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 __visible DEFINE_PER_CPU(unsigned long, rsp_scratch);
  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         set_tsk_thread_flag(p, TIF_FORK);
 165         p->thread.io_bitmap_ptr = NULL;
 166
 167         savesegment(gs, p->thread.gsindex);
 168         p->thread.gs = p->thread.gsindex ? 0 : me->thread.gs;
 169         savesegment(fs, p->thread.fsindex);
 170         p->thread.fs = p->thread.fsindex ? 0 : me->thread.fs;
 171         savesegment(es, p->thread.es);
 172         savesegment(ds, p->thread.ds);
 173         memset(p->thread.ptrace_bps, 0, sizeof(p->thread.ptrace_bps));
 174
 175         if (unlikely(p->flags & PF_KTHREAD)) {
 176                 /* kernel thread */
 177                 memset(childregs, 0, sizeof(struct pt_regs));
 178                 childregs->sp = (unsigned long)childregs;
 179                 childregs->ss = __KERNEL_DS;
 180                 childregs->bx = sp; /* function */
 181                 childregs->bp = arg;
 182                 childregs->orig_ax = -1;
 183                 childregs->cs = __KERNEL_CS | get_kernel_rpl();
 184                 childregs->flags = X86_EFLAGS_IF | X86_EFLAGS_FIXED;
 185                 return 0;
 186         }
 187         *childregs = *current_pt_regs();
 188
 189         childregs->ax = 0;
 190         if (sp)
 191                 childregs->sp = sp;
 192
 193         err = -ENOMEM;
 194         if (unlikely(test_tsk_thread_flag(me, TIF_IO_BITMAP))) {
 195                 p->thread.io_bitmap_ptr = kmemdup(me->thread.io_bitmap_ptr,
 196                                                   IO_BITMAP_BYTES, GFP_KERNEL);
 197                 if (!p->thread.io_bitmap_ptr) {
 198                         p->thread.io_bitmap_max = 0;
 199                         return -ENOMEM;
 200                 }
 201                 set_tsk_thread_flag(p, TIF_IO_BITMAP);
 202         }
 203
 204         /*
 205          * Set a new TLS for the child thread?
 206          */
 207         if (clone_flags & CLONE_SETTLS) {
 208 #ifdef CONFIG_IA32_EMULATION
 209                 if (is_ia32_task())
 210                         err = do_set_thread_area(p, -1,
 211                                 (struct user_desc __user *)childregs->si, 0);
 212                 else
 213 #endif
 214                         err = do_arch_prctl(p, ARCH_SET_FS, childregs->r8);
 215                 if (err)
 216                         goto out;
 217         }
 218         err = 0;
 219 out:
 220         if (err && p->thread.io_bitmap_ptr) {
 221                 kfree(p->thread.io_bitmap_ptr);
 222                 p->thread.io_bitmap_max = 0;
 223         }
 224
 225         return err;
 226 }
 227
 228 static void
 229 start_thread_common(struct pt_regs *regs, unsigned long new_ip,
 230                     unsigned long new_sp,
 231                     unsigned int _cs, unsigned int _ss, unsigned int _ds)
 232 {
 233         loadsegment(fs, 0);
 234         loadsegment(es, _ds);
 235         loadsegment(ds, _ds);
 236         load_gs_index(0);
 237         regs->ip                = new_ip;
 238         regs->sp                = new_sp;
 239         regs->cs                = _cs;
 240         regs->ss                = _ss;
 241         regs->flags             = X86_EFLAGS_IF;
 242         force_iret();
 243 }
 244
 245 void
 246 start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp)
 247 {
 248         start_thread_common(regs, new_ip, new_sp,
 249                             __USER_CS, __USER_DS, 0);
 250 }
 251
 252 #ifdef CONFIG_IA32_EMULATION
 253 void start_thread_ia32(struct pt_regs *regs, u32 new_ip, u32 new_sp)
 254 {
 255         start_thread_common(regs, new_ip, new_sp,
 256                             test_thread_flag(TIF_X32)
 257                             ? __USER_CS : __USER32_CS,
 258                             __USER_DS, __USER_DS);
 259 }
 260 #endif
 261
 262 /*
 263  *      switch_to(x,y) should switch tasks from x to y.
 264  *
 265  * This could still be optimized:
 266  * - fold all the options into a flag word and test it with a single test.
