arch/x86/include/asm/system.h

   1 #ifndef _ASM_X86_SYSTEM_H
   2 #define _ASM_X86_SYSTEM_H
   3
   4 #include <asm/asm.h>
   5 #include <asm/segment.h>
   6 #include <asm/cpufeature.h>
   7 #include <asm/cmpxchg.h>
   8 #include <asm/nops.h>
   9
  10 #include <linux/kernel.h>
  11 #include <linux/irqflags.h>
  12
  13 /* entries in ARCH_DLINFO: */
  14 #ifdef CONFIG_IA32_EMULATION
  15 # define AT_VECTOR_SIZE_ARCH 2
  16 #else
  17 # define AT_VECTOR_SIZE_ARCH 1
  18 #endif
  19
  20 struct task_struct; /* one of the stranger aspects of C forward declarations */
  21 struct task_struct *__switch_to(struct task_struct *prev,
  22                                 struct task_struct *next);
  23 struct tss_struct;
  24 void __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p,
  25                       struct tss_struct *tss);
  26
  27 #ifdef CONFIG_X86_32
  28
  29 #ifdef CONFIG_CC_STACKPROTECTOR
  30 #define __switch_canary                                                 \
  31         "movl %P[task_canary](%[next]), %%ebx\n\t"                      \
  32         "movl %%ebx, "__percpu_arg([stack_canary])"\n\t"
  33 #define __switch_canary_oparam                                          \
  34         , [stack_canary] "=m" (per_cpu_var(stack_canary.canary))
  35 #define __switch_canary_iparam                                          \
  36         , [task_canary] "i" (offsetof(struct task_struct, stack_canary))
  37 #else   /* CC_STACKPROTECTOR */
  38 #define __switch_canary
  39 #define __switch_canary_oparam
  40 #define __switch_canary_iparam
  41 #endif  /* CC_STACKPROTECTOR */
  42
  43 /*
  44  * Saving eflags is important. It switches not only IOPL between tasks,
  45  * it also protects other tasks from NT leaking through sysenter etc.
  46  */
  47 #define switch_to(prev, next, last)                                     \
  48 do {                                                                    \
  49         /*                                                              \
  50          * Context-switching clobbers all registers, so we clobber      \
  51          * them explicitly, via unused output variables.                \
  52          * (EAX and EBP is not listed because EBP is saved/restored     \
  53          * explicitly for wchan access and EAX is the return value of   \
  54          * __switch_to())                                               \
  55          */                                                             \
  56         unsigned long ebx, ecx, edx, esi, edi;                          \
  57                                                                         \
  58         asm volatile("pushfl\n\t"               /* save    flags */     \
  59                      "pushl %%ebp\n\t"          /* save    EBP   */     \
  60                      "movl %%esp,%[prev_sp]\n\t"        /* save    ESP   */ \
  61                      "movl %[next_sp],%%esp\n\t"        /* restore ESP   */ \
  62                      "movl $1f,%[prev_ip]\n\t"  /* save    EIP   */     \
  63                      "pushl %[next_ip]\n\t"     /* restore EIP   */     \
  64                      __switch_canary                                    \
  65                      "jmp __switch_to\n"        /* regparm call  */     \
  66                      "1:\t"                                             \
  67                      "popl %%ebp\n\t"           /* restore EBP   */     \
  68                      "popfl\n"                  /* restore flags */     \
  69                                                                         \
  70                      /* output parameters */                            \
  71                      : [prev_sp] "=m" (prev->thread.sp),                \
  72                        [prev_ip] "=m" (prev->thread.ip),                \
  73                        "=a" (last),                                     \
  74                                                                         \
  75                        /* clobbered output registers: */                \
  76                        "=b" (ebx), "=c" (ecx), "=d" (edx),              \
  77                        "=S" (esi), "=D" (edi)                           \
  78                                                                         \
  79                        __switch_canary_oparam                           \
  80                                                                         \
  81                        /* input parameters: */                          \
  82                      : [next_sp]  "m" (next->thread.sp),                \
  83                        [next_ip]  "m" (next->thread.ip),                \
  84                                                                         \
  85                        /* regparm parameters for __switch_to(): */      \
  86                        [prev]     "a" (prev),                           \
  87                        [next]     "d" (next)                            \
  88                                                                         \
  89                        __switch_canary_iparam                           \
  90                                                                         \
  91                      : /* reloaded segment registers */                 \
  92                         "memory");                                      \
  93 } while (0)
  94
  95 /*
  96  * disable hlt during certain critical i/o operations
  97  */
  98 #define HAVE_DISABLE_HLT
  99 #else
 100 #define __SAVE(reg, offset) "movq %%" #reg ",(14-" #offset ")*8(%%rsp)\n\t"
 101 #define __RESTORE(reg, offset) "movq (14-" #offset ")*8(%%rsp),%%" #reg "\n\t"
 102
 103 /* frame pointer must be last for get_wchan */
 104 #define SAVE_CONTEXT    "pushf ; pushq %%rbp ; movq %%rsi,%%rbp\n\t"
 105 #define RESTORE_CONTEXT "movq %%rbp,%%rsi ; popq %%rbp ; popf\t"
 106
 107 #define __EXTRA_CLOBBER  \
 108         , "rcx", "rbx", "rdx", "r8", "r9", "r10", "r11", \
 109           "r12", "r13", "r14", "r15"
 110
 111 #ifdef CONFIG_CC_STACKPROTECTOR
 112 #define __switch_canary                                                   \
 113         "movq %P[task_canary](%%rsi),%%r8\n\t"                            \
 114         "movq %%r8,"__percpu_arg([gs_canary])"\n\t"
 115 #define __switch_canary_oparam                                            \
 116         , [gs_canary] "=m" (per_cpu_var(irq_stack_union.stack_canary))
 117 #define __switch_canary_iparam                                            \
 118         , [task_canary] "i" (offsetof(struct task_struct, stack_canary))
 119 #else   /* CC_STACKPROTECTOR */
 120 #define __switch_canary
 121 #define __switch_canary_oparam
 122 #define __switch_canary_iparam
 123 #endif  /* CC_STACKPROTECTOR */
 124
 125 /* Save restore flags to clear handle leaking NT */
 126 #define switch_to(prev, next, last) \
 127         asm volatile(SAVE_CONTEXT                                         \
 128              "movq %%rsp,%P[threadrsp](%[prev])\n\t" /* save RSP */       \
 129              "movq %P[threadrsp](%[next]),%%rsp\n\t" /* restore RSP */    \
 130              "call __switch_to\n\t"                                       \
 131              ".globl thread_return\n"                                     \
 132              "thread_return:\n\t"                                         \
 133              "movq "__percpu_arg([current_task])",%%rsi\n\t"              \
 134              __switch_canary                                              \
 135              "movq %P[thread_info](%%rsi),%%r8\n\t"                       \
 136              "movq %%rax,%%rdi\n\t"                                       \
 137              "testl  %[_tif_fork],%P[ti_flags](%%r8)\n\t"         \
 138              "jnz   ret_from_fork\n\t"                                    \
 139              RESTORE_CONTEXT                                              \
 140              : "=a" (last)                                                \
 141                __switch_canary_oparam                                     \
 142              : [next] "S" (next), [prev] "D" (prev),                      \
 143                [threadrsp] "i" (offsetof(struct task_struct, thread.sp)), \
 144                [ti_flags] "i" (offsetof(struct thread_info, flags)),      \
 145                [_tif_fork] "i" (_TIF_FORK),                               \
 146                [thread_info] "i" (offsetof(struct task_struct, stack)),   \
 147                [current_task] "m" (per_cpu_var(current_task))             \
 148                __switch_canary_iparam                                     \
 149              : "memory", "cc" __EXTRA_CLOBBER)
 150 #endif
 151
 152 #ifdef __KERNEL__
 153
 154 extern void native_load_gs_index(unsigned);
 155
 156 /*
 157  * Load a segment. Fall back on loading the zero
 158  * segment if something goes wrong..
 159  */
 160 #define loadsegment(seg, value)                 \
 161         asm volatile("\n"                       \
 162                      "1:\t"                     \
 163                      "movl %k0,%%" #seg "\n"    \
 164                      "2:\n"                     \
 165                      ".section .fixup,\"ax\"\n" \
 166                      "3:\t"                     \
 167                      "movl %k1, %%" #seg "\n\t" \
 168                      "jmp 2b\n"                 \
 169                      ".previous\n"              \
 170                      _ASM_EXTABLE(1b,3b)        \
 171                      : :"r" (value), "r" (0) : "memory")
 172
 173
 174 /*
 175  * Save a segment register away
 176  */
 177 #define savesegment(seg, value)                         \
 178         asm("mov %%" #seg ",%0":"=r" (value) : : "memory")
 179
 180 /*
 181  * x86_32 user gs accessors.
 182  */
 183 #ifdef CONFIG_X86_32
 184 #ifdef CONFIG_X86_32_LAZY_GS
 185 #define get_user_gs(regs)       (u16)({unsigned long v; savesegment(gs, v); v;})
 186 #define set_user_gs(regs, v)    loadsegment(gs, (unsigned long)(v))
 187 #define task_user_gs(tsk)       ((tsk)->thread.gs)
 188 #define lazy_save_gs(v)         savesegment(gs, (v))
 189 #define lazy_load_gs(v)         loadsegment(gs, (v))
 190 #else   /* X86_32_LAZY_GS */
 191 #define get_user_gs(regs)       (u16)((regs)->gs)
 192 #define set_user_gs(regs, v)    do { (regs)->gs = (v); } while (0)
 193 #define task_user_gs(tsk)       (task_pt_regs(tsk)->gs)
 194 #define lazy_save_gs(v)         do { } while (0)
 195 #define lazy_load_gs(v)         do { } while (0)
 196 #endif  /* X86_32_LAZY_GS */
 197 #endif  /* X86_32 */
 198
 199 static inline unsigned long get_limit(unsigned long segment)
 200 {
 201         unsigned long __limit;
 202         asm("lsll %1,%0" : "=r" (__limit) : "r" (segment));
 203         return __limit + 1;
 204 }
 205
 206 static inline void native_clts(void)
 207 {
 208         asm volatile("clts");
 209 }
 210
 211 /*
 212  * Volatile isn't enough to prevent the compiler from reordering the
 213  * read/write functions for the control registers and messing everything up.
 214  * A memory clobber would solve the problem, but would prevent reordering of
 215  * all loads stores around it, which can hurt performance. Solution is to
 216  * use a variable and mimic reads and writes to it to enforce serialization
 217  */
 218 static unsigned long __force_order;
 219
 220 static inline unsigned long native_read_cr0(void)
 221 {
 222         unsigned long val;
 223         asm volatile("mov %%cr0,%0\n\t" : "=r" (val), "=m" (__force_order));
 224         return val;
 225 }
 226
 227 static inline void native_write_cr0(unsigned long val)
 228 {
 229         asm volatile("mov %0,%%cr0": : "r" (val), "m" (__force_order));
 230 }
 231
 232 static inline unsigned long native_read_cr2(void)
 233 {
 234         unsigned long val;
 235         asm volatile("mov %%cr2,%0\n\t" : "=r" (val), "=m" (__force_order));
 236         return val;
 237 }
 238
 239 static inline void native_write_cr2(unsigned long val)
 240 {
 241         asm volatile("mov %0,%%cr2": : "r" (val), "m" (__force_order));
 242 }
 243
 244 static inline unsigned long native_read_cr3(void)
 245 {
 246         unsigned long val;
 247         asm volatile("mov %%cr3,%0\n\t" : "=r" (val), "=m" (__force_order));
 248         return val;
 249 }
 250
 251 static inline void native_write_cr3(unsigned long val)
 252 {
 253         asm volatile("mov %0,%%cr3": : "r" (val), "m" (__force_order));
 254 }
 255
 256 static inline unsigned long native_read_cr4(void)
 257 {
 258         unsigned long val;
 259         asm volatile("mov %%cr4,%0\n\t" : "=r" (val), "=m" (__force_order));
 260         return val;
 261 }
 262
 263 static inline unsigned long native_read_cr4_safe(void)
 264 {
 265         unsigned long val;
 266         /* This could fault if %cr4 does not exist. In x86_64, a cr4 always
 267          * exists, so it will never fail. */
 268 #ifdef CONFIG_X86_32
 269         asm volatile("1: mov %%cr4, %0\n"
 270                      "2:\n"
 271                      _ASM_EXTABLE(1b, 2b)
 272                      : "=r" (val), "=m" (__force_order) : "0" (0));
 273 #else
 274         val = native_read_cr4();
 275 #endif
 276         return val;
 277 }
 278
 279 static inline void native_write_cr4(unsigned long val)
 280 {
 281         asm volatile("mov %0,%%cr4": : "r" (val), "m" (__force_order));
 282 }
 283
 284 #ifdef CONFIG_X86_64
 285 static inline unsigned long native_read_cr8(void)
 286 {
 287         unsigned long cr8;
 288         asm volatile("movq %%cr8,%0" : "=r" (cr8));
 289         return cr8;
 290 }
 291
 292 static inline void native_write_cr8(unsigned long val)
 293 {
 294         asm volatile("movq %0,%%cr8" :: "r" (val) : "memory");
 295 }
 296 #endif
 297
 298 static inline void native_wbinvd(void)
 299 {
 300         asm volatile("wbinvd": : :"memory");
 301 }
 302
 303 #ifdef CONFIG_PARAVIRT
 304 #include <asm/paravirt.h>
 305 #else
 306 #define read_cr0()      (native_read_cr0())
 307 #define write_cr0(x)    (native_write_cr0(x))
 308 #define read_cr2()      (native_read_cr2())
 309 #define write_cr2(x)    (native_write_cr2(x))
 310 #define read_cr3()      (native_read_cr3())
 311 #define write_cr3(x)    (native_write_cr3(x))
 312 #define read_cr4()      (native_read_cr4())
 313 #define read_cr4_safe() (native_read_cr4_safe())
 314 #define write_cr4(x)    (native_write_cr4(x))
 315 #define wbinvd()        (native_wbinvd())
 316 #ifdef CONFIG_X86_64
 317 #define read_cr8()      (native_read_cr8())
 318 #define write_cr8(x)    (native_write_cr8(x))
 319 #define load_gs_index   native_load_gs_index
 320 #endif
 321
 322 /* Clear the 'TS' bit */
 323 #define clts()          (native_clts())
 324
 325 #endif/* CONFIG_PARAVIRT */
 326
 327 #define stts() write_cr0(read_cr0() | X86_CR0_TS)
 328
 329 #endif /* __KERNEL__ */
 330
 331 static inline void clflush(volatile void *__p)
 332 {
 333         asm volatile("clflush %0" : "+m" (*(volatile char __force *)__p));
 334 }
 335
 336 #define nop() asm volatile ("nop")
 337
 338 void disable_hlt(void);
 339 void enable_hlt(void);
 340
 341 void cpu_idle_wait(void);
 342
 343 extern unsigned long arch_align_stack(unsigned long sp);
 344 extern void free_init_pages(char *what, unsigned long begin, unsigned long end);
 345
 346 void default_idle(void);
 347
 348 void stop_this_cpu(void *dummy);
 349
 350 /*
 351  * Force strict CPU ordering.
 352  * And yes, this is required on UP too when we're talking
 353  * to devices.
 354  */
 355 #ifdef CONFIG_X86_32
 356 /*
 357  * Some non-Intel clones support out of order store. wmb() ceases to be a
 358  * nop for these.
 359  */
 360 #define mb() alternative("lock; addl $0,0(%%esp)", "mfence", X86_FEATURE_XMM2)
 361 #define rmb() alternative("lock; addl $0,0(%%esp)", "lfence", X86_FEATURE_XMM2)
 362 #define wmb() alternative("lock; addl $0,0(%%esp)", "sfence", X86_FEATURE_XMM)
 363 #else
 364 #define mb()    asm volatile("mfence":::"memory")
 365 #define rmb()   asm volatile("lfence":::"memory")
 366 #define wmb()   asm volatile("sfence" ::: "memory")
 367 #endif
 368
 369 /**
 370  * read_barrier_depends - Flush all pending reads that subsequents reads
 371  * depend on.
 372  *
 373  * No data-dependent reads from memory-like regions are ever reordered
 374  * over this barrier.  All reads preceding this primitive are guaranteed
 375  * to access memory (but not necessarily other CPUs' caches) before any
 376  * reads following this primitive that depend on the data return by
 377  * any of the preceding reads.  This primitive is much lighter weight than
 378  * rmb() on most CPUs, and is never heavier weight than is
 379  * rmb().
 380  *
 381  * These ordering constraints are respected by both the local CPU
 382  * and the compiler.
 383  *
 384  * Ordering is not guaranteed by anything other than these primitives,
 385  * not even by data dependencies.  See the documentation for
 386  * memory_barrier() for examples and URLs to more information.
 387  *
 388  * For example, the following code would force ordering (the initial
 389  * value of "a" is zero, "b" is one, and "p" is "&a"):
 390  *
 391  * <programlisting>
 392  *      CPU 0                           CPU 1
 393  *
 394  *      b = 2;
 395  *      memory_barrier();
 396  *      p = &b;                         q = p;
 397  *                                      read_barrier_depends();
 398  *                                      d = *q;
 399  * </programlisting>
 400  *
 401  * because the read of "*q" depends on the read of "p" and these
 402  * two reads are separated by a read_barrier_depends().  However,
 403  * the following code, with the same initial values for "a" and "b":
 404  *
 405  * <programlisting>
 406  *      CPU 0                           CPU 1
 407  *
 408  *      a = 2;
 409  *      memory_barrier();
 410  *      b = 3;                          y = b;
 411  *                                      read_barrier_depends();
 412  *                                      x = a;
 413  * </programlisting>
 414  *
 415  * does not enforce ordering, since there is no data dependency between
 416  * the read of "a" and the read of "b".  Therefore, on some CPUs, such
 417  * as Alpha, "y" could be set to 3 and "x" to 0.  Use rmb()
 418  * in cases like this where there are no data dependencies.
 419  **/
 420
 421 #define read_barrier_depends()  do { } while (0)
 422
 423 #ifdef CONFIG_SMP
 424 #define smp_mb()        mb()
 425 #ifdef CONFIG_X86_PPRO_FENCE
 426 # define smp_rmb()      rmb()
 427 #else
 428 # define smp_rmb()      barrier()
 429 #endif
 430 #ifdef CONFIG_X86_OOSTORE
 431 # define smp_wmb()      wmb()
 432 #else
 433 # define smp_wmb()      barrier()
 434 #endif
 435 #define smp_read_barrier_depends()      read_barrier_depends()
 436 #define set_mb(var, value) do { (void)xchg(&var, value); } while (0)
 437 #else
 438 #define smp_mb()        barrier()
 439 #define smp_rmb()       barrier()
 440 #define smp_wmb()       barrier()
 441 #define smp_read_barrier_depends()      do { } while (0)
 442 #define set_mb(var, value) do { var = value; barrier(); } while (0)
 443 #endif
 444
 445 /*
 446  * Stop RDTSC speculation. This is needed when you need to use RDTSC
 447  * (or get_cycles or vread that possibly accesses the TSC) in a defined
 448  * code region.
 449  *
 450  * (Could use an alternative three way for this if there was one.)
 451  */
 452 static inline void rdtsc_barrier(void)
 453 {
 454         alternative(ASM_NOP3, "mfence", X86_FEATURE_MFENCE_RDTSC);
 455         alternative(ASM_NOP3, "lfence", X86_FEATURE_LFENCE_RDTSC);
 456 }
 457
 458 #endif /* _ASM_X86_SYSTEM_H */