arch/x86_64/kernel/process.c

   1 /*
   2  *  linux/arch/x86-64/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  *  X86-64 port
  10  *      Andi Kleen.
  11  *
  12  *      CPU hotplug support - ashok.raj@intel.com
  13  */
  14
  15 /*
  16  * This file handles the architecture-dependent parts of process handling..
  17  */
  18
  19 #include <stdarg.h>
  20
  21 #include <linux/cpu.h>
  22 #include <linux/errno.h>
  23 #include <linux/sched.h>
  24 #include <linux/kernel.h>
  25 #include <linux/mm.h>
  26 #include <linux/elfcore.h>
  27 #include <linux/smp.h>
  28 #include <linux/slab.h>
  29 #include <linux/user.h>
  30 #include <linux/module.h>
  31 #include <linux/a.out.h>
  32 #include <linux/interrupt.h>
  33 #include <linux/delay.h>
  34 #include <linux/ptrace.h>
  35 #include <linux/utsname.h>
  36 #include <linux/random.h>
  37 #include <linux/notifier.h>
  38 #include <linux/kprobes.h>
  39
  40 #include <asm/uaccess.h>
  41 #include <asm/pgtable.h>
  42 #include <asm/system.h>
  43 #include <asm/io.h>
  44 #include <asm/processor.h>
  45 #include <asm/i387.h>
  46 #include <asm/mmu_context.h>
  47 #include <asm/pda.h>
  48 #include <asm/prctl.h>
  49 #include <asm/kdebug.h>
  50 #include <asm/desc.h>
  51 #include <asm/proto.h>
  52 #include <asm/ia32.h>
  53 #include <asm/idle.h>
  54
  55 asmlinkage extern void ret_from_fork(void);
  56
  57 unsigned long kernel_thread_flags =
  58         CLONE_VM | CLONE_UNTRACED | CLONE_KTHREAD;
  59
  60 unsigned long boot_option_idle_override = 0;
  61 EXPORT_SYMBOL(boot_option_idle_override);
  62
  63 /*
  64  * Powermanagement idle function, if any..
  65  */
  66 void (*pm_idle)(void);
  67 EXPORT_SYMBOL(pm_idle);
  68 static DEFINE_PER_CPU(unsigned int, cpu_idle_state);
  69
  70 static ATOMIC_NOTIFIER_HEAD(idle_notifier);
  71
  72 void idle_notifier_register(struct notifier_block *n)
  73 {
  74         atomic_notifier_chain_register(&idle_notifier, n);
  75 }
  76 EXPORT_SYMBOL_GPL(idle_notifier_register);
  77
  78 void idle_notifier_unregister(struct notifier_block *n)
  79 {
  80         atomic_notifier_chain_unregister(&idle_notifier, n);
  81 }
  82 EXPORT_SYMBOL(idle_notifier_unregister);
  83
  84 void enter_idle(void)
  85 {
  86         write_pda(isidle, 1);
  87         atomic_notifier_call_chain(&idle_notifier, IDLE_START, NULL);
  88 }
  89
  90 static void __exit_idle(void)
  91 {
  92         if (test_and_clear_bit_pda(0, isidle) == 0)
  93                 return;
  94         atomic_notifier_call_chain(&idle_notifier, IDLE_END, NULL);
  95 }
  96
  97 /* Called from interrupts to signify idle end */
  98 void exit_idle(void)
  99 {
 100         /* idle loop has pid 0 */
 101         if (current->pid)
 102                 return;
 103         __exit_idle();
 104 }
 105
 106 /*
 107  * We use this if we don't have any better
 108  * idle routine..
 109  */
 110 static void default_idle(void)
 111 {
 112         current_thread_info()->status &= ~TS_POLLING;
 113         /*
 114          * TS_POLLING-cleared state must be visible before we
 115          * test NEED_RESCHED:
 116          */
 117         smp_mb();
 118         local_irq_disable();
 119         if (!need_resched()) {
 120                 /* Enables interrupts one instruction before HLT.
