err.no Git - linux-2.6/blob - 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 = 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 EXPORT_SYMBOL_GPL(idle_notifier_register);
  76
  77 void idle_notifier_unregister(struct notifier_block *n)
  78 {
  79         atomic_notifier_chain_unregister(&idle_notifier, n);
  80 }
  81 EXPORT_SYMBOL(idle_notifier_unregister);
  82
  83 enum idle_state { CPU_IDLE, CPU_NOT_IDLE };
  84 static DEFINE_PER_CPU(enum idle_state, idle_state) = CPU_NOT_IDLE;
  85
  86 void enter_idle(void)
  87 {
  88         __get_cpu_var(idle_state) = CPU_IDLE;
  89         atomic_notifier_call_chain(&idle_notifier, IDLE_START, NULL);
  90 }
  91
  92 static void __exit_idle(void)
  93 {
  94         __get_cpu_var(idle_state) = CPU_NOT_IDLE;
  95         atomic_notifier_call_chain(&idle_notifier, IDLE_END, NULL);
  96 }
  97
  98 /* Called from interrupts to signify idle end */
  99 void exit_idle(void)
 100 {
 101         if (current->pid | read_pda(irqcount))
 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         local_irq_enable();
 113
 114         current_thread_info()->status &= ~TS_POLLING;
 115         smp_mb__after_clear_bit();
 116         while (!need_resched()) {
 117                 local_irq_disable();
 118                 if (!need_resched())
 119                         safe_halt();
 120                 else
 121                         local_irq_enable();
 122         }
 123         current_thread_info()->status |= TS_POLLING;
 124 }
 125
 126 /*
 127  * On SMP it's slightly faster (but much more power-consuming!)
 128  * to poll the ->need_resched flag instead of waiting for the
 129  * cross-CPU IPI to arrive. Use this option with caution.
 130  */
 131 static void poll_idle (void)
 132 {
 133         local_irq_enable();
 134
 135         asm volatile(
 136                 "2:"
 137                 "testl %0,%1;"
 138                 "rep; nop;"
 139                 "je 2b;"
 140                 : :
 141                 "i" (_TIF_NEED_RESCHED),
 142                 "m" (current_thread_info()->flags));
 143 }
 144
 145 void cpu_idle_wait(void)
 146 {
 147         unsigned int cpu, this_cpu = get_cpu();
 148         cpumask_t map;
 149
 150         set_cpus_allowed(current, cpumask_of_cpu(this_cpu));
 151         put_cpu();
 152
 153         cpus_clear(map);
 154         for_each_online_cpu(cpu) {
 155                 per_cpu(cpu_idle_state, cpu) = 1;
 156                 cpu_set(cpu, map);
 157         }
 158
 159         __get_cpu_var(cpu_idle_state) = 0;
 160
 161         wmb();
 162         do {
 163                 ssleep(1);
 164                 for_each_online_cpu(cpu) {
 165                         if (cpu_isset(cpu, map) &&
 166                                         !per_cpu(cpu_idle_state, cpu))
 167                                 cpu_clear(cpu, map);
 168                 }
 169                 cpus_and(map, map, cpu_online_map);
 170         } while (!cpus_empty(map));
 171 }
 172 EXPORT_SYMBOL_GPL(cpu_idle_wait);
 173
 174 #ifdef CONFIG_HOTPLUG_CPU
 175 DECLARE_PER_CPU(int, cpu_state);
 176
 177 #include <asm/nmi.h>
 178 /* We halt the CPU with physical CPU hotplug */
 179 static inline void play_dead(void)
 180 {
 181         idle_task_exit();
 182         wbinvd();
 183         mb();
 184         /* Ack it */
 185         __get_cpu_var(cpu_state) = CPU_DEAD;
 186
 187         local_irq_disable();
 188         while (1)
 189                 halt();
 190 }
 191 #else
 192 static inline void play_dead(void)
 193 {
 194         BUG();
 195 }
 196 #endif /* CONFIG_HOTPLUG_CPU */
 197
 198 /*
 199  * The idle thread. There's no useful work to be
 200  * done, so just try to conserve power and have a
 201  * low exit latency (ie sit in a loop waiting for
 202  * somebody to say that they'd like to reschedule)
 203  */
 204 void cpu_idle (void)
 205 {
 206         current_thread_info()->status |= TS_POLLING;
 207         /* endless idle loop with no priority at all */
 208         while (1) {
 209                 while (!need_resched()) {
 210                         void (*idle)(void);
 211
 212                         if (__get_cpu_var(cpu_idle_state))
 213                                 __get_cpu_var(cpu_idle_state) = 0;
 214
 215                         rmb();
 216                         idle = pm_idle;
 217                         if (!idle)
 218                                 idle = default_idle;
 219                         if (cpu_is_offline(smp_processor_id()))
 220                                 play_dead();
 221                         enter_idle();
 222                         idle();
 223                         __exit_idle();
 224                 }
 225
 226                 preempt_enable_no_resched();
 227                 schedule();
 228                 preempt_disable();
 229         }
 230 }
 231
 232 /*
 233  * This uses new MONITOR/MWAIT instructions on P4 processors with PNI,
 234  * which can obviate IPI to trigger checking of need_resched.
