Description of problem:

During our internal tests of HP Ibrix file system and NFS, we found the Linux kernel lock contention for dcache_lock under low memory pressure.
I can submit the system crashdump, if needed - let me know where you want me to upload it.
The symptoms are:

  SYSTEM MAP: ./System.map-2.6.18-194.el5                              
DEBUG KERNEL: ./vmlinux (2.6.18-194.el5)
    DUMPFILE: ./vmcore
        CPUS: 16
        DATE: Mon Aug 29 10:49:51 2011
      UPTIME: 03:04:02
LOAD AVERAGE: 25.35, 17.47, 11.32
       TASKS: 854
    NODENAME: ib48-123
     RELEASE: 2.6.18-194.el5
     VERSION: #1 SMP Tue Mar 16 21:52:39 EDT 2010
     MACHINE: x86_64  (2665 Mhz)
      MEMORY: 47.3 GB
       PANIC: "Kernel panic - not syncing: hung_task: blocked tasks"
         PID: 903
     COMMAND: "khungtaskd"
        TASK: ffff810c1fb6d040  [THREAD_INFO: ffff81061f0fe000]
         CPU: 8
       STATE: TASK_RUNNING (PANIC)

INFO: task nfsd:11087 blocked for more than 120 seconds.
"echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
nfsd          D ffffffff80151248     0 11087      1         11098 11094 (L-TLB)
 ffff810609359900 0000000000000046 ffff810bd1398bc0 ffff81061b420000
 0000000000000000 000000000000000a ffff81060e4080c0 ffff810c1fc36820
 00000a138f8ff7bc 00000000000157eb ffff81060e4082a8 0000000ebb4eb380
Call Trace:
 [<ffffffff800322bb>] ip_output+0x29a/0x2dd
 [<ffffffff80064c6f>] __mutex_lock_slowpath+0x60/0x9b
 [<ffffffff888eff12>] :ibrix:unreference_internal+0x142/0x280
 [<ffffffff80064cb9>] .text.lock.mutex+0xf/0x14
 [<ffffffff888f0095>] :ibrix:unreference+0x45/0x60
 [<ffffffff888d74c4>] :ibrix:setInodeWithDelegInContextByIndexEx+0x7e4/0x1450
 [<ffffffff88889253>] :ibrix:put_delegation+0x643/0x6d0
 [<ffffffff888d8d61>] :ibrix:setInodeWithDelegInContextByIndex+0x61/0xa0
 [<ffffffff88916c80>] :ibrix:ibrix_aio_write_internal+0x330/0x1c40
 [<ffffffff888802ce>] :ibrix:get_ibr_time+0xe/0x30
 [<ffffffff888824f3>] :ibrix:ibr_release_resource+0x5f3/0x670
 [<ffffffff888d5f1c>] :ibrix:removeSegmentFromContext+0x1bc/0x320
 [<ffffffff889ef56d>] :ibrix:j_errorInjector+0x2d/0x3d0
 [<ffffffff889187b1>] :ibrix:ibrix_aio_write+0x221/0x260
 [<ffffffff88924ea0>] :ibrix:ibrix_file_writev+0xe0/0x130
 [<ffffffff8001819f>] do_sync_write+0x0/0x104
 [<ffffffff800e1368>] do_readv_writev+0x172/0x291
 [<ffffffff8001819f>] do_sync_write+0x0/0x104
 [<ffffffff888d6cb2>] :ibrix:release_ds_context_internal+0x202/0x230
 [<ffffffff88941106>] :ibrix:ibrix_open_file+0x756/0x770
 [<ffffffff89cb4711>] :nfsd:nfsd_vfs_write+0xf2/0x2e1
 [<ffffffff889409b0>] :ibrix:ibrix_open_file+0x0/0x770
 [<ffffffff8001e9e2>] __dentry_open+0x101/0x1dc
 [<ffffffff89cb5027>] :nfsd:nfsd_write+0xb5/0xd5
 [<ffffffff89cbbb1a>] :nfsd:nfsd3_proc_write+0xea/0x109
 [<ffffffff89cb11db>] :nfsd:nfsd_dispatch+0xd8/0x1d6
 [<ffffffff88630651>] :sunrpc:svc_process+0x454/0x71b
 [<ffffffff80065644>] __down_read+0x12/0x92
 [<ffffffff89cb15a1>] :nfsd:nfsd+0x0/0x2cb
 [<ffffffff89cb1746>] :nfsd:nfsd+0x1a5/0x2cb
 [<ffffffff8005efb1>] child_rip+0xa/0x11
 [<ffffffff89cb15a1>] :nfsd:nfsd+0x0/0x2cb
 [<ffffffff89cb15a1>] :nfsd:nfsd+0x0/0x2cb
 [<ffffffff8005efa7>] child_rip+0x0/0x11

