?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 3.15-rc1
[linux/fpc-iii.git] / arch / mips / math-emu / cp1emu.c
blob7b3c9acae6895120e751dc4efe19b7ee8a6572ea
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 case mm_mfhc1_op:
421 case mm_mthc1_op:
422 if (insn.mm_fp1_format.op == mm_mfc1_op)
423 op = mfc_op;
424 else if (insn.mm_fp1_format.op == mm_mtc1_op)
425 op = mtc_op;
426 else if (insn.mm_fp1_format.op == mm_cfc1_op)
427 op = cfc_op;
428 else if (insn.mm_fp1_format.op == mm_ctc1_op)
429 op = ctc_op;
430 else if (insn.mm_fp1_format.op == mm_mfhc1_op)
431 op = mfhc_op;
432 else
433 op = mthc_op;
434 mips32_insn.fp1_format.opcode = cop1_op;
435 mips32_insn.fp1_format.op = op;
436 mips32_insn.fp1_format.rt =
437 insn.mm_fp1_format.rt;
438 mips32_insn.fp1_format.fs =
439 insn.mm_fp1_format.fs;
440 mips32_insn.fp1_format.fd = 0;
441 mips32_insn.fp1_format.func = 0;
442 break;
443 default:
444 return SIGILL;
445 }
446 break;
447 case mm_32f_74_op: /* c.cond.fmt */
448 mips32_insn.fp0_format.opcode = cop1_op;
449 mips32_insn.fp0_format.fmt =
450 sdps_format[insn.mm_fp4_format.fmt];
451 mips32_insn.fp0_format.ft = insn.mm_fp4_format.rt;
452 mips32_insn.fp0_format.fs = insn.mm_fp4_format.fs;
453 mips32_insn.fp0_format.fd = insn.mm_fp4_format.cc << 2;
454 mips32_insn.fp0_format.func =
455 insn.mm_fp4_format.cond | MM_MIPS32_COND_FC;
456 break;
457 default:
458 return SIGILL;
459 }
460 break;
461 default:
462 return SIGILL;
463 }
464
465 *insn_ptr = mips32_insn;
466 return 0;
467 }
468
469 int mm_isBranchInstr(struct pt_regs *regs, struct mm_decoded_insn dec_insn,
470 unsigned long *contpc)
471 {
472 union mips_instruction insn = (union mips_instruction)dec_insn.insn;
473 int bc_false = 0;
474 unsigned int fcr31;
475 unsigned int bit;
476
477 if (!cpu_has_mmips)
478 return 0;
479
480 switch (insn.mm_i_format.opcode) {
481 case mm_pool32a_op:
482 if ((insn.mm_i_format.simmediate & MM_POOL32A_MINOR_MASK) ==
483 mm_pool32axf_op) {
484 switch (insn.mm_i_format.simmediate >>
485 MM_POOL32A_MINOR_SHIFT) {
486 case mm_jalr_op:
487 case mm_jalrhb_op:
488 case mm_jalrs_op:
489 case mm_jalrshb_op:
490 if (insn.mm_i_format.rt != 0) /* Not mm_jr */
491 regs->regs[insn.mm_i_format.rt] =
492 regs->cp0_epc +
493 dec_insn.pc_inc +
494 dec_insn.next_pc_inc;
495 *contpc = regs->regs[insn.mm_i_format.rs];
496 return 1;
497 }
498 }
499 break;
500 case mm_pool32i_op:
501 switch (insn.mm_i_format.rt) {
502 case mm_bltzals_op:
503 case mm_bltzal_op:
504 regs->regs[31] = regs->cp0_epc +
505 dec_insn.pc_inc +
506 dec_insn.next_pc_inc;
507 /* Fall through */
508 case mm_bltz_op:
509 if ((long)regs->regs[insn.mm_i_format.rs] < 0)
510 *contpc = regs->cp0_epc +
511 dec_insn.pc_inc +
512 (insn.mm_i_format.simmediate << 1);
513 else
514 *contpc = regs->cp0_epc +
515 dec_insn.pc_inc +
516 dec_insn.next_pc_inc;
517 return 1;
518 case mm_bgezals_op:
519 case mm_bgezal_op:
520 regs->regs[31] = regs->cp0_epc +
521 dec_insn.pc_inc +
522 dec_insn.next_pc_inc;
523 /* Fall through */
524 case mm_bgez_op:
525 if ((long)regs->regs[insn.mm_i_format.rs] >= 0)
526 *contpc = regs->cp0_epc +
527 dec_insn.pc_inc +
528 (insn.mm_i_format.simmediate << 1);
529 else
530 *contpc = regs->cp0_epc +
531 dec_insn.pc_inc +
532 dec_insn.next_pc_inc;
533 return 1;
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 case mm_bgtz_op:
545 if ((long)regs->regs[insn.mm_i_format.rs] <= 0)
546 *contpc = regs->cp0_epc +
547 dec_insn.pc_inc +
548 (insn.mm_i_format.simmediate << 1);
549 else
550 *contpc = regs->cp0_epc +
551 dec_insn.pc_inc +
552 dec_insn.next_pc_inc;
553 return 1;
554 case mm_bc2f_op:
555 case mm_bc1f_op:
556 bc_false = 1;
557 /* Fall through */
558 case mm_bc2t_op:
559 case mm_bc1t_op:
560 preempt_disable();
561 if (is_fpu_owner())
562 asm volatile("cfc1\t%0,$31" : "=r" (fcr31));
563 else
564 fcr31 = current->thread.fpu.fcr31;
565 preempt_enable();
566
567 if (bc_false)
568 fcr31 = ~fcr31;
569
570 bit = (insn.mm_i_format.rs >> 2);
571 bit += (bit != 0);
572 bit += 23;
573 if (fcr31 & (1 << bit))
574 *contpc = regs->cp0_epc +
575 dec_insn.pc_inc +
576 (insn.mm_i_format.simmediate << 1);
577 else
578 *contpc = regs->cp0_epc +
579 dec_insn.pc_inc + dec_insn.next_pc_inc;
580 return 1;
581 }
582 break;
583 case mm_pool16c_op:
584 switch (insn.mm_i_format.rt) {
585 case mm_jalr16_op:
586 case mm_jalrs16_op:
587 regs->regs[31] = regs->cp0_epc +
588 dec_insn.pc_inc + dec_insn.next_pc_inc;
589 /* Fall through */
590 case mm_jr16_op:
591 *contpc = regs->regs[insn.mm_i_format.rs];
592 return 1;
593 }
594 break;
595 case mm_beqz16_op:
596 if ((long)regs->regs[reg16to32map[insn.mm_b1_format.rs]] == 0)
597 *contpc = regs->cp0_epc +
598 dec_insn.pc_inc +
599 (insn.mm_b1_format.simmediate << 1);
600 else
601 *contpc = regs->cp0_epc +
602 dec_insn.pc_inc + dec_insn.next_pc_inc;
603 return 1;
604 case mm_bnez16_op:
605 if ((long)regs->regs[reg16to32map[insn.mm_b1_format.rs]] != 0)
606 *contpc = regs->cp0_epc +
607 dec_insn.pc_inc +
608 (insn.mm_b1_format.simmediate << 1);
609 else
610 *contpc = regs->cp0_epc +
611 dec_insn.pc_inc + dec_insn.next_pc_inc;
612 return 1;
613 case mm_b16_op:
614 *contpc = regs->cp0_epc + dec_insn.pc_inc +
615 (insn.mm_b0_format.simmediate << 1);
616 return 1;
617 case mm_beq32_op:
618 if (regs->regs[insn.mm_i_format.rs] ==
619 regs->regs[insn.mm_i_format.rt])
620 *contpc = regs->cp0_epc +
621 dec_insn.pc_inc +
622 (insn.mm_i_format.simmediate << 1);
623 else
624 *contpc = regs->cp0_epc +
625 dec_insn.pc_inc +
626 dec_insn.next_pc_inc;
627 return 1;
628 case mm_bne32_op:
629 if (regs->regs[insn.mm_i_format.rs] !=
630 regs->regs[insn.mm_i_format.rt])
631 *contpc = regs->cp0_epc +
632 dec_insn.pc_inc +
633 (insn.mm_i_format.simmediate << 1);
634 else
635 *contpc = regs->cp0_epc +
636 dec_insn.pc_inc + dec_insn.next_pc_inc;
637 return 1;
638 case mm_jalx32_op:
639 regs->regs[31] = regs->cp0_epc +
640 dec_insn.pc_inc + dec_insn.next_pc_inc;
641 *contpc = regs->cp0_epc + dec_insn.pc_inc;
642 *contpc >>= 28;
643 *contpc <<= 28;
644 *contpc |= (insn.j_format.target << 2);
645 return 1;
646 case mm_jals32_op:
647 case mm_jal32_op:
648 regs->regs[31] = regs->cp0_epc +
649 dec_insn.pc_inc + dec_insn.next_pc_inc;
650 /* Fall through */
651 case mm_j32_op:
652 *contpc = regs->cp0_epc + dec_insn.pc_inc;
653 *contpc >>= 27;
654 *contpc <<= 27;
655 *contpc |= (insn.j_format.target << 1);
656 set_isa16_mode(*contpc);
657 return 1;
658 }
659 return 0;
660 }
661
662 /*
663 * Redundant with logic already in kernel/branch.c,
664 * embedded in compute_return_epc. At some point,
665 * a single subroutine should be used across both
666 * modules.
