arch/i386/kernel/process.c

   1 /*
   2  *  linux/arch/i386/kernel/process.c
   3  *
   4  *  Copyright (C) 1995  Linus Torvalds
   5  *
   6  *  Pentium III FXSR, SSE support
   7  *      Gareth Hughes <gareth@valinux.com>, May 2000
   8  */
   9
  10 /*
  11  * This file handles the architecture-dependent parts of process handling..
  12  */
  13
  14 #include <stdarg.h>
  15
  16 #include <linux/cpu.h>
  17 #include <linux/errno.h>
  18 #include <linux/sched.h>
  19 #include <linux/fs.h>
  20 #include <linux/kernel.h>
  21 #include <linux/mm.h>
  22 #include <linux/elfcore.h>
  23 #include <linux/smp.h>
  24 #include <linux/smp_lock.h>
  25 #include <linux/stddef.h>
  26 #include <linux/slab.h>
  27 #include <linux/vmalloc.h>
  28 #include <linux/user.h>
  29 #include <linux/a.out.h>
  30 #include <linux/interrupt.h>
  31 #include <linux/config.h>
  32 #include <linux/utsname.h>
  33 #include <linux/delay.h>
  34 #include <linux/reboot.h>
  35 #include <linux/init.h>
  36 #include <linux/mc146818rtc.h>
  37 #include <linux/module.h>
  38 #include <linux/kallsyms.h>
  39 #include <linux/ptrace.h>
  40 #include <linux/random.h>
  41 #include <linux/kprobes.h>
  42
  43 #include <asm/uaccess.h>
  44 #include <asm/pgtable.h>
  45 #include <asm/system.h>
  46 #include <asm/io.h>
  47 #include <asm/ldt.h>
  48 #include <asm/processor.h>
  49 #include <asm/i387.h>
  50 #include <asm/desc.h>
  51 #include <asm/vm86.h>
  52 #ifdef CONFIG_MATH_EMULATION
  53 #include <asm/math_emu.h>
  54 #endif
  55
  56 #include <linux/err.h>
  57
  58 #include <asm/tlbflush.h>
  59 #include <asm/cpu.h>
  60
  61 asmlinkage void ret_from_fork(void) __asm__("ret_from_fork");
  62
  63 static int hlt_counter;
  64
  65 unsigned long boot_option_idle_override = 0;
  66 EXPORT_SYMBOL(boot_option_idle_override);
  67
  68 /*
  69  * Return saved PC of a blocked thread.
  70  */
  71 unsigned long thread_saved_pc(struct task_struct *tsk)
  72 {
  73         return ((unsigned long *)tsk->thread.esp)[3];
  74 }
  75
  76 /*
  77  * Powermanagement idle function, if any..
  78  */
  79 void (*pm_idle)(void);
  80 EXPORT_SYMBOL(pm_idle);
  81 static DEFINE_PER_CPU(unsigned int, cpu_idle_state);
  82
  83 void disable_hlt(void)
  84 {
  85         hlt_counter++;
  86 }
  87
  88 EXPORT_SYMBOL(disable_hlt);
  89
  90 void enable_hlt(void)
  91 {
  92         hlt_counter--;
  93 }
  94
  95 EXPORT_SYMBOL(enable_hlt);
  96
  97 /*
  98  * We use this if we don't have any better
  99  * idle routine..
 100  */
 101 void default_idle(void)
 102 {
 103         local_irq_enable();
 104
 105         if (!hlt_counter && boot_cpu_data.hlt_works_ok) {
 106                 clear_thread_flag(TIF_POLLING_NRFLAG);
 107                 smp_mb__after_clear_bit();
 108                 while (!need_resched()) {
 109                         local_irq_disable();
 110                         if (!need_resched())
 111                                 safe_halt();
 112                         else
 113                                 local_irq_enable();
 114                 }
 115                 set_thread_flag(TIF_POLLING_NRFLAG);
 116         } else {
 117                 while (!need_resched())
 118                         cpu_relax();
 119         }
 120 }
 121 #ifdef CONFIG_APM_MODULE
 122 EXPORT_SYMBOL(default_idle);
 123 #endif
 124
 125 /*
 126  * On SMP it's slightly faster (but much more power-consuming!)
