<HdkR>
`[CRTC:40:crtc-1] vblank wait timed out` That line seems to imply it
<ventYl>
so, it is an issue of rk3399 alternate mode DP driver?
<HdkR>
Hard to tell until someone digs in to it :)
<ventYl>
I can try to dig into it, but I have to know where to start :)
kaspter has quit [Ping timeout: 260 seconds]
kaspter has joined #panfrost
camus1 has joined #panfrost
kaspter has quit [Ping timeout: 246 seconds]
camus1 is now known as kaspter
<icecream95>
ventYl: If you update mesa to 20.1 (it should be in the stable repos by now), does the problem still happen?
<icecream95>
It looks like you already tried mesa-git, so that probably wouldn't help
<ventYl>
icecream95: i have tried mesa-git and just 10 minutes ago rebooted with latest stable
yann|work has joined #panfrost
raster has joined #panfrost
camus1 has joined #panfrost
kaspter has quit [Ping timeout: 246 seconds]
camus1 is now known as kaspter
stikonas has joined #panfrost
raster has quit [Quit: Gettin' stinky!]
<warpme_>
guys: fyi: I give run for g31 on mythv & current mesa-master on amlogic sm1 tvbox. Impressive!. GLES based GUI is fully usable and even hw accelerated video playback partially works (it fails after few sec like in this screen recording: http://warped.inet2.org/sample-panfrost.mov). Fantastic work!
<HdkR>
Nice
<warpme_>
for url delete parenthesis and dot at end of url....
<HdkR>
irssi is cool enough to do that for me already
<robmur01>
alyssa: according to the datasheet the industrial-grade "K" variant is also "W version", so I suppose it could be knocking around in recent digital signage boxes etc.
<robmur01>
Bootlin are pushing clock patches upstream, so I'd assume they have a customer using it
<robmur01>
ventYl: given that the upstream RK3399 Type C phy driver only works on Chromebooks with legacy extcon crap, and Manjaro seem to be using dirty hacks to get DP altmode semi-working, I'd point the finger there ;)
<icecream95>
warpme_: It looks like I forgot to make sure 7ef648c0f309 works with Bifrost. You can try reverting it, and it will probably fix things
<daniels>
hmm, is mythtv not using dmabuf then?
<tomeu>
robmur01: bootlin? I thought it was collabora
<tomeu>
iirc for a customer who uses it in their custom boards
icecream95 has quit [Quit: leaving]
<mmind00>
robmur01: actually it looks like cros-ec now also somehow can talk to the typc-c framework of the kernel - see recent cros-ec platform patches ... so I guess it's "just" waiting on someone finding the time to dive into this ;-)
<robmur01>
tomeu: oops, my bad - apparently my mental model for which names work at which consultancy firms is a bit out of date :)
<robmur01>
mmind00: I did have a look, but as far as I could tell seemed like it was just adding a layer to expose status information rather than reworking any of the fundamental control stuff to the new API :(
<mmind00>
robmur01: that's what I was trying to say ... needs someone to do the conversion to type-c framework for the rk3399 typc-c driver
<mmind00>
I remember seeing a mail about that flying through my inbox some time ago, but not sure what became of it
<warpme_>
icecream95: indeed reverting 7ef648c0f309 makes OpenGL playback with software decode working well. For playback with hw decode (drmprime with gles_dmabuf_export from video decoder to gles surface) playback has following distortions: http://warped.inet2.org/bifrost-g31-hw-decode.mov As exactly the same sw stack plays ok with hw decode on lima/midgard - issue seems to be bifrost specific. Anyway: fantastic work with
<warpme_>
indeed - progress i see with foss support for this soc is incredible!. still some rough edges but progress is really nice! i'm very surprised how quickly you team was able to provide really working bifrost!!!
<tomeu>
it was mostly alyssa :)
* alyssa
blush
<alyssa>
warpme_: glad to hear :)
<alyssa>
not sure why that commit would break on bifrost
<alyssa>
---Although the video is telling yes :p
<alyssa>
fafc305600cb8d54ce18494e2b0d961d694b63c8 is buggy due to midgard_bo/bifrost_bo
<alyssa>
though tbh those fields should be merged anyway.
<alyssa>
oh, the updates only run in the midgard branch.
