perf tools: Add a test for decoding of new x86 instructions

Add a new test titled:

	Test x86 instruction decoder - new instructions

The purpose of this test is to check the instruction decoder after new
instructions have been added.  Initially, MPX instructions are tested
which are already supported, but the definitions in x86-opcode-map.txt
will be tweaked in a subsequent patch, after which this test can be run
to verify those changes.

The data for the test comes from assembly language instructions in
insn-x86-dat-src.c which is converted into bytes by the scripts
gen-insn-x86-dat.sh and gen-insn-x86-dat.awk, and included into the test
program insn-x86.c as insn-x86-dat-32.c and insn-x86-dat-64.c.

The conversion is not done as part of the perf tools build because the
test data must be under (git) change control in order for the test to be
repeatably-correct.  Also it may require a recent version of binutils.

Commiter notes:

Using it:

  # perf test decoder
  39: Test x86 instruction decoder - new instructions          : Ok
  # perf test -v decoder
  39: Test x86 instruction decoder - new instructions          :
  --- start ---
  test child forked, pid 21970
  Decoded ok: 0f 31                	rdtsc
  Decoded ok: f3 0f 1b 00          	bndmk  (%eax),%bnd0
  Decoded ok: f3 0f 1b 05 78 56 34 12 	bndmk  0x12345678,%bnd0
  Decoded ok: f3 0f 1b 18          	bndmk  (%eax),%bnd3
  <SNIP>
  Decoded ok: f2 e9 00 00 00 00    	bnd jmpq 402 <main+0x402>
  Decoded ok: f2 e9 00 00 00 00    	bnd jmpq 408 <main+0x408>
  Decoded ok: 67 f2 ff 21          	bnd jmpq *(%ecx)
  Decoded ok: f2 0f 85 00 00 00 00 	bnd jne 413 <main+0x413>
  test child finished with 0
  ---- end ----
  Test x86 instruction decoder - new instructions: Ok
  #
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Acked-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Qiaowei Ren <qiaowei.ren@intel.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/1441196131-20632-3-git-send-email-adrian.hunter@intel.comSigned-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>

tools/perf/tests/Build

@@ -35,6 +35,9 @@ perf-y += thread-map.o
 perf-y += llvm.o
 
 perf-$(CONFIG_X86) += perf-time-to-tsc.o
+ifdef CONFIG_AUXTRACE
+perf-$(CONFIG_X86) += insn-x86.o
+endif
 
 ifeq ($(ARCH),$(filter $(ARCH),x86 arm arm64))
 perf-$(CONFIG_DWARF_UNWIND) += dwarf-unwind.o

tools/perf/tests/builtin-test.c

@@ -178,6 +178,14 @@ static struct test {
 		.desc = "Test LLVM searching and compiling",
 		.func = test__llvm,
 	},
+#ifdef HAVE_AUXTRACE_SUPPORT
+#if defined(__x86_64__) || defined(__i386__)
+	{
+		.desc = "Test x86 instruction decoder - new instructions",
+		.func = test__insn_x86,
+	},
+#endif
+#endif
 	{
 		.func = NULL,
 	},

tools/perf/tests/gen-insn-x86-dat.awk

+#!/bin/awk -f
+# gen-insn-x86-dat.awk: script to convert data for the insn-x86 test
+# Copyright (c) 2015, Intel Corporation.
+#
+# This program is free software; you can redistribute it and/or modify it
+# under the terms and conditions of the GNU General Public License,
+# version 2, as published by the Free Software Foundation.
+#
+# This program is distributed in the hope 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.
+
+BEGIN {
+	print "/*"
+	print " * Generated by gen-insn-x86-dat.sh and gen-insn-x86-dat.awk"
+	print " * from insn-x86-dat-src.c for inclusion by insn-x86.c"
+	print " * Do not change this code."
+	print "*/\n"
+	op = ""
+	branch = ""
+	rel = 0
+	going = 0
+}
+
+/ Start here / {
+	going = 1
+}
+
+/ Stop here / {
+	going = 0
+}
+
+/^\s*[0-9a-fA-F]+\:/ {
+	if (going) {
+		colon_pos = index($0, ":")
+		useful_line = substr($0, colon_pos + 1)
+		first_pos = match(useful_line, "[0-9a-fA-F]")
+		useful_line = substr(useful_line, first_pos)
+		gsub("\t", "\\t", useful_line)
+		printf "{{"
+		len = 0
+		for (i = 2; i <= NF; i++) {
+			if (match($i, "^[0-9a-fA-F][0-9a-fA-F]$")) {
+				printf "0x%s, ", $i
+				len += 1
+			} else {
+				break
+			}
+		}
+		printf "}, %d, %s, \"%s\", \"%s\",", len, rel, op, branch
+		printf "\n\"%s\",},\n", useful_line
+		op = ""
+		branch = ""
+		rel = 0
+	}
+}
+
+/ Expecting: / {
+	expecting_str = " Expecting: "
+	expecting_len = length(expecting_str)
+	expecting_pos = index($0, expecting_str)
+	useful_line = substr($0, expecting_pos + expecting_len)
+	for (i = 1; i <= NF; i++) {
+		if ($i == "Expecting:") {
+			i++
+			op = $i
+			i++
+			branch = $i
+			i++
+			rel = $i
+			break
+		}
+	}
+}