 267  * - could test fs/gs bitsliced
 268  *
 269  * Kprobes not supported here. Set the probe on schedule instead.
 270  * Function graph tracer not supported too.
 271  */
 272 __visible __notrace_funcgraph struct task_struct *
 273 __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
 274 {
 275         struct thread_struct *prev = &prev_p->thread;
 276         struct thread_struct *next = &next_p->thread;
 277         int cpu = smp_processor_id();
 278         struct tss_struct *tss = &per_cpu(cpu_tss, cpu);
 279         unsigned fsindex, gsindex;
 280         fpu_switch_t fpu;
 281
 282         fpu = switch_fpu_prepare(prev_p, next_p, cpu);
 283
 284         /* We must save %fs and %gs before load_TLS() because
 285          * %fs and %gs may be cleared by load_TLS().
 286          *
 287          * (e.g. xen_load_tls())
 288          */
 289         savesegment(fs, fsindex);
 290         savesegment(gs, gsindex);
 291
 292         /*
 293          * Load TLS before restoring any segments so that segment loads
 294          * reference the correct GDT entries.
 295          */
 296         load_TLS(next, cpu);
 297
 298         /*
 299          * Leave lazy mode, flushing any hypercalls made here.  This
 300          * must be done after loading TLS entries in the GDT but before
 301          * loading segments that might reference them, and and it must
 302          * be done before math_state_restore, so the TS bit is up to
 303          * date.
 304          */
 305         arch_end_context_switch(next_p);
 306
 307         /* Switch DS and ES.
 308          *
 309          * Reading them only returns the selectors, but writing them (if
 310          * nonzero) loads the full descriptor from the GDT or LDT.  The
 311          * LDT for next is loaded in switch_mm, and the GDT is loaded
 312          * above.
 313          *
 314          * We therefore need to write new values to the segment
 315          * registers on every context switch unless both the new and old
 316          * values are zero.
 317          *
 318          * Note that we don't need to do anything for CS and SS, as
 319          * those are saved and restored as part of pt_regs.
 320          */
 321         savesegment(es, prev->es);
 322         if (unlikely(next->es | prev->es))
 323                 loadsegment(es, next->es);
 324
 325         savesegment(ds, prev->ds);
 326         if (unlikely(next->ds | prev->ds))
 327                 loadsegment(ds, next->ds);
 328
 329         /*
 330          * Switch FS and GS.
 331          *
 332          * These are even more complicated than FS and GS: they have
 333          * 64-bit bases are that controlled by arch_prctl.  Those bases
 334          * only differ from the values in the GDT or LDT if the selector
 335          * is 0.
 336          *
 337          * Loading the segment register resets the hidden base part of
 338          * the register to 0 or the value from the GDT / LDT.  If the
 339          * next base address zero, writing 0 to the segment register is
 340          * much faster than using wrmsr to explicitly zero the base.
 341          *
 342          * The thread_struct.fs and thread_struct.gs values are 0
 343          * if the fs and gs bases respectively are not overridden
 344          * from the values implied by fsindex and gsindex.  They
 345          * are nonzero, and store the nonzero base addresses, if
 346          * the bases are overridden.
 347          *
 348          * (fs != 0 && fsindex != 0) || (gs != 0 && gsindex != 0) should
 349          * be impossible.
 350          *
 351          * Therefore we need to reload the segment registers if either
 352          * the old or new selector is nonzero, and we need to override
 353          * the base address if next thread expects it to be overridden.
 354          *
 355          * This code is unnecessarily slow in the case where the old and
 356          * new indexes are zero and the new base is nonzero -- it will
 357          * unnecessarily write 0 to the selector before writing the new
 358          * base address.
 359          *
 360          * Note: This all depends on arch_prctl being the only way that
 361          * user code can override the segment base.  Once wrfsbase and
 362          * wrgsbase are enabled, most of this code will need to change.
 363          */
 364         if (unlikely(fsindex | next->fsindex | prev->fs)) {
 365                 loadsegment(fs, next->fsindex);
 366
 367                 /*
 368                  * If user code wrote a nonzero value to FS, then it also
 369                  * cleared the overridden base address.
 370                  *
 371                  * XXX: if user code wrote 0 to FS and cleared the base
 372                  * address itself, we won't notice and we'll incorrectly
 373                  * restore the prior base address next time we reschdule
 374                  * the process.
