arch/x86/kernel/process_32.c

   1 /*
   2  *  Copyright (C) 1995  Linus Torvalds
   3  *
   4  *  Pentium III FXSR, SSE support
   5  *      Gareth Hughes <gareth@valinux.com>, May 2000
   6  */
   7
   8 /*
   9  * This file handles the architecture-dependent parts of process handling..
  10  */
  11
  12 #include <stdarg.h>
  13
  14 #include <linux/cpu.h>
  15 #include <linux/errno.h>
  16 #include <linux/sched.h>
  17 #include <linux/fs.h>
  18 #include <linux/kernel.h>
  19 #include <linux/mm.h>
  20 #include <linux/elfcore.h>
  21 #include <linux/smp.h>
  22 #include <linux/stddef.h>
  23 #include <linux/slab.h>
  24 #include <linux/vmalloc.h>
  25 #include <linux/user.h>
  26 #include <linux/a.out.h>
  27 #include <linux/interrupt.h>
  28 #include <linux/utsname.h>
  29 #include <linux/delay.h>
  30 #include <linux/reboot.h>
  31 #include <linux/init.h>
  32 #include <linux/mc146818rtc.h>
  33 #include <linux/module.h>
  34 #include <linux/kallsyms.h>
  35 #include <linux/ptrace.h>
  36 #include <linux/random.h>
  37 #include <linux/personality.h>
  38 #include <linux/tick.h>
  39 #include <linux/percpu.h>
  40
  41 #include <asm/uaccess.h>
  42 #include <asm/pgtable.h>
  43 #include <asm/system.h>
  44 #include <asm/io.h>
  45 #include <asm/ldt.h>
  46 #include <asm/processor.h>
  47 #include <asm/i387.h>
  48 #include <asm/desc.h>
  49 #include <asm/vm86.h>
  50 #ifdef CONFIG_MATH_EMULATION
  51 #include <asm/math_emu.h>
  52 #endif
  53
  54 #include <linux/err.h>
  55
  56 #include <asm/tlbflush.h>
  57 #include <asm/cpu.h>
  58
  59 asmlinkage void ret_from_fork(void) __asm__("ret_from_fork");
  60
  61 static int hlt_counter;
  62
  63 unsigned long boot_option_idle_override = 0;
  64 EXPORT_SYMBOL(boot_option_idle_override);
  65
  66 DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task;
  67 EXPORT_PER_CPU_SYMBOL(current_task);
  68
  69 DEFINE_PER_CPU(int, cpu_number);
  70 EXPORT_PER_CPU_SYMBOL(cpu_number);
  71
  72 /*
  73  * Return saved PC of a blocked thread.
  74  */
  75 unsigned long thread_saved_pc(struct task_struct *tsk)
  76 {
  77         return ((unsigned long *)tsk->thread.esp)[3];
  78 }
  79
  80 /*
  81  * Powermanagement idle function, if any..
  82  */
  83 void (*pm_idle)(void);
  84 EXPORT_SYMBOL(pm_idle);
  85 static DEFINE_PER_CPU(unsigned int, cpu_idle_state);
  86
  87 void disable_hlt(void)
  88 {
  89         hlt_counter++;
  90 }
  91
  92 EXPORT_SYMBOL(disable_hlt);
  93
  94 void enable_hlt(void)
  95 {
  96         hlt_counter--;
  97 }
  98
  99 EXPORT_SYMBOL(enable_hlt);
 100
 101 /*
 102  * We use this if we don't have any better
 103  * idle routine..
 104  */
 105 void default_idle(void)
 106 {
 107         if (!hlt_counter && boot_cpu_data.hlt_works_ok) {
 108                 current_thread_info()->status &= ~TS_POLLING;
 109                 /*
 110                  * TS_POLLING-cleared state must be visible before we
 111                  * test NEED_RESCHED:
 112                  */
 113                 smp_mb();
 114
 115                 local_irq_disable();
 116                 if (!need_resched())
 117                         safe_halt();    /* enables interrupts racelessly */
 118                 else
 119                         local_irq_enable();
 120                 current_thread_info()->status |= TS_POLLING;
 121         } else {
 122                 /* loop is done by the caller */
 123                 cpu_relax();
 124         }
 125 }
 126 #ifdef CONFIG_APM_MODULE
 127 EXPORT_SYMBOL(default_idle);
 128 #endif
 129
 130 /*
 131  * On SMP it's slightly faster (but much more power-consuming!)
 132  * to poll the ->work.need_resched flag instead of waiting for the
 133  * cross-CPU IPI to arrive. Use this option with caution.
