/* tc-avr.c -- Assembler code for the ATMEL AVR
- Copyright (C) 1999-2015 Free Software Foundation, Inc.
+ Copyright (C) 1999-2020 Free Software Foundation, Inc.
Contributed by Denis Chertykov <denisc@overta.ru>
This file is part of GAS, the GNU Assembler.
struct avr_opcodes_s
{
- char * name;
- char * constraints;
- char * opcode;
+ const char * name;
+ const char * constraints;
+ const char * opcode;
int insn_size; /* In words. */
int isa;
unsigned int bin_opcode;
{NULL, NULL, NULL, 0, 0, 0}
};
+
+/* Stuff for the `__gcc_isr' pseudo instruction.
+
+ Purpose of the pseudo instruction is to emit more efficient ISR prologues
+ and epilogues than GCC currently does. GCC has no explicit (on RTL level)
+ modelling of SREG, TMP_REG or ZERO_REG. These regs are used implicitly
+ during instruction printing. That doesn't hurt too much for ordinary
+ functions, however for small ISRs there might be some overhead.
+
+ As implementing http://gcc.gnu.org/PR20296 would imply an almost complete
+ rewite of GCC's AVR back-end (which might pop up less optimized code in
+ other places), we provide a pseudo-instruction which is resolved by GAS
+ into ISR prologue / epilogue as expected by GCC.
+
+ Using GAS for this purpose has the additional benefit that it can scan
+ code emit by inline asm which is opaque to GCC.
+
+ The pseudo-instruction is only supposed to handle the starting of
+ prologue and the ending of epilogues (without RETI) which deal with
+ SREG, TMP_REG and ZERO_REG and one additional, optional general purpose
+ register.
+
+ __gcc_isr consists of 3 different "chunks":
+
+ __gcc_isr 1
+ Chunk 1 (ISR_CHUNK_Prologue)
+ Start the ISR code. Will be replaced by ISR prologue by next Done chunk.
+ Must be the 1st chunk in a file or follow a Done chunk from previous
+ ISR (which has been patched already).
+
+ It will finish the current frag and emit a new frag of
+ type rs_machine_dependent, subtype ISR_CHUNK_Prologue.
+
+ __gcc_isr 2
+ Chunk 2 (ISR_CHUNK_Epilogue)
+ Will be replaced by ISR epilogue by next Done chunk. Must follow
+ chunk 1 (Prologue) or chunk 2 (Epilogue). Functions might come
+ without epilogue or with more than one epilogue, and even code
+ located statically after the last epilogue might belong to a function.
+
+ It will finish the current frag and emit a new frag of
+ type rs_machine_dependent, subtype ISR_CHUNK_Epilogue.
+
+ __gcc_isr 0, Rx
+ Chunk 0 (ISR_CHUNK_Done)
+ Must follow chunk 1 (Prologue) or chunk 2 (Epilogue) and finishes
+ the ISR code. Only GCC can know where a function's code ends.
+
+ It triggers the patch-up of all rs_machine_dependent frags in the
+ current frag chain and turns them into ordinary rs_fill code frags.
+
+ If Rx is a register > ZERO_REG then GCC also wants to push / pop Rx.
+ If neither TMP_REG nor ZERO_REG are needed, Rx will be used in
+ the push / pop sequence avoiding the need for TMP_REG / ZERO_REG.
+ If Rx <= ZERO_REG then GCC doesn't assume anything about Rx.
+
+ Assumptions:
+
+ o GCC takes care of code that is opaque to GAS like tail calls
+ or non-local goto.
+
+ o Using SEI / CLI does not count as clobbering SREG. This is
+ because a final RETI will restore the I-flag.
+
+ o Using OUT or ST* is supposed not to clobber SREG. Sequences like
+
+ IN-SREG + CLI + Atomic-Code + OUT-SREG
+
+ will still work as expected because the scan will reveal any
+ clobber of SREG other than I-flag and emit PUSH / POP of SREG.
+*/
+
+enum
+ {
+ ISR_CHUNK_Done = 0,
+ ISR_CHUNK_Prologue = 1,
+ ISR_CHUNK_Epilogue = 2
+ };
+
+static struct
+{
+ /* Previous __gcc_isr chunk (one of the enums above)
+ and it's location for diagnostics. */
+ int prev_chunk;
+ unsigned line;
+ const char *file;
+ /* Replacer for __gcc_isr.n_pushed once we know how many regs are
+ pushed by the Prologue chunk. */
+ symbolS *sym_n_pushed;
+
+ /* Set and used during parse from chunk 1 (Prologue) up to chunk 0 (Done).
