xref: /openbmc/linux/arch/sparc/kernel/visemul.c (revision fb574682)
1 // SPDX-License-Identifier: GPL-2.0
2 /* visemul.c: Emulation of VIS instructions.
3  *
4  * Copyright (C) 2006 David S. Miller (davem@davemloft.net)
5  */
6 #include <linux/kernel.h>
7 #include <linux/errno.h>
8 #include <linux/thread_info.h>
9 #include <linux/perf_event.h>
10 
11 #include <asm/ptrace.h>
12 #include <asm/pstate.h>
13 #include <asm/fpumacro.h>
14 #include <linux/uaccess.h>
15 #include <asm/cacheflush.h>
16 
17 /* OPF field of various VIS instructions.  */
18 
19 /* 000111011 - four 16-bit packs  */
20 #define FPACK16_OPF	0x03b
21 
22 /* 000111010 - two 32-bit packs  */
23 #define FPACK32_OPF	0x03a
24 
25 /* 000111101 - four 16-bit packs  */
26 #define FPACKFIX_OPF	0x03d
27 
28 /* 001001101 - four 16-bit expands  */
29 #define FEXPAND_OPF	0x04d
30 
31 /* 001001011 - two 32-bit merges */
32 #define FPMERGE_OPF	0x04b
33 
34 /* 000110001 - 8-by-16-bit partitioned product  */
35 #define FMUL8x16_OPF	0x031
36 
37 /* 000110011 - 8-by-16-bit upper alpha partitioned product  */
38 #define FMUL8x16AU_OPF	0x033
39 
40 /* 000110101 - 8-by-16-bit lower alpha partitioned product  */
41 #define FMUL8x16AL_OPF	0x035
42 
43 /* 000110110 - upper 8-by-16-bit partitioned product  */
44 #define FMUL8SUx16_OPF	0x036
45 
46 /* 000110111 - lower 8-by-16-bit partitioned product  */
47 #define FMUL8ULx16_OPF	0x037
48 
49 /* 000111000 - upper 8-by-16-bit partitioned product  */
50 #define FMULD8SUx16_OPF	0x038
51 
52 /* 000111001 - lower unsigned 8-by-16-bit partitioned product  */
53 #define FMULD8ULx16_OPF	0x039
54 
55 /* 000101000 - four 16-bit compare; set rd if src1 > src2  */
56 #define FCMPGT16_OPF	0x028
57 
58 /* 000101100 - two 32-bit compare; set rd if src1 > src2  */
59 #define FCMPGT32_OPF	0x02c
60 
61 /* 000100000 - four 16-bit compare; set rd if src1 <= src2  */
62 #define FCMPLE16_OPF	0x020
63 
64 /* 000100100 - two 32-bit compare; set rd if src1 <= src2  */
65 #define FCMPLE32_OPF	0x024
66 
67 /* 000100010 - four 16-bit compare; set rd if src1 != src2  */
68 #define FCMPNE16_OPF	0x022
69 
70 /* 000100110 - two 32-bit compare; set rd if src1 != src2  */
71 #define FCMPNE32_OPF	0x026
72 
73 /* 000101010 - four 16-bit compare; set rd if src1 == src2  */
74 #define FCMPEQ16_OPF	0x02a
75 
76 /* 000101110 - two 32-bit compare; set rd if src1 == src2  */
77 #define FCMPEQ32_OPF	0x02e
78 
79 /* 000000000 - Eight 8-bit edge boundary processing  */
80 #define EDGE8_OPF	0x000
81 
82 /* 000000001 - Eight 8-bit edge boundary processing, no CC */
83 #define EDGE8N_OPF	0x001
84 
85 /* 000000010 - Eight 8-bit edge boundary processing, little-endian  */
86 #define EDGE8L_OPF	0x002
87 
88 /* 000000011 - Eight 8-bit edge boundary processing, little-endian, no CC  */
89 #define EDGE8LN_OPF	0x003
90 
91 /* 000000100 - Four 16-bit edge boundary processing  */
92 #define EDGE16_OPF	0x004
93 
94 /* 000000101 - Four 16-bit edge boundary processing, no CC  */
95 #define EDGE16N_OPF	0x005
96 
97 /* 000000110 - Four 16-bit edge boundary processing, little-endian  */
