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 <linux/user-return-notifier.h>
  13 #include <linux/dmi.h>
  14 #include <linux/utsname.h>
  15 #include <trace/events/power.h>
  16 #include <linux/hw_breakpoint.h>
  17 #include <asm/cpu.h>
  18 #include <asm/system.h>
  19 #include <asm/apic.h>
  20 #include <asm/syscalls.h>
  21 #include <asm/idle.h>
  22 #include <asm/uaccess.h>
  23 #include <asm/i387.h>
  24 #include <asm/debugreg.h>
  25
  26 struct kmem_cache *task_xstate_cachep;
  27 EXPORT_SYMBOL_GPL(task_xstate_cachep);
  28
  29 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
  30 {
  31         int ret;
  32
  33         *dst = *src;
  34         if (fpu_allocated(&src->thread.fpu)) {
  35                 memset(&dst->thread.fpu, 0, sizeof(dst->thread.fpu));
  36                 ret = fpu_alloc(&dst->thread.fpu);
  37                 if (ret)
  38                         return ret;
  39                 fpu_copy(&dst->thread.fpu, &src->thread.fpu);
  40         }
  41         return 0;
  42 }
  43
  44 void free_thread_xstate(struct task_struct *tsk)
  45 {
  46         fpu_free(&tsk->thread.fpu);
  47 }
  48
  49 void free_thread_info(struct thread_info *ti)
  50 {
  51         free_thread_xstate(ti->task);
  52         free_pages((unsigned long)ti, get_order(THREAD_SIZE));
  53 }
  54
  55 void arch_task_cache_init(void)
  56 {
  57         task_xstate_cachep =
  58                 kmem_cache_create("task_xstate", xstate_size,
  59                                   __alignof__(union thread_xstate),
  60                                   SLAB_PANIC | SLAB_NOTRACK, NULL);
  61 }
  62
  63 /*
  64  * Free current thread data structures etc..
  65  */
  66 void exit_thread(void)
  67 {
  68         struct task_struct *me = current;
  69         struct thread_struct *t = &me->thread;
  70         unsigned long *bp = t->io_bitmap_ptr;
  71
  72         if (bp) {
  73                 struct tss_struct *tss = &per_cpu(init_tss, get_cpu());
  74
  75                 t->io_bitmap_ptr = NULL;
  76                 clear_thread_flag(TIF_IO_BITMAP);
  77                 /*
  78                  * Careful, clear this in the TSS too:
  79                  */
  80                 memset(tss->io_bitmap, 0xff, t->io_bitmap_max);
  81                 t->io_bitmap_max = 0;
  82                 put_cpu();
  83                 kfree(bp);
  84         }
  85 }
  86
  87 void show_regs(struct pt_regs *regs)
  88 {
  89         show_registers(regs);
  90         show_trace(NULL, regs, (unsigned long *)kernel_stack_pointer(regs), 0);
  91 }
  92
  93 void show_regs_common(void)
  94 {
  95         const char *vendor, *product, *board;
  96
  97         vendor = dmi_get_system_info(DMI_SYS_VENDOR);
  98         if (!vendor)
  99                 vendor = "";
 100         product = dmi_get_system_info(DMI_PRODUCT_NAME);
 101         if (!product)
 102                 product = "";
 103
 104         /* Board Name is optional */
 105         board = dmi_get_system_info(DMI_BOARD_NAME);
 106
 107         printk(KERN_CONT "\n");
 108         printk(KERN_DEFAULT "Pid: %d, comm: %.20s %s %s %.*s",
 109                 current->pid, current->comm, print_tainted(),
 110                 init_utsname()->release,
 111                 (int)strcspn(init_utsname()->version, " "),
 112                 init_utsname()->version);
 113         printk(KERN_CONT " %s %s", vendor, product);
 114         if (board)
 115                 printk(KERN_CONT "/%s", board);
 116         printk(KERN_CONT "\n");
 117 }
 118
 119 void flush_thread(void)
 120 {
 121         struct task_struct *tsk = current;
 122
 123         flush_ptrace_hw_breakpoint(tsk);
 124         memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
 125         /*
 126          * Forget coprocessor state..
