xref: /openbmc/linux/arch/mips/math-emu/cp1emu.c (revision ee8a99bd)
1 /*
2  * cp1emu.c: a MIPS coprocessor 1 (fpu) instruction emulator
3  *
4  * MIPS floating point support
5  * Copyright (C) 1994-2000 Algorithmics Ltd.
6  *
7  * Kevin D. Kissell, kevink@mips.com and Carsten Langgaard, carstenl@mips.com
8  * Copyright (C) 2000  MIPS Technologies, Inc.
9  *
10  *  This program is free software; you can distribute it and/or modify it
11  *  under the terms of the GNU General Public License (Version 2) as
12  *  published by the Free Software Foundation.
13  *
14  *  This program is distributed in the hope it will be useful, but WITHOUT
15  *  ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
16  *  FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
17  *  for more details.
18  *
19  *  You should have received a copy of the GNU General Public License along
20  *  with this program; if not, write to the Free Software Foundation, Inc.,
21  *  59 Temple Place - Suite 330, Boston MA 02111-1307, USA.
22  *
23  * A complete emulator for MIPS coprocessor 1 instructions.  This is
24  * required for #float(switch) or #float(trap), where it catches all
25  * COP1 instructions via the "CoProcessor Unusable" exception.
26  *
27  * More surprisingly it is also required for #float(ieee), to help out
28  * the hardware fpu at the boundaries of the IEEE-754 representation
29  * (denormalised values, infinities, underflow, etc).  It is made
30  * quite nasty because emulation of some non-COP1 instructions is
31  * required, e.g. in branch delay slots.
32  *
33  * Note if you know that you won't have an fpu, then you'll get much
34  * better performance by compiling with -msoft-float!
35  */
36 #include <linux/sched.h>
37 #include <linux/module.h>
38 #include <linux/debugfs.h>
39 #include <linux/perf_event.h>
40 
41 #include <asm/inst.h>
42 #include <asm/bootinfo.h>
43 #include <asm/processor.h>
44 #include <asm/ptrace.h>
45 #include <asm/signal.h>
46 #include <asm/mipsregs.h>
47 #include <asm/fpu_emulator.h>
48 #include <asm/fpu.h>
49 #include <asm/uaccess.h>
50 #include <asm/branch.h>
51 
52 #include "ieee754.h"
53 
54 /* Strap kernel emulator for full MIPS IV emulation */
55 
56 #ifdef __mips
57 #undef __mips
58 #endif
59 #define __mips 4
60 
61 /* Function which emulates a floating point instruction. */
62 
63 static int fpu_emu(struct pt_regs *, struct mips_fpu_struct *,
64 	mips_instruction);
65 
66 #if __mips >= 4 && __mips != 32
67 static int fpux_emu(struct pt_regs *,
68 	struct mips_fpu_struct *, mips_instruction, void *__user *);
69 #endif
70 
71 /* Further private data for which no space exists in mips_fpu_struct */
72 
73 #ifdef CONFIG_DEBUG_FS
74 DEFINE_PER_CPU(struct mips_fpu_emulator_stats, fpuemustats);
75 #endif
76 
77 /* Control registers */
78 
79 #define FPCREG_RID	0	/* $0  = revision id */
80 #define FPCREG_CSR	31	/* $31 = csr */
81 
82 /* Determine rounding mode from the RM bits of the FCSR */
83 #define modeindex(v) ((v) & FPU_CSR_RM)
84 
85 /* microMIPS bitfields */
86 #define MM_POOL32A_MINOR_MASK	0x3f
87 #define MM_POOL32A_MINOR_SHIFT	0x6
88 #define MM_MIPS32_COND_FC	0x30
89 
90 /* Convert Mips rounding mode (0..3) to IEEE library modes. */
91 static const unsigned char ieee_rm[4] = {
92 	[FPU_CSR_RN] = IEEE754_RN,
93 	[FPU_CSR_RZ] = IEEE754_RZ,
94 	[FPU_CSR_RU] = IEEE754_RU,
95 	[FPU_CSR_RD] = IEEE754_RD,
96 };
97 /* Convert IEEE library modes to Mips rounding mode (0..3). */
98 static const unsigned char mips_rm[4] = {
99 	[IEEE754_RN] = FPU_CSR_RN,
100 	[IEEE754_RZ] = FPU_CSR_RZ,
101 	[IEEE754_RD] = FPU_CSR_RD,
102 	[IEEE754_RU] = FPU_CSR_RU,
103 };
104 
105 #if __mips >= 4
106 /* convert condition code register number to csr bit */
107 static const unsigned int fpucondbit[8] = {
108 	FPU_CSR_COND0,
109 	FPU_CSR_COND1,
110 	FPU_CSR_COND2,
111 	FPU_CSR_COND3,
112 	FPU_CSR_COND4,
113 	FPU_CSR_COND5,
114 	FPU_CSR_COND6,
115 	FPU_CSR_COND7
116 };
117 #endif
118 
119 /* (microMIPS) Convert 16-bit register encoding to 32-bit register encoding. */
120 static const unsigned int reg16to32map[8] = {16, 17, 2, 3, 4, 5, 6, 7};
121 
122 /* (microMIPS) Convert certain microMIPS instructions to MIPS32 format. */
123 static const int sd_format[] = {16, 17, 0, 0, 0, 0, 0, 0};
124 static const int sdps_format[] = {16, 17, 22, 0, 0, 0, 0, 0};
125 static const int dwl_format[] = {17, 20, 21, 0, 0, 0, 0, 0};
126 static const int swl_format[] = {16, 20, 21, 0, 0, 0, 0, 0};
127 
128 /*
129  * This functions translates a 32-bit microMIPS instruction
130  * into a 32-bit MIPS32 instruction. Returns 0 on success
131  * and SIGILL otherwise.
132  */
133 static int microMIPS32_to_MIPS32(union mips_instruction *insn_ptr)
134 {
135 	union mips_instruction insn = *insn_ptr;
136 	union mips_instruction mips32_insn = insn;
137 	int func, fmt, op;
138 
139 	switch (insn.mm_i_format.opcode) {
140 	case mm_ldc132_op:
141 		mips32_insn.mm_i_format.opcode = ldc1_op;
142 		mips32_insn.mm_i_format.rt = insn.mm_i_format.rs;
143 		mips32_insn.mm_i_format.rs = insn.mm_i_format.rt;
144 		break;
145 	case mm_lwc132_op:
146 		mips32_insn.mm_i_format.opcode = lwc1_op;
147 		mips32_insn.mm_i_format.rt = insn.mm_i_format.rs;
148 		mips32_insn.mm_i_format.rs = insn.mm_i_format.rt;
149 		break;
150 	case mm_sdc132_op:
151 		mips32_insn.mm_i_format.opcode = sdc1_op;
152 		mips32_insn.mm_i_format.rt = insn.mm_i_format.rs;
153 		mips32_insn.mm_i_format.rs = insn.mm_i_format.rt;
154 		break;
155 	case mm_swc132_op:
156 		mips32_insn.mm_i_format.opcode = swc1_op;
157 		mips32_insn.mm_i_format.rt = insn.mm_i_format.rs;
158 		mips32_insn.mm_i_format.rs = insn.mm_i_format.rt;
159 		break;
160 	case mm_pool32i_op:
161 		/* NOTE: offset is << by 1 if in microMIPS mode. */
162 		if ((insn.mm_i_format.rt == mm_bc1f_op) ||
163 		    (insn.mm_i_format.rt == mm_bc1t_op)) {
164 			mips32_insn.fb_format.opcode = cop1_op;
165 			mips32_insn.fb_format.bc = bc_op;
166 			mips32_insn.fb_format.flag =
167 				(insn.mm_i_format.rt == mm_bc1t_op) ? 1 : 0;
168 		} else
169 			return SIGILL;
170 		break;
171 	case mm_pool32f_op:
172 		switch (insn.mm_fp0_format.func) {
173 		case mm_32f_01_op:
174 		case mm_32f_11_op:
175 		case mm_32f_02_op:
176 		case mm_32f_12_op:
177 		case mm_32f_41_op:
178 		case mm_32f_51_op:
179 		case mm_32f_42_op:
180 		case mm_32f_52_op:
181 			op = insn.mm_fp0_format.func;
182 			if (op == mm_32f_01_op)
183 				func = madd_s_op;
184 			else if (op == mm_32f_11_op)
185 				func = madd_d_op;
186 			else if (op == mm_32f_02_op)
187 				func = nmadd_s_op;
188 			else if (op == mm_32f_12_op)
189 				func = nmadd_d_op;
190 			else if (op == mm_32f_41_op)
191 				func = msub_s_op;
192 			else if (op == mm_32f_51_op)
193 				func = msub_d_op;
194 			else if (op == mm_32f_42_op)
195 				func = nmsub_s_op;
196 			else
197 				func = nmsub_d_op;
198 			mips32_insn.fp6_format.opcode = cop1x_op;
199 			mips32_insn.fp6_format.fr = insn.mm_fp6_format.fr;
200 			mips32_insn.fp6_format.ft = insn.mm_fp6_format.ft;
201 			mips32_insn.fp6_format.fs = insn.mm_fp6_format.fs;
202 			mips32_insn.fp6_format.fd = insn.mm_fp6_format.fd;
203 			mips32_insn.fp6_format.func = func;
204 			break;
205 		case mm_32f_10_op:
206 			func = -1;	/* Invalid */
207 			op = insn.mm_fp5_format.op & 0x7;
208 			if (op == mm_ldxc1_op)
209 				func = ldxc1_op;
210 			else if (op == mm_sdxc1_op)
211 				func = sdxc1_op;
212 			else if (op == mm_lwxc1_op)
213 				func = lwxc1_op;
214 			else if (op == mm_swxc1_op)
215 				func = swxc1_op;
216 
217 			if (func != -1) {
218 				mips32_insn.r_format.opcode = cop1x_op;
219 				mips32_insn.r_format.rs =
220 					insn.mm_fp5_format.base;
221 				mips32_insn.r_format.rt =
222 					insn.mm_fp5_format.index;
223 				mips32_insn.r_format.rd = 0;
224 				mips32_insn.r_format.re = insn.mm_fp5_format.fd;
225 				mips32_insn.r_format.func = func;
226 			} else
227 				return SIGILL;
228 			break;
229 		case mm_32f_40_op:
230 			op = -1;	/* Invalid */
231 			if (insn.mm_fp2_format.op == mm_fmovt_op)
232 				op = 1;
233 			else if (insn.mm_fp2_format.op == mm_fmovf_op)
234 				op = 0;
235 			if (op != -1) {
236 				mips32_insn.fp0_format.opcode = cop1_op;
237 				mips32_insn.fp0_format.fmt =
238 					sdps_format[insn.mm_fp2_format.fmt];
239 				mips32_insn.fp0_format.ft =
240 					(insn.mm_fp2_format.cc<<2) + op;
241 				mips32_insn.fp0_format.fs =
242 					insn.mm_fp2_format.fs;
243 				mips32_insn.fp0_format.fd =
244 					insn.mm_fp2_format.fd;
245 				mips32_insn.fp0_format.func = fmovc_op;
246 			} else
247 				return SIGILL;
248 			break;
249 		case mm_32f_60_op:
250 			func = -1;	/* Invalid */
251 			if (insn.mm_fp0_format.op == mm_fadd_op)
252 				func = fadd_op;
253 			else if (insn.mm_fp0_format.op == mm_fsub_op)
254 				func = fsub_op;
255 			else if (insn.mm_fp0_format.op == mm_fmul_op)
256 				func = fmul_op;
257 			else if (insn.mm_fp0_format.op == mm_fdiv_op)
258 				func = fdiv_op;
259 			if (func != -1) {
260 				mips32_insn.fp0_format.opcode = cop1_op;
261 				mips32_insn.fp0_format.fmt =
262 					sdps_format[insn.mm_fp0_format.fmt];
263 				mips32_insn.fp0_format.ft =
264 					insn.mm_fp0_format.ft;
265 				mips32_insn.fp0_format.fs =
266 					insn.mm_fp0_format.fs;
267 				mips32_insn.fp0_format.fd =
268 					insn.mm_fp0_format.fd;
269 				mips32_insn.fp0_format.func = func;
270 			} else
271 				return SIGILL;
272 			break;
273 		case mm_32f_70_op:
274 			func = -1;	/* Invalid */
275 			if (insn.mm_fp0_format.op == mm_fmovn_op)
276 				func = fmovn_op;
277 			else if (insn.mm_fp0_format.op == mm_fmovz_op)
278 				func = fmovz_op;
279 			if (func != -1) {
280 				mips32_insn.fp0_format.opcode = cop1_op;
281 				mips32_insn.fp0_format.fmt =
282 					sdps_format[insn.mm_fp0_format.fmt];
283 				mips32_insn.fp0_format.ft =
284 					insn.mm_fp0_format.ft;
285 				mips32_insn.fp0_format.fs =
286 					insn.mm_fp0_format.fs;
287 				mips32_insn.fp0_format.fd =
288 					insn.mm_fp0_format.fd;
289 				mips32_insn.fp0_format.func = func;
290 			} else
291 				return SIGILL;
292 			break;
293 		case mm_32f_73_op:    /* POOL32FXF */
294 			switch (insn.mm_fp1_format.op) {
295 			case mm_movf0_op:
296 			case mm_movf1_op:
297 			case mm_movt0_op:
298 			case mm_movt1_op:
299 				if ((insn.mm_fp1_format.op & 0x7f) ==
300 				    mm_movf0_op)
301 					op = 0;
302 				else
303 					op = 1;
304 				mips32_insn.r_format.opcode = spec_op;
305 				mips32_insn.r_format.rs = insn.mm_fp4_format.fs;
306 				mips32_insn.r_format.rt =
307 					(insn.mm_fp4_format.cc << 2) + op;
308 				mips32_insn.r_format.rd = insn.mm_fp4_format.rt;
309 				mips32_insn.r_format.re = 0;
310 				mips32_insn.r_format.func = movc_op;
311 				break;
312 			case mm_fcvtd0_op:
313 			case mm_fcvtd1_op:
314 			case mm_fcvts0_op:
315 			case mm_fcvts1_op:
316 				if ((insn.mm_fp1_format.op & 0x7f) ==
317 				    mm_fcvtd0_op) {
318 					func = fcvtd_op;
319 					fmt = swl_format[insn.mm_fp3_format.fmt];
320 				} else {
321 					func = fcvts_op;
322 					fmt = dwl_format[insn.mm_fp3_format.fmt];
323 				}
324 				mips32_insn.fp0_format.opcode = cop1_op;
325 				mips32_insn.fp0_format.fmt = fmt;
326 				mips32_insn.fp0_format.ft = 0;
327 				mips32_insn.fp0_format.fs =
328 					insn.mm_fp3_format.fs;
329 				mips32_insn.fp0_format.fd =
330 					insn.mm_fp3_format.rt;
331 				mips32_insn.fp0_format.func = func;
332 				break;
333 			case mm_fmov0_op:
334 			case mm_fmov1_op:
335 			case mm_fabs0_op:
336 			case mm_fabs1_op:
337 			case mm_fneg0_op:
338 			case mm_fneg1_op:
339 				if ((insn.mm_fp1_format.op & 0x7f) ==
340 				    mm_fmov0_op)
341 					func = fmov_op;
342 				else if ((insn.mm_fp1_format.op & 0x7f) ==
343 					 mm_fabs0_op)
344 					func = fabs_op;
345 				else
346 					func = fneg_op;
347 				mips32_insn.fp0_format.opcode = cop1_op;
348 				mips32_insn.fp0_format.fmt =
349 					sdps_format[insn.mm_fp3_format.fmt];
350 				mips32_insn.fp0_format.ft = 0;
351 				mips32_insn.fp0_format.fs =
352 					insn.mm_fp3_format.fs;
353 				mips32_insn.fp0_format.fd =
354 					insn.mm_fp3_format.rt;
355 				mips32_insn.fp0_format.func = func;
356 				break;
357 			case mm_ffloorl_op:
358 			case mm_ffloorw_op:
359 			case mm_fceill_op:
360 			case mm_fceilw_op:
361 			case mm_ftruncl_op:
362 			case mm_ftruncw_op:
363 			case mm_froundl_op:
364 			case mm_froundw_op:
365 			case mm_fcvtl_op:
366 			case mm_fcvtw_op:
367 				if (insn.mm_fp1_format.op == mm_ffloorl_op)
368 					func = ffloorl_op;
369 				else if (insn.mm_fp1_format.op == mm_ffloorw_op)
370 					func = ffloor_op;
371 				else if (insn.mm_fp1_format.op == mm_fceill_op)
372 					func = fceill_op;
373 				else if (insn.mm_fp1_format.op == mm_fceilw_op)
374 					func = fceil_op;
375 				else if (insn.mm_fp1_format.op == mm_ftruncl_op)
376 					func = ftruncl_op;
377 				else if (insn.mm_fp1_format.op == mm_ftruncw_op)
378 					func = ftrunc_op;
379 				else if (insn.mm_fp1_format.op == mm_froundl_op)
380 					func = froundl_op;
381 				else if (insn.mm_fp1_format.op == mm_froundw_op)
382 					func = fround_op;
383 				else if (insn.mm_fp1_format.op == mm_fcvtl_op)
384 					func = fcvtl_op;
385 				else
386 					func = fcvtw_op;
387 				mips32_insn.fp0_format.opcode = cop1_op;
388 				mips32_insn.fp0_format.fmt =
389 					sd_format[insn.mm_fp1_format.fmt];
390 				mips32_insn.fp0_format.ft = 0;
391 				mips32_insn.fp0_format.fs =
392 					insn.mm_fp1_format.fs;
393 				mips32_insn.fp0_format.fd =
394 					insn.mm_fp1_format.rt;
395 				mips32_insn.fp0_format.func = func;
396 				break;
397 			case mm_frsqrt_op:
398 			case mm_fsqrt_op:
399 			case mm_frecip_op:
400 				if (insn.mm_fp1_format.op == mm_frsqrt_op)
401 					func = frsqrt_op;
402 				else if (insn.mm_fp1_format.op == mm_fsqrt_op)
403 					func = fsqrt_op;
404 				else
405 					func = frecip_op;
406 				mips32_insn.fp0_format.opcode = cop1_op;
407 				mips32_insn.fp0_format.fmt =
408 					sdps_format[insn.mm_fp1_format.fmt];
409 				mips32_insn.fp0_format.ft = 0;
410 				mips32_insn.fp0_format.fs =
411 					insn.mm_fp1_format.fs;
412 				mips32_insn.fp0_format.fd =
413 					insn.mm_fp1_format.rt;
414 				mips32_insn.fp0_format.func = func;
415 				break;
416 			case mm_mfc1_op:
417 			case mm_mtc1_op:
418 			case mm_cfc1_op:
419 			case mm_ctc1_op:
420 				if (insn.mm_fp1_format.op == mm_mfc1_op)
421 					op = mfc_op;
422 				else if (insn.mm_fp1_format.op == mm_mtc1_op)
423 					op = mtc_op;
424 				else if (insn.mm_fp1_format.op == mm_cfc1_op)
425 					op = cfc_op;
426 				else
427 					op = ctc_op;
428 				mips32_insn.fp1_format.opcode = cop1_op;
429 				mips32_insn.fp1_format.op = op;
430 				mips32_insn.fp1_format.rt =
431 					insn.mm_fp1_format.rt;
432 				mips32_insn.fp1_format.fs =
433 					insn.mm_fp1_format.fs;
434 				mips32_insn.fp1_format.fd = 0;
435 				mips32_insn.fp1_format.func = 0;
436 				break;
437 			default:
438 				return SIGILL;
439 				break;
440 			}
441 			break;
442 		case mm_32f_74_op:	/* c.cond.fmt */
443 			mips32_insn.fp0_format.opcode = cop1_op;
444 			mips32_insn.fp0_format.fmt =
445 				sdps_format[insn.mm_fp4_format.fmt];
446 			mips32_insn.fp0_format.ft = insn.mm_fp4_format.rt;
447 			mips32_insn.fp0_format.fs = insn.mm_fp4_format.fs;
448 			mips32_insn.fp0_format.fd = insn.mm_fp4_format.cc << 2;
449 			mips32_insn.fp0_format.func =
450 				insn.mm_fp4_format.cond | MM_MIPS32_COND_FC;
451 			break;
452 		default:
453 			return SIGILL;
454 			break;
455 		}
456 		break;
457 	default:
458 		return SIGILL;
459 		break;
460 	}
461 
462 	*insn_ptr = mips32_insn;
463 	return 0;
464 }
465 
466 int mm_isBranchInstr(struct pt_regs *regs, struct mm_decoded_insn dec_insn,
467 		     unsigned long *contpc)
468 {
469 	union mips_instruction insn = (union mips_instruction)dec_insn.insn;
470 	int bc_false = 0;
471 	unsigned int fcr31;
472 	unsigned int bit;
473 
474 	if (!cpu_has_mmips)
475 		return 0;
476 
477 	switch (insn.mm_i_format.opcode) {
478 	case mm_pool32a_op:
479 		if ((insn.mm_i_format.simmediate & MM_POOL32A_MINOR_MASK) ==
480 		    mm_pool32axf_op) {
481 			switch (insn.mm_i_format.simmediate >>
482 				MM_POOL32A_MINOR_SHIFT) {
483 			case mm_jalr_op:
484 			case mm_jalrhb_op:
485 			case mm_jalrs_op:
486 			case mm_jalrshb_op:
487 				if (insn.mm_i_format.rt != 0)	/* Not mm_jr */
488 					regs->regs[insn.mm_i_format.rt] =
489 						regs->cp0_epc +
490 						dec_insn.pc_inc +
491 						dec_insn.next_pc_inc;
492 				*contpc = regs->regs[insn.mm_i_format.rs];
493 				return 1;
494 				break;
495 			}
496 		}
497 		break;
498 	case mm_pool32i_op:
499 		switch (insn.mm_i_format.rt) {
500 		case mm_bltzals_op:
501 		case mm_bltzal_op:
502 			regs->regs[31] = regs->cp0_epc +
503 				dec_insn.pc_inc +
504 				dec_insn.next_pc_inc;
505 			/* Fall through */
506 		case mm_bltz_op:
507 			if ((long)regs->regs[insn.mm_i_format.rs] < 0)
508 				*contpc = regs->cp0_epc +
509 					dec_insn.pc_inc +
510 					(insn.mm_i_format.simmediate << 1);
511 			else
512 				*contpc = regs->cp0_epc +
513 					dec_insn.pc_inc +
514 					dec_insn.next_pc_inc;
515 			return 1;
516 			break;
517 		case mm_bgezals_op:
518 		case mm_bgezal_op:
519 			regs->regs[31] = regs->cp0_epc +
520 					dec_insn.pc_inc +
521 					dec_insn.next_pc_inc;
522 			/* Fall through */
523 		case mm_bgez_op:
524 			if ((long)regs->regs[insn.mm_i_format.rs] >= 0)
525 				*contpc = regs->cp0_epc +
526 					dec_insn.pc_inc +
527 					(insn.mm_i_format.simmediate << 1);
528 			else
529 				*contpc = regs->cp0_epc +
530 					dec_insn.pc_inc +
531 					dec_insn.next_pc_inc;
532 			return 1;
533 			break;
534 		case mm_blez_op:
535 			if ((long)regs->regs[insn.mm_i_format.rs] <= 0)
536 				*contpc = regs->cp0_epc +
537 					dec_insn.pc_inc +
538 					(insn.mm_i_format.simmediate << 1);
539 			else
540 				*contpc = regs->cp0_epc +
541 					dec_insn.pc_inc +
542 					dec_insn.next_pc_inc;
543 			return 1;
544 			break;
545 		case mm_bgtz_op:
546 			if ((long)regs->regs[insn.mm_i_format.rs] <= 0)
547 				*contpc = regs->cp0_epc +
548 					dec_insn.pc_inc +
549 					(insn.mm_i_format.simmediate << 1);
550 			else
551 				*contpc = regs->cp0_epc +
552 					dec_insn.pc_inc +
553 					dec_insn.next_pc_inc;
554 			return 1;
555 			break;
556 		case mm_bc2f_op:
557 		case mm_bc1f_op:
558 			bc_false = 1;
559 			/* Fall through */
560 		case mm_bc2t_op:
561 		case mm_bc1t_op:
562 			preempt_disable();
563 			if (is_fpu_owner())
564 				asm volatile("cfc1\t%0,$31" : "=r" (fcr31));
565 			else
566 				fcr31 = current->thread.fpu.fcr31;
567 			preempt_enable();
568 
569 			if (bc_false)
570 				fcr31 = ~fcr31;
571 
572 			bit = (insn.mm_i_format.rs >> 2);
573 			bit += (bit != 0);
574 			bit += 23;
575 			if (fcr31 & (1 << bit))
576 				*contpc = regs->cp0_epc +
577 					dec_insn.pc_inc +
578 					(insn.mm_i_format.simmediate << 1);
579 			else
580 				*contpc = regs->cp0_epc +
581 					dec_insn.pc_inc + dec_insn.next_pc_inc;
582 			return 1;
583 			break;
584 		}
585 		break;
586 	case mm_pool16c_op:
587 		switch (insn.mm_i_format.rt) {
588 		case mm_jalr16_op:
589 		case mm_jalrs16_op:
590 			regs->regs[31] = regs->cp0_epc +
591 				dec_insn.pc_inc + dec_insn.next_pc_inc;
592 			/* Fall through */
593 		case mm_jr16_op:
594 			*contpc = regs->regs[insn.mm_i_format.rs];
595 			return 1;
596 			break;
597 		}
598 		break;
599 	case mm_beqz16_op:
600 		if ((long)regs->regs[reg16to32map[insn.mm_b1_format.rs]] == 0)
601 			*contpc = regs->cp0_epc +
602 				dec_insn.pc_inc +
603 				(insn.mm_b1_format.simmediate << 1);
604 		else
605 			*contpc = regs->cp0_epc +
606 				dec_insn.pc_inc + dec_insn.next_pc_inc;
607 		return 1;
608 		break;
609 	case mm_bnez16_op:
610 		if ((long)regs->regs[reg16to32map[insn.mm_b1_format.rs]] != 0)
611 			*contpc = regs->cp0_epc +
612 				dec_insn.pc_inc +
613 				(insn.mm_b1_format.simmediate << 1);
614 		else
615 			*contpc = regs->cp0_epc +
616 				dec_insn.pc_inc + dec_insn.next_pc_inc;
617 		return 1;
618 		break;
619 	case mm_b16_op:
620 		*contpc = regs->cp0_epc + dec_insn.pc_inc +
621 			 (insn.mm_b0_format.simmediate << 1);
622 		return 1;
623 		break;
624 	case mm_beq32_op:
625 		if (regs->regs[insn.mm_i_format.rs] ==
626 		    regs->regs[insn.mm_i_format.rt])
627 			*contpc = regs->cp0_epc +
628 				dec_insn.pc_inc +
629 				(insn.mm_i_format.simmediate << 1);
630 		else
631 			*contpc = regs->cp0_epc +
632 				dec_insn.pc_inc +
633 				dec_insn.next_pc_inc;
634 		return 1;
635 		break;
636 	case mm_bne32_op:
637 		if (regs->regs[insn.mm_i_format.rs] !=
638 		    regs->regs[insn.mm_i_format.rt])
639 			*contpc = regs->cp0_epc +
640 				dec_insn.pc_inc +
641 				(insn.mm_i_format.simmediate << 1);
642 		else
643 			*contpc = regs->cp0_epc +
644 				dec_insn.pc_inc + dec_insn.next_pc_inc;
645 		return 1;
646 		break;
647 	case mm_jalx32_op:
648 		regs->regs[31] = regs->cp0_epc +
649 			dec_insn.pc_inc + dec_insn.next_pc_inc;
650 		*contpc = regs->cp0_epc + dec_insn.pc_inc;
651 		*contpc >>= 28;
652 		*contpc <<= 28;
653 		*contpc |= (insn.j_format.target << 2);
654 		return 1;
655 		break;
656 	case mm_jals32_op:
657 	case mm_jal32_op:
658 		regs->regs[31] = regs->cp0_epc +
659 			dec_insn.pc_inc + dec_insn.next_pc_inc;
660 		/* Fall through */
661 	case mm_j32_op:
662 		*contpc = regs->cp0_epc + dec_insn.pc_inc;
663 		*contpc >>= 27;
664 		*contpc <<= 27;
665 		*contpc |= (insn.j_format.target << 1);
666 		set_isa16_mode(*contpc);
667 		return 1;
668 		break;
669 	}
670 	return 0;
671 }
672 
673 /*
674  * Redundant with logic already in kernel/branch.c,
675  * embedded in compute_return_epc.  At some point,
676  * a single subroutine should be used across both
677  * modules.
678  */
679 static int isBranchInstr(struct pt_regs *regs, struct mm_decoded_insn dec_insn,
680 			 unsigned long *contpc)
681 {
682 	union mips_instruction insn = (union mips_instruction)dec_insn.insn;
683 	unsigned int fcr31;
684 	unsigned int bit = 0;
685 
686 	switch (insn.i_format.opcode) {
687 	case spec_op:
688 		switch (insn.r_format.func) {
689 		case jalr_op:
690 			regs->regs[insn.r_format.rd] =
691 				regs->cp0_epc + dec_insn.pc_inc +
692 				dec_insn.next_pc_inc;
693 			/* Fall through */
694 		case jr_op:
695 			*contpc = regs->regs[insn.r_format.rs];
696 			return 1;
697 			break;
698 		}
699 		break;
700 	case bcond_op:
701 		switch (insn.i_format.rt) {
702 		case bltzal_op:
703 		case bltzall_op:
704 			regs->regs[31] = regs->cp0_epc +
705 				dec_insn.pc_inc +
706 				dec_insn.next_pc_inc;
707 			/* Fall through */
708 		case bltz_op:
709 		case bltzl_op:
710 			if ((long)regs->regs[insn.i_format.rs] < 0)
711 				*contpc = regs->cp0_epc +
712 					dec_insn.pc_inc +
713 					(insn.i_format.simmediate << 2);
714 			else
715 				*contpc = regs->cp0_epc +
716 					dec_insn.pc_inc +
717 					dec_insn.next_pc_inc;
718 			return 1;
719 			break;
720 		case bgezal_op:
721 		case bgezall_op:
722 			regs->regs[31] = regs->cp0_epc +
723 				dec_insn.pc_inc +
724 				dec_insn.next_pc_inc;
725 			/* Fall through */
726 		case bgez_op:
727 		case bgezl_op:
728 			if ((long)regs->regs[insn.i_format.rs] >= 0)
729 				*contpc = regs->cp0_epc +
730 					dec_insn.pc_inc +
731 					(insn.i_format.simmediate << 2);
732 			else
733 				*contpc = regs->cp0_epc +
734 					dec_insn.pc_inc +
735 					dec_insn.next_pc_inc;
736 			return 1;
737 			break;
738 		}
739 		break;
740 	case jalx_op:
741 		set_isa16_mode(bit);
742 	case jal_op:
743 		regs->regs[31] = regs->cp0_epc +
744 			dec_insn.pc_inc +
745 			dec_insn.next_pc_inc;
746 		/* Fall through */
747 	case j_op:
748 		*contpc = regs->cp0_epc + dec_insn.pc_inc;
749 		*contpc >>= 28;
750 		*contpc <<= 28;
751 		*contpc |= (insn.j_format.target << 2);
752 		/* Set microMIPS mode bit: XOR for jalx. */
753 		*contpc ^= bit;
754 		return 1;
755 		break;
756 	case beq_op:
757 	case beql_op:
758 		if (regs->regs[insn.i_format.rs] ==
759 		    regs->regs[insn.i_format.rt])
760 			*contpc = regs->cp0_epc +
761 				dec_insn.pc_inc +
762 				(insn.i_format.simmediate << 2);
763 		else
764 			*contpc = regs->cp0_epc +
765 				dec_insn.pc_inc +
766 				dec_insn.next_pc_inc;
767 		return 1;
768 		break;
769 	case bne_op:
770 	case bnel_op:
771 		if (regs->regs[insn.i_format.rs] !=
772 		    regs->regs[insn.i_format.rt])
773 			*contpc = regs->cp0_epc +
774 				dec_insn.pc_inc +
775 				(insn.i_format.simmediate << 2);
776 		else
777 			*contpc = regs->cp0_epc +
778 				dec_insn.pc_inc +
779 				dec_insn.next_pc_inc;
780 		return 1;
781 		break;
782 	case blez_op:
783 	case blezl_op:
784 		if ((long)regs->regs[insn.i_format.rs] <= 0)
785 			*contpc = regs->cp0_epc +
786 				dec_insn.pc_inc +
787 				(insn.i_format.simmediate << 2);
788 		else
789 			*contpc = regs->cp0_epc +
790 				dec_insn.pc_inc +
791 				dec_insn.next_pc_inc;
792 		return 1;
793 		break;
794 	case bgtz_op:
795 	case bgtzl_op:
796 		if ((long)regs->regs[insn.i_format.rs] > 0)
797 			*contpc = regs->cp0_epc +
798 				dec_insn.pc_inc +
799 				(insn.i_format.simmediate << 2);
800 		else
801 			*contpc = regs->cp0_epc +
802 				dec_insn.pc_inc +
803 				dec_insn.next_pc_inc;
804 		return 1;
805 		break;
806 	case cop0_op:
807 	case cop1_op:
808 	case cop2_op:
809 	case cop1x_op:
810 		if (insn.i_format.rs == bc_op) {
811 			preempt_disable();
812 			if (is_fpu_owner())
813 				asm volatile("cfc1\t%0,$31" : "=r" (fcr31));
814 			else
815 				fcr31 = current->thread.fpu.fcr31;
816 			preempt_enable();
817 
818 			bit = (insn.i_format.rt >> 2);
819 			bit += (bit != 0);
820 			bit += 23;
821 			switch (insn.i_format.rt & 3) {
822 			case 0:	/* bc1f */
823 			case 2:	/* bc1fl */
824 				if (~fcr31 & (1 << bit))
825 					*contpc = regs->cp0_epc +
826 						dec_insn.pc_inc +
827 						(insn.i_format.simmediate << 2);
828 				else
829 					*contpc = regs->cp0_epc +
830 						dec_insn.pc_inc +
831 						dec_insn.next_pc_inc;
832 				return 1;
833 				break;
834 			case 1:	/* bc1t */
835 			case 3:	/* bc1tl */
836 				if (fcr31 & (1 << bit))
837 					*contpc = regs->cp0_epc +
838 						dec_insn.pc_inc +
839 						(insn.i_format.simmediate << 2);
840 				else
841 					*contpc = regs->cp0_epc +
842 						dec_insn.pc_inc +
843 						dec_insn.next_pc_inc;
844 				return 1;
845 				break;
846 			}
847 		}
848 		break;
849 	}
850 	return 0;
851 }
852 
853 /*
854  * In the Linux kernel, we support selection of FPR format on the
855  * basis of the Status.FR bit.	If an FPU is not present, the FR bit
856  * is hardwired to zero, which would imply a 32-bit FPU even for
857  * 64-bit CPUs so we rather look at TIF_32BIT_REGS.
