/* SPU specific support for 32-bit ELF
- Copyright 2006, 2007, 2008 Free Software Foundation, Inc.
+ Copyright (C) 2006-2020 Free Software Foundation, Inc.
This file is part of BFD, the Binary File Descriptor library.
#include "elf/spu.h"
#include "elf32-spu.h"
+/* All users of this file have bfd_octets_per_byte (abfd, sec) == 1. */
+#define OCTETS_PER_BYTE(ABFD, SEC) 1
+
/* We use RELA style relocs. Don't define USE_REL. */
static bfd_reloc_status_type spu_elf_rel9 (bfd *, arelent *, asymbol *,
array, so it must be declared in the order of that type. */
static reloc_howto_type elf_howto_table[] = {
- HOWTO (R_SPU_NONE, 0, 0, 0, FALSE, 0, complain_overflow_dont,
+ HOWTO (R_SPU_NONE, 0, 3, 0, FALSE, 0, complain_overflow_dont,
bfd_elf_generic_reloc, "SPU_NONE",
FALSE, 0, 0x00000000, FALSE),
- HOWTO (R_SPU_ADDR10, 4, 2, 10, FALSE, 14, complain_overflow_bitfield,
+ HOWTO (R_SPU_ADDR10, 4, 2, 10, FALSE, 14, complain_overflow_bitfield,
bfd_elf_generic_reloc, "SPU_ADDR10",
FALSE, 0, 0x00ffc000, FALSE),
- HOWTO (R_SPU_ADDR16, 2, 2, 16, FALSE, 7, complain_overflow_bitfield,
+ HOWTO (R_SPU_ADDR16, 2, 2, 16, FALSE, 7, complain_overflow_bitfield,
bfd_elf_generic_reloc, "SPU_ADDR16",
FALSE, 0, 0x007fff80, FALSE),
HOWTO (R_SPU_ADDR16_HI, 16, 2, 16, FALSE, 7, complain_overflow_bitfield,
HOWTO (R_SPU_ADDR16_LO, 0, 2, 16, FALSE, 7, complain_overflow_dont,
bfd_elf_generic_reloc, "SPU_ADDR16_LO",
FALSE, 0, 0x007fff80, FALSE),
- HOWTO (R_SPU_ADDR18, 0, 2, 18, FALSE, 7, complain_overflow_bitfield,
+ HOWTO (R_SPU_ADDR18, 0, 2, 18, FALSE, 7, complain_overflow_bitfield,
bfd_elf_generic_reloc, "SPU_ADDR18",
FALSE, 0, 0x01ffff80, FALSE),
- HOWTO (R_SPU_ADDR32, 0, 2, 32, FALSE, 0, complain_overflow_dont,
+ HOWTO (R_SPU_ADDR32, 0, 2, 32, FALSE, 0, complain_overflow_dont,
bfd_elf_generic_reloc, "SPU_ADDR32",
FALSE, 0, 0xffffffff, FALSE),
- HOWTO (R_SPU_REL16, 2, 2, 16, TRUE, 7, complain_overflow_bitfield,
+ HOWTO (R_SPU_REL16, 2, 2, 16, TRUE, 7, complain_overflow_bitfield,
bfd_elf_generic_reloc, "SPU_REL16",
FALSE, 0, 0x007fff80, TRUE),
- HOWTO (R_SPU_ADDR7, 0, 2, 7, FALSE, 14, complain_overflow_dont,
+ HOWTO (R_SPU_ADDR7, 0, 2, 7, FALSE, 14, complain_overflow_dont,
bfd_elf_generic_reloc, "SPU_ADDR7",
FALSE, 0, 0x001fc000, FALSE),
- HOWTO (R_SPU_REL9, 2, 2, 9, TRUE, 0, complain_overflow_signed,
- spu_elf_rel9, "SPU_REL9",
+ HOWTO (R_SPU_REL9, 2, 2, 9, TRUE, 0, complain_overflow_signed,
+ spu_elf_rel9, "SPU_REL9",
FALSE, 0, 0x0180007f, TRUE),
- HOWTO (R_SPU_REL9I, 2, 2, 9, TRUE, 0, complain_overflow_signed,
- spu_elf_rel9, "SPU_REL9I",
+ HOWTO (R_SPU_REL9I, 2, 2, 9, TRUE, 0, complain_overflow_signed,
+ spu_elf_rel9, "SPU_REL9I",
FALSE, 0, 0x0000c07f, TRUE),
- HOWTO (R_SPU_ADDR10I, 0, 2, 10, FALSE, 14, complain_overflow_signed,
+ HOWTO (R_SPU_ADDR10I, 0, 2, 10, FALSE, 14, complain_overflow_signed,
bfd_elf_generic_reloc, "SPU_ADDR10I",
FALSE, 0, 0x00ffc000, FALSE),
- HOWTO (R_SPU_ADDR16I, 0, 2, 16, FALSE, 7, complain_overflow_signed,
+ HOWTO (R_SPU_ADDR16I, 0, 2, 16, FALSE, 7, complain_overflow_signed,
bfd_elf_generic_reloc, "SPU_ADDR16I",
FALSE, 0, 0x007fff80, FALSE),
- HOWTO (R_SPU_REL32, 0, 2, 32, TRUE, 0, complain_overflow_dont,
+ HOWTO (R_SPU_REL32, 0, 2, 32, TRUE, 0, complain_overflow_dont,
bfd_elf_generic_reloc, "SPU_REL32",
FALSE, 0, 0xffffffff, TRUE),
- HOWTO (R_SPU_ADDR16X, 0, 2, 16, FALSE, 7, complain_overflow_bitfield,
+ HOWTO (R_SPU_ADDR16X, 0, 2, 16, FALSE, 7, complain_overflow_bitfield,
bfd_elf_generic_reloc, "SPU_ADDR16X",
FALSE, 0, 0x007fff80, FALSE),
- HOWTO (R_SPU_PPU32, 0, 2, 32, FALSE, 0, complain_overflow_dont,
+ HOWTO (R_SPU_PPU32, 0, 2, 32, FALSE, 0, complain_overflow_dont,
bfd_elf_generic_reloc, "SPU_PPU32",
FALSE, 0, 0xffffffff, FALSE),
- HOWTO (R_SPU_PPU64, 0, 4, 64, FALSE, 0, complain_overflow_dont,
+ HOWTO (R_SPU_PPU64, 0, 4, 64, FALSE, 0, complain_overflow_dont,
bfd_elf_generic_reloc, "SPU_PPU64",
FALSE, 0, -1, FALSE),
+ HOWTO (R_SPU_ADD_PIC, 0, 0, 0, FALSE, 0, complain_overflow_dont,
+ bfd_elf_generic_reloc, "SPU_ADD_PIC",
+ FALSE, 0, 0x00000000, FALSE),
};
static struct bfd_elf_special_section const spu_elf_special_sections[] = {
switch (code)
{
default:
+ return (enum elf_spu_reloc_type) -1;
+ case BFD_RELOC_NONE:
return R_SPU_NONE;
case BFD_RELOC_SPU_IMM10W:
return R_SPU_ADDR10;
return R_SPU_PPU32;
case BFD_RELOC_SPU_PPU64:
return R_SPU_PPU64;
+ case BFD_RELOC_SPU_ADD_PIC:
+ return R_SPU_ADD_PIC;
}
}
-static void
-spu_elf_info_to_howto (bfd *abfd ATTRIBUTE_UNUSED,
+static bfd_boolean
+spu_elf_info_to_howto (bfd *abfd,
arelent *cache_ptr,
Elf_Internal_Rela *dst)
{
enum elf_spu_reloc_type r_type;
r_type = (enum elf_spu_reloc_type) ELF32_R_TYPE (dst->r_info);
- BFD_ASSERT (r_type < R_SPU_max);
+ /* PR 17512: file: 90c2a92e. */
+ if (r_type >= R_SPU_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_howto_table[(int) r_type];
+ return TRUE;
}
static reloc_howto_type *
{
enum elf_spu_reloc_type r_type = spu_elf_bfd_to_reloc_type (code);
- if (r_type == R_SPU_NONE)
+ if (r_type == (enum elf_spu_reloc_type) -1)
return NULL;
return elf_howto_table + r_type;
if (reloc_entry->address > bfd_get_section_limit (abfd, input_section))
return bfd_reloc_outofrange;
- octets = reloc_entry->address * bfd_octets_per_byte (abfd);
+ octets = reloc_entry->address * OCTETS_PER_BYTE (abfd, input_section);
/* Get symbol value. */
val = 0;
{
Elf_Internal_Shdr *shdr = elf_elfsections (abfd)[j];
- if (ELF_IS_SECTION_IN_SEGMENT_MEMORY (shdr, phdr))
+ if (ELF_SECTION_SIZE (shdr, phdr) != 0
+ && ELF_SECTION_IN_SEGMENT (shdr, phdr))
{
asection *sec = shdr->bfd_section;
spu_elf_section_data (sec)->u.o.ovl_index = num_ovl;
{
struct elf_link_hash_table elf;
+ struct spu_elf_params *params;
+
/* Shortcuts to overlay sections. */
asection *ovtab;
+ asection *init;
asection *toe;
asection **ovl_sec;
/* The stub section for each overlay section. */
asection **stub_sec;
- struct elf_link_hash_entry *ovly_load;
- struct elf_link_hash_entry *ovly_return;
- unsigned long ovly_load_r_symndx;
+ struct elf_link_hash_entry *ovly_entry[2];
/* Number of overlay buffers. */
unsigned int num_buf;
/* Total number of overlays. */
unsigned int num_overlays;
+ /* For soft icache. */
+ unsigned int line_size_log2;
+ unsigned int num_lines_log2;
+ unsigned int fromelem_size_log2;
+
/* How much memory we have. */
unsigned int local_store;
- /* Local store --auto-overlay should reserve for non-overlay
- functions and data. */
- unsigned int overlay_fixed;
- /* Local store --auto-overlay should reserve for stack and heap. */
- unsigned int reserved;
- /* If reserved is not specified, stack analysis will calculate a value
- for the stack. This parameter adjusts that value to allow for
- negative sp access (the ABI says 2000 bytes below sp are valid,
- and the overlay manager uses some of this area). */
- int extra_stack_space;
+
/* Count of overlay stubs needed in non-overlay area. */
unsigned int non_ovly_stub;
- /* Stash various callbacks for --auto-overlay. */
- void (*spu_elf_load_ovl_mgr) (void);
- FILE *(*spu_elf_open_overlay_script) (void);
- void (*spu_elf_relink) (void);
-
- /* Bit 0 set if --auto-overlay.
- Bit 1 set if --auto-relink.
- Bit 2 set if --overlay-rodata. */
- unsigned int auto_overlay : 3;
-#define AUTO_OVERLAY 1
-#define AUTO_RELINK 2
-#define OVERLAY_RODATA 4
-
- /* Set if we should emit symbols for stubs. */
- unsigned int emit_stub_syms:1;
-
- /* Set if we want stubs on calls out of overlay regions to
- non-overlay regions. */
- unsigned int non_overlay_stubs : 1;
+ /* Pointer to the fixup section */
+ asection *sfixup;
/* Set on error. */
unsigned int stub_err : 1;
-
- /* Set if stack size analysis should be done. */
- unsigned int stack_analysis : 1;
-
- /* Set if __stack_* syms will be emitted. */
- unsigned int emit_stack_syms : 1;
};
/* Hijack the generic got fields for overlay stub accounting. */
{
struct got_entry *next;
unsigned int ovl;
- bfd_vma addend;
+ union {
+ bfd_vma addend;
+ bfd_vma br_addr;
+ };
bfd_vma stub_addr;
};
#define spu_hash_table(p) \
- ((struct spu_link_hash_table *) ((p)->hash))
+ (elf_hash_table_id ((struct elf_link_hash_table *) ((p)->hash)) \
+ == SPU_ELF_DATA ? ((struct spu_link_hash_table *) ((p)->hash)) : NULL)
+
+struct call_info
+{
+ struct function_info *fun;
+ struct call_info *next;
+ unsigned int count;
+ unsigned int max_depth;
+ unsigned int is_tail : 1;
+ unsigned int is_pasted : 1;
+ unsigned int broken_cycle : 1;
+ unsigned int priority : 13;
+};
+
+struct function_info
+{
+ /* List of functions called. Also branches to hot/cold part of
+ function. */
+ struct call_info *call_list;
+ /* For hot/cold part of function, point to owner. */
+ struct function_info *start;
+ /* Symbol at start of function. */
+ union {
+ Elf_Internal_Sym *sym;
+ struct elf_link_hash_entry *h;
+ } u;
+ /* Function section. */
+ asection *sec;
+ asection *rodata;
+ /* Where last called from, and number of sections called from. */
+ asection *last_caller;
+ unsigned int call_count;
+ /* Address range of (this part of) function. */
+ bfd_vma lo, hi;
+ /* Offset where we found a store of lr, or -1 if none found. */
+ bfd_vma lr_store;
+ /* Offset where we found the stack adjustment insn. */
+ bfd_vma sp_adjust;
+ /* Stack usage. */
+ int stack;
+ /* Distance from root of call tree. Tail and hot/cold branches
+ count as one deeper. We aren't counting stack frames here. */
+ unsigned int depth;
+ /* Set if global symbol. */
+ unsigned int global : 1;
+ /* Set if known to be start of function (as distinct from a hunk
+ in hot/cold section. */
+ unsigned int is_func : 1;
+ /* Set if not a root node. */
+ unsigned int non_root : 1;
+ /* Flags used during call tree traversal. It's cheaper to replicate
+ the visit flags than have one which needs clearing after a traversal. */
+ unsigned int visit1 : 1;
+ unsigned int visit2 : 1;
+ unsigned int marking : 1;
+ unsigned int visit3 : 1;
+ unsigned int visit4 : 1;
+ unsigned int visit5 : 1;
+ unsigned int visit6 : 1;
+ unsigned int visit7 : 1;
+};
+
+struct spu_elf_stack_info
+{
+ int num_fun;
+ int max_fun;
+ /* Variable size array describing functions, one per contiguous
+ address range belonging to a function. */
+ struct function_info fun[1];
+};
+
+static struct function_info *find_function (asection *, bfd_vma,
+ struct bfd_link_info *);
/* Create a spu ELF linker hash table. */
{
struct spu_link_hash_table *htab;
- htab = bfd_malloc (sizeof (*htab));
+ htab = bfd_zmalloc (sizeof (*htab));
if (htab == NULL)
return NULL;
if (!_bfd_elf_link_hash_table_init (&htab->elf, abfd,
_bfd_elf_link_hash_newfunc,
- sizeof (struct elf_link_hash_entry)))
+ sizeof (struct elf_link_hash_entry),
+ SPU_ELF_DATA))
{
free (htab);
return NULL;
}
- memset (&htab->ovtab, 0,
- sizeof (*htab) - offsetof (struct spu_link_hash_table, ovtab));
-
htab->elf.init_got_refcount.refcount = 0;
htab->elf.init_got_refcount.glist = NULL;
htab->elf.init_got_offset.offset = 0;
return &htab->elf.root;
}
+void
+spu_elf_setup (struct bfd_link_info *info, struct spu_elf_params *params)
+{
+ bfd_vma max_branch_log2;
+
+ struct spu_link_hash_table *htab = spu_hash_table (info);
+ htab->params = params;
+ htab->line_size_log2 = bfd_log2 (htab->params->line_size);
+ htab->num_lines_log2 = bfd_log2 (htab->params->num_lines);
+
+ /* For the software i-cache, we provide a "from" list whose size
+ is a power-of-two number of quadwords, big enough to hold one
+ byte per outgoing branch. Compute this number here. */
+ max_branch_log2 = bfd_log2 (htab->params->max_branch);
+ htab->fromelem_size_log2 = max_branch_log2 > 4 ? max_branch_log2 - 4 : 0;
+}
+
/* Find the symbol for the given R_SYMNDX in IBFD and set *HP and *SYMP
to (hash, NULL) for global symbols, and (NULL, sym) for locals. Set
*SYMSECP to the symbol's section. *LOCSYMSP caches local syms. */
that the linker maps the sections to the right place in the output. */
bfd_boolean
-spu_elf_create_sections (struct bfd_link_info *info,
- int stack_analysis,
- int emit_stack_syms)
+spu_elf_create_sections (struct bfd_link_info *info)
{
- bfd *ibfd;
struct spu_link_hash_table *htab = spu_hash_table (info);
+ bfd *ibfd;
- /* Stash some options away where we can get at them later. */
- htab->stack_analysis = stack_analysis;
- htab->emit_stack_syms = emit_stack_syms;
-
- for (ibfd = info->input_bfds; ibfd != NULL; ibfd = ibfd->link_next)
+ for (ibfd = info->input_bfds; ibfd != NULL; ibfd = ibfd->link.next)
if (bfd_get_section_by_name (ibfd, SPU_PTNOTE_SPUNAME) != NULL)
break;
flagword flags;
ibfd = info->input_bfds;
+ /* This should really be SEC_LINKER_CREATED, but then we'd need
+ to write out the section ourselves. */
flags = SEC_LOAD | SEC_READONLY | SEC_HAS_CONTENTS | SEC_IN_MEMORY;
s = bfd_make_section_anyway_with_flags (ibfd, SPU_PTNOTE_SPUNAME, flags);
if (s == NULL
- || !bfd_set_section_alignment (ibfd, s, 4))
+ || !bfd_set_section_alignment (s, 4))
return FALSE;
+ /* Because we didn't set SEC_LINKER_CREATED we need to set the
+ proper section type. */
+ elf_section_type (s) = SHT_NOTE;
name_len = strlen (bfd_get_filename (info->output_bfd)) + 1;
size = 12 + ((sizeof (SPU_PLUGIN_NAME) + 3) & -4);
size += (name_len + 3) & -4;
- if (!bfd_set_section_size (ibfd, s, size))
+ if (!bfd_set_section_size (s, size))
return FALSE;
data = bfd_zalloc (ibfd, size);
s->contents = data;
}
+ if (htab->params->emit_fixups)
+ {
+ asection *s;
+ flagword flags;
+
+ if (htab->elf.dynobj == NULL)
+ htab->elf.dynobj = ibfd;
+ ibfd = htab->elf.dynobj;
+ flags = (SEC_LOAD | SEC_ALLOC | SEC_READONLY | SEC_HAS_CONTENTS
+ | SEC_IN_MEMORY | SEC_LINKER_CREATED);
+ s = bfd_make_section_anyway_with_flags (ibfd, ".fixup", flags);
+ if (s == NULL || !bfd_set_section_alignment (s, 2))
+ return FALSE;
+ htab->sfixup = s;
+ }
+
return TRUE;
}
return (*s1)->index - (*s2)->index;
}
-/* Identify overlays in the output bfd, and number them. */
+/* Identify overlays in the output bfd, and number them.
+ Returns 0 on error, 1 if no overlays, 2 if overlays. */
-bfd_boolean
+int
spu_elf_find_overlays (struct bfd_link_info *info)
{
struct spu_link_hash_table *htab = spu_hash_table (info);
unsigned int i, n, ovl_index, num_buf;
asection *s;
bfd_vma ovl_end;
+ static const char *const entry_names[2][2] = {
+ { "__ovly_load", "__icache_br_handler" },
+ { "__ovly_return", "__icache_call_handler" }
+ };
if (info->output_bfd->section_count < 2)
- return FALSE;
+ return 1;
alloc_sec
= bfd_malloc (info->output_bfd->section_count * sizeof (*alloc_sec));
if (alloc_sec == NULL)
- return FALSE;
+ return 0;
/* Pick out all the alloced sections. */
for (n = 0, s = info->output_bfd->sections; s != NULL; s = s->next)
if (n == 0)
{
free (alloc_sec);
- return FALSE;
+ return 1;
}
/* Sort them by vma. */
qsort (alloc_sec, n, sizeof (*alloc_sec), sort_sections);
- /* Look for overlapping vmas. Any with overlap must be overlays.
