We've been seeing some issues with ptrace signal delivery on some utrace enabled kernels. The issue might have been fixed already--we've seen it on Fedora Core 8 and a RedHat 2.6.18-33 kernel, but not in a RHEL 5.1 2.6.18-55 kernel. I'm wondering if anyone recognizes these symptoms as belonging to a specific bug, as I'd like to understand the issue to see if we can implement a user space work-around in our tool.
So far this has happened when our debugger process is attached to more than one process. At some point we should recieve notice of a signal delivery (such as SIGTRAP or SIGSEGV) via wait(), but instead get a notice of signal 0 being delivered. We can try to continue with PTRACE_CONT, but then recieve notice of another signal 0. After ~6000 signal 0/PTRACE_CONT we seem to get the original expected signal, but a subsequent PTRACE_GETREGS gives what looks like incorrect register values.
Does this ring any bells? If not, I'll try to build a test and do some debugging.
Thanks. -Matthew LeGendre
