loonggpu 1.0.1 的几个问题

tsingkong

测试环境：

OS: Debian GNU/Linux trixie/sid loongarch64
Host: XC-LS3A6M
Kernel: Linux 6.12.10-loong64
Uptime: 9 hours, 53 mins
Packages: 4995 (dpkg)
Shell: bash 5.2.37
Display (28E850): 1920x1080 @ 120 Hz in 22" [External]
DE: KDE Plasma 6.2.5
WM: KWin (X11)
WM Theme: Breeze 微风
Theme: Breeze (KDEStoryDark) [Qt], Breeze-Dark [GTK2], Breeze [GTK3]
Icons: We10X-black [Qt], We10X-black [GTK2/3/4]
Font: 宋体 (9pt) [Qt], 宋体 (9pt) [GTK2/3/4]
Cursor: KDE_Classic (24px)
Terminal: konsole 24.12.0
Terminal Font: 宋体 (9pt)
CPU: Loongson-3A6000-HV loongarch64 (8) @ 2.50 GHz
GPU: Loongson Technology LLC Display Controller
Memory: 8.96 GiB / 31.68 GiB (28%)

不兼容 4k page size

2月 04 16:08:32 debian kernel: [drm] GPU Family: LG100 series LG110, Feature:0x00000000
2月 04 16:08:32 debian kernel: [drm] Firmware Version: 1.0.0
2月 04 16:08:32 debian kernel: [drm] aper_base 0xe0030000000 SIZE 0x10000000 bytes 
2月 04 16:08:32 debian kernel: loonggpu 0000:00:06.0: Video RAM: 256M 0x0001000000000000 - 0x000100000FFFFFFF (256M used)
2月 04 16:08:32 debian kernel: loonggpu 0000:00:06.0: GART: 256M 0x0000000000000000 - 0x000000000FFFFFFF
2月 04 16:08:32 debian kernel: [drm] gsgpu: 3072M of GTT memory ready.
2月 04 16:08:32 debian kernel: [drm:gsgpu_gart_init [loonggpu]] *ERROR* Page size is smaller than GPU page size!
2月 04 16:08:32 debian kernel: [drm:gsgpu_device_init [loonggpu]] *ERROR* sw_init of IP block <mmu> failed -22
2月 04 16:08:32 debian kernel: loonggpu 0000:00:06.0: gsgpu_device_ip_init failed
2月 04 16:08:32 debian kernel: loonggpu 0000:00:06.0: Fatal error during GPU init
2月 04 16:08:32 debian kernel: gsgpu-dc 0000:00:06.1: probe with driver gsgpu-dc failed with error -22

6.13 内核下工作不正常，使用 aosc 的补丁(https://github.com/AOSC-Tracking/loonggpu-kernel-dkms/tree/aosc/v1.0.1-alpha-lnd25.5)后可以编译内核驱动，但是 libEGL_loonggpu.so 报错:

[   179.584] (EE) Backtrace:
[   179.584] (EE) 0: /usr/lib/xorg/Xorg (xorg_backtrace+0x58) [0x5555586e5c38]
[   179.584] (EE) 1: /usr/lib/xorg/Xorg (0x5555584e0000+0x20a4c4) [0x5555586ea4c4]
[   179.584] (EE) 2: linux-vdso.so.1 (__vdso_rt_sigreturn+0x0) [0x7ffffc448f94]
[   179.584] (EE) 3: /lib/loongarch64-linux-gnu/libc.so.6 (0x7ffff2d40000+0xa94cc) [0x7ffff2de94cc]
[   179.584] (EE) 4: /lib/loongarch64-linux-gnu/libc.so.6 (__strdup+0x18) [0x7ffff2de08b8]
[   179.584] (EE) 5: /lib/loongarch64-linux-gnu/libEGL_loonggpu.so.0 (0x7fffe3c3c000+0x24924) [0x7fffe3c60924]
[   179.584] (EE) 6: /lib/loongarch64-linux-gnu/libEGL_loonggpu.so.0 (0x7fffe3c3c000+0x1d1e4) [0x7fffe3c591e4]
[   179.584] (EE) 7: /lib/loongarch64-linux-gnu/libEGL_loonggpu.so.0 (0x7fffe3c3c000+0x19b28) [0x7fffe3c55b28]
[   179.584] (EE) 8: /lib/loongarch64-linux-gnu/libEGL_loonggpu.so.0 (eglInitialize+0x140) [0x7fffe3c508e8]
[   179.584] (EE) 9: /usr/lib/xorg/modules/libglamoregl.so (glamor_egl_init+0xc0) [0x7ffff1998d68]
[   179.584] (EE) 10: /usr/lib/xorg/modules/drivers/modesetting_drv.so (0x7ffff1a20000+0xb4bc) [0x7ffff1a2b4bc]
[   179.584] (EE) 11: /usr/lib/xorg/Xorg (InitOutput+0x92c) [0x5555585948d4]
[   179.584] (EE) 12: /usr/lib/xorg/Xorg (0x5555584e0000+0x6e9bc) [0x55555854e9bc]
[   179.584] (EE) 13: /lib/loongarch64-linux-gnu/libc.so.6 (0x7ffff2d40000+0x1d8b0) [0x7ffff2d5d8b0]
[   179.584] (EE) 14: /lib/loongarch64-linux-gnu/libc.so.6 (__libc_start_main+0xa8) [0x7ffff2d5d998]
[   179.584] (EE) 15: /usr/lib/xorg/Xorg (_start+0x48) [0x555558533ab0]

