From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (qmail 3841 invoked by alias); 11 Apr 2005 18:55:34 -0000 Mailing-List: contact binutils-help@sources.redhat.com; run by ezmlm Precedence: bulk List-Subscribe: List-Archive: List-Post: List-Help: , Sender: binutils-owner@sources.redhat.com Received: (qmail 3743 invoked from network); 11 Apr 2005 18:55:23 -0000 Received: from unknown (HELO mail.codesourcery.com) (65.74.133.9) by sourceware.org with SMTP; 11 Apr 2005 18:55:23 -0000 Received: (qmail 15093 invoked from network); 11 Apr 2005 18:55:22 -0000 Received: from localhost (HELO ?10.1.1.125?) (julian@127.0.0.1) by mail.codesourcery.com with SMTP; 11 Apr 2005 18:55:22 -0000 Message-ID: <425AC810.5040408@codesourcery.com> Date: Mon, 11 Apr 2005 18:55:00 -0000 From: Julian Brown User-Agent: Debian Thunderbird 1.0 (X11/20050116) MIME-Version: 1.0 To: binutils@sources.redhat.com CC: Julian Brown Subject: [PATCH] Enable use of ARMv5t BLX for Thumb-mode calls via PLT Content-Type: multipart/mixed; boundary="------------050909080806070205080305" X-SW-Source: 2005-04/txt/msg00248.txt.bz2 This is a multi-part message in MIME format. --------------050909080806070205080305 Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit Content-length: 1794 Hi, This patch allows the ARMv5t Thumb BLX instruction to be used (optionally) to call PLT entries, which should provide a slight speed boost. A previous, SymbianOS-specific version of the patch was posted here: http://sourceware.org/ml/binutils/2005-03/msg00433.html With this patch, the option "--use-thumb-blx" can be passed to the linker to enable the use of the BLX instruction when the target supports it. The format of the PLT itself isn't changed by this patch: there is still a Thumb stub before each ARM PLT entry, but it is now unused. The ARM mode entry is called directly instead. Additionally, an instruction-encoding bug is fixed from the previous unapplied version of the patch, and an uninitialised value is now initialised (fix_v4bx field in elf32_arm_link_hash_table_create, just cosmetic I think). Tested on arm-none-eabi and arm-none-symbianelf. OK to apply? ChangeLog: bfd: * bfd-in.h (bfd_elf32_arm_set_target_relocs): Update prototype. * bfd-in2.h: Regenerate. * elf32-arm.c (elf32_arm_link_hash_table): New field, 'use_thumb_blx'. (elf32_arm_link_hash_table_create): Initialise fix_v4bx, use_thumb_blx. (bfd_elf32_arm_set_target_relocs): Handle use_thumb_blx. (elf32_arm_final_link_relocate): Use Thumb BLX for R_ARM_THM_PC22 relocations if requested to. (elf32_arm_symbian_link_hash_table_create): Enable use_thumb_blx by default for SymbianOS. ld: * ld.texinfo: Document --use-thumb-blx. * emultempl/armelf.em (use_thumb_blx): New variable. (arm_elf_create_output_section_statements): Communicate value of use_thumb_blx to bfd. (PARSE_AND_LIST_PROLOGUE): Add OPTION_USE_THUMB_BLX. (PARSE_AND_LIST_OPTIONS): Add --use-thumb-blx option. (PARSE_AND_LIST_ARGS_CASES): Add OPTION_USE_THUMB_BLX case. --------------050909080806070205080305 Content-Type: text/plain; name="patch-2" Content-Transfer-Encoding: 7bit Content-Disposition: inline; filename="patch-2" Content-length: 8456 Index: bfd/bfd-in.h =================================================================== RCS file: /cvs/src/src/bfd/bfd-in.h,v retrieving revision 1.97 diff -c -p -r1.97 bfd-in.h *** bfd/bfd-in.h 30 Mar 2005 17:19:27 -0000 1.97 --- bfd/bfd-in.h 11 Apr 2005 18:49:01 -0000 *************** extern bfd_boolean bfd_elf32_arm_process *** 816,822 **** (bfd *, struct bfd_link_info *, int); void bfd_elf32_arm_set_target_relocs ! (struct bfd_link_info *, int, char *, int); extern bfd_boolean bfd_elf32_arm_get_bfd_for_interworking (bfd *, struct bfd_link_info *); --- 816,822 ---- (bfd *, struct bfd_link_info *, int); void bfd_elf32_arm_set_target_relocs ! (struct bfd_link_info *, int, char *, int, int); extern bfd_boolean bfd_elf32_arm_get_bfd_for_interworking (bfd *, struct bfd_link_info *); Index: bfd/elf32-arm.c =================================================================== RCS file: /cvs/src/src/bfd/elf32-arm.c,v retrieving revision 1.31 diff -c -p -r1.31 elf32-arm.c *** bfd/elf32-arm.c 8 Apr 2005 11:47:59 -0000 1.31 --- bfd/elf32-arm.c 11 Apr 2005 18:49:01 -0000 *************** struct elf32_arm_link_hash_table *** 1324,1329 **** --- 1324,1332 ---- /* Nonzero to fix BX instructions for ARMv4 targets. */ int fix_v4bx; + /* Nonzero if the Thumb BLX instruction is available for use. */ + int use_thumb_blx; + /* The number of bytes in the initial entry in the PLT. */ bfd_size_type plt_header_size; *************** elf32_arm_link_hash_table_create (bfd *a *** 1559,1564 **** --- 1562,1569 ---- ret->plt_header_size = 20; ret->plt_entry_size = 12; #endif + ret->fix_v4bx = 0; + ret->use_thumb_blx = 0; ret->symbian_p = 0; ret->use_rel = 1; ret->sym_sec.abfd = NULL; *************** void *** 2125,2131 **** bfd_elf32_arm_set_target_relocs (struct bfd_link_info *link_info, int target1_is_rel, char * target2_type, ! int fix_v4bx) { struct elf32_arm_link_hash_table *globals; --- 2130,2137 ---- bfd_elf32_arm_set_target_relocs (struct bfd_link_info *link_info, int target1_is_rel, char * target2_type, ! int fix_v4bx, ! int use_thumb_blx) { struct elf32_arm_link_hash_table *globals; *************** bfd_elf32_arm_set_target_relocs (struct *** 2144,2149 **** --- 2150,2156 ---- target2_type); } globals->fix_v4bx = fix_v4bx; + globals->use_thumb_blx = use_thumb_blx; } #endif *************** elf32_arm_final_link_relocate (reloc_how *** 2893,2898 **** --- 2900,2906 ---- bfd_signed_vma reloc_signed_min = ~ reloc_signed_max; bfd_vma check; bfd_signed_vma signed_check; + bfd_boolean thumb_plt_call = FALSE; /* Need to refetch the addend and squish the two 11 bit pieces together. */ *************** elf32_arm_final_link_relocate (reloc_how *** 2942,2949 **** value = (splt->output_section->vma + splt->output_offset + h->plt.offset); ! /* Target the Thumb stub before the ARM PLT entry. */ ! value -= 4; *unresolved_reloc_p = FALSE; } --- 2950,2968 ---- value = (splt->output_section->vma + splt->output_offset + h->plt.offset); ! if (globals->use_thumb_blx) ! { ! /* If the Thumb BLX instruction is available, convert the ! BL to a BLX instruction to call the ARM-mode PLT entry. */ ! if ((lower_insn & (0x3 << 11)) == 0x3 << 11) ! { ! lower_insn = (lower_insn & ~(0x3 << 11)) | 0x1 << 11; ! thumb_plt_call = TRUE; ! } ! } ! else ! /* Target the Thumb stub before the ARM PLT entry. */ ! value -= PLT_THUMB_STUB_SIZE; *unresolved_reloc_p = FALSE; } *************** elf32_arm_final_link_relocate (reloc_how *** 2967,2974 **** overflow = TRUE; #ifndef OLD_ARM_ABI ! if (r_type == R_ARM_THM_XPC22 ! && ((lower_insn & 0x1800) == 0x0800)) /* For a BLX instruction, make sure that the relocation is rounded up to a word boundary. This follows the semantics of the instruction which specifies that bit 1 of the target address will come from bit --- 2986,2994 ---- overflow = TRUE; #ifndef OLD_ARM_ABI ! if ((r_type == R_ARM_THM_XPC22 ! && ((lower_insn & 0x1800) == 0x0800)) ! || thumb_plt_call) /* For a BLX instruction, make sure that the relocation is rounded up to a word boundary. This follows the