[PATCH 0/2] [GDBserver] Fix compiling conditional expressions accessing registers

[PATCH 0/2] [GDBserver] Fix compiling conditional expressions accessing registers

[Pierre Langlois]

Hi all,

I noticed compiled conditions for fast tracepoints failed to read
registers.  We can uncover the issue by compiling and executing the
following expression: `$rip == *set_point' on x86_64, with `set_point'
being the tracepoint symbol.  The expression should always be true but this
shows it isn't:

~~~
(gdb) ftrace set_point if $rip == *set_point
Fast tracepoint 2 at 0x4006b9: ...
(gdb) tstart
(gdb) continue
Continuing.

Breakpoint 3, end () at ..
(gdb) tstatus
Trace is running on the target.
Collected 0 trace frames.
^^^^^^^^^^^^^^^^^^^^^^^^^
Trace buffer has 5242880 bytes of 5242880 bytes free (0% full).
Trace will stop if GDB disconnects.
Not looking at any trace frame.
Trace started at 1441727485.544642 secs.
~~~

The following patches address this particular issue and adds a test case.
However, the test cases uncovered other issues which have been marked as
known failures, linking to PR/18955.

I have ran the test suite on both x86_64-linux and i386-linux, in a remote
configuration.

Thanks,
Pierre

Pierre Langlois (2):
  [GDBserver] Fix compiling conditional expressions accessing registers
  [testsuite] Add test case for tracepoints with conditions

 gdb/gdbserver/linux-amd64-ipa.c             |   9 --
 gdb/gdbserver/linux-i386-ipa.c              |  15 ---
 gdb/gdbserver/linux-x86-low.c               |   4 +-
 gdb/gdbserver/tracepoint.c                  |  42 ++++---
 gdb/gdbserver/tracepoint.h                  |   9 +-
 gdb/testsuite/gdb.trace/trace-condition.c   |  66 +++++++++++
 gdb/testsuite/gdb.trace/trace-condition.exp | 167 ++++++++++++++++++++++++++++
 7 files changed, 263 insertions(+), 49 deletions(-)
 create mode 100644 gdb/testsuite/gdb.trace/trace-condition.c
 create mode 100644 gdb/testsuite/gdb.trace/trace-condition.exp

-- 
2.4.6

[PATCH 2/2] [testsuite] Add test case for tracepoints with conditions

[Pierre Langlois]

This patch adds a test case for tracepoints with a condition expression.
Each case will test a condition against the number of frames that should
have been traced.  Some of these tests fail on x86_64 and others on
i386, which have been marked as known failures for now, see PR/18955.

gdb/testsuite/ChangeLog:

	* gdb.trace/trace-condition.c: New file.
	* gdb.trace/trace-condition.exp: New file.
---
 gdb/testsuite/gdb.trace/trace-condition.c   |  66 +++++++++++
 gdb/testsuite/gdb.trace/trace-condition.exp | 167 ++++++++++++++++++++++++++++
 2 files changed, 233 insertions(+)
 create mode 100644 gdb/testsuite/gdb.trace/trace-condition.c
 create mode 100644 gdb/testsuite/gdb.trace/trace-condition.exp