6.12.10内核中，kde及大部分应用运行正常，目前只发现一个应用(evolution)启动异常:

Stack trace of thread 4848:
#0  0x00007fffe9e194cc __strlen_lasx (libc.so.6 + 0xa94cc)
#1  0x00007fffe9e108b8 __GI___strdup (libc.so.6 + 0xa08b8)
#2  0x00007fffbc05c924 n/a (libEGL_loonggpu.so.0 + 0x24924)
#3  0x00007fffbc0551e4 n/a (libEGL_loonggpu.so.0 + 0x1d1e4)
#4  0x00007fffbc051b28 n/a (libEGL_loonggpu.so.0 + 0x19b28)
#5  0x00007fffbc04c8e8 eglInitialize (libEGL_loonggpu.so.0 + 0x148e8)
#6  0x00007fffed723b78 n/a (libwebkit2gtk-4.1.so.0 + 0x3233b78)
#7  0x00007fffed75316c n/a (libwebkit2gtk-4.1.so.0 + 0x326316c)
#8  0x00007fffeb2ad570 n/a (libwebkit2gtk-4.1.so.0 + 0xdbd570)
#9  0x00007fffe9dfbc44 __pthread_once_slow (libc.so.6 + 0x8bc44)
#10 0x00007fffe9dfbd0c ___pthread_once (libc.so.6 + 0x8bd0c)
#11 0x00007fffeb2aa380 n/a (libwebkit2gtk-4.1.so.0 + 0xdba380)
#12 0x00007fffeb2a9020 n/a (libwebkit2gtk-4.1.so.0 + 0xdb9020)
#13 0x00007fffeb2ba2a4 n/a (libwebkit2gtk-4.1.so.0 + 0xdca2a4)
#14 0x00007fffeb2ba358 n/a (libwebkit2gtk-4.1.so.0 + 0xdca358)
#15 0x00007fffeb2c82d4 n/a (libwebkit2gtk-4.1.so.0 + 0xdd82d4)
#16 0x00007fffeb128880 n/a (libwebkit2gtk-4.1.so.0 + 0xc38880)
#17 0x00007fffeb1c0144 n/a (libwebkit2gtk-4.1.so.0 + 0xcd0144)
#18 0x00007fffef76c888 g_type_create_instance (libgobject-2.0.so.0 + 0x3c888)
#19 0x00007fffef74e8e0 n/a (libgobject-2.0.so.0 + 0x1e8e0)
#20 0x00007fffef75161c g_object_new_valist (libgobject-2.0.so.0 + 0x2161c)
#21 0x00007fffeb1c0a50 webkit_settings_new_with_settings (libwebkit2gtk-4.1.so.0 + 0xcd0a50)
#22 0x00007fffea14bd68 e_web_view_get_default_webkit_settings (libevolution-util.so.0 + 0x1fbd68)
#23 0x00007fffea14bf08 n/a (libevolution-util.so.0 + 0x1fbf08)

tsingkong

更新一下问题2，不管是 6.12.x 还是 6.13.x , kde 大概率起不来，目前只能暂时用 xfce。
问题 4, 运行 eglinfo 异常：

$ eglinfo
EGL client extensions string:
    EGL_EXT_client_extensions, EGL_EXT_device_base, 
    EGL_EXT_device_enumeration, EGL_EXT_device_query, EGL_EXT_explicit_device, 
    EGL_EXT_platform_base, EGL_EXT_platform_device, EGL_EXT_platform_wayland, 
    EGL_EXT_platform_x11, EGL_EXT_platform_xcb, 
    EGL_KHR_client_get_all_proc_addresses, EGL_KHR_debug, 
    EGL_KHR_platform_gbm, EGL_KHR_platform_wayland, EGL_KHR_platform_x11, 
    EGL_MESA_platform_gbm, EGL_MESA_platform_surfaceless

GBM platform:
gsgpu_device_initialize: gsgpu_query_info(ACCEL_WORKING) failed (-13)
gsgpu_winsys_create: gsgpu_device_initialize failed [-13]
段错误（核心已转储）

journal log:

