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/irq.h>
  51 #include <asm/desc.h>
  52 #ifdef CONFIG_MATH_EMULATION
  53 #include <asm/math_emu.h>
  54 #endif
  55
  56 #include <linux/irq.h>
  57 #include <linux/err.h>
  58
  59 #include <asm/tlbflush.h>
  60 #include <asm/cpu.h>
  61
  62 asmlinkage void ret_from_fork(void) __asm__("ret_from_fork");
  63
  64 static int hlt_counter;
  65
  66 unsigned long boot_option_idle_override = 0;
  67 EXPORT_SYMBOL(boot_option_idle_override);
  68
  69 /*
  70  * Return saved PC of a blocked thread.
  71  */
  72 unsigned long thread_saved_pc(struct task_struct *tsk)
  73 {
  74         return ((unsigned long *)tsk->thread.esp)[3];
  75 }
  76
  77 /*
  78  * Powermanagement idle function, if any..
  79  */
  80 void (*pm_idle)(void);
  81 EXPORT_SYMBOL(pm_idle);
  82 static DEFINE_PER_CPU(unsigned int, cpu_idle_state);
  83
  84 void disable_hlt(void)
  85 {
  86         hlt_counter++;
  87 }
  88
  89 EXPORT_SYMBOL(disable_hlt);
  90
  91 void enable_hlt(void)
  92 {
  93         hlt_counter--;
  94 }
  95
  96 EXPORT_SYMBOL(enable_hlt);
  97
  98 /*
  99  * We use this if we don't have any better
 100  * idle routine..
 101  */
 102 void default_idle(void)
 103 {
 104         if (!hlt_counter && boot_cpu_data.hlt_works_ok) {
 105                 local_irq_disable();
 106                 if (!need_resched())
 107                         safe_halt();
 108                 else
 109                         local_irq_enable();
 110         } else {
 111                 cpu_relax();
 112         }
 113 }
 114 #ifdef CONFIG_APM_MODULE
 115 EXPORT_SYMBOL(default_idle);
 116 #endif
 117
 118 /*
 119  * On SMP it's slightly faster (but much more power-consuming!)
 120  * to poll the ->work.need_resched flag instead of waiting for the
 121  * cross-CPU IPI to arrive. Use this option with caution.
 122  */
 123 static void poll_idle (void)
 124 {
 125         int oldval;
 126
 127         local_irq_enable();
 128
 129         /*
 130          * Deal with another CPU just having chosen a thread to
 131          * run here:
 132          */
 133         oldval = test_and_clear_thread_flag(TIF_NEED_RESCHED);
 134
 135         if (!oldval) {
 136                 set_thread_flag(TIF_POLLING_NRFLAG);
 137                 asm volatile(
 138                         "2:"
 139                         "testl %0, %1;"
 140                         "rep; nop;"
 141                         "je 2b;"
 142                         : : "i"(_TIF_NEED_RESCHED), "m" (current_thread_info()->flags));
 143
 144                 clear_thread_flag(TIF_POLLING_NRFLAG);
 145         } else {
 146                 set_need_resched();
 147         }
 148 }
 149
 150 #ifdef CONFIG_HOTPLUG_CPU
 151 #include <asm/nmi.h>
 152 /* We don't actually take CPU down, just spin without interrupts. */
 153 static inline void play_dead(void)
 154 {
 155         /* This must be done before dead CPU ack */
 156         cpu_exit_clear();
 157         wbinvd();
 158         mb();
 159         /* Ack it */
 160         __get_cpu_var(cpu_state) = CPU_DEAD;
 161
 162         /*
 163          * With physical CPU hotplug, we should halt the cpu
 164          */
 165         local_irq_disable();
 166         while (1)
 167                 __asm__ __volatile__("hlt":::"memory");
 168 }
 169 #else
 170 static inline void play_dead(void)
 171 {
 172         BUG();
 173 }
 174 #endif /* CONFIG_HOTPLUG_CPU */
 175
 176 /*
 177  * The idle thread. There's no useful work to be
 178  * done, so just try to conserve power and have a
 179  * low exit latency (ie sit in a loop waiting for
 180  * somebody to say that they'd like to reschedule)
 181  */
 182 void cpu_idle(void)
 183 {