- Count them. Also count the number of overlay regions. */
ovl_end = alloc_sec[0]->vma + alloc_sec[0]->size;
- for (ovl_index = 0, num_buf = 0, i = 1; i < n; i++)
+ if (htab->params->ovly_flavour == ovly_soft_icache)
{
- s = alloc_sec[i];
- if (s->vma < ovl_end)
+ unsigned int prev_buf = 0, set_id = 0;
+
+ /* Look for an overlapping vma to find the first overlay section. */
+ bfd_vma vma_start = 0;
+
+ for (i = 1; i < n; i++)
+ {
+ s = alloc_sec[i];
+ if (s->vma < ovl_end)
+ {
+ asection *s0 = alloc_sec[i - 1];
+ vma_start = s0->vma;
+ ovl_end = (s0->vma
+ + ((bfd_vma) 1
+ << (htab->num_lines_log2 + htab->line_size_log2)));
+ --i;
+ break;
+ }
+ else
+ ovl_end = s->vma + s->size;
+ }
+
+ /* Now find any sections within the cache area. */
+ for (ovl_index = 0, num_buf = 0; i < n; i++)
{
- asection *s0 = alloc_sec[i - 1];
+ s = alloc_sec[i];
+ if (s->vma >= ovl_end)
+ break;
+
+ /* A section in an overlay area called .ovl.init is not
+ an overlay, in the sense that it might be loaded in
+ by the overlay manager, but rather the initial
+ section contents for the overlay buffer. */
+ if (strncmp (s->name, ".ovl.init", 9) != 0)
+ {
+ num_buf = ((s->vma - vma_start) >> htab->line_size_log2) + 1;
+ set_id = (num_buf == prev_buf)? set_id + 1 : 0;
+ prev_buf = num_buf;
+
+ if ((s->vma - vma_start) & (htab->params->line_size - 1))
+ {
+ info->callbacks->einfo (_("%X%P: overlay section %pA "
+ "does not start on a cache line\n"),
+ s);
+ bfd_set_error (bfd_error_bad_value);
+ return 0;
+ }
+ else if (s->size > htab->params->line_size)
+ {
+ info->callbacks->einfo (_("%X%P: overlay section %pA "
+ "is larger than a cache line\n"),
+ s);
+ bfd_set_error (bfd_error_bad_value);
+ return 0;
+ }
- if (spu_elf_section_data (s0)->u.o.ovl_index == 0)
+ alloc_sec[ovl_index++] = s;
+ spu_elf_section_data (s)->u.o.ovl_index
+ = (set_id << htab->num_lines_log2) + num_buf;
+ spu_elf_section_data (s)->u.o.ovl_buf = num_buf;
+ }
+ }
+
+ /* Ensure there are no more overlay sections. */
+ for ( ; i < n; i++)
+ {
+ s = alloc_sec[i];
+ if (s->vma < ovl_end)
{
- alloc_sec[ovl_index] = s0;
- spu_elf_section_data (s0)->u.o.ovl_index = ++ovl_index;
- spu_elf_section_data (s0)->u.o.ovl_buf = ++num_buf;
+ info->callbacks->einfo (_("%X%P: overlay section %pA "
+ "is not in cache area\n"),
+ alloc_sec[i-1]);
+ bfd_set_error (bfd_error_bad_value);
+ return 0;
}
- alloc_sec[ovl_index] = s;
- spu_elf_section_data (s)->u.o.ovl_index = ++ovl_index;
- spu_elf_section_data (s)->u.o.ovl_buf = num_buf;
- if (s0->vma != s->vma)
+ else
+ ovl_end = s->vma + s->size;
+ }
+ }
+ else
+ {
+ /* Look for overlapping vmas. Any with overlap must be overlays.
+ Count them. Also count the number of overlay regions. */
+ for (ovl_index = 0, num_buf = 0, i = 1; i < n; i++)
+ {
+ s = alloc_sec[i];
+ if (s->vma < ovl_end)
{
- info->callbacks->einfo (_("%X%P: overlay sections %A and %A "
- "do not start at the same address.\n"),
- s0, s);
- return FALSE;
+ asection *s0 = alloc_sec[i - 1];
+
+ if (spu_elf_section_data (s0)->u.o.ovl_index == 0)
+ {
+ ++num_buf;
+ if (strncmp (s0->name, ".ovl.init", 9) != 0)
+ {
+ alloc_sec[ovl_index] = s0;
+ spu_elf_section_data (s0)->u.o.ovl_index = ++ovl_index;
+ spu_elf_section_data (s0)->u.o.ovl_buf = num_buf;
+ }
+ else
+ ovl_end = s->vma + s->size;
+ }
+ if (strncmp (s->name, ".ovl.init", 9) != 0)
+ {
+ alloc_sec[ovl_index] = s;
+ spu_elf_section_data (s)->u.o.ovl_index = ++ovl_index;
+ spu_elf_section_data (s)->u.o.ovl_buf = num_buf;
+ if (s0->vma != s->vma)
+ {
+ /* xgettext:c-format */
+ info->callbacks->einfo (_("%X%P: overlay sections %pA "
+ "and %pA do not start at the "
+ "same address\n"),
+ s0, s);
+ bfd_set_error (bfd_error_bad_value);
+ return 0;
+ }
+ if (ovl_end < s->vma + s->size)
+ ovl_end = s->vma + s->size;
+ }
}
- if (ovl_end < s->vma + s->size)
+ else
ovl_end = s->vma + s->size;
}
- else
- ovl_end = s->vma + s->size;
}
htab->num_overlays = ovl_index;
htab->num_buf = num_buf;
htab->ovl_sec = alloc_sec;
- htab->ovly_load = elf_link_hash_lookup (&htab->elf, "__ovly_load",
- FALSE, FALSE, FALSE);
- htab->ovly_return = elf_link_hash_lookup (&htab->elf, "__ovly_return",
- FALSE, FALSE, FALSE);
- return ovl_index != 0;
+
+ if (ovl_index == 0)
+ return 1;
+
+ for (i = 0; i < 2; i++)
+ {
+ const char *name;
+ struct elf_link_hash_entry *h;
+
+ name = entry_names[i][htab->params->ovly_flavour];
+ h = elf_link_hash_lookup (&htab->elf, name, TRUE, FALSE, FALSE);
+ if (h == NULL)
+ return 0;
+
+ if (h->root.type == bfd_link_hash_new)
+ {
+ h->root.type = bfd_link_hash_undefined;
+ h->ref_regular = 1;
+ h->ref_regular_nonweak = 1;
+ h->non_elf = 0;
+ }
+ htab->ovly_entry[i] = h;
+ }
+
+ return 2;
}
-/* Support two sizes of overlay stubs, a slower more compact stub of two
- intructions, and a faster stub of four instructions. */
-#ifndef OVL_STUB_SIZE
-/* Default to faster. */
-#define OVL_STUB_SIZE 16
-/* #define OVL_STUB_SIZE 8 */
-#endif
-#define BRSL 0x33000000
+/* Non-zero to use bra in overlay stubs rather than br. */
+#define BRA_STUBS 0
+
+#define BRA 0x30000000
+#define BRASL 0x31000000
#define BR 0x32000000
+#define BRSL 0x33000000
#define NOP 0x40200000
#define LNOP 0x00200000
#define ILA 0x42000000
/* True if INPUT_SECTION might need overlay stubs. */
static bfd_boolean
-maybe_needs_stubs (asection *input_section, bfd *output_bfd)
+maybe_needs_stubs (asection *input_section)
{
/* No stubs for debug sections and suchlike. */
if ((input_section->flags & SEC_ALLOC) == 0)
return FALSE;
/* No stubs for link-once sections that will be discarded. */
- if (input_section->output_section == NULL
- || input_section->output_section->owner != output_bfd)
+ if (input_section->output_section == bfd_abs_section_ptr)
return FALSE;
/* Don't create stubs for .eh_frame references. */
enum _stub_type
{
no_stub,
- ovl_stub,
+ call_ovl_stub,
+ br000_ovl_stub,
+ br001_ovl_stub,
+ br010_ovl_stub,
+ br011_ovl_stub,
+ br100_ovl_stub,
+ br101_ovl_stub,
+ br110_ovl_stub,
+ br111_ovl_stub,
nonovl_stub,
stub_error
};
struct spu_link_hash_table *htab = spu_hash_table (info);
enum elf_spu_reloc_type r_type;
unsigned int sym_type;
- bfd_boolean branch;
+ bfd_boolean branch, hint, call;
enum _stub_type ret = no_stub;
+ bfd_byte insn[4];
if (sym_sec == NULL
- || sym_sec->output_section == NULL
- || sym_sec->output_section->owner != info->output_bfd
+ || sym_sec->output_section == bfd_abs_section_ptr
|| spu_elf_section_data (sym_sec->output_section) == NULL)
return ret;
if (h != NULL)
{
/* Ensure no stubs for user supplied overlay manager syms. */
- if (h == htab->ovly_load || h == htab->ovly_return)
+ if (h == htab->ovly_entry[0] || h == htab->ovly_entry[1])
return ret;
/* setjmp always goes via an overlay stub, because then the return
makes setjmp/longjmp between overlays work. */
if (strncmp (h->root.root.string, "setjmp", 6) == 0
&& (h->root.root.string[6] == '\0' || h->root.root.string[6] == '@'))
- ret = ovl_stub;
+ ret = call_ovl_stub;
}
- /* Usually, symbols in non-overlay sections don't need stubs. */
- if (spu_elf_section_data (sym_sec->output_section)->u.o.ovl_index == 0
- && !htab->non_overlay_stubs)
- return ret;
-
if (h != NULL)
sym_type = h->type;
else
r_type = ELF32_R_TYPE (irela->r_info);
branch = FALSE;
+ hint = FALSE;
+ call = FALSE;
if (r_type == R_SPU_REL16 || r_type == R_SPU_ADDR16)
{
- bfd_byte insn[4];
-
if (contents == NULL)
{
contents = insn;
else
contents += irela->r_offset;
- if (is_branch (contents) || is_hint (contents))
+ branch = is_branch (contents);
+ hint = is_hint (contents);
+ if (branch || hint)
{
- branch = TRUE;
- if ((contents[0] & 0xfd) == 0x31
+ call = (contents[0] & 0xfd) == 0x31;
+ if (call
&& sym_type != STT_FUNC
&& contents != insn)
{
sym,
sym_sec);
}
- (*_bfd_error_handler) (_("warning: call to non-function"
- " symbol %s defined in %B"),
- sym_sec->owner, sym_name);
+ _bfd_error_handler
+ /* xgettext:c-format */
+ (_("warning: call to non-function symbol %s defined in %pB"),
+ sym_name, sym_sec->owner);
}
}
}
- if (sym_type != STT_FUNC
- && !branch
- && (sym_sec->flags & SEC_CODE) == 0)
+ if ((!branch && htab->params->ovly_flavour == ovly_soft_icache)
+ || (sym_type != STT_FUNC
+ && !(branch || hint)
+ && (sym_sec->flags & SEC_CODE) == 0))
+ return no_stub;
+
+ /* Usually, symbols in non-overlay sections don't need stubs. */
+ if (spu_elf_section_data (sym_sec->output_section)->u.o.ovl_index == 0
+ && !htab->params->non_overlay_stubs)
return ret;
/* A reference from some other section to a symbol in an overlay
section needs a stub. */
if (spu_elf_section_data (sym_sec->output_section)->u.o.ovl_index
!= spu_elf_section_data (input_section->output_section)->u.o.ovl_index)
- ret = ovl_stub;
+ {
+ unsigned int lrlive = 0;
+ if (branch)
+ lrlive = (contents[1] & 0x70) >> 4;
+
+ if (!lrlive && (call || sym_type == STT_FUNC))
+ ret = call_ovl_stub;
+ else
+ ret = br000_ovl_stub + lrlive;
+ }
/* If this insn isn't a branch then we are possibly taking the
- address of a function and passing it out somehow. */
- return !branch && sym_type == STT_FUNC ? nonovl_stub : ret;
+ address of a function and passing it out somehow. Soft-icache code
+ always generates inline code to do indirect branches. */
+ if (!(branch || hint)
+ && sym_type == STT_FUNC
+ && htab->params->ovly_flavour != ovly_soft_icache)
+ ret = nonovl_stub;
+
+ return ret;
}
static bfd_boolean
head = elf_local_got_ents (ibfd) + ELF32_R_SYM (irela->r_info);
}
+ if (htab->params->ovly_flavour == ovly_soft_icache)
+ {
+ htab->stub_count[ovl] += 1;
+ return TRUE;
+ }
+
addend = 0;
if (irela != NULL)
addend = irela->r_addend;
return TRUE;
}
+/* Support two sizes of overlay stubs, a slower more compact stub of two
+ instructions, and a faster stub of four instructions.
+ Soft-icache stubs are four or eight words. */
+
+static unsigned int
+ovl_stub_size (struct spu_elf_params *params)
+{
+ return 16 << params->ovly_flavour >> params->compact_stub;
+}
+
+static unsigned int
+ovl_stub_size_log2 (struct spu_elf_params *params)
+{
+ return 4 + params->ovly_flavour - params->compact_stub;
+}
+
/* Two instruction overlay stubs look like:
brsl $75,__ovly_load
ila $78,ovl_number
lnop
ila $79,target_address
- br __ovly_load */
+ br __ovly_load
+
+ Software icache stubs are:
+
+ .word target_index
+ .word target_ia;
+ .word lrlive_branchlocalstoreaddr;
+ brasl $75,__icache_br_handler
+ .quad xor_pattern
+*/
static bfd_boolean
-build_stub (struct spu_link_hash_table *htab,
+build_stub (struct bfd_link_info *info,
bfd *ibfd,
asection *isec,
enum _stub_type stub_type,
bfd_vma dest,
asection *dest_sec)
{
- unsigned int ovl;
+ struct spu_link_hash_table *htab = spu_hash_table (info);
+ unsigned int ovl, dest_ovl, set_id;
struct got_entry *g, **head;
asection *sec;
- bfd_vma addend, val, from, to;
+ bfd_vma addend, from, to, br_dest, patt;
+ unsigned int lrlive;
ovl = 0;
if (stub_type != nonovl_stub)
if (irela != NULL)
addend = irela->r_addend;
- for (g = *head; g != NULL; g = g->next)
- if (g->addend == addend && (g->ovl == ovl || g->ovl == 0))
- break;
- if (g == NULL)
- abort ();
+ if (htab->params->ovly_flavour == ovly_soft_icache)
+ {
+ g = bfd_malloc (sizeof *g);
+ if (g == NULL)
+ return FALSE;
+ g->ovl = ovl;
+ g->br_addr = 0;
+ if (irela != NULL)
+ g->br_addr = (irela->r_offset
+ + isec->output_offset
+ + isec->output_section->vma);
+ g->next = *head;
+ *head = g;
+ }
+ else
+ {
+ for (g = *head; g != NULL; g = g->next)
+ if (g->addend == addend && (g->ovl == ovl || g->ovl == 0))
+ break;
+ if (g == NULL)
+ abort ();
- if (g->ovl == 0 && ovl != 0)
- return TRUE;
+ if (g->ovl == 0 && ovl != 0)
+ return TRUE;
- if (g->stub_addr != (bfd_vma) -1)
- return TRUE;
+ if (g->stub_addr != (bfd_vma) -1)
+ return TRUE;
+ }
sec = htab->stub_sec[ovl];
dest += dest_sec->output_offset + dest_sec->output_section->vma;
from = sec->size + sec->output_offset + sec->output_section->vma;
g->stub_addr = from;
- to = (htab->ovly_load->root.u.def.value
- + htab->ovly_load->root.u.def.section->output_offset
- + htab->ovly_load->root.u.def.section->output_section->vma);
- val = to - from;
- if (OVL_STUB_SIZE == 16)
- val -= 12;
- if (((dest | to | from) & 3) != 0
- || val + 0x40000 >= 0x80000)
+ to = (htab->ovly_entry[0]->root.u.def.value
+ + htab->ovly_entry[0]->root.u.def.section->output_offset
+ + htab->ovly_entry[0]->root.u.def.section->output_section->vma);
+
+ if (((dest | to | from) & 3) != 0)
{
htab->stub_err = 1;
return FALSE;
}
- ovl = spu_elf_section_data (dest_sec->output_section)->u.o.ovl_index;
+ dest_ovl = spu_elf_section_data (dest_sec->output_section)->u.o.ovl_index;
- if (OVL_STUB_SIZE == 16)
+ if (htab->params->ovly_flavour == ovly_normal
+ && !htab->params->compact_stub)
{
- bfd_put_32 (sec->owner, ILA + ((ovl << 7) & 0x01ffff80) + 78,
+ bfd_put_32 (sec->owner, ILA + ((dest_ovl << 7) & 0x01ffff80) + 78,
sec->contents + sec->size);
bfd_put_32 (sec->owner, LNOP,
sec->contents + sec->size + 4);
bfd_put_32 (sec->owner, ILA + ((dest << 7) & 0x01ffff80) + 79,
sec->contents + sec->size + 8);
- bfd_put_32 (sec->owner, BR + ((val << 5) & 0x007fff80),
- sec->contents + sec->size + 12);
+ if (!BRA_STUBS)
+ bfd_put_32 (sec->owner, BR + (((to - (from + 12)) << 5) & 0x007fff80),
+ sec->contents + sec->size + 12);
+ else
+ bfd_put_32 (sec->owner, BRA + ((to << 5) & 0x007fff80),
+ sec->contents + sec->size + 12);
}
- else if (OVL_STUB_SIZE == 8)
+ else if (htab->params->ovly_flavour == ovly_normal
+ && htab->params->compact_stub)
{
- bfd_put_32 (sec->owner, BRSL + ((val << 5) & 0x007fff80) + 75,
- sec->contents + sec->size);
-
- val = (dest & 0x3ffff) | (ovl << 18);
- bfd_put_32 (sec->owner, val,
+ if (!BRA_STUBS)
+ bfd_put_32 (sec->owner, BRSL + (((to - from) << 5) & 0x007fff80) + 75,
+ sec->contents + sec->size);
+ else
+ bfd_put_32 (sec->owner, BRASL + ((to << 5) & 0x007fff80) + 75,
+ sec->contents + sec->size);
+ bfd_put_32 (sec->owner, (dest & 0x3ffff) | (dest_ovl << 18),
sec->contents + sec->size + 4);
}
- else
- abort ();
- sec->size += OVL_STUB_SIZE;
-
- if (htab->emit_stub_syms)
+ else if (htab->params->ovly_flavour == ovly_soft_icache
+ && htab->params->compact_stub)
{
- size_t len;
- char *name;
- int add;
-
- len = 8 + sizeof (".ovl_call.") - 1;
- if (h != NULL)
- len += strlen (h->root.root.string);
- else
- len += 8 + 1 + 8;
- add = 0;
- if (irela != NULL)
- add = (int) irela->r_addend & 0xffffffff;
- if (add != 0)
- len += 1 + 8;
- name = bfd_malloc (len);
- if (name == NULL)
- return FALSE;
+ lrlive = 0;
+ if (stub_type == nonovl_stub)
+ ;
+ else if (stub_type == call_ovl_stub)
+ /* A brsl makes lr live and *(*sp+16) is live.
