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/export.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 #include <asm/vm86.h>
  57
  58 #include "process.h"
  59
  60 void __show_regs(struct pt_regs *regs, int all)
  61 {
  62         unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L;
  63         unsigned long d0, d1, d2, d3, d6, d7;
  64         unsigned long sp;
  65         unsigned short ss, gs;
  66
  67         if (user_mode(regs)) {
  68                 sp = regs->sp;
  69                 ss = regs->ss & 0xffff;
  70                 gs = get_user_gs(regs);
  71         } else {
  72                 sp = kernel_stack_pointer(regs);
  73                 savesegment(ss, ss);
  74                 savesegment(gs, gs);
  75         }
  76
  77         printk(KERN_DEFAULT "EIP: %04x:[<%08lx>] EFLAGS: %08lx CPU: %d\n",
  78                         (u16)regs->cs, regs->ip, regs->flags,
  79                         smp_processor_id());
  80         print_symbol("EIP is at %s\n", regs->ip);
  81
  82         printk(KERN_DEFAULT "EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
  83                 regs->ax, regs->bx, regs->cx, regs->dx);
  84         printk(KERN_DEFAULT "ESI: %08lx EDI: %08lx EBP: %08lx ESP: %08lx\n",
  85                 regs->si, regs->di, regs->bp, sp);
  86         printk(KERN_DEFAULT " DS: %04x ES: %04x FS: %04x GS: %04x SS: %04x\n",
  87                (u16)regs->ds, (u16)regs->es, (u16)regs->fs, gs, ss);
  88
  89         if (!all)
  90                 return;
  91
  92         cr0 = read_cr0();
  93         cr2 = read_cr2();
  94         cr3 = read_cr3();
  95         cr4 = __read_cr4();
  96         printk(KERN_DEFAULT "CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n",
  97                         cr0, cr2, cr3, cr4);
  98
  99         get_debugreg(d0, 0);
 100         get_debugreg(d1, 1);
 101         get_debugreg(d2, 2);
 102         get_debugreg(d3, 3);
 103         get_debugreg(d6, 6);
 104         get_debugreg(d7, 7);
 105
 106         /* Only print out debug registers if they are in their non-default state. */
 107         if ((d0 == 0) && (d1 == 0) && (d2 == 0) && (d3 == 0) &&
 108             (d6 == DR6_RESERVED) && (d7 == 0x400))
 109                 return;
 110
 111         printk(KERN_DEFAULT "DR0: %08lx DR1: %08lx DR2: %08lx DR3: %08lx\n",
 112                         d0, d1, d2, d3);
 113         printk(KERN_DEFAULT "DR6: %08lx DR7: %08lx\n",
 114                         d6, d7);
 115 }
 116
 117 void release_thread(struct task_struct *dead_task)
 118 {
 119         BUG_ON(dead_task->mm);
 120         release_vm86_irqs(dead_task);
 121 }
 122
 123 int copy_thread_tls(unsigned long clone_flags, unsigned long sp,
 124         unsigned long arg, struct task_struct *p, unsigned long tls)
 125 {
 126         struct pt_regs *childregs = task_pt_regs(p);
 127         struct fork_frame *fork_frame = container_of(childregs, struct fork_frame, regs);
 128         struct inactive_task_frame *frame = &fork_frame->frame;
 129         struct task_struct *tsk;
 130         int err;
 131
 132         /*
 133          * For a new task use the RESET flags value since there is no before.
 134          * All the status flags are zero; DF and all the system flags must also
 135          * be 0, specifically IF must be 0 because we context switch to the new
 136          * task with interrupts disabled.
 137          */
 138         frame->flags = X86_EFLAGS_FIXED;
 139         frame->bp = 0;
 140         frame->ret_addr = (unsigned long) ret_from_fork;
 141         p->thread.sp = (unsigned long) fork_frame;
 142         p->thread.sp0 = (unsigned long) (childregs+1);
 143         memset(p->thread.ptrace_bps, 0, sizeof(p->thread.ptrace_bps));
 144
 145         if (unlikely(p->flags & PF_KTHREAD)) {
 146                 /* kernel thread */
 147                 memset(childregs, 0, sizeof(struct pt_regs));
 148                 frame->bx = sp;         /* function */
 149                 frame->di = arg;
 150                 p->thread.io_bitmap_ptr = NULL;
 151                 return 0;
 152         }
 153         frame->bx = 0;
 154         *childregs = *current_pt_regs();
 155         childregs->ax = 0;
 156         if (sp)
 157                 childregs->sp = sp;
 158
 159         task_user_gs(p) = get_user_gs(current_pt_regs());
 160
 161         p->thread.io_bitmap_ptr = NULL;
 162         tsk = current;
 163         err = -ENOMEM;
 164
 165         if (unlikely(test_tsk_thread_flag(tsk, TIF_IO_BITMAP))) {
 166                 p->thread.io_bitmap_ptr = kmemdup(tsk->thread.io_bitmap_ptr,
 167                                                 IO_BITMAP_BYTES, GFP_KERNEL);
 168                 if (!p->thread.io_bitmap_ptr) {
 169                         p->thread.io_bitmap_max = 0;
 170                         return -ENOMEM;
 171                 }
 172                 set_tsk_thread_flag(p, TIF_IO_BITMAP);
 173         }
 174
 175         err = 0;
 176
 177         /*
 178          * Set a new TLS for the child thread?
 179          */
 180         if (clone_flags & CLONE_SETTLS)
 181                 err = do_set_thread_area(p, -1,
 182                         (struct user_desc __user *)tls, 0);
 183
 184         if (err && p->thread.io_bitmap_ptr) {
 185                 kfree(p->thread.io_bitmap_ptr);
 186                 p->thread.io_bitmap_max = 0;
 187         }
 188         return err;
 189 }
 190
 191 void
 192 start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp)
 193 {
 194         set_user_gs(regs, 0);
 195         regs->fs                = 0;
 196         regs->ds                = __USER_DS;
 197         regs->es                = __USER_DS;
 198         regs->ss                = __USER_DS;
 199         regs->cs                = __USER_CS;
 200         regs->ip                = new_ip;
 201         regs->sp                = new_sp;
 202         regs->flags             = X86_EFLAGS_IF;
 203         force_iret();
 204 }
 205 EXPORT_SYMBOL_GPL(start_thread);
 206
 207
 208 /*
 209  *      switch_to(x,y) should switch tasks from x to y.
 210  *
 211  * We fsave/fwait so that an exception goes off at the right time
 212  * (as a call from the fsave or fwait in effect) rather than to
 213  * the wrong process. Lazy FP saving no longer makes any sense
 214  * with modern CPU's, and this simplifies a lot of things (SMP
 215  * and UP become the same).
 216  *
 217  * NOTE! We used to use the x86 hardware context switching. The
 218  * reason for not using it any more becomes apparent when you
 219  * try to recover gracefully from saved state that is no longer
 220  * valid (stale segment register values in particular). With the
 221  * hardware task-switch, there is no way to fix up bad state in
 222  * a reasonable manner.
 223  *
 224  * The fact that Intel documents the hardware task-switching to
 225  * be slow is a fairly red herring - this code is not noticeably
 226  * faster. However, there _is_ some room for improvement here,
 227  * so the performance issues may eventually be a valid point.
 228  * More important, however, is the fact that this allows us much
 229  * more flexibility.
 230  *
 231  * The return value (in %ax) will be the "prev" task after
 232  * the task-switch, and shows up in ret_from_fork in entry.S,
 233  * for example.
 234  */
 235 __visible __notrace_funcgraph struct task_struct *
 236 __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
 237 {
 238         struct thread_struct *prev = &prev_p->thread,
 239                              *next = &next_p->thread;
 240         struct fpu *prev_fpu = &prev->fpu;
 241         struct fpu *next_fpu = &next->fpu;
 242         int cpu = smp_processor_id();
 243         struct tss_struct *tss = &per_cpu(cpu_tss, cpu);
 244         fpu_switch_t fpu_switch;
 245
 246         /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
 247
 248         fpu_switch = switch_fpu_prepare(prev_fpu, next_fpu, cpu);
 249
 250         /*
 251          * Save away %gs. No need to save %fs, as it was saved on the
 252          * stack on entry.  No need to save %es and %ds, as those are
 253          * always kernel segments while inside the kernel.  Doing this
 254          * before setting the new TLS descriptors avoids the situation
 255          * where we temporarily have non-reloadable segments in %fs
 256          * and %gs.  This could be an issue if the NMI handler ever
 257          * used %fs or %gs (it does not today), or if the kernel is
 258          * running inside of a hypervisor layer.
 259          */
 260         lazy_save_gs(prev->gs);
 261
 262         /*
 263          * Load the per-thread Thread-Local Storage descriptor.
 264          */
 265         load_TLS(next, cpu);
 266
 267         /*
 268          * Restore IOPL if needed.  In normal use, the flags restore
 269          * in the switch assembly will handle this.  But if the kernel
 270          * is running virtualized at a non-zero CPL, the popf will
 271          * not restore flags, so it must be done in a separate step.
 272          */
 273         if (get_kernel_rpl() && unlikely(prev->iopl != next->iopl))
 274                 set_iopl_mask(next->iopl);
 275
 276         switch_to_extra(prev_p, next_p);
 277
 278         /*
 279          * Leave lazy mode, flushing any hypercalls made here.
 280          * This must be done before restoring TLS segments so
 281          * the GDT and LDT are properly updated, and must be
 282          * done before fpu__restore(), so the TS bit is up
 283          * to date.
 284          */
 285         arch_end_context_switch(next_p);
 286
 287         /*
 288          * Reload esp0 and cpu_current_top_of_stack.  This changes
 289          * current_thread_info().
 290          */
 291         load_sp0(tss, next);
 292         this_cpu_write(cpu_current_top_of_stack,
 293                        (unsigned long)task_stack_page(next_p) +
 294                        THREAD_SIZE);
 295
 296         /*
 297          * Restore %gs if needed (which is common)
 298          */
 299         if (prev->gs | next->gs)
 300                 lazy_load_gs(next->gs);
 301
 302         switch_fpu_finish(next_fpu, fpu_switch);
 303
 304         this_cpu_write(current_task, next_p);
 305
 306         return prev_p;
 307 }