Re: [PATCH 2/2] xfs: mark the xfs-alloc workqueue as high priority

[Eric Sandeen <sandeen@sandeen.net>]

On 1/12/15 5:37 PM, Tejun Heo wrote:
> Hello, Eric.
> 
> On Mon, Jan 12, 2015 at 05:12:34PM -0600, Eric Sandeen wrote:
>> The only tasks in RU state are the one I mentioned, and swapper/X.
>>
>> crash> ps -m | grep RU
>> [0 00:00:00.159] [RU]  PID: 17056  TASK: c000000111cc0000  CPU: 8   COMMAND: "kworker/u112:1"
>> [4 19:11:44.360] [RU]  PID: 0      TASK: c0000000010416f0  CPU: 0   COMMAND: "swapper/0"
>> [4 19:11:44.192] [RU]  PID: 0      TASK: c000000633280000  CPU: 1   COMMAND: "swapper/1"
>> [4 19:11:44.192] [RU]  PID: 0      TASK: c0000006332c0000  CPU: 2   COMMAND: "swapper/2"
>> ...
>>
>> and I don't see anything interesting in those. :)
> 
> Heh, I don't get it.  Can you please print out worker_pool->nr_running
> of the pool that the hung work item is queued on?

crash> struct worker_pool c000000001cd7b00
struct worker_pool {
  lock = {
    {
      rlock = {
        raw_lock = {
          slock = 0
        }
      }
    }
  }, 
  cpu = 27, 
  node = 0, 
  id = 54, 
  flags = 0, 
  worklist = {
    next = 0xc0000006182ceee0, 
    prev = 0xc000000000fe0470 <release_agent_work+8>
  }, 
  nr_workers = 15, 
  nr_idle = 0, 
...

And that worklist is the one w/ my stuck aio-stress btree split
worker:

crash> list -H 0xc000000001cd7b18 -o work_struct.entry -s work_struct
...
c00000003ad030a0
struct work_struct {
  data = {
    counter = -4611686018397087739
  }, 
  entry = {
    next = 0xc0000006182c5e08, 
    prev = 0xc0000006182c0750
  }, 
  func = 0xd000000005c58ef0 <xfs_btree_split_worker>
}
...

-Eric

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs