2 * Copyright (C) 1995 Linus Torvalds
4 * Pentium III FXSR, SSE support
5 * Gareth Hughes <gareth@valinux.com>, May 2000
9 * This file handles the architecture-dependent parts of process handling..
14 #include <linux/cpu.h>
15 #include <linux/errno.h>
16 #include <linux/sched.h>
18 #include <linux/kernel.h>
20 #include <linux/elfcore.h>
21 #include <linux/smp.h>
22 #include <linux/stddef.h>
23 #include <linux/slab.h>
24 #include <linux/vmalloc.h>
25 #include <linux/user.h>
26 #include <linux/a.out.h>
27 #include <linux/interrupt.h>
28 #include <linux/utsname.h>
29 #include <linux/delay.h>
30 #include <linux/reboot.h>
31 #include <linux/init.h>
32 #include <linux/mc146818rtc.h>
33 #include <linux/module.h>
34 #include <linux/kallsyms.h>
35 #include <linux/ptrace.h>
36 #include <linux/random.h>
37 #include <linux/personality.h>
38 #include <linux/tick.h>
39 #include <linux/percpu.h>
41 #include <asm/uaccess.h>
42 #include <asm/pgtable.h>
43 #include <asm/system.h>
46 #include <asm/processor.h>
50 #ifdef CONFIG_MATH_EMULATION
51 #include <asm/math_emu.h>
54 #include <linux/err.h>
56 #include <asm/tlbflush.h>
58 #include <asm/kdebug.h>
60 asmlinkage void ret_from_fork(void) __asm__("ret_from_fork");
62 static int hlt_counter;
64 unsigned long boot_option_idle_override = 0;
65 EXPORT_SYMBOL(boot_option_idle_override);
67 DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task;
68 EXPORT_PER_CPU_SYMBOL(current_task);
70 DEFINE_PER_CPU(int, cpu_number);
71 EXPORT_PER_CPU_SYMBOL(cpu_number);
74 * Return saved PC of a blocked thread.
76 unsigned long thread_saved_pc(struct task_struct *tsk)
78 return ((unsigned long *)tsk->thread.sp)[3];
82 * Powermanagement idle function, if any..
84 void (*pm_idle)(void);
85 EXPORT_SYMBOL(pm_idle);
86 static DEFINE_PER_CPU(unsigned int, cpu_idle_state);
88 void disable_hlt(void)
93 EXPORT_SYMBOL(disable_hlt);
100 EXPORT_SYMBOL(enable_hlt);
103 * We use this if we don't have any better
106 void default_idle(void)
108 if (!hlt_counter && boot_cpu_data.hlt_works_ok) {
109 current_thread_info()->status &= ~TS_POLLING;
111 * TS_POLLING-cleared state must be visible before we
117 if (!need_resched()) {
122 t0n = ktime_to_ns(t0);
123 safe_halt(); /* enables interrupts racelessly */
126 t1n = ktime_to_ns(t1);
127 sched_clock_idle_wakeup_event(t1n - t0n);
130 current_thread_info()->status |= TS_POLLING;
132 /* loop is done by the caller */
136 #ifdef CONFIG_APM_MODULE
137 EXPORT_SYMBOL(default_idle);
141 * On SMP it's slightly faster (but much more power-consuming!)
142 * to poll the ->work.need_resched flag instead of waiting for the
143 * cross-CPU IPI to arrive. Use this option with caution.
145 static void poll_idle(void)
150 #ifdef CONFIG_HOTPLUG_CPU
152 /* We don't actually take CPU down, just spin without interrupts. */
153 static inline void play_dead(void)
155 /* This must be done before dead CPU ack */
160 __get_cpu_var(cpu_state) = CPU_DEAD;
163 * With physical CPU hotplug, we should halt the cpu
170 static inline void play_dead(void)
174 #endif /* CONFIG_HOTPLUG_CPU */
177 * The idle thread. There's no useful work to be
178 * done, so just try to conserve power and have a
179 * low exit latency (ie sit in a loop waiting for
180 * somebody to say that they'd like to reschedule)
184 int cpu = smp_processor_id();
186 current_thread_info()->status |= TS_POLLING;
188 /* endless idle loop with no priority at all */
190 tick_nohz_stop_sched_tick();
191 while (!need_resched()) {
194 if (__get_cpu_var(cpu_idle_state))
195 __get_cpu_var(cpu_idle_state) = 0;
201 if (rcu_pending(cpu))
202 rcu_check_callbacks(cpu, 0);
207 if (cpu_is_offline(cpu))
210 __get_cpu_var(irq_stat).idle_timestamp = jiffies;
213 tick_nohz_restart_sched_tick();
214 preempt_enable_no_resched();
220 static void do_nothing(void *unused)
224 void cpu_idle_wait(void)
226 unsigned int cpu, this_cpu = get_cpu();
227 cpumask_t map, tmp = current->cpus_allowed;
229 set_cpus_allowed(current, cpumask_of_cpu(this_cpu));
233 for_each_online_cpu(cpu) {
234 per_cpu(cpu_idle_state, cpu) = 1;
238 __get_cpu_var(cpu_idle_state) = 0;
243 for_each_online_cpu(cpu) {
244 if (cpu_isset(cpu, map) && !per_cpu(cpu_idle_state, cpu))
247 cpus_and(map, map, cpu_online_map);
249 * We waited 1 sec, if a CPU still did not call idle
250 * it may be because it is in idle and not waking up
251 * because it has nothing to do.
252 * Give all the remaining CPUS a kick.
254 smp_call_function_mask(map, do_nothing, 0, 0);
255 } while (!cpus_empty(map));
257 set_cpus_allowed(current, tmp);
259 EXPORT_SYMBOL_GPL(cpu_idle_wait);
262 * This uses new MONITOR/MWAIT instructions on P4 processors with PNI,
263 * which can obviate IPI to trigger checking of need_resched.
264 * We execute MONITOR against need_resched and enter optimized wait state
265 * through MWAIT. Whenever someone changes need_resched, we would be woken
266 * up from MWAIT (without an IPI).
268 * New with Core Duo processors, MWAIT can take some hints based on CPU
271 void mwait_idle_with_hints(unsigned long ax, unsigned long cx)
273 if (!need_resched()) {
274 __monitor((void *)¤t_thread_info()->flags, 0, 0);
281 /* Default MONITOR/MWAIT with no hints, used for default C1 state */
282 static void mwait_idle(void)
285 mwait_idle_with_hints(0, 0);
288 void __cpuinit select_idle_routine(const struct cpuinfo_x86 *c)
290 if (cpu_has(c, X86_FEATURE_MWAIT)) {
291 printk("monitor/mwait feature present.\n");
293 * Skip, if setup has overridden idle.
294 * One CPU supports mwait => All CPUs supports mwait
297 printk("using mwait in idle threads.\n");
298 pm_idle = mwait_idle;
303 static int __init idle_setup(char *str)
305 if (!strcmp(str, "poll")) {
306 printk("using polling idle threads.\n");
308 #ifdef CONFIG_X86_SMP
309 if (smp_num_siblings > 1)
310 printk("WARNING: polling idle and HT enabled, performance may degrade.\n");
312 } else if (!strcmp(str, "mwait"))
317 boot_option_idle_override = 1;
320 early_param("idle", idle_setup);
322 void __show_registers(struct pt_regs *regs, int all)
324 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L;
325 unsigned long d0, d1, d2, d3, d6, d7;
327 unsigned short ss, gs;
329 if (user_mode_vm(regs)) {
331 ss = regs->ss & 0xffff;
334 sp = (unsigned long) (®s->sp);
340 printk("Pid: %d, comm: %s %s (%s %.*s)\n",
341 task_pid_nr(current), current->comm,
