Skip to contentSkip to conversation

net: phy: Unbind/bind fixes

PATCH In review Florian Fainelli 4 participants 9 messages Feb 8, 2017

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

v2 cover net: phy: Unbind/bind fixes

Hi all,

This patch series addresses the inability to safely unbind and bind PHY drivers by making the appropriate checks throught PHYLIB where we may be directly responding to user-space queries, as well as from within the kernel state machine.

The second patch makes the unbind -> bind working by taking care of the PHY state machine state.

Changes in v2:

- fixed net: phy: Fix lack of reference count on PHY driver against the Generic PHY driver which is special

Florian Fainelli (3):
net: phy: Fix PHY module checks
net: phy: Check phydev->drv
net: phy: Fix PHY driver bind and unbind events

 drivers/net/phy/phy.c        | 26 ++++++++++++++++++++++----
 drivers/net/phy/phy_device.c | 35 ++++++++++++++++++++++++++++++-----
 include/linux/phy.h          |  3 +++
 3 files changed, 55 insertions(+), 9 deletions(-)
Signature · 1 line
2.9.3
#1 lore raw

PATCH v2 1/3 net: phy: Fix PHY module checks

The Generic PHY drivers gets assigned after we checked that the current PHY driver is NULL, so we need to check a few things before we can safely derference d->driver. Update phy_attach_direct() and phy_detach() accordingly to be resilient to these cases.

Even though the Generic PHY driver defaults to phy_probe() which can hardly fail at the moment, let's fix the label so we don't call phy_detach() on a network device we have not attached yet.

Fixes cafe8df8b9bc ("net: phy: Fix lack of reference count on PHY driver")
---
 drivers/net/phy/phy_device.c | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)
drivers/net/phy/phy_device.c +5 -3
index 0d8f4d3847f6..bde240bf8d7b 100644
@@ -920,7 +920,7 @@ int phy_attach_direct(struct net_device *dev, struct phy_device *phydev,
 		return -EIO;
 	}
 
-	if (!try_module_get(d->driver->owner)) {
+	if (d->driver && !try_module_get(d->driver->owner)) {
 		dev_err(&dev->dev, "failed to get the device driver module\n");
 		return -EIO;
 	}
@@ -943,7 +943,7 @@ int phy_attach_direct(struct net_device *dev, struct phy_device *phydev,
 			err = device_bind_driver(d);
 
 		if (err)
-			goto error;
+			goto error_put_device;
 	}
 
 	if (phydev->attached_dev) {
@@ -981,6 +981,7 @@ int phy_attach_direct(struct net_device *dev, struct phy_device *phydev,
 
 error:
 	phy_detach(phydev);
+error_put_device:
 	put_device(d);
 	module_put(d->driver->owner);
 	if (ndev_owner != bus->owner)
@@ -1065,7 +1066,8 @@ void phy_detach(struct phy_device *phydev)
 	bus = phydev->mdio.bus;
 
 	put_device(&phydev->mdio.dev);
-	module_put(phydev->mdio.dev.driver->owner);
+	if (phydev->mdio.dev.driver)
+		module_put(phydev->mdio.dev.driver->owner);
 	if (ndev_owner != bus->owner)
 		module_put(bus->owner);
 }
Signature · 1 line
2.9.3
#2 lore raw

RE: [PATCH net v2 1/3] net: phy: Fix PHY module checks

Quoted 30 lines

-----Original Message-----
From: Florian Fainelli [mailto:[email protected]]
Sent: Wednesday, February 08, 2017 3:38 PM
To: [email protected]
Cc: [email protected]; [email protected]; [email protected]; maowenan; Florian Fainelli
Subject: [PATCH net v2 1/3] net: phy: Fix PHY module checks

The Generic PHY drivers gets assigned after we checked that the current PHY driver is NULL, so we need to check a few things before we can safely derference d->driver. Update phy_attach_direct() and phy_detach() accordingly to be resilient to these cases.

Even though the Generic PHY driver defaults to phy_probe() which can hardly fail at the moment, let's fix the label so we don't call phy_detach() on a network device we have not attached yet.

