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/cpu.h>
  18 #include <linux/errno.h>
  19 #include <linux/sched.h>
  20 #include <linux/fs.h>
  21 #include <linux/kernel.h>
  22 #include <linux/mm.h>
  23 #include <linux/elfcore.h>
  24 #include <linux/smp.h>
  25 #include <linux/slab.h>
  26 #include <linux/user.h>
  27 #include <linux/interrupt.h>
  28 #include <linux/delay.h>
  29 #include <linux/module.h>
  30 #include <linux/ptrace.h>
  31 #include <linux/notifier.h>
  32 #include <linux/kprobes.h>
  33 #include <linux/kdebug.h>
  34 #include <linux/prctl.h>
  35 #include <linux/uaccess.h>
  36 #include <linux/io.h>
  37 #include <linux/ftrace.h>
  38
  39 #include <asm/pgtable.h>
  40 #include <asm/processor.h>
  41 #include <asm/i387.h>
  42 #include <asm/fpu-internal.h>
  43 #include <asm/mmu_context.h>
  44 #include <asm/prctl.h>
  45 #include <asm/desc.h>
  46 #include <asm/proto.h>
  47 #include <asm/ia32.h>
  48 #include <asm/idle.h>
  49 #include <asm/syscalls.h>
  50 #include <asm/debugreg.h>
  51 #include <asm/switch_to.h>
  52
  53 asmlinkage extern void ret_from_fork(void);
  54
  55 asmlinkage DEFINE_PER_CPU(unsigned long, old_rsp);
  56
  57 /* Prints also some state that isn't saved in the pt_regs */
  58 void __show_regs(struct pt_regs *regs, int all)
  59 {
  60         unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L, fs, gs, shadowgs;
  61         unsigned long d0, d1, d2, d3, d6, d7;
  62         unsigned int fsindex, gsindex;
  63         unsigned int ds, cs, es;
  64
  65         printk(KERN_DEFAULT "RIP: %04lx:[<%016lx>] ", regs->cs & 0xffff, regs->ip);
  66         printk_address(regs->ip);
  67         printk(KERN_DEFAULT "RSP: %04lx:%016lx  EFLAGS: %08lx\n", regs->ss,
  68                         regs->sp, regs->flags);
  69         printk(KERN_DEFAULT "RAX: %016lx RBX: %016lx RCX: %016lx\n",
  70                regs->ax, regs->bx, regs->cx);
  71         printk(KERN_DEFAULT "RDX: %016lx RSI: %016lx RDI: %016lx\n",
  72                regs->dx, regs->si, regs->di);
  73         printk(KERN_DEFAULT "RBP: %016lx R08: %016lx R09: %016lx\n",
  74                regs->bp, regs->r8, regs->r9);
  75         printk(KERN_DEFAULT "R10: %016lx R11: %016lx R12: %016lx\n",
  76                regs->r10, regs->r11, regs->r12);
  77         printk(KERN_DEFAULT "R13: %016lx R14: %016lx R15: %016lx\n",
  78                regs->r13, regs->r14, regs->r15);
  79
  80         asm("movl %%ds,%0" : "=r" (ds));
  81         asm("movl %%cs,%0" : "=r" (cs));
  82         asm("movl %%es,%0" : "=r" (es));
  83         asm("movl %%fs,%0" : "=r" (fsindex));
  84         asm("movl %%gs,%0" : "=r" (gsindex));
  85
  86         rdmsrl(MSR_FS_BASE, fs);
  87         rdmsrl(MSR_GS_BASE, gs);
  88         rdmsrl(MSR_KERNEL_GS_BASE, shadowgs);
  89
  90         if (!all)
  91                 return;
  92
  93         cr0 = read_cr0();
  94         cr2 = read_cr2();
  95         cr3 = read_cr3();
  96         cr4 = read_cr4();
  97
  98         printk(KERN_DEFAULT "FS:  %016lx(%04x) GS:%016lx(%04x) knlGS:%016lx\n",
  99                fs, fsindex, gs, gsindex, shadowgs);
 100         printk(KERN_DEFAULT "CS:  %04x DS: %04x ES: %04x CR0: %016lx\n", cs, ds,
 101                         es, cr0);
 102         printk(KERN_DEFAULT "CR2: %016lx CR3: %016lx CR4: %016lx\n", cr2, cr3,
 103                         cr4);
 104
 105         get_debugreg(d0, 0);
 106         get_debugreg(d1, 1);
 107         get_debugreg(d2, 2);
 108         get_debugreg(d3, 3);
 109         get_debugreg(d6, 6);
 110         get_debugreg(d7, 7);
 111
 112         /* Only print out debug registers if they are in their non-default state. */
 113         if ((d0 == 0) && (d1 == 0) && (d2 == 0) && (d3 == 0) &&
 114             (d6 == DR6_RESERVED) && (d7 == 0x400))
 115                 return;
 116
 117         printk(KERN_DEFAULT "DR0: %016lx DR1: %016lx DR2: %016lx\n", d0, d1, d2);
 118         printk(KERN_DEFAULT "DR3: %016lx DR6: %016lx DR7: %016lx\n", d3, d6, d7);
 119
 120 }
 121
 122 void release_thread(struct task_struct *dead_task)
 123 {
 124         if (dead_task->mm) {
 125                 if (dead_task->mm->context.size) {
 126                         pr_warn("WARNING: dead process %s still has LDT? <%p/%d>\n",
 127                                 dead_task->comm,
 128                                 dead_task->mm->context.ldt,
 129                                 dead_task->mm->context.size);
 130                         BUG();
 131                 }
 132         }
 133 }
 134
 135 static inline void set_32bit_tls(struct task_struct *t, int tls, u32 addr)
 136 {
 137         struct user_desc ud = {
 138                 .base_addr = addr,
 139                 .limit = 0xfffff,
 140                 .seg_32bit = 1,
 141                 .limit_in_pages = 1,
 142                 .useable = 1,
 143         };
 144         struct desc_struct *desc = t->thread.tls_array;
 145         desc += tls;
 146         fill_ldt(desc, &ud);
 147 }
 148
 149 static inline u32 read_32bit_tls(struct task_struct *t, int tls)
 150 {
 151         return get_desc_base(&t->thread.tls_array[tls]);
 152 }
 153
 154 int copy_thread(unsigned long clone_flags, unsigned long sp,
 155                 unsigned long arg, struct task_struct *p)
 156 {
 157         int err;
 158         struct pt_regs *childregs;
 159         struct task_struct *me = current;
 160
 161         p->thread.sp0 = (unsigned long)task_stack_page(p) + THREAD_SIZE;
 162         childregs = task_pt_regs(p);
 163         p->thread.sp = (unsigned long) childregs;
 164         p->thread.usersp = me->thread.usersp;
 165         set_tsk_thread_flag(p, TIF_FORK);
 166         p->thread.fpu_counter = 0;
 167         p->thread.io_bitmap_ptr = NULL;
 168
 169         savesegment(gs, p->thread.gsindex);
 170         p->thread.gs = p->thread.gsindex ? 0 : me->thread.gs;
 171         savesegment(fs, p->thread.fsindex);
 172         p->thread.fs = p->thread.fsindex ? 0 : me->thread.fs;
 173         savesegment(es, p->thread.es);
 174         savesegment(ds, p->thread.ds);
 175         memset(p->thread.ptrace_bps, 0, sizeof(p->thread.ptrace_bps));
 176
 177         if (unlikely(p->flags & PF_KTHREAD)) {
 178                 /* kernel thread */
 179                 memset(childregs, 0, sizeof(struct pt_regs));
 180                 childregs->sp = (unsigned long)childregs;
 181                 childregs->ss = __KERNEL_DS;
 182                 childregs->bx = sp; /* function */
 183                 childregs->bp = arg;
 184                 childregs->orig_ax = -1;
 185                 childregs->cs = __KERNEL_CS | get_kernel_rpl();
 186                 childregs->flags = X86_EFLAGS_IF | X86_EFLAGS_FIXED;
 187                 return 0;
 188         }
 189         *childregs = *current_pt_regs();
 190
 191         childregs->ax = 0;
 192         if (sp)
 193                 childregs->sp = sp;
 194
 195         err = -ENOMEM;
 196         memset(p->thread.ptrace_bps, 0, sizeof(p->thread.ptrace_bps));
 197
 198         if (unlikely(test_tsk_thread_flag(me, TIF_IO_BITMAP))) {
 199                 p->thread.io_bitmap_ptr = kmemdup(me->thread.io_bitmap_ptr,
 200                                                   IO_BITMAP_BYTES, GFP_KERNEL);
 201                 if (!p->thread.io_bitmap_ptr) {
 202                         p->thread.io_bitmap_max = 0;
 203                         return -ENOMEM;
 204                 }
 205                 set_tsk_thread_flag(p, TIF_IO_BITMAP);
 206         }
 207
 208         /*
 209          * Set a new TLS for the child thread?