 134  */
 135 static void poll_idle (void)
 136 {
 137         cpu_relax();
 138 }
 139
 140 #ifdef CONFIG_HOTPLUG_CPU
 141 #include <asm/nmi.h>
 142 /* We don't actually take CPU down, just spin without interrupts. */
 143 static inline void play_dead(void)
 144 {
 145         /* This must be done before dead CPU ack */
 146         cpu_exit_clear();
 147         wbinvd();
 148         mb();
 149         /* Ack it */
 150         __get_cpu_var(cpu_state) = CPU_DEAD;
 151
 152         /*
 153          * With physical CPU hotplug, we should halt the cpu
 154          */
 155         local_irq_disable();
 156         while (1)
 157                 halt();
 158 }
 159 #else
 160 static inline void play_dead(void)
 161 {
 162         BUG();
 163 }
 164 #endif /* CONFIG_HOTPLUG_CPU */
 165
 166 /*
 167  * The idle thread. There's no useful work to be
 168  * done, so just try to conserve power and have a
 169  * low exit latency (ie sit in a loop waiting for
 170  * somebody to say that they'd like to reschedule)
 171  */
 172 void cpu_idle(void)
 173 {
 174         int cpu = smp_processor_id();
 175
 176         current_thread_info()->status |= TS_POLLING;
 177
 178         /* endless idle loop with no priority at all */
 179         while (1) {
 180                 tick_nohz_stop_sched_tick();
 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                         check_pgt_cache();
 188                         rmb();
 189                         idle = pm_idle;
 190
 191                         if (!idle)
 192                                 idle = default_idle;
 193
 194                         if (cpu_is_offline(cpu))
 195                                 play_dead();
 196
 197                         __get_cpu_var(irq_stat).idle_timestamp = jiffies;
 198                         idle();
 199                 }
 200                 tick_nohz_restart_sched_tick();
 201                 preempt_enable_no_resched();
 202                 schedule();
 203                 preempt_disable();
 204         }
 205 }
 206
 207 static void do_nothing(void *unused)
 208 {
 209 }
 210
 211 void cpu_idle_wait(void)
 212 {
 213         unsigned int cpu, this_cpu = get_cpu();
 214         cpumask_t map, tmp = current->cpus_allowed;
 215
 216         set_cpus_allowed(current, cpumask_of_cpu(this_cpu));
 217         put_cpu();
 218
 219         cpus_clear(map);
 220         for_each_online_cpu(cpu) {
 221                 per_cpu(cpu_idle_state, cpu) = 1;
 222                 cpu_set(cpu, map);
 223         }
 224
 225         __get_cpu_var(cpu_idle_state) = 0;
 226
 227         wmb();
 228         do {
 229                 ssleep(1);
 230                 for_each_online_cpu(cpu) {
 231                         if (cpu_isset(cpu, map) && !per_cpu(cpu_idle_state, cpu))
 232                                 cpu_clear(cpu, map);
 233                 }
 234                 cpus_and(map, map, cpu_online_map);
 235                 /*
 236                  * We waited 1 sec, if a CPU still did not call idle
 237                  * it may be because it is in idle and not waking up
 238                  * because it has nothing to do.
 239                  * Give all the remaining CPUS a kick.
 240                  */
 241                 smp_call_function_mask(map, do_nothing, 0, 0);
 242         } while (!cpus_empty(map));
 243
 244         set_cpus_allowed(current, tmp);
 245 }
 246 EXPORT_SYMBOL_GPL(cpu_idle_wait);
 247
 248 /*
 249  * This uses new MONITOR/MWAIT instructions on P4 processors with PNI,
 250  * which can obviate IPI to trigger checking of need_resched.
 251  * We execute MONITOR against need_resched and enter optimized wait state
 252  * through MWAIT. Whenever someone changes need_resched, we would be woken
 253  * up from MWAIT (without an IPI).
 254  *
 255  * New with Core Duo processors, MWAIT can take some hints based on CPU
 256  * capability.
 257  */
 258 void mwait_idle_with_hints(unsigned long eax, unsigned long ecx)
 259 {
 260         if (!need_resched()) {
 261                 __monitor((void *)&current_thread_info()->flags, 0, 0);
 262                 smp_mb();
 263                 if (!need_resched())
 264                         __mwait(eax, ecx);
 265         }
 266 }
 267
 268 /* Default MONITOR/MWAIT with no hints, used for default C1 state */
 269 static void mwait_idle(void)
 270 {
 271         local_irq_enable();
 272         mwait_idle_with_hints(0, 0);
 273 }
 274
 275 void __cpuinit select_idle_routine(const struct cpuinfo_x86 *c)
 276 {
 277         if (cpu_has(c, X86_FEATURE_MWAIT)) {
 278                 printk("monitor/mwait feature present.\n");
 279                 /*
 280                  * Skip, if setup has overridden idle.