 235  * We execute MONITOR against need_resched and enter optimized wait state
 236  * through MWAIT. Whenever someone changes need_resched, we would be woken
 237  * up from MWAIT (without an IPI).
 238  */
 239 static void mwait_idle(void)
 240 {
 241         local_irq_enable();
 242
 243         while (!need_resched()) {
 244                 __monitor((void *)&current_thread_info()->flags, 0, 0);
 245                 smp_mb();
 246                 if (need_resched())
 247                         break;
 248                 __mwait(0, 0);
 249         }
 250 }
 251
 252 void __cpuinit select_idle_routine(const struct cpuinfo_x86 *c)
 253 {
 254         static int printed;
 255         if (cpu_has(c, X86_FEATURE_MWAIT)) {
 256                 /*
 257                  * Skip, if setup has overridden idle.
 258                  * One CPU supports mwait => All CPUs supports mwait
 259                  */
 260                 if (!pm_idle) {
 261                         if (!printed) {
 262                                 printk("using mwait in idle threads.\n");
 263                                 printed = 1;
 264                         }
 265                         pm_idle = mwait_idle;
 266                 }
 267         }
 268 }
 269
 270 static int __init idle_setup (char *str)
 271 {
 272         if (!strncmp(str, "poll", 4)) {
 273                 printk("using polling idle threads.\n");
 274                 pm_idle = poll_idle;
 275         }
 276
 277         boot_option_idle_override = 1;
 278         return 1;
 279 }
 280
 281 __setup("idle=", idle_setup);
 282
 283 /* Prints also some state that isn't saved in the pt_regs */
 284 void __show_regs(struct pt_regs * regs)
 285 {
 286         unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L, fs, gs, shadowgs;
 287         unsigned int fsindex,gsindex;
 288         unsigned int ds,cs,es;
 289
 290         printk("\n");
 291         print_modules();
 292         printk("Pid: %d, comm: %.20s %s %s %.*s\n",
 293                 current->pid, current->comm, print_tainted(),
 294                 system_utsname.release,
 295                 (int)strcspn(system_utsname.version, " "),
 296                 system_utsname.version);
 297         printk("RIP: %04lx:[<%016lx>] ", regs->cs & 0xffff, regs->rip);
 298         printk_address(regs->rip);
 299         printk("RSP: %04lx:%016lx  EFLAGS: %08lx\n", regs->ss, regs->rsp,
 300                 regs->eflags);
 301         printk("RAX: %016lx RBX: %016lx RCX: %016lx\n",
 302                regs->rax, regs->rbx, regs->rcx);
 303         printk("RDX: %016lx RSI: %016lx RDI: %016lx\n",
 304                regs->rdx, regs->rsi, regs->rdi);
 305         printk("RBP: %016lx R08: %016lx R09: %016lx\n",
 306                regs->rbp, regs->r8, regs->r9);
 307         printk("R10: %016lx R11: %016lx R12: %016lx\n",
 308                regs->r10, regs->r11, regs->r12);
 309         printk("R13: %016lx R14: %016lx R15: %016lx\n",
 310                regs->r13, regs->r14, regs->r15);
 311
 312         asm("movl %%ds,%0" : "=r" (ds));
 313         asm("movl %%cs,%0" : "=r" (cs));
 314         asm("movl %%es,%0" : "=r" (es));
 315         asm("movl %%fs,%0" : "=r" (fsindex));
 316         asm("movl %%gs,%0" : "=r" (gsindex));
 317
 318         rdmsrl(MSR_FS_BASE, fs);
 319         rdmsrl(MSR_GS_BASE, gs);
 320         rdmsrl(MSR_KERNEL_GS_BASE, shadowgs);
 321
 322         asm("movq %%cr0, %0": "=r" (cr0));
 323         asm("movq %%cr2, %0": "=r" (cr2));
 324         asm("movq %%cr3, %0": "=r" (cr3));
 325         asm("movq %%cr4, %0": "=r" (cr4));
 326
