Re: perf on 2.6.38-rc4 wedges my box

[Peter Zijlstra]

On Wed, 2011-02-09 at 12:38 -0500, Jeff Moyer wrote:
> Hi,
> 
> I'm trying out willy's ata_ram driver[1], and in so doing have managed to
> wedge my box while using perf record on an aio-stress run:
> 
> [root@metallica ~]# modprobe ata_ram capacity=2097152 preallocate=1
> [root@metallica ~]# ./aio-stress -O -o 0 -r 4 -d 32 -b 16 /dev/sds
> adding stage write
> starting with write
> file size 1024MB, record size 4KB, depth 32, ios per iteration 8
> max io_submit 16, buffer alignment set to 4KB
> threads 1 files 1 contexts 1 context offset 2MB verification off
> adding file /dev/sds thread 0
> write on /dev/sds (621.30 MB/s) 1024.00 MB in 1.65s
> thread 0 write totals (621.27 MB/s) 1024.00 MB in 1.65s
> [root@metallica ~]# perf record -- ./aio-stress -O -o 0 -r 4 -d 32 -b 16
> /dev/sds
> adding stage write
> starting with write
> file size 1024MB, record size 4KB, depth 32, ios per iteration 8
> max io_submit 16, buffer alignment set to 4KB
> threads 1 files 1 contexts 1 context offset 2MB verification off
> adding file /dev/sds thread 0
> 
> and there it sits.  On the console, I see:
> 
> NOHZ: local_softirq_pending 100
> NOHZ: local_softirq_pending 100
> NOHZ: local_softirq_pending 100
> NOHZ: local_softirq_pending 100
> NOHZ: local_softirq_pending 100
> 
> The number of messages varies, but this is the most I've seen (it
> doesn't keep repeating).  At this point, the machine does not respond to
> pings.  As I don't have physical access at the moment, I can't try
> alt-sysrq, but might be able to do that tomorrow.  It's probably worth
> noting that I've witnessed similar behavior with real devices, so it's
> not just the ata_ram driver.
> 
> Any ideas on how to track this down?

So I took linus' tree of about half an hour ago, added
git://git.kernel.org/pub/scm/linux/kernel/git/willy/misc.git ata-ram
(fixed up some Kconfig/Makefile rejects), google'd aio-stress
(http://fsbench.filesystems.org/bench/aio-stress.c) and set out to
reproduce the above..

Sadly it all seems to work here, its spending ~15% in
_raw_spin_lock_irq, which when I use -g looks to break down like:

-     14.13%  aio-stress  [kernel.kallsyms]  [k] _raw_spin_lock_irq 
   - _raw_spin_lock_irq                                             
      + 44.14% __make_request                                       
      + 20.91% __aio_get_req                                        
      + 10.15% aio_run_iocb                                         
      + 7.37% do_io_submit                                          
      + 6.55% scsi_request_fn                                       
      + 5.48% generic_unplug_device                                 
      + 3.58% aio_put_req                                           
      + 0.92% generic_make_request                                  
      + 0.91% __generic_unplug_device



--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/