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 #include <linux/cpuidle.h>
  42
  43 #include <asm/pgtable.h>
  44 #include <asm/system.h>
  45 #include <asm/ldt.h>
  46 #include <asm/processor.h>
  47 #include <asm/i387.h>
  48 #include <asm/desc.h>
  49 #ifdef CONFIG_MATH_EMULATION
  50 #include <asm/math_emu.h>
  51 #endif
  52
  53 #include <linux/err.h>
  54
  55 #include <asm/tlbflush.h>
  56 #include <asm/cpu.h>
  57 #include <asm/idle.h>
  58 #include <asm/syscalls.h>
  59 #include <asm/debugreg.h>
  60 #include <asm/nmi.h>
  61
  62 asmlinkage void ret_from_fork(void) __asm__("ret_from_fork");
  63
  64 /*
  65  * Return saved PC of a blocked thread.
  66  */
  67 unsigned long thread_saved_pc(struct task_struct *tsk)
  68 {
  69         return ((unsigned long *)tsk->thread.sp)[3];
  70 }
  71
  72 #ifndef CONFIG_SMP
  73 static inline void play_dead(void)
  74 {
  75         BUG();
  76 }
  77 #endif
  78
  79 /*
  80  * The idle thread. There's no useful work to be
  81  * done, so just try to conserve power and have a
  82  * low exit latency (ie sit in a loop waiting for
  83  * somebody to say that they'd like to reschedule)
  84  */
  85 void cpu_idle(void)
  86 {
  87         int cpu = smp_processor_id();
  88
  89         /*
  90          * If we're the non-boot CPU, nothing set the stack canary up
  91          * for us.  CPU0 already has it initialized but no harm in
  92          * doing it again.  This is a good place for updating it, as
  93          * we wont ever return from this function (so the invalid
  94          * canaries already on the stack wont ever trigger).
  95          */
  96         boot_init_stack_canary();
  97
  98         current_thread_info()->status |= TS_POLLING;
  99
 100         /* endless idle loop with no priority at all */
 101         while (1) {
 102                 tick_nohz_idle_enter();
 103                 rcu_idle_enter();
 104                 while (!need_resched()) {
 105
 106                         check_pgt_cache();
 107                         rmb();
 108
 109                         if (cpu_is_offline(cpu))
 110                                 play_dead();
 111
 112                         local_touch_nmi();
 113                         local_irq_disable();
 114                         /* Don't trace irqs off for idle */
 115                         stop_critical_timings();
 116                         if (cpuidle_idle_call())
 117                                 pm_idle();
 118                         start_critical_timings();
 119                 }
 120                 rcu_idle_exit();
 121                 tick_nohz_idle_exit();
 122                 preempt_enable_no_resched();
 123                 schedule();
 124                 preempt_disable();
 125         }
 126 }
 127
 128 void __show_regs(struct pt_regs *regs, int all)
 129 {
 130         unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L;
 131         unsigned long d0, d1, d2, d3, d6, d7;
 132         unsigned long sp;
 133         unsigned short ss, gs;
 134
 135         if (user_mode_vm(regs)) {
 136                 sp = regs->sp;
 137                 ss = regs->ss & 0xffff;
 138                 gs = get_user_gs(regs);
 139         } else {
 140                 sp = kernel_stack_pointer(regs);
 141                 savesegment(ss, ss);
 142                 savesegment(gs, gs);
 143         }
 144
 145         show_regs_common();
 146
 147         printk(KERN_DEFAULT "EIP: %04x:[<%08lx>] EFLAGS: %08lx CPU: %d\n",
 148                         (u16)regs->cs, regs->ip, regs->flags,
 149                         smp_processor_id());
 150         print_symbol("EIP is at %s\n", regs->ip);
 151
 152         printk(KERN_DEFAULT "EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
 153                 regs->ax, regs->bx, regs->cx, regs->dx);
 154         printk(KERN_DEFAULT "ESI: %08lx EDI: %08lx EBP: %08lx ESP: %08lx\n",
 155                 regs->si, regs->di, regs->bp, sp);
 156         printk(KERN_DEFAULT " DS: %04x ES: %04x FS: %04x GS: %04x SS: %04x\n",
 157                (u16)regs->ds, (u16)regs->es, (u16)regs->fs, gs, ss);
 158
 159         if (!all)
 160                 return;
 161
 162         cr0 = read_cr0();
 163         cr2 = read_cr2();
 164         cr3 = read_cr3();
 165         cr4 = read_cr4_safe();
 166         printk(KERN_DEFAULT "CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n",
 167                         cr0, cr2, cr3, cr4);
 168
 169         get_debugreg(d0, 0);
 170         get_debugreg(d1, 1);
 171         get_debugreg(d2, 2);
 172         get_debugreg(d3, 3);
 173         printk(KERN_DEFAULT "DR0: %08lx DR1: %08lx DR2: %08lx DR3: %08lx\n",
 174                         d0, d1, d2, d3);
 175
 176         get_debugreg(d6, 6);
 177         get_debugreg(d7, 7);
 178         printk(KERN_DEFAULT "DR6: %08lx DR7: %08lx\n",
 179                         d6, d7);
 180 }
 181
 182 void release_thread(struct task_struct *dead_task)
 183 {
 184         BUG_ON(dead_task->mm);
 185         release_vm86_irqs(dead_task);
 186 }
 187
 188 /*
 189  * This gets called before we allocate a new thread and copy
 190  * the current task into it.
 191  */
 192 void prepare_to_copy(struct task_struct *tsk)
 193 {
 194         unlazy_fpu(tsk);
 195 }
 196
 197 int copy_thread(unsigned long clone_flags, unsigned long sp,
 198         unsigned long unused,
 199         struct task_struct *p, struct pt_regs *regs)
 200 {
 201         struct pt_regs *childregs;
 202         struct task_struct *tsk;
 203         int err;
 204
 205         childregs = task_pt_regs(p);
 206         *childregs = *regs;
 207         childregs->ax = 0;
 208         childregs->sp = sp;
 209
 210         p->thread.sp = (unsigned long) childregs;
 211         p->thread.sp0 = (unsigned long) (childregs+1);
 212
 213         p->thread.ip = (unsigned long) ret_from_fork;
 214
 215         task_user_gs(p) = get_user_gs(regs);
 216
 217         p->thread.io_bitmap_ptr = NULL;
 218         tsk = current;
 219         err = -ENOMEM;
 220
 221         memset(p->thread.ptrace_bps, 0, sizeof(p->thread.ptrace_bps));
 222
 223         if (unlikely(test_tsk_thread_flag(tsk, TIF_IO_BITMAP))) {
 224                 p->thread.io_bitmap_ptr = kmemdup(tsk->thread.io_bitmap_ptr,
 225                                                 IO_BITMAP_BYTES, GFP_KERNEL);
 226                 if (!p->thread.io_bitmap_ptr) {
 227                         p->thread.io_bitmap_max = 0;
 228                         return -ENOMEM;
 229                 }
 230                 set_tsk_thread_flag(p, TIF_IO_BITMAP);
 231         }
 232
 233         err = 0;
 234
 235         /*
 236          * Set a new TLS for the child thread?
 237          */
 238         if (clone_flags & CLONE_SETTLS)
 239                 err = do_set_thread_area(p, -1,
 240                         (struct user_desc __user *)childregs->si, 0);
 241
 242         if (err && p->thread.io_bitmap_ptr) {
 243                 kfree(p->thread.io_bitmap_ptr);
 244                 p->thread.io_bitmap_max = 0;
 245         }
 246         return err;
 247 }
 248
 249 void
 250 start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp)
 251 {
 252         set_user_gs(regs, 0);
 253         regs->fs                = 0;
 254         regs->ds                = __USER_DS;
 255         regs->es                = __USER_DS;
 256         regs->ss                = __USER_DS;
 257         regs->cs                = __USER_CS;
 258         regs->ip                = new_ip;
 259         regs->sp                = new_sp;
 260         /*
 261          * Free the old FP and other extended state
 262          */
 263         free_thread_xstate(current);
 264 }
 265 EXPORT_SYMBOL_GPL(start_thread);
 266
 267
 268 /*
 269  *      switch_to(x,y) should switch tasks from x to y.
 270  *
 271  * We fsave/fwait so that an exception goes off at the right time
 272  * (as a call from the fsave or fwait in effect) rather than to
 273  * the wrong process. Lazy FP saving no longer makes any sense
 274  * with modern CPU's, and this simplifies a lot of things (SMP
 275  * and UP become the same).
 276  *
 277  * NOTE! We used to use the x86 hardware context switching. The
 278  * reason for not using it any more becomes apparent when you
 279  * try to recover gracefully from saved state that is no longer
 280  * valid (stale segment register values in particular). With the
 281  * hardware task-switch, there is no way to fix up bad state in
 282  * a reasonable manner.
 283  *
 284  * The fact that Intel documents the hardware task-switching to
 285  * be slow is a fairly red herring - this code is not noticeably
 286  * faster. However, there _is_ some room for improvement here,
 287  * so the performance issues may eventually be a valid point.
