Wizzup | freemangordon: nice, opening pdf from email just worked | 09:45 |
---|---|---|
freemangordon | not on my freshly upgraded d4 | 09:47 |
Wizzup | weird | 09:47 |
freemangordon | yeah, but I am not going to dig into that | 09:47 |
freemangordon | my concern is modesetting/glamor | 09:48 |
freemangordon | I think those are FUBAR | 09:48 |
freemangordon | the point is: it seems that PRESENT implementation in modesetting is more or less broken | 09:49 |
freemangordon | maybe I shall pull latest master and see it is is better | 09:49 |
freemangordon | what I see on n900 - for some reason 'older' frames are drawn 'after' newer | 09:50 |
freemangordon | TFP is broken as well | 09:50 |
freemangordon | CMA area is 16 MB - how much more is needed for MS/GLAMOR to work? | 09:51 |
freemangordon | hmm, I have to try some video playback | 09:51 |
freemangordon | but given that performance is 3/4th of that without xorg, I doubt it will look nice | 09:52 |
freemangordon | Wizzup: I built mesa with dri3 enabled, does it make sense to build it with dri2? | 09:54 |
freemangordon | ah, where is uvos, he knows a bit about hat | 09:54 |
freemangordon | *that | 09:54 |
freemangordon | tmlind: does glmark-es2-drm works for you on d4? | 09:56 |
Wizzup | freemangordon: which xorg is this? | 09:57 |
Wizzup | it is possible some of the modeset problems are kms problems | 09:58 |
freemangordon | lemme check | 09:58 |
Wizzup | esp. the frame stuff/trails you mention | 09:58 |
freemangordon | 5e3900904ddc27f3d5580ce3a07929469d82fb5e | 09:59 |
Wizzup | I do not know regarding dri2, probably not | 09:59 |
Wizzup | I would really suggest separating glamor and ms if we can in our evaluation | 10:00 |
Wizzup | modesetting I think we really want to use for the mode setting part | 10:00 |
Wizzup | there is no reason not to | 10:00 |
Wizzup | glamor if potentially a different story | 10:00 |
Wizzup | is* | 10:00 |
Wizzup | freemangordon: cannot check the commit hash atm but iirc tmlind had some fixes for xorg modeset and rotation | 10:01 |
Wizzup | if that is what you mean with present | 10:01 |
freemangordon | hmm, upstream has lots of fixes for glamor/ms | 10:02 |
Wizzup | I use modesetting on my laptop and it does not seem broken | 10:02 |
freemangordon | worths rtying latest master | 10:02 |
Wizzup | right | 10:02 |
Wizzup | we also patch our xorg but only with my patch iirc | 10:02 |
Wizzup | (xrexcorcd related) | 10:03 |
freemangordon | we are at https://gitlab.freedesktop.org/xorg/xserver/-/commits/master?search=Use+different+scroll+increment+than+libinput+in+inputtest+drv | 10:03 |
freemangordon | since that there is a pile of commits | 10:04 |
freemangordon | so I am going to try latest master | 10:04 |
Wizzup | ok | 10:09 |
freemangordon | hmm, ABI_INPUT has changed, lets hope there will be no issue with that | 10:09 |
freemangordon | ABI_XINPUT_VERSION that is | 10:10 |
freemangordon | only minor is changed though | 10:10 |
Wizzup | I think the xorg server builds some drivers as well no | 10:12 |
Wizzup | so you should just get them | 10:12 |
freemangordon | those are in separate repos | 10:13 |
Wizzup | ok | 10:13 |
freemangordon | but, for video, it is no issue as we use only modesetting | 10:13 |
freemangordon | and input ABI major is still 24, lets see | 10:13 |
freemangordon | Requested 'fixesproto >= 6.0' but version of FixesProto is 5.0 | 10:15 |
freemangordon | Requested 'inputproto >= 2.3.99.1' but version of InputProto is 2.3.2 | 10:15 |
freemangordon | No package 'libxcvt' found | 10:15 |
freemangordon | :( | 10:15 |
Wizzup | so this is a buster/beowulf+1 thing? | 10:21 |
freemangordon | mhm | 10:21 |
Wizzup | it could also be that master is prepped for some rc | 10:22 |
freemangordon | just installed one package from buster, with no issue | 10:22 |
freemangordon | lets see | 10:22 |
Wizzup | they never merged this either https://gitlab.freedesktop.org/xorg/xserver/-/merge_requests/568 | 10:22 |
freemangordon | yeah, but we don;t need it with chromeos mesa | 10:23 |
Wizzup | https://gitlab.freedesktop.org/xorg/xserver/-/merge_requests?scope=all&state=opened&search=glamor heh | 10:23 |
freemangordon | we need x11proto-dev_2021.5-1_all.deb | 10:24 |
freemangordon | https://salsa.debian.org/xorg-team/lib/libxcvt.git | 10:27 |
freemangordon | tmlind: yeah, glmark doesn't work on droid4-pending-pvr-omapdrm-v5.15 | 10:35 |
freemangordon | maybe the same "Fix page fault handling..." patch | 10:36 |
freemangordon | ok, upstream xorg is being build, fingers crossed | 10:37 |
Wizzup | :) | 10:40 |
freemangordon | glmark2 Score: 21 | 12:34 |
freemangordon | trying to start hildon-desktop, xorg segfault | 12:36 |
freemangordon | Thread 1 "Xorg" received signal SIGSEGV, Segmentation fault. | 12:39 |
freemangordon | 0xb55eb612 in ?? () from /root/mesa/sgx/libGLESv2_PVR_MESA.so | 12:39 |
Wizzup | debug symbols? | 12:48 |
freemangordon | no | 12:54 |
freemangordon | this is ion the blob | 12:54 |
freemangordon | *in the | 12:54 |
freemangordon | I am trying some other thing ATM | 12:55 |
freemangordon | yes, with this h-d starts | 13:07 |
freemangordon | not that it works | 13:12 |
freemangordon | but at least good thing is that I can rmmod/modprobe on hang | 13:13 |
freemangordon | ok, TFP seems to work, somehow :) | 13:21 |
freemangordon | seems we need GL_OES_texture_border_clamp or replacement | 13:24 |
freemangordon | otherwise all repaints are fullscreen | 13:24 |
Wizzup | cool @ tfp | 13:30 |
* Wizzup is excited | 13:34 | |
freemangordon | but, I don;t know how to implement GL_OES_texture_border_clamp replacement | 13:34 |
freemangordon | without that this is useless | 13:34 |
Wizzup | wonder if newer x will work with lima better | 13:34 |
Wizzup | right | 13:35 |
freemangordon | trying to play video: | 13:47 |
freemangordon | ERROR: 0:1: Syntax error, GLSL Version 120 not supported | 13:47 |
freemangordon | I think this is a dead-end | 13:47 |
bencoh | oh | 13:48 |
bencoh | what are you playing that video with? | 13:48 |
freemangordon | gst-launch | 13:48 |
bencoh | hmm | 13:48 |
bencoh | is glsl mandatory? that's odd | 13:48 |
bencoh | I bet it's for yuv/rgb conversion | 13:48 |
freemangordon | this is error from Xorg | 13:48 |
freemangordon | modesetting/glamor | 13:48 |
bencoh | ah | 13:49 |
freemangordon | yeah, I would guess the same (yuv/rgb) | 13:49 |
freemangordon | I can try to force gst to do it in SW, but what is th epoint | 13:49 |
bencoh | I'd rather try disabling it in glamor, but yeah | 13:50 |
bencoh | or backport shader to older glsl | 13:50 |
freemangordon | and where it should be done? | 13:50 |
freemangordon | yeah, porting is ok | 13:50 |
bencoh | I mean, glsl shaders have been working for ... years, I doubt they just recently added something that enables yuv/rgb :) | 13:50 |
freemangordon | but doesn;t really make sense, given that PRV2d supports tens of types of surfaces in HW | 13:51 |
bencoh | what do you mean by that? | 13:51 |
freemangordon | it is just that SGX 3D driver functionality is limited to ES2 | 13:51 |
freemangordon | I mean that we can try to not use glamor for EXA | 13:52 |
bencoh | yeah | 13:52 |
bencoh | ah, I see | 13:52 |
freemangordon | too many pieces are missing IIUC | 13:52 |
mighty17[m] | <freemangordon> "but, I don;t know how to..." <- Is the bgra extension working? | 13:52 |
