From mboxrd@z Thu Jan  1 00:00:00 1970
Return-Path: <gcc-patches-return-305889-listarch-gcc-patches=gcc.gnu.org@gcc.gnu.org>
Received: (qmail 12940 invoked by alias); 28 Oct 2011 15:56:23 -0000
Received: (qmail 12845 invoked by uid 22791); 28 Oct 2011 15:56:10 -0000
X-SWARE-Spam-Status: No, hits=-2.2 required=5.0	tests=AWL,BAYES_00,DKIM_SIGNED,DKIM_VALID,DKIM_VALID_AU,RP_MATCHES_RCVD,SPF_HELO_PASS,TW_CP,TW_TM,TW_VP,TW_VT
X-Spam-Check-By: sourceware.org
Received: from smtp-out.google.com (HELO smtp-out.google.com) (74.125.121.67)    by sourceware.org (qpsmtpd/0.43rc1) with ESMTP; Fri, 28 Oct 2011 15:55:46 +0000
Received: from wpaz29.hot.corp.google.com (wpaz29.hot.corp.google.com [172.24.198.93])	by smtp-out.google.com with ESMTP id p9SFthCc007141	for <gcc-patches@gcc.gnu.org>; Fri, 28 Oct 2011 08:55:44 -0700
Received: from vcbfl15 (vcbfl15.prod.google.com [10.220.204.79])	by wpaz29.hot.corp.google.com with ESMTP id p9SFpPtx000573	(version=TLSv1/SSLv3 cipher=RC4-SHA bits=128 verify=NOT)	for <gcc-patches@gcc.gnu.org>; Fri, 28 Oct 2011 08:55:43 -0700
Received: by vcbfl15 with SMTP id fl15so5667064vcb.5        for <gcc-patches@gcc.gnu.org>; Fri, 28 Oct 2011 08:55:42 -0700 (PDT)
Received: by 10.150.205.19 with SMTP id c19mr3252374ybg.58.1319817342837;        Fri, 28 Oct 2011 08:55:42 -0700 (PDT)
MIME-Version: 1.0
Received: by 10.150.205.19 with SMTP id c19mr3252326ybg.58.1319817342394; Fri, 28 Oct 2011 08:55:42 -0700 (PDT)
Received: by 10.150.218.16 with HTTP; Fri, 28 Oct 2011 08:55:42 -0700 (PDT)
In-Reply-To: <CAAkRFZ+epzvx2fs24B92GxrB0eP+6rYVwpodR-YnXTUtt1sdaw@mail.gmail.com>
References: <20111028132511.873EAC1B4C@1024cores.msk.corp.google.com>	<CAAkRFZ+epzvx2fs24B92GxrB0eP+6rYVwpodR-YnXTUtt1sdaw@mail.gmail.com>
Date: Fri, 28 Oct 2011 16:32:00 -0000
Message-ID: <CAD_=9DTSRhOj8thOiYOJj8kyvYxwTKBghYeF+MLuDvDyDkv30Q@mail.gmail.com>
Subject: Re: [C-compiler-team] [google] ThreadSanitizer instrumentation pass (issue5303083)
From: Diego Novillo <dnovillo@google.com>
To: Xinliang David Li <davidxl@google.com>
Cc: Dmitriy Vyukov <dvyukov@google.com>, reply@codereview.appspotmail.com,        kcc@google.com, gcc-patches@gcc.gnu.org
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: quoted-printable
X-System-Of-Record: true
X-IsSubscribed: yes
Mailing-List: contact gcc-patches-help@gcc.gnu.org; run by ezmlm
Precedence: bulk
List-Id: <gcc-patches.gcc.gnu.org>
List-Archive: <http://gcc.gnu.org/ml/gcc-patches/>
List-Post: <mailto:gcc-patches@gcc.gnu.org>
List-Help: <mailto:gcc-patches-help@gcc.gnu.org>
Sender: gcc-patches-owner@gcc.gnu.org
X-SW-Source: 2011-10/txt/msg02687.txt.bz2

[ Removing closed list from CCs, please trim CC in future replies.  Thanks.=
  ]

On Fri, Oct 28, 2011 at 08:53, Xinliang David Li <davidxl@google.com> wrote:
> Dmitriy, I will review it next week.
>
> Do you have some runtime overhead data ?
>
> thanks,
>
> David
>
> On Fri, Oct 28, 2011 at 6:25 AM, Dmitriy Vyukov <dvyukov@google.com> wrot=
e:
>> The patch is for google/main branch.
>> ThreadSanitizer is a data race detector for C/C++ programs.
>> http://code.google.com/p/data-race-test/wiki/ThreadSanitizer
>>
>> The tool consists of two parts:
>> instrumentation module (this file) and a run-time library.
>> The instrumentation module mainintains shadow call stacks
>> and intercepts interesting memory accesses.
>> The instrumentation is enabled with -ftsan flag.
>>
>> Instrumentation for shadow stack maintainance is as follows:
>> void somefunc ()
>> {
>> =C2=A0__tsan_shadow_stack [-1] =3D __builtin_return_address (0);
>> =C2=A0__tsan_shadow_stack++;
>> =C2=A0// function body
>> =C2=A0__tsan_shadow_stack--;
>> }
>>
>> Interception for memory access interception is as follows:
>> *addr =3D 1;
>> __tsan_handle_mop (addr, flags);
>> where flags are (is_sblock | (is_store << 1) | ((sizeof (*addr) - 1) << =
2).
>> is_sblock is used merely for optimization purposes and can always
>> be set to 1, see comments in instrument_mops function.
>>
>> Ignore files can be used to selectively non instrument some functions.
>> Ignore file is specified with -ftsan-ignore=3Dfilename flag.
>> There are 3 types of ignores: (1) do not instrument memory accesses
>> in the function, (2) do not create sblocks in the function
>> and (3) recursively ignore memory accesses in the function.
>> That last ignore type requires additional instrumentation of the form:
>> void somefunc ()
>> {
>> =C2=A0__tsan_thread_ignore++;
>> =C2=A0// function body
>> =C2=A0__tsan_thread_ignore--;
>> }
>>
>> The run-time library provides __tsan_handle_mop function,
>> definitions of __tsan_shadow_stack and __tsan_thread_ignore variables,
>> and intercepts synchronization related functions.
>>
>> 2011-10-28 =C2=A0 Dmitriy Vyukov =C2=A0<dvyukov@google.com>
>>
>> =C2=A0 =C2=A0 =C2=A0 =C2=A0* gcc/doc/invoke.texi:
>> =C2=A0 =C2=A0 =C2=A0 =C2=A0* gcc/tree-tsan.c (enum tsan_ignore_e):
>> =C2=A0 =C2=A0 =C2=A0 =C2=A0(enum bb_state_e):
>> =C2=A0 =C2=A0 =C2=A0 =C2=A0(struct bb_data_t):
>> =C2=A0 =C2=A0 =C2=A0 =C2=A0(struct mop_desc_t):
>> =C2=A0 =C2=A0 =C2=A0 =C2=A0(struct tsan_ignore_desc_t):
>> =C2=A0 =C2=A0 =C2=A0 =C2=A0(lookup_name):
>> =C2=A0 =C2=A0 =C2=A0 =C2=A0(shadow_stack_def):
>> =C2=A0 =C2=A0 =C2=A0 =C2=A0(thread_ignore_def):
>> =C2=A0 =C2=A0 =C2=A0 =C2=A0(rtl_mop_def):
>> =C2=A0 =C2=A0 =C2=A0 =C2=A0(ignore_append):
>> =C2=A0 =C2=A0 =C2=A0 =C2=A0(ignore_match):
>> =C2=A0 =C2=A0 =C2=A0 =C2=A0(ignore_load):
>> =C2=A0 =C2=A0 =C2=A0 =C2=A0(tsan_ignore):
>> =C2=A0 =C2=A0 =C2=A0 =C2=A0(decl_name):
>> =C2=A0 =C2=A0 =C2=A0 =C2=A0(build_stack_op):
>> =C2=A0 =C2=A0 =C2=A0 =C2=A0(build_rec_ignore_op):
>> =C2=A0 =C2=A0 =C2=A0 =C2=A0(build_stack_assign):
>> =C2=A0 =C2=A0 =C2=A0 =C2=A0(instr_mop):
>> =C2=A0 =C2=A0 =C2=A0 =C2=A0(instr_vptr_store):
>> =C2=A0 =C2=A0 =C2=A0 =C2=A0(instr_func):
>> =C2=A0 =C2=A0 =C2=A0 =C2=A0(set_location):
>> =C2=A0 =C2=A0 =C2=A0 =C2=A0(is_dtor_vptr_store):
>> =C2=A0 =C2=A0 =C2=A0 =C2=A0(is_vtbl_read):
>> =C2=A0 =C2=A0 =C2=A0 =C2=A0(is_load_of_const):
>> =C2=A0 =C2=A0 =C2=A0 =C2=A0(handle_expr):
>> =C2=A0 =C2=A0 =C2=A0 =C2=A0(handle_gimple):
>> =C2=A0 =C2=A0 =C2=A0 =C2=A0(instrument_bblock):
>> =C2=A0 =C2=A0 =C2=A0 =C2=A0(instrument_mops):
>> =C2=A0 =C2=A0 =C2=A0 =C2=A0(instrument_function):
>> =C2=A0 =C2=A0 =C2=A0 =C2=A0(tsan_pass):
>> =C2=A0 =C2=A0 =C2=A0 =C2=A0(tsan_gate):
>> =C2=A0 =C2=A0 =C2=A0 =C2=A0* gcc/tree-pass.h:
>> =C2=A0 =C2=A0 =C2=A0 =C2=A0* gcc/testsuite/gcc.dg/tsan-ignore.ignore:
>> =C2=A0 =C2=A0 =C2=A0 =C2=A0* gcc/testsuite/gcc.dg/tsan.h (__tsan_init):
>> =C2=A0 =C2=A0 =C2=A0 =C2=A0(__tsan_expect_mop):
>> =C2=A0 =C2=A0 =C2=A0 =C2=A0(__tsan_handle_mop):
>> =C2=A0 =C2=A0 =C2=A0 =C2=A0* gcc/testsuite/gcc.dg/tsan-ignore.c (foo):
>> =C2=A0 =C2=A0 =C2=A0 =C2=A0(int bar):
>> =C2=A0 =C2=A0 =C2=A0 =C2=A0(int baz):
>> =C2=A0 =C2=A0 =C2=A0 =C2=A0(int bla):
>> =C2=A0 =C2=A0 =C2=A0 =C2=A0(int xxx):
>> =C2=A0 =C2=A0 =C2=A0 =C2=A0(main):
>> =C2=A0 =C2=A0 =C2=A0 =C2=A0* gcc/testsuite/gcc.dg/tsan-ignore.h (in_tsan=
_ignore_header):
>> =C2=A0 =C2=A0 =C2=A0 =C2=A0* gcc/testsuite/gcc.dg/tsan-stack.c (foobar):
>> =C2=A0 =C2=A0 =C2=A0 =C2=A0* gcc/testsuite/gcc.dg/tsan-mop.c:
>> =C2=A0 =C2=A0 =C2=A0 =C2=A0* gcc/common.opt:
>> =C2=A0 =C2=A0 =C2=A0 =C2=A0* gcc/Makefile.in:
>> =C2=A0 =C2=A0 =C2=A0 =C2=A0* gcc/passes.c:
>>
>> Index: gcc/doc/invoke.texi
>> =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
>> --- gcc/doc/invoke.texi (revision 180522)
>> +++ gcc/doc/invoke.texi (working copy)
>> @@ -308,6 +308,7 @@
>> =C2=A0-fdump-tree-ssa@r{[}-@var{n}@r{]} -fdump-tree-pre@r{[}-@var{n}@r{]=
} @gol
>> =C2=A0-fdump-tree-ccp@r{[}-@var{n}@r{]} -fdump-tree-dce@r{[}-@var{n}@r{]=
} @gol
>> =C2=A0-fdump-tree-gimple@r{[}-raw@r{]} -fdump-tree-mudflap@r{[}-@var{n}@=
r{]} @gol
>> +-fdump-tree-tsan@r{[}-@var{n}@r{]} @gol
>> =C2=A0-fdump-tree-dom@r{[}-@var{n}@r{]} @gol
>> =C2=A0-fdump-tree-dse@r{[}-@var{n}@r{]} @gol
>> =C2=A0-fdump-tree-phiprop@r{[}-@var{n}@r{]} @gol
>> @@ -381,8 +382,8 @@
>> =C2=A0-floop-parallelize-all -flto -flto-compression-level @gol
>> =C2=A0-flto-partition=3D@var{alg} -flto-report -fmerge-all-constants @gol
>> =C2=A0-fmerge-constants -fmodulo-sched -fmodulo-sched-allow-regmoves @gol
>> --fmove-loop-invariants fmudflap -fmudflapir -fmudflapth -fno-branch-cou=
nt-reg @gol
>> --fno-default-inline @gol
>> +-fmove-loop-invariants -fmudflap -fmudflapir -fmudflapth -fno-branch-co=
unt-reg @gol
>> +-ftsan -ftsan-ignore -fno-default-inline @gol
>> =C2=A0-fno-defer-pop -fno-function-cse -fno-guess-branch-probability @gol
>> =C2=A0-fno-inline -fno-math-errno -fno-peephole -fno-peephole2 @gol
>> =C2=A0-fno-sched-interblock -fno-sched-spec -fno-signed-zeros @gol
>> @@ -5896,6 +5897,11 @@
>> =C2=A0Dump each function after adding mudflap instrumentation. =C2=A0The=
 file name is
>> =C2=A0made by appending @file{.mudflap} to the source file name.
