fpu.c 13.0 KB
Newer Older
1
/*
L
Linus Torvalds 已提交
2 3 4 5 6 7 8
 * Save/restore floating point context for signal handlers.
 *
 * This file is subject to the terms and conditions of the GNU General Public
 * License.  See the file "COPYING" in the main directory of this archive
 * for more details.
 *
 * Copyright (C) 1999, 2000  Kaz Kojima & Niibe Yutaka
9
 * Copyright (C) 2006  ST Microelectronics Ltd. (denorm support)
L
Linus Torvalds 已提交
10
 *
11
 * FIXME! These routines have not been tested for big endian case.
L
Linus Torvalds 已提交
12 13 14
 */
#include <linux/sched.h>
#include <linux/signal.h>
15 16
#include <linux/io.h>
#include <asm/cpu/fpu.h>
L
Linus Torvalds 已提交
17
#include <asm/processor.h>
18
#include <asm/system.h>
L
Linus Torvalds 已提交
19 20 21 22 23 24 25

/* The PR (precision) bit in the FP Status Register must be clear when
 * an frchg instruction is executed, otherwise the instruction is undefined.
 * Executing frchg with PR set causes a trap on some SH4 implementations.
 */

#define FPSCR_RCHG 0x00000000
26 27 28 29 30 31 32 33 34 35 36 37
extern unsigned long long float64_div(unsigned long long a,
				      unsigned long long b);
extern unsigned long int float32_div(unsigned long int a, unsigned long int b);
extern unsigned long long float64_mul(unsigned long long a,
				      unsigned long long b);
extern unsigned long int float32_mul(unsigned long int a, unsigned long int b);
extern unsigned long long float64_add(unsigned long long a,
				      unsigned long long b);
extern unsigned long int float32_add(unsigned long int a, unsigned long int b);
extern unsigned long long float64_sub(unsigned long long a,
				      unsigned long long b);
extern unsigned long int float32_sub(unsigned long int a, unsigned long int b);
L
Linus Torvalds 已提交
38

39
static unsigned int fpu_exception_flags;
L
Linus Torvalds 已提交
40 41 42 43 44

/*
 * Save FPU registers onto task structure.
 * Assume called with FPU enabled (SR.FD=0).
 */
45
void save_fpu(struct task_struct *tsk, struct pt_regs *regs)
L
Linus Torvalds 已提交
46 47 48 49 50
{
	unsigned long dummy;

	clear_tsk_thread_flag(tsk, TIF_USEDFPU);
	enable_fpu();
51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91
	asm volatile ("sts.l	fpul, @-%0\n\t"
		      "sts.l	fpscr, @-%0\n\t"
		      "lds	%2, fpscr\n\t"
		      "frchg\n\t"
		      "fmov.s	fr15, @-%0\n\t"
		      "fmov.s	fr14, @-%0\n\t"
		      "fmov.s	fr13, @-%0\n\t"
		      "fmov.s	fr12, @-%0\n\t"
		      "fmov.s	fr11, @-%0\n\t"
		      "fmov.s	fr10, @-%0\n\t"
		      "fmov.s	fr9, @-%0\n\t"
		      "fmov.s	fr8, @-%0\n\t"
		      "fmov.s	fr7, @-%0\n\t"
		      "fmov.s	fr6, @-%0\n\t"
		      "fmov.s	fr5, @-%0\n\t"
		      "fmov.s	fr4, @-%0\n\t"
		      "fmov.s	fr3, @-%0\n\t"
		      "fmov.s	fr2, @-%0\n\t"
		      "fmov.s	fr1, @-%0\n\t"
		      "fmov.s	fr0, @-%0\n\t"
		      "frchg\n\t"
		      "fmov.s	fr15, @-%0\n\t"
		      "fmov.s	fr14, @-%0\n\t"
		      "fmov.s	fr13, @-%0\n\t"
		      "fmov.s	fr12, @-%0\n\t"
		      "fmov.s	fr11, @-%0\n\t"
		      "fmov.s	fr10, @-%0\n\t"
		      "fmov.s	fr9, @-%0\n\t"
		      "fmov.s	fr8, @-%0\n\t"
		      "fmov.s	fr7, @-%0\n\t"
		      "fmov.s	fr6, @-%0\n\t"
		      "fmov.s	fr5, @-%0\n\t"
		      "fmov.s	fr4, @-%0\n\t"
		      "fmov.s	fr3, @-%0\n\t"
		      "fmov.s	fr2, @-%0\n\t"
		      "fmov.s	fr1, @-%0\n\t"
		      "fmov.s	fr0, @-%0\n\t"
		      "lds	%3, fpscr\n\t":"=r" (dummy)
		      :"0"((char *)(&tsk->thread.fpu.hard.status)),
		      "r"(FPSCR_RCHG), "r"(FPSCR_INIT)
		      :"memory");
L
Linus Torvalds 已提交
92

