solved!!! VDSO support, 2.6.20-1.2925 hangs after unmounting old proc and sys

Roland McGrath roland at redhat.com
Thu Apr 12 21:58:24 UTC 2007


I may have missed some of the context, so forgive me if I'm confused.  It
sounds like the report was of CONFIG_COMPAT_VDSO=y causing problems, the
problem being fixed by using CONFIG_COMPAT_VDSO=n.  Is that right, or inverted?

CONFIG_COMPAT_VDSO=y has never been set in Fedora configs AFAIK.  It is not
desireable.  It's been cleaned up and fixed upstream, some by me and some
not by me, fairly recently (probably after 2.6.19, don't really remember).
It may well never have been reconciled with the exec-shield patch, and I
wouldn't lay odds on whether the current upstream code jibes or not, though
I don't know of any interaction problems there off hand.  

With the very latest code (certainly after 2.6.20), now perhaps in -mm and
not yet all the way in upstream, it will become possible to compile in
support for the compatibility mode without undesireable effects when it's
not enabled at runtime (by "vdso=2" or a sysctl equivalent).  When those
changes go in, it will make sense to compile in the compatibility support
but not enable it at boot, and this will also likely be the default
upstream configuration.

As to the state of some existing or older Fedora kernel, I don't really
know off hand (and won't be in a position to try any debugging of my own
until I get home next week).  print-fatal-signals might shed some light.


Thanks,
Roland




More information about the Fedora-kernel-list mailing list