Kernel panic - not syncing: hung_task: blocked tasks
The system was low on memory, and a thread was blocked on a RW lock
waiting for other thread to finish write. That other thread was waiting in prune_dcache() for the spinlock:

--- <NMI exception stack> ---
 #3 [ffff81060275d618] .text.lock.spinlock at ffffffff80065bfc (via _spin_lock)
 #4 [ffff81060275d618] prune_dcache at ffffffff8002ea85
 #5 [ffff81060275d648] shrink_dcache_memory at ffffffff800ec75f
 #6 [ffff81060275d658] shrink_slab at ffffffff8003f7aa
 #7 [ffff81060275d698] try_to_free_pages at ffffffff800ccbb7
 #8 [ffff81060275d728] __alloc_pages at ffffffff8000f47f
 #9 [ffff81060275d798] grab_cache_page_write_begin at ffffffff800c6d70
#10 [ffff81060275d7c8] generic_file_buffered_write at ffffffff8000fed3
#11 [ffff81060275d8b8] __generic_file_aio_write_nolock at ffffffff80016641
#12 [ffff81060275d968] generic_file_aio_write_nolock at ffffffff800c75cc
#13 [ffff81060275d998] EsIbx_Write at ffffffff887d4882
#14 [ffff81060275da18] ibrix_aio_write_internal at ffffffff88917c82
#15 [ffff81060275db78] ibrix_aio_write at ffffffff889187b1
#16 [ffff81060275dbc8] ibrix_file_writev at ffffffff88924ea0
#17 [ffff81060275dc08] do_readv_writev at ffffffff800e1368
#18 [ffff81060275dce8] nfsd_vfs_write at ffffffff89cb4711
#19 [ffff81060275dd98] nfsd_write at ffffffff89cb5027
#20 [ffff81060275ddf8] nfsd3_proc_write at ffffffff89cbbb1a
#21 [ffff81060275de48] nfsd_dispatch at ffffffff89cb11db
#22 [ffff81060275de78] svc_process at ffffffff88630651
#23 [ffff81060275dee8] nfsd at ffffffff89cb1746
#24 [ffff81060275df48] kernel_thread at ffffffff8005efb1

crash> kmem -i
              PAGES        TOTAL      PERCENTAGE
 TOTAL MEM  12330931        47 GB         ----
      FREE   360301       1.4 GB    2% of TOTAL MEM
      USED  11970630      45.7 GB   97% of TOTAL MEM
    SHARED  10529542      40.2 GB   85% of TOTAL MEM
   BUFFERS  10265044      39.2 GB   83% of TOTAL MEM
    CACHED  1175243       4.5 GB    9% of TOTAL MEM
      SLAB   314137       1.2 GB    2% of TOTAL MEM

TOTAL HIGH        0            0    0% of TOTAL MEM
 FREE HIGH        0            0    0% of TOTAL HIGH
 TOTAL LOW  12330931        47 GB  100% of TOTAL MEM
  FREE LOW   360301       1.4 GB    2% of TOTAL LOW

TOTAL SWAP   524286         2 GB         ----
 SWAP USED      311       1.2 MB    0% of TOTAL SWAP
 SWAP FREE   523975         2 GB   99% of TOTAL SWAP


