Re: detecting all possible deadlock conditions
Posted Jun 8, 2006 8:53 UTC (Thu) by jmansion
In reply to: Re: detecting all possible deadlock conditions
Parent article: The kernel lock validator
>In other words, we've reduced a highly probabilistic
>parallel QA (and manual code review) task to a
>comparatively _much_ simpler task of serial code coverage.
Hardly. If you can legitimately *ever* lock two objects of the same type then there will be an exception that turns off the 'false positive' that will also turn off checking whether you acquire A then B and also on some code paths B then A. Which is exactly the sort of deadlock you normally get in a database server, for example.
Which is not to say that the approach isn't useful, just that its not a panacea. It would have been handy to have a compare/contrast with FreeBSD's WITNESS facility.
to post comments)