Skip to contentSkip to conversation

net: phy: Fix lack of reference count on PHY driver

PATCH In review Florian Fainelli 6 participants 16 messages Feb 1, 2017 – Feb 9, 2017

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

PATCH 1/1 net: phy: Fix lack of reference count on PHY driver

From: Mao Wenan <[email protected]>

There is currently no reference count being held on the PHY driver, which makes it possible to remove the PHY driver module while the PHY state machine is running and polling the PHY. This could cause crashes similar to this one to show up:

[   43.361162] BUG: unable to handle kernel NULL pointer dereference at 0000000000000140
[   43.361162] IP: phy_state_machine+0x32/0x490
[   43.361162] PGD 59dc067
[   43.361162] PUD 0
[   43.361162]
[   43.361162] Oops: 0000 [#1] SMP
[   43.361162] Modules linked in: dsa_loop [last unloaded: broadcom]
[   43.361162] CPU: 0 PID: 1299 Comm: kworker/0:3 Not tainted 4.10.0-rc5+ #415
[   43.361162] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996),
BIOS Ubuntu-1.8.2-1ubuntu2 04/01/2014
[   43.361162] Workqueue: events_power_efficient phy_state_machine
[   43.361162] task: ffff880006782b80 task.stack: ffffc90000184000
[   43.361162] RIP: 0010:phy_state_machine+0x32/0x490
[   43.361162] RSP: 0018:ffffc90000187e18 EFLAGS: 00000246
[   43.361162] RAX: 0000000000000000 RBX: ffff8800059e53c0 RCX:
ffff880006a15c60
[   43.361162] RDX: ffff880006782b80 RSI: 0000000000000000 RDI:
ffff8800059e5428
[   43.361162] RBP: ffffc90000187e48 R08: ffff880006a15c40 R09:
0000000000000000
[   43.361162] R10: 0000000000000000 R11: 0000000000000000 R12:
ffff8800059e5428
[   43.361162] R13: ffff8800059e5000 R14: 0000000000000000 R15:
ffff880006a15c40
[   43.361162] FS:  0000000000000000(0000) GS:ffff880006a00000(0000)
knlGS:0000000000000000
[   43.361162] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[   43.361162] CR2: 0000000000000140 CR3: 0000000005979000 CR4:
00000000000006f0
[   43.361162] Call Trace:
[   43.361162]  process_one_work+0x1b4/0x3e0
[   43.361162]  worker_thread+0x43/0x4d0
[   43.361162]  ? __schedule+0x17f/0x4e0
[   43.361162]  kthread+0xf7/0x130
[   43.361162]  ? process_one_work+0x3e0/0x3e0
[   43.361162]  ? kthread_create_on_node+0x40/0x40
[   43.361162]  ret_from_fork+0x29/0x40
[   43.361162] Code: 56 41 55 41 54 4c 8d 67 68 53 4c 8d af 40 fc ff ff

48 89 fb 4c 89 e7 48 83 ec 08 e8 c9 9d 27 00 48 8b 83 60 ff ff ff 44 8b 73 98 <48> 8b 90 40 01 00 00 44 89 f0 48 85 d2 74 08 4c 89 ef ff d2 8b

Keep references on the PHY driver module right before we are going to utilize it in phy_attach_direct(), and conversely when we don't use it anymore in phy_detach().

[florian: rebase, rework commit message]
---
 drivers/net/phy/phy_device.c | 7 +++++++
 1 file changed, 7 insertions(+)
drivers/net/phy/phy_device.c +7 -0
index 92b08383cafa..0d8f4d3847f6 100644
@@ -920,6 +920,11 @@ int phy_attach_direct(struct net_device *dev, struct phy_device *phydev,
 		return -EIO;
 	}
 
+	if (!try_module_get(d->driver->owner)) {
+		dev_err(&dev->dev, "failed to get the device driver module\n");
+		return -EIO;
+	}
+
 	get_device(d);
 
 	/* Assume that if there is no driver, that it doesn't
@@ -977,6 +982,7 @@ int phy_attach_direct(struct net_device *dev, struct phy_device *phydev,
 error:
 	phy_detach(phydev);
 	put_device(d);
+	module_put(d->driver->owner);
 	if (ndev_owner != bus->owner)
 		module_put(bus->owner);
 	return err;
@@ -1059,6 +1065,7 @@ void phy_detach(struct phy_device *phydev)
 	bus = phydev->mdio.bus;
 
 	put_device(&phydev->mdio.dev);
+	module_put(phydev->mdio.dev.driver->owner);
 	if (ndev_owner != bus->owner)
 		module_put(bus->owner);
 }
Signature · 1 line
2.9.3
#1 lore raw
Quoted 6 lines · Florian Fainelli

From: Mao Wenan <[email protected]>

There is currently no reference count being held on the PHY driver, which makes it possible to remove the PHY driver module while the PHY state machine is running and polling the PHY. This could cause crashes similar to this one to show up:

Does this really solve the problem? What if you use sysfs to unbind the driver but without removing the module?