 121                    x86 special cases this so there is no race. */
 122                 safe_halt();
 123         } else
 124                 local_irq_enable();
 125         current_thread_info()->status |= TS_POLLING;
 126 }
 127
 128 /*
 129  * On SMP it's slightly faster (but much more power-consuming!)
 130  * to poll the ->need_resched flag instead of waiting for the
 131  * cross-CPU IPI to arrive. Use this option with caution.
 132  */
 133 static void poll_idle (void)
 134 {
 135         local_irq_enable();
 136         cpu_relax();
 137 }
 138
 139 void cpu_idle_wait(void)
 140 {
 141         unsigned int cpu, this_cpu = get_cpu();
 142         cpumask_t map, tmp = current->cpus_allowed;
 143
 144         set_cpus_allowed(current, cpumask_of_cpu(this_cpu));
 145         put_cpu();
 146
 147         cpus_clear(map);
 148         for_each_online_cpu(cpu) {
 149                 per_cpu(cpu_idle_state, cpu) = 1;
 150                 cpu_set(cpu, map);
 151         }
 152
 153         __get_cpu_var(cpu_idle_state) = 0;
 154
 155         wmb();
 156         do {
 157                 ssleep(1);
 158                 for_each_online_cpu(cpu) {
 159                         if (cpu_isset(cpu, map) &&
 160                                         !per_cpu(cpu_idle_state, cpu))
 161                                 cpu_clear(cpu, map);
 162                 }
 163                 cpus_and(map, map, cpu_online_map);
 164         } while (!cpus_empty(map));
 165
 166         set_cpus_allowed(current, tmp);
 167 }
 168 EXPORT_SYMBOL_GPL(cpu_idle_wait);
 169
 170 #ifdef CONFIG_HOTPLUG_CPU
 171 DECLARE_PER_CPU(int, cpu_state);
 172
 173 #include <asm/nmi.h>
 174 /* We halt the CPU with physical CPU hotplug */
 175 static inline void play_dead(void)
 176 {
 177         idle_task_exit();
 178         wbinvd();
 179         mb();
 180         /* Ack it */
 181         __get_cpu_var(cpu_state) = CPU_DEAD;
 182
 183         local_irq_disable();
 184         while (1)
 185                 halt();
 186 }
 187 #else
 188 static inline void play_dead(void)
 189 {
 190         BUG();
 191 }
 192 #endif /* CONFIG_HOTPLUG_CPU */
 193
 194 /*
 195  * The idle thread. There's no useful work to be
 196  * done, so just try to conserve power and have a
 197  * low exit latency (ie sit in a loop waiting for
 198  * somebody to say that they'd like to reschedule)
 199  */
 200 void cpu_idle (void)
 201 {
 202         current_thread_info()->status |= TS_POLLING;
 203         /* endless idle loop with no priority at all */
 204         while (1) {
 205                 while (!need_resched()) {
 206                         void (*idle)(void);
 207
 208                         if (__get_cpu_var(cpu_idle_state))
 209                                 __get_cpu_var(cpu_idle_state) = 0;
 210
 211                         rmb();
 212                         idle = pm_idle;
 213                         if (!idle)
 214                                 idle = default_idle;
 215                         if (cpu_is_offline(smp_processor_id()))
 216                                 play_dead();
 217                         /*
 218                          * Idle routines should keep interrupts disabled
 219                          * from here on, until they go to idle.
 220                          * Otherwise, idle callbacks can misfire.
