图形驱动框架
整体框架
驱动方案选择
K3 平台支持两种互斥的 PowerVR GPU 驱动方案：
- 闭源方案（推荐）：性能更强，功能完整，支持 DVFS 和温控
- 开源方案：基于 Mesa 和开源内核驱动，内核版本需 >= 6.12，Mesa 版本需 >= 26.1.0
闭源驱动方案
Kernel 配置
进入 Linux Kernel 源码目录，在 config 中启用 PowerVR Rogue：
CONFIG_POWERVR_ROGUE=y # 启用 PowerVR Rogue
或通过 make menuconfig 在 Device Drivers -> Graphics support -> Direct Rendering Manager 路径下启用：
<*> PowerVR GPU
<*> PowerVR GPU DVFS
<*> PowerVR GPU Thermal
< > null drm disp
设备树（DTS）配置：
imggpu: imggpu@cac00000 {
compatible = "img,rgx";
interrupt-names = "rgxirq";
interrupt-parent = <&saplic>;
interrupts = <75 IRQ_TYPE_LEVEL_HIGH>;
reg = <0x0 0xcac00000 0x0 0x80000>;
reg-names = "rgxregs";
power-domains = <&power K3_PMU_GPU_PWR_DOMAIN>;
clocks = <&syscon_apmu CLK_APMU_GPU>;
clock-names = "clk_rgx";
resets = <&syscon_apmu RESET_APMU_GPU>;
#cooling-cells = <2>;
thermal-zone = "thermal_gpu";
status = "okay";
};
PVR DDK 配置
由于版权限制，PVR GPU 的 DDK（Device Development Kit）无法直接提供。用户层闭源代码以 so 动态库形式提供 OpenGLES、Vulkan 和 OpenCL API。
Buildroot
编译 buildroot 时，在配置文件中启用：
BR2_PACKAGE_IMG_GPU_POWERVR=y
或通过 make menuconfig 手动勾选：
External options -> Bianbu config -> img-gpu-powervr
[*] rtk hciattach
*** spacemit mpp package ***
-*- spacemit mpp
[*] img-gpu-powervr # 启用 img-gpu-powervr
Output option (Wayland) --->
[ ] install examples
Bianbu
闭源 GPU 代码以 .so 动态库形式通过 img-gpu-powervr deb 包提供：
sudo apt install img-gpu-powervr
注意：GPU 驱动分为内核态驱动和用户态驱动，二者版本必须保持一致，否则可能导致 GPU 功能异常。
查看版本方法：
# 查看内核层驱动
➜ ~ journalctl -b | grep "Initialized pvr"
[drm] Initialized pvr 24.2.6603887 for cac00000.imggpu on minor 0 # 版本为24.2
# 查看用户层驱动
➜ ~ dpkg -s img-gpu-powervr
Package: img-gpu-powervr
Status: install ok installed
Priority: optional
Section: graphics
Installed-Size: 70314
Maintainer: bianbu <bo.deng@spacemit.com>
Architecture: all
Version: 24.2-6603887bb1 # 版本为24.2
Mesa3D 配置
Mesa3D 提供 OpenGL ES API 接口及硬件加速渲染实现。
Buildroot
在配置文件中启用以下选项：
Symbol: BR2_PACKAGE_MESA3D [=y]
Symbol: BR2_PACKAGE_MESA3D_DRIVER [=y]
Symbol: BR2_PACKAGE_MESA3D_GALLIUM_DRIVER [=y]
Symbol: BR2_PACKAGE_MESA3D_GALLIUM_DRIVER_PVR [=y]
Symbol: BR2_PACKAGE_MESA3D_GBM [=y]
Symbol: BR2_PACKAGE_MESA3D_OPENGL_EGL [=y]
或在 make menuconfig 中手动勾选： Target packages -> Graphic libraries and applications -> mesa3d
-*- Gallium pvr driver
*** Gallium VDPAU state tracker needs X.org and gallium drivers r300, r600, radeonsi or nouveau ***
*** Vulkan drivers ***
*** Off-screen Rendering ***
[ ] OSMesa (Gallium) library
*** OpenGL API Support ***
-*- gbm
*** OpenGL GLX support needs X11 ***
-*- OpenGL EGL
[ ] OpenGL ES
Bianbu
按以下顺序安装 Mesa deb 包：
sudo apt install \
libgl1-mesa-dev \
libegl1-mesa \
libegl1-mesa-dev \
