arch/x86/kernel/xsave.c

   1 /*
   2  * xsave/xrstor support.
   3  *
   4  * Author: Suresh Siddha <suresh.b.siddha@intel.com>
   5  */
   6 #include <linux/bootmem.h>
   7 #include <linux/compat.h>
   8 #include <asm/i387.h>
   9 #ifdef CONFIG_IA32_EMULATION
  10 #include <asm/sigcontext32.h>
  11 #endif
  12 #include <asm/xcr.h>
  13
  14 /*
  15  * Supported feature mask by the CPU and the kernel.
  16  */
  17 u64 pcntxt_mask;
  18
  19 struct _fpx_sw_bytes fx_sw_reserved;
  20 #ifdef CONFIG_IA32_EMULATION
  21 struct _fpx_sw_bytes fx_sw_reserved_ia32;
  22 #endif
  23
  24 static unsigned int *xstate_offsets, *xstate_sizes, xstate_features;
  25
  26 /*
  27  * If a processor implementation discern that a processor state component is
  28  * in its initialized state it may modify the corresponding bit in the
  29  * xsave_hdr.xstate_bv as '0', with out modifying the corresponding memory
  30  * layout in the case of xsaveopt. While presenting the xstate information to
  31  * the user, we always ensure that the memory layout of a feature will be in
  32  * the init state if the corresponding header bit is zero. This is to ensure
  33  * that the user doesn't see some stale state in the memory layout during
  34  * signal handling, debugging etc.
  35  */
  36 void __sanitize_i387_state(struct task_struct *tsk)
  37 {
  38         u64 xstate_bv;
  39         int feature_bit = 0x2;
  40         struct i387_fxsave_struct *fx = &tsk->thread.fpu.state->fxsave;
  41
  42         if (!fx)
  43                 return;
  44
  45         BUG_ON(task_thread_info(tsk)->status & TS_USEDFPU);
  46
  47         xstate_bv = tsk->thread.fpu.state->xsave.xsave_hdr.xstate_bv;
  48
  49         /*
  50          * None of the feature bits are in init state. So nothing else
  51          * to do for us, as the memory layout is upto date.
  52          */
  53         if ((xstate_bv & pcntxt_mask) == pcntxt_mask)
  54                 return;
  55
  56         /*
  57          * FP is in init state
  58          */
  59         if (!(xstate_bv & XSTATE_FP)) {
  60                 fx->cwd = 0x37f;
  61                 fx->swd = 0;
  62                 fx->twd = 0;
  63                 fx->fop = 0;
  64                 fx->rip = 0;
  65                 fx->rdp = 0;
  66                 memset(&fx->st_space[0], 0, 128);
  67         }
  68
  69         /*
  70          * SSE is in init state
  71          */
  72         if (!(xstate_bv & XSTATE_SSE))
  73                 memset(&fx->xmm_space[0], 0, 256);
  74
  75         xstate_bv = (pcntxt_mask & ~xstate_bv) >> 2;
  76
  77         /*
  78          * Update all the other memory layouts for which the corresponding
  79          * header bit is in the init state.
  80          */
  81         while (xstate_bv) {
  82                 if (xstate_bv & 0x1) {
  83                         int offset = xstate_offsets[feature_bit];
  84                         int size = xstate_sizes[feature_bit];
  85
  86                         memcpy(((void *) fx) + offset,
  87                                ((void *) init_xstate_buf) + offset,
  88                                size);
  89                 }
  90
  91                 xstate_bv >>= 1;
  92                 feature_bit++;
  93         }
  94 }
  95
  96 /*
  97  * Check for the presence of extended state information in the
  98  * user fpstate pointer in the sigcontext.
  99  */
 100 int check_for_xstate(struct i387_fxsave_struct __user *buf,
 101                      void __user *fpstate,
 102                      struct _fpx_sw_bytes *fx_sw_user)
 103 {
 104         int min_xstate_size = sizeof(struct i387_fxsave_struct) +
 105                               sizeof(struct xsave_hdr_struct);
 106         unsigned int magic2;
 107         int err;
 108
 109         err = __copy_from_user(fx_sw_user, &buf->sw_reserved[0],
 110                                sizeof(struct _fpx_sw_bytes));
 111         if (err)
 112                 return -EFAULT;
 113
 114         /*
 115          * First Magic check failed.
