net: phy: clear phydev->devlink when the link is deleted
Subsystems (4)
PATCH v2 1/1 net: phy: clear phydev->devlink when the link is deleted
There is a potential crash issue when disabling and re-enabling the network port. When disabling the network port, phy_detach() calls device_link_del() to remove the device link, but it does not clear phydev->devlink, so phydev->devlink is not a NULL pointer. Then the network port is re-enabled, but if phy_attach_direct() fails before calling device_link_add(), the code jumps to the "error" label and calls phy_detach(). Since phydev->devlink retains the old value from the previous attach/detach cycle, device_link_del() uses the old value, which accesses a NULL pointer and causes a crash. The simplified crash log is as follows.
[ 24.702421] Call trace: [ 24.704856] device_link_put_kref+0x20/0x120 [ 24.709124] device_link_del+0x30/0x48 [ 24.712864] phy_detach+0x24/0x168 [ 24.716261] phy_attach_direct+0x168/0x3a4 [ 24.720352] phylink_fwnode_phy_connect+0xc8/0x14c [ 24.725140] phylink_of_phy_connect+0x1c/0x34
Therefore, phydev->devlink needs to be cleared when the device link is deleted.
---
v2: Improve the commit message.
--- drivers/net/phy/phy_device.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)
drivers/net/phy/phy_device.c +3 -1
@@ -1727,8 +1727,10 @@ void phy_detach(struct phy_device *phydev)struct module *ndev_owner =NULL ;struct mii_bus *bus; -if (phydev->devlink) +if (phydev->devlink) {device_link_del (phydev->devlink); + phydev->devlink =NULL ; + }if (phydev->sysfs_links) {if (dev)
Signature · 1 line
2.34.1
Quoted 19 lines · Wei Fang
There is a potential crash issue when disabling and re-enabling the network port. When disabling the network port,
phy_detach()callsdevice_link_del()to remove the device link, but it does not clearphydev->devlink, sophydev->devlinkis not a NULL pointer. Then the network port is re-enabled, but ifphy_attach_direct()fails before callingdevice_link_add(), the code jumps to the "error" label and callsphy_detach(). Sincephydev->devlinkretains the old value from the previous attach/detach cycle,device_link_del()uses the old value, which accesses a NULL pointer and causes a crash. The simplified crash log is as follows.[ 24.702421] Call trace: [ 24.704856] device_link_put_kref+0x20/0x120 [ 24.709124] device_link_del+0x30/0x48 [ 24.712864] phy_detach+0x24/0x168 [ 24.716261] phy_attach_direct+0x168/0x3a4 [ 24.720352] phylink_fwnode_phy_connect+0xc8/0x14c [ 24.725140] phylink_of_phy_connect+0x1c/0x34
Quoted 5 lines
Therefore,
phydev->devlinkneeds to be cleared when the device link is deleted.Fixes bc66fa87d4fd ("net: phy: Add link between phy dev and mac dev")Signed-off-by Wei Fang <[email protected]>
Andrew
Quoted 19 lines · Wei Fang
There is a potential crash issue when disabling and re-enabling the network port. When disabling the network port,
phy_detach()callsdevice_link_del()to remove the device link, but it does not clearphydev->devlink, sophydev->devlinkis not a NULL pointer. Then the network port is re-enabled, but ifphy_attach_direct()fails before callingdevice_link_add(), the code jumps to the "error" label and callsphy_detach(). Sincephydev->devlinkretains the old value from the previous attach/detach cycle,device_link_del()uses the old value, which accesses a NULL pointer and causes a crash. The simplified crash log is as follows.[ 24.702421] Call trace: [ 24.704856] device_link_put_kref+0x20/0x120 [ 24.709124] device_link_del+0x30/0x48 [ 24.712864] phy_detach+0x24/0x168 [ 24.716261] phy_attach_direct+0x168/0x3a4 [ 24.720352] phylink_fwnode_phy_connect+0xc8/0x14c [ 24.725140] phylink_of_phy_connect+0x1c/0x34
Quoted 5 lines
Therefore,
phydev->devlinkneeds to be cleared when the device link is deleted.Fixes bc66fa87d4fd ("net: phy: Add link between phy dev and mac dev")Signed-off-by Wei Fang <[email protected]>
Signature · 1 line
