arch/x86/kernel/process_32.c

   1 /*
   2  *  Copyright (C) 1995  Linus Torvalds
   3  *
   4  *  Pentium III FXSR, SSE support
   5  *      Gareth Hughes <gareth@valinux.com>, May 2000
   6  */
   7
   8 /*
   9  * This file handles the architecture-dependent parts of process handling..
  10  */
  11
  12 #include <stdarg.h>
  13
  14 #include <linux/cpu.h>
  15 #include <linux/errno.h>
  16 #include <linux/sched.h>
  17 #include <linux/fs.h>
  18 #include <linux/kernel.h>
  19 #include <linux/mm.h>
  20 #include <linux/elfcore.h>
  21 #include <linux/smp.h>
  22 #include <linux/stddef.h>
  23 #include <linux/slab.h>
  24 #include <linux/vmalloc.h>
  25 #include <linux/user.h>
  26 #include <linux/a.out.h>
  27 #include <linux/interrupt.h>
  28 #include <linux/utsname.h>
  29 #include <linux/delay.h>
  30 #include <linux/reboot.h>
  31 #include <linux/init.h>
  32 #include <linux/mc146818rtc.h>
  33 #include <linux/module.h>
  34 #include <linux/kallsyms.h>
  35 #include <linux/ptrace.h>
  36 #include <linux/random.h>
  37 #include <linux/personality.h>
  38 #include <linux/tick.h>
  39 #include <linux/percpu.h>
  40
  41 #include <asm/uaccess.h>
  42 #include <asm/pgtable.h>
  43 #include <asm/system.h>
  44 #include <asm/io.h>
  45 #include <asm/ldt.h>
  46 #include <asm/processor.h>
  47 #include <asm/i387.h>
  48 #include <asm/desc.h>
  49 #include <asm/vm86.h>
  50 #ifdef CONFIG_MATH_EMULATION
  51 #include <asm/math_emu.h>
  52 #endif
  53
  54 #include <linux/err.h>
  55
  56 #include <asm/tlbflush.h>
  57 #include <asm/cpu.h>
  58 #include <asm/kdebug.h>
  59
  60 asmlinkage void ret_from_fork(void) __asm__("ret_from_fork");
  61
  62 static int hlt_counter;
  63
  64 unsigned long boot_option_idle_override = 0;
  65 EXPORT_SYMBOL(boot_option_idle_override);
  66
  67 DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task;
  68 EXPORT_PER_CPU_SYMBOL(current_task);
  69
  70 DEFINE_PER_CPU(int, cpu_number);
  71 EXPORT_PER_CPU_SYMBOL(cpu_number);
  72
  73 /*
  74  * Return saved PC of a blocked thread.
  75  */
  76 unsigned long thread_saved_pc(struct task_struct *tsk)
  77 {
  78         return ((unsigned long *)tsk->thread.sp)[3];
  79 }
  80
  81 /*
  82  * Powermanagement idle function, if any..
  83  */
  84 void (*pm_idle)(void);
  85 EXPORT_SYMBOL(pm_idle);
  86 static DEFINE_PER_CPU(unsigned int, cpu_idle_state);
  87
  88 void disable_hlt(void)
  89 {
  90         hlt_counter++;
  91 }
  92
  93 EXPORT_SYMBOL(disable_hlt);
  94
  95 void enable_hlt(void)
  96 {
  97         hlt_counter--;
  98 }
  99
 100 EXPORT_SYMBOL(enable_hlt);
 101
 102 /*
 103  * We use this if we don't have any better
 104  * idle routine..
 105  */
 106 void default_idle(void)
 107 {
 108         if (!hlt_counter && boot_cpu_data.hlt_works_ok) {
 109                 current_thread_info()->status &= ~TS_POLLING;
 110                 /*
 111                  * TS_POLLING-cleared state must be visible before we
 112                  * test NEED_RESCHED:
 113                  */
 114                 smp_mb();
 115
 116                 local_irq_disable();
 117                 if (!need_resched()) {
 118                         ktime_t t0, t1;
 119                         u64 t0n, t1n;
 120
 121                         t0 = ktime_get();
 122                         t0n = ktime_to_ns(t0);
 123                         safe_halt();    /* enables interrupts racelessly */
 124                         local_irq_disable();
 125                         t1 = ktime_get();
 126                         t1n = ktime_to_ns(t1);
 127                         sched_clock_idle_wakeup_event(t1n - t0n);
 128                 }
 129                 local_irq_enable();
 130                 current_thread_info()->status |= TS_POLLING;
 131         } else {
 132                 /* loop is done by the caller */
 133                 cpu_relax();
 134         }
 135 }
 136 #ifdef CONFIG_APM_MODULE
 137 EXPORT_SYMBOL(default_idle);
 138 #endif
 139
 140 /*
 141  * On SMP it's slightly faster (but much more power-consuming!)