 221                          */
 222                         local_irq_disable();
 223                         enter_idle();
 224                         idle();
 225                         /* In many cases the interrupt that ended idle
 226                            has already called exit_idle. But some idle
 227                            loops can be woken up without interrupt. */
 228                         __exit_idle();
 229                 }
 230
 231                 preempt_enable_no_resched();
 232                 schedule();
 233                 preempt_disable();
 234         }
 235 }
 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         if (!need_resched()) {
 261                 __monitor((void *)&current_thread_info()->flags, 0, 0);
 262                 smp_mb();
 263                 if (!need_resched())
 264                         __sti_mwait(0, 0);
 265                 else
 266                         local_irq_enable();
 267         } else {
 268                 local_irq_enable();
 269         }
 270 }
 271
 272 void __cpuinit select_idle_routine(const struct cpuinfo_x86 *c)
 273 {
 274         static int printed;
 275         if (cpu_has(c, X86_FEATURE_MWAIT)) {
 276                 /*
 277                  * Skip, if setup has overridden idle.
 278                  * One CPU supports mwait => All CPUs supports mwait
 279                  */
 280                 if (!pm_idle) {
 281                         if (!printed) {
 282                                 printk("using mwait in idle threads.\n");
 283                                 printed = 1;
 284                         }
 285                         pm_idle = mwait_idle;
 286                 }
 287         }
 288 }
 289
 290 static int __init idle_setup (char *str)
 291 {
 292         if (!strncmp(str, "poll", 4)) {
 293                 printk("using polling idle threads.\n");
 294                 pm_idle = poll_idle;
 295         }
 296
 297         boot_option_idle_override = 1;
 298         return 1;
 299 }
 300
 301 __setup("idle=", idle_setup);
 302
 303 /* Prints also some state that isn't saved in the pt_regs */
 304 void __show_regs(struct pt_regs * regs)
 305 {
 306         unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L, fs, gs, shadowgs;
 307         unsigned int fsindex,gsindex;
 308         unsigned int ds,cs,es;
 309
 310         printk("\n");
 311         print_modules();
 312         printk("Pid: %d:#%u, comm: %.20s %s %s %.*s\n",
 313                 current->pid, current->xid, current->comm, print_tainted(),
 314                 init_utsname()->release,
 315                 (int)strcspn(init_utsname()->version, " "),
 316                 init_utsname()->version);
 317         printk("RIP: %04lx:[<%016lx>] ", regs->cs & 0xffff, regs->rip);
 318         printk_address(regs->rip);
 319         printk("RSP: %04lx:%016lx  EFLAGS: %08lx\n", regs->ss, regs->rsp,
 320                 regs->eflags);
 321         printk("RAX: %016lx RBX: %016lx RCX: %016lx\n",
 322                regs->rax, regs->rbx, regs->rcx);
 323         printk("RDX: %016lx RSI: %016lx RDI: %016lx\n",
 324                regs->rdx, regs->rsi, regs->rdi);
 325         printk("RBP: %016lx R08: %016lx R09: %016lx\n",
 326                regs->rbp, regs->r8, regs->r9);
 327         printk("R10: %016lx R11: %016lx R12: %016lx\n",
 328                regs->r10, regs->r11, regs->r12);
 329         printk("R13: %016lx R14: %016lx R15: %016lx\n",
 330                regs->r13, regs->r14, regs->r15);
 331
 332         asm("movl %%ds,%0" : "=r" (ds));
 333         asm("movl %%cs,%0" : "=r" (cs));
 334         asm("movl %%es,%0" : "=r" (es));
 335         asm("movl %%fs,%0" : "=r" (fsindex));
 336         asm("movl %%gs,%0" : "=r" (gsindex));
 337
 338         rdmsrl(MSR_FS_BASE, fs);
 339         rdmsrl(MSR_GS_BASE, gs);
 340         rdmsrl(MSR_KERNEL_GS_BASE, shadowgs);
 341
 342         asm("movq %%cr0, %0": "=r" (cr0));
 343         asm("movq %%cr2, %0": "=r" (cr2));
 344         asm("movq %%cr3, %0": "=r" (cr3));
 345         asm("movq %%cr4, %0": "=r" (cr4));
 346
 347         printk("FS:  %016lx(%04x) GS:%016lx(%04x) knlGS:%016lx\n",
 348                fs,fsindex,gs,gsindex,shadowgs);
 349         printk("CS:  %04x DS: %04x ES: %04x CR0: %016lx\n", cs, ds, es, cr0);
 350         printk("CR2: %016lx CR3: %016lx CR4: %016lx\n", cr2, cr3, cr4);
 351 }
 352
 353 void show_regs(struct pt_regs *regs)
 354 {
 355         printk("CPU %d:", smp_processor_id());
 356         __show_regs(regs);
 357         show_trace(NULL, regs, (void *)(regs + 1));
 358 }
 359
 360 /*
 361  * Free current thread data structures etc..
