This is the structure of a relocation entry:
struct reloc_cache_entry { /* A pointer into the canonical table of pointers. */ struct bfd_symbol **sym_ptr_ptr; /* offset in section. */ bfd_size_type address; /* addend for relocation value. */ bfd_vma addend; /* Pointer to how to perform the required relocation. */ reloc_howto_type *howto; };
Here is a description of each of the fields within an arelent
:
sym_ptr_ptr
The symbol table pointer points to a pointer to the symbol
associated with the relocation request. It is the pointer
into the table returned by the back end’s
canonicalize_symtab
action. See Symbols. The symbol is
referenced through a pointer to a pointer so that tools like
the linker can fix up all the symbols of the same name by
modifying only one pointer. The relocation routine looks in
the symbol and uses the base of the section the symbol is
attached to and the value of the symbol as the initial
relocation offset. If the symbol pointer is zero, then the
section provided is looked up.
address
The address
field gives the offset in bytes from the base of
the section data which owns the relocation record to the first
byte of relocatable information. The actual data relocated
will be relative to this point; for example, a relocation
type which modifies the bottom two bytes of a four byte word
would not touch the first byte pointed to in a big endian
world.
addend
The addend
is a value provided by the back end to be added (!)
to the relocation offset. Its interpretation is dependent upon
the howto. For example, on the 68k the code:
char foo[]; main() { return foo[0x12345678]; }
Could be compiled into:
linkw fp,#-4 moveb @#12345678,d0 extbl d0 unlk fp rts
This could create a reloc pointing to foo
, but leave the
offset in the data, something like:
RELOCATION RECORDS FOR [.text]: offset type value 00000006 32 _foo 00000000 4e56 fffc ; linkw fp,#-4 00000004 1039 1234 5678 ; moveb @#12345678,d0 0000000a 49c0 ; extbl d0 0000000c 4e5e ; unlk fp 0000000e 4e75 ; rts
Using coff and an 88k, some instructions don’t have enough space in them to represent the full address range, and pointers have to be loaded in two parts. So you’d get something like:
or.u r13,r0,hi16(_foo+0x12345678) ld.b r2,r13,lo16(_foo+0x12345678) jmp r1
This should create two relocs, both pointing to _foo
, and with
0x12340000 in their addend field. The data would consist of:
RELOCATION RECORDS FOR [.text]: offset type value 00000002 HVRT16 _foo+0x12340000 00000006 LVRT16 _foo+0x12340000 00000000 5da05678 ; or.u r13,r0,0x5678 00000004 1c4d5678 ; ld.b r2,r13,0x5678 00000008 f400c001 ; jmp r1
The relocation routine digs out the value from the data, adds
it to the addend to get the original offset, and then adds the
value of _foo
. Note that all 32 bits have to be kept around
somewhere, to cope with carry from bit 15 to bit 16.
One further example is the sparc and the a.out format. The sparc has a similar problem to the 88k, in that some instructions don’t have room for an entire offset, but on the sparc the parts are created in odd sized lumps. The designers of the a.out format chose to not use the data within the section for storing part of the offset; all the offset is kept within the reloc. Anything in the data should be ignored.
save %sp,-112,%sp sethi %hi(_foo+0x12345678),%g2 ldsb [%g2+%lo(_foo+0x12345678)],%i0 ret restore
Both relocs contain a pointer to foo
, and the offsets
contain junk.
RELOCATION RECORDS FOR [.text]: offset type value 00000004 HI22 _foo+0x12345678 00000008 LO10 _foo+0x12345678 00000000 9de3bf90 ; save %sp,-112,%sp 00000004 05000000 ; sethi %hi(_foo+0),%g2 00000008 f048a000 ; ldsb [%g2+%lo(_foo+0)],%i0 0000000c 81c7e008 ; ret 00000010 81e80000 ; restore
howto
The howto
field can be imagined as a
relocation instruction. It is a pointer to a structure which
contains information on what to do with all of the other
information in the reloc record and data section. A back end
would normally have a relocation instruction set and turn
relocations into pointers to the correct structure on input -
but it would be possible to create each howto field on demand.
enum complain_overflow
reloc_howto_type
The HOWTO Macro
arelent_chain
bfd_check_overflow
bfd_reloc_offset_in_range
bfd_perform_relocation
bfd_install_relocation
enum complain_overflow
Indicates what sort of overflow checking should be done when performing a relocation.
enum complain_overflow { /* Do not complain on overflow. */ complain_overflow_dont, /* Complain if the value overflows when considered as a signed number one bit larger than the field. ie. A bitfield of N bits is allowed to represent -2**n to 2**n-1. */ complain_overflow_bitfield, /* Complain if the value overflows when considered as a signed number. */ complain_overflow_signed, /* Complain if the value overflows when considered as an unsigned number. */ complain_overflow_unsigned };
reloc_howto_type
The reloc_howto_type
is a structure which contains all the
information that libbfd needs to know to tie up a back end’s data.