667 */
668 static int isBranchInstr(struct pt_regs *regs, struct mm_decoded_insn dec_insn,
669 unsigned long *contpc)
670 {
671 union mips_instruction insn = (union mips_instruction)dec_insn.insn;
672 unsigned int fcr31;
673 unsigned int bit = 0;
674
675 switch (insn.i_format.opcode) {
676 case spec_op:
677 switch (insn.r_format.func) {
678 case jalr_op:
679 regs->regs[insn.r_format.rd] =
680 regs->cp0_epc + dec_insn.pc_inc +
681 dec_insn.next_pc_inc;
682 /* Fall through */
683 case jr_op:
684 *contpc = regs->regs[insn.r_format.rs];
685 return 1;
686 }
687 break;
688 case bcond_op:
689 switch (insn.i_format.rt) {
690 case bltzal_op:
691 case bltzall_op:
692 regs->regs[31] = regs->cp0_epc +
693 dec_insn.pc_inc +
694 dec_insn.next_pc_inc;
695 /* Fall through */
696 case bltz_op:
697 case bltzl_op:
698 if ((long)regs->regs[insn.i_format.rs] < 0)
699 *contpc = regs->cp0_epc +
700 dec_insn.pc_inc +
701 (insn.i_format.simmediate << 2);
702 else
703 *contpc = regs->cp0_epc +
704 dec_insn.pc_inc +
705 dec_insn.next_pc_inc;
706 return 1;
707 case bgezal_op:
708 case bgezall_op:
709 regs->regs[31] = regs->cp0_epc +
710 dec_insn.pc_inc +
711 dec_insn.next_pc_inc;
712 /* Fall through */
713 case bgez_op:
714 case bgezl_op:
715 if ((long)regs->regs[insn.i_format.rs] >= 0)
716 *contpc = regs->cp0_epc +
717 dec_insn.pc_inc +
718 (insn.i_format.simmediate << 2);
719 else
720 *contpc = regs->cp0_epc +
721 dec_insn.pc_inc +
722 dec_insn.next_pc_inc;
723 return 1;
724 }
725 break;
726 case jalx_op:
727 set_isa16_mode(bit);
728 case jal_op:
729 regs->regs[31] = regs->cp0_epc +
730 dec_insn.pc_inc +
731 dec_insn.next_pc_inc;
732 /* Fall through */
733 case j_op:
734 *contpc = regs->cp0_epc + dec_insn.pc_inc;
735 *contpc >>= 28;
736 *contpc <<= 28;
737 *contpc |= (insn.j_format.target << 2);
738 /* Set microMIPS mode bit: XOR for jalx. */
739 *contpc ^= bit;
740 return 1;
741 case beq_op:
742 case beql_op:
743 if (regs->regs[insn.i_format.rs] ==
744 regs->regs[insn.i_format.rt])
745 *contpc = regs->cp0_epc +
746 dec_insn.pc_inc +
747 (insn.i_format.simmediate << 2);
748 else
749 *contpc = regs->cp0_epc +
750 dec_insn.pc_inc +
751 dec_insn.next_pc_inc;
752 return 1;
753 case bne_op:
754 case bnel_op:
755 if (regs->regs[insn.i_format.rs] !=
756 regs->regs[insn.i_format.rt])
757 *contpc = regs->cp0_epc +
758 dec_insn.pc_inc +
759 (insn.i_format.simmediate << 2);
760 else
761 *contpc = regs->cp0_epc +
762 dec_insn.pc_inc +
763 dec_insn.next_pc_inc;
764 return 1;
765 case blez_op:
766 case blezl_op:
767 if ((long)regs->regs[insn.i_format.rs] <= 0)
768 *contpc = regs->cp0_epc +
769 dec_insn.pc_inc +
770 (insn.i_format.simmediate << 2);
771 else
772 *contpc = regs->cp0_epc +
773 dec_insn.pc_inc +
774 dec_insn.next_pc_inc;
775 return 1;
776 case bgtz_op:
777 case bgtzl_op:
778 if ((long)regs->regs[insn.i_format.rs] > 0)
779 *contpc = regs->cp0_epc +
780 dec_insn.pc_inc +
781 (insn.i_format.simmediate << 2);
782 else
783 *contpc = regs->cp0_epc +
784 dec_insn.pc_inc +
785 dec_insn.next_pc_inc;
786 return 1;
787 #ifdef CONFIG_CPU_CAVIUM_OCTEON
788 case lwc2_op: /* This is bbit0 on Octeon */
789 if ((regs->regs[insn.i_format.rs] & (1ull<<insn.i_format.rt)) == 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 ldc2_op: /* This is bbit032 on Octeon */
795 if ((regs->regs[insn.i_format.rs] & (1ull<<(insn.i_format.rt + 32))) == 0)
796 *contpc = regs->cp0_epc + 4 + (insn.i_format.simmediate << 2);
797 else
798 *contpc = regs->cp0_epc + 8;
799 return 1;
800 case swc2_op: /* This is bbit1 on Octeon */
801 if (regs->regs[insn.i_format.rs] & (1ull<<insn.i_format.rt))
802 *contpc = regs->cp0_epc + 4 + (insn.i_format.simmediate << 2);
803 else
804 *contpc = regs->cp0_epc + 8;
805 return 1;
806 case sdc2_op: /* This is bbit132 on Octeon */
807 if (regs->regs[insn.i_format.rs] & (1ull<<(insn.i_format.rt + 32)))
808 *contpc = regs->cp0_epc + 4 + (insn.i_format.simmediate << 2);
809 else
810 *contpc = regs->cp0_epc + 8;
811 return 1;
812 #endif
813 case cop0_op:
814 case cop1_op:
815 case cop2_op:
816 case cop1x_op:
817 if (insn.i_format.rs == bc_op) {
818 preempt_disable();
819 if (is_fpu_owner())
820 asm volatile("cfc1\t%0,$31" : "=r" (fcr31));
821 else
822 fcr31 = current->thread.fpu.fcr31;
823 preempt_enable();
824
825 bit = (insn.i_format.rt >> 2);
826 bit += (bit != 0);
827 bit += 23;
828 switch (insn.i_format.rt & 3) {
829 case 0: /* bc1f */
830 case 2: /* bc1fl */
831 if (~fcr31 & (1 << bit))
832 *contpc = regs->cp0_epc +
833 dec_insn.pc_inc +
834 (insn.i_format.simmediate << 2);
835 else
836 *contpc = regs->cp0_epc +
837 dec_insn.pc_inc +
838 dec_insn.next_pc_inc;
839 return 1;
840 case 1: /* bc1t */
841 case 3: /* bc1tl */
842 if (fcr31 & (1 << bit))
843 *contpc = regs->cp0_epc +
844 dec_insn.pc_inc +
845 (insn.i_format.simmediate << 2);
846 else
847 *contpc = regs->cp0_epc +
848 dec_insn.pc_inc +
849 dec_insn.next_pc_inc;
850 return 1;
851 }
852 }
853 break;
854 }
855 return 0;
856 }
857
858 /*
859 * In the Linux kernel, we support selection of FPR format on the
860 * basis of the Status.FR bit. If an FPU is not present, the FR bit
861 * is hardwired to zero, which would imply a 32-bit FPU even for
862 * 64-bit CPUs so we rather look at TIF_32BIT_FPREGS.
863 * FPU emu is slow and bulky and optimizing this function offers fairly
864 * sizeable benefits so we try to be clever and make this function return
865 * a constant whenever possible, that is on 64-bit kernels without O32
866 * compatibility enabled and on 32-bit without 64-bit FPU support.
