2 * Copyright (C) 1995 Linus Torvalds
4 * Pentium III FXSR, SSE support
5 * Gareth Hughes <gareth@valinux.com>, May 2000
10 * CPU hotplug support - ashok.raj@intel.com
14 * This file handles the architecture-dependent parts of process handling..
19 #include <linux/cpu.h>
20 #include <linux/errno.h>
21 #include <linux/sched.h>
23 #include <linux/kernel.h>
25 #include <linux/elfcore.h>
26 #include <linux/smp.h>
27 #include <linux/slab.h>
28 #include <linux/user.h>
29 #include <linux/interrupt.h>
30 #include <linux/utsname.h>
31 #include <linux/delay.h>
32 #include <linux/module.h>
33 #include <linux/ptrace.h>
34 #include <linux/random.h>
35 #include <linux/notifier.h>
36 #include <linux/kprobes.h>
37 #include <linux/kdebug.h>
38 #include <linux/tick.h>
40 #include <asm/uaccess.h>
41 #include <asm/pgtable.h>
42 #include <asm/system.h>
44 #include <asm/processor.h>
46 #include <asm/mmu_context.h>
48 #include <asm/prctl.h>
50 #include <asm/proto.h>
54 asmlinkage extern void ret_from_fork(void);
56 unsigned long kernel_thread_flags = CLONE_VM | CLONE_UNTRACED;
58 unsigned long boot_option_idle_override = 0;
59 EXPORT_SYMBOL(boot_option_idle_override);
62 * Powermanagement idle function, if any..
64 void (*pm_idle)(void);
65 EXPORT_SYMBOL(pm_idle);
66 static DEFINE_PER_CPU(unsigned int, cpu_idle_state);
68 static ATOMIC_NOTIFIER_HEAD(idle_notifier);
70 void idle_notifier_register(struct notifier_block *n)
72 atomic_notifier_chain_register(&idle_notifier, n);
78 atomic_notifier_call_chain(&idle_notifier, IDLE_START, NULL);
81 static void __exit_idle(void)
83 if (test_and_clear_bit_pda(0, isidle) == 0)
85 atomic_notifier_call_chain(&idle_notifier, IDLE_END, NULL);
88 /* Called from interrupts to signify idle end */
91 /* idle loop has pid 0 */
98 * We use this if we don't have any better
101 void default_idle(void)
103 current_thread_info()->status &= ~TS_POLLING;
105 * TS_POLLING-cleared state must be visible before we
110 if (!need_resched()) {
115 t0n = ktime_to_ns(t0);
116 safe_halt(); /* enables interrupts racelessly */
119 t1n = ktime_to_ns(t1);
120 sched_clock_idle_wakeup_event(t1n - t0n);
123 current_thread_info()->status |= TS_POLLING;
127 * On SMP it's slightly faster (but much more power-consuming!)
128 * to poll the ->need_resched flag instead of waiting for the
129 * cross-CPU IPI to arrive. Use this option with caution.
131 static void poll_idle(void)
137 #ifdef CONFIG_HOTPLUG_CPU
138 DECLARE_PER_CPU(int, cpu_state);
141 /* We halt the CPU with physical CPU hotplug */
142 static inline void play_dead(void)
148 __get_cpu_var(cpu_state) = CPU_DEAD;
155 static inline void play_dead(void)
159 #endif /* CONFIG_HOTPLUG_CPU */
162 * The idle thread. There's no useful work to be
163 * done, so just try to conserve power and have a
164 * low exit latency (ie sit in a loop waiting for
165 * somebody to say that they'd like to reschedule)
169 current_thread_info()->status |= TS_POLLING;
170 /* endless idle loop with no priority at all */
172 tick_nohz_stop_sched_tick();
173 while (!need_resched()) {
176 if (__get_cpu_var(cpu_idle_state))
177 __get_cpu_var(cpu_idle_state) = 0;
183 if (cpu_is_offline(smp_processor_id()))
186 * Idle routines should keep interrupts disabled
187 * from here on, until they go to idle.
