Eraser: A Dynamic Data Race Detector for
Multithreaded Programs (Savage et al., 1997)
Presented by Scott Fleming
A Paper Presentation for
CSE 891 - Formal Methods in Software Development: Reliable
Computing With Threads
Fall Semester, 2008
Prepare for your exams
Study with the several resources on Docsity
Earn points to download
Earn points by helping other students or get them with a premium plan
Guidelines and tips
Prepare for your exams
Study with the several resources on Docsity
Earn points to download
Earn points by helping other students or get them with a premium plan
Eraser: A Dynamic Data Race Detector for Multithreaded Programs, Papers of Computer Science
Eraser is a dynamic data race detector for multithreaded programs that uses lockset algorithms to detect potential data races. It assumes lock-based synchronization and checks that all shared-memory accesses follow a consistent locking discipline. The lockset algorithm, its application, and three problematic practices related to data races. Eraser is implemented by automatically instrumenting the program binary with calls to the eraser runtime.
Typology: Papers
Pre 2010
1 / 18
This page cannot be seen from the preview
Don't miss anything!
Related documents
Partial preview of the text
Download Eraser: A Dynamic Data Race Detector for Multithreaded Programs and more Papers Computer Science in PDF only on Docsity!
Eraser: A Dynamic Data Race Detector for
Multithreaded Programs (Savage et al., 1997)
Presented by Scott Fleming
A Paper Presentation for CSE 891 - Formal Methods in Software Development: Reliable Computing With Threads Fall Semester, 2008
What Does Eraser Do?
Function: Dynamically detects the occurrence of data races in multithreaded programs
Data race: Occurs when ๎ (^) two or more threads concurrently access the same memory location, ๎ (^) at least one of the threads is writing, and ๎ (^) no explicit mechanism is used to prevent the accesses from being simultaneous
Dynamic detection: Detects as program executes ๎ (^) Occurrence of data race depends on what code is executed ๎ (^) Therefore, may depend on thread schedule ๎ (^) May miss potential data races
Producer-Consumer with General Race
Producer Consumer
<
q : Queue
empty() : Boolean full() : Boolean push(item : T) pull() : T
q.enqueue(item); for (;;) { if (!buf->full()) buf->push(produce()); }
How Does Eraser Do It?
Assumes: Lock-based synchronization
Approach: Checks that all shared-memory accesses follow a consistent locking discipline
Lock: Primitive synchronization object used for mutual exclusion (also called a mutex ) ๎ (^) States: available or owned (by a thread) ๎ (^) Operations: (atomic) lock and unlock
Locking discipline: Protocol of lock acquisition and release to prevent data races
Example Application of Lockset Algorithm
t1 t2 locks_held(t1) locks_held(t2) C(v) { } { } { l1, l2 } lock(l1) { l1 } ++v { l1 } unlock(l1) { } lock(l2) { l2 } ++v { } unlock(l2) { }
Three Problematic Practices
Initialization: Shared variables initialized without locking
Read-shared data: Shared constant variables
Read-write locks: Read-write locks allow multiple readers to access a shared variable, but allow only one writer
Solution to Read-Shared Data Problem
Only report races after an initialized variable has been written
State Transitions of a Shared Variable
Virgin
Shared
Exclusive
Shared- Modified
read/write, 1st thread write
read, new thread
read
write
write, new thread
read/write
Example of R/W Lock Problem
t1 t2 locks_held(t1) locks_held(t2) C(v) { } { } { l1 } rdlock(l1) { l1 } rdlock(l1) { l1 } ++v { l1 } unlock(l1) { } read(v) { l1 } unlock(l1) { }
Solution to Read-Write Locks Problem
Require that for each variable v , ๎ (^) some lock l protects v , and ๎ (^) l is held in some mode for every read of v
Definition: l protects v if l is held in write mode for every write of v
Example of R/W Lock Solution
t1 t2 l_held(t1) l_held(t2) C(v) { } { } { } { } { l1 } { l1 } { l1 } ++v { } unlock(l1) { } read(v) unlock(l1) { } { } { }
wl_held(t1) wl_held(t2)
rdlock(l1) rdlock(l1)
Implementation Details
Approach: Automatically instruments program binary with calls to Eraser runtime ๎ (^) memory loads and stores ๎ (^) lock acquires and releases ๎ (^) storage allocators
Warning messages: ๎ (^) Source file and line number where race discovered ๎ (^) Backtrace listing all active stack frames ๎ (^) Thread ID ๎ (^) Memory address ๎ (^) Type of memory access ๎ (^) Register values such as program counter and stack pointer