1 /*
2 ** DynASM ARM encoding engine.
3 ** Copyright (C) 2005-2021 Mike Pall. All rights reserved.
4 ** Released under the MIT license. See dynasm.lua for full copyright notice.
5 */
6 
7 #include <stddef.h>
8 #include <stdarg.h>
9 #include <string.h>
10 #include <stdlib.h>
11 
12 #define DASM_ARCH		"arm"
13 
14 #ifndef DASM_EXTERN
15 #define DASM_EXTERN(a,b,c,d)	0
16 #endif
17 
18 /* Action definitions. */
19 enum {
20   DASM_STOP, DASM_SECTION, DASM_ESC, DASM_REL_EXT,
21   /* The following actions need a buffer position. */
22   DASM_ALIGN, DASM_REL_LG, DASM_LABEL_LG,
23   /* The following actions also have an argument. */
24   DASM_REL_PC, DASM_LABEL_PC,
25   DASM_IMM, DASM_IMM12, DASM_IMM16, DASM_IMML8, DASM_IMML12, DASM_IMMV8,
26   DASM__MAX
27 };
28 
29 /* Maximum number of section buffer positions for a single dasm_put() call. */
30 #define DASM_MAXSECPOS		25
31 
32 /* DynASM encoder status codes. Action list offset or number are or'ed in. */
33 #define DASM_S_OK		0x00000000
34 #define DASM_S_NOMEM		0x01000000
35 #define DASM_S_PHASE		0x02000000
36 #define DASM_S_MATCH_SEC	0x03000000
37 #define DASM_S_RANGE_I		0x11000000
38 #define DASM_S_RANGE_SEC	0x12000000
39 #define DASM_S_RANGE_LG		0x13000000
40 #define DASM_S_RANGE_PC		0x14000000
41 #define DASM_S_RANGE_REL	0x15000000
42 #define DASM_S_UNDEF_LG		0x21000000
43 #define DASM_S_UNDEF_PC		0x22000000
44 
45 /* Macros to convert positions (8 bit section + 24 bit index). */
46 #define DASM_POS2IDX(pos)	((pos)&0x00ffffff)
47 #define DASM_POS2BIAS(pos)	((pos)&0xff000000)
48 #define DASM_SEC2POS(sec)	((sec)<<24)
49 #define DASM_POS2SEC(pos)	((pos)>>24)
50 #define DASM_POS2PTR(D, pos)	(D->sections[DASM_POS2SEC(pos)].rbuf + (pos))
51 
52 /* Action list type. */
53 typedef const unsigned int *dasm_ActList;
54 
55 /* Per-section structure. */
56 typedef struct dasm_Section {
57   int *rbuf;		/* Biased buffer pointer (negative section bias). */
58   int *buf;		/* True buffer pointer. */
59   size_t bsize;		/* Buffer size in bytes. */
60   int pos;		/* Biased buffer position. */
61   int epos;		/* End of biased buffer position - max single put. */
62   int ofs;		/* Byte offset into section. */
63 } dasm_Section;
64 
65 /* Core structure holding the DynASM encoding state. */
66 struct dasm_State {
67   size_t psize;			/* Allocated size of this structure. */
68   dasm_ActList actionlist;	/* Current actionlist pointer. */
69   int *lglabels;		/* Local/global chain/pos ptrs. */
70   size_t lgsize;
71   int *pclabels;		/* PC label chains/pos ptrs. */
72   size_t pcsize;
73   void **globals;		/* Array of globals (bias -10). */
74   dasm_Section *section;	/* Pointer to active section. */
75   size_t codesize;		/* Total size of all code sections. */
76   int maxsection;		/* 0 <= sectionidx < maxsection. */
77   int status;			/* Status code. */
78   dasm_Section sections[1];	/* All sections. Alloc-extended. */
79 };
80 
81 /* The size of the core structure depends on the max. number of sections. */
82 #define DASM_PSZ(ms)	(sizeof(dasm_State)+(ms-1)*sizeof(dasm_Section))
83 
84 
85 /* Initialize DynASM state. */
dasm_init(Dst_DECL,int maxsection)86 void dasm_init(Dst_DECL, int maxsection)
87 {
88   dasm_State *D;
