2 * arch/powerpc/math-emu/math_efp.c
4 * Copyright (C) 2006-2008, 2010 Freescale Semiconductor, Inc.
6 * Author: Ebony Zhu, <ebony.zhu@freescale.com>
7 * Yu Liu, <yu.liu@freescale.com>
9 * Derived from arch/alpha/math-emu/math.c
10 * arch/powerpc/math-emu/math.c
13 * This file is the exception handler to make E500 SPE instructions
14 * fully comply with IEEE-754 floating point standard.
16 * This program is free software; you can redistribute it and/or
17 * modify it under the terms of the GNU General Public License
18 * as published by the Free Software Foundation; either version
19 * 2 of the License, or (at your option) any later version.
22 #include <linux/types.h>
24 #include <asm/uaccess.h>
27 #define FP_EX_BOOKE_E500_SPE
28 #include <asm/sfp-machine.h>
30 #include <math-emu/soft-fp.h>
31 #include <math-emu/single.h>
32 #include <math-emu/double.h>
47 #define EFSCMPGT 0x2cc
48 #define EFSCMPLT 0x2cd
49 #define EFSCMPEQ 0x2ce
56 #define EFSCTUIZ 0x2d8
57 #define EFSCTSIZ 0x2da
62 #define EVFSNABS 0x285
66 #define EVFSCMPGT 0x28c
67 #define EVFSCMPLT 0x28d
68 #define EVFSCMPEQ 0x28e
69 #define EVFSCTUI 0x294
70 #define EVFSCTSI 0x295
71 #define EVFSCTUF 0x296
72 #define EVFSCTSF 0x297
73 #define EVFSCTUIZ 0x298
74 #define EVFSCTSIZ 0x29a
83 #define EFDCTUIDZ 0x2ea
84 #define EFDCTSIDZ 0x2eb
85 #define EFDCMPGT 0x2ec
86 #define EFDCMPLT 0x2ed
87 #define EFDCMPEQ 0x2ee
93 #define EFDCTUIZ 0x2f8
94 #define EFDCTSIZ 0x2fa
102 #define SIGN_BIT_S (1UL << 31)
103 #define SIGN_BIT_D (1ULL << 63)
104 #define FP_EX_MASK (FP_EX_INEXACT | FP_EX_INVALID | FP_EX_DIVZERO | \
105 FP_EX_UNDERFLOW | FP_EX_OVERFLOW)
107 static int have_e500_cpu_a005_erratum
;
114 static unsigned long insn_type(unsigned long speinsn
)
116 unsigned long ret
= NOTYPE
;
118 switch (speinsn
& 0x7ff) {
119 case EFSABS
: ret
= XA
; break;
120 case EFSADD
: ret
= AB
; break;
121 case EFSCFD
: ret
= XB
; break;
122 case EFSCMPEQ
: ret
= XCR
; break;
123 case EFSCMPGT
: ret
= XCR
; break;
124 case EFSCMPLT
: ret
= XCR
; break;
125 case EFSCTSF
: ret
= XB
; break;
126 case EFSCTSI
: ret
= XB
; break;
127 case EFSCTSIZ
: ret
= XB
; break;
128 case EFSCTUF
: ret
= XB
; break;
129 case EFSCTUI
: ret
= XB
; break;
130 case EFSCTUIZ
: ret
= XB
; break;
131 case EFSDIV
: ret
= AB
; break;
132 case EFSMUL
: ret
= AB
; break;
133 case EFSNABS
: ret
= XA
; break;
134 case EFSNEG
: ret
= XA
; break;
135 case EFSSUB
: ret
= AB
; break;
136 case EFSCFSI
: ret
= XB
; break;
138 case EVFSABS
: ret
= XA
; break;
139 case EVFSADD
: ret
= AB
; break;
140 case EVFSCMPEQ
: ret
= XCR
; break;
141 case EVFSCMPGT
: ret
= XCR
; break;
142 case EVFSCMPLT
: ret
= XCR
; break;
143 case EVFSCTSF
: ret
= XB
; break;
144 case EVFSCTSI
: ret
= XB
; break;
145 case EVFSCTSIZ
: ret
= XB
; break;
146 case EVFSCTUF
: ret
= XB
; break;
147 case EVFSCTUI
: ret