 281                  * One CPU supports mwait => All CPUs supports mwait
 282                  */
 283                 if (!pm_idle) {
 284                         printk("using mwait in idle threads.\n");
 285                         pm_idle = mwait_idle;
 286                 }
 287         }
 288 }
 289
 290 static int __init idle_setup(char *str)
 291 {
 292         if (!strcmp(str, "poll")) {
 293                 printk("using polling idle threads.\n");
 294                 pm_idle = poll_idle;
 295 #ifdef CONFIG_X86_SMP
 296                 if (smp_num_siblings > 1)
 297                         printk("WARNING: polling idle and HT enabled, performance may degrade.\n");
 298 #endif
 299         } else if (!strcmp(str, "mwait"))
 300                 force_mwait = 1;
 301         else
 302                 return -1;
 303
 304         boot_option_idle_override = 1;
 305         return 0;
 306 }
 307 early_param("idle", idle_setup);
 308
 309 void __show_registers(struct pt_regs *regs, int all)
 310 {
 311         unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L;
 312         unsigned long d0, d1, d2, d3, d6, d7;
 313         unsigned long esp;
 314         unsigned short ss, gs;
 315
 316         if (user_mode_vm(regs)) {
 317                 esp = regs->esp;
 318                 ss = regs->xss & 0xffff;
 319                 savesegment(gs, gs);
 320         } else {
 321                 esp = (unsigned long) (&regs->esp);
 322                 savesegment(ss, ss);
 323                 savesegment(gs, gs);
 324         }
 325
 326         printk("\n");
 327         printk("Pid: %d, comm: %s %s (%s %.*s)\n",
 328                         task_pid_nr(current), current->comm,
 329                         print_tainted(), init_utsname()->release,
 330                         (int)strcspn(init_utsname()->version, " "),
 331                         init_utsname()->version);
 332
 333         printk("EIP: %04x:[<%08lx>] EFLAGS: %08lx CPU: %d\n",
 334                         0xffff & regs->xcs, regs->eip, regs->eflags,
 335                         smp_processor_id());
 336         print_symbol("EIP is at %s\n", regs->eip);
 337
 338         printk("EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
 339                 regs->eax, regs->ebx, regs->ecx, regs->edx);
 340         printk("ESI: %08lx EDI: %08lx EBP: %08lx ESP: %08lx\n",
 341                 regs->esi, regs->edi, regs->ebp, esp);
 342         printk(" DS: %04x ES: %04x FS: %04x GS: %04x SS: %04x\n",
 343                regs->xds & 0xffff, regs->xes & 0xffff,
 344                regs->xfs & 0xffff, gs, ss);
 345
 346         if (!all)
 347                 return;
 348
 349         cr0 = read_cr0();
 350         cr2 = read_cr2();
 351         cr3 = read_cr3();
 352         cr4 = read_cr4_safe();
 353         printk("CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n",
 354                         cr0, cr2, cr3, cr4);
 355
 356         get_debugreg(d0, 0);
 357         get_debugreg(d1, 1);
 358         get_debugreg(d2, 2);
 359         get_debugreg(d3, 3);
 360         printk("DR0: %08lx DR1: %08lx DR2: %08lx DR3: %08lx\n",
 361                         d0, d1, d2, d3);
 362
 363         get_debugreg(d6, 6);
 364         get_debugreg(d7, 7);
 365         printk("DR6: %08lx DR7: %08lx\n",
 366                         d6, d7);
 367 }
 368
 369 void show_regs(struct pt_regs *regs)
 370 {
 371         __show_registers(regs, 1);
 372         show_trace(NULL, regs, &regs->esp);
 373 }
 374
 375 /*
 376  * This gets run with %ebx containing the
 377  * function to call, and %edx containing
 378  * the "args".
