git.oblomov.eu 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  *  $Id: process.c,v 1.38 2002/01/15 10:08:03 ak Exp $
  14  */
  15
  16 /*
  17  * This file handles the architecture-dependent parts of process handling..
  18  */
  19
  20 #include <stdarg.h>
  21
  22 #include <linux/cpu.h>
  23 #include <linux/errno.h>
  24 #include <linux/sched.h>
  25 #include <linux/kernel.h>
  26 #include <linux/mm.h>
  27 #include <linux/elfcore.h>
  28 #include <linux/smp.h>
  29 #include <linux/slab.h>
  30 #include <linux/user.h>
  31 #include <linux/module.h>
  32 #include <linux/a.out.h>
  33 #include <linux/interrupt.h>
  34 #include <linux/delay.h>
  35 #include <linux/ptrace.h>
  36 #include <linux/utsname.h>
  37 #include <linux/random.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
  54 asmlinkage extern void ret_from_fork(void);
  55
  56 unsigned long kernel_thread_flags = CLONE_VM | CLONE_UNTRACED;
  57
  58 static atomic_t hlt_counter = ATOMIC_INIT(0);
  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 static DEFINE_PER_CPU(unsigned int, cpu_idle_state);
  68
  69 void disable_hlt(void)
  70 {
  71         atomic_inc(&hlt_counter);
  72 }
  73
  74 EXPORT_SYMBOL(disable_hlt);
  75
  76 void enable_hlt(void)
  77 {
  78         atomic_dec(&hlt_counter);
  79 }
  80
  81 EXPORT_SYMBOL(enable_hlt);
  82
  83 /*
  84  * We use this if we don't have any better
  85  * idle routine..
  86  */
  87 void default_idle(void)
  88 {
  89         local_irq_enable();
  90
  91         if (!atomic_read(&hlt_counter)) {
  92                 clear_thread_flag(TIF_POLLING_NRFLAG);
  93                 smp_mb__after_clear_bit();
  94                 while (!need_resched()) {
  95                         local_irq_disable();
  96                         if (!need_resched())
  97                                 safe_halt();
  98                         else
  99                                 local_irq_enable();
 100                 }
 101                 set_thread_flag(TIF_POLLING_NRFLAG);
 102         } else {
 103                 while (!need_resched())
 104                         cpu_relax();
 105         }
 106 }
 107
 108 /*
 109  * On SMP it's slightly faster (but much more power-consuming!)
 110  * to poll the ->need_resched flag instead of waiting for the
 111  * cross-CPU IPI to arrive. Use this option with caution.
 112  */
 113 static void poll_idle (void)
 114 {
 115         local_irq_enable();
 116
 117         asm volatile(
 118                 "2:"
 119                 "testl %0,%1;"
 120                 "rep; nop;"
 121                 "je 2b;"
 122                 : :
 123                 "i" (_TIF_NEED_RESCHED),
 124                 "m" (current_thread_info()->flags));
 125 }
 126
 127 void cpu_idle_wait(void)
 128 {
 129         unsigned int cpu, this_cpu = get_cpu();
 130         cpumask_t map;
 131
 132         set_cpus_allowed(current, cpumask_of_cpu(this_cpu));
 133         put_cpu();
 134
 135         cpus_clear(map);
 136         for_each_online_cpu(cpu) {
 137                 per_cpu(cpu_idle_state, cpu) = 1;
 138                 cpu_set(cpu, map);
 139         }
 140
 141         __get_cpu_var(cpu_idle_state) = 0;
 142
 143         wmb();
 144         do {
 145                 ssleep(1);
 146                 for_each_online_cpu(cpu) {
 147                         if (cpu_isset(cpu, map) &&
 148                                         !per_cpu(cpu_idle_state, cpu))
 149                                 cpu_clear(cpu, map);
 150                 }
 151                 cpus_and(map, map, cpu_online_map);
 152         } while (!cpus_empty(map));
 153 }
 154 EXPORT_SYMBOL_GPL(cpu_idle_wait);
 155
 156 #ifdef CONFIG_HOTPLUG_CPU
 157 DECLARE_PER_CPU(int, cpu_state);
 158
 159 #include <asm/nmi.h>
