blob: 99c550632d44a541f29a7c70ee7052ae09869fa9 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * cp1emu.c: a MIPS coprocessor 1 (fpu) instruction emulator
3 *
4 * MIPS floating point support
5 * Copyright (C) 1994-2000 Algorithmics Ltd.
6 * http://www.algor.co.uk
7 *
8 * Kevin D. Kissell, kevink@mips.com and Carsten Langgaard, carstenl@mips.com
9 * Copyright (C) 2000 MIPS Technologies, Inc.
10 *
11 * This program is free software; you can distribute it and/or modify it
12 * under the terms of the GNU General Public License (Version 2) as
13 * published by the Free Software Foundation.
14 *
15 * This program is distributed in the hope it will be useful, but WITHOUT
16 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
17 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
18 * for more details.
19 *
20 * You should have received a copy of the GNU General Public License along
21 * with this program; if not, write to the Free Software Foundation, Inc.,
22 * 59 Temple Place - Suite 330, Boston MA 02111-1307, USA.
23 *
24 * A complete emulator for MIPS coprocessor 1 instructions. This is
25 * required for #float(switch) or #float(trap), where it catches all
26 * COP1 instructions via the "CoProcessor Unusable" exception.
27 *
28 * More surprisingly it is also required for #float(ieee), to help out
29 * the hardware fpu at the boundaries of the IEEE-754 representation
30 * (denormalised values, infinities, underflow, etc). It is made
31 * quite nasty because emulation of some non-COP1 instructions is
32 * required, e.g. in branch delay slots.
33 *
34 * Note if you know that you won't have an fpu, then you'll get much
35 * better performance by compiling with -msoft-float!
36 */
37#include <linux/sched.h>
38
39#include <asm/inst.h>
40#include <asm/bootinfo.h>
41#include <asm/cpu.h>
42#include <asm/cpu-features.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/uaccess.h>
49#include <asm/branch.h>
50
51#include "ieee754.h"
52#include "dsemul.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
63static int fpu_emu(struct pt_regs *, struct mips_fpu_soft_struct *,
64 mips_instruction);
65
66#if __mips >= 4 && __mips != 32
67static int fpux_emu(struct pt_regs *,
68 struct mips_fpu_soft_struct *, mips_instruction);
69#endif
70
71/* Further private data for which no space exists in mips_fpu_soft_struct */
72
73struct mips_fpu_emulator_private fpuemuprivate;
74
75/* Control registers */
76
77#define FPCREG_RID 0 /* $0 = revision id */
78#define FPCREG_CSR 31 /* $31 = csr */
79
80/* Convert Mips rounding mode (0..3) to IEEE library modes. */
81static const unsigned char ieee_rm[4] = {
82 IEEE754_RN, IEEE754_RZ, IEEE754_RU, IEEE754_RD
83};
84
85#if __mips >= 4
86/* convert condition code register number to csr bit */
87static const unsigned int fpucondbit[8] = {
88 FPU_CSR_COND0,
89 FPU_CSR_COND1,
90 FPU_CSR_COND2,
91 FPU_CSR_COND3,
92 FPU_CSR_COND4,
93 FPU_CSR_COND5,
94 FPU_CSR_COND6,
95 FPU_CSR_COND7
96};
97#endif
98
99
100/*
101 * Redundant with logic already in kernel/branch.c,
102 * embedded in compute_return_epc. At some point,
103 * a single subroutine should be used across both
104 * modules.
105 */
106static int isBranchInstr(mips_instruction * i)
107{
108 switch (MIPSInst_OPCODE(*i)) {
109 case spec_op:
110 switch (MIPSInst_FUNC(*i)) {
111 case jalr_op:
112 case jr_op:
113 return 1;
114 }
115 break;
116
117 case bcond_op:
118 switch (MIPSInst_RT(*i)) {
119 case bltz_op:
120 case bgez_op:
121 case bltzl_op:
122 case bgezl_op:
123 case bltzal_op:
124 case bgezal_op:
125 case bltzall_op:
126 case bgezall_op:
127 return 1;
128 }
129 break;
130
131 case j_op:
132 case jal_op:
133 case jalx_op:
134 case beq_op:
135 case bne_op:
136 case blez_op:
137 case bgtz_op:
138 case beql_op:
139 case bnel_op:
140 case blezl_op:
141 case bgtzl_op:
142 return 1;
143
144 case cop0_op:
145 case cop1_op:
146 case cop2_op:
147 case cop1x_op:
148 if (MIPSInst_RS(*i) == bc_op)
149 return 1;
150 break;
151 }
152
153 return 0;
154}
155
156/*
157 * In the Linux kernel, we support selection of FPR format on the
158 * basis of the Status.FR bit. This does imply that, if a full 32
159 * FPRs are desired, there needs to be a flip-flop that can be written
160 * to one at that bit position. In any case, O32 MIPS ABI uses
161 * only the even FPRs (Status.FR = 0).
162 */
163
164#define CP0_STATUS_FR_SUPPORT
165
166#ifdef CP0_STATUS_FR_SUPPORT
167#define FR_BIT ST0_FR
168#else
169#define FR_BIT 0
170#endif
171
172#define SIFROMREG(si,x) ((si) = \
173 (xcp->cp0_status & FR_BIT) || !(x & 1) ? \
174 (int)ctx->fpr[x] : \
175 (int)(ctx->fpr[x & ~1] >> 32 ))
176#define SITOREG(si,x) (ctx->fpr[x & ~((xcp->cp0_status & FR_BIT) == 0)] = \
177 (xcp->cp0_status & FR_BIT) || !(x & 1) ? \
178 ctx->fpr[x & ~1] >> 32 << 32 | (u32)(si) : \
179 ctx->fpr[x & ~1] << 32 >> 32 | (u64)(si) << 32)
180
181#define DIFROMREG(di,x) ((di) = \
182 ctx->fpr[x & ~((xcp->cp0_status & FR_BIT) == 0)])
183#define DITOREG(di,x) (ctx->fpr[x & ~((xcp->cp0_status & FR_BIT) == 0)] \
184 = (di))
185
186#define SPFROMREG(sp,x) SIFROMREG((sp).bits,x)
187#define SPTOREG(sp,x) SITOREG((sp).bits,x)
188#define DPFROMREG(dp,x) DIFROMREG((dp).bits,x)
189#define DPTOREG(dp,x) DITOREG((dp).bits,x)
190
191/*
192 * Emulate the single floating point instruction pointed at by EPC.
