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 #include <stdint.h> 46 47 /* Environment information of floating point unit */ 48 struct env87 { 49 int32_t en_cw; /* control word (16bits) */ 50 int32_t en_sw; /* status word (16bits) */ 51 int32_t en_tw; /* tag word (16bits) */ 52 int32_t en_fip; /* floating point instruction pointer */ 53 uint16_t en_fcs; /* floating code segment selector */ 54 uint16_t en_opcode; /* opcode last executed (11 bits ) */ 55 int32_t en_foo; /* floating operand offset */ 56 int32_t en_fos; /* floating operand segment selector */ 57 }; 58 59 /* Contents of each x87 floating point accumulator */ 60 struct fpacc87 { 61 uint8_t fp_bytes[10]; 62 }; 63 64 /* Floating point context (i386 fnsave/frstor) */ 65 struct save87 { 66 struct env87 sv_env; /* floating point control/status */ 67 struct fpacc87 sv_ac[8]; /* accumulator contents, 0-7 */ 68 uint8_t sv_pad0[4]; /* saved status word (now unused) */ 69 /* 70 * Bogus padding for emulators. Emulators should use their own 71 * struct and arrange to store into this struct (ending here) 72 * before it is inspected for ptracing or for core dumps. Some 73 * emulators overwrite the whole struct. We have no good way of 74 * knowing how much padding to leave. Leave just enough for the 75 * GPL emulator's i387_union (176 bytes total). 76 */ 77 uint8_t sv_pad[64]; 78 }; 79 80 struct envxmm { 81 uint16_t en_cw; /* control word (16bits) */ 82 uint16_t en_sw; /* status word (16bits) */ 83 uint16_t en_tw; /* tag word (16bits) */ 84 uint16_t en_opcode; /* opcode last executed (11 bits) */ 85 uint32_t en_fip; /* fp instruction pointer */ 86 uint16_t en_fcs; /* fp code segment selector */ 87 uint16_t en_pad0; /* padding */ 88 uint32_t en_foo; /* fp operand offset */ 89 uint16_t en_fos; /* fp operand segment selector */ 90 uint16_t en_pad1; /* padding */ 91 uint32_t en_mxcsr; /* SSE control/status register */ 92 uint32_t en_mxcsr_mask; /* valid bits in mxcsr */ 93 }; 94 95 struct envxmm64 { 96 uint16_t en_cw; /* control word (16bits) */ 97 uint16_t en_sw; /* status word (16bits) */ 98 uint8_t en_tw; /* tag word (8bits) */ 99 uint8_t en_zero; 100 uint16_t en_opcode; /* opcode last executed (11 bits ) */ 101 uint64_t en_rip; /* fp instruction pointer */ 102 uint64_t en_rdp; /* fp operand pointer */ 103 uint32_t en_mxcsr; /* SSE control/status register */ 104 uint32_t en_mxcsr_mask; /* valid bits in mxcsr */ 105 }; 106 107 /* Contents of each SSE extended accumulator */ 108 struct xmmacc { 109 uint8_t xmm_bytes[16]; 110 }; 111 112 /* Contents of the upper 16 bytes of each AVX extended accumulator */ 113 struct ymmacc { 114 uint8_t ymm_bytes[16]; 115 }; 116 117 /* 118 * Floating point context. (i386 fxsave/fxrstor) 119 * savexmm is a 512-byte structure 120 */ 121 struct savexmm { 122 struct envxmm sv_env; /* 32 */ 123 struct { 124 struct fpacc87 fp_acc; /* -- 10 */ 125 uint8_t fp_pad[6]; /* -- 6 */ 126 } sv_fp[8]; /* 128 */ 127 struct xmmacc sv_xmm[8]; /* 