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/utsname.h>
  32 #include <linux/delay.h>
  33 #include <linux/reboot.h>
  34 #include <linux/init.h>
  35 #include <linux/mc146818rtc.h>
  36 #include <linux/module.h>
  37 #include <linux/kallsyms.h>
  38 #include <linux/ptrace.h>
  39 #include <linux/random.h>
  40 #include <linux/personality.h>
  41
  42 #include <asm/uaccess.h>
  43 #include <asm/pgtable.h>
  44 #include <asm/system.h>
  45 #include <asm/io.h>
  46 #include <asm/ldt.h>
  47 #include <asm/processor.h>
  48 #include <asm/i387.h>
  49 #include <asm/desc.h>
  50 #include <asm/vm86.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                 current_thread_info()->status &= ~TS_POLLING;
 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                 current_thread_info()->status |= TS_POLLING;
 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         current_thread_info()->status |= TS_POLLING;
 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_vm(regs))
 298                 printk(" ESP: %04x:%08lx",0xffff & regs->xss,regs->esp);
 299         printk(" EFLAGS: %08lx    %s  (%s %.*s)\n",
 300                regs->eflags, print_tainted(), system_utsname.release,
 301                (int)strcspn(system_utsname.version, " "),
 302                system_utsname.version);
 303         printk("EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
 304                 regs->eax,regs->ebx,regs->ecx,regs->edx);
 305         printk("ESI: %08lx EDI: %08lx EBP: %08lx",
 306                 regs->esi, regs->edi, regs->ebp);
 307         printk(" DS: %04x ES: %04x\n",
 308                 0xffff & regs->xds,0xffff & regs->xes);
 309
 310         cr0 = read_cr0();
 311         cr2 = read_cr2();
 312         cr3 = read_cr3();
 313         cr4 = read_cr4_safe();
 314         printk("CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n", cr0, cr2, cr3, cr4);
 315         show_trace(NULL, regs, &regs->esp);
 316 }
 317
 318 /*
 319  * This gets run with %ebx containing the
 320  * function to call, and %edx containing
 321  * the "args".
 322  */
 323 extern void kernel_thread_helper(void);
 324
 325 /*
 326  * Create a kernel thread
 327  */
 328 int kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
 329 {
 330         struct pt_regs regs;
 331
 332         memset(&regs, 0, sizeof(regs));
 333
 334         regs.ebx = (unsigned long) fn;
 335         regs.edx = (unsigned long) arg;
 336
 337         regs.xds = __USER_DS;
 338         regs.xes = __USER_DS;
 339         regs.orig_eax = -1;
 340         regs.eip = (unsigned long) kernel_thread_helper;
 341         regs.xcs = __KERNEL_CS | get_kernel_rpl();
 342         regs.eflags = X86_EFLAGS_IF | X86_EFLAGS_SF | X86_EFLAGS_PF | 0x2;
 343
 344         /* Ok, create the new process.. */
 345         return do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0, &regs, 0, NULL, NULL);
 346 }
 347 EXPORT_SYMBOL(kernel_thread);
 348
 349 /*
 350  * Free current thread data structures etc..
 351  */
 352 void exit_thread(void)
 353 {
 354         /* The process may have allocated an io port bitmap... nuke it. */
 355         if (unlikely(test_thread_flag(TIF_IO_BITMAP))) {
 356                 struct task_struct *tsk = current;
 357                 struct thread_struct *t = &tsk->thread;
 358                 int cpu = get_cpu();
 359                 struct tss_struct *tss = &per_cpu(init_tss, cpu);
 360
 361                 kfree(t->io_bitmap_ptr);
 362                 t->io_bitmap_ptr = NULL;
 363                 clear_thread_flag(TIF_IO_BITMAP);
 364                 /*
 365                  * Careful, clear this in the TSS too:
 366                  */
 367                 memset(tss->io_bitmap, 0xff, tss->io_bitmap_max);
 368                 t->io_bitmap_max = 0;
 369                 tss->io_bitmap_owner = NULL;
 370                 tss->io_bitmap_max = 0;
 371                 tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
 372                 put_cpu();
 373         }
 374 }
 375
 376 void flush_thread(void)
 377 {
 378         struct task_struct *tsk = current;
 379
 380         memset(tsk->thread.debugreg, 0, sizeof(unsigned long)*8);
 381         memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
 382         clear_tsk_thread_flag(tsk, TIF_DEBUG);
 383         /*
 384          * Forget coprocessor state..
