arch/x86/kernel/process.c

   1 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
   2
   3 #include <linux/errno.h>
   4 #include <linux/kernel.h>
   5 #include <linux/mm.h>
   6 #include <linux/smp.h>
   7 #include <linux/prctl.h>
   8 #include <linux/slab.h>
   9 #include <linux/sched.h>
  10 #include <linux/module.h>
  11 #include <linux/pm.h>
  12 #include <linux/tick.h>
  13 #include <linux/random.h>
  14 #include <linux/user-return-notifier.h>
  15 #include <linux/dmi.h>
  16 #include <linux/utsname.h>
  17 #include <linux/stackprotector.h>
  18 #include <linux/tick.h>
  19 #include <linux/cpuidle.h>
  20 #include <trace/events/power.h>
  21 #include <linux/hw_breakpoint.h>
  22 #include <asm/cpu.h>
  23 #include <asm/apic.h>
  24 #include <asm/syscalls.h>
  25 #include <asm/idle.h>
  26 #include <asm/uaccess.h>
  27 #include <asm/mwait.h>
  28 #include <asm/fpu/internal.h>
  29 #include <asm/debugreg.h>
  30 #include <asm/nmi.h>
  31 #include <asm/tlbflush.h>
  32 #include <asm/mce.h>
  33 #include <asm/vm86.h>
  34
  35 /*
  36  * per-CPU TSS segments. Threads are completely 'soft' on Linux,
  37  * no more per-task TSS's. The TSS size is kept cacheline-aligned
  38  * so they are allowed to end up in the .data..cacheline_aligned
  39  * section. Since TSS's are completely CPU-local, we want them
  40  * on exact cacheline boundaries, to eliminate cacheline ping-pong.
  41  */
  42 __visible DEFINE_PER_CPU_SHARED_ALIGNED(struct tss_struct, cpu_tss) = {
  43         .x86_tss = {
  44                 .sp0 = TOP_OF_INIT_STACK,
  45 #ifdef CONFIG_X86_32
  46                 .ss0 = __KERNEL_DS,
  47                 .ss1 = __KERNEL_CS,
  48                 .io_bitmap_base = INVALID_IO_BITMAP_OFFSET,
  49 #endif
  50          },
  51 #ifdef CONFIG_X86_32
  52          /*
  53           * Note that the .io_bitmap member must be extra-big. This is because
  54           * the CPU will access an additional byte beyond the end of the IO
  55           * permission bitmap. The extra byte must be all 1 bits, and must
  56           * be within the limit.
  57           */
  58         .io_bitmap              = { [0 ... IO_BITMAP_LONGS] = ~0 },
  59 #endif
  60 };
  61 EXPORT_PER_CPU_SYMBOL(cpu_tss);
  62
  63 #ifdef CONFIG_X86_64
  64 static DEFINE_PER_CPU(unsigned char, is_idle);
  65 static ATOMIC_NOTIFIER_HEAD(idle_notifier);
  66
  67 void idle_notifier_register(struct notifier_block *n)
  68 {
  69         atomic_notifier_chain_register(&idle_notifier, n);
  70 }
  71 EXPORT_SYMBOL_GPL(idle_notifier_register);
  72
  73 void idle_notifier_unregister(struct notifier_block *n)
  74 {
  75         atomic_notifier_chain_unregister(&idle_notifier, n);
  76 }
  77 EXPORT_SYMBOL_GPL(idle_notifier_unregister);
  78 #endif
  79
  80 /*
  81  * this gets called so that we can store lazy state into memory and copy the
  82  * current task into the new thread.
  83  */
  84 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
  85 {
  86         memcpy(dst, src, arch_task_struct_size);
  87
  88         return fpu__copy(&dst->thread.fpu, &src->thread.fpu);
  89 }
  90
  91 /*
  92  * Free current thread data structures etc..
