[PATCH] spi: mvebu: use wait_on_timeout instead of a loop with udelay.

[PATCH] spi: mvebu: use wait_on_timeout instead of a loop with udelay.

[Uwe Kleine-König]

This looks nicer and reduces the time to transfer 40 MB at 50 MHz from
203 seconds to 87 seconds.

Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
---
Side note:

When using

	while(!(readl(p->base + SPI_IF_CTRL) & IF_READ_READY))
		;
	return 0;

instead, the time goes down to 24 seconds.

I think the culprit is that get_time_ns (which is used in
wait_on_timeout) handles conversion from timer values to ns.

I didn't look into that (yet) but I think this could be improved by
changing from:

	int __ret = 0;
	uint64_t __to_start = get_time_ns();

	while (!(condition)) {
		if (is_timeout(__to_start, (timeout))) {
			__ret = -ETIMEOUT;
			break;
		}
	}
	return __ret;

to something like:

	int __ret = 0;
	uint64_t __to_start_cyc = get_time_cyc();
	uint64_t __timeout_cyc = ns2cyc(timeout);

	while (!(condition)) {
		if (is_timeout_cyc(__to_start_cyc, __timeout_cyc)) {
			__ret = -ETIMEOUT;
			break;
		}
	}
	return __ret;

Here only a single conversion between cyc and ns is done and
is_timeout_cyc is cheap as it doesn't need to convert.

The obvious downside is that time_ns isn't updated that often. But it
could still be worth the effort.

Best regards
Uwe

 drivers/spi/mvebu_spi.c | 12 +++++-------
 1 file changed, 5 insertions(+), 7 deletions(-)

diff --git a/drivers/spi/mvebu_spi.c b/drivers/spi/mvebu_spi.c
index 68838c0ce184..f24508a7bc5d 100644
--- a/drivers/spi/mvebu_spi.c
+++ b/drivers/spi/mvebu_spi.c
@@ -249,13 +249,11 @@ static int mvebu_spi_setup(struct spi_device *spi)
 
 static inline int mvebu_spi_wait_for_read_ready(struct mvebu_spi *p)
 {
-	int timeout = 100;
-	while ((readl(p->base + SPI_IF_CTRL) & IF_READ_READY) == 0 &&
-		timeout--)
-		udelay(1);
-	if (timeout < 0)
-		return -EIO;
-	return 0;
+	int ret;
+
+	ret = wait_on_timeout(100 * USECOND,
+			      readl(p->base + SPI_IF_CTRL) & IF_READ_READY);
+	return ret;
 }
 
 static int mvebu_spi_do_transfer(struct spi_device *spi,
-- 
2.10.2


_______________________________________________
barebox mailing list
barebox@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/barebox

Re: [PATCH] spi: mvebu: use wait_on_timeout instead of a loop with udelay.

[Sascha Hauer]

On Thu, Dec 08, 2016 at 11:20:50AM +0100, Uwe Kleine-König wrote:
> This looks nicer and reduces the time to transfer 40 MB at 50 MHz from
> 203 seconds to 87 seconds.
> 
> Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
> ---
> Side note:
> 
> When using
> 
> 	while(!(readl(p->base + SPI_IF_CTRL) & IF_READ_READY))
> 		;
> 	return 0;
> 
> instead, the time goes down to 24 seconds.

Have you tried how many loop iterations it takes to complete? Maybe it's
worth it to try a few times and only after that start a real timeout
loop.

Sascha


-- 
Pengutronix e.K.                           |                             |
Industrial Linux Solutions                 | http://www.pengutronix.de/  |
Peiner Str. 6-8, 31137 Hildesheim, Germany | Phone: +49-5121-206917-0    |
Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-5555 |

_______________________________________________
barebox mailing list
barebox@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/barebox