 385          */
 386         clear_fpu(tsk);
 387         clear_used_math();
 388 }
 389
 390 void release_thread(struct task_struct *dead_task)
 391 {
 392         BUG_ON(dead_task->mm);
 393         release_vm86_irqs(dead_task);
 394 }
 395
 396 /*
 397  * This gets called before we allocate a new thread and copy
 398  * the current task into it.
 399  */
 400 void prepare_to_copy(struct task_struct *tsk)
 401 {
 402         unlazy_fpu(tsk);
 403 }
 404
 405 int copy_thread(int nr, unsigned long clone_flags, unsigned long esp,
 406         unsigned long unused,
 407         struct task_struct * p, struct pt_regs * regs)
 408 {
 409         struct pt_regs * childregs;
 410         struct task_struct *tsk;
 411         int err;
 412
 413         childregs = task_pt_regs(p);
 414         *childregs = *regs;
 415         childregs->eax = 0;
 416         childregs->esp = esp;
 417
 418         p->thread.esp = (unsigned long) childregs;
 419         p->thread.esp0 = (unsigned long) (childregs+1);
 420
 421         p->thread.eip = (unsigned long) ret_from_fork;
 422
 423         savesegment(fs,p->thread.fs);
 424         savesegment(gs,p->thread.gs);
 425
 426         tsk = current;
 427         if (unlikely(test_tsk_thread_flag(tsk, TIF_IO_BITMAP))) {
 428                 p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL);
 429                 if (!p->thread.io_bitmap_ptr) {
 430                         p->thread.io_bitmap_max = 0;
 431                         return -ENOMEM;
 432                 }
 433                 memcpy(p->thread.io_bitmap_ptr, tsk->thread.io_bitmap_ptr,
 434                         IO_BITMAP_BYTES);
 435                 set_tsk_thread_flag(p, TIF_IO_BITMAP);
 436         }
 437
 438         /*
 439          * Set a new TLS for the child thread?
 440          */
 441         if (clone_flags & CLONE_SETTLS) {
 442                 struct desc_struct *desc;
 443                 struct user_desc info;
 444                 int idx;
 445
 446                 err = -EFAULT;
 447                 if (copy_from_user(&info, (void __user *)childregs->esi, sizeof(info)))
 448                         goto out;
 449                 err = -EINVAL;
 450                 if (LDT_empty(&info))
 451                         goto out;
 452
 453                 idx = info.entry_number;
 454                 if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
 455                         goto out;
 456
 457                 desc = p->thread.tls_array + idx - GDT_ENTRY_TLS_MIN;
 458                 desc->a = LDT_entry_a(&info);
 459                 desc->b = LDT_entry_b(&info);
 460         }
 461
 462         err = 0;
 463  out:
 464         if (err && p->thread.io_bitmap_ptr) {
 465                 kfree(p->thread.io_bitmap_ptr);
 466                 p->thread.io_bitmap_max = 0;
 467         }
 468         return err;
 469 }
 470
 471 /*
 472  * fill in the user structure for a core dump..