  93  */
  94 void exit_thread(void)
  95 {
  96         struct task_struct *me = current;
  97         struct thread_struct *t = &me->thread;
  98         unsigned long *bp = t->io_bitmap_ptr;
  99         struct fpu *fpu = &t->fpu;
 100
 101         if (bp) {
 102                 struct tss_struct *tss = &per_cpu(cpu_tss, get_cpu());
 103
 104                 t->io_bitmap_ptr = NULL;
 105                 clear_thread_flag(TIF_IO_BITMAP);
 106                 /*
 107                  * Careful, clear this in the TSS too:
 108                  */
 109                 memset(tss->io_bitmap, 0xff, t->io_bitmap_max);
 110                 t->io_bitmap_max = 0;
 111                 put_cpu();
 112                 kfree(bp);
 113         }
 114
 115         free_vm86(t);
 116
 117         fpu__drop(fpu);
 118 }
 119
 120 void flush_thread(void)
 121 {
 122         struct task_struct *tsk = current;
 123
 124         flush_ptrace_hw_breakpoint(tsk);
 125         memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
 126
 127         fpu__clear(&tsk->thread.fpu);
 128 }
 129
 130 static void hard_disable_TSC(void)
 131 {
 132         cr4_set_bits(X86_CR4_TSD);
 133 }
 134
 135 void disable_TSC(void)
 136 {
 137         preempt_disable();
 138         if (!test_and_set_thread_flag(TIF_NOTSC))
 139                 /*
 140                  * Must flip the CPU state synchronously with
 141                  * TIF_NOTSC in the current running context.
 142                  */
 143                 hard_disable_TSC();
 144         preempt_enable();
 145 }
 146
 147 static void hard_enable_TSC(void)
 148 {
 149         cr4_clear_bits(X86_CR4_TSD);
 150 }
 151
 152 static void enable_TSC(void)
 153 {
 154         preempt_disable();
 155         if (test_and_clear_thread_flag(TIF_NOTSC))
 156                 /*
 157                  * Must flip the CPU state synchronously with
 158                  * TIF_NOTSC in the current running context.
 159                  */
 160                 hard_enable_TSC();
 161         preempt_enable();
 162 }
 163
 164 int get_tsc_mode(unsigned long adr)
 165 {
 166         unsigned int val;
 167
 168         if (test_thread_flag(TIF_NOTSC))
 169                 val = PR_TSC_SIGSEGV;
 170         else
 171                 val = PR_TSC_ENABLE;
 172
 173         return put_user(val, (unsigned int __user *)adr);
 174 }
 175
 176 int set_tsc_mode(unsigned int val)
 177 {
 178         if (val == PR_TSC_SIGSEGV)
 179                 disable_TSC();
 180         else if (val == PR_TSC_ENABLE)
 181                 enable_TSC();
 182         else
 183                 return -EINVAL;
 184
 185         return 0;
 186 }
 187
 188 void __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p,
 189                       struct tss_struct *tss)
 190 {
 191         struct thread_struct *prev, *next;
 192
 193         prev = &prev_p->thread;
 194         next = &next_p->thread;
 195
 196         if (test_tsk_thread_flag(prev_p, TIF_BLOCKSTEP) ^
 197             test_tsk_thread_flag(next_p, TIF_BLOCKSTEP)) {
 198                 unsigned long debugctl = get_debugctlmsr();
 199
 200                 debugctl &= ~DEBUGCTLMSR_BTF;
 201                 if (test_tsk_thread_flag(next_p, TIF_BLOCKSTEP))
 202                         debugctl |= DEBUGCTLMSR_BTF;
 203
 204                 update_debugctlmsr(debugctl);
 205         }
 206
 207         if (test_tsk_thread_flag(prev_p, TIF_NOTSC) ^
 208             test_tsk_thread_flag(next_p, TIF_NOTSC)) {
 209                 /* prev and next are different */
 210                 if (test_tsk_thread_flag(next_p, TIF_NOTSC))
 211                         hard_disable_TSC();
 212                 else
 213                         hard_enable_TSC();
 214         }
 215
 216         if (test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
 217                 /*
 218                  * Copy the relevant range of the IO bitmap.
