root / dyngen.h @ 189cd4d4
History | View | Annotate | Download (13.1 kB)
| 1 |
/*
|
|---|---|
| 2 |
* dyngen helpers
|
| 3 |
*
|
| 4 |
* Copyright (c) 2003 Fabrice Bellard
|
| 5 |
*
|
| 6 |
* This library is free software; you can redistribute it and/or
|
| 7 |
* modify it under the terms of the GNU Lesser General Public
|
| 8 |
* License as published by the Free Software Foundation; either
|
| 9 |
* version 2 of the License, or (at your option) any later version.
|
| 10 |
*
|
| 11 |
* This library is distributed in the hope that it will be useful,
|
| 12 |
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
| 13 |
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
| 14 |
* Lesser General Public License for more details.
|
| 15 |
*
|
| 16 |
* You should have received a copy of the GNU Lesser General Public
|
| 17 |
* License along with this library; if not, write to the Free Software
|
| 18 |
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
|
| 19 |
*/
|
| 20 |
|
| 21 |
int __op_param1, __op_param2, __op_param3;
|
| 22 |
#if defined(__sparc__) || defined(__arm__)
|
| 23 |
void __op_gen_label1(){}
|
| 24 |
void __op_gen_label2(){}
|
| 25 |
void __op_gen_label3(){}
|
| 26 |
#else
|
| 27 |
int __op_gen_label1, __op_gen_label2, __op_gen_label3;
|
| 28 |
#endif
|
| 29 |
int __op_jmp0, __op_jmp1, __op_jmp2, __op_jmp3;
|
| 30 |
|
| 31 |
#if defined(__i386__) || defined(__x86_64__) || defined(__s390__)
|
| 32 |
static inline void flush_icache_range(unsigned long start, unsigned long stop) |
| 33 |
{
|
| 34 |
} |
| 35 |
#elif defined(__ia64__)
|
| 36 |
static inline void flush_icache_range(unsigned long start, unsigned long stop) |
| 37 |
{
|
| 38 |
while (start < stop) {
|
| 39 |
asm volatile ("fc %0" :: "r"(start)); |
| 40 |
start += 32;
|
| 41 |
} |
| 42 |
asm volatile (";;sync.i;;srlz.i;;"); |
| 43 |
} |
| 44 |
#elif defined(__powerpc__)
|
| 45 |
|
| 46 |
#define MIN_CACHE_LINE_SIZE 8 /* conservative value */ |
| 47 |
|
| 48 |
static inline void flush_icache_range(unsigned long start, unsigned long stop) |
| 49 |
{
|
| 50 |
unsigned long p; |
| 51 |
|
| 52 |
start &= ~(MIN_CACHE_LINE_SIZE - 1);
|
| 53 |
stop = (stop + MIN_CACHE_LINE_SIZE - 1) & ~(MIN_CACHE_LINE_SIZE - 1); |
| 54 |
|
| 55 |
for (p = start; p < stop; p += MIN_CACHE_LINE_SIZE) {
|
| 56 |
asm volatile ("dcbst 0,%0" : : "r"(p) : "memory"); |
| 57 |
} |
| 58 |
asm volatile ("sync" : : : "memory"); |
| 59 |
for (p = start; p < stop; p += MIN_CACHE_LINE_SIZE) {
|
| 60 |
asm volatile ("icbi 0,%0" : : "r"(p) : "memory"); |
| 61 |
} |
| 62 |
asm volatile ("sync" : : : "memory"); |
| 63 |
asm volatile ("isync" : : : "memory"); |
| 64 |
} |
| 65 |
#elif defined(__alpha__)
|
| 66 |
