Ilario Gelmetti
2022-Jan-08 00:22 UTC
[Nouveau] Piglit results for NV46 and glitches using GeForce 7500LE
Dear all, thanks for your work on Nouveau!!! I have been a user, but I never touched the Nouveau code or testing suite before. I am building a desktop PC with some rather old hardware and a nVidia GeForce 7500LE 64MB TC V/D/VO PN 88-1N01-0C-PB (NV46, G72) graphic card (see lshw output here [1]). I observed frequent glitches with some common user experiences: * Ubuntu 21.10 * Ubuntu 18.04 * Debian 11 stable with GDM3+Gnome * Debian 10 oldstable with GDM3+Gnome with errors in dmesg like this [2] and glitches like these [3] * Debian 10 oldstable with GDM3+XFCE with usually no errors in dmesg [4] and glitches like these [5]. But sometimes runs smoothly The situation improved replacing GDM3 by LightDM (why??), so now I have these two setups where under normal operation I see no glitches (but glitches appear after running piglit): * Arch without DM and with XFCE * Debian oldstable with LightDM and Gnome So I run piglit with various combinations of software versions and distros, and X crashed with some of these (the only responsive thing was the moving-but-not-shapeshifting mouse arrow, like this [6]). Here I detail the piglit tests and the results: * Debian 10 oldstable (kernel 4.19) + LightDM + Gnome + piglit compiled from source + libdrm2 (2.4.97) + libdrm-nouveau (2.4.97) + xserver-xorg-core 1.20.4 + xserver-xorg-video-nouveau (1.0.16) no glitches observed during normal operation both piglit from Debian oldstable repository and piglit compiled from source run to completion but at the end there are some glitches at the screen, like this [7], or sometimes just a black screen or all the windows as black rectangles. Here you can find the result of the tests in these conditions: [8] * Arch Linux + kernel (tested both 5.15.12 and 5.16.0-rc8 from linux-git package) + nouveau (tested both 1.0.17 and 1.0.17.r2 from xf86-video-nouveau-git package) + libdrm (tested both 2.4.109 and 2.4.109.r7 from libdrm-git package) + mesa (tested both 21.3.3 and 22.0.0_devel from mesa-git package) + XFCE no glitches observed during normal operation piglit-git (r11511.6c4da153b) makes X crash (or something similar, the kernel seems not panicked as magic-sysrq work). This happens reproducibly always after the getteximage-depth test (which itself succeeds but the next one, any next one, gets stuck in the "running" status) and very often also when the other "*getteximage*" or "*vertex*" or "generatemipmap-base-change" or "arb_get_texture_sub_image" or "texture with glblitframebuffer" or "fbo-finish-deleted" or "ext_image_dma_buf_import_sample_argb8888" or "depthstencil-render-mpilevels" or "readpixels-24_8" (I noticed that when it does not make X crash it just results in dmesg-fail output, maybe this happens also to the others, I didn't check) or "texsubimage" or "draw-pixels". The glitches caused by running piglit look like these [9]. When I manage to get out of the crashed X (using Alt+SysRq+R,E) I can get some error messages from dmesg, like these [10]. And here are also some coredumps obtained via "coredumpctl dump" command [11]. After many reboots and adding many exclusions to the tests list I gave up trying to reach the end of the tests list, so here you have a couple of incomplete lists with the results that got to actually run: [12] Let me know if I can help. Thanks for your help and for your precious work! :) Ilario Gelmetti [1]: https://uz.sns.it/~ilario/nouveau-nv46/lshw [2]: https://uz.sns.it/~ilario/nouveau-nv46/debian/dmesg/20220107-nouveau-gdm3-gnome [3]: https://uz.sns.it/~ilario/nouveau-nv46/debian/glitches/IMG_20220108_002132.jpg https://uz.sns.it/~ilario/nouveau-nv46/debian/glitches/IMG_20220108_002140.jpg [4]: https://uz.sns.it/~ilario/nouveau-nv46/debian/dmesg/20220107-nouveau-gdm3-xfce [5]: https://uz.sns.it/~ilario/nouveau-nv46/debian/glitches/IMG_20220108_001511.jpg [6]: https://uz.sns.it/~ilario/nouveau-nv46/arch/glitches/IMG_20220104_130045.jpg [7]: https://uz.sns.it/~ilario/nouveau-nv46/debian/glitches/IMG_20220104_231136.jpg [8]: https://uz.sns.it/~ilario/nouveau-nv46/nv46-2022-01-05_1020-ilario.json.bz2 [9]: https://uz.sns.it/~ilario/nouveau-nv46/arch/glitches/ [10]: https://uz.sns.it/~ilario/nouveau-nv46/arch/dmesg/ [11]: https://uz.sns.it/~ilario/nouveau-nv46/arch/coredumps/ [12]: https://uz.sns.it/~ilario/nouveau-nv46/arch/nv46-2022-01-06_1321-ilario-incomplete.tar.xz https://uz.sns.it/~ilario/nouveau-nv46/arch/nv46-2022-01-07_2210-ilario-incomplete.tar.xz -- Ilario iochesonome at gmail.com ilario at sindominio.net
Karol Herbst
2022-Jan-11 10:14 UTC
[Nouveau] Piglit results for NV46 and glitches using GeForce 7500LE
