arch/x86/kernel/process_64.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  *  X86-64 port
   8  *      Andi Kleen.
   9  *
  10  *      CPU hotplug support - ashok.raj@intel.com
  11  */
  12
  13 /*
  14  * This file handles the architecture-dependent parts of process handling..
  15  */
  16
  17 #include <stdarg.h>
  18
  19 #include <linux/cpu.h>
  20 #include <linux/errno.h>
  21 #include <linux/sched.h>
  22 #include <linux/fs.h>
  23 #include <linux/kernel.h>
  24 #include <linux/mm.h>
  25 #include <linux/elfcore.h>
  26 #include <linux/smp.h>
  27 #include <linux/slab.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/module.h>
  34 #include <linux/ptrace.h>
  35 #include <linux/random.h>
  36 #include <linux/notifier.h>
  37 #include <linux/kprobes.h>
  38 #include <linux/kdebug.h>
  39 #include <linux/tick.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/processor.h>
  46 #include <asm/i387.h>
  47 #include <asm/mmu_context.h>
  48 #include <asm/pda.h>
  49 #include <asm/prctl.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 = CLONE_VM | CLONE_UNTRACED;
  58
  59 unsigned long boot_option_idle_override = 0;
  60 EXPORT_SYMBOL(boot_option_idle_override);
  61
  62 /*
  63  * Powermanagement idle function, if any..
  64  */
  65 void (*pm_idle)(void);
  66 EXPORT_SYMBOL(pm_idle);
  67 static DEFINE_PER_CPU(unsigned int, cpu_idle_state);
  68
  69 static ATOMIC_NOTIFIER_HEAD(idle_notifier);
  70
  71 void idle_notifier_register(struct notifier_block *n)
  72 {
  73         atomic_notifier_chain_register(&idle_notifier, n);
  74 }
  75
  76 void enter_idle(void)
  77 {
  78         write_pda(isidle, 1);
  79         atomic_notifier_call_chain(&idle_notifier, IDLE_START, NULL);
  80 }
  81
  82 static void __exit_idle(void)
  83 {
  84         if (test_and_clear_bit_pda(0, isidle) == 0)
  85                 return;
  86         atomic_notifier_call_chain(&idle_notifier, IDLE_END, NULL);
  87 }
  88
  89 /* Called from interrupts to signify idle end */
  90 void exit_idle(void)
  91 {
  92         /* idle loop has pid 0 */
  93         if (current->pid)
  94                 return;
  95         __exit_idle();
  96 }
  97
  98 /*
  99  * We use this if we don't have any better
 100  * idle routine..
 101  */
 102 void default_idle(void)
 103 {
 104         current_thread_info()->status &= ~TS_POLLING;
 105         /*
 106          * TS_POLLING-cleared state must be visible before we
 107          * test NEED_RESCHED:
 108          */
 109         smp_mb();
 110         local_irq_disable();
 111         if (!need_resched()) {
 112                 ktime_t t0, t1;
 113                 u64 t0n, t1n;
 114
 115                 t0 = ktime_get();
 116                 t0n = ktime_to_ns(t0);
 117                 safe_halt();    /* enables interrupts racelessly */
 118                 local_irq_disable();
 119                 t1 = ktime_get();
 120                 t1n = ktime_to_ns(t1);
 121                 sched_clock_idle_wakeup_event(t1n - t0n);
 122         }
 123         local_irq_enable();
 124         current_thread_info()->status |= TS_POLLING;
 125 }
 126
 127 /*
 128  * On SMP it's slightly faster (but much more power-consuming!)
 129  * to poll the ->need_resched flag instead of waiting for the
 130  * cross-CPU IPI to arrive. Use this option with caution.