 160 /* We halt the CPU with physical CPU hotplug */
 161 static inline void play_dead(void)
 162 {
 163         idle_task_exit();
 164         wbinvd();
 165         mb();
 166         /* Ack it */
 167         __get_cpu_var(cpu_state) = CPU_DEAD;
 168
 169         local_irq_disable();
 170         while (1)
 171                 halt();
 172 }
 173 #else
 174 static inline void play_dead(void)
 175 {
 176         BUG();
 177 }
 178 #endif /* CONFIG_HOTPLUG_CPU */
 179
 180 /*
 181  * The idle thread. There's no useful work to be
 182  * done, so just try to conserve power and have a
 183  * low exit latency (ie sit in a loop waiting for
 184  * somebody to say that they'd like to reschedule)
 185  */
 186 void cpu_idle (void)
 187 {
 188         set_thread_flag(TIF_POLLING_NRFLAG);
 189
 190         /* endless idle loop with no priority at all */
 191         while (1) {
 192                 while (!need_resched()) {
 193                         void (*idle)(void);
 194
 195                         if (__get_cpu_var(cpu_idle_state))
 196                                 __get_cpu_var(cpu_idle_state) = 0;
 197
 198                         rmb();
 199                         idle = pm_idle;
 200                         if (!idle)
 201                                 idle = default_idle;
 202                         if (cpu_is_offline(smp_processor_id()))
 203                                 play_dead();
 204                         idle();
 205                 }
 206
 207                 preempt_enable_no_resched();
 208                 schedule();
 209                 preempt_disable();
 210         }
 211 }
 212
 213 /*
 214  * This uses new MONITOR/MWAIT instructions on P4 processors with PNI,
 215  * which can obviate IPI to trigger checking of need_resched.
 216  * We execute MONITOR against need_resched and enter optimized wait state
 217  * through MWAIT. Whenever someone changes need_resched, we would be woken
 218  * up from MWAIT (without an IPI).
 219  */
 220 static void mwait_idle(void)
 221 {
 222         local_irq_enable();
 223
 224         while (!need_resched()) {
 225                 __monitor((void *)&current_thread_info()->flags, 0, 0);
 226                 smp_mb();
 227                 if (need_resched())
 228                         break;
 229                 __mwait(0, 0);
 230         }
 231 }
 232
 233 void __cpuinit select_idle_routine(const struct cpuinfo_x86 *c)
 234 {
 235         static int printed;
 236         if (cpu_has(c, X86_FEATURE_MWAIT)) {
 237                 /*
 238                  * Skip, if setup has overridden idle.
 239                  * One CPU supports mwait => All CPUs supports mwait
 240                  */
 241                 if (!pm_idle) {
 242                         if (!printed) {
 243                                 printk("using mwait in idle threads.\n");
 244                                 printed = 1;
 245                         }
 246                         pm_idle = mwait_idle;
 247                 }
 248         }
 249 }
 250
 251 static int __init idle_setup (char *str)
 252 {
 253         if (!strncmp(str, "poll", 4)) {
 254                 printk("using polling idle threads.\n");
 255                 pm_idle = poll_idle;
 256         }
 257
 258         boot_option_idle_override = 1;
 259         return 1;
 260 }
 261
 262 __setup("idle=", idle_setup);
 263
 264 /* Prints also some state that isn't saved in the pt_regs */
 265 void __show_regs(struct pt_regs * regs)
 266 {
 267         unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L, fs, gs, shadowgs;
 268         unsigned int fsindex,gsindex;
 269         unsigned int ds,cs,es;
 270
 271         printk("\n");
 272         print_modules();