diff --git a/gdb/testsuite/gdb.trace/trace-condition.c b/gdb/testsuite/gdb.trace/trace-condition.c
new file mode 100644
index 0000000..2e965c9
--- /dev/null
+++ b/gdb/testsuite/gdb.trace/trace-condition.c
@@ -0,0 +1,66 @@
+/* This testcase is part of GDB, the GNU debugger.
+
+   Copyright 2011-2015 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/>.  */
+
+#ifdef SYMBOL_PREFIX
+#define SYMBOL(str)     SYMBOL_PREFIX #str
+#else
+#define SYMBOL(str)     #str
+#endif
+
+int globvar;
+
+static void
+begin (void)
+{
+}
+
+/* Called from asm.  */
+static void __attribute__((used))
+func (void)
+{
+}
+
+static void
+marker (int anarg)
+{
+  /* `set_point' is the label at which to set a fast tracepoint.  The
+     insn at the label must be large enough to fit a fast tracepoint
+     jump.  */
+  asm ("    .global " SYMBOL (set_point) "\n"
+       SYMBOL (set_point) ":\n"
+#if (defined __x86_64__ || defined __i386__)
+       "    call " SYMBOL (func) "\n"
+#endif
+       );
+}
+
+static void
+end (void)
+{
+}
+
+int
+main ()
+{
+  begin ();
+
+  for (globvar = 1; globvar < 11; ++globvar)
+    marker (globvar * 100);
+
+  end ();
+  return 0;
+}
diff --git a/gdb/testsuite/gdb.trace/trace-condition.exp b/gdb/testsuite/gdb.trace/trace-condition.exp
new file mode 100644
index 0000000..8bfd34a
--- /dev/null
+++ b/gdb/testsuite/gdb.trace/trace-condition.exp
@@ -0,0 +1,167 @@
+# Copyright 2011-2015 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/>.
+
+load_lib "trace-support.exp"
+
+standard_testfile
+set executable $testfile
+set expfile $testfile.exp
+
+# Some targets have leading underscores on assembly symbols.
+set additional_flags [gdb_target_symbol_prefix_flags]
+
+if [is_amd64_regs_target] {
+    set pcreg "\$rip"
+} elseif [is_x86_like_target] {
+    set pcreg "\$eip"
+} else {
+    set pcreg "\$pc"
+}
+
+if [prepare_for_testing $expfile $executable $srcfile \
+	[list debug $additional_flags]] {
+    untested "failed to prepare for trace tests"
+    return -1
+}
+
+if ![runto_main] {
+    fail "Can't run to main to check for trace support"
+    return -1
+}
+
+if ![gdb_target_supports_trace] {
+    unsupported "target does not support trace"
+    return -1
+}
+
+set libipa [get_in_proc_agent]
+gdb_load_shlibs $libipa
+
+# Can't use prepare_for_testing, because that splits compiling into
+# building objects and then linking, and we'd fail with "linker input
+# file unused because linking not done" when building the object.
+
+if { [gdb_compile "$srcdir/$subdir/$srcfile" $binfile \
+	  executable [list debug $additional_flags shlib=$libipa] ] != "" } {
+    untested "failed to compile ftrace tests"
+    return -1
+}
+
+clean_restart ${executable}
+
+if ![runto_main] {
+    fail "Can't run to main for ftrace tests"
+    return 0
+}
+
+gdb_reinitialize_dir $srcdir/$subdir
+
+if { [gdb_test "info sharedlibrary" ".*${libipa}.*" "IPA loaded"] != 0 } {
+    untested "Could not find IPA lib loaded"
+    return 1
+}
+
+proc test_tracepoints { trace_command condition num_frames { kfail_proc 0 } } {
+    global executable
+
+    clean_restart ${executable}
+
+    if ![runto_main] {
+	fail "Can't run to main for ftrace tests"
+	return 0
+    }
+
+    gdb_test "break begin" ".*" ""
+
+    gdb_test "break end" ".*" ""
+
+    gdb_test "${trace_command} set_point if ${condition}" "\(Fast t|T\)racepoint .*" \
+	"${trace_command} with condition $condition"
+
+    gdb_test "continue" \
+	".*Breakpoint \[0-9\]+, begin .*" \
+	"advance to trace begin"
+
+    gdb_test_no_output "tstart" "start trace experiment"
+
+    gdb_test_multiple "continue" "advance through tracing" {
+	-re ".*Breakpoint \[0-9\]+, end .*" {
+	    pass "advance through tracing"
+	}
+	-re "Program received signal SIGSEGV, Segmentation fault\\..*" {
+	    if { $kfail_proc != 0 } {
+		$kfail_proc $trace_command
+	    }
+	    fail "advance through tracing"
+	}
+    }
+
+    if { $kfail_proc != 0 } {
+	$kfail_proc $trace_command
+    }
+    gdb_test "tstatus" \
+	".*Trace .*Collected $num_frames .*" \
+	"check $num_frames frames were collected."
+
+    gdb_test "tstop" "" ""
+}
+
+# These callbacks identify known failures for certain architectures.  They
+# are called either if GDBserver crashes or has not traced the correct
+# number of frames.
+
+proc 18955_x86_64_failure { trace_command } {
+    if { $trace_command == "ftrace" } {
+	setup_kfail "gdb/18955" "x86_64-*-linux*"
+    }
+}
+
+proc 18955_i386_failure { trace_command } {
+    if { $trace_command == "ftrace" } {
+	setup_kfail "gdb/18955" "i\[34567\]86-*-*"
+    }
+}
+
+foreach trace_command { "trace" "ftrace" } {
+    # This condition is always true as the PC should be set to the tracepoint
+    # address when hit.
+    test_tracepoints $trace_command "$pcreg == *set_point" 10
+
+    # Can we read local variables?
+    test_tracepoints $trace_command "anarg == 100 || anarg == 200" 2 18955_x86_64_failure
+    # Can we read global variables?
+    test_tracepoints $trace_command "anarg == 100 && globvar == 1" 1 18955_x86_64_failure
+
+    # Test various operations to cover as many opcodes as possible.
+    test_tracepoints $trace_command "21 + 21 == 42" 10
+    test_tracepoints $trace_command "21 - 21 == 0" 10
+    test_tracepoints $trace_command "21 * 2 == 42" 10
+    test_tracepoints $trace_command "21 << 1 == 42" 10
+    test_tracepoints $trace_command "42 >> 1 == 21" 10
+    test_tracepoints $trace_command "-21 << 1 == -42" 10
+    test_tracepoints $trace_command "-42 >> 1 == -21" 10
+    test_tracepoints $trace_command "(0xabababab & 0x0000ffff) == 0xabab" 10
+    test_tracepoints $trace_command "(0xabababab | 0x0000ffff) == 0xababffff" 10
+    test_tracepoints $trace_command "(0xaaaaaaaa ^ 0x55555555) == 0xffffffff" 10
+    test_tracepoints $trace_command "~0xaaaaaaaa == 0x55555555" 10
+    test_tracepoints $trace_command "21 < 42" 10
+    test_tracepoints $trace_command "42 <= 42" 10
+    test_tracepoints $trace_command "42 >= 42" 10
+    test_tracepoints $trace_command "42 > 21" 10
+    test_tracepoints $trace_command "(21 < 42 ? 0 : 1) == 0" 10 18955_i386_failure
+    test_tracepoints $trace_command "(42 <= 42 ? 0 : 1) == 0" 10
+    test_tracepoints $trace_command "(42 >= 42 ? 0 : 1) == 0" 10
+    test_tracepoints $trace_command "(42 > 21 ? 0 : 1) == 0" 10 18955_i386_failure
+    test_tracepoints $trace_command "\$trace_timestamp >= 0" 10
+}
-- 
2.4.6

