1 /*- 2 * Copyright (c) 1990 The Regents of the University of California. 3 * Copyright (c) 2008 The DragonFly Project. 4 * All rights reserved. 5 * 6 * This code is derived from software contributed to Berkeley by 7 * William Jolitz. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 1. Redistributions of source code must retain the above copyright 13 * notice, this list of conditions and the following disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in the 16 * documentation and/or other materials provided with the distribution. 17 * 3. Neither the name of the University nor the names of its contributors 18 * may be used to endorse or promote products derived from this software 19 * without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 * 33 * from: @(#)npx.h 5.3 (Berkeley) 1/18/91 34 * $FreeBSD: src/sys/i386/include/npx.h,v 1.18.2.1 2001/08/15 01:23:52 peter Exp $ 35 */ 36 37 /* 38 * 287/387 NPX Coprocessor Data Structures and Constants 39 * W. Jolitz 1/90 40 */ 41 42 #ifndef _CPU_NPX_H_ 43 #define _CPU_NPX_H_ 44 45 #ifndef _SYS_STDINT_H_ 46 #include <sys/stdint.h> 47 #endif 48 49 /* Environment information of floating point unit */ 50 struct env87 { 51 int32_t en_cw; /* control word (16bits) */ 52 int32_t en_sw; /* status word (16bits) */ 53 int32_t en_tw; /* tag word (16bits) */ 54 int32_t en_fip; /* floating point instruction pointer */ 55 uint16_t en_fcs; /* floating code segment selector */ 56 uint16_t en_opcode; /* opcode last executed (11 bits ) */ 57 int32_t en_foo; /* floating operand offset */ 58 int32_t en_fos; /* floating operand segment selector */ 59 }; 60 61 /* Contents of each x87 floating point accumulator */ 62 struct fpacc87 { 63 uint8_t fp_bytes[10]; 64 }; 65 66 /* Floating point context (i386 fnsave/frstor) */ 67 struct save87 { 68 struct env87 sv_env; /* floating point control/status */ 69 struct fpacc87 sv_ac[8]; /* accumulator contents, 0-7 */ 70 uint8_t sv_pad0[4]; /* saved status word (now unused) */ 71 /* 72 * Bogus padding for emulators. Emulators should use their own 73 * struct and arrange to store into this struct (ending here) 74 * before it is inspected for ptracing or for core dumps. Some 75 * emulators overwrite the whole struct. We have no good way of 76 * knowing how much padding to leave. Leave just enough for the 77 * GPL emulator's i387_union (176 bytes total). 78 */ 79 uint8_t sv_pad[64]; 80 }; 81 82 struct envxmm { 83 uint16_t en_cw; /* control word (16bits) */ 84 uint16_t en_sw; /* status word (16bits) */ 85 uint16_t en_tw; /* tag word (16bits) */ 86 uint16_t en_opcode; /* opcode last executed (11 bits) */ 87 uint32_t en_fip; /* fp instruction pointer */ 88 uint16_t en_fcs; /* fp code segment selector */ 89 uint16_t en_pad0; /* padding */ 90 uint32_t en_foo; /* fp operand offset */ 91 uint16_t en_fos; /* fp operand segment selector */ 92 uint16_t en_pad1; /* padding */ 93 uint32_t en_mxcsr; /* SSE control/status register */ 94 uint32_t en_mxcsr_mask; /* valid bits in mxcsr */ 95 }; 96 97 struct envxmm64 { 98 uint16_t en_cw; /* control word (16bits) */ 99 uint16_t en_sw; /* status word (16bits) */ 100 uint8_t en_tw; /* tag word (8bits) */ 101 uint8_t en_zero; 102 uint16_t en_opcode; /* opcode last executed (11 bits ) */ 103 uint64_t en_rip; /* fp instruction pointer */ 104 uint64_t en_rdp; /* fp operand pointer */ 105 uint32_t en_mxcsr; /* SSE control/status register */ 106 uint32_t en_mxcsr_mask; /* valid bits in mxcsr */ 107 }; 108 109 /* Contents