Florian
Quoted 22 lines · Florian Fainelli
Quoted 19 lines · Wei Fang
There is a potential crash issue when disabling and re-enabling the network port. When disabling the network port,
phy_detach()callsdevice_link_del()to remove the device link, but it does not clearphydev->devlink, sophydev->devlinkis not a NULL pointer. Then the network port is re-enabled, but ifphy_attach_direct()fails before callingdevice_link_add(), the code jumps to the "error" label and callsphy_detach(). Sincephydev->devlinkretains the old value from the previous attach/detach cycle,device_link_del()uses the old value, which accesses a NULL pointer and causes a crash. The simplified crash log is as follows.[ 24.702421] Call trace:
[ 24.704856]device_link_put_kref+0x20/0x120[ 24.709124]device_link_del+0x30/0x48
[ 24.712864]phy_detach+0x24/0x168
[ 24.716261]phy_attach_direct+0x168/0x3a4[ 24.720352]phylink_fwnode_phy_connect+0xc8/0x14c[ 24.725140]phylink_of_phy_connect+0x1c/0x34
Quoted 6 lines
Quoted 5 lines
Therefore,
phydev->devlinkneeds to be cleared when the device link is deleted.Fixes bc66fa87d4fd ("net: phy: Add link between phy dev and mac dev")Signed-off-by Wei Fang <[email protected]>
@Wei
What happens in case of shared mdio ?
1. Device 23040000 has the mdio node of both the ethernet phy and device 23000000 references the phy-handle present in the Device 23040000
2. When rmmod of the driver happens
3. the parent devlink is already deleted.
4. This cause the child mdio to access an entry causing a corruption.
5. Thought this fix would help but i see that its not helping the case.
Wondering if this is a legacy issue with shared mdio framework.
43369.232799: qcom-ethqos 23040000.ethernet eth1: stmmac_dvr_remove: removing driver 43369.233782: stmmac_pcs: Link Down 43369.258337: qcom-ethqos 23040000.ethernet eth1: FPE workqueue stop 43369.258522: br1: port 1(eth1) entered disabled state 43369.758779: qcom-ethqos 23040000.ethernet eth1: Timeout accessing MAC_VLAN_Tag_Filter 43369.758789: qcom-ethqos 23040000.ethernet eth1: failed to kill vid 0081/0 43369.759270: qcom-ethqos 23040000.ethernet eth1 (unregistering): left allmulticast mode 43369.759275: qcom-ethqos 23040000.ethernet eth1 (unregistering): left promiscuous mode 43369.759301: br1: port 1(eth1) entered disabled state 43370.259618: qcom-ethqos 23040000.ethernet eth1 (unregistering): Timeout accessing MAC_VLAN_Tag_Filter 43370.309863: qcom-ethqos 23000000.ethernet eth0: stmmac_dvr_remove: removing driver 43370.310019: list_del corruption, ffffff80c6ec9408->prev is LIST_POISON2 (dead000000000122) 43370.310034: ------------[ cut here ]------------ 43370.310035: kernel BUG at lib/list_debug.c:59! 43370.310119: CPU: 4 PID: 3067767 Comm: rmmod Tainted: G W OE 6.6.65-rt47-debug #1 43370.310122: Hardware name: Qualcomm Technologies, Inc. SA8775P Ride (DT) 43370.310165: Call trace: 43370.310166: __list_del_entry_valid_or_report+0xa8/0xe0 43370.310168: __device_link_del+0x40/0xf0 43370.310172: device_link_put_kref+0xb4/0xc8 43370.310174: device_link_del+0x38/0x58 43370.310176: phy_detach+0x2c/0x170 43370.310181: phy_disconnect+0x4c/0x70 43370.310184: phylink_disconnect_phy+0x6c/0xc0 [phylink] 43370.310194: stmmac_release+0x60/0x358 [stmmac] 43370.310210: __dev_close_many+0xb4/0x160 43370.310213: dev_close_many+0xbc/0x1a0 43370.310215: unregister_netdevice_many_notify+0x178/0x870 43370.310218: unregister_netdevice_queue+0xf8/0x140 43370.310221: unregister_netdev+0x2c/0x48 43370.310223: stmmac_dvr_remove+0xd0/0x1b0 [stmmac] 43370.310233: devm_stmmac_pltfr_remove+0x2c/0x58 [stmmac_platform]
Quoted 1 line
Reviewed-by Florian Fainelli <[email protected]>
Russell King (Oracle)
reviewerQuoted 34 lines · Abhishek Chauhan (ABC)
Quoted 22 lines · Florian Fainelli
Quoted 19 lines · Wei Fang
There is a potential crash issue when disabling and re-enabling the network port. When disabling the network port,