 273         printk("Pid: %d, comm: %.20s %s %s %.*s\n",
 274                 current->pid, current->comm, print_tainted(),
 275                 system_utsname.release,
 276                 (int)strcspn(system_utsname.version, " "),
 277                 system_utsname.version);
 278         printk("RIP: %04lx:[<%016lx>] ", regs->cs & 0xffff, regs->rip);
 279         printk_address(regs->rip);
 280         printk("\nRSP: %04lx:%016lx  EFLAGS: %08lx\n", regs->ss, regs->rsp,
 281                 regs->eflags);
 282         printk("RAX: %016lx RBX: %016lx RCX: %016lx\n",
 283                regs->rax, regs->rbx, regs->rcx);
 284         printk("RDX: %016lx RSI: %016lx RDI: %016lx\n",
 285                regs->rdx, regs->rsi, regs->rdi);
 286         printk("RBP: %016lx R08: %016lx R09: %016lx\n",
 287                regs->rbp, regs->r8, regs->r9);
 288         printk("R10: %016lx R11: %016lx R12: %016lx\n",
 289                regs->r10, regs->r11, regs->r12);
 290         printk("R13: %016lx R14: %016lx R15: %016lx\n",
 291                regs->r13, regs->r14, regs->r15);
 292
 293         asm("movl %%ds,%0" : "=r" (ds));
 294         asm("movl %%cs,%0" : "=r" (cs));
 295         asm("movl %%es,%0" : "=r" (es));
 296         asm("movl %%fs,%0" : "=r" (fsindex));
 297         asm("movl %%gs,%0" : "=r" (gsindex));
 298
 299         rdmsrl(MSR_FS_BASE, fs);
 300         rdmsrl(MSR_GS_BASE, gs);
 301         rdmsrl(MSR_KERNEL_GS_BASE, shadowgs);
 302
 303         asm("movq %%cr0, %0": "=r" (cr0));
 304         asm("movq %%cr2, %0": "=r" (cr2));
 305         asm("movq %%cr3, %0": "=r" (cr3));
 306         asm("movq %%cr4, %0": "=r" (cr4));
 307
 308         printk("FS:  %016lx(%04x) GS:%016lx(%04x) knlGS:%016lx\n",
 309                fs,fsindex,gs,gsindex,shadowgs);
 310         printk("CS:  %04x DS: %04x ES: %04x CR0: %016lx\n", cs, ds, es, cr0);
 311         printk("CR2: %016lx CR3: %016lx CR4: %016lx\n", cr2, cr3, cr4);
 312 }
 313
 314 void show_regs(struct pt_regs *regs)
 315 {
 316         printk("CPU %d:", smp_processor_id());
 317         __show_regs(regs);
 318         show_trace(&regs->rsp);
 319 }
 320
 321 /*
 322  * Free current thread data structures etc..
 323  */
 324 void exit_thread(void)
 325 {
 326         struct task_struct *me = current;
 327         struct thread_struct *t = &me->thread;
 328
 329         /*
 330          * Remove function-return probe instances associated with this task
 331          * and put them back on the free list. Do not insert an exit probe for
 332          * this function, it will be disabled by kprobe_flush_task if you do.
 333          */
 334         kprobe_flush_task(me);
 335
 336         if (me->thread.io_bitmap_ptr) {
 337                 struct tss_struct *tss = &per_cpu(init_tss, get_cpu());
 338
 339                 kfree(t->io_bitmap_ptr);
 340                 t->io_bitmap_ptr = NULL;
 341                 /*
 342                  * Careful, clear this in the TSS too:
 343                  */
 344                 memset(tss->io_bitmap, 0xff, t->io_bitmap_max);
 345                 t->io_bitmap_max = 0;
 346                 put_cpu();
 347         }
 348 }
 349
 350 void flush_thread(void)
 351 {
 352         struct task_struct *tsk = current;
 353         struct thread_info *t = current_thread_info();
 354
 355         if (t->flags & _TIF_ABI_PENDING)
 356                 t->flags ^= (_TIF_ABI_PENDING | _TIF_IA32);
 357
 358         tsk->thread.debugreg0 = 0;
 359         tsk->thread.debugreg1 = 0;
 360         tsk->thread.debugreg2 = 0;
 361         tsk->thread.debugreg3 = 0;
 362         tsk->thread.debugreg6 = 0;
 363         tsk->thread.debugreg7 = 0;
 364         memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
 365         /*
 366          * Forget coprocessor state..
