arch/x86/kernel/process_64.c

   1 /*
   2  *  Copyright (C) 1995  Linus Torvalds
   3  *
   4  *  Pentium III FXSR, SSE support
   5  *      Gareth Hughes <gareth@valinux.com>, May 2000
   6  *
   7  *  X86-64 port
   8  *      Andi Kleen.
   9  *
  10  *      CPU hotplug support - ashok.raj@intel.com
  11  */
  12
  13 /*
  14  * This file handles the architecture-dependent parts of process handling..
  15  */
  16
  17 #include <linux/stackprotector.h>
  18 #include <linux/cpu.h>
  19 #include <linux/errno.h>
  20 #include <linux/sched.h>
  21 #include <linux/fs.h>
  22 #include <linux/kernel.h>
  23 #include <linux/mm.h>
  24 #include <linux/elfcore.h>
  25 #include <linux/smp.h>
  26 #include <linux/slab.h>
  27 #include <linux/user.h>
  28 #include <linux/interrupt.h>
  29 #include <linux/delay.h>
  30 #include <linux/module.h>
  31 #include <linux/ptrace.h>
  32 #include <linux/notifier.h>
  33 #include <linux/kprobes.h>
  34 #include <linux/kdebug.h>
  35 #include <linux/tick.h>
  36 #include <linux/prctl.h>
  37 #include <linux/uaccess.h>
  38 #include <linux/io.h>
  39 #include <linux/ftrace.h>
  40
  41 #include <asm/pgtable.h>
  42 #include <asm/system.h>
  43 #include <asm/processor.h>
  44 #include <asm/i387.h>
  45 #include <asm/mmu_context.h>
  46 #include <asm/prctl.h>
  47 #include <asm/desc.h>
  48 #include <asm/proto.h>
  49 #include <asm/ia32.h>
  50 #include <asm/idle.h>
  51 #include <asm/syscalls.h>
  52 #include <asm/debugreg.h>
  53
  54 #include <trace/events/power.h>
  55
  56 asmlinkage extern void ret_from_fork(void);
  57
  58 DEFINE_PER_CPU(unsigned long, old_rsp);
  59 static DEFINE_PER_CPU(unsigned char, is_idle);
  60
  61 static ATOMIC_NOTIFIER_HEAD(idle_notifier);
  62
  63 void idle_notifier_register(struct notifier_block *n)
  64 {
  65         atomic_notifier_chain_register(&idle_notifier, n);
  66 }
  67 EXPORT_SYMBOL_GPL(idle_notifier_register);
  68
  69 void idle_notifier_unregister(struct notifier_block *n)
  70 {
  71         atomic_notifier_chain_unregister(&idle_notifier, n);
  72 }
  73 EXPORT_SYMBOL_GPL(idle_notifier_unregister);
  74
  75 void enter_idle(void)
  76 {
  77         percpu_write(is_idle, 1);
  78         atomic_notifier_call_chain(&idle_notifier, IDLE_START, NULL);
  79 }
  80
  81 static void __exit_idle(void)
  82 {
  83         if (x86_test_and_clear_bit_percpu(0, is_idle) == 0)
  84                 return;
  85         atomic_notifier_call_chain(&idle_notifier, IDLE_END, NULL);
  86 }
  87
  88 /* Called from interrupts to signify idle end */
  89 void exit_idle(void)
  90 {
  91         /* idle loop has pid 0 */
  92         if (current->pid)
  93                 return;
  94         __exit_idle();
  95 }
  96
  97 #ifndef CONFIG_SMP
  98 static inline void play_dead(void)
  99 {
 100         BUG();
 101 }
 102 #endif
 103
 104 /*
 105  * The idle thread. There's no useful work to be
 106  * done, so just try to conserve power and have a
 107  * low exit latency (ie sit in a loop waiting for
 108  * somebody to say that they'd like to reschedule)
 109  */
 110 void cpu_idle(void)
 111 {
 112         current_thread_info()->status |= TS_POLLING;
 113
 114         /*
 115          * If we're the non-boot CPU, nothing set the stack canary up
 116          * for us.  CPU0 already has it initialized but no harm in
 117          * doing it again.  This is a good place for updating it, as
 118          * we wont ever return from this function (so the invalid
 119          * canaries already on the stack wont ever trigger).
 120          */
 121         boot_init_stack_canary();
 122
 123         /* endless idle loop with no priority at all */
 124         while (1) {
 125                 tick_nohz_stop_sched_tick(1);
 126                 while (!need_resched()) {
 127
 128                         rmb();
 129
 130                         if (cpu_is_offline(smp_processor_id()))
 131                                 play_dead();
 132                         /*
 133                          * Idle routines should keep interrupts disabled
 134                          * from here on, until they go to idle.
 135                          * Otherwise, idle callbacks can misfire.
 136                          */
 137                         local_irq_disable();
 138                         enter_idle();
 139                         /* Don't trace irqs off for idle */
 140                         stop_critical_timings();
 141                         pm_idle();
 142                         start_critical_timings();
 143
 144                         trace_power_end(smp_processor_id());
 145
 146                         /* In many cases the interrupt that ended idle
 147                            has already called exit_idle. But some idle
 148                            loops can be woken up without interrupt. */
 149                         __exit_idle();
 150                 }
 151
 152                 tick_nohz_restart_sched_tick();
 153                 preempt_enable_no_resched();
 154                 schedule();
 155                 preempt_disable();
 156         }
 157 }
 158
 159 /* Prints also some state that isn't saved in the pt_regs */
 160 void __show_regs(struct pt_regs *regs, int all)
 161 {
 162         unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L, fs, gs, shadowgs;
 163         unsigned long d0, d1, d2, d3, d6, d7;
 164         unsigned int fsindex, gsindex;
 165         unsigned int ds, cs, es;
 166
 167         show_regs_common();
 168         printk(KERN_DEFAULT "RIP: %04lx:[<%016lx>] ", regs->cs & 0xffff, regs->ip);
 169         printk_address(regs->ip, 1);
 170         printk(KERN_DEFAULT "RSP: %04lx:%016lx  EFLAGS: %08lx\n", regs->ss,
 171                         regs->sp, regs->flags);
 172         printk(KERN_DEFAULT "RAX: %016lx RBX: %016lx RCX: %016lx\n",
 173                regs->ax, regs->bx, regs->cx);
 174         printk(KERN_DEFAULT "RDX: %016lx RSI: %016lx RDI: %016lx\n",
 175                regs->dx, regs->si, regs->di);
 176         printk(KERN_DEFAULT "RBP: %016lx R08: %016lx R09: %016lx\n",
 177                regs->bp, regs->r8, regs->r9);
 178         printk(KERN_DEFAULT "R10: %016lx R11: %016lx R12: %016lx\n",
 179                regs->r10, regs->r11, regs->r12);
 180         printk(KERN_DEFAULT "R13: %016lx R14: %016lx R15: %016lx\n",
 181                regs->r13, regs->r14, regs->r15);
 182
 183         asm("movl %%ds,%0" : "=r" (ds));
 184         asm("movl %%cs,%0" : "=r" (cs));
 185         asm("movl %%es,%0" : "=r" (es));
 186         asm("movl %%fs,%0" : "=r" (fsindex));
 187         asm("movl %%gs,%0" : "=r" (gsindex));
 188
 189         rdmsrl(MSR_FS_BASE, fs);
 190         rdmsrl(MSR_GS_BASE, gs);
 191         rdmsrl(MSR_KERNEL_GS_BASE, shadowgs);
 192
 193         if (!all)
 194                 return;
 195
 196         cr0 = read_cr0();
 197         cr2 = read_cr2();
 198         cr3 = read_cr3();
 199         cr4 = read_cr4();
 200
 201         printk(KERN_DEFAULT "FS:  %016lx(%04x) GS:%016lx(%04x) knlGS:%016lx\n",
 202                fs, fsindex, gs, gsindex, shadowgs);
 203         printk(KERN_DEFAULT "CS:  %04x DS: %04x ES: %04x CR0: %016lx\n", cs, ds,
 204                         es, cr0);
 205         printk(KERN_DEFAULT "CR2: %016lx CR3: %016lx CR4: %016lx\n", cr2, cr3,
 206                         cr4);
 207
 208         get_debugreg(d0, 0);
 209         get_debugreg(d1, 1);
 210         get_debugreg(d2, 2);
 211         printk(KERN_DEFAULT "DR0: %016lx DR1: %016lx DR2: %016lx\n", d0, d1, d2);
 212         get_debugreg(d3, 3);
 213         get_debugreg(d6, 6);
 214         get_debugreg(d7, 7);
 215         printk(KERN_DEFAULT "DR3: %016lx DR6: %016lx DR7: %016lx\n", d3, d6, d7);
 216 }
 217
 218 void release_thread(struct task_struct *dead_task)
 219 {
 220         if (dead_task->mm) {
 221                 if (dead_task->mm->context.size) {
 222                         printk("WARNING: dead process %8s still has LDT? <%p/%d>\n",
 223                                         dead_task->comm,
 224                                         dead_task->mm->context.ldt,
 225                                         dead_task->mm->context.size);
 226                         BUG();
 227                 }
 228         }
 229 }
 230
 231 static inline void set_32bit_tls(struct task_struct *t, int tls, u32 addr)
 232 {
 233         struct user_desc ud = {
 234                 .base_addr = addr,
 235                 .limit = 0xfffff,
 236                 .seg_32bit = 1,
 237                 .limit_in_pages = 1,
 238                 .useable = 1,
 239         };
 240         struct desc_struct *desc = t->thread.tls_array;
 241         desc += tls;
 242         fill_ldt(desc, &ud);
 243 }
 244
 245 static inline u32 read_32bit_tls(struct task_struct *t, int tls)
 246 {
 247         return get_desc_base(&t->thread.tls_array[tls]);
 248 }
 249
 250 /*
 251  * This gets called before we allocate a new thread and copy
 252  * the current task into it.
 253  */
 254 void prepare_to_copy(struct task_struct *tsk)
 255 {
 256         unlazy_fpu(tsk);
 257 }
 258
 259 int copy_thread(unsigned long clone_flags, unsigned long sp,
 260                 unsigned long unused,
 261         struct task_struct *p, struct pt_regs *regs)
 262 {
 263         int err;
 264         struct pt_regs *childregs;
 265         struct task_struct *me = current;
 266
 267         childregs = ((struct pt_regs *)
 268                         (THREAD_SIZE + task_stack_page(p))) - 1;
 269         *childregs = *regs;
 270
 271         childregs->ax = 0;
 272         if (user_mode(regs))
 273                 childregs->sp = sp;
 274         else
 275                 childregs->sp = (unsigned long)childregs;
 276
 277         p->thread.sp = (unsigned long) childregs;
 278         p->thread.sp0 = (unsigned long) (childregs+1);
 279         p->thread.usersp = me->thread.usersp;
 280
 281         set_tsk_thread_flag(p, TIF_FORK);
 282
 283         p->thread.io_bitmap_ptr = NULL;
 284
 285         savesegment(gs, p->thread.gsindex);
 286         p->thread.gs = p->thread.gsindex ? 0 : me->thread.gs;
 287         savesegment(fs, p->thread.fsindex);
 288         p->thread.fs = p->thread.fsindex ? 0 : me->thread.fs;
 289         savesegment(es, p->thread.es);
 290         savesegment(ds, p->thread.ds);
 291
 292         err = -ENOMEM;
 293         memset(p->thread.ptrace_bps, 0, sizeof(p->thread.ptrace_bps));
 294
 295         if (unlikely(test_tsk_thread_flag(me, TIF_IO_BITMAP))) {
 296                 p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL);
 297                 if (!p->thread.io_bitmap_ptr) {
 298                         p->thread.io_bitmap_max = 0;
 299                         return -ENOMEM;
 300                 }
 301                 memcpy(p->thread.io_bitmap_ptr, me->thread.io_bitmap_ptr,
 302                                 IO_BITMAP_BYTES);
 303                 set_tsk_thread_flag(p, TIF_IO_BITMAP);
 304         }
 305
 306         /*
 307          * Set a new TLS for the child thread?
 308          */
 309         if (clone_flags & CLONE_SETTLS) {
 310 #ifdef CONFIG_IA32_EMULATION
 311                 if (test_thread_flag(TIF_IA32))
 312                         err = do_set_thread_area(p, -1,
 313                                 (struct user_desc __user *)childregs->si, 0);
 314                 else
 315 #endif
 316                         err = do_arch_prctl(p, ARCH_SET_FS, childregs->r8);
 317                 if (err)
 318                         goto out;
 319         }
 320         err = 0;
 321 out:
 322         if (err && p->thread.io_bitmap_ptr) {
 323                 kfree(p->thread.io_bitmap_ptr);
 324                 p->thread.io_bitmap_max = 0;
 325         }
 326
 327         return err;
 328 }
 329
 330 static void
 331 start_thread_common(struct pt_regs *regs, unsigned long new_ip,
 332                     unsigned long new_sp,
 333                     unsigned int _cs, unsigned int _ss, unsigned int _ds)
 334 {
 335         loadsegment(fs, 0);
 336         loadsegment(es, _ds);
 337         loadsegment(ds, _ds);
 338         load_gs_index(0);
 339         regs->ip                = new_ip;
 340         regs->sp                = new_sp;
 341         percpu_write(old_rsp, new_sp);
 342         regs->cs                = _cs;
 343         regs->ss                = _ss;
 344         regs->flags             = X86_EFLAGS_IF;
 345         set_fs(USER_DS);
 346         /*
 347          * Free the old FP and other extended state
 348          */
 349         free_thread_xstate(current);
 350 }
 351
 352 void
 353 start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp)
 354 {
 355         start_thread_common(regs, new_ip, new_sp,
 356                             __USER_CS, __USER_DS, 0);
 357 }
 358
 359 #ifdef CONFIG_IA32_EMULATION
 360 void start_thread_ia32(struct pt_regs *regs, u32 new_ip, u32 new_sp)
 361 {
 362         start_thread_common(regs, new_ip, new_sp,
 363                             __USER32_CS, __USER32_DS, __USER32_DS);
 364 }
 365 #endif
 366
 367 /*
 368  *      switch_to(x,y) should switch tasks from x to y.
 369  *
 370  * This could still be optimized:
 371  * - fold all the options into a flag word and test it with a single test.
 372  * - could test fs/gs bitsliced
 373  *
 374  * Kprobes not supported here. Set the probe on schedule instead.
 375  * Function graph tracer not supported too.
 376  */
 377 __notrace_funcgraph struct task_struct *
 378 __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
 379 {
 380         struct thread_struct *prev = &prev_p->thread;
 381         struct thread_struct *next = &next_p->thread;
 382         int cpu = smp_processor_id();
 383         struct tss_struct *tss = &per_cpu(init_tss, cpu);
 384         unsigned fsindex, gsindex;
 385         bool preload_fpu;
 386
 387         /*
 388          * If the task has used fpu the last 5 timeslices, just do a full
 389          * restore of the math state immediately to avoid the trap; the
 390          * chances of needing FPU soon are obviously high now
 391          */
 392         preload_fpu = tsk_used_math(next_p) && next_p->fpu_counter > 5;
 393
 394         /* we're going to use this soon, after a few expensive things */
 395         if (preload_fpu)
 396                 prefetch(next->fpu.state);
 397
 398         /*
 399          * Reload esp0, LDT and the page table pointer:
 400          */
 401         load_sp0(tss, next);
 402
 403         /*
 404          * Switch DS and ES.
 405          * This won't pick up thread selector changes, but I guess that is ok.
 406          */
 407         savesegment(es, prev->es);
 408         if (unlikely(next->es | prev->es))
 409                 loadsegment(es, next->es);
 410
 411         savesegment(ds, prev->ds);
 412         if (unlikely(next->ds | prev->ds))
 413                 loadsegment(ds, next->ds);
 414
 415
 416         /* We must save %fs and %gs before load_TLS() because
 417          * %fs and %gs may be cleared by load_TLS().
 418          *
 419          * (e.g. xen_load_tls())
 420          */
 421         savesegment(fs, fsindex);
 422         savesegment(gs, gsindex);
 423
 424         load_TLS(next, cpu);
 425
 426         /* Must be after DS reload */
 427         unlazy_fpu(prev_p);
 428
 429         /* Make sure cpu is ready for new context */
 430         if (preload_fpu)
 431                 clts();
 432
 433         /*
 434          * Leave lazy mode, flushing any hypercalls made here.
 435          * This must be done before restoring TLS segments so
 436          * the GDT and LDT are properly updated, and must be
 437          * done before math_state_restore, so the TS bit is up
 438          * to date.
 439          */
 440         arch_end_context_switch(next_p);
 441
 442         /*
 443          * Switch FS and GS.
 444          *
 445          * Segment register != 0 always requires a reload.  Also
 446          * reload when it has changed.  When prev process used 64bit
 447          * base always reload to avoid an information leak.
 448          */
 449         if (unlikely(fsindex | next->fsindex | prev->fs)) {
 450                 loadsegment(fs, next->fsindex);
 451                 /*
 452                  * Check if the user used a selector != 0; if yes
 453                  *  clear 64bit base, since overloaded base is always
 454                  *  mapped to the Null selector
 455                  */
 456                 if (fsindex)
 457                         prev->fs = 0;
 458         }
 459         /* when next process has a 64bit base use it */
 460         if (next->fs)
 461                 wrmsrl(MSR_FS_BASE, next->fs);
 462         prev->fsindex = fsindex;
 463
 464         if (unlikely(gsindex | next->gsindex | prev->gs)) {
 465                 load_gs_index(next->gsindex);
 466                 if (gsindex)
 467                         prev->gs = 0;
 468         }
 469         if (next->gs)
 470                 wrmsrl(MSR_KERNEL_GS_BASE, next->gs);
 471         prev->gsindex = gsindex;
 472
 473         /*
 474          * Switch the PDA and FPU contexts.
 475          */
 476         prev->usersp = percpu_read(old_rsp);
 477         percpu_write(old_rsp, next->usersp);
 478         percpu_write(current_task, next_p);
 479
 480         percpu_write(kernel_stack,
 481                   (unsigned long)task_stack_page(next_p) +
 482                   THREAD_SIZE - KERNEL_STACK_OFFSET);
 483
 484         /*
 485          * Now maybe reload the debug registers and handle I/O bitmaps
 486          */
 487         if (unlikely(task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT ||
 488                      task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV))
 489                 __switch_to_xtra(prev_p, next_p, tss);
 490
 491         /*
 492          * Preload the FPU context, now that we've determined that the
 493          * task is likely to be using it.
 494          */
 495         if (preload_fpu)
 496                 __math_state_restore();
 497
 498         return prev_p;
 499 }
 500
 501 void set_personality_64bit(void)
 502 {
 503         /* inherit personality from parent */
 504
 505         /* Make sure to be in 64bit mode */
 506         clear_thread_flag(TIF_IA32);
 507
 508         /* TBD: overwrites user setup. Should have two bits.
 509            But 64bit processes have always behaved this way,
 510            so it's not too bad. The main problem is just that
 511            32bit childs are affected again. */
 512         current->personality &= ~READ_IMPLIES_EXEC;
 513 }
 514
 515 void set_personality_ia32(void)
 516 {
 517         /* inherit personality from parent */
 518
 519         /* Make sure to be in 32bit mode */
 520         set_thread_flag(TIF_IA32);
 521         current->personality |= force_personality32;
 522
 523         /* Prepare the first "return" to user space */
 524         current_thread_info()->status |= TS_COMPAT;
 525 }
 526
 527 unsigned long get_wchan(struct task_struct *p)
 528 {
 529         unsigned long stack;
 530         u64 fp, ip;
 531         int count = 0;
 532
 533         if (!p || p == current || p->state == TASK_RUNNING)
 534                 return 0;
 535         stack = (unsigned long)task_stack_page(p);
 536         if (p->thread.sp < stack || p->thread.sp >= stack+THREAD_SIZE)
 537                 return 0;
 538         fp = *(u64 *)(p->thread.sp);
 539         do {
 540                 if (fp < (unsigned long)stack ||
 541                     fp >= (unsigned long)stack+THREAD_SIZE)
 542                         return 0;
 543                 ip = *(u64 *)(fp+8);
 544                 if (!in_sched_functions(ip))
 545                         return ip;
 546                 fp = *(u64 *)fp;
 547         } while (count++ < 16);
 548         return 0;
 549 }
 550
 551 long do_arch_prctl(struct task_struct *task, int code, unsigned long addr)
