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/init.h>
  28 #include <linux/mc146818rtc.h>
  29 #include <linux/module.h>
  30 #include <linux/kallsyms.h>
  31 #include <linux/ptrace.h>
  32 #include <linux/personality.h>
  33 #include <linux/percpu.h>
  34 #include <linux/prctl.h>
  35 #include <linux/ftrace.h>
  36 #include <linux/uaccess.h>
  37 #include <linux/io.h>
  38 #include <linux/kdebug.h>
  39
  40 #include <asm/pgtable.h>
  41 #include <asm/ldt.h>
  42 #include <asm/processor.h>
  43 #include <asm/i387.h>
  44 #include <asm/fpu-internal.h>
  45 #include <asm/desc.h>
  46 #ifdef CONFIG_MATH_EMULATION
  47 #include <asm/math_emu.h>
  48 #endif
  49
  50 #include <linux/err.h>
  51
  52 #include <asm/tlbflush.h>
  53 #include <asm/cpu.h>
  54 #include <asm/idle.h>
  55 #include <asm/syscalls.h>
  56 #include <asm/debugreg.h>
  57 #include <asm/switch_to.h>
  58
  59 asmlinkage void ret_from_fork(void) __asm__("ret_from_fork");
  60
  61 /*
  62  * Return saved PC of a blocked thread.
  63  */
  64 unsigned long thread_saved_pc(struct task_struct *tsk)
  65 {
  66         return ((unsigned long *)tsk->thread.sp)[3];
  67 }
  68
  69 void __show_regs(struct pt_regs *regs, int all)
  70 {
  71         unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L;
  72         unsigned long d0, d1, d2, d3, d6, d7;
  73         unsigned long sp;
  74         unsigned short ss, gs;
  75
  76         if (user_mode_vm(regs)) {
  77                 sp = regs->sp;
  78                 ss = regs->ss & 0xffff;
  79                 gs = get_user_gs(regs);
  80         } else {
  81                 sp = kernel_stack_pointer(regs);
  82                 savesegment(ss, ss);
  83                 savesegment(gs, gs);
  84         }
  85
  86         show_regs_common();
  87
  88         printk(KERN_DEFAULT "EIP: %04x:[<%08lx>] EFLAGS: %08lx CPU: %d\n",
  89                         (u16)regs->cs, regs->ip, regs->flags,
  90                         smp_processor_id());
  91         print_symbol("EIP is at %s\n", regs->ip);
  92
  93         printk(KERN_DEFAULT "EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
  94                 regs->ax, regs->bx, regs->cx, regs->dx);
  95         printk(KERN_DEFAULT "ESI: %08lx EDI: %08lx EBP: %08lx ESP: %08lx\n",
  96                 regs->si, regs->di, regs->bp, sp);
  97         printk(KERN_DEFAULT " DS: %04x ES: %04x FS: %04x GS: %04x SS: %04x\n",
  98                (u16)regs->ds, (u16)regs->es, (u16)regs->fs, gs, ss);
  99
 100         if (!all)
 101                 return;
 102
 103         cr0 = read_cr0();
 104         cr2 = read_cr2();
 105         cr3 = read_cr3();
 106         cr4 = read_cr4_safe();
 107         printk(KERN_DEFAULT "CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n",
 108                         cr0, cr2, cr3, cr4);
 109
 110         get_debugreg(d0, 0);
 111         get_debugreg(d1, 1);
 112         get_debugreg(d2, 2);
 113         get_debugreg(d3, 3);
 114         printk(KERN_DEFAULT "DR0: %08lx DR1: %08lx DR2: %08lx DR3: %08lx\n",
 115                         d0, d1, d2, d3);
 116
 117         get_debugreg(d6, 6);
 118         get_debugreg(d7, 7);
 119         printk(KERN_DEFAULT "DR6: %08lx DR7: %08lx\n",
 120                         d6, d7);
 121 }
 122
 123 void release_thread(struct task_struct *dead_task)
 124 {
 125         BUG_ON(dead_task->mm);
 126         release_vm86_irqs(dead_task);
 127 }
 128
 129 int copy_thread(unsigned long clone_flags, unsigned long sp,
 130         unsigned long unused,
