arch/x86_64/kernel/process-xen.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  *  Jun Nakajima <jun.nakajima@intel.com>
  15  *     Modified for Xen
  16  */
  17
  18 /*
  19  * This file handles the architecture-dependent parts of process handling..
  20  */
  21
  22 #include <stdarg.h>
  23
  24 #include <linux/cpu.h>
  25 #include <linux/errno.h>
  26 #include <linux/sched.h>
  27 #include <linux/kernel.h>
  28 #include <linux/mm.h>
  29 #include <linux/elfcore.h>
  30 #include <linux/smp.h>
  31 #include <linux/slab.h>
  32 #include <linux/user.h>
  33 #include <linux/module.h>
  34 #include <linux/a.out.h>
  35 #include <linux/interrupt.h>
  36 #include <linux/delay.h>
  37 #include <linux/ptrace.h>
  38 #include <linux/utsname.h>
  39 #include <linux/random.h>
  40 #include <linux/notifier.h>
  41 #include <linux/kprobes.h>
  42
  43 #include <asm/uaccess.h>
  44 #include <asm/pgtable.h>
  45 #include <asm/system.h>
  46 #include <asm/io.h>
  47 #include <asm/processor.h>
  48 #include <asm/i387.h>
  49 #include <asm/mmu_context.h>
  50 #include <asm/pda.h>
  51 #include <asm/prctl.h>
  52 #include <asm/kdebug.h>
  53 #include <xen/interface/dom0_ops.h>
  54 #include <xen/interface/physdev.h>
  55 #include <xen/interface/vcpu.h>
  56 #include <asm/desc.h>
  57 #include <asm/proto.h>
  58 #include <asm/hardirq.h>
  59 #include <asm/ia32.h>
  60 #include <asm/idle.h>
  61
  62 #include <xen/cpu_hotplug.h>
  63
  64 asmlinkage extern void ret_from_fork(void);
  65
  66 unsigned long kernel_thread_flags = CLONE_VM | CLONE_UNTRACED;
  67
  68 unsigned long boot_option_idle_override = 0;
  69 EXPORT_SYMBOL(boot_option_idle_override);
  70
  71 /*
  72  * Powermanagement idle function, if any..
  73  */
  74 void (*pm_idle)(void);
  75 static DEFINE_PER_CPU(unsigned int, cpu_idle_state);
  76
  77 static ATOMIC_NOTIFIER_HEAD(idle_notifier);
  78
  79 void idle_notifier_register(struct notifier_block *n)
  80 {
  81         atomic_notifier_chain_register(&idle_notifier, n);
  82 }
  83 EXPORT_SYMBOL_GPL(idle_notifier_register);
  84
  85 void idle_notifier_unregister(struct notifier_block *n)
  86 {
  87         atomic_notifier_chain_unregister(&idle_notifier, n);
  88 }
  89 EXPORT_SYMBOL(idle_notifier_unregister);
  90
  91 enum idle_state { CPU_IDLE, CPU_NOT_IDLE };
  92 static DEFINE_PER_CPU(enum idle_state, idle_state) = CPU_NOT_IDLE;
  93
  94 void enter_idle(void)
  95 {
  96         __get_cpu_var(idle_state) = CPU_IDLE;
  97         atomic_notifier_call_chain(&idle_notifier, IDLE_START, NULL);
  98 }
  99
 100 static void __exit_idle(void)
 101 {
 102         __get_cpu_var(idle_state) = CPU_NOT_IDLE;
 103         atomic_notifier_call_chain(&idle_notifier, IDLE_END, NULL);
 104 }
 105
 106 /* Called from interrupts to signify idle end */
 107 void exit_idle(void)
 108 {
 109         if (current->pid | read_pda(irqcount))
 110                 return;
 111         __exit_idle();
 112 }
 113
 114 /* XXX XEN doesn't use default_idle(), poll_idle(). Use xen_idle() instead. */
 115 void xen_idle(void)
 116 {
 117         local_irq_disable();
 118
 119         if (need_resched())
 120                 local_irq_enable();
 121         else {
 122                 current_thread_info()->status &= ~TS_POLLING;
 123                 smp_mb__after_clear_bit();
 124                 safe_halt();
 125                 current_thread_info()->status |= TS_POLLING;
