bonding: make slave status notifications GFP_ATOMIC
authorVeaceslav Falico <vfalico@redhat.com>
Thu, 6 Mar 2014 14:33:22 +0000 (15:33 +0100)
committerDavid S. Miller <davem@davemloft.net>
Thu, 6 Mar 2014 20:19:43 +0000 (15:19 -0500)
commit072256d1f2b8ba0bbb265d590c703f3d57a39d6a
tree0eef5201115e3d8487a7b9a4e1fac6fa1c8183ed
parente90c14835ba2e1d3f82b8f24e429b971725afff5
bonding: make slave status notifications GFP_ATOMIC

Currently we're using GFP_KERNEL, however there are some path(s) where we
can hold some spinlocks, specifically bond->curr_slave_lock:

[    4.722916] BUG: sleeping function called from invalid context at mm/slub.c:965
[    4.724438] in_atomic(): 1, irqs_disabled(): 0, pid: 940, name: ifup-eth
[    4.726034] 5 locks held by ifup-eth/940:
...snip...
[    4.734646]  #4:  (&bond->curr_slave_lock){+...+.}, at: [<ffffffffa00badc6>] bond_enslave+0xda6/0xdd0 [bonding]
...snip...
[    4.759081]  [<ffffffffa00b6f11>] bond_change_active_slave+0x191/0x3b0 [bonding]
[    4.760917]  [<ffffffffa00b7227>] bond_select_active_slave+0xf7/0x1d0 [bonding]
[    4.762751]  [<ffffffffa00badce>] bond_enslave+0xdae/0xdd0 [bonding]
...snip...

As it's out of hot path and is a really rare event - change the gfp_t flags
to GFP_ATOMIC to avoid sleeping under spinlock.

v2: convert new notify calls to GFP_ATOMIC.

CC: Thomas Glanzmann <thomas@glanzmann.de>
CC: Ding Tianhong <dingtianhong@huawei.com>
CC: Jay Vosburgh <fubar@us.ibm.com>
CC: Andy Gospodarek <andy@greyhouse.net>
Signed-off-by: Veaceslav Falico <vfalico@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/bonding/bonding.h