Re: [PATCH 2/2] [testsuite] Add test case for tracepoints with conditions

[Yao Qi]

Pierre Langlois <pierre.langlois@arm.com> writes:

> +clean_restart ${executable}
> +
> +if ![runto_main] {
> +    fail "Can't run to main for ftrace tests"
> +    return 0
> +}
> +
> +gdb_reinitialize_dir $srcdir/$subdir

Don't need gdb_reinitialize_dir, as clean_restart did that already.

The patch is OK.

-- 
Yao (齐尧)

Re: [PATCH 2/2] [testsuite] Add test case for tracepoints with conditions

[Yao Qi]

Pierre Langlois <pierre.langlois@arm.com> writes:

[Pierre finished his 3-month rotation program on GDB and is assigned to
some other project.  I take over his patches today.]

> +proc test_tracepoints { trace_command condition num_frames { kfail_proc 0 } } {
> +    global executable
> +
> +    clean_restart ${executable}
> +
> +    if ![runto_main] {
> +	fail "Can't run to main for ftrace tests"
> +	return 0
> +    }
> +
> +    gdb_test "break begin" ".*" ""
> +
> +    gdb_test "break end" ".*" ""
> +
> +    gdb_test "${trace_command} set_point if ${condition}" "\(Fast t|T\)racepoint .*" \
> +	"${trace_command} with condition $condition"
> +
> +    gdb_test "continue" \
> +	".*Breakpoint \[0-9\]+, begin .*" \
> +	"advance to trace begin"

This will generate duplicated test results entries in gdb.sum.  We need
to use with_test_prefix to avoid that.

> +
> +    gdb_test_no_output "tstart" "start trace experiment"
> +
> +    gdb_test_multiple "continue" "advance through tracing" {
> +	-re ".*Breakpoint \[0-9\]+, end .*" {
> +	    pass "advance through tracing"
> +	}
> +	-re "Program received signal SIGSEGV, Segmentation fault\\..*" {
> +	    if { $kfail_proc != 0 } {
> +		$kfail_proc $trace_command
> +	    }
> +	    fail "advance through tracing"
> +	}
> +    }

This is racy, "$gdb_prompt $" is needed at the end of each pattern.
Patch below fixes all of them, and I'll push it in.

-- 
Yao (齐尧)

From 5aa114981fab1c3b80fe0a23fe3eee21fb3c0c99 Mon Sep 17 00:00:00 2001
From: Pierre Langlois <pierre.langlois@arm.com>
Date: Fri, 11 Sep 2015 18:25:01 +0100
Subject: [PATCH] Add test case for tracepoints with conditions

This patch adds a test case for tracepoints with a condition expression.
Each case will test a condition against the number of frames that should
have been traced.  Some of these tests fail on x86_64 and others on
i386, which have been marked as known failures for now, see PR/18955.

gdb/testsuite/ChangeLog:

2015-09-17  Pierre Langlois  <pierre.langlois@arm.com>
	    Yao Qi  <yao.qi@linaro.org>

	* gdb.trace/trace-condition.c: New file.
	* gdb.trace/trace-condition.exp: New file.
---
 gdb/testsuite/gdb.trace/trace-condition.c   |  66 +++++++++++
 gdb/testsuite/gdb.trace/trace-condition.exp | 168 ++++++++++++++++++++++++++++
 2 files changed, 234 insertions(+)
 create mode 100644 gdb/testsuite/gdb.trace/trace-condition.c
 create mode 100644 gdb/testsuite/gdb.trace/trace-condition.exp

