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 #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         if (!hlt_counter && boot_cpu_data.hlt_works_ok) {
 103                 local_irq_disable();
 104                 if (!need_resched())
 105                         safe_halt();
 106                 else
 107                         local_irq_enable();
 108         } else {
 109                 cpu_relax();
 110         }
 111 }
 112 #ifdef CONFIG_APM_MODULE
 113 EXPORT_SYMBOL(default_idle);
 114 #endif
 115
 116 /*
 117  * On SMP it's slightly faster (but much more power-consuming!)
 118  * to poll the ->work.need_resched flag instead of waiting for the
 119  * cross-CPU IPI to arrive. Use this option with caution.
 120  */
 121 static void poll_idle (void)
 122 {
 123         int oldval;
 124
 125         local_irq_enable();
 126
 127         /*
 128          * Deal with another CPU just having chosen a thread to
 129          * run here:
 130          */
 131         oldval = test_and_clear_thread_flag(TIF_NEED_RESCHED);
 132
 133         if (!oldval) {
 134                 set_thread_flag(TIF_POLLING_NRFLAG);
 135                 asm volatile(
 136                         "2:"
 137                         "testl %0, %1;"
 138                         "rep; nop;"
 139                         "je 2b;"
 140                         : : "i"(_TIF_NEED_RESCHED), "m" (current_thread_info()->flags));
 141
 142                 clear_thread_flag(TIF_POLLING_NRFLAG);
 143         } else {
 144                 set_need_resched();
 145         }
 146 }
 147
 148 #ifdef CONFIG_HOTPLUG_CPU
 149 #include <asm/nmi.h>
 150 /* We don't actually take CPU down, just spin without interrupts. */
 151 static inline void play_dead(void)
 152 {
 153         /* This must be done before dead CPU ack */
 154         cpu_exit_clear();
 155         wbinvd();
 156         mb();
 157         /* Ack it */
 158         __get_cpu_var(cpu_state) = CPU_DEAD;
 159
 160         /*
 161          * With physical CPU hotplug, we should halt the cpu
 162          */
 163         local_irq_disable();
 164         while (1)
 165                 halt();
 166 }
 167 #else
 168 static inline void play_dead(void)
 169 {
 170         BUG();
 171 }
 172 #endif /* CONFIG_HOTPLUG_CPU */
 173
 174 /*
 175  * The idle thread. There's no useful work to be
 176  * done, so just try to conserve power and have a
 177  * low exit latency (ie sit in a loop waiting for
 178  * somebody to say that they'd like to reschedule)
 179  */
 180 void cpu_idle(void)
 181 {
 182         int cpu = raw_smp_processor_id();
 183
 184         /* endless idle loop with no priority at all */
 185         while (1) {
 186                 while (!need_resched()) {
 187                         void (*idle)(void);
 188
 189                         if (__get_cpu_var(cpu_idle_state))
 190                                 __get_cpu_var(cpu_idle_state) = 0;
 191
 192                         rmb();
 193                         idle = pm_idle;
 194
 195                         if (!idle)
 196                                 idle = default_idle;
 197
 198                         if (cpu_is_offline(cpu))
 199                                 play_dead();
 200
 201                         __get_cpu_var(irq_stat).idle_timestamp = jiffies;
 202                         idle();
 203                 }
 204                 schedule();
 205         }
 206 }
 207
 208 void cpu_idle_wait(void)
 209 {
 210         unsigned int cpu, this_cpu = get_cpu();
 211         cpumask_t map;
 212
 213         set_cpus_allowed(current, cpumask_of_cpu(this_cpu));
 214         put_cpu();
 215
 216         cpus_clear(map);
 217         for_each_online_cpu(cpu) {
 218                 per_cpu(cpu_idle_state, cpu) = 1;
 219                 cpu_set(cpu, map);
 220         }
 221
 222         __get_cpu_var(cpu_idle_state) = 0;
 223
 224         wmb();
 225         do {
 226                 ssleep(1);
 227                 for_each_online_cpu(cpu) {
 228                         if (cpu_isset(cpu, map) && !per_cpu(cpu_idle_state, cpu))
 229                                 cpu_clear(cpu, map);
 230                 }
 231                 cpus_and(map, map, cpu_online_map);
 232         } while (!cpus_empty(map));
 233 }
 234 EXPORT_SYMBOL_GPL(cpu_idle_wait);
 235
 236 /*
 237  * This uses new MONITOR/MWAIT instructions on P4 processors with PNI,
 238  * which can obviate IPI to trigger checking of need_resched.