193 * Two instructions if the instruction is in a branch delay slot.
194 */
195
196static int cop1Emulate(struct pt_regs *xcp, struct mips_fpu_soft_struct *ctx)
197{
198 mips_instruction ir;
199 vaddr_t emulpc, contpc;
200 unsigned int cond;
201
202 if (get_user(ir, (mips_instruction *) xcp->cp0_epc)) {
203 fpuemuprivate.stats.errors++;
204 return SIGBUS;
205 }
206
207 /* XXX NEC Vr54xx bug workaround */
208 if ((xcp->cp0_cause & CAUSEF_BD) && !isBranchInstr(&ir))
209 xcp->cp0_cause &= ~CAUSEF_BD;
210
211 if (xcp->cp0_cause & CAUSEF_BD) {
212 /*
213 * The instruction to be emulated is in a branch delay slot
214 * which means that we have to emulate the branch instruction
215 * BEFORE we do the cop1 instruction.
216 *
217 * This branch could be a COP1 branch, but in that case we
218 * would have had a trap for that instruction, and would not
219 * come through this route.
220 *
221 * Linux MIPS branch emulator operates on context, updating the
222 * cp0_epc.
223 */
224 emulpc = REG_TO_VA(xcp->cp0_epc + 4); /* Snapshot emulation target */
225
226 if (__compute_return_epc(xcp)) {
227#ifdef CP1DBG
228 printk("failed to emulate branch at %p\n",
229 REG_TO_VA(xcp->cp0_epc));
230#endif
231 return SIGILL;
232 }
233 if (get_user(ir, (mips_instruction *) emulpc)) {
234 fpuemuprivate.stats.errors++;
235 return SIGBUS;
236 }
237 /* __compute_return_epc() will have updated cp0_epc */
238 contpc = REG_TO_VA xcp->cp0_epc;
239 /* In order not to confuse ptrace() et al, tweak context */
240 xcp->cp0_epc = VA_TO_REG emulpc - 4;
241 }
242 else {
243 emulpc = REG_TO_VA xcp->cp0_epc;
244 contpc = REG_TO_VA(xcp->cp0_epc + 4);
245 }
246
247 emul:
248 fpuemuprivate.stats.emulated++;
249 switch (MIPSInst_OPCODE(ir)) {
250#ifndef SINGLE_ONLY_FPU
251 case ldc1_op:{
252 u64 *va = REG_TO_VA(xcp->regs[MIPSInst_RS(ir)] +
253 MIPSInst_SIMM(ir));
254 u64 val;
255
256 fpuemuprivate.stats.loads++;
257 if (get_user(val, va)) {
258 fpuemuprivate.stats.errors++;
259 return SIGBUS;
260 }
261 DITOREG(val, MIPSInst_RT(ir));
262 break;
263 }
264
265 case sdc1_op:{
266 u64 *va = REG_TO_VA(xcp->regs[MIPSInst_RS(ir)] +
267 MIPSInst_SIMM(ir));
268 u64 val;
269
270 fpuemuprivate.stats.stores++;
271 DIFROMREG(val, MIPSInst_RT(ir));
272 if (put_user(val, va)) {
273 fpuemuprivate.stats.errors++;
274 return SIGBUS;
275 }
276 break;
277 }
278#endif
279
280 case lwc1_op:{
281 u32 *va = REG_TO_VA(xcp->regs[MIPSInst_RS(ir)] +
282 MIPSInst_SIMM(ir));
283 u32 val;
284
285 fpuemuprivate.stats.loads++;
286 if (get_user(val, va)) {
287 fpuemuprivate.stats.errors++;
288 return SIGBUS;
289 }
290#ifdef SINGLE_ONLY_FPU
291 if (MIPSInst_RT(ir) & 1) {
292 /* illegal register in single-float mode */
293 return SIGILL;
294 }
295#endif
296 SITOREG(val, MIPSInst_RT(ir));
297 break;
298 }
299
300 case swc1_op:{
301 u32 *va = REG_TO_VA(xcp->regs[MIPSInst_RS(ir)] +
302 MIPSInst_SIMM(ir));
303 u32 val;
304
305 fpuemuprivate.stats.stores++;
306#ifdef SINGLE_ONLY_FPU
307 if (MIPSInst_RT(ir) & 1) {
308 /* illegal register in single-float mode */
309 return SIGILL;
310 }
311#endif
312 SIFROMREG(val, MIPSInst_RT(ir));
313 if (put_user(val, va)) {
314 fpuemuprivate.stats.errors++;
315 return SIGBUS;
316 }
317 break;
318 }
319
320 case cop1_op:
321 switch (MIPSInst_RS(ir)) {
322
Yoichi Yuasa766160c2005-09-03 15:56:22 -0700323#if defined(__mips64) && !defined(SINGLE_ONLY_FPU)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 case dmfc_op:
325 /* copregister fs -> gpr[rt] */
326 if (MIPSInst_RT(ir) != 0) {
327 DIFROMREG(xcp->regs[MIPSInst_RT(ir)],
