Re: [PATCH] Honour SIGILL and SIGSEGV in cancel breakpoint

[Yao Qi <yao at codesourcery dot com>]

Pedro Alves <palves@redhat.com> writes:

>> count_events_callback and select_event_lwp_callback in GDBServer need to
>> honour SIGILL and SIGSEGV too.  I write a patch to call
>> lp_status_is_sigtrap_like_event in them, but regression test result
>> isn't changed, which is a surprise to me.  I thought some fails should
>> be fixed.  I'll look into it deeply.
>
> Maybe you're getting lucky with scheduling.
> pthreads.exp and schedlock.exp I think are the most sensitive to this.

I run them ten times, the results aren't changed.

>
> See:
>  https://www.sourceware.org/ml/gdb-patches/2001-06/msg00250.html

Randomly selecting event lwp was added in the url above you gave, to
prevent the starvation of threads.  However, in my configuration
(arm-linux with SIGILL), event lwp selection does nothing, but no test
fails are caused.  GDBserver processes events like this:

 1. When GDBServer gets a breakpoint event from waitpid (-1, ),
 2. GDBserver will stop_all_lwps, in which wait_for_sigstop will drain
 all pending reports from kernel.
 3. GDBserver selects one lwp and cancels the breakpoint on the rest.  If
 event lwp selection does nothing, it is the lwp GDBserver gets in step 1.
 4. GDBserver steps over the breakpoint, and resumes all the threads.
 Go back to step 1, wait until any threads hit breakpoint,

As we can see, if waitpid (-1, ) (in step #1) returns event lwp randomly,
we don't have to randomly select event lwp again in step #3.  IMO, it is
naturally random that one thread hits the breakpoint first in a
multi-thread program.  That is the reason why no test fails are caused
without event lwp selection in my experiments.  IOW, on the platform
that waitpid (-1, ) returns event lwp randomly, we don't need such lwp
random selection at all.  However, if waitpid kernel implementation
always iterate over a list children in the fixed order, it is possible
that event of the lwp in the front of the list is reported and the rest
lwps may be starved.  In this case, we still have to reply on random
selection inside GDB/GDBserver to avoid starvation.

The patch below is updated to call lp_status_maybe_breakpoint in both
breakpoint cancellation and event lwp selection.

-- 
Yao (éå)

Subject: [PATCH] Honour SIGILL and SIGSEGV in cancel breakpoint and event lwp selection

I see the following fail on arm-none-linux-gnueabi testing,

(gdb) continue^M
Continuing.^M
^M
Program received signal SIGILL, Illegal instruction.^M
[Switching to Thread 1003]^M
handler (signo=10) at
/scratch/yqi/arm-none-linux-gnueabi/src/gdb-trunk/gdb/testsuite/gdb.threads/sigstep-threads.c:33^M
33        tgkill (getpid (), gettid (), SIGUSR1);       /* step-2 */^M
(gdb) FAIL: gdb.threads/sigstep-threads.exp: continue

the cause is that GDBserver doesn't cancel the breakpoint if the stop
signal is SIGILL.  The kernel used here is a little old, 2.6.x, and
doesn't translate SIGILL to SIGTRAP when program hits breakpoint
instruction (which is an illegal instruction actually).  GDB and
GDBserver can translate SIGILL to SIGTRAP under certain circumstance,
so it is not a problem here.  See gdbserver/linux-low.c:linux_wait_1

  /* If this event was not handled before, and is not a SIGTRAP, we
     report it.  SIGILL and SIGSEGV are also treated as traps in case
     a breakpoint is inserted at the current PC.  If this target does
     not support internal breakpoints at all, we also report the
     SIGTRAP without further processing; it's of no concern to us.  */
  maybe_internal_trap
    = (supports_breakpoints ()
       && (WSTOPSIG (w) == SIGTRAP
	   || ((WSTOPSIG (w) == SIGILL
		|| WSTOPSIG (w) == SIGSEGV)
	       && (*the_low_target.breakpoint_at) (event_child->stop_pc))));

However, SIGILL and SIGSEGV is not considered when cancelling
breakpoint, which causes the fail above.  That is, when GDB is doing
software single step on address ADDR, both thread A and thread B hits the
software single step breakpoint, and get SIGILL.  GDB selects the event
from thread A, removes the software single step breakpoint, and resume
the program.  The event (SIGILL) from thread B is reported to GDB, but
GDB doesn't regard this SIGILL as SIGTRAP, because the breakpoint on
address ADDR was removed, so GDB reports "Program received signal
SIGILL".