 367          */
 368         clear_fpu(tsk);
 369         clear_used_math();
 370 }
 371
 372 void release_thread(struct task_struct *dead_task)
 373 {
 374         if (dead_task->mm) {
 375                 if (dead_task->mm->context.size) {
 376                         printk("WARNING: dead process %8s still has LDT? <%p/%d>\n",
 377                                         dead_task->comm,
 378                                         dead_task->mm->context.ldt,
 379                                         dead_task->mm->context.size);
 380                         BUG();
 381                 }
 382         }
 383 }
 384
 385 static inline void set_32bit_tls(struct task_struct *t, int tls, u32 addr)
 386 {
 387         struct user_desc ud = {
 388                 .base_addr = addr,
 389                 .limit = 0xfffff,
 390                 .seg_32bit = 1,
 391                 .limit_in_pages = 1,
 392                 .useable = 1,
 393         };
 394         struct n_desc_struct *desc = (void *)t->thread.tls_array;
 395         desc += tls;
 396         desc->a = LDT_entry_a(&ud);
 397         desc->b = LDT_entry_b(&ud);
 398 }
 399
 400 static inline u32 read_32bit_tls(struct task_struct *t, int tls)
 401 {
 402         struct desc_struct *desc = (void *)t->thread.tls_array;
 403         desc += tls;
 404         return desc->base0 |
 405                 (((u32)desc->base1) << 16) |
 406                 (((u32)desc->base2) << 24);
 407 }
 408
 409 /*
 410  * This gets called before we allocate a new thread and copy
 411  * the current task into it.
 412  */
 413 void prepare_to_copy(struct task_struct *tsk)
 414 {
 415         unlazy_fpu(tsk);
 416 }
 417
 418 int copy_thread(int nr, unsigned long clone_flags, unsigned long rsp,
 419                 unsigned long unused,
 420         struct task_struct * p, struct pt_regs * regs)
 421 {
 422         int err;
 423         struct pt_regs * childregs;
 424         struct task_struct *me = current;
 425
 426         childregs = ((struct pt_regs *)
 427                         (THREAD_SIZE + (unsigned long) p->thread_info)) - 1;
 428         *childregs = *regs;
 429
 430         childregs->rax = 0;
 431         childregs->rsp = rsp;
 432         if (rsp == ~0UL)
 433                 childregs->rsp = (unsigned long)childregs;
 434
 435         p->thread.rsp = (unsigned long) childregs;
 436         p->thread.rsp0 = (unsigned long) (childregs+1);
 437         p->thread.userrsp = me->thread.userrsp;
 438
 439         set_ti_thread_flag(p->thread_info, TIF_FORK);
 440
 441         p->thread.fs = me->thread.fs;
 442         p->thread.gs = me->thread.gs;
 443
 444         asm("mov %%gs,%0" : "=m" (p->thread.gsindex));
 445         asm("mov %%fs,%0" : "=m" (p->thread.fsindex));
 446         asm("mov %%es,%0" : "=m" (p->thread.es));
 447         asm("mov %%ds,%0" : "=m" (p->thread.ds));
 448
 449         if (unlikely(me->thread.io_bitmap_ptr != NULL)) {
 450                 p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL);
 451                 if (!p->thread.io_bitmap_ptr) {
 452                         p->thread.io_bitmap_max = 0;
 453                         return -ENOMEM;
 454                 }
 455                 memcpy(p->thread.io_bitmap_ptr, me->thread.io_bitmap_ptr,
 456                                 IO_BITMAP_BYTES);
 457         }
 458
 459         /*
 460          * Set a new TLS for the child thread?
 461          */
 462         if (clone_flags & CLONE_SETTLS) {
 463 #ifdef CONFIG_IA32_EMULATION
 464                 if (test_thread_flag(TIF_IA32))
 465                         err = ia32_child_tls(p, childregs);
 466                 else
 467 #endif
 468                         err = do_arch_prctl(p, ARCH_SET_FS, childregs->r8);
 469                 if (err)
 470                         goto out;
 471         }
 472         err = 0;
 473 out:
 474         if (err && p->thread.io_bitmap_ptr) {
 475                 kfree(p->thread.io_bitmap_ptr);
 476                 p->thread.io_bitmap_max = 0;
 477         }
 478         return err;
 479 }
 480
 481 /*
 482  * This special macro can be used to load a debugging register
 483  */
 484 #define loaddebug(thread,r) set_debug(thread->debugreg ## r, r)
 485
 486 /*
 487  *      switch_to(x,y) should switch tasks from x to y.
