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/cpu.h>
  13 #include <linux/errno.h>
  14 #include <linux/sched.h>
  15 #include <linux/fs.h>
  16 #include <linux/kernel.h>
  17 #include <linux/mm.h>
  18 #include <linux/elfcore.h>
  19 #include <linux/smp.h>
  20 #include <linux/stddef.h>
  21 #include <linux/slab.h>
  22 #include <linux/vmalloc.h>
  23 #include <linux/user.h>
  24 #include <linux/interrupt.h>
  25 #include <linux/delay.h>
  26 #include <linux/reboot.h>
  27 #include <linux/mc146818rtc.h>
  28 #include <linux/module.h>
  29 #include <linux/kallsyms.h>
  30 #include <linux/ptrace.h>
  31 #include <linux/personality.h>
  32 #include <linux/percpu.h>
  33 #include <linux/prctl.h>
  34 #include <linux/ftrace.h>
  35 #include <linux/uaccess.h>
  36 #include <linux/io.h>
  37 #include <linux/kdebug.h>
  38
  39 #include <asm/pgtable.h>
  40 #include <asm/ldt.h>
  41 #include <asm/processor.h>
  42 #include <asm/fpu/internal.h>
  43 #include <asm/desc.h>
  44 #ifdef CONFIG_MATH_EMULATION
  45 #include <asm/math_emu.h>
  46 #endif
  47
  48 #include <linux/err.h>
  49
  50 #include <asm/tlbflush.h>
  51 #include <asm/cpu.h>
  52 #include <asm/idle.h>
  53 #include <asm/syscalls.h>
  54 #include <asm/debugreg.h>
  55 #include <asm/switch_to.h>
  56
  57 asmlinkage void ret_from_fork(void) __asm__("ret_from_fork");
  58 asmlinkage void ret_from_kernel_thread(void) __asm__("ret_from_kernel_thread");
  59
  60 /*
  61  * Return saved PC of a blocked thread.
  62  */
  63 unsigned long thread_saved_pc(struct task_struct *tsk)
  64 {
  65         return ((unsigned long *)tsk->thread.sp)[3];
  66 }
  67
  68 void __show_regs(struct pt_regs *regs, int all)
  69 {
  70         unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L;
  71         unsigned long d0, d1, d2, d3, d6, d7;
  72         unsigned long sp;
  73         unsigned short ss, gs;
  74
  75         if (user_mode(regs)) {
  76                 sp = regs->sp;
  77                 ss = regs->ss & 0xffff;
  78                 gs = get_user_gs(regs);
  79         } else {
  80                 sp = kernel_stack_pointer(regs);
  81                 savesegment(ss, ss);
  82                 savesegment(gs, gs);
  83         }
  84
  85         printk(KERN_DEFAULT "EIP: %04x:[<%08lx>] EFLAGS: %08lx CPU: %d\n",
  86                         (u16)regs->cs, regs->ip, regs->flags,
  87                         smp_processor_id());
  88         print_symbol("EIP is at %s\n", regs->ip);
  89
  90         printk(KERN_DEFAULT "EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
  91                 regs->ax, regs->bx, regs->cx, regs->dx);
  92         printk(KERN_DEFAULT "ESI: %08lx EDI: %08lx EBP: %08lx ESP: %08lx\n",
  93                 regs->si, regs->di, regs->bp, sp);
  94         printk(KERN_DEFAULT " DS: %04x ES: %04x FS: %04x GS: %04x SS: %04x\n",
  95                (u16)regs->ds, (u16)regs->es, (u16)regs->fs, gs, ss);
  96
  97         if (!all)
  98                 return;
  99
 100         cr0 = read_cr0();
 101         cr2 = read_cr2();
 102         cr3 = read_cr3();
 103         cr4 = __read_cr4_safe();
 104         printk(KERN_DEFAULT "CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n",
 105                         cr0, cr2, cr3, cr4);
 106
 107         get_debugreg(d0, 0);
 108         get_debugreg(d1, 1);
 109         get_debugreg(d2, 2);
 110         get_debugreg(d3, 3);
 111         get_debugreg(d6, 6);
 112         get_debugreg(d7, 7);
 113
 114         /* Only print out debug registers if they are in their non-default state. */
 115         if ((d0 == 0) && (d1 == 0) && (d2 == 0) && (d3 == 0) &&
 116             (d6 == DR6_RESERVED) && (d7 == 0x400))
 117                 return;
 118
 119         printk(KERN_DEFAULT "DR0: %08lx DR1: %08lx DR2: %08lx DR3: %08lx\n",
 120                         d0, d1, d2, d3);
 121         printk(KERN_DEFAULT "DR6: %08lx DR7: %08lx\n",
 122                         d6, d7);
 123 }
 124
 125 void release_thread(struct task_struct *dead_task)
 126 {
 127         BUG_ON(dead_task->mm);
 128         release_vm86_irqs(dead_task);
 129 }
 130
 131 int copy_thread(unsigned long clone_flags, unsigned long sp,
 132         unsigned long arg, struct task_struct *p)
 133 {
 134         struct pt_regs *childregs = task_pt_regs(p);
 135         struct task_struct *tsk;
