Adding vlan to DSA port causes lockdep splat

Adding vlan to DSA port causes lockdep splat

[Maxim Uvarov]

Is there any progress on subj issue?

I see it was investigated here:
https://www.spinics.net/lists/netdev/msg361434.html

But it still exist on later kernels:

[   37.320301] ip/1047 is trying to acquire lock:
[   37.324764]  (_xmit_ETHER/1){+.....}, at: [<c06ad228>] dev_mc_sync+0x4c/0x88
[   37.331882]
[   37.331882] but task is already holding lock:
[   37.337738]  (_xmit_ETHER/1){+.....}, at: [<c06ad228>] dev_mc_sync+0x4c/0x88
[   37.344828]
[   37.344828] other info that might help us debug this:
[   37.351384]  Possible unsafe locking scenario:
[   37.351384]
[   37.357326]        CPU0
[   37.359778]        ----
[   37.362230]   lock(_xmit_ETHER/1);
[   37.365650]   lock(_xmit_ETHER/1);
[   37.369069]
[   37.369069]  *** DEADLOCK ***
[   37.369069]
[   37.375013]  May be due to missing lock nesting notation
[   37.375013]
[   37.381830] 3 locks held by ip/1047:
[   37.385416]  #0:  (rtnl_mutex){+.+.+.}, at: [<c06b8248>]
rtnetlink_rcv+0x1c/0x38
[   37.392860]  #1:  (&vlan_netdev_addr_lock_key/1){+.....}, at:
[<c06a4f94>] dev_set_rx_mode+0x1c/0x30
[   37.402046]  #2:  (_xmit_ETHER/1){+.....}, at: [<c06ad228>]
dev_mc_sync+0x4c/0x88
[   37.409574]
[   37.409574] stack backtrace:
[   37.413952] CPU: 0 PID: 1047 Comm: ip Not tainted
4.10.0maxdebug-00008-g9d55486 #22
[   37.421639] Hardware name: Generic AM33XX (Flattened Device Tree)
[   37.427756] Backtrace:
[   37.430237] [<c010bf78>] (dump_backtrace) from [<c010c220>]
(show_stack+0x18/0x1c)
[   37.437842]  r7:c140f7ec r6:c13e29e0 r5:dc316780 r4:c0db8408
[   37.443537] [<c010c208>] (show_stack) from [<c0402ab0>]
(dump_stack+0x20/0x28)
[   37.450800] [<c0402a90>] (dump_stack) from [<c016b294>]
(__lock_acquire+0x15d4/0x18ec)
[   37.458753] [<c0169cc0>] (__lock_acquire) from [<c016b958>]
(lock_acquire+0x74/0x94)
[   37.466532]  r10:dc2260c0 r9:dd747910 r8:00000000 r7:00000001
r6:00000001 r5:600d0013
[   37.474393]  r4:00000000
[   37.476946] [<c016b8e4>] (lock_acquire) from [<c07f802c>]
(_raw_spin_lock_nested+0x44/0x54)
[   37.485334]  r7:00001002 r6:dd4fd988 r5:dc1f3000 r4:dd4fd988
[   37.491022] [<c07f7fe8>] (_raw_spin_lock_nested) from [<c06ad228>]
(dev_mc_sync+0x4c/0x88)
[   37.499320]  r4:dd4fd800
[   37.501870] [<c06ad1dc>] (dev_mc_sync) from [<c0796b60>]
(dsa_slave_set_rx_mode+0x28/0x38)
[   37.510171]  r7:00001002 r6:00000000 r5:dd4fd800 r4:dc1f3000
[   37.515856] [<c0796b38>] (dsa_slave_set_rx_mode) from [<c06a4f40>]
(__dev_set_rx_mode+0x64/0x9c)

-- 
Best regards,
Maxim Uvarov

Re: Adding vlan to DSA port causes lockdep splat

[Andrew Lunn]

On Tue, Feb 21, 2017 at 11:38:24AM +0300, Maxim Uvarov wrote:
> Is there any progress on subj issue?
> 
> I see it was investigated here:
> https://www.spinics.net/lists/netdev/msg361434.html

Hi Maxim

Nobody made further investigations. And since i don't see it in my
tests, i've never been annoyed enough to go try find a fix.