Fixes cafe8df8b9bc ("net: phy: Fix lack of reference count on PHY driver")
---
 drivers/net/phy/phy_device.c | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)
drivers/net/phy/phy_device.c index renamed from drivers/net/phy/phy_device.c +0 -0

0d8f4d3847f6..bde240bf8d7b 100644

drivers/net/phy/phy_device.c int phy_attach_direct(struct net_device *dev, struct · line 926

Quoted 8 lines
drivers/net/phy/phy_device.c +1 -1
@@ -920,7 +920,7 @@ int phy_attach_direct(struct net_device *dev, struct
 phy_device *phydev,
 		return -EIO;
 	}
 
-	if (!try_module_get(d->driver->owner)) {
+	if (d->driver && !try_module_get(d->driver->owner)) {
 		dev_err(&dev->dev, "failed to get the device driver module\n");
 		return -EIO;

Hi Florian,
Here "return -EIO" will miss decreasing bus->owner reference count.

Mao Wenan

Quoted 32 lines
drivers/net/phy/phy_device.c +1 -1
 	}
@@ -943,7 +943,7 @@ int phy_attach_direct(struct net_device *dev, struct
 phy_device *phydev,
 			err = device_bind_driver(d);
 
 		if (err)
-			goto error;
+			goto error_put_device;
 	}
 
drivers/net/phy/phy_device.c +1 -0
 	if (phydev->attached_dev) {
@@ -981,6 +981,7 @@ int phy_attach_direct(struct net_device *dev, struct
 phy_device *phydev,
 
 error:
 	phy_detach(phydev);
+error_put_device:
 	put_device(d);
 	module_put(d->driver->owner);
drivers/net/phy/phy_device.c +2 -2
 	if (ndev_owner != bus->owner)
@@ -1065,7 +1066,8 @@ void phy_detach(struct phy_device *phydev)
 	bus = phydev->mdio.bus;
 
 	put_device(&phydev->mdio.dev);
-	module_put(phydev->mdio.dev.driver->owner);
+	if (phydev->mdio.dev.driver)
+		module_put(phydev->mdio.dev.driver->owner);
 	if (ndev_owner != bus->owner)
 		module_put(bus->owner);
 }
--

2.9.3

#3 lore raw

Re: [PATCH net v2 1/3] net: phy: Fix PHY module checks

Quoted 34 lines · maowenan
Quoted 34 lines

-----Original Message-----
From: Florian Fainelli [mailto:[email protected]]
Sent: Wednesday, February 08, 2017 3:38 PM
To: [email protected]
Cc: [email protected]; [email protected]; [email protected]; maowenan; Florian Fainelli
Subject: [PATCH net v2 1/3] net: phy: Fix PHY module checks

The Generic PHY drivers gets assigned after we checked that the current PHY driver is NULL, so we need to check a few things before we can safely derference d->driver. Update phy_attach_direct() and phy_detach() accordingly to be resilient to these cases.

Even though the Generic PHY driver defaults to phy_probe() which can hardly fail at the moment, let's fix the label so we don't call phy_detach() on a network device we have not attached yet.

Fixes cafe8df8b9bc ("net: phy: Fix lack of reference count on PHY driver")
---
 drivers/net/phy/phy_device.c | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)
drivers/net/phy/phy_device.c index renamed from drivers/net/phy/phy_device.c +0 -0

0d8f4d3847f6..bde240bf8d7b 100644

drivers/net/phy/phy_device.c +0 -0
@@ -920,7 +920,7 @@ int phy_attach_direct(struct net_device *dev, struct
 phy_device *phydev,
 		return -EIO;
 	}
 
Quoted 7 lines

drivers/net/phy/phy_device.c int phy_attach_direct(struct net_device *dev, struct phy_device *phydev, · line 925

Quoted 4 lines
drivers/net/phy/phy_device.c +1 -1
-	if (!try_module_get(d->driver->owner)) {
+	if (d->driver && !try_module_get(d->driver->owner)) {
 		dev_err(&dev->dev, "failed to get the device driver module\n");
 		return -EIO;

Hi Florian,
Here "return -EIO" will miss decreasing bus->owner reference count.

Good catch thanks!

Signature · 1 line
Florian
Tested-by on v2 1/3
#4 lore raw

Re: [PATCH net v2 1/3] net: phy: Fix PHY module checks

In reply to Florian Fainelli
Quoted 11 lines · Florian Fainelli

The Generic PHY drivers gets assigned after we checked that the current PHY driver is NULL, so we need to check a few things before we can safely derference d->driver. Update phy_attach_direct() and phy_detach() accordingly to be resilient to these cases.

Even though the Generic PHY driver defaults to phy_probe() which can hardly fail at the moment, let's fix the label so we don't call phy_detach() on a network device we have not attached yet.