 136         int err;
 137
 138         p->thread.sp = (unsigned long) childregs;
 139         p->thread.sp0 = (unsigned long) (childregs+1);
 140         memset(p->thread.ptrace_bps, 0, sizeof(p->thread.ptrace_bps));
 141
 142         if (unlikely(p->flags & PF_KTHREAD)) {
 143                 /* kernel thread */
 144                 memset(childregs, 0, sizeof(struct pt_regs));
 145                 p->thread.ip = (unsigned long) ret_from_kernel_thread;
 146                 task_user_gs(p) = __KERNEL_STACK_CANARY;
 147                 childregs->ds = __USER_DS;
 148                 childregs->es = __USER_DS;
 149                 childregs->fs = __KERNEL_PERCPU;
 150                 childregs->bx = sp;     /* function */
 151                 childregs->bp = arg;
 152                 childregs->orig_ax = -1;
 153                 childregs->cs = __KERNEL_CS | get_kernel_rpl();
 154                 childregs->flags = X86_EFLAGS_IF | X86_EFLAGS_FIXED;
 155                 p->thread.io_bitmap_ptr = NULL;
 156                 return 0;
 157         }
 158         *childregs = *current_pt_regs();
 159         childregs->ax = 0;
 160         if (sp)
 161                 childregs->sp = sp;
 162
 163         p->thread.ip = (unsigned long) ret_from_fork;
 164         task_user_gs(p) = get_user_gs(current_pt_regs());
 165
 166         p->thread.io_bitmap_ptr = NULL;
 167         tsk = current;
 168         err = -ENOMEM;
 169
 170         if (unlikely(test_tsk_thread_flag(tsk, TIF_IO_BITMAP))) {
 171                 p->thread.io_bitmap_ptr = kmemdup(tsk->thread.io_bitmap_ptr,
 172                                                 IO_BITMAP_BYTES, GFP_KERNEL);
 173                 if (!p->thread.io_bitmap_ptr) {
 174                         p->thread.io_bitmap_max = 0;
 175                         return -ENOMEM;
 176                 }
 177                 set_tsk_thread_flag(p, TIF_IO_BITMAP);
 178         }
 179
 180         err = 0;
 181
 182         /*
 183          * Set a new TLS for the child thread?
 184          */
 185         if (clone_flags & CLONE_SETTLS)
 186                 err = do_set_thread_area(p, -1,
 187                         (struct user_desc __user *)childregs->si, 0);
 188
 189         if (err && p->thread.io_bitmap_ptr) {
 190                 kfree(p->thread.io_bitmap_ptr);
 191                 p->thread.io_bitmap_max = 0;
 192         }
 193         return err;
 194 }
 195
 196 void
 197 start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp)
 198 {
 199         set_user_gs(regs, 0);
 200         regs->fs                = 0;
 201         regs->ds                = __USER_DS;
 202         regs->es                = __USER_DS;
 203         regs->ss                = __USER_DS;
 204         regs->cs                = __USER_CS;
 205         regs->ip                = new_ip;
 206         regs->sp                = new_sp;
 207         regs->flags             = X86_EFLAGS_IF;
 208         force_iret();
 209 }
 210 EXPORT_SYMBOL_GPL(start_thread);
 211
 212
 213 /*
 214  *      switch_to(x,y) should switch tasks from x to y.
 215  *
 216  * We fsave/fwait so that an exception goes off at the right time
 217  * (as a call from the fsave or fwait in effect) rather than to
 218  * the wrong process. Lazy FP saving no longer makes any sense
 219  * with modern CPU's, and this simplifies a lot of things (SMP
 220  * and UP become the same).
 221  *
 222  * NOTE! We used to use the x86 hardware context switching. The
 223  * reason for not using it any more becomes apparent when you
 224  * try to recover gracefully from saved state that is no longer
 225  * valid (stale segment register values in particular). With the
 226  * hardware task-switch, there is no way to fix up bad state in
 227  * a reasonable manner.
 228  *
 229  * The fact that Intel documents the hardware task-switching to
 230  * be slow is a fairly red herring - this code is not noticeably
 231  * faster. However, there _is_ some room for improvement here,
 232  * so the performance issues may eventually be a valid point.
 233  * More important, however, is the fact that this allows us much
 234  * more flexibility.
 235  *
 236  * The return value (in %ax) will be the "prev" task after
 237  * the task-switch, and shows up in ret_from_fork in entry.S,
 238  * for example.
