2016-11-25 11:11:09

by Anthony Romano

Subject: [PATCH] mt7601u: wait for clear rxq when stopping mac

mt7601u_mac_stop_hw should stop polling the rxq once it remains empty
but instead continues polling after the rxq status stays clear; bringing
down the interface takes about six seconds from this alone.

Speed up path by exiting rxq loop once status repeatedly polls empty.

Signed-off-by: Anthony Romano <[email protected]>
---
drivers/net/wireless/mediatek/mt7601u/init.c | 14 +++++++-------
drivers/net/wireless/mediatek/mt7601u/regs.h | 3 +++
2 files changed, 10 insertions(+), 7 deletions(-)

diff --git a/drivers/net/wireless/mediatek/mt7601u/init.c b/drivers/net/wireless/mediatek/mt7601u/init.c
index 44d46e25db80..a6e901766226 100644
--- a/drivers/net/wireless/mediatek/mt7601u/init.c
+++ b/drivers/net/wireless/mediatek/mt7601u/init.c
@@ -293,13 +293,13 @@ static void mt7601u_mac_stop_hw(struct mt7601u_dev *dev)
ok = 0;
i = 200;
while (i--) {
- if ((mt76_rr(dev, 0x0430) & 0x00ff0000) ||
- (mt76_rr(dev, 0x0a30) & 0xffffffff) ||
- (mt76_rr(dev, 0x0a34) & 0xffffffff))
- ok++;
- if (ok > 6)
- break;
-
+ if (!(mt76_rr(dev, MT_RXQ_STA) & 0x00ff0000) &&
+ !mt76_rr(dev, 0x0a30) &&
+ !mt76_rr(dev, 0x0a34)) {
+ if (ok++ > 5)
+ break;
+ continue;
+ }
msleep(1);
}

diff --git a/drivers/net/wireless/mediatek/mt7601u/regs.h b/drivers/net/wireless/mediatek/mt7601u/regs.h
index 27a429d90cec..2a8837002f00 100644
--- a/drivers/net/wireless/mediatek/mt7601u/regs.h
+++ b/drivers/net/wireless/mediatek/mt7601u/regs.h
@@ -192,6 +192,9 @@
#define MT_BCN_OFFSET_BASE 0x041c
#define MT_BCN_OFFSET(_n) (MT_BCN_OFFSET_BASE + ((_n) << 2))

+#define MT_RXQ_STA 0x0430
+#define MT_TXQ_STA 0x0434
+
#define MT_RF_CSR_CFG 0x0500
#define MT_RF_CSR_CFG_DATA GENMASK(7, 0)
#define MT_RF_CSR_CFG_REG_ID GENMASK(13, 8)
--
2.11.0.rc2

2016-11-25 17:13:23

by Jakub Kicinski

Subject: Re: [PATCH] mt7601u: wait for clear rxq when stopping mac

On Fri, 25 Nov 2016 03:13:34 -0800, Anthony Romano wrote:
> mt7601u_mac_stop_hw should stop polling the rxq once it remains empty
> but instead continues polling after the rxq status stays clear; bringing
> down the interface takes about six seconds from this alone.
>
> Speed up path by exiting rxq loop once status repeatedly polls empty.
>
> Signed-off-by: Anthony Romano <[email protected]>

Looks correct, the condition was inverted and we don't have to
msleep() when we see values go to zero. Thanks!

Reviewed-by: Jakub Kicinski <[email protected]>

2016-11-29 15:32:55

by Kalle Valo

Subject: Re: mt7601u: wait for clear rxq when stopping mac

Anthony Romano <[email protected]> wrote:
> mt7601u_mac_stop_hw should stop polling the rxq once it remains empty
> but instead continues polling after the rxq status stays clear; bringing
> down the interface takes about six seconds from this alone.
>
> Speed up path by exiting rxq loop once status repeatedly polls empty.
>
> Signed-off-by: Anthony Romano <[email protected]>
> Reviewed-by: Jakub Kicinski <[email protected]>

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

05db221e30a9 mt7601u: wait for clear rxq when stopping mac

--
https://patchwork.kernel.org/patch/9447371/

Documentation about submitting wireless patches and checking status
from patchwork:

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