+ Set by `avr_update_gccisr' and used by `avr_patch_gccisr_frag'. */
+ int need_reg_tmp;
+ int need_reg_zero;
+ int need_sreg;
+} avr_isr;
+
+static void avr_gccisr_operands (struct avr_opcodes_s*, char**);
+static void avr_update_gccisr (struct avr_opcodes_s*, int, int);
+static struct avr_opcodes_s *avr_gccisr_opcode;
+
const char comment_chars[] = ";";
const char line_comment_chars[] = "#";
const char line_separator_chars[] = "$";
const char *md_shortopts = "m:";
struct mcu_type_s
{
- char *name;
+ const char *name;
int isa;
int mach;
};
static struct mcu_type_s mcu_types[] =
{
{"avr1", AVR_ISA_AVR1, bfd_mach_avr1},
-/* TODO: insruction set for avr2 architecture should be AVR_ISA_AVR2,
+/* TODO: instruction set for avr2 architecture should be AVR_ISA_AVR2,
but set to AVR_ISA_AVR25 for some following version
of GCC (from 4.3) for backward compatibility. */
{"avr2", AVR_ISA_AVR25, bfd_mach_avr2},
{"avr25", AVR_ISA_AVR25, bfd_mach_avr25},
-/* TODO: insruction set for avr3 architecture should be AVR_ISA_AVR3,
+/* TODO: instruction set for avr3 architecture should be AVR_ISA_AVR3,
but set to AVR_ISA_AVR3_ALL for some following version
of GCC (from 4.3) for backward compatibility. */
{"avr3", AVR_ISA_AVR3_ALL, bfd_mach_avr3},
{"avr31", AVR_ISA_AVR31, bfd_mach_avr31},
{"avr35", AVR_ISA_AVR35, bfd_mach_avr35},
{"avr4", AVR_ISA_AVR4, bfd_mach_avr4},
-/* TODO: insruction set for avr5 architecture should be AVR_ISA_AVR5,
+/* TODO: instruction set for avr5 architecture should be AVR_ISA_AVR5,
but set to AVR_ISA_AVR51 for some following version
of GCC (from 4.3) for backward compatibility. */
{"avr5", AVR_ISA_AVR51, bfd_mach_avr5},
{"atxmega16e5", AVR_ISA_XMEGA, bfd_mach_avrxmega2},
{"atxmega8e5", AVR_ISA_XMEGA, bfd_mach_avrxmega2},
{"atxmega32x1", AVR_ISA_XMEGA, bfd_mach_avrxmega2},
+ {"attiny212", AVR_ISA_XMEGA, bfd_mach_avrxmega3},
+ {"attiny214", AVR_ISA_XMEGA, bfd_mach_avrxmega3},
+ {"attiny412", AVR_ISA_XMEGA, bfd_mach_avrxmega3},
+ {"attiny414", AVR_ISA_XMEGA, bfd_mach_avrxmega3},
+ {"attiny416", AVR_ISA_XMEGA, bfd_mach_avrxmega3},
+ {"attiny417", AVR_ISA_XMEGA, bfd_mach_avrxmega3},
+ {"attiny814", AVR_ISA_XMEGA, bfd_mach_avrxmega3},
+ {"attiny816", AVR_ISA_XMEGA, bfd_mach_avrxmega3},
+ {"attiny817", AVR_ISA_XMEGA, bfd_mach_avrxmega3},
+ {"attiny1614", AVR_ISA_XMEGA, bfd_mach_avrxmega3},
+ {"attiny1616", AVR_ISA_XMEGA, bfd_mach_avrxmega3},
+ {"attiny1617", AVR_ISA_XMEGA, bfd_mach_avrxmega3},
+ {"attiny3214", AVR_ISA_XMEGA, bfd_mach_avrxmega3},
+ {"attiny3216", AVR_ISA_XMEGA, bfd_mach_avrxmega3},
+ {"attiny3217", AVR_ISA_XMEGA, bfd_mach_avrxmega3},
{"atxmega64a3", AVR_ISA_XMEGA, bfd_mach_avrxmega4},
{"atxmega64a3u",AVR_ISA_XMEGAU, bfd_mach_avrxmega4},
{"atxmega64a4u",AVR_ISA_XMEGAU, bfd_mach_avrxmega4},
int no_wrap; /* -mno-wrap: reject rjmp/rcall with 8K wrap-around. */
int no_link_relax; /* -mno-link-relax / -mlink-relax: generate (or not)
relocations for linker relaxation. */
+ int have_gccisr; /* Whether "__gcc_isr" is a known (pseudo) insn. */
};
-static struct avr_opt_s avr_opt = { 0, 0, 0, 0 };
+static struct avr_opt_s avr_opt = { 0, 0, 0, 0, 0 };
const char EXP_CHARS[] = "eE";
const char FLT_CHARS[] = "dD";
struct exp_mod_s
{
- char * name;
+ const char * name;
bfd_reloc_code_real_type reloc;
bfd_reloc_code_real_type neg_reloc;
int have_pm;
{"hhi8", BFD_RELOC_AVR_MS8_LDI, BFD_RELOC_AVR_MS8_LDI_NEG, 0},
};
-/* A union used to store indicies into the exp_mod[] array
+/* A union used to store indices into the exp_mod[] array
in a hash table which expects void * data types. */
typedef union
{
/* Reloc modifiers hash control (hh8,hi8,lo8,pm_xx). */
static struct hash_control *avr_mod_hash;
+/* Whether some opcode does not change SREG. */
+static struct hash_control *avr_no_sreg_hash;
+
+static const char* const avr_no_sreg[] =
+ {
+ /* Arithmetic */
+ "ldi", "swap", "mov", "movw",
+ /* Special instructions. I-Flag will be restored by RETI, and we don't
+ consider I-Flag as being clobbered when changed. */
+ "sei", "cli", "reti", "brie", "brid",
+ "nop", "wdr", "sleep",
+ /* Load / Store */
+ "ld", "ldd", "lds", "pop", "in", "lpm", "elpm",
+ "st", "std", "sts", "push", "out",
+ /* Jumps and Calls. Calls might call code that changes SREG.
