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/init.h>
  11 #include <linux/export.h>
  12 #include <linux/pm.h>
  13 #include <linux/tick.h>
  14 #include <linux/random.h>
  15 #include <linux/user-return-notifier.h>
  16 #include <linux/dmi.h>
  17 #include <linux/utsname.h>
  18 #include <linux/stackprotector.h>
  19 #include <linux/tick.h>
  20 #include <linux/cpuidle.h>
  21 #include <trace/events/power.h>
  22 #include <linux/hw_breakpoint.h>
  23 #include <asm/cpu.h>
  24 #include <asm/apic.h>
  25 #include <asm/syscalls.h>
  26 #include <asm/idle.h>
  27 #include <asm/uaccess.h>
  28 #include <asm/mwait.h>
  29 #include <asm/fpu/internal.h>
  30 #include <asm/debugreg.h>
  31 #include <asm/nmi.h>
  32 #include <asm/tlbflush.h>
  33 #include <asm/mce.h>
  34 #include <asm/vm86.h>
  35
  36 /*
  37  * per-CPU TSS segments. Threads are completely 'soft' on Linux,
  38  * no more per-task TSS's. The TSS size is kept cacheline-aligned
  39  * so they are allowed to end up in the .data..cacheline_aligned
  40  * section. Since TSS's are completely CPU-local, we want them
  41  * on exact cacheline boundaries, to eliminate cacheline ping-pong.
  42  */
  43 __visible DEFINE_PER_CPU_SHARED_ALIGNED(struct tss_struct, cpu_tss) = {
  44         .x86_tss = {
  45                 .sp0 = TOP_OF_INIT_STACK,
  46 #ifdef CONFIG_X86_32
  47                 .ss0 = __KERNEL_DS,
  48                 .ss1 = __KERNEL_CS,
  49                 .io_bitmap_base = INVALID_IO_BITMAP_OFFSET,
  50 #endif
  51          },
  52 #ifdef CONFIG_X86_32
  53          /*
  54           * Note that the .io_bitmap member must be extra-big. This is because
  55           * the CPU will access an additional byte beyond the end of the IO
  56           * permission bitmap. The extra byte must be all 1 bits, and must
  57           * be within the limit.
  58           */
  59         .io_bitmap              = { [0 ... IO_BITMAP_LONGS] = ~0 },
  60 #endif
  61 #ifdef CONFIG_X86_32
  62         .SYSENTER_stack_canary  = STACK_END_MAGIC,
  63 #endif
  64 };
  65 EXPORT_PER_CPU_SYMBOL(cpu_tss);
  66
  67 #ifdef CONFIG_X86_64
  68 static DEFINE_PER_CPU(unsigned char, is_idle);
  69 static ATOMIC_NOTIFIER_HEAD(idle_notifier);
  70
  71 void idle_notifier_register(struct notifier_block *n)
  72 {
  73         atomic_notifier_chain_register(&idle_notifier, n);
  74 }
  75 EXPORT_SYMBOL_GPL(idle_notifier_register);
  76
  77 void idle_notifier_unregister(struct notifier_block *n)
  78 {
  79         atomic_notifier_chain_unregister(&idle_notifier, n);
  80 }
  81 EXPORT_SYMBOL_GPL(idle_notifier_unregister);
  82 #endif
  83
  84 /*
  85  * this gets called so that we can store lazy state into memory and copy the
  86  * current task into the new thread.
  87  */
  88 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
  89 {
  90         memcpy(dst, src, arch_task_struct_size);
  91 #ifdef CONFIG_VM86
  92         dst->thread.vm86 = NULL;
  93 #endif
  94
  95         return fpu__copy(&dst->thread.fpu, &src->thread.fpu);
  96 }
  97
  98 /*
  99  * Free current thread data structures etc..
