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 2015 Nexenta Systems, Inc. All rights reserved. 23 * Copyright 2017 Joyent, Inc. 24 * 25 * Copyright (c) 1992, 2010, Oracle and/or its affiliates. All rights reserved. 26 */ 27 28 /* Copyright (c) 1990, 1991 UNIX System Laboratories, Inc. */ 29 /* Copyright (c) 1984, 1986, 1987, 1988, 1989, 1990 AT&T */ 30 /* All Rights Reserved */ 31 32 #ifndef _SYS_FP_H 33 #define _SYS_FP_H 34 35 #ifdef __cplusplus 36 extern "C" { 37 #endif 38 39 /* 40 * 80287/80387 and SSE/SSE2 floating point processor definitions 41 */ 42 43 /* 44 * values that go into fp_kind 45 */ 46 #define FP_NO 0 /* no fp chip, no emulator (no fp support) */ 47 #define FP_SW 1 /* no fp chip, using software emulator */ 48 #define FP_HW 2 /* chip present bit */ 49 #define FP_287 2 /* 80287 chip present */ 50 #define FP_387 3 /* 80387 chip present */ 51 #define FP_487 6 /* 80487 chip present */ 52 #define FP_486 6 /* 80486 chip present */ 53 /* 54 * The following values are bit flags instead of actual values. 55 * E.g. to know if we are using SSE, test (value & __FP_SSE) instead 56 * of (value == __FP_SSE). 57 */ 58 #define __FP_SSE 0x100 /* .. plus SSE-capable CPU */ 59 #define __FP_AVX 0x200 /* .. plus AVX-capable CPU */ 60 61 /* 62 * values that go into fp_save_mech 63 */ 64 #define FP_FNSAVE 1 /* fnsave/frstor instructions */ 65 #define FP_FXSAVE 2 /* fxsave/fxrstor instructions */ 66 #define FP_XSAVE 3 /* xsave/xrstor instructions */ 67 68 /* 69 * masks for 80387 control word 70 */ 71 #define FPIM 0x00000001 /* invalid operation */ 72 #define FPDM 0x00000002 /* denormalized operand */ 73 #define FPZM 0x00000004 /* zero divide */ 74 #define FPOM 0x00000008 /* overflow */ 75 #define FPUM 0x00000010 /* underflow */ 76 #define FPPM 0x00000020 /* precision */ 77 #define FPPC 0x00000300 /* precision control */ 78 #define FPRC 0x00000C00 /* rounding control */ 79 #define FPIC 0x00001000 /* infinity control */ 80 #define WFPDE 0x00000080 /* data chain exception */ 81 82 /* 83 * (Old symbol compatibility) 84 */ 85 #define FPINV FPIM 86 #define FPDNO FPDM 87 #define FPZDIV FPZM 88 #define FPOVR FPOM 89 #define FPUNR FPUM 90 #define FPPRE FPPM 91 92 /* 93 * precision, rounding, and infinity options in control word 94 */ 95 #define FPSIG24 0x00000000 /* 24-bit significand precision (short) */ 96 #define FPSIG53 0x00000200 /* 53-bit significand precision (long) */ 97 #define FPSIG64 0x00000300 /* 64-bit significand precision (temp) */ 98 #define FPRTN 0x00000000 /* round to nearest or even */ 99 #define FPRD 0x00000400 /* round down */ 100 #define FPRU 0x00000800 /* round up */ 101 #define FPCHOP 0x00000C00 /* chop (truncate toward zero) */ 102 #define FPP 0x00000000 /* projective infinity */ 103 #define FPA 0x00001000 /* affine infinity */ 104 #define WFPB17 0x00020000 /* bit 17 */ 105 #define WFPB24 0x00040000 /* bit 24 */ 106 107 /* 108 * masks for 80387 status word 109 */ 110 #define FPS_IE 0x00000001 /* invalid operation */ 111 #define FPS_DE 0x00000002 /* denormalized operand */ 112 #define FPS_ZE 0x00000004 /* zero divide */ 113 #define FPS_OE 0x00000008 /* overflow */ 114 #define FPS_UE 0x00000010 /* underflow */ 115 #define FPS_PE 0x00000020 /* precision */ 116 #define FPS_SF 0x00000040 /* stack fault */ 117 #define FPS_ES 0x00000080 /* error summary bit */ 118 #define FPS_C0 0x00000100 /* C0 bit */ 119 #define FPS_C1 0x00000200 /* C1 bit */ 120 #define FPS_C2 0x00000400 /* C2 bit */ 121 #define FPS_TOP 0x00003800 /* top of stack pointer */ 122 #define FPS_C3 0x00004000 /* C3 bit */ 123 #define FPS_B 0x00008000 /* busy bit */ 124 125 /* 126 * Exception flags manually cleared during x87 exception handling. 