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