nfsd lockups with xfs during SPEC SFS testing

["HABBINGA,ERIK (HP-Loveland,ex1)" <erik_habbinga@xxxxxx>]

I'm running linux 2.4.17 with a version of XFS downloaded via CVS on Jan
30th.  When I run the SPEC SFS NFS test against this kernel, nfsd stops
responding after awhile.  I captured the state of all of the system
processes via magic sysrq, and found 24 nfsd processes locked up in various
stages of the nfsd_lookup code:

- 20 of them were locked up in the fh_lock call before lookup_one_len in
nfsd_lookup().  
- 2 processes were locked up in the _pagebuf_grab_lock call inside
_pagebuf_find_lockable_buffer().
- 2 processes were locked up in the pagebuf_iowait() call in
pagebuf_iostart()

Any ideas on what may be wrong, and how I can help debug and solve this
problem?  I've attached the call traces for the locked up nfsd processes.  I
can provide vmlinux and System.map for this kernel to help debugging.

Thanks,
Erik Habbinga
Hewlett Packard

nfsd processes locked up in nfsd_lookup()->fh_lock()
task: nfsd
        c0105ab4: c0105a48 T __down
        c0105c50: c0105c48 T __down_failed
        c02c9bab: c02c6214 T stext_lock
        c0168048: c0167f74 t nfsd3_proc_lookup
        c015fa13: c015f940 t nfsd_dispatch
        c02bd775: c02bd4e8 T svc_process
        c015f80f: c015f618 t nfsd
        c0105594: c010556c T kernel_thread

nfsd processes locked up in
_pagebuf_find_lockable_buffer()->_pagebuf_grab_lock call() 
task: nfsd
        c0105ab4: c0105a48 T __down
        c0105c50: c0105c48 T __down_failed
        c02ca95c: c02c6214 T stext_lock
        c01db8e6: c01db7e4 T _pagebuf_find_lockable_buffer
        c01e5d92: c01e5d58 T kmem_zone_alloc
        c01a7545: c01a7510 t xfs_dir2_block_lookup_int
        c01a7545: c01a7510 t xfs_dir2_block_lookup_int
        c01dba19: c01db9e4 T _pagebuf_get_lockable_buffer
        c01d829f: c01d8268 T pagebuf_get
        c01cc374: c01cc338 T xfs_trans_read_buf
        c01b9078: c01b8f78 T xfs_itobp
        c01ba0cc: c01ba05c T xfs_iread
        c01b80c2: c01b7ebc T xfs_iget_core
        c0147ebe: c0147df0 T icreate
        c01b8488: c01b8404 T xfs_iget
        c01cd61c: c01cd4f8 T xfs_dir_lookup_int
        c01d1c07: c01d1b78 t xfs_lookup
        c01df729: c01df6c4 T linvfs_lookup
        c013df41: c013dea8 T lookup_hash
        c013dfd7: c013df80 T lookup_one_len
        c016263d: c0162370 T nfsd_lookup
        c0168048: c0167f74 t nfsd3_proc_lookup
        c015fa13: c015f940 t nfsd_dispatch
        c02bd775: c02bd4e8 T svc_process
        c015f80f: c015f618 t nfsd
        c0105594: c010556c T kernel_thread

0xc02ca95c <stext_lock+18248>:  jmp    0xc01db7e1 <_pagebuf_grab_lock+17>

nfsd processes locked up in pagebuf_iostart()->pagebuf_iowait()
task: nfsd
        c0105ab4: c0105a48 T __down
        c0105c50: c0105c48 T __down_failed
        c02ca8b3: c02c6214 T stext_lock
        c01d88cd: c01d8844 T pagebuf_iostart
        c01d830d: c01d8268 T pagebuf_get
        c01cc374: c01cc338 T xfs_trans_read_buf
        c01b9078: c01b8f78 T xfs_itobp
        c01ba0cc: c01ba05c T xfs_iread
        c01b80c2: c01b7ebc T xfs_iget_core
        c0147ebe: c0147df0 T icreate
        c01b8488: c01b8404 T xfs_iget
        c01cd61c: c01cd4f8 T xfs_dir_lookup_int
        c01d1c07: c01d1b78 t xfs_lookup
        c01df729: c01df6c4 T linvfs_lookup
        c013df41: c013dea8 T lookup_hash
        c013dfd7: c013df80 T lookup_one_len
        c016263d: c0162370 T nfsd_lookup
        c0168048: c0167f74 t nfsd3_proc_lookup
        c015fa13: c015f940 t nfsd_dispatch
        c02bd775: c02bd4e8 T svc_process
        c015f80f: c015f618 t nfsd
        c0105594: c010556c T kernel_thread

0xc02ca8b3 <stext_lock+18079>:  jmp    0xc01d9109 <pagebuf_iowait+49>