 219                  * Normally this is 128 bytes or less:
 220                  */
 221                 memcpy(tss->io_bitmap, next->io_bitmap_ptr,
 222                        max(prev->io_bitmap_max, next->io_bitmap_max));
 223         } else if (test_tsk_thread_flag(prev_p, TIF_IO_BITMAP)) {
 224                 /*
 225                  * Clear any possible leftover bits:
 226                  */
 227                 memset(tss->io_bitmap, 0xff, prev->io_bitmap_max);
 228         }
 229         propagate_user_return_notify(prev_p, next_p);
 230 }
 231
 232 /*
 233  * Idle related variables and functions
 234  */
 235 unsigned long boot_option_idle_override = IDLE_NO_OVERRIDE;
 236 EXPORT_SYMBOL(boot_option_idle_override);
 237
 238 static void (*x86_idle)(void);
 239
 240 #ifndef CONFIG_SMP
 241 static inline void play_dead(void)
 242 {
 243         BUG();
 244 }
 245 #endif
 246
 247 #ifdef CONFIG_X86_64
 248 void enter_idle(void)
 249 {
 250         this_cpu_write(is_idle, 1);
 251         atomic_notifier_call_chain(&idle_notifier, IDLE_START, NULL);
 252 }
 253
 254 static void __exit_idle(void)
 255 {
 256         if (x86_test_and_clear_bit_percpu(0, is_idle) == 0)
 257                 return;
 258         atomic_notifier_call_chain(&idle_notifier, IDLE_END, NULL);
 259 }
 260
 261 /* Called from interrupts to signify idle end */
 262 void exit_idle(void)
 263 {
 264         /* idle loop has pid 0 */
 265         if (current->pid)
 266                 return;
 267         __exit_idle();
 268 }
 269 #endif
 270
 271 void arch_cpu_idle_enter(void)
 272 {
 273         local_touch_nmi();
 274         enter_idle();
 275 }
 276
 277 void arch_cpu_idle_exit(void)
 278 {
 279         __exit_idle();
 280 }
 281
 282 void arch_cpu_idle_dead(void)
 283 {
 284         play_dead();
 285 }
 286
 287 /*
 288  * Called from the generic idle code.
 289  */
 290 void arch_cpu_idle(void)
 291 {
 292         x86_idle();
 293 }
 294
 295 /*
 296  * We use this if we don't have any better idle routine..
 297  */
 298 void default_idle(void)
 299 {
 300         trace_cpu_idle_rcuidle(1, smp_processor_id());
 301         safe_halt();
 302         trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, smp_processor_id());
 303 }
 304 #ifdef CONFIG_APM_MODULE
 305 EXPORT_SYMBOL(default_idle);
 306 #endif
 307
 308 #ifdef CONFIG_XEN
 309 bool xen_set_default_idle(void)
 310 {
 311         bool ret = !!x86_idle;
 312
 313         x86_idle = default_idle;
 314
 315         return ret;
 316 }
 317 #endif
 318 void stop_this_cpu(void *dummy)
 319 {
 320         local_irq_disable();
 321         /*
 322          * Remove this CPU:
 323          */
 324         set_cpu_online(smp_processor_id(), false);
 325         disable_local_APIC();
 326         mcheck_cpu_clear(this_cpu_ptr(&cpu_info));
 327
 328         for (;;)
 329                 halt();
 330 }
 331
 332 bool amd_e400_c1e_detected;
 333 EXPORT_SYMBOL(amd_e400_c1e_detected);
 334
 335 static cpumask_var_t amd_e400_c1e_mask;
 336
 337 void amd_e400_remove_cpu(int cpu)
 338 {
 339         if (amd_e400_c1e_mask != NULL)
 340                 cpumask_clear_cpu(cpu, amd_e400_c1e_mask);
 341 }
 342
 343 /*
 344  * AMD Erratum 400 aware idle routine. We check for C1E active in the interrupt
 345  * pending message MSR. If we detect C1E, then we handle it the same
 346  * way as C3 power states (local apic timer and TSC stop)
 347  */
 348 static void amd_e400_idle(void)
 349 {
 350         if (!amd_e400_c1e_detected) {
 351                 u32 lo, hi;
 352
 353                 rdmsr(MSR_K8_INT_PENDING_MSG, lo, hi);
 354
 355                 if (lo & K8_INTP_C1E_ACTIVE_MASK) {
 356                         amd_e400_c1e_detected = true;
 357                         if (!boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
 358                                 mark_tsc_unstable("TSC halt in AMD C1E");
 359                         pr_info("System has AMD C1E enabled\n");
 360                 }
 361         }
 362
 363         if (amd_e400_c1e_detected) {
 364                 int cpu = smp_processor_id();
 365
 366                 if (!cpumask_test_cpu(cpu, amd_e400_c1e_mask)) {
 367                         cpumask_set_cpu(cpu, amd_e400_c1e_mask);
 368                         /* Force broadcast so ACPI can not interfere. */
 369                         tick_broadcast_force();
 370                         pr_info("Switch to broadcast mode on CPU%d\n", cpu);
 371                 }
 372                 tick_broadcast_enter();
 373
 374                 default_idle();
 375
 376                 /*
 377                  * The switch back from broadcast mode needs to be
 378                  * called with interrupts disabled.
