[ofa-general] Trouble with IPoIB stack ?

[Bart Van Assche]

Hello,

While running a stress test of Linux' iSCSI initiator I noticed that
the test ran fine over an Ethernet network but that strange memory
allocation problems were reported after I switched from Ethernet to
IPoIB. A few minutes later the whole system froze. Anyone any hints on
how I can pinpoint the cause of this issue ?

>From the kernel messages:

...
[  552.238283] scsi_wq_25: page allocation failure. order:0, mode:0x20
[  551.878249] scsi_wq_39: page allocation failure. order:0, mode:0x20
...
[  553.369853] Call Trace:
[  551.878264] Pid: 7860, comm: scsi_wq_39 Not tainted 2.6.25.6 #3
[  552.958273] Pid: 7761, comm: scsi_wq_36 Not tainted 2.6.25.6 #3
[  553.369866]  [__alloc_pages+0x2e8/0x3d0] __alloc_pages+0x2e8/0x3d0
[  552.418332]  [__alloc_pages+0x2e8/0x3d0] __alloc_pages+0x2e8/0x3d0
[  553.369875]  [ib_ipoib:dev_queue_xmit+0xed/0xb410] dev_queue_xmit+0xed/0x380
[  552.418343]  [new_slab+0x22d/0x2b0] new_slab+0x22d/0x2b0
[  552.418349]  [__slab_alloc+0x1f5/0x4c0] __slab_alloc+0x1f5/0x4c0
[  553.369894]  [new_slab+0x22d/0x2b0] new_slab+0x22d/0x2b0
[  552.778357] Pid: 7306, comm: scsi_wq_23 Not tainted 2.6.25.6 #3
...

See also http://bugzilla.kernel.org/show_bug.cgi?id=10890.

Bart.