Fixes cafe8df8b9bc ("net: phy: Fix lack of reference count on PHY driver")

This fixes the kernel crash:

maintainer
#5 lore raw

Re: [PATCH net v2 1/3] net: phy: Fix PHY module checks

In reply to Florian Fainelli
Quoted 21 lines
drivers/net/phy/phy_device.c +2 -2
@@ -920,7 +920,7 @@ int phy_attach_direct(struct net_device *dev, struct phy_device *phydev,
 		return -EIO;
 	}
 
-	if (!try_module_get(d->driver->owner)) {
+	if (d->driver && !try_module_get(d->driver->owner)) {
 		dev_err(&dev->dev, "failed to get the device driver module\n");
 		return -EIO;
 	}
@@ -943,7 +943,7 @@ int phy_attach_direct(struct net_device *dev, struct phy_device *phydev,
 			err = device_bind_driver(d);
 
 		if (err)
-			goto error;
+			goto error_put_device;
 	}
 
 	if (phydev->attached_dev) {

drivers/net/phy/phy_device.c int phy_attach_direct(struct net_device *dev, struct phy_device *phydev, · line 986

Quoted 6 lines
drivers/net/phy/phy_device.c +1 -0
@@ -981,6 +981,7 @@ int phy_attach_direct(struct net_device *dev, struct phy_device *phydev,
 
 error:
 	phy_detach(phydev);
+error_put_device:
 	put_device(d);
 	module_put(d->driver->owner);

Can we get into problems here? Maybe we did not do a get?

drivers/net/phy/phy_device.c void phy_detach(struct phy_device *phydev) · line 1070

Quoted 8 lines
drivers/net/phy/phy_device.c +2 -1
 	if (ndev_owner != bus->owner)
@@ -1065,7 +1066,8 @@ void phy_detach(struct phy_device *phydev)
 	bus = phydev->mdio.bus;
 
 	put_device(&phydev->mdio.dev);
-	module_put(phydev->mdio.dev.driver->owner);
+	if (phydev->mdio.dev.driver)
+		module_put(phydev->mdio.dev.driver->owner);

Humm. By this point, havn't we assigned d->driver to genphy_driver? So this decrements something we never incremented?

To me, it seems better to move the try_module_get() after assigning genphy_driver if needed. We then don't have any special conditions.

Andrew

#6 lore raw

Re: [PATCH net v2 1/3] net: phy: Fix PHY module checks

Quoted 22 lines · Andrew Lunn
Quoted 22 lines
drivers/net/phy/phy_device.c +2 -2
@@ -920,7 +920,7 @@ int phy_attach_direct(struct net_device *dev, struct phy_device *phydev,
 		return -EIO;
 	}
 
-	if (!try_module_get(d->driver->owner)) {
+	if (d->driver && !try_module_get(d->driver->owner)) {
 		dev_err(&dev->dev, "failed to get the device driver module\n");
 		return -EIO;
 	}
@@ -943,7 +943,7 @@ int phy_attach_direct(struct net_device *dev, struct phy_device *phydev,
 			err = device_bind_driver(d);
 
 		if (err)
-			goto error;
+			goto error_put_device;
 	}
 
 	if (phydev->attached_dev) {
@@ -981,6 +981,7 @@ int phy_attach_direct(struct net_device *dev, struct phy_device *phydev,
 
Quoted 7 lines

drivers/net/phy/phy_device.c int phy_attach_direct(struct net_device *dev, struct phy_device *phydev, · line 986

Quoted 5 lines
drivers/net/phy/phy_device.c +1 -0
 error:
 	phy_detach(phydev);
+error_put_device:
 	put_device(d);
 	module_put(d->driver->owner);

Can we get into problems here? Maybe we did not do a get?

Humm yes, that seems possible too.

Quoted 6 lines
Quoted 6 lines
drivers/net/phy/phy_device.c +0 -0
 	if (ndev_owner != bus->owner)
@@ -1065,7 +1066,8 @@ void phy_detach(struct phy_device *phydev)
 	bus = phydev->mdio.bus;
 
Quoted 7 lines

drivers/net/phy/phy_device.c void phy_detach(struct phy_device *phydev) · line 1070

Quoted 4 lines
drivers/net/phy/phy_device.c +2 -1
 	put_device(&phydev->mdio.dev);
-	module_put(phydev->mdio.dev.driver->owner);
+	if (phydev->mdio.dev.driver)
+		module_put(phydev->mdio.dev.driver->owner);

Humm. By this point, havn't we assigned d->driver to genphy_driver? So this decrements something we never incremented?