P
Paul Mundt 已提交
93 94
	disable_fpu();
	release_fpu(regs);
L
Linus Torvalds 已提交
95 96
}

97
static void restore_fpu(struct task_struct *tsk)
L
Linus Torvalds 已提交
98 99 100
{
	unsigned long dummy;

P
Paul Mundt 已提交
101
	enable_fpu();
102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141
	asm volatile ("lds	%2, fpscr\n\t"
		      "fmov.s	@%0+, fr0\n\t"
		      "fmov.s	@%0+, fr1\n\t"
		      "fmov.s	@%0+, fr2\n\t"
		      "fmov.s	@%0+, fr3\n\t"
		      "fmov.s	@%0+, fr4\n\t"
		      "fmov.s	@%0+, fr5\n\t"
		      "fmov.s	@%0+, fr6\n\t"
		      "fmov.s	@%0+, fr7\n\t"
		      "fmov.s	@%0+, fr8\n\t"
		      "fmov.s	@%0+, fr9\n\t"
		      "fmov.s	@%0+, fr10\n\t"
		      "fmov.s	@%0+, fr11\n\t"
		      "fmov.s	@%0+, fr12\n\t"
		      "fmov.s	@%0+, fr13\n\t"
		      "fmov.s	@%0+, fr14\n\t"
		      "fmov.s	@%0+, fr15\n\t"
		      "frchg\n\t"
		      "fmov.s	@%0+, fr0\n\t"
		      "fmov.s	@%0+, fr1\n\t"
		      "fmov.s	@%0+, fr2\n\t"
		      "fmov.s	@%0+, fr3\n\t"
		      "fmov.s	@%0+, fr4\n\t"
		      "fmov.s	@%0+, fr5\n\t"
		      "fmov.s	@%0+, fr6\n\t"
		      "fmov.s	@%0+, fr7\n\t"
		      "fmov.s	@%0+, fr8\n\t"
		      "fmov.s	@%0+, fr9\n\t"
		      "fmov.s	@%0+, fr10\n\t"
		      "fmov.s	@%0+, fr11\n\t"
		      "fmov.s	@%0+, fr12\n\t"
		      "fmov.s	@%0+, fr13\n\t"
		      "fmov.s	@%0+, fr14\n\t"
		      "fmov.s	@%0+, fr15\n\t"
		      "frchg\n\t"
		      "lds.l	@%0+, fpscr\n\t"
		      "lds.l	@%0+, fpul\n\t"
		      :"=r" (dummy)
		      :"0"(&tsk->thread.fpu), "r"(FPSCR_RCHG)
		      :"memory");
L
Linus Torvalds 已提交
142 143 144 145 146 147
	disable_fpu();
}

/*
 * Load the FPU with signalling NANS.  This bit pattern we're using
 * has the property that no matter wether considered as single or as
P
Paul Mundt 已提交
148
 * double precision represents signaling NANS.
L
Linus Torvalds 已提交
149 150
 */

151
static void fpu_init(void)
L
Linus Torvalds 已提交
152 153
{
	enable_fpu();
154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192
	asm volatile (	"lds	%0, fpul\n\t"
			"lds	%1, fpscr\n\t"
			"fsts	fpul, fr0\n\t"
			"fsts	fpul, fr1\n\t"
			"fsts	fpul, fr2\n\t"
			"fsts	fpul, fr3\n\t"
			"fsts	fpul, fr4\n\t"
			"fsts	fpul, fr5\n\t"
			"fsts	fpul, fr6\n\t"
			"fsts	fpul, fr7\n\t"
			"fsts	fpul, fr8\n\t"
			"fsts	fpul, fr9\n\t"
			"fsts	fpul, fr10\n\t"
			"fsts	fpul, fr11\n\t"
			"fsts	fpul, fr12\n\t"
			"fsts	fpul, fr13\n\t"
			"fsts	fpul, fr14\n\t"
			"fsts	fpul, fr15\n\t"
			"frchg\n\t"
			"fsts	fpul, fr0\n\t"
			"fsts	fpul, fr1\n\t"
			"fsts	fpul, fr2\n\t"
			"fsts	fpul, fr3\n\t"
			"fsts	fpul, fr4\n\t"
			"fsts	fpul, fr5\n\t"
			"fsts	fpul, fr6\n\t"
			"fsts	fpul, fr7\n\t"
			"fsts	fpul, fr8\n\t"
			"fsts	fpul, fr9\n\t"
			"fsts	fpul, fr10\n\t"
			"fsts	fpul, fr11\n\t"
			"fsts	fpul, fr12\n\t"
			"fsts	fpul, fr13\n\t"
			"fsts	fpul, fr14\n\t"
			"fsts	fpul, fr15\n\t"
			"frchg\n\t"
			"lds	%2, fpscr\n\t"
			:	/* no output */
			:"r" (0), "r"(FPSCR_RCHG), "r"(FPSCR_INIT));
P
Paul Mundt 已提交
193
	disable_fpu();
L
Linus Torvalds 已提交
194 195 196
}

