maxikrie commented on PR #21347:
URL: https://github.com/apache/nuttx/pull/21347#issuecomment-6987165054

   Ran this on a NUCLEO-H753ZI on master fc8d354f9a (plus #22023, without which 
`nucleo-h753zi:socketcan` registers no CAN interface), FDCAN1 in internal 
loopback (`CONFIG_STM32_FDCAN_LOOPBACK_INTERNAL`), `CONFIG_STM32_FDCAN_HPWORK`, 
`CONFIG_IOB_NBUFFERS=64`. A test app sends numbered classic frames from one 
thread with blocking `write()` and receives them on a second socket, checking 
id, payload and order.
   
   | | 3 x 20000 frames |
   |---|---|
   | master | all 60000 sent and received once, in order, no corrupt payload, 
no send errors |
   | this PR | same |
   
   So no regression, but I could not reproduce the race on master this way: the 
writer's `devif_poll()` (called directly from `fdcan_txavail()`) and 
`fdcan_receive_work()` on the HP queue did not collide visibly here. The 
`fdcan_error()` path mentioned in review was not exercised (no bus errors in 
loopback).
   
   With the default 24 IOBs and the socket's default `CAN_RAW_LOOPBACK`, about 
half of the blocking writes failed with `ENOMEM` (from `devif_send()` in the 
unbuffered send path) on both master and this PR, unrelated to this change.


-- 
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