Signature · 3 lines
RMK's Patch system: http://www.armlinux.org.uk/developer/patches/
FTTC broadband for 0.8mile line: currently at 9.6Mbps down 400kbps up
according to speedtest.net.
#2 lore raw
Quoted 10 lines · Russell King - ARM Linux
Quoted 6 lines · Florian Fainelli

From: Mao Wenan <[email protected]>

There is currently no reference count being held on the PHY driver, which makes it possible to remove the PHY driver module while the PHY state machine is running and polling the PHY. This could cause crashes similar to this one to show up:

Does this really solve the problem? What if you use sysfs to unbind the driver but without removing the module?

I think that's a problem, and the patch is just solving a symptom of it.

If a phy driver is unbound from a device, phy_remove() will be called. This will set the state to PHY_DOWN (under the mutex) before calling the driver's remove function (if any), and finally setting phydev->drv to NULL.

If phy_state_machine() is called after that point, then:

void phy_state_machine(struct work_struct *work)
{
...

        if (phydev->drv->link_change_notify)
                phydev->drv->link_change_notify(phydev);

which happens unconditionally, causes a NULL pointer dereference, which is probably the same NULL pointer dereference given in Mao Wenan's patch description.

It looks to me as if that's the only case where this can happen, so maybe the above needs to be:

        if (phydev->drv && phydev->drv->link_change_notify)
                phydev->drv->link_change_notify(phydev);

Also, I'd suggest making sure that the workqueue is flushed in phy_remove() after setting phydev->drv to NULL to ensure that the workqueue isn't running while the phy driver is being unbound, which should also make module removal safe(r). I haven't fully analysed that though.

Signature · 3 lines
RMK's Patch system: http://www.armlinux.org.uk/developer/patches/
FTTC broadband for 0.8mile line: currently at 9.6Mbps down 400kbps up
according to speedtest.net.
#3 lore raw
Quoted 9 lines · Russell King - ARM Linux
Quoted 8 lines · Russell King - ARM Linux
Quoted 6 lines · Florian Fainelli

From: Mao Wenan <[email protected]>

There is currently no reference count being held on the PHY driver, which makes it possible to remove the PHY driver module while the PHY state machine is running and polling the PHY. This could cause crashes similar to this one to show up:

Quoted 5 lines
Quoted 2 lines

Does this really solve the problem? What if you use sysfs to unbind the driver but without removing the module?

I think that's a problem, and the patch is just solving a symptom of it.

You are right, but there is still a fundamental problem IMHO in that you should not be able to rmmod a PHY driver as long as a network device is attached to the PHY, and if the PHY driver is attached from several different network devices, they should all have a way to prevent a PHY driver rmmod, each of them incrementing the driver refcount, which is what the patche from Maowan does here.

Quoted 11 lines

If a phy driver is unbound from a device, phy_remove() will be called. This will set the state to PHY_DOWN (under the mutex) before calling the driver's remove function (if any), and finally setting phydev->drv to NULL.

If phy_state_machine() is called after that point, then:

void phy_state_machine(struct work_struct *work)
{
...

Quoted 6 lines
        if (phydev->drv->link_change_notify)
                phydev->drv->link_change_notify(phydev);

which happens unconditionally, causes a NULL pointer dereference, which is probably the same NULL pointer dereference given in Mao Wenan's patch description.

Yep, that's exactly the location, but then after fixing that, we can still crash in other locations, e.g: if we bring down an interface that was attached to the PHY we would now crash in phy_suspend -> phy_ethtool_get_wol

All of that can be fixed, and actually should be fixed, but it still feels like we should have an easier way to prevent the driver removal IMHO.

