Armada XP (MV78460): BUG in netdevice.h with maxcpus=2

[Stefan Roese]

Hi Thomas,

I'm currently trying to boot latest Linux on the DB-MV784MP-GP
Marvell board. This works just fine for the normal
configuration. But when I try to boot with only 2 CPUs enabled
(maxcpus=2), mounting the rootfs via NFS crashes with this
log:

[   16.408285] ------------[ cut here ]------------
[   16.412915] kernel BUG at include/linux/netdevice.h:507!
[   16.418239] Internal error: Oops - BUG: 0 [#1] SMP ARM
[   16.423390] Modules linked in:
[   16.426464] CPU: 0 PID: 1424 Comm: systemd-udevd Not tainted 4.4.0-rc8-00045-gcbf389a #37
[   16.434660] Hardware name: Marvell Armada 370/XP (Device Tree)
[   16.440507] task: ee9d3100 ti: ee164000 task.ti: ee164000
[   16.445927] PC is at mvneta_percpu_notifier+0x358/0x35c
[   16.451167] LR is at mvneta_percpu_notifier+0x27c/0x35c
[   16.456405] pc : [<c0310d60>]    lr : [<c0310c84>]    psr: 60010013
[   16.456405] sp : ee165e18  ip : 00000003  fp : c077c494
[   16.467914] r10: c077cd58  r9 : 00000002  r8 : c077ce08
[   16.473154] r7 : eea8a520  r6 : ff7ea400  r5 : 00000004  r4 : 00000000
[   16.479697] r3 : 00000000  r2 : 00000008  r1 : 00000004  r0 : 00000004
[   16.486243] Flags: nZCv  IRQs on  FIQs on  Mode SVC_32  ISA ARM  Segment none
[   16.493402] Control: 10c5387d  Table: 2e16806a  DAC: 00000051
[   16.499168] Process systemd-udevd (pid: 1424, stack limit = 0xee164220)
[   16.505801] Stack: (0xee165e18 to 0xee166000)
[   16.510170] 5e00:                                                       c07b89c0 eea8a500
[   16.518373] 5e20: 00000002 ffffffdb 00000002 00000002 00000000 c0782aa0 00000000 ee82a5c0
[   16.526575] 5e40: 00000000 c003fb44 00000000 00000000 c077c494 00000000 00000001 c0024f68
[   16.534774] 5e60: 00000000 00000000 00000002 c0025574 ee165e7c 00000022 00000000 00000028
[   16.542977] 5e80: 00000002 eefca010 eefca044 ee165f88 edc61240 b6f07000 00000000 c00255fc
[   16.551178] 5ea0: c02907fc c028c520 eefca010 00000001 00000001 c028c5ac 00000000 01eada00
[   16.559381] 5ec0: 00000001 edeada00 edc6124c c01324f0 00000000 00000000 00000800 ee9c6f40
[   16.567582] 5ee0: c0132438 b6f07000 ee165f88 c000f7c4 ee164000 00000000 00039388 c00d6bd0
[   16.575782] 5f00: 00000001 00000000 0002b2cc c00092f4 00001000 00001000 b6f07000 00000003
[   16.583981] 5f20: 00000000 edc22580 00000001 c077ce2c 00000073 c00c3994 000b6f07 00000000
[   16.592180] 5f40: beef1270 ffffffff 48400000 ee9c6f40 00000001 b6f07000 ee165f88 c000f7c4
[   16.600385] 5f60: ee164000 c00d7414 0002b2cc c00b3a88 ee9c6f40 ee9c6f40 b6f07000 00000001
[   16.608588] 5f80: c000f7c4 c00d7c38 00000000 00000000 48400000 00000001 b6f07000 00046c70
[   16.616797] 5fa0: 00000004 c000f600 00000001 b6f07000 00000005 b6f07000 00000001 00000000
[   16.625003] 5fc0: 00000001 b6f07000 00046c70 00000004 00000001 00000000 00039860 00039388
[   16.633208] 5fe0: 00000000 beef181c b6d6bf81 b6da5066 400d0030 00000005 00180011 000022be
[   16.641431] [<c0310d60>] (mvneta_percpu_notifier) from [<c003fb44>] (notifier_call_chain+0x44/0x84)
[   16.650513] [<c003fb44>] (notifier_call_chain) from [<c0024f68>] (cpu_notify+0x24/0x40)
[   16.658544] [<c0024f68>] (cpu_notify) from [<c0025574>] (_cpu_up+0x14c/0x164)
[   16.665700] [<c0025574>] (_cpu_up) from [<c00255fc>] (cpu_up+0x70/0x94)
[   16.672333] [<c00255fc>] (cpu_up) from [<c028c520>] (device_online+0x64/0x88)
[   16.679490] [<c028c520>] (device_online) from [<c028c5ac>] (online_store+0x68/0x74)
[   16.687174] [<c028c5ac>] (online_store) from [<c01324f0>] (kernfs_fop_write+0xb8/0x1b4)
[   16.695206] [<c01324f0>] (kernfs_fop_write) from [<c00d6bd0>] (__vfs_write+0x1c/0xd8)
[   16.703061] [<c00d6bd0>] (__vfs_write) from [<c00d7414>] (vfs_write+0x90/0x16c)
[   16.710391] [<c00d7414>] (vfs_write) from [<c00d7c38>] (SyS_write+0x44/0x9c)
[   16.717464] [<c00d7c38>] (SyS_write) from [<c000f600>] (ret_fast_syscall+0x0/0x3c)
[   16.725057] Code: 3afffff5 e3a00001 e28dd00c e8bd8ff0 (e7f001f2) 
[   16.731167] ---[ end trace 1451bd9f24838b0f ]---


Again, this only happen, if I boot the system with 2 CPUs online.
Without this kernel commandline arg (and all 4 CPUs online),
mounting the NFS rootfs works just fine.

Is this a known problem? Do you have any quick hints how to fix
this issue?

Thanks,
Stefan