 288  * More important, however, is the fact that this allows us much
 289  * more flexibility.
 290  *
 291  * The return value (in %ax) will be the "prev" task after
 292  * the task-switch, and shows up in ret_from_fork in entry.S,
 293  * for example.
 294  */
 295 __notrace_funcgraph struct task_struct *
 296 __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
 297 {
 298         struct thread_struct *prev = &prev_p->thread,
 299                                  *next = &next_p->thread;
 300         int cpu = smp_processor_id();
 301         struct tss_struct *tss = &per_cpu(init_tss, cpu);
 302         bool preload_fpu;
 303
 304         /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
 305
 306         /*
 307          * If the task has used fpu the last 5 timeslices, just do a full
 308          * restore of the math state immediately to avoid the trap; the
 309          * chances of needing FPU soon are obviously high now
 310          */
 311         preload_fpu = tsk_used_math(next_p) && next_p->fpu_counter > 5;
 312
 313         __unlazy_fpu(prev_p);
 314
 315         /* we're going to use this soon, after a few expensive things */
 316         if (preload_fpu)
 317                 prefetch(next->fpu.state);
 318
 319         /*
 320          * Reload esp0.
 321          */
 322         load_sp0(tss, next);
 323
 324         /*
 325          * Save away %gs. No need to save %fs, as it was saved on the
 326          * stack on entry.  No need to save %es and %ds, as those are
 327          * always kernel segments while inside the kernel.  Doing this
 328          * before setting the new TLS descriptors avoids the situation
 329          * where we temporarily have non-reloadable segments in %fs
 330          * and %gs.  This could be an issue if the NMI handler ever
 331          * used %fs or %gs (it does not today), or if the kernel is
 332          * running inside of a hypervisor layer.
 333          */
 334         lazy_save_gs(prev->gs);
 335
 336         /*
 337          * Load the per-thread Thread-Local Storage descriptor.
 338          */
 339         load_TLS(next, cpu);
 340
 341         /*
 342          * Restore IOPL if needed.  In normal use, the flags restore
 343          * in the switch assembly will handle this.  But if the kernel
 344          * is running virtualized at a non-zero CPL, the popf will
 345          * not restore flags, so it must be done in a separate step.
 346          */
 347         if (get_kernel_rpl() && unlikely(prev->iopl != next->iopl))
 348                 set_iopl_mask(next->iopl);
 349
 350         /*
 351          * Now maybe handle debug registers and/or IO bitmaps
 352          */
 353         if (unlikely(task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV ||
 354                      task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT))
 355                 __switch_to_xtra(prev_p, next_p, tss);
 356
 357         /* If we're going to preload the fpu context, make sure clts
 358            is run while we're batching the cpu state updates. */
 359         if (preload_fpu)
 360                 clts();
 361
 362         /*
 363          * Leave lazy mode, flushing any hypercalls made here.
 364          * This must be done before restoring TLS segments so
 365          * the GDT and LDT are properly updated, and must be
 366          * done before math_state_restore, so the TS bit is up
 367          * to date.
 368          */
 369         arch_end_context_switch(next_p);
 370
 371         if (preload_fpu)
 372                 __math_state_restore();
 373
 374         /*
 375          * Restore %gs if needed (which is common)
 376          */
 377         if (prev->gs | next->gs)
 378                 lazy_load_gs(next->gs);
 379
 380         percpu_write(current_task, next_p);
 381
 382         return prev_p;
 383 }
 384
 385 #define top_esp                (THREAD_SIZE - sizeof(unsigned long))
 386 #define top_ebp                (THREAD_SIZE - 2*sizeof(unsigned long))
 387
 388 unsigned long get_wchan(struct task_struct *p)
 389 {
 390         unsigned long bp, sp, ip;
 391         unsigned long stack_page;
 392         int count = 0;
 393         if (!p || p == current || p->state == TASK_RUNNING)
 394                 return 0;
 395         stack_page = (unsigned long)task_stack_page(p);
 396         sp = p->thread.sp;
 397         if (!stack_page || sp < stack_page || sp > top_esp+stack_page)
 398                 return 0;
 399         /* include/asm-i386/system.h:switch_to() pushes bp last. */
 400         bp = *(unsigned long *) sp;
 401         do {
 402                 if (bp < stack_page || bp > top_ebp+stack_page)
 403                         return 0;
 404                 ip = *(unsigned long *) (bp+4);
 405                 if (!in_sched_functions(ip))
 406                         return ip;
 407                 bp = *(unsigned long *) bp;
 408         } while (count++ < 16);
 409         return 0;
 410 }
 411