arch/x86/kernel/process.c

   1 #include <linux/errno.h>
   2 #include <linux/kernel.h>
   3 #include <linux/mm.h>
   4 #include <linux/smp.h>
   5 #include <linux/prctl.h>
   6 #include <linux/slab.h>
   7 #include <linux/sched.h>
   8 #include <linux/module.h>
   9 #include <linux/pm.h>
  10 #include <linux/clockchips.h>
  11 #include <linux/random.h>
  12 #include <trace/events/power.h>
  13 #include <linux/hw_breakpoint.h>
  14 #include <asm/system.h>
  15 #include <asm/apic.h>
  16 #include <asm/syscalls.h>
  17 #include <asm/idle.h>
  18 #include <asm/uaccess.h>
  19 #include <asm/i387.h>
  20 #include <asm/ds.h>
  21 #include <asm/debugreg.h>
  22
  23 unsigned long idle_halt;
  24 EXPORT_SYMBOL(idle_halt);
  25 unsigned long idle_nomwait;
  26 EXPORT_SYMBOL(idle_nomwait);
  27
  28 struct kmem_cache *task_xstate_cachep;
  29
  30 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
  31 {
  32         *dst = *src;
  33         if (src->thread.xstate) {
  34                 dst->thread.xstate = kmem_cache_alloc(task_xstate_cachep,
  35                                                       GFP_KERNEL);
  36                 if (!dst->thread.xstate)
  37                         return -ENOMEM;
  38                 WARN_ON((unsigned long)dst->thread.xstate & 15);
  39                 memcpy(dst->thread.xstate, src->thread.xstate, xstate_size);
  40         }
  41         return 0;
  42 }
  43
  44 void free_thread_xstate(struct task_struct *tsk)
  45 {
  46         if (tsk->thread.xstate) {
  47                 kmem_cache_free(task_xstate_cachep, tsk->thread.xstate);
  48                 tsk->thread.xstate = NULL;
  49         }
  50
  51         WARN(tsk->thread.ds_ctx, "leaking DS context\n");
  52 }
  53
  54 void free_thread_info(struct thread_info *ti)
  55 {
  56         free_thread_xstate(ti->task);
  57         free_pages((unsigned long)ti, get_order(THREAD_SIZE));
  58 }
  59
  60 void arch_task_cache_init(void)
  61 {
  62         task_xstate_cachep =
  63                 kmem_cache_create("task_xstate", xstate_size,
  64                                   __alignof__(union thread_xstate),
  65                                   SLAB_PANIC | SLAB_NOTRACK, NULL);
  66 }
  67
  68 /*
  69  * Free current thread data structures etc..
  70  */
  71 void exit_thread(void)
  72 {
  73         struct task_struct *me = current;
  74         struct thread_struct *t = &me->thread;
  75         unsigned long *bp = t->io_bitmap_ptr;
  76
  77         if (bp) {
  78                 struct tss_struct *tss = &per_cpu(init_tss, get_cpu());
  79
  80                 t->io_bitmap_ptr = NULL;
  81                 clear_thread_flag(TIF_IO_BITMAP);
  82                 /*
  83                  * Careful, clear this in the TSS too:
  84                  */
  85                 memset(tss->io_bitmap, 0xff, t->io_bitmap_max);
  86                 t->io_bitmap_max = 0;
  87                 put_cpu();
  88                 kfree(bp);
  89         }
  90 }
  91
  92 void flush_thread(void)
  93 {
  94         struct task_struct *tsk = current;
  95
  96 #ifdef CONFIG_X86_64
  97         if (test_tsk_thread_flag(tsk, TIF_ABI_PENDING)) {
  98                 clear_tsk_thread_flag(tsk, TIF_ABI_PENDING);
  99                 if (test_tsk_thread_flag(tsk, TIF_IA32)) {
 100                         clear_tsk_thread_flag(tsk, TIF_IA32);
 101                 } else {
 102                         set_tsk_thread_flag(tsk, TIF_IA32);
 103                         current_thread_info()->status |= TS_COMPAT;
 104                 }
 105         }
 106 #endif
 107
 108         flush_ptrace_hw_breakpoint(tsk);
 109         memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
 110         /*
 111          * Forget coprocessor state..
 112          */
 113         tsk->fpu_counter = 0;
 114         clear_fpu(tsk);
 115         clear_used_math();
 116 }
 117
 118 static void hard_disable_TSC(void)
 119 {
 120         write_cr4(read_cr4() | X86_CR4_TSD);
 121 }
 122
 123 void disable_TSC(void)
 124 {
 125         preempt_disable();
 126         if (!test_and_set_thread_flag(TIF_NOTSC))
 127                 /*
 128                  * Must flip the CPU state synchronously with
 129                  * TIF_NOTSC in the current running context.
 130                  */
 131                 hard_disable_TSC();
 132         preempt_enable();
 133 }
 134
 135 static void hard_enable_TSC(void)
 136 {
 137         write_cr4(read_cr4() & ~X86_CR4_TSD);
 138 }
 139
 140 static void enable_TSC(void)
 141 {
 142         preempt_disable();
 143         if (test_and_clear_thread_flag(TIF_NOTSC))
 144                 /*
 145                  * Must flip the CPU state synchronously with
 146                  * TIF_NOTSC in the current running context.
 147                  */
 148                 hard_enable_TSC();
 149         preempt_enable();
 150 }
 151
 152 int get_tsc_mode(unsigned long adr)
 153 {
 154         unsigned int val;
 155
 156         if (test_thread_flag(TIF_NOTSC))
 157                 val = PR_TSC_SIGSEGV;
 158         else
 159                 val = PR_TSC_ENABLE;
 160
 161         return put_user(val, (unsigned int __user *)adr);
 162 }
 163
 164 int set_tsc_mode(unsigned int val)
 165 {
 166         if (val == PR_TSC_SIGSEGV)
 167                 disable_TSC();
 168         else if (val == PR_TSC_ENABLE)
 169                 enable_TSC();
 170         else
 171                 return -EINVAL;
 172
 173         return 0;
 174 }
 175
 176 void __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p,
 177                       struct tss_struct *tss)
 178 {
 179         struct thread_struct *prev, *next;
 180
 181         prev = &prev_p->thread;
 182         next = &next_p->thread;
 183
 184         if (test_tsk_thread_flag(next_p, TIF_DS_AREA_MSR) ||
 185             test_tsk_thread_flag(prev_p, TIF_DS_AREA_MSR))
 186                 ds_switch_to(prev_p, next_p);
 187         else if (next->debugctlmsr != prev->debugctlmsr)
 188                 update_debugctlmsr(next->debugctlmsr);
 189
 190         if (test_tsk_thread_flag(prev_p, TIF_NOTSC) ^
 191             test_tsk_thread_flag(next_p, TIF_NOTSC)) {
 192                 /* prev and next are different */
 193                 if (test_tsk_thread_flag(next_p, TIF_NOTSC))
 194                         hard_disable_TSC();
 195                 else
 196                         hard_enable_TSC();
 197         }
 198
 199         if (test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
 200                 /*
 201                  * Copy the relevant range of the IO bitmap.
 202                  * Normally this is 128 bytes or less:
 203                  */
 204                 memcpy(tss->io_bitmap, next->io_bitmap_ptr,
 205                        max(prev->io_bitmap_max, next->io_bitmap_max));
 206         } else if (test_tsk_thread_flag(prev_p, TIF_IO_BITMAP)) {
 207                 /*
 208                  * Clear any possible leftover bits:
 209                  */
 210                 memset(tss->io_bitmap, 0xff, prev->io_bitmap_max);
 211         }
 212 }
 213
 214 int sys_fork(struct pt_regs *regs)
 215 {
 216         return do_fork(SIGCHLD, regs->sp, regs, 0, NULL, NULL);
 217 }
 218
 219 /*
 220  * This is trivial, and on the face of it looks like it
 221  * could equally well be done in user mode.
 222  *
 223  * Not so, for quite unobvious reasons - register pressure.
 224  * In user mode vfork() cannot have a stack frame, and if
 225  * done by calling the "clone()" system call directly, you
 226  * do not have enough call-clobbered registers to hold all
 227  * the information you need.
 228  */
 229 int sys_vfork(struct pt_regs *regs)
 230 {
 231         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->sp, regs, 0,
 232                        NULL, NULL);
 233 }
 234
 235
 236 /*
 237  * Idle related variables and functions
 238  */
 239 unsigned long boot_option_idle_override = 0;
 240 EXPORT_SYMBOL(boot_option_idle_override);
 241
 242 /*
 243  * Powermanagement idle function, if any..
 244  */
 245 void (*pm_idle)(void);
 246 EXPORT_SYMBOL(pm_idle);
 247
 248 #ifdef CONFIG_X86_32
 249 /*
 250  * This halt magic was a workaround for ancient floppy DMA
 251  * wreckage. It should be safe to remove.
 252  */
 253 static int hlt_counter;
 254 void disable_hlt(void)
 255 {
 256         hlt_counter++;
 257 }
 258 EXPORT_SYMBOL(disable_hlt);
 259
 260 void enable_hlt(void)
 261 {
 262         hlt_counter--;
 263 }
 264 EXPORT_SYMBOL(enable_hlt);
 265
 266 static inline int hlt_use_halt(void)
 267 {
 268         return (!hlt_counter && boot_cpu_data.hlt_works_ok);
 269 }
 270 #else
 271 static inline int hlt_use_halt(void)
 272 {
 273         return 1;
 274 }
 275 #endif
 276
 277 /*
 278  * We use this if we don't have any better
 279  * idle routine..
 280  */
 281 void default_idle(void)
 282 {
 283         if (hlt_use_halt()) {
 284                 trace_power_start(POWER_CSTATE, 1);
 285                 current_thread_info()->status &= ~TS_POLLING;
 286                 /*
 287                  * TS_POLLING-cleared state must be visible before we
 288                  * test NEED_RESCHED:
 289                  */
 290                 smp_mb();
 291
 292                 if (!need_resched())
 293                         safe_halt();    /* enables interrupts racelessly */
 294                 else
 295                         local_irq_enable();
 296                 current_thread_info()->status |= TS_POLLING;
 297         } else {
 298                 local_irq_enable();
 299                 /* loop is done by the caller */
 300                 cpu_relax();
 301         }
 302 }
 303 #ifdef CONFIG_APM_MODULE
 304 EXPORT_SYMBOL(default_idle);
 305 #endif
 306
 307 void stop_this_cpu(void *dummy)
 308 {
 309         local_irq_disable();
 310         /*
 311          * Remove this CPU:
 312          */
 313         set_cpu_online(smp_processor_id(), false);
 314         disable_local_APIC();
 315
 316         for (;;) {
 317                 if (hlt_works(smp_processor_id()))
 318                         halt();
 319         }
 320 }
 321
 322 static void do_nothing(void *unused)
 323 {
 324 }
 325
 326 /*
 327  * cpu_idle_wait - Used to ensure that all the CPUs discard old value of
 328  * pm_idle and update to new pm_idle value. Required while changing pm_idle
 329  * handler on SMP systems.
 330  *
 331  * Caller must have changed pm_idle to the new value before the call. Old
 332  * pm_idle value will not be used by any CPU after the return of this function.
 333  */
 334 void cpu_idle_wait(void)
 335 {
 336         smp_mb();
 337         /* kick all the CPUs so that they exit out of pm_idle */
 338         smp_call_function(do_nothing, NULL, 1);
 339 }
 340 EXPORT_SYMBOL_GPL(cpu_idle_wait);
 341
 342 /*
 343  * This uses new MONITOR/MWAIT instructions on P4 processors with PNI,
 344  * which can obviate IPI to trigger checking of need_resched.
 345  * We execute MONITOR against need_resched and enter optimized wait state
 346  * through MWAIT. Whenever someone changes need_resched, we would be woken
 347  * up from MWAIT (without an IPI).
 348  *
 349  * New with Core Duo processors, MWAIT can take some hints based on CPU
 350  * capability.
 351  */
 352 void mwait_idle_with_hints(unsigned long ax, unsigned long cx)
 353 {
 354         trace_power_start(POWER_CSTATE, (ax>>4)+1);
 355         if (!need_resched()) {
 356                 if (cpu_has(&current_cpu_data, X86_FEATURE_CLFLUSH_MONITOR))
 357                         clflush((void *)&current_thread_info()->flags);
 358
 359                 __monitor((void *)&current_thread_info()->flags, 0, 0);
 360                 smp_mb();
 361                 if (!need_resched())
 362                         __mwait(ax, cx);
 363         }
 364 }
 365
 366 /* Default MONITOR/MWAIT with no hints, used for default C1 state */
 367 static void mwait_idle(void)
 368 {
 369         if (!need_resched()) {
 370                 trace_power_start(POWER_CSTATE, 1);
 371                 if (cpu_has(&current_cpu_data, X86_FEATURE_CLFLUSH_MONITOR))
 372                         clflush((void *)&current_thread_info()->flags);
 373
 374                 __monitor((void *)&current_thread_info()->flags, 0, 0);
 375                 smp_mb();
 376                 if (!need_resched())
 377                         __sti_mwait(0, 0);
 378                 else
 379                         local_irq_enable();
 380         } else
 381                 local_irq_enable();
 382 }
 383
 384 /*
 385  * On SMP it's slightly faster (but much more power-consuming!)
 386  * to poll the ->work.need_resched flag instead of waiting for the
 387  * cross-CPU IPI to arrive. Use this option with caution.
 388  */
 389 static void poll_idle(void)
 390 {
 391         trace_power_start(POWER_CSTATE, 0);
 392         local_irq_enable();
 393         while (!need_resched())
 394                 cpu_relax();
 395         trace_power_end(0);
 396 }
 397
 398 /*
 399  * mwait selection logic:
 400  *
 401  * It depends on the CPU. For AMD CPUs that support MWAIT this is
 402  * wrong. Family 0x10 and 0x11 CPUs will enter C1 on HLT. Powersavings
 403  * then depend on a clock divisor and current Pstate of the core. If
 404  * all cores of a processor are in halt state (C1) the processor can
 405  * enter the C1E (C1 enhanced) state. If mwait is used this will never
 406  * happen.
 407  *
 408  * idle=mwait overrides this decision and forces the usage of mwait.
 409  */
 410 static int __cpuinitdata force_mwait;
 411
 412 #define MWAIT_INFO                      0x05
 413 #define MWAIT_ECX_EXTENDED_INFO         0x01
 414 #define MWAIT_EDX_C1                    0xf0
 415
 416 static int __cpuinit mwait_usable(const struct cpuinfo_x86 *c)
 417 {
 418         u32 eax, ebx, ecx, edx;
 419
 420         if (force_mwait)
 421                 return 1;
 422
 423         if (c->cpuid_level < MWAIT_INFO)
 424                 return 0;
 425
 426         cpuid(MWAIT_INFO, &eax, &ebx, &ecx, &edx);
 427         /* Check, whether EDX has extended info about MWAIT */
 428         if (!(ecx & MWAIT_ECX_EXTENDED_INFO))
 429                 return 1;
 430
 431         /*
 432          * edx enumeratios MONITOR/MWAIT extensions. Check, whether
 433          * C1  supports MWAIT
 434          */
 435         return (edx & MWAIT_EDX_C1);
 436 }
 437
 438 /*
 439  * Check for AMD CPUs, which have potentially C1E support
 440  */
 441 static int __cpuinit check_c1e_idle(const struct cpuinfo_x86 *c)
 442 {
 443         if (c->x86_vendor != X86_VENDOR_AMD)
 444                 return 0;
 445
 446         if (c->x86 < 0x0F)
 447                 return 0;
 448
 449         /* Family 0x0f models < rev F do not have C1E */
 450         if (c->x86 == 0x0f && c->x86_model < 0x40)
 451                 return 0;
 452
 453         return 1;
 454 }
 455
 456 static cpumask_var_t c1e_mask;
 457 static int c1e_detected;
 458
 459 void c1e_remove_cpu(int cpu)
 460 {
 461         if (c1e_mask != NULL)
 462                 cpumask_clear_cpu(cpu, c1e_mask);
 463 }
 464
 465 /*
 466  * C1E aware idle routine. We check for C1E active in the interrupt
 467  * pending message MSR. If we detect C1E, then we handle it the same
 468  * way as C3 power states (local apic timer and TSC stop)
 469  */
 470 static void c1e_idle(void)
 471 {
 472         if (need_resched())
 473                 return;
 474
 475         if (!c1e_detected) {
 476                 u32 lo, hi;
 477
 478                 rdmsr(MSR_K8_INT_PENDING_MSG, lo, hi);
 479                 if (lo & K8_INTP_C1E_ACTIVE_MASK) {
 480                         c1e_detected = 1;
 481                         if (!boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
 482                                 mark_tsc_unstable("TSC halt in AMD C1E");
 483                         printk(KERN_INFO "System has AMD C1E enabled\n");
 484                         set_cpu_cap(&boot_cpu_data, X86_FEATURE_AMDC1E);
 485                 }
 486         }
 487
 488         if (c1e_detected) {
 489                 int cpu = smp_processor_id();
 490
 491                 if (!cpumask_test_cpu(cpu, c1e_mask)) {
 492                         cpumask_set_cpu(cpu, c1e_mask);
 493                         /*
 494                          * Force broadcast so ACPI can not interfere.
