1、在用戶設備樹中 pl-custom中

/delete-node/ &axi_dma_0;
/ {
axi_dma_0: dma@a0000000 {
#dma-cells = ;
clock-names = "s_axi_lite_aclk", "m_axi_mm2s_aclk", "m_axi_s2mm_aclk";
clocks = , , ;
compatible = "xlnx,axi-dma-7.1", "xlnx,axi-dma-1.00.a";
interrupt-names = "mm2s_introut", "s2mm_introut";
interrupt-parent = ;
interrupts = ;
reg = ;
xlnx,addrwidth = ;
xlnx,sg-length-width = ;
dma-channel@a0000000 {
compatible = "xlnx,axi-dma-mm2s-channel";
dma-channels = ;
interrupts = ;
xlnx,datawidth = ;
xlnx,device-id = ;
xlnx,include-dre ;
};
dma-channel@a0000030 {
compatible = "xlnx,axi-dma-s2mm-channel";
dma-channels = ;
interrupts = ;
xlnx,datawidth = ;
xlnx,device-id = ;
xlnx,include-dre ;
};
};
};

2 、在用戶system-user中

/include/ "system-conf.dtsi"
/include/ "pl-custom.dtsi"
/ {
axidma_chrdev: axidma_chrdev@0 {
compatible = "xlnx,axidma-chrdev";
dmas = ;
dma-names = "tx_channel", "rx_channel";
};
};

&axi_dma_0 {
dma-coherent;
status = "okay";
};

3、device-tree.bbappend

SRC_URI += "file://system-user.dtsi"
SRC_URI += "file://pl-custom.dtsi"

4 、teminal

petalinux-create -t modules -n xilinx-axidma --enable

5 、github上下載dma驅動，xilinx_axidma-master.zip 并解壓，由于內核的改變，對下列部分文件做出改變。
(1) makefile

DRIVER_NAME = xilinx-axidma
$(DRIVER_NAME)-objs = axi_dma.o axidma_chrdev.o axidma_dma.o axidma_of.o
obj-m := $(DRIVER_NAME).o

(2)xilinx-axidma.bb

SRC_URI = "file://Makefile \
file://axi_dma.c \
file://axidma_chrdev.c \
file://axidma_dma.c \
file://axidma_of.c \
file://axidma.h \
file://axidma_ioctl.h \
file://COPYING \

(3)axi_dma.c

#include
// new

(4)axidma_chrdev.c
linux 內核4.x到5.x，參數減少，查看內核函數，無影響
https://elixir.bootlin.com/linux/v4.9.251/source/include/linux/of_device.h
https://elixir.bootlin.com/linux/v4.9.251/source/include/linux/uaccess.h

在 static bool axidma_access_ok(const void __user *arg, size_t size, bool readonly)函數中

// if (!readonly && !access_ok(VERIFY_WRITE, arg, size)) {
if (!readonly && !access_ok(arg, size)) {

// } else if (!access_ok(VERIFY_READ, arg, size)) {
} else if (!access_ok(arg, size)) {

//of_dma_configure(struct device->dev, NULL);
of_dma_configure(struct device->dev, NULL,true);

(5)axidma_dma.c

#include
// new

static void axidma_dma_callback(void *data)
struct kernel_siginfo sig_info;

6、petalinux-build petalinux-build --sdk petalinux-package --sysroot

7、配置qt的環境。

8 qt 將example中的axidma_benchmark.c復制到qt，并添加其他所需文件。

(1)**.pro
SOURCES += \
gpio.cpp \
libaxidma.c \
main.cpp \
util.c

HEADERS += \
axidma_ioctl.h \
conversion.h \
gpio.h \
libaxidma.h \
util.h

(2)main.cpp() 由于是將C文件改為c++文件，所以此文件中這個函數需要做改變（（void *）和（char *）相同）

static int single_transfer_test(axidma_dev_t dev, int tx_channel, void *tx_buf,
int tx_size, struct axidma_video_frame *tx_frame, int rx_channel,
void *rx_buf, int rx_size, struct axidma_video_frame *rx_frame)
{
int rc;

// Initialize the buffer region we're going to transmit
init_data((char *)tx_buf, (char *)rx_buf, tx_size, rx_size);

// Perform the DMA transaction
rc = axidma_twoway_transfer(dev, tx_channel, (char *)tx_buf, tx_size, tx_frame,
rx_channel, (char *)rx_buf, rx_size, rx_frame, true);
if (rc return rc;
}

// Verify that the data in the buffer changed
return verify_data((char *)tx_buf, (char *)rx_buf, tx_size, rx_size);
}

(3)在所有的 **.h中，為了使用QT的類功能，盡量用C++編程，對C文件的頭文件做如下改變。

#ifdef __cplusplus
extern "C"{
#endif

......

#endif /* UTIL_H_ */

#ifdef __cplusplus
}
#endif

9、加載驅動

root@u3_save:/lib/modules/5.4.0-xilinx-v2020.1/extra# insmod xilinx-axidma.ko
[ 1637.846215] axidma: axidma_dma.c: axidma_dma_init: 725: DMA: Found 1 transmit channels and 1 receive channels.
[ 1637.856240] axidma: axidma_dma.c: axidma_dma_init: 727: VDMA: Found 0 transmit channels and 0 receive channels.