 379  */
 380 extern void kernel_thread_helper(void);
 381
 382 /*
 383  * Create a kernel thread
 384  */
 385 int kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
 386 {
 387         struct pt_regs regs;
 388
 389         memset(&regs, 0, sizeof(regs));
 390
 391         regs.ebx = (unsigned long) fn;
 392         regs.edx = (unsigned long) arg;
 393
 394         regs.xds = __USER_DS;
 395         regs.xes = __USER_DS;
 396         regs.xfs = __KERNEL_PERCPU;
 397         regs.orig_eax = -1;
 398         regs.eip = (unsigned long) kernel_thread_helper;
 399         regs.xcs = __KERNEL_CS | get_kernel_rpl();
 400         regs.eflags = X86_EFLAGS_IF | X86_EFLAGS_SF | X86_EFLAGS_PF | 0x2;
 401
 402         /* Ok, create the new process.. */
 403         return do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0, &regs, 0, NULL, NULL);
 404 }
 405 EXPORT_SYMBOL(kernel_thread);
 406
 407 /*
 408  * Free current thread data structures etc..
 409  */
 410 void exit_thread(void)
 411 {
 412         /* The process may have allocated an io port bitmap... nuke it. */
 413         if (unlikely(test_thread_flag(TIF_IO_BITMAP))) {
 414                 struct task_struct *tsk = current;
 415                 struct thread_struct *t = &tsk->thread;
 416                 int cpu = get_cpu();
 417                 struct tss_struct *tss = &per_cpu(init_tss, cpu);
 418
 419                 kfree(t->io_bitmap_ptr);
 420                 t->io_bitmap_ptr = NULL;
 421                 clear_thread_flag(TIF_IO_BITMAP);
 422                 /*
 423                  * Careful, clear this in the TSS too:
 424                  */
 425                 memset(tss->io_bitmap, 0xff, tss->io_bitmap_max);
 426                 t->io_bitmap_max = 0;
 427                 tss->io_bitmap_owner = NULL;
 428                 tss->io_bitmap_max = 0;
 429                 tss->x86_tss.io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
 430                 put_cpu();
 431         }
 432 }
 433
 434 void flush_thread(void)
 435 {
 436         struct task_struct *tsk = current;
 437
 438         memset(tsk->thread.debugreg, 0, sizeof(unsigned long)*8);
 439         memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
 440         clear_tsk_thread_flag(tsk, TIF_DEBUG);
 441         /*
 442          * Forget coprocessor state..
 443          */
 444         clear_fpu(tsk);
 445         clear_used_math();
 446 }
 447
 448 void release_thread(struct task_struct *dead_task)
 449 {
 450         BUG_ON(dead_task->mm);
 451         release_vm86_irqs(dead_task);
 452 }
 453
 454 /*
 455  * This gets called before we allocate a new thread and copy
 456  * the current task into it.
 457  */
 458 void prepare_to_copy(struct task_struct *tsk)
 459 {
 460         unlazy_fpu(tsk);
 461 }
 462
 463 int copy_thread(int nr, unsigned long clone_flags, unsigned long esp,
 464         unsigned long unused,
 465         struct task_struct * p, struct pt_regs * regs)
 466 {
 467         struct pt_regs * childregs;
 468         struct task_struct *tsk;
 469         int err;
 470
 471         childregs = task_pt_regs(p);
 472         *childregs = *regs;
 473         childregs->eax = 0;
 474         childregs->esp = esp;
 475
 476         p->thread.esp = (unsigned long) childregs;
 477         p->thread.esp0 = (unsigned long) (childregs+1);
 478
 479         p->thread.eip = (unsigned long) ret_from_fork;
 480
 481         savesegment(gs,p->thread.gs);
 482
 483         tsk = current;
 484         if (unlikely(test_tsk_thread_flag(tsk, TIF_IO_BITMAP))) {
 485                 p->thread.io_bitmap_ptr = kmemdup(tsk->thread.io_bitmap_ptr,
 486                                                 IO_BITMAP_BYTES, GFP_KERNEL);
 487                 if (!p->thread.io_bitmap_ptr) {
 488                         p->thread.io_bitmap_max = 0;
 489                         return -ENOMEM;
 490                 }
 491                 set_tsk_thread_flag(p, TIF_IO_BITMAP);
 492         }
 493
 494         /*
 495          * Set a new TLS for the child thread?
 496          */
 497         if (clone_flags & CLONE_SETTLS) {
 498                 struct desc_struct *desc;
 499                 struct user_desc info;
 500                 int idx;
 501
 502                 err = -EFAULT;
 503                 if (copy_from_user(&info, (void __user *)childregs->esi, sizeof(info)))
 504                         goto out;
 505                 err = -EINVAL;
 506                 if (LDT_empty(&info))
 507                         goto out;
 508
 509                 idx = info.entry_number;
 510                 if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
 511                         goto out;
 512
 513                 desc = p->thread.tls_array + idx - GDT_ENTRY_TLS_MIN;
 514                 desc->a = LDT_entry_a(&info);
 515                 desc->b = LDT_entry_b(&info);
 516         }
 517
 518         err = 0;
 519  out:
 520         if (err && p->thread.io_bitmap_ptr) {
 521                 kfree(p->thread.io_bitmap_ptr);
 522                 p->thread.io_bitmap_max = 0;
 523         }
 524         return err;
 525 }
 526
 527 /*
 528  * fill in the user structure for a core dump..