 362  */
 363 void exit_thread(void)
 364 {
 365         struct task_struct *me = current;
 366         struct thread_struct *t = &me->thread;
 367
 368         if (me->thread.io_bitmap_ptr) {
 369                 struct tss_struct *tss = &per_cpu(init_tss, get_cpu());
 370
 371                 kfree(t->io_bitmap_ptr);
 372                 t->io_bitmap_ptr = NULL;
 373                 clear_thread_flag(TIF_IO_BITMAP);
 374                 /*
 375                  * Careful, clear this in the TSS too:
 376                  */
 377                 memset(tss->io_bitmap, 0xff, t->io_bitmap_max);
 378                 t->io_bitmap_max = 0;
 379                 put_cpu();
 380         }
 381 }
 382
 383 void flush_thread(void)
 384 {
 385         struct task_struct *tsk = current;
 386         struct thread_info *t = current_thread_info();
 387
 388         if (t->flags & _TIF_ABI_PENDING) {
 389                 t->flags ^= (_TIF_ABI_PENDING | _TIF_IA32);
 390                 if (t->flags & _TIF_IA32)
 391                         current_thread_info()->status |= TS_COMPAT;
 392         }
 393         t->flags &= ~_TIF_DEBUG;
 394
 395         tsk->thread.debugreg0 = 0;
 396         tsk->thread.debugreg1 = 0;
 397         tsk->thread.debugreg2 = 0;
 398         tsk->thread.debugreg3 = 0;
 399         tsk->thread.debugreg6 = 0;
 400         tsk->thread.debugreg7 = 0;
 401         memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
 402         /*
 403          * Forget coprocessor state..
 404          */
 405         clear_fpu(tsk);
 406         clear_used_math();
 407 }
 408
 409 void release_thread(struct task_struct *dead_task)
 410 {
 411         if (dead_task->mm) {
 412                 if (dead_task->mm->context.size) {
 413                         printk("WARNING: dead process %8s still has LDT? <%p/%d>\n",
 414                                         dead_task->comm,
 415                                         dead_task->mm->context.ldt,
 416                                         dead_task->mm->context.size);
 417                         BUG();
 418                 }
 419         }
 420 }
 421
 422 static inline void set_32bit_tls(struct task_struct *t, int tls, u32 addr)
 423 {
 424         struct user_desc ud = {
 425                 .base_addr = addr,
 426                 .limit = 0xfffff,
 427                 .seg_32bit = 1,
 428                 .limit_in_pages = 1,
 429                 .useable = 1,
 430         };
 431         struct n_desc_struct *desc = (void *)t->thread.tls_array;
 432         desc += tls;
 433         desc->a = LDT_entry_a(&ud);
 434         desc->b = LDT_entry_b(&ud);
 435 }
 436
 437 static inline u32 read_32bit_tls(struct task_struct *t, int tls)
 438 {
 439         struct desc_struct *desc = (void *)t->thread.tls_array;
 440         desc += tls;
 441         return desc->base0 |
 442                 (((u32)desc->base1) << 16) |
 443                 (((u32)desc->base2) << 24);
 444 }
 445
 446 /*
 447  * This gets called before we allocate a new thread and copy
 448  * the current task into it.