98 #define EDGE16L_OPF	0x006
99 
100 /* 000000111 - Four 16-bit edge boundary processing, little-endian, no CC  */
101 #define EDGE16LN_OPF	0x007
102 
103 /* 000001000 - Two 32-bit edge boundary processing  */
104 #define EDGE32_OPF	0x008
105 
106 /* 000001001 - Two 32-bit edge boundary processing, no CC  */
107 #define EDGE32N_OPF	0x009
108 
109 /* 000001010 - Two 32-bit edge boundary processing, little-endian  */
110 #define EDGE32L_OPF	0x00a
111 
112 /* 000001011 - Two 32-bit edge boundary processing, little-endian, no CC  */
113 #define EDGE32LN_OPF	0x00b
114 
115 /* 000111110 - distance between 8 8-bit components  */
116 #define PDIST_OPF	0x03e
117 
118 /* 000010000 - convert 8-bit 3-D address to blocked byte address  */
119 #define ARRAY8_OPF	0x010
120 
121 /* 000010010 - convert 16-bit 3-D address to blocked byte address  */
122 #define ARRAY16_OPF	0x012
123 
124 /* 000010100 - convert 32-bit 3-D address to blocked byte address  */
125 #define ARRAY32_OPF	0x014
126 
127 /* 000011001 - Set the GSR.MASK field in preparation for a BSHUFFLE  */
128 #define BMASK_OPF	0x019
129 
130 /* 001001100 - Permute bytes as specified by GSR.MASK  */
131 #define BSHUFFLE_OPF	0x04c
132 
133 #define VIS_OPF_SHIFT	5
134 #define VIS_OPF_MASK	(0x1ff << VIS_OPF_SHIFT)
135 
136 #define RS1(INSN)	(((INSN) >> 14) & 0x1f)
137 #define RS2(INSN)	(((INSN) >>  0) & 0x1f)
138 #define RD(INSN)	(((INSN) >> 25) & 0x1f)
139 
140 static inline void maybe_flush_windows(unsigned int rs1, unsigned int rs2,
141 				       unsigned int rd, int from_kernel)
142 {
143 	if (rs2 >= 16 || rs1 >= 16 || rd >= 16) {
144 		if (from_kernel != 0)
145 			__asm__ __volatile__("flushw");
146 		else
147 			flushw_user();
148 	}
149 }
150 
151 static unsigned long fetch_reg(unsigned int reg, struct pt_regs *regs)
152 {
153 	unsigned long value, fp;
154 
155 	if (reg < 16)
156 		return (!reg ? 0 : regs->u_regs[reg]);
157 
158 	fp = regs->u_regs[UREG_FP];
159 
160 	if (regs->tstate & TSTATE_PRIV) {
161 		struct reg_window *win;
162 		win = (struct reg_window *)(fp + STACK_BIAS);
163 		value = win->locals[reg - 16];
164 	} else if (!test_thread_64bit_stack(fp)) {
165 		struct reg_window32 __user *win32;
166 		win32 = (struct reg_window32 __user *)((unsigned long)((u32)fp));
167 		get_user(value, &win32->locals[reg - 16]);
168 	} else {
169 		struct reg_window __user *win;
170 		win = (struct reg_window __user *)(fp + STACK_BIAS);
171 		get_user(value, &win->locals[reg - 16]);
172 	}
173 	return value;
174 }
175 
176 static inline unsigned long __user *__fetch_reg_addr_user(unsigned int reg,
177 							  struct pt_regs *regs)
178 {
179 	unsigned long fp = regs->u_regs[UREG_FP];
180 
181 	BUG_ON(reg < 16);
182 	BUG_ON(regs->tstate & TSTATE_PRIV);
183 
184 	if (!test_thread_64bit_stack(fp)) {
185 		struct reg_window32 __user *win32;
186 		win32 = (struct reg_window32 __user *)((unsigned long)((u32)fp));
187 		return (unsigned long __user *)&win32->locals[reg - 16];
188 	} else {
189 		struct reg_window __user *win;
190 		win = (struct reg_window __user *)(fp + STACK_BIAS);
191 		return &win->locals[reg - 16];
192 	}
193 }
194 
195 static inline unsigned long *__fetch_reg_addr_kern(unsigned int reg,
196 						   struct pt_regs *regs)
197 {
198 	BUG_ON(reg >= 16);
199 	BUG_ON(regs->tstate & TSTATE_PRIV);
200 
201 	return &regs->u_regs[reg];
202 }
203 
204 static void store_reg(struct pt_regs *regs, unsigned long val, unsigned long rd)
205 {
206 	if (rd < 16) {
207 		unsigned long *rd_kern = __fetch_reg_addr_kern(rd, regs);
208 
209 		*rd_kern = val;
210 	} else {
211 		unsigned long __user *rd_user = __fetch_reg_addr_user(rd, regs);
212 
213 		if (!test_thread_64bit_stack(regs->u_regs[UREG_FP]))
214 			__put_user((u32)val, (u32 __user *)rd_user);
215 		else
216 			__put_user(val, rd_user);
217 	}
218 }
219 
220 static inline unsigned long fpd_regval(struct fpustate *f,
221 				       unsigned int insn_regnum)
222 {
223 	insn_regnum = (((insn_regnum & 1) << 5) |
224 		       (insn_regnum & 0x1e));
225 
226 	return *(unsigned long *) &f->regs[insn_regnum];
227 }
228 
229 static inline unsigned long *fpd_regaddr(struct fpustate *f,
230 					 unsigned int insn_regnum)
231 {
232 	insn_regnum = (((insn_regnum & 1) << 5) |
233 		       (insn_regnum & 0x1e));
234 
235 	return (unsigned long *) &f->regs[insn_regnum];
236 }
237 
238 static inline unsigned int fps_regval(struct fpustate *f,
239 				      unsigned int insn_regnum)
240 {
241 	return f->regs[insn_regnum];
242 }
243 
244 static inline unsigned int *fps_regaddr(struct fpustate *f,
245 					unsigned int insn_regnum)
246 {
247 	return &f->regs[insn_regnum];
248 }
249 
250 struct edge_tab {
251 	u16 left, right;
252 };
253 static struct edge_tab edge8_tab[8] = {
254 	{ 0xff, 0x80 },
255 	{ 0x7f, 0xc0 },
256 	{ 0x3f, 0xe0 },
257 	{ 0x1f, 0xf0 },
258 	{ 0x0f, 0xf8 },
259 	{ 0x07, 0xfc },
260 	{ 0x03, 0xfe },
261 	{ 0x01, 0xff },
262 };
263 static struct edge_tab edge8_tab_l[8] = {
264 	{ 0xff, 0x01 },
265 	{ 0xfe, 0x03 },
266 	{ 0xfc, 0x07 },
267 	{ 0xf8, 0x0f },
268 	{ 0xf0, 0x1f },
269 	{ 0xe0, 0x3f },
270 	{ 0xc0, 0x7f },
271 	{ 0x80, 0xff },
272 };
273 static struct edge_tab edge16_tab[4] = {
274 	{ 0xf, 0x8 },
275 	{ 0x7, 0xc },
276 	{ 0x3, 0xe },
277 	{ 0x1, 0xf },
278 };
279 static struct edge_tab edge16_tab_l[4] = {
280 	{ 0xf, 0x1 },
281 	{ 0xe, 0x3 },
282 	{ 0xc, 0x7 },
283 	{ 0x8, 0xf },
284 };
285 static struct edge_tab edge32_tab[2] = {
286 	{ 0x3, 0x2 },
287 	{ 0x1, 0x3 },
288 };
289 static struct edge_tab edge32_tab_l[2] = {
290 	{ 0x3, 0x1 },
291 	{ 0x2, 0x3 },
292 };
293 
294 static void edge(struct pt_regs *regs, unsigned int insn, unsigned int opf)
295 {
296 	unsigned long orig_rs1, rs1, orig_rs2, rs2, rd_val;
297 	u16 left, right;
298 
299 	maybe_flush_windows(RS1(insn), RS2(insn), RD(insn), 0);
300 	orig_rs1 = rs1 = fetch_reg(RS1(insn), regs);
301 	orig_rs2 = rs2 = fetch_reg(RS2(insn), regs);
302 
303 	if (test_thread_flag(TIF_32BIT)) {
304 		rs1 = rs1 & 0xffffffff;
305 		rs2 = rs2 & 0xffffffff;
306 	}
307 	switch (opf) {
308 	default:
309 	case EDGE8_OPF:
310 	case EDGE8N_OPF:
311 		left = edge8_tab[rs1 & 0x7].left;
312 		right = edge8_tab[rs2 & 0x7].right;
313 		break;
314 	case EDGE8L_OPF:
315 	case EDGE8LN_OPF:
316 		left = edge8_tab_l[rs1 & 0x7].left;
317 		right = edge8_tab_l[rs2 & 0x7].right;
318 		break;
319 
320 	case EDGE16_OPF:
321 	case EDGE16N_OPF:
322 		left = edge16_tab[(rs1 >> 1) & 0x3].left;
323 		right = edge16_tab[(rs2 >> 1) & 0x3].right;
324 		break;
325 
326 	case EDGE16L_OPF:
327 	case EDGE16LN_OPF:
328 		left = edge16_tab_l[(rs1 >> 1) & 0x3].left;
329 		right = edge16_tab_l[(rs2 >> 1) & 0x3].right;
330 		break;
331 
332 	case EDGE32_OPF:
333 	case EDGE32N_OPF:
334 		left = edge32_tab[(rs1 >> 2) & 0x1].left;
335 		right = edge32_tab[(rs2 >> 2) & 0x1].right;
336 		break;
337 
338 	case EDGE32L_OPF:
339 	case EDGE32LN_OPF:
340 		left = edge32_tab_l[(rs1 >> 2) & 0x1].left;
341 		right = edge32_tab_l[(rs2 >> 2) & 0x1].right;
342 		break;
343 	}
344 
345 	if ((rs1 & ~0x7UL) == (rs2 & ~0x7UL))
346 		rd_val = right & left;
347 	else
348 		rd_val = left;
349 
350 	store_reg(regs, rd_val, RD(insn));
351 
352 	switch (opf) {
353 	case EDGE8_OPF:
354 	case EDGE8L_OPF:
355 	case EDGE16_OPF:
356 	case EDGE16L_OPF:
357 	case EDGE32_OPF:
358 	case EDGE32L_OPF: {
359 		unsigned long ccr, tstate;
360 
361 		__asm__ __volatile__("subcc	%1, %2, %%g0\n\t"
362 				     "rd	%%ccr, %0"
363 				     : "=r" (ccr)
364 				     : "r" (orig_rs1), "r" (orig_rs2)
365 				     : "cc");
366 		tstate = regs->tstate & ~(TSTATE_XCC | TSTATE_ICC);
367 		regs->tstate = tstate | (ccr << 32UL);
368 	}
369 	}
370 }
371 
372 static void array(struct pt_regs *regs, unsigned int insn, unsigned int opf)
373 {
374 	unsigned long rs1, rs2, rd_val;
375 	unsigned int bits, bits_mask;
376 
377 	maybe_flush_windows(RS1(insn), RS2(insn), RD(insn), 0);
378 	rs1 = fetch_reg(RS1(insn), regs);
379 	rs2 = fetch_reg(RS2(insn), regs);
380 
381 	bits = (rs2 > 5 ? 5 : rs2);
382 	bits_mask = (1UL << bits) - 1UL;
383 
384 	rd_val = ((((rs1 >> 11) & 0x3) <<  0) |
385 		  (((rs1 >> 33) & 0x3) <<  2) |
386 		  (((rs1 >> 55) & 0x1) <<  4) |
387 		  (((rs1 >> 13) & 0xf) <<  5) |
388 		  (((rs1 >> 35) & 0xf) <<  9) |
389 		  (((rs1 >> 56) & 0xf) << 13) |
390 		  (((rs1 >> 17) & bits_mask) << 17) |
391 		  (((rs1 >> 39) & bits_mask) << (17 + bits)) |
392 		  (((rs1 >> 60) & 0xf)       << (17 + (2*bits))));
393 
394 	switch (opf) {
395 	case ARRAY16_OPF:
396 		rd_val <<= 1;
397 		break;
398 
399 	case ARRAY32_OPF:
400 		rd_val <<= 2;
401 	}
402 
403 	store_reg(regs, rd_val, RD(insn));
404 }
405 
406 static void bmask(struct pt_regs *regs, unsigned int insn)
407 {
408 	unsigned long rs1, rs2, rd_val, gsr;
409 
410 	maybe_flush_windows(RS1(insn), RS2(insn), RD(insn), 0);
411 	rs1 = fetch_reg(RS1(insn), regs);
412 	rs2 = fetch_reg(RS2(insn), regs);
413 	rd_val = rs1 + rs2;
414 
415 	store_reg(regs, rd_val, RD(insn));
416 
417 	gsr = current_thread_info()->gsr[0] & 0xffffffff;
418 	gsr |= rd_val << 32UL;
419 	current_thread_info()->gsr[0] = gsr;
420 }
421 
422 static void bshuffle(struct pt_regs *regs, unsigned int insn)
423 {
424 	struct fpustate *f = FPUSTATE;
425 	unsigned long rs1, rs2, rd_val;
426 	unsigned long bmask, i;
427 
428 	bmask = current_thread_info()->gsr[0] >> 32UL;
429 
430 	rs1 = fpd_regval(f, RS1(insn));
431 	rs2 = fpd_regval(f, RS2(insn));
432 
433 	rd_val = 0UL;
434 	for (i = 0; i < 8; i++) {
435 		unsigned long which = (bmask >> (i * 4)) & 0xf;
436 		unsigned long byte;
437 
438 		if (which < 8)
439 			byte = (rs1 >> (which * 8)) & 0xff;
440 		else
441 			byte = (rs2 >> ((which-8)*8)) & 0xff;
442 		rd_val |= (byte << (i * 8));
443 	}
444 
445 	*fpd_regaddr(f, RD(insn)) = rd_val;
446 }
447 
448 static void pdist(struct pt_regs *regs, unsigned int insn)
449 {
450 	struct fpustate *f = FPUSTATE;
451 	unsigned long rs1, rs2, *rd, rd_val;
452 	unsigned long i;
453 
454 	rs1 = fpd_regval(f, RS1(insn));
455 	rs2 = fpd_regval(f, RS2(insn));
456 	rd = fpd_regaddr(f, RD(insn));
457 
458 	rd_val = *rd;
459 
460 	for (i = 0; i < 8; i++) {
461 		s16 s1, s2;
462 
463 		s1 = (rs1 >> (56 - (i * 8))) & 0xff;
464 		s2 = (rs2 >> (56 - (i * 8))) & 0xff;
465 
466 		/* Absolute value of difference. */
467 		s1 -= s2;
468 		if (s1 < 0)
469 			s1 = ~s1 + 1;
470 
471 		rd_val += s1;
472 	}
473 
474 	*rd = rd_val;
475 }
476 
477 static void pformat(struct pt_regs *regs, unsigned int insn, unsigned int opf)
478 {
479 	struct fpustate *f = FPUSTATE;
480 	unsigned long rs1, rs2, gsr, scale, rd_val;
481 
482 	gsr = current_thread_info()->gsr[0];
483 	scale = (gsr >> 3) & (opf == FPACK16_OPF ? 0xf : 0x1f);
484 	switch (opf) {
485 	case FPACK16_OPF: {
486 		unsigned long byte;
487 
488 		rs2 = fpd_regval(f, RS2(insn));
489 		rd_val = 0;
490 		for (byte = 0; byte < 4; byte++) {
491 			unsigned int val;
492 			s16 src = (rs2 >> (byte * 16UL)) & 0xffffUL;
493 			int scaled = src << scale;
494 			int from_fixed = scaled >> 7;
495 
496 			val = ((from_fixed < 0) ?
497 			       0 :
498 			       (from_fixed > 255) ?
499 			       255 : from_fixed);
500 
501 			rd_val |= (val << (8 * byte));
502 		}
503 		*fps_regaddr(f, RD(insn)) = rd_val;
504 		break;
505 	}
506 
507 	case FPACK32_OPF: {
508 		unsigned long word;
509 
510 		rs1 = fpd_regval(f, RS1(insn));
511 		rs2 = fpd_regval(f, RS2(insn));
512 		rd_val = (rs1 << 8) & ~(0x000000ff000000ffUL);
513 		for (word = 0; word < 2; word++) {
514 			unsigned long val;
515 			s32 src = (rs2 >> (word * 32UL));
516 			s64 scaled = src << scale;
517 			s64 from_fixed = scaled >> 23;
518 
519 			val = ((from_fixed < 0) ?
520 			       0 :
521 			       (from_fixed > 255) ?
522 			       255 : from_fixed);
523 
524 			rd_val |= (val << (32 * word));
525 		}
526 		*fpd_regaddr(f, RD(insn)) = rd_val;
527 		break;
528 	}
529 
530 	case FPACKFIX_OPF: {
531 		unsigned long word;
532 
533 		rs2 = fpd_regval(f, RS2(insn));
534 
535 		rd_val = 0;
536 		for (word = 0; word < 2; word++) {
537 			long val;
538 			s32 src = (rs2 >> (word * 32UL));
539 			s64 scaled = src << scale;
540 			s64 from_fixed = scaled >> 16;
541 
542 			val = ((from_fixed < -32768) ?
543 			       -32768 :
544 			       (from_fixed > 32767) ?
545 			       32767 : from_fixed);
546 
547 			rd_val |= ((val & 0xffff) << (word * 16));
548 		}
549 		*fps_regaddr(f, RD(insn)) = rd_val;
550 		break;
551 	}
552 
553 	case FEXPAND_OPF: {
554 		unsigned long byte;
555 
556 		rs2 = fps_regval(f, RS2(insn));
557 
558 		rd_val = 0;
559 		for (byte = 0; byte < 4; byte++) {
560 			unsigned long val;
561 			u8 src = (rs2 >> (byte * 8)) & 0xff;
562 
563 			val = src << 4;
564 
565 			rd_val |= (val << (byte * 16));
566 		}
567 		*fpd_regaddr(f, RD(insn)) = rd_val;
568 		break;
569 	}
570 
571 	case FPMERGE_OPF: {
572 		rs1 = fps_regval(f, RS1(insn));
573 		rs2 = fps_regval(f, RS2(insn));
574 
575 		rd_val = (((rs2 & 0x000000ff) <<  0) |
576 			  ((rs1 & 0x000000ff) <<  8) |
577 			  ((rs2 & 0x0000ff00) <<  8) |
578 			  ((rs1 & 0x0000ff00) << 16) |
579 			  ((rs2 & 0x00ff0000) << 16) |
580 			  ((rs1 & 0x00ff0000) << 24) |
581 			  ((rs2 & 0xff000000) << 24) |
582 			  ((rs1 & 0xff000000) << 32));
583 		*fpd_regaddr(f, RD(insn)) = rd_val;
584 		break;
585 	}
586 	}
587 }
588 
589 static void pmul(struct pt_regs *regs, unsigned int insn, unsigned int opf)
590 {
591 	struct fpustate *f = FPUSTATE;
592 	unsigned long rs1, rs2, rd_val;
593 
594 	switch (opf) {
595 	case FMUL8x16_OPF: {
596 		unsigned long byte;
597 
598 		rs1 = fps_regval(f, RS1(insn));
599 		rs2 = fpd_regval(f, RS2(insn));
600 
601 		rd_val = 0;
602 		for (byte = 0; byte < 4; byte++) {
603 			u16 src1 = (rs1 >> (byte *  8)) & 0x00ff;
604 			s16 src2 = (rs2 >> (byte * 16)) & 0xffff;
605 			u32 prod = src1 * src2;
606 			u16 scaled = ((prod & 0x00ffff00) >> 8);
607 
608 			/* Round up.  */
609 			if (prod & 0x80)
610 				scaled++;
611 			rd_val |= ((scaled & 0xffffUL) << (byte * 16UL));
612 		}
613 
614 		*fpd_regaddr(f, RD(insn)) = rd_val;
615 		break;
616 	}
617 
618 	case FMUL8x16AU_OPF:
619 	case FMUL8x16AL_OPF: {
620 		unsigned long byte;
621 		s16 src2;
622 
623 		rs1 = fps_regval(f, RS1(insn));
624 		rs2 = fps_regval(f, RS2(insn));
625 
626 		rd_val = 0;
627 		src2 = rs2 >> (opf == FMUL8x16AU_OPF ? 16 : 0);
628 		for (byte = 0; byte < 4; byte++) {
629 			u16 src1 = (rs1 >> (byte * 8)) & 0x00ff;
630 			u32 prod = src1 * src2;
631 			u16 scaled = ((prod & 0x00ffff00) >> 8);
632 
633 			/* Round up.  */
634 			if (prod & 0x80)
635 				scaled++;
636 			rd_val |= ((scaled & 0xffffUL) << (byte * 16UL));
637 		}
638 
639 		*fpd_regaddr(f, RD(insn)) = rd_val;
640 		break;
641 	}
642 
643 	case FMUL8SUx16_OPF:
644 	case FMUL8ULx16_OPF: {
645 		unsigned long byte, ushift;
646 
647 		rs1 = fpd_regval(f, RS1(insn));
648 		rs2 = fpd_regval(f, RS2(insn));
649 
650 		rd_val = 0;
651 		ushift = (opf == FMUL8SUx16_OPF) ? 8 : 0;
652 		for (byte = 0; byte < 4; byte++) {
653 			u16 src1;
654 			s16 src2;
655 			u32 prod;
656 			u16 scaled;
657 
658 			src1 = ((rs1 >> ((16 * byte) + ushift)) & 0x00ff);
659 			src2 = ((rs2 >> (16 * byte)) & 0xffff);
660 			prod = src1 * src2;
661 			scaled = ((prod & 0x00ffff00) >> 8);
662 
663 			/* Round up.  */
664 			if (prod & 0x80)
665 				scaled++;
666 			rd_val |= ((scaled & 0xffffUL) << (byte * 16UL));