>>
>> +@item tsan
>> +@opindex fdump-tree-tsan
>> +Dump each function after adding ThreadSanitizer instrumentation. =C2=A0=
The file name is
>> +made by appending @file{.tsan} to the source file name.
>> +
>> =C2=A0@item sra
>> =C2=A0@opindex fdump-tree-sra
>> =C2=A0Dump each function after performing scalar replacement of aggregat=
es. =C2=A0The
>> @@ -6674,6 +6680,12 @@
>> =C2=A0some protection against outright memory corrupting writes, but all=
ows
>> =C2=A0erroneously read data to propagate within a program.
>>
>> +@item -ftsan -ftsan-ignore
>> +@opindex ftsan
>> +@opindex ftsan-ignore
>> +Add ThreadSanitizer instrumentation. Use @option{-ftsan-ignore} to spec=
ify
>> +an ignore file. Refer to http://go/tsan for details.
>> +
>> =C2=A0@item -fthread-jumps
>> =C2=A0@opindex fthread-jumps
>> =C2=A0Perform optimizations where we check to see if a jump branches to a
>> Index: gcc/tree-tsan.c
>> =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
>> --- gcc/tree-tsan.c =C2=A0 =C2=A0 (revision 0)
>> +++ gcc/tree-tsan.c =C2=A0 =C2=A0 (revision 0)
>> @@ -0,0 +1,1204 @@
>> +/* ThreadSanitizer instrumentation pass.
>> + =C2=A0 http://code.google.com/p/data-race-test
>> + =C2=A0 Copyright (C) 2011
>> + =C2=A0 Free Software Foundation, Inc.
>> + =C2=A0 Contributed by Dmitry Vyukov <dvyukov@google.com>
>> +
>> +This file is part of GCC.
>> +
>> +GCC 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, or (at your option)
>> +any later version.
>> +
>> +GCC 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. =C2=A0See the
>> +GNU General Public License for more details.
>> +
>> +You should have received a copy of the GNU General Public License
>> +along with GCC; see the file COPYING3. =C2=A0If not see
>> +<http://www.gnu.org/licenses/>. =C2=A0*/
>> +
>> +#include "config.h"
>> +#include "system.h"
>> +#include "coretypes.h"
>> +#include "tree.h"
>> +#include "intl.h"
>> +#include "tm.h"
>> +#include "basic-block.h"
>> +#include "gimple.h"
>> +#include "function.h"
>> +#include "tree-flow.h"
>> +#include "tree-pass.h"
>> +#include "cfghooks.h"
>> +#include "langhooks.h"
>> +#include "output.h"
>> +#include "options.h"
>> +
>> +/* The file can be compiled either as compiler pass or plugin. =C2=A0*/
>> +#ifdef GCC_PLG
>> +# include "c-common.h"
>> +#else
>> +# include "c-family/c-common.h"
>> +#endif
>> +
>> +#include "diagnostic.h"
>> +
>> +#include <stdlib.h>
>> +#include <stdio.h>
>> +
>> +/* ThreadSanitizer is a data race detector for C/C++ programs.
>> + =C2=A0 http://code.google.com/p/data-race-test/wiki/ThreadSanitizer
>> +
>> + =C2=A0 The tool consists of two parts:
>> + =C2=A0 instrumentation module (this file) and a run-time library.
>> + =C2=A0 The instrumentation module mainintains shadow call stacks
>> + =C2=A0 and intercepts interesting memory accesses.
>> + =C2=A0 The instrumentation is enabled with -ftsan flag.
>> +
>> + =C2=A0 Instrumentation for shadow stack maintainance is as follows:
>> + =C2=A0 void somefunc ()
>> + =C2=A0 {
>> + =C2=A0 =C2=A0 __tsan_shadow_stack [-1] =3D __builtin_return_address (0=
);
>> + =C2=A0 =C2=A0 __tsan_shadow_stack++;
>> + =C2=A0 =C2=A0 // function body
>> + =C2=A0 =C2=A0 __tsan_shadow_stack--;
>> + =C2=A0 }
>> +
>> + =C2=A0 Interception for memory access interception is as follows:
>> + =C2=A0 *addr =3D 1;
>> + =C2=A0 __tsan_handle_mop (addr, flags);
>> + =C2=A0 where flags are (is_sblock | (is_store << 1) | ((sizeof (*addr)=
 - 1) << 2).
>> + =C2=A0 is_sblock is used merely for optimization purposes and can alwa=
ys
>> + =C2=A0 be set to 1, see comments in instrument_mops function.
>> +
>> + =C2=A0 Ignore files can be used to selectively non instrument some fun=
ctions.
>> + =C2=A0 Ignore file is specified with -ftsan-ignore=3Dfilename flag.
>> + =C2=A0 There are 3 types of ignores: (1) do not instrument memory acce=
sses
>> + =C2=A0 in the function, (2) do not create sblocks in the function
>> + =C2=A0 and (3) recursively ignore memory accesses in the function.
>> + =C2=A0 That last ignore type requires additional instrumentation of th=
e form:
>> + =C2=A0 void somefunc ()
>> + =C2=A0 {
>> + =C2=A0 =C2=A0 __tsan_thread_ignore++;
>> + =C2=A0 =C2=A0 // function body
>> + =C2=A0 =C2=A0 __tsan_thread_ignore--;
>> + =C2=A0 }
>> +
>> + =C2=A0 The run-time library provides __tsan_handle_mop function,
>> + =C2=A0 definitions of __tsan_shadow_stack and __tsan_thread_ignore var=
iables,
>> + =C2=A0 and intercepts synchronization related functions. =C2=A0*/
>> +
>> +#define RTL_IGNORE "__tsan_thread_ignore"
>> +#define RTL_STACK "__tsan_shadow_stack"
>> +#define RTL_MOP "__tsan_handle_mop"
>> +#define RTL_PERFIX "__tsan_"
>> +#define MAX_MOP_BYTES 16
>> +#define SBLOCK_SIZE 5
>> +
>> +enum tsan_ignore_e
>> +{
>> + =C2=A0tsan_ignore_none =C2=A0=3D 1 << 0, /* Do not ignore. =C2=A0*/
>> + =C2=A0tsan_ignore_func =C2=A0=3D 1 << 1, /* Completely ignore the whol=
e func. =C2=A0*/
>> + =C2=A0tsan_ignore_mop =C2=A0 =3D 1 << 2, /* Do not instrument memory a=
ccesses. =C2=A0*/
>> + =C2=A0tsan_ignore_rec =C2=A0 =3D 1 << 3, /* Do not instrument memory a=
ccesses recursively. =C2=A0*/
>> + =C2=A0tsan_ignore_hist =C2=A0=3D 1 << 4 =C2=A0/* Do not create superbl=
ocks. =C2=A0*/
>> +};
>> +
>> +/* Basic block state during CFG traversal. =C2=A0*/
>> +enum bb_state_e
>> +{
>> + =C2=A0bb_not_visited,
>> + =C2=A0bb_candidate,
>> + =C2=A0bb_visited
>> +};
>> +
>> +/* Info associated with each basic block.
