uses?
19:57fdobridge: <airlied> Yeah I'd say 200 is closer than a couple, but I haven't any real numbers near me, would need to add some printks
20:04fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> `nvidia-open` works fine
20:14fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> And the VRAM usage is fine
20:14fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> https://cdn.discordapp.com/attachments/1034184951790305330/1098340771951820820/Screenshot_20230419_231348.png
20:15fdobridge: <karolherbst🐧🦀> try a GPU with more VRAM 😄
20:16fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> How do I upgrade my laptop GPU? 😅
20:17fdobridge: <karolherbst🐧🦀> download more VRAM
20:19fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> But I have enough VRAM to run older games at 4K 😅
20:20fdobridge: <karolherbst🐧🦀> so there is this weird bug I've hit with the nvidia driver
20:20fdobridge: <karolherbst🐧🦀> they kinda pre allocate a bunch of VRAM flat for each application
20:20fdobridge: <karolherbst🐧🦀> and that depends on the size of total VRAM
20:20fdobridge: <karolherbst🐧🦀> anyway, the more VRAM you have, the more likely you run into out of memory situations with 32 bit games 🙃
20:21fdobridge: <karolherbst🐧🦀> it's also CPU mapped memory, that's why it causes issues in the first place
20:21fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> I'm actually playing a 32-bit game
20:21fdobridge: <karolherbst🐧🦀> yeah, but I'm talking 24GB of VRAM
20:21fdobridge: <karolherbst🐧🦀> or 48GB even
20:22fdobridge: <karolherbst🐧🦀> or maybe they fixed it in the meantime 😄 point is, the more VRAM you have, the higher the memory consumption
20:33fdobridge: <![NVK Whacker] Echo (she) 🇱🇹> But anyway (since I'm running nvidia-open now), I should finally do some overdue things 🐸