667 		}
668 
669 		*fpd_regaddr(f, RD(insn)) = rd_val;
670 		break;
671 	}
672 
673 	case FMULD8SUx16_OPF:
674 	case FMULD8ULx16_OPF: {
675 		unsigned long byte, ushift;
676 
677 		rs1 = fps_regval(f, RS1(insn));
678 		rs2 = fps_regval(f, RS2(insn));
679 
680 		rd_val = 0;
681 		ushift = (opf == FMULD8SUx16_OPF) ? 8 : 0;
682 		for (byte = 0; byte < 2; byte++) {
683 			u16 src1;
684 			s16 src2;
685 			u32 prod;
686 			u16 scaled;
687 
688 			src1 = ((rs1 >> ((16 * byte) + ushift)) & 0x00ff);
689 			src2 = ((rs2 >> (16 * byte)) & 0xffff);
690 			prod = src1 * src2;
691 			scaled = ((prod & 0x00ffff00) >> 8);
692 
693 			/* Round up.  */
694 			if (prod & 0x80)
695 				scaled++;
696 			rd_val |= ((scaled & 0xffffUL) <<
697 				   ((byte * 32UL) + 7UL));
698 		}
699 		*fpd_regaddr(f, RD(insn)) = rd_val;
700 		break;
701 	}
702 	}
703 }
704 
705 static void pcmp(struct pt_regs *regs, unsigned int insn, unsigned int opf)
706 {
707 	struct fpustate *f = FPUSTATE;
708 	unsigned long rs1, rs2, rd_val, i;
709 
710 	rs1 = fpd_regval(f, RS1(insn));
711 	rs2 = fpd_regval(f, RS2(insn));
712 
713 	rd_val = 0;
714 
715 	switch (opf) {
716 	case FCMPGT16_OPF:
717 		for (i = 0; i < 4; i++) {
718 			s16 a = (rs1 >> (i * 16)) & 0xffff;
719 			s16 b = (rs2 >> (i * 16)) & 0xffff;
720 
721 			if (a > b)
722 				rd_val |= 8 >> i;
723 		}
724 		break;
725 
726 	case FCMPGT32_OPF:
727 		for (i = 0; i < 2; i++) {
728 			s32 a = (rs1 >> (i * 32)) & 0xffffffff;
729 			s32 b = (rs2 >> (i * 32)) & 0xffffffff;
730 
731 			if (a > b)
732 				rd_val |= 2 >> i;
733 		}
734 		break;
735 
736 	case FCMPLE16_OPF:
737 		for (i = 0; i < 4; i++) {
738 			s16 a = (rs1 >> (i * 16)) & 0xffff;
739 			s16 b = (rs2 >> (i * 16)) & 0xffff;
740 
741 			if (a <= b)
742 				rd_val |= 8 >> i;
743 		}
744 		break;
745 
746 	case FCMPLE32_OPF:
747 		for (i = 0; i < 2; i++) {
748 			s32 a = (rs1 >> (i * 32)) & 0xffffffff;
749 			s32 b = (rs2 >> (i * 32)) & 0xffffffff;
750 
751 			if (a <= b)
752 				rd_val |= 2 >> i;
753 		}
754 		break;
755 
756 	case FCMPNE16_OPF:
757 		for (i = 0; i < 4; i++) {
758 			s16 a = (rs1 >> (i * 16)) & 0xffff;
759 			s16 b = (rs2 >> (i * 16)) & 0xffff;
760 
761 			if (a != b)
762 				rd_val |= 8 >> i;
763 		}
764 		break;
765 
766 	case FCMPNE32_OPF:
767 		for (i = 0; i < 2; i++) {
768 			s32 a = (rs1 >> (i * 32)) & 0xffffffff;
769 			s32 b = (rs2 >> (i * 32)) & 0xffffffff;
770 
771 			if (a != b)
772 				rd_val |= 2 >> i;
773 		}
774 		break;
775 
776 	case FCMPEQ16_OPF:
777 		for (i = 0; i < 4; i++) {
778 			s16 a = (rs1 >> (i * 16)) & 0xffff;
779 			s16 b = (rs2 >> (i * 16)) & 0xffff;
780 
781 			if (a == b)
782 				rd_val |= 8 >> i;
783 		}
784 		break;
785 
786 	case FCMPEQ32_OPF:
787 		for (i = 0; i < 2; i++) {
788 			s32 a = (rs1 >> (i * 32)) & 0xffffffff;
789 			s32 b = (rs2 >> (i * 32)) & 0xffffffff;
790 
791 			if (a == b)
792 				rd_val |= 2 >> i;
793 		}
794 		break;
795 	}
796 
797 	maybe_flush_windows(0, 0, RD(insn), 0);
798 	store_reg(regs, rd_val, RD(insn));
799 }
800 
801 /* Emulate the VIS instructions which are not implemented in
802  * hardware on Niagara.
