Develop on a chip like MT7628 is full of adventure…For new version of firmware I plan to use opensource MT76 driver instead of MT7628 driver. Weird thing is MT76 works well on some devices, but not well on some others, include VoCore2.

I guess it is antenna issue, because for normal WIFI4 routers default has two antennas works at same time for 300Mbps, 802.11n mode; but for VoCore2, default antenna 2 is disabled to save power and avoid cross talk when no second antenna attached.

Thanks to github MT76 contributors, I have some clues to fix this: https://github.com/openwrt/mt76/issues/423

After apply that patch, change mt76-2020-03-10-08054d5a/mt7603/init.c line 538

- dev->mt76.antenna_mask = 1;
+ dev->mt76.antenna_mask = 3;

I get a good result, 5minute no disconnect anymore.

...
[  5] 297.00-298.00 sec  6.31 MBytes  52.9 Mbits/sec                  
[  5] 298.00-299.00 sec  6.38 MBytes  53.5 Mbits/sec                  
[  5] 299.00-300.00 sec  6.41 MBytes  53.7 Mbits/sec                  
- - - - - - - - - - - - - - - - - - - - - - - - -
[ ID] Interval           Transfer     Bitrate
[  5]   0.00-300.00 sec  1.75 GBytes  50.2 Mbits/sec                  sender
[  5]   0.00-300.07 sec  1.75 GBytes  50.2 Mbits/sec                  receiver

So this should be the solution, just need test more. Currently we store 1T1R and 2T2R parameter to factory setting partition in flash, so there are two ways, one is to add factory setting patch to change it, another is to add DTS setting patch to change it, emm, which way is better? hard choice.

To be continue…

Here is the compressed firmware link http://vonger.cn/misc/screen/20200628.touch.bin.xz, upgrade then it is able to work directly 🙂 Just need two files to work, one is the fbusb.ko driver for touch and USB data transfer, another is /root/demo which is the littlevgl demo application.

New Screen with touch API is simpler, please check https://vocore.io/screen.html for more details.

mt7628a has a wafer/silicon bug on SPI, we can not use its SPI in full duplex mode, but we can use bitbang for that mode. For some device do not support half-duplex, maybe this is the only way.

To enable bitbang on VoCore2, we need to modify Linux kernel code, VOCORE2.DTS and kernel setting.

Setup Kernel Setting

1. call “make kernel_menuconfig”
2. In menu, select “Device Drivers” => “SPI Support” => <*>   GPIO-based bitbanging SPI Master

VOCORE2.dts

This file is in target/linux/ramips/dts, example of the DTS. We need to set the driver to “spi-gpio” and pinctrl set the pins to GPIO mode.

/dts-v1/;

#include "VOCORE2.dtsi"

#include <dt-bindings/gpio/gpio.h>

/ {
	compatible = "vocore,vocore2", "mediatek,mt7628an-soc";
	model = "VoCore2";

	gpio-leds {
		compatible = "gpio-leds";

		status {
			label = "vocore2:fuchsia:status";
			gpios = <&gpio1 12 GPIO_ACTIVE_HIGH>;
		};
	};

	spi1: spi-gpio {
		status = "okay";

		compatible = "spi-gpio";

		gpio-sck = <&gpio0 7 1>;
        	gpio-miso = <&gpio0 9 1>;
        	gpio-mosi = <&gpio0 8 1>;

        	cs-gpios = <&gpio0 10 1>, <&gpio0 6 1>;
        	num-chipselects = <2>;
	};
};

&pinctrl {
        state_default: pinctrl0 {
                spi {
                        ralink,group = "spi";
                        ralink,function = "gpio";
                };
                spi_cs1 {
                        ralink,group = "spi cs1";
                        ralink,function = "gpio";
                };
        };
};

&i2c {
	status = "okay";
};

&spi1 {
	m25p80@0 {
		#address-cells = <1>;
		#size-cells = <1>;
		compatible = "jedec,spi-nor";
		reg = <0>;
		spi-max-frequency = <10000000>;
		m25p,chunked-io = <32>;

		partition@0 {
			label = "u-boot";
			reg = <0x0 0x30000>;
			read-only;
		};

		partition@30000 {
			label = "u-boot-env";
			reg = <0x30000 0x10000>;
			read-only;
		};

		factory: partition@40000 {
			label = "factory";
			reg = <0x40000 0x10000>;
			read-only;
		};

		partition@50000 {
			label = "firmware";
			reg = <0x50000 0xfb0000>;
		};
	};

	spidev@1 {
		#address-cells = <1>;
		#size-cells = <1>;
		compatible = "rohm,dh2228fv";
		reg = <1>;
		spi-max-frequency = <10000000>;
	};
};

