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         local_irq_enable();
 103
 104         if (!hlt_counter && boot_cpu_data.hlt_works_ok) {
 105                 clear_thread_flag(TIF_POLLING_NRFLAG);
 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                 set_thread_flag(TIF_POLLING_NRFLAG);
 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         set_thread_flag(TIF_POLLING_NRFLAG);
 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 static void mwait_idle(void)
 241 {
 242         local_irq_enable();
 243
 244         while (!need_resched()) {
 245                 __monitor((void *)&current_thread_info()->flags, 0, 0);
 246                 smp_mb();
 247                 if (need_resched())
 248                         break;
 249                 __mwait(0, 0);
 250         }
 251 }
 252
 253 void __devinit select_idle_routine(const struct cpuinfo_x86 *c)
 254 {
 255         if (cpu_has(c, X86_FEATURE_MWAIT)) {
 256                 printk("monitor/mwait feature present.\n");
 257                 /*
 258                  * Skip, if setup has overridden idle.
 259                  * One CPU supports mwait => All CPUs supports mwait
 260                  */
 261                 if (!pm_idle) {
 262                         printk("using mwait in idle threads.\n");
 263                         pm_idle = mwait_idle;
 264                 }
 265         }
 266 }
 267
 268 static int __init idle_setup (char *str)
 269 {
 270         if (!strncmp(str, "poll", 4)) {
 271                 printk("using polling idle threads.\n");
 272                 pm_idle = poll_idle;
 273 #ifdef CONFIG_X86_SMP
 274                 if (smp_num_siblings > 1)
 275                         printk("WARNING: polling idle and HT enabled, performance may degrade.\n");
 276 #endif
 277         } else if (!strncmp(str, "halt", 4)) {
 278                 printk("using halt in idle threads.\n");
 279                 pm_idle = default_idle;
 280         }
 281
 282         boot_option_idle_override = 1;
 283         return 1;
 284 }
 285
 286 __setup("idle=", idle_setup);
 287
 288 void show_regs(struct pt_regs * regs)
 289 {
 290         unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L;
 291
 292         printk("\n");
 293         printk("Pid: %d, comm: %20s\n", current->pid, current->comm);
 294         printk("EIP: %04x:[<%08lx>] CPU: %d\n",0xffff & regs->xcs,regs->eip, smp_processor_id());
 295         print_symbol("EIP is at %s\n", regs->eip);
 296
 297         if (user_mode(regs))
 298                 printk(" ESP: %04x:%08lx",0xffff & regs->xss,regs->esp);
 299         printk(" EFLAGS: %08lx    %s  (%s)\n",
 300                regs->eflags, print_tainted(), system_utsname.release);
 301         printk("EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
 302                 regs->eax,regs->ebx,regs->ecx,regs->edx);
 303         printk("ESI: %08lx EDI: %08lx EBP: %08lx",
 304                 regs->esi, regs->edi, regs->ebp);
 305         printk(" DS: %04x ES: %04x\n",
 306                 0xffff & regs->xds,0xffff & regs->xes);
 307
 308         cr0 = read_cr0();
 309         cr2 = read_cr2();
 310         cr3 = read_cr3();
 311         cr4 = read_cr4_safe();
 312         printk("CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n", cr0, cr2, cr3, cr4);
 313         show_trace(NULL, &regs->esp);
 314 }
 315
 316 /*
 317  * This gets run with %ebx containing the
 318  * function to call, and %edx containing
 319  * the "args".
 320  */
 321 extern void kernel_thread_helper(void);
 322 __asm__(".section .text\n"
 323         ".align 4\n"
 324         "kernel_thread_helper:\n\t"
 325         "movl %edx,%eax\n\t"
 326         "pushl %edx\n\t"
 327         "call *%ebx\n\t"
 328         "pushl %eax\n\t"
 329         "call do_exit\n"
 330         ".previous");
 331
 332 /*
 333  * Create a kernel thread
 334  */
 335 int kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
 336 {
 337         struct pt_regs regs;
 338
 339         memset(&regs, 0, sizeof(regs));
 340
 341         regs.ebx = (unsigned long) fn;
 342         regs.edx = (unsigned long) arg;
 343
 344         regs.xds = __USER_DS;
 345         regs.xes = __USER_DS;
 346         regs.orig_eax = -1;
 347         regs.eip = (unsigned long) kernel_thread_helper;
 348         regs.xcs = __KERNEL_CS;
 349         regs.eflags = X86_EFLAGS_IF | X86_EFLAGS_SF | X86_EFLAGS_PF | 0x2;
 350
 351         /* Ok, create the new process.. */
 352         return do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0, &regs, 0, NULL, NULL);
 353 }
 354 EXPORT_SYMBOL(kernel_thread);
 355
 356 /*
 357  * Free current thread data structures etc..