 473  */
 474 void dump_thread(struct pt_regs * regs, struct user * dump)
 475 {
 476         int i;
 477
 478 /* changed the size calculations - should hopefully work better. lbt */
 479         dump->magic = CMAGIC;
 480         dump->start_code = 0;
 481         dump->start_stack = regs->esp & ~(PAGE_SIZE - 1);
 482         dump->u_tsize = ((unsigned long) current->mm->end_code) >> PAGE_SHIFT;
 483         dump->u_dsize = ((unsigned long) (current->mm->brk + (PAGE_SIZE-1))) >> PAGE_SHIFT;
 484         dump->u_dsize -= dump->u_tsize;
 485         dump->u_ssize = 0;
 486         for (i = 0; i < 8; i++)
 487                 dump->u_debugreg[i] = current->thread.debugreg[i];
 488
 489         if (dump->start_stack < TASK_SIZE)
 490                 dump->u_ssize = ((unsigned long) (TASK_SIZE - dump->start_stack)) >> PAGE_SHIFT;
 491
 492         dump->regs.ebx = regs->ebx;
 493         dump->regs.ecx = regs->ecx;
 494         dump->regs.edx = regs->edx;
 495         dump->regs.esi = regs->esi;
 496         dump->regs.edi = regs->edi;
 497         dump->regs.ebp = regs->ebp;
 498         dump->regs.eax = regs->eax;
 499         dump->regs.ds = regs->xds;
 500         dump->regs.es = regs->xes;
 501         savesegment(fs,dump->regs.fs);
 502         savesegment(gs,dump->regs.gs);
 503         dump->regs.orig_eax = regs->orig_eax;
 504         dump->regs.eip = regs->eip;
 505         dump->regs.cs = regs->xcs;
 506         dump->regs.eflags = regs->eflags;
 507         dump->regs.esp = regs->esp;
 508         dump->regs.ss = regs->xss;
 509
 510         dump->u_fpvalid = dump_fpu (regs, &dump->i387);
 511 }
 512 EXPORT_SYMBOL(dump_thread);
 513
 514 /*
 515  * Capture the user space registers if the task is not running (in user space)
 516  */
 517 int dump_task_regs(struct task_struct *tsk, elf_gregset_t *regs)
 518 {
 519         struct pt_regs ptregs = *task_pt_regs(tsk);
 520         ptregs.xcs &= 0xffff;
 521         ptregs.xds &= 0xffff;
 522         ptregs.xes &= 0xffff;
 523         ptregs.xss &= 0xffff;
 524
 525         elf_core_copy_regs(regs, &ptregs);
 526
 527         return 1;
 528 }
 529
 530 static noinline void __switch_to_xtra(struct task_struct *next_p,
 531                                     struct tss_struct *tss)
 532 {
 533         struct thread_struct *next;
 534
 535         next = &next_p->thread;
 536
 537         if (test_tsk_thread_flag(next_p, TIF_DEBUG)) {
 538                 set_debugreg(next->debugreg[0], 0);
 539                 set_debugreg(next->debugreg[1], 1);
 540                 set_debugreg(next->debugreg[2], 2);
 541                 set_debugreg(next->debugreg[3], 3);
 542                 /* no 4 and 5 */
 543                 set_debugreg(next->debugreg[6], 6);
 544                 set_debugreg(next->debugreg[7], 7);
 545         }
 546
 547         if (!test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
 548                 /*
 549                  * Disable the bitmap via an invalid offset. We still cache
 550                  * the previous bitmap owner and the IO bitmap contents:
 551                  */
 552                 tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
 553                 return;
 554         }
 555
 556         if (likely(next == tss->io_bitmap_owner)) {
 557                 /*
 558                  * Previous owner of the bitmap (hence the bitmap content)
 559                  * matches the next task, we dont have to do anything but
 560                  * to set a valid offset in the TSS:
 561                  */
 562                 tss->io_bitmap_base = IO_BITMAP_OFFSET;
 563                 return;
 564         }
 565         /*
 566          * Lazy TSS's I/O bitmap copy. We set an invalid offset here
 567          * and we let the task to get a GPF in case an I/O instruction
 568          * is performed.  The handler of the GPF will verify that the
 569          * faulting task has a valid I/O bitmap and, it true, does the
 570          * real copy and restart the instruction.  This will save us
 571          * redundant copies when the currently switched task does not
 572          * perform any I/O during its timeslice.
 573          */
 574         tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET_LAZY;
 575 }
 576
 577 /*
 578  * This function selects if the context switch from prev to next
 579  * has to tweak the TSC disable bit in the cr4.