 126         }
 127 }
 128
 129 #ifdef CONFIG_HOTPLUG_CPU
 130 static inline void play_dead(void)
 131 {
 132         idle_task_exit();
 133         local_irq_disable();
 134         cpu_clear(smp_processor_id(), cpu_initialized);
 135         preempt_enable_no_resched();
 136         HYPERVISOR_vcpu_op(VCPUOP_down, smp_processor_id(), NULL);
 137         cpu_bringup();
 138 }
 139 #else
 140 static inline void play_dead(void)
 141 {
 142         BUG();
 143 }
 144 #endif /* CONFIG_HOTPLUG_CPU */
 145
 146 /*
 147  * The idle thread. There's no useful work to be
 148  * done, so just try to conserve power and have a
 149  * low exit latency (ie sit in a loop waiting for
 150  * somebody to say that they'd like to reschedule)
 151  */
 152 void cpu_idle (void)
 153 {
 154         current_thread_info()->status |= TS_POLLING;
 155         /* endless idle loop with no priority at all */
 156         while (1) {
 157                 while (!need_resched()) {
 158                         if (__get_cpu_var(cpu_idle_state))
 159                                 __get_cpu_var(cpu_idle_state) = 0;
 160                         rmb();
 161
 162                         if (cpu_is_offline(smp_processor_id()))
 163                                 play_dead();
 164                         enter_idle();
 165                         xen_idle();
 166                         __exit_idle();
 167                 }
 168
 169                 preempt_enable_no_resched();
 170                 schedule();
 171                 preempt_disable();
 172         }
 173 }
 174
 175 void cpu_idle_wait(void)
 176 {
 177         unsigned int cpu, this_cpu = get_cpu();
 178         cpumask_t map;
 179
 180         set_cpus_allowed(current, cpumask_of_cpu(this_cpu));
 181         put_cpu();
 182
 183         cpus_clear(map);
 184         for_each_online_cpu(cpu) {
 185                 per_cpu(cpu_idle_state, cpu) = 1;
 186                 cpu_set(cpu, map);
 187         }
 188
 189         __get_cpu_var(cpu_idle_state) = 0;
 190
 191         wmb();
 192         do {
 193                 ssleep(1);
 194                 for_each_online_cpu(cpu) {
 195                         if (cpu_isset(cpu, map) &&
 196                                         !per_cpu(cpu_idle_state, cpu))
 197                                 cpu_clear(cpu, map);
 198                 }
 199                 cpus_and(map, map, cpu_online_map);
 200         } while (!cpus_empty(map));
 201 }
 202 EXPORT_SYMBOL_GPL(cpu_idle_wait);
 203
 204 /* XXX XEN doesn't use mwait_idle(), select_idle_routine(), idle_setup(). */
 205 /* Always use xen_idle() instead. */
 206 void __cpuinit select_idle_routine(const struct cpuinfo_x86 *c) {}
 207
 208 /* Prints also some state that isn't saved in the pt_regs */
 209 void __show_regs(struct pt_regs * regs)
 210 {
 211         unsigned long fs, gs, shadowgs;
 212         unsigned int fsindex,gsindex;
 213         unsigned int ds,cs,es;
 214
 215         printk("\n");
 216         print_modules();
 217         printk("Pid: %d, comm: %.20s %s %s %.*s\n",
 218                 current->pid, current->comm, print_tainted(),
 219                 system_utsname.release,
 220                 (int)strcspn(system_utsname.version, " "),
 221                 system_utsname.version);
 222         printk("RIP: %04lx:[<%016lx>] ", regs->cs & 0xffff, regs->rip);