10、測試 因為dma的輸入輸出并沒有還起來，所以只能看到發出去，看不到接收的

root@u3_save:/mnt# ./u3_dma_save
AXI DMA Benchmark Parameters:
Transmit Buffer Size: 7.91 MiB
R[ 1661.465441] axidma axidma: DMA mask not set
eceive Buffer Size: 7.91 MiB
Number of DMA Transfers: 1000 transfers

Using transmit channel 0 and receive channel 1.
[ 1671.710879] axidma: axidma_dma.c: axidma_start_transfer: 309: DMA receive transaction timed out.
[ 1672.719678] xilinx-vdma a0000000.dma: Cannot stop channel 00000000145aa465: 0
Failed to perform the AXI DMA read-write transfer: Timer expired

11、當使用環回時

root@u3_dma_test:/# ./mnt/u3_dma_save
AXI DMA Benchmark Parameters:
Transmit Buffer Size: 7.91 MiB
Receive Buffer Size: 7.91 MiB
Number of DMA Transfers: 1000 transfers

Using transmit channel 0 and receive channel 1.
Single transfer test successfully completed!
Beginning performance analysis of the DMA engine.

DMA Timing Statistics:
Elapsed Time: 41.49 s
Transmit Throughput: 190.66 MiB/s
Receive Throughput: 190.66 MiB/s
Total Throughput: 381.32 MiB/s
root@u3_dma_test:/#

13、qt and opencv 的配置
(1)安裝sdk.sh
(2)在qt的pro中添加

INCLUDEPATH += /home/lcl/Soft/Sdk/sysroots/aarch64-xilinx-linux/usr/include

LIBS += \
-lopencv_imgcodecs \
-lopencv_highgui \
-lopencv_imgproc \
-lopencv_core \

即可。

14、接口改為128

root@u3_dma_test:/# ./mnt/u3_dma_save
AXI DMA Benchmark Parameters:
Transmit Buffer Size: 16.00 MiB
Receive Buffer Size: 16.00 MiB
Number of DMA Transfers: 1000 transfers

Using transmit channel 0 and receive channel 1.
Elapsed Time: 0.25 s
Single transfer test successfully completed!
Beginning performance analysis of the DMA engine.

DMA Timing Statistics:
Elapsed Time: 12.39 s
Transmit Throughput: 1291.32 MiB/s
Receive Throughput: 1291.32 MiB/s
Total Throughput: 2582.65 MiB/s
root@u3_dma_test:/#

15、第一個Elapsed Time: 0.25 s是當我把數據從緩沖區復制到其他位置消耗的時間，也就是16MB的數據需要250ms的時間，速率也就在60M左右，也驗證以前的貼子，dma的實現方式（UIO或者內核驅動）與數據從緩沖區copy到別的地方的速率是沒有關系的。

16、速度慢是緩沖區的問題 https://github.com/bperez77/xilinx_axidma/issues/69
（1）在設備樹中
/include/ "system-conf.dtsi"
/include/ "pl-custom.dtsi"
/ {

axidma_chrdev: axidma_chrdev@0 {
compatible = "xlnx,axidma-chrdev";
dmas = ;
dma-names = "tx_channel", "rx_channel";
dma-coherent;
};
&axi_dma_0 {
dma-coherent;
status = "okay";
};

在測試代碼中加速加入測速代碼
......
rc = single_transfer_test(axidma_dev, tx_channel, tx_buf, tx_size,
tx_frame, rx_channel, rx_buf, rx_size, rx_frame);
if (rc goto free_rx_buf;
}

struct timeval start_time0, end_time0;
double elapsed_time0;
// Begin timing
gettimeofday(&start_time0, NULL);
memcpy(image_mid0, rx_buf, 2048*4096*4);
gettimeofday(&end_time0, NULL);
elapsed_time0 = TVAL_TO_SEC(end_time0) - TVAL_TO_SEC(start_time0);
printf("\tElapsed Time0: %0.2f s\n", elapsed_time0);

struct timeval start_time1, end_time1;
double elapsed_time1;
// Begin timing
gettimeofday(&start_time1, NULL);
memcpy(image_mid1, image_mid0, 2048*4096*4);
gettimeofday(&end_time1, NULL);
elapsed_time1 = TVAL_TO_SEC(end_time1) - TVAL_TO_SEC(start_time1);
printf("\tElapsed Time1: %0.2f s\n", elapsed_time1);

（3）測試結果
root@u3_dma_test:/# ./mnt/u3_dma_save
AXI DMA Benchmark Parameters:
Transmit Buffer Size: 32.00 MiB
[ 84.709677] axidma axidma: DMA mask not set
Receive Buffer Size: 32.00 MiB
Number of DMA Transfers: 1000 transfers

Using transmit channel 0 and receive channel 1.
Elapsed Time0: 0.07 s
Elapsed Time1: 0.07 s
Single transfer test successfully completed!
Beginning performance analysis of the DMA engine.

DMA Timing Statistics:
Elapsed Time: 24.75 s
Transmit Throughput: 1292.82 MiB/s
Receive Throughput: 1292.82 MiB/s
Total Throughput: 2585.64 MiB/s
root@u3_dma_test:/#

最后測得速率 32MB/70ms=450MB。
17、雖然速度是提高了，但是讀寫的數據是不對的
https://forums.xilinx.com/t5/%E5%B5%8C%E5%85%A5%E5%BC%8F-%E7%A1%AC%E4%BB...
https://xilinx-wiki.atlassian.net/wiki/spaces/A/pages/18842098/Zynq+Ultr...

審核編輯：符乾江