 580  */
 581 static inline void disable_tsc(struct task_struct *prev_p,
 582                                struct task_struct *next_p)
 583 {
 584         struct thread_info *prev, *next;
 585
 586         /*
 587          * gcc should eliminate the ->thread_info dereference if
 588          * has_secure_computing returns 0 at compile time (SECCOMP=n).
 589          */
 590         prev = task_thread_info(prev_p);
 591         next = task_thread_info(next_p);
 592
 593         if (has_secure_computing(prev) || has_secure_computing(next)) {
 594                 /* slow path here */
 595                 if (has_secure_computing(prev) &&
 596                     !has_secure_computing(next)) {
 597                         write_cr4(read_cr4() & ~X86_CR4_TSD);
 598                 } else if (!has_secure_computing(prev) &&
 599                            has_secure_computing(next))
 600                         write_cr4(read_cr4() | X86_CR4_TSD);
 601         }
 602 }
 603
 604 /*
 605  *      switch_to(x,yn) should switch tasks from x to y.
 606  *
 607  * We fsave/fwait so that an exception goes off at the right time
 608  * (as a call from the fsave or fwait in effect) rather than to
 609  * the wrong process. Lazy FP saving no longer makes any sense
 610  * with modern CPU's, and this simplifies a lot of things (SMP
 611  * and UP become the same).
 612  *
 613  * NOTE! We used to use the x86 hardware context switching. The
 614  * reason for not using it any more becomes apparent when you
 615  * try to recover gracefully from saved state that is no longer
 616  * valid (stale segment register values in particular). With the
 617  * hardware task-switch, there is no way to fix up bad state in
 618  * a reasonable manner.
 619  *
 620  * The fact that Intel documents the hardware task-switching to
 621  * be slow is a fairly red herring - this code is not noticeably
 622  * faster. However, there _is_ some room for improvement here,
 623  * so the performance issues may eventually be a valid point.
 624  * More important, however, is the fact that this allows us much
 625  * more flexibility.
 626  *
 627  * The return value (in %eax) will be the "prev" task after
 628  * the task-switch, and shows up in ret_from_fork in entry.S,
 629  * for example.
 630  */
 631 struct task_struct fastcall * __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
 632 {
 633         struct thread_struct *prev = &prev_p->thread,
 634                                  *next = &next_p->thread;
 635         int cpu = smp_processor_id();
 636         struct tss_struct *tss = &per_cpu(init_tss, cpu);
 637
 638         /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
 639
 640         __unlazy_fpu(prev_p);
 641
 642         /*
 643          * Reload esp0.
 644          */
 645         load_esp0(tss, next);
 646
 647         /*
 648          * Save away %fs and %gs. No need to save %es and %ds, as
 649          * those are always kernel segments while inside the kernel.
 650          * Doing this before setting the new TLS descriptors avoids
 651          * the situation where we temporarily have non-reloadable
 652          * segments in %fs and %gs.  This could be an issue if the
 653          * NMI handler ever used %fs or %gs (it does not today), or
 654          * if the kernel is running inside of a hypervisor layer.
 655          */
 656         savesegment(fs, prev->fs);
 657         savesegment(gs, prev->gs);
 658
 659         /*
 660          * Load the per-thread Thread-Local Storage descriptor.
 661          */
 662         load_TLS(next, cpu);
 663
 664         /*
 665          * Restore %fs and %gs if needed.
 666          *
 667          * Glibc normally makes %fs be zero, and %gs is one of
 668          * the TLS segments.
 669          */
 670         if (unlikely(prev->fs | next->fs))
 671                 loadsegment(fs, next->fs);
 672
 673         if (prev->gs | next->gs)
 674                 loadsegment(gs, next->gs);
 675
 676         /*
 677          * Restore IOPL if needed.