 223         printk_address(regs->rip);
 224         printk("RSP: %04lx:%016lx  EFLAGS: %08lx\n", regs->ss, regs->rsp,
 225                 regs->eflags);
 226         printk("RAX: %016lx RBX: %016lx RCX: %016lx\n",
 227                regs->rax, regs->rbx, regs->rcx);
 228         printk("RDX: %016lx RSI: %016lx RDI: %016lx\n",
 229                regs->rdx, regs->rsi, regs->rdi);
 230         printk("RBP: %016lx R08: %016lx R09: %016lx\n",
 231                regs->rbp, regs->r8, regs->r9);
 232         printk("R10: %016lx R11: %016lx R12: %016lx\n",
 233                regs->r10, regs->r11, regs->r12);
 234         printk("R13: %016lx R14: %016lx R15: %016lx\n",
 235                regs->r13, regs->r14, regs->r15);
 236
 237         asm("mov %%ds,%0" : "=r" (ds));
 238         asm("mov %%cs,%0" : "=r" (cs));
 239         asm("mov %%es,%0" : "=r" (es));
 240         asm("mov %%fs,%0" : "=r" (fsindex));
 241         asm("mov %%gs,%0" : "=r" (gsindex));
 242
 243         rdmsrl(MSR_FS_BASE, fs);
 244         rdmsrl(MSR_GS_BASE, gs);
 245         rdmsrl(MSR_KERNEL_GS_BASE, shadowgs);
 246
 247         printk("FS:  %016lx(%04x) GS:%016lx(%04x) knlGS:%016lx\n",
 248                fs,fsindex,gs,gsindex,shadowgs);
 249         printk("CS:  %04x DS: %04x ES: %04x\n", cs, ds, es);
 250
 251 }
 252
 253 void show_regs(struct pt_regs *regs)
 254 {
 255         printk("CPU %d:", smp_processor_id());
 256         __show_regs(regs);
 257         show_trace(NULL, regs, &regs->rsp);
 258 }
 259
 260 /*
 261  * Free current thread data structures etc..
 262  */
 263 void exit_thread(void)
 264 {
 265         struct task_struct *me = current;
 266         struct thread_struct *t = &me->thread;
 267
 268         if (me->thread.io_bitmap_ptr) {
 269 #ifndef CONFIG_X86_NO_TSS
 270                 struct tss_struct *tss = &per_cpu(init_tss, get_cpu());
 271 #endif
 272 #ifdef CONFIG_XEN
 273                 struct physdev_set_iobitmap iobmp_op = { 0 };
 274 #endif
 275
 276                 kfree(t->io_bitmap_ptr);
 277                 t->io_bitmap_ptr = NULL;
 278                 /*
 279                  * Careful, clear this in the TSS too:
 280                  */
 281 #ifndef CONFIG_X86_NO_TSS
 282                 memset(tss->io_bitmap, 0xff, t->io_bitmap_max);
 283                 put_cpu();
 284 #endif
 285 #ifdef CONFIG_XEN
 286                 HYPERVISOR_physdev_op(PHYSDEVOP_set_iobitmap, &iobmp_op);
 287 #endif
 288                 t->io_bitmap_max = 0;
 289         }
 290 }
 291
 292 void load_gs_index(unsigned gs)
 293 {
 294         HYPERVISOR_set_segment_base(SEGBASE_GS_USER_SEL, gs);
 295 }
 296
 297 void flush_thread(void)
 298 {
 299         struct task_struct *tsk = current;
 300         struct thread_info *t = current_thread_info();
 301
 302         if (t->flags & _TIF_ABI_PENDING) {
 303                 t->flags ^= (_TIF_ABI_PENDING | _TIF_IA32);
 304                 if (t->flags & _TIF_IA32)
 305                         current_thread_info()->status |= TS_COMPAT;
 306         }
 307
 308
 309         tsk->thread.debugreg0 = 0;
 310         tsk->thread.debugreg1 = 0;
 311         tsk->thread.debugreg2 = 0;
 312         tsk->thread.debugreg3 = 0;
 313         tsk->thread.debugreg6 = 0;
 314         tsk->thread.debugreg7 = 0;
 315         memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
 316         /*
 317          * Forget coprocessor state..