 131         struct task_struct *p, struct pt_regs *regs)
 132 {
 133         struct pt_regs *childregs;
 134         struct task_struct *tsk;
 135         int err;
 136
 137         childregs = task_pt_regs(p);
 138         *childregs = *regs;
 139         childregs->ax = 0;
 140         childregs->sp = sp;
 141
 142         p->thread.sp = (unsigned long) childregs;
 143         p->thread.sp0 = (unsigned long) (childregs+1);
 144
 145         p->thread.ip = (unsigned long) ret_from_fork;
 146
 147         task_user_gs(p) = get_user_gs(regs);
 148
 149         p->fpu_counter = 0;
 150         p->thread.io_bitmap_ptr = NULL;
 151         tsk = current;
 152         err = -ENOMEM;
 153
 154         memset(p->thread.ptrace_bps, 0, sizeof(p->thread.ptrace_bps));
 155
 156         if (unlikely(test_tsk_thread_flag(tsk, TIF_IO_BITMAP))) {
 157                 p->thread.io_bitmap_ptr = kmemdup(tsk->thread.io_bitmap_ptr,
 158                                                 IO_BITMAP_BYTES, GFP_KERNEL);
 159                 if (!p->thread.io_bitmap_ptr) {
 160                         p->thread.io_bitmap_max = 0;
 161                         return -ENOMEM;
 162                 }
 163                 set_tsk_thread_flag(p, TIF_IO_BITMAP);
 164         }
 165
 166         err = 0;
 167
 168         /*
 169          * Set a new TLS for the child thread?
 170          */
 171         if (clone_flags & CLONE_SETTLS)
 172                 err = do_set_thread_area(p, -1,
 173                         (struct user_desc __user *)childregs->si, 0);
 174
 175         if (err && p->thread.io_bitmap_ptr) {
 176                 kfree(p->thread.io_bitmap_ptr);
 177                 p->thread.io_bitmap_max = 0;
 178         }
 179         return err;
 180 }
 181
 182 void
 183 start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp)
 184 {
 185         set_user_gs(regs, 0);
 186         regs->fs                = 0;
 187         regs->ds                = __USER_DS;
 188         regs->es                = __USER_DS;
 189         regs->ss                = __USER_DS;
 190         regs->cs                = __USER_CS;
 191         regs->ip                = new_ip;
 192         regs->sp                = new_sp;
 193         /*
 194          * Free the old FP and other extended state
 195          */
 196         free_thread_xstate(current);
 197 }
 198 EXPORT_SYMBOL_GPL(start_thread);
 199
 200
 201 /*
 202  *      switch_to(x,y) should switch tasks from x to y.
 203  *
 204  * We fsave/fwait so that an exception goes off at the right time
 205  * (as a call from the fsave or fwait in effect) rather than to
 206  * the wrong process. Lazy FP saving no longer makes any sense
 207  * with modern CPU's, and this simplifies a lot of things (SMP
 208  * and UP become the same).
 209  *
 210  * NOTE! We used to use the x86 hardware context switching. The
 211  * reason for not using it any more becomes apparent when you
 212  * try to recover gracefully from saved state that is no longer
 213  * valid (stale segment register values in particular). With the
 214  * hardware task-switch, there is no way to fix up bad state in
 215  * a reasonable manner.
 216  *
 217  * The fact that Intel documents the hardware task-switching to
 218  * be slow is a fairly red herring - this code is not noticeably
 219  * faster. However, there _is_ some room for improvement here,
 220  * so the performance issues may eventually be a valid point.
 221  * More important, however, is the fact that this allows us much
 222  * more flexibility.