diff --git a/gdb/testsuite/gdb.trace/trace-condition.c b/gdb/testsuite/gdb.trace/trace-condition.c
new file mode 100644
index 0000000..2e965c9
--- /dev/null
+++ b/gdb/testsuite/gdb.trace/trace-condition.c
@@ -0,0 +1,66 @@
+/* This testcase is part of GDB, the GNU debugger.
+
+   Copyright 2011-2015 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/>.  */
+
+#ifdef SYMBOL_PREFIX
+#define SYMBOL(str)     SYMBOL_PREFIX #str
+#else
+#define SYMBOL(str)     #str
+#endif
+
+int globvar;
+
+static void
+begin (void)
+{
+}
+
+/* Called from asm.  */
+static void __attribute__((used))
+func (void)
+{
+}
+
+static void
+marker (int anarg)
+{
+  /* `set_point' is the label at which to set a fast tracepoint.  The
+     insn at the label must be large enough to fit a fast tracepoint
+     jump.  */
+  asm ("    .global " SYMBOL (set_point) "\n"
+       SYMBOL (set_point) ":\n"
+#if (defined __x86_64__ || defined __i386__)
+       "    call " SYMBOL (func) "\n"
+#endif
+       );
+}
+
+static void
+end (void)
+{
+}
+
+int
+main ()
+{
+  begin ();
+
+  for (globvar = 1; globvar < 11; ++globvar)
+    marker (globvar * 100);
+
+  end ();
+  return 0;
+}
diff --git a/gdb/testsuite/gdb.trace/trace-condition.exp b/gdb/testsuite/gdb.trace/trace-condition.exp
new file mode 100644
index 0000000..d10fa9a
--- /dev/null
+++ b/gdb/testsuite/gdb.trace/trace-condition.exp
@@ -0,0 +1,168 @@
+# Copyright 2011-2015 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/>.
+
+load_lib "trace-support.exp"
+
+standard_testfile
+set executable $testfile
+set expfile $testfile.exp
+
+# Some targets have leading underscores on assembly symbols.
+set additional_flags [gdb_target_symbol_prefix_flags]
+
+if [is_amd64_regs_target] {
+    set pcreg "\$rip"
+} elseif [is_x86_like_target] {
+    set pcreg "\$eip"
+} else {
+    set pcreg "\$pc"
+}
+
+if [prepare_for_testing $expfile $executable $srcfile \
+	[list debug $additional_flags]] {
+    untested "failed to prepare for trace tests"
+    return -1
+}
+
+if ![runto_main] {
+    fail "Can't run to main to check for trace support"
+    return -1
+}
+
+if ![gdb_target_supports_trace] {
+    unsupported "target does not support trace"
+    return -1
+}
+
+set libipa [get_in_proc_agent]
+gdb_load_shlibs $libipa
+
+# Can't use prepare_for_testing, because that splits compiling into
+# building objects and then linking, and we'd fail with "linker input
+# file unused because linking not done" when building the object.
+
+if { [gdb_compile "$srcdir/$subdir/$srcfile" $binfile \
+	  executable [list debug $additional_flags shlib=$libipa] ] != "" } {
+    untested "failed to compile ftrace tests"
+    return -1
+}
+
+clean_restart ${executable}
+
+if ![runto_main] {
+    fail "Can't run to main for ftrace tests"
+    return 0
+}
+
+if { [gdb_test "info sharedlibrary" ".*${libipa}.*" "IPA loaded"] != 0 } {
+    untested "Could not find IPA lib loaded"
+    return 1
+}
+
+proc test_tracepoints { trace_command condition num_frames { kfail_proc 0 } } {
+    global executable gdb_prompt
+
+    clean_restart ${executable}
+
+    if ![runto_main] {
+	fail "Can't run to main for ftrace tests"
+	return 0
+    }
+
+    gdb_test "break begin" ".*" ""
+
+    gdb_test "break end" ".*" ""
+
+    with_test_prefix "${trace_command}: ${condition}" {
+
+	gdb_test "${trace_command} set_point if ${condition}" \
+	    "\(Fast t|T\)racepoint .*" \
+	    "set tracepoint"
+
+	gdb_test "continue" ".*Breakpoint \[0-9\]+, begin .*" \
+	    "advance to trace begin"
+
+	gdb_test_no_output "tstart" "start trace experiment"
+
+	gdb_test_multiple "continue" "advance through tracing" {
+	    -re ".*Breakpoint \[0-9\]+, end .*$gdb_prompt $" {
+		pass "advance through tracing"
+	    }
+	    -re "Program received signal SIGSEGV, Segmentation fault\\..*$gdb_prompt $" {
+		if { $kfail_proc != 0 } {
+		    $kfail_proc $trace_command
+		}
+		fail "advance through tracing"
+	    }
+	}
+
+	if { $kfail_proc != 0 } {
+	    $kfail_proc $trace_command
+	}
+	gdb_test "tstatus" \
+	    ".*Trace .*Collected $num_frames .*" \
+	    "check $num_frames frames were collected."
+
+	gdb_test "tstop" "" ""
+    }
+}
+
+# These callbacks identify known failures for certain architectures.  They
+# are called either if GDBserver crashes or has not traced the correct
+# number of frames.
+
+proc 18955_x86_64_failure { trace_command } {
+    if { $trace_command == "ftrace" } {
+	setup_kfail "gdb/18955" "x86_64-*-linux*"
+    }
+}
+
+proc 18955_i386_failure { trace_command } {
+    if { $trace_command == "ftrace" } {
+	setup_kfail "gdb/18955" "i\[34567\]86-*-*"
+    }
+}
+
+foreach trace_command { "trace" "ftrace" } {
+    # This condition is always true as the PC should be set to the tracepoint
+    # address when hit.
+    test_tracepoints $trace_command "$pcreg == *set_point" 10
+
+    # Can we read local variables?
+    test_tracepoints $trace_command "anarg == 100 || anarg == 200" 2 18955_x86_64_failure
+    # Can we read global variables?
+    test_tracepoints $trace_command "anarg == 100 && globvar == 1" 1 18955_x86_64_failure
+
+    # Test various operations to cover as many opcodes as possible.
+    test_tracepoints $trace_command "21 + 21 == 42" 10
+    test_tracepoints $trace_command "21 - 21 == 0" 10
+    test_tracepoints $trace_command "21 * 2 == 42" 10
+    test_tracepoints $trace_command "21 << 1 == 42" 10
+    test_tracepoints $trace_command "42 >> 1 == 21" 10
+    test_tracepoints $trace_command "-21 << 1 == -42" 10
+    test_tracepoints $trace_command "-42 >> 1 == -21" 10
+    test_tracepoints $trace_command "(0xabababab & 0x0000ffff) == 0xabab" 10
+    test_tracepoints $trace_command "(0xabababab | 0x0000ffff) == 0xababffff" 10
+    test_tracepoints $trace_command "(0xaaaaaaaa ^ 0x55555555) == 0xffffffff" 10
+    test_tracepoints $trace_command "~0xaaaaaaaa == 0x55555555" 10
+    test_tracepoints $trace_command "21 < 42" 10
+    test_tracepoints $trace_command "42 <= 42" 10
+    test_tracepoints $trace_command "42 >= 42" 10
+    test_tracepoints $trace_command "42 > 21" 10
+    test_tracepoints $trace_command "(21 < 42 ? 0 : 1) == 0" 10 18955_i386_failure
+    test_tracepoints $trace_command "(42 <= 42 ? 0 : 1) == 0" 10
+    test_tracepoints $trace_command "(42 >= 42 ? 0 : 1) == 0" 10
+    test_tracepoints $trace_command "(42 > 21 ? 0 : 1) == 0" 10 18955_i386_failure
+    test_tracepoints $trace_command "\$trace_timestamp >= 0" 10
+}
-- 
1.9.1