867 */
868 static inline int cop1_64bit(struct pt_regs *xcp)
869 {
870 #if defined(CONFIG_64BIT) && !defined(CONFIG_MIPS32_O32)
871 return 1;
872 #elif defined(CONFIG_32BIT) && !defined(CONFIG_MIPS_O32_FP64_SUPPORT)
873 return 0;
874 #else
875 return !test_thread_flag(TIF_32BIT_FPREGS);
876 #endif
877 }
878
879 #define SIFROMREG(si, x) do { \
880 if (cop1_64bit(xcp)) \
881 (si) = get_fpr32(&ctx->fpr[x], 0); \
882 else \
883 (si) = get_fpr32(&ctx->fpr[(x) & ~1], (x) & 1); \
884 } while (0)
885
886 #define SITOREG(si, x) do { \
887 if (cop1_64bit(xcp)) { \
888 unsigned i; \
889 set_fpr32(&ctx->fpr[x], 0, si); \
890 for (i = 1; i < ARRAY_SIZE(ctx->fpr[x].val32); i++) \
891 set_fpr32(&ctx->fpr[x], i, 0); \
892 } else { \
893 set_fpr32(&ctx->fpr[(x) & ~1], (x) & 1, si); \
894 } \
895 } while (0)
896
897 #define SIFROMHREG(si, x) ((si) = get_fpr32(&ctx->fpr[x], 1))
898
899 #define SITOHREG(si, x) do { \
900 unsigned i; \
901 set_fpr32(&ctx->fpr[x], 1, si); \
902 for (i = 2; i < ARRAY_SIZE(ctx->fpr[x].val32); i++) \
903 set_fpr32(&ctx->fpr[x], i, 0); \
904 } while (0)
905
906 #define DIFROMREG(di, x) \
907 ((di) = get_fpr64(&ctx->fpr[(x) & ~(cop1_64bit(xcp) == 0)], 0))
908
909 #define DITOREG(di, x) do { \
910 unsigned fpr, i; \
911 fpr = (x) & ~(cop1_64bit(xcp) == 0); \
912 set_fpr64(&ctx->fpr[fpr], 0, di); \
913 for (i = 1; i < ARRAY_SIZE(ctx->fpr[x].val64); i++) \
914 set_fpr64(&ctx->fpr[fpr], i, 0); \
915 } while (0)
916
917 #define SPFROMREG(sp, x) SIFROMREG((sp).bits, x)
918 #define SPTOREG(sp, x) SITOREG((sp).bits, x)
919 #define DPFROMREG(dp, x) DIFROMREG((dp).bits, x)
920 #define DPTOREG(dp, x) DITOREG((dp).bits, x)
921
922 /*
923 * Emulate the single floating point instruction pointed at by EPC.
924 * Two instructions if the instruction is in a branch delay slot.
925 */
926
927 static int cop1Emulate(struct pt_regs *xcp, struct mips_fpu_struct *ctx,
928 struct mm_decoded_insn dec_insn, void *__user *fault_addr)
929 {
930 mips_instruction ir;
931 unsigned long contpc = xcp->cp0_epc + dec_insn.pc_inc;
932 unsigned int cond;
933 int pc_inc;
934
935 /* XXX NEC Vr54xx bug workaround */
936 if (xcp->cp0_cause & CAUSEF_BD) {
937 if (dec_insn.micro_mips_mode) {
938 if (!mm_isBranchInstr(xcp, dec_insn, &contpc))
939 xcp->cp0_cause &= ~CAUSEF_BD;
940 } else {
941 if (!isBranchInstr(xcp, dec_insn, &contpc))
942 xcp->cp0_cause &= ~CAUSEF_BD;
943 }
944 }
945
946 if (xcp->cp0_cause & CAUSEF_BD) {
947 /*
948 * The instruction to be emulated is in a branch delay slot
949 * which means that we have to emulate the branch instruction
950 * BEFORE we do the cop1 instruction.
951 *
952 * This branch could be a COP1 branch, but in that case we
953 * would have had a trap for that instruction, and would not
954 * come through this route.
955 *
956 * Linux MIPS branch emulator operates on context, updating the
957 * cp0_epc.
958 */
959 ir = dec_insn.next_insn; /* process delay slot instr */
960 pc_inc = dec_insn.next_pc_inc;
961 } else {
962 ir = dec_insn.insn; /* process current instr */
963 pc_inc = dec_insn.pc_inc;
964 }
965
966 /*
967 * Since microMIPS FPU instructios are a subset of MIPS32 FPU
968 * instructions, we want to convert microMIPS FPU instructions
969 * into MIPS32 instructions so that we could reuse all of the
970 * FPU emulation code.
971 *
972 * NOTE: We cannot do this for branch instructions since they
973 * are not a subset. Example: Cannot emulate a 16-bit
974 * aligned target address with a MIPS32 instruction.
975 */
976 if (dec_insn.micro_mips_mode) {
977 /*
978 * If next instruction is a 16-bit instruction, then it
979 * it cannot be a FPU instruction. This could happen
980 * since we can be called for non-FPU instructions.
981 */
982 if ((pc_inc == 2) ||
983 (microMIPS32_to_MIPS32((union mips_instruction *)&ir)
984 == SIGILL))
985 return SIGILL;
986 }
987
988 emul:
989 perf_sw_event(PERF_COUNT_SW_EMULATION_FAULTS, 1, xcp, 0);
990 MIPS_FPU_EMU_INC_STATS(emulated);
991 switch (MIPSInst_OPCODE(ir)) {
992 case ldc1_op:{
993 u64 __user *va = (u64 __user *) (xcp->regs[MIPSInst_RS(ir)] +
994 MIPSInst_SIMM(ir));
995 u64 val;
996
997 MIPS_FPU_EMU_INC_STATS(loads);
998
999 if (!access_ok(VERIFY_READ, va, sizeof(u64))) {
1000 MIPS_FPU_EMU_INC_STATS(errors);
1001 *fault_addr = va;
1002 return SIGBUS;
1003 }
1004 if (__get_user(val, va)) {
1005 MIPS_FPU_EMU_INC_STATS(errors);
1006 *fault_addr = va;
1007 return SIGSEGV;
1008 }
1009 DITOREG(val, MIPSInst_RT(ir));
1010 break;
1011 }
1012
1013 case sdc1_op:{
1014 u64 __user *va = (u64 __user *) (xcp->regs[MIPSInst_RS(ir)] +
1015 MIPSInst_SIMM(ir));
1016 u64 val;
1017
1018 MIPS_FPU_EMU_INC_STATS(stores);
1019 DIFROMREG(val, MIPSInst_RT(ir));
1020 if (!access_ok(VERIFY_WRITE, va, sizeof(u64))) {
1021 MIPS_FPU_EMU_INC_STATS(errors);
1022 *fault_addr = va;
1023 return SIGBUS;
1024 }
1025 if (__put_user(val, va)) {
1026 MIPS_FPU_EMU_INC_STATS(errors);
1027 *fault_addr = va;
1028 return SIGSEGV;
1029 }
1030 break;
1031 }
1032
1033 case lwc1_op:{
1034 u32 __user *va = (u32 __user *) (xcp->regs[MIPSInst_RS(ir)] +
1035 MIPSInst_SIMM(ir));
1036 u32 val;
1037
1038 MIPS_FPU_EMU_INC_STATS(loads);
1039 if (!access_ok(VERIFY_READ, va, sizeof(u32))) {
1040 MIPS_FPU_EMU_INC_STATS(errors);
1041 *fault_addr = va;
1042 return SIGBUS;
1043 }
1044 if (__get_user(val, va)) {
1045 MIPS_FPU_EMU_INC_STATS(errors);
1046 *fault_addr = va;
1047 return SIGSEGV;
1048 }
1049 SITOREG(val, MIPSInst_RT(ir));
1050 break;
1051 }
1052
1053 case swc1_op:{
1054 u32 __user *va = (u32 __user *) (xcp->regs[MIPSInst_RS(ir)] +
1055 MIPSInst_SIMM(ir));
1056 u32 val;
1057
1058 MIPS_FPU_EMU_INC_STATS(stores);
1059 SIFROMREG(val, MIPSInst_RT(ir));
1060 if (!access_ok(VERIFY_WRITE, va, sizeof(u32))) {
1061 MIPS_FPU_EMU_INC_STATS(errors);
1062 *fault_addr = va;
1063 return SIGBUS;
1064 }
1065 if (__put_user(val, va)) {
1066 MIPS_FPU_EMU_INC_STATS(errors);
1067 *fault_addr = va;
1068 return SIGSEGV;
1069 }
1070 break;
1071 }
1072
1073 case cop1_op:
1074 switch (MIPSInst_RS(ir)) {
1075
1076 #if defined(__mips64)
1077 case dmfc_op:
1078 /* copregister fs -> gpr[rt] */