+ GCC has to filter out ABI calls. The non-ABI transparent calls
+ must use [R]CALL and are filtered out now by not mentioning them. */
+ "rjmp", "jmp", "ijmp", "ret",
+ /* Skipping. Branches need SREG to be set, hence we regard them
+ as if they changed SREG and don't list them here. */
+ "sbrc", "sbrs", "sbic", "sbis", "cpse",
+ /* I/O Manipulation */
+ "sbi", "cbi",
+ /* Read-Modify-Write */
+ "lac", "las", "lat", "xch"
+ };
+
#define OPTION_MMCU 'm'
enum options
{
OPTION_NO_WRAP,
OPTION_ISA_RMW,
OPTION_LINK_RELAX,
- OPTION_NO_LINK_RELAX
+ OPTION_NO_LINK_RELAX,
+ OPTION_HAVE_GCCISR
};
struct option md_longopts[] =
{ "mrmw", no_argument, NULL, OPTION_ISA_RMW },
{ "mlink-relax", no_argument, NULL, OPTION_LINK_RELAX },
{ "mno-link-relax", no_argument, NULL, OPTION_NO_LINK_RELAX },
+ { "mgcc-isr", no_argument, NULL, OPTION_HAVE_GCCISR },
{ NULL, no_argument, NULL, 0 }
};
" avr51 - enhanced AVR core with up to 128K program memory\n"
" avr6 - enhanced AVR core with up to 256K program memory\n"
" avrxmega2 - XMEGA, > 8K, < 64K FLASH, < 64K RAM\n"
- " avrxmega3 - XMEGA, > 8K, <= 64K FLASH, > 64K RAM\n"
+ " avrxmega3 - XMEGA, RAM + FLASH < 64K, Flash visible in RAM\n"
" avrxmega4 - XMEGA, > 64K, <= 128K FLASH, <= 64K RAM\n"
" avrxmega5 - XMEGA, > 64K, <= 128K FLASH, > 64K RAM\n"
" avrxmega6 - XMEGA, > 128K, <= 256K FLASH, <= 64K RAM\n"
" -mrmw accept Read-Modify-Write instructions\n"
" -mlink-relax generate relocations for linker relaxation (default)\n"
" -mno-link-relax don't generate relocations for linker relaxation.\n"
+ " -mgcc-isr accept the __gcc_isr pseudo-instruction.\n"
));
show_mcu_list (stream);
}
}
int
-md_parse_option (int c, char *arg)
+md_parse_option (int c, const char *arg)
{
switch (c)
{
case OPTION_MMCU:
{
int i;
- char *s = alloca (strlen (arg) + 1);
-
- {
- char *t = s;
- char *arg1 = arg;
-
- do
- *t = TOLOWER (*arg1++);
- while (*t++);
- }
for (i = 0; mcu_types[i].name; ++i)
- if (strcmp (mcu_types[i].name, s) == 0)
+ if (strcasecmp (mcu_types[i].name, arg) == 0)
break;
if (!mcu_types[i].name)
type - this for allows passing -mmcu=... via gcc ASM_SPEC as well
as .arch ... in the asm output at the same time. */
if (avr_mcu == &default_mcu || avr_mcu->mach == mcu_types[i].mach)
- {
- specified_mcu.name = mcu_types[i].name;
- specified_mcu.isa |= mcu_types[i].isa;
- specified_mcu.mach = mcu_types[i].mach;
- avr_mcu = &specified_mcu;
- }
+ {
+ specified_mcu.name = mcu_types[i].name;
+ specified_mcu.isa |= mcu_types[i].isa;
+ specified_mcu.mach = mcu_types[i].mach;
+ avr_mcu = &specified_mcu;
+ }
else
as_fatal (_("redefinition of mcu type `%s' to `%s'"),
avr_mcu->name, mcu_types[i].name);
case OPTION_NO_LINK_RELAX:
avr_opt.no_link_relax = 1;
return 1;
+ case OPTION_HAVE_GCCISR:
+ avr_opt.have_gccisr = 1;
+ return 1;
}
return 0;
}
+
+/* Implement `md_undefined_symbol' */
+/* If we are in `__gcc_isr' chunk, pop up `__gcc_isr.n_pushed.<NUM>'
+ instead of `__gcc_isr.n_pushed'. This will be resolved by the Done
+ chunk in `avr_patch_gccisr_frag' to the number of PUSHes produced by
+ the Prologue chunk. */
+
symbolS *
-md_undefined_symbol (char *name ATTRIBUTE_UNUSED)
+avr_undefined_symbol (char *name)
{
+ if (ISR_CHUNK_Done != avr_isr.prev_chunk
+ && 0 == strcmp (name, "__gcc_isr.n_pushed"))
+ {
+ if (!avr_isr.sym_n_pushed)
+ {
+ static unsigned suffix;
+ char xname[30];
+ sprintf (xname, "%s.%03u", name, (++suffix) % 1000);
+ avr_isr.sym_n_pushed = symbol_new (xname, undefined_section,
+ (valueT) 0, &zero_address_frag);
+ }
+ return avr_isr.sym_n_pushed;
+ }
+
return NULL;
}
-char *
+const char *
md_atof (int type, char *litP, int *sizeP)
{
return ieee_md_atof (type, litP, sizeP, FALSE);
hash_insert (avr_mod_hash, EXP_MOD_NAME (i), m.ptr);
}
+ avr_no_sreg_hash = hash_new ();
+
+ for (i = 0; i < ARRAY_SIZE (avr_no_sreg); ++i)
+ {
+ gas_assert (hash_find (avr_hash, avr_no_sreg[i]));
+ hash_insert (avr_no_sreg_hash, avr_no_sreg[i], (char*) 4 /* dummy */);
+ }
+
