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 #include <linux/personality.h>
  41
  42 #include <asm/uaccess.h>
  43 #include <asm/pgtable.h>
  44 #include <asm/system.h>
  45 #include <asm/io.h>
  46 #include <asm/ldt.h>
  47 #include <asm/processor.h>
  48 #include <asm/i387.h>
  49 #include <asm/desc.h>
  50 #include <asm/vm86.h>
  51 #ifdef CONFIG_MATH_EMULATION
  52 #include <asm/math_emu.h>
  53 #endif
  54
  55 #include <linux/err.h>
  56
  57 #include <asm/tlbflush.h>
  58 #include <asm/cpu.h>
  59
  60 asmlinkage void ret_from_fork(void) __asm__("ret_from_fork");
  61
  62 static int hlt_counter;
  63
  64 unsigned long boot_option_idle_override = 0;
  65 EXPORT_SYMBOL(boot_option_idle_override);
  66
  67 /*
  68  * Return saved PC of a blocked thread.
  69  */
  70 unsigned long thread_saved_pc(struct task_struct *tsk)
  71 {
  72         return ((unsigned long *)tsk->thread.esp)[3];
  73 }
  74
  75 /*
  76  * Powermanagement idle function, if any..
  77  */
  78 void (*pm_idle)(void);
  79 EXPORT_SYMBOL(pm_idle);
  80 static DEFINE_PER_CPU(unsigned int, cpu_idle_state);
  81
  82 void disable_hlt(void)
  83 {
  84         hlt_counter++;
  85 }
  86
  87 EXPORT_SYMBOL(disable_hlt);
  88
  89 void enable_hlt(void)
  90 {
  91         hlt_counter--;
  92 }
  93
  94 EXPORT_SYMBOL(enable_hlt);
  95
  96 /*
  97  * We use this if we don't have any better
  98  * idle routine..
  99  */
 100 void default_idle(void)
 101 {
 102         local_irq_enable();
 103
 104         if (!hlt_counter && boot_cpu_data.hlt_works_ok) {
 105                 current_thread_info()->status &= ~TS_POLLING;
 106                 smp_mb__after_clear_bit();
 107                 while (!need_resched()) {
 108                         local_irq_disable();
 109                         if (!need_resched())
 110                                 safe_halt();
 111                         else
 112                                 local_irq_enable();
 113                 }
 114                 current_thread_info()->status |= TS_POLLING;
 115         } else {
 116                 while (!need_resched())
 117                         cpu_relax();
 118         }
 119 }
 120 #ifdef CONFIG_APM_MODULE
 121 EXPORT_SYMBOL(default_idle);
 122 #endif
 123
 124 /*
 125  * On SMP it's slightly faster (but much more power-consuming!)
 126  * to poll the ->work.need_resched flag instead of waiting for the
 127  * cross-CPU IPI to arrive. Use this option with caution.
