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
  59 asmlinkage void ret_from_fork(void) __asm__("ret_from_fork");
  60
  61 static int hlt_counter;
  62
  63 unsigned long boot_option_idle_override = 0;
  64 EXPORT_SYMBOL(boot_option_idle_override);
  65
  66 DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task;
  67 EXPORT_PER_CPU_SYMBOL(current_task);
  68
  69 DEFINE_PER_CPU(int, cpu_number);
  70 EXPORT_PER_CPU_SYMBOL(cpu_number);
  71
  72 /*
  73  * Return saved PC of a blocked thread.
  74  */
  75 unsigned long thread_saved_pc(struct task_struct *tsk)
  76 {
  77         return ((unsigned long *)tsk->thread.esp)[3];
  78 }
  79
  80 /*
  81  * Powermanagement idle function, if any..
  82  */
  83 void (*pm_idle)(void);
  84 EXPORT_SYMBOL(pm_idle);
  85 static DEFINE_PER_CPU(unsigned int, cpu_idle_state);
  86
  87 void disable_hlt(void)
  88 {
  89         hlt_counter++;
  90 }
  91
  92 EXPORT_SYMBOL(disable_hlt);
  93
  94 void enable_hlt(void)
  95 {
  96         hlt_counter--;
  97 }
  98
  99 EXPORT_SYMBOL(enable_hlt);
 100
 101 /*
 102  * We use this if we don't have any better
 103  * idle routine..
 104  */
 105 void default_idle(void)
 106 {
 107         if (!hlt_counter && boot_cpu_data.hlt_works_ok) {
 108                 current_thread_info()->status &= ~TS_POLLING;
 109                 /*
 110                  * TS_POLLING-cleared state must be visible before we
 111                  * test NEED_RESCHED:
 112                  */
 113                 smp_mb();
 114
 115                 local_irq_disable();
 116                 if (!need_resched())
 117                         safe_halt();    /* enables interrupts racelessly */
 118                 else
 119                         local_irq_enable();
 120                 current_thread_info()->status |= TS_POLLING;
 121         } else {
 122                 /* loop is done by the caller */
 123                 cpu_relax();
 124         }
 125 }
 126 #ifdef CONFIG_APM_MODULE
 127 EXPORT_SYMBOL(default_idle);
 128 #endif
 129
 130 /*
 131  * On SMP it's slightly faster (but much more power-consuming!)
 132  * to poll the ->work.need_resched flag instead of waiting for the
 133  * cross-CPU IPI to arrive. Use this option with caution.
 134  */
 135 static void poll_idle (void)
 136 {
 137         cpu_relax();
 138 }
 139
 140 #ifdef CONFIG_HOTPLUG_CPU
 141 #include <asm/nmi.h>
 142 /* We don't actually take CPU down, just spin without interrupts. */
 143 static inline void play_dead(void)
 144 {
 145         /* This must be done before dead CPU ack */
 146         cpu_exit_clear();
 147         wbinvd();
 148         mb();
 149         /* Ack it */
 150         __get_cpu_var(cpu_state) = CPU_DEAD;
 151
 152         /*
 153          * With physical CPU hotplug, we should halt the cpu
 154          */
 155         local_irq_disable();
 156         while (1)
 157                 halt();
 158 }
 159 #else
 160 static inline void play_dead(void)
 161 {
 162         BUG();
 163 }
 164 #endif /* CONFIG_HOTPLUG_CPU */
 165
 166 /*
 167  * The idle thread. There's no useful work to be
 168  * done, so just try to conserve power and have a
 169  * low exit latency (ie sit in a loop waiting for
 170  * somebody to say that they'd like to reschedule)
 171  */
 172 void cpu_idle(void)
 173 {
 174         int cpu = smp_processor_id();
 175
 176         current_thread_info()->status |= TS_POLLING;
 177
 178         /* endless idle loop with no priority at all */
 179         while (1) {
 180                 tick_nohz_stop_sched_tick();
 181                 while (!need_resched()) {
 182                         void (*idle)(void);
 183
 184                         if (__get_cpu_var(cpu_idle_state))
 185                                 __get_cpu_var(cpu_idle_state) = 0;
 186
 187                         check_pgt_cache();
 188                         rmb();
 189                         idle = pm_idle;
 190
 191                         if (!idle)
 192                                 idle = default_idle;
 193
 194                         if (cpu_is_offline(cpu))
 195                                 play_dead();
 196
 197                         __get_cpu_var(irq_stat).idle_timestamp = jiffies;
 198                         idle();
 199                 }
 200                 tick_nohz_restart_sched_tick();
 201                 preempt_enable_no_resched();
 202                 schedule();
 203                 preempt_disable();
 204         }
 205 }
 206
 207 void cpu_idle_wait(void)
 208 {
 209         unsigned int cpu, this_cpu = get_cpu();
 210         cpumask_t map, tmp = current->cpus_allowed;
 211
 212         set_cpus_allowed(current, cpumask_of_cpu(this_cpu));
 213         put_cpu();
 214
 215         cpus_clear(map);
 216         for_each_online_cpu(cpu) {
 217                 per_cpu(cpu_idle_state, cpu) = 1;
 218                 cpu_set(cpu, map);
 219         }
 220
 221         __get_cpu_var(cpu_idle_state) = 0;
 222
 223         wmb();
 224         do {
 225                 ssleep(1);
 226                 for_each_online_cpu(cpu) {
 227                         if (cpu_isset(cpu, map) && !per_cpu(cpu_idle_state, cpu))
 228                                 cpu_clear(cpu, map);
 229                 }
 230                 cpus_and(map, map, cpu_online_map);
 231         } while (!cpus_empty(map));
 232
 233         set_cpus_allowed(current, tmp);
 234 }
 235 EXPORT_SYMBOL_GPL(cpu_idle_wait);
 236
 237 /*
 238  * This uses new MONITOR/MWAIT instructions on P4 processors with PNI,
 239  * which can obviate IPI to trigger checking of need_resched.
 240  * We execute MONITOR against need_resched and enter optimized wait state
 241  * through MWAIT. Whenever someone changes need_resched, we would be woken
 242  * up from MWAIT (without an IPI).
 243  *
 244  * New with Core Duo processors, MWAIT can take some hints based on CPU
 245  * capability.
 246  */
 247 void mwait_idle_with_hints(unsigned long eax, unsigned long ecx)
 248 {
 249         if (!need_resched()) {
 250                 __monitor((void *)&current_thread_info()->flags, 0, 0);
 251                 smp_mb();
 252                 if (!need_resched())
 253                         __mwait(eax, ecx);
 254         }
 255 }
 256
 257 /* Default MONITOR/MWAIT with no hints, used for default C1 state */
 258 static void mwait_idle(void)
 259 {
 260         local_irq_enable();
 261         mwait_idle_with_hints(0, 0);
 262 }
 263
 264 void __devinit select_idle_routine(const struct cpuinfo_x86 *c)
 265 {
 266         if (cpu_has(c, X86_FEATURE_MWAIT)) {
 267                 printk("monitor/mwait feature present.\n");
 268                 /*
 269                  * Skip, if setup has overridden idle.
 270                  * One CPU supports mwait => All CPUs supports mwait
 271                  */
 272                 if (!pm_idle) {
 273                         printk("using mwait in idle threads.\n");
 274                         pm_idle = mwait_idle;
 275                 }
 276         }
 277 }
 278
 279 static int __init idle_setup(char *str)
 280 {
 281         if (!strcmp(str, "poll")) {
 282                 printk("using polling idle threads.\n");
 283                 pm_idle = poll_idle;
 284 #ifdef CONFIG_X86_SMP
 285                 if (smp_num_siblings > 1)
 286                         printk("WARNING: polling idle and HT enabled, performance may degrade.\n");
 287 #endif
 288         } else if (!strcmp(str, "mwait"))
 289                 force_mwait = 1;
 290         else
 291                 return -1;
 292
 293         boot_option_idle_override = 1;
 294         return 0;
 295 }
 296 early_param("idle", idle_setup);
 297
 298 void show_regs(struct pt_regs * regs)
 299 {
 300         unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L;
 301         unsigned long d0, d1, d2, d3, d6, d7;
 302
 303         printk("\n");
 304         printk("Pid: %d, comm: %20s\n", task_pid_nr(current), current->comm);
 305         printk("EIP: %04x:[<%08lx>] CPU: %d\n",0xffff & regs->xcs,regs->eip, smp_processor_id());
 306         print_symbol("EIP is at %s\n", regs->eip);
 307
 308         if (user_mode_vm(regs))
 309                 printk(" ESP: %04x:%08lx",0xffff & regs->xss,regs->esp);
 310         printk(" EFLAGS: %08lx    %s  (%s %.*s)\n",
 311                regs->eflags, print_tainted(), init_utsname()->release,
 312                (int)strcspn(init_utsname()->version, " "),
 313                init_utsname()->version);
 314         printk("EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
 315                 regs->eax,regs->ebx,regs->ecx,regs->edx);
 316         printk("ESI: %08lx EDI: %08lx EBP: %08lx",
 317                 regs->esi, regs->edi, regs->ebp);
 318         printk(" DS: %04x ES: %04x FS: %04x\n",
 319                0xffff & regs->xds,0xffff & regs->xes, 0xffff & regs->xfs);
 320
 321         cr0 = read_cr0();
 322         cr2 = read_cr2();
 323         cr3 = read_cr3();
 324         cr4 = read_cr4_safe();
 325         printk("CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n", cr0, cr2, cr3, cr4);
 326
 327         get_debugreg(d0, 0);
 328         get_debugreg(d1, 1);
 329         get_debugreg(d2, 2);
 330         get_debugreg(d3, 3);
 331         printk("DR0: %08lx DR1: %08lx DR2: %08lx DR3: %08lx\n",
 332                         d0, d1, d2, d3);
 333         get_debugreg(d6, 6);
 334         get_debugreg(d7, 7);
 335         printk("DR6: %08lx DR7: %08lx\n", d6, d7);
 336
 337         show_trace(NULL, regs, &regs->esp);
 338 }
 339
 340 /*
 341  * This gets run with %ebx containing the
 342  * function to call, and %edx containing
 343  * the "args".
 344  */
 345 extern void kernel_thread_helper(void);
 346
 347 /*
 348  * Create a kernel thread
 349  */
 350 int kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
 351 {
 352         struct pt_regs regs;
 353
 354         memset(&regs, 0, sizeof(regs));
 355
 356         regs.ebx = (unsigned long) fn;
 357         regs.edx = (unsigned long) arg;
 358
 359         regs.xds = __USER_DS;
 360         regs.xes = __USER_DS;
 361         regs.xfs = __KERNEL_PERCPU;
 362         regs.orig_eax = -1;
 363         regs.eip = (unsigned long) kernel_thread_helper;
 364         regs.xcs = __KERNEL_CS | get_kernel_rpl();
 365         regs.eflags = X86_EFLAGS_IF | X86_EFLAGS_SF | X86_EFLAGS_PF | 0x2;
 366
 367         /* Ok, create the new process.. */
 368         return do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0, &regs, 0, NULL, NULL);
 369 }
 370 EXPORT_SYMBOL(kernel_thread);
 371
 372 /*
 373  * Free current thread data structures etc..
 374  */
 375 void exit_thread(void)
 376 {
 377         /* The process may have allocated an io port bitmap... nuke it. */
 378         if (unlikely(test_thread_flag(TIF_IO_BITMAP))) {
 379                 struct task_struct *tsk = current;
 380                 struct thread_struct *t = &tsk->thread;
 381                 int cpu = get_cpu();
 382                 struct tss_struct *tss = &per_cpu(init_tss, cpu);
 383
 384                 kfree(t->io_bitmap_ptr);
 385                 t->io_bitmap_ptr = NULL;
 386                 clear_thread_flag(TIF_IO_BITMAP);
 387                 /*
 388                  * Careful, clear this in the TSS too:
 389                  */
 390                 memset(tss->io_bitmap, 0xff, tss->io_bitmap_max);
 391                 t->io_bitmap_max = 0;
 392                 tss->io_bitmap_owner = NULL;
 393                 tss->io_bitmap_max = 0;
 394                 tss->x86_tss.io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
 395                 put_cpu();
 396         }
 397 }
 398
 399 void flush_thread(void)
 400 {
 401         struct task_struct *tsk = current;
 402
 403         memset(tsk->thread.debugreg, 0, sizeof(unsigned long)*8);
 404         memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
 405         clear_tsk_thread_flag(tsk, TIF_DEBUG);
 406         /*
 407          * Forget coprocessor state..
 408          */
 409         clear_fpu(tsk);
 410         clear_used_math();
 411 }
 412
 413 void release_thread(struct task_struct *dead_task)
 414 {
 415         BUG_ON(dead_task->mm);
 416         release_vm86_irqs(dead_task);
 417 }
 418
 419 /*
 420  * This gets called before we allocate a new thread and copy
 421  * the current task into it.
 422  */
 423 void prepare_to_copy(struct task_struct *tsk)
 424 {
 425         unlazy_fpu(tsk);
 426 }
 427
 428 int copy_thread(int nr, unsigned long clone_flags, unsigned long esp,
 429         unsigned long unused,
 430         struct task_struct * p, struct pt_regs * regs)
 431 {
 432         struct pt_regs * childregs;
 433         struct task_struct *tsk;
 434         int err;
 435
 436         childregs = task_pt_regs(p);
 437         *childregs = *regs;
 438         childregs->eax = 0;
 439         childregs->esp = esp;
 440
 441         p->thread.esp = (unsigned long) childregs;
 442         p->thread.esp0 = (unsigned long) (childregs+1);
 443
 444         p->thread.eip = (unsigned long) ret_from_fork;
 445
 446         savesegment(gs,p->thread.gs);
 447
 448         tsk = current;
 449         if (unlikely(test_tsk_thread_flag(tsk, TIF_IO_BITMAP))) {
 450                 p->thread.io_bitmap_ptr = kmemdup(tsk->thread.io_bitmap_ptr,
 451                                                 IO_BITMAP_BYTES, GFP_KERNEL);
 452                 if (!p->thread.io_bitmap_ptr) {
 453                         p->thread.io_bitmap_max = 0;
 454                         return -ENOMEM;
 455                 }
 456                 set_tsk_thread_flag(p, TIF_IO_BITMAP);
 457         }
 458
 459         /*
 460          * Set a new TLS for the child thread?
 461          */
 462         if (clone_flags & CLONE_SETTLS) {
 463                 struct desc_struct *desc;
 464                 struct user_desc info;
 465                 int idx;
 466
 467                 err = -EFAULT;
 468                 if (copy_from_user(&info, (void __user *)childregs->esi, sizeof(info)))
 469                         goto out;
 470                 err = -EINVAL;
 471                 if (LDT_empty(&info))
 472                         goto out;
 473
 474                 idx = info.entry_number;
 475                 if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
 476                         goto out;
 477
 478                 desc = p->thread.tls_array + idx - GDT_ENTRY_TLS_MIN;
 479                 desc->a = LDT_entry_a(&info);
 480                 desc->b = LDT_entry_b(&info);
 481         }
 482
 483         err = 0;
 484  out:
 485         if (err && p->thread.io_bitmap_ptr) {
 486                 kfree(p->thread.io_bitmap_ptr);
 487                 p->thread.io_bitmap_max = 0;
 488         }
 489         return err;
 490 }
 491
 492 /*
 493  * fill in the user structure for a core dump..
 494  */
 495 void dump_thread(struct pt_regs * regs, struct user * dump)
 496 {
 497         int i;
 498
 499 /* changed the size calculations - should hopefully work better. lbt */
 500         dump->magic = CMAGIC;
 501         dump->start_code = 0;
 502         dump->start_stack = regs->esp & ~(PAGE_SIZE - 1);
 503         dump->u_tsize = ((unsigned long) current->mm->end_code) >> PAGE_SHIFT;
 504         dump->u_dsize = ((unsigned long) (current->mm->brk + (PAGE_SIZE-1))) >> PAGE_SHIFT;
 505         dump->u_dsize -= dump->u_tsize;
 506         dump->u_ssize = 0;
 507         for (i = 0; i < 8; i++)
 508                 dump->u_debugreg[i] = current->thread.debugreg[i];
 509
 510         if (dump->start_stack < TASK_SIZE)
 511                 dump->u_ssize = ((unsigned long) (TASK_SIZE - dump->start_stack)) >> PAGE_SHIFT;
 512
 513         dump->regs.ebx = regs->ebx;
 514         dump->regs.ecx = regs->ecx;
 515         dump->regs.edx = regs->edx;
 516         dump->regs.esi = regs->esi;
 517         dump->regs.edi = regs->edi;
 518         dump->regs.ebp = regs->ebp;
 519         dump->regs.eax = regs->eax;
 520         dump->regs.ds = regs->xds;
 521         dump->regs.es = regs->xes;
 522         dump->regs.fs = regs->xfs;
 523         savesegment(gs,dump->regs.gs);
 524         dump->regs.orig_eax = regs->orig_eax;
 525         dump->regs.eip = regs->eip;
 526         dump->regs.cs = regs->xcs;
 527         dump->regs.eflags = regs->eflags;
 528         dump->regs.esp = regs->esp;
 529         dump->regs.ss = regs->xss;
 530
 531         dump->u_fpvalid = dump_fpu (regs, &dump->i387);
 532 }
 533 EXPORT_SYMBOL(dump_thread);
 534
 535 /*
 536  * Capture the user space registers if the task is not running (in user space)
 537  */
 538 int dump_task_regs(struct task_struct *tsk, elf_gregset_t *regs)
 539 {
 540         struct pt_regs ptregs = *task_pt_regs(tsk);
 541         ptregs.xcs &= 0xffff;
 542         ptregs.xds &= 0xffff;
 543         ptregs.xes &= 0xffff;
 544         ptregs.xss &= 0xffff;
 545
 546         elf_core_copy_regs(regs, &ptregs);
 547
 548         return 1;
 549 }
 550
 551 #ifdef CONFIG_SECCOMP
 552 void hard_disable_TSC(void)
 553 {
 554         write_cr4(read_cr4() | X86_CR4_TSD);
 555 }
 556 void disable_TSC(void)
 557 {
 558         preempt_disable();
 559         if (!test_and_set_thread_flag(TIF_NOTSC))
 560                 /*
 561                  * Must flip the CPU state synchronously with
 562                  * TIF_NOTSC in the current running context.
 563                  */
 564                 hard_disable_TSC();
 565         preempt_enable();
 566 }
 567 void hard_enable_TSC(void)
 568 {
 569         write_cr4(read_cr4() & ~X86_CR4_TSD);
 570 }
 571 #endif /* CONFIG_SECCOMP */
 572
 573 static noinline void
 574 __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p,
 575                  struct tss_struct *tss)
 576 {
 577         struct thread_struct *next;
 578
 579         next = &next_p->thread;
 580
 581         if (test_tsk_thread_flag(next_p, TIF_DEBUG)) {
 582                 set_debugreg(next->debugreg[0], 0);
 583                 set_debugreg(next->debugreg[1], 1);
 584                 set_debugreg(next->debugreg[2], 2);
 585                 set_debugreg(next->debugreg[3], 3);
 586                 /* no 4 and 5 */
 587                 set_debugreg(next->debugreg[6], 6);
 588                 set_debugreg(next->debugreg[7], 7);
 589         }
 590
 591 #ifdef CONFIG_SECCOMP
 592         if (test_tsk_thread_flag(prev_p, TIF_NOTSC) ^
 593             test_tsk_thread_flag(next_p, TIF_NOTSC)) {
 594                 /* prev and next are different */
 595                 if (test_tsk_thread_flag(next_p, TIF_NOTSC))
 596                         hard_disable_TSC();
 597                 else
 598                         hard_enable_TSC();
 599         }
 600 #endif
 601
 602         if (!test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
 603                 /*
 604                  * Disable the bitmap via an invalid offset. We still cache
 605                  * the previous bitmap owner and the IO bitmap contents:
 606                  */
 607                 tss->x86_tss.io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
 608                 return;
 609         }
 610
 611         if (likely(next == tss->io_bitmap_owner)) {
 612                 /*
 613                  * Previous owner of the bitmap (hence the bitmap content)
 614                  * matches the next task, we dont have to do anything but
 615                  * to set a valid offset in the TSS:
 616                  */
 617                 tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET;
 618                 return;
 619         }
 620         /*
 621          * Lazy TSS's I/O bitmap copy. We set an invalid offset here
 622          * and we let the task to get a GPF in case an I/O instruction
 623          * is performed.  The handler of the GPF will verify that the
 624          * faulting task has a valid I/O bitmap and, it true, does the
 625          * real copy and restart the instruction.  This will save us
 626          * redundant copies when the currently switched task does not
 627          * perform any I/O during its timeslice.