 100  */
 101 void exit_thread(struct task_struct *tsk)
 102 {
 103         struct thread_struct *t = &tsk->thread;
 104         unsigned long *bp = t->io_bitmap_ptr;
 105         struct fpu *fpu = &t->fpu;
 106
 107         if (bp) {
 108                 struct tss_struct *tss = &per_cpu(cpu_tss, get_cpu());
 109
 110                 t->io_bitmap_ptr = NULL;
 111                 clear_thread_flag(TIF_IO_BITMAP);
 112                 /*
 113                  * Careful, clear this in the TSS too:
 114                  */
 115                 memset(tss->io_bitmap, 0xff, t->io_bitmap_max);
 116                 t->io_bitmap_max = 0;
 117                 put_cpu();
 118                 kfree(bp);
 119         }
 120
 121         free_vm86(t);
 122
 123         fpu__drop(fpu);
 124 }
 125
 126 void flush_thread(void)
 127 {
 128         struct task_struct *tsk = current;
 129
 130         flush_ptrace_hw_breakpoint(tsk);
 131         memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
 132
 133         fpu__clear(&tsk->thread.fpu);
 134 }
 135
 136 static void hard_disable_TSC(void)
 137 {
 138         cr4_set_bits(X86_CR4_TSD);
 139 }
 140
 141 void disable_TSC(void)
 142 {
 143         preempt_disable();
 144         if (!test_and_set_thread_flag(TIF_NOTSC))
 145                 /*
 146                  * Must flip the CPU state synchronously with
 147                  * TIF_NOTSC in the current running context.
 148                  */
 149                 hard_disable_TSC();
 150         preempt_enable();
 151 }
 152
 153 static void hard_enable_TSC(void)
 154 {
 155         cr4_clear_bits(X86_CR4_TSD);
 156 }
 157
 158 static void enable_TSC(void)
 159 {
 160         preempt_disable();
 161         if (test_and_clear_thread_flag(TIF_NOTSC))
 162                 /*
 163                  * Must flip the CPU state synchronously with
 164                  * TIF_NOTSC in the current running context.
 165                  */
 166                 hard_enable_TSC();
 167         preempt_enable();
 168 }
 169
 170 int get_tsc_mode(unsigned long adr)
 171 {
 172         unsigned int val;
 173
 174         if (test_thread_flag(TIF_NOTSC))
 175                 val = PR_TSC_SIGSEGV;
 176         else
 177                 val = PR_TSC_ENABLE;
 178
 179         return put_user(val, (unsigned int __user *)adr);
 180 }
 181
 182 int set_tsc_mode(unsigned int val)
 183 {
 184         if (val == PR_TSC_SIGSEGV)
 185                 disable_TSC();
 186         else if (val == PR_TSC_ENABLE)
 187                 enable_TSC();
 188         else
 189                 return -EINVAL;
 190
 191         return 0;
 192 }
 193
 194 void __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p,
 195                       struct tss_struct *tss)
 196 {
 197         struct thread_struct *prev, *next;
 198
 199         prev = &prev_p->thread;
 200         next = &next_p->thread;
 201
 202         if (test_tsk_thread_flag(prev_p, TIF_BLOCKSTEP) ^
 203             test_tsk_thread_flag(next_p, TIF_BLOCKSTEP)) {
 204                 unsigned long debugctl = get_debugctlmsr();
 205
 206                 debugctl &= ~DEBUGCTLMSR_BTF;
 207                 if (test_tsk_thread_flag(next_p, TIF_BLOCKSTEP))
 208                         debugctl |= DEBUGCTLMSR_BTF;
 209
 210                 update_debugctlmsr(debugctl);
 211         }
 212
 213         if (test_tsk_thread_flag(prev_p, TIF_NOTSC) ^
 214             test_tsk_thread_flag(next_p, TIF_NOTSC)) {
 215                 /* prev and next are different */
 216                 if (test_tsk_thread_flag(next_p, TIF_NOTSC))
 217                         hard_disable_TSC();
 218                 else
 219                         hard_enable_TSC();
 220         }
 221
 222         if (test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
 223                 /*
 224                  * Copy the relevant range of the IO bitmap.
