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);
86 void disable_hlt(void)
91 EXPORT_SYMBOL(disable_hlt);
98 EXPORT_SYMBOL(enable_hlt);
101 * We use this if we don't have any better
104 void default_idle(void)
106 if (!hlt_counter && boot_cpu_data.hlt_works_ok) {
107 current_thread_info()->status &= ~TS_POLLING;
109 * TS_POLLING-cleared state must be visible before we
115 if (!need_resched()) {
116 safe_halt(); /* enables interrupts racelessly */
120 current_thread_info()->status |= TS_POLLING;
123 /* loop is done by the caller */
127 #ifdef CONFIG_APM_MODULE
128 EXPORT_SYMBOL(default_idle);
132 * On SMP it's slightly faster (but much more power-consuming!)
133 * to poll the ->work.need_resched flag instead of waiting for the
134 * cross-CPU IPI to arrive. Use this option with caution.
136 static void poll_idle(void)
142 #ifdef CONFIG_HOTPLUG_CPU
144 /* We don't actually take CPU down, just spin without interrupts. */
145 static inline void play_dead(void)
147 /* This must be done before dead CPU ack */
152 __get_cpu_var(cpu_state) = CPU_DEAD;
155 * With physical CPU hotplug, we should halt the cpu
162 static inline void play_dead(void)
166 #endif /* CONFIG_HOTPLUG_CPU */
169 * The idle thread. There's no useful work to be
170 * done, so just try to conserve power and have a
171 * low exit latency (ie sit in a loop waiting for
172 * somebody to say that they'd like to reschedule)
176 int cpu = smp_processor_id();
178 current_thread_info()->status |= TS_POLLING;
180 /* endless idle loop with no priority at all */
182 tick_nohz_stop_sched_tick();
183 while (!need_resched()) {
190 if (rcu_pending(cpu))
191 rcu_check_callbacks(cpu, 0);
196 if (cpu_is_offline(cpu))
199 __get_cpu_var(irq_stat).idle_timestamp = jiffies;
202 tick_nohz_restart_sched_tick();
203 preempt_enable_no_resched();
209 static void do_nothing(void *unused)
214 * cpu_idle_wait - Used to ensure that all the CPUs discard old value of
215 * pm_idle and update to new pm_idle value. Required while changing pm_idle
216 * handler on SMP systems.
218 * Caller must have changed pm_idle to the new value before the call. Old
219 * pm_idle value will not be used by any CPU after the return of this function.
221 void cpu_idle_wait(void)
224 /* kick all the CPUs so that they exit out of pm_idle */
225 smp_call_function(do_nothing, NULL, 0, 1);
227 EXPORT_SYMBOL_GPL(cpu_idle_wait);
230 * This uses new MONITOR/MWAIT instructions on P4 processors with PNI,
231 * which can obviate IPI to trigger checking of need_resched.
232 * We execute MONITOR against need_resched and enter optimized wait state
233 * through MWAIT. Whenever someone changes need_resched, we would be woken
234 * up from MWAIT (without an IPI).
236 * New with Core Duo processors, MWAIT can take some hints based on CPU
239 void mwait_idle_with_hints(unsigned long ax, unsigned long cx)
241 if (!need_resched()) {
242 __monitor((void *)¤t_thread_info()->flags, 0, 0);
252 /* Default MONITOR/MWAIT with no hints, used for default C1 state */
253 static void mwait_idle(void)
256 mwait_idle_with_hints(0, 0);
259 static int __cpuinit mwait_usable(const struct cpuinfo_x86 *c)
263 /* Any C1 states supported? */
264 return c->cpuid_level >= 5 && ((cpuid_edx(5) >> 4) & 0xf) > 0;
267 void __cpuinit select_idle_routine(const struct cpuinfo_x86 *c)
273 #ifdef CONFIG_X86_SMP
274 if (pm_idle == poll_idle && smp_num_siblings > 1) {
275 printk(KERN_WARNING "WARNING: polling idle and HT enabled,"
276 " performance may degrade.\n");
279 if (cpu_has(c, X86_FEATURE_MWAIT) && mwait_usable(c)) {
281 * Skip, if setup has overridden idle.
282 * One CPU supports mwait => All CPUs supports mwait
285 printk(KERN_INFO "using mwait in idle threads.\n");
286 pm_idle = mwait_idle;
292 static int __init idle_setup(char *str)
294 if (!strcmp(str, "poll")) {
295 printk("using polling idle threads.\n");
297 } else if (!strcmp(str, "mwait"))
302 boot_option_idle_override = 1;
305 early_param("idle", idle_setup);
307 void __show_registers(struct pt_regs *regs, int all)
309 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L;
310 unsigned long d0, d1, d2, d3, d6, d7;
312 unsigned short ss, gs;
314 if (user_mode_vm(regs)) {
316 ss = regs->ss & 0xffff;
319 sp = (unsigned long) (®s->sp);
325 printk("Pid: %d, comm: %s %s (%s %.*s)\n",
326 task_pid_nr(current), current->comm,
327 print_tainted(), init_utsname()->release,
328 (int)strcspn(init_utsname()->version, " "),
329 init_utsname()->version);
331 printk("EIP: %04x:[<%08lx>] EFLAGS: %08lx CPU: %d\n",
332 (u16)regs->cs, regs->ip, regs->flags,
334 print_symbol("EIP is at %s\n", regs->ip);
336 printk("EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
337 regs->ax, regs->bx, regs->cx, regs->dx);
338 printk("ESI: %08lx EDI: %08lx EBP: %08lx ESP: %08lx\n",
339 regs->si, regs->di, regs->bp, sp);
340 printk(" DS: %04x ES: %04x FS: %04x GS: %04x SS: %04x\n",
341 (u16)regs->ds, (u16)regs->es, (u16)regs->fs, gs, ss);
349 cr4 = read_cr4_safe();
350 printk("CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n",
357 printk("DR0: %08lx DR1: %08lx DR2: %08lx DR3: %08lx\n",
362 printk("DR6: %08lx DR7: %08lx\n",
366 void show_regs(struct pt_regs *regs)
368 __show_registers(regs, 1);
369 show_trace(NULL, regs, ®s->sp, regs->bp);
373 * This gets run with %bx containing the
374 * function to call, and %dx containing
377 extern void kernel_thread_helper(void);
380 * Create a kernel thread
382 int kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
386 memset(®s, 0, sizeof(regs));
388 regs.bx = (unsigned long) fn;
389 regs.dx = (unsigned long) arg;
393 regs.fs = __KERNEL_PERCPU;
395 regs.ip = (unsigned long) kernel_thread_helper;
396 regs.cs = __KERNEL_CS | get_kernel_rpl();
397 regs.flags = X86_EFLAGS_IF | X86_EFLAGS_SF | X86_EFLAGS_PF | 0x2;
399 /* Ok, create the new process.. */
400 return do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0, ®s, 0, NULL, NULL);
402 EXPORT_SYMBOL(kernel_thread);
405 * Free current thread data structures etc..
407 void exit_thread(void)
409 /* The process may have allocated an io port bitmap... nuke it. */
410 if (unlikely(test_thread_flag(TIF_IO_BITMAP))) {
411 struct task_struct *tsk = current;
412 struct thread_struct *t = &tsk->thread;
414 struct tss_struct *tss = &per_cpu(init_tss, cpu);
416 kfree(t->io_bitmap_ptr);
417 t->io_bitmap_ptr = NULL;
418 clear_thread_flag(TIF_IO_BITMAP);
420 * Careful, clear this in the TSS too:
422 memset(tss->io_bitmap, 0xff, tss->io_bitmap_max);
423 t->io_bitmap_max = 0;
424 tss->io_bitmap_owner = NULL;
425 tss->io_bitmap_max = 0;
426 tss->x86_tss.io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
431 void flush_thread(void)
433 struct task_struct *tsk = current;
435 tsk->thread.debugreg0 = 0;
436 tsk->thread.debugreg1 = 0;
437 tsk->thread.debugreg2 = 0;
438 tsk->thread.debugreg3 = 0;
439 tsk->thread.debugreg6 = 0;
440 tsk->thread.debugreg7 = 0;
441 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
442 clear_tsk_thread_flag(tsk, TIF_DEBUG);
444 * Forget coprocessor state..