freemangordon | yes | 13:52 |
freemangordon | mighty17[m]: sec, I'll pastebin the patch | 13:53 |
freemangordon | mighty17[m]: https://pastebin.com/XiYgriJ1 | 13:54 |
Wizzup | not sure if relevant https://gitlab.freedesktop.org/xorg/xserver/-/merge_requests/404 | 13:54 |
bencoh | huh, https://github.com/freedesktop/glamor/commit/b48fb31d70a16976b1cff60604e39ac042d1efac | 13:55 |
mighty17[m] | freemangordon you're a savior | 13:55 |
bencoh | I wonder if this got dropped at some point | 13:55 |
freemangordon | mighty17[m]: this alone is not enough | 13:55 |
freemangordon | at least for modesetting/glamor | 13:56 |
mighty17[m] | Can we try to patch gles instead of glamor? | 13:56 |
freemangordon | you need https://pastebin.com/ciGS2zJt too | 13:56 |
freemangordon | mighty17[m]: patch what exactly? | 13:56 |
freemangordon | uplift SGX driver to ES3.0? | 13:57 |
mighty17[m] | freemangordon: Heh that's impossible | 13:58 |
freemangordon | mhm | 13:58 |
mighty17[m] | freemangordon: Like wrapper around libegl or smth | 13:58 |
mighty17[m] | Else we'll have to patch all apps for support for the missing gles extensions | 13:59 |
freemangordon | bencoh: the point is that SGX driver seems to perform very well for 3D stuff, but glamor is FUBAR on gles2 | 13:59 |
bencoh | freemangordon: oh, I see | 14:00 |
bencoh | well then there is really no point in keeping it | 14:00 |
freemangordon | so, we can use modesetting without glamor but instead RE pvr_drv from old IMG DDKs | 14:00 |
freemangordon | but, I want to hear from uvos on that, I am not really experienced in that shit | 14:01 |
bencoh | wait, you really want to re the pvr driver? | 14:01 |
freemangordon | it is ~50k and I have a version with debug symbols | 14:01 |
freemangordon | a week or so | 14:01 |
bencoh | oh | 14:01 |
bencoh | well | 14:02 |
Wizzup | this also doens't seem too bad: $ wc -l ./src/sgx_exa.c | 14:04 |
Wizzup | 1176 ./src/sgx_exa.c | 14:04 |
freemangordon | what is this? | 14:05 |
Wizzup | xf86-video-pvrsgx driver | 14:05 |
Wizzup | based on fbdeb, but I assume that exa plugin arch didn't change much | 14:05 |
Wizzup | fbdev* | 14:05 |
Wizzup | https://github.com/maemo-leste/xf86-video-pvrsgx/blob/master/src/sgx_exa.c | 14:05 |
freemangordon | does it call into libpvr2d? | 14:06 |
Wizzup | https://github.com/maemo-leste/xf86-video-pvrsgx/blob/master/src/sgx_pvr2d.c | 14:06 |
Wizzup | maybe it contains the pvr2d stuff, I can't really tell without looking at the .so | 14:07 |
freemangordon | I think it does | 14:07 |
freemangordon | yes, it does | 14:08 |
freemangordon | it includes pvr2d.h | 14:09 |
freemangordon | which I guess comes from SDK/DDK | 14:09 |
freemangordon | actually, if we build that, it should work against the new sdk | 14:09 |
freemangordon | did we try it? | 14:09 |
freemangordon | anyway, going to a walk, ttyl | 14:10 |
mighty17[m] | I'll try to add bgra patch to wayland (wlroots) ig, is that even possible | 14:22 |
Wizzup | freemangordon: we do not want to and cannot use most of xf86-video-pvrsgx | 14:41 |
Wizzup | but the exa part we can use in modesetting | 14:41 |
Wizzup | freemangordon: yeah pvrsgx_drv_la_LDFLAGS = -module -avoid-version -lm -lpvr2d | 14:43 |
Wizzup | freemangordon: most of the stuff in there I don't think we want/need, a lot of it is related to old omapfb and the really weird api with overlays | 14:54 |
bencoh | unused symbols can't hurt though | 14:55 |
Wizzup | the code is a mess | 14:55 |
Wizzup | we're trying to move away so we don't have to touch it again | 14:55 |
bencoh | ah | 14:56 |
Wizzup | I don't think there's much of value there | 14:56 |
