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  *  $Id: process.c,v 1.38 2002/01/15 10:08:03 ak Exp $
  13  */
  14
  15 /*
  16  * This file handles the architecture-dependent parts of process handling..
  17  */
  18
  19 #include <stdarg.h>
  20
  21 #include <linux/errno.h>
  22 #include <linux/sched.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/module.h>
  30 #include <linux/a.out.h>
  31 #include <linux/interrupt.h>
  32 #include <linux/delay.h>
  33 #include <linux/irq.h>
  34 #include <linux/ptrace.h>
  35 #include <linux/utsname.h>
  36 #include <linux/random.h>
  37
  38 #include <asm/uaccess.h>
  39 #include <asm/pgtable.h>
  40 #include <asm/system.h>
  41 #include <asm/io.h>
  42 #include <asm/processor.h>
  43 #include <asm/i387.h>
  44 #include <asm/mmu_context.h>
  45 #include <asm/pda.h>
  46 #include <asm/prctl.h>
  47 #include <asm/kdebug.h>
  48 #include <asm/desc.h>
  49 #include <asm/proto.h>
  50 #include <asm/ia32.h>
  51
  52 asmlinkage extern void ret_from_fork(void);
  53
  54 unsigned long kernel_thread_flags = CLONE_VM | CLONE_UNTRACED;
  55
  56 static atomic_t hlt_counter = ATOMIC_INIT(0);
  57
  58 unsigned long boot_option_idle_override = 0;
  59 EXPORT_SYMBOL(boot_option_idle_override);
  60
  61 /*
  62  * Powermanagement idle function, if any..
  63  */
  64 void (*pm_idle)(void);
  65 static DEFINE_PER_CPU(unsigned int, cpu_idle_state);
  66
  67 void disable_hlt(void)
  68 {
  69         atomic_inc(&hlt_counter);
  70 }
  71
  72 EXPORT_SYMBOL(disable_hlt);
  73
  74 void enable_hlt(void)
  75 {
  76         atomic_dec(&hlt_counter);
  77 }
  78
  79 EXPORT_SYMBOL(enable_hlt);
  80
  81 /*
  82  * We use this if we don't have any better
  83  * idle routine..
  84  */
  85 void default_idle(void)
  86 {
  87         if (!atomic_read(&hlt_counter)) {
  88                 local_irq_disable();
  89                 if (!need_resched())
  90                         safe_halt();
  91                 else
  92                         local_irq_enable();
  93         }
  94 }
  95
  96 /*
  97  * On SMP it's slightly faster (but much more power-consuming!)
  98  * to poll the ->need_resched flag instead of waiting for the
  99  * cross-CPU IPI to arrive. Use this option with caution.
 100  */
 101 static void poll_idle (void)
 102 {
 103         int oldval;
 104
 105         local_irq_enable();
 106
 107         /*
 108          * Deal with another CPU just having chosen a thread to
 109          * run here:
 110          */
 111         oldval = test_and_clear_thread_flag(TIF_NEED_RESCHED);
 112
 113         if (!oldval) {
 114                 set_thread_flag(TIF_POLLING_NRFLAG);
 115                 asm volatile(
 116                         "2:"
 117                         "testl %0,%1;"
 118                         "rep; nop;"
 119                         "je 2b;"
 120                         : :
 121                         "i" (_TIF_NEED_RESCHED),
 122                         "m" (current_thread_info()->flags));
 123         } else {
 124                 set_need_resched();
 125         }
 126 }
 127
 128 void cpu_idle_wait(void)
 129 {
 130         unsigned int cpu, this_cpu = get_cpu();
 131         cpumask_t map;
 132
 133         set_cpus_allowed(current, cpumask_of_cpu(this_cpu));
 134         put_cpu();
 135
 136         cpus_clear(map);
 137         for_each_online_cpu(cpu) {
 138                 per_cpu(cpu_idle_state, cpu) = 1;
 139                 cpu_set(cpu, map);
 140         }
 141
 142         __get_cpu_var(cpu_idle_state) = 0;
 143
 144         wmb();
 145         do {
 146                 ssleep(1);
 147                 for_each_online_cpu(cpu) {
 148                         if (cpu_isset(cpu, map) && !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 /*
 157  * The idle thread. There's no useful work to be
 158  * done, so just try to conserve power and have a
 159  * low exit latency (ie sit in a loop waiting for
 160  * somebody to say that they'd like to reschedule)
 161  */
 162 void cpu_idle (void)
 163 {
 164         /* endless idle loop with no priority at all */
 165         while (1) {
 166                 while (!need_resched()) {
 167                         void (*idle)(void);
 168
 169                         if (__get_cpu_var(cpu_idle_state))
 170                                 __get_cpu_var(cpu_idle_state) = 0;
 171
 172                         rmb();
 173                         idle = pm_idle;
 174                         if (!idle)
 175                                 idle = default_idle;
 176                         idle();
 177                 }
 178
 179                 schedule();
 180         }
 181 }
 182
 183 /*
 184  * This uses new MONITOR/MWAIT instructions on P4 processors with PNI,
 185  * which can obviate IPI to trigger checking of need_resched.
 186  * We execute MONITOR against need_resched and enter optimized wait state
 187  * through MWAIT. Whenever someone changes need_resched, we would be woken
 188  * up from MWAIT (without an IPI).
 189  */
 190 static void mwait_idle(void)
 191 {
 192         local_irq_enable();
 193
 194         if (!need_resched()) {
 195                 set_thread_flag(TIF_POLLING_NRFLAG);
 196                 do {
 197                         __monitor((void *)&current_thread_info()->flags, 0, 0);
 198                         if (need_resched())
 199                                 break;
 200                         __mwait(0, 0);
 201                 } while (!need_resched());
 202                 clear_thread_flag(TIF_POLLING_NRFLAG);
 203         }
 204 }
 205
 206 void __init select_idle_routine(const struct cpuinfo_x86 *c)
 207 {
 208         static int printed;
 209         if (cpu_has(c, X86_FEATURE_MWAIT)) {
 210                 /*
 211                  * Skip, if setup has overridden idle.
 212                  * One CPU supports mwait => All CPUs supports mwait
 213                  */
 214                 if (!pm_idle) {
 215                         if (!printed) {
 216                                 printk("using mwait in idle threads.\n");
 217                                 printed = 1;
 218                         }
 219                         pm_idle = mwait_idle;
 220                 }
 221         }
 222 }
 223
 224 static int __init idle_setup (char *str)
 225 {
 226         if (!strncmp(str, "poll", 4)) {
 227                 printk("using polling idle threads.\n");
 228                 pm_idle = poll_idle;
 229         }
 230
 231         boot_option_idle_override = 1;
 232         return 1;
 233 }
 234
 235 __setup("idle=", idle_setup);
 236
 237 /* Prints also some state that isn't saved in the pt_regs */
 238 void __show_regs(struct pt_regs * regs)
 239 {
 240         unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L, fs, gs, shadowgs;
 241         unsigned int fsindex,gsindex;
 242         unsigned int ds,cs,es;
 243
 244         printk("\n");
 245         print_modules();
 246         printk("Pid: %d, comm: %.20s %s %s\n",
 247                current->pid, current->comm, print_tainted(), system_utsname.release);
 248         printk("RIP: %04lx:[<%016lx>] ", regs->cs & 0xffff, regs->rip);
 249         printk_address(regs->rip);
 250         printk("\nRSP: %04lx:%016lx  EFLAGS: %08lx\n", regs->ss, regs->rsp, regs->eflags);
 251         printk("RAX: %016lx RBX: %016lx RCX: %016lx\n",
 252                regs->rax, regs->rbx, regs->rcx);
 253         printk("RDX: %016lx RSI: %016lx RDI: %016lx\n",
 254                regs->rdx, regs->rsi, regs->rdi);
 255         printk("RBP: %016lx R08: %016lx R09: %016lx\n",
 256                regs->rbp, regs->r8, regs->r9);
 257         printk("R10: %016lx R11: %016lx R12: %016lx\n",
 258                regs->r10, regs->r11, regs->r12);
 259         printk("R13: %016lx R14: %016lx R15: %016lx\n",
 260                regs->r13, regs->r14, regs->r15);
 261
 262         asm("movl %%ds,%0" : "=r" (ds));
 263         asm("movl %%cs,%0" : "=r" (cs));
 264         asm("movl %%es,%0" : "=r" (es));
 265         asm("movl %%fs,%0" : "=r" (fsindex));
 266         asm("movl %%gs,%0" : "=r" (gsindex));
 267
 268         rdmsrl(MSR_FS_BASE, fs);
 269         rdmsrl(MSR_GS_BASE, gs);
 270         rdmsrl(MSR_KERNEL_GS_BASE, shadowgs);
 271
 272         asm("movq %%cr0, %0": "=r" (cr0));
 273         asm("movq %%cr2, %0": "=r" (cr2));
 274         asm("movq %%cr3, %0": "=r" (cr3));
 275         asm("movq %%cr4, %0": "=r" (cr4));
 276
 277         printk("FS:  %016lx(%04x) GS:%016lx(%04x) knlGS:%016lx\n",
 278                fs,fsindex,gs,gsindex,shadowgs);
 279         printk("CS:  %04x DS: %04x ES: %04x CR0: %016lx\n", cs, ds, es, cr0);
 280         printk("CR2: %016lx CR3: %016lx CR4: %016lx\n", cr2, cr3, cr4);
 281 }
 282
 283 void show_regs(struct pt_regs *regs)
 284 {
 285         __show_regs(regs);
 286         show_trace(&regs->rsp);
 287 }
 288
 289 EXPORT_SYMBOL_GPL(show_regs);
 290
 291 /*
 292  * Free current thread data structures etc..
 293  */
 294 void exit_thread(void)
 295 {
 296         struct task_struct *me = current;
 297         struct thread_struct *t = &me->thread;
 298         if (me->thread.io_bitmap_ptr) {
 299                 struct tss_struct *tss = &per_cpu(init_tss, get_cpu());
 300
 301                 kfree(t->io_bitmap_ptr);
 302                 t->io_bitmap_ptr = NULL;
 303                 /*
 304                  * Careful, clear this in the TSS too:
 305                  */
 306                 memset(tss->io_bitmap, 0xff, t->io_bitmap_max);
 307                 t->io_bitmap_max = 0;
 308                 put_cpu();
 309         }
 310 }
 311
 312 void flush_thread(void)
 313 {
 314         struct task_struct *tsk = current;
 315         struct thread_info *t = current_thread_info();
 316
 317         if (t->flags & _TIF_ABI_PENDING)
 318                 t->flags ^= (_TIF_ABI_PENDING | _TIF_IA32);
 319
 320         tsk->thread.debugreg0 = 0;
 321         tsk->thread.debugreg1 = 0;
 322         tsk->thread.debugreg2 = 0;
 323         tsk->thread.debugreg3 = 0;
 324         tsk->thread.debugreg6 = 0;
 325         tsk->thread.debugreg7 = 0;
 326         memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
 327         /*
 328          * Forget coprocessor state..
 329          */
 330         clear_fpu(tsk);
 331         clear_used_math();
 332 }
 333
 334 void release_thread(struct task_struct *dead_task)
 335 {
 336         if (dead_task->mm) {
 337                 if (dead_task->mm->context.size) {
 338                         printk("WARNING: dead process %8s still has LDT? <%p/%d>\n",
 339                                         dead_task->comm,
 340                                         dead_task->mm->context.ldt,
 341                                         dead_task->mm->context.size);
 342                         BUG();
 343                 }
 344         }
 345 }
 346
 347 static inline void set_32bit_tls(struct task_struct *t, int tls, u32 addr)
 348 {
 349         struct user_desc ud = {
 350                 .base_addr = addr,
 351                 .limit = 0xfffff,
 352                 .seg_32bit = 1,
 353                 .limit_in_pages = 1,
 354                 .useable = 1,
 355         };
 356         struct n_desc_struct *desc = (void *)t->thread.tls_array;
 357         desc += tls;
 358         desc->a = LDT_entry_a(&ud);
 359         desc->b = LDT_entry_b(&ud);
 360 }
 361
 362 static inline u32 read_32bit_tls(struct task_struct *t, int tls)
 363 {
 364         struct desc_struct *desc = (void *)t->thread.tls_array;
 365         desc += tls;
 366         return desc->base0 |
 367                 (((u32)desc->base1) << 16) |
 368                 (((u32)desc->base2) << 24);
 369 }
 370
 371 /*
 372  * This gets called before we allocate a new thread and copy
 373  * the current task into it.
 374  */
 375 void prepare_to_copy(struct task_struct *tsk)
 376 {
 377         unlazy_fpu(tsk);
 378 }
 379
 380 int copy_thread(int nr, unsigned long clone_flags, unsigned long rsp,
 381                 unsigned long unused,
 382         struct task_struct * p, struct pt_regs * regs)
 383 {
 384         int err;
 385         struct pt_regs * childregs;
 386         struct task_struct *me = current;
 387
 388         childregs = ((struct pt_regs *) (THREAD_SIZE + (unsigned long) p->thread_info)) - 1;
 389
 390         *childregs = *regs;
 391
 392         childregs->rax = 0;
 393         childregs->rsp = rsp;
 394         if (rsp == ~0UL) {
 395                 childregs->rsp = (unsigned long)childregs;
 396         }
 397
 398         p->thread.rsp = (unsigned long) childregs;
 399         p->thread.rsp0 = (unsigned long) (childregs+1);
 400         p->thread.userrsp = me->thread.userrsp;
 401
 402         set_ti_thread_flag(p->thread_info, TIF_FORK);
 403
 404         p->thread.fs = me->thread.fs;
 405         p->thread.gs = me->thread.gs;
 406
 407         asm("mov %%gs,%0" : "=m" (p->thread.gsindex));
 408         asm("mov %%fs,%0" : "=m" (p->thread.fsindex));
 409         asm("mov %%es,%0" : "=m" (p->thread.es));
 410         asm("mov %%ds,%0" : "=m" (p->thread.ds));
 411
 412         if (unlikely(me->thread.io_bitmap_ptr != NULL)) {
 413                 p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL);
 414                 if (!p->thread.io_bitmap_ptr) {
 415                         p->thread.io_bitmap_max = 0;
 416                         return -ENOMEM;
 417                 }
 418                 memcpy(p->thread.io_bitmap_ptr, me->thread.io_bitmap_ptr, IO_BITMAP_BYTES);
 419         }
 420
 421         /*
 422          * Set a new TLS for the child thread?
 423          */
 424         if (clone_flags & CLONE_SETTLS) {
 425 #ifdef CONFIG_IA32_EMULATION
 426                 if (test_thread_flag(TIF_IA32))
 427                         err = ia32_child_tls(p, childregs);
 428                 else
 429 #endif
 430                         err = do_arch_prctl(p, ARCH_SET_FS, childregs->r8);
 431                 if (err)
 432                         goto out;
 433         }
 434         err = 0;
 435 out:
 436         if (err && p->thread.io_bitmap_ptr) {
 437                 kfree(p->thread.io_bitmap_ptr);
 438                 p->thread.io_bitmap_max = 0;
 439         }
 440         return err;
 441 }
 442
 443 /*
 444  * This special macro can be used to load a debugging register
 445  */
 446 #define loaddebug(thread,r) set_debug(thread->debugreg ## r, r)
 447
 448 /*
 449  *      switch_to(x,y) should switch tasks from x to y.
 450  *
 451  * This could still be optimized:
 452  * - fold all the options into a flag word and test it with a single test.
 453  * - could test fs/gs bitsliced
 454  */
 455 struct task_struct *__switch_to(struct task_struct *prev_p, struct task_struct *next_p)
 456 {
 457         struct thread_struct *prev = &prev_p->thread,
 458                                  *next = &next_p->thread;
 459         int cpu = smp_processor_id();
 460         struct tss_struct *tss = &per_cpu(init_tss, cpu);
 461
 462         unlazy_fpu(prev_p);
 463
 464         /*
 465          * Reload esp0, LDT and the page table pointer:
 466          */
 467         tss->rsp0 = next->rsp0;
 468
 469         /*
 470          * Switch DS and ES.
 471          * This won't pick up thread selector changes, but I guess that is ok.
 472          */
 473         asm volatile("mov %%es,%0" : "=m" (prev->es));
 474         if (unlikely(next->es | prev->es))
 475                 loadsegment(es, next->es);
 476
 477         asm volatile ("mov %%ds,%0" : "=m" (prev->ds));
 478         if (unlikely(next->ds | prev->ds))
 479                 loadsegment(ds, next->ds);
 480
 481         load_TLS(next, cpu);
 482
 483         /*
 484          * Switch FS and GS.
 485          */
 486         {
 487                 unsigned fsindex;
 488                 asm volatile("movl %%fs,%0" : "=r" (fsindex));
 489                 /* segment register != 0 always requires a reload.
 490                    also reload when it has changed.
 491                    when prev process used 64bit base always reload
 492                    to avoid an information leak. */
 493                 if (unlikely(fsindex | next->fsindex | prev->fs)) {
 494                         loadsegment(fs, next->fsindex);
 495                         /* check if the user used a selector != 0
 496                          * if yes clear 64bit base, since overloaded base
 497                          * is always mapped to the Null selector
 498                          */
 499                         if (fsindex)
 500                         prev->fs = 0;
 501                 }
 502                 /* when next process has a 64bit base use it */
 503                 if (next->fs)
 504                         wrmsrl(MSR_FS_BASE, next->fs);
 505                 prev->fsindex = fsindex;
 506         }
 507         {
 508                 unsigned gsindex;
 509                 asm volatile("movl %%gs,%0" : "=r" (gsindex));
 510                 if (unlikely(gsindex | next->gsindex | prev->gs)) {
 511                         load_gs_index(next->gsindex);
 512                         if (gsindex)
 513                         prev->gs = 0;
 514                 }
 515                 if (next->gs)
 516                         wrmsrl(MSR_KERNEL_GS_BASE, next->gs);
 517                 prev->gsindex = gsindex;
 518         }
 519
 520         /*
 521          * Switch the PDA context.
 522          */
 523         prev->userrsp = read_pda(oldrsp);
 524         write_pda(oldrsp, next->userrsp);
 525         write_pda(pcurrent, next_p);
 526         write_pda(kernelstack, (unsigned long)next_p->thread_info + THREAD_SIZE - PDA_STACKOFFSET);
 527
 528         /*
 529          * Now maybe reload the debug registers
 530          */
 531         if (unlikely(next->debugreg7)) {
 532                 loaddebug(next, 0);
 533                 loaddebug(next, 1);
 534                 loaddebug(next, 2);
 535                 loaddebug(next, 3);
 536                 /* no 4 and 5 */
 537                 loaddebug(next, 6);
 538                 loaddebug(next, 7);
 539         }
 540
 541
 542         /*
 543          * Handle the IO bitmap
 544          */
 545         if (unlikely(prev->io_bitmap_ptr || next->io_bitmap_ptr)) {
 546                 if (next->io_bitmap_ptr)
 547                         /*
 548                          * Copy the relevant range of the IO bitmap.
 549                          * Normally this is 128 bytes or less:
 550                          */
 551                         memcpy(tss->io_bitmap, next->io_bitmap_ptr,
 552                                 max(prev->io_bitmap_max, next->io_bitmap_max));
 553                 else {
 554                         /*
 555                          * Clear any possible leftover bits:
 556                          */
 557                         memset(tss->io_bitmap, 0xff, prev->io_bitmap_max);
 558                 }
 559         }
 560
 561         return prev_p;
 562 }
 563
 564 /*
 565  * sys_execve() executes a new program.
 566  */
 567 asmlinkage
 568 long sys_execve(char __user *name, char __user * __user *argv,
 569                 char __user * __user *envp, struct pt_regs regs)
 570 {
 571         long error;
 572         char * filename;
 573
 574         filename = getname(name);
 575         error = PTR_ERR(filename);
 576         if (IS_ERR(filename))
 577                 return error;
 578         error = do_execve(filename, argv, envp, &regs);
 579         if (error == 0) {
 580                 task_lock(current);
 581                 current->ptrace &= ~PT_DTRACE;
 582                 task_unlock(current);
 583         }
 584         putname(filename);
 585         return error;
 586 }
 587
 588 void set_personality_64bit(void)
 589 {
 590         /* inherit personality from parent */
 591
 592         /* Make sure to be in 64bit mode */
 593         clear_thread_flag(TIF_IA32);
 594 }
 595
 596 asmlinkage long sys_fork(struct pt_regs *regs)
 597 {
 598         return do_fork(SIGCHLD, regs->rsp, regs, 0, NULL, NULL);
 599 }
 600
 601 asmlinkage long sys_clone(unsigned long clone_flags, unsigned long newsp, void __user *parent_tid, void __user *child_tid, struct pt_regs *regs)
 602 {
 603         if (!newsp)
 604                 newsp = regs->rsp;
 605         return do_fork(clone_flags, newsp, regs, 0, parent_tid, child_tid);
 606 }
 607
 608 /*
 609  * This is trivial, and on the face of it looks like it
 610  * could equally well be done in user mode.
 611  *
 612  * Not so, for quite unobvious reasons - register pressure.
 613  * In user mode vfork() cannot have a stack frame, and if
 614  * done by calling the "clone()" system call directly, you
 615  * do not have enough call-clobbered registers to hold all
 616  * the information you need.
 617  */
 618 asmlinkage long sys_vfork(struct pt_regs *regs)
 619 {
 620         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->rsp, regs, 0,
 621                     NULL, NULL);
 622 }
 623
 624 unsigned long get_wchan(struct task_struct *p)
 625 {
 626         unsigned long stack;
 627         u64 fp,rip;
 628         int count = 0;
 629
 630         if (!p || p == current || p->state==TASK_RUNNING)
 631                 return 0;
 632         stack = (unsigned long)p->thread_info;
 633         if (p->thread.rsp < stack || p->thread.rsp > stack+THREAD_SIZE)
 634                 return 0;
 635         fp = *(u64 *)(p->thread.rsp);
 636         do {
 637                 if (fp < (unsigned long)stack || fp > (unsigned long)stack+THREAD_SIZE)
 638                         return 0;
 639                 rip = *(u64 *)(fp+8);
 640                 if (!in_sched_functions(rip))
 641                         return rip;
 642                 fp = *(u64 *)fp;
 643         } while (count++ < 16);
 644         return 0;
 645 }
 646
 647 long do_arch_prctl(struct task_struct *task, int code, unsigned long addr)
 648 {
 649         int ret = 0;
 650         int doit = task == current;
 651         int cpu;
 652
 653         switch (code) {
 654         case ARCH_SET_GS:
 655                 if (addr >= TASK_SIZE)
 656                         return -EPERM;
 657                 cpu = get_cpu();
 658                 /* handle small bases via the GDT because that's faster to
 659                    switch. */