450 void release_thread(struct task_struct *dead_task)
452 BUG_ON(dead_task->mm);
453 release_vm86_irqs(dead_task);
457 * This gets called before we allocate a new thread and copy
458 * the current task into it.
460 void prepare_to_copy(struct task_struct *tsk)
465 int copy_thread(int nr, unsigned long clone_flags, unsigned long sp,
466 unsigned long unused,
467 struct task_struct * p, struct pt_regs * regs)
469 struct pt_regs * childregs;
470 struct task_struct *tsk;
473 childregs = task_pt_regs(p);
478 p->thread.sp = (unsigned long) childregs;
479 p->thread.sp0 = (unsigned long) (childregs+1);
481 p->thread.ip = (unsigned long) ret_from_fork;
483 savesegment(gs, p->thread.gs);
486 if (unlikely(test_tsk_thread_flag(tsk, TIF_IO_BITMAP))) {
487 p->thread.io_bitmap_ptr = kmemdup(tsk->thread.io_bitmap_ptr,
488 IO_BITMAP_BYTES, GFP_KERNEL);
489 if (!p->thread.io_bitmap_ptr) {
490 p->thread.io_bitmap_max = 0;
493 set_tsk_thread_flag(p, TIF_IO_BITMAP);
499 * Set a new TLS for the child thread?
501 if (clone_flags & CLONE_SETTLS)
502 err = do_set_thread_area(p, -1,
503 (struct user_desc __user *)childregs->si, 0);
505 if (err && p->thread.io_bitmap_ptr) {
506 kfree(p->thread.io_bitmap_ptr);
507 p->thread.io_bitmap_max = 0;
513 start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp)
515 __asm__("movl %0, %%gs" :: "r"(0));
518 regs->ds = __USER_DS;
519 regs->es = __USER_DS;
520 regs->ss = __USER_DS;
521 regs->cs = __USER_CS;
525 EXPORT_SYMBOL_GPL(start_thread);
527 #ifdef CONFIG_SECCOMP
528 static void hard_disable_TSC(void)
530 write_cr4(read_cr4() | X86_CR4_TSD);
532 void disable_TSC(void)
535 if (!test_and_set_thread_flag(TIF_NOTSC))
537 * Must flip the CPU state synchronously with
538 * TIF_NOTSC in the current running context.
