Paul E. McKenney | a241ec6 | 2005-10-30 15:03:12 -0800 | [diff] [blame] | 1 | RCU Torture Test Operation |
| 2 | |
| 3 | |
| 4 | CONFIG_RCU_TORTURE_TEST |
| 5 | |
| 6 | The CONFIG_RCU_TORTURE_TEST config option is available for all RCU |
| 7 | implementations. It creates an rcutorture kernel module that can |
| 8 | be loaded to run a torture test. The test periodically outputs |
| 9 | status messages via printk(), which can be examined via the dmesg |
| 10 | command (perhaps grepping for "rcutorture"). The test is started |
| 11 | when the module is loaded, and stops when the module is unloaded. |
| 12 | |
| 13 | However, actually setting this config option to "y" results in the system |
| 14 | running the test immediately upon boot, and ending only when the system |
| 15 | is taken down. Normally, one will instead want to build the system |
| 16 | with CONFIG_RCU_TORTURE_TEST=m and to use modprobe and rmmod to control |
| 17 | the test, perhaps using a script similar to the one shown at the end of |
| 18 | this document. Note that you will need CONFIG_MODULE_UNLOAD in order |
| 19 | to be able to end the test. |
| 20 | |
| 21 | |
| 22 | MODULE PARAMETERS |
| 23 | |
| 24 | This module has the following parameters: |
| 25 | |
| 26 | nreaders This is the number of RCU reading threads supported. |
| 27 | The default is twice the number of CPUs. Why twice? |
| 28 | To properly exercise RCU implementations with preemptible |
| 29 | read-side critical sections. |
| 30 | |
| 31 | stat_interval The number of seconds between output of torture |
| 32 | statistics (via printk()). Regardless of the interval, |
| 33 | statistics are printed when the module is unloaded. |
| 34 | Setting the interval to zero causes the statistics to |
| 35 | be printed -only- when the module is unloaded, and this |
| 36 | is the default. |
| 37 | |
| 38 | verbose Enable debug printk()s. Default is disabled. |
| 39 | |
| 40 | |
| 41 | OUTPUT |
| 42 | |
| 43 | The statistics output is as follows: |
| 44 | |
| 45 | rcutorture: --- Start of test: nreaders=16 stat_interval=0 verbose=0 |
| 46 | rcutorture: rtc: 0000000000000000 ver: 1916 tfle: 0 rta: 1916 rtaf: 0 rtf: 1915 |
| 47 | rcutorture: Reader Pipe: 1466408 9747 0 0 0 0 0 0 0 0 0 |
| 48 | rcutorture: Reader Batch: 1464477 11678 0 0 0 0 0 0 0 0 |
| 49 | rcutorture: Free-Block Circulation: 1915 1915 1915 1915 1915 1915 1915 1915 1915 1915 0 |
| 50 | rcutorture: --- End of test |
| 51 | |
| 52 | The command "dmesg | grep rcutorture:" will extract this information on |
| 53 | most systems. On more esoteric configurations, it may be necessary to |
| 54 | use other commands to access the output of the printk()s used by |
| 55 | the RCU torture test. The printk()s use KERN_ALERT, so they should |
| 56 | be evident. ;-) |
| 57 | |
| 58 | The entries are as follows: |
| 59 | |
| 60 | o "ggp": The number of counter flips (or batches) since boot. |
| 61 | |
| 62 | o "rtc": The hexadecimal address of the structure currently visible |
| 63 | to readers. |
| 64 | |
| 65 | o "ver": The number of times since boot that the rcutw writer task |
| 66 | has changed the structure visible to readers. |
| 67 | |
| 68 | o "tfle": If non-zero, indicates that the "torture freelist" |
| 69 | containing structure to be placed into the "rtc" area is empty. |
| 70 | This condition is important, since it can fool you into thinking |
| 71 | that RCU is working when it is not. :-/ |
| 72 | |
| 73 | o "rta": Number of structures allocated from the torture freelist. |
| 74 | |
| 75 | o "rtaf": Number of allocations from the torture freelist that have |
| 76 | failed due to the list being empty. |
| 77 | |
| 78 | o "rtf": Number of frees into the torture freelist. |
| 79 | |
| 80 | o "Reader Pipe": Histogram of "ages" of structures seen by readers. |
| 81 | If any entries past the first two are non-zero, RCU is broken. |
| 82 | And rcutorture prints the error flag string "!!!" to make sure |
| 83 | you notice. The age of a newly allocated structure is zero, |
| 84 | it becomes one when removed from reader visibility, and is |
| 85 | incremented once per grace period subsequently -- and is freed |
| 86 | after passing through (RCU_TORTURE_PIPE_LEN-2) grace periods. |
| 87 | |
| 88 | The output displayed above was taken from a correctly working |
| 89 | RCU. If you want to see what it looks like when broken, break |
| 90 | it yourself. ;-) |
| 91 | |
| 92 | o "Reader Batch": Another histogram of "ages" of structures seen |
| 93 | by readers, but in terms of counter flips (or batches) rather |
| 94 | than in terms of grace periods. The legal number of non-zero |
| 95 | entries is again two. The reason for this separate view is |
| 96 | that it is easier to get the third entry to show up in the |
| 97 | "Reader Batch" list than in the "Reader Pipe" list. |
| 98 | |
| 99 | o "Free-Block Circulation": Shows the number of torture structures |
| 100 | that have reached a given point in the pipeline. The first element |
| 101 | should closely correspond to the number of structures allocated, |
| 102 | the second to the number that have been removed from reader view, |
| 103 | and all but the last remaining to the corresponding number of |
| 104 | passes through a grace period. The last entry should be zero, |
| 105 | as it is only incremented if a torture structure's counter |
| 106 | somehow gets incremented farther than it should. |
| 107 | |
| 108 | |
| 109 | USAGE |
| 110 | |
| 111 | The following script may be used to torture RCU: |
| 112 | |
| 113 | #!/bin/sh |
| 114 | |
| 115 | modprobe rcutorture |
| 116 | sleep 100 |
| 117 | rmmod rcutorture |
| 118 | dmesg | grep rcutorture: |
| 119 | |
| 120 | The output can be manually inspected for the error flag of "!!!". |
| 121 | One could of course create a more elaborate script that automatically |
| 122 | checked for such errors. |