| 1 | // SPDX-License-Identifier: GPL-2.0 |
| 2 | /* |
| 3 | * FPU register's regset abstraction, for ptrace, core dumps, etc. |
| 4 | */ |
| 5 | #include <linux/sched/task_stack.h> |
| 6 | #include <linux/vmalloc.h> |
| 7 | |
| 8 | #include <asm/fpu/api.h> |
| 9 | #include <asm/fpu/signal.h> |
| 10 | #include <asm/fpu/regset.h> |
| 11 | #include <asm/prctl.h> |
| 12 | |
| 13 | #include "context.h" |
| 14 | #include "internal.h" |
| 15 | #include "legacy.h" |
| 16 | #include "xstate.h" |
| 17 | |
| 18 | /* |
| 19 | * The xstateregs_active() routine is the same as the regset_fpregs_active() routine, |
| 20 | * as the "regset->n" for the xstate regset will be updated based on the feature |
| 21 | * capabilities supported by the xsave. |
| 22 | */ |
| 23 | int regset_fpregs_active(struct task_struct *target, const struct user_regset *regset) |
| 24 | { |
| 25 | return regset->n; |
| 26 | } |
| 27 | |
| 28 | int regset_xregset_fpregs_active(struct task_struct *target, const struct user_regset *regset) |
| 29 | { |
| 30 | if (boot_cpu_has(X86_FEATURE_FXSR)) |
| 31 | return regset->n; |
| 32 | else |
| 33 | return 0; |
| 34 | } |
| 35 | |
| 36 | /* |
| 37 | * The regset get() functions are invoked from: |
| 38 | * |
| 39 | * - coredump to dump the current task's fpstate. If the current task |
| 40 | * owns the FPU then the memory state has to be synchronized and the |
| 41 | * FPU register state preserved. Otherwise fpstate is already in sync. |
| 42 | * |
| 43 | * - ptrace to dump fpstate of a stopped task, in which case the registers |
| 44 | * have already been saved to fpstate on context switch. |
| 45 | */ |
| 46 | static void sync_fpstate(struct fpu *fpu) |
| 47 | { |
| 48 | if (fpu == x86_task_fpu(current)) |
| 49 | fpu_sync_fpstate(fpu); |
| 50 | } |
| 51 | |
| 52 | /* |
| 53 | * Invalidate cached FPU registers before modifying the stopped target |
| 54 | * task's fpstate. |
| 55 | * |
| 56 | * This forces the target task on resume to restore the FPU registers from |
| 57 | * modified fpstate. Otherwise the task might skip the restore and operate |
| 58 | * with the cached FPU registers which discards the modifications. |
| 59 | */ |
| 60 | static void fpu_force_restore(struct fpu *fpu) |
| 61 | { |
| 62 | /* |
| 63 | * Only stopped child tasks can be used to modify the FPU |
| 64 | * state in the fpstate buffer: |
| 65 | */ |
| 66 | WARN_ON_FPU(fpu == x86_task_fpu(current)); |
| 67 | |
| 68 | __fpu_invalidate_fpregs_state(fpu); |
| 69 | } |
| 70 | |
| 71 | int xfpregs_get(struct task_struct *target, const struct user_regset *regset, |
| 72 | struct membuf to) |
| 73 | { |
| 74 | struct fpu *fpu = x86_task_fpu(task: target); |
| 75 | |
| 76 | if (!cpu_feature_enabled(X86_FEATURE_FXSR)) |