858  * FPU emu is slow and bulky and optimizing this function offers fairly
859  * sizeable benefits so we try to be clever and make this function return
860  * a constant whenever possible, that is on 64-bit kernels without O32
861  * compatibility enabled and on 32-bit kernels.
862  */
863 static inline int cop1_64bit(struct pt_regs *xcp)
864 {
865 #if defined(CONFIG_64BIT) && !defined(CONFIG_MIPS32_O32)
866 	return 1;
867 #elif defined(CONFIG_64BIT) && defined(CONFIG_MIPS32_O32)
868 	return !test_thread_flag(TIF_32BIT_REGS);
869 #else
870 	return 0;
871 #endif
872 }
873 
874 #define SIFROMREG(si, x) ((si) = cop1_64bit(xcp) || !(x & 1) ? \
875 			(int)ctx->fpr[x] : (int)(ctx->fpr[x & ~1] >> 32))
876 
877 #define SITOREG(si, x)	(ctx->fpr[x & ~(cop1_64bit(xcp) == 0)] = \
878 			cop1_64bit(xcp) || !(x & 1) ? \
879 			ctx->fpr[x & ~1] >> 32 << 32 | (u32)(si) : \
880 			ctx->fpr[x & ~1] << 32 >> 32 | (u64)(si) << 32)
881 
882 #define DIFROMREG(di, x) ((di) = ctx->fpr[x & ~(cop1_64bit(xcp) == 0)])
883 #define DITOREG(di, x)	(ctx->fpr[x & ~(cop1_64bit(xcp) == 0)] = (di))
884 
885 #define SPFROMREG(sp, x) SIFROMREG((sp).bits, x)
886 #define SPTOREG(sp, x)	SITOREG((sp).bits, x)
887 #define DPFROMREG(dp, x)	DIFROMREG((dp).bits, x)
888 #define DPTOREG(dp, x)	DITOREG((dp).bits, x)
889 
890 /*
891  * Emulate the single floating point instruction pointed at by EPC.
892  * Two instructions if the instruction is in a branch delay slot.
893  */
894 
895 static int cop1Emulate(struct pt_regs *xcp, struct mips_fpu_struct *ctx,
896 		struct mm_decoded_insn dec_insn, void *__user *fault_addr)
897 {
898 	mips_instruction ir;
899 	unsigned long contpc = xcp->cp0_epc + dec_insn.pc_inc;
900 	unsigned int cond;
901 	int pc_inc;
902 
903 	/* XXX NEC Vr54xx bug workaround */
904 	if (xcp->cp0_cause & CAUSEF_BD) {
905 		if (dec_insn.micro_mips_mode) {
906 			if (!mm_isBranchInstr(xcp, dec_insn, &contpc))
907 				xcp->cp0_cause &= ~CAUSEF_BD;
908 		} else {
909 			if (!isBranchInstr(xcp, dec_insn, &contpc))
910 				xcp->cp0_cause &= ~CAUSEF_BD;
911 		}
912 	}
913 
914 	if (xcp->cp0_cause & CAUSEF_BD) {
915 		/*
916 		 * The instruction to be emulated is in a branch delay slot
917 		 * which means that we have to	emulate the branch instruction
918 		 * BEFORE we do the cop1 instruction.
919 		 *
920 		 * This branch could be a COP1 branch, but in that case we
921 		 * would have had a trap for that instruction, and would not
922 		 * come through this route.
923 		 *
924 		 * Linux MIPS branch emulator operates on context, updating the
925 		 * cp0_epc.
926 		 */
927 		ir = dec_insn.next_insn;  /* process delay slot instr */
928 		pc_inc = dec_insn.next_pc_inc;
929 	} else {
930 		ir = dec_insn.insn;       /* process current instr */
931 		pc_inc = dec_insn.pc_inc;
932 	}
933 
934 	/*
935 	 * Since microMIPS FPU instructios are a subset of MIPS32 FPU
936 	 * instructions, we want to convert microMIPS FPU instructions
937 	 * into MIPS32 instructions so that we could reuse all of the
938 	 * FPU emulation code.
939 	 *
940 	 * NOTE: We cannot do this for branch instructions since they
941 	 *       are not a subset. Example: Cannot emulate a 16-bit
942 	 *       aligned target address with a MIPS32 instruction.
943 	 */
944 	if (dec_insn.micro_mips_mode) {
945 		/*
946 		 * If next instruction is a 16-bit instruction, then it
947 		 * it cannot be a FPU instruction. This could happen
948 		 * since we can be called for non-FPU instructions.
949 		 */
950 		if ((pc_inc == 2) ||
951 			(microMIPS32_to_MIPS32((union mips_instruction *)&ir)
952 			 == SIGILL))
953 			return SIGILL;
954 	}
955 
956       emul:
957 	perf_sw_event(PERF_COUNT_SW_EMULATION_FAULTS, 1, xcp, 0);
958 	MIPS_FPU_EMU_INC_STATS(emulated);
959 	switch (MIPSInst_OPCODE(ir)) {
960 	case ldc1_op:{
961 		u64 __user *va = (u64 __user *) (xcp->regs[MIPSInst_RS(ir)] +
962 			MIPSInst_SIMM(ir));
963 		u64 val;
964 
965 		MIPS_FPU_EMU_INC_STATS(loads);
966 
967 		if (!access_ok(VERIFY_READ, va, sizeof(u64))) {
968 			MIPS_FPU_EMU_INC_STATS(errors);
969 			*fault_addr = va;
970 			return SIGBUS;
971 		}
972 		if (__get_user(val, va)) {
973 			MIPS_FPU_EMU_INC_STATS(errors);
974 			*fault_addr = va;
975 			return SIGSEGV;
976 		}
977 		DITOREG(val, MIPSInst_RT(ir));
978 		break;
979 	}
980 
981 	case sdc1_op:{
982 		u64 __user *va = (u64 __user *) (xcp->regs[MIPSInst_RS(ir)] +
983 			MIPSInst_SIMM(ir));
984 		u64 val;
985 
986 		MIPS_FPU_EMU_INC_STATS(stores);
987 		DIFROMREG(val, MIPSInst_RT(ir));
988 		if (!access_ok(VERIFY_WRITE, va, sizeof(u64))) {
989 			MIPS_FPU_EMU_INC_STATS(errors);
990 			*fault_addr = va;
991 			return SIGBUS;
992 		}
993 		if (__put_user(val, va)) {
994 			MIPS_FPU_EMU_INC_STATS(errors);
995 			*fault_addr = va;
996 			return SIGSEGV;
997 		}
998 		break;
999 	}
1000 
1001 	case lwc1_op:{
1002 		u32 __user *va = (u32 __user *) (xcp->regs[MIPSInst_RS(ir)] +
1003 			MIPSInst_SIMM(ir));
1004 		u32 val;
1005 
1006 		MIPS_FPU_EMU_INC_STATS(loads);
1007 		if (!access_ok(VERIFY_READ, va, sizeof(u32))) {
1008 			MIPS_FPU_EMU_INC_STATS(errors);
1009 			*fault_addr = va;
1010 			return SIGBUS;
1011 		}
1012 		if (__get_user(val, va)) {
1013 			MIPS_FPU_EMU_INC_STATS(errors);
1014 			*fault_addr = va;
1015 			return SIGSEGV;
1016 		}
1017 		SITOREG(val, MIPSInst_RT(ir));
1018 		break;
1019 	}
1020 
1021 	case swc1_op:{
1022 		u32 __user *va = (u32 __user *) (xcp->regs[MIPSInst_RS(ir)] +
1023 			MIPSInst_SIMM(ir));
1024 		u32 val;
1025 
1026 		MIPS_FPU_EMU_INC_STATS(stores);
1027 		SIFROMREG(val, MIPSInst_RT(ir));
1028 		if (!access_ok(VERIFY_WRITE, va, sizeof(u32))) {
1029 			MIPS_FPU_EMU_INC_STATS(errors);
1030 			*fault_addr = va;
1031 			return SIGBUS;
1032 		}
1033 		if (__put_user(val, va)) {
1034 			MIPS_FPU_EMU_INC_STATS(errors);
1035 			*fault_addr = va;
1036 			return SIGSEGV;
1037 		}
1038 		break;
1039 	}
1040 
1041 	case cop1_op:
1042 		switch (MIPSInst_RS(ir)) {
1043 
1044 #if defined(__mips64)
1045 		case dmfc_op:
1046 			/* copregister fs -> gpr[rt] */
1047 			if (MIPSInst_RT(ir) != 0) {
1048 				DIFROMREG(xcp->regs[MIPSInst_RT(ir)],
1049 					MIPSInst_RD(ir));
1050 			}
1051 			break;
1052 
1053 		case dmtc_op:
1054 			/* copregister fs <- rt */
1055 			DITOREG(xcp->regs[MIPSInst_RT(ir)], MIPSInst_RD(ir));
1056 			break;
1057 #endif
1058 
1059 		case mfc_op:
1060 			/* copregister rd -> gpr[rt] */
1061 			if (MIPSInst_RT(ir) != 0) {
1062 				SIFROMREG(xcp->regs[MIPSInst_RT(ir)],
1063 					MIPSInst_RD(ir));
1064 			}
1065 			break;
1066 
1067 		case mtc_op:
1068 			/* copregister rd <- rt */
1069 			SITOREG(xcp->regs[MIPSInst_RT(ir)], MIPSInst_RD(ir));
1070 			break;
1071 
1072 		case cfc_op:{
1073 			/* cop control register rd -> gpr[rt] */
1074 			u32 value;
1075 
1076 			if (MIPSInst_RD(ir) == FPCREG_CSR) {
1077 				value = ctx->fcr31;
1078 				value = (value & ~FPU_CSR_RM) |
1079 					mips_rm[modeindex(value)];
1080 #ifdef CSRTRACE
1081 				printk("%p gpr[%d]<-csr=%08x\n",
1082 					(void *) (xcp->cp0_epc),
1083 					MIPSInst_RT(ir), value);
1084 #endif
1085 			}
1086 			else if (MIPSInst_RD(ir) == FPCREG_RID)
1087 				value = 0;
1088 			else
1089 				value = 0;
1090 			if (MIPSInst_RT(ir))
1091 				xcp->regs[MIPSInst_RT(ir)] = value;
1092 			break;
1093 		}
1094 
1095 		case ctc_op:{
1096 			/* copregister rd <- rt */
1097 			u32 value;
1098 
1099 			if (MIPSInst_RT(ir) == 0)
1100 				value = 0;
1101 			else
1102 				value = xcp->regs[MIPSInst_RT(ir)];
1103 
1104 			/* we only have one writable control reg
1105 			 */
1106 			if (MIPSInst_RD(ir) == FPCREG_CSR) {
1107 #ifdef CSRTRACE
1108 				printk("%p gpr[%d]->csr=%08x\n",
1109 					(void *) (xcp->cp0_epc),
1110 					MIPSInst_RT(ir), value);
1111 #endif
1112 
1113 				/*
1114 				 * Don't write reserved bits,
1115 				 * and convert to ieee library modes
1116 				 */
1117 				ctx->fcr31 = (value &
1118 						~(FPU_CSR_RSVD | FPU_CSR_RM)) |
1119 						ieee_rm[modeindex(value)];
1120 			}
1121 			if ((ctx->fcr31 >> 5) & ctx->fcr31 & FPU_CSR_ALL_E) {
1122 				return SIGFPE;
1123 			}
1124 			break;
1125 		}
1126 
1127 		case bc_op:{
1128 			int likely = 0;
1129 
1130 			if (xcp->cp0_cause & CAUSEF_BD)
1131 				return SIGILL;
1132 
1133 #if __mips >= 4
1134 			cond = ctx->fcr31 & fpucondbit[MIPSInst_RT(ir) >> 2];
1135 #else
1136 			cond = ctx->fcr31 & FPU_CSR_COND;
1137 #endif
1138 			switch (MIPSInst_RT(ir) & 3) {
1139 			case bcfl_op:
1140 				likely = 1;
1141 			case bcf_op:
1142 				cond = !cond;
1143 				break;
1144 			case bctl_op:
1145 				likely = 1;
1146 			case bct_op:
1147 				break;
1148 			default:
1149 				/* thats an illegal instruction */
1150 				return SIGILL;
1151 			}
1152 
1153 			xcp->cp0_cause |= CAUSEF_BD;
1154 			if (cond) {
1155 				/* branch taken: emulate dslot
1156 				 * instruction
1157 				 */
1158 				xcp->cp0_epc += dec_insn.pc_inc;
1159 
1160 				contpc = MIPSInst_SIMM(ir);
1161 				ir = dec_insn.next_insn;
1162 				if (dec_insn.micro_mips_mode) {
1163 					contpc = (xcp->cp0_epc + (contpc << 1));
1164 
1165 					/* If 16-bit instruction, not FPU. */
1166 					if ((dec_insn.next_pc_inc == 2) ||
1167 						(microMIPS32_to_MIPS32((union mips_instruction *)&ir) == SIGILL)) {
1168 
1169 						/*
1170 						 * Since this instruction will
1171 						 * be put on the stack with
1172 						 * 32-bit words, get around
1173 						 * this problem by putting a
1174 						 * NOP16 as the second one.
1175 						 */
1176 						if (dec_insn.next_pc_inc == 2)
1177 							ir = (ir & (~0xffff)) | MM_NOP16;
1178 
1179 						/*
1180 						 * Single step the non-CP1
1181 						 * instruction in the dslot.
1182 						 */
1183 						return mips_dsemul(xcp, ir, contpc);
1184 					}
1185 				} else
1186 					contpc = (xcp->cp0_epc + (contpc << 2));
1187 
1188 				switch (MIPSInst_OPCODE(ir)) {
1189 				case lwc1_op:
1190 				case swc1_op:
1191 #if (__mips >= 2 || defined(__mips64))
1192 				case ldc1_op:
1193 				case sdc1_op:
1194 #endif
1195 				case cop1_op:
1196 #if __mips >= 4 && __mips != 32
1197 				case cop1x_op:
1198 #endif
1199 					/* its one of ours */
1200 					goto emul;
1201 #if __mips >= 4
1202 				case spec_op:
1203 					if (MIPSInst_FUNC(ir) == movc_op)
1204 						goto emul;
1205 					break;
1206 #endif
1207 				}
1208 
1209 				/*
1210 				 * Single step the non-cp1
1211 				 * instruction in the dslot
1212 				 */
1213 				return mips_dsemul(xcp, ir, contpc);
1214 			}
1215 			else {
1216 				/* branch not taken */
1217 				if (likely) {
1218 					/*
1219 					 * branch likely nullifies
1220 					 * dslot if not taken
1221 					 */
1222 					xcp->cp0_epc += dec_insn.pc_inc;
1223 					contpc += dec_insn.pc_inc;
1224 					/*
1225 					 * else continue & execute
1226 					 * dslot as normal insn
1227 					 */
1228 				}
1229 			}
1230 			break;
1231 		}
1232 
1233 		default:
1234 			if (!(MIPSInst_RS(ir) & 0x10))
1235 				return SIGILL;
1236 			{
1237 				int sig;
1238 
1239 				/* a real fpu computation instruction */
1240 				if ((sig = fpu_emu(xcp, ctx, ir)))
1241 					return sig;
1242 			}
1243 		}
1244 		break;
1245 
1246 #if __mips >= 4 && __mips != 32
1247 	case cop1x_op:{
1248 		int sig = fpux_emu(xcp, ctx, ir, fault_addr);
1249 		if (sig)
1250 			return sig;
1251 		break;
1252 	}
1253 #endif
1254 
1255 #if __mips >= 4
1256 	case spec_op:
1257 		if (MIPSInst_FUNC(ir) != movc_op)
1258 			return SIGILL;
1259 		cond = fpucondbit[MIPSInst_RT(ir) >> 2];
1260 		if (((ctx->fcr31 & cond) != 0) == ((MIPSInst_RT(ir) & 1) != 0))
1261 			xcp->regs[MIPSInst_RD(ir)] =
1262 				xcp->regs[MIPSInst_RS(ir)];
1263 		break;
1264 #endif
1265 
1266 	default:
1267 		return SIGILL;
1268 	}
1269 
1270 	/* we did it !! */
1271 	xcp->cp0_epc = contpc;
1272 	xcp->cp0_cause &= ~CAUSEF_BD;
1273 
1274 	return 0;
1275 }
1276 
1277 /*
1278  * Conversion table from MIPS compare ops 48-63
1279  * cond = ieee754dp_cmp(x,y,IEEE754_UN,sig);
1280  */
1281 static const unsigned char cmptab[8] = {
1282 	0,			/* cmp_0 (sig) cmp_sf */
1283 	IEEE754_CUN,		/* cmp_un (sig) cmp_ngle */
1284 	IEEE754_CEQ,		/* cmp_eq (sig) cmp_seq */
1285 	IEEE754_CEQ | IEEE754_CUN,	/* cmp_ueq (sig) cmp_ngl  */
1286 	IEEE754_CLT,		/* cmp_olt (sig) cmp_lt */
1287 	IEEE754_CLT | IEEE754_CUN,	/* cmp_ult (sig) cmp_nge */
1288 	IEEE754_CLT | IEEE754_CEQ,	/* cmp_ole (sig) cmp_le */
1289 	IEEE754_CLT | IEEE754_CEQ | IEEE754_CUN,	/* cmp_ule (sig) cmp_ngt */
1290 };
1291 
1292 
1293 #if __mips >= 4 && __mips != 32
1294 
1295 /*
1296  * Additional MIPS4 instructions
1297  */
1298 
1299 #define DEF3OP(name, p, f1, f2, f3) \
1300 static ieee754##p fpemu_##p##_##name(ieee754##p r, ieee754##p s, \
1301     ieee754##p t) \
1302 { \
1303 	struct _ieee754_csr ieee754_csr_save; \
1304 	s = f1(s, t); \
1305 	ieee754_csr_save = ieee754_csr; \
1306 	s = f2(s, r); \
1307 	ieee754_csr_save.cx |= ieee754_csr.cx; \
1308 	ieee754_csr_save.sx |= ieee754_csr.sx; \
1309 	s = f3(s); \
1310 	ieee754_csr.cx |= ieee754_csr_save.cx; \
1311 	ieee754_csr.sx |= ieee754_csr_save.sx; \
1312 	return s; \
1313 }
1314 
1315 static ieee754dp fpemu_dp_recip(ieee754dp d)
1316 {
1317 	return ieee754dp_div(ieee754dp_one(0), d);
1318 }
1319 
1320 static ieee754dp fpemu_dp_rsqrt(ieee754dp d)
1321 {
1322 	return ieee754dp_div(ieee754dp_one(0), ieee754dp_sqrt(d));
1323 }
1324 
1325 static ieee754sp fpemu_sp_recip(ieee754sp s)
1326 {
1327 	return ieee754sp_div(ieee754sp_one(0), s);
1328 }
1329 
1330 static ieee754sp fpemu_sp_rsqrt(ieee754sp s)
1331 {
1332 	return ieee754sp_div(ieee754sp_one(0), ieee754sp_sqrt(s));
1333 }
1334 
1335 DEF3OP(madd, sp, ieee754sp_mul, ieee754sp_add, );
1336 DEF3OP(msub, sp, ieee754sp_mul, ieee754sp_sub, );
1337 DEF3OP(nmadd, sp, ieee754sp_mul, ieee754sp_add, ieee754sp_neg);
1338 DEF3OP(nmsub, sp, ieee754sp_mul, ieee754sp_sub, ieee754sp_neg);
1339 DEF3OP(madd, dp, ieee754dp_mul, ieee754dp_add, );
1340 DEF3OP(msub, dp, ieee754dp_mul, ieee754dp_sub, );
1341 DEF3OP(nmadd, dp, ieee754dp_mul, ieee754dp_add, ieee754dp_neg);
1342 DEF3OP(nmsub, dp, ieee754dp_mul, ieee754dp_sub, ieee754dp_neg);
1343 
1344 static int fpux_emu(struct pt_regs *xcp, struct mips_fpu_struct *ctx,
1345 	mips_instruction ir, void *__user *fault_addr)
1346 {
1347 	unsigned rcsr = 0;	/* resulting csr */
1348 
1349 	MIPS_FPU_EMU_INC_STATS(cp1xops);
1350 
1351 	switch (MIPSInst_FMA_FFMT(ir)) {
1352 	case s_fmt:{		/* 0 */
1353 
1354 		ieee754sp(*handler) (ieee754sp, ieee754sp, ieee754sp);
1355 		ieee754sp fd, fr, fs, ft;
1356 		u32 __user *va;
1357 		u32 val;
1358 
1359 		switch (MIPSInst_FUNC(ir)) {
1360 		case lwxc1_op:
1361 			va = (void __user *) (xcp->regs[MIPSInst_FR(ir)] +
1362 				xcp->regs[MIPSInst_FT(ir)]);
1363 
1364 			MIPS_FPU_EMU_INC_STATS(loads);
1365 			if (!access_ok(VERIFY_READ, va, sizeof(u32))) {
1366 				MIPS_FPU_EMU_INC_STATS(errors);
1367 				*fault_addr = va;
1368 				return SIGBUS;
1369 			}
1370 			if (__get_user(val, va)) {
1371 				MIPS_FPU_EMU_INC_STATS(errors);
1372 				*fault_addr = va;
1373 				return SIGSEGV;
1374 			}
1375 			SITOREG(val, MIPSInst_FD(ir));
1376 			break;
1377 
1378 		case swxc1_op:
1379 			va = (void __user *) (xcp->regs[MIPSInst_FR(ir)] +
1380 				xcp->regs[MIPSInst_FT(ir)]);
1381 
1382 			MIPS_FPU_EMU_INC_STATS(stores);
1383 
1384 			SIFROMREG(val, MIPSInst_FS(ir));
1385 			if (!access_ok(VERIFY_WRITE, va, sizeof(u32))) {
1386 				MIPS_FPU_EMU_INC_STATS(errors);
1387 				*fault_addr = va;
1388 				return SIGBUS;
1389 			}
1390 			if (put_user(val, va)) {
1391 				MIPS_FPU_EMU_INC_STATS(errors);
1392 				*fault_addr = va;
1393 				return SIGSEGV;
1394 			}
1395 			break;
1396 
1397 		case madd_s_op:
1398 			handler = fpemu_sp_madd;
1399 			goto scoptop;
1400 		case msub_s_op:
1401 			handler = fpemu_sp_msub;
1402 			goto scoptop;
1403 		case nmadd_s_op:
1404 			handler = fpemu_sp_nmadd;
1405 			goto scoptop;
1406 		case nmsub_s_op:
1407 			handler = fpemu_sp_nmsub;
1408 			goto scoptop;
1409 
1410 		      scoptop:
1411 			SPFROMREG(fr, MIPSInst_FR(ir));
1412 			SPFROMREG(fs, MIPSInst_FS(ir));
1413 			SPFROMREG(ft, MIPSInst_FT(ir));
1414 			fd = (*handler) (fr, fs, ft);
1415 			SPTOREG(fd, MIPSInst_FD(ir));
1416 
1417 		      copcsr:
1418 			if (ieee754_cxtest(IEEE754_INEXACT))
1419 				rcsr |= FPU_CSR_INE_X | FPU_CSR_INE_S;
1420 			if (ieee754_cxtest(IEEE754_UNDERFLOW))
1421 				rcsr |= FPU_CSR_UDF_X | FPU_CSR_UDF_S;
1422 			if (ieee754_cxtest(IEEE754_OVERFLOW))
1423 				rcsr |= FPU_CSR_OVF_X | FPU_CSR_OVF_S;
1424 			if (ieee754_cxtest(IEEE754_INVALID_OPERATION))
1425 				rcsr |= FPU_CSR_INV_X | FPU_CSR_INV_S;
1426 
1427 			ctx->fcr31 = (ctx->fcr31 & ~FPU_CSR_ALL_X) | rcsr;
1428 			if ((ctx->fcr31 >> 5) & ctx->fcr31 & FPU_CSR_ALL_E) {
1429 				/*printk ("SIGFPE: fpu csr = %08x\n",
1430 				   ctx->fcr31); */
1431 				return SIGFPE;
1432 			}
1433 
1434 			break;
1435 
1436 		default:
1437 			return SIGILL;
1438 		}
1439 		break;
1440 	}
1441 
1442 	case d_fmt:{		/* 1 */
1443 		ieee754dp(*handler) (ieee754dp, ieee754dp, ieee754dp);
1444 		ieee754dp fd, fr, fs, ft;
1445 		u64 __user *va;
1446 		u64 val;
1447 
1448 		switch (MIPSInst_FUNC(ir)) {
1449 		case ldxc1_op:
1450 			va = (void __user *) (xcp->regs[MIPSInst_FR(ir)] +
1451 				xcp->regs[MIPSInst_FT(ir)]);
1452 
1453 			MIPS_FPU_EMU_INC_STATS(loads);
1454 			if (!access_ok(VERIFY_READ, va, sizeof(u64))) {
1455 				MIPS_FPU_EMU_INC_STATS(errors);
1456 				*fault_addr = va;
1457 				return SIGBUS;
1458 			}
1459 			if (__get_user(val, va)) {
1460 				MIPS_FPU_EMU_INC_STATS(errors);
1461 				*fault_addr = va;
1462 				return SIGSEGV;
1463 			}
1464 			DITOREG(val, MIPSInst_FD(ir));
1465 			break;
1466 
1467 		case sdxc1_op:
1468 			va = (void __user *) (xcp->regs[MIPSInst_FR(ir)] +
1469 				xcp->regs[MIPSInst_FT(ir)]);
1470 
1471 			MIPS_FPU_EMU_INC_STATS(stores);
1472 			DIFROMREG(val, MIPSInst_FS(ir));
1473 			if (!access_ok(VERIFY_WRITE, va, sizeof(u64))) {
1474 				MIPS_FPU_EMU_INC_STATS(errors);
1475 				*fault_addr = va;
1476 				return SIGBUS;
1477 			}
1478 			if (__put_user(val, va)) {
1479 				MIPS_FPU_EMU_INC_STATS(errors);
1480 				*fault_addr = va;
1481 				return SIGSEGV;
1482 			}
1483 			break;
1484 
1485 		case madd_d_op:
1486 			handler = fpemu_dp_madd;
1487 			goto dcoptop;
1488 		case msub_d_op:
1489 			handler = fpemu_dp_msub;
1490 			goto dcoptop;
1491 		case nmadd_d_op:
1492 			handler = fpemu_dp_nmadd;
1493 			goto dcoptop;
1494 		case nmsub_d_op:
1495 			handler = fpemu_dp_nmsub;
1496 			goto dcoptop;
1497 
1498 		      dcoptop:
1499 			DPFROMREG(fr, MIPSInst_FR(ir));
1500 			DPFROMREG(fs, MIPSInst_FS(ir));
1501 			DPFROMREG(ft, MIPSInst_FT(ir));
1502 			fd = (*handler) (fr, fs, ft);
1503 			DPTOREG(fd, MIPSInst_FD(ir));
1504 			goto copcsr;
1505 
1506 		default:
1507 			return SIGILL;
1508 		}
1509 		break;
1510 	}
1511 
1512 	case 0x7:		/* 7 */
1513 		if (MIPSInst_FUNC(ir) != pfetch_op) {
1514 			return SIGILL;
1515 		}
1516 		/* ignore prefx operation */
1517 		break;
1518 
1519 	default:
1520 		return SIGILL;
1521 	}
1522 
1523 	return 0;
1524 }
1525 #endif
1526 
1527 
1528 
1529 /*
1530  * Emulate a single COP1 arithmetic instruction.
