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/utsname.h>
  32 #include <linux/delay.h>
  33 #include <linux/reboot.h>
  34 #include <linux/init.h>
  35 #include <linux/mc146818rtc.h>
  36 #include <linux/module.h>
  37 #include <linux/kallsyms.h>
  38 #include <linux/ptrace.h>
  39 #include <linux/random.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 /*
  67  * Return saved PC of a blocked thread.
  68  */
  69 unsigned long thread_saved_pc(struct task_struct *tsk)
  70 {
  71         return ((unsigned long *)tsk->thread.esp)[3];
  72 }
  73
  74 /*
  75  * Powermanagement idle function, if any..
  76  */
  77 void (*pm_idle)(void);
  78 EXPORT_SYMBOL(pm_idle);
  79 static DEFINE_PER_CPU(unsigned int, cpu_idle_state);
  80
  81 void disable_hlt(void)
  82 {
  83         hlt_counter++;
  84 }
  85
  86 EXPORT_SYMBOL(disable_hlt);
  87
  88 void enable_hlt(void)
  89 {
  90         hlt_counter--;
  91 }
  92
  93 EXPORT_SYMBOL(enable_hlt);
  94
  95 /*
  96  * We use this if we don't have any better
  97  * idle routine..
  98  */
  99 void default_idle(void)
 100 {
 101         local_irq_enable();
 102
 103         if (!hlt_counter && boot_cpu_data.hlt_works_ok) {
 104                 current_thread_info()->status &= ~TS_POLLING;
 105                 smp_mb__after_clear_bit();
 106                 while (!need_resched()) {
 107                         local_irq_disable();
 108                         if (!need_resched())
 109                                 safe_halt();
 110                         else
 111                                 local_irq_enable();
 112                 }
 113                 current_thread_info()->status |= TS_POLLING;
 114         } else {
 115                 while (!need_resched())
 116                         cpu_relax();
 117         }
 118 }
 119 #ifdef CONFIG_APM_MODULE
 120 EXPORT_SYMBOL(default_idle);
 121 #endif
 122
 123 /*
 124  * On SMP it's slightly faster (but much more power-consuming!)
 125  * to poll the ->work.need_resched flag instead of waiting for the
 126  * cross-CPU IPI to arrive. Use this option with caution.
 127  */
 128 static void poll_idle (void)
 129 {
 130         local_irq_enable();
 131
 132         asm volatile(
 133                 "2:"
 134                 "testl %0, %1;"
 135                 "rep; nop;"
 136                 "je 2b;"
 137                 : : "i"(_TIF_NEED_RESCHED), "m" (current_thread_info()->flags));
 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                 while (!need_resched()) {
 181                         void (*idle)(void);
 182
 183                         if (__get_cpu_var(cpu_idle_state))
 184                                 __get_cpu_var(cpu_idle_state) = 0;
 185
 186                         rmb();
 187                         idle = pm_idle;
 188
 189                         if (!idle)
 190                                 idle = default_idle;
 191
 192                         if (cpu_is_offline(cpu))
 193                                 play_dead();
 194
 195                         __get_cpu_var(irq_stat).idle_timestamp = jiffies;
 196                         idle();
 197                 }
 198                 preempt_enable_no_resched();
 199                 schedule();
 200                 preempt_disable();
 201         }
 202 }
 203
 204 void cpu_idle_wait(void)
 205 {
 206         unsigned int cpu, this_cpu = get_cpu();
 207         cpumask_t map;
 208
 209         set_cpus_allowed(current, cpumask_of_cpu(this_cpu));
 210         put_cpu();
 211
 212         cpus_clear(map);
 213         for_each_online_cpu(cpu) {
 214                 per_cpu(cpu_idle_state, cpu) = 1;
 215                 cpu_set(cpu, map);
 216         }
 217
 218         __get_cpu_var(cpu_idle_state) = 0;
 219
 220         wmb();
 221         do {
 222                 ssleep(1);
 223                 for_each_online_cpu(cpu) {
 224                         if (cpu_isset(cpu, map) && !per_cpu(cpu_idle_state, cpu))
 225                                 cpu_clear(cpu, map);
 226                 }
 227                 cpus_and(map, map, cpu_online_map);
 228         } while (!cpus_empty(map));
 229 }
 230 EXPORT_SYMBOL_GPL(cpu_idle_wait);
 231
 232 /*
 233  * This uses new MONITOR/MWAIT instructions on P4 processors with PNI,
 234  * which can obviate IPI to trigger checking of need_resched.
