[rfa/testsuite] Signals vs handler entry-point

[rfa/testsuite] Signals vs handler entry-point

[Andrew Cagney]

[-- Attachment #1: Type: text/plain, Size: 409 bytes --]

Hello,

On GNU/Linux systems, when deliverying a signal, the process is resumed 
at the first instruction of the handler, and not the first instruction 
of the signal trampoline.  This checks that a breakpoint on that first 
handler instruction still works.

Tested on a i386 GNU/Linux, it 1738 kfailed when single-stepping.
Tested on a patched PPC/NetBSD, it passed (doesn't have that feature).

Ok?
Andrew


[-- Attachment #2: diffs --]
[-- Type: text/plain, Size: 4123 bytes --]

Index: ChangeLog
2004-08-30  Andrew Cagney  <cagney@gnu.org>

	* gdb.base/sigstep.exp (breakpoint_to_handler_entry)
	(skip_to_handler_entry): New procedures.  Test stepping into a
	handler when the breakpoint is at the handler's entry point.

Index: gdb.base/sigstep.exp
===================================================================
RCS file: /cvs/src/src/gdb/testsuite/gdb.base/sigstep.exp,v
retrieving revision 1.6
diff -p -u -r1.6 sigstep.exp
--- gdb.base/sigstep.exp	30 Aug 2004 16:59:45 -0000	1.6
+++ gdb.base/sigstep.exp	30 Aug 2004 17:37:42 -0000
@@ -235,6 +235,47 @@ skip_to_handler step
 skip_to_handler next
 skip_to_handler continue
 
+# Try stepping when there's a signal pending, and a breakpoint at the
+# handler's entry-point.  Should step into the signal handler stopping
+# at the entry-point.
+
+# Some systems (e.x., GNU/Linux as of 2004-08-30), when delivering a
+# signal, resume the process at the first instruction of the signal
+# handler and not the first instruction of the signal trampoline.  The
+# stack is constructed such that the signal handler still appears to
+# have been called by the trampoline code.  This test checks that it
+# is possible to stop the inferior, even at that first instruction.
+
+proc skip_to_handler_entry { i } {
+    global gdb_prompt
+    global infinite_loop
+    set prefix "$i to handler entry"
+    
+    # Run around to the done
+    set test "$prefix; resync"
+    gdb_test_multiple "continue" "$test" {
+	-re "done = 0.*$gdb_prompt " {
+	    pass "$test"
+	}
+	# other patterns can go here
+    }
+    
+    # Advance to the infinite loop
+    gdb_test "advance $infinite_loop" "" "$prefix; advance to infinite loop"
+
+    # Make the signal pending
+    sleep 1
+    
+    # Insert / remove the handler breakpoint.
+    gdb_test "break *handler" "" "$prefix; break handler"
+    gdb_test "$i" " handler .*" "$prefix; performing $i"
+    gdb_test "clear *handler" "" "$prefix; clear handler"
+}
+
+skip_to_handler_entry step
+skip_to_handler_entry next
+skip_to_handler_entry continue
+
 # Try stepping when there's a signal pending but no breakpoints.
 # Should skip the handler advancing to the next line.
 
@@ -302,6 +343,51 @@ breakpoint_to_handler step
 breakpoint_to_handler next
 breakpoint_to_handler continue
 
+# Try stepping when there's a signal pending, and a breakpoint at the
+# handler's entry instruction and a breakpoint at the current
+# instruction.  Should step into the signal handler and breakpoint at
+# that entry instruction.
+
+# Some systems (e.x., GNU/Linux as of 2004-08-30), when delivering a
+# signal, resume the process at the first instruction of the signal
+# handler and not the first instruction of the signal trampoline.  The
+# stack is constructed such that the signal handler still appears to
+# have been called by the trampoline code.  This test checks that it
+# is possible to stop the inferior, even at that first instruction.
+
+proc breakpoint_to_handler_entry { i } {
+    global gdb_prompt
+    global infinite_loop
+    set prefix "$i on breakpoint, to handler entry"
+    
+    # Run around to the done
+    set test "$prefix; resync"
+    gdb_test_multiple "continue" "$test" {
+	-re "done = 0.*$gdb_prompt " {
+	    pass "$test"
+	}
+	# other patterns can go here
+    }
+    
+    gdb_test "break $infinite_loop" "" "$prefix; break infinite loop"
+    gdb_test "break *handler" "" "$prefix; break handler"
+
+    # Continue to the infinite loop
+    gdb_test "continue" "while ..done.*" "$prefix; continue to infinite loop"
+
+    # Make the signal pending
+    sleep 1
+    
+    setup_kfail "i*86-*-*" gdb/1738
+    gdb_test "$i" " handler .*" "$prefix; performing $i"
+    gdb_test "clear $infinite_loop" "" "$prefix; clear infinite loop"
+    gdb_test "clear *handler" "" "$prefix; clear handler"
+}
+
+breakpoint_to_handler_entry step
+breakpoint_to_handler_entry next
+breakpoint_to_handler_entry continue
+
 # Try stepping when there's a signal pending, and a pre-existing
 # breakpoint at the current instruction, and no breakpoint in the
 # handler.  Should advance to the next line.

Re: [rfa/testsuite] Signals vs handler entry-point

[Michael Chastain]

Looks okay to me.

This patch is approved, with a 24-hour delay for anyone else
to object or ask for more time.

===

2004-08-30  Andrew Cagney  <cagney@gnu.org>

	* gdb.base/sigstep.exp (breakpoint_to_handler_entry)
	(skip_to_handler_entry): New procedures.  Test stepping into a
	handler when the breakpoint is at the handler's entry point.

Re: [rfa/testsuite] Signals vs handler entry-point

[Andrew Cagney]

> Looks okay to me.
> 
> This patch is approved, with a 24-hour delay for anyone else
> to object or ask for more time.
> 
> ===
> 
> 2004-08-30  Andrew Cagney  <cagney@gnu.org>
> 
> 	* gdb.base/sigstep.exp (breakpoint_to_handler_entry)
> 	(skip_to_handler_entry): New procedures.  Test stepping into a
> 	handler when the breakpoint is at the handler's entry point.

It's now in.

Andrew