>> + =C2=A0 Used to determine super-blocks (see instrument_mops ()). =C2=A0=
*/
>> +struct bb_data_t
>> +{
>> + =C2=A0enum bb_state_e =C2=A0 =C2=A0 =C2=A0 state;
>> + =C2=A0int =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=
=A0 has_sb;
>> + =C2=A0const char =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 *sb_file;
>> + =C2=A0int =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=
=A0 sb_line_min;
>> + =C2=A0int =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=
=A0 sb_line_max;
>> +};
>> +
>> +/* Memory access descriptor. =C2=A0*/
>> +struct mop_desc_t
>> +{
>> + =C2=A0int =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=
=A0 is_call;
>> + =C2=A0gimple_stmt_iterator =C2=A0gsi;
>> + =C2=A0tree =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=
=A0expr;
>> + =C2=A0tree =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=
=A0dtor_vptr_expr;
>> + =C2=A0int =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=
=A0 is_store;
>> +};
>> +
>> +struct tsan_ignore_desc_t
>> +{
>> + =C2=A0struct tsan_ignore_desc_t *next;
>> + =C2=A0enum tsan_ignore_e =C2=A0 =C2=A0 =C2=A0 =C2=A0 type;
>> + =C2=A0char =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=
=A0 =C2=A0 =C2=A0*name;
>> +};
>> +
>> +/* Number of instrumented memory accesses in the current function. =C2=
=A0*/
>> +static int func_mops;
>> +/* Number of function calls in the current function. =C2=A0*/
>> +static int func_calls;
>> +/* Ignore status for the current function (see tsan_ignore_e). =C2=A0*/
>> +static enum tsan_ignore_e func_ignore;
>> +
>> +static int ignore_init =3D 0;
>> +static struct tsan_ignore_desc_t *ignore_head;
>> +
>> +typedef struct mop_desc_t mop_desc_t;
>> +DEF_VEC_O (mop_desc_t);
>> +DEF_VEC_ALLOC_O (mop_desc_t, heap);
>> +static VEC (mop_desc_t, heap) *mop_list;
>> +
>> +/* The function is not available in some modules. =C2=A0*/
>> +tree __attribute__((weak))
>> +lookup_name (tree t)
>> +{
>> + =C2=A0(void)t;
>> + =C2=A0return NULL_TREE;
>> +}
>> +
>> +/* Builds the following decl
>> + =C2=A0 extern __thread void **__tsan_shadow_stack; */
>> +static tree
>> +shadow_stack_def (void)
>> +{
>> + =C2=A0static tree def;
>> +
>> + =C2=A0if (def !=3D NULL)
>> + =C2=A0 =C2=A0return def;
>> +
>> + =C2=A0/* Check if a user has defined it for testing */
>> + =C2=A0def =3D lookup_name (get_identifier (RTL_STACK));
>> + =C2=A0if (def !=3D NULL)
>> + =C2=A0 =C2=A0return def;
>> +
>> + =C2=A0def =3D build_decl (UNKNOWN_LOCATION, VAR_DECL,
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 get_ide=
ntifier (RTL_STACK),
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 build_p=
ointer_type (ptr_type_node));
>> + =C2=A0TREE_STATIC (def) =3D 1;
>> + =C2=A0TREE_PUBLIC (def) =3D 1;
>> + =C2=A0DECL_EXTERNAL (def) =3D 1;
>> + =C2=A0DECL_TLS_MODEL (def) =3D decl_default_tls_model (def);
>> + =C2=A0TREE_USED (def) =3D 1;
>> + =C2=A0TREE_THIS_VOLATILE (def) =3D 1;
>> + =C2=A0SET_DECL_ASSEMBLER_NAME (def, get_identifier (RTL_STACK));
>> + =C2=A0return def;
>> +}
>> +
>> +/* Builds the following decl
>> + =C2=A0 extern __thread int __tsan_thread_ignore; */
>> +static tree
>> +thread_ignore_def (void)
>> +{
>> + =C2=A0static tree def;
>> +
>> + =C2=A0if (def !=3D NULL)
>> + =C2=A0 =C2=A0return def;
>> +
>> + =C2=A0/* Check if a user has defined it for testing */
>> + =C2=A0def =3D lookup_name (get_identifier (RTL_IGNORE));
>> + =C2=A0if (def !=3D NULL)
>> + =C2=A0 =C2=A0return def;
>> +
>> + =C2=A0def =3D build_decl (UNKNOWN_LOCATION, VAR_DECL,
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 get_ide=
ntifier (RTL_IGNORE),
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 integer=
_type_node);
>> + =C2=A0TREE_STATIC (def) =3D 1;
>> + =C2=A0TREE_PUBLIC (def) =3D 1;
>> + =C2=A0DECL_EXTERNAL (def) =3D 1;
>> + =C2=A0DECL_TLS_MODEL (def) =3D decl_default_tls_model (def);
>> + =C2=A0TREE_USED (def) =3D 1;
>> + =C2=A0TREE_THIS_VOLATILE (def) =3D 1;
>> + =C2=A0SET_DECL_ASSEMBLER_NAME (def, get_identifier (RTL_IGNORE));
>> + =C2=A0return def;
>> +}
>> +
>> +/* Builds the following decl
>> + =C2=A0 void __tsan_handle_mop (void *addr, unsigned flags); */
>> +static tree
>> +rtl_mop_def (void)
>> +{
>> + =C2=A0tree fn_type;
>> +
>> + =C2=A0static tree def;
>> +
>> + =C2=A0if (def !=3D NULL)
>> + =C2=A0 =C2=A0return def;
>> +
>> + =C2=A0/* Check if a user has defined it for testing */
>> + =C2=A0def =3D lookup_name (get_identifier (RTL_MOP));
>> + =C2=A0if (def !=3D NULL)
>> + =C2=A0 =C2=A0return def;
>> +
>> + =C2=A0fn_type =3D build_function_type_list (void_type_node, ptr_type_n=
ode,
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =
=C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0integer_type_=
node , NULL_TREE);
>> + =C2=A0def =3D build_fn_decl (RTL_MOP, fn_type);
>> + =C2=A0TREE_NOTHROW (def) =3D 1;
>> + =C2=A0DECL_ATTRIBUTES (def) =3D tree_cons (get_identifier ("leaf"),
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =
=C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 NULL, DECL_ATTRIBUT=
ES (def));
>> + =C2=A0DECL_ASSEMBLER_NAME (def);
>> + =C2=A0return def;
>> +}
>> +
>> +/* Adds new ignore definition to the global list */
>> +static void
>> +ignore_append (enum tsan_ignore_e type, char *name)
>> +{
>> + =C2=A0struct tsan_ignore_desc_t *desc;
>> +
>> + =C2=A0desc =3D (struct tsan_ignore_desc_t*)xmalloc (sizeof (*desc));
>> + =C2=A0desc->type =3D type;
>> + =C2=A0desc->name =3D xstrdup (name);
>> + =C2=A0desc->next =3D ignore_head;
>> + =C2=A0ignore_head =3D desc;
>> +}
>> +
>> +/* Checks as to whether identifier 'str' matches template 'templ'.
>> + =C2=A0 Templates can only contain '*', e.g. 'std*string*insert'.
>> + =C2=A0 Templates implicitly start and end with '*'
>> + =C2=A0 since they are matched against mangled names. =C2=A0*/
>> +static int
>> +ignore_match (char *templ, const char *str)
>> +{
>> + =C2=A0char *tpos;
>> + =C2=A0const char *spos;
>> +
>> + =C2=A0while (templ && templ [0])
>> + =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0if (templ [0] =3D=3D '*')
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0templ++;
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0continue;
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0}
>> + =C2=A0 =C2=A0 =C2=A0if (str [0] =3D=3D 0)
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0return 0;
>> + =C2=A0 =C2=A0 =C2=A0tpos =3D strchr (templ, '*');
>> + =C2=A0 =C2=A0 =C2=A0if (tpos !=3D NULL)
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0tpos [0] =3D 0;
>> + =C2=A0 =C2=A0 =C2=A0spos =3D strstr (str, templ);
>> + =C2=A0 =C2=A0 =C2=A0str =3D spos + strlen (templ);
>> + =C2=A0 =C2=A0 =C2=A0templ =3D tpos;
>> + =C2=A0 =C2=A0 =C2=A0if (tpos !=3D NULL)
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0tpos [0] =3D '*';
>> + =C2=A0 =C2=A0 =C2=A0if (spos =3D=3D NULL)
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0return 0;
>> + =C2=A0 =C2=A0}
>> + =C2=A0return 1;
>> +}
>> +
>> +/* Loads ignore definitions from the file specified by -ftsan-ignore=3D=
filename.
>> + =C2=A0 Ignore files have the following format:
>> +
>> +# This is a comment - ignored
>> +
>> +# The below line says to not instrument memory accesses
>> +# in all functions that match 'std*string*insert'
>> +fun:std*string*insert
>> +
>> +# The below line says to not instrument memory accesses
>> +# in the function called 'foobar' *and* in all functions
>> +# that it calls recursively
>> +fun_r:foobar
>> +
>> +# The below line says to not create superblocks
>> +# in the function called 'barbaz'
>> +fun_hist:barbaz
>> +
>> +# Ignore all functions in the source file
>> +src:atomic.c
>> +
>> +# Everything else is uninteresting for us (e.g. obj:)
>> +*/
>> +static void
>> +ignore_load (void)
>> +{
>> + =C2=A0FILE *f;
>> + =C2=A0char *line;
>> + =C2=A0size_t linesz;
>> + =C2=A0ssize_t sz;
>> + =C2=A0char buf [PATH_MAX];
>> +
>> + =C2=A0if (flag_tsan_ignore =3D=3D NULL || flag_tsan_ignore [0] =3D=3D =
0)
>> + =C2=A0 =C2=A0return;
>> +
>> + =C2=A0f =3D fopen (flag_tsan_ignore, "r");
>> + =C2=A0if (f =3D=3D NULL)
>> + =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0/* Try to open it relative to main_input_filename.=
 =C2=A0*/
>> + =C2=A0 =C2=A0 =C2=A0strncpy (buf, main_input_filename, sizeof (buf));
>> + =C2=A0 =C2=A0 =C2=A0buf [sizeof (buf) - 1] =3D 0;
>> + =C2=A0 =C2=A0 =C2=A0line =3D strrchr (buf, '/');
>> + =C2=A0 =C2=A0 =C2=A0if (line !=3D NULL)
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0line++;
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0strncpy (line, flag_tsan_ignore, siz=
eof (buf) - (line - buf));
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0buf [sizeof (buf) - 1] =3D 0;
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0f =3D fopen (buf, "r");
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0}
>> + =C2=A0 =C2=A0}
>> + =C2=A0if (f =3D=3D NULL)
>> + =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0printf ("failed to open ignore file '%s'\n", flag_=
tsan_ignore);
>> + =C2=A0 =C2=A0 =C2=A0exit (1);
>> + =C2=A0 =C2=A0}
>> +
>> + =C2=A0line =3D 0;
>> + =C2=A0linesz =3D 0;
>> + =C2=A0while ((sz =3D getline (&line, &linesz, f)) !=3D -1)
>> + =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0if (sz =3D=3D 0)
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0continue;
>> + =C2=A0 =C2=A0 =C2=A0/* strip line terminator */
>> + =C2=A0 =C2=A0 =C2=A0if (line [sz-1] =3D=3D '\r' || line [sz-1] =3D=3D =
'\n')
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0line [sz-1] =3D 0;
>> + =C2=A0 =C2=A0 =C2=A0if (strncmp (line, "src:", sizeof ("src:")-1) =3D=
=3D 0)
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0ignore_append (tsan_ignore_func, line + siz=
eof ("src:")-1);
>> + =C2=A0 =C2=A0 =C2=A0else if (strncmp (line, "fun:", sizeof ("fun:")-1)=
 =3D=3D 0)
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0ignore_append (tsan_ignore_mop, line + size=
of ("fun:")-1);
>> + =C2=A0 =C2=A0 =C2=A0else if (strncmp (line, "fun_r:", sizeof ("fun_r:"=
)-1) =3D=3D 0)
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0ignore_append (tsan_ignore_rec, line + size=
of ("fun_r:")-1);
>> + =C2=A0 =C2=A0 =C2=A0else if (strncmp (line, "fun_hist:", sizeof ("fun_=
hist:")-1) =3D=3D 0)
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0ignore_append (tsan_ignore_hist, line + siz=
eof ("fun_hist:")-1);
>> + =C2=A0 =C2=A0 =C2=A0/* other lines are not interesting */
>> + =C2=A0 =C2=A0}
>> +
>> + =C2=A0free (line);
>> + =C2=A0fclose (f);
>> +}
>> +
>> +/* Returns ignore status for the current function */
>> +static enum tsan_ignore_e
>> +tsan_ignore (void)
>> +{
>> + =C2=A0const char *func_name;
>> + =C2=A0const char *src_name;
>> + =C2=A0struct tsan_ignore_desc_t *desc;
>> +
>> + =C2=A0if (ignore_init =3D=3D 0)
>> + =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0ignore_load ();
>> + =C2=A0 =C2=A0 =C2=A0ignore_init =3D 1;
>> + =C2=A0 =C2=A0}
>> +
>> + =C2=A0src_name =3D expand_location(cfun->function_start_locus).file;
>> + =C2=A0if (src_name =3D=3D NULL)
>> + =C2=A0 =C2=A0src_name =3D "";
>> +
>> + =C2=A0func_name =3D IDENTIFIER_POINTER (DECL_ASSEMBLER_NAME (cfun->dec=
l));
>> + =C2=A0/* Ignore all functions starting with __tsan_ - intended for tes=
ting */
>> + =C2=A0if (strncmp (func_name, RTL_PERFIX, sizeof (RTL_PERFIX) - 1) =3D=
=3D 0)
