Messages in this thread | | | Subject | [RFC PATCH 0/2] ipc: do not hold ipc lock more than necessary | From | Davidlohr Bueso <> | Date | Fri, 01 Mar 2013 16:16:32 -0800 |
| |
The following set of not-thoroughly-tested patches are based on the discussion of holding the ipc lock unnecessarily, such as for permissions and security checks:
https://lkml.org/lkml/2013/2/28/540
Patch 0/1: Introduces new functions, analogous to ipc_lock and ipc_lock_check in the ipc utility code, allowing to obtain the ipc object without holding the lock.
Patch 0/2: Use the new functions and only acquire the ipc lock when needed.
With Rik's semop-multi.c microbenchmark we can see the following results:
256 sems without patches: + 59.40% a.out [kernel.kallsyms] [k] _raw_spin_lock + 6.14% a.out [kernel.kallsyms] [k] sys_semtimedop + 3.84% a.out [kernel.kallsyms] [k] avc_has_perm_flags + 3.64% a.out [kernel.kallsyms] [k] __audit_syscall_exit + 2.06% a.out [kernel.kallsyms] [k] copy_user_enhanced_fast_string + 1.86% a.out [kernel.kallsyms] [k] ipc_lock + 1.75% a.out [kernel.kallsyms] [k] __audit_syscall_entry + 1.69% a.out [kernel.kallsyms] [k] ipc_has_perm.isra.21 + 1.47% a.out [kernel.kallsyms] [k] do_smart_update + 1.43% a.out [kernel.kallsyms] [k] pid_vnr + 1.39% a.out [kernel.kallsyms] [k] try_atomic_semop.isra.5
total operations: 151452270, ops/sec 5048409
256 sems with patches: + 17.47% a.out [kernel.kallsyms] [k] _raw_spin_lock + 11.08% a.out [kernel.kallsyms] [k] sys_semtimedop + 8.81% a.out [kernel.kallsyms] [k] avc_has_perm_flags + 7.96% a.out [kernel.kallsyms] [k] ipc_has_perm.isra.21 + 6.50% a.out [kernel.kallsyms] [k] __audit_syscall_exit + 4.67% a.out [kernel.kallsyms] [k] ipc_obtain_object_check + 4.19% a.out [kernel.kallsyms] [k] ipcperms + 3.75% a.out [kernel.kallsyms] [k] copy_user_enhanced_fast_string + 3.38% a.out [kernel.kallsyms] [k] system_call + 3.05% a.out [kernel.kallsyms] [k] try_atomic_semop.isra.5 + 2.70% a.out [kernel.kallsyms] [k] do_smart_update + 2.60% a.out [kernel.kallsyms] [k] __audit_syscall_entry
total operations: 266502912, ops/sec 8883430
While the _raw_spin_lock time is drastically reduced, others do increase. This results in an overall speedup of ~1.7x regarding ops/sec.
| |