 184         int cpu = raw_smp_processor_id();
 185
 186         /* endless idle loop with no priority at all */
 187         while (1) {
 188                 while (!need_resched()) {
 189                         void (*idle)(void);
 190
 191                         if (__get_cpu_var(cpu_idle_state))
 192                                 __get_cpu_var(cpu_idle_state) = 0;
 193
 194                         rmb();
 195                         idle = pm_idle;
 196
 197                         if (!idle)
 198                                 idle = default_idle;
 199
 200                         if (cpu_is_offline(cpu))
 201                                 play_dead();
 202
 203                         __get_cpu_var(irq_stat).idle_timestamp = jiffies;
 204                         idle();
 205                 }
 206                 schedule();
 207         }
 208 }
 209
 210 void cpu_idle_wait(void)
 211 {
 212         unsigned int cpu, this_cpu = get_cpu();
 213         cpumask_t map;
 214
 215         set_cpus_allowed(current, cpumask_of_cpu(this_cpu));
 216         put_cpu();
 217
 218         cpus_clear(map);
 219         for_each_online_cpu(cpu) {
 220                 per_cpu(cpu_idle_state, cpu) = 1;
 221                 cpu_set(cpu, map);
 222         }
 223
 224         __get_cpu_var(cpu_idle_state) = 0;
 225
 226         wmb();
 227         do {
 228                 ssleep(1);
 229                 for_each_online_cpu(cpu) {
 230                         if (cpu_isset(cpu, map) && !per_cpu(cpu_idle_state, cpu))
 231                                 cpu_clear(cpu, map);
 232                 }
 233                 cpus_and(map, map, cpu_online_map);
 234         } while (!cpus_empty(map));
 235 }
 236 EXPORT_SYMBOL_GPL(cpu_idle_wait);
 237
 238 /*
 239  * This uses new MONITOR/MWAIT instructions on P4 processors with PNI,
 240  * which can obviate IPI to trigger checking of need_resched.
 241  * We execute MONITOR against need_resched and enter optimized wait state
 242  * through MWAIT. Whenever someone changes need_resched, we would be woken
 243  * up from MWAIT (without an IPI).
 244  */
 245 static void mwait_idle(void)
 246 {
 247         local_irq_enable();
 248
 249         if (!need_resched()) {
 250                 set_thread_flag(TIF_POLLING_NRFLAG);
 251                 do {
 252                         __monitor((void *)&current_thread_info()->flags, 0, 0);
 253                         if (need_resched())
 254                                 break;
 255                         __mwait(0, 0);
 256                 } while (!need_resched());
 257                 clear_thread_flag(TIF_POLLING_NRFLAG);
 258         }
 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(regs))
 306                 printk(" ESP: %04x:%08lx",0xffff & regs->xss,regs->esp);
 307         printk(" EFLAGS: %08lx    %s  (%s)\n",
 308                regs->eflags, print_tainted(), system_utsname.release);
 309         printk("EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
 310                 regs->eax,regs->ebx,regs->ecx,regs->edx);
 311         printk("ESI: %08lx EDI: %08lx EBP: %08lx",
 312                 regs->esi, regs->edi, regs->ebp);
 313         printk(" DS: %04x ES: %04x\n",
 314                 0xffff & regs->xds,0xffff & regs->xes);
 315
 316         cr0 = read_cr0();
 317         cr2 = read_cr2();
 318         cr3 = read_cr3();
 319         if (current_cpu_data.x86 > 4) {
 320                 cr4 = read_cr4();
 321         }
 322         printk("CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n", cr0, cr2, cr3, cr4);
 323         show_trace(NULL, &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 __asm__(".section .text\n"
 333         ".align 4\n"
 334         "kernel_thread_helper:\n\t"
