[RFA/7.8] user breakpoint not inserted if software-single-step at same location
Commit Message
On 05/30/2014 02:26 PM, Joel Brobecker wrote:
>>>> - if there's still a non-sss breakpoint inserted at the
>>>> same address, then don't actually remove the breakpoint
>>>> off of the target, just wipe it from gdb's list.
>>>
>>> It seems to me that we'd need to merge your initial recommendation
>>> into your summary above, right?
>>
>> I admit I don't know what recommendation you're referring to. :-)
>
> Sorry! This one:
>
> | but we'll need to create/clone the location and its shadow buffer,
> | and then still handle the issue in the "remove" path.
Ah. Yes.
> I am wondering how to create that test, because it would be
> a little tricky. We need to set ourselves into a situation
> where we single-step out of a breakpoint with the second SSS
> breakpoint being at the same address as one of the user breakpoints,
> that second SSS not being the one that gets hit during that
> first single-step-out-of-breakpoint.
Yeah. Hmmm.
*thinks*
I have a very simple idea, around "jump" + "always-inserted".
E.g., with, where b+ indicates a user breakpoint:
00001 nop <- PC
b+ 00002 nop
- enable breakpoints always inserted mode
- step to 00002
- gdb removes the sss breakpoint.
- due to always inserted mode, gdb does not remove b+, but
due to the bug, it's actually no longer planted.
- the b+ breakpoint should be reported to the user.
- now do "jump $pc".
- expected:
The breakpoint should trigger immediately again.
- what we get on sss targets:
GDB loses control, and program runs to end.
Ah, I just went ahead and tried that against my by sss-on-x86
branch, and indeed it fails here, while it passes on
pristine mainline / hardware stepping.
8<----------
From e13bf4d64bf299111193a1f27a0bbc194d9b34f4 Mon Sep 17 00:00:00 2001
From: Pedro Alves <palves@redhat.com>
Date: Fri, 30 May 2014 16:52:36 +0100
Subject: [PATCH] test for sss breakpoints bug
---
gdb/testsuite/gdb.base/sss-bp-on-user-bp.c | 30 ++++++++++++++++
gdb/testsuite/gdb.base/sss-bp-on-user-bp.exp | 51 ++++++++++++++++++++++++++++
2 files changed, 81 insertions(+)
create mode 100644 gdb/testsuite/gdb.base/sss-bp-on-user-bp.c
create mode 100644 gdb/testsuite/gdb.base/sss-bp-on-user-bp.exp
Comments
> I have a very simple idea, around "jump" + "always-inserted".
>
> E.g., with, where b+ indicates a user breakpoint:
>
> 00001 nop <- PC
> b+ 00002 nop
>
> - enable breakpoints always inserted mode
> - step to 00002
> - gdb removes the sss breakpoint.
> - due to always inserted mode, gdb does not remove b+, but
> due to the bug, it's actually no longer planted.
> - the b+ breakpoint should be reported to the user.
> - now do "jump $pc".
> - expected:
> The breakpoint should trigger immediately again.
> - what we get on sss targets:
> GDB loses control, and program runs to end.
>
> Ah, I just went ahead and tried that against my by sss-on-x86
> branch, and indeed it fails here, while it passes on
> pristine mainline / hardware stepping.
That is very clever! WDYT about pushing this new testcase now?
On 05/30/2014 05:19 PM, Joel Brobecker wrote:
> That is very clever! WDYT about pushing this new testcase now?
Done. :-)
On 05/30/2014 11:57 PM, Pedro Alves wrote:
> +int
> +main (void)
> +{
> + /* Assume writes to integers compile to a single instruction. */
This assumption is wrong on arm at least.
> + volatile int i = 0;
> +
> + i = 1; /* set foo break here */
> + i = 2; /* set bar break here */
> + return 0;
> +}
Each line is compiled to two instructions.
