xref: /linux/arch/mips/math-emu/cp1emu.c (revision c4ee0af3fa0dc65f690fc908f02b8355f9576ea0)
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 			}
440 			break;
441 		case mm_32f_74_op:	/* c.cond.fmt */
442 			mips32_insn.fp0_format.opcode = cop1_op;
443 			mips32_insn.fp0_format.fmt =
444 				sdps_format[insn.mm_fp4_format.fmt];
445 			mips32_insn.fp0_format.ft = insn.mm_fp4_format.rt;
446 			mips32_insn.fp0_format.fs = insn.mm_fp4_format.fs;
447 			mips32_insn.fp0_format.fd = insn.mm_fp4_format.cc << 2;
448 			mips32_insn.fp0_format.func =
449 				insn.mm_fp4_format.cond | MM_MIPS32_COND_FC;
450 			break;
451 		default:
452 			return SIGILL;
453 		}
454 		break;
455 	default:
456 		return SIGILL;
457 	}
458 
459 	*insn_ptr = mips32_insn;
460 	return 0;
461 }
462 
463 int mm_isBranchInstr(struct pt_regs *regs, struct mm_decoded_insn dec_insn,
464 		     unsigned long *contpc)
465 {
466 	union mips_instruction insn = (union mips_instruction)dec_insn.insn;
467 	int bc_false = 0;
468 	unsigned int fcr31;
469 	unsigned int bit;
470 
471 	if (!cpu_has_mmips)
472 		return 0;
473 
474 	switch (insn.mm_i_format.opcode) {
475 	case mm_pool32a_op:
476 		if ((insn.mm_i_format.simmediate & MM_POOL32A_MINOR_MASK) ==
477 		    mm_pool32axf_op) {
478 			switch (insn.mm_i_format.simmediate >>
479 				MM_POOL32A_MINOR_SHIFT) {
480 			case mm_jalr_op:
481 			case mm_jalrhb_op:
482 			case mm_jalrs_op:
483 			case mm_jalrshb_op:
484 				if (insn.mm_i_format.rt != 0)	/* Not mm_jr */
485 					regs->regs[insn.mm_i_format.rt] =
486 						regs->cp0_epc +
487 						dec_insn.pc_inc +
488 						dec_insn.next_pc_inc;
489 				*contpc = regs->regs[insn.mm_i_format.rs];
490 				return 1;
491 			}
492 		}
493 		break;
494 	case mm_pool32i_op:
495 		switch (insn.mm_i_format.rt) {
496 		case mm_bltzals_op:
497 		case mm_bltzal_op:
498 			regs->regs[31] = regs->cp0_epc +
499 				dec_insn.pc_inc +
500 				dec_insn.next_pc_inc;
501 			/* Fall through */
502 		case mm_bltz_op:
503 			if ((long)regs->regs[insn.mm_i_format.rs] < 0)
504 				*contpc = regs->cp0_epc +
505 					dec_insn.pc_inc +
506 					(insn.mm_i_format.simmediate << 1);
507 			else
508 				*contpc = regs->cp0_epc +
509 					dec_insn.pc_inc +
510 					dec_insn.next_pc_inc;
511 			return 1;
512 		case mm_bgezals_op:
513 		case mm_bgezal_op:
514 			regs->regs[31] = regs->cp0_epc +
515 					dec_insn.pc_inc +
516 					dec_insn.next_pc_inc;
517 			/* Fall through */
518 		case mm_bgez_op:
519 			if ((long)regs->regs[insn.mm_i_format.rs] >= 0)
520 				*contpc = regs->cp0_epc +
521 					dec_insn.pc_inc +
522 					(insn.mm_i_format.simmediate << 1);
523 			else
524 				*contpc = regs->cp0_epc +
525 					dec_insn.pc_inc +
526 					dec_insn.next_pc_inc;
527 			return 1;
528 		case mm_blez_op:
529 			if ((long)regs->regs[insn.mm_i_format.rs] <= 0)
530 				*contpc = regs->cp0_epc +
531 					dec_insn.pc_inc +
532 					(insn.mm_i_format.simmediate << 1);
533 			else
534 				*contpc = regs->cp0_epc +
535 					dec_insn.pc_inc +
536 					dec_insn.next_pc_inc;
537 			return 1;
538 		case mm_bgtz_op:
539 			if ((long)regs->regs[insn.mm_i_format.rs] <= 0)
540 				*contpc = regs->cp0_epc +
541 					dec_insn.pc_inc +
542 					(insn.mm_i_format.simmediate << 1);
543 			else
544 				*contpc = regs->cp0_epc +
545 					dec_insn.pc_inc +
546 					dec_insn.next_pc_inc;
547 			return 1;
548 		case mm_bc2f_op:
549 		case mm_bc1f_op:
550 			bc_false = 1;
551 			/* Fall through */
552 		case mm_bc2t_op:
553 		case mm_bc1t_op:
554 			preempt_disable();
555 			if (is_fpu_owner())
556 				asm volatile("cfc1\t%0,$31" : "=r" (fcr31));
557 			else
558 				fcr31 = current->thread.fpu.fcr31;
559 			preempt_enable();
560 
561 			if (bc_false)
562 				fcr31 = ~fcr31;
563 
564 			bit = (insn.mm_i_format.rs >> 2);
565 			bit += (bit != 0);
566 			bit += 23;
567 			if (fcr31 & (1 << bit))
568 				*contpc = regs->cp0_epc +
569 					dec_insn.pc_inc +
570 					(insn.mm_i_format.simmediate << 1);
571 			else
572 				*contpc = regs->cp0_epc +
573 					dec_insn.pc_inc + dec_insn.next_pc_inc;
574 			return 1;
575 		}
576 		break;
577 	case mm_pool16c_op:
578 		switch (insn.mm_i_format.rt) {
579 		case mm_jalr16_op:
580 		case mm_jalrs16_op:
581 			regs->regs[31] = regs->cp0_epc +
582 				dec_insn.pc_inc + dec_insn.next_pc_inc;
583 			/* Fall through */
584 		case mm_jr16_op:
585 			*contpc = regs->regs[insn.mm_i_format.rs];
586 			return 1;
587 		}
588 		break;
589 	case mm_beqz16_op:
590 		if ((long)regs->regs[reg16to32map[insn.mm_b1_format.rs]] == 0)
591 			*contpc = regs->cp0_epc +
592 				dec_insn.pc_inc +
593 				(insn.mm_b1_format.simmediate << 1);
594 		else
595 			*contpc = regs->cp0_epc +
596 				dec_insn.pc_inc + dec_insn.next_pc_inc;
597 		return 1;
598 	case mm_bnez16_op:
599 		if ((long)regs->regs[reg16to32map[insn.mm_b1_format.rs]] != 0)
600 			*contpc = regs->cp0_epc +
601 				dec_insn.pc_inc +
602 				(insn.mm_b1_format.simmediate << 1);
603 		else
604 			*contpc = regs->cp0_epc +
605 				dec_insn.pc_inc + dec_insn.next_pc_inc;
606 		return 1;
607 	case mm_b16_op:
608 		*contpc = regs->cp0_epc + dec_insn.pc_inc +
609 			 (insn.mm_b0_format.simmediate << 1);
610 		return 1;
611 	case mm_beq32_op:
612 		if (regs->regs[insn.mm_i_format.rs] ==
613 		    regs->regs[insn.mm_i_format.rt])
614 			*contpc = regs->cp0_epc +
615 				dec_insn.pc_inc +
616 				(insn.mm_i_format.simmediate << 1);
617 		else
618 			*contpc = regs->cp0_epc +
619 				dec_insn.pc_inc +
620 				dec_insn.next_pc_inc;
621 		return 1;
622 	case mm_bne32_op:
623 		if (regs->regs[insn.mm_i_format.rs] !=
624 		    regs->regs[insn.mm_i_format.rt])
625 			*contpc = regs->cp0_epc +
626 				dec_insn.pc_inc +
627 				(insn.mm_i_format.simmediate << 1);
628 		else
629 			*contpc = regs->cp0_epc +
630 				dec_insn.pc_inc + dec_insn.next_pc_inc;
631 		return 1;
632 	case mm_jalx32_op:
633 		regs->regs[31] = regs->cp0_epc +
634 			dec_insn.pc_inc + dec_insn.next_pc_inc;
635 		*contpc = regs->cp0_epc + dec_insn.pc_inc;
636 		*contpc >>= 28;
637 		*contpc <<= 28;
638 		*contpc |= (insn.j_format.target << 2);
639 		return 1;
640 	case mm_jals32_op:
641 	case mm_jal32_op:
642 		regs->regs[31] = regs->cp0_epc +
643 			dec_insn.pc_inc + dec_insn.next_pc_inc;
644 		/* Fall through */
645 	case mm_j32_op:
646 		*contpc = regs->cp0_epc + dec_insn.pc_inc;
647 		*contpc >>= 27;
648 		*contpc <<= 27;
649 		*contpc |= (insn.j_format.target << 1);
650 		set_isa16_mode(*contpc);
651 		return 1;
652 	}
653 	return 0;
654 }
655 
656 /*
657  * Redundant with logic already in kernel/branch.c,
658  * embedded in compute_return_epc.  At some point,
659  * a single subroutine should be used across both
660  * modules.