+ avr_gccisr_opcode = (struct avr_opcodes_s*) hash_find (avr_hash, "__gcc_isr");
+ gas_assert (avr_gccisr_opcode);
+
bfd_set_arch_mach (stdoutput, TARGET_ARCH, avr_mcu->mach);
linkrelax = !avr_opt.no_link_relax;
}
static unsigned int
avr_operand (struct avr_opcodes_s *opcode,
int where,
- char *op,
- char **line)
+ const char *op,
+ char **line,
+ int *pregno)
{
expressionS op_expr;
unsigned int op_mask = 0;
}
}
+ if (pregno)
+ *pregno = op_mask;
+
if (avr_mcu->mach == bfd_mach_avrtiny)
{
if (op_mask < 16 || op_mask > 31)
if (*str == '+')
{
++str;
- char *s;
+ const char *s;
for (s = opcode->opcode; *s; ++s)
{
if (*s == '+')
}
break;
+ case 'N':
+ {
+ unsigned int x;
+
+ x = avr_get_constant (str, 255);
+ str = input_line_pointer;
+ op_mask = x;
+ }
+ break;
+
case 'K':
input_line_pointer = str;
avr_offset_expression (& op_expr);
return op_mask;
}
+/* TC_FRAG_INIT hook */
+
+void
+avr_frag_init (fragS *frag)
+{
+ memset (& frag->tc_frag_data, 0, sizeof frag->tc_frag_data);
+}
+
+
/* Parse instruction operands.
Return binary opcode. */
static unsigned int
avr_operands (struct avr_opcodes_s *opcode, char **line)
{
- char *op = opcode->constraints;
+ const char *op = opcode->constraints;
unsigned int bin = opcode->bin_opcode;
char *frag = frag_more (opcode->insn_size * 2);
char *str = *line;
int where = frag - frag_now->fr_literal;
- static unsigned int prev = 0; /* Previous opcode. */
+ int regno1 = -2;
+ int regno2 = -2;
/* Opcode have operands. */
if (*op)
/* Parse first operand. */
if (REGISTER_P (*op))
reg1_present = 1;
- reg1 = avr_operand (opcode, where, op, &str);
+ reg1 = avr_operand (opcode, where, op, &str, ®no1);
++op;
/* Parse second operand. */
{
reg2 = reg1;
reg2_present = 1;
+ regno2 = regno1;
}
else
{
as_bad (_("`,' required"));
str = skip_space (str);
- reg2 = avr_operand (opcode, where, op, &str);
+ reg2 = avr_operand (opcode, where, op, &str, ®no2);
}
if (reg1_present && reg2_present)
bin |= reg1 | reg2;
}
+ if (avr_opt.have_gccisr)
+ avr_update_gccisr (opcode, regno1, regno2);
+
/* Detect undefined combinations (like ld r31,Z+). */
if (!avr_opt.all_opcodes && AVR_UNDEF_P (bin))
as_warn (_("undefined combination of operands"));
(AVR core bug, fixed in the newer devices). */
if (!(avr_opt.no_skip_bug ||
(avr_mcu->isa & (AVR_ISA_MUL | AVR_ISA_MOVW)))
- && AVR_SKIP_P (prev))
+ && AVR_SKIP_P (frag_now->tc_frag_data.prev_opcode))
as_warn (_("skipping two-word instruction"));
bfd_putl32 ((bfd_vma) bin, frag);
else
bfd_putl16 ((bfd_vma) bin, frag);
- prev = bin;
+ frag_now->tc_frag_data.prev_opcode = bin;
*line = str;
return bin;
}
valueT
md_section_align (asection *seg, valueT addr)
{
- int align = bfd_get_section_alignment (stdoutput, seg);
- return ((addr + (1 << align) - 1) & (-1 << align));
+ int align = bfd_section_alignment (seg);
+ return ((addr + (1 << align) - 1) & (-1UL << align));
}
/* If you define this macro, it should return the offset between the
fixS *fixp)
{
arelent *reloc;
+ bfd_reloc_code_real_type code = fixp->fx_r_type;
if (fixp->fx_subsy != NULL)
{
return NULL;
}
- reloc = xmalloc (sizeof (arelent));
+ reloc = XNEW (arelent);
- reloc->sym_ptr_ptr = xmalloc (sizeof (asymbol *));
+ reloc->sym_ptr_ptr = XNEW (asymbol *);
*reloc->sym_ptr_ptr = symbol_get_bfdsym (fixp->fx_addsy);
reloc->address = fixp->fx_frag->fr_address + fixp->fx_where;
- reloc->howto = bfd_reloc_type_lookup (stdoutput, fixp->fx_r_type);
+
+ if ((fixp->fx_r_type == BFD_RELOC_32) && (fixp->fx_pcrel))
+ {
+ if (seg->use_rela_p)
+ fixp->fx_offset -= md_pcrel_from_section (fixp, seg);
+ else
+ fixp->fx_offset = reloc->address;
+
+ code = BFD_RELOC_32_PCREL;
+ }
+
+ reloc->addend = fixp->fx_offset;
+
+ reloc->howto = bfd_reloc_type_lookup (stdoutput, code);
+
if (reloc->howto == (reloc_howto_type *) NULL)
{
as_bad_where (fixp->fx_file, fixp->fx_line,
|| fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
reloc->address = fixp->fx_offset;
- reloc->addend = fixp->fx_offset;
return reloc;
}
