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/a.out.h>
30 #include <linux/interrupt.h>
31 #include <linux/utsname.h>
32 #include <linux/delay.h>
33 #include <linux/module.h>
34 #include <linux/ptrace.h>
35 #include <linux/random.h>
36 #include <linux/notifier.h>
37 #include <linux/kprobes.h>
38 #include <linux/kdebug.h>
39 #include <linux/tick.h>
41 #include <asm/uaccess.h>
42 #include <asm/pgtable.h>
43 #include <asm/system.h>
45 #include <asm/processor.h>
47 #include <asm/mmu_context.h>
49 #include <asm/prctl.h>
51 #include <asm/proto.h>
55 asmlinkage extern void ret_from_fork(void);
57 unsigned long kernel_thread_flags = CLONE_VM | CLONE_UNTRACED;
59 unsigned long boot_option_idle_override = 0;
60 EXPORT_SYMBOL(boot_option_idle_override);
63 * Powermanagement idle function, if any..
65 void (*pm_idle)(void);
66 EXPORT_SYMBOL(pm_idle);
67 static DEFINE_PER_CPU(unsigned int, cpu_idle_state);
69 static ATOMIC_NOTIFIER_HEAD(idle_notifier);
71 void idle_notifier_register(struct notifier_block *n)
73 atomic_notifier_chain_register(&idle_notifier, n);
79 atomic_notifier_call_chain(&idle_notifier, IDLE_START, NULL);
82 static void __exit_idle(void)
84 if (test_and_clear_bit_pda(0, isidle) == 0)
86 atomic_notifier_call_chain(&idle_notifier, IDLE_END, NULL);
89 /* Called from interrupts to signify idle end */
92 /* idle loop has pid 0 */
99 * We use this if we don't have any better
102 void default_idle(void)
104 current_thread_info()->status &= ~TS_POLLING;
106 * TS_POLLING-cleared state must be visible before we
111 if (!need_resched()) {
116 t0n = ktime_to_ns(t0);
117 safe_halt(); /* enables interrupts racelessly */
120 t1n = ktime_to_ns(t1);
121 sched_clock_idle_wakeup_event(t1n - t0n);
124 current_thread_info()->status |= TS_POLLING;
128 * On SMP it's slightly faster (but much more power-consuming!)
129 * to poll the ->need_resched flag instead of waiting for the
130 * cross-CPU IPI to arrive. Use this option with caution.
132 static void poll_idle(void)
138 #ifdef CONFIG_HOTPLUG_CPU
139 DECLARE_PER_CPU(int, cpu_state);
142 /* We halt the CPU with physical CPU hotplug */
143 static inline void play_dead(void)
149 __get_cpu_var(cpu_state) = CPU_DEAD;
156 static inline void play_dead(void)
160 #endif /* CONFIG_HOTPLUG_CPU */
163 * The idle thread. There's no useful work to be
164 * done, so just try to conserve power and have a
165 * low exit latency (ie sit in a loop waiting for
166 * somebody to say that they'd like to reschedule)
170 current_thread_info()->status |= TS_POLLING;
171 /* endless idle loop with no priority at all */
173 tick_nohz_stop_sched_tick();
174 while (!need_resched()) {
177 if (__get_cpu_var(cpu_idle_state))
178 __get_cpu_var(cpu_idle_state) = 0;
184 if (cpu_is_offline(smp_processor_id()))
187 * Idle routines should keep interrupts disabled
188 * from here on, until they go to idle.
189 * Otherwise, idle callbacks can misfire.
194 /* In many cases the interrupt that ended idle
195 has already called exit_idle. But some idle
196 loops can be woken up without interrupt. */
200 tick_nohz_restart_sched_tick();
201 preempt_enable_no_resched();
207 static void do_nothing(void *unused)
211 void cpu_idle_wait(void)
213 unsigned int cpu, this_cpu = get_cpu();
214 cpumask_t map, tmp = current->cpus_allowed;
216 set_cpus_allowed(current, cpumask_of_cpu(this_cpu));
220 for_each_online_cpu(cpu) {
221 per_cpu(cpu_idle_state, cpu) = 1;
225 __get_cpu_var(cpu_idle_state) = 0;
230 for_each_online_cpu(cpu) {
231 if (cpu_isset(cpu, map) && !per_cpu(cpu_idle_state, cpu))
234 cpus_and(map, map, cpu_online_map);
236 * We waited 1 sec, if a CPU still did not call idle
237 * it may be because it is in idle and not waking up
238 * because it has nothing to do.
239 * Give all the remaining CPUS a kick.