661  */
662 static int isBranchInstr(struct pt_regs *regs, struct mm_decoded_insn dec_insn,
663 			 unsigned long *contpc)
664 {
665 	union mips_instruction insn = (union mips_instruction)dec_insn.insn;
666 	unsigned int fcr31;
667 	unsigned int bit = 0;
668 
669 	switch (insn.i_format.opcode) {
670 	case spec_op:
671 		switch (insn.r_format.func) {
672 		case jalr_op:
673 			regs->regs[insn.r_format.rd] =
674 				regs->cp0_epc + dec_insn.pc_inc +
675 				dec_insn.next_pc_inc;
676 			/* Fall through */
677 		case jr_op:
678 			*contpc = regs->regs[insn.r_format.rs];
679 			return 1;
680 		}
681 		break;
682 	case bcond_op:
683 		switch (insn.i_format.rt) {
684 		case bltzal_op:
685 		case bltzall_op:
686 			regs->regs[31] = regs->cp0_epc +
687 				dec_insn.pc_inc +
688 				dec_insn.next_pc_inc;
689 			/* Fall through */
690 		case bltz_op:
691 		case bltzl_op:
692 			if ((long)regs->regs[insn.i_format.rs] < 0)
693 				*contpc = regs->cp0_epc +
694 					dec_insn.pc_inc +
695 					(insn.i_format.simmediate << 2);
696 			else
697 				*contpc = regs->cp0_epc +
698 					dec_insn.pc_inc +
699 					dec_insn.next_pc_inc;
700 			return 1;
701 		case bgezal_op:
702 		case bgezall_op:
703 			regs->regs[31] = regs->cp0_epc +
704 				dec_insn.pc_inc +
705 				dec_insn.next_pc_inc;
706 			/* Fall through */
707 		case bgez_op:
708 		case bgezl_op:
709 			if ((long)regs->regs[insn.i_format.rs] >= 0)
710 				*contpc = regs->cp0_epc +
711 					dec_insn.pc_inc +
712 					(insn.i_format.simmediate << 2);
713 			else
714 				*contpc = regs->cp0_epc +
715 					dec_insn.pc_inc +
716 					dec_insn.next_pc_inc;
717 			return 1;
718 		}
719 		break;
720 	case jalx_op:
721 		set_isa16_mode(bit);
722 	case jal_op:
723 		regs->regs[31] = regs->cp0_epc +
724 			dec_insn.pc_inc +
725 			dec_insn.next_pc_inc;
726 		/* Fall through */
727 	case j_op:
728 		*contpc = regs->cp0_epc + dec_insn.pc_inc;
729 		*contpc >>= 28;
730 		*contpc <<= 28;
731 		*contpc |= (insn.j_format.target << 2);
732 		/* Set microMIPS mode bit: XOR for jalx. */
733 		*contpc ^= bit;
734 		return 1;
735 	case beq_op:
736 	case beql_op:
737 		if (regs->regs[insn.i_format.rs] ==
738 		    regs->regs[insn.i_format.rt])
739 			*contpc = regs->cp0_epc +
740 				dec_insn.pc_inc +
741 				(insn.i_format.simmediate << 2);
742 		else
743 			*contpc = regs->cp0_epc +
744 				dec_insn.pc_inc +
745 				dec_insn.next_pc_inc;
746 		return 1;
747 	case bne_op:
748 	case bnel_op:
749 		if (regs->regs[insn.i_format.rs] !=
750 		    regs->regs[insn.i_format.rt])
751 			*contpc = regs->cp0_epc +
752 				dec_insn.pc_inc +
753 				(insn.i_format.simmediate << 2);
754 		else
755 			*contpc = regs->cp0_epc +
756 				dec_insn.pc_inc +
757 				dec_insn.next_pc_inc;
758 		return 1;
759 	case blez_op:
760 	case blezl_op:
761 		if ((long)regs->regs[insn.i_format.rs] <= 0)
762 			*contpc = regs->cp0_epc +
763 				dec_insn.pc_inc +
764 				(insn.i_format.simmediate << 2);
765 		else
766 			*contpc = regs->cp0_epc +
767 				dec_insn.pc_inc +
768 				dec_insn.next_pc_inc;
769 		return 1;
770 	case bgtz_op:
771 	case bgtzl_op:
772 		if ((long)regs->regs[insn.i_format.rs] > 0)
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 #ifdef CONFIG_CPU_CAVIUM_OCTEON
782 	case lwc2_op: /* This is bbit0 on Octeon */
783 		if ((regs->regs[insn.i_format.rs] & (1ull<<insn.i_format.rt)) == 0)
784 			*contpc = regs->cp0_epc + 4 + (insn.i_format.simmediate << 2);
785 		else
786 			*contpc = regs->cp0_epc + 8;
787 		return 1;
788 	case ldc2_op: /* This is bbit032 on Octeon */
789 		if ((regs->regs[insn.i_format.rs] & (1ull<<(insn.i_format.rt + 32))) == 0)
790 			*contpc = regs->cp0_epc + 4 + (insn.i_format.simmediate << 2);
791 		else
792 			*contpc = regs->cp0_epc + 8;
793 		return 1;
794 	case swc2_op: /* This is bbit1 on Octeon */
795 		if (regs->regs[insn.i_format.rs] & (1ull<<insn.i_format.rt))
796 			*contpc = regs->cp0_epc + 4 + (insn.i_format.simmediate << 2);
797 		else
798 			*contpc = regs->cp0_epc + 8;
799 		return 1;
800 	case sdc2_op: /* This is bbit132 on Octeon */
801 		if (regs->regs[insn.i_format.rs] & (1ull<<(insn.i_format.rt + 32)))
802 			*contpc = regs->cp0_epc + 4 + (insn.i_format.simmediate << 2);
803 		else
804 			*contpc = regs->cp0_epc + 8;
805 		return 1;
806 #endif
807 	case cop0_op:
808 	case cop1_op:
809 	case cop2_op:
810 	case cop1x_op:
811 		if (insn.i_format.rs == bc_op) {
812 			preempt_disable();
813 			if (is_fpu_owner())
814 				asm volatile("cfc1\t%0,$31" : "=r" (fcr31));
815 			else
816 				fcr31 = current->thread.fpu.fcr31;
817 			preempt_enable();
818 
819 			bit = (insn.i_format.rt >> 2);
820 			bit += (bit != 0);
821 			bit += 23;
822 			switch (insn.i_format.rt & 3) {
823 			case 0:	/* bc1f */
824 			case 2:	/* bc1fl */
825 				if (~fcr31 & (1 << bit))
826 					*contpc = regs->cp0_epc +
827 						dec_insn.pc_inc +
828 						(insn.i_format.simmediate << 2);
829 				else
830 					*contpc = regs->cp0_epc +
831 						dec_insn.pc_inc +
832 						dec_insn.next_pc_inc;
833 				return 1;
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 			}
846 		}
847 		break;
848 	}
849 	return 0;
850 }
851 
852 /*
853  * In the Linux kernel, we support selection of FPR format on the
854  * basis of the Status.FR bit.	If an FPU is not present, the FR bit
855  * is hardwired to zero, which would imply a 32-bit FPU even for
856  * 64-bit CPUs so we rather look at TIF_32BIT_REGS.
857  * FPU emu is slow and bulky and optimizing this function offers fairly
858  * sizeable benefits so we try to be clever and make this function return
859  * a constant whenever possible, that is on 64-bit kernels without O32
860  * compatibility enabled and on 32-bit kernels.