of each SSE extended accumulator */ 110 struct xmmacc { 111 uint8_t xmm_bytes[16]; 112 }; 113 114 /* Contents of the upper 16 bytes of each AVX extended accumulator */ 115 struct ymmacc { 116 uint8_t ymm_bytes[16]; 117 }; 118 119 /* 120 * Floating point context. (i386 fxsave/fxrstor) 121 * savexmm is a 512-byte structure 122 */ 123 struct savexmm { 124 struct envxmm sv_env; /* 32 */ 125 struct { 126 struct fpacc87 fp_acc; /* -- 10 */ 127 uint8_t fp_pad[6]; /* -- 6 */ 128 } sv_fp[8]; /* 128 */ 129 struct xmmacc sv_xmm[8]; /* 128 */ 130 uint8_t sv_pad[224]; /* 224 (padding) */ 131 } __aligned(16); 132 133 /* 134 * Floating point context. (amd64 fxsave/fxrstor) 135 * savexmm64 is a 512-byte structure 136 */ 137 struct savexmm64 { 138 struct envxmm64 sv_env; /* 32 */ 139 struct { 140 struct fpacc87 fp_acc; 141 uint8_t fp_pad[6]; 142 } sv_fp[8]; /* 128 */ 143 struct xmmacc sv_xmm[8]; /* 128 */ 144 uint8_t sv_pad[224]; /* 224 */ 145 } __aligned(16); 146 147 /* xstate_hdr is a 64-byte structure */ 148 struct xstate_hdr { 149 uint64_t xstate_bv; 150 uint64_t xstate_xcomp_bv; 151 uint8_t xstate_rsrv0[8]; 152 uint8_t xstate_rsrv[40]; 153 }; 154 #define XSTATE_XCOMP_BV_COMPACT (1ULL << 63) 155 156 /* savexmm_xstate is a 320-byte structure (64 + 256) */ 157 struct savexmm_xstate { 158 struct xstate_hdr sx_hd; 159 struct ymmacc sx_ymm[16]; 160 }; 161 162 /* saveymm is a 832-byte structure (i386) */ 163 struct saveymm { 164 struct envxmm sv_env; /* 32 */ 165 struct { 166 struct fpacc87 fp_acc; 167 uint8_t fp_pad[6]; 168 } sv_fp[8]; /* 128 */ 169 struct xmmacc sv_xmm[16]; /* 256 */ 170 uint8_t sv_pad[96]; /* 96 */ 171 struct savexmm_xstate sv_xstate; /* 320 */ 172 } __aligned(64); 173 174 /* saveymm64 is a 832-byte structure (amd64) */ 175 struct saveymm64 { 176 struct envxmm64 sv_env; /* 32 */ 177 struct { 178 struct fpacc87 fp_acc; 179 int8_t fp_pad[6]; 180 } sv_fp[8]; /* 128 */ 181 struct xmmacc sv_xmm[16]; /* 256 */ 182 uint8_t sv_pad[96]; /* 96 */ 183 struct savexmm_xstate sv_xstate; /* 320 */ 184 } __aligned(64); 185 186 union savefpu { 187 struct save87 sv_87; 188 struct savexmm sv_xmm; 189 struct saveymm sv_ymm; 190 struct savexmm64 sv_xmm64; 191 struct saveymm64 sv_ymm64; 192 char sv_savearea[1024]; /* see mcontext_t */ 193 }; 194 195 /* 196 * The hardware default control word for i387's and later coprocessors is 197 * 0x37F, giving: 198 * 199 * round to nearest 200 * 64-bit precision 201 * all exceptions masked. 202 * 203 * We modify the affine mode bit and precision bits in this to give: 204 * 205 * affine mode for 287's (if they work at all) (1 in bitfield 1<<12) 206 * 53-bit precision (2 in bitfield 3<<8) 207 * 208 * 64-bit precision often gives bad results with high level languages 209 * because it makes the results of calculations depend on whether 210 * intermediate values are stored in memory or in FPU registers. 211 */ 212 #define __INITIAL_NPXCW__ 0x127F 213 214 #define __INITIAL_FPUCW__ 0x037F /* used by libm/arch/x86_64/fenv.c */ 215 #define __INITIAL_FPUCW_I386__ 0x127F 216 #define __INITIAL_MXCSR__ 0x1F80 /* used by libm/arch/x86_64/fenv.c */ 217 #define __INITIAL_MXCSR_MASK__ 0xFFBF 218 219 #ifdef _KERNEL 220 221 struct proc; 222 struct trapframe; 223 224 extern uint32_t npx_mxcsr_mask; 225 extern uint64_t npx_xcr0_mask; 226 227 void npxprobemask (void); 228 void npxexit (void); 229 void npxinit (void); 230 void npxsave (union savefpu *addr); 231 #endif 232 233 #endif /* !_CPU_NPX_H_ */ 234