releases/3.11.6/ipc-shm-introduce-lockless-functions-to-obtain-the-ipc-object.patch

   1 From 8b8d52ac382b17a19906b930cd69e2edb0aca8ba Mon Sep 17 00:00:00 2001
   2 From: Davidlohr Bueso <davidlohr.bueso@hp.com>
   3 Date: Wed, 11 Sep 2013 14:26:15 -0700
   4 Subject: ipc,shm: introduce lockless functions to obtain the ipc object
   5
   6 From: Davidlohr Bueso <davidlohr.bueso@hp.com>
   7
   8 commit 8b8d52ac382b17a19906b930cd69e2edb0aca8ba upstream.
   9
  10 This is the third and final patchset that deals with reducing the amount
  11 of contention we impose on the ipc lock (kern_ipc_perm.lock).  These
  12 changes mostly deal with shared memory, previous work has already been
  13 done for semaphores and message queues:
  14
  15   http://lkml.org/lkml/2013/3/20/546 (sems)
  16   http://lkml.org/lkml/2013/5/15/584 (mqueues)
  17
  18 With these patches applied, a custom shm microbenchmark stressing shmctl
  19 doing IPC_STAT with 4 threads a million times, reduces the execution
  20 time by 50%.  A similar run, this time with IPC_SET, reduces the
  21 execution time from 3 mins and 35 secs to 27 seconds.
  22
  23 Patches 1-8: replaces blindly taking the ipc lock for a smarter
  24 combination of rcu and ipc_obtain_object, only acquiring the spinlock
  25 when updating.
  26
  27 Patch 9: renames the ids rw_mutex to rwsem, which is what it already was.
  28
  29 Patch 10: is a trivial mqueue leftover cleanup
  30
  31 Patch 11: adds a brief lock scheme description, requested by Andrew.
  32
  33 This patch:
  34
  35 Add shm_obtain_object() and shm_obtain_object_check(), which will allow us
  36 to get the ipc object without acquiring the lock.  Just as with other
  37 forms of ipc, these functions are basically wrappers around
  38 ipc_obtain_object*().
  39
  40 Signed-off-by: Davidlohr Bueso <davidlohr.bueso@hp.com>
  41 Tested-by: Sedat Dilek <sedat.dilek@gmail.com>
  42 Cc: Rik van Riel <riel@redhat.com>
  43 Cc: Manfred Spraul <manfred@colorfullife.com>
  44 Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
  45 Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
  46 Cc: Mike Galbraith <efault@gmx.de>
  47 Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
  48
  49 ---
  50  ipc/shm.c |   20 ++++++++++++++++++++
  51  1 file changed, 20 insertions(+)
  52
  53 --- a/ipc/shm.c
  54 +++ b/ipc/shm.c
  55 @@ -124,6 +124,26 @@ void __init shm_init (void)
  56                                 IPC_SHM_IDS, sysvipc_shm_proc_show);
  57  }
  58
  59 +static inline struct shmid_kernel *shm_obtain_object(struct ipc_namespace *ns, int id)
  60 +{
  61 +       struct kern_ipc_perm *ipcp = ipc_obtain_object(&shm_ids(ns), id);
  62 +
  63 +       if (IS_ERR(ipcp))
  64 +               return ERR_CAST(ipcp);
  65 +
  66 +       return container_of(ipcp, struct shmid_kernel, shm_perm);
  67 +}
  68 +
  69 +static inline struct shmid_kernel *shm_obtain_object_check(struct ipc_namespace *ns, int id)
  70 +{
  71 +       struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&shm_ids(ns), id);
  72 +
  73 +       if (IS_ERR(ipcp))
  74 +               return ERR_CAST(ipcp);
  75 +
  76 +       return container_of(ipcp, struct shmid_kernel, shm_perm);
  77 +}
  78 +
  79  /*
  80   * shm_lock_(check_) routines are called in the paths where the rw_mutex
  81   * is not necessarily held.