static inline void flush_icache_range(unsigned long start, unsigned long stop) |
| 67 |
{
|
| 68 |
asm ("imb"); |
| 69 |
} |
| 70 |
#elif defined(__sparc__)
|
| 71 |
static inline void flush_icache_range(unsigned long start, unsigned long stop) |
| 72 |
{
|
| 73 |
unsigned long p; |
| 74 |
|
| 75 |
p = start & ~(8UL - 1UL); |
| 76 |
stop = (stop + (8UL - 1UL)) & ~(8UL - 1UL); |
| 77 |
|
| 78 |
for (; p < stop; p += 8) |
| 79 |
__asm__ __volatile__("flush\t%0" : : "r" (p)); |
| 80 |
} |
| 81 |
#elif defined(__arm__)
|
| 82 |
static inline void flush_icache_range(unsigned long start, unsigned long stop) |
| 83 |
{
|
| 84 |
register unsigned long _beg __asm ("a1") = start; |
| 85 |
register unsigned long _end __asm ("a2") = stop; |
| 86 |
register unsigned long _flg __asm ("a3") = 0; |
| 87 |
__asm __volatile__ ("swi 0x9f0002" : : "r" (_beg), "r" (_end), "r" (_flg)); |
| 88 |
} |
| 89 |
#elif defined(__mc68000)
|
| 90 |
|
| 91 |
# include <asm/cachectl.h> |
| 92 |
static inline void flush_icache_range(unsigned long start, unsigned long stop) |
| 93 |
{
|
| 94 |
cacheflush(start,FLUSH_SCOPE_LINE,FLUSH_CACHE_BOTH,stop-start+16);
|
| 95 |
} |
| 96 |
#elif defined(__mips__)
|
| 97 |
|
| 98 |
#include <sys/cachectl.h> |
| 99 |
static inline void flush_icache_range(unsigned long start, unsigned long stop) |
| 100 |
{
|
| 101 |
_flush_cache ((void *)start, stop - start, BCACHE);
|
| 102 |
} |
| 103 |
#else
|
| 104 |
#error unsupported CPU
|
| 105 |
#endif
|
| 106 |
|
| 107 |
#ifdef __alpha__
|
| 108 |
|
| 109 |
register int gp asm("$29"); |
| 110 |
|
| 111 |
static inline void immediate_ldah(void *p, int val) { |
| 112 |
uint32_t *dest = p; |
| 113 |
long high = ((val >> 16) + ((val >> 15) & 1)) & 0xffff; |
| 114 |
|
| 115 |
*dest &= ~0xffff;
|
| 116 |
*dest |= high; |
| 117 |
*dest |= 31 << 16; |
| 118 |
} |
| 119 |
static inline void immediate_lda(void *dest, int val) { |
| 120 |
*(uint16_t *) dest = val; |
| 121 |
} |
| 122 |
void fix_bsr(void *p, int offset) { |
| 123 |
uint32_t *dest = p; |
| 124 |
*dest &= ~((1 << 21) - 1); |
| 125 |
*dest |= (offset >> 2) & ((1 << 21) - 1); |
| 126 |
} |
| 127 |
|
| 128 |
#endif /* __alpha__ */ |
| 129 |
|
| 130 |
#ifdef __arm__
|
| 131 |
|
| 132 |
#define ARM_LDR_TABLE_SIZE 1024 |
| 133 |
|
| 134 |
typedef struct LDREntry { |
| 135 |
uint8_t *ptr; |
| 136 |
uint32_t *data_ptr; |
| 137 |
unsigned type:2; |
| 138 |
} LDREntry; |
| 139 |
|
| 140 |
static LDREntry arm_ldr_table[1024]; |
| 141 |
static uint32_t arm_data_table[ARM_LDR_TABLE_SIZE];
|
| 142 |
|
| 143 |
extern char exec_loop; |
| 144 |
|
| 145 |
static inline void arm_reloc_pc24(uint32_t *ptr, uint32_t insn, int val) |
| 146 |
{
|
| 147 |
*ptr = (insn & ~0xffffff) | ((insn + ((val - (int)ptr) >> 2)) & 0xffffff); |