89   size_t psz = 0;
90   int i;
91   Dst_REF = NULL;
92   DASM_M_GROW(Dst, struct dasm_State, Dst_REF, psz, DASM_PSZ(maxsection));
93   D = Dst_REF;
94   D->psize = psz;
95   D->lglabels = NULL;
96   D->lgsize = 0;
97   D->pclabels = NULL;
98   D->pcsize = 0;
99   D->globals = NULL;
100   D->maxsection = maxsection;
101   for (i = 0; i < maxsection; i++) {
102     D->sections[i].buf = NULL;  /* Need this for pass3. */
103     D->sections[i].rbuf = D->sections[i].buf - DASM_SEC2POS(i);
104     D->sections[i].bsize = 0;
105     D->sections[i].epos = 0;  /* Wrong, but is recalculated after resize. */
106   }
107 }
108 
109 /* Free DynASM state. */
dasm_free(Dst_DECL)110 void dasm_free(Dst_DECL)
111 {
112   dasm_State *D = Dst_REF;
113   int i;
114   for (i = 0; i < D->maxsection; i++)
115     if (D->sections[i].buf)
116       DASM_M_FREE(Dst, D->sections[i].buf, D->sections[i].bsize);
117   if (D->pclabels) DASM_M_FREE(Dst, D->pclabels, D->pcsize);
118   if (D->lglabels) DASM_M_FREE(Dst, D->lglabels, D->lgsize);
119   DASM_M_FREE(Dst, D, D->psize);
120 }
121 
122 /* Setup global label array. Must be called before dasm_setup(). */
dasm_setupglobal(Dst_DECL,void ** gl,unsigned int maxgl)123 void dasm_setupglobal(Dst_DECL, void **gl, unsigned int maxgl)
124 {
125   dasm_State *D = Dst_REF;
126   D->globals = gl - 10;  /* Negative bias to compensate for locals. */
127   DASM_M_GROW(Dst, int, D->lglabels, D->lgsize, (10+maxgl)*sizeof(int));
128 }
129 
130 /* Grow PC label array. Can be called after dasm_setup(), too. */
dasm_growpc(Dst_DECL,unsigned int maxpc)131 void dasm_growpc(Dst_DECL, unsigned int maxpc)
132 {
133   dasm_State *D = Dst_REF;
134   size_t osz = D->pcsize;
135   DASM_M_GROW(Dst, int, D->pclabels, D->pcsize, maxpc*sizeof(int));
136   memset((void *)(((unsigned char *)D->pclabels)+osz), 0, D->pcsize-osz);
137 }
138 
139 /* Setup encoder. */
dasm_setup(Dst_DECL,const void * actionlist)140 void dasm_setup(Dst_DECL, const void *actionlist)
141 {
142   dasm_State *D = Dst_REF;
143   int i;
144   D->actionlist = (dasm_ActList)actionlist;
145   D->status = DASM_S_OK;
146   D->section = &D->sections[0];
147   memset((void *)D->lglabels, 0, D->lgsize);
148   if (D->pclabels) memset((void *)D->pclabels, 0, D->pcsize);
149   for (i = 0; i < D->maxsection; i++) {
150     D->sections[i].pos = DASM_SEC2POS(i);
151     D->sections[i].ofs = 0;
152   }
153 }
154 
155 
156 #ifdef DASM_CHECKS
157 #define CK(x, st) \
158   do { if (!(x)) { \
159     D->status = DASM_S_##st|(p-D->actionlist-1); return; } } while (0)
160 #define CKPL(kind, st) \
161   do { if ((size_t)((char *)pl-(char *)D->kind##labels) >= D->kind##size) { \
162     D->status = DASM_S_RANGE_##st|(p-D->actionlist-1); return; } } while (0)
163 #else
164 #define CK(x, st)	((void)0)
165 #define CKPL(kind, st)	((void)0)
166 #endif
167 
dasm_imm12(unsigned int n)168 static int dasm_imm12(unsigned int n)
169 {
170   int i;
171   for (i = 0; i < 16; i++, n = (n << 2) | (n >> 30))
172     if (n <= 255) return (int)(n + (i << 8));
173   return -1;
174 }
175 
176 /* Pass 1: Store actions and args, link branches/labels, estimate offsets. */
dasm_put(Dst_DECL,int start,...)177 void dasm_put(Dst_DECL, int start, ...)
