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/interrupt.h>
27 #include <linux/utsname.h>
28 #include <linux/delay.h>
29 #include <linux/reboot.h>
30 #include <linux/init.h>
31 #include <linux/mc146818rtc.h>
32 #include <linux/module.h>
33 #include <linux/kallsyms.h>
34 #include <linux/ptrace.h>
35 #include <linux/random.h>
36 #include <linux/personality.h>
37 #include <linux/tick.h>
38 #include <linux/percpu.h>
40 #include <asm/uaccess.h>
41 #include <asm/pgtable.h>
42 #include <asm/system.h>
45 #include <asm/processor.h>
49 #ifdef CONFIG_MATH_EMULATION
50 #include <asm/math_emu.h>
53 #include <linux/err.h>
55 #include <asm/tlbflush.h>
57 #include <asm/kdebug.h>
59 asmlinkage void ret_from_fork(void) __asm__("ret_from_fork");
61 static int hlt_counter;
63 unsigned long boot_option_idle_override = 0;
64 EXPORT_SYMBOL(boot_option_idle_override);
66 DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task;
67 EXPORT_PER_CPU_SYMBOL(current_task);
69 DEFINE_PER_CPU(int, cpu_number);
70 EXPORT_PER_CPU_SYMBOL(cpu_number);
73 * Return saved PC of a blocked thread.
75 unsigned long thread_saved_pc(struct task_struct *tsk)
77 return ((unsigned long *)tsk->thread.sp)[3];
81 * Powermanagement idle function, if any..
83 void (*pm_idle)(void);
84 EXPORT_SYMBOL(pm_idle);
85 static DEFINE_PER_CPU(unsigned int, cpu_idle_state);
87 void disable_hlt(void)
92 EXPORT_SYMBOL(disable_hlt);
99 EXPORT_SYMBOL(enable_hlt);
102 * We use this if we don't have any better
105 void default_idle(void)
107 if (!hlt_counter && boot_cpu_data.hlt_works_ok) {
108 current_thread_info()->status &= ~TS_POLLING;
110 * TS_POLLING-cleared state must be visible before we
116 if (!need_resched()) {
121 t0n = ktime_to_ns(t0);
122 safe_halt(); /* enables interrupts racelessly */
125 t1n = ktime_to_ns(t1);
126 sched_clock_idle_wakeup_event(t1n - t0n);
129 current_thread_info()->status |= TS_POLLING;
131 /* loop is done by the caller */
135 #ifdef CONFIG_APM_MODULE
136 EXPORT_SYMBOL(default_idle);
140 * On SMP it's slightly faster (but much more power-consuming!)
141 * to poll the ->work.need_resched flag instead of waiting for the
142 * cross-CPU IPI to arrive. Use this option with caution.
144 static void poll_idle(void)
149 #ifdef CONFIG_HOTPLUG_CPU
151 /* We don't actually take CPU down, just spin without interrupts. */
152 static inline void play_dead(void)
154 /* This must be done before dead CPU ack */
159 __get_cpu_var(cpu_state) = CPU_DEAD;
162 * With physical CPU hotplug, we should halt the cpu
169 static inline void play_dead(void)
173 #endif /* CONFIG_HOTPLUG_CPU */
176 * The idle thread. There's no useful work to be
177 * done, so just try to conserve power and have a
178 * low exit latency (ie sit in a loop waiting for
179 * somebody to say that they'd like to reschedule)
183 int cpu = smp_processor_id();
185 current_thread_info()->status |= TS_POLLING;
187 /* endless idle loop with no priority at all */
189 tick_nohz_stop_sched_tick();
190 while (!need_resched()) {
193 if (__get_cpu_var(cpu_idle_state))
194 __get_cpu_var(cpu_idle_state) = 0;
200 if (rcu_pending(cpu))
201 rcu_check_callbacks(cpu, 0);
206 if (cpu_is_offline(cpu))
209 __get_cpu_var(irq_stat).idle_timestamp = jiffies;
212 tick_nohz_restart_sched_tick();
213 preempt_enable_no_resched();
219 static void do_nothing(void *unused)
223 void cpu_idle_wait(void)
225 unsigned int cpu, this_cpu = get_cpu();
226 cpumask_t map, tmp = current->cpus_allowed;
228 set_cpus_allowed(current, cpumask_of_cpu(this_cpu));
232 for_each_online_cpu(cpu) {
233 per_cpu(cpu_idle_state, cpu) = 1;
237 __get_cpu_var(cpu_idle_state) = 0;
242 for_each_online_cpu(cpu) {
243 if (cpu_isset(cpu, map) && !per_cpu(cpu_idle_state, cpu))
246 cpus_and(map, map, cpu_online_map);
248 * We waited 1 sec, if a CPU still did not call idle
249 * it may be because it is in idle and not waking up
250 * because it has nothing to do.
