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