Quoted 4 lines

It looks to me as if that's the only case where this can happen, so maybe the above needs to be:

Quoted 8 lines
        if (phydev->drv && phydev->drv->link_change_notify)
                phydev->drv->link_change_notify(phydev);

Also, I'd suggest making sure that the workqueue is flushed in phy_remove() after setting phydev->drv to NULL to ensure that the workqueue isn't running while the phy driver is being unbound, which should also make module removal safe(r). I haven't fully analysed that though.

That is reasonable to do as well, thanks!

Signature · 1 line
Florian
maintainer
#4 lore raw

From: Florian Fainelli <[email protected]>
Date: Wed, 1 Feb 2017 10:55:46 -0800

Quoted 6 lines

You are right, but there is still a fundamental problem IMHO in that you should not be able to rmmod a PHY driver as long as a network device is attached to the PHY, and if the PHY driver is attached from several different network devices, they should all have a way to prevent a PHY driver rmmod, each of them incrementing the driver refcount, which is what the patche from Maowan does here.

It briefly occurred to me that we might want to be able to disconnect PHYs to allow an unload using notifiers, the same way that when you take a netdevice down we emit notifiers so that all of the references to the netdevice can release themselves.

I have no idea how well that would work, or whether it is valuable or not. But it is another way to handle this.

But that is a longer-term thing even if we want to go that way, and thus grabbing the proper refs is the right things to do for now.

#5 lore raw
Quoted 15 lines · David Miller

From: Florian Fainelli <[email protected]>
Date: Wed, 1 Feb 2017 10:55:46 -0800

Quoted 6 lines

You are right, but there is still a fundamental problem IMHO in that you should not be able to rmmod a PHY driver as long as a network device is attached to the PHY, and if the PHY driver is attached from several different network devices, they should all have a way to prevent a PHY driver rmmod, each of them incrementing the driver refcount, which is what the patche from Maowan does here.

It briefly occurred to me that we might want to be able to disconnect PHYs to allow an unload using notifiers, the same way that when you take a netdevice down we emit notifiers so that all of the references to the netdevice can release themselves.

Quoted 5 lines

I have no idea how well that would work, or whether it is valuable or not. But it is another way to handle this.

But that is a longer-term thing even if we want to go that way, and thus grabbing the proper refs is the right things to do for now.

It's something I'm effectively already doing as part of my phylink project for SFP support, since you can hot-unplug a copper SFP module, and the PHY on the copper SFP module will be gone. phylink, however, sits between the network driver and phylib, which isn't ideal.

Signature · 3 lines
RMK's Patch system: http://www.armlinux.org.uk/developer/patches/
FTTC broadband for 0.8mile line: currently at 9.6Mbps down 400kbps up
according to speedtest.net.
#6 lore raw
Quoted 19 lines · Russell King - ARM Linux
Quoted 18 lines · David Miller

From: Florian Fainelli <[email protected]>
Date: Wed, 1 Feb 2017 10:55:46 -0800

Quoted 6 lines

You are right, but there is still a fundamental problem IMHO in that you should not be able to rmmod a PHY driver as long as a network device is attached to the PHY, and if the PHY driver is attached from several different network devices, they should all have a way to prevent a PHY driver rmmod, each of them incrementing the driver refcount, which is what the patche from Maowan does here.

It briefly occurred to me that we might want to be able to disconnect PHYs to allow an unload using notifiers, the same way that when you take a netdevice down we emit notifiers so that all of the references to the netdevice can release themselves.

I have no idea how well that would work, or whether it is valuable or not. But it is another way to handle this.

Quoted 8 lines
Quoted 2 lines

But that is a longer-term thing even if we want to go that way, and thus grabbing the proper refs is the right things to do for now.

It's something I'm effectively already doing as part of my phylink project for SFP support, since you can hot-unplug a copper SFP module, and the PHY on the copper SFP module will be gone. phylink, however, sits between the network driver and phylib, which isn't ideal.

So, for the "net" tree what should we do? I don't really think that we should be able to let the PHY state machine run without a PHY driver bound to the device, at best nothing happens, at worse, we just crash and burn without further chance of recovering.

Signature · 1 line
Florian
#7 lore raw
In reply to Russell King - ARM Linux
Quoted 11 lines · Russell King - ARM Linux

It looks to me as if that's the only case where this can happen, so maybe the above needs to be:

        if (phydev->drv && phydev->drv->link_change_notify)
                phydev->drv->link_change_notify(phydev);

Also, I'd suggest making sure that the workqueue is flushed in phy_remove() after setting phydev->drv to NULL to ensure that the workqueue isn't running while the phy driver is being unbound, which should also make module removal safe(r). I haven't fully analysed that though.