Wizzup | apart from the exa over pvr2d stuff | 14:56 |
Wizzup | freemangordon: does 3d work if you disable glamor, or does it require glamor to be loaded for dri3 | 14:58 |
freemangordon | I think glamor is required because of dri3 | 15:57 |
freemangordon | going to try without glamor | 15:57 |
freemangordon | hmm, it works with out glamor | 15:59 |
bencoh | neat | 16:00 |
freemangordon | but, veeeery slow | 16:00 |
freemangordon | GL_RENDERER: llvmpipe (LLVM 7.0.1, 128 bits) | 16:01 |
freemangordon | :) | 16:01 |
bencoh | huhu | 16:01 |
freemangordon | yes, because of dri3 | 16:01 |
freemangordon | but, I think I can compile mese for dri2 | 16:01 |
freemangordon | *mesa | 16:01 |
freemangordon | Wizzup: yes, we don't want xf86-video-pvrsgx, that's why I want to RE pvr_drv | 16:05 |
freemangordon | xf86-video-pvrsgx doesn't use 2D engine, iiuc | 16:05 |
freemangordon | this fbdev stuff we will want someday, because of XV stuff | 16:06 |
freemangordon | ofc it will need to be ported to drm | 16:06 |
Wizzup | freemangordon: what 2d engine is that | 18:03 |
freemangordon | the one on SGX | 18:08 |
freemangordon | "Advanced and standard 2D operations (that is, vector graphics, BLTs (fixed blitters), ROPs (raster operating processors operations))" | 18:12 |
freemangordon | Wizzup: at least blit is in HW, IIUC | 18:16 |
freemangordon | Wizzup: look at pvr2d.h for type of ROPs it can do | 18:18 |
Wizzup | ok, so this is part of the gpu then | 18:18 |
freemangordon | yeah | 18:18 |
freemangordon | and this is the public SDK | 18:19 |
freemangordon | it seems it can do more then that, at least by the looks of pvr_drv | 18:19 |
Wizzup | ok | 18:20 |
freemangordon | maybe it can do vektor stuff in HW | 18:20 |
freemangordon | *vactor | 18:20 |
freemangordon | aaah | 18:20 |
freemangordon | vector :) | 18:21 |
Wizzup | I need to go for some time of the eve, but will hopefully finish the droid3 kexecboot tomorrow | 18:26 |
uvos | hi | 20:13 |
uvos | i am back around | 20:13 |
freemangordon | hi | 20:13 |
uvos | whats the state on pvr-xorg | 20:13 |
freemangordon | bad | 20:13 |
uvos | also any suggestions on what to do about the n900 wrt wayland | 20:14 |
freemangordon | basically nothing works ok | 20:14 |
freemangordon | what do you mean? | 20:14 |
uvos | dose someone have the patched blobs for me to use with a leste image? | 20:14 |
freemangordon | sure | 20:14 |
uvos | or shal i work to get wlan working on debian 11 | 20:14 |
freemangordon | I will provide blobs, including mesa | 20:14 |
freemangordon | chromeos mesa that is | 20:14 |
uvos | ok | 20:14 |
freemangordon | BTW, I had x11 working with that | 20:15 |
freemangordon | but we hit various bug in modesetting/glamor | 20:15 |
uvos | yeah i saw that | 20:15 |
freemangordon | *bugs | 20:15 |
freemangordon | ok | 20:15 |
uvos | cool :) | 20:15 |
uvos | so the x11 emulation path in mesa works on pvr | 20:15 |
freemangordon | yeah | 20:15 |
freemangordon | so, now the plan is - I am REing pvr_drv, in hope we will have working EXA | 20:16 |
freemangordon | and then, somehow will implement PRESENT in it | 20:16 |
uvos | ok, taking the exa code from the old fbdev video-pvr driver is not an option? | 20:16 |
uvos | not for present ofc | 20:16 |
freemangordon | Wizzup: doesn;t like it | 20:17 |
freemangordon | me neuther, it doesnt really do any 2d accel | 20:17 |
freemangordon | IIUC | 20:17 |
uvos | Wizzup: can you comment on that? | 20:17 |
uvos | ok | 20:17 |
freemangordon | well, ofc I am not sure there will be any use of pvr_frv | 20:18 |
freemangordon | *pvd_drv | 20:18 |
freemangordon | what the?!? | 20:18 |
freemangordon | PVR_DRV | 20:18 |
