RHEL 5.7 appears to hangs when vm.dirty_bytes and vm.dirty_ratio are lowered and a high I/O is applied.

Issue

• When vm.dirty_bytes and vm.dirty_ratio are lowered and a moderate to heavy IO operation is done, the system appears to hang e.g:

  vm.dirty_bytes = 512000
  vm.dirty_ratio = 0
  

• Incomplete vmcore is generated by nfs in kdump.

• Following type of trace can be observed,

  "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
  ls            D ffff81000503eaa0     0  2937   2896                     (NOTLB)
   ffff810008d17d98 0000000000000086 ffff81000c146620 ffff81000c146218
   ffff81000898a460 0000000000000007 ffff81000be53860 ffff81000fc36100
   0000012fb24ffad2 00000000000023c9 ffff81000be53a48 0000000108d17e48
  Call Trace:
   [<ffffffff8000a851>] __link_path_walk+0xfa8/0xfd1
   [<ffffffff80063c53>] __mutex_lock_slowpath+0x60/0x9b
   [<ffffffff80063c9d>] .text.lock.mutex+0xf/0x14
   [<ffffffff886bf787>] :nfs:nfs_getattr+0x45/0xd9
   [<ffffffff8003efd6>] vfs_lstat_fd+0x2f/0x47
   [<ffffffff8002ad86>] sys_newlstat+0x19/0x31
   [<ffffffff80038a15>] sys_getdents+0xaf/0xbd
   [<ffffffff8005dde9>] error_exit+0x0/0x84
   [<ffffffff8005d116>] system_call+0x7e/0x83