snikeguo opened a new pull request, #16463: URL: https://github.com/apache/nuttx/pull/16463
*Note: Please adhere to [Contributing Guidelines](https://github.com/apache/nuttx/blob/master/CONTRIBUTING.md).* ## Summary 1. Added support for STM32H723. 2. ARM: Added SP register initialization during startup. 3. Added FDCAN3 support. 4. Fixed UART compilation issue (removed macro definition restrictions related to the "up_restoreusartint undefined" error). 5. Adjusted interrupt initialization timing: interrupts should be configured as soon as the chip starts up to prevent abnormal initialization after the OS starts, which could otherwise result in the BootLoader’s interrupt handler being triggered. 6. Updated CMake to add ELF file suffix and fixed a bug in CMake (to be investigated). 7. Added DMA align manager component. 8. Optimized mmcsd code logic. 9. Removed FAT layer DMA configuration; DMA logic should reside in the device driver layer. ## Impact Regarding point ⑤, I believe that interrupt initialization is too late. We should configure interrupts as soon as the chip starts up to prevent abnormal initialization after the OS starts, which could cause exceptions to be handled by the BootLoader’s interrupt handlers (if a bootloader is present). The previous initialization sequence was: ``` Power on; Initialize bss and data sections; Some hardware initialization, such as clock, FPU, CACHE, etc.; tasklist_initialize(); idle_task_initialize(); ... irq_initialize(); // Interrupts are only set up here. ``` The problem with this approach is that, if an exception occurs before this point, it is handled by the BootLoader’s handler (if there is a bootloader). I believe the sequence should be: ``` Power on(); Disable interrupts(); Initialize bss and data sections(); Configure interrupts(); ... ... Enable interrupts(); Start scheduler(); ``` Regarding point ⑥, some debuggers only recognize files with the .elf suffix. Also, sometimes the check if(NOT EXISTS "${NUTTX_APPS_DIR}") incorrectly determines that the APPDIR folder does not exist. Regarding point ⑦, I believe DMA management should be handled by device drivers, not by the FAT layer. For more details, see: @apache/nuttx/issues/16415 Regarding point ⑧, during MMC/TF card enumeration, DMA should not be used because the data volume is only a few bytes. Linux is also designed this way. Regarding point ⑨, the file system should not have to consider DMA issues. ## Testing MMC Write Logic: mmcsd_writemultiple() { .... mmcsd_transferready(); ... SDIO_DMASENDSETUP(); } In the current code (https://github.com/apache/nuttx/commit/b05ec7729d9b7fe85a6f11bfbb2da979255f44b7), I found that after writing multiple sectors, when sending ACMD51 and then querying R1, the error rate is very high. test code: ``` int r1_error_cnt=0; mmcsd_transferready() { ... ret = mmcsd_get_r1(priv, &r1); if (ret != OK) { ferr("ERROR: mmcsd_get_r1 failed: %d\n", ret); return ret; } /* Now check if the card is in the expected transfer state. */ if (IS_STATE(r1, MMCSD_R1_STATE_TRAN)) { /* Yes.. return Success */ priv->wrbusy = false; return OK; } extern int wm_called; if(wm_called) { r1_error_cnt++; } } ```  With my code, the test results show no errors.  STM32H7 -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: commits-unsubscr...@nuttx.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org
