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/clockchips.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/i387.h>
  28 #include <asm/fpu-internal.h>
  29 #include <asm/debugreg.h>
  30 #include <asm/nmi.h>
  31
  32 /*
  33  * per-CPU TSS segments. Threads are completely 'soft' on Linux,
  34  * no more per-task TSS's. The TSS size is kept cacheline-aligned
  35  * so they are allowed to end up in the .data..cacheline_aligned
  36  * section. Since TSS's are completely CPU-local, we want them
  37  * on exact cacheline boundaries, to eliminate cacheline ping-pong.
  38  */
  39 DEFINE_PER_CPU_SHARED_ALIGNED(struct tss_struct, init_tss) = INIT_TSS;
  40
  41 #ifdef CONFIG_X86_64
  42 static DEFINE_PER_CPU(unsigned char, is_idle);
  43 static ATOMIC_NOTIFIER_HEAD(idle_notifier);
  44
  45 void idle_notifier_register(struct notifier_block *n)
  46 {
  47         atomic_notifier_chain_register(&idle_notifier, n);
  48 }
  49 EXPORT_SYMBOL_GPL(idle_notifier_register);
  50
  51 void idle_notifier_unregister(struct notifier_block *n)
  52 {
  53         atomic_notifier_chain_unregister(&idle_notifier, n);
  54 }
  55 EXPORT_SYMBOL_GPL(idle_notifier_unregister);
  56 #endif
  57
  58 struct kmem_cache *task_xstate_cachep;
  59 EXPORT_SYMBOL_GPL(task_xstate_cachep);
  60
  61 /*
  62  * this gets called so that we can store lazy state into memory and copy the
  63  * current task into the new thread.
  64  */
  65 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
  66 {
  67         int ret;
  68
  69         *dst = *src;
  70         if (fpu_allocated(&src->thread.fpu)) {
  71                 memset(&dst->thread.fpu, 0, sizeof(dst->thread.fpu));
  72                 ret = fpu_alloc(&dst->thread.fpu);
  73                 if (ret)
  74                         return ret;
  75                 fpu_copy(dst, src);
  76         }
  77         return 0;
  78 }
  79
  80 void free_thread_xstate(struct task_struct *tsk)
  81 {
  82         fpu_free(&tsk->thread.fpu);
  83 }
  84
  85 void arch_release_task_struct(struct task_struct *tsk)
  86 {
  87         free_thread_xstate(tsk);
  88 }
  89
  90 void arch_task_cache_init(void)
  91 {
  92         task_xstate_cachep =
  93                 kmem_cache_create("task_xstate", xstate_size,
  94                                   __alignof__(union thread_xstate),
  95                                   SLAB_PANIC | SLAB_NOTRACK, NULL);
  96 }
  97
  98 /*
  99  * Free current thread data structures etc..
 100  */
 101 void exit_thread(void)
 102 {
 103         struct task_struct *me = current;
 104         struct thread_struct *t = &me->thread;
 105         unsigned long *bp = t->io_bitmap_ptr;
 106
 107         if (bp) {
 108                 struct tss_struct *tss = &per_cpu(init_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         drop_fpu(me);
 122 }
 123
 124 void flush_thread(void)
 125 {
 126         struct task_struct *tsk = current;
 127
 128         flush_ptrace_hw_breakpoint(tsk);
 129         memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
 130         drop_init_fpu(tsk);
 131         /*
 132          * Free the FPU state for non xsave platforms. They get reallocated
 133          * lazily at the first use.
 134          */
 135         if (!use_eager_fpu())
 136                 free_thread_xstate(tsk);
 137 }
 138
 139 static void hard_disable_TSC(void)
 140 {
 141         write_cr4(read_cr4() | X86_CR4_TSD);
 142 }
 143
 144 void disable_TSC(void)
 145 {
 146         preempt_disable();
 147         if (!test_and_set_thread_flag(TIF_NOTSC))
 148                 /*
 149                  * Must flip the CPU state synchronously with
 150                  * TIF_NOTSC in the current running context.
 151                  */
 152                 hard_disable_TSC();
 153         preempt_enable();
 154 }
 155
 156 static void hard_enable_TSC(void)
 157 {
 158         write_cr4(read_cr4() & ~X86_CR4_TSD);
 159 }
 160
 161 static void enable_TSC(void)
 162 {
 163         preempt_disable();
 164         if (test_and_clear_thread_flag(TIF_NOTSC))
 165                 /*
 166                  * Must flip the CPU state synchronously with
 167                  * TIF_NOTSC in the current running context.
 168                  */
 169                 hard_enable_TSC();
 170         preempt_enable();
 171 }
 172
 173 int get_tsc_mode(unsigned long adr)
 174 {
 175         unsigned int val;
 176
 177         if (test_thread_flag(TIF_NOTSC))
 178                 val = PR_TSC_SIGSEGV;
 179         else
 180                 val = PR_TSC_ENABLE;
 181
 182         return put_user(val, (unsigned int __user *)adr);
 183 }
 184
 185 int set_tsc_mode(unsigned int val)
 186 {
 187         if (val == PR_TSC_SIGSEGV)
 188                 disable_TSC();
 189         else if (val == PR_TSC_ENABLE)
 190                 enable_TSC();
 191         else
 192                 return -EINVAL;
 193
 194         return 0;
 195 }
 196
 197 void __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p,
 198                       struct tss_struct *tss)
 199 {
 200         struct thread_struct *prev, *next;
 201
 202         prev = &prev_p->thread;
 203         next = &next_p->thread;
 204
 205         if (test_tsk_thread_flag(prev_p, TIF_BLOCKSTEP) ^
 206             test_tsk_thread_flag(next_p, TIF_BLOCKSTEP)) {
 207                 unsigned long debugctl = get_debugctlmsr();
 208
 209                 debugctl &= ~DEBUGCTLMSR_BTF;
 210                 if (test_tsk_thread_flag(next_p, TIF_BLOCKSTEP))
 211                         debugctl |= DEBUGCTLMSR_BTF;
 212
 213                 update_debugctlmsr(debugctl);
 214         }
 215
 216         if (test_tsk_thread_flag(prev_p, TIF_NOTSC) ^
 217             test_tsk_thread_flag(next_p, TIF_NOTSC)) {
 218                 /* prev and next are different */
 219                 if (test_tsk_thread_flag(next_p, TIF_NOTSC))
 220                         hard_disable_TSC();
 221                 else
 222                         hard_enable_TSC();
 223         }
 224
 225         if (test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
 226                 /*
 227                  * Copy the relevant range of the IO bitmap.
 228                  * Normally this is 128 bytes or less:
 229                  */
 230                 memcpy(tss->io_bitmap, next->io_bitmap_ptr,
 231                        max(prev->io_bitmap_max, next->io_bitmap_max));
 232         } else if (test_tsk_thread_flag(prev_p, TIF_IO_BITMAP)) {
 233                 /*
 234                  * Clear any possible leftover bits:
 235                  */
 236                 memset(tss->io_bitmap, 0xff, prev->io_bitmap_max);
 237         }
 238         propagate_user_return_notify(prev_p, next_p);
 239 }
 240
 241 /*
 242  * Idle related variables and functions
 243  */
 244 unsigned long boot_option_idle_override = IDLE_NO_OVERRIDE;
 245 EXPORT_SYMBOL(boot_option_idle_override);
 246
 247 static void (*x86_idle)(void);
 248
 249 #ifndef CONFIG_SMP
 250 static inline void play_dead(void)
 251 {
 252         BUG();
 253 }
 254 #endif
 255
 256 #ifdef CONFIG_X86_64
 257 void enter_idle(void)
 258 {
 259         this_cpu_write(is_idle, 1);
 260         atomic_notifier_call_chain(&idle_notifier, IDLE_START, NULL);
 261 }
 262
 263 static void __exit_idle(void)
 264 {
 265         if (x86_test_and_clear_bit_percpu(0, is_idle) == 0)
 266                 return;
 267         atomic_notifier_call_chain(&idle_notifier, IDLE_END, NULL);
 268 }
 269
 270 /* Called from interrupts to signify idle end */
 271 void exit_idle(void)
 272 {
 273         /* idle loop has pid 0 */
 274         if (current->pid)
 275                 return;
 276         __exit_idle();
 277 }
 278 #endif
 279
 280 void arch_cpu_idle_prepare(void)
 281 {
 282         /*
 283          * If we're the non-boot CPU, nothing set the stack canary up
 284          * for us.  CPU0 already has it initialized but no harm in
 285          * doing it again.  This is a good place for updating it, as
 286          * we wont ever return from this function (so the invalid
 287          * canaries already on the stack wont ever trigger).
 288          */
 289         boot_init_stack_canary();
 290 }
 291
 292 void arch_cpu_idle_enter(void)
 293 {
 294         local_touch_nmi();
 295         enter_idle();
 296 }
 297
 298 void arch_cpu_idle_exit(void)
 299 {
 300         __exit_idle();
 301 }
 302
 303 void arch_cpu_idle_dead(void)
 304 {
 305         play_dead();
 306 }
 307
 308 /*
 309  * Called from the generic idle code.
 310  */
 311 void arch_cpu_idle(void)
 312 {
 313         if (cpuidle_idle_call())
 314                 x86_idle();
 315 }
 316
 317 /*
 318  * We use this if we don't have any better idle routine..
 319  */
 320 void default_idle(void)
 321 {
 322         trace_cpu_idle_rcuidle(1, smp_processor_id());
 323         safe_halt();
 324         trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, smp_processor_id());
 325 }
 326 #ifdef CONFIG_APM_MODULE
 327 EXPORT_SYMBOL(default_idle);
 328 #endif
 329
 330 #ifdef CONFIG_XEN
 331 bool xen_set_default_idle(void)
 332 {
 333         bool ret = !!x86_idle;
 334
 335         x86_idle = default_idle;
 336
 337         return ret;
 338 }
 339 #endif
 340 void stop_this_cpu(void *dummy)
 341 {
 342         local_irq_disable();
 343         /*
 344          * Remove this CPU:
 345          */
 346         set_cpu_online(smp_processor_id(), false);
 347         disable_local_APIC();
 348
 349         for (;;)
 350                 halt();
 351 }
 352
 353 bool amd_e400_c1e_detected;
 354 EXPORT_SYMBOL(amd_e400_c1e_detected);
 355
 356 static cpumask_var_t amd_e400_c1e_mask;
 357
 358 void amd_e400_remove_cpu(int cpu)
 359 {
 360         if (amd_e400_c1e_mask != NULL)
 361                 cpumask_clear_cpu(cpu, amd_e400_c1e_mask);
 362 }
 363
 364 /*
 365  * AMD Erratum 400 aware idle routine. We check for C1E active in the interrupt
 366  * pending message MSR. If we detect C1E, then we handle it the same
 367  * way as C3 power states (local apic timer and TSC stop)
 368  */
 369 static void amd_e400_idle(void)
 370 {
 371         if (need_resched())
 372                 return;
 373
 374         if (!amd_e400_c1e_detected) {
 375                 u32 lo, hi;
 376
 377                 rdmsr(MSR_K8_INT_PENDING_MSG, lo, hi);
 378
 379                 if (lo & K8_INTP_C1E_ACTIVE_MASK) {
 380                         amd_e400_c1e_detected = true;
 381                         if (!boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
 382                                 mark_tsc_unstable("TSC halt in AMD C1E");
 383                         pr_info("System has AMD C1E enabled\n");
 384                 }
 385         }
 386
 387         if (amd_e400_c1e_detected) {
 388                 int cpu = smp_processor_id();
 389
 390                 if (!cpumask_test_cpu(cpu, amd_e400_c1e_mask)) {
 391                         cpumask_set_cpu(cpu, amd_e400_c1e_mask);
 392                         /*
 393                          * Force broadcast so ACPI can not interfere.