I suspect nobody has actually ever tested that, because it's pretty badly broken at the moment... working on it.

Signature · 1 line
Florian
maintainer
#8 lore raw
In reply to Florian Fainelli

From: Florian Fainelli <[email protected]>
Date: Tue, 31 Jan 2017 18:46:43 -0800

Quoted 6 lines

From: Mao Wenan <[email protected]>

There is currently no reference count being held on the PHY driver, which makes it possible to remove the PHY driver module while the PHY state machine is running and polling the PHY. This could cause crashes similar to this one to show up:

...

Quoted 7 lines

Keep references on the PHY driver module right before we are going to utilize it in phy_attach_direct(), and conversely when we don't use it anymore in phy_detach().

[florian: rebase, rework commit message]

I think grabbing the module reference is the only easy fix we can do for now.

Hot plugging PHYs and notifications and all of that business is net-next material.

Florian, do you need to respin this with the workqueue or whatever suggestion Russell made?

Thanks.

#9 lore raw
Quoted 18 lines · David Miller

From: Florian Fainelli <[email protected]>
Date: Tue, 31 Jan 2017 18:46:43 -0800

Quoted 6 lines

From: Mao Wenan <[email protected]>

There is currently no reference count being held on the PHY driver, which makes it possible to remove the PHY driver module while the PHY state machine is running and polling the PHY. This could cause crashes similar to this one to show up:

...

Quoted 7 lines

Keep references on the PHY driver module right before we are going to utilize it in phy_attach_direct(), and conversely when we don't use it anymore in phy_detach().

[florian: rebase, rework commit message]
Quoted 8 lines

I think grabbing the module reference is the only easy fix we can do for now.

Hot plugging PHYs and notifications and all of that business is net-next material.

Florian, do you need to respin this with the workqueue or whatever suggestion Russell made?

That seems to deserve a separate fix of its own it seems, because it becomes tangential to this particular problem.

Signature · 1 line
Florian
maintainer
#10 lore raw

From: Florian Fainelli <[email protected]>
Date: Thu, 2 Feb 2017 19:47:43 -0800

Quoted 22 lines · David Miller
Quoted 21 lines

From: Florian Fainelli <[email protected]>
Date: Tue, 31 Jan 2017 18:46:43 -0800

Quoted 6 lines

From: Mao Wenan <[email protected]>

There is currently no reference count being held on the PHY driver, which makes it possible to remove the PHY driver module while the PHY state machine is running and polling the PHY. This could cause crashes similar to this one to show up:

...

Quoted 7 lines

Keep references on the PHY driver module right before we are going to utilize it in phy_attach_direct(), and conversely when we don't use it anymore in phy_detach().

[florian: rebase, rework commit message]

I think grabbing the module reference is the only easy fix we can do for now.

Quoted 8 lines
Quoted 5 lines

Hot plugging PHYs and notifications and all of that business is net-next material.

Florian, do you need to respin this with the workqueue or whatever suggestion Russell made?

That seems to deserve a separate fix of its own it seems, because it becomes tangential to this particular problem.

Ok, patch applied then.

Thanks!

#11 lore raw
In reply to David Miller
Quoted 2 lines · David Miller

Hot plugging PHYs and notifications and all of that business is net-next material.

I was talking more about unbinding of the driver, which is something that can be done today, eg:

$ ls -l /sys/bus/mdio_bus/drivers/Atheros\ 8035\ ethernet/
total 0
lrwxrwxrwx 1 root root    0 Feb  3 09:49 2188000.ethernet:00 -> ../../../../devices/soc0/soc/2100000.aips-bus/2188000.ethernet/mdio_bus/2188000.ethernet/2188000.ethernet:00
--w------- 1 root root 4096 Feb  3 09:49 bind
--w------- 1 root root 4096 Feb  3 09:49 uevent
--w------- 1 root root 4096 Feb  3 09:49 unbind
$ echo 2188000.ethernet:00 > /sys/bus/mdio_bus/drivers/Atheros\ 8035\ ethernet/unbind

is all it takes, and the same oops will happen. Try it on a box you don't care about crashing. :)

This is my point - locking the module into the kernel using try_module_get() doesn't actually fix the problem where drivers are concerned, it just has the illusion of being safe.