 327         printk("FS:  %016lx(%04x) GS:%016lx(%04x) knlGS:%016lx\n",
 328                fs,fsindex,gs,gsindex,shadowgs);
 329         printk("CS:  %04x DS: %04x ES: %04x CR0: %016lx\n", cs, ds, es, cr0);
 330         printk("CR2: %016lx CR3: %016lx CR4: %016lx\n", cr2, cr3, cr4);
 331 }
 332
 333 void show_regs(struct pt_regs *regs)
 334 {
 335         printk("CPU %d:", smp_processor_id());
 336         __show_regs(regs);
 337         show_trace(NULL, regs, (void *)(regs + 1));
 338 }
 339
 340 /*
 341  * Free current thread data structures etc..
 342  */
 343 void exit_thread(void)
 344 {
 345         struct task_struct *me = current;
 346         struct thread_struct *t = &me->thread;
 347
 348         if (me->thread.io_bitmap_ptr) {
 349                 struct tss_struct *tss = &per_cpu(init_tss, get_cpu());
 350
 351                 kfree(t->io_bitmap_ptr);
 352                 t->io_bitmap_ptr = NULL;
 353                 clear_thread_flag(TIF_IO_BITMAP);
 354                 /*
 355                  * Careful, clear this in the TSS too:
 356                  */
 357                 memset(tss->io_bitmap, 0xff, t->io_bitmap_max);
 358                 t->io_bitmap_max = 0;
 359                 put_cpu();
 360         }
 361 }
 362
 363 void flush_thread(void)
 364 {
 365         struct task_struct *tsk = current;
 366         struct thread_info *t = current_thread_info();
 367
 368         if (t->flags & _TIF_ABI_PENDING) {
 369                 t->flags ^= (_TIF_ABI_PENDING | _TIF_IA32);
 370                 if (t->flags & _TIF_IA32)
 371                         current_thread_info()->status |= TS_COMPAT;
 372         }
 373         t->flags &= ~_TIF_DEBUG;
 374
 375         tsk->thread.debugreg0 = 0;
 376         tsk->thread.debugreg1 = 0;
 377         tsk->thread.debugreg2 = 0;
 378         tsk->thread.debugreg3 = 0;
 379         tsk->thread.debugreg6 = 0;
 380         tsk->thread.debugreg7 = 0;
 381         memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
 382         /*
 383          * Forget coprocessor state..
 384          */
 385         clear_fpu(tsk);
 386         clear_used_math();
 387 }
 388
 389 void release_thread(struct task_struct *dead_task)
 390 {
 391         if (dead_task->mm) {
 392                 if (dead_task->mm->context.size) {
 393                         printk("WARNING: dead process %8s still has LDT? <%p/%d>\n",
 394                                         dead_task->comm,
 395                                         dead_task->mm->context.ldt,
 396                                         dead_task->mm->context.size);
 397                         BUG();
 398                 }
 399         }
 400 }
 401
 402 static inline void set_32bit_tls(struct task_struct *t, int tls, u32 addr)
 403 {
 404         struct user_desc ud = {
 405                 .base_addr = addr,
 406                 .limit = 0xfffff,
 407                 .seg_32bit = 1,
 408                 .limit_in_pages = 1,
 409                 .useable = 1,
 410         };
 411         struct n_desc_struct *desc = (void *)t->thread.tls_array;
 412         desc += tls;
 413         desc->a = LDT_entry_a(&ud);
 414         desc->b = LDT_entry_b(&ud);
 415 }
 416
 417 static inline u32 read_32bit_tls(struct task_struct *t, int tls)
 418 {
 419         struct desc_struct *desc = (void *)t->thread.tls_array;
 420         desc += tls;
 421         return desc->base0 |
 422                 (((u32)desc->base1) << 16) |
 423                 (((u32)desc->base2) << 24);
 424 }
 425
 426 /*
 427  * This gets called before we allocate a new thread and copy
 428  * the current task into it.