 116          */
 117         if (fx_sw_user->magic1 != FP_XSTATE_MAGIC1)
 118                 return -EINVAL;
 119
 120         /*
 121          * Check for error scenarios.
 122          */
 123         if (fx_sw_user->xstate_size < min_xstate_size ||
 124             fx_sw_user->xstate_size > xstate_size ||
 125             fx_sw_user->xstate_size > fx_sw_user->extended_size)
 126                 return -EINVAL;
 127
 128         err = __get_user(magic2, (__u32 *) (((void *)fpstate) +
 129                                             fx_sw_user->extended_size -
 130                                             FP_XSTATE_MAGIC2_SIZE));
 131         if (err)
 132                 return err;
 133         /*
 134          * Check for the presence of second magic word at the end of memory
 135          * layout. This detects the case where the user just copied the legacy
 136          * fpstate layout with out copying the extended state information
 137          * in the memory layout.
 138          */
 139         if (magic2 != FP_XSTATE_MAGIC2)
 140                 return -EFAULT;
 141
 142         return 0;
 143 }
 144
 145 #ifdef CONFIG_X86_64
 146 /*
 147  * Signal frame handlers.
 148  */
 149
 150 int save_i387_xstate(void __user *buf)
 151 {
 152         struct task_struct *tsk = current;
 153         int err = 0;
 154
 155         if (!access_ok(VERIFY_WRITE, buf, sig_xstate_size))
 156                 return -EACCES;
 157
 158         BUG_ON(sig_xstate_size < xstate_size);
 159
 160         if ((unsigned long)buf % 64)
 161                 printk("save_i387_xstate: bad fpstate %p\n", buf);
 162
 163         if (!used_math())
 164                 return 0;
 165
 166         if (task_thread_info(tsk)->status & TS_USEDFPU) {
 167                 /*
 168                  * Start with clearing the user buffer. This will present a
 169                  * clean context for the bytes not touched by the fxsave/xsave.
 170                  */
 171                 err = __clear_user(buf, sig_xstate_size);
 172                 if (err)
 173                         return err;
 174
 175                 if (use_xsave())
 176                         err = xsave_user(buf);
 177                 else
 178                         err = fxsave_user(buf);
 179
 180                 if (err)
 181                         return err;
 182                 task_thread_info(tsk)->status &= ~TS_USEDFPU;
 183                 stts();
 184         } else {
 185                 sanitize_i387_state(tsk);
 186                 if (__copy_to_user(buf, &tsk->thread.fpu.state->fxsave,
 187                                    xstate_size))
 188                         return -1;
 189         }
 190
 191         clear_used_math(); /* trigger finit */
 192
 193         if (use_xsave()) {
 194                 struct _fpstate __user *fx = buf;
 195                 struct _xstate __user *x = buf;
 196                 u64 xstate_bv;
 197
 198                 err = __copy_to_user(&fx->sw_reserved, &fx_sw_reserved,
 199                                      sizeof(struct _fpx_sw_bytes));
 200
 201                 err |= __put_user(FP_XSTATE_MAGIC2,
 202                                   (__u32 __user *) (buf + sig_xstate_size
 203                                                     - FP_XSTATE_MAGIC2_SIZE));
 204
 205                 /*
 206                  * Read the xstate_bv which we copied (directly from the cpu or
 207                  * from the state in task struct) to the user buffers and
 208                  * set the FP/SSE bits.
 209                  */
 210                 err |= __get_user(xstate_bv, &x->xstate_hdr.xstate_bv);
 211
 212                 /*
 213                  * For legacy compatible, we always set FP/SSE bits in the bit
 214                  * vector while saving the state to the user context. This will
 215                  * enable us capturing any changes(during sigreturn) to
 216                  * the FP/SSE bits by the legacy applications which don't touch
 217                  * xstate_bv in the xsave header.
 218                  *
 219                  * xsave aware apps can change the xstate_bv in the xsave
 220                  * header as well as change any contents in the memory layout.
 221                  * xrestore as part of sigreturn will capture all the changes.
