Messages in this thread | | | Date | Fri, 8 Sep 2006 09:52:41 -0700 | From | Andrew Morton <> | Subject | Re: [Problem] System hang when I run pounder and syscall test on kernel 2.6.18-rc5 |
| |
On Fri, 8 Sep 2006 19:36:40 +0800 Shu Qing Yang <yangshuq@cn.ibm.com> wrote:
> Andrew Morton <akpm@osdl.org> wrote on 2006-09-08 10:14:34: > > > On Thu, 7 Sep 2006 12:35:09 +0800 > > Shu Qing Yang <yangshuq@cn.ibm.com> wrote: > > > > > Problem description: > > > I run pounder, scsi_debug on a machine. Then start 200 random > syscall > > > test > > > simultaneously. Tens of minutes later, the system hang. > > > > What is "pounder" and from where can it be obtained? > > > Thanks for your reply. > > Pounder is part of ltp and locate in LTPROOT/testcases/pounder21. > It is a suit of test cases including mem_alloc, random_syscall, bonnie++, > etc.
OK, thanks.
> > Running two tests at the same time complicates things. The next step > > should be to determine whether it is reproducible. If it is, then see > if > > it is reproducible with just one test running (presumably pounder?) > > > Running multiple cases simultaneously is to stress kernel more. And > because of > lack of machine resource I have no chance to reproduce it. > > > It would be helpful to provide sufficient information to give others a > > chance of reproducing it: amount of memory, method for configuring the > > scsi-debug "disks", method for invoking pounder, etc. > > > The machine belongs to IBM p-Series with power5+ cpu and 2GB memory. > Run LTPROOT/testscript/ltp-scsi_debug.sh and > LTPROOT/testscript/pounder21/pounder directly. > No extra parameters. The command to load scsi_debug module is: > modprobe scsi_debug max_luns=2 num_tgts=2 add_host=2 dev_size_mb=20 > > ... > > I can not excute sysrq command now. But I can get memory allocation > information from xmon, > which indicates your guess may be right. > > 1:mon> mi > Mem-info: > DMA per-cpu: > cpu 0 hot: high 6, batch 1 used:5 > cpu 0 cold: high 2, batch 1 used:1 > cpu 1 hot: high 6, batch 1 used:5 > cpu 1 cold: high 2, batch 1 used:1 > cpu 2 hot: high 6, batch 1 used:5 > cpu 2 cold: high 2, batch 1 used:1 > cpu 3 hot: high 6, batch 1 used:3 > cpu 3 cold: high 2, batch 1 used:1 > cpu 4 hot: high 6, batch 1 used:5 > cpu 4 cold: high 2, batch 1 used:1 > cpu 5 hot: high 6, batch 1 used:4 > cpu 5 cold: high 2, batch 1 used:0 > DMA32 per-cpu: empty > Normal per-cpu: empty > HighMem per-cpu: empty > Free pages: 6976kB (0kB HighMem) > Active:6141 inactive:11012 dirty:4742 writeback:0 unstable:0 free:109 > slab:11925 mapped:7 pagetables:7061 > DMA free:6976kB min:5760kB low:7168kB high:8640kB active:393024kB > inactive:704768kB present:2097152kB pages_scanned:5172 all_unreclaimable? > no > lowmem_reserve[]: 0 0 0 0 > DMA32 free:0kB min:0kB low:0kB high:0kB active:0kB inactive:0kB > present:0kB pages_scanned:0 all_unreclaimable? no > lowmem_reserve[]: 0 0 0 0 > Normal free:0kB min:0kB low:0kB high:0kB active:0kB inactive:0kB > present:0kB pages_scanned:0 all_unreclaimable? no > lowmem_reserve[]: 0 0 0 0 > HighMem free:0kB min:2048kB low:2048kB high:2048kB active:0kB inactive:0kB > present:0kB pages_scanned:0 all_unreclaimable? no > lowmem_reserve[]: 0 0 0 0 > DMA: 19*64kB 1*128kB 2*256kB 0*512kB 1*1024kB 0*2048kB 1*4096kB 0*8192kB > 0*16384kB = 6976kB > DMA32: empty > Normal: empty > HighMem: empty > Swap cache: add 439156, delete 439156, find 50391/101032, race 26+79 > Free swap = 0kB > Total swap = 855552kB > Free swap: 0kB > 32768 pages of RAM > 408 reserved pages > 6834 pages shared > 0 pages swap cached
So we ran out of memory and we ran out of swap.
Possibly what has happened here is that the machine is doing a huge amount of work scanning pages and pretty soon it will enter the oom-killer to kill some userspace process. But before that happened, the softlockup detector triggered.
But the machine _should_ have recovered. If it hung for more than a few seconds then that's bad behaviour. If it hung for more than a few minutes then that should be considered a bug. If it hung for ever then that's definitely a bug.
Do you recall approximately how long the machine spent in this state? - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
| |