/**
197 198
 *      denormal_to_double - Given denormalized float number,
 *                           store double float
L
Linus Torvalds 已提交
199
 *
200 201
 *      @fpu: Pointer to sh_fpu_hard structure
 *      @n: Index to FP register
L
Linus Torvalds 已提交
202
 */
203
static void denormal_to_double(struct sh_fpu_hard_struct *fpu, int n)
L
Linus Torvalds 已提交
204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219
{
	unsigned long du, dl;
	unsigned long x = fpu->fpul;
	int exp = 1023 - 126;

	if (x != 0 && (x & 0x7f800000) == 0) {
		du = (x & 0x80000000);
		while ((x & 0x00800000) == 0) {
			x <<= 1;
			exp--;
		}
		x &= 0x007fffff;
		du |= (exp << 20) | (x >> 3);
		dl = x << 29;

		fpu->fp_regs[n] = du;
220
		fpu->fp_regs[n + 1] = dl;
L
Linus Torvalds 已提交
221 222 223 224 225 226 227 228 229 230
	}
}

/**
 *	ieee_fpe_handler - Handle denormalized number exception
 *
 *	@regs: Pointer to register structure
 *
 *	Returns 1 when it's handled (should not cause exception).
 */
231
static int ieee_fpe_handler(struct pt_regs *regs)
L
Linus Torvalds 已提交
232
{
233
	unsigned short insn = *(unsigned short *)regs->pc;
L
Linus Torvalds 已提交
234 235 236 237 238 239
	unsigned short finsn;
	unsigned long nextpc;
	int nib[4] = {
		(insn >> 12) & 0xf,
		(insn >> 8) & 0xf,
		(insn >> 4) & 0xf,
240 241 242 243 244 245 246 247 248 249 250 251
		insn & 0xf
	};

	if (nib[0] == 0xb || (nib[0] == 0x4 && nib[2] == 0x0 && nib[3] == 0xb))
		regs->pr = regs->pc + 4;  /* bsr & jsr */

	if (nib[0] == 0xa || nib[0] == 0xb) {
		/* bra & bsr */
		nextpc = regs->pc + 4 + ((short)((insn & 0xfff) << 4) >> 3);
		finsn = *(unsigned short *)(regs->pc + 2);
	} else if (nib[0] == 0x8 && nib[1] == 0xd) {
		/* bt/s */
L
Linus Torvalds 已提交
252
		if (regs->sr & 1)
253
			nextpc = regs->pc + 4 + ((char)(insn & 0xff) << 1);
L
Linus Torvalds 已提交
254 255
		else
			nextpc = regs->pc + 4;
256 257 258
		finsn = *(unsigned short *)(regs->pc + 2);
	} else if (nib[0] == 0x8 && nib[1] == 0xf) {
		/* bf/s */
L
Linus Torvalds 已提交
259 260 261
		if (regs->sr & 1)
			nextpc = regs->pc + 4;
		else
262 263
			nextpc = regs->pc + 4 + ((char)(insn & 0xff) << 1);
		finsn = *(unsigned short *)(regs->pc + 2);
L
Linus Torvalds 已提交
264
	} else if (nib[0] == 0x4 && nib[3] == 0xb &&
265 266
		   (nib[2] == 0x0 || nib[2] == 0x2)) {
		/* jmp & jsr */
L
Linus Torvalds 已提交
267
		nextpc = regs->regs[nib[1]];
268
		finsn = *(unsigned short *)(regs->pc + 2);
L
Linus Torvalds 已提交
269
	} else if (nib[0] == 0x0 && nib[3] == 0x3 &&
270 271
		   (nib[2] == 0x0 || nib[2] == 0x2)) {
		/* braf & bsrf */
L
Linus Torvalds 已提交
272
		nextpc = regs->pc + 4 + regs->regs[nib[1]];
273 274 275
		finsn = *(unsigned short *)(regs->pc + 2);
	} else if (insn == 0x000b) {
		/* rts */
L
Linus Torvalds 已提交
276
		nextpc = regs->pr;
277
		finsn = *(unsigned short *)(regs->pc + 2);
L
Linus Torvalds 已提交
278
	} else {
279
		nextpc = regs->pc + instruction_size(insn);
L
Linus Torvalds 已提交
280 281 282
		finsn = insn;
	}

