LXR linux/arch/x86/kernel/process

   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#include <asm/vm86.h>
  57
  58asmlinkage void ret_from_fork(void) __asm__("ret_from_fork");
  59asmlinkage void ret_from_kernel_thread(void) __asm__("ret_from_kernel_thread");
  60
  61/*
  62 * Return saved PC of a blocked thread.
  63 */
  64unsigned long thread_saved_pc(struct task_struct *tsk)
  65{
  66        return ((unsigned long *)tsk->thread.sp)[3];
  67}
  68
  69void __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(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        printk(KERN_DEFAULT "EIP: %04x:[<%08lx>] EFLAGS: %08lx CPU: %d\n",
  87                        (u16)regs->cs, regs->ip, regs->flags,
  88                        smp_processor_id());
  89        print_symbol("EIP is at %s\n", regs->ip);
  90
  91        printk(KERN_DEFAULT "EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
  92                regs->ax, regs->bx, regs->cx, regs->dx);
  93        printk(KERN_DEFAULT "ESI: %08lx EDI: %08lx EBP: %08lx ESP: %08lx\n",
  94                regs->si, regs->di, regs->bp, sp);
  95        printk(KERN_DEFAULT " DS: %04x ES: %04x FS: %04x GS: %04x SS: %04x\n",
  96               (u16)regs->ds, (u16)regs->es, (u16)regs->fs, gs, ss);
  97
  98        if (!all)
  99                return;
 100
 101        cr0 = read_cr0();
 102        cr2 = read_cr2();
 103        cr3 = read_cr3();
 104        cr4 = __read_cr4_safe();
 105        printk(KERN_DEFAULT "CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n",
 106                        cr0, cr2, cr3, cr4);
 107
 108        get_debugreg(d0, 0);
 109        get_debugreg(d1, 1);
 110        get_debugreg(d2, 2);
 111        get_debugreg(d3, 3);
 112        get_debugreg(d6, 6);
 113        get_debugreg(d7, 7);
 114
 115        /* Only print out debug registers if they are in their non-default state. */
 116        if ((d0 == 0) && (d1 == 0) && (d2 == 0) && (d3 == 0) &&
 117            (d6 == DR6_RESERVED) && (d7 == 0x400))
 118                return;
 119
 120        printk(KERN_DEFAULT "DR0: %08lx DR1: %08lx DR2: %08lx DR3: %08lx\n",
 121                        d0, d1, d2, d3);
 122        printk(KERN_DEFAULT "DR6: %08lx DR7: %08lx\n",
 123                        d6, d7);
 124}
 125
 126void release_thread(struct task_struct *dead_task)
 127{
 128        BUG_ON(dead_task->mm);
 129        release_vm86_irqs(dead_task);
 130}
 131
 132int copy_thread_tls(unsigned long clone_flags, unsigned long sp,
 133        unsigned long arg, struct task_struct *p, unsigned long tls)
 134{
 135        struct pt_regs *childregs = task_pt_regs(p);
 136        struct task_struct *tsk;
 137        int err;
 138
 139        p->thread.sp = (unsigned long) childregs;
 140        p->thread.sp0 = (unsigned long) (childregs+1);
 141        memset(p->thread.ptrace_bps, 0, sizeof(p->thread.ptrace_bps));
 142
 143        if (unlikely(p->flags & PF_KTHREAD)) {
 144                /* kernel thread */
 145                memset(childregs, 0, sizeof(struct pt_regs));
 146                p->thread.ip = (unsigned long) ret_from_kernel_thread;
 147                task_user_gs(p) = __KERNEL_STACK_CANARY;
 148                childregs->ds = __USER_DS;
 149                childregs->es = __USER_DS;
 150                childregs->fs = __KERNEL_PERCPU;
 151                childregs->bx = sp;     /* function */
 152                childregs->bp = arg;
 153                childregs->orig_ax = -1;
 154                childregs->cs = __KERNEL_CS | get_kernel_rpl();
 155                childregs->flags = X86_EFLAGS_IF | X86_EFLAGS_FIXED;
 156                p->thread.io_bitmap_ptr = NULL;
 157                return 0;
 158        }
 159        *childregs = *current_pt_regs();
 160        childregs->ax = 0;
 161        if (sp)
 162                childregs->sp = sp;
 163
 164        p->thread.ip = (unsigned long) ret_from_fork;
 165        task_user_gs(p) = get_user_gs(current_pt_regs());
 166
 167        p->thread.io_bitmap_ptr = NULL;
 168        tsk = current;
 169        err = -ENOMEM;
 170
 171        if (unlikely(test_tsk_thread_flag(tsk, TIF_IO_BITMAP))) {
 172                p->thread.io_bitmap_ptr = kmemdup(tsk->thread.io_bitmap_ptr,
 173                                                IO_BITMAP_BYTES, GFP_KERNEL);
 174                if (!p->thread.io_bitmap_ptr) {
 175                        p->thread.io_bitmap_max = 0;
 176                        return -ENOMEM;
 177                }
 178                set_tsk_thread_flag(p, TIF_IO_BITMAP);
 179        }
 180
 181        err = 0;
 182
 183        /*
 184         * Set a new TLS for the child thread?
