On 25.04.14 16:52:05, Peter Zijlstra wrote: > But no, I don't think that helps, its still true that the moment you get > a fd another thread can immediately close(). That would drop the last > ref and free it, meanwhile perf_event_open() is happily poking at it. > > Now I think you could cure this by adding an extra ref before calling > your perf_get_fd() and dropping that extra ref at the end, where we used > to have fd_install().
Yes, right. I have a solution now which increments the event's ref count before creating the file descriptor using try_get_event()/ put_event(). The patch also does not remove get_unused_fd_flags() and the err_fd error handler. Have an update already of a rebase version but still need to test it. Would it be ok to split the patch set and send in a first step only the first 4 patches that refactor the perf mmap code? Thanks, -Robert -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