| 148 |
} |
| 149 |
|
| 150 |
static uint8_t *arm_flush_ldr(uint8_t *gen_code_ptr,
|
| 151 |
LDREntry *ldr_start, LDREntry *ldr_end, |
| 152 |
uint32_t *data_start, uint32_t *data_end, |
| 153 |
int gen_jmp)
|
| 154 |
{
|
| 155 |
LDREntry *le; |
| 156 |
uint32_t *ptr; |
| 157 |
int offset, data_size, target;
|
| 158 |
uint8_t *data_ptr; |
| 159 |
uint32_t insn; |
| 160 |
uint32_t mask; |
| 161 |
|
| 162 |
data_size = (data_end - data_start) << 2;
|
| 163 |
|
| 164 |
if (gen_jmp) {
|
| 165 |
/* generate branch to skip the data */
|
| 166 |
if (data_size == 0) |
| 167 |
return gen_code_ptr;
|
| 168 |
target = (long)gen_code_ptr + data_size + 4; |
| 169 |
arm_reloc_pc24((uint32_t *)gen_code_ptr, 0xeafffffe, target);
|
| 170 |
gen_code_ptr += 4;
|
| 171 |
} |
| 172 |
|
| 173 |
/* copy the data */
|
| 174 |
data_ptr = gen_code_ptr; |
| 175 |
memcpy(gen_code_ptr, data_start, data_size); |
| 176 |
gen_code_ptr += data_size; |
| 177 |
|
| 178 |
/* patch the ldr to point to the data */
|
| 179 |
for(le = ldr_start; le < ldr_end; le++) {
|
| 180 |
ptr = (uint32_t *)le->ptr; |
| 181 |
offset = ((unsigned long)(le->data_ptr) - (unsigned long)data_start) + |
| 182 |
(unsigned long)data_ptr - |
| 183 |
(unsigned long)ptr - 8; |
| 184 |
if (offset < 0) { |
| 185 |
fprintf(stderr, "Negative constant pool offset\n");
|
| 186 |
abort(); |
| 187 |
} |
| 188 |
switch (le->type) {
|
| 189 |
case 0: /* ldr */ |
| 190 |
mask = ~0x00800fff;
|
| 191 |
if (offset >= 4096) { |
| 192 |
fprintf(stderr, "Bad ldr offset\n");
|
| 193 |
abort(); |
| 194 |
} |
| 195 |
break;
|
| 196 |
case 1: /* ldc */ |
| 197 |
mask = ~0x008000ff;
|
| 198 |
if (offset >= 1024 ) { |
| 199 |
fprintf(stderr, "Bad ldc offset\n");
|
| 200 |
abort(); |
| 201 |
} |
| 202 |
break;
|
| 203 |
case 2: /* add */ |
| 204 |
mask = ~0xfff;
|
| 205 |
if (offset >= 1024 ) { |
| 206 |
fprintf(stderr, "Bad add offset\n");
|
| 207 |
abort(); |
| 208 |
} |
| 209 |
break;
|
| 210 |
default:
|
| 211 |
fprintf(stderr, "Bad pc relative fixup\n");
|
| 212 |
abort(); |
| 213 |
} |
| 214 |
insn = *ptr & mask; |
| 215 |
switch (le->type) {
|
| 216 |
case 0: /* ldr */ |
| 217 |
insn |= offset | 0x00800000;
|
| 218 |
break;
|
| 219 |
case 1: /* ldc */ |
| 220 |
insn |= (offset >> 2) | 0x00800000; |
| 221 |
break;
|
| 222 |
case 2: /* add */ |
| 223 |
insn |= (offset >> 2) | 0xf00; |
| 224 |
break;
|
| 225 |
} |
| 226 |
*ptr = insn; |
| 227 |
} |
| 228 |
return gen_code_ptr;
|
| 229 |
} |
| 230 |
|
| 231 |
#endif /* __arm__ */ |
| 232 |
|
| 233 |
#ifdef __ia64
|
| 234 |
|
| 235 |
/* Patch instruction with "val" where "mask" has 1 bits. */
|
| 236 |
static inline void ia64_patch (uint64_t insn_addr, uint64_t mask, uint64_t val) |
| 237 |
{
|