= XB
; break;
148 case EVFSCTUIZ
: ret
= XB
; break;
149 case EVFSDIV
: ret
= AB
; break;
150 case EVFSMUL
: ret
= AB
; break;
151 case EVFSNABS
: ret
= XA
; break;
152 case EVFSNEG
: ret
= XA
; break;
153 case EVFSSUB
: ret
= AB
; break;
155 case EFDABS
: ret
= XA
; break;
156 case EFDADD
: ret
= AB
; break;
157 case EFDCFS
: ret
= XB
; break;
158 case EFDCMPEQ
: ret
= XCR
; break;
159 case EFDCMPGT
: ret
= XCR
; break;
160 case EFDCMPLT
: ret
= XCR
; break;
161 case EFDCTSF
: ret
= XB
; break;
162 case EFDCTSI
: ret
= XB
; break;
163 case EFDCTSIDZ
: ret
= XB
; break;
164 case EFDCTSIZ
: ret
= XB
; break;
165 case EFDCTUF
: ret
= XB
; break;
166 case EFDCTUI
: ret
= XB
; break;
167 case EFDCTUIDZ
: ret
= XB
; break;
168 case EFDCTUIZ
: ret
= XB
; break;
169 case EFDDIV
: ret
= AB
; break;
170 case EFDMUL
: ret
= AB
; break;
171 case EFDNABS
: ret
= XA
; break;
172 case EFDNEG
: ret
= XA
; break;
173 case EFDSUB
: ret
= AB
; break;
176 printk(KERN_ERR
"\nOoops! SPE instruction no type found.");
177 printk(KERN_ERR
"\ninst code: %08lx\n", speinsn
);
183 int do_spe_mathemu(struct pt_regs
*regs
)
188 unsigned long type
, func
, fc
, fa
, fb
, src
, speinsn
;
189 union dw_union vc
, va
, vb
;
191 if (get_user(speinsn
, (unsigned int __user
*) regs
->nip
))
193 if ((speinsn
>> 26) != EFAPU
)
194 return -EINVAL
; /* not an spe instruction */
196 type
= insn_type(speinsn
);
200 func
= speinsn
& 0x7ff;
201 fc
= (speinsn
>> 21) & 0x1f;
202 fa
= (speinsn
>> 16) & 0x1f;
203 fb
= (speinsn
>> 11) & 0x1f;
204 src
= (speinsn
>> 5) & 0x7;
206 vc
.wp
[0] = current
->thread
.evr
[fc
];
207 vc
.wp
[1] = regs
->gpr
[fc
];
208 va
.wp
[0] = current
->thread
.evr
[fa
];
209 va
.wp
[1] = regs
->gpr
[fa
];
210 vb
.wp
[0] = current
->thread
.evr
[fb
];
211 vb
.wp
[1] = regs
->gpr
[fb
];
213 __FPU_FPSCR
= mfspr(SPRN_SPEFSCR
);
216 printk("speinsn:%08lx spefscr:%08lx\n", speinsn
, __FPU_FPSCR
);
217 printk("vc: %08x %08x\n", vc
.wp
[0], vc
.wp
[1]);
218 printk("va: %08x %08x\n", va
.wp
[0], va
.wp
[1]);
219 printk("vb: %08x %08x\n", vb
.wp
[0], vb
.wp
[1]);
224 FP_DECL_S(SA
); FP_DECL_S(SB
); FP_DECL_S(SR
);
229 FP_UNPACK_SP(SA
, va
.wp
+ 1);
231 FP_UNPACK_SP(SB
, vb
.wp
+ 1);
234 FP_UNPACK_SP(SA
, va
.wp
+ 1);
239 printk("SA: %ld %08lx %ld (%ld)\n", SA_s
, SA_f
, SA_e
, SA_c
);
240 printk("SB: %ld %08lx %ld (%ld)\n", SB_s
, SB_f
, SB_e
, SB_c
);
245 vc
.wp
[1] = va
.wp
[1] & ~SIGN_BIT_S
;
249 vc
.wp
[1] = va
.wp
[1] | SIGN_BIT_S
;
253 vc
.wp
[1] = va
.wp
[1] ^ SIGN_BIT_S
;
257 FP_ADD_S(SR
, SA
, SB
);
261 FP_SUB_S(SR
, SA
, SB
);
265 FP_MUL_S(SR
, SA
, SB
);
269 FP_DIV_S(SR
, SA
, SB
);
286 if (!((vb
.wp
[1] >> 23) == 0xff && ((vb
.wp
[1] & 0x7fffff) > 0))) {
288 if (((vb
.wp
[1] >> 23) & 0xff) == 0) {
291 } else if ((vb
.wp
[1] >> 31) == 0) {
292 /* positive normal */
293 vc
.wp
[1] = (func
== EFSCTSF
) ?