 335         "movl %edx,%eax\n\t"
 336         "pushl %edx\n\t"
 337         "call *%ebx\n\t"
 338         "pushl %eax\n\t"
 339         "call do_exit\n"
 340         ".previous");
 341
 342 /*
 343  * Create a kernel thread
 344  */
 345 int kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
 346 {
 347         struct pt_regs regs;
 348
 349         memset(&regs, 0, sizeof(regs));
 350
 351         regs.ebx = (unsigned long) fn;
 352         regs.edx = (unsigned long) arg;
 353
 354         regs.xds = __USER_DS;
 355         regs.xes = __USER_DS;
 356         regs.orig_eax = -1;
 357         regs.eip = (unsigned long) kernel_thread_helper;
 358         regs.xcs = __KERNEL_CS;
 359         regs.eflags = X86_EFLAGS_IF | X86_EFLAGS_SF | X86_EFLAGS_PF | 0x2;
 360
 361         /* Ok, create the new process.. */
 362         return do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0, &regs, 0, NULL, NULL);
 363 }
 364 EXPORT_SYMBOL(kernel_thread);
 365
 366 /*
 367  * Free current thread data structures etc..
 368  */
 369 void exit_thread(void)
 370 {
 371         struct task_struct *tsk = current;
 372         struct thread_struct *t = &tsk->thread;
 373
 374         /*
 375          * Remove function-return probe instances associated with this task
 376          * and put them back on the free list. Do not insert an exit probe for
 377          * this function, it will be disabled by kprobe_flush_task if you do.
 378          */
 379         kprobe_flush_task(tsk);
 380
 381         /* The process may have allocated an io port bitmap... nuke it. */
 382         if (unlikely(NULL != t->io_bitmap_ptr)) {
 383                 int cpu = get_cpu();
 384                 struct tss_struct *tss = &per_cpu(init_tss, cpu);
 385
 386                 kfree(t->io_bitmap_ptr);
 387                 t->io_bitmap_ptr = NULL;
 388                 /*
 389                  * Careful, clear this in the TSS too:
 390                  */
 391                 memset(tss->io_bitmap, 0xff, tss->io_bitmap_max);
 392                 t->io_bitmap_max = 0;
 393                 tss->io_bitmap_owner = NULL;
 394                 tss->io_bitmap_max = 0;
 395                 tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
 396                 put_cpu();
 397         }
 398 }
 399
 400 void flush_thread(void)
 401 {
 402         struct task_struct *tsk = current;
 403
 404         /*
 405          * Remove function-return probe instances associated with this task
 406          * and put them back on the free list. Do not insert an exit probe for
 407          * this function, it will be disabled by kprobe_flush_task if you do.
 408          */
 409         kprobe_flush_task(tsk);
 410
 411         memset(tsk->thread.debugreg, 0, sizeof(unsigned long)*8);
 412         memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
 413         /*
 414          * Forget coprocessor state..
 415          */
 416         clear_fpu(tsk);
 417         clear_used_math();
 418 }
 419
 420 void release_thread(struct task_struct *dead_task)
 421 {
 422         if (dead_task->mm) {
 423                 // temporary debugging check
 424                 if (dead_task->mm->context.size) {
 425                         printk("WARNING: dead process %8s still has LDT? <%p/%d>\n",
 426                                         dead_task->comm,
 427                                         dead_task->mm->context.ldt,
 428                                         dead_task->mm->context.size);
 429                         BUG();
 430                 }
 431         }
 432
 433         release_vm86_irqs(dead_task);
 434 }
 435
 436 /*
 437  * This gets called before we allocate a new thread and copy
 438  * the current task into it.
 439  */
 440 void prepare_to_copy(struct task_struct *tsk)
 441 {
 442         unlazy_fpu(tsk);
 443 }
 444
 445 int copy_thread(int nr, unsigned long clone_flags, unsigned long esp,
 446         unsigned long unused,
 447         struct task_struct * p, struct pt_regs * regs)
 448 {
 449         struct pt_regs * childregs;
 450         struct task_struct *tsk;
 451         int err;
 452
 453         childregs = ((struct pt_regs *) (THREAD_SIZE + (unsigned long) p->thread_info)) - 1;
 454         /*
 455          * The below -8 is to reserve 8 bytes on top of the ring0 stack.
 456          * This is necessary to guarantee that the entire "struct pt_regs"
 457          * is accessable even if the CPU haven't stored the SS/ESP registers
 458          * on the stack (interrupt gate does not save these registers
 459          * when switching to the same priv ring).
 460          * Therefore beware: accessing the xss/esp fields of the
 461          * "struct pt_regs" is possible, but they may contain the
 462          * completely wrong values.
 463          */
 464         childregs = (struct pt_regs *) ((unsigned long) childregs - 8);
 465         *childregs = *regs;
 466         childregs->eax = 0;
 467         childregs->esp = esp;
 468
 469         p->thread.esp = (unsigned long) childregs;
 470         p->thread.esp0 = (unsigned long) (childregs+1);
 471
 472         p->thread.eip = (unsigned long) ret_from_fork;
 473
 474         savesegment(fs,p->thread.fs);
 475         savesegment(gs,p->thread.gs);
 476
 477         tsk = current;
 478         if (unlikely(NULL != tsk->thread.io_bitmap_ptr)) {
 479                 p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL);
 480                 if (!p->thread.io_bitmap_ptr) {
 481                         p->thread.io_bitmap_max = 0;
 482                         return -ENOMEM;
 483                 }
 484                 memcpy(p->thread.io_bitmap_ptr, tsk->thread.io_bitmap_ptr,
 485                         IO_BITMAP_BYTES);
 486         }
 487
 488         /*
 489          * Set a new TLS for the child thread?
 490          */
 491         if (clone_flags & CLONE_SETTLS) {
 492                 struct desc_struct *desc;
 493                 struct user_desc info;
 494                 int idx;
 495
 496                 err = -EFAULT;
 497                 if (copy_from_user(&info, (void __user *)childregs->esi, sizeof(info)))
 498                         goto out;
 499                 err = -EINVAL;
 500                 if (LDT_empty(&info))
 501                         goto out;
 502
 503                 idx = info.entry_number;
 504                 if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
 505                         goto out;
 506
 507                 desc = p->thread.tls_array + idx - GDT_ENTRY_TLS_MIN;
 508                 desc->a = LDT_entry_a(&info);
 509                 desc->b = LDT_entry_b(&info);
 510         }
 511
 512         err = 0;
 513  out:
 514         if (err && p->thread.io_bitmap_ptr) {
 515                 kfree(p->thread.io_bitmap_ptr);
 516                 p->thread.io_bitmap_max = 0;
 517         }
 518         return err;
 519 }
 520
 521 /*
 522  * fill in the user structure for a core dump..
 523  */
 524 void dump_thread(struct pt_regs * regs, struct user * dump)
 525 {
 526         int i;
 527
 528 /* changed the size calculations - should hopefully work better. lbt */
 529         dump->magic = CMAGIC;
 530         dump->start_code = 0;
 531         dump->start_stack = regs->esp & ~(PAGE_SIZE - 1);
 532         dump->u_tsize = ((unsigned long) current->mm->end_code) >> PAGE_SHIFT;
 533         dump->u_dsize = ((unsigned long) (current->mm->brk + (PAGE_SIZE-1))) >> PAGE_SHIFT;
 534         dump->u_dsize -= dump->u_tsize;
 535         dump->u_ssize = 0;
 536         for (i = 0; i < 8; i++)
 537                 dump->u_debugreg[i] = current->thread.debugreg[i];
 538
 539         if (dump->start_stack < TASK_SIZE)
 540                 dump->u_ssize = ((unsigned long) (TASK_SIZE - dump->start_stack)) >> PAGE_SHIFT;
 541
 542         dump->regs.ebx = regs->ebx;
 543         dump->regs.ecx = regs->ecx;
 544         dump->regs.edx = regs->edx;
 545         dump->regs.esi = regs->esi;
 546         dump->regs.edi = regs->edi;
 547         dump->regs.ebp = regs->ebp;
 548         dump->regs.eax = regs->eax;
 549         dump->regs.ds = regs->xds;
 550         dump->regs.es = regs->xes;
 551         savesegment(fs,dump->regs.fs);
 552         savesegment(gs,dump->regs.gs);
 553         dump->regs.orig_eax = regs->orig_eax;
 554         dump->regs.eip = regs->eip;
 555         dump->regs.cs = regs->xcs;
 556         dump->regs.eflags = regs->eflags;
 557         dump->regs.esp = regs->esp;
 558         dump->regs.ss = regs->xss;
 559
 560         dump->u_fpvalid = dump_fpu (regs, &dump->i387);
 561 }
 562 EXPORT_SYMBOL(dump_thread);
 563
 564 /*
 565  * Capture the user space registers if the task is not running (in user space)
 566  */
 567 int dump_task_regs(struct task_struct *tsk, elf_gregset_t *regs)
 568 {
 569         struct pt_regs ptregs;
 570
 571         ptregs = *(struct pt_regs *)
 572                 ((unsigned long)tsk->thread_info+THREAD_SIZE - sizeof(ptregs));
 573         ptregs.xcs &= 0xffff;
 574         ptregs.xds &= 0xffff;
 575         ptregs.xes &= 0xffff;
 576         ptregs.xss &= 0xffff;
 577
 578         elf_core_copy_regs(regs, &ptregs);
 579
 580         return 1;
 581 }
 582
 583 static inline void
 584 handle_io_bitmap(struct thread_struct *next, struct tss_struct *tss)
 585 {
 586         if (!next->io_bitmap_ptr) {
 587                 /*
 588                  * Disable the bitmap via an invalid offset. We still cache
 589                  * the previous bitmap owner and the IO bitmap contents:
 590                  */
 591                 tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
 592                 return;
 593         }
 594         if (likely(next == tss->io_bitmap_owner)) {
 595                 /*
 596                  * Previous owner of the bitmap (hence the bitmap content)
 597                  * matches the next task, we dont have to do anything but
 598                  * to set a valid offset in the TSS:
 599                  */
 600                 tss->io_bitmap_base = IO_BITMAP_OFFSET;
 601                 return;
 602         }
 603         /*
 604          * Lazy TSS's I/O bitmap copy. We set an invalid offset here
 605          * and we let the task to get a GPF in case an I/O instruction
 606          * is performed.  The handler of the GPF will verify that the
 607          * faulting task has a valid I/O bitmap and, it true, does the
 608          * real copy and restart the instruction.  This will save us
 609          * redundant copies when the currently switched task does not
 610          * perform any I/O during its timeslice.
 611          */
 612         tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET_LAZY;
 613 }
 614
 615 /*
 616  * This function selects if the context switch from prev to next
 617  * has to tweak the TSC disable bit in the cr4.
 618  */
 619 static inline void disable_tsc(struct task_struct *prev_p,
 620                                struct task_struct *next_p)
 621 {
 622         struct thread_info *prev, *next;
 623
 624         /*
 625          * gcc should eliminate the ->thread_info dereference if
 626          * has_secure_computing returns 0 at compile time (SECCOMP=n).
