|
From: Julian S. <js...@ac...> - 2005-11-01 20:25:35
|
On Tuesday 01 November 2005 19:48, Tom Hughes wrote: > In message <Pin...@ch...> > > Nicholas Nethercote <nj...@cs...> wrote: > > This machine exhibits a lot of test noise -- a number of tests fail and > > then work again for no apparent reason, and not just the common such ones > > (ie. 'tls' and 'yield'). Tom, any idea why? > > I believe the main cause of this is that because the sysinfo page is > placed at a random address by the kernel it was occasionally placed > at the preferred load address of ld.so which then stopped us mapping > the interpreter. > > What should happen in that case is that the interpreter is mapped at > a different address instead, and I have just committed a fix that > should achieve that. Both 'tls' and 'yield' are sensitive to scheduling. Yield figures out whether a spinloop doing 'rep nop's progresses more slowly than one that doesn't (and is a bit nonsensical). 'tls' I'm less clear about but I did look at it before and thought it might be susceptible to races. J |