1 /* 2 * FPU data structures: 3 */ 4 #ifndef _ASM_X86_FPU_H 5 #define _ASM_X86_FPU_H 6 7 /* 8 * The legacy x87 FPU state format, as saved by FSAVE and 9 * restored by the FRSTOR instructions: 10 */ 11 struct fregs_state { 12 u32 cwd; /* FPU Control Word */ 13 u32 swd; /* FPU Status Word */ 14 u32 twd; /* FPU Tag Word */ 15 u32 fip; /* FPU IP Offset */ 16 u32 fcs; /* FPU IP Selector */ 17 u32 foo; /* FPU Operand Pointer Offset */ 18 u32 fos; /* FPU Operand Pointer Selector */ 19 20 /* 8*10 bytes for each FP-reg = 80 bytes: */ 21 u32 st_space[20]; 22 23 /* Software status information [not touched by FSAVE]: */ 24 u32 status; 25 }; 26 27 /* 28 * The legacy fx SSE/MMX FPU state format, as saved by FXSAVE and 29 * restored by the FXRSTOR instructions. It's similar to the FSAVE 30 * format, but differs in some areas, plus has extensions at 31 * the end for the XMM registers. 32 */ 33 struct fxregs_state { 34 u16 cwd; /* Control Word */ 35 u16 swd; /* Status Word */ 36 u16 twd; /* Tag Word */ 37 u16 fop; /* Last Instruction Opcode */ 38 union { 39 struct { 40 u64 rip; /* Instruction Pointer */ 41 u64 rdp; /* Data Pointer */ 42 }; 43 struct { 44 u32 fip; /* FPU IP Offset */ 45 u32 fcs; /* FPU IP Selector */ 46 u32 foo; /* FPU Operand Offset */ 47 u32 fos; /* FPU Operand Selector */ 48 }; 49 }; 50 u32 mxcsr; /* MXCSR Register State */ 51 u32 mxcsr_mask; /* MXCSR Mask */ 52 53 /* 8*16 bytes for each FP-reg = 128 bytes: */ 54 u32 st_space[32]; 55 56 /* 16*16 bytes for each XMM-reg = 256 bytes: */ 57 u32 xmm_space[64]; 58 59 u32 padding[12]; 60 61 union { 62 u32 padding1[12]; 63 u32 sw_reserved[12]; 64 }; 65 66 } __attribute__((aligned(16))); 67 68 /* Default value for fxregs_state.mxcsr: */ 69 #define MXCSR_DEFAULT 0x1f80 70 71 /* 72 * Software based FPU emulation state. This is arbitrary really, 73 * it matches the x87 format to make it easier to understand: 74 */ 75 struct swregs_state { 76 u32 cwd; 77 u32 swd; 78 u32 twd; 79 u32 fip; 80 u32 fcs; 81 u32 foo; 82 u32 fos; 83 /* 8*10 bytes for each FP-reg = 80 bytes: */ 84 u32 st_space[20]; 85 u8 ftop; 86 u8 changed; 87 u8 lookahead; 88 u8 no_update; 89 u8 rm; 90 u8 alimit; 91 struct math_emu_info *info; 92 u32 entry_eip; 93 }; 94 95 /* 96 * List of XSAVE features Linux knows about: 97 */ 98 enum xfeature { 99 XFEATURE_FP, 100 XFEATURE_SSE, 101 /* 102 * Values above here are "legacy states". 103 * Those below are "extended states". 104 */ 105 XFEATURE_YMM, 106 XFEATURE_BNDREGS, 107 XFEATURE_BNDCSR, 108 XFEATURE_OPMASK, 109 XFEATURE_ZMM_Hi256, 110 XFEATURE_Hi16_ZMM, 111 XFEATURE_PT_UNIMPLEMENTED_SO_FAR, 112 XFEATURE_PKRU, 113 114 XFEATURE_MAX, 115 }; 116 117 #define XFEATURE_MASK_FP (1 << XFEATURE_FP) 118 #define XFEATURE_MASK_SSE (1 << XFEATURE_SSE) 119 #define XFEATURE_MASK_YMM (1 << XFEATURE_YMM) 120 #define XFEATURE_MASK_BNDREGS (1 << XFEATURE_BNDREGS) 121 #define XFEATURE_MASK_BNDCSR (1 << XFEATURE_BNDCSR) 122 #define XFEATURE_MASK_OPMASK (1 << XFEATURE_OPMASK) 123 #define XFEATURE_MASK_ZMM_Hi256 (1 << XFEATURE_ZMM_Hi256) 124 #define XFEATURE_MASK_Hi16_ZMM (1 << XFEATURE_Hi16_ZMM) 125 #define XFEATURE_MASK_PT (1 << XFEATURE_PT_UNIMPLEMENTED_SO_FAR) 126 #define XFEATURE_MASK_PKRU (1 << XFEATURE_PKRU) 127 128 #define XFEATURE_MASK_FPSSE (XFEATURE_MASK_FP | XFEATURE_MASK_SSE) 129 #define XFEATURE_MASK_AVX512 (XFEATURE_MASK_OPMASK \ 130 | XFEATURE_MASK_ZMM_Hi256 \ 131 | XFEATURE_MASK_Hi16_ZMM) 132 133 #define FIRST_EXTENDED_XFEATURE XFEATURE_YMM 134 135 struct reg_128_bit { 136 u8 regbytes[128/8]; 137 }; 138 struct reg_256_bit { 139 u8 regbytes[256/8]; 140 }; 141 struct reg_512_bit { 142 u8 regbytes[512/8]; 143 }; 144 145 /* 146 * State component 2: 147 * 148 * There are 16x 256-bit AVX registers named YMM0-YMM15. 149 * The low 128 bits are aliased to the 16 SSE registers (XMM0-XMM15) 150 * and are stored in 'struct fxregs_state::xmm_space[]' in the 151 * "legacy" area. 152 * 153 * The high 128 bits are stored here. 154 */ 155 struct ymmh_struct { 156 struct reg_128_bit hi_ymm[16]; 157 } __packed; 158 159 /* Intel MPX support: */ 160 161 struct mpx_bndreg { 162 u64 lower_bound; 163 u64 upper_bound; 164 } __packed; 165 /* 166 * State component 3 is used for the 4 128-bit bounds registers 167 */ 168 struct mpx_bndreg_state { 169 struct mpx_bndreg bndreg[4]; 170 } __packed; 171 172 /* 173 * State component 4 is used for the 64-bit user-mode MPX 174 * configuration register BNDCFGU and the 64-bit MPX status 175 * register BNDSTATUS. We call the pair "BNDCSR". 176 */ 177 struct mpx_bndcsr { 178 u64 bndcfgu; 179 u64 bndstatus; 180 } __packed; 181 182 /* 183 * The BNDCSR state is padded out to be 64-bytes in size. 184 */ 185 struct mpx_bndcsr_state { 186 union { 187 struct mpx_bndcsr bndcsr; 188 u8 pad_to_64_bytes[64]; 189 }; 190 } __packed; 191 192 /* AVX-512 Components: */ 193 194 /* 195 * State component 5 is used for the 8 64-bit opmask registers 196 * k0-k7 (opmask state). 197 */ 198 struct avx_512_opmask_state { 199 u64 opmask_reg[8]; 200 } __packed; 201 202 /* 203 * State component 6 is used for the upper 256 bits of the 204 * registers ZMM0-ZMM15. These 16 256-bit values are denoted 205 * ZMM0_H-ZMM15_H (ZMM_Hi256 state). 206 */ 207 struct avx_512_zmm_uppers_state { 208 struct reg_256_bit zmm_upper[16]; 209 } __packed; 210 211 /* 212 * State component 7 is used for the 16 512-bit registers 213 * ZMM16-ZMM31 (Hi16_ZMM state). 214 */ 215 struct avx_512_hi16_state { 216 struct reg_512_bit hi16_zmm[16]; 217 } __packed; 218 219 /* 220 * State component 9: 32-bit PKRU register. The state is 221 * 8 bytes long but only 4 bytes is used currently. 222 */ 223 struct pkru_state { 224 u32 pkru; 225 u32 pad; 226 } __packed; 227 228 struct xstate_header { 229 u64 xfeatures; 230 u64 xcomp_bv; 231 u64 reserved[6]; 232 } __attribute__((packed)); 233 234 /* 235 * xstate_header.xcomp_bv[63] indicates that the extended_state_area 236 * is in compacted format. 237 */ 238 #define XCOMP_BV_COMPACTED_FORMAT ((u64)1 << 63) 239 240 /* 241 * This is our most modern FPU state format, as saved by the XSAVE 242 * and restored by the XRSTOR instructions. 