328 MIPSInst_RD(ir));
329 }
330 break;
331
332 case dmtc_op:
333 /* copregister fs <- rt */
334 DITOREG(xcp->regs[MIPSInst_RT(ir)], MIPSInst_RD(ir));
335 break;
336#endif
337
338 case mfc_op:
339 /* copregister rd -> gpr[rt] */
340#ifdef SINGLE_ONLY_FPU
341 if (MIPSInst_RD(ir) & 1) {
342 /* illegal register in single-float mode */
343 return SIGILL;
344 }
345#endif
346 if (MIPSInst_RT(ir) != 0) {
347 SIFROMREG(xcp->regs[MIPSInst_RT(ir)],
348 MIPSInst_RD(ir));
349 }
350 break;
351
352 case mtc_op:
353 /* copregister rd <- rt */
354#ifdef SINGLE_ONLY_FPU
355 if (MIPSInst_RD(ir) & 1) {
356 /* illegal register in single-float mode */
357 return SIGILL;
358 }
359#endif
360 SITOREG(xcp->regs[MIPSInst_RT(ir)], MIPSInst_RD(ir));
361 break;
362
363 case cfc_op:{
364 /* cop control register rd -> gpr[rt] */
365 u32 value;
366
367 if (ir == CP1UNDEF) {
368 return do_dsemulret(xcp);
369 }
370 if (MIPSInst_RD(ir) == FPCREG_CSR) {
371 value = ctx->fcr31;
372#ifdef CSRTRACE
373 printk("%p gpr[%d]<-csr=%08x\n",
374 REG_TO_VA(xcp->cp0_epc),
375 MIPSInst_RT(ir), value);
376#endif
377 }
378 else if (MIPSInst_RD(ir) == FPCREG_RID)
379 value = 0;
380 else
381 value = 0;
382 if (MIPSInst_RT(ir))
383 xcp->regs[MIPSInst_RT(ir)] = value;
384 break;
385 }
386
387 case ctc_op:{
388 /* copregister rd <- rt */
389 u32 value;
390
391 if (MIPSInst_RT(ir) == 0)
392 value = 0;
393 else
394 value = xcp->regs[MIPSInst_RT(ir)];
395
396 /* we only have one writable control reg
397 */
398 if (MIPSInst_RD(ir) == FPCREG_CSR) {
399#ifdef CSRTRACE
400 printk("%p gpr[%d]->csr=%08x\n",
401 REG_TO_VA(xcp->cp0_epc),
402 MIPSInst_RT(ir), value);
403#endif
404 ctx->fcr31 = value;
405 /* copy new rounding mode and
406 flush bit to ieee library state! */
407 ieee754_csr.nod = (ctx->fcr31 & 0x1000000) != 0;
408 ieee754_csr.rm = ieee_rm[value & 0x3];
409 }
410 if ((ctx->fcr31 >> 5) & ctx->fcr31 & FPU_CSR_ALL_E) {
411 return SIGFPE;
412 }
413 break;
414 }
415
416 case bc_op:{
417 int likely = 0;
418
419 if (xcp->cp0_cause & CAUSEF_BD)
420 return SIGILL;
421
422#if __mips >= 4
423 cond = ctx->fcr31 & fpucondbit[MIPSInst_RT(ir) >> 2];
424#else
425 cond = ctx->fcr31 & FPU_CSR_COND;
426#endif
427 switch (MIPSInst_RT(ir) & 3) {
428 case bcfl_op:
429 likely = 1;
430 case bcf_op:
431 cond = !cond;
432 break;
433 case bctl_op:
434 likely = 1;
435 case bct_op:
436 break;
437 default:
438 /* thats an illegal instruction */
439 return SIGILL;
440 }
441
442 xcp->cp0_cause |= CAUSEF_BD;
443 if (cond) {
444 /* branch taken: emulate dslot
445 * instruction
446 */
447 xcp->cp0_epc += 4;
448 contpc = REG_TO_VA
449 (xcp->cp0_epc +
450 (MIPSInst_SIMM(ir) << 2));
451
452 if (get_user(ir, (mips_instruction *)
453 REG_TO_VA xcp->cp0_epc)) {
454 fpuemuprivate.stats.errors++;
455 return SIGBUS;
456 }
457
458 switch (MIPSInst_OPCODE(ir)) {
459 case lwc1_op:
460 case swc1_op:
461#if (__mips >= 2 || __mips64) && !defined(SINGLE_ONLY_FPU)
462 case ldc1_op:
463 case sdc1_op:
464#endif
465 case cop1_op:
466#if __mips >= 4 && __mips != 32
467 case cop1x_op:
468#endif
469 /* its one of ours */
470 goto emul;
471#if __mips >= 4
472 case spec_op:
473 if (MIPSInst_FUNC(ir) == movc_op)
474 goto emul;
475 break;
476#endif
477 }
478
479 /*
480 * Single step the non-cp1
481 * instruction in the dslot
482 */
483 return mips_dsemul(xcp, ir, VA_TO_REG contpc);
484 }
485 else {
486 /* branch not taken */
487 if (likely) {
488 /*
489 * branch likely nullifies
490 * dslot if not taken
491 */
492 xcp->cp0_epc += 4;
493 contpc += 4;
494 /*
495 * else continue & execute
496 * dslot as normal insn
497 */
498 }
499 }
500 break;
501 }
502
503 default:
504 if (!(MIPSInst_RS(ir) & 0x10))
505 return SIGILL;
506 {
507 int sig;
508
509 /* a real fpu computation instruction */
510 if ((sig = fpu_emu(xcp, ctx, ir)))
511 return sig;
512 }
513 }
514 break;
515
516#if __mips >= 4 && __mips != 32
517 case cop1x_op:{
518 int sig;
519
520 if ((sig = fpux_emu(xcp, ctx, ir)))
521 return sig;
522 break;
523 }
524#endif
525
526#if __mips >= 4
527 case spec_op:
528 if (MIPSInst_FUNC(ir) != movc_op)
529 return SIGILL;
530 cond = fpucondbit[MIPSInst_RT(ir) >> 2];
531 if (((ctx->fcr31 & cond) != 0) == ((MIPSInst_RT(ir) & 1) != 0))
532 xcp->regs[MIPSInst_RD(ir)] =
533 xcp->regs[MIPSInst_RS(ir)];
534 break;
535#endif
536
537 default:
538 return SIGILL;
539 }
540
541 /* we did it !! */
542 xcp->cp0_epc = VA_TO_REG(contpc);
543 xcp->cp0_cause &= ~CAUSEF_BD;
544 return 0;
545}
546
547/*
548 * Conversion table from MIPS compare ops 48-63
549 * cond = ieee754dp_cmp(x,y,IEEE754_UN,sig);
550 */
551static const unsigned char cmptab[8] = {
552 0, /* cmp_0 (sig) cmp_sf */
553 IEEE754_CUN, /* cmp_un (sig) cmp_ngle */
554 IEEE754_CEQ, /* cmp_eq (sig) cmp_seq */
555 IEEE754_CEQ | IEEE754_CUN, /* cmp_ueq (sig) cmp_ngl */
556 IEEE754_CLT, /* cmp_olt (sig) cmp_lt */
557 IEEE754_CLT | IEEE754_CUN, /* cmp_ult (sig) cmp_nge */
558 IEEE754_CLT | IEEE754_CEQ, /* cmp_ole (sig) cmp_le */
559 IEEE754_CLT | IEEE754_CEQ | IEEE754_CUN, /* cmp_ule (sig) cmp_ngt */
560};
561
562
563#if __mips >= 4 && __mips != 32
564
565/*
566 * Additional MIPS4 instructions
567 */
568
569#define DEF3OP(name, p, f1, f2, f3) \
570static ieee754##p fpemu_##p##_##name (ieee754##p r, ieee754##p s, \
571 ieee754##p t) \
572{ \
573 struct ieee754_csr ieee754_csr_save; \
574 s = f1 (s, t); \
575 ieee754_csr_save = ieee754_csr; \
576 s = f2 (s, r); \
577 ieee754_csr_save.cx |= ieee754_csr.cx; \
578 ieee754_csr_save.sx |= ieee754_csr.sx; \
579 s = f3 (s); \
580 ieee754_csr.cx |= ieee754_csr_save.cx; \
581 ieee754_csr.sx |= ieee754_csr_save.sx; \
582 return s; \
583}
584
585static ieee754dp fpemu_dp_recip(ieee754dp d)
586{
587 return ieee754dp_div(ieee754dp_one(0), d);
588}
589
590static ieee754dp fpemu_dp_rsqrt(ieee754dp d)
591{
592 return ieee754dp_div(ieee754dp_one(0), ieee754dp_sqrt(d));
593}
594
595static ieee754sp fpemu_sp_recip(ieee754sp s)
596{
597 return ieee754sp_div(ieee754sp_one(0), s);
598}
599
600static ieee754sp fpemu_sp_rsqrt(ieee754sp s)
601{
602 return ieee754sp_div(ieee754sp_one(0), ieee754sp_sqrt(s));
603}
604
605DEF3OP(madd, sp, ieee754sp_mul, ieee754sp_add,);
606DEF3OP(msub, sp, ieee754sp_mul, ieee754sp_sub,);
607DEF3OP(nmadd, sp, ieee754sp_mul, ieee754sp_add, ieee754sp_neg);
608DEF3OP(nmsub, sp, ieee754sp_mul, ieee754sp_sub, ieee754sp_neg);
609DEF3OP(madd, dp, ieee754dp_mul, ieee754dp_add,);
610DEF3OP(msub, dp, ieee754dp_mul, ieee754dp_sub,);
611DEF3OP(nmadd, dp, ieee754dp_mul, ieee754dp_add, ieee754dp_neg);
612DEF3OP(nmsub, dp, ieee754dp_mul, ieee754dp_sub, ieee754dp_neg);
613
614static int fpux_emu(struct pt_regs *xcp, struct mips_fpu_soft_struct *ctx,
615 mips_instruction ir)
616{
617 unsigned rcsr = 0; /* resulting csr */
618
619 fpuemuprivate.stats.cp1xops++;
620
621 switch (MIPSInst_FMA_FFMT(ir)) {
622 case s_fmt:{ /* 0 */
623
624 ieee754sp(*handler) (ieee754sp, ieee754sp, ieee754sp);
625 ieee754sp fd, fr, fs, ft;
626 u32 *va;
627 u32 val;
628
629 switch (MIPSInst_FUNC(ir)) {
630 case lwxc1_op:
631 va = REG_TO_VA(xcp->regs[MIPSInst_FR(ir)] +
632 xcp->regs[MIPSInst_FT(ir)]);