Signature · 3 lines
RMK's Patch system: http://www.armlinux.org.uk/developer/patches/
FTTC broadband for 0.8mile line: currently at 9.6Mbps down 400kbps up
according to speedtest.net.
#12 lore raw
Quoted 15 lines · Russell King - ARM Linux
Quoted 2 lines · David Miller

Hot plugging PHYs and notifications and all of that business is net-next material.

I was talking more about unbinding of the driver, which is something that can be done today, eg:

$ ls -l /sys/bus/mdio_bus/drivers/Atheros\ 8035\ ethernet/
total 0
lrwxrwxrwx 1 root root    0 Feb  3 09:49 2188000.ethernet:00 -> ../../../../devices/soc0/soc/2100000.aips-bus/2188000.ethernet/mdio_bus/2188000.ethernet/2188000.ethernet:00
--w------- 1 root root 4096 Feb  3 09:49 bind
--w------- 1 root root 4096 Feb  3 09:49 uevent
--w------- 1 root root 4096 Feb  3 09:49 unbind
$ echo 2188000.ethernet:00 > /sys/bus/mdio_bus/drivers/Atheros\ 8035\ ethernet/unbind
Quoted 6 lines

is all it takes, and the same oops will happen. Try it on a box you don't care about crashing. :)

This is my point - locking the module into the kernel using try_module_get() doesn't actually fix the problem where drivers are concerned, it just has the illusion of being safe.

I have some patches for that too, along with making bind, unbind work again, because the state machine seems to be hosed (another thing that was very well tested...).

Signature · 1 line
Florian
#13 lore raw
In reply to Florian Fainelli

Hi all,

We're seeing a new boot-time crash[1] on SMSC911x hardware from this patch in today's HEAD (as cafe8df8b9bc)...

Quoted 64 lines · Florian Fainelli

From: Mao Wenan <[email protected]>

There is currently no reference count being held on the PHY driver, which makes it possible to remove the PHY driver module while the PHY state machine is running and polling the PHY. This could cause crashes similar to this one to show up:

[   43.361162] BUG: unable to handle kernel NULL pointer dereference at 0000000000000140
[   43.361162] IP: phy_state_machine+0x32/0x490
[   43.361162] PGD 59dc067
[   43.361162] PUD 0
[   43.361162]
[   43.361162] Oops: 0000 [#1] SMP
[   43.361162] Modules linked in: dsa_loop [last unloaded: broadcom]
[   43.361162] CPU: 0 PID: 1299 Comm: kworker/0:3 Not tainted 4.10.0-rc5+ #415
[   43.361162] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996),
BIOS Ubuntu-1.8.2-1ubuntu2 04/01/2014
[   43.361162] Workqueue: events_power_efficient phy_state_machine
[   43.361162] task: ffff880006782b80 task.stack: ffffc90000184000
[   43.361162] RIP: 0010:phy_state_machine+0x32/0x490
[   43.361162] RSP: 0018:ffffc90000187e18 EFLAGS: 00000246
[   43.361162] RAX: 0000000000000000 RBX: ffff8800059e53c0 RCX:
ffff880006a15c60
[   43.361162] RDX: ffff880006782b80 RSI: 0000000000000000 RDI:
ffff8800059e5428
[   43.361162] RBP: ffffc90000187e48 R08: ffff880006a15c40 R09:
0000000000000000
[   43.361162] R10: 0000000000000000 R11: 0000000000000000 R12:
ffff8800059e5428
[   43.361162] R13: ffff8800059e5000 R14: 0000000000000000 R15:
ffff880006a15c40
[   43.361162] FS:  0000000000000000(0000) GS:ffff880006a00000(0000)
knlGS:0000000000000000
[   43.361162] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[   43.361162] CR2: 0000000000000140 CR3: 0000000005979000 CR4:
00000000000006f0
[   43.361162] Call Trace:
[   43.361162]  process_one_work+0x1b4/0x3e0
[   43.361162]  worker_thread+0x43/0x4d0
[   43.361162]  ? __schedule+0x17f/0x4e0
[   43.361162]  kthread+0xf7/0x130
[   43.361162]  ? process_one_work+0x3e0/0x3e0
[   43.361162]  ? kthread_create_on_node+0x40/0x40
[   43.361162]  ret_from_fork+0x29/0x40
[   43.361162] Code: 56 41 55 41 54 4c 8d 67 68 53 4c 8d af 40 fc ff ff

48 89 fb 4c 89 e7 48 83 ec 08 e8 c9 9d 27 00 48 8b 83 60 ff ff ff 44 8b 73 98 <48> 8b 90 40 01 00 00 44 89 f0 48 85 d2 74 08 4c 89 ef ff d2 8b

Keep references on the PHY driver module right before we are going to utilize it in phy_attach_direct(), and conversely when we don't use it anymore in phy_detach().