 529  */
 530 void dump_thread(struct pt_regs * regs, struct user * dump)
 531 {
 532         int i;
 533
 534 /* changed the size calculations - should hopefully work better. lbt */
 535         dump->magic = CMAGIC;
 536         dump->start_code = 0;
 537         dump->start_stack = regs->esp & ~(PAGE_SIZE - 1);
 538         dump->u_tsize = ((unsigned long) current->mm->end_code) >> PAGE_SHIFT;
 539         dump->u_dsize = ((unsigned long) (current->mm->brk + (PAGE_SIZE-1))) >> PAGE_SHIFT;
 540         dump->u_dsize -= dump->u_tsize;
 541         dump->u_ssize = 0;
 542         for (i = 0; i < 8; i++)
 543                 dump->u_debugreg[i] = current->thread.debugreg[i];
 544
 545         if (dump->start_stack < TASK_SIZE)
 546                 dump->u_ssize = ((unsigned long) (TASK_SIZE - dump->start_stack)) >> PAGE_SHIFT;
 547
 548         dump->regs.ebx = regs->ebx;
 549         dump->regs.ecx = regs->ecx;
 550         dump->regs.edx = regs->edx;
 551         dump->regs.esi = regs->esi;
 552         dump->regs.edi = regs->edi;
 553         dump->regs.ebp = regs->ebp;
 554         dump->regs.eax = regs->eax;
 555         dump->regs.ds = regs->xds;
 556         dump->regs.es = regs->xes;
 557         dump->regs.fs = regs->xfs;
 558         savesegment(gs,dump->regs.gs);
 559         dump->regs.orig_eax = regs->orig_eax;
 560         dump->regs.eip = regs->eip;
 561         dump->regs.cs = regs->xcs;
 562         dump->regs.eflags = regs->eflags;
 563         dump->regs.esp = regs->esp;
 564         dump->regs.ss = regs->xss;
 565
 566         dump->u_fpvalid = dump_fpu (regs, &dump->i387);
 567 }
 568 EXPORT_SYMBOL(dump_thread);
 569
 570 /*
 571  * Capture the user space registers if the task is not running (in user space)
 572  */
 573 int dump_task_regs(struct task_struct *tsk, elf_gregset_t *regs)
 574 {
 575         struct pt_regs ptregs = *task_pt_regs(tsk);
 576         ptregs.xcs &= 0xffff;
 577         ptregs.xds &= 0xffff;
 578         ptregs.xes &= 0xffff;
 579         ptregs.xss &= 0xffff;
 580
 581         elf_core_copy_regs(regs, &ptregs);
 582
 583         return 1;
 584 }
 585
 586 #ifdef CONFIG_SECCOMP
 587 void hard_disable_TSC(void)
 588 {
 589         write_cr4(read_cr4() | X86_CR4_TSD);
 590 }
 591 void disable_TSC(void)
 592 {
 593         preempt_disable();
 594         if (!test_and_set_thread_flag(TIF_NOTSC))
 595                 /*
 596                  * Must flip the CPU state synchronously with
 597                  * TIF_NOTSC in the current running context.