 318          */
 319         clear_fpu(tsk);
 320         clear_used_math();
 321 }
 322
 323 void release_thread(struct task_struct *dead_task)
 324 {
 325         if (dead_task->mm) {
 326                 if (dead_task->mm->context.size) {
 327                         printk("WARNING: dead process %8s still has LDT? <%p/%d>\n",
 328                                         dead_task->comm,
 329                                         dead_task->mm->context.ldt,
 330                                         dead_task->mm->context.size);
 331                         BUG();
 332                 }
 333         }
 334 }
 335
 336 static inline void set_32bit_tls(struct task_struct *t, int tls, u32 addr)
 337 {
 338         struct user_desc ud = {
 339                 .base_addr = addr,
 340                 .limit = 0xfffff,
 341                 .seg_32bit = 1,
 342                 .limit_in_pages = 1,
 343                 .useable = 1,
 344         };
 345         struct n_desc_struct *desc = (void *)t->thread.tls_array;
 346         desc += tls;
 347         desc->a = LDT_entry_a(&ud);
 348         desc->b = LDT_entry_b(&ud);
 349 }
 350
 351 static inline u32 read_32bit_tls(struct task_struct *t, int tls)
 352 {
 353         struct desc_struct *desc = (void *)t->thread.tls_array;
 354         desc += tls;
 355         return desc->base0 |
 356                 (((u32)desc->base1) << 16) |
 357                 (((u32)desc->base2) << 24);
 358 }
 359
 360 /*
 361  * This gets called before we allocate a new thread and copy
 362  * the current task into it.
 363  */
 364 void prepare_to_copy(struct task_struct *tsk)
 365 {
 366         unlazy_fpu(tsk);
 367 }
 368
 369 int copy_thread(int nr, unsigned long clone_flags, unsigned long rsp,
 370                 unsigned long unused,
 371         struct task_struct * p, struct pt_regs * regs)
 372 {
 373         int err;
 374         struct pt_regs * childregs;
 375         struct task_struct *me = current;
 376
 377         childregs = ((struct pt_regs *)
 378                         (THREAD_SIZE + task_stack_page(p))) - 1;
 379         *childregs = *regs;
 380
 381         childregs->rax = 0;
 382         childregs->rsp = rsp;
 383         if (rsp == ~0UL)
 384                 childregs->rsp = (unsigned long)childregs;
 385
 386         p->thread.rsp = (unsigned long) childregs;
 387         p->thread.rsp0 = (unsigned long) (childregs+1);
 388         p->thread.userrsp = me->thread.userrsp;
 389
 390         set_tsk_thread_flag(p, TIF_FORK);
 391
 392         p->thread.fs = me->thread.fs;
 393         p->thread.gs = me->thread.gs;
 394
 395         asm("mov %%gs,%0" : "=m" (p->thread.gsindex));
 396         asm("mov %%fs,%0" : "=m" (p->thread.fsindex));
 397         asm("mov %%es,%0" : "=m" (p->thread.es));
 398         asm("mov %%ds,%0" : "=m" (p->thread.ds));
 399
 400         if (unlikely(me->thread.io_bitmap_ptr != NULL)) {
 401                 p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL);
 402                 if (!p->thread.io_bitmap_ptr) {
 403                         p->thread.io_bitmap_max = 0;
 404                         return -ENOMEM;
 405                 }
 406                 memcpy(p->thread.io_bitmap_ptr, me->thread.io_bitmap_ptr,
 407                                 IO_BITMAP_BYTES);
 408         }
 409
 410         /*
 411          * Set a new TLS for the child thread?
 412          */
 413         if (clone_flags & CLONE_SETTLS) {
 414 #ifdef CONFIG_IA32_EMULATION
 415                 if (test_thread_flag(TIF_IA32))
 416                         err = ia32_child_tls(p, childregs);
 417                 else
 418 #endif
 419                         err = do_arch_prctl(p, ARCH_SET_FS, childregs->r8);
 420                 if (err)
 421                         goto out;
 422         }
 423         p->thread.iopl = current->thread.iopl;
 424
 425         err = 0;
 426 out:
 427         if (err && p->thread.io_bitmap_ptr) {
 428                 kfree(p->thread.io_bitmap_ptr);
 429                 p->thread.io_bitmap_max = 0;
 430         }
 431         return err;
 432 }
 433
 434 static inline void __save_init_fpu( struct task_struct *tsk )
 435 {
 436         asm volatile( "rex64 ; fxsave %0 ; fnclex"
 437                       : "=m" (tsk->thread.i387.fxsave));
 438         tsk->thread_info->status &= ~TS_USEDFPU;
 439 }
 440
 441 /*
 442  *      switch_to(x,y) should switch tasks from x to y.
 443  *
 444  * This could still be optimized:
 445  * - fold all the options into a flag word and test it with a single test.
 446  * - could test fs/gs bitsliced
 447  *
 448  * Kprobes not supported here. Set the probe on schedule instead.