 429  */
 430 void prepare_to_copy(struct task_struct *tsk)
 431 {
 432         unlazy_fpu(tsk);
 433 }
 434
 435 int copy_thread(int nr, unsigned long clone_flags, unsigned long rsp,
 436                 unsigned long unused,
 437         struct task_struct * p, struct pt_regs * regs)
 438 {
 439         int err;
 440         struct pt_regs * childregs;
 441         struct task_struct *me = current;
 442
 443         childregs = ((struct pt_regs *)
 444                         (THREAD_SIZE + task_stack_page(p))) - 1;
 445         *childregs = *regs;
 446
 447         childregs->rax = 0;
 448         childregs->rsp = rsp;
 449         if (rsp == ~0UL)
 450                 childregs->rsp = (unsigned long)childregs;
 451
 452         p->thread.rsp = (unsigned long) childregs;
 453         p->thread.rsp0 = (unsigned long) (childregs+1);
 454         p->thread.userrsp = me->thread.userrsp;
 455
 456         set_tsk_thread_flag(p, TIF_FORK);
 457
 458         p->thread.fs = me->thread.fs;
 459         p->thread.gs = me->thread.gs;
 460
 461         asm("mov %%gs,%0" : "=m" (p->thread.gsindex));
 462         asm("mov %%fs,%0" : "=m" (p->thread.fsindex));
 463         asm("mov %%es,%0" : "=m" (p->thread.es));
 464         asm("mov %%ds,%0" : "=m" (p->thread.ds));
 465
 466         if (unlikely(test_tsk_thread_flag(me, TIF_IO_BITMAP))) {
 467                 p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL);
 468                 if (!p->thread.io_bitmap_ptr) {
 469                         p->thread.io_bitmap_max = 0;
 470                         return -ENOMEM;
 471                 }
 472                 memcpy(p->thread.io_bitmap_ptr, me->thread.io_bitmap_ptr,
 473                                 IO_BITMAP_BYTES);
 474                 set_tsk_thread_flag(p, TIF_IO_BITMAP);
 475         }
 476
 477         /*
 478          * Set a new TLS for the child thread?
 479          */
 480         if (clone_flags & CLONE_SETTLS) {
 481 #ifdef CONFIG_IA32_EMULATION
 482                 if (test_thread_flag(TIF_IA32))
 483                         err = ia32_child_tls(p, childregs);
 484                 else
 485 #endif
 486                         err = do_arch_prctl(p, ARCH_SET_FS, childregs->r8);
 487                 if (err)
 488                         goto out;
 489         }
 490         err = 0;
 491 out:
 492         if (err && p->thread.io_bitmap_ptr) {
 493                 kfree(p->thread.io_bitmap_ptr);
 494                 p->thread.io_bitmap_max = 0;
 495         }
 496         return err;
 497 }
 498
 499 /*
 500  * This special macro can be used to load a debugging register
 501  */
 502 #define loaddebug(thread,r) set_debugreg(thread->debugreg ## r, r)
 503
 504 static inline void __switch_to_xtra(struct task_struct *prev_p,
 505                                     struct task_struct *next_p,
 506                                     struct tss_struct *tss)
 507 {
 508         struct thread_struct *prev, *next;
 509
 510         prev = &prev_p->thread,
 511         next = &next_p->thread;
 512
 513         if (test_tsk_thread_flag(next_p, TIF_DEBUG)) {
 514                 loaddebug(next, 0);
 515                 loaddebug(next, 1);
 516                 loaddebug(next, 2);
 517                 loaddebug(next, 3);
 518                 /* no 4 and 5 */
 519                 loaddebug(next, 6);
 520                 loaddebug(next, 7);
 521         }
 522
 523         if (test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
 524                 /*
 525                  * Copy the relevant range of the IO bitmap.