251 * Give all the remaining CPUS a kick.
253 smp_call_function_mask(map, do_nothing, NULL, 0);
254 } while (!cpus_empty(map));
256 set_cpus_allowed(current, tmp);
258 EXPORT_SYMBOL_GPL(cpu_idle_wait);
261 * This uses new MONITOR/MWAIT instructions on P4 processors with PNI,
262 * which can obviate IPI to trigger checking of need_resched.
263 * We execute MONITOR against need_resched and enter optimized wait state
264 * through MWAIT. Whenever someone changes need_resched, we would be woken
265 * up from MWAIT (without an IPI).
267 * New with Core Duo processors, MWAIT can take some hints based on CPU
270 void mwait_idle_with_hints(unsigned long ax, unsigned long cx)
272 if (!need_resched()) {
273 __monitor((void *)¤t_thread_info()->flags, 0, 0);
280 /* Default MONITOR/MWAIT with no hints, used for default C1 state */
281 static void mwait_idle(void)
284 mwait_idle_with_hints(0, 0);
287 static int __cpuinit mwait_usable(const struct cpuinfo_x86 *c)
291 /* Any C1 states supported? */
292 return c->cpuid_level >= 5 && ((cpuid_edx(5) >> 4) & 0xf) > 0;
295 void __cpuinit select_idle_routine(const struct cpuinfo_x86 *c)
301 #ifdef CONFIG_X86_SMP
302 if (pm_idle == poll_idle && smp_num_siblings > 1) {
303 printk(KERN_WARNING "WARNING: polling idle and HT enabled,"
304 " performance may degrade.\n");
307 if (cpu_has(c, X86_FEATURE_MWAIT) && mwait_usable(c)) {
309 * Skip, if setup has overridden idle.
310 * One CPU supports mwait => All CPUs supports mwait
313 printk(KERN_INFO "using mwait in idle threads.\n");
314 pm_idle = mwait_idle;
320 static int __init idle_setup(char *str)
322 if (!strcmp(str, "poll")) {
323 printk("using polling idle threads.\n");
325 } else if (!strcmp(str, "mwait"))
330 boot_option_idle_override = 1;
333 early_param("idle", idle_setup);
335 void __show_registers(struct pt_regs *regs, int all)
337 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L;
338 unsigned long d0, d1, d2, d3, d6, d7;
340 unsigned short ss, gs;
342 if (user_mode_vm(regs)) {
344 ss = regs->ss & 0xffff;
347 sp = (unsigned long) (®s->sp);
353 printk("Pid: %d, comm: %s %s (%s %.*s)\n",
354 task_pid_nr(current), current->comm,
355 print_tainted(), init_utsname()->release,
356 (int)strcspn(init_utsname()->version, " "),
357 init_utsname()->version);
359 printk("EIP: %04x:[<%08lx>] EFLAGS: %08lx CPU: %d\n",
360 0xffff & regs->cs, regs->ip, regs->flags,
362 print_symbol("EIP is at %s\n", regs->ip);
364 printk("EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
365 regs->ax, regs->bx, regs->cx, regs->dx);
366 printk("ESI: %08lx EDI: %08lx EBP: %08lx ESP: %08lx\n",
367 regs->si, regs->di, regs->bp, sp);
368 printk(" DS: %04x ES: %04x FS: %04x GS: %04x SS: %04x\n",
369 regs->ds & 0xffff, regs->es & 0xffff,
370 regs->fs & 0xffff, gs, ss);
378 cr4 = read_cr4_safe();
379 printk("CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n",
386 printk("DR0: %08lx DR1: %08lx DR2: %08lx DR3: %08lx\n",
391 printk("DR6: %08lx DR7: %08lx\n",
395 void show_regs(struct pt_regs *regs)
397 __show_registers(regs, 1);
398 show_trace(NULL, regs, ®s->sp, regs->bp);
402 * This gets run with %bx containing the
403 * function to call, and %dx containing
406 extern void kernel_thread_helper(void);
409 * Create a kernel thread
411 int kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
415 memset(®s, 0, sizeof(regs));
417 regs.bx = (unsigned long) fn;
418 regs.dx = (unsigned long) arg;
422 regs.fs = __KERNEL_PERCPU;
424 regs.ip = (unsigned long) kernel_thread_helper;
425 regs.cs = __KERNEL_CS | get_kernel_rpl();
426 regs.flags = X86_EFLAGS_IF | X86_EFLAGS_SF | X86_EFLAGS_PF | 0x2;
428 /* Ok, create the new process.. */
429 return do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0, ®s, 0, NULL, NULL);
431 EXPORT_SYMBOL(kernel_thread);
434 * Free current thread data structures etc..