 449  */
 450 void prepare_to_copy(struct task_struct *tsk)
 451 {
 452         unlazy_fpu(tsk);
 453 }
 454
 455 int copy_thread(int nr, unsigned long clone_flags, unsigned long rsp,
 456                 unsigned long unused,
 457         struct task_struct * p, struct pt_regs * regs)
 458 {
 459         int err;
 460         struct pt_regs * childregs;
 461         struct task_struct *me = current;
 462
 463         childregs = ((struct pt_regs *)
 464                         (THREAD_SIZE + task_stack_page(p))) - 1;
 465         *childregs = *regs;
 466
 467         childregs->rax = 0;
 468         childregs->rsp = rsp;
 469         if (rsp == ~0UL)
 470                 childregs->rsp = (unsigned long)childregs;
 471
 472         p->thread.rsp = (unsigned long) childregs;
 473         p->thread.rsp0 = (unsigned long) (childregs+1);
 474         p->thread.userrsp = me->thread.userrsp;
 475
 476         set_tsk_thread_flag(p, TIF_FORK);
 477
 478         p->thread.fs = me->thread.fs;
 479         p->thread.gs = me->thread.gs;
 480
 481         asm("mov %%gs,%0" : "=m" (p->thread.gsindex));
 482         asm("mov %%fs,%0" : "=m" (p->thread.fsindex));
 483         asm("mov %%es,%0" : "=m" (p->thread.es));
 484         asm("mov %%ds,%0" : "=m" (p->thread.ds));
 485
 486         if (unlikely(test_tsk_thread_flag(me, TIF_IO_BITMAP))) {
 487                 p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL);
 488                 if (!p->thread.io_bitmap_ptr) {
 489                         p->thread.io_bitmap_max = 0;
 490                         return -ENOMEM;
 491                 }
 492                 memcpy(p->thread.io_bitmap_ptr, me->thread.io_bitmap_ptr,
 493                                 IO_BITMAP_BYTES);
 494                 set_tsk_thread_flag(p, TIF_IO_BITMAP);
 495         }
 496
 497         /*
 498          * Set a new TLS for the child thread?
 499          */
 500         if (clone_flags & CLONE_SETTLS) {
 501 #ifdef CONFIG_IA32_EMULATION
 502                 if (test_thread_flag(TIF_IA32))
 503                         err = ia32_child_tls(p, childregs);
 504                 else
 505 #endif
 506                         err = do_arch_prctl(p, ARCH_SET_FS, childregs->r8);
 507                 if (err)
 508                         goto out;
 509         }
 510         err = 0;
 511 out:
 512         if (err && p->thread.io_bitmap_ptr) {
 513                 kfree(p->thread.io_bitmap_ptr);
 514                 p->thread.io_bitmap_max = 0;
 515         }
 516         return err;
 517 }
 518
 519 /*
 520  * This special macro can be used to load a debugging register
 521  */
 522 #define loaddebug(thread,r) set_debugreg(thread->debugreg ## r, r)
 523
 524 static inline void __switch_to_xtra(struct task_struct *prev_p,
 525                                     struct task_struct *next_p,
 526                                     struct tss_struct *tss)
 527 {
 528         struct thread_struct *prev, *next;
 529
 530         prev = &prev_p->thread,
 531         next = &next_p->thread;
 532
 533         if (test_tsk_thread_flag(next_p, TIF_DEBUG)) {
 534                 loaddebug(next, 0);
 535                 loaddebug(next, 1);
 536                 loaddebug(next, 2);
 537                 loaddebug(next, 3);
 538                 /* no 4 and 5 */
 539                 loaddebug(next, 6);
 540                 loaddebug(next, 7);
 541         }
 542
 543         if (test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
 544                 /*
 545                  * Copy the relevant range of the IO bitmap.
 546                  * Normally this is 128 bytes or less:
 547                  */
 548                 memcpy(tss->io_bitmap, next->io_bitmap_ptr,
 549                        max(prev->io_bitmap_max, next->io_bitmap_max));
 550         } else if (test_tsk_thread_flag(prev_p, TIF_IO_BITMAP)) {
 551                 /*
 552                  * Clear any possible leftover bits:
 553                  */
 554                 memset(tss->io_bitmap, 0xff, prev->io_bitmap_max);
 555         }
 556 }
 557
 558 /*
 559  *      switch_to(x,y) should switch tasks from x to y.
