arch/x86/kernel/process_32.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
   8 /*
   9  * This file handles the architecture-dependent parts of process handling..
  10  */
  11
  12 #include <linux/stackprotector.h>
  13 #include <linux/cpu.h>
  14 #include <linux/errno.h>
  15 #include <linux/sched.h>
  16 #include <linux/fs.h>
  17 #include <linux/kernel.h>
  18 #include <linux/mm.h>
  19 #include <linux/elfcore.h>
  20 #include <linux/smp.h>
  21 #include <linux/stddef.h>
  22 #include <linux/slab.h>
  23 #include <linux/vmalloc.h>
  24 #include <linux/user.h>
  25 #include <linux/interrupt.h>
  26 #include <linux/delay.h>
  27 #include <linux/reboot.h>
  28 #include <linux/init.h>
  29 #include <linux/mc146818rtc.h>
  30 #include <linux/module.h>
  31 #include <linux/kallsyms.h>
  32 #include <linux/ptrace.h>
  33 #include <linux/personality.h>
  34 #include <linux/tick.h>
  35 #include <linux/percpu.h>
  36 #include <linux/prctl.h>
  37 #include <linux/ftrace.h>
  38 #include <linux/uaccess.h>
  39 #include <linux/io.h>
  40 #include <linux/kdebug.h>
  41
  42 #include <asm/pgtable.h>
  43 #include <asm/system.h>
  44 #include <asm/ldt.h>
  45 #include <asm/processor.h>
  46 #include <asm/i387.h>
  47 #include <asm/desc.h>
  48 #ifdef CONFIG_MATH_EMULATION
  49 #include <asm/math_emu.h>
  50 #endif
  51
  52 #include <linux/err.h>
  53
  54 #include <asm/tlbflush.h>
  55 #include <asm/cpu.h>
  56 #include <asm/idle.h>
  57 #include <asm/syscalls.h>
  58 #include <asm/debugreg.h>
  59
  60 asmlinkage void ret_from_fork(void) __asm__("ret_from_fork");
  61
  62 /*
  63  * Return saved PC of a blocked thread.
  64  */
  65 unsigned long thread_saved_pc(struct task_struct *tsk)
  66 {
  67         return ((unsigned long *)tsk->thread.sp)[3];
  68 }
  69
  70 #ifndef CONFIG_SMP
  71 static inline void play_dead(void)
  72 {
  73         BUG();
  74 }
  75 #endif
  76
  77 /*
  78  * The idle thread. There's no useful work to be
  79  * done, so just try to conserve power and have a
  80  * low exit latency (ie sit in a loop waiting for
  81  * somebody to say that they'd like to reschedule)
  82  */
  83 void cpu_idle(void)
  84 {
  85         int cpu = smp_processor_id();
  86
  87         /*
  88          * If we're the non-boot CPU, nothing set the stack canary up
  89          * for us.  CPU0 already has it initialized but no harm in
  90          * doing it again.  This is a good place for updating it, as
  91          * we wont ever return from this function (so the invalid
  92          * canaries already on the stack wont ever trigger).
  93          */
  94         boot_init_stack_canary();
  95
  96         current_thread_info()->status |= TS_POLLING;
  97
  98         /* endless idle loop with no priority at all */
  99         while (1) {
 100                 tick_nohz_stop_sched_tick(1);
 101                 while (!need_resched()) {
 102
 103                         check_pgt_cache();
 104                         rmb();
 105
 106                         if (cpu_is_offline(cpu))
 107                                 play_dead();
 108
 109                         local_irq_disable();
 110                         /* Don't trace irqs off for idle */
 111                         stop_critical_timings();
 112                         pm_idle();
 113                         start_critical_timings();
 114                 }
 115                 tick_nohz_restart_sched_tick();
 116                 preempt_enable_no_resched();
 117                 schedule();
 118                 preempt_disable();
 119         }
 120 }
 121
 122 void __show_regs(struct pt_regs *regs, int all)
 123 {
 124         unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L;
 125         unsigned long d0, d1, d2, d3, d6, d7;
 126         unsigned long sp;
 127         unsigned short ss, gs;
 128
 129         if (user_mode_vm(regs)) {
 130                 sp = regs->sp;
 131                 ss = regs->ss & 0xffff;
 132                 gs = get_user_gs(regs);
 133         } else {
 134                 sp = kernel_stack_pointer(regs);
 135                 savesegment(ss, ss);
 136                 savesegment(gs, gs);
 137         }
 138
 139         show_regs_common();
 140
 141         printk(KERN_DEFAULT "EIP: %04x:[<%08lx>] EFLAGS: %08lx CPU: %d\n",
 142                         (u16)regs->cs, regs->ip, regs->flags,
 143                         smp_processor_id());
 144         print_symbol("EIP is at %s\n", regs->ip);
 145
 146         printk(KERN_DEFAULT "EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
 147                 regs->ax, regs->bx, regs->cx, regs->dx);