436 void exit_thread(void)
438 /* The process may have allocated an io port bitmap... nuke it. */
439 if (unlikely(test_thread_flag(TIF_IO_BITMAP))) {
440 struct task_struct *tsk = current;
441 struct thread_struct *t = &tsk->thread;
443 struct tss_struct *tss = &per_cpu(init_tss, cpu);
445 kfree(t->io_bitmap_ptr);
446 t->io_bitmap_ptr = NULL;
447 clear_thread_flag(TIF_IO_BITMAP);
449 * Careful, clear this in the TSS too:
451 memset(tss->io_bitmap, 0xff, tss->io_bitmap_max);
452 t->io_bitmap_max = 0;
453 tss->io_bitmap_owner = NULL;
454 tss->io_bitmap_max = 0;
455 tss->x86_tss.io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
460 void flush_thread(void)
462 struct task_struct *tsk = current;
464 tsk->thread.debugreg0 = 0;
465 tsk->thread.debugreg1 = 0;
466 tsk->thread.debugreg2 = 0;
467 tsk->thread.debugreg3 = 0;
468 tsk->thread.debugreg6 = 0;
469 tsk->thread.debugreg7 = 0;
470 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
471 clear_tsk_thread_flag(tsk, TIF_DEBUG);
473 * Forget coprocessor state..
479 void release_thread(struct task_struct *dead_task)
481 BUG_ON(dead_task->mm);
482 release_vm86_irqs(dead_task);
486 * This gets called before we allocate a new thread and copy
487 * the current task into it.
489 void prepare_to_copy(struct task_struct *tsk)
494 int copy_thread(int nr, unsigned long clone_flags, unsigned long sp,
495 unsigned long unused,
496 struct task_struct * p, struct pt_regs * regs)
498 struct pt_regs * childregs;
499 struct task_struct *tsk;
502 childregs = task_pt_regs(p);
507 p->thread.sp = (unsigned long) childregs;
508 p->thread.sp0 = (unsigned long) (childregs+1);
510 p->thread.ip = (unsigned long) ret_from_fork;
512 savesegment(gs, p->thread.gs);
515 if (unlikely(test_tsk_thread_flag(tsk, TIF_IO_BITMAP))) {
516 p->thread.io_bitmap_ptr = kmemdup(tsk->thread.io_bitmap_ptr,
517 IO_BITMAP_BYTES, GFP_KERNEL);
518 if (!p->thread.io_bitmap_ptr) {
519 p->thread.io_bitmap_max = 0;
522 set_tsk_thread_flag(p, TIF_IO_BITMAP);
528 * Set a new TLS for the child thread?
530 if (clone_flags & CLONE_SETTLS)
531 err = do_set_thread_area(p, -1,
532 (struct user_desc __user *)childregs->si, 0);
534 if (err && p->thread.io_bitmap_ptr) {
535 kfree(p->thread.io_bitmap_ptr);
536 p->thread.io_bitmap_max = 0;
541 #ifdef CONFIG_SECCOMP
542 static void hard_disable_TSC(void)
544 write_cr4(read_cr4() | X86_CR4_TSD);
546 void disable_TSC(void)
549 if (!test_and_set_thread_flag(TIF_NOTSC))
551 * Must flip the CPU state synchronously with
552 * TIF_NOTSC in the current running context.