+ Tail calls have the same liveness. */
+ lrlive = 5;
+ else if (!htab->params->lrlive_analysis)
+ /* Assume stack frame and lr save. */
+ lrlive = 1;
+ else if (irela != NULL)
+ {
+ /* Analyse branch instructions. */
+ struct function_info *caller;
+ bfd_vma off;
- sprintf (name, "%08x.ovl_call.", g->ovl);
- if (h != NULL)
+ caller = find_function (isec, irela->r_offset, info);
+ if (caller->start == NULL)
+ off = irela->r_offset;
+ else
+ {
+ struct function_info *found = NULL;
+
+ /* Find the earliest piece of this function that
+ has frame adjusting instructions. We might
+ see dynamic frame adjustment (eg. for alloca)
+ in some later piece, but functions using
+ alloca always set up a frame earlier. Frame
+ setup instructions are always in one piece. */
+ if (caller->lr_store != (bfd_vma) -1
+ || caller->sp_adjust != (bfd_vma) -1)
+ found = caller;
+ while (caller->start != NULL)
+ {
+ caller = caller->start;
+ if (caller->lr_store != (bfd_vma) -1
+ || caller->sp_adjust != (bfd_vma) -1)
+ found = caller;
+ }
+ if (found != NULL)
+ caller = found;
+ off = (bfd_vma) -1;
+ }
+
+ if (off > caller->sp_adjust)
+ {
+ if (off > caller->lr_store)
+ /* Only *(*sp+16) is live. */
+ lrlive = 1;
+ else
+ /* If no lr save, then we must be in a
+ leaf function with a frame.
+ lr is still live. */
+ lrlive = 4;
+ }
+ else if (off > caller->lr_store)
+ {
+ /* Between lr save and stack adjust. */
+ lrlive = 3;
+ /* This should never happen since prologues won't
+ be split here. */
+ BFD_ASSERT (0);
+ }
+ else
+ /* On entry to function. */
+ lrlive = 5;
+
+ if (stub_type != br000_ovl_stub
+ && lrlive != stub_type - br000_ovl_stub)
+ /* xgettext:c-format */
+ info->callbacks->einfo (_("%pA:0x%v lrlive .brinfo (%u) differs "
+ "from analysis (%u)\n"),
+ isec, irela->r_offset, lrlive,
+ stub_type - br000_ovl_stub);
+ }
+
+ /* If given lrlive info via .brinfo, use it. */
+ if (stub_type > br000_ovl_stub)
+ lrlive = stub_type - br000_ovl_stub;
+
+ if (ovl == 0)
+ to = (htab->ovly_entry[1]->root.u.def.value
+ + htab->ovly_entry[1]->root.u.def.section->output_offset
+ + htab->ovly_entry[1]->root.u.def.section->output_section->vma);
+
+ /* The branch that uses this stub goes to stub_addr + 4. We'll
+ set up an xor pattern that can be used by the icache manager
+ to modify this branch to go directly to its destination. */
+ g->stub_addr += 4;
+ br_dest = g->stub_addr;
+ if (irela == NULL)
+ {
+ /* Except in the case of _SPUEAR_ stubs, the branch in
+ question is the one in the stub itself. */
+ BFD_ASSERT (stub_type == nonovl_stub);
+ g->br_addr = g->stub_addr;
+ br_dest = to;
+ }
+
+ set_id = ((dest_ovl - 1) >> htab->num_lines_log2) + 1;
+ bfd_put_32 (sec->owner, (set_id << 18) | (dest & 0x3ffff),
+ sec->contents + sec->size);
+ bfd_put_32 (sec->owner, BRASL + ((to << 5) & 0x007fff80) + 75,
+ sec->contents + sec->size + 4);
+ bfd_put_32 (sec->owner, (lrlive << 29) | (g->br_addr & 0x3ffff),
+ sec->contents + sec->size + 8);
+ patt = dest ^ br_dest;
+ if (irela != NULL && ELF32_R_TYPE (irela->r_info) == R_SPU_REL16)
+ patt = (dest - g->br_addr) ^ (br_dest - g->br_addr);
+ bfd_put_32 (sec->owner, (patt << 5) & 0x007fff80,
+ sec->contents + sec->size + 12);
+
+ if (ovl == 0)
+ /* Extra space for linked list entries. */
+ sec->size += 16;
+ }
+ else
+ abort ();
+
+ sec->size += ovl_stub_size (htab->params);
+
+ if (htab->params->emit_stub_syms)
+ {
+ size_t len;
+ char *name;
+ int add;
+
+ len = 8 + sizeof (".ovl_call.") - 1;
+ if (h != NULL)
+ len += strlen (h->root.root.string);
+ else
+ len += 8 + 1 + 8;
+ add = 0;
+ if (irela != NULL)
+ add = (int) irela->r_addend & 0xffffffff;
+ if (add != 0)
+ len += 1 + 8;
+ name = bfd_malloc (len + 1);
+ if (name == NULL)
+ return FALSE;
+
+ sprintf (name, "%08x.ovl_call.", g->ovl);
+ if (h != NULL)
strcpy (name + 8 + sizeof (".ovl_call.") - 1, h->root.root.string);
else
sprintf (name + 8 + sizeof (".ovl_call.") - 1, "%x:%x",
{
h->root.type = bfd_link_hash_defined;
h->root.u.def.section = sec;
- h->root.u.def.value = sec->size - OVL_STUB_SIZE;
- h->size = OVL_STUB_SIZE;
+ h->size = ovl_stub_size (htab->params);
+ h->root.u.def.value = sec->size - h->size;
h->type = STT_FUNC;
h->ref_regular = 1;
h->def_regular = 1;
&& h->def_regular
&& strncmp (h->root.root.string, "_SPUEAR_", 8) == 0
&& (sym_sec = h->root.u.def.section) != NULL
- && sym_sec->output_section != NULL
- && sym_sec->output_section->owner == info->output_bfd
+ && sym_sec->output_section != bfd_abs_section_ptr
&& spu_elf_section_data (sym_sec->output_section) != NULL
&& (spu_elf_section_data (sym_sec->output_section)->u.o.ovl_index != 0
- || htab->non_overlay_stubs))
+ || htab->params->non_overlay_stubs))
{
return count_stub (htab, NULL, NULL, nonovl_stub, h, NULL);
}
-
+
return TRUE;
}
&& h->def_regular
&& strncmp (h->root.root.string, "_SPUEAR_", 8) == 0
&& (sym_sec = h->root.u.def.section) != NULL
- && sym_sec->output_section != NULL
- && sym_sec->output_section->owner == info->output_bfd
+ && sym_sec->output_section != bfd_abs_section_ptr
&& spu_elf_section_data (sym_sec->output_section) != NULL
&& (spu_elf_section_data (sym_sec->output_section)->u.o.ovl_index != 0
- || htab->non_overlay_stubs))
+ || htab->params->non_overlay_stubs))
{
- return build_stub (htab, NULL, NULL, nonovl_stub, h, NULL,
+ return build_stub (info, NULL, NULL, nonovl_stub, h, NULL,
h->root.u.def.value, sym_sec);
}
-
+
return TRUE;
}
struct spu_link_hash_table *htab = spu_hash_table (info);
bfd *ibfd;
- for (ibfd = info->input_bfds; ibfd != NULL; ibfd = ibfd->link_next)
+ for (ibfd = info->input_bfds; ibfd != NULL; ibfd = ibfd->link.next)
{
- extern const bfd_target bfd_elf32_spu_vec;
+ extern const bfd_target spu_elf32_vec;
Elf_Internal_Shdr *symtab_hdr;
asection *isec;
Elf_Internal_Sym *local_syms = NULL;
- if (ibfd->xvec != &bfd_elf32_spu_vec)
+ if (ibfd->xvec != &spu_elf32_vec)
continue;
/* We'll need the symbol table in a second. */
|| isec->reloc_count == 0)
continue;
- if (!maybe_needs_stubs (isec, info->output_bfd))
+ if (!maybe_needs_stubs (isec))
continue;
/* Get the relocs. */
if (elf_section_data (isec)->relocs != internal_relocs)
free (internal_relocs);
error_ret_free_local:
- if (local_syms != NULL
- && (symtab_hdr->contents
- != (unsigned char *) local_syms))
+ if (symtab_hdr->contents != (unsigned char *) local_syms)
free (local_syms);
return FALSE;
}
else
dest = sym->st_value;
dest += irela->r_addend;
- if (!build_stub (htab, ibfd, isec, stub_type, h, irela,
+ if (!build_stub (info, ibfd, isec, stub_type, h, irela,
dest, sym_sec))
goto error_ret_free_internal;
}
return TRUE;
}
-/* Allocate space for overlay call and return stubs. */
+/* Allocate space for overlay call and return stubs.
+ Return 0 on error, 1 if no overlays, 2 otherwise. */
int
-spu_elf_size_stubs (struct bfd_link_info *info,
- void (*place_spu_section) (asection *, asection *,
- const char *),
- int non_overlay_stubs)
+spu_elf_size_stubs (struct bfd_link_info *info)
{
- struct spu_link_hash_table *htab = spu_hash_table (info);
+ struct spu_link_hash_table *htab;
bfd *ibfd;
bfd_size_type amt;
flagword flags;
unsigned int i;
asection *stub;
- htab->non_overlay_stubs = non_overlay_stubs;
if (!process_stubs (info, FALSE))
return 0;
+ htab = spu_hash_table (info);
elf_link_hash_traverse (&htab->elf, allocate_spuear_stubs, info);
if (htab->stub_err)
return 0;
- if (htab->stub_count == NULL)
- return 1;
-
ibfd = info->input_bfds;
- amt = (htab->num_overlays + 1) * sizeof (*htab->stub_sec);
- htab->stub_sec = bfd_zmalloc (amt);
- if (htab->stub_sec == NULL)
- return 0;
-
- flags = (SEC_ALLOC | SEC_LOAD | SEC_CODE | SEC_READONLY
- | SEC_HAS_CONTENTS | SEC_IN_MEMORY);
- stub = bfd_make_section_anyway_with_flags (ibfd, ".stub", flags);
- htab->stub_sec[0] = stub;
- if (stub == NULL
- || !bfd_set_section_alignment (ibfd, stub, 3 + (OVL_STUB_SIZE > 8)))
- return 0;
- stub->size = htab->stub_count[0] * OVL_STUB_SIZE;
- (*place_spu_section) (stub, NULL, ".text");
-
- for (i = 0; i < htab->num_overlays; ++i)
+ if (htab->stub_count != NULL)
{
- asection *osec = htab->ovl_sec[i];
- unsigned int ovl = spu_elf_section_data (osec)->u.o.ovl_index;
+ amt = (htab->num_overlays + 1) * sizeof (*htab->stub_sec);
+ htab->stub_sec = bfd_zmalloc (amt);
+ if (htab->stub_sec == NULL)
+ return 0;
+
+ flags = (SEC_ALLOC | SEC_LOAD | SEC_CODE | SEC_READONLY
+ | SEC_HAS_CONTENTS | SEC_IN_MEMORY);
stub = bfd_make_section_anyway_with_flags (ibfd, ".stub", flags);
- htab->stub_sec[ovl] = stub;
+ htab->stub_sec[0] = stub;
if (stub == NULL
- || !bfd_set_section_alignment (ibfd, stub, 3 + (OVL_STUB_SIZE > 8)))
+ || !bfd_set_section_alignment (stub,
+ ovl_stub_size_log2 (htab->params)))
return 0;
- stub->size = htab->stub_count[ovl] * OVL_STUB_SIZE;
- (*place_spu_section) (stub, osec, NULL);
+ stub->size = htab->stub_count[0] * ovl_stub_size (htab->params);
+ if (htab->params->ovly_flavour == ovly_soft_icache)
+ /* Extra space for linked list entries. */
+ stub->size += htab->stub_count[0] * 16;
+
+ for (i = 0; i < htab->num_overlays; ++i)
+ {
+ asection *osec = htab->ovl_sec[i];
+ unsigned int ovl = spu_elf_section_data (osec)->u.o.ovl_index;
+ stub = bfd_make_section_anyway_with_flags (ibfd, ".stub", flags);
+ htab->stub_sec[ovl] = stub;
+ if (stub == NULL
+ || !bfd_set_section_alignment (stub,
+ ovl_stub_size_log2 (htab->params)))
+ return 0;
+ stub->size = htab->stub_count[ovl] * ovl_stub_size (htab->params);
+ }
}
- /* htab->ovtab consists of two arrays.
- . struct {
- . u32 vma;
- . u32 size;
- . u32 file_off;
- . u32 buf;
- . } _ovly_table[];
- .
- . struct {
- . u32 mapped;
- . } _ovly_buf_table[];
- . */
-
- flags = (SEC_ALLOC | SEC_LOAD
- | SEC_HAS_CONTENTS | SEC_IN_MEMORY);
- htab->ovtab = bfd_make_section_anyway_with_flags (ibfd, ".ovtab", flags);
- if (htab->ovtab == NULL
- || !bfd_set_section_alignment (ibfd, htab->ovtab, 4))
- return 0;
+ if (htab->params->ovly_flavour == ovly_soft_icache)
+ {
+ /* Space for icache manager tables.
+ a) Tag array, one quadword per cache line.
+ b) Rewrite "to" list, one quadword per cache line.
+ c) Rewrite "from" list, one byte per outgoing branch (rounded up to
+ a power-of-two number of full quadwords) per cache line. */
+
+ flags = SEC_ALLOC;
+ htab->ovtab = bfd_make_section_anyway_with_flags (ibfd, ".ovtab", flags);
+ if (htab->ovtab == NULL
+ || !bfd_set_section_alignment (htab->ovtab, 4))
+ return 0;
+
+ htab->ovtab->size = (16 + 16 + (16 << htab->fromelem_size_log2))
+ << htab->num_lines_log2;
- htab->ovtab->size = htab->num_overlays * 16 + 16 + htab->num_buf * 4;
- (*place_spu_section) (htab->ovtab, NULL, ".data");
+ flags = SEC_ALLOC | SEC_LOAD | SEC_HAS_CONTENTS | SEC_IN_MEMORY;
+ htab->init = bfd_make_section_anyway_with_flags (ibfd, ".ovini", flags);
+ if (htab->init == NULL
+ || !bfd_set_section_alignment (htab->init, 4))
+ return 0;
+
+ htab->init->size = 16;
+ }
+ else if (htab->stub_count == NULL)
+ return 1;
+ else
+ {
+ /* htab->ovtab consists of two arrays.
+ . struct {
+ . u32 vma;
+ . u32 size;
+ . u32 file_off;
+ . u32 buf;
+ . } _ovly_table[];
+ .