 131  */
 132 static void poll_idle(void)
 133 {
 134         local_irq_enable();
 135         cpu_relax();
 136 }
 137
 138 #ifdef CONFIG_HOTPLUG_CPU
 139 DECLARE_PER_CPU(int, cpu_state);
 140
 141 #include <asm/nmi.h>
 142 /* We halt the CPU with physical CPU hotplug */
 143 static inline void play_dead(void)
 144 {
 145         idle_task_exit();
 146         wbinvd();
 147         mb();
 148         /* Ack it */
 149         __get_cpu_var(cpu_state) = CPU_DEAD;
 150
 151         local_irq_disable();
 152         while (1)
 153                 halt();
 154 }
 155 #else
 156 static inline void play_dead(void)
 157 {
 158         BUG();
 159 }
 160 #endif /* CONFIG_HOTPLUG_CPU */
 161
 162 /*
 163  * The idle thread. There's no useful work to be
 164  * done, so just try to conserve power and have a
 165  * low exit latency (ie sit in a loop waiting for
 166  * somebody to say that they'd like to reschedule)
 167  */
 168 void cpu_idle(void)
 169 {
 170         current_thread_info()->status |= TS_POLLING;
 171         /* endless idle loop with no priority at all */
 172         while (1) {
 173                 while (!need_resched()) {
 174                         void (*idle)(void);
 175
 176                         if (__get_cpu_var(cpu_idle_state))
 177                                 __get_cpu_var(cpu_idle_state) = 0;
 178
 179                         tick_nohz_stop_sched_tick();
 180
 181                         rmb();
 182                         idle = pm_idle;
 183                         if (!idle)
 184                                 idle = default_idle;
 185                         if (cpu_is_offline(smp_processor_id()))
 186                                 play_dead();
 187                         /*
 188                          * Idle routines should keep interrupts disabled
 189                          * from here on, until they go to idle.
 190                          * Otherwise, idle callbacks can misfire.
 191                          */
 192                         local_irq_disable();
 193                         enter_idle();
 194                         idle();
 195                         /* In many cases the interrupt that ended idle
 196                            has already called exit_idle. But some idle
 197                            loops can be woken up without interrupt. */
 198                         __exit_idle();
 199                 }
 200
 201                 tick_nohz_restart_sched_tick();
 202                 preempt_enable_no_resched();
 203                 schedule();
 204                 preempt_disable();
 205         }
 206 }
 207
 208 static void do_nothing(void *unused)
 209 {
 210 }
 211
 212 void cpu_idle_wait(void)
 213 {
 214         unsigned int cpu, this_cpu = get_cpu();
 215         cpumask_t map, tmp = current->cpus_allowed;
 216
 217         set_cpus_allowed(current, cpumask_of_cpu(this_cpu));
 218         put_cpu();
 219
 220         cpus_clear(map);
 221         for_each_online_cpu(cpu) {
 222                 per_cpu(cpu_idle_state, cpu) = 1;
 223                 cpu_set(cpu, map);
 224         }
 225
 226         __get_cpu_var(cpu_idle_state) = 0;
 227
 228         wmb();
 229         do {
 230                 ssleep(1);
 231                 for_each_online_cpu(cpu) {
 232                         if (cpu_isset(cpu, map) && !per_cpu(cpu_idle_state, cpu))
 233                                 cpu_clear(cpu, map);
 234                 }
 235                 cpus_and(map, map, cpu_online_map);
 236                 /*
 237                  * We waited 1 sec, if a CPU still did not call idle
 238                  * it may be because it is in idle and not waking up
 239                  * because it has nothing to do.
 240                  * Give all the remaining CPUS a kick.
 241                  */
 242                 smp_call_function_mask(map, do_nothing, 0, 0);
 243         } while (!cpus_empty(map));
 244
 245         set_cpus_allowed(current, tmp);
 246 }
 247 EXPORT_SYMBOL_GPL(cpu_idle_wait);
 248
 249 /*
 250  * This uses new MONITOR/MWAIT instructions on P4 processors with PNI,
 251  * which can obviate IPI to trigger checking of need_resched.
 252  * We execute MONITOR against need_resched and enter optimized wait state
 253  * through MWAIT. Whenever someone changes need_resched, we would be woken
 254  * up from MWAIT (without an IPI).
 255  *
 256  * New with Core Duo processors, MWAIT can take some hints based on CPU
 257  * capability.
 258  */
 259 void mwait_idle_with_hints(unsigned long ax, unsigned long cx)
 260 {
 261         if (!need_resched()) {
 262                 __monitor((void *)&current_thread_info()->flags, 0, 0);
 263                 smp_mb();
 264                 if (!need_resched())
 265                         __mwait(ax, cx);
 266         }
 267 }
 268
 269 /* Default MONITOR/MWAIT with no hints, used for default C1 state */
 270 static void mwait_idle(void)
 271 {
 272         if (!need_resched()) {
 273                 __monitor((void *)&current_thread_info()->flags, 0, 0);
 274                 smp_mb();
 275                 if (!need_resched())
 276                         __sti_mwait(0, 0);
 277                 else
 278                         local_irq_enable();
 279         } else {
 280                 local_irq_enable();
 281         }
 282 }
 283
 284 void __cpuinit select_idle_routine(const struct cpuinfo_x86 *c)
 285 {
 286         static int printed;
 287         if (cpu_has(c, X86_FEATURE_MWAIT)) {
 288                 /*
 289                  * Skip, if setup has overridden idle.
 290                  * One CPU supports mwait => All CPUs supports mwait
 291                  */
 292                 if (!pm_idle) {
 293                         if (!printed) {
 294                                 printk(KERN_INFO "using mwait in idle threads.\n");
 295                                 printed = 1;
 296                         }
 297                         pm_idle = mwait_idle;
 298                 }
 299         }
 300 }
 301
 302 static int __init idle_setup(char *str)
 303 {
 304         if (!strcmp(str, "poll")) {
 305                 printk("using polling idle threads.\n");
 306                 pm_idle = poll_idle;
 307         } else if (!strcmp(str, "mwait"))
 308                 force_mwait = 1;
 309         else
 310                 return -1;
 311
 312         boot_option_idle_override = 1;
 313         return 0;
 314 }
 315 early_param("idle", idle_setup);
 316
 317 /* Prints also some state that isn't saved in the pt_regs */
 318 void __show_regs(struct pt_regs * regs)
 319 {
 320         unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L, fs, gs, shadowgs;
 321         unsigned long d0, d1, d2, d3, d6, d7;
 322         unsigned int fsindex, gsindex;
 323         unsigned int ds, cs, es;
 324
 325         printk("\n");
 326         print_modules();
 327         printk("Pid: %d, comm: %.20s %s %s %.*s\n",
 328                 current->pid, current->comm, print_tainted(),
 329                 init_utsname()->release,
 330                 (int)strcspn(init_utsname()->version, " "),
 331                 init_utsname()->version);
 332         printk("RIP: %04lx:[<%016lx>] ", regs->cs & 0xffff, regs->ip);
 333         printk_address(regs->ip);
 334         printk("RSP: %04lx:%016lx  EFLAGS: %08lx\n", regs->ss, regs->sp,
 335                 regs->flags);
 336         printk("RAX: %016lx RBX: %016lx RCX: %016lx\n",
 337                regs->ax, regs->bx, regs->cx);
 338         printk("RDX: %016lx RSI: %016lx RDI: %016lx\n",
 339                regs->dx, regs->si, regs->di);
 340         printk("RBP: %016lx R08: %016lx R09: %016lx\n",
 341                regs->bp, regs->r8, regs->r9);
 342         printk("R10: %016lx R11: %016lx R12: %016lx\n",
 343                regs->r10, regs->r11, regs->r12);
 344         printk("R13: %016lx R14: %016lx R15: %016lx\n",
 345                regs->r13, regs->r14, regs->r15);
 346
 347         asm("movl %%ds,%0" : "=r" (ds));
 348         asm("movl %%cs,%0" : "=r" (cs));
 349         asm("movl %%es,%0" : "=r" (es));
 350         asm("movl %%fs,%0" : "=r" (fsindex));
 351         asm("movl %%gs,%0" : "=r" (gsindex));
 352
 353         rdmsrl(MSR_FS_BASE, fs);
 354         rdmsrl(MSR_GS_BASE, gs);
 355         rdmsrl(MSR_KERNEL_GS_BASE, shadowgs);
 356
 357         cr0 = read_cr0();
 358         cr2 = read_cr2();
 359         cr3 = read_cr3();
 360         cr4 = read_cr4();
 361
 362         printk("FS:  %016lx(%04x) GS:%016lx(%04x) knlGS:%016lx\n",
 363                fs,fsindex,gs,gsindex,shadowgs);
 364         printk("CS:  %04x DS: %04x ES: %04x CR0: %016lx\n", cs, ds, es, cr0);
 365         printk("CR2: %016lx CR3: %016lx CR4: %016lx\n", cr2, cr3, cr4);
 366
 367         get_debugreg(d0, 0);
 368         get_debugreg(d1, 1);
 369         get_debugreg(d2, 2);
 370         printk("DR0: %016lx DR1: %016lx DR2: %016lx\n", d0, d1, d2);
 371         get_debugreg(d3, 3);
 372         get_debugreg(d6, 6);
 373         get_debugreg(d7, 7);
 374         printk("DR3: %016lx DR6: %016lx DR7: %016lx\n", d3, d6, d7);
 375 }
 376
 377 void show_regs(struct pt_regs *regs)
 378 {
 379         printk("CPU %d:", smp_processor_id());
 380         __show_regs(regs);
 381         show_trace(NULL, regs, (void *)(regs + 1));
 382 }
 383
 384 /*
 385  * Free current thread data structures etc..