633
634 fpuemuprivate.stats.loads++;
635 if (get_user(val, va)) {
636 fpuemuprivate.stats.errors++;
637 return SIGBUS;
638 }
639#ifdef SINGLE_ONLY_FPU
640 if (MIPSInst_FD(ir) & 1) {
641 /* illegal register in single-float
642 * mode
643 */
644 return SIGILL;
645 }
646#endif
647 SITOREG(val, MIPSInst_FD(ir));
648 break;
649
650 case swxc1_op:
651 va = REG_TO_VA(xcp->regs[MIPSInst_FR(ir)] +
652 xcp->regs[MIPSInst_FT(ir)]);
653
654 fpuemuprivate.stats.stores++;
655#ifdef SINGLE_ONLY_FPU
656 if (MIPSInst_FS(ir) & 1) {
657 /* illegal register in single-float
658 * mode
659 */
660 return SIGILL;
661 }
662#endif
663
664 SIFROMREG(val, MIPSInst_FS(ir));
665 if (put_user(val, va)) {
666 fpuemuprivate.stats.errors++;
667 return SIGBUS;
668 }
669 break;
670
671 case madd_s_op:
672 handler = fpemu_sp_madd;
673 goto scoptop;
674 case msub_s_op:
675 handler = fpemu_sp_msub;
676 goto scoptop;
677 case nmadd_s_op:
678 handler = fpemu_sp_nmadd;
679 goto scoptop;
680 case nmsub_s_op:
681 handler = fpemu_sp_nmsub;
682 goto scoptop;
683
684 scoptop:
685 SPFROMREG(fr, MIPSInst_FR(ir));
686 SPFROMREG(fs, MIPSInst_FS(ir));
687 SPFROMREG(ft, MIPSInst_FT(ir));
688 fd = (*handler) (fr, fs, ft);
689 SPTOREG(fd, MIPSInst_FD(ir));
690
691 copcsr:
692 if (ieee754_cxtest(IEEE754_INEXACT))
693 rcsr |= FPU_CSR_INE_X | FPU_CSR_INE_S;
694 if (ieee754_cxtest(IEEE754_UNDERFLOW))
695 rcsr |= FPU_CSR_UDF_X | FPU_CSR_UDF_S;
696 if (ieee754_cxtest(IEEE754_OVERFLOW))
697 rcsr |= FPU_CSR_OVF_X | FPU_CSR_OVF_S;
698 if (ieee754_cxtest(IEEE754_INVALID_OPERATION))
699 rcsr |= FPU_CSR_INV_X | FPU_CSR_INV_S;
700
701 ctx->fcr31 = (ctx->fcr31 & ~FPU_CSR_ALL_X) | rcsr;
702 if (ieee754_csr.nod)
703 ctx->fcr31 |= 0x1000000;
704 if ((ctx->fcr31 >> 5) & ctx->fcr31 & FPU_CSR_ALL_E) {
705 /*printk ("SIGFPE: fpu csr = %08x\n",
706 ctx->fcr31); */
707 return SIGFPE;
708 }
709
710 break;
711
712 default:
713 return SIGILL;
714 }
715 break;
716 }
717
718#ifndef SINGLE_ONLY_FPU
719 case d_fmt:{ /* 1 */
720 ieee754dp(*handler) (ieee754dp, ieee754dp, ieee754dp);
721 ieee754dp fd, fr, fs, ft;
722 u64 *va;
723 u64 val;
724
725 switch (MIPSInst_FUNC(ir)) {
726 case ldxc1_op:
727 va = REG_TO_VA(xcp->regs[MIPSInst_FR(ir)] +
728 xcp->regs[MIPSInst_FT(ir)]);
729
730 fpuemuprivate.stats.loads++;
731 if (get_user(val, va)) {
732 fpuemuprivate.stats.errors++;
733 return SIGBUS;
734 }
735 DITOREG(val, MIPSInst_FD(ir));
736 break;
737
738 case sdxc1_op:
739 va = REG_TO_VA(xcp->regs[MIPSInst_FR(ir)] +
740 xcp->regs[MIPSInst_FT(ir)]);
741
742 fpuemuprivate.stats.stores++;
743 DIFROMREG(val, MIPSInst_FS(ir));
744 if (put_user(val, va)) {
745 fpuemuprivate.stats.errors++;
746 return SIGBUS;
747 }
748 break;
749
750 case madd_d_op:
751 handler = fpemu_dp_madd;
752 goto dcoptop;
753 case msub_d_op:
754 handler = fpemu_dp_msub;
755 goto dcoptop;
756 case nmadd_d_op:
757 handler = fpemu_dp_nmadd;
758 goto dcoptop;
759 case nmsub_d_op:
760 handler = fpemu_dp_nmsub;
761 goto dcoptop;
762
763 dcoptop:
764 DPFROMREG(fr, MIPSInst_FR(ir));
765 DPFROMREG(fs, MIPSInst_FS(ir));
766 DPFROMREG(ft, MIPSInst_FT(ir));
767 fd = (*handler) (fr, fs, ft);
768 DPTOREG(fd, MIPSInst_FD(ir));
769 goto copcsr;
770
771 default:
772 return SIGILL;
773 }
774 break;
775 }
776#endif
777
778 case 0x7: /* 7 */
779 if (MIPSInst_FUNC(ir) != pfetch_op) {
780 return SIGILL;
781 }
782 /* ignore prefx operation */
783 break;
784
785 default:
786 return SIGILL;
787 }
788
789 return 0;
790}
791#endif
792
793
794
795/*
796 * Emulate a single COP1 arithmetic instruction.
