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/ds.h>
  59 #include <asm/debugreg.h>
  60
  61 asmlinkage void ret_from_fork(void) __asm__("ret_from_fork");
  62
  63 /*
  64  * Return saved PC of a blocked thread.
  65  */
  66 unsigned long thread_saved_pc(struct task_struct *tsk)
  67 {
  68         return ((unsigned long *)tsk->thread.sp)[3];
  69 }
  70
  71 #ifndef CONFIG_SMP
  72 static inline void play_dead(void)
  73 {
  74         BUG();
  75 }
  76 #endif
  77
  78 /*
  79  * The idle thread. There's no useful work to be
  80  * done, so just try to conserve power and have a
  81  * low exit latency (ie sit in a loop waiting for
  82  * somebody to say that they'd like to reschedule)
  83  */
  84 void cpu_idle(void)
  85 {
  86         int cpu = smp_processor_id();
  87
  88         /*
  89          * If we're the non-boot CPU, nothing set the stack canary up
  90          * for us.  CPU0 already has it initialized but no harm in
  91          * doing it again.  This is a good place for updating it, as
  92          * we wont ever return from this function (so the invalid
  93          * canaries already on the stack wont ever trigger).
  94          */
  95         boot_init_stack_canary();
  96
  97         current_thread_info()->status |= TS_POLLING;
  98
  99         /* endless idle loop with no priority at all */
 100         while (1) {
 101                 tick_nohz_stop_sched_tick(1);
 102                 while (!need_resched()) {
 103
 104                         check_pgt_cache();
 105                         rmb();
 106
 107                         if (cpu_is_offline(cpu))
 108                                 play_dead();
 109
 110                         local_irq_disable();
 111                         /* Don't trace irqs off for idle */
 112                         stop_critical_timings();
 113                         pm_idle();
 114                         start_critical_timings();
 115                 }
 116                 tick_nohz_restart_sched_tick();
 117                 preempt_enable_no_resched();
 118                 schedule();
 119                 preempt_disable();
 120         }
 121 }
 122
 123 void __show_regs(struct pt_regs *regs, int all)
 124 {
 125         unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L;
 126         unsigned long d0, d1, d2, d3, d6, d7;
 127         unsigned long sp;
 128         unsigned short ss, gs;
 129
 130         if (user_mode_vm(regs)) {
 131                 sp = regs->sp;
 132                 ss = regs->ss & 0xffff;
 133                 gs = get_user_gs(regs);
 134         } else {
 135                 sp = kernel_stack_pointer(regs);
 136                 savesegment(ss, ss);
 137                 savesegment(gs, gs);
 138         }
 139
 140         show_regs_common();
 141
 142         printk(KERN_DEFAULT "EIP: %04x:[<%08lx>] EFLAGS: %08lx CPU: %d\n",
 143                         (u16)regs->cs, regs->ip, regs->flags,
 144                         smp_processor_id());
 145         print_symbol("EIP is at %s\n", regs->ip);
 146
 147         printk(KERN_DEFAULT "EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
 148                 regs->ax, regs->bx, regs->cx, regs->dx);
 149         printk(KERN_DEFAULT "ESI: %08lx EDI: %08lx EBP: %08lx ESP: %08lx\n",
 150                 regs->si, regs->di, regs->bp, sp);
 151         printk(KERN_DEFAULT " DS: %04x ES: %04x FS: %04x GS: %04x SS: %04x\n",
 152                (u16)regs->ds, (u16)regs->es, (u16)regs->fs, gs, ss);
 153
 154         if (!all)
 155                 return;
 156
 157         cr0 = read_cr0();
 158         cr2 = read_cr2();
 159         cr3 = read_cr3();
 160         cr4 = read_cr4_safe();
 161         printk(KERN_DEFAULT "CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n",
 162                         cr0, cr2, cr3, cr4);
 163
 164         get_debugreg(d0, 0);
 165         get_debugreg(d1, 1);
 166         get_debugreg(d2, 2);
 167         get_debugreg(d3, 3);
 168         printk(KERN_DEFAULT "DR0: %08lx DR1: %08lx DR2: %08lx DR3: %08lx\n",
 169                         d0, d1, d2, d3);
 170
 171         get_debugreg(d6, 6);
 172         get_debugreg(d7, 7);
 173         printk(KERN_DEFAULT "DR6: %08lx DR7: %08lx\n",
 174                         d6, d7);
 175 }
 176
 177 void release_thread(struct task_struct *dead_task)
 178 {
 179         BUG_ON(dead_task->mm);
 180         release_vm86_irqs(dead_task);
 181 }
 182
 183 /*
 184  * This gets called before we allocate a new thread and copy
 185  * the current task into it.
