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 /* Layout for code/data pointers relating to FP exceptions. */ 98 union fp_addr { 99 uint64_t fa_64; /* linear address for 64-bit system */ 100 struct { 101 uint32_t fa_off; /* linear address for 32-bit system */ 102 uint16_t fa_seg; /* code/data (etc) segment */ 103 uint16_t fa_opcode; /* last opcode (sometimes) */ 104 } fa_32; 105 }; 106 107 struct envxmm64 { 108 uint16_t en_cw; /* control word (16bits) */ 109 uint16_t en_sw; /* status word (16bits) */ 110 uint8_t en_tw; /* tag word (8bits) */ 111 uint8_t en_zero; 112 uint16_t en_opcode; /* opcode last executed (11 bits ) */ 113 union { /* fp instruction pointer */ 114 uint64_t en_rip; 115 union fp_addr en_ip64; 116 }; 117 union { /* fp operand pointer */ 118 uint64_t en_rdp; 119 union fp_addr en_dp64; 120 }; 121 uint32_t en_mxcsr; /* SSE control/status register */ 122 uint32_t en_mxcsr_mask; /* valid bits in mxcsr */ 123 }; 124 125 /* Contents of each SSE extended accumulator */ 126 struct xmmacc { 127 uint8_t xmm_bytes[16]; 128 }; 129 130 /* Contents of the upper 16 bytes of each AVX extended accumulator */ 131 struct ymmacc { 132 uint8_t ymm_bytes[16]; 133 }; 134 135 /* 136 * Floating point context. (i386 fxsave/fxrstor) 137 * savexmm is a 512-byte structure 138 */ 139 struct savexmm { 140 struct envxmm sv_env; /* 32 */ 141 struct { 142 struct fpacc87 fp_acc; 143 uint8_t fp_pad[6]; 144 } sv_fp[8]; /* 128 */ 145 struct xmmacc sv_xmm[8]; /* 128 */ 146 uint8_t sv_pad[224]; /* 224 */ 147 } __aligned(16); 148 149 /* 150 * Floating point context. (amd64 fxsave/fxrstor) 151 * savexmm64 is a 512-byte structure 152 */ 153 struct savexmm64 { 154 struct envxmm64 sv_env; /* 32 */ 155 struct { 156 struct fpacc87 fp_acc; 157 uint8_t fp_pad[6]; 158 } sv_fp[8]; /* 128 */ 159 struct xmmacc sv_xmm[16]; /* 256 */ 160 uint8_t sv_pad[96]; /* 96 */ 161 } __aligned(16); 162 163 /* xstate_hdr is a 64-byte structure */ 164 struct xstate_hdr { 165 uint64_t xstate_bv; 166 uint64_t xstate_xcomp_bv; 167 uint8_t xstate_rsrv0[8]; 168 uint8_t xstate_rsrv[40]; 169 }; 170 #define XSTATE_XCOMP_BV_COMPACT (1ULL << 63) 171 172 /* savexmm_xstate is a 320-byte structure (64 + 256) */ 173 struct savexmm_xstate { 174 struct xstate_hdr sx_hd; 175 struct ymmacc sx_ymm[16]; 176 }; 177 178 /* saveymm is a 832-byte structure (i386) */ 179 struct saveymm { 180 struct envxmm sv_env; /* 32 */ 181 struct { 182 struct fpacc87 fp_acc; 183 uint8_t fp_pad[6]; 184 } sv_fp[8]; /* 128 */ 185 struct xmmacc sv_xmm[16]; /* 256 */ 186 uint8_t sv_pad[96]; /* 96 */ 187 struct savexmm_xstate sv_xstate; /* 320 */ 188 } __aligned(64); 189 190 /* saveymm64 is a 832-byte structure (amd64) */ 191 struct saveymm64 { 192 struct envxmm64 sv_env; /* 32 */ 193 struct { 194 struct fpacc87 fp_acc; 195 int8_t fp_pad[6]; 196 } sv_fp[8]; /* 128 */ 197 struct xmmacc sv_xmm[16]; /* 256 */ 198 uint8_t sv_pad[96]; /* 96 */ 199 struct savexmm_xstate sv_xstate; /* 320 */ 200 } __aligned(64); 201 202 union savefpu { 203 struct save87 sv_87; 204 struct savexmm sv_xmm; 205 struct saveymm sv_ymm; 206 struct savexmm64 sv_xmm64; 207 struct saveymm64 sv_ymm64; 208 char sv_savearea[1024]; /* see mcontext_t */ 209 }; 210 211 /* 212 * The hardware default control word for i387's and later coprocessors is 213 * 0x37F, giving: 214 * 215 * round to nearest 216 * 64-bit precision 217 * all exceptions masked. 218 * 219 * We modify the affine mode bit and precision bits in this to give: 220 * 221 * affine mode for 287's (if they work at all) (1 in bitfield 1<<12) 222 * 53-bit precision (2 in bitfield 3<<8) 223 * 224 * 64-bit precision often gives bad results with high level languages 225 * because it makes the results of calculations depend on whether 226 * intermediate values are stored in memory or in FPU registers. 227 */ 228 #define __INITIAL_NPXCW__ 0x127F 229 230 #define __INITIAL_FPUCW__ 0x037F /* used by libm/arch/x86_64/fenv.c */ 231 #define __INITIAL_FPUCW_I386__ 0x127F 232 #define __INITIAL_MXCSR__ 0x1F80 /* used by libm/arch/x86_64/fenv.c */ 233 #define __INITIAL_MXCSR_MASK__ 0xFFBF 234 235 #ifdef _KERNEL 236 237 struct proc; 238 struct trapframe; 239 240 extern uint32_t npx_mxcsr_mask; 241 extern uint64_t npx_xcr0_mask; 242 243 void npxprobemask (void); 244 void npxexit (void); 245 void npxinit (void); 246 void npxsave (union savefpu *addr); 247 void fpusave (union savefpu *addr, uint64_t mask); 248 void fpurstor (union savefpu *addr, uint64_t mask); 249 #endif 250 251 #endif /* !_CPU_NPX_H_ */ 252