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 struct kmem_cache *task_xstate_cachep;
  26 EXPORT_SYMBOL_GPL(task_xstate_cachep);
  27
  28 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
  29 {
  30         int ret;
  31
  32         *dst = *src;
  33         if (fpu_allocated(&src->thread.fpu)) {
  34                 memset(&dst->thread.fpu, 0, sizeof(dst->thread.fpu));
  35                 ret = fpu_alloc(&dst->thread.fpu);
  36                 if (ret)
  37                         return ret;
  38                 fpu_copy(&dst->thread.fpu, &src->thread.fpu);
  39         }
  40         return 0;
  41 }
  42
  43 void free_thread_xstate(struct task_struct *tsk)
  44 {
  45         fpu_free(&tsk->thread.fpu);
  46 }
  47
  48 void free_thread_info(struct thread_info *ti)
  49 {
  50         free_thread_xstate(ti->task);
  51         free_pages((unsigned long)ti, get_order(THREAD_SIZE));
  52 }
  53
  54 void arch_task_cache_init(void)
  55 {
  56         task_xstate_cachep =
  57                 kmem_cache_create("task_xstate", xstate_size,
  58                                   __alignof__(union thread_xstate),
  59                                   SLAB_PANIC | SLAB_NOTRACK, NULL);
  60 }
  61
  62 /*
  63  * Free current thread data structures etc..
  64  */
  65 void exit_thread(void)
  66 {
  67         struct task_struct *me = current;
  68         struct thread_struct *t = &me->thread;
  69         unsigned long *bp = t->io_bitmap_ptr;
  70
  71         if (bp) {
  72                 struct tss_struct *tss = &per_cpu(init_tss, get_cpu());
  73
  74                 t->io_bitmap_ptr = NULL;
  75                 clear_thread_flag(TIF_IO_BITMAP);
  76                 /*
  77                  * Careful, clear this in the TSS too:
  78                  */
  79                 memset(tss->io_bitmap, 0xff, t->io_bitmap_max);
  80                 t->io_bitmap_max = 0;
  81                 put_cpu();
  82                 kfree(bp);
  83         }
  84 }
  85
  86 void show_regs(struct pt_regs *regs)
  87 {
  88         show_registers(regs);
  89         show_trace(NULL, regs, (unsigned long *)kernel_stack_pointer(regs));
  90 }
  91
  92 void show_regs_common(void)
  93 {
  94         const char *board, *product;
  95
  96         board = dmi_get_system_info(DMI_BOARD_NAME);
  97         if (!board)
  98                 board = "";
  99         product = dmi_get_system_info(DMI_PRODUCT_NAME);
 100         if (!product)
 101                 product = "";
 102
 103         printk(KERN_CONT "\n");
 104         printk(KERN_DEFAULT "Pid: %d, comm: %.20s %s %s %.*s %s/%s\n",
 105                 current->pid, current->comm, print_tainted(),
 106                 init_utsname()->release,
 107                 (int)strcspn(init_utsname()->version, " "),
 108                 init_utsname()->version, board, product);
 109 }
 110
 111 void flush_thread(void)
 112 {
 113         struct task_struct *tsk = current;
 114
 115         flush_ptrace_hw_breakpoint(tsk);
 116         memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
 117         /*
 118          * Forget coprocessor state..
 119          */
 120         tsk->fpu_counter = 0;
 121         clear_fpu(tsk);
 122         clear_used_math();
 123 }
 124
 125 static void hard_disable_TSC(void)
 126 {
 127         write_cr4(read_cr4() | X86_CR4_TSD);
 128 }
 129
 130 void disable_TSC(void)
 131 {
 132         preempt_disable();
 133         if (!test_and_set_thread_flag(TIF_NOTSC))
 134                 /*
 135                  * Must flip the CPU state synchronously with
 136                  * TIF_NOTSC in the current running context.
 137                  */
 138                 hard_disable_TSC();
 139         preempt_enable();
 140 }
 141
 142 static void hard_enable_TSC(void)
 143 {
 144         write_cr4(read_cr4() & ~X86_CR4_TSD);
 145 }
 146
 147 static void enable_TSC(void)
 148 {
 149         preempt_disable();
 150         if (test_and_clear_thread_flag(TIF_NOTSC))
 151                 /*
 152                  * Must flip the CPU state synchronously with
 153                  * TIF_NOTSC in the current running context.
