xref: /illumos-gate/usr/src/uts/sparc/sys/fpu/fpu_simulator.h (revision a38ddfee9c8c6b6c5a2947ff52fd2338362a4444)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #ifndef	_SYS_FPU_FPU_SIMULATOR_H
27 #define	_SYS_FPU_FPU_SIMULATOR_H
28 
29 /* SunOS-4.0 1.10	*/
30 
31 /*
32  * sparc floating-point simulator definitions.
33  */
34 
35 #ifndef	_ASM
36 #include <sys/types.h>
37 #include <sys/ieeefp.h>
38 #include <vm/seg.h>
39 #include <sys/kstat.h>
40 #endif /* _ASM */
41 
42 #ifdef	__cplusplus
43 extern "C" {
44 #endif
45 
46 /*
47  * Constants to decode/extract "fitos" instruction fields
48  */
49 #define	FITOS_INSTR_MASK	0xc1f83fe0
50 #define	FITOS_INSTR		0x81a01880
51 #define	FITOS_RS2_SHIFT		0
52 #define	FITOS_RD_SHIFT		25
53 #define	FITOS_REG_MASK		0x1f
54 
55 #ifndef _ASM
56 /*	PUBLIC TYPES	*/
57 
58 enum fcc_type {			/* relationships */
59 	fcc_equal	= 0,
60 	fcc_less	= 1,
61 	fcc_greater	= 2,
62 	fcc_unordered	= 3
63 };
64 
65 enum cc_type {			/* icc/fcc number */
66 	fcc_0	= 0,
67 	fcc_1	= 1,
68 	fcc_2	= 2,
69 	fcc_3	= 3,
70 	icc	= 4,
71 	xcc	= 6
72 };
73 
74 /* FSR types. */
75 
76 enum ftt_type {			/* types of traps */
77 	ftt_none	= 0,
78 	ftt_ieee	= 1,
79 	ftt_unfinished	= 2,
80 	ftt_unimplemented = 3,
81 	ftt_sequence	= 4,
82 	ftt_alignment	= 5,	/* defined by software convention only */
83 	ftt_fault	= 6,	/* defined by software convention only */
84 	ftt_7		= 7
85 };
86 
87 typedef	struct {		/* sparc V9 FSR. */
88 	unsigned int			: 26;
89 	unsigned int		fcc3	: 2;	/* fp condition code 3 */
90 	unsigned int		fcc2	: 2;	/* fp condition code 2 */
91 	unsigned int		fcc1	: 2;	/* fp condition code 1 */
92 						/* enum fp_direction_type */
93 	unsigned int		rnd	: 2;	/* rounding direction */
94 	unsigned int		rnp	: 2;	/* for v7 compatibility only */
95 	unsigned int		tem	: 5;	/* trap enable mask */
96 	unsigned int		 ns	: 1;	/* non-standard */
97 	unsigned int			: 5;
98 						/* enum ftt_type */
99 	unsigned int 		ftt	: 3;	/* FPU trap type */
100 	unsigned int		qne	: 1;	/* FPQ not empty */
101 	unsigned int		 pr	: 1;	/* partial result */
102 						/* enum fcc_type */
103 	unsigned int 		fcc	: 2;	/* fp condition code 0 */
104 	unsigned int		aexc	: 5;	/* accumulated exceptions */
105 	unsigned int		cexc	: 5;	/* current exception */
106 } fsr_types;
107 
108 /*
109  * The C compiler and the C spec do not support bitfields in a long long,
110  * as per fsr_types above, so don't hold your breath waiting for this
111  * workaround cruft to disappear.
112  */
113 
114 typedef union {
115 	fsr_types	fsr;
116 	uint64_t	ll;
117 } fsr_type;
118 
119 #define	fcc3	fsr.fcc3
120 #define	fcc2	fsr.fcc2
121 #define	fcc1	fsr.fcc1
122 #define	fcc0	fsr.fcc
123 #define	rnd	fsr.rnd
124 #define	rnp	fsr.rnp
125 #define	tem	fsr.tem
126 #define	aexc	fsr.aexc
127 #define	cexc	fsr.cexc
128 
129 typedef			/* FPU register viewed as single components. */
130 	struct {
131 	uint32_t sign :		1;
132 	uint32_t exponent :	8;
133 	uint32_t significand : 23;
134 } single_type;
135 
136 typedef			/* FPU register viewed as double components. */
137 	struct {
138 	uint32_t sign :		1;
139 	uint32_t exponent :    11;
140 	uint32_t significand : 20;
141 } double_type;
142 
143 typedef			/* FPU register viewed as extended components. */
144 	struct {
145 	uint32_t sign :		 1;
146 	uint32_t exponent :	15;
147 	uint32_t significand :	16;
148 } extended_type;
149 
150 typedef			/* FPU register with multiple data views. */
151 	union {
152 	int32_t		int32_reg;
153 	int64_t		int64_reg;
154 	uint32_t	uint32_reg;
155 	uint64_t	uint64_reg;
156 	float		float_reg;
157 	single_type	single_reg;
158 	double_type	double_reg;
159 	extended_type	extended_reg;
160 } freg_type;
161 
162 enum fp_op_type {		/* Type specifiers in FPU instructions. */
163 	fp_op_int32	= 0,	/* Not in hardware, but convenient to define. */
164 	fp_op_single	= 1,
165 	fp_op_double	= 2,
166 	fp_op_extended	= 3,
167 	fp_op_int64	= 4
168 };
169 
170 enum fp_opcode {	/* FPU op codes, minus precision and leading 0. */
171 	fmovs		= 0x0,
172 	fnegs		= 0x1,
173 	fabss		= 0x2,
174 	fp_op_3 = 3, fp_op_4 = 4, fp_op_5 = 5, fp_op_6 = 6, fp_op_7 = 7,
175 	fp_op_8		= 0x8,
176 	fp_op_9		= 0x9,
177 	fsqrt		= 0xa,
178 	fp_op_b = 0xb, fp_op_c = 0xc, fp_op_d = 0xd,
179 	fp_op_e = 0xe, fp_op_f = 0xf,
180 	fadd		= 0x10,
181 	fsub		= 0x11,
182 	fmul		= 0x12,
183 	fdiv		= 0x13,
184 	fcmp		= 0x14,
185 	fcmpe		= 0x15,
186 	fp_op_16 = 0x16, fp_op_17 = 0x17,
187 	fp_op_18	= 0x18,
188 	fp_op_19	= 0x19,
189 	fsmuld		= 0x1a,
190 	fdmulx		= 0x1b,
191 	ftoll		= 0x20,
192 	flltos		= 0x21,
193 	flltod		= 0x22,
194 	flltox		= 0x23,
195 	fp_op_24 = 0x24, fp_op_25 = 0x25, fp_op_26 = 0x26, fp_op_27 = 0x27,
196 	fp_op_28 = 0x28, fp_op_29 = 0x29, fp_op_2a = 0x2a, fp_op_2b = 0x2b,
197 	fp_op_2c = 0x2c, fp_op_2d = 0x2d, fp_op_2e = 0x2e, fp_op_2f = 0x2f,
198 	fp_op_30	= 0x30,
199 	fitos		= 0x31,
200 	fitod		= 0x32,
201 	fitox		= 0x33,
202 	ftoi		= 0x34,
203 	fp_op_35 = 0x35, fp_op_36 = 0x36, fp_op_37 = 0x37,
204 	ft_op_38	= 0x38,
205 	fp_op_39 = 0x39, fp_op_3a = 0x3a, fp_op_3b = 0x3b,
206 	fp_op_3c	= 0x3c,
207 	fp_op_3d = 0x3d, fp_op_3e = 0x3e, fmau = 0x3f
208 };
209 
210 typedef			/* FPU instruction. */
211 	struct {
212 	uint32_t		hibits	: 2;	/* Top two bits. */
213 	uint32_t		rd	: 5;	/* Destination. */
214 	uint32_t		op3	: 6;	/* Main op code. */
215 	uint32_t		rs1	: 5;	/* First operand. */
216 	uint32_t		ibit	: 1;	/* I format bit. */
217 	uint32_t /* enum fp_opcode */  opcode : 6; /* Floating-point op code. */
218 	uint32_t /* enum fp_op_type */ prec   : 2; /* Precision. */
219 	uint32_t		rs2	: 5;	/* Second operand. */
220 } fp_inst_type;
221 
222 enum fp_op_fma_var {	/* FMA-fused/unfused instr. variations */
223 	fmadd	=	0,
224 	fmsub	=	1,
225 	fnmsub	=	2,
226 	fnmadd	=	3
227 };
228 
229 typedef		/* FPU FMA-fused/unfused instructions. */
230 	struct {
