[PATCH v2] serial: qcom-geni: Set UPSTAT_AUTOCTS for hardware flow control

[Michael McAllister]

qcom_geni_serial_set_termios() clears UART_CTS_MASK when CRTSCTS is set,
enabling hardware CTS flow control. However, it never sets UPSTAT_AUTOCTS,
so uart_softcts_mode() also selects software-assisted CTS flow control.

If CTS is deasserted when uart_change_line_settings() samples the modem
status, serial core sets hw_stopped and stops transmission. The GENI
driver does not handle CTS-change interrupts, so asserting CTS again does
not clear hw_stopped or restart the transmitter.

On a Motorola Moto G10 with WCN3950, this can stall Bluetooth setup in
tty_wait_until_sent() while qca_setup() closes the serial port. A captured
stall had one byte left in the software transmit queue, hw_stopped set,
and CTS asserted, with the UART powered and its interrupt enabled.

Set UPSTAT_AUTOCTS under the port lock when CRTSCTS is enabled, and clear
it when CRTSCTS is disabled, to match the hardware configuration.

Fixes: 8a8a66a1a18a ("tty: serial: qcom_geni_serial: Add support for flow control")
Cc: stable@xxxxxxxxxxxxxxx
Assisted-by: LLM
Signed-off-by: Michael McAllister <me@xxxxxxxxxxxxxxxxxxxxx>
---
Changes in v2:
- Use scoped_guard(uart_port_lock_irqsave, uport), as suggested by Jiri Slaby,
  removing the flags variable and explicit lock/unlock calls.

v1: https://lore.kernel.org/linux-serial/20261021202603.42888-1-me@xxxxxxxxxxxxxxxxxxxxx/

Hardware validation used v1 with explicit lock/unlock calls in the Motorola
Moto G10 Linux 7.2.9 port. This revision preserves the same lock scope.

Before the fix, a Bluetooth power-cycle test captured qca_setup() blocked
in tty_wait_until_sent() through serdev_device_close(), with:

  software TX pending: 1 byte
  hw_stopped:          1
  UART status:        0x1 (UPSTAT_CTS_ENABLE without UPSTAT_AUTOCTS)
  SE_GENI_IOS:         0x1 (IO2_DATA_IN clear, CTS asserted)
  UART IRQ disabled:  0
  UART IRQ masked:    0

After the fix RTC suspend/resume worked, followed by Bluetooth discovery.

Built drivers/tty/serial/qcom_geni_serial.o for arm64 with W=1 and
CONFIG_SERIAL_QCOM_GENI_CONSOLE=y against the base below.

AI assistance: Codex analyzed the saved UART diagnostics and serial-core
source and ran the build and style checks.

 drivers/tty/serial/qcom_geni_serial.c | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/drivers/tty/serial/qcom_geni_serial.c b/drivers/tty/serial/qcom_geni_serial.c
index a851b8f0ef..aff26a8351 100644
--- a/drivers/tty/serial/qcom_geni_serial.c
+++ b/drivers/tty/serial/qcom_geni_serial.c
@@ -1596,6 +1596,14 @@ static void qcom_geni_serial_set_termios(struct uart_port *uport,
 	writel(bits_per_char, uport->membase + SE_UART_RX_WORD_LEN);
 	writel(stop_bit_len, uport->membase + SE_UART_TX_STOP_BIT_LEN);
 
+	/* CTS gates TX in hardware; no CTS interrupt updates hw_stopped. */
+	scoped_guard(uart_port_lock_irqsave, uport) {
+		if (termios->c_cflag & CRTSCTS)
+			uport->status |= UPSTAT_AUTOCTS;
+		else
+			uport->status &= ~UPSTAT_AUTOCTS;
+	}
+
 	trace_geni_serial_set_termios(uport->dev, baud, bits_per_char,
 				      tx_trans_cfg, tx_parity_cfg, rx_trans_cfg,
 				      rx_parity_cfg, stop_bit_len);

base-commit: a90ee4305c4a5df72c11b31dacfdc76e00fcf78a
-- 
2.50.1 (Apple Git-155)