[florian: rebase, rework commit message]
---
 drivers/net/phy/phy_device.c | 7 +++++++
 1 file changed, 7 insertions(+)
drivers/net/phy/phy_device.c +0 -0
index 92b08383cafa..0d8f4d3847f6 100644

drivers/net/phy/phy_device.c int phy_attach_direct(struct net_device *dev, struct phy_device *phydev, · line 923

Quoted 4 lines
drivers/net/phy/phy_device.c +1 -0
@@ -920,6 +920,11 @@ int phy_attach_direct(struct net_device *dev, struct phy_device *phydev,
 		return -EIO;
 	}
 
+	if (!try_module_get(d->driver->owner)) {

...because d->driver is NULL here. I'm a little surprised static checking hasn't picked this up, because right below we test "if (d->driver)".

I won't pretend to understand this code and its interaction with the SMSC driver anywhere near enough to suggest a patch myself, so consider this just a panic bug report in the hope of preventing 4.10 being horribly broken.

Thanks,
Robin.

[1]:

[    4.689360] Unable to handle kernel NULL pointer dereference at
virtual address 00000010
[    4.697371] pgd = ffffff80092bd000
[    4.700740] [00000010] *pgd=00000009ffffe003, *pud=00000009ffffe003,
*pmd=0000000000000000
[    4.708936] Internal error: Oops: 96000005 [#1] PREEMPT SMP
[    4.714446] Modules linked in:
[    4.717467] CPU: 2 PID: 1 Comm: swapper/0 Tainted: G        W
4.10.0-rc7+ #1577
[    4.725214] Hardware name: ARM Juno development board (r1) (DT)
[    4.731070] task: ffffffc976850000 task.stack: ffffffc976858000
[    4.736929] PC is at phy_attach_direct+0x98/0x1c8
[    4.741581] LR is at phy_attach_direct+0x94/0x1c8
[    4.746233] pc : [<ffffff80085e0340>] lr : [<ffffff80085e033c>]
pstate: 60000145
[    4.753548] sp : ffffffc97685bb30
[    4.756823] x29: ffffffc97685bb30 x28: ffffff8008c16040
[    4.762078] x27: ffffff8008e65c20 x26: ffffff8008c9045c
[    4.767333] x25: 0000000000000000 x24: 0000000000000001
[    4.772587] x23: ffffffc976b25000 x22: 0000000000000000
[    4.777841] x21: ffffff8008684818 x20: ffffffc976b25800
[    4.783095] x19: ffffffc976825000 x18: 0000000000000010
[    4.788349] x17: 0000000000000000 x16: 0000000000000001
[    4.793602] x15: 0000000000000006 x14: ffffff8088e5ad4f
[    4.798855] x13: ffffff8008e5ad5d x12: ffffff8008e5d160
[    4.804109] x11: ffffffc97685b890 x10: ffffff8008d889d8
[    4.809362] x9 : ffffff80084bee48 x8 : 2020202020203030
[    4.814616] x7 : 3835326236373963 x6 : 00000000000001e2
[    4.819870] x5 : ffffff8008e5c9e8 x4 : 0000000000000000
[    4.825123] x3 : 0000000000000000 x2 : ffffff8008d9ee68
[    4.830376] x1 : ffffffc976850000 x0 : 0000000000000000
[    4.835629]
[    4.837098] Process swapper/0 (pid: 1, stack limit = 0xffffffc976858000)
[    4.843727] Stack: (0xffffffc97685bb30 to 0xffffffc97685c000)
[    4.849411] bb20:                                   ffffffc97685bb80
ffffff80085e0570
[    4.857158] bb40: ffffffc976b25800 ffffffc976825800 ffffff8008684818
ffffffc976825048
[    4.864906] bb60: 0000000000001002 ffffffc976825000 0000000000001002
0000000000000000
[    4.872653] bb80: ffffffc97685bbb0 ffffff8008685e68 ffffffc976825000
ffffffc976825800
[    4.880400] bba0: ffffff80089db328 ffffffc976825800 ffffffc97685bc40
ffffff800880aff8
[    4.888147] bbc0: ffffffc976825000 0000000000000000 ffffff80089db328
ffffffc976825048
[    4.895894] bbe0: 0000000000001002 ffffff8008a0c000 0000000000001002
ffffff8008c9045c
[    4.903640] bc00: ffffff8008e65c20 ffffff8008c16040 ffffffc97685bc40
ffffff800880af78
[    4.911387] bc20: ffffffc976b25800 0000000000001003 ffffff80089db328
0000000000000000
[    4.919134] bc40: ffffffc97685bc80 ffffff800880b2d0 ffffffc976825000
0000000000001003
[    4.926881] bc60: 0000000000000001 0000000000000000 ffffffc976825000
ffffffc976825000
[    4.934628] bc80: ffffffc97685bcc0 ffffff800880b3a0 ffffffc976825000
0000000000001002
[    4.942375] bca0: ffffff8008d11488 0000000000000000 ffffff8008e40000
ffffff8008c11ba8
[    4.950121] bcc0: ffffffc97685bcf0 ffffff8008cda728 ffffff8008d11000
ffffffc976825000
[    4.957868] bce0: ffffff8008d11488 0000000000000001 ffffffc97685bdd0
ffffff80080830f8
