[PATCH] net: dsa: mv88e6060: Fix false positive lockdep splat

[PATCH] net: dsa: mv88e6060: Fix false positive lockdep splat

[Neil Armstrong]

Like the change made for mv88e6xxx, use mutex_lock_nested() to avoid
lockdep to give false positives because of nested MDIO busses.

The false positive was observed using a mv88e6060 from a TI816X SoC.

Signed-off-by: Neil Armstrong <narmstrong@baylibre.com>
---
 drivers/net/dsa/mv88e6060.c | 19 +++++++++++++++++--
 1 file changed, 17 insertions(+), 2 deletions(-)

diff --git a/drivers/net/dsa/mv88e6060.c b/drivers/net/dsa/mv88e6060.c
index c29aebe..b1db460 100644
--- a/drivers/net/dsa/mv88e6060.c
+++ b/drivers/net/dsa/mv88e6060.c
@@ -19,14 +19,24 @@
 #define REG_PORT(p)		(8 + (p))
 #define REG_GLOBAL		0x0f

+/* MDIO bus access can be nested in the case of PHYs connected to the
+ * internal MDIO bus of the switch, which is accessed via MDIO bus of
+ * the Ethernet interface. Avoid lockdep false positives by using
+ * mutex_lock_nested().
+ */
 static int reg_read(struct dsa_switch *ds, int addr, int reg)
 {
+	int ret;
 	struct mii_bus *bus = dsa_host_dev_to_mii_bus(ds->master_dev);

 	if (bus == NULL)
 		return -EINVAL;

-	return mdiobus_read(bus, ds->pd->sw_addr + addr, reg);
+	mutex_lock_nested(&bus->mdio_lock, SINGLE_DEPTH_NESTING);
+	ret = bus->read(bus, ds->pd->sw_addr, reg);
+	mutex_unlock(&bus->mdio_lock);
+
+	return ret;
 }

 #define REG_READ(addr, reg)					\
@@ -42,12 +52,17 @@ static int reg_read(struct dsa_switch *ds, int addr, int reg)

 static int reg_write(struct dsa_switch *ds, int addr, int reg, u16 val)
 {
+	int ret;
 	struct mii_bus *bus = dsa_host_dev_to_mii_bus(ds->master_dev);

 	if (bus == NULL)
 		return -EINVAL;

-	return mdiobus_write(bus, ds->pd->sw_addr + addr, reg, val);
+	mutex_lock_nested(&bus->mdio_lock, SINGLE_DEPTH_NESTING);
+	ret = bus->write(bus, ds->pd->sw_addr, reg, val);
+	mutex_unlock(&bus->mdio_lock);
+
+	return ret;
 }

 #define REG_WRITE(addr, reg, val)				\
-- 
1.9.1

Re: [PATCH] net: dsa: mv88e6060: Fix false positive lockdep splat

[Andrew Lunn]

On Wed, Oct 21, 2015 at 05:37:45PM +0200, Neil Armstrong wrote:
> Like the change made for mv88e6xxx, use mutex_lock_nested() to avoid
> lockdep to give false positives because of nested MDIO busses.

Hi Neil

We now have three instances of this, since mdio-mux.c has the same
code. Maybe now would be a good time to refactor this code into
mdiobus_read_nested() and mdiobus_write_nested() in mdio_bus.c?  At
the same time, add BUG_ON(in_interrupt()) similar to the non-nested
versions?

	Andrew

Re: [PATCH] net: dsa: mv88e6060: Fix false positive lockdep splat

[Neil Armstrong]

Hi Andrew,

On 10/21/2015 06:14 PM, Andrew Lunn wrote:
> On Wed, Oct 21, 2015 at 05:37:45PM +0200, Neil Armstrong wrote:
>> Like the change made for mv88e6xxx, use mutex_lock_nested() to avoid
>> lockdep to give false positives because of nested MDIO busses.
> 
> Hi Neil
> 
> We now have three instances of this, since mdio-mux.c has the same
> code. Maybe now would be a good time to refactor this code into
> mdiobus_read_nested() and mdiobus_write_nested() in mdio_bus.c?  At
> the same time, add BUG_ON(in_interrupt()) similar to the non-nested
> versions?
> 
> 	Andrew
> 
Indeed, you are right, I will post a serie with this refactoring.

Neil

Re: [PATCH] net: dsa: mv88e6060: Fix false positive lockdep splat

[Neil Armstrong]

On 10/21/2015 06:14 PM, Andrew Lunn wrote:
> On Wed, Oct 21, 2015 at 05:37:45PM +0200, Neil Armstrong wrote:
>> Like the change made for mv88e6xxx, use mutex_lock_nested() to avoid
>> lockdep to give false positives because of nested MDIO busses.
> 
> Hi Neil
> 
> We now have three instances of this, since mdio-mux.c has the same
> code. Maybe now would be a good time to refactor this code into
> mdiobus_read_nested() and mdiobus_write_nested() in mdio_bus.c?  At
> the same time, add BUG_ON(in_interrupt()) similar to the non-nested
> versions?
> 
> 	Andrew
> 
Well, mdio-mux also calls switch_fn inside the mdio_lock, clean refactoring
would introduce a separate lock and call the nested variants.
Is that ok ? Can someone test mdio-mux is I make the change ?


Neil

Re: [PATCH] net: dsa: mv88e6060: Fix false positive lockdep splat

[Neil Armstrong]

On 10/21/2015 06:14 PM, Andrew Lunn wrote:
> On Wed, Oct 21, 2015 at 05:37:45PM +0200, Neil Armstrong wrote:
>> Like the change made for mv88e6xxx, use mutex_lock_nested() to avoid
>> lockdep to give false positives because of nested MDIO busses.
> 
> Hi Neil
> 
> We now have three instances of this, since mdio-mux.c has the same
> code. Maybe now would be a good time to refactor this code into
> mdiobus_read_nested() and mdiobus_write_nested() in mdio_bus.c?  At
> the same time, add BUG_ON(in_interrupt()) similar to the non-nested
> versions?
> 
> 	Andrew
> 
Well, mdio-mux also calls switch_fn inside the mdio_lock, clean refactoring
would introduce a separate lock and call the nested variants.
Is that ok ? Can someone test mdio-mux if I make the change ?


Neil

Re: [PATCH] net: dsa: mv88e6060: Fix false positive lockdep splat

[Andrew Lunn]

> Well, mdio-mux also calls switch_fn inside the mdio_lock, clean refactoring
> would introduce a separate lock and call the nested variants.
> Is that ok ? Can someone test mdio-mux if I make the change ?

Hi Neil

I would not touch mdio-mux. As you said, it does more than lock, read,
unlock. It is not something sufficiently generic to place into shared
code.

	Andrew