>> + =C2=A0 =C2=A0return tsan_ignore_func;
>> +
>> + =C2=A0for (desc =3D ignore_head; desc; desc =3D desc->next)
>> + =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0if (desc->type =3D=3D tsan_ignore_func)
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0if (ignore_match (desc->name, src_na=
me))
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 return desc->type;
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0}
>> + =C2=A0 =C2=A0 =C2=A0else if (ignore_match (desc->name, func_name))
>> + =C2=A0 =C2=A0 =C2=A0 return desc->type;
>> + =C2=A0 =C2=A0}
>> + =C2=A0return tsan_ignore_none;
>> +}
>> +
>> +static const char *
>> +decl_name (tree decl)
>> +{
>> + =C2=A0tree id;
>> + =C2=A0const char *name;
>> +
>> + =C2=A0if (decl !=3D 0 && DECL_P (decl))
>> + =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0id =3D DECL_NAME (decl);
>> + =C2=A0 =C2=A0 =C2=A0if (id !=3D NULL)
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0name =3D IDENTIFIER_POINTER (id);
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0if (name !=3D NULL)
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0return name;
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0}
>> + =C2=A0 =C2=A0}
>> + =C2=A0return "<unknown>";
>> +}
>> +
>> +/* Builds either (__tsan_shadow_stack +=3D 1) or (__tsan_shadow_stack -=
=3D 1) expression
>> + =C2=A0 depending on 'do_dec' parameter. =C2=A0Appends the result to se=
q. =C2=A0*/
>> +static void
>> +build_stack_op (gimple_seq *seq, bool do_dec)
>> +{
>> + =C2=A0tree op_size;
>> + =C2=A0double_int op_size_cst;
>> + =C2=A0unsigned long long size_val;
>> + =C2=A0unsigned long long size_valhi;
>> + =C2=A0tree op_expr;
>> + =C2=A0tree assign;
>> + =C2=A0tree rtl_stack;
>> + =C2=A0gimple_seq s;
>> +
>> + =C2=A0op_size =3D TYPE_SIZE (ptr_type_node);
>> + =C2=A0op_size_cst =3D tree_to_double_int (op_size);
>> + =C2=A0size_val =3D op_size_cst.low / BITS_PER_UNIT;
>> + =C2=A0size_valhi =3D 0;
>> + =C2=A0if (do_dec)
>> + =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0size_val =3D -size_val;
>> + =C2=A0 =C2=A0 =C2=A0size_valhi =3D -1;
>> + =C2=A0 =C2=A0}
>> + =C2=A0op_size =3D build_int_cst_wide (sizetype, size_val, size_valhi);
>> + =C2=A0rtl_stack =3D shadow_stack_def ();
>> + =C2=A0op_expr =3D build2 (POINTER_PLUS_EXPR, ptr_type_node, rtl_stack,=
 op_size);
>> + =C2=A0assign =3D build2 (MODIFY_EXPR, ptr_type_node, rtl_stack, op_exp=
r);
>> + =C2=A0s =3D NULL;
>> + =C2=A0force_gimple_operand (assign, &s, true, NULL_TREE);
>> + =C2=A0gimple_seq_add_seq (seq, s);
>> +}
>> +
>> +/* Builds either (__tsan_thread_ignore +=3D 1) or (__tsan_thread_ignore=
 -=3D 1) expression
>> + =C2=A0 depending on op parameter. =C2=A0Stores the result in seq. =C2=
=A0*/
>> +static void
>> +build_rec_ignore_op (gimple_seq *seq, enum tree_code op)
>> +{
>> + =C2=A0tree rec_expr;
>> + =C2=A0gimple_seq rec_inc;
>> + =C2=A0gimple rec_assign;
>> + =C2=A0tree rtl_ignore;
>> +
>> + =C2=A0rtl_ignore =3D thread_ignore_def ();
>> + =C2=A0rec_expr =3D build2 (op, integer_type_node, rtl_ignore, integer_=
one_node);
>> + =C2=A0rec_inc =3D NULL;
>> + =C2=A0rec_expr =3D force_gimple_operand (rec_expr, &rec_inc, true, NUL=
L_TREE);
>> + =C2=A0rec_assign =3D gimple_build_assign (rtl_ignore, rec_expr);
>> + =C2=A0gimple_seq_add_seq (seq, rec_inc);
>> + =C2=A0gimple_seq_add_stmt (seq, rec_assign);
>> +}
>> +
>> +/* Build the following gimple sequence:
>> + =C2=A0 __tsan_shadow_stack [-1] =3D __builtin_return_address (0);
>> + =C2=A0 Stores the result in seq. =C2=A0*/
>> +static void
>> +build_stack_assign (gimple_seq *seq)
>> +{
>> + =C2=A0tree pc_addr;
>> + =C2=A0tree op_size;
>> + =C2=A0tree op_expr;
>> + =C2=A0tree stack_op;
>> + =C2=A0tree assign;
>> + =C2=A0tree rtl_retaddr;
>> +
>> + =C2=A0rtl_retaddr =3D implicit_built_in_decls [BUILT_IN_RETURN_ADDRESS=
];
>> + =C2=A0pc_addr =3D build_call_expr (rtl_retaddr, 1, integer_zero_node);
>> + =C2=A0op_size =3D build_int_cst_wide (sizetype, -(POINTER_SIZE / BITS_=
PER_UNIT), -1);
>> + =C2=A0op_expr =3D build2 (POINTER_PLUS_EXPR, ptr_type_node,
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =
=C2=A0 =C2=A0shadow_stack_def (), op_size);
>> + =C2=A0stack_op =3D build1 (INDIRECT_REF, ptr_type_node, op_expr);
>> + =C2=A0assign =3D build2 (MODIFY_EXPR, ptr_type_node, stack_op, pc_addr=
);
>> + =C2=A0force_gimple_operand (assign, seq, true, NULL_TREE);
>> +}
>> +
>> +/* Builds the following gimple sequence:
>> + =C2=A0 __tsan_handle_mop (&expr, (is_sblock | (is_store << 1) | ((size=
of (expr)-1) << 2)
>> + =C2=A0 The result is stored in gseq. =C2=A0*/
>> +static void
>> +instr_mop (tree expr, int is_store, int is_sblock, gimple_seq *gseq)
>> +{
>> + =C2=A0tree addr_expr;
>> + =C2=A0tree expr_type;
>> + =C2=A0unsigned size;
>> + =C2=A0unsigned flags;
>> + =C2=A0tree flags_expr;
>> + =C2=A0tree call_expr;
>> +
>> + =C2=A0gcc_assert (gseq !=3D 0 && *gseq =3D=3D 0);
>> + =C2=A0gcc_assert (is_gimple_addressable (expr));
>> +
>> + =C2=A0addr_expr =3D build_addr (expr, current_function_decl);
>> + =C2=A0expr_type =3D TREE_TYPE (expr);
>> + =C2=A0while (TREE_CODE (expr_type) =3D=3D ARRAY_TYPE)
>> + =C2=A0 =C2=A0expr_type =3D TREE_TYPE (expr_type);
>> + =C2=A0size =3D TREE_INT_CST_LOW (TYPE_SIZE (expr_type));
>> + =C2=A0size =3D size / BITS_PER_UNIT;
>> + =C2=A0if (size > MAX_MOP_BYTES)
>> + =C2=A0 =C2=A0size =3D MAX_MOP_BYTES;
>> + =C2=A0size -=3D 1;
>> + =C2=A0flags =3D ((!!is_sblock << 0) + (!!is_store << 1) + (size << 2));
>> + =C2=A0flags_expr =3D build_int_cst (unsigned_type_node, flags);
>> + =C2=A0call_expr =3D build_call_expr (rtl_mop_def (), 2, addr_expr, fla=
gs_expr);
>> + =C2=A0force_gimple_operand (call_expr, gseq, true, 0);
>> +}
>> +
>> +/* Builds the following gimple sequence:
>> + =C2=A0 int is_store =3D (expr !=3D rhs); // the temp is not actually i=
ntroduced
>> + =C2=A0 __tsan_handle_mop (&expr, (is_sblock | (is_store << 1) | ((size=
of (expr)-1) << 2)
>> + =C2=A0 The result is stored in gseq. =C2=A0*/
>> +static void
>> +instr_vptr_store (tree expr, tree rhs, int is_sblock, gimple_seq *gseq)
>> +{
>> + =C2=A0tree expr_ptr;
>> + =C2=A0tree addr_expr;
>> + =C2=A0tree expr_type;
>> + =C2=A0tree expr_size;
>> + =C2=A0double_int size;
>> + =C2=A0unsigned flags;
>> + =C2=A0tree flags_expr;
>> + =C2=A0gimple_seq flags_seq;
>> + =C2=A0gimple collect;
>> + =C2=A0tree is_store_expr;
>> +
>> + =C2=A0expr_ptr =3D build_addr (expr, current_function_decl);
>> + =C2=A0addr_expr =3D force_gimple_operand (expr_ptr, gseq, true, NULL_T=
REE);
>> + =C2=A0expr_type =3D TREE_TYPE (expr);
>> + =C2=A0while (TREE_CODE (expr_type) =3D=3D ARRAY_TYPE)
>> + =C2=A0 =C2=A0expr_type =3D TREE_TYPE (expr_type);
>> + =C2=A0expr_size =3D TYPE_SIZE (expr_type);
>> + =C2=A0size =3D tree_to_double_int (expr_size);
>> + =C2=A0gcc_assert (size.high =3D=3D 0 && size.low !=3D 0);
>> + =C2=A0if (size.low > 128)
>> + =C2=A0 =C2=A0size.low =3D 128;
>> + =C2=A0size.low =3D (size.low / 8) - 1;
>> + =C2=A0flags =3D ((!!is_sblock << 0) + (size.low << 2));
>> + =C2=A0flags_expr =3D build_int_cst (unsigned_type_node, flags);
>> + =C2=A0is_store_expr =3D build2 (NE_EXPR, integer_type_node,
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =
=C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0build1 (VIEW_CONVERT_EXPR, size_type_node=
, expr),
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =
=C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0build1 (VIEW_CONVERT_EXPR, size_type_node=
, rhs));
>> + =C2=A0is_store_expr =3D build2 (LSHIFT_EXPR, integer_type_node,
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =
=C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0is_store_expr, integer_one_node);
>> + =C2=A0flags_expr =3D build2 (BIT_IOR_EXPR, integer_type_node,
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =
=C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0is_store_expr, flags_expr);
>> + =C2=A0flags_seq =3D 0;
>> + =C2=A0flags_expr =3D force_gimple_operand (flags_expr, &flags_seq, tru=
e, NULL_TREE);
>> + =C2=A0gimple_seq_add_seq (gseq, flags_seq);
>> + =C2=A0collect =3D gimple_build_call (
>> + =C2=A0 =C2=A0 =C2=A0rtl_mop_def (), 2, addr_expr, flags_expr);
>> + =C2=A0gimple_seq_add_stmt (gseq, collect);
>> +}
>> +
>> +/* Builds gimple sequences that must be inserted at function entry (pre)
>> + =C2=A0 and before function exit (post). =C2=A0*/
>> +static void
>> +instr_func (gimple_seq *pre, gimple_seq *post)
>> +{
>> + =C2=A0/* In this case we need no instrumentation for the function */
>> + =C2=A0if (func_calls =3D=3D 0 && func_mops =3D=3D 0)
>> + =C2=A0 =C2=A0return;
>> +
>> + =C2=A0if (func_ignore !=3D tsan_ignore_rec)
>> + =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0build_stack_assign (pre);
>> + =C2=A0 =C2=A0 =C2=A0build_stack_op (pre, false);
>> + =C2=A0 =C2=A0 =C2=A0build_stack_op (post, true);
>> + =C2=A0 =C2=A0}
>> +
>> + =C2=A0if (func_ignore =3D=3D tsan_ignore_rec && func_calls !=3D 0)
>> + =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0build_rec_ignore_op (pre, PLUS_EXPR);
>> + =C2=A0 =C2=A0 =C2=A0build_rec_ignore_op (post, MINUS_EXPR);
>> + =C2=A0 =C2=A0}
>> +}
>> +
>> +/* Sets location for all gimples in the seq. =C2=A0*/
>> +static void
>> +set_location (gimple_seq seq, location_t loc)
>> +{
>> + =C2=A0gimple_seq_node n;
>> +
>> + =C2=A0for (n =3D gimple_seq_first (seq); n !=3D NULL; n =3D n->next)
>> + =C2=A0 =C2=A0gimple_set_location (n->stmt, loc);
>> +}
>> +
>> +/* Check as to whether expr refers to a store to vptr. =C2=A0*/
>> +static tree
>> +is_dtor_vptr_store (gimple stmt, tree expr, int is_store)
>> +{
>> + =C2=A0if (is_store =3D=3D 1
>> + =C2=A0 =C2=A0 =C2=A0&& TREE_CODE (expr) =3D=3D COMPONENT_REF
>> + =C2=A0 =C2=A0 =C2=A0&& gimple_assign_single_p (stmt)
>> + =C2=A0 =C2=A0 =C2=A0&& strcmp (decl_name (cfun->decl), "__base_dtor ")=
 =3D=3D 0)
>> + =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0tree comp =3D expr->exp.operands [0];
>> + =C2=A0 =C2=A0 =C2=A0while (TREE_CODE (comp) =3D=3D COMPONENT_REF)
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0comp =3D comp->exp.operands [0];
>> + =C2=A0 =C2=A0 =C2=A0if (TREE_CODE (comp) =3D=3D INDIRECT_REF || TREE_C=
ODE (comp) =3D=3D MEM_REF)
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0comp =3D comp->exp.operands [0];
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0if (TREE_CODE (comp) =3D=3D SSA_NAME)
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0comp =3D SSA_NAME_VAR (comp);
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0if (strcmp (decl_name (comp), "this"=
) =3D=3D 0)
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0tree field =3D expr->e=
xp.operands [1];
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0if (TREE_CODE (field) =
=3D=3D FIELD_DECL
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0&& strnc=
mp (decl_name (field),
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =
=C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0"_vptr.", sizeof ("_vptr.") - 1) =3D=3D 0)
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0return gimple_a=
ssign_rhs1 (stmt);
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0}
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0}
>> + =C2=A0 =C2=A0}
>> + =C2=A0return 0;
>> +}
>> +
>> +/* Checks as to whether expr refers to a read from vtlb.