543 static void hard_enable_TSC(void)
545 write_cr4(read_cr4() & ~X86_CR4_TSD);
547 #endif /* CONFIG_SECCOMP */
550 __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p,
551 struct tss_struct *tss)
553 struct thread_struct *prev, *next;
554 unsigned long debugctl;
556 prev = &prev_p->thread;
557 next = &next_p->thread;
559 debugctl = prev->debugctlmsr;
560 if (next->ds_area_msr != prev->ds_area_msr) {
561 /* we clear debugctl to make sure DS
562 * is not in use when we change it */
564 update_debugctlmsr(0);
565 wrmsr(MSR_IA32_DS_AREA, next->ds_area_msr, 0);
568 if (next->debugctlmsr != debugctl)
569 update_debugctlmsr(next->debugctlmsr);
571 if (test_tsk_thread_flag(next_p, TIF_DEBUG)) {
572 set_debugreg(next->debugreg0, 0);
573 set_debugreg(next->debugreg1, 1);
574 set_debugreg(next->debugreg2, 2);
575 set_debugreg(next->debugreg3, 3);
577 set_debugreg(next->debugreg6, 6);
578 set_debugreg(next->debugreg7, 7);
581 #ifdef CONFIG_SECCOMP
582 if (test_tsk_thread_flag(prev_p, TIF_NOTSC) ^
583 test_tsk_thread_flag(next_p, TIF_NOTSC)) {
584 /* prev and next are different */
585 if (test_tsk_thread_flag(next_p, TIF_NOTSC))
593 if (test_tsk_thread_flag(prev_p, TIF_BTS_TRACE_TS))
594 ptrace_bts_take_timestamp(prev_p, BTS_TASK_DEPARTS);
596 if (test_tsk_thread_flag(next_p, TIF_BTS_TRACE_TS))
597 ptrace_bts_take_timestamp(next_p, BTS_TASK_ARRIVES);
601 if (!test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
603 * Disable the bitmap via an invalid offset. We still cache
604 * the previous bitmap owner and the IO bitmap contents:
606 tss->x86_tss.io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
610 if (likely(next == tss->io_bitmap_owner)) {
612 * Previous owner of the bitmap (hence the bitmap content)
613 * matches the next task, we dont have to do anything but
614 * to set a valid offset in the TSS:
616 tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET;
620 * Lazy TSS's I/O bitmap copy. We set an invalid offset here
621 * and we let the task to get a GPF in case an I/O instruction
622 * is performed. The handler of the GPF will verify that the
623 * faulting task has a valid I/O bitmap and, it true, does the
624 * real copy and restart the instruction. This will save us
625 * redundant copies when the currently switched task does not
626 * perform any I/O during its timeslice.
628 tss->x86_tss.io_bitmap_base = INVALID_IO_BITMAP_OFFSET_LAZY;
632 * switch_to(x,yn) should switch tasks from x to y.
634 * We fsave/fwait so that an exception goes off at the right time
635 * (as a call from the fsave or fwait in effect) rather than to
636 * the wrong process. Lazy FP saving no longer makes any sense
637 * with modern CPU's, and this simplifies a lot of things (SMP
638 * and UP become the same).
640 * NOTE! We used to use the x86 hardware context switching. The
641 * reason for not using it any more becomes apparent when you
642 * try to recover gracefully from saved state that is no longer
643 * valid (stale segment register values in particular). With the
644 * hardware task-switch, there is no way to fix up bad state in
645 * a reasonable manner.
647 * The fact that Intel documents the hardware task-switching to
648 * be slow is a fairly red herring - this code is not noticeably
649 * faster. However, there _is_ some room for improvement here,
650 * so the performance issues may eventually be a valid point.
651 * More important, however, is the fact that this allows us much
654 * The return value (in %ax) will be the "prev" task after
655 * the task-switch, and shows up in ret_from_fork in entry.S,
658 struct task_struct * __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
660 struct thread_struct *prev = &prev_p->thread,
661 *next = &next_p->thread;
662 int cpu = smp_processor_id();
663 struct tss_struct *tss = &per_cpu(init_tss, cpu);
665 /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
667 __unlazy_fpu(prev_p);
670 /* we're going to use this soon, after a few expensive things */
671 if (next_p->fpu_counter > 5)
672 prefetch(&next->i387.fxsave);
680 * Save away %gs. No need to save %fs, as it was saved on the
681 * stack on entry. No need to save %es and %ds, as those are
682 * always kernel segments while inside the kernel. Doing this
683 * before setting the new TLS descriptors avoids the situation
684 * where we temporarily have non-reloadable segments in %fs
685 * and %gs. This could be an issue if the NMI handler ever
686 * used %fs or %gs (it does not today), or if the kernel is
687 * running inside of a hypervisor layer.
689 savesegment(gs, prev->gs);
692 * Load the per-thread Thread-Local Storage descriptor.