| 238 |
uint64_t m0, m1, v0, v1, b0, b1, *b = (uint64_t *) (insn_addr & -16);
|
| 239 |
# define insn_mask ((1UL << 41) - 1) |
| 240 |
unsigned long shift; |
| 241 |
|
| 242 |
b0 = b[0]; b1 = b[1]; |
| 243 |
shift = 5 + 41 * (insn_addr % 16); /* 5 template, 3 x 41-bit insns */ |
| 244 |
if (shift >= 64) { |
| 245 |
m1 = mask << (shift - 64);
|
| 246 |
v1 = val << (shift - 64);
|
| 247 |
} else {
|
| 248 |
m0 = mask << shift; m1 = mask >> (64 - shift);
|
| 249 |
v0 = val << shift; v1 = val >> (64 - shift);
|
| 250 |
b[0] = (b0 & ~m0) | (v0 & m0);
|
| 251 |
} |
| 252 |
b[1] = (b1 & ~m1) | (v1 & m1);
|
| 253 |
} |
| 254 |
|
| 255 |
static inline void ia64_patch_imm60 (uint64_t insn_addr, uint64_t val) |
| 256 |
{
|
| 257 |
ia64_patch(insn_addr, |
| 258 |
0x011ffffe000UL,
|
| 259 |
( ((val & 0x0800000000000000UL) >> 23) /* bit 59 -> 36 */ |
| 260 |
| ((val & 0x00000000000fffffUL) << 13) /* bit 0 -> 13 */)); |
| 261 |
ia64_patch(insn_addr - 1, 0x1fffffffffcUL, val >> 18); |
| 262 |
} |
| 263 |
|
| 264 |
static inline void ia64_imm64 (void *insn, uint64_t val) |
| 265 |
{
|
| 266 |
/* Ignore the slot number of the relocation; GCC and Intel
|
| 267 |
toolchains differed for some time on whether IMM64 relocs are
|
| 268 |
against slot 1 (Intel) or slot 2 (GCC). */
|
| 269 |
uint64_t insn_addr = (uint64_t) insn & ~3UL;
|
| 270 |
|
| 271 |
ia64_patch(insn_addr + 2,
|
| 272 |
0x01fffefe000UL,
|
| 273 |
( ((val & 0x8000000000000000UL) >> 27) /* bit 63 -> 36 */ |
| 274 |
| ((val & 0x0000000000200000UL) << 0) /* bit 21 -> 21 */ |
| 275 |
| ((val & 0x00000000001f0000UL) << 6) /* bit 16 -> 22 */ |
| 276 |
| ((val & 0x000000000000ff80UL) << 20) /* bit 7 -> 27 */ |
| 277 |
| ((val & 0x000000000000007fUL) << 13) /* bit 0 -> 13 */) |
| 278 |
); |
| 279 |
ia64_patch(insn_addr + 1, 0x1ffffffffffUL, val >> 22); |
| 280 |
} |
| 281 |
|
| 282 |
static inline void ia64_imm60b (void *insn, uint64_t val) |
| 283 |
{
|
| 284 |
/* Ignore the slot number of the relocation; GCC and Intel
|
| 285 |
toolchains differed for some time on whether IMM64 relocs are
|
| 286 |
against slot 1 (Intel) or slot 2 (GCC). */
|
| 287 |
uint64_t insn_addr = (uint64_t) insn & ~3UL;
|
| 288 |
|
| 289 |
if (val + ((uint64_t) 1 << 59) >= (1UL << 60)) |
| 290 |
fprintf(stderr, "%s: value %ld out of IMM60 range\n",
|
| 291 |
__FUNCTION__, (int64_t) val); |
| 292 |
ia64_patch_imm60(insn_addr + 2, val);
|
| 293 |
} |
| 294 |
|
| 295 |
static inline void ia64_imm22 (void *insn, uint64_t val) |
| 296 |
{
|
| 297 |
if (val + (1 << 21) >= (1 << 22)) |
| 298 |
fprintf(stderr, "%s: value %li out of IMM22 range\n",
|
| 299 |
__FUNCTION__, (int64_t)val); |
| 300 |
ia64_patch((uint64_t) insn, 0x01fffcfe000UL,
|
| 301 |