 552 {
 553         int ret = 0;
 554         int doit = task == current;
 555         int cpu;
 556
 557         switch (code) {
 558         case ARCH_SET_GS:
 559                 if (addr >= TASK_SIZE_OF(task))
 560                         return -EPERM;
 561                 cpu = get_cpu();
 562                 /* handle small bases via the GDT because that's faster to
 563                    switch. */
 564                 if (addr <= 0xffffffff) {
 565                         set_32bit_tls(task, GS_TLS, addr);
 566                         if (doit) {
 567                                 load_TLS(&task->thread, cpu);
 568                                 load_gs_index(GS_TLS_SEL);
 569                         }
 570                         task->thread.gsindex = GS_TLS_SEL;
 571                         task->thread.gs = 0;
 572                 } else {
 573                         task->thread.gsindex = 0;
 574                         task->thread.gs = addr;
 575                         if (doit) {
 576                                 load_gs_index(0);
 577                                 ret = checking_wrmsrl(MSR_KERNEL_GS_BASE, addr);
 578                         }
 579                 }
 580                 put_cpu();
 581                 break;
 582         case ARCH_SET_FS:
 583                 /* Not strictly needed for fs, but do it for symmetry
 584                    with gs */
 585                 if (addr >= TASK_SIZE_OF(task))
 586                         return -EPERM;
 587                 cpu = get_cpu();
 588                 /* handle small bases via the GDT because that's faster to
 589                    switch. */
 590                 if (addr <= 0xffffffff) {
 591                         set_32bit_tls(task, FS_TLS, addr);
 592                         if (doit) {
 593                                 load_TLS(&task->thread, cpu);
 594                                 loadsegment(fs, FS_TLS_SEL);
 595                         }
 596                         task->thread.fsindex = FS_TLS_SEL;
 597                         task->thread.fs = 0;
 598                 } else {
 599                         task->thread.fsindex = 0;
 600                         task->thread.fs = addr;
 601                         if (doit) {
 602                                 /* set the selector to 0 to not confuse
 603                                    __switch_to */
 604                                 loadsegment(fs, 0);
 605                                 ret = checking_wrmsrl(MSR_FS_BASE, addr);
 606                         }
 607                 }
 608                 put_cpu();
 609                 break;
 610         case ARCH_GET_FS: {
 611                 unsigned long base;
 612                 if (task->thread.fsindex == FS_TLS_SEL)
 613                         base = read_32bit_tls(task, FS_TLS);
 614                 else if (doit)
 615                         rdmsrl(MSR_FS_BASE, base);
 616                 else
 617                         base = task->thread.fs;
 618                 ret = put_user(base, (unsigned long __user *)addr);
 619                 break;
 620         }
 621         case ARCH_GET_GS: {
 622                 unsigned long base;
 623                 unsigned gsindex;
 624                 if (task->thread.gsindex == GS_TLS_SEL)
 625                         base = read_32bit_tls(task, GS_TLS);
 626                 else if (doit) {
 627                         savesegment(gs, gsindex);
 628                         if (gsindex)
 629                                 rdmsrl(MSR_KERNEL_GS_BASE, base);
 630                         else
 631                                 base = task->thread.gs;
 632                 } else
 633                         base = task->thread.gs;
 634                 ret = put_user(base, (unsigned long __user *)addr);
 635                 break;
 636         }
 637
 638         default:
 639                 ret = -EINVAL;
 640                 break;
 641         }
 642
 643         return ret;
 644 }
 645
 646 long sys_arch_prctl(int code, unsigned long addr)
 647 {
 648         return do_arch_prctl(current, code, addr);
 649 }
 650
 651 unsigned long KSTK_ESP(struct task_struct *task)
 652 {
 653         return (test_tsk_thread_flag(task, TIF_IA32)) ?
 654                         (task_pt_regs(task)->sp) : ((task)->thread.usersp);
 655 }