]> git.ipfire.org Git - thirdparty/kernel/linux.git/commit
serial: qcom-geni: fix TX DMA buffer flush
authorJan Sebastian Götte <linux@jaseg.de>
Wed, 29 Jul 2026 17:41:05 +0000 (19:41 +0200)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Thu, 30 Jul 2026 14:46:29 +0000 (16:46 +0200)
commite3c04834ae1ab5e9cfbe8ac54ec734aa4774249d
tree290778a19100d94ccfaba26215dcb8e1b34cb4a6
parente2fe6a0efecbef00e3ecc2db64dd5afa8c212b41
serial: qcom-geni: fix TX DMA buffer flush

When transmit flushing a qcom-geni UART during an ongoing TX DMA, the
UART gets stuck infinitely repeating corrupted TX DMA frames.

The DMA-mode uart_ops does not provide a flush_buffer callback, so an
in-flight transfer can complete after serial core has reset the transmit
kfifo, underflowing its length and resubmitting page-sized transfers
indefinitely. Add one that stops the transfer and clears tx_remaining
and tx_queued.

The stop path was also broken: it unmapped the buffer while the serial
engine could still read it, and never reset the TX DMA state machine.
Cancel the main sequencer command first, then reset the state machine
and wait for it before unmapping. Drop the early return so a pending
mapping is also cleaned up when the main command is inactive.

The bug can be triggered from userspace with a large write immediately
followed by TCOFLUSH. A following tcdrain will hang forever. The bug was
reproduced and this fix was validated on Arduino Uno Q (QRB2210)
using /dev/ttyHS1.

Assisted-by: Claude:claude-5-opus Codex:gpt-5
Signed-off-by: Jan Sebastian Götte <linux@jaseg.de>
Fixes: 2aaa43c70778 ("tty: serial: qcom-geni-serial: add support for serial engine DMA")
Cc: stable <stable@kernel.org>
Reviewed-by: Praveen Talari <praveen.talari@oss.qualcomm.com>
Link: https://patch.msgid.link/20260729174105.21838-2-git@jaseg.de
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drivers/tty/serial/qcom_geni_serial.c