struct reloc_howto_struct { /* The type field has mainly a documentary use - the back end can do what it wants with it, though normally the back end's idea of an external reloc number is stored in this field. */ unsigned int type; /* The size of the item to be relocated in bytes. */ unsigned int size:4; /* The number of bits in the field to be relocated. This is used when doing overflow checking. */ unsigned int bitsize:7; /* The value the final relocation is shifted right by. This drops unwanted data from the relocation. */ unsigned int rightshift:6; /* The bit position of the reloc value in the destination. The relocated value is left shifted by this amount. */ unsigned int bitpos:6; /* What type of overflow error should be checked for when relocating. */ ENUM_BITFIELD (complain_overflow) complain_on_overflow:2; /* The relocation value should be negated before applying. */ unsigned int negate:1; /* The relocation is relative to the item being relocated. */ unsigned int pc_relative:1; /* Some formats record a relocation addend in the section contents rather than with the relocation. For ELF formats this is the distinction between USE_REL and USE_RELA (though the code checks for USE_REL == 1/0). The value of this field is TRUE if the addend is recorded with the section contents; when performing a partial link (ld -r) the section contents (the data) will be modified. The value of this field is FALSE if addends are recorded with the relocation (in arelent.addend); when performing a partial link the relocation will be modified. All relocations for all ELF USE_RELA targets should set this field to FALSE (values of TRUE should be looked on with suspicion). However, the converse is not true: not all relocations of all ELF USE_REL targets set this field to TRUE. Why this is so is peculiar to each particular target. For relocs that aren't used in partial links (e.g. GOT stuff) it doesn't matter what this is set to. */ unsigned int partial_inplace:1; /* When some formats create PC relative instructions, they leave the value of the pc of the place being relocated in the offset slot of the instruction, so that a PC relative relocation can be made just by adding in an ordinary offset (e.g., sun3 a.out). Some formats leave the displacement part of an instruction empty (e.g., ELF); this flag signals the fact. */ unsigned int pcrel_offset:1; /* Whether bfd_install_relocation should just install the addend, or should follow the practice of some older object formats and install a value including the symbol. */ unsigned int install_addend:1; /* src_mask selects the part of the instruction (or data) to be used in the relocation sum. If the target relocations don't have an addend in the reloc, eg. ELF USE_REL, src_mask will normally equal dst_mask to extract the addend from the section contents. If relocations do have an addend in the reloc, eg. ELF USE_RELA, this field should normally be zero. Non-zero values for ELF USE_RELA targets should be viewed with suspicion as normally the value in the dst_mask part of the section contents should be ignored. */ bfd_vma src_mask; /* dst_mask selects which parts of the instruction (or data) are replaced with a relocated value. */ bfd_vma dst_mask; /* If this field is non null, then the supplied function is called rather than the normal function. This allows really strange relocation methods to be accommodated. */ bfd_reloc_status_type (*special_function) (bfd *, arelent *, struct bfd_symbol *, void *, asection *, bfd *, char **); /* The textual name of the relocation type. */ const char *name; };
The HOWTO Macro
The HOWTO macro fills in a reloc_howto_type (a typedef for const struct reloc_howto_struct).
#define HOWTO_INSTALL_ADDEND 0 #define HOWTO_RSIZE(sz) ((sz) < 0 ? -(sz) : (sz)) #define HOWTO(type, right, size, bits, pcrel, left, ovf, func, name, \ inplace, src_mask, dst_mask, pcrel_off) \ { (unsigned) type, HOWTO_RSIZE (size), bits, right, left, ovf, \ size < 0, pcrel, inplace, pcrel_off, HOWTO_INSTALL_ADDEND, \ src_mask, dst_mask, func, name }
This is used to fill in an empty howto entry in an array.
#define EMPTY_HOWTO(C) \ HOWTO ((C), 0, 1, 0, false, 0, complain_overflow_dont, NULL, \ NULL, false, 0, 0, false) static inline unsigned int bfd_get_reloc_size (reloc_howto_type *howto) { return howto->size; }
arelent_chain
How relocs are tied together in an asection
:
typedef struct relent_chain { arelent relent; struct relent_chain *next; } arelent_chain;
bfd_check_overflow
bfd_reloc_status_type
bfd_check_overflow (enum complain_overflow how, unsigned int bitsize, unsigned int rightshift, unsigned int addrsize, bfd_vma relocation);
¶Perform overflow checking on relocation which has
bitsize significant bits and will be shifted right by
rightshift bits, on a machine with addresses containing
addrsize significant bits. The result is either of
bfd_reloc_ok
or bfd_reloc_overflow
.
bfd_reloc_offset_in_range
bool
bfd_reloc_offset_in_range (reloc_howto_type *howto, bfd *abfd, asection *section, bfd_size_type offset);
¶Returns TRUE if the reloc described by HOWTO can be applied at OFFSET octets in SECTION.
bfd_perform_relocation
bfd_reloc_status_type
bfd_perform_relocation (bfd *abfd, arelent *reloc_entry, void *data, asection *input_section, bfd *output_bfd, char **error_message);
¶If output_bfd is supplied to this function, the
generated image will be relocatable; the relocations are
copied to the output file after they have been changed to
reflect the new state of the world. There are two ways of
reflecting the results of partial linkage in an output file:
by modifying the output data in place, and by modifying the
relocation record. Some native formats (e.g., basic a.out and
basic coff) have no way of specifying an addend in the
relocation type, so the addend has to go in the output data.
This is no big deal since in these formats the output data
slot will always be big enough for the addend. Complex reloc
types with addends were invented to solve just this problem.
The error_message argument is set to an error message if
this return bfd_reloc_dangerous
.
bfd_install_relocation
bfd_reloc_status_type
bfd_install_relocation (bfd *abfd, arelent *reloc_entry, void *data, bfd_vma data_start, asection *input_section, char **error_message);
¶This looks remarkably like bfd_perform_relocation
, except it
does not expect that the section contents have been filled in.
I.e., it’s suitable for use when creating, rather than applying
a relocation.
For now, this function should be considered reserved for the assembler.