[PATCH 1/2] [GDBserver] Fix compiling conditional expressions accessing registers

[Pierre Langlois]

In order to compile conditions on fast tracepoints, we need to emit code
to read registers.  This is done with the `reg' opcode, and when
generating native code for it we emit a call to `gdb_agent_get_raw_reg',
which is a shared symbol between GDBserver and the IPA.

However, compiled conditions accessing registers do not work at the
moment.  For example, while debugging the ftrace test case on x86_64
using GDBserver:

~~~
(gdb) ftrace set_point if $rip == *set_point
Fast tracepoint 2 at 0x4006b9: ...
(gdb) tstart
(gdb) continue
Continuing.

Breakpoint 3, end () at ..
(gdb) tstatus
Trace is running on the target.
Collected 0 trace frames.
Trace buffer has 5242880 bytes of 5242880 bytes free (0% full).
Trace will stop if GDB disconnects.
Not looking at any trace frame.
Trace started at 1441727485.544642 secs.
~~~

The condition on the tracepoint should always evaluate to TRUE, $rip
being the program counter.

It turns out there is an error when emitting a function call to
`gdb_agent_get_raw_reg'.  The call emitted by `amd64_emit_reg' should
pass a pointer to the raw registers saved on the stack as a first
argument, but instead it passes the fast tracepoint context object
(`struct fast_tracepoint_ctx *').  As a result, `gdb_agent_get_raw_reg'
always returns the wrong values.

We can see this by setting a breakpoint on both
`condition_true_at_tracepoint' and `gdb_agent_get_raw_reg'.  We can see
the pointer being passed to the latter is the context:
~~~
(gdb) ftrace set_point if $rip == *set_point
Fast tracepoint 2 at 0x4006b9: ...
(gdb) b condition_true_at_tracepoint
Breakpoint 4 at ...
(gdb) b gdb_agent_get_raw_reg
Breakpoint 5 at ...
(gdb) tstart
(gdb) continue
Continuing.

Breakpoint 4, condition_true_at_tracepoint (ctx=0x7fffffffc038,
                                                ^^^^^^^^^^^^^^
    tpoint=0x7ffff68e2010) at ...
(gdb) continue

Breakpoint 5, gdb_agent_get_raw_reg (raw_regs=0x7fffffffc038 "\001",
                                              ^^^^^^^^^^^^^^
    regnum=16) at ...

(gdb)
~~~

This patch fixes this issue by replacing `gdb_agent_get_raw_reg' with a
`gdb_agent_get_reg' function which takes the tracepoint context object
as argument instead of a raw buffer.  Additionally, this patch makes
this function architecture independent by initializing the context's
regcache early and making `gdb_agent_get_reg' use `collect_register'.
As a result, the fast tracepoint context object does not need to
contain the raw register buffer.

gdb/gdbserver/ChangeLog:

	* linux-amd64-ipa.c (gdb_agent_get_raw_reg): Remove.
	* linux-i386-ipa.c (gdb_agent_get_raw_reg): Remove.
	* linux-x86-low.c (amd64_emit_reg): Call get_reg_func_addr
	instead of get_raw_reg_func_addr.
	(i386_emit_reg): Likewise.
	* tracepoint.c (get_raw_reg): Rename to ...
	(get_reg): ... this.
	(struct ipa_sym_addresses) <addr_get_raw_reg>: Rename to ...
	(struct ipa_sym_addresses) <add_get_reg>: ... this.
	(symbol_list): Rename get_raw_reg to get_reg.
	(struct fast_tracepoint_ctx) <regcache_initted>, <regs>: Remove.
	(get_context_regcache): Do not initialize fctx->regcache.
	(gdb_collect): Remove ctx.regs and ctx.regcache_initted.
	Initialize ctx.regcache.
	(gdb_agent_get_reg): New exported function.
	(get_raw_reg_func_addr): Rename to ...
	(get_reg_func_addr): ... this.
	* tracepoint.h (gdb_agent_get_raw_reg): Remove.
	(get_raw_reg_func_addr): Rename to ...
	(get_reg_func_addr): ... this.
---
 gdb/gdbserver/linux-amd64-ipa.c |  9 ---------
 gdb/gdbserver/linux-i386-ipa.c  | 15 ---------------
 gdb/gdbserver/linux-x86-low.c   |  4 ++--
 gdb/gdbserver/tracepoint.c      | 42 +++++++++++++++++++++++++----------------
 gdb/gdbserver/tracepoint.h      |  9 ++-------
 5 files changed, 30 insertions(+), 49 deletions(-)

diff --git a/gdb/gdbserver/linux-amd64-ipa.c b/gdb/gdbserver/linux-amd64-ipa.c
index a6dfb03..78125d7 100644
--- a/gdb/gdbserver/linux-amd64-ipa.c
+++ b/gdb/gdbserver/linux-amd64-ipa.c
@@ -68,15 +68,6 @@ supply_fast_tracepoint_registers (struct regcache *regcache,
 		     ((char *) buf) + x86_64_ft_collect_regmap[i]);
 }
 
-IP_AGENT_EXPORT_FUNC ULONGEST
-gdb_agent_get_raw_reg (const unsigned char *raw_regs, int regnum)
-{
-  if (regnum >= X86_64_NUM_FT_COLLECT_GREGS)
-    return 0;
-
-  return *(ULONGEST *) (raw_regs + x86_64_ft_collect_regmap[regnum]);
-}
-
 #ifdef HAVE_UST
 
 #include <ust/processor.h>
diff --git a/gdb/gdbserver/linux-i386-ipa.c b/gdb/gdbserver/linux-i386-ipa.c
index de8d394..6be87ad 100644
--- a/gdb/gdbserver/linux-i386-ipa.c
+++ b/gdb/gdbserver/linux-i386-ipa.c
@@ -98,21 +98,6 @@ supply_fast_tracepoint_registers (struct regcache *regcache,
     }
 }
 
-IP_AGENT_EXPORT_FUNC ULONGEST
-gdb_agent_get_raw_reg (const unsigned char *raw_regs, int regnum)
-{
-  /* This should maybe be allowed to return an error code, or perhaps
-     better, have the emit_reg detect this, and emit a constant zero,
-     or something.  */
-
-  if (regnum > i386_num_regs)
-    return 0;
-  else if (regnum >= I386_CS_REGNUM && regnum <= I386_GS_REGNUM)
-    return *(short *) (raw_regs + i386_ft_collect_regmap[regnum]);
-  else
-    return *(int *) (raw_regs + i386_ft_collect_regmap[regnum]);
-}
-
 #ifdef HAVE_UST
 
 #include <ust/processor.h>
diff --git a/gdb/gdbserver/linux-x86-low.c b/gdb/gdbserver/linux-x86-low.c
index 20d4257..3fb447d 100644
--- a/gdb/gdbserver/linux-x86-low.c
+++ b/gdb/gdbserver/linux-x86-low.c
@@ -2287,7 +2287,7 @@ amd64_emit_reg (int reg)
   i += 4;
   append_insns (&buildaddr, i, buf);
   current_insn_ptr = buildaddr;
-  amd64_emit_call (get_raw_reg_func_addr ());
+  amd64_emit_call (get_reg_func_addr ());
 }
 