231 	uint32_t		hibits	: 2;	/* Top two bits. */
232 	uint32_t		rd	: 5;	/* Destination. */
233 	uint32_t		op3	: 6;	/* Main op code. */
234 	uint32_t		rs1	: 5;	/* First operand. */
235 	uint32_t		rs3	: 5;	/* Third operand */
236 	uint32_t /* enum fp_op_fma_var */ var : 2; /* Instr. variation */
237 	uint32_t		sz	: 2;	/* Size */
238 	uint32_t		rs2	: 5;	/* Second operand. */
239 } fp_fma_inst_type;
240 
241 typedef			/* Integer condition code. */
242 	struct {
243 	uint32_t			: 28;	/* the unused part */
244 	uint32_t		n	: 1;	/* Negative bit. */
245 	uint32_t		z	: 1;	/* Zero bit. */
246 	uint32_t		v	: 1;	/* Overflow bit. */
247 	uint32_t		c	: 1;	/* Carry bit. */
248 } ccr_type;
249 
250 typedef			/* FPU data used by simulator. */
251 	struct {
252 	uint_t			fp_fsrtem;
253 	enum fp_direction_type	fp_direction;
254 	enum fp_precision_type	fp_precision;
255 	uint_t			fp_current_exceptions;
256 	kfpu_t			*fp_current_pfregs;
257 	void			(*fp_current_read_freg) ();
258 	void			(*fp_current_write_freg) ();
259 	void			(*fp_current_read_dreg) ();
260 	void			(*fp_current_write_dreg) ();
261 	uint64_t		(*fp_current_read_gsr) (kfpu_t *);
262 	void			(*fp_current_write_gsr) (uint64_t, kfpu_t *);
263 	int			fp_trapcode;
264 	char			*fp_trapaddr;
265 	struct	regs		*fp_traprp;
266 	enum	seg_rw		fp_traprw;
267 } fp_simd_type;
268 
269 /*
270  * FPU related kstat structures
271  */
272 struct fpustat_kstat  {
273 	struct kstat_named		fpu_ieee_traps;
274 	struct kstat_named		fpu_unfinished_traps;
275 	struct kstat_named		fpu_unimplemented_traps;
276 };
277 
278 struct fpuinfo_kstat {
279 	struct kstat_named		fpu_sim_fmovs;
280 	struct kstat_named		fpu_sim_fmovd;
281 	struct kstat_named		fpu_sim_fmovq;
282 	struct kstat_named		fpu_sim_fnegs;
283 	struct kstat_named		fpu_sim_fnegd;
284 	struct kstat_named		fpu_sim_fnegq;
285 	struct kstat_named		fpu_sim_fabss;
286 	struct kstat_named		fpu_sim_fabsd;
287 	struct kstat_named		fpu_sim_fabsq;
288 	struct kstat_named		fpu_sim_fsqrts;
289 	struct kstat_named		fpu_sim_fsqrtd;
290 	struct kstat_named		fpu_sim_fsqrtq;
291 	struct kstat_named		fpu_sim_fadds;
292 	struct kstat_named		fpu_sim_faddd;
293 	struct kstat_named		fpu_sim_faddq;
294 	struct kstat_named		fpu_sim_fsubs;
295 	struct kstat_named		fpu_sim_fsubd;
296 	struct kstat_named		fpu_sim_fsubq;
297 	struct kstat_named		fpu_sim_fmuls;
298 	struct kstat_named		fpu_sim_fmuld;
299 	struct kstat_named		fpu_sim_fmulq;
300 	struct kstat_named		fpu_sim_fdivs;
301 	struct kstat_named		fpu_sim_fdivd;
302 	struct kstat_named		fpu_sim_fdivq;
303 	struct kstat_named		fpu_sim_fcmps;
304 	struct kstat_named		fpu_sim_fcmpd;
305 	struct kstat_named		fpu_sim_fcmpq;
306 	struct kstat_named		fpu_sim_fcmpes;
307 	struct kstat_named		fpu_sim_fcmped;
308 	struct kstat_named		fpu_sim_fcmpeq;
309 	struct kstat_named		fpu_sim_fsmuld;
310 	struct kstat_named		fpu_sim_fdmulx;
311 	struct kstat_named		fpu_sim_fstox;
312 	struct kstat_named		fpu_sim_fdtox;
313 	struct kstat_named		fpu_sim_fqtox;
314 	struct kstat_named		fpu_sim_fxtos;
315 	struct kstat_named		fpu_sim_fxtod;
316 	struct kstat_named		fpu_sim_fxtoq;
317 	struct kstat_named		fpu_sim_fitos;
318 	struct kstat_named		fpu_sim_fitod;
319 	struct kstat_named		fpu_sim_fitoq;
320 	struct kstat_named		fpu_sim_fstoi;
321 	struct kstat_named		fpu_sim_fdtoi;
322 	struct kstat_named		fpu_sim_fqtoi;
323 	struct kstat_named		fpu_sim_fmovcc;