342 print_tainted(), init_utsname()->release,
343 (int)strcspn(init_utsname()->version, " "),
344 init_utsname()->version);
346 printk("EIP: %04x:[<%08lx>] EFLAGS: %08lx CPU: %d\n",
347 0xffff & regs->cs, regs->ip, regs->flags,
349 print_symbol("EIP is at %s\n", regs->ip);
351 printk("EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
352 regs->ax, regs->bx, regs->cx, regs->dx);
353 printk("ESI: %08lx EDI: %08lx EBP: %08lx ESP: %08lx\n",
354 regs->si, regs->di, regs->bp, sp);
355 printk(" DS: %04x ES: %04x FS: %04x GS: %04x SS: %04x\n",
356 regs->ds & 0xffff, regs->es & 0xffff,
357 regs->fs & 0xffff, gs, ss);
365 cr4 = read_cr4_safe();
366 printk("CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n",
373 printk("DR0: %08lx DR1: %08lx DR2: %08lx DR3: %08lx\n",
378 printk("DR6: %08lx DR7: %08lx\n",
382 void show_regs(struct pt_regs *regs)
384 __show_registers(regs, 1);
385 show_trace(NULL, regs, ®s->sp, regs->bp);
389 * This gets run with %bx containing the
390 * function to call, and %dx containing
393 extern void kernel_thread_helper(void);
396 * Create a kernel thread
398 int kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
402 memset(®s, 0, sizeof(regs));
404 regs.bx = (unsigned long) fn;
405 regs.dx = (unsigned long) arg;
409 regs.fs = __KERNEL_PERCPU;
411 regs.ip = (unsigned long) kernel_thread_helper;
412 regs.cs = __KERNEL_CS | get_kernel_rpl();
413 regs.flags = X86_EFLAGS_IF | X86_EFLAGS_SF | X86_EFLAGS_PF | 0x2;
415 /* Ok, create the new process.. */
416 return do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0, ®s, 0, NULL, NULL);
418 EXPORT_SYMBOL(kernel_thread);
421 * Free current thread data structures etc..
423 void exit_thread(void)
425 /* The process may have allocated an io port bitmap... nuke it. */
426 if (unlikely(test_thread_flag(TIF_IO_BITMAP))) {
427 struct task_struct *tsk = current;
428 struct thread_struct *t = &tsk->thread;
430 struct tss_struct *tss = &per_cpu(init_tss, cpu);
432 kfree(t->io_bitmap_ptr);
433 t->io_bitmap_ptr = NULL;
434 clear_thread_flag(TIF_IO_BITMAP);
436 * Careful, clear this in the TSS too:
438 memset(tss->io_bitmap, 0xff, tss->io_bitmap_max);
439 t->io_bitmap_max = 0;
440 tss->io_bitmap_owner = NULL;
441 tss->io_bitmap_max = 0;
442 tss->x86_tss.io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
447 void flush_thread(void)
449 struct task_struct *tsk = current;
451 tsk->thread.debugreg0 = 0;
452 tsk->thread.debugreg1 = 0;
453 tsk->thread.debugreg2 = 0;
454 tsk->thread.debugreg3 = 0;
455 tsk->thread.debugreg6 = 0;
456 tsk->thread.debugreg7 = 0;
457 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
458 clear_tsk_thread_flag(tsk, TIF_DEBUG);
460 * Forget coprocessor state..
466 void release_thread(struct task_struct *dead_task)
468 BUG_ON(dead_task->mm);
469 release_vm86_irqs(dead_task);
473 * This gets called before we allocate a new thread and copy
474 * the current task into it.
476 void prepare_to_copy(struct task_struct *tsk)
481 int copy_thread(int nr, unsigned long clone_flags, unsigned long sp,
482 unsigned long unused,
483 struct task_struct * p, struct pt_regs * regs)
485 struct pt_regs * childregs;
486 struct task_struct *tsk;
489 childregs = task_pt_regs(p);
494 p->thread.sp = (unsigned long) childregs;
495 p->thread.sp0 = (unsigned long) (childregs+1);
497 p->thread.ip = (unsigned long) ret_from_fork;
499 savesegment(gs, p->thread.gs);
502 if (unlikely(test_tsk_thread_flag(tsk, TIF_IO_BITMAP))) {
503 p->thread.io_bitmap_ptr = kmemdup(tsk->thread.io_bitmap_ptr,
504 IO_BITMAP_BYTES, GFP_KERNEL);
505 if (!p->thread.io_bitmap_ptr) {
506 p->thread.io_bitmap_max = 0;
509 set_tsk_thread_flag(p, TIF_IO_BITMAP);
515 * Set a new TLS for the child thread?
517 if (clone_flags & CLONE_SETTLS)
518 err = do_set_thread_area(p, -1,
519 (struct user_desc __user *)childregs->si, 0);
521 if (err && p->thread.io_bitmap_ptr) {
522 kfree(p->thread.io_bitmap_ptr);
523 p->thread.io_bitmap_max = 0;
529 * fill in the user structure for a core dump..
531 void dump_thread(struct pt_regs * regs, struct user * dump)
535 /* changed the size calculations - should hopefully work better. lbt */
536 dump->magic = CMAGIC;
537 dump->start_code = 0;
538 dump->start_stack = regs->sp & ~(PAGE_SIZE - 1);
539 dump->u_tsize = ((unsigned long) current->mm->end_code) >> PAGE_SHIFT;
540 dump->u_dsize = ((unsigned long) (current->mm->brk + (PAGE_SIZE-1))) >> PAGE_SHIFT;
541 dump->u_dsize -= dump->u_tsize;
543 dump->u_debugreg[0] = current->thread.debugreg0;
544 dump->u_debugreg[1] = current->thread.debugreg1;
545 dump->u_debugreg[2] = current->thread.debugreg2;
546 dump->u_debugreg[3] = current->thread.debugreg3;
547 dump->u_debugreg[4] = 0;
548 dump->u_debugreg[5] = 0;
549 dump->u_debugreg[6] = current->thread.debugreg6;
550 dump->u_debugreg[7] = current->thread.debugreg7;
552 if (dump->start_stack < TASK_SIZE)
553 dump->u_ssize = ((unsigned long) (TASK_SIZE - dump->start_stack)) >> PAGE_SHIFT;
555 dump->regs.bx = regs->bx;
556 dump->regs.cx = regs->cx;
557 dump->regs.dx = regs->dx;
558 dump->regs.si = regs->si;
559 dump->regs.di = regs->di;
560 dump->regs.bp = regs->bp;
561 dump->regs.ax = regs->ax;
562 dump->regs.ds = (u16)regs->ds;
563 dump->regs.es = (u16)regs->es;
564 dump->regs.fs = (u16)regs->fs;
566 dump->regs.orig_ax = regs->orig_ax;
567 dump->regs.ip = regs->ip;
568 dump->regs.cs = (u16)regs->cs;
569 dump->regs.flags = regs->flags;
570 dump->regs.sp = regs->sp;
571 dump->regs.ss = (u16)regs->ss;
573 dump->u_fpvalid = dump_fpu (regs, &dump->i387);
575 EXPORT_SYMBOL(dump_thread);
577 #ifdef CONFIG_SECCOMP
578 static void hard_disable_TSC(void)
580 write_cr4(read_cr4() | X86_CR4_TSD);
582 void disable_TSC(void)
585 if (!test_and_set_thread_flag(TIF_NOTSC))
587 * Must flip the CPU state synchronously with
588 * TIF_NOTSC in the current running context.
