00:35 fdobridge: <a​irlied> @rinlovesyou I think setting the precedent is what @gfxstrand will be doing 🙂
00:36 fdobridge: <r​inlovesyou> :salute:
00:36 fdobridge: <g​fxstrand> Setting precedent is basically my job. 😂
00:37 fdobridge: <a​irlied> @gfxstrand https://patchwork.freedesktop.org/patch/588344/ can you add that in your next few CTS runs, and see do you ever see the NULL ptr oops
00:37 fdobridge: <r​inlovesyou> strictly talking about converting existing code to rust i think even I will be able to contribute something :Hehe:
00:37 fdobridge: <g​fxstrand> Sure. Let my pull it quick
00:38 fdobridge: <S​id> ~~btw faith I sent you a connection req or however it's called on linkedin~~
00:38 fdobridge: <a​irlied> finding memory barrier bugs is not the fun experience everyone makes it out to be 😛
00:39 fdobridge: <S​id> just so you know it's me 😅
00:41 fdobridge: <a​irlied> also not 100% confident in my analysis yet, so hopefully dakr can't validate my sanity
00:44 fdobridge: <g​fxstrand> Accepted, I think. 😅
00:45 fdobridge: <S​id> yup
00:45 fdobridge: <S​id> :saigeheart:
00:45 fdobridge: <g​fxstrand> 👍🏻
00:57 fdobridge: <g​fxstrand> Built and I'm CTSing now. Of course, who knows when I'll actually trigger it. I'll let you know if I see something again.
01:01 fdobridge: <a​irlied> I had it reproduce at least once nearly every CTS run on the kernel I'm on now, so hopefully I can spot it
01:01 fdobridge: <a​irlied> I changed some config options a few days ago, and it went away completely, had to work out how to get back to the broken config 😛
01:14 fdobridge: <g​fxstrand> Woof
04:31 fdobridge: <r​inlovesyou> meow
04:44 fdobridge: <S​id> tcp - transfem communication protocol
06:14 fdobridge: <r​inlovesyou> what does that make UDP?
06:56 fdobridge: <a​huillet> FWIW, it seems I was wrong and GSP actually does implement backlight logic, so NV0073 should be working. it would likely make sense to debug why it's not rather than writes the registers directly
06:56 fdobridge: <a​huillet> (that's what I get for trying to help by reading code I don't know...)
07:01 ad__: Lyude: so trying to help, posted a first patch for ada lovelace eDP backlight not working. (forgot a printk inside :) ) . Waiting for feedbacks and ready for v2 as needed.
07:04 fdobridge: <a​huillet> without your patch, is this going through GSP? are you getting an error of any kind?
07:08 airlied: should look at how openrm does it, whether it uses displayport or gsp
07:18 ad__: if you want to go for the openrm way, looks like much more job has to be done
07:19 fdobridge: <a​huillet> I think Ben had already hooked up the logic to call GSP with NV0073?
07:19 ad__: while your "nouveau" way for eDP (drm layer) with minimal adjustments works great
07:20 ad__: https://lists.freedesktop.org/archives/nouveau/2024-April/044376.html
07:21 ad__: so my laptop now is ok. Studying in nouveau logic looks very complex to me, and cannot access eDP dpdc specs, but happy to help you as needed.
07:24 yusisamerican: ad__: Sorry for butting in but, how do you know that the max_brightness = 2048 magic isnt just something specific to your device?
07:27 ad__: yusisamerican: 496 is the max (<< 4). I don't know, becouse cannot access embedded display port DPDC spec pdf, that seems to be at payment, sent initial patch te get feedbacks from you.
07:27 ad__: *4096
07:28 ad__: at least, since _ADA switch/case was missing, there is something. But collecting feedbacks, otherwise the amount of things to study is huge.
07:29 ad__: Btw, i suspect an issue in the way drm helper calculate the max brightness.
07:37 yusisamerican: So your saying that you guess the spec's maximum value is 4096? Then why use 2048, half the value? You also are missing out on a lot of the logic in nv50_backlight_init, did something fail? Also I am pretty sure you *really* want to use RMAPI, looking at it at first glance it has a verrrrry nice system for handling this.
07:39 yusisamerican: ctrl0073specific.h's BACKLIGHT_BRIGHTNESS has a neat percentage mode you can use to get and set your brightness, max brightness is just 100%. It really looks like a neater way of doing things. See nvKmsGetBacklight/nvKmsSetBacklight.
07:41 yusisamerican: sorry for the text wall...
08:01 yusisamerican: ad__: ping just incase
08:01 ad__: yusisamerican: well, backlight was not working, so i would appreciate any contribute that is "better of nothing". Where do you read that "i guess" ? It's 256 << 4 so 4096 from kcalc. As i said, i am happy to help, but i am total noob on nouveau and extremly complex for me.