 185         */
 186        if (clone_flags & CLONE_SETTLS)
 187                err = do_set_thread_area(p, -1,
 188                        (struct user_desc __user *)tls, 0);
 189
 190        if (err && p->thread.io_bitmap_ptr) {
 191                kfree(p->thread.io_bitmap_ptr);
 192                p->thread.io_bitmap_max = 0;
 193        }
 194        return err;
 195}
 196
 197void
 198start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp)
 199{
 200        set_user_gs(regs, 0);
 201        regs->fs                = 0;
 202        regs->ds                = __USER_DS;
 203        regs->es                = __USER_DS;
 204        regs->ss                = __USER_DS;
 205        regs->cs                = __USER_CS;
 206        regs->ip                = new_ip;
 207        regs->sp                = new_sp;
 208        regs->flags             = X86_EFLAGS_IF;
 209        force_iret();
 210}
 211EXPORT_SYMBOL_GPL(start_thread);
 212
 213
 214/*
 215 *      switch_to(x,y) should switch tasks from x to y.
 216 *
 217 * We fsave/fwait so that an exception goes off at the right time
 218 * (as a call from the fsave or fwait in effect) rather than to
 219 * the wrong process. Lazy FP saving no longer makes any sense
 220 * with modern CPU's, and this simplifies a lot of things (SMP
 221 * and UP become the same).
 222 *
 223 * NOTE! We used to use the x86 hardware context switching. The
 224 * reason for not using it any more becomes apparent when you
 225 * try to recover gracefully from saved state that is no longer
 226 * valid (stale segment register values in particular). With the
 227 * hardware task-switch, there is no way to fix up bad state in
 228 * a reasonable manner.
 229 *
 230 * The fact that Intel documents the hardware task-switching to
 231 * be slow is a fairly red herring - this code is not noticeably
 232 * faster. However, there _is_ some room for improvement here,
 233 * so the performance issues may eventually be a valid point.
 234 * More important, however, is the fact that this allows us much
 235 * more flexibility.
 236 *
 237 * The return value (in %ax) will be the "prev" task after
 238 * the task-switch, and shows up in ret_from_fork in entry.S,
 239 * for example.
 240 */
 241__visible __notrace_funcgraph struct task_struct *
 242__switch_to(struct task_struct *prev_p, struct task_struct *next_p)
 243{
 244        struct thread_struct *prev = &prev_p->thread,
 245                             *next = &next_p->thread;
 246        struct fpu *prev_fpu = &prev->fpu;
 247        struct fpu *next_fpu = &next->fpu;
 248        int cpu = smp_processor_id();
 249        struct tss_struct *tss = &per_cpu(cpu_tss, cpu);
 250        fpu_switch_t fpu_switch;
 251
 252        /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
 253
 254        fpu_switch = switch_fpu_prepare(prev_fpu, next_fpu, cpu);
 255
 256        /*
 257         * Save away %gs. No need to save %fs, as it was saved on the
 258         * stack on entry.  No need to save %es and %ds, as those are
 259         * always kernel segments while inside the kernel.  Doing this
 260         * before setting the new TLS descriptors avoids the situation
 261         * where we temporarily have non-reloadable segments in %fs
 262         * and %gs.  This could be an issue if the NMI handler ever
 263         * used %fs or %gs (it does not today), or if the kernel is
 264         * running inside of a hypervisor layer.
 265         */
 266        lazy_save_gs(prev->gs);
 267
 268        /*
 269         * Load the per-thread Thread-Local Storage descriptor.
 270         */
 271        load_TLS(next, cpu);
 272
 273        /*
 274         * Restore IOPL if needed.  In normal use, the flags restore
 275         * in the switch assembly will handle this.  But if the kernel
 276         * is running virtualized at a non-zero CPL, the popf will
 277         * not restore flags, so it must be done in a separate step.
 278         */
 279        if (get_kernel_rpl() && unlikely(prev->iopl != next->iopl))
 280                set_iopl_mask(next->iopl);
 281
 282        /*
 283         * If it were not for PREEMPT_ACTIVE we could guarantee that the
 284         * preempt_count of all tasks was equal here and this would not be
 285         * needed.
 286         */
 287        task_thread_info(prev_p)->saved_preempt_count = this_cpu_read(__preempt_count);
 288        this_cpu_write(__preempt_count, task_thread_info(next_p)->saved_preempt_count);
 289
 290        /*
 291         * Now maybe handle debug registers and/or IO bitmaps
 292         */
 293        if (unlikely(task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV ||
 294                     task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT))
 295                __switch_to_xtra(prev_p, next_p, tss);
 296
 297        /*
 298         * Leave lazy mode, flushing any hypercalls made here.
 299         * This must be done before restoring TLS segments so
 300         * the GDT and LDT are properly updated, and must be
 301         * done before fpu__restore(), so the TS bit is up
 302         * to date.
 303         */
 304        arch_end_context_switch(next_p);
 305
 306        /*
 307         * Reload esp0 and cpu_current_top_of_stack.  This changes
 308         * current_thread_info().
 309         */
 310        load_sp0(tss, next);
 311        this_cpu_write(cpu_current_top_of_stack,
 312                       (unsigned long)task_stack_page(next_p) +
 313                       THREAD_SIZE);
 314
 315        /*
 316         * Restore %gs if needed (which is common)
 317         */
 318        if (prev->gs | next->gs)
 319                lazy_load_gs(next->gs);
 320
 321        switch_fpu_finish(next_fpu, fpu_switch);
 322
 323        this_cpu_write(current_task, next_p);
 324
 325        return prev_p;
 326}
 327