libglapi-mesa \
libgbm1 \
libgbm-dev \
libegl-mesa0 \
libgl1-mesa-dri \
libgles2-mesa \
libgles2-mesa-dev \
libgl1-mesa-glx \
libglx-mesa0 \
libosmesa6 \
libosmesa6-dev \
libwayland-egl1-mesa \
mesa-common-dev \
libglvnd
开源驱动方案
Kernel 配置
内核版本需 >= 6.12。在 config 中启用 PowerVR DRM 驱动：
CONFIG_DRM_POWERVR=y # 启用 PowerVR DRM 驱动
或通过 make menuconfig 在 Device Drivers -> Graphics support -> Direct Rendering Manager 路径下启用：
<*> Imagination Technologies PowerVR (Series 6 and later) & IMG Graphics
< > PowerVR GPU
< > null drm disp
适配 k3 平台
为适配 SpacemiT K3 平台，需对内核驱动做如下修改：
index 571b56a11161..636f8ab3a54f 100644
--- a/arch/riscv/boot/dts/spacemit/k3.dtsi
+++ b/arch/riscv/boot/dts/spacemit/k3.dtsi
@@ -1659,7 +1659,7 @@ gmac_axi_setup: stmmac-axi-config {
};
imggpu: imggpu@cac00000 {
- compatible = "img,rgx";
+ compatible = "img,img-rogue";
interrupt-names = "rgxirq";
interrupt-parent = <&saplic>;
interrupts = <75 IRQ_TYPE_LEVEL_HIGH>;
@@ -1667,11 +1667,12 @@ imggpu: imggpu@cac00000 {
reg-names = "rgxregs";
power-domains = <&power K3_PMU_GPU_PWR_DOMAIN>;
clocks = <&syscon_apmu CLK_APMU_GPU>;
- clock-names = "clk_rgx";
+ clock-names = "core";
resets = <&syscon_apmu RESET_APMU_GPU>;
status = "okay";
+ img,core-clock-rate-hz = <1228000000>;
};
diff --git a/drivers/gpu/drm/imagination/pvr_device.c b/drivers/gpu/drm/imagination/pvr_device.c
index 78d6b8a0a450..8e490b3921f7 100644
--- a/drivers/gpu/drm/imagination/pvr_device.c
+++ b/drivers/gpu/drm/imagination/pvr_device.c
@@ -99,6 +99,9 @@ static int pvr_device_clk_init(struct pvr_device *pvr_dev)
struct clk *core_clk;
struct clk *sys_clk;
struct clk *mem_clk;
+ struct device *dev = drm_dev->dev;
+ u32 core_clk_rate = PVR_CORE_CLK_RATE_HZ;
+ int err;
core_clk = devm_clk_get(drm_dev->dev, "core");
if (IS_ERR(core_clk))
@@ -115,6 +118,22 @@ static int pvr_device_clk_init(struct pvr_device *pvr_dev)
return dev_err_probe(drm_dev->dev, PTR_ERR(mem_clk),
"failed to get mem clock\n");
+ /* Optional: set core clock rate from device tree or fallback macro. */
+ if (!of_property_read_u32(dev->of_node, "img,core-clock-rate-hz", &core_clk_rate) &&
+ core_clk_rate > 0) {
+ err = clk_set_rate(core_clk, core_clk_rate);
+ if (err)
+ return dev_err_probe(dev, err,
+ "failed to set core clock rate (%u Hz)\n",
+ core_clk_rate);
+ } else if (core_clk_rate > 0) {
+ err = clk_set_rate(core_clk, core_clk_rate);
+ if (err)
+ return dev_err_probe(dev, err,
+ "failed to set core clock rate (%u Hz)\n",
+ core_clk_rate);
+ }
+
pvr_dev->core_clk = core_clk;
pvr_dev->sys_clk = sys_clk;
pvr_dev->mem_clk = mem_clk;
diff --git a/drivers/gpu/drm/imagination/pvr_device.h b/drivers/gpu/drm/imagination/pvr_device.h
