Hi, >> As long as _dl_debug_state() has enough space to patch in the jump to our >> code > > A single 'ret' is a single byte, which isn't enough space to patch anything > (which was my point). > >> The dynamic linkers I looked seem to be built with enough code padding >> that there's space to replace the ret with a jump. > > You are likely saved by '-falign-functions=16', which leaves you 15 bytes > to play with. > > You can't count on it though, and I believe that I *have* seen dynamic > loaders where there literally was only one byte available.
Yeah, exactly. I'm slightly pragmatic here, because every once in a while we get a dead-lock. So I am looking for practical alternatives. The systems we use are aged (RHEL5 / x86_64) with little hope of picking up fresh work like Linux kernel or GLIBC. We still try to campaign for those for future, but we also need a practical solution for today. Regards, Lassi _______________________________________________ Libunwind-devel mailing list [email protected] http://lists.nongnu.org/mailman/listinfo/libunwind-devel