tools/perf/tests/gen-insn-x86-dat.sh

+#!/bin/sh
+# gen-insn-x86-dat: generate data for the insn-x86 test
+# Copyright (c) 2015, Intel Corporation.
+#
+# This program is free software; you can redistribute it and/or modify it
+# under the terms and conditions of the GNU General Public License,
+# version 2, as published by the Free Software Foundation.
+#
+# This program is distributed in the hope 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.
+
+set -e
+
+if [ "$(uname -m)" != "x86_64" ]; then
+	echo "ERROR: This script only works on x86_64"
+	exit 1
+fi
+
+cd $(dirname $0)
+
+trap 'echo "Might need a more recent version of binutils"' EXIT
+
+echo "Compiling insn-x86-dat-src.c to 64-bit object"
+
+gcc -g -c insn-x86-dat-src.c
+
+objdump -dSw insn-x86-dat-src.o | awk -f gen-insn-x86-dat.awk > insn-x86-dat-64.c
+
+rm -f insn-x86-dat-src.o
+
+echo "Compiling insn-x86-dat-src.c to 32-bit object"
+
+gcc -g -c -m32 insn-x86-dat-src.c
+
+objdump -dSw insn-x86-dat-src.o | awk -f gen-insn-x86-dat.awk > insn-x86-dat-32.c
+
+rm -f insn-x86-dat-src.o
+
+trap - EXIT
+
+echo "Done (use git diff to see the changes)"

