1 | /*
|
---|
2 | * Tiny Code Generator for QEMU
|
---|
3 | *
|
---|
4 | * Copyright (c) 2008 Fabrice Bellard
|
---|
5 | *
|
---|
6 | * Permission is hereby granted, free of charge, to any person obtaining a copy
|
---|
7 | * of this software and associated documentation files (the "Software"), to deal
|
---|
8 | * in the Software without restriction, including without limitation the rights
|
---|
9 | * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
|
---|
10 | * copies of the Software, and to permit persons to whom the Software is
|
---|
11 | * furnished to do so, subject to the following conditions:
|
---|
12 | *
|
---|
13 | * The above copyright notice and this permission notice shall be included in
|
---|
14 | * all copies or substantial portions of the Software.
|
---|
15 | *
|
---|
16 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
---|
17 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
---|
18 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
|
---|
19 | * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
|
---|
20 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
|
---|
21 | * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
|
---|
22 | * THE SOFTWARE.
|
---|
23 | */
|
---|
24 |
|
---|
25 | #ifndef VBOX
|
---|
26 | #include <assert.h>
|
---|
27 | #include <stdarg.h>
|
---|
28 | #include <stdlib.h>
|
---|
29 | #include <stdio.h>
|
---|
30 | #include <string.h>
|
---|
31 | #include <inttypes.h>
|
---|
32 | #else
|
---|
33 | # include <stdio.h>
|
---|
34 | # include "osdep.h"
|
---|
35 | #endif
|
---|
36 |
|
---|
37 | #include "config.h"
|
---|
38 | #include "osdep.h"
|
---|
39 |
|
---|
40 | #include "tcg.h"
|
---|
41 |
|
---|
42 | int __op_param1, __op_param2, __op_param3;
|
---|
43 | #if defined(__sparc__) || defined(__arm__)
|
---|
44 | void __op_gen_label1(){}
|
---|
45 | void __op_gen_label2(){}
|
---|
46 | void __op_gen_label3(){}
|
---|
47 | #else
|
---|
48 | int __op_gen_label1, __op_gen_label2, __op_gen_label3;
|
---|
49 | #endif
|
---|
50 | int __op_jmp0, __op_jmp1, __op_jmp2, __op_jmp3;
|
---|
51 |
|
---|
52 | #if 0
|
---|
53 | #if defined(__s390__)
|
---|
54 | static inline void flush_icache_range(unsigned long start, unsigned long stop)
|
---|
55 | {
|
---|
56 | }
|
---|
57 | #elif defined(__ia64__)
|
---|
58 | static inline void flush_icache_range(unsigned long start, unsigned long stop)
|
---|
59 | {
|
---|
60 | while (start < stop) {
|
---|
61 | asm volatile ("fc %0" :: "r"(start));
|
---|
62 | start += 32;
|
---|
63 | }
|
---|
64 | asm volatile (";;sync.i;;srlz.i;;");
|
---|
65 | }
|
---|
66 | #elif defined(__powerpc__)
|
---|
67 |
|
---|
68 | #define MIN_CACHE_LINE_SIZE 8 /* conservative value */
|
---|
69 |
|
---|
70 | static inline void flush_icache_range(unsigned long start, unsigned long stop)
|
---|
71 | {
|
---|
72 | unsigned long p;
|
---|
73 |
|
---|
