Michel Salim wrote: > I have Chez Scheme installed from this tarball: > http://www.scheme.com/csv7.0a/pcsv7.0a-ti3le.tar.gz > > and it works fine up to 2.6.15-1.2054_FC5 as shipped with FC5, both > the i686 SMP kernel and the x86_64 kernel. > > When I rebooted to 2.6.16-1.2080_FC5, though, on both my i686 and > x86_64 machines I now get the following: > (5 #f "invalid memory reference") > > I have SELinux set to permissive + targeted, so what change in the > kernel could have caused this? Should I bugzilla this? It's possible that changes introduced with kernel 2080_FC5 to deal with bugzilla #162797 may be interacting with the application. [What is the address of the attempted bad access, and what is the memory map there?] First, try the application under the values 1, 3, 0, 11, and 9 of /proc/sys/kernel/exec-shield, when running 2080_FC5. That is: # echo 1 > /proc/sys/kernel/exec-shield ## note: as root $ run-the-application ## note: as ordinary user then # echo 3 > /proc/sys/kernel/exec-shield $ run-the-application then use 0, 11, and 9. [The default behavior of kernel 2054_FC5 is very close to the behavior of 2080_FC5 when exec-shield is 1.] If the problem persists under both 11 and 9, but not under any of 1, 3, and 0, then there may be a kernel bug, or the application might be making an assumption about virtual address space that is no longer true in kernel 2080_FC5. Set exec-shield back to 11 (the new default), run the application under strace, and attach the output from strace when you enter the problem in bugzilla. [Refer to bz #162797, please. Perhaps even re-open that one.] If the problem appears with any of exec-shield values 1, 3, or 0, then there is a good chance of a bug in the application. Run under strace, and look for colliding mmap(), etc. Does the application know about AT_SYSINFO_EHDR? -- -- fedora-test-list mailing list fedora-test-list@xxxxxxxxxx To unsubscribe: https://www.redhat.com/mailman/listinfo/fedora-test-list