593 static void hard_enable_TSC(void)
595 write_cr4(read_cr4() & ~X86_CR4_TSD);
597 #endif /* CONFIG_SECCOMP */
600 __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p,
601 struct tss_struct *tss)
603 struct thread_struct *prev, *next;
604 unsigned long debugctl;
606 prev = &prev_p->thread;
607 next = &next_p->thread;
609 debugctl = prev->debugctlmsr;
610 if (next->ds_area_msr != prev->ds_area_msr) {
611 /* we clear debugctl to make sure DS
612 * is not in use when we change it */
614 wrmsrl(MSR_IA32_DEBUGCTLMSR, 0);
615 wrmsr(MSR_IA32_DS_AREA, next->ds_area_msr, 0);
618 if (next->debugctlmsr != debugctl)
619 wrmsr(MSR_IA32_DEBUGCTLMSR, next->debugctlmsr, 0);
621 if (test_tsk_thread_flag(next_p, TIF_DEBUG)) {
622 set_debugreg(next->debugreg0, 0);
623 set_debugreg(next->debugreg1, 1);
624 set_debugreg(next->debugreg2, 2);
625 set_debugreg(next->debugreg3, 3);
627 set_debugreg(next->debugreg6, 6);
628 set_debugreg(next->debugreg7, 7);
631 #ifdef CONFIG_SECCOMP
632 if (test_tsk_thread_flag(prev_p, TIF_NOTSC) ^
633 test_tsk_thread_flag(next_p, TIF_NOTSC)) {
634 /* prev and next are different */
635 if (test_tsk_thread_flag(next_p, TIF_NOTSC))
642 if (test_tsk_thread_flag(prev_p, TIF_BTS_TRACE_TS))
643 ptrace_bts_take_timestamp(prev_p, BTS_TASK_DEPARTS);
645 if (test_tsk_thread_flag(next_p, TIF_BTS_TRACE_TS))
646 ptrace_bts_take_timestamp(next_p, BTS_TASK_ARRIVES);
649 if (!test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
651 * Disable the bitmap via an invalid offset. We still cache
652 * the previous bitmap owner and the IO bitmap contents:
654 tss->x86_tss.io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
658 if (likely(next == tss->io_bitmap_owner)) {
660 * Previous owner of the bitmap (hence the bitmap content)
661 * matches the next task, we dont have to do anything but
662 * to set a valid offset in the TSS:
664 tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET;
668 * Lazy TSS's I/O bitmap copy. We set an invalid offset here
669 * and we let the task to get a GPF in case an I/O instruction
670 * is performed. The handler of the GPF will verify that the
671 * faulting task has a valid I/O bitmap and, it true, does the
672 * real copy and restart the instruction. This will save us
673 * redundant copies when the currently switched task does not
674 * perform any I/O during its timeslice.
676 tss->x86_tss.io_bitmap_base = INVALID_IO_BITMAP_OFFSET_LAZY;
680 * switch_to(x,yn) should switch tasks from x to y.
682 * We fsave/fwait so that an exception goes off at the right time
683 * (as a call from the fsave or fwait in effect) rather than to
684 * the wrong process. Lazy FP saving no longer makes any sense
685 * with modern CPU's, and this simplifies a lot of things (SMP
686 * and UP become the same).
688 * NOTE! We used to use the x86 hardware context switching. The
689 * reason for not using it any more becomes apparent when you
690 * try to recover gracefully from saved state that is no longer
691 * valid (stale segment register values in particular). With the
692 * hardware task-switch, there is no way to fix up bad state in
693 * a reasonable manner.
695 * The fact that Intel documents the hardware task-switching to
696 * be slow is a fairly red herring - this code is not noticeably
697 * faster. However, there _is_ some room for improvement here,
698 * so the performance issues may eventually be a valid point.
699 * More important, however, is the fact that this allows us much
702 * The return value (in %ax) will be the "prev" task after
703 * the task-switch, and shows up in ret_from_fork in entry.S,
706 struct task_struct * __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
708 struct thread_struct *prev = &prev_p->thread,
709 *next = &next_p->thread;
710 int cpu = smp_processor_id();
711 struct tss_struct *tss = &per_cpu(init_tss, cpu);
713 /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
715 __unlazy_fpu(prev_p);
718 /* we're going to use this soon, after a few expensive things */
719 if (next_p->fpu_counter > 5)
720 prefetch(&next->i387.fxsave);
728 * Save away %gs. No need to save %fs, as it was saved on the
729 * stack on entry. No need to save %es and %ds, as those are
730 * always kernel segments while inside the kernel. Doing this
731 * before setting the new TLS descriptors avoids the situation
732 * where we temporarily have non-reloadable segments in %fs
733 * and %gs. This could be an issue if the NMI handler ever
734 * used %fs or %gs (it does not today), or if the kernel is
735 * running inside of a hypervisor layer.