+ . struct {
+ . u32 mapped;
+ . } _ovly_buf_table[];
+ . */
+
+ flags = SEC_ALLOC | SEC_LOAD | SEC_HAS_CONTENTS | SEC_IN_MEMORY;
+ htab->ovtab = bfd_make_section_anyway_with_flags (ibfd, ".ovtab", flags);
+ if (htab->ovtab == NULL
+ || !bfd_set_section_alignment (htab->ovtab, 4))
+ return 0;
+
+ htab->ovtab->size = htab->num_overlays * 16 + 16 + htab->num_buf * 4;
+ }
htab->toe = bfd_make_section_anyway_with_flags (ibfd, ".toe", SEC_ALLOC);
if (htab->toe == NULL
- || !bfd_set_section_alignment (ibfd, htab->toe, 4))
+ || !bfd_set_section_alignment (htab->toe, 4))
return 0;
htab->toe->size = 16;
- (*place_spu_section) (htab->toe, NULL, ".toe");
return 2;
}
+/* Called from ld to place overlay manager data sections. This is done
+ after the overlay manager itself is loaded, mainly so that the
+ linker's htab->init section is placed after any other .ovl.init
+ sections. */
+
+void
+spu_elf_place_overlay_data (struct bfd_link_info *info)
+{
+ struct spu_link_hash_table *htab = spu_hash_table (info);
+ unsigned int i;
+
+ if (htab->stub_sec != NULL)
+ {
+ (*htab->params->place_spu_section) (htab->stub_sec[0], NULL, ".text");
+
+ for (i = 0; i < htab->num_overlays; ++i)
+ {
+ asection *osec = htab->ovl_sec[i];
+ unsigned int ovl = spu_elf_section_data (osec)->u.o.ovl_index;
+ (*htab->params->place_spu_section) (htab->stub_sec[ovl], osec, NULL);
+ }
+ }
+
+ if (htab->params->ovly_flavour == ovly_soft_icache)
+ (*htab->params->place_spu_section) (htab->init, NULL, ".ovl.init");
+
+ if (htab->ovtab != NULL)
+ {
+ const char *ovout = ".data";
+ if (htab->params->ovly_flavour == ovly_soft_icache)
+ ovout = ".bss";
+ (*htab->params->place_spu_section) (htab->ovtab, NULL, ovout);
+ }
+
+ if (htab->toe != NULL)
+ (*htab->params->place_spu_section) (htab->toe, NULL, ".toe");
+}
+
/* Functions to handle embedded spu_ovl.o object. */
static void *
return count;
}
+static int
+ovl_mgr_stat (struct bfd *abfd ATTRIBUTE_UNUSED,
+ void *stream,
+ struct stat *sb)
+{
+ struct _ovl_stream *os = (struct _ovl_stream *) stream;
+
+ memset (sb, 0, sizeof (*sb));
+ sb->st_size = (const char *) os->end - (const char *) os->start;
+ return 0;
+}
+
bfd_boolean
spu_elf_open_builtin_lib (bfd **ovl_bfd, const struct _ovl_stream *stream)
{
(void *) stream,
ovl_mgr_pread,
NULL,
- NULL);
+ ovl_mgr_stat);
return *ovl_bfd != NULL;
}
+static unsigned int
+overlay_index (asection *sec)
+{
+ if (sec == NULL
+ || sec->output_section == bfd_abs_section_ptr)
+ return 0;
+ return spu_elf_section_data (sec->output_section)->u.o.ovl_index;
+}
+
/* Define an STT_OBJECT symbol. */
static struct elf_link_hash_entry *
h->ref_regular_nonweak = 1;
h->non_elf = 0;
}
+ else if (h->root.u.def.section->owner != NULL)
+ {
+ /* xgettext:c-format */
+ _bfd_error_handler (_("%pB is not allowed to define %s"),
+ h->root.u.def.section->owner,
+ h->root.root.string);
+ bfd_set_error (bfd_error_bad_value);
+ return NULL;
+ }
else
{
- (*_bfd_error_handler) (_("%B is not allowed to define %s"),
- h->root.u.def.section->owner,
- h->root.root.string);
+ _bfd_error_handler (_("you are not allowed to define %s in a script"),
+ h->root.root.string);
bfd_set_error (bfd_error_bad_value);
return NULL;
}
/* Fill in all stubs and the overlay tables. */
-bfd_boolean
-spu_elf_build_stubs (struct bfd_link_info *info, int emit_syms)
+static bfd_boolean
+spu_elf_build_stubs (struct bfd_link_info *info)
{
struct spu_link_hash_table *htab = spu_hash_table (info);
struct elf_link_hash_entry *h;
bfd *obfd;
unsigned int i;
- htab->emit_stub_syms = emit_syms;
- if (htab->stub_count == NULL)
- return TRUE;
-
- for (i = 0; i <= htab->num_overlays; i++)
- if (htab->stub_sec[i]->size != 0)
- {
- htab->stub_sec[i]->contents = bfd_zalloc (htab->stub_sec[i]->owner,
- htab->stub_sec[i]->size);
- if (htab->stub_sec[i]->contents == NULL)
- return FALSE;
- htab->stub_sec[i]->rawsize = htab->stub_sec[i]->size;
- htab->stub_sec[i]->size = 0;
- }
-
- h = elf_link_hash_lookup (&htab->elf, "__ovly_load", FALSE, FALSE, FALSE);
- htab->ovly_load = h;
- BFD_ASSERT (h != NULL
+ if (htab->num_overlays != 0)
+ {
+ for (i = 0; i < 2; i++)
+ {
+ h = htab->ovly_entry[i];
+ if (h != NULL
&& (h->root.type == bfd_link_hash_defined
|| h->root.type == bfd_link_hash_defweak)
- && h->def_regular);
-
- s = h->root.u.def.section->output_section;
- if (spu_elf_section_data (s)->u.o.ovl_index)
- {
- (*_bfd_error_handler) (_("%s in overlay section"),
- h->root.root.string);
- bfd_set_error (bfd_error_bad_value);
- return FALSE;
+ && h->def_regular)
+ {
+ s = h->root.u.def.section->output_section;
+ if (spu_elf_section_data (s)->u.o.ovl_index)
+ {
+ _bfd_error_handler (_("%s in overlay section"),
+ h->root.root.string);
+ bfd_set_error (bfd_error_bad_value);
+ return FALSE;
+ }
+ }
+ }
}
- h = elf_link_hash_lookup (&htab->elf, "__ovly_return", FALSE, FALSE, FALSE);
- htab->ovly_return = h;
-
- /* Fill in all the stubs. */
- process_stubs (info, TRUE);
- if (!htab->stub_err)
- elf_link_hash_traverse (&htab->elf, build_spuear_stubs, info);
-
- if (htab->stub_err)
+ if (htab->stub_sec != NULL)
{
- (*_bfd_error_handler) (_("overlay stub relocation overflow"));
- bfd_set_error (bfd_error_bad_value);
- return FALSE;
- }
+ for (i = 0; i <= htab->num_overlays; i++)
+ if (htab->stub_sec[i]->size != 0)
+ {
+ htab->stub_sec[i]->contents = bfd_zalloc (htab->stub_sec[i]->owner,
+ htab->stub_sec[i]->size);
+ if (htab->stub_sec[i]->contents == NULL)
+ return FALSE;
+ htab->stub_sec[i]->rawsize = htab->stub_sec[i]->size;
+ htab->stub_sec[i]->size = 0;
+ }
- for (i = 0; i <= htab->num_overlays; i++)
- {
- if (htab->stub_sec[i]->size != htab->stub_sec[i]->rawsize)
+ /* Fill in all the stubs. */
+ process_stubs (info, TRUE);
+ if (!htab->stub_err)
+ elf_link_hash_traverse (&htab->elf, build_spuear_stubs, info);
+
+ if (htab->stub_err)
{
- (*_bfd_error_handler) (_("stubs don't match calculated size"));
+ _bfd_error_handler (_("overlay stub relocation overflow"));
bfd_set_error (bfd_error_bad_value);
return FALSE;
}
- htab->stub_sec[i]->rawsize = 0;
+
+ for (i = 0; i <= htab->num_overlays; i++)
+ {
+ if (htab->stub_sec[i]->size != htab->stub_sec[i]->rawsize)
+ {
+ _bfd_error_handler (_("stubs don't match calculated size"));
+ bfd_set_error (bfd_error_bad_value);
+ return FALSE;
+ }
+ htab->stub_sec[i]->rawsize = 0;
+ }
}
+ if (htab->ovtab == NULL || htab->ovtab->size == 0)
+ return TRUE;
+
htab->ovtab->contents = bfd_zalloc (htab->ovtab->owner, htab->ovtab->size);
if (htab->ovtab->contents == NULL)
return FALSE;
- /* Write out _ovly_table. */
p = htab->ovtab->contents;
- /* set low bit of .size to mark non-overlay area as present. */
- p[7] = 1;
- obfd = htab->ovtab->output_section->owner;
- for (s = obfd->sections; s != NULL; s = s->next)
+ if (htab->params->ovly_flavour == ovly_soft_icache)
{
- unsigned int ovl_index = spu_elf_section_data (s)->u.o.ovl_index;
+ bfd_vma off;
+
+ h = define_ovtab_symbol (htab, "__icache_tag_array");
+ if (h == NULL)
+ return FALSE;
+ h->root.u.def.value = 0;
+ h->size = 16 << htab->num_lines_log2;
+ off = h->size;
+
+ h = define_ovtab_symbol (htab, "__icache_tag_array_size");
+ if (h == NULL)
+ return FALSE;
+ h->root.u.def.value = 16 << htab->num_lines_log2;
+ h->root.u.def.section = bfd_abs_section_ptr;
+
+ h = define_ovtab_symbol (htab, "__icache_rewrite_to");
+ if (h == NULL)
+ return FALSE;
+ h->root.u.def.value = off;
+ h->size = 16 << htab->num_lines_log2;
+ off += h->size;
+
+ h = define_ovtab_symbol (htab, "__icache_rewrite_to_size");
+ if (h == NULL)
+ return FALSE;
+ h->root.u.def.value = 16 << htab->num_lines_log2;
+ h->root.u.def.section = bfd_abs_section_ptr;
- if (ovl_index != 0)
+ h = define_ovtab_symbol (htab, "__icache_rewrite_from");
+ if (h == NULL)
+ return FALSE;
+ h->root.u.def.value = off;
+ h->size = 16 << (htab->fromelem_size_log2 + htab->num_lines_log2);
+ off += h->size;
+
+ h = define_ovtab_symbol (htab, "__icache_rewrite_from_size");
+ if (h == NULL)
+ return FALSE;
+ h->root.u.def.value = 16 << (htab->fromelem_size_log2
+ + htab->num_lines_log2);
+ h->root.u.def.section = bfd_abs_section_ptr;
+
+ h = define_ovtab_symbol (htab, "__icache_log2_fromelemsize");
+ if (h == NULL)
+ return FALSE;
+ h->root.u.def.value = htab->fromelem_size_log2;
+ h->root.u.def.section = bfd_abs_section_ptr;
+
+ h = define_ovtab_symbol (htab, "__icache_base");
+ if (h == NULL)
+ return FALSE;
+ h->root.u.def.value = htab->ovl_sec[0]->vma;
+ h->root.u.def.section = bfd_abs_section_ptr;
+ h->size = htab->num_buf << htab->line_size_log2;
+
+ h = define_ovtab_symbol (htab, "__icache_linesize");
+ if (h == NULL)
+ return FALSE;
+ h->root.u.def.value = 1 << htab->line_size_log2;
+ h->root.u.def.section = bfd_abs_section_ptr;
+
+ h = define_ovtab_symbol (htab, "__icache_log2_linesize");
+ if (h == NULL)
+ return FALSE;
+ h->root.u.def.value = htab->line_size_log2;
+ h->root.u.def.section = bfd_abs_section_ptr;
+
+ h = define_ovtab_symbol (htab, "__icache_neg_log2_linesize");
+ if (h == NULL)
+ return FALSE;
+ h->root.u.def.value = -htab->line_size_log2;
+ h->root.u.def.section = bfd_abs_section_ptr;
+
+ h = define_ovtab_symbol (htab, "__icache_cachesize");
+ if (h == NULL)
+ return FALSE;
+ h->root.u.def.value = 1 << (htab->num_lines_log2 + htab->line_size_log2);
+ h->root.u.def.section = bfd_abs_section_ptr;
+
+ h = define_ovtab_symbol (htab, "__icache_log2_cachesize");
+ if (h == NULL)
+ return FALSE;
+ h->root.u.def.value = htab->num_lines_log2 + htab->line_size_log2;
+ h->root.u.def.section = bfd_abs_section_ptr;
+
+ h = define_ovtab_symbol (htab, "__icache_neg_log2_cachesize");
+ if (h == NULL)
+ return FALSE;
+ h->root.u.def.value = -(htab->num_lines_log2 + htab->line_size_log2);
+ h->root.u.def.section = bfd_abs_section_ptr;
+
+ if (htab->init != NULL && htab->init->size != 0)
{
- unsigned long off = ovl_index * 16;
- unsigned int ovl_buf = spu_elf_section_data (s)->u.o.ovl_buf;
+ htab->init->contents = bfd_zalloc (htab->init->owner,
+ htab->init->size);
+ if (htab->init->contents == NULL)
+ return FALSE;
- bfd_put_32 (htab->ovtab->owner, s->vma, p + off);
- bfd_put_32 (htab->ovtab->owner, (s->size + 15) & -16, p + off + 4);
- /* file_off written later in spu_elf_modify_program_headers. */
- bfd_put_32 (htab->ovtab->owner, ovl_buf, p + off + 12);
+ h = define_ovtab_symbol (htab, "__icache_fileoff");
+ if (h == NULL)
+ return FALSE;
+ h->root.u.def.value = 0;
+ h->root.u.def.section = htab->init;
+ h->size = 8;
}
}
+ else
+ {
+ /* Write out _ovly_table. */
+ /* set low bit of .size to mark non-overlay area as present. */
+ p[7] = 1;
+ obfd = htab->ovtab->output_section->owner;
+ for (s = obfd->sections; s != NULL; s = s->next)
+ {
+ unsigned int ovl_index = spu_elf_section_data (s)->u.o.ovl_index;
- h = define_ovtab_symbol (htab, "_ovly_table");
- if (h == NULL)
- return FALSE;
- h->root.u.def.value = 16;
- h->size = htab->num_overlays * 16;
+ if (ovl_index != 0)
+ {
+ unsigned long off = ovl_index * 16;
+ unsigned int ovl_buf = spu_elf_section_data (s)->u.o.ovl_buf;
+
+ bfd_put_32 (htab->ovtab->owner, s->vma, p + off);
+ bfd_put_32 (htab->ovtab->owner, (s->size + 15) & -16,
+ p + off + 4);
+ /* file_off written later in spu_elf_modify_headers. */
+ bfd_put_32 (htab->ovtab->owner, ovl_buf, p + off + 12);
+ }
+ }
- h = define_ovtab_symbol (htab, "_ovly_table_end");
- if (h == NULL)
- return FALSE;
- h->root.u.def.value = htab->num_overlays * 16 + 16;
- h->size = 0;
+ h = define_ovtab_symbol (htab, "_ovly_table");
+ if (h == NULL)
+ return FALSE;
+ h->root.u.def.value = 16;
+ h->size = htab->num_overlays * 16;
- h = define_ovtab_symbol (htab, "_ovly_buf_table");
- if (h == NULL)
- return FALSE;
- h->root.u.def.value = htab->num_overlays * 16 + 16;
- h->size = htab->num_buf * 4;
+ h = define_ovtab_symbol (htab, "_ovly_table_end");
+ if (h == NULL)
+ return FALSE;
+ h->root.u.def.value = htab->num_overlays * 16 + 16;
+ h->size = 0;
- h = define_ovtab_symbol (htab, "_ovly_buf_table_end");
- if (h == NULL)
- return FALSE;
- h->root.u.def.value = htab->num_overlays * 16 + 16 + htab->num_buf * 4;
- h->size = 0;
+ h = define_ovtab_symbol (htab, "_ovly_buf_table");
+ if (h == NULL)
+ return FALSE;
+ h->root.u.def.value = htab->num_overlays * 16 + 16;
+ h->size = htab->num_buf * 4;
+
+ h = define_ovtab_symbol (htab, "_ovly_buf_table_end");
+ if (h == NULL)
+ return FALSE;
+ h->root.u.def.value = htab->num_overlays * 16 + 16 + htab->num_buf * 4;
+ h->size = 0;
+ }
h = define_ovtab_symbol (htab, "_EAR_");
if (h == NULL)
LO .. HI inclusive, and stash some parameters for --auto-overlay. */
asection *
-spu_elf_check_vma (struct bfd_link_info *info,
- int auto_overlay,
- unsigned int lo,
- unsigned int hi,
- unsigned int overlay_fixed,
- unsigned int reserved,
- int extra_stack_space,
- void (*spu_elf_load_ovl_mgr) (void),
- FILE *(*spu_elf_open_overlay_script) (void),
- void (*spu_elf_relink) (void))
+spu_elf_check_vma (struct bfd_link_info *info)
{
struct elf_segment_map *m;
unsigned int i;
struct spu_link_hash_table *htab = spu_hash_table (info);
bfd *abfd = info->output_bfd;
+ bfd_vma hi = htab->params->local_store_hi;
+ bfd_vma lo = htab->params->local_store_lo;
- if (auto_overlay & AUTO_OVERLAY)
- htab->auto_overlay = auto_overlay;
htab->local_store = hi + 1 - lo;
- htab->overlay_fixed = overlay_fixed;
- htab->reserved = reserved;
- htab->extra_stack_space = extra_stack_space;
- htab->spu_elf_load_ovl_mgr = spu_elf_load_ovl_mgr;
- htab->spu_elf_open_overlay_script = spu_elf_open_overlay_script;
- htab->spu_elf_relink = spu_elf_relink;
-
- for (m = elf_tdata (abfd)->segment_map; m != NULL; m = m->next)
+
+ for (m = elf_seg_map (abfd); m != NULL; m = m->next)
if (m->p_type == PT_LOAD)
for (i = 0; i < m->count; i++)
if (m->sections[i]->size != 0
|| m->sections[i]->vma + m->sections[i]->size - 1 > hi))
return m->sections[i];
- /* No need for overlays if it all fits. */
- htab->auto_overlay = 0;
return NULL;
}
/* OFFSET in SEC (presumably) is the beginning of a function prologue.
- Search for stack adjusting insns, and return the sp delta. */
+ Search for stack adjusting insns, and return the sp delta.
+ If a store of lr is found save the instruction offset to *LR_STORE.
+ If a stack adjusting instruction is found, save that offset to
+ *SP_ADJUST. */
static int
-find_function_stack_adjust (asection *sec, bfd_vma offset)
+find_function_stack_adjust (asection *sec,
+ bfd_vma offset,
+ bfd_vma *lr_store,
+ bfd_vma *sp_adjust)
{
- int unrecog;
int reg[128];
memset (reg, 0, sizeof (reg));
- for (unrecog = 0; offset + 4 <= sec->size && unrecog < 32; offset += 4)
+ for ( ; offset + 4 <= sec->size; offset += 4)
{
unsigned char buf[4];
int rt, ra;
if (!bfd_get_section_contents (sec->owner, sec, buf, offset, 4))
break;
- if (buf[0] == 0x24 /* stqd */)
- continue;
-
rt = buf[3] & 0x7f;
ra = ((buf[2] & 0x3f) << 1) | (buf[3] >> 7);
+
+ if (buf[0] == 0x24 /* stqd */)
+ {
+ if (rt == 0 /* lr */ && ra == 1 /* sp */)
+ *lr_store = offset;
+ continue;
+ }
+
/* Partly decoded immediate field. */
imm = (buf[1] << 9) | (buf[2] << 1) | (buf[3] >> 7);
if (rt == 1 /* sp */)