178 {
179   va_list ap;
180   dasm_State *D = Dst_REF;
181   dasm_ActList p = D->actionlist + start;
182   dasm_Section *sec = D->section;
183   int pos = sec->pos, ofs = sec->ofs;
184   int *b;
185 
186   if (pos >= sec->epos) {
187     DASM_M_GROW(Dst, int, sec->buf, sec->bsize,
188       sec->bsize + 2*DASM_MAXSECPOS*sizeof(int));
189     sec->rbuf = sec->buf - DASM_POS2BIAS(pos);
190     sec->epos = (int)sec->bsize/sizeof(int) - DASM_MAXSECPOS+DASM_POS2BIAS(pos);
191   }
192 
193   b = sec->rbuf;
194   b[pos++] = start;
195 
196   va_start(ap, start);
197   while (1) {
198     unsigned int ins = *p++;
199     unsigned int action = (ins >> 16);
200     if (action >= DASM__MAX) {
201       ofs += 4;
202     } else {
203       int *pl, n = action >= DASM_REL_PC ? va_arg(ap, int) : 0;
204       switch (action) {
205       case DASM_STOP: goto stop;
206       case DASM_SECTION:
207 	n = (ins & 255); CK(n < D->maxsection, RANGE_SEC);
208 	D->section = &D->sections[n]; goto stop;
209       case DASM_ESC: p++; ofs += 4; break;
210       case DASM_REL_EXT: break;
211       case DASM_ALIGN: ofs += (ins & 255); b[pos++] = ofs; break;
212       case DASM_REL_LG:
213 	n = (ins & 2047) - 10; pl = D->lglabels + n;
214 	/* Bkwd rel or global. */
215 	if (n >= 0) { CK(n>=10||*pl<0, RANGE_LG); CKPL(lg, LG); goto putrel; }
216 	pl += 10; n = *pl;
217 	if (n < 0) n = 0;  /* Start new chain for fwd rel if label exists. */
218 	goto linkrel;
219       case DASM_REL_PC:
220 	pl = D->pclabels + n; CKPL(pc, PC);
221       putrel:
222 	n = *pl;
223 	if (n < 0) {  /* Label exists. Get label pos and store it. */
224 	  b[pos] = -n;
225 	} else {
226       linkrel:
227 	  b[pos] = n;  /* Else link to rel chain, anchored at label. */
228 	  *pl = pos;
229 	}
230 	pos++;
231 	break;
232       case DASM_LABEL_LG:
233 	pl = D->lglabels + (ins & 2047) - 10; CKPL(lg, LG); goto putlabel;
234       case DASM_LABEL_PC:
235 	pl = D->pclabels + n; CKPL(pc, PC);
236       putlabel:
237 	n = *pl;  /* n > 0: Collapse rel chain and replace with label pos. */
238 	while (n > 0) { int *pb = DASM_POS2PTR(D, n); n = *pb; *pb = pos;
239 	}
240 	*pl = -pos;  /* Label exists now. */
241 	b[pos++] = ofs;  /* Store pass1 offset estimate. */
242 	break;
243       case DASM_IMM:
244       case DASM_IMM16:
245 #ifdef DASM_CHECKS
246 	CK((n & ((1<<((ins>>10)&31))-1)) == 0, RANGE_I);
247 	if ((ins & 0x8000))
248 	  CK(((n + (1<<(((ins>>5)&31)-1)))>>((ins>>5)&31)) == 0, RANGE_I);
249 	else
250 	  CK((n>>((ins>>5)&31)) == 0, RANGE_I);
251 #endif
252 	b[pos++] = n;
253 	break;
254       case DASM_IMMV8:
255 	CK((n & 3) == 0, RANGE_I);
256 	n >>= 2;
257 	/* fallthrough */
258       case DASM_IMML8:
259       case DASM_IMML12:
260 	CK(n >= 0 ? ((n>>((ins>>5)&31)) == 0) :
261 		    (((-n)>>((ins>>5)&31)) == 0), RANGE_I);
262 	b[pos++] = n;
263 	break;
264       case DASM_IMM12:
265 	CK(dasm_imm12((unsigned int)n) != -1, RANGE_I);
266 	b[pos++] = n;
267 	break;
268       }
269     }
270   }
271 stop:
272   va_end(ap);
273   sec->pos = pos;
274   sec->ofs = ofs;
275 }
276 #undef CK
277 
