openharmony系统移植之gpu mesa3d适配

1. 环境说明

• 芯片平台： 第三方soc cortex-A7

• gpu：mail-T628

• 系统：openharmony-v4.0-release tag

gpu mesa3d适配的前提是先完成cpu点屏，再进行gpu适配，整个调试过程困难重重，感谢DIEMIT，Algoldeas大佬的指导。

下面主要描述整个移植过程和一些经验分享。

前期参考DIEMIT大佬的视频从零开始移植OpenHarmony_第4节_CPU亮屏，已经完成openharmony下cpu点屏，在此过程中遇到了显示花屏也困扰很久，具体解决情况见链接openharmony4.0 cpu点屏图像显示花屏。

2. gpu内核panfrost驱动

panfrost的是对ARM 系列GPU驱动的开源实现，它的功能主要是完成对GPU硬件的初始化，以及以job的方式，完成对渲染数据硬件处理。对应应用层，GPU相关的配置，渲染管理等都是通过Mesa3D对panfrost ioctl来实现的。

在drivers/gpu/drm/panfrost/panfrost_drv.c源码中dt_match可以确定panfrost是支持mail-t628 gpu的，那么下一步主要是defconfig中打开panfrost驱动配置，以及dts相关的配置。

static const struct of_device_id dt_match[] = {
        /* Set first to probe before the generic compatibles */
        { .compatible = "amlogic,meson-gxm-mali",
          .data = &amlogic_data, },
        { .compatible = "amlogic,meson-g12a-mali",
          .data = &amlogic_data, },
        { .compatible = "arm,mali-t604", .data = &default_data, },
        { .compatible = "arm,mali-t624", .data = &default_data, },
        { .compatible = "arm,mali-t628", .data = &default_data, },
        { .compatible = "arm,mali-t720", .data = &default_data, },
        { .compatible = "arm,mali-t760", .data = &default_data, },
        { .compatible = "arm,mali-t820", .data = &default_data, },
        { .compatible = "arm,mali-t830", .data = &default_data, },
        { .compatible = "arm,mali-t860", .data = &default_data, },
        { .compatible = "arm,mali-t880", .data = &default_data, },
        { .compatible = "arm,mali-bifrost", .data = &default_data, },
        {}
};
MODULE_DEVICE_TABLE(of, dt_match);

2.1 使能panfrost驱动

在对应的defconfig中配置如下即可

CONFIG_DRM_PANFROST=y

2.2 panfrost dts配置

在kernel源码中搜索"arm,mali-t628"，可以看到arch/arm/boot/dts/exynos5420.dtsi中有实现，刚好可以作为参考。

arch/arm/boot/dts/exynos5420.dtsi 参考如下。主要是配置中断，寄存器基地址，供电，clk和opp-table。

                gpu: gpu@11800000 {
                        compatible = "samsung,exynos5420-mali", "arm,mali-t628";
                        reg = <0x11800000 0x5000>;
                        interrupts = <GIC_SPI 219 IRQ_TYPE_LEVEL_HIGH>,
                                     <GIC_SPI 74 IRQ_TYPE_LEVEL_HIGH>,
                                     <GIC_SPI 117 IRQ_TYPE_LEVEL_HIGH>;
                        interrupt-names = "job", "mmu", "gpu";

                        clocks = <&clock CLK_G3D>;
                        clock-names = "core";
                        power-domains = <&g3d_pd>;
                        operating-points-v2 = <&gpu_opp_table>;

                        status = "disabled";
                        #cooling-cells = <2>;

                        gpu_opp_table: opp-table {
                                compatible = "operating-points-v2";

                                opp-177000000 {
                                        opp-hz = /bits/ 64 <177000000>;
                                        opp-microvolt = <812500>;
                                };
                                opp-266000000 {
                                        opp-hz = /bits/ 64 <266000000>;
                                        opp-microvolt = <862500>;
                                };
                                opp-350000000 {
                                        opp-hz = /bits/ 64 <350000000>;
                                        opp-microvolt = <912500>;
                                };
                                opp-420000000 {
                                        opp-hz = /bits/ 64 <420000000>;
                                        opp-microvolt = <962500>;
                                };
                                opp-480000000 {
                                        opp-hz = /bits/ 64 <480000000>;
                                        opp-microvolt = <1000000>;
                                };
                                opp-543000000 {
                                        opp-hz = /bits/ 64 <543000000>;
                                        opp-microvolt = <1037500>;
                                };
                                opp-600000000 {
                                        opp-hz = /bits/ 64 <600000000>;
                                        opp-microvolt = <1150000>;
                                };
                        };
                };