 127          */
 128         tsk->fpu_counter = 0;
 129         clear_fpu(tsk);
 130         clear_used_math();
 131 }
 132
 133 static void hard_disable_TSC(void)
 134 {
 135         write_cr4(read_cr4() | X86_CR4_TSD);
 136 }
 137
 138 void disable_TSC(void)
 139 {
 140         preempt_disable();
 141         if (!test_and_set_thread_flag(TIF_NOTSC))
 142                 /*
 143                  * Must flip the CPU state synchronously with
 144                  * TIF_NOTSC in the current running context.
 145                  */
 146                 hard_disable_TSC();
 147         preempt_enable();
 148 }
 149
 150 static void hard_enable_TSC(void)
 151 {
 152         write_cr4(read_cr4() & ~X86_CR4_TSD);
 153 }
 154
 155 static void enable_TSC(void)
 156 {
 157         preempt_disable();
 158         if (test_and_clear_thread_flag(TIF_NOTSC))
 159                 /*
 160                  * Must flip the CPU state synchronously with
 161                  * TIF_NOTSC in the current running context.
 162                  */
 163                 hard_enable_TSC();
 164         preempt_enable();
 165 }
 166
 167 int get_tsc_mode(unsigned long adr)
 168 {
 169         unsigned int val;
 170
 171         if (test_thread_flag(TIF_NOTSC))
 172                 val = PR_TSC_SIGSEGV;
 173         else
 174                 val = PR_TSC_ENABLE;
 175
 176         return put_user(val, (unsigned int __user *)adr);
 177 }
 178
 179 int set_tsc_mode(unsigned int val)
 180 {
 181         if (val == PR_TSC_SIGSEGV)
 182                 disable_TSC();
 183         else if (val == PR_TSC_ENABLE)
 184                 enable_TSC();
 185         else
 186                 return -EINVAL;
 187
 188         return 0;
 189 }
 190
 191 void __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p,
 192                       struct tss_struct *tss)
 193 {
 194         struct thread_struct *prev, *next;
 195
 196         prev = &prev_p->thread;
 197         next = &next_p->thread;
 198
 199         if (test_tsk_thread_flag(prev_p, TIF_BLOCKSTEP) ^
 200             test_tsk_thread_flag(next_p, TIF_BLOCKSTEP)) {
 201                 unsigned long debugctl = get_debugctlmsr();
 202
 203                 debugctl &= ~DEBUGCTLMSR_BTF;
 204                 if (test_tsk_thread_flag(next_p, TIF_BLOCKSTEP))
 205                         debugctl |= DEBUGCTLMSR_BTF;
 206
 207                 update_debugctlmsr(debugctl);
 208         }
 209
 210         if (test_tsk_thread_flag(prev_p, TIF_NOTSC) ^
 211             test_tsk_thread_flag(next_p, TIF_NOTSC)) {
 212                 /* prev and next are different */
 213                 if (test_tsk_thread_flag(next_p, TIF_NOTSC))
 214                         hard_disable_TSC();
 215                 else
 216                         hard_enable_TSC();
 217         }
 218
 219         if (test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
 220                 /*
 221                  * Copy the relevant range of the IO bitmap.
 222                  * Normally this is 128 bytes or less:
 223                  */
 224                 memcpy(tss->io_bitmap, next->io_bitmap_ptr,
 225                        max(prev->io_bitmap_max, next->io_bitmap_max));
 226         } else if (test_tsk_thread_flag(prev_p, TIF_IO_BITMAP)) {
 227                 /*
 228                  * Clear any possible leftover bits:
 229                  */
 230                 memset(tss->io_bitmap, 0xff, prev->io_bitmap_max);
 231         }
 232         propagate_user_return_notify(prev_p, next_p);
 233 }
 234
 235 int sys_fork(struct pt_regs *regs)
 236 {
 237         return do_fork(SIGCHLD, regs->sp, regs, 0, NULL, NULL);
 238 }
 239
 240 /*
 241  * This is trivial, and on the face of it looks like it
 242  * could equally well be done in user mode.
 243  *
 244  * Not so, for quite unobvious reasons - register pressure.
 245  * In user mode vfork() cannot have a stack frame, and if
 246  * done by calling the "clone()" system call directly, you
 247  * do not have enough call-clobbered registers to hold all
 248  * the information you need.
 249  */
 250 int sys_vfork(struct pt_regs *regs)
 251 {
 252         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->sp, regs, 0,
 253                        NULL, NULL);
 254 }
 255
 256 long
 257 sys_clone(unsigned long clone_flags, unsigned long newsp,
 258           void __user *parent_tid, void __user *child_tid, struct pt_regs *regs)
 259 {
 260         if (!newsp)
 261                 newsp = regs->sp;
 262         return do_fork(clone_flags, newsp, regs, 0, parent_tid, child_tid);
 263 }
 264
 265 /*
 266  * This gets run with %si containing the
 267  * function to call, and %di containing
 268  * the "args".
 269  */
 270 extern void kernel_thread_helper(void);
 271
 272 /*
 273  * Create a kernel thread
 274  */
 275 int kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
 276 {
 277         struct pt_regs regs;
 278
 279         memset(&regs, 0, sizeof(regs));
 280
 281         regs.si = (unsigned long) fn;
 282         regs.di = (unsigned long) arg;
 283
 284 #ifdef CONFIG_X86_32
 285         regs.ds = __USER_DS;
 286         regs.es = __USER_DS;
 287         regs.fs = __KERNEL_PERCPU;
 288         regs.gs = __KERNEL_STACK_CANARY;
 289 #else
 290         regs.ss = __KERNEL_DS;
 291 #endif
 292
 293         regs.orig_ax = -1;
 294         regs.ip = (unsigned long) kernel_thread_helper;
 295         regs.cs = __KERNEL_CS | get_kernel_rpl();
 296         regs.flags = X86_EFLAGS_IF | 0x2;
 297
 298         /* Ok, create the new process.. */
 299         return do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0, &regs, 0, NULL, NULL);
 300 }
 301 EXPORT_SYMBOL(kernel_thread);
 302
 303 /*
 304  * sys_execve() executes a new program.
 305  */
 306 long sys_execve(const char __user *name,
 307                 const char __user *const __user *argv,
 308                 const char __user *const __user *envp, struct pt_regs *regs)
 309 {
 310         long error;
 311         char *filename;
 312
 313         filename = getname(name);
 314         error = PTR_ERR(filename);
 315         if (IS_ERR(filename))
 316                 return error;
 317         error = do_execve(filename, argv, envp, regs);
 318
 319 #ifdef CONFIG_X86_32
 320         if (error == 0) {
 321                 /* Make sure we don't return using sysenter.. */
 322                 set_thread_flag(TIF_IRET);
 323         }
 324 #endif
 325
 326         putname(filename);
 327         return error;
 328 }
 329
 330 /*
 331  * Idle related variables and functions
 332  */
 333 unsigned long boot_option_idle_override = IDLE_NO_OVERRIDE;
 334 EXPORT_SYMBOL(boot_option_idle_override);
 335
 336 /*
 337  * Powermanagement idle function, if any..
 338  */
 339 void (*pm_idle)(void);
 340 #ifdef CONFIG_APM_MODULE
 341 EXPORT_SYMBOL(pm_idle);
 342 #endif
 343
 344 #ifdef CONFIG_X86_32
 345 /*
 346  * This halt magic was a workaround for ancient floppy DMA
 347  * wreckage. It should be safe to remove.
 348  */
 349 static int hlt_counter;
 350 void disable_hlt(void)
 351 {
 352         hlt_counter++;
 353 }
 354 EXPORT_SYMBOL(disable_hlt);
 355
 356 void enable_hlt(void)
 357 {
 358         hlt_counter--;
 359 }
 360 EXPORT_SYMBOL(enable_hlt);
 361
 362 static inline int hlt_use_halt(void)
 363 {
 364         return (!hlt_counter && boot_cpu_data.hlt_works_ok);
 365 }
 366 #else
 367 static inline int hlt_use_halt(void)
 368 {
 369         return 1;
 370 }
 371 #endif
 372
 373 /*
 374  * We use this if we don't have any better
 375  * idle routine..
 376  */
 377 void default_idle(void)
 378 {
 379         if (hlt_use_halt()) {
 380                 trace_power_start(POWER_CSTATE, 1, smp_processor_id());
 381                 trace_cpu_idle(1, smp_processor_id());
 382                 current_thread_info()->status &= ~TS_POLLING;
 383                 /*
 384                  * TS_POLLING-cleared state must be visible before we
 385                  * test NEED_RESCHED:
 386                  */
 387                 smp_mb();
 388
 389                 if (!need_resched())
 390                         safe_halt();    /* enables interrupts racelessly */
 391                 else
 392                         local_irq_enable();
 393                 current_thread_info()->status |= TS_POLLING;
 394                 trace_power_end(smp_processor_id());
 395                 trace_cpu_idle(PWR_EVENT_EXIT, smp_processor_id());
 396         } else {
 397                 local_irq_enable();
 398                 /* loop is done by the caller */
 399                 cpu_relax();
 400         }
 401 }
 402 #ifdef CONFIG_APM_MODULE
 403 EXPORT_SYMBOL(default_idle);
 404 #endif
 405
 406 void stop_this_cpu(void *dummy)
 407 {
 408         local_irq_disable();
 409         /*
 410          * Remove this CPU:
 411          */
 412         set_cpu_online(smp_processor_id(), false);
 413         disable_local_APIC();
 414
 415         for (;;) {
 416                 if (hlt_works(smp_processor_id()))
 417                         halt();
 418         }
 419 }
 420
 421 static void do_nothing(void *unused)
 422 {
 423 }
 424
 425 /*
 426  * cpu_idle_wait - Used to ensure that all the CPUs discard old value of
 427  * pm_idle and update to new pm_idle value. Required while changing pm_idle
 428  * handler on SMP systems.
 429  *
 430  * Caller must have changed pm_idle to the new value before the call. Old
 431  * pm_idle value will not be used by any CPU after the return of this function.
 432  */
 433 void cpu_idle_wait(void)
 434 {
 435         smp_mb();
 436         /* kick all the CPUs so that they exit out of pm_idle */
 437         smp_call_function(do_nothing, NULL, 1);
 438 }
 439 EXPORT_SYMBOL_GPL(cpu_idle_wait);
 440
 441 /* Default MONITOR/MWAIT with no hints, used for default C1 state */
 442 static void mwait_idle(void)
 443 {
 444         if (!need_resched()) {
 445                 trace_power_start(POWER_CSTATE, 1, smp_processor_id());
 446                 trace_cpu_idle(1, smp_processor_id());
 447                 if (this_cpu_has(X86_FEATURE_CLFLUSH_MONITOR))
 448                         clflush((void *)&current_thread_info()->flags);
 449
 450                 __monitor((void *)&current_thread_info()->flags, 0, 0);
 451                 smp_mb();
 452                 if (!need_resched())
 453                         __sti_mwait(0, 0);
 454                 else
 455                         local_irq_enable();
 456                 trace_power_end(smp_processor_id());
 457                 trace_cpu_idle(PWR_EVENT_EXIT, smp_processor_id());
 458         } else
 459                 local_irq_enable();
 460 }
 461
 462 /*
 463  * On SMP it's slightly faster (but much more power-consuming!)
 464  * to poll the ->work.need_resched flag instead of waiting for the
 465  * cross-CPU IPI to arrive. Use this option with caution.
 466  */
 467 static void poll_idle(void)
 468 {
 469         trace_power_start(POWER_CSTATE, 0, smp_processor_id());
 470         trace_cpu_idle(0, smp_processor_id());
 471         local_irq_enable();
 472         while (!need_resched())
 473                 cpu_relax();
 474         trace_power_end(smp_processor_id());
 475         trace_cpu_idle(PWR_EVENT_EXIT, smp_processor_id());
 476 }
 477
 478 /*
 479  * mwait selection logic:
 480  *
 481  * It depends on the CPU. For AMD CPUs that support MWAIT this is
 482  * wrong. Family 0x10 and 0x11 CPUs will enter C1 on HLT. Powersavings
 483  * then depend on a clock divisor and current Pstate of the core. If
 484  * all cores of a processor are in halt state (C1) the processor can
 485  * enter the C1E (C1 enhanced) state. If mwait is used this will never
 486  * happen.
 487  *
 488  * idle=mwait overrides this decision and forces the usage of mwait.
 489  */
 490
 491 #define MWAIT_INFO                      0x05
 492 #define MWAIT_ECX_EXTENDED_INFO         0x01
 493 #define MWAIT_EDX_C1                    0xf0
 494
 495 int mwait_usable(const struct cpuinfo_x86 *c)
 496 {
 497         u32 eax, ebx, ecx, edx;
 498
 499         if (boot_option_idle_override == IDLE_FORCE_MWAIT)
 500                 return 1;
 501
 502         if (c->cpuid_level < MWAIT_INFO)
 503                 return 0;
 504
 505         cpuid(MWAIT_INFO, &eax, &ebx, &ecx, &edx);
 506         /* Check, whether EDX has extended info about MWAIT */
 507         if (!(ecx & MWAIT_ECX_EXTENDED_INFO))
 508                 return 1;
 509
 510         /*
 511          * edx enumeratios MONITOR/MWAIT extensions. Check, whether
 512          * C1  supports MWAIT
 513          */
 514         return (edx & MWAIT_EDX_C1);
 515 }
 516
 517 bool amd_e400_c1e_detected;
 518 EXPORT_SYMBOL(amd_e400_c1e_detected);
 519
 520 static cpumask_var_t amd_e400_c1e_mask;
 521
 522 void amd_e400_remove_cpu(int cpu)
 523 {
 524         if (amd_e400_c1e_mask != NULL)
 525                 cpumask_clear_cpu(cpu, amd_e400_c1e_mask);
 526 }
 527
 528 /*
 529  * AMD Erratum 400 aware idle routine. We check for C1E active in the interrupt
 530  * pending message MSR. If we detect C1E, then we handle it the same
 531  * way as C3 power states (local apic timer and TSC stop)
 532  */
 533 static void amd_e400_idle(void)
 534 {
 535         if (need_resched())
 536                 return;
 537
 538         if (!amd_e400_c1e_detected) {
 539                 u32 lo, hi;
 540
 541                 rdmsr(MSR_K8_INT_PENDING_MSG, lo, hi);
 542
 543                 if (lo & K8_INTP_C1E_ACTIVE_MASK) {
 544                         amd_e400_c1e_detected = true;
 545                         if (!boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
 546                                 mark_tsc_unstable("TSC halt in AMD C1E");
 547                         printk(KERN_INFO "System has AMD C1E enabled\n");
 548                 }
 549         }
 550
 551         if (amd_e400_c1e_detected) {
 552                 int cpu = smp_processor_id();
 553
 554                 if (!cpumask_test_cpu(cpu, amd_e400_c1e_mask)) {
 555                         cpumask_set_cpu(cpu, amd_e400_c1e_mask);
 556                         /*
 557                          * Force broadcast so ACPI can not interfere.