 210          */
 211         if (clone_flags & CLONE_SETTLS) {
 212 #ifdef CONFIG_IA32_EMULATION
 213                 if (test_thread_flag(TIF_IA32))
 214                         err = do_set_thread_area(p, -1,
 215                                 (struct user_desc __user *)childregs->si, 0);
 216                 else
 217 #endif
 218                         err = do_arch_prctl(p, ARCH_SET_FS, childregs->r8);
 219                 if (err)
 220                         goto out;
 221         }
 222         err = 0;
 223 out:
 224         if (err && p->thread.io_bitmap_ptr) {
 225                 kfree(p->thread.io_bitmap_ptr);
 226                 p->thread.io_bitmap_max = 0;
 227         }
 228
 229         return err;
 230 }
 231
 232 static void
 233 start_thread_common(struct pt_regs *regs, unsigned long new_ip,
 234                     unsigned long new_sp,
 235                     unsigned int _cs, unsigned int _ss, unsigned int _ds)
 236 {
 237         loadsegment(fs, 0);
 238         loadsegment(es, _ds);
 239         loadsegment(ds, _ds);
 240         load_gs_index(0);
 241         current->thread.usersp  = new_sp;
 242         regs->ip                = new_ip;
 243         regs->sp                = new_sp;
 244         this_cpu_write(old_rsp, new_sp);
 245         regs->cs                = _cs;
 246         regs->ss                = _ss;
 247         regs->flags             = X86_EFLAGS_IF;
 248 }
 249
 250 void
 251 start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp)
 252 {
 253         start_thread_common(regs, new_ip, new_sp,
 254                             __USER_CS, __USER_DS, 0);
 255 }
 256
 257 #ifdef CONFIG_IA32_EMULATION
 258 void start_thread_ia32(struct pt_regs *regs, u32 new_ip, u32 new_sp)
 259 {
 260         start_thread_common(regs, new_ip, new_sp,
 261                             test_thread_flag(TIF_X32)
 262                             ? __USER_CS : __USER32_CS,
 263                             __USER_DS, __USER_DS);
 264 }
 265 #endif
 266
 267 /*
 268  *      switch_to(x,y) should switch tasks from x to y.
 269  *
 270  * This could still be optimized:
 271  * - fold all the options into a flag word and test it with a single test.
 272  * - could test fs/gs bitsliced
 273  *
 274  * Kprobes not supported here. Set the probe on schedule instead.
 275  * Function graph tracer not supported too.
 276  */
 277 __visible __notrace_funcgraph struct task_struct *
 278 __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
 279 {
 280         struct thread_struct *prev = &prev_p->thread;
 281         struct thread_struct *next = &next_p->thread;
 282         int cpu = smp_processor_id();
 283         struct tss_struct *tss = &per_cpu(init_tss, cpu);
 284         unsigned fsindex, gsindex;
 285         fpu_switch_t fpu;
 286
 287         fpu = switch_fpu_prepare(prev_p, next_p, cpu);
 288
 289         /*
 290          * Reload esp0, LDT and the page table pointer:
 291          */
 292         load_sp0(tss, next);
 293
 294         /*
 295          * Switch DS and ES.
 296          * This won't pick up thread selector changes, but I guess that is ok.
 297          */
 298         savesegment(es, prev->es);
 299         if (unlikely(next->es | prev->es))
 300                 loadsegment(es, next->es);
 301
 302         savesegment(ds, prev->ds);
 303         if (unlikely(next->ds | prev->ds))
 304                 loadsegment(ds, next->ds);
 305
 306
 307         /* We must save %fs and %gs before load_TLS() because
 308          * %fs and %gs may be cleared by load_TLS().
 309          *
 310          * (e.g. xen_load_tls())
 311          */
 312         savesegment(fs, fsindex);
 313         savesegment(gs, gsindex);
 314
 315         load_TLS(next, cpu);
 316
 317         /*
 318          * Leave lazy mode, flushing any hypercalls made here.
 319          * This must be done before restoring TLS segments so
 320          * the GDT and LDT are properly updated, and must be
 321          * done before math_state_restore, so the TS bit is up
 322          * to date.
 323          */
 324         arch_end_context_switch(next_p);
 325
 326         /*
 327          * Switch FS and GS.
 328          *
 329          * Segment register != 0 always requires a reload.  Also
 330          * reload when it has changed.  When prev process used 64bit
 331          * base always reload to avoid an information leak.