241 smp_call_function_mask(map, do_nothing, 0, 0);
242 } while (!cpus_empty(map));
244 set_cpus_allowed(current, tmp);
246 EXPORT_SYMBOL_GPL(cpu_idle_wait);
249 * This uses new MONITOR/MWAIT instructions on P4 processors with PNI,
250 * which can obviate IPI to trigger checking of need_resched.
251 * We execute MONITOR against need_resched and enter optimized wait state
252 * through MWAIT. Whenever someone changes need_resched, we would be woken
253 * up from MWAIT (without an IPI).
255 * New with Core Duo processors, MWAIT can take some hints based on CPU
258 void mwait_idle_with_hints(unsigned long ax, unsigned long cx)
260 if (!need_resched()) {
261 __monitor((void *)¤t_thread_info()->flags, 0, 0);
268 /* Default MONITOR/MWAIT with no hints, used for default C1 state */
269 static void mwait_idle(void)
271 if (!need_resched()) {
272 __monitor((void *)¤t_thread_info()->flags, 0, 0);
283 void __cpuinit select_idle_routine(const struct cpuinfo_x86 *c)
286 if (cpu_has(c, X86_FEATURE_MWAIT)) {
288 * Skip, if setup has overridden idle.
289 * One CPU supports mwait => All CPUs supports mwait
293 printk(KERN_INFO "using mwait in idle threads.\n");
296 pm_idle = mwait_idle;
301 static int __init idle_setup(char *str)
303 if (!strcmp(str, "poll")) {
304 printk("using polling idle threads.\n");
306 } else if (!strcmp(str, "mwait"))
311 boot_option_idle_override = 1;
314 early_param("idle", idle_setup);
316 /* Prints also some state that isn't saved in the pt_regs */
317 void __show_regs(struct pt_regs * regs)
319 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L, fs, gs, shadowgs;
320 unsigned long d0, d1, d2, d3, d6, d7;
321 unsigned int fsindex, gsindex;
322 unsigned int ds, cs, es;
326 printk("Pid: %d, comm: %.20s %s %s %.*s\n",
327 current->pid, current->comm, print_tainted(),
328 init_utsname()->release,
329 (int)strcspn(init_utsname()->version, " "),
330 init_utsname()->version);
331 printk("RIP: %04lx:[<%016lx>] ", regs->cs & 0xffff, regs->ip);
332 printk_address(regs->ip, 1);
333 printk("RSP: %04lx:%016lx EFLAGS: %08lx\n", regs->ss, regs->sp,
335 printk("RAX: %016lx RBX: %016lx RCX: %016lx\n",
336 regs->ax, regs->bx, regs->cx);
337 printk("RDX: %016lx RSI: %016lx RDI: %016lx\n",
338 regs->dx, regs->si, regs->di);
339 printk("RBP: %016lx R08: %016lx R09: %016lx\n",
340 regs->bp, regs->r8, regs->r9);
341 printk("R10: %016lx R11: %016lx R12: %016lx\n",
342 regs->r10, regs->r11, regs->r12);
343 printk("R13: %016lx R14: %016lx R15: %016lx\n",
344 regs->r13, regs->r14, regs->r15);
346 asm("movl %%ds,%0" : "=r" (ds));
347 asm("movl %%cs,%0" : "=r" (cs));
348 asm("movl %%es,%0" : "=r" (es));
349 asm("movl %%fs,%0" : "=r" (fsindex));
350 asm("movl %%gs,%0" : "=r" (gsindex));
352 rdmsrl(MSR_FS_BASE, fs);
353 rdmsrl(MSR_GS_BASE, gs);
354 rdmsrl(MSR_KERNEL_GS_BASE, shadowgs);
361 printk("FS: %016lx(%04x) GS:%016lx(%04x) knlGS:%016lx\n",
362 fs,fsindex,gs,gsindex,shadowgs);
363 printk("CS: %04x DS: %04x ES: %04x CR0: %016lx\n", cs, ds, es, cr0);
364 printk("CR2: %016lx CR3: %016lx CR4: %016lx\n", cr2, cr3, cr4);
369 printk("DR0: %016lx DR1: %016lx DR2: %016lx\n", d0, d1, d2);
373 printk("DR3: %016lx DR6: %016lx DR7: %016lx\n", d3, d6, d7);
376 void show_regs(struct pt_regs *regs)
378 printk("CPU %d:", smp_processor_id());
380 show_trace(NULL, regs, (void *)(regs + 1), regs->bp);
384 * Free current thread data structures etc..
386 void exit_thread(void)
388 struct task_struct *me = current;
389 struct thread_struct *t = &me->thread;
391 if (me->thread.io_bitmap_ptr) