/* MMIX-specific support for 64-bit ELF. Copyright (C) 2001-2024 Free Software Foundation, Inc. Contributed by Hans-Peter Nilsson This file is part of BFD, the Binary File Descriptor library. This program 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 of the License, or (at your option) any later version. This program 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 this program; if not, write to the Free Software Foundation, Inc., 51 Franklin Street - Fifth Floor, Boston, MA 02110-1301, USA. */ /* No specific ABI or "processor-specific supplement" defined. */ /* TODO: - "Traditional" linker relaxation (shrinking whole sections). - Merge reloc stubs jumping to same location. - GETA stub relaxation (call a stub for out of range new R_MMIX_GETA_STUBBABLE). */ #include "sysdep.h" #include "bfd.h" #include "libbfd.h" #include "elf-bfd.h" #include "elf/mmix.h" #include "opcode/mmix.h" #define MINUS_ONE (((bfd_vma) 0) - 1) #define MAX_PUSHJ_STUB_SIZE (5 * 4) /* Put these everywhere in new code. */ #define FATAL_DEBUG \ _bfd_abort (__FILE__, __LINE__, \ "Internal: Non-debugged code (test-case missing)") #define BAD_CASE(x) \ _bfd_abort (__FILE__, __LINE__, \ "bad case for " #x) struct _mmix_elf_section_data { struct bfd_elf_section_data elf; union { struct bpo_reloc_section_info *reloc; struct bpo_greg_section_info *greg; } bpo; struct pushj_stub_info { /* Maximum number of stubs needed for this section. */ bfd_size_type n_pushj_relocs; /* Size of stubs after a mmix_elf_relax_section round. */ bfd_size_type stubs_size_sum; /* Per-reloc stubs_size_sum information. The stubs_size_sum member is the sum of these. Allocated in mmix_elf_check_common_relocs. */ bfd_size_type *stub_size; /* Offset of next stub during relocation. Somewhat redundant with the above: error coverage is easier and we don't have to reset the stubs_size_sum for relocation. */ bfd_size_type stub_offset; } pjs; /* Whether there has been a warning that this section could not be linked due to a specific cause. FIXME: a way to access the linker info or output section, then stuff the limiter guard there. */ bool has_warned_bpo; bool has_warned_pushj; }; #define mmix_elf_section_data(sec) \ ((struct _mmix_elf_section_data *) elf_section_data (sec)) /* For each section containing a base-plus-offset (BPO) reloc, we attach this struct as mmix_elf_section_data (section)->bpo, which is otherwise NULL. */ struct bpo_reloc_section_info { /* The base is 1; this is the first number in this section. */ size_t first_base_plus_offset_reloc; /* Number of BPO-relocs in this section. */ size_t n_bpo_relocs_this_section; /* Running index, used at relocation time. */ size_t bpo_index; /* We don't have access to the bfd_link_info struct in mmix_final_link_relocate. What we really want to get at is the global single struct greg_relocation, so we stash it here. */ asection *bpo_greg_section; }; /* Helper struct (in global context) for the one below. There's one of these created for every BPO reloc. */ struct bpo_reloc_request { bfd_vma value; /* Valid after relaxation. The base is 0; the first register number must be added. The offset is in range 0..255. */ size_t regindex; size_t offset; /* The order number for this BPO reloc, corresponding to the order in which BPO relocs were found. Used to create an index after reloc requests are sorted. */ size_t bpo_reloc_no; /* Set when the value is computed. Better than coding "guard values" into the other members. Is FALSE only for BPO relocs in a GC:ed section. */ bool valid; }; /* We attach this as mmix_elf_section_data (sec)->bpo in the linker-allocated greg contents section (MMIX_LD_ALLOCATED_REG_CONTENTS_SECTION_NAME), which is linked into the register contents section (MMIX_REG_CONTENTS_SECTION_NAME). This section is created by the linker; using the same hook as for usual with BPO relocs does not collide. */ struct bpo_greg_section_info { /* After GC, this reflects the number of remaining, non-excluded BPO-relocs. */ size_t n_bpo_relocs; /* This is the number of allocated bpo_reloc_requests; the size of sorted_indexes. Valid after the check.*relocs functions are called for all incoming sections. It includes the number of BPO relocs in sections that were GC:ed. */ size_t n_max_bpo_relocs; /* A counter used to find out when to fold the BPO gregs, since we don't have a single "after-relaxation" hook. */ size_t n_remaining_bpo_relocs_this_relaxation_round; /* The number of linker-allocated GREGs resulting from BPO relocs. This is an approximation after _bfd_mmix_before_linker_allocation and supposedly accurate after mmix_elf_relax_section is called for all incoming non-collected sections. */ size_t n_allocated_bpo_gregs; /* Index into reloc_request[], sorted on increasing "value", secondary by increasing index for strict sorting order. */ size_t *bpo_reloc_indexes; /* An array of all relocations, with the "value" member filled in by the relaxation function. */ struct bpo_reloc_request *reloc_request; }; extern bool mmix_elf_final_link (bfd *, struct bfd_link_info *); extern void mmix_elf_symbol_processing (bfd *, asymbol *); /* Only intended to be called from a debugger. */ extern void mmix_dump_bpo_gregs (struct bfd_link_info *, void (*) (const char *, ...)); static void mmix_set_relaxable_size (bfd *, asection *, void *); static bfd_reloc_status_type mmix_elf_reloc (bfd *, arelent *, asymbol *, void *, asection *, bfd *, char **); static bfd_reloc_status_type mmix_final_link_relocate (reloc_howto_type *, asection *, bfd_byte *, bfd_vma, bfd_signed_vma, bfd_vma, const char *, asection *, char **); /* Watch out: this currently needs to have elements with the same index as their R_MMIX_ number. */ static reloc_howto_type elf_mmix_howto_table[] = { /* This reloc does nothing. */ HOWTO (R_MMIX_NONE, /* type */ 0, /* rightshift */ 0, /* size */ 0, /* bitsize */ false, /* pc_relative */ 0, /* bitpos */ complain_overflow_dont, /* complain_on_overflow */ bfd_elf_generic_reloc, /* special_function */ "R_MMIX_NONE", /* name */ false, /* partial_inplace */ 0, /* src_mask */ 0, /* dst_mask */ false), /* pcrel_offset */ /* An 8 bit absolute relocation. */ HOWTO (R_MMIX_8, /* type */ 0, /* rightshift */ 1, /* size */ 8, /* bitsize */ false, /* pc_relative */ 0, /* bitpos */ complain_overflow_bitfield, /* complain_on_overflow */ bfd_elf_generic_reloc, /* special_function */ "R_MMIX_8", /* name */ false, /* partial_inplace */ 0, /* src_mask */ 0xff, /* dst_mask */ false), /* pcrel_offset */ /* An 16 bit absolute relocation. */ HOWTO (R_MMIX_16, /* type */ 0, /* rightshift */ 2, /* size */ 16, /* bitsize */ false, /* pc_relative */ 0, /* bitpos */ complain_overflow_bitfield, /* complain_on_overflow */ bfd_elf_generic_reloc, /* special_function */ "R_MMIX_16", /* name */ false, /* partial_inplace */ 0, /* src_mask */ 0xffff, /* dst_mask */ false), /* pcrel_offset */ /* An 24 bit absolute relocation. */ HOWTO (R_MMIX_24, /* type */ 0, /* rightshift */ 4, /* size */ 24, /* bitsize */ false, /* pc_relative */ 0, /* bitpos */ complain_overflow_bitfield, /* complain_on_overflow */ bfd_elf_generic_reloc, /* special_function */ "R_MMIX_24", /* name */ false, /* partial_inplace */ ~0xffffff, /* src_mask */ 0xffffff, /* dst_mask */ false), /* pcrel_offset */ /* A 32 bit absolute relocation. */ HOWTO (R_MMIX_32, /* type */ 0, /* rightshift */ 4, /* size */ 32, /* bitsize */ false, /* pc_relative */ 0, /* bitpos */ complain_overflow_bitfield, /* complain_on_overflow */ bfd_elf_generic_reloc, /* special_function */ "R_MMIX_32", /* name */ false, /* partial_inplace */ 0, /* src_mask */ 0xffffffff, /* dst_mask */ false), /* pcrel_offset */ /* 64 bit relocation. */ HOWTO (R_MMIX_64, /* type */ 0, /* rightshift */ 8, /* size */ 64, /* bitsize */ false, /* pc_relative */ 0, /* bitpos */ complain_overflow_bitfield, /* complain_on_overflow */ bfd_elf_generic_reloc, /* special_function */ "R_MMIX_64", /* name */ false, /* partial_inplace */ 0, /* src_mask */ MINUS_ONE, /* dst_mask */ false), /* pcrel_offset */ /* An 8 bit PC-relative relocation. */ HOWTO (R_MMIX_PC_8, /* type */ 0, /* rightshift */ 1, /* size */ 8, /* bitsize */ true, /* pc_relative */ 0, /* bitpos */ complain_overflow_bitfield, /* complain_on_overflow */ bfd_elf_generic_reloc, /* special_function */ "R_MMIX_PC_8", /* name */ false, /* partial_inplace */ 0, /* src_mask */ 0xff, /* dst_mask */ true), /* pcrel_offset */ /* An 16 bit PC-relative relocation. */ HOWTO (R_MMIX_PC_16, /* type */ 0, /* rightshift */ 2, /* size */ 16, /* bitsize */ true, /* pc_relative */ 0, /* bitpos */ complain_overflow_bitfield, /* complain_on_overflow */ bfd_elf_generic_reloc, /* special_function */ "R_MMIX_PC_16", /* name */ false, /* partial_inplace */ 0, /* src_mask */ 0xffff, /* dst_mask */ true), /* pcrel_offset */ /* An 24 bit PC-relative relocation. */ HOWTO (R_MMIX_PC_24, /* type */ 0, /* rightshift */ 4, /* size */ 24, /* bitsize */ true, /* pc_relative */ 0, /* bitpos */ complain_overflow_bitfield, /* complain_on_overflow */ bfd_elf_generic_reloc, /* special_function */ "R_MMIX_PC_24", /* name */ false, /* partial_inplace */ ~0xffffff, /* src_mask */ 0xffffff, /* dst_mask */ true), /* pcrel_offset */ /* A 32 bit absolute PC-relative relocation. */ HOWTO (R_MMIX_PC_32, /* type */ 0, /* rightshift */ 4, /* size */ 32, /* bitsize */ true, /* pc_relative */ 0, /* bitpos */ complain_overflow_bitfield, /* complain_on_overflow */ bfd_elf_generic_reloc, /* special_function */ "R_MMIX_PC_32", /* name */ false, /* partial_inplace */ 0, /* src_mask */ 0xffffffff, /* dst_mask */ true), /* pcrel_offset */ /* 64 bit PC-relative relocation. */ HOWTO (R_MMIX_PC_64, /* type */ 0, /* rightshift */ 8, /* size */ 64, /* bitsize */ true, /* pc_relative */ 0, /* bitpos */ complain_overflow_bitfield, /* complain_on_overflow */ bfd_elf_generic_reloc, /* special_function */ "R_MMIX_PC_64", /* name */ false, /* partial_inplace */ 0, /* src_mask */ MINUS_ONE, /* dst_mask */ true), /* pcrel_offset */ /* GNU extension to record C++ vtable hierarchy. */ HOWTO (R_MMIX_GNU_VTINHERIT, /* type */ 0, /* rightshift */ 0, /* size */ 0, /* bitsize */ false, /* pc_relative */ 0, /* bitpos */ complain_overflow_dont, /* complain_on_overflow */ NULL, /* special_function */ "R_MMIX_GNU_VTINHERIT", /* name */ false, /* partial_inplace */ 0, /* src_mask */ 0, /* dst_mask */ true), /* pcrel_offset */ /* GNU extension to record C++ vtable member usage. */ HOWTO (R_MMIX_GNU_VTENTRY, /* type */ 0, /* rightshift */ 0, /* size */ 0, /* bitsize */ false, /* pc_relative */ 0, /* bitpos */ complain_overflow_dont, /* complain_on_overflow */ _bfd_elf_rel_vtable_reloc_fn, /* special_function */ "R_MMIX_GNU_VTENTRY", /* name */ false, /* partial_inplace */ 0, /* src_mask */ 0, /* dst_mask */ false), /* pcrel_offset */ /* The GETA relocation is supposed to get any address that could possibly be reached by the GETA instruction. It can silently expand to get a 64-bit operand, but will complain if any of the two least significant bits are set. The howto members reflect a simple GETA. */ HOWTO (R_MMIX_GETA, /* type */ 2, /* rightshift */ 4, /* size */ 19, /* bitsize */ true, /* pc_relative */ 0, /* bitpos */ complain_overflow_signed, /* complain_on_overflow */ mmix_elf_reloc, /* special_function */ "R_MMIX_GETA", /* name */ false, /* partial_inplace */ ~0x0100ffff, /* src_mask */ 0x0100ffff, /* dst_mask */ true), /* pcrel_offset */ HOWTO (R_MMIX_GETA_1, /* type */ 2, /* rightshift */ 4, /* size */ 19, /* bitsize */ true, /* pc_relative */ 0, /* bitpos */ complain_overflow_signed, /* complain_on_overflow */ mmix_elf_reloc, /* special_function */ "R_MMIX_GETA_1", /* name */ false, /* partial_inplace */ ~0x0100ffff, /* src_mask */ 0x0100ffff, /* dst_mask */ true), /* pcrel_offset */ HOWTO (R_MMIX_GETA_2, /* type */ 2, /* rightshift */ 4, /* size */ 19, /* bitsize */ true, /* pc_relative */ 0, /* bitpos */ complain_overflow_signed, /* complain_on_overflow */ mmix_elf_reloc, /* special_function */ "R_MMIX_GETA_2", /* name */ false, /* partial_inplace */ ~0x0100ffff, /* src_mask */ 0x0100ffff, /* dst_mask */ true), /* pcrel_offset */ HOWTO (R_MMIX_GETA_3, /* type */ 2, /* rightshift */ 4, /* size */ 19, /* bitsize */ true, /* pc_relative */ 0, /* bitpos */ complain_overflow_signed, /* complain_on_overflow */ mmix_elf_reloc, /* special_function */ "R_MMIX_GETA_3", /* name */ false, /* partial_inplace */ ~0x0100ffff, /* src_mask */ 0x0100ffff, /* dst_mask */ true), /* pcrel_offset */ /* The conditional branches are supposed to reach any (code) address. It can silently expand to a 64-bit operand, but will emit an error if any of the two least significant bits are set. The howto members reflect a simple branch. */ HOWTO (R_MMIX_CBRANCH, /* type */ 2, /* rightshift */ 4, /* size */ 19, /* bitsize */ true, /* pc_relative */ 0, /* bitpos */ complain_overflow_signed, /* complain_on_overflow */ mmix_elf_reloc, /* special_function */ "R_MMIX_CBRANCH", /* name */ false, /* partial_inplace */ ~0x0100ffff, /* src_mask */ 0x0100ffff, /* dst_mask */ true), /* pcrel_offset */ HOWTO (R_MMIX_CBRANCH_J, /* type */ 2, /* rightshift */ 4, /* size */ 19, /* bitsize */ true, /* pc_relative */ 0, /* bitpos */ complain_overflow_signed, /* complain_on_overflow */ mmix_elf_reloc, /* special_function */ "R_MMIX_CBRANCH_J", /* name */ false, /* partial_inplace */ ~0x0100ffff, /* src_mask */ 0x0100ffff, /* dst_mask */ true), /* pcrel_offset */ HOWTO (R_MMIX_CBRANCH_1, /* type */ 2, /* rightshift */ 4, /* size */ 19, /* bitsize */ true, /* pc_relative */ 0, /* bitpos */ complain_overflow_signed, /* complain_on_overflow */ mmix_elf_reloc, /* special_function */ "R_MMIX_CBRANCH_1", /* name */ false, /* partial_inplace */ ~0x0100ffff, /* src_mask */ 0x0100ffff, /* dst_mask */ true), /* pcrel_offset */ HOWTO (R_MMIX_CBRANCH_2, /* type */ 2, /* rightshift */ 4, /* size */ 19, /* bitsize */ true, /* pc_relative */ 0, /* bitpos */ complain_overflow_signed, /* complain_on_overflow */ mmix_elf_reloc, /* special_function */ "R_MMIX_CBRANCH_2", /* name */ false, /* partial_inplace */ ~0x0100ffff, /* src_mask */ 0x0100ffff, /* dst_mask */ true), /* pcrel_offset */ HOWTO (R_MMIX_CBRANCH_3, /* type */ 2, /* rightshift */ 4, /* size */ 19, /* bitsize */ true, /* pc_relative */ 0, /* bitpos */ complain_overflow_signed, /* complain_on_overflow */ mmix_elf_reloc, /* special_function */ "R_MMIX_CBRANCH_3", /* name */ false, /* partial_inplace */ ~0x0100ffff, /* src_mask */ 0x0100ffff, /* dst_mask */ true), /* pcrel_offset */ /* The PUSHJ instruction can reach any (code) address, as long as it's the beginning of a function (no usable restriction). It can silently expand to a 64-bit operand, but will emit an error if any of the two least significant bits are set. It can also expand into a call to a stub; see R_MMIX_PUSHJ_STUBBABLE. The howto members reflect a simple PUSHJ. */ HOWTO (R_MMIX_PUSHJ, /* type */ 2, /* rightshift */ 4, /* size */ 19, /* bitsize */ true, /* pc_relative */ 0, /* bitpos */ complain_overflow_signed, /* complain_on_overflow */ mmix_elf_reloc, /* special_function */ "R_MMIX_PUSHJ", /* name */ false, /* partial_inplace */ ~0x0100ffff, /* src_mask */ 0x0100ffff, /* dst_mask */ true), /* pcrel_offset */ HOWTO (R_MMIX_PUSHJ_1, /* type */ 2, /* rightshift */ 4, /* size */ 19, /* bitsize */ true, /* pc_relative */ 0, /* bitpos */ complain_overflow_signed, /* complain_on_overflow */ mmix_elf_reloc, /* special_function */ "R_MMIX_PUSHJ_1", /* name */ false, /* partial_inplace */ ~0x0100ffff, /* src_mask */ 0x0100ffff, /* dst_mask */ true), /* pcrel_offset */ HOWTO (R_MMIX_PUSHJ_2, /* type */ 2, /* rightshift */ 4, /* size */ 19, /* bitsize */ true, /* pc_relative */ 0, /* bitpos */ complain_overflow_signed, /* complain_on_overflow */ mmix_elf_reloc, /* special_function */ "R_MMIX_PUSHJ_2", /* name */ false, /* partial_inplace */ ~0x0100ffff, /* src_mask */ 0x0100ffff, /* dst_mask */ true), /* pcrel_offset */ HOWTO (R_MMIX_PUSHJ_3, /* type */ 2, /* rightshift */ 4, /* size */ 19, /* bitsize */ true, /* pc_relative */ 0, /* bitpos */ complain_overflow_signed, /* complain_on_overflow */ mmix_elf_reloc, /* special_function */ "R_MMIX_PUSHJ_3", /* name */ false, /* partial_inplace */ ~0x0100ffff, /* src_mask */ 0x0100ffff, /* dst_mask */ true), /* pcrel_offset */ /* A JMP is supposed to reach any (code) address. By itself, it can reach +-64M; the expansion can reach all 64 bits. Note that the 64M limit is soon reached if you link the program in wildly different memory segments. The howto members reflect a trivial JMP. */ HOWTO (R_MMIX_JMP, /* type */ 2, /* rightshift */ 4, /* size */ 27, /* bitsize */ true, /* pc_relative */ 0, /* bitpos */ complain_overflow_signed, /* complain_on_overflow */ mmix_elf_reloc, /* special_function */ "R_MMIX_JMP", /* name */ false, /* partial_inplace */ ~0x1ffffff, /* src_mask */ 0x1ffffff, /* dst_mask */ true), /* pcrel_offset */ HOWTO (R_MMIX_JMP_1, /* type */ 2, /* rightshift */ 4, /* size */ 27, /* bitsize */ true, /* pc_relative */ 0, /* bitpos */ complain_overflow_signed, /* complain_on_overflow */ mmix_elf_reloc, /* special_function */ "R_MMIX_JMP_1", /* name */ false, /* partial_inplace */ ~0x1ffffff, /* src_mask */ 0x1ffffff, /* dst_mask */ true), /* pcrel_offset */ HOWTO (R_MMIX_JMP_2, /* type */ 2, /* rightshift */ 4, /* size */ 27, /* bitsize */ true, /* pc_relative */ 0, /* bitpos */ complain_overflow_signed, /* complain_on_overflow */ mmix_elf_reloc, /* special_function */ "R_MMIX_JMP_2", /* name */ false, /* partial_inplace */ ~0x1ffffff, /* src_mask */ 0x1ffffff, /* dst_mask */ true), /* pcrel_offset */ HOWTO (R_MMIX_JMP_3, /* type */ 2, /* rightshift */ 4, /* size */ 27, /* bitsize */ true, /* pc_relative */ 0, /* bitpos */ complain_overflow_signed, /* complain_on_overflow */ mmix_elf_reloc, /* special_function */ "R_MMIX_JMP_3", /* name */ false, /* partial_inplace */ ~0x1ffffff, /* src_mask */ 0x1ffffff, /* dst_mask */ true), /* pcrel_offset */ /* When we don't emit link-time-relaxable code from the assembler, or when relaxation has done all it can do, these relocs are used. For GETA/PUSHJ/branches. */ HOWTO (R_MMIX_ADDR19, /* type */ 2, /* rightshift */ 4, /* size */ 19, /* bitsize */ true, /* pc_relative */ 0, /* bitpos */ complain_overflow_signed, /* complain_on_overflow */ mmix_elf_reloc, /* special_function */ "R_MMIX_ADDR19", /* name */ false, /* partial_inplace */ ~0x0100ffff, /* src_mask */ 0x0100ffff, /* dst_mask */ true), /* pcrel_offset */ /* For JMP. */ HOWTO (R_MMIX_ADDR27, /* type */ 2, /* rightshift */ 4, /* size */ 27, /* bitsize */ true, /* pc_relative */ 0, /* bitpos */ complain_overflow_signed, /* complain_on_overflow */ mmix_elf_reloc, /* special_function */ "R_MMIX_ADDR27", /* name */ false, /* partial_inplace */ ~0x1ffffff, /* src_mask */ 0x1ffffff, /* dst_mask */ true), /* pcrel_offset */ /* A general register or the value 0..255. If a value, then the instruction (offset -3) needs adjusting. */ HOWTO (R_MMIX_REG_OR_BYTE, /* type */ 0, /* rightshift */ 2, /* size */ 8, /* bitsize */ false, /* pc_relative */ 0, /* bitpos */ complain_overflow_bitfield, /* complain_on_overflow */ mmix_elf_reloc, /* special_function */ "R_MMIX_REG_OR_BYTE", /* name */ false, /* partial_inplace */ 0, /* src_mask */ 0xff, /* dst_mask */ false), /* pcrel_offset */ /* A general register. */ HOWTO (R_MMIX_REG, /* type */ 0, /* rightshift */ 2, /* size */ 8, /* bitsize */ false, /* pc_relative */ 0, /* bitpos */ complain_overflow_bitfield, /* complain_on_overflow */ mmix_elf_reloc, /* special_function */ "R_MMIX_REG", /* name */ false, /* partial_inplace */ 0, /* src_mask */ 0xff, /* dst_mask */ false), /* pcrel_offset */ /* A register plus an index, corresponding to the relocation expression. The sizes must correspond to the valid range of the expression, while the bitmasks correspond to what we store in the image. */ HOWTO (R_MMIX_BASE_PLUS_OFFSET, /* type */ 0, /* rightshift */ 8, /* size */ 64, /* bitsize */ false, /* pc_relative */ 0, /* bitpos */ complain_overflow_bitfield, /* complain_on_overflow */ mmix_elf_reloc, /* special_function */ "R_MMIX_BASE_PLUS_OFFSET", /* name */ false, /* partial_inplace */ 0, /* src_mask */ 0xffff, /* dst_mask */ false), /* pcrel_offset */ /* A "magic" relocation for a LOCAL expression, asserting that the expression is less than the number of global registers. No actual modification of the contents is done. Implementing this as a relocation was less intrusive than e.g. putting such expressions in a section to discard *after* relocation. */ HOWTO (R_MMIX_LOCAL, /* type */ 0, /* rightshift */ 0, /* size */ 0, /* bitsize */ false, /* pc_relative */ 0, /* bitpos */ complain_overflow_dont, /* complain_on_overflow */ mmix_elf_reloc, /* special_function */ "R_MMIX_LOCAL", /* name */ false, /* partial_inplace */ 0, /* src_mask */ 0, /* dst_mask */ false), /* pcrel_offset */ HOWTO (R_MMIX_PUSHJ_STUBBABLE, /* type */ 2, /* rightshift */ 4, /* size */ 19, /* bitsize */ true, /* pc_relative */ 0, /* bitpos */ complain_overflow_signed, /* complain_on_overflow */ mmix_elf_reloc, /* special_function */ "R_MMIX_PUSHJ_STUBBABLE", /* name */ false, /* partial_inplace */ ~0x0100ffff, /* src_mask */ 0x0100ffff, /* dst_mask */ true) /* pcrel_offset */ }; /* Map BFD reloc types to MMIX ELF reloc types. */ struct mmix_reloc_map { bfd_reloc_code_real_type bfd_reloc_val; enum elf_mmix_reloc_type elf_reloc_val; }; static const struct mmix_reloc_map mmix_reloc_map[] = { {BFD_RELOC_NONE, R_MMIX_NONE}, {BFD_RELOC_8, R_MMIX_8}, {BFD_RELOC_16, R_MMIX_16}, {BFD_RELOC_24, R_MMIX_24}, {BFD_RELOC_32, R_MMIX_32}, {BFD_RELOC_64, R_MMIX_64}, {BFD_RELOC_8_PCREL, R_MMIX_PC_8}, {BFD_RELOC_16_PCREL, R_MMIX_PC_16}, {BFD_RELOC_24_PCREL, R_MMIX_PC_24}, {BFD_RELOC_32_PCREL, R_MMIX_PC_32}, {BFD_RELOC_64_PCREL, R_MMIX_PC_64}, {BFD_RELOC_VTABLE_INHERIT, R_MMIX_GNU_VTINHERIT}, {BFD_RELOC_VTABLE_ENTRY, R_MMIX_GNU_VTENTRY}, {BFD_RELOC_MMIX_GETA, R_MMIX_GETA}, {BFD_RELOC_MMIX_CBRANCH, R_MMIX_CBRANCH}, {BFD_RELOC_MMIX_PUSHJ, R_MMIX_PUSHJ}, {BFD_RELOC_MMIX_JMP, R_MMIX_JMP}, {BFD_RELOC_MMIX_ADDR19, R_MMIX_ADDR19}, {BFD_RELOC_MMIX_ADDR27, R_MMIX_ADDR27}, {BFD_RELOC_MMIX_REG_OR_BYTE, R_MMIX_REG_OR_BYTE}, {BFD_RELOC_MMIX_REG, R_MMIX_REG}, {BFD_RELOC_MMIX_BASE_PLUS_OFFSET, R_MMIX_BASE_PLUS_OFFSET}, {BFD_RELOC_MMIX_LOCAL, R_MMIX_LOCAL}, {BFD_RELOC_MMIX_PUSHJ_STUBBABLE, R_MMIX_PUSHJ_STUBBABLE} }; static reloc_howto_type * bfd_elf64_bfd_reloc_type_lookup (bfd *abfd ATTRIBUTE_UNUSED, bfd_reloc_code_real_type code) { unsigned int i; for (i = 0; i < sizeof (mmix_reloc_map) / sizeof (mmix_reloc_map[0]); i++) { if (mmix_reloc_map[i].bfd_reloc_val == code) return &elf_mmix_howto_table[mmix_reloc_map[i].elf_reloc_val]; } return NULL; } static reloc_howto_type * bfd_elf64_bfd_reloc_name_lookup (bfd *abfd ATTRIBUTE_UNUSED, const char *r_name) { unsigned int i; for (i = 0; i < sizeof (elf_mmix_howto_table) / sizeof (elf_mmix_howto_table[0]); i++) if (elf_mmix_howto_table[i].name != NULL && strcasecmp (elf_mmix_howto_table[i].name, r_name) == 0) return &elf_mmix_howto_table[i]; return NULL; } static bool mmix_elf_new_section_hook (bfd *abfd, asection *sec) { struct _mmix_elf_section_data *sdata; sdata = bfd_zalloc (abfd, sizeof (*sdata)); if (sdata == NULL) return false; sec->used_by_bfd = sdata; return _bfd_elf_new_section_hook (abfd, sec); } /* This function performs the actual bitfiddling and sanity check for a final relocation. Each relocation gets its *worst*-case expansion in size when it arrives here; any reduction in size should have been caught in linker relaxation earlier. When we get here, the relocation looks like the smallest instruction with SWYM:s (nop:s) appended to the max size. We fill in those nop:s. R_MMIX_GETA: (FIXME: Relaxation should break this up in 1, 2, 3 tetra) GETA $N,foo -> SETL $N,foo & 0xffff INCML $N,(foo >> 16) & 0xffff INCMH $N,(foo >> 32) & 0xffff INCH $N,(foo >> 48) & 0xffff R_MMIX_CBRANCH: (FIXME: Relaxation should break this up, but condbranches needing relaxation might be rare enough to not be worthwhile.) [P]Bcc $N,foo -> [~P]B~cc $N,.+20 SETL $255,foo & ... INCML ... INCMH ... INCH ... GO $255,$255,0 R_MMIX_PUSHJ: (FIXME: Relaxation...) PUSHJ $N,foo -> SETL $255,foo & ... INCML ... INCMH ... INCH ... PUSHGO $N,$255,0 R_MMIX_JMP: (FIXME: Relaxation...) JMP foo -> SETL $255,foo & ... INCML ... INCMH ... INCH ... GO $255,$255,0 R_MMIX_ADDR19 and R_MMIX_ADDR27 are just filled in. */ static bfd_reloc_status_type mmix_elf_perform_relocation (asection *isec, reloc_howto_type *howto, void *datap, bfd_vma addr, bfd_vma value, char **error_message) { bfd *abfd = isec->owner; bfd_reloc_status_type flag = bfd_reloc_ok; bfd_reloc_status_type r; int offs = 0; int reg = 255; /* The worst case bits are all similar SETL/INCML/INCMH/INCH sequences. We handle the differences here and the common sequence later. */ switch (howto->type) { case R_MMIX_GETA: offs = 0; reg = bfd_get_8 (abfd, (bfd_byte *) datap + 1); /* We change to an absolute value. */ value += addr; break; case R_MMIX_CBRANCH: { int in1 = bfd_get_16 (abfd, (bfd_byte *) datap) << 16; /* Invert the condition and prediction bit, and set the offset to five instructions ahead. We *can* do better if we want to. If the branch is found to be within limits, we could leave the branch as is; there'll just be a bunch of NOP:s after it. But we shouldn't see this sequence often enough that it's worth doing it. */ bfd_put_32 (abfd, (((in1 ^ ((PRED_INV_BIT | COND_INV_BIT) << 24)) & ~0xffff) | (24/4)), (bfd_byte *) datap); /* Put a "GO $255,$255,0" after the common sequence. */ bfd_put_32 (abfd, ((GO_INSN_BYTE | IMM_OFFSET_BIT) << 24) | 0xffff00, (bfd_byte *) datap + 20); /* Common sequence starts at offset 4. */ offs = 4; /* We change to an absolute value. */ value += addr; } break; case R_MMIX_PUSHJ_STUBBABLE: /* If the address fits, we're fine. */ if ((value & 3) == 0 /* Note rightshift 0; see R_MMIX_JMP case below. */ && (r = bfd_check_overflow (complain_overflow_signed, howto->bitsize, 0, bfd_arch_bits_per_address (abfd), value)) == bfd_reloc_ok) goto pcrel_mmix_reloc_fits; else { bfd_size_type size = isec->rawsize ? isec->rawsize : isec->size; /* We have the bytes at the PUSHJ insn and need to get the position for the stub. There's supposed to be room allocated for the stub. */ bfd_byte *stubcontents = ((bfd_byte *) datap - (addr - (isec->output_section->vma + isec->output_offset)) + size + mmix_elf_section_data (isec)->pjs.stub_offset); bfd_vma stubaddr; if (mmix_elf_section_data (isec)->pjs.n_pushj_relocs == 0) { /* This shouldn't happen when linking to ELF or mmo, so this is an attempt to link to "binary", right? We can't access the output bfd, so we can't verify that assumption. We only know that the critical mmix_elf_check_common_relocs has not been called, which happens when the output format is different from the input format (and is not mmo). */ if (! mmix_elf_section_data (isec)->has_warned_pushj) { /* For the first such error per input section, produce a verbose message. */ *error_message = _("invalid input relocation when producing" " non-ELF, non-mmo format output;" " please use the objcopy program to convert from" " ELF or mmo," " or assemble using" " \"-no-expand\" (for gcc, \"-Wa,-no-expand\""); mmix_elf_section_data (isec)->has_warned_pushj = true; return bfd_reloc_dangerous; } /* For subsequent errors, return this one, which is rate-limited but looks a little bit different, hopefully without affecting user-friendliness. */ return bfd_reloc_overflow; } /* The address doesn't fit, so redirect the PUSHJ to the location of the stub. */ r = mmix_elf_perform_relocation (isec, &elf_mmix_howto_table [R_MMIX_ADDR19], datap, addr, isec->output_section->vma + isec->output_offset + size + (mmix_elf_section_data (isec) ->pjs.stub_offset) - addr, error_message); if (r != bfd_reloc_ok) return r; stubaddr = (isec->output_section->vma + isec->output_offset + size + mmix_elf_section_data (isec)->pjs.stub_offset); /* We generate a simple JMP if that suffices, else the whole 5 insn stub. */ if (bfd_check_overflow (complain_overflow_signed, elf_mmix_howto_table[R_MMIX_ADDR27].bitsize, 0, bfd_arch_bits_per_address (abfd), addr + value - stubaddr) == bfd_reloc_ok) { bfd_put_32 (abfd, JMP_INSN_BYTE << 24, stubcontents); r = mmix_elf_perform_relocation (isec, &elf_mmix_howto_table [R_MMIX_ADDR27], stubcontents, stubaddr, value + addr - stubaddr, error_message); mmix_elf_section_data (isec)->pjs.stub_offset += 4; if (size + mmix_elf_section_data (isec)->pjs.stub_offset > isec->size) abort (); return r; } else { /* Put a "GO $255,0" after the common sequence. */ bfd_put_32 (abfd, ((GO_INSN_BYTE | IMM_OFFSET_BIT) << 24) | 0xff00, (bfd_byte *) stubcontents + 16); /* Prepare for the general code to set the first part of the linker stub, and */ value += addr; datap = stubcontents; mmix_elf_section_data (isec)->pjs.stub_offset += MAX_PUSHJ_STUB_SIZE; } } break; case R_MMIX_PUSHJ: { int inreg = bfd_get_8 (abfd, (bfd_byte *) datap + 1); /* Put a "PUSHGO $N,$255,0" after the common sequence. */ bfd_put_32 (abfd, ((PUSHGO_INSN_BYTE | IMM_OFFSET_BIT) << 24) | (inreg << 16) | 0xff00, (bfd_byte *) datap + 16); /* We change to an absolute value. */ value += addr; } break; case R_MMIX_JMP: /* This one is a little special. If we get here on a non-relaxing link, and the destination is actually in range, we don't need to execute the nops. If so, we fall through to the bit-fiddling relocs. FIXME: bfd_check_overflow seems broken; the relocation is rightshifted before testing, so supply a zero rightshift. */ if (! ((value & 3) == 0 && (r = bfd_check_overflow (complain_overflow_signed, howto->bitsize, 0, bfd_arch_bits_per_address (abfd), value)) == bfd_reloc_ok)) { /* If the relocation doesn't fit in a JMP, we let the NOP:s be modified below, and put a "GO $255,$255,0" after the address-loading sequence. */ bfd_put_32 (abfd, ((GO_INSN_BYTE | IMM_OFFSET_BIT) << 24) | 0xffff00, (bfd_byte *) datap + 16); /* We change to an absolute value. */ value += addr; break; } /* FALLTHROUGH. */ case R_MMIX_ADDR19: case R_MMIX_ADDR27: pcrel_mmix_reloc_fits: /* These must be in range, or else we emit an error. */ if ((value & 3) == 0 /* Note rightshift 0; see above. */ && (r = bfd_check_overflow (complain_overflow_signed, howto->bitsize, 0, bfd_arch_bits_per_address (abfd), value)) == bfd_reloc_ok) { bfd_vma in1 = bfd_get_32 (abfd, (bfd_byte *) datap); bfd_vma highbit; if ((bfd_signed_vma) value < 0) { highbit = 1 << 24; value += (1 << (howto->bitsize - 1)); } else highbit = 0; value >>= 2; bfd_put_32 (abfd, (in1 & howto->src_mask) | highbit | (value & howto->dst_mask), (bfd_byte *) datap); return bfd_reloc_ok; } else return bfd_reloc_overflow; case R_MMIX_BASE_PLUS_OFFSET: { struct bpo_reloc_section_info *bpodata = mmix_elf_section_data (isec)->bpo.reloc; asection *bpo_greg_section; struct bpo_greg_section_info *gregdata; size_t bpo_index; if (bpodata == NULL) { /* This shouldn't happen when linking to ELF or mmo, so this is an attempt to link to "binary", right? We can't access the output bfd, so we can't verify that assumption. We only know that the critical mmix_elf_check_common_relocs has not been called, which happens when the output format is different from the input format (and is not mmo). */ if (! mmix_elf_section_data (isec)->has_warned_bpo) { /* For the first such error per input section, produce a verbose message. */ *error_message = _("invalid input relocation when producing" " non-ELF, non-mmo format output;" " please use the objcopy program to convert from" " ELF or mmo," " or compile using the gcc-option" " \"-mno-base-addresses\"."); mmix_elf_section_data (isec)->has_warned_bpo = true; return bfd_reloc_dangerous; } /* For subsequent errors, return this one, which is rate-limited but looks a little bit different, hopefully without affecting user-friendliness. */ return bfd_reloc_overflow; } bpo_greg_section = bpodata->bpo_greg_section; gregdata = mmix_elf_section_data (bpo_greg_section)->bpo.greg; bpo_index = gregdata->bpo_reloc_indexes[bpodata->bpo_index++]; /* A consistency check: The value we now have in "relocation" must be the same as the value we stored for that relocation. It doesn't cost much, so can be left in at all times. */ if (value != gregdata->reloc_request[bpo_index].value) { _bfd_error_handler /* xgettext:c-format */ (_("%pB: Internal inconsistency error for value for\n\ linker-allocated global register: linked: %#" PRIx64 " != relaxed: %#" PRIx64 ""), isec->owner, (uint64_t) value, (uint64_t) gregdata->reloc_request[bpo_index].value); bfd_set_error (bfd_error_bad_value); return bfd_reloc_overflow; } /* Then store the register number and offset for that register into datap and datap + 1 respectively. */ bfd_put_8 (abfd, gregdata->reloc_request[bpo_index].regindex + bpo_greg_section->output_section->vma / 8, datap); bfd_put_8 (abfd, gregdata->reloc_request[bpo_index].offset, ((unsigned char *) datap) + 1); return bfd_reloc_ok; } case R_MMIX_REG_OR_BYTE: case R_MMIX_REG: if (value > 255) return bfd_reloc_overflow; bfd_put_8 (abfd, value, datap); return bfd_reloc_ok; default: BAD_CASE (howto->type); } /* This code adds the common SETL/INCML/INCMH/INCH worst-case sequence. */ /* Lowest two bits must be 0. We return bfd_reloc_overflow for everything that looks strange. */ if (value & 3) flag = bfd_reloc_overflow; bfd_put_32 (abfd, (SETL_INSN_BYTE << 24) | (value & 0xffff) | (reg << 16), (bfd_byte *) datap + offs); bfd_put_32 (abfd, (INCML_INSN_BYTE << 24) | ((value >> 16) & 0xffff) | (reg << 16), (bfd_byte *) datap + offs + 4); bfd_put_32 (abfd, (INCMH_INSN_BYTE << 24) | ((value >> 32) & 0xffff) | (reg << 16), (bfd_byte *) datap + offs + 8); bfd_put_32 (abfd, (INCH_INSN_BYTE << 24) | ((value >> 48) & 0xffff) | (reg << 16), (bfd_byte *) datap + offs + 12); return flag; } /* Set the howto pointer for an MMIX ELF reloc (type RELA). */ static bool mmix_info_to_howto_rela (bfd *abfd, arelent *cache_ptr, Elf_Internal_Rela *dst) { unsigned int r_type; r_type = ELF64_R_TYPE (dst->r_info); if (r_type >= (unsigned int) R_MMIX_max) { /* xgettext:c-format */ _bfd_error_handler (_("%pB: unsupported relocation type %#x"), abfd, r_type); bfd_set_error (bfd_error_bad_value); return false; } cache_ptr->howto = &elf_mmix_howto_table[r_type]; return true; } /* Any MMIX-specific relocation gets here at assembly time or when linking to other formats (such as mmo); this is the relocation function from the reloc_table. We don't get here for final pure ELF linking. */ static bfd_reloc_status_type mmix_elf_reloc (bfd *abfd, arelent *reloc_entry, asymbol *symbol, void * data, asection *input_section, bfd *output_bfd, char **error_message) { bfd_vma relocation; bfd_reloc_status_type r; asection *reloc_target_output_section; bfd_reloc_status_type flag = bfd_reloc_ok; bfd_vma output_base = 0; r = bfd_elf_generic_reloc (abfd, reloc_entry, symbol, data, input_section, output_bfd, error_message); /* If that was all that was needed (i.e. this isn't a final link, only some segment adjustments), we're done. */ if (r != bfd_reloc_continue) return r; if (bfd_is_und_section (symbol->section) && (symbol->flags & BSF_WEAK) == 0 && output_bfd == (bfd *) NULL) return bfd_reloc_undefined; /* Is the address of the relocation really within the section? */ if (reloc_entry->address > bfd_get_section_limit (abfd, input_section)) return bfd_reloc_outofrange; /* Work out which section the relocation is targeted at and the initial relocation command value. */ /* Get symbol value. (Common symbols are special.) */ if (bfd_is_com_section (symbol->section)) relocation = 0; else relocation = symbol->value; reloc_target_output_section = bfd_asymbol_section (symbol)->output_section; /* Here the variable relocation holds the final address of the symbol we are relocating against, plus any addend. */ if (output_bfd) output_base = 0; else output_base = reloc_target_output_section->vma; relocation += output_base + symbol->section->output_offset; if (output_bfd != (bfd *) NULL) { /* Add in supplied addend. */ relocation += reloc_entry->addend; /* This is a partial relocation, and we want to apply the relocation to the reloc entry rather than the raw data. Modify the reloc inplace to reflect what we now know. */ reloc_entry->addend = relocation; reloc_entry->address += input_section->output_offset; return flag; } return mmix_final_link_relocate (reloc_entry->howto, input_section, data, reloc_entry->address, reloc_entry->addend, relocation, bfd_asymbol_name (symbol), reloc_target_output_section, error_message); } /* Relocate an MMIX ELF section. Modified from elf32-fr30.c; look to it for guidance if you're thinking of copying this. */ static int mmix_elf_relocate_section (bfd *output_bfd ATTRIBUTE_UNUSED, struct bfd_link_info *info, bfd *input_bfd, asection *input_section, bfd_byte *contents, Elf_Internal_Rela *relocs, Elf_Internal_Sym *local_syms, asection **local_sections) { Elf_Internal_Shdr *symtab_hdr; struct elf_link_hash_entry **sym_hashes; Elf_Internal_Rela *rel; Elf_Internal_Rela *relend; bfd_size_type size; size_t pjsno = 0; size = input_section->rawsize ? input_section->rawsize : input_section->size; symtab_hdr = &elf_tdata (input_bfd)->symtab_hdr; sym_hashes = elf_sym_hashes (input_bfd); relend = relocs + input_section->reloc_count; /* Zero the stub area before we start. */ if (input_section->rawsize != 0 && input_section->size > input_section->rawsize) memset (contents + input_section->rawsize, 0, input_section->size - input_section->rawsize); for (rel = relocs; rel < relend; rel ++) { reloc_howto_type *howto; unsigned long r_symndx; Elf_Internal_Sym *sym; asection *sec; struct elf_link_hash_entry *h; bfd_vma relocation; bfd_reloc_status_type r; const char *name = NULL; int r_type; bool undefined_signalled = false; r_type = ELF64_R_TYPE (rel->r_info); if (r_type == R_MMIX_GNU_VTINHERIT || r_type == R_MMIX_GNU_VTENTRY) continue; r_symndx = ELF64_R_SYM (rel->r_info); howto = elf_mmix_howto_table + ELF64_R_TYPE (rel->r_info); h = NULL; sym = NULL; sec = NULL; if (r_symndx < symtab_hdr->sh_info) { sym = local_syms + r_symndx; sec = local_sections [r_symndx]; relocation = _bfd_elf_rela_local_sym (output_bfd, sym, &sec, rel); name = bfd_elf_string_from_elf_section (input_bfd, symtab_hdr->sh_link, sym->st_name); if (name == NULL) name = bfd_section_name (sec); } else { bool unresolved_reloc, ignored; RELOC_FOR_GLOBAL_SYMBOL (info, input_bfd, input_section, rel, r_symndx, symtab_hdr, sym_hashes, h, sec, relocation, unresolved_reloc, undefined_signalled, ignored); name = h->root.root.string; } if (sec != NULL && discarded_section (sec)) RELOC_AGAINST_DISCARDED_SECTION (info, input_bfd, input_section, rel, 1, relend, howto, 0, contents); if (bfd_link_relocatable (info)) { /* This is a relocatable link. For most relocs we don't have to change anything, unless the reloc is against a section symbol, in which case we have to adjust according to where the section symbol winds up in the output section. */ if (sym != NULL && ELF_ST_TYPE (sym->st_info) == STT_SECTION) rel->r_addend += sec->output_offset; /* For PUSHJ stub relocs however, we may need to change the reloc and the section contents, if the reloc doesn't reach beyond the end of the output section and previous stubs. Then we change the section contents to be a PUSHJ to the end of the input section plus stubs (we can do that without using a reloc), and then we change the reloc to be a R_MMIX_PUSHJ at the stub location. */ if (r_type == R_MMIX_PUSHJ_STUBBABLE) { /* We've already checked whether we need a stub; use that knowledge. */ if (mmix_elf_section_data (input_section)->pjs.stub_size[pjsno] != 0) { Elf_Internal_Rela relcpy; if (mmix_elf_section_data (input_section) ->pjs.stub_size[pjsno] != MAX_PUSHJ_STUB_SIZE) abort (); /* There's already a PUSHJ insn there, so just fill in the offset bits to the stub. */ if (mmix_final_link_relocate (elf_mmix_howto_table + R_MMIX_ADDR19, input_section, contents, rel->r_offset, 0, input_section ->output_section->vma + input_section->output_offset + size + mmix_elf_section_data (input_section) ->pjs.stub_offset, NULL, NULL, NULL) != bfd_reloc_ok) return false; /* Put a JMP insn at the stub; it goes with the R_MMIX_JMP reloc. */ bfd_put_32 (output_bfd, JMP_INSN_BYTE << 24, contents + size + mmix_elf_section_data (input_section) ->pjs.stub_offset); /* Change the reloc to be at the stub, and to a full R_MMIX_JMP reloc. */ rel->r_info = ELF64_R_INFO (r_symndx, R_MMIX_JMP); rel->r_offset = (size + mmix_elf_section_data (input_section) ->pjs.stub_offset); mmix_elf_section_data (input_section)->pjs.stub_offset += MAX_PUSHJ_STUB_SIZE; /* Shift this reloc to the end of the relocs to maintain the r_offset sorted reloc order. */ relcpy = *rel; memmove (rel, rel + 1, (char *) relend - (char *) rel); relend[-1] = relcpy; /* Back up one reloc, or else we'd skip the next reloc in turn. */ rel--; } pjsno++; } continue; } r = mmix_final_link_relocate (howto, input_section, contents, rel->r_offset, rel->r_addend, relocation, name, sec, NULL); if (r != bfd_reloc_ok) { const char * msg = (const char *) NULL; switch (r) { case bfd_reloc_overflow: info->callbacks->reloc_overflow (info, (h ? &h->root : NULL), name, howto->name, (bfd_vma) 0, input_bfd, input_section, rel->r_offset); break; case bfd_reloc_undefined: /* We may have sent this message above. */ if (! undefined_signalled) info->callbacks->undefined_symbol (info, name, input_bfd, input_section, rel->r_offset, true); undefined_signalled = true; break; case bfd_reloc_outofrange: msg = _("internal error: out of range error"); break; case bfd_reloc_notsupported: msg = _("internal error: unsupported relocation error"); break; case bfd_reloc_dangerous: msg = _("internal error: dangerous relocation"); break; default: msg = _("internal error: unknown error"); break; } if (msg) (*info->callbacks->warning) (info, msg, name, input_bfd, input_section, rel->r_offset); } } return true; } /* Perform a single relocation. By default we use the standard BFD routines. A few relocs we have to do ourselves. */ static bfd_reloc_status_type mmix_final_link_relocate (reloc_howto_type *howto, asection *input_section, bfd_byte *contents, bfd_vma r_offset, bfd_signed_vma r_addend, bfd_vma relocation, const char *symname, asection *symsec, char **error_message) { bfd_reloc_status_type r = bfd_reloc_ok; bfd_vma addr = (input_section->output_section->vma + input_section->output_offset + r_offset); bfd_signed_vma srel = (bfd_signed_vma) relocation + r_addend; switch (howto->type) { /* All these are PC-relative. */ case R_MMIX_PUSHJ_STUBBABLE: case R_MMIX_PUSHJ: case R_MMIX_CBRANCH: case R_MMIX_ADDR19: case R_MMIX_GETA: case R_MMIX_ADDR27: case R_MMIX_JMP: contents += r_offset; srel -= (input_section->output_section->vma + input_section->output_offset + r_offset); r = mmix_elf_perform_relocation (input_section, howto, contents, addr, srel, error_message); break; case R_MMIX_BASE_PLUS_OFFSET: if (symsec == NULL) return bfd_reloc_undefined; /* Check that we're not relocating against a register symbol. */ if (strcmp (bfd_section_name (symsec), MMIX_REG_CONTENTS_SECTION_NAME) == 0 || strcmp (bfd_section_name (symsec), MMIX_REG_SECTION_NAME) == 0) { /* Note: This is separated out into two messages in order to ease the translation into other languages. */ if (symname == NULL || *symname == 0) _bfd_error_handler /* xgettext:c-format */ (_("%pB: base-plus-offset relocation against register symbol:" " (unknown) in %pA"), input_section->owner, symsec); else _bfd_error_handler /* xgettext:c-format */ (_("%pB: base-plus-offset relocation against register symbol:" " %s in %pA"), input_section->owner, symname, symsec); return bfd_reloc_overflow; } goto do_mmix_reloc; case R_MMIX_REG_OR_BYTE: case R_MMIX_REG: /* For now, we handle these alike. They must refer to an register symbol, which is either relative to the register section and in the range 0..255, or is in the register contents section with vma regno * 8. */ /* FIXME: A better way to check for reg contents section? FIXME: Postpone section->scaling to mmix_elf_perform_relocation? */ if (symsec == NULL) return bfd_reloc_undefined; if (strcmp (bfd_section_name (symsec), MMIX_REG_CONTENTS_SECTION_NAME) == 0) { if ((srel & 7) != 0 || srel < 32*8 || srel > 255*8) { /* The bfd_reloc_outofrange return value, though intuitively a better value, will not get us an error. */ return bfd_reloc_overflow; } srel /= 8; } else if (strcmp (bfd_section_name (symsec), MMIX_REG_SECTION_NAME) == 0) { if (srel < 0 || srel > 255) /* The bfd_reloc_outofrange return value, though intuitively a better value, will not get us an error. */ return bfd_reloc_overflow; } else { /* Note: This is separated out into two messages in order to ease the translation into other languages. */ if (symname == NULL || *symname == 0) _bfd_error_handler /* xgettext:c-format */ (_("%pB: register relocation against non-register symbol:" " (unknown) in %pA"), input_section->owner, symsec); else _bfd_error_handler /* xgettext:c-format */ (_("%pB: register relocation against non-register symbol:" " %s in %pA"), input_section->owner, symname, symsec); /* The bfd_reloc_outofrange return value, though intuitively a better value, will not get us an error. */ return bfd_reloc_overflow; } do_mmix_reloc: contents += r_offset; r = mmix_elf_perform_relocation (input_section, howto, contents, addr, srel, error_message); break; case R_MMIX_LOCAL: /* This isn't a real relocation, it's just an assertion that the final relocation value corresponds to a local register. We ignore the actual relocation; nothing is changed. */ { asection *regsec = bfd_get_section_by_name (input_section->output_section->owner, MMIX_REG_CONTENTS_SECTION_NAME); bfd_vma first_global; /* Check that this is an absolute value, or a reference to the register contents section or the register (symbol) section. Absolute numbers can get here as undefined section. Undefined symbols are signalled elsewhere, so there's no conflict in us accidentally handling it. */ if (!bfd_is_abs_section (symsec) && !bfd_is_und_section (symsec) && strcmp (bfd_section_name (symsec), MMIX_REG_CONTENTS_SECTION_NAME) != 0 && strcmp (bfd_section_name (symsec), MMIX_REG_SECTION_NAME) != 0) { _bfd_error_handler (_("%pB: directive LOCAL valid only with a register or absolute value"), input_section->owner); return bfd_reloc_overflow; } /* If we don't have a register contents section, then $255 is the first global register. */ if (regsec == NULL) first_global = 255; else { first_global = bfd_section_vma (regsec) / 8; if (strcmp (bfd_section_name (symsec), MMIX_REG_CONTENTS_SECTION_NAME) == 0) { if ((srel & 7) != 0 || srel < 32*8 || srel > 255*8) /* The bfd_reloc_outofrange return value, though intuitively a better value, will not get us an error. */ return bfd_reloc_overflow; srel /= 8; } } if ((bfd_vma) srel >= first_global) { /* FIXME: Better error message. */ _bfd_error_handler /* xgettext:c-format */ (_("%pB: LOCAL directive: " "register $%" PRId64 " is not a local register;" " first global register is $%" PRId64), input_section->owner, (int64_t) srel, (int64_t) first_global); return bfd_reloc_overflow; } } r = bfd_reloc_ok; break; default: r = _bfd_final_link_relocate (howto, input_section->owner, input_section, contents, r_offset, relocation, r_addend); } return r; } /* Return the section that should be marked against GC for a given relocation. */ static asection * mmix_elf_gc_mark_hook (asection *sec, struct bfd_link_info *info, Elf_Internal_Rela *rel, struct elf_link_hash_entry *h, Elf_Internal_Sym *sym) { if (h != NULL) switch (ELF64_R_TYPE (rel->r_info)) { case R_MMIX_GNU_VTINHERIT: case R_MMIX_GNU_VTENTRY: return NULL; } return _bfd_elf_gc_mark_hook (sec, info, rel, h, sym); } /* Sort register relocs to come before expanding relocs. */ static int mmix_elf_sort_relocs (const void * p1, const void * p2) { const Elf_Internal_Rela *r1 = (const Elf_Internal_Rela *) p1; const Elf_Internal_Rela *r2 = (const Elf_Internal_Rela *) p2; int r1_is_reg, r2_is_reg; /* Sort primarily on r_offset & ~3, so relocs are done to consecutive insns. */ if ((r1->r_offset & ~(bfd_vma) 3) > (r2->r_offset & ~(bfd_vma) 3)) return 1; else if ((r1->r_offset & ~(bfd_vma) 3) < (r2->r_offset & ~(bfd_vma) 3)) return -1; r1_is_reg = (ELF64_R_TYPE (r1->r_info) == R_MMIX_REG_OR_BYTE || ELF64_R_TYPE (r1->r_info) == R_MMIX_REG); r2_is_reg = (ELF64_R_TYPE (r2->r_info) == R_MMIX_REG_OR_BYTE || ELF64_R_TYPE (r2->r_info) == R_MMIX_REG); if (r1_is_reg != r2_is_reg) return r2_is_reg - r1_is_reg; /* Neither or both are register relocs. Then sort on full offset. */ if (r1->r_offset > r2->r_offset) return 1; else if (r1->r_offset < r2->r_offset) return -1; return 0; } /* Subset of mmix_elf_check_relocs, common to ELF and mmo linking. */ static bool mmix_elf_check_common_relocs (bfd *abfd, struct bfd_link_info *info, asection *sec, const Elf_Internal_Rela *relocs) { bfd *bpo_greg_owner = NULL; asection *allocated_gregs_section = NULL; struct bpo_greg_section_info *gregdata = NULL; struct bpo_reloc_section_info *bpodata = NULL; const Elf_Internal_Rela *rel; const Elf_Internal_Rela *rel_end; /* We currently have to abuse this COFF-specific member, since there's no target-machine-dedicated member. There's no alternative outside the bfd_link_info struct; we can't specialize a hash-table since they're different between ELF and mmo. */ bpo_greg_owner = (bfd *) info->base_file; rel_end = relocs + sec->reloc_count; for (rel = relocs; rel < rel_end; rel++) { switch (ELF64_R_TYPE (rel->r_info)) { /* This relocation causes a GREG allocation. We need to count them, and we need to create a section for them, so we need an object to fake as the owner of that section. We can't use the ELF dynobj for this, since the ELF bits assume lots of DSO-related stuff if that member is non-NULL. */ case R_MMIX_BASE_PLUS_OFFSET: /* We don't do anything with this reloc for a relocatable link. */ if (bfd_link_relocatable (info)) break; if (bpo_greg_owner == NULL) { bpo_greg_owner = abfd; info->base_file = bpo_greg_owner; } if (allocated_gregs_section == NULL) allocated_gregs_section = bfd_get_section_by_name (bpo_greg_owner, MMIX_LD_ALLOCATED_REG_CONTENTS_SECTION_NAME); if (allocated_gregs_section == NULL) { allocated_gregs_section = bfd_make_section_with_flags (bpo_greg_owner, MMIX_LD_ALLOCATED_REG_CONTENTS_SECTION_NAME, (SEC_HAS_CONTENTS | SEC_IN_MEMORY | SEC_LINKER_CREATED)); /* Setting both SEC_ALLOC and SEC_LOAD means the section is treated like any other section, and we'd get errors for address overlap with the text section. Let's set none of those flags, as that is what currently happens for usual GREG allocations, and that works. */ if (allocated_gregs_section == NULL || !bfd_set_section_alignment (allocated_gregs_section, 3)) return false; gregdata = (struct bpo_greg_section_info *) bfd_zalloc (bpo_greg_owner, sizeof (struct bpo_greg_section_info)); if (gregdata == NULL) return false; mmix_elf_section_data (allocated_gregs_section)->bpo.greg = gregdata; } else if (gregdata == NULL) gregdata = mmix_elf_section_data (allocated_gregs_section)->bpo.greg; /* Get ourselves some auxiliary info for the BPO-relocs. */ if (bpodata == NULL) { /* No use doing a separate iteration pass to find the upper limit - just use the number of relocs. */ bpodata = (struct bpo_reloc_section_info *) bfd_alloc (bpo_greg_owner, sizeof (struct bpo_reloc_section_info) * (sec->reloc_count + 1)); if (bpodata == NULL) return false; mmix_elf_section_data (sec)->bpo.reloc = bpodata; bpodata->first_base_plus_offset_reloc = bpodata->bpo_index = gregdata->n_max_bpo_relocs; bpodata->bpo_greg_section = allocated_gregs_section; bpodata->n_bpo_relocs_this_section = 0; } bpodata->n_bpo_relocs_this_section++; gregdata->n_max_bpo_relocs++; /* We don't get another chance to set this before GC; we've not set up any hook that runs before GC. */ gregdata->n_bpo_relocs = gregdata->n_max_bpo_relocs; break; case R_MMIX_PUSHJ_STUBBABLE: mmix_elf_section_data (sec)->pjs.n_pushj_relocs++; break; } } /* Allocate per-reloc stub storage and initialize it to the max stub size. */ if (mmix_elf_section_data (sec)->pjs.n_pushj_relocs != 0) { size_t i; mmix_elf_section_data (sec)->pjs.stub_size = bfd_alloc (abfd, mmix_elf_section_data (sec)->pjs.n_pushj_relocs * sizeof (mmix_elf_section_data (sec) ->pjs.stub_size[0])); if (mmix_elf_section_data (sec)->pjs.stub_size == NULL) return false; for (i = 0; i < mmix_elf_section_data (sec)->pjs.n_pushj_relocs; i++) mmix_elf_section_data (sec)->pjs.stub_size[i] = MAX_PUSHJ_STUB_SIZE; } return true; } /* Look through the relocs for a section during the first phase. */ static bool mmix_elf_check_relocs (bfd *abfd, struct bfd_link_info *info, asection *sec, const Elf_Internal_Rela *relocs) { Elf_Internal_Shdr *symtab_hdr; struct elf_link_hash_entry **sym_hashes; const Elf_Internal_Rela *rel; const Elf_Internal_Rela *rel_end; symtab_hdr = &elf_tdata (abfd)->symtab_hdr; sym_hashes = elf_sym_hashes (abfd); /* First we sort the relocs so that any register relocs come before expansion-relocs to the same insn. FIXME: Not done for mmo. */ qsort ((void *) relocs, sec->reloc_count, sizeof (Elf_Internal_Rela), mmix_elf_sort_relocs); /* Do the common part. */ if (!mmix_elf_check_common_relocs (abfd, info, sec, relocs)) return false; if (bfd_link_relocatable (info)) return true; rel_end = relocs + sec->reloc_count; for (rel = relocs; rel < rel_end; rel++) { struct elf_link_hash_entry *h; unsigned long r_symndx; r_symndx = ELF64_R_SYM (rel->r_info); if (r_symndx < symtab_hdr->sh_info) h = NULL; else { h = sym_hashes[r_symndx - symtab_hdr->sh_info]; while (h->root.type == bfd_link_hash_indirect || h->root.type == bfd_link_hash_warning) h = (struct elf_link_hash_entry *) h->root.u.i.link; } switch (ELF64_R_TYPE (rel->r_info)) { /* This relocation describes the C++ object vtable hierarchy. Reconstruct it for later use during GC. */ case R_MMIX_GNU_VTINHERIT: if (!bfd_elf_gc_record_vtinherit (abfd, sec, h, rel->r_offset)) return false; break; /* This relocation describes which C++ vtable entries are actually used. Record for later use during GC. */ case R_MMIX_GNU_VTENTRY: if (!bfd_elf_gc_record_vtentry (abfd, sec, h, rel->r_addend)) return false; break; } } return true; } /* Wrapper for mmix_elf_check_common_relocs, called when linking to mmo. Copied from elf_link_add_object_symbols. */ bool _bfd_mmix_check_all_relocs (bfd *abfd, struct bfd_link_info *info) { asection *o; for (o = abfd->sections; o != NULL; o = o->next) { Elf_Internal_Rela *internal_relocs; bool ok; if ((o->flags & SEC_RELOC) == 0 || o->reloc_count == 0 || ((info->strip == strip_all || info->strip == strip_debugger) && (o->flags & SEC_DEBUGGING) != 0) || bfd_is_abs_section (o->output_section)) continue; internal_relocs = _bfd_elf_link_read_relocs (abfd, o, NULL, (Elf_Internal_Rela *) NULL, info->keep_memory); if (internal_relocs == NULL) return false; ok = mmix_elf_check_common_relocs (abfd, info, o, internal_relocs); if (! info->keep_memory) free (internal_relocs); if (! ok) return false; } return true; } /* Change symbols relative to the reg contents section to instead be to the register section, and scale them down to correspond to the register number. */ static int mmix_elf_link_output_symbol_hook (struct bfd_link_info *info ATTRIBUTE_UNUSED, const char *name ATTRIBUTE_UNUSED, Elf_Internal_Sym *sym, asection *input_sec, struct elf_link_hash_entry *h ATTRIBUTE_UNUSED) { if (input_sec != NULL && input_sec->name != NULL && ELF_ST_TYPE (sym->st_info) != STT_SECTION && strcmp (input_sec->name, MMIX_REG_CONTENTS_SECTION_NAME) == 0) { sym->st_value /= 8; sym->st_shndx = SHN_REGISTER; } return 1; } /* We fake a register section that holds values that are register numbers. Having a SHN_REGISTER and register section translates better to other formats (e.g. mmo) than for example a STT_REGISTER attribute. This section faking is based on a construct in elf32-mips.c. */ static asection mmix_elf_reg_section; static const asymbol mmix_elf_reg_section_symbol = GLOBAL_SYM_INIT (MMIX_REG_SECTION_NAME, &mmix_elf_reg_section); static asection mmix_elf_reg_section = BFD_FAKE_SECTION (mmix_elf_reg_section, &mmix_elf_reg_section_symbol, MMIX_REG_SECTION_NAME, 0, SEC_NO_FLAGS); /* Handle the special section numbers that a symbol may use. */ void mmix_elf_symbol_processing (bfd *abfd ATTRIBUTE_UNUSED, asymbol *asym) { elf_symbol_type *elfsym; elfsym = (elf_symbol_type *) asym; switch (elfsym->internal_elf_sym.st_shndx) { case SHN_REGISTER: asym->section = &mmix_elf_reg_section; break; default: break; } } /* Given a BFD section, try to locate the corresponding ELF section index. */ static bool mmix_elf_section_from_bfd_section (bfd * abfd ATTRIBUTE_UNUSED, asection * sec, int * retval) { if (strcmp (bfd_section_name (sec), MMIX_REG_SECTION_NAME) == 0) *retval = SHN_REGISTER; else return false; return true; } /* Hook called by the linker routine which adds symbols from an object file. We must handle the special SHN_REGISTER section number here. We also check that we only have *one* each of the section-start symbols, since otherwise having two with the same value would cause them to be "merged", but with the contents serialized. */ static bool mmix_elf_add_symbol_hook (bfd *abfd, struct bfd_link_info *info ATTRIBUTE_UNUSED, Elf_Internal_Sym *sym, const char **namep ATTRIBUTE_UNUSED, flagword *flagsp ATTRIBUTE_UNUSED, asection **secp, bfd_vma *valp ATTRIBUTE_UNUSED) { if (sym->st_shndx == SHN_REGISTER) { *secp = bfd_make_section_old_way (abfd, MMIX_REG_SECTION_NAME); (*secp)->flags |= SEC_LINKER_CREATED; } else if ((*namep)[0] == '_' && (*namep)[1] == '_' && (*namep)[2] == '.' && startswith (*namep, MMIX_LOC_SECTION_START_SYMBOL_PREFIX)) { /* See if we have another one. */ struct bfd_link_hash_entry *h = bfd_link_hash_lookup (info->hash, *namep, false, false, false); if (h != NULL && h->type != bfd_link_hash_undefined) { /* How do we get the asymbol (or really: the filename) from h? h->u.def.section->owner is NULL. */ _bfd_error_handler /* xgettext:c-format */ (_("%pB: error: multiple definition of `%s'; start of %s " "is set in a earlier linked file"), abfd, *namep, *namep + strlen (MMIX_LOC_SECTION_START_SYMBOL_PREFIX)); bfd_set_error (bfd_error_bad_value); return false; } } return true; } /* We consider symbols matching "L.*:[0-9]+" to be local symbols. */ static bool mmix_elf_is_local_label_name (bfd *abfd, const char *name) { const char *colpos; int digits; /* Also include the default local-label definition. */ if (_bfd_elf_is_local_label_name (abfd, name)) return true; if (*name != 'L') return false; /* If there's no ":", or more than one, it's not a local symbol. */ colpos = strchr (name, ':'); if (colpos == NULL || strchr (colpos + 1, ':') != NULL) return false; /* Check that there are remaining characters and that they are digits. */ if (colpos[1] == 0) return false; digits = strspn (colpos + 1, "0123456789"); return digits != 0 && colpos[1 + digits] == 0; } /* We get rid of the register section here. */ bool mmix_elf_final_link (bfd *abfd, struct bfd_link_info *info) { /* We never output a register section, though we create one for temporary measures. Check that nobody entered contents into it. */ asection *reg_section; reg_section = bfd_get_section_by_name (abfd, MMIX_REG_SECTION_NAME); if (reg_section != NULL) { /* FIXME: Pass error state gracefully. */ if (bfd_section_flags (reg_section) & SEC_HAS_CONTENTS) _bfd_abort (__FILE__, __LINE__, _("register section has contents\n")); /* Really remove the section, if it hasn't already been done. */ if (!bfd_section_removed_from_list (abfd, reg_section)) { bfd_section_list_remove (abfd, reg_section); --abfd->section_count; } } if (! bfd_elf_final_link (abfd, info)) return false; /* Since this section is marked SEC_LINKER_CREATED, it isn't output by the regular linker machinery. We do it here, like other targets with special sections. */ if (info->base_file != NULL) { asection *greg_section = bfd_get_section_by_name ((bfd *) info->base_file, MMIX_LD_ALLOCATED_REG_CONTENTS_SECTION_NAME); if (!bfd_set_section_contents (abfd, greg_section->output_section, greg_section->contents, (file_ptr) greg_section->output_offset, greg_section->size)) return false; } return true; } /* We need to include the maximum size of PUSHJ-stubs in the initial section size. This is expected to shrink during linker relaxation. */ static void mmix_set_relaxable_size (bfd *abfd ATTRIBUTE_UNUSED, asection *sec, void *ptr) { struct bfd_link_info *info = ptr; /* Make sure we only do this for section where we know we want this, otherwise we might end up resetting the size of COMMONs. */ if (mmix_elf_section_data (sec)->pjs.n_pushj_relocs == 0) return; sec->rawsize = sec->size; sec->size += (mmix_elf_section_data (sec)->pjs.n_pushj_relocs * MAX_PUSHJ_STUB_SIZE); /* For use in relocatable link, we start with a max stubs size. See mmix_elf_relax_section. */ if (bfd_link_relocatable (info) && sec->output_section) mmix_elf_section_data (sec->output_section)->pjs.stubs_size_sum += (mmix_elf_section_data (sec)->pjs.n_pushj_relocs * MAX_PUSHJ_STUB_SIZE); } /* Initialize stuff for the linker-generated GREGs to match R_MMIX_BASE_PLUS_OFFSET relocs seen by the linker. */ bool _bfd_mmix_before_linker_allocation (bfd *abfd ATTRIBUTE_UNUSED, struct bfd_link_info *info) { asection *bpo_gregs_section; bfd *bpo_greg_owner; struct bpo_greg_section_info *gregdata; size_t n_gregs; bfd_vma gregs_size; size_t i; size_t *bpo_reloc_indexes; bfd *ibfd; /* Set the initial size of sections. */ for (ibfd = info->input_bfds; ibfd != NULL; ibfd = ibfd->link.next) bfd_map_over_sections (ibfd, mmix_set_relaxable_size, info); /* The bpo_greg_owner bfd is supposed to have been set by mmix_elf_check_relocs when the first R_MMIX_BASE_PLUS_OFFSET is seen. If there is no such object, there was no R_MMIX_BASE_PLUS_OFFSET. */ bpo_greg_owner = (bfd *) info->base_file; if (bpo_greg_owner == NULL) return true; bpo_gregs_section = bfd_get_section_by_name (bpo_greg_owner, MMIX_LD_ALLOCATED_REG_CONTENTS_SECTION_NAME); if (bpo_gregs_section == NULL) return true; /* We use the target-data handle in the ELF section data. */ gregdata = mmix_elf_section_data (bpo_gregs_section)->bpo.greg; if (gregdata == NULL) return false; n_gregs = gregdata->n_bpo_relocs; gregdata->n_allocated_bpo_gregs = n_gregs; /* When this reaches zero during relaxation, all entries have been filled in and the size of the linker gregs can be calculated. */ gregdata->n_remaining_bpo_relocs_this_relaxation_round = n_gregs; /* Set the zeroth-order estimate for the GREGs size. */ gregs_size = n_gregs * 8; if (!bfd_set_section_size (bpo_gregs_section, gregs_size)) return false; /* Allocate and set up the GREG arrays. They're filled in at relaxation time. Note that we must use the max number ever noted for the array, since the index numbers were created before GC. */ gregdata->reloc_request = bfd_zalloc (bpo_greg_owner, sizeof (struct bpo_reloc_request) * gregdata->n_max_bpo_relocs); gregdata->bpo_reloc_indexes = bpo_reloc_indexes = bfd_alloc (bpo_greg_owner, gregdata->n_max_bpo_relocs * sizeof (size_t)); if (bpo_reloc_indexes == NULL) return false; /* The default order is an identity mapping. */ for (i = 0; i < gregdata->n_max_bpo_relocs; i++) { bpo_reloc_indexes[i] = i; gregdata->reloc_request[i].bpo_reloc_no = i; } return true; } /* Fill in contents in the linker allocated gregs. Everything is calculated at this point; we just move the contents into place here. */ bool _bfd_mmix_after_linker_allocation (bfd *abfd ATTRIBUTE_UNUSED, struct bfd_link_info *link_info) { asection *bpo_gregs_section; bfd *bpo_greg_owner; struct bpo_greg_section_info *gregdata; size_t n_gregs; size_t i, j; size_t lastreg; bfd_byte *contents; /* The bpo_greg_owner bfd is supposed to have been set by mmix_elf_check_relocs when the first R_MMIX_BASE_PLUS_OFFSET is seen. If there is no such object, there was no R_MMIX_BASE_PLUS_OFFSET. */ bpo_greg_owner = (bfd *) link_info->base_file; if (bpo_greg_owner == NULL) return true; bpo_gregs_section = bfd_get_section_by_name (bpo_greg_owner, MMIX_LD_ALLOCATED_REG_CONTENTS_SECTION_NAME); /* This can't happen without DSO handling. When DSOs are handled without any R_MMIX_BASE_PLUS_OFFSET seen, there will be no such section. */ if (bpo_gregs_section == NULL) return true; /* We use the target-data handle in the ELF section data. */ gregdata = mmix_elf_section_data (bpo_gregs_section)->bpo.greg; if (gregdata == NULL) return false; n_gregs = gregdata->n_allocated_bpo_gregs; bpo_gregs_section->contents = contents = bfd_alloc (bpo_greg_owner, bpo_gregs_section->size); if (contents == NULL) return false; /* Sanity check: If these numbers mismatch, some relocation has not been accounted for and the rest of gregdata is probably inconsistent. It's a bug, but it's more helpful to identify it than segfaulting below. */ if (gregdata->n_remaining_bpo_relocs_this_relaxation_round != gregdata->n_bpo_relocs) { _bfd_error_handler /* xgettext:c-format */ (_("internal inconsistency: remaining %lu != max %lu;" " please report this bug"), (unsigned long) gregdata->n_remaining_bpo_relocs_this_relaxation_round, (unsigned long) gregdata->n_bpo_relocs); return false; } for (lastreg = 255, i = 0, j = 0; j < n_gregs; i++) if (gregdata->reloc_request[i].regindex != lastreg) { bfd_put_64 (bpo_greg_owner, gregdata->reloc_request[i].value, contents + j * 8); lastreg = gregdata->reloc_request[i].regindex; j++; } return true; } /* Sort valid relocs to come before non-valid relocs, then on increasing value. */ static int bpo_reloc_request_sort_fn (const void * p1, const void * p2) { const struct bpo_reloc_request *r1 = (const struct bpo_reloc_request *) p1; const struct bpo_reloc_request *r2 = (const struct bpo_reloc_request *) p2; /* Primary function is validity; non-valid relocs sorted after valid ones. */ if (r1->valid != r2->valid) return r2->valid - r1->valid; /* Then sort on value. Don't simplify and return just the difference of the values: the upper bits of the 64-bit value would be truncated on a host with 32-bit ints. */ if (r1->value != r2->value) return r1->value > r2->value ? 1 : -1; /* As a last re-sort, use the relocation number, so we get a stable sort. The *addresses* aren't stable since items are swapped during sorting. It depends on the qsort implementation if this actually happens. */ return r1->bpo_reloc_no > r2->bpo_reloc_no ? 1 : (r1->bpo_reloc_no < r2->bpo_reloc_no ? -1 : 0); } /* For debug use only. Dumps the global register allocations resulting from base-plus-offset relocs. */ void mmix_dump_bpo_gregs (struct bfd_link_info *link_info, void (*pf) (const char *fmt, ...)) { bfd *bpo_greg_owner; asection *bpo_gregs_section; struct bpo_greg_section_info *gregdata; unsigned int i; if (link_info == NULL || link_info->base_file == NULL) return; bpo_greg_owner = (bfd *) link_info->base_file; bpo_gregs_section = bfd_get_section_by_name (bpo_greg_owner, MMIX_LD_ALLOCATED_REG_CONTENTS_SECTION_NAME); if (bpo_gregs_section == NULL) return; gregdata = mmix_elf_section_data (bpo_gregs_section)->bpo.greg; if (gregdata == NULL) return; if (pf == NULL) pf = _bfd_error_handler; /* These format strings are not translated. They are for debug purposes only and never displayed to an end user. Should they escape, we surely want them in original. */ (*pf) (" n_bpo_relocs: %u\n n_max_bpo_relocs: %u\n n_remain...round: %u\n\ n_allocated_bpo_gregs: %u\n", gregdata->n_bpo_relocs, gregdata->n_max_bpo_relocs, gregdata->n_remaining_bpo_relocs_this_relaxation_round, gregdata->n_allocated_bpo_gregs); if (gregdata->reloc_request) for (i = 0; i < gregdata->n_max_bpo_relocs; i++) (*pf) ("%4u (%4u)/%4u#%u: 0x%08lx%08lx r: %3u o: %3u\n", i, (gregdata->bpo_reloc_indexes != NULL ? gregdata->bpo_reloc_indexes[i] : (size_t) -1), gregdata->reloc_request[i].bpo_reloc_no, gregdata->reloc_request[i].valid, (unsigned long) (gregdata->reloc_request[i].value >> 32), (unsigned long) gregdata->reloc_request[i].value, gregdata->reloc_request[i].regindex, gregdata->reloc_request[i].offset); } /* This links all R_MMIX_BASE_PLUS_OFFSET relocs into a special array, and when the last such reloc is done, an index-array is sorted according to the values and iterated over to produce register numbers (indexed by 0 from the first allocated register number) and offsets for use in real relocation. (N.B.: Relocatable runs are handled, not just punted.) PUSHJ stub accounting is also done here. Symbol- and reloc-reading infrastructure copied from elf-m10200.c. */ static bool mmix_elf_relax_section (bfd *abfd, asection *sec, struct bfd_link_info *link_info, bool *again) { Elf_Internal_Shdr *symtab_hdr; Elf_Internal_Rela *internal_relocs; Elf_Internal_Rela *irel, *irelend; asection *bpo_gregs_section = NULL; struct bpo_greg_section_info *gregdata; struct bpo_reloc_section_info *bpodata = mmix_elf_section_data (sec)->bpo.reloc; /* The initialization is to quiet compiler warnings. The value is to spot a missing actual initialization. */ size_t bpono = (size_t) -1; size_t pjsno = 0; size_t pjsno_undefs = 0; Elf_Internal_Sym *isymbuf = NULL; bfd_size_type size = sec->rawsize ? sec->rawsize : sec->size; mmix_elf_section_data (sec)->pjs.stubs_size_sum = 0; /* Assume nothing changes. */ *again = false; /* We don't have to do anything if this section does not have relocs, or if this is not a code section. */ if ((sec->flags & SEC_RELOC) == 0 || sec->reloc_count == 0 || (sec->flags & SEC_CODE) == 0 || (sec->flags & SEC_LINKER_CREATED) != 0 /* If no R_MMIX_BASE_PLUS_OFFSET relocs and no PUSHJ-stub relocs, then nothing to do. */ || (bpodata == NULL && mmix_elf_section_data (sec)->pjs.n_pushj_relocs == 0)) return true; symtab_hdr = &elf_tdata (abfd)->symtab_hdr; if (bpodata != NULL) { bpo_gregs_section = bpodata->bpo_greg_section; gregdata = mmix_elf_section_data (bpo_gregs_section)->bpo.greg; bpono = bpodata->first_base_plus_offset_reloc; } else gregdata = NULL; /* Get a copy of the native relocations. */ internal_relocs = _bfd_elf_link_read_relocs (abfd, sec, NULL, (Elf_Internal_Rela *) NULL, link_info->keep_memory); if (internal_relocs == NULL) goto error_return; /* Walk through them looking for relaxing opportunities. */ irelend = internal_relocs + sec->reloc_count; for (irel = internal_relocs; irel < irelend; irel++) { bfd_vma symval; struct elf_link_hash_entry *h = NULL; /* We only process two relocs. */ if (ELF64_R_TYPE (irel->r_info) != (int) R_MMIX_BASE_PLUS_OFFSET && ELF64_R_TYPE (irel->r_info) != (int) R_MMIX_PUSHJ_STUBBABLE) continue; /* We process relocs in a distinctly different way when this is a relocatable link (for one, we don't look at symbols), so we avoid mixing its code with that for the "normal" relaxation. */ if (bfd_link_relocatable (link_info)) { /* The only transformation in a relocatable link is to generate a full stub at the location of the stub calculated for the input section, if the relocated stub location, the end of the output section plus earlier stubs, cannot be reached. Thus relocatable linking can only lead to worse code, but it still works. */ if (ELF64_R_TYPE (irel->r_info) == R_MMIX_PUSHJ_STUBBABLE) { /* If we can reach the end of the output-section and beyond any current stubs, then we don't need a stub for this reloc. The relaxed order of output stub allocation may not exactly match the straightforward order, so we always assume presence of output stubs, which will allow relaxation only on relocations indifferent to the presence of output stub allocations for other relocations and thus the order of output stub allocation. */ if (bfd_check_overflow (complain_overflow_signed, 19, 0, bfd_arch_bits_per_address (abfd), /* Output-stub location. */ sec->output_section->rawsize + (mmix_elf_section_data (sec ->output_section) ->pjs.stubs_size_sum) /* Location of this PUSHJ reloc. */ - (sec->output_offset + irel->r_offset) /* Don't count *this* stub twice. */ - (mmix_elf_section_data (sec) ->pjs.stub_size[pjsno] + MAX_PUSHJ_STUB_SIZE)) == bfd_reloc_ok) mmix_elf_section_data (sec)->pjs.stub_size[pjsno] = 0; mmix_elf_section_data (sec)->pjs.stubs_size_sum += mmix_elf_section_data (sec)->pjs.stub_size[pjsno]; pjsno++; } continue; } /* Get the value of the symbol referred to by the reloc. */ if (ELF64_R_SYM (irel->r_info) < symtab_hdr->sh_info) { /* A local symbol. */ Elf_Internal_Sym *isym; asection *sym_sec; /* Read this BFD's local symbols if we haven't already. */ if (isymbuf == NULL) { isymbuf = (Elf_Internal_Sym *) symtab_hdr->contents; if (isymbuf == NULL) isymbuf = bfd_elf_get_elf_syms (abfd, symtab_hdr, symtab_hdr->sh_info, 0, NULL, NULL, NULL); if (isymbuf == 0) goto error_return; } isym = isymbuf + ELF64_R_SYM (irel->r_info); if (isym->st_shndx == SHN_UNDEF) sym_sec = bfd_und_section_ptr; else if (isym->st_shndx == SHN_ABS) sym_sec = bfd_abs_section_ptr; else if (isym->st_shndx == SHN_COMMON) sym_sec = bfd_com_section_ptr; else sym_sec = bfd_section_from_elf_index (abfd, isym->st_shndx); symval = (isym->st_value + sym_sec->output_section->vma + sym_sec->output_offset); } else { unsigned long indx; /* An external symbol. */ indx = ELF64_R_SYM (irel->r_info) - symtab_hdr->sh_info; h = elf_sym_hashes (abfd)[indx]; BFD_ASSERT (h != NULL); if (h->root.type == bfd_link_hash_undefweak) /* FIXME: for R_MMIX_PUSHJ_STUBBABLE, there are alternatives to the canonical value 0 for an unresolved weak symbol to consider: as the debug-friendly approach, resolve to "abort" (or a port-specific function), or as the space-friendly approach resolve to the next instruction (like some other ports, notably ARM and AArch64). These alternatives require matching code in mmix_elf_perform_relocation or its caller. */ symval = 0; else if (h->root.type == bfd_link_hash_defined || h->root.type == bfd_link_hash_defweak) symval = (h->root.u.def.value + h->root.u.def.section->output_section->vma + h->root.u.def.section->output_offset); else { /* This appears to be a reference to an undefined symbol. Just ignore it--it will be caught by the regular reloc processing. We need to keep BPO reloc accounting consistent, though else we'll abort instead of emitting an error message. */ if (ELF64_R_TYPE (irel->r_info) == R_MMIX_BASE_PLUS_OFFSET && gregdata != NULL) { gregdata->n_remaining_bpo_relocs_this_relaxation_round--; bpono++; } /* Similarly, keep accounting consistent for PUSHJ referring to an undefined symbol. */ if (ELF64_R_TYPE (irel->r_info) == R_MMIX_PUSHJ_STUBBABLE) pjsno_undefs++; continue; } } if (ELF64_R_TYPE (irel->r_info) == (int) R_MMIX_PUSHJ_STUBBABLE) { bfd_vma value = symval + irel->r_addend; bfd_vma dot = (sec->output_section->vma + sec->output_offset + irel->r_offset); bfd_vma stubaddr = (sec->output_section->vma + sec->output_offset + size + mmix_elf_section_data (sec)->pjs.stubs_size_sum); if ((value & 3) == 0 && bfd_check_overflow (complain_overflow_signed, 19, 0, bfd_arch_bits_per_address (abfd), value - dot - (value > dot ? mmix_elf_section_data (sec) ->pjs.stub_size[pjsno] : 0)) == bfd_reloc_ok) /* If the reloc fits, no stub is needed. */ mmix_elf_section_data (sec)->pjs.stub_size[pjsno] = 0; else /* Maybe we can get away with just a JMP insn? */ if ((value & 3) == 0 && bfd_check_overflow (complain_overflow_signed, 27, 0, bfd_arch_bits_per_address (abfd), value - stubaddr - (value > dot ? mmix_elf_section_data (sec) ->pjs.stub_size[pjsno] - 4 : 0)) == bfd_reloc_ok) /* Yep, account for a stub consisting of a single JMP insn. */ mmix_elf_section_data (sec)->pjs.stub_size[pjsno] = 4; else /* Nope, go for the full insn stub. It doesn't seem useful to emit the intermediate sizes; those will only be useful for a >64M program assuming contiguous code. */ mmix_elf_section_data (sec)->pjs.stub_size[pjsno] = MAX_PUSHJ_STUB_SIZE; mmix_elf_section_data (sec)->pjs.stubs_size_sum += mmix_elf_section_data (sec)->pjs.stub_size[pjsno]; pjsno++; continue; } /* We're looking at a R_MMIX_BASE_PLUS_OFFSET reloc. */ gregdata->reloc_request[gregdata->bpo_reloc_indexes[bpono]].value = symval + irel->r_addend; gregdata->reloc_request[gregdata->bpo_reloc_indexes[bpono++]].valid = true; gregdata->n_remaining_bpo_relocs_this_relaxation_round--; } /* Check if that was the last BPO-reloc. If so, sort the values and calculate how many registers we need to cover them. Set the size of the linker gregs, and if the number of registers changed, indicate that we need to relax some more because we have more work to do. */ if (gregdata != NULL && gregdata->n_remaining_bpo_relocs_this_relaxation_round == 0) { size_t i; bfd_vma prev_base; size_t regindex; /* First, reset the remaining relocs for the next round. */ gregdata->n_remaining_bpo_relocs_this_relaxation_round = gregdata->n_bpo_relocs; qsort (gregdata->reloc_request, gregdata->n_max_bpo_relocs, sizeof (struct bpo_reloc_request), bpo_reloc_request_sort_fn); /* Recalculate indexes. When we find a change (however unlikely after the initial iteration), we know we need to relax again, since items in the GREG-array are sorted by increasing value and stored in the relaxation phase. */ for (i = 0; i < gregdata->n_max_bpo_relocs; i++) if (gregdata->bpo_reloc_indexes[gregdata->reloc_request[i].bpo_reloc_no] != i) { gregdata->bpo_reloc_indexes[gregdata->reloc_request[i].bpo_reloc_no] = i; *again = true; } /* Allocate register numbers (indexing from 0). Stop at the first non-valid reloc. */ for (i = 0, regindex = 0, prev_base = gregdata->reloc_request[0].value; i < gregdata->n_bpo_relocs; i++) { if (gregdata->reloc_request[i].value > prev_base + 255) { regindex++; prev_base = gregdata->reloc_request[i].value; } gregdata->reloc_request[i].regindex = regindex; gregdata->reloc_request[i].offset = gregdata->reloc_request[i].value - prev_base; } /* If it's not the same as the last time, we need to relax again, because the size of the section has changed. I'm not sure we actually need to do any adjustments since the shrinking happens at the start of this section, but better safe than sorry. */ if (gregdata->n_allocated_bpo_gregs != regindex + 1) { gregdata->n_allocated_bpo_gregs = regindex + 1; *again = true; } bpo_gregs_section->size = (regindex + 1) * 8; } if (isymbuf != NULL && (unsigned char *) isymbuf != symtab_hdr->contents) { if (! link_info->keep_memory) free (isymbuf); else { /* Cache the symbols for elf_link_input_bfd. */ symtab_hdr->contents = (unsigned char *) isymbuf; } } BFD_ASSERT(pjsno + pjsno_undefs == mmix_elf_section_data (sec)->pjs.n_pushj_relocs); if (elf_section_data (sec)->relocs != internal_relocs) free (internal_relocs); if (sec->size < size + mmix_elf_section_data (sec)->pjs.stubs_size_sum) abort (); if (sec->size > size + mmix_elf_section_data (sec)->pjs.stubs_size_sum) { sec->size = size + mmix_elf_section_data (sec)->pjs.stubs_size_sum; *again = true; } return true; error_return: if ((unsigned char *) isymbuf != symtab_hdr->contents) free (isymbuf); if (elf_section_data (sec)->relocs != internal_relocs) free (internal_relocs); return false; } #define ELF_ARCH bfd_arch_mmix #define ELF_MACHINE_CODE EM_MMIX #define ELF_TARGET_ID MMIX_ELF_DATA /* According to mmix-doc page 36 (paragraph 45), this should be (1LL << 48LL). However, that's too much for something somewhere in the linker part of BFD; perhaps the start-address has to be a non-zero multiple of this number, or larger than this number. The symptom is that the linker complains: "warning: allocated section `.text' not in segment". We settle for 64k; the page-size used in examples is 8k. #define ELF_MAXPAGESIZE 0x10000 Unfortunately, this causes excessive padding in the supposedly small for-education programs that are the expected usage (where people would inspect output). We stick to 256 bytes just to have *some* default alignment. */ #define ELF_MAXPAGESIZE 0x100 #define TARGET_BIG_SYM mmix_elf64_vec #define TARGET_BIG_NAME "elf64-mmix" #define elf_info_to_howto_rel NULL #define elf_info_to_howto mmix_info_to_howto_rela #define elf_backend_relocate_section mmix_elf_relocate_section #define elf_backend_gc_mark_hook mmix_elf_gc_mark_hook #define elf_backend_link_output_symbol_hook \ mmix_elf_link_output_symbol_hook #define elf_backend_add_symbol_hook mmix_elf_add_symbol_hook #define elf_backend_check_relocs mmix_elf_check_relocs #define elf_backend_symbol_processing mmix_elf_symbol_processing #define elf_backend_omit_section_dynsym _bfd_elf_omit_section_dynsym_all #define bfd_elf64_bfd_copy_link_hash_symbol_type \ _bfd_generic_copy_link_hash_symbol_type #define bfd_elf64_bfd_is_local_label_name \ mmix_elf_is_local_label_name #define elf_backend_may_use_rel_p 0 #define elf_backend_may_use_rela_p 1 #define elf_backend_default_use_rela_p 1 #define elf_backend_can_gc_sections 1 #define elf_backend_section_from_bfd_section \ mmix_elf_section_from_bfd_section #define bfd_elf64_new_section_hook mmix_elf_new_section_hook #define bfd_elf64_bfd_final_link mmix_elf_final_link #define bfd_elf64_bfd_relax_section mmix_elf_relax_section #include "elf64-target.h"