Re: Ap mode regression in linux-5.3-rc1 in rt2800usb

Re: Ap mode regression in linux-5.3-rc1 in rt2800usb

[Stanislaw Gruszka]

(cc linux-wireless mailing list)

On Tue, Aug 13, 2019 at 09:50:00PM +0300, Emil Karlson wrote:
> Greetings
> 
> After upgrading my ap running rt2800usb to linux-5.3-rc1 I noticed an
> unusual problem of not being able to connect to my ap with my android
> devices (nexus7/flo and nexus5x/bullhead), from tcpdump it seemed ap
> was receiving packets from the android devices after successful
> association, but android devices were not seeing the dhcp replies.
> 
> I reverted drivers/net/wireless/ralink to the state it is in v5.2.8 and
> android clients can connect again normally. I did not explicitly set
> watchdog parameter to any value.
> 
> Do you have any insights or is this already fixed?

Most likely is not fixed. We have those new commits in 5.3:

41a531ffa4c5 rt2x00usb: fix rx queue hang
0f47aeeada2a rt2800: do not enable watchdog by default
e403fa31ed71 rt2x00: add restart hw
710e6cc1595e rt2800: do not nullify initialization vector data
09db3b000619 rt2800: add pre_reset_hw callback
759c5b599cf4 rt2800: initial watchdog implementation
2034afe4db4a rt2800: add helpers for reading dma done index
9f3e3323e996 rt2x00: allow to specify watchdog interval

Most suspicious are 710e6cc1595e and 41a531ffa4c5 .

Could you test by reverting one single commit on 5.3
(first 710e6cc1595e and if that not help 41a531ffa4c5)
and check if it makes the problem gone?

Stanislaw

Re: Ap mode regression in linux-5.3-rc1 in rt2800usb

[Emil Karlson]

Greetings

On Wed, 14 Aug 2019 10:50:19 +0200
Stanislaw Gruszka <sgruszka@redhat.com> wrote:

> (cc linux-wireless mailing list)
> 
> On Tue, Aug 13, 2019 at 09:50:00PM +0300, Emil Karlson wrote:
> > Greetings
> > 
> > After upgrading my ap running rt2800usb to linux-5.3-rc1 I noticed
> > an unusual problem of not being able to connect to my ap with my
> > android devices (nexus7/flo and nexus5x/bullhead), from tcpdump it
> > seemed ap was receiving packets from the android devices after
> > successful association, but android devices were not seeing the
> > dhcp replies.
> > 
> > I reverted drivers/net/wireless/ralink to the state it is in v5.2.8
> > and android clients can connect again normally. I did not
> > explicitly set watchdog parameter to any value.

> Most suspicious are 710e6cc1595e and 41a531ffa4c5 .

It seems to me that reverting only
710e6cc1595e25378c4b9977f7a8b4ad4a72a109
allows all my android devices to successfully connect to the internet.

Best Regards
-Emil

Re: Ap mode regression in linux-5.3-rc1 in rt2800usb

[Stanislaw Gruszka]

[-- Attachment #1: Type: text/plain, Size: 1212 bytes --]

On Fri, Aug 16, 2019 at 11:00:12AM +0300, Emil Karlson wrote:
> Greetings
> 
> On Wed, 14 Aug 2019 10:50:19 +0200
> Stanislaw Gruszka <sgruszka@redhat.com> wrote:
> 
> > (cc linux-wireless mailing list)
> > 
> > On Tue, Aug 13, 2019 at 09:50:00PM +0300, Emil Karlson wrote:
> > > Greetings
> > > 
> > > After upgrading my ap running rt2800usb to linux-5.3-rc1 I noticed
> > > an unusual problem of not being able to connect to my ap with my
> > > android devices (nexus7/flo and nexus5x/bullhead), from tcpdump it
> > > seemed ap was receiving packets from the android devices after
> > > successful association, but android devices were not seeing the
> > > dhcp replies.
> > > 
> > > I reverted drivers/net/wireless/ralink to the state it is in v5.2.8
> > > and android clients can connect again normally. I did not
> > > explicitly set watchdog parameter to any value.
> 
> > Most suspicious are 710e6cc1595e and 41a531ffa4c5 .
> 
> It seems to me that reverting only
> 710e6cc1595e25378c4b9977f7a8b4ad4a72a109
> allows all my android devices to successfully connect to the internet.

Please test attached patch as proposed fix for
710e6cc1595e25378c4b9977f7a8b4ad4a72a109 and report back. Thanks.