 598                  */
 599                 hard_disable_TSC();
 600         preempt_enable();
 601 }
 602 void hard_enable_TSC(void)
 603 {
 604         write_cr4(read_cr4() & ~X86_CR4_TSD);
 605 }
 606 #endif /* CONFIG_SECCOMP */
 607
 608 static noinline void
 609 __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p,
 610                  struct tss_struct *tss)
 611 {
 612         struct thread_struct *next;
 613
 614         next = &next_p->thread;
 615
 616         if (test_tsk_thread_flag(next_p, TIF_DEBUG)) {
 617                 set_debugreg(next->debugreg[0], 0);
 618                 set_debugreg(next->debugreg[1], 1);
 619                 set_debugreg(next->debugreg[2], 2);
 620                 set_debugreg(next->debugreg[3], 3);
 621                 /* no 4 and 5 */
 622                 set_debugreg(next->debugreg[6], 6);
 623                 set_debugreg(next->debugreg[7], 7);
 624         }
 625
 626 #ifdef CONFIG_SECCOMP
 627         if (test_tsk_thread_flag(prev_p, TIF_NOTSC) ^
 628             test_tsk_thread_flag(next_p, TIF_NOTSC)) {
 629                 /* prev and next are different */
 630                 if (test_tsk_thread_flag(next_p, TIF_NOTSC))
 631                         hard_disable_TSC();
 632                 else
 633                         hard_enable_TSC();
 634         }
 635 #endif
 636
 637         if (!test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
 638                 /*
 639                  * Disable the bitmap via an invalid offset. We still cache
 640                  * the previous bitmap owner and the IO bitmap contents:
 641                  */
 642                 tss->x86_tss.io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
 643                 return;
 644         }
 645
 646         if (likely(next == tss->io_bitmap_owner)) {
 647                 /*
 648                  * Previous owner of the bitmap (hence the bitmap content)
 649                  * matches the next task, we dont have to do anything but
 650                  * to set a valid offset in the TSS:
 651                  */
 652                 tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET;
 653                 return;
 654         }
 655         /*
 656          * Lazy TSS's I/O bitmap copy. We set an invalid offset here
 657          * and we let the task to get a GPF in case an I/O instruction
 658          * is performed.  The handler of the GPF will verify that the
 659          * faulting task has a valid I/O bitmap and, it true, does the
 660          * real copy and restart the instruction.  This will save us
 661          * redundant copies when the currently switched task does not
 662          * perform any I/O during its timeslice.
 663          */
 664         tss->x86_tss.io_bitmap_base = INVALID_IO_BITMAP_OFFSET_LAZY;
 665 }
 666
 667 /*
 668  *      switch_to(x,yn) should switch tasks from x to y.
 669  *
 670  * We fsave/fwait so that an exception goes off at the right time
 671  * (as a call from the fsave or fwait in effect) rather than to
 672  * the wrong process. Lazy FP saving no longer makes any sense
 673  * with modern CPU's, and this simplifies a lot of things (SMP
 674  * and UP become the same).
 675  *
 676  * NOTE! We used to use the x86 hardware context switching. The
 677  * reason for not using it any more becomes apparent when you
 678  * try to recover gracefully from saved state that is no longer
 679  * valid (stale segment register values in particular). With the
 680  * hardware task-switch, there is no way to fix up bad state in
 681  * a reasonable manner.
 682  *
 683  * The fact that Intel documents the hardware task-switching to
 684  * be slow is a fairly red herring - this code is not noticeably
 685  * faster. However, there _is_ some room for improvement here,
 686  * so the performance issues may eventually be a valid point.
 687  * More important, however, is the fact that this allows us much
 688  * more flexibility.
 689  *
 690  * The return value (in %eax) will be the "prev" task after
 691  * the task-switch, and shows up in ret_from_fork in entry.S,
 692  * for example.
 693  */
 694 struct task_struct fastcall * __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
 695 {
 696         struct thread_struct *prev = &prev_p->thread,
 697                                  *next = &next_p->thread;
 698         int cpu = smp_processor_id();
 699         struct tss_struct *tss = &per_cpu(init_tss, cpu);
 700
 701         /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
 702
 703         __unlazy_fpu(prev_p);
 704
 705
 706         /* we're going to use this soon, after a few expensive things */
 707         if (next_p->fpu_counter > 5)
 708                 prefetch(&next->i387.fxsave);
 709
 710         /*
 711          * Reload esp0.
 712          */
 713         load_esp0(tss, next);
 714
 715         /*
 716          * Save away %gs. No need to save %fs, as it was saved on the
 717          * stack on entry.  No need to save %es and %ds, as those are
 718          * always kernel segments while inside the kernel.  Doing this
 719          * before setting the new TLS descriptors avoids the situation
 720          * where we temporarily have non-reloadable segments in %fs
 721          * and %gs.  This could be an issue if the NMI handler ever
 722          * used %fs or %gs (it does not today), or if the kernel is
 723          * running inside of a hypervisor layer.
 724          */
 725         savesegment(gs, prev->gs);
 726
 727         /*
 728          * Load the per-thread Thread-Local Storage descriptor.
 729          */
 730         load_TLS(next, cpu);
 731
 732         /*
 733          * Restore IOPL if needed.  In normal use, the flags restore
 734          * in the switch assembly will handle this.  But if the kernel
 735          * is running virtualized at a non-zero CPL, the popf will
 736          * not restore flags, so it must be done in a separate step.