861  */
862 static inline int cop1_64bit(struct pt_regs *xcp)
863 {
864 #if defined(CONFIG_64BIT) && !defined(CONFIG_MIPS32_O32)
865 	return 1;
866 #elif defined(CONFIG_64BIT) && defined(CONFIG_MIPS32_O32)
867 	return !test_thread_flag(TIF_32BIT_REGS);
868 #else
869 	return 0;
870 #endif
871 }
872 
873 #define SIFROMREG(si, x) ((si) = cop1_64bit(xcp) || !(x & 1) ? \
874 			(int)ctx->fpr[x] : (int)(ctx->fpr[x & ~1] >> 32))
875 
876 #define SITOREG(si, x)	(ctx->fpr[x & ~(cop1_64bit(xcp) == 0)] = \
877 			cop1_64bit(xcp) || !(x & 1) ? \
878 			ctx->fpr[x & ~1] >> 32 << 32 | (u32)(si) : \
879 			ctx->fpr[x & ~1] << 32 >> 32 | (u64)(si) << 32)
880 
881 #define DIFROMREG(di, x) ((di) = ctx->fpr[x & ~(cop1_64bit(xcp) == 0)])
882 #define DITOREG(di, x)	(ctx->fpr[x & ~(cop1_64bit(xcp) == 0)] = (di))
883 
884 #define SPFROMREG(sp, x) SIFROMREG((sp).bits, x)
885 #define SPTOREG(sp, x)	SITOREG((sp).bits, x)
886 #define DPFROMREG(dp, x)	DIFROMREG((dp).bits, x)
887 #define DPTOREG(dp, x)	DITOREG((dp).bits, x)
888 
889 /*
890  * Emulate the single floating point instruction pointed at by EPC.
891  * Two instructions if the instruction is in a branch delay slot.
892  */
893 
894 static int cop1Emulate(struct pt_regs *xcp, struct mips_fpu_struct *ctx,
895 		struct mm_decoded_insn dec_insn, void *__user *fault_addr)
896 {
897 	mips_instruction ir;
898 	unsigned long contpc = xcp->cp0_epc + dec_insn.pc_inc;
899 	unsigned int cond;
900 	int pc_inc;
901 
902 	/* XXX NEC Vr54xx bug workaround */
903 	if (xcp->cp0_cause & CAUSEF_BD) {
904 		if (dec_insn.micro_mips_mode) {
905 			if (!mm_isBranchInstr(xcp, dec_insn, &contpc))
906 				xcp->cp0_cause &= ~CAUSEF_BD;
907 		} else {
908 			if (!isBranchInstr(xcp, dec_insn, &contpc))
909 				xcp->cp0_cause &= ~CAUSEF_BD;
910 		}
911 	}
912 
913 	if (xcp->cp0_cause & CAUSEF_BD) {
914 		/*
915 		 * The instruction to be emulated is in a branch delay slot
916 		 * which means that we have to	emulate the branch instruction
917 		 * BEFORE we do the cop1 instruction.
918 		 *
919 		 * This branch could be a COP1 branch, but in that case we
920 		 * would have had a trap for that instruction, and would not
921 		 * come through this route.
922 		 *
923 		 * Linux MIPS branch emulator operates on context, updating the
924 		 * cp0_epc.
925 		 */
926 		ir = dec_insn.next_insn;  /* process delay slot instr */
927 		pc_inc = dec_insn.next_pc_inc;
928 	} else {
929 		ir = dec_insn.insn;       /* process current instr */
930 		pc_inc = dec_insn.pc_inc;
931 	}
932 
933 	/*
934 	 * Since microMIPS FPU instructios are a subset of MIPS32 FPU
935 	 * instructions, we want to convert microMIPS FPU instructions
936 	 * into MIPS32 instructions so that we could reuse all of the
937 	 * FPU emulation code.
938 	 *
939 	 * NOTE: We cannot do this for branch instructions since they
940 	 *       are not a subset. Example: Cannot emulate a 16-bit
941 	 *       aligned target address with a MIPS32 instruction.
942 	 */
943 	if (dec_insn.micro_mips_mode) {
944 		/*
945 		 * If next instruction is a 16-bit instruction, then it
946 		 * it cannot be a FPU instruction. This could happen
947 		 * since we can be called for non-FPU instructions.
948 		 */
949 		if ((pc_inc == 2) ||
950 			(microMIPS32_to_MIPS32((union mips_instruction *)&ir)
951 			 == SIGILL))
952 			return SIGILL;
953 	}
954 
955       emul:
956 	perf_sw_event(PERF_COUNT_SW_EMULATION_FAULTS, 1, xcp, 0);
957 	MIPS_FPU_EMU_INC_STATS(emulated);
958 	switch (MIPSInst_OPCODE(ir)) {
959 	case ldc1_op:{
960 		u64 __user *va = (u64 __user *) (xcp->regs[MIPSInst_RS(ir)] +
961 			MIPSInst_SIMM(ir));
962 		u64 val;
963 
964 		MIPS_FPU_EMU_INC_STATS(loads);
965 
966 		if (!access_ok(VERIFY_READ, va, sizeof(u64))) {
967 			MIPS_FPU_EMU_INC_STATS(errors);
968 			*fault_addr = va;
969 			return SIGBUS;
970 		}
971 		if (__get_user(val, va)) {
972 			MIPS_FPU_EMU_INC_STATS(errors);
973 			*fault_addr = va;
974 			return SIGSEGV;
975 		}
976 		DITOREG(val, MIPSInst_RT(ir));
977 		break;
978 	}
979 
980 	case sdc1_op:{
981 		u64 __user *va = (u64 __user *) (xcp->regs[MIPSInst_RS(ir)] +
982 			MIPSInst_SIMM(ir));
983 		u64 val;
984 
985 		MIPS_FPU_EMU_INC_STATS(stores);
986 		DIFROMREG(val, MIPSInst_RT(ir));
987 		if (!access_ok(VERIFY_WRITE, va, sizeof(u64))) {
988 			MIPS_FPU_EMU_INC_STATS(errors);
989 			*fault_addr = va;
990 			return SIGBUS;
991 		}
992 		if (__put_user(val, va)) {
993 			MIPS_FPU_EMU_INC_STATS(errors);
994 			*fault_addr = va;
995 			return SIGSEGV;
996 		}
997 		break;
998 	}
999 
1000 	case lwc1_op:{
1001 		u32 __user *va = (u32 __user *) (xcp->regs[MIPSInst_RS(ir)] +
1002 			MIPSInst_SIMM(ir));
1003 		u32 val;
1004 
1005 		MIPS_FPU_EMU_INC_STATS(loads);
1006 		if (!access_ok(VERIFY_READ, va, sizeof(u32))) {
1007 			MIPS_FPU_EMU_INC_STATS(errors);
1008 			*fault_addr = va;
1009 			return SIGBUS;
1010 		}
1011 		if (__get_user(val, va)) {
1012 			MIPS_FPU_EMU_INC_STATS(errors);
1013 			*fault_addr = va;
1014 			return SIGSEGV;
1015 		}
1016 		SITOREG(val, MIPSInst_RT(ir));
1017 		break;
1018 	}
1019 
1020 	case swc1_op:{
1021 		u32 __user *va = (u32 __user *) (xcp->regs[MIPSInst_RS(ir)] +
1022 			MIPSInst_SIMM(ir));
1023 		u32 val;
1024 
1025 		MIPS_FPU_EMU_INC_STATS(stores);
1026 		SIFROMREG(val, MIPSInst_RT(ir));
1027 		if (!access_ok(VERIFY_WRITE, va, sizeof(u32))) {
1028 			MIPS_FPU_EMU_INC_STATS(errors);
1029 			*fault_addr = va;
1030 			return SIGBUS;
1031 		}
1032 		if (__put_user(val, va)) {
1033 			MIPS_FPU_EMU_INC_STATS(errors);
1034 			*fault_addr = va;
1035 			return SIGSEGV;
1036 		}
1037 		break;
1038 	}
1039 
1040 	case cop1_op:
1041 		switch (MIPSInst_RS(ir)) {
1042 
1043 #if defined(__mips64)
1044 		case dmfc_op:
1045 			/* copregister fs -> gpr[rt] */
1046 			if (MIPSInst_RT(ir) != 0) {
1047 				DIFROMREG(xcp->regs[MIPSInst_RT(ir)],
1048 					MIPSInst_RD(ir));
1049 			}
1050 			break;
1051 
1052 		case dmtc_op:
1053 			/* copregister fs <- rt */
1054 			DITOREG(xcp->regs[MIPSInst_RT(ir)], MIPSInst_RD(ir));
1055 			break;
1056 #endif
1057 
1058 		case mfc_op:
1059 			/* copregister rd -> gpr[rt] */
1060 			if (MIPSInst_RT(ir) != 0) {
1061 				SIFROMREG(xcp->regs[MIPSInst_RT(ir)],
1062 					MIPSInst_RD(ir));
1063 			}
1064 			break;
1065 
1066 		case mtc_op:
1067 			/* copregister rd <- rt */
1068 			SITOREG(xcp->regs[MIPSInst_RT(ir)], MIPSInst_RD(ir));
1069 			break;
1070 
1071 		case cfc_op:{
1072 			/* cop control register rd -> gpr[rt] */
1073 			u32 value;
1074 
1075 			if (MIPSInst_RD(ir) == FPCREG_CSR) {
1076 				value = ctx->fcr31;
1077 				value = (value & ~FPU_CSR_RM) |
1078 					mips_rm[modeindex(value)];
1079 #ifdef CSRTRACE
1080 				printk("%p gpr[%d]<-csr=%08x\n",
1081 					(void *) (xcp->cp0_epc),
1082 					MIPSInst_RT(ir), value);
1083 #endif
1084 			}
1085 			else if (MIPSInst_RD(ir) == FPCREG_RID)
1086 				value = 0;
1087 			else
1088 				value = 0;
1089 			if (MIPSInst_RT(ir))
1090 				xcp->regs[MIPSInst_RT(ir)] = value;
1091 			break;
1092 		}
1093 
1094 		case ctc_op:{
1095 			/* copregister rd <- rt */
1096 			u32 value;
1097 
1098 			if (MIPSInst_RT(ir) == 0)
1099 				value = 0;
1100 			else
1101 				value = xcp->regs[MIPSInst_RT(ir)];
1102 
1103 			/* we only have one writable control reg
1104 			 */
1105 			if (MIPSInst_RD(ir) == FPCREG_CSR) {
1106 #ifdef CSRTRACE
1107 				printk("%p gpr[%d]->csr=%08x\n",
1108 					(void *) (xcp->cp0_epc),
1109 					MIPSInst_RT(ir), value);
1110 #endif
1111 
1112 				/*
1113 				 * Don't write reserved bits,
1114 				 * and convert to ieee library modes
1115 				 */
1116 				ctx->fcr31 = (value &
1117 						~(FPU_CSR_RSVD | FPU_CSR_RM)) |
1118 						ieee_rm[modeindex(value)];
1119 			}
1120 			if ((ctx->fcr31 >> 5) & ctx->fcr31 & FPU_CSR_ALL_E) {
1121 				return SIGFPE;
1122 			}
1123 			break;
1124 		}
1125 
1126 		case bc_op:{
1127 			int likely = 0;
1128 
1129 			if (xcp->cp0_cause & CAUSEF_BD)
1130 				return SIGILL;
1131 
1132 #if __mips >= 4
1133 			cond = ctx->fcr31 & fpucondbit[MIPSInst_RT(ir) >> 2];
1134 #else
1135 			cond = ctx->fcr31 & FPU_CSR_COND;
1136 #endif
1137 			switch (MIPSInst_RT(ir) & 3) {
1138 			case bcfl_op:
1139 				likely = 1;
1140 			case bcf_op:
1141 				cond = !cond;
1142 				break;
1143 			case bctl_op:
1144 				likely = 1;
1145 			case bct_op:
1146 				break;
1147 			default:
1148 				/* thats an illegal instruction */
1149 				return SIGILL;
1150 			}
1151 
1152 			xcp->cp0_cause |= CAUSEF_BD;
1153 			if (cond) {
1154 				/* branch taken: emulate dslot
1155 				 * instruction
1156 				 */
1157 				xcp->cp0_epc += dec_insn.pc_inc;
1158 
1159 				contpc = MIPSInst_SIMM(ir);
1160 				ir = dec_insn.next_insn;
1161 				if (dec_insn.micro_mips_mode) {
1162 					contpc = (xcp->cp0_epc + (contpc << 1));
1163 
1164 					/* If 16-bit instruction, not FPU. */
1165 					if ((dec_insn.next_pc_inc == 2) ||
1166 						(microMIPS32_to_MIPS32((union mips_instruction *)&ir) == SIGILL)) {
1167 
1168 						/*
1169 						 * Since this instruction will
1170 						 * be put on the stack with
1171 						 * 32-bit words, get around
1172 						 * this problem by putting a
1173 						 * NOP16 as the second one.
1174 						 */
1175 						if (dec_insn.next_pc_inc == 2)
1176 							ir = (ir & (~0xffff)) | MM_NOP16;
1177 
1178 						/*
1179 						 * Single step the non-CP1
1180 						 * instruction in the dslot.
1181 						 */
1182 						return mips_dsemul(xcp, ir, contpc);
1183 					}
1184 				} else
1185 					contpc = (xcp->cp0_epc + (contpc << 2));
1186 
1187 				switch (MIPSInst_OPCODE(ir)) {
1188 				case lwc1_op:
1189 				case swc1_op:
1190 #if (__mips >= 2 || defined(__mips64))
1191 				case ldc1_op:
1192 				case sdc1_op:
1193 #endif
1194 				case cop1_op:
1195 #if __mips >= 4 && __mips != 32
1196 				case cop1x_op:
1197 #endif
1198 					/* its one of ours */
1199 					goto emul;
1200 #if __mips >= 4
1201 				case spec_op:
1202 					if (MIPSInst_FUNC(ir) == movc_op)
1203 						goto emul;
1204 					break;
1205 #endif
1206 				}
1207 
1208 				/*
1209 				 * Single step the non-cp1
1210 				 * instruction in the dslot
1211 				 */
1212 				return mips_dsemul(xcp, ir, contpc);
1213 			}
1214 			else {
1215 				/* branch not taken */
1216 				if (likely) {
1217 					/*
1218 					 * branch likely nullifies
1219 					 * dslot if not taken
1220 					 */
1221 					xcp->cp0_epc += dec_insn.pc_inc;
1222 					contpc += dec_insn.pc_inc;
1223 					/*
1224 					 * else continue & execute
1225 					 * dslot as normal insn
1226 					 */
1227 				}
1228 			}
1229 			break;
1230 		}
1231 
1232 		default:
1233 			if (!(MIPSInst_RS(ir) & 0x10))
1234 				return SIGILL;
1235 			{
1236 				int sig;
1237 
1238 				/* a real fpu computation instruction */
1239 				if ((sig = fpu_emu(xcp, ctx, ir)))
1240 					return sig;
1241 			}
1242 		}
1243 		break;
1244 
1245 #if __mips >= 4 && __mips != 32
1246 	case cop1x_op:{
1247 		int sig = fpux_emu(xcp, ctx, ir, fault_addr);
1248 		if (sig)
1249 			return sig;
1250 		break;
1251 	}
1252 #endif
1253 
1254 #if __mips >= 4
1255 	case spec_op:
1256 		if (MIPSInst_FUNC(ir) != movc_op)
1257 			return SIGILL;
1258 		cond = fpucondbit[MIPSInst_RT(ir) >> 2];
1259 		if (((ctx->fcr31 & cond) != 0) == ((MIPSInst_RT(ir) & 1) != 0))
1260 			xcp->regs[MIPSInst_RD(ir)] =
1261 				xcp->regs[MIPSInst_RS(ir)];
1262 		break;
1263 #endif
1264 
1265 	default:
1266 		return SIGILL;
1267 	}
1268 
1269 	/* we did it !! */
1270 	xcp->cp0_epc = contpc;
1271 	xcp->cp0_cause &= ~CAUSEF_BD;
1272 
1273 	return 0;
1274 }
1275 
1276 /*
1277  * Conversion table from MIPS compare ops 48-63
1278  * cond = ieee754dp_cmp(x,y,IEEE754_UN,sig);
1279  */
1280 static const unsigned char cmptab[8] = {
1281 	0,			/* cmp_0 (sig) cmp_sf */
1282 	IEEE754_CUN,		/* cmp_un (sig) cmp_ngle */
1283 	IEEE754_CEQ,		/* cmp_eq (sig) cmp_seq */
1284 	IEEE754_CEQ | IEEE754_CUN,	/* cmp_ueq (sig) cmp_ngl  */
1285 	IEEE754_CLT,		/* cmp_olt (sig) cmp_lt */
1286 	IEEE754_CLT | IEEE754_CUN,	/* cmp_ult (sig) cmp_nge */
1287 	IEEE754_CLT | IEEE754_CEQ,	/* cmp_ole (sig) cmp_le */
1288 	IEEE754_CLT | IEEE754_CEQ | IEEE754_CUN,	/* cmp_ule (sig) cmp_ngt */
1289 };
1290 
1291 
1292 #if __mips >= 4 && __mips != 32
1293 
1294 /*
1295  * Additional MIPS4 instructions
1296  */
1297 
1298 #define DEF3OP(name, p, f1, f2, f3) \
1299 static ieee754##p fpemu_##p##_##name(ieee754##p r, ieee754##p s, \
1300     ieee754##p t) \
1301 { \
1302 	struct _ieee754_csr ieee754_csr_save; \
1303 	s = f1(s, t); \
1304 	ieee754_csr_save = ieee754_csr; \
1305 	s = f2(s, r); \
1306 	ieee754_csr_save.cx |= ieee754_csr.cx; \