697 * Restore IOPL if needed. In normal use, the flags restore
698 * in the switch assembly will handle this. But if the kernel
699 * is running virtualized at a non-zero CPL, the popf will
700 * not restore flags, so it must be done in a separate step.
702 if (get_kernel_rpl() && unlikely(prev->iopl != next->iopl))
703 set_iopl_mask(next->iopl);
706 * Now maybe handle debug registers and/or IO bitmaps
708 if (unlikely(task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV ||
709 task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT))
710 __switch_to_xtra(prev_p, next_p, tss);
713 * Leave lazy mode, flushing any hypercalls made here.
714 * This must be done before restoring TLS segments so
715 * the GDT and LDT are properly updated, and must be
716 * done before math_state_restore, so the TS bit is up
719 arch_leave_lazy_cpu_mode();
721 /* If the task has used fpu the last 5 timeslices, just do a full
722 * restore of the math state immediately to avoid the trap; the
723 * chances of needing FPU soon are obviously high now
725 if (next_p->fpu_counter > 5)
726 math_state_restore();
729 * Restore %gs if needed (which is common)
731 if (prev->gs | next->gs)
732 loadsegment(gs, next->gs);
734 x86_write_percpu(current_task, next_p);
739 asmlinkage int sys_fork(struct pt_regs regs)
741 return do_fork(SIGCHLD, regs.sp, ®s, 0, NULL, NULL);
744 asmlinkage int sys_clone(struct pt_regs regs)
746 unsigned long clone_flags;
748 int __user *parent_tidptr, *child_tidptr;
750 clone_flags = regs.bx;
752 parent_tidptr = (int __user *)regs.dx;
753 child_tidptr = (int __user *)regs.di;
756 return do_fork(clone_flags, newsp, ®s, 0, parent_tidptr, child_tidptr);
760 * This is trivial, and on the face of it looks like it
761 * could equally well be done in user mode.
763 * Not so, for quite unobvious reasons - register pressure.
764 * In user mode vfork() cannot have a stack frame, and if
765 * done by calling the "clone()" system call directly, you
766 * do not have enough call-clobbered registers to hold all
767 * the information you need.
769 asmlinkage int sys_vfork(struct pt_regs regs)
771 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs.sp, ®s, 0, NULL, NULL);
775 * sys_execve() executes a new program.
777 asmlinkage int sys_execve(struct pt_regs regs)
782 filename = getname((char __user *) regs.bx);
783 error = PTR_ERR(filename);
784 if (IS_ERR(filename))
786 error = do_execve(filename,
787 (char __user * __user *) regs.cx,
788 (char __user * __user *) regs.dx,
791 /* Make sure we don't return using sysenter.. */
792 set_thread_flag(TIF_IRET);
799 #define top_esp (THREAD_SIZE - sizeof(unsigned long))
800 #define top_ebp (THREAD_SIZE - 2*sizeof(unsigned long))
802 unsigned long get_wchan(struct task_struct *p)
804 unsigned long bp, sp, ip;
805 unsigned long stack_page;
807 if (!p || p == current || p->state == TASK_RUNNING)
809 stack_page = (unsigned long)task_stack_page(p);
811 if (!stack_page || sp < stack_page || sp > top_esp+stack_page)
813 /* include/asm-i386/system.h:switch_to() pushes bp last. */
814 bp = *(unsigned long *) sp;
816 if (bp < stack_page || bp > top_ebp+stack_page)
818 ip = *(unsigned long *) (bp+4);
819 if (!in_sched_functions(ip))
821 bp = *(unsigned long *) bp;
822 } while (count++ < 16);
826 unsigned long arch_align_stack(unsigned long sp)
828 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
829 sp -= get_random_int() % 8192;
833 unsigned long arch_randomize_brk(struct mm_struct *mm)
835 unsigned long range_end = mm->brk + 0x02000000;
836 return randomize_range(mm->brk, range_end, 0) ? : mm->brk;