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/system.h>
  18 #include <asm/apic.h>
  19 #include <asm/syscalls.h>
  20 #include <asm/idle.h>
  21 #include <asm/uaccess.h>
  22 #include <asm/i387.h>
  23 #include <asm/debugreg.h>
  24
  25 unsigned long idle_halt;
  26 EXPORT_SYMBOL(idle_halt);
  27 unsigned long idle_nomwait;
  28 EXPORT_SYMBOL(idle_nomwait);
  29
  30 struct kmem_cache *task_xstate_cachep;
  31
  32 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
  33 {
  34         int ret;
  35
  36         *dst = *src;
  37         if (fpu_allocated(&src->thread.fpu)) {
  38                 memset(&dst->thread.fpu, 0, sizeof(dst->thread.fpu));
  39                 ret = fpu_alloc(&dst->thread.fpu);
  40                 if (ret)
  41                         return ret;
  42                 fpu_copy(&dst->thread.fpu, &src->thread.fpu);
  43         }
  44         return 0;
  45 }
  46
  47 void free_thread_xstate(struct task_struct *tsk)
  48 {
  49         fpu_free(&tsk->thread.fpu);
  50 }
  51
  52 void free_thread_info(struct thread_info *ti)
  53 {
  54         free_thread_xstate(ti->task);
  55         free_pages((unsigned long)ti, get_order(THREAD_SIZE));
  56 }
  57
  58 void arch_task_cache_init(void)
  59 {
  60         task_xstate_cachep =
  61                 kmem_cache_create("task_xstate", xstate_size,
  62                                   __alignof__(union thread_xstate),
  63                                   SLAB_PANIC | SLAB_NOTRACK, NULL);
  64 }
  65
  66 /*
  67  * Free current thread data structures etc..
  68  */
  69 void exit_thread(void)
  70 {
  71         struct task_struct *me = current;
  72         struct thread_struct *t = &me->thread;
  73         unsigned long *bp = t->io_bitmap_ptr;
  74
  75         if (bp) {
  76                 struct tss_struct *tss = &per_cpu(init_tss, get_cpu());
  77
  78                 t->io_bitmap_ptr = NULL;
  79                 clear_thread_flag(TIF_IO_BITMAP);
  80                 /*
  81                  * Careful, clear this in the TSS too:
  82                  */
  83                 memset(tss->io_bitmap, 0xff, t->io_bitmap_max);
  84                 t->io_bitmap_max = 0;
  85                 put_cpu();
  86                 kfree(bp);
  87         }
  88 }
  89
  90 void show_regs(struct pt_regs *regs)
  91 {
  92         show_registers(regs);
  93         show_trace(NULL, regs, (unsigned long *)kernel_stack_pointer(regs),
  94                    regs->bp);
  95 }
  96
  97 void show_regs_common(void)
  98 {
  99         const char *board, *product;
 100
 101         board = dmi_get_system_info(DMI_BOARD_NAME);
 102         if (!board)
 103                 board = "";
 104         product = dmi_get_system_info(DMI_PRODUCT_NAME);
 105         if (!product)
 106                 product = "";
 107
 108         printk(KERN_CONT "\n");
 109         printk(KERN_DEFAULT "Pid: %d, comm: %.20s %s %s %.*s %s/%s\n",
 110                 current->pid, current->comm, print_tainted(),
 111                 init_utsname()->release,
 112                 (int)strcspn(init_utsname()->version, " "),
 113                 init_utsname()->version, board, product);
 114 }
 115
 116 void flush_thread(void)
 117 {
 118         struct task_struct *tsk = current;
 119
 120         flush_ptrace_hw_breakpoint(tsk);
 121         memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
 122         /*
 123          * Forget coprocessor state..
 124          */
 125         tsk->fpu_counter = 0;
 126         clear_fpu(tsk);
 127         clear_used_math();
 128 }
 129
 130 static void hard_disable_TSC(void)
 131 {
 132         write_cr4(read_cr4() | 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         write_cr4(read_cr4() & ~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 int sys_fork(struct pt_regs *regs)
 233 {
 234         return do_fork(SIGCHLD, regs->sp, regs, 0, NULL, NULL);
 235 }
 236
 237 /*
 238  * This is trivial, and on the face of it looks like it
 239  * could equally well be done in user mode.
 240  *
 241  * Not so, for quite unobvious reasons - register pressure.
 242  * In user mode vfork() cannot have a stack frame, and if
 243  * done by calling the "clone()" system call directly, you
 244  * do not have enough call-clobbered registers to hold all
 245  * the information you need.
 246  */
 247 int sys_vfork(struct pt_regs *regs)
 248 {
 249         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->sp, regs, 0,
 250                        NULL, NULL);
 251 }
 252
 253 long
 254 sys_clone(unsigned long clone_flags, unsigned long newsp,
 255           void __user *parent_tid, void __user *child_tid, struct pt_regs *regs)
 256 {
 257         if (!newsp)
 258                 newsp = regs->sp;
 259         return do_fork(clone_flags, newsp, regs, 0, parent_tid, child_tid);
 260 }
 261
 262 /*
 263  * This gets run with %si containing the
 264  * function to call, and %di containing
 265  * the "args".
 266  */
 267 extern void kernel_thread_helper(void);
 268
 269 /*
 270  * Create a kernel thread
 271  */
 272 int kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
 273 {
 274         struct pt_regs regs;
 275
 276         memset(&regs, 0, sizeof(regs));
 277
 278         regs.si = (unsigned long) fn;
 279         regs.di = (unsigned long) arg;
 280
 281 #ifdef CONFIG_X86_32
 282         regs.ds = __USER_DS;
 283         regs.es = __USER_DS;
 284         regs.fs = __KERNEL_PERCPU;
 285         regs.gs = __KERNEL_STACK_CANARY;
 286 #else
 287         regs.ss = __KERNEL_DS;
 288 #endif
 289
 290         regs.orig_ax = -1;
 291         regs.ip = (unsigned long) kernel_thread_helper;
 292         regs.cs = __KERNEL_CS | get_kernel_rpl();
 293         regs.flags = X86_EFLAGS_IF | 0x2;
 294
 295         /* Ok, create the new process.. */
 296         return do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0, &regs, 0, NULL, NULL);
 297 }
 298 EXPORT_SYMBOL(kernel_thread);
 299
 300 /*
 301  * sys_execve() executes a new program.
 302  */
 303 long sys_execve(char __user *name, char __user * __user *argv,
 304                 char __user * __user *envp, struct pt_regs *regs)
 305 {
 306         long error;
 307         char *filename;
 308
 309         filename = getname(name);
 310         error = PTR_ERR(filename);
 311         if (IS_ERR(filename))
 312                 return error;
 313         error = do_execve(filename, argv, envp, regs);
 314
 315 #ifdef CONFIG_X86_32
 316         if (error == 0) {
 317                 /* Make sure we don't return using sysenter.. */
 318                 set_thread_flag(TIF_IRET);
 319         }
 320 #endif
 321
 322         putname(filename);
 323         return error;
 324 }
 325
 326 /*
 327  * Idle related variables and functions
 328  */
 329 unsigned long boot_option_idle_override = 0;
 330 EXPORT_SYMBOL(boot_option_idle_override);
 331
 332 /*
 333  * Powermanagement idle function, if any..
 334  */
 335 void (*pm_idle)(void);
 336 EXPORT_SYMBOL(pm_idle);
 337
 338 #ifdef CONFIG_X86_32
 339 /*
 340  * This halt magic was a workaround for ancient floppy DMA
 341  * wreckage. It should be safe to remove.
 342  */
 343 static int hlt_counter;
 344 void disable_hlt(void)
 345 {
 346         hlt_counter++;
 347 }
 348 EXPORT_SYMBOL(disable_hlt);
 349
 350 void enable_hlt(void)
 351 {
 352         hlt_counter--;
 353 }
 354 EXPORT_SYMBOL(enable_hlt);
 355
 356 static inline int hlt_use_halt(void)
 357 {
 358         return (!hlt_counter && boot_cpu_data.hlt_works_ok);
 359 }
 360 #else
 361 static inline int hlt_use_halt(void)
 362 {
 363         return 1;
 364 }
 365 #endif
 366
 367 /*
 368  * We use this if we don't have any better
 369  * idle routine..
 370  */
 371 void default_idle(void)
 372 {
 373         if (hlt_use_halt()) {
 374                 trace_power_start(POWER_CSTATE, 1);
 375                 current_thread_info()->status &= ~TS_POLLING;
 376                 /*
 377                  * TS_POLLING-cleared state must be visible before we
 378                  * test NEED_RESCHED:
 379                  */
 380                 smp_mb();
 381
 382                 if (!need_resched())
 383                         safe_halt();    /* enables interrupts racelessly */
 384                 else
 385                         local_irq_enable();
 386                 current_thread_info()->status |= TS_POLLING;
 387         } else {
 388                 local_irq_enable();
 389                 /* loop is done by the caller */
 390                 cpu_relax();
 391         }
 392 }
 393 #ifdef CONFIG_APM_MODULE
 394 EXPORT_SYMBOL(default_idle);
 395 #endif
 396
 397 void stop_this_cpu(void *dummy)
 398 {
 399         local_irq_disable();
 400         /*
 401          * Remove this CPU:
 402          */
 403         set_cpu_online(smp_processor_id(), false);
 404         disable_local_APIC();
 405
 406         for (;;) {
 407                 if (hlt_works(smp_processor_id()))
 408                         halt();
 409         }
 410 }
 411
 412 static void do_nothing(void *unused)
 413 {
 414 }
 415
 416 /*
 417  * cpu_idle_wait - Used to ensure that all the CPUs discard old value of
 418  * pm_idle and update to new pm_idle value. Required while changing pm_idle
 419  * handler on SMP systems.
 420  *
 421  * Caller must have changed pm_idle to the new value before the call. Old
 422  * pm_idle value will not be used by any CPU after the return of this function.
 423  */
 424 void cpu_idle_wait(void)
 425 {
 426         smp_mb();
 427         /* kick all the CPUs so that they exit out of pm_idle */
 428         smp_call_function(do_nothing, NULL, 1);
 429 }
 430 EXPORT_SYMBOL_GPL(cpu_idle_wait);
 431
 432 /*
 433  * This uses new MONITOR/MWAIT instructions on P4 processors with PNI,
 434  * which can obviate IPI to trigger checking of need_resched.
 435  * We execute MONITOR against need_resched and enter optimized wait state
 436  * through MWAIT. Whenever someone changes need_resched, we would be woken
 437  * up from MWAIT (without an IPI).
 438  *
 439  * New with Core Duo processors, MWAIT can take some hints based on CPU
 440  * capability.
 441  */
 442 void mwait_idle_with_hints(unsigned long ax, unsigned long cx)
 443 {
 444         trace_power_start(POWER_CSTATE, (ax>>4)+1);
 445         if (!need_resched()) {
 446                 if (cpu_has(&current_cpu_data, X86_FEATURE_CLFLUSH_MONITOR))
 447                         clflush((void *)&current_thread_info()->flags);
 448
 449                 __monitor((void *)&current_thread_info()->flags, 0, 0);
 450                 smp_mb();
 451                 if (!need_resched())
 452                         __mwait(ax, cx);
 453         }
 454 }
 455
 456 /* Default MONITOR/MWAIT with no hints, used for default C1 state */
 457 static void mwait_idle(void)
 458 {
 459         if (!need_resched()) {
 460                 trace_power_start(POWER_CSTATE, 1);
 461                 if (cpu_has(&current_cpu_data, X86_FEATURE_CLFLUSH_MONITOR))
 462                         clflush((void *)&current_thread_info()->flags);
 463
 464                 __monitor((void *)&current_thread_info()->flags, 0, 0);
 465                 smp_mb();
 466                 if (!need_resched())
 467                         __sti_mwait(0, 0);
 468                 else
 469                         local_irq_enable();
 470         } else
 471                 local_irq_enable();
 472 }
 473
 474 /*
 475  * On SMP it's slightly faster (but much more power-consuming!)
 476  * to poll the ->work.need_resched flag instead of waiting for the
 477  * cross-CPU IPI to arrive. Use this option with caution.
 478  */
 479 static void poll_idle(void)
 480 {
 481         trace_power_start(POWER_CSTATE, 0);
 482         local_irq_enable();
 483         while (!need_resched())
 484                 cpu_relax();
 485         trace_power_end(0);
 486 }
 487
 488 /*
 489  * mwait selection logic:
 490  *
 491  * It depends on the CPU. For AMD CPUs that support MWAIT this is
 492  * wrong. Family 0x10 and 0x11 CPUs will enter C1 on HLT. Powersavings
 493  * then depend on a clock divisor and current Pstate of the core. If
 494  * all cores of a processor are in halt state (C1) the processor can
 495  * enter the C1E (C1 enhanced) state. If mwait is used this will never
 496  * happen.
 497  *
 498  * idle=mwait overrides this decision and forces the usage of mwait.
 499  */
 500 static int __cpuinitdata force_mwait;
 501
 502 #define MWAIT_INFO                      0x05
 503 #define MWAIT_ECX_EXTENDED_INFO         0x01
 504 #define MWAIT_EDX_C1                    0xf0
 505
 506 static int __cpuinit mwait_usable(const struct cpuinfo_x86 *c)
 507 {
 508         u32 eax, ebx, ecx, edx;
 509
 510         if (force_mwait)
 511                 return 1;
 512
 513         if (c->cpuid_level < MWAIT_INFO)
 514                 return 0;
 515
 516         cpuid(MWAIT_INFO, &eax, &ebx, &ecx, &edx);
 517         /* Check, whether EDX has extended info about MWAIT */
 518         if (!(ecx & MWAIT_ECX_EXTENDED_INFO))
 519                 return 1;
 520
 521         /*
 522          * edx enumeratios MONITOR/MWAIT extensions. Check, whether
 523          * C1  supports MWAIT
 524          */
 525         return (edx & MWAIT_EDX_C1);
 526 }
 527
 528 static cpumask_var_t c1e_mask;
 529 static int c1e_detected;
 530
 531 void c1e_remove_cpu(int cpu)
 532 {
 533         if (c1e_mask != NULL)
 534                 cpumask_clear_cpu(cpu, c1e_mask);
 535 }
 536
 537 /*
 538  * C1E aware idle routine. We check for C1E active in the interrupt
 539  * pending message MSR. If we detect C1E, then we handle it the same
 540  * way as C3 power states (local apic timer and TSC stop)
 541  */
 542 static void c1e_idle(void)
 543 {
 544         if (need_resched())
 545                 return;
 546
 547         if (!c1e_detected) {
 548                 u32 lo, hi;
 549
 550                 rdmsr(MSR_K8_INT_PENDING_MSG, lo, hi);
 551                 if (lo & K8_INTP_C1E_ACTIVE_MASK) {
 552                         c1e_detected = 1;
 553                         if (!boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
 554                                 mark_tsc_unstable("TSC halt in AMD C1E");
 555                         printk(KERN_INFO "System has AMD C1E enabled\n");
 556                         set_cpu_cap(&boot_cpu_data, X86_FEATURE_AMDC1E);
 557                 }
 558         }
 559
 560         if (c1e_detected) {
 561                 int cpu = smp_processor_id();
 562
 563                 if (!cpumask_test_cpu(cpu, c1e_mask)) {
 564                         cpumask_set_cpu(cpu, c1e_mask);
 565                         /*
 566                          * Force broadcast so ACPI can not interfere.