结合芯片手册，实际实现如下：

                gpu: gpu@a0100000 {
                        compatible = "arm,mali-t628";
                        reg = <0xa0100000 0x3fff>;
                        interrupts = <GIC_SPI 11 IRQ_TYPE_LEVEL_HIGH>,
                                <GIC_SPI 12 IRQ_TYPE_LEVEL_HIGH>,
                                <GIC_SPI 13 IRQ_TYPE_LEVEL_HIGH>;
                        interrupt-names = "job", "mmu", "gpu";

                        clocks = <&gpu_mclk>, <&gpu_adb_aclkmgt>, <&xxx>;
                        clock-names = "core", "bus", "xxx";
                        mali-supply = <&xxx>;
                        #cooling-cells = <2>;
                        operating-points-v2 = <&gpu_opp_table>;

                        gpu_opp_table: opp-table {
                                compatible = "operating-points-v2";
                                opp-156000000 {
                                        opp-hz = /bits/ 64 <156000000>;
                                        opp-microvolt = <xxx>;
                                };
                                ......
                        };
                };

其中mali-supply需要配置，power-domains方式供电，暂不支持，实现我们在代码中panfrost初始化过程中配置上电。panfrost加载日志如下。

3. buildroot下测试gpu驱动

panfrost驱动，目前还无法保证是否正常，需要使用mesa3d glmark2来做验证测试。buildroot比较方便，参考Openharmony之GPU Mesa3D移植一（weston 老框架）来编译测试panfrost驱动。

3.1 buildroot配置编译

使用buildroot-2023.02源码，编译完成后生成root.ext4文件烧录到system分区，原来kernel是加载ramdisk的，现在需要改为加载system分区。将uboot bootargs中进行修改。

删除initrd

initrd=xxx,xxx

修改原来的root=/dev/ram0 init=/init为

root=/dev/mmcblk0p20 init=/linuxrc rootfstype=ext4 rw rootwait

这样kernel启动就会直接加载system分区文件系统，方便测试。

glmark2测试命令如下

mkdir /tmp/xdg  
export XDG_RUNTIME_DIR=/tmp/xdg  
weston --tty 1 &  
glmark2-es2-wayland  

在实际运行中weston --tty 1 & 跑不起来，报错退出。

信息如下：

failed to bind extensions
failed to initialize egl

查询了各种资料，未能解决，此时就先在ohos下mesa3d编译适配（见下一节），依旧glmark2跑不起了。后来想到决定直接使用最新的buildroot源码来再来测试看看。

git clone一份最新官方最新的源码，git check 2025.02 ，切换到最新的稳定tag标签2025.02。

增加configs/mesa3d_defconfig， 内容如下：

BR2_arm=y
BR2_cortex_a7=y
BR2_ARM_FPU_NEON_VFPV4=y
BR2_ARM_EABI=y
BR2_TOOLCHAIN_BUILDROOT_MUSL=y
#BR2_TOOLCHAIN_BUILDROOT_GLIBC=y
BR2_KERNEL_HEADERS_5_10=y
BR2_TOOLCHAIN_BUILDROOT_CXX=y
BR2_ROOTFS_DEVICE_CREATION_DYNAMIC_EUDEV=y
BR2_PACKAGE_MESA3D_GALLIUM_DRIVER_PANFROST=y
BR2_PACKAGE_MESA3D_GALLIUM_DRIVER_KMSRO=y
BR2_PACKAGE_MESA3D=y
BR2_PACKAGE_MESA3D_OPENGL_ES=y
BR2_PACKAGE_MESA3D_OPENGL_EGL=y
BR2_PACKAGE_GLMARK2=y
BR2_PACKAGE_WESTON=y
BR2_PACKAGE_WESTON_DEFAULT_DRM=y
BR2_PACKAGE_LIBDRM=y
BR2_PACKAGE_LIBDRM_INSTALL_TESTS=y
BR2_TARGET_ROOTFS_EXT2=y
BR2_TARGET_ROOTFS_EXT2_4=y
BR2_TARGET_ROOTFS_EXT2_SIZE="100M"

在调试过程，发现还需参考其他芯片增加芯片平台对应的支持。

增加package/mesa3d/0005-gallium-drm-support.patch，内容如下，其中virtsoc需要和实际芯片drm显示驱动中name相同。

rom 390a3b9e146243d706b38e76438490197edea61a Mon Sep 17 00:00:00 2001
From: songze_lee <songze_lee@163.com>
Date: Sun, 20 Apr 2025 10:10:27 +0800
Subject: [PATCH] add virtsoc support

---
 src/gallium/targets/dri/meson.build | 1 +
 src/gallium/targets/dri/target.c    | 1 +
 2 files changed, 2 insertions(+)

diff --git a/src/gallium/targets/dri/meson.build b/src/gallium/targets/dri/meson.build
index 8392524..46ac77a 100644
--- a/src/gallium/targets/dri/meson.build
+++ b/src/gallium/targets/dri/meson.build
@@ -101,6 +101,7 @@ foreach d : [[with_gallium_kmsro, [
                'stm_dri.so',
                'sun4i-drm_dri.so',
                'udl_dri.so',
+               'virtsoc_dri.so',
              ]],
              [with_gallium_radeonsi, 'radeonsi_dri.so'],
              [with_gallium_nouveau, 'nouveau_dri.so'],
diff --git a/src/gallium/targets/dri/target.c b/src/gallium/targets/dri/target.c
index 415e494..4b82358 100644
--- a/src/gallium/targets/dri/target.c
+++ b/src/gallium/targets/dri/target.c
@@ -129,6 +129,7 @@ DEFINE_LOADER_DRM_ENTRYPOINT(sti)
 DEFINE_LOADER_DRM_ENTRYPOINT(stm)
 DEFINE_LOADER_DRM_ENTRYPOINT(sun4i_drm)
 DEFINE_LOADER_DRM_ENTRYPOINT(udl)
+DEFINE_LOADER_DRM_ENTRYPOINT(virtsoc)
 #endif

 #if defined(GALLIUM_LIMA)
--
2.17.1

glmark2测试命令继续进行测试，weston --tty 1 & 可以跑起来，显示屏有实际显示。但glmark2-es2-wayland 跑不起了。各种痛苦折腾查资料，偶尔看见Panfrost使能润和DAYU200（RK3568）使用的是glmark2-es2-drm程序测试的，因此使用glmark2-es2-drm实测，终于出现了令人激动人心的画面。

glmark2-es2-drm显示效果如下：

至此，基本验证的底层panfrost驱动是正常的。

4. ohos下mesa3d适配

本节主要参考DIEMIT大佬的视频从零开始移植OpenHarmony_第5节_GPU适配。

主要涉及三部分：

• third_party/mesa3d编译

• native_window_wrapper编译

• glmark2 编译

native_window_wrapper主要作用为转接glmark2渲染到oh窗口

glmark2源码使用：git clone https://gitee.com/diemit/glmark2_2 -b OH4.0

下面主要描述GPU适配过程中实际遇到的问题。

4.1 ohos下mesa3d编译调试

4.1.2 编译

使用build_ohos.py 编译

命令如下

python ohos/build_ohos.py 参数1  参数2 参数3

参数1：ohos源码目录绝对路径

参数2：产品名称

参数3：mesa3d源码目录绝对路径

视频中适配的为树莓派4B gpu，mesa3D中具体的gallium-drivers需要根据具体芯片情况修改，我们使用的panfrost。

diff --git a/ohos/build_ohos.py b/ohos/build_ohos.py
index 8f2485922ed..f2312bcf806 100644
--- a/ohos/build_ohos.py
+++ b/ohos/build_ohos.py
@@ -34,7 +34,7 @@ if __name__ == '__main__':
     run_build_cmd = 'PKG_CONFIG_PATH=./pkgconfig '
     run_build_cmd += 'meson setup '+ sys.argv[3] + ' build-ohos '
     run_build_cmd += '-Dplatforms=ohos -Degl-native-platform=ohos -Ddri-drivers= -Dgallium-drivers=panfrost \
-                      -Dvulkan-drivers= -Dgbm=enabled -Degl=enabled -Dcpp_rtti=false -Dglx=disabled -Dtools=panfrost -Ddri-search-path=/system/lib '
+                      -Dvulkan-drivers= -Dgbm=enabled -Degl=enabled -Dgles1=enabled -Dgles2=enabled -Dcpp_rtti=false -Dglx=disabled -Dtools= -Ddri-search-path=/vendor/lib/chipsetsdk '

4.1.3 glmark2测试

ohos下测试glmark2命令为

glmark2-es2-ohos --data-path /lib/glmark2/data

实测中出现eglInitialize 初始化失败，返回0x3001，如下所示。

参考博客OpenHarmony富设备移植指南（6.2）GPU测试程序编译尝试移植native_window_ohos, 使用gpu简单绘制三角形程序，测试看看情况，发现还是不行。

此问题也困扰了很久，只能正面加打印分析源码，分析到dri2_dpy->image_driver->createNewScreen2后出现失败，源码复杂，createNewScreen2不清楚具体调用到哪里，各种查资料中，文章Android环境下Mesa初始化流程重学习之eglInitialize,梳理了eglInitialize，经过实际分析后，得到结论，createNewScreen2 最终会调用panfrost_create_screen函数，在此函数加打印分析，panfrost: Unsupported model 0x620, 源码中debug_printf(“panfrost: Unsupported model %X”, dev->gpu_id); debug没有放开，导致关键出错信息没有显示。大致分析出gpu_id不支持。分析源码最终发现src/panfrost/lib/pan_props.c中panfrost_model_list没有T620。

/* Table of supported Mali GPUs */
const struct panfrost_model panfrost_model_list[] = {
        MODEL(0x720, "T720", "T72x", NO_ANISO, 8192, { .no_hierarchical_tiling = true }),
        MODEL(0x750, "T760", "T76x", NO_ANISO, 8192, {}),
        MODEL(0x820, "T820", "T82x", NO_ANISO, 8192, { .no_hierarchical_tiling = true }),
        MODEL(0x830, "T830", "T83x", NO_ANISO, 8192, { .no_hierarchical_tiling = true }),
        MODEL(0x860, "T860", "T86x", NO_ANISO, 8192, {}),
        MODEL(0x880, "T880", "T88x", NO_ANISO, 8192, {}),

        MODEL(0x6000, "G71", "TMIx", NO_ANISO, 8192, {}),
        MODEL(0x6221, "G72", "THEx", 0x0030 /* r0p3 */, 16384, {}),
        MODEL(0x7090, "G51", "TSIx", 0x1010 /* r1p1 */, 16384, {}),
        MODEL(0x7093, "G31", "TDVx", HAS_ANISO, 16384, {}),
        MODEL(0x7211, "G76", "TNOx", HAS_ANISO, 16384, {}),
        MODEL(0x7212, "G52", "TGOx", HAS_ANISO, 16384, {}),
        MODEL(0x7402, "G52 r1", "TGOx", HAS_ANISO, 16384, {}),
        MODEL(0x9093, "G57", "TNAx", HAS_ANISO, 16384, {}),
};

并且在docs/drivers/panfrost.rst中提示Other Midgard and Bifrost chips (T604, T628, G71) are not yet supported，此版本暂不支持T620 GPU。

ohos下third_party/mesa3d源码和buildroot-2025.02中mesa3d源码进行对比，ohos下mesa3d版本为mesa-22.2.4，buildroot-2025.02中版本为mesa-24.0.9，版本差异较大，以及查看src/panfrost/lib/pan_props.c中panfrost_model_list和docs/drivers/panfrost.rst发现已经支持T620 GPU。

此时git clone了一份mesa3d官方源码查看src/panfrost/lib/pan_props.c修改记录，如下

另外查看这笔提交只修改了panfrost_model_list，没有其他文件修改，另外再git reset --hard到这一笔提交，查看mesa3d版本为22.3.0-devel，和hos下mesa3d版本为mesa-22.2.4比较接近。因此尝试在ohos源码下仅修改src/panfrost/lib/pan_props.c panfrost_model_list测试。

