On 31/12/17 02:50, Guenter Roeck wrote:
> The piix4 i2c driver is extremely slow. Replacing msleep()
> with usleep_range() increases its speed substantially.
>
> Signed-off-by: Guenter Roeck <linux@xxxxxxxxxxxx>
> ---
> drivers/i2c/busses/i2c-piix4.c | 6 +++---
> 1 file changed, 3 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/i2c/busses/i2c-piix4.c b/drivers/i2c/busses/i2c-piix4.c
> index 78dd5951d6e7..52a8b1c5c110 100644
> --- a/drivers/i2c/busses/i2c-piix4.c
> +++ b/drivers/i2c/busses/i2c-piix4.c
> @@ -467,13 +467,13 @@ static int piix4_transaction(struct i2c_adapter *piix4_adapter)
>
> /* We will always wait for a fraction of a second! (See PIIX4 docs errata) */
> if (srvrworks_csb5_delay) /* Extra delay for SERVERWORKS_CSB5 */
> - msleep(2);
> + usleep_range(2000, 2000);
> else
> - msleep(1);
> + usleep_range(500, 1000);
>
> while ((++timeout < MAX_TIMEOUT) &&
> ((temp = inb_p(SMBHSTSTS)) & 0x01))
> - msleep(1);
> + usleep_range(200, 500);
>
> /* If the SMBus is still busy, we give up */
> if (timeout == MAX_TIMEOUT) {
>
Thanks for this patch.
FWIW, this also makes a noticeable difference on AMD Family 16h Model 30h, used in embedded designs and also commonly available as AMD GX-412TC SoC in the PC Engines APU2.
Among the tests I did were reading from the SoC temperature sensor in a loop:
while [ true ] ; do i2cget -y 0 0x4C 0x01 ; done
and scanning for peripherals in a loop:
while [ true ] ; do i2cdetect -y 0 ; done
These tests may be artificial and trivial, but the speedup matters to us because we have more than one bus master and the embedded controller needs to poll multiple sensors.
Tested-by: Andrew Cooks <andrew.cooks@xxxxxxxxxxxx>