specifications with same mnemonic who's ISA bits matches.
This requires include/opcode/avr.h to have the instructions with
- same mnenomic to be specified in sequence. */
+ same mnemonic to be specified in sequence. */
while ((opcode->isa & avr_mcu->isa) != opcode->isa)
{
return;
}
+ if (opcode == avr_gccisr_opcode
+ && !avr_opt.have_gccisr)
+ {
+ as_bad (_("pseudo instruction `%s' not supported"), op);
+ return;
+ }
+
/* Special case for opcodes with optional operands (lpm, elpm) -
version with operands exists in avr_opcodes[] in the next entry. */
{
char *t = input_line_pointer;
- avr_operands (opcode, &str);
+ if (opcode == avr_gccisr_opcode)
+ avr_gccisr_operands (opcode, &str);
+ else
+ avr_operands (opcode, &str);
if (*skip_space (str))
as_bad (_("garbage at end of line"));
input_line_pointer = t;
{ "hi8", 1, BFD_RELOC_AVR_8_HI, "`hi8' " },
{ "hlo8", 1, BFD_RELOC_AVR_8_HLO, "`hlo8' " },
{ "hh8", 1, BFD_RELOC_AVR_8_HLO, "`hh8' " },
- /* End of list. */
- { NULL, 0, 0, NULL }
};
/* Parse special CONS expression: pm (expression) or alternatively
const exp_mod_data_t *
avr_parse_cons_expression (expressionS *exp, int nbytes)
{
- const exp_mod_data_t *pexp = &exp_mod_data[0];
char *tmp;
+ unsigned int i;
tmp = input_line_pointer = skip_space (input_line_pointer);
/* The first entry of exp_mod_data[] contains an entry if no
expression modifier is present. Skip it. */
- for (pexp++; pexp->name; pexp++)
+ for (i = 0; i < ARRAY_SIZE (exp_mod_data); i++)
{
+ const exp_mod_data_t *pexp = &exp_mod_data[i];
int len = strlen (pexp->name);
if (nbytes == pexp->nbytes
}
if (bad)
- as_bad (_("illegal %srelocation size: %d"), pexp_mod_data->error, nbytes);
+ as_bad (_("illegal %s relocation size: %d"), pexp_mod_data->error, nbytes);
}
static bfd_boolean
cfi_add_CFA_def_cfa (32, return_size);
/* Note that AVR consistently uses post-decrement, which means that things
- do not line up the same way as for targers that use pre-decrement. */
+ do not line up the same way as for targets that use pre-decrement. */
cfi_add_CFA_offset (DWARF2_DEFAULT_RETURN_COLUMN, 1-return_size);
}
case RECORD_ALIGN_AND_FILL:
md_number_to_chars (frag_ptr, record->data.align.bytes, 4);
- md_number_to_chars (frag_ptr, record->data.align.fill, 4);
+ md_number_to_chars (frag_ptr + 4, record->data.align.fill, 4);
frag_ptr += 8;
break;
sec = bfd_make_section (stdoutput, section_name);
if (sec == NULL)
as_fatal (_("Failed to create property section `%s'\n"), section_name);
- bfd_set_section_flags (stdoutput, sec, flags);
+ bfd_set_section_flags (sec, flags);
sec->output_section = sec;
return sec;
}
alignment mechanism. */
if ((fragP->fr_type == rs_align
|| fragP->fr_type == rs_align_code)
- && fragP->fr_address > 0
&& fragP->fr_offset > 0)
{
+ char *p = fragP->fr_literal + fragP->fr_fix;
+
fragP->tc_frag_data.is_align = TRUE;
fragP->tc_frag_data.alignment = fragP->fr_offset;
+ fragP->tc_frag_data.fill = *p;
+ fragP->tc_frag_data.has_fill = (fragP->tc_frag_data.fill != 0);
}
if (fragP->fr_type == rs_org && fragP->fr_offset > 0)
{
struct avr_property_record_link *prop_rec_link;
- prop_rec_link = xmalloc (sizeof (struct avr_property_record_link));
- memset (prop_rec_link, 0, sizeof (*prop_rec_link));
+ prop_rec_link = XCNEW (struct avr_property_record_link);
+ gas_assert (fragP->fr_next != NULL);
if (fragP->tc_frag_data.is_org)
{
}
else
{
- prop_rec_link->record.offset = fragP->fr_address;
+ prop_rec_link->record.offset = fragP->fr_next->fr_address;
prop_rec_link->record.section = sec;
gas_assert (fragP->tc_frag_data.is_align);
return;
prop_sec = avr_create_property_section ();
- bfd_set_section_size (stdoutput, prop_sec, sec_size);
+ bfd_set_section_size (prop_sec, sec_size);
subseg_set (prop_sec, 0);
frag_base = frag_more (sec_size);
{
avr_create_and_fill_property_section ();
}
+
+
+/* Accumulate information about instruction sequence to `avr_isr':
+ wheter TMP_REG, ZERO_REG and SREG might be touched. Used during parse.