1307 	ieee754_csr_save.sx |= ieee754_csr.sx; \
1308 	s = f3(s); \
1309 	ieee754_csr.cx |= ieee754_csr_save.cx; \
1310 	ieee754_csr.sx |= ieee754_csr_save.sx; \
1311 	return s; \
1312 }
1313 
1314 static ieee754dp fpemu_dp_recip(ieee754dp d)
1315 {
1316 	return ieee754dp_div(ieee754dp_one(0), d);
1317 }
1318 
1319 static ieee754dp fpemu_dp_rsqrt(ieee754dp d)
1320 {
1321 	return ieee754dp_div(ieee754dp_one(0), ieee754dp_sqrt(d));
1322 }
1323 
1324 static ieee754sp fpemu_sp_recip(ieee754sp s)
1325 {
1326 	return ieee754sp_div(ieee754sp_one(0), s);
1327 }
1328 
1329 static ieee754sp fpemu_sp_rsqrt(ieee754sp s)
1330 {
1331 	return ieee754sp_div(ieee754sp_one(0), ieee754sp_sqrt(s));
1332 }
1333 
1334 DEF3OP(madd, sp, ieee754sp_mul, ieee754sp_add, );
1335 DEF3OP(msub, sp, ieee754sp_mul, ieee754sp_sub, );
1336 DEF3OP(nmadd, sp, ieee754sp_mul, ieee754sp_add, ieee754sp_neg);
1337 DEF3OP(nmsub, sp, ieee754sp_mul, ieee754sp_sub, ieee754sp_neg);
1338 DEF3OP(madd, dp, ieee754dp_mul, ieee754dp_add, );
1339 DEF3OP(msub, dp, ieee754dp_mul, ieee754dp_sub, );
1340 DEF3OP(nmadd, dp, ieee754dp_mul, ieee754dp_add, ieee754dp_neg);
1341 DEF3OP(nmsub, dp, ieee754dp_mul, ieee754dp_sub, ieee754dp_neg);
1342 
1343 static int fpux_emu(struct pt_regs *xcp, struct mips_fpu_struct *ctx,
1344 	mips_instruction ir, void *__user *fault_addr)
1345 {
1346 	unsigned rcsr = 0;	/* resulting csr */
1347 
1348 	MIPS_FPU_EMU_INC_STATS(cp1xops);
1349 
1350 	switch (MIPSInst_FMA_FFMT(ir)) {
1351 	case s_fmt:{		/* 0 */
1352 
1353 		ieee754sp(*handler) (ieee754sp, ieee754sp, ieee754sp);
1354 		ieee754sp fd, fr, fs, ft;
1355 		u32 __user *va;
1356 		u32 val;
1357 
1358 		switch (MIPSInst_FUNC(ir)) {
1359 		case lwxc1_op:
1360 			va = (void __user *) (xcp->regs[MIPSInst_FR(ir)] +
1361 				xcp->regs[MIPSInst_FT(ir)]);
1362 
1363 			MIPS_FPU_EMU_INC_STATS(loads);
1364 			if (!access_ok(VERIFY_READ, va, sizeof(u32))) {
1365 				MIPS_FPU_EMU_INC_STATS(errors);
1366 				*fault_addr = va;
1367 				return SIGBUS;
1368 			}
1369 			if (__get_user(val, va)) {
1370 				MIPS_FPU_EMU_INC_STATS(errors);
1371 				*fault_addr = va;
1372 				return SIGSEGV;
1373 			}
1374 			SITOREG(val, MIPSInst_FD(ir));
1375 			break;
1376 
1377 		case swxc1_op:
1378 			va = (void __user *) (xcp->regs[MIPSInst_FR(ir)] +
1379 				xcp->regs[MIPSInst_FT(ir)]);
1380 
1381 			MIPS_FPU_EMU_INC_STATS(stores);
1382 
1383 			SIFROMREG(val, MIPSInst_FS(ir));
1384 			if (!access_ok(VERIFY_WRITE, va, sizeof(u32))) {
1385 				MIPS_FPU_EMU_INC_STATS(errors);
1386 				*fault_addr = va;
1387 				return SIGBUS;
1388 			}
1389 			if (put_user(val, va)) {
1390 				MIPS_FPU_EMU_INC_STATS(errors);
1391 				*fault_addr = va;
1392 				return SIGSEGV;
1393 			}
1394 			break;
1395 
1396 		case madd_s_op:
1397 			handler = fpemu_sp_madd;
1398 			goto scoptop;
1399 		case msub_s_op:
1400 			handler = fpemu_sp_msub;
1401 			goto scoptop;
1402 		case nmadd_s_op:
1403 			handler = fpemu_sp_nmadd;
1404 			goto scoptop;
1405 		case nmsub_s_op:
1406 			handler = fpemu_sp_nmsub;
1407 			goto scoptop;
1408 
1409 		      scoptop:
1410 			SPFROMREG(fr, MIPSInst_FR(ir));
1411 			SPFROMREG(fs, MIPSInst_FS(ir));
1412 			SPFROMREG(ft, MIPSInst_FT(ir));
1413 			fd = (*handler) (fr, fs, ft);
1414 			SPTOREG(fd, MIPSInst_FD(ir));
1415 
1416 		      copcsr:
1417 			if (ieee754_cxtest(IEEE754_INEXACT))
1418 				rcsr |= FPU_CSR_INE_X | FPU_CSR_INE_S;
1419 			if (ieee754_cxtest(IEEE754_UNDERFLOW))
1420 				rcsr |= FPU_CSR_UDF_X | FPU_CSR_UDF_S;
1421 			if (ieee754_cxtest(IEEE754_OVERFLOW))
1422 				rcsr |= FPU_CSR_OVF_X | FPU_CSR_OVF_S;
1423 			if (ieee754_cxtest(IEEE754_INVALID_OPERATION))
1424 				rcsr |= FPU_CSR_INV_X | FPU_CSR_INV_S;
1425 
1426 			ctx->fcr31 = (ctx->fcr31 & ~FPU_CSR_ALL_X) | rcsr;
1427 			if ((ctx->fcr31 >> 5) & ctx->fcr31 & FPU_CSR_ALL_E) {
1428 				/*printk ("SIGFPE: fpu csr = %08x\n",
1429 				   ctx->fcr31); */
1430 				return SIGFPE;
1431 			}
1432 
1433 			break;
1434 
1435 		default:
1436 			return SIGILL;
1437 		}
1438 		break;
1439 	}
1440 
1441 	case d_fmt:{		/* 1 */
1442 		ieee754dp(*handler) (ieee754dp, ieee754dp, ieee754dp);
1443 		ieee754dp fd, fr, fs, ft;
1444 		u64 __user *va;
1445 		u64 val;
1446 
1447 		switch (MIPSInst_FUNC(ir)) {
1448 		case ldxc1_op:
1449 			va = (void __user *) (xcp->regs[MIPSInst_FR(ir)] +
1450 				xcp->regs[MIPSInst_FT(ir)]);
1451 
1452 			MIPS_FPU_EMU_INC_STATS(loads);
1453 			if (!access_ok(VERIFY_READ, va, sizeof(u64))) {
1454 				MIPS_FPU_EMU_INC_STATS(errors);
1455 				*fault_addr = va;
1456 				return SIGBUS;
1457 			}
1458 			if (__get_user(val, va)) {
1459 				MIPS_FPU_EMU_INC_STATS(errors);
1460 				*fault_addr = va;
1461 				return SIGSEGV;
1462 			}
1463 			DITOREG(val, MIPSInst_FD(ir));
1464 			break;
1465 
1466 		case sdxc1_op:
1467 			va = (void __user *) (xcp->regs[MIPSInst_FR(ir)] +
1468 				xcp->regs[MIPSInst_FT(ir)]);
1469 
1470 			MIPS_FPU_EMU_INC_STATS(stores);
1471 			DIFROMREG(val, MIPSInst_FS(ir));
1472 			if (!access_ok(VERIFY_WRITE, va, sizeof(u64))) {
1473 				MIPS_FPU_EMU_INC_STATS(errors);
1474 				*fault_addr = va;
1475 				return SIGBUS;
1476 			}
1477 			if (__put_user(val, va)) {
1478 				MIPS_FPU_EMU_INC_STATS(errors);
1479 				*fault_addr = va;
1480 				return SIGSEGV;
1481 			}
1482 			break;
1483 
1484 		case madd_d_op:
1485 			handler = fpemu_dp_madd;
1486 			goto dcoptop;
1487 		case msub_d_op:
1488 			handler = fpemu_dp_msub;
1489 			goto dcoptop;
1490 		case nmadd_d_op:
1491 			handler = fpemu_dp_nmadd;
1492 			goto dcoptop;
1493 		case nmsub_d_op:
1494 			handler = fpemu_dp_nmsub;
1495 			goto dcoptop;
1496 
1497 		      dcoptop:
1498 			DPFROMREG(fr, MIPSInst_FR(ir));
1499 			DPFROMREG(fs, MIPSInst_FS(ir));
1500 			DPFROMREG(ft, MIPSInst_FT(ir));
1501 			fd = (*handler) (fr, fs, ft);
1502 			DPTOREG(fd, MIPSInst_FD(ir));
1503 			goto copcsr;
1504 
1505 		default:
1506 			return SIGILL;
1507 		}
1508 		break;
1509 	}
1510 
1511 	case 0x7:		/* 7 */
1512 		if (MIPSInst_FUNC(ir) != pfetch_op) {
1513 			return SIGILL;
1514 		}
1515 		/* ignore prefx operation */
1516 		break;
1517 
1518 	default:
1519 		return SIGILL;
1520 	}
1521 
1522 	return 0;
1523 }
1524 #endif
1525 
1526 
1527 
1528 /*
1529  * Emulate a single COP1 arithmetic instruction.