Several threads were spinning for dcache_lock spinlock in
shrink_slab()->shrink_dcache_memory()->prune_dcache()->_spinlock()

crash> bt -a |grep prune_dcache
 #4 [ffff81061f549d58] prune_dcache at ffffffff8002ea85
 #4 [ffff8101e0a6ba80] prune_dcache at ffffffff8002ea85
 #4 [ffff8105adef3a80] prune_dcache at ffffffff8002ea85
 #4 [ffff8104d509ba80] prune_dcache at ffffffff8002ea85
 #4 [ffff810609f6fcf8] prune_dcache at ffffffff8002ea85
 #4 [ffff81060c12dd38] prune_dcache at ffffffff8002ea85
 #4 [ffff8101e6aeda80] prune_dcache at ffffffff8002ea85
 #4 [ffff81061f5ddd58] prune_dcache at ffffffff8002ea85
 #4 [ffff81060275d618] prune_dcache at ffffffff8002ea85
 #4 [ffff810b9b443a80] prune_dcache at ffffffff8002ea85
 #4 [ffff8103ea0b5a80] prune_dcache at ffffffff8002ea85

11 processors out of 16 were stuck contending for dcache_lock, like this one:

PID: 6782   TASK: ffff8101e0a6e820  CPU: 15  COMMAND: "mkibfs"
 #0 [ffff810c1fccaf20] crash_nmi_callback at ffffffff8007bf44
 #1 [ffff810c1fccaf40] do_nmi at ffffffff8006688a
 #2 [ffff810c1fccaf50] nmi at ffffffff80065eef
    [exception RIP: .text.lock.spinlock+2]
    RIP: ffffffff80065bfc  RSP: ffff8103ea0b5a80  RFLAGS: 00000282
    RAX: 00000000ffffffff  RBX: ffff810c1fffcac0  RCX: 0000000000000064
    RDX: 0000000000000000  RSI: 0000000000000000  RDI: ffffffff803f2480
    RBP: 00000000ffc6c470   R8: 0000000000000246   R9: 0000000000000030
    R10: ffff81063511d490  R11: 00000000ffc6c470  R12: 0000000000000000
    R13: 0000000000000080  R14: 0000000000000000  R15: 0000000000002b80
    ORIG_RAX: ffffffffffffffff  CS: 0010  SS: 0018
--- <NMI exception stack> ---
 #3 [ffff8103ea0b5a80] .text.lock.spinlock at ffffffff80065bfc (via _spin_lock)
 #4 [ffff8103ea0b5a80] prune_dcache at ffffffff8002ea85
 #5 [ffff8103ea0b5ab0] shrink_dcache_memory at ffffffff800ec75f
 #6 [ffff8103ea0b5ac0] shrink_slab at ffffffff8003f7aa
 #7 [ffff8103ea0b5b00] try_to_free_pages at ffffffff800ccbb7
 #8 [ffff8103ea0b5b90] __alloc_pages at ffffffff8000f47f
 #9 [ffff8103ea0b5c00] grab_cache_page_write_begin at ffffffff800c6d70
#10 [ffff8103ea0b5c30] generic_file_buffered_write at ffffffff8000fed3
#11 [ffff8103ea0b5d20] __generic_file_aio_write_nolock at ffffffff80016641
#12 [ffff8103ea0b5dd0] generic_file_aio_write_nolock at ffffffff800c75cc
#13 [ffff8103ea0b5e00] generic_file_write_nolock at ffffffff800c79a3
#14 [ffff8103ea0b5ef0] blkdev_file_write at ffffffff800e560d
#15 [ffff8103ea0b5f10] vfs_write at ffffffff80016a49
#16 [ffff8103ea0b5f40] sys_write at ffffffff80017316
#17 [ffff8103ea0b5f80] tracesys at ffffffff8005e28d (via system_call)

The problem is similar to what has been described in
RedHat bug 526612
https://bugzilla.redhat.com/show_bug.cgi?id=526612
It is supposed be fixed according to
http://rhn.redhat.com/errata/RHSA-2010-0178.html
in the kernel 2.6.18-194 and that is the version of the kernel we have installed.

Indeed it looks like an incomplete fix for bug 526612.