2 * Save/restore floating point context for signal handlers.
4 * This file is subject to the terms and conditions of the GNU General Public
5 * License. See the file "COPYING" in the main directory of this archive
8 * Copyright (C) 1999, 2000 Kaz Kojima & Niibe Yutaka
9 * Copyright (C) 2006 ST Microelectronics Ltd. (denorm support)
11 * FIXME! These routines have not been tested for big endian case.
13 #include <linux/sched.h>
14 #include <linux/signal.h>
17 #include <asm/processor.h>
19 #include <asm/traps.h>
21 /* The PR (precision) bit in the FP Status Register must be clear when
22 * an frchg instruction is executed, otherwise the instruction is undefined.
23 * Executing frchg with PR set causes a trap on some SH4 implementations.
26 #define FPSCR_RCHG 0x00000000
27 extern unsigned long long float64_div(unsigned long long a
,
28 unsigned long long b
);
29 extern unsigned long int float32_div(unsigned long int a
, unsigned long int b
);
30 extern unsigned long long float64_mul(unsigned long long a
,
31 unsigned long long b
);
32 extern unsigned long int float32_mul(unsigned long int a
, unsigned long int b
);
33 extern unsigned long long float64_add(unsigned long long a
,
34 unsigned long long b
);
35 extern unsigned long int float32_add(unsigned long int a
, unsigned long int b
);
36 extern unsigned long long float64_sub(unsigned long long a
,
37 unsigned long long b
);
38 extern unsigned long int float32_sub(unsigned long int a
, unsigned long int b
);
39 extern unsigned long int float64_to_float32(unsigned long long a
);
40 static unsigned int fpu_exception_flags
;
43 * Save FPU registers onto task structure.
45 void save_fpu(struct task_struct
*tsk
)
50 asm volatile ("sts.l fpul, @-%0\n\t"
51 "sts.l fpscr, @-%0\n\t"
54 "fmov.s fr15, @-%0\n\t"
55 "fmov.s fr14, @-%0\n\t"
56 "fmov.s fr13, @-%0\n\t"
57 "fmov.s fr12, @-%0\n\t"
58 "fmov.s fr11, @-%0\n\t"
59 "fmov.s fr10, @-%0\n\t"
60 "fmov.s fr9, @-%0\n\t"
61 "fmov.s fr8, @-%0\n\t"
62 "fmov.s fr7, @-%0\n\t"
63 "fmov.s fr6, @-%0\n\t"
64 "fmov.s fr5, @-%0\n\t"
65 "fmov.s fr4, @-%0\n\t"
66 "fmov.s fr3, @-%0\n\t"
67 "fmov.s fr2, @-%0\n\t"
68 "fmov.s fr1, @-%0\n\t"
69 "fmov.s fr0, @-%0\n\t"
71 "fmov.s fr15, @-%0\n\t"
72 "fmov.s fr14, @-%0\n\t"
73 "fmov.s fr13, @-%0\n\t"
74 "fmov.s fr12, @-%0\n\t"
75 "fmov.s fr11, @-%0\n\t"
76 "fmov.s fr10, @-%0\n\t"
77 "fmov.s fr9, @-%0\n\t"
78 "fmov.s fr8, @-%0\n\t"
79 "fmov.s fr7, @-%0\n\t"
80 "fmov.s fr6, @-%0\n\t"
81 "fmov.s fr5, @-%0\n\t"
82 "fmov.s fr4, @-%0\n\t"
83 "fmov.s fr3, @-%0\n\t"
84 "fmov.s fr2, @-%0\n\t"
85 "fmov.s fr1, @-%0\n\t"
86 "fmov.s fr0, @-%0\n\t"
87 "lds %3, fpscr\n\t":"=r" (dummy
)
88 :"0"((char *)(&tsk
->thread
.xstate
->hardfpu
.status
)),
89 "r"(FPSCR_RCHG
), "r"(FPSCR_INIT
)
95 void restore_fpu(struct task_struct