Patch Linux spi-gpio.c

This is because CS1 pins has a default pull low resistor. If we do not patch it, CS1 will be low at startup, conflict with the flash driver who use CS0. So we must at startup setup CS1 pin to output and value to high.

                if (!SPI_N_CHIPSEL)
                        spi_gpio->cs_gpios[0] = SPI_GPIO_NO_CHIPSELECT;
                else
                        for (i = 0; i < SPI_N_CHIPSEL; i++) {
                                status = of_get_named_gpio(np, "cs-gpios", i);
                                if (status < 0) {
                                        dev_err(&pdev->dev,
                                                "invalid cs-gpios property\n");
                                        goto gpio_free;
                                }
                                spi_gpio->cs_gpios[i] = status;

+                               gpio_request(status, dev_name(&pdev->dev));
+                               gpio_direction_output(status, 1);
+                               gpio_free(status);
                        }
        }
#endif

        spi_gpio->bitbang.master = master;
        spi_gpio->bitbang.chipselect = spi_gpio_chipselect;

That is all 🙂 Now the firmware is able to support SPI device in full duplex mode.

Git is good, but for beginner like me, always feeling no way to give the first bite. I am confused about fork, branch, signed-off-by, push, merge and many other commands, maybe it is pretty clear for the clever boys but not for me. Thanks to adrianschmutzler comment at https://github.com/openwrt/openwrt/pull/3022, I start to understand how it works.

Tutorial

1. click the “Fork” button on github.com/openwrt/openwrt
2. goto your forked openwrt, click on “Clone or download”, get the address and clone to your local.
3. make the change on your local source code.
4. setup your name and email on your local by command git config –global user.email xxxx@xxxxx.xxx and git config –global user.name “your name”
5. submit your change by command “git commit -s“, remember to keep that Signed-off-By line and add the reason of the changes to comment.
6. push the commit by command “git push origin“
7. finally, go to github, create a merge request.

Signed-Off-By is just a text line in the comment to tell other who submit that patch, I am confused about that a long time, I thought it is added by some magic :p

PS: In old good time I only use git init, git commit and git clone. Now good new day comes.

Recently I try latest stable version openwrt, looks like the open source MT76 driver is much stable and fast now.

iperf3 -c 192.168.1.1
Connecting to host 192.168.1.1, port 5201
[  5] local 192.168.1.231 port 60121 connected to 192.168.1.1 port 5201
[ ID] Interval           Transfer     Bitrate
[  5]   0.00-1.00   sec  5.33 MBytes  44.7 Mbits/sec                  
[  5]   1.00-2.00   sec  5.50 MBytes  46.1 Mbits/sec                  
[  5]   2.00-3.00   sec  5.34 MBytes  44.7 Mbits/sec                  
[  5]   3.00-4.00   sec  4.67 MBytes  39.3 Mbits/sec                  
[  5]   4.00-5.00   sec  5.47 MBytes  45.8 Mbits/sec                  
[  5]   5.00-6.00   sec  5.03 MBytes  42.3 Mbits/sec                  
[  5]   6.00-7.00   sec  4.93 MBytes  41.3 Mbits/sec                  
[  5]   7.00-8.00   sec  5.07 MBytes  42.6 Mbits/sec                  
[  5]   8.00-9.00   sec  4.95 MBytes  41.5 Mbits/sec                  
[  5]   9.00-10.00  sec  5.13 MBytes  43.0 Mbits/sec                  
- - - - - - - - - - - - - - - - - - - - - - - - -
[ ID] Interval           Transfer     Bitrate
[  5]   0.00-10.00  sec  51.4 MBytes  43.1 Mbits/sec                  sender
[  5]   0.00-10.09  sec  51.3 MBytes  42.7 Mbits/sec                  receiver

iperf Done.

This is good 🙂 Next firmware will base on this driver and I do not have to patch it anymore.

Pin share or I call it pinmux, it is a common feature of most MCU chips. Normally a pin we can use it as GPIO or some special functions.

In picture, the orange number is each pins GPIO number. And in the color block, there is its special function.

Here is a table from vocore.io/v2.html, for example, the pin 29 on the chip has three functions, one is used as GPIO, one is system reference clock, final one is SDcard CLK.

On VoCore2 Ultimate, ES8388 do not need external clock because we can use VoCore2 system reference clock as its source, in most situation, that will save cost of the BOM.

Set the Pin Function

now, question come, how do we setup every pins function? Sometimes you may want use it as GPIO but sometimes we need to use as clock.

