[PATCH] rtw88: use shorter delay time to poll PS state

[PATCH] rtw88: use shorter delay time to poll PS state

[yhchuang]

From: Yan-Hsuan Chuang <yhchuang@realtek.com>

When TX packet arrives, driver should leave deep PS state to make
sure the DMA is working. After requested to leave deep PS state,
driver needs to poll the PS state to check if the mode has been
changed successfully. The driver used to check the state of the
hardware every 20 msecs, which means upon the first failure of
state check, the CPU is delayed 20 msecs for next check. This is
harmful for some time-sensitive applications such as media players.

So, use shorter delay time each check from 20 msecs to 100 usecs.
The state should be changed in several tries. But we still need
to reserve ~15 msecs in total in case of the state just took too
long to be changed successfully. If the states of driver and the
hardware is not synchronized, the power state could be locked
forever, which mean we could never enter/leave the PS state.

Signed-off-by: Yan-Hsuan Chuang <yhchuang@realtek.com>
---
 drivers/net/wireless/realtek/rtw88/ps.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/drivers/net/wireless/realtek/rtw88/ps.c b/drivers/net/wireless/realtek/rtw88/ps.c
index 913e6f47130f..7a189a9926fe 100644
--- a/drivers/net/wireless/realtek/rtw88/ps.c
+++ b/drivers/net/wireless/realtek/rtw88/ps.c
@@ -91,11 +91,11 @@ void rtw_power_mode_change(struct rtw_dev *rtwdev, bool enter)
 			return;
 
 		/* check confirm power mode has left power save state */
-		for (polling_cnt = 0; polling_cnt < 3; polling_cnt++) {
+		for (polling_cnt = 0; polling_cnt < 50; polling_cnt++) {
 			polling = rtw_read8(rtwdev, rtwdev->hci.cpwm_addr);
 			if ((polling ^ confirm) & BIT_RPWM_TOGGLE)
 				return;
-			mdelay(20);
+			udelay(100);
 		}
 
 		/* in case of fw/hw missed the request, retry */
-- 
2.17.1

Re: [PATCH] rtw88: use shorter delay time to poll PS state

[Chris Chiu]

On Tue, Jan 7, 2020 at 10:27 PM <yhchuang@realtek.com> wrote:
>
> From: Yan-Hsuan Chuang <yhchuang@realtek.com>
>
> When TX packet arrives, driver should leave deep PS state to make
> sure the DMA is working. After requested to leave deep PS state,
> driver needs to poll the PS state to check if the mode has been
> changed successfully. The driver used to check the state of the
> hardware every 20 msecs, which means upon the first failure of
> state check, the CPU is delayed 20 msecs for next check. This is
> harmful for some time-sensitive applications such as media players.
>
> So, use shorter delay time each check from 20 msecs to 100 usecs.
> The state should be changed in several tries. But we still need
> to reserve ~15 msecs in total in case of the state just took too
> long to be changed successfully. If the states of driver and the
> hardware is not synchronized, the power state could be locked
> forever, which mean we could never enter/leave the PS state.
>
> Signed-off-by: Yan-Hsuan Chuang <yhchuang@realtek.com>
Reviewed-by: Chris Chiu <chiu@endlessm.com>
> ---
>  drivers/net/wireless/realtek/rtw88/ps.c | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/net/wireless/realtek/rtw88/ps.c b/drivers/net/wireless/realtek/rtw88/ps.c
> index 913e6f47130f..7a189a9926fe 100644
> --- a/drivers/net/wireless/realtek/rtw88/ps.c
> +++ b/drivers/net/wireless/realtek/rtw88/ps.c
> @@ -91,11 +91,11 @@ void rtw_power_mode_change(struct rtw_dev *rtwdev, bool enter)
>                         return;
>
>                 /* check confirm power mode has left power save state */
> -               for (polling_cnt = 0; polling_cnt < 3; polling_cnt++) {
> +               for (polling_cnt = 0; polling_cnt < 50; polling_cnt++) {
>                         polling = rtw_read8(rtwdev, rtwdev->hci.cpwm_addr);
>                         if ((polling ^ confirm) & BIT_RPWM_TOGGLE)
>                                 return;
> -                       mdelay(20);
> +                       udelay(100);
>                 }
>
>                 /* in case of fw/hw missed the request, retry */
> --
> 2.17.1
>

Re: [PATCH] rtw88: use shorter delay time to poll PS state

[Kalle Valo]

<yhchuang@realtek.com> wrote:

> From: Yan-Hsuan Chuang <yhchuang@realtek.com>
> 
> When TX packet arrives, driver should leave deep PS state to make
> sure the DMA is working. After requested to leave deep PS state,
> driver needs to poll the PS state to check if the mode has been
> changed successfully. The driver used to check the state of the
> hardware every 20 msecs, which means upon the first failure of
> state check, the CPU is delayed 20 msecs for next check. This is
> harmful for some time-sensitive applications such as media players.
> 
> So, use shorter delay time each check from 20 msecs to 100 usecs.
> The state should be changed in several tries. But we still need
> to reserve ~15 msecs in total in case of the state just took too
> long to be changed successfully. If the states of driver and the
> hardware is not synchronized, the power state could be locked
> forever, which mean we could never enter/leave the PS state.
> 
> Signed-off-by: Yan-Hsuan Chuang <yhchuang@realtek.com>
> Reviewed-by: Chris Chiu <chiu@endlessm.com>

Patch applied to wireless-drivers-next.git, thanks.

6b6c150b8464 rtw88: use shorter delay time to poll PS state

-- 
https://patchwork.kernel.org/patch/11320947/

https://wireless.wiki.kernel.org/en/developers/documentation/submittingpatches