phy_detach()callsdevice_link_del()to remove the device link, but it does not clearphydev->devlink, sophydev->devlinkis not a NULL pointer. Then the network port is re-enabled, but ifphy_attach_direct()fails before callingdevice_link_add(), the code jumps to the "error" label and callsphy_detach(). Sincephydev->devlinkretains the old value from the previous attach/detach cycle,device_link_del()uses the old value, which accesses a NULL pointer and causes a crash. The simplified crash log is as follows.[ 24.702421] Call trace:
[ 24.704856]device_link_put_kref+0x20/0x120[ 24.709124]device_link_del+0x30/0x48
[ 24.712864]phy_detach+0x24/0x168
[ 24.716261]phy_attach_direct+0x168/0x3a4[ 24.720352]phylink_fwnode_phy_connect+0xc8/0x14c[ 24.725140]phylink_of_phy_connect+0x1c/0x34Quoted 6 lines
Quoted 5 lines
Therefore,
phydev->devlinkneeds to be cleared when the device link is deleted.Fixes bc66fa87d4fd ("net: phy: Add link between phy dev and mac dev")Signed-off-by Wei Fang <[email protected]>@Wei
What happens in case of shared mdio ?
Quoted 7 lines
1. Device 23040000 has the mdio node of both the ethernet phy and device 23000000 references the phy-handle present in the Device 23040000
2. When rmmod of the driver happens
3. the parent devlink is already deleted.
4. This cause the child mdio to access an entry causing a corruption.
5. Thought this fix would help but i see that its not helping the case.Wondering if this is a legacy issue with shared mdio framework.
The device link does nothing for this as it has DL_FLAG_STATELESS set, which only affects suspend/resume/shutdown ordering, and with DL_FLAG_PM_RUNTIME also set, runtime PM.
The device probe/removal ordering is unaffected. Maybe that's a problem, but it needs careful consideration to change.
Signature · 2 lines
RMK's Patch system: https://www.armlinux.org.uk/developer/patches/ FTTP is here! 80Mbps down 10Mbps up. Decent connectivity at last!
Quoted 29 lines
Quoted 20 lines · Wei Fang
Quoted 19 lines
There is a potential crash issue when disabling and re-enabling the network port. When disabling the network port,
phy_detach()callsdevice_link_del()to remove the device link, but it does not clearphydev->devlink, sophydev->devlinkis not a NULL pointer. Then the network port is re-enabled, but ifphy_attach_direct()fails before callingdevice_link_add(), the code jumps to the "error" label and callsphy_detach(). Sincephydev->devlinkretains the old value from the previous attach/detach cycle,device_link_del()uses the old value, which accesses a NULL pointer and causes a crash. The simplified crash log is as follows.[ 24.702421] Call trace:
[ 24.704856]device_link_put_kref+0x20/0x120[ 24.709124]device_link_del+0x30/0x48
[ 24.712864]phy_detach+0x24/0x168
[ 24.716261]phy_attach_direct+0x168/0x3a4[ 24.720352]phylink_fwnode_phy_connect+0xc8/0x14c[ 24.725140]phylink_of_phy_connect+0x1c/0x34Quoted 6 lines
Quoted 5 lines
Therefore,
phydev->devlinkneeds to be cleared when the device link is deleted.Fixes bc66fa87d4fd ("net: phy: Add link between phy dev and mac dev")Signed-off-by Wei Fang <[email protected]>@Wei
What happens in case of shared mdio ?
Quoted 7 lines
1. Device 23040000 has the mdio node of both the ethernet phy and device 23000000 references the phy-handle present in the Device 23040000
2. When rmmod of the driver happens
3. the parent devlink is already deleted.
4. This cause the child mdio to access an entry causing a corruption.
5. Thought this fix would help but i see that its not helping the case.
My patch is only to fix the potential crash issue when re-enabling the network interface. phy_detach() is not called when the MDIO controller driver is removed. So phydev->devlink is not cleared, but actually the device link has been removed by phy_device_remove() --> device_del(). Therefore, it will cause the crash when the MAC controller driver is removed.