1530  */
1531 static int fpu_emu(struct pt_regs *xcp, struct mips_fpu_struct *ctx,
1532 	mips_instruction ir)
1533 {
1534 	int rfmt;		/* resulting format */
1535 	unsigned rcsr = 0;	/* resulting csr */
1536 	unsigned cond;
1537 	union {
1538 		ieee754dp d;
1539 		ieee754sp s;
1540 		int w;
1541 #ifdef __mips64
1542 		s64 l;
1543 #endif
1544 	} rv;			/* resulting value */
1545 
1546 	MIPS_FPU_EMU_INC_STATS(cp1ops);
1547 	switch (rfmt = (MIPSInst_FFMT(ir) & 0xf)) {
1548 	case s_fmt:{		/* 0 */
1549 		union {
1550 			ieee754sp(*b) (ieee754sp, ieee754sp);
1551 			ieee754sp(*u) (ieee754sp);
1552 		} handler;
1553 
1554 		switch (MIPSInst_FUNC(ir)) {
1555 			/* binary ops */
1556 		case fadd_op:
1557 			handler.b = ieee754sp_add;
1558 			goto scopbop;
1559 		case fsub_op:
1560 			handler.b = ieee754sp_sub;
1561 			goto scopbop;
1562 		case fmul_op:
1563 			handler.b = ieee754sp_mul;
1564 			goto scopbop;
1565 		case fdiv_op:
1566 			handler.b = ieee754sp_div;
1567 			goto scopbop;
1568 
1569 			/* unary  ops */
1570 #if __mips >= 2 || defined(__mips64)
1571 		case fsqrt_op:
1572 			handler.u = ieee754sp_sqrt;
1573 			goto scopuop;
1574 #endif
1575 #if __mips >= 4 && __mips != 32
1576 		case frsqrt_op:
1577 			handler.u = fpemu_sp_rsqrt;
1578 			goto scopuop;
1579 		case frecip_op:
1580 			handler.u = fpemu_sp_recip;
1581 			goto scopuop;
1582 #endif
1583 #if __mips >= 4
1584 		case fmovc_op:
1585 			cond = fpucondbit[MIPSInst_FT(ir) >> 2];
1586 			if (((ctx->fcr31 & cond) != 0) !=
1587 				((MIPSInst_FT(ir) & 1) != 0))
1588 				return 0;
1589 			SPFROMREG(rv.s, MIPSInst_FS(ir));
1590 			break;
1591 		case fmovz_op:
1592 			if (xcp->regs[MIPSInst_FT(ir)] != 0)
1593 				return 0;
1594 			SPFROMREG(rv.s, MIPSInst_FS(ir));
1595 			break;
1596 		case fmovn_op:
1597 			if (xcp->regs[MIPSInst_FT(ir)] == 0)
1598 				return 0;
1599 			SPFROMREG(rv.s, MIPSInst_FS(ir));
1600 			break;
1601 #endif
1602 		case fabs_op:
1603 			handler.u = ieee754sp_abs;
1604 			goto scopuop;
1605 		case fneg_op:
1606 			handler.u = ieee754sp_neg;
1607 			goto scopuop;
1608 		case fmov_op:
1609 			/* an easy one */
1610 			SPFROMREG(rv.s, MIPSInst_FS(ir));
1611 			goto copcsr;
1612 
1613 			/* binary op on handler */
1614 		      scopbop:
1615 			{
1616 				ieee754sp fs, ft;
1617 
1618 				SPFROMREG(fs, MIPSInst_FS(ir));
1619 				SPFROMREG(ft, MIPSInst_FT(ir));
1620 
1621 				rv.s = (*handler.b) (fs, ft);
1622 				goto copcsr;
1623 			}
1624 		      scopuop:
1625 			{
1626 				ieee754sp fs;
1627 
1628 				SPFROMREG(fs, MIPSInst_FS(ir));
1629 				rv.s = (*handler.u) (fs);
1630 				goto copcsr;
1631 			}
1632 		      copcsr:
1633 			if (ieee754_cxtest(IEEE754_INEXACT))
1634 				rcsr |= FPU_CSR_INE_X | FPU_CSR_INE_S;
1635 			if (ieee754_cxtest(IEEE754_UNDERFLOW))
1636 				rcsr |= FPU_CSR_UDF_X | FPU_CSR_UDF_S;
1637 			if (ieee754_cxtest(IEEE754_OVERFLOW))
1638 				rcsr |= FPU_CSR_OVF_X | FPU_CSR_OVF_S;
1639 			if (ieee754_cxtest(IEEE754_ZERO_DIVIDE))
1640 				rcsr |= FPU_CSR_DIV_X | FPU_CSR_DIV_S;
1641 			if (ieee754_cxtest(IEEE754_INVALID_OPERATION))
1642 				rcsr |= FPU_CSR_INV_X | FPU_CSR_INV_S;
1643 			break;
1644 
1645 			/* unary conv ops */
1646 		case fcvts_op:
1647 			return SIGILL;	/* not defined */
1648 		case fcvtd_op:{
1649 			ieee754sp fs;
1650 
1651 			SPFROMREG(fs, MIPSInst_FS(ir));
1652 			rv.d = ieee754dp_fsp(fs);
1653 			rfmt = d_fmt;
1654 			goto copcsr;
1655 		}
1656 		case fcvtw_op:{
1657 			ieee754sp fs;
1658 
1659 			SPFROMREG(fs, MIPSInst_FS(ir));
1660 			rv.w = ieee754sp_tint(fs);
1661 			rfmt = w_fmt;
1662 			goto copcsr;
1663 		}
1664 
1665 #if __mips >= 2 || defined(__mips64)
1666 		case fround_op:
1667 		case ftrunc_op:
1668 		case fceil_op:
1669 		case ffloor_op:{
1670 			unsigned int oldrm = ieee754_csr.rm;
1671 			ieee754sp fs;
1672 
1673 			SPFROMREG(fs, MIPSInst_FS(ir));
1674 			ieee754_csr.rm = ieee_rm[modeindex(MIPSInst_FUNC(ir))];
1675 			rv.w = ieee754sp_tint(fs);
1676 			ieee754_csr.rm = oldrm;
1677 			rfmt = w_fmt;
1678 			goto copcsr;
1679 		}
1680 #endif /* __mips >= 2 */
1681 
1682 #if defined(__mips64)
1683 		case fcvtl_op:{
1684 			ieee754sp fs;
1685 
1686 			SPFROMREG(fs, MIPSInst_FS(ir));
1687 			rv.l = ieee754sp_tlong(fs);
1688 			rfmt = l_fmt;
1689 			goto copcsr;
1690 		}
1691 
1692 		case froundl_op:
1693 		case ftruncl_op:
1694 		case fceill_op:
1695 		case ffloorl_op:{
1696 			unsigned int oldrm = ieee754_csr.rm;
1697 			ieee754sp fs;
1698 
1699 			SPFROMREG(fs, MIPSInst_FS(ir));
1700 			ieee754_csr.rm = ieee_rm[modeindex(MIPSInst_FUNC(ir))];
1701 			rv.l = ieee754sp_tlong(fs);
1702 			ieee754_csr.rm = oldrm;
1703 			rfmt = l_fmt;
1704 			goto copcsr;
1705 		}
1706 #endif /* defined(__mips64) */
1707 
1708 		default:
1709 			if (MIPSInst_FUNC(ir) >= fcmp_op) {
1710 				unsigned cmpop = MIPSInst_FUNC(ir) - fcmp_op;
1711 				ieee754sp fs, ft;
1712 
1713 				SPFROMREG(fs, MIPSInst_FS(ir));
1714 				SPFROMREG(ft, MIPSInst_FT(ir));
1715 				rv.w = ieee754sp_cmp(fs, ft,
1716 					cmptab[cmpop & 0x7], cmpop & 0x8);
1717 				rfmt = -1;
1718 				if ((cmpop & 0x8) && ieee754_cxtest
1719 					(IEEE754_INVALID_OPERATION))
1720 					rcsr = FPU_CSR_INV_X | FPU_CSR_INV_S;
1721 				else
1722 					goto copcsr;
1723 
1724 			}
1725 			else {
1726 				return SIGILL;
1727 			}
1728 			break;
1729 		}
1730 		break;
1731 	}
1732 
1733 	case d_fmt:{
1734 		union {
1735 			ieee754dp(*b) (ieee754dp, ieee754dp);
1736 			ieee754dp(*u) (ieee754dp);