+ REG1 is either -1 or a register number used by the instruction as input
+ or output operand. Similar for REG2. */
+
+static void
+avr_update_gccisr (struct avr_opcodes_s *opcode, int reg1, int reg2)
+{
+ const int tiny_p = avr_mcu->mach == bfd_mach_avrtiny;
+ const int reg_tmp = tiny_p ? 16 : 0;
+ const int reg_zero = 1 + reg_tmp;
+
+ if (ISR_CHUNK_Done == avr_isr.prev_chunk
+ || (avr_isr.need_sreg
+ && avr_isr.need_reg_tmp
+ && avr_isr.need_reg_zero))
+ {
+ /* Nothing (more) to do */
+ return;
+ }
+
+ /* SREG: Look up instructions that don't clobber SREG. */
+
+ if (!avr_isr.need_sreg
+ && !hash_find (avr_no_sreg_hash, opcode->name))
+ {
+ avr_isr.need_sreg = 1;
+ }
+
+ /* Handle explicit register operands. Record *any* use as clobber.
+ This is because TMP_REG and ZERO_REG are not global and using
+ them makes no sense without a previous set. */
+
+ avr_isr.need_reg_tmp |= reg1 == reg_tmp || reg2 == reg_tmp;
+ avr_isr.need_reg_zero |= reg1 == reg_zero || reg2 == reg_zero;
+
+ /* Handle implicit register operands and some opaque stuff. */
+
+ if (strstr (opcode->name, "lpm")
+ && '?' == *opcode->constraints)
+ {
+ avr_isr.need_reg_tmp = 1;
+ }
+
+ if (strstr (opcode->name, "call")
+ || strstr (opcode->name, "mul")
+ || 0 == strcmp (opcode->name, "des")
+ || (0 == strcmp (opcode->name, "movw")
+ && (reg1 == reg_tmp || reg2 == reg_tmp)))
+ {
+ avr_isr.need_reg_tmp = 1;
+ avr_isr.need_reg_zero = 1;
+ }
+}
+
+
+/* Emit some 1-word instruction to **PWHERE and advance *PWHERE by the number
+ of octets written. INSN specifies the desired instruction and REG is the
+ register used by it. This function is only used with restricted subset of
+ instructions as might be emit by `__gcc_isr'. IN / OUT will use SREG
+ and LDI loads 0. */
+
+static void
+avr_emit_insn (const char *insn, int reg, char **pwhere)
+{
+ const int sreg = 0x3f;
+ unsigned bin = 0;
+ const struct avr_opcodes_s *op
+ = (struct avr_opcodes_s*) hash_find (avr_hash, insn);
+
+ /* We only have to deal with: IN, OUT, PUSH, POP, CLR, LDI 0. All of
+ these deal with at least one Reg and are 1-word instructions. */
+
+ gas_assert (op && 1 == op->insn_size);
+ gas_assert (reg >= 0 && reg <= 31);
+
+ if (strchr (op->constraints, 'r'))
+ {
+ bin = op->bin_opcode | (reg << 4);
+ }
+ else if (strchr (op->constraints, 'd'))
+ {
+ gas_assert (reg >= 16);
+ bin = op->bin_opcode | ((reg & 0xf) << 4);
+ }
+ else
+ abort();
+
+ if (strchr (op->constraints, 'P'))
+ {
+ bin |= ((sreg & 0x30) << 5) | (sreg & 0x0f);
+ }
+ else if (0 == strcmp ("r=r", op->constraints))
+ {
+ bin |= ((reg & 0x10) << 5) | (reg & 0x0f);
+ }
+ else
+ gas_assert (0 == strcmp ("r", op->constraints)
+ || 0 == strcmp ("ldi", op->name));
+
+ bfd_putl16 ((bfd_vma) bin, *pwhere);
+ (*pwhere) += 2 * op->insn_size;
+}
+
+
+/* Turn rs_machine_dependent frag *FR into an ordinary rs_fill code frag,
+ using information gathered in `avr_isr'. REG is the register number as
+ supplied by Done chunk "__gcc_isr 0,REG". */
+
+static void
+avr_patch_gccisr_frag (fragS *fr, int reg)
+{
+ int treg;
+ int n_pushed = 0;
+ char *where = fr->fr_literal;
+ const int tiny_p = avr_mcu->mach == bfd_mach_avrtiny;
+ const int reg_tmp = tiny_p ? 16 : 0;
+ const int reg_zero = 1 + reg_tmp;
+
+ /* Clearing ZERO_REG on non-Tiny needs CLR which clobbers SREG. */
+
