[BOLT] Add unit tests for negate_ra_state cfi handling

- also add match_dwarf.py, a tool used by the unit tests.

Author: bgergely0

bolt/test/AArch64/incorrect-negate-ra-state.s

@@ -0,0 +1,41 @@
+# RUN: llvm-mc -filetype=obj -triple aarch64-unknown-unknown %s -o %t.o
+# RUN: %clang %cflags  %t.o -o %t.exe -Wl,-q
+# RUN: llvm-bolt %t.exe -o %t.exe.bolt | FileCheck %s
+# check that the output is listing foo as incorrect
+# CHECK: BOLT-INFO: inconsistent RAStates in function foo
+
+# check that foo got Ignored, so it's not in the new .text section
+# llvm-objdump %t.exe -d -j .text > %t.exe.dump
+# RUN: not grep "<foo>:" %t.exe.dump
+
+
+  .text
+  .globl  foo
+  .p2align        2
+  .type   foo,@function
+foo:
+  .cfi_startproc
+  hint    #25
+  .cfi_negate_ra_state
+  sub     sp, sp, #16
+  stp     x29, x30, [sp, #16]             // 16-byte Folded Spill
+  .cfi_def_cfa_offset 16
+  str     w0, [sp, #12]
+  ldr     w8, [sp, #12]
+  .cfi_negate_ra_state
+  add     w0, w8, #1
+  ldp     x29, x30, [sp, #16]             // 16-byte Folded Reload
+  add     sp, sp, #16
+  hint    #29
+  .cfi_negate_ra_state
+  ret
+.Lfunc_end1:
+  .size   foo, .Lfunc_end1-foo
+  .cfi_endproc
+
+  .global _start
+  .type _start, %function
+_start:
+  b foo
+
+.reloc 0, R_AARCH64_NONE

bolt/test/AArch64/simple-negate-ra-state.s

@@ -0,0 +1,42 @@
+# RUN: llvm-mc -filetype=obj -triple aarch64-unknown-unknown %s -o %t.o
+# RUN: %clang %cflags  %t.o -o %t.exe -Wl,-q
+
+# RUN: llvm-objdump %t.exe -d > %t.exe.dump
+# RUN: llvm-objdump --dwarf=frames %t.exe -D > %t.exe.dump-dwarf
+# RUN: match-dwarf %t.exe.dump %t.exe.dump-dwarf foo > orig.txt
+
+# RUN: llvm-bolt %t.exe -o %t.exe.bolt
+
+# RUN: llvm-objdump %t.exe.bolt -d > %t.exe.bolt.dump
+# RUN: llvm-objdump --dwarf=frames %t.exe.bolt  > %t.exe.bolt.dump-dwarf
+# RUN: match-dwarf %t.exe.bolt.dump %t.exe.bolt.dump-dwarf foo > bolted.txt
+
+# RUN: diff orig.txt bolted.txt
+
+	.text
+	.globl	foo
+	.p2align	2
+	.type	foo,@function
+foo:
+	.cfi_startproc
+	hint	#25
+	.cfi_negate_ra_state
+	sub	sp, sp, #16
+	stp	x29, x30, [sp, #16]             // 16-byte Folded Spill
+	.cfi_def_cfa_offset 16
+	str	w0, [sp, #12]
+	ldr	w8, [sp, #12]
+	add	w0, w8, #1
+	ldp	x29, x30, [sp, #16]             // 16-byte Folded Reload
+	add	sp, sp, #16
+	hint	#29
+	.cfi_negate_ra_state
+	ret
+.Lfunc_end1:
+	.size	foo, .Lfunc_end1-foo
+	.cfi_endproc
+
+	.global _start
+	.type _start, %function
+_start:
+	b foo

bolt/test/lit.cfg.py

@@ -89,6 +89,7 @@
 config.substitutions.append(("%cxxflags", ""))
 
 link_fdata_cmd = os.path.join(config.test_source_root, "link_fdata.py")
+match_dwarf_cmd = os.path.join(config.test_source_root, "match_dwarf.py")
 
 tool_dirs = [config.llvm_tools_dir, config.test_source_root]
 