 379                  */
 380                 local_irq_disable();
 381                 tick_broadcast_exit();
 382                 local_irq_enable();
 383         } else
 384                 default_idle();
 385 }
 386
 387 /*
 388  * Intel Core2 and older machines prefer MWAIT over HALT for C1.
 389  * We can't rely on cpuidle installing MWAIT, because it will not load
 390  * on systems that support only C1 -- so the boot default must be MWAIT.
 391  *
 392  * Some AMD machines are the opposite, they depend on using HALT.
 393  *
 394  * So for default C1, which is used during boot until cpuidle loads,
 395  * use MWAIT-C1 on Intel HW that has it, else use HALT.
 396  */
 397 static int prefer_mwait_c1_over_halt(const struct cpuinfo_x86 *c)
 398 {
 399         if (c->x86_vendor != X86_VENDOR_INTEL)
 400                 return 0;
 401
 402         if (!cpu_has(c, X86_FEATURE_MWAIT))
 403                 return 0;
 404
 405         return 1;
 406 }
 407
 408 /*
 409  * MONITOR/MWAIT with no hints, used for default C1 state. This invokes MWAIT
 410  * with interrupts enabled and no flags, which is backwards compatible with the
 411  * original MWAIT implementation.
 412  */
 413 static void mwait_idle(void)
 414 {
 415         if (!current_set_polling_and_test()) {
 416                 trace_cpu_idle_rcuidle(1, smp_processor_id());
 417                 if (this_cpu_has(X86_BUG_CLFLUSH_MONITOR)) {
 418                         smp_mb(); /* quirk */
 419                         clflush((void *)&current_thread_info()->flags);
 420                         smp_mb(); /* quirk */
 421                 }
 422
 423                 __monitor((void *)&current_thread_info()->flags, 0, 0);
 424                 if (!need_resched())
 425                         __sti_mwait(0, 0);
 426                 else
 427                         local_irq_enable();
 428                 trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, smp_processor_id());
 429         } else {
 430                 local_irq_enable();
 431         }
 432         __current_clr_polling();
 433 }
 434
 435 void select_idle_routine(const struct cpuinfo_x86 *c)
 436 {
 437 #ifdef CONFIG_SMP
 438         if (boot_option_idle_override == IDLE_POLL && smp_num_siblings > 1)
 439                 pr_warn_once("WARNING: polling idle and HT enabled, performance may degrade\n");
 440 #endif
 441         if (x86_idle || boot_option_idle_override == IDLE_POLL)
 442                 return;
 443
 444         if (cpu_has_bug(c, X86_BUG_AMD_APIC_C1E)) {
 445                 /* E400: APIC timer interrupt does not wake up CPU from C1e */
 446                 pr_info("using AMD E400 aware idle routine\n");
 447                 x86_idle = amd_e400_idle;
 448         } else if (prefer_mwait_c1_over_halt(c)) {
 449                 pr_info("using mwait in idle threads\n");
 450                 x86_idle = mwait_idle;
 451         } else
 452                 x86_idle = default_idle;
 453 }
 454
 455 void __init init_amd_e400_c1e_mask(void)
 456 {
 457         /* If we're using amd_e400_idle, we need to allocate amd_e400_c1e_mask. */
 458         if (x86_idle == amd_e400_idle)
 459                 zalloc_cpumask_var(&amd_e400_c1e_mask, GFP_KERNEL);
 460 }
 461
 462 static int __init idle_setup(char *str)
 463 {
 464         if (!str)
 465                 return -EINVAL;
 466
 467         if (!strcmp(str, "poll")) {
 468                 pr_info("using polling idle threads\n");
 469                 boot_option_idle_override = IDLE_POLL;
 470                 cpu_idle_poll_ctrl(true);
 471         } else if (!strcmp(str, "halt")) {
 472                 /*
 473                  * When the boot option of idle=halt is added, halt is
 474                  * forced to be used for CPU idle. In such case CPU C2/C3
 475                  * won't be used again.