 358  */
 359 void exit_thread(void)
 360 {
 361         struct task_struct *tsk = current;
 362         struct thread_struct *t = &tsk->thread;
 363
 364         /*
 365          * Remove function-return probe instances associated with this task
 366          * and put them back on the free list. Do not insert an exit probe for
 367          * this function, it will be disabled by kprobe_flush_task if you do.
 368          */
 369         kprobe_flush_task(tsk);
 370
 371         /* The process may have allocated an io port bitmap... nuke it. */
 372         if (unlikely(NULL != t->io_bitmap_ptr)) {
 373                 int cpu = get_cpu();
 374                 struct tss_struct *tss = &per_cpu(init_tss, cpu);
 375
 376                 kfree(t->io_bitmap_ptr);
 377                 t->io_bitmap_ptr = NULL;
 378                 /*
 379                  * Careful, clear this in the TSS too:
 380                  */
 381                 memset(tss->io_bitmap, 0xff, tss->io_bitmap_max);
 382                 t->io_bitmap_max = 0;
 383                 tss->io_bitmap_owner = NULL;
 384                 tss->io_bitmap_max = 0;
 385                 tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
 386                 put_cpu();
 387         }
 388 }
 389
 390 void flush_thread(void)
 391 {
 392         struct task_struct *tsk = current;
 393
 394         memset(tsk->thread.debugreg, 0, sizeof(unsigned long)*8);
 395         memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
 396         /*
 397          * Forget coprocessor state..
 398          */
 399         clear_fpu(tsk);
 400         clear_used_math();
 401 }
 402
 403 void release_thread(struct task_struct *dead_task)
 404 {
 405         BUG_ON(dead_task->mm);
 406         release_vm86_irqs(dead_task);
 407 }
 408
 409 /*
 410  * This gets called before we allocate a new thread and copy
 411  * the current task into it.
 412  */
 413 void prepare_to_copy(struct task_struct *tsk)
 414 {
 415         unlazy_fpu(tsk);
 416 }
 417
 418 int copy_thread(int nr, unsigned long clone_flags, unsigned long esp,
 419         unsigned long unused,
 420         struct task_struct * p, struct pt_regs * regs)
 421 {
 422         struct pt_regs * childregs;
 423         struct task_struct *tsk;
 424         int err;
 425
 426         childregs = ((struct pt_regs *) (THREAD_SIZE + (unsigned long) p->thread_info)) - 1;
 427         /*
 428          * The below -8 is to reserve 8 bytes on top of the ring0 stack.
 429          * This is necessary to guarantee that the entire "struct pt_regs"
 430          * is accessable even if the CPU haven't stored the SS/ESP registers
 431          * on the stack (interrupt gate does not save these registers
 432          * when switching to the same priv ring).
 433          * Therefore beware: accessing the xss/esp fields of the
 434          * "struct pt_regs" is possible, but they may contain the
 435          * completely wrong values.
 436          */
 437         childregs = (struct pt_regs *) ((unsigned long) childregs - 8);
 438         *childregs = *regs;
 439         childregs->eax = 0;
 440         childregs->esp = esp;
 441
 442         p->thread.esp = (unsigned long) childregs;
 443         p->thread.esp0 = (unsigned long) (childregs+1);
 444
 445         p->thread.eip = (unsigned long) ret_from_fork;
 446
 447         savesegment(fs,p->thread.fs);
 448         savesegment(gs,p->thread.gs);
 449
 450         tsk = current;
 451         if (unlikely(NULL != tsk->thread.io_bitmap_ptr)) {
 452                 p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL);
 453                 if (!p->thread.io_bitmap_ptr) {
 454                         p->thread.io_bitmap_max = 0;
 455                         return -ENOMEM;
 456                 }
 457                 memcpy(p->thread.io_bitmap_ptr, tsk->thread.io_bitmap_ptr,
 458                         IO_BITMAP_BYTES);
 459         }
 460
 461         /*
 462          * Set a new TLS for the child thread?