 222                  */
 223                 xstate_bv |= XSTATE_FPSSE;
 224
 225                 err |= __put_user(xstate_bv, &x->xstate_hdr.xstate_bv);
 226
 227                 if (err)
 228                         return err;
 229         }
 230
 231         return 1;
 232 }
 233
 234 /*
 235  * Restore the extended state if present. Otherwise, restore the FP/SSE
 236  * state.
 237  */
 238 static int restore_user_xstate(void __user *buf)
 239 {
 240         struct _fpx_sw_bytes fx_sw_user;
 241         u64 mask;
 242         int err;
 243
 244         if (((unsigned long)buf % 64) ||
 245              check_for_xstate(buf, buf, &fx_sw_user))
 246                 goto fx_only;
 247
 248         mask = fx_sw_user.xstate_bv;
 249
 250         /*
 251          * restore the state passed by the user.
 252          */
 253         err = xrestore_user(buf, mask);
 254         if (err)
 255                 return err;
 256
 257         /*
 258          * init the state skipped by the user.
 259          */
 260         mask = pcntxt_mask & ~mask;
 261
 262         xrstor_state(init_xstate_buf, mask);
 263
 264         return 0;
 265
 266 fx_only:
 267         /*
 268          * couldn't find the extended state information in the
 269          * memory layout. Restore just the FP/SSE and init all
 270          * the other extended state.
 271          */
 272         xrstor_state(init_xstate_buf, pcntxt_mask & ~XSTATE_FPSSE);
 273         return fxrstor_checking((__force struct i387_fxsave_struct *)buf);
 274 }
 275
 276 /*
 277  * This restores directly out of user space. Exceptions are handled.
 278  */
 279 int restore_i387_xstate(void __user *buf)
 280 {
 281         struct task_struct *tsk = current;
 282         int err = 0;
 283
 284         if (!buf) {
 285                 if (used_math())
 286                         goto clear;
 287                 return 0;
 288         } else
 289                 if (!access_ok(VERIFY_READ, buf, sig_xstate_size))
 290                         return -EACCES;
 291
 292         if (!used_math()) {
 293                 err = init_fpu(tsk);
 294                 if (err)
 295                         return err;
 296         }
 297
 298         if (!(task_thread_info(current)->status & TS_USEDFPU)) {
 299                 clts();
 300                 task_thread_info(current)->status |= TS_USEDFPU;
 301         }
 302         if (use_xsave())
 303                 err = restore_user_xstate(buf);
 304         else
 305                 err = fxrstor_checking((__force struct i387_fxsave_struct *)
 306                                        buf);
 307         if (unlikely(err)) {
 308                 /*
 309                  * Encountered an error while doing the restore from the
 310                  * user buffer, clear the fpu state.
 311                  */
 312 clear:
 313                 clear_fpu(tsk);
 314                 clear_used_math();
 315         }
 316         return err;
 317 }
 318 #endif
 319
 320 /*
 321  * Prepare the SW reserved portion of the fxsave memory layout, indicating
 322  * the presence of the extended state information in the memory layout
 323  * pointed by the fpstate pointer in the sigcontext.
 324  * This will be saved when ever the FP and extended state context is
 325  * saved on the user stack during the signal handler delivery to the user.
 326  */
 327 static void prepare_fx_sw_frame(void)
 328 {
 329         int size_extended = (xstate_size - sizeof(struct i387_fxsave_struct)) +
 330                              FP_XSTATE_MAGIC2_SIZE;
 331
 332         sig_xstate_size = sizeof(struct _fpstate) + size_extended;
 333
 334 #ifdef CONFIG_IA32_EMULATION
 335         sig_xstate_ia32_size = sizeof(struct _fpstate_ia32) + size_extended;
 336 #endif
 337
 338         memset(&fx_sw_reserved, 0, sizeof(fx_sw_reserved));
 339
 340         fx_sw_reserved.magic1 = FP_XSTATE_MAGIC1;
 341         fx_sw_reserved.extended_size = sig_xstate_size;
 342         fx_sw_reserved.xstate_bv = pcntxt_mask;
 343         fx_sw_reserved.xstate_size = xstate_size;
 344 #ifdef CONFIG_IA32_EMULATION
 345         memcpy(&fx_sw_reserved_ia32, &fx_sw_reserved,
 346                sizeof(struct _fpx_sw_bytes));
 347         fx_sw_reserved_ia32.extended_size = sig_xstate_ia32_size;
 348 #endif
 349 }
 350
 351 /*
 352  * Represents init state for the supported extended state.
 353  */
 354 struct xsave_struct *init_xstate_buf;
 355
 356 #ifdef CONFIG_X86_64
 357 unsigned int sig_xstate_size = sizeof(struct _fpstate);
 358 #endif
 359
 360 /*
 361  * Enable the extended processor state save/restore feature
 362  */
 363 static void __cpuinit __xsave_init(void)
 364 {
 365         set_in_cr4(X86_CR4_OSXSAVE);
 366
 367         /*
 368          * Enable all the features that the HW is capable of
 369          * and the Linux kernel is aware of.
 370          */
 371         xsetbv(XCR_XFEATURE_ENABLED_MASK, pcntxt_mask);
 372 }
 373
 374 /*
 375  * Record the offsets and sizes of different state managed by the xsave
 376  * memory layout.
 377  */
 378 static void setup_xstate_features(void)
 379 {
 380         int eax, ebx, ecx, edx, leaf = 0x2;
 381
 382         xstate_features = fls64(pcntxt_mask);
 383         xstate_offsets = alloc_bootmem(xstate_features * sizeof(int));
 384         xstate_sizes = alloc_bootmem(xstate_features * sizeof(int));
 385
 386         do {
 387                 cpuid_count(0xd, leaf, &eax, &ebx, &ecx, &edx);
 388
 389                 if (eax == 0)
 390                         break;
 391
 392                 xstate_offsets[leaf] = ebx;
 393                 xstate_sizes[leaf] = eax;
 394
 395                 leaf++;
 396         } while (1);
 397 }
 398
 399 /*
 400  * setup the xstate image representing the init state
 401  */
 402 static void __init setup_xstate_init(void)
 403 {
 404         setup_xstate_features();
 405
 406         /*
 407          * Setup init_xstate_buf to represent the init state of
 408          * all the features managed by the xsave
 409          */
 410         init_xstate_buf = alloc_bootmem(xstate_size);
 411         init_xstate_buf->i387.mxcsr = MXCSR_DEFAULT;
 412
 413         clts();
 414         /*
 415          * Init all the features state with header_bv being 0x0
 416          */
 417         xrstor_state(init_xstate_buf, -1);
 418         /*
 419          * Dump the init state again. This is to identify the init state
 420          * of any feature which is not represented by all zero's.
 421          */
 422         xsave_state(init_xstate_buf, -1);
 423         stts();
 424 }
 425
 426 /*
 427  * Enable and initialize the xsave feature.
 428  */
 429 static void __cpuinit xsave_cntxt_init(void)
 430 {
 431         unsigned int eax, ebx, ecx, edx;
 432
 433         cpuid_count(0xd, 0, &eax, &ebx, &ecx, &edx);
 434         pcntxt_mask = eax + ((u64)edx << 32);
 435
 436         if ((pcntxt_mask & XSTATE_FPSSE) != XSTATE_FPSSE) {
 437                 printk(KERN_ERR "FP/SSE not shown under xsave features 0x%llx\n",
 438                        pcntxt_mask);
 439                 BUG();
 440         }
 441
 442         /*
 443          * Support only the state known to OS.
 444          */
 445         pcntxt_mask = pcntxt_mask & XCNTXT_MASK;
 446         __xsave_init();
 447
 448         /*
 449          * Recompute the context size for enabled features
 450          */
 451         cpuid_count(0xd, 0, &eax, &ebx, &ecx, &edx);
 452         xstate_size = ebx;
 453
 454         update_regset_xstate_info(xstate_size, pcntxt_mask);
 455         prepare_fx_sw_frame();
 456
 457         setup_xstate_init();
 458
 459         printk(KERN_INFO "xsave/xrstor: enabled xstate_bv 0x%llx, "
 460                "cntxt size 0x%x\n",
 461                pcntxt_mask, xstate_size);
 462 }
 463
 464 void __cpuinit xsave_init(void)
 465 {
 466         if (!cpu_has_xsave)
 467                 return;
 468
 469         /*
 470          * Boot processor to setup the FP and extended state context info.
 471          */
 472         if (!smp_processor_id())
 473                 xsave_cntxt_init();
 474         __xsave_init();
 475 }