 627          */
 628         prev = prev_p->thread_info;
 629         next = next_p->thread_info;
 630
 631         if (has_secure_computing(prev) || has_secure_computing(next)) {
 632                 /* slow path here */
 633                 if (has_secure_computing(prev) &&
 634                     !has_secure_computing(next)) {
 635                         write_cr4(read_cr4() & ~X86_CR4_TSD);
 636                 } else if (!has_secure_computing(prev) &&
 637                            has_secure_computing(next))
 638                         write_cr4(read_cr4() | X86_CR4_TSD);
 639         }
 640 }
 641
 642 /*
 643  *      switch_to(x,yn) should switch tasks from x to y.
 644  *
 645  * We fsave/fwait so that an exception goes off at the right time
 646  * (as a call from the fsave or fwait in effect) rather than to
 647  * the wrong process. Lazy FP saving no longer makes any sense
 648  * with modern CPU's, and this simplifies a lot of things (SMP
 649  * and UP become the same).
 650  *
 651  * NOTE! We used to use the x86 hardware context switching. The
 652  * reason for not using it any more becomes apparent when you
 653  * try to recover gracefully from saved state that is no longer
 654  * valid (stale segment register values in particular). With the
 655  * hardware task-switch, there is no way to fix up bad state in
 656  * a reasonable manner.
 657  *
 658  * The fact that Intel documents the hardware task-switching to
 659  * be slow is a fairly red herring - this code is not noticeably
 660  * faster. However, there _is_ some room for improvement here,
 661  * so the performance issues may eventually be a valid point.
 662  * More important, however, is the fact that this allows us much
 663  * more flexibility.
 664  *
 665  * The return value (in %eax) will be the "prev" task after
 666  * the task-switch, and shows up in ret_from_fork in entry.S,
 667  * for example.
 668  */
 669 struct task_struct fastcall * __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
 670 {
 671         struct thread_struct *prev = &prev_p->thread,
 672                                  *next = &next_p->thread;
 673         int cpu = smp_processor_id();
 674         struct tss_struct *tss = &per_cpu(init_tss, cpu);
 675
 676         /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
 677
 678         __unlazy_fpu(prev_p);
 679
 680         /*
 681          * Reload esp0.
 682          */
 683         load_esp0(tss, next);
 684
 685         /*
 686          * Save away %fs and %gs. No need to save %es and %ds, as
 687          * those are always kernel segments while inside the kernel.
 688          * Doing this before setting the new TLS descriptors avoids
 689          * the situation where we temporarily have non-reloadable
 690          * segments in %fs and %gs.  This could be an issue if the
 691          * NMI handler ever used %fs or %gs (it does not today), or
 692          * if the kernel is running inside of a hypervisor layer.
 693          */
 694         savesegment(fs, prev->fs);
 695         savesegment(gs, prev->gs);
 696
 697         /*
 698          * Load the per-thread Thread-Local Storage descriptor.
 699          */
 700         load_TLS(next, cpu);
 701
 702         /*
 703          * Restore %fs and %gs if needed.
 704          *
 705          * Glibc normally makes %fs be zero, and %gs is one of
 706          * the TLS segments.
 707          */
 708         if (unlikely(prev->fs | next->fs))
 709                 loadsegment(fs, next->fs);
 710
 711         if (prev->gs | next->gs)
 712                 loadsegment(gs, next->gs);
 713
 714         /*
 715          * Now maybe reload the debug registers
 716          */
 717         if (unlikely(next->debugreg[7])) {
 718                 set_debugreg(next->debugreg[0], 0);
 719                 set_debugreg(next->debugreg[1], 1);
 720                 set_debugreg(next->debugreg[2], 2);
 721                 set_debugreg(next->debugreg[3], 3);
 722                 /* no 4 and 5 */
 723                 set_debugreg(next->debugreg[6], 6);
 724                 set_debugreg(next->debugreg[7], 7);
 725         }
 726
 727         if (unlikely(prev->io_bitmap_ptr || next->io_bitmap_ptr))
 728                 handle_io_bitmap(next, tss);
 729
 730         disable_tsc(prev_p, next_p);
 731
 732         return prev_p;
 733 }
 734
 735 asmlinkage int sys_fork(struct pt_regs regs)
 736 {
 737         return do_fork(SIGCHLD, regs.esp, &regs, 0, NULL, NULL);
 738 }
 739
 740 asmlinkage int sys_clone(struct pt_regs regs)
 741 {
 742         unsigned long clone_flags;
 743         unsigned long newsp;
 744         int __user *parent_tidptr, *child_tidptr;
 745
 746         clone_flags = regs.ebx;
 747         newsp = regs.ecx;
 748         parent_tidptr = (int __user *)regs.edx;
 749         child_tidptr = (int __user *)regs.edi;
 750         if (!newsp)
 751                 newsp = regs.esp;
 752         return do_fork(clone_flags, newsp, &regs, 0, parent_tidptr, child_tidptr);
 753 }
 754
 755 /*
 756  * This is trivial, and on the face of it looks like it
 757  * could equally well be done in user mode.