 463          */
 464         if (clone_flags & CLONE_SETTLS) {
 465                 struct desc_struct *desc;
 466                 struct user_desc info;
 467                 int idx;
 468
 469                 err = -EFAULT;
 470                 if (copy_from_user(&info, (void __user *)childregs->esi, sizeof(info)))
 471                         goto out;
 472                 err = -EINVAL;
 473                 if (LDT_empty(&info))
 474                         goto out;
 475
 476                 idx = info.entry_number;
 477                 if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
 478                         goto out;
 479
 480                 desc = p->thread.tls_array + idx - GDT_ENTRY_TLS_MIN;
 481                 desc->a = LDT_entry_a(&info);
 482                 desc->b = LDT_entry_b(&info);
 483         }
 484
 485         err = 0;
 486  out:
 487         if (err && p->thread.io_bitmap_ptr) {
 488                 kfree(p->thread.io_bitmap_ptr);
 489                 p->thread.io_bitmap_max = 0;
 490         }
 491         return err;
 492 }
 493
 494 /*
 495  * fill in the user structure for a core dump..
 496  */
 497 void dump_thread(struct pt_regs * regs, struct user * dump)
 498 {
 499         int i;
 500
 501 /* changed the size calculations - should hopefully work better. lbt */
 502         dump->magic = CMAGIC;
 503         dump->start_code = 0;
 504         dump->start_stack = regs->esp & ~(PAGE_SIZE - 1);
 505         dump->u_tsize = ((unsigned long) current->mm->end_code) >> PAGE_SHIFT;
 506         dump->u_dsize = ((unsigned long) (current->mm->brk + (PAGE_SIZE-1))) >> PAGE_SHIFT;
 507         dump->u_dsize -= dump->u_tsize;
 508         dump->u_ssize = 0;
 509         for (i = 0; i < 8; i++)
 510                 dump->u_debugreg[i] = current->thread.debugreg[i];
 511
 512         if (dump->start_stack < TASK_SIZE)
 513                 dump->u_ssize = ((unsigned long) (TASK_SIZE - dump->start_stack)) >> PAGE_SHIFT;
 514
 515         dump->regs.ebx = regs->ebx;
 516         dump->regs.ecx = regs->ecx;
 517         dump->regs.edx = regs->edx;
 518         dump->regs.esi = regs->esi;
 519         dump->regs.edi = regs->edi;
 520         dump->regs.ebp = regs->ebp;
 521         dump->regs.eax = regs->eax;
 522         dump->regs.ds = regs->xds;
 523         dump->regs.es = regs->xes;
 524         savesegment(fs,dump->regs.fs);
 525         savesegment(gs,dump->regs.gs);
 526         dump->regs.orig_eax = regs->orig_eax;
 527         dump->regs.eip = regs->eip;
 528         dump->regs.cs = regs->xcs;
 529         dump->regs.eflags = regs->eflags;
 530         dump->regs.esp = regs->esp;
 531         dump->regs.ss = regs->xss;
 532
 533         dump->u_fpvalid = dump_fpu (regs, &dump->i387);
 534 }
 535 EXPORT_SYMBOL(dump_thread);
 536
 537 /*
 538  * Capture the user space registers if the task is not running (in user space)
 539  */
 540 int dump_task_regs(struct task_struct *tsk, elf_gregset_t *regs)
 541 {
 542         struct pt_regs ptregs;
 543
 544         ptregs = *(struct pt_regs *)
 545                 ((unsigned long)tsk->thread_info +
 546                 /* see comments in copy_thread() about -8 */
 547                 THREAD_SIZE - sizeof(ptregs) - 8);
 548         ptregs.xcs &= 0xffff;
 549         ptregs.xds &= 0xffff;
 550         ptregs.xes &= 0xffff;
 551         ptregs.xss &= 0xffff;
 552
 553         elf_core_copy_regs(regs, &ptregs);
 554
 555         return 1;
 556 }
 557
 558 static inline void
 559 handle_io_bitmap(struct thread_struct *next, struct tss_struct *tss)
 560 {
 561         if (!next->io_bitmap_ptr) {
 562                 /*
 563                  * Disable the bitmap via an invalid offset. We still cache
 564                  * the previous bitmap owner and the IO bitmap contents:
 565                  */
 566                 tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
 567                 return;
 568         }
 569         if (likely(next == tss->io_bitmap_owner)) {
 570                 /*
 571                  * Previous owner of the bitmap (hence the bitmap content)
 572                  * matches the next task, we dont have to do anything but
 573                  * to set a valid offset in the TSS:
 574                  */
 575                 tss->io_bitmap_base = IO_BITMAP_OFFSET;
 576                 return;
 577         }
 578         /*
 579          * Lazy TSS's I/O bitmap copy. We set an invalid offset here
 580          * and we let the task to get a GPF in case an I/O instruction
 581          * is performed.  The handler of the GPF will verify that the
 582          * faulting task has a valid I/O bitmap and, it true, does the
 583          * real copy and restart the instruction.  This will save us
 584          * redundant copies when the currently switched task does not
 585          * perform any I/O during its timeslice.