>> + =C2=A0 Vtlbs are immutable, so don't bother to instrument them. =C2=A0=
*/
>> +static int
>> +is_vtbl_read (tree expr, int is_store)
>> +{
>> + =C2=A0/* We may not instrument reads from vtbl, because the data is co=
nstant.
>> + =C2=A0 =C2=A0 vtbl read is of the form:
>> + =C2=A0 =C2=A0 =C2=A0 gimple_assign <component_ref, D.2133, x->_vptr.X,=
 NULL>
>> + =C2=A0 =C2=A0 =C2=A0 gimple_assign <indirect_ref, D.2134, *D.2133, NUL=
L>
>> + =C2=A0 =C2=A0 or:
>> + =C2=A0 =C2=A0 =C2=A0 gimple_assign <component_ref, D.2133, x->_vptr.X,=
 NULL>
>> + =C2=A0 =C2=A0 =C2=A0 gimple_assign <pointer_plus_expr, D.2135, D.2133,=
 8>
>> + =C2=A0 =C2=A0 =C2=A0 gimple_assign <indirect_ref, D.2136, *D.2135, NUL=
L> */
>> +
>> + =C2=A0if (is_store =3D=3D 0
>> + =C2=A0 =C2=A0 =C2=A0&& TREE_CODE (expr) =3D=3D INDIRECT_REF)
>> + =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0tree ref_target =3D expr->exp.operands [0];
>> + =C2=A0 =C2=A0 =C2=A0if (TREE_CODE (ref_target) =3D=3D SSA_NAME)
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0gimple ref_stmt =3D ref_target->ssa_=
name.def_stmt;
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0if (gimple_code (ref_stmt) =3D=3D GI=
MPLE_ASSIGN)
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0if (gimple_expr_code (=
ref_stmt) =3D=3D POINTER_PLUS_EXPR)
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0tree tmp=
 =3D ref_stmt->gsmem.op [1];
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0if (TREE=
_CODE (tmp) =3D=3D SSA_NAME
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =
=C2=A0&& gimple_code (tmp->ssa_name.def_stmt) =3D=3D GIMPLE_ASSIGN)
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0r=
ef_stmt =3D tmp->ssa_name.def_stmt;
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0}
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0if (gimple_expr_code (=
ref_stmt) =3D=3D COMPONENT_REF
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0&=
& gimple_assign_single_p (ref_stmt))
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0tree com=
p_expr =3D ref_stmt->gsmem.op [1];
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0tree fie=
ld_expr =3D comp_expr->exp.operands [1];
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0if (TREE=
_CODE (field_expr) =3D=3D FIELD_DECL
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =
=C2=A0&& strncmp (decl_name (field_expr),
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =
=C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0"_vptr.", sizeof ("_vptr.")=
 - 1) =3D=3D 0)
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0r=
eturn 1;
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0}
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0}
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0}
>> + =C2=A0 =C2=A0}
>> +
>> + =C2=A0return 0;
>> +}
>> +
>> +/* Checks as to whether expr refers to constant var/field/param.
>> + =C2=A0 Don't bother to instrument them. =C2=A0*/
>> +static int
>> +is_load_of_const (tree expr, int is_store)
>> +{
>> + =C2=A0if (is_store =3D=3D 0)
>> + =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0if (TREE_CODE (expr) =3D=3D COMPONENT_REF)
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0expr =3D expr->exp.operands [1];
>> + =C2=A0 =C2=A0 =C2=A0if (TREE_CODE (expr) =3D=3D VAR_DECL
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0|| TREE_CODE (expr) =3D=3D PARM_DECL
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0|| TREE_CODE (expr) =3D=3D FIELD_DEC=
L)
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0if (TREE_READONLY (expr))
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0return 1;
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0}
>> + =C2=A0 =C2=A0}
>> + =C2=A0return 0;
>> +}
>> +
>> +static void
>> +handle_expr (gimple stmt, gimple_stmt_iterator gsi,
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 tree expr, int is_store, VEC=
 (mop_desc_t, heap) **mop_list)
>> +{
>> + =C2=A0enum tree_code tcode;
>> + =C2=A0struct mop_desc_t mop;
>> + =C2=A0unsigned fld_off;
>> + =C2=A0unsigned fld_size;
>> +
>> + =C2=A0/* map SSA name to real name */
>> + =C2=A0if (TREE_CODE (expr) =3D=3D SSA_NAME)
>> + =C2=A0 =C2=A0expr =3D SSA_NAME_VAR (expr);
>> +
>> + =C2=A0tcode =3D TREE_CODE (expr);
>> +
>> + =C2=A0/* Below are things we do NOT want to instrument. =C2=A0*/
>> + =C2=A0if (func_ignore & (tsan_ignore_mop | tsan_ignore_rec))
>> + =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0return;
>> + =C2=A0 =C2=A0}
>> + =C2=A0else if (TREE_CODE_CLASS (tcode) =3D=3D tcc_constant)
>> + =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0/* various constant literals */
>> + =C2=A0 =C2=A0 =C2=A0return;
>> + =C2=A0 =C2=A0}
>> + =C2=A0else if (TREE_CODE_CLASS (tcode) =3D=3D tcc_declaration
>> + =C2=A0 =C2=A0 =C2=A0&& DECL_ARTIFICIAL (expr))
>> + =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0/* compiler-emitted artificial variables */
>> + =C2=A0 =C2=A0 =C2=A0return;
>> + =C2=A0 =C2=A0}
>> + =C2=A0if (tcode =3D=3D RESULT_DECL)
>> + =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0/* store to function result */
>> + =C2=A0 =C2=A0 =C2=A0return;
>> + =C2=A0 =C2=A0}
>> + =C2=A0else if (tcode =3D=3D VAR_DECL
>> + =C2=A0 =C2=A0 =C2=A0&& TREE_ADDRESSABLE (expr) =3D=3D 0
>> + =C2=A0 =C2=A0 =C2=A0&& TREE_STATIC (expr) =3D=3D 0)
>> + =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0/* the var does not live in memory -> no possibili=
ty of races */
>> + =C2=A0 =C2=A0 =C2=A0return;
>> + =C2=A0 =C2=A0}
>> + =C2=A0else if (TREE_CODE (TREE_TYPE (expr)) =3D=3D RECORD_TYPE)
>> + =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0/* TODO (dvyukov): implement me */
>> + =C2=A0 =C2=A0 =C2=A0return;
>> + =C2=A0 =C2=A0}
>> + =C2=A0else if (tcode =3D=3D CONSTRUCTOR)
>> + =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0/* TODO (dvyukov): implement me */
>> + =C2=A0 =C2=A0 =C2=A0return;
>> + =C2=A0 =C2=A0}
>> + =C2=A0else if (tcode =3D=3D PARM_DECL)
>> + =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0/* TODO (dvyukov): implement me */
>> + =C2=A0 =C2=A0 =C2=A0return;
>> + =C2=A0 =C2=A0}
>> + =C2=A0else if (is_load_of_const (expr, is_store))
>> + =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0/* load of a const variable/parameter/field */
>> + =C2=A0 =C2=A0 =C2=A0return;
>> + =C2=A0 =C2=A0}
>> + =C2=A0else if (is_vtbl_read (expr, is_store))
>> + =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0/* vtbl read */
>> + =C2=A0 =C2=A0 =C2=A0return;
>> + =C2=A0 =C2=A0}
>> + =C2=A0else if (tcode =3D=3D COMPONENT_REF)
>> + =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0tree field =3D expr->exp.operands [1];
>> + =C2=A0 =C2=A0 =C2=A0if (TREE_CODE (field) =3D=3D FIELD_DECL)
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0fld_off =3D field->field_decl.bit_of=
fset->int_cst.int_cst.low;
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0fld_size =3D field->decl_common.size=
->int_cst.int_cst.low;
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0if (((fld_off % BITS_PER_UNIT) !=3D =
0)
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0|| ((fld_size % BITS_P=
ER_UNIT) !=3D 0))
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0/* As of now it crashe=
s compilation.