 758  *
 759  * Not so, for quite unobvious reasons - register pressure.
 760  * In user mode vfork() cannot have a stack frame, and if
 761  * done by calling the "clone()" system call directly, you
 762  * do not have enough call-clobbered registers to hold all
 763  * the information you need.
 764  */
 765 asmlinkage int sys_vfork(struct pt_regs regs)
 766 {
 767         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs.esp, &regs, 0, NULL, NULL);
 768 }
 769
 770 /*
 771  * sys_execve() executes a new program.
 772  */
 773 asmlinkage int sys_execve(struct pt_regs regs)
 774 {
 775         int error;
 776         char * filename;
 777
 778         filename = getname((char __user *) regs.ebx);
 779         error = PTR_ERR(filename);
 780         if (IS_ERR(filename))
 781                 goto out;
 782         error = do_execve(filename,
 783                         (char __user * __user *) regs.ecx,
 784                         (char __user * __user *) regs.edx,
 785                         &regs);
 786         if (error == 0) {
 787                 task_lock(current);
 788                 current->ptrace &= ~PT_DTRACE;
 789                 task_unlock(current);
 790                 /* Make sure we don't return using sysenter.. */
 791                 set_thread_flag(TIF_IRET);
 792         }
 793         putname(filename);
 794 out:
 795         return error;
 796 }
 797
 798 #define top_esp                (THREAD_SIZE - sizeof(unsigned long))
 799 #define top_ebp                (THREAD_SIZE - 2*sizeof(unsigned long))
 800
 801 unsigned long get_wchan(struct task_struct *p)
 802 {
 803         unsigned long ebp, esp, eip;
 804         unsigned long stack_page;
 805         int count = 0;
 806         if (!p || p == current || p->state == TASK_RUNNING)
 807                 return 0;
 808         stack_page = (unsigned long)p->thread_info;
 809         esp = p->thread.esp;
 810         if (!stack_page || esp < stack_page || esp > top_esp+stack_page)
 811                 return 0;
 812         /* include/asm-i386/system.h:switch_to() pushes ebp last. */
 813         ebp = *(unsigned long *) esp;
 814         do {
 815                 if (ebp < stack_page || ebp > top_ebp+stack_page)
 816                         return 0;
 817                 eip = *(unsigned long *) (ebp+4);
 818                 if (!in_sched_functions(eip))
 819                         return eip;
 820                 ebp = *(unsigned long *) ebp;
 821         } while (count++ < 16);
 822         return 0;
 823 }
 824 EXPORT_SYMBOL(get_wchan);
 825
 826 /*
 827  * sys_alloc_thread_area: get a yet unused TLS descriptor index.