1079 if (MIPSInst_RT(ir) != 0) {
1080 DIFROMREG(xcp->regs[MIPSInst_RT(ir)],
1081 MIPSInst_RD(ir));
1082 }
1083 break;
1084
1085 case dmtc_op:
1086 /* copregister fs <- rt */
1087 DITOREG(xcp->regs[MIPSInst_RT(ir)], MIPSInst_RD(ir));
1088 break;
1089 #endif
1090
1091 case mfhc_op:
1092 if (!cpu_has_mips_r2)
1093 goto sigill;
1094
1095 /* copregister rd -> gpr[rt] */
1096 if (MIPSInst_RT(ir) != 0) {
1097 SIFROMHREG(xcp->regs[MIPSInst_RT(ir)],
1098 MIPSInst_RD(ir));
1099 }
1100 break;
1101
1102 case mthc_op:
1103 if (!cpu_has_mips_r2)
1104 goto sigill;
1105
1106 /* copregister rd <- gpr[rt] */
1107 SITOHREG(xcp->regs[MIPSInst_RT(ir)], MIPSInst_RD(ir));
1108 break;
1109
1110 case mfc_op:
1111 /* copregister rd -> gpr[rt] */
1112 if (MIPSInst_RT(ir) != 0) {
1113 SIFROMREG(xcp->regs[MIPSInst_RT(ir)],
1114 MIPSInst_RD(ir));
1115 }
1116 break;
1117
1118 case mtc_op:
1119 /* copregister rd <- rt */
1120 SITOREG(xcp->regs[MIPSInst_RT(ir)], MIPSInst_RD(ir));
1121 break;
1122
1123 case cfc_op:{
1124 /* cop control register rd -> gpr[rt] */
1125 u32 value;
1126
1127 if (MIPSInst_RD(ir) == FPCREG_CSR) {
1128 value = ctx->fcr31;
1129 value = (value & ~FPU_CSR_RM) |
1130 mips_rm[modeindex(value)];
1131 #ifdef CSRTRACE
1132 printk("%p gpr[%d]<-csr=%08x\n",
1133 (void *) (xcp->cp0_epc),
1134 MIPSInst_RT(ir), value);
1135 #endif
1136 }
1137 else if (MIPSInst_RD(ir) == FPCREG_RID)
1138 value = 0;
1139 else
1140 value = 0;
1141 if (MIPSInst_RT(ir))
1142 xcp->regs[MIPSInst_RT(ir)] = value;
1143 break;
1144 }
1145
1146 case ctc_op:{
1147 /* copregister rd <- rt */
1148 u32 value;
1149
1150 if (MIPSInst_RT(ir) == 0)
1151 value = 0;
1152 else
1153 value = xcp->regs[MIPSInst_RT(ir)];
1154
1155 /* we only have one writable control reg
1156 */
1157 if (MIPSInst_RD(ir) == FPCREG_CSR) {
1158 #ifdef CSRTRACE
1159 printk("%p gpr[%d]->csr=%08x\n",
1160 (void *) (xcp->cp0_epc),
1161 MIPSInst_RT(ir), value);
1162 #endif
1163
1164 /*
1165 * Don't write reserved bits,
1166 * and convert to ieee library modes
1167 */
1168 ctx->fcr31 = (value &
1169 ~(FPU_CSR_RSVD | FPU_CSR_RM)) |
1170 ieee_rm[modeindex(value)];
1171 }
1172 if ((ctx->fcr31 >> 5) & ctx->fcr31 & FPU_CSR_ALL_E) {
1173 return SIGFPE;
1174 }
1175 break;
1176 }
1177
1178 case bc_op:{
1179 int likely = 0;
1180
1181 if (xcp->cp0_cause & CAUSEF_BD)
1182 return SIGILL;
1183
1184 #if __mips >= 4
1185 cond = ctx->fcr31 & fpucondbit[MIPSInst_RT(ir) >> 2];
1186 #else
1187 cond = ctx->fcr31 & FPU_CSR_COND;
1188 #endif
1189 switch (MIPSInst_RT(ir) & 3) {
1190 case bcfl_op:
1191 likely = 1;
1192 case bcf_op:
1193 cond = !cond;
1194 break;
1195 case bctl_op:
1196 likely = 1;
1197 case bct_op:
1198 break;
1199 default:
1200 /* thats an illegal instruction */
1201 return SIGILL;
1202 }
1203
1204 xcp->cp0_cause |= CAUSEF_BD;
1205 if (cond) {
1206 /* branch taken: emulate dslot
1207 * instruction
1208 */
1209 xcp->cp0_epc += dec_insn.pc_inc;
1210
1211 contpc = MIPSInst_SIMM(ir);
1212 ir = dec_insn.next_insn;
1213 if (dec_insn.micro_mips_mode) {
1214 contpc = (xcp->cp0_epc + (contpc << 1));
1215
1216 /* If 16-bit instruction, not FPU. */
1217 if ((dec_insn.next_pc_inc == 2) ||
1218 (microMIPS32_to_MIPS32((union mips_instruction *)&ir) == SIGILL)) {
1219
1220 /*
1221 * Since this instruction will
1222 * be put on the stack with
1223 * 32-bit words, get around
1224 * this problem by putting a
1225 * NOP16 as the second one.
1226 */
1227 if (dec_insn.next_pc_inc == 2)
1228 ir = (ir & (~0xffff)) | MM_NOP16;
1229
1230 /*
1231 * Single step the non-CP1
1232 * instruction in the dslot.
1233 */
1234 return mips_dsemul(xcp, ir, contpc);
1235 }
1236 } else
1237 contpc = (xcp->cp0_epc + (contpc << 2));
1238
1239 switch (MIPSInst_OPCODE(ir)) {
1240 case lwc1_op:
1241 case swc1_op:
1242 #if (__mips >= 2 || defined(__mips64))
1243 case ldc1_op:
1244 case sdc1_op:
1245 #endif
1246 case cop1_op:
1247 #if __mips >= 4 && __mips != 32
1248 case cop1x_op:
1249 #endif
1250 /* its one of ours */
1251 goto emul;
1252 #if __mips >= 4
1253 case spec_op:
1254 if (MIPSInst_FUNC(ir) == movc_op)
1255 goto emul;
1256 break;
1257 #endif
1258 }
1259
1260 /*
1261 * Single step the non-cp1
1262 * instruction in the dslot
1263 */
1264 return mips_dsemul(xcp, ir, contpc);
1265 }
1266 else {
1267 /* branch not taken */
1268 if (likely) {
1269 /*
1270 * branch likely nullifies
1271 * dslot if not taken
1272 */
1273 xcp->cp0_epc += dec_insn.pc_inc;
1274 contpc += dec_insn.pc_inc;
1275 /*
1276 * else continue & execute
1277 * dslot as normal insn
1278 */
1279 }
1280 }
1281 break;
1282 }
1283
1284 default:
1285 if (!(MIPSInst_RS(ir) & 0x10))
1286 return SIGILL;
1287 {
1288 int sig;
1289
1290 /* a real fpu computation instruction */
1291 if ((sig = fpu_emu(xcp, ctx, ir)))
1292 return sig;
1293 }
1294 }
1295 break;
1296
1297 #if __mips >= 4 && __mips != 32
1298 case cop1x_op:{
1299 int sig = fpux_emu(xcp, ctx, ir, fault_addr);
1300 if (sig)
1301 return sig;
1302 break;
1303 }
1304 #endif
1305
1306 #if __mips >= 4
1307 case spec_op:
1308 if (MIPSInst_FUNC(ir) != movc_op)
1309 return SIGILL;
1310 cond = fpucondbit[MIPSInst_RT(ir) >> 2];
1311 if (((ctx->fcr31 & cond) != 0) == ((MIPSInst_RT(ir) & 1) != 0))
1312 xcp->regs[MIPSInst_RD(ir)] =
1313 xcp->regs[MIPSInst_RS(ir)];
1314 break;
1315 #endif
1316
1317 default:
1318 sigill:
1319 return SIGILL;
1320 }
1321
1322 /* we did it !! */
1323 xcp->cp0_epc = contpc;
1324 xcp->cp0_cause &= ~CAUSEF_BD;
1325
1326 return 0;
1327 }
1328
1329 /*
1330 * Conversion table from MIPS compare ops 48-63
1331 * cond = ieee754dp_cmp(x,y,IEEE754_UN,sig);
1332 */
1333 static const unsigned char cmptab[8] = {
1334 0, /* cmp_0 (sig) cmp_sf */
1335 IEEE754_CUN, /* cmp_un (sig) cmp_ngle */
1336 IEEE754_CEQ, /* cmp_eq (sig) cmp_seq */
1337 IEEE754_CEQ | IEEE754_CUN, /* cmp_ueq (sig) cmp_ngl */
1338 IEEE754_CLT, /* cmp_olt (sig) cmp_lt */
1339 IEEE754_CLT | IEEE754_CUN, /* cmp_ult (sig) cmp_nge */
1340 IEEE754_CLT | IEEE754_CEQ, /* cmp_ole (sig) cmp_le */
1341 IEEE754_CLT | IEEE754_CEQ | IEEE754_CUN, /* cmp_ule (sig) cmp_ngt */
1342 };
1343
1344
1345 #if __mips >= 4 && __mips != 32
1346
1347 /*