557 static void hard_enable_TSC(void)
559 write_cr4(read_cr4() & ~X86_CR4_TSD);
561 #endif /* CONFIG_SECCOMP */
564 __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p,
565 struct tss_struct *tss)
567 struct thread_struct *prev, *next;
568 unsigned long debugctl;
570 prev = &prev_p->thread;
571 next = &next_p->thread;
573 debugctl = prev->debugctlmsr;
574 if (next->ds_area_msr != prev->ds_area_msr) {
575 /* we clear debugctl to make sure DS
576 * is not in use when we change it */
578 wrmsrl(MSR_IA32_DEBUGCTLMSR, 0);
579 wrmsr(MSR_IA32_DS_AREA, next->ds_area_msr, 0);
582 if (next->debugctlmsr != debugctl)
583 wrmsr(MSR_IA32_DEBUGCTLMSR, next->debugctlmsr, 0);
585 if (test_tsk_thread_flag(next_p, TIF_DEBUG)) {
586 set_debugreg(next->debugreg0, 0);
587 set_debugreg(next->debugreg1, 1);
588 set_debugreg(next->debugreg2, 2);
589 set_debugreg(next->debugreg3, 3);
591 set_debugreg(next->debugreg6, 6);
592 set_debugreg(next->debugreg7, 7);
595 #ifdef CONFIG_SECCOMP
596 if (test_tsk_thread_flag(prev_p, TIF_NOTSC) ^
597 test_tsk_thread_flag(next_p, TIF_NOTSC)) {
598 /* prev and next are different */
599 if (test_tsk_thread_flag(next_p, TIF_NOTSC))
606 if (test_tsk_thread_flag(prev_p, TIF_BTS_TRACE_TS))
607 ptrace_bts_take_timestamp(prev_p, BTS_TASK_DEPARTS);
609 if (test_tsk_thread_flag(next_p, TIF_BTS_TRACE_TS))
610 ptrace_bts_take_timestamp(next_p, BTS_TASK_ARRIVES);
613 if (!test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
615 * Disable the bitmap via an invalid offset. We still cache
616 * the previous bitmap owner and the IO bitmap contents:
618 tss->x86_tss.io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
622 if (likely(next == tss->io_bitmap_owner)) {
624 * Previous owner of the bitmap (hence the bitmap content)
625 * matches the next task, we dont have to do anything but
626 * to set a valid offset in the TSS:
628 tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET;
632 * Lazy TSS's I/O bitmap copy. We set an invalid offset here
633 * and we let the task to get a GPF in case an I/O instruction
634 * is performed. The handler of the GPF will verify that the
635 * faulting task has a valid I/O bitmap and, it true, does the
636 * real copy and restart the instruction. This will save us
637 * redundant copies when the currently switched task does not
638 * perform any I/O during its timeslice.
640 tss->x86_tss.io_bitmap_base = INVALID_IO_BITMAP_OFFSET_LAZY;
644 * switch_to(x,yn) should switch tasks from x to y.
646 * We fsave/fwait so that an exception goes off at the right time
647 * (as a call from the fsave or fwait in effect) rather than to
648 * the wrong process. Lazy FP saving no longer makes any sense
649 * with modern CPU's, and this simplifies a lot of things (SMP
650 * and UP become the same).
652 * NOTE! We used to use the x86 hardware context switching. The
653 * reason for not using it any more becomes apparent when you
654 * try to recover gracefully from saved state that is no longer
655 * valid (stale segment register values in particular). With the
656 * hardware task-switch, there is no way to fix up bad state in
657 * a reasonable manner.
659 * The fact that Intel documents the hardware task-switching to
660 * be slow is a fairly red herring - this code is not noticeably
661 * faster. However, there _is_ some room for improvement here,
662 * so the performance issues may eventually be a valid point.
663 * More important, however, is the fact that this allows us much
666 * The return value (in %ax) will be the "prev" task after
667 * the task-switch, and shows up in ret_from_fork in entry.S,
670 struct task_struct * __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
672 struct thread_struct *prev = &prev_p->thread,
673 *next = &next_p->thread;
674 int cpu = smp_processor_id();
675 struct tss_struct *tss = &per_cpu(init_tss, cpu);
677 /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
679 __unlazy_fpu(prev_p);
682 /* we're going to use this soon, after a few expensive things */
683 if (next_p->fpu_counter > 5)
684 prefetch(&next->i387.fxsave);
692 * Save away %gs. No need to save %fs, as it was saved on the
693 * stack on entry. No need to save %es and %ds, as those are
694 * always kernel segments while inside the kernel. Doing this
695 * before setting the new TLS descriptors avoids the situation
696 * where we temporarily have non-reloadable segments in %fs
697 * and %gs. This could be an issue if the NMI handler ever
698 * used %fs or %gs (it does not today), or if the kernel is
699 * running inside of a hypervisor layer.