 127  * to poll the ->work.need_resched flag instead of waiting for the
 128  * cross-CPU IPI to arrive. Use this option with caution.
 129  */
 130 static void poll_idle (void)
 131 {
 132         local_irq_enable();
 133
 134         asm volatile(
 135                 "2:"
 136                 "testl %0, %1;"
 137                 "rep; nop;"
 138                 "je 2b;"
 139                 : : "i"(_TIF_NEED_RESCHED), "m" (current_thread_info()->flags));
 140 }
 141
 142 #ifdef CONFIG_HOTPLUG_CPU
 143 #include <asm/nmi.h>
 144 /* We don't actually take CPU down, just spin without interrupts. */
 145 static inline void play_dead(void)
 146 {
 147         /* This must be done before dead CPU ack */
 148         cpu_exit_clear();
 149         wbinvd();
 150         mb();
 151         /* Ack it */
 152         __get_cpu_var(cpu_state) = CPU_DEAD;
 153
 154         /*
 155          * With physical CPU hotplug, we should halt the cpu
 156          */
 157         local_irq_disable();
 158         while (1)
 159                 halt();
 160 }
 161 #else
 162 static inline void play_dead(void)
 163 {
 164         BUG();
 165 }
 166 #endif /* CONFIG_HOTPLUG_CPU */
 167
 168 /*
 169  * The idle thread. There's no useful work to be
 170  * done, so just try to conserve power and have a
 171  * low exit latency (ie sit in a loop waiting for
 172  * somebody to say that they'd like to reschedule)
 173  */
 174 void cpu_idle(void)
 175 {
 176         int cpu = smp_processor_id();
 177
 178         set_thread_flag(TIF_POLLING_NRFLAG);
 179
 180         /* endless idle loop with no priority at all */
 181         while (1) {
 182                 while (!need_resched()) {
 183                         void (*idle)(void);
 184
 185                         if (__get_cpu_var(cpu_idle_state))
 186                                 __get_cpu_var(cpu_idle_state) = 0;
 187
 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                 preempt_enable_no_resched();
 201                 schedule();
 202                 preempt_disable();
 203         }
 204 }
 205
 206 void cpu_idle_wait(void)
 207 {
 208         unsigned int cpu, this_cpu = get_cpu();
 209         cpumask_t map;
 210
 211         set_cpus_allowed(current, cpumask_of_cpu(this_cpu));
 212         put_cpu();
 213
 214         cpus_clear(map);
 215         for_each_online_cpu(cpu) {
 216                 per_cpu(cpu_idle_state, cpu) = 1;
 217                 cpu_set(cpu, map);
 218         }
 219
 220         __get_cpu_var(cpu_idle_state) = 0;
 221
 222         wmb();
 223         do {
 224                 ssleep(1);
 225                 for_each_online_cpu(cpu) {
 226                         if (cpu_isset(cpu, map) && !per_cpu(cpu_idle_state, cpu))
 227                                 cpu_clear(cpu, map);
 228                 }
 229                 cpus_and(map, map, cpu_online_map);
 230         } while (!cpus_empty(map));
 231 }
 232 EXPORT_SYMBOL_GPL(cpu_idle_wait);
 233
 234 /*
 235  * This uses new MONITOR/MWAIT instructions on P4 processors with PNI,
 236  * which can obviate IPI to trigger checking of need_resched.
 237  * We execute MONITOR against need_resched and enter optimized wait state
 238  * through MWAIT. Whenever someone changes need_resched, we would be woken
 239  * up from MWAIT (without an IPI).
 240  */
 241 static void mwait_idle(void)
 242 {
 243         local_irq_enable();
 244
 245         while (!need_resched()) {
 246                 __monitor((void *)&current_thread_info()->flags, 0, 0);
 247                 smp_mb();
 248                 if (need_resched())
 249                         break;
 250                 __mwait(0, 0);
 251         }
 252 }
 253
 254 void __devinit select_idle_routine(const struct cpuinfo_x86 *c)
 255 {
 256         if (cpu_has(c, X86_FEATURE_MWAIT)) {
 257                 printk("monitor/mwait feature present.\n");
 258                 /*
 259                  * Skip, if setup has overridden idle.
