arch/x86/kernel/xsave.c

   1 /*
   2  * xsave/xrstor support.
   3  *
   4  * Author: Suresh Siddha <suresh.b.siddha@intel.com>
   5  */
   6
   7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
   8
   9 #include <linux/bootmem.h>
  10 #include <linux/compat.h>
  11 #include <asm/i387.h>
  12 #include <asm/fpu-internal.h>
  13 #ifdef CONFIG_IA32_EMULATION
  14 #include <asm/sigcontext32.h>
  15 #endif
  16 #include <asm/xcr.h>
  17
  18 /*
  19  * Supported feature mask by the CPU and the kernel.
  20  */
  21 u64 pcntxt_mask;
  22
  23 /*
  24  * Represents init state for the supported extended state.
  25  */
  26 static struct xsave_struct *init_xstate_buf;
  27
  28 struct _fpx_sw_bytes fx_sw_reserved;
  29 #ifdef CONFIG_IA32_EMULATION
  30 struct _fpx_sw_bytes fx_sw_reserved_ia32;
  31 #endif
  32
  33 static unsigned int *xstate_offsets, *xstate_sizes, xstate_features;
  34
  35 /*
  36  * If a processor implementation discern that a processor state component is
  37  * in its initialized state it may modify the corresponding bit in the
  38  * xsave_hdr.xstate_bv as '0', with out modifying the corresponding memory
  39  * layout in the case of xsaveopt. While presenting the xstate information to
  40  * the user, we always ensure that the memory layout of a feature will be in
  41  * the init state if the corresponding header bit is zero. This is to ensure
  42  * that the user doesn't see some stale state in the memory layout during
  43  * signal handling, debugging etc.
  44  */
  45 void __sanitize_i387_state(struct task_struct *tsk)
  46 {
  47         u64 xstate_bv;
  48         int feature_bit = 0x2;
  49         struct i387_fxsave_struct *fx = &tsk->thread.fpu.state->fxsave;
  50
  51         if (!fx)
  52                 return;
  53
  54         xstate_bv = tsk->thread.fpu.state->xsave.xsave_hdr.xstate_bv;
  55
  56         /*
  57          * None of the feature bits are in init state. So nothing else
  58          * to do for us, as the memory layout is up to date.
  59          */
  60         if ((xstate_bv & pcntxt_mask) == pcntxt_mask)
  61                 return;
  62
  63         /*
  64          * FP is in init state
  65          */
  66         if (!(xstate_bv & XSTATE_FP)) {
  67                 fx->cwd = 0x37f;
  68                 fx->swd = 0;
  69                 fx->twd = 0;
  70                 fx->fop = 0;
  71                 fx->rip = 0;
  72                 fx->rdp = 0;
  73                 memset(&fx->st_space[0], 0, 128);
  74         }
  75
  76         /*
  77          * SSE is in init state
  78          */
  79         if (!(xstate_bv & XSTATE_SSE))
  80                 memset(&fx->xmm_space[0], 0, 256);
  81
  82         xstate_bv = (pcntxt_mask & ~xstate_bv) >> 2;
  83
  84         /*
  85          * Update all the other memory layouts for which the corresponding
  86          * header bit is in the init state.
  87          */
  88         while (xstate_bv) {
  89                 if (xstate_bv & 0x1) {
  90                         int offset = xstate_offsets[feature_bit];
  91                         int size = xstate_sizes[feature_bit];
  92
  93                         memcpy(((void *) fx) + offset,
  94                                ((void *) init_xstate_buf) + offset,
  95                                size);
  96                 }
  97
  98                 xstate_bv >>= 1;
  99                 feature_bit++;
 100         }
 101 }
 102
 103 /*
 104  * Check for the presence of extended state information in the
 105  * user fpstate pointer in the sigcontext.
 106  */
 107 int check_for_xstate(struct i387_fxsave_struct __user *buf,
 108                      void __user *fpstate,
 109                      struct _fpx_sw_bytes *fx_sw_user)
 110 {
 111         int min_xstate_size = sizeof(struct i387_fxsave_struct) +
 112                               sizeof(struct xsave_hdr_struct);
 113         unsigned int magic2;
 114         int err;
 115
 116         err = __copy_from_user(fx_sw_user, &buf->sw_reserved[0],
 117                                sizeof(struct _fpx_sw_bytes));
 118         if (err)
 119                 return -EFAULT;
 120
 121         /*
 122          * First Magic check failed.
 123          */
 124         if (fx_sw_user->magic1 != FP_XSTATE_MAGIC1)
 125                 return -EINVAL;
 126
 127         /*
 128          * Check for error scenarios.
 129          */
 130         if (fx_sw_user->xstate_size < min_xstate_size ||
 131             fx_sw_user->xstate_size > xstate_size ||
 132             fx_sw_user->xstate_size > fx_sw_user->extended_size)
 133                 return -EINVAL;
 134
 135         err = __get_user(magic2, (__u32 *) (((void *)fpstate) +
 136                                             fx_sw_user->extended_size -
 137                                             FP_XSTATE_MAGIC2_SIZE));
 138         if (err)
 139                 return err;
 140         /*
 141          * Check for the presence of second magic word at the end of memory
 142          * layout. This detects the case where the user just copied the legacy
 143          * fpstate layout with out copying the extended state information
 144          * in the memory layout.
 145          */
 146         if (magic2 != FP_XSTATE_MAGIC2)
 147                 return -EFAULT;
 148
 149         return 0;
 150 }
 151
 152 #ifdef CONFIG_X86_64
 153 /*
 154  * Signal frame handlers.
 155  */
 156
 157 int save_i387_xstate(void __user *buf)
 158 {
 159         struct task_struct *tsk = current;
 160         int err = 0;
 161
 162         if (!access_ok(VERIFY_WRITE, buf, sig_xstate_size))
 163                 return -EACCES;
 164
 165         BUG_ON(sig_xstate_size < xstate_size);
 166
 167         if ((unsigned long)buf % 64)
 168                 pr_err("%s: bad fpstate %p\n", __func__, buf);
 169
 170         if (!used_math())
 171                 return 0;
 172
 173         if (user_has_fpu()) {
 174                 if (use_xsave())
 175                         err = xsave_user(buf);
 176                 else
 177                         err = fxsave_user(buf);
 178
 179                 if (err)
 180                         return err;
 181                 user_fpu_end();
 182         } else {
 183                 sanitize_i387_state(tsk);
 184                 if (__copy_to_user(buf, &tsk->thread.fpu.state->fxsave,
 185                                    xstate_size))
 186                         return -1;
 187         }
 188
 189         clear_used_math(); /* trigger finit */
 190
 191         if (use_xsave()) {
 192                 struct _fpstate __user *fx = buf;
 193                 struct _xstate __user *x = buf;
 194                 u64 xstate_bv;
 195
 196                 err = __copy_to_user(&fx->sw_reserved, &fx_sw_reserved,
 197                                      sizeof(struct _fpx_sw_bytes));
 198
 199                 err |= __put_user(FP_XSTATE_MAGIC2,
 200                                   (__u32 __user *) (buf + sig_xstate_size
 201                                                     - FP_XSTATE_MAGIC2_SIZE));
 202
 203                 /*
 204                  * Read the xstate_bv which we copied (directly from the cpu or
 205                  * from the state in task struct) to the user buffers and
 206                  * set the FP/SSE bits.
 207                  */
 208                 err |= __get_user(xstate_bv, &x->xstate_hdr.xstate_bv);
 209
 210                 /*
 211                  * For legacy compatible, we always set FP/SSE bits in the bit
 212                  * vector while saving the state to the user context. This will
 213                  * enable us capturing any changes(during sigreturn) to
 214                  * the FP/SSE bits by the legacy applications which don't touch
 215                  * xstate_bv in the xsave header.
 216                  *
 217                  * xsave aware apps can change the xstate_bv in the xsave
 218                  * header as well as change any contents in the memory layout.
 219                  * xrestore as part of sigreturn will capture all the changes.
 220                  */
 221                 xstate_bv |= XSTATE_FPSSE;
 222
 223                 err |= __put_user(xstate_bv, &x->xstate_hdr.xstate_bv);
 224
 225                 if (err)
 226                         return err;
 227         }
 228
 229         return 1;
 230 }
 231
 232 /*
 233  * Restore the extended state if present. Otherwise, restore the FP/SSE
 234  * state.
 235  */
 236 static int restore_user_xstate(void __user *buf)
 237 {
 238         struct _fpx_sw_bytes fx_sw_user;
 239         u64 mask;
 240         int err;
 241
 242         if (((unsigned long)buf % 64) ||
 243              check_for_xstate(buf, buf, &fx_sw_user))
 244                 goto fx_only;
 245
 246         mask = fx_sw_user.xstate_bv;
 247
 248         /*
 249          * restore the state passed by the user.
 250          */
 251         err = xrestore_user(buf, mask);
 252         if (err)
 253                 return err;
 254
 255         /*
 256          * init the state skipped by the user.
 257          */
 258         mask = pcntxt_mask & ~mask;
 259         if (unlikely(mask))
 260                 xrstor_state(init_xstate_buf, mask);
 261
 262         return 0;
 263
 264 fx_only:
 265         /*
 266          * couldn't find the extended state information in the
 267          * memory layout. Restore just the FP/SSE and init all
 268          * the other extended state.
 269          */
 270         xrstor_state(init_xstate_buf, pcntxt_mask & ~XSTATE_FPSSE);
 271         return fxrstor_checking((__force struct i387_fxsave_struct *)buf);
 272 }
 273
 274 /*
 275  * This restores directly out of user space. Exceptions are handled.
 276  */
 277 int restore_i387_xstate(void __user *buf)
 278 {
 279         struct task_struct *tsk = current;
 280         int err = 0;
 281
 282         if (!buf) {
 283                 if (used_math())
 284                         goto clear;
 285                 return 0;
 286         } else
 287                 if (!access_ok(VERIFY_READ, buf, sig_xstate_size))
 288                         return -EACCES;
 289
 290         if (!used_math()) {
 291                 err = init_fpu(tsk);
 292                 if (err)
 293                         return err;
 294         }
 295
 296         user_fpu_begin();
 297         if (use_xsave())
 298                 err = restore_user_xstate(buf);
 299         else
 300                 err = fxrstor_checking((__force struct i387_fxsave_struct *)
 301                                        buf);
 302         if (unlikely(err)) {
 303                 /*
 304                  * Encountered an error while doing the restore from the
 305                  * user buffer, clear the fpu state.
 306                  */
 307 clear:
 308                 clear_fpu(tsk);
 309                 clear_used_math();
 310         }
 311         return err;
 312 }
 313 #endif
 314
 315 /*
 316  * Prepare the SW reserved portion of the fxsave memory layout, indicating
 317  * the presence of the extended state information in the memory layout
 318  * pointed by the fpstate pointer in the sigcontext.
 319  * This will be saved when ever the FP and extended state context is
 320  * saved on the user stack during the signal handler delivery to the user.
 321  */
 322 static void prepare_fx_sw_frame(void)
 323 {
 324         int size_extended = (xstate_size - sizeof(struct i387_fxsave_struct)) +
 325                              FP_XSTATE_MAGIC2_SIZE;
 326
 327         sig_xstate_size = sizeof(struct _fpstate) + size_extended;
 328
 329 #ifdef CONFIG_IA32_EMULATION
 330         sig_xstate_ia32_size = sizeof(struct _fpstate_ia32) + size_extended;
 331 #endif
 332
 333         memset(&fx_sw_reserved, 0, sizeof(fx_sw_reserved));
 334
 335         fx_sw_reserved.magic1 = FP_XSTATE_MAGIC1;
 336         fx_sw_reserved.extended_size = sig_xstate_size;
 337         fx_sw_reserved.xstate_bv = pcntxt_mask;
 338         fx_sw_reserved.xstate_size = xstate_size;
 339 #ifdef CONFIG_IA32_EMULATION
 340         memcpy(&fx_sw_reserved_ia32, &fx_sw_reserved,
 341                sizeof(struct _fpx_sw_bytes));
 342         fx_sw_reserved_ia32.extended_size = sig_xstate_ia32_size;
 343 #endif
 344 }
 345
 346 #ifdef CONFIG_X86_64
 347 unsigned int sig_xstate_size = sizeof(struct _fpstate);
 348 #endif
 349
 350 /*
 351  * Enable the extended processor state save/restore feature
 352  */
 353 static inline void xstate_enable(void)
 354 {
 355         set_in_cr4(X86_CR4_OSXSAVE);
 356         xsetbv(XCR_XFEATURE_ENABLED_MASK, pcntxt_mask);
 357 }
 358
 359 /*
 360  * Record the offsets and sizes of different state managed by the xsave
 361  * memory layout.
 362  */
 363 static void __init setup_xstate_features(void)
 364 {
 365         int eax, ebx, ecx, edx, leaf = 0x2;
 366
 367         xstate_features = fls64(pcntxt_mask);
 368         xstate_offsets = alloc_bootmem(xstate_features * sizeof(int));
 369         xstate_sizes = alloc_bootmem(xstate_features * sizeof(int));
 370
 371         do {
 372                 cpuid_count(XSTATE_CPUID, leaf, &eax, &ebx, &ecx, &edx);
 373
 374                 if (eax == 0)
 375                         break;
 376
 377                 xstate_offsets[leaf] = ebx;
 378                 xstate_sizes[leaf] = eax;
 379
 380                 leaf++;
 381         } while (1);
 382 }
 383
 384 /*
 385  * setup the xstate image representing the init state
 386  */
 387 static void __init setup_xstate_init(void)
 388 {
 389         setup_xstate_features();
 390
 391         /*
 392          * Setup init_xstate_buf to represent the init state of
 393          * all the features managed by the xsave
 394          */
 395         init_xstate_buf = alloc_bootmem_align(xstate_size,
 396                                               __alignof__(struct xsave_struct));
 397         init_xstate_buf->i387.mxcsr = MXCSR_DEFAULT;
 398
 399         clts();
 400         /*
 401          * Init all the features state with header_bv being 0x0
 402          */
 403         xrstor_state(init_xstate_buf, -1);
 404         /*
 405          * Dump the init state again. This is to identify the init state
 406          * of any feature which is not represented by all zero's.
 407          */
 408         xsave_state(init_xstate_buf, -1);
 409         stts();
 410 }
 411
 412 /*
 413  * Enable and initialize the xsave feature.
 414  */
 415 static void __init xstate_enable_boot_cpu(void)
 416 {
 417         unsigned int eax, ebx, ecx, edx;
 418
 419         if (boot_cpu_data.cpuid_level < XSTATE_CPUID) {
 420                 WARN(1, KERN_ERR "XSTATE_CPUID missing\n");
 421                 return;
 422         }
 423
 424         cpuid_count(XSTATE_CPUID, 0, &eax, &ebx, &ecx, &edx);
 425         pcntxt_mask = eax + ((u64)edx << 32);
 426
 427         if ((pcntxt_mask & XSTATE_FPSSE) != XSTATE_FPSSE) {
 428                 pr_err("FP/SSE not shown under xsave features 0x%llx\n",
 429                        pcntxt_mask);
 430                 BUG();
 431         }
 432
 433         /*
 434          * Support only the state known to OS.
 435          */
 436         pcntxt_mask = pcntxt_mask & XCNTXT_MASK;
 437
 438         xstate_enable();
 439
 440         /*
 441          * Recompute the context size for enabled features
 442          */
 443         cpuid_count(XSTATE_CPUID, 0, &eax, &ebx, &ecx, &edx);
 444         xstate_size = ebx;
 445
 446         update_regset_xstate_info(xstate_size, pcntxt_mask);
 447         prepare_fx_sw_frame();
 448
 449         setup_xstate_init();
 450
 451         pr_info("enabled xstate_bv 0x%llx, cntxt size 0x%x\n",
 452                 pcntxt_mask, xstate_size);
 453 }
 454
 455 /*
 456  * For the very first instance, this calls xstate_enable_boot_cpu();
 457  * for all subsequent instances, this calls xstate_enable().
 458  *
 459  * This is somewhat obfuscated due to the lack of powerful enough
 460  * overrides for the section checks.
 461  */
 462 void __cpuinit xsave_init(void)
 463 {
 464         static __refdata void (*next_func)(void) = xstate_enable_boot_cpu;
 465         void (*this_func)(void);
 466
 467         if (!cpu_has_xsave)
 468                 return;
 469
 470         this_func = next_func;
 471         next_func = xstate_enable;
 472         this_func();
 473 }