 558                          */
 559                         clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_FORCE,
 560                                            &cpu);
 561                         printk(KERN_INFO "Switch to broadcast mode on CPU%d\n",
 562                                cpu);
 563                 }
 564                 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER, &cpu);
 565
 566                 default_idle();
 567
 568                 /*
 569                  * The switch back from broadcast mode needs to be
 570                  * called with interrupts disabled.
 571                  */
 572                  local_irq_disable();
 573                  clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_EXIT, &cpu);
 574                  local_irq_enable();
 575         } else
 576                 default_idle();
 577 }
 578
 579 void __cpuinit select_idle_routine(const struct cpuinfo_x86 *c)
 580 {
 581 #ifdef CONFIG_SMP
 582         if (pm_idle == poll_idle && smp_num_siblings > 1) {
 583                 printk_once(KERN_WARNING "WARNING: polling idle and HT enabled,"
 584                         " performance may degrade.\n");
 585         }
 586 #endif
 587         if (pm_idle)
 588                 return;
 589
 590         if (cpu_has(c, X86_FEATURE_MWAIT) && mwait_usable(c)) {
 591                 /*
 592                  * One CPU supports mwait => All CPUs supports mwait
 593                  */
 594                 printk(KERN_INFO "using mwait in idle threads.\n");
 595                 pm_idle = mwait_idle;
 596         } else if (cpu_has_amd_erratum(amd_erratum_400)) {
 597                 /* E400: APIC timer interrupt does not wake up CPU from C1e */
 598                 printk(KERN_INFO "using AMD E400 aware idle routine\n");
 599                 pm_idle = amd_e400_idle;
 600         } else
 601                 pm_idle = default_idle;
 602 }
 603
 604 void __init init_amd_e400_c1e_mask(void)
 605 {
 606         /* If we're using amd_e400_idle, we need to allocate amd_e400_c1e_mask. */
 607         if (pm_idle == amd_e400_idle)
 608                 zalloc_cpumask_var(&amd_e400_c1e_mask, GFP_KERNEL);
 609 }
 610
 611 static int __init idle_setup(char *str)
 612 {
 613         if (!str)
 614                 return -EINVAL;
 615
 616         if (!strcmp(str, "poll")) {
 617                 printk("using polling idle threads.\n");
 618                 pm_idle = poll_idle;
 619                 boot_option_idle_override = IDLE_POLL;
 620         } else if (!strcmp(str, "mwait")) {
 621                 boot_option_idle_override = IDLE_FORCE_MWAIT;
 622                 WARN_ONCE(1, "\"idle=mwait\" will be removed in 2012\n");
 623         } else if (!strcmp(str, "halt")) {
 624                 /*
 625                  * When the boot option of idle=halt is added, halt is
 626                  * forced to be used for CPU idle. In such case CPU C2/C3
 627                  * won't be used again.
 628                  * To continue to load the CPU idle driver, don't touch
 629                  * the boot_option_idle_override.
 630                  */
 631                 pm_idle = default_idle;
 632                 boot_option_idle_override = IDLE_HALT;
 633         } else if (!strcmp(str, "nomwait")) {
 634                 /*
 635                  * If the boot option of "idle=nomwait" is added,
 636                  * it means that mwait will be disabled for CPU C2/C3
 637                  * states. In such case it won't touch the variable
 638                  * of boot_option_idle_override.
 639                  */
 640                 boot_option_idle_override = IDLE_NOMWAIT;
 641         } else
 642                 return -1;
 643
 644         return 0;
 645 }
 646 early_param("idle", idle_setup);
 647
 648 unsigned long arch_align_stack(unsigned long sp)
 649 {
 650         if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
 651                 sp -= get_random_int() % 8192;
 652         return sp & ~0xf;
 653 }
 654
 655 unsigned long arch_randomize_brk(struct mm_struct *mm)
 656 {
 657         unsigned long range_end = mm->brk + 0x02000000;
 658         return randomize_range(mm->brk, range_end, 0) ? : mm->brk;
 659 }
 660