74 | start &= ~(MIN_CACHE_LINE_SIZE - 1);
|
---|
75 | stop = (stop + MIN_CACHE_LINE_SIZE - 1) & ~(MIN_CACHE_LINE_SIZE - 1);
|
---|
76 |
|
---|
77 | for (p = start; p < stop; p += MIN_CACHE_LINE_SIZE) {
|
---|
78 | asm volatile ("dcbst 0,%0" : : "r"(p) : "memory");
|
---|
79 | }
|
---|
80 | asm volatile ("sync" : : : "memory");
|
---|
81 | for (p = start; p < stop; p += MIN_CACHE_LINE_SIZE) {
|
---|
82 | asm volatile ("icbi 0,%0" : : "r"(p) : "memory");
|
---|
83 | }
|
---|
84 | asm volatile ("sync" : : : "memory");
|
---|
85 | asm volatile ("isync" : : : "memory");
|
---|
86 | }
|
---|
87 | #elif defined(__alpha__)
|
---|
88 | static inline void flush_icache_range(unsigned long start, unsigned long stop)
|
---|
89 | {
|
---|
90 | asm ("imb");
|
---|
91 | }
|
---|
92 | #elif defined(__sparc__)
|
---|
93 | static inline void flush_icache_range(unsigned long start, unsigned long stop)
|
---|
94 | {
|
---|
95 | unsigned long p;
|
---|
96 |
|
---|
97 | p = start & ~(8UL - 1UL);
|
---|
98 | stop = (stop + (8UL - 1UL)) & ~(8UL - 1UL);
|
---|
99 |
|
---|
100 | for (; p < stop; p += 8)
|
---|
101 | __asm__ __volatile__("flush\t%0" : : "r" (p));
|
---|
102 | }
|
---|
103 | #elif defined(__arm__)
|
---|
104 | static inline void flush_icache_range(unsigned long start, unsigned long stop)
|
---|
105 | {
|
---|
106 | register unsigned long _beg __asm ("a1") = start;
|
---|
107 | register unsigned long _end __asm ("a2") = stop;
|
---|
108 | register unsigned long _flg __asm ("a3") = 0;
|
---|
109 | __asm __volatile__ ("swi 0x9f0002" : : "r" (_beg), "r" (_end), "r" (_flg));
|
---|
110 | }
|
---|
111 | #elif defined(__mc68000)
|
---|
112 |
|
---|
113 | # include <asm/cachectl.h>
|
---|
114 | static inline void flush_icache_range(unsigned long start, unsigned long stop)
|
---|
115 | {
|
---|
116 | cacheflush(start,FLUSH_SCOPE_LINE,FLUSH_CACHE_BOTH,stop-start+16);
|
---|
117 | }
|
---|
118 | #elif defined(__mips__)
|
---|
119 |
|
---|
120 | #include <sys/cachectl.h>
|
---|
121 | static inline void flush_icache_range(unsigned long start, unsigned long stop)
|
---|
122 | {
|
---|
123 | _flush_cache ((void *)start, stop - start, BCACHE);
|
---|
124 | }
|
---|
125 | #else
|
---|
126 | #error unsupported CPU
|
---|
127 | #endif
|
---|
128 |
|
---|
129 | #ifdef __alpha__
|
---|
130 |
|
---|
131 | register int gp asm("$29");
|
---|
132 |
|
---|
133 | static inline void immediate_ldah(void *p, int val) {
|
---|
134 | uint32_t *dest = p;
|
---|
135 | long high = ((val >> 16) + ((val >> 15) & 1)) & 0xffff;
|
---|
136 |
|
---|
137 | *dest &= ~0xffff;
|
---|
138 | *dest |= high;
|
---|
139 | *dest |= 31 << 16;
|
---|
140 | }
|
---|
141 | static inline void immediate_lda(void *dest, int val) {
|
---|
142 | *(uint16_t *) dest = val;
|
---|
143 | }
|
---|
144 | void fix_bsr(void *p, int offset) {
|
---|
145 | uint32_t *dest = p;
|
---|