701 savesegment(gs, prev->gs);
704 * Load the per-thread Thread-Local Storage descriptor.
709 * Restore IOPL if needed. In normal use, the flags restore
710 * in the switch assembly will handle this. But if the kernel
711 * is running virtualized at a non-zero CPL, the popf will
712 * not restore flags, so it must be done in a separate step.
714 if (get_kernel_rpl() && unlikely(prev->iopl != next->iopl))
715 set_iopl_mask(next->iopl);
718 * Now maybe handle debug registers and/or IO bitmaps
720 if (unlikely(task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV ||
721 task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT))
722 __switch_to_xtra(prev_p, next_p, tss);
725 * Leave lazy mode, flushing any hypercalls made here.
726 * This must be done before restoring TLS segments so
727 * the GDT and LDT are properly updated, and must be
728 * done before math_state_restore, so the TS bit is up
731 arch_leave_lazy_cpu_mode();
733 /* If the task has used fpu the last 5 timeslices, just do a full
734 * restore of the math state immediately to avoid the trap; the
735 * chances of needing FPU soon are obviously high now
737 if (next_p->fpu_counter > 5)
738 math_state_restore();
741 * Restore %gs if needed (which is common)
743 if (prev->gs | next->gs)
744 loadsegment(gs, next->gs);
746 x86_write_percpu(current_task, next_p);
751 asmlinkage int sys_fork(struct pt_regs regs)
753 return do_fork(SIGCHLD, regs.sp, ®s, 0, NULL, NULL);
756 asmlinkage int sys_clone(struct pt_regs regs)
758 unsigned long clone_flags;
760 int __user *parent_tidptr, *child_tidptr;
762 clone_flags = regs.bx;
764 parent_tidptr = (int __user *)regs.dx;
765 child_tidptr = (int __user *)regs.di;
768 return do_fork(clone_flags, newsp, ®s, 0, parent_tidptr, child_tidptr);
772 * This is trivial, and on the face of it looks like it
773 * could equally well be done in user mode.
775 * Not so, for quite unobvious reasons - register pressure.
776 * In user mode vfork() cannot have a stack frame, and if
777 * done by calling the "clone()" system call directly, you
778 * do not have enough call-clobbered registers to hold all
779 * the information you need.
781 asmlinkage int sys_vfork(struct pt_regs regs)
783 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs.sp, ®s, 0, NULL, NULL);
787 * sys_execve() executes a new program.
789 asmlinkage int sys_execve(struct pt_regs regs)
794 filename = getname((char __user *) regs.bx);
795 error = PTR_ERR(filename);
796 if (IS_ERR(filename))
798 error = do_execve(filename,
799 (char __user * __user *) regs.cx,
800 (char __user * __user *) regs.dx,
803 /* Make sure we don't return using sysenter.. */
804 set_thread_flag(TIF_IRET);
811 #define top_esp (THREAD_SIZE - sizeof(unsigned long))
812 #define top_ebp (THREAD_SIZE - 2*sizeof(unsigned long))
814 unsigned long get_wchan(struct task_struct *p)
816 unsigned long bp, sp, ip;
817 unsigned long stack_page;
819 if (!p || p == current || p->state == TASK_RUNNING)
821 stack_page = (unsigned long)task_stack_page(p);
823 if (!stack_page || sp < stack_page || sp > top_esp+stack_page)
825 /* include/asm-i386/system.h:switch_to() pushes bp last. */
826 bp = *(unsigned long *) sp;
828 if (bp < stack_page || bp > top_ebp+stack_page)
830 ip = *(unsigned long *) (bp+4);
831 if (!in_sched_functions(ip))
833 bp = *(unsigned long *) bp;
834 } while (count++ < 16);
838 unsigned long arch_align_stack(unsigned long sp)
840 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
841 sp -= get_random_int() % 8192;
845 unsigned long arch_randomize_brk(struct mm_struct *mm)
847 unsigned long range_end = mm->brk + 0x02000000;
848 return randomize_range(mm->brk, range_end, 0) ? : mm->brk;