libpthread/nptl/sysdeps/x86_64/tls.h

   1 /* Definition for thread-local data handling.  nptl/x86_64 version.
   2    Copyright (C) 2002-2007, 2008, 2009 Free Software Foundation, Inc.
   3    This file is part of the GNU C Library.
   4
   5    The GNU C Library is free software; you can redistribute it and/or
   6    modify it under the terms of the GNU Lesser General Public
   7    License as published by the Free Software Foundation; either
   8    version 2.1 of the License, or (at your option) any later version.
   9
  10    The GNU C Library is distributed in the hope that it will be useful,
  11    but WITHOUT ANY WARRANTY; without even the implied warranty of
  12    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13    Lesser General Public License for more details.
  14
  15    You should have received a copy of the GNU Lesser General Public
  16    License along with the GNU C Library; if not, see
  17    <http://www.gnu.org/licenses/>.  */
  18
  19 #ifndef _TLS_H
  20 #define _TLS_H  1
  21
  22 #ifndef __ASSEMBLER__
  23 # include <asm/prctl.h> /* For ARCH_SET_FS.  */
  24 # include <stdbool.h>
  25 # include <stddef.h>
  26 # include <stdint.h>
  27 # include <stdlib.h>
  28 # include <sysdep.h>
  29 # include <bits/kernel-features.h>
  30 # include <bits/wordsize.h>
  31 # include <xmmintrin.h>
  32
  33
  34 /* Type for the dtv.  */
  35 typedef union dtv
  36 {
  37   size_t counter;
  38   struct
  39   {
  40     void *val;
  41     bool is_static;
  42   } pointer;
  43 } dtv_t;
  44
  45
  46 typedef struct
  47 {
  48   void *tcb;            /* Pointer to the TCB.  Not necessarily the
  49                            thread descriptor used by libpthread.  */
  50   dtv_t *dtv;
  51   void *self;           /* Pointer to the thread descriptor.  */
  52   int multiple_threads;
  53   int gscope_flag;
  54   uintptr_t sysinfo;
  55   uintptr_t stack_guard;
  56   uintptr_t pointer_guard;
  57   unsigned long int vgetcpu_cache[2];
  58 # ifndef __ASSUME_PRIVATE_FUTEX
  59   int private_futex;
  60 # else
  61   int __unused1;
  62 # endif
  63 # if __WORDSIZE == 64
  64   int rtld_must_xmm_save;
  65 # endif
  66   /* Reservation of some values for the TM ABI.  */
  67   void *__private_tm[5];
  68 # if __WORDSIZE == 64
  69   long int __unused2;
  70   /* Have space for the post-AVX register size.  */
  71   __m128 rtld_savespace_sse[8][4];
  72
  73   void *__padding[8];
  74 # endif
  75 } tcbhead_t;
  76
  77 #else /* __ASSEMBLER__ */
  78 # include <tcb-offsets.h>
  79 #endif
  80
  81
  82 /* We require TLS support in the tools.  */
  83 #define HAVE_TLS_SUPPORT                1
  84 #define HAVE___THREAD 1
  85 #define HAVE_TLS_MODEL_ATTRIBUTE 1
  86
  87 /* Signal that TLS support is available.  */
  88 #define USE_TLS        1
  89
  90 /* Alignment requirement for the stack.  */
  91 #define STACK_ALIGN     16
  92
  93
  94 #ifndef __ASSEMBLER__
  95 /* Get system call information.  */
  96 # include <sysdep.h>
  97
  98
  99 /* Get the thread descriptor definition.  */
 100 # include <descr.h>
 101
 102 #ifndef LOCK_PREFIX
 103 # ifdef UP
 104 #  define LOCK_PREFIX   /* nothing */
 105 # else
 106 #  define LOCK_PREFIX   "lock;"
 107 # endif
 108 #endif
 109
 110 /* This is the size of the initial TCB.  Can't be just sizeof (tcbhead_t),
 111    because NPTL getpid, __libc_alloca_cutoff etc. need (almost) the whole
 112    struct pthread even when not linked with -lpthread.  */
 113 # define TLS_INIT_TCB_SIZE sizeof (struct pthread)
 114
 115 /* Alignment requirements for the initial TCB.  */
 116 # define TLS_INIT_TCB_ALIGN __alignof__ (struct pthread)
 117
 118 /* This is the size of the TCB.  */
 119 # define TLS_TCB_SIZE sizeof (struct pthread)
 120
 121 /* Alignment requirements for the TCB.  */
 122 //# define TLS_TCB_ALIGN __alignof__ (struct pthread)
 123 // Normally the above would be correct  But we have to store post-AVX
 124 // vector registers in the TCB and we want the storage to be aligned.
 125 // unfortunately there isn't yet a type for these values and hence no
 126 // 32-byte alignment requirement.  Make this explicit, for now.
 127 # define TLS_TCB_ALIGN 32
 128
 129 /* The TCB can have any size and the memory following the address the
 130    thread pointer points to is unspecified.  Allocate the TCB there.  */
 131 # define TLS_TCB_AT_TP  1
 132
 133
 134 /* Install the dtv pointer.  The pointer passed is to the element with
 135    index -1 which contain the length.  */
 136 # define INSTALL_DTV(descr, dtvp) \
 137   ((tcbhead_t *) (descr))->dtv = (dtvp) + 1
 138
 139 /* Install new dtv for current thread.  */
 140 # define INSTALL_NEW_DTV(dtvp) \
 141   ({ struct pthread *__pd;                                                    \
 142      THREAD_SETMEM (__pd, header.dtv, (dtvp)); })
 143
 144 /* Return dtv of given thread descriptor.  */
 145 # define GET_DTV(descr) \
 146   (((tcbhead_t *) (descr))->dtv)
 147
 148
 149 /* Macros to load from and store into segment registers.  */
 150 # define TLS_GET_FS() \
 151   ({ int __seg; __asm__ ("movl %%fs, %0" : "=q" (__seg)); __seg; })
 152 # define TLS_SET_FS(val) \
 153   __asm__ ("movl %0, %%fs" :: "q" (val))
 154
 155
 156 /* Code to initially initialize the thread pointer.  This might need
 157    special attention since 'errno' is not yet available and if the
 158    operation can cause a failure 'errno' must not be touched.
 159
 160    We have to make the syscall for both uses of the macro since the
 161    address might be (and probably is) different.  */
 162 # define TLS_INIT_TP(thrdescr, secondcall) \
 163   ({ void *_thrdescr = (thrdescr);                                            \
 164      tcbhead_t *_head = _thrdescr;                                            \
 165      int _result;                                                             \
 166                                                                               \
 167      _head->tcb = _thrdescr;                                                  \
 168      /* For now the thread descriptor is at the same address.  */             \
 169      _head->self = _thrdescr;                                                 \
 170                                                                               \
 171      /* It is a simple syscall to set the %fs value for the thread.  */       \
 172      __asm__ __volatile__ ("syscall"                                                  \
 173                    : "=a" (_result)                                           \
 174                    : "0" ((unsigned long int) __NR_arch_prctl),               \
 175                      "D" ((unsigned long int) ARCH_SET_FS),                   \
 176                      "S" (_thrdescr)                                          \
 177                    : "memory", "cc", "r11", "cx");                            \
 178                                                                               \
 179     _result ? "cannot set %fs base address for thread-local storage" : 0;     \
 180   })
 181
 182
 183 /* Return the address of the dtv for the current thread.  */
 184 # define THREAD_DTV() \
 185   ({ struct pthread *__pd;                                                    \
 186      THREAD_GETMEM (__pd, header.dtv); })
 187
 188
 189 /* Return the thread descriptor for the current thread.
 190
 191    The contained asm must *not* be marked __volatile__ since otherwise
 192    assignments like
 193         pthread_descr self = thread_self();
 194    do not get optimized away.  */
 195 # define THREAD_SELF \
 196   ({ struct pthread *__self;                                                  \
 197      __asm__ ("movq %%fs:%c1,%q0" : "=r" (__self)                                     \
 198           : "i" (offsetof (struct pthread, header.self)));                    \
 199      __self;})
 200
 201 /* Magic for libthread_db to know how to do THREAD_SELF.  */
 202 # define DB_THREAD_SELF_INCLUDE  <sys/reg.h> /* For the FS constant.  */
 203 # define DB_THREAD_SELF CONST_THREAD_AREA (64, FS)
 204
 205 /* Read member of the thread descriptor directly.  */
 206 # define THREAD_GETMEM(descr, member) \
 207   ({ __typeof (descr->member) __value;                                        \
 208      if (sizeof (__value) == 1)                                               \
 209        __asm__ __volatile__ ("movb %%fs:%P2,%b0"                                      \
 210                      : "=q" (__value)                                         \
 211                      : "0" (0), "i" (offsetof (struct pthread, member)));     \
 212      else if (sizeof (__value) == 4)                                          \
 213        __asm__ __volatile__ ("movl %%fs:%P1,%0"                                       \
 214                      : "=r" (__value)                                         \
 215                      : "i" (offsetof (struct pthread, member)));              \
 216      else                                                                     \
 217        {                                                                      \
 218          if (sizeof (__value) != 8)                                           \
 219            /* There should not be any value with a size other than 1,         \
 220               4 or 8.  */                                                     \
 221            abort ();                                                          \
 222                                                                               \
 223          __asm__ __volatile__ ("movq %%fs:%P1,%q0"                                    \
 224                        : "=r" (__value)                                       \
 225                        : "i" (offsetof (struct pthread, member)));            \
 226        }                                                                      \
 227      __value; })
 228
 229
 230 /* Same as THREAD_GETMEM, but the member offset can be non-constant.  */
 231 # define THREAD_GETMEM_NC(descr, member, idx) \
 232   ({ __typeof (descr->member[0]) __value;                                     \
 233      if (sizeof (__value) == 1)                                               \
 234        __asm__ __volatile__ ("movb %%fs:%P2(%q3),%b0"                                 \
 235                      : "=q" (__value)                                         \
 236                      : "0" (0), "i" (offsetof (struct pthread, member[0])),   \
 237                        "r" (idx));                                            \
 238      else if (sizeof (__value) == 4)                                          \
 239        __asm__ __volatile__ ("movl %%fs:%P1(,%q2,4),%0"                               \
 240                      : "=r" (__value)                                         \
 241                      : "i" (offsetof (struct pthread, member[0])), "r" (idx));\
 242      else                                                                     \
 243        {                                                                      \
 244          if (sizeof (__value) != 8)                                           \
 245            /* There should not be any value with a size other than 1,         \
 246               4 or 8.  */                                                     \
 247            abort ();                                                          \
 248                                                                               \
 249          __asm__ __volatile__ ("movq %%fs:%P1(,%q2,8),%q0"                            \
 250                        : "=r" (__value)                                       \
 251                        : "i" (offsetof (struct pthread, member[0])),          \
 252                          "r" (idx));                                          \
 253        }                                                                      \
 254      __value; })
 255
 256
 257 /* Loading addresses of objects on x86-64 needs to be treated special
 258    when generating PIC code.  */
 259 #ifdef __pic__
 260 # define IMM_MODE "nr"
 261 #else
 262 # define IMM_MODE "ir"
 263 #endif
 264
 265
 266 /* Same as THREAD_SETMEM, but the member offset can be non-constant.  */
 267 # define THREAD_SETMEM(descr, member, value) \
 268   ({ if (sizeof (descr->member) == 1)                                         \
 269        __asm__ __volatile__ ("movb %b0,%%fs:%P1" :                                    \
 270                      : "iq" (value),                                          \
 271                        "i" (offsetof (struct pthread, member)));              \
 272      else if (sizeof (descr->member) == 4)                                    \
 273        __asm__ __volatile__ ("movl %0,%%fs:%P1" :                                     \
 274                      : IMM_MODE (value),                                      \
 275                        "i" (offsetof (struct pthread, member)));              \
 276      else                                                                     \
 277        {                                                                      \
 278          if (sizeof (descr->member) != 8)                                     \
 279            /* There should not be any value with a size other than 1,         \
 280               4 or 8.  */                                                     \
 281            abort ();                                                          \
 282                                                                               \
 283          __asm__ __volatile__ ("movq %q0,%%fs:%P1" :                                  \
 284                        : IMM_MODE ((unsigned long int) value),                \
 285                          "i" (offsetof (struct pthread, member)));            \
 286        }})
 287
 288
 289 /* Set member of the thread descriptor directly.  */
 290 # define THREAD_SETMEM_NC(descr, member, idx, value) \
 291   ({ if (sizeof (descr->member[0]) == 1)                                      \
 292        __asm__ __volatile__ ("movb %b0,%%fs:%P1(%q2)" :                               \
 293                      : "iq" (value),                                          \
 294                        "i" (offsetof (struct pthread, member[0])),            \
 295                        "r" (idx));                                            \
 296      else if (sizeof (descr->member[0]) == 4)                                 \
 297        __asm__ __volatile__ ("movl %0,%%fs:%P1(,%q2,4)" :                             \
 298                      : IMM_MODE (value),                                      \
 299                        "i" (offsetof (struct pthread, member[0])),            \
 300                        "r" (idx));                                            \
 301      else                                                                     \
 302        {                                                                      \
 303          if (sizeof (descr->member[0]) != 8)                                  \
 304            /* There should not be any value with a size other than 1,         \
 305               4 or 8.  */                                                     \
 306            abort ();                                                          \
 307                                                                               \
 308          __asm__ __volatile__ ("movq %q0,%%fs:%P1(,%q2,8)" :                          \
 309                        : IMM_MODE ((unsigned long int) value),                \
 310                          "i" (offsetof (struct pthread, member[0])),          \
 311                          "r" (idx));                                          \
 312        }})
 313
 314
 315 /* Atomic compare and exchange on TLS, returning old value.  */
 316 # define THREAD_ATOMIC_CMPXCHG_VAL(descr, member, newval, oldval) \
 317   ({ __typeof (descr->member) __ret;                                          \
 318      __typeof (oldval) __old = (oldval);                                      \
 319      if (sizeof (descr->member) == 4)                                         \
 320        __asm__ __volatile__ (LOCK_PREFIX "cmpxchgl %2, %%fs:%P3"                      \
 321                      : "=a" (__ret)                                           \
 322                      : "0" (__old), "r" (newval),                             \
 323                        "i" (offsetof (struct pthread, member)));              \
 324      else                                                                     \
 325        /* Not necessary for other sizes in the moment.  */                    \
 326        abort ();                                                              \
 327      __ret; })
 328
 329
 330 /* Atomic logical and.  */
 331 # define THREAD_ATOMIC_AND(descr, member, val) \
 332   (void) ({ if (sizeof ((descr)->member) == 4)                                \
 333               __asm__ __volatile__ (LOCK_PREFIX "andl %1, %%fs:%P0"                   \
 334                             :: "i" (offsetof (struct pthread, member)),       \
 335                                "ir" (val));                                   \
 336             else                                                              \
 337               /* Not necessary for other sizes in the moment.  */             \
 338               abort (); })
 339
 340
 341 /* Atomic set bit.  */
 342 # define THREAD_ATOMIC_BIT_SET(descr, member, bit) \
 343   (void) ({ if (sizeof ((descr)->member) == 4)                                \
 344               __asm__ __volatile__ (LOCK_PREFIX "orl %1, %%fs:%P0"                    \
 345                             :: "i" (offsetof (struct pthread, member)),       \
 346                                "ir" (1 << (bit)));                            \
 347             else                                                              \
 348               /* Not necessary for other sizes in the moment.  */             \
 349               abort (); })
 350
 351
 352 # define CALL_THREAD_FCT(descr) \
 353   ({ void *__res;                                                             \
 354      __asm__ __volatile__ ("movq %%fs:%P2, %%rdi\n\t"                                 \
 355                    "callq *%%fs:%P1"                                          \
 356                    : "=a" (__res)                                             \
 357                    : "i" (offsetof (struct pthread, start_routine)),          \
 358                      "i" (offsetof (struct pthread, arg))                     \
 359                    : "di", "si", "cx", "dx", "r8", "r9", "r10", "r11",        \
 360                      "memory", "cc");                                         \
 361      __res; })
 362
 363
 364 /* Set the stack guard field in TCB head.  */
 365 # define THREAD_SET_STACK_GUARD(value) \
 366     THREAD_SETMEM (THREAD_SELF, header.stack_guard, value)
 367 # define THREAD_COPY_STACK_GUARD(descr) \
 368     ((descr)->header.stack_guard                                              \
 369      = THREAD_GETMEM (THREAD_SELF, header.stack_guard))
 370
 371
 372 /* Set the pointer guard field in the TCB head.  */
 373 # define THREAD_SET_POINTER_GUARD(value) \
 374   THREAD_SETMEM (THREAD_SELF, header.pointer_guard, value)
 375 # define THREAD_COPY_POINTER_GUARD(descr) \
 376   ((descr)->header.pointer_guard                                              \
 377    = THREAD_GETMEM (THREAD_SELF, header.pointer_guard))
 378
 379
 380 /* Get and set the global scope generation counter in the TCB head.  */
 381 # define THREAD_GSCOPE_FLAG_UNUSED 0
 382 # define THREAD_GSCOPE_FLAG_USED   1
 383 # define THREAD_GSCOPE_FLAG_WAIT   2
 384 # define THREAD_GSCOPE_RESET_FLAG() \
 385   do                                                                          \
 386     { int __res;                                                              \
 387       __asm__ __volatile__ ("xchgl %0, %%fs:%P1"                                      \
 388                     : "=r" (__res)                                            \
 389                     : "i" (offsetof (struct pthread, header.gscope_flag)),    \
 390                       "0" (THREAD_GSCOPE_FLAG_UNUSED));                       \
 391       if (__res == THREAD_GSCOPE_FLAG_WAIT)                                   \
 392         lll_futex_wake (&THREAD_SELF->header.gscope_flag, 1, LLL_PRIVATE);    \
 393     }                                                                         \
 394   while (0)
 395 # define THREAD_GSCOPE_SET_FLAG() \
 396   THREAD_SETMEM (THREAD_SELF, header.gscope_flag, THREAD_GSCOPE_FLAG_USED)
 397 # define THREAD_GSCOPE_WAIT() \
 398   GL(dl_wait_lookup_done) ()
 399
 400
 401 # ifdef SHARED
 402 /* Defined in dl-trampoline.S.  */
 403 extern void _dl_x86_64_save_sse (void);
 404 extern void _dl_x86_64_restore_sse (void);
 405
 406 # define RTLD_CHECK_FOREIGN_CALL \
 407   (THREAD_GETMEM (THREAD_SELF, header.rtld_must_xmm_save) != 0)
 408
 409 /* NB: Don't use the xchg operation because that would imply a lock
 410    prefix which is expensive and unnecessary.  The cache line is also
 411    not contested at all.  */
 412 #  define RTLD_ENABLE_FOREIGN_CALL \
 413   int old_rtld_must_xmm_save = THREAD_GETMEM (THREAD_SELF,                    \
 414                                               header.rtld_must_xmm_save);     \
 415   THREAD_SETMEM (THREAD_SELF, header.rtld_must_xmm_save, 1)
 416
 417 #  define RTLD_PREPARE_FOREIGN_CALL \
 418   do if (THREAD_GETMEM (THREAD_SELF, header.rtld_must_xmm_save))              \
 419     {                                                                         \
 420       _dl_x86_64_save_sse ();                                                 \
 421       THREAD_SETMEM (THREAD_SELF, header.rtld_must_xmm_save, 0);              \
 422     }                                                                         \
 423   while (0)
 424
 425 #  define RTLD_FINALIZE_FOREIGN_CALL \
 426   do {                                                                        \
 427     if (THREAD_GETMEM (THREAD_SELF, header.rtld_must_xmm_save) == 0)          \
 428       _dl_x86_64_restore_sse ();                                              \
 429     THREAD_SETMEM (THREAD_SELF, header.rtld_must_xmm_save,                    \
 430                    old_rtld_must_xmm_save);                                   \
 431   } while (0)
 432 # endif
 433
 434
 435 #endif /* __ASSEMBLER__ */
 436
 437 #endif  /* tls.h */