 678          */
 679         if (unlikely(prev->iopl != next->iopl))
 680                 set_iopl_mask(next->iopl);
 681
 682         /*
 683          * Now maybe handle debug registers and/or IO bitmaps
 684          */
 685         if (unlikely((task_thread_info(next_p)->flags & _TIF_WORK_CTXSW)
 686             || test_tsk_thread_flag(prev_p, TIF_IO_BITMAP)))
 687                 __switch_to_xtra(next_p, tss);
 688
 689         disable_tsc(prev_p, next_p);
 690
 691         return prev_p;
 692 }
 693
 694 asmlinkage int sys_fork(struct pt_regs regs)
 695 {
 696         return do_fork(SIGCHLD, regs.esp, &regs, 0, NULL, NULL);
 697 }
 698
 699 asmlinkage int sys_clone(struct pt_regs regs)
 700 {
 701         unsigned long clone_flags;
 702         unsigned long newsp;
 703         int __user *parent_tidptr, *child_tidptr;
 704
 705         clone_flags = regs.ebx;
 706         newsp = regs.ecx;
 707         parent_tidptr = (int __user *)regs.edx;
 708         child_tidptr = (int __user *)regs.edi;
 709         if (!newsp)
 710                 newsp = regs.esp;
 711         return do_fork(clone_flags, newsp, &regs, 0, parent_tidptr, child_tidptr);
 712 }
 713
 714 /*
 715  * This is trivial, and on the face of it looks like it
 716  * could equally well be done in user mode.
 717  *
 718  * Not so, for quite unobvious reasons - register pressure.
 719  * In user mode vfork() cannot have a stack frame, and if
 720  * done by calling the "clone()" system call directly, you
 721  * do not have enough call-clobbered registers to hold all
 722  * the information you need.
 723  */
 724 asmlinkage int sys_vfork(struct pt_regs regs)
 725 {
 726         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs.esp, &regs, 0, NULL, NULL);
 727 }
 728
 729 /*
 730  * sys_execve() executes a new program.
 731  */
 732 asmlinkage int sys_execve(struct pt_regs regs)
 733 {
 734         int error;
 735         char * filename;
 736
 737         filename = getname((char __user *) regs.ebx);
 738         error = PTR_ERR(filename);
 739         if (IS_ERR(filename))
 740                 goto out;
 741         error = do_execve(filename,
 742                         (char __user * __user *) regs.ecx,
 743                         (char __user * __user *) regs.edx,
 744                         &regs);
 745         if (error == 0) {
 746                 task_lock(current);
 747                 current->ptrace &= ~PT_DTRACE;
 748                 task_unlock(current);
 749                 /* Make sure we don't return using sysenter.. */
 750                 set_thread_flag(TIF_IRET);
 751         }
 752         putname(filename);
 753 out:
 754         return error;
 755 }
 756
 757 #define top_esp                (THREAD_SIZE - sizeof(unsigned long))
 758 #define top_ebp                (THREAD_SIZE - 2*sizeof(unsigned long))
 759
 760 unsigned long get_wchan(struct task_struct *p)
 761 {
 762         unsigned long ebp, esp, eip;
 763         unsigned long stack_page;
 764         int count = 0;
 765         if (!p || p == current || p->state == TASK_RUNNING)
 766                 return 0;
 767         stack_page = (unsigned long)task_stack_page(p);
 768         esp = p->thread.esp;
 769         if (!stack_page || esp < stack_page || esp > top_esp+stack_page)
 770                 return 0;
 771         /* include/asm-i386/system.h:switch_to() pushes ebp last. */
 772         ebp = *(unsigned long *) esp;
 773         do {
 774                 if (ebp < stack_page || ebp > top_ebp+stack_page)
 775                         return 0;
 776                 eip = *(unsigned long *) (ebp+4);
 777                 if (!in_sched_functions(eip))
 778                         return eip;
 779                 ebp = *(unsigned long *) ebp;
 780         } while (count++ < 16);
 781         return 0;
 782 }
 783
 784 /*
 785  * sys_alloc_thread_area: get a yet unused TLS descriptor index.