 828  */
 829 static int get_free_idx(void)
 830 {
 831         struct thread_struct *t = &current->thread;
 832         int idx;
 833
 834         for (idx = 0; idx < GDT_ENTRY_TLS_ENTRIES; idx++)
 835                 if (desc_empty(t->tls_array + idx))
 836                         return idx + GDT_ENTRY_TLS_MIN;
 837         return -ESRCH;
 838 }
 839
 840 /*
 841  * Set a given TLS descriptor:
 842  */
 843 asmlinkage int sys_set_thread_area(struct user_desc __user *u_info)
 844 {
 845         struct thread_struct *t = &current->thread;
 846         struct user_desc info;
 847         struct desc_struct *desc;
 848         int cpu, idx;
 849
 850         if (copy_from_user(&info, u_info, sizeof(info)))
 851                 return -EFAULT;
 852         idx = info.entry_number;
 853
 854         /*
 855          * index -1 means the kernel should try to find and
 856          * allocate an empty descriptor:
 857          */
 858         if (idx == -1) {
 859                 idx = get_free_idx();
 860                 if (idx < 0)
 861                         return idx;
 862                 if (put_user(idx, &u_info->entry_number))
 863                         return -EFAULT;
 864         }
 865
 866         if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
 867                 return -EINVAL;
 868
 869         desc = t->tls_array + idx - GDT_ENTRY_TLS_MIN;
 870
 871         /*
 872          * We must not get preempted while modifying the TLS.
 873          */
 874         cpu = get_cpu();
 875
 876         if (LDT_empty(&info)) {
 877                 desc->a = 0;
 878                 desc->b = 0;
 879         } else {
 880                 desc->a = LDT_entry_a(&info);
 881                 desc->b = LDT_entry_b(&info);
 882         }
 883         load_TLS(t, cpu);
 884
 885         put_cpu();
 886
 887         return 0;
 888 }
 889
 890 /*
 891  * Get the current Thread-Local Storage area:
 892  */
 893
 894 #define GET_BASE(desc) ( \
 895         (((desc)->a >> 16) & 0x0000ffff) | \
 896         (((desc)->b << 16) & 0x00ff0000) | \
 897         ( (desc)->b        & 0xff000000)   )
 898
 899 #define GET_LIMIT(desc) ( \
 900         ((desc)->a & 0x0ffff) | \
 901          ((desc)->b & 0xf0000) )
 902
 903 #define GET_32BIT(desc)         (((desc)->b >> 22) & 1)
 904 #define GET_CONTENTS(desc)      (((desc)->b >> 10) & 3)
 905 #define GET_WRITABLE(desc)      (((desc)->b >>  9) & 1)
 906 #define GET_LIMIT_PAGES(desc)   (((desc)->b >> 23) & 1)
 907 #define GET_PRESENT(desc)       (((desc)->b >> 15) & 1)
 908 #define GET_USEABLE(desc)       (((desc)->b >> 20) & 1)
 909
 910 asmlinkage int sys_get_thread_area(struct user_desc __user *u_info)
 911 {
 912         struct user_desc info;
 913         struct desc_struct *desc;
 914         int idx;
 915
 916         if (get_user(idx, &u_info->entry_number))
 917                 return -EFAULT;
 918         if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
 919                 return -EINVAL;
 920
 921         memset(&info, 0, sizeof(info));
 922
 923         desc = current->thread.tls_array + idx - GDT_ENTRY_TLS_MIN;
 924
 925         info.entry_number = idx;
 926         info.base_addr = GET_BASE(desc);
 927         info.limit = GET_LIMIT(desc);
 928         info.seg_32bit = GET_32BIT(desc);
 929         info.contents = GET_CONTENTS(desc);
 930         info.read_exec_only = !GET_WRITABLE(desc);
 931         info.limit_in_pages = GET_LIMIT_PAGES(desc);
 932         info.seg_not_present = !GET_PRESENT(desc);
 933         info.useable = GET_USEABLE(desc);
 934
 935         if (copy_to_user(u_info, &info, sizeof(info)))
 936                 return -EFAULT;
 937         return 0;
 938 }
 939
 940 unsigned long arch_align_stack(unsigned long sp)
 941 {
 942         if (randomize_va_space)
 943                 sp -= get_random_int() % 8192;
 944         return sp & ~0xf;
 945 }