Quoted 2 lines
Wondering if this is a legacy issue with shared mdio framework.
I think this issue is also introduced by the commit bc66fa87d4fd ("net: phy: Add link between phy dev and mac dev"). I suggested to change the DL_FLAG_STATELESS flag to DL_FLAG_AUTOREMOVE_SUPPLIER to solve this issue, so that the consumer (MAC controller) driver will be automatically removed when the link is removed. The changes are as follows.
drivers/net/phy/phy_device.c +11 -8
@@ -1515,6 +1515,7 @@ int phy_attach_direct(struct net_device *dev, struct phy_device *phydev,struct mii_bus *bus = phydev->mdio.bus;struct device *d = &phydev->mdio.dev;struct module *ndev_owner =NULL ; +struct device_link *devlink;bool using_genphy = false;int err;
@@ -1646,9 +1647,16 @@ int phy_attach_direct(struct net_device *dev, struct phy_device *phydev,* another mac interface, so we should create a device link between * phy dev and mac dev. */ -if (dev && phydev->mdio.bus->parent && dev->dev.parent != phydev->mdio.bus->parent) - phydev->devlink =device_link_add (dev->dev.parent, &phydev->mdio.dev, -DL_FLAG_PM_RUNTIME |DL_FLAG_STATELESS ); +if (dev && phydev->mdio.bus->parent && + dev->dev.parent != phydev->mdio.bus->parent) { + devlink =device_link_add (dev->dev.parent, &phydev->mdio.dev, +DL_FLAG_PM_RUNTIME | +DL_FLAG_AUTOREMOVE_SUPPLIER ); +if (!devlink) { + err = -ENOMEM ; + goto error; + } + }return err;
@@ -1749,11 +1757,6 @@ void phy_detach(struct phy_device *phydev)struct module *ndev_owner =NULL ;struct mii_bus *bus; -if (phydev->devlink) { -device_link_del (phydev->devlink); - phydev->devlink =NULL ; - } -if (phydev->sysfs_links) {if (dev)sysfs_remove_link (&dev->dev.kobj,"phydev" );
include/linux/phy.h +0 -4
@@ -505,8 +505,6 @@ struct macsec_ops;* * @mdio: MDIO bus this PHY is on * @drv: Pointer to the driver for this PHY instance -* @devlink: Create a link between phy dev and mac dev, if the external phy -* used by current mac interface is managed by another mac interface. * @phyindex: Unique id across the phy's parent tree of phys to address the PHY * from userspace, similar to ifindex. A zero index means the PHY * wasn't assigned an id yet.
@@ -610,8 +608,6 @@ struct phy_device {/* And management functions */ const struct phy_driver *drv; -struct device_link *devlink; -u32 phyindex;u32 phy_id;
Quoted 47 lines · Wei Fang
Quoted 29 lines
Quoted 20 lines · Wei Fang
Quoted 19 lines
There is a potential crash issue when disabling and re-enabling the network port. When disabling the network port,
phy_detach()callsdevice_link_del()to remove the device link, but it does not clearphydev->devlink, sophydev->devlinkis not a NULL pointer. Then the network port is re-enabled, but ifphy_attach_direct()fails before callingdevice_link_add(), the code jumps to the "error" label and callsphy_detach(). Sincephydev->devlinkretains the old value from the previous attach/detach cycle,device_link_del()uses the old value, which accesses a NULL pointer and causes a crash. The simplified crash log is as follows.[ 24.702421] Call trace:
[ 24.704856]device_link_put_kref+0x20/0x120[ 24.709124]device_link_del+0x30/0x48
[ 24.712864]phy_detach+0x24/0x168
[ 24.716261]phy_attach_direct+0x168/0x3a4[ 24.720352]phylink_fwnode_phy_connect+0xc8/0x14c[ 24.725140]phylink_of_phy_connect+0x1c/0x34Quoted 6 lines
Quoted 5 lines
Therefore,
phydev->devlinkneeds to be cleared when the device link is deleted.Fixes bc66fa87d4fd ("net: phy: Add link between phy dev and mac dev")Signed-off-by Wei Fang <[email protected]>@Wei
What happens in case of shared mdio ?Quoted 7 lines
1. Device 23040000 has the mdio node of both the ethernet phy and device 23000000 references the phy-handle present in the Device 23040000
2. When rmmod of the driver happens
3. the parent devlink is already deleted.