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 TODO (dvyukov)=
: handle bit-fields as if touching the whole field */
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0return;
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0}
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0}
>> + =C2=A0 =C2=A0}
>> +
>> + =C2=A0/* TODO (dvyukov): handle other cases
>> + =C2=A0 =C2=A0 (FIELD_DECL, MEM_REF, ARRAY_RANGE_REF, TARGET_MEM_REF, A=
DDR_EXPR) */
>> + =C2=A0if (tcode !=3D ARRAY_REF
>> + =C2=A0 =C2=A0 =C2=A0&& tcode !=3D VAR_DECL
>> + =C2=A0 =C2=A0 =C2=A0&& tcode !=3D COMPONENT_REF
>> + =C2=A0 =C2=A0 =C2=A0&& tcode !=3D INDIRECT_REF
>> + =C2=A0 =C2=A0 =C2=A0&& tcode !=3D MEM_REF)
>> + =C2=A0 =C2=A0return;
>> +
>> + =C2=A0mop.is_call =3D 0;
>> + =C2=A0mop.gsi =3D gsi;
>> + =C2=A0mop.expr =3D expr;
>> + =C2=A0mop.dtor_vptr_expr =3D is_dtor_vptr_store (stmt, expr, is_store);
>> + =C2=A0mop.is_store =3D is_store;
>> + =C2=A0VEC_safe_push (mop_desc_t, heap, *mop_list, &mop);
>> +}
>> +
>> +static void
>> +handle_gimple (gimple_stmt_iterator gsi, VEC (mop_desc_t, heap) **mop_l=
ist)
>> +{
>> + =C2=A0unsigned i;
>> + =C2=A0struct mop_desc_t mop;
>> + =C2=A0gimple stmt;
>> + =C2=A0enum gimple_code gcode;
>> + =C2=A0location_t loc;
>> + =C2=A0tree rhs;
>> + =C2=A0tree lhs;
>> +
>> + =C2=A0stmt =3D gsi_stmt (gsi);
>> + =C2=A0gcode =3D gimple_code (stmt);
>> + =C2=A0if (gcode >=3D LAST_AND_UNUSED_GIMPLE_CODE)
>> + =C2=A0 =C2=A0return;
>> +
>> + =C2=A0loc =3D gimple_location (stmt);
>> +
>> + =C2=A0switch (gcode)
>> + =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0/* TODO (dvyukov): handle GIMPLE_COND (can it acce=
ss memmory?) */
>> + =C2=A0 =C2=A0 =C2=A0case GIMPLE_CALL:
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0func_calls +=3D 1;
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0/* Handle call arguments as loads */
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0for (i =3D 0; i < gimple_call_num_ar=
gs (stmt); i++)
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0rhs =3D gimple_call_ar=
g (stmt, i);
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0handle_expr (stmt, gsi=
, rhs, 0, mop_list);
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0}
>> +
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0memset (&mop, 0, sizeof (mop));
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0mop.is_call =3D 1;
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0VEC_safe_push (mop_desc_t, heap, *mo=
p_list, &mop);
>> +
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0/* Handle assignment lhs as store */
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0lhs =3D gimple_call_lhs (stmt);
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0if (lhs !=3D 0)
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0handle_expr (stmt, gsi, lhs, =
1, mop_list);
>> +
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0break;
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0}
>> +
>> + =C2=A0 =C2=A0 =C2=A0case GIMPLE_ASSIGN:
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0/* Handle assignment lhs as store */
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0lhs =3D gimple_assign_lhs (stmt);
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0handle_expr (stmt, gsi, lhs, 1, mop_=
list);
>> +
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0/* Handle operands as loads */
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0for (i =3D 1; i < gimple_num_ops (st=
mt); i++)
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0rhs =3D gimple_op (stm=
t, i);
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0handle_expr (stmt, gsi=
, rhs, 0, mop_list);
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0}
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0break;
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0}
>> +
>> + =C2=A0 =C2=A0 =C2=A0case GIMPLE_BIND:
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0gcc_assert (!"there should be no GIM=
PLE_BIND on this level");
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0break;
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0}
>> +
>> + =C2=A0 =C2=A0 =C2=A0default:
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0break;
>> + =C2=A0 =C2=A0}
>> +}
>> +
>> +/* Instruments single basic block. =C2=A0*/
>> +static void
>> +instrument_bblock (struct bb_data_t *bbd, basic_block bb)
>> +{
>> + =C2=A0int ix;
>> + =C2=A0int is_sblock;
>> + =C2=A0gimple_stmt_iterator gsi;
>> + =C2=A0struct mop_desc_t *mop;
>> + =C2=A0gimple stmt;
>> + =C2=A0location_t loc;
>> + =C2=A0expanded_location eloc;
>> + =C2=A0gimple_seq instr_seq;
>> +
>> + =C2=A0/* Iterate over all gimples and collect interesting mops into mo=
p_list. =C2=A0*/
>> + =C2=A0VEC_free (mop_desc_t, heap, mop_list);
>> + =C2=A0for (gsi =3D gsi_start_bb (bb); !gsi_end_p (gsi); gsi_next (&gsi=
))
>> + =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0handle_gimple (gsi, &mop_list);
>> + =C2=A0 =C2=A0}
>> +
>> + =C2=A0mop =3D 0;
>> + =C2=A0for (ix =3D 0; VEC_iterate (mop_desc_t, mop_list, ix, mop); ix +=
=3D 1)
>> + =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0if (mop->is_call !=3D 0)
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0/* After a function call we must sta=
rt a brand new sblock,
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 because the function can con=
tain synchronization. =C2=A0*/
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0bbd->has_sb =3D 0;
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0continue;
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0}
>> +
>> + =C2=A0 =C2=A0 =C2=A0func_mops +=3D 1;
>> + =C2=A0 =C2=A0 =C2=A0stmt =3D gsi_stmt (mop->gsi);
>> + =C2=A0 =C2=A0 =C2=A0loc =3D gimple_location (stmt);
>> + =C2=A0 =C2=A0 =C2=A0eloc =3D expand_location (loc);
>> +
>> + =C2=A0 =C2=A0 =C2=A0/* Check as to whether we may not set sblock flag
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 for the access */
>> + =C2=A0 =C2=A0 =C2=A0is_sblock =3D (bbd->has_sb =3D=3D 0
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0|| !(eloc.file !=3D 0
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0&& bbd->sb_file !=3D 0
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0&& strcmp (eloc.file, =
bbd->sb_file) =3D=3D 0
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0&& eloc.line >=3D bbd-=
>sb_line_min
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0&& eloc.line <=3D bbd-=
>sb_line_max));
>> +
>> + =C2=A0 =C2=A0 =C2=A0if (func_ignore =3D=3D tsan_ignore_hist)
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0is_sblock =3D 0;
>> +
>> + =C2=A0 =C2=A0 =C2=A0if (is_sblock)
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0/* Start new sblock with new source =
info. =C2=A0*/
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0bbd->has_sb =3D 1;
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0bbd->sb_file =3D eloc.file;
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0bbd->sb_line_min =3D eloc.line;
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0bbd->sb_line_max =3D eloc.line + SBL=
OCK_SIZE;
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0}
>> +
>> + =C2=A0 =C2=A0 =C2=A0instr_seq =3D 0;
>> + =C2=A0 =C2=A0 =C2=A0if (mop->dtor_vptr_expr =3D=3D 0)
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0instr_mop (mop->expr, mop->is_store, is_sbl=
ock, &instr_seq);
>> + =C2=A0 =C2=A0 =C2=A0else
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0instr_vptr_store (mop->expr, mop->dtor_vptr=
_expr, is_sblock, &instr_seq);
>> + =C2=A0 =C2=A0 =C2=A0gcc_assert (instr_seq !=3D 0);
>> + =C2=A0 =C2=A0 =C2=A0set_location (instr_seq, loc);
>> + =C2=A0 =C2=A0 =C2=A0/* Instrumentation for assignment of a function re=
sult
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 must be inserted after the call. =C2=A0Ins=
trumentation for
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 reads of function arguments must be insert=
ed before the call.
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 That's because the call can contain synchr=
onization. =C2=A0*/
>> + =C2=A0 =C2=A0 =C2=A0if (is_gimple_call (stmt) && mop->is_store =3D=3D =
1)
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0gsi_insert_seq_after (&mop->gsi, instr_seq,=
 GSI_NEW_STMT);
>> + =C2=A0 =C2=A0 =C2=A0else
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0gsi_insert_seq_before (&mop->gsi, instr_seq=
, GSI_SAME_STMT);
>> + =C2=A0 =C2=A0}
>> +}
>> +
>> +/* Instruments all interesting memory accesses in the function */
>> +static void
>> +instrument_mops (void)
>> +{
>> + =C2=A0int sb_line_min;
>> + =C2=A0int sb_line_max;
>> + =C2=A0int bb_cnt;
>> + =C2=A0int eidx;
>> + =C2=A0basic_block bb;
>> + =C2=A0basic_block entry_bb;
>> + =C2=A0basic_block cur_bb;
>> + =C2=A0basic_block any_bb;
>> + =C2=A0struct bb_data_t *pred;
>> + =C2=A0struct bb_data_t *succ;
>> + =C2=A0struct bb_data_t *bb_data;
>> + =C2=A0struct bb_data_t *bbd;
>> + =C2=A0edge entry_edge;
>> + =C2=A0edge e;
>> +
>> + =C2=A0/* The function does breadth-first traversal of CFG.
>> + =C2=A0 =C2=A0 BB is visited preferably if all its predecessors are vis=
ited.
>> + =C2=A0 =C2=A0 Such order is required to properly mark super-blocks.
>> + =C2=A0 =C2=A0 The idea behind super-blocks is as follows.
>> + =C2=A0 =C2=A0 If several memory accesses happen within SBLOCK_SIZE sou=
rce code lines
>> + =C2=A0 =C2=A0 from each other, then we only mark the first access as S=
BLOCK.
>> + =C2=A0 =C2=A0 This allows runtime library to memorize stack trace
>> + =C2=A0 =C2=A0 only for the first access and do not memorize for others.
>> + =C2=A0 =C2=A0 This significantly reduces memory consumption in exchang=
e for slightly
>> + =C2=A0 =C2=A0 imprecise stack traces for previous accesses. =C2=A0*/
>> +
>> + =C2=A0/* First, mark all blocks as not visited, and entry block as can=
didate. =C2=A0*/
>> + =C2=A0bb_cnt =3D cfun->cfg->x_n_basic_blocks;
>> + =C2=A0bb_data =3D (struct bb_data_t*) xcalloc (bb_cnt, sizeof (struct =
bb_data_t));
>> + =C2=A0entry_bb =3D ENTRY_BLOCK_PTR;
>> + =C2=A0entry_edge =3D single_succ_edge (entry_bb);
>> + =C2=A0entry_bb =3D entry_edge->dest;
>> + =C2=A0bb =3D 0;
>> + =C2=A0FOR_EACH_BB (bb)
>> + =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0bb_data [bb->index].state =3D (bb =3D=3D entry_bb)=
 ? bb_candidate : bb_not_visited;
>> + =C2=A0 =C2=A0}
>> +
>> + =C2=A0/* Until all blocks are visited. =C2=A0*/
>> + =C2=A0for (; ; )
>> + =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0cur_bb =3D 0;
>> + =C2=A0 =C2=A0 =C2=A0any_bb =3D 0;
>> + =C2=A0 =C2=A0 =C2=A0/* Look for a candidate with all visited predecess=
ors. =C2=A0*/
>> + =C2=A0 =C2=A0 =C2=A0FOR_EACH_BB (bb)
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0bbd =3D &bb_data [bb->index];
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0if (bbd->state =3D=3D bb_candidate)
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0cur_bb =3D bb;
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0any_bb =3D bb;
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0e =3D 0;
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0for (eidx =3D 0; VEC_i=
terate (edge, bb->preds, eidx, e); eidx++)
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0pred =3D=
 &bb_data [e->src->index];
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0if (pred=
->state !=3D bb_visited)
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =
=C2=A0cur_bb =3D 0;
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =
=C2=A0break;
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0}
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0}
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0}
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0if (cur_bb !=3D 0)
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0break;
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0}
>> + =C2=A0 =C2=A0 =C2=A0/* All blocks are visited. =C2=A0*/
>> + =C2=A0 =C2=A0 =C2=A0if (any_bb =3D=3D 0)
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0break;
>> + =C2=A0 =C2=A0 =C2=A0/* If no blocks with all visited predecessors, cho=
ose any candidate.