 332          */
 333         if (unlikely(fsindex | next->fsindex | prev->fs)) {
 334                 loadsegment(fs, next->fsindex);
 335                 /*
 336                  * Check if the user used a selector != 0; if yes
 337                  *  clear 64bit base, since overloaded base is always
 338                  *  mapped to the Null selector
 339                  */
 340                 if (fsindex)
 341                         prev->fs = 0;
 342         }
 343         /* when next process has a 64bit base use it */
 344         if (next->fs)
 345                 wrmsrl(MSR_FS_BASE, next->fs);
 346         prev->fsindex = fsindex;
 347
 348         if (unlikely(gsindex | next->gsindex | prev->gs)) {
 349                 load_gs_index(next->gsindex);
 350                 if (gsindex)
 351                         prev->gs = 0;
 352         }
 353         if (next->gs)
 354                 wrmsrl(MSR_KERNEL_GS_BASE, next->gs);
 355         prev->gsindex = gsindex;
 356
 357         switch_fpu_finish(next_p, fpu);
 358
 359         /*
 360          * Switch the PDA and FPU contexts.
 361          */
 362         prev->usersp = this_cpu_read(old_rsp);
 363         this_cpu_write(old_rsp, next->usersp);
 364         this_cpu_write(current_task, next_p);
 365
 366         /*
 367          * If it were not for PREEMPT_ACTIVE we could guarantee that the
 368          * preempt_count of all tasks was equal here and this would not be
 369          * needed.
 370          */
 371         task_thread_info(prev_p)->saved_preempt_count = this_cpu_read(__preempt_count);
 372         this_cpu_write(__preempt_count, task_thread_info(next_p)->saved_preempt_count);
 373
 374         this_cpu_write(kernel_stack,
 375                   (unsigned long)task_stack_page(next_p) +
 376                   THREAD_SIZE - KERNEL_STACK_OFFSET);
 377
 378         /*
 379          * Now maybe reload the debug registers and handle I/O bitmaps
 380          */
 381         if (unlikely(task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT ||
 382                      task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV))
 383                 __switch_to_xtra(prev_p, next_p, tss);
 384
 385         return prev_p;
 386 }
 387
 388 void set_personality_64bit(void)
 389 {
 390         /* inherit personality from parent */
 391
 392         /* Make sure to be in 64bit mode */
 393         clear_thread_flag(TIF_IA32);
 394         clear_thread_flag(TIF_ADDR32);
 395         clear_thread_flag(TIF_X32);
 396
 397         /* Ensure the corresponding mm is not marked. */
 398         if (current->mm)
 399                 current->mm->context.ia32_compat = 0;
 400
 401         /* TBD: overwrites user setup. Should have two bits.
 402            But 64bit processes have always behaved this way,
 403            so it's not too bad. The main problem is just that
 404            32bit childs are affected again. */
 405         current->personality &= ~READ_IMPLIES_EXEC;
 406 }
 407
 408 void set_personality_ia32(bool x32)
 409 {
 410         /* inherit personality from parent */
 411
 412         /* Make sure to be in 32bit mode */
 413         set_thread_flag(TIF_ADDR32);
 414
 415         /* Mark the associated mm as containing 32-bit tasks. */
 416         if (current->mm)
 417                 current->mm->context.ia32_compat = 1;
 418
 419         if (x32) {
 420                 clear_thread_flag(TIF_IA32);
 421                 set_thread_flag(TIF_X32);
 422                 current->personality &= ~READ_IMPLIES_EXEC;
 423                 /* is_compat_task() uses the presence of the x32
 424                    syscall bit flag to determine compat status */
 425                 current_thread_info()->status &= ~TS_COMPAT;
 426         } else {
 427                 set_thread_flag(TIF_IA32);
 428                 clear_thread_flag(TIF_X32);
 429                 current->personality |= force_personality32;
 430                 /* Prepare the first "return" to user space */
 431                 current_thread_info()->status |= TS_COMPAT;
 432         }
 433 }
 434 EXPORT_SYMBOL_GPL(set_personality_ia32);
 435
 436 unsigned long get_wchan(struct task_struct *p)
 437 {
 438         unsigned long stack;
 439         u64 fp, ip;
 440         int count = 0;
 441
 442         if (!p || p == current || p->state == TASK_RUNNING)
 443                 return 0;
 444         stack = (unsigned long)task_stack_page(p);
 445         if (p->thread.sp < stack || p->thread.sp >= stack+THREAD_SIZE)
 446                 return 0;
 447         fp = *(u64 *)(p->thread.sp);
 448         do {
 449                 if (fp < (unsigned long)stack ||
 450                     fp >= (unsigned long)stack+THREAD_SIZE)
