Download presentation
Presentation is loading. Please wait.
1
ADVERSARIAL MEMORY FOR DETECTING DESTRUCTIVE RACES Cormac Flanagan & Stephen Freund UC Santa Cruz Williams College PLDI 2010 Slides by Michelle Goodstein LBA Reading Group, June 2 2010
2
Motivation Multi-threaded programs often contain data races Hardware with relaxed memory consistency models may still behave like SC most of the time Hard to classify data races as benign or destructive New dynamic analysis technique: Adversarial Memory
3
Outline Motivation Review of Memory Models High-level idea of adversarial memory Will be skipping the formalisms; they are in the paper Results
4
Memory Models Sequential Consistency (SC): Once x is non-null, the conditional in Thread 2 will evaluate to true Java Relaxed Memory Model (JMM): Each of thread 2’s reads of x is independently null/non-null Initially: T2 reads x non-null, passes conditional Then x appears null, and x.draw() throws exception
5
Memory Models Trace: sequence of ops performed by threads Happens-Before Memory Model (HB): A read(x) operation A in a trace can return the value written by any write(x) operation B so long as B is either concurrent or happens before A (B doesn’t occur after A) no write C exists such that B < C < A in the trace (< :happens-before) Progressive Java Memory Model (PJ): A read(x) operation A in a trace can return the value written by any write(x) operation B so long as B executes before A in the trace No intervening write(x) C exists where B < C < A JMM: Happens-Before + Causality
6
Memory Models JMM, HBMM allow a potential future value to be read. PJMM only allows values def. in past to be read
7
Adversarial Memory Hardware is often SC-like even when it doesn’t guarantee SC Hard to see where races can truly be problematic Stress-test racy Java code Return old but still valid values (according to consistency model) Maintain write buffer to each shared variables involved in races On read Compute set of visible values that do not violate consistency model Return “worst case” according to heuristic
8
Adversarial Memory
9
Authors provide operational semantics Skipping here On reads, looks within write buffers for any write that could still be visible Only one write will be returned Use heuristics to choose “Most recent” write—very SC-like “Oldest” write—further from SC
10
Adversarial Memory Example per-thread vector clocks lock’s vector clock write buffer for location x: @ list “t0 writes value 13 to x at clock ” Available : 42@, 13@, 0@ Available : 42@
11
Adversarial Memory Heuristics Sequentially Consistent: Return most recent write Oldest: Return oldest value Intuition: staler the value, the likelier to cause problems Oldest-but-different Consider if(x != null) {x.draw();} What if x always reads null ? Gets out of infinite loop Random Random-but-different
12
Implementation JUMBLE: Java-based implementation, on RoadRunner framework Use precise race detector to discover racy shared vars Focus on one location at a time Special Cases Arrays: Sample indices, and only jumble accesses to a few indices Long/Double: Treat 8B as 2 non-atomic 4B accesses
13
Experimental Setup Examined 10 race conditions discovered by FASTTRACK Compared performance under 6 different memory implementations: No Jumble SC Oldest Oldest-but-different Random Random-but-different
14
Experimental Setup For each race & configuration 100 tests to detect how frequently race caused error Race on fields: jumbled reads from all instances of field Race on arrays: jumbled reads from all arrays at indices 0 & 1
15
Custom Benchmarks
16
Experimental Results: Efficacy
17
Some Discussion (More in Paper) montecarlo: Writes same value to global mtrt: threadcounter is incremented by parent, decremented by child. Never used elsewhere, so corruption of this variable does not matter.
18
Figure 8: null-ptr exception generated, since both null and non-null are available for x. Oldest fails due to infinite loop Figure 2: p can be initialized before p.x becomes non-zero, causing a divide-by-zero at line 17
19
Performance Results Performance of other heuristics similar to SC, except in degenerate cases EMPTY: 1.2x-1.5x (instrumentation) JUMBLE slowdown similar to EMPTY except: tsp, sor, moldyn Compression can greatly shrink size needed for write buffer
20
Eclipse Results FASTTRACK found 27 races Ran Jumble once/race 4 races: null ptr exceptions 4 races: non-deterministic reads, no bug Remaining fields: no non-deterministic reads detected Races on fields where the same value is written
21
Conclusions Data races are problematic Novel dynamic analysis to expose destructive data races Complements statically checking all valid SC interleavings
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.