freemangordon | but at least we will have a better understanding on what is under the hood | 20:18 |
uvos | ok | 20:19 |
uvos | Wizzup: with kexecboot on d3 | 20:20 |
uvos | if bspw is just 2 mb on d3 i dont think its viable to use for kexecboot | 20:20 |
uvos | just put it on cdrom | 20:20 |
uvos | android dosent need that | 20:21 |
uvos | cdrom is pretty big | 20:21 |
uvos | 212mb on xt894 | 20:23 |
uvos | hmm | 21:14 |
uvos | having touble compileing a hildon application | 21:14 |
uvos | hildon-gtk.h includes maemo-gtk-compat.h | 21:14 |
uvos | maemo-gtk-compat.h dosent appear to exist on my system | 21:14 |
uvos | at least acording to find | 21:15 |
uvos | oh i need to define MAEMO_GTK | 21:17 |
uvos | can we drop this | 21:17 |
uvos | since maemo-gtk-compat.h appears gohne | 21:17 |
freemangordon | [ 41324.198] (II) pvr: Driver for PowerVR chipsets: PowerVR SGX | 21:25 |
freemangordon | [ 41324.308] (EE) ERROR: Couldn't get PVR Services status | 21:25 |
uvos | freemangordon: context? | 21:26 |
freemangordon | REed driver | 21:27 |
freemangordon | most of it is still stubs though | 21:27 |
uvos | neat | 21:28 |
Wizzup | freemangordon: uvos: hi, I don't think I said I don't like the exa code for xf86-video-pvrsgx, but the fbdev code I don't like | 21:52 |
Wizzup | I cannot comment on whether it actually uses the 2d unit/api/whatever, that requires some more examination at leats from me | 21:52 |
Wizzup | uvos: I was able to load kexecboot in there I think | 21:52 |
Wizzup | it seems to fit at least | 21:52 |
uvos | well even if it dosent use it execpt to blit | 21:52 |
uvos | its still usefullt | 21:53 |
Wizzup | yeah | 21:53 |
uvos | (as we would start with pvr2d just doing blits too) | 21:53 |
uvos | Wizzup: fastboot dosent check sizes if you flash:raw | 21:53 |
Wizzup | uvos: re: d3, I think I am almost there, the only thing that is missing is mmcblk1 not showing up | 21:53 |
Wizzup | uvos: ls -ls seems to indicate it matches | 21:53 |
Wizzup | afaik | 21:54 |
uvos | Wizzup: omap4 has 3 mmc controllers iirc | 21:54 |
uvos | Wizzup: might be on a different one than on d4/bionic | 21:54 |
uvos | Wizzup: btw im massively refacotring sphone rn | 21:56 |
uvos | Wizzup: to make it modular | 21:56 |
uvos | dont work on it untill that drops | 21:56 |
Wizzup | uvos: no because it works on 5.11 | 21:58 |
Wizzup | it shows up as mmcblk2 | 21:58 |
uvos | oh ok | 21:58 |
Wizzup | so I could build 5.11 statically instead, I suppose | 21:58 |
Wizzup | just for the proof of concept | 21:58 |
Wizzup | but there are some traces in 5.14.9, so maybe that is the problem | 21:58 |
uvos | i would kinda prefer a lts kernel for the bootloader | 21:58 |
uvos | so 5.10 | 21:58 |
uvos | but yeah fixing 5.14+ is good in any case | 21:59 |
uvos | dose mmc work on d4 on 5.14? i dont think i have checked recently | 21:59 |
uvos | i wouldent notice | 21:59 |
Wizzup | not sure, haven't checked either | 22:00 |
uvos | sec ill boot the sway d4 | 22:00 |
Wizzup | right, and I mean internal mmc, but you got that | 22:00 |
uvos | works fine on d4 | 22:01 |
uvos | its mmcblk1 | 22:01 |
uvos | (5.15-rc2) | 22:02 |
uvos | maybe is broken in 5.14 only | 22:02 |
uvos | had a 5.14 kernel on it too | 22:04 |
uvos | works just the same | 22:04 |
uvos | its d3 specific | 22:04 |
uvos | or there is patch about this that i pulled in via tmlids -pending | 22:05 |
uvos | into my tree | 22:05 |
uvos | tmlind: ^^^ | 22:05 |
uvos | Wizzup: your d3 kernel is vanilla mainline right? | 22:07 |
Wizzup | yes | 22:10 |
Wizzup | just the tar from kernel.org | 22:10 |
Generated by irclog2html.py 2.17.0 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!