1348 * Additional MIPS4 instructions
1349 */
1350
1351 #define DEF3OP(name, p, f1, f2, f3) \
1352 static ieee754##p fpemu_##p##_##name(ieee754##p r, ieee754##p s, \
1353 ieee754##p t) \
1354 { \
1355 struct _ieee754_csr ieee754_csr_save; \
1356 s = f1(s, t); \
1357 ieee754_csr_save = ieee754_csr; \
1358 s = f2(s, r); \
1359 ieee754_csr_save.cx |= ieee754_csr.cx; \
1360 ieee754_csr_save.sx |= ieee754_csr.sx; \
1361 s = f3(s); \
1362 ieee754_csr.cx |= ieee754_csr_save.cx; \
1363 ieee754_csr.sx |= ieee754_csr_save.sx; \
1364 return s; \
1365 }
1366
1367 static ieee754dp fpemu_dp_recip(ieee754dp d)
1368 {
1369 return ieee754dp_div(ieee754dp_one(0), d);
1370 }
1371
1372 static ieee754dp fpemu_dp_rsqrt(ieee754dp d)
1373 {
1374 return ieee754dp_div(ieee754dp_one(0), ieee754dp_sqrt(d));
1375 }
1376
1377 static ieee754sp fpemu_sp_recip(ieee754sp s)
1378 {
1379 return ieee754sp_div(ieee754sp_one(0), s);
1380 }
1381
1382 static ieee754sp fpemu_sp_rsqrt(ieee754sp s)
1383 {
1384 return ieee754sp_div(ieee754sp_one(0), ieee754sp_sqrt(s));
1385 }
1386
1387 DEF3OP(madd, sp, ieee754sp_mul, ieee754sp_add, );
1388 DEF3OP(msub, sp, ieee754sp_mul, ieee754sp_sub, );
1389 DEF3OP(nmadd, sp, ieee754sp_mul, ieee754sp_add, ieee754sp_neg);
1390 DEF3OP(nmsub, sp, ieee754sp_mul, ieee754sp_sub, ieee754sp_neg);
1391 DEF3OP(madd, dp, ieee754dp_mul, ieee754dp_add, );
1392 DEF3OP(msub, dp, ieee754dp_mul, ieee754dp_sub, );
1393 DEF3OP(nmadd, dp, ieee754dp_mul, ieee754dp_add, ieee754dp_neg);
1394 DEF3OP(nmsub, dp, ieee754dp_mul, ieee754dp_sub, ieee754dp_neg);
1395
1396 static int fpux_emu(struct pt_regs *xcp, struct mips_fpu_struct *ctx,
1397 mips_instruction ir, void *__user *fault_addr)
1398 {
1399 unsigned rcsr = 0; /* resulting csr */
1400
1401 MIPS_FPU_EMU_INC_STATS(cp1xops);
1402
1403 switch (MIPSInst_FMA_FFMT(ir)) {
1404 case s_fmt:{ /* 0 */
1405
1406 ieee754sp(*handler) (ieee754sp, ieee754sp, ieee754sp);
1407 ieee754sp fd, fr, fs, ft;
1408 u32 __user *va;
1409 u32 val;
1410
1411 switch (MIPSInst_FUNC(ir)) {
1412 case lwxc1_op:
1413 va = (void __user *) (xcp->regs[MIPSInst_FR(ir)] +
1414 xcp->regs[MIPSInst_FT(ir)]);
1415
1416 MIPS_FPU_EMU_INC_STATS(loads);
1417 if (!access_ok(VERIFY_READ, va, sizeof(u32))) {
1418 MIPS_FPU_EMU_INC_STATS(errors);
1419 *fault_addr = va;
1420 return SIGBUS;
1421 }
1422 if (__get_user(val, va)) {
1423 MIPS_FPU_EMU_INC_STATS(errors);
1424 *fault_addr = va;
1425 return SIGSEGV;
1426 }
1427 SITOREG(val, MIPSInst_FD(ir));
1428 break;
1429
1430 case swxc1_op:
1431 va = (void __user *) (xcp->regs[MIPSInst_FR(ir)] +
1432 xcp->regs[MIPSInst_FT(ir)]);
1433
1434 MIPS_FPU_EMU_INC_STATS(stores);
1435
1436 SIFROMREG(val, MIPSInst_FS(ir));
1437 if (!access_ok(VERIFY_WRITE, va, sizeof(u32))) {
1438 MIPS_FPU_EMU_INC_STATS(errors);
1439 *fault_addr = va;
1440 return SIGBUS;
1441 }
1442 if (put_user(val, va)) {
1443 MIPS_FPU_EMU_INC_STATS(errors);
1444 *fault_addr = va;
1445 return SIGSEGV;
1446 }
1447 break;
1448
1449 case madd_s_op:
1450 handler = fpemu_sp_madd;
1451 goto scoptop;
1452 case msub_s_op:
1453 handler = fpemu_sp_msub;
1454 goto scoptop;
1455 case nmadd_s_op:
1456 handler = fpemu_sp_nmadd;
1457 goto scoptop;
1458 case nmsub_s_op:
1459 handler = fpemu_sp_nmsub;
1460 goto scoptop;
1461
1462 scoptop:
1463 SPFROMREG(fr, MIPSInst_FR(ir));
1464 SPFROMREG(fs, MIPSInst_FS(ir));
1465 SPFROMREG(ft, MIPSInst_FT(ir));
1466 fd = (*handler) (fr, fs, ft);
1467 SPTOREG(fd, MIPSInst_FD(ir));
1468
1469 copcsr:
1470 if (ieee754_cxtest(IEEE754_INEXACT))
1471 rcsr |= FPU_CSR_INE_X | FPU_CSR_INE_S;
1472 if (ieee754_cxtest(IEEE754_UNDERFLOW))
1473 rcsr |= FPU_CSR_UDF_X | FPU_CSR_UDF_S;
1474 if (ieee754_cxtest(IEEE754_OVERFLOW))
1475 rcsr |= FPU_CSR_OVF_X | FPU_CSR_OVF_S;
1476 if (ieee754_cxtest(IEEE754_INVALID_OPERATION))
1477 rcsr |= FPU_CSR_INV_X | FPU_CSR_INV_S;
1478
1479 ctx->fcr31 = (ctx->fcr31 & ~FPU_CSR_ALL_X) | rcsr;
1480 if ((ctx->fcr31 >> 5) & ctx->fcr31 & FPU_CSR_ALL_E) {
1481 /*printk ("SIGFPE: fpu csr = %08x\n",
1482 ctx->fcr31); */
1483 return SIGFPE;
1484 }
1485
1486 break;
1487
1488 default:
1489 return SIGILL;
1490 }
1491 break;
1492 }
1493
1494 case d_fmt:{ /* 1 */
1495 ieee754dp(*handler) (ieee754dp, ieee754dp, ieee754dp);
1496 ieee754dp fd, fr, fs, ft;
1497 u64 __user *va;
1498 u64 val;
1499
1500 switch (MIPSInst_FUNC(ir)) {
1501 case ldxc1_op:
1502 va = (void __user *) (xcp->regs[MIPSInst_FR(ir)] +
1503 xcp->regs[MIPSInst_FT(ir)]);
1504
1505 MIPS_FPU_EMU_INC_STATS(loads);
1506 if (!access_ok(VERIFY_READ, va, sizeof(u64))) {
1507 MIPS_FPU_EMU_INC_STATS(errors);
1508 *fault_addr = va;
1509 return SIGBUS;
1510 }
1511 if (__get_user(val, va)) {
1512 MIPS_FPU_EMU_INC_STATS(errors);
1513 *fault_addr = va;
1514 return SIGSEGV;
1515 }
1516 DITOREG(val, MIPSInst_FD(ir));
1517 break;
1518
1519 case sdxc1_op:
1520 va = (void __user *) (xcp->regs[MIPSInst_FR(ir)] +
1521 xcp->regs[MIPSInst_FT(ir)]);
1522
1523 MIPS_FPU_EMU_INC_STATS(stores);
1524 DIFROMREG(val, MIPSInst_FS(ir));
1525 if (!access_ok(VERIFY_WRITE, va, sizeof(u64))) {
1526 MIPS_FPU_EMU_INC_STATS(errors);
1527 *fault_addr = va;
1528 return SIGBUS;
1529 }
1530 if (__put_user(val, va)) {
1531 MIPS_FPU_EMU_INC_STATS(errors);
1532 *fault_addr = va;
1533 return SIGSEGV;
1534 }
1535 break;
1536
1537 case madd_d_op:
1538 handler = fpemu_dp_madd;
1539 goto dcoptop;
1540 case msub_d_op:
1541 handler = fpemu_dp_msub;
1542 goto dcoptop;
1543 case nmadd_d_op:
1544 handler = fpemu_dp_nmadd;
1545 goto dcoptop;
1546 case nmsub_d_op:
1547 handler = fpemu_dp_nmsub;
1548 goto dcoptop;
1549
1550 dcoptop:
1551 DPFROMREG(fr, MIPSInst_FR(ir));
1552 DPFROMREG(fs, MIPSInst_FS(ir));
1553 DPFROMREG(ft, MIPSInst_FT(ir));
1554 fd = (*handler) (fr, fs, ft);
1555 DPTOREG(fd, MIPSInst_FD(ir));
1556 goto copcsr;
1557
1558 default:
1559 return SIGILL;
1560 }
1561 break;
1562 }
1563
1564 case 0x3:
1565 if (MIPSInst_FUNC(ir) != pfetch_op)
1566 return SIGILL;
1567
1568 /* ignore prefx operation */
1569 break;
1570
1571 default:
1572 return SIGILL;
1573 }
1574
1575 return 0;
1576 }
1577 #endif
1578
1579
1580
1581 /*
1582 * Emulate a single COP1 arithmetic instruction.