[    4.958250] atkbd serio0: keyboard reset failed on 1c060000.kmi
[    4.971466] bd00: ffffff8008cda554 ffffffc976850000 0000000000000000
ffffff8008ce0ad8
[    4.979213] bd20: ffffff8008c7e8f0 ffffff8008ce0ad0 ffffff8008e53000
ffffff8008c9045c
[    4.986960] bd40: ffffff8008d4d2f0 0000000000000000 0000000000000000
ffffff8008ce0ad8
[    4.994707] bd60: ffffff8008c7e8f0 ffffff800834f940 ffffffc97685bda0
ffffff800823a964
[    5.002454] bd80: 000000027685bdc0 ffffff8008e40a18 ffffff8008cd8d58
0000000176850000
[    5.010200] bda0: 0000000000000000 ffffff8008ce0ad8 ffffff8008c7e8f0
ffffff80080830f8
[    5.017947] bdc0: ffffffc97685bdd0 ffffff80080830f8 ffffffc97685be40
ffffff8008c90ca0
[    5.025694] bde0: 0000000000000133 ffffff8008e53000 0000000000000007
ffffff8008ce0ad8
[    5.033441] be00: ffffff8008d4d100 0000000000000000 0000000000000000
ffffff8008b5ae60
[    5.041188] be20: 0000000700000007 ffffff8008c9045c 0000000000000000
ffffff8008c7e8f0
[    5.048934] be40: ffffffc97685bea0 ffffff80088cdca0 ffffff80088cdc90
0000000000000000
[    5.056681] be60: 0000000000000000 0000000000000000 0000000000000000
0000000000000000
[    5.064427] be80: 0000000000000000 0000000000000000 0000000000000000
0000000000000000
[    5.072174] bea0: 0000000000000000 ffffff8008082ec0 ffffff80088cdc90
0000000000000000
[    5.079920] bec0: 0000000000000000 0000000000000000 0000000000000000
0000000000000000
[    5.087666] bee0: 0000000000000000 0000000000000000 0000000000000000
0000000000000000
[    5.095412] bf00: 0000000000000000 0000000000000000 0000000000000000
0000000000000000
[    5.103159] bf20: 0000000000000000 0000000000000000 0000000000000000
0000000000000000
[    5.110905] bf40: 0000000000000000 0000000000000000 0000000000000000
0000000000000000
[    5.118651] bf60: 0000000000000000 0000000000000000 0000000000000000
0000000000000000
[    5.126397] bf80: 0000000000000000 0000000000000000 0000000000000000
0000000000000000
[    5.134144] bfa0: 0000000000000000 0000000000000000 0000000000000000
0000000000000000
[    5.141890] bfc0: 0000000000000000 0000000000000005 0000000000000000
0000000000000000
[    5.149637] bfe0: 0000000000000000 0000000000000000 0800410001480110
0010001001080000
[    5.157382] Call trace:
[    5.159797] Exception stack(0xffffffc97685b960 to 0xffffffc97685ba90)
[    5.166169] b960: ffffffc976825000 0000008000000000 ffffffc97685bb30
ffffff80085e0340
[    5.173916] b980: ffffff8008b5c3e8 ffffff8008d4e000 ffffff8008e5c9e8
ffffff8008e11760
[    5.181662] b9a0: ffffff8008e5ad50 0000000108e5a7e8 ffffffc97685ba50
ffffff80081012b4
[    5.189409] b9c0: ffffffc976825000 ffffffc976b25800 ffffff8008684818
0000000000000000
[    5.197156] b9e0: ffffffc976b25000 0000000000000001 0000000000000000
ffffff8008c9045c
[    5.204902] ba00: 0000000000000000 ffffffc976850000 ffffff8008d9ee68
0000000000000000
[    5.212649] ba20: 0000000000000000 ffffff8008e5c9e8 00000000000001e2
3835326236373963
[    5.220396] ba40: 2020202020203030 ffffff80084bee48 ffffff8008d889d8
ffffffc97685b890
[    5.228143] ba60: ffffff8008e5d160 ffffff8008e5ad5d ffffff8088e5ad4f
0000000000000006
[    5.235888] ba80: 0000000000000001 0000000000000000
[    5.240713] [<ffffff80085e0340>] phy_attach_direct+0x98/0x1c8
[    5.246396] [<ffffff80085e0570>] phy_connect_direct+0x20/0x78
[    5.252081] [<ffffff8008685e68>] smsc911x_open+0x578/0xa50
[    5.257508] [<ffffff800880aff8>] __dev_open+0xb8/0x128
[    5.262589] [<ffffff800880b2d0>] __dev_change_flags+0x98/0x148
[    5.268358] [<ffffff800880b3a0>] dev_change_flags+0x20/0x60
[    5.273871] [<ffffff8008cda728>] ip_auto_config+0x1d4/0xeb0
[    5.279384] [<ffffff80080830f8>] do_one_initcall+0x38/0x120
[    5.284896] [<ffffff8008c90ca0>] kernel_init_freeable+0x144/0x1e8
[    5.290923] [<ffffff80088cdca0>] kernel_init+0x10/0x100
[    5.296090] [<ffffff8008082ec0>] ret_from_fork+0x10/0x50
[    5.301344] Code: 90002f80 91328000 97edd06a f9404680 (f9400800)
[    5.307423] ---[ end trace 69bc6e46c6317a9c ]---
[    5.312053] Kernel panic - not syncing: Attempted to kill init!