146 | *dest &= ~((1 << 21) - 1);
|
---|
147 | *dest |= (offset >> 2) & ((1 << 21) - 1);
|
---|
148 | }
|
---|
149 |
|
---|
150 | #endif /* __alpha__ */
|
---|
151 |
|
---|
152 | #ifdef __ia64
|
---|
153 |
|
---|
154 | /* Patch instruction with "val" where "mask" has 1 bits. */
|
---|
155 | static inline void ia64_patch (uint64_t insn_addr, uint64_t mask, uint64_t val)
|
---|
156 | {
|
---|
157 | uint64_t m0, m1, v0, v1, b0, b1, *b = (uint64_t *) (insn_addr & -16);
|
---|
158 | # define insn_mask ((1UL << 41) - 1)
|
---|
159 | unsigned long shift;
|
---|
160 |
|
---|
161 | b0 = b[0]; b1 = b[1];
|
---|
162 | shift = 5 + 41 * (insn_addr % 16); /* 5 template, 3 x 41-bit insns */
|
---|
163 | if (shift >= 64) {
|
---|
164 | m1 = mask << (shift - 64);
|
---|
165 | v1 = val << (shift - 64);
|
---|
166 | } else {
|
---|
167 | m0 = mask << shift; m1 = mask >> (64 - shift);
|
---|
168 | v0 = val << shift; v1 = val >> (64 - shift);
|
---|
169 | b[0] = (b0 & ~m0) | (v0 & m0);
|
---|
170 | }
|
---|
171 | b[1] = (b1 & ~m1) | (v1 & m1);
|
---|
172 | }
|
---|
173 |
|
---|
174 | static inline void ia64_patch_imm60 (uint64_t insn_addr, uint64_t val)
|
---|
175 | {
|
---|
176 | ia64_patch(insn_addr,
|
---|
177 | 0x011ffffe000UL,
|
---|
178 | ( ((val & 0x0800000000000000UL) >> 23) /* bit 59 -> 36 */
|
---|
179 | | ((val & 0x00000000000fffffUL) << 13) /* bit 0 -> 13 */));
|
---|
180 | ia64_patch(insn_addr - 1, 0x1fffffffffcUL, val >> 18);
|
---|
181 | }
|
---|
182 |
|
---|
183 | static inline void ia64_imm64 (void *insn, uint64_t val)
|
---|
184 | {
|
---|
185 | /* Ignore the slot number of the relocation; GCC and Intel
|
---|
186 | toolchains differed for some time on whether IMM64 relocs are
|
---|
187 | against slot 1 (Intel) or slot 2 (GCC). */
|
---|
188 | uint64_t insn_addr = (uint64_t) insn & ~3UL;
|
---|
189 |
|
---|
190 | ia64_patch(insn_addr + 2,
|
---|
191 | 0x01fffefe000UL,
|
---|
192 | ( ((val & 0x8000000000000000UL) >> 27) /* bit 63 -> 36 */
|
---|
193 | | ((val & 0x0000000000200000UL) << 0) /* bit 21 -> 21 */
|
---|
194 | | ((val & 0x00000000001f0000UL) << 6) /* bit 16 -> 22 */
|
---|
195 | | ((val & 0x000000000000ff80UL) << 20) /* bit 7 -> 27 */
|
---|
196 | | ((val & 0x000000000000007fUL) << 13) /* bit 0 -> 13 */)
|
---|
197 | );
|
---|
198 | ia64_patch(insn_addr + 1, 0x1ffffffffffUL, val >> 22);
|
---|
199 | }
|
---|
200 |
|
---|
201 | static inline void ia64_imm60b (void *insn, uint64_t val)
|
---|
202 | {
|
---|
203 | /* Ignore the slot number of the relocation; GCC and Intel
|
---|
204 | toolchains differed for some time on whether IMM64 relocs are
|
---|
205 | against slot 1 (Intel) or slot 2 (GCC). */
|
---|
206 | uint64_t insn_addr = (uint64_t) insn & ~3UL;
|
---|
207 |
|
---|
208 | if (val + ((uint64_t) 1 << 59) >= (1UL << 60))
|
---|
209 | fprintf(stderr, "%s: value %ld out of IMM60 range\n",
|
---|
210 | __FUNCTION__, (int64_t) val);