| 77 | return -ENODEV; |
| 78 | |
| 79 | sync_fpstate(fpu); |
| 80 | |
| 81 | if (!use_xsave()) { |
| 82 | return membuf_write(s: &to, v: &fpu->fpstate->regs.fxsave, |
| 83 | size: sizeof(fpu->fpstate->regs.fxsave)); |
| 84 | } |
| 85 | |
| 86 | copy_xstate_to_uabi_buf(to, tsk: target, mode: XSTATE_COPY_FX); |
| 87 | return 0; |
| 88 | } |
| 89 | |
| 90 | int xfpregs_set(struct task_struct *target, const struct user_regset *regset, |
| 91 | unsigned int pos, unsigned int count, |
| 92 | const void *kbuf, const void __user *ubuf) |
| 93 | { |
| 94 | struct fpu *fpu = x86_task_fpu(task: target); |
| 95 | struct fxregs_state newstate; |
| 96 | int ret; |
| 97 | |
| 98 | if (!cpu_feature_enabled(X86_FEATURE_FXSR)) |
| 99 | return -ENODEV; |
| 100 | |
| 101 | /* No funny business with partial or oversized writes is permitted. */ |
| 102 | if (pos != 0 || count != sizeof(newstate)) |
| 103 | return -EINVAL; |
| 104 | |
| 105 | ret = user_regset_copyin(pos: &pos, count: &count, kbuf: &kbuf, ubuf: &ubuf, data: &newstate, start_pos: 0, end_pos: -1); |
| 106 | if (ret) |
| 107 | return ret; |
| 108 | |
| 109 | /* Do not allow an invalid MXCSR value. */ |
| 110 | if (newstate.mxcsr & ~mxcsr_feature_mask) |
| 111 | return -EINVAL; |
| 112 | |
| 113 | fpu_force_restore(fpu); |
| 114 | |
| 115 | /* Copy the state */ |
| 116 | memcpy(&fpu->fpstate->regs.fxsave, &newstate, sizeof(newstate)); |
| 117 | |
| 118 | /* Clear xmm8..15 for 32-bit callers */ |
| 119 | BUILD_BUG_ON(sizeof(fpu->__fpstate.regs.fxsave.xmm_space) != 16 * 16); |
| 120 | if (in_ia32_syscall()) |
| 121 | memset(&fpu->fpstate->regs.fxsave.xmm_space[8*4], 0, 8 * 16); |
| 122 | |
| 123 | /* Mark FP and SSE as in use when XSAVE is enabled */ |
| 124 | if (use_xsave()) |
| 125 | fpu->fpstate->regs.xsave.header.xfeatures |= XFEATURE_MASK_FPSSE; |
| 126 | |
| 127 | return 0; |
| 128 | } |
| 129 | |
| 130 | int xstateregs_get(struct task_struct *target, const struct user_regset *regset, |
| 131 | struct membuf to) |
| 132 | { |
| 133 | if (!cpu_feature_enabled(X86_FEATURE_XSAVE)) |
| 134 | return -ENODEV; |
| 135 | |
| 136 | sync_fpstate(fpu: x86_task_fpu(task: target)); |
| 137 | |
| 138 | copy_xstate_to_uabi_buf(to, tsk: target, mode: XSTATE_COPY_XSAVE); |
| 139 | return 0; |
| 140 | } |
| 141 | |
| 142 | int xstateregs_set(struct task_struct *target, const struct user_regset *regset, |
| 143 | unsigned int pos, unsigned int count, |
| 144 | const void *kbuf, const void __user *ubuf) |
| 145 | { |
| 146 | struct fpu *fpu = x86_task_fpu(task: target); |
| 147 | struct xregs_state *tmpbuf = NULL; |
| 148 | int ret; |
| 149 | |
| 150 | if (!cpu_feature_enabled(X86_FEATURE_XSAVE)) |
| 151 | return -ENODEV; |
| 152 | |
| 153 | /* |
