From patchwork Thu Jun 24 19:38:15 2010 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Steven Bosscher X-Patchwork-Id: 56825 Return-Path: X-Original-To: incoming@patchwork.ozlabs.org Delivered-To: patchwork-incoming@bilbo.ozlabs.org Received: from sourceware.org (server1.sourceware.org [209.132.180.131]) by ozlabs.org (Postfix) with SMTP id D0ED4B6F1A for ; Fri, 25 Jun 2010 05:39:04 +1000 (EST) Received: (qmail 31058 invoked by alias); 24 Jun 2010 19:39:02 -0000 Received: (qmail 31011 invoked by uid 22791); 24 Jun 2010 19:38:52 -0000 X-SWARE-Spam-Status: No, hits=-0.4 required=5.0 tests=AWL, BAYES_50, DKIM_SIGNED, DKIM_VALID, DKIM_VALID_AU, FREEMAIL_FROM, RCVD_IN_DNSWL_NONE, TW_CP X-Spam-Check-By: sourceware.org Received: from mail-pv0-f175.google.com (HELO mail-pv0-f175.google.com) (74.125.83.175) by sourceware.org (qpsmtpd/0.43rc1) with ESMTP; Thu, 24 Jun 2010 19:38:20 +0000 Received: by pvg3 with SMTP id 3so1125526pvg.20 for ; Thu, 24 Jun 2010 12:38:18 -0700 (PDT) MIME-Version: 1.0 Received: by 10.115.80.7 with SMTP id h7mr10118370wal.159.1277408295867; Thu, 24 Jun 2010 12:38:15 -0700 (PDT) Received: by 10.151.154.11 with HTTP; Thu, 24 Jun 2010 12:38:15 -0700 (PDT) Date: Thu, 24 Jun 2010 21:38:15 +0200 Message-ID: Subject: [patch][4.5] Backport Mach-O LTO support From: Steven Bosscher To: GCC Patches , Richard Guenther , Jakub Jelinek Mailing-List: contact gcc-patches-help@gcc.gnu.org; run by ezmlm Precedence: bulk List-Id: List-Unsubscribe: List-Archive: List-Post: List-Help: Sender: gcc-patches-owner@gcc.gnu.org Delivered-To: mailing list gcc-patches@gcc.gnu.org Hi, Pretty straight-forward port of the patches from trunk to gcc 4.5.1. There are new test suite failures if configured with --enable-lto but those are the same as on trunk (where some test cases have been patched to xfail ELF-isms in the lto test suite). Bootstrapped & tested on x86_64-unknown-linux-gnu. OK for GCC 4.5? Ciao! Steven Backport of r159173 and r159527 to gcc 4.5.1. ChangeLog: * configure.ac (--enable-lto): Add x86_64-apple-darwin* as a platform that supports LTO. * configure: Regenerate. gcc/ChangeLog: * config.gcc (i[34567]86-*-darwin*, x86_64-*-darwin*): Add lto-macho as lto_binary_reader. * target.h (struct gcc_target): New hooks lto_start and lto_end. * target-def.h (TARGET_ASM_LTO_START, TARGET_ASM_LTO_END): Define. * cgraphunit.c (ipa_passes): Wrap LTO assembler output generation in lto_start and lto_end calls. (is_elf_or_coff): Rename to maybe_lto_object_file. Add Mach-O magic numbers. (scan_prog_file): Update is_elf_or_coff call. * doc/tm.text (TARGET_ASM_LTO_START, TARGET_ASM_LTO_END): Document. * collect2.c (main): Fix enum comparison. * config/darwin-protos.h (darwin_asm_lto_start, darwin_asm_lto_end): Add prototypes. * darwin9.h (LINK_COMMAND_SPEC): Pass -flto and -fwhopr to the linker. * darwin.h (LINK_COMMAND_SPEC): Likewise. Define TARGET_ASM_LTO_START and TARGET_ASM_LTO_END. * darwin.c: Include obstack.h and lto-streamer.h. (lto_section_names_offset, lto_section_names_obstack, lto_asm_out_file, lto_asm_out_name, saved_asm_out_file): New static global variables. (LTO_SEGMENT_NAME, LTO_NAMES_SECTION): New defines. (darwin_asm_lto_start): New function. Redirect output to asm_out_file to a temporary file. (darwin_asm_lto_end): New function. Restore asm_out_file. (darwin_asm_named_section): For LTO sections, replace the name with the offset of the section name in a string table, and build this table. (darwin_file_start): Initialize global vars for LTO support. (darwin_file_end): If output to asm_out_file was redirected, append it to the proper asm_out_file here. Add the section names section. lto/ChangeLog: * lto.h (struct lto_file_struct): Document offset member. * lto-endian.h: New file. * lto-macho.h: New file. * lto-macho.c: New file. * Make-lang.in: Add rule for lto-macho.o. Index: configure.ac =================================================================== --- configure.ac (revision 161328) +++ configure.ac (working copy) @@ -1784,6 +1784,7 @@ fi],[if test x"$default_enable_lto" = x"yes" ; the # -flto it won't be needed until after installation anyway. case $target in *-cygwin*|*-mingw*) ;; + x86_64-apple-darwin*) ;; *) if test x"$enable_lto" = x"yes"; then AC_MSG_ERROR([LTO support is not enabled for this target.]) fi Index: configure =================================================================== --- configure (revision 161328) +++ configure (working copy) @@ -6667,6 +6667,7 @@ else # -flto it won't be needed until after installation anyway. case $target in *-cygwin*|*-mingw*) ;; + x86_64-apple-darwin*) ;; *) if test x"$enable_lto" = x"yes"; then as_fn_error "LTO support is not enabled for this target." "$LINENO" 5 fi Index: gcc/config.gcc =================================================================== --- gcc/config.gcc (revision 161328) +++ gcc/config.gcc (working copy) @@ -1094,11 +1094,13 @@ i[34567]86-*-darwin*) # support. with_cpu=${with_cpu:-generic} tmake_file="${tmake_file} t-slibgcc-darwin i386/t-crtpc i386/t-crtfm" + lto_binary_reader=lto-macho ;; x86_64-*-darwin*) with_cpu=${with_cpu:-generic} tmake_file="${tmake_file} ${cpu_type}/t-darwin64 t-slibgcc-darwin i386/t-crtpc i386/t-crtfm" tm_file="${tm_file} ${cpu_type}/darwin64.h" + lto_binary_reader=lto-macho ;; i[34567]86-*-elf*) tm_file="${tm_file} i386/unix.h i386/att.h dbxelf.h elfos.h newlib-stdint.h i386/i386elf.h" Index: gcc/target.h =================================================================== --- gcc/target.h (revision 161328) +++ gcc/target.h (working copy) @@ -235,6 +235,14 @@ struct gcc_target translation unit. */ void (*file_end) (void); + /* Output any boilerplate text needed at the beginning of an + LTO output stream. */ + void (*lto_start) (void); + + /* Output any boilerplate text needed at the end of an + LTO output stream. */ + void (*lto_end) (void); + /* Output any boilerplace text needed at the end of a translation unit before debug and unwind info is emitted. */ void (*code_end) (void); Index: gcc/target-def.h =================================================================== --- gcc/target-def.h (revision 161328) +++ gcc/target-def.h (working copy) @@ -204,6 +204,14 @@ #define TARGET_ASM_FILE_END hook_void_void #endif +#ifndef TARGET_ASM_LTO_START +#define TARGET_ASM_LTO_START hook_void_void +#endif + +#ifndef TARGET_ASM_LTO_END +#define TARGET_ASM_LTO_END hook_void_void +#endif + #ifndef TARGET_ASM_CODE_END #define TARGET_ASM_CODE_END hook_void_void #endif @@ -296,6 +304,8 @@ TARGET_ASM_CAN_OUTPUT_MI_THUNK, \ TARGET_ASM_FILE_START, \ TARGET_ASM_FILE_END, \ + TARGET_ASM_LTO_START, \ + TARGET_ASM_LTO_END, \ TARGET_ASM_CODE_END, \ TARGET_ASM_EXTERNAL_LIBCALL, \ TARGET_ASM_MARK_DECL_PRESERVED, \ Index: gcc/cgraphunit.c =================================================================== --- gcc/cgraphunit.c (revision 161328) +++ gcc/cgraphunit.c (working copy) @@ -1813,11 +1813,19 @@ ipa_passes (void) execute_ipa_summary_passes ((struct ipa_opt_pass_d *) all_regular_ipa_passes); } + + /* Some targets need to handle LTO assembler output specially. */ + if (flag_generate_lto) + targetm.asm_out.lto_start (); + execute_ipa_summary_passes ((struct ipa_opt_pass_d *) all_lto_gen_passes); if (!in_lto_p) ipa_write_summaries (); + if (flag_generate_lto) + targetm.asm_out.lto_end (); + if (!flag_ltrans) execute_ipa_pass_list (all_regular_ipa_passes); invoke_plugin_callbacks (PLUGIN_ALL_IPA_PASSES_END, NULL); Index: gcc/doc/tm.texi =================================================================== --- gcc/doc/tm.texi (revision 161328) +++ gcc/doc/tm.texi (working copy) @@ -7149,6 +7149,18 @@ need to do other things in that hook, have your ho this function. @end deftypefun +@deftypefn {Target Hook} void TARGET_ASM_LTO_START (void) +Output to @code{asm_out_file} any text which the assembler expects +to find at the start of an LTO section. The default is to output +nothing. +@end deftypefn + +@deftypefn {Target Hook} void TARGET_ASM_LTO_END (void) +Output to @code{asm_out_file} any text which the assembler expects +to find at the end of an LTO section. The default is to output +nothing. +@end deftypefn + @deftypefn {Target Hook} void TARGET_ASM_CODE_END (void) Output to @code{asm_out_file} any text which is needed before emitting unwind info and debug info at the end of a file. Some targets emit Index: gcc/collect2.c =================================================================== --- gcc/collect2.c (revision 161328) +++ gcc/collect2.c (working copy) @@ -1802,7 +1802,7 @@ main (int argc, char **argv) if (export_file != 0 && export_file[0]) maybe_unlink (export_file); #endif - if (lto_mode) + if (lto_mode != LTO_MODE_NONE) maybe_run_lto_and_relink (ld1_argv, object_lst, object, false); maybe_unlink (c_file); @@ -2548,25 +2548,39 @@ write_aix_file (FILE *stream, struct id *list) #ifdef OBJECT_FORMAT_NONE -/* Check to make sure the file is an ELF file. LTO objects must - be in ELF format. */ +/* Check to make sure the file is an LTO object file. */ static bool -is_elf_or_coff (const char *prog_name) +maybe_lto_object_file (const char *prog_name) { FILE *f; - char buf[4]; - static char magic[4] = { 0x7f, 'E', 'L', 'F' }; - static char coffmag[2] = { 0x4c, 0x01 }; + unsigned char buf[4]; + int i; + static unsigned char elfmagic[4] = { 0x7f, 'E', 'L', 'F' }; + static unsigned char coffmagic[2] = { 0x4c, 0x01 }; + static unsigned char machomagic[4][4] = { + { 0xcf, 0xfa, 0xed, 0xfe }, + { 0xce, 0xfa, 0xed, 0xfe }, + { 0xfe, 0xed, 0xfa, 0xcf }, + { 0xfe, 0xed, 0xfa, 0xce } + }; + f = fopen (prog_name, "rb"); if (f == NULL) return false; if (fread (buf, sizeof (buf), 1, f) != 1) buf[0] = 0; fclose (f); - return memcmp (buf, magic, sizeof (magic)) == 0 - || memcmp (buf, coffmag, sizeof (coffmag)) == 0; + + if (memcmp (buf, elfmagic, sizeof (elfmagic)) == 0 + || memcmp (buf, coffmagic, sizeof (coffmagic)) == 0) + return true; + for (i = 0; i < 4; i++) + if (memcmp (buf, machomagic[i], sizeof (machomagic[i])) == 0) + return true; + + return false; } /* Generic version to scan the name list of the loaded program for @@ -2596,7 +2610,7 @@ scan_prog_file (const char *prog_name, scanpass wh /* LTO objects must be in a known format. This check prevents us from accepting an archive containing LTO objects, which gcc cannnot currently handle. */ - if (which_pass == PASS_LTOINFO && !is_elf_or_coff (prog_name)) + if (which_pass == PASS_LTOINFO && !maybe_lto_object_file (prog_name)) return; /* If we do not have an `nm', complain. */ Index: gcc/config/darwin-protos.h =================================================================== --- gcc/config/darwin-protos.h (revision 161328) +++ gcc/config/darwin-protos.h (working copy) @@ -71,6 +71,9 @@ extern void darwin_pragma_ms_struct (struct cpp_re extern void darwin_file_start (void); extern void darwin_file_end (void); +extern void darwin_asm_lto_start (void); +extern void darwin_asm_lto_end (void); + extern void darwin_mark_decl_preserved (const char *); extern tree darwin_handle_kext_attribute (tree *, tree, tree, int, bool *); Index: gcc/config/darwin9.h =================================================================== --- gcc/config/darwin9.h (revision 161328) +++ gcc/config/darwin9.h (working copy) @@ -32,6 +32,7 @@ along with GCC; see the file COPYING3. If not see %{o*}%{!o:-o a.out} \ %{!A:%{!nostdlib:%{!nostartfiles:%S}}} \ %{L*} %(link_libgcc) %o %{fprofile-arcs|fprofile-generate*|coverage:-lgcov} \ + %{flto} %{fwhopr} \ %{fopenmp|ftree-parallelize-loops=*: \ %{static|static-libgcc|static-libstdc++|static-libgfortran: libgomp.a%s; : -lgomp } } \ %{!nostdlib:%{!nodefaultlibs: %(link_ssp) %G %L }} \ Index: gcc/config/darwin.h =================================================================== --- gcc/config/darwin.h (revision 161328) +++ gcc/config/darwin.h (working copy) @@ -273,6 +273,7 @@ extern GTY(()) int darwin_ms_struct; %{o*}%{!o:-o a.out} \ %{!A:%{!nostdlib:%{!nostartfiles:%S}}} \ %{L*} %(link_libgcc) %o %{fprofile-arcs|fprofile-generate*|coverage:-lgcov} \ + %{flto} %{fwhopr} \ %{fopenmp|ftree-parallelize-loops=*: \ %{static|static-libgcc|static-libstdc++|static-libgfortran: libgomp.a%s; : -lgomp } } \ %{!nostdlib:%{!nodefaultlibs: %(link_ssp) %G %L }} \ @@ -581,6 +582,16 @@ extern GTY(()) int darwin_ms_struct; #undef TARGET_ASM_FILE_END #define TARGET_ASM_FILE_END darwin_file_end +/* Because Mach-O relocations have a counter from 1 to 255 for the + section number they apply to, it is necessary to output all + normal sections before the LTO sections, to make sure that the + sections that may have relocations always have a section number + smaller than 255. */ +#undef TARGET_ASM_LTO_START +#define TARGET_ASM_LTO_START darwin_asm_lto_start +#undef TARGET_ASM_LTO_END +#define TARGET_ASM_LTO_END darwin_asm_lto_end + #define ASM_OUTPUT_SKIP(FILE,SIZE) \ fprintf (FILE, "\t.space "HOST_WIDE_INT_PRINT_UNSIGNED"\n", SIZE) Index: gcc/config/darwin.c =================================================================== --- gcc/config/darwin.c (revision 161328) +++ gcc/config/darwin.c (working copy) @@ -46,6 +46,8 @@ along with GCC; see the file COPYING3. If not see #include "hashtab.h" #include "df.h" #include "debug.h" +#include "obstack.h" +#include "lto-streamer.h" /* Darwin supports a feature called fix-and-continue, which is used for rapid turn around debugging. When code is compiled with the @@ -1391,12 +1393,89 @@ darwin_label_is_anonymous_local_objc_name (const c return (!strncmp ((const char *)p, "_OBJC_", 6)); } +/* LTO support for Mach-O. */ + +/* Section names for LTO sections. */ +static unsigned int lto_section_names_offset = 0; + +/* This is the obstack which we use to allocate the many strings. */ +static struct obstack lto_section_names_obstack; + +/* Segment name for LTO sections. */ +#define LTO_SEGMENT_NAME "__GNU_LTO" + +/* Section name for LTO section names section. */ +#define LTO_NAMES_SECTION "__section_names" + +/* File to temporarily store LTO data. This is appended to asm_out_file + in darwin_end_file. */ +static FILE *lto_asm_out_file, *saved_asm_out_file; +static char *lto_asm_out_name; + +/* Prepare asm_out_file for LTO output. For darwin, this means hiding + asm_out_file and switching to an alternative output file. */ void +darwin_asm_lto_start (void) +{ + gcc_assert (! saved_asm_out_file); + saved_asm_out_file = asm_out_file; + if (! lto_asm_out_name) + lto_asm_out_name = make_temp_file (".lto.s"); + lto_asm_out_file = fopen (lto_asm_out_name, "a"); + if (lto_asm_out_file == NULL) + fatal_error ("failed to open temporary file %s for LTO output", + lto_asm_out_name); + asm_out_file = lto_asm_out_file; +} + +/* Restore asm_out_file. */ +void +darwin_asm_lto_end (void) +{ + gcc_assert (saved_asm_out_file); + fclose (lto_asm_out_file); + asm_out_file = saved_asm_out_file; + saved_asm_out_file = NULL; +} + +void darwin_asm_named_section (const char *name, - unsigned int flags ATTRIBUTE_UNUSED, + unsigned int flags, tree decl ATTRIBUTE_UNUSED) { - fprintf (asm_out_file, "\t.section %s\n", name); + /* LTO sections go in a special segment __GNU_LTO. We want to replace the + section name with something we can use to represent arbitrary-length + names (section names in Mach-O are at most 16 characters long). */ + if (strncmp (name, LTO_SECTION_NAME_PREFIX, + strlen (LTO_SECTION_NAME_PREFIX)) == 0) + { + /* We expect certain flags to be set... */ + gcc_assert ((flags & (SECTION_DEBUG | SECTION_NAMED)) + == (SECTION_DEBUG | SECTION_NAMED)); + + /* Add the section name to the things to output when we end the + current assembler output file. + This is all not very efficient, but that doesn't matter -- this + shouldn't be a hot path in the compiler... */ + obstack_1grow (<o_section_names_obstack, '\t'); + obstack_grow (<o_section_names_obstack, ".ascii ", 7); + obstack_1grow (<o_section_names_obstack, '"'); + obstack_grow (<o_section_names_obstack, name, strlen (name)); + obstack_grow (<o_section_names_obstack, "\\0\"\n", 4); + + /* Output the dummy section name. */ + fprintf (asm_out_file, "\t# %s\n", name); + fprintf (asm_out_file, "\t.section %s,__%08X,regular,debug\n", + LTO_SEGMENT_NAME, lto_section_names_offset); + + /* Update the offset for the next section name. Make sure we stay + within reasonable length. */ + lto_section_names_offset += strlen (name) + 1; + gcc_assert (lto_section_names_offset > 0 + && lto_section_names_offset < ((unsigned) 1 << 31)); + } + else + fprintf (asm_out_file, "\t.section %s\n", name); } void @@ -1589,7 +1668,8 @@ darwin_asm_output_dwarf_delta (FILE *file, int siz fprintf (file, "\n\t%s L$set$%d", directive, darwin_dwarf_label_counter++); } -/* Output labels for the start of the DWARF sections if necessary. */ +/* Output labels for the start of the DWARF sections if necessary. + Initialize the stuff we need for LTO long section names support. */ void darwin_file_start (void) { @@ -1624,6 +1704,11 @@ darwin_file_start (void) fprintf (asm_out_file, "Lsection%.*s:\n", namelen, debugnames[i] + 8); } } + + /* We fill this obstack with the complete section text for the lto section + names to write in darwin_file_end. */ + obstack_init (<o_section_names_obstack); + lto_section_names_offset = 0; } /* Output an offset in a DWARF section on Darwin. On Darwin, DWARF section @@ -1650,6 +1735,8 @@ darwin_asm_output_dwarf_offset (FILE *file, int si void darwin_file_end (void) { + const char *lto_section_names; + machopic_finish (asm_out_file); if (strcmp (lang_hooks.name, "GNU C++") == 0) { @@ -1657,6 +1744,52 @@ darwin_file_end (void) switch_to_section (darwin_sections[destructor_section]); ASM_OUTPUT_ALIGN (asm_out_file, 1); } + + /* If there was LTO assembler output, append it to asm_out_file. */ + if (lto_asm_out_name) + { + int n; + char *buf, *lto_asm_txt; + + /* Shouldn't be here if we failed to switch back. */ + gcc_assert (! saved_asm_out_file); + + lto_asm_out_file = fopen (lto_asm_out_name, "r"); + if (lto_asm_out_file == NULL) + fatal_error ("failed to open temporary file %s with LTO output", + lto_asm_out_name); + fseek (lto_asm_out_file, 0, SEEK_END); + n = ftell (lto_asm_out_file); + if (n > 0) + { + fseek (lto_asm_out_file, 0, SEEK_SET); + lto_asm_txt = buf = (char *) xmalloc (n + 1); + while (fgets (lto_asm_txt, n, lto_asm_out_file)) + fputs (lto_asm_txt, asm_out_file); + } + + /* Remove the temporary file. */ + fclose (lto_asm_out_file); + unlink_if_ordinary (lto_asm_out_name); + free (lto_asm_out_name); + } + + /* Finish the LTO section names obstack. Don't output anything if + there are no recorded section names. */ + obstack_1grow (<o_section_names_obstack, '\0'); + lto_section_names = XOBFINISH (<o_section_names_obstack, const char *); + if (strlen (lto_section_names) > 0) + { + fprintf (asm_out_file, + "\t.section %s,%s,regular,debug\n", + LTO_SEGMENT_NAME, LTO_NAMES_SECTION); + fprintf (asm_out_file, + "\t# Section names in %s are offsets into this table\n", + LTO_SEGMENT_NAME); + fprintf (asm_out_file, "%s\n", lto_section_names); + } + obstack_free (<o_section_names_obstack, NULL); + fprintf (asm_out_file, "\t.subsections_via_symbols\n"); } Index: gcc/lto/lto.h =================================================================== --- gcc/lto/lto.h (revision 161328) +++ gcc/lto/lto.h (working copy) @@ -1,5 +1,5 @@ /* LTO declarations. - Copyright 2009 Free Software Foundation, Inc. + Copyright 2009, 2010 Free Software Foundation, Inc. Contributed by CodeSourcery, Inc. This file is part of GCC. @@ -28,6 +28,7 @@ typedef struct lto_file_struct { /* The name of the file. */ const char *filename; + /* The offset for the object inside an ar archive file (or zero). */ off_t offset; } lto_file; Index: gcc/lto/lto-endian.h =================================================================== --- gcc/lto/lto-endian.h (revision 0) +++ gcc/lto/lto-endian.h (revision 0) @@ -0,0 +1,205 @@ +/* Very simple endian-ness layer for LTO object file handling + Copyright 2010 Free Software Foundation, Inc. + +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. See 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. If not see +. */ + +/* This header file provides a simple way to handle object files in + another endian-ness than the host machine. This is necesarry to + enable cross-compilation with LTO enabled. Targets that use the + ELF binary object format do not need this (libelf already handles + endian-ness) but for COFF and Mach-O the functions in this header + are used in the minimal binary object reader/writer. + + For all functions in this header, the user is responsible for + making sure that the memory accesses are valid. */ + +#ifndef GCC_LTO_ENDIAN_H +#define GCC_LTO_ENDIAN_H + +#include +#include + +static inline uint16_t +get_uint16_le (const unsigned char *ptr) +{ + return ptr[0] | (ptr[1] << 8); +} + +static inline uint32_t +get_uint32_le (const unsigned char *ptr) +{ + return ptr[0] | (ptr[1] << 8) | (ptr[2] << 16) | (ptr[3] << 24); +} + +static inline uint64_t +get_uint64_le (const unsigned char *ptr_) +{ +#define ptr (uint64_t) ptr_ + return ptr[0] | (ptr[1] << 8) | (ptr[2] << 16) | (ptr[3] << 24) + | (ptr[4] << 32) | (ptr[5] << 40) | (ptr[6] << 48) | (ptr[7] << 56); +#undef ptr +} + +static inline uint16_t +get_uint16_be (const unsigned char *ptr) +{ + return ptr[1] | (ptr[2] << 8); +} + +static inline uint32_t +get_uint32_be (const unsigned char *ptr) +{ + return ptr[3] | (ptr[2] << 8) | (ptr[1] << 16) | (ptr[0] << 24); +} + +static inline uint64_t +get_uint64_be (const unsigned char *ptr_) +{ +#define ptr (uint64_t) ptr_ + return ptr[7] | (ptr[6] << 8) | (ptr[5] << 16) | (ptr[4] << 24) + | (ptr[3] << 32) | (ptr[2] << 40) | (ptr[1] << 48) | (ptr[0] << 56); +#undef ptr +} + +static inline void +put_uint16_le (unsigned char *ptr, uint16_t data) +{ + ptr[0] = data & 0xff; + ptr[1] = (data >> 8) & 0xff; +} + +static inline void +put_uint32_le (unsigned char *ptr, uint32_t data) +{ + ptr[0] = data & 0xff; + ptr[1] = (data >> 8) & 0xff; + ptr[2] = (data >> 16) & 0xff; + ptr[3] = (data >> 24) & 0xff; +} + +static inline void +put_uint64_le (unsigned char *ptr, uint64_t data) +{ + ptr[0] = data & 0xff; + ptr[1] = (data >> 8) & 0xff; + ptr[2] = (data >> 16) & 0xff; + ptr[3] = (data >> 24) & 0xff; + ptr[4] = (data >> 32) & 0xff; + ptr[5] = (data >> 40) & 0xff; + ptr[6] = (data >> 48) & 0xff; + ptr[7] = (data >> 56) & 0xff; +} + +static inline void +put_uint16_be (unsigned char *ptr, uint16_t data) +{ + ptr[1] = data & 0xff; + ptr[0] = (data >> 8) & 0xff; +} + +static inline void +put_uint32_be (unsigned char *ptr, uint32_t data) +{ + ptr[3] = data & 0xff; + ptr[2] = (data >> 8) & 0xff; + ptr[1] = (data >> 16) & 0xff; + ptr[0] = (data >> 24) & 0xff; +} + +static inline void +put_uint64_be (unsigned char *ptr, uint64_t data) +{ + ptr[7] = data & 0xff; + ptr[6] = (data >> 8) & 0xff; + ptr[5] = (data >> 16) & 0xff; + ptr[4] = (data >> 24) & 0xff; + ptr[3] = (data >> 32) & 0xff; + ptr[2] = (data >> 40) & 0xff; + ptr[1] = (data >> 48) & 0xff; + ptr[0] = (data >> 56) & 0xff; +} + +static inline void +get_string (unsigned char *ptr, char *dest, size_t len) +{ + memcpy (dest, ptr, len); +} + +static inline void +put_string (unsigned char *ptr, char *src, size_t len) +{ + memcpy (ptr, src, len); +} + +/* Use the target macro BYTES_BIG_ENDIAN to choose. */ + +static inline uint16_t +get_uint16 (const unsigned char *ptr) +{ + if (BYTES_BIG_ENDIAN) + return get_uint16_be (ptr); + else + return get_uint16_le (ptr); +} + +static inline uint32_t +get_uint32 (const unsigned char *ptr) +{ + if (BYTES_BIG_ENDIAN) + return get_uint32_be (ptr); + else + return get_uint32_le (ptr); +} + +static inline uint64_t +get_uint64 (const unsigned char *ptr) +{ + if (BYTES_BIG_ENDIAN) + return get_uint64_be (ptr); + else + return get_uint64_le (ptr); +} + +static inline void +put_uint16 (unsigned char *ptr, uint16_t data) +{ + if (BYTES_BIG_ENDIAN) + put_uint16_be (ptr, data); + else + put_uint16_le (ptr, data); +} + +static inline void +put_uint32 (unsigned char *ptr, uint32_t data) +{ + if (BYTES_BIG_ENDIAN) + put_uint32_be (ptr, data); + else + put_uint32_le (ptr, data); +} + +static inline void +put_uint64 (unsigned char *ptr, uint64_t data) +{ + if (BYTES_BIG_ENDIAN) + put_uint64_be (ptr, data); + else + put_uint64_le (ptr, data); +} + +#endif /* GCC_LTO_ENDIAN_H */ + Index: gcc/lto/lto-macho.h =================================================================== --- gcc/lto/lto-macho.h (revision 0) +++ gcc/lto/lto-macho.h (revision 0) @@ -0,0 +1,251 @@ +/* LTO routines for Mach-O object files. + Copyright 2010 Free Software Foundation, Inc. + Contributed by Steven Bosscher. + +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. See 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. If not see +. */ + +#ifndef LTO_MACH_O_H +#define LTO_MACH_O_H + +/* On-disk file structures. */ + +/* Mach-O header (32 bits version). */ +struct mach_o_header_32 +{ + unsigned char magic[4]; /* Magic number. */ + unsigned char cputype[4]; /* CPU that this object is for. */ + unsigned char cpusubtype[4]; /* CPU subtype. */ + unsigned char filetype[4]; /* Type of file. */ + unsigned char ncmds[4]; /* Number of load commands. */ + unsigned char sizeofcmds[4]; /* Total size of load commands. */ + unsigned char flags[4]; /* Flags for special featues. */ +}; +typedef struct mach_o_header_32 mach_o_header_32; + +/* Mach-O header (64 bits version). */ +struct mach_o_header_64 +{ + unsigned char magic[4]; /* Magic number. */ + unsigned char cputype[4]; /* CPU that this object is for. */ + unsigned char cpusubtype[4]; /* CPU subtype. */ + unsigned char filetype[4]; /* Type of file. */ + unsigned char ncmds[4]; /* Number of load commands. */ + unsigned char sizeofcmds[4]; /* Total size of load commands. */ + unsigned char flags[4]; /* Flags for special featues. */ + unsigned char reserved[4]; /* Reserved. Duh. */ +}; +typedef struct mach_o_header_64 mach_o_header_64; + +/* Magic number. */ +#define MACH_O_MH_MAGIC 0xfeedface +#define MACH_O_MH_CIGAM 0xcefaedfe +#define MACH_O_MH_MAGIC_64 0xfeedfacf +#define MACH_O_MH_CIGAM_64 0xcffaedfe + +/* Supported CPU types. */ +#define MACH_O_CPU_TYPE_I386 7 +#define MACH_O_CPU_TYPE_X86_64 7 + 0x1000000 +#define MACH_O_CPU_TYPE_POWERPC 18 +#define MACH_O_CPU_TYPE_POWERPC_64 18 + 0x1000000 + +/* Supported file types. */ +#define MACH_O_MH_OBJECT 0x01 + +/* Mach-O load command data structure. */ +struct mach_o_load_command +{ + unsigned char cmd[4]; /* The type of load command. */ + unsigned char cmdsize[4]; /* Size in bytes of load command data structure. */ +}; +typedef struct mach_o_load_command mach_o_load_command; + +/* Supported load commands. We support only the segment load commands. */ +#define MACH_O_LC_SEGMENT 0x01 +#define MACH_O_LC_SEGMENT_64 0x19 + +/* LC_SEGMENT load command. */ +struct mach_o_segment_command_32 +{ + unsigned char cmd[4]; /* The type of load command (LC_SEGMENT). */ + unsigned char cmdsize[4]; /* Size in bytes of load command data structure. */ + unsigned char segname[16]; /* Name of this segment. */ + unsigned char vmaddr[4]; /* Virtual memory address of this segment. */ + unsigned char vmsize[4]; /* Size there, in bytes. */ + unsigned char fileoff[4]; /* Offset in bytes of the data to be mapped. */ + unsigned char filesize[4]; /* Size in bytes on disk. */ + unsigned char maxprot[4]; /* Maximum permitted vmem protection. */ + unsigned char initprot[4]; /* Initial vmem protection. */ + unsigned char nsects[4]; /* Number of sections in this segment. */ + unsigned char flags[4]; /* Flags that affect the loading. */ +}; +typedef struct mach_o_segment_command_32 mach_o_segment_command_32; + +/* LC_SEGMENT_64 load command. Only nsects matters for us, really. */ +struct mach_o_segment_command_64 +{ + unsigned char cmd[4]; /* The type of load command (LC_SEGMENT_64). */ + unsigned char cmdsize[4]; /* Size in bytes of load command data structure. */ + unsigned char segname[16]; /* Name of this segment. */ + unsigned char vmaddr[8]; /* Virtual memory address of this segment. */ + unsigned char vmsize[8]; /* Size there, in bytes. */ + unsigned char fileoff[8]; /* Offset in bytes of the data to be mapped. */ + unsigned char filesize[8]; /* Size in bytes on disk. */ + unsigned char maxprot[4]; /* Maximum permitted vmem protection. */ + unsigned char initprot[4]; /* Initial vmem protection. */ + unsigned char nsects[4]; /* Number of sections in this segment. */ + unsigned char flags[4]; /* Flags that affect the loading. */ +}; +typedef struct mach_o_segment_command_64 mach_o_segment_command_64; + +/* A Mach-O 32-bits section. */ +struct mach_o_section_32 +{ + unsigned char sectname[16]; /* Section name. */ + unsigned char segname[16]; /* Segment that the section belongs to. */ + unsigned char addr[4]; /* Address of this section in memory. */ + unsigned char size[4]; /* Size in bytes of this section. */ + unsigned char offset[4]; /* File offset of this section. */ + unsigned char align[4]; /* log2 of this section's alignment. */ + unsigned char reloff[4]; /* File offset of this section's relocs. */ + unsigned char nreloc[4]; /* Number of relocs for this section. */ + unsigned char flags[4]; /* Section flags/attributes. */ + unsigned char reserved1[4]; + unsigned char reserved2[4]; +}; +typedef struct mach_o_section_32 mach_o_section_32; + +/* A Mach-O 64-bits section. */ +struct mach_o_section_64 +{ + unsigned char sectname[16]; /* Section name. */ + unsigned char segname[16]; /* Segment that the section belongs to. */ + unsigned char addr[8]; /* Address of this section in memory. */ + unsigned char size[8]; /* Size in bytes of this section. */ + unsigned char offset[4]; /* File offset of this section. */ + unsigned char align[4]; /* log2 of this section's alignment. */ + unsigned char reloff[4]; /* File offset of this section's relocs. */ + unsigned char nreloc[4]; /* Number of relocs for this section. */ + unsigned char flags[4]; /* Section flags/attributes. */ + unsigned char reserved1[4]; + unsigned char reserved2[4]; + unsigned