4. This cause the child mdio to access an entry causing a corruption.
5. Thought this fix would help but i see that its not helping the case.My patch is only to fix the potential crash issue when re-enabling the network interface.
phy_detach()is not called when the MDIO controller driver is removed. Sophydev->devlinkis not cleared, but actually the device link has been removed byphy_device_remove()-->device_del(). Therefore, it will cause the crash when the MAC controller driver is removed.Quoted 2 lines
Wondering if this is a legacy issue with shared mdio framework.
Quoted 7 lines
I think this issue is also introduced by the commit
bc66fa87d4fd("net: phy: Add link between phy dev and mac dev"). I suggested to change theDL_FLAG_STATELESSflag toDL_FLAG_AUTOREMOVE_SUPPLIERto solve this issue, so that the consumer (MAC controller) driver will be automatically removed when the link is removed. The changes are as follows.
thanks a lot , Russell and Wei for your prompt response. I appreciate your help. let me test this change and get back.
Quoted 66 lines
drivers/net/phy/phy_device.c +11
-8@@ -1515,6 +1515,7 @@ int phy_attach_direct(struct net_device *dev, struct phy_device *phydev,struct mii_bus *bus = phydev->mdio.bus;struct device *d = &phydev->mdio.dev;struct module *ndev_owner =NULL ; +struct device_link *devlink;bool using_genphy = false;int err;@@ -1646,9 +1647,16 @@ int phy_attach_direct(struct net_device *dev, struct phy_device *phydev,* another mac interface, so we should create a device link between * phy dev and mac dev. */ -if (dev && phydev->mdio.bus->parent && dev->dev.parent != phydev->mdio.bus->parent) - phydev->devlink =device_link_add (dev->dev.parent, &phydev->mdio.dev, -DL_FLAG_PM_RUNTIME |DL_FLAG_STATELESS ); +if (dev && phydev->mdio.bus->parent && + dev->dev.parent != phydev->mdio.bus->parent) { + devlink =device_link_add (dev->dev.parent, &phydev->mdio.dev, +DL_FLAG_PM_RUNTIME | +DL_FLAG_AUTOREMOVE_SUPPLIER ); +if (!devlink) { + err = -ENOMEM ; + goto error; + } + }return err;@@ -1749,11 +1757,6 @@ void phy_detach(struct phy_device *phydev)struct module *ndev_owner =NULL ;struct mii_bus *bus; -if (phydev->devlink) { -device_link_del (phydev->devlink); - phydev->devlink =NULL ; - } -if (phydev->sysfs_links) {if (dev)sysfs_remove_link (&dev->dev.kobj,"phydev" );include/linux/phy.h +0
-4@@ -505,8 +505,6 @@ struct macsec_ops;* * @mdio: MDIO bus this PHY is on * @drv: Pointer to the driver for this PHY instance -* @devlink: Create a link between phy dev and mac dev, if the external phy -* used by current mac interface is managed by another mac interface. * @phyindex: Unique id across the phy's parent tree of phys to address the PHY * from userspace, similar to ifindex. A zero index means the PHY * wasn't assigned an id yet. @@ -610,8 +608,6 @@ struct phy_device {/* And management functions */ const struct phy_driver *drv; -struct device_link *devlink; -u32 phyindex;u32 phy_id;
Quoted 58 lines · Abhishek Chauhan (ABC)
Quoted 47 lines · Wei Fang
Quoted 29 lines
Quoted 20 lines · Wei Fang
Quoted 19 lines
There is a potential crash issue when disabling and re-enabling the network port. When disabling the network port,
phy_detach()callsdevice_link_del()to remove the device link, but it does not clearphydev->devlink, sophydev->devlinkis not a NULL pointer. Then the network port is re-enabled, but ifphy_attach_direct()fails before callingdevice_link_add(), the code jumps to the "error" label and callsphy_detach(). Sincephydev->devlinkretains the old value from the previous attach/detach cycle,device_link_del()uses the old value, which accesses a NULL pointer and causes a crash. The simplified crash log is as follows.[ 24.702421] Call trace:
[ 24.704856]device_link_put_kref+0x20/0x120[ 24.709124]device_link_del+0x30/0x48
[ 24.712864]phy_detach+0x24/0x168
[ 24.716261]phy_attach_direct+0x168/0x3a4[ 24.720352]phylink_fwnode_phy_connect+0xc8/0x14c[ 24.725140]phylink_of_phy_connect+0x1c/0x34Quoted 6 lines
Quoted 5 lines
Therefore,
phydev->devlinkneeds to be cleared when the device link is deleted.Fixes bc66fa87d4fd ("net: phy: Add link between phy dev and mac dev")Signed-off-by Wei Fang <[email protected]>@Wei
What happens in case of shared mdio ?Quoted 7 lines
1. Device 23040000 has the mdio node of both the ethernet phy and device 23000000 references the phy-handle present in the Device 23040000
2. When rmmod of the driver happens
3. the parent devlink is already deleted.
