[Bug gdb/27830] New: Attempt to resume already running thread

["andrew.burgess at embecosm dot com" <sourceware-bugzilla@sourceware.org>]

https://sourceware.org/bugzilla/show_bug.cgi?id=27830

            Bug ID: 27830
           Summary: Attempt to resume already running thread
           Product: gdb
           Version: HEAD
            Status: NEW
          Severity: normal
          Priority: P2
         Component: gdb
          Assignee: unassigned at sourceware dot org
          Reporter: andrew.burgess at embecosm dot com
  Target Milestone: ---

Created attachment 13423
  --> https://sourceware.org/bugzilla/attachment.cgi?id=13423&action=edit
reproducer

When using 'target remote', 'set non-stop on' and 'set displaced-stepping off',
if we step over an instruction that creates a new thread then GDB ends up
trying to resume a thread that is already running.

To reproduce, download the attached tar file then:

  $ tar -xf thr-resume-issue.tar.xz
  $ cd thr-resume-issue
  $ make
  .....snip....
  [New Thread 3878916.3878917]
  cmd.gdb:12: Error in sourced command file:
  PC register is not available
  **** FAILED ****

This test case is only going to work on an x86-64/Linux setup.

Notice the "PC register is not available" error.  This is caused by GDB trying
to read the $pc from the new running thread.

In more detail what happens is that GDB stops at the syscall instruction that
will spawn a new thread.  At this point the inferior is single threaded.

Next we 'stepi', as displaced stepping is off we do inplace stepping.

When the step is completed we end up in finish_step_over (infrun.c), this then
calls restart_threads.

In restart_threads we call update_thread_list and spot the new thread.  The
thread is created with executing == true and state == THREAD_RUNNING, but,
resumed == false.

Back in restart_threads we then loop over all threads and try to put the thread
"back to what they were trying to do back when we paused them for an in-line
step-over."

For our new thread we try to resume the thread (even though it's already
running), which requires a read of $pc, which causes the error we see.

-- 
You are receiving this mail because:
You are on the CC list for the bug.