 737          */
 738         if (get_kernel_rpl() && unlikely(prev->iopl != next->iopl))
 739                 set_iopl_mask(next->iopl);
 740
 741         /*
 742          * Now maybe handle debug registers and/or IO bitmaps
 743          */
 744         if (unlikely(task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV ||
 745                      task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT))
 746                 __switch_to_xtra(prev_p, next_p, tss);
 747
 748         /*
 749          * Leave lazy mode, flushing any hypercalls made here.
 750          * This must be done before restoring TLS segments so
 751          * the GDT and LDT are properly updated, and must be
 752          * done before math_state_restore, so the TS bit is up
 753          * to date.
 754          */
 755         arch_leave_lazy_cpu_mode();
 756
 757         /* If the task has used fpu the last 5 timeslices, just do a full
 758          * restore of the math state immediately to avoid the trap; the
 759          * chances of needing FPU soon are obviously high now
 760          */
 761         if (next_p->fpu_counter > 5)
 762                 math_state_restore();
 763
 764         /*
 765          * Restore %gs if needed (which is common)
 766          */
 767         if (prev->gs | next->gs)
 768                 loadsegment(gs, next->gs);
 769
 770         x86_write_percpu(current_task, next_p);
 771
 772         return prev_p;
 773 }
 774
 775 asmlinkage int sys_fork(struct pt_regs regs)
 776 {
 777         return do_fork(SIGCHLD, regs.esp, &regs, 0, NULL, NULL);
 778 }
 779
 780 asmlinkage int sys_clone(struct pt_regs regs)
 781 {
 782         unsigned long clone_flags;
 783         unsigned long newsp;
 784         int __user *parent_tidptr, *child_tidptr;
 785
 786         clone_flags = regs.ebx;
 787         newsp = regs.ecx;
 788         parent_tidptr = (int __user *)regs.edx;
 789         child_tidptr = (int __user *)regs.edi;
 790         if (!newsp)
 791                 newsp = regs.esp;
 792         return do_fork(clone_flags, newsp, &regs, 0, parent_tidptr, child_tidptr);
 793 }
 794
 795 /*
 796  * This is trivial, and on the face of it looks like it
 797  * could equally well be done in user mode.
 798  *
 799  * Not so, for quite unobvious reasons - register pressure.
 800  * In user mode vfork() cannot have a stack frame, and if
 801  * done by calling the "clone()" system call directly, you
 802  * do not have enough call-clobbered registers to hold all
 803  * the information you need.
 804  */
 805 asmlinkage int sys_vfork(struct pt_regs regs)
 806 {
 807         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs.esp, &regs, 0, NULL, NULL);
 808 }
 809
 810 /*
 811  * sys_execve() executes a new program.
 812  */
 813 asmlinkage int sys_execve(struct pt_regs regs)
 814 {
 815         int error;
 816         char * filename;
 817
 818         filename = getname((char __user *) regs.ebx);
 819         error = PTR_ERR(filename);
 820         if (IS_ERR(filename))
 821                 goto out;
 822         error = do_execve(filename,
 823                         (char __user * __user *) regs.ecx,
 824                         (char __user * __user *) regs.edx,
 825                         &regs);
 826         if (error == 0) {
 827                 task_lock(current);
 828                 current->ptrace &= ~PT_DTRACE;
 829                 task_unlock(current);
 830                 /* Make sure we don't return using sysenter.. */
 831                 set_thread_flag(TIF_IRET);
 832         }
 833         putname(filename);
 834 out:
 835         return error;
 836 }
 837
 838 #define top_esp                (THREAD_SIZE - sizeof(unsigned long))
 839 #define top_ebp                (THREAD_SIZE - 2*sizeof(unsigned long))
 840
 841 unsigned long get_wchan(struct task_struct *p)
 842 {
 843         unsigned long ebp, esp, eip;
 844         unsigned long stack_page;
 845         int count = 0;
 846         if (!p || p == current || p->state == TASK_RUNNING)
 847                 return 0;
 848         stack_page = (unsigned long)task_stack_page(p);
 849         esp = p->thread.esp;
 850         if (!stack_page || esp < stack_page || esp > top_esp+stack_page)
 851                 return 0;
 852         /* include/asm-i386/system.h:switch_to() pushes ebp last. */
 853         ebp = *(unsigned long *) esp;
 854         do {
 855                 if (ebp < stack_page || ebp > top_ebp+stack_page)
 856                         return 0;
 857                 eip = *(unsigned long *) (ebp+4);
 858                 if (!in_sched_functions(eip))
 859                         return eip;
 860                 ebp = *(unsigned long *) ebp;
 861         } while (count++ < 16);
 862         return 0;
 863 }
 864
 865 /*
 866  * sys_alloc_thread_area: get a yet unused TLS descriptor index.
