[Xen-devel] dom0 crashing on extreme I/O

[Diwaker Gupta <diwaker.lists@xxxxxxxxx>]

I have 3 VMs, two running webservers and the 3rd running
netperf/iperf. This is a multi-cpu setup, with dom0 on CPU-0 and all
the remaining VMs on a separate CPU.

Currently my dom0 has 528M of memory, while each VM has around 160M.
Under high loads, the system crashes. I'm pasting a representative crash here:

file=grant_table.c, line=729) gnttab_transfer: out-of-range or xen
frame 2f016001
(XEN) (file=grant_table.c, line=729) gnttab_transfer: out-of-range or
xen frame 2f017001
(XEN) (file=grant_table.c, line=729) gnttab_transfer: out-of-range or
xen frame 18fca001
(XEN) (file=grant_table.c, line=729) gnttab_transfer: out-of-range or
xen frame 18fcb001
(XEN) (file=grant_table.c, line=729) gnttab_transfer: out-of-range or
xen frame 2270c001
(XEN) (file=grant_table.c, line=729) gnttab_transfer: out-of-range or
xen frame 2270d001
------------[ cut here ]------------
kernel BUG at drivers/xen/netback/netback.c:335!
invalid operand: 0000 [#1]
Modules linked in: ipt_physdev iptable_filter ip_tables video thermal
processor fan button battery ac md sworks_agp agpgart dm_snapshot
dm_zero dm_mirror ext3 jbd dm_mod mptscsih mptbase sd_mod scsi_mod
CPU:    0
EIP:    0061:[<c02c6782>]    Not tainted VLI
EFLAGS: 00010246   (2.6.12.6-xen0)
EIP is at net_rx_action+0x4c2/0x4f0
eax: 0000fff7   ebx: df26b620   ecx: 00000042   edx: c04b8920
esi: dc073480   edi: 00000000   ebp: c04b3900   esp: c0a23d28
ds: 007b   es: 007b   ss: 0069
Process ksoftirqd/0 (pid: 2, threadinfo=c0a22000 task=c0a16510)
Stack: c04b38e0 c0362d90 80000000 c0363b36 dbad7d80 db6fee80 df26b400 5cb34f36
       00000088 00000000 0003d700 db6ff012 c04b8920 00000106 00a23e2c c05e5000
       00000000 c0363a90 00000001 00000000 00000000 00000001 c0a16510 00000000
Call Trace:
 [<c0362d90>] br_forward_finish+0x0/0x80
 [<c0363b36>] br_handle_frame_finish+0xa6/0x160
 [<c0363a90>] br_handle_frame_finish+0x0/0x160
 [<c01423a5>] kmem_getpages+0x65/0x90
 [<c013ece2>] __rmqueue+0xb2/0xf0
 [<c032302d>] nf_iterate+0x5d/0x90
 [<c0367aa0>] br_nf_pre_routing_finish+0x0/0x420
 [<c0367aa0>] br_nf_pre_routing_finish+0x0/0x420
 [<c032336e>] nf_hook_slow+0x6e/0x120
 [<c0367aa0>] br_nf_pre_routing_finish+0x0/0x420
 [<c0363a90>] br_handle_frame_finish+0x0/0x160
 [<c0368549>] br_nf_pre_routing+0x319/0x4a0
 [<c0367aa0>] br_nf_pre_routing_finish+0x0/0x420
 [<c032302d>] nf_iterate+0x5d/0x90
 [<c0363a90>] br_handle_frame_finish+0x0/0x160
 [<c0363a90>] br_handle_frame_finish+0x0/0x160
 [<c032336e>] nf_hook_slow+0x6e/0x120
 [<c0363a90>] br_handle_frame_finish+0x0/0x160
 [<c0363db3>] br_handle_frame+0x1c3/0x260
 [<c0363a90>] br_handle_frame_finish+0x0/0x160
 [<c03188d3>] netif_receive_skb+0x113/0x230
 [<c02820bf>] tg3_rx+0x2cf/0x490
 [<c027e246>] tg3_restart_ints+0x26/0xa0
 [<c02823a6>] tg3_poll+0x126/0x1a0
 [<c0121660>] ksoftirqd+0x0/0xa0
 [<c0121660>] ksoftirqd+0x0/0xa0
 [<c01214ff>] tasklet_action+0x5f/0xa0
 [<c0121152>] __do_softirq+0x52/0xc0
 [<c0121207>] do_softirq+0x47/0x60
 [<c01216b9>] ksoftirqd+0x59/0xa0
 [<c013079d>] kthread+0xad/0xf0
 [<c01306f0>] kthread+0x0/0xf0
 [<c0106855>] kernel_thread_helper+0x5/0x10
Code: 0f 0b 44 01 38 19 3a c0 90 e9 5a fe ff ff b8 74 64 40 c0 e8 31
ac e5 ff eb 8f c7 04 24 9c 10 3b c0 e8 b3 60 e5 ff 8d 76 00 eb 92 <0f>
0b 4f 01 38 19 3a c0 e9 4c fe ff ff 0f 0b 2a 01 38 19 3a c0
 <0>Kernel panic - not syncing: Fatal exception in interrupt
 (XEN) Domain 0 shutdown: rebooting machine.

NOTE: the line number in netback.c (335) might not be very useful for
reference. I have some additional instrumentation in netback, so the
line number might not match the files in xen-unstable.hg

Will increasing dom0 memory further help? Or increasing the size of the rings?
--
Web/Blog/Gallery: http://floatingsun.net

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel