xref: /linux/arch/powerpc/math-emu/math.c (revision ca55b2fef3a9373fcfc30f82fd26bc7fccbda732)
1 /*
2  * Copyright (C) 1999  Eddie C. Dost  (ecd@atecom.com)
3  */
4 
5 #include <linux/types.h>
6 #include <linux/sched.h>
7 
8 #include <asm/uaccess.h>
9 #include <asm/reg.h>
10 #include <asm/switch_to.h>
11 
12 #include <asm/sfp-machine.h>
13 #include <math-emu/double.h>
14 
15 #define FLOATFUNC(x)	extern int x(void *, void *, void *, void *)
16 
17 /* The instructions list which may be not implemented by a hardware FPU */
18 FLOATFUNC(fre);
19 FLOATFUNC(frsqrtes);
20 FLOATFUNC(fsqrt);
21 FLOATFUNC(fsqrts);
22 FLOATFUNC(mtfsf);
23 FLOATFUNC(mtfsfi);
24 
25 #ifdef CONFIG_MATH_EMULATION_HW_UNIMPLEMENTED
26 #undef FLOATFUNC(x)
27 #define FLOATFUNC(x)	static inline int x(void *op1, void *op2, void *op3, \
28 						 void *op4) { }
29 #endif
30 
31 FLOATFUNC(fadd);
32 FLOATFUNC(fadds);
33 FLOATFUNC(fdiv);
34 FLOATFUNC(fdivs);
35 FLOATFUNC(fmul);
36 FLOATFUNC(fmuls);
37 FLOATFUNC(fsub);
38 FLOATFUNC(fsubs);
39 
40 FLOATFUNC(fmadd);
41 FLOATFUNC(fmadds);
42 FLOATFUNC(fmsub);
43 FLOATFUNC(fmsubs);
44 FLOATFUNC(fnmadd);
45 FLOATFUNC(fnmadds);
46 FLOATFUNC(fnmsub);
47 FLOATFUNC(fnmsubs);
48 
49 FLOATFUNC(fctiw);
50 FLOATFUNC(fctiwz);
51 FLOATFUNC(frsp);
52 
53 FLOATFUNC(fcmpo);
54 FLOATFUNC(fcmpu);
55 
56 FLOATFUNC(mcrfs);
57 FLOATFUNC(mffs);
58 FLOATFUNC(mtfsb0);
59 FLOATFUNC(mtfsb1);
60 
61 FLOATFUNC(lfd);
62 FLOATFUNC(lfs);
63 
64 FLOATFUNC(stfd);
65 FLOATFUNC(stfs);
66 FLOATFUNC(stfiwx);
67 
68 FLOATFUNC(fabs);
69 FLOATFUNC(fmr);
70 FLOATFUNC(fnabs);
71 FLOATFUNC(fneg);
72 
73 /* Optional */
74 FLOATFUNC(fres);
75 FLOATFUNC(frsqrte);
76 FLOATFUNC(fsel);
77 
78 
79 #define OP31		0x1f		/*   31 */
80 #define LFS		0x30		/*   48 */
81 #define LFSU		0x31		/*   49 */
82 #define LFD		0x32		/*   50 */
83 #define LFDU		0x33		/*   51 */
84 #define STFS		0x34		/*   52 */
85 #define STFSU		0x35		/*   53 */
86 #define STFD		0x36		/*   54 */
87 #define STFDU		0x37		/*   55 */
88 #define OP59		0x3b		/*   59 */
89 #define OP63		0x3f		/*   63 */
90 
91 /* Opcode 31: */
92 /* X-Form: */
93 #define LFSX		0x217		/*  535 */
94 #define LFSUX		0x237		/*  567 */
95 #define LFDX		0x257		/*  599 */
96 #define LFDUX		0x277		/*  631 */
97 #define STFSX		0x297		/*  663 */
98 #define STFSUX		0x2b7		/*  695 */
99 #define STFDX		0x2d7		/*  727 */
100 #define STFDUX		0x2f7		/*  759 */
101 #define STFIWX		0x3d7		/*  983 */
102 
103 /* Opcode 59: */
104 /* A-Form: */
105 #define FDIVS		0x012		/*   18 */
106 #define FSUBS		0x014		/*   20 */
107 #define FADDS		0x015		/*   21 */
108 #define FSQRTS		0x016		/*   22 */
109 #define FRES		0x018		/*   24 */
110 #define FMULS		0x019		/*   25 */
111 #define FRSQRTES	0x01a		/*   26 */
112 #define FMSUBS		0x01c		/*   28 */
113 #define FMADDS		0x01d		/*   29 */
114 #define FNMSUBS		0x01e		/*   30 */
115 #define FNMADDS		0x01f		/*   31 */
116 
117 /* Opcode 63: */
118 /* A-Form: */
119 #define FDIV		0x012		/*   18 */
120 #define FSUB		0x014		/*   20 */
121 #define FADD		0x015		/*   21 */
122 #define FSQRT		0x016		/*   22 */
123 #define FSEL		0x017		/*   23 */
124 #define FRE		0x018		/*   24 */
125 #define FMUL		0x019		/*   25 */
126 #define FRSQRTE		0x01a		/*   26 */
127 #define FMSUB		0x01c		/*   28 */
128 #define FMADD		0x01d		/*   29 */
129 #define FNMSUB		0x01e		/*   30 */
130 #define FNMADD		0x01f		/*   31 */
131 
132 /* X-Form: */
133 #define FCMPU		0x000		/*    0	*/
134 #define FRSP		0x00c		/*   12 */
135 #define FCTIW		0x00e		/*   14 */
136 #define FCTIWZ		0x00f		/*   15 */
137 #define FCMPO		0x020		/*   32 */
138 #define MTFSB1		0x026		/*   38 */
139 #define FNEG		0x028		/*   40 */
140 #define MCRFS		0x040		/*   64 */
141 #define MTFSB0		0x046		/*   70 */
142 #define FMR		0x048		/*   72 */
143 #define MTFSFI		0x086		/*  134 */
144 #define FNABS		0x088		/*  136 */
145 #define FABS		0x108		/*  264 */
146 #define MFFS		0x247		/*  583 */
147 #define MTFSF		0x2c7		/*  711 */
148 
149 
150 #define AB	2
151 #define AC	3
152 #define ABC	4
153 #define D	5
154 #define DU	6
155 #define X	7
156 #define XA	8
157 #define XB	9
158 #define XCR	11
159 #define XCRB	12
160 #define XCRI	13
161 #define XCRL	16
162 #define XE	14
163 #define XEU	15
164 #define XFLB	10
165 
166 static int
167 record_exception(struct pt_regs *regs, int eflag)
168 {
169 	u32 fpscr;
170 
171 	fpscr = __FPU_FPSCR;
172 
173 	if (eflag) {
174 		fpscr |= FPSCR_FX;
175 		if (eflag & EFLAG_OVERFLOW)
176 			fpscr |= FPSCR_OX;
177 		if (eflag & EFLAG_UNDERFLOW)
178 			fpscr |= FPSCR_UX;
179 		if (eflag & EFLAG_DIVZERO)
180 			fpscr |= FPSCR_ZX;
181 		if (eflag & EFLAG_INEXACT)
182 			fpscr |= FPSCR_XX;
183 		if (eflag & EFLAG_INVALID)
184 			fpscr |= FPSCR_VX;
185 		if (eflag & EFLAG_VXSNAN)
186 			fpscr |= FPSCR_VXSNAN;
187 		if (eflag & EFLAG_VXISI)
188 			fpscr |= FPSCR_VXISI;
189 		if (eflag & EFLAG_VXIDI)
190 			fpscr |= FPSCR_VXIDI;
191 		if (eflag & EFLAG_VXZDZ)
192 			fpscr |= FPSCR_VXZDZ;
193 		if (eflag & EFLAG_VXIMZ)
194 			fpscr |= FPSCR_VXIMZ;
195 		if (eflag & EFLAG_VXVC)
196 			fpscr |= FPSCR_VXVC;
197 		if (eflag & EFLAG_VXSOFT)
198 			fpscr |= FPSCR_VXSOFT;
199 		if (eflag & EFLAG_VXSQRT)
200 			fpscr |= FPSCR_VXSQRT;
201 		if (eflag & EFLAG_VXCVI)
202 			fpscr |= FPSCR_VXCVI;
203 	}
204 
205 //	fpscr &= ~(FPSCR_VX);
206 	if (fpscr & (FPSCR_VXSNAN | FPSCR_VXISI | FPSCR_VXIDI |
207 		     FPSCR_VXZDZ | FPSCR_VXIMZ | FPSCR_VXVC |
208 		     FPSCR_VXSOFT | FPSCR_VXSQRT | FPSCR_VXCVI))
209 		fpscr |= FPSCR_VX;
210 
211 	fpscr &= ~(FPSCR_FEX);
212 	if (((fpscr & FPSCR_VX) && (fpscr & FPSCR_VE)) ||
213 	    ((fpscr & FPSCR_OX) && (fpscr & FPSCR_OE)) ||
214 	    ((fpscr & FPSCR_UX) && (fpscr & FPSCR_UE)) ||
215 	    ((fpscr & FPSCR_ZX) && (fpscr & FPSCR_ZE)) ||
216 	    ((fpscr & FPSCR_XX) && (fpscr & FPSCR_XE)))
217 		fpscr |= FPSCR_FEX;
218 
219 	__FPU_FPSCR = fpscr;
220 
221 	return (fpscr & FPSCR_FEX) ? 1 : 0;
222 }
223 
224 int
225 do_mathemu(struct pt_regs *regs)
226 {
227 	void *op0 = 0, *op1 = 0, *op2 = 0, *op3 = 0;
228 	unsigned long pc = regs->nip;
229 	signed short sdisp;
230 	u32 insn = 0;
231 	int idx = 0;
232 	int (*func)(void *, void *, void *, void *);
233 	int type = 0;
234 	int eflag, trap;
235 
236 	if (get_user(insn, (u32 *)pc))
237 		return -EFAULT;
238 
239 	switch (insn >> 26) {
240 	case LFS:	func = lfs;	type = D;	break;
241 	case LFSU:	func = lfs;	type = DU;	break;
242 	case LFD:	func = lfd;	type = D;	break;
243 	case LFDU:	func = lfd;	type = DU;	break;
244 	case STFS:	func = stfs;	type = D;	break;
245 	case STFSU:	func = stfs;	type = DU;	break;
246 	case STFD:	func = stfd;	type = D;	break;
247 	case STFDU:	func = stfd;	type = DU;	break;
248 
249 	case OP31:
250 		switch ((insn >> 1) & 0x3ff) {
251 		case LFSX:	func = lfs;	type = XE;	break;
252 		case LFSUX:	func = lfs;	type = XEU;	break;
253 		case LFDX:	func = lfd;	type = XE;	break;
254 		case LFDUX:	func = lfd;	type = XEU;	break;
255 		case STFSX:	func = stfs;	type = XE;	break;
256 		case STFSUX:	func = stfs;	type = XEU;	break;
257 		case STFDX:	func = stfd;	type = XE;	break;
258 		case STFDUX:	func = stfd;	type = XEU;	break;
259 		case STFIWX:	func = stfiwx;	type = XE;	break;
260 		default:
261 			goto illegal;
262 		}
263 		break;
264 
265 	case OP59:
266 		switch ((insn >> 1) & 0x1f) {
267 		case FDIVS:	func = fdivs;	type = AB;	break;
268 		case FSUBS:	func = fsubs;	type = AB;	break;
269 		case FADDS:	func = fadds;	type = AB;	break;
270 		case FSQRTS:	func = fsqrts;	type = XB;	break;
271 		case FRES:	func = fres;	type = XB;	break;
272 		case FMULS:	func = fmuls;	type = AC;	break;
273 		case FRSQRTES:	func = frsqrtes;type = XB;	break;
274 		case FMSUBS:	func = fmsubs;	type = ABC;	break;
275 		case FMADDS:	func = fmadds;	type = ABC;	break;
276 		case FNMSUBS:	func = fnmsubs;	type = ABC;	break;
277 		case FNMADDS:	func = fnmadds;	type = ABC;	break;
278 		default:
279 			goto illegal;
280 		}
281 		break;
282 
283 	case OP63:
284 		if (insn & 0x20) {
285 			switch ((insn >> 1) & 0x1f) {
286 			case FDIV:	func = fdiv;	type = AB;	break;
287 			case FSUB:	func = fsub;	type = AB;	break;
288 			case FADD:	func = fadd;	type = AB;	break;
289 			case FSQRT:	func = fsqrt;	type = XB;	break;
290 			case FRE:	func = fre;	type = XB;	break;
291 			case FSEL:	func = fsel;	type = ABC;	break;
292 			case FMUL:	func = fmul;	type = AC;	break;
293 			case FRSQRTE:	func = frsqrte;	type = XB;	break;
294 			case FMSUB:	func = fmsub;	type = ABC;	break;
295 			case FMADD:	func = fmadd;	type = ABC;	break;
296 			case FNMSUB:	func = fnmsub;	type = ABC;	break;
297 			case FNMADD:	func = fnmadd;	type = ABC;	break;
298 			default:
299 				goto illegal;
300 			}
301 			break;
302 		}
303 
304 		switch ((insn >> 1) & 0x3ff) {
305 		case FCMPU:	func = fcmpu;	type = XCR;	break;
306 		case FRSP:	func = frsp;	type = XB;	break;
307 		case FCTIW:	func = fctiw;	type = XB;	break;
308 		case FCTIWZ:	func = fctiwz;	type = XB;	break;
309 		case FCMPO:	func = fcmpo;	type = XCR;	break;
310 		case MTFSB1:	func = mtfsb1;	type = XCRB;	break;
311 		case FNEG:	func = fneg;	type = XB;	break;
312 		case MCRFS:	func = mcrfs;	type = XCRL;	break;
313 		case MTFSB0:	func = mtfsb0;	type = XCRB;	break;
314 		case FMR:	func = fmr;	type = XB;	break;
315 		case MTFSFI:	func = mtfsfi;	type = XCRI;	break;
316 		case FNABS:	func = fnabs;	type = XB;	break;
317 		case FABS:	func = fabs;	type = XB;	break;
318 		case MFFS:	func = mffs;	type = X;	break;
319 		case MTFSF:	func = mtfsf;	type = XFLB;	break;
320 		default:
321 			goto illegal;
322 		}
323 		break;
324 
325 	default:
326 		goto illegal;
327 	}
328 
329 	switch (type) {
330 	case AB:
331 		op0 = (void *)&current->thread.TS_FPR((insn >> 21) & 0x1f);
332 		op1 = (void *)&current->thread.TS_FPR((insn >> 16) & 0x1f);
333 		op2 = (void *)&current->thread.TS_FPR((insn >> 11) & 0x1f);
334 		break;
335 
336 	case AC:
337 		op0 = (void *)&current->thread.TS_FPR((insn >> 21) & 0x1f);
338 		op1 = (void *)&current->thread.TS_FPR((insn >> 16) & 0x1f);
339 		op2 = (void *)&current->thread.TS_FPR((insn >>  6) & 0x1f);
340 		break;
341 
342 	case ABC:
343 		op0 = (void *)&current->thread.TS_FPR((insn >> 21) & 0x1f);
344 		op1 = (void *)&current->thread.TS_FPR((insn >> 16) & 0x1f);
345 		op2 = (void *)&current->thread.TS_FPR((insn >> 11) & 0x1f);
346 		op3 = (void *)&current->thread.TS_FPR((insn >>  6) & 0x1f);
347 		break;
348 
349 	case D:
350 		idx = (insn >> 16) & 0x1f;
351 		sdisp = (insn & 0xffff);
352 		op0 = (void *)&current->thread.TS_FPR((insn >> 21) & 0x1f);
353 		op1 = (void *)((idx ? regs->gpr[idx] : 0) + sdisp);
354 		break;
355 
356 	case DU:
357 		idx = (insn >> 16) & 0x1f;
358 		if (!idx)
359 			goto illegal;
360 
361 		sdisp = (insn & 0xffff);
362 		op0 = (void *)&current->thread.TS_FPR((insn >> 21) & 0x1f);
363 		op1 = (void *)(regs->gpr[idx] + sdisp);
364 		break;
365 
366 	case X:
367 		op0 = (void *)&current->thread.TS_FPR((insn >> 21) & 0x1f);
368 		break;
369 
370 	case XA:
371 		op0 = (void *)&current->thread.TS_FPR((insn >> 21) & 0x1f);
372 		op1 = (void *)&current->thread.TS_FPR((insn >> 16) & 0x1f);
373 		break;
374 
375 	case XB:
376 		op0 = (void *)&current->thread.TS_FPR((insn >> 21) & 0x1f);
377 		op1 = (void *)&current->thread.TS_FPR((insn >> 11) & 0x1f);
378 		break;
379 
380 	case XE:
381 		idx = (insn >> 16) & 0x1f;
382 		op0 = (void *)&current->thread.TS_FPR((insn >> 21) & 0x1f);
383 		op1 = (void *)((idx ? regs->gpr[idx] : 0)
384 				+ regs->gpr[(insn >> 11) & 0x1f]);
385 		break;
386 
387 	case XEU:
388 		idx = (insn >> 16) & 0x1f;
389 		if (!idx)
390 			goto illegal;
391 		op0 = (void *)&current->thread.TS_FPR((insn >> 21) & 0x1f);
392 		op1 = (void *)(regs->gpr[idx]
393 				+ regs->gpr[(insn >> 11) & 0x1f]);
394 		break;
395 
396 	case XCR:
397 		op0 = (void *)&regs->ccr;
398 		op1 = (void *)((insn >> 23) & 0x7);
399 		op2 = (void *)&current->thread.TS_FPR((insn >> 16) & 0x1f);
400 		op3 = (void *)&current->thread.TS_FPR((insn >> 11) & 0x1f);
401 		break;
402 
403 	case XCRL:
404 		op0 = (void *)&regs->ccr;
405 		op1 = (void *)((insn >> 23) & 0x7);
406 		op2 = (void *)((insn >> 18) & 0x7);
407 		break;
408 
409 	case XCRB:
410 		op0 = (void *)((insn >> 21) & 0x1f);
411 		break;
412 
413 	case XCRI:
414 		op0 = (void *)((insn >> 23) & 0x7);
415 		op1 = (void *)((insn >> 12) & 0xf);
416 		break;
417 
418 	case XFLB:
419 		op0 = (void *)((insn >> 17) & 0xff);
420 		op1 = (void *)&current->thread.TS_FPR((insn >> 11) & 0x1f);
421 		break;
422 
423 	default:
424 		goto illegal;
425 	}
426 
427 	/*
428 	 * If we support a HW FPU, we need to ensure the FP state
429 	 * is flushed into the thread_struct before attempting
430 	 * emulation
431 	 */
432 	flush_fp_to_thread(current);
433 
434 	eflag = func(op0, op1, op2, op3);
435 
436 	if (insn & 1) {
437 		regs->ccr &= ~(0x0f000000);
438 		regs->ccr |= (__FPU_FPSCR >> 4) & 0x0f000000;
439 	}
440 
441 	trap = record_exception(regs, eflag);
442 	if (trap)
443 		return 1;
444 
445 	switch (type) {
446 	case DU:
447 	case XEU:
448 		regs->gpr[idx] = (unsigned long)op1;
449 		break;
450 
451 	default:
452 		break;
453 	}
454 
455 	regs->nip += 4;
456 	return 0;
457 
458 illegal:
459 	return -ENOSYS;
460 }
461