128 */ 128 uint8_t sv_pad[224]; /* 224 (padding) */ 129 } __attribute__((aligned(16))); 130 131 /* 132 * Floating point context. (amd64 fxsave/fxrstor) 133 * savexmm64 is a 512-byte structure 134 */ 135 struct savexmm64 { 136 struct envxmm64 sv_env; /* 32 */ 137 struct { 138 struct fpacc87 fp_acc; 139 uint8_t fp_pad[6]; 140 } sv_fp[8]; /* 128 */ 141 struct xmmacc sv_xmm[8]; /* 128 */ 142 uint8_t sv_pad[224]; /* 224 */ 143 } __attribute__((aligned(16))); 144 145 /* xstate_hdr is a 64-byte structure */ 146 struct xstate_hdr { 147 uint64_t xstate_bv; 148 uint64_t xstate_xcomp_bv; 149 uint8_t xstate_rsrv0[8]; 150 uint8_t xstate_rsrv[40]; 151 }; 152 #define XSTATE_XCOMP_BV_COMPACT (1ULL << 63) 153 154 /* savexmm_xstate is a 320-byte structure (64 + 256) */ 155 struct savexmm_xstate { 156 struct xstate_hdr sx_hd; 157 struct ymmacc sx_ymm[16]; 158 }; 159 160 /* saveymm is a 832-byte structure (i386) */ 161 struct saveymm { 162 struct envxmm sv_env; /* 32 */ 163 struct { 164 struct fpacc87 fp_acc; 165 uint8_t fp_pad[6]; 166 } sv_fp[8]; /* 128 */ 167 struct xmmacc sv_xmm[16]; /* 256 */ 168 uint8_t sv_pad[96]; /* 96 */ 169 struct savexmm_xstate sv_xstate; /* 320 */ 170 } __attribute__((aligned(64))); 171 172 /* saveymm64 is a 832-byte structure (amd64) */ 173 struct saveymm64 { 174 struct envxmm64 sv_env; /* 32 */ 175 struct { 176 struct fpacc87 fp_acc; 177 int8_t fp_pad[6]; 178 } sv_fp[8]; /* 128 */ 179 struct xmmacc sv_xmm[16]; /* 256 */ 180 uint8_t sv_pad[96]; /* 96 */ 181 struct savexmm_xstate sv_xstate; /* 320 */ 182 } __attribute__((aligned(64))); 183 184 union savefpu { 185 struct save87 sv_87; 186 struct savexmm sv_xmm; 187 struct saveymm sv_ymm; 188 struct savexmm64 sv_xmm64; 189 struct saveymm64 sv_ymm64; 190 char sv_savearea[1024]; /* see mcontext_t */ 191 }; 192 193 /* 194 * The hardware default control word for i387's and later coprocessors is 195 * 0x37F, giving: 196 * 197 * round to nearest 198 * 64-bit precision 199 * all exceptions masked. 200 * 201 * We modify the affine mode bit and precision bits in this to give: 202 * 203 * affine mode for 287's (if they work at all) (1 in bitfield 1<<12) 204 * 53-bit precision (2 in bitfield 3<<8) 205 * 206 * 64-bit precision often gives bad results with high level languages 207 * because it makes the results of calculations depend on whether 208 * intermediate values are stored in memory or in FPU registers. 209 */ 210 #define __INITIAL_NPXCW__ 0x127F 211 212 #define __INITIAL_FPUCW__ 0x037F /* used by libm/arch/x86_64/fenv.c */ 213 #define __INITIAL_FPUCW_I386__ 0x127F 214 #define __INITIAL_MXCSR__ 0x1F80 /* used by libm/arch/x86_64/fenv.c */ 215 #define __INITIAL_MXCSR_MASK__ 0xFFBF 216 217 #ifdef _KERNEL 218 219 struct proc; 220 struct trapframe; 221 222 extern uint32_t npx_mxcsr_mask; 223 224 void npxprobemask (void); 225 void npxexit (void); 226 void npxinit (void); 227 void npxsave (union savefpu *addr); 228 #endif 229 230 #endif /* !_CPU_NPX_H_ */ 231