 386  */
 387 void exit_thread(void)
 388 {
 389         struct task_struct *me = current;
 390         struct thread_struct *t = &me->thread;
 391
 392         if (me->thread.io_bitmap_ptr) {
 393                 struct tss_struct *tss = &per_cpu(init_tss, get_cpu());
 394
 395                 kfree(t->io_bitmap_ptr);
 396                 t->io_bitmap_ptr = NULL;
 397                 clear_thread_flag(TIF_IO_BITMAP);
 398                 /*
 399                  * Careful, clear this in the TSS too:
 400                  */
 401                 memset(tss->io_bitmap, 0xff, t->io_bitmap_max);
 402                 t->io_bitmap_max = 0;
 403                 put_cpu();
 404         }
 405 }
 406
 407 void flush_thread(void)
 408 {
 409         struct task_struct *tsk = current;
 410
 411         if (test_tsk_thread_flag(tsk, TIF_ABI_PENDING)) {
 412                 clear_tsk_thread_flag(tsk, TIF_ABI_PENDING);
 413                 if (test_tsk_thread_flag(tsk, TIF_IA32)) {
 414                         clear_tsk_thread_flag(tsk, TIF_IA32);
 415                 } else {
 416                         set_tsk_thread_flag(tsk, TIF_IA32);
 417                         current_thread_info()->status |= TS_COMPAT;
 418                 }
 419         }
 420         clear_tsk_thread_flag(tsk, TIF_DEBUG);
 421
 422         tsk->thread.debugreg0 = 0;
 423         tsk->thread.debugreg1 = 0;
 424         tsk->thread.debugreg2 = 0;
 425         tsk->thread.debugreg3 = 0;
 426         tsk->thread.debugreg6 = 0;
 427         tsk->thread.debugreg7 = 0;
 428         memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
 429         /*
 430          * Forget coprocessor state..
 431          */
 432         clear_fpu(tsk);
 433         clear_used_math();
 434 }
 435
 436 void release_thread(struct task_struct *dead_task)
 437 {
 438         if (dead_task->mm) {
 439                 if (dead_task->mm->context.size) {
 440                         printk("WARNING: dead process %8s still has LDT? <%p/%d>\n",
 441                                         dead_task->comm,
 442                                         dead_task->mm->context.ldt,
 443                                         dead_task->mm->context.size);
 444                         BUG();
 445                 }
 446         }
 447 }
 448
 449 static inline void set_32bit_tls(struct task_struct *t, int tls, u32 addr)
 450 {
 451         struct user_desc ud = {
 452                 .base_addr = addr,
 453                 .limit = 0xfffff,
 454                 .seg_32bit = 1,
 455                 .limit_in_pages = 1,
 456                 .useable = 1,
 457         };
 458         struct desc_struct *desc = (void *)t->thread.tls_array;
 459         desc += tls;
 460         fill_ldt(desc, &ud);
 461 }
 462
 463 static inline u32 read_32bit_tls(struct task_struct *t, int tls)
 464 {
 465         return get_desc_base(&t->thread.tls_array[tls]);
 466 }
 467
 468 /*
 469  * This gets called before we allocate a new thread and copy
 470  * the current task into it.