 128  */
 129 static void poll_idle (void)
 130 {
 131         local_irq_enable();
 132
 133         asm volatile(
 134                 "2:"
 135                 "testl %0, %1;"
 136                 "rep; nop;"
 137                 "je 2b;"
 138                 : : "i"(_TIF_NEED_RESCHED), "m" (current_thread_info()->flags));
 139 }
 140
 141 #ifdef CONFIG_HOTPLUG_CPU
 142 #include <asm/nmi.h>
 143 /* We don't actually take CPU down, just spin without interrupts. */
 144 static inline void play_dead(void)
 145 {
 146         /* This must be done before dead CPU ack */
 147         cpu_exit_clear();
 148         wbinvd();
 149         mb();
 150         /* Ack it */
 151         __get_cpu_var(cpu_state) = CPU_DEAD;
 152
 153         /*
 154          * With physical CPU hotplug, we should halt the cpu
 155          */
 156         local_irq_disable();
 157         while (1)
 158                 halt();
 159 }
 160 #else
 161 static inline void play_dead(void)
 162 {
 163         BUG();
 164 }
 165 #endif /* CONFIG_HOTPLUG_CPU */
 166
 167 /*
 168  * The idle thread. There's no useful work to be
 169  * done, so just try to conserve power and have a
 170  * low exit latency (ie sit in a loop waiting for
 171  * somebody to say that they'd like to reschedule)
 172  */
 173 void cpu_idle(void)
 174 {
 175         int cpu = smp_processor_id();
 176
 177         current_thread_info()->status |= TS_POLLING;
 178
 179         /* endless idle loop with no priority at all */
 180         while (1) {
 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                         rmb();
 188                         idle = pm_idle;
 189
 190                         if (!idle)
 191                                 idle = default_idle;
 192
 193                         if (cpu_is_offline(cpu))
 194                                 play_dead();
 195
 196                         __get_cpu_var(irq_stat).idle_timestamp = jiffies;
 197                         idle();
 198                 }
 199                 preempt_enable_no_resched();
 200                 schedule();
 201                 preempt_disable();
 202         }
 203 }
 204
 205 void cpu_idle_wait(void)
 206 {
 207         unsigned int cpu, this_cpu = get_cpu();
 208         cpumask_t map;
 209
 210         set_cpus_allowed(current, cpumask_of_cpu(this_cpu));
 211         put_cpu();
 212
 213         cpus_clear(map);
 214         for_each_online_cpu(cpu) {
 215                 per_cpu(cpu_idle_state, cpu) = 1;
 216                 cpu_set(cpu, map);
 217         }
 218
 219         __get_cpu_var(cpu_idle_state) = 0;
 220
 221         wmb();
 222         do {
 223                 ssleep(1);
 224                 for_each_online_cpu(cpu) {
 225                         if (cpu_isset(cpu, map) && !per_cpu(cpu_idle_state, cpu))
 226                                 cpu_clear(cpu, map);
 227                 }
 228                 cpus_and(map, map, cpu_online_map);
 229         } while (!cpus_empty(map));
 230 }
 231 EXPORT_SYMBOL_GPL(cpu_idle_wait);
 232
 233 /*
 234  * This uses new MONITOR/MWAIT instructions on P4 processors with PNI,
 235  * which can obviate IPI to trigger checking of need_resched.
 236  * We execute MONITOR against need_resched and enter optimized wait state
 237  * through MWAIT. Whenever someone changes need_resched, we would be woken
 238  * up from MWAIT (without an IPI).
 239  *
 240  * New with Core Duo processors, MWAIT can take some hints based on CPU
 241  * capability.
 242  */
 243 void mwait_idle_with_hints(unsigned long eax, unsigned long ecx)
 244 {
 245         if (!need_resched()) {
 246                 __monitor((void *)&current_thread_info()->flags, 0, 0);
 247                 smp_mb();
 248                 if (!need_resched())
 249                         __mwait(eax, ecx);
 250         }
 251 }
 252
 253 /* Default MONITOR/MWAIT with no hints, used for default C1 state */
 254 static void mwait_idle(void)
 255 {
 256         local_irq_enable();
 257         while (!need_resched())
 258                 mwait_idle_with_hints(0, 0);
 259 }
 260
 261 void __devinit select_idle_routine(const struct cpuinfo_x86 *c)
 262 {
 263         if (cpu_has(c, X86_FEATURE_MWAIT)) {
 264                 printk("monitor/mwait feature present.\n");
 265                 /*
 266                  * Skip, if setup has overridden idle.