index ec53ff275541..419d4bbf2df2 100644
--- a/drivers/gpu/drm/imagination/pvr_device.h
+++ b/drivers/gpu/drm/imagination/pvr_device.h
@@ -68,6 +68,16 @@ struct pvr_device_data {
const struct pvr_power_sequence_ops *pwr_ops;
};
+#ifndef PVR_CORE_CLK_RATE_HZ
+#define PVR_CORE_CLK_RATE_HZ 1228 * 1000 * 1000U
+#endif
+
/**
* struct pvr_device - powervr-specific wrapper for &struct drm_device
*/
diff --git a/drivers/gpu/drm/imagination/pvr_drv.c b/drivers/gpu/drm/imagination/pvr_drv.c
index 916b40ced7eb..4306c182d99d 100644
--- a/drivers/gpu/drm/imagination/pvr_drv.c
+++ b/drivers/gpu/drm/imagination/pvr_drv.c
@@ -1532,4 +1534,5 @@ MODULE_LICENSE("Dual MIT/GPL");
MODULE_IMPORT_NS("DMA_BUF");
MODULE_FIRMWARE("powervr/rogue_33.15.11.3_v1.fw");
MODULE_FIRMWARE("powervr/rogue_36.52.104.182_v1.fw");
+MODULE_FIRMWARE("powervr/rogue_36.56.104.183_v1.fw");
MODULE_FIRMWARE("powervr/rogue_36.53.104.796_v1.fw");
固件文件：rogue_36.56.104.183_v1.fw 固件文件需放置在 /lib/firmware/powervr/ 目录。
Mesa3D 配置
Buildroot
最新 Mesa 已支持 IMG BXM-4-64 GPU。下载 Mesa 源码 覆盖 buildroot-k3/package-src/mesa 后编译即可。
Bianbu
确保系统中安装的 Mesa deb 包版本>=26.1.0。例如：
sudo sudo apt install libgl1-mesa-dev=26.1.0-2ubuntu1
闭源驱动的高级配置
注意：本部分仅适用于闭源驱动方案。开源驱动不支持 DVFS、温控和 BlobCache。
BlobCache 使用
GPU 的 BlobCache 是一种用于存储和重用 GPU 编译结果的数据缓存机制。这种缓存机制旨在提高性能，减少重复计算和数据传输的时间，从而加快渲染速度。
配置方法
打开配置文件/etc/powervr.ini并设置如下, 即可启用 BlobCache 功能：
[default]
EnableBlobCache=1 # 启用 BlobCache 功能
[mpv]
EnableWorkaroundMPVScreenTearing=1
[totem]
EnableWorkaroundMPVScreenTearing=1
[gst-launch-1.0]
EnableWorkaroundMPVScreenTearing=1
GPU DVFS
K3 平台的 GPU 内核驱动通过 Linux devfreq 框架实现了 DVFS（动态频率电压调节），默认使用 simple_ondemand 调频策略，根据 GPU 负载动态调整频率（硬件实现上仅支持动态调频）。
支持以下频率档位，通过 DTS 中的 OPP（Operating Performance Points）表进行配置：
| 档位 | 频率 |
|---|---|
| 1 | 409 MHz |
| 2 | 491 MHz |
| 3 | 600 MHz |
| 4 | 614 MHz |
| 5 | 750 MHz |
| 6 | 819 MHz |
| 7 | 1000 MHz |
| 8 | 1228 MHz |
DTS 配置示例（k3.dtsi）：
imggpu: imggpu@cac00000 {
/* ... 其他属性 ... */
clocks = <&syscon_apmu CLK_APMU_GPU>;
clock-names = "clk_rgx";
resets = <&syscon_apmu RESET_APMU_GPU>;
#cooling-cells = <2>;
thermal-zone = "thermal_gpu";
operating-points-v2 = <&imggpu_opp_table>;
status = "okay";
imggpu_opp_table: opp-table {
compatible = "operating-points-v2";
opp-409000000 {
opp-hz = /bits/ 64 <409000000>;
opp-microvolt = <800000>; /* OPP 框架要求的固定值，实际仅调频 */
clock-latency-ns = <100000>;
};
opp-614000000 {
opp-hz = /bits/ 64 <614000000>;
opp-microvolt = <900000>;
clock-latency-ns = <100000>;
};
opp-819000000 {
opp-hz = /bits/ 64 <819000000>;
opp-microvolt = <900000>;
clock-latency-ns = <100000>;