Actually it is same as other MCU(like STM32 and ATMEGA–Arduino) , we have register for every function.

Use this REFCLK pin as example, we can find on MT7628AN datasheet page 117, 2.2 System Control Section, 0x10000060 is that control register.

Its register is 1:0 two bits. (note: on this datasheet, its name is GPIO mode)

We can write 10b to this register, then it will be REFCLK mode.

and after it into REFCLK mode, we can change its frequency, at page 110, 0x1000002c register, bit 11:9 REFCLK0_RATE

And you can use same way to change other pins function 🙂

Also I have a simple tool to do this, please download at vocore.io/v2.html, Directly write to memory/register Section.

mjpeg-streamer is an easy way, but it only able to push mjpeg data, it is a waste because VoCam264 already has h264 stream inside.

So I plan to push h264 data to a RTMP server like nginx-rtmp(-module).

The first way in my mind is to use ffmpeg directly.

Just one command, and it should work: ffmpeg -re -f v4l2 -i /dev/video1 -c:v copy -f flv rtmp://192.168.0.xxx/live

My server is setup at rtmp://192.168.0.xxx/live

Note1: old version ffmpeg do not work well, I am using version 4.2.2. Currently openwrt package has ffmpeg, but it is 3.4.2, we must modify the Makefile and compile again to use latest version. I ave upload Makefile to github.com/vonger/vocore2, in utils folder.

Trying to set codec:h264 pix_fmt:none
parser not found for codec h264, packets or times may be invalid.

Note2: default Makefile will show error about parser, we need to enable h264 parser.

Then ffmpeg is able to push h264 video from camera to server.

Unfortunately, the speed is very slow and video is laggy…Maybe there is something wrong, because it works well on a Linux PC..

I have to find some other way.

My compiled ffmpeg ipk download at here: http://vonger.cn/misc/vocore2/ipk/ffmpeg/

To be continue…

Forget Keil, I love GCC. Makefile is enough for the ARM based chip GD32F150 🙂

Link to my source code:https://github.com/Vonger/gd32tools

To get AD values from UART2. Here is the Makefile. Source code please use project/adc1.

NAME = $(notdir $(CURDIR))
CMSIS = $(CURDIR)/../../GD32F1x0_Firmware_Library_v3.1.0/Firmware/CMSIS
PERIP = $(CURDIR)/../../GD32F1x0_Firmware_Library_v3.1.0/Firmware/GD32F1x0_standard_peripheral
TOOLCHAIN = $(CURDIR)/../../toolchain/mac/bin/arm-none-eabi

CC = $(TOOLCHAIN)-gcc
CP = $(TOOLCHAIN)-objcopy

DEFINES = -DGD32F130_150 -DUSE_STDPERIPH_DRIVER -DUSE_HSI_8M

INCLUDES = \
	-I$(CURDIR)/../core \
	-I$(CMSIS)/GD/GD32F1x0/Include \
	-I$(PERIP)/Include \
	-I$(CURDIR)

SOURCES = \
	$(CURDIR)/../core/core_cm3.c \
	$(CURDIR)/../core/startup_gd32f1x0.s \
	$(CMSIS)/GD/GD32F1x0/Source/system_gd32f1x0.c \
	$(wildcard $(PERIP)/Source/*.c) \
	$(wildcard $(CURDIR)/*.c)

OBJECTS = $(SOURCES:%.c=%.o)

CFLAGS = \
	-mcpu=cortex-m3 -mthumb -mlittle-endian -mthumb-interwork \
	-ffast-math -fdata-sections -ffunction-sections \
	-Wl,-T,$(CURDIR)/../core/gd32f150g8.ld,-Map,$(NAME).map,--gc-sections \
	-Wall -std=gnu99 -O2 $(DEFINES) $(INCLUDES) 

$(NAME): $(SOURCES)
	@$(CC) $(CFLAGS) $^ -lm -lnosys -o $(CURDIR)/$@
	@$(CP) -O ihex $(CURDIR)/$@ $(CURDIR)/$@.hex

test:
	@echo $(OBJECTS)

clean:
	@rm -f $(CURDIR)/$(NAME)
	@rm -f $(CURDIR)/$(NAME).map

GD32F1x0_Firmware_Library_v3.1.0 this library is from GigaDevice. Download here: http://vonger.cn/misc/vocore2/GD32F1x0_Firmware_Library_v3.1.0.rar; For toolchain, please check http://vonger.cn/?p=14891, I have a link there.

The library is for Keil originally. I make some patches, all changes are in project/core folder.