 660                 if (addr <= 0xffffffff) {
 661                         set_32bit_tls(task, GS_TLS, addr);
 662                         if (doit) {
 663                                 load_TLS(&task->thread, cpu);
 664                                 load_gs_index(GS_TLS_SEL);
 665                         }
 666                         task->thread.gsindex = GS_TLS_SEL;
 667                         task->thread.gs = 0;
 668                 } else {
 669                         task->thread.gsindex = 0;
 670                         task->thread.gs = addr;
 671                         if (doit) {
 672                 load_gs_index(0);
 673                 ret = checking_wrmsrl(MSR_KERNEL_GS_BASE, addr);
 674                         }
 675                 }
 676                 put_cpu();
 677                 break;
 678         case ARCH_SET_FS:
 679                 /* Not strictly needed for fs, but do it for symmetry
 680                    with gs */
 681                 if (addr >= TASK_SIZE)
 682                         return -EPERM;
 683                 cpu = get_cpu();
 684                 /* handle small bases via the GDT because that's faster to
 685                    switch. */
 686                 if (addr <= 0xffffffff) {
 687                         set_32bit_tls(task, FS_TLS, addr);
 688                         if (doit) {
 689                                 load_TLS(&task->thread, cpu);
 690                                 asm volatile("movl %0,%%fs" :: "r" (FS_TLS_SEL));
 691                         }
 692                         task->thread.fsindex = FS_TLS_SEL;
 693                         task->thread.fs = 0;
 694                 } else {
 695                         task->thread.fsindex = 0;
 696                         task->thread.fs = addr;
 697                         if (doit) {
 698                                 /* set the selector to 0 to not confuse
 699                                    __switch_to */
 700                 asm volatile("movl %0,%%fs" :: "r" (0));
 701                 ret = checking_wrmsrl(MSR_FS_BASE, addr);
 702                         }
 703                 }
 704                 put_cpu();
 705                 break;
 706         case ARCH_GET_FS: {
 707                 unsigned long base;
 708                 if (task->thread.fsindex == FS_TLS_SEL)
 709                         base = read_32bit_tls(task, FS_TLS);
 710                 else if (doit) {
 711                         rdmsrl(MSR_FS_BASE, base);
 712                 } else
 713                         base = task->thread.fs;
 714                 ret = put_user(base, (unsigned long __user *)addr);
 715                 break;
 716         }
 717         case ARCH_GET_GS: {
 718                 unsigned long base;
 719                 if (task->thread.gsindex == GS_TLS_SEL)
 720                         base = read_32bit_tls(task, GS_TLS);
 721                 else if (doit) {
 722                         rdmsrl(MSR_KERNEL_GS_BASE, base);
 723                 } else
 724                         base = task->thread.gs;
 725                 ret = put_user(base, (unsigned long __user *)addr);
 726                 break;
 727         }
 728
 729         default:
 730                 ret = -EINVAL;
 731                 break;
 732         }
 733
 734         return ret;
 735 }
 736
 737 long sys_arch_prctl(int code, unsigned long addr)
 738 {
 739         return do_arch_prctl(current, code, addr);
 740 }
 741
 742 /*
 743  * Capture the user space registers if the task is not running (in user space)
 744  */
 745 int dump_task_regs(struct task_struct *tsk, elf_gregset_t *regs)
 746 {
 747         struct pt_regs *pp, ptregs;
 748
 749         pp = (struct pt_regs *)(tsk->thread.rsp0);
 750         --pp;
 751
 752         ptregs = *pp;
 753         ptregs.cs &= 0xffff;
 754         ptregs.ss &= 0xffff;
 755
 756         elf_core_copy_regs(regs, &ptregs);
 757
 758         return 1;
 759 }