1583 */
1584 static int fpu_emu(struct pt_regs *xcp, struct mips_fpu_struct *ctx,
1585 mips_instruction ir)
1586 {
1587 int rfmt; /* resulting format */
1588 unsigned rcsr = 0; /* resulting csr */
1589 unsigned cond;
1590 union {
1591 ieee754dp d;
1592 ieee754sp s;
1593 int w;
1594 #ifdef __mips64
1595 s64 l;
1596 #endif
1597 } rv; /* resulting value */
1598
1599 MIPS_FPU_EMU_INC_STATS(cp1ops);
1600 switch (rfmt = (MIPSInst_FFMT(ir) & 0xf)) {
1601 case s_fmt:{ /* 0 */
1602 union {
1603 ieee754sp(*b) (ieee754sp, ieee754sp);
1604 ieee754sp(*u) (ieee754sp);
1605 } handler;
1606
1607 switch (MIPSInst_FUNC(ir)) {
1608 /* binary ops */
1609 case fadd_op:
1610 handler.b = ieee754sp_add;
1611 goto scopbop;
1612 case fsub_op:
1613 handler.b = ieee754sp_sub;
1614 goto scopbop;
1615 case fmul_op:
1616 handler.b = ieee754sp_mul;
1617 goto scopbop;
1618 case fdiv_op:
1619 handler.b = ieee754sp_div;
1620 goto scopbop;
1621
1622 /* unary ops */
1623 #if __mips >= 2 || defined(__mips64)
1624 case fsqrt_op:
1625 handler.u = ieee754sp_sqrt;
1626 goto scopuop;
1627 #endif
1628 #if __mips >= 4 && __mips != 32
1629 case frsqrt_op:
1630 handler.u = fpemu_sp_rsqrt;
1631 goto scopuop;
1632 case frecip_op:
1633 handler.u = fpemu_sp_recip;
1634 goto scopuop;
1635 #endif
1636 #if __mips >= 4
1637 case fmovc_op:
1638 cond = fpucondbit[MIPSInst_FT(ir) >> 2];
1639 if (((ctx->fcr31 & cond) != 0) !=
1640 ((MIPSInst_FT(ir) & 1) != 0))
1641 return 0;
1642 SPFROMREG(rv.s, MIPSInst_FS(ir));
1643 break;
1644 case fmovz_op:
1645 if (xcp->regs[MIPSInst_FT(ir)] != 0)
1646 return 0;
1647 SPFROMREG(rv.s, MIPSInst_FS(ir));
1648 break;
1649 case fmovn_op:
1650 if (xcp->regs[MIPSInst_FT(ir)] == 0)
1651 return 0;
1652 SPFROMREG(rv.s, MIPSInst_FS(ir));
1653 break;
1654 #endif
1655 case fabs_op:
1656 handler.u = ieee754sp_abs;
1657 goto scopuop;
1658 case fneg_op:
1659 handler.u = ieee754sp_neg;
1660 goto scopuop;
1661 case fmov_op:
1662 /* an easy one */
1663 SPFROMREG(rv.s, MIPSInst_FS(ir));
1664 goto copcsr;
1665
1666 /* binary op on handler */
1667 scopbop:
1668 {
1669 ieee754sp fs, ft;
1670
1671 SPFROMREG(fs, MIPSInst_FS(ir));
1672 SPFROMREG(ft, MIPSInst_FT(ir));
1673
1674 rv.s = (*handler.b) (fs, ft);
1675 goto copcsr;
1676 }
1677 scopuop:
1678 {
1679 ieee754sp fs;
1680
1681 SPFROMREG(fs, MIPSInst_FS(ir));
1682 rv.s = (*handler.u) (fs);
1683 goto copcsr;
1684 }
1685 copcsr:
1686 if (ieee754_cxtest(IEEE754_INEXACT))
1687 rcsr |= FPU_CSR_INE_X | FPU_CSR_INE_S;
1688 if (ieee754_cxtest(IEEE754_UNDERFLOW))
1689 rcsr |= FPU_CSR_UDF_X | FPU_CSR_UDF_S;
1690 if (ieee754_cxtest(IEEE754_OVERFLOW))
1691 rcsr |= FPU_CSR_OVF_X | FPU_CSR_OVF_S;
1692 if (ieee754_cxtest(IEEE754_ZERO_DIVIDE))
1693 rcsr |= FPU_CSR_DIV_X | FPU_CSR_DIV_S;
1694 if (ieee754_cxtest(IEEE754_INVALID_OPERATION))
1695 rcsr |= FPU_CSR_INV_X | FPU_CSR_INV_S;
1696 break;
1697
1698 /* unary conv ops */
1699 case fcvts_op:
1700 return SIGILL; /* not defined */
1701 case fcvtd_op:{
1702 ieee754sp fs;
1703
1704 SPFROMREG(fs, MIPSInst_FS(ir));
1705 rv.d = ieee754dp_fsp(fs);
1706 rfmt = d_fmt;
1707 goto copcsr;
1708 }
1709 case fcvtw_op:{
1710 ieee754sp fs;
1711
1712 SPFROMREG(fs, MIPSInst_FS(ir));
1713 rv.w = ieee754sp_tint(fs);
1714 rfmt = w_fmt;
1715 goto copcsr;
1716 }
1717
1718 #if __mips >= 2 || defined(__mips64)
1719 case fround_op:
1720 case ftrunc_op:
1721 case fceil_op:
1722 case ffloor_op:{
1723 unsigned int oldrm = ieee754_csr.rm;
1724 ieee754sp fs;
1725
1726 SPFROMREG(fs, MIPSInst_FS(ir));
1727 ieee754_csr.rm = ieee_rm[modeindex(MIPSInst_FUNC(ir))];
1728 rv.w = ieee754sp_tint(fs);
1729 ieee754_csr.rm = oldrm;
1730 rfmt = w_fmt;
1731 goto copcsr;
1732 }
1733 #endif /* __mips >= 2 */
1734
1735 #if defined(__mips64)
1736 case fcvtl_op:{
1737 ieee754sp fs;
1738
1739 SPFROMREG(fs, MIPSInst_FS(ir));
1740 rv.l = ieee754sp_tlong(fs);
1741 rfmt = l_fmt;
1742 goto copcsr;
1743 }
1744
1745 case froundl_op:
1746 case ftruncl_op:
1747 case fceill_op:
1748 case ffloorl_op:{
1749 unsigned int oldrm = ieee754_csr.rm;
1750 ieee754sp fs;
1751
1752 SPFROMREG(fs, MIPSInst_FS(ir));
1753 ieee754_csr.rm = ieee_rm[modeindex(MIPSInst_FUNC(ir))];
1754 rv.l = ieee754sp_tlong(fs);
1755 ieee754_csr.rm = oldrm;
1756 rfmt = l_fmt;
1757 goto copcsr;
1758 }
1759 #endif /* defined(__mips64) */
1760
1761 default:
1762 if (MIPSInst_FUNC(ir) >= fcmp_op) {
1763 unsigned cmpop = MIPSInst_FUNC(ir) - fcmp_op;
1764 ieee754sp fs, ft;
1765
1766 SPFROMREG(fs, MIPSInst_FS(ir));
1767 SPFROMREG(ft, MIPSInst_FT(ir));
1768 rv.w = ieee754sp_cmp(fs, ft,
1769 cmptab[cmpop & 0x7], cmpop & 0x8);
1770 rfmt = -1;
1771 if ((cmpop & 0x8) && ieee754_cxtest
1772 (IEEE754_INVALID_OPERATION))
1773 rcsr = FPU_CSR_INV_X | FPU_CSR_INV_S;
1774 else
1775 goto copcsr;
1776
1777 }
1778 else {
1779 return SIGILL;
1780 }
1781 break;
1782 }
1783 break;
1784 }
1785
1786 case d_fmt:{
1787 union {
1788 ieee754dp(*b) (ieee754dp, ieee754dp);
1789 ieee754dp(*u) (ieee754dp);