1737 		} handler;
1738 
1739 		switch (MIPSInst_FUNC(ir)) {
1740 			/* binary ops */
1741 		case fadd_op:
1742 			handler.b = ieee754dp_add;
1743 			goto dcopbop;
1744 		case fsub_op:
1745 			handler.b = ieee754dp_sub;
1746 			goto dcopbop;
1747 		case fmul_op:
1748 			handler.b = ieee754dp_mul;
1749 			goto dcopbop;
1750 		case fdiv_op:
1751 			handler.b = ieee754dp_div;
1752 			goto dcopbop;
1753 
1754 			/* unary  ops */
1755 #if __mips >= 2 || defined(__mips64)
1756 		case fsqrt_op:
1757 			handler.u = ieee754dp_sqrt;
1758 			goto dcopuop;
1759 #endif
1760 #if __mips >= 4 && __mips != 32
1761 		case frsqrt_op:
1762 			handler.u = fpemu_dp_rsqrt;
1763 			goto dcopuop;
1764 		case frecip_op:
1765 			handler.u = fpemu_dp_recip;
1766 			goto dcopuop;
1767 #endif
1768 #if __mips >= 4
1769 		case fmovc_op:
1770 			cond = fpucondbit[MIPSInst_FT(ir) >> 2];
1771 			if (((ctx->fcr31 & cond) != 0) !=
1772 				((MIPSInst_FT(ir) & 1) != 0))
1773 				return 0;
1774 			DPFROMREG(rv.d, MIPSInst_FS(ir));
1775 			break;
1776 		case fmovz_op:
1777 			if (xcp->regs[MIPSInst_FT(ir)] != 0)
1778 				return 0;
1779 			DPFROMREG(rv.d, MIPSInst_FS(ir));
1780 			break;
1781 		case fmovn_op:
1782 			if (xcp->regs[MIPSInst_FT(ir)] == 0)
1783 				return 0;
1784 			DPFROMREG(rv.d, MIPSInst_FS(ir));
1785 			break;
1786 #endif
1787 		case fabs_op:
1788 			handler.u = ieee754dp_abs;
1789 			goto dcopuop;
1790 
1791 		case fneg_op:
1792 			handler.u = ieee754dp_neg;
1793 			goto dcopuop;
1794 
1795 		case fmov_op:
1796 			/* an easy one */
1797 			DPFROMREG(rv.d, MIPSInst_FS(ir));
1798 			goto copcsr;
1799 
1800 			/* binary op on handler */
1801 		      dcopbop:{
1802 				ieee754dp fs, ft;
1803 
1804 				DPFROMREG(fs, MIPSInst_FS(ir));
1805 				DPFROMREG(ft, MIPSInst_FT(ir));
1806 
1807 				rv.d = (*handler.b) (fs, ft);
1808 				goto copcsr;
1809 			}
1810 		      dcopuop:{
1811 				ieee754dp fs;
1812 
1813 				DPFROMREG(fs, MIPSInst_FS(ir));
1814 				rv.d = (*handler.u) (fs);
1815 				goto copcsr;
1816 			}
1817 
1818 			/* unary conv ops */
1819 		case fcvts_op:{
1820 			ieee754dp fs;
1821 
1822 			DPFROMREG(fs, MIPSInst_FS(ir));
1823 			rv.s = ieee754sp_fdp(fs);
1824 			rfmt = s_fmt;
1825 			goto copcsr;
1826 		}
1827 		case fcvtd_op:
1828 			return SIGILL;	/* not defined */
1829 
1830 		case fcvtw_op:{
1831 			ieee754dp fs;
1832 
1833 			DPFROMREG(fs, MIPSInst_FS(ir));
1834 			rv.w = ieee754dp_tint(fs);	/* wrong */
1835 			rfmt = w_fmt;
1836 			goto copcsr;
1837 		}
1838 
1839 #if __mips >= 2 || defined(__mips64)
1840 		case fround_op:
1841 		case ftrunc_op:
1842 		case fceil_op:
1843 		case ffloor_op:{
1844 			unsigned int oldrm = ieee754_csr.rm;
1845 			ieee754dp fs;
1846 
1847 			DPFROMREG(fs, MIPSInst_FS(ir));
1848 			ieee754_csr.rm = ieee_rm[modeindex(MIPSInst_FUNC(ir))];
1849 			rv.w = ieee754dp_tint(fs);
1850 			ieee754_csr.rm = oldrm;
1851 			rfmt = w_fmt;
1852 			goto copcsr;
1853 		}
1854 #endif
1855 
1856 #if defined(__mips64)
1857 		case fcvtl_op:{
1858 			ieee754dp fs;
1859 
1860 			DPFROMREG(fs, MIPSInst_FS(ir));
1861 			rv.l = ieee754dp_tlong(fs);
1862 			rfmt = l_fmt;
1863 			goto copcsr;
1864 		}
1865 
1866 		case froundl_op:
1867 		case ftruncl_op:
1868 		case fceill_op:
1869 		case ffloorl_op:{
1870 			unsigned int oldrm = ieee754_csr.rm;
1871 			ieee754dp fs;
1872 
1873 			DPFROMREG(fs, MIPSInst_FS(ir));
1874 			ieee754_csr.rm = ieee_rm[modeindex(MIPSInst_FUNC(ir))];
1875 			rv.l = ieee754dp_tlong(fs);
1876 			ieee754_csr.rm = oldrm;
1877 			rfmt = l_fmt;
1878 			goto copcsr;
1879 		}
1880 #endif /* __mips >= 3 */
1881 
1882 		default:
1883 			if (MIPSInst_FUNC(ir) >= fcmp_op) {
1884 				unsigned cmpop = MIPSInst_FUNC(ir) - fcmp_op;
1885 				ieee754dp fs, ft;
1886 
1887 				DPFROMREG(fs, MIPSInst_FS(ir));
1888 				DPFROMREG(ft, MIPSInst_FT(ir));
1889 				rv.w = ieee754dp_cmp(fs, ft,
1890 					cmptab[cmpop & 0x7], cmpop & 0x8);
1891 				rfmt = -1;
1892 				if ((cmpop & 0x8)
1893 					&&
1894 					ieee754_cxtest
1895 					(IEEE754_INVALID_OPERATION))
1896 					rcsr = FPU_CSR_INV_X | FPU_CSR_INV_S;
1897 				else
1898 					goto copcsr;
1899 
1900 			}
1901 			else {
1902 				return SIGILL;
1903 			}
1904 			break;
1905 		}
1906 		break;
1907 	}
1908 
1909 	case w_fmt:{
1910 		ieee754sp fs;
1911 
1912 		switch (MIPSInst_FUNC(ir)) {
1913 		case fcvts_op:
1914 			/* convert word to single precision real */
1915 			SPFROMREG(fs, MIPSInst_FS(ir));
1916 			rv.s = ieee754sp_fint(fs.bits);
1917 			rfmt = s_fmt;
1918 			goto copcsr;
1919 		case fcvtd_op:
1920 			/* convert word to double precision real */
1921 			SPFROMREG(fs, MIPSInst_FS(ir));
1922 			rv.d = ieee754dp_fint(fs.bits);
1923 			rfmt = d_fmt;
1924 			goto copcsr;
1925 		default:
1926 			return SIGILL;
1927 		}
1928 		break;
1929 	}
1930 
1931 #if defined(__mips64)
1932 	case l_fmt:{
1933 		switch (MIPSInst_FUNC(ir)) {
1934 		case fcvts_op:
1935 			/* convert long to single precision real */
1936 			rv.s = ieee754sp_flong(ctx->fpr[MIPSInst_FS(ir)]);
1937 			rfmt = s_fmt;
1938 			goto copcsr;
1939 		case fcvtd_op:
1940 			/* convert long to double precision real */
1941 			rv.d = ieee754dp_flong(ctx->fpr[MIPSInst_FS(ir)]);
1942 			rfmt = d_fmt;
1943 			goto copcsr;
1944 		default:
1945 			return SIGILL;
1946 		}
1947 		break;
1948 	}
1949 #endif
1950 
1951 	default:
1952 		return SIGILL;
1953 	}
1954 
1955 	/*
1956 	 * Update the fpu CSR register for this operation.