 375                  */
 376                 if (fsindex)
 377                         prev->fs = 0;
 378         }
 379         if (next->fs)
 380                 wrmsrl(MSR_FS_BASE, next->fs);
 381         prev->fsindex = fsindex;
 382
 383         if (unlikely(gsindex | next->gsindex | prev->gs)) {
 384                 load_gs_index(next->gsindex);
 385
 386                 /* This works (and fails) the same way as fsindex above. */
 387                 if (gsindex)
 388                         prev->gs = 0;
 389         }
 390         if (next->gs)
 391                 wrmsrl(MSR_KERNEL_GS_BASE, next->gs);
 392         prev->gsindex = gsindex;
 393
 394         switch_fpu_finish(next_p, fpu);
 395
 396         /*
 397          * Switch the PDA and FPU contexts.
 398          */
 399         this_cpu_write(current_task, next_p);
 400
 401         /*
 402          * If it were not for PREEMPT_ACTIVE we could guarantee that the
 403          * preempt_count of all tasks was equal here and this would not be
 404          * needed.
 405          */
 406         task_thread_info(prev_p)->saved_preempt_count = this_cpu_read(__preempt_count);
 407         this_cpu_write(__preempt_count, task_thread_info(next_p)->saved_preempt_count);
 408
 409         /* Reload esp0 and ss1.  This changes current_thread_info(). */
 410         load_sp0(tss, next);
 411
 412         this_cpu_write(kernel_stack,
 413                 (unsigned long)task_stack_page(next_p) + THREAD_SIZE);
 414
 415         /*
 416          * Now maybe reload the debug registers and handle I/O bitmaps
 417          */
 418         if (unlikely(task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT ||
 419                      task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV))
 420                 __switch_to_xtra(prev_p, next_p, tss);
 421
 422         return prev_p;
 423 }
 424
 425 void set_personality_64bit(void)
 426 {
 427         /* inherit personality from parent */
 428
 429         /* Make sure to be in 64bit mode */
 430         clear_thread_flag(TIF_IA32);
 431         clear_thread_flag(TIF_ADDR32);
 432         clear_thread_flag(TIF_X32);
 433
 434         /* Ensure the corresponding mm is not marked. */
 435         if (current->mm)
 436                 current->mm->context.ia32_compat = 0;
 437
 438         /* TBD: overwrites user setup. Should have two bits.
 439            But 64bit processes have always behaved this way,
 440            so it's not too bad. The main problem is just that
 441            32bit childs are affected again. */
 442         current->personality &= ~READ_IMPLIES_EXEC;
 443 }
 444
 445 void set_personality_ia32(bool x32)
 446 {
 447         /* inherit personality from parent */
 448
 449         /* Make sure to be in 32bit mode */
 450         set_thread_flag(TIF_ADDR32);
 451
 452         /* Mark the associated mm as containing 32-bit tasks. */
 453         if (x32) {
 454                 clear_thread_flag(TIF_IA32);
 455                 set_thread_flag(TIF_X32);
 456                 if (current->mm)
 457                         current->mm->context.ia32_compat = TIF_X32;
 458                 current->personality &= ~READ_IMPLIES_EXEC;
 459                 /* is_compat_task() uses the presence of the x32
 460                    syscall bit flag to determine compat status */
 461                 current_thread_info()->status &= ~TS_COMPAT;
 462         } else {
 463                 set_thread_flag(TIF_IA32);
 464                 clear_thread_flag(TIF_X32);
 465                 if (current->mm)
 466                         current->mm->context.ia32_compat = TIF_IA32;
 467                 current->personality |= force_personality32;
 468                 /* Prepare the first "return" to user space */
 469                 current_thread_info()->status |= TS_COMPAT;
 470         }
 471 }
 472 EXPORT_SYMBOL_GPL(set_personality_ia32);
 473
 474 unsigned long get_wchan(struct task_struct *p)
 475 {
 476         unsigned long stack;
 477         u64 fp, ip;
 478         int count = 0;
 479
 480         if (!p || p == current || p->state == TASK_RUNNING)
 481                 return 0;
 482         stack = (unsigned long)task_stack_page(p);
 483         if (p->thread.sp < stack || p->thread.sp >= stack+THREAD_SIZE)
 484                 return 0;
 485         fp = *(u64 *)(p->thread.sp);
 486         do {
 487                 if (fp < (unsigned long)stack ||
 488                     fp >= (unsigned long)stack+THREAD_SIZE)
 489                         return 0;