797 */
798static int fpu_emu(struct pt_regs *xcp, struct mips_fpu_soft_struct *ctx,
799 mips_instruction ir)
800{
801 int rfmt; /* resulting format */
802 unsigned rcsr = 0; /* resulting csr */
803 unsigned cond;
804 union {
805 ieee754dp d;
806 ieee754sp s;
807 int w;
Yoichi Yuasa766160c2005-09-03 15:56:22 -0700808#ifdef __mips64
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809 s64 l;
810#endif
811 } rv; /* resulting value */
812
813 fpuemuprivate.stats.cp1ops++;
814 switch (rfmt = (MIPSInst_FFMT(ir) & 0xf)) {
815 case s_fmt:{ /* 0 */
816 union {
817 ieee754sp(*b) (ieee754sp, ieee754sp);
818 ieee754sp(*u) (ieee754sp);
819 } handler;
820
821 switch (MIPSInst_FUNC(ir)) {
822 /* binary ops */
823 case fadd_op:
824 handler.b = ieee754sp_add;
825 goto scopbop;
826 case fsub_op:
827 handler.b = ieee754sp_sub;
828 goto scopbop;
829 case fmul_op:
830 handler.b = ieee754sp_mul;
831 goto scopbop;
832 case fdiv_op:
833 handler.b = ieee754sp_div;
834 goto scopbop;
835
836 /* unary ops */
837#if __mips >= 2 || __mips64
838 case fsqrt_op:
839 handler.u = ieee754sp_sqrt;
840 goto scopuop;
841#endif
842#if __mips >= 4 && __mips != 32
843 case frsqrt_op:
844 handler.u = fpemu_sp_rsqrt;
845 goto scopuop;
846 case frecip_op:
847 handler.u = fpemu_sp_recip;
848 goto scopuop;
849#endif
850#if __mips >= 4
851 case fmovc_op:
852 cond = fpucondbit[MIPSInst_FT(ir) >> 2];
853 if (((ctx->fcr31 & cond) != 0) !=
854 ((MIPSInst_FT(ir) & 1) != 0))
855 return 0;
856 SPFROMREG(rv.s, MIPSInst_FS(ir));
857 break;
858 case fmovz_op:
859 if (xcp->regs[MIPSInst_FT(ir)] != 0)
860 return 0;
861 SPFROMREG(rv.s, MIPSInst_FS(ir));
862 break;
863 case fmovn_op:
864 if (xcp->regs[MIPSInst_FT(ir)] == 0)
865 return 0;
866 SPFROMREG(rv.s, MIPSInst_FS(ir));
867 break;
868#endif
869 case fabs_op:
870 handler.u = ieee754sp_abs;
871 goto scopuop;
872 case fneg_op:
873 handler.u = ieee754sp_neg;
874 goto scopuop;
875 case fmov_op:
876 /* an easy one */
877 SPFROMREG(rv.s, MIPSInst_FS(ir));
878 goto copcsr;
879
880 /* binary op on handler */
881 scopbop:
882 {
883 ieee754sp fs, ft;
884
885 SPFROMREG(fs, MIPSInst_FS(ir));
886 SPFROMREG(ft, MIPSInst_FT(ir));
887
888 rv.s = (*handler.b) (fs, ft);
889 goto copcsr;
890 }
891 scopuop:
892 {
893 ieee754sp fs;
894
895 SPFROMREG(fs, MIPSInst_FS(ir));
896 rv.s = (*handler.u) (fs);
897 goto copcsr;
898 }
899 copcsr:
900 if (ieee754_cxtest(IEEE754_INEXACT))
901 rcsr |= FPU_CSR_INE_X | FPU_CSR_INE_S;
902 if (ieee754_cxtest(IEEE754_UNDERFLOW))
903 rcsr |= FPU_CSR_UDF_X | FPU_CSR_UDF_S;
904 if (ieee754_cxtest(IEEE754_OVERFLOW))
905 rcsr |= FPU_CSR_OVF_X | FPU_CSR_OVF_S;
906 if (ieee754_cxtest(IEEE754_ZERO_DIVIDE))
907 rcsr |= FPU_CSR_DIV_X | FPU_CSR_DIV_S;
908 if (ieee754_cxtest(IEEE754_INVALID_OPERATION))
909 rcsr |= FPU_CSR_INV_X | FPU_CSR_INV_S;
910 break;
911
912 /* unary conv ops */
913 case fcvts_op:
914 return SIGILL; /* not defined */
915 case fcvtd_op:{
916#ifdef SINGLE_ONLY_FPU
917 return SIGILL; /* not defined */
918#else
919 ieee754sp fs;
920
921 SPFROMREG(fs, MIPSInst_FS(ir));
922 rv.d = ieee754dp_fsp(fs);
923 rfmt = d_fmt;
924 goto copcsr;
925 }
926#endif
927 case fcvtw_op:{
928 ieee754sp fs;
929
930 SPFROMREG(fs, MIPSInst_FS(ir));
931 rv.w = ieee754sp_tint(fs);
932 rfmt = w_fmt;
933 goto copcsr;
934 }
935
936#if __mips >= 2 || __mips64
937 case fround_op:
938 case ftrunc_op:
939 case fceil_op:
940 case ffloor_op:{
941 unsigned int oldrm = ieee754_csr.rm;
942 ieee754sp fs;
943
944 SPFROMREG(fs, MIPSInst_FS(ir));
945 ieee754_csr.rm = ieee_rm[MIPSInst_FUNC(ir) & 0x3];
946 rv.w = ieee754sp_tint(fs);
947 ieee754_csr.rm = oldrm;
948 rfmt = w_fmt;
949 goto copcsr;
950 }
951#endif /* __mips >= 2 */
952
Yoichi Yuasa766160c2005-09-03 15:56:22 -0700953#if defined(__mips64) && !defined(SINGLE_ONLY_FPU)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954 case fcvtl_op:{
955 ieee754sp fs;
956
957 SPFROMREG(fs, MIPSInst_FS(ir));
958 rv.l = ieee754sp_tlong(fs);
959 rfmt = l_fmt;
960 goto copcsr;
961 }
962
963 case froundl_op:
964 case ftruncl_op:
965 case fceill_op:
966 case ffloorl_op:{
967 unsigned int oldrm = ieee754_csr.rm;
968 ieee754sp fs;
969
970 SPFROMREG(fs, MIPSInst_FS(ir));