I've fixed a really really really old bug in mesa which impacted gnome: https://gitlab.freedesktop.org/mesa/mesa/-/commit/1387d1d41103b3120d40f93f66a7cfe00304bfd7 this one is part of 21.3 and should have made it into 21.1 and 21.2 stable releases. We also have two pending merge requests which should fix other issues related to gnome and random other stuff: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests?scope=all&state=opened&label_name[]=nv30 I'd suggest if you want to start digging into issues you do so with a fixed mesa and those merge requests applied. On Mon, Jan 10, 2022 at 4:05 PM Ilario Gelmetti <iochesonome at gmail.com> wrote:> > Dear all, > thanks for your work on Nouveau!!! > I have been a user, but I never touched the Nouveau code or testing > suite before. > > I am building a desktop PC with some rather old hardware and a nVidia > GeForce 7500LE 64MB TC V/D/VO PN 88-1N01-0C-PB (NV46, G72) graphic card > (see lshw output here [1]). > > I observed frequent glitches with some common user experiences: > * Ubuntu 21.10 > * Ubuntu 18.04 > * Debian 11 stable with GDM3+Gnome > * Debian 10 oldstable with GDM3+Gnome with errors in dmesg like this [2] > and glitches like these [3] > * Debian 10 oldstable with GDM3+XFCE with usually no errors in dmesg [4] > and glitches like these [5]. But sometimes runs smoothly > > The situation improved replacing GDM3 by LightDM (why??), so now I have > these two setups where under normal operation I see no glitches (but > glitches appear after running piglit): > * Arch without DM and with XFCE > * Debian oldstable with LightDM and Gnome > > So I run piglit with various combinations of software versions and > distros, and X crashed with some of these (the only responsive thing was > the moving-but-not-shapeshifting mouse arrow, like this [6]). > > Here I detail the piglit tests and the results: > > * Debian 10 oldstable (kernel 4.19) + LightDM + Gnome + piglit compiled > from source + libdrm2 (2.4.97) + libdrm-nouveau (2.4.97) + > xserver-xorg-core 1.20.4 + xserver-xorg-video-nouveau (1.0.16) > > no glitches observed during normal operation > both piglit from Debian oldstable repository and piglit compiled from > source run to completion but at the end there are some glitches > at the screen, like this [7], or sometimes just a black screen or all > the windows as black rectangles. > Here you can find the result of the tests in these conditions: [8] > > * Arch Linux + kernel (tested both 5.15.12 and 5.16.0-rc8 from linux-git > package) + nouveau (tested both 1.0.17 and 1.0.17.r2 from > xf86-video-nouveau-git package) + libdrm (tested both 2.4.109 and > 2.4.109.r7 from libdrm-git package) + mesa (tested both 21.3.3 and > 22.0.0_devel from mesa-git package) + XFCE > > no glitches observed during normal operation > piglit-git (r11511.6c4da153b) makes X crash (or something similar, the > kernel seems not panicked as magic-sysrq work). This happens > reproducibly always after the getteximage-depth test (which itself > succeeds but the next one, any next one, gets stuck in the "running" > status) and very often also when the other "*getteximage*" or "*vertex*" > or "generatemipmap-base-change" or "arb_get_texture_sub_image" or > "texture with glblitframebuffer" or "fbo-finish-deleted" or > "ext_image_dma_buf_import_sample_argb8888" or > "depthstencil-render-mpilevels" or "readpixels-24_8" (I noticed that > when it does not make X crash it just results in dmesg-fail output, > maybe this happens also to the others, I didn't check) or "texsubimage" > or "draw-pixels". > The glitches caused by running piglit look like these [9]. > When I manage to get out of the crashed X (using Alt+SysRq+R,E) I can > get some error messages from dmesg, like these [10]. > And here are also some coredumps obtained via "coredumpctl dump" command > [11]. > After many reboots and adding many exclusions to the tests list I gave > up trying to reach the end of the tests list, so here you have a couple > of incomplete lists with the results that got to actually run: [12] > > Let me know if I can help. > Thanks for your help and for your precious work! :) > Ilario Gelmetti > > [1]: https://uz.sns.it/~ilario/nouveau-nv46/lshw > > [2]: > https://uz.sns.it/~ilario/nouveau-nv46/debian/dmesg/20220107-nouveau-gdm3-gnome > > [3]: > https://uz.sns.it/~ilario/nouveau-nv46/debian/glitches/IMG_20220108_002132.jpg > https://uz.sns.it/~ilario/nouveau-nv46/debian/glitches/IMG_20220108_002140.jpg > > [4]: > https://uz.sns.it/~ilario/nouveau-nv46/debian/dmesg/20220107-nouveau-gdm3-xfce > > [5]: > https://uz.sns.it/~ilario/nouveau-nv46/debian/glitches/IMG_20220108_001511.jpg > > [6]: > https://uz.sns.it/~ilario/nouveau-nv46/arch/glitches/IMG_20220104_130045.jpg > > [7]: > https://uz.sns.it/~ilario/nouveau-nv46/debian/glitches/IMG_20220104_231136.jpg > > [8]: > https://uz.sns.it/~ilario/nouveau-nv46/nv46-2022-01-05_1020-ilario.json.bz2 > > [9]: https://uz.sns.it/~ilario/nouveau-nv46/arch/glitches/ > > [10]: https://uz.sns.it/~ilario/nouveau-nv46/arch/dmesg/ > > [11]: https://uz.sns.it/~ilario/nouveau-nv46/arch/coredumps/ > > [12]: > https://uz.sns.it/~ilario/nouveau-nv46/arch/nv46-2022-01-06_1321-ilario-incomplete.tar.xz > https://uz.sns.it/~ilario/nouveau-nv46/arch/nv46-2022-01-07_2210-ilario-incomplete.tar.xz > > > > -- > Ilario > iochesonome at gmail.com > ilario at sindominio.net > >