【分享】使用 AXI performance monitors (APM)测试MPSoC DDR访问带宽

MPSoC的DDR控制器的数据通道上集成了 AXI performance monitors (APM)。具体情况，可以参考Xilinx UG1085 (v2.2)中Chapter 15的“Figure 15‐1: PS Interconnect”或者Figure 17-1。

1. APM驱动程序

对于AXI performance monitors (APM)，xilinx提供的驱动程序是drivers\uio\uio_xilinx_apm.c，相关匹配字符串和主要数据结构如下：

static const struct of_device_id xapm_of_match[] = {
        { .compatible = "xlnx,axi-perf-monitor", },
        { /* end of table*/ }
};

MODULE_DEVICE_TABLE(of, xapm_of_match);
static struct platform_driver xapm_driver = {
        .driver = {
                 .name = "xilinx-axipmon",
                 .of_match_table = xapm_of_match,
                 .pm = &xapm_dev_pm_ops,
        },
        .probe = xapm_probe,
        .remove = xapm_remove,
};

APM驱动程序最终会注册UIO设备。
在Linux内核里，需要使能UIO_XILINX_APM。

在每个单板的device tree里，会支持DDR的APM。比如，在VCU TRD里，为DDR的APM定义了下面字段。

         perf-monitor@fd0b0000 {
                    compatible = "xlnx,axi-perf-monitor";
                    reg = <0x00 0xfd0b0000 0x00 0x10000>;
                    interrupts = <0x00 0x7b 0x04>;
                    interrupt-parent = <0x04>;
                    xlnx,enable-profile = <0x00>;
                    xlnx,enable-trace = <0x00>;
                    xlnx,num-monitor-slots = <0x06>;
                    xlnx,enable-event-count = <0x01>;
                    xlnx,enable-event-log = <0x00>;
                    xlnx,have-sampled-metric-cnt = <0x01>;
                    xlnx,num-of-counters = <0x0a>;
                    xlnx,metric-count-width = <0x20>;
                    xlnx,metrics-sample-count-width = <0x20>;
                    xlnx,global-count-width = <0x20>;
                    xlnx,metric-count-scale = <0x01>;
                    clocks = <0x03 0x1c>;
                    phandle = <0x9b>;
         };

     perf_monitor_ddr = "/amba/perf-monitor@fd0b0000";

OCM、CCI等模块也有APM。

     perf_monitor_ocm = "/amba/perf-monitor@ffa00000";
     perf_monitor_cci = "/amba/perf-monitor@fd490000";
     perf_monitor_lpd = "/amba/perf-monitor@ffa10000";

单板Linux启动后，能查询到对应的UIO设备。

root@vcu_trd:~# ls /dev/uio*
/dev/uio0  /dev/uio1  /dev/uio2  /dev/uio3  /dev/uio4

2. APM用户态程序

对于AXI performance monitors (APM)，xilinx也提供了用户态程序。Xilinx的代码在https://github.com/Xilinx/linux-xlnx/blob/master/samples/xilinx_apm/main.c。编译后得到可执行文件，apm_main.elf。使用apm_main.elf，可以查询APM监视到的DDR带宽利用情况。

root@vcu_trd:~# chmod +x apm_main.elf
root@vcu_trd:~# ./apm_main.elf -d /dev/uio0
AXI PMON is in Advanced Mode
port 1 read: 0 0.00 MBps
port 1 write: 0 0.00 MBps
port 2 read: 0 0.00 MBps
port 2 write: 0 0.00 MBps
port 4 read: 0 0.00 MBps
port 4 write: 0 0.00 MBps
port 5 read: 0 0.00 MBps
port 5 write: 0 0.00 MBps
root@vcu_trd:~#
root@vcu_trd:~# ./apm_main.elf -d /dev/uio1
AXI PMON is in Advanced Mode
port 1 read: 12523440 12.52 MBps
port 1 write: 7564856 7.56 MBps
port 2 read: 13290992 13.29 MBps
port 2 write: 7493209 7.49 MBps
port 4 read: 0 0.00 MBps
port 4 write: 331253760 331.25 MBps
port 5 read: 667254864 667.25 MBps
port 5 write: 320656878 320.66 MBps
root@vcu_trd:~# ./apm_main.elf -d /dev/uio2
AXI PMON is in Advanced Mode
port 1 read: 0 0.00 MBps
port 1 write: 0 0.00 MBps
port 2 read: 0 0.00 MBps
port 2 write: 0 0.00 MBps
port 4 read: 0 0.00 MBps
port 4 write: 0 0.00 MBps
port 5 read: 0 0.00 MBps
port 5 write: 0 0.00 MBps
root@vcu_trd:~# ./apm_main.elf -d /dev/uio3
AXI PMON is in Advanced Mode
port 1 read: 0 0.00 MBps
port 1 write: 0 0.00 MBps
port 2 read: 0 0.00 MBps
port 2 write: 0 0.00 MBps
port 4 read: 0 0.00 MBps
port 4 write: 0 0.00 MBps
port 5 read: 0 0.00 MBps
port 5 write: 0 0.00 MBps

上述设备“/dev/uio1”代表DDR APM1。在使用VCU TRD进行1080p编码时，port 4写带宽331.25 MBps，port 5读带宽667.25 MBps，port 5写带宽320.66 MBps。每个DDR port对应的硬件模块，请参考ug1085的Figure 17-1。port 4连接到了HP 1/HP 2。port 5连接到了HP 3。

参考文档： https://xilinx-wiki.atlassian.net/wiki/spaces/A/pages/18842046/APM