4. This cause the child mdio to access an entry causing a corruption.
5. Thought this fix would help but i see that its not helping the case.My patch is only to fix the potential crash issue when re-enabling the network interface.
phy_detach()is not called when the MDIO controller driver is removed. Sophydev->devlinkis not cleared, but actually the device link has been removed byphy_device_remove()-->device_del(). Therefore, it will cause the crash when the MAC controller driver is removed.Quoted 2 lines
Wondering if this is a legacy issue with shared mdio framework.
Quoted 7 lines
I think this issue is also introduced by the commit
bc66fa87d4fd("net: phy: Add link between phy dev and mac dev"). I suggested to change theDL_FLAG_STATELESSflag toDL_FLAG_AUTOREMOVE_SUPPLIERto solve this issue, so that the consumer (MAC controller) driver will be automatically removed when the link is removed. The changes are as follows.
Quoted 3 lines
thanks a lot , Russell and Wei for your prompt response. I appreciate your help. let me test this change and get back.
Myself and Abhishek tested this suggestion from Wei and we now observe that the driver removal is successful without a crash. It looks like the flag 'DL_FLAG_AUTOREMOVE_SUPPLIER' is helping. Wei and Russell, please suggest the next steps.
Quoted 1 line
Review
Quoted 1 line
(quoted diff) +0
-0
b/drivers/net/phy/phy_device.c
Quoted 65 lines
Quoted 65 lines
index
73f9cb2e2844..a6d7acd73391100644drivers/net/phy/phy_device.c +11
-8@@ -1515,6 +1515,7 @@ int phy_attach_direct(struct net_device *dev, struct phy_device *phydev,struct mii_bus *bus = phydev->mdio.bus;struct device *d = &phydev->mdio.dev;struct module *ndev_owner =NULL ; +struct device_link *devlink;bool using_genphy = false;int err;@@ -1646,9 +1647,16 @@ int phy_attach_direct(struct net_device *dev, struct phy_device *phydev,* another mac interface, so we should create a device link between * phy dev and mac dev. */ -if (dev && phydev->mdio.bus->parent && dev->dev.parent != phydev->mdio.bus->parent) - phydev->devlink =device_link_add (dev->dev.parent, &phydev->mdio.dev, -DL_FLAG_PM_RUNTIME |DL_FLAG_STATELESS ); +if (dev && phydev->mdio.bus->parent && + dev->dev.parent != phydev->mdio.bus->parent) { + devlink =device_link_add (dev->dev.parent, &phydev->mdio.dev, +DL_FLAG_PM_RUNTIME | +DL_FLAG_AUTOREMOVE_SUPPLIER ); +if (!devlink) { + err = -ENOMEM ; + goto error; + } + }return err;@@ -1749,11 +1757,6 @@ void phy_detach(struct phy_device *phydev)struct module *ndev_owner =NULL ;struct mii_bus *bus; -if (phydev->devlink) { -device_link_del (phydev->devlink); - phydev->devlink =NULL ; - } -if (phydev->sysfs_links) {if (dev)sysfs_remove_link (&dev->dev.kobj,"phydev" );include/linux/phy.h +0
-4@@ -505,8 +505,6 @@ struct macsec_ops;* * @mdio: MDIO bus this PHY is on * @drv: Pointer to the driver for this PHY instance -* @devlink: Create a link between phy dev and mac dev, if the external phy -* used by current mac interface is managed by another mac interface. * @phyindex: Unique id across the phy's parent tree of phys to address the PHY * from userspace, similar to ifindex. A zero index means the PHY * wasn't assigned an id yet. @@ -610,8 +608,6 @@ struct phy_device {/* And management functions */ const struct phy_driver *drv; -struct device_link *devlink; -u32 phyindex;u32 phy_id;
Russell King (Oracle)
reviewerQuoted 6 lines · Wei Fang
I think this issue is also introduced by the commit
bc66fa87d4fd("net: phy: Add link between phy dev and mac dev"). I suggested to change theDL_FLAG_STATELESSflag toDL_FLAG_AUTOREMOVE_SUPPLIERto solve this issue, so that the consumer (MAC controller) driver will be automatically removed when the link is removed. The changes are as follows.