 142  * to poll the ->work.need_resched flag instead of waiting for the
 143  * cross-CPU IPI to arrive. Use this option with caution.
 144  */
 145 static void poll_idle(void)
 146 {
 147         cpu_relax();
 148 }
 149
 150 #ifdef CONFIG_HOTPLUG_CPU
 151 #include <asm/nmi.h>
 152 /* We don't actually take CPU down, just spin without interrupts. */
 153 static inline void play_dead(void)
 154 {
 155         /* This must be done before dead CPU ack */
 156         cpu_exit_clear();
 157         wbinvd();
 158         mb();
 159         /* Ack it */
 160         __get_cpu_var(cpu_state) = CPU_DEAD;
 161
 162         /*
 163          * With physical CPU hotplug, we should halt the cpu
 164          */
 165         local_irq_disable();
 166         while (1)
 167                 halt();
 168 }
 169 #else
 170 static inline void play_dead(void)
 171 {
 172         BUG();
 173 }
 174 #endif /* CONFIG_HOTPLUG_CPU */
 175
 176 /*
 177  * The idle thread. There's no useful work to be
 178  * done, so just try to conserve power and have a
 179  * low exit latency (ie sit in a loop waiting for
 180  * somebody to say that they'd like to reschedule)
 181  */
 182 void cpu_idle(void)
 183 {
 184         int cpu = smp_processor_id();
 185
 186         current_thread_info()->status |= TS_POLLING;
 187
 188         /* endless idle loop with no priority at all */
 189         while (1) {
 190                 tick_nohz_stop_sched_tick();
 191                 while (!need_resched()) {
 192                         void (*idle)(void);
 193
 194                         if (__get_cpu_var(cpu_idle_state))
 195                                 __get_cpu_var(cpu_idle_state) = 0;
 196
 197                         check_pgt_cache();
 198                         rmb();
 199                         idle = pm_idle;
 200
 201                         if (rcu_pending(cpu))
 202                                 rcu_check_callbacks(cpu, 0);
 203
 204                         if (!idle)
 205                                 idle = default_idle;
 206
 207                         if (cpu_is_offline(cpu))
 208                                 play_dead();
 209
 210                         __get_cpu_var(irq_stat).idle_timestamp = jiffies;
 211                         idle();
 212                 }
 213                 tick_nohz_restart_sched_tick();
 214                 preempt_enable_no_resched();
 215                 schedule();
 216                 preempt_disable();
 217         }
 218 }
 219
 220 static void do_nothing(void *unused)
 221 {
 222 }
 223
 224 void cpu_idle_wait(void)
 225 {
 226         unsigned int cpu, this_cpu = get_cpu();
 227         cpumask_t map, tmp = current->cpus_allowed;
 228
 229         set_cpus_allowed(current, cpumask_of_cpu(this_cpu));
 230         put_cpu();
 231
 232         cpus_clear(map);
 233         for_each_online_cpu(cpu) {
 234                 per_cpu(cpu_idle_state, cpu) = 1;
 235                 cpu_set(cpu, map);
 236         }
 237
 238         __get_cpu_var(cpu_idle_state) = 0;
 239
 240         wmb();
 241         do {
 242                 ssleep(1);
 243                 for_each_online_cpu(cpu) {
 244                         if (cpu_isset(cpu, map) && !per_cpu(cpu_idle_state, cpu))
 245                                 cpu_clear(cpu, map);
 246                 }
 247                 cpus_and(map, map, cpu_online_map);
 248                 /*
 249                  * We waited 1 sec, if a CPU still did not call idle
 250                  * it may be because it is in idle and not waking up
 251                  * because it has nothing to do.
 252                  * Give all the remaining CPUS a kick.
 253                  */
 254                 smp_call_function_mask(map, do_nothing, 0, 0);
 255         } while (!cpus_empty(map));
 256
 257         set_cpus_allowed(current, tmp);
 258 }
 259 EXPORT_SYMBOL_GPL(cpu_idle_wait);
 260
 261 /*
 262  * This uses new MONITOR/MWAIT instructions on P4 processors with PNI,
 263  * which can obviate IPI to trigger checking of need_resched.
 264  * We execute MONITOR against need_resched and enter optimized wait state
 265  * through MWAIT. Whenever someone changes need_resched, we would be woken
 266  * up from MWAIT (without an IPI).
 267  *
 268  * New with Core Duo processors, MWAIT can take some hints based on CPU
 269  * capability.
 270  */
 271 void mwait_idle_with_hints(unsigned long ax, unsigned long cx)
 272 {
 273         if (!need_resched()) {
 274                 __monitor((void *)&current_thread_info()->flags, 0, 0);
 275                 smp_mb();
 276                 if (!need_resched())
 277                         __mwait(ax, cx);
 278         }
 279 }
 280
 281 /* Default MONITOR/MWAIT with no hints, used for default C1 state */
 282 static void mwait_idle(void)
 283 {
 284         local_irq_enable();
 285         mwait_idle_with_hints(0, 0);
 286 }
 287
 288 static int mwait_usable(const struct cpuinfo_x86 *c)
 289 {
 290         if (force_mwait)
 291                 return 1;
 292         /* Any C1 states supported? */
 293         return c->cpuid_level >= 5 && ((cpuid_edx(5) >> 4) & 0xf) > 0;
 294 }
 295
 296 void __cpuinit select_idle_routine(const struct cpuinfo_x86 *c)
 297 {
 298         static int selected;
 299
 300         if (selected)
 301                 return;
 302 #ifdef CONFIG_X86_SMP
 303         if (pm_idle == poll_idle && smp_num_siblings > 1) {
 304                 printk(KERN_WARNING "WARNING: polling idle and HT enabled,"
 305                         " performance may degrade.\n");
 306         }
 307 #endif
 308         if (cpu_has(c, X86_FEATURE_MWAIT) && mwait_usable(c)) {
 309                 /*
 310                  * Skip, if setup has overridden idle.
 311                  * One CPU supports mwait => All CPUs supports mwait
 312                  */
 313                 if (!pm_idle) {
 314                         printk(KERN_INFO "using mwait in idle threads.\n");
 315                         pm_idle = mwait_idle;
 316                 }
 317         }
 318         selected = 1;
 319 }
 320
 321 static int __init idle_setup(char *str)
 322 {
 323         if (!strcmp(str, "poll")) {
 324                 printk("using polling idle threads.\n");
 325                 pm_idle = poll_idle;
 326         } else if (!strcmp(str, "mwait"))
 327                 force_mwait = 1;
 328         else
 329                 return -1;
 330
 331         boot_option_idle_override = 1;
 332         return 0;
 333 }
 334 early_param("idle", idle_setup);
 335
 336 void __show_registers(struct pt_regs *regs, int all)
 337 {
 338         unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L;
 339         unsigned long d0, d1, d2, d3, d6, d7;
 340         unsigned long sp;
 341         unsigned short ss, gs;
 342
 343         if (user_mode_vm(regs)) {
 344                 sp = regs->sp;
 345                 ss = regs->ss & 0xffff;
 346                 savesegment(gs, gs);
 347         } else {
 348                 sp = (unsigned long) (&regs->sp);
 349                 savesegment(ss, ss);
 350                 savesegment(gs, gs);
 351         }
 352
 353         printk("\n");
 354         printk("Pid: %d, comm: %s %s (%s %.*s)\n",
 355                         task_pid_nr(current), current->comm,
 356                         print_tainted(), init_utsname()->release,
 357                         (int)strcspn(init_utsname()->version, " "),
 358                         init_utsname()->version);
 359
 360         printk("EIP: %04x:[<%08lx>] EFLAGS: %08lx CPU: %d\n",
 361                         0xffff & regs->cs, regs->ip, regs->flags,
 362                         smp_processor_id());
 363         print_symbol("EIP is at %s\n", regs->ip);
 364
 365         printk("EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
 366                 regs->ax, regs->bx, regs->cx, regs->dx);
 367         printk("ESI: %08lx EDI: %08lx EBP: %08lx ESP: %08lx\n",
 368                 regs->si, regs->di, regs->bp, sp);
 369         printk(" DS: %04x ES: %04x FS: %04x GS: %04x SS: %04x\n",
 370                regs->ds & 0xffff, regs->es & 0xffff,
 371                regs->fs & 0xffff, gs, ss);
 372
 373         if (!all)
 374                 return;
 375
 376         cr0 = read_cr0();
 377         cr2 = read_cr2();
 378         cr3 = read_cr3();
 379         cr4 = read_cr4_safe();
 380         printk("CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n",
 381                         cr0, cr2, cr3, cr4);
 382
 383         get_debugreg(d0, 0);
 384         get_debugreg(d1, 1);
 385         get_debugreg(d2, 2);
 386         get_debugreg(d3, 3);
 387         printk("DR0: %08lx DR1: %08lx DR2: %08lx DR3: %08lx\n",
 388                         d0, d1, d2, d3);
 389
 390         get_debugreg(d6, 6);
 391         get_debugreg(d7, 7);
 392         printk("DR6: %08lx DR7: %08lx\n",
 393                         d6, d7);
 394 }
 395
 396 void show_regs(struct pt_regs *regs)
 397 {
 398         __show_registers(regs, 1);
 399         show_trace(NULL, regs, &regs->sp, regs->bp);
 400 }
 401
 402 /*
 403  * This gets run with %bx containing the
 404  * function to call, and %dx containing
 405  * the "args".
 406  */
 407 extern void kernel_thread_helper(void);
 408
 409 /*
 410  * Create a kernel thread
 411  */
 412 int kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
 413 {
 414         struct pt_regs regs;
 415
 416         memset(&regs, 0, sizeof(regs));
 417
 418         regs.bx = (unsigned long) fn;
 419         regs.dx = (unsigned long) arg;
 420
 421         regs.ds = __USER_DS;
 422         regs.es = __USER_DS;
 423         regs.fs = __KERNEL_PERCPU;
 424         regs.orig_ax = -1;
 425         regs.ip = (unsigned long) kernel_thread_helper;
 426         regs.cs = __KERNEL_CS | get_kernel_rpl();
 427         regs.flags = X86_EFLAGS_IF | X86_EFLAGS_SF | X86_EFLAGS_PF | 0x2;
 428
 429         /* Ok, create the new process.. */
 430         return do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0, &regs, 0, NULL, NULL);
 431 }
 432 EXPORT_SYMBOL(kernel_thread);
 433
 434 /*
 435  * Free current thread data structures etc..
 436  */
 437 void exit_thread(void)
 438 {
 439         /* The process may have allocated an io port bitmap... nuke it. */
 440         if (unlikely(test_thread_flag(TIF_IO_BITMAP))) {
 441                 struct task_struct *tsk = current;
 442                 struct thread_struct *t = &tsk->thread;
 443                 int cpu = get_cpu();
 444                 struct tss_struct *tss = &per_cpu(init_tss, cpu);
 445
 446                 kfree(t->io_bitmap_ptr);
 447                 t->io_bitmap_ptr = NULL;
 448                 clear_thread_flag(TIF_IO_BITMAP);
 449                 /*
 450                  * Careful, clear this in the TSS too:
 451                  */
 452                 memset(tss->io_bitmap, 0xff, tss->io_bitmap_max);
 453                 t->io_bitmap_max = 0;
 454                 tss->io_bitmap_owner = NULL;
 455                 tss->io_bitmap_max = 0;
 456                 tss->x86_tss.io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
 457                 put_cpu();
 458         }
 459 }
 460
 461 void flush_thread(void)
 462 {
 463         struct task_struct *tsk = current;
 464
 465         tsk->thread.debugreg0 = 0;
 466         tsk->thread.debugreg1 = 0;
 467         tsk->thread.debugreg2 = 0;
 468         tsk->thread.debugreg3 = 0;
 469         tsk->thread.debugreg6 = 0;
 470         tsk->thread.debugreg7 = 0;
 471         memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
 472         clear_tsk_thread_flag(tsk, TIF_DEBUG);
 473         /*
 474          * Forget coprocessor state..