 449  */
 450 __kprobes struct task_struct *
 451 __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
 452 {
 453         struct thread_struct *prev = &prev_p->thread,
 454                                  *next = &next_p->thread;
 455         int cpu = smp_processor_id();
 456 #ifndef CONFIG_X86_NO_TSS
 457         struct tss_struct *tss = &per_cpu(init_tss, cpu);
 458 #endif
 459         struct physdev_set_iopl iopl_op;
 460         struct physdev_set_iobitmap iobmp_op;
 461         multicall_entry_t _mcl[8], *mcl = _mcl;
 462
 463         /*
 464          * Reload esp0, LDT and the page table pointer:
 465          */
 466         mcl->op      = __HYPERVISOR_stack_switch;
 467         mcl->args[0] = __KERNEL_DS;
 468         mcl->args[1] = next->rsp0;
 469         mcl++;
 470
 471         /*
 472          * Load the per-thread Thread-Local Storage descriptor.
 473          * This is load_TLS(next, cpu) with multicalls.
 474          */
 475 #define C(i) do {                                                       \
 476         if (unlikely(next->tls_array[i] != prev->tls_array[i])) {       \
 477                 mcl->op      = __HYPERVISOR_update_descriptor;          \
 478                 mcl->args[0] = virt_to_machine(                         \
 479                         &cpu_gdt(cpu)[GDT_ENTRY_TLS_MIN + i]);          \
 480                 mcl->args[1] = next->tls_array[i];                      \
 481                 mcl++;                                                  \
 482         }                                                               \
 483 } while (0)
 484         C(0); C(1); C(2);
 485 #undef C
 486
 487         if (unlikely(prev->iopl != next->iopl)) {
 488                 iopl_op.iopl = (next->iopl == 0) ? 1 : next->iopl;
 489                 mcl->op      = __HYPERVISOR_physdev_op;
 490                 mcl->args[0] = PHYSDEVOP_set_iopl;
 491                 mcl->args[1] = (unsigned long)&iopl_op;
 492                 mcl++;
 493         }
 494
 495         if (unlikely(prev->io_bitmap_ptr || next->io_bitmap_ptr)) {
 496                 iobmp_op.bitmap   = (char *)next->io_bitmap_ptr;
 497                 iobmp_op.nr_ports = next->io_bitmap_ptr ? IO_BITMAP_BITS : 0;
 498                 mcl->op      = __HYPERVISOR_physdev_op;
 499                 mcl->args[0] = PHYSDEVOP_set_iobitmap;
 500                 mcl->args[1] = (unsigned long)&iobmp_op;
 501                 mcl++;
 502         }
 503
 504         (void)HYPERVISOR_multicall(_mcl, mcl - _mcl);
 505         /*
 506          * Switch DS and ES.
 507          * This won't pick up thread selector changes, but I guess that is ok.
 508          */
 509         if (unlikely(next->es))
 510                 loadsegment(es, next->es);
 511
 512         if (unlikely(next->ds))
 513                 loadsegment(ds, next->ds);
 514
 515         /*
 516          * Switch FS and GS.
 517          */
 518         if (unlikely(next->fsindex))
 519                 loadsegment(fs, next->fsindex);
 520
 521         if (next->fs)
 522                 HYPERVISOR_set_segment_base(SEGBASE_FS, next->fs);
 523
 524         if (unlikely(next->gsindex))
 525                 load_gs_index(next->gsindex);
 526
 527         if (next->gs)
 528                 HYPERVISOR_set_segment_base(SEGBASE_GS_USER, next->gs);
 529
 530         /*
 531          * Switch the PDA and FPU context.
 532          */
 533         prev->userrsp = read_pda(oldrsp);
 534         write_pda(oldrsp, next->userrsp);
 535         write_pda(pcurrent, next_p);
 536
 537         /* This must be here to ensure both math_state_restore() and
 538            kernel_fpu_begin() work consistently.
 539            And the AMD workaround requires it to be after DS reload. */
 540         /*
 541          * This is basically '__unlazy_fpu', except that we queue a
 542          * multicall to indicate FPU task switch, rather than
 543          * synchronously trapping to Xen.