*tsk
)
100 asm volatile ("lds %2, fpscr\n\t"
101 "fmov.s @%0+, fr0\n\t"
102 "fmov.s @%0+, fr1\n\t"
103 "fmov.s @%0+, fr2\n\t"
104 "fmov.s @%0+, fr3\n\t"
105 "fmov.s @%0+, fr4\n\t"
106 "fmov.s @%0+, fr5\n\t"
107 "fmov.s @%0+, fr6\n\t"
108 "fmov.s @%0+, fr7\n\t"
109 "fmov.s @%0+, fr8\n\t"
110 "fmov.s @%0+, fr9\n\t"
111 "fmov.s @%0+, fr10\n\t"
112 "fmov.s @%0+, fr11\n\t"
113 "fmov.s @%0+, fr12\n\t"
114 "fmov.s @%0+, fr13\n\t"
115 "fmov.s @%0+, fr14\n\t"
116 "fmov.s @%0+, fr15\n\t"
118 "fmov.s @%0+, fr0\n\t"
119 "fmov.s @%0+, fr1\n\t"
120 "fmov.s @%0+, fr2\n\t"
121 "fmov.s @%0+, fr3\n\t"
122 "fmov.s @%0+, fr4\n\t"
123 "fmov.s @%0+, fr5\n\t"
124 "fmov.s @%0+, fr6\n\t"
125 "fmov.s @%0+, fr7\n\t"
126 "fmov.s @%0+, fr8\n\t"
127 "fmov.s @%0+, fr9\n\t"
128 "fmov.s @%0+, fr10\n\t"
129 "fmov.s @%0+, fr11\n\t"
130 "fmov.s @%0+, fr12\n\t"
131 "fmov.s @%0+, fr13\n\t"
132 "fmov.s @%0+, fr14\n\t"
133 "fmov.s @%0+, fr15\n\t"
135 "lds.l @%0+, fpscr\n\t"
136 "lds.l @%0+, fpul\n\t"
138 :"0" (tsk
->thread
.xstate
), "r" (FPSCR_RCHG
)
144 * denormal_to_double - Given denormalized float number,
147 * @fpu: Pointer to sh_fpu_hard structure
148 * @n: Index to FP register
150 static void denormal_to_double(struct sh_fpu_hard_struct
*fpu
, int n
)
152 unsigned long du
, dl
;
153 unsigned long x
= fpu
->fpul
;
154 int exp
= 1023 - 126;
156 if (x
!= 0 && (x
& 0x7f800000) == 0) {
157 du
= (x
& 0x80000000);
158 while ((x
& 0x00800000) == 0) {
163 du
|= (exp
<< 20) | (x
>> 3);
166 fpu
->fp_regs
[n
] = du
;
167 fpu
->fp_regs
[n
+ 1] = dl
;
172 * ieee_fpe_handler - Handle denormalized number exception
174 * @regs: Pointer to register structure
176 * Returns 1 when it's handled (should not cause exception).
178 static int ieee_fpe_handler(struct pt_regs
*regs
)
180 unsigned short insn
= *(unsigned short *)regs
->pc
;
181 unsigned short finsn
;
182 unsigned long nextpc
;
190 if (nib
[0] == 0xb || (nib
[0] == 0x4 && nib
[2] == 0x0 && nib
[3] == 0xb))
191 regs
->pr
= regs
->pc
+ 4; /* bsr & jsr */
193 if (nib
[0] == 0xa || nib
[0] == 0xb) {
195 nextpc
= regs
->pc
+ 4 + ((short)((insn
& 0xfff) << 4) >> 3);
196 finsn
= *(unsigned short *)(regs
->pc
+ 2);
197 } else if (nib
[0] == 0x8 && nib
[1] == 0xd) {
200 nextpc
= regs
->pc
+ 4 + ((char)(insn
& 0xff) << 1);
202 nextpc
= regs
->pc
+ 4;
203 finsn
= *(unsigned short *)(regs
->pc
+ 2);
204 } else if (nib
[0] == 0x8 && nib
[1] == 0xf) {
207 nextpc
= regs
->pc
+ 4;
209 nextpc
= regs
->pc
+ 4 + ((char)(insn
& 0xff) << 1);
210 finsn
= *(unsigned short *)(regs
->pc
+ 2);
211 } else if (nib
[0] == 0x4 && nib
[3] == 0xb &&
212 (nib
[2] == 0x0 || nib
[2] == 0x2)) {
214 nextpc
= regs
->regs
[nib
[1]];
215 finsn
= *(unsigned short *)(regs
->pc
+ 2);
216 } else if (nib
[0] == 0x0 && nib
[3] == 0x3 &&
217 (nib
[2] == 0x0 || nib
[2] == 0x2)) {
219 nextpc
= regs
->pc
+ 4 + regs
->regs
[nib
[1]];
220 finsn
= *(unsigned short *)(regs
->pc
+ 2);
221 } else if (insn
== 0x000b) {
224 finsn
= *(unsigned short *)(regs
->pc
+ 2);
226 nextpc
= regs
->pc
+ instruction_size(insn
);
230 if ((finsn
& 0xf1ff) == 0xf0ad) {
232 struct task_struct
*tsk
= current
;
234 if ((tsk
->thread
.xstate
->hardfpu
.fpscr
& FPSCR_CAUSE_ERROR
))
236 denormal_to_double(&tsk
->thread
.xstate
->hardfpu
,
243 } else if ((finsn
& 0xf00f) == 0xf002) {
245 struct task_struct
*tsk
= current
;
250 n
= (finsn
>> 8) & 0xf;
251 m
= (finsn
>> 4) & 0xf;
252 hx
= tsk
->thread
.xstate
->hardfpu
.fp_regs
[n
];
253 hy
= tsk
->thread
.xstate
->hardfpu
.fp_regs
[m
];
254 fpscr
= tsk
->thread
.xstate
->hardfpu
.fpscr
;
255 prec
= fpscr
& FPSCR_DBL_PRECISION
;
257 if ((fpscr
& FPSCR_CAUSE_ERROR
)
258 && (prec
&& ((hx
& 0x7fffffff) < 0x00100000
259 || (hy
& 0x7fffffff) < 0x00100000))) {
262 /* FPU error because of denormal (doubles) */
263 llx
= ((long long)hx
<< 32)
264 | tsk
->thread
.xstate
->hardfpu
.fp_regs
[n
+ 1];
265 lly
= ((long long)hy
<< 32)
266 | tsk
->thread
.xstate
->hardfpu
.fp_regs
[m
+ 1];
267 llx
= float64_mul(llx
, lly
);
268 tsk
->thread
.xstate
->hardfpu
.fp_regs
[n
] = llx
>> 32;
269 tsk
->thread
.xstate
->hardfpu
.fp_regs
[n
+ 1] = llx
& 0xffffffff;
270 } else if ((fpscr
& FPSCR_CAUSE_ERROR
)
271 && (!prec
&& ((hx
& 0x7fffffff) < 0x00800000
272 || (hy
& 0x7fffffff) < 0x00800000))) {
273 /* FPU error because of denormal (floats) */
274 hx
= float32_mul(hx
, hy
);
275 tsk
->thread
.xstate
->hardfpu
.fp_regs
[n
] = hx
;
281 } else if ((finsn
& 0xf00e) == 0xf000) {
283 struct task_struct
*tsk
= current
;
288 n
= (finsn
>> 8) & 0xf;
289 m
= (finsn
>> 4) & 0xf;
290 hx
= tsk
->thread
.xstate
->hardfpu
.fp_regs
[n
];
291 hy
= tsk
->thread
.xstate
->hardfpu
.fp_regs
[m
];
292 fpscr
= tsk
->thread
.xstate
->hardfpu
.fpscr
;
293 prec
= fpscr
& FPSCR_DBL_PRECISION
;
295 if ((fpscr
& FPSCR_CAUSE_ERROR
)
296 && (prec
&& ((hx
& 0x7fffffff) < 0x00100000
297 || (hy
& 0x7fffffff) < 0x00100000))) {
300 /* FPU error because of denormal (doubles) */
301 llx
= ((long long)hx
<< 32)
302 | tsk
->thread
.xstate
->hardfpu
.fp_regs
[n
+ 1];
303 lly
= ((long long)hy
<< 32)
304 | tsk
->thread
.xstate
->hardfpu
.fp_regs
[m
+ 1];
305 if ((finsn
& 0xf00f) == 0xf000)
306 llx
= float64_add(llx
, lly
);
308 llx
= float64_sub(llx
, lly
);
309 tsk
->thread
.xstate
->hardfpu
.fp_regs
[n
] = llx
>> 32;
310 tsk
->thread
.xstate
->hardfpu
.fp_regs
[n
+ 1] = llx
& 0xffffffff;
311 } else if ((fpscr
& FPSCR_CAUSE_ERROR
)
312 && (!prec
&& ((hx
& 0x7fffffff) < 0x00800000
313 || (hy
& 0x7fffffff) < 0x00800000))) {
314 /* FPU error because of denormal (floats) */
315 if ((finsn
& 0xf00f) == 0xf000)