Not really, the generic PHY drivers are built into libphy, so all PHY drivers de facto going to depend on libphy.ko being loaded first, that's why we can take shortcuts and not increment the refcount on the genphy/libphy module here.

Quoted 3 lines

To me, it seems better to move the try_module_get() after assigning genphy_driver if needed. We then don't have any special conditions.

Sounds fair, let me fix that (and the problem reported by Mao about error handling).

Thanks!

Signature · 1 line
Florian
#7 lore raw

PATCH v2 2/3 net: phy: Check phydev->drv

In reply to Florian Fainelli

In preparation for supporting driver bind/unbind properly, sprinkle checks on phydev->drv where we may call into PHYLIB from user-space or other parts of the kernel.

---
 drivers/net/phy/phy.c | 26 ++++++++++++++++++++++----
 include/linux/phy.h   |  3 +++
 2 files changed, 25 insertions(+), 4 deletions(-)
drivers/net/phy/phy.c +22 -4
index 7cc1b7dcfe05..d6f7838455dd 100644
@@ -580,7 +580,7 @@ int phy_mii_ioctl(struct phy_device *phydev, struct ifreq *ifr, int cmd)
 		return 0;
 
 	case SIOCSHWTSTAMP:
-		if (phydev->drv->hwtstamp)
+		if (phydev->drv && phydev->drv->hwtstamp)
 			return phydev->drv->hwtstamp(phydev, ifr);
 		/* fall through */
 
@@ -603,6 +603,9 @@ int phy_start_aneg(struct phy_device *phydev)
 {
 	int err;
 
+	if (!phydev->drv)
+		return -EIO;
+
 	mutex_lock(&phydev->lock);
 
 	if (AUTONEG_DISABLE == phydev->autoneg)
@@ -975,7 +978,7 @@ void phy_state_machine(struct work_struct *work)
 
 	old_state = phydev->state;
 
-	if (phydev->drv->link_change_notify)
+	if (phydev->drv && phydev->drv->link_change_notify)
 		phydev->drv->link_change_notify(phydev);
 
 	switch (phydev->state) {
@@ -1286,6 +1289,9 @@ EXPORT_SYMBOL(phy_write_mmd_indirect);
  */
 int phy_init_eee(struct phy_device *phydev, bool clk_stop_enable)
 {
+	if (!phydev->drv)
+		return -EIO;
+
 	/* According to 802.3az,the EEE is supported only in full duplex-mode.
 	 * Also EEE feature is active when core is operating with MII, GMII
 	 * or RGMII (all kinds). Internal PHYs are also allowed to proceed and
@@ -1363,6 +1369,9 @@ EXPORT_SYMBOL(phy_init_eee);
  */
 int phy_get_eee_err(struct phy_device *phydev)
 {
+	if (!phydev->drv)
+		return -EIO;
+
 	return phy_read_mmd_indirect(phydev, MDIO_PCS_EEE_WK_ERR, MDIO_MMD_PCS);
 }
 EXPORT_SYMBOL(phy_get_eee_err);
@@ -1379,6 +1388,9 @@ int phy_ethtool_get_eee(struct phy_device *phydev, struct ethtool_eee *data)
 {
 	int val;
 
+	if (!phydev->drv)
+		return -EIO;
+
 	/* Get Supported EEE */
 	val = phy_read_mmd_indirect(phydev, MDIO_PCS_EEE_ABLE, MDIO_MMD_PCS);
 	if (val < 0)
@@ -1412,6 +1424,9 @@ int phy_ethtool_set_eee(struct phy_device *phydev, struct ethtool_eee *data)
 {
 	int val = ethtool_adv_to_mmd_eee_adv_t(data->advertised);
 
+	if (!phydev->drv)
+		return -EIO;
+
 	/* Mask prohibited EEE modes */
 	val &= ~phydev->eee_broken_modes;
 