 544          */
 545         if (prev_p->thread_info->status & TS_USEDFPU) {
 546                 __save_init_fpu(prev_p); /* _not_ save_init_fpu() */
 547                 HYPERVISOR_fpu_taskswitch(1);
 548         }
 549         write_pda(kernelstack,
 550                   task_stack_page(next_p) + THREAD_SIZE - PDA_STACKOFFSET);
 551
 552         /*
 553          * Now maybe reload the debug registers
 554          */
 555         if (unlikely(next->debugreg7)) {
 556                 set_debugreg(next->debugreg0, 0);
 557                 set_debugreg(next->debugreg1, 1);
 558                 set_debugreg(next->debugreg2, 2);
 559                 set_debugreg(next->debugreg3, 3);
 560                 /* no 4 and 5 */
 561                 set_debugreg(next->debugreg6, 6);
 562                 set_debugreg(next->debugreg7, 7);
 563         }
 564
 565         return prev_p;
 566 }
 567
 568 /*
 569  * sys_execve() executes a new program.
 570  */
 571 asmlinkage
 572 long sys_execve(char __user *name, char __user * __user *argv,
 573                 char __user * __user *envp, struct pt_regs regs)
 574 {
 575         long error;
 576         char * filename;
 577
 578         filename = getname(name);
 579         error = PTR_ERR(filename);
 580         if (IS_ERR(filename))
 581                 return error;
 582         error = do_execve(filename, argv, envp, &regs);
 583         putname(filename);
 584         return error;
 585 }
 586
 587 void set_personality_64bit(void)
 588 {
 589         /* inherit personality from parent */
 590
 591         /* Make sure to be in 64bit mode */
 592         clear_thread_flag(TIF_IA32);
 593 }
 594
 595 asmlinkage long sys_fork(struct pt_regs *regs)
 596 {
 597         return do_fork(SIGCHLD, regs->rsp, regs, 0, NULL, NULL);
 598 }
 599
 600 asmlinkage long
 601 sys_clone(unsigned long clone_flags, unsigned long newsp,
 602           void __user *parent_tid, void __user *child_tid, struct pt_regs *regs)
 603 {
 604         if (!newsp)
 605                 newsp = regs->rsp;
 606         return do_fork(clone_flags, newsp, regs, 0, parent_tid, child_tid);
 607 }
 608
 609 /*
 610  * This is trivial, and on the face of it looks like it
 611  * could equally well be done in user mode.
 612  *
 613  * Not so, for quite unobvious reasons - register pressure.
 614  * In user mode vfork() cannot have a stack frame, and if
 615  * done by calling the "clone()" system call directly, you
 616  * do not have enough call-clobbered registers to hold all
 617  * the information you need.
 618  */
 619 asmlinkage long sys_vfork(struct pt_regs *regs)
 620 {
 621         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->rsp, regs, 0,
 622                     NULL, NULL);
 623 }
 624
 625 unsigned long get_wchan(struct task_struct *p)
 626 {
 627         unsigned long stack;
 628         u64 fp,rip;
 629         int count = 0;
 630
 631         if (!p || p == current || p->state==TASK_RUNNING)
 632                 return 0;
 633         stack = (unsigned long)task_stack_page(p);
 634         if (p->thread.rsp < stack || p->thread.rsp > stack+THREAD_SIZE)
 635                 return 0;
 636         fp = *(u64 *)(p->thread.rsp);
 637         do {
 638                 if (fp < (unsigned long)stack ||
 639                     fp > (unsigned long)stack+THREAD_SIZE)
 640                         return 0;
 641                 rip = *(u64 *)(fp+8);
 642                 if (!in_sched_functions(rip))
 643                         return rip;
 644                 fp = *(u64 *)fp;
 645         } while (count++ < 16);
 646         return 0;
 647 }
 648
 649 long do_arch_prctl(struct task_struct *task, int code, unsigned long addr)
 650 {
 651         int ret = 0;
 652         int doit = task == current;
 653         int cpu;
 654
 655         switch (code) {
 656         case ARCH_SET_GS:
 657                 if (addr >= TASK_SIZE_OF(task))
 658                         return -EPERM;
 659                 cpu = get_cpu();
 660                 /* handle small bases via the GDT because that's faster to
 661                    switch. */
 662                 if (addr <= 0xffffffff) {
 663                         set_32bit_tls(task, GS_TLS, addr);
 664                         if (doit) {
 665                                 load_TLS(&task->thread, cpu);
 666                                 load_gs_index(GS_TLS_SEL);
 667                         }
 668                         task->thread.gsindex = GS_TLS_SEL;
 669                         task->thread.gs = 0;
 670                 } else {
 671                         task->thread.gsindex = 0;
 672                         task->thread.gs = addr;
 673                         if (doit) {
 674                                 load_gs_index(0);
 675                                 ret = HYPERVISOR_set_segment_base(
 676                                         SEGBASE_GS_USER, addr);
 677                         }
 678                 }
 679                 put_cpu();
 680                 break;
 681         case ARCH_SET_FS:
 682                 /* Not strictly needed for fs, but do it for symmetry
 683                    with gs */
 684                 if (addr >= TASK_SIZE_OF(task))
 685                         return -EPERM;
 686                 cpu = get_cpu();
 687                 /* handle small bases via the GDT because that's faster to
 688                    switch. */
 689                 if (addr <= 0xffffffff) {
 690                         set_32bit_tls(task, FS_TLS, addr);
 691                         if (doit) {
 692                                 load_TLS(&task->thread, cpu);
 693                                 asm volatile("movl %0,%%fs" :: "r"(FS_TLS_SEL));
 694                         }
 695                         task->thread.fsindex = FS_TLS_SEL;
 696                         task->thread.fs = 0;
 697                 } else {
 698                         task->thread.fsindex = 0;
 699                         task->thread.fs = addr;
 700                         if (doit) {
 701                                 /* set the selector to 0 to not confuse
 702                                    __switch_to */
 703                                 asm volatile("movl %0,%%fs" :: "r" (0));
 704                                 ret = HYPERVISOR_set_segment_base(SEGBASE_FS,
 705                                                                   addr);
 706                         }
 707                 }
 708                 put_cpu();
 709                 break;
 710         case ARCH_GET_FS: {
 711                 unsigned long base;
 712                 if (task->thread.fsindex == FS_TLS_SEL)
 713                         base = read_32bit_tls(task, FS_TLS);
 714                 else if (doit)
 715                         rdmsrl(MSR_FS_BASE, base);
 716                 else
 717                         base = task->thread.fs;
 718                 ret = put_user(base, (unsigned long __user *)addr);
 719                 break;
 720         }
 721         case ARCH_GET_GS: {
 722                 unsigned long base;
 723                 unsigned gsindex;
 724                 if (task->thread.gsindex == GS_TLS_SEL)
 725                         base = read_32bit_tls(task, GS_TLS);
 726                 else if (doit) {
 727                         asm("movl %%gs,%0" : "=r" (gsindex));
 728                         if (gsindex)
 729                                 rdmsrl(MSR_KERNEL_GS_BASE, base);
 730                         else
 731                                 base = task->thread.gs;
 732                 }
 733                 else
 734                         base = task->thread.gs;
 735                 ret = put_user(base, (unsigned long __user *)addr);
 736                 break;
 737         }
 738
 739         default:
 740                 ret = -EINVAL;
 741                 break;
 742         }
 743
 744         return ret;
 745 }
 746
 747 long sys_arch_prctl(int code, unsigned long addr)
 748 {
 749         return do_arch_prctl(current, code, addr);
 750 }
 751
 752 /*
 753  * Capture the user space registers if the task is not running (in user space)
 754  */
 755 int dump_task_regs(struct task_struct *tsk, elf_gregset_t *regs)
 756 {
 757         struct pt_regs *pp, ptregs;
 758
 759         pp = task_pt_regs(tsk);
 760
 761         ptregs = *pp;
 762         ptregs.cs &= 0xffff;
 763         ptregs.ss &= 0xffff;
 764
 765         elf_core_copy_regs(regs, &ptregs);
 766
 767         boot_option_idle_override = 1;
 768         return 1;
 769 }
 770
 771 unsigned long arch_align_stack(unsigned long sp)
 772 {
 773         if (randomize_va_space)
 774                 sp -= get_random_int() % 8192;
 775         return sp & ~0xf;
 776 }
 777
 778 #ifndef CONFIG_SMP
 779 void _restore_vcpu(void)
 780 {
 781 }
 782 #endif