 static void
@@ -2896,7 +2896,7 @@ i386_emit_reg (int reg)
 	    "mov %eax,4(%esp)\n\t"
 	    "mov 8(%ebp),%eax\n\t"
 	    "mov %eax,(%esp)");
-  i386_emit_call (get_raw_reg_func_addr ());
+  i386_emit_call (get_reg_func_addr ());
   EMIT_ASM32 (i386_reg_c,
 	    "xor %ebx,%ebx\n\t"
 	    "lea 0x8(%esp),%esp");
diff --git a/gdb/gdbserver/tracepoint.c b/gdb/gdbserver/tracepoint.c
index fd010ae..113848c 100644
--- a/gdb/gdbserver/tracepoint.c
+++ b/gdb/gdbserver/tracepoint.c
@@ -126,7 +126,7 @@ trace_vdebug (const char *fmt, ...)
 # define traceframe_write_count IPA_SYM_EXPORTED_NAME (traceframe_write_count)
 # define traceframes_created IPA_SYM_EXPORTED_NAME (traceframes_created)
 # define trace_state_variables IPA_SYM_EXPORTED_NAME (trace_state_variables)
-# define get_raw_reg IPA_SYM_EXPORTED_NAME (get_raw_reg)
+# define get_reg IPA_SYM_EXPORTED_NAME (get_reg)
 # define get_trace_state_variable_value \
   IPA_SYM_EXPORTED_NAME (get_trace_state_variable_value)
 # define set_trace_state_variable_value \
@@ -167,7 +167,7 @@ struct ipa_sym_addresses
   CORE_ADDR addr_traceframe_write_count;
   CORE_ADDR addr_traceframes_created;
   CORE_ADDR addr_trace_state_variables;
-  CORE_ADDR addr_get_raw_reg;
+  CORE_ADDR addr_get_reg;
   CORE_ADDR addr_get_trace_state_variable_value;
   CORE_ADDR addr_set_trace_state_variable_value;
   CORE_ADDR addr_ust_loaded;
@@ -203,7 +203,7 @@ static struct
   IPA_SYM(traceframe_write_count),
   IPA_SYM(traceframes_created),
   IPA_SYM(trace_state_variables),
-  IPA_SYM(get_raw_reg),
+  IPA_SYM(get_reg),
   IPA_SYM(get_trace_state_variable_value),
   IPA_SYM(set_trace_state_variable_value),
   IPA_SYM(ust_loaded),
@@ -1306,10 +1306,8 @@ struct fast_tracepoint_ctx
   struct tracepoint_hit_ctx base;
 
   struct regcache regcache;
-  int regcache_initted;
   unsigned char *regspace;
 
-  unsigned char *regs;
   struct tracepoint *tpoint;
 };
 
@@ -4724,13 +4722,6 @@ get_context_regcache (struct tracepoint_hit_ctx *ctx)
   if (ctx->type == fast_tracepoint)
     {
       struct fast_tracepoint_ctx *fctx = (struct fast_tracepoint_ctx *) ctx;
-      if (!fctx->regcache_initted)
-	{
-	  fctx->regcache_initted = 1;
-	  init_register_cache (&fctx->regcache, ipa_tdesc, fctx->regspace);
-	  supply_regblock (&fctx->regcache, NULL);
-	  supply_fast_tracepoint_registers (&fctx->regcache, fctx->regs);
-	}
       regcache = &fctx->regcache;
     }
 #ifdef HAVE_UST
@@ -5796,11 +5787,16 @@ gdb_collect (struct tracepoint *tpoint, unsigned char *regs)
     return;
 
   ctx.base.type = fast_tracepoint;
-  ctx.regs = regs;
-  ctx.regcache_initted = 0;
   /* Wrap the regblock in a register cache (in the stack, we don't
      want to malloc here).  */
   ctx.regspace = alloca (ipa_tdesc->registers_size);
+
+  /* Initialize the regcache with registers from the stack set up by jump
+     pad.  */
+  init_register_cache (&ctx.regcache, ipa_tdesc, ctx.regspace);
+  supply_regblock (&ctx.regcache, NULL);
+  supply_fast_tracepoint_registers (&ctx.regcache, regs);
+
   if (ctx.regspace == NULL)
     {
       trace_debug ("Trace buffer block allocation failed, skipping");
@@ -5853,14 +5849,28 @@ gdb_collect (struct tracepoint *tpoint, unsigned char *regs)
     }
 }
 
+/* This function is called from native code generated for the emit_reg
+   agent expression bytecode.  It returns the value of register number
+   REGNUM read from regcache contained into CTX.  */
+
+IP_AGENT_EXPORT_FUNC ULONGEST
+gdb_agent_get_reg (struct fast_tracepoint_ctx *ctx, int regnum)
+{
+  ULONGEST reg;
+
+  collect_register (&ctx->regcache, regnum, &reg);
+
+  return reg;
+}
+
 #endif
 
 #ifndef IN_PROCESS_AGENT
 
 CORE_ADDR
-get_raw_reg_func_addr (void)
+get_reg_func_addr (void)
 {
-  return ipa_sym_addrs.addr_get_raw_reg;
+  return ipa_sym_addrs.addr_get_reg;
 }
 
 CORE_ADDR
diff --git a/gdb/gdbserver/tracepoint.h b/gdb/gdbserver/tracepoint.h
index 30d0b58..3a96487 100644
--- a/gdb/gdbserver/tracepoint.h
+++ b/gdb/gdbserver/tracepoint.h
@@ -163,13 +163,8 @@ int agent_mem_read_string (struct eval_agent_expr_context *ctx,
 			   CORE_ADDR from,
 			   ULONGEST len);
 
-/* The prototype the get_raw_reg function in the IPA.  Each arch's
-   bytecode compiler emits calls to this function.  */
-IP_AGENT_EXPORT_FUNC ULONGEST gdb_agent_get_raw_reg
-  (const unsigned char *raw_regs, int regnum);
-
-/* Returns the address of the get_raw_reg function in the IPA.  */
-CORE_ADDR get_raw_reg_func_addr (void);
+/* Returns the address of the get_reg function in the IPA.  */
+CORE_ADDR get_reg_func_addr (void);
 /* Returns the address of the get_trace_state_variable_value
    function in the IPA.  */
 CORE_ADDR get_get_tsv_func_addr (void);
-- 
2.4.6

