[PATCH] mt76x0: run calibration after scanning

[PATCH] mt76x0: run calibration after scanning

[Stanislaw Gruszka]

If we are associated and scanning is performed , sw_scan_complete callback
is done after we get back to operating channel, so we do not perform
phy calibration and queue cal work. Fix this by run calibration from
sw_scan_complete().

Fixes: bbd10586f0df ("mt76x0: phy: do not run calibration during channel switch")
Signed-off-by: Stanislaw Gruszka <sgruszka@redhat.com>
---
This is for 4.20

 drivers/net/wireless/mediatek/mt76/mt76x0/main.c   | 19 +++++++++++++++++++
 drivers/net/wireless/mediatek/mt76/mt76x0/mt76x0.h |  2 ++
 drivers/net/wireless/mediatek/mt76/mt76x0/pci.c    |  2 +-
 drivers/net/wireless/mediatek/mt76/mt76x0/usb.c    |  2 +-
 4 files changed, 23 insertions(+), 2 deletions(-)

diff --git a/drivers/net/wireless/mediatek/mt76/mt76x0/main.c b/drivers/net/wireless/mediatek/mt76/mt76x0/main.c
index a803a9b6a4c5..896a9d46bb76 100644
--- a/drivers/net/wireless/mediatek/mt76/mt76x0/main.c
+++ b/drivers/net/wireless/mediatek/mt76/mt76x0/main.c
@@ -78,3 +78,22 @@ int mt76x0_config(struct ieee80211_hw *hw, u32 changed)
 	return ret;
 }
 EXPORT_SYMBOL_GPL(mt76x0_config);
+
+void mt76x0_sw_scan_complete(struct ieee80211_hw *hw,
+			     struct ieee80211_vif *vif)
+{
+	struct mt76x02_dev *dev = hw->priv;
+
+	mutex_lock(&dev->mt76.mutex);
+	if (vif->bss_conf.assoc) {
+		mt76x0_phy_calibrate(dev, false);
+		mt76x0_phy_set_txpower(dev);
+
+		ieee80211_queue_delayed_work(dev->mt76.hw, &dev->cal_work,
+					     MT_CALIBRATE_INTERVAL);
+	}
+	mutex_unlock(&dev->mt76.mutex);
+
+	mt76x02_sw_scan_complete(hw, vif);
+}
+EXPORT_SYMBOL_GPL(mt76x0_sw_scan_complete);
diff --git a/drivers/net/wireless/mediatek/mt76/mt76x0/mt76x0.h b/drivers/net/wireless/mediatek/mt76/mt76x0/mt76x0.h
index 46629f61673b..d850a6702598 100644
--- a/drivers/net/wireless/mediatek/mt76/mt76x0/mt76x0.h
+++ b/drivers/net/wireless/mediatek/mt76/mt76x0/mt76x0.h
@@ -62,6 +62,8 @@ int mt76x0_mac_start(struct mt76x02_dev *dev);
 void mt76x0_mac_stop(struct mt76x02_dev *dev);
 
 int mt76x0_config(struct ieee80211_hw *hw, u32 changed);
+void mt76x0_sw_scan_complete(struct ieee80211_hw *hw,
+			     struct ieee80211_vif *vif);
 
 /* PHY */
 void mt76x0_phy_init(struct mt76x02_dev *dev);
diff --git a/drivers/net/wireless/mediatek/mt76/mt76x0/pci.c b/drivers/net/wireless/mediatek/mt76/mt76x0/pci.c
index 757816d2b1ab..854d882f775b 100644
--- a/drivers/net/wireless/mediatek/mt76/mt76x0/pci.c
+++ b/drivers/net/wireless/mediatek/mt76/mt76x0/pci.c
@@ -95,7 +95,7 @@ static const struct ieee80211_ops mt76x0e_ops = {
 	.set_key = mt76x02_set_key,
 	.conf_tx = mt76x02_conf_tx,
 	.sw_scan_start = mt76x02_sw_scan,
-	.sw_scan_complete = mt76x02_sw_scan_complete,
+	.sw_scan_complete = mt76x0_sw_scan_complete,
 	.ampdu_action = mt76x02_ampdu_action,
 	.sta_rate_tbl_update = mt76x02_sta_rate_tbl_update,
 	.wake_tx_queue = mt76_wake_tx_queue,
diff --git a/drivers/net/wireless/mediatek/mt76/mt76x0/usb.c b/drivers/net/wireless/mediatek/mt76/mt76x0/usb.c
index 35b98661632a..9a83d9f62012 100644
--- a/drivers/net/wireless/mediatek/mt76/mt76x0/usb.c
+++ b/drivers/net/wireless/mediatek/mt76/mt76x0/usb.c
@@ -151,7 +151,7 @@ static const struct ieee80211_ops mt76x0u_ops = {
 	.set_key = mt76x02_set_key,
 	.conf_tx = mt76x02_conf_tx,
 	.sw_scan_start = mt76x02_sw_scan,
-	.sw_scan_complete = mt76x02_sw_scan_complete,
+	.sw_scan_complete = mt76x0_sw_scan_complete,
 	.ampdu_action = mt76x02_ampdu_action,
 	.sta_rate_tbl_update = mt76x02_sta_rate_tbl_update,
 	.set_rts_threshold = mt76x02_set_rts_threshold,
-- 
2.7.5

Re: [PATCH] mt76x0: run calibration after scanning

[Kalle Valo]

Stanislaw Gruszka <sgruszka@redhat.com> writes:

> If we are associated and scanning is performed , sw_scan_complete callback
> is done after we get back to operating channel, so we do not perform
> phy calibration and queue cal work. Fix this by run calibration from
> sw_scan_complete().
>
> Fixes: bbd10586f0df ("mt76x0: phy: do not run calibration during channel switch")
> Signed-off-by: Stanislaw Gruszka <sgruszka@redhat.com>
> ---
> This is for 4.20

So what are the symptoms from user's point of view? Is this a
regression?

