blob: 2967a65269d8af7702dbe76f4780e14d91dd9061 [file] [log] [blame]
Paul E. McKenneya241ec62005-10-30 15:03:12 -08001RCU Torture Test Operation
2
3
4CONFIG_RCU_TORTURE_TEST
5
6The CONFIG_RCU_TORTURE_TEST config option is available for all RCU
7implementations. It creates an rcutorture kernel module that can
8be loaded to run a torture test. The test periodically outputs
9status messages via printk(), which can be examined via the dmesg
Paul E. McKenney72e9bb52006-06-27 02:54:03 -070010command (perhaps grepping for "torture"). The test is started
Paul E. McKenneya241ec62005-10-30 15:03:12 -080011when the module is loaded, and stops when the module is unloaded.
12
13However, actually setting this config option to "y" results in the system
14running the test immediately upon boot, and ending only when the system
15is taken down. Normally, one will instead want to build the system
16with CONFIG_RCU_TORTURE_TEST=m and to use modprobe and rmmod to control
17the test, perhaps using a script similar to the one shown at the end of
18this document. Note that you will need CONFIG_MODULE_UNLOAD in order
19to be able to end the test.
20
21
22MODULE PARAMETERS
23
24This module has the following parameters:
25
26nreaders 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
Josh Triplettb772e1d2006-10-04 02:17:13 -070031nfakewriters This is the number of RCU fake writer threads to run. Fake
32 writer threads repeatedly use the synchronous "wait for
33 current readers" function of the interface selected by
34 torture_type, with a delay between calls to allow for various
35 different numbers of writers running in parallel.
36 nfakewriters defaults to 4, which provides enough parallelism
37 to trigger special cases caused by multiple writers, such as
38 the synchronize_srcu() early return optimization.
39
Paul E. McKenneya241ec62005-10-30 15:03:12 -080040stat_interval The number of seconds between output of torture
41 statistics (via printk()). Regardless of the interval,
42 statistics are printed when the module is unloaded.
43 Setting the interval to zero causes the statistics to
44 be printed -only- when the module is unloaded, and this
45 is the default.
46
Paul E. McKenney29766f12006-06-27 02:54:02 -070047shuffle_interval
48 The number of seconds to keep the test threads affinitied
Paul E. McKenneyf85d6c72008-01-25 21:08:25 +010049 to a particular subset of the CPUs, defaults to 5 seconds.
50 Used in conjunction with test_no_idle_hz.
Paul E. McKenney29766f12006-06-27 02:54:02 -070051
52test_no_idle_hz Whether or not to test the ability of RCU to operate in
53 a kernel that disables the scheduling-clock interrupt to
54 idle CPUs. Boolean parameter, "1" to test, "0" otherwise.
Paul E. McKenneyf85d6c72008-01-25 21:08:25 +010055 Defaults to omitting this test.
Paul E. McKenney29766f12006-06-27 02:54:02 -070056
Josh Triplett20d2e422006-10-04 02:17:15 -070057torture_type The type of RCU to test: "rcu" for the rcu_read_lock() API,
58 "rcu_sync" for rcu_read_lock() with synchronous reclamation,
Josh Triplett11a14702006-10-04 02:17:16 -070059 "rcu_bh" for the rcu_read_lock_bh() API, "rcu_bh_sync" for
Josh Triplett4b6c2cc2006-10-04 02:17:16 -070060 rcu_read_lock_bh() with synchronous reclamation, "srcu" for
61 the "srcu_read_lock()" API, and "sched" for the use of
62 preempt_disable() together with synchronize_sched().
Paul E. McKenney72e9bb52006-06-27 02:54:03 -070063
Paul E. McKenneya241ec62005-10-30 15:03:12 -080064verbose Enable debug printk()s. Default is disabled.
65
66
67OUTPUT
68
69The statistics output is as follows:
70
Paul E. McKenney72e9bb52006-06-27 02:54:03 -070071 rcu-torture: --- Start of test: nreaders=16 stat_interval=0 verbose=0
72 rcu-torture: rtc: 0000000000000000 ver: 1916 tfle: 0 rta: 1916 rtaf: 0 rtf: 1915
73 rcu-torture: Reader Pipe: 1466408 9747 0 0 0 0 0 0 0 0 0
74 rcu-torture: Reader Batch: 1464477 11678 0 0 0 0 0 0 0 0
75 rcu-torture: Free-Block Circulation: 1915 1915 1915 1915 1915 1915 1915 1915 1915 1915 0
76 rcu-torture: --- End of test
Paul E. McKenneya241ec62005-10-30 15:03:12 -080077
Paul E. McKenney72e9bb52006-06-27 02:54:03 -070078The command "dmesg | grep torture:" will extract this information on
Paul E. McKenneya241ec62005-10-30 15:03:12 -080079most systems. On more esoteric configurations, it may be necessary to
80use other commands to access the output of the printk()s used by
81the RCU torture test. The printk()s use KERN_ALERT, so they should
82be evident. ;-)
83
84The entries are as follows:
85
Paul E. McKenneya241ec62005-10-30 15:03:12 -080086o "rtc": The hexadecimal address of the structure currently visible
87 to readers.