( ((val & 0x200000UL) << 15) /* bit 21 -> 36 */ |
| 302 |
| ((val & 0x1f0000UL) << 6) /* bit 16 -> 22 */ |
| 303 |
| ((val & 0x00ff80UL) << 20) /* bit 7 -> 27 */ |
| 304 |
| ((val & 0x00007fUL) << 13) /* bit 0 -> 13 */)); |
| 305 |
} |
| 306 |
|
| 307 |
/* Like ia64_imm22(), but also clear bits 20-21. For addl, this has
|
| 308 |
the effect of turning "addl rX=imm22,rY" into "addl
|
| 309 |
rX=imm22,r0". */
|
| 310 |
static inline void ia64_imm22_r0 (void *insn, uint64_t val) |
| 311 |
{
|
| 312 |
if (val + (1 << 21) >= (1 << 22)) |
| 313 |
fprintf(stderr, "%s: value %li out of IMM22 range\n",
|
| 314 |
__FUNCTION__, (int64_t)val); |
| 315 |
ia64_patch((uint64_t) insn, 0x01fffcfe000UL | (0x3UL << 20), |
| 316 |
( ((val & 0x200000UL) << 15) /* bit 21 -> 36 */ |
| 317 |
| ((val & 0x1f0000UL) << 6) /* bit 16 -> 22 */ |
| 318 |
| ((val & 0x00ff80UL) << 20) /* bit 7 -> 27 */ |
| 319 |
| ((val & 0x00007fUL) << 13) /* bit 0 -> 13 */)); |
| 320 |
} |
| 321 |
|
| 322 |
static inline void ia64_imm21b (void *insn, uint64_t val) |
| 323 |
{
|
| 324 |
if (val + (1 << 20) >= (1 << 21)) |
| 325 |
fprintf(stderr, "%s: value %li out of IMM21b range\n",
|
| 326 |
__FUNCTION__, (int64_t)val); |
| 327 |
ia64_patch((uint64_t) insn, 0x11ffffe000UL,
|
| 328 |
( ((val & 0x100000UL) << 16) /* bit 20 -> 36 */ |
| 329 |
| ((val & 0x0fffffUL) << 13) /* bit 0 -> 13 */)); |
| 330 |
} |
| 331 |
|
| 332 |
static inline void ia64_nop_b (void *insn) |
| 333 |
{
|
| 334 |
ia64_patch((uint64_t) insn, (1UL << 41) - 1, 2UL << 37); |
| 335 |
} |
| 336 |
|
| 337 |
static inline void ia64_ldxmov(void *insn, uint64_t val) |
| 338 |
{
|
| 339 |
if (val + (1 << 21) < (1 << 22)) |
| 340 |
ia64_patch((uint64_t) insn, 0x1fff80fe000UL, 8UL << 37); |
| 341 |
} |
| 342 |
|
| 343 |
static inline int ia64_patch_ltoff(void *insn, uint64_t val, |
| 344 |
int relaxable)
|
| 345 |
{
|
| 346 |
if (relaxable && (val + (1 << 21) < (1 << 22))) { |
| 347 |
ia64_imm22_r0(insn, val); |
| 348 |
return 0; |
| 349 |
} |
| 350 |
return 1; |
| 351 |
} |
| 352 |
|
| 353 |
struct ia64_fixup {
|
| 354 |
struct ia64_fixup *next;
|
| 355 |
void *addr; /* address that needs to be patched */ |
| 356 |
long value;
|
| 357 |
}; |
| 358 |
|
| 359 |
#define IA64_PLT(insn, plt_index) \
|
| 360 |
do { \
|
| 361 |
struct ia64_fixup *fixup = alloca(sizeof(*fixup)); \ |
| 362 |
fixup->next = plt_fixes; \ |
| 363 |
plt_fixes = fixup; \ |
| 364 |
fixup->addr = (insn); \ |
| 365 |
fixup->value = (plt_index); \ |
| 366 |
plt_offset[(plt_index)] = 1; \
|
| 367 |
} while (0) |
| 368 |
|
| 369 |
#define IA64_LTOFF(insn, val, relaxable) \
|
| 370 |
do { \
|
| 371 |
if (ia64_patch_ltoff(insn, val, relaxable)) { \
|
| 372 |
struct ia64_fixup *fixup = alloca(sizeof(*fixup)); \ |