The patch is to allow calling cancel_breakpoint if the signal is
SIGILL and SIGSEGV.  This patch fixes the fail above.  Likewise, event
lwp selection should honour SIGILL and SIGSEGV too.

gdb/gdbserver:

2014-09-23  Yao Qi  <yao@codesourcery.com>

	* linux-low.c (lp_status_maybe_breakpoint): New function.
	(linux_low_filter_event): Call lp_status_maybe_breakpoint.
	(count_events_callback): Likewise.
	(select_event_lwp_callback): Likewise.
	(cancel_breakpoints_callback): Likewise.

diff --git a/gdb/gdbserver/linux-low.c b/gdb/gdbserver/linux-low.c
index 705edde..2f860e9 100644
--- a/gdb/gdbserver/linux-low.c
+++ b/gdb/gdbserver/linux-low.c
@@ -1739,6 +1739,20 @@ cancel_breakpoint (struct lwp_info *lwp)
   return 0;
 }
 
+/* Return true if the event in LP may be caused by breakpoint.  */
+
+static int
+lp_status_maybe_breakpoint (struct lwp_info *lp)
+{
+  return (lp->status_pending_p
+	  && WIFSTOPPED (lp->status_pending)
+	  && (WSTOPSIG (lp->status_pending) == SIGTRAP
+	      /* SIGILL and SIGSEGV are also treated as traps in case a
+		 breakpoint is inserted at the current PC.  */
+	      || WSTOPSIG (lp->status_pending) == SIGILL
+	      || WSTOPSIG (lp->status_pending) == SIGSEGV));
+}
+
 /* Do low-level handling of the event, and check if we should go on
    and pass it to caller code.  Return the affected lwp if we are, or
    NULL otherwise.  */
@@ -1936,7 +1950,7 @@ linux_low_filter_event (ptid_t filter_ptid, int lwpid, int wstat)
 		 the core before this one is handled.  All-stop always
 		 cancels breakpoint hits in all threads.  */
 	      if (non_stop
-		  && WSTOPSIG (wstat) == SIGTRAP
+		  && lp_status_maybe_breakpoint (child)
 		  && cancel_breakpoint (child))
 		{
 		  /* Throw away the SIGTRAP.  */
@@ -2197,9 +2211,7 @@ count_events_callback (struct inferior_list_entry *entry, void *data)
      should be reported to GDB.  */
   if (thread->last_status.kind == TARGET_WAITKIND_IGNORE
       && thread->last_resume_kind != resume_stop
-      && lp->status_pending_p
-      && WIFSTOPPED (lp->status_pending)
-      && WSTOPSIG (lp->status_pending) == SIGTRAP
+      && lp_status_maybe_breakpoint (lp)
       && !breakpoint_inserted_here (lp->stop_pc))
     (*count)++;
 
@@ -2237,9 +2249,7 @@ select_event_lwp_callback (struct inferior_list_entry *entry, void *data)
   /* Select only resumed LWPs that have a SIGTRAP event pending. */
   if (thread->last_resume_kind != resume_stop
       && thread->last_status.kind == TARGET_WAITKIND_IGNORE
-      && lp->status_pending_p
-      && WIFSTOPPED (lp->status_pending)
-      && WSTOPSIG (lp->status_pending) == SIGTRAP
+      && lp_status_maybe_breakpoint (lp)
       && !breakpoint_inserted_here (lp->stop_pc))
     if ((*selector)-- == 0)
       return 1;
@@ -2271,9 +2281,7 @@ cancel_breakpoints_callback (struct inferior_list_entry *entry, void *data)
 
   if (thread->last_resume_kind != resume_stop
       && thread->last_status.kind == TARGET_WAITKIND_IGNORE
-      && lp->status_pending_p
-      && WIFSTOPPED (lp->status_pending)
-      && WSTOPSIG (lp->status_pending) == SIGTRAP
+      && lp_status_maybe_breakpoint (lp)
       && !lp->stepping
       && !lp->stopped_by_watchpoint
       && cancel_breakpoint (lp))