+ avr_isr.need_sreg |= !tiny_p && avr_isr.need_reg_zero;
+
+ /* A working register to PUSH / POP the SREG. We might use the register
+ as supplied by ISR_CHUNK_Done for that purpose as GCC wants to push
+ it anyways. If GCC passes ZERO_REG or TMP_REG, it has no clue (and
+ no additional regs to safe) and we use that reg. */
+
+ treg
+ = avr_isr.need_reg_tmp ? reg_tmp
+ : avr_isr.need_reg_zero ? reg_zero
+ : avr_isr.need_sreg ? reg
+ : reg > reg_zero ? reg
+ : -1;
+
+ if (treg >= 0)
+ {
+ /* Non-empty prologue / epilogue */
+
+ if (ISR_CHUNK_Prologue == fr->fr_subtype)
+ {
+ avr_emit_insn ("push", treg, &where);
+ n_pushed++;
+
+ if (avr_isr.need_sreg)
+ {
+ avr_emit_insn ("in", treg, &where);
+ avr_emit_insn ("push", treg, &where);
+ n_pushed++;
+ }
+
+ if (avr_isr.need_reg_zero)
+ {
+ if (reg_zero != treg)
+ {
+ avr_emit_insn ("push", reg_zero, &where);
+ n_pushed++;
+ }
+ avr_emit_insn (tiny_p ? "ldi" : "clr", reg_zero, &where);
+ }
+
+ if (reg > reg_zero && reg != treg)
+ {
+ avr_emit_insn ("push", reg, &where);
+ n_pushed++;
+ }
+ }
+ else if (ISR_CHUNK_Epilogue == fr->fr_subtype)
+ {
+ /* Same logic as in Prologue but in reverse order and with counter
+ parts of either instruction: POP instead of PUSH and OUT instead
+ of IN. Clearing ZERO_REG has no couter part. */
+
+ if (reg > reg_zero && reg != treg)
+ avr_emit_insn ("pop", reg, &where);
+
+ if (avr_isr.need_reg_zero
+ && reg_zero != treg)
+ avr_emit_insn ("pop", reg_zero, &where);
+
+ if (avr_isr.need_sreg)
+ {
+ avr_emit_insn ("pop", treg, &where);
+ avr_emit_insn ("out", treg, &where);
+ }
+
+ avr_emit_insn ("pop", treg, &where);
+ }
+ else
+ abort();
+ } /* treg >= 0 */
+
+ if (ISR_CHUNK_Prologue == fr->fr_subtype
+ && avr_isr.sym_n_pushed)
+ {
+ symbolS *sy = avr_isr.sym_n_pushed;
+ /* Turn magic `__gcc_isr.n_pushed' into its now known value. */
+
+ S_SET_VALUE (sy, n_pushed);
+ S_SET_SEGMENT (sy, expr_section);
+ avr_isr.sym_n_pushed = NULL;
+ }
+
+ /* Turn frag into ordinary code frag of now known size. */
+
+ fr->fr_var = 0;
+ fr->fr_fix = where - fr->fr_literal;
+ gas_assert (fr->fr_fix <= (valueT) fr->fr_offset);
+ fr->fr_offset = 0;
+ fr->fr_type = rs_fill;
+ fr->fr_subtype = 0;
+}
+
+
+/* Implements `__gcc_isr' pseudo-instruction. For Prologue and Epilogue
+ chunks, emit a new rs_machine_dependent frag. For Done chunks, traverse
+ the current segment and patch all rs_machine_dependent frags to become
+ appropriate rs_fill code frags. If chunks are seen in an odd ordering,
+ throw an error instead. */
+
+static void
+avr_gccisr_operands (struct avr_opcodes_s *opcode, char **line)
+{
+ int bad = 0;
+ int chunk, reg = 0;
+ char *str = *line;
+
+ gas_assert (avr_opt.have_gccisr);
+
+ /* We only use operands "N" and "r" which don't pop new fix-ups. */
+
+ /* 1st operand: Which chunk of __gcc_isr: 0...2. */
+
+ chunk = avr_operand (opcode, -1, "N", &str, NULL);
+ if (chunk < 0 || chunk > 2)
+ as_bad (_("%s requires value 0-2 as operand 1"), opcode->name);
+
+ if (ISR_CHUNK_Done == chunk)
+ {
+ /* 2nd operand: A register to push / pop. */
+
+ str = skip_space (str);
+ if (*str == '\0' || *str++ != ',')
+ as_bad (_("`,' required"));
+ else
+ avr_operand (opcode, -1, "r", &str, ®);
+ }
+
+ *line = str;
+
+ /* Chunks must follow in a specific order:
+ - Prologue: Exactly one
+ - Epilogue: Any number