243 * 244 * It consists of a legacy fxregs portion, an xstate header and 245 * subsequent areas as defined by the xstate header. Not all CPUs 246 * support all the extensions, so the size of the extended area 247 * can vary quite a bit between CPUs. 248 */ 249 struct xregs_state { 250 struct fxregs_state i387; 251 struct xstate_header header; 252 u8 extended_state_area[0]; 253 } __attribute__ ((packed, aligned (64))); 254 255 /* 256 * This is a union of all the possible FPU state formats 257 * put together, so that we can pick the right one runtime. 258 * 259 * The size of the structure is determined by the largest 260 * member - which is the xsave area. The padding is there 261 * to ensure that statically-allocated task_structs (just 262 * the init_task today) have enough space. 263 */ 264 union fpregs_state { 265 struct fregs_state fsave; 266 struct fxregs_state fxsave; 267 struct swregs_state soft; 268 struct xregs_state xsave; 269 u8 __padding[PAGE_SIZE]; 270 }; 271 272 /* 273 * Highest level per task FPU state data structure that 274 * contains the FPU register state plus various FPU 275 * state fields: 276 */ 277 struct fpu { 278 /* 279 * @last_cpu: 280 * 281 * Records the last CPU on which this context was loaded into 282 * FPU registers. (In the lazy-restore case we might be 283 * able to reuse FPU registers across multiple context switches 284 * this way, if no intermediate task used the FPU.) 285 * 286 * A value of -1 is used to indicate that the FPU state in context 287 * memory is newer than the FPU state in registers, and that the 288 * FPU state should be reloaded next time the task is run. 289 */ 290 unsigned int last_cpu; 291 292 /* 293 * @fpstate_active: 294 * 295 * This flag indicates whether this context is active: if the task 296 * is not running then we can restore from this context, if the task 297 * is running then we should save into this context. 298 */ 299 unsigned char fpstate_active; 300 301 /* 302 * @fpregs_active: 303 * 304 * This flag determines whether a given context is actively 305 * loaded into the FPU's registers and that those registers 306 * represent the task's current FPU state. 307 * 308 * Note the interaction with fpstate_active: 309 * 310 * # task does not use the FPU: 311 * fpstate_active == 0 312 * 313 * # task uses the FPU and regs are active: 314 * fpstate_active == 1 && fpregs_active == 1 315 * 316 * # the regs are inactive but still match fpstate: 317 * fpstate_active == 1 && fpregs_active == 0 && fpregs_owner == fpu 318 * 319 * The third state is what we use for the lazy restore optimization 320 * on lazy-switching CPUs. 321 */ 322 unsigned char fpregs_active; 323 324 /* 325 * @state: 326 * 327 * In-memory copy of all FPU registers that we save/restore 328 * over context switches. If the task is using the FPU then 329 * the registers in the FPU are more recent than this state 330 * copy. If the task context-switches away then they get 331 * saved here and represent the FPU state. 332 */ 333 union fpregs_state state; 334 /* 335 * WARNING: 'state' is dynamically-sized. Do not put 336 * anything after it here. 337 */ 338 }; 339 340 #endif /* _ASM_X86_FPU_H */ 341