/* ehopt.c--optimize gcc exception frame information. Copyright 1998, 2000, 2001 Free Software Foundation, Inc. Written by Ian Lance Taylor . This file is part of GAS, the GNU Assembler. GAS 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 2, or (at your option) any later version. GAS 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 GAS; see the file COPYING. If not, write to the Free Software Foundation, 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. */ #include "as.h" #include "subsegs.h" /* We include this ELF file, even though we may not be assembling for ELF, since the exception frame information is always in a format derived from DWARF. */ #include "elf/dwarf2.h" /* Try to optimize gcc 2.8 exception frame information. Exception frame information is emitted for every function in the .eh_frame or .debug_frame sections. Simple information for a function with no exceptions looks like this: __FRAME_BEGIN__: .4byte .LLCIE1 / Length of Common Information Entry .LSCIE1: #if .eh_frame .4byte 0x0 / CIE Identifier Tag #elif .debug_frame .4byte 0xffffffff / CIE Identifier Tag #endif .byte 0x1 / CIE Version .byte 0x0 / CIE Augmentation (none) .byte 0x1 / ULEB128 0x1 (CIE Code Alignment Factor) .byte 0x7c / SLEB128 -4 (CIE Data Alignment Factor) .byte 0x8 / CIE RA Column .byte 0xc / DW_CFA_def_cfa .byte 0x4 / ULEB128 0x4 .byte 0x4 / ULEB128 0x4 .byte 0x88 / DW_CFA_offset, column 0x8 .byte 0x1 / ULEB128 0x1 .align 4 .LECIE1: .set .LLCIE1,.LECIE1-.LSCIE1 / CIE Length Symbol .4byte .LLFDE1 / FDE Length .LSFDE1: .4byte .LSFDE1-__FRAME_BEGIN__ / FDE CIE offset .4byte .LFB1 / FDE initial location .4byte .LFE1-.LFB1 / FDE address range .byte 0x4 / DW_CFA_advance_loc4 .4byte .LCFI0-.LFB1 .byte 0xe / DW_CFA_def_cfa_offset .byte 0x8 / ULEB128 0x8 .byte 0x85 / DW_CFA_offset, column 0x5 .byte 0x2 / ULEB128 0x2 .byte 0x4 / DW_CFA_advance_loc4 .4byte .LCFI1-.LCFI0 .byte 0xd / DW_CFA_def_cfa_register .byte 0x5 / ULEB128 0x5 .byte 0x4 / DW_CFA_advance_loc4 .4byte .LCFI2-.LCFI1 .byte 0x2e / DW_CFA_GNU_args_size .byte 0x4 / ULEB128 0x4 .byte 0x4 / DW_CFA_advance_loc4 .4byte .LCFI3-.LCFI2 .byte 0x2e / DW_CFA_GNU_args_size .byte 0x0 / ULEB128 0x0 .align 4 .LEFDE1: .set .LLFDE1,.LEFDE1-.LSFDE1 / FDE Length Symbol The immediate issue we can address in the assembler is the DW_CFA_advance_loc4 followed by a four byte value. The value is the difference of two addresses in the function. Since gcc does not know this value, it always uses four bytes. We will know the value at the end of assembly, so we can do better. */ struct cie_info { unsigned code_alignment; int z_augmentation; }; static int get_cie_info PARAMS ((struct cie_info *)); /* Extract information from the CIE. */ static int get_cie_info (info) struct cie_info *info; { fragS *f; fixS *fix; int offset; char CIE_id; char augmentation[10]; int iaug; int code_alignment = 0; /* We should find the CIE at the start of the section. */ #if defined (BFD_ASSEMBLER) || defined (MANY_SEGMENTS) f = seg_info (now_seg)->frchainP->frch_root; #else f = frchain_now->frch_root; #endif #ifdef BFD_ASSEMBLER fix = seg_info (now_seg)->frchainP->fix_root; #else fix = *seg_fix_rootP; #endif /* Look through the frags of the section to find the code alignment. */ /* First make sure that the CIE Identifier Tag is 0/-1. */ if (strcmp (segment_name (now_seg), ".debug_frame") == 0) CIE_id = (char)0xff; else CIE_id = 0; offset = 4; while (f != NULL && offset >= f->fr_fix) { offset -= f->fr_fix; f = f->fr_next; } if (f == NULL || f->fr_fix - offset < 4 || f->fr_literal[offset] != CIE_id || f->fr_literal[offset + 1] != CIE_id || f->fr_literal[offset + 2] != CIE_id || f->fr_literal[offset + 3] != CIE_id) return 0; /* Next make sure the CIE version number is 1. */ offset += 4; while (f != NULL && offset >= f->fr_fix) { offset -= f->fr_fix; f = f->fr_next; } if (f == NULL || f->fr_fix - offset < 1 || f->fr_literal[offset] != 1) return 0; /* Skip the augmentation (a null terminated string). */ iaug = 0; ++offset; while (1) { while (f != NULL && offset >= f->fr_fix) { offset -= f->fr_fix; f = f->fr_next; } if (f == NULL) return 0; while (offset < f->fr_fix && f->fr_literal[offset] != '\0') { if ((size_t) iaug < (sizeof augmentation) - 1) { augmentation[iaug] = f->fr_literal[offset]; ++iaug; } ++offset; } if (offset < f->fr_fix) break; } ++offset; while (f != NULL && offset >= f->fr_fix) { offset -= f->fr_fix; f = f->fr_next; } if (f == NULL) return 0; augmentation[iaug] = '\0'; if (augmentation[0] == '\0') { /* No augmentation. */ } else if (strcmp (augmentation, "eh") == 0) { /* We have to skip a pointer. Unfortunately, we don't know how large it is. We find out by looking for a matching fixup. */ while (fix != NULL && (fix->fx_frag != f || fix->fx_where != offset)) fix = fix->fx_next; if (fix == NULL) offset += 4; else offset += fix->fx_size; while (f != NULL && offset >= f->fr_fix) { offset -= f->fr_fix; f = f->fr_next; } if (f == NULL) return 0; } else if (augmentation[0] != 'z') return 0; /* We're now at the code alignment factor, which is a ULEB128. If it isn't a single byte, forget it. */ code_alignment = f->fr_literal[offset] & 0xff; if ((code_alignment & 0x80) != 0) code_alignment = 0; info->code_alignment = code_alignment; info->z_augmentation = (augmentation[0] == 'z'); return 1; } /* This function is called from emit_expr. It looks for cases which we can optimize. Rather than try to parse all this information as we read it, we look for a single byte DW_CFA_advance_loc4 followed by a 4 byte difference. We turn that into a rs_cfa_advance frag, and handle those frags at the end of the assembly. If the gcc output changes somewhat, this optimization may stop working. This function returns non-zero if it handled the expression and emit_expr should not do anything, or zero otherwise. It can also change *EXP and *PNBYTES. */ int check_eh_frame (exp, pnbytes) expressionS *exp; unsigned int *pnbytes; { struct frame_data { enum frame_state { state_idle, state_saw_size, state_saw_cie_offset, state_saw_pc_begin, state_seeing_aug_size, state_skipping_aug, state_wait_loc4, state_saw_loc4, state_error, } state; int cie_info_ok; struct cie_info cie_info; symbolS *size_end_sym; fragS *loc4_frag; int loc4_fix; int aug_size; int aug_shift; }; static struct frame_data eh_frame_data; static struct frame_data debug_frame_data; struct frame_data *d; /* Don't optimize. */ if (flag_traditional_format) return 0; /* Select the proper section data. */ if (strcmp (segment_name (now_seg), ".eh_frame") == 0) d = &eh_frame_data; else if (strcmp (segment_name (now_seg), ".debug_frame") == 0) d = &debug_frame_data; else return 0; if (d->state >= state_saw_size && S_IS_DEFINED (d->size_end_sym)) { /* We have come to the end of the CIE or FDE. See below where we set saw_size. We must check this first because we may now be looking at the next size. */ d->state = state_idle; } switch (d->state) { case state_idle: if (*pnbytes == 4) { /* This might be the size of the CIE or FDE. We want to know the size so that we don't accidentally optimize across an FDE boundary. We recognize the size in one of two forms: a symbol which will later be defined as a difference, or a subtraction of two symbols. Either way, we can tell when we are at the end of the FDE because the symbol becomes defined (in the case of a subtraction, the end symbol, from which the start symbol is being subtracted). Other ways of describing the size will not be optimized. */ if ((exp->X_op == O_symbol || exp->X_op == O_subtract) && ! S_IS_DEFINED (exp->X_add_symbol)) { d->state = state_saw_size; d->size_end_sym = exp->X_add_symbol; } } break; case state_saw_size: case state_saw_cie_offset: /* Assume whatever form it appears in, it appears atomically. */ d->state += 1; break; case state_saw_pc_begin: /* Decide whether we should see an augmentation. */ if (! d->cie_info_ok && ! (d->cie_info_ok = get_cie_info (&d->cie_info))) d->state = state_error; else if (d->cie_info.z_augmentation) { d->state = state_seeing_aug_size; d->aug_size = 0; d->aug_shift = 0; } else d->state = state_wait_loc4; break; case state_seeing_aug_size: /* Bytes == -1 means this comes from an leb128 directive. */ if ((int)*pnbytes == -1 && exp->X_op == O_constant) { d->aug_size = exp->X_add_number; d->state = state_skipping_aug; } else if (*pnbytes == 1 && exp->X_op == O_constant) { unsigned char byte = exp->X_add_number; d->aug_size |= (byte & 0x7f) << d->aug_shift; d->aug_shift += 7; if ((byte & 0x80) == 0) d->state = state_skipping_aug; } else d->state = state_error; break; case state_skipping_aug: if ((int)*pnbytes < 0) d->state = state_error; else { int left = (d->aug_size -= *pnbytes); if (left == 0) d->state = state_wait_loc4; else if (left < 0) d->state = state_error; } break; case state_wait_loc4: if (*pnbytes == 1 && exp->X_op == O_constant && exp->X_add_number == DW_CFA_advance_loc4) { /* This might be a DW_CFA_advance_loc4. Record the frag and the position within the frag, so that we can change it later. */ frag_grow (1); d->state = state_saw_loc4; d->loc4_frag = frag_now; d->loc4_fix = frag_now_fix (); } break; case state_saw_loc4: d->state = state_wait_loc4; if (*pnbytes != 4) break; if (exp->X_op == O_constant) { /* This is a case which we can optimize. The two symbols being subtracted were in the same frag and the expression was reduced to a constant. We can do the optimization entirely in this function. */ if (d->cie_info.code_alignment > 0 && exp->X_add_number % d->cie_info.code_alignment == 0 && exp->X_add_number / d->cie_info.code_alignment < 0x40) { d->loc4_frag->fr_literal[d->loc4_fix] = DW_CFA_advance_loc | (exp->X_add_number / d->cie_info.code_alignment); /* No more bytes needed. */ return 1; } else if (exp->X_add_number < 0x100) { d->loc4_frag->fr_literal[d->loc4_fix] = DW_CFA_advance_loc1; *pnbytes = 1; } else if (exp->X_add_number < 0x10000) { d->loc4_frag->fr_literal[d->loc4_fix] = DW_CFA_advance_loc2; *pnbytes = 2; } } else if (exp->X_op == O_subtract) { /* This is a case we can optimize. The expression was not reduced, so we can not finish the optimization until the end of the assembly. We set up a variant frag which we handle later. */ int fr_subtype; if (d->cie_info.code_alignment > 0) fr_subtype = d->cie_info.code_alignment << 3; else fr_subtype = 0; frag_var (rs_cfa, 4, 0, fr_subtype, make_expr_symbol (exp), d->loc4_fix, (char *) d->loc4_frag); return 1; } break; case state_error: /* Just skipping everything. */ break; } return 0; } /* The function estimates the size of a rs_cfa variant frag based on the current values of the symbols. It is called before the relaxation loop. We set fr_subtype{0:2} to the expected length. */ int eh_frame_estimate_size_before_relax (frag) fragS *frag; { offsetT diff; int ca = frag->fr_subtype >> 3; int ret; diff = resolve_symbol_value (frag->fr_symbol, 0); if (ca > 0 && diff % ca == 0 && diff / ca < 0x40) ret = 0; else if (diff < 0x100) ret = 1; else if (diff < 0x10000) ret = 2; else ret = 4; frag->fr_subtype = (frag->fr_subtype & ~7) | ret; return ret; } /* This function relaxes a rs_cfa variant frag based on the current values of the symbols. fr_subtype{0:2} is the current length of the frag. This returns the change in frag length. */ int eh_frame_relax_frag (frag) fragS *frag; { int oldsize, newsize; oldsize = frag->fr_subtype & 7; newsize = eh_frame_estimate_size_before_relax (frag); return newsize - oldsize; } /* This function converts a rs_cfa variant frag into a normal fill frag. This is called after all relaxation has been done. fr_subtype{0:2} will be the desired length of the frag. */ void eh_frame_convert_frag (frag) fragS *frag; { offsetT diff; fragS *loc4_frag; int loc4_fix; loc4_frag = (fragS *) frag->fr_opcode; loc4_fix = (int) frag->fr_offset; diff = resolve_symbol_value (frag->fr_symbol, finalize_syms); switch (frag->fr_subtype & 7) { case 0: { int ca = frag->fr_subtype >> 3; assert (ca > 0 && diff % ca == 0 && diff / ca < 0x40); loc4_frag->fr_literal[loc4_fix] = DW_CFA_advance_loc | (diff / ca); } break; case 1: assert (diff < 0x100); loc4_frag->fr_literal[loc4_fix] = DW_CFA_advance_loc1; frag->fr_literal[frag->fr_fix] = diff; break; case 2: assert (diff < 0x10000); loc4_frag->fr_literal[loc4_fix] = DW_CFA_advance_loc2; md_number_to_chars (frag->fr_literal + frag->fr_fix, diff, 2); break; default: md_number_to_chars (frag->fr_literal + frag->fr_fix, diff, 4); break; } frag->fr_fix += frag->fr_subtype & 7; frag->fr_type = rs_fill; frag->fr_subtype = 0; frag->fr_offset = 0; }