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 Must be a loop. =C2=A0*/
>> + =C2=A0 =C2=A0 =C2=A0cur_bb =3D cur_bb ? cur_bb : any_bb;
>> + =C2=A0 =C2=A0 =C2=A0bbd =3D &bb_data [cur_bb->index];
>> + =C2=A0 =C2=A0 =C2=A0gcc_assert (bbd->state =3D=3D bb_candidate);
>> + =C2=A0 =C2=A0 =C2=A0bbd->state =3D bb_visited;
>> +
>> + =C2=A0 =C2=A0 =C2=A0/* Iterate over all predecessors and merge their s=
block info. =C2=A0*/
>> + =C2=A0 =C2=A0 =C2=A0e =3D 0;
>> + =C2=A0 =C2=A0 =C2=A0for (eidx =3D 0; VEC_iterate (edge, cur_bb->preds,=
 eidx, e); eidx++)
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0pred =3D &bb_data [e->src->index];
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0if ((pred->state !=3D bb_visited)
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0|| (pred->has_sb =3D=
=3D 0)
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0|| (pred =3D=3D bbd))
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0/* If there is a not v=
isited predecessor,
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 or a predecess=
or with no active sblock info,
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 or a self-loop=
, then we will have to start
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 a brand new sb=
lock on next memory access. =C2=A0*/
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0bbd->has_sb =3D 0;
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0break;
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0}
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0else if (bbd->has_sb =3D=3D 0)
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0/* If it's a first pre=
decessor, just copy the info. =C2=A0*/
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0bbd->has_sb =3D 1;
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0bbd->sb_file =3D pred-=
>sb_file;
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0bbd->sb_line_min =3D p=
red->sb_line_min;
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0bbd->sb_line_max =3D p=
red->sb_line_max;
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0}
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0else
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0/* Otherwise, find the=
 interception
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 between two sb=
lock descriptors. =C2=A0*/
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0bbd->has_sb =3D 0;
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0if (bbd->sb_file !=3D =
0 && pred->sb_file !=3D 0
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0&& strcm=
p (bbd->sb_file, pred->sb_file) =3D=3D 0)
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0sb_line_=
min =3D MAX (bbd->sb_line_min, pred->sb_line_min);
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0sb_line_=
max =3D MIN (bbd->sb_line_max, pred->sb_line_max);
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0if (sb_l=
ine_min <=3D sb_line_max)
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =
=C2=A0bbd->has_sb =3D 1;
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =
=C2=A0bbd->sb_line_min =3D sb_line_min;
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =
=C2=A0bbd->sb_line_max =3D sb_line_max;
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0}
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0}
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0/* No interception, ha=
ve to start new sblock. =C2=A0*/
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0if (bbd->has_sb =3D=3D=
 0)
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0break;
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0}
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0}
>> +
>> + =C2=A0 =C2=A0 =C2=A0/* Finally, instrument the block. =C2=A0*/
>> + =C2=A0 =C2=A0 =C2=A0instrument_bblock (bbd, cur_bb);
>> +
>> + =C2=A0 =C2=A0 =C2=A0/* Mark all successors as candidates. =C2=A0*/
>> + =C2=A0 =C2=A0 =C2=A0for (eidx =3D 0; VEC_iterate (edge, cur_bb->succs,=
 eidx, e); eidx++)
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0succ =3D &bb_data [e->dest->index];
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0if (succ->state =3D=3D bb_not_visite=
d)
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0succ->state =3D bb_candidate;
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0}
>> + =C2=A0 =C2=A0}
>> +}
>> +
>> +/* Instruments function entry and exit, if necessary. =C2=A0*/
>> +static void
>> +instrument_function (void)
>> +{
>> + =C2=A0location_t loc;
>> + =C2=A0gimple_seq pre_func_seq;
>> + =C2=A0gimple_seq post_func_seq;
>> + =C2=A0basic_block entry_bb;
>> + =C2=A0basic_block first_bb;
>> + =C2=A0basic_block bb;
>> + =C2=A0edge entry_edge;
>> + =C2=A0gimple_stmt_iterator first_gsi;
>> + =C2=A0gimple_stmt_iterator gsi;
>> + =C2=A0gimple_stmt_iterator gsi2;
>> + =C2=A0gimple first_stmt;
>> + =C2=A0gimple stmt;
>> +
>> + =C2=A0pre_func_seq =3D 0;
>> + =C2=A0post_func_seq =3D 0;
>> + =C2=A0instr_func (&pre_func_seq, &post_func_seq);
>> +
>> + =C2=A0if (pre_func_seq !=3D 0)
>> + =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0/* Insert new BB before the first BB. =C2=A0*/
>> + =C2=A0 =C2=A0 =C2=A0entry_bb =3D ENTRY_BLOCK_PTR;
>> + =C2=A0 =C2=A0 =C2=A0entry_edge =3D single_succ_edge (entry_bb);
>> + =C2=A0 =C2=A0 =C2=A0first_bb =3D entry_edge->dest;
>> + =C2=A0 =C2=A0 =C2=A0first_gsi =3D gsi_start_bb (first_bb);
>> + =C2=A0 =C2=A0 =C2=A0if (!gsi_end_p (first_gsi))
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0first_stmt =3D gsi_stmt (first_gsi);
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0loc =3D gimple_location (first_stmt);
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0set_location (pre_func_seq, loc);
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0}
>> + =C2=A0 =C2=A0 =C2=A0entry_bb =3D split_edge (entry_edge);
>> + =C2=A0 =C2=A0 =C2=A0gsi =3D gsi_start_bb (entry_bb);
>> + =C2=A0 =C2=A0 =C2=A0gsi_insert_seq_after (&gsi, pre_func_seq, GSI_NEW_=
STMT);
>> + =C2=A0 =C2=A0}
>> +
>> + =C2=A0if (post_func_seq !=3D 0)
>> + =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0/* Find all function exits. =C2=A0*/
>> + =C2=A0 =C2=A0 =C2=A0FOR_EACH_BB (bb)
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0gsi2 =3D gsi_start_bb (bb);
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0for (; ; )
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0gsi =3D gsi2;
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0if (gsi_end_p (gsi))
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0break;
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0gsi_next (&gsi2);
>> +
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0stmt =3D gsi_stmt (gsi=
);
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0loc =3D gimple_locatio=
n (stmt);
>> +
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0if (gimple_code (stmt)=
 =3D=3D GIMPLE_RETURN)
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0set_loca=
tion (post_func_seq, loc);
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0gsi_inse=
rt_seq_before (&gsi, post_func_seq, GSI_SAME_STMT);
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0}
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0}
>> + =C2=A0 =C2=A0 =C2=A0 =C2=A0}
>> + =C2=A0 =C2=A0}
>> +}
>> +
>> +static unsigned
>> +tsan_pass (void)
>> +{
>> + =C2=A0if (errorcount !=3D 0 || sorrycount !=3D 0)
>> + =C2=A0 =C2=A0return 0;
>> +
>> + =C2=A0func_ignore =3D tsan_ignore ();
>> + =C2=A0if (func_ignore =3D=3D tsan_ignore_func)
>> + =C2=A0 =C2=A0return 0;
>> +
>> + =C2=A0func_calls =3D 0;
>> + =C2=A0func_mops =3D 0;
>> +
>> + =C2=A0instrument_mops ();
>> + =C2=A0instrument_function ();
>> +
>> + =C2=A0return 0;
>> +}
>> +
>> +static bool
>> +tsan_gate (void)
>> +{
>> + =C2=A0return flag_tsan !=3D 0;
>> +}
>> +
>> +struct gimple_opt_pass pass_tsan =3D {{
>> + =C2=A0GIMPLE_PASS,
>> + =C2=A0"tsan", =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =
=C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 /* name */
>> + =C2=A0tsan_gate, =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=
=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0/* gate */
>> + =C2=A0tsan_pass, =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=
=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0/* execute */
>> + =C2=A0NULL, =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =
=C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 /* sub */
>> + =C2=A0NULL, =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =
=C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 /* next */
>> + =C2=A00, =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=
=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0/* static=
_pass_number */
>> + =C2=A0TV_NONE, =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0=
 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0/* tv_id */
>> + =C2=A0PROP_trees | PROP_cfg, =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0=
 =C2=A0 =C2=A0/* properties_required */
>> + =C2=A00, =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=
=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0/* proper=
ties_provided */
>> + =C2=A00, =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=
=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0/* proper=
ties_destroyed */
>> + =C2=A00, =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=
=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0 =C2=A0/* todo_f=
lags_start */
>> + =C2=A0TODO_dump_cgraph | TODO_dump_func | TODO_verify_all
>> + =C2=A0 =C2=A0| TODO_update_ssa | TODO_update_address_taken /* todo_fla=
gs_finish */
>> +}};
>> +
>>
>> Property changes on: gcc/tree-tsan.c
>> ___________________________________________________________________
>> Added: svn:eol-style
>> =C2=A0 + LF
>>
>> Index: gcc/tree-pass.h
>> =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
>> --- gcc/tree-pass.h =C2=A0 =C2=A0 (revision 180522)
>> +++ gcc/tree-pass.h =C2=A0 =C2=A0 (working copy)
>> @@ -352,6 +352,7 @@
>>
>> =C2=A0extern struct gimple_opt_pass pass_mudflap_1;
>> =C2=A0extern struct gimple_opt_pass pass_mudflap_2;
>> +extern struct gimple_opt_pass pass_tsan;
>> =C2=A0extern struct gimple_opt_pass pass_lower_cf;
>> =C2=A0extern struct gimple_opt_pass pass_refactor_eh;
>> =C2=A0extern struct gimple_opt_pass pass_lower_eh;
>> Index: gcc/testsuite/gcc.dg/tsan-ignore.ignore
>> =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
>> --- gcc/testsuite/gcc.dg/tsan-ignore.ignore =C2=A0 =C2=A0 (revision 0)
>> +++ gcc/testsuite/gcc.dg/tsan-ignore.ignore =C2=A0 =C2=A0 (revision 0)
>> @@ -0,0 +1,7 @@
>> +#comment
>> +fun:foo
>> +fun:*bar
>> +fun:baz*
>> +fun:*bla*
>> +fun:x*x
>> +src:tsan-ignore.h
>> Index: gcc/testsuite/gcc.dg/tsan.h
>> =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
>> --- gcc/testsuite/gcc.dg/tsan.h (revision 0)
>> +++ gcc/testsuite/gcc.dg/tsan.h (revision 0)
>> @@ -0,0 +1,87 @@
>> +/* Helper declarations and functions for ThreadSanitizer instrumentatio=
n (-ftsan) testing */
>> +
>> +int printf (char *str, ...);
>> +void exit (int);
>> +
>> +/* Variables referenced by the instrumentation */
>> +__thread void **__tsan_shadow_stack;
>> +__thread int __tsan_thread_ignore;
>> +
>> +/* Local helper vars */
>> +__thread void *shadow_stack[1024];
>> +__thread int mop_expect;
>> +__thread int mop_depth;
>> +__thread void* mop_addr;
>> +__thread unsigned long long mop_pc;
>> +__thread unsigned mop_flags;
>> +
>> +/* Setups shadow stack var (not instrumented) */
>> +void __attribute__ ((constructor))
>> +__tsan_init (void)
>> +{
>> + =C2=A0__tsan_shadow_stack =3D shadow_stack;
>> +}
>> +
>> +/* Declare that we expect an instrumented memory access (not instrument=
ed).
>> + =C2=A0 depth - stack depth of the mop (0 - main, 1 - func called from =
main and so on).
>> + =C2=A0 addr - memory access address.
>> + =C2=A0 is_store - store/load.
>> + =C2=A0 is_sblock - superblock flag of the access.