 148         printk(KERN_DEFAULT "ESI: %08lx EDI: %08lx EBP: %08lx ESP: %08lx\n",
 149                 regs->si, regs->di, regs->bp, sp);
 150         printk(KERN_DEFAULT " DS: %04x ES: %04x FS: %04x GS: %04x SS: %04x\n",
 151                (u16)regs->ds, (u16)regs->es, (u16)regs->fs, gs, ss);
 152
 153         if (!all)
 154                 return;
 155
 156         cr0 = read_cr0();
 157         cr2 = read_cr2();
 158         cr3 = read_cr3();
 159         cr4 = read_cr4_safe();
 160         printk(KERN_DEFAULT "CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n",
 161                         cr0, cr2, cr3, cr4);
 162
 163         get_debugreg(d0, 0);
 164         get_debugreg(d1, 1);
 165         get_debugreg(d2, 2);
 166         get_debugreg(d3, 3);
 167         printk(KERN_DEFAULT "DR0: %08lx DR1: %08lx DR2: %08lx DR3: %08lx\n",
 168                         d0, d1, d2, d3);
 169
 170         get_debugreg(d6, 6);
 171         get_debugreg(d7, 7);
 172         printk(KERN_DEFAULT "DR6: %08lx DR7: %08lx\n",
 173                         d6, d7);
 174 }
 175
 176 void release_thread(struct task_struct *dead_task)
 177 {
 178         BUG_ON(dead_task->mm);
 179         release_vm86_irqs(dead_task);
 180 }
 181
 182 /*
 183  * This gets called before we allocate a new thread and copy
 184  * the current task into it.
 185  */
 186 void prepare_to_copy(struct task_struct *tsk)
 187 {
 188         unlazy_fpu(tsk);
 189 }
 190
 191 int copy_thread(unsigned long clone_flags, unsigned long sp,
 192         unsigned long unused,
 193         struct task_struct *p, struct pt_regs *regs)
 194 {
 195         struct pt_regs *childregs;
 196         struct task_struct *tsk;
 197         int err;
 198
 199         childregs = task_pt_regs(p);
 200         *childregs = *regs;
 201         childregs->ax = 0;
 202         childregs->sp = sp;
 203
 204         p->thread.sp = (unsigned long) childregs;
 205         p->thread.sp0 = (unsigned long) (childregs+1);
 206
 207         p->thread.ip = (unsigned long) ret_from_fork;
 208
 209         task_user_gs(p) = get_user_gs(regs);
 210
 211         p->thread.io_bitmap_ptr = NULL;
 212         tsk = current;
 213         err = -ENOMEM;
 214
 215         memset(p->thread.ptrace_bps, 0, sizeof(p->thread.ptrace_bps));
 216
 217         if (unlikely(test_tsk_thread_flag(tsk, TIF_IO_BITMAP))) {
 218                 p->thread.io_bitmap_ptr = kmemdup(tsk->thread.io_bitmap_ptr,
 219                                                 IO_BITMAP_BYTES, GFP_KERNEL);
 220                 if (!p->thread.io_bitmap_ptr) {
 221                         p->thread.io_bitmap_max = 0;
 222                         return -ENOMEM;
 223                 }
 224                 set_tsk_thread_flag(p, TIF_IO_BITMAP);
 225         }
 226
 227         err = 0;
 228
 229         /*
 230          * Set a new TLS for the child thread?
 231          */
 232         if (clone_flags & CLONE_SETTLS)
 233                 err = do_set_thread_area(p, -1,
 234                         (struct user_desc __user *)childregs->si, 0);
 235
 236         if (err && p->thread.io_bitmap_ptr) {
 237                 kfree(p->thread.io_bitmap_ptr);
 238                 p->thread.io_bitmap_max = 0;
 239         }
 240         return err;
 241 }
 242
 243 void
 244 start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp)
 245 {
 246         set_user_gs(regs, 0);
 247         regs->fs                = 0;
 248         set_fs(USER_DS);
 249         regs->ds                = __USER_DS;
 250         regs->es                = __USER_DS;
 251         regs->ss                = __USER_DS;
 252         regs->cs                = __USER_CS;
 253         regs->ip                = new_ip;
 254         regs->sp                = new_sp;
 255         /*
 256          * Free the old FP and other extended state
 257          */
 258         free_thread_xstate(current);
 259 }
 260 EXPORT_SYMBOL_GPL(start_thread);
 261
 262
 263 /*
 264  *      switch_to(x,yn) should switch tasks from x to y.
 265  *
 266  * We fsave/fwait so that an exception goes off at the right time
 267  * (as a call from the fsave or fwait in effect) rather than to
 268  * the wrong process. Lazy FP saving no longer makes any sense
 269  * with modern CPU's, and this simplifies a lot of things (SMP
 270  * and UP become the same).
 271  *
 272  * NOTE! We used to use the x86 hardware context switching. The
 273  * reason for not using it any more becomes apparent when you
 274  * try to recover gracefully from saved state that is no longer
 275  * valid (stale segment register values in particular). With the
 276  * hardware task-switch, there is no way to fix up bad state in
 277  * a reasonable manner.
 278  *
 279  * The fact that Intel documents the hardware task-switching to
 280  * be slow is a fairly red herring - this code is not noticeably
 281  * faster. However, there _is_ some room for improvement here,
 282  * so the performance issues may eventually be a valid point.
 283  * More important, however, is the fact that this allows us much
 284  * more flexibility.
 285  *
 286  * The return value (in %ax) will be the "prev" task after
 287  * the task-switch, and shows up in ret_from_fork in entry.S,
 288  * for example.
 289  */
 290 __notrace_funcgraph struct task_struct *
 291 __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
 292 {
 293         struct thread_struct *prev = &prev_p->thread,
 294                                  *next = &next_p->thread;
 295         int cpu = smp_processor_id();
 296         struct tss_struct *tss = &per_cpu(init_tss, cpu);
 297         bool preload_fpu;
 298
 299         /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
 300
 301         /*
 302          * If the task has used fpu the last 5 timeslices, just do a full
 303          * restore of the math state immediately to avoid the trap; the
 304          * chances of needing FPU soon are obviously high now
 305          */
 306         preload_fpu = tsk_used_math(next_p) && next_p->fpu_counter > 5;
 307
 308         __unlazy_fpu(prev_p);
 309
 310         /* we're going to use this soon, after a few expensive things */
 311         if (preload_fpu)
 312                 prefetch(next->fpu.state);
 313
 314         /*
 315          * Reload esp0.
 316          */
 317         load_sp0(tss, next);
 318
 319         /*
 320          * Save away %gs. No need to save %fs, as it was saved on the
 321          * stack on entry.  No need to save %es and %ds, as those are
 322          * always kernel segments while inside the kernel.  Doing this
 323          * before setting the new TLS descriptors avoids the situation
 324          * where we temporarily have non-reloadable segments in %fs
 325          * and %gs.  This could be an issue if the NMI handler ever
 326          * used %fs or %gs (it does not today), or if the kernel is
 327          * running inside of a hypervisor layer.
 328          */
 329         lazy_save_gs(prev->gs);
 330
 331         /*
 332          * Load the per-thread Thread-Local Storage descriptor.
 333          */
 334         load_TLS(next, cpu);
 335
 336         /*
 337          * Restore IOPL if needed.  In normal use, the flags restore
 338          * in the switch assembly will handle this.  But if the kernel
 339          * is running virtualized at a non-zero CPL, the popf will
 340          * not restore flags, so it must be done in a separate step.
 341          */
 342         if (get_kernel_rpl() && unlikely(prev->iopl != next->iopl))
 343                 set_iopl_mask(next->iopl);
 344
 345         /*
 346          * Now maybe handle debug registers and/or IO bitmaps
 347          */
 348         if (unlikely(task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV ||
 349                      task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT))
 350                 __switch_to_xtra(prev_p, next_p, tss);
 351
 352         /* If we're going to preload the fpu context, make sure clts
 353            is run while we're batching the cpu state updates. */
 354         if (preload_fpu)
 355                 clts();
 356
 357         /*
 358          * Leave lazy mode, flushing any hypercalls made here.
 359          * This must be done before restoring TLS segments so
 360          * the GDT and LDT are properly updated, and must be
 361          * done before math_state_restore, so the TS bit is up
 362          * to date.
 363          */
 364         arch_end_context_switch(next_p);
 365
 366         if (preload_fpu)
 367                 __math_state_restore();
 368
 369         /*
 370          * Restore %gs if needed (which is common)
 371          */
 372         if (prev->gs | next->gs)
 373                 lazy_load_gs(next->gs);
 374
 375         percpu_write(current_task, next_p);
 376
 377         return prev_p;
 378 }
 379
 380 #define top_esp                (THREAD_SIZE - sizeof(unsigned long))
 381 #define top_ebp                (THREAD_SIZE - 2*sizeof(unsigned long))
 382
 383 unsigned long get_wchan(struct task_struct *p)
 384 {
 385         unsigned long bp, sp, ip;
 386         unsigned long stack_page;
 387         int count = 0;
 388         if (!p || p == current || p->state == TASK_RUNNING)
 389                 return 0;
 390         stack_page = (unsigned long)task_stack_page(p);
 391         sp = p->thread.sp;
 392         if (!stack_page || sp < stack_page || sp > top_esp+stack_page)
 393                 return 0;
 394         /* include/asm-i386/system.h:switch_to() pushes bp last. */
 395         bp = *(unsigned long *) sp;
 396         do {
 397                 if (bp < stack_page || bp > top_ebp+stack_page)
 398                         return 0;
 399                 ip = *(unsigned long *) (bp+4);
 400                 if (!in_sched_functions(ip))
 401                         return ip;
 402                 bp = *(unsigned long *) bp;
 403         } while (count++ < 16);
 404         return 0;
 405 }
 406