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 #include <linux/notifier.h>
  40
  41 #include <asm/uaccess.h>
  42 #include <asm/pgtable.h>
  43 #include <asm/system.h>
  44 #include <asm/io.h>
  45 #include <asm/processor.h>
  46 #include <asm/i387.h>
  47 #include <asm/mmu_context.h>
  48 #include <asm/pda.h>
  49 #include <asm/prctl.h>
  50 #include <asm/kdebug.h>
  51 #include <asm/desc.h>
  52 #include <asm/proto.h>
  53 #include <asm/ia32.h>
  54 #include <asm/idle.h>
  55
  56 asmlinkage extern void ret_from_fork(void);
  57
  58 unsigned long kernel_thread_flags = CLONE_VM | CLONE_UNTRACED;
  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 static struct notifier_block *idle_notifier;
  70 static DEFINE_SPINLOCK(idle_notifier_lock);
  71
  72 void idle_notifier_register(struct notifier_block *n)
  73 {
  74         unsigned long flags;
  75         spin_lock_irqsave(&idle_notifier_lock, flags);
  76         notifier_chain_register(&idle_notifier, n);
  77         spin_unlock_irqrestore(&idle_notifier_lock, flags);
  78 }
  79 EXPORT_SYMBOL_GPL(idle_notifier_register);
  80
  81 void idle_notifier_unregister(struct notifier_block *n)
  82 {
  83         unsigned long flags;
  84         spin_lock_irqsave(&idle_notifier_lock, flags);
  85         notifier_chain_unregister(&idle_notifier, n);
  86         spin_unlock_irqrestore(&idle_notifier_lock, flags);
  87 }
  88 EXPORT_SYMBOL(idle_notifier_unregister);
  89
  90 enum idle_state { CPU_IDLE, CPU_NOT_IDLE };
  91 static DEFINE_PER_CPU(enum idle_state, idle_state) = CPU_NOT_IDLE;
  92
  93 void enter_idle(void)
  94 {
  95         __get_cpu_var(idle_state) = CPU_IDLE;
  96         notifier_call_chain(&idle_notifier, IDLE_START, NULL);
  97 }
  98
  99 static void __exit_idle(void)
 100 {
 101         __get_cpu_var(idle_state) = CPU_NOT_IDLE;
 102         notifier_call_chain(&idle_notifier, IDLE_END, NULL);
 103 }
 104
 105 /* Called from interrupts to signify idle end */
 106 void exit_idle(void)
 107 {
 108         if (current->pid | read_pda(irqcount))
 109                 return;
 110         __exit_idle();
 111 }
 112
 113 /*
 114  * We use this if we don't have any better
 115  * idle routine..
 116  */
 117 void default_idle(void)
 118 {
 119         local_irq_enable();
 120
 121         clear_thread_flag(TIF_POLLING_NRFLAG);
 122         smp_mb__after_clear_bit();
 123         while (!need_resched()) {
 124                 local_irq_disable();
 125                 if (!need_resched())
 126                         safe_halt();
 127                 else
 128                         local_irq_enable();
 129         }
 130         set_thread_flag(TIF_POLLING_NRFLAG);
 131 }
 132
 133 /*
 134  * On SMP it's slightly faster (but much more power-consuming!)
 135  * to poll the ->need_resched flag instead of waiting for the
 136  * cross-CPU IPI to arrive. Use this option with caution.
 137  */
 138 static void poll_idle (void)
 139 {
 140         local_irq_enable();
 141
 142         asm volatile(
 143                 "2:"
 144                 "testl %0,%1;"
 145                 "rep; nop;"
 146                 "je 2b;"
 147                 : :
 148                 "i" (_TIF_NEED_RESCHED),
 149                 "m" (current_thread_info()->flags));
 150 }
 151
 152 void cpu_idle_wait(void)
 153 {
 154         unsigned int cpu, this_cpu = get_cpu();
 155         cpumask_t map;
 156
 157         set_cpus_allowed(current, cpumask_of_cpu(this_cpu));
 158         put_cpu();
 159
 160         cpus_clear(map);
 161         for_each_online_cpu(cpu) {
 162                 per_cpu(cpu_idle_state, cpu) = 1;
 163                 cpu_set(cpu, map);
 164         }
 165
 166         __get_cpu_var(cpu_idle_state) = 0;
 167
 168         wmb();
 169         do {
 170                 ssleep(1);
 171                 for_each_online_cpu(cpu) {
 172                         if (cpu_isset(cpu, map) &&
 173                                         !per_cpu(cpu_idle_state, cpu))
 174                                 cpu_clear(cpu, map);
 175                 }
 176                 cpus_and(map, map, cpu_online_map);
 177         } while (!cpus_empty(map));
 178 }
 179 EXPORT_SYMBOL_GPL(cpu_idle_wait);
 180
 181 #ifdef CONFIG_HOTPLUG_CPU
 182 DECLARE_PER_CPU(int, cpu_state);
 183
 184 #include <asm/nmi.h>
 185 /* We halt the CPU with physical CPU hotplug */
 186 static inline void play_dead(void)
 187 {
 188         idle_task_exit();
 189         wbinvd();
 190         mb();
 191         /* Ack it */
 192         __get_cpu_var(cpu_state) = CPU_DEAD;
 193
 194         local_irq_disable();
 195         while (1)
 196                 halt();
 197 }
 198 #else
 199 static inline void play_dead(void)
 200 {
 201         BUG();
 202 }
 203 #endif /* CONFIG_HOTPLUG_CPU */
 204
 205 /*
 206  * The idle thread. There's no useful work to be
 207  * done, so just try to conserve power and have a
 208  * low exit latency (ie sit in a loop waiting for
 209  * somebody to say that they'd like to reschedule)
 210  */
 211 void cpu_idle (void)
 212 {
 213         set_thread_flag(TIF_POLLING_NRFLAG);
 214
 215         /* endless idle loop with no priority at all */
 216         while (1) {
 217                 while (!need_resched()) {
 218                         void (*idle)(void);
 219
 220                         if (__get_cpu_var(cpu_idle_state))
 221                                 __get_cpu_var(cpu_idle_state) = 0;
 222
 223                         rmb();
 224                         idle = pm_idle;
 225                         if (!idle)
 226                                 idle = default_idle;
 227                         if (cpu_is_offline(smp_processor_id()))
 228                                 play_dead();
 229                         enter_idle();
 230                         idle();
 231                         __exit_idle();
 232                 }
 233
 234                 preempt_enable_no_resched();
 235                 schedule();
 236                 preempt_disable();
 237         }
 238 }
 239
 240 /*
 241  * This uses new MONITOR/MWAIT instructions on P4 processors with PNI,
 242  * which can obviate IPI to trigger checking of need_resched.
 243  * We execute MONITOR against need_resched and enter optimized wait state
 244  * through MWAIT. Whenever someone changes need_resched, we would be woken
 245  * up from MWAIT (without an IPI).
 246  */
 247 static void mwait_idle(void)
 248 {
 249         local_irq_enable();
 250
 251         while (!need_resched()) {
 252                 __monitor((void *)&current_thread_info()->flags, 0, 0);
 253                 smp_mb();
 254                 if (need_resched())
 255                         break;
 256                 __mwait(0, 0);
 257         }
 258 }
 259
 260 void __cpuinit select_idle_routine(const struct cpuinfo_x86 *c)
 261 {
 262         static int printed;
 263         if (cpu_has(c, X86_FEATURE_MWAIT)) {
 264                 /*
 265                  * Skip, if setup has overridden idle.
 266                  * One CPU supports mwait => All CPUs supports mwait
 267                  */
 268                 if (!pm_idle) {
 269                         if (!printed) {
 270                                 printk("using mwait in idle threads.\n");
 271                                 printed = 1;
 272                         }
 273                         pm_idle = mwait_idle;
 274                 }
 275         }
 276 }
 277
 278 static int __init idle_setup (char *str)
 279 {
 280         if (!strncmp(str, "poll", 4)) {
 281                 printk("using polling idle threads.\n");
 282                 pm_idle = poll_idle;
 283         }
 284
 285         boot_option_idle_override = 1;
 286         return 1;
 287 }
 288
 289 __setup("idle=", idle_setup);
 290
 291 /* Prints also some state that isn't saved in the pt_regs */
 292 void __show_regs(struct pt_regs * regs)
 293 {
 294         unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L, fs, gs, shadowgs;
 295         unsigned int fsindex,gsindex;
 296         unsigned int ds,cs,es;
 297
 298         printk("\n");
 299         print_modules();
 300         printk("Pid: %d, comm: %.20s %s %s %.*s\n",
 301                 current->pid, current->comm, print_tainted(),
 302                 system_utsname.release,
 303                 (int)strcspn(system_utsname.version, " "),
 304                 system_utsname.version);
 305         printk("RIP: %04lx:[<%016lx>] ", regs->cs & 0xffff, regs->rip);
 306         printk_address(regs->rip);
 307         printk("\nRSP: %04lx:%016lx  EFLAGS: %08lx\n", regs->ss, regs->rsp,
 308                 regs->eflags);
 309         printk("RAX: %016lx RBX: %016lx RCX: %016lx\n",
 310                regs->rax, regs->rbx, regs->rcx);
 311         printk("RDX: %016lx RSI: %016lx RDI: %016lx\n",
 312                regs->rdx, regs->rsi, regs->rdi);
 313         printk("RBP: %016lx R08: %016lx R09: %016lx\n",
 314                regs->rbp, regs->r8, regs->r9);
 315         printk("R10: %016lx R11: %016lx R12: %016lx\n",
 316                regs->r10, regs->r11, regs->r12);
 317         printk("R13: %016lx R14: %016lx R15: %016lx\n",
 318                regs->r13, regs->r14, regs->r15);
 319
 320         asm("movl %%ds,%0" : "=r" (ds));
 321         asm("movl %%cs,%0" : "=r" (cs));
 322         asm("movl %%es,%0" : "=r" (es));
 323         asm("movl %%fs,%0" : "=r" (fsindex));
 324         asm("movl %%gs,%0" : "=r" (gsindex));
 325
 326         rdmsrl(MSR_FS_BASE, fs);
 327         rdmsrl(MSR_GS_BASE, gs);
 328         rdmsrl(MSR_KERNEL_GS_BASE, shadowgs);
 329
 330         asm("movq %%cr0, %0": "=r" (cr0));
 331         asm("movq %%cr2, %0": "=r" (cr2));
 332         asm("movq %%cr3, %0": "=r" (cr3));
 333         asm("movq %%cr4, %0": "=r" (cr4));
 334
 335         printk("FS:  %016lx(%04x) GS:%016lx(%04x) knlGS:%016lx\n",
 336                fs,fsindex,gs,gsindex,shadowgs);
 337         printk("CS:  %04x DS: %04x ES: %04x CR0: %016lx\n", cs, ds, es, cr0);
 338         printk("CR2: %016lx CR3: %016lx CR4: %016lx\n", cr2, cr3, cr4);
 339 }
 340
 341 void show_regs(struct pt_regs *regs)
 342 {
 343         printk("CPU %d:", smp_processor_id());
 344         __show_regs(regs);
 345         show_trace(&regs->rsp);
 346 }
 347
 348 /*
 349  * Free current thread data structures etc..
 350  */
 351 void exit_thread(void)
 352 {
 353         struct task_struct *me = current;
 354         struct thread_struct *t = &me->thread;
 355
 356         /*
 357          * Remove function-return probe instances associated with this task
 358          * and put them back on the free list. Do not insert an exit probe for
 359          * this function, it will be disabled by kprobe_flush_task if you do.
 360          */
 361         kprobe_flush_task(me);
 362
 363         if (me->thread.io_bitmap_ptr) {
 364                 struct tss_struct *tss = &per_cpu(init_tss, get_cpu());
 365
 366                 kfree(t->io_bitmap_ptr);
 367                 t->io_bitmap_ptr = NULL;
 368                 /*
 369                  * Careful, clear this in the TSS too:
 370                  */
 371                 memset(tss->io_bitmap, 0xff, t->io_bitmap_max);
 372                 t->io_bitmap_max = 0;
 373                 put_cpu();
 374         }
 375 }
 376
 377 void flush_thread(void)
 378 {
 379         struct task_struct *tsk = current;
 380         struct thread_info *t = current_thread_info();
 381
 382         if (t->flags & _TIF_ABI_PENDING)
 383                 t->flags ^= (_TIF_ABI_PENDING | _TIF_IA32);
 384
 385         tsk->thread.debugreg0 = 0;
 386         tsk->thread.debugreg1 = 0;
 387         tsk->thread.debugreg2 = 0;
 388         tsk->thread.debugreg3 = 0;
 389         tsk->thread.debugreg6 = 0;
 390         tsk->thread.debugreg7 = 0;
 391         memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
 392         /*
 393          * Forget coprocessor state..
 394          */
 395         clear_fpu(tsk);
 396         clear_used_math();
 397 }
 398
 399 void release_thread(struct task_struct *dead_task)
 400 {
 401         if (dead_task->mm) {
 402                 if (dead_task->mm->context.size) {
 403                         printk("WARNING: dead process %8s still has LDT? <%p/%d>\n",
 404                                         dead_task->comm,
 405                                         dead_task->mm->context.ldt,
 406                                         dead_task->mm->context.size);
 407                         BUG();
 408                 }
 409         }
 410 }
 411
 412 static inline void set_32bit_tls(struct task_struct *t, int tls, u32 addr)
 413 {
 414         struct user_desc ud = {
 415                 .base_addr = addr,
 416                 .limit = 0xfffff,
 417                 .seg_32bit = 1,
 418                 .limit_in_pages = 1,
 419                 .useable = 1,
 420         };
 421         struct n_desc_struct *desc = (void *)t->thread.tls_array;
 422         desc += tls;
 423         desc->a = LDT_entry_a(&ud);
 424         desc->b = LDT_entry_b(&ud);
 425 }
 426
 427 static inline u32 read_32bit_tls(struct task_struct *t, int tls)
 428 {
 429         struct desc_struct *desc = (void *)t->thread.tls_array;
 430         desc += tls;
 431         return desc->base0 |
 432                 (((u32)desc->base1) << 16) |
 433                 (((u32)desc->base2) << 24);
 434 }
 435
 436 /*
 437  * This gets called before we allocate a new thread and copy
 438  * the current task into it.
 439  */
 440 void prepare_to_copy(struct task_struct *tsk)
 441 {
 442         unlazy_fpu(tsk);
 443 }
 444
 445 int copy_thread(int nr, unsigned long clone_flags, unsigned long rsp,
 446                 unsigned long unused,
 447         struct task_struct * p, struct pt_regs * regs)
 448 {
 449         int err;
 450         struct pt_regs * childregs;
 451         struct task_struct *me = current;
 452
 453         childregs = ((struct pt_regs *)
 454                         (THREAD_SIZE + task_stack_page(p))) - 1;
 455         *childregs = *regs;
 456
 457         childregs->rax = 0;
 458         childregs->rsp = rsp;
 459         if (rsp == ~0UL)
 460                 childregs->rsp = (unsigned long)childregs;
 461
 462         p->thread.rsp = (unsigned long) childregs;
 463         p->thread.rsp0 = (unsigned long) (childregs+1);
 464         p->thread.userrsp = me->thread.userrsp;
 465
 466         set_tsk_thread_flag(p, TIF_FORK);
 467
 468         p->thread.fs = me->thread.fs;
 469         p->thread.gs = me->thread.gs;
 470
 471         asm("mov %%gs,%0" : "=m" (p->thread.gsindex));
 472         asm("mov %%fs,%0" : "=m" (p->thread.fsindex));
 473         asm("mov %%es,%0" : "=m" (p->thread.es));
 474         asm("mov %%ds,%0" : "=m" (p->thread.ds));
 475
 476         if (unlikely(me->thread.io_bitmap_ptr != NULL)) {
 477                 p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL);
 478                 if (!p->thread.io_bitmap_ptr) {
 479                         p->thread.io_bitmap_max = 0;
 480                         return -ENOMEM;
 481                 }
 482                 memcpy(p->thread.io_bitmap_ptr, me->thread.io_bitmap_ptr,
 483                                 IO_BITMAP_BYTES);
 484         }
 485
 486         /*
 487          * Set a new TLS for the child thread?
 488          */
 489         if (clone_flags & CLONE_SETTLS) {
 490 #ifdef CONFIG_IA32_EMULATION
 491                 if (test_thread_flag(TIF_IA32))
 492                         err = ia32_child_tls(p, childregs);
 493                 else
 494 #endif
 495                         err = do_arch_prctl(p, ARCH_SET_FS, childregs->r8);
 496                 if (err)
 497                         goto out;
 498         }
 499         err = 0;
 500 out:
 501         if (err && p->thread.io_bitmap_ptr) {
 502                 kfree(p->thread.io_bitmap_ptr);
 503                 p->thread.io_bitmap_max = 0;
 504         }
 505         return err;
 506 }
 507
 508 /*
 509  * This special macro can be used to load a debugging register
 510  */
 511 #define loaddebug(thread,r) set_debug(thread->debugreg ## r, r)
 512
 513 /*
 514  *      switch_to(x,y) should switch tasks from x to y.
 515  *
 516  * This could still be optimized:
 517  * - fold all the options into a flag word and test it with a single test.
 518  * - could test fs/gs bitsliced
 519  */
 520 struct task_struct *
 521 __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
 522 {
 523         struct thread_struct *prev = &prev_p->thread,
 524                                  *next = &next_p->thread;
 525         int cpu = smp_processor_id();
 526         struct tss_struct *tss = &per_cpu(init_tss, cpu);
 527
 528         unlazy_fpu(prev_p);
 529
 530         /*
 531          * Reload esp0, LDT and the page table pointer:
 532          */
 533         tss->rsp0 = next->rsp0;
 534
 535         /*
 536          * Switch DS and ES.
 537          * This won't pick up thread selector changes, but I guess that is ok.
 538          */
 539         asm volatile("mov %%es,%0" : "=m" (prev->es));
 540         if (unlikely(next->es | prev->es))
 541                 loadsegment(es, next->es);
 542
 543         asm volatile ("mov %%ds,%0" : "=m" (prev->ds));
 544         if (unlikely(next->ds | prev->ds))
 545                 loadsegment(ds, next->ds);
 546
 547         load_TLS(next, cpu);
 548
 549         /*
 550          * Switch FS and GS.
 551          */
 552         {
 553                 unsigned fsindex;
 554                 asm volatile("movl %%fs,%0" : "=r" (fsindex));
 555                 /* segment register != 0 always requires a reload.
 556                    also reload when it has changed.
 557                    when prev process used 64bit base always reload
 558                    to avoid an information leak. */
 559                 if (unlikely(fsindex | next->fsindex | prev->fs)) {
 560                         loadsegment(fs, next->fsindex);
 561                         /* check if the user used a selector != 0
 562                          * if yes clear 64bit base, since overloaded base
 563                          * is always mapped to the Null selector
 564                          */
 565                         if (fsindex)
 566                         prev->fs = 0;
 567                 }
 568                 /* when next process has a 64bit base use it */
 569                 if (next->fs)
 570                         wrmsrl(MSR_FS_BASE, next->fs);
 571                 prev->fsindex = fsindex;
 572         }
 573         {
 574                 unsigned gsindex;
 575                 asm volatile("movl %%gs,%0" : "=r" (gsindex));
 576                 if (unlikely(gsindex | next->gsindex | prev->gs)) {
 577                         load_gs_index(next->gsindex);
 578                         if (gsindex)
 579                         prev->gs = 0;
 580                 }
 581                 if (next->gs)
 582                         wrmsrl(MSR_KERNEL_GS_BASE, next->gs);
 583                 prev->gsindex = gsindex;
 584         }
 585
 586         /*
 587          * Switch the PDA context.
 588          */
 589         prev->userrsp = read_pda(oldrsp);
 590         write_pda(oldrsp, next->userrsp);
 591         write_pda(pcurrent, next_p);
 592         write_pda(kernelstack,
 593                   task_stack_page(next_p) + THREAD_SIZE - PDA_STACKOFFSET);
 594
 595         /*
 596          * Now maybe reload the debug registers
 597          */
 598         if (unlikely(next->debugreg7)) {
 599                 loaddebug(next, 0);
 600                 loaddebug(next, 1);
 601                 loaddebug(next, 2);
 602                 loaddebug(next, 3);
 603                 /* no 4 and 5 */
 604                 loaddebug(next, 6);
 605                 loaddebug(next, 7);
 606         }
 607
 608
 609         /*
 610          * Handle the IO bitmap
 611          */
 612         if (unlikely(prev->io_bitmap_ptr || next->io_bitmap_ptr)) {
 613                 if (next->io_bitmap_ptr)
 614                         /*
 615                          * Copy the relevant range of the IO bitmap.
 616                          * Normally this is 128 bytes or less:
 617                          */
 618                         memcpy(tss->io_bitmap, next->io_bitmap_ptr,
 619                                 max(prev->io_bitmap_max, next->io_bitmap_max));
 620                 else {
 621                         /*
 622                          * Clear any possible leftover bits:
 623                          */
 624                         memset(tss->io_bitmap, 0xff, prev->io_bitmap_max);
 625                 }
 626         }
 627
 628         return prev_p;
 629 }
 630
 631 /*
 632  * sys_execve() executes a new program.
 633  */
 634 asmlinkage
 635 long sys_execve(char __user *name, char __user * __user *argv,
 636                 char __user * __user *envp, struct pt_regs regs)
 637 {
 638         long error;
 639         char * filename;
 640
 641         filename = getname(name);
 642         error = PTR_ERR(filename);
 643         if (IS_ERR(filename))
 644                 return error;
 645         error = do_execve(filename, argv, envp, &regs);
 646         if (error == 0) {
 647                 task_lock(current);
 648                 current->ptrace &= ~PT_DTRACE;
 649                 task_unlock(current);
 650         }
 651         putname(filename);
 652         return error;
 653 }
 654
 655 void set_personality_64bit(void)
 656 {
 657         /* inherit personality from parent */
 658
 659         /* Make sure to be in 64bit mode */
 660         clear_thread_flag(TIF_IA32);
 661
 662         /* TBD: overwrites user setup. Should have two bits.
 663            But 64bit processes have always behaved this way,
 664            so it's not too bad. The main problem is just that
 665            32bit childs are affected again. */
 666         current->personality &= ~READ_IMPLIES_EXEC;
 667 }
 668
 669 asmlinkage long sys_fork(struct pt_regs *regs)
 670 {
 671         return do_fork(SIGCHLD, regs->rsp, regs, 0, NULL, NULL);
 672 }
 673
 674 asmlinkage long
 675 sys_clone(unsigned long clone_flags, unsigned long newsp,
 676           void __user *parent_tid, void __user *child_tid, struct pt_regs *regs)
 677 {
 678         if (!newsp)
 679                 newsp = regs->rsp;
 680         return do_fork(clone_flags, newsp, regs, 0, parent_tid, child_tid);
 681 }
 682
 683 /*
 684  * This is trivial, and on the face of it looks like it
 685  * could equally well be done in user mode.
 686  *
 687  * Not so, for quite unobvious reasons - register pressure.
 688  * In user mode vfork() cannot have a stack frame, and if
 689  * done by calling the "clone()" system call directly, you
 690  * do not have enough call-clobbered registers to hold all
 691  * the information you need.
 692  */
 693 asmlinkage long sys_vfork(struct pt_regs *regs)
 694 {
 695         return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->rsp, regs, 0,
 696                     NULL, NULL);
 697 }
 698
 699 unsigned long get_wchan(struct task_struct *p)
 700 {
 701         unsigned long stack;
 702         u64 fp,rip;
 703         int count = 0;
 704
 705         if (!p || p == current || p->state==TASK_RUNNING)
 706                 return 0;
 707         stack = (unsigned long)task_stack_page(p);
 708         if (p->thread.rsp < stack || p->thread.rsp > stack+THREAD_SIZE)
 709                 return 0;
 710         fp = *(u64 *)(p->thread.rsp);
 711         do {
 712                 if (fp < (unsigned long)stack ||
 713                     fp > (unsigned long)stack+THREAD_SIZE)
 714                         return 0;
 715                 rip = *(u64 *)(fp+8);
 716                 if (!in_sched_functions(rip))
 717                         return rip;
 718                 fp = *(u64 *)fp;
 719         } while (count++ < 16);
 720         return 0;
 721 }
 722
 723 long do_arch_prctl(struct task_struct *task, int code, unsigned long addr)
 724 {
 725         int ret = 0;
 726         int doit = task == current;
 727         int cpu;
 728
 729         switch (code) {
 730         case ARCH_SET_GS:
 731                 if (addr >= TASK_SIZE_OF(task))
 732                         return -EPERM;
 733                 cpu = get_cpu();
 734                 /* handle small bases via the GDT because that's faster to
 735                    switch. */
 736                 if (addr <= 0xffffffff) {
 737                         set_32bit_tls(task, GS_TLS, addr);
 738                         if (doit) {
 739                                 load_TLS(&task->thread, cpu);
 740                                 load_gs_index(GS_TLS_SEL);
 741                         }
 742                         task->thread.gsindex = GS_TLS_SEL;
 743                         task->thread.gs = 0;
 744                 } else {
 745                         task->thread.gsindex = 0;
 746                         task->thread.gs = addr;
 747                         if (doit) {
 748                                 load_gs_index(0);
 749                                 ret = checking_wrmsrl(MSR_KERNEL_GS_BASE, addr);
 750                         }
 751                 }
 752                 put_cpu();
 753                 break;
 754         case ARCH_SET_FS:
 755                 /* Not strictly needed for fs, but do it for symmetry
 756                    with gs */
 757                 if (addr >= TASK_SIZE_OF(task))
 758                         return -EPERM;
 759                 cpu = get_cpu();
 760                 /* handle small bases via the GDT because that's faster to
 761                    switch. */
 762                 if (addr <= 0xffffffff) {
 763                         set_32bit_tls(task, FS_TLS, addr);
 764                         if (doit) {
 765                                 load_TLS(&task->thread, cpu);
 766                                 asm volatile("movl %0,%%fs" :: "r"(FS_TLS_SEL));
 767                         }
 768                         task->thread.fsindex = FS_TLS_SEL;
 769                         task->thread.fs = 0;
 770                 } else {
 771                         task->thread.fsindex = 0;
 772                         task->thread.fs = addr;
 773                         if (doit) {
 774                                 /* set the selector to 0 to not confuse
 775                                    __switch_to */
 776                                 asm volatile("movl %0,%%fs" :: "r" (0));
 777                                 ret = checking_wrmsrl(MSR_FS_BASE, addr);
 778                         }
 779                 }
 780                 put_cpu();
 781                 break;
 782         case ARCH_GET_FS: {
 783                 unsigned long base;
 784                 if (task->thread.fsindex == FS_TLS_SEL)
 785                         base = read_32bit_tls(task, FS_TLS);
 786                 else if (doit)
 787                         rdmsrl(MSR_FS_BASE, base);
 788                 else
 789                         base = task->thread.fs;
 790                 ret = put_user(base, (unsigned long __user *)addr);
 791                 break;
 792         }
 793         case ARCH_GET_GS: {
 794                 unsigned long base;
 795                 if (task->thread.gsindex == GS_TLS_SEL)
 796                         base = read_32bit_tls(task, GS_TLS);
 797                 else if (doit)
 798                         rdmsrl(MSR_KERNEL_GS_BASE, base);
 799                 else
 800                         base = task->thread.gs;
 801                 ret = put_user(base, (unsigned long __user *)addr);
 802                 break;
 803         }
 804
 805         default:
 806                 ret = -EINVAL;
 807                 break;
 808         }
 809
 810         return ret;
 811 }
 812
 813 long sys_arch_prctl(int code, unsigned long addr)
 814 {
 815         return do_arch_prctl(current, code, addr);
 816 }
 817
 818 /*
 819  * Capture the user space registers if the task is not running (in user space)
 820  */
 821 int dump_task_regs(struct task_struct *tsk, elf_gregset_t *regs)
 822 {
 823         struct pt_regs *pp, ptregs;
 824
 825         pp = task_pt_regs(tsk);
 826
 827         ptregs = *pp;
 828         ptregs.cs &= 0xffff;
 829         ptregs.ss &= 0xffff;
 830
 831         elf_core_copy_regs(regs, &ptregs);
 832
 833         return 1;
 834 }
 835
 836 unsigned long arch_align_stack(unsigned long sp)
 837 {
 838         if (randomize_va_space)
 839                 sp -= get_random_int() % 8192;
 840         return sp & ~0xf;
 841 }