Hi Please see the bug from the mentioned URL! PS: I know this is also because of some memory management "tricks" valgrind people do and they want to recode the memory management for this and other reasons but whould be nice if gentoo whould come with some fix/workarround as gentoo kernel makes the problem very worse (maybe gentoo kernel people can look into it too). Thanks!
It seems some of gentoo-sources' patches are causing a delay in m(un)map for valgrind. If anyone in the kernel team has some time to take a look at the kde bug report (see url field), then please do.
Has anyone confirmed that its actually the gentoo patchset making a difference (i.e. compared vanilla-sources-2.6.12 to gentoo-sources-2.6.12)?
Christian, I'm sure you don't mind being put on the cc of this bug as well. Can you answer dsd's question? Does the problem disappear with a vanilla kernel?
Heya, so far I can confirm that the behavior still occurs on * hardened-sources-2.6.11-r15, and * gentoo-source-2.6.12-gentoo-r6 I need to get a vanilla built up first; (man, I hate rebooting ;) hold on.
vanilla-sources-2.6.13_rc6 doesn't speedup VG either :(
Uh... sorry, I got confused from the original bug report that somehow it was a gentoo kernel only issue, I should have tested vanilla before posting here. I posted this bug because I thought it's a gentoo kernel issue or some kind of gentoo particular issue because so far on valgrind bugtracker there are 2 persons who complained about this, both having gentoo :) I also commented on the valgrind bug tracker that it affects vanilla, this should raise the priority of this problem (but still whould be interesting to find out why the FC kernel doesn't seem to display this bug).
Alrighty then, marking RESOLVED/UPSTREAM. Please add any extra non-gentoo specific comments to their bug report.
*** Bug 103898 has been marked as a duplicate of this bug. ***
Reopening to fix status
Should be fixed in valgrind-3.0.1, which was just added to portage. Thanks for reporting the problem, Dizzy.
are you sure that 3.0.1 has a fix for it? I mean, I remember some VG dev telling me, that this fix *will* require major changes to the address manager of VG, though, this wouldn't be part of 3.0.x-line of course (it's a seperate branch even). But this address manager rewrite isn't the only reason for rewriting it (think of the glibc independency in stage2 etc)... can someone approve it is working for him by now? (having 3.0.1 installed of course)
It's not fixed in valgrind's source, it's fixed in the ebuild by not building with PIE anymore.
I can confirm it works very nice now :) (only used it on my new project for 5 minutes to test it out and already found a bug hehe)