2月 07 14:02:21 debian kernel: do_page_fault(): sending SIGSEGV to eglinfo for invalid read access from 00000000000000b0
2月 07 14:02:21 debian kernel: era = 00007fffe2863240 in gsgpu_dri.so[16b240,7fffe26f8000+7f4000]
2月 07 14:02:21 debian kernel: ra  = 00007fffe2863240 in gsgpu_dri.so[16b240,7fffe26f8000+7f4000]
2月 07 14:02:21 debian systemd-coredump[856925]: Process 856924 (eglinfo) of user 1000 terminated abnormally with signal 11/SEGV, processing...
2月 07 14:02:21 debian systemd[1]: Started systemd-coredump@52-856925-0.service - Process Core Dump (PID 856925/UID 0).
2月 07 14:02:21 debian systemd[1]: Started drkonqi-coredump-processor@52-856925-0.service - Pass systemd-coredump journal entries to relevant user for potential DrKonqi handling.
2月 07 14:02:21 debian systemd-coredump[856926]: [🡕] Process 856924 (eglinfo) of user 1000 dumped core.
          Module libgcc_s.so.1 from deb gcc-14-14.2.0-16.loong64
          Module libstdc++.so.6 from deb gcc-14-14.2.0-16.loong64
          Module libzstd.so.1 from deb libzstd-1.5.6+dfsg-2.loong64
          Stack trace of thread 856924:
          #0  0x00007fffe2863240 n/a (gsgpu_dri.so + 0x16b240)
          #1  0x00007fffe2770c5c n/a (gsgpu_dri.so + 0x78c5c)
          #2  0x00007fffe2819ae4 n/a (gsgpu_dri.so + 0x121ae4)
          #3  0x00007fffe2774dc4 n/a (gsgpu_dri.so + 0x7cdc4)
          #4  0x00007fffe277219c n/a (gsgpu_dri.so + 0x7a19c)
          #5  0x00007ffff10a03d8 n/a (libgbm.so.1 + 0x43d8)
          #6  0x00007ffff10a0c38 n/a (libgbm.so.1 + 0x4c38)
          #7  0x00007ffff109e930 gbm_create_device (libgbm.so.1 + 0x2930)
          #8  0x00007ffff10d8cb4 n/a (libEGL_loonggpu.so.0 + 0x24cb4)
          #9  0x00007ffff10d11e4 n/a (libEGL_loonggpu.so.0 + 0x1d1e4)
          #10 0x00007ffff10cdb28 n/a (libEGL_loonggpu.so.0 + 0x19b28)
          #11 0x00007ffff10c88e8 eglInitialize (libEGL_loonggpu.so.0 + 0x148e8)
          #12 0x0000555555f09ac0 n/a (n/a + 0x0)
          ELF object binary architecture: LoongArch

tsingkong

关于问题2 的追踪分析：

华硕主板上有 VGA 接口和 HDMI 接口，这两个接口我都接了显示器；
/usr/lib/loongarch64-linux-gnu/qt6/plugins/kf6/kscreen/KSC_XRandR.so 异常，导致 kscreen_backend_launcher 崩溃；
kscreen_backend_launcher 崩溃导致 Xorg 异常退出，从而整个 KDE 无法启动；
临时解决方法（二选一）：
卸载 libkscreen-bin 及依赖它的 kscreen, powerdevil；
只接一个显示器；

tsingkong

关于问题3 的临时解决办法：
设置环境变量 WEBKIT_DISABLE_DMABUF_RENDERER=1

tsingkong

关于问题2 ，有一个方法可以保留 kscreen, powerdevil ：
创建 ~/.config/plasma-workspace/env/kwin_vars.sh, 内容:

#!/bin/bash 
xrandr --output HDMI-1 --same-as VGA-1

liudidi

求助，我在6.12.17内核下编译有大量的出错提示，查了一下，有些包含的头文件在6.12中已经没有了，请问你在6.12下编译是正常的吗？

tsingkong

liudidi 6.12 应该是正常的呀。另外，建议用 aosc 修改后的版本，可以用在 6.13 和 6.14 上。

tsingkong

liudidi 如果缺少 ttm 相关的头，可能是因为编译内核的时候没有选中 ttm 支持。
根据我的经验，需要至少选中 DRM 中的一个驱动，amdgpu, 或者xe，或者 radeon

liudidi

tsingkong 就是aosc的版本，仔细看了一下，发现不是没有ttm的头文件，而是在找以前内核的头文件。conftest出来的结果是不对的。我手工修改了大量的宏定义，到是能通过编译了，但是insmod出错，应该是还没有改全。我再看一下，谢谢。

liudidi

tsingkong 知道原因了，我用的gcc15，默认的c23不行的，改成gnu17就能编译了。

liudidi

折腾了好几天，还是不行，fedora的底层可能和deepin区别太大了，光ncurses就折腾了两天才行，但是也不知道什么原因，运行weston就段错误，运行startx就是unw_get_proc_name failed: no unwind info found。算了，不折腾了，还是用rx580吧。

鄂ICP备2022017735号

此论坛为纯技术类论坛, 且为"独立运作，风险自控", 请用户在论坛中不要讨论无关话题, 以免影响论坛正常运行