|
From: Bart V. A. <bar...@gm...> - 2006-08-12 06:58:15
|
For who is interested in data-race detection, the conclusion of Muhlenfeld's paper points out a fundamental limitation of Helgrind and other tools based on the Eraser algorithm: these tools do not recognize user-built synchronization constructs. E.g. in many applications reader-writer locks are implemented as a mutex + counter instead of calling the pthread_rwlock_...() functions. >From the conclusion: Further improvements could have a similar impact on the detection abilities of the data race checker, but require more effort. The weaknesses with regard to common multi-threading patterns should be addressed. Common concurrent patterns often rely on higher level constructs for synchronization that the lock-set algorithm is unaware of. On 8/12/06, sv...@va... <sv...@va...> wrote: > > Author: njn > Date: 2006-08-12 05:33:35 +0100 (Sat, 12 Aug 2006) > New Revision: 293 > > Log: > Added two papers by others about Valgrind. > > Added: > trunk/docs/muehlenfeld2006.pdf > trunk/docs/newsome2005.pdf > Modified: > trunk/docs/pubs.html ... + <li><p> > + <b><a href="/docs/muehlenfeld2006.pdf">Fault Detection in Multi-Threaded > C++ > + Server Applications.</a><br> > + Arndt Muehlenfeld and Franz Wotawa.<br> > + Informal Proceedings of the International Workshop on Multithreading in > + Hardware and Software (TV06), Seattle, Washington, USA, August > 2006.</b><br> > + This paper is about some improvements to Helgrind, the data-race > detector. > + </p></li> > |