I suspect this still has problems. This is fine if the PHY device is going away and as you say device_del() is called.
However, you need to consider the case where a MAC driver attaches the PHY during .ndo_open and releases it during .ndo_release. These will happen multiple times.
Each time the MAC driver attaches to the PHY via .ndo_open, we will call device_link_add(), but the device link will not be removed when .ndo_release is called.
Either device_link_add() will fail, or we will eat memory each time the device is closed and re-opened.
If that is correct, then we're trading one problem for another.
Signature · 2 lines
RMK's Patch system: https://www.armlinux.org.uk/developer/patches/ FTTP is here! 80Mbps down 10Mbps up. Decent connectivity at last!
Quoted 8 lines · Wei Fang
Quoted 6 lines
I think this issue is also introduced by the commit
bc66fa87d4fd("net: phy: Add link between phy dev and mac dev"). I suggested to change theDL_FLAG_STATELESSflag toDL_FLAG_AUTOREMOVE_SUPPLIERto solve this issue, so that the consumer (MAC controller) driver will be automatically removed when the link is removed. The changes are as follows.
Quoted 6 lines
I suspect this still has problems. This is fine if the PHY device is going away and as you say
device_del()is called.However, you need to consider the case where a MAC driver attaches the PHY during .ndo_open and releases it during .ndo_release. These will happen multiple times.
.ndo_release? Do you mean .ndo_stop?
Quoted 7 lines
Each time the MAC driver attaches to the PHY via .ndo_open, we will call
device_link_add(), but the device link will not be removed when .ndo_release is called.Either
device_link_add()will fail, or we will eat memory each time the device is closed and re-opened.
Below is what I find in the kernel doc of device_link_add().
https://elixir.bootlin.com/linux/v6.15/source/drivers/base/core.c#L711
if a device link between the given @consumer and @supplier pair exists already when this function is called for them, the existing link will be returned regardless of its current type and status.
Therefore, it will not create new link each time the netdev is re-opened.
Quoted 3 lines
If that is correct, then we're trading one problem for another.
Quoted 25 lines · Wei Fang
Quoted 8 lines · Wei Fang
Quoted 6 lines
I think this issue is also introduced by the commit
bc66fa87d4fd("net: phy: Add link between phy dev and mac dev"). I suggested to change theDL_FLAG_STATELESSflag toDL_FLAG_AUTOREMOVE_SUPPLIERto solve this issue, so that the consumer (MAC controller) driver will be automatically removed when the link is removed. The changes are as follows.Quoted 6 lines
I suspect this still has problems. This is fine if the PHY device is going away and as you say
device_del()is called.However, you need to consider the case where a MAC driver attaches the PHY during .ndo_open and releases it during .ndo_release. These will happen multiple times.
.ndo_release? Do you mean .ndo_stop?
Quoted 7 lines
Each time the MAC driver attaches to the PHY via .ndo_open, we will call
device_link_add(), but the device link will not be removed when .ndo_release is called.Either
device_link_add()will fail, or we will eat memory each time the device is closed and re-opened.
Quoted 9 lines
Below is what I find in the kernel doc of
device_link_add().
https://elixir.bootlin.com/linux/v6.15/source/drivers/base/core.c#L711if a device link between the given @consumer and @supplier pair exists already when this function is called for them, the existing link will be returned regardless of its current type and status.
Therefore, it will not create new link each time the netdev is re-opened.
@Wei,
We were able to verify the suggestion what you gave us. No crash is seen now.