|
---|
211 | ia64_patch_imm60(insn_addr + 2, val);
|
---|
212 | }
|
---|
213 |
|
---|
214 | static inline void ia64_imm22 (void *insn, uint64_t val)
|
---|
215 | {
|
---|
216 | if (val + (1 << 21) >= (1 << 22))
|
---|
217 | fprintf(stderr, "%s: value %li out of IMM22 range\n",
|
---|
218 | __FUNCTION__, (int64_t)val);
|
---|
219 | ia64_patch((uint64_t) insn, 0x01fffcfe000UL,
|
---|
220 | ( ((val & 0x200000UL) << 15) /* bit 21 -> 36 */
|
---|
221 | | ((val & 0x1f0000UL) << 6) /* bit 16 -> 22 */
|
---|
222 | | ((val & 0x00ff80UL) << 20) /* bit 7 -> 27 */
|
---|
223 | | ((val & 0x00007fUL) << 13) /* bit 0 -> 13 */));
|
---|
224 | }
|
---|
225 |
|
---|
226 | /* Like ia64_imm22(), but also clear bits 20-21. For addl, this has
|
---|
227 | the effect of turning "addl rX=imm22,rY" into "addl
|
---|
228 | rX=imm22,r0". */
|
---|
229 | static inline void ia64_imm22_r0 (void *insn, uint64_t val)
|
---|
230 | {
|
---|
231 | if (val + (1 << 21) >= (1 << 22))
|
---|
232 | fprintf(stderr, "%s: value %li out of IMM22 range\n",
|
---|
233 | __FUNCTION__, (int64_t)val);
|
---|
234 | ia64_patch((uint64_t) insn, 0x01fffcfe000UL | (0x3UL << 20),
|
---|
235 | ( ((val & 0x200000UL) << 15) /* bit 21 -> 36 */
|
---|
236 | | ((val & 0x1f0000UL) << 6) /* bit 16 -> 22 */
|
---|
237 | | ((val & 0x00ff80UL) << 20) /* bit 7 -> 27 */
|
---|
238 | | ((val & 0x00007fUL) << 13) /* bit 0 -> 13 */));
|
---|
239 | }
|
---|
240 |
|
---|
241 | static inline void ia64_imm21b (void *insn, uint64_t val)
|
---|
242 | {
|
---|
243 | if (val + (1 << 20) >= (1 << 21))
|
---|
244 | fprintf(stderr, "%s: value %li out of IMM21b range\n",
|
---|
245 | __FUNCTION__, (int64_t)val);
|
---|
246 | ia64_patch((uint64_t) insn, 0x11ffffe000UL,
|
---|
247 | ( ((val & 0x100000UL) << 16) /* bit 20 -> 36 */
|
---|
248 | | ((val & 0x0fffffUL) << 13) /* bit 0 -> 13 */));
|
---|
249 | }
|
---|
250 |
|
---|
251 | static inline void ia64_nop_b (void *insn)
|
---|
252 | {
|
---|
253 | ia64_patch((uint64_t) insn, (1UL << 41) - 1, 2UL << 37);
|
---|
254 | }
|
---|
255 |
|
---|
256 | static inline void ia64_ldxmov(void *insn, uint64_t val)
|
---|
257 | {
|
---|
258 | if (val + (1 << 21) < (1 << 22))
|
---|
259 | ia64_patch((uint64_t) insn, 0x1fff80fe000UL, 8UL << 37);
|
---|
260 | }
|
---|
261 |
|
---|
262 | static inline int ia64_patch_ltoff(void *insn, uint64_t val,
|
---|
263 | int relaxable)
|
---|
264 | {
|
---|
265 | if (relaxable && (val + (1 << 21) < (1 << 22))) {
|
---|
266 | ia64_imm22_r0(insn, val);
|
---|
267 | return 0;
|
---|
268 | }
|
---|
269 | return 1;
|
---|
270 | }
|
---|
271 |
|
---|
272 | struct ia64_fixup {
|
---|
273 | struct ia64_fixup *next;
|
---|
274 | void *addr; /* address that needs to be patched */
|
---|
275 | long value;
|
---|
276 | };
|
---|
277 |
|
---|
278 | #define IA64_PLT(insn, plt_index) \
|
---|
279 | do { \
|
---|