 475          */
 476         clear_fpu(tsk);
 477         clear_used_math();
 478 }
 479
 480 void release_thread(struct task_struct *dead_task)
 481 {
 482         BUG_ON(dead_task->mm);
 483         release_vm86_irqs(dead_task);
 484 }
 485
 486 /*
 487  * This gets called before we allocate a new thread and copy
 488  * the current task into it.
 489  */
 490 void prepare_to_copy(struct task_struct *tsk)
 491 {
 492         unlazy_fpu(tsk);
 493 }
 494
 495 int copy_thread(int nr, unsigned long clone_flags, unsigned long sp,
 496         unsigned long unused,
 497         struct task_struct * p, struct pt_regs * regs)
 498 {
 499         struct pt_regs * childregs;
 500         struct task_struct *tsk;
 501         int err;
 502
 503         childregs = task_pt_regs(p);
 504         *childregs = *regs;
 505         childregs->ax = 0;
 506         childregs->sp = sp;
 507
 508         p->thread.sp = (unsigned long) childregs;
 509         p->thread.sp0 = (unsigned long) (childregs+1);
 510
 511         p->thread.ip = (unsigned long) ret_from_fork;
 512
 513         savesegment(gs, p->thread.gs);
 514
 515         tsk = current;
 516         if (unlikely(test_tsk_thread_flag(tsk, TIF_IO_BITMAP))) {
 517                 p->thread.io_bitmap_ptr = kmemdup(tsk->thread.io_bitmap_ptr,
 518                                                 IO_BITMAP_BYTES, GFP_KERNEL);
 519                 if (!p->thread.io_bitmap_ptr) {
 520                         p->thread.io_bitmap_max = 0;
 521                         return -ENOMEM;
 522                 }
 523                 set_tsk_thread_flag(p, TIF_IO_BITMAP);
 524         }
 525
 526         err = 0;
 527
 528         /*
 529          * Set a new TLS for the child thread?
 530          */
 531         if (clone_flags & CLONE_SETTLS)
 532                 err = do_set_thread_area(p, -1,
 533                         (struct user_desc __user *)childregs->si, 0);
 534
 535         if (err && p->thread.io_bitmap_ptr) {
 536                 kfree(p->thread.io_bitmap_ptr);
 537                 p->thread.io_bitmap_max = 0;
 538         }
 539         return err;
 540 }
 541
 542 /*
 543  * fill in the user structure for a core dump..
 544  */
 545 void dump_thread(struct pt_regs * regs, struct user * dump)
 546 {
 547         u16 gs;
 548
 549 /* changed the size calculations - should hopefully work better. lbt */
 550         dump->magic = CMAGIC;
 551         dump->start_code = 0;
 552         dump->start_stack = regs->sp & ~(PAGE_SIZE - 1);
 553         dump->u_tsize = ((unsigned long) current->mm->end_code) >> PAGE_SHIFT;
 554         dump->u_dsize = ((unsigned long) (current->mm->brk + (PAGE_SIZE-1))) >> PAGE_SHIFT;
 555         dump->u_dsize -= dump->u_tsize;
 556         dump->u_ssize = 0;
 557         dump->u_debugreg[0] = current->thread.debugreg0;
 558         dump->u_debugreg[1] = current->thread.debugreg1;
 559         dump->u_debugreg[2] = current->thread.debugreg2;
 560         dump->u_debugreg[3] = current->thread.debugreg3;
 561         dump->u_debugreg[4] = 0;
 562         dump->u_debugreg[5] = 0;
 563         dump->u_debugreg[6] = current->thread.debugreg6;
 564         dump->u_debugreg[7] = current->thread.debugreg7;
 565
 566         if (dump->start_stack < TASK_SIZE)
 567                 dump->u_ssize = ((unsigned long) (TASK_SIZE - dump->start_stack)) >> PAGE_SHIFT;
 568
 569         dump->regs.bx = regs->bx;
 570         dump->regs.cx = regs->cx;
 571         dump->regs.dx = regs->dx;
 572         dump->regs.si = regs->si;
 573         dump->regs.di = regs->di;
 574         dump->regs.bp = regs->bp;
 575         dump->regs.ax = regs->ax;
 576         dump->regs.ds = (u16)regs->ds;
 577         dump->regs.es = (u16)regs->es;
 578         dump->regs.fs = (u16)regs->fs;
 579         savesegment(gs,gs);
 580         dump->regs.orig_ax = regs->orig_ax;
 581         dump->regs.ip = regs->ip;
 582         dump->regs.cs = (u16)regs->cs;
 583         dump->regs.flags = regs->flags;
 584         dump->regs.sp = regs->sp;
 585         dump->regs.ss = (u16)regs->ss;
 586
 587         dump->u_fpvalid = dump_fpu (regs, &dump->i387);
 588 }
 589 EXPORT_SYMBOL(dump_thread);
 590
 591 #ifdef CONFIG_SECCOMP
 592 static void hard_disable_TSC(void)
 593 {
 594         write_cr4(read_cr4() | X86_CR4_TSD);
 595 }
 596 void disable_TSC(void)
 597 {
 598         preempt_disable();
 599         if (!test_and_set_thread_flag(TIF_NOTSC))
 600                 /*
 601                  * Must flip the CPU state synchronously with
 602                  * TIF_NOTSC in the current running context.