+ - Done: Exactly one. */
+ bad |= ISR_CHUNK_Prologue == chunk && avr_isr.prev_chunk != ISR_CHUNK_Done;
+ bad |= ISR_CHUNK_Epilogue == chunk && avr_isr.prev_chunk == ISR_CHUNK_Done;
+ bad |= ISR_CHUNK_Done == chunk && avr_isr.prev_chunk == ISR_CHUNK_Done;
+ if (bad)
+ {
+ if (avr_isr.file)
+ as_bad (_("`%s %d' after `%s %d' from %s:%u"), opcode->name, chunk,
+ opcode->name, avr_isr.prev_chunk, avr_isr.file, avr_isr.line);
+ else
+ as_bad (_("`%s %d' but no chunk open yet"), opcode->name, chunk);
+ }
+
+ if (!had_errors())
+ {
+ /* The longest sequence (prologue) might have up to 6 insns (words):
+
+ push R0
+ in R0, SREG
+ push R0
+ push R1
+ clr R1
+ push Rx
+ */
+ unsigned int size = 2 * 6;
+ fragS *fr;
+
+ switch (chunk)
+ {
+ case ISR_CHUNK_Prologue:
+ avr_isr.need_reg_tmp = 0;
+ avr_isr.need_reg_zero = 0;
+ avr_isr.need_sreg = 0;
+ avr_isr.sym_n_pushed = NULL;
+ /* FALLTHRU */
+
+ case ISR_CHUNK_Epilogue:
+ /* Emit a new rs_machine_dependent fragment into the fragment chain.
+ It will be patched and cleaned up once we see the matching
+ ISR_CHUNK_Done. */
+ frag_wane (frag_now);
+ frag_new (0);
+ frag_more (size);
+
+ frag_now->fr_var = 1;
+ frag_now->fr_offset = size;
+ frag_now->fr_fix = 0;
+ frag_now->fr_type = rs_machine_dependent;
+ frag_now->fr_subtype = chunk;
+ frag_new (size);
+ break;
+
+ case ISR_CHUNK_Done:
+ /* Traverse all frags of the current subseg and turn ones of type
+ rs_machine_dependent into ordinary code as expected by GCC. */
+
+ for (fr = frchain_now->frch_root; fr; fr = fr->fr_next)
+ if (fr->fr_type == rs_machine_dependent)
+ avr_patch_gccisr_frag (fr, reg);
+ break;
+
+ default:
+ abort();
+ break;
+ }
+ } /* !had_errors */
+
+ avr_isr.prev_chunk = chunk;
+ avr_isr.file = as_where (&avr_isr.line);
+}
+
+
+/* Callback used by the function below. Diagnose any dangling stuff from
+ `__gcc_isr', i.e. frags of type rs_machine_dependent. Such frags should
+ have been resolved during parse by ISR_CHUNK_Done. If such a frag is
+ seen, report an error and turn it into something harmless. */
+
+static void
+avr_check_gccisr_done (bfd *abfd ATTRIBUTE_UNUSED,
+ segT section,
+ void *xxx ATTRIBUTE_UNUSED)
+{
+ segment_info_type *info = seg_info (section);
+
+ if (SEG_NORMAL (section)
+ /* BFD may have introduced its own sections without using
+ subseg_new, so it is possible that seg_info is NULL. */
+ && info)
+ {
+ fragS *fr;
+ frchainS *frch;
+
+ for (frch = info->frchainP; frch; frch = frch->frch_next)
+ for (fr = frch->frch_root; fr; fr = fr->fr_next)
+ if (fr->fr_type == rs_machine_dependent)
+ {
+ if (avr_isr.file)
+ as_bad_where (avr_isr.file, avr_isr.line,
+ _("dangling `__gcc_isr %d'"), avr_isr.prev_chunk);
+ else if (!had_errors())
+ as_bad (_("dangling `__gcc_isr'"));
+
+ avr_isr.file = NULL;
+
+ /* Avoid Internal errors due to rs_machine_dependent in the
+ remainder: Turn frag into something harmless. */
+ fr->fr_var = 0;
+ fr->fr_fix = 0;
+ fr->fr_offset = 0;
+ fr->fr_type = rs_fill;
+ fr->fr_subtype = 0;
+ }
+ }
+}
+
+
+/* Implement `md_pre_output_hook' */
+/* Run over all relevant sections and diagnose any dangling `__gcc_isr'.
+ This runs after parsing all inputs but before relaxing and writing. */
+
+void
+avr_pre_output_hook (void)
+{
+ if (avr_opt.have_gccisr)
+ bfd_map_over_sections (stdoutput, avr_check_gccisr_done, NULL);
+}