 225                  * Normally this is 128 bytes or less:
 226                  */
 227                 memcpy(tss->io_bitmap, next->io_bitmap_ptr,
 228                        max(prev->io_bitmap_max, next->io_bitmap_max));
 229         } else if (test_tsk_thread_flag(prev_p, TIF_IO_BITMAP)) {
 230                 /*
 231                  * Clear any possible leftover bits:
 232                  */
 233                 memset(tss->io_bitmap, 0xff, prev->io_bitmap_max);
 234         }
 235         propagate_user_return_notify(prev_p, next_p);
 236 }
 237
 238 /*
 239  * Idle related variables and functions
 240  */
 241 unsigned long boot_option_idle_override = IDLE_NO_OVERRIDE;
 242 EXPORT_SYMBOL(boot_option_idle_override);
 243
 244 static void (*x86_idle)(void);
 245
 246 #ifndef CONFIG_SMP
 247 static inline void play_dead(void)
 248 {
 249         BUG();
 250 }
 251 #endif
 252
 253 #ifdef CONFIG_X86_64
 254 void enter_idle(void)
 255 {
 256         this_cpu_write(is_idle, 1);
 257         atomic_notifier_call_chain(&idle_notifier, IDLE_START, NULL);
 258 }
 259
 260 static void __exit_idle(void)
 261 {
 262         if (x86_test_and_clear_bit_percpu(0, is_idle) == 0)
 263                 return;
 264         atomic_notifier_call_chain(&idle_notifier, IDLE_END, NULL);
 265 }
 266
 267 /* Called from interrupts to signify idle end */
 268 void exit_idle(void)
 269 {
 270         /* idle loop has pid 0 */
 271         if (current->pid)
 272                 return;
 273         __exit_idle();
 274 }
 275 #endif
 276
 277 void arch_cpu_idle_enter(void)
 278 {
 279         local_touch_nmi();
 280         enter_idle();
 281 }
 282
 283 void arch_cpu_idle_exit(void)
 284 {
 285         __exit_idle();
 286 }
 287
 288 void arch_cpu_idle_dead(void)
 289 {
 290         play_dead();
 291 }
 292
 293 /*
 294  * Called from the generic idle code.
 295  */
 296 void arch_cpu_idle(void)
 297 {
 298         x86_idle();
 299 }
 300
 301 /*
 302  * We use this if we don't have any better idle routine..
 303  */
 304 void __cpuidle default_idle(void)
 305 {
 306         trace_cpu_idle_rcuidle(1, smp_processor_id());
 307         safe_halt();
 308         trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, smp_processor_id());
 309 }
 310 #ifdef CONFIG_APM_MODULE
 311 EXPORT_SYMBOL(default_idle);
 312 #endif
 313
 314 #ifdef CONFIG_XEN
 315 bool xen_set_default_idle(void)
 316 {
 317         bool ret = !!x86_idle;
 318
 319         x86_idle = default_idle;
 320
 321         return ret;
 322 }
 323 #endif
 324 void stop_this_cpu(void *dummy)
 325 {
 326         local_irq_disable();
 327         /*
 328          * Remove this CPU:
 329          */
 330         set_cpu_online(smp_processor_id(), false);
 331         disable_local_APIC();
 332         mcheck_cpu_clear(this_cpu_ptr(&cpu_info));
 333
 334         for (;;)
 335                 halt();
 336 }
 337
 338 bool amd_e400_c1e_detected;
 339 EXPORT_SYMBOL(amd_e400_c1e_detected);
 340
 341 static cpumask_var_t amd_e400_c1e_mask;
 342
 343 void amd_e400_remove_cpu(int cpu)
 344 {
 345         if (amd_e400_c1e_mask != NULL)
 346                 cpumask_clear_cpu(cpu, amd_e400_c1e_mask);
 347 }
 348
 349 /*
 350  * AMD Erratum 400 aware idle routine. We check for C1E active in the interrupt
 351  * pending message MSR. If we detect C1E, then we handle it the same
 352  * way as C3 power states (local apic timer and TSC stop)
 353  */
 354 static void amd_e400_idle(void)
 355 {
 356         if (!amd_e400_c1e_detected) {
 357                 u32 lo, hi;
 358
 359                 rdmsr(MSR_K8_INT_PENDING_MSG, lo, hi);
 360
 361                 if (lo & K8_INTP_C1E_ACTIVE_MASK) {
 362                         amd_e400_c1e_detected = true;
 363                         if (!boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
 364                                 mark_tsc_unstable("TSC halt in AMD C1E");
 365                         pr_info("System has AMD C1E enabled\n");
 366                 }
 367         }
 368
 369         if (amd_e400_c1e_detected) {
 370                 int cpu = smp_processor_id();
 371
 372                 if (!cpumask_test_cpu(cpu, amd_e400_c1e_mask)) {
 373                         cpumask_set_cpu(cpu, amd_e400_c1e_mask);
 374                         /* Force broadcast so ACPI can not interfere. */
 375                         tick_broadcast_force();
 376                         pr_info("Switch to broadcast mode on CPU%d\n", cpu);
 377                 }
 378                 tick_broadcast_enter();
 379
 380                 default_idle();
 381
 382                 /*
 383                  * The switch back from broadcast mode needs to be
 384                  * called with interrupts disabled.