 488  *
 489  * This could still be optimized:
 490  * - fold all the options into a flag word and test it with a single test.
 491  * - could test fs/gs bitsliced
 492  */
 493 struct task_struct *
 494 __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
 495 {
 496         struct thread_struct *prev = &prev_p->thread,
 497                                  *next = &next_p->thread;
 498         int cpu = smp_processor_id();
 499         struct tss_struct *tss = &per_cpu(init_tss, cpu);
 500
 501         unlazy_fpu(prev_p);
 502
 503         /*
 504          * Reload esp0, LDT and the page table pointer:
 505          */
 506         tss->rsp0 = next->rsp0;
 507
 508         /*
 509          * Switch DS and ES.
 510          * This won't pick up thread selector changes, but I guess that is ok.
 511          */
 512         asm volatile("mov %%es,%0" : "=m" (prev->es));
 513         if (unlikely(next->es | prev->es))
 514                 loadsegment(es, next->es);
 515
 516         asm volatile ("mov %%ds,%0" : "=m" (prev->ds));
 517         if (unlikely(next->ds | prev->ds))
 518                 loadsegment(ds, next->ds);
 519
 520         load_TLS(next, cpu);
 521
 522         /*
 523          * Switch FS and GS.
 524          */
 525         {
 526                 unsigned fsindex;
 527                 asm volatile("movl %%fs,%0" : "=r" (fsindex));
 528                 /* segment register != 0 always requires a reload.
 529                    also reload when it has changed.
 530                    when prev process used 64bit base always reload
 531                    to avoid an information leak. */
 532                 if (unlikely(fsindex | next->fsindex | prev->fs)) {
 533                         loadsegment(fs, next->fsindex);
 534                         /* check if the user used a selector != 0
 535                          * if yes clear 64bit base, since overloaded base
 536                          * is always mapped to the Null selector
 537                          */
 538                         if (fsindex)
 539                         prev->fs = 0;
 540                 }
 541                 /* when next process has a 64bit base use it */
 542                 if (next->fs)
 543                         wrmsrl(MSR_FS_BASE, next->fs);
 544                 prev->fsindex = fsindex;
 545         }
 546         {
 547                 unsigned gsindex;
 548                 asm volatile("movl %%gs,%0" : "=r" (gsindex));
 549                 if (unlikely(gsindex | next->gsindex | prev->gs)) {
 550                         load_gs_index(next->gsindex);
 551                         if (gsindex)
 552                         prev->gs = 0;
 553                 }
 554                 if (next->gs)
 555                         wrmsrl(MSR_KERNEL_GS_BASE, next->gs);
 556                 prev->gsindex = gsindex;
 557         }
 558
 559         /*
 560          * Switch the PDA context.
 561          */
 562         prev->userrsp = read_pda(oldrsp);
 563         write_pda(oldrsp, next->userrsp);
 564         write_pda(pcurrent, next_p);
 565         write_pda(kernelstack,
 566             (unsigned long)next_p->thread_info + THREAD_SIZE - PDA_STACKOFFSET);
 567
 568         /*
 569          * Now maybe reload the debug registers
 570          */
 571         if (unlikely(next->debugreg7)) {
 572                 loaddebug(next, 0);
 573                 loaddebug(next, 1);
 574                 loaddebug(next, 2);
 575                 loaddebug(next, 3);
 576                 /* no 4 and 5 */
 577                 loaddebug(next, 6);
 578                 loaddebug(next, 7);
 579         }
 580
 581
 582         /*
 583          * Handle the IO bitmap
 584          */
 585         if (unlikely(prev->io_bitmap_ptr || next->io_bitmap_ptr)) {
 586                 if (next->io_bitmap_ptr)
 587                         /*
 588                          * Copy the relevant range of the IO bitmap.