 260                  * One CPU supports mwait => All CPUs supports mwait
 261                  */
 262                 if (!pm_idle) {
 263                         printk("using mwait in idle threads.\n");
 264                         pm_idle = mwait_idle;
 265                 }
 266         }
 267 }
 268
 269 static int __init idle_setup (char *str)
 270 {
 271         if (!strncmp(str, "poll", 4)) {
 272                 printk("using polling idle threads.\n");
 273                 pm_idle = poll_idle;
 274 #ifdef CONFIG_X86_SMP
 275                 if (smp_num_siblings > 1)
 276                         printk("WARNING: polling idle and HT enabled, performance may degrade.\n");
 277 #endif
 278         } else if (!strncmp(str, "halt", 4)) {
 279                 printk("using halt in idle threads.\n");
 280                 pm_idle = default_idle;
 281         }
 282
 283         boot_option_idle_override = 1;
 284         return 1;
 285 }
 286
 287 __setup("idle=", idle_setup);
 288
 289 void show_regs(struct pt_regs * regs)
 290 {
 291         unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L;
 292
 293         printk("\n");
 294         printk("Pid: %d, comm: %20s\n", current->pid, current->comm);
 295         printk("EIP: %04x:[<%08lx>] CPU: %d\n",0xffff & regs->xcs,regs->eip, smp_processor_id());
 296         print_symbol("EIP is at %s\n", regs->eip);
 297
 298         if (user_mode_vm(regs))
 299                 printk(" ESP: %04x:%08lx",0xffff & regs->xss,regs->esp);
 300         printk(" EFLAGS: %08lx    %s  (%s %.*s)\n",
 301                regs->eflags, print_tainted(), system_utsname.release,
 302                (int)strcspn(system_utsname.version, " "),
 303                system_utsname.version);
 304         printk("EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
 305                 regs->eax,regs->ebx,regs->ecx,regs->edx);
 306         printk("ESI: %08lx EDI: %08lx EBP: %08lx",
 307                 regs->esi, regs->edi, regs->ebp);
 308         printk(" DS: %04x ES: %04x\n",
 309                 0xffff & regs->xds,0xffff & regs->xes);
 310
 311         cr0 = read_cr0();
 312         cr2 = read_cr2();
 313         cr3 = read_cr3();
 314         cr4 = read_cr4_safe();
 315         printk("CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n", cr0, cr2, cr3, cr4);
 316         show_trace(NULL, &regs->esp);
 317 }
 318
 319 /*
 320  * This gets run with %ebx containing the
 321  * function to call, and %edx containing
 322  * the "args".
 323  */
 324 extern void kernel_thread_helper(void);
 325 __asm__(".section .text\n"
 326         ".align 4\n"
 327         "kernel_thread_helper:\n\t"
 328         "movl %edx,%eax\n\t"
 329         "pushl %edx\n\t"
 330         "call *%ebx\n\t"
 331         "pushl %eax\n\t"
 332         "call do_exit\n"
 333         ".previous");
 334
 335 /*
 336  * Create a kernel thread
 337  */
 338 int kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
 339 {
 340         struct pt_regs regs;
 341
 342         memset(&regs, 0, sizeof(regs));
 343
 344         regs.ebx = (unsigned long) fn;
 345         regs.edx = (unsigned long) arg;
 346
 347         regs.xds = __USER_DS;
 348         regs.xes = __USER_DS;
 349         regs.orig_eax = -1;
 350         regs.eip = (unsigned long) kernel_thread_helper;
 351         regs.xcs = __KERNEL_CS;
 352         regs.eflags = X86_EFLAGS_IF | X86_EFLAGS_SF | X86_EFLAGS_PF | 0x2;
 353
 354         /* Ok, create the new process.. */
 355         return do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0, &regs, 0, NULL, NULL);
 356 }
 357 EXPORT_SYMBOL(kernel_thread);
 358
 359 /*
 360  * Free current thread data structures etc..
 361  */
 362 void exit_thread(void)
 363 {
 364         struct task_struct *tsk = current;
 365         struct thread_struct *t = &tsk->thread;
 366
 367         /*
 368          * Remove function-return probe instances associated with this task
 369          * and put them back on the free list. Do not insert an exit probe for
 370          * this function, it will be disabled by kprobe_flush_task if you do.
 371          */
 372         kprobe_flush_task(tsk);
 373
 374         /* The process may have allocated an io port bitmap... nuke it. */
 375         if (unlikely(NULL != t->io_bitmap_ptr)) {
 376                 int cpu = get_cpu();
 377                 struct tss_struct *tss = &per_cpu(init_tss, cpu);
 378
 379                 kfree(t->io_bitmap_ptr);
 380                 t->io_bitmap_ptr = NULL;
 381                 /*
 382                  * Careful, clear this in the TSS too:
 383                  */
 384                 memset(tss->io_bitmap, 0xff, tss->io_bitmap_max);
 385                 t->io_bitmap_max = 0;
 386                 tss->io_bitmap_owner = NULL;
 387                 tss->io_bitmap_max = 0;
 388                 tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
 389                 put_cpu();
 390         }
 391 }
 392
 393 void flush_thread(void)
 394 {
 395         struct task_struct *tsk = current;
 396
 397         memset(tsk->thread.debugreg, 0, sizeof(unsigned long)*8);
 398         memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
 399         /*
 400          * Forget coprocessor state..
 401          */
 402         clear_fpu(tsk);
 403         clear_used_math();
 404 }
 405
 406 void release_thread(struct task_struct *dead_task)
 407 {
 408         BUG_ON(dead_task->mm);
 409         release_vm86_irqs(dead_task);
 410 }
 411
 412 /*
 413  * This gets called before we allocate a new thread and copy
 414  * the current task into it.
 415  */
 416 void prepare_to_copy(struct task_struct *tsk)
 417 {
 418         unlazy_fpu(tsk);
 419 }
 420
 421 int copy_thread(int nr, unsigned long clone_flags, unsigned long esp,
 422         unsigned long unused,
 423         struct task_struct * p, struct pt_regs * regs)
 424 {
 425         struct pt_regs * childregs;
 426         struct task_struct *tsk;
 427         int err;
 428
 429         childregs = task_pt_regs(p);
 430         *childregs = *regs;
 431         childregs->eax = 0;
 432         childregs->esp = esp;
 433
 434         p->thread.esp = (unsigned long) childregs;
 435         p->thread.esp0 = (unsigned long) (childregs+1);
 436
 437         p->thread.eip = (unsigned long) ret_from_fork;
 438
 439         savesegment(fs,p->thread.fs);
 440         savesegment(gs,p->thread.gs);
 441
 442         tsk = current;
 443         if (unlikely(NULL != tsk->thread.io_bitmap_ptr)) {
 444                 p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL);
 445                 if (!p->thread.io_bitmap_ptr) {
 446                         p->thread.io_bitmap_max = 0;
 447                         return -ENOMEM;
 448                 }
 449                 memcpy(p->thread.io_bitmap_ptr, tsk->thread.io_bitmap_ptr,
 450                         IO_BITMAP_BYTES);
 451         }
 452
 453         /*
 454          * Set a new TLS for the child thread?
 455          */
 456         if (clone_flags & CLONE_SETTLS) {
 457                 struct desc_struct *desc;
 458                 struct user_desc info;
 459                 int idx;
 460
 461                 err = -EFAULT;
 462                 if (copy_from_user(&info, (void __user *)childregs->esi, sizeof(info)))
 463                         goto out;
 464                 err = -EINVAL;
 465                 if (LDT_empty(&info))
 466                         goto out;
 467
 468                 idx = info.entry_number;
 469                 if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
 470                         goto out;
 471
 472                 desc = p->thread.tls_array + idx - GDT_ENTRY_TLS_MIN;
 473                 desc->a = LDT_entry_a(&info);
 474                 desc->b = LDT_entry_b(&info);
 475         }
 476
 477         err = 0;
 478  out:
 479         if (err && p->thread.io_bitmap_ptr) {
 480                 kfree(p->thread.io_bitmap_ptr);
 481                 p->thread.io_bitmap_max = 0;
 482         }
 483         return err;
 484 }
 485
 486 /*
 487  * fill in the user structure for a core dump..
 488  */
 489 void dump_thread(struct pt_regs * regs, struct user * dump)
 490 {
 491         int i;
 492
 493 /* changed the size calculations - should hopefully work better. lbt */
 494         dump->magic = CMAGIC;
 495         dump->start_code = 0;
 496         dump->start_stack = regs->esp & ~(PAGE_SIZE - 1);
 497         dump->u_tsize = ((unsigned long) current->mm->end_code) >> PAGE_SHIFT;
 498         dump->u_dsize = ((unsigned long) (current->mm->brk + (PAGE_SIZE-1))) >> PAGE_SHIFT;
 499         dump->u_dsize -= dump->u_tsize;
 500         dump->u_ssize = 0;
 501         for (i = 0; i < 8; i++)
 502                 dump->u_debugreg[i] = current->thread.debugreg[i];
 503
 504         if (dump->start_stack < TASK_SIZE)
 505                 dump->u_ssize = ((unsigned long) (TASK_SIZE - dump->start_stack)) >> PAGE_SHIFT;
 506
 507         dump->regs.ebx = regs->ebx;
 508         dump->regs.ecx = regs->ecx;
 509         dump->regs.edx = regs->edx;
 510         dump->regs.esi = regs->esi;
 511         dump->regs.edi = regs->edi;
 512         dump->regs.ebp = regs->ebp;
 513         dump->regs.eax = regs->eax;
 514         dump->regs.ds = regs->xds;
 515         dump->regs.es = regs->xes;
 516         savesegment(fs,dump->regs.fs);
 517         savesegment(gs,dump->regs.gs);
 518         dump->regs.orig_eax = regs->orig_eax;
 519         dump->regs.eip = regs->eip;
 520         dump->regs.cs = regs->xcs;
 521         dump->regs.eflags = regs->eflags;
 522         dump->regs.esp = regs->esp;
 523         dump->regs.ss = regs->xss;
 524
 525         dump->u_fpvalid = dump_fpu (regs, &dump->i387);
 526 }
 527 EXPORT_SYMBOL(dump_thread);
 528
 529 /*
 530  * Capture the user space registers if the task is not running (in user space)
 531  */
 532 int dump_task_regs(struct task_struct *tsk, elf_gregset_t *regs)
 533 {
 534         struct pt_regs ptregs = *task_pt_regs(tsk);
 535         ptregs.xcs &= 0xffff;
 536         ptregs.xds &= 0xffff;
 537         ptregs.xes &= 0xffff;
 538         ptregs.xss &= 0xffff;
 539
 540         elf_core_copy_regs(regs, &ptregs);
 541
 542         return 1;
 543 }
 544
 545 static inline void
 546 handle_io_bitmap(struct thread_struct *next, struct tss_struct *tss)
 547 {
 548         if (!next->io_bitmap_ptr) {
 549                 /*
 550                  * Disable the bitmap via an invalid offset. We still cache
 551                  * the previous bitmap owner and the IO bitmap contents:
 552                  */
 553                 tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
 554                 return;
 555         }
 556         if (likely(next == tss->io_bitmap_owner)) {
 557                 /*
 558                  * Previous owner of the bitmap (hence the bitmap content)
 559                  * matches the next task, we dont have to do anything but
 560                  * to set a valid offset in the TSS:
 561                  */
 562                 tss->io_bitmap_base = IO_BITMAP_OFFSET;
 563                 return;
 564         }
 565         /*
 566          * Lazy TSS's I/O bitmap copy. We set an invalid offset here
 567          * and we let the task to get a GPF in case an I/O instruction
 568          * is performed.  The handler of the GPF will verify that the
 569          * faulting task has a valid I/O bitmap and, it true, does the
 570          * real copy and restart the instruction.  This will save us
 571          * redundant copies when the currently switched task does not
 572          * perform any I/O during its timeslice.
 573          */
 574         tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET_LAZY;
 575 }
 576
 577 /*
 578  * This function selects if the context switch from prev to next
 579  * has to tweak the TSC disable bit in the cr4.
 580  */
 581 static inline void disable_tsc(struct task_struct *prev_p,
 582                                struct task_struct *next_p)
 583 {
 584         struct thread_info *prev, *next;
 585
 586         /*
 587          * gcc should eliminate the ->thread_info dereference if
 588          * has_secure_computing returns 0 at compile time (SECCOMP=n).
 589          */
 590         prev = task_thread_info(prev_p);
 591         next = task_thread_info(next_p);
 592
 593         if (has_secure_computing(prev) || has_secure_computing(next)) {
 594                 /* slow path here */
 595                 if (has_secure_computing(prev) &&
 596                     !has_secure_computing(next)) {
 597                         write_cr4(read_cr4() & ~X86_CR4_TSD);
 598                 } else if (!has_secure_computing(prev) &&
 599                            has_secure_computing(next))
 600                         write_cr4(read_cr4() | X86_CR4_TSD);
 601         }
 602 }
 603
 604 /*
 605  *      switch_to(x,yn) should switch tasks from x to y.
 606  *
 607  * We fsave/fwait so that an exception goes off at the right time
 608  * (as a call from the fsave or fwait in effect) rather than to
 609  * the wrong process. Lazy FP saving no longer makes any sense
 610  * with modern CPU's, and this simplifies a lot of things (SMP
 611  * and UP become the same).
 612  *
 613  * NOTE! We used to use the x86 hardware context switching. The
 614  * reason for not using it any more becomes apparent when you
 615  * try to recover gracefully from saved state that is no longer
 616  * valid (stale segment register values in particular). With the
 617  * hardware task-switch, there is no way to fix up bad state in
 618  * a reasonable manner.
 619  *
 620  * The fact that Intel documents the hardware task-switching to
 621  * be slow is a fairly red herring - this code is not noticeably
 622  * faster. However, there _is_ some room for improvement here,
 623  * so the performance issues may eventually be a valid point.
 624  * More important, however, is the fact that this allows us much
 625  * more flexibility.
 626  *
 627  * The return value (in %eax) will be the "prev" task after
 628  * the task-switch, and shows up in ret_from_fork in entry.S,
 629  * for example.
 630  */
 631 struct task_struct fastcall * __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
 632 {
 633         struct thread_struct *prev = &prev_p->thread,
 634                                  *next = &next_p->thread;
 635         int cpu = smp_processor_id();
 636         struct tss_struct *tss = &per_cpu(init_tss, cpu);
 637
 638         /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
 639
 640         __unlazy_fpu(prev_p);
 641
 642         /*
 643          * Reload esp0.
 644          */
 645         load_esp0(tss, next);
 646
 647         /*
 648          * Save away %fs and %gs. No need to save %es and %ds, as
 649          * those are always kernel segments while inside the kernel.
 650          * Doing this before setting the new TLS descriptors avoids
 651          * the situation where we temporarily have non-reloadable
 652          * segments in %fs and %gs.  This could be an issue if the
 653          * NMI handler ever used %fs or %gs (it does not today), or
 654          * if the kernel is running inside of a hypervisor layer.
 655          */
 656         savesegment(fs, prev->fs);
 657         savesegment(gs, prev->gs);
 658
 659         /*
 660          * Load the per-thread Thread-Local Storage descriptor.
 661          */
 662         load_TLS(next, cpu);
 663
 664         /*
 665          * Restore %fs and %gs if needed.
 666          *
 667          * Glibc normally makes %fs be zero, and %gs is one of
 668          * the TLS segments.
 669          */
 670         if (unlikely(prev->fs | next->fs))
 671                 loadsegment(fs, next->fs);
 672
 673         if (prev->gs | next->gs)
 674                 loadsegment(gs, next->gs);
 675
 676         /*
 677          * Restore IOPL if needed.
 678          */
 679         if (unlikely(prev->iopl != next->iopl))
 680                 set_iopl_mask(next->iopl);
 681
 682         /*
 683          * Now maybe reload the debug registers
 684          */
 685         if (unlikely(next->debugreg[7])) {
 686                 set_debugreg(next->debugreg[0], 0);
 687                 set_debugreg(next->debugreg[1], 1);
 688                 set_debugreg(next->debugreg[2], 2);
 689                 set_debugreg(next->debugreg[3], 3);
 690                 /* no 4 and 5 */
 691                 set_debugreg(next->debugreg[6], 6);
 692                 set_debugreg(next->debugreg[7], 7);
 693         }
 694
 695         if (unlikely(prev->io_bitmap_ptr || next->io_bitmap_ptr))
 696                 handle_io_bitmap(next, tss);
 697
 698         disable_tsc(prev_p, next_p);
 699
 700         return prev_p;
 701 }
 702
 703 asmlinkage int sys_fork(struct pt_regs regs)
 704 {
 705         return do_fork(SIGCHLD, regs.esp, &regs, 0, NULL, NULL);
 706 }
 707
 708 asmlinkage int sys_clone(struct pt_regs regs)
 709 {
 710         unsigned long clone_flags;
 711         unsigned long newsp;
 712         int __user *parent_tidptr, *child_tidptr;
 713
 714         clone_flags = regs.ebx;
 715         newsp = regs.ecx;
 716         parent_tidptr = (int __user *)regs.edx;
 717         child_tidptr = (int __user *)regs.edi;
 718         if (!newsp)
 719                 newsp = regs.esp;
 720         return do_fork(clone_flags, newsp, &regs, 0, parent_tidptr, child_tidptr);
 721 }
 722
 723 /*
 724  * This is trivial, and on the face of it looks like it
 725  * could equally well be done in user mode.
 726  *
 727  * Not so, for quite unobvious reasons - register pressure.
 728  * In user mode vfork() cannot have a stack frame, and if
 729  * done by calling the "clone()" system call directly, you
 730  * do not have enough call-clobbered registers to hold all
 731  * the information you need.
 732  */
 733 asmlinkage int sys_vfork(struct pt_regs regs)
 734 {
 735         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs.esp, &regs, 0, NULL, NULL);
 736 }
 737
 738 /*
 739  * sys_execve() executes a new program.
 740  */
 741 asmlinkage int sys_execve(struct pt_regs regs)
 742 {
 743         int error;
 744         char * filename;
 745
 746         filename = getname((char __user *) regs.ebx);
 747         error = PTR_ERR(filename);
 748         if (IS_ERR(filename))
 749                 goto out;
 750         error = do_execve(filename,
 751                         (char __user * __user *) regs.ecx,
 752                         (char __user * __user *) regs.edx,
 753                         &regs);
 754         if (error == 0) {
 755                 task_lock(current);
 756                 current->ptrace &= ~PT_DTRACE;
 757                 task_unlock(current);
 758                 /* Make sure we don't return using sysenter.. */
 759                 set_thread_flag(TIF_IRET);
 760         }
 761         putname(filename);
 762 out:
 763         return error;
 764 }
 765
 766 #define top_esp                (THREAD_SIZE - sizeof(unsigned long))
 767 #define top_ebp                (THREAD_SIZE - 2*sizeof(unsigned long))
 768
 769 unsigned long get_wchan(struct task_struct *p)
 770 {
 771         unsigned long ebp, esp, eip;
 772         unsigned long stack_page;
 773         int count = 0;
 774         if (!p || p == current || p->state == TASK_RUNNING)
 775                 return 0;
 776         stack_page = (unsigned long)task_stack_page(p);
 777         esp = p->thread.esp;
 778         if (!stack_page || esp < stack_page || esp > top_esp+stack_page)
 779                 return 0;
 780         /* include/asm-i386/system.h:switch_to() pushes ebp last. */
 781         ebp = *(unsigned long *) esp;
 782         do {
 783                 if (ebp < stack_page || ebp > top_ebp+stack_page)
 784                         return 0;
 785                 eip = *(unsigned long *) (ebp+4);
 786                 if (!in_sched_functions(eip))
 787                         return eip;
 788                 ebp = *(unsigned long *) ebp;
 789         } while (count++ < 16);
 790         return 0;
 791 }
 792 EXPORT_SYMBOL(get_wchan);
 793
 794 /*
 795  * sys_alloc_thread_area: get a yet unused TLS descriptor index.
 796  */
 797 static int get_free_idx(void)
 798 {
 799         struct thread_struct *t = &current->thread;
 800         int idx;
 801
 802         for (idx = 0; idx < GDT_ENTRY_TLS_ENTRIES; idx++)
 803                 if (desc_empty(t->tls_array + idx))
 804                         return idx + GDT_ENTRY_TLS_MIN;
 805         return -ESRCH;
 806 }
 807
 808 /*
 809  * Set a given TLS descriptor:
 810  */
 811 asmlinkage int sys_set_thread_area(struct user_desc __user *u_info)
 812 {
 813         struct thread_struct *t = &current->thread;
 814         struct user_desc info;
 815         struct desc_struct *desc;
 816         int cpu, idx;
 817
 818         if (copy_from_user(&info, u_info, sizeof(info)))
 819                 return -EFAULT;
 820         idx = info.entry_number;
 821
 822         /*
 823          * index -1 means the kernel should try to find and
 824          * allocate an empty descriptor:
 825          */
 826         if (idx == -1) {
 827                 idx = get_free_idx();
 828                 if (idx < 0)
 829                         return idx;
 830                 if (put_user(idx, &u_info->entry_number))
 831                         return -EFAULT;
 832         }
 833
 834         if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
 835                 return -EINVAL;
 836
 837         desc = t->tls_array + idx - GDT_ENTRY_TLS_MIN;
 838
 839         /*
 840          * We must not get preempted while modifying the TLS.
 841          */
 842         cpu = get_cpu();
 843
 844         if (LDT_empty(&info)) {
 845                 desc->a = 0;
 846                 desc->b = 0;
 847         } else {
 848                 desc->a = LDT_entry_a(&info);
 849                 desc->b = LDT_entry_b(&info);
 850         }
 851         load_TLS(t, cpu);
 852
 853         put_cpu();
 854
 855         return 0;
 856 }
 857
 858 /*
 859  * Get the current Thread-Local Storage area:
 860  */
 861
 862 #define GET_BASE(desc) ( \
 863         (((desc)->a >> 16) & 0x0000ffff) | \
 864         (((desc)->b << 16) & 0x00ff0000) | \
 865         ( (desc)->b        & 0xff000000)   )
 866
 867 #define GET_LIMIT(desc) ( \
 868         ((desc)->a & 0x0ffff) | \
 869          ((desc)->b & 0xf0000) )
 870
 871 #define GET_32BIT(desc)         (((desc)->b >> 22) & 1)
 872 #define GET_CONTENTS(desc)      (((desc)->b >> 10) & 3)
 873 #define GET_WRITABLE(desc)      (((desc)->b >>  9) & 1)
 874 #define GET_LIMIT_PAGES(desc)   (((desc)->b >> 23) & 1)
 875 #define GET_PRESENT(desc)       (((desc)->b >> 15) & 1)
 876 #define GET_USEABLE(desc)       (((desc)->b >> 20) & 1)
 877
 878 asmlinkage int sys_get_thread_area(struct user_desc __user *u_info)
 879 {
 880         struct user_desc info;
 881         struct desc_struct *desc;
 882         int idx;
 883
 884         if (get_user(idx, &u_info->entry_number))
 885                 return -EFAULT;
 886         if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
 887                 return -EINVAL;
 888
 889         memset(&info, 0, sizeof(info));
 890
 891         desc = current->thread.tls_array + idx - GDT_ENTRY_TLS_MIN;
 892
 893         info.entry_number = idx;
 894         info.base_addr = GET_BASE(desc);
 895         info.limit = GET_LIMIT(desc);
 896         info.seg_32bit = GET_32BIT(desc);
 897         info.contents = GET_CONTENTS(desc);
 898         info.read_exec_only = !GET_WRITABLE(desc);
 899         info.limit_in_pages = GET_LIMIT_PAGES(desc);
 900         info.seg_not_present = !GET_PRESENT(desc);
 901         info.useable = GET_USEABLE(desc);
 902
 903         if (copy_to_user(u_info, &info, sizeof(info)))
 904                 return -EFAULT;
 905         return 0;
 906 }
 907
 908 unsigned long arch_align_stack(unsigned long sp)
 909 {
 910         if (randomize_va_space)
 911                 sp -= get_random_int() % 8192;
 912         return sp & ~0xf;
 913 }