 526                  * Normally this is 128 bytes or less:
 527                  */
 528                 memcpy(tss->io_bitmap, next->io_bitmap_ptr,
 529                        max(prev->io_bitmap_max, next->io_bitmap_max));
 530         } else if (test_tsk_thread_flag(prev_p, TIF_IO_BITMAP)) {
 531                 /*
 532                  * Clear any possible leftover bits:
 533                  */
 534                 memset(tss->io_bitmap, 0xff, prev->io_bitmap_max);
 535         }
 536 }
 537
 538 /*
 539  *      switch_to(x,y) should switch tasks from x to y.
 540  *
 541  * This could still be optimized:
 542  * - fold all the options into a flag word and test it with a single test.
 543  * - could test fs/gs bitsliced
 544  *
 545  * Kprobes not supported here. Set the probe on schedule instead.
 546  */
 547 __kprobes struct task_struct *
 548 __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
 549 {
 550         struct thread_struct *prev = &prev_p->thread,
 551                                  *next = &next_p->thread;
 552         int cpu = smp_processor_id();
 553         struct tss_struct *tss = &per_cpu(init_tss, cpu);
 554
 555         /* we're going to use this soon, after a few expensive things */
 556         if (next_p->fpu_counter>5)
 557                 prefetch(&next->i387.fxsave);
 558
 559         /*
 560          * Reload esp0, LDT and the page table pointer:
 561          */
 562         tss->rsp0 = next->rsp0;
 563
 564         /*
 565          * Switch DS and ES.
 566          * This won't pick up thread selector changes, but I guess that is ok.
 567          */
 568         asm volatile("mov %%es,%0" : "=m" (prev->es));
 569         if (unlikely(next->es | prev->es))
 570                 loadsegment(es, next->es);
 571
 572         asm volatile ("mov %%ds,%0" : "=m" (prev->ds));
 573         if (unlikely(next->ds | prev->ds))
 574                 loadsegment(ds, next->ds);
 575
 576         load_TLS(next, cpu);
 577
 578         /*
 579          * Switch FS and GS.
 580          */
 581         {
 582                 unsigned fsindex;
 583                 asm volatile("movl %%fs,%0" : "=r" (fsindex));
 584                 /* segment register != 0 always requires a reload.
 585                    also reload when it has changed.
 586                    when prev process used 64bit base always reload
 587                    to avoid an information leak. */
 588                 if (unlikely(fsindex | next->fsindex | prev->fs)) {
 589                         loadsegment(fs, next->fsindex);
 590                         /* check if the user used a selector != 0
 591                          * if yes clear 64bit base, since overloaded base
 592                          * is always mapped to the Null selector
 593                          */
 594                         if (fsindex)
 595                         prev->fs = 0;
 596                 }
 597                 /* when next process has a 64bit base use it */
 598                 if (next->fs)
 599                         wrmsrl(MSR_FS_BASE, next->fs);
 600                 prev->fsindex = fsindex;
 601         }
 602         {
 603                 unsigned gsindex;
 604                 asm volatile("movl %%gs,%0" : "=r" (gsindex));
 605                 if (unlikely(gsindex | next->gsindex | prev->gs)) {
 606                         load_gs_index(next->gsindex);
 607                         if (gsindex)
 608                         prev->gs = 0;
 609                 }
 610                 if (next->gs)
 611                         wrmsrl(MSR_KERNEL_GS_BASE, next->gs);
 612                 prev->gsindex = gsindex;
 613         }
 614
 615         /*
 616          * Switch the PDA and FPU contexts.
 617          */
 618         prev->userrsp = read_pda(oldrsp);
 619         write_pda(oldrsp, next->userrsp);
 620         write_pda(pcurrent, next_p);
 621
 622         /* This must be here to ensure both math_state_restore() and
 623            kernel_fpu_begin() work consistently.
