dakejahl opened a new pull request, #20008:
URL: https://github.com/apache/nuttx/pull/20008
## Summary
Two changes to the i.MX RT FlexCAN SocketCAN driver, plus the network-layer
plumbing the second one needs.
1. `imxrt_txdone_work()` and `imxrt_txtimeout_work()` shared
`priv->irqwork`. `work_queue()` cancels a pending callback when its `work_s` is
reused, so whichever was queued second silently replaced the first: deadlines
left set, `IMASK1` TX bits left off, or expired frames never aborted. Both
paths now queue one worker, `imxrt_tx_work()`, which retires completions before
aborting expired mailboxes.
2. A new `SIOCGCANERRORS` ioctl (`_SIOC(0x0045)`, behind
`CONFIG_NETDEV_CAN_ERROR_IOCTL`) returns `struct can_ioctl_errors_s`:
fault-confinement state, TEC, REC, a monotonic bus-error count and the RX
mailbox overrun count. `SIOCGCANSTATE` reports sleep/operational, not fault
confinement, so this is a distinct command. The driver counts bus errors by
sampling the clear-on-read `ESR1` error flags at every driver entry (TX work,
RX work, the ioctl) instead of enabling `ERRINT`, which fires per error frame
and storms at bus rate on a dead bus. Frames the CAN socket layer drops for
want of an IOB now also count as `rx_dropped` in the per-device netdev
statistics.
## Impact
- New feature: YES, `SIOCGCANERRORS` and `CONFIG_NETDEV_CAN_ERROR_IOCTL`
(default n), implemented for i.MX RT FlexCAN only; other SocketCAN drivers
return `-ENOTTY`.
- User: NO change for existing applications; the ioctl is opt-in.
- Build: NO change with the option disabled; `include/net/if.h` gains
`struct can_ioctl_errors_s` and one union member in `ifreq`/`lifreq`.
- Hardware: i.MX RT FlexCAN only (the TX worker change affects every i.MX RT
SocketCAN user).
- Documentation: NO documentation of the CAN ioctls exists to update.
- Security: NO.
- Compatibility: YES, backward compatible; `ifreq` grows by the size of the
new union member only if it becomes the largest, which it is not.
## Testing
Build: `imxrt1170-evk:can` with `CONFIG_NETDEV_CAN_ERROR_IOCTL=y`,
`CONFIG_NETDEV_STATISTICS=y`, `CONFIG_NET_STATISTICS=y`, arm-none-eabi-gcc
13.2.1, host Linux.
Runtime: i.MX RT1176 (ARK FMU-v6XRT) running PX4 with this change carried on
the PX4 NuttX fork (PX4/NuttX#401), two DroneCAN nodes, one per FlexCAN
interface, 1 Mbit/s.
Before (stub `getErrorCount()`, `isInBusOffState()`): an error-passive
controller is invisible from the socket API; `uavcan status` prints `HW errors:
0` regardless.
After, unplugging the CAN2 node for 10 s and reading the ioctl every 2 s
(console) while `ECR`/`ESR1` were read over SWD at 20 Hz:
```
64.1s CAN1 error-active TEC 0 REC 0 hw 0 | CAN2 error-passive TEC 128
REC 0 hw 379
66.8s CAN1 error-active TEC 0 REC 0 hw 0 | CAN2 error-passive TEC 128
REC 0 hw 451
74.7s CAN1 error-active TEC 0 REC 0 hw 0 | CAN2 error-passive TEC 128
REC 0 hw 637
77.4s CAN1 error-active TEC 0 REC 0 hw 0 | CAN2 error-passive TEC 152
REC 0 hw 659
85.3s CAN1 error-active TEC 0 REC 0 hw 0 | CAN2 error-active TEC 17
REC 0 hw 660
SWD: CAN2 TXERRCNT 128 from 66.9 s, ESR1 fault confinement error-passive,
302 samples over 23 s, 259 with ACKERR; CAN1 0 / error-active on all 5932
samples
```
With the two nodes moved to 500 kbit/s while the controller stayed at 1
Mbit/s, the ioctl reads `error-passive (TEC 0, REC 128)` while the controller
transmits nothing and `(TEC 128, REC 133)` while it does. With
`NETDEV_STATISTICS` enabled, `/proc/net/can0` showed `Dropped` advancing (445
frames in 300 s at 3260 frames/s with a 24-buffer IOB pool), which the global
CAN statistics alone did not attribute to an interface.
--
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]