 186  */
 187 void prepare_to_copy(struct task_struct *tsk)
 188 {
 189         unlazy_fpu(tsk);
 190 }
 191
 192 int copy_thread(unsigned long clone_flags, unsigned long sp,
 193         unsigned long unused,
 194         struct task_struct *p, struct pt_regs *regs)
 195 {
 196         struct pt_regs *childregs;
 197         struct task_struct *tsk;
 198         int err;
 199
 200         childregs = task_pt_regs(p);
 201         *childregs = *regs;
 202         childregs->ax = 0;
 203         childregs->sp = sp;
 204
 205         p->thread.sp = (unsigned long) childregs;
 206         p->thread.sp0 = (unsigned long) (childregs+1);
 207
 208         p->thread.ip = (unsigned long) ret_from_fork;
 209
 210         task_user_gs(p) = get_user_gs(regs);
 211
 212         p->thread.io_bitmap_ptr = NULL;
 213         tsk = current;
 214         err = -ENOMEM;
 215
 216         memset(p->thread.ptrace_bps, 0, sizeof(p->thread.ptrace_bps));
 217
 218         if (unlikely(test_tsk_thread_flag(tsk, TIF_IO_BITMAP))) {
 219                 p->thread.io_bitmap_ptr = kmemdup(tsk->thread.io_bitmap_ptr,
 220                                                 IO_BITMAP_BYTES, GFP_KERNEL);
 221                 if (!p->thread.io_bitmap_ptr) {
 222                         p->thread.io_bitmap_max = 0;
 223                         return -ENOMEM;
 224                 }
 225                 set_tsk_thread_flag(p, TIF_IO_BITMAP);
 226         }
 227
 228         err = 0;
 229
 230         /*
 231          * Set a new TLS for the child thread?
 232          */
 233         if (clone_flags & CLONE_SETTLS)
 234                 err = do_set_thread_area(p, -1,
 235                         (struct user_desc __user *)childregs->si, 0);
 236
 237         if (err && p->thread.io_bitmap_ptr) {
 238                 kfree(p->thread.io_bitmap_ptr);
 239                 p->thread.io_bitmap_max = 0;
 240         }
 241
 242         clear_tsk_thread_flag(p, TIF_DS_AREA_MSR);
 243         p->thread.ds_ctx = NULL;
 244
 245         clear_tsk_thread_flag(p, TIF_DEBUGCTLMSR);
 246         p->thread.debugctlmsr = 0;
 247
 248         return err;
 249 }
 250
 251 void
 252 start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp)
 253 {
 254         set_user_gs(regs, 0);
 255         regs->fs                = 0;
 256         set_fs(USER_DS);
 257         regs->ds                = __USER_DS;
 258         regs->es                = __USER_DS;
 259         regs->ss                = __USER_DS;
 260         regs->cs                = __USER_CS;
 261         regs->ip                = new_ip;
 262         regs->sp                = new_sp;
 263         /*
 264          * Free the old FP and other extended state
 265          */
 266         free_thread_xstate(current);
 267 }
 268 EXPORT_SYMBOL_GPL(start_thread);
 269
 270
 271 /*
 272  *      switch_to(x,yn) should switch tasks from x to y.
 273  *
 274  * We fsave/fwait so that an exception goes off at the right time
 275  * (as a call from the fsave or fwait in effect) rather than to
 276  * the wrong process. Lazy FP saving no longer makes any sense
 277  * with modern CPU's, and this simplifies a lot of things (SMP
 278  * and UP become the same).
 279  *
 280  * NOTE! We used to use the x86 hardware context switching. The
 281  * reason for not using it any more becomes apparent when you
 282  * try to recover gracefully from saved state that is no longer
 283  * valid (stale segment register values in particular). With the
 284  * hardware task-switch, there is no way to fix up bad state in
 285  * a reasonable manner.
 286  *
 287  * The fact that Intel documents the hardware task-switching to
 288  * be slow is a fairly red herring - this code is not noticeably
 289  * faster. However, there _is_ some room for improvement here,
 290  * so the performance issues may eventually be a valid point.
 291  * More important, however, is the fact that this allows us much
 292  * more flexibility.