283 284
	if ((finsn & 0xf1ff) == 0xf0ad) {
		/* fcnvsd */
L
Linus Torvalds 已提交
285 286 287
		struct task_struct *tsk = current;

		save_fpu(tsk, regs);
288
		if ((tsk->thread.fpu.hard.fpscr & FPSCR_CAUSE_ERROR))
L
Linus Torvalds 已提交
289
			/* FPU error */
290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413
			denormal_to_double(&tsk->thread.fpu.hard,
					   (finsn >> 8) & 0xf);
		else
			return 0;

		regs->pc = nextpc;
		return 1;
	} else if ((finsn & 0xf00f) == 0xf002) {
		/* fmul */
		struct task_struct *tsk = current;
		int fpscr;
		int n, m, prec;
		unsigned int hx, hy;

		n = (finsn >> 8) & 0xf;
		m = (finsn >> 4) & 0xf;
		hx = tsk->thread.fpu.hard.fp_regs[n];
		hy = tsk->thread.fpu.hard.fp_regs[m];
		fpscr = tsk->thread.fpu.hard.fpscr;
		prec = fpscr & FPSCR_DBL_PRECISION;

		if ((fpscr & FPSCR_CAUSE_ERROR)
		    && (prec && ((hx & 0x7fffffff) < 0x00100000
				 || (hy & 0x7fffffff) < 0x00100000))) {
			long long llx, lly;

			/* FPU error because of denormal (doubles) */
			llx = ((long long)hx << 32)
			    | tsk->thread.fpu.hard.fp_regs[n + 1];
			lly = ((long long)hy << 32)
			    | tsk->thread.fpu.hard.fp_regs[m + 1];
			llx = float64_mul(llx, lly);
			tsk->thread.fpu.hard.fp_regs[n] = llx >> 32;
			tsk->thread.fpu.hard.fp_regs[n + 1] = llx & 0xffffffff;
		} else if ((fpscr & FPSCR_CAUSE_ERROR)
			   && (!prec && ((hx & 0x7fffffff) < 0x00800000
					 || (hy & 0x7fffffff) < 0x00800000))) {
			/* FPU error because of denormal (floats) */
			hx = float32_mul(hx, hy);
			tsk->thread.fpu.hard.fp_regs[n] = hx;
		} else
			return 0;

		regs->pc = nextpc;
		return 1;
	} else if ((finsn & 0xf00e) == 0xf000) {
		/* fadd, fsub */
		struct task_struct *tsk = current;
		int fpscr;
		int n, m, prec;
		unsigned int hx, hy;

		n = (finsn >> 8) & 0xf;
		m = (finsn >> 4) & 0xf;
		hx = tsk->thread.fpu.hard.fp_regs[n];
		hy = tsk->thread.fpu.hard.fp_regs[m];
		fpscr = tsk->thread.fpu.hard.fpscr;
		prec = fpscr & FPSCR_DBL_PRECISION;

		if ((fpscr & FPSCR_CAUSE_ERROR)
		    && (prec && ((hx & 0x7fffffff) < 0x00100000
				 || (hy & 0x7fffffff) < 0x00100000))) {
			long long llx, lly;

			/* FPU error because of denormal (doubles) */
			llx = ((long long)hx << 32)
			    | tsk->thread.fpu.hard.fp_regs[n + 1];
			lly = ((long long)hy << 32)
			    | tsk->thread.fpu.hard.fp_regs[m + 1];
			if ((finsn & 0xf00f) == 0xf000)
				llx = float64_add(llx, lly);
			else
				llx = float64_sub(llx, lly);
			tsk->thread.fpu.hard.fp_regs[n] = llx >> 32;
			tsk->thread.fpu.hard.fp_regs[n + 1] = llx & 0xffffffff;
		} else if ((fpscr & FPSCR_CAUSE_ERROR)
			   && (!prec && ((hx & 0x7fffffff) < 0x00800000
					 || (hy & 0x7fffffff) < 0x00800000))) {
			/* FPU error because of denormal (floats) */
			if ((finsn & 0xf00f) == 0xf000)
				hx = float32_add(hx, hy);
			else
				hx = float32_sub(hx, hy);
			tsk->thread.fpu.hard.fp_regs[n] = hx;
		} else
			return 0;