1531  */
1532 static int fpu_emu(struct pt_regs *xcp, struct mips_fpu_struct *ctx,
1533 	mips_instruction ir)
1534 {
1535 	int rfmt;		/* resulting format */
1536 	unsigned rcsr = 0;	/* resulting csr */
1537 	unsigned cond;
1538 	union {
1539 		ieee754dp d;
1540 		ieee754sp s;
1541 		int w;
1542 #ifdef __mips64
1543 		s64 l;
1544 #endif
1545 	} rv;			/* resulting value */
1546 
1547 	MIPS_FPU_EMU_INC_STATS(cp1ops);
1548 	switch (rfmt = (MIPSInst_FFMT(ir) & 0xf)) {
1549 	case s_fmt:{		/* 0 */
1550 		union {
1551 			ieee754sp(*b) (ieee754sp, ieee754sp);
1552 			ieee754sp(*u) (ieee754sp);
1553 		} handler;
1554 
1555 		switch (MIPSInst_FUNC(ir)) {
1556 			/* binary ops */
1557 		case fadd_op:
1558 			handler.b = ieee754sp_add;
1559 			goto scopbop;
1560 		case fsub_op:
1561 			handler.b = ieee754sp_sub;
1562 			goto scopbop;
1563 		case fmul_op:
1564 			handler.b = ieee754sp_mul;
1565 			goto scopbop;
1566 		case fdiv_op:
1567 			handler.b = ieee754sp_div;
1568 			goto scopbop;
1569 
1570 			/* unary  ops */
1571 #if __mips >= 2 || defined(__mips64)
1572 		case fsqrt_op:
1573 			handler.u = ieee754sp_sqrt;
1574 			goto scopuop;
1575 #endif
1576 #if __mips >= 4 && __mips != 32
1577 		case frsqrt_op:
1578 			handler.u = fpemu_sp_rsqrt;
1579 			goto scopuop;
1580 		case frecip_op:
1581 			handler.u = fpemu_sp_recip;
1582 			goto scopuop;
1583 #endif
1584 #if __mips >= 4
1585 		case fmovc_op:
1586 			cond = fpucondbit[MIPSInst_FT(ir) >> 2];
1587 			if (((ctx->fcr31 & cond) != 0) !=
1588 				((MIPSInst_FT(ir) & 1) != 0))
1589 				return 0;
1590 			SPFROMREG(rv.s, MIPSInst_FS(ir));
1591 			break;
1592 		case fmovz_op:
1593 			if (xcp->regs[MIPSInst_FT(ir)] != 0)
1594 				return 0;
1595 			SPFROMREG(rv.s, MIPSInst_FS(ir));
1596 			break;
1597 		case fmovn_op:
1598 			if (xcp->regs[MIPSInst_FT(ir)] == 0)
1599 				return 0;
1600 			SPFROMREG(rv.s, MIPSInst_FS(ir));
1601 			break;
1602 #endif
1603 		case fabs_op:
1604 			handler.u = ieee754sp_abs;
1605 			goto scopuop;
1606 		case fneg_op:
1607 			handler.u = ieee754sp_neg;
1608 			goto scopuop;
1609 		case fmov_op:
1610 			/* an easy one */
1611 			SPFROMREG(rv.s, MIPSInst_FS(ir));
1612 			goto copcsr;
1613 
1614 			/* binary op on handler */
1615 		      scopbop:
1616 			{
1617 				ieee754sp fs, ft;
1618 
1619 				SPFROMREG(fs, MIPSInst_FS(ir));
1620 				SPFROMREG(ft, MIPSInst_FT(ir));
1621 
1622 				rv.s = (*handler.b) (fs, ft);
1623 				goto copcsr;
1624 			}
1625 		      scopuop:
1626 			{
1627 				ieee754sp fs;
1628 
1629 				SPFROMREG(fs, MIPSInst_FS(ir));
1630 				rv.s = (*handler.u) (fs);
1631 				goto copcsr;
1632 			}
1633 		      copcsr:
1634 			if (ieee754_cxtest(IEEE754_INEXACT))
1635 				rcsr |= FPU_CSR_INE_X | FPU_CSR_INE_S;
1636 			if (ieee754_cxtest(IEEE754_UNDERFLOW))
1637 				rcsr |= FPU_CSR_UDF_X | FPU_CSR_UDF_S;
1638 			if (ieee754_cxtest(IEEE754_OVERFLOW))
1639 				rcsr |= FPU_CSR_OVF_X | FPU_CSR_OVF_S;
1640 			if (ieee754_cxtest(IEEE754_ZERO_DIVIDE))
1641 				rcsr |= FPU_CSR_DIV_X | FPU_CSR_DIV_S;
1642 			if (ieee754_cxtest(IEEE754_INVALID_OPERATION))
1643 				rcsr |= FPU_CSR_INV_X | FPU_CSR_INV_S;
1644 			break;
1645 
1646 			/* unary conv ops */
1647 		case fcvts_op:
1648 			return SIGILL;	/* not defined */
1649 		case fcvtd_op:{
1650 			ieee754sp fs;
1651 
1652 			SPFROMREG(fs, MIPSInst_FS(ir));
1653 			rv.d = ieee754dp_fsp(fs);
1654 			rfmt = d_fmt;
1655 			goto copcsr;
1656 		}
1657 		case fcvtw_op:{
1658 			ieee754sp fs;
1659 
1660 			SPFROMREG(fs, MIPSInst_FS(ir));
1661 			rv.w = ieee754sp_tint(fs);
1662 			rfmt = w_fmt;
1663 			goto copcsr;
1664 		}
1665 
1666 #if __mips >= 2 || defined(__mips64)
1667 		case fround_op:
1668 		case ftrunc_op:
1669 		case fceil_op:
1670 		case ffloor_op:{
1671 			unsigned int oldrm = ieee754_csr.rm;
1672 			ieee754sp fs;
1673 
1674 			SPFROMREG(fs, MIPSInst_FS(ir));
1675 			ieee754_csr.rm = ieee_rm[modeindex(MIPSInst_FUNC(ir))];
1676 			rv.w = ieee754sp_tint(fs);
1677 			ieee754_csr.rm = oldrm;
1678 			rfmt = w_fmt;
1679 			goto copcsr;
1680 		}
1681 #endif /* __mips >= 2 */
1682 
1683 #if defined(__mips64)
1684 		case fcvtl_op:{
1685 			ieee754sp fs;
1686 
1687 			SPFROMREG(fs, MIPSInst_FS(ir));
1688 			rv.l = ieee754sp_tlong(fs);
1689 			rfmt = l_fmt;
1690 			goto copcsr;
1691 		}
1692 
1693 		case froundl_op:
1694 		case ftruncl_op:
1695 		case fceill_op:
1696 		case ffloorl_op:{
1697 			unsigned int oldrm = ieee754_csr.rm;
1698 			ieee754sp fs;
1699 
1700 			SPFROMREG(fs, MIPSInst_FS(ir));
1701 			ieee754_csr.rm = ieee_rm[modeindex(MIPSInst_FUNC(ir))];
1702 			rv.l = ieee754sp_tlong(fs);
1703 			ieee754_csr.rm = oldrm;
1704 			rfmt = l_fmt;
1705 			goto copcsr;
1706 		}
1707 #endif /* defined(__mips64) */
1708 
1709 		default:
1710 			if (MIPSInst_FUNC(ir) >= fcmp_op) {
1711 				unsigned cmpop = MIPSInst_FUNC(ir) - fcmp_op;
1712 				ieee754sp fs, ft;
1713 
1714 				SPFROMREG(fs, MIPSInst_FS(ir));
1715 				SPFROMREG(ft, MIPSInst_FT(ir));
1716 				rv.w = ieee754sp_cmp(fs, ft,
1717 					cmptab[cmpop & 0x7], cmpop & 0x8);
1718 				rfmt = -1;
1719 				if ((cmpop & 0x8) && ieee754_cxtest
1720 					(IEEE754_INVALID_OPERATION))
1721 					rcsr = FPU_CSR_INV_X | FPU_CSR_INV_S;
1722 				else
1723 					goto copcsr;
1724 
1725 			}
1726 			else {
1727 				return SIGILL;
1728 			}
1729 			break;
1730 		}
1731 		break;
1732 	}
1733 
1734 	case d_fmt:{
1735 		union {
1736 			ieee754dp(*b) (ieee754dp, ieee754dp);
1737 			ieee754dp(*u) (ieee754dp);
1738 		} handler;
1739 
1740 		switch (MIPSInst_FUNC(ir)) {
1741 			/* binary ops */
1742 		case fadd_op:
1743 			handler.b = ieee754dp_add;
1744 			goto dcopbop;
1745 		case fsub_op:
1746 			handler.b = ieee754dp_sub;
1747 			goto dcopbop;
1748 		case fmul_op:
1749 			handler.b = ieee754dp_mul;
1750 			goto dcopbop;
1751 		case fdiv_op:
1752 			handler.b = ieee754dp_div;
1753 			goto dcopbop;
1754 
1755 			/* unary  ops */
1756 #if __mips >= 2 || defined(__mips64)
1757 		case fsqrt_op:
1758 			handler.u = ieee754dp_sqrt;
1759 			goto dcopuop;
1760 #endif
1761 #if __mips >= 4 && __mips != 32
1762 		case frsqrt_op:
1763 			handler.u = fpemu_dp_rsqrt;
1764 			goto dcopuop;
1765 		case frecip_op:
1766 			handler.u = fpemu_dp_recip;
1767 			goto dcopuop;
1768 #endif
1769 #if __mips >= 4
1770 		case fmovc_op:
1771 			cond = fpucondbit[MIPSInst_FT(ir) >> 2];
1772 			if (((ctx->fcr31 & cond) != 0) !=
1773 				((MIPSInst_FT(ir) & 1) != 0))
1774 				return 0;
1775 			DPFROMREG(rv.d, MIPSInst_FS(ir));
1776 			break;
1777 		case fmovz_op:
1778 			if (xcp->regs[MIPSInst_FT(ir)] != 0)
1779 				return 0;
1780 			DPFROMREG(rv.d, MIPSInst_FS(ir));
1781 			break;
1782 		case fmovn_op:
1783 			if (xcp->regs[MIPSInst_FT(ir)] == 0)
1784 				return 0;
1785 			DPFROMREG(rv.d, MIPSInst_FS(ir));
1786 			break;
1787 #endif
1788 		case fabs_op:
1789 			handler.u = ieee754dp_abs;
1790 			goto dcopuop;
1791 
1792 		case fneg_op:
1793 			handler.u = ieee754dp_neg;
1794 			goto dcopuop;
1795 
1796 		case fmov_op:
1797 			/* an easy one */
1798 			DPFROMREG(rv.d, MIPSInst_FS(ir));
1799 			goto copcsr;
1800 
1801 			/* binary op on handler */
1802 		      dcopbop:{
1803 				ieee754dp fs, ft;
1804 
1805 				DPFROMREG(fs, MIPSInst_FS(ir));
1806 				DPFROMREG(ft, MIPSInst_FT(ir));
1807 
1808 				rv.d = (*handler.b) (fs, ft);
1809 				goto copcsr;
1810 			}
1811 		      dcopuop:{
1812 				ieee754dp fs;
1813 
1814 				DPFROMREG(fs, MIPSInst_FS(ir));
1815 				rv.d = (*handler.u) (fs);
1816 				goto copcsr;
1817 			}
1818 
1819 			/* unary conv ops */
1820 		case fcvts_op:{
1821 			ieee754dp fs;
1822 
1823 			DPFROMREG(fs, MIPSInst_FS(ir));
1824 			rv.s = ieee754sp_fdp(fs);
1825 			rfmt = s_fmt;
1826 			goto copcsr;
1827 		}
1828 		case fcvtd_op:
1829 			return SIGILL;	/* not defined */
1830 
1831 		case fcvtw_op:{
1832 			ieee754dp fs;
1833 
1834 			DPFROMREG(fs, MIPSInst_FS(ir));
1835 			rv.w = ieee754dp_tint(fs);	/* wrong */
1836 			rfmt = w_fmt;
1837 			goto copcsr;
1838 		}
1839 
1840 #if __mips >= 2 || defined(__mips64)
1841 		case fround_op:
1842 		case ftrunc_op:
1843 		case fceil_op:
1844 		case ffloor_op:{
1845 			unsigned int oldrm = ieee754_csr.rm;
1846 			ieee754dp fs;
1847 
1848 			DPFROMREG(fs, MIPSInst_FS(ir));
1849 			ieee754_csr.rm = ieee_rm[modeindex(MIPSInst_FUNC(ir))];
1850 			rv.w = ieee754dp_tint(fs);
1851 			ieee754_csr.rm = oldrm;
1852 			rfmt = w_fmt;
1853 			goto copcsr;
1854 		}
1855 #endif
1856 
1857 #if defined(__mips64)
1858 		case fcvtl_op:{
1859 			ieee754dp fs;
1860 
1861 			DPFROMREG(fs, MIPSInst_FS(ir));
1862 			rv.l = ieee754dp_tlong(fs);
1863 			rfmt = l_fmt;
1864 			goto copcsr;
1865 		}
1866 
1867 		case froundl_op:
1868 		case ftruncl_op:
1869 		case fceill_op:
1870 		case ffloorl_op:{
1871 			unsigned int oldrm = ieee754_csr.rm;
1872 			ieee754dp fs;
1873 
1874 			DPFROMREG(fs, MIPSInst_FS(ir));
1875 			ieee754_csr.rm = ieee_rm[modeindex(MIPSInst_FUNC(ir))];
1876 			rv.l = ieee754dp_tlong(fs);
1877 			ieee754_csr.rm = oldrm;
1878 			rfmt = l_fmt;
1879 			goto copcsr;
1880 		}
1881 #endif /* __mips >= 3 */
1882 
1883 		default:
1884 			if (MIPSInst_FUNC(ir) >= fcmp_op) {
1885 				unsigned cmpop = MIPSInst_FUNC(ir) - fcmp_op;
1886 				ieee754dp fs, ft;
1887 
1888 				DPFROMREG(fs, MIPSInst_FS(ir));
1889 				DPFROMREG(ft, MIPSInst_FT(ir));
1890 				rv.w = ieee754dp_cmp(fs, ft,
1891 					cmptab[cmpop & 0x7], cmpop & 0x8);
1892 				rfmt = -1;
1893 				if ((cmpop & 0x8)
1894 					&&
1895 					ieee754_cxtest
1896 					(IEEE754_INVALID_OPERATION))
1897 					rcsr = FPU_CSR_INV_X | FPU_CSR_INV_S;
1898 				else
1899 					goto copcsr;
1900 
1901 			}
1902 			else {
1903 				return SIGILL;
1904 			}
1905 			break;
1906 		}
1907 		break;
1908 	}
1909 
1910 	case w_fmt:{
1911 		ieee754sp fs;
1912 
1913 		switch (MIPSInst_FUNC(ir)) {
1914 		case fcvts_op:
1915 			/* convert word to single precision real */
1916 			SPFROMREG(fs, MIPSInst_FS(ir));
1917 			rv.s = ieee754sp_fint(fs.bits);
1918 			rfmt = s_fmt;
1919 			goto copcsr;
1920 		case fcvtd_op:
1921 			/* convert word to double precision real */
1922 			SPFROMREG(fs, MIPSInst_FS(ir));
1923 			rv.d = ieee754dp_fint(fs.bits);
1924 			rfmt = d_fmt;
1925 			goto copcsr;
1926 		default:
1927 			return SIGILL;
1928 		}
1929 		break;
1930 	}
1931 
1932 #if defined(__mips64)
1933 	case l_fmt:{
1934 		switch (MIPSInst_FUNC(ir)) {
1935 		case fcvts_op:
1936 			/* convert long to single precision real */
1937 			rv.s = ieee754sp_flong(ctx->fpr[MIPSInst_FS(ir)]);
1938 			rfmt = s_fmt;
1939 			goto copcsr;
1940 		case fcvtd_op:
1941 			/* convert long to double precision real */
1942 			rv.d = ieee754dp_flong(ctx->fpr[MIPSInst_FS(ir)]);
1943 			rfmt = d_fmt;
1944 			goto copcsr;
1945 		default:
1946 			return SIGILL;
1947 		}
1948 		break;
1949 	}
1950 #endif
1951 
1952 	default:
1953 		return SIGILL;
1954 	}
1955 
1956 	/*
1957 	 * Update the fpu CSR register for this operation.