 624            And the AMD workaround requires it to be after DS reload. */
 625         unlazy_fpu(prev_p);
 626         write_pda(kernelstack,
 627                   task_stack_page(next_p) + THREAD_SIZE - PDA_STACKOFFSET);
 628
 629         /*
 630          * Now maybe reload the debug registers and handle I/O bitmaps
 631          */
 632         if (unlikely((task_thread_info(next_p)->flags & _TIF_WORK_CTXSW))
 633             || test_tsk_thread_flag(prev_p, TIF_IO_BITMAP))
 634                 __switch_to_xtra(prev_p, next_p, tss);
 635
 636         /* If the task has used fpu the last 5 timeslices, just do a full
 637          * restore of the math state immediately to avoid the trap; the
 638          * chances of needing FPU soon are obviously high now
 639          */
 640         if (next_p->fpu_counter>5)
 641                 math_state_restore();
 642         return prev_p;
 643 }
 644
 645 /*
 646  * sys_execve() executes a new program.
 647  */
 648 asmlinkage
 649 long sys_execve(char __user *name, char __user * __user *argv,
 650                 char __user * __user *envp, struct pt_regs regs)
 651 {
 652         long error;
 653         char * filename;
 654
 655         filename = getname(name);
 656         error = PTR_ERR(filename);
 657         if (IS_ERR(filename))
 658                 return error;
 659         error = do_execve(filename, argv, envp, &regs);
 660         if (error == 0) {
 661                 task_lock(current);
 662                 current->ptrace &= ~PT_DTRACE;
 663                 task_unlock(current);
 664         }
 665         putname(filename);
 666         return error;
 667 }
 668
 669 void set_personality_64bit(void)
 670 {
 671         /* inherit personality from parent */
 672
 673         /* Make sure to be in 64bit mode */
 674         clear_thread_flag(TIF_IA32);
 675
 676         /* TBD: overwrites user setup. Should have two bits.
 677            But 64bit processes have always behaved this way,
 678            so it's not too bad. The main problem is just that
 679            32bit childs are affected again. */
 680         current->personality &= ~READ_IMPLIES_EXEC;
 681 }
 682
 683 asmlinkage long sys_fork(struct pt_regs *regs)
 684 {
 685         return do_fork(SIGCHLD, regs->rsp, regs, 0, NULL, NULL);
 686 }
 687
 688 asmlinkage long
 689 sys_clone(unsigned long clone_flags, unsigned long newsp,
 690           void __user *parent_tid, void __user *child_tid, struct pt_regs *regs)
 691 {
 692         if (!newsp)
 693                 newsp = regs->rsp;
 694         return do_fork(clone_flags, newsp, regs, 0, parent_tid, child_tid);
 695 }
 696
 697 /*
 698  * This is trivial, and on the face of it looks like it
 699  * could equally well be done in user mode.
 700  *
 701  * Not so, for quite unobvious reasons - register pressure.
 702  * In user mode vfork() cannot have a stack frame, and if
 703  * done by calling the "clone()" system call directly, you
 704  * do not have enough call-clobbered registers to hold all
 705  * the information you need.
 706  */
 707 asmlinkage long sys_vfork(struct pt_regs *regs)
 708 {
 709         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->rsp, regs, 0,
 710                     NULL, NULL);
 711 }
 712
 713 unsigned long get_wchan(struct task_struct *p)
 714 {
 715         unsigned long stack;
 716         u64 fp,rip;
 717         int count = 0;
 718
 719         if (!p || p == current || p->state==TASK_RUNNING)
 720                 return 0;
 721         stack = (unsigned long)task_stack_page(p);
 722         if (p->thread.rsp < stack || p->thread.rsp > stack+THREAD_SIZE)
 723                 return 0;
 724         fp = *(u64 *)(p->thread.rsp);
 725         do {
 726                 if (fp < (unsigned long)stack ||
 727                     fp > (unsigned long)stack+THREAD_SIZE)
 728                         return 0;
 729                 rip = *(u64 *)(fp+8);
 730                 if (!in_sched_functions(rip))
 731                         return rip;
 732                 fp = *(u64 *)fp;
 733         } while (count++ < 16);
 734         return 0;
 735 }
 736
 737 long do_arch_prctl(struct task_struct *task, int code, unsigned long addr)
 738 {
 739         int ret = 0;
 740         int doit = task == current;
 741         int cpu;
 742
 743         switch (code) {
 744         case ARCH_SET_GS:
 745                 if (addr >= TASK_SIZE_OF(task))
 746                         return -EPERM;
 747                 cpu = get_cpu();
 748                 /* handle small bases via the GDT because that's faster to
