[TW#23413] spi master management of byte order when the size of not byte aligned is crazy

[Jegeva]

----------------------------- Delete below -----------------------------

If your issue is a general question, starts similar to "How do I..", or is related to 3rd party development kits/libs, please discuss this on our community forum at esp32.com instead.

INSTRUCTIONS

----------------------------- Delete above -----------------------------

Environment

• Development Kit: [ESP32-Wrover-Kit|]
• Kit version (for WroverKit/PicoKit/DevKitC): [NA]
• Core (if using chip or module): [NA]
• IDF version (git rev-parse --short HEAD to get the commit id.): 0e501e5
• Development Env: [Make]
• Operating System: [Debian]
• Power Supply: [USB]

Problem Description

//Detailed problem description goes here.
while sending data with the spi master :

void send_9(uint16_t data){
esp_err_t ret;
spi_transaction_t t;
uint16_t workaround = data;
memset(&t, 0, sizeof(t)); //Zero out the transaction
t.length=9; //Command is >9< bits
t.tx_buffer=&workaround; //The data is the cmd itself
ret=spi_device_transmit(spi, &t);//Transmit!
assert(ret==ESP_OK);
}

Expected Behavior

data_bit8, data_bit7,...,data_bit0 as expected

Actual Behavior

data_bit7,...,data_bit0,data_bit15 is sent (check with a logic analyser)

Steps to repropduce

hook an esp32 to a LA,
set it up to use VSPI

send 0x00ff with a size of 9 : 0b111111110 is sent
send 0x80ff with a size of 9 : 0b111111111 is sent

this makes no sense

// It helps if you attach a picture of your setup/wiring here.

Code to reproduce this issue

-> use the spi master exemple, change the size, test the patterns i explained with a LA.

// the code should be wrapped in the ```cpp tag so that it will be displayed better.
#include "esp_log.h"

void app_main()
{
    
}

// If your code is longer than 30 lines, GIST is preferred.

Debug Logs

Debug log goes here, should contain the backtrace, as well as the reset source if it is a crash.
Please copy the plain text here for us to search the error log. Or attach the complete logs but leave the main part here if the log is *too* long.

Other items if possible

• sdkconfig file (attach the sdkconfig file from your project folder)
• elf file in the build folder (note this may contain all the code details and symbols of your project.)
• coredump (This provides stacks of tasks.)

[xiongyumail]

Hi,@Jegeva
Can you provide a completed test program for us to use? Let us solve your problem faster.

[FayeY]

Any update?

Hi @Jegeva,

Could you provide a test project so we can reproduce the problem and help solve it?

Thanks.

[ginkgm]

Hi @Jegeva ,
ESP32 is a little endian system, and this is a common issue existing in little endien systems.

In such systems, an unit32_t variable a=0x12345678 is placed like following in the memory.
0x78, 0x56, 0x34, 0x12

However the DMA don't know how your data is organized (as uint8_t, uint16_t or uint32_t), it always assume your data are uint_8s. The DMA takes data from the memory from low to high address, byte by byte. i.e. 0x78->0x56->0x34->0x12, and send from the MSB to LSB, namely:
01111000->01010110->00110100->00010010

This is how ESP32 hardware works.

To send uint16_t and uint32_t and even random length of data, please:

1. Move your data to start from the MSB of your variable (bit15 for uint16_t and bit31 for uint32_t).
2. use __builtin_bswap16 or __builtin_bswap32 to re-order the MSB byte to the lowest address

To resolve data received, please do in the reverse way.

Example:

#define SPI_SHIFT_DATA_8(data, len) (data<<(8-len))
#define SPI_SHIFT_DATA_16(data, len) __builtin_bswap16(data<<(16-len))
#define SPI_SHIFT_DATA_32(data, len) __builtin_bswap32(data<<(32-len))

#define SPI_GET_DATA_8(data, len) (data>>(8-len))
#define SPI_GET_DATA_16(data, len) (__builtin_bswap16(data)>>(16-len))
#define SPI_GET_DATA_32(data, len) (__builtin_bswap32(data)>>(32-len))

void send_9(uint16_t data){
    esp_err_t ret;
    spi_transaction_t t;
    uint16_t workaround = SPI_SHIFT_DATA_16(data, 9);
    memset(&t, 0, sizeof(t)); //Zero out the transaction
    t.length=9; //Command is >9< bits
    t.tx_buffer=&workaround; //The data is the cmd itself
    ret=spi_device_transmit(spi, &t);//Transmit!
    assert(ret==ESP_OK);
}

We plan to provide macros in the future version.

Another notice, please take care that your variable sent to the driver should be 32-bit aligned.

[ginkgm]

BTW, there're such issues with command and address fields, and are re-ordered by the driver:

https://github.com/espressif/esp-idf/blob/master/components/driver/spi_master.c#L730

[Jegeva]

Dears, Have been busy, I’ll make something soon Sincerely Le ven. 22 juin 2018 à 14:09, FayeY <notifications@github.com> a écrit :

…

Any update? Hi @Jegeva <https://github.com/Jegeva>, Could you provide a test project so we can reproduce the problem and help solve it? Thanks. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#2062 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AEtfWYzRA9AykT4uHVK_XBH_hY04PfpVks5t_N5zgaJpZM4UmgCm> .

[ginkgm]

Update:

#define SPI_SHIFT_DATA(data, len) __builtin_bswap32((uint32_t)data<<(32-len))
#define SPI_REARRANGE_DATA(data, len) (__builtin_bswap32(data)>>(32-len))

[ginkgm]

Hi @Jegeva ,
The IDF is updated according to your problem. Please update and have a try.
If there's any issue, please come and feel free to reopen this.