 154                  */
 155                 hard_enable_TSC();
 156         preempt_enable();
 157 }
 158
 159 int get_tsc_mode(unsigned long adr)
 160 {
 161         unsigned int val;
 162
 163         if (test_thread_flag(TIF_NOTSC))
 164                 val = PR_TSC_SIGSEGV;
 165         else
 166                 val = PR_TSC_ENABLE;
 167
 168         return put_user(val, (unsigned int __user *)adr);
 169 }
 170
 171 int set_tsc_mode(unsigned int val)
 172 {
 173         if (val == PR_TSC_SIGSEGV)
 174                 disable_TSC();
 175         else if (val == PR_TSC_ENABLE)
 176                 enable_TSC();
 177         else
 178                 return -EINVAL;
 179
 180         return 0;
 181 }
 182
 183 void __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p,
 184                       struct tss_struct *tss)
 185 {
 186         struct thread_struct *prev, *next;
 187
 188         prev = &prev_p->thread;
 189         next = &next_p->thread;
 190
 191         if (test_tsk_thread_flag(prev_p, TIF_BLOCKSTEP) ^
 192             test_tsk_thread_flag(next_p, TIF_BLOCKSTEP)) {
 193                 unsigned long debugctl = get_debugctlmsr();
 194
 195                 debugctl &= ~DEBUGCTLMSR_BTF;
 196                 if (test_tsk_thread_flag(next_p, TIF_BLOCKSTEP))
 197                         debugctl |= DEBUGCTLMSR_BTF;
 198
 199                 update_debugctlmsr(debugctl);
 200         }
 201
 202         if (test_tsk_thread_flag(prev_p, TIF_NOTSC) ^
 203             test_tsk_thread_flag(next_p, TIF_NOTSC)) {
 204                 /* prev and next are different */
 205                 if (test_tsk_thread_flag(next_p, TIF_NOTSC))
 206                         hard_disable_TSC();
 207                 else
 208                         hard_enable_TSC();
 209         }
 210
 211         if (test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
 212                 /*
 213                  * Copy the relevant range of the IO bitmap.
 214                  * Normally this is 128 bytes or less:
 215                  */
 216                 memcpy(tss->io_bitmap, next->io_bitmap_ptr,
 217                        max(prev->io_bitmap_max, next->io_bitmap_max));
 218         } else if (test_tsk_thread_flag(prev_p, TIF_IO_BITMAP)) {
 219                 /*
 220                  * Clear any possible leftover bits:
 221                  */
 222                 memset(tss->io_bitmap, 0xff, prev->io_bitmap_max);
 223         }
 224         propagate_user_return_notify(prev_p, next_p);
 225 }
 226
 227 int sys_fork(struct pt_regs *regs)
 228 {
 229         return do_fork(SIGCHLD, regs->sp, regs, 0, NULL, NULL);
 230 }
 231
 232 /*
 233  * This is trivial, and on the face of it looks like it
 234  * could equally well be done in user mode.
 235  *
 236  * Not so, for quite unobvious reasons - register pressure.
 237  * In user mode vfork() cannot have a stack frame, and if
 238  * done by calling the "clone()" system call directly, you
 239  * do not have enough call-clobbered registers to hold all
 240  * the information you need.
 241  */
 242 int sys_vfork(struct pt_regs *regs)
 243 {
 244         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->sp, regs, 0,
 245                        NULL, NULL);
 246 }
 247
 248 long
 249 sys_clone(unsigned long clone_flags, unsigned long newsp,
 250           void __user *parent_tid, void __user *child_tid, struct pt_regs *regs)
 251 {
 252         if (!newsp)
 253                 newsp = regs->sp;
 254         return do_fork(clone_flags, newsp, regs, 0, parent_tid, child_tid);
 255 }
 256
 257 /*
 258  * This gets run with %si containing the
 259  * function to call, and %di containing
 260  * the "args".
 261  */
 262 extern void kernel_thread_helper(void);
 263
 264 /*
 265  * Create a kernel thread
 266  */
 267 int kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
 268 {
 269         struct pt_regs regs;
 270
 271         memset(&regs, 0, sizeof(regs));
 272
 273         regs.si = (unsigned long) fn;
 274         regs.di = (unsigned long) arg;
 275
 276 #ifdef CONFIG_X86_32
 277         regs.ds = __USER_DS;
 278         regs.es = __USER_DS;
 279         regs.fs = __KERNEL_PERCPU;
 280         regs.gs = __KERNEL_STACK_CANARY;
 281 #else
 282         regs.ss = __KERNEL_DS;
 283 #endif
 284
 285         regs.orig_ax = -1;
 286         regs.ip = (unsigned long) kernel_thread_helper;
 287         regs.cs = __KERNEL_CS | get_kernel_rpl();
 288         regs.flags = X86_EFLAGS_IF | 0x2;
 289
 290         /* Ok, create the new process.. */
 291         return do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0, &regs, 0, NULL, NULL);
 292 }
 293 EXPORT_SYMBOL(kernel_thread);
 294
 295 /*
 296  * sys_execve() executes a new program.
 297  */
 298 long sys_execve(const char __user *name,
 299                 const char __user *const __user *argv,
 300                 const char __user *const __user *envp, struct pt_regs *regs)
 301 {
 302         long error;
 303         char *filename;
 304
 305         filename = getname(name);
 306         error = PTR_ERR(filename);
 307         if (IS_ERR(filename))
 308                 return error;
 309         error = do_execve(filename, argv, envp, regs);
 310
 311 #ifdef CONFIG_X86_32
 312         if (error == 0) {
 313                 /* Make sure we don't return using sysenter.. */
 314                 set_thread_flag(TIF_IRET);
 315         }
 316 #endif
 317
 318         putname(filename);
 319         return error;
 320 }
 321
 322 /*
 323  * Idle related variables and functions
 324  */
 325 unsigned long boot_option_idle_override = IDLE_NO_OVERRIDE;
 326 EXPORT_SYMBOL(boot_option_idle_override);
 327
 328 /*
 329  * Powermanagement idle function, if any..
 330  */
 331 void (*pm_idle)(void);
 332 EXPORT_SYMBOL(pm_idle);
 333
 334 #ifdef CONFIG_X86_32
 335 /*
 336  * This halt magic was a workaround for ancient floppy DMA
 337  * wreckage. It should be safe to remove.
 338  */
 339 static int hlt_counter;
 340 void disable_hlt(void)
 341 {
 342         hlt_counter++;
 343 }
 344 EXPORT_SYMBOL(disable_hlt);
 345
 346 void enable_hlt(void)
 347 {
 348         hlt_counter--;
 349 }
 350 EXPORT_SYMBOL(enable_hlt);
 351
 352 static inline int hlt_use_halt(void)
 353 {
 354         return (!hlt_counter && boot_cpu_data.hlt_works_ok);
 355 }
 356 #else
 357 static inline int hlt_use_halt(void)
 358 {
 359         return 1;
 360 }
 361 #endif
 362
 363 /*
 364  * We use this if we don't have any better
 365  * idle routine..
 366  */
 367 void default_idle(void)
 368 {
 369         if (hlt_use_halt()) {
 370                 trace_power_start(POWER_CSTATE, 1, smp_processor_id());
 371                 trace_cpu_idle(1, smp_processor_id());
 372                 current_thread_info()->status &= ~TS_POLLING;
 373                 /*
 374                  * TS_POLLING-cleared state must be visible before we
 375                  * test NEED_RESCHED:
 376                  */
 377                 smp_mb();
 378
 379                 if (!need_resched())
 380                         safe_halt();    /* enables interrupts racelessly */
 381                 else
 382                         local_irq_enable();
 383                 current_thread_info()->status |= TS_POLLING;
 384         } else {
 385                 local_irq_enable();
 386                 /* loop is done by the caller */
 387                 cpu_relax();
 388         }
 389 }
 390 #ifdef CONFIG_APM_MODULE
 391 EXPORT_SYMBOL(default_idle);
 392 #endif
 393
 394 void stop_this_cpu(void *dummy)
 395 {
 396         local_irq_disable();
 397         /*
 398          * Remove this CPU:
 399          */
 400         set_cpu_online(smp_processor_id(), false);
 401         disable_local_APIC();
 402
 403         for (;;) {
 404                 if (hlt_works(smp_processor_id()))
 405                         halt();
 406         }
 407 }
 408
 409 static void do_nothing(void *unused)
 410 {
 411 }
 412
 413 /*
 414  * cpu_idle_wait - Used to ensure that all the CPUs discard old value of
 415  * pm_idle and update to new pm_idle value. Required while changing pm_idle
 416  * handler on SMP systems.
