HEADS UP: filedesc_sx patch in CVS HEAD

I've committed the below to the tree; Kris has performed quite a lot of 
performance and stability testing, but since he tends to run with specific 
workloads, I wouldn't be surprised if there are minor (and hopefully quickly 
corrected) issues reported.  If you experience hangs or other problems, please 
make sure to run with INVARIANTS and WITNESS, which will help with debugging.

This patch represents a significant part of the performance improvements for 
improved scalability on 7-CURRENT with respect to threaded databases, and is 
only possible because of the long hours of work Attilio, Kris, John, and 
others have put in preparing the sxlock optimizations this patch depends on, 
as well as reviewing and testing the patch.

Please let me know if you experience any problems.

Thanks,

Robert N M Watson
Computer Laboratory
University of Cambridge

---------- Forwarded message ----------
Date: Wed, 4 Apr 2007 09:11:34 +0000 (UTC)
From: Robert Watson <rwatson_at_FreeBSD.org>
To: src-committers_at_FreeBSD.org, cvs-src_at_FreeBSD.org, cvs-all_at_FreeBSD.org
Subject: cvs commit: src/sys/compat/linux linux_file.c src/sys/compat/svr4
        svr4_filio.c src/sys/dev/streams streams.c src/sys/fs/devfs
     devfs_vnops.c src/sys/fs/fdescfs fdesc_vfsops.c fdesc_vnops.c
     src/sys/fs/fifofs fifo_vnops.c src/sys/fs/unionfs union_subr.c ...

rwatson     2007-04-04 09:11:34 UTC

   FreeBSD src repository

   Modified files:
     sys/compat/linux     linux_file.c
     sys/compat/svr4      svr4_filio.c
     sys/dev/streams      streams.c
     sys/fs/devfs         devfs_vnops.c
     sys/fs/fdescfs       fdesc_vfsops.c fdesc_vnops.c
     sys/fs/fifofs        fifo_vnops.c
     sys/fs/unionfs       union_subr.c
     sys/kern             kern_descrip.c kern_event.c kern_fork.c
                          subr_witness.c sys_generic.c
                          uipc_mqueue.c uipc_syscalls.c
                          uipc_usrreq.c vfs_cache.c vfs_lookup.c
                          vfs_mount.c vfs_syscalls.c
     sys/netsmb           smb_dev.c
     sys/opencrypto       cryptodev.c
     sys/security/audit   audit_bsm_klib.c
     sys/sys              filedesc.h
   Log:
   Replace custom file descriptor array sleep lock constructed using a mutex
   and flags with an sxlock.  This leads to a significant and measurable
   performance improvement as a result of access to shared locking for
   frequent lookup operations, reduced general overhead, and reduced overhead
   in the event of contention.  All of these are imported for threaded
   applications where simultaneous access to a shared file descriptor array
   occurs frequently.  Kris has reported 2x-4x transaction rate improvements
   on 8-core MySQL benchmarks; smaller improvements can be expected for many
   workloads as a result of reduced overhead.

   - Generally eliminate the distinction between "fast" and regular
     acquisisition of the filedesc lock; the plan is that they will now all
     be fast.  Change all locking instances to either shared or exclusive
     locks.

   - Correct a bug (pointed out by kib) in fdfree() where previously msleep()
     was called without the mutex held; sx_sleep() is now always called with
     the sxlock held exclusively.

   - Universally hold the struct file lock over changes to struct file,
     rather than the filedesc lock or no lock.  Always update the f_ops
     field last. A further memory barrier is required here in the future
     (discussed with jhb).

   - Improve locking and reference management in linux_at(), which fails to
     properly acquire vnode references before using vnode pointers.  Annotate
     improper use of vn_fullpath(), which will be replaced at a future date.

   In fcntl(), we conservatively acquire an exclusive lock, even though in
   some cases a shared lock may be sufficient, which should be revisited.
   The dropping of the filedesc lock in fdgrowtable() is no longer required
   as the sxlock can be held over the sleep operation; we should consider
   removing that (pointed out by attilio).

   Tested by:      kris
   Discussed with: jhb, kris, attilio, jeff

   Revision  Changes    Path
   1.103     +17 -4     src/sys/compat/linux/linux_file.c
   1.35      +4 -4      src/sys/compat/svr4/svr4_filio.c
   1.55      +2 -2      src/sys/dev/streams/streams.c
   1.143     +3 -1      src/sys/fs/devfs/devfs_vnops.c
   1.56      +2 -2      src/sys/fs/fdescfs/fdesc_vfsops.c
   1.104     +5 -5      src/sys/fs/fdescfs/fdesc_vnops.c
   1.136     +3 -1      src/sys/fs/fifofs/fifo_vnops.c
   1.91      +2 -2      src/sys/fs/unionfs/union_subr.c
   1.307     +174 -170  src/sys/kern/kern_descrip.c
   1.109     +9 -9      src/sys/kern/kern_event.c
   1.270     +2 -2      src/sys/kern/kern_fork.c
   1.228     +0 -2      src/sys/kern/subr_witness.c
   1.155     +11 -12    src/sys/kern/sys_generic.c
   1.21      +10 -11    src/sys/kern/uipc_mqueue.c
   1.250     +14 -9     src/sys/kern/uipc_syscalls.c
   1.201     +10 -9     src/sys/kern/uipc_usrreq.c
   1.108     +4 -4      src/sys/kern/vfs_cache.c
   1.100     +2 -2      src/sys/kern/vfs_lookup.c
   1.252     +2 -2      src/sys/kern/vfs_mount.c
   1.436     +26 -25    src/sys/kern/vfs_syscalls.c
   1.32      +3 -3      src/sys/netsmb/smb_dev.c
   1.33      +3 -1      src/sys/opencrypto/cryptodev.c
   1.6       +2 -2      src/sys/security/audit/audit_bsm_klib.c
   1.76      +15 -61    src/sys/sys/filedesc.h