 239  */
 240 __visible __notrace_funcgraph struct task_struct *
 241 __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
 242 {
 243         struct thread_struct *prev = &prev_p->thread,
 244                              *next = &next_p->thread;
 245         struct fpu *prev_fpu = &prev->fpu;
 246         struct fpu *next_fpu = &next->fpu;
 247         int cpu = smp_processor_id();
 248         struct tss_struct *tss = &per_cpu(cpu_tss, cpu);
 249         fpu_switch_t fpu_switch;
 250
 251         /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
 252
 253         fpu_switch = switch_fpu_prepare(prev_fpu, next_fpu, cpu);
 254
 255         /*
 256          * Save away %gs. No need to save %fs, as it was saved on the
 257          * stack on entry.  No need to save %es and %ds, as those are
 258          * always kernel segments while inside the kernel.  Doing this
 259          * before setting the new TLS descriptors avoids the situation
 260          * where we temporarily have non-reloadable segments in %fs
 261          * and %gs.  This could be an issue if the NMI handler ever
 262          * used %fs or %gs (it does not today), or if the kernel is
 263          * running inside of a hypervisor layer.
 264          */
 265         lazy_save_gs(prev->gs);
 266
 267         /*
 268          * Load the per-thread Thread-Local Storage descriptor.
 269          */
 270         load_TLS(next, cpu);
 271
 272         /*
 273          * Restore IOPL if needed.  In normal use, the flags restore
 274          * in the switch assembly will handle this.  But if the kernel
 275          * is running virtualized at a non-zero CPL, the popf will
 276          * not restore flags, so it must be done in a separate step.
 277          */
 278         if (get_kernel_rpl() && unlikely(prev->iopl != next->iopl))
 279                 set_iopl_mask(next->iopl);
 280
 281         /*
 282          * If it were not for PREEMPT_ACTIVE we could guarantee that the
 283          * preempt_count of all tasks was equal here and this would not be
 284          * needed.
 285          */
 286         task_thread_info(prev_p)->saved_preempt_count = this_cpu_read(__preempt_count);
 287         this_cpu_write(__preempt_count, task_thread_info(next_p)->saved_preempt_count);
 288
 289         /*
 290          * Now maybe handle debug registers and/or IO bitmaps
 291          */
 292         if (unlikely(task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV ||
 293                      task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT))
 294                 __switch_to_xtra(prev_p, next_p, tss);
 295
 296         /*
 297          * Leave lazy mode, flushing any hypercalls made here.
 298          * This must be done before restoring TLS segments so
 299          * the GDT and LDT are properly updated, and must be
 300          * done before fpu__restore(), so the TS bit is up
 301          * to date.
 302          */
 303         arch_end_context_switch(next_p);
 304
 305         /*
 306          * Reload esp0, kernel_stack, and current_top_of_stack.  This changes
 307          * current_thread_info().
 308          */
 309         load_sp0(tss, next);
 310         this_cpu_write(kernel_stack,
 311                        (unsigned long)task_stack_page(next_p) +
 312                        THREAD_SIZE);
 313         this_cpu_write(cpu_current_top_of_stack,
 314                        (unsigned long)task_stack_page(next_p) +
 315                        THREAD_SIZE);
 316
 317         /*
 318          * Restore %gs if needed (which is common)
 319          */
 320         if (prev->gs | next->gs)
 321                 lazy_load_gs(next->gs);
 322
 323         switch_fpu_finish(next_fpu, fpu_switch);
 324
 325         this_cpu_write(current_task, next_p);
 326
 327         return prev_p;
 328 }
 329
 330 #define top_esp                (THREAD_SIZE - sizeof(unsigned long))
 331 #define top_ebp                (THREAD_SIZE - 2*sizeof(unsigned long))
 332
 333 unsigned long get_wchan(struct task_struct *p)
 334 {
 335         unsigned long bp, sp, ip;
 336         unsigned long stack_page;
 337         int count = 0;
 338         if (!p || p == current || p->state == TASK_RUNNING)
 339                 return 0;
 340         stack_page = (unsigned long)task_stack_page(p);
 341         sp = p->thread.sp;
 342         if (!stack_page || sp < stack_page || sp > top_esp+stack_page)
 343                 return 0;
 344         /* include/asm-i386/system.h:switch_to() pushes bp last. */
 345         bp = *(unsigned long *) sp;
 346         do {
 347                 if (bp < stack_page || bp > top_ebp+stack_page)
 348                         return 0;
 349                 ip = *(unsigned long *) (bp+4);
 350                 if (!in_sched_functions(ip))
 351                         return ip;
 352                 bp = *(unsigned long *) bp;
 353         } while (count++ < 16);
 354         return 0;
 355 }
 356