Should we raise a patch or shall we wait until this is clarified and Russell has no more further questions. ?
I am okay with anything.
Quoted 4 lines
Quoted 3 lines
If that is correct, then we're trading one problem for another.
Best Regards,
Wei Fang
Quoted 47 lines
Quoted 22 lines
Quoted 8 lines · Wei Fang
Quoted 6 lines
I think this issue is also introduced by the commit
bc66fa87d4fd("net: phy: Add link between phy dev and mac dev"). I suggested to change theDL_FLAG_STATELESSflag toDL_FLAG_AUTOREMOVE_SUPPLIERto solve this issue, so that the consumer (MAC controller) driver will be automatically removed when the link is removed. The changes are as follows.Quoted 6 lines
I suspect this still has problems. This is fine if the PHY device is going away and as you say
device_del()is called.However, you need to consider the case where a MAC driver attaches the PHY during .ndo_open and releases it during .ndo_release. These will happen multiple times.
.ndo_release? Do you mean .ndo_stop?
Quoted 5 lines
Each time the MAC driver attaches to the PHY via .ndo_open, we will call
device_link_add(), but the device link will not be removed when .ndo_release is called.Quoted 6 lines
Quoted 2 lines
Either
device_link_add()will fail, or we will eat memory each time the device is closed and re-opened.Below is what I find in the kernel doc of
device_link_add().
https://eur01.safelinks.protection.outlook.com/?url=https%3A%2F%2Felixir.bootlin.com%2Flinux%2Fv6.15%2Fsource%2Fdrivers%2Fbase%2Fcore.c%23L7
Quoted 1 line
11&data=05%7C02%7Cwei.fang%40nxp.com%7C078b59e45bd54f13cfea08dda 49477a
Quoted 1 line
4%7C686ea1d3bc2b4c6fa92cd99c5c301635%7C0%7C0%7C638847679768356 722%7CUn
Quoted 1 line
known%7CTWFpbGZsb3d8eyJFbXB0eU1hcGkiOnRydWUsIlYiOiIwLjAuMDAwMCI sIlAiOi
Quoted 1 line
JXaW4zMiIsIkFOIjoiTWFpbCIsIldUIjoyfQ%3D%3D%7C0%7C%7C%7C&sdata=iFtK o%2F
Quoted 6 lines
NfwPVYbxLBtZM8Y15Aa9eomUNhPTRfnEb0L7c%3D&reserved=0
if a device link between the given @consumer and @supplier pair exists already when this function is called for them, the existing link will be returned regardless of its current type and status.
Quoted 9 lines
Quoted 2 lines
Therefore, it will not create new link each time the netdev is re-opened.
@Wei,
We were able to verify the suggestion what you gave us.
No crash is seen now.
Should we raise a patch or shall we wait until this is clarified and Russell has no more further questions. ?
I'm okay, but I don't know if Russel has any other concerns. And if there are no more comments, you can raise a patch. :)
Quoted 6 lines
I am okay with anything.
Quoted 4 lines
Quoted 3 lines
If that is correct, then we're trading one problem for another.
This patch was applied to netdev/net.git (main) by Jakub Kicinski <[email protected]>:
Hello:
This patch was applied to netdev/net.git (main)
by Jakub Kicinski <[email protected]>:
Quoted 11 lines · On Fri, 23 May 2025 16:37:59 +0800 you wrote:
There is a potential crash issue when disabling and re-enabling the network port. When disabling the network port,
phy_detach()callsdevice_link_del()to remove the device link, but it does not clearphydev->devlink, sophydev->devlinkis not a NULL pointer. Then the network port is re-enabled, but ifphy_attach_direct()fails before callingdevice_link_add(), the code jumps to the "error" label and callsphy_detach(). Sincephydev->devlinkretains the old value from the previous attach/detach cycle,device_link_del()uses the old value, which accesses a NULL pointer and causes a crash. The simplified crash log is as follows.
Quoted 1 line
[...]
Here is the summary with links:
- [v2,net] net: phy: clear phydev->devlink when the link is deleted
https://git.kernel.org/netdev/net/c/0795b05a59b1
You are awesome, thank you!
Signature · 2 lines
Deet-doot-dot, I am a bot. https://korg.docs.kernel.org/patchwork/pwbot.html