803  */
804 int vis_emul(struct pt_regs *regs, unsigned int insn)
805 {
806 	unsigned long pc = regs->tpc;
807 	unsigned int opf;
808 
809 	BUG_ON(regs->tstate & TSTATE_PRIV);
810 
811 	perf_sw_event(PERF_COUNT_SW_EMULATION_FAULTS, 1, regs, 0);
812 
813 	if (test_thread_flag(TIF_32BIT))
814 		pc = (u32)pc;
815 
816 	if (get_user(insn, (u32 __user *) pc))
817 		return -EFAULT;
818 
819 	save_and_clear_fpu();
820 
821 	opf = (insn & VIS_OPF_MASK) >> VIS_OPF_SHIFT;
822 	switch (opf) {
823 	default:
824 		return -EINVAL;
825 
826 	/* Pixel Formatting Instructions.  */
827 	case FPACK16_OPF:
828 	case FPACK32_OPF:
829 	case FPACKFIX_OPF:
830 	case FEXPAND_OPF:
831 	case FPMERGE_OPF:
832 		pformat(regs, insn, opf);
833 		break;
834 
835 	/* Partitioned Multiply Instructions  */
836 	case FMUL8x16_OPF:
837 	case FMUL8x16AU_OPF:
838 	case FMUL8x16AL_OPF:
839 	case FMUL8SUx16_OPF:
840 	case FMUL8ULx16_OPF:
841 	case FMULD8SUx16_OPF:
842 	case FMULD8ULx16_OPF:
843 		pmul(regs, insn, opf);
844 		break;
845 
846 	/* Pixel Compare Instructions  */
847 	case FCMPGT16_OPF:
848 	case FCMPGT32_OPF:
849 	case FCMPLE16_OPF:
850 	case FCMPLE32_OPF:
851 	case FCMPNE16_OPF:
852 	case FCMPNE32_OPF:
853 	case FCMPEQ16_OPF:
854 	case FCMPEQ32_OPF:
855 		pcmp(regs, insn, opf);
856 		break;
857 
858 	/* Edge Handling Instructions  */
859 	case EDGE8_OPF:
860 	case EDGE8N_OPF:
861 	case EDGE8L_OPF:
862 	case EDGE8LN_OPF:
863 	case EDGE16_OPF:
864 	case EDGE16N_OPF:
865 	case EDGE16L_OPF:
866 	case EDGE16LN_OPF:
867 	case EDGE32_OPF:
868 	case EDGE32N_OPF:
869 	case EDGE32L_OPF:
870 	case EDGE32LN_OPF:
871 		edge(regs, insn, opf);
872 		break;
873 
874 	/* Pixel Component Distance  */
875 	case PDIST_OPF:
876 		pdist(regs, insn);
877 		break;
878 
879 	/* Three-Dimensional Array Addressing Instructions  */
880 	case ARRAY8_OPF:
881 	case ARRAY16_OPF:
882 	case ARRAY32_OPF:
883 		array(regs, insn, opf);
884 		break;
885 
886 	/* Byte Mask and Shuffle Instructions  */
887 	case BMASK_OPF:
888 		bmask(regs, insn);
889 		break;
890 
891 	case BSHUFFLE_OPF:
892 		bshuffle(regs, insn);
893 		break;
894 	}
895 
896 	regs->tpc = regs->tnpc;
897 	regs->tnpc += 4;
898 	return 0;
899 }
900