 603                  */
 604                 hard_disable_TSC();
 605         preempt_enable();
 606 }
 607 static void hard_enable_TSC(void)
 608 {
 609         write_cr4(read_cr4() & ~X86_CR4_TSD);
 610 }
 611 #endif /* CONFIG_SECCOMP */
 612
 613 static noinline void
 614 __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p,
 615                  struct tss_struct *tss)
 616 {
 617         struct thread_struct *prev, *next;
 618         unsigned long debugctl;
 619
 620         prev = &prev_p->thread;
 621         next = &next_p->thread;
 622
 623         debugctl = prev->debugctlmsr;
 624         if (next->ds_area_msr != prev->ds_area_msr) {
 625                 /* we clear debugctl to make sure DS
 626                  * is not in use when we change it */
 627                 debugctl = 0;
 628                 wrmsrl(MSR_IA32_DEBUGCTLMSR, 0);
 629                 wrmsr(MSR_IA32_DS_AREA, next->ds_area_msr, 0);
 630         }
 631
 632         if (next->debugctlmsr != debugctl)
 633                 wrmsr(MSR_IA32_DEBUGCTLMSR, next->debugctlmsr, 0);
 634
 635         if (test_tsk_thread_flag(next_p, TIF_DEBUG)) {
 636                 set_debugreg(next->debugreg0, 0);
 637                 set_debugreg(next->debugreg1, 1);
 638                 set_debugreg(next->debugreg2, 2);
 639                 set_debugreg(next->debugreg3, 3);
 640                 /* no 4 and 5 */
 641                 set_debugreg(next->debugreg6, 6);
 642                 set_debugreg(next->debugreg7, 7);
 643         }
 644
 645 #ifdef CONFIG_SECCOMP
 646         if (test_tsk_thread_flag(prev_p, TIF_NOTSC) ^
 647             test_tsk_thread_flag(next_p, TIF_NOTSC)) {
 648                 /* prev and next are different */
 649                 if (test_tsk_thread_flag(next_p, TIF_NOTSC))
 650                         hard_disable_TSC();
 651                 else
 652                         hard_enable_TSC();
 653         }
 654 #endif
 655
 656         if (test_tsk_thread_flag(prev_p, TIF_BTS_TRACE_TS))
 657                 ptrace_bts_take_timestamp(prev_p, BTS_TASK_DEPARTS);
 658
 659         if (test_tsk_thread_flag(next_p, TIF_BTS_TRACE_TS))
 660                 ptrace_bts_take_timestamp(next_p, BTS_TASK_ARRIVES);
 661
 662
 663         if (!test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
 664                 /*
 665                  * Disable the bitmap via an invalid offset. We still cache
 666                  * the previous bitmap owner and the IO bitmap contents:
 667                  */
 668                 tss->x86_tss.io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
 669                 return;
 670         }
 671
 672         if (likely(next == tss->io_bitmap_owner)) {
 673                 /*
 674                  * Previous owner of the bitmap (hence the bitmap content)
 675                  * matches the next task, we dont have to do anything but
 676                  * to set a valid offset in the TSS:
 677                  */
 678                 tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET;
 679                 return;
 680         }
 681         /*
 682          * Lazy TSS's I/O bitmap copy. We set an invalid offset here
 683          * and we let the task to get a GPF in case an I/O instruction
 684          * is performed.  The handler of the GPF will verify that the
 685          * faulting task has a valid I/O bitmap and, it true, does the
 686          * real copy and restart the instruction.  This will save us
 687          * redundant copies when the currently switched task does not
 688          * perform any I/O during its timeslice.
