Call Trace: page allocation failure - is it normal behaviour?
Hello,
I noticed some call trace when testing box under heavy load. To create a load following jobs have been running simultaneously.
ab2 -c 200 -n 10000000 http://192.168.114.239/
fsx-linux -l 900000000 fsx-data3
dbench 100
adt root # w
19:24:32 up 14:58, 6 users, load average: 90.92, 83.97, 84.28
Some times after dmesg has shown multiple call traces of two types:
swapper: page allocation failure. order:2, mode:0x20
Call Trace: [<c014059c>] __alloc_pages+0x30c/0x350 [<c0140605>] __get_free_pages+0x25/0x40 [<c01435a7>] cache_grow+0xc7/0x310 [<c01438fe>] cache_alloc_refill+0x10e/0x2c0 [<c0143e01>] __kmalloc+0x71/0x80 [<c0266697>] alloc_skb+0x47/0xe0 [<c0294e7e>] tcp_fragment+0x5e/0x340 [<c02975b8>] tcp_write_wakeup+0xe8/0x280 [<c0298870>] tcp_write_timer+0x0/0x130 [<c029776d>] tcp_send_probe0+0x1d/0x110 [<c0298933>] tcp_write_timer+0xc3/0x130 [<c01298f7>] run_timer_softirq+0xe7/0x1d0 [<c0124e2a>] do_softirq+0xca/0xd0 [<c01162e7>] smp_apic_timer_interrupt+0xd7/0x150 [<c0106d90>] default_idle+0x0/0x40 [<c0109dea>] apic_timer_interrupt+0x1a/0x20 [<c0106d90>] default_idle+0x0/0x40 [<c0106dbd>] default_idle+0x2d/0x40 [<c0106e47>] cpu_idle+0x37/0x40 [<c012111d>] printk+0x17d/0x1d0
ab2: page allocation failure. order:2, mode:0x20
Call Trace: [<c014059c>] __alloc_pages+0x30c/0x350 [<c0140605>] __get_free_pages+0x25/0x40 [<c01435a7>] cache_grow+0xc7/0x310 [<c01438fe>] cache_alloc_refill+0x10e/0x2c0 [<c0143e01>] __kmalloc+0x71/0x80 [<c0266697>] alloc_skb+0x47/0xe0 [<c0294e7e>] tcp_fragment+0x5e/0x340 [<c0294680>] tcp_transmit_skb+0x3e0/0x600 [<c02975b8>] tcp_write_wakeup+0xe8/0x280 [<c0298870>] tcp_write_timer+0x0/0x130 [<c029776d>] tcp_send_probe0+0x1d/0x110 [<c0287d45>] __tcp_mem_reclaim+0x25/0x60 [<c0298933>] tcp_write_timer+0xc3/0x130 [<c01298f7>] run_timer_softirq+0xe7/0x1d0 [<c0124e2a>] do_softirq+0xca/0xd0 [<c01162e7>] smp_apic_timer_interrupt+0xd7/0x150 [<c0109dea>] apic_timer_interrupt+0x1a/0x20 [<c0289fc5>] tcp_sendmsg+0x955/0x12b0 [<c02acc2d>] inet_sendmsg+0x4d/0x60 [<c0262a5d>] sock_aio_write+0xbd/0xe0 [<c015b02b>] do_sync_write+0x8b/0xc0 [<c0140613>] __get_free_pages+0x33/0x40 [<c0287dd4>] tcp_poll+0x34/0x190 [<c0262ff9>] sock_poll+0x29/0x40 [<c015b15e>] vfs_write+0xfe/0x130 [<c015b242>] sys_write+0x42/0x70 [<c01093fb>] syscall_call+0x7/0xb
bwt I've noticed no visible harm to system and question ruther is whether this behaviour is normal under such circumstances?
dmesg(full): http://sysadminday.org.ru/call_trace/dmesg .config: http://sysadminday.org.ru/call_trace/config /proc/meminfo: http://sysadminday.org.ru/call_trace/meminfo /proc/cpuinfo: http://sysadminday.org.ru/call_trace/cpuinfo lspsi: http://sysadminday.org.ru/call_trace/lspci lspci -vvn: http://sysadminday.org.ru/call_trace/lspci_-vvn
Signature · 8 lines
Best regards.
Alexander Y. Fomichev <[email protected]>
Public PGP key: http://sysadminday.org.ru/gluk.asc
--
Best regards.
Alexander Y. Fomichev <[email protected]>
Public PGP key: http://sysadminday.org.ru/gluk.ascQuoted 31 lines · Alexander Y. Fomichev
Hello,
I noticed some call trace when testing box under heavy load. To create a load following jobs have been running simultaneously.
ab2 -c 200 -n 10000000 http://192.168.114.239/
fsx-linux -l 900000000 fsx-data3
dbench 100adt root # w
19:24:32 up 14:58, 6 users, load average: 90.92, 83.97, 84.28Some times after dmesg has shown multiple call traces of two types:
swapper: page allocation failure. order:2, mode:0x20
Call Trace: [<c014059c>] __alloc_pages+0x30c/0x350 [<c0140605>] __get_free_pages+0x25/0x40 [<c01435a7>] cache_grow+0xc7/0x310 [<c01438fe>] cache_alloc_refill+0x10e/0x2c0 [<c0143e01>] __kmalloc+0x71/0x80 [<c0266697>] alloc_skb+0x47/0xe0 [<c0294e7e>] tcp_fragment+0x5e/0x340 [<c02975b8>] tcp_write_wakeup+0xe8/0x280 [<c0298870>] tcp_write_timer+0x0/0x130 [<c029776d>] tcp_send_probe0+0x1d/0x110 [<c0298933>] tcp_write_timer+0xc3/0x130 [<c01298f7>] run_timer_softirq+0xe7/0x1d0 [<c0124e2a>] do_softirq+0xca/0xd0
Quoted 3 lines
...
bwt I've noticed no visible harm to system and question ruther is whether this behaviour is normal under such circumstances?
Yes, it is expected and the networking stack will recover. We'll remove that debug code at some point.