 567                          */
 568                         clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_FORCE,
 569                                            &cpu);
 570                         printk(KERN_INFO "Switch to broadcast mode on CPU%d\n",
 571                                cpu);
 572                 }
 573                 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER, &cpu);
 574
 575                 default_idle();
 576
 577                 /*
 578                  * The switch back from broadcast mode needs to be
 579                  * called with interrupts disabled.
 580                  */
 581                  local_irq_disable();
 582                  clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_EXIT, &cpu);
 583                  local_irq_enable();
 584         } else
 585                 default_idle();
 586 }
 587
 588 void __cpuinit select_idle_routine(const struct cpuinfo_x86 *c)
 589 {
 590 #ifdef CONFIG_SMP
 591         if (pm_idle == poll_idle && smp_num_siblings > 1) {
 592                 printk_once(KERN_WARNING "WARNING: polling idle and HT enabled,"
 593                         " performance may degrade.\n");
 594         }
 595 #endif
 596         if (pm_idle)
 597                 return;
 598
 599         if (cpu_has(c, X86_FEATURE_MWAIT) && mwait_usable(c)) {
 600                 /*
 601                  * One CPU supports mwait => All CPUs supports mwait
 602                  */
 603                 printk(KERN_INFO "using mwait in idle threads.\n");
 604                 pm_idle = mwait_idle;
 605         } else if (cpu_has_amd_erratum(amd_erratum_400)) {
 606                 /* E400: APIC timer interrupt does not wake up CPU from C1e */
 607                 printk(KERN_INFO "using C1E aware idle routine\n");
 608                 pm_idle = c1e_idle;
 609         } else
 610                 pm_idle = default_idle;
 611 }
 612
 613 void __init init_c1e_mask(void)
 614 {
 615         /* If we're using c1e_idle, we need to allocate c1e_mask. */
 616         if (pm_idle == c1e_idle)
 617                 zalloc_cpumask_var(&c1e_mask, GFP_KERNEL);
 618 }
 619
 620 static int __init idle_setup(char *str)
 621 {
 622         if (!str)
 623                 return -EINVAL;
 624
 625         if (!strcmp(str, "poll")) {
 626                 printk("using polling idle threads.\n");
 627                 pm_idle = poll_idle;
 628         } else if (!strcmp(str, "mwait"))
 629                 force_mwait = 1;
 630         else if (!strcmp(str, "halt")) {
 631                 /*
 632                  * When the boot option of idle=halt is added, halt is
 633                  * forced to be used for CPU idle. In such case CPU C2/C3
 634                  * won't be used again.
 635                  * To continue to load the CPU idle driver, don't touch
 636                  * the boot_option_idle_override.
 637                  */
 638                 pm_idle = default_idle;
 639                 idle_halt = 1;
 640                 return 0;
 641         } else if (!strcmp(str, "nomwait")) {
 642                 /*
 643                  * If the boot option of "idle=nomwait" is added,
 644                  * it means that mwait will be disabled for CPU C2/C3
 645                  * states. In such case it won't touch the variable
 646                  * of boot_option_idle_override.
 647                  */
 648                 idle_nomwait = 1;
 649                 return 0;
 650         } else
 651                 return -1;
 652
 653         boot_option_idle_override = 1;
 654         return 0;
 655 }
 656 early_param("idle", idle_setup);
 657
 658 unsigned long arch_align_stack(unsigned long sp)
 659 {
 660         if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
 661                 sp -= get_random_int() % 8192;
 662         return sp & ~0xf;
 663 }
 664
 665 unsigned long arch_randomize_brk(struct mm_struct *mm)
 666 {
 667         unsigned long range_end = mm->brk + 0x02000000;
 668         return randomize_range(mm->brk, range_end, 0) ? : mm->brk;
 669 }
 670