semantics of the instruction which specifies that bit 1 of the target address will come from bit *************** elf32_arm_symbian_link_hash_table_create *** 6505,6510 **** --- 6525,6532 ---- /* The PLT entries are each three instructions. */ htab->plt_entry_size = 4 * NUM_ELEM (elf32_arm_symbian_plt_entry); htab->symbian_p = 1; + /* Symbian uses armv5t or above, so use_thumb_blx is always true. */ + htab->use_thumb_blx = 1; htab->root.is_relocatable_executable = 1; } return ret; Index: ld/ld.texinfo =================================================================== RCS file: /cvs/src/src/ld/ld.texinfo,v retrieving revision 1.140 diff -c -p -r1.140 ld.texinfo *** ld/ld.texinfo 1 Feb 2005 17:31:01 -0000 1.140 --- ld/ld.texinfo 11 Apr 2005 18:49:01 -0000 *************** linker, which causes v4t @code{BX rM} in *** 5262,5267 **** --- 5262,5276 ---- In the former case, the switch should not be used, and @samp{R_ARM_V4BX} relocations are ignored. + @cindex USE_THUMB_BLX + @kindex --use-thumb-blx + The @samp{--use-thumb-blx} switch enables calls via the PLT in Thumb + code to use the BLX instruction, as defined in ARMv5t and above. This + should lead to such calls executing slightly faster. + + This option is enabled implicitly for SymbianOS, so there is no need to + specify it if you are using that target. + @ifclear GENERIC @lowersections @end ifclear Index: ld/emultempl/armelf.em =================================================================== RCS file: /cvs/src/src/ld/emultempl/armelf.em,v retrieving revision 1.44 diff -c -p -r1.44 armelf.em *** ld/emultempl/armelf.em 3 Mar 2005 11:52:04 -0000 1.44 --- ld/emultempl/armelf.em 11 Apr 2005 18:49:01 -0000 *************** static int byteswap_code = 0; *** 32,37 **** --- 32,38 ---- static int target1_is_rel = 0${TARGET1_IS_REL}; static char *target2_type = "${TARGET2_TYPE}"; static int fix_v4bx = 0; + static int use_thumb_blx = 0; static void gld${EMULATION_NAME}_before_parse (void) *************** static void *** 192,198 **** arm_elf_create_output_section_statements (void) { bfd_elf32_arm_set_target_relocs (&link_info, target1_is_rel, target2_type, ! fix_v4bx); } EOF --- 193,199 ---- arm_elf_create_output_section_statements (void) { bfd_elf32_arm_set_target_relocs (&link_info, target1_is_rel, target2_type, ! fix_v4bx, use_thumb_blx); } EOF *************** PARSE_AND_LIST_PROLOGUE=' *** 207,212 **** --- 208,214 ---- #define OPTION_TARGET1_ABS 304 #define OPTION_TARGET2 305 #define OPTION_FIX_V4BX 306 + #define OPTION_USE_THUMB_BLX 307 ' PARSE_AND_LIST_SHORTOPTS=p *************** PARSE_AND_LIST_LONGOPTS=' *** 219,224 **** --- 221,227 ---- { "target1-abs", no_argument, NULL, OPTION_TARGET1_ABS}, { "target2", required_argument, NULL, OPTION_TARGET2}, { "fix-v4bx", no_argument, NULL, OPTION_FIX_V4BX}, + { "use-thumb-blx", no_argument, NULL, OPTION_USE_THUMB_BLX}, ' PARSE_AND_LIST_OPTIONS=' *************** PARSE_AND_LIST_OPTIONS=' *** 228,233 **** --- 231,237 ---- fprintf (file, _(" --target1=abs Interpret R_ARM_TARGET1 as R_ARM_ABS32\n")); fprintf (file, _(" --target2= Specify definition of R_ARM_TARGET2\n")); fprintf (file, _(" --fix-v4bx Rewrite BX rn as MOV pc, rn for ARMv4\n")); + fprintf (file, _(" --use-thumb-blx Enable use of Thumb BLX instruction\n")); ' PARSE_AND_LIST_ARGS_CASES=' *************** PARSE_AND_LIST_ARGS_CASES=' *** 258,263 **** --- 262,271 ---- case OPTION_FIX_V4BX: fix_v4bx = 1; break; + + case OPTION_USE_THUMB_BLX: + use_thumb_blx = 1; + break; ' # We have our own after_open and before_allocation functions, but they call --------------050909080806070205080305--