{
- if (imm > 0)
+ if (reg[rt] > 0)
break;
+ *sp_adjust = offset;
return reg[rt];
}
}
reg[rt] = reg[ra] + reg[rb];
if (rt == 1)
- return reg[rt];
+ {
+ if (reg[rt] > 0)
+ break;
+ *sp_adjust = offset;
+ return reg[rt];
+ }
+ }
+ else if (buf[0] == 0x08 && (buf[1] & 0xe0) == 0 /* sf */)
+ {
+ int rb = ((buf[1] & 0x1f) << 2) | ((buf[2] & 0xc0) >> 6);
+
+ reg[rt] = reg[rb] - reg[ra];
+ if (rt == 1)
+ {
+ if (reg[rt] > 0)
+ break;
+ *sp_adjust = offset;
+ return reg[rt];
+ }
}
else if ((buf[0] & 0xfc) == 0x40 /* il, ilh, ilhu, ila */)
{
if (buf[0] == 0x40 /* il */)
{
if ((buf[1] & 0x80) == 0)
- goto unknown_insn;
+ continue;
imm = (imm ^ 0x8000) - 0x8000;
}
else if ((buf[1] & 0x80) == 0 /* ilhu */)
reg[rt] = reg[ra] | imm;
continue;
}
- else if ((buf[0] == 0x33 && imm == 1 /* brsl .+4 */)
- || (buf[0] == 0x08 && (buf[1] & 0xe0) == 0 /* sf */))
+ else if (buf[0] == 0x32 && (buf[1] & 0x80) != 0 /* fsmbi */)
+ {
+ reg[rt] = ( ((imm & 0x8000) ? 0xff000000 : 0)
+ | ((imm & 0x4000) ? 0x00ff0000 : 0)
+ | ((imm & 0x2000) ? 0x0000ff00 : 0)
+ | ((imm & 0x1000) ? 0x000000ff : 0));
+ continue;
+ }
+ else if (buf[0] == 0x16 /* andbi */)
{
- /* Used in pic reg load. Say rt is trashed. */
+ imm >>= 7;
+ imm &= 0xff;
+ imm |= imm << 8;
+ imm |= imm << 16;
+ reg[rt] = reg[ra] & imm;
+ continue;
+ }
+ else if (buf[0] == 0x33 && imm == 1 /* brsl .+4 */)
+ {
+ /* Used in pic reg load. Say rt is trashed. Won't be used
+ in stack adjust, but we need to continue past this branch. */
reg[rt] = 0;
continue;
}
else if (is_branch (buf) || is_indirect_branch (buf))
/* If we hit a branch then we must be out of the prologue. */
break;
- unknown_insn:
- ++unrecog;
}
return 0;
return *s1 < *s2 ? -1 : 1;
}
-struct call_info
-{
- struct function_info *fun;
- struct call_info *next;
- unsigned int count;
- unsigned int max_depth;
- unsigned int is_tail : 1;
- unsigned int is_pasted : 1;
-};
-
-struct function_info
-{
- /* List of functions called. Also branches to hot/cold part of
- function. */
- struct call_info *call_list;
- /* For hot/cold part of function, point to owner. */
- struct function_info *start;
- /* Symbol at start of function. */
- union {
- Elf_Internal_Sym *sym;
- struct elf_link_hash_entry *h;
- } u;
- /* Function section. */
- asection *sec;
- asection *rodata;
- /* Where last called from, and number of sections called from. */
- asection *last_caller;
- unsigned int call_count;
- /* Address range of (this part of) function. */
- bfd_vma lo, hi;
- /* Stack usage. */
- int stack;
- /* Distance from root of call tree. Tail and hot/cold branches
- count as one deeper. We aren't counting stack frames here. */
- unsigned int depth;
- /* Set if global symbol. */
- unsigned int global : 1;
- /* Set if known to be start of function (as distinct from a hunk
- in hot/cold section. */
- unsigned int is_func : 1;
- /* Set if not a root node. */
- unsigned int non_root : 1;
- /* Flags used during call tree traversal. It's cheaper to replicate
- the visit flags than have one which needs clearing after a traversal. */
- unsigned int visit1 : 1;
- unsigned int visit2 : 1;
- unsigned int marking : 1;
- unsigned int visit3 : 1;
- unsigned int visit4 : 1;
- unsigned int visit5 : 1;
- unsigned int visit6 : 1;
- unsigned int visit7 : 1;
-};
-
-struct spu_elf_stack_info
-{
- int num_fun;
- int max_fun;
- /* Variable size array describing functions, one per contiguous
- address range belonging to a function. */
- struct function_info fun[1];
-};
-
/* Allocate a struct spu_elf_stack_info with MAX_FUN struct function_info
entries for section SEC. */
sinfo->fun[i].u.sym = sym_h;
sinfo->fun[i].lo = off;
sinfo->fun[i].hi = off + size;
- sinfo->fun[i].stack = -find_function_stack_adjust (sec, off);
+ sinfo->fun[i].lr_store = -1;
+ sinfo->fun[i].sp_adjust = -1;
+ sinfo->fun[i].stack = -find_function_stack_adjust (sec, off,
+ &sinfo->fun[i].lr_store,
+ &sinfo->fun[i].sp_adjust);
sinfo->num_fun += 1;
return &sinfo->fun[i];
}
const char *f1 = func_name (&sinfo->fun[i - 1]);
const char *f2 = func_name (&sinfo->fun[i]);
+ /* xgettext:c-format */
info->callbacks->einfo (_("warning: %s overlaps %s\n"), f1, f2);
sinfo->fun[i - 1].hi = sinfo->fun[i].lo;
}
else
return &sinfo->fun[mid];
}
- info->callbacks->einfo (_("%A:0x%v not found in function table\n"),
+ /* xgettext:c-format */
+ info->callbacks->einfo (_("%pA:0x%v not found in function table\n"),
sec, offset);
+ bfd_set_error (bfd_error_bad_value);
return NULL;
}
p->fun->start = NULL;
p->fun->is_func = TRUE;
}
- p->count += 1;
+ p->count += callee->count;
/* Reorder list so most recent call is first. */
*pp = p->next;
p->next = caller->call_list;
return FALSE;
}
callee->next = caller->call_list;
- callee->count += 1;
caller->call_list = callee;
return TRUE;
}
overlay stub sections. */
static bfd_boolean
-interesting_section (asection *s, bfd *obfd)
+interesting_section (asection *s)
{
- return (s->output_section != NULL
- && s->output_section->owner == obfd
+ return (s->output_section != bfd_abs_section_ptr
&& ((s->flags & (SEC_ALLOC | SEC_LOAD | SEC_CODE | SEC_IN_MEMORY))
== (SEC_ALLOC | SEC_LOAD | SEC_CODE))
&& s->size != 0);
Elf_Internal_Rela *internal_relocs, *irelaend, *irela;
Elf_Internal_Shdr *symtab_hdr;
void *psyms;
+ unsigned int priority = 0;
static bfd_boolean warned;
- if (!interesting_section (sec, info->output_bfd)
+ if (!interesting_section (sec)
|| sec->reloc_count == 0)
return TRUE;
Elf_Internal_Sym *sym;
struct elf_link_hash_entry *h;
bfd_vma val;
- bfd_boolean reject, is_call;
+ bfd_boolean nonbranch, is_call;
struct function_info *caller;
struct call_info *callee;
- reject = FALSE;
r_type = ELF32_R_TYPE (irela->r_info);
- if (r_type != R_SPU_REL16
- && r_type != R_SPU_ADDR16)
- {
- reject = TRUE;
- if (!(call_tree && spu_hash_table (info)->auto_overlay))
- continue;
- }
+ nonbranch = r_type != R_SPU_REL16 && r_type != R_SPU_ADDR16;
r_indx = ELF32_R_SYM (irela->r_info);
if (!get_sym_h (&h, &sym, &sym_sec, psyms, r_indx, sec->owner))
return FALSE;
if (sym_sec == NULL
- || sym_sec->output_section == NULL
- || sym_sec->output_section->owner != info->output_bfd)
+ || sym_sec->output_section == bfd_abs_section_ptr)
continue;
is_call = FALSE;
- if (!reject)
+ if (!nonbranch)
{
unsigned char insn[4];
if (is_branch (insn))
{
is_call = (insn[0] & 0xfd) == 0x31;
+ priority = insn[1] & 0x0f;
+ priority <<= 8;
+ priority |= insn[2];
+ priority <<= 8;
+ priority |= insn[3];
+ priority >>= 7;
if ((sym_sec->flags & (SEC_ALLOC | SEC_LOAD | SEC_CODE))
!= (SEC_ALLOC | SEC_LOAD | SEC_CODE))
{
if (!warned)
info->callbacks->einfo
- (_("%B(%A+0x%v): call to non-code section"
- " %B(%A), analysis incomplete\n"),
+ /* xgettext:c-format */
+ (_("%pB(%pA+0x%v): call to non-code section"
+ " %pB(%pA), analysis incomplete\n"),
sec->owner, sec, irela->r_offset,
sym_sec->owner, sym_sec);
warned = TRUE;
}
else
{
- reject = TRUE;
- if (!(call_tree && spu_hash_table (info)->auto_overlay)
- || is_hint (insn))
+ nonbranch = TRUE;
+ if (is_hint (insn))
continue;
}
}
- if (reject)
+ if (nonbranch)
{
/* For --auto-overlay, count possible stubs we need for
function pointer references. */
else
sym_type = ELF_ST_TYPE (sym->st_info);
if (sym_type == STT_FUNC)
- spu_hash_table (info)->non_ovly_stub += 1;
- continue;
+ {
+ if (call_tree && spu_hash_table (info)->params->auto_overlay)
+ spu_hash_table (info)->non_ovly_stub += 1;
+ /* If the symbol type is STT_FUNC then this must be a
+ function pointer initialisation. */
+ continue;
+ }
+ /* Ignore data references. */
+ if ((sym_sec->flags & (SEC_ALLOC | SEC_LOAD | SEC_CODE))
+ != (SEC_ALLOC | SEC_LOAD | SEC_CODE))
+ continue;
+ /* Otherwise we probably have a jump table reloc for
+ a switch statement or some other reference to a
+ code label. */
}
if (h)
return FALSE;
callee->is_tail = !is_call;
callee->is_pasted = FALSE;
- callee->count = 0;
+ callee->broken_cycle = FALSE;
+ callee->priority = priority;
+ callee->count = nonbranch? 0 : 1;
if (callee->fun->last_caller != sec)
{
callee->fun->last_caller = sec;
callee->fun->is_func = TRUE;
}
else if (callee->fun->start == NULL)
- callee->fun->start = caller;
+ {
+ struct function_info *caller_start = caller;
+ while (caller_start->start)
+ caller_start = caller_start->start;
+
+ if (caller_start != callee->fun)
+ callee->fun->start = caller_start;
+ }
else
{
struct function_info *callee_start;
These sections are pasted together to form a single function. */
static bfd_boolean
-pasted_function (asection *sec, struct bfd_link_info *info)
+pasted_function (asection *sec)
{
struct bfd_link_order *l;
struct _spu_elf_section_data *sec_data;
callee->fun = fun;
callee->is_tail = TRUE;
callee->is_pasted = TRUE;
- callee->count = 0;
+ callee->broken_cycle = FALSE;
+ callee->priority = 0;
+ callee->count = 1;
if (!insert_callee (fun_start, callee))
free (callee);
return TRUE;
fun_start = &sinfo->fun[sinfo->num_fun - 1];
}
- info->callbacks->einfo (_("%A link_order not found\n"), sec);
- return FALSE;
+ /* Don't return an error if we did not find a function preceding this
+ section. The section may have incorrect flags. */
+ return TRUE;
}
/* Map address ranges in code sections to functions. */
bfd_boolean gaps = FALSE;
bfd_idx = 0;
- for (ibfd = info->input_bfds; ibfd != NULL; ibfd = ibfd->link_next)
+ for (ibfd = info->input_bfds; ibfd != NULL; ibfd = ibfd->link.next)
bfd_idx++;
psym_arr = bfd_zmalloc (bfd_idx * sizeof (*psym_arr));
if (sec_arr == NULL)
return FALSE;
-
for (ibfd = info->input_bfds, bfd_idx = 0;
ibfd != NULL;
- ibfd = ibfd->link_next, bfd_idx++)
+ ibfd = ibfd->link.next, bfd_idx++)
{
- extern const bfd_target bfd_elf32_spu_vec;
+ extern const bfd_target spu_elf32_vec;
Elf_Internal_Shdr *symtab_hdr;
asection *sec;
size_t symcount;
Elf_Internal_Sym *syms, *sy, **psyms, **psy;
asection **psecs, **p;
- if (ibfd->xvec != &bfd_elf32_spu_vec)
+ if (ibfd->xvec != &spu_elf32_vec)
continue;
/* Read all the symbols. */
{
if (!gaps)
for (sec = ibfd->sections; sec != NULL && !gaps; sec = sec->next)
- if (interesting_section (sec, info->output_bfd))
+ if (interesting_section (sec))
{
gaps = TRUE;
break;
continue;
}
- if (symtab_hdr->contents != NULL)
- {
- /* Don't use cached symbols since the generic ELF linker
- code only reads local symbols, and we need globals too. */
- free (symtab_hdr->contents);
- symtab_hdr->contents = NULL;
- }
+ /* Don't use cached symbols since the generic ELF linker
+ code only reads local symbols, and we need globals too. */
+ free (symtab_hdr->contents);
+ symtab_hdr->contents = NULL;
syms = bfd_elf_get_elf_syms (ibfd, symtab_hdr, symcount, 0,
NULL, NULL, NULL);
symtab_hdr->contents = (void *) syms;
asection *s;
*p = s = bfd_section_from_elf_index (ibfd, sy->st_shndx);
- if (s != NULL && interesting_section (s, info->output_bfd))
+ if (s != NULL && interesting_section (s))
*psy++ = sy;
}
symcount = psy - psyms;
}
for (sec = ibfd->sections; sec != NULL && !gaps; sec = sec->next)
- if (interesting_section (sec, info->output_bfd))
+ if (interesting_section (sec))
gaps |= check_function_ranges (sec, info);
}
relocations. */
for (ibfd = info->input_bfds, bfd_idx = 0;
ibfd != NULL;
- ibfd = ibfd->link_next, bfd_idx++)
+ ibfd = ibfd->link.next, bfd_idx++)
{
asection *sec;
for (ibfd = info->input_bfds, bfd_idx = 0;
ibfd != NULL;
- ibfd = ibfd->link_next, bfd_idx++)
+ ibfd = ibfd->link.next, bfd_idx++)
{
Elf_Internal_Shdr *symtab_hdr;
asection *sec;
gaps = FALSE;
for (sec = ibfd->sections; sec != NULL && !gaps; sec = sec->next)
- if (interesting_section (sec, info->output_bfd))
+ if (interesting_section (sec))
gaps |= check_function_ranges (sec, info);
if (!gaps)
continue;
}
}
- for (ibfd = info->input_bfds; ibfd != NULL; ibfd = ibfd->link_next)
+ for (ibfd = info->input_bfds; ibfd != NULL; ibfd = ibfd->link.next)
{
- extern const bfd_target bfd_elf32_spu_vec;
+ extern const bfd_target spu_elf32_vec;
asection *sec;
- if (ibfd->xvec != &bfd_elf32_spu_vec)
+ if (ibfd->xvec != &spu_elf32_vec)
continue;
/* Some of the symbols we've installed as marking the
the range of such functions to the beginning of the
next symbol of interest. */
for (sec = ibfd->sections; sec != NULL; sec = sec->next)
- if (interesting_section (sec, info->output_bfd))
+ if (interesting_section (sec))
{
struct _spu_elf_section_data *sec_data;
struct spu_elf_stack_info *sinfo;
sec_data = spu_elf_section_data (sec);
sinfo = sec_data->u.i.stack_info;
- if (sinfo != NULL)
+ if (sinfo != NULL && sinfo->num_fun != 0)
{
int fun_idx;
bfd_vma hi = sec->size;
sinfo->fun[fun_idx].hi = hi;
hi = sinfo->fun[fun_idx].lo;
}
+
+ sinfo->fun[0].lo = 0;
}
/* No symbols in this section. Must be .init or .fini
or something similar. */
- else if (!pasted_function (sec, info))
+ else if (!pasted_function (sec))
return FALSE;
}
}
for (ibfd = info->input_bfds, bfd_idx = 0;
ibfd != NULL;
- ibfd = ibfd->link_next, bfd_idx++)
+ ibfd = ibfd->link.next, bfd_idx++)
{
if (psym_arr[bfd_idx] == NULL)
continue;
{
bfd *ibfd;
- for (ibfd = info->input_bfds; ibfd != NULL; ibfd = ibfd->link_next)
+ for (ibfd = info->input_bfds; ibfd != NULL; ibfd = ibfd->link.next)
{
- extern const bfd_target bfd_elf32_spu_vec;
+ extern const bfd_target spu_elf32_vec;
asection *sec;
- if (ibfd->xvec != &bfd_elf32_spu_vec)
+ if (ibfd->xvec != &spu_elf32_vec)
continue;
for (sec = ibfd->sections; sec != NULL; sec = sec->next)
callp = &fun->call_list;
while ((call = *callp) != NULL)
{
+ call->max_depth = depth + !call->is_pasted;
if (!call->fun->visit2)
{
- call->max_depth = depth + !call->is_pasted;
if (!remove_cycles (call->fun, info, &call->max_depth))
return FALSE;
if (max_depth < call->max_depth)
}
else if (call->fun->marking)
{
- if (!spu_hash_table (info)->auto_overlay)
+ struct spu_link_hash_table *htab = spu_hash_table (info);
+
+ if (!htab->params->auto_overlay
+ && htab->params->stack_analysis)
{
const char *f1 = func_name (fun);
const char *f2 = func_name (call->fun);
- info->callbacks->info (_("Stack analysis will ignore the call "
+ /* xgettext:c-format */
+ info->callbacks->info (_("stack analysis will ignore the call "
"from %s to %s\n"),
f1, f2);
}
- *callp = call->next;
- free (call);
- continue;
+
+ call->broken_cycle = TRUE;
}
callp = &call->next;
}
return TRUE;
}
+/* Check that we actually visited all nodes in remove_cycles. If we
+ didn't, then there is some cycle in the call graph not attached to
+ any root node. Arbitrarily choose a node in the cycle as a new
+ root and break the cycle. */
+
+static bfd_boolean
+mark_detached_root (struct function_info *fun,
+ struct bfd_link_info *info,
+ void *param)
+{
+ if (fun->visit2)
+ return TRUE;
+ fun->non_root = FALSE;
+ *(unsigned int *) param = 0;
+ return remove_cycles (fun, info, param);
+}
+
/* Populate call_list for each function. */
static bfd_boolean
bfd *ibfd;
unsigned int depth;
- for (ibfd = info->input_bfds; ibfd != NULL; ibfd = ibfd->link_next)
+ for (ibfd = info->input_bfds; ibfd != NULL; ibfd = ibfd->link.next)
{
- extern const bfd_target bfd_elf32_spu_vec;
+ extern const bfd_target spu_elf32_vec;
asection *sec;
- if (ibfd->xvec != &bfd_elf32_spu_vec)
