[PATCH v2] serial: imx: reduce RX interrupt frequency

[PATCH v2] serial: imx: reduce RX interrupt frequency

[Tomasz Moń]

Triggering RX interrupt for every byte defeats the purpose of aging
timer and leads to interrupt storm at high baud rates. The interrupt
storm can starve line discipline worker and prevent tty throttling,
rendering hardware/software flow control useless.

Increase receiver trigger level to 8 to increase the minimum period
between RX interrupts to 8 characters time. The tradeoff is increased
latency.

Aging timer resets with every received character. Worst case scenario
happens when RX data intercharacter delay is slightly less than the
aging timer timeout (8 characters time). The upper bound of the time
a character can wait in RxFIFO before being read is:
  (RXTL - 1) * (8 character time timeout + received 1 character time)

Usually the data is received in frames, with low intercharacter delay.
In such case the latency increase is 8 characters time at the end of
the frame with probability (RXTL - 1) / RXTL.

Signed-off-by: Tomasz Moń <tomasz.mon@camlingroup.com>
Acked-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
---
changes in v2:
  - reworded commit message based on review comments
  - added Acked-by by Uwe tag from V1 review

v1: https://lore.kernel.org/linux-serial/20220104103203.2033673-1-tomasz.mon@camlingroup.com/
---
 drivers/tty/serial/imx.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/tty/serial/imx.c b/drivers/tty/serial/imx.c
index 90f82e6c54e4..3c812c47ecc0 100644
--- a/drivers/tty/serial/imx.c
+++ b/drivers/tty/serial/imx.c
@@ -1255,7 +1255,7 @@ static void imx_uart_clear_rx_errors(struct imx_port *sport)
 }
 
 #define TXTL_DEFAULT 2 /* reset default */
-#define RXTL_DEFAULT 1 /* reset default */
+#define RXTL_DEFAULT 8 /* 8 characters or aging timer */
 #define TXTL_DMA 8 /* DMA burst setting */
 #define RXTL_DMA 9 /* DMA burst setting */
 
-- 
2.25.1

Re: [PATCH v2] serial: imx: reduce RX interrupt frequency

[Tomasz Moń]

On 14.01.2022 08:44, Tomasz Moń wrote:
> Aging timer resets with every received character. Worst case scenario
> happens when RX data intercharacter delay is slightly less than the
> aging timer timeout (8 characters time). The upper bound of the time
> a character can wait in RxFIFO before being read is:
>   (RXTL - 1) * (8 character time timeout + received 1 character time)

s/being read/interrupt is raised/

This description is not precise. The character can wait in RxFIFO longer
than the mentioned upper bound, as the time it takes to actually read it
from RxFIFO is the sum of how long it takes to raise the interrupt and
actual interrupt handling time.

Best Regards,
Tomasz Mon