       Andrew

Re: Adding vlan to DSA port causes lockdep splat

[Florian Fainelli]

On 02/21/2017 12:38 AM, Maxim Uvarov wrote:
> Is there any progress on subj issue?
> 
> I see it was investigated here:
> https://www.spinics.net/lists/netdev/msg361434.html

I was just able to reproduce this and will try to look into it. At first
glance, the reason why we usually don't see it, is that VLAN
configuration goes through bridge vlan, which exercises a slightly
different code path.

FWIW: using bridge vlan (VLAN filtering) allows you to define your
switch's default VLAN (ID and tag/untag flag) whereas the traditional
vlan device does not really allow you to do that.

> 
> But it still exist on later kernels:
> 
> [   37.320301] ip/1047 is trying to acquire lock:
> [   37.324764]  (_xmit_ETHER/1){+.....}, at: [<c06ad228>] dev_mc_sync+0x4c/0x88
> [   37.331882]
> [   37.331882] but task is already holding lock:
> [   37.337738]  (_xmit_ETHER/1){+.....}, at: [<c06ad228>] dev_mc_sync+0x4c/0x88
> [   37.344828]
> [   37.344828] other info that might help us debug this:
> [   37.351384]  Possible unsafe locking scenario:
> [   37.351384]
> [   37.357326]        CPU0
> [   37.359778]        ----
> [   37.362230]   lock(_xmit_ETHER/1);
> [   37.365650]   lock(_xmit_ETHER/1);
> [   37.369069]
> [   37.369069]  *** DEADLOCK ***
> [   37.369069]
> [   37.375013]  May be due to missing lock nesting notation
> [   37.375013]
> [   37.381830] 3 locks held by ip/1047:
> [   37.385416]  #0:  (rtnl_mutex){+.+.+.}, at: [<c06b8248>]
> rtnetlink_rcv+0x1c/0x38
> [   37.392860]  #1:  (&vlan_netdev_addr_lock_key/1){+.....}, at:
> [<c06a4f94>] dev_set_rx_mode+0x1c/0x30
> [   37.402046]  #2:  (_xmit_ETHER/1){+.....}, at: [<c06ad228>]
> dev_mc_sync+0x4c/0x88
> [   37.409574]
> [   37.409574] stack backtrace:
> [   37.413952] CPU: 0 PID: 1047 Comm: ip Not tainted
> 4.10.0maxdebug-00008-g9d55486 #22
> [   37.421639] Hardware name: Generic AM33XX (Flattened Device Tree)
> [   37.427756] Backtrace:
> [   37.430237] [<c010bf78>] (dump_backtrace) from [<c010c220>]
> (show_stack+0x18/0x1c)
> [   37.437842]  r7:c140f7ec r6:c13e29e0 r5:dc316780 r4:c0db8408
> [   37.443537] [<c010c208>] (show_stack) from [<c0402ab0>]
> (dump_stack+0x20/0x28)
> [   37.450800] [<c0402a90>] (dump_stack) from [<c016b294>]
> (__lock_acquire+0x15d4/0x18ec)
> [   37.458753] [<c0169cc0>] (__lock_acquire) from [<c016b958>]
> (lock_acquire+0x74/0x94)
> [   37.466532]  r10:dc2260c0 r9:dd747910 r8:00000000 r7:00000001
> r6:00000001 r5:600d0013
> [   37.474393]  r4:00000000
> [   37.476946] [<c016b8e4>] (lock_acquire) from [<c07f802c>]
> (_raw_spin_lock_nested+0x44/0x54)
> [   37.485334]  r7:00001002 r6:dd4fd988 r5:dc1f3000 r4:dd4fd988
> [   37.491022] [<c07f7fe8>] (_raw_spin_lock_nested) from [<c06ad228>]
> (dev_mc_sync+0x4c/0x88)
> [   37.499320]  r4:dd4fd800
> [   37.501870] [<c06ad1dc>] (dev_mc_sync) from [<c0796b60>]
> (dsa_slave_set_rx_mode+0x28/0x38)
> [   37.510171]  r7:00001002 r6:00000000 r5:dd4fd800 r4:dc1f3000
> [   37.515856] [<c0796b38>] (dsa_slave_set_rx_mode) from [<c06a4f40>]
> (__dev_set_rx_mode+0x64/0x9c)
> 


-- 
Florian

Re: Adding vlan to DSA port causes lockdep splat

[Russell King - ARM Linux]

On Wed, Mar 29, 2017 at 09:45:16AM -0700, Florian Fainelli wrote:
> On 02/21/2017 12:38 AM, Maxim Uvarov wrote:
> > Is there any progress on subj issue?