+ if (ibfd->xvec != &spu_elf32_vec)
continue;
for (sec = ibfd->sections; sec != NULL; sec = sec->next)
/* Transfer call info from hot/cold section part of function
to main entry. */
- if (!spu_hash_table (info)->auto_overlay
+ if (!spu_hash_table (info)->params->auto_overlay
&& !for_each_node (transfer_calls, info, 0, FALSE))
return FALSE;
/* Remove cycles from the call graph. We start from the root node(s)
so that we break cycles in a reasonable place. */
depth = 0;
- return for_each_node (remove_cycles, info, &depth, TRUE);
+ if (!for_each_node (remove_cycles, info, &depth, TRUE))
+ return FALSE;
+
+ return for_each_node (mark_detached_root, info, &depth, FALSE);
}
-/* qsort predicate to sort calls by max_depth then count. */
+/* qsort predicate to sort calls by priority, max_depth then count. */
static int
sort_calls (const void *a, const void *b)
struct call_info *const *c2 = b;
int delta;
+ delta = (*c2)->priority - (*c1)->priority;
+ if (delta != 0)
+ return delta;
+
delta = (*c2)->max_depth - (*c1)->max_depth;
if (delta != 0)
return delta;
if (delta != 0)
return delta;
- return c1 - c2;
+ return (char *) c1 - (char *) c2;
}
struct _mos_param {
struct call_info *call;
unsigned int count;
struct _mos_param *mos_param = param;
+ struct spu_link_hash_table *htab = spu_hash_table (info);
if (fun->visit4)
return TRUE;
fun->visit4 = TRUE;
- if (!fun->sec->linker_mark)
+ if (!fun->sec->linker_mark
+ && (htab->params->ovly_flavour != ovly_soft_icache
+ || htab->params->non_ia_text
+ || strncmp (fun->sec->name, ".text.ia.", 9) == 0
+ || strcmp (fun->sec->name, ".init") == 0
+ || strcmp (fun->sec->name, ".fini") == 0))
{
unsigned int size;
this flag to differentiate the two overlay section types. */
fun->sec->flags |= SEC_CODE;
- if (spu_hash_table (info)->auto_overlay & OVERLAY_RODATA)
+ size = fun->sec->size;
+ if (htab->params->auto_overlay & OVERLAY_RODATA)
{
char *name = NULL;
fun->rodata = rodata;
if (fun->rodata)
{
- fun->rodata->linker_mark = 1;
- fun->rodata->gc_mark = 1;
- fun->rodata->flags &= ~SEC_CODE;
+ size += fun->rodata->size;
+ if (htab->params->line_size != 0
+ && size > htab->params->line_size)
+ {
+ size -= fun->rodata->size;
+ fun->rodata = NULL;
+ }
+ else
+ {
+ fun->rodata->linker_mark = 1;
+ fun->rodata->gc_mark = 1;
+ fun->rodata->flags &= ~SEC_CODE;
+ }
}
free (name);
}
}
- size = fun->sec->size;
- if (fun->rodata)
- size += fun->rodata->size;
if (mos_param->max_overlay_size < size)
mos_param->max_overlay_size = size;
}
BFD_ASSERT (!fun->sec->segment_mark);
fun->sec->segment_mark = 1;
}
- if (!mark_overlay_section (call->fun, info, param))
+ if (!call->broken_cycle
+ && !mark_overlay_section (call->fun, info, param))
return FALSE;
}
/* Don't put entry code into an overlay. The overlay manager needs
- a stack! */
+ a stack! Also, don't mark .ovl.init as an overlay. */
if (fun->lo + fun->sec->output_offset + fun->sec->output_section->vma
- == info->output_bfd->start_address)
+ == info->output_bfd->start_address
+ || strncmp (fun->sec->output_section->name, ".ovl.init", 9) == 0)
{
fun->sec->linker_mark = 0;
if (fun->rodata != NULL)
}
for (call = fun->call_list; call != NULL; call = call->next)
- if (!unmark_overlay_section (call->fun, info, param))
+ if (!call->broken_cycle
+ && !unmark_overlay_section (call->fun, info, param))
return FALSE;
if (RECURSE_UNMARK)
size = fun->sec->size;
if (fun->rodata)
size += fun->rodata->size;
- if (size > lib_param->lib_size)
- return TRUE;
- *lib_param->lib_sections++ = fun->sec;
- fun->sec->gc_mark = 0;
- if (fun->rodata && fun->rodata->linker_mark && fun->rodata->gc_mark)
+ if (size <= lib_param->lib_size)
{
- *lib_param->lib_sections++ = fun->rodata;
- fun->rodata->gc_mark = 0;
+ *lib_param->lib_sections++ = fun->sec;
+ fun->sec->gc_mark = 0;
+ if (fun->rodata && fun->rodata->linker_mark && fun->rodata->gc_mark)
+ {
+ *lib_param->lib_sections++ = fun->rodata;
+ fun->rodata->gc_mark = 0;
+ }
+ else
+ *lib_param->lib_sections++ = NULL;
}
- else
- *lib_param->lib_sections++ = NULL;
for (call = fun->call_list; call != NULL; call = call->next)
- collect_lib_sections (call->fun, info, param);
+ if (!call->broken_cycle)
+ collect_lib_sections (call->fun, info, param);
return TRUE;
}
unsigned int i, lib_count;
struct _cl_param collect_lib_param;
struct function_info dummy_caller;
+ struct spu_link_hash_table *htab;
memset (&dummy_caller, 0, sizeof (dummy_caller));
lib_count = 0;
- for (ibfd = info->input_bfds; ibfd != NULL; ibfd = ibfd->link_next)
+ for (ibfd = info->input_bfds; ibfd != NULL; ibfd = ibfd->link.next)
{
- extern const bfd_target bfd_elf32_spu_vec;
+ extern const bfd_target spu_elf32_vec;
asection *sec;
- if (ibfd->xvec != &bfd_elf32_spu_vec)
+ if (ibfd->xvec != &spu_elf32_vec)
continue;
for (sec = ibfd->sections; sec != NULL; sec = sec->next)
if (lib_count > 1)
qsort (lib_sections, lib_count, 2 * sizeof (*lib_sections), sort_lib);
+ htab = spu_hash_table (info);
for (i = 0; i < lib_count; i++)
{
unsigned int tmp, stub_size;
if (p->fun == call->fun)
break;
if (!p)
- stub_size += OVL_STUB_SIZE;
+ stub_size += ovl_stub_size (htab->params);
}
}
if (tmp + stub_size < lib_size)
while ((p = *pp) != NULL)
if (!p->fun->sec->linker_mark)
{
- lib_size += OVL_STUB_SIZE;
+ lib_size += ovl_stub_size (htab->params);
*pp = p->next;
free (p);
}
fun->visit7 = TRUE;
for (call = fun->call_list; call != NULL; call = call->next)
- if (!call->is_pasted)
+ if (!call->is_pasted && !call->broken_cycle)
{
if (!collect_overlays (call->fun, info, ovly_sections))
return FALSE;
}
for (call = fun->call_list; call != NULL; call = call->next)
- if (!collect_overlays (call->fun, info, ovly_sections))
+ if (!call->broken_cycle
+ && !collect_overlays (call->fun, info, ovly_sections))
return FALSE;
if (added_fun)
max = NULL;
for (call = fun->call_list; call; call = call->next)
{
+ if (call->broken_cycle)
+ continue;
if (!call->is_pasted)
has_call = TRUE;
if (!sum_stack (call->fun, info, sum_stack_param))
sum_stack_param->overall_stack = cum_stack;
htab = spu_hash_table (info);
- if (htab->auto_overlay)
+ if (htab->params->auto_overlay)
return TRUE;
f1 = func_name (fun);
- if (!fun->non_root)
- info->callbacks->info (_(" %s: 0x%v\n"), f1, (bfd_vma) cum_stack);
- info->callbacks->minfo (_("%s: 0x%v 0x%v\n"),
- f1, (bfd_vma) stack, (bfd_vma) cum_stack);
-
- if (has_call)
+ if (htab->params->stack_analysis)
{
- info->callbacks->minfo (_(" calls:\n"));
- for (call = fun->call_list; call; call = call->next)
- if (!call->is_pasted)
- {
- const char *f2 = func_name (call->fun);
- const char *ann1 = call->fun == max ? "*" : " ";
- const char *ann2 = call->is_tail ? "t" : " ";
+ if (!fun->non_root)
+ info->callbacks->info (" %s: 0x%v\n", f1, (bfd_vma) cum_stack);
+ info->callbacks->minfo ("%s: 0x%v 0x%v\n",
+ f1, (bfd_vma) stack, (bfd_vma) cum_stack);
- info->callbacks->minfo (_(" %s%s %s\n"), ann1, ann2, f2);
- }
+ if (has_call)
+ {
+ info->callbacks->minfo (_(" calls:\n"));
+ for (call = fun->call_list; call; call = call->next)
+ if (!call->is_pasted && !call->broken_cycle)
+ {
+ const char *f2 = func_name (call->fun);
+ const char *ann1 = call->fun == max ? "*" : " ";
+ const char *ann2 = call->is_tail ? "t" : " ";
+
+ info->callbacks->minfo (" %s%s %s\n", ann1, ann2, f2);
+ }
+ }
}
if (sum_stack_param->emit_stack_syms)
bfd *const *abfd1 = a;
bfd *const *abfd2 = b;
- return strcmp ((*abfd1)->filename, (*abfd2)->filename);
+ return filename_cmp (bfd_get_filename (*abfd1), bfd_get_filename (*abfd2));
}
-/* Handle --auto-overlay. */
+static unsigned int
+print_one_overlay_section (FILE *script,
+ unsigned int base,
+ unsigned int count,
+ unsigned int ovlynum,
+ unsigned int *ovly_map,
+ asection **ovly_sections,
+ struct bfd_link_info *info)
+{
+ unsigned int j;
+
+ for (j = base; j < count && ovly_map[j] == ovlynum; j++)
+ {
+ asection *sec = ovly_sections[2 * j];
+
+ if (fprintf (script, " %s%c%s (%s)\n",
+ (sec->owner->my_archive != NULL
+ ? bfd_get_filename (sec->owner->my_archive) : ""),
+ info->path_separator,
+ bfd_get_filename (sec->owner),
+ sec->name) <= 0)
+ return -1;
+ if (sec->segment_mark)
+ {
+ struct call_info *call = find_pasted_call (sec);
+ while (call != NULL)
+ {
+ struct function_info *call_fun = call->fun;
+ sec = call_fun->sec;
+ if (fprintf (script, " %s%c%s (%s)\n",
+ (sec->owner->my_archive != NULL
+ ? bfd_get_filename (sec->owner->my_archive) : ""),
+ info->path_separator,
+ bfd_get_filename (sec->owner),
+ sec->name) <= 0)
+ return -1;
+ for (call = call_fun->call_list; call; call = call->next)
+ if (call->is_pasted)
+ break;
+ }
+ }
+ }
-static void spu_elf_auto_overlay (struct bfd_link_info *, void (*) (void))
- ATTRIBUTE_NORETURN;
+ for (j = base; j < count && ovly_map[j] == ovlynum; j++)
+ {
+ asection *sec = ovly_sections[2 * j + 1];
+ if (sec != NULL
+ && fprintf (script, " %s%c%s (%s)\n",
+ (sec->owner->my_archive != NULL
+ ? bfd_get_filename (sec->owner->my_archive) : ""),
+ info->path_separator,
+ bfd_get_filename (sec->owner),
+ sec->name) <= 0)
+ return -1;
+
+ sec = ovly_sections[2 * j];
+ if (sec->segment_mark)
+ {
+ struct call_info *call = find_pasted_call (sec);
+ while (call != NULL)
+ {
+ struct function_info *call_fun = call->fun;
+ sec = call_fun->rodata;
+ if (sec != NULL
+ && fprintf (script, " %s%c%s (%s)\n",
+ (sec->owner->my_archive != NULL
+ ? bfd_get_filename (sec->owner->my_archive) : ""),
+ info->path_separator,
+ bfd_get_filename (sec->owner),
+ sec->name) <= 0)
+ return -1;
+ for (call = call_fun->call_list; call; call = call->next)
+ if (call->is_pasted)
+ break;
+ }
+ }
+ }
+
+ return j;
+}
+
+/* Handle --auto-overlay. */
static void
-spu_elf_auto_overlay (struct bfd_link_info *info,
- void (*spu_elf_load_ovl_mgr) (void))
+spu_elf_auto_overlay (struct bfd_link_info *info)
{
bfd *ibfd;
bfd **bfd_arr;
struct elf_segment_map *m;
unsigned int fixed_size, lo, hi;
+ unsigned int reserved;
struct spu_link_hash_table *htab;
unsigned int base, i, count, bfd_count;
- int ovlynum;
+ unsigned int region, ovlynum;
asection **ovly_sections, **ovly_p;
+ unsigned int *ovly_map;
FILE *script;
unsigned int total_overlay_size, overlay_size;
+ const char *ovly_mgr_entry;
struct elf_link_hash_entry *h;
struct _mos_param mos_param;
struct _uos_param uos_param;
/* Find the extents of our loadable image. */
lo = (unsigned int) -1;
hi = 0;
- for (m = elf_tdata (info->output_bfd)->segment_map; m != NULL; m = m->next)
+ for (m = elf_seg_map (info->output_bfd); m != NULL; m = m->next)
if (m->p_type == PT_LOAD)
for (i = 0; i < m->count; i++)
if (m->sections[i]->size != 0)
if (!build_call_tree (info))
goto err_exit;
+ htab = spu_hash_table (info);
+ reserved = htab->params->auto_overlay_reserved;
+ if (reserved == 0)
+ {
+ struct _sum_stack_param sum_stack_param;
+
+ sum_stack_param.emit_stack_syms = 0;
+ sum_stack_param.overall_stack = 0;
+ if (!for_each_node (sum_stack, info, &sum_stack_param, TRUE))
+ goto err_exit;
+ reserved = (sum_stack_param.overall_stack
+ + htab->params->extra_stack_space);
+ }
+
+ /* No need for overlays if everything already fits. */
+ if (fixed_size + reserved <= htab->local_store
+ && htab->params->ovly_flavour != ovly_soft_icache)
+ {
+ htab->params->auto_overlay = 0;
+ return;
+ }
+
uos_param.exclude_input_section = 0;
uos_param.exclude_output_section
= bfd_get_section_by_name (info->output_bfd, ".interrupt");
- htab = spu_hash_table (info);
- h = elf_link_hash_lookup (&htab->elf, "__ovly_load",
+ ovly_mgr_entry = "__ovly_load";
+ if (htab->params->ovly_flavour == ovly_soft_icache)
+ ovly_mgr_entry = "__icache_br_handler";
+ h = elf_link_hash_lookup (&htab->elf, ovly_mgr_entry,
FALSE, FALSE, FALSE);
if (h != NULL
&& (h->root.type == bfd_link_hash_defined
{
/* If no user overlay manager, spu_elf_load_ovl_mgr will add our
builtin version to .text, and will adjust .text size. */
- asection *text = bfd_get_section_by_name (info->output_bfd, ".text");
- if (text != NULL)
- fixed_size -= text->size;
- spu_elf_load_ovl_mgr ();
- text = bfd_get_section_by_name (info->output_bfd, ".text");
- if (text != NULL)
- fixed_size += text->size;
+ fixed_size += (*htab->params->spu_elf_load_ovl_mgr) ();
}
/* Mark overlay sections, and find max overlay section size. */
goto err_exit;
bfd_count = 0;
- for (ibfd = info->input_bfds; ibfd != NULL; ibfd = ibfd->link_next)
+ for (ibfd = info->input_bfds; ibfd != NULL; ibfd = ibfd->link.next)
++bfd_count;
bfd_arr = bfd_malloc (bfd_count * sizeof (*bfd_arr));
if (bfd_arr == NULL)
count = 0;
bfd_count = 0;
total_overlay_size = 0;
- for (ibfd = info->input_bfds; ibfd != NULL; ibfd = ibfd->link_next)
+ for (ibfd = info->input_bfds; ibfd != NULL; ibfd = ibfd->link.next)
{
- extern const bfd_target bfd_elf32_spu_vec;
+ extern const bfd_target spu_elf32_vec;
asection *sec;
unsigned int old_count;
- if (ibfd->xvec != &bfd_elf32_spu_vec)
+ if (ibfd->xvec != &spu_elf32_vec)
continue;
old_count = count;
fixed_size -= sec->size;
total_overlay_size += sec->size;
}
+ else if ((sec->flags & (SEC_ALLOC | SEC_LOAD)) == (SEC_ALLOC | SEC_LOAD)
+ && sec->output_section->owner == info->output_bfd
+ && strncmp (sec->output_section->name, ".ovl.init", 9) == 0)
+ fixed_size -= sec->size;
if (count != old_count)
bfd_arr[bfd_count++] = ibfd;
}
qsort (bfd_arr, bfd_count, sizeof (*bfd_arr), sort_bfds);
for (i = 1; i < bfd_count; ++i)
- if (strcmp (bfd_arr[i - 1]->filename, bfd_arr[i]->filename) == 0)
+ if (filename_cmp (bfd_get_filename (bfd_arr[i - 1]),
+ bfd_get_filename (bfd_arr[i])) == 0)
{
if (bfd_arr[i - 1]->my_archive == bfd_arr[i]->my_archive)
{
if (bfd_arr[i - 1]->my_archive && bfd_arr[i]->my_archive)
+ /* xgettext:c-format */
info->callbacks->einfo (_("%s duplicated in %s\n"),
- bfd_arr[i]->filename,
- bfd_arr[i]->my_archive->filename);
+ bfd_get_filename (bfd_arr[i]),
+ bfd_get_filename (bfd_arr[i]->my_archive));
else
info->callbacks->einfo (_("%s duplicated\n"),
- bfd_arr[i]->filename);
+ bfd_get_filename (bfd_arr[i]));
ok = FALSE;
}
}
}
free (bfd_arr);
- if (htab->reserved == 0)
- {
- struct _sum_stack_param sum_stack_param;
-
- sum_stack_param.emit_stack_syms = 0;
- sum_stack_param.overall_stack = 0;
- if (!for_each_node (sum_stack, info, &sum_stack_param, TRUE))
- goto err_exit;
- htab->reserved = sum_stack_param.overall_stack + htab->extra_stack_space;
- }
- fixed_size += htab->reserved;
- fixed_size += htab->non_ovly_stub * OVL_STUB_SIZE;
+ fixed_size += reserved;
+ fixed_size += htab->non_ovly_stub * ovl_stub_size (htab->params);
if (fixed_size + mos_param.max_overlay_size <= htab->local_store)
{
- /* Guess number of overlays. Assuming overlay buffer is on
- average only half full should be conservative. */
- ovlynum = total_overlay_size * 2 / (htab->local_store - fixed_size);
- /* Space for _ovly_table[], _ovly_buf_table[] and toe. */
- fixed_size += ovlynum * 16 + 16 + 4 + 16;
+ if (htab->params->ovly_flavour == ovly_soft_icache)
+ {
+ /* Stubs in the non-icache area are bigger. */
+ fixed_size += htab->non_ovly_stub * 16;
+ /* Space for icache manager tables.
+ a) Tag array, one quadword per cache line.
