net: dsa: mt7530: disable learning on standalone ports

This is a partial backport of commit 5a30833b9a16f8d1aa15de06636f9317ca51f9df
("net: dsa: mt7530: support MDB and bridge flag operations") upstream.

Make sure that the standalone ports start up with learning disabled.

Signed-off-by: DENG Qingfang <dqfext@gmail.com>
---
 drivers/net/dsa/mt7530.c | 12 ++++++++++--
 1 file changed, 10 insertions(+), 2 deletions(-)

diff --git a/drivers/net/dsa/mt7530.c b/drivers/net/dsa/mt7530.c
index 6335c4ea0957..67dfab774618 100644
--- a/drivers/net/dsa/mt7530.c
+++ b/drivers/net/dsa/mt7530.c

@@ -803,6 +803,8 @@ mt7530_port_bridge_join(struct dsa_switch *ds, int port,
 			   PCR_MATRIX_MASK, PCR_MATRIX(port_bitmap));
 	priv->ports[port].pm |= PCR_MATRIX(port_bitmap);
 
+	mt7530_clear(priv, MT7530_PSC_P(port), SA_DIS);
+
 	mutex_unlock(&priv->reg_mutex);
 
 	return 0;

@@ -907,6 +909,8 @@ mt7530_port_bridge_leave(struct dsa_switch *ds, int port,
 
 	mt7530_port_set_vlan_unaware(ds, port);
 
+	mt7530_set(priv, MT7530_PSC_P(port), SA_DIS);
+
 	mutex_unlock(&priv->reg_mutex);
 }
 

@@ -1287,11 +1291,15 @@ mt7530_setup(struct dsa_switch *ds)
 		mt7530_rmw(priv, MT7530_PCR_P(i), PCR_MATRIX_MASK,
 			   PCR_MATRIX_CLR);
 
-		if (dsa_is_cpu_port(ds, i))
+		if (dsa_is_cpu_port(ds, i)) {
 			mt7530_cpu_port_enable(priv, i);
-		else
+		} else {
 			mt7530_port_disable(ds, i, NULL);
 
+			/* Disable learning by default on all user ports */
+			mt7530_set(priv, MT7530_PSC_P(i), SA_DIS);
+		}
+
 		/* Enable consistent egress tag */
 		mt7530_rmw(priv, MT7530_PVC_P(i), PVC_EG_TAG_MASK,
 			   PVC_EG_TAG(MT7530_VLAN_EG_CONSISTENT));

-- 
2.25.1

On Tue, Aug 24, 2021 at 01:55:08PM +0800, DENG Qingfang wrote:

This is a partial backport of commit 5a30833b9a16f8d1aa15de06636f9317ca51f9df
("net: dsa: mt7530: support MDB and bridge flag operations") upstream.

Make sure that the standalone ports start up with learning disabled.

What's the reasoning behind:

1. Backporting this patch?
2. A partial backport of this patch?

-- 
Thanks,
Sasha

Hi Sasha,

On Tue, Aug 24, 2021 at 11:29 PM Sasha Levin [off-list ref] wrote:

What's the reasoning behind:

1. Backporting this patch?

Standalone ports should have address learning disabled, according to
the documentation:
https://www.kernel.org/doc/html/v5.14-rc7/networking/dsa/dsa.html#bridge-layer
dsa_switch_ops on 5.10 or earlier does not have .port_bridge_flags
function so it has to be done differently.

I've identified an issue related to this.

2. A partial backport of this patch?

The other part does not actually fix anything.

--
Thanks,
Sasha

On Tue, Aug 24, 2021 at 11:57:53PM +0800, DENG Qingfang wrote:

Hi Sasha,

On Tue, Aug 24, 2021 at 11:29 PM Sasha Levin [off-list ref] wrote:

quoted

What's the reasoning behind:

1. Backporting this patch?

Standalone ports should have address learning disabled, according to
the documentation:
https://www.kernel.org/doc/html/v5.14-rc7/networking/dsa/dsa.html#bridge-layer
dsa_switch_ops on 5.10 or earlier does not have .port_bridge_flags
function so it has to be done differently.

I've identified an issue related to this.

What issue is that?  Where was it reported?

quoted

2. A partial backport of this patch?

The other part does not actually fix anything.

Then why is it not ok to just take the whole thing?

When backporting not-identical-patches, something almost always goes
wrong, so we prefer to take the original commit when ever possible.

thanks,

greg k-h

On Fri, Aug 27, 2021 at 12:19:23PM +0200, Greg KH wrote:

On Tue, Aug 24, 2021 at 11:57:53PM +0800, DENG Qingfang wrote:

quoted

Standalone ports should have address learning disabled, according to
the documentation:
https://www.kernel.org/doc/html/v5.14-rc7/networking/dsa/dsa.html#bridge-layer
dsa_switch_ops on 5.10 or earlier does not have .port_bridge_flags
function so it has to be done differently.

I've identified an issue related to this.

What issue is that?  Where was it reported?

See Florian's message here
https://lore.kernel.org/stable/20210317003549.3964522-2-f.fainelli@gmail.com/

quoted

quoted

2. A partial backport of this patch?

The other part does not actually fix anything.

Then why is it not ok to just take the whole thing?

When backporting not-identical-patches, something almost always goes
wrong, so we prefer to take the original commit when ever possible.