| 154 | * A whole standard-format XSAVE buffer is needed: |
| 155 | */ |
| 156 | if (pos != 0 || count != fpu_user_cfg.max_size) |
| 157 | return -EFAULT; |
| 158 | |
| 159 | if (!kbuf) { |
| 160 | tmpbuf = vmalloc(count); |
| 161 | if (!tmpbuf) |
| 162 | return -ENOMEM; |
| 163 | |
| 164 | if (copy_from_user(to: tmpbuf, from: ubuf, n: count)) { |
| 165 | ret = -EFAULT; |
| 166 | goto out; |
| 167 | } |
| 168 | } |
| 169 | |
| 170 | fpu_force_restore(fpu); |
| 171 | ret = copy_uabi_from_kernel_to_xstate(fpstate: fpu->fpstate, kbuf: kbuf ?: tmpbuf, pkru: &target->thread.pkru); |
| 172 | |
| 173 | out: |
| 174 | vfree(addr: tmpbuf); |
| 175 | return ret; |
| 176 | } |
| 177 | |
| 178 | #ifdef CONFIG_X86_USER_SHADOW_STACK |
| 179 | int ssp_active(struct task_struct *target, const struct user_regset *regset) |
| 180 | { |
| 181 | if (target->thread.features & ARCH_SHSTK_SHSTK) |
| 182 | return regset->n; |
| 183 | |
| 184 | return 0; |
| 185 | } |
| 186 | |
| 187 | int ssp_get(struct task_struct *target, const struct user_regset *regset, |
| 188 | struct membuf to) |
| 189 | { |
| 190 | struct fpu *fpu = x86_task_fpu(task: target); |
| 191 | struct cet_user_state *cetregs; |
| 192 | |
| 193 | if (!cpu_feature_enabled(X86_FEATURE_USER_SHSTK) || |
| 194 | !ssp_active(target, regset)) |
| 195 | return -ENODEV; |
| 196 | |
| 197 | sync_fpstate(fpu); |
| 198 | cetregs = get_xsave_addr(xsave: &fpu->fpstate->regs.xsave, xfeature_nr: XFEATURE_CET_USER); |
| 199 | if (WARN_ON(!cetregs)) { |
| 200 | /* |
| 201 | * This shouldn't ever be NULL because shadow stack was |
| 202 | * verified to be enabled above. This means |
| 203 | * MSR_IA32_U_CET.CET_SHSTK_EN should be 1 and so |
| 204 | * XFEATURE_CET_USER should not be in the init state. |
| 205 | */ |
| 206 | return -ENODEV; |
| 207 | } |
| 208 | |
| 209 | return membuf_write(s: &to, v: (unsigned long *)&cetregs->user_ssp, |
| 210 | size: sizeof(cetregs->user_ssp)); |
| 211 | } |
| 212 | |
| 213 | int ssp_set(struct task_struct *target, const struct user_regset *regset, |
| 214 | unsigned int pos, unsigned int count, |
| 215 | const void *kbuf, const void __user *ubuf) |
| 216 | { |
| 217 | struct fpu *fpu = x86_task_fpu(task: target); |
| 218 | struct xregs_state *xsave = &fpu->fpstate->regs.xsave; |
| 219 | struct cet_user_state *cetregs; |
| 220 | unsigned long user_ssp; |
| 221 | int r; |
| 222 | |
| 223 | if (!cpu_feature_enabled(X86_FEATURE_USER_SHSTK) || |
| 224 | !ssp_active(target, regset)) |
| 225 | return -ENODEV; |
| 226 | |
| 227 | if (pos != 0 || count != sizeof(user_ssp)) |
| 228 | return -EINVAL; |
| 229 | |
| 230 | r = user_regset_copyin(pos: &pos, count: &count, kbuf: &kbuf, ubuf: &ubuf, data: &user_ssp, start_pos: 0, end_pos: -1); |