 239  * We execute MONITOR against need_resched and enter optimized wait state
 240  * through MWAIT. Whenever someone changes need_resched, we would be woken
 241  * up from MWAIT (without an IPI).
 242  */
 243 static void mwait_idle(void)
 244 {
 245         local_irq_enable();
 246
 247         if (!need_resched()) {
 248                 set_thread_flag(TIF_POLLING_NRFLAG);
 249                 do {
 250                         __monitor((void *)&current_thread_info()->flags, 0, 0);
 251                         if (need_resched())
 252                                 break;
 253                         __mwait(0, 0);
 254                 } while (!need_resched());
 255                 clear_thread_flag(TIF_POLLING_NRFLAG);
 256         }
 257 }
 258
 259 void __devinit select_idle_routine(const struct cpuinfo_x86 *c)
 260 {
 261         if (cpu_has(c, X86_FEATURE_MWAIT)) {
 262                 printk("monitor/mwait feature present.\n");
 263                 /*
 264                  * Skip, if setup has overridden idle.
 265                  * One CPU supports mwait => All CPUs supports mwait
 266                  */
 267                 if (!pm_idle) {
 268                         printk("using mwait in idle threads.\n");
 269                         pm_idle = mwait_idle;
 270                 }
 271         }
 272 }
 273
 274 static int __init idle_setup (char *str)
 275 {
 276         if (!strncmp(str, "poll", 4)) {
 277                 printk("using polling idle threads.\n");
 278                 pm_idle = poll_idle;
 279 #ifdef CONFIG_X86_SMP
 280                 if (smp_num_siblings > 1)
 281                         printk("WARNING: polling idle and HT enabled, performance may degrade.\n");
 282 #endif
 283         } else if (!strncmp(str, "halt", 4)) {
 284                 printk("using halt in idle threads.\n");
 285                 pm_idle = default_idle;
 286         }
 287
 288         boot_option_idle_override = 1;
 289         return 1;
 290 }
 291
 292 __setup("idle=", idle_setup);
 293
 294 void show_regs(struct pt_regs * regs)
 295 {
 296         unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L;
 297
 298         printk("\n");
 299         printk("Pid: %d, comm: %20s\n", current->pid, current->comm);
 300         printk("EIP: %04x:[<%08lx>] CPU: %d\n",0xffff & regs->xcs,regs->eip, smp_processor_id());
 301         print_symbol("EIP is at %s\n", regs->eip);
 302
 303         if (user_mode(regs))
 304                 printk(" ESP: %04x:%08lx",0xffff & regs->xss,regs->esp);
 305         printk(" EFLAGS: %08lx    %s  (%s)\n",
 306                regs->eflags, print_tainted(), system_utsname.release);
 307         printk("EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
 308                 regs->eax,regs->ebx,regs->ecx,regs->edx);
 309         printk("ESI: %08lx EDI: %08lx EBP: %08lx",
 310                 regs->esi, regs->edi, regs->ebp);
 311         printk(" DS: %04x ES: %04x\n",
 312                 0xffff & regs->xds,0xffff & regs->xes);
 313
 314         cr0 = read_cr0();
 315         cr2 = read_cr2();
 316         cr3 = read_cr3();
 317         if (current_cpu_data.x86 > 4) {
 318                 cr4 = read_cr4();
 319         }
 320         printk("CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n", cr0, cr2, cr3, cr4);
 321         show_trace(NULL, &regs->esp);
 322 }
 323
 324 /*
 325  * This gets run with %ebx containing the
 326  * function to call, and %edx containing
 327  * the "args".