 495                          */
 496                         clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_FORCE,
 497                                            &cpu);
 498                         printk(KERN_INFO "Switch to broadcast mode on CPU%d\n",
 499                                cpu);
 500                 }
 501                 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER, &cpu);
 502
 503                 default_idle();
 504
 505                 /*
 506                  * The switch back from broadcast mode needs to be
 507                  * called with interrupts disabled.
 508                  */
 509                  local_irq_disable();
 510                  clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_EXIT, &cpu);
 511                  local_irq_enable();
 512         } else
 513                 default_idle();
 514 }
 515
 516 void __cpuinit select_idle_routine(const struct cpuinfo_x86 *c)
 517 {
 518 #ifdef CONFIG_SMP
 519         if (pm_idle == poll_idle && smp_num_siblings > 1) {
 520                 printk(KERN_WARNING "WARNING: polling idle and HT enabled,"
 521                         " performance may degrade.\n");
 522         }
 523 #endif
 524         if (pm_idle)
 525                 return;
 526
 527         if (cpu_has(c, X86_FEATURE_MWAIT) && mwait_usable(c)) {
 528                 /*
 529                  * One CPU supports mwait => All CPUs supports mwait
 530                  */
 531                 printk(KERN_INFO "using mwait in idle threads.\n");
 532                 pm_idle = mwait_idle;
 533         } else if (check_c1e_idle(c)) {
 534                 printk(KERN_INFO "using C1E aware idle routine\n");
 535                 pm_idle = c1e_idle;
 536         } else
 537                 pm_idle = default_idle;
 538 }
 539
 540 void __init init_c1e_mask(void)
 541 {
 542         /* If we're using c1e_idle, we need to allocate c1e_mask. */
 543         if (pm_idle == c1e_idle)
 544                 zalloc_cpumask_var(&c1e_mask, GFP_KERNEL);
 545 }
 546
 547 static int __init idle_setup(char *str)
 548 {
 549         if (!str)
 550                 return -EINVAL;
 551
 552         if (!strcmp(str, "poll")) {
 553                 printk("using polling idle threads.\n");
 554                 pm_idle = poll_idle;
 555         } else if (!strcmp(str, "mwait"))
 556                 force_mwait = 1;
 557         else if (!strcmp(str, "halt")) {
 558                 /*
 559                  * When the boot option of idle=halt is added, halt is
 560                  * forced to be used for CPU idle. In such case CPU C2/C3
 561                  * won't be used again.
 562                  * To continue to load the CPU idle driver, don't touch
 563                  * the boot_option_idle_override.
 564                  */
 565                 pm_idle = default_idle;
 566                 idle_halt = 1;
 567                 return 0;
 568         } else if (!strcmp(str, "nomwait")) {
 569                 /*
 570                  * If the boot option of "idle=nomwait" is added,
 571                  * it means that mwait will be disabled for CPU C2/C3
 572                  * states. In such case it won't touch the variable
 573                  * of boot_option_idle_override.
 574                  */
 575                 idle_nomwait = 1;
 576                 return 0;
 577         } else
 578                 return -1;
 579
 580         boot_option_idle_override = 1;
 581         return 0;
 582 }
 583 early_param("idle", idle_setup);
 584
 585 unsigned long arch_align_stack(unsigned long sp)
 586 {
 587         if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
 588                 sp -= get_random_int() % 8192;
 589         return sp & ~0xf;
 590 }
 591
 592 unsigned long arch_randomize_brk(struct mm_struct *mm)
 593 {
 594         unsigned long range_end = mm->brk + 0x02000000;
 595         return randomize_range(mm->brk, range_end, 0) ? : mm->brk;
 596 }
 597