278 /* Pass 2: Link sections, shrink aligns, fix label offsets. */
dasm_link(Dst_DECL,size_t * szp)279 int dasm_link(Dst_DECL, size_t *szp)
280 {
281   dasm_State *D = Dst_REF;
282   int secnum;
283   int ofs = 0;
284 
285 #ifdef DASM_CHECKS
286   *szp = 0;
287   if (D->status != DASM_S_OK) return D->status;
288   {
289     int pc;
290     for (pc = 0; pc*sizeof(int) < D->pcsize; pc++)
291       if (D->pclabels[pc] > 0) return DASM_S_UNDEF_PC|pc;
292   }
293 #endif
294 
295   { /* Handle globals not defined in this translation unit. */
296     int idx;
297     for (idx = 10; idx*sizeof(int) < D->lgsize; idx++) {
298       int n = D->lglabels[idx];
299       /* Undefined label: Collapse rel chain and replace with marker (< 0). */
300       while (n > 0) { int *pb = DASM_POS2PTR(D, n); n = *pb; *pb = -idx; }
301     }
302   }
303 
304   /* Combine all code sections. No support for data sections (yet). */
305   for (secnum = 0; secnum < D->maxsection; secnum++) {
306     dasm_Section *sec = D->sections + secnum;
307     int *b = sec->rbuf;
308     int pos = DASM_SEC2POS(secnum);
309     int lastpos = sec->pos;
310 
311     while (pos != lastpos) {
312       dasm_ActList p = D->actionlist + b[pos++];
313       while (1) {
314 	unsigned int ins = *p++;
315 	unsigned int action = (ins >> 16);
316 	switch (action) {
317 	case DASM_STOP: case DASM_SECTION: goto stop;
318 	case DASM_ESC: p++; break;
319 	case DASM_REL_EXT: break;
320 	case DASM_ALIGN: ofs -= (b[pos++] + ofs) & (ins & 255); break;
321 	case DASM_REL_LG: case DASM_REL_PC: pos++; break;
322 	case DASM_LABEL_LG: case DASM_LABEL_PC: b[pos++] += ofs; break;
323 	case DASM_IMM: case DASM_IMM12: case DASM_IMM16:
324 	case DASM_IMML8: case DASM_IMML12: case DASM_IMMV8: pos++; break;
325 	}
326       }
327       stop: (void)0;
328     }
329     ofs += sec->ofs;  /* Next section starts right after current section. */
330   }
331 
332   D->codesize = ofs;  /* Total size of all code sections */
333   *szp = ofs;
334   return DASM_S_OK;
335 }
336 
337 #ifdef DASM_CHECKS
338 #define CK(x, st) \
339   do { if (!(x)) return DASM_S_##st|(p-D->actionlist-1); } while (0)
340 #else
341 #define CK(x, st)	((void)0)
342 #endif
343 
344 /* Pass 3: Encode sections. */
dasm_encode(Dst_DECL,void * buffer)345 int dasm_encode(Dst_DECL, void *buffer)
346 {
347   dasm_State *D = Dst_REF;
348   char *base = (char *)buffer;
349   unsigned int *cp = (unsigned int *)buffer;
350   int secnum;
351 
352   /* Encode all code sections. No support for data sections (yet). */
353   for (secnum = 0; secnum < D->maxsection; secnum++) {
354     dasm_Section *sec = D->sections + secnum;
355     int *b = sec->buf;
356     int *endb = sec->rbuf + sec->pos;
357 
358     while (b != endb) {
359       dasm_ActList p = D->actionlist + *b++;
360       while (1) {
361 	unsigned int ins = *p++;
362 	unsigned int action = (ins >> 16);
363 	int n = (action >= DASM_ALIGN && action < DASM__MAX) ? *b++ : 0;
364 	switch (action) {
365 	case DASM_STOP: case DASM_SECTION: goto stop;
366 	case DASM_ESC: *cp++ = *p++; break;