 451                         return 0;
 452                 ip = *(u64 *)(fp+8);
 453                 if (!in_sched_functions(ip))
 454                         return ip;
 455                 fp = *(u64 *)fp;
 456         } while (count++ < 16);
 457         return 0;
 458 }
 459
 460 long do_arch_prctl(struct task_struct *task, int code, unsigned long addr)
 461 {
 462         int ret = 0;
 463         int doit = task == current;
 464         int cpu;
 465
 466         switch (code) {
 467         case ARCH_SET_GS:
 468                 if (addr >= TASK_SIZE_OF(task))
 469                         return -EPERM;
 470                 cpu = get_cpu();
 471                 /* handle small bases via the GDT because that's faster to
 472                    switch. */
 473                 if (addr <= 0xffffffff) {
 474                         set_32bit_tls(task, GS_TLS, addr);
 475                         if (doit) {
 476                                 load_TLS(&task->thread, cpu);
 477                                 load_gs_index(GS_TLS_SEL);
 478                         }
 479                         task->thread.gsindex = GS_TLS_SEL;
 480                         task->thread.gs = 0;
 481                 } else {
 482                         task->thread.gsindex = 0;
 483                         task->thread.gs = addr;
 484                         if (doit) {
 485                                 load_gs_index(0);
 486                                 ret = wrmsrl_safe(MSR_KERNEL_GS_BASE, addr);
 487                         }
 488                 }
 489                 put_cpu();
 490                 break;
 491         case ARCH_SET_FS:
 492                 /* Not strictly needed for fs, but do it for symmetry
 493                    with gs */
 494                 if (addr >= TASK_SIZE_OF(task))
 495                         return -EPERM;
 496                 cpu = get_cpu();
 497                 /* handle small bases via the GDT because that's faster to
 498                    switch. */
 499                 if (addr <= 0xffffffff) {
 500                         set_32bit_tls(task, FS_TLS, addr);
 501                         if (doit) {
 502                                 load_TLS(&task->thread, cpu);
 503                                 loadsegment(fs, FS_TLS_SEL);
 504                         }
 505                         task->thread.fsindex = FS_TLS_SEL;
 506                         task->thread.fs = 0;
 507                 } else {
 508                         task->thread.fsindex = 0;
 509                         task->thread.fs = addr;
 510                         if (doit) {
 511                                 /* set the selector to 0 to not confuse
 512                                    __switch_to */
 513                                 loadsegment(fs, 0);
 514                                 ret = wrmsrl_safe(MSR_FS_BASE, addr);
 515                         }
 516                 }
 517                 put_cpu();
 518                 break;
 519         case ARCH_GET_FS: {
 520                 unsigned long base;
 521                 if (task->thread.fsindex == FS_TLS_SEL)
 522                         base = read_32bit_tls(task, FS_TLS);
 523                 else if (doit)
 524                         rdmsrl(MSR_FS_BASE, base);
 525                 else
 526                         base = task->thread.fs;
 527                 ret = put_user(base, (unsigned long __user *)addr);
 528                 break;
 529         }
 530         case ARCH_GET_GS: {
 531                 unsigned long base;
 532                 unsigned gsindex;
 533                 if (task->thread.gsindex == GS_TLS_SEL)
 534                         base = read_32bit_tls(task, GS_TLS);
 535                 else if (doit) {
 536                         savesegment(gs, gsindex);
 537                         if (gsindex)
 538                                 rdmsrl(MSR_KERNEL_GS_BASE, base);
 539                         else
 540                                 base = task->thread.gs;
 541                 } else
 542                         base = task->thread.gs;
 543                 ret = put_user(base, (unsigned long __user *)addr);
 544                 break;
 545         }
 546
 547         default:
 548                 ret = -EINVAL;
 549                 break;
 550         }
 551
 552         return ret;
 553 }
 554
 555 long sys_arch_prctl(int code, unsigned long addr)
 556 {
 557         return do_arch_prctl(current, code, addr);
 558 }
 559
 560 unsigned long KSTK_ESP(struct task_struct *task)
 561 {
 562         return (test_tsk_thread_flag(task, TIF_IA32)) ?
 563                         (task_pt_regs(task)->sp) : ((task)->thread.usersp);
 564 }