 223  *
 224  * The return value (in %ax) will be the "prev" task after
 225  * the task-switch, and shows up in ret_from_fork in entry.S,
 226  * for example.
 227  */
 228 __notrace_funcgraph struct task_struct *
 229 __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
 230 {
 231         struct thread_struct *prev = &prev_p->thread,
 232                                  *next = &next_p->thread;
 233         int cpu = smp_processor_id();
 234         struct tss_struct *tss = &per_cpu(init_tss, cpu);
 235         fpu_switch_t fpu;
 236
 237         /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
 238
 239         fpu = switch_fpu_prepare(prev_p, next_p, cpu);
 240
 241         /*
 242          * Reload esp0.
 243          */
 244         load_sp0(tss, next);
 245
 246         /*
 247          * Save away %gs. No need to save %fs, as it was saved on the
 248          * stack on entry.  No need to save %es and %ds, as those are
 249          * always kernel segments while inside the kernel.  Doing this
 250          * before setting the new TLS descriptors avoids the situation
 251          * where we temporarily have non-reloadable segments in %fs
 252          * and %gs.  This could be an issue if the NMI handler ever
 253          * used %fs or %gs (it does not today), or if the kernel is
 254          * running inside of a hypervisor layer.
 255          */
 256         lazy_save_gs(prev->gs);
 257
 258         /*
 259          * Load the per-thread Thread-Local Storage descriptor.
 260          */
 261         load_TLS(next, cpu);
 262
 263         /*
 264          * Restore IOPL if needed.  In normal use, the flags restore
 265          * in the switch assembly will handle this.  But if the kernel
 266          * is running virtualized at a non-zero CPL, the popf will
 267          * not restore flags, so it must be done in a separate step.
 268          */
 269         if (get_kernel_rpl() && unlikely(prev->iopl != next->iopl))
 270                 set_iopl_mask(next->iopl);
 271
 272         /*
 273          * Now maybe handle debug registers and/or IO bitmaps
 274          */
 275         if (unlikely(task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV ||
 276                      task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT))
 277                 __switch_to_xtra(prev_p, next_p, tss);
 278
 279         /*
 280          * Leave lazy mode, flushing any hypercalls made here.
 281          * This must be done before restoring TLS segments so
 282          * the GDT and LDT are properly updated, and must be
 283          * done before math_state_restore, so the TS bit is up
 284          * to date.
 285          */
 286         arch_end_context_switch(next_p);
 287
 288         /*
 289          * Restore %gs if needed (which is common)
 290          */
 291         if (prev->gs | next->gs)
 292                 lazy_load_gs(next->gs);
 293
 294         switch_fpu_finish(next_p, fpu);
 295
 296         this_cpu_write(current_task, next_p);
 297
 298         return prev_p;
 299 }
 300
 301 #define top_esp                (THREAD_SIZE - sizeof(unsigned long))
 302 #define top_ebp                (THREAD_SIZE - 2*sizeof(unsigned long))
 303
 304 unsigned long get_wchan(struct task_struct *p)
 305 {
 306         unsigned long bp, sp, ip;
 307         unsigned long stack_page;
 308         int count = 0;
 309         if (!p || p == current || p->state == TASK_RUNNING)
 310                 return 0;
 311         stack_page = (unsigned long)task_stack_page(p);
 312         sp = p->thread.sp;
 313         if (!stack_page || sp < stack_page || sp > top_esp+stack_page)
 314                 return 0;
 315         /* include/asm-i386/system.h:switch_to() pushes bp last. */
 316         bp = *(unsigned long *) sp;
 317         do {
 318                 if (bp < stack_page || bp > top_ebp+stack_page)
 319                         return 0;
 320                 ip = *(unsigned long *) (bp+4);
 321                 if (!in_sched_functions(ip))
 322                         return ip;
 323                 bp = *(unsigned long *) bp;
 324         } while (count++ < 16);
 325         return 0;
 326 }
 327