 417  *
 418  * Caller must have changed pm_idle to the new value before the call. Old
 419  * pm_idle value will not be used by any CPU after the return of this function.
 420  */
 421 void cpu_idle_wait(void)
 422 {
 423         smp_mb();
 424         /* kick all the CPUs so that they exit out of pm_idle */
 425         smp_call_function(do_nothing, NULL, 1);
 426 }
 427 EXPORT_SYMBOL_GPL(cpu_idle_wait);
 428
 429 /*
 430  * This uses new MONITOR/MWAIT instructions on P4 processors with PNI,
 431  * which can obviate IPI to trigger checking of need_resched.
 432  * We execute MONITOR against need_resched and enter optimized wait state
 433  * through MWAIT. Whenever someone changes need_resched, we would be woken
 434  * up from MWAIT (without an IPI).
 435  *
 436  * New with Core Duo processors, MWAIT can take some hints based on CPU
 437  * capability.
 438  */
 439 void mwait_idle_with_hints(unsigned long ax, unsigned long cx)
 440 {
 441         trace_power_start(POWER_CSTATE, (ax>>4)+1, smp_processor_id());
 442         trace_cpu_idle((ax>>4)+1, smp_processor_id());
 443         if (!need_resched()) {
 444                 if (cpu_has(__this_cpu_ptr(&cpu_info), X86_FEATURE_CLFLUSH_MONITOR))
 445                         clflush((void *)&current_thread_info()->flags);
 446
 447                 __monitor((void *)&current_thread_info()->flags, 0, 0);
 448                 smp_mb();
 449                 if (!need_resched())
 450                         __mwait(ax, cx);
 451         }
 452 }
 453
 454 /* Default MONITOR/MWAIT with no hints, used for default C1 state */
 455 static void mwait_idle(void)
 456 {
 457         if (!need_resched()) {
 458                 trace_power_start(POWER_CSTATE, 1, smp_processor_id());
 459                 trace_cpu_idle(1, smp_processor_id());
 460                 if (cpu_has(__this_cpu_ptr(&cpu_info), X86_FEATURE_CLFLUSH_MONITOR))
 461                         clflush((void *)&current_thread_info()->flags);
 462
 463                 __monitor((void *)&current_thread_info()->flags, 0, 0);
 464                 smp_mb();
 465                 if (!need_resched())
 466                         __sti_mwait(0, 0);
 467                 else
 468                         local_irq_enable();
 469         } else
 470                 local_irq_enable();
 471 }
 472
 473 /*
 474  * On SMP it's slightly faster (but much more power-consuming!)
 475  * to poll the ->work.need_resched flag instead of waiting for the
 476  * cross-CPU IPI to arrive. Use this option with caution.
 477  */
 478 static void poll_idle(void)
 479 {
 480         trace_power_start(POWER_CSTATE, 0, smp_processor_id());
 481         trace_cpu_idle(0, smp_processor_id());
 482         local_irq_enable();
 483         while (!need_resched())
 484                 cpu_relax();
 485         trace_power_end(smp_processor_id());
 486         trace_cpu_idle(PWR_EVENT_EXIT, smp_processor_id());
 487 }
 488
 489 /*
 490  * mwait selection logic:
 491  *
 492  * It depends on the CPU. For AMD CPUs that support MWAIT this is
 493  * wrong. Family 0x10 and 0x11 CPUs will enter C1 on HLT. Powersavings
 494  * then depend on a clock divisor and current Pstate of the core. If
 495  * all cores of a processor are in halt state (C1) the processor can
 496  * enter the C1E (C1 enhanced) state. If mwait is used this will never
 497  * happen.
 498  *
 499  * idle=mwait overrides this decision and forces the usage of mwait.