 628          */
 629         tss->x86_tss.io_bitmap_base = INVALID_IO_BITMAP_OFFSET_LAZY;
 630 }
 631
 632 /*
 633  *      switch_to(x,yn) should switch tasks from x to y.
 634  *
 635  * We fsave/fwait so that an exception goes off at the right time
 636  * (as a call from the fsave or fwait in effect) rather than to
 637  * the wrong process. Lazy FP saving no longer makes any sense
 638  * with modern CPU's, and this simplifies a lot of things (SMP
 639  * and UP become the same).
 640  *
 641  * NOTE! We used to use the x86 hardware context switching. The
 642  * reason for not using it any more becomes apparent when you
 643  * try to recover gracefully from saved state that is no longer
 644  * valid (stale segment register values in particular). With the
 645  * hardware task-switch, there is no way to fix up bad state in
 646  * a reasonable manner.
 647  *
 648  * The fact that Intel documents the hardware task-switching to
 649  * be slow is a fairly red herring - this code is not noticeably
 650  * faster. However, there _is_ some room for improvement here,
 651  * so the performance issues may eventually be a valid point.
 652  * More important, however, is the fact that this allows us much
 653  * more flexibility.
 654  *
 655  * The return value (in %eax) will be the "prev" task after
 656  * the task-switch, and shows up in ret_from_fork in entry.S,
 657  * for example.
 658  */
 659 struct task_struct fastcall * __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
 660 {
 661         struct thread_struct *prev = &prev_p->thread,
 662                                  *next = &next_p->thread;
 663         int cpu = smp_processor_id();
 664         struct tss_struct *tss = &per_cpu(init_tss, cpu);
 665
 666         /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
 667
 668         __unlazy_fpu(prev_p);
 669
 670
 671         /* we're going to use this soon, after a few expensive things */
 672         if (next_p->fpu_counter > 5)
 673                 prefetch(&next->i387.fxsave);
 674
 675         /*
 676          * Reload esp0.
 677          */
 678         load_esp0(tss, next);
 679
 680         /*
 681          * Save away %gs. No need to save %fs, as it was saved on the
 682          * stack on entry.  No need to save %es and %ds, as those are
 683          * always kernel segments while inside the kernel.  Doing this
 684          * before setting the new TLS descriptors avoids the situation
 685          * where we temporarily have non-reloadable segments in %fs
 686          * and %gs.  This could be an issue if the NMI handler ever
 687          * used %fs or %gs (it does not today), or if the kernel is
 688          * running inside of a hypervisor layer.
 689          */
 690         savesegment(gs, prev->gs);
 691
 692         /*
 693          * Load the per-thread Thread-Local Storage descriptor.
 694          */
 695         load_TLS(next, cpu);
 696
 697         /*
 698          * Restore IOPL if needed.  In normal use, the flags restore
 699          * in the switch assembly will handle this.  But if the kernel
 700          * is running virtualized at a non-zero CPL, the popf will
 701          * not restore flags, so it must be done in a separate step.
 702          */
 703         if (get_kernel_rpl() && unlikely(prev->iopl != next->iopl))
 704                 set_iopl_mask(next->iopl);
 705
 706         /*
 707          * Now maybe handle debug registers and/or IO bitmaps
 708          */
 709         if (unlikely(task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV ||
 710                      task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT))
 711                 __switch_to_xtra(prev_p, next_p, tss);
 712
 713         /*
 714          * Leave lazy mode, flushing any hypercalls made here.
 715          * This must be done before restoring TLS segments so
 716          * the GDT and LDT are properly updated, and must be
 717          * done before math_state_restore, so the TS bit is up
 718          * to date.
