设备树覆盖手柄

Question

我正在研究带有双 Cortex-A9 处理器的 Altera Cyclone V SOC FPGA。嵌入式系统（linux-socfpga 4.16）是用Buildroot-2018.05创建的。

我在处理器组件的启动时使用 "top" 设备树，并使用设备树覆盖来配置组件的 FPGA 部分并加载相关的驱动程序。叠加层将附加到顶部 DT 的 base_fpga_region。

顶级设备树

/dts-v1/;

/ {
    model = "MY_PROJECT";   /* appended from boardinfo */
    compatible = "altr,socfpga-cyclone5", "altr,socfpga";   /* appended from boardinfo */
    #address-cells = <1>;
    #size-cells = <1>;

    cpus {
        [...]
    }; //end cpus

    memory {
        device_type = "memory";
        reg = <0xffff0000 0x00010000>,
            <0x00000000 0x80000000>;
    }; //end memory

    reserved-memory {
        #address-cells = <1>;
        #size-cells = <1>;
        ranges;

        mem_dma_reserved {
            compatible = "shared-dma-pool";
            no-map;
            reg = <0x78000000 0x8000000>;
        };
    };

    soc: soc {
        device_type = "soc";
        ranges;
        #address-cells = <1>;
        #size-cells = <1>;
        compatible = "altr,avalon", "simple-bus";
        bus-frequency = <0>;

        fpgabridge1: fpgabridge@1 {
            compatible = "altr,socfpga-lwhps2fpga-bridge";
            resets = <&hps_rstmgr 97>;  /* appended from boardinfo */
            clocks = <&l4_main_clk>;    /* appended from boardinfo */           
            #address-cells = <1>;
            #size-cells = <1>;
            ranges;
            bridge-enable = <1>;

            label = "lwhps2fpga";
            reset-names = "lwhps2fpga";
            reg = <0xff200000 0x200000>;
            reg-names = "axi_h2f_lw";
        }; //end fpgabridge@1 (fpgabridge1)

        base_fpga_region: base_fpga_region  {
            compatible = "fpga-region";
            #address-cells = <0x2>;
            #size-cells = <0x1>;
            fpga-mgr = <&hps_fpgamgr>;
            fpga-bridges = <&fpgabridge1>;          
            ranges = <0x0 0x0 0xff200000 0x200000>;
        }; //end base_fpga_region (base_fpga_region)

etc......

设备树覆盖

/dts-v1/ /plugin/; 

/{ 
    fragment@0 { 
        target-path = "/soc/base_fpga_region"; 
        #address-cells = <2>; 
        #size-cells = <1>; 
        __overlay__ { 
            #address-cells = <2>; 
            #size-cells = <1>; 

            firmware-name = "my_project.rbf";

            my_dma_0: dma@0x000000000 {
                compatible = "my_company,my_dma-0.1";
                reg = <0x00000000 0x0000000 0x00000014>;
                memory-region = <&mem_dma_reserved>;
            }; //end dma@0x000000000 (my_dma_0)      
        }; 
    }; 
};

我的问题是 link 从顶部 DT 到覆盖层中的 memory-region mem_dma_reserved。

我假设在使用 -@ 选项将 dts 转换为 dtbo 时，叠加层应使用 __fixups__ 选项获得 mem_dma_reserved 的 phandle。我已经创建了 dtbo 文件并在 dts 中再次转换它以查看编译过程中做了什么：

dtc -@ -I dts -O dtb -o overlay.dtbo overlay.dts
dtc -I dtb -O dts -o overlay_recovery.dts overlay.dtbo

设备树覆盖重新生成

/dts-v1/;    
/ {

    fragment@0 {
        target-path = "/soc/base_fpga_region";
        #address-cells = <0x2>;
        #size-cells = <0x1>;

        __overlay__ {
            #address-cells = <0x2>;
            #size-cells = <0x1>;
            firmware-name = "my_project.rbf";

            dma@0x000000000 {
                compatible = "my_company,my_dma-0.1";
                reg = <0x0 0x0 0x14>;
                memory-region = <0xffffffff>;  // phandle converted to 0xffffffff, cannot resolve unless the __fixup__ function does it.
                linux,phandle = <0x2>;
                phandle = <0x2>;
            };
        };
    };

    __symbols__ {
        my_dma_0 = "/fragment@0/__overlay__/dma@0x000000000";
    };

    __fixups__ {
        mem_dma_reserved = "/fragment@0/__overlay__/dma@0x000000000:memory-region:0";
    };
};

我们可以看到内存区域的 phandle 是 0xFFFFFFFF，因为叠加层不知道 <&mem_dma_reserved> 节点。修复部分应该能够在加载时取回 phandle，但它不工作，我得到这个错误：

[ 27.434730] OF: resolver: of_resolve_phandles: no symbols in root of device tree.
[ 27.440401] OF: resolver: overlay phandle fixup failed: -22
[ 27.445993] create_overlay: Failed to resolve tree

我在顶部 DT 上进行了从 dtb 到 dts 的相同再生。我看到保留内存的phandle其实是0x6。我在设备树覆盖层中写了 <0x6> 而不是 <&mem_dma_reserved> 并且使用此配置，一切都加载了！

如何让叠加层自动找到 <&mem_dma_reserved> 而无需手动操作？

编辑

正如 ikwzm 所指出的，我在顶部设备树中添加了以下几行：

reserved-memory {
    #address-cells = <1>;
    #size-cells = <1>;
    ranges;

    mem_dma_reserved: mem_dma_reserved { // add a label
        compatible = "shared-dma-pool";
        no-map;
        reg = <0x78000000 0x8000000>;
    };
};    

[...]    

// add the whole symbol block
__symbols__ { 
    mem_dma_reserved = "/reserved-memory/mem_dma_reserved ";
};

现在错误消失了，但是：

我期待在操作期间加载 my_dma 的驱动程序。

我检查了设备树覆盖是否很好地考虑到了：

ls /sys/firmware/devicetree/base/soc/base_fpga_region/
my_dma_0
other_stuff

cat /sys/firmware/devicetree/base/soc/base_fpga_region/my_dma_0/memory-region
//nothing//

似乎没有附加内存区域。

我错过了什么？

Answer 1

制作top dtb时，是否在dtc命令中附加了选项--symbol（或-@）？

给mem_dma_reserved添加一个标签（符号）如下。

reserved-memory {
    #address-cells = <1>;
    #size-cells = <1>;
    ranges;

    mem_dma_reserved: mem_dma_reserved {
        compatible = "shared-dma-pool";
        no-map;
        reg = <0x78000000 0x8000000>;
    };
};

使用--symbol选项dtc命令创建dtb时，添加__symbols__{...};如下，mem_dma_reserved = "/reserved-memory/ mem_dma_reserved";应该在里面

__symbols__ {
    :
    :
    :
    mem_dma_reserved = "/reserved-memory/mem_dma_reserved";
    usb_phy0 = "/phy0";
};

设备树覆盖手柄

device tree overlay phandle

linux-kernel

device-tree

顶级设备树

设备树覆盖

设备树覆盖重新生成

编辑