Messages in this thread | | | Subject | Re: [PATCH 00/11] sysv mqueue: do not hold the ipc lock unnecessarily | From | Davidlohr Bueso <> | Date | Thu, 23 May 2013 18:45:30 -0700 |
| |
ping, Andrew?
On Wed, 2013-05-15 at 18:07 -0700, Davidlohr Bueso wrote: > This patchset continues the work that began in the sysv ipc semaphore scaling > series: https://lkml.org/lkml/2013/3/20/546 > > Just like semaphores used to be, sysv shared memory and msg queues also abuse the ipc > lock, unnecessarily holding it for operations such as permission and security checks. This > patchset mostly deals with mqueues, and while shared mem can be done in a very similar way, > I want to get these patches out in the open first. It also does some pending cleanups, > mostly focused on the two level locking we have in ipc code, taking care of ipc_addid() > and ipcctl_pre_down_nolock() - yes there are still functions that need to be updated as well. > > I have tried to split each patch to be as readable and specific as possible, specially when > shortening the critical regions, going one function at a time. > > Patch 1 moves the locking to be explicitly done by the callers of ipc_addid. > It updates msg, sem and shm. > > Patches 2-3: are just wrappers around the ipc lock, initially suggested by Linus. > > Patch 4 is similar to patch 1, moving the rcu and rw_mutex locking out of > ipcctl_pre_down_nolock so that the callers explicitly deals with them. It updates msg, sem > and shm. > > Patch 5 shortens the critical region in msgctl_down(), using the lockless > ipcctl_pre_down() function and only acquiring the ipc lock for RMID and SET commands. > > Patch 6 simply moves the what-should-be lockless logic of *_INFO and *_STAT commands > out of msgctl() into a new function, msgctl_lockless(). > > Patch 7 introduces the necessary wrappers around ipc_obtain_object[_check]() > that will later enable us to separately lookup the ipc object without holding the lock. > > Patch 8 updates the previously added msgctl_nolock() to actually be lockless, reducing > the critical region for the STAT commands. > > Patch 9 redoes the locking for msgsend(). > > Patch 10 redoes the locking for msgrcv(). > > Patch 11 removes the now unused msg_lock and msg_lock_check functions, replaced by > a smarter combination of rcu, ipc_obtain_object and ipc_object_lock. > > Davidlohr Bueso (11): > ipc: move rcu lock out of ipc_addid > ipc: introduce ipc object locking helpers > ipc: close open coded spin lock calls > ipc: move locking out of ipcctl_pre_down_nolock > ipc,msg: shorten critical region in semctl_down > ipc,msg: introduce msgctl_nolock > ipc,msg: introduce lockless functions to obtain the ipc object > ipc,msg: make msgctl_nolock lockless > ipc,msg: reduce critical region in msgsnd > ipc,msg: make shorten critical region in msgrcv > ipc: remove unused functions > > ipc/msg.c | 227 ++++++++++++++++++++++++++++++++++++++----------------------- > ipc/sem.c | 42 +++++++----- > ipc/shm.c | 32 ++++++--- > ipc/util.c | 25 ++----- > ipc/util.h | 22 ++++-- > 5 files changed, 211 insertions(+), 137 deletions(-) > > Thanks, > Davidlohr >
| |