 500  */
 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 (boot_option_idle_override == IDLE_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 bool c1e_detected;
 529 EXPORT_SYMBOL(c1e_detected);
 530
 531 static cpumask_var_t c1e_mask;
 532
 533 void c1e_remove_cpu(int cpu)
 534 {
 535         if (c1e_mask != NULL)
 536                 cpumask_clear_cpu(cpu, c1e_mask);
 537 }
 538
 539 /*
 540  * C1E aware idle routine. We check for C1E active in the interrupt
 541  * pending message MSR. If we detect C1E, then we handle it the same
 542  * way as C3 power states (local apic timer and TSC stop)
 543  */
 544 static void c1e_idle(void)
 545 {
 546         if (need_resched())
 547                 return;
 548
 549         if (!c1e_detected) {
 550                 u32 lo, hi;
 551
 552                 rdmsr(MSR_K8_INT_PENDING_MSG, lo, hi);
 553
 554                 if (lo & K8_INTP_C1E_ACTIVE_MASK) {
 555                         c1e_detected = true;
 556                         if (!boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
 557                                 mark_tsc_unstable("TSC halt in AMD C1E");
 558                         printk(KERN_INFO "System has AMD C1E enabled\n");
 559                 }
 560         }
 561
 562         if (c1e_detected) {
 563                 int cpu = smp_processor_id();
 564
 565                 if (!cpumask_test_cpu(cpu, c1e_mask)) {
 566                         cpumask_set_cpu(cpu, c1e_mask);
 567                         /*
 568                          * Force broadcast so ACPI can not interfere.
 569                          */
 570                         clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_FORCE,
 571                                            &cpu);
 572                         printk(KERN_INFO "Switch to broadcast mode on CPU%d\n",
 573                                cpu);
 574                 }
 575                 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER, &cpu);
 576
 577                 default_idle();
 578
 579                 /*
 580                  * The switch back from broadcast mode needs to be
 581                  * called with interrupts disabled.
 582                  */
 583                  local_irq_disable();
 584                  clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_EXIT, &cpu);
 585                  local_irq_enable();
 586         } else
 587                 default_idle();
 588 }
 589
 590 void __cpuinit select_idle_routine(const struct cpuinfo_x86 *c)
 591 {
 592 #ifdef CONFIG_SMP
 593         if (pm_idle == poll_idle && smp_num_siblings > 1) {
 594                 printk_once(KERN_WARNING "WARNING: polling idle and HT enabled,"
 595                         " performance may degrade.\n");
 596         }
 597 #endif
 598         if (pm_idle)
 599                 return;
 600
 601         if (cpu_has(c, X86_FEATURE_MWAIT) && mwait_usable(c)) {
 602                 /*
 603                  * One CPU supports mwait => All CPUs supports mwait
 604                  */
 605                 printk(KERN_INFO "using mwait in idle threads.\n");
 606                 pm_idle = mwait_idle;
 607         } else if (cpu_has_amd_erratum(amd_erratum_400)) {
 608                 /* E400: APIC timer interrupt does not wake up CPU from C1e */
 609                 printk(KERN_INFO "using C1E aware idle routine\n");
 610                 pm_idle = c1e_idle;
 611         } else
 612                 pm_idle = default_idle;
 613 }
 614
 615 void __init init_c1e_mask(void)
 616 {
 617         /* If we're using c1e_idle, we need to allocate c1e_mask. */
 618         if (pm_idle == c1e_idle)
 619                 zalloc_cpumask_var(&c1e_mask, GFP_KERNEL);
 620 }
 621
 622 static int __init idle_setup(char *str)
 623 {
 624         if (!str)
 625                 return -EINVAL;
 626
 627         if (!strcmp(str, "poll")) {
 628                 printk("using polling idle threads.\n");
 629                 pm_idle = poll_idle;
 630                 boot_option_idle_override = IDLE_POLL;
 631         } else if (!strcmp(str, "mwait")) {
 632                 boot_option_idle_override = IDLE_FORCE_MWAIT;
 633         } else if (!strcmp(str, "halt")) {
 634                 /*
 635                  * When the boot option of idle=halt is added, halt is
 636                  * forced to be used for CPU idle. In such case CPU C2/C3
 637                  * won't be used again.
 638                  * To continue to load the CPU idle driver, don't touch
 639                  * the boot_option_idle_override.
 640                  */
 641                 pm_idle = default_idle;
 642                 boot_option_idle_override = IDLE_HALT;
 643         } else if (!strcmp(str, "nomwait")) {
 644                 /*
 645                  * If the boot option of "idle=nomwait" is added,
 646                  * it means that mwait will be disabled for CPU C2/C3
 647                  * states. In such case it won't touch the variable
 648                  * of boot_option_idle_override.
 649                  */
 650                 boot_option_idle_override = IDLE_NOMWAIT;
 651         } else
 652                 return -1;
 653
 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