 471  */
 472 void prepare_to_copy(struct task_struct *tsk)
 473 {
 474         unlazy_fpu(tsk);
 475 }
 476
 477 int copy_thread(int nr, unsigned long clone_flags, unsigned long sp,
 478                 unsigned long unused,
 479         struct task_struct * p, struct pt_regs * regs)
 480 {
 481         int err;
 482         struct pt_regs * childregs;
 483         struct task_struct *me = current;
 484
 485         childregs = ((struct pt_regs *)
 486                         (THREAD_SIZE + task_stack_page(p))) - 1;
 487         *childregs = *regs;
 488
 489         childregs->ax = 0;
 490         childregs->sp = sp;
 491         if (sp == ~0UL)
 492                 childregs->sp = (unsigned long)childregs;
 493
 494         p->thread.sp = (unsigned long) childregs;
 495         p->thread.sp0 = (unsigned long) (childregs+1);
 496         p->thread.usersp = me->thread.usersp;
 497
 498         set_tsk_thread_flag(p, TIF_FORK);
 499
 500         p->thread.fs = me->thread.fs;
 501         p->thread.gs = me->thread.gs;
 502
 503         asm("mov %%gs,%0" : "=m" (p->thread.gsindex));
 504         asm("mov %%fs,%0" : "=m" (p->thread.fsindex));
 505         asm("mov %%es,%0" : "=m" (p->thread.es));
 506         asm("mov %%ds,%0" : "=m" (p->thread.ds));
 507
 508         if (unlikely(test_tsk_thread_flag(me, TIF_IO_BITMAP))) {
 509                 p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL);
 510                 if (!p->thread.io_bitmap_ptr) {
 511                         p->thread.io_bitmap_max = 0;
 512                         return -ENOMEM;
 513                 }
 514                 memcpy(p->thread.io_bitmap_ptr, me->thread.io_bitmap_ptr,
 515                                 IO_BITMAP_BYTES);
 516                 set_tsk_thread_flag(p, TIF_IO_BITMAP);
 517         }
 518
 519         /*
 520          * Set a new TLS for the child thread?
 521          */
 522         if (clone_flags & CLONE_SETTLS) {
 523 #ifdef CONFIG_IA32_EMULATION
 524                 if (test_thread_flag(TIF_IA32))
 525                         err = do_set_thread_area(p, -1,
 526                                 (struct user_desc __user *)childregs->si, 0);
 527                 else
 528 #endif
 529                         err = do_arch_prctl(p, ARCH_SET_FS, childregs->r8);
 530                 if (err)
 531                         goto out;
 532         }
 533         err = 0;
 534 out:
 535         if (err && p->thread.io_bitmap_ptr) {
 536                 kfree(p->thread.io_bitmap_ptr);
 537                 p->thread.io_bitmap_max = 0;
 538         }
 539         return err;
 540 }
 541
 542 /*
 543  * This special macro can be used to load a debugging register
 544  */
 545 #define loaddebug(thread, r) set_debugreg(thread->debugreg ## r, r)
 546
 547 /*
 548  * Capture the user space registers if the task is not running (in user space)
 549  */
 550 int dump_task_regs(struct task_struct *tsk, elf_gregset_t *regs)
 551 {
 552         struct pt_regs *pp, ptregs;
 553
 554         pp = task_pt_regs(tsk);
 555
 556         ptregs = *pp;
 557         ptregs.cs &= 0xffff;
 558         ptregs.ss &= 0xffff;
 559
 560         elf_core_copy_regs(regs, &ptregs);
 561
 562         return 1;
 563 }
 564
 565 static inline void __switch_to_xtra(struct task_struct *prev_p,
 566                                     struct task_struct *next_p,
 567                                     struct tss_struct *tss)
 568 {
 569         struct thread_struct *prev, *next;
 570         unsigned long debugctl;
 571
 572         prev = &prev_p->thread,
 573         next = &next_p->thread;
 574
 575         debugctl = prev->debugctlmsr;
 576         if (next->ds_area_msr != prev->ds_area_msr) {
 577                 /* we clear debugctl to make sure DS
 578                  * is not in use when we change it */
 579                 debugctl = 0;
 580                 wrmsrl(MSR_IA32_DEBUGCTLMSR, 0);
 581                 wrmsrl(MSR_IA32_DS_AREA, next->ds_area_msr);
 582         }
 583
 584         if (next->debugctlmsr != debugctl)
 585                 wrmsrl(MSR_IA32_DEBUGCTLMSR, next->debugctlmsr);
 586
 587         if (test_tsk_thread_flag(next_p, TIF_DEBUG)) {
 588                 loaddebug(next, 0);
 589                 loaddebug(next, 1);
 590                 loaddebug(next, 2);
 591                 loaddebug(next, 3);
 592                 /* no 4 and 5 */
 593                 loaddebug(next, 6);
 594                 loaddebug(next, 7);
 595         }
 596
 597         if (test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
 598                 /*
 599                  * Copy the relevant range of the IO bitmap.