324 	struct kstat_named		fpu_sim_fmovr;
325 	struct kstat_named		fpu_sim_fmadds;
326 	struct kstat_named		fpu_sim_fmaddd;
327 	struct kstat_named		fpu_sim_fmsubs;
328 	struct kstat_named		fpu_sim_fmsubd;
329 	struct kstat_named		fpu_sim_fnmadds;
330 	struct kstat_named		fpu_sim_fnmaddd;
331 	struct kstat_named		fpu_sim_fnmsubs;
332 	struct kstat_named		fpu_sim_fnmsubd;
333 	struct kstat_named		fpu_sim_fumadds;
334 	struct kstat_named		fpu_sim_fumaddd;
335 	struct kstat_named		fpu_sim_fumsubs;
336 	struct kstat_named		fpu_sim_fumsubd;
337 	struct kstat_named		fpu_sim_fnumadds;
338 	struct kstat_named		fpu_sim_fnumaddd;
339 	struct kstat_named		fpu_sim_fnumsubs;
340 	struct kstat_named		fpu_sim_fnumsubd;
341 	struct kstat_named		fpu_sim_invalid;
342 };
343 
344 struct visinfo_kstat {
345 	struct kstat_named		vis_edge8;
346 	struct kstat_named		vis_edge8n;
347 	struct kstat_named		vis_edge8l;
348 	struct kstat_named		vis_edge8ln;
349 	struct kstat_named		vis_edge16;
350 	struct kstat_named		vis_edge16n;
351 	struct kstat_named		vis_edge16l;
352 	struct kstat_named		vis_edge16ln;
353 	struct kstat_named		vis_edge32;
354 	struct kstat_named		vis_edge32n;
355 	struct kstat_named		vis_edge32l;
356 	struct kstat_named		vis_edge32ln;
357 	struct kstat_named		vis_array8;
358 	struct kstat_named		vis_array16;
359 	struct kstat_named		vis_array32;
360 	struct kstat_named		vis_bmask;
361 	struct kstat_named		vis_fcmple16;
362 	struct kstat_named		vis_fcmpne16;
363 	struct kstat_named		vis_fcmpgt16;
364 	struct kstat_named		vis_fcmpeq16;
365 	struct kstat_named		vis_fcmple32;
366 	struct kstat_named		vis_fcmpne32;
367 	struct kstat_named		vis_fcmpgt32;
368 	struct kstat_named		vis_fcmpeq32;
369 	struct kstat_named		vis_fmul8x16;
370 	struct kstat_named		vis_fmul8x16au;
371 	struct kstat_named		vis_fmul8x16al;
372 	struct kstat_named		vis_fmul8sux16;
373 	struct kstat_named		vis_fmul8ulx16;
374 	struct kstat_named		vis_fmuld8sux16;
375 	struct kstat_named		vis_fmuld8ulx16;
376 	struct kstat_named		vis_fpack16;
377 	struct kstat_named		vis_fpack32;
378 	struct kstat_named		vis_fpackfix;
379 	struct kstat_named		vis_fexpand;
380 	struct kstat_named		vis_fpmerge;
381 	struct kstat_named		vis_pdist;
382 	struct kstat_named		vis_bshuffle;
383 };
384 
385 #define	VISINFO_KSTAT(opcode)	{					\
386 	extern void __dtrace_probe___visinfo_##opcode(uint64_t *);	\
387 	uint64_t *stataddr = &visinfo.opcode.value.ui64;		\
388 	__dtrace_probe___visinfo_##opcode(stataddr);       		\
389 	atomic_add_64(&visinfo.opcode.value.ui64, 1);			\
390 }
391 
392 
393 /* PUBLIC FUNCTIONS */
394 
395 #ifdef	__STDC__
396 
397 /*
398  * fpu_vis_sim simulates FPU VIS Partial load store instructions; reads and
399  * writes FPU data registers directly or works with the PCB image if fpu_exists
400  * is 0.
401  */
402 extern enum ftt_type fpu_vis_sim(fp_simd_type *pfpsd, fp_inst_type *pinst,
403 	struct regs *pregs, fsr_type *pfsr, uint64_t gsr, uint32_t inst);
404 /*
405  * fpu_simulator simulates FPU instructions only; reads and writes FPU data
406  * registers directly.
407  */
408 extern enum ftt_type fpu_simulator(fp_simd_type *pfpsd, fp_inst_type *pinst,
409 	fsr_type *pfsr, uint64_t gsr, uint32_t inst);
410 /*
411  * fp_emulator simulates FPU and CPU-FPU instructions; reads and writes FPU
412  * data registers from image in pfpu.