 589                          * Normally this is 128 bytes or less:
 590                          */
 591                         memcpy(tss->io_bitmap, next->io_bitmap_ptr,
 592                                 max(prev->io_bitmap_max, next->io_bitmap_max));
 593                 else {
 594                         /*
 595                          * Clear any possible leftover bits:
 596                          */
 597                         memset(tss->io_bitmap, 0xff, prev->io_bitmap_max);
 598                 }
 599         }
 600
 601         return prev_p;
 602 }
 603
 604 /*
 605  * sys_execve() executes a new program.
 606  */
 607 asmlinkage
 608 long sys_execve(char __user *name, char __user * __user *argv,
 609                 char __user * __user *envp, struct pt_regs regs)
 610 {
 611         long error;
 612         char * filename;
 613
 614         filename = getname(name);
 615         error = PTR_ERR(filename);
 616         if (IS_ERR(filename))
 617                 return error;
 618         error = do_execve(filename, argv, envp, &regs);
 619         if (error == 0) {
 620                 task_lock(current);
 621                 current->ptrace &= ~PT_DTRACE;
 622                 task_unlock(current);
 623         }
 624         putname(filename);
 625         return error;
 626 }
 627
 628 void set_personality_64bit(void)
 629 {
 630         /* inherit personality from parent */
 631
 632         /* Make sure to be in 64bit mode */
 633         clear_thread_flag(TIF_IA32);
 634
 635         /* TBD: overwrites user setup. Should have two bits.
 636            But 64bit processes have always behaved this way,
 637            so it's not too bad. The main problem is just that
 638            32bit childs are affected again. */
 639         current->personality &= ~READ_IMPLIES_EXEC;
 640 }
 641
 642 asmlinkage long sys_fork(struct pt_regs *regs)
 643 {
 644         return do_fork(SIGCHLD, regs->rsp, regs, 0, NULL, NULL);
 645 }
 646
 647 asmlinkage long
 648 sys_clone(unsigned long clone_flags, unsigned long newsp,
 649           void __user *parent_tid, void __user *child_tid, struct pt_regs *regs)
 650 {
 651         if (!newsp)
 652                 newsp = regs->rsp;
 653         return do_fork(clone_flags, newsp, regs, 0, parent_tid, child_tid);
 654 }
 655
 656 /*
 657  * This is trivial, and on the face of it looks like it
 658  * could equally well be done in user mode.
 659  *
 660  * Not so, for quite unobvious reasons - register pressure.
 661  * In user mode vfork() cannot have a stack frame, and if
 662  * done by calling the "clone()" system call directly, you
 663  * do not have enough call-clobbered registers to hold all
 664  * the information you need.
 665  */
 666 asmlinkage long sys_vfork(struct pt_regs *regs)
 667 {
 668         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->rsp, regs, 0,
 669                     NULL, NULL);
 670 }
 671
 672 unsigned long get_wchan(struct task_struct *p)
 673 {
 674         unsigned long stack;
 675         u64 fp,rip;
 676         int count = 0;
 677
 678         if (!p || p == current || p->state==TASK_RUNNING)
 679                 return 0;
 680         stack = (unsigned long)p->thread_info;
 681         if (p->thread.rsp < stack || p->thread.rsp > stack+THREAD_SIZE)
 682                 return 0;
 683         fp = *(u64 *)(p->thread.rsp);
 684         do {
 685                 if (fp < (unsigned long)stack ||
 686                     fp > (unsigned long)stack+THREAD_SIZE)
 687                         return 0;
 688                 rip = *(u64 *)(fp+8);
 689                 if (!in_sched_functions(rip))
 690                         return rip;
 691                 fp = *(u64 *)fp;
 692         } while (count++ < 16);
 693         return 0;
 694 }
 695
 696 long do_arch_prctl(struct task_struct *task, int code, unsigned long addr)
 697 {
 698         int ret = 0;
 699         int doit = task == current;
 700         int cpu;
 701
 702         switch (code) {
 703         case ARCH_SET_GS:
 704                 if (addr >= TASK_SIZE_OF(task))
 705                         return -EPERM;
 706                 cpu = get_cpu();
 707                 /* handle small bases via the GDT because that's faster to