 600                  * Normally this is 128 bytes or less:
 601                  */
 602                 memcpy(tss->io_bitmap, next->io_bitmap_ptr,
 603                        max(prev->io_bitmap_max, next->io_bitmap_max));
 604         } else if (test_tsk_thread_flag(prev_p, TIF_IO_BITMAP)) {
 605                 /*
 606                  * Clear any possible leftover bits:
 607                  */
 608                 memset(tss->io_bitmap, 0xff, prev->io_bitmap_max);
 609         }
 610
 611         /*
 612          * Last branch recording recofiguration of trace hardware and
 613          * disentangling of trace data per task.
 614          */
 615         if (test_tsk_thread_flag(prev_p, TIF_BTS_TRACE_TS))
 616                 ptrace_bts_take_timestamp(prev_p, BTS_TASK_DEPARTS);
 617
 618         if (test_tsk_thread_flag(next_p, TIF_BTS_TRACE_TS))
 619                 ptrace_bts_take_timestamp(next_p, BTS_TASK_ARRIVES);
 620 }
 621
 622 /*
 623  *      switch_to(x,y) should switch tasks from x to y.
 624  *
 625  * This could still be optimized:
 626  * - fold all the options into a flag word and test it with a single test.
 627  * - could test fs/gs bitsliced
 628  *
 629  * Kprobes not supported here. Set the probe on schedule instead.
 630  */
 631 struct task_struct *
 632 __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
 633 {
 634         struct thread_struct *prev = &prev_p->thread,
 635                                  *next = &next_p->thread;
 636         int cpu = smp_processor_id();
 637         struct tss_struct *tss = &per_cpu(init_tss, cpu);
 638
 639         /* we're going to use this soon, after a few expensive things */
 640         if (next_p->fpu_counter>5)
 641                 prefetch(&next->i387.fxsave);
 642
 643         /*
 644          * Reload esp0, LDT and the page table pointer:
 645          */
 646         tss->sp0 = next->sp0;
 647
 648         /*
 649          * Switch DS and ES.
 650          * This won't pick up thread selector changes, but I guess that is ok.
 651          */
 652         asm volatile("mov %%es,%0" : "=m" (prev->es));
 653         if (unlikely(next->es | prev->es))
 654                 loadsegment(es, next->es);
 655
 656         asm volatile ("mov %%ds,%0" : "=m" (prev->ds));
 657         if (unlikely(next->ds | prev->ds))
 658                 loadsegment(ds, next->ds);
 659
 660         load_TLS(next, cpu);
 661
 662         /*
 663          * Switch FS and GS.
 664          */
 665         {
 666                 unsigned fsindex;
 667                 asm volatile("movl %%fs,%0" : "=r" (fsindex));
 668                 /* segment register != 0 always requires a reload.
 669                    also reload when it has changed.
 670                    when prev process used 64bit base always reload
 671                    to avoid an information leak. */
 672                 if (unlikely(fsindex | next->fsindex | prev->fs)) {
 673                         loadsegment(fs, next->fsindex);
 674                         /* check if the user used a selector != 0
 675                          * if yes clear 64bit base, since overloaded base
 676                          * is always mapped to the Null selector
 677                          */
 678                         if (fsindex)
 679                         prev->fs = 0;
 680                 }
 681                 /* when next process has a 64bit base use it */
 682                 if (next->fs)
 683                         wrmsrl(MSR_FS_BASE, next->fs);
 684                 prev->fsindex = fsindex;
 685         }
 686         {
 687                 unsigned gsindex;
 688                 asm volatile("movl %%gs,%0" : "=r" (gsindex));
 689                 if (unlikely(gsindex | next->gsindex | prev->gs)) {
 690                         load_gs_index(next->gsindex);
 691                         if (gsindex)
 692                         prev->gs = 0;
 693                 }
 694                 if (next->gs)
 695                         wrmsrl(MSR_KERNEL_GS_BASE, next->gs);
 696                 prev->gsindex = gsindex;
 697         }
 698
 699         /* Must be after DS reload */
 700         unlazy_fpu(prev_p);
 701
 702         /*
 703          * Switch the PDA and FPU contexts.
 704          */
 705         prev->usersp = read_pda(oldrsp);
 706         write_pda(oldrsp, next->usersp);
 707         write_pda(pcurrent, next_p);
 708
 709         write_pda(kernelstack,
 710         (unsigned long)task_stack_page(next_p) + THREAD_SIZE - PDA_STACKOFFSET);
 711 #ifdef CONFIG_CC_STACKPROTECTOR
 712         write_pda(stack_canary, next_p->stack_canary);
 713         /*
 714          * Build time only check to make sure the stack_canary is at
 715          * offset 40 in the pda; this is a gcc ABI requirement
 716          */
 717         BUILD_BUG_ON(offsetof(struct x8664_pda, stack_canary) != 40);
 718 #endif
 719
 720         /*
 721          * Now maybe reload the debug registers and handle I/O bitmaps
 722          */
 723         if (unlikely(task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT ||
 724                      task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV))
 725                 __switch_to_xtra(prev_p, next_p, tss);
 726
 727         /* If the task has used fpu the last 5 timeslices, just do a full
 728          * restore of the math state immediately to avoid the trap; the
 729          * chances of needing FPU soon are obviously high now
 730          */
 731         if (next_p->fpu_counter>5)
 732                 math_state_restore();
 733         return prev_p;
 734 }
 735
 736 /*
 737  * sys_execve() executes a new program.
 738  */
 739 asmlinkage
 740 long sys_execve(char __user *name, char __user * __user *argv,
 741                 char __user * __user *envp, struct pt_regs regs)
 742 {
 743         long error;
 744         char * filename;
 745
 746         filename = getname(name);
 747         error = PTR_ERR(filename);
 748         if (IS_ERR(filename))
 749                 return error;
 750         error = do_execve(filename, argv, envp, &regs);
 751         putname(filename);
 752         return error;
 753 }
 754
 755 void set_personality_64bit(void)
 756 {
 757         /* inherit personality from parent */
 758
 759         /* Make sure to be in 64bit mode */
 760         clear_thread_flag(TIF_IA32);
 761
 762         /* TBD: overwrites user setup. Should have two bits.
 763            But 64bit processes have always behaved this way,
 764            so it's not too bad. The main problem is just that
 765            32bit childs are affected again. */
 766         current->personality &= ~READ_IMPLIES_EXEC;
 767 }
 768
 769 asmlinkage long sys_fork(struct pt_regs *regs)
 770 {
 771         return do_fork(SIGCHLD, regs->sp, regs, 0, NULL, NULL);
 772 }
 773
 774 asmlinkage long
 775 sys_clone(unsigned long clone_flags, unsigned long newsp,
 776           void __user *parent_tid, void __user *child_tid, struct pt_regs *regs)
 777 {
 778         if (!newsp)
 779                 newsp = regs->sp;
 780         return do_fork(clone_flags, newsp, regs, 0, parent_tid, child_tid);
 781 }
 782
 783 /*
 784  * This is trivial, and on the face of it looks like it
 785  * could equally well be done in user mode.
 786  *
 787  * Not so, for quite unobvious reasons - register pressure.
 788  * In user mode vfork() cannot have a stack frame, and if
 789  * done by calling the "clone()" system call directly, you
 790  * do not have enough call-clobbered registers to hold all
 791  * the information you need.