127 */ 128 #define FPS_SW_EFLAGS \ 129 (FPS_IE|FPS_DE|FPS_ZE|FPS_OE|FPS_UE|FPS_PE|FPS_SF|FPS_ES|FPS_B) 130 131 /* 132 * Initial value of FPU control word as per 4th ed. ABI document 133 * - affine infinity 134 * - round to nearest or even 135 * - 64-bit double precision 136 * - all exceptions masked 137 */ 138 #define FPU_CW_INIT 0x133f 139 140 /* 141 * masks and flags for SSE/SSE2 MXCSR 142 */ 143 #define SSE_IE 0x00000001 /* invalid operation */ 144 #define SSE_DE 0x00000002 /* denormalized operand */ 145 #define SSE_ZE 0x00000004 /* zero divide */ 146 #define SSE_OE 0x00000008 /* overflow */ 147 #define SSE_UE 0x00000010 /* underflow */ 148 #define SSE_PE 0x00000020 /* precision */ 149 #define SSE_DAZ 0x00000040 /* denormals are zero */ 150 #define SSE_IM 0x00000080 /* invalid op exception mask */ 151 #define SSE_DM 0x00000100 /* denormalize exception mask */ 152 #define SSE_ZM 0x00000200 /* zero-divide exception mask */ 153 #define SSE_OM 0x00000400 /* overflow exception mask */ 154 #define SSE_UM 0x00000800 /* underflow exception mask */ 155 #define SSE_PM 0x00001000 /* precision exception mask */ 156 #define SSE_RC 0x00006000 /* rounding control */ 157 #define SSE_RD 0x00002000 /* rounding control: round down */ 158 #define SSE_RU 0x00004000 /* rounding control: round up */ 159 #define SSE_FZ 0x00008000 /* flush to zero for masked underflow */ 160 161 #define SSE_MXCSR_EFLAGS \ 162 (SSE_IE|SSE_DE|SSE_ZE|SSE_OE|SSE_UE|SSE_PE) /* 0x3f */ 163 164 #define SSE_MXCSR_INIT \ 165 (SSE_IM|SSE_DM|SSE_ZM|SSE_OM|SSE_UM|SSE_PM) /* 0x1f80 */ 166 167 #define SSE_MXCSR_MASK_DEFAULT \ 168 (0xffff & ~SSE_DAZ) /* 0xffbf */ 169 170 #define SSE_FMT_MXCSR \ 171 "\20\20fz\17ru\16rd\15pm\14um\13om\12zm\11dm" \ 172 "\10im\7daz\6pe\5ue\4oe\3ze\2de\1ie" 173 174 /* 175 * This structure is written to memory by an 'fnsave' instruction 176 */ 177 struct fnsave_state { 178 uint16_t f_fcw; 179 uint16_t __f_ign0; 180 uint16_t f_fsw; 181 uint16_t __f_ign1; 182 uint16_t f_ftw; 183 uint16_t __f_ign2; 184 uint32_t f_eip; 185 uint16_t f_cs; 186 uint16_t f_fop; 187 uint32_t f_dp; 188 uint16_t f_ds; 189 uint16_t __f_ign3; 190 union { 191 uint16_t fpr_16[5]; /* 80-bits of x87 state */ 192 } f_st[8]; 193 }; /* 108 bytes */ 194 195 /* 196 * This structure is written to memory by an 'fxsave' instruction 197 * Note the variant behaviour of this instruction between long mode 198 * and legacy environments! 199 */ 200 struct fxsave_state { 201 uint16_t fx_fcw; 202 uint16_t fx_fsw; 203 uint16_t fx_fctw; /* compressed tag word */ 204 uint16_t fx_fop; 205 #if defined(__amd64) 206 uint64_t fx_rip; 207 uint64_t fx_rdp; 208 #else 209 uint32_t fx_eip; 210 uint16_t fx_cs; 211 uint16_t __fx_ign0; 212 uint32_t fx_dp; 213 uint16_t fx_ds; 214 uint16_t __fx_ign1; 215 #endif 216 uint32_t fx_mxcsr; 217 uint32_t fx_mxcsr_mask; 218 union { 219 uint16_t fpr_16[5]; /* 80-bits of x87 state */ 220 u_longlong_t fpr_mmx; /* 64-bit mmx register */ 221 uint32_t __fpr_pad[4]; /* (pad out to 128-bits) */ 222 } fx_st[8]; 223 #if defined(__amd64) 224 upad128_t fx_xmm[16]; /* 128-bit registers */ 225 upad128_t __fx_ign2[6]; 226 #else 227 upad128_t fx_xmm[8]; /* 128-bit registers */ 228 upad128_t __fx_ign2[14]; 229 #endif 230 }; /* 512 bytes */ 231 232 /* 233 * This structure is written to memory by one of the 'xsave' instruction 234 * variants. The first 512 bytes are compatible with the format of the 'fxsave' 235 * area. The header portion of the xsave layout is documented in section 236 * 13.4.2 of the Intel 64 and IA-32 Architectures Software Developer’s Manual, 237 * Volume 1 (IASDv1). The extended portion is documented in section 13.4.3. 238 * 239 * Our size is at least AVX_XSAVE_SIZE (832 bytes), asserted in fpnoextflt(). 