maxikrie opened a new pull request, #22022:
URL: https://github.com/apache/nuttx/pull/22022

   ## Summary
   
   On STM32H7, opening a serial port with RX or TX DMA crashes the system if 
its DMA controller has no free stream left.
   
   `stm32_dmachannel()` on the H7 (DMAMUX) hands out the first free stream of 
the requested controller and returns NULL when there is none; it does not block 
as on F4/F7. `up_dma_setup()` did not check the handle and passed it to 
`stm32_dmasetup()`, which asserts (`DEBUGASSERT(handle != NULL)`) or, without 
assertions, dereferences NULL.
   
   This PR makes `up_dma_setup()` return `-EBUSY` in that case, so `open()` 
fails with `EBUSY`. On failure it releases a TX stream that was already taken 
(when only the RX stream is missing) and undoes `up_setup()` for a non-console 
port (clock, pins), so a later `open()` works once a stream is free. The H5 
driver already returns `-EBUSY` here.
   
   This is item 3 of #20244, which says it "can land first", independent of the 
build-time budget proposed there.
   
   Not changed here: the H7 SPI driver also only `DEBUGASSERT`s its DMA handles 
in `spi_bus_initialize()`. With assertions off it already falls back to 
`spi_exchange_nodma()` when a handle is NULL, so it does not crash; with 
assertions on it panics at boot. Its comment still says `stm32_dmachannel()` 
blocks, which is the F4/F7 behaviour.
   
   ## Impact
   
   STM32H7 U(S)ARTs with `CONFIG_USARTn_RXDMA`/`CONFIG_UARTn_TXDMA`. Only the 
path where no DMA stream is available changes: `open()` now returns `EBUSY` 
instead of crashing. No change when a stream is available. No Kconfig, API or 
build changes.
   
   ## Testing
   
   NUCLEO-H753ZI, master fc8d354f9a, `nucleo-h753zi:nsh` + 
`CONFIG_STM32_USART6`, `CONFIG_STM32_DMA1`, `CONFIG_USART6_RXDMA`, 
`CONFIG_USART6_TXDMA`, `CONFIG_DEBUG_ASSERTIONS`. For the test, `board.h` maps 
USART6 to DMA1 (`DMAMAP_USART6_RX/TX = DMAMAP_DMA12_USART6xX_0`), and a 
test-only NSH app `dmahog [n]` takes n (default: all) DMA1 streams with 
`stm32_dmachannel()` and `dmahog free` releases them. Neither is part of this 
PR.
   Toolchain: Arm GNU Toolchain 13.3.Rel1. `tools/checkpatch.sh -g HEAD` passes.
   
   | step | master | this PR |
   |---|---|---|
   | `echo hello > /dev/ttyS1` (USART6) | ok | ok |
   | `dmahog` (all 8 DMA1 streams), then `echo hello > /dev/ttyS1` | assertion 
in `stm32_dmasetup()`, system panics | `open failed: 16` (EBUSY), repeatable |
   | `dmahog free`, then `echo hello > /dev/ttyS1` | - | ok |
   | `dmahog 7` (TX gets the last stream, RX fails), `echo x > /dev/ttyS1` | - 
| `open failed: 16`; afterwards `dmahog` again gets all 8 streams (TX stream 
released), RCC_APB2ENR.USART6EN = 0 (port shut down) |
   
   ## Logs
   
   <details><summary>master</summary>
   
   ```
   nsh> dmahog
   dmahog: holding 8 DMA1 streams
   nsh> echo hello > /dev/ttyS1
   dump_assert_info: Current Version: NuttX  13.1.0 fc8d354f9a-dirty Nov 18 
2026 17:58:01 arm
   dump_assert_info: Assertion failed panic: at file: :0 task: <noname> 
process: <noname> 0x8011c4d
   up_dump_register: R0: 00000000 R1: 38001354 R2: 38001354  R3: 00000000
   ...
   up_dump_register: IP: 00000000 SP: 38001330 LR: 08000eb3  PC: 0800227a
   ```
   
   PC 0x0800227a is `stm32_dmasetup()`, LR in `up_dma_setup()` 
(stm32_serial.c:2261), called from `uart_open()`.
   </details>
   
   <details><summary>this PR</summary>
   
   ```
   nsh> dmahog
   dmahog: holding 8 DMA1 streams
   nsh> echo hello > /dev/ttyS1
   nsh: echo: open failed: 16
   nsh> echo again > /dev/ttyS1
   nsh: echo: open failed: 16
   nsh> dmahog free
   dmahog: released all
   nsh> echo hello > /dev/ttyS1
   nsh>
   ```
   
   ```
   nsh> dmahog 7
   dmahog: holding 7 DMA1 streams
   nsh> echo x > /dev/ttyS1
   nsh: echo: open failed: 16
   nsh> dmahog free
   dmahog: released all
   nsh> dmahog
   dmahog: holding 8 DMA1 streams
   ```
   </details>


-- 
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: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to