char reserved3[4]; +}; +typedef struct mach_o_section_64 mach_o_section_64; + +/* Flags for Mach-O sections. LTO sections are marked with S_ATTR_DEBUG + to instruct the linker to ignore the sections. */ +#define MACH_O_S_ATTR_DEBUG 0x02000000 + +/* In-memory file structures. */ + +/* Section data in output files is made of these. */ +struct lto_mach_o_data_d +{ + /* Pointer to data block. */ + void *d_buf; + + /* Size of data block. */ + ssize_t d_size; + + /* Next data block for this section. */ + struct lto_mach_o_data_d *next; +}; +typedef struct lto_mach_o_data_d *lto_mach_o_data; + +/* This struct tracks the data for a section. */ +struct lto_mach_o_section_d +{ + /* Singly-linked list of section's data blocks. */ + lto_mach_o_data data_chain; + + /* Offset in string table of the section name. */ + size_t strtab_offs; + + /* Section name. */ + const char *name; + + /* Number of trailing padding bytes needed. */ + ssize_t pad_needed; + + /* Raw section header data. */ + size_t section_size; + union { + struct { + char sectname[16]; + char segname[16]; + } section; + mach_o_section_32 section_32; + mach_o_section_64 section_64; + } u; + + /* Next section for this file. */ + struct lto_mach_o_section_d *next; +}; +typedef struct lto_mach_o_section_d *lto_mach_o_section; +DEF_VEC_P (lto_mach_o_section); +DEF_VEC_ALLOC_P (lto_mach_o_section, heap); + +/* A Mach-O file. */ +struct lto_mach_o_file_d +{ + /* The base information. */ + lto_file base; + + /* Common file members: */ + + /* The system file descriptor for the file. */ + int fd; + + /* The file's overall header. */ + union { + /* We make use here of the fact that section_32 and section_64 + have the same layout (except for section_64.reserved3). We + read the struct of proper size, but only address the first + member of this union. */ + mach_o_header_64 header; + mach_o_header_32 header_32; + mach_o_header_64 header_64; + } u; + + /* All sections in a varray. */ + VEC(lto_mach_o_section, heap) *section_vec; + + /* Readable file members: */ + + /* File total size. */ + off_t file_size; + + /* True if this file is open for writing. */ + bool writable; + + /* Section containing the __section_names section. */ + lto_mach_o_section section_names_section; + + /* Writable file members: */ + + /* The currently active section. */ + lto_mach_o_section scn; + + /* Linked list of data which must be freed *after* the file has been + closed. This is an annoying limitation of libelf. Which has been + faithfully reproduced here. */ + struct lto_char_ptr_base *data; +}; +typedef struct lto_mach_o_file_d lto_mach_o_file; + +#endif /* LTO_MACH_O_H */ + Index: gcc/lto/lto-macho.c =================================================================== --- gcc/lto/lto-macho.c (revision 0) +++ gcc/lto/lto-macho.c (revision 0) @@ -0,0 +1,969 @@ +/* LTO routines for Mach-O object files. + Copyright 2010 Free Software Foundation, Inc. + Contributed by Steven Bosscher. + +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. See 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. If not see +. */ + +#include "config.h" +#include "system.h" +#include "coretypes.h" +#include "toplev.h" +#include "lto.h" +#include "tm.h" +#include "libiberty.h" +#include "lto-streamer.h" +#include "lto/lto-endian.h" +#include "lto/lto-macho.h" + +/* Rather than implementing a libmacho to match libelf, or attempting to + integrate libbfd into GCC, this file is a self-contained (and very + minimal) Mach-O format object file reader/writer. The generated files + will contain a Mach-O header, a number of Mach-O load commands an + section headers, the section data itself, and a trailing string table + for section names. */ + +/* This needs to be kept in sync with darwin.c. Better yet, lto-macho.c + and lto-macho.h should be moved to config/, and likewise for lto-coff.* + and lto-elf.*. */ + +/* Segment name for LTO sections. */ +#define LTO_SEGMENT_NAME "__GNU_LTO" + +/* Section name for LTO section names section. */ +#define LTO_NAMES_SECTION "__section_names" + +/* Handle opening elf files on hosts, such as Windows, that may use + text file handling that will break binary access. */ +#ifndef O_BINARY +# define O_BINARY 0 +#endif + +/* Cached object file header. We use a header_64 for this, since all + the fields we need are in there, in the same position as header_32. */ +mach_o_header_64 cached_mach_o_header; +uint32_t cached_mach_o_magic; + +/* The current output file. */ +static lto_file *current_out_file; + + +/* Is this a 32-bits or 64-bits Mach-O object file? */ +static int +mach_o_word_size (void) +{ + gcc_assert (cached_mach_o_magic != 0); + return (cached_mach_o_magic == MACH_O_MH_MAGIC_64 + || cached_mach_o_magic == MACH_O_MH_CIGAM_64) ? 64 : 32; +} + +/* Sets the current output file to FILE. Returns the old output file or + NULL. */ + +lto_file * +lto_set_current_out_file (lto_file *file) +{ + lto_file *old_file = current_out_file; + current_out_file = file; + return old_file; +} + + +/* Returns the current output file. */ + +lto_file * +lto_get_current_out_file (void) +{ + return current_out_file; +} + +/* Mach-O section structure constructor. */ + +static lto_mach_o_section +mach_o_new_section (lto_mach_o_file *mach_o_file, const char *name) +{ + lto_mach_o_section ptr; + + /* FIXME We could allocate these things on an obstack. */ + ptr = XCNEW (struct lto_mach_o_section_d); + if (name) + { + if (strncmp (name, LTO_SECTION_NAME_PREFIX, + strlen(LTO_SECTION_NAME_PREFIX)) != 0) + sorry ("not implemented: Mach-O writer for non-LTO sections"); + ptr->name = xstrdup (name); + } + + VEC_safe_push (lto_mach_o_section, heap, mach_o_file->section_vec, ptr); + + return ptr; +} + +/* Mach-O section data block structure constructor. */ + +static lto_mach_o_data +mach_o_new_data (lto_mach_o_section sec) +{ + lto_mach_o_data ptr, *chain_ptr_ptr; + + /* FIXME We could allocate these things on an obstack. */ + ptr = XCNEW (struct lto_mach_o_data_d); + + chain_ptr_ptr = &sec->data_chain; + while (*chain_ptr_ptr) + chain_ptr_ptr = &(*chain_ptr_ptr)->next; + *chain_ptr_ptr = ptr; + + return ptr; +} + +/* Initialize FILE, an LTO file object for FILENAME. Offset is the + offset into FILE where the object is located (e.g. in an archive). */ + +static void +lto_file_init (lto_file *file, const char *filename, off_t offset) +{ + file->filename = filename; + file->offset = offset; +} + +/* Returns a hash code for P. */ + +static hashval_t +hash_name (const void *p) +{ + const struct lto_section_slot *s = (const struct lto_section_slot *) p; + return (hashval_t) htab_hash_string (s->name); +} + +/* Returns nonzero if P1 and P2 are equal. */ + +static int +eq_name (const void *p1, const void *p2) +{ + const struct lto_section_slot *s1 = + (const struct lto_section_slot *) p1; + const struct lto_section_slot *s2 = + (const struct lto_section_slot *) p2; + + return strcmp (s1->name, s2->name) == 0; +} + +/* Build a hash table whose key is the section names and whose data is + the start and size of each section in the .o file. */ + +htab_t +lto_obj_build_section_table (lto_file *lto_file) +{ + lto_mach_o_file *mach_o_file = (lto_mach_o_file *)lto_file; + lto_mach_o_section sec; + htab_t section_hash_table; + off_t strtab_offs; + ssize_t strtab_size; + char *strtab = NULL; + int i; + + section_hash_table = htab_create (37, hash_name, eq_name, free); + + /* Seek the string table. */ + /* FIXME The segment name should be in darwin.h, but can we include it + here