1790 } handler;
1791
1792 switch (MIPSInst_FUNC(ir)) {
1793 /* binary ops */
1794 case fadd_op:
1795 handler.b = ieee754dp_add;
1796 goto dcopbop;
1797 case fsub_op:
1798 handler.b = ieee754dp_sub;
1799 goto dcopbop;
1800 case fmul_op:
1801 handler.b = ieee754dp_mul;
1802 goto dcopbop;
1803 case fdiv_op:
1804 handler.b = ieee754dp_div;
1805 goto dcopbop;
1806
1807 /* unary ops */
1808 #if __mips >= 2 || defined(__mips64)
1809 case fsqrt_op:
1810 handler.u = ieee754dp_sqrt;
1811 goto dcopuop;
1812 #endif
1813 #if __mips >= 4 && __mips != 32
1814 case frsqrt_op:
1815 handler.u = fpemu_dp_rsqrt;
1816 goto dcopuop;
1817 case frecip_op:
1818 handler.u = fpemu_dp_recip;
1819 goto dcopuop;
1820 #endif
1821 #if __mips >= 4
1822 case fmovc_op:
1823 cond = fpucondbit[MIPSInst_FT(ir) >> 2];
1824 if (((ctx->fcr31 & cond) != 0) !=
1825 ((MIPSInst_FT(ir) & 1) != 0))
1826 return 0;
1827 DPFROMREG(rv.d, MIPSInst_FS(ir));
1828 break;
1829 case fmovz_op:
1830 if (xcp->regs[MIPSInst_FT(ir)] != 0)
1831 return 0;
1832 DPFROMREG(rv.d, MIPSInst_FS(ir));
1833 break;
1834 case fmovn_op:
1835 if (xcp->regs[MIPSInst_FT(ir)] == 0)
1836 return 0;
1837 DPFROMREG(rv.d, MIPSInst_FS(ir));
1838 break;
1839 #endif
1840 case fabs_op:
1841 handler.u = ieee754dp_abs;
1842 goto dcopuop;
1843
1844 case fneg_op:
1845 handler.u = ieee754dp_neg;
1846 goto dcopuop;
1847
1848 case fmov_op:
1849 /* an easy one */
1850 DPFROMREG(rv.d, MIPSInst_FS(ir));
1851 goto copcsr;
1852
1853 /* binary op on handler */
1854 dcopbop:{
1855 ieee754dp fs, ft;
1856
1857 DPFROMREG(fs, MIPSInst_FS(ir));
1858 DPFROMREG(ft, MIPSInst_FT(ir));
1859
1860 rv.d = (*handler.b) (fs, ft);
1861 goto copcsr;
1862 }
1863 dcopuop:{
1864 ieee754dp fs;
1865
1866 DPFROMREG(fs, MIPSInst_FS(ir));
1867 rv.d = (*handler.u) (fs);
1868 goto copcsr;
1869 }
1870
1871 /* unary conv ops */
1872 case fcvts_op:{
1873 ieee754dp fs;
1874
1875 DPFROMREG(fs, MIPSInst_FS(ir));
1876 rv.s = ieee754sp_fdp(fs);
1877 rfmt = s_fmt;
1878 goto copcsr;
1879 }
1880 case fcvtd_op:
1881 return SIGILL; /* not defined */
1882
1883 case fcvtw_op:{
1884 ieee754dp fs;
1885
1886 DPFROMREG(fs, MIPSInst_FS(ir));
1887 rv.w = ieee754dp_tint(fs); /* wrong */
1888 rfmt = w_fmt;
1889 goto copcsr;
1890 }
1891
1892 #if __mips >= 2 || defined(__mips64)
1893 case fround_op:
1894 case ftrunc_op:
1895 case fceil_op:
1896 case ffloor_op:{
1897 unsigned int oldrm = ieee754_csr.rm;
1898 ieee754dp fs;
1899
1900 DPFROMREG(fs, MIPSInst_FS(ir));
1901 ieee754_csr.rm = ieee_rm[modeindex(MIPSInst_FUNC(ir))];
1902 rv.w = ieee754dp_tint(fs);
1903 ieee754_csr.rm = oldrm;
1904 rfmt = w_fmt;
1905 goto copcsr;
1906 }
1907 #endif
1908
1909 #if defined(__mips64)
1910 case fcvtl_op:{
1911 ieee754dp fs;
1912
1913 DPFROMREG(fs, MIPSInst_FS(ir));
1914 rv.l = ieee754dp_tlong(fs);
1915 rfmt = l_fmt;
1916 goto copcsr;
1917 }
1918
1919 case froundl_op:
1920 case ftruncl_op:
1921 case fceill_op:
1922 case ffloorl_op:{
1923 unsigned int oldrm = ieee754_csr.rm;
1924 ieee754dp fs;
1925
1926 DPFROMREG(fs, MIPSInst_FS(ir));
1927 ieee754_csr.rm = ieee_rm[modeindex(MIPSInst_FUNC(ir))];
1928 rv.l = ieee754dp_tlong(fs);
1929 ieee754_csr.rm = oldrm;
1930 rfmt = l_fmt;
1931 goto copcsr;
1932 }
1933 #endif /* __mips >= 3 */
1934
1935 default:
1936 if (MIPSInst_FUNC(ir) >= fcmp_op) {
1937 unsigned cmpop = MIPSInst_FUNC(ir) - fcmp_op;
1938 ieee754dp fs, ft;
1939
1940 DPFROMREG(fs, MIPSInst_FS(ir));
1941 DPFROMREG(ft, MIPSInst_FT(ir));
1942 rv.w = ieee754dp_cmp(fs, ft,
1943 cmptab[cmpop & 0x7], cmpop & 0x8);
1944 rfmt = -1;
1945 if ((cmpop & 0x8)
1946 &&
1947 ieee754_cxtest
1948 (IEEE754_INVALID_OPERATION))
1949 rcsr = FPU_CSR_INV_X | FPU_CSR_INV_S;
1950 else
1951 goto copcsr;
1952
1953 }
1954 else {
1955 return SIGILL;
1956 }
1957 break;
1958 }
1959 break;
1960 }
1961
1962 case w_fmt:{
1963 ieee754sp fs;
1964
1965 switch (MIPSInst_FUNC(ir)) {
1966 case fcvts_op:
1967 /* convert word to single precision real */
1968 SPFROMREG(fs, MIPSInst_FS(ir));
1969 rv.s = ieee754sp_fint(fs.bits);
1970 rfmt = s_fmt;
1971 goto copcsr;
1972 case fcvtd_op:
1973 /* convert word to double precision real */
1974 SPFROMREG(fs, MIPSInst_FS(ir));
1975 rv.d = ieee754dp_fint(fs.bits);
1976 rfmt = d_fmt;
1977 goto copcsr;
1978 default:
1979 return SIGILL;
1980 }
1981 break;
1982 }
1983
1984 #if defined(__mips64)
1985 case l_fmt:{
1986 u64 bits;
1987 DIFROMREG(bits, MIPSInst_FS(ir));
1988
1989 switch (MIPSInst_FUNC(ir)) {
1990 case fcvts_op:
1991 /* convert long to single precision real */
1992 rv.s = ieee754sp_flong(bits);
1993 rfmt = s_fmt;
1994 goto copcsr;
1995 case fcvtd_op:
1996 /* convert long to double precision real */
1997 rv.d = ieee754dp_flong(bits);
1998 rfmt = d_fmt;
1999 goto copcsr;
2000 default:
2001 return SIGILL;
2002 }
2003 break;
2004 }
2005 #endif
2006
2007 default:
2008 return SIGILL;
2009 }
2010
2011 /*
2012 * Update the fpu CSR register for this operation.