+ - word 0: ia address of present line, init to zero. */
+ fixed_size += 16 << htab->num_lines_log2;
+ /* b) Rewrite "to" list, one quadword per cache line. */
+ fixed_size += 16 << htab->num_lines_log2;
+ /* c) Rewrite "from" list, one byte per outgoing branch (rounded up
+ to a power-of-two number of full quadwords) per cache line. */
+ fixed_size += 16 << (htab->fromelem_size_log2
+ + htab->num_lines_log2);
+ /* d) Pointer to __ea backing store (toe), 1 quadword. */
+ fixed_size += 16;
+ }
+ else
+ {
+ /* Guess number of overlays. Assuming overlay buffer is on
+ average only half full should be conservative. */
+ ovlynum = (total_overlay_size * 2 * htab->params->num_lines
+ / (htab->local_store - fixed_size));
+ /* Space for _ovly_table[], _ovly_buf_table[] and toe. */
+ fixed_size += ovlynum * 16 + 16 + 4 + 16;
+ }
}
if (fixed_size + mos_param.max_overlay_size > htab->local_store)
+ /* xgettext:c-format */
info->callbacks->einfo (_("non-overlay size of 0x%v plus maximum overlay "
"size of 0x%v exceeds local store\n"),
(bfd_vma) fixed_size,
(bfd_vma) mos_param.max_overlay_size);
/* Now see if we should put some functions in the non-overlay area. */
- if (fixed_size < htab->overlay_fixed
- && htab->overlay_fixed + mos_param.max_overlay_size < htab->local_store)
+ else if (fixed_size < htab->params->auto_overlay_fixed)
{
- unsigned int lib_size = htab->overlay_fixed - fixed_size;
+ unsigned int max_fixed, lib_size;
+
+ max_fixed = htab->local_store - mos_param.max_overlay_size;
+ if (max_fixed > htab->params->auto_overlay_fixed)
+ max_fixed = htab->params->auto_overlay_fixed;
+ lib_size = max_fixed - fixed_size;
lib_size = auto_ovl_lib_functions (info, lib_size);
if (lib_size == (unsigned int) -1)
goto err_exit;
- fixed_size = htab->overlay_fixed - lib_size;
+ fixed_size = max_fixed - lib_size;
}
/* Build an array of sections, suitably sorted to place into
if (!for_each_node (collect_overlays, info, &ovly_p, TRUE))
goto err_exit;
count = (size_t) (ovly_p - ovly_sections) / 2;
-
- script = htab->spu_elf_open_overlay_script ();
-
- if (fprintf (script, "SECTIONS\n{\n OVERLAY :\n {\n") <= 0)
- goto file_err;
+ ovly_map = bfd_malloc (count * sizeof (*ovly_map));
+ if (ovly_map == NULL)
+ goto err_exit;
memset (&dummy_caller, 0, sizeof (dummy_caller));
- overlay_size = htab->local_store - fixed_size;
+ overlay_size = (htab->local_store - fixed_size) / htab->params->num_lines;
+ if (htab->params->line_size != 0)
+ overlay_size = htab->params->line_size;
base = 0;
ovlynum = 0;
while (base < count)
{
- unsigned int size = 0;
- unsigned int j;
+ unsigned int size = 0, rosize = 0, roalign = 0;
for (i = base; i < count; i++)
{
- asection *sec;
- unsigned int tmp;
- unsigned int stub_size;
+ asection *sec, *rosec;
+ unsigned int tmp, rotmp;
+ unsigned int num_stubs;
struct call_info *call, *pasty;
struct _spu_elf_section_data *sec_data;
struct spu_elf_stack_info *sinfo;
- int k;
+ unsigned int k;
/* See whether we can add this section to the current
overlay without overflowing our overlay buffer. */
sec = ovly_sections[2 * i];
- tmp = size + sec->size;
- if (ovly_sections[2 * i + 1])
- tmp += ovly_sections[2 * i + 1]->size;
- if (tmp > overlay_size)
+ tmp = align_power (size, sec->alignment_power) + sec->size;
+ rotmp = rosize;
+ rosec = ovly_sections[2 * i + 1];
+ if (rosec != NULL)
+ {
+ rotmp = align_power (rotmp, rosec->alignment_power) + rosec->size;
+ if (roalign < rosec->alignment_power)
+ roalign = rosec->alignment_power;
+ }
+ if (align_power (tmp, roalign) + rotmp > overlay_size)
break;
if (sec->segment_mark)
{
/* Pasted sections must stay together, so add their
sizes too. */
- struct call_info *pasty = find_pasted_call (sec);
+ pasty = find_pasted_call (sec);
while (pasty != NULL)
{
struct function_info *call_fun = pasty->fun;
- tmp += call_fun->sec->size;
+ tmp = (align_power (tmp, call_fun->sec->alignment_power)
+ + call_fun->sec->size);
if (call_fun->rodata)
- tmp += call_fun->rodata->size;
+ {
+ rotmp = (align_power (rotmp,
+ call_fun->rodata->alignment_power)
+ + call_fun->rodata->size);
+ if (roalign < rosec->alignment_power)
+ roalign = rosec->alignment_power;
+ }
for (pasty = call_fun->call_list; pasty; pasty = pasty->next)
if (pasty->is_pasted)
break;
}
}
- if (tmp > overlay_size)
+ if (align_power (tmp, roalign) + rotmp > overlay_size)
break;
/* If we add this section, we might need new overlay call
pasty = NULL;
sec_data = spu_elf_section_data (sec);
sinfo = sec_data->u.i.stack_info;
- for (k = 0; k < sinfo->num_fun; ++k)
+ for (k = 0; k < (unsigned) sinfo->num_fun; ++k)
for (call = sinfo->fun[k].call_list; call; call = call->next)
if (call->is_pasted)
{
}
/* Calculate call stub size. */
- stub_size = 0;
+ num_stubs = 0;
for (call = dummy_caller.call_list; call; call = call->next)
{
- unsigned int k;
+ unsigned int stub_delta = 1;
+
+ if (htab->params->ovly_flavour == ovly_soft_icache)
+ stub_delta = call->count;
+ num_stubs += stub_delta;
- stub_size += OVL_STUB_SIZE;
/* If the call is within this overlay, we won't need a
stub. */
for (k = base; k < i + 1; k++)
if (call->fun->sec == ovly_sections[2 * k])
{
- stub_size -= OVL_STUB_SIZE;
+ num_stubs -= stub_delta;
break;
}
}
- if (tmp + stub_size > overlay_size)
+ if (htab->params->ovly_flavour == ovly_soft_icache
+ && num_stubs > htab->params->max_branch)
+ break;
+ if (align_power (tmp, roalign) + rotmp
+ + num_stubs * ovl_stub_size (htab->params) > overlay_size)
break;
-
size = tmp;
+ rosize = rotmp;
}
if (i == base)
{
- info->callbacks->einfo (_("%B:%A%s exceeds overlay size\n"),
+ /* xgettext:c-format */
+ info->callbacks->einfo (_("%pB:%pA%s exceeds overlay size\n"),
ovly_sections[2 * i]->owner,
ovly_sections[2 * i],
ovly_sections[2 * i + 1] ? " + rodata" : "");
goto err_exit;
}
- if (fprintf (script, " .ovly%d {\n", ++ovlynum) <= 0)
- goto file_err;
- for (j = base; j < i; j++)
+ while (dummy_caller.call_list != NULL)
{
- asection *sec = ovly_sections[2 * j];
-
- if (fprintf (script, " %s%c%s (%s)\n",
- (sec->owner->my_archive != NULL
- ? sec->owner->my_archive->filename : ""),
- info->path_separator,
- sec->owner->filename,
- sec->name) <= 0)
- goto file_err;
- if (sec->segment_mark)
- {
- struct call_info *call = find_pasted_call (sec);
- while (call != NULL)
- {
- struct function_info *call_fun = call->fun;
- sec = call_fun->sec;
- if (fprintf (script, " %s%c%s (%s)\n",
- (sec->owner->my_archive != NULL
- ? sec->owner->my_archive->filename : ""),
- info->path_separator,
- sec->owner->filename,
- sec->name) <= 0)
- goto file_err;
- for (call = call_fun->call_list; call; call = call->next)
- if (call->is_pasted)
- break;
- }
- }
+ struct call_info *call = dummy_caller.call_list;
+ dummy_caller.call_list = call->next;
+ free (call);
}
- for (j = base; j < i; j++)
+ ++ovlynum;
+ while (base < i)
+ ovly_map[base++] = ovlynum;
+ }
+
+ script = htab->params->spu_elf_open_overlay_script ();
+
+ if (htab->params->ovly_flavour == ovly_soft_icache)
+ {
+ if (fprintf (script, "SECTIONS\n{\n") <= 0)
+ goto file_err;
+
+ if (fprintf (script,
+ " . = ALIGN (%u);\n"
+ " .ovl.init : { *(.ovl.init) }\n"
+ " . = ABSOLUTE (ADDR (.ovl.init));\n",
+ htab->params->line_size) <= 0)
+ goto file_err;
+
+ base = 0;
+ ovlynum = 1;
+ while (base < count)
{
- asection *sec = ovly_sections[2 * j + 1];
- if (sec != NULL
- && fprintf (script, " %s%c%s (%s)\n",
- (sec->owner->my_archive != NULL
- ? sec->owner->my_archive->filename : ""),
- info->path_separator,
- sec->owner->filename,
- sec->name) <= 0)
+ unsigned int indx = ovlynum - 1;
+ unsigned int vma, lma;
+
+ vma = (indx & (htab->params->num_lines - 1)) << htab->line_size_log2;
+ lma = vma + (((indx >> htab->num_lines_log2) + 1) << 18);
+
+ if (fprintf (script, " .ovly%u ABSOLUTE (ADDR (.ovl.init)) + %u "
+ ": AT (LOADADDR (.ovl.init) + %u) {\n",
+ ovlynum, vma, lma) <= 0)
goto file_err;
- sec = ovly_sections[2 * j];
- if (sec->segment_mark)
- {
- struct call_info *call = find_pasted_call (sec);
- while (call != NULL)
- {
- struct function_info *call_fun = call->fun;
- sec = call_fun->rodata;
- if (sec != NULL
- && fprintf (script, " %s%c%s (%s)\n",
- (sec->owner->my_archive != NULL
- ? sec->owner->my_archive->filename : ""),
- info->path_separator,
- sec->owner->filename,
- sec->name) <= 0)
- goto file_err;
- for (call = call_fun->call_list; call; call = call->next)
- if (call->is_pasted)
- break;
- }
- }
+ base = print_one_overlay_section (script, base, count, ovlynum,
+ ovly_map, ovly_sections, info);
+ if (base == (unsigned) -1)
+ goto file_err;
+
+ if (fprintf (script, " }\n") <= 0)
+ goto file_err;
+
+ ovlynum++;
}
- if (fprintf (script, " }\n") <= 0)
+ if (fprintf (script, " . = ABSOLUTE (ADDR (.ovl.init)) + %u;\n",
+ 1 << (htab->num_lines_log2 + htab->line_size_log2)) <= 0)
goto file_err;
- while (dummy_caller.call_list != NULL)
+ if (fprintf (script, "}\nINSERT AFTER .toe;\n") <= 0)
+ goto file_err;
+ }
+ else
+ {
+ if (fprintf (script, "SECTIONS\n{\n") <= 0)
+ goto file_err;
+
+ if (fprintf (script,
+ " . = ALIGN (16);\n"
+ " .ovl.init : { *(.ovl.init) }\n"
+ " . = ABSOLUTE (ADDR (.ovl.init));\n") <= 0)
+ goto file_err;
+
+ for (region = 1; region <= htab->params->num_lines; region++)
{
- struct call_info *call = dummy_caller.call_list;
- dummy_caller.call_list = call->next;
- free (call);
+ ovlynum = region;
+ base = 0;
+ while (base < count && ovly_map[base] < ovlynum)
+ base++;
+
+ if (base == count)
+ break;
+
+ if (region == 1)
+ {
+ /* We need to set lma since we are overlaying .ovl.init. */
+ if (fprintf (script,
+ " OVERLAY : AT (ALIGN (LOADADDR (.ovl.init) + SIZEOF (.ovl.init), 16))\n {\n") <= 0)
+ goto file_err;
+ }
+ else
+ {
+ if (fprintf (script, " OVERLAY :\n {\n") <= 0)
+ goto file_err;
+ }
+
+ while (base < count)
+ {
+ if (fprintf (script, " .ovly%u {\n", ovlynum) <= 0)
+ goto file_err;
+
+ base = print_one_overlay_section (script, base, count, ovlynum,
+ ovly_map, ovly_sections, info);
+ if (base == (unsigned) -1)
+ goto file_err;
+
+ if (fprintf (script, " }\n") <= 0)
+ goto file_err;
+
+ ovlynum += htab->params->num_lines;
+ while (base < count && ovly_map[base] < ovlynum)
+ base++;
+ }
+
+ if (fprintf (script, " }\n") <= 0)
+ goto file_err;
}
- base = i;
+ if (fprintf (script, "}\nINSERT BEFORE .text;\n") <= 0)
+ goto file_err;
}
+
+ free (ovly_map);
free (ovly_sections);
- if (fprintf (script, " }\n}\nINSERT AFTER .text;\n") <= 0)
- goto file_err;
if (fclose (script) != 0)
goto file_err;
- if (htab->auto_overlay & AUTO_RELINK)
- htab->spu_elf_relink ();
+ if (htab->params->auto_overlay & AUTO_RELINK)
+ (*htab->params->spu_elf_relink) ();
xexit (0);
file_err:
bfd_set_error (bfd_error_system_call);
err_exit:
- info->callbacks->einfo ("%F%P: auto overlay error: %E\n");
+ info->callbacks->einfo (_("%F%P: auto overlay error: %E\n"));
xexit (1);
}
/* Provide an estimate of total stack required. */
static bfd_boolean
-spu_elf_stack_analysis (struct bfd_link_info *info, int emit_stack_syms)
+spu_elf_stack_analysis (struct bfd_link_info *info)
{
+ struct spu_link_hash_table *htab;
struct _sum_stack_param sum_stack_param;
if (!discover_functions (info))
if (!build_call_tree (info))
return FALSE;
- info->callbacks->info (_("Stack size for call graph root nodes.\n"));
- info->callbacks->minfo (_("\nStack size for functions. "
- "Annotations: '*' max stack, 't' tail call\n"));
+ htab = spu_hash_table (info);
+ if (htab->params->stack_analysis)
+ {
+ info->callbacks->info (_("Stack size for call graph root nodes.\n"));
+ info->callbacks->minfo (_("\nStack size for functions. "
+ "Annotations: '*' max stack, 't' tail call\n"));
+ }
- sum_stack_param.emit_stack_syms = emit_stack_syms;
+ sum_stack_param.emit_stack_syms = htab->params->emit_stack_syms;
sum_stack_param.overall_stack = 0;
if (!for_each_node (sum_stack, info, &sum_stack_param, TRUE))
return FALSE;
- info->callbacks->info (_("Maximum stack required is 0x%v\n"),
- (bfd_vma) sum_stack_param.overall_stack);
+ if (htab->params->stack_analysis)
+ info->callbacks->info (_("Maximum stack required is 0x%v\n"),
+ (bfd_vma) sum_stack_param.overall_stack);
return TRUE;
}
{
struct spu_link_hash_table *htab = spu_hash_table (info);
- if (htab->auto_overlay)
- spu_elf_auto_overlay (info, htab->spu_elf_load_ovl_mgr);
+ if (htab->params->auto_overlay)
+ spu_elf_auto_overlay (info);
- if (htab->stack_analysis
- && !spu_elf_stack_analysis (info, htab->emit_stack_syms))
- info->callbacks->einfo ("%X%P: stack analysis error: %E\n");
+ if ((htab->params->stack_analysis
+ || (htab->params->ovly_flavour == ovly_soft_icache
+ && htab->params->lrlive_analysis))
+ && !spu_elf_stack_analysis (info))
+ info->callbacks->einfo (_("%X%P: stack/lrlive analysis error: %E\n"));
+
+ if (!spu_elf_build_stubs (info))
+ info->callbacks->einfo (_("%F%P: can not build overlay stubs: %E\n"));
return bfd_elf_final_link (output_bfd, info);
}
-/* Called when not normally emitting relocs, ie. !info->relocatable
+/* Called when not normally emitting relocs, ie. !bfd_link_relocatable (info)
and !info->emitrelocations. Returns a count of special relocs
that need to be emitted. */
static unsigned int
-spu_elf_count_relocs (asection *sec, Elf_Internal_Rela *relocs)
+spu_elf_count_relocs (struct bfd_link_info *info, asection *sec)
{
+ Elf_Internal_Rela *relocs;
unsigned int count = 0;
- Elf_Internal_Rela *relend = relocs + sec->reloc_count;
- for (; relocs < relend; relocs++)
+ relocs = _bfd_elf_link_read_relocs (sec->owner, sec, NULL, NULL,
+ info->keep_memory);
+ if (relocs != NULL)
{
- int r_type = ELF32_R_TYPE (relocs->r_info);
- if (r_type == R_SPU_PPU32 || r_type == R_SPU_PPU64)
- ++count;
+ Elf_Internal_Rela *rel;
+ Elf_Internal_Rela *relend = relocs + sec->reloc_count;
+
+ for (rel = relocs; rel < relend; rel++)
+ {
+ int r_type = ELF32_R_TYPE (rel->r_info);
+ if (r_type == R_SPU_PPU32 || r_type == R_SPU_PPU64)
+ ++count;
+ }
+
+ if (elf_section_data (sec)->relocs != relocs)
+ free (relocs);
}
return count;
}
+/* Functions for adding fixup records to .fixup */
+
+#define FIXUP_RECORD_SIZE 4
+
+#define FIXUP_PUT(output_bfd,htab,index,addr) \
+ bfd_put_32 (output_bfd, addr, \
+ htab->sfixup->contents + FIXUP_RECORD_SIZE * (index))
+#define FIXUP_GET(output_bfd,htab,index) \
+ bfd_get_32 (output_bfd, \
+ htab->sfixup->contents + FIXUP_RECORD_SIZE * (index))
+
+/* Store OFFSET in .fixup. This assumes it will be called with an
+ increasing OFFSET. When this OFFSET fits with the last base offset,
+ it just sets a bit, otherwise it adds a new fixup record. */
+static void
+spu_elf_emit_fixup (bfd * output_bfd, struct bfd_link_info *info,
+ bfd_vma offset)
+{
+ struct spu_link_hash_table *htab = spu_hash_table (info);
+ asection *sfixup = htab->sfixup;
+ bfd_vma qaddr = offset & ~(bfd_vma) 15;
+ bfd_vma bit = ((bfd_vma) 8) >> ((offset & 15) >> 2);
+ if (sfixup->reloc_count == 0)
+ {
+ FIXUP_PUT (output_bfd, htab, 0, qaddr | bit);
+ sfixup->reloc_count++;
+ }
+ else
+ {
+ bfd_vma base = FIXUP_GET (output_bfd, htab, sfixup->reloc_count - 1);
+ if (qaddr != (base & ~(bfd_vma) 15))
+ {
+ if ((sfixup->reloc_count + 1) * FIXUP_RECORD_SIZE > sfixup->size)
+ _bfd_error_handler (_("fatal error while creating .fixup"));
+ FIXUP_PUT (output_bfd, htab, sfixup->reloc_count, qaddr | bit);
+ sfixup->reloc_count++;
+ }
+ else
+ FIXUP_PUT (output_bfd, htab, sfixup->reloc_count - 1, base | bit);
+ }
+}
+
/* Apply RELOCS to CONTENTS of INPUT_SECTION from INPUT_BFD. */
static int
struct elf_link_hash_entry **sym_hashes;
Elf_Internal_Rela *rel, *relend;
struct spu_link_hash_table *htab;
- asection *ea = bfd_get_section_by_name (output_bfd, "._ea");
+ asection *ea;
int ret = TRUE;
bfd_boolean emit_these_relocs = FALSE;
bfd_boolean is_ea_sym;
bfd_boolean stubs;
+ unsigned int iovl = 0;
htab = spu_hash_table (info);
stubs = (htab->stub_sec != NULL
- && maybe_needs_stubs (input_section, output_bfd));
+ && maybe_needs_stubs (input_section));
+ iovl = overlay_index (input_section);
+ ea = bfd_get_section_by_name (output_bfd, "._ea");
symtab_hdr = &elf_tdata (input_bfd)->symtab_hdr;
sym_hashes = (struct elf_link_hash_entry **) (elf_sym_hashes (input_bfd));
bfd_vma addend;
bfd_reloc_status_type r;
bfd_boolean unresolved_reloc;
- bfd_boolean warned;
enum _stub_type stub_type;
r_symndx = ELF32_R_SYM (rel->r_info);
r_type = ELF32_R_TYPE (rel->r_info);
howto = elf_howto_table + r_type;
unresolved_reloc = FALSE;
- warned = FALSE;
h = NULL;
sym = NULL;
sec = NULL;
h = sym_hashes[r_symndx - symtab_hdr->sh_info];
+ if (info->wrap_hash != NULL
+ && (input_section->flags & SEC_DEBUGGING) != 0)
+ h = ((struct elf_link_hash_entry *)
+ unwrap_hash_lookup (info, input_bfd, &h->root));
+
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;
else if (info->unresolved_syms_in_objects == RM_IGNORE
&& ELF_ST_VISIBILITY (h->other) == STV_DEFAULT)
;
- else if (!info->relocatable
+ else if (!bfd_link_relocatable (info)
&& !(r_type == R_SPU_PPU32 || r_type == R_SPU_PPU64))
{
bfd_boolean err;
- err = (info->unresolved_syms_in_objects == RM_GENERATE_ERROR
- || ELF_ST_VISIBILITY (h->other) != STV_DEFAULT);
- if (!info->callbacks->undefined_symbol (info,
- h->root.root.string,
- input_bfd,
- input_section,
- rel->r_offset, err))
- return FALSE;
- warned = TRUE;
+
+ err = (info->unresolved_syms_in_objects == RM_DIAGNOSE
+ && !info->warn_unresolved_syms)
+ || ELF_ST_VISIBILITY (h->other) != STV_DEFAULT;
+
+ info->callbacks->undefined_symbol
+ (info, h->root.root.string, input_bfd,
+ input_section, rel->r_offset, err);
}
sym_name = h->root.root.string;
}