@@ -131,6 +132,12 @@
     ToolSubst("llvm-readobj", unresolved="fatal"),
     ToolSubst("llvm-dwp", unresolved="fatal"),
     ToolSubst("split-file", unresolved="fatal"),
+    ToolSubst(
+        "match-dwarf",
+        command=sys.executable,
+        unresolved="fatal",
+        extra_args=[match_dwarf_cmd],
+    ),
 ]
 llvm_config.add_tool_substitutions(tools, tool_dirs)
 

bolt/test/match_dwarf.py

@@ -0,0 +1,114 @@
+#!/usr/bin/env python3
+
+# This tool helps matching dwarf dumps
+# (= the output from running llvm-objdump --dwarf=frames),
+# by address to function names (which are parsed from a normal objdump).
+# The script is used for checking if .cfi_negate_ra_state CFIs
+# are generated by BOLT the same way they are generated by LLVM.
+
+import argparse
+import subprocess
+import sys
+import re
+
+
+class NameDwarfPair(object):
+    def __init__(self, name, body):
+        self.name = name
+        self.body = body
+        self.finalized = False
+
+    def append(self, body_line):
+        # only store elements into the body until the first whitespace line is encountered.
+        if body_line.isspace():
+            self.finalized = True
+        if not self.finalized:
+            self.body += body_line
+
+    def print(self):
+        print(self.name)
+        print(self.body)
+
+    def parse_negates(self):
+        negate_offsets = []
+        loc = 0
+        # TODO: make sure this is not printed in hex
+        re_advloc = f"DW_CFA_advance_loc: (\d+)"
+
+        for line in self.body.splitlines():
+            # if line matches advance_loc int
+            match = re.search(re_advloc, line)
+            if match:
+                loc += int(match.group(1))
+            if "DW_CFA_AARCH64_negate_ra_state" in line:
+                negate_offsets.append(loc)
+
+        self.negate_offsets = negate_offsets
+
+    def __eq__(self, other):
+        return self.name == other.name and self.negate_offsets == other.negate_offsets
+
+
+def parse_objdump(objdump):
+    """
+    Parse and return address-to-name dictionary from objdump file
+    """
+    addr_name_dict = dict()
+    re_function = re.compile(r"^([0-9a-fA-F]+)\s<(.*)>:$")
+    with open(objdump, "r") as f:
+        for line in f.readlines():
+            match = re_function.match(line)
+            if not match:
+                continue
+            m_addr = match.groups()[0]
+            m_name = match.groups()[1]
+            addr_name_dict[int(m_addr, 16)] = m_name
+
+    return addr_name_dict
+
+
+def parse_dwarf(dwarfdump, addr_name_dict):
+    """
+    Parse dwarf dump, and match names to blocks using the dict from the objdump.
+    Return a list of NameDwarfPairs.
+    """
+    re_address_line = re.compile(r".*pc=([0-9a-fA-F]{8})\.\.\.([0-9a-fA-F]{8})")
+    with open(dwarfdump, "r") as dw:
+        functions = []
+        for line in dw.readlines():
+            match = re_address_line.match(line)
+            if not match:
+                if len(functions) > 0:
+                    functions[-1].append(line)
+                continue
+            pc_start_address = match.groups()[0]
+            name = addr_name_dict.get(int(pc_start_address, 16))
+            functions.append(NameDwarfPair(name, ""))
+
+        return functions
+
+
+def main():
+    parser = argparse.ArgumentParser()
+    parser.add_argument("objdump", help="Objdump file")
+    parser.add_argument(
+        "dwarfdump", help="dwarf dump file created with 'llvm-objdump --dwarf=frames'"
+    )
+    parser.add_argument("function", help="Function to search CFIs in.")
+
+    args = parser.parse_args()
+
+    addr_name_dict = parse_objdump(args.objdump)
+    functions = parse_dwarf(args.dwarfdump, addr_name_dict)
+
+    for f in functions:
+        if f.name == args.function:
+            f.parse_negates()
+            print(f.negate_offsets)
+            break
+    else:
+        print(f"{args.function} not found")
+        exit(-1)
+
+
+main()