• startup_gd32f1x0.s this mainly change the interrupt vector, we need the interrupt callback function name same.
• gd32f150g8.ld this is used to create bin or hex file which is used to load into flash.

For VoCore2 Ultimate, we use internal 8M clock, so comment the line 46 #define __SYSTEM_CLOCK_72M_PLL_HXTAL and uncomment line 47 #define __SYSTEM_CLOCK_72M_PLL_IRC8M_DIV2 in GD32F1x0_Firmware_Library_v3.1.0/Firmware/CMSIS/GD/GD32F1x0/Source/system_gd32f1x0.c

Now we are ready to make. in adc1 folder, call make, it will create adc1.hex.

Connect BOOT pin on the dock to 3.3V pin on VoCore2, then power on them from microUSB.

Once VoCore2 ready, run gd32up to load adc1.hex. After upload done, power off it and disconnect BOOT from 3.3V (We do not export its RST pin, so have to power off to reset).

After you connect it back to power, the GD32F150 will in ADC mode, UART2 on VoCore2 is not for USB2TTL anymore, it will output ADC value to VoCore’s UART2. We can use minicom to view it at VoCore2 console, 115200, 8n1.

This is a tutorial for using C/C++ compile application in VoCore2.

1. Prepare a USB disk or SD card, at least 256MB, because GCC takes around 110MB. USB disk or SD card must be EXT4 format.

For macOS or Linux, call mkfs.ext4 /dev/disk2 to do this. /dev/disk2 is the USB disk name on my computer, please change to the name on your computer. Or another way, directly format it in VoCore2, need to install e2fsprogs.

opkg update
opkg install e2fsprogs
mkfs.ext4 /dev/sda

NOTE: opkg updaterequires internet, check vocore.io/v2.html, AP+Client mode for wireless connection. Or modify network switch settings, set ethernet port0 to VLAN 2(wan).

This is an example, directly patch /etc/config/network. Another way, you can change virtual network switch in LuCI, Network->Switch.

 config switch_vlan     
         option device 'switch0'
-        option vlan '1'  
+        option vlan '2'    
         option ports '0 6t'

2. Mount SD/USB storage to /overlay

Patch /etc/config/fstab, add target /overlay to make sure once storage is detected, system will auto mount it to /overlay.

config 'global'
        option  anon_swap       '0'
        option  anon_mount      '0'
        option  auto_swap       '1'
        option  auto_mount      '1'
        option  delay_root      '5'
        option  check_fs        '0'

config 'mount'
        option  device  '/dev/sda'
        option  target  '/overlay'
        option  enabled '1'

config 'mount'
        option  device  '/dev/mmcblk0'
        option  target  '/overlay'
        option  enabled '1'

/dev/sda is for USB disk, /dev/mmcblk0 is for SD card.

3. Reboot

Make sure /overlay is valid. You can check by command df.

root@OpenWrt:~# df
Filesystem           1K-blocks      Used Available Use% Mounted on
/dev/root                12800     12800         0 100% /rom
tmpfs                    62492       660     61832   1% /tmp
/dev/mmcblk0          30363504    148636  28649428   1% /overlay
overlayfs:/overlay    30363504    148636  28649428   1% /
tmpfs                      512         0       512   0% /dev

Or another way, directly manually mount /dev/sda or /dev/mmcblk0 to /overlay folder by command mount /dev/sda /overlay.

4. Now we can install GCC

This part is easy, just call

opkg update
opkg install gcc

It will install gcc, ar, binutils,libbfd, objdump, libopcodes packages from openwrt server.

Then we can compile C source code, try gcc yourcode.c -o out. Speed is not very fast, but works.

NOTE: compile better in /tmp folder(it is memory virtual disk) or in /overlay(it is the SD card we inserted). Rest path will store in NOR flash who has very limited write times and very little free space.

Finally I make this:

It works very smooth, just like smart phone. And color is very well too. 🙂 I really like this small device, I guess it should be the lowest cost WVGA(800×480) UI solution or HMI solution. 1K units should be less than 10USD, so with VoCore2, the full solution cost can be less than 20USD.

Now I am busy on doing the final debug and fix. Hopefully in next month, I can get first batch of the production and put on vocore.io.

Because currently I am updating the fully screen every frame, so its FPS is pretty low, only 30fps, but I find a way to update the screen partially, so once I finish the next patch, the screen will reach 60fps, can be run on VoCore, Android smart phone and Windows computer as an extend screen.

Qt recently public MCU1.0, running on a very expansive platform over 30USD with low speed CPU and only 1MB~4MB memory. I guess it will be perfect with VoCore and this screen. I will be the first guy port it. :p