 560  *
 561  * This could still be optimized:
 562  * - fold all the options into a flag word and test it with a single test.
 563  * - could test fs/gs bitsliced
 564  *
 565  * Kprobes not supported here. Set the probe on schedule instead.
 566  */
 567 __kprobes struct task_struct *
 568 __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
 569 {
 570         struct thread_struct *prev = &prev_p->thread,
 571                                  *next = &next_p->thread;
 572         int cpu = smp_processor_id();
 573         struct tss_struct *tss = &per_cpu(init_tss, cpu);
 574
 575         /* we're going to use this soon, after a few expensive things */
 576         if (next_p->fpu_counter>5)
 577                 prefetch(&next->i387.fxsave);
 578
 579         /*
 580          * Reload esp0, LDT and the page table pointer:
 581          */
 582         tss->rsp0 = next->rsp0;
 583
 584         /*
 585          * Switch DS and ES.
 586          * This won't pick up thread selector changes, but I guess that is ok.
 587          */
 588         asm volatile("mov %%es,%0" : "=m" (prev->es));
 589         if (unlikely(next->es | prev->es))
 590                 loadsegment(es, next->es);
 591
 592         asm volatile ("mov %%ds,%0" : "=m" (prev->ds));
 593         if (unlikely(next->ds | prev->ds))
 594                 loadsegment(ds, next->ds);
 595
 596         load_TLS(next, cpu);
 597
 598         /*
 599          * Switch FS and GS.
 600          */
 601         {
 602                 unsigned fsindex;
 603                 asm volatile("movl %%fs,%0" : "=r" (fsindex));
 604                 /* segment register != 0 always requires a reload.
 605                    also reload when it has changed.
 606                    when prev process used 64bit base always reload
 607                    to avoid an information leak. */
 608                 if (unlikely(fsindex | next->fsindex | prev->fs)) {
 609                         loadsegment(fs, next->fsindex);
 610                         /* check if the user used a selector != 0
 611                          * if yes clear 64bit base, since overloaded base
 612                          * is always mapped to the Null selector
 613                          */
 614                         if (fsindex)
 615                         prev->fs = 0;
 616                 }
 617                 /* when next process has a 64bit base use it */
 618                 if (next->fs)
 619                         wrmsrl(MSR_FS_BASE, next->fs);
 620                 prev->fsindex = fsindex;
 621         }
 622         {
 623                 unsigned gsindex;
 624                 asm volatile("movl %%gs,%0" : "=r" (gsindex));
 625                 if (unlikely(gsindex | next->gsindex | prev->gs)) {
 626                         load_gs_index(next->gsindex);
 627                         if (gsindex)
 628                         prev->gs = 0;
 629                 }
 630                 if (next->gs)
 631                         wrmsrl(MSR_KERNEL_GS_BASE, next->gs);
 632                 prev->gsindex = gsindex;
 633         }
 634
 635         /* Must be after DS reload */
 636         unlazy_fpu(prev_p);
 637
 638         /*
 639          * Switch the PDA and FPU contexts.
 640          */
 641         prev->userrsp = read_pda(oldrsp);
 642         write_pda(oldrsp, next->userrsp);
 643         write_pda(pcurrent, next_p);
 644
 645         write_pda(kernelstack,
 646         (unsigned long)task_stack_page(next_p) + THREAD_SIZE - PDA_STACKOFFSET);
 647 #ifdef CONFIG_CC_STACKPROTECTOR
 648         write_pda(stack_canary, next_p->stack_canary);
 649         /*
 650          * Build time only check to make sure the stack_canary is at
 651          * offset 40 in the pda; this is a gcc ABI requirement
 652          */
 653         BUILD_BUG_ON(offsetof(struct x8664_pda, stack_canary) != 40);
 654 #endif
 655
 656         /*
 657          * Now maybe reload the debug registers and handle I/O bitmaps
 658          */
 659         if (unlikely((task_thread_info(next_p)->flags & _TIF_WORK_CTXSW))
 660             || test_tsk_thread_flag(prev_p, TIF_IO_BITMAP))
 661                 __switch_to_xtra(prev_p, next_p, tss);
 662
 663         /* If the task has used fpu the last 5 timeslices, just do a full
 664          * restore of the math state immediately to avoid the trap; the
 665          * chances of needing FPU soon are obviously high now
 666          */
 667         if (next_p->fpu_counter>5)
 668                 math_state_restore();
 669         return prev_p;
 670 }
 671
 672 /*
 673  * sys_execve() executes a new program.