08:01 yusisamerican: Oh I see
08:02 ad__: using rm apis would probably result in much more code, so i have seen your "nouveau" way was near to work, and started from there
08:03 ad__: initially, i looked in nv50_backlight_init
08:03 ad__: issue there was if (nvif_outp_bl_get(&nv_encoder->outp) < 0 ||
08:03 ad__: outp was 0
08:04 yusisamerican: oh
08:04 ad__: so could not use that function as is, but with minimal changes it worked
08:04 yusisamerican: you know why outp was 0?
08:04 ad__: anyway, i can reimplement the fix, hope to have time to study in nouveu, actually i work in zephyros :)
08:05 ad__: no, i can investigate forther, sure
08:05 ad__: *further
08:06 ad__: bu i suspect the whole nvif stuff is not working from this chip
08:06 ad__: also acpi backlight was not working at all, and those pass through nvif ioctl (seems)
08:07 ad__: btw, this ADA LOVELACE chip is the "mobile" variant, may be a bit different form other ADA LOVELACE, but at least there is an initial patch wirth working backlight that should not harm
08:08 fdobridge: <a​huillet> ad__: ping
08:08 ad__: becouse for other laptom., acpi_backlight=vendor may work, (not on this)
08:09 ad__: sorry for typos, switching between job and nouveau :)
08:24 yusisamerican: ad__: The patch you sent seems at least to me as something that might not work on all systems. It might be better to investigate the cause of outp being 0 and your brightness maximum being completely off.
08:28 yusisamerican: In my opinion rmapi would be best, especially as the logic can pretty much be copied from open-gpu-kernel-modules. I would do it myself but I dont have backlight supporting nvidia hw.
08:35 ad__: yusisamerican: well, i can adventure on that, only issue is that seems not easy to call such apis from that nouveau-backlight.c point, at least i see much more code to be added. But i can try, sure. Btw, apart for that mistery (4096, 256 << 4) that i cannot clarify without eDP spec, it works great.
08:38 fdobridge: <a​huillet> https://cgit.freedesktop.org/drm/drm-misc/tree/drivers/gpu/drm/nouveau/nvkm/engine/disp/r535.c#n263
08:38 ahuillet: ad__ : the code is already there AFAICT (not sure you're reading me?)
08:41 ad__: ahuillet: i have seen you joined 10 minurtes ago, but not your msg. Btw, good. Tonight i will try that way, happy to help :)
08:41 ahuillet: missing the "fdobridge" stuff maybe?
08:41 ahuillet: <fdobridge> <a​huillet> https://cgit.freedesktop.org/drm/drm-misc/tree/drivers/gpu/drm/nouveau/nvkm/engine/disp/r535.c#n263
08:41 ad__: ok, great, thanks a lot
08:42 ahuillet: I have no idea how to get that called, but it seems like it would do what you want, so maybe you should try that :)
08:42 yusisamerican: Does it have to be plugged into backlight_ops?
08:43 ad__: that r535 looks to have some issue with this chipset, btw. In the case of acpi fallback, it was not working and i had to #undef CONFIG_ACPI inside that r535.c to have backlight up, otherwise, black screen
08:44 ad__: like "probing" was disabling backlight, but that may be a efi fw issue of this laptop
08:44 ad__: will try that way, this evening :)
11:19 fdobridge: <!​DodoNVK (she) 🇱🇹> I actually had to add 2 Rust depnedencies
11:19 fdobridge: <!​DodoNVK (she) 🇱🇹> *dependencies
15:38 fdobridge: <g​fxstrand> Why is the OpenGL CTS looking for WGL?!?
15:42 fdobridge: <g​fxstrand> Never mind. It's normal. I just don't know how to read GL CTS logs
15:43 fdobridge: <g​fxstrand> Let's see if we can pass the 4.6 CTS
15:44 fdobridge: <g​fxstrand> (My two fixes have been merged so I can CTS for realz now)
15:58 fdobridge: <!​DodoNVK (she) 🇱🇹> `skip("WGL is not supported on this system.\n");` (the Wine test way)
16:05 fdobridge: <g​fxstrand> Oh, it was even dumber than that. It's supposed to print a warning. I was reading the first thing and not the thing 3 lines later that told me my output directory was missing. :blobcatnotlikethis:
19:06 ad__: ok, backlight time now :)
20:35 juri_: karolherbst: is there a specific board i should acquire that is easiest to get on-gpu-compute going on?
21:21 fdobridge: <g​fxstrand> Finally working through the review backlog. Sorry to everyone who's had stuff blocked in me for like a month. 😢
21:21 fdobridge: <g​fxstrand> Especially @mhenning
22:00 Lyude: ad__: will take a look first thing tomorrow!