188 * Otherwise, idle callbacks can misfire.
193 /* In many cases the interrupt that ended idle
194 has already called exit_idle. But some idle
195 loops can be woken up without interrupt. */
199 tick_nohz_restart_sched_tick();
200 preempt_enable_no_resched();
206 static void do_nothing(void *unused)
210 void cpu_idle_wait(void)
212 unsigned int cpu, this_cpu = get_cpu();
213 cpumask_t map, tmp = current->cpus_allowed;
215 set_cpus_allowed(current, cpumask_of_cpu(this_cpu));
219 for_each_online_cpu(cpu) {
220 per_cpu(cpu_idle_state, cpu) = 1;
224 __get_cpu_var(cpu_idle_state) = 0;
229 for_each_online_cpu(cpu) {
230 if (cpu_isset(cpu, map) && !per_cpu(cpu_idle_state, cpu))
233 cpus_and(map, map, cpu_online_map);
235 * We waited 1 sec, if a CPU still did not call idle
236 * it may be because it is in idle and not waking up
237 * because it has nothing to do.
238 * Give all the remaining CPUS a kick.
240 smp_call_function_mask(map, do_nothing, 0, 0);
241 } while (!cpus_empty(map));
243 set_cpus_allowed(current, tmp);
245 EXPORT_SYMBOL_GPL(cpu_idle_wait);
248 * This uses new MONITOR/MWAIT instructions on P4 processors with PNI,
249 * which can obviate IPI to trigger checking of need_resched.
250 * We execute MONITOR against need_resched and enter optimized wait state
251 * through MWAIT. Whenever someone changes need_resched, we would be woken
252 * up from MWAIT (without an IPI).
254 * New with Core Duo processors, MWAIT can take some hints based on CPU
257 void mwait_idle_with_hints(unsigned long ax, unsigned long cx)
259 if (!need_resched()) {
260 __monitor((void *)¤t_thread_info()->flags, 0, 0);
267 /* Default MONITOR/MWAIT with no hints, used for default C1 state */
268 static void mwait_idle(void)
270 if (!need_resched()) {
271 __monitor((void *)¤t_thread_info()->flags, 0, 0);
283 static int __cpuinit mwait_usable(const struct cpuinfo_x86 *c)
287 /* Any C1 states supported? */
288 return c->cpuid_level >= 5 && ((cpuid_edx(5) >> 4) & 0xf) > 0;
291 void __cpuinit select_idle_routine(const struct cpuinfo_x86 *c)
297 #ifdef CONFIG_X86_SMP
298 if (pm_idle == poll_idle && smp_num_siblings > 1) {
299 printk(KERN_WARNING "WARNING: polling idle and HT enabled,"
300 " performance may degrade.\n");
303 if (cpu_has(c, X86_FEATURE_MWAIT) && mwait_usable(c)) {
305 * Skip, if setup has overridden idle.
306 * One CPU supports mwait => All CPUs supports mwait
309 printk(KERN_INFO "using mwait in idle threads.\n");
310 pm_idle = mwait_idle;
316 static int __init idle_setup(char *str)
318 if (!strcmp(str, "poll")) {
319 printk("using polling idle threads.\n");
321 } else if (!strcmp(str, "mwait"))
326 boot_option_idle_override = 1;
329 early_param("idle", idle_setup);
331 /* Prints also some state that isn't saved in the pt_regs */
332 void __show_regs(struct pt_regs * regs)
334 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L, fs, gs, shadowgs;
335 unsigned long d0, d1, d2, d3, d6, d7;
336 unsigned int fsindex, gsindex;
337 unsigned int ds, cs, es;
341 printk("Pid: %d, comm: %.20s %s %s %.*s\n",
342 current->pid, current->comm, print_tainted(),
343 init_utsname()->release,
344 (int)strcspn(init_utsname()->version, " "),
345 init_utsname()->version);
346 printk("RIP: %04lx:[<%016lx>] ", regs->cs & 0xffff, regs->ip);
347 printk_address(regs->ip, 1);
348 printk("RSP: %04lx:%016lx EFLAGS: %08lx\n", regs->ss, regs->sp,
350 printk("RAX: %016lx RBX: %016lx RCX: %016lx\n",
351 regs->ax, regs->bx, regs->cx);
352 printk("RDX: %016lx RSI: %016lx RDI: %016lx\n",
353 regs->dx, regs->si, regs->di);
354 printk("RBP: %016lx R08: %016lx R09: %016lx\n",
355 regs->bp, regs->r8, regs->r9);
356 printk("R10: %016lx R11: %016lx R12: %016lx\n",
357 regs->r10, regs->r11, regs->r12);
358 printk("R13: %016lx R14: %016lx R15: %016lx\n",
359 regs->r13, regs->r14, regs->r15);
361 asm("movl %%ds,%0" : "=r" (ds));
362 asm("movl %%cs,%0" : "=r" (cs));
363 asm("movl %%es,%0" : "=r" (es));
364 asm("movl %%fs,%0" : "=r" (fsindex));
365 asm("movl %%gs,%0" : "=r" (gsindex));
367 rdmsrl(MSR_FS_BASE, fs);
368 rdmsrl(MSR_GS_BASE, gs);
369 rdmsrl(MSR_KERNEL_GS_BASE, shadowgs);
376 printk("FS: %016lx(%04x) GS:%016lx(%04x) knlGS:%016lx\n",
377 fs,fsindex,gs,gsindex,shadowgs);
378 printk("CS: %04x DS: %04x ES: %04x CR0: %016lx\n", cs, ds, es, cr0);
379 printk("CR2: %016lx CR3: %016lx CR4: %016lx\n", cr2, cr3, cr4);
384 printk("DR0: %016lx DR1: %016lx DR2: %016lx\n", d0, d1, d2);
388 printk("DR3: %016lx DR6: %016lx DR7: %016lx\n", d3, d6, d7);
391 void show_regs(struct pt_regs *regs)
393 printk("CPU %d:", smp_processor_id());
395 show_trace(NULL, regs, (void *)(regs + 1), regs->bp);
399 * Free current thread data structures etc..
401 void exit_thread(void)
403 struct task_struct *me = current;
404 struct thread_struct *t = &me->thread;
406 if (me->thread.io_bitmap_ptr) {
407 struct tss_struct *tss = &per_cpu(init_tss, get_cpu());
409 kfree(t->io_bitmap_ptr);
410 t->io_bitmap_ptr = NULL;
411 clear_thread_flag(TIF_IO_BITMAP);
413 * Careful, clear this in the TSS too:
415 memset(tss->io_bitmap, 0xff, t->io_bitmap_max);
416 t->io_bitmap_max = 0;
421 void flush_thread(void)
423 struct task_struct *tsk = current;
425 if (test_tsk_thread_flag(tsk, TIF_ABI_PENDING)) {
426 clear_tsk_thread_flag(tsk, TIF_ABI_PENDING);
427 if (test_tsk_thread_flag(tsk, TIF_IA32)) {
428 clear_tsk_thread_flag(tsk, TIF_IA32);
430 set_tsk_thread_flag(tsk, TIF_IA32);
431 current_thread_info()->status |= TS_COMPAT;
434 clear_tsk_thread_flag(tsk, TIF_DEBUG);
436 tsk->thread.debugreg0 = 0;
437 tsk->thread.debugreg1 = 0;
438 tsk->thread.debugreg2 = 0;
439 tsk->thread.debugreg3 = 0;
440 tsk->thread.debugreg6 = 0;
441 tsk->thread.debugreg7 = 0;
442 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
444 * Forget coprocessor state..