Re: [PATCH 1/2] [GDBserver] Fix compiling conditional expressions accessing registers

[Yao Qi]

Pierre Langlois <pierre.langlois@arm.com> writes:

> This patch fixes this issue by replacing `gdb_agent_get_raw_reg' with a
> `gdb_agent_get_reg' function which takes the tracepoint context object
> as argument instead of a raw buffer.  Additionally, this patch makes
> this function architecture independent by initializing the context's
> regcache early and making `gdb_agent_get_reg' use `collect_register'.
> As a result, the fast tracepoint context object does not need to
> contain the raw register buffer.

The fix looks reasonable to me as it makes no sense to keep two copies
of registers.  I go through this patch, and it looks good to me.

This patch invalidates Wei-cheng's patch here
<https://sourceware.org/ml/gdb-patches/2015-06/msg00587.html> which is
already approved but not committed.

If Ulirch/Wei-cheng have no objections, this patch can go in.

-- 
Yao (齐尧)

Re: [PATCH 1/2] [GDBserver] Fix compiling conditional expressions accessing registers

[Ulrich Weigand]

Yao Qi wrote:
> Pierre Langlois <pierre.langlois@arm.com> writes:
> 
> > This patch fixes this issue by replacing `gdb_agent_get_raw_reg' with a
> > `gdb_agent_get_reg' function which takes the tracepoint context object
> > as argument instead of a raw buffer.  Additionally, this patch makes
> > this function architecture independent by initializing the context's
> > regcache early and making `gdb_agent_get_reg' use `collect_register'.
> > As a result, the fast tracepoint context object does not need to
> > contain the raw register buffer.
> 
> The fix looks reasonable to me as it makes no sense to keep two copies
> of registers.  I go through this patch, and it looks good to me.

It seems to me this was intended as performance optimization to avoid
having to do the full regcache setup every time a tracepoint is hit,
in case we're not actually tracing registers ...

Not sure whether this is a real performance concern for actual use
cases though.  I don't have any actual measurements ...
 
> This patch invalidates Wei-cheng's patch here
> <https://sourceware.org/ml/gdb-patches/2015-06/msg00587.html> which is
> already approved but not committed.
> 
> If Ulirch/Wei-cheng have no objections, this patch can go in.

Except for the performance question above, I have no objections.

Bye,
Ulrich

-- 
  Dr. Ulrich Weigand
  GNU/Linux compilers and toolchain
  Ulrich.Weigand@de.ibm.com

Re: [PATCH 1/2] [GDBserver] Fix compiling conditional expressions accessing registers

[Pedro Alves]

On 09/16/2015 12:50 PM, Ulrich Weigand wrote:
> Yao Qi wrote:
>> Pierre Langlois <pierre.langlois@arm.com> writes:
>>
>>> This patch fixes this issue by replacing `gdb_agent_get_raw_reg' with a
>>> `gdb_agent_get_reg' function which takes the tracepoint context object
>>> as argument instead of a raw buffer.  Additionally, this patch makes
>>> this function architecture independent by initializing the context's
>>> regcache early and making `gdb_agent_get_reg' use `collect_register'.
>>> As a result, the fast tracepoint context object does not need to
>>> contain the raw register buffer.
>>
>> The fix looks reasonable to me as it makes no sense to keep two copies
>> of registers.  I go through this patch, and it looks good to me.
> 
> It seems to me this was intended as performance optimization to avoid
> having to do the full regcache setup every time a tracepoint is hit,
> in case we're not actually tracing registers ...
> 
> Not sure whether this is a real performance concern for actual use
> cases though.  I don't have any actual measurements ...

Yes, I was just now reading this, and was going to say the same.

E.g., the original fast tracepoints code was tuned to avoid as much
work as possible in the condition-fails scenario, when the condition
only accesses a global.  That is, e.g., "trace foo if debug_knob == 1".
I no longer have the numbers captured at the time though.

Thanks,
Pedro Alves

Re: [PATCH 1/2] [GDBserver] Fix compiling conditional expressions accessing registers

[Yao Qi]

Pedro Alves <palves@redhat.com> writes:

>> Not sure whether this is a real performance concern for actual use
>> cases though.  I don't have any actual measurements ...
>
> Yes, I was just now reading this, and was going to say the same.
>
> E.g., the original fast tracepoints code was tuned to avoid as much
> work as possible in the condition-fails scenario, when the condition
> only accesses a global.  That is, e.g., "trace foo if debug_knob == 1".
> I no longer have the numbers captured at the time though.

I run tspeed.exp, and it shows this patch slows down fast tracepoint
from 29ns to 144ns, so we need to commit Wei-chang's patch
<https://sourceware.org/ml/gdb-patches/2015-06/msg00587.html> instead.

-- 
Yao (齐尧)