280 | struct ia64_fixup *fixup = alloca(sizeof(*fixup)); \
|
---|
281 | fixup->next = plt_fixes; \
|
---|
282 | plt_fixes = fixup; \
|
---|
283 | fixup->addr = (insn); \
|
---|
284 | fixup->value = (plt_index); \
|
---|
285 | plt_offset[(plt_index)] = 1; \
|
---|
286 | } while (0)
|
---|
287 |
|
---|
288 | #define IA64_LTOFF(insn, val, relaxable) \
|
---|
289 | do { \
|
---|
290 | if (ia64_patch_ltoff(insn, val, relaxable)) { \
|
---|
291 | struct ia64_fixup *fixup = alloca(sizeof(*fixup)); \
|
---|
292 | fixup->next = ltoff_fixes; \
|
---|
293 | ltoff_fixes = fixup; \
|
---|
294 | fixup->addr = (insn); \
|
---|
295 | fixup->value = (val); \
|
---|
296 | } \
|
---|
297 | } while (0)
|
---|
298 |
|
---|
299 | static inline void ia64_apply_fixes (uint8_t **gen_code_pp,
|
---|
300 | struct ia64_fixup *ltoff_fixes,
|
---|
301 | uint64_t gp,
|
---|
302 | struct ia64_fixup *plt_fixes,
|
---|
303 | int num_plts,
|
---|
304 | unsigned long *plt_target,
|
---|
305 | unsigned int *plt_offset)
|
---|
306 | {
|
---|
307 | static const uint8_t plt_bundle[] = {
|
---|
308 | 0x04, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, /* nop 0; movl r1=GP */
|
---|
309 | 0x00, 0x00, 0x00, 0x20, 0x00, 0x00, 0x00, 0x60,
|
---|
310 |
|
---|
311 | 0x05, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, /* nop 0; brl IP */
|
---|
312 | 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xc0
|
---|
313 | };
|
---|
314 | uint8_t *gen_code_ptr = *gen_code_pp, *plt_start, *got_start;
|
---|
315 | uint64_t *vp;
|
---|
316 | struct ia64_fixup *fixup;
|
---|
317 | unsigned int offset = 0;
|
---|
318 | struct fdesc {
|
---|
319 | long ip;
|
---|
320 | long gp;
|
---|
321 | } *fdesc;
|
---|
322 | int i;
|
---|
323 |
|
---|
324 | if (plt_fixes) {
|
---|
325 | plt_start = gen_code_ptr;
|
---|
326 |
|
---|
327 | for (i = 0; i < num_plts; ++i) {
|
---|
328 | if (plt_offset[i]) {
|
---|
329 | plt_offset[i] = offset;
|
---|
330 | offset += sizeof(plt_bundle);
|
---|
331 |
|
---|
332 | fdesc = (struct fdesc *) plt_target[i];
|
---|
333 | memcpy(gen_code_ptr, plt_bundle, sizeof(plt_bundle));
|
---|
334 | ia64_imm64 (gen_code_ptr + 0x02, fdesc->gp);
|
---|
335 | ia64_imm60b(gen_code_ptr + 0x12,
|
---|
336 | (fdesc->ip - (long) (gen_code_ptr + 0x10)) >> 4);
|
---|
337 | gen_code_ptr += sizeof(plt_bundle);
|
---|
338 | }
|
---|
339 | }
|
---|
340 |
|
---|
341 | for (fixup = plt_fixes; fixup; fixup = fixup->next)
|
---|
342 | ia64_imm21b(fixup->addr,
|
---|
343 | ((long) plt_start + plt_offset[fixup->value]
|
---|
344 | - ((long) fixup->addr & ~0xf)) >> 4);
|
---|
345 | }
|
---|
346 |
|
---|
347 | got_start = gen_code_ptr;
|
---|
348 |
|
---|
349 | /* First, create the GOT: */
|
---|
350 | for (fixup = ltoff_fixes; fixup; fixup = fixup->next) {
|
---|
351 | /* first check if we already have this value in the GOT: */
|
---|
352 | for (vp = (uint64_t *) got_start; vp < (uint64_t *) gen_code_ptr; ++vp)