 267                  * One CPU supports mwait => All CPUs supports mwait
 268                  */
 269                 if (!pm_idle) {
 270                         printk("using mwait in idle threads.\n");
 271                         pm_idle = mwait_idle;
 272                 }
 273         }
 274 }
 275
 276 static int __init idle_setup (char *str)
 277 {
 278         if (!strncmp(str, "poll", 4)) {
 279                 printk("using polling idle threads.\n");
 280                 pm_idle = poll_idle;
 281 #ifdef CONFIG_X86_SMP
 282                 if (smp_num_siblings > 1)
 283                         printk("WARNING: polling idle and HT enabled, performance may degrade.\n");
 284 #endif
 285         } else if (!strncmp(str, "halt", 4)) {
 286                 printk("using halt in idle threads.\n");
 287                 pm_idle = default_idle;
 288         }
 289
 290         boot_option_idle_override = 1;
 291         return 1;
 292 }
 293
 294 __setup("idle=", idle_setup);
 295
 296 void show_regs(struct pt_regs * regs)
 297 {
 298         unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L;
 299
 300         printk("\n");
 301         printk("Pid: %d, comm: %20s\n", current->pid, current->comm);
 302         printk("EIP: %04x:[<%08lx>] CPU: %d\n",0xffff & regs->xcs,regs->eip, smp_processor_id());
 303         print_symbol("EIP is at %s\n", regs->eip);
 304
 305         if (user_mode_vm(regs))
 306                 printk(" ESP: %04x:%08lx",0xffff & regs->xss,regs->esp);
 307         printk(" EFLAGS: %08lx    %s  (%s %.*s)\n",
 308                regs->eflags, print_tainted(), init_utsname()->release,
 309                (int)strcspn(init_utsname()->version, " "),
 310                init_utsname()->version);
 311         printk("EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
 312                 regs->eax,regs->ebx,regs->ecx,regs->edx);
 313         printk("ESI: %08lx EDI: %08lx EBP: %08lx",
 314                 regs->esi, regs->edi, regs->ebp);
 315         printk(" DS: %04x ES: %04x\n",
 316                 0xffff & regs->xds,0xffff & regs->xes);
 317
 318         cr0 = read_cr0();
 319         cr2 = read_cr2();
 320         cr3 = read_cr3();
 321         cr4 = read_cr4_safe();
 322         printk("CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n", cr0, cr2, cr3, cr4);
 323         show_trace(NULL, regs, &regs->esp);
 324 }
 325
 326 /*
 327  * This gets run with %ebx containing the
 328  * function to call, and %edx containing
 329  * the "args".
 330  */
 331 extern void kernel_thread_helper(void);
 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         int err;
 340
 341         memset(&regs, 0, sizeof(regs));
 342
 343         regs.ebx = (unsigned long) fn;
 344         regs.edx = (unsigned long) arg;
 345
 346         regs.xds = __USER_DS;
 347         regs.xes = __USER_DS;
 348         regs.orig_eax = -1;
 349         regs.eip = (unsigned long) kernel_thread_helper;
 350         regs.xcs = __KERNEL_CS | get_kernel_rpl();
 351         regs.eflags = X86_EFLAGS_IF | X86_EFLAGS_SF | X86_EFLAGS_PF | 0x2;
 352
 353         /* Ok, create the new process.. */
 354         err = do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0, &regs, 0, NULL, NULL);
 355         if (err == 0) /* terminate kernel stack */
 356                 task_pt_regs(current)->eip = 0;
 357         return err;
 358 }
 359 EXPORT_SYMBOL(kernel_thread);
 360
 361 /*
 362  * Free current thread data structures etc..
 363  */
 364 void exit_thread(void)
 365 {
 366         /* The process may have allocated an io port bitmap... nuke it. */
 367         if (unlikely(test_thread_flag(TIF_IO_BITMAP))) {
 368                 struct task_struct *tsk = current;
 369                 struct thread_struct *t = &tsk->thread;
 370                 int cpu = get_cpu();
 371                 struct tss_struct *tss = &per_cpu(init_tss, cpu);
 372
 373                 kfree(t->io_bitmap_ptr);
 374                 t->io_bitmap_ptr = NULL;
 375                 clear_thread_flag(TIF_IO_BITMAP);
 376                 /*
 377                  * Careful, clear this in the TSS too:
 378                  */
 379                 memset(tss->io_bitmap, 0xff, tss->io_bitmap_max);
 380                 t->io_bitmap_max = 0;
 381                 tss->io_bitmap_owner = NULL;
 382                 tss->io_bitmap_max = 0;
 383                 tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
 384                 put_cpu();
 385         }
 386 }
 387
 388 void flush_thread(void)
 389 {
 390         struct task_struct *tsk = current;
 391
 392         memset(tsk->thread.debugreg, 0, sizeof(unsigned long)*8);
 393         memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
 394         clear_tsk_thread_flag(tsk, TIF_DEBUG);
 395         /*
 396          * Forget coprocessor state..
 397          */
 398         clear_fpu(tsk);
 399         clear_used_math();
 400 }
 401
 402 void release_thread(struct task_struct *dead_task)
 403 {
 404         BUG_ON(dead_task->mm);
 405         release_vm86_irqs(dead_task);
 406 }
 407
 408 /*
 409  * This gets called before we allocate a new thread and copy
 410  * the current task into it.
 411  */
 412 void prepare_to_copy(struct task_struct *tsk)
 413 {
 414         unlazy_fpu(tsk);
 415 }
 416
 417 int copy_thread(int nr, unsigned long clone_flags, unsigned long esp,
 418         unsigned long unused,
 419         struct task_struct * p, struct pt_regs * regs)
 420 {
 421         struct pt_regs * childregs;
 422         struct task_struct *tsk;
 423         int err;
 424
 425         childregs = task_pt_regs(p);
 426         *childregs = *regs;
 427         childregs->eax = 0;
 428         childregs->esp = esp;
 429
 430         p->thread.esp = (unsigned long) childregs;
 431         p->thread.esp0 = (unsigned long) (childregs+1);
 432
 433         p->thread.eip = (unsigned long) ret_from_fork;
 434
 435         savesegment(fs,p->thread.fs);
 436         savesegment(gs,p->thread.gs);
 437
 438         tsk = current;
 439         if (unlikely(test_tsk_thread_flag(tsk, TIF_IO_BITMAP))) {
 440                 p->thread.io_bitmap_ptr = kmemdup(tsk->thread.io_bitmap_ptr,
 441                                                 IO_BITMAP_BYTES, GFP_KERNEL);
 442                 if (!p->thread.io_bitmap_ptr) {
 443                         p->thread.io_bitmap_max = 0;
 444                         return -ENOMEM;
 445                 }
 446                 set_tsk_thread_flag(p, TIF_IO_BITMAP);
 447         }
 448
 449         /*
 450          * Set a new TLS for the child thread?
 451          */
 452         if (clone_flags & CLONE_SETTLS) {
 453                 struct desc_struct *desc;
 454                 struct user_desc info;
 455                 int idx;
 456
 457                 err = -EFAULT;
 458                 if (copy_from_user(&info, (void __user *)childregs->esi, sizeof(info)))
 459                         goto out;
 460                 err = -EINVAL;
 461                 if (LDT_empty(&info))
 462                         goto out;
 463
 464                 idx = info.entry_number;
 465                 if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
 466                         goto out;
 467
 468                 desc = p->thread.tls_array + idx - GDT_ENTRY_TLS_MIN;
 469                 desc->a = LDT_entry_a(&info);
 470                 desc->b = LDT_entry_b(&info);
 471         }
 472
 473         err = 0;
 474  out:
 475         if (err && p->thread.io_bitmap_ptr) {
 476                 kfree(p->thread.io_bitmap_ptr);
 477                 p->thread.io_bitmap_max = 0;
 478         }
 479         return err;
 480 }
 481
 482 /*
 483  * fill in the user structure for a core dump..
 484  */
 485 void dump_thread(struct pt_regs * regs, struct user * dump)
 486 {
 487         int i;
 488
 489 /* changed the size calculations - should hopefully work better. lbt */
 490         dump->magic = CMAGIC;
 491         dump->start_code = 0;
 492         dump->start_stack = regs->esp & ~(PAGE_SIZE - 1);
 493         dump->u_tsize = ((unsigned long) current->mm->end_code) >> PAGE_SHIFT;
 494         dump->u_dsize = ((unsigned long) (current->mm->brk + (PAGE_SIZE-1))) >> PAGE_SHIFT;
 495         dump->u_dsize -= dump->u_tsize;
 496         dump->u_ssize = 0;
 497         for (i = 0; i < 8; i++)
 498                 dump->u_debugreg[i] = current->thread.debugreg[i];
 499
 500         if (dump->start_stack < TASK_SIZE)
 501                 dump->u_ssize = ((unsigned long) (TASK_SIZE - dump->start_stack)) >> PAGE_SHIFT;
 502
 503         dump->regs.ebx = regs->ebx;
 504         dump->regs.ecx = regs->ecx;
 505         dump->regs.edx = regs->edx;
 506         dump->regs.esi = regs->esi;
 507         dump->regs.edi = regs->edi;
 508         dump->regs.ebp = regs->ebp;
 509         dump->regs.eax = regs->eax;
 510         dump->regs.ds = regs->xds;
 511         dump->regs.es = regs->xes;
 512         savesegment(fs,dump->regs.fs);
 513         savesegment(gs,dump->regs.gs);
 514         dump->regs.orig_eax = regs->orig_eax;
 515         dump->regs.eip = regs->eip;
 516         dump->regs.cs = regs->xcs;
 517         dump->regs.eflags = regs->eflags;
 518         dump->regs.esp = regs->esp;
 519         dump->regs.ss = regs->xss;
 520
 521         dump->u_fpvalid = dump_fpu (regs, &dump->i387);
 522 }
 523 EXPORT_SYMBOL(dump_thread);
 524
 525 /*
 526  * Capture the user space registers if the task is not running (in user space)
 527  */
 528 int dump_task_regs(struct task_struct *tsk, elf_gregset_t *regs)
 529 {
 530         struct pt_regs ptregs = *task_pt_regs(tsk);
 531         ptregs.xcs &= 0xffff;
 532         ptregs.xds &= 0xffff;
 533         ptregs.xes &= 0xffff;
 534         ptregs.xss &= 0xffff;
 535
 536         elf_core_copy_regs(regs, &ptregs);
 537
 538         return 1;
 539 }
 540
 541 static noinline void __switch_to_xtra(struct task_struct *next_p,
 542                                     struct tss_struct *tss)
 543 {
 544         struct thread_struct *next;
 545
 546         next = &next_p->thread;
 547
 548         if (test_tsk_thread_flag(next_p, TIF_DEBUG)) {
 549                 set_debugreg(next->debugreg[0], 0);
 550                 set_debugreg(next->debugreg[1], 1);
 551                 set_debugreg(next->debugreg[2], 2);
 552                 set_debugreg(next->debugreg[3], 3);
 553                 /* no 4 and 5 */
 554                 set_debugreg(next->debugreg[6], 6);
 555                 set_debugreg(next->debugreg[7], 7);
 556         }
 557
 558         if (!test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
 559                 /*
 560                  * Disable the bitmap via an invalid offset. We still cache
 561                  * the previous bitmap owner and the IO bitmap contents:
 562                  */
 563                 tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
 564                 return;
 565         }
 566
 567         if (likely(next == tss->io_bitmap_owner)) {
 568                 /*
 569                  * Previous owner of the bitmap (hence the bitmap content)
 570                  * matches the next task, we dont have to do anything but
 571                  * to set a valid offset in the TSS:
 572                  */
 573                 tss->io_bitmap_base = IO_BITMAP_OFFSET;
 574                 return;
 575         }
 576         /*
 577          * Lazy TSS's I/O bitmap copy. We set an invalid offset here
 578          * and we let the task to get a GPF in case an I/O instruction
 579          * is performed.  The handler of the GPF will verify that the
 580          * faulting task has a valid I/O bitmap and, it true, does the
 581          * real copy and restart the instruction.  This will save us
 582          * redundant copies when the currently switched task does not
 583          * perform any I/O during its timeslice.
 584          */
 585         tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET_LAZY;
 586 }
 587
 588 /*
 589  * This function selects if the context switch from prev to next
 590  * has to tweak the TSC disable bit in the cr4.
 591  */
 592 static inline void disable_tsc(struct task_struct *prev_p,
 593                                struct task_struct *next_p)
 594 {
 595         struct thread_info *prev, *next;
 596
 597         /*
 598          * gcc should eliminate the ->thread_info dereference if
 599          * has_secure_computing returns 0 at compile time (SECCOMP=n).
 600          */
 601         prev = task_thread_info(prev_p);
 602         next = task_thread_info(next_p);
 603
 604         if (has_secure_computing(prev) || has_secure_computing(next)) {
 605                 /* slow path here */
 606                 if (has_secure_computing(prev) &&
 607                     !has_secure_computing(next)) {
 608                         write_cr4(read_cr4() & ~X86_CR4_TSD);
 609                 } else if (!has_secure_computing(prev) &&
 610                            has_secure_computing(next))
 611                         write_cr4(read_cr4() | X86_CR4_TSD);
 612         }
 613 }
 614
 615 /*
 616  *      switch_to(x,yn) should switch tasks from x to y.
 617  *
 618  * We fsave/fwait so that an exception goes off at the right time
 619  * (as a call from the fsave or fwait in effect) rather than to
 620  * the wrong process. Lazy FP saving no longer makes any sense
 621  * with modern CPU's, and this simplifies a lot of things (SMP
 622  * and UP become the same).
 623  *
 624  * NOTE! We used to use the x86 hardware context switching. The
 625  * reason for not using it any more becomes apparent when you
 626  * try to recover gracefully from saved state that is no longer
 627  * valid (stale segment register values in particular). With the
 628  * hardware task-switch, there is no way to fix up bad state in
 629  * a reasonable manner.
 630  *
 631  * The fact that Intel documents the hardware task-switching to
 632  * be slow is a fairly red herring - this code is not noticeably
 633  * faster. However, there _is_ some room for improvement here,
 634  * so the performance issues may eventually be a valid point.
 635  * More important, however, is the fact that this allows us much
 636  * more flexibility.
 637  *
 638  * The return value (in %eax) will be the "prev" task after
 639  * the task-switch, and shows up in ret_from_fork in entry.S,
 640  * for example.
 641  */
 642 struct task_struct fastcall * __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
 643 {
 644         struct thread_struct *prev = &prev_p->thread,
 645                                  *next = &next_p->thread;
 646         int cpu = smp_processor_id();
 647         struct tss_struct *tss = &per_cpu(init_tss, cpu);
 648
 649         /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
 650
 651         __unlazy_fpu(prev_p);
 652
 653         /*
 654          * Reload esp0.
 655          */
 656         load_esp0(tss, next);
 657
 658         /*
 659          * Save away %fs and %gs. No need to save %es and %ds, as
 660          * those are always kernel segments while inside the kernel.
 661          * Doing this before setting the new TLS descriptors avoids
 662          * the situation where we temporarily have non-reloadable
 663          * segments in %fs and %gs.  This could be an issue if the
 664          * NMI handler ever used %fs or %gs (it does not today), or
 665          * if the kernel is running inside of a hypervisor layer.
 666          */
 667         savesegment(fs, prev->fs);
 668         savesegment(gs, prev->gs);
 669
 670         /*
 671          * Load the per-thread Thread-Local Storage descriptor.
 672          */
 673         load_TLS(next, cpu);
 674
 675         /*
 676          * Restore %fs and %gs if needed.
 677          *
 678          * Glibc normally makes %fs be zero, and %gs is one of
 679          * the TLS segments.
 680          */
 681         if (unlikely(prev->fs | next->fs))
 682                 loadsegment(fs, next->fs);
 683
 684         if (prev->gs | next->gs)
 685                 loadsegment(gs, next->gs);
 686
 687         /*
 688          * Restore IOPL if needed.
 689          */
 690         if (unlikely(prev->iopl != next->iopl))
 691                 set_iopl_mask(next->iopl);
 692
 693         /*
 694          * Now maybe handle debug registers and/or IO bitmaps
 695          */
 696         if (unlikely((task_thread_info(next_p)->flags & _TIF_WORK_CTXSW)
 697             || test_tsk_thread_flag(prev_p, TIF_IO_BITMAP)))
 698                 __switch_to_xtra(next_p, tss);
 699
 700         disable_tsc(prev_p, next_p);
 701
 702         return prev_p;
 703 }
 704
 705 asmlinkage int sys_fork(struct pt_regs regs)
 706 {
 707         return do_fork(SIGCHLD, regs.esp, &regs, 0, NULL, NULL);
 708 }
 709
 710 asmlinkage int sys_clone(struct pt_regs regs)
 711 {
 712         unsigned long clone_flags;
 713         unsigned long newsp;
 714         int __user *parent_tidptr, *child_tidptr;
 715
 716         clone_flags = regs.ebx;
 717         newsp = regs.ecx;
 718         parent_tidptr = (int __user *)regs.edx;
 719         child_tidptr = (int __user *)regs.edi;
 720         if (!newsp)
 721                 newsp = regs.esp;
 722         return do_fork(clone_flags, newsp, &regs, 0, parent_tidptr, child_tidptr);
 723 }
 724
 725 /*
 726  * This is trivial, and on the face of it looks like it
 727  * could equally well be done in user mode.
 728  *
 729  * Not so, for quite unobvious reasons - register pressure.
 730  * In user mode vfork() cannot have a stack frame, and if
 731  * done by calling the "clone()" system call directly, you
 732  * do not have enough call-clobbered registers to hold all
 733  * the information you need.
 734  */
 735 asmlinkage int sys_vfork(struct pt_regs regs)
 736 {
 737         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs.esp, &regs, 0, NULL, NULL);
 738 }
 739
 740 /*
 741  * sys_execve() executes a new program.
 742  */
 743 asmlinkage int sys_execve(struct pt_regs regs)
 744 {
 745         int error;
 746         char * filename;
 747
 748         filename = getname((char __user *) regs.ebx);
 749         error = PTR_ERR(filename);
 750         if (IS_ERR(filename))
 751                 goto out;
 752         error = do_execve(filename,
 753                         (char __user * __user *) regs.ecx,
 754                         (char __user * __user *) regs.edx,
 755                         &regs);
 756         if (error == 0) {
 757                 task_lock(current);
 758                 current->ptrace &= ~PT_DTRACE;
 759                 task_unlock(current);
 760                 /* Make sure we don't return using sysenter.. */
 761                 set_thread_flag(TIF_IRET);
 762         }
 763         putname(filename);
 764 out:
 765         return error;
 766 }
 767
 768 #define top_esp                (THREAD_SIZE - sizeof(unsigned long))
 769 #define top_ebp                (THREAD_SIZE - 2*sizeof(unsigned long))
 770
 771 unsigned long get_wchan(struct task_struct *p)
 772 {
 773         unsigned long ebp, esp, eip;
 774         unsigned long stack_page;
 775         int count = 0;
 776         if (!p || p == current || p->state == TASK_RUNNING)
 777                 return 0;
 778         stack_page = (unsigned long)task_stack_page(p);
 779         esp = p->thread.esp;
 780         if (!stack_page || esp < stack_page || esp > top_esp+stack_page)
 781                 return 0;
 782         /* include/asm-i386/system.h:switch_to() pushes ebp last. */
 783         ebp = *(unsigned long *) esp;
 784         do {
 785                 if (ebp < stack_page || ebp > top_ebp+stack_page)
 786                         return 0;
 787                 eip = *(unsigned long *) (ebp+4);
 788                 if (!in_sched_functions(eip))
 789                         return eip;
 790                 ebp = *(unsigned long *) ebp;
 791         } while (count++ < 16);
 792         return 0;
 793 }
 794
 795 /*
 796  * sys_alloc_thread_area: get a yet unused TLS descriptor index.
 797  */
 798 static int get_free_idx(void)
 799 {
 800         struct thread_struct *t = &current->thread;
 801         int idx;
 802
 803         for (idx = 0; idx < GDT_ENTRY_TLS_ENTRIES; idx++)
 804                 if (desc_empty(t->tls_array + idx))
 805                         return idx + GDT_ENTRY_TLS_MIN;
 806         return -ESRCH;
 807 }
 808
 809 /*
 810  * Set a given TLS descriptor:
 811  */
 812 asmlinkage int sys_set_thread_area(struct user_desc __user *u_info)
 813 {
 814         struct thread_struct *t = &current->thread;
 815         struct user_desc info;
 816         struct desc_struct *desc;
 817         int cpu, idx;
 818
 819         if (copy_from_user(&info, u_info, sizeof(info)))
 820                 return -EFAULT;
 821         idx = info.entry_number;
 822
 823         /*
 824          * index -1 means the kernel should try to find and
 825          * allocate an empty descriptor:
 826          */
 827         if (idx == -1) {
 828                 idx = get_free_idx();
 829                 if (idx < 0)
 830                         return idx;
 831                 if (put_user(idx, &u_info->entry_number))
 832                         return -EFAULT;
 833         }
 834
 835         if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
 836                 return -EINVAL;
 837
 838         desc = t->tls_array + idx - GDT_ENTRY_TLS_MIN;
 839
 840         /*
 841          * We must not get preempted while modifying the TLS.
 842          */
 843         cpu = get_cpu();
 844
 845         if (LDT_empty(&info)) {
 846                 desc->a = 0;
 847                 desc->b = 0;
 848         } else {
 849                 desc->a = LDT_entry_a(&info);
 850                 desc->b = LDT_entry_b(&info);
 851         }
 852         load_TLS(t, cpu);
 853
 854         put_cpu();
 855
 856         return 0;
 857 }
 858
 859 /*
 860  * Get the current Thread-Local Storage area:
 861  */
 862
 863 #define GET_BASE(desc) ( \
 864         (((desc)->a >> 16) & 0x0000ffff) | \
 865         (((desc)->b << 16) & 0x00ff0000) | \
 866         ( (desc)->b        & 0xff000000)   )
 867
 868 #define GET_LIMIT(desc) ( \
 869         ((desc)->a & 0x0ffff) | \
 870          ((desc)->b & 0xf0000) )
 871
 872 #define GET_32BIT(desc)         (((desc)->b >> 22) & 1)
 873 #define GET_CONTENTS(desc)      (((desc)->b >> 10) & 3)
 874 #define GET_WRITABLE(desc)      (((desc)->b >>  9) & 1)
 875 #define GET_LIMIT_PAGES(desc)   (((desc)->b >> 23) & 1)
 876 #define GET_PRESENT(desc)       (((desc)->b >> 15) & 1)
 877 #define GET_USEABLE(desc)       (((desc)->b >> 20) & 1)
 878
 879 asmlinkage int sys_get_thread_area(struct user_desc __user *u_info)
 880 {
 881         struct user_desc info;
 882         struct desc_struct *desc;
 883         int idx;
 884
 885         if (get_user(idx, &u_info->entry_number))
 886                 return -EFAULT;
 887         if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
 888                 return -EINVAL;
 889
 890         memset(&info, 0, sizeof(info));
 891
 892         desc = current->thread.tls_array + idx - GDT_ENTRY_TLS_MIN;
 893
 894         info.entry_number = idx;
 895         info.base_addr = GET_BASE(desc);
 896         info.limit = GET_LIMIT(desc);
 897         info.seg_32bit = GET_32BIT(desc);
 898         info.contents = GET_CONTENTS(desc);
 899         info.read_exec_only = !GET_WRITABLE(desc);
 900         info.limit_in_pages = GET_LIMIT_PAGES(desc);
 901         info.seg_not_present = !GET_PRESENT(desc);
 902         info.useable = GET_USEABLE(desc);
 903
 904         if (copy_to_user(u_info, &info, sizeof(info)))
 905                 return -EFAULT;
 906         return 0;
 907 }
 908
 909 unsigned long arch_align_stack(unsigned long sp)
 910 {
 911         if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
 912                 sp -= get_random_int() % 8192;
 913         return sp & ~0xf;
 914 }