 394                          */
 395                         clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_FORCE,
 396                                            &cpu);
 397                         pr_info("Switch to broadcast mode on CPU%d\n", cpu);
 398                 }
 399                 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER, &cpu);
 400
 401                 default_idle();
 402
 403                 /*
 404                  * The switch back from broadcast mode needs to be
 405                  * called with interrupts disabled.
 406                  */
 407                  local_irq_disable();
 408                  clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_EXIT, &cpu);
 409                  local_irq_enable();
 410         } else
 411                 default_idle();
 412 }
 413
 414 void __cpuinit select_idle_routine(const struct cpuinfo_x86 *c)
 415 {
 416 #ifdef CONFIG_SMP
 417         if (boot_option_idle_override == IDLE_POLL && smp_num_siblings > 1)
 418                 pr_warn_once("WARNING: polling idle and HT enabled, performance may degrade\n");
 419 #endif
 420         if (x86_idle || boot_option_idle_override == IDLE_POLL)
 421                 return;
 422
 423         if (cpu_has_bug(c, X86_BUG_AMD_APIC_C1E)) {
 424                 /* E400: APIC timer interrupt does not wake up CPU from C1e */
 425                 pr_info("using AMD E400 aware idle routine\n");
 426                 x86_idle = amd_e400_idle;
 427         } else
 428                 x86_idle = default_idle;
 429 }
 430
 431 void __init init_amd_e400_c1e_mask(void)
 432 {
 433         /* If we're using amd_e400_idle, we need to allocate amd_e400_c1e_mask. */
 434         if (x86_idle == amd_e400_idle)
 435                 zalloc_cpumask_var(&amd_e400_c1e_mask, GFP_KERNEL);
 436 }
 437
 438 static int __init idle_setup(char *str)
 439 {
 440         if (!str)
 441                 return -EINVAL;
 442
 443         if (!strcmp(str, "poll")) {
 444                 pr_info("using polling idle threads\n");
 445                 boot_option_idle_override = IDLE_POLL;
 446                 cpu_idle_poll_ctrl(true);
 447         } else if (!strcmp(str, "halt")) {
 448                 /*
 449                  * When the boot option of idle=halt is added, halt is
 450                  * forced to be used for CPU idle. In such case CPU C2/C3
 451                  * won't be used again.
 452                  * To continue to load the CPU idle driver, don't touch
 453                  * the boot_option_idle_override.
 454                  */
 455                 x86_idle = default_idle;
 456                 boot_option_idle_override = IDLE_HALT;
 457         } else if (!strcmp(str, "nomwait")) {
 458                 /*
 459                  * If the boot option of "idle=nomwait" is added,
 460                  * it means that mwait will be disabled for CPU C2/C3
 461                  * states. In such case it won't touch the variable
 462                  * of boot_option_idle_override.
 463                  */
 464                 boot_option_idle_override = IDLE_NOMWAIT;
 465         } else
 466                 return -1;
 467
 468         return 0;
 469 }
 470 early_param("idle", idle_setup);
 471
 472 unsigned long arch_align_stack(unsigned long sp)
 473 {
 474         if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
 475                 sp -= get_random_int() % 8192;
 476         return sp & ~0xf;
 477 }
 478
 479 unsigned long arch_randomize_brk(struct mm_struct *mm)
 480 {
 481         unsigned long range_end = mm->brk + 0x02000000;
 482         return randomize_range(mm->brk, range_end, 0) ? : mm->brk;
 483 }
 484