240 * Enabling additional xsave-related CPU features requires an increase in the 241 * size. We dynamically allocate the per-lwp xsave area at runtime, based on 242 * the size needed for the CPU-specific features. This xsave_state structure 243 * simply defines our historical layout for the beginning of the xsave area. The 244 * locations and size of new, extended, components is determined dynamically by 245 * querying the CPU. See the xsave_info structure in cpuid.c. 246 * 247 * xsave component usage is tracked using bits in the xs_xstate_bv field. The 248 * components are documented in section 13.1 of IASDv1. For easy reference, 249 * this is a summary of the currently defined component bit definitions: 250 * x87 0x0001 251 * SSE 0x0002 252 * AVX 0x0004 253 * bndreg (MPX) 0x0008 254 * bndcsr (MPX) 0x0010 255 * opmask (AVX512) 0x0020 256 * zmm hi256 (AVX512) 0x0040 257 * zmm hi16 (AVX512) 0x0080 258 * PT 0x0100 259 * PKRU 0x0200 260 * When xsaveopt_ctxt is being used to save into the xsave_state area, the 261 * xs_xstate_bv field is updated by the xsaveopt instruction to indicate which 262 * elements of the xsave area are active. 263 * 264 * xs_xcomp_bv should always be 0, since we do not currently use the compressed 265 * form of xsave (xsavec). 266 */ 267 struct xsave_state { 268 struct fxsave_state xs_fxsave; /* 0-511 legacy region */ 269 uint64_t xs_xstate_bv; /* 512-519 start xsave header */ 270 uint64_t xs_xcomp_bv; /* 520-527 */ 271 uint64_t xs_reserved[6]; /* 528-575 end xsave header */ 272 upad128_t xs_ymm[16]; /* 576 AVX component */ 273 }; 274 275 /* 276 * Kernel's FPU save area 277 */ 278 typedef struct { 279 union _kfpu_u { 280 void *kfpu_generic; 281 struct fxsave_state *kfpu_fx; 282 #if defined(__i386) 283 struct fnsave_state *kfpu_fn; 284 #endif 285 struct xsave_state *kfpu_xs; 286 } kfpu_u; 287 uint32_t kfpu_status; /* saved at #mf exception */ 288 uint32_t kfpu_xstatus; /* saved at #xm exception */ 289 } kfpu_t; 290 291 extern int fp_kind; /* kind of fp support */ 292 extern int fp_save_mech; /* fp save/restore mechanism */ 293 extern int fpu_exists; /* FPU hw exists */ 294 295 #ifdef _KERNEL 296 297 extern int fpu_ignored; 298 extern int fpu_pentium_fdivbug; 299 300 extern uint32_t sse_mxcsr_mask; 301 302 extern void fpu_probe(void); 303 extern uint_t fpu_initial_probe(void); 304 extern int fpu_probe_pentium_fdivbug(void); 305 306 extern void fpnsave_ctxt(void *); 307 extern void fpxsave_ctxt(void *); 308 extern void xsave_ctxt(void *); 309 extern void xsaveopt_ctxt(void *); 310 extern void fpxsave_excp_clr_ctxt(void *); 311 extern void xsave_excp_clr_ctxt(void *); 312 extern void xsaveopt_excp_clr_ctxt(void *); 313 extern void (*fpsave_ctxt)(void *); 314 extern void (*xsavep)(struct xsave_state *, uint64_t); 315 316 extern void fxsave_insn(struct fxsave_state *); 317 extern void fpsave(struct fnsave_state *); 318 extern void fprestore(struct fnsave_state *); 319 extern void fpxsave(struct fxsave_state *); 320 extern void fpxrestore(struct fxsave_state *); 321 extern void xsave(struct xsave_state *, uint64_t); 322 extern void xsaveopt(struct xsave_state *, uint64_t); 323 extern void xrestore(struct xsave_state *, uint64_t); 324 325 extern void fpenable(void); 326 extern void fpdisable(void); 327 extern void fpinit(void); 328 329 extern uint32_t fperr_reset(void); 330 extern uint32_t fpxerr_reset(void); 331 332 extern uint32_t fpgetcwsw(void); 333 extern uint32_t fpgetmxcsr(void); 334 335 struct regs; 336 extern int fpnoextflt(struct regs *); 337 extern int fpextovrflt(struct regs *); 338 extern int fpexterrflt(struct regs *); 339 extern int fpsimderrflt(struct regs *); 340 extern void fpsetcw(uint16_t, uint32_t); 341 struct _klwp; 342 extern void fp_lwp_init(struct _klwp *); 343 extern void fp_lwp_cleanup(struct _klwp *); 344 extern void fp_lwp_dup(struct _klwp *); 345 346 #endif /* _KERNEL */ 347 348 #ifdef __cplusplus 349 } 350 #endif 351 352 #endif /* _SYS_FP_H */ 353