| 231 | if (r) |
| 232 | return r; |
| 233 | |
| 234 | /* |
| 235 | * Some kernel instructions (IRET, etc) can cause exceptions in the case |
| 236 | * of disallowed CET register values. Just prevent invalid values. |
| 237 | */ |
| 238 | if (user_ssp >= TASK_SIZE_MAX || !IS_ALIGNED(user_ssp, 8)) |
| 239 | return -EINVAL; |
| 240 | |
| 241 | fpu_force_restore(fpu); |
| 242 | |
| 243 | cetregs = get_xsave_addr(xsave, xfeature_nr: XFEATURE_CET_USER); |
| 244 | if (WARN_ON(!cetregs)) { |
| 245 | /* |
| 246 | * This shouldn't ever be NULL because shadow stack was |
| 247 | * verified to be enabled above. This means |
| 248 | * MSR_IA32_U_CET.CET_SHSTK_EN should be 1 and so |
| 249 | * XFEATURE_CET_USER should not be in the init state. |
| 250 | */ |
| 251 | return -ENODEV; |
| 252 | } |
| 253 | |
| 254 | cetregs->user_ssp = user_ssp; |
| 255 | return 0; |
| 256 | } |
| 257 | #endif /* CONFIG_X86_USER_SHADOW_STACK */ |
| 258 | |
| 259 | #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION |
| 260 | |
| 261 | /* |
| 262 | * FPU tag word conversions. |
| 263 | */ |
| 264 | |
| 265 | static inline unsigned short twd_i387_to_fxsr(unsigned short twd) |
| 266 | { |
| 267 | unsigned int tmp; /* to avoid 16 bit prefixes in the code */ |
| 268 | |
| 269 | /* Transform each pair of bits into 01 (valid) or 00 (empty) */ |
| 270 | tmp = ~twd; |
| 271 | tmp = (tmp | (tmp>>1)) & 0x5555; /* 0V0V0V0V0V0V0V0V */ |
| 272 | /* and move the valid bits to the lower byte. */ |
| 273 | tmp = (tmp | (tmp >> 1)) & 0x3333; /* 00VV00VV00VV00VV */ |
| 274 | tmp = (tmp | (tmp >> 2)) & 0x0f0f; /* 0000VVVV0000VVVV */ |
| 275 | tmp = (tmp | (tmp >> 4)) & 0x00ff; /* 00000000VVVVVVVV */ |
| 276 | |
| 277 | return tmp; |
| 278 | } |
| 279 | |
| 280 | #define FPREG_ADDR(f, n) ((void *)&(f)->st_space + (n) * 16) |
| 281 | #define FP_EXP_TAG_VALID 0 |
| 282 | #define FP_EXP_TAG_ZERO 1 |
| 283 | #define FP_EXP_TAG_SPECIAL 2 |
| 284 | #define FP_EXP_TAG_EMPTY 3 |
| 285 | |
| 286 | static inline u32 twd_fxsr_to_i387(struct fxregs_state *fxsave) |
| 287 | { |
| 288 | struct _fpxreg *st; |
| 289 | u32 tos = (fxsave->swd >> 11) & 7; |
| 290 | u32 twd = (unsigned long) fxsave->twd; |
| 291 | u32 tag; |
| 292 | u32 ret = 0xffff0000u; |
| 293 | int i; |
| 294 | |
| 295 | for (i = 0; i < 8; i++, twd >>= 1) { |
| 296 | if (twd & 0x1) { |
| 297 | st = FPREG_ADDR(fxsave, (i - tos) & 7); |
| 298 | |
| 299 | switch (st->exponent & 0x7fff) { |
| 300 | case 0x7fff: |
| 301 | tag = FP_EXP_TAG_SPECIAL; |
| 302 | break; |
| 303 | case 0x0000: |
| 304 | if (!st->significand[0] && |
| 305 | !st->significand[1] && |
| 306 | !st->significand[2] && |
| 307 | !st->significand[3]) |
| 308 | tag = FP_EXP_TAG_ZERO; |
| 309 | else |
| 310 | tag = FP_EXP_TAG_SPECIAL; |
| 311 | break; |
| 312 | default: |
| 313 | if (st->significand[3] & 0x8000) |
| 314 | tag = FP_EXP_TAG_VALID; |
| 315 | else |
| 316 | tag = FP_EXP_TAG_SPECIAL; |
| 317 | break; |
| 318 | } |
| 319 | } else { |
| 320 | tag = FP_EXP_TAG_EMPTY; |
| 321 | } |
| 322 | ret |= tag << (2 * i); |
| 323 | } |
| 324 | return ret; |
| 325 | } |
| 326 | |
| 327 | /* |
| 328 | * FXSR floating point environment conversions. |
| 329 | */ |
| 330 | |
| 331 | static void __convert_from_fxsr(struct user_i387_ia32_struct *env, |
| 332 | struct task_struct *tsk, |
| 333 | struct fxregs_state *fxsave) |
| 334 | { |
| 335 | struct _fpreg *to = (struct _fpreg *) &env->st_space[0]; |
| 336 | struct _fpxreg *from = (struct _fpxreg *) &fxsave->st_space[0]; |
| 337 | int i; |
| 338 | |
| 339 | env->cwd = fxsave->cwd | 0xffff0000u; |
| 340 | env->swd = fxsave->swd | 0xffff0000u; |
| 341 | env->twd = twd_fxsr_to_i387(fxsave); |
| 342 | |
| 343 | #ifdef CONFIG_X86_64 |
| 344 | env->fip = fxsave->rip; |
| 345 | env->foo = fxsave->rdp; |
| 346 | /* |
| 347 | * should be actually ds/cs at fpu exception time, but |
| 348 | * that information is not available in 64bit mode. |
| 349 | */ |
| 350 | env->fcs = task_pt_regs(tsk)->cs; |
| 351 | if (tsk == current) { |
| 352 | savesegment(ds, env->fos); |
| 353 | } else { |
| 354 | env->fos = tsk->thread.ds; |
| 355 | } |
| 356 | env->fos |= 0xffff0000; |
| 357 | #else |
| 358 | env->fip = fxsave->fip; |
| 359 | env->fcs = (u16) fxsave->fcs | ((u32) fxsave->fop << 16); |
| 360 | env->foo = fxsave->foo; |
| 361 | env->fos = fxsave->fos; |
| 362 | #endif |
| 363 | |
| 364 | for (i = 0; i < 8; ++i) |
| 365 | memcpy(&to[i], &from[i], sizeof(to[0])); |
| 366 | } |
| 367 | |
| 368 | void |
| 369 | convert_from_fxsr(struct user_i387_ia32_struct *env, struct task_struct *tsk) |
| 370 | { |
| 371 | __convert_from_fxsr(env, tsk, fxsave: &x86_task_fpu(task: tsk)->fpstate->regs.fxsave); |
| 372 | } |
| 373 | |
| 374 | void convert_to_fxsr(struct fxregs_state *fxsave, |
| 375 | const struct user_i387_ia32_struct *env) |
| 376 | |
| 377 | { |
| 378 | struct _fpreg *from = (struct _fpreg *) &env->st_space[0]; |
| 379 | struct _fpxreg *to = (struct _fpxreg *) &fxsave->st_space[0]; |
| 380 | int i; |
| 381 | |
| 382 | fxsave->cwd = env->cwd; |
| 383 | fxsave->swd = env->swd; |
| 384 | fxsave->twd = twd_i387_to_fxsr(twd: env->twd); |
| 385 | fxsave->fop = (u16) ((u32) env->fcs >> 16); |
| 386 | #ifdef CONFIG_X86_64 |
| 387 | fxsave->rip = env->fip; |
| 388 | fxsave->rdp = env->foo; |
| 389 | /* cs and ds ignored */ |
| 390 | #else |
| 391 | fxsave->fip = env->fip; |
| 392 | fxsave->fcs = (env->fcs & 0xffff); |
| 393 | fxsave->foo = env->foo; |
| 394 | fxsave->fos = env->fos; |
| 395 | #endif |
| 396 | |
| 397 | for (i = 0; i < 8; ++i) |
| 398 | memcpy(&to[i], &from[i], sizeof(from[0])); |
| 399 | } |
| 400 | |
| 401 | int fpregs_get(struct task_struct *target, const struct user_regset *regset, |
| 402 | struct membuf to) |
| 403 | { |
| 404 | struct fpu *fpu = x86_task_fpu(task: target); |
| 405 | struct user_i387_ia32_struct env; |
| 406 | struct fxregs_state fxsave, *fx; |
| 407 | |
| 408 | sync_fpstate(fpu); |
| 409 | |
| 410 | if (!cpu_feature_enabled(X86_FEATURE_FPU)) |
| 411 | return fpregs_soft_get(target, regset, to); |
| 412 | |
| 413 | if (!cpu_feature_enabled(X86_FEATURE_FXSR)) { |
| 414 | return membuf_write(s: &to, v: &fpu->fpstate->regs.fsave, |
| 415 | size: sizeof(struct fregs_state)); |
| 416 | } |
| 417 | |
| 418 | if (use_xsave()) { |
| 419 | struct membuf mb = { .p = &fxsave, .left = sizeof(fxsave) }; |
| 420 | |
| 421 | /* Handle init state optimized xstate correctly */ |
| 422 | copy_xstate_to_uabi_buf(to: mb, tsk: target, mode: XSTATE_COPY_FP); |
| 423 | fx = &fxsave; |
| 424 | } else { |
| 425 | fx = &fpu->fpstate->regs.fxsave; |
| 426 | } |
| 427 | |
| 428 | __convert_from_fxsr(env: &env, tsk: target, fxsave: fx); |
| 429 | return membuf_write(s: &to, v: &env, size: sizeof(env)); |
| 430 | } |
| 431 | |
| 432 | int fpregs_set(struct task_struct *target, const struct user_regset *regset, |
| 433 | unsigned int pos, unsigned int count, |
| 434 | const void *kbuf, const void __user *ubuf) |
| 435 | { |
| 436 | struct fpu *fpu = x86_task_fpu(task: target); |
| 437 | struct user_i387_ia32_struct env; |
| 438 | int ret; |
| 439 | |
| 440 | /* No funny business with partial or oversized writes is permitted. */ |
| 441 | if (pos != 0 || count != sizeof(struct user_i387_ia32_struct)) |
| 442 | return -EINVAL; |
| 443 | |
| 444 | if (!cpu_feature_enabled(X86_FEATURE_FPU)) |
| 445 | return fpregs_soft_set(target, regset, pos, count, kbuf, ubuf); |
| 446 | |
| 447 | ret = user_regset_copyin(pos: &pos, count: &count, kbuf: &kbuf, ubuf: &ubuf, data: &env, start_pos: 0, end_pos: -1); |
| 448 | if (ret) |
| 449 | return ret; |
| 450 | |
| 451 | fpu_force_restore(fpu); |
| 452 | |
| 453 | if (cpu_feature_enabled(X86_FEATURE_FXSR)) |
| 454 | convert_to_fxsr(fxsave: &fpu->fpstate->regs.fxsave, env: &env); |
| 455 | else |
| 456 | memcpy(&fpu->fpstate->regs.fsave, &env, sizeof(env)); |
| 457 | |
| 458 | /* |
| 459 | * Update the header bit in the xsave header, indicating the |
| 460 | * presence of FP. |
| 461 | */ |
| 462 | if (cpu_feature_enabled(X86_FEATURE_XSAVE)) |
| 463 | fpu->fpstate->regs.xsave.header.xfeatures |= XFEATURE_MASK_FP; |
| 464 | |
| 465 | return 0; |
| 466 | } |
| 467 | |
| 468 | #endif /* CONFIG_X86_32 || CONFIG_IA32_EMULATION */ |
| 469 | |