367 	case DASM_REL_EXT:
368 	  n = DASM_EXTERN(Dst, (unsigned char *)cp, (ins&2047), !(ins&2048));
369 	  goto patchrel;
370 	case DASM_ALIGN:
371 	  ins &= 255; while ((((char *)cp - base) & ins)) *cp++ = 0xe1a00000;
372 	  break;
373 	case DASM_REL_LG:
374 	  if (n < 0) {
375 	    n = (int)((ptrdiff_t)D->globals[-n] - (ptrdiff_t)cp - 4);
376 	    goto patchrel;
377 	  }
378 	  /* fallthrough */
379 	case DASM_REL_PC:
380 	  CK(n >= 0, UNDEF_PC);
381 	  n = *DASM_POS2PTR(D, n) - (int)((char *)cp - base) - 4;
382 	patchrel:
383 	  if ((ins & 0x800) == 0) {
384 	    CK((n & 3) == 0 && ((n+0x02000000) >> 26) == 0, RANGE_REL);
385 	    cp[-1] |= ((n >> 2) & 0x00ffffff);
386 	  } else if ((ins & 0x1000)) {
387 	    CK((n & 3) == 0 && -256 <= n && n <= 256, RANGE_REL);
388 	    goto patchimml8;
389 	  } else if ((ins & 0x2000) == 0) {
390 	    CK((n & 3) == 0 && -4096 <= n && n <= 4096, RANGE_REL);
391 	    goto patchimml;
392 	  } else {
393 	    CK((n & 3) == 0 && -1020 <= n && n <= 1020, RANGE_REL);
394 	    n >>= 2;
395 	    goto patchimml;
396 	  }
397 	  break;
398 	case DASM_LABEL_LG:
399 	  ins &= 2047; if (ins >= 20) D->globals[ins-10] = (void *)(base + n);
400 	  break;
401 	case DASM_LABEL_PC: break;
402 	case DASM_IMM:
403 	  cp[-1] |= ((n>>((ins>>10)&31)) & ((1<<((ins>>5)&31))-1)) << (ins&31);
404 	  break;
405 	case DASM_IMM12:
406 	  cp[-1] |= dasm_imm12((unsigned int)n);
407 	  break;
408 	case DASM_IMM16:
409 	  cp[-1] |= ((n & 0xf000) << 4) | (n & 0x0fff);
410 	  break;
411 	case DASM_IMML8: patchimml8:
412 	  cp[-1] |= n >= 0 ? (0x00800000 | (n & 0x0f) | ((n & 0xf0) << 4)) :
413 			     ((-n & 0x0f) | ((-n & 0xf0) << 4));
414 	  break;
415 	case DASM_IMML12: case DASM_IMMV8: patchimml:
416 	  cp[-1] |= n >= 0 ? (0x00800000 | n) : (-n);
417 	  break;
418 	default: *cp++ = ins; break;
419 	}
420       }
421       stop: (void)0;
422     }
423   }
424 
425   if (base + D->codesize != (char *)cp)  /* Check for phase errors. */
426     return DASM_S_PHASE;
427   return DASM_S_OK;
428 }
429 #undef CK
430 
431 /* Get PC label offset. */
dasm_getpclabel(Dst_DECL,unsigned int pc)432 int dasm_getpclabel(Dst_DECL, unsigned int pc)
433 {
434   dasm_State *D = Dst_REF;
435   if (pc*sizeof(int) < D->pcsize) {
436     int pos = D->pclabels[pc];
437     if (pos < 0) return *DASM_POS2PTR(D, -pos);
438     if (pos > 0) return -1;  /* Undefined. */
439   }
440   return -2;  /* Unused or out of range. */
441 }
442 
443 #ifdef DASM_CHECKS
444 /* Optional sanity checker to call between isolated encoding steps. */
dasm_checkstep(Dst_DECL,int secmatch)445 int dasm_checkstep(Dst_DECL, int secmatch)
446 {
447   dasm_State *D = Dst_REF;
448   if (D->status == DASM_S_OK) {
449     int i;
450     for (i = 1; i <= 9; i++) {
451       if (D->lglabels[i] > 0) { D->status = DASM_S_UNDEF_LG|i; break; }
452       D->lglabels[i] = 0;
453     }
454   }
455   if (D->status == DASM_S_OK && secmatch >= 0 &&
456       D->section != &D->sections[secmatch])
457     D->status = DASM_S_MATCH_SEC|(D->section-D->sections);
458   return D->status;
459 }
460 #endif
461 
462