>> + =C2=A0 size - memory access size in bytes. */
>> +void
>> +__tsan_expect_mop (int depth, void *addr, int is_store, int is_sblock, =
int size)
>> +{
>> + =C2=A0if (mop_expect)
>> + =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0printf ("missed mop: addr=3D%p pc=3D%d\n", mop_add=
r, mop_pc);
>> + =C2=A0 =C2=A0 =C2=A0exit (1);
>> + =C2=A0 =C2=A0}
>> +
>> + =C2=A0mop_expect =3D 1;
>> + =C2=A0mop_depth =3D depth;
>> + =C2=A0mop_addr =3D addr;
>> + =C2=A0mop_pc =3D (unsigned long long)__builtin_return_address(0);
>> + =C2=A0mop_flags =3D !!is_sblock | (!!is_store << 1) | ((size - 1) << 2=
);
>> +}
>> +
>> +/* Memory access function (referenced by instrumentation, not instrumen=
ted). */
>> +void
>> +__tsan_handle_mop (void *addr, unsigned flags)
>> +{
>> + =C2=A0unsigned long long pc;
>> + =C2=A0int depth;
>> +
>> + =C2=A0printf ("mop: addr=3D%p flags=3D%x called from %p/%p\n", addr, f=
lags, __builtin_return_address(1), __tsan_shadow_stack[-2]);
>> + =C2=A0if (mop_expect =3D=3D 0)
>> + =C2=A0 =C2=A0return;
>> +
>> + =C2=A0/* Verify parameters with what we expect. */
>> +
>> + =C2=A0if (addr !=3D mop_addr)
>> + =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0printf ("incorrect mop addr: %p/%p\n", addr, =C2=
=A0mop_addr);
>> + =C2=A0 =C2=A0 =C2=A0exit (1);
>> + =C2=A0 =C2=A0}
>> +
>> + =C2=A0pc =3D (unsigned long long)__builtin_return_address(0);
>> + =C2=A0if (pc < mop_pc - 100 || pc > mop_pc + 100)
>> + =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0printf ("incorrect mop pc: %p/%p\n", (void*)pc, (v=
oid*)mop_pc);
>> + =C2=A0 =C2=A0 =C2=A0exit (1);
>> + =C2=A0 =C2=A0}
>> +
>> + =C2=A0depth =3D __tsan_shadow_stack - shadow_stack - 1;
>> + =C2=A0if (depth !=3D mop_depth)
>> + =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0printf ("incorrect mop depth: %d/%d\n", depth, mop=
_depth);
>> + =C2=A0 =C2=A0 =C2=A0exit (1);
>> + =C2=A0 =C2=A0}
>> +
>> + =C2=A0if (flags !=3D mop_flags)
>> + =C2=A0 =C2=A0{
>> + =C2=A0 =C2=A0 =C2=A0printf ("incorrect mop flags: %x/%x\n", flags, mop=
_flags);
>> + =C2=A0 =C2=A0 =C2=A0exit (1);
>> + =C2=A0 =C2=A0}
>> +
>> + =C2=A0mop_expect =3D 0;
>> +}
>>
>> Property changes on: gcc/testsuite/gcc.dg/tsan.h
>> ___________________________________________________________________
>> Added: svn:eol-style
>> =C2=A0 + LF
>>
>> Index: gcc/testsuite/gcc.dg/tsan-ignore.c
>> =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
>> --- gcc/testsuite/gcc.dg/tsan-ignore.c =C2=A0(revision 0)
>> +++ gcc/testsuite/gcc.dg/tsan-ignore.c =C2=A0(revision 0)
>> @@ -0,0 +1,49 @@
>> +/* { dg-do run } */
>> +/* { dg-options "-ftsan -ftsan-ignore=3Dtsan-ignore.ignore" } */
>> +#include "tsan.h"
>> +#include "tsan-ignore.h"
>> +
>> +/* Check ignore file handling. */
>> +
>> +int
>> +foo (int *p)
>> +{
>> + =C2=A0p [0] =3D 1;
>> +}
>> +
>> +int bar (int *p)
>> +{
>> + =C2=A0p [0] =3D 1;
>> +}
>> +
>> +int baz (int *p)
>> +{
>> + =C2=A0p [0] =3D 1;
>> +}
>> +
>> +int bla (int *p)
>> +{
>> + =C2=A0p [0] =3D 1;
>> +}
>> +
>> +int xxx (int *p)
>> +{
>> + =C2=A0p [0] =3D 1;
>> +}
>> +
>> +int
>> +main (void)
>> +{
>> + =C2=A0int p, x;
>> +
>> + =C2=A0__tsan_expect_mop(0, &p, 1, 1, sizeof(p));
>> + =C2=A0/* All these functions must be ignored. */
>> + =C2=A0foo (&x);
>> + =C2=A0bar (&x);
>> + =C2=A0baz (&x);
>> + =C2=A0bla (&x);
>> + =C2=A0xxx (&x);
>> + =C2=A0in_tsan_ignore_header (&x);
>> + =C2=A0p =3D 0;
>> + =C2=A0return 0;
>> +}
>>
>> Property changes on: gcc/testsuite/gcc.dg/tsan-ignore.c
>> ___________________________________________________________________
>> Added: svn:eol-style
>> =C2=A0 + LF
>>
>> Index: gcc/testsuite/gcc.dg/tsan-ignore.h
>> =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
>> --- gcc/testsuite/gcc.dg/tsan-ignore.h =C2=A0(revision 0)
>> +++ gcc/testsuite/gcc.dg/tsan-ignore.h =C2=A0(revision 0)
>> @@ -0,0 +1,5 @@
>> +int
>> +in_tsan_ignore_header (int *p)
>> +{
>> + =C2=A0p [0] =3D 1;
>> +}
>>
>> Property changes on: gcc/testsuite/gcc.dg/tsan-ignore.h
>> ___________________________________________________________________
>> Added: svn:eol-style
>> =C2=A0 + LF
>>
>> Index: gcc/testsuite/gcc.dg/tsan-stack.c
>> =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
>> --- gcc/testsuite/gcc.dg/tsan-stack.c =C2=A0 (revision 0)
>> +++ gcc/testsuite/gcc.dg/tsan-stack.c =C2=A0 (revision 0)
>> @@ -0,0 +1,23 @@
>> +/* { dg-do run } */
>> +/* { dg-options "-ftsan" } */
>> +#include "tsan.h"
>> +
>> +/* Check shadow stack maintance. */
>> +
>> +int
>> +foobar (int *p)
>> +{
>> + =C2=A0__tsan_expect_mop(1, p, 1, 1, sizeof(*p));
>> + =C2=A0p[0] =3D 1;
>> +}
>> +
>> +int
>> +main (void)
>> +{
>> + =C2=A0int p;
>> +
>> + =C2=A0__tsan_expect_mop(0, &p, 1, 1, sizeof(p));
>> + =C2=A0p =3D 0;
>> + =C2=A0foobar (&p);
>> + =C2=A0return 0;
>> +}
>>
>> Property changes on: gcc/testsuite/gcc.dg/tsan-stack.c
>> ___________________________________________________________________
>> Added: svn:eol-style
>> =C2=A0 + LF
>>
>> Index: gcc/testsuite/gcc.dg/tsan-mop.c
>> =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
>> --- gcc/testsuite/gcc.dg/tsan-mop.c =C2=A0 =C2=A0 (revision 0)
>> +++ gcc/testsuite/gcc.dg/tsan-mop.c =C2=A0 =C2=A0 (revision 0)
>> @@ -0,0 +1,29 @@
>> +/* { dg-do run } */
>> +/* { dg-options "-ftsan" } */
>> +#include "tsan.h"
>> +
>> +/* Sanity check for memory accesses instrumentation. */
>> +
>> +int
>> +foobar (int *p)
>> +{
>> + =C2=A0__tsan_expect_mop(1, p, 1, 1, sizeof(*p));
>> + =C2=A0p[0] =3D 1;
>> +
>> + =C2=A0__tsan_expect_mop(1, p, 1, 1, sizeof(*p));
>> + =C2=A0*p =3D 2;
>> +
>> + =C2=A0__tsan_expect_mop(1, (char*)p+3, 1, 1, 1);
>> + =C2=A0*((char*)p+3) =3D 3;
>> +}
>> +
>> +int
>> +main (void)
>> +{
>> + =C2=A0int p;
>> +
>> + =C2=A0__tsan_expect_mop(0, &p, 1, 1, sizeof(p));
>> + =C2=A0p =3D 0;
>> + =C2=A0foobar (&p);
>> + =C2=A0return 0;
>> +}
>>
>> Property changes on: gcc/testsuite/gcc.dg/tsan-mop.c
>> ___________________________________________________________________
>> Added: svn:eol-style
>> =C2=A0 + LF
>>
>> Index: gcc/common.opt
>> =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
>> --- gcc/common.opt =C2=A0 =C2=A0 =C2=A0(revision 180522)
>> +++ gcc/common.opt =C2=A0 =C2=A0 =C2=A0(working copy)
>> @@ -1547,6 +1547,14 @@
>> =C2=A0Common RejectNegative Report Var(flag_mudflap_ignore_reads)
>> =C2=A0Ignore read operations when inserting mudflap instrumentation
>>
>> +ftsan
>> +Common RejectNegative Report Var(flag_tsan)
>> +Add ThreadSanitizer instrumentation
>> +
>> +ftsan-ignore=3D
>> +Common RejectNegative Joined Var(flag_tsan_ignore)
>> +-ftsan-ignore=3Dfilename ThreadSanitizer ignore file
>> +
>> =C2=A0fdce
>> =C2=A0Common Var(flag_dce) Init(1) Optimization
>> =C2=A0Use the RTL dead code elimination pass
>> Index: gcc/Makefile.in
>> =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
>> --- gcc/Makefile.in =C2=A0 =C2=A0 (revision 180522)
>> +++ gcc/Makefile.in =C2=A0 =C2=A0 (working copy)
>> @@ -1494,6 +1494,7 @@
>> =C2=A0 =C2=A0 =C2=A0 =C2=A0tree-streamer-out.o \
>> =C2=A0 =C2=A0 =C2=A0 =C2=A0tree-tailcall.o \
>> =C2=A0 =C2=A0 =C2=A0 =C2=A0tree-threadsafe-analyze.o \
>> + =C2=A0 =C2=A0 =C2=A0 tree-tsan.o \
>> =C2=A0 =C2=A0 =C2=A0 =C2=A0tree-vect-generic.o \
>> =C2=A0 =C2=A0 =C2=A0 =C2=A0tree-vect-patterns.o \
>> =C2=A0 =C2=A0 =C2=A0 =C2=A0tree-vect-data-refs.o \
>> @@ -2814,6 +2815,12 @@
>> =C2=A0 =C2=A0$(C_TREE_H) $(C_COMMON_H) $(GIMPLE_H) $(DIAGNOSTIC_H) $(HAS=
HTAB_H) \
>> =C2=A0 =C2=A0output.h langhooks.h tree-mudflap.h $(TM_H) coretypes.h \
>> =C2=A0 =C2=A0$(GGC_H) gt-tree-mudflap.h $(TREE_PASS_H) $(DIAGNOSTIC_CORE=
_H)
>> +tree-tsan.o : $(CONFIG_H) $(SYSTEM_H) $(TREE_H) $(TREE_INLINE_H) \
>> + =C2=A0 $(GIMPLE_H) $(DIAGNOSTIC_H) langhooks.h \
>> + =C2=A0 $(TM_H) coretypes.h $(TREE_DUMP_H) $(TREE_PASS_H) $(CGRAPH_H) $=
(GGC_H) \
>> + =C2=A0 $(BASIC_BLOCK_H) $(FLAGS_H) $(FUNCTION_H) \
>> + =C2=A0 $(TM_P_H) $(TREE_FLOW_H) $(DIAGNOSTIC_CORE_H) $(GIMPLE_H) tree-=
iterator.h \
>> + =C2=A0 intl.h cfghooks.h output.h options.h c-family/c-common.h
>> =C2=A0tree-pretty-print.o : tree-pretty-print.c $(CONFIG_H) $(SYSTEM_H) \
>> =C2=A0 =C2=A0$(TREE_H) $(DIAGNOSTIC_H) $(HASHTAB_H) $(TREE_FLOW_H) \
>> =C2=A0 =C2=A0$(TM_H) coretypes.h tree-iterator.h $(SCEV_H) langhooks.h \
>> Index: gcc/passes.c
>> =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
>> --- gcc/passes.c =C2=A0 =C2=A0 =C2=A0 =C2=A0(revision 180522)
>> +++ gcc/passes.c =C2=A0 =C2=A0 =C2=A0 =C2=A0(working copy)
>> @@ -1420,6 +1420,7 @@
>> =C2=A0 NEXT_PASS (pass_lower_resx);
>> =C2=A0 NEXT_PASS (pass_nrv);
>> =C2=A0 NEXT_PASS (pass_mudflap_2);
>> + =C2=A0NEXT_PASS (pass_tsan);
>> =C2=A0 NEXT_PASS (pass_cleanup_cfg_post_optimizing);
>> =C2=A0 NEXT_PASS (pass_warn_function_noreturn);
>>
>>
>> --
>> This patch is available for review at http://codereview.appspot.com/5303=
083
>>
>> --
>> You received this message because you are subscribed to the Google Group=
s "C-compiler-team" group.
>> To post to this group, send email to c-compiler-team@google.com.
>> To unsubscribe from this group, send email to c-compiler-team+unsubscrib=
e@google.com.
>> For more options, visit this group at http://groups.google.com/a/google.=
com/group/c-compiler-team/?hl=3Den.
>>
>>
>
> --
> You received this message because you are subscribed to the Google Groups=
 "C-compiler-team" group.
> To post to this group, send email to c-compiler-team@google.com.
> To unsubscribe from this group, send email to c-compiler-team+unsubscribe=
@google.com.
> For more options, visit this group at http://groups.google.com/a/google.c=
om/group/c-compiler-team/?hl=3Den.
>
>