修改如下：

diff --git a/src/panfrost/lib/pan_props.c b/src/panfrost/lib/pan_props.c
old mode 100644
new mode 100755
index c013ee4edb1..5de9a288bf1
--- a/src/panfrost/lib/pan_props.c
+++ b/src/panfrost/lib/pan_props.c
@@ -54,6 +54,7 @@

 /* Table of supported Mali GPUs */
 const struct panfrost_model panfrost_model_list[] = {
+        MODEL(0x620, "T620", "T62x", NO_ANISO, 8192, {}),
         MODEL(0x720, "T720", "T72x", NO_ANISO, 8192, { .no_hierarchical_tiling = true }),
         MODEL(0x750, "T760", "T76x", NO_ANISO, 8192, {}),
         MODEL(0x820, "T820", "T82x", NO_ANISO, 8192, { .no_hierarchical_tiling = true }),

实测比较幸运，native_window_ohos绘制三角形程序和glmark2-es2-ohos正常了。

native_window_ohos绘制三角形程序效果如下：

glmark2-es2-ohos显示效果和buildroot下glmark2-es2-drm一样。

4.2 ohos使能gpu

4.2.1 使能gpu

参考视频中主要做以下修改。

vendor中对应config.json中graphic_2d参数使能gpu，如下所示

            "graphic_2d_feature_ace_enable_gpu = true",
            "graphic_2d_feature_rs_enable_eglimage = true"

third_mesa3d 修改如下：

diff --git a/src/egl/main/eglapi.c b/src/egl/main/eglapi.c
old mode 100644
new mode 100755
index c1a1d8540af..831d0b258ee
--- a/src/egl/main/eglapi.c
+++ b/src/egl/main/eglapi.c

@@ -1764,7 +1794,7 @@ _eglCreateImageCommon(_EGLDisplay *disp, EGLContext ctx, EGLenum target,
    RETURN_EGL_EVAL(disp, ret);
 }

-static EGLImage EGLAPIENTRY
+EGLAPI EGLImage EGLAPIENTRY
 eglCreateImageKHR(EGLDisplay dpy, EGLContext ctx, EGLenum target,
                   EGLClientBuffer buffer, const EGLint *attr_list)
 {
@@ -1820,7 +1850,7 @@ eglDestroyImage(EGLDisplay dpy, EGLImage image)
    return _eglDestroyImageCommon(disp, img);
 }

-static EGLBoolean EGLAPIENTRY
+EGLAPI EGLBoolean EGLAPIENTRY
 eglDestroyImageKHR(EGLDisplay dpy, EGLImage image)
 {
    _EGLDisplay *disp = _eglLockDisplay(dpy);
@@ -1896,7 +1926,7 @@ _eglCreateSync(_EGLDisplay *disp, EGLenum type, const EGLAttrib *attrib_list,
 }


-static EGLSync EGLAPIENTRY
+EGLAPI EGLSync EGLAPIENTRY
 eglCreateSyncKHR(EGLDisplay dpy, EGLenum type, const EGLint *int_list)
 {
    _EGLDisplay *disp = _eglLockDisplay(dpy);
@@ -1970,7 +2000,7 @@ eglDestroySync(EGLDisplay dpy, EGLSync sync)
    return _eglDestroySync(disp, s);
 }

-static EGLBoolean EGLAPIENTRY
+EGLAPI EGLBoolean EGLAPIENTRY
 eglDestroySyncKHR(EGLDisplay dpy, EGLSync sync)
 {
    _EGLDisplay *disp = _eglLockDisplay(dpy);
@@ -2024,7 +2054,7 @@ eglClientWaitSync(EGLDisplay dpy, EGLSync sync,
    return _eglClientWaitSyncCommon(disp, dpy, s, flags, timeout);
 }

-static EGLint EGLAPIENTRY
+EGLAPI EGLint EGLAPIENTRY
 eglClientWaitSyncKHR(EGLDisplay dpy, EGLSync sync,
                      EGLint flags, EGLTime timeout)
 {
@@ -2059,7 +2089,7 @@ _eglWaitSyncCommon(_EGLDisplay *disp, _EGLSync *s, EGLint flags)
    RETURN_EGL_EVAL(disp, ret);
 }

