On Do, 2006-07-06 at 23:36 +0200, Johannes Berg wrote: > On Thu, 2006-07-06 at 23:22 +0200, Sebastian Dröge wrote: > > > Ok, it was really helpful to change this as this gave me useful output > > for the crashes... see http://bugzilla.ximian.com/show_bug.cgi?id=78782 > > Ok. > > > Can you reproduce this? You need a gcc and glibc compiled with SSP for > > this. mono can be without SSP. > > I can try though I'm not too keen on recompiling my libc :)
Ok, after a second thought (it was really late yesterday) compiling gcc 4.1 with SSP is enough. Only the update of gcc (libgcc1) to a version with SSP broke mono. > Now, can you explain something to me? > The replacement I suggested (g_malloc -> codeman reserve) shouldn't make > any difference on ppc32. Was the stacktrace before from a ppc64 machine? > Or from a ppc32? All my tests were on a PPC32 (I said before that it worked for me there but I wasn't up to date on that machine...). Without the codeman stuff I get a segfault in the segfault handler which gives us then no stacktrace and one million segfaults. When running with gdb the first segfault has the same backtrace is the one I've pasted below in the bugreport. When using the codeman stuff instead of malloc the segfault handler doesn't segfault anymore (TlsGetValue() returns non-NULL) and the segfault handler of mono writes the correct backtrace (i.e. the same that gdb writes). Bye
signature.asc
Description: This is a digitally signed message part