27 i = 1; /* set foo break here */
0x0000025c <+20>: mov r3, #1
0x00000260 <+24>: str r3, [r11, #-8]
28 i = 2; /* set bar break here */
0x00000264 <+28>: mov r3, #2
0x00000268 <+32>: str r3, [r11, #-8]
> +# On software single-step targets, this step will want to momentarily
> +# place a single-step breakpoint over the bar breakpoint, and then
> +# remove it. But, a regular breakpoint it planted there already, and
> +# with always-inserted on, should remain planted when the step
> +# finishes.
> +gdb_test "si" "Breakpoint .* bar break .*"
this test will fail, because it still stops at "foo break".
On 06/03/2014 12:53 PM, Yao Qi wrote:
> On 05/30/2014 11:57 PM, Pedro Alves wrote:
>> +int
>> +main (void)
>> +{
>> + /* Assume writes to integers compile to a single instruction. */
>
> This assumption is wrong on arm at least.
Ah, thanks. We need to replace then with asm("nop") then.
Pedro Alves <palves@redhat.com> writes:
> Ah, thanks. We need to replace then with asm("nop") then.
nop isn't portable.
Andreas.
On 06/03/2014 01:12 PM, Andreas Schwab wrote:
> Pedro Alves <palves@redhat.com> writes:
>
>> Ah, thanks. We need to replace then with asm("nop") then.
>
> nop isn't portable.
Yes, but it doesn't matter what the instruction is as
long as it's a single instruction that doesn't do much.
For archs that don't have "nop" (like e.g., IA64), we can
just use #ifdef to pick another insn.
new file mode 100644
@@ -0,0 +1,30 @@
+/* This testcase is part of GDB, the GNU debugger.
+
+ Copyright 2014 Free Software Foundation, Inc.
+
+ This program is free software; you can redistribute it and/or modify
+ it under the terms of the GNU General Public License as published by
+ the Free Software Foundation; either version 3 of the License, or
+ (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ GNU General Public License for more details.
+
+ You should have received a copy of the GNU General Public License
+ along with this program. If not, see <http://www.gnu.org/licenses/>. */
+
+#include <signal.h>
+#include <unistd.h>
+
+int
+main (void)
+{
+ /* Assume writes to integers compile to a single instruction. */
+ volatile int i = 0;
+
+ i = 1; /* set foo break here */
+ i = 2; /* set bar break here */
+ return 0;
+}
new file mode 100644
@@ -0,0 +1,51 @@
+# Copyright 2014 Free Software Foundation, Inc.
+
+# This program is free software; you can redistribute it and/or modify
+# it under the terms of the GNU General Public License as published by
+# the Free Software Foundation; either version 3 of the License, or
+# (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU General Public License
+# along with this program. If not, see <http://www.gnu.org/licenses/>.
+
+# Test that removing a single-step breakpoint that is placed at the
+# same address as another regular breakpoint leaves the regular
+# breakpoint inserted.
+
+standard_testfile
+set executable ${testfile}
+
+if {[prepare_for_testing "failed to prepare" $testfile $srcfile debug]} {
+ return -1
+}
+
+if ![runto_main] then {
+ fail "Can't run to main"
+ return 0
+}
+
+gdb_breakpoint [gdb_get_line_number "set foo break here"]
+gdb_continue_to_breakpoint "first breakpoint" ".* set foo break here .*"
+
+gdb_breakpoint [gdb_get_line_number "set bar break here"]
+
+# So that GDB doesn't try to remove the regular breakpoint when the
+# step finishes.
+gdb_test_no_output "set breakpoint always-inserted on"
+
+# On software single-step targets, this step will want to momentarily
+# place a single-step breakpoint over the bar breakpoint, and then
+# remove it. But, a regular breakpoint it planted there already, and
+# with always-inserted on, should remain planted when the step
+# finishes.
+gdb_test "si" "Breakpoint .* bar break .*"
+
+# If the breakpoint is still correctly inserted, then this jump should
+# re-trigger it. Otherwise, GDB will lose control and the program
+# will exit.
+gdb_test "jump *\$pc" "Breakpoint .* bar break .*"