@@ -1423,7 +1438,7 @@ EXPORT_SYMBOL(phy_ethtool_set_eee);
 
 int phy_ethtool_set_wol(struct phy_device *phydev, struct ethtool_wolinfo *wol)
 {
-	if (phydev->drv->set_wol)
+	if (phydev->drv && phydev->drv->set_wol)
 		return phydev->drv->set_wol(phydev, wol);
 
 	return -EOPNOTSUPP;
@@ -1432,7 +1447,7 @@ EXPORT_SYMBOL(phy_ethtool_set_wol);
 
 void phy_ethtool_get_wol(struct phy_device *phydev, struct ethtool_wolinfo *wol)
 {
-	if (phydev->drv->get_wol)
+	if (phydev->drv && phydev->drv->get_wol)
 		phydev->drv->get_wol(phydev, wol);
 }
 EXPORT_SYMBOL(phy_ethtool_get_wol);
@@ -1468,6 +1483,9 @@ int phy_ethtool_nway_reset(struct net_device *ndev)
 	if (!phydev)
 		return -ENODEV;
 
+	if (!phydev->drv)
+		return -EIO;
+
 	return genphy_restart_aneg(phydev);
 }
 EXPORT_SYMBOL(phy_ethtool_nway_reset);
include/linux/phy.h +3 -0
index 7fc1105605bf..231e07bb0d76 100644
@@ -802,6 +802,9 @@ int phy_stop_interrupts(struct phy_device *phydev);
 
 static inline int phy_read_status(struct phy_device *phydev)
 {
+	if (!phydev->drv)
+		return -EIO;
+
 	return phydev->drv->read_status(phydev);
 }
 
Signature · 1 line
2.9.3
#8 lore raw

PATCH v2 3/3 net: phy: Fix PHY driver bind and unbind events

In reply to Florian Fainelli

The PHY library does not deal very well with bind and unbind events. The first thing we would see is that we were not properly canceling the PHY state machine workqueue, so we would be crashing while dereferencing phydev->drv since there is no driver attached anymore.

Once we fix that, there are several things that did not quite work as expected:

- if the PHY state machine was running, we were not stopping it properly, and the state machine state would not be marked as such
- when we rebind the driver, nothing would happen, since we would not know which state we were before the unbind

This patch takes the following approach:

- if the PHY was attached, and the state machine was running we would stop it, remember where we left, and schedule the state machine for restart upong driver bind
- if the PHY was attached, but HALTED, we would let it in that state, and do not alter the state upon driver bind
- in all other cases (detached) we would keep the PHY in DOWN state waiting for a network driver to show up, and set PHY_READY on driver bind

---
 drivers/net/phy/phy_device.c | 27 +++++++++++++++++++++++++--
 1 file changed, 25 insertions(+), 2 deletions(-)
drivers/net/phy/phy_device.c +25 -2
index bde240bf8d7b..5314e764a387 100644
@@ -1711,6 +1711,7 @@ static int phy_probe(struct device *dev)
 	struct phy_device *phydev = to_phy_device(dev);
 	struct device_driver *drv = phydev->mdio.dev.driver;
 	struct phy_driver *phydrv = to_phy_driver(drv);
+	bool should_start = false;
 	int err = 0;
 
 	phydev->drv = phydrv;
@@ -1760,24 +1761,46 @@ static int phy_probe(struct device *dev)
 	}
 
 	/* Set the state to READY by default */
-	phydev->state = PHY_READY;
+	if (phydev->state > PHY_UP && phydev->state != PHY_HALTED)
+		should_start = true;
+	else
+		phydev->state = PHY_READY;
 
 	if (phydev->drv->probe)
 		err = phydev->drv->probe(phydev);
 
 	mutex_unlock(&phydev->lock);
 
+	if (should_start)
+		phy_start(phydev);
+
 	return err;
 }
 
 static int phy_remove(struct device *dev)
 {
 	struct phy_device *phydev = to_phy_device(dev);
+	bool should_stop = false;
+	enum phy_state state;
+
+	cancel_delayed_work_sync(&phydev->state_queue);
 
 	mutex_lock(&phydev->lock);
-	phydev->state = PHY_DOWN;
+	state = phydev->state;
+	if (state > PHY_UP && state != PHY_HALTED)
+		should_stop = true;
+	else
+		phydev->state = PHY_DOWN;
 	mutex_unlock(&phydev->lock);
 
+	/* phy_stop() sets the state to HALTED, undo that for the ->probe() function
+	 * to have a chance to resume where we left
+	 */
+	if (should_stop) {
+		phy_stop(phydev);
+		phydev->state = state;
+	}
+
 	if (phydev->drv->remove)
 		phydev->drv->remove(phydev);
 	phydev->drv = NULL;
Signature · 1 line
2.9.3