 719          */
 720         arch_leave_lazy_cpu_mode();
 721
 722         /* If the task has used fpu the last 5 timeslices, just do a full
 723          * restore of the math state immediately to avoid the trap; the
 724          * chances of needing FPU soon are obviously high now
 725          */
 726         if (next_p->fpu_counter > 5)
 727                 math_state_restore();
 728
 729         /*
 730          * Restore %gs if needed (which is common)
 731          */
 732         if (prev->gs | next->gs)
 733                 loadsegment(gs, next->gs);
 734
 735         x86_write_percpu(current_task, next_p);
 736
 737         return prev_p;
 738 }
 739
 740 asmlinkage int sys_fork(struct pt_regs regs)
 741 {
 742         return do_fork(SIGCHLD, regs.esp, &regs, 0, NULL, NULL);
 743 }
 744
 745 asmlinkage int sys_clone(struct pt_regs regs)
 746 {
 747         unsigned long clone_flags;
 748         unsigned long newsp;
 749         int __user *parent_tidptr, *child_tidptr;
 750
 751         clone_flags = regs.ebx;
 752         newsp = regs.ecx;
 753         parent_tidptr = (int __user *)regs.edx;
 754         child_tidptr = (int __user *)regs.edi;
 755         if (!newsp)
 756                 newsp = regs.esp;
 757         return do_fork(clone_flags, newsp, &regs, 0, parent_tidptr, child_tidptr);
 758 }
 759
 760 /*
 761  * This is trivial, and on the face of it looks like it
 762  * could equally well be done in user mode.
 763  *
 764  * Not so, for quite unobvious reasons - register pressure.
 765  * In user mode vfork() cannot have a stack frame, and if
 766  * done by calling the "clone()" system call directly, you
 767  * do not have enough call-clobbered registers to hold all
 768  * the information you need.
 769  */
 770 asmlinkage int sys_vfork(struct pt_regs regs)
 771 {
 772         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs.esp, &regs, 0, NULL, NULL);
 773 }
 774
 775 /*
 776  * sys_execve() executes a new program.
 777  */
 778 asmlinkage int sys_execve(struct pt_regs regs)
 779 {
 780         int error;
 781         char * filename;
 782
 783         filename = getname((char __user *) regs.ebx);
 784         error = PTR_ERR(filename);
 785         if (IS_ERR(filename))
 786                 goto out;
 787         error = do_execve(filename,
 788                         (char __user * __user *) regs.ecx,
 789                         (char __user * __user *) regs.edx,
 790                         &regs);
 791         if (error == 0) {
 792                 task_lock(current);
 793                 current->ptrace &= ~PT_DTRACE;
 794                 task_unlock(current);
 795                 /* Make sure we don't return using sysenter.. */
 796                 set_thread_flag(TIF_IRET);
 797         }
 798         putname(filename);
 799 out:
 800         return error;
 801 }
 802
 803 #define top_esp                (THREAD_SIZE - sizeof(unsigned long))
 804 #define top_ebp                (THREAD_SIZE - 2*sizeof(unsigned long))
 805
 806 unsigned long get_wchan(struct task_struct *p)
 807 {
 808         unsigned long ebp, esp, eip;
 809         unsigned long stack_page;
 810         int count = 0;
 811         if (!p || p == current || p->state == TASK_RUNNING)
 812                 return 0;
 813         stack_page = (unsigned long)task_stack_page(p);
 814         esp = p->thread.esp;
 815         if (!stack_page || esp < stack_page || esp > top_esp+stack_page)
 816                 return 0;
 817         /* include/asm-i386/system.h:switch_to() pushes ebp last. */
 818         ebp = *(unsigned long *) esp;
 819         do {
 820                 if (ebp < stack_page || ebp > top_ebp+stack_page)
 821                         return 0;
 822                 eip = *(unsigned long *) (ebp+4);
 823                 if (!in_sched_functions(eip))
 824                         return eip;
 825                 ebp = *(unsigned long *) ebp;
 826         } while (count++ < 16);
 827         return 0;
 828 }
 829
 830 /*
 831  * sys_alloc_thread_area: get a yet unused TLS descriptor index.