-static EGLint EGLAPIENTRY
+EGLAPI EGLint EGLAPIENTRY
 eglWaitSyncKHR(EGLDisplay dpy, EGLSync sync, EGLint flags)
 {
    _EGLDisplay *disp = _eglLockDisplay(dpy);
@@ -2129,7 +2159,7 @@ eglGetSyncAttrib(EGLDisplay dpy, EGLSync sync, EGLint attribute, EGLAttrib *valu
 }


-static EGLBoolean EGLAPIENTRY
+EGLAPI EGLBoolean EGLAPIENTRY
 eglGetSyncAttribKHR(EGLDisplay dpy, EGLSync sync, EGLint attribute, EGLint *value)
 {
    _EGLDisplay *disp = _eglLockDisplay(dpy);
@@ -2156,7 +2186,7 @@ eglGetSyncAttribKHR(EGLDisplay dpy, EGLSync sync, EGLint attribute, EGLint *valu
    return result;
 }

-static EGLint EGLAPIENTRY
+EGLAPI EGLint EGLAPIENTRY
 eglDupNativeFenceFDANDROID(EGLDisplay dpy, EGLSync sync)
 {
    _EGLDisplay *disp = _eglLockDisplay(dpy);

foundation/graphic/graphic_2d修改如下：

diff --git a/frameworks/surfaceimage/src/surface_image.cpp b/frameworks/surfaceimage/src/surface_image.cpp
index a4839f88b5..485f746d7d 100644
--- a/frameworks/surfaceimage/src/surface_image.cpp
+++ b/frameworks/surfaceimage/src/surface_image.cpp
@@ -160,9 +160,9 @@ SurfaceError SurfaceImage::UpdateSurfaceImage()

     uint32_t seqNum = buffer->GetSeqNum();
     BLOGI("seqNum %{public}d", seqNum);
-    EGLImageKHR img = imageCacheSeqs_[seqNum].eglImage_;
+    //EGLImageKHR img = imageCacheSeqs_[seqNum].eglImage_;
     glBindTexture(textureTarget_, textureId_);
-    glEGLImageTargetTexture2DOES(textureTarget_, static_cast<GLeglImageOES>(img));
+    //glEGLImageTargetTexture2DOES(textureTarget_, static_cast<GLeglImageOES>(img));

     while (glGetError() != GL_NO_ERROR) {
         BLOGE("glEGLImageTargetTexture2DOES error");
diff --git a/graphic_config.gni b/graphic_config.gni
index e156025e5b..7ad24df2bc 100644
--- a/graphic_config.gni
+++ b/graphic_config.gni
@@ -13,7 +13,7 @@

 declare_args() {
   graphic_2d_feature_bootanimation_enable = true
-  graphic_2d_feature_ace_enable_gpu = false
+  graphic_2d_feature_ace_enable_gpu = true
   graphic_2d_feature_color_gamut_enable = false
   graphic_2d_feature_rs_enable_eglimage = false
   graphic_2d_feature_rs_enable_uni_render = false
diff --git a/rosen/modules/render_service_base/src/pipeline/rs_draw_cmd.cpp b/rosen/modules/render_service_base/src/pipeline/rs_draw_cmd.cpp
index 77c40da430..04f832e1f9 100644
--- a/rosen/modules/render_service_base/src/pipeline/rs_draw_cmd.cpp
+++ b/rosen/modules/render_service_base/src/pipeline/rs_draw_cmd.cpp
@@ -785,7 +785,7 @@ sk_sp<SkImage> ImageWithParmOpItem::GetSkImageFromSurfaceBuffer(SkCanvas& canvas
         glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR);
         glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
         glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
-        glEGLImageTargetTexture2DOES(GL_TEXTURE_2D, static_cast<GLeglImageOES>(eglImage_));
+        //glEGLImageTargetTexture2DOES(GL_TEXTURE_2D, static_cast<GLeglImageOES>(eglImage_));
     }