I'm afraid the report got completely ignored.  Hope you have better luck.

> > I see it was investigated here:
> > https://www.spinics.net/lists/netdev/msg361434.html
> 
> I was just able to reproduce this and will try to look into it. At first
> glance, the reason why we usually don't see it, is that VLAN
> configuration goes through bridge vlan, which exercises a slightly
> different code path.
> 
> FWIW: using bridge vlan (VLAN filtering) allows you to define your
> switch's default VLAN (ID and tag/untag flag) whereas the traditional
> vlan device does not really allow you to do that.
> 
> > 
> > But it still exist on later kernels:
> > 
> > [   37.320301] ip/1047 is trying to acquire lock:
> > [   37.324764]  (_xmit_ETHER/1){+.....}, at: [<c06ad228>] dev_mc_sync+0x4c/0x88
> > [   37.331882]
> > [   37.331882] but task is already holding lock:
> > [   37.337738]  (_xmit_ETHER/1){+.....}, at: [<c06ad228>] dev_mc_sync+0x4c/0x88
> > [   37.344828]
> > [   37.344828] other info that might help us debug this:
> > [   37.351384]  Possible unsafe locking scenario:
> > [   37.351384]
> > [   37.357326]        CPU0
> > [   37.359778]        ----
> > [   37.362230]   lock(_xmit_ETHER/1);
> > [   37.365650]   lock(_xmit_ETHER/1);
> > [   37.369069]
> > [   37.369069]  *** DEADLOCK ***
> > [   37.369069]
> > [   37.375013]  May be due to missing lock nesting notation
> > [   37.375013]
> > [   37.381830] 3 locks held by ip/1047:
> > [   37.385416]  #0:  (rtnl_mutex){+.+.+.}, at: [<c06b8248>]
> > rtnetlink_rcv+0x1c/0x38
> > [   37.392860]  #1:  (&vlan_netdev_addr_lock_key/1){+.....}, at:
> > [<c06a4f94>] dev_set_rx_mode+0x1c/0x30
> > [   37.402046]  #2:  (_xmit_ETHER/1){+.....}, at: [<c06ad228>]
> > dev_mc_sync+0x4c/0x88
> > [   37.409574]
> > [   37.409574] stack backtrace:
> > [   37.413952] CPU: 0 PID: 1047 Comm: ip Not tainted
> > 4.10.0maxdebug-00008-g9d55486 #22
> > [   37.421639] Hardware name: Generic AM33XX (Flattened Device Tree)
> > [   37.427756] Backtrace:
> > [   37.430237] [<c010bf78>] (dump_backtrace) from [<c010c220>]
> > (show_stack+0x18/0x1c)
> > [   37.437842]  r7:c140f7ec r6:c13e29e0 r5:dc316780 r4:c0db8408
> > [   37.443537] [<c010c208>] (show_stack) from [<c0402ab0>]
> > (dump_stack+0x20/0x28)
> > [   37.450800] [<c0402a90>] (dump_stack) from [<c016b294>]
> > (__lock_acquire+0x15d4/0x18ec)
> > [   37.458753] [<c0169cc0>] (__lock_acquire) from [<c016b958>]
> > (lock_acquire+0x74/0x94)
> > [   37.466532]  r10:dc2260c0 r9:dd747910 r8:00000000 r7:00000001
> > r6:00000001 r5:600d0013
> > [   37.474393]  r4:00000000
> > [   37.476946] [<c016b8e4>] (lock_acquire) from [<c07f802c>]
> > (_raw_spin_lock_nested+0x44/0x54)
> > [   37.485334]  r7:00001002 r6:dd4fd988 r5:dc1f3000 r4:dd4fd988
> > [   37.491022] [<c07f7fe8>] (_raw_spin_lock_nested) from [<c06ad228>]
> > (dev_mc_sync+0x4c/0x88)
> > [   37.499320]  r4:dd4fd800
> > [   37.501870] [<c06ad1dc>] (dev_mc_sync) from [<c0796b60>]
> > (dsa_slave_set_rx_mode+0x28/0x38)
> > [   37.510171]  r7:00001002 r6:00000000 r5:dd4fd800 r4:dc1f3000
> > [   37.515856] [<c0796b38>] (dsa_slave_set_rx_mode) from [<c06a4f40>]
> > (__dev_set_rx_mode+0x64/0x9c)
> > 
> 
> 
> -- 
> Florian

-- 
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.