 792  */
 793 asmlinkage long sys_vfork(struct pt_regs *regs)
 794 {
 795         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->sp, regs, 0,
 796                     NULL, NULL);
 797 }
 798
 799 unsigned long get_wchan(struct task_struct *p)
 800 {
 801         unsigned long stack;
 802         u64 fp,ip;
 803         int count = 0;
 804
 805         if (!p || p == current || p->state==TASK_RUNNING)
 806                 return 0;
 807         stack = (unsigned long)task_stack_page(p);
 808         if (p->thread.sp < stack || p->thread.sp > stack+THREAD_SIZE)
 809                 return 0;
 810         fp = *(u64 *)(p->thread.sp);
 811         do {
 812                 if (fp < (unsigned long)stack ||
 813                     fp > (unsigned long)stack+THREAD_SIZE)
 814                         return 0;
 815                 ip = *(u64 *)(fp+8);
 816                 if (!in_sched_functions(ip))
 817                         return ip;
 818                 fp = *(u64 *)fp;
 819         } while (count++ < 16);
 820         return 0;
 821 }
 822
 823 long do_arch_prctl(struct task_struct *task, int code, unsigned long addr)
 824 {
 825         int ret = 0;
 826         int doit = task == current;
 827         int cpu;
 828
 829         switch (code) {
 830         case ARCH_SET_GS:
 831                 if (addr >= TASK_SIZE_OF(task))
 832                         return -EPERM;
 833                 cpu = get_cpu();
 834                 /* handle small bases via the GDT because that's faster to
 835                    switch. */
 836                 if (addr <= 0xffffffff) {
 837                         set_32bit_tls(task, GS_TLS, addr);
 838                         if (doit) {
 839                                 load_TLS(&task->thread, cpu);
 840                                 load_gs_index(GS_TLS_SEL);
 841                         }
 842                         task->thread.gsindex = GS_TLS_SEL;
 843                         task->thread.gs = 0;
 844                 } else {
 845                         task->thread.gsindex = 0;
 846                         task->thread.gs = addr;
 847                         if (doit) {
 848                                 load_gs_index(0);
 849                                 ret = checking_wrmsrl(MSR_KERNEL_GS_BASE, addr);
 850                         }
 851                 }
 852                 put_cpu();
 853                 break;
 854         case ARCH_SET_FS:
 855                 /* Not strictly needed for fs, but do it for symmetry
 856                    with gs */
 857                 if (addr >= TASK_SIZE_OF(task))
 858                         return -EPERM;
 859                 cpu = get_cpu();
 860                 /* handle small bases via the GDT because that's faster to
 861                    switch. */
 862                 if (addr <= 0xffffffff) {
 863                         set_32bit_tls(task, FS_TLS, addr);
 864                         if (doit) {
 865                                 load_TLS(&task->thread, cpu);
 866                                 asm volatile("movl %0,%%fs" :: "r"(FS_TLS_SEL));
 867                         }
 868                         task->thread.fsindex = FS_TLS_SEL;
 869                         task->thread.fs = 0;
 870                 } else {
 871                         task->thread.fsindex = 0;
 872                         task->thread.fs = addr;
 873                         if (doit) {
 874                                 /* set the selector to 0 to not confuse
 875                                    __switch_to */
 876                                 asm volatile("movl %0,%%fs" :: "r" (0));
 877                                 ret = checking_wrmsrl(MSR_FS_BASE, addr);
 878                         }
 879                 }
 880                 put_cpu();
 881                 break;
 882         case ARCH_GET_FS: {
 883                 unsigned long base;
 884                 if (task->thread.fsindex == FS_TLS_SEL)
 885                         base = read_32bit_tls(task, FS_TLS);
 886                 else if (doit)
 887                         rdmsrl(MSR_FS_BASE, base);
 888                 else
 889                         base = task->thread.fs;
 890                 ret = put_user(base, (unsigned long __user *)addr);
 891                 break;
 892         }
 893         case ARCH_GET_GS: {
 894                 unsigned long base;
 895                 unsigned gsindex;
 896                 if (task->thread.gsindex == GS_TLS_SEL)
 897                         base = read_32bit_tls(task, GS_TLS);
 898                 else if (doit) {
 899                         asm("movl %%gs,%0" : "=r" (gsindex));
 900                         if (gsindex)
 901                                 rdmsrl(MSR_KERNEL_GS_BASE, base);
 902                         else
 903                                 base = task->thread.gs;
 904                 }
 905                 else
 906                         base = task->thread.gs;
 907                 ret = put_user(base, (unsigned long __user *)addr);
 908                 break;
 909         }
 910
 911         default:
 912                 ret = -EINVAL;
 913                 break;
 914         }
 915
 916         return ret;
 917 }
 918
 919 long sys_arch_prctl(int code, unsigned long addr)
 920 {
 921         return do_arch_prctl(current, code, addr);
 922 }
 923
 924 unsigned long arch_align_stack(unsigned long sp)
 925 {
 926         if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
 927                 sp -= get_random_int() % 8192;
 928         return sp & ~0xf;
 929 }
 930
 931 unsigned long arch_randomize_brk(struct mm_struct *mm)
 932 {
 933         unsigned long range_end = mm->brk + 0x02000000;
 934         return randomize_range(mm->brk, range_end, 0) ? : mm->brk;
 935 }
 936