     GrGLTextureInfo textureInfo = { GL_TEXTURE_2D, texId_, GL_RGBA8_OES };
@@ -1027,7 +1027,7 @@ void SurfaceBufferOpItem::Draw(RSPaintFilterCanvas& canvas, const SkRect*) const
     glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR);
     glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
     glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
-    glEGLImageTargetTexture2DOES(GL_TEXTURE_2D, static_cast<GLeglImageOES>(eglImage_));
+    //glEGLImageTargetTexture2DOES(GL_TEXTURE_2D, static_cast<GLeglImageOES>(eglImage_));

     // restore
     glBindTexture(GL_TEXTURE_2D, originTexture);
(END)

     ret = eglChooseConfig(eglDisplay_, config_attribs, &config_, 1, &count);
     if (!(ret && static_cast<unsigned int>(count) >= 1)) {
diff --git a/rosen/modules/render_service_base/src/pipeline/rs_draw_cmd.cpp b/rosen/modules/render_service_base/src/pipeline/rs_draw_cmd.cpp
index 77c40da430..04f832e1f9 100644
--- a/rosen/modules/render_service_base/src/pipeline/rs_draw_cmd.cpp
+++ b/rosen/modules/render_service_base/src/pipeline/rs_draw_cmd.cpp
@@ -785,7 +785,7 @@ sk_sp<SkImage> ImageWithParmOpItem::GetSkImageFromSurfaceBuffer(SkCanvas& canvas
         glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR);
         glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
         glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
-        glEGLImageTargetTexture2DOES(GL_TEXTURE_2D, static_cast<GLeglImageOES>(eglImage_));
+        //glEGLImageTargetTexture2DOES(GL_TEXTURE_2D, static_cast<GLeglImageOES>(eglImage_));
     }

     GrGLTextureInfo textureInfo = { GL_TEXTURE_2D, texId_, GL_RGBA8_OES };
@@ -1027,7 +1027,7 @@ void SurfaceBufferOpItem::Draw(RSPaintFilterCanvas& canvas, const SkRect*) const
     glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR);
     glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
     glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
-    glEGLImageTargetTexture2DOES(GL_TEXTURE_2D, static_cast<GLeglImageOES>(eglImage_));
+    //glEGLImageTargetTexture2DOES(GL_TEXTURE_2D, static_cast<GLeglImageOES>(eglImage_));

     // restore
     glBindTexture(GL_TEXTURE_2D, originTexture);
~

4.2.2 使能gpu后开机启动黑屏问题解决

上述修改使能gpu后开机启动出现黑屏，rendor_serivce 会异常退出，通过命令敲hilog查看日志，敲命令感觉滞后看不到有效信息，非常需要将开机过程中的日志存储到文件系统导出来分析。参考hilog落盘修改后，日志能够正常存储。

分析日志有报错：Failed to eglChooseConfig

查看芯片手册后，分析gpu不支持EGL_OPENGL_ES3_BIT，改为EGL_OPENGL_ES2_BIT后eglChooseConfig执行正常。
InitializeEglContext初始正常。

2DGraphics: InitializeEglContext: Create EGL context successfully, version 1.4

但后续流程出现报错：2DGraphics: SetUpGrContext: SetUpGrContext failed to make native interface

分析源码，SetUpGrContext glInterface函数会调用到third_party/skia源码中，在 skia中分析GrGLMakeAssembledInterface函数const char* verStr = reinterpret_cast<const char*>(GetString(GR_GL_VERSION)); 返回为null， 分析mesa3d _mesa_GetString( GLenum name )函数，参考https://laval.csdn.net/user/discuss/668bd1d35c462a3f4fd478bf 解决方式

+++ b/meson.build
@@ -505,7 +505,7 @@ foreach platform : _platforms
  pre_args += '-DHAVE_@0@_PLATFORM'.format(platform.to_upper())
 endforeach

-if with_platform_android and get_option('platform-sdk-version') >= 29
+if with_platform_ohos or with_platform_android and get_option('platform-sdk-version') >= 29

其本质是增加了

c_args += ‘-fno-emulated-tls’
cpp_args += ‘-fno-emulated-tls’

此问题已解决，开机能够进入桌面。

后续遇到使能gpu后内存不足问题，具体解决见zram内存压缩 mkswap报错。

至此，mesa3d适配完成，过程困难重重，希望后续他人能够避免踩坑。