NXP drivers: spi: spi_mcux_lpspi: enable CONFIG_SPI_ASYNC mode with DMA #78194
Conversation
DerekSnell
added
area: SPI
DerekSnell
force-pushed
the
LPSPI_ASYNC
branch
from
35119cb to
703d3c3
Compare
zephyrbot
requested review from
danieldegrasse,
dbaluta,
decsny,
manuargue,
MarkWangChinese,
tbursztyka and
teburd
| @@ -532,6 +532,12 @@ static int dma_mcux_edma_reload(const struct device *dev, uint32_t channel, | |||
| ret = -EFAULT; | |||
| } | |||
|
|
|||
| if (data->edma_handle.tcdPool == NULL) { | |||
| /* Not using scatter/gather mode, re-enable the hardware request */ | |||
| EDMA_EnableChannelRequest(DEV_EDMA_HANDLE(dev, channel)->base, | |||
There was a problem hiding this comment.
can you give an explanation of "why" in the commit message at least, maybe also in the comment
There was a problem hiding this comment.
added details in comment and commit message
drivers/spi/spi_mcux_lpspi.c
Outdated
| dma_size = MIN(data->ctx.tx_len, data->ctx.rx_len); | ||
|
|
||
| if (dma_size == 0) { | ||
| dma_size = MAX(data->ctx.tx_len, data->ctx.rx_len); |
There was a problem hiding this comment.
Hi @decsny ,
Making this change (always MAX) here in spi_mcux_dma_callback(), the spi_loopback test passes. However, if I make this same change in spi_mcux_dma_rxtx_load(), the "half start" test fails. I believe this existing code is keeping the RX and TX transfers equal size until one completes, and then it finishes the other.
With some compiler optimizations, compiler error occurs when trying to test tx_buf_size in BUILD_ASSERT. Signed-off-by: Derek Snell <[email protected]>
If not using scatter/gather mode, re-enable the hardware channel requests. The fsl_edma driver uses the eDMA DREQ bit to automatically disable the hardware DMA request at the end of the major loop. Signed-off-by: Derek Snell <[email protected]>
DMA scatter/gather mode was unused in this driver. Signed-off-by: Derek Snell <[email protected]>
If CONFIG_SPI_ASYNC, select the DMA feature Signed-off-by: Derek Snell <[email protected]>
add DMA async support. Signed-off-by: Derek Snell <[email protected]>
…ards Enable testing async mode on all the NXP boards with the LPSPI peripherals. Signed-off-by: Derek Snell <[email protected]>
DerekSnell
force-pushed
the
LPSPI_ASYNC
branch
from
703d3c3 to
cd2b208
Compare
| } | ||
|
|
||
| /* Restart the TX DMA */ | ||
| ret = dma_reload(dev, |
There was a problem hiding this comment.
Have we verified this works correctly when the SPI is only sending or only receiving data? If memory serves, we usually configure the TX DMA or RX DMA with a step size of zero and a dummy buffer in these cases, so that DMA will run on both channels. My concern is that if we (for example) are sending TX data and not receiving anything, the LPSPI RX FIFO will overflow since the DMA will not be reading from it.
There was a problem hiding this comment.
Good point, @danieldegrasse .
No, that has not been tested here. the spi_loopback test does not enable RX without TX.
Adding DNM label until we test further.
Thanks
|
Hello, I'm testing with mimxrt1060_evk. The tests are passing, but the functionality is incorrect. I am getting chip select for every byte when I should only get chip select for one block. |
what do you mean by this, are you saying that PR breaks it? because if so, that's not my observations. |
|
I'm using main branch of zephyr, and executing
If I remove the
I think the flag is correct because otherwise, you could only send one byte commands. The documentation says that to use this function, you need to send extra command to deselect chip-select.
|
|
This pull request has been marked as stale because it has been open (more than) 60 days with no activity. Remove the stale label or add a comment saying that you would like to have the label removed otherwise this pull request will automatically be closed in 14 days. Note, that you can always re-open a closed pull request at any time. |
Updates LPSPI driver to support async mode with DMA. Enables
spi_loopbacktesting async mode on all the NXP boards with the LPSPI peripherals. Resolves #74907.spi_loopback test with
CONFIG_SPI_ASYNCpasses on the following boards. The test failed on 1 board which needs more investigation, a separate issue created for that board at #78238.