- /* This function is used when following both the parent and child
- of a fork. In this case, the debugger clones itself. The original
- debugger follows the parent, the clone follows the child. The
- original detaches from the child, delivering a SIGSTOP to it to
- keep it from running away until the clone can attach itself.
+ /* Detach from the child. */
+ printf_unfiltered ("Detaching after fork from %s\n",
+ target_pid_to_str (pid_to_ptid (child_pid)));
+ hppa_require_detach (child_pid, 0);
+
+ /* The parent and child of a vfork share the same address space.
+ Also, on some targets the order in which vfork and exec events
+ are received for parent in child requires some delicate handling
+ of the events.
+
+ For instance, on ptrace-based HPUX we receive the child's vfork
+ event first, at which time the parent has been suspended by the
+ OS and is essentially untouchable until the child's exit or second
+ exec event arrives. At that time, the parent's vfork event is
+ delivered to us, and that's when we see and decide how to follow
+ the vfork. But to get to that point, we must continue the child
+ until it execs or exits. To do that smoothly, all breakpoints
+ must be removed from the child, in case there are any set between
+ the vfork() and exec() calls. But removing them from the child
+ also removes them from the parent, due to the shared-address-space
+ nature of a vfork'd parent and child. On HPUX, therefore, we must
+ take care to restore the bp's to the parent before we continue it.
+ Else, it's likely that we may not stop in the expected place. (The
+ worst scenario is when the user tries to step over a vfork() call;
+ the step-resume bp must be restored for the step to properly stop
+ in the parent after the call completes!)
+
+ Sequence of events, as reported to gdb from HPUX:
+
+ Parent Child Action for gdb to take
+ -------------------------------------------------------
+ 1 VFORK Continue child
+ 2 EXEC
+ 3 EXEC or EXIT
+ 4 VFORK
+
+ Now that the child has safely exec'd or exited, we must restore
+ the parent's breakpoints before we continue it. Else, we may
+ cause it run past expected stopping points. */
+
+ if (has_vforked)
+ reattach_breakpoints (parent_pid);
+ }
+ else
+ {
+ /* Needed to keep the breakpoint lists in sync. */
+ if (! has_vforked)
+ detach_breakpoints (child_pid);