|
---|
353 | if (*vp == fixup->value)
|
---|
354 | break;
|
---|
355 | if (vp == (uint64_t *) gen_code_ptr) {
|
---|
356 | /* Nope, we need to put the value in the GOT: */
|
---|
357 | *vp = fixup->value;
|
---|
358 | gen_code_ptr += 8;
|
---|
359 | }
|
---|
360 | ia64_imm22(fixup->addr, (long) vp - gp);
|
---|
361 | }
|
---|
362 | /* Keep code ptr aligned. */
|
---|
363 | if ((long) gen_code_ptr & 15)
|
---|
364 | gen_code_ptr += 8;
|
---|
365 | *gen_code_pp = gen_code_ptr;
|
---|
366 | }
|
---|
367 | #endif
|
---|
368 | #endif
|
---|
369 |
|
---|
370 | #ifdef CONFIG_DYNGEN_OP
|
---|
371 |
|
---|
372 | #if defined __hppa__
|
---|
373 | struct hppa_branch_stub {
|
---|
374 | uint32_t *location;
|
---|
375 | long target;
|
---|
376 | struct hppa_branch_stub *next;
|
---|
377 | };
|
---|
378 |
|
---|
379 | #define HPPA_RECORD_BRANCH(LIST, LOC, TARGET) \
|
---|
380 | do { \
|
---|
381 | struct hppa_branch_stub *stub = alloca(sizeof(struct hppa_branch_stub)); \
|
---|
382 | stub->location = LOC; \
|
---|
383 | stub->target = TARGET; \
|
---|
384 | stub->next = LIST; \
|
---|
385 | LIST = stub; \
|
---|
386 | } while (0)
|
---|
387 |
|
---|
388 | static inline void hppa_process_stubs(struct hppa_branch_stub *stub,
|
---|
389 | uint8_t **gen_code_pp)
|
---|
390 | {
|
---|
391 | uint32_t *s = (uint32_t *)*gen_code_pp;
|
---|
392 | uint32_t *p = s + 1;
|
---|
393 |
|
---|
394 | if (!stub) return;
|
---|
395 |
|
---|
396 | for (; stub != NULL; stub = stub->next) {
|
---|
397 | unsigned long l = (unsigned long)p;
|
---|
398 | /* stub:
|
---|
399 | * ldil L'target, %r1
|
---|
400 | * be,n R'target(%sr4,%r1)
|
---|
401 | */
|
---|
402 | *p++ = 0x20200000 | reassemble_21(lrsel(stub->target, 0));
|
---|
403 | *p++ = 0xe0202002 | (reassemble_17(rrsel(stub->target, 0) >> 2));
|
---|
404 | hppa_patch17f(stub->location, l, 0);
|
---|
405 | }
|
---|
406 | /* b,l,n stub,%r0 */
|
---|
407 | *s = 0xe8000002 | reassemble_17((p - s) - 2);
|
---|
408 | *gen_code_pp = (uint8_t *)p;
|
---|
409 | }
|
---|
410 | #endif /* __hppa__ */
|
---|
411 |
|
---|
412 | const TCGArg *dyngen_op(TCGContext *s, int opc, const TCGArg *opparam_ptr)
|
---|
413 | {
|
---|
414 | uint8_t *gen_code_ptr;
|
---|
415 |
|
---|
416 | #ifdef __hppa__
|
---|
417 | struct hppa_branch_stub *hppa_stubs = NULL;
|
---|
418 | #endif
|
---|
419 |
|
---|
420 | gen_code_ptr = s->code_ptr;
|
---|
421 | switch(opc) {
|
---|
422 |
|
---|
423 | /* op.h is dynamically generated by dyngen.c from op.c */
|
---|
424 | #include "op.h"
|
---|
425 |
|
---|
426 | default:
|
---|
427 | tcg_abort();
|
---|
428 | }
|
---|
429 |
|
---|
430 | #ifdef __hppa__
|
---|
431 | hppa_process_stubs(hppa_stubs, &gen_code_ptr);
|
---|
432 | #endif
|
---|
433 |
|
---|
434 | s->code_ptr = gen_code_ptr;
|
---|
435 | return opparam_ptr;
|
---|
436 | }
|
---|
437 | #endif
|
---|