The way you have written the code is that each periph defines a dma_descr_t when it wants to use DMA. Eg for I2C it defines {dma_I2C, 1, DMA_TX_TRANSFER}. Then whenever dma_init or dma_invalidate_channel is called it must search the big table in dma.c (dma_transfer_info) for a match to this dma_descr_t info. This seems inefficient in speed and code size.

As an alternative, how about the following: expose each entry of the dma_transfer_info table in dma.h with a unique name, and then point to this from the peripheral that wants to use it. For example:

dma.h:
    typedef struct _dma_p2dma_t dma_p2dma_t;
    extern dma_p2dma_t dma_I2C_1_RX;
    extern dma_p2dma_t dma_SPI_3_RX;
    extern dma_p2dma_t dma_I2C_3_RX;
    extern dma_p2dma_t dma_I2C_2_RX;

dma.c:
    struct _dma_p2dma_t {
        // no need for dma_descr_t
    #if defined(MCU_SERIES_F4) || defined(MCU_SERIES_F7)
        DMA_Stream_TypeDef  *instance;
    #elif defined(MCU_SERIES_L4)
        DMA_Channel_TypeDef *instance;
    #else
    #error "Unsupported Processor"
    #endif
        uint32_t        sub_instance;
        dma_id_t        id; 
        const DMA_InitTypeDef *init;
    } dma_p2dma_t;

    const dma_p2dma_t dma_I2C_1_RX = {DMA1_Stream0, DMA_CHANNEL_1, dma_id_0 , &dma_init_struct_spi_i2c},
    const dma_p2dma_t dma_SPI_3_RX = {DMA1_Stream2, DMA_CHANNEL_0, dma_id_2  , &dma_init_struct_spi_i2c},
    const dma_p2dma_t dma_I2C_3_RX = {DMA1_Stream2, DMA_CHANNEL_3, dma_id_2  , &dma_init_struct_spi_i2c},
    const dma_p2dma_t dma_I2C_2_RX = {DMA1_Stream2, DMA_CHANNEL_7, dma_id_2  , &dma_init_struct_spi_i2c},

i2c.c:
    typedef struct _pyb_i2c_obj_t {
        mp_obj_base_t base;
        I2C_HandleTypeDef *i2c;
        dma_p2dma_t *tx_dma;
        dma_p2dma_t *rx_dma;
    } pyb_i2c_obj_t;

    STATIC const pyb_i2c_obj_t pyb_i2c_obj[] = { 
        {{&pyb_i2c_type}, &I2CHandle1, &dma_I2C_1_TX, &dma_I2C1_RX},
        ...
    };

Does that make sense?

Thank you - good point. I was as well not satisfied with my solution.

Sometimes you have different alternatives for a periphery. Shall I call them eg. dma_I2C_1_RX_1 or just not declare them?

Sometimes you have different alternatives for a periphery. Shall I call them eg. dma_I2C_1_RX_1 or just not declare them?

If they're not used, don't declare them, just leave them as commented-out code in case they are needed later. Otherwise, if they are used, put a postfix of _A or _B.

I have now a solution at hand and debugging it.

Is there an automated build, deploy and test framework for directly testing the code on a hardware? For the DMA I started to build some code which:

1. Checkout a given branch if the code is not dirty.
2. Builds the code.
3. Deploys the code to the correct board (Board must match the build target) even if there are several boards connected to the PC.
4. Runs python test code over serial connection on the correct board and checks the code result.
   Of course there is need for some polishing but as concept my python code works.

Is there a function in mp where I can get the information which is returned during a cold reset eg:

PYB: sync filesystems
PYB: soft reboot
MicroPython v1.7-240-gd7cd205-dirty on 2016-04-21; L476-DISCO with STM32L476
Type "help()" for more information.

What I am interested in is the version, date, board and CPU information.

Is there a function in mp where I can get the information which is returned during a cold reset eg:

Try os.uname()

Thank you.

I modified the PR according to your suggestions and collected all concerned file in this PR. I did some test (spi/i2c transfer on stm32f4disco and stm32l476disco) which passed.

I hope that it can be merged now.

I see that there are modifications to adc.c. Are they relevant to DMA? If not can you please separate the ADC changes into a separate commit (can be within this PR, just a separate commit to the DMA changes). Thanks!

Ops - that somehow sneaked in.

The adc changes are already merged in d49a547.

I did an update.

Updated the code and did some clean ups. Did some simple i2c/spi test on stm32f4disco, stm32F429disco, stm32l4Disco, limifrog.

I did a little bit of tidying up (mostly white space changes and reordering some things in dma.c) and merged it in e64032d. Thank you!