88
89o "ver": The number of times since boot that the rcutw writer task
90 has changed the structure visible to readers.
91
92o "tfle": If non-zero, indicates that the "torture freelist"
93 containing structure to be placed into the "rtc" area is empty.
94 This condition is important, since it can fool you into thinking
95 that RCU is working when it is not. :-/
96
97o "rta": Number of structures allocated from the torture freelist.
98
99o "rtaf": Number of allocations from the torture freelist that have
100 failed due to the list being empty.
101
102o "rtf": Number of frees into the torture freelist.
103
104o "Reader Pipe": Histogram of "ages" of structures seen by readers.
105 If any entries past the first two are non-zero, RCU is broken.
106 And rcutorture prints the error flag string "!!!" to make sure
107 you notice. The age of a newly allocated structure is zero,
108 it becomes one when removed from reader visibility, and is
109 incremented once per grace period subsequently -- and is freed
110 after passing through (RCU_TORTURE_PIPE_LEN-2) grace periods.
111
112 The output displayed above was taken from a correctly working
113 RCU. If you want to see what it looks like when broken, break
114 it yourself. ;-)
115
116o "Reader Batch": Another histogram of "ages" of structures seen
117 by readers, but in terms of counter flips (or batches) rather
118 than in terms of grace periods. The legal number of non-zero
Paul E. McKenneyf85d6c72008-01-25 21:08:25 +0100119 entries is again two. The reason for this separate view is that
120 it is sometimes easier to get the third entry to show up in the
Paul E. McKenneya241ec62005-10-30 15:03:12 -0800121 "Reader Batch" list than in the "Reader Pipe" list.
122
123o "Free-Block Circulation": Shows the number of torture structures
124 that have reached a given point in the pipeline. The first element
125 should closely correspond to the number of structures allocated,
126 the second to the number that have been removed from reader view,
127 and all but the last remaining to the corresponding number of
128 passes through a grace period. The last entry should be zero,
129 as it is only incremented if a torture structure's counter
130 somehow gets incremented farther than it should.
131
Paul E. McKenneyb2896d22006-10-04 02:17:03 -0700132Different implementations of RCU can provide implementation-specific
133additional information. For example, SRCU provides the following:
134
135 srcu-torture: rtc: f8cf46a8 ver: 355 tfle: 0 rta: 356 rtaf: 0 rtf: 346 rtmbe: 0
136 srcu-torture: Reader Pipe: 559738 939 0 0 0 0 0 0 0 0 0
137 srcu-torture: Reader Batch: 560434 243 0 0 0 0 0 0 0 0
138 srcu-torture: Free-Block Circulation: 355 354 353 352 351 350 349 348 347 346 0
139 srcu-torture: per-CPU(idx=1): 0(0,1) 1(0,1) 2(0,0) 3(0,1)
140
141The first four lines are similar to those for RCU. The last line shows
142the per-CPU counter state. The numbers in parentheses are the values
143of the "old" and "current" counters for the corresponding CPU. The
144"idx" value maps the "old" and "current" values to the underlying array,
145and is useful for debugging.
146
Paul E. McKenneya241ec62005-10-30 15:03:12 -0800147
148USAGE
149
150The following script may be used to torture RCU:
151
152 #!/bin/sh
153
154 modprobe rcutorture
155 sleep 100
156 rmmod rcutorture
Paul E. McKenney72e9bb52006-06-27 02:54:03 -0700157 dmesg | grep torture:
Paul E. McKenneya241ec62005-10-30 15:03:12 -0800158
159The output can be manually inspected for the error flag of "!!!".
160One could of course create a more elaborate script that automatically
Paul E. McKenney29766f12006-06-27 02:54:02 -0700161checked for such errors. The "rmmod" command forces a "SUCCESS" or
162"FAILURE" indication to be printk()ed.