 689          */
 690         tss->x86_tss.io_bitmap_base = INVALID_IO_BITMAP_OFFSET_LAZY;
 691 }
 692
 693 /*
 694  *      switch_to(x,yn) should switch tasks from x to y.
 695  *
 696  * We fsave/fwait so that an exception goes off at the right time
 697  * (as a call from the fsave or fwait in effect) rather than to
 698  * the wrong process. Lazy FP saving no longer makes any sense
 699  * with modern CPU's, and this simplifies a lot of things (SMP
 700  * and UP become the same).
 701  *
 702  * NOTE! We used to use the x86 hardware context switching. The
 703  * reason for not using it any more becomes apparent when you
 704  * try to recover gracefully from saved state that is no longer
 705  * valid (stale segment register values in particular). With the
 706  * hardware task-switch, there is no way to fix up bad state in
 707  * a reasonable manner.
 708  *
 709  * The fact that Intel documents the hardware task-switching to
 710  * be slow is a fairly red herring - this code is not noticeably
 711  * faster. However, there _is_ some room for improvement here,
 712  * so the performance issues may eventually be a valid point.
 713  * More important, however, is the fact that this allows us much
 714  * more flexibility.
 715  *
 716  * The return value (in %ax) will be the "prev" task after
 717  * the task-switch, and shows up in ret_from_fork in entry.S,
 718  * for example.
 719  */
 720 struct task_struct * __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
 721 {
 722         struct thread_struct *prev = &prev_p->thread,
 723                                  *next = &next_p->thread;
 724         int cpu = smp_processor_id();
 725         struct tss_struct *tss = &per_cpu(init_tss, cpu);
 726
 727         /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
 728
 729         __unlazy_fpu(prev_p);
 730
 731
 732         /* we're going to use this soon, after a few expensive things */
 733         if (next_p->fpu_counter > 5)
 734                 prefetch(&next->i387.fxsave);
 735
 736         /*
 737          * Reload esp0.
 738          */
 739         load_sp0(tss, next);
 740
 741         /*
 742          * Save away %gs. No need to save %fs, as it was saved on the
 743          * stack on entry.  No need to save %es and %ds, as those are
 744          * always kernel segments while inside the kernel.  Doing this
 745          * before setting the new TLS descriptors avoids the situation
 746          * where we temporarily have non-reloadable segments in %fs
 747          * and %gs.  This could be an issue if the NMI handler ever
 748          * used %fs or %gs (it does not today), or if the kernel is
 749          * running inside of a hypervisor layer.
 750          */
 751         savesegment(gs, prev->gs);
 752
 753         /*
 754          * Load the per-thread Thread-Local Storage descriptor.
 755          */
 756         load_TLS(next, cpu);
 757
 758         /*
 759          * Restore IOPL if needed.  In normal use, the flags restore
 760          * in the switch assembly will handle this.  But if the kernel
 761          * is running virtualized at a non-zero CPL, the popf will
 762          * not restore flags, so it must be done in a separate step.
 763          */
 764         if (get_kernel_rpl() && unlikely(prev->iopl != next->iopl))
 765                 set_iopl_mask(next->iopl);
 766
 767         /*
 768          * Now maybe handle debug registers and/or IO bitmaps
 769          */
 770         if (unlikely(task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV ||
 771                      task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT))
 772                 __switch_to_xtra(prev_p, next_p, tss);
 773
 774         /*
 775          * Leave lazy mode, flushing any hypercalls made here.
 776          * This must be done before restoring TLS segments so
 777          * the GDT and LDT are properly updated, and must be
 778          * done before math_state_restore, so the TS bit is up
 779          * to date.