I'm just trying to understand if this really is needed for 4.20. There
are so many mt76 patches flowing to -next that the bar for mt76 -rc
patches is even higher than normally.

-- 
Kalle Valo

Re: [PATCH] mt76x0: run calibration after scanning

[Stanislaw Gruszka]

On Mon, Oct 29, 2018 at 04:25:31PM +0200, Kalle Valo wrote:
> Stanislaw Gruszka <sgruszka@redhat.com> writes:
> 
> > If we are associated and scanning is performed , sw_scan_complete callback
> > is done after we get back to operating channel, so we do not perform
> > phy calibration and queue cal work. Fix this by run calibration from
> > sw_scan_complete().
> >
> > Fixes: bbd10586f0df ("mt76x0: phy: do not run calibration during channel switch")
> > Signed-off-by: Stanislaw Gruszka <sgruszka@redhat.com>
> > ---
> > This is for 4.20
> 
> So what are the symptoms from user's point of view? Is this a
> regression?

We do not perform gain calibration any longer if somebody will
request scan after association. This is formally a regression,
but calibration code changed a lot, so perhaps this can be dropped
for 4.20 and eventually go through -stable if I can confirm
it fixes performance problems.

Thanks
Stanislaw

Re: [PATCH] mt76x0: run calibration after scanning

[Stanislaw Gruszka]

On Mon, Oct 29, 2018 at 11:16:53AM +0100, Stanislaw Gruszka wrote:
> If we are associated and scanning is performed , sw_scan_complete callback
> is done after we get back to operating channel, so we do not perform
> phy calibration and queue cal work. Fix this by run calibration from
> sw_scan_complete().
> 
> Fixes: bbd10586f0df ("mt76x0: phy: do not run calibration during channel switch")
> Signed-off-by: Stanislaw Gruszka <sgruszka@redhat.com>
> ---
> This is for 4.20
> 
>  drivers/net/wireless/mediatek/mt76/mt76x0/main.c   | 19 +++++++++++++++++++
>  drivers/net/wireless/mediatek/mt76/mt76x0/mt76x0.h |  2 ++
>  drivers/net/wireless/mediatek/mt76/mt76x0/pci.c    |  2 +-
>  drivers/net/wireless/mediatek/mt76/mt76x0/usb.c    |  2 +-
>  4 files changed, 23 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/net/wireless/mediatek/mt76/mt76x0/main.c b/drivers/net/wireless/mediatek/mt76/mt76x0/main.c
> index a803a9b6a4c5..896a9d46bb76 100644
> --- a/drivers/net/wireless/mediatek/mt76/mt76x0/main.c
> +++ b/drivers/net/wireless/mediatek/mt76/mt76x0/main.c
> @@ -78,3 +78,22 @@ int mt76x0_config(struct ieee80211_hw *hw, u32 changed)
>  	return ret;
>  }
>  EXPORT_SYMBOL_GPL(mt76x0_config);
> +
> +void mt76x0_sw_scan_complete(struct ieee80211_hw *hw,
> +			     struct ieee80211_vif *vif)
> +{
> +	struct mt76x02_dev *dev = hw->priv;
> +
> +	mutex_lock(&dev->mt76.mutex);
> +	if (vif->bss_conf.assoc) {
> +		mt76x0_phy_calibrate(dev, false);
> +		mt76x0_phy_set_txpower(dev);

Actually I don't think this part is needed, it is done after channel
switch just before we associate, and that seems to be sufficient.

> +
> +		ieee80211_queue_delayed_work(dev->mt76.hw, &dev->cal_work,
> +					     MT_CALIBRATE_INTERVAL);

We just have to schedule this work, and mt76x2 suffer from similar
issue, so this can be done in mt76x02_sw_scan_complete(). Will
post v2.

Thanks
Stanislaw

Re: [PATCH] mt76x0: run calibration after scanning

[Kalle Valo]

Stanislaw Gruszka <sgruszka@redhat.com> writes:

> On Mon, Oct 29, 2018 at 04:25:31PM +0200, Kalle Valo wrote:
>> Stanislaw Gruszka <sgruszka@redhat.com> writes:
>> 
>> > If we are associated and scanning is performed , sw_scan_complete callback
>> > is done after we get back to operating channel, so we do not perform
>> > phy calibration and queue cal work. Fix this by run calibration from
>> > sw_scan_complete().
>> >
>> > Fixes: bbd10586f0df ("mt76x0: phy: do not run calibration during channel switch")
>> > Signed-off-by: Stanislaw Gruszka <sgruszka@redhat.com>
>> > ---
>> > This is for 4.20
>> 
>> So what are the symptoms from user's point of view? Is this a
>> regression?
>
> We do not perform gain calibration any longer if somebody will
> request scan after association. This is formally a regression,
> but calibration code changed a lot, so perhaps this can be dropped
> for 4.20 and eventually go through -stable if I can confirm
> it fixes performance problems.

To me that would be a much better option, less conflicts that way.

-- 
Kalle Valo