in this file? */ + for (i = 0; + VEC_iterate (lto_mach_o_section, mach_o_file->section_vec, i, sec); + i++) + { + if (strncmp (sec->u.section.segname, "__GNU_LTO", 16) != 0) + continue; + if (strncmp (sec->u.section.sectname, "__section_names", 16) == 0) + break; + } + if (! sec) + { + error ("invalid Mach-O LTO object file: no __section_names section found"); + goto done; + } + mach_o_file->section_names_section = sec; + + if (mach_o_word_size () == 64) + { + strtab_offs = (off_t) get_uint32 (&sec->u.section_64.offset[0]); + strtab_size = (size_t) get_uint64 (&sec->u.section_64.size[0]); + } + else + { + strtab_offs = (off_t) get_uint32 (&sec->u.section_32.offset[0]); + strtab_size = (size_t) get_uint32 (&sec->u.section_32.size[0]); + } + + /* Seek to start of string table. */ + if (strtab_offs != lseek (mach_o_file->fd, + mach_o_file->base.offset + strtab_offs, + SEEK_SET)) + { + error ("altered or invalid Mach-O object file"); + goto done; + } + + strtab = XNEWVEC (char, strtab_size); + if (read (mach_o_file->fd, strtab, strtab_size) != strtab_size) + { + error ("invalid Mach-O LTO object file __section_names section"); + goto done; + } + + /* Scan sections looking at names. */ + for (i = 0; + VEC_iterate (lto_mach_o_section, mach_o_file->section_vec, i, sec); + i++) + { + struct lto_section_slot s_slot; + void **slot; + char *new_name; + unsigned long stringoffset; + char name[17]; + + /* Ignore non-LTO sections. Also ignore the __section_names section + which does not need renaming. */ + if (strncmp (sec->u.section.segname, "__GNU_LTO", 16) != 0) + continue; + if (sec == mach_o_file->section_names_section) + continue; + + /* Try to extract the offset of the real name for this section from + __section_names. */ + memcpy (&name[0], sec->u.section.sectname, 16); + name[16] = '\0'; + if (name[0] != '_' || name[1] != '_' + || sscanf (&name[2], "%08lX", &stringoffset) != 1 + || strtab_size < (ssize_t) stringoffset) + { + error ("invalid Mach-O LTO section name string: %s", name); + continue; + } + + new_name = XNEWVEC (char, strlen (strtab + stringoffset) + 1); + strcpy (new_name, strtab + stringoffset); + s_slot.name = new_name; + slot = htab_find_slot (section_hash_table, &s_slot, INSERT); + if (*slot == NULL) + { + struct lto_section_slot *new_slot = XNEW (struct lto_section_slot); + + new_slot->name = new_name; + if (mach_o_word_size() == 64) + { + new_slot->start = + (intptr_t) get_uint32 (&sec->u.section_64.offset[0]); + new_slot->len = + (size_t) get_uint64 (&sec->u.section_64.size[0]); + } + else + { + new_slot->start = + (intptr_t) get_uint32 (&sec->u.section_32.offset[0]); + new_slot->len = + (size_t) get_uint32 (&sec->u.section_32.size[0]); + } + + *slot = new_slot; + } + else + { + error ("two or more sections for %s:", new_name); + goto done; + } + } + + done: + if (strtab) + free (strtab); + return section_hash_table; +} + + +/* Begin a new Mach-O section named NAME in the current output file. */ + +void +lto_obj_begin_section (const char *name) +{ + lto_mach_o_file *file; + + if (strncmp (name, LTO_SECTION_NAME_PREFIX, + strlen(LTO_SECTION_NAME_PREFIX)) != 0) + sorry ("not implemented: Mach-O writer for non-LTO sections"); + + /* Grab the current output file and do some basic assertion checking. */ + file = (lto_mach_o_file *) lto_get_current_out_file (), + gcc_assert (file && file->writable && !file->scn); + + /* Create a new section. */ + file->scn = mach_o_new_section (file, name); + if (!file->scn) + fatal_error ("could not create a new Mach-O section: %m"); +} + + +/* Append DATA of length LEN to the current output section. BASE is a pointer + to the output page containing DATA. It is freed once the output file has + been written. */ + +void +lto_obj_append_data (const void *data, size_t len, void *block) +{ + lto_mach_o_file *file; + lto_mach_o_data mach_o_data; + struct lto_char_ptr_base *base = (struct lto_char_ptr_base *) block; + + /* Grab the current output file and do some basic assertion checking. */ + file = (lto_mach_o_file *) lto_get_current_out_file (); + gcc_assert (file); + gcc_assert (file->scn); + + mach_o_data = mach_o_new_data (file->scn); + if (!mach_o_data) + fatal_error ("could not append data to Mach-O section: %m"); + + mach_o_data->d_buf = CONST_CAST (void *, data); + mach_o_data->d_size = len; + + /* Chain all data blocks (from all sections) on one singly-linked + list for freeing en masse after the file is closed. */ + base->ptr = (char *)file->data; + file->data = base; +} + + +/* End the current output section. This just does some assertion checking + and sets the current output file's scn member to NULL. */ + +void +lto_obj_end_section (void) +{ + lto_mach_o_file *file; + + /* Grab the current output file and validate some basic assertions. */ + file = (lto_mach_o_file *) lto_get_current_out_file (); + gcc_assert (file); + gcc_assert (file->scn); + + file->scn = NULL; +} + + +/* Read a Mach-O header from MACH_O_FILE and validate it. + The file descriptor in MACH_O_FILE points at the start of the file. + If cached_mach_o_header is uninitialized, caches the results. + On succes, returns true and moves file pointer to the start of the + load commands. On failure, returns false. */ + +static bool +validate_mach_o_header (lto_mach_o_file *mach_o_file) +{ + ssize_t i, n; + unsigned char magic[4]; + uint32_t cputype; + off_t startpos; + + /* Known header magics for validation, as an array. */ + static const unsigned int mach_o_known_formats[] = { + MACH_O_MH_MAGIC, + MACH_O_MH_CIGAM, + MACH_O_MH_MAGIC_64, + MACH_O_MH_CIGAM_64, + }; +#define MACH_O_NUM_KNOWN_FORMATS \ + ((ssize_t) ARRAY_SIZE (mach_o_known_formats)) + + startpos = lseek (mach_o_file->fd, 0, SEEK_CUR); + if (read (mach_o_file->fd, &magic, sizeof (magic)) != 4 + || lseek (mach_o_file->fd, -4, SEEK_CUR) != startpos) + { + error ("cannot read file %s", mach_o_file->base.filename); + return false; + } + + for (i = 0; i < MACH_O_NUM_KNOWN_FORMATS; ++i) + if (get_uint32 (&magic[0]) == mach_o_known_formats[i]) + break; + if (i == MACH_O_NUM_KNOWN_FORMATS) + goto not_for_target; + + /* Check the endian-ness. */ + if (BYTES_BIG_ENDIAN && magic[0] != 0xfe) + goto not_for_target; + + /* Set or check cached magic number. */ + if (cached_mach_o_magic == 0) + cached_mach_o_magic = get_uint32 (&magic[0]); + else if (cached_mach_o_magic != get_uint32 (&magic[0])) + goto not_for_target; + + n = mach_o_word_size () == 64 + ? sizeof (mach_o_header_64) : sizeof (mach_o_header_32); + if (read (mach_o_file->fd, &mach_o_file->u.header, n) != n) + goto not_for_target; + + /* Is this a supported CPU? */ + /* ??? Would be nice to validate the exact target architecture. */ + cputype = get_uint32 (&mach_o_file->u.header.cputype[0]); + if (cputype == MACH_O_CPU_TYPE_I386 + || cputype == MACH_O_CPU_TYPE_POWERPC) + { + if (mach_o_word_size () != 32) + goto not_for_target; + } + else if (cputype == MACH_O_CPU_TYPE_X86_64 + || cputype == MACH_O_CPU_TYPE_POWERPC_64) + { + if (mach_o_word_size () != 64) + goto not_for_target; + } + + /* Is this an MH_OBJECT file? */ + if (get_uint32 (&mach_o_file->u.header.filetype[0]) != MACH_O_MH_OBJECT) + error ("Mach-O file %s is not an MH_OBJECT file", + mach_o_file->base.filename); + + /* Save the header for future use. */ + memcpy (&cached_mach_o_header, &mach_o_file->u.header, + sizeof (cached_mach_o_header)); + + return true; + + not_for_target: + error ("file %s is not a Mach-O object file for target", + mach_o_file->base.filename); + return