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