262 lines · plain
1============2LITMUS TESTS3============4 5CoRR+poonceonce+Once.litmus6 Test of read-read coherence, that is, whether or not two7 successive reads from the same variable are ordered.8 9CoRW+poonceonce+Once.litmus10 Test of read-write coherence, that is, whether or not a read11 from a given variable followed by a write to that same variable12 are ordered.13 14CoWR+poonceonce+Once.litmus15 Test of write-read coherence, that is, whether or not a write16 to a given variable followed by a read from that same variable17 are ordered.18 19CoWW+poonceonce.litmus20 Test of write-write coherence, that is, whether or not two21 successive writes to the same variable are ordered.22 23IRIW+fencembonceonces+OnceOnce.litmus24 Test of independent reads from independent writes with smp_mb()25 between each pairs of reads. In other words, is smp_mb()26 sufficient to cause two different reading processes to agree on27 the order of a pair of writes, where each write is to a different28 variable by a different process? This litmus test is forbidden29 by LKMM's propagation rule.30 31IRIW+poonceonces+OnceOnce.litmus32 Test of independent reads from independent writes with nothing33 between each pairs of reads. In other words, is anything at all34 needed to cause two different reading processes to agree on the35 order of a pair of writes, where each write is to a different36 variable by a different process?37 38ISA2+pooncelock+pooncelock+pombonce.litmus39 Tests whether the ordering provided by a lock-protected S40 litmus test is visible to an external process whose accesses are41 separated by smp_mb(). This addition of an external process to42 S is otherwise known as ISA2.43 44ISA2+poonceonces.litmus45 As below, but with store-release replaced with WRITE_ONCE()46 and load-acquire replaced with READ_ONCE().47 48ISA2+pooncerelease+poacquirerelease+poacquireonce.litmus49 Can a release-acquire chain order a prior store against50 a later load?51 52LB+fencembonceonce+ctrlonceonce.litmus53 Does a control dependency and an smp_mb() suffice for the54 load-buffering litmus test, where each process reads from one55 of two variables then writes to the other?56 57LB+poacquireonce+pooncerelease.litmus58 Does a release-acquire pair suffice for the load-buffering59 litmus test, where each process reads from one of two variables then60 writes to the other?61 62LB+poonceonces.litmus63 As above, but with store-release replaced with WRITE_ONCE()64 and load-acquire replaced with READ_ONCE().65 66LB+unlocklockonceonce+poacquireonce.litmus67 Does a unlock+lock pair provides ordering guarantee between a68 load and a store?69 70MP+onceassign+derefonce.litmus71 As below, but with rcu_assign_pointer() and an rcu_dereference().72 73MP+polockmbonce+poacquiresilsil.litmus74 Protect the access with a lock and an smp_mb__after_spinlock()75 in one process, and use an acquire load followed by a pair of76 spin_is_locked() calls in the other process.77 78MP+polockonce+poacquiresilsil.litmus79 Protect the access with a lock in one process, and use an80 acquire load followed by a pair of spin_is_locked() calls81 in the other process.82 83MP+polocks.litmus84 As below, but with the second access of the writer process85 and the first access of reader process protected by a lock.86 87MP+poonceonces.litmus88 As below, but without the smp_rmb() and smp_wmb().89 90MP+pooncerelease+poacquireonce.litmus91 As below, but with a release-acquire chain.92 93MP+porevlocks.litmus94 As below, but with the first access of the writer process95 and the second access of reader process protected by a lock.96 97MP+unlocklockonceonce+fencermbonceonce.litmus98 Does a unlock+lock pair provides ordering guarantee between a99 store and another store?100 101MP+fencewmbonceonce+fencermbonceonce.litmus102 Does a smp_wmb() (between the stores) and an smp_rmb() (between103 the loads) suffice for the message-passing litmus test, where one104 process writes data and then a flag, and the other process reads105 the flag and then the data. (This is similar to the ISA2 tests,106 but with two processes instead of three.)107 108R+fencembonceonces.litmus109 This is the fully ordered (via smp_mb()) version of one of110 the classic counterintuitive litmus tests that illustrates the111 effects of store propagation delays.112 113R+poonceonces.litmus114 As above, but without the smp_mb() invocations.115 116SB+fencembonceonces.litmus117 This is the fully ordered (again, via smp_mb() version of store118 buffering, which forms the core of Dekker's mutual-exclusion119 algorithm.120 121SB+poonceonces.litmus122 As above, but without the smp_mb() invocations.123 124SB+rfionceonce-poonceonces.litmus125 This litmus test demonstrates that LKMM is not fully multicopy126 atomic. (Neither is it other multicopy atomic.) This litmus test127 also demonstrates the "locations" debugging aid, which designates128 additional registers and locations to be printed out in the dump129 of final states in the herd7 output. Without the "locations"130 statement, only those registers and locations mentioned in the131 "exists" clause will be printed.132 133S+poonceonces.litmus134 