- if (sec != NULL && elf_discarded_section (sec))
- {
- /* For relocs against symbols from removed linkonce sections,
- or sections discarded by a linker script, we just want the
- section contents zeroed. Avoid any special processing. */
- _bfd_clear_contents (howto, input_bfd, contents + rel->r_offset);
- rel->r_info = 0;
- rel->r_addend = 0;
- continue;
- }
+ if (sec != NULL && discarded_section (sec))
+ RELOC_AGAINST_DISCARDED_SECTION (info, input_bfd, input_section,
+ rel, 1, relend, howto, 0, contents);
- if (info->relocatable)
+ if (bfd_link_relocatable (info))
continue;
- is_ea_sym = (ea != NULL
- && sec != NULL
- && sec->output_section == ea);
-
- if (r_type == R_SPU_PPU32 || r_type == R_SPU_PPU64)
+ /* Change "a rt,ra,rb" to "ai rt,ra,0". */
+ if (r_type == R_SPU_ADD_PIC
+ && h != NULL
+ && !(h->def_regular || ELF_COMMON_DEF_P (h)))
{
- if (is_ea_sym)
- {
- /* ._ea is a special section that isn't allocated in SPU
- memory, but rather occupies space in PPU memory as
- part of an embedded ELF image. If this reloc is
- against a symbol defined in ._ea, then transform the
- reloc into an equivalent one without a symbol
- relative to the start of the ELF image. */
- rel->r_addend += (relocation
- - ea->vma
- + elf_section_data (ea)->this_hdr.sh_offset);
- rel->r_info = ELF32_R_INFO (0, r_type);
- }
- emit_these_relocs = TRUE;
- continue;
+ bfd_byte *loc = contents + rel->r_offset;
+ loc[0] = 0x1c;
+ loc[1] = 0x00;
+ loc[2] &= 0x3f;
}
- if (is_ea_sym)
- unresolved_reloc = TRUE;
-
- if (unresolved_reloc)
- {
- (*_bfd_error_handler)
- (_("%B(%s+0x%lx): unresolvable %s relocation against symbol `%s'"),
- input_bfd,
- bfd_get_section_name (input_bfd, input_section),
- (long) rel->r_offset,
- howto->name,
- sym_name);
- ret = FALSE;
- }
+ is_ea_sym = (ea != NULL
+ && sec != NULL
+ && sec->output_section == ea);
/* If this symbol is in an overlay area, we may need to relocate
to the overlay stub. */
addend = rel->r_addend;
if (stubs
+ && !is_ea_sym
&& (stub_type = needs_ovl_stub (h, sym, sec, input_section, rel,
contents, info)) != no_stub)
{
struct got_entry *g, **head;
if (stub_type != nonovl_stub)
- ovl = (spu_elf_section_data (input_section->output_section)
- ->u.o.ovl_index);
+ ovl = iovl;
if (h != NULL)
head = &h->got.glist;
head = elf_local_got_ents (input_bfd) + r_symndx;
for (g = *head; g != NULL; g = g->next)
- if (g->addend == addend && (g->ovl == ovl || g->ovl == 0))
+ if (htab->params->ovly_flavour == ovly_soft_icache
+ ? (g->ovl == ovl
+ && g->br_addr == (rel->r_offset
+ + input_section->output_offset
+ + input_section->output_section->vma))
+ : g->addend == addend && (g->ovl == ovl || g->ovl == 0))
break;
if (g == NULL)
abort ();
relocation = g->stub_addr;
addend = 0;
}
+ else
+ {
+ /* For soft icache, encode the overlay index into addresses. */
+ if (htab->params->ovly_flavour == ovly_soft_icache
+ && (r_type == R_SPU_ADDR16_HI
+ || r_type == R_SPU_ADDR32 || r_type == R_SPU_REL32)
+ && !is_ea_sym)
+ {
+ unsigned int ovl = overlay_index (sec);
+ if (ovl != 0)
+ {
+ unsigned int set_id = ((ovl - 1) >> htab->num_lines_log2) + 1;
+ relocation += set_id << 18;
+ }
+ }
+ }
+
+ if (htab->params->emit_fixups && !bfd_link_relocatable (info)
+ && (input_section->flags & SEC_ALLOC) != 0
+ && r_type == R_SPU_ADDR32)
+ {
+ bfd_vma offset;
+ offset = rel->r_offset + input_section->output_section->vma
+ + input_section->output_offset;
+ spu_elf_emit_fixup (output_bfd, info, offset);
+ }
+
+ if (unresolved_reloc)
+ ;
+ else if (r_type == R_SPU_PPU32 || r_type == R_SPU_PPU64)
+ {
+ if (is_ea_sym)
+ {
+ /* ._ea is a special section that isn't allocated in SPU
+ memory, but rather occupies space in PPU memory as
+ part of an embedded ELF image. If this reloc is
+ against a symbol defined in ._ea, then transform the
+ reloc into an equivalent one without a symbol
+ relative to the start of the ELF image. */
+ rel->r_addend += (relocation
+ - ea->vma
+ + elf_section_data (ea)->this_hdr.sh_offset);
+ rel->r_info = ELF32_R_INFO (0, r_type);
+ }
+ emit_these_relocs = TRUE;
+ continue;
+ }
+ else if (is_ea_sym)
+ unresolved_reloc = TRUE;
+
+ if (unresolved_reloc
+ && _bfd_elf_section_offset (output_bfd, info, input_section,
+ rel->r_offset) != (bfd_vma) -1)
+ {
+ _bfd_error_handler
+ /* xgettext:c-format */
+ (_("%pB(%s+%#" PRIx64 "): "
+ "unresolvable %s relocation against symbol `%s'"),
+ input_bfd,
+ bfd_section_name (input_section),
+ (uint64_t) rel->r_offset,
+ howto->name,
+ sym_name);
+ ret = FALSE;
+ }
r = _bfd_final_link_relocate (howto,
input_bfd,
switch (r)
{
case bfd_reloc_overflow:
- if (!((*info->callbacks->reloc_overflow)
- (info, (h ? &h->root : NULL), sym_name, howto->name,
- (bfd_vma) 0, input_bfd, input_section, rel->r_offset)))
- return FALSE;
+ (*info->callbacks->reloc_overflow)
+ (info, (h ? &h->root : NULL), sym_name, howto->name,
+ (bfd_vma) 0, input_bfd, input_section, rel->r_offset);
break;
case bfd_reloc_undefined:
- if (!((*info->callbacks->undefined_symbol)
- (info, sym_name, input_bfd, input_section,
- rel->r_offset, TRUE)))
- return FALSE;
+ (*info->callbacks->undefined_symbol)
+ (info, sym_name, input_bfd, input_section, rel->r_offset, TRUE);
break;
case bfd_reloc_outofrange:
common_error:
ret = FALSE;
- if (!((*info->callbacks->warning)
- (info, msg, sym_name, input_bfd, input_section,
- rel->r_offset)))
- return FALSE;
+ (*info->callbacks->warning) (info, msg, sym_name, input_bfd,
+ input_section, rel->r_offset);
break;
}
}
}
input_section->reloc_count = wrel - relocs;
/* Backflips for _bfd_elf_link_output_relocs. */
- rel_hdr = &elf_section_data (input_section)->rel_hdr;
+ rel_hdr = _bfd_elf_single_rel_hdr (input_section);
rel_hdr->sh_size = input_section->reloc_count * rel_hdr->sh_entsize;
ret = 2;
}
return ret;
}
+static bfd_boolean
+spu_elf_finish_dynamic_sections (bfd *output_bfd ATTRIBUTE_UNUSED,
+ struct bfd_link_info *info ATTRIBUTE_UNUSED)
+{
+ return TRUE;
+}
+
/* Adjust _SPUEAR_ syms to point at their overlay stubs. */
-static bfd_boolean
+static int
spu_elf_output_symbol_hook (struct bfd_link_info *info,
const char *sym_name ATTRIBUTE_UNUSED,
Elf_Internal_Sym *sym,
{
struct spu_link_hash_table *htab = spu_hash_table (info);
- if (!info->relocatable
+ if (!bfd_link_relocatable (info)
&& htab->stub_sec != NULL
&& h != NULL
&& (h->root.type == bfd_link_hash_defined
struct got_entry *g;
for (g = h->got.glist; g != NULL; g = g->next)
- if (g->addend == 0 && g->ovl == 0)
+ if (htab->params->ovly_flavour == ovly_soft_icache
+ ? g->br_addr == g->stub_addr
+ : g->addend == 0 && g->ovl == 0)
{
sym->st_shndx = (_bfd_elf_section_from_bfd_section
(htab->stub_sec[0]->output_section->owner,
}
}
- return TRUE;
+ return 1;
}
static int spu_plugin = 0;
/* Set ELF header e_type for plugins. */
-static void
-spu_elf_post_process_headers (bfd *abfd,
- struct bfd_link_info *info ATTRIBUTE_UNUSED)
+static bfd_boolean
+spu_elf_init_file_header (bfd *abfd, struct bfd_link_info *info)
{
+ if (!_bfd_elf_init_file_header (abfd, info))
+ return FALSE;
+
if (spu_plugin)
{
Elf_Internal_Ehdr *i_ehdrp = elf_elfheader (abfd);
i_ehdrp->e_type = ET_DYN;
}
+ return TRUE;
}
/* We may add an extra PT_LOAD segment for .toe. We also need extra
static int
spu_elf_additional_program_headers (bfd *abfd, struct bfd_link_info *info)
{
- struct spu_link_hash_table *htab = spu_hash_table (info);
- int extra = htab->num_overlays;
+ int extra = 0;
asection *sec;
+ if (info != NULL)
+ {
+ struct spu_link_hash_table *htab = spu_hash_table (info);
+ extra = htab->num_overlays;
+ }
+
if (extra)
++extra;
spu_elf_modify_segment_map (bfd *abfd, struct bfd_link_info *info)
{
asection *toe, *s;
- struct elf_segment_map *m;
+ struct elf_segment_map *m, *m_overlay;
+ struct elf_segment_map **p, **p_overlay, **first_load;
unsigned int i;
if (info == NULL)
return TRUE;
toe = bfd_get_section_by_name (abfd, ".toe");
- for (m = elf_tdata (abfd)->segment_map; m != NULL; m = m->next)
+ for (m = elf_seg_map (abfd); m != NULL; m = m->next)
if (m->p_type == PT_LOAD && m->count > 1)
for (i = 0; i < m->count; i++)
if ((s = m->sections[i]) == toe
break;
}
+
+ /* Some SPU ELF loaders ignore the PF_OVERLAY flag and just load all
+ PT_LOAD segments. This can cause the .ovl.init section to be
+ overwritten with the contents of some overlay segment. To work
+ around this issue, we ensure that all PF_OVERLAY segments are
+ sorted first amongst the program headers; this ensures that even
+ with a broken loader, the .ovl.init section (which is not marked
+ as PF_OVERLAY) will be placed into SPU local store on startup. */
+
+ /* Move all overlay segments onto a separate list. */
+ p = &elf_seg_map (abfd);
+ p_overlay = &m_overlay;
+ m_overlay = NULL;
+ first_load = NULL;
+ while (*p != NULL)
+ {
+ if ((*p)->p_type == PT_LOAD)
+ {
+ if (!first_load)
+ first_load = p;
+ if ((*p)->count == 1
+ && spu_elf_section_data ((*p)->sections[0])->u.o.ovl_index != 0)
+ {
+ m = *p;
+ m->no_sort_lma = 1;
+ *p = m->next;
+ *p_overlay = m;
+ p_overlay = &m->next;
+ continue;
+ }
+ }
+ p = &((*p)->next);
+ }
+
+ /* Re-insert overlay segments at the head of the segment map. */
+ if (m_overlay != NULL)
+ {
+ p = first_load;
+ if (*p != NULL && (*p)->p_type == PT_LOAD && (*p)->includes_filehdr)
+ /* It doesn't really make sense for someone to include the ELF
+ file header into an spu image, but if they do the code that
+ assigns p_offset needs to see the segment containing the
+ header first. */
+ p = &(*p)->next;
+ *p_overlay = *p;
+ *p = m_overlay;
+ }
+
return TRUE;
}
/* Tweak phdrs before writing them out. */
static int
-spu_elf_modify_program_headers (bfd *abfd, struct bfd_link_info *info)
+spu_elf_modify_headers (bfd *abfd, struct bfd_link_info *info)
{
- const struct elf_backend_data *bed;
- struct elf_obj_tdata *tdata;
- Elf_Internal_Phdr *phdr, *last;
- struct spu_link_hash_table *htab;
- unsigned int count;
- unsigned int i;
+ if (info != NULL)
+ {
+ const struct elf_backend_data *bed;
+ struct elf_obj_tdata *tdata;
+ Elf_Internal_Phdr *phdr, *last;
+ struct spu_link_hash_table *htab;
+ unsigned int count;
+ unsigned int i;
+
+ bed = get_elf_backend_data (abfd);
+ tdata = elf_tdata (abfd);
+ phdr = tdata->phdr;
+ count = elf_program_header_size (abfd) / bed->s->sizeof_phdr;
+ htab = spu_hash_table (info);
+ if (htab->num_overlays != 0)
+ {
+ struct elf_segment_map *m;
+ unsigned int o;
- if (info == NULL)
- return TRUE;
+ for (i = 0, m = elf_seg_map (abfd); m; ++i, m = m->next)
+ if (m->count != 0
+ && ((o = spu_elf_section_data (m->sections[0])->u.o.ovl_index)
+ != 0))
+ {
+ /* Mark this as an overlay header. */
+ phdr[i].p_flags |= PF_OVERLAY;
- bed = get_elf_backend_data (abfd);
- tdata = elf_tdata (abfd);
- phdr = tdata->phdr;
- count = tdata->program_header_size / bed->s->sizeof_phdr;
- htab = spu_hash_table (info);
- if (htab->num_overlays != 0)
- {
- struct elf_segment_map *m;
- unsigned int o;
+ if (htab->ovtab != NULL && htab->ovtab->size != 0
+ && htab->params->ovly_flavour != ovly_soft_icache)
+ {
+ bfd_byte *p = htab->ovtab->contents;
+ unsigned int off = o * 16 + 8;
+
+ /* Write file_off into _ovly_table. */
+ bfd_put_32 (htab->ovtab->owner, phdr[i].p_offset, p + off);
+ }
+ }
+ /* Soft-icache has its file offset put in .ovl.init. */
+ if (htab->init != NULL && htab->init->size != 0)
+ {
+ bfd_vma val
+ = elf_section_data (htab->ovl_sec[0])->this_hdr.sh_offset;
+
+ bfd_put_32 (htab->init->owner, val, htab->init->contents + 4);
+ }
+ }
- for (i = 0, m = elf_tdata (abfd)->segment_map; m; ++i, m = m->next)
- if (m->count != 0
- && (o = spu_elf_section_data (m->sections[0])->u.o.ovl_index) != 0)
+ /* Round up p_filesz and p_memsz of PT_LOAD segments to multiples
+ of 16. This should always be possible when using the standard
+ linker scripts, but don't create overlapping segments if
+ someone is playing games with linker scripts. */
+ last = NULL;
+ for (i = count; i-- != 0; )
+ if (phdr[i].p_type == PT_LOAD)
{
- /* Mark this as an overlay header. */
- phdr[i].p_flags |= PF_OVERLAY;
+ unsigned adjust;
- if (htab->ovtab != NULL && htab->ovtab->size != 0)
- {
- bfd_byte *p = htab->ovtab->contents;
- unsigned int off = o * 16 + 8;
+ adjust = -phdr[i].p_filesz & 15;
+ if (adjust != 0
+ && last != NULL
+ && (phdr[i].p_offset + phdr[i].p_filesz
+ > last->p_offset - adjust))
+ break;
- /* Write file_off into _ovly_table. */
- bfd_put_32 (htab->ovtab->owner, phdr[i].p_offset, p + off);
- }
+ adjust = -phdr[i].p_memsz & 15;
+ if (adjust != 0
+ && last != NULL
+ && phdr[i].p_filesz != 0
+ && phdr[i].p_vaddr + phdr[i].p_memsz > last->p_vaddr - adjust
+ && phdr[i].p_vaddr + phdr[i].p_memsz <= last->p_vaddr)
+ break;
+
+ if (phdr[i].p_filesz != 0)
+ last = &phdr[i];
}
- }
- /* Round up p_filesz and p_memsz of PT_LOAD segments to multiples
- of 16. This should always be possible when using the standard
- linker scripts, but don't create overlapping segments if
- someone is playing games with linker scripts. */
- last = NULL;
- for (i = count; i-- != 0; )
- if (phdr[i].p_type == PT_LOAD)
- {
- unsigned adjust;
+ if (i == (unsigned int) -1)
+ for (i = count; i-- != 0; )
+ if (phdr[i].p_type == PT_LOAD)
+ {
+ unsigned adjust;
- adjust = -phdr[i].p_filesz & 15;
- if (adjust != 0
- && last != NULL
- && phdr[i].p_offset + phdr[i].p_filesz > last->p_offset - adjust)
- break;
+ adjust = -phdr[i].p_filesz & 15;
+ phdr[i].p_filesz += adjust;
- adjust = -phdr[i].p_memsz & 15;
- if (adjust != 0
- && last != NULL
- && phdr[i].p_filesz != 0
- && phdr[i].p_vaddr + phdr[i].p_memsz > last->p_vaddr - adjust
- && phdr[i].p_vaddr + phdr[i].p_memsz <= last->p_vaddr)
- break;
+ adjust = -phdr[i].p_memsz & 15;
+ phdr[i].p_memsz += adjust;
+ }
+ }
- if (phdr[i].p_filesz != 0)
- last = &phdr[i];
- }
+ return _bfd_elf_modify_headers (abfd, info);
+}
+
+bfd_boolean
+spu_elf_size_sections (bfd *obfd ATTRIBUTE_UNUSED, struct bfd_link_info *info)
+{
+ struct spu_link_hash_table *htab = spu_hash_table (info);
+ if (htab->params->emit_fixups)
+ {
+ asection *sfixup = htab->sfixup;
+ int fixup_count = 0;
+ bfd *ibfd;
+ size_t size;
- if (i == (unsigned int) -1)
- for (i = count; i-- != 0; )
- if (phdr[i].p_type == PT_LOAD)
+ for (ibfd = info->input_bfds; ibfd != NULL; ibfd = ibfd->link.next)
{
- unsigned adjust;
+ asection *isec;
- adjust = -phdr[i].p_filesz & 15;
- phdr[i].p_filesz += adjust;
+ if (bfd_get_flavour (ibfd) != bfd_target_elf_flavour)
+ continue;
- adjust = -phdr[i].p_memsz & 15;
- phdr[i].p_memsz += adjust;
- }
+ /* Walk over each section attached to the input bfd. */
+ for (isec = ibfd->sections; isec != NULL; isec = isec->next)
+ {
+ Elf_Internal_Rela *internal_relocs, *irelaend, *irela;
+ bfd_vma base_end;
+
+ /* If there aren't any relocs, then there's nothing more
+ to do. */
+ if ((isec->flags & SEC_ALLOC) == 0
+ || (isec->flags & SEC_RELOC) == 0
+ || isec->reloc_count == 0)
+ continue;
+
+ /* Get the relocs. */
+ internal_relocs =
+ _bfd_elf_link_read_relocs (ibfd, isec, NULL, NULL,
+ info->keep_memory);
+ if (internal_relocs == NULL)
+ return FALSE;
+ /* 1 quadword can contain up to 4 R_SPU_ADDR32
+ relocations. They are stored in a single word by
+ saving the upper 28 bits of the address and setting the
+ lower 4 bits to a bit mask of the words that have the
+ relocation. BASE_END keeps track of the next quadword. */
+ irela = internal_relocs;
+ irelaend = irela + isec->reloc_count;
+ base_end = 0;
+ for (; irela < irelaend; irela++)
+ if (ELF32_R_TYPE (irela->r_info) == R_SPU_ADDR32
+ && irela->r_offset >= base_end)
+ {
+ base_end = (irela->r_offset & ~(bfd_vma) 15) + 16;
+ fixup_count++;
+ }
+ }
+ }
+
+ /* We always have a NULL fixup as a sentinel */
+ size = (fixup_count + 1) * FIXUP_RECORD_SIZE;
+ if (!bfd_set_section_size (sfixup, size))
+ return FALSE;
+ sfixup->contents = (bfd_byte *) bfd_zalloc (info->input_bfds, size);
+ if (sfixup->contents == NULL)
+ return FALSE;
+ }
return TRUE;
}
-#define TARGET_BIG_SYM bfd_elf32_spu_vec
+#define TARGET_BIG_SYM spu_elf32_vec
#define TARGET_BIG_NAME "elf32-spu"
#define ELF_ARCH bfd_arch_spu
+#define ELF_TARGET_ID SPU_ELF_DATA
#define ELF_MACHINE_CODE EM_SPU
/* This matches the alignment need for DMA. */
#define ELF_MAXPAGESIZE 0x80
-#define elf_backend_rela_normal 1
+#define elf_backend_rela_normal 1
#define elf_backend_can_gc_sections 1
#define bfd_elf32_bfd_reloc_type_lookup spu_elf_reloc_type_lookup
-#define bfd_elf32_bfd_reloc_name_lookup spu_elf_reloc_name_lookup
+#define bfd_elf32_bfd_reloc_name_lookup spu_elf_reloc_name_lookup
#define elf_info_to_howto spu_elf_info_to_howto
#define elf_backend_count_relocs spu_elf_count_relocs
#define elf_backend_relocate_section spu_elf_relocate_section
+#define elf_backend_finish_dynamic_sections spu_elf_finish_dynamic_sections
#define elf_backend_symbol_processing spu_elf_backend_symbol_processing
#define elf_backend_link_output_symbol_hook spu_elf_output_symbol_hook
#define elf_backend_object_p spu_elf_object_p
#define elf_backend_additional_program_headers spu_elf_additional_program_headers
#define elf_backend_modify_segment_map spu_elf_modify_segment_map
-#define elf_backend_modify_program_headers spu_elf_modify_program_headers
-#define elf_backend_post_process_headers spu_elf_post_process_headers
+#define elf_backend_modify_headers spu_elf_modify_headers
+#define elf_backend_init_file_header spu_elf_init_file_header
#define elf_backend_fake_sections spu_elf_fake_sections
#define elf_backend_special_sections spu_elf_special_sections
#define bfd_elf32_bfd_final_link spu_elf_final_link
projects / deliverable / binutils-gdb.git / blobdiff