413  */
414 extern enum ftt_type fp_emulator(fp_simd_type *pfpsd, fp_inst_type *pinst,
415 	struct regs *rp, void *prw, kfpu_t *pfpu);
416 /*
417  * fp_traps handles passing exception conditions to the kernel.
418  * It is called after fp_simulator or fp_emulator fail (return a non-zero ftt).
419  */
420 extern void fp_traps(fp_simd_type *pfpsd, enum ftt_type ftt, struct regs *rp);
421 
422 /*
423  * fp_kstat_update tracks fpu exception conditions.
424  * It is called after a hardware trap returns a non-zero ftt.
425  */
426 extern void fp_kstat_update(enum ftt_type ftt);
427 
428 /*
429  * fp_precise handles floating point unimplemented and unfinished traps,
430  * for sparc V9 hardware. These traps are normally passed along to the
431  * fpu_simulator, to see if it can run the unimplemented instruction or
432  * finish the unfinished instruction. Needless to say, this takes time.
433  */
434 extern void fp_precise(struct regs *rp);
435 
436 /*
437  * fpu_trap handles V9 floating point ieee and other floating point traps.
438  * It is called after fp_simulator or fp_emulator fail (return a non-zero ftt),
439  * and from the _fp_ieee_exception trap handler.
440  */
441 extern void fpu_trap(struct regs *rp, caddr_t addr, uint32_t type,
442 			uint32_t code);
443 
444 #else	/* ! __STDC__ */
445 
446 /*
447  * fpu_simulator simulates FPU instructions only; reads and writes FPU data
448  * registers directly.
449  */
450 extern enum ftt_type fpu_simulator(
451 	fp_simd_type	*pfpsd,	 /* Pointer to FPU simulator data */
452 	fp_inst_type	*pinst,	 /* Pointer to FPU instruction to simulate. */
453 	fsr_type	*pfsr,	 /* Pointer to image of FSR to read & write. */
454 	int		instr);	 /* Instruction to emulate. */
455 
456 /*
457  * fp_emulator simulates FPU and CPU-FPU instructions; reads and writes FPU
458  * data registers from image in pfpu.
459  */
460 extern enum ftt_type fp_emulator(
461 	fp_simd_type	*pfpsd,	   /* Pointer to FPU simulator data */
462 	fp_inst_type	*pinst,    /* Pointer to FPU instruction to simulate. */
463 	struct regs	*pregs,    /* Pointer to PCB image of registers. */
464 	struct rwindow	*pwindow,  /* Pointer to locals and ins. */
465 	struct fpu	*pfpu);	   /* Pointer to FPU register block. */
466 
467 /*
468  * fp_traps handles passing exception conditions to the kernel.
469  * It is called after fp_simulator or fp_emulator fail (return a non-zero ftt).
470  */
471 extern void fp_traps(
472 	fp_simd_type	*pfpsd,	 /* Pointer to FPU simulator data */
473 	enum ftt_type	ftt,	 /* Type of trap. */
474 	struct regs	*rp);	 /* Pointer to PCB image of registers. */
475 
476 /*
477  * fp_kstat_update tracks fpu exception conditions.
478  * It is called after a hardware trap returns a non-zero ftt.
479  */
480 extern void fp_kstat_update(enum ftt_type ftt);	/* Type of trap. */
481 
482 /*
483  * fp_precise handles floating point unimplemented and unfinished traps,
484  * for sparc V9 hardware. These traps are normally passed along to the
485  * fpu_simulator, to see if it can run the unimplemented instruction or
486  * finish the unfinished instruction. Needless to say, this takes time.
487  */
488 extern void fp_precise(
489 	struct regs *rp);	/* Pointer to PCB image of registers. */
490 
491 /*
492  * fpu_trap handles V9 floating point ieee and other floating point traps.
493  * It is called after fp_simulator or fp_emulator fail (return a non-zero ftt),
494  * and from the _fp_ieee_exception trap handler.
495  */
496 extern void fpu_trap(
497 	struct regs *rp,	/* Pointer to PCB image of registers. */
498 	caddr_t addr,		/* Address of trapping instruction. */
499 	uint32_t type,		/* Type of trapping exception. */
500 	uint32_t code);		/* Trap code -> si_code. */
501 
502 #endif	/* __STDC__ */
503 #endif /* _ASM */
504 
505 #ifdef	__cplusplus
506 }
507 #endif
508 
509 #endif	/* _SYS_FPU_FPU_SIMULATOR_H */
510