<warpme_>
alyssa: million thx for this. i'll test this as soon as possible (now looking on no audio on amlogic sm1 soc). will report so stay tuned :p
<alyssa>
ack
<alyssa>
tomeu: If I'm OoO when that's tested (or it needs further work/bugfixes), please make sure that gets landed. Thank you :-)
<tomeu>
np!
<alyssa>
Interesting also that amlogic sm1 doesn't seem to have hte usual stack of amlogic issues
<alyssa>
maybe that *only* affects the g52 chips and the kodi bug is something much more benign
<alyssa>
(that would be good? since then at least we can get the g31 stuff working)
<alyssa>
warpme_: those artefacts for hw decode are odd alright
<robmur01>
anyone know whether "drm/panfrost: Make sure MMU context lifetime is not bound to panfrost_priv" is still relevant or superseded by something else?
<warpme_>
alysa: re: "maybe that *only* affects the g52 chips and the kodi bug is something much more benign" - pls note Kodi and mythtv are using different rendering method: KODI goes with drm planes where video decoder output is mixed with UI & OSD on crct controller (handled by kernel DRM subsystem). mythtv uses GL engine and GL engine is mixing UI surface and GLES video surface imported with dmabuf. In this contest mythtv
<warpme_>
is more demanding from GL engine as mythtv provides multiple surfaces and GL is expected to transform size/pixel format/deinterlace via GLSL and provide common mixed output format to single DRM plane.
<alyssa>
warpme_: Yeah, the Kodi issue is a text rendering goofiness :)
<ventYl>
robmur01: i would expect that device disconnect / reconnect and/or reboot would reset the status, but it seems, that it breaks and fixes itself randomly
<robmur01>
ventYl: indeed that should be expected if the DP phy, type C phy, and type C port controller drivers were all properly working together and in sync; I'm suggesting that likely isn't the case ;)
<ventYl>
robmur01: is there any, slightest, chance to get hands on some kind of datasheet?
Elpaulo has quit [Quit: Elpaulo]
<alyssa>
But... Kodi text is rendering fine from this trace.
<alyssa>
Is it possible newer Kodi changed something?
<robmur01>
ventYl: the publicly-released RK3399 TRM doesn't cover the relevant parts, but there are leaked copies of the full thing floating around if you look hard enough; for FUSB302 a full datasheet is readily available
<robmur01>
either way this doesn't really have much to do with the GPU, since it's all downstream of the display controller
<robmur01>
the hardware aspects of the individual drivers are there already; the main thing is making them all work together via the kernel typec APIs
<ventYl>
so, the error will be more like purely of SW nature, how these drivers are wired within linux kernel?
<ventYl>
this way it is possible i'll be able to fix it
<alyssa>
maybe blaming mipmapping
macc24 has quit [Quit: WeeChat 2.8]
<chewitt>
Kodi has seen huge changes for video rendering and such in recent times (GBM/V4L2) but not the GUI (GLES) bits
<alyssa>
chewitt: mm
<alyssa>
unless debian kodi is that old
<chewitt>
Debian Kodi has a reputation for being old, but it would have to be spectacularly old
<chewitt>
and the same code runs on Midgard fine, so the core of panfrost should be good
<alyssa>
right
<alyssa>
I don't suppose it's possible to get an apitrace from le?
<alyssa>
mipmapping does not appear to be used by kodi for anything
<chewitt>
sorry.. have been busy tweeting (must be a twit)
<chewitt>
having a w00t! moment :)
agrisis has joined #panfrost
<chewitt>
G52 also runs with the same change, but still shows the 'faults' and there are some visual glitches (flashes of triangles, like lightning) that I don't see on G31
<chewitt>
but, progress!
guillaume_g has quit [Quit: Konversation terminated!]
<agrisis>
any updates on if/how people got this kernel booted on odroid go advance?
<agrisis>
last I tried I got no video and I don't think it completed the boot
Elpaulo has joined #panfrost
davidlt has quit [Ping timeout: 260 seconds]
raster has joined #panfrost
macc24_ has joined #panfrost
macc24 has quit [Disconnected by services]
macc24_ is now known as macc24
macc24 has quit [Quit: WeeChat 2.8]
macc24 has joined #panfrost
<alyssa>
chewitt: uh, huh, which patch did it?
<alyssa>
As for G52... that *could* be aml related
<alyssa>
Might also be tiler configuration related
<alyssa>
again, I'm not excited to debug G52 until aml stuff stabilizes
raster has quit [Quit: Gettin' stinky!]
warpme_ has quit [Quit: Connection closed for inactivity]