316 hx
= float32_add(hx
, hy
);
318 hx
= float32_sub(hx
, hy
);
319 tsk
->thread
.xstate
->hardfpu
.fp_regs
[n
] = hx
;
325 } else if ((finsn
& 0xf003) == 0xf003) {
327 struct task_struct
*tsk
= current
;
332 n
= (finsn
>> 8) & 0xf;
333 m
= (finsn
>> 4) & 0xf;
334 hx
= tsk
->thread
.xstate
->hardfpu
.fp_regs
[n
];
335 hy
= tsk
->thread
.xstate
->hardfpu
.fp_regs
[m
];
336 fpscr
= tsk
->thread
.xstate
->hardfpu
.fpscr
;
337 prec
= fpscr
& FPSCR_DBL_PRECISION
;
339 if ((fpscr
& FPSCR_CAUSE_ERROR
)
340 && (prec
&& ((hx
& 0x7fffffff) < 0x00100000
341 || (hy
& 0x7fffffff) < 0x00100000))) {
344 /* FPU error because of denormal (doubles) */
345 llx
= ((long long)hx
<< 32)
346 | tsk
->thread
.xstate
->hardfpu
.fp_regs
[n
+ 1];
347 lly
= ((long long)hy
<< 32)
348 | tsk
->thread
.xstate
->hardfpu
.fp_regs
[m
+ 1];
350 llx
= float64_div(llx
, lly
);
352 tsk
->thread
.xstate
->hardfpu
.fp_regs
[n
] = llx
>> 32;
353 tsk
->thread
.xstate
->hardfpu
.fp_regs
[n
+ 1] = llx
& 0xffffffff;
354 } else if ((fpscr
& FPSCR_CAUSE_ERROR
)
355 && (!prec
&& ((hx
& 0x7fffffff) < 0x00800000
356 || (hy
& 0x7fffffff) < 0x00800000))) {
357 /* FPU error because of denormal (floats) */
358 hx
= float32_div(hx
, hy
);
359 tsk
->thread
.xstate
->hardfpu
.fp_regs
[n
] = hx
;
365 } else if ((finsn
& 0xf0bd) == 0xf0bd) {
366 /* fcnvds - double to single precision convert */
367 struct task_struct
*tsk
= current
;
371 m
= (finsn
>> 8) & 0x7;
372 hx
= tsk
->thread
.xstate
->hardfpu
.fp_regs
[m
];
374 if ((tsk
->thread
.xstate
->hardfpu
.fpscr
& FPSCR_CAUSE_ERROR
)
375 && ((hx
& 0x7fffffff) < 0x00100000)) {
376 /* subnormal double to float conversion */
379 llx
= ((long long)tsk
->thread
.xstate
->hardfpu
.fp_regs
[m
] << 32)
380 | tsk
->thread
.xstate
->hardfpu
.fp_regs
[m
+ 1];
382 tsk
->thread
.xstate
->hardfpu
.fpul
= float64_to_float32(llx
);
393 void float_raise(unsigned int flags
)
395 fpu_exception_flags
|= flags
;
398 int float_rounding_mode(void)
400 struct task_struct
*tsk
= current
;
401 int roundingMode
= FPSCR_ROUNDING_MODE(tsk
->thread
.xstate
->hardfpu
.fpscr
);
405 BUILD_TRAP_HANDLER(fpu_error
)
407 struct task_struct
*tsk
= current
;
410 __unlazy_fpu(tsk
, regs
);
411 fpu_exception_flags
= 0;
412 if (ieee_fpe_handler(regs
)) {
413 tsk
->thread
.xstate
->hardfpu
.fpscr
&=
414 ~(FPSCR_CAUSE_MASK
| FPSCR_FLAG_MASK
);
415 tsk
->thread
.xstate
->hardfpu
.fpscr
|= fpu_exception_flags
;
416 /* Set the FPSCR flag as well as cause bits - simply
417 * replicate the cause */
418 tsk
->thread
.xstate
->hardfpu
.fpscr
|= (fpu_exception_flags
>> 10);
421 task_thread_info(tsk
)->status
|= TS_USEDFPU
;
422 if ((((tsk
->thread
.xstate
->hardfpu
.fpscr
& FPSCR_ENABLE_MASK
) >> 7) &
423 (fpu_exception_flags
>> 2)) == 0) {
428 force_sig(SIGFPE
, tsk
);