};
opp-1000000000 {
opp-hz = /bits/ 64 <1000000000>;
opp-microvolt = <950000>;
clock-latency-ns = <100000>;
};
opp-1228000000 {
opp-hz = /bits/ 64 <1228000000>;
opp-microvolt = <950000>;
clock-latency-ns = <100000>;
};
};
};
可通过修改 DTS 中的 OPP 表实现频率档位的定制化。
DVFS 调试
可通过 devfreq 节点查看和调整 GPU 频率：
# 查看当前频率
cat /sys/class/devfreq/cac00000.imggpu/cur_freq
# 查看可用频率
cat /sys/class/devfreq/cac00000.imggpu/available_frequencies
# 查看当前调频策略
cat /sys/class/devfreq/cac00000.imggpu/governor
# 查看可用调频策略
cat /sys/class/devfreq/cac00000.imggpu/available_governors
# 配置最小频率为 819MHz
echo 819000000 > /sys/class/devfreq/cac00000.imggpu/min_freq
# 配置最大频率为 1228MHz
echo 1228000000 > /sys/class/devfreq/cac00000.imggpu/max_freq
GPU 温控
K3 平台的 GPU 温控通过 Linux thermal 框架与 devfreq cooling 机制实现。当 GPU 温度达到阈值时，系统会自动降低 GPU 频率以控制温度。
温控策略通过 DTS 中的 thermal zone 配置，包含多级温度阈值和对应的降频映射。以 k3_com260.dts 为例：
| 级别 | 温度阈值 | 回滞温度 | 类型 | 降频等级 |
|---|---|---|---|---|
| gpu_alert0 | 65°C | 2°C | passive | cooling 0~1 |
| gpu_alert1 | 75°C | 2°C | passive | cooling 1~2 |
| gpu_alert2 | 85°C | 2°C | passive | cooling 2~3 |
| gpu_crit | 105°C | 2°C | critical | 关闭 GPU |
DTS 配置示例（k3_com260.dts）：
thermal_gpu {
polling-delay = <1000>;
polling-delay-passive = <500>;
thermal-sensors = <&thermal 3>;
trips {
gpu_alert0: gpu-alert0 {
temperature = <65000>;
hysteresis = <2000>;
type = "passive";
};
gpu_alert1: gpu-alert1 {
temperature = <75000>;
hysteresis = <2000>;
type = "passive";
};
gpu_alert2: gpu-alert2 {
temperature = <85000>;
hysteresis = <2000>;
type = "passive";
};
gpu_crit: gpu-crit {
temperature = <105000>;
hysteresis = <2000>;
type = "critical";
};
};
cooling-maps {
map0 {
trip = <&gpu_alert0>;
cooling-device = <&imggpu 0 1>;
};
map1 {
trip = <&gpu_alert1>;
cooling-device = <&imggpu 1 2>;
};
map2 {
trip = <&gpu_alert2>;
cooling-device = <&imggpu 2 3>;
};
};
};
注意：不同板型的温控阈值可能不同。例如
k3_deb1.dts中的阈值为 80°C / 85°C / 90°C / 105°C（critical），具体以实际板型 DTS 配置为准。
温控调试
可通过 thermal 节点查看 GPU 温度和温控状态：
# 查看 GPU 当前温度
cat /sys/class/thermal/thermal_zone*/type # 找到 thermal_gpu 对应的 zone
cat /sys/class/thermal/thermal_zone*/temp # 查看温度（单位：毫摄氏度）
GPU 节点调试
通过对 GPU 节点的监控，可以实时查看 GPU 的运行状态：
~ su
~ cd /sys/kernel/debug/pvr/
~ watch cat status (按 Ctrl + C 退出)
如下所示：
Every 2.0s: cat /sys/kernel/debug/pvr/status bianbu-pc: Thu Mar 05 10:00:00 2026
Driver Status: OK
Device ID: 0:128
Firmware Status: OK
Server Errors: 0
HWR Event Count: 0
CRR Event Count: 0
SLR Event Count: 0
WGP Error Count: 0
TRP Error Count: 0
FWF Event Count: 0
APM Event Count: 2026
GPU Utilisation: 0%
DM Utilisation: VM0
2D: 0%
GEOM: 0%
3D: 0%
CDM: 0%
RAY: 0%
GEOM2: 0%