| Peter Zijlstra | a560aa4 | 2007-10-07 00:24:33 -0700 | [diff] [blame] | 1 | |
| 2 | LOCK STATISTICS |
| 3 | |
| 4 | - WHAT |
| 5 | |
| 6 | As the name suggests, it provides statistics on locks. |
| 7 | |
| 8 | - WHY |
| 9 | |
| 10 | Because things like lock contention can severely impact performance. |
| 11 | |
| 12 | - HOW |
| 13 | |
| 14 | Lockdep already has hooks in the lock functions and maps lock instances to |
| 15 | lock classes. We build on that. The graph below shows the relation between |
| 16 | the lock functions and the various hooks therein. |
| 17 | |
| 18 | __acquire |
| 19 | | |
| 20 | lock _____ |
| 21 | | \ |
| 22 | | __contended |
| 23 | | | |
| 24 | | <wait> |
| 25 | | _______/ |
| 26 | |/ |
| 27 | | |
| 28 | __acquired |
| 29 | | |
| 30 | . |
| 31 | <hold> |
| 32 | . |
| 33 | | |
| 34 | __release |
| 35 | | |
| 36 | unlock |
| 37 | |
| 38 | lock, unlock - the regular lock functions |
| 39 | __* - the hooks |
| 40 | <> - states |
| 41 | |
| 42 | With these hooks we provide the following statistics: |
| 43 | |
| 44 | con-bounces - number of lock contention that involved x-cpu data |
| 45 | contentions - number of lock acquisitions that had to wait |
| 46 | wait time min - shortest (non-0) time we ever had to wait for a lock |
| 47 | max - longest time we ever had to wait for a lock |
| 48 | total - total time we spend waiting on this lock |
| 49 | acq-bounces - number of lock acquisitions that involved x-cpu data |
| 50 | acquisitions - number of times we took the lock |
| 51 | hold time min - shortest (non-0) time we ever held the lock |
| 52 | max - longest time we ever held the lock |
| 53 | total - total time this lock was held |
| 54 | |
| 55 | From these number various other statistics can be derived, such as: |
| 56 | |
| 57 | hold time average = hold time total / acquisitions |
| 58 | |
| 59 | These numbers are gathered per lock class, per read/write state (when |
| 60 | applicable). |
| 61 | |
| 62 | It also tracks 4 contention points per class. A contention point is a call site |
| 63 | that had to wait on lock acquisition. |
| 64 | |
| 65 | - USAGE |
| 66 | |
| 67 | Look at the current lock statistics: |
| 68 | |
| 69 | ( line numbers not part of actual output, done for clarity in the explanation |
| 70 | below ) |
| 71 | |
| 72 | # less /proc/lock_stat |
| 73 | |
| 74 | 01 lock_stat version 0.2 |
| 75 | 02 ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| 76 | 03 class name con-bounces contentions waittime-min waittime-max waittime-total acq-bounces acquisitions holdtime-min holdtime-max holdtime-total |
| 77 | 04 ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| 78 | 05 |
| 79 | 06 &inode->i_data.tree_lock-W: 15 21657 0.18 1093295.30 11547131054.85 58 10415 0.16 87.51 6387.60 |
| 80 | 07 &inode->i_data.tree_lock-R: 0 0 0.00 0.00 0.00 23302 231198 0.25 8.45 98023.38 |
| 81 | 08 -------------------------- |
| 82 | 09 &inode->i_data.tree_lock 0 [<ffffffff8027c08f>] add_to_page_cache+0x5f/0x190 |
| 83 | 10 |
| 84 | 11 ............................................................................................................................................................................................... |
| 85 | 12 |
| 86 | 13 dcache_lock: 1037 1161 0.38 45.32 774.51 6611 243371 0.15 306.48 77387.24 |
| 87 | 14 ----------- |
| 88 | 15 dcache_lock 180 [<ffffffff802c0d7e>] sys_getcwd+0x11e/0x230 |
| 89 | 16 dcache_lock 165 [<ffffffff802c002a>] d_alloc+0x15a/0x210 |
| 90 | 17 dcache_lock 33 [<ffffffff8035818d>] _atomic_dec_and_lock+0x4d/0x70 |
| 91 | 18 dcache_lock 1 [<ffffffff802beef8>] shrink_dcache_parent+0x18/0x130 |
| 92 | |
| 93 | This excerpt shows the first two lock class statistics. Line 01 shows the |
| 94 | output version - each time the format changes this will be updated. Line 02-04 |
| 95 | show the header with column descriptions. Lines 05-10 and 13-18 show the actual |
| 96 | statistics. These statistics come in two parts; the actual stats separated by a |
| 97 | short separator (line 08, 14) from the contention points. |
| 98 | |
| 99 | The first lock (05-10) is a read/write lock, and shows two lines above the |
| 100 | short separator. The contention points don't match the column descriptors, |
| 101 | they have two: contentions and [<IP>] symbol. |
| 102 | |
| 103 | |
| 104 | View the top contending locks: |
| 105 | |
| 106 | # grep : /proc/lock_stat | head |
| 107 | &inode->i_data.tree_lock-W: 15 21657 0.18 1093295.30 11547131054.85 58 10415 0.16 87.51 6387.60 |
| 108 | &inode->i_data.tree_lock-R: 0 0 0.00 0.00 0.00 23302 231198 0.25 8.45 98023.38 |
| 109 | dcache_lock: 1037 1161 0.38 45.32 774.51 6611 243371 0.15 306.48 77387.24 |
| 110 | &inode->i_mutex: 161 286 18446744073709 62882.54 1244614.55 3653 20598 18446744073709 62318.60 1693822.74 |
| 111 | &zone->lru_lock: 94 94 0.53 7.33 92.10 4366 32690 0.29 59.81 16350.06 |
| 112 | &inode->i_data.i_mmap_lock: 79 79 0.40 3.77 53.03 11779 87755 0.28 116.93 29898.44 |
| 113 | &q->__queue_lock: 48 50 0.52 31.62 86.31 774 13131 0.17 113.08 12277.52 |
| 114 | &rq->rq_lock_key: 43 47 0.74 68.50 170.63 3706 33929 0.22 107.99 17460.62 |
| 115 | &rq->rq_lock_key#2: 39 46 0.75 6.68 49.03 2979 32292 0.17 125.17 17137.63 |
| 116 | tasklist_lock-W: 15 15 1.45 10.87 32.70 1201 7390 0.58 62.55 13648.47 |
| 117 | |
| 118 | Clear the statistics: |
| 119 | |
| 120 | # echo 0 > /proc/lock_stat |