 674  */
 675 asmlinkage
 676 long sys_execve(char __user *name, char __user * __user *argv,
 677                 char __user * __user *envp, struct pt_regs regs)
 678 {
 679         long error;
 680         char * filename;
 681
 682         filename = getname(name);
 683         error = PTR_ERR(filename);
 684         if (IS_ERR(filename))
 685                 return error;
 686         error = do_execve(filename, argv, envp, &regs);
 687         putname(filename);
 688         return error;
 689 }
 690
 691 void set_personality_64bit(void)
 692 {
 693         /* inherit personality from parent */
 694
 695         /* Make sure to be in 64bit mode */
 696         clear_thread_flag(TIF_IA32);
 697 }
 698
 699 asmlinkage long sys_fork(struct pt_regs *regs)
 700 {
 701         return do_fork(SIGCHLD, regs->rsp, regs, 0, NULL, NULL);
 702 }
 703
 704 asmlinkage long
 705 sys_clone(unsigned long clone_flags, unsigned long newsp,
 706           void __user *parent_tid, void __user *child_tid, struct pt_regs *regs)
 707 {
 708         if (!newsp)
 709                 newsp = regs->rsp;
 710         return do_fork(clone_flags, newsp, regs, 0, parent_tid, child_tid);
 711 }
 712
 713 /*
 714  * This is trivial, and on the face of it looks like it
 715  * could equally well be done in user mode.
 716  *
 717  * Not so, for quite unobvious reasons - register pressure.
 718  * In user mode vfork() cannot have a stack frame, and if
 719  * done by calling the "clone()" system call directly, you
 720  * do not have enough call-clobbered registers to hold all
 721  * the information you need.
 722  */
 723 asmlinkage long sys_vfork(struct pt_regs *regs)
 724 {
 725         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->rsp, regs, 0,
 726                     NULL, NULL);
 727 }
 728
 729 unsigned long get_wchan(struct task_struct *p)
 730 {
 731         unsigned long stack;
 732         u64 fp,rip;
 733         int count = 0;
 734
 735         if (!p || p == current || p->state==TASK_RUNNING)
 736                 return 0;
 737         stack = (unsigned long)task_stack_page(p);
 738         if (p->thread.rsp < stack || p->thread.rsp > stack+THREAD_SIZE)
 739                 return 0;
 740         fp = *(u64 *)(p->thread.rsp);
 741         do {
 742                 if (fp < (unsigned long)stack ||
 743                     fp > (unsigned long)stack+THREAD_SIZE)
 744                         return 0;
 745                 rip = *(u64 *)(fp+8);
 746                 if (!in_sched_functions(rip))
 747                         return rip;
 748                 fp = *(u64 *)fp;
 749         } while (count++ < 16);
 750         return 0;
 751 }
 752
 753 long do_arch_prctl(struct task_struct *task, int code, unsigned long addr)
 754 {
 755         int ret = 0;
 756         int doit = task == current;
 757         int cpu;
 758
 759         switch (code) {
 760         case ARCH_SET_GS:
 761                 if (addr >= TASK_SIZE_OF(task))
 762                         return -EPERM;
 763                 cpu = get_cpu();
 764                 /* handle small bases via the GDT because that's faster to
 765                    switch. */
 766                 if (addr <= 0xffffffff) {
 767                         set_32bit_tls(task, GS_TLS, addr);
 768                         if (doit) {
 769                                 load_TLS(&task->thread, cpu);
 770                                 load_gs_index(GS_TLS_SEL);
 771                         }
 772                         task->thread.gsindex = GS_TLS_SEL;
 773                         task->thread.gs = 0;
 774                 } else {
 775                         task->thread.gsindex = 0;
 776                         task->thread.gs = addr;
 777                         if (doit) {
 778                                 load_gs_index(0);