Stanislaw

[-- Attachment #2: rt2x00_clear_ivs_on_start.patch --]
[-- Type: text/plain, Size: 2698 bytes --]

diff --git a/drivers/net/wireless/ralink/rt2x00/rt2800lib.c b/drivers/net/wireless/ralink/rt2x00/rt2800lib.c
index 4a996550288e..cbec2131e943 100644
--- a/drivers/net/wireless/ralink/rt2x00/rt2800lib.c
+++ b/drivers/net/wireless/ralink/rt2x00/rt2800lib.c
@@ -6095,6 +6095,15 @@ static int rt2800_init_registers(struct rt2x00_dev *rt2x00dev)
 	}
 
 	/*
+	 * Clear encryption initialization vectors on start, but keep them
+	 * for watchdog reset. Otherwise we will have wrong IVs and not be
+	 * able to keep connections after reset.
+	 */
+	if (!test_bit(DEVICE_STATE_RESET, &rt2x00dev->flags))
+		for (i = 0; i < 256; i++)
+			rt2800_register_write(rt2x00dev, MAC_IVEIV_ENTRY(i), 0);
+
+	/*
 	 * Clear all beacons
 	 */
 	for (i = 0; i < 8; i++)
diff --git a/drivers/net/wireless/ralink/rt2x00/rt2x00.h b/drivers/net/wireless/ralink/rt2x00/rt2x00.h
index d35ef06c5c7a..1dd54a0d083d 100644
--- a/drivers/net/wireless/ralink/rt2x00/rt2x00.h
+++ b/drivers/net/wireless/ralink/rt2x00/rt2x00.h
@@ -659,6 +659,7 @@ enum rt2x00_state_flags {
 	DEVICE_STATE_ENABLED_RADIO,
 	DEVICE_STATE_SCANNING,
 	DEVICE_STATE_FLUSHING,
+	DEVICE_STATE_RESET,
 
 	/*
 	 * Driver configuration
diff --git a/drivers/net/wireless/ralink/rt2x00/rt2x00dev.c b/drivers/net/wireless/ralink/rt2x00/rt2x00dev.c
index ad063c920323..c3eab767bc21 100644
--- a/drivers/net/wireless/ralink/rt2x00/rt2x00dev.c
+++ b/drivers/net/wireless/ralink/rt2x00/rt2x00dev.c
@@ -1253,13 +1253,14 @@ static int rt2x00lib_initialize(struct rt2x00_dev *rt2x00dev)
 
 int rt2x00lib_start(struct rt2x00_dev *rt2x00dev)
 {
-	int retval;
+	int retval = 0;
 
 	if (test_bit(DEVICE_STATE_STARTED, &rt2x00dev->flags)) {
 		/*
 		 * This is special case for ieee80211_restart_hw(), otherwise
 		 * mac80211 never call start() two times in row without stop();
 		 */
+		set_bit(DEVICE_STATE_RESET, &rt2x00dev->flags);
 		rt2x00dev->ops->lib->pre_reset_hw(rt2x00dev);
 		rt2x00lib_stop(rt2x00dev);
 	}
@@ -1270,14 +1271,14 @@ int rt2x00lib_start(struct rt2x00_dev *rt2x00dev)
 	 */
 	retval = rt2x00lib_load_firmware(rt2x00dev);
 	if (retval)
-		return retval;
+		goto out;
 
 	/*
 	 * Initialize the device.
 	 */
 	retval = rt2x00lib_initialize(rt2x00dev);
 	if (retval)
-		return retval;
+		goto out;
 
 	rt2x00dev->intf_ap_count = 0;
 	rt2x00dev->intf_sta_count = 0;
@@ -1286,11 +1287,13 @@ int rt2x00lib_start(struct rt2x00_dev *rt2x00dev)
 	/* Enable the radio */
 	retval = rt2x00lib_enable_radio(rt2x00dev);
 	if (retval)
-		return retval;
+		goto out;
 
 	set_bit(DEVICE_STATE_STARTED, &rt2x00dev->flags);
 
-	return 0;
+out:
+	clear_bit(DEVICE_STATE_RESET, &rt2x00dev->flags);
+	return retval;
 }
 
 void rt2x00lib_stop(struct rt2x00_dev *rt2x00dev)

Re: Ap mode regression in linux-5.3-rc1 in rt2800usb

[Emil Karlson]

Hi

On Fri, 16 Aug 2019 11:33:02 +0200
Stanislaw Gruszka <sgruszka@redhat.com> wrote:

> > > > After upgrading my ap running rt2800usb to linux-5.3-rc1 I
> > > > noticed an unusual problem of not being able to connect to my
> > > > ap with my android devices (nexus7/flo and nexus5x/bullhead),
> > > > from tcpdump it seemed ap was receiving packets from the
> > > > android devices after successful association, but android
> > > > devices were not seeing the dhcp replies.
> > > > 
> > > > I reverted drivers/net/wireless/ralink to the state it is in
> > > > v5.2.8 and android clients can connect again normally. I did not
> > > > explicitly set watchdog parameter to any value.
> > 
> > > Most suspicious are 710e6cc1595e and 41a531ffa4c5 .
> > 
> > It seems to me that reverting only
> > 710e6cc1595e25378c4b9977f7a8b4ad4a72a109
> > allows all my android devices to successfully connect to the
> > internet.
> 
> Please test attached patch as proposed fix for
> 710e6cc1595e25378c4b9977f7a8b4ad4a72a109 and report back. Thanks.

After some testing it seems attached patch fixes the regression with no
observed negative side effects. Thanks.

Best Regards
-Emil