The test case from the ARM Linux kernel thread does not work for me; GCC allocates the VFP use in main to d8-d15 (since main is not a leaf function). The signal handler is allocated d0-d7 to corrupt, so main does not notice. It may depend on the GCC version.
To force the register allocation, I coded a similar testcase in assembler. Like the original testcase, alarm(2) is used to schedule a signal. main is coded to call no functions in between setting d0 and checking it (since the ABI allows d0-d7 to be corrupted in this case). * If the signal handler manages to corrupt d0, a message is printed and the test terminates. * If no corruption is observed, main will simply loop forever and the test will not terminate. I get: $ gcc -o vfp-sigtest vfp-sigtest.s $ ./vfp-sigtest Waiting for d0 to be corrupted... <2 seconds later> d0 corrupted. ** Attachment added: "Test case for VFP signal handler corruption" http://launchpadlibrarian.net/38752212/vfp-sigtest.s -- VFP/NEON state is not preserved around signal handlers, causing state corruption between user processes https://bugs.launchpad.net/bugs/507503 You received this bug notification because you are a member of Ubuntu Bugs, which is subscribed to Ubuntu. -- ubuntu-bugs mailing list ubuntu-bugs@lists.ubuntu.com https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