exitcode=0x0000000b

Quoted 24 lines
drivers/net/phy/phy_device.c +6 -0
+		dev_err(&dev->dev, "failed to get the device driver module\n");
+		return -EIO;
+	}
+
 	get_device(d);
 
 	/* Assume that if there is no driver, that it doesn't
@@ -977,6 +982,7 @@ int phy_attach_direct(struct net_device *dev, struct phy_device *phydev,
 error:
 	phy_detach(phydev);
 	put_device(d);
+	module_put(d->driver->owner);
 	if (ndev_owner != bus->owner)
 		module_put(bus->owner);
 	return err;
@@ -1059,6 +1065,7 @@ void phy_detach(struct phy_device *phydev)
 	bus = phydev->mdio.bus;
 
 	put_device(&phydev->mdio.dev);
+	module_put(phydev->mdio.dev.driver->owner);
 	if (ndev_owner != bus->owner)
 		module_put(bus->owner);
 }
maintainer
#14 lore raw
Quoted 4 lines · Robin Murphy

Hi all,

We're seeing a new boot-time crash[1] on SMSC911x hardware from this patch in today's HEAD (as cafe8df8b9bc)...

Hi Robin

Thank for the report. See the discussion on netdev under the subject "Kernel crashes in phy_attach_direct()"

Andrew

#15 lore raw
Quoted 15 lines

-----Original Message-----
From: Andrew Lunn [mailto:[email protected]]
Sent: Thursday, February 09, 2017 12:24 AM
To: Robin Murphy
Cc: Florian Fainelli; [email protected]; [email protected]; [email protected]; maowenan; Catalin Marinas
Subject: Re: [net] net: phy: Fix lack of reference count on PHY driver

Quoted 4 lines · Robin Murphy

Hi all,

We're seeing a new boot-time crash[1] on SMSC911x hardware from this patch in today's HEAD (as cafe8df8b9bc)...

Quoted 6 lines

Hi Robin

Thank for the report. See the discussion on netdev under the subject "Kernel crashes in phy_attach_direct()"

Andrew

There is bug report from Dan Carpenter([email protected]) who ran static analysis to find this issue. Thanks a lot. [bug report] net: phy: Fix lack of reference count on PHY driver