 490                 ip = *(u64 *)(fp+8);
 491                 if (!in_sched_functions(ip))
 492                         return ip;
 493                 fp = *(u64 *)fp;
 494         } while (count++ < 16);
 495         return 0;
 496 }
 497
 498 long do_arch_prctl(struct task_struct *task, int code, unsigned long addr)
 499 {
 500         int ret = 0;
 501         int doit = task == current;
 502         int cpu;
 503
 504         switch (code) {
 505         case ARCH_SET_GS:
 506                 if (addr >= TASK_SIZE_OF(task))
 507                         return -EPERM;
 508                 cpu = get_cpu();
 509                 /* handle small bases via the GDT because that's faster to
 510                    switch. */
 511                 if (addr <= 0xffffffff) {
 512                         set_32bit_tls(task, GS_TLS, addr);
 513                         if (doit) {
 514                                 load_TLS(&task->thread, cpu);
 515                                 load_gs_index(GS_TLS_SEL);
 516                         }
 517                         task->thread.gsindex = GS_TLS_SEL;
 518                         task->thread.gs = 0;
 519                 } else {
 520                         task->thread.gsindex = 0;
 521                         task->thread.gs = addr;
 522                         if (doit) {
 523                                 load_gs_index(0);
 524                                 ret = wrmsrl_safe(MSR_KERNEL_GS_BASE, addr);
 525                         }
 526                 }
 527                 put_cpu();
 528                 break;
 529         case ARCH_SET_FS:
 530                 /* Not strictly needed for fs, but do it for symmetry
 531                    with gs */
 532                 if (addr >= TASK_SIZE_OF(task))
 533                         return -EPERM;
 534                 cpu = get_cpu();
 535                 /* handle small bases via the GDT because that's faster to
 536                    switch. */
 537                 if (addr <= 0xffffffff) {
 538                         set_32bit_tls(task, FS_TLS, addr);
 539                         if (doit) {
 540                                 load_TLS(&task->thread, cpu);
 541                                 loadsegment(fs, FS_TLS_SEL);
 542                         }
 543                         task->thread.fsindex = FS_TLS_SEL;
 544                         task->thread.fs = 0;
 545                 } else {
 546                         task->thread.fsindex = 0;
 547                         task->thread.fs = addr;
 548                         if (doit) {
 549                                 /* set the selector to 0 to not confuse
 550                                    __switch_to */
 551                                 loadsegment(fs, 0);
 552                                 ret = wrmsrl_safe(MSR_FS_BASE, addr);
 553                         }
 554                 }
 555                 put_cpu();
 556                 break;
 557         case ARCH_GET_FS: {
 558                 unsigned long base;
 559                 if (task->thread.fsindex == FS_TLS_SEL)
 560                         base = read_32bit_tls(task, FS_TLS);
 561                 else if (doit)
 562                         rdmsrl(MSR_FS_BASE, base);
 563                 else
 564                         base = task->thread.fs;
 565                 ret = put_user(base, (unsigned long __user *)addr);
 566                 break;
 567         }
 568         case ARCH_GET_GS: {
 569                 unsigned long base;
 570                 unsigned gsindex;
 571                 if (task->thread.gsindex == GS_TLS_SEL)
 572                         base = read_32bit_tls(task, GS_TLS);
 573                 else if (doit) {
 574                         savesegment(gs, gsindex);
 575                         if (gsindex)
 576                                 rdmsrl(MSR_KERNEL_GS_BASE, base);
 577                         else
 578                                 base = task->thread.gs;
 579                 } else
 580                         base = task->thread.gs;
 581                 ret = put_user(base, (unsigned long __user *)addr);
 582                 break;
 583         }
 584
 585         default:
 586                 ret = -EINVAL;
 587                 break;
 588         }
 589
 590         return ret;
 591 }
 592
 593 long sys_arch_prctl(int code, unsigned long addr)
 594 {
 595         return do_arch_prctl(current, code, addr);
 596 }
 597
 598 unsigned long KSTK_ESP(struct task_struct *task)
 599 {
 600         return task_pt_regs(task)->sp;
 601 }