294 0x7fffffff : 0xffffffff;
295 } else { /* negative normal */
296 vc
.wp
[1] = (func
== EFSCTSF
) ?
299 } else { /* rB is NaN */
307 FP_UNPACK_DP(DB
, vb
.dp
);
309 printk("DB: %ld %08lx %08lx %ld (%ld)\n",
310 DB_s
, DB_f1
, DB_f0
, DB_e
, DB_c
);
312 FP_CONV(S
, D
, 1, 2, SR
, DB
);
323 _FP_ROUND_ZERO(1, SB
);
325 FP_TO_INT_S(vc
.wp
[1], SB
, 32,
326 (((func
& 0x3) != 0) || SB_s
));
336 printk("SR: %ld %08lx %ld (%ld)\n", SR_s
, SR_f
, SR_e
, SR_c
);
338 FP_PACK_SP(vc
.wp
+ 1, SR
);
342 FP_CMP_S(IR
, SA
, SB
, 3);
343 if (IR
== 3 && (FP_ISSIGNAN_S(SA
) || FP_ISSIGNAN_S(SB
)))
344 FP_SET_EXCEPTION(FP_EX_INVALID
);
354 FP_DECL_D(DA
); FP_DECL_D(DB
); FP_DECL_D(DR
);
359 FP_UNPACK_DP(DA
, va
.dp
);
361 FP_UNPACK_DP(DB
, vb
.dp
);
364 FP_UNPACK_DP(DA
, va
.dp
);
369 printk("DA: %ld %08lx %08lx %ld (%ld)\n",
370 DA_s
, DA_f1
, DA_f0
, DA_e
, DA_c
);
371 printk("DB: %ld %08lx %08lx %ld (%ld)\n",
372 DB_s
, DB_f1
, DB_f0
, DB_e
, DB_c
);
377 vc
.dp
[0] = va
.dp
[0] & ~SIGN_BIT_D
;
381 vc
.dp
[0] = va
.dp
[0] | SIGN_BIT_D
;
385 vc
.dp
[0] = va
.dp
[0] ^ SIGN_BIT_D
;
389 FP_ADD_D(DR
, DA
, DB
);
393 FP_SUB_D(DR
, DA
, DB
);
397 FP_MUL_D(DR
, DA
, DB
);
401 FP_DIV_D(DR
, DA
, DB
);
418 if (!((vb
.wp
[0] >> 20) == 0x7ff &&
419 ((vb
.wp
[0] & 0xfffff) > 0 || (vb
.wp
[1] > 0)))) {
421 if (((vb
.wp
[0] >> 20) & 0x7ff) == 0) {
424 } else if ((vb
.wp
[0] >> 31) == 0) {
425 /* positive normal */
426 vc
.wp
[1] = (func
== EFDCTSF
) ?
427 0x7fffffff : 0xffffffff;
428 } else { /* negative normal */
429 vc
.wp
[1] = (func
== EFDCTSF
) ?
440 FP_UNPACK_SP(SB
, vb
.wp
+ 1);
442 printk("SB: %ld %08lx %ld (%ld)\n",
443 SB_s
, SB_f
, SB_e
, SB_c
);
445 FP_CONV(D
, S
, 2, 1, DR
, SB
);
451 _FP_ROUND_ZERO(2, DB
);
452 FP_TO_INT_D(vc
.dp
[0], DB
, 64, ((func
& 0x1) == 0));
462 _FP_ROUND_ZERO(2, DB
);
464 FP_TO_INT_D(vc
.wp
[1], DB
, 32,
465 (((func
& 0x3) != 0) || DB_s
));
475 printk("DR: %ld %08lx %08lx %ld (%ld)\n",
476 DR_s
, DR_f1
, DR_f0
, DR_e
, DR_c
);
478 FP_PACK_DP(vc
.dp
, DR
);
482 FP_CMP_D(IR
, DA
, DB
, 3);
483 if (IR
== 3 && (FP_ISSIGNAN_D(DA
) || FP_ISSIGNAN_D(DB
)))
484 FP_SET_EXCEPTION(FP_EX_INVALID
);
495 FP_DECL_S(SA0
); FP_DECL_S(SB0
); FP_DECL_S(SR0
);
496 FP_DECL_S(SA1
); FP_DECL_S(SB1
); FP_DECL_S(SR1
);
502 FP_UNPACK_SP(SA0
, va
.wp
);
503 FP_UNPACK_SP(SA1
, va
.wp
+ 1);
505 FP_UNPACK_SP(SB0
, vb
.wp
);
506 FP_UNPACK_SP(SB1
, vb
.wp
+ 1);
509 FP_UNPACK_SP(SA0
, va
.wp
);