false; +} + + +/* Read a Mach-O LC_SEGMENT command (32 bits) from MACH_O_FILE and + validate it. + The file descriptor in MACH_O_FILE points at the start of the load + command. On sucess, returns true and advances the file pointer + past the end of the load command. On failure, returns false. */ + +static bool +validate_mach_o_segment_command_32 (lto_mach_o_file *mach_o_file) +{ + mach_o_segment_command_32 seg_cmd_32; + unsigned int i; + ssize_t n; + off_t startpos; + + /* Fields we're interested in. */ + uint32_t cmd; + uint32_t cmdsize; + uint32_t nsects; + + startpos = lseek (mach_o_file->fd, 0, SEEK_CUR); + + n = sizeof (mach_o_segment_command_32); + if (read (mach_o_file->fd, (void *) &seg_cmd_32, n) != n) + goto fail; + + cmd = get_uint32 (&seg_cmd_32.cmd[0]); + cmdsize = get_uint32 (&seg_cmd_32.cmdsize[0]); + nsects = get_uint32 (&seg_cmd_32.nsects[0]); + gcc_assert (cmd == MACH_O_LC_SEGMENT); + + /* Validate section table entries. */ + for (i = 0; i < nsects; i++) + { + mach_o_section_32 sec_32; + lto_mach_o_section ltosec; + + n = sizeof (mach_o_section_32); + if (read (mach_o_file->fd, &sec_32, n) != n) + goto fail; + + /* ??? Perform some checks. */ + + /* Looks ok, so record its details. We don't read the + string table or set up names yet; we'll do that when + we build the hash table. */ + ltosec = mach_o_new_section (mach_o_file, NULL); + memcpy (<osec->u.section_32, &sec_32, sizeof (sec_32)); + } + + if (lseek (mach_o_file->fd, 0, SEEK_CUR) != startpos + cmdsize) + goto fail; + + return true; + + fail: + error ("could not read LC_SEGMENT command in Mach-O file %s", + mach_o_file->base.filename); + return false; +} + + +/* Read a Mach-O LC_SEGMENT_64 command from MACH_O_FILE and validate it. + The file descriptor in MACH_O_FILE points at the start of the load + command. On sucess, returns true and advances the file pointer + past the end of the load command. On failure, returns false. */ + +static bool +validate_mach_o_segment_command_64 (lto_mach_o_file *mach_o_file) +{ + mach_o_segment_command_64 seg_cmd_64; + unsigned int i; + ssize_t n; + off_t startpos; + + /* Fields we're interested in. */ + uint32_t cmd; + uint32_t cmdsize; + uint32_t nsects; + + startpos = lseek (mach_o_file->fd, 0, SEEK_CUR); + + n = sizeof (mach_o_segment_command_64); + if (read (mach_o_file->fd, (void *) &seg_cmd_64, n) != n) + goto fail; + + cmd = get_uint32 (&seg_cmd_64.cmd[0]); + cmdsize = get_uint32 (&seg_cmd_64.cmdsize[0]); + nsects = get_uint32 (&seg_cmd_64.nsects[0]); + gcc_assert (cmd == MACH_O_LC_SEGMENT_64); + + /* Validate section table entries. */ + for (i = 0; i < nsects; i++) + { + mach_o_section_64 sec_64; + lto_mach_o_section ltosec; + + n = sizeof (mach_o_section_64); + if (read (mach_o_file->fd, &sec_64, n) != n) + goto fail; + + /* ??? Perform some checks. */ + + /* Looks ok, so record its details. We don't read the + string table or set up names yet; we'll do that when + we build the hash table. */ + ltosec = mach_o_new_section (mach_o_file, NULL); + memcpy (<osec->u.section_64, &sec_64, sizeof (sec_64)); + } + + if (lseek (mach_o_file->fd, 0, SEEK_CUR) != startpos + cmdsize) + goto fail; + + return true; + + fail: + error ("could not read LC_SEGMENT_64 command in Mach-O file %s", + mach_o_file->base.filename); + return false; +} + +/* Read a Mach-O load commands from MACH_O_FILE and validate it. + The file descriptor in MACH_O_FILE points at the start of the load + command. On sucess, returns true and advances the file pointer + past the end of the load command. On failure, returns false. */ + +static bool +validate_mach_o_load_command (lto_mach_o_file *mach_o_file) +{ + mach_o_load_command load_command; + uint32_t cmd; + uint32_t cmdsize; + ssize_t n; + + n = sizeof (load_command); + if (read (mach_o_file->fd, &load_command, n) != n) + { + error ("could not read load commands in Mach-O file %s", + mach_o_file->base.filename); + return false; + } + lseek (mach_o_file->fd, -1 * (off_t) sizeof (load_command), SEEK_CUR); + + cmd = get_uint32 (&load_command.cmd[0]); + cmdsize = get_uint32 (&load_command.cmdsize[0]); + switch (cmd) + { + case MACH_O_LC_SEGMENT: + return validate_mach_o_segment_command_32 (mach_o_file); + case MACH_O_LC_SEGMENT_64: + return validate_mach_o_segment_command_64 (mach_o_file); + + default: + /* Just skip over it. */ + lseek (mach_o_file->fd, cmdsize, SEEK_CUR); + return true; + } +} + +/* Validate's MACH_O_FILE's executable header and, if cached_mach_o_header is + uninitialized, caches the results. Also records the section header string + table's section index. Returns true on success, false on failure. */ + +static bool +validate_file (lto_mach_o_file *mach_o_file) +{ + uint32_t i, ncmds; + + /* Read and sanity check the raw header. */ + if (! validate_mach_o_header (mach_o_file)) + return false; + + ncmds = get_uint32 (&mach_o_file->u.header.ncmds[0]); + for (i = 0; i < ncmds; ++i) + if (! validate_mach_o_load_command (mach_o_file)) + return false; + + return true; +} + +/* Initialize MACH_O_FILE's executable header using cached data from previously + read files. */ + +static void +init_mach_o_header (lto_mach_o_file *mach_o_file) +{ + gcc_assert (cached_mach_o_magic != 0); + memcpy (&mach_o_file->u.header, + &cached_mach_o_header, + sizeof (mach_o_file->u.header)); + put_uint32 (&mach_o_file->u.header.ncmds[0], 0); + put_uint32 (&mach_o_file->u.header.sizeofcmds[0], 0); +} + +/* Open Mach-O file FILENAME. If WRITABLE is true, the file is opened for write + and, if necessary, created. Otherwise, the file is opened for reading. + Returns the opened file. */ + +lto_file * +lto_obj_file_open (const char *filename, bool writable) +{ + lto_mach_o_file *mach_o_file; + lto_file *result = NULL; + off_t offset; + const char *offset_p; + char *fname; + struct stat statbuf; + + offset_p = strchr (filename, '@'); + if (!offset_p) + { + fname = xstrdup (filename); + offset = 0; + } + else + { + /* The file started with '@' is a file containing command line + options. Stop if it doesn't exist. */ + if (offset_p == filename) + fatal_error ("command line option file '%s' does not exist", + filename); + + fname = (char *) xmalloc (offset_p - filename + 1); + memcpy (fname, filename, offset_p - filename); + fname[offset_p - filename] = '\0'; + offset_p += 3; /* skip the @0x */ + offset = lto_parse_hex (offset_p); + } + + /* Set up. */ + mach_o_file = XCNEW (lto_mach_o_file); + result = (lto_file *) mach_o_file; + lto_file_init (result, fname, offset); + mach_o_file->fd = -1; + mach_o_file->writable = writable; + + /* Open the file. */ + mach_o_file->fd = open (fname, + O_BINARY | (writable ? O_WRONLY | O_CREAT | O_TRUNC : O_RDONLY), 0666); + + if (mach_o_file->fd == -1) + { + error ("could not open file %s", fname); + goto fail; + } + + if (stat (fname, &statbuf) < 0) + { + error ("could not stat file %s", fname); + goto fail; + } + + mach_o_file->file_size = statbuf.st_size; + + /* If the object is in an archive, get it out. */ + if (offset != 0) + { + char ar_tail[12]; + int size; + + /* Surely not? */ + gcc_assert (!writable); + + /* Seek to offset, or error. */ + if (lseek (mach_o_file->fd, offset, SEEK_SET) != (ssize_t) offset) + { + error ("could not find archive member @0x%lx", (long) offset); + goto fail; + } + + /* Now seek back 12 chars and read the tail of the AR header to + find the length of the member file. */ + if (lseek (mach_o_file->fd, -12, SEEK_CUR) < 0 + || read (mach_o_file->fd, ar_tail, 12) != 12 + || lseek (mach_o_file->fd, 0, SEEK_CUR) != (ssize_t) offset + || ar_tail[10] != '`' || ar_tail[11] != '\n') + { + error ("could not find archive header @0x%lx", (long) offset); + goto fail; + } + + ar_tail[11] = 0; + if (sscanf (ar_tail, "%d", &size) != 1) + { + error ("invalid archive header @0x%lx", (long) offset); + goto fail; + } + mach_o_file->file_size = size; + } + + if (writable) + { + init_mach_o_header (mach_o_file); + } + else + if (! validate_file (mach_o_file)) + goto fail; + + return result; + + fail: + if (result) + lto_obj_file_close (result); + return NULL; +} + + +/* Write the data in MACH_O_FILE to a real Mach-O binary object. + We write a header, a segment load command, and section data. */ + +static bool +mach_o_write_object_file (lto_mach_o_file *mach_o_file) +{ + lto_mach_o_section sec, snsec; + lto_mach_o_data snsec_data; + ssize_t hdrsize, cmdsize, secsize; + size_t num_sections, snsec_size, total_sec_size; + unsigned int sec_offs, strtab_offs; + int i; + bool write_err = false; + + /* The number of sections we will write is the number of sections added by + the streamer, plus 1 for the section names section. */ + num_sections = VEC_length (lto_mach_o_section, mach_o_file->section_vec) + 1; + + /* Calculate the size of the basic data structures on disk. */ + if (mach_o_word_size () == 64) + { + hdrsize = sizeof (mach_o_header_64); + secsize = sizeof (mach_o_section_64); + cmdsize = sizeof (mach_o_segment_command_64) + num_sections * secsize; + } + else + { + hdrsize = sizeof (mach_o_header_32); + secsize = sizeof (mach_o_section_32); + cmdsize = sizeof (mach_o_segment_command_32) + num_sections * secsize; + } + + /* Allocate the section names section. */ + snsec_size = 0; + for (i = 0; + VEC_iterate (lto_mach_o_section, mach_o_file->section_vec, i, sec); + i++) + snsec_size += strlen (sec->name) + 1; + snsec = mach_o_new_section (mach_o_file, NULL); + snsec->name = LTO_NAMES_SECTION; + snsec_data = mach_o_new_data (snsec); + snsec_data->d_buf = XCNEWVEC (char, snsec_size); + snsec_data->d_size = snsec_size; + + /* Position all the sections, and fill out their headers. */ + sec_offs = hdrsize + cmdsize; + strtab_offs = 0; + total_sec_size = 0; + for (i = 0; + VEC_iterate (lto_mach_o_section, mach_o_file->section_vec, i, sec); + i++) + { + lto_mach_o_data data; + size_t data_size; + /* Put the section and segment names. Add the section name to the + section names section (unless, of course, this *is* the section + names section). */ + if (sec == snsec) + snprintf (sec->u.section.sectname, 16, "%s", LTO_NAMES_SECTION); + else + { + sprintf (sec->u.section.sectname, "__%08X", strtab_offs); + memcpy ((char *) snsec_data->d_buf + strtab_offs, sec->name, strlen (sec->name)); + } + memcpy (&sec->u.section.segname[0], + LTO_SEGMENT_NAME, strlen (LTO_SEGMENT_NAME)); + + /* Add layout and attributes. */ + for (data = sec->data_chain, data_size = 0; data; data = data->next) + data_size += data->d_size; + if (mach_o_word_size () == 64) + { + put_uint64 (&sec->u.section_64.addr[0], total_sec_size); + put_uint64 (&sec->u.section_64.size[0], data_size); + put_uint32 (&sec->u.section_64.offset[0], sec_offs); + put_uint32 (&sec->u.section_64.flags[0], MACH_O_S_ATTR_DEBUG); + } + else + { + put_uint32 (&sec->u.section_64.addr[0], total_sec_size); + put_uint32 (&sec->u.section_32.size[0], data_size); + put_uint32 (&sec->u.section_32.offset[0], sec_offs); + put_uint32 (&sec->u.section_32.flags[0], MACH_O_S_ATTR_DEBUG); + } + + sec_offs += data_size; + total_sec_size += data_size; + strtab_offs += strlen (sec->name) + 1; + } + + /* We can write the data now. As there's no way to indicate an error return + from this hook, error handling is limited to not wasting our time doing + any more writes in the event that any one fails. */ + + /* Write the header. */ + put_uint32 (&mach_o_file->u.header.ncmds[0], 1); + put_uint32 (&mach_o_file->u.header.sizeofcmds[0], cmdsize); + write_err = (write (mach_o_file->fd, + &mach_o_file->u.header, hdrsize) != hdrsize); + /* Write the segment load command. */ + if (mach_o_word_size () == 64) + { + mach_o_segment_command_64 lc; + ssize_t lc_size = sizeof (lc); + memset (&lc, 0, lc_size); + put_uint32 (&lc.cmd[0], MACH_O_LC_SEGMENT_64); + put_uint32 (&lc.cmdsize[0], cmdsize); + put_uint64 (&lc.fileoff[0], hdrsize + cmdsize); + put_uint64 (&lc.filesize[0], total_sec_size); + put_uint32 (&lc.nsects[0], num_sections); + write_err = (write (mach_o_file->fd, &lc, lc_size) != lc_size); + } + else + { + mach_o_segment_command_32 lc; + ssize_t lc_size = sizeof (lc); + memset (&lc, 0, lc_size); + put_uint32 (&lc.cmd[0], MACH_O_LC_SEGMENT); + put_uint32 (&lc.cmdsize[0], cmdsize); + put_uint32 (&lc.fileoff[0], hdrsize + cmdsize); + put_uint32 (&lc.filesize[0], total_sec_size); + put_uint32 (&lc.nsects[0], num_sections); + write_err = (write (mach_o_file->fd, &lc, lc_size) != lc_size); + } + for (i = 0; + !write_err + && VEC_iterate (lto_mach_o_section, mach_o_file->section_vec, i, sec); + i++) + write_err = (write (mach_o_file->fd, + &sec->u.section, secsize) != secsize); + + gcc_assert (lseek (mach_o_file->fd, 0, SEEK_CUR) == hdrsize + cmdsize); + + /* Write the section data. */ + for (i = 0; + !write_err + && VEC_iterate (lto_mach_o_section, mach_o_file->section_vec, i, sec); + i++) + { + lto_mach_o_data data; + + for (data = sec->data_chain; data; data = data->next) + { + if (!write_err) + write_err = (write (mach_o_file->fd, data->d_buf, data->d_size) + != data->d_size); + else + break; + } + } + + return !write_err; +} + +/* Close Mach-O file FILE and clean up any associated data structures. If FILE + was opened for writing, the file's Mach-O data is written at this time. Any + cached data buffers are freed. */ + +void +lto_obj_file_close (lto_file *file) +{ + lto_mach_o_file *mach_o_file = (lto_mach_o_file *) file; + struct lto_char_ptr_base *cur, *tmp; + lto_mach_o_section sec; + bool write_err = false; + int i; + + /* If this file is open for writing, write a Mach-O object file. */ + if (mach_o_file->writable) + { + if (! mach_o_write_object_file (mach_o_file)) + fatal_error ("cannot write Mach-O object file"); + } + + /* Close the file, we're done. */ + if (mach_o_file->fd != -1) + close (mach_o_file->fd); + + /* Free any data buffers. */ + cur = mach_o_file->data; + while (cur) + { + tmp = cur; + cur = (struct lto_char_ptr_base *) cur->ptr; + free (tmp); + } + + /* Free any sections and their data chains. */ + for (i = 0; + VEC_iterate (lto_mach_o_section, mach_o_file->section_vec, i, sec); + i++) + { + lto_mach_o_data curdata, nextdata; + curdata = sec->data_chain; + while (curdata) + { + nextdata = curdata->next; + free (curdata); + curdata = nextdata; + } + free (sec); + } + VEC_free (lto_mach_o_section, heap, mach_o_file->section_vec); + + free (file); + + /* If there was an error, mention it. */ + if (write_err) + error ("I/O error writing Mach-O output file"); +} + Index: gcc/lto/Make-lang.in =================================================================== --- gcc/lto/Make-lang.in (revision 161328) +++ gcc/lto/Make-lang.in (working copy) @@ -91,3 +91,6 @@ lto/lto-elf.o: lto/lto-elf.c $(CONFIG_H) coretypes lto/lto-coff.o: lto/lto-coff.c $(CONFIG_H) coretypes.h $(SYSTEM_H) \ toplev.h $(LTO_H) $(TM_H) $(LIBIBERTY_H) $(GGC_H) $(LTO_STREAMER_H) \ lto/lto-coff.h +lto/lto-macho.o: lto/lto-macho.c $(CONFIG_H) coretypes.h $(SYSTEM_H) \ + toplev.h $(LTO_H) $(TM_H) $(LIBIBERTY_H) $(GGC_H) $(LTO_STREAMER_H) \ + lto/lto-macho.h lto/lto-endian.h