[4/6] native/Linux: internal error if inferior disappears after stopped by breakpoint
Commit Message
If the inferior disappears just after it was stopped at a breakpoint,
GDB internal errors on next resume:
Executing on target: kill -9 11605 (timeout = 300)
spawn -ignore SIGHUP kill -9 11605
continue
Continuing.
/home/pedro/gdb/mygit/src/gdb/linux-nat.c:2590: internal-error: status_callback: Assertion `lp->status != 0' failed.
This is because the thread had stopped for a breakpoint, and had
already reported the event, so its ->status flag was cleared. The
lwp's stopped, etc., flags should only be cleared when we're sure the
LWP was successfully resumed (see PR gdb/15713, git 8817a6f2). So the
next resume hits an ESRCH error which throws before those flags are
cleared. GDB core prints the error, and ends up calling target_wait
to poll remaining events. We then trip on the assertion.
Fix this by bailing out earlier. GDBserver is already doing this.
A follow up patch will add a test that exercises this
(gdb.base/killed-outside.exp).
gdb/ChangeLog:
2015-03-06 Pedro Alves <palves@redhat.com>
* linux-nat.c (status_callback): Return early if the LWP has no
status pending.
---
gdb/linux-nat.c | 8 ++++----
1 file changed, 4 insertions(+), 4 deletions(-)
@@ -2574,6 +2574,9 @@ status_callback (struct lwp_info *lp, void *data)
if (!lp->resumed)
return 0;
+ if (!lwp_status_pending_p (lp))
+ return 0;
+
if (lp->stop_reason == TARGET_STOPPED_BY_SW_BREAKPOINT
|| lp->stop_reason == TARGET_STOPPED_BY_HW_BREAKPOINT)
{
@@ -2582,8 +2585,6 @@ status_callback (struct lwp_info *lp, void *data)
CORE_ADDR pc;
int discard = 0;
- gdb_assert (lp->status != 0);
-
pc = regcache_read_pc (regcache);
if (pc != lp->stop_pc)
@@ -2621,10 +2622,9 @@ status_callback (struct lwp_info *lp, void *data)
linux_resume_one_lwp (lp, lp->step, GDB_SIGNAL_0);
return 0;
}
- return 1;
}
- return lwp_status_pending_p (lp);
+ return 1;
}
/* Return non-zero if LP isn't stopped. */