Okay. MDB and tag ops can be backported as is, and broadcast/multicast
flooding can be implemented in .port_egress_floods. 

thanks,

greg k-h

On Thu, Sep 02, 2021 at 01:36:19PM +0800, DENG Qingfang wrote:

On Fri, Aug 27, 2021 at 12:19:23PM +0200, Greg KH wrote:

quoted

On Tue, Aug 24, 2021 at 11:57:53PM +0800, DENG Qingfang wrote:

quoted

Standalone ports should have address learning disabled, according to
the documentation:
https://www.kernel.org/doc/html/v5.14-rc7/networking/dsa/dsa.html#bridge-layer
dsa_switch_ops on 5.10 or earlier does not have .port_bridge_flags
function so it has to be done differently.

I've identified an issue related to this.

What issue is that?  Where was it reported?

See Florian's message here
https://lore.kernel.org/stable/20210317003549.3964522-2-f.fainelli@gmail.com/

THat is just the patch changelog text, or is it unique to this
stable-only patch?  It is not obvious at all.

quoted

quoted

quoted

2. A partial backport of this patch?

The other part does not actually fix anything.

Then why is it not ok to just take the whole thing?

When backporting not-identical-patches, something almost always goes
wrong, so we prefer to take the original commit when ever possible.

Okay. MDB and tag ops can be backported as is, and broadcast/multicast
flooding can be implemented in .port_egress_floods. 

So what are we supposed to do here?

totally confused,

greg k-h

On Thu, Sep 02, 2021 at 07:58:37AM +0200, Greg KH wrote:

On Thu, Sep 02, 2021 at 01:36:19PM +0800, DENG Qingfang wrote:

quoted

On Fri, Aug 27, 2021 at 12:19:23PM +0200, Greg KH wrote:

quoted

On Tue, Aug 24, 2021 at 11:57:53PM +0800, DENG Qingfang wrote:

quoted

Standalone ports should have address learning disabled, according to
the documentation:
https://www.kernel.org/doc/html/v5.14-rc7/networking/dsa/dsa.html#bridge-layer
dsa_switch_ops on 5.10 or earlier does not have .port_bridge_flags
function so it has to be done differently.

I've identified an issue related to this.

What issue is that?  Where was it reported?

See Florian's message here
https://lore.kernel.org/stable/20210317003549.3964522-2-f.fainelli@gmail.com/

THat is just the patch changelog text, or is it unique to this
stable-only patch?  It is not obvious at all.

The issue is with all DSA drivers that do not disable address learning
on standalone ports.

"With learning enabled we would end up with the switch having
incorrectly learned the address of the CPU port which typically results
in a complete break down of network connectivity until the address
learned ages out and gets re-learned, from the correct port this time."

quoted

quoted

quoted

quoted

2. A partial backport of this patch?

The other part does not actually fix anything.

Then why is it not ok to just take the whole thing?

When backporting not-identical-patches, something almost always goes
wrong, so we prefer to take the original commit when ever possible.

Okay. MDB and tag ops can be backported as is, and broadcast/multicast
flooding can be implemented in .port_egress_floods. 

So what are we supposed to do here?

Function port_egress_floods is refactored to port_bridge_flags in commit
a8b659e7ff75 ("net: dsa: act as passthrough for bridge port flags"). I can
backport the mt7530_port_bridge_flags function as port_egress_floods.

totally confused,

greg k-h

On Fri, Sep 03, 2021 at 05:14:30PM +0800, DENG Qingfang wrote:

On Thu, Sep 02, 2021 at 07:58:37AM +0200, Greg KH wrote:

quoted

On Thu, Sep 02, 2021 at 01:36:19PM +0800, DENG Qingfang wrote:

quoted

On Fri, Aug 27, 2021 at 12:19:23PM +0200, Greg KH wrote:

quoted

On Tue, Aug 24, 2021 at 11:57:53PM +0800, DENG Qingfang wrote:

quoted

Standalone ports should have address learning disabled, according to
the documentation:
https://www.kernel.org/doc/html/v5.14-rc7/networking/dsa/dsa.html#bridge-layer
dsa_switch_ops on 5.10 or earlier does not have .port_bridge_flags
function so it has to be done differently.

I've identified an issue related to this.

What issue is that?  Where was it reported?

See Florian's message here
https://lore.kernel.org/stable/20210317003549.3964522-2-f.fainelli@gmail.com/

THat is just the patch changelog text, or is it unique to this
stable-only patch?  It is not obvious at all.

The issue is with all DSA drivers that do not disable address learning
on standalone ports.

"With learning enabled we would end up with the switch having
incorrectly learned the address of the CPU port which typically results
in a complete break down of network connectivity until the address
learned ages out and gets re-learned, from the correct port this time."

quoted

quoted

quoted

quoted

quoted

2. A partial backport of this patch?

The other part does not actually fix anything.

Then why is it not ok to just take the whole thing?

When backporting not-identical-patches, something almost always goes
wrong, so we prefer to take the original commit when ever possible.

Okay. MDB and tag ops can be backported as is, and broadcast/multicast
flooding can be implemented in .port_egress_floods. 

6 more quoted lines

I am sorry, I still do not understand what to do here.

Ideally we want to take the original patches as get merged into Linus's
tree.  If that is not possible for some reason, we need to have it
documented very well why that is so, and to get everyone to agree with
the different patch that is submitted.

thanks,

greg k-h