 586          */
 587         tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET_LAZY;
 588 }
 589
 590 /*
 591  * This function selects if the context switch from prev to next
 592  * has to tweak the TSC disable bit in the cr4.
 593  */
 594 static inline void disable_tsc(struct task_struct *prev_p,
 595                                struct task_struct *next_p)
 596 {
 597         struct thread_info *prev, *next;
 598
 599         /*
 600          * gcc should eliminate the ->thread_info dereference if
 601          * has_secure_computing returns 0 at compile time (SECCOMP=n).
 602          */
 603         prev = prev_p->thread_info;
 604         next = next_p->thread_info;
 605
 606         if (has_secure_computing(prev) || has_secure_computing(next)) {
 607                 /* slow path here */
 608                 if (has_secure_computing(prev) &&
 609                     !has_secure_computing(next)) {
 610                         write_cr4(read_cr4() & ~X86_CR4_TSD);
 611                 } else if (!has_secure_computing(prev) &&
 612                            has_secure_computing(next))
 613                         write_cr4(read_cr4() | X86_CR4_TSD);
 614         }
 615 }
 616
 617 /*
 618  *      switch_to(x,yn) should switch tasks from x to y.
 619  *
 620  * We fsave/fwait so that an exception goes off at the right time
 621  * (as a call from the fsave or fwait in effect) rather than to
 622  * the wrong process. Lazy FP saving no longer makes any sense
 623  * with modern CPU's, and this simplifies a lot of things (SMP
 624  * and UP become the same).
 625  *
 626  * NOTE! We used to use the x86 hardware context switching. The
 627  * reason for not using it any more becomes apparent when you
 628  * try to recover gracefully from saved state that is no longer
 629  * valid (stale segment register values in particular). With the
 630  * hardware task-switch, there is no way to fix up bad state in
 631  * a reasonable manner.
 632  *
 633  * The fact that Intel documents the hardware task-switching to
 634  * be slow is a fairly red herring - this code is not noticeably
 635  * faster. However, there _is_ some room for improvement here,
 636  * so the performance issues may eventually be a valid point.
 637  * More important, however, is the fact that this allows us much
 638  * more flexibility.
 639  *
 640  * The return value (in %eax) will be the "prev" task after
 641  * the task-switch, and shows up in ret_from_fork in entry.S,
 642  * for example.
 643  */
 644 struct task_struct fastcall * __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
 645 {
 646         struct thread_struct *prev = &prev_p->thread,
 647                                  *next = &next_p->thread;
 648         int cpu = smp_processor_id();
 649         struct tss_struct *tss = &per_cpu(init_tss, cpu);
 650
 651         /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
 652
 653         __unlazy_fpu(prev_p);
 654
 655         /*
 656          * Reload esp0.
 657          */
 658         load_esp0(tss, next);
 659
 660         /*
 661          * Save away %fs and %gs. No need to save %es and %ds, as
 662          * those are always kernel segments while inside the kernel.
 663          * Doing this before setting the new TLS descriptors avoids
 664          * the situation where we temporarily have non-reloadable
 665          * segments in %fs and %gs.  This could be an issue if the
 666          * NMI handler ever used %fs or %gs (it does not today), or
 667          * if the kernel is running inside of a hypervisor layer.
 668          */
 669         savesegment(fs, prev->fs);
 670         savesegment(gs, prev->gs);
 671
 672         /*
 673          * Load the per-thread Thread-Local Storage descriptor.
 674          */
 675         load_TLS(next, cpu);
 676
 677         /*
 678          * Restore %fs and %gs if needed.
 679          *
 680          * Glibc normally makes %fs be zero, and %gs is one of
 681          * the TLS segments.
 682          */
 683         if (unlikely(prev->fs | next->fs))
 684                 loadsegment(fs, next->fs);
 685
 686         if (prev->gs | next->gs)
 687                 loadsegment(gs, next->gs);
 688
 689         /*
 690          * Restore IOPL if needed.