2013 * If an exception is required, generate a tidy SIGFPE exception,
2014 * without updating the result register.
2015 * Note: cause exception bits do not accumulate, they are rewritten
2016 * for each op; only the flag/sticky bits accumulate.
2017 */
2018 ctx->fcr31 = (ctx->fcr31 & ~FPU_CSR_ALL_X) | rcsr;
2019 if ((ctx->fcr31 >> 5) & ctx->fcr31 & FPU_CSR_ALL_E) {
2020 /*printk ("SIGFPE: fpu csr = %08x\n",ctx->fcr31); */
2021 return SIGFPE;
2022 }
2023
2024 /*
2025 * Now we can safely write the result back to the register file.
2026 */
2027 switch (rfmt) {
2028 case -1:{
2029 #if __mips >= 4
2030 cond = fpucondbit[MIPSInst_FD(ir) >> 2];
2031 #else
2032 cond = FPU_CSR_COND;
2033 #endif
2034 if (rv.w)
2035 ctx->fcr31 |= cond;
2036 else
2037 ctx->fcr31 &= ~cond;
2038 break;
2039 }
2040 case d_fmt:
2041 DPTOREG(rv.d, MIPSInst_FD(ir));
2042 break;
2043 case s_fmt:
2044 SPTOREG(rv.s, MIPSInst_FD(ir));
2045 break;
2046 case w_fmt:
2047 SITOREG(rv.w, MIPSInst_FD(ir));
2048 break;
2049 #if defined(__mips64)
2050 case l_fmt:
2051 DITOREG(rv.l, MIPSInst_FD(ir));
2052 break;
2053 #endif
2054 default:
2055 return SIGILL;
2056 }
2057
2058 return 0;
2059 }
2060
2061 int fpu_emulator_cop1Handler(struct pt_regs *xcp, struct mips_fpu_struct *ctx,
2062 int has_fpu, void *__user *fault_addr)
2063 {
2064 unsigned long oldepc, prevepc;
2065 struct mm_decoded_insn dec_insn;
2066 u16 instr[4];
2067 u16 *instr_ptr;
2068 int sig = 0;
2069
2070 oldepc = xcp->cp0_epc;
2071 do {
2072 prevepc = xcp->cp0_epc;
2073
2074 if (get_isa16_mode(prevepc) && cpu_has_mmips) {
2075 /*
2076 * Get next 2 microMIPS instructions and convert them
2077 * into 32-bit instructions.
2078 */
2079 if ((get_user(instr[0], (u16 __user *)msk_isa16_mode(xcp->cp0_epc))) ||
2080 (get_user(instr[1], (u16 __user *)msk_isa16_mode(xcp->cp0_epc + 2))) ||
2081 (get_user(instr[2], (u16 __user *)msk_isa16_mode(xcp->cp0_epc + 4))) ||
2082 (get_user(instr[3], (u16 __user *)msk_isa16_mode(xcp->cp0_epc + 6)))) {
2083 MIPS_FPU_EMU_INC_STATS(errors);
2084 return SIGBUS;
2085 }
2086 instr_ptr = instr;
2087
2088 /* Get first instruction. */
2089 if (mm_insn_16bit(*instr_ptr)) {
2090 /* Duplicate the half-word. */
2091 dec_insn.insn = (*instr_ptr << 16) |
2092 (*instr_ptr);
2093 /* 16-bit instruction. */
2094 dec_insn.pc_inc = 2;
2095 instr_ptr += 1;
2096 } else {
2097 dec_insn.insn = (*instr_ptr << 16) |
2098 *(instr_ptr+1);
2099 /* 32-bit instruction. */
2100 dec_insn.pc_inc = 4;
2101 instr_ptr += 2;
2102 }
2103 /* Get second instruction. */
2104 if (mm_insn_16bit(*instr_ptr)) {
2105 /* Duplicate the half-word. */
2106 dec_insn.next_insn = (*instr_ptr << 16) |
2107 (*instr_ptr);
2108 /* 16-bit instruction. */
2109 dec_insn.next_pc_inc = 2;
2110 } else {
2111 dec_insn.next_insn = (*instr_ptr << 16) |
2112 *(instr_ptr+1);
2113 /* 32-bit instruction. */
2114 dec_insn.next_pc_inc = 4;
2115 }
2116 dec_insn.micro_mips_mode = 1;
2117 } else {
2118 if ((get_user(dec_insn.insn,
2119 (mips_instruction __user *) xcp->cp0_epc)) ||
2120 (get_user(dec_insn.next_insn,
2121 (mips_instruction __user *)(xcp->cp0_epc+4)))) {
2122 MIPS_FPU_EMU_INC_STATS(errors);
2123 return SIGBUS;
2124 }
2125 dec_insn.pc_inc = 4;
2126 dec_insn.next_pc_inc = 4;
2127 dec_insn.micro_mips_mode = 0;
2128 }
2129
2130 if ((dec_insn.insn == 0) ||
2131 ((dec_insn.pc_inc == 2) &&
2132 ((dec_insn.insn & 0xffff) == MM_NOP16)))
2133 xcp->cp0_epc += dec_insn.pc_inc; /* Skip NOPs */
2134 else {
2135 /*
2136 * The 'ieee754_csr' is an alias of
2137 * ctx->fcr31. No need to copy ctx->fcr31 to
2138 * ieee754_csr. But ieee754_csr.rm is ieee
2139 * library modes. (not mips rounding mode)
2140 */
2141 /* convert to ieee library modes */
2142 ieee754_csr.rm = ieee_rm[ieee754_csr.rm];
2143 sig = cop1Emulate(xcp, ctx, dec_insn, fault_addr);
2144 /* revert to mips rounding mode */
2145 ieee754_csr.rm = mips_rm[ieee754_csr.rm];
2146 }
2147
2148 if (has_fpu)
2149 break;
2150 if (sig)
2151 break;
2152
2153 cond_resched();
2154 } while (xcp->cp0_epc > prevepc);
2155
2156 /* SIGILL indicates a non-fpu instruction */
2157 if (sig == SIGILL && xcp->cp0_epc != oldepc)
2158 /* but if epc has advanced, then ignore it */
2159 sig = 0;
2160
2161 return sig;
2162 }
2163
2164 #ifdef CONFIG_DEBUG_FS
2165
2166 static int fpuemu_stat_get(void *data, u64 *val)
2167 {
2168 int cpu;
2169 unsigned long sum = 0;
2170 for_each_online_cpu(cpu) {
2171 struct mips_fpu_emulator_stats *ps;
2172 local_t *pv;
2173 ps = &per_cpu(fpuemustats, cpu);
2174 pv = (void *)ps + (unsigned long)data;
2175 sum += local_read(pv);
2176 }
2177 *val = sum;
2178 return 0;
2179 }
2180 DEFINE_SIMPLE_ATTRIBUTE(fops_fpuemu_stat, fpuemu_stat_get, NULL, "%llu\n");
2181
2182 extern struct dentry *mips_debugfs_dir;
2183 static int __init debugfs_fpuemu(void)
2184 {
2185 struct dentry *d, *dir;
2186
2187 if (!mips_debugfs_dir)
2188 return -ENODEV;
2189 dir = debugfs_create_dir("fpuemustats", mips_debugfs_dir);
2190 if (!dir)
2191 return -ENOMEM;
2192
2193 #define FPU_STAT_CREATE(M) \
2194 do { \
2195 d = debugfs_create_file(#M , S_IRUGO, dir, \
2196 (void *)offsetof(struct mips_fpu_emulator_stats, M), \
2197 &fops_fpuemu_stat); \
2198 if (!d) \
2199 return -ENOMEM; \
2200 } while (0)
2201
2202 FPU_STAT_CREATE(emulated);
2203 FPU_STAT_CREATE(loads);
2204 FPU_STAT_CREATE(stores);
2205 FPU_STAT_CREATE(cp1ops);
2206 FPU_STAT_CREATE(cp1xops);
2207 FPU_STAT_CREATE(errors);
2208
2209 return 0;
2210 }
2211 __initcall(debugfs_fpuemu);
2212 #endif
Linux with added FPC-III support