 786  */
 787 static int get_free_idx(void)
 788 {
 789         struct thread_struct *t = &current->thread;
 790         int idx;
 791
 792         for (idx = 0; idx < GDT_ENTRY_TLS_ENTRIES; idx++)
 793                 if (desc_empty(t->tls_array + idx))
 794                         return idx + GDT_ENTRY_TLS_MIN;
 795         return -ESRCH;
 796 }
 797
 798 /*
 799  * Set a given TLS descriptor:
 800  */
 801 asmlinkage int sys_set_thread_area(struct user_desc __user *u_info)
 802 {
 803         struct thread_struct *t = &current->thread;
 804         struct user_desc info;
 805         struct desc_struct *desc;
 806         int cpu, idx;
 807
 808         if (copy_from_user(&info, u_info, sizeof(info)))
 809                 return -EFAULT;
 810         idx = info.entry_number;
 811
 812         /*
 813          * index -1 means the kernel should try to find and
 814          * allocate an empty descriptor:
 815          */
 816         if (idx == -1) {
 817                 idx = get_free_idx();
 818                 if (idx < 0)
 819                         return idx;
 820                 if (put_user(idx, &u_info->entry_number))
 821                         return -EFAULT;
 822         }
 823
 824         if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
 825                 return -EINVAL;
 826
 827         desc = t->tls_array + idx - GDT_ENTRY_TLS_MIN;
 828
 829         /*
 830          * We must not get preempted while modifying the TLS.
 831          */
 832         cpu = get_cpu();
 833
 834         if (LDT_empty(&info)) {
 835                 desc->a = 0;
 836                 desc->b = 0;
 837         } else {
 838                 desc->a = LDT_entry_a(&info);
 839                 desc->b = LDT_entry_b(&info);
 840         }
 841         load_TLS(t, cpu);
 842
 843         put_cpu();
 844
 845         return 0;
 846 }
 847
 848 /*
 849  * Get the current Thread-Local Storage area:
 850  */
 851
 852 #define GET_BASE(desc) ( \
 853         (((desc)->a >> 16) & 0x0000ffff) | \
 854         (((desc)->b << 16) & 0x00ff0000) | \
 855         ( (desc)->b        & 0xff000000)   )
 856
 857 #define GET_LIMIT(desc) ( \
 858         ((desc)->a & 0x0ffff) | \
 859          ((desc)->b & 0xf0000) )
 860
 861 #define GET_32BIT(desc)         (((desc)->b >> 22) & 1)
 862 #define GET_CONTENTS(desc)      (((desc)->b >> 10) & 3)
 863 #define GET_WRITABLE(desc)      (((desc)->b >>  9) & 1)
 864 #define GET_LIMIT_PAGES(desc)   (((desc)->b >> 23) & 1)
 865 #define GET_PRESENT(desc)       (((desc)->b >> 15) & 1)
 866 #define GET_USEABLE(desc)       (((desc)->b >> 20) & 1)
 867
 868 asmlinkage int sys_get_thread_area(struct user_desc __user *u_info)
 869 {
 870         struct user_desc info;
 871         struct desc_struct *desc;
 872         int idx;
 873
 874         if (get_user(idx, &u_info->entry_number))
 875                 return -EFAULT;
 876         if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
 877                 return -EINVAL;
 878
 879         memset(&info, 0, sizeof(info));
 880
 881         desc = current->thread.tls_array + idx - GDT_ENTRY_TLS_MIN;
 882
 883         info.entry_number = idx;
 884         info.base_addr = GET_BASE(desc);
 885         info.limit = GET_LIMIT(desc);
 886         info.seg_32bit = GET_32BIT(desc);
 887         info.contents = GET_CONTENTS(desc);
 888         info.read_exec_only = !GET_WRITABLE(desc);
 889         info.limit_in_pages = GET_LIMIT_PAGES(desc);
 890         info.seg_not_present = !GET_PRESENT(desc);
 891         info.useable = GET_USEABLE(desc);
 892
 893         if (copy_to_user(u_info, &info, sizeof(info)))
 894                 return -EFAULT;
 895         return 0;
 896 }
 897
 898 unsigned long arch_align_stack(unsigned long sp)
 899 {
 900         if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
 901                 sp -= get_random_int() % 8192;
 902         return sp & ~0xf;
 903 }