450 void release_thread(struct task_struct *dead_task)
453 if (dead_task->mm->context.size) {
454 printk("WARNING: dead process %8s still has LDT? <%p/%d>\n",
456 dead_task->mm->context.ldt,
457 dead_task->mm->context.size);
463 static inline void set_32bit_tls(struct task_struct *t, int tls, u32 addr)
465 struct user_desc ud = {
472 struct desc_struct *desc = t->thread.tls_array;
477 static inline u32 read_32bit_tls(struct task_struct *t, int tls)
479 return get_desc_base(&t->thread.tls_array[tls]);
483 * This gets called before we allocate a new thread and copy
484 * the current task into it.
486 void prepare_to_copy(struct task_struct *tsk)
491 int copy_thread(int nr, unsigned long clone_flags, unsigned long sp,
492 unsigned long unused,
493 struct task_struct * p, struct pt_regs * regs)
496 struct pt_regs * childregs;
497 struct task_struct *me = current;
499 childregs = ((struct pt_regs *)
500 (THREAD_SIZE + task_stack_page(p))) - 1;
506 childregs->sp = (unsigned long)childregs;
508 p->thread.sp = (unsigned long) childregs;
509 p->thread.sp0 = (unsigned long) (childregs+1);
510 p->thread.usersp = me->thread.usersp;
512 set_tsk_thread_flag(p, TIF_FORK);
514 p->thread.fs = me->thread.fs;
515 p->thread.gs = me->thread.gs;
517 asm("mov %%gs,%0" : "=m" (p->thread.gsindex));
518 asm("mov %%fs,%0" : "=m" (p->thread.fsindex));
519 asm("mov %%es,%0" : "=m" (p->thread.es));
520 asm("mov %%ds,%0" : "=m" (p->thread.ds));
522 if (unlikely(test_tsk_thread_flag(me, TIF_IO_BITMAP))) {
523 p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL);
524 if (!p->thread.io_bitmap_ptr) {
525 p->thread.io_bitmap_max = 0;
528 memcpy(p->thread.io_bitmap_ptr, me->thread.io_bitmap_ptr,
530 set_tsk_thread_flag(p, TIF_IO_BITMAP);
534 * Set a new TLS for the child thread?
536 if (clone_flags & CLONE_SETTLS) {
537 #ifdef CONFIG_IA32_EMULATION
538 if (test_thread_flag(TIF_IA32))
539 err = do_set_thread_area(p, -1,
540 (struct user_desc __user *)childregs->si, 0);
543 err = do_arch_prctl(p, ARCH_SET_FS, childregs->r8);
549 if (err && p->thread.io_bitmap_ptr) {
550 kfree(p->thread.io_bitmap_ptr);
551 p->thread.io_bitmap_max = 0;
557 * This special macro can be used to load a debugging register
559 #define loaddebug(thread, r) set_debugreg(thread->debugreg ## r, r)
561 static inline void __switch_to_xtra(struct task_struct *prev_p,
562 struct task_struct *next_p,
563 struct tss_struct *tss)
565 struct thread_struct *prev, *next;
566 unsigned long debugctl;
568 prev = &prev_p->thread,
569 next = &next_p->thread;
571 debugctl = prev->debugctlmsr;
572 if (next->ds_area_msr != prev->ds_area_msr) {
573 /* we clear debugctl to make sure DS
574 * is not in use when we change it */
576 wrmsrl(MSR_IA32_DEBUGCTLMSR, 0);
577 wrmsrl(MSR_IA32_DS_AREA, next->ds_area_msr);
580 if (next->debugctlmsr != debugctl)
581 wrmsrl(MSR_IA32_DEBUGCTLMSR, next->debugctlmsr);
583 if (test_tsk_thread_flag(next_p, TIF_DEBUG)) {
593 if (test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
595 * Copy the relevant range of the IO bitmap.