		regs->pc = nextpc;
		return 1;
	} else if ((finsn & 0xf003) == 0xf003) {
		/* fdiv */
		struct task_struct *tsk = current;
		int fpscr;
		int n, m, prec;
		unsigned int hx, hy;

		n = (finsn >> 8) & 0xf;
		m = (finsn >> 4) & 0xf;
		hx = tsk->thread.fpu.hard.fp_regs[n];
		hy = tsk->thread.fpu.hard.fp_regs[m];
		fpscr = tsk->thread.fpu.hard.fpscr;
		prec = fpscr & FPSCR_DBL_PRECISION;

		if ((fpscr & FPSCR_CAUSE_ERROR)
		    && (prec && ((hx & 0x7fffffff) < 0x00100000
				 || (hy & 0x7fffffff) < 0x00100000))) {
			long long llx, lly;

			/* FPU error because of denormal (doubles) */
			llx = ((long long)hx << 32)
			    | tsk->thread.fpu.hard.fp_regs[n + 1];
			lly = ((long long)hy << 32)
			    | tsk->thread.fpu.hard.fp_regs[m + 1];

			llx = float64_div(llx, lly);

			tsk->thread.fpu.hard.fp_regs[n] = llx >> 32;
			tsk->thread.fpu.hard.fp_regs[n + 1] = llx & 0xffffffff;
		} else if ((fpscr & FPSCR_CAUSE_ERROR)
			   && (!prec && ((hx & 0x7fffffff) < 0x00800000
					 || (hy & 0x7fffffff) < 0x00800000))) {
			/* FPU error because of denormal (floats) */
			hx = float32_div(hx, hy);
			tsk->thread.fpu.hard.fp_regs[n] = hx;
414
		} else
415
			return 0;
L
Linus Torvalds 已提交
416 417 418 419 420 421 422 423

		regs->pc = nextpc;
		return 1;
	}

	return 0;
}

424 425 426 427 428 429 430 431 432 433 434 435
void float_raise(unsigned int flags)
{
	fpu_exception_flags |= flags;
}

int float_rounding_mode(void)
{
	struct task_struct *tsk = current;
	int roundingMode = FPSCR_ROUNDING_MODE(tsk->thread.fpu.hard.fpscr);
	return roundingMode;
}

P
Paul Mundt 已提交
436
BUILD_TRAP_HANDLER(fpu_error)
L
Linus Torvalds 已提交
437 438
{
	struct task_struct *tsk = current;
P
Paul Mundt 已提交
439
	TRAP_HANDLER_DECL;
L
Linus Torvalds 已提交
440

S
Stuart Menefy 已提交
441
	save_fpu(tsk, regs);
442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458
	fpu_exception_flags = 0;
	if (ieee_fpe_handler(regs)) {
		tsk->thread.fpu.hard.fpscr &=
		    ~(FPSCR_CAUSE_MASK | FPSCR_FLAG_MASK);
		tsk->thread.fpu.hard.fpscr |= fpu_exception_flags;
		/* Set the FPSCR flag as well as cause bits - simply
		 * replicate the cause */
		tsk->thread.fpu.hard.fpscr |= (fpu_exception_flags >> 10);
		grab_fpu(regs);
		restore_fpu(tsk);
		set_tsk_thread_flag(tsk, TIF_USEDFPU);
		if ((((tsk->thread.fpu.hard.fpscr & FPSCR_ENABLE_MASK) >> 7) &
		     (fpu_exception_flags >> 2)) == 0) {
			return;
		}
	}

L
Linus Torvalds 已提交
459 460 461
	force_sig(SIGFPE, tsk);
}

P
Paul Mundt 已提交
462
BUILD_TRAP_HANDLER(fpu_state_restore)
L
Linus Torvalds 已提交
463 464
{
	struct task_struct *tsk = current;
P
Paul Mundt 已提交
465
	TRAP_HANDLER_DECL;
L
Linus Torvalds 已提交
466

S
Stuart Menefy 已提交
467 468
	grab_fpu(regs);
	if (!user_mode(regs)) {
L
Linus Torvalds 已提交
469 470 471 472 473 474 475
		printk(KERN_ERR "BUG: FPU is used in kernel mode.\n");
		return;
	}

	if (used_math()) {
		/* Using the FPU again.  */
		restore_fpu(tsk);
476
	} else {
L
Linus Torvalds 已提交
477 478 479 480 481 482
		/* First time FPU user.  */
		fpu_init();
		set_used_math();
	}
	set_tsk_thread_flag(tsk, TIF_USEDFPU);
}