 691          */
 692         if (unlikely(prev->iopl != next->iopl))
 693                 set_iopl_mask(next->iopl);
 694
 695         /*
 696          * Now maybe reload the debug registers
 697          */
 698         if (unlikely(next->debugreg[7])) {
 699                 set_debugreg(next->debugreg[0], 0);
 700                 set_debugreg(next->debugreg[1], 1);
 701                 set_debugreg(next->debugreg[2], 2);
 702                 set_debugreg(next->debugreg[3], 3);
 703                 /* no 4 and 5 */
 704                 set_debugreg(next->debugreg[6], 6);
 705                 set_debugreg(next->debugreg[7], 7);
 706         }
 707
 708         if (unlikely(prev->io_bitmap_ptr || next->io_bitmap_ptr))
 709                 handle_io_bitmap(next, tss);
 710
 711         disable_tsc(prev_p, next_p);
 712
 713         return prev_p;
 714 }
 715
 716 asmlinkage int sys_fork(struct pt_regs regs)
 717 {
 718         return do_fork(SIGCHLD, regs.esp, &regs, 0, NULL, NULL);
 719 }
 720
 721 asmlinkage int sys_clone(struct pt_regs regs)
 722 {
 723         unsigned long clone_flags;
 724         unsigned long newsp;
 725         int __user *parent_tidptr, *child_tidptr;
 726
 727         clone_flags = regs.ebx;
 728         newsp = regs.ecx;
 729         parent_tidptr = (int __user *)regs.edx;
 730         child_tidptr = (int __user *)regs.edi;
 731         if (!newsp)
 732                 newsp = regs.esp;
 733         return do_fork(clone_flags, newsp, &regs, 0, parent_tidptr, child_tidptr);
 734 }
 735
 736 /*
 737  * This is trivial, and on the face of it looks like it
 738  * could equally well be done in user mode.
 739  *
 740  * Not so, for quite unobvious reasons - register pressure.
 741  * In user mode vfork() cannot have a stack frame, and if
 742  * done by calling the "clone()" system call directly, you
 743  * do not have enough call-clobbered registers to hold all
 744  * the information you need.
 745  */
 746 asmlinkage int sys_vfork(struct pt_regs regs)
 747 {
 748         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs.esp, &regs, 0, NULL, NULL);
 749 }
 750
 751 /*
 752  * sys_execve() executes a new program.
 753  */
 754 asmlinkage int sys_execve(struct pt_regs regs)
 755 {
 756         int error;
 757         char * filename;
 758
 759         filename = getname((char __user *) regs.ebx);
 760         error = PTR_ERR(filename);
 761         if (IS_ERR(filename))
 762                 goto out;
 763         error = do_execve(filename,
 764                         (char __user * __user *) regs.ecx,
 765                         (char __user * __user *) regs.edx,
 766                         &regs);
 767         if (error == 0) {
 768                 task_lock(current);
 769                 current->ptrace &= ~PT_DTRACE;
 770                 task_unlock(current);
 771                 /* Make sure we don't return using sysenter.. */
 772                 set_thread_flag(TIF_IRET);
 773         }
 774         putname(filename);
 775 out:
 776         return error;
 777 }
 778
 779 #define top_esp                (THREAD_SIZE - sizeof(unsigned long))
 780 #define top_ebp                (THREAD_SIZE - 2*sizeof(unsigned long))
 781
 782 unsigned long get_wchan(struct task_struct *p)
 783 {
 784         unsigned long ebp, esp, eip;
 785         unsigned long stack_page;
 786         int count = 0;
 787         if (!p || p == current || p->state == TASK_RUNNING)
 788                 return 0;
 789         stack_page = (unsigned long)p->thread_info;
 790         esp = p->thread.esp;
 791         if (!stack_page || esp < stack_page || esp > top_esp+stack_page)
 792                 return 0;
 793         /* include/asm-i386/system.h:switch_to() pushes ebp last. */
 794         ebp = *(unsigned long *) esp;
 795         do {
 796                 if (ebp < stack_page || ebp > top_ebp+stack_page)
 797                         return 0;
 798                 eip = *(unsigned long *) (ebp+4);
 799                 if (!in_sched_functions(eip))
 800                         return eip;
 801                 ebp = *(unsigned long *) ebp;
 802         } while (count++ < 16);
 803         return 0;
 804 }
 805 EXPORT_SYMBOL(get_wchan);
 806
 807 /*
 808  * sys_alloc_thread_area: get a yet unused TLS descriptor index.
 809  */
 810 static int get_free_idx(void)
 811 {
 812         struct thread_struct *t = &current->thread;
 813         int idx;
 814
 815         for (idx = 0; idx < GDT_ENTRY_TLS_ENTRIES; idx++)
 816                 if (desc_empty(t->tls_array + idx))
 817                         return idx + GDT_ENTRY_TLS_MIN;
 818         return -ESRCH;
 819 }
 820
 821 /*
 822  * Set a given TLS descriptor:
 823  */
 824 asmlinkage int sys_set_thread_area(struct user_desc __user *u_info)
 825 {
 826         struct thread_struct *t = &current->thread;
 827         struct user_desc info;
 828         struct desc_struct *desc;
 829         int cpu, idx;
 830
 831         if (copy_from_user(&info, u_info, sizeof(info)))
 832                 return -EFAULT;
 833         idx = info.entry_number;
 834
 835         /*
 836          * index -1 means the kernel should try to find and
 837          * allocate an empty descriptor:
 838          */
 839         if (idx == -1) {
 840                 idx = get_free_idx();
 841                 if (idx < 0)
 842                         return idx;
 843                 if (put_user(idx, &u_info->entry_number))
 844                         return -EFAULT;
 845         }
 846
 847         if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
 848                 return -EINVAL;
 849
 850         desc = t->tls_array + idx - GDT_ENTRY_TLS_MIN;
 851
 852         /*
 853          * We must not get preempted while modifying the TLS.
 854          */
 855         cpu = get_cpu();
 856
 857         if (LDT_empty(&info)) {
 858                 desc->a = 0;
 859                 desc->b = 0;
 860         } else {
 861                 desc->a = LDT_entry_a(&info);
 862                 desc->b = LDT_entry_b(&info);
 863         }
 864         load_TLS(t, cpu);
 865
 866         put_cpu();
 867
 868         return 0;
 869 }
 870
 871 /*
 872  * Get the current Thread-Local Storage area:
 873  */
 874
 875 #define GET_BASE(desc) ( \
 876         (((desc)->a >> 16) & 0x0000ffff) | \
 877         (((desc)->b << 16) & 0x00ff0000) | \
 878         ( (desc)->b        & 0xff000000)   )
 879
 880 #define GET_LIMIT(desc) ( \
 881         ((desc)->a & 0x0ffff) | \
 882          ((desc)->b & 0xf0000) )
 883
 884 #define GET_32BIT(desc)         (((desc)->b >> 22) & 1)
 885 #define GET_CONTENTS(desc)      (((desc)->b >> 10) & 3)
 886 #define GET_WRITABLE(desc)      (((desc)->b >>  9) & 1)
 887 #define GET_LIMIT_PAGES(desc)   (((desc)->b >> 23) & 1)
 888 #define GET_PRESENT(desc)       (((desc)->b >> 15) & 1)
 889 #define GET_USEABLE(desc)       (((desc)->b >> 20) & 1)
 890
 891 asmlinkage int sys_get_thread_area(struct user_desc __user *u_info)
 892 {
 893         struct user_desc info;
 894         struct desc_struct *desc;
 895         int idx;
 896
 897         if (get_user(idx, &u_info->entry_number))
 898                 return -EFAULT;
 899         if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
 900                 return -EINVAL;
 901
 902         memset(&info, 0, sizeof(info));
 903
 904         desc = current->thread.tls_array + idx - GDT_ENTRY_TLS_MIN;
 905
 906         info.entry_number = idx;
 907         info.base_addr = GET_BASE(desc);
 908         info.limit = GET_LIMIT(desc);
 909         info.seg_32bit = GET_32BIT(desc);
 910         info.contents = GET_CONTENTS(desc);
 911         info.read_exec_only = !GET_WRITABLE(desc);
 912         info.limit_in_pages = GET_LIMIT_PAGES(desc);
 913         info.seg_not_present = !GET_PRESENT(desc);
 914         info.useable = GET_USEABLE(desc);
 915
 916         if (copy_to_user(u_info, &info, sizeof(info)))
 917                 return -EFAULT;
 918         return 0;
 919 }
 920
 921 unsigned long arch_align_stack(unsigned long sp)
 922 {
 923         if (randomize_va_space)
 924                 sp -= get_random_int() % 8192;
 925         return sp & ~0xf;
 926 }