 749                    switch. */
 750                 if (addr <= 0xffffffff) {
 751                         set_32bit_tls(task, GS_TLS, addr);
 752                         if (doit) {
 753                                 load_TLS(&task->thread, cpu);
 754                                 load_gs_index(GS_TLS_SEL);
 755                         }
 756                         task->thread.gsindex = GS_TLS_SEL;
 757                         task->thread.gs = 0;
 758                 } else {
 759                         task->thread.gsindex = 0;
 760                         task->thread.gs = addr;
 761                         if (doit) {
 762                                 load_gs_index(0);
 763                                 ret = checking_wrmsrl(MSR_KERNEL_GS_BASE, addr);
 764                         }
 765                 }
 766                 put_cpu();
 767                 break;
 768         case ARCH_SET_FS:
 769                 /* Not strictly needed for fs, but do it for symmetry
 770                    with gs */
 771                 if (addr >= TASK_SIZE_OF(task))
 772                         return -EPERM;
 773                 cpu = get_cpu();
 774                 /* handle small bases via the GDT because that's faster to
 775                    switch. */
 776                 if (addr <= 0xffffffff) {
 777                         set_32bit_tls(task, FS_TLS, addr);
 778                         if (doit) {
 779                                 load_TLS(&task->thread, cpu);
 780                                 asm volatile("movl %0,%%fs" :: "r"(FS_TLS_SEL));
 781                         }
 782                         task->thread.fsindex = FS_TLS_SEL;
 783                         task->thread.fs = 0;
 784                 } else {
 785                         task->thread.fsindex = 0;
 786                         task->thread.fs = addr;
 787                         if (doit) {
 788                                 /* set the selector to 0 to not confuse
 789                                    __switch_to */
 790                                 asm volatile("movl %0,%%fs" :: "r" (0));
 791                                 ret = checking_wrmsrl(MSR_FS_BASE, addr);
 792                         }
 793                 }
 794                 put_cpu();
 795                 break;
 796         case ARCH_GET_FS: {
 797                 unsigned long base;
 798                 if (task->thread.fsindex == FS_TLS_SEL)
 799                         base = read_32bit_tls(task, FS_TLS);
 800                 else if (doit)
 801                         rdmsrl(MSR_FS_BASE, base);
 802                 else
 803                         base = task->thread.fs;
 804                 ret = put_user(base, (unsigned long __user *)addr);
 805                 break;
 806         }
 807         case ARCH_GET_GS: {
 808                 unsigned long base;
 809                 unsigned gsindex;
 810                 if (task->thread.gsindex == GS_TLS_SEL)
 811                         base = read_32bit_tls(task, GS_TLS);
 812                 else if (doit) {
 813                         asm("movl %%gs,%0" : "=r" (gsindex));
 814                         if (gsindex)
 815                                 rdmsrl(MSR_KERNEL_GS_BASE, base);
 816                         else
 817                                 base = task->thread.gs;
 818                 }
 819                 else
 820                         base = task->thread.gs;
 821                 ret = put_user(base, (unsigned long __user *)addr);
 822                 break;
 823         }
 824
 825         default:
 826                 ret = -EINVAL;
 827                 break;
 828         }
 829
 830         return ret;
 831 }
 832
 833 long sys_arch_prctl(int code, unsigned long addr)
 834 {
 835         return do_arch_prctl(current, code, addr);
 836 }
 837
 838 /*
 839  * Capture the user space registers if the task is not running (in user space)
 840  */
 841 int dump_task_regs(struct task_struct *tsk, elf_gregset_t *regs)
 842 {
 843         struct pt_regs *pp, ptregs;
 844
 845         pp = task_pt_regs(tsk);
 846
 847         ptregs = *pp;
 848         ptregs.cs &= 0xffff;
 849         ptregs.ss &= 0xffff;
 850
 851         elf_core_copy_regs(regs, &ptregs);
 852
 853         return 1;
 854 }
 855
 856 unsigned long arch_align_stack(unsigned long sp)
 857 {
 858         if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
 859                 sp -= get_random_int() % 8192;
 860         return sp & ~0xf;
 861 }