971 ieee754_csr.rm = ieee_rm[MIPSInst_FUNC(ir) & 0x3];
972 rv.l = ieee754sp_tlong(fs);
973 ieee754_csr.rm = oldrm;
974 rfmt = l_fmt;
975 goto copcsr;
976 }
977#endif /* __mips64 && !fpu(single) */
978
979 default:
980 if (MIPSInst_FUNC(ir) >= fcmp_op) {
981 unsigned cmpop = MIPSInst_FUNC(ir) - fcmp_op;
982 ieee754sp fs, ft;
983
984 SPFROMREG(fs, MIPSInst_FS(ir));
985 SPFROMREG(ft, MIPSInst_FT(ir));
986 rv.w = ieee754sp_cmp(fs, ft,
987 cmptab[cmpop & 0x7], cmpop & 0x8);
988 rfmt = -1;
989 if ((cmpop & 0x8) && ieee754_cxtest
990 (IEEE754_INVALID_OPERATION))
991 rcsr = FPU_CSR_INV_X | FPU_CSR_INV_S;
992 else
993 goto copcsr;
994
995 }
996 else {
997 return SIGILL;
998 }
999 break;
1000 }
1001 break;
1002 }
1003
1004#ifndef SINGLE_ONLY_FPU
1005 case d_fmt:{
1006 union {
1007 ieee754dp(*b) (ieee754dp, ieee754dp);
1008 ieee754dp(*u) (ieee754dp);
1009 } handler;
1010
1011 switch (MIPSInst_FUNC(ir)) {
1012 /* binary ops */
1013 case fadd_op:
1014 handler.b = ieee754dp_add;
1015 goto dcopbop;
1016 case fsub_op:
1017 handler.b = ieee754dp_sub;
1018 goto dcopbop;
1019 case fmul_op:
1020 handler.b = ieee754dp_mul;
1021 goto dcopbop;
1022 case fdiv_op:
1023 handler.b = ieee754dp_div;
1024 goto dcopbop;
1025
1026 /* unary ops */
1027#if __mips >= 2 || __mips64
1028 case fsqrt_op:
1029 handler.u = ieee754dp_sqrt;
1030 goto dcopuop;
1031#endif
1032#if __mips >= 4 && __mips != 32
1033 case frsqrt_op:
1034 handler.u = fpemu_dp_rsqrt;
1035 goto dcopuop;
1036 case frecip_op:
1037 handler.u = fpemu_dp_recip;
1038 goto dcopuop;
1039#endif
1040#if __mips >= 4
1041 case fmovc_op:
1042 cond = fpucondbit[MIPSInst_FT(ir) >> 2];
1043 if (((ctx->fcr31 & cond) != 0) !=
1044 ((MIPSInst_FT(ir) & 1) != 0))
1045 return 0;
1046 DPFROMREG(rv.d, MIPSInst_FS(ir));
1047 break;
1048 case fmovz_op:
1049 if (xcp->regs[MIPSInst_FT(ir)] != 0)
1050 return 0;
1051 DPFROMREG(rv.d, MIPSInst_FS(ir));
1052 break;
1053 case fmovn_op:
1054 if (xcp->regs[MIPSInst_FT(ir)] == 0)
1055 return 0;
1056 DPFROMREG(rv.d, MIPSInst_FS(ir));
1057 break;
1058#endif
1059 case fabs_op:
1060 handler.u = ieee754dp_abs;
1061 goto dcopuop;
1062
1063 case fneg_op:
1064 handler.u = ieee754dp_neg;
1065 goto dcopuop;
1066
1067 case fmov_op:
1068 /* an easy one */
1069 DPFROMREG(rv.d, MIPSInst_FS(ir));
1070 goto copcsr;
1071
1072 /* binary op on handler */
1073 dcopbop:{
1074 ieee754dp fs, ft;
1075
1076 DPFROMREG(fs, MIPSInst_FS(ir));
1077 DPFROMREG(ft, MIPSInst_FT(ir));
1078
1079 rv.d = (*handler.b) (fs, ft);
1080 goto copcsr;
1081 }
1082 dcopuop:{
1083 ieee754dp fs;
1084
1085 DPFROMREG(fs, MIPSInst_FS(ir));
1086 rv.d = (*handler.u) (fs);
1087 goto copcsr;
1088 }
1089
1090 /* unary conv ops */
1091 case fcvts_op:{
1092 ieee754dp fs;
1093
1094 DPFROMREG(fs, MIPSInst_FS(ir));
1095 rv.s = ieee754sp_fdp(fs);
1096 rfmt = s_fmt;
1097 goto copcsr;
1098 }
1099 case fcvtd_op:
1100 return SIGILL; /* not defined */
1101
1102 case fcvtw_op:{
1103 ieee754dp fs;
1104
1105 DPFROMREG(fs, MIPSInst_FS(ir));
1106 rv.w = ieee754dp_tint(fs); /* wrong */
1107 rfmt = w_fmt;
1108 goto copcsr;
1109 }
1110
1111#if __mips >= 2 || __mips64
1112 case fround_op:
1113 case ftrunc_op:
1114 case fceil_op:
1115 case ffloor_op:{
1116 unsigned int oldrm = ieee754_csr.rm;
1117 ieee754dp fs;
1118
1119 DPFROMREG(fs, MIPSInst_FS(ir));
1120 ieee754_csr.rm = ieee_rm[MIPSInst_FUNC(ir) & 0x3];
1121 rv.w = ieee754dp_tint(fs);
1122 ieee754_csr.rm = oldrm;
1123 rfmt = w_fmt;
1124 goto copcsr;
1125 }
1126#endif
1127
Yoichi Yuasa766160c2005-09-03 15:56:22 -07001128#if defined(__mips64) && !defined(SINGLE_ONLY_FPU)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129 case fcvtl_op:{
1130 ieee754dp fs;
1131
1132 DPFROMREG(fs, MIPSInst_FS(ir));
1133 rv.l = ieee754dp_tlong(fs);
1134 rfmt = l_fmt;
1135 goto copcsr;
1136 }
1137
1138 case froundl_op:
1139 case ftruncl_op:
1140 case fceill_op:
1141 case ffloorl_op:{
1142 unsigned int oldrm = ieee754_csr.rm;
1143 ieee754dp fs;
1144
1145 DPFROMREG(fs, MIPSInst_FS(ir));
1146 ieee754_csr.rm = ieee_rm[MIPSInst_FUNC(ir) & 0x3];
1147 rv.l = ieee754dp_tlong(fs);
1148 ieee754_csr.rm = oldrm;
1149 rfmt = l_fmt;
1150 goto copcsr;
1151 }
1152#endif /* __mips >= 3 && !fpu(single) */
1153
1154 default:
1155 if (MIPSInst_FUNC(ir) >= fcmp_op) {
1156 unsigned cmpop = MIPSInst_FUNC(ir) - fcmp_op;
1157 ieee754dp fs, ft;
1158
1159 DPFROMREG(fs, MIPSInst_FS(ir));
1160 DPFROMREG(ft, MIPSInst_FT(ir));
1161 rv.w = ieee754dp_cmp(fs, ft,
1162 cmptab[cmpop & 0x7], cmpop & 0x8);
1163 rfmt = -1;
1164 if ((cmpop & 0x8)
1165 &&
1166 ieee754_cxtest
1167 (IEEE754_INVALID_OPERATION))
1168 rcsr = FPU_CSR_INV_X | FPU_CSR_INV_S;
1169 else
1170 goto copcsr;
1171
1172 }
1173 else {
1174 return SIGILL;
1175 }
1176 break;
1177 }
1178 break;
1179 }
1180#endif /* ifndef SINGLE_ONLY_FPU */
1181
1182 case w_fmt:{
1183 ieee754sp fs;
1184
1185 switch (MIPSInst_FUNC(ir)) {
1186 case fcvts_op:
1187 /* convert word to single precision real */
1188 SPFROMREG(fs, MIPSInst_FS(ir));
1189 rv.s = ieee754sp_fint(fs.bits);
1190 rfmt = s_fmt;
1191 goto copcsr;
1192#ifndef SINGLE_ONLY_FPU
1193 case fcvtd_op:
1194 /* convert word to double precision real */
1195 SPFROMREG(fs, MIPSInst_FS(ir));
1196 rv.d = ieee754dp_fint(fs.bits);
1197 rfmt = d_fmt;
1198 goto copcsr;
1199#endif
1200 default:
1201 return SIGILL;
1202 }
1203 break;
1204 }
1205
Yoichi Yuasa766160c2005-09-03 15:56:22 -07001206#if defined(__mips64) && !defined(SINGLE_ONLY_FPU)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207 case l_fmt:{
1208 switch (MIPSInst_FUNC(ir)) {
1209 case fcvts_op:
1210 /* convert long to single precision real */
1211 rv.s = ieee754sp_flong(ctx->fpr[MIPSInst_FS(ir)]);
1212 rfmt = s_fmt;
1213 goto copcsr;
1214 case fcvtd_op:
1215 /* convert long to double precision real */
1216 rv.d = ieee754dp_flong(ctx->fpr[MIPSInst_FS(ir)]);
1217 rfmt = d_fmt;
1218 goto copcsr;
1219 default:
1220 return SIGILL;
1221 }
1222 break;
1223 }
1224#endif
1225
1226 default:
1227 return SIGILL;
1228 }
1229
1230 /*
1231 * Update the fpu CSR register for this operation.
1232 * If an exception is required, generate a tidy SIGFPE exception,
1233 * without updating the result register.
1234 * Note: cause exception bits do not accumulate, they are rewritten
1235 * for each op; only the flag/sticky bits accumulate.
1236 */
1237 ctx->fcr31 = (ctx->fcr31 & ~FPU_CSR_ALL_X) | rcsr;
1238 if ((ctx->fcr31 >> 5) & ctx->fcr31 & FPU_CSR_ALL_E) {
1239 /*printk ("SIGFPE: fpu csr = %08x\n",ctx->fcr31); */
1240 return SIGFPE;
1241 }
1242
1243 /*
1244 * Now we can safely write the result back to the register file.
1245 */
1246 switch (rfmt) {
1247 case -1:{
1248#if __mips >= 4
1249 cond = fpucondbit[MIPSInst_FD(ir) >> 2];
1250#else
1251 cond = FPU_CSR_COND;
1252#endif
1253 if (rv.w)
1254 ctx->fcr31 |= cond;
1255 else
1256 ctx->fcr31 &= ~cond;
1257 break;
1258 }
1259#ifndef SINGLE_ONLY_FPU
1260 case d_fmt:
1261 DPTOREG(rv.d, MIPSInst_FD(ir));
1262 break;
1263#endif
1264 case s_fmt:
1265 SPTOREG(rv.s, MIPSInst_FD(ir));
1266 break;
1267 case w_fmt:
1268 SITOREG(rv.w, MIPSInst_FD(ir));
1269 break;
Yoichi Yuasa766160c2005-09-03 15:56:22 -07001270#if defined(__mips64) && !defined(SINGLE_ONLY_FPU)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271 case l_fmt:
1272 DITOREG(rv.l, MIPSInst_FD(ir));
1273 break;
1274#endif
1275 default:
1276 return SIGILL;
1277 }
1278
1279 return 0;
1280}
1281
1282int fpu_emulator_cop1Handler(int xcptno, struct pt_regs *xcp,
1283 struct mips_fpu_soft_struct *ctx)
1284{
1285 gpreg_t oldepc, prevepc;
1286 mips_instruction insn;
1287 int sig = 0;
1288
1289 oldepc = xcp->cp0_epc;
1290 do {
1291 prevepc = xcp->cp0_epc;
1292
1293 if (get_user(insn, (mips_instruction *) xcp->cp0_epc)) {
1294 fpuemuprivate.stats.errors++;
1295 return SIGBUS;
1296 }
1297 if (insn == 0)
1298 xcp->cp0_epc += 4; /* skip nops */
1299 else {
1300 /* Update ieee754_csr. Only relevant if we have a
1301 h/w FPU */
1302 ieee754_csr.nod = (ctx->fcr31 & 0x1000000) != 0;
1303 ieee754_csr.rm = ieee_rm[ctx->fcr31 & 0x3];
1304 ieee754_csr.cx = (ctx->fcr31 >> 12) & 0x1f;
1305 sig = cop1Emulate(xcp, ctx);
1306 }
1307
1308 if (cpu_has_fpu)
1309 break;
1310 if (sig)
1311 break;
1312
1313 cond_resched();
1314 } while (xcp->cp0_epc > prevepc);
1315
1316 /* SIGILL indicates a non-fpu instruction */
1317 if (sig == SIGILL && xcp->cp0_epc != oldepc)
1318 /* but if epc has advanced, then ignore it */
1319 sig = 0;
1320
1321 return sig;
1322}