Repository navigation
Conversation
nxp_nci_remove() takes info_lock and then calls nci_unregister_device(). If the device is up, nci_unregister_device() closes it through nci_close_device() and the driver's close hook, nxp_nci_close(), which takes info_lock again. The mutex is not recursive, so removing a powered device hangs: the unbind or rmmod never returns, and a later shutdown blocks in device_shutdown(). Seen on a Pixel 9a whose NFC controller is driven through this I2C transport: with the controller powered by neard, the device's removal during a suspend attempt hung with nxp_nci_close() under nxp_nci_remove(). Unregister first, outside the lock, then put the chip in cold mode under the lock, and free the device last. Signed-off-by: W4HO <87742609+W4HO-ham@users.noreply.github.com>
|
Thanks for checking, but this one is already on The same goes for #67: your pinctrl commit is |
This is #61 again. You merged it on 2026-10-26 at 23:23 UTC, but the omamobile branch on GitHub doesn't have it now: the merge commit isn't an ancestor of omamobile, and the file still has the old code. My guess is that the branch was republished after the merge from a tree without it.
Same commit, cherry-picked onto the current omamobile with no conflicts. checkpatch --strict is clean and the file builds with W=1. #61 and #59 were merged together and are both missing, so I'm resubmitting each.