510 FP_UNPACK_SP(SA1
, va
.wp
+ 1);
515 printk("SA0: %ld %08lx %ld (%ld)\n", SA0_s
, SA0_f
, SA0_e
, SA0_c
);
516 printk("SA1: %ld %08lx %ld (%ld)\n", SA1_s
, SA1_f
, SA1_e
, SA1_c
);
517 printk("SB0: %ld %08lx %ld (%ld)\n", SB0_s
, SB0_f
, SB0_e
, SB0_c
);
518 printk("SB1: %ld %08lx %ld (%ld)\n", SB1_s
, SB1_f
, SB1_e
, SB1_c
);
523 vc
.wp
[0] = va
.wp
[0] & ~SIGN_BIT_S
;
524 vc
.wp
[1] = va
.wp
[1] & ~SIGN_BIT_S
;
528 vc
.wp
[0] = va
.wp
[0] | SIGN_BIT_S
;
529 vc
.wp
[1] = va
.wp
[1] | SIGN_BIT_S
;
533 vc
.wp
[0] = va
.wp
[0] ^ SIGN_BIT_S
;
534 vc
.wp
[1] = va
.wp
[1] ^ SIGN_BIT_S
;
538 FP_ADD_S(SR0
, SA0
, SB0
);
539 FP_ADD_S(SR1
, SA1
, SB1
);
543 FP_SUB_S(SR0
, SA0
, SB0
);
544 FP_SUB_S(SR1
, SA1
, SB1
);
548 FP_MUL_S(SR0
, SA0
, SB0
);
549 FP_MUL_S(SR1
, SA1
, SB1
);
553 FP_DIV_S(SR0
, SA0
, SB0
);
554 FP_DIV_S(SR1
, SA1
, SB1
);
570 __asm__
__volatile__ ("mtspr 512, %4\n"
573 : "=r" (vc
.wp
[0]), "=r" (vc
.wp
[1])
574 : "r" (vb
.wp
[0]), "r" (vb
.wp
[1]), "r" (0));
578 __asm__
__volatile__ ("mtspr 512, %4\n"
581 : "=r" (vc
.wp
[0]), "=r" (vc
.wp
[1])
582 : "r" (vb
.wp
[0]), "r" (vb
.wp
[1]), "r" (0));
593 _FP_ROUND_ZERO(1, SB0
);
594 _FP_ROUND_ZERO(1, SB1
);
596 FP_TO_INT_S(vc
.wp
[0], SB0
, 32,
597 (((func
& 0x3) != 0) || SB0_s
));
598 FP_TO_INT_S(vc
.wp
[1], SB1
, 32,
599 (((func
& 0x3) != 0) || SB1_s
));
609 printk("SR0: %ld %08lx %ld (%ld)\n", SR0_s
, SR0_f
, SR0_e
, SR0_c
);
610 printk("SR1: %ld %08lx %ld (%ld)\n", SR1_s
, SR1_f
, SR1_e
, SR1_c
);
612 FP_PACK_SP(vc
.wp
, SR0
);
613 FP_PACK_SP(vc
.wp
+ 1, SR1
);
620 FP_CMP_S(IR0
, SA0
, SB0
, 3);
621 FP_CMP_S(IR1
, SA1
, SB1
, 3);
622 if (IR0
== 3 && (FP_ISSIGNAN_S(SA0
) || FP_ISSIGNAN_S(SB0
)))
623 FP_SET_EXCEPTION(FP_EX_INVALID
);
624 if (IR1
== 3 && (FP_ISSIGNAN_S(SA1
) || FP_ISSIGNAN_S(SB1
)))
625 FP_SET_EXCEPTION(FP_EX_INVALID
);
626 ch
= (IR0
== cmp
) ? 1 : 0;
627 cl
= (IR1
== cmp
) ? 1 : 0;
628 IR
= (ch
<< 3) | (cl
<< 2) | ((ch
| cl
) << 1) |
638 regs
->ccr
&= ~(15 << ((7 - ((speinsn
>> 23) & 0x7)) << 2));
639 regs
->ccr
|= (IR
<< ((7 - ((speinsn
>> 23) & 0x7)) << 2));
642 __FPU_FPSCR
&= ~FP_EX_MASK
;
643 __FPU_FPSCR
|= (FP_CUR_EXCEPTIONS
& FP_EX_MASK
);
644 mtspr(SPRN_SPEFSCR
, __FPU_FPSCR
);
646 current
->thread
.evr
[fc
] = vc
.wp
[0];
647 regs
->gpr
[fc
] = vc
.wp
[1];
650 printk("ccr = %08lx\n", regs
->ccr
);
651 printk("cur exceptions = %08x spefscr = %08lx\n",
652 FP_CUR_EXCEPTIONS
, __FPU_FPSCR
);