 235  * We execute MONITOR against need_resched and enter optimized wait state
 236  * through MWAIT. Whenever someone changes need_resched, we would be woken
 237  * up from MWAIT (without an IPI).
 238  */
 239 static void mwait_idle(void)
 240 {
 241         local_irq_enable();
 242
 243         while (!need_resched()) {
 244                 __monitor((void *)&current_thread_info()->flags, 0, 0);
 245                 smp_mb();
 246                 if (need_resched())
 247                         break;
 248                 __mwait(0, 0);
 249         }
 250 }
 251
 252 void __devinit select_idle_routine(const struct cpuinfo_x86 *c)
 253 {
 254         if (cpu_has(c, X86_FEATURE_MWAIT)) {
 255                 printk("monitor/mwait feature present.\n");
 256                 /*
 257                  * Skip, if setup has overridden idle.
 258                  * One CPU supports mwait => All CPUs supports mwait
 259                  */
 260                 if (!pm_idle) {
 261                         printk("using mwait in idle threads.\n");
 262                         pm_idle = mwait_idle;
 263                 }
 264         }
 265 }
 266
 267 static int __init idle_setup (char *str)
 268 {
 269         if (!strncmp(str, "poll", 4)) {
 270                 printk("using polling idle threads.\n");
 271                 pm_idle = poll_idle;
 272 #ifdef CONFIG_X86_SMP
 273                 if (smp_num_siblings > 1)
 274                         printk("WARNING: polling idle and HT enabled, performance may degrade.\n");
 275 #endif
 276         } else if (!strncmp(str, "halt", 4)) {
 277                 printk("using halt in idle threads.\n");
 278                 pm_idle = default_idle;
 279         }
 280
 281         boot_option_idle_override = 1;
 282         return 1;
 283 }
 284
 285 __setup("idle=", idle_setup);
 286
 287 void show_regs(struct pt_regs * regs)
 288 {
 289         unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L;
 290
 291         printk("\n");
 292         printk("Pid: %d, comm: %20s\n", current->pid, current->comm);
 293         printk("EIP: %04x:[<%08lx>] CPU: %d\n",0xffff & regs->xcs,regs->eip, smp_processor_id());
 294         print_symbol("EIP is at %s\n", regs->eip);
 295
 296         if (user_mode_vm(regs))
 297                 printk(" ESP: %04x:%08lx",0xffff & regs->xss,regs->esp);
 298         printk(" EFLAGS: %08lx    %s  (%s %.*s)\n",
 299                regs->eflags, print_tainted(), system_utsname.release,
 300                (int)strcspn(system_utsname.version, " "),
 301                system_utsname.version);
 302         printk("EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
 303                 regs->eax,regs->ebx,regs->ecx,regs->edx);
 304         printk("ESI: %08lx EDI: %08lx EBP: %08lx",
 305                 regs->esi, regs->edi, regs->ebp);
 306         printk(" DS: %04x ES: %04x\n",
 307                 0xffff & regs->xds,0xffff & regs->xes);
 308
 309         cr0 = read_cr0();
 310         cr2 = read_cr2();
 311         cr3 = read_cr3();
 312         cr4 = read_cr4_safe();
 313         printk("CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n", cr0, cr2, cr3, cr4);
 314         show_trace(NULL, regs, &regs->esp);
 315 }
 316
 317 /*
 318  * This gets run with %ebx containing the
 319  * function to call, and %edx containing
 320  * the "args".
 321  */
 322 extern void kernel_thread_helper(void);
 323 __asm__(".section .text\n"
 324         ".align 4\n"
 325         "kernel_thread_helper:\n\t"
 326         "movl %edx,%eax\n\t"
 327         "pushl %edx\n\t"
 328         "call *%ebx\n\t"
 329         "pushl %eax\n\t"
 330         "call do_exit\n"
 331         ".previous");
 332
 333 /*
 334  * Create a kernel thread
 335  */
 336 int kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
 337 {
 338         struct pt_regs regs;
 339
 340         memset(&regs, 0, sizeof(regs));
 341
 342         regs.ebx = (unsigned long) fn;
 343         regs.edx = (unsigned long) arg;
 344
 345         regs.xds = __USER_DS;
 346         regs.xes = __USER_DS;
 347         regs.orig_eax = -1;
 348         regs.eip = (unsigned long) kernel_thread_helper;
 349         regs.xcs = __KERNEL_CS;
 350         regs.eflags = X86_EFLAGS_IF | X86_EFLAGS_SF | X86_EFLAGS_PF | 0x2;
 351
 352         /* Ok, create the new process.. */
 353         return do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0, &regs, 0, NULL, NULL);
 354 }
 355 EXPORT_SYMBOL(kernel_thread);
 356
 357 /*
 358  * Free current thread data structures etc..
 359  */
 360 void exit_thread(void)
 361 {
 362         /* The process may have allocated an io port bitmap... nuke it. */
 363         if (unlikely(test_thread_flag(TIF_IO_BITMAP))) {
 364                 struct task_struct *tsk = current;
 365                 struct thread_struct *t = &tsk->thread;
 366                 int cpu = get_cpu();
 367                 struct tss_struct *tss = &per_cpu(init_tss, cpu);
 368
 369                 kfree(t->io_bitmap_ptr);
 370                 t->io_bitmap_ptr = NULL;
 371                 clear_thread_flag(TIF_IO_BITMAP);
 372                 /*
 373                  * Careful, clear this in the TSS too:
 374                  */
 375                 memset(tss->io_bitmap, 0xff, tss->io_bitmap_max);
 376                 t->io_bitmap_max = 0;
 377                 tss->io_bitmap_owner = NULL;
 378                 tss->io_bitmap_max = 0;
 379                 tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
 380                 put_cpu();
 381         }
 382 }
 383
 384 void flush_thread(void)
 385 {
 386         struct task_struct *tsk = current;
 387
 388         memset(tsk->thread.debugreg, 0, sizeof(unsigned long)*8);
 389         memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
 390         clear_tsk_thread_flag(tsk, TIF_DEBUG);
 391         /*
 392          * Forget coprocessor state..
 393          */
 394         clear_fpu(tsk);
 395         clear_used_math();
 396 }
 397
 398 void release_thread(struct task_struct *dead_task)
 399 {
 400         BUG_ON(dead_task->mm);
 401         release_vm86_irqs(dead_task);
 402 }
 403
 404 /*
 405  * This gets called before we allocate a new thread and copy
 406  * the current task into it.
 407  */
 408 void prepare_to_copy(struct task_struct *tsk)
 409 {
 410         unlazy_fpu(tsk);
 411 }
 412
 413 int copy_thread(int nr, unsigned long clone_flags, unsigned long esp,
 414         unsigned long unused,
 415         struct task_struct * p, struct pt_regs * regs)
 416 {
 417         struct pt_regs * childregs;
 418         struct task_struct *tsk;
 419         int err;
 420
 421         childregs = task_pt_regs(p);
 422         *childregs = *regs;
 423         childregs->eax = 0;
 424         childregs->esp = esp;
 425
 426         p->thread.esp = (unsigned long) childregs;
 427         p->thread.esp0 = (unsigned long) (childregs+1);
 428
 429         p->thread.eip = (unsigned long) ret_from_fork;
 430
 431         savesegment(fs,p->thread.fs);
 432         savesegment(gs,p->thread.gs);
 433
 434         tsk = current;
 435         if (unlikely(test_tsk_thread_flag(tsk, TIF_IO_BITMAP))) {
 436                 p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL);
 437                 if (!p->thread.io_bitmap_ptr) {
 438                         p->thread.io_bitmap_max = 0;
 439                         return -ENOMEM;
 440                 }
 441                 memcpy(p->thread.io_bitmap_ptr, tsk->thread.io_bitmap_ptr,
 442                         IO_BITMAP_BYTES);
 443                 set_tsk_thread_flag(p, TIF_IO_BITMAP);
 444         }
 445
 446         /*
 447          * Set a new TLS for the child thread?
 448          */
 449         if (clone_flags & CLONE_SETTLS) {
 450                 struct desc_struct *desc;
 451                 struct user_desc info;
 452                 int idx;
 453
 454                 err = -EFAULT;
 455                 if (copy_from_user(&info, (void __user *)childregs->esi, sizeof(info)))
 456                         goto out;
 457                 err = -EINVAL;
 458                 if (LDT_empty(&info))
 459                         goto out;
 460
 461                 idx = info.entry_number;
 462                 if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
 463                         goto out;
 464
 465                 desc = p->thread.tls_array + idx - GDT_ENTRY_TLS_MIN;
 466                 desc->a = LDT_entry_a(&info);
 467                 desc->b = LDT_entry_b(&info);
 468         }
 469
 470         err = 0;
 471  out:
 472         if (err && p->thread.io_bitmap_ptr) {
 473                 kfree(p->thread.io_bitmap_ptr);
 474                 p->thread.io_bitmap_max = 0;
 475         }
 476         return err;
 477 }
 478
 479 /*
 480  * fill in the user structure for a core dump..
 481  */
 482 void dump_thread(struct pt_regs * regs, struct user * dump)
 483 {
 484         int i;
 485
 486 /* changed the size calculations - should hopefully work better. lbt */
 487         dump->magic = CMAGIC;
 488         dump->start_code = 0;
 489         dump->start_stack = regs->esp & ~(PAGE_SIZE - 1);
 490         dump->u_tsize = ((unsigned long) current->mm->end_code) >> PAGE_SHIFT;
 491         dump->u_dsize = ((unsigned long) (current->mm->brk + (PAGE_SIZE-1))) >> PAGE_SHIFT;
 492         dump->u_dsize -= dump->u_tsize;
 493         dump->u_ssize = 0;
 494         for (i = 0; i < 8; i++)
 495                 dump->u_debugreg[i] = current->thread.debugreg[i];
 496
 497         if (dump->start_stack < TASK_SIZE)
 498                 dump->u_ssize = ((unsigned long) (TASK_SIZE - dump->start_stack)) >> PAGE_SHIFT;
 499
 500         dump->regs.ebx = regs->ebx;
 501         dump->regs.ecx = regs->ecx;
 502         dump->regs.edx = regs->edx;
 503         dump->regs.esi = regs->esi;
 504         dump->regs.edi = regs->edi;
 505         dump->regs.ebp = regs->ebp;
 506         dump->regs.eax = regs->eax;
 507         dump->regs.ds = regs->xds;
 508         dump->regs.es = regs->xes;
 509         savesegment(fs,dump->regs.fs);
 510         savesegment(gs,dump->regs.gs);
 511         dump->regs.orig_eax = regs->orig_eax;
 512         dump->regs.eip = regs->eip;
 513         dump->regs.cs = regs->xcs;
 514         dump->regs.eflags = regs->eflags;
 515         dump->regs.esp = regs->esp;
 516         dump->regs.ss = regs->xss;
 517
 518         dump->u_fpvalid = dump_fpu (regs, &dump->i387);
 519 }
 520 EXPORT_SYMBOL(dump_thread);
 521
 522 /*
 523  * Capture the user space registers if the task is not running (in user space)
 524  */
 525 int dump_task_regs(struct task_struct *tsk, elf_gregset_t *regs)
 526 {
 527         struct pt_regs ptregs = *task_pt_regs(tsk);
 528         ptregs.xcs &= 0xffff;
 529         ptregs.xds &= 0xffff;
 530         ptregs.xes &= 0xffff;
 531         ptregs.xss &= 0xffff;
 532
 533         elf_core_copy_regs(regs, &ptregs);
 534
 535         return 1;
 536 }
 537
 538 static noinline void __switch_to_xtra(struct task_struct *next_p,
 539                                     struct tss_struct *tss)
 540 {
 541         struct thread_struct *next;
 542
 543         next = &next_p->thread;
 544
 545         if (test_tsk_thread_flag(next_p, TIF_DEBUG)) {
 546                 set_debugreg(next->debugreg[0], 0);
 547                 set_debugreg(next->debugreg[1], 1);
 548                 set_debugreg(next->debugreg[2], 2);
 549                 set_debugreg(next->debugreg[3], 3);
 550                 /* no 4 and 5 */
 551                 set_debugreg(next->debugreg[6], 6);
 552                 set_debugreg(next->debugreg[7], 7);
 553         }
 554
 555         if (!test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
 556                 /*
 557                  * Disable the bitmap via an invalid offset. We still cache
 558                  * the previous bitmap owner and the IO bitmap contents:
 559                  */
 560                 tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
 561                 return;
 562         }
 563
 564         if (likely(next == tss->io_bitmap_owner)) {
 565                 /*
 566                  * Previous owner of the bitmap (hence the bitmap content)
 567                  * matches the next task, we dont have to do anything but
 568                  * to set a valid offset in the TSS:
 569                  */
 570                 tss->io_bitmap_base = IO_BITMAP_OFFSET;
 571                 return;
 572         }
 573         /*
 574          * Lazy TSS's I/O bitmap copy. We set an invalid offset here
 575          * and we let the task to get a GPF in case an I/O instruction
 576          * is performed.  The handler of the GPF will verify that the
 577          * faulting task has a valid I/O bitmap and, it true, does the
 578          * real copy and restart the instruction.  This will save us
 579          * redundant copies when the currently switched task does not
 580          * perform any I/O during its timeslice.
 581          */
 582         tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET_LAZY;
 583 }
 584
 585 /*
 586  * This function selects if the context switch from prev to next
 587  * has to tweak the TSC disable bit in the cr4.
 588  */
 589 static inline void disable_tsc(struct task_struct *prev_p,
 590                                struct task_struct *next_p)
 591 {
 592         struct thread_info *prev, *next;
 593
 594         /*
 595          * gcc should eliminate the ->thread_info dereference if
 596          * has_secure_computing returns 0 at compile time (SECCOMP=n).
 597          */
 598         prev = task_thread_info(prev_p);
 599         next = task_thread_info(next_p);
 600
 601         if (has_secure_computing(prev) || has_secure_computing(next)) {
 602                 /* slow path here */
 603                 if (has_secure_computing(prev) &&
 604                     !has_secure_computing(next)) {
 605                         write_cr4(read_cr4() & ~X86_CR4_TSD);
 606                 } else if (!has_secure_computing(prev) &&
 607                            has_secure_computing(next))
 608                         write_cr4(read_cr4() | X86_CR4_TSD);
 609         }
 610 }
 611
 612 /*
 613  *      switch_to(x,yn) should switch tasks from x to y.
 614  *
 615  * We fsave/fwait so that an exception goes off at the right time
 616  * (as a call from the fsave or fwait in effect) rather than to
 617  * the wrong process. Lazy FP saving no longer makes any sense
 618  * with modern CPU's, and this simplifies a lot of things (SMP
 619  * and UP become the same).
 620  *
 621  * NOTE! We used to use the x86 hardware context switching. The
 622  * reason for not using it any more becomes apparent when you
 623  * try to recover gracefully from saved state that is no longer
 624  * valid (stale segment register values in particular). With the
 625  * hardware task-switch, there is no way to fix up bad state in
 626  * a reasonable manner.
 627  *
 628  * The fact that Intel documents the hardware task-switching to
 629  * be slow is a fairly red herring - this code is not noticeably
 630  * faster. However, there _is_ some room for improvement here,
 631  * so the performance issues may eventually be a valid point.
 632  * More important, however, is the fact that this allows us much
 633  * more flexibility.
 634  *
 635  * The return value (in %eax) will be the "prev" task after
 636  * the task-switch, and shows up in ret_from_fork in entry.S,
 637  * for example.
 638  */
 639 struct task_struct fastcall * __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
 640 {
 641         struct thread_struct *prev = &prev_p->thread,
 642                                  *next = &next_p->thread;
 643         int cpu = smp_processor_id();
 644         struct tss_struct *tss = &per_cpu(init_tss, cpu);
 645
 646         /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
 647
 648         __unlazy_fpu(prev_p);
 649
 650         /*
 651          * Reload esp0.
 652          */
 653         load_esp0(tss, next);
 654
 655         /*
 656          * Save away %fs and %gs. No need to save %es and %ds, as
 657          * those are always kernel segments while inside the kernel.
 658          * Doing this before setting the new TLS descriptors avoids
 659          * the situation where we temporarily have non-reloadable
 660          * segments in %fs and %gs.  This could be an issue if the
 661          * NMI handler ever used %fs or %gs (it does not today), or
 662          * if the kernel is running inside of a hypervisor layer.
 663          */
 664         savesegment(fs, prev->fs);
 665         savesegment(gs, prev->gs);
 666
 667         /*
 668          * Load the per-thread Thread-Local Storage descriptor.
 669          */
 670         load_TLS(next, cpu);
 671
 672         /*
 673          * Restore %fs and %gs if needed.
 674          *
 675          * Glibc normally makes %fs be zero, and %gs is one of
 676          * the TLS segments.
 677          */
 678         if (unlikely(prev->fs | next->fs))
 679                 loadsegment(fs, next->fs);
 680
 681         if (prev->gs | next->gs)
 682                 loadsegment(gs, next->gs);
 683
 684         /*
 685          * Restore IOPL if needed.
 686          */
 687         if (unlikely(prev->iopl != next->iopl))
 688                 set_iopl_mask(next->iopl);
 689
 690         /*
 691          * Now maybe handle debug registers and/or IO bitmaps
 692          */
 693         if (unlikely((task_thread_info(next_p)->flags & _TIF_WORK_CTXSW))
 694             || test_tsk_thread_flag(prev_p, TIF_IO_BITMAP))
 695                 __switch_to_xtra(next_p, tss);
 696
 697         disable_tsc(prev_p, next_p);
 698
 699         return prev_p;
 700 }
 701
 702 asmlinkage int sys_fork(struct pt_regs regs)
 703 {
 704         return do_fork(SIGCHLD, regs.esp, &regs, 0, NULL, NULL);
 705 }
 706
 707 asmlinkage int sys_clone(struct pt_regs regs)
 708 {
 709         unsigned long clone_flags;
 710         unsigned long newsp;
 711         int __user *parent_tidptr, *child_tidptr;
 712
 713         clone_flags = regs.ebx;
 714         newsp = regs.ecx;
 715         parent_tidptr = (int __user *)regs.edx;
 716         child_tidptr = (int __user *)regs.edi;
 717         if (!newsp)
 718                 newsp = regs.esp;
 719         return do_fork(clone_flags, newsp, &regs, 0, parent_tidptr, child_tidptr);
 720 }
 721
 722 /*
 723  * This is trivial, and on the face of it looks like it
 724  * could equally well be done in user mode.
 725  *
 726  * Not so, for quite unobvious reasons - register pressure.
 727  * In user mode vfork() cannot have a stack frame, and if
 728  * done by calling the "clone()" system call directly, you
 729  * do not have enough call-clobbered registers to hold all
 730  * the information you need.
 731  */
 732 asmlinkage int sys_vfork(struct pt_regs regs)
 733 {
 734         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs.esp, &regs, 0, NULL, NULL);
 735 }
 736
 737 /*
 738  * sys_execve() executes a new program.
 739  */
 740 asmlinkage int sys_execve(struct pt_regs regs)
 741 {
 742         int error;
 743         char * filename;
 744
 745         filename = getname((char __user *) regs.ebx);
 746         error = PTR_ERR(filename);
 747         if (IS_ERR(filename))
 748                 goto out;
 749         error = do_execve(filename,
 750                         (char __user * __user *) regs.ecx,
 751                         (char __user * __user *) regs.edx,
 752                         &regs);
 753         if (error == 0) {
 754                 task_lock(current);
 755                 current->ptrace &= ~PT_DTRACE;
 756                 task_unlock(current);
 757                 /* Make sure we don't return using sysenter.. */
 758                 set_thread_flag(TIF_IRET);
 759         }
 760         putname(filename);
 761 out:
 762         return error;
 763 }
 764
 765 #define top_esp                (THREAD_SIZE - sizeof(unsigned long))
 766 #define top_ebp                (THREAD_SIZE - 2*sizeof(unsigned long))
 767
 768 unsigned long get_wchan(struct task_struct *p)
 769 {
 770         unsigned long ebp, esp, eip;
 771         unsigned long stack_page;
 772         int count = 0;
 773         if (!p || p == current || p->state == TASK_RUNNING)
 774                 return 0;
 775         stack_page = (unsigned long)task_stack_page(p);
 776         esp = p->thread.esp;
 777         if (!stack_page || esp < stack_page || esp > top_esp+stack_page)
 778                 return 0;
 779         /* include/asm-i386/system.h:switch_to() pushes ebp last. */
 780         ebp = *(unsigned long *) esp;
 781         do {
 782                 if (ebp < stack_page || ebp > top_ebp+stack_page)
 783                         return 0;
 784                 eip = *(unsigned long *) (ebp+4);
 785                 if (!in_sched_functions(eip))
 786                         return eip;
 787                 ebp = *(unsigned long *) ebp;
 788         } while (count++ < 16);
 789         return 0;
 790 }
 791
 792 /*
 793  * sys_alloc_thread_area: get a yet unused TLS descriptor index.
 794  */
 795 static int get_free_idx(void)
 796 {
 797         struct thread_struct *t = &current->thread;
 798         int idx;
 799
 800         for (idx = 0; idx < GDT_ENTRY_TLS_ENTRIES; idx++)
 801                 if (desc_empty(t->tls_array + idx))
 802                         return idx + GDT_ENTRY_TLS_MIN;
 803         return -ESRCH;
 804 }
 805
 806 /*
 807  * Set a given TLS descriptor:
 808  */
 809 asmlinkage int sys_set_thread_area(struct user_desc __user *u_info)
 810 {
 811         struct thread_struct *t = &current->thread;
 812         struct user_desc info;
 813         struct desc_struct *desc;
 814         int cpu, idx;
 815
 816         if (copy_from_user(&info, u_info, sizeof(info)))
 817                 return -EFAULT;
 818         idx = info.entry_number;
 819
 820         /*
 821          * index -1 means the kernel should try to find and
 822          * allocate an empty descriptor:
 823          */
 824         if (idx == -1) {
 825                 idx = get_free_idx();
 826                 if (idx < 0)
 827                         return idx;
 828                 if (put_user(idx, &u_info->entry_number))
 829                         return -EFAULT;
 830         }
 831
 832         if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
 833                 return -EINVAL;
 834
 835         desc = t->tls_array + idx - GDT_ENTRY_TLS_MIN;
 836
 837         /*
 838          * We must not get preempted while modifying the TLS.
 839          */
 840         cpu = get_cpu();
 841
 842         if (LDT_empty(&info)) {
 843                 desc->a = 0;
 844                 desc->b = 0;
 845         } else {
 846                 desc->a = LDT_entry_a(&info);
 847                 desc->b = LDT_entry_b(&info);
 848         }
 849         load_TLS(t, cpu);
 850
 851         put_cpu();
 852
 853         return 0;
 854 }
 855
 856 /*
 857  * Get the current Thread-Local Storage area:
 858  */
 859
 860 #define GET_BASE(desc) ( \
 861         (((desc)->a >> 16) & 0x0000ffff) | \
 862         (((desc)->b << 16) & 0x00ff0000) | \
 863         ( (desc)->b        & 0xff000000)   )
 864
 865 #define GET_LIMIT(desc) ( \
 866         ((desc)->a & 0x0ffff) | \
 867          ((desc)->b & 0xf0000) )
 868
 869 #define GET_32BIT(desc)         (((desc)->b >> 22) & 1)
 870 #define GET_CONTENTS(desc)      (((desc)->b >> 10) & 3)
 871 #define GET_WRITABLE(desc)      (((desc)->b >>  9) & 1)
 872 #define GET_LIMIT_PAGES(desc)   (((desc)->b >> 23) & 1)
 873 #define GET_PRESENT(desc)       (((desc)->b >> 15) & 1)
 874 #define GET_USEABLE(desc)       (((desc)->b >> 20) & 1)
 875
 876 asmlinkage int sys_get_thread_area(struct user_desc __user *u_info)
 877 {
 878         struct user_desc info;
 879         struct desc_struct *desc;
 880         int idx;
 881
 882         if (get_user(idx, &u_info->entry_number))
 883                 return -EFAULT;
 884         if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
 885                 return -EINVAL;
 886
 887         memset(&info, 0, sizeof(info));
 888
 889         desc = current->thread.tls_array + idx - GDT_ENTRY_TLS_MIN;
 890
 891         info.entry_number = idx;
 892         info.base_addr = GET_BASE(desc);
 893         info.limit = GET_LIMIT(desc);
 894         info.seg_32bit = GET_32BIT(desc);
 895         info.contents = GET_CONTENTS(desc);
 896         info.read_exec_only = !GET_WRITABLE(desc);
 897         info.limit_in_pages = GET_LIMIT_PAGES(desc);
 898         info.seg_not_present = !GET_PRESENT(desc);
 899         info.useable = GET_USEABLE(desc);
 900
 901         if (copy_to_user(u_info, &info, sizeof(info)))
 902                 return -EFAULT;
 903         return 0;
 904 }
 905
 906 unsigned long arch_align_stack(unsigned long sp)
 907 {
 908         if (randomize_va_space)
 909                 sp -= get_random_int() % 8192;
 910         return sp & ~0xf;
 911 }