1957 	 * If an exception is required, generate a tidy SIGFPE exception,
1958 	 * without updating the result register.
1959 	 * Note: cause exception bits do not accumulate, they are rewritten
1960 	 * for each op; only the flag/sticky bits accumulate.
1961 	 */
1962 	ctx->fcr31 = (ctx->fcr31 & ~FPU_CSR_ALL_X) | rcsr;
1963 	if ((ctx->fcr31 >> 5) & ctx->fcr31 & FPU_CSR_ALL_E) {
1964 		/*printk ("SIGFPE: fpu csr = %08x\n",ctx->fcr31); */
1965 		return SIGFPE;
1966 	}
1967 
1968 	/*
1969 	 * Now we can safely write the result back to the register file.
1970 	 */
1971 	switch (rfmt) {
1972 	case -1:{
1973 #if __mips >= 4
1974 		cond = fpucondbit[MIPSInst_FD(ir) >> 2];
1975 #else
1976 		cond = FPU_CSR_COND;
1977 #endif
1978 		if (rv.w)
1979 			ctx->fcr31 |= cond;
1980 		else
1981 			ctx->fcr31 &= ~cond;
1982 		break;
1983 	}
1984 	case d_fmt:
1985 		DPTOREG(rv.d, MIPSInst_FD(ir));
1986 		break;
1987 	case s_fmt:
1988 		SPTOREG(rv.s, MIPSInst_FD(ir));
1989 		break;
1990 	case w_fmt:
1991 		SITOREG(rv.w, MIPSInst_FD(ir));
1992 		break;
1993 #if defined(__mips64)
1994 	case l_fmt:
1995 		DITOREG(rv.l, MIPSInst_FD(ir));
1996 		break;
1997 #endif
1998 	default:
1999 		return SIGILL;
2000 	}
2001 
2002 	return 0;
2003 }
2004 
2005 int fpu_emulator_cop1Handler(struct pt_regs *xcp, struct mips_fpu_struct *ctx,
2006 	int has_fpu, void *__user *fault_addr)
2007 {
2008 	unsigned long oldepc, prevepc;
2009 	struct mm_decoded_insn dec_insn;
2010 	u16 instr[4];
2011 	u16 *instr_ptr;
2012 	int sig = 0;
2013 
2014 	oldepc = xcp->cp0_epc;
2015 	do {
2016 		prevepc = xcp->cp0_epc;
2017 
2018 		if (get_isa16_mode(prevepc) && cpu_has_mmips) {
2019 			/*
2020 			 * Get next 2 microMIPS instructions and convert them
2021 			 * into 32-bit instructions.
2022 			 */
2023 			if ((get_user(instr[0], (u16 __user *)msk_isa16_mode(xcp->cp0_epc))) ||
2024 			    (get_user(instr[1], (u16 __user *)msk_isa16_mode(xcp->cp0_epc + 2))) ||
2025 			    (get_user(instr[2], (u16 __user *)msk_isa16_mode(xcp->cp0_epc + 4))) ||
2026 			    (get_user(instr[3], (u16 __user *)msk_isa16_mode(xcp->cp0_epc + 6)))) {
2027 				MIPS_FPU_EMU_INC_STATS(errors);
2028 				return SIGBUS;
2029 			}
2030 			instr_ptr = instr;
2031 
2032 			/* Get first instruction. */
2033 			if (mm_insn_16bit(*instr_ptr)) {
2034 				/* Duplicate the half-word. */
2035 				dec_insn.insn = (*instr_ptr << 16) |
2036 					(*instr_ptr);
2037 				/* 16-bit instruction. */
2038 				dec_insn.pc_inc = 2;
2039 				instr_ptr += 1;
2040 			} else {
2041 				dec_insn.insn = (*instr_ptr << 16) |
2042 					*(instr_ptr+1);
2043 				/* 32-bit instruction. */
2044 				dec_insn.pc_inc = 4;
2045 				instr_ptr += 2;
2046 			}
2047 			/* Get second instruction. */
2048 			if (mm_insn_16bit(*instr_ptr)) {
2049 				/* Duplicate the half-word. */
2050 				dec_insn.next_insn = (*instr_ptr << 16) |
2051 					(*instr_ptr);
2052 				/* 16-bit instruction. */
2053 				dec_insn.next_pc_inc = 2;
2054 			} else {
2055 				dec_insn.next_insn = (*instr_ptr << 16) |
2056 					*(instr_ptr+1);
2057 				/* 32-bit instruction. */
2058 				dec_insn.next_pc_inc = 4;
2059 			}
2060 			dec_insn.micro_mips_mode = 1;
2061 		} else {
2062 			if ((get_user(dec_insn.insn,
2063 			    (mips_instruction __user *) xcp->cp0_epc)) ||
2064 			    (get_user(dec_insn.next_insn,
2065 			    (mips_instruction __user *)(xcp->cp0_epc+4)))) {
2066 				MIPS_FPU_EMU_INC_STATS(errors);
2067 				return SIGBUS;
2068 			}
2069 			dec_insn.pc_inc = 4;
2070 			dec_insn.next_pc_inc = 4;
2071 			dec_insn.micro_mips_mode = 0;
2072 		}
2073 
2074 		if ((dec_insn.insn == 0) ||
2075 		   ((dec_insn.pc_inc == 2) &&
2076 		   ((dec_insn.insn & 0xffff) == MM_NOP16)))
2077 			xcp->cp0_epc += dec_insn.pc_inc;	/* Skip NOPs */
2078 		else {
2079 			/*
2080 			 * The 'ieee754_csr' is an alias of
2081 			 * ctx->fcr31.	No need to copy ctx->fcr31 to
2082 			 * ieee754_csr.	 But ieee754_csr.rm is ieee
2083 			 * library modes. (not mips rounding mode)
2084 			 */
2085 			/* convert to ieee library modes */
2086 			ieee754_csr.rm = ieee_rm[ieee754_csr.rm];
2087 			sig = cop1Emulate(xcp, ctx, dec_insn, fault_addr);
2088 			/* revert to mips rounding mode */
2089 			ieee754_csr.rm = mips_rm[ieee754_csr.rm];
2090 		}
2091 
2092 		if (has_fpu)
2093 			break;
2094 		if (sig)
2095 			break;
2096 
2097 		cond_resched();
2098 	} while (xcp->cp0_epc > prevepc);
2099 
2100 	/* SIGILL indicates a non-fpu instruction */
2101 	if (sig == SIGILL && xcp->cp0_epc != oldepc)
2102 		/* but if epc has advanced, then ignore it */
2103 		sig = 0;
2104 
2105 	return sig;
2106 }
2107 
2108 #ifdef CONFIG_DEBUG_FS
2109 
2110 static int fpuemu_stat_get(void *data, u64 *val)
2111 {
2112 	int cpu;
2113 	unsigned long sum = 0;
2114 	for_each_online_cpu(cpu) {
2115 		struct mips_fpu_emulator_stats *ps;
2116 		local_t *pv;
2117 		ps = &per_cpu(fpuemustats, cpu);
2118 		pv = (void *)ps + (unsigned long)data;
2119 		sum += local_read(pv);
2120 	}
2121 	*val = sum;
2122 	return 0;
2123 }
2124 DEFINE_SIMPLE_ATTRIBUTE(fops_fpuemu_stat, fpuemu_stat_get, NULL, "%llu\n");
2125 
2126 extern struct dentry *mips_debugfs_dir;
2127 static int __init debugfs_fpuemu(void)
2128 {
2129 	struct dentry *d, *dir;
2130 
2131 	if (!mips_debugfs_dir)
2132 		return -ENODEV;
2133 	dir = debugfs_create_dir("fpuemustats", mips_debugfs_dir);
2134 	if (!dir)
2135 		return -ENOMEM;
2136 
2137 #define FPU_STAT_CREATE(M)						\
2138 	do {								\
2139 		d = debugfs_create_file(#M , S_IRUGO, dir,		\
2140 			(void *)offsetof(struct mips_fpu_emulator_stats, M), \
2141 			&fops_fpuemu_stat);				\
2142 		if (!d)							\
2143 			return -ENOMEM;					\
2144 	} while (0)
2145 
2146 	FPU_STAT_CREATE(emulated);
2147 	FPU_STAT_CREATE(loads);
2148 	FPU_STAT_CREATE(stores);
2149 	FPU_STAT_CREATE(cp1ops);
2150 	FPU_STAT_CREATE(cp1xops);
2151 	FPU_STAT_CREATE(errors);
2152 
2153 	return 0;
2154 }
2155 __initcall(debugfs_fpuemu);
2156 #endif
2157