 385                  */
 386                 local_irq_disable();
 387                 tick_broadcast_exit();
 388                 local_irq_enable();
 389         } else
 390                 default_idle();
 391 }
 392
 393 /*
 394  * Intel Core2 and older machines prefer MWAIT over HALT for C1.
 395  * We can't rely on cpuidle installing MWAIT, because it will not load
 396  * on systems that support only C1 -- so the boot default must be MWAIT.
 397  *
 398  * Some AMD machines are the opposite, they depend on using HALT.
 399  *
 400  * So for default C1, which is used during boot until cpuidle loads,
 401  * use MWAIT-C1 on Intel HW that has it, else use HALT.
 402  */
 403 static int prefer_mwait_c1_over_halt(const struct cpuinfo_x86 *c)
 404 {
 405         if (c->x86_vendor != X86_VENDOR_INTEL)
 406                 return 0;
 407
 408         if (!cpu_has(c, X86_FEATURE_MWAIT) || static_cpu_has_bug(X86_BUG_MONITOR))
 409                 return 0;
 410
 411         return 1;
 412 }
 413
 414 /*
 415  * MONITOR/MWAIT with no hints, used for default C1 state. This invokes MWAIT
 416  * with interrupts enabled and no flags, which is backwards compatible with the
 417  * original MWAIT implementation.
 418  */
 419 static __cpuidle void mwait_idle(void)
 420 {
 421         if (!current_set_polling_and_test()) {
 422                 trace_cpu_idle_rcuidle(1, smp_processor_id());
 423                 if (this_cpu_has(X86_BUG_CLFLUSH_MONITOR)) {
 424                         mb(); /* quirk */
 425                         clflush((void *)&current_thread_info()->flags);
 426                         mb(); /* quirk */
 427                 }
 428
 429                 __monitor((void *)&current_thread_info()->flags, 0, 0);
 430                 if (!need_resched())
 431                         __sti_mwait(0, 0);
 432                 else
 433                         local_irq_enable();
 434                 trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, smp_processor_id());
 435         } else {
 436                 local_irq_enable();
 437         }
 438         __current_clr_polling();
 439 }
 440
 441 void select_idle_routine(const struct cpuinfo_x86 *c)
 442 {
 443 #ifdef CONFIG_SMP
 444         if (boot_option_idle_override == IDLE_POLL && smp_num_siblings > 1)
 445                 pr_warn_once("WARNING: polling idle and HT enabled, performance may degrade\n");
 446 #endif
 447         if (x86_idle || boot_option_idle_override == IDLE_POLL)
 448                 return;
 449
 450         if (cpu_has_bug(c, X86_BUG_AMD_APIC_C1E)) {
 451                 /* E400: APIC timer interrupt does not wake up CPU from C1e */
 452                 pr_info("using AMD E400 aware idle routine\n");
 453                 x86_idle = amd_e400_idle;
 454         } else if (prefer_mwait_c1_over_halt(c)) {
 455                 pr_info("using mwait in idle threads\n");
 456                 x86_idle = mwait_idle;
 457         } else
 458                 x86_idle = default_idle;
 459 }
 460
 461 void __init init_amd_e400_c1e_mask(void)
 462 {
 463         /* If we're using amd_e400_idle, we need to allocate amd_e400_c1e_mask. */
 464         if (x86_idle == amd_e400_idle)
 465                 zalloc_cpumask_var(&amd_e400_c1e_mask, GFP_KERNEL);
 466 }
 467
 468 static int __init idle_setup(char *str)
 469 {
 470         if (!str)
 471                 return -EINVAL;
 472
 473         if (!strcmp(str, "poll")) {
 474                 pr_info("using polling idle threads\n");
 475                 boot_option_idle_override = IDLE_POLL;
 476                 cpu_idle_poll_ctrl(true);
 477         } else if (!strcmp(str, "halt")) {
 478                 /*
 479                  * When the boot option of idle=halt is added, halt is
 480                  * forced to be used for CPU idle. In such case CPU C2/C3
 481                  * won't be used again.
 482                  * To continue to load the CPU idle driver, don't touch
 483                  * the boot_option_idle_override.
 484                  */
 485                 x86_idle = default_idle;
 486                 boot_option_idle_override = IDLE_HALT;
 487         } else if (!strcmp(str, "nomwait")) {
 488                 /*
 489                  * If the boot option of "idle=nomwait" is added,
 490                  * it means that mwait will be disabled for CPU C2/C3
 491                  * states. In such case it won't touch the variable
 492                  * of boot_option_idle_override.
 493                  */
 494                 boot_option_idle_override = IDLE_NOMWAIT;
 495         } else
 496                 return -1;
 497
 498         return 0;
 499 }
 500 early_param("idle", idle_setup);
 501
 502 unsigned long arch_align_stack(unsigned long sp)
 503 {
 504         if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
 505                 sp -= get_random_int() % 8192;
 506         return sp & ~0xf;
 507 }
 508
 509 unsigned long arch_randomize_brk(struct mm_struct *mm)
 510 {
 511         return randomize_page(mm->brk, 0x02000000);
 512 }
 513
 514 /*
 515  * Called from fs/proc with a reference on @p to find the function
 516  * which called into schedule(). This needs to be done carefully
 517  * because the task might wake up and we might look at a stack
 518  * changing under us.
 519  */
 520 unsigned long get_wchan(struct task_struct *p)
 521 {
 522         unsigned long start, bottom, top, sp, fp, ip;
 523         int count = 0;
 524
 525         if (!p || p == current || p->state == TASK_RUNNING)
 526                 return 0;
 527
 528         start = (unsigned long)task_stack_page(p);
 529         if (!start)
 530                 return 0;
 531
 532         /*
 533          * Layout of the stack page:
 534          *
 535          * ----------- topmax = start + THREAD_SIZE - sizeof(unsigned long)
 536          * PADDING
 537          * ----------- top = topmax - TOP_OF_KERNEL_STACK_PADDING
 538          * stack
 539          * ----------- bottom = start + sizeof(thread_info)
 540          * thread_info
 541          * ----------- start
 542          *
 543          * The tasks stack pointer points at the location where the
 544          * framepointer is stored. The data on the stack is:
 545          * ... IP FP ... IP FP
 546          *
 547          * We need to read FP and IP, so we need to adjust the upper
 548          * bound by another unsigned long.
 549          */
 550         top = start + THREAD_SIZE - TOP_OF_KERNEL_STACK_PADDING;
 551         top -= 2 * sizeof(unsigned long);
 552         bottom = start + sizeof(struct thread_info);
 553
 554         sp = READ_ONCE(p->thread.sp);
 555         if (sp < bottom || sp > top)
 556                 return 0;
 557
 558         fp = READ_ONCE_NOCHECK(*(unsigned long *)sp);
 559         do {
 560                 if (fp < bottom || fp > top)
 561                         return 0;
 562                 ip = READ_ONCE_NOCHECK(*(unsigned long *)(fp + sizeof(unsigned long)));
 563                 if (!in_sched_functions(ip))
 564                         return ip;
 565                 fp = READ_ONCE_NOCHECK(*(unsigned long *)fp);
 566         } while (count++ < 16 && p->state != TASK_RUNNING);
 567         return 0;
 568 }