 779                                 ret = checking_wrmsrl(MSR_KERNEL_GS_BASE, addr);
 780                         }
 781                 }
 782                 put_cpu();
 783                 break;
 784         case ARCH_SET_FS:
 785                 /* Not strictly needed for fs, but do it for symmetry
 786                    with gs */
 787                 if (addr >= TASK_SIZE_OF(task))
 788                         return -EPERM;
 789                 cpu = get_cpu();
 790                 /* handle small bases via the GDT because that's faster to
 791                    switch. */
 792                 if (addr <= 0xffffffff) {
 793                         set_32bit_tls(task, FS_TLS, addr);
 794                         if (doit) {
 795                                 load_TLS(&task->thread, cpu);
 796                                 asm volatile("movl %0,%%fs" :: "r"(FS_TLS_SEL));
 797                         }
 798                         task->thread.fsindex = FS_TLS_SEL;
 799                         task->thread.fs = 0;
 800                 } else {
 801                         task->thread.fsindex = 0;
 802                         task->thread.fs = addr;
 803                         if (doit) {
 804                                 /* set the selector to 0 to not confuse
 805                                    __switch_to */
 806                                 asm volatile("movl %0,%%fs" :: "r" (0));
 807                                 ret = checking_wrmsrl(MSR_FS_BASE, addr);
 808                         }
 809                 }
 810                 put_cpu();
 811                 break;
 812         case ARCH_GET_FS: {
 813                 unsigned long base;
 814                 if (task->thread.fsindex == FS_TLS_SEL)
 815                         base = read_32bit_tls(task, FS_TLS);
 816                 else if (doit)
 817                         rdmsrl(MSR_FS_BASE, base);
 818                 else
 819                         base = task->thread.fs;
 820                 ret = put_user(base, (unsigned long __user *)addr);
 821                 break;
 822         }
 823         case ARCH_GET_GS: {
 824                 unsigned long base;
 825                 unsigned gsindex;
 826                 if (task->thread.gsindex == GS_TLS_SEL)
 827                         base = read_32bit_tls(task, GS_TLS);
 828                 else if (doit) {
 829                         asm("movl %%gs,%0" : "=r" (gsindex));
 830                         if (gsindex)
 831                                 rdmsrl(MSR_KERNEL_GS_BASE, base);
 832                         else
 833                                 base = task->thread.gs;
 834                 }
 835                 else
 836                         base = task->thread.gs;
 837                 ret = put_user(base, (unsigned long __user *)addr);
 838                 break;
 839         }
 840
 841         default:
 842                 ret = -EINVAL;
 843                 break;
 844         }
 845
 846         return ret;
 847 }
 848
 849 long sys_arch_prctl(int code, unsigned long addr)
 850 {
 851         return do_arch_prctl(current, code, addr);
 852 }
 853
 854 /*
 855  * Capture the user space registers if the task is not running (in user space)
 856  */
 857 int dump_task_regs(struct task_struct *tsk, elf_gregset_t *regs)
 858 {
 859         struct pt_regs *pp, ptregs;
 860
 861         pp = task_pt_regs(tsk);
 862
 863         ptregs = *pp;
 864         ptregs.cs &= 0xffff;
 865         ptregs.ss &= 0xffff;
 866
 867         elf_core_copy_regs(regs, &ptregs);
 868
 869         return 1;
 870 }
 871
 872 unsigned long arch_align_stack(unsigned long sp)
 873 {
 874         if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
 875                 sp -= get_random_int() % 8192;
 876         return sp & ~0xf;
 877 }