596 * Normally this is 128 bytes or less:
598 memcpy(tss->io_bitmap, next->io_bitmap_ptr,
599 max(prev->io_bitmap_max, next->io_bitmap_max));
600 } else if (test_tsk_thread_flag(prev_p, TIF_IO_BITMAP)) {
602 * Clear any possible leftover bits:
604 memset(tss->io_bitmap, 0xff, prev->io_bitmap_max);
608 if (test_tsk_thread_flag(prev_p, TIF_BTS_TRACE_TS))
609 ptrace_bts_take_timestamp(prev_p, BTS_TASK_DEPARTS);
611 if (test_tsk_thread_flag(next_p, TIF_BTS_TRACE_TS))
612 ptrace_bts_take_timestamp(next_p, BTS_TASK_ARRIVES);
617 * switch_to(x,y) should switch tasks from x to y.
619 * This could still be optimized:
620 * - fold all the options into a flag word and test it with a single test.
621 * - could test fs/gs bitsliced
623 * Kprobes not supported here. Set the probe on schedule instead.
626 __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
628 struct thread_struct *prev = &prev_p->thread,
629 *next = &next_p->thread;
630 int cpu = smp_processor_id();
631 struct tss_struct *tss = &per_cpu(init_tss, cpu);
633 /* we're going to use this soon, after a few expensive things */
634 if (next_p->fpu_counter>5)
635 prefetch(&next->i387.fxsave);
638 * Reload esp0, LDT and the page table pointer:
644 * This won't pick up thread selector changes, but I guess that is ok.
646 asm volatile("mov %%es,%0" : "=m" (prev->es));
647 if (unlikely(next->es | prev->es))
648 loadsegment(es, next->es);
650 asm volatile ("mov %%ds,%0" : "=m" (prev->ds));
651 if (unlikely(next->ds | prev->ds))
652 loadsegment(ds, next->ds);
661 asm volatile("movl %%fs,%0" : "=r" (fsindex));
662 /* segment register != 0 always requires a reload.
663 also reload when it has changed.
664 when prev process used 64bit base always reload
665 to avoid an information leak. */
666 if (unlikely(fsindex | next->fsindex | prev->fs)) {
667 loadsegment(fs, next->fsindex);
668 /* check if the user used a selector != 0
669 * if yes clear 64bit base, since overloaded base
670 * is always mapped to the Null selector
675 /* when next process has a 64bit base use it */
677 wrmsrl(MSR_FS_BASE, next->fs);
678 prev->fsindex = fsindex;
682 asm volatile("movl %%gs,%0" : "=r" (gsindex));
683 if (unlikely(gsindex | next->gsindex | prev->gs)) {
684 load_gs_index(next->gsindex);
689 wrmsrl(MSR_KERNEL_GS_BASE, next->gs);
690 prev->gsindex = gsindex;
693 /* Must be after DS reload */
697 * Switch the PDA and FPU contexts.
699 prev->usersp = read_pda(oldrsp);
700 write_pda(oldrsp, next->usersp);
701 write_pda(pcurrent, next_p);
703 write_pda(kernelstack,
704 (unsigned long)task_stack_page(next_p) + THREAD_SIZE - PDA_STACKOFFSET);
705 #ifdef CONFIG_CC_STACKPROTECTOR
706 write_pda(stack_canary, next_p->stack_canary);
708 * Build time only check to make sure the stack_canary is at
709 * offset 40 in the pda; this is a gcc ABI requirement
711 BUILD_BUG_ON(offsetof(struct x8664_pda, stack_canary) != 40);
715 * Now maybe reload the debug registers and handle I/O bitmaps
717 if (unlikely(task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT ||
718 task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV))
719 __switch_to_xtra(prev_p, next_p, tss);
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();
731 * sys_execve() executes a new program.
734 long sys_execve(char __user *name, char __user * __user *argv,
735 char __user * __user *envp, struct pt_regs *regs)
740 filename = getname(name);
741 error = PTR_ERR(filename);
742 if (IS_ERR(filename))
744 error = do_execve(filename, argv, envp, regs);
749 void set_personality_64bit(void)
751 /* inherit personality from parent */
753 /* Make sure to be in 64bit mode */
754 clear_thread_flag(TIF_IA32);
756 /* TBD: overwrites user setup. Should have two bits.