As below, but without the smp_wmb() and acquire load.135 136S+fencewmbonceonce+poacquireonce.litmus137 Can a smp_wmb(), instead of a release, and an acquire order138 a prior store against a subsequent store?139 140WRC+poonceonces+Once.litmus141WRC+pooncerelease+fencermbonceonce+Once.litmus142 These two are members of an extension of the MP litmus-test143 class in which the first write is moved to a separate process.144 The second is forbidden because smp_store_release() is145 A-cumulative in LKMM.146 147Z6.0+pooncelock+pooncelock+pombonce.litmus148 Is the ordering provided by a spin_unlock() and a subsequent149 spin_lock() sufficient to make ordering apparent to accesses150 by a process not holding the lock?151 152Z6.0+pooncelock+poonceLock+pombonce.litmus153 As above, but with smp_mb__after_spinlock() immediately154 following the spin_lock().155 156Z6.0+pooncerelease+poacquirerelease+fencembonceonce.litmus157 Is the ordering provided by a release-acquire chain sufficient158 to make ordering apparent to accesses by a process that does159 not participate in that release-acquire chain?160 161A great many more litmus tests are available here:162 163 https://github.com/paulmckrcu/litmus164 165==================166LITMUS TEST NAMING167==================168 169Litmus tests are usually named based on their contents, which means that170looking at the name tells you what the litmus test does. The naming171scheme covers litmus tests having a single cycle that passes through172each process exactly once, so litmus tests not fitting this description173are named on an ad-hoc basis.174 175The structure of a litmus-test name is the litmus-test class, a plus176sign ("+"), and one string for each process, separated by plus signs.177The end of the name is ".litmus".178 179The litmus-test classes may be found in the infamous test6.pdf:180https://www.cl.cam.ac.uk/~pes20/ppc-supplemental/test6.pdf181Each class defines the pattern of accesses and of the variables accessed.182For example, if the one process writes to a pair of variables, and183the other process reads from these same variables, the corresponding184litmus-test class is "MP" (message passing), which may be found on the185left-hand end of the second row of tests on page one of test6.pdf.186 187The strings used to identify the actions carried out by each process are188complex due to a desire to have short(er) names. Thus, there is a tool to189generate these strings from a given litmus test's actions. For example,190consider the processes from SB+rfionceonce-poonceonces.litmus:191 192 P0(int *x, int *y)193 {194 int r1;195 int r2;196 197 WRITE_ONCE(*x, 1);198 r1 = READ_ONCE(*x);199 r2 = READ_ONCE(*y);200 }201 202 P1(int *x, int *y)203 {204 int r3;205 int r4;206 207 WRITE_ONCE(*y, 1);208 r3 = READ_ONCE(*y);209 r4 = READ_ONCE(*x);210 }211 212The next step is to construct a space-separated list of descriptors,213interleaving descriptions of the relation between a pair of consecutive214accesses with descriptions of the second access in the pair.215 216P0()'s WRITE_ONCE() is read by its first READ_ONCE(), which is a217reads-from link (rf) and internal to the P0() process. This is218"rfi", which is an abbreviation for "reads-from internal". Because219some of the tools string these abbreviations together with space220characters separating processes, the first character is capitalized,221resulting in "Rfi".222 223P0()'s second access is a READ_ONCE(), as opposed to (for example)224smp_load_acquire(), so next is "Once". Thus far, we have "Rfi Once".225 226P0()'s third access is also a READ_ONCE(), but to y rather than x.227This is related to P0()'s second access by program order ("po"),228to a different variable ("d"), and both accesses are reads ("RR").229The resulting descriptor is "PodRR". Because P0()'s third access is230READ_ONCE(), we add another "Once" descriptor.231 232A from-read ("fre") relation links P0()'s third to P1()'s first233access, and the resulting descriptor is "Fre". P1()'s first access is234WRITE_ONCE(), which as before gives the descriptor "Once". The string235thus far is thus "Rfi Once PodRR Once Fre Once".236 237The remainder of P1() is similar to P0(), which means we add238"Rfi Once PodRR Once". Another fre links P1()'s last access to239P0()'s first access, which is WRITE_ONCE(), so we add "Fre Once".240The full string is thus:241 242 Rfi Once PodRR Once Fre Once Rfi Once PodRR Once Fre Once243 244This string can be given to the "norm7" and "classify7" tools to245produce the name:246 247 $ norm7 -bell linux-kernel.bell \248 Rfi Once PodRR Once Fre Once Rfi Once PodRR Once Fre Once | \249 sed -e 's/:.*//g'250 SB+rfionceonce-poonceonces251 252Adding the ".litmus" suffix: SB+rfionceonce-poonceonces.litmus253 254The descriptors that describe connections between consecutive accesses255within the cycle through a given litmus test can be provided by the herd7256tool (Rfi, Po, Fre, and so on) or by the linux-kernel.bell file (Once,257Release, Acquire, and so on).258 259To see the full list of descriptors, execute the following command:260 261 $ diyone7 -bell linux-kernel.bell -show edges262