1958 	 * If an exception is required, generate a tidy SIGFPE exception,
1959 	 * without updating the result register.
1960 	 * Note: cause exception bits do not accumulate, they are rewritten
1961 	 * for each op; only the flag/sticky bits accumulate.
1962 	 */
1963 	ctx->fcr31 = (ctx->fcr31 & ~FPU_CSR_ALL_X) | rcsr;
1964 	if ((ctx->fcr31 >> 5) & ctx->fcr31 & FPU_CSR_ALL_E) {
1965 		/*printk ("SIGFPE: fpu csr = %08x\n",ctx->fcr31); */
1966 		return SIGFPE;
1967 	}
1968 
1969 	/*
1970 	 * Now we can safely write the result back to the register file.
1971 	 */
1972 	switch (rfmt) {
1973 	case -1:{
1974 #if __mips >= 4
1975 		cond = fpucondbit[MIPSInst_FD(ir) >> 2];
1976 #else
1977 		cond = FPU_CSR_COND;
1978 #endif
1979 		if (rv.w)
1980 			ctx->fcr31 |= cond;
1981 		else
1982 			ctx->fcr31 &= ~cond;
1983 		break;
1984 	}
1985 	case d_fmt:
1986 		DPTOREG(rv.d, MIPSInst_FD(ir));
1987 		break;
1988 	case s_fmt:
1989 		SPTOREG(rv.s, MIPSInst_FD(ir));
1990 		break;
1991 	case w_fmt:
1992 		SITOREG(rv.w, MIPSInst_FD(ir));
1993 		break;
1994 #if defined(__mips64)
1995 	case l_fmt:
1996 		DITOREG(rv.l, MIPSInst_FD(ir));
1997 		break;
1998 #endif
1999 	default:
2000 		return SIGILL;
2001 	}
2002 
2003 	return 0;
2004 }
2005 
2006 int fpu_emulator_cop1Handler(struct pt_regs *xcp, struct mips_fpu_struct *ctx,
2007 	int has_fpu, void *__user *fault_addr)
2008 {
2009 	unsigned long oldepc, prevepc;
2010 	struct mm_decoded_insn dec_insn;
2011 	u16 instr[4];
2012 	u16 *instr_ptr;
2013 	int sig = 0;
2014 
2015 	oldepc = xcp->cp0_epc;
2016 	do {
2017 		prevepc = xcp->cp0_epc;
2018 
2019 		if (get_isa16_mode(prevepc) && cpu_has_mmips) {
2020 			/*
2021 			 * Get next 2 microMIPS instructions and convert them
2022 			 * into 32-bit instructions.
2023 			 */
2024 			if ((get_user(instr[0], (u16 __user *)msk_isa16_mode(xcp->cp0_epc))) ||
2025 			    (get_user(instr[1], (u16 __user *)msk_isa16_mode(xcp->cp0_epc + 2))) ||
2026 			    (get_user(instr[2], (u16 __user *)msk_isa16_mode(xcp->cp0_epc + 4))) ||
2027 			    (get_user(instr[3], (u16 __user *)msk_isa16_mode(xcp->cp0_epc + 6)))) {
2028 				MIPS_FPU_EMU_INC_STATS(errors);
2029 				return SIGBUS;
2030 			}
2031 			instr_ptr = instr;
2032 
2033 			/* Get first instruction. */
2034 			if (mm_insn_16bit(*instr_ptr)) {
2035 				/* Duplicate the half-word. */
2036 				dec_insn.insn = (*instr_ptr << 16) |
2037 					(*instr_ptr);
2038 				/* 16-bit instruction. */
2039 				dec_insn.pc_inc = 2;
2040 				instr_ptr += 1;
2041 			} else {
2042 				dec_insn.insn = (*instr_ptr << 16) |
2043 					*(instr_ptr+1);
2044 				/* 32-bit instruction. */
2045 				dec_insn.pc_inc = 4;
2046 				instr_ptr += 2;
2047 			}
2048 			/* Get second instruction. */
2049 			if (mm_insn_16bit(*instr_ptr)) {
2050 				/* Duplicate the half-word. */
2051 				dec_insn.next_insn = (*instr_ptr << 16) |
2052 					(*instr_ptr);
2053 				/* 16-bit instruction. */
2054 				dec_insn.next_pc_inc = 2;
2055 			} else {
2056 				dec_insn.next_insn = (*instr_ptr << 16) |
2057 					*(instr_ptr+1);
2058 				/* 32-bit instruction. */
2059 				dec_insn.next_pc_inc = 4;
2060 			}
2061 			dec_insn.micro_mips_mode = 1;
2062 		} else {
2063 			if ((get_user(dec_insn.insn,
2064 			    (mips_instruction __user *) xcp->cp0_epc)) ||
2065 			    (get_user(dec_insn.next_insn,
2066 			    (mips_instruction __user *)(xcp->cp0_epc+4)))) {
2067 				MIPS_FPU_EMU_INC_STATS(errors);
2068 				return SIGBUS;
2069 			}
2070 			dec_insn.pc_inc = 4;
2071 			dec_insn.next_pc_inc = 4;
2072 			dec_insn.micro_mips_mode = 0;
2073 		}
2074 
2075 		if ((dec_insn.insn == 0) ||
2076 		   ((dec_insn.pc_inc == 2) &&
2077 		   ((dec_insn.insn & 0xffff) == MM_NOP16)))
2078 			xcp->cp0_epc += dec_insn.pc_inc;	/* Skip NOPs */
2079 		else {
2080 			/*
2081 			 * The 'ieee754_csr' is an alias of
2082 			 * ctx->fcr31.	No need to copy ctx->fcr31 to
2083 			 * ieee754_csr.	 But ieee754_csr.rm is ieee
2084 			 * library modes. (not mips rounding mode)
2085 			 */
2086 			/* convert to ieee library modes */
2087 			ieee754_csr.rm = ieee_rm[ieee754_csr.rm];
2088 			sig = cop1Emulate(xcp, ctx, dec_insn, fault_addr);
2089 			/* revert to mips rounding mode */
2090 			ieee754_csr.rm = mips_rm[ieee754_csr.rm];
2091 		}
2092 
2093 		if (has_fpu)
2094 			break;
2095 		if (sig)
2096 			break;
2097 
2098 		cond_resched();
2099 	} while (xcp->cp0_epc > prevepc);
2100 
2101 	/* SIGILL indicates a non-fpu instruction */
2102 	if (sig == SIGILL && xcp->cp0_epc != oldepc)
2103 		/* but if epc has advanced, then ignore it */
2104 		sig = 0;
2105 
2106 	return sig;
2107 }
2108 
2109 #ifdef CONFIG_DEBUG_FS
2110 
2111 static int fpuemu_stat_get(void *data, u64 *val)
2112 {
2113 	int cpu;
2114 	unsigned long sum = 0;
2115 	for_each_online_cpu(cpu) {
2116 		struct mips_fpu_emulator_stats *ps;
2117 		local_t *pv;
2118 		ps = &per_cpu(fpuemustats, cpu);
2119 		pv = (void *)ps + (unsigned long)data;
2120 		sum += local_read(pv);
2121 	}
2122 	*val = sum;
2123 	return 0;
2124 }
2125 DEFINE_SIMPLE_ATTRIBUTE(fops_fpuemu_stat, fpuemu_stat_get, NULL, "%llu\n");
2126 
2127 extern struct dentry *mips_debugfs_dir;
2128 static int __init debugfs_fpuemu(void)
2129 {
2130 	struct dentry *d, *dir;
2131 
2132 	if (!mips_debugfs_dir)
2133 		return -ENODEV;
2134 	dir = debugfs_create_dir("fpuemustats", mips_debugfs_dir);
2135 	if (!dir)
2136 		return -ENOMEM;
2137 
2138 #define FPU_STAT_CREATE(M)						\
2139 	do {								\
2140 		d = debugfs_create_file(#M , S_IRUGO, dir,		\
2141 			(void *)offsetof(struct mips_fpu_emulator_stats, M), \
2142 			&fops_fpuemu_stat);				\
2143 		if (!d)							\
2144 			return -ENOMEM;					\
2145 	} while (0)
2146 
2147 	FPU_STAT_CREATE(emulated);
2148 	FPU_STAT_CREATE(loads);
2149 	FPU_STAT_CREATE(stores);
2150 	FPU_STAT_CREATE(cp1ops);
2151 	FPU_STAT_CREATE(cp1xops);
2152 	FPU_STAT_CREATE(errors);
2153 
2154 	return 0;
2155 }
2156 __initcall(debugfs_fpuemu);
2157 #endif
2158