757 But 64bit processes have always behaved this way,
758 so it's not too bad. The main problem is just that
759 32bit childs are affected again. */
760 current->personality &= ~READ_IMPLIES_EXEC;
763 asmlinkage long sys_fork(struct pt_regs *regs)
765 return do_fork(SIGCHLD, regs->sp, regs, 0, NULL, NULL);
769 sys_clone(unsigned long clone_flags, unsigned long newsp,
770 void __user *parent_tid, void __user *child_tid, struct pt_regs *regs)
774 return do_fork(clone_flags, newsp, regs, 0, parent_tid, child_tid);
778 * This is trivial, and on the face of it looks like it
779 * could equally well be done in user mode.
781 * Not so, for quite unobvious reasons - register pressure.
782 * In user mode vfork() cannot have a stack frame, and if
783 * done by calling the "clone()" system call directly, you
784 * do not have enough call-clobbered registers to hold all
785 * the information you need.
787 asmlinkage long sys_vfork(struct pt_regs *regs)
789 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->sp, regs, 0,
793 unsigned long get_wchan(struct task_struct *p)
799 if (!p || p == current || p->state==TASK_RUNNING)
801 stack = (unsigned long)task_stack_page(p);
802 if (p->thread.sp < stack || p->thread.sp > stack+THREAD_SIZE)
804 fp = *(u64 *)(p->thread.sp);
806 if (fp < (unsigned long)stack ||
807 fp > (unsigned long)stack+THREAD_SIZE)
810 if (!in_sched_functions(ip))
813 } while (count++ < 16);
817 long do_arch_prctl(struct task_struct *task, int code, unsigned long addr)
820 int doit = task == current;
825 if (addr >= TASK_SIZE_OF(task))
828 /* handle small bases via the GDT because that's faster to
830 if (addr <= 0xffffffff) {
831 set_32bit_tls(task, GS_TLS, addr);
833 load_TLS(&task->thread, cpu);
834 load_gs_index(GS_TLS_SEL);
836 task->thread.gsindex = GS_TLS_SEL;
839 task->thread.gsindex = 0;
840 task->thread.gs = addr;
843 ret = checking_wrmsrl(MSR_KERNEL_GS_BASE, addr);
849 /* Not strictly needed for fs, but do it for symmetry
851 if (addr >= TASK_SIZE_OF(task))
854 /* handle small bases via the GDT because that's faster to
856 if (addr <= 0xffffffff) {
857 set_32bit_tls(task, FS_TLS, addr);
859 load_TLS(&task->thread, cpu);
860 asm volatile("movl %0,%%fs" :: "r"(FS_TLS_SEL));
862 task->thread.fsindex = FS_TLS_SEL;
865 task->thread.fsindex = 0;
866 task->thread.fs = addr;
868 /* set the selector to 0 to not confuse
870 asm volatile("movl %0,%%fs" :: "r" (0));
871 ret = checking_wrmsrl(MSR_FS_BASE, addr);
878 if (task->thread.fsindex == FS_TLS_SEL)
879 base = read_32bit_tls(task, FS_TLS);
881 rdmsrl(MSR_FS_BASE, base);
883 base = task->thread.fs;
884 ret = put_user(base, (unsigned long __user *)addr);
890 if (task->thread.gsindex == GS_TLS_SEL)
891 base = read_32bit_tls(task, GS_TLS);
893 asm("movl %%gs,%0" : "=r" (gsindex));
895 rdmsrl(MSR_KERNEL_GS_BASE, base);
897 base = task->thread.gs;
900 base = task->thread.gs;
901 ret = put_user(base, (unsigned long __user *)addr);
913 long sys_arch_prctl(int code, unsigned long addr)
915 return do_arch_prctl(current, code, addr);
918 unsigned long arch_align_stack(unsigned long sp)
920 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
921 sp -= get_random_int() % 8192;
925 unsigned long arch_randomize_brk(struct mm_struct *mm)
927 unsigned long range_end = mm->brk + 0x02000000;
928 return randomize_range(mm->brk, range_end, 0) ? : mm->brk;