 328  */
 329 extern void kernel_thread_helper(void);
 330 __asm__(".section .text\n"
 331         ".align 4\n"
 332         "kernel_thread_helper:\n\t"
 333         "movl %edx,%eax\n\t"
 334         "pushl %edx\n\t"
 335         "call *%ebx\n\t"
 336         "pushl %eax\n\t"
 337         "call do_exit\n"
 338         ".previous");
 339
 340 /*
 341  * Create a kernel thread
 342  */
 343 int kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
 344 {
 345         struct pt_regs regs;
 346
 347         memset(&regs, 0, sizeof(regs));
 348
 349         regs.ebx = (unsigned long) fn;
 350         regs.edx = (unsigned long) arg;
 351
 352         regs.xds = __USER_DS;
 353         regs.xes = __USER_DS;
 354         regs.orig_eax = -1;
 355         regs.eip = (unsigned long) kernel_thread_helper;
 356         regs.xcs = __KERNEL_CS;
 357         regs.eflags = X86_EFLAGS_IF | X86_EFLAGS_SF | X86_EFLAGS_PF | 0x2;
 358
 359         /* Ok, create the new process.. */
 360         return do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0, &regs, 0, NULL, NULL);
 361 }
 362 EXPORT_SYMBOL(kernel_thread);
 363
 364 /*
 365  * Free current thread data structures etc..
 366  */
 367 void exit_thread(void)
 368 {
 369         struct task_struct *tsk = current;
 370         struct thread_struct *t = &tsk->thread;
 371
 372         /*
 373          * Remove function-return probe instances associated with this task
 374          * and put them back on the free list. Do not insert an exit probe for
 375          * this function, it will be disabled by kprobe_flush_task if you do.
 376          */
 377         kprobe_flush_task(tsk);
 378
 379         /* The process may have allocated an io port bitmap... nuke it. */
 380         if (unlikely(NULL != t->io_bitmap_ptr)) {
 381                 int cpu = get_cpu();
 382                 struct tss_struct *tss = &per_cpu(init_tss, cpu);
 383
 384                 kfree(t->io_bitmap_ptr);
 385                 t->io_bitmap_ptr = NULL;
 386                 /*
 387                  * Careful, clear this in the TSS too:
 388                  */
 389                 memset(tss->io_bitmap, 0xff, tss->io_bitmap_max);
 390                 t->io_bitmap_max = 0;
 391                 tss->io_bitmap_owner = NULL;
 392                 tss->io_bitmap_max = 0;
 393                 tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
 394                 put_cpu();
 395         }
 396 }
 397
 398 void flush_thread(void)
 399 {
 400         struct task_struct *tsk = current;
 401
 402         /*
 403          * Remove function-return probe instances associated with this task
 404          * and put them back on the free list. Do not insert an exit probe for
 405          * this function, it will be disabled by kprobe_flush_task if you do.
 406          */
 407         kprobe_flush_task(tsk);
 408
 409         memset(tsk->thread.debugreg, 0, sizeof(unsigned long)*8);
 410         memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
 411         /*
 412          * Forget coprocessor state..
 413          */
 414         clear_fpu(tsk);
 415         clear_used_math();
 416 }
 417
 418 void release_thread(struct task_struct *dead_task)
 419 {
 420         if (dead_task->mm) {
 421                 // temporary debugging check
 422                 if (dead_task->mm->context.size) {
 423                         printk("WARNING: dead process %8s still has LDT? <%p/%d>\n",
 424                                         dead_task->comm,
 425                                         dead_task->mm->context.ldt,
 426                                         dead_task->mm->context.size);
 427                         BUG();
 428                 }
 429         }
 430
 431         release_vm86_irqs(dead_task);
 432 }
 433
 434 /*
 435  * This gets called before we allocate a new thread and copy
 436  * the current task into it.
 437  */
 438 void prepare_to_copy(struct task_struct *tsk)
 439 {
 440         unlazy_fpu(tsk);
 441 }
 442
 443 int copy_thread(int nr, unsigned long clone_flags, unsigned long esp,
 444         unsigned long unused,
 445         struct task_struct * p, struct pt_regs * regs)
 446 {
 447         struct pt_regs * childregs;
 448         struct task_struct *tsk;
 449         int err;
 450
 451         childregs = ((struct pt_regs *) (THREAD_SIZE + (unsigned long) p->thread_info)) - 1;
 452         /*
 453          * The below -8 is to reserve 8 bytes on top of the ring0 stack.
 454          * This is necessary to guarantee that the entire "struct pt_regs"
 455          * is accessable even if the CPU haven't stored the SS/ESP registers
 456          * on the stack (interrupt gate does not save these registers
 457          * when switching to the same priv ring).
 458          * Therefore beware: accessing the xss/esp fields of the
 459          * "struct pt_regs" is possible, but they may contain the
 460          * completely wrong values.
 461          */
 462         childregs = (struct pt_regs *) ((unsigned long) childregs - 8);
 463         *childregs = *regs;
 464         childregs->eax = 0;
 465         childregs->esp = esp;
 466
 467         p->thread.esp = (unsigned long) childregs;
 468         p->thread.esp0 = (unsigned long) (childregs+1);
 469
 470         p->thread.eip = (unsigned long) ret_from_fork;
 471
 472         savesegment(fs,p->thread.fs);
 473         savesegment(gs,p->thread.gs);
 474
 475         tsk = current;
 476         if (unlikely(NULL != tsk->thread.io_bitmap_ptr)) {
 477                 p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL);
 478                 if (!p->thread.io_bitmap_ptr) {
 479                         p->thread.io_bitmap_max = 0;
 480                         return -ENOMEM;
 481                 }
 482                 memcpy(p->thread.io_bitmap_ptr, tsk->thread.io_bitmap_ptr,
 483                         IO_BITMAP_BYTES);
 484         }
 485
 486         /*
 487          * Set a new TLS for the child thread?
 488          */
 489         if (clone_flags & CLONE_SETTLS) {
 490                 struct desc_struct *desc;
 491                 struct user_desc info;
 492                 int idx;
 493
 494                 err = -EFAULT;
 495                 if (copy_from_user(&info, (void __user *)childregs->esi, sizeof(info)))
 496                         goto out;
 497                 err = -EINVAL;
 498                 if (LDT_empty(&info))
 499                         goto out;
 500
 501                 idx = info.entry_number;
 502                 if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
 503                         goto out;
 504
 505                 desc = p->thread.tls_array + idx - GDT_ENTRY_TLS_MIN;
 506                 desc->a = LDT_entry_a(&info);
 507                 desc->b = LDT_entry_b(&info);
 508         }
 509
 510         err = 0;
 511  out:
 512         if (err && p->thread.io_bitmap_ptr) {
 513                 kfree(p->thread.io_bitmap_ptr);
 514                 p->thread.io_bitmap_max = 0;
 515         }
 516         return err;
 517 }
 518
 519 /*
 520  * fill in the user structure for a core dump..
 521  */
 522 void dump_thread(struct pt_regs * regs, struct user * dump)
 523 {
 524         int i;
 525
 526 /* changed the size calculations - should hopefully work better. lbt */
 527         dump->magic = CMAGIC;
 528         dump->start_code = 0;
 529         dump->start_stack = regs->esp & ~(PAGE_SIZE - 1);
 530         dump->u_tsize = ((unsigned long) current->mm->end_code) >> PAGE_SHIFT;
 531         dump->u_dsize = ((unsigned long) (current->mm->brk + (PAGE_SIZE-1))) >> PAGE_SHIFT;
 532         dump->u_dsize -= dump->u_tsize;
 533         dump->u_ssize = 0;
 534         for (i = 0; i < 8; i++)
 535                 dump->u_debugreg[i] = current->thread.debugreg[i];
 536
 537         if (dump->start_stack < TASK_SIZE)
 538                 dump->u_ssize = ((unsigned long) (TASK_SIZE - dump->start_stack)) >> PAGE_SHIFT;
 539
 540         dump->regs.ebx = regs->ebx;
 541         dump->regs.ecx = regs->ecx;
 542         dump->regs.edx = regs->edx;
 543         dump->regs.esi = regs->esi;
 544         dump->regs.edi = regs->edi;
 545         dump->regs.ebp = regs->ebp;
 546         dump->regs.eax = regs->eax;
 547         dump->regs.ds = regs->xds;
 548         dump->regs.es = regs->xes;
 549         savesegment(fs,dump->regs.fs);
 550         savesegment(gs,dump->regs.gs);
 551         dump->regs.orig_eax = regs->orig_eax;
 552         dump->regs.eip = regs->eip;
 553         dump->regs.cs = regs->xcs;
 554         dump->regs.eflags = regs->eflags;
 555         dump->regs.esp = regs->esp;
 556         dump->regs.ss = regs->xss;
 557
 558         dump->u_fpvalid = dump_fpu (regs, &dump->i387);
 559 }
 560 EXPORT_SYMBOL(dump_thread);
 561
 562 /*
 563  * Capture the user space registers if the task is not running (in user space)
 564  */
 565 int dump_task_regs(struct task_struct *tsk, elf_gregset_t *regs)
 566 {
 567         struct pt_regs ptregs;
 568
 569         ptregs = *(struct pt_regs *)
 570                 ((unsigned long)tsk->thread_info+THREAD_SIZE - sizeof(ptregs));
 571         ptregs.xcs &= 0xffff;
 572         ptregs.xds &= 0xffff;
 573         ptregs.xes &= 0xffff;
 574         ptregs.xss &= 0xffff;
 575
 576         elf_core_copy_regs(regs, &ptregs);
 577
 578         return 1;
 579 }
 580
 581 static inline void
 582 handle_io_bitmap(struct thread_struct *next, struct tss_struct *tss)
 583 {
 584         if (!next->io_bitmap_ptr) {
 585                 /*
 586                  * Disable the bitmap via an invalid offset. We still cache
 587                  * the previous bitmap owner and the IO bitmap contents:
 588                  */
 589                 tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
 590                 return;
 591         }
 592         if (likely(next == tss->io_bitmap_owner)) {
 593                 /*
 594                  * Previous owner of the bitmap (hence the bitmap content)
 595                  * matches the next task, we dont have to do anything but
 596                  * to set a valid offset in the TSS:
 597                  */
 598                 tss->io_bitmap_base = IO_BITMAP_OFFSET;
 599                 return;
 600         }
 601         /*
 602          * Lazy TSS's I/O bitmap copy. We set an invalid offset here
 603          * and we let the task to get a GPF in case an I/O instruction
 604          * is performed.  The handler of the GPF will verify that the
 605          * faulting task has a valid I/O bitmap and, it true, does the
 606          * real copy and restart the instruction.  This will save us
 607          * redundant copies when the currently switched task does not
 608          * perform any I/O during its timeslice.
 609          */
 610         tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET_LAZY;
 611 }
 612
 613 /*
 614  * This function selects if the context switch from prev to next
 615  * has to tweak the TSC disable bit in the cr4.
 616  */
 617 static inline void disable_tsc(struct task_struct *prev_p,
 618                                struct task_struct *next_p)
 619 {
 620         struct thread_info *prev, *next;
 621
 622         /*
 623          * gcc should eliminate the ->thread_info dereference if
 624          * has_secure_computing returns 0 at compile time (SECCOMP=n).
 625          */
 626         prev = prev_p->thread_info;
 627         next = next_p->thread_info;
 628
 629         if (has_secure_computing(prev) || has_secure_computing(next)) {
 630                 /* slow path here */
 631                 if (has_secure_computing(prev) &&
 632                     !has_secure_computing(next)) {
 633                         write_cr4(read_cr4() & ~X86_CR4_TSD);
 634                 } else if (!has_secure_computing(prev) &&
 635                            has_secure_computing(next))
 636                         write_cr4(read_cr4() | X86_CR4_TSD);
 637         }
 638 }
 639
 640 /*
 641  *      switch_to(x,yn) should switch tasks from x to y.
 642  *
 643  * We fsave/fwait so that an exception goes off at the right time
 644  * (as a call from the fsave or fwait in effect) rather than to
 645  * the wrong process. Lazy FP saving no longer makes any sense
 646  * with modern CPU's, and this simplifies a lot of things (SMP
 647  * and UP become the same).
 648  *
 649  * NOTE! We used to use the x86 hardware context switching. The
 650  * reason for not using it any more becomes apparent when you
 651  * try to recover gracefully from saved state that is no longer
 652  * valid (stale segment register values in particular). With the
 653  * hardware task-switch, there is no way to fix up bad state in
 654  * a reasonable manner.
 655  *
 656  * The fact that Intel documents the hardware task-switching to
 657  * be slow is a fairly red herring - this code is not noticeably
 658  * faster. However, there _is_ some room for improvement here,
 659  * so the performance issues may eventually be a valid point.
 660  * More important, however, is the fact that this allows us much
 661  * more flexibility.
 662  *
 663  * The return value (in %eax) will be the "prev" task after
 664  * the task-switch, and shows up in ret_from_fork in entry.S,
 665  * for example.
 666  */
 667 struct task_struct fastcall * __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
 668 {
 669         struct thread_struct *prev = &prev_p->thread,
 670                                  *next = &next_p->thread;
 671         int cpu = smp_processor_id();
 672         struct tss_struct *tss = &per_cpu(init_tss, cpu);
 673
 674         /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
 675
 676         __unlazy_fpu(prev_p);
 677
 678         /*
 679          * Reload esp0.
 680          */
 681         load_esp0(tss, next);
 682
 683         /*
 684          * Save away %fs and %gs. No need to save %es and %ds, as
 685          * those are always kernel segments while inside the kernel.
 686          * Doing this before setting the new TLS descriptors avoids
 687          * the situation where we temporarily have non-reloadable
 688          * segments in %fs and %gs.  This could be an issue if the
 689          * NMI handler ever used %fs or %gs (it does not today), or
 690          * if the kernel is running inside of a hypervisor layer.
 691          */
 692         savesegment(fs, prev->fs);
 693         savesegment(gs, prev->gs);
 694
 695         /*
 696          * Load the per-thread Thread-Local Storage descriptor.
 697          */
 698         load_TLS(next, cpu);
 699
 700         /*
 701          * Restore %fs and %gs if needed.
 702          *
 703          * Glibc normally makes %fs be zero, and %gs is one of
 704          * the TLS segments.
 705          */
 706         if (unlikely(prev->fs | next->fs))
 707                 loadsegment(fs, next->fs);
 708
 709         if (prev->gs | next->gs)
 710                 loadsegment(gs, next->gs);
 711
 712         /*
 713          * Restore IOPL if needed.
 714          */
 715         if (unlikely(prev->iopl != next->iopl))
 716                 set_iopl_mask(next->iopl);
 717
 718         /*
 719          * Now maybe reload the debug registers
 720          */
 721         if (unlikely(next->debugreg[7])) {
 722                 set_debugreg(next->debugreg[0], 0);
 723                 set_debugreg(next->debugreg[1], 1);
 724                 set_debugreg(next->debugreg[2], 2);
 725                 set_debugreg(next->debugreg[3], 3);
 726                 /* no 4 and 5 */
 727                 set_debugreg(next->debugreg[6], 6);
 728                 set_debugreg(next->debugreg[7], 7);
 729         }
 730
 731         if (unlikely(prev->io_bitmap_ptr || next->io_bitmap_ptr))
 732                 handle_io_bitmap(next, tss);
 733
 734         disable_tsc(prev_p, next_p);
 735
 736         return prev_p;
 737 }
 738
 739 asmlinkage int sys_fork(struct pt_regs regs)
 740 {
 741         return do_fork(SIGCHLD, regs.esp, &regs, 0, NULL, NULL);
 742 }
 743
 744 asmlinkage int sys_clone(struct pt_regs regs)
 745 {
 746         unsigned long clone_flags;
 747         unsigned long newsp;
 748         int __user *parent_tidptr, *child_tidptr;
 749
 750         clone_flags = regs.ebx;
 751         newsp = regs.ecx;
 752         parent_tidptr = (int __user *)regs.edx;
 753         child_tidptr = (int __user *)regs.edi;
 754         if (!newsp)
 755                 newsp = regs.esp;
 756         return do_fork(clone_flags, newsp, &regs, 0, parent_tidptr, child_tidptr);
 757 }
 758
 759 /*
 760  * This is trivial, and on the face of it looks like it
 761  * could equally well be done in user mode.
 762  *
 763  * Not so, for quite unobvious reasons - register pressure.
 764  * In user mode vfork() cannot have a stack frame, and if
 765  * done by calling the "clone()" system call directly, you
 766  * do not have enough call-clobbered registers to hold all
 767  * the information you need.
 768  */
 769 asmlinkage int sys_vfork(struct pt_regs regs)
 770 {
 771         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs.esp, &regs, 0, NULL, NULL);
 772 }
 773
 774 /*
 775  * sys_execve() executes a new program.
 776  */
 777 asmlinkage int sys_execve(struct pt_regs regs)
 778 {
 779         int error;
 780         char * filename;
 781
 782         filename = getname((char __user *) regs.ebx);
 783         error = PTR_ERR(filename);
 784         if (IS_ERR(filename))
 785                 goto out;
 786         error = do_execve(filename,
 787                         (char __user * __user *) regs.ecx,
 788                         (char __user * __user *) regs.edx,
 789                         &regs);
 790         if (error == 0) {
 791                 task_lock(current);
 792                 current->ptrace &= ~PT_DTRACE;
 793                 task_unlock(current);
 794                 /* Make sure we don't return using sysenter.. */
 795                 set_thread_flag(TIF_IRET);
 796         }
 797         putname(filename);
 798 out:
 799         return error;
 800 }
 801
 802 #define top_esp                (THREAD_SIZE - sizeof(unsigned long))
 803 #define top_ebp                (THREAD_SIZE - 2*sizeof(unsigned long))
 804
 805 unsigned long get_wchan(struct task_struct *p)
 806 {
 807         unsigned long ebp, esp, eip;
 808         unsigned long stack_page;
 809         int count = 0;
 810         if (!p || p == current || p->state == TASK_RUNNING)
 811                 return 0;
 812         stack_page = (unsigned long)p->thread_info;
 813         esp = p->thread.esp;
 814         if (!stack_page || esp < stack_page || esp > top_esp+stack_page)
 815                 return 0;
 816         /* include/asm-i386/system.h:switch_to() pushes ebp last. */
 817         ebp = *(unsigned long *) esp;
 818         do {
 819                 if (ebp < stack_page || ebp > top_ebp+stack_page)
 820                         return 0;
 821                 eip = *(unsigned long *) (ebp+4);
 822                 if (!in_sched_functions(eip))
 823                         return eip;
 824                 ebp = *(unsigned long *) ebp;
 825         } while (count++ < 16);
 826         return 0;
 827 }
 828 EXPORT_SYMBOL(get_wchan);
 829
 830 /*
 831  * sys_alloc_thread_area: get a yet unused TLS descriptor index.
 832  */
 833 static int get_free_idx(void)
 834 {
 835         struct thread_struct *t = &current->thread;
 836         int idx;
 837
 838         for (idx = 0; idx < GDT_ENTRY_TLS_ENTRIES; idx++)
 839                 if (desc_empty(t->tls_array + idx))
 840                         return idx + GDT_ENTRY_TLS_MIN;
 841         return -ESRCH;
 842 }
 843
 844 /*
 845  * Set a given TLS descriptor:
 846  */
 847 asmlinkage int sys_set_thread_area(struct user_desc __user *u_info)
 848 {
 849         struct thread_struct *t = &current->thread;
 850         struct user_desc info;
 851         struct desc_struct *desc;
 852         int cpu, idx;
 853
 854         if (copy_from_user(&info, u_info, sizeof(info)))
 855                 return -EFAULT;
 856         idx = info.entry_number;
 857
 858         /*
 859          * index -1 means the kernel should try to find and
 860          * allocate an empty descriptor:
 861          */
 862         if (idx == -1) {
 863                 idx = get_free_idx();
 864                 if (idx < 0)
 865                         return idx;
 866                 if (put_user(idx, &u_info->entry_number))
 867                         return -EFAULT;
 868         }
 869
 870         if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
 871                 return -EINVAL;
 872
 873         desc = t->tls_array + idx - GDT_ENTRY_TLS_MIN;
 874
 875         /*
 876          * We must not get preempted while modifying the TLS.
 877          */
 878         cpu = get_cpu();
 879
 880         if (LDT_empty(&info)) {
 881                 desc->a = 0;
 882                 desc->b = 0;
 883         } else {
 884                 desc->a = LDT_entry_a(&info);
 885                 desc->b = LDT_entry_b(&info);
 886         }
 887         load_TLS(t, cpu);
 888
 889         put_cpu();
 890
 891         return 0;
 892 }
 893
 894 /*
 895  * Get the current Thread-Local Storage area:
 896  */
 897
 898 #define GET_BASE(desc) ( \
 899         (((desc)->a >> 16) & 0x0000ffff) | \
 900         (((desc)->b << 16) & 0x00ff0000) | \
 901         ( (desc)->b        & 0xff000000)   )
 902
 903 #define GET_LIMIT(desc) ( \
 904         ((desc)->a & 0x0ffff) | \
 905          ((desc)->b & 0xf0000) )
 906
 907 #define GET_32BIT(desc)         (((desc)->b >> 22) & 1)
 908 #define GET_CONTENTS(desc)      (((desc)->b >> 10) & 3)
 909 #define GET_WRITABLE(desc)      (((desc)->b >>  9) & 1)
 910 #define GET_LIMIT_PAGES(desc)   (((desc)->b >> 23) & 1)
 911 #define GET_PRESENT(desc)       (((desc)->b >> 15) & 1)
 912 #define GET_USEABLE(desc)       (((desc)->b >> 20) & 1)
 913
 914 asmlinkage int sys_get_thread_area(struct user_desc __user *u_info)
 915 {
 916         struct user_desc info;
 917         struct desc_struct *desc;
 918         int idx;
 919
 920         if (get_user(idx, &u_info->entry_number))
 921                 return -EFAULT;
 922         if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
 923                 return -EINVAL;
 924
 925         memset(&info, 0, sizeof(info));
 926
 927         desc = current->thread.tls_array + idx - GDT_ENTRY_TLS_MIN;
 928
 929         info.entry_number = idx;
 930         info.base_addr = GET_BASE(desc);
 931         info.limit = GET_LIMIT(desc);
 932         info.seg_32bit = GET_32BIT(desc);
 933         info.contents = GET_CONTENTS(desc);
 934         info.read_exec_only = !GET_WRITABLE(desc);
 935         info.limit_in_pages = GET_LIMIT_PAGES(desc);
 936         info.seg_not_present = !GET_PRESENT(desc);
 937         info.useable = GET_USEABLE(desc);
 938
 939         if (copy_to_user(u_info, &info, sizeof(info)))
 940                 return -EFAULT;
 941         return 0;
 942 }
 943
 944 unsigned long arch_align_stack(unsigned long sp)
 945 {
 946         if (randomize_va_space)
 947                 sp -= get_random_int() % 8192;
 948         return sp & ~0xf;
 949 }