On Fri, Sep 26, 2008 at 08:34:12PM +0900, Izumi Tsutsui wrote: > martin%duskware.de@localhost wrote: > > Anyway, we should have certain benchmarks on revivesa as > regression tests (not for performance) as Andrew did on 1:1 > http://mail-index.NetBSD.org/tech-kern/2007/09/28/0014.html > and I'm afraid it can't be done in a reasonable period > for the next release. We do have regression tests, specifically the ones from 4.X. And revivesa is passing all of them. And has been for quite a while. We have a simple named regression test which is to run named w/o a configuration file. It should exit complaining about no config file. We have this "test" due to testing & messages on this list. In my test env, the 4.X named is almost always fine(*). Manuel's 3.X named (9.3.0) with a 3.X libpthread (in a chroot) is NOT fine. However the same named binary is fine when not in a chroot, i.e. with the 4.X libpthread. named is dying with an illegal instruction error, and I'm getting a core that gdb can't do much with. (*) by almost fine, I mean that after boot, all repeated invocations of /usr/sbin/named run fine. My boot logs show that the first invocation, at boot, crashed. I think the problem is that something's wrong with how user contexts are handled in upcall delivery. This _used_ to work, so I'm not sure what changed to trigger the problem.I am also not sure what's different to cause repeated applications with the 4.1 libpthread to not have problems that the 3.0 libpthread barfs on. I would appreciate debugging assistance. The exact specifics of upcall delivery are things that I'd hoped to leave as-was & have work. It is the one area I'm not too strong at. Note: AFAIK we have no panics or DOS regressions in the SA code relative to 4.X. As such, we could merge revivesa and continue debugging on HEAD. I've forgotten the specifics of the release plan we had in mind, but we coudl do this in order to un-block 6.0 feature work. Take care, Bill
Attachment:
pgpkhQdK06HNY.pgp
Description: PGP signature