tools/perf/tests/insn-x86.c

+#include <linux/types.h>
+
+#include "debug.h"
+#include "tests.h"
+
+#include "intel-pt-decoder/insn.h"
+#include "intel-pt-decoder/intel-pt-insn-decoder.h"
+
+struct test_data {
+	u8 data[MAX_INSN_SIZE];
+	int expected_length;
+	int expected_rel;
+	const char *expected_op_str;
+	const char *expected_branch_str;
+	const char *asm_rep;
+};
+
+struct test_data test_data_32[] = {
+#include "insn-x86-dat-32.c"
+	{{0}, 0, 0, NULL, NULL, NULL},
+};
+
+struct test_data test_data_64[] = {
+#include "insn-x86-dat-64.c"
+	{{0}, 0, 0, NULL, NULL, NULL},
+};
+
+static int get_op(const char *op_str)
+{
+	struct val_data {
+		const char *name;
+		int val;
+	} vals[] = {
+		{"other",   INTEL_PT_OP_OTHER},
+		{"call",    INTEL_PT_OP_CALL},
+		{"ret",     INTEL_PT_OP_RET},
+		{"jcc",     INTEL_PT_OP_JCC},
+		{"jmp",     INTEL_PT_OP_JMP},
+		{"loop",    INTEL_PT_OP_LOOP},
+		{"iret",    INTEL_PT_OP_IRET},
+		{"int",     INTEL_PT_OP_INT},
+		{"syscall", INTEL_PT_OP_SYSCALL},
+		{"sysret",  INTEL_PT_OP_SYSRET},
+		{NULL, 0},
+	};
+	struct val_data *val;
+
+	if (!op_str || !strlen(op_str))
+		return 0;
+
+	for (val = vals; val->name; val++) {
+		if (!strcmp(val->name, op_str))
+			return val->val;
+	}
+
+	pr_debug("Failed to get op\n");
+
+	return -1;
+}
+
+static int get_branch(const char *branch_str)
+{
+	struct val_data {
+		const char *name;
+		int val;
+	} vals[] = {
+		{"no_branch",     INTEL_PT_BR_NO_BRANCH},
+		{"indirect",      INTEL_PT_BR_INDIRECT},
+		{"conditional",   INTEL_PT_BR_CONDITIONAL},
+		{"unconditional", INTEL_PT_BR_UNCONDITIONAL},
+		{NULL, 0},
+	};
+	struct val_data *val;
+
+	if (!branch_str || !strlen(branch_str))
+		return 0;
+
+	for (val = vals; val->name; val++) {
+		if (!strcmp(val->name, branch_str))
+			return val->val;
+	}
+
+	pr_debug("Failed to get branch\n");
+
+	return -1;
+}
+
+static int test_data_item(struct test_data *dat, int x86_64)
+{
+	struct intel_pt_insn intel_pt_insn;
+	struct insn insn;
+	int op, branch;
+
+	insn_init(&insn, dat->data, MAX_INSN_SIZE, x86_64);
+	insn_get_length(&insn);
+
+	if (!insn_complete(&insn)) {
+		pr_debug("Failed to decode: %s\n", dat->asm_rep);
+		return -1;
+	}
+
+	if (insn.length != dat->expected_length) {
+		pr_debug("Failed to decode length (%d vs expected %d): %s\n",
+			 insn.length, dat->expected_length, dat->asm_rep);
+		return -1;
+	}
+
+	op = get_op(dat->expected_op_str);
+	branch = get_branch(dat->expected_branch_str);
+
+	if (intel_pt_get_insn(dat->data, MAX_INSN_SIZE, x86_64, &intel_pt_insn)) {
+		pr_debug("Intel PT failed to decode: %s\n", dat->asm_rep);
+		return -1;
+	}
+
+	if ((int)intel_pt_insn.op != op) {
+		pr_debug("Failed to decode 'op' value (%d vs expected %d): %s\n",
+			 intel_pt_insn.op, op, dat->asm_rep);
+		return -1;
+	}
+
+	if ((int)intel_pt_insn.branch != branch) {
+		pr_debug("Failed to decode 'branch' value (%d vs expected %d): %s\n",
+			 intel_pt_insn.branch, branch, dat->asm_rep);
+		return -1;
+	}
+
+	if (intel_pt_insn.rel != dat->expected_rel) {
+		pr_debug("Failed to decode 'rel' value (%#x vs expected %#x): %s\n",
+			 intel_pt_insn.rel, dat->expected_rel, dat->asm_rep);
+		return -1;
+	}
+
+	pr_debug("Decoded ok: %s\n", dat->asm_rep);
+
+	return 0;
+}
+
+static int test_data_set(struct test_data *dat_set, int x86_64)
+{
+	struct test_data *dat;
+	int ret = 0;
+
+	for (dat = dat_set; dat->expected_length; dat++) {
+		if (test_data_item(dat, x86_64))
+			ret = -1;
+	}
+
+	return ret;
+}
+
+/**
+ * test__insn_x86 - test x86 instruction decoder - new instructions.
+ *
+ * This function implements a test that decodes a selection of instructions and
+ * checks the results.  The Intel PT function that further categorizes
+ * instructions (i.e. intel_pt_get_insn()) is also checked.
+ *
+ * The instructions are originally in insn-x86-dat-src.c which has been
+ * processed by scripts gen-insn-x86-dat.sh and gen-insn-x86-dat.awk to produce
+ * insn-x86-dat-32.c and insn-x86-dat-64.c which are included into this program.
+ * i.e. to add new instructions to the test, edit insn-x86-dat-src.c, run the
+ * gen-insn-x86-dat.sh script, make perf, and then run the test.
+ *
+ * If the test passes %0 is returned, otherwise %-1 is returned.  Use the
+ * verbose (-v) option to see all the instructions and whether or not they
+ * decoded successfuly.
+ */
+int test__insn_x86(void)
+{
+	int ret = 0;
+
+	if (test_data_set(test_data_32, 0))
+		ret = -1;
+
+	if (test_data_set(test_data_64, 1))
+		ret = -1;
+
+	return ret;
+}

tools/perf/tests/tests.h

@@ -63,6 +63,7 @@ int test__fdarray__add(void);
 int test__kmod_path__parse(void);
 int test__thread_map(void);
 int test__llvm(void);
+int test__insn_x86(void);
 
 #if defined(__x86_64__) || defined(__i386__) || defined(__arm__) || defined(__aarch64__)
 #ifdef HAVE_DWARF_UNWIND_SUPPORT