 780          */
 781         arch_leave_lazy_cpu_mode();
 782
 783         /* If the task has used fpu the last 5 timeslices, just do a full
 784          * restore of the math state immediately to avoid the trap; the
 785          * chances of needing FPU soon are obviously high now
 786          */
 787         if (next_p->fpu_counter > 5)
 788                 math_state_restore();
 789
 790         /*
 791          * Restore %gs if needed (which is common)
 792          */
 793         if (prev->gs | next->gs)
 794                 loadsegment(gs, next->gs);
 795
 796         x86_write_percpu(current_task, next_p);
 797
 798         return prev_p;
 799 }
 800
 801 asmlinkage int sys_fork(struct pt_regs regs)
 802 {
 803         return do_fork(SIGCHLD, regs.sp, &regs, 0, NULL, NULL);
 804 }
 805
 806 asmlinkage int sys_clone(struct pt_regs regs)
 807 {
 808         unsigned long clone_flags;
 809         unsigned long newsp;
 810         int __user *parent_tidptr, *child_tidptr;
 811
 812         clone_flags = regs.bx;
 813         newsp = regs.cx;
 814         parent_tidptr = (int __user *)regs.dx;
 815         child_tidptr = (int __user *)regs.di;
 816         if (!newsp)
 817                 newsp = regs.sp;
 818         return do_fork(clone_flags, newsp, &regs, 0, parent_tidptr, child_tidptr);
 819 }
 820
 821 /*
 822  * This is trivial, and on the face of it looks like it
 823  * could equally well be done in user mode.
 824  *
 825  * Not so, for quite unobvious reasons - register pressure.
 826  * In user mode vfork() cannot have a stack frame, and if
 827  * done by calling the "clone()" system call directly, you
 828  * do not have enough call-clobbered registers to hold all
 829  * the information you need.
 830  */
 831 asmlinkage int sys_vfork(struct pt_regs regs)
 832 {
 833         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs.sp, &regs, 0, NULL, NULL);
 834 }
 835
 836 /*
 837  * sys_execve() executes a new program.
 838  */
 839 asmlinkage int sys_execve(struct pt_regs regs)
 840 {
 841         int error;
 842         char * filename;
 843
 844         filename = getname((char __user *) regs.bx);
 845         error = PTR_ERR(filename);
 846         if (IS_ERR(filename))
 847                 goto out;
 848         error = do_execve(filename,
 849                         (char __user * __user *) regs.cx,
 850                         (char __user * __user *) regs.dx,
 851                         &regs);
 852         if (error == 0) {
 853                 /* Make sure we don't return using sysenter.. */
 854                 set_thread_flag(TIF_IRET);
 855         }
 856         putname(filename);
 857 out:
 858         return error;
 859 }
 860
 861 #define top_esp                (THREAD_SIZE - sizeof(unsigned long))
 862 #define top_ebp                (THREAD_SIZE - 2*sizeof(unsigned long))
 863
 864 unsigned long get_wchan(struct task_struct *p)
 865 {
 866         unsigned long bp, sp, ip;
 867         unsigned long stack_page;
 868         int count = 0;
 869         if (!p || p == current || p->state == TASK_RUNNING)
 870                 return 0;
 871         stack_page = (unsigned long)task_stack_page(p);
 872         sp = p->thread.sp;
 873         if (!stack_page || sp < stack_page || sp > top_esp+stack_page)
 874                 return 0;
 875         /* include/asm-i386/system.h:switch_to() pushes bp last. */
 876         bp = *(unsigned long *) sp;
 877         do {
 878                 if (bp < stack_page || bp > top_ebp+stack_page)
 879                         return 0;
 880                 ip = *(unsigned long *) (bp+4);
 881                 if (!in_sched_functions(ip))
 882                         return ip;
 883                 bp = *(unsigned long *) bp;
 884         } while (count++ < 16);
 885         return 0;
 886 }
 887
 888 unsigned long arch_align_stack(unsigned long sp)
 889 {
 890         if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
 891                 sp -= get_random_int() % 8192;
 892         return sp & ~0xf;
 893 }
 894
 895 unsigned long arch_randomize_brk(struct mm_struct *mm)
 896 {
 897         unsigned long range_end = mm->brk + 0x02000000;
 898         return randomize_range(mm->brk, range_end, 0) ? : mm->brk;
 899 }