| 373 |
fixup->next = ltoff_fixes; \ |
| 374 |
ltoff_fixes = fixup; \ |
| 375 |
fixup->addr = (insn); \ |
| 376 |
fixup->value = (val); \ |
| 377 |
} \ |
| 378 |
} while (0) |
| 379 |
|
| 380 |
static inline void ia64_apply_fixes (uint8_t **gen_code_pp, |
| 381 |
struct ia64_fixup *ltoff_fixes,
|
| 382 |
uint64_t gp, |
| 383 |
struct ia64_fixup *plt_fixes,
|
| 384 |
int num_plts,
|
| 385 |
unsigned long *plt_target, |
| 386 |
unsigned int *plt_offset) |
| 387 |
{
|
| 388 |
static const uint8_t plt_bundle[] = { |
| 389 |
0x04, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, /* nop 0; movl r1=GP */ |
| 390 |
0x00, 0x00, 0x00, 0x20, 0x00, 0x00, 0x00, 0x60, |
| 391 |
|
| 392 |
0x05, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, /* nop 0; brl IP */ |
| 393 |
0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xc0 |
| 394 |
}; |
| 395 |
uint8_t *gen_code_ptr = *gen_code_pp, *plt_start, *got_start; |
| 396 |
uint64_t *vp; |
| 397 |
struct ia64_fixup *fixup;
|
| 398 |
unsigned int offset = 0; |
| 399 |
struct fdesc {
|
| 400 |
long ip;
|
| 401 |
long gp;
|
| 402 |
} *fdesc; |
| 403 |
int i;
|
| 404 |
|
| 405 |
if (plt_fixes) {
|
| 406 |
plt_start = gen_code_ptr; |
| 407 |
|
| 408 |
for (i = 0; i < num_plts; ++i) { |
| 409 |
if (plt_offset[i]) {
|
| 410 |
plt_offset[i] = offset; |
| 411 |
offset += sizeof(plt_bundle);
|
| 412 |
|
| 413 |
fdesc = (struct fdesc *) plt_target[i];
|
| 414 |
memcpy(gen_code_ptr, plt_bundle, sizeof(plt_bundle));
|
| 415 |
ia64_imm64 (gen_code_ptr + 0x02, fdesc->gp);
|
| 416 |
ia64_imm60b(gen_code_ptr + 0x12,
|
| 417 |
(fdesc->ip - (long) (gen_code_ptr + 0x10)) >> 4); |
| 418 |
gen_code_ptr += sizeof(plt_bundle);
|
| 419 |
} |
| 420 |
} |
| 421 |
|
| 422 |
for (fixup = plt_fixes; fixup; fixup = fixup->next)
|
| 423 |
ia64_imm21b(fixup->addr, |
| 424 |
((long) plt_start + plt_offset[fixup->value]
|
| 425 |
- ((long) fixup->addr & ~0xf)) >> 4); |
| 426 |
} |
| 427 |
|
| 428 |
got_start = gen_code_ptr; |
| 429 |
|
| 430 |
/* First, create the GOT: */
|
| 431 |
for (fixup = ltoff_fixes; fixup; fixup = fixup->next) {
|
| 432 |
/* first check if we already have this value in the GOT: */
|
| 433 |
for (vp = (uint64_t *) got_start; vp < (uint64_t *) gen_code_ptr; ++vp)
|
| 434 |
if (*vp == fixup->value)
|
| 435 |
break;
|
| 436 |
if (vp == (uint64_t *) gen_code_ptr) {
|
| 437 |
/* Nope, we need to put the value in the GOT: */
|
| 438 |
*vp = fixup->value; |
| 439 |
gen_code_ptr += 8;
|
| 440 |
} |
| 441 |
ia64_imm22(fixup->addr, (long) vp - gp);
|
| 442 |
} |
| 443 |
/* Keep code ptr aligned. */
|
| 444 |
if ((long) gen_code_ptr & 15) |
| 445 |
gen_code_ptr += 8;
|
| 446 |
*gen_code_pp = gen_code_ptr; |
| 447 |
} |
| 448 |
|
| 449 |
#endif
|