 708                    switch. */
 709                 if (addr <= 0xffffffff) {
 710                         set_32bit_tls(task, GS_TLS, addr);
 711                         if (doit) {
 712                                 load_TLS(&task->thread, cpu);
 713                                 load_gs_index(GS_TLS_SEL);
 714                         }
 715                         task->thread.gsindex = GS_TLS_SEL;
 716                         task->thread.gs = 0;
 717                 } else {
 718                         task->thread.gsindex = 0;
 719                         task->thread.gs = addr;
 720                         if (doit) {
 721                                 load_gs_index(0);
 722                                 ret = checking_wrmsrl(MSR_KERNEL_GS_BASE, addr);
 723                         }
 724                 }
 725                 put_cpu();
 726                 break;
 727         case ARCH_SET_FS:
 728                 /* Not strictly needed for fs, but do it for symmetry
 729                    with gs */
 730                 if (addr >= TASK_SIZE_OF(task))
 731                         return -EPERM;
 732                 cpu = get_cpu();
 733                 /* handle small bases via the GDT because that's faster to
 734                    switch. */
 735                 if (addr <= 0xffffffff) {
 736                         set_32bit_tls(task, FS_TLS, addr);
 737                         if (doit) {
 738                                 load_TLS(&task->thread, cpu);
 739                                 asm volatile("movl %0,%%fs" :: "r"(FS_TLS_SEL));
 740                         }
 741                         task->thread.fsindex = FS_TLS_SEL;
 742                         task->thread.fs = 0;
 743                 } else {
 744                         task->thread.fsindex = 0;
 745                         task->thread.fs = addr;
 746                         if (doit) {
 747                                 /* set the selector to 0 to not confuse
 748                                    __switch_to */
 749                                 asm volatile("movl %0,%%fs" :: "r" (0));
 750                                 ret = checking_wrmsrl(MSR_FS_BASE, addr);
 751                         }
 752                 }
 753                 put_cpu();
 754                 break;
 755         case ARCH_GET_FS: {
 756                 unsigned long base;
 757                 if (task->thread.fsindex == FS_TLS_SEL)
 758                         base = read_32bit_tls(task, FS_TLS);
 759                 else if (doit)
 760                         rdmsrl(MSR_FS_BASE, base);
 761                 else
 762                         base = task->thread.fs;
 763                 ret = put_user(base, (unsigned long __user *)addr);
 764                 break;
 765         }
 766         case ARCH_GET_GS: {
 767                 unsigned long base;
 768                 if (task->thread.gsindex == GS_TLS_SEL)
 769                         base = read_32bit_tls(task, GS_TLS);
 770                 else if (doit)
 771                         rdmsrl(MSR_KERNEL_GS_BASE, base);
 772                 else
 773                         base = task->thread.gs;
 774                 ret = put_user(base, (unsigned long __user *)addr);
 775                 break;
 776         }
 777
 778         default:
 779                 ret = -EINVAL;
 780                 break;
 781         }
 782
 783         return ret;
 784 }
 785
 786 long sys_arch_prctl(int code, unsigned long addr)
 787 {
 788         return do_arch_prctl(current, code, addr);
 789 }
 790
 791 /*
 792  * Capture the user space registers if the task is not running (in user space)
 793  */
 794 int dump_task_regs(struct task_struct *tsk, elf_gregset_t *regs)
 795 {
 796         struct pt_regs *pp, ptregs;
 797
 798         pp = (struct pt_regs *)(tsk->thread.rsp0);
 799         --pp;
 800
 801         ptregs = *pp;
 802         ptregs.cs &= 0xffff;
 803         ptregs.ss &= 0xffff;
 804
 805         elf_core_copy_regs(regs, &ptregs);
 806
 807         return 1;
 808 }
 809
 810 unsigned long arch_align_stack(unsigned long sp)
 811 {
 812         if (randomize_va_space)
 813                 sp -= get_random_int() % 8192;
 814         return sp & ~0xf;
 815 }