 293  *
 294  * The return value (in %ax) will be the "prev" task after
 295  * the task-switch, and shows up in ret_from_fork in entry.S,
 296  * for example.
 297  */
 298 __notrace_funcgraph struct task_struct *
 299 __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
 300 {
 301         struct thread_struct *prev = &prev_p->thread,
 302                                  *next = &next_p->thread;
 303         int cpu = smp_processor_id();
 304         struct tss_struct *tss = &per_cpu(init_tss, cpu);
 305         bool preload_fpu;
 306
 307         /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
 308
 309         /*
 310          * If the task has used fpu the last 5 timeslices, just do a full
 311          * restore of the math state immediately to avoid the trap; the
 312          * chances of needing FPU soon are obviously high now
 313          */
 314         preload_fpu = tsk_used_math(next_p) && next_p->fpu_counter > 5;
 315
 316         __unlazy_fpu(prev_p);
 317
 318         /* we're going to use this soon, after a few expensive things */
 319         if (preload_fpu)
 320                 prefetch(next->xstate);
 321
 322         /*
 323          * Reload esp0.
 324          */
 325         load_sp0(tss, next);
 326
 327         /*
 328          * Save away %gs. No need to save %fs, as it was saved on the
 329          * stack on entry.  No need to save %es and %ds, as those are
 330          * always kernel segments while inside the kernel.  Doing this
 331          * before setting the new TLS descriptors avoids the situation
 332          * where we temporarily have non-reloadable segments in %fs
 333          * and %gs.  This could be an issue if the NMI handler ever
 334          * used %fs or %gs (it does not today), or if the kernel is
 335          * running inside of a hypervisor layer.
 336          */
 337         lazy_save_gs(prev->gs);
 338
 339         /*
 340          * Load the per-thread Thread-Local Storage descriptor.
 341          */
 342         load_TLS(next, cpu);
 343
 344         /*
 345          * Restore IOPL if needed.  In normal use, the flags restore
 346          * in the switch assembly will handle this.  But if the kernel
 347          * is running virtualized at a non-zero CPL, the popf will
 348          * not restore flags, so it must be done in a separate step.
 349          */
 350         if (get_kernel_rpl() && unlikely(prev->iopl != next->iopl))
 351                 set_iopl_mask(next->iopl);
 352
 353         /*
 354          * Now maybe handle debug registers and/or IO bitmaps
 355          */
 356         if (unlikely(task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV ||
 357                      task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT))
 358                 __switch_to_xtra(prev_p, next_p, tss);
 359
 360         /* If we're going to preload the fpu context, make sure clts
 361            is run while we're batching the cpu state updates. */
 362         if (preload_fpu)
 363                 clts();
 364
 365         /*
 366          * Leave lazy mode, flushing any hypercalls made here.
 367          * This must be done before restoring TLS segments so
 368          * the GDT and LDT are properly updated, and must be
 369          * done before math_state_restore, so the TS bit is up
 370          * to date.
 371          */
 372         arch_end_context_switch(next_p);
 373
 374         if (preload_fpu)
 375                 __math_state_restore();
 376
 377         /*
 378          * Restore %gs if needed (which is common)
 379          */
 380         if (prev->gs | next->gs)
 381                 lazy_load_gs(next->gs);
 382
 383         percpu_write(current_task, next_p);
 384
 385         return prev_p;
 386 }
 387
 388 #define top_esp                (THREAD_SIZE - sizeof(unsigned long))
 389 #define top_ebp                (THREAD_SIZE - 2*sizeof(unsigned long))
 390
 391 unsigned long get_wchan(struct task_struct *p)
 392 {
 393         unsigned long bp, sp, ip;
 394         unsigned long stack_page;
 395         int count = 0;
 396         if (!p || p == current || p->state == TASK_RUNNING)
 397                 return 0;
 398         stack_page = (unsigned long)task_stack_page(p);
 399         sp = p->thread.sp;
 400         if (!stack_page || sp < stack_page || sp > top_esp+stack_page)
 401                 return 0;
 402         /* include/asm-i386/system.h:switch_to() pushes bp last. */
 403         bp = *(unsigned long *) sp;
 404         do {
 405                 if (bp < stack_page || bp > top_ebp+stack_page)
 406                         return 0;
 407                 ip = *(unsigned long *) (bp+4);
 408                 if (!in_sched_functions(ip))
 409                         return ip;
 410                 bp = *(unsigned long *) bp;
 411         } while (count++ < 16);
 412         return 0;
 413 }
 414