653 printk("vc: %08x %08x\n", vc
.wp
[0], vc
.wp
[1]);
654 printk("va: %08x %08x\n", va
.wp
[0], va
.wp
[1]);
655 printk("vb: %08x %08x\n", vb
.wp
[0], vb
.wp
[1]);
661 if (have_e500_cpu_a005_erratum
) {
662 /* according to e500 cpu a005 erratum, reissue efp inst */
665 printk(KERN_DEBUG
"re-issue efp inst: %08lx\n", speinsn
);
670 printk(KERN_ERR
"\nOoops! IEEE-754 compliance handler encountered un-supported instruction.\ninst code: %08lx\n", speinsn
);
674 int speround_handler(struct pt_regs
*regs
)
678 unsigned long speinsn
, type
, fc
;
680 if (get_user(speinsn
, (unsigned int __user
*) regs
->nip
))
682 if ((speinsn
>> 26) != 4)
683 return -EINVAL
; /* not an spe instruction */
685 type
= insn_type(speinsn
& 0x7ff);
686 if (type
== XCR
) return -ENOSYS
;
688 fc
= (speinsn
>> 21) & 0x1f;
689 s_lo
= regs
->gpr
[fc
] & SIGN_BIT_S
;
690 s_hi
= current
->thread
.evr
[fc
] & SIGN_BIT_S
;
691 fgpr
.wp
[0] = current
->thread
.evr
[fc
];
692 fgpr
.wp
[1] = regs
->gpr
[fc
];
694 __FPU_FPSCR
= mfspr(SPRN_SPEFSCR
);
696 switch ((speinsn
>> 5) & 0x7) {
697 /* Since SPE instructions on E500 core can handle round to nearest
698 * and round toward zero with IEEE-754 complied, we just need
699 * to handle round toward +Inf and round toward -Inf by software.
702 if ((FP_ROUNDMODE
) == FP_RND_PINF
) {
703 if (!s_lo
) fgpr
.wp
[1]++; /* Z > 0, choose Z1 */
704 } else { /* round to -Inf */
705 if (s_lo
) fgpr
.wp
[1]++; /* Z < 0, choose Z2 */
710 if (FP_ROUNDMODE
== FP_RND_PINF
) {
711 if (!s_hi
) fgpr
.dp
[0]++; /* Z > 0, choose Z1 */
712 } else { /* round to -Inf */
713 if (s_hi
) fgpr
.dp
[0]++; /* Z < 0, choose Z2 */
718 if (FP_ROUNDMODE
== FP_RND_PINF
) {
719 if (!s_lo
) fgpr
.wp
[1]++; /* Z_low > 0, choose Z1 */
720 if (!s_hi
) fgpr
.wp
[0]++; /* Z_high word > 0, choose Z1 */
721 } else { /* round to -Inf */
722 if (s_lo
) fgpr
.wp
[1]++; /* Z_low < 0, choose Z2 */
723 if (s_hi
) fgpr
.wp
[0]++; /* Z_high < 0, choose Z2 */
731 current
->thread
.evr
[fc
] = fgpr
.wp
[0];
732 regs
->gpr
[fc
] = fgpr
.wp
[1];
737 int __init
spe_mathemu_init(void)
741 pvr
= mfspr(SPRN_PVR
);
743 if ((PVR_VER(pvr
) == PVR_VER_E500V1
) ||
744 (PVR_VER(pvr
) == PVR_VER_E500V2
)) {
749 * E500 revision below 1.1, 2.3, 3.1, 4.1, 5.1
750 * need cpu a005 errata workaround
755 have_e500_cpu_a005_erratum
= 1;
759 have_e500_cpu_a005_erratum
= 1;
765 have_e500_cpu_a005_erratum
= 1;
775 module_init(spe_mathemu_init
);