737 savesegment(gs, prev->gs);
740 * Load the per-thread Thread-Local Storage descriptor.
745 * Restore IOPL if needed. In normal use, the flags restore
746 * in the switch assembly will handle this. But if the kernel
747 * is running virtualized at a non-zero CPL, the popf will
748 * not restore flags, so it must be done in a separate step.
750 if (get_kernel_rpl() && unlikely(prev->iopl != next->iopl))
751 set_iopl_mask(next->iopl);
754 * Now maybe handle debug registers and/or IO bitmaps
756 if (unlikely(task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV ||
757 task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT))
758 __switch_to_xtra(prev_p, next_p, tss);
761 * Leave lazy mode, flushing any hypercalls made here.
762 * This must be done before restoring TLS segments so
763 * the GDT and LDT are properly updated, and must be
764 * done before math_state_restore, so the TS bit is up
767 arch_leave_lazy_cpu_mode();
769 /* If the task has used fpu the last 5 timeslices, just do a full
770 * restore of the math state immediately to avoid the trap; the
771 * chances of needing FPU soon are obviously high now
773 if (next_p->fpu_counter > 5)
774 math_state_restore();
777 * Restore %gs if needed (which is common)
779 if (prev->gs | next->gs)
780 loadsegment(gs, next->gs);
782 x86_write_percpu(current_task, next_p);
787 asmlinkage int sys_fork(struct pt_regs regs)
789 return do_fork(SIGCHLD, regs.sp, ®s, 0, NULL, NULL);
792 asmlinkage int sys_clone(struct pt_regs regs)
794 unsigned long clone_flags;
796 int __user *parent_tidptr, *child_tidptr;
798 clone_flags = regs.bx;
800 parent_tidptr = (int __user *)regs.dx;
801 child_tidptr = (int __user *)regs.di;
804 return do_fork(clone_flags, newsp, ®s, 0, parent_tidptr, child_tidptr);
808 * This is trivial, and on the face of it looks like it
809 * could equally well be done in user mode.
811 * Not so, for quite unobvious reasons - register pressure.
812 * In user mode vfork() cannot have a stack frame, and if
813 * done by calling the "clone()" system call directly, you
814 * do not have enough call-clobbered registers to hold all
815 * the information you need.
817 asmlinkage int sys_vfork(struct pt_regs regs)
819 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs.sp, ®s, 0, NULL, NULL);
823 * sys_execve() executes a new program.
825 asmlinkage int sys_execve(struct pt_regs regs)
830 filename = getname((char __user *) regs.bx);
831 error = PTR_ERR(filename);
832 if (IS_ERR(filename))
834 error = do_execve(filename,
835 (char __user * __user *) regs.cx,
836 (char __user * __user *) regs.dx,
839 /* Make sure we don't return using sysenter.. */
840 set_thread_flag(TIF_IRET);
847 #define top_esp (THREAD_SIZE - sizeof(unsigned long))
848 #define top_ebp (THREAD_SIZE - 2*sizeof(unsigned long))
850 unsigned long get_wchan(struct task_struct *p)
852 unsigned long bp, sp, ip;
853 unsigned long stack_page;
855 if (!p || p == current || p->state == TASK_RUNNING)
857 stack_page = (unsigned long)task_stack_page(p);
859 if (!stack_page || sp < stack_page || sp > top_esp+stack_page)
861 /* include/asm-i386/system.h:switch_to() pushes bp last. */
862 bp = *(unsigned long *) sp;
864 if (bp < stack_page || bp > top_ebp+stack_page)
866 ip = *(unsigned long *) (bp+4);
867 if (!in_sched_functions(ip))
869 bp = *(unsigned long *) bp;
870 } while (count++ < 16);
874 unsigned long arch_align_stack(unsigned long sp)
876 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
877 sp -= get_random_int() % 8192;
881 unsigned long arch_randomize_brk(struct mm_struct *mm)
883 unsigned long range_end = mm->brk + 0x02000000;
884 return randomize_range(mm->brk, range_end, 0) ? : mm->brk;