 867  */
 868 static int get_free_idx(void)
 869 {
 870         struct thread_struct *t = &current->thread;
 871         int idx;
 872
 873         for (idx = 0; idx < GDT_ENTRY_TLS_ENTRIES; idx++)
 874                 if (desc_empty(t->tls_array + idx))
 875                         return idx + GDT_ENTRY_TLS_MIN;
 876         return -ESRCH;
 877 }
 878
 879 /*
 880  * Set a given TLS descriptor:
 881  */
 882 asmlinkage int sys_set_thread_area(struct user_desc __user *u_info)
 883 {
 884         struct thread_struct *t = &current->thread;
 885         struct user_desc info;
 886         struct desc_struct *desc;
 887         int cpu, idx;
 888
 889         if (copy_from_user(&info, u_info, sizeof(info)))
 890                 return -EFAULT;
 891         idx = info.entry_number;
 892
 893         /*
 894          * index -1 means the kernel should try to find and
 895          * allocate an empty descriptor:
 896          */
 897         if (idx == -1) {
 898                 idx = get_free_idx();
 899                 if (idx < 0)
 900                         return idx;
 901                 if (put_user(idx, &u_info->entry_number))
 902                         return -EFAULT;
 903         }
 904
 905         if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
 906                 return -EINVAL;
 907
 908         desc = t->tls_array + idx - GDT_ENTRY_TLS_MIN;
 909
 910         /*
 911          * We must not get preempted while modifying the TLS.
 912          */
 913         cpu = get_cpu();
 914
 915         if (LDT_empty(&info)) {
 916                 desc->a = 0;
 917                 desc->b = 0;
 918         } else {
 919                 desc->a = LDT_entry_a(&info);
 920                 desc->b = LDT_entry_b(&info);
 921         }
 922         load_TLS(t, cpu);
 923
 924         put_cpu();
 925
 926         return 0;
 927 }
 928
 929 /*
 930  * Get the current Thread-Local Storage area:
 931  */
 932
 933 #define GET_BASE(desc) ( \
 934         (((desc)->a >> 16) & 0x0000ffff) | \
 935         (((desc)->b << 16) & 0x00ff0000) | \
 936         ( (desc)->b        & 0xff000000)   )
 937
 938 #define GET_LIMIT(desc) ( \
 939         ((desc)->a & 0x0ffff) | \
 940          ((desc)->b & 0xf0000) )
 941
 942 #define GET_32BIT(desc)         (((desc)->b >> 22) & 1)
 943 #define GET_CONTENTS(desc)      (((desc)->b >> 10) & 3)
 944 #define GET_WRITABLE(desc)      (((desc)->b >>  9) & 1)
 945 #define GET_LIMIT_PAGES(desc)   (((desc)->b >> 23) & 1)
 946 #define GET_PRESENT(desc)       (((desc)->b >> 15) & 1)
 947 #define GET_USEABLE(desc)       (((desc)->b >> 20) & 1)
 948
 949 asmlinkage int sys_get_thread_area(struct user_desc __user *u_info)
 950 {
 951         struct user_desc info;
 952         struct desc_struct *desc;
 953         int idx;
 954
 955         if (get_user(idx, &u_info->entry_number))
 956                 return -EFAULT;
 957         if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
 958                 return -EINVAL;
 959
 960         memset(&info, 0, sizeof(info));
 961
 962         desc = current->thread.tls_array + idx - GDT_ENTRY_TLS_MIN;
 963
 964         info.entry_number = idx;
 965         info.base_addr = GET_BASE(desc);
 966         info.limit = GET_LIMIT(desc);
 967         info.seg_32bit = GET_32BIT(desc);
 968         info.contents = GET_CONTENTS(desc);
 969         info.read_exec_only = !GET_WRITABLE(desc);
 970         info.limit_in_pages = GET_LIMIT_PAGES(desc);
 971         info.seg_not_present = !GET_PRESENT(desc);
 972         info.useable = GET_USEABLE(desc);
 973
 974         if (copy_to_user(u_info, &info, sizeof(info)))
 975                 return -EFAULT;
 976         return 0;
 977 }
 978
 979 unsigned long arch_align_stack(unsigned long sp)
 980 {
 981         if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
 982                 sp -= get_random_int() % 8192;
 983         return sp & ~0xf;
 984 }