serial: core: Clean up uart_update_timeout() function

Rename the variable size to temp and change its data type from
unsigned int to u64 to avoid type casting in multiplication. Remove the
intermediate variable frame_time and use temp instead to accommodate
the nanoseconds(ns). port->frame_time is an unsigned int, therefore an
explicit cast is used to improve readability. Having said this unsigned
int is sufficinet to hold frame time duration in nanoseconds for all
the standard baudrates.

Consider 9600 baud, it takes 1/9600 seconds for one bit, for a total of
10 bits (start, 8-bit data, stop) 10/9600=1.04 ms for 1 byte transfer,
frame_time here is 1041667ns. As baudrate increases frame_time
decreases, say for 115200 baud it is 86806ns.

To avoid costly 64-bit arithmetic we do not upconvert the type for
variable frame_time as overflow happens for extremely low baudrates
which are impractical and are not used in real-world applications.

Signed-off-by: Vamshi Gajjela <vamshigajjela@google.com>
Link: https://lore.kernel.org/r/20231109063417.3971005-3-vamshigajjela@google.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

1 files changed, 3 insertions, 4 deletions

diff --git a/drivers/tty/serial/serial_core.c b/drivers/tty/serial/serial_core.c
index 80085b151b34..1147fb1b39f2 100644
--- a/drivers/tty/serial/serial_core.c
+++ b/drivers/tty/serial/serial_core.c
@@ -410,11 +410,10 @@ void
 uart_update_timeout(struct uart_port *port, unsigned int cflag,
 		    unsigned int baud)
 {
-	unsigned int size = tty_get_frame_size(cflag);
-	u64 frame_time;
+	u64 temp = tty_get_frame_size(cflag);
 
-	frame_time = (u64)size * NSEC_PER_SEC;
-	port->frame_time = DIV64_U64_ROUND_UP(frame_time, baud);
+	temp *= NSEC_PER_SEC;
+	port->frame_time = (unsigned int)DIV64_U64_ROUND_UP(temp, baud);
 }
 EXPORT_SYMBOL(uart_update_timeout);