 476                  * To continue to load the CPU idle driver, don't touch
 477                  * the boot_option_idle_override.
 478                  */
 479                 x86_idle = default_idle;
 480                 boot_option_idle_override = IDLE_HALT;
 481         } else if (!strcmp(str, "nomwait")) {
 482                 /*
 483                  * If the boot option of "idle=nomwait" is added,
 484                  * it means that mwait will be disabled for CPU C2/C3
 485                  * states. In such case it won't touch the variable
 486                  * of boot_option_idle_override.
 487                  */
 488                 boot_option_idle_override = IDLE_NOMWAIT;
 489         } else
 490                 return -1;
 491
 492         return 0;
 493 }
 494 early_param("idle", idle_setup);
 495
 496 unsigned long arch_align_stack(unsigned long sp)
 497 {
 498         if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
 499                 sp -= get_random_int() % 8192;
 500         return sp & ~0xf;
 501 }
 502
 503 unsigned long arch_randomize_brk(struct mm_struct *mm)
 504 {
 505         unsigned long range_end = mm->brk + 0x02000000;
 506         return randomize_range(mm->brk, range_end, 0) ? : mm->brk;
 507 }
 508
 509 /*
 510  * Called from fs/proc with a reference on @p to find the function
 511  * which called into schedule(). This needs to be done carefully
 512  * because the task might wake up and we might look at a stack
 513  * changing under us.
 514  */
 515 unsigned long get_wchan(struct task_struct *p)
 516 {
 517         unsigned long start, bottom, top, sp, fp, ip;
 518         int count = 0;
 519
 520         if (!p || p == current || p->state == TASK_RUNNING)
 521                 return 0;
 522
 523         start = (unsigned long)task_stack_page(p);
 524         if (!start)
 525                 return 0;
 526
 527         /*
 528          * Layout of the stack page:
 529          *
 530          * ----------- topmax = start + THREAD_SIZE - sizeof(unsigned long)
 531          * PADDING
 532          * ----------- top = topmax - TOP_OF_KERNEL_STACK_PADDING
 533          * stack
 534          * ----------- bottom = start + sizeof(thread_info)
 535          * thread_info
 536          * ----------- start
 537          *
 538          * The tasks stack pointer points at the location where the
 539          * framepointer is stored. The data on the stack is:
 540          * ... IP FP ... IP FP
 541          *
 542          * We need to read FP and IP, so we need to adjust the upper
 543          * bound by another unsigned long.
 544          */
 545         top = start + THREAD_SIZE - TOP_OF_KERNEL_STACK_PADDING;
 546         top -= 2 * sizeof(unsigned long);
 547         bottom = start + sizeof(struct thread_info);
 548
 549         sp = READ_ONCE(p->thread.sp);
 550         if (sp < bottom || sp > top)
 551                 return 0;
 552
 553         fp = READ_ONCE(*(unsigned long *)sp);
 554         do {
 555                 if (fp < bottom || fp > top)
 556                         return 0;
 557                 ip = READ_ONCE(*(unsigned long *)(fp + sizeof(unsigned long)));
 558                 if (!in_sched_functions(ip))
 559                         return ip;
 560                 fp = READ_ONCE(*(unsigned long *)fp);
 561         } while (count++ < 16 && p->state != TASK_RUNNING);
 562         return 0;
 563 }