 832  */
 833 static int get_free_idx(void)
 834 {
 835         struct thread_struct *t = &current->thread;
 836         int idx;
 837
 838         for (idx = 0; idx < GDT_ENTRY_TLS_ENTRIES; idx++)
 839                 if (desc_empty(t->tls_array + idx))
 840                         return idx + GDT_ENTRY_TLS_MIN;
 841         return -ESRCH;
 842 }
 843
 844 /*
 845  * Set a given TLS descriptor:
 846  */
 847 asmlinkage int sys_set_thread_area(struct user_desc __user *u_info)
 848 {
 849         struct thread_struct *t = &current->thread;
 850         struct user_desc info;
 851         struct desc_struct *desc;
 852         int cpu, idx;
 853
 854         if (copy_from_user(&info, u_info, sizeof(info)))
 855                 return -EFAULT;
 856         idx = info.entry_number;
 857
 858         /*
 859          * index -1 means the kernel should try to find and
 860          * allocate an empty descriptor:
 861          */
 862         if (idx == -1) {
 863                 idx = get_free_idx();
 864                 if (idx < 0)
 865                         return idx;
 866                 if (put_user(idx, &u_info->entry_number))
 867                         return -EFAULT;
 868         }
 869
 870         if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
 871                 return -EINVAL;
 872
 873         desc = t->tls_array + idx - GDT_ENTRY_TLS_MIN;
 874
 875         /*
 876          * We must not get preempted while modifying the TLS.
 877          */
 878         cpu = get_cpu();
 879
 880         if (LDT_empty(&info)) {
 881                 desc->a = 0;
 882                 desc->b = 0;
 883         } else {
 884                 desc->a = LDT_entry_a(&info);
 885                 desc->b = LDT_entry_b(&info);
 886         }
 887         load_TLS(t, cpu);
 888
 889         put_cpu();
 890
 891         return 0;
 892 }
 893
 894 /*
 895  * Get the current Thread-Local Storage area:
 896  */
 897
 898 #define GET_BASE(desc) ( \
 899         (((desc)->a >> 16) & 0x0000ffff) | \
 900         (((desc)->b << 16) & 0x00ff0000) | \
 901         ( (desc)->b        & 0xff000000)   )
 902
 903 #define GET_LIMIT(desc) ( \
 904         ((desc)->a & 0x0ffff) | \
 905          ((desc)->b & 0xf0000) )
 906
 907 #define GET_32BIT(desc)         (((desc)->b >> 22) & 1)
 908 #define GET_CONTENTS(desc)      (((desc)->b >> 10) & 3)
 909 #define GET_WRITABLE(desc)      (((desc)->b >>  9) & 1)
 910 #define GET_LIMIT_PAGES(desc)   (((desc)->b >> 23) & 1)
 911 #define GET_PRESENT(desc)       (((desc)->b >> 15) & 1)
 912 #define GET_USEABLE(desc)       (((desc)->b >> 20) & 1)
 913
 914 asmlinkage int sys_get_thread_area(struct user_desc __user *u_info)
 915 {
 916         struct user_desc info;
 917         struct desc_struct *desc;
 918         int idx;
 919
 920         if (get_user(idx, &u_info->entry_number))
 921                 return -EFAULT;
 922         if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
 923                 return -EINVAL;
 924
 925         memset(&info, 0, sizeof(info));
 926
 927         desc = current->thread.tls_array + idx - GDT_ENTRY_TLS_MIN;
 928
 929         info.entry_number = idx;
 930         info.base_addr = GET_BASE(desc);
 931         info.limit = GET_LIMIT(desc);
 932         info.seg_32bit = GET_32BIT(desc);
 933         info.contents = GET_CONTENTS(desc);
 934         info.read_exec_only = !GET_WRITABLE(desc);
 935         info.limit_in_pages = GET_LIMIT_PAGES(desc);
 936         info.seg_not_present = !GET_PRESENT(desc);
 937         info.useable = GET_USEABLE(desc);
 938
 939         if (copy_to_user(u_info, &info, sizeof(info)))
 940                 return -EFAULT;
 941         return 0;
 942 }
 943
 944 unsigned long arch_align_stack(unsigned long sp)
 945 {
 946         if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
 947                 sp -= get_random_int() % 8192;
 948         return sp & ~0xf;
 949 }