Hi Petro, Thanks for the feedback! CONFIG_ARMV7M_DCACHE = y //only this option related to DCACHE CONFIG_STM32F7_DMACAPABLE = y CONFIG_STM32F7_SPI_DMATHRESHOLD = 8
no CONFIG_ARMV7M_DCACHE_WRITETHROUGH no CONFIG_SPI_TRIGGER no CONFIG_STM32F7_SPI1_DMA_BUFFER I've tried before with CONFIG_STM32F7_SPI1_DMA_BUFFER=1024 since I used STM32F4 <https://github.com/PX4/PX4-Autopilot/issues/16254#issuecomment-731768238>, and also now with STM32F7, but DMAtransfers stops work at all (there is empty data read and written for some reason), while I have such config buffer for SPI3 and it's working correctly. I haven't found out the reason for this and just skipped it. Could this be related to current problems and I should debug this more deeply? I've tried to use up_clean_dcach() in stm32f7/stm32_spi.c as you suggested. It doesn't help, there is still data corruption, but subjectively it seems that less often than before with the same actions I repeat trying to reproduce the issue. Just in case, there are no "extra" bytes, everything in place, but a couple bytes are corrupted to zero, as I see, in the same places, but I haven't noticed some pattern or specific offsets of corruption. [image: изображение.png] --- With best regards, Oleg. пт, 23 июн. 2023 г. в 00:07, Petro Karashchenko < petro.karashche...@gmail.com>: > Hi, > > Could you please add some more information about the configuration like > "CONFIG_STM32F7_DMACAPABLE", "CONFIG_SPI_TRIGGER", > "CONFIG_STM32F7_SPIx_DMA_BUFFER" and "CONFIG_STM32F7_SPI_DMATHRESHOLD". > Also the D-cache configuration like "CONFIG_ARMV7M_DCACHE_WRITETHROUGH". > > I do not see obvious issue except mirror improvement where > "up_clean_dcache((uintptr_t)txbuffer, (uintptr_t)txbuffer + nbytes);" can > be used instead of "up_flush_dcache((uintptr_t)txbuffer, > (uintptr_t)txbuffer + nbytes);", but that obviously should not lead to > random extra bytes. > > Best regards, > Petro > > чт, 22 черв. 2023 р. о 16:14 Oleg <ev.m...@gmail.com> пише: > > > Hi all, > > > > I'm working with custom stm32f7 board and PX4 old fork, where I have my > > module that writes in several chunks some data (289 bytes) inside one of > > the partition of MTD mx25lx at SPI1, while there is PX4 params module > that > > autosave parameters on changes in another partition of the same MTD. > > And sometimes data in my part of MTD (at least) gets corrupted by a > couple > > of zeroed bytes, I can check it by rereading right after writing. > > > > It does not always happen and it's quite tricky to catch. There is no > issue > > if no parameters change in parallel or if I disable > > CONFIG_STM32F7_SPI1_DMA. > > > > As for me looks like potential issue with dcache, especially taking into > > account I'm quite "lucky" for such issues (: previously were "SD FAT32: > > Write failed with err 28 (no space left on device)" that was fixed in > > https://github.com/apache/nuttx/pull/5062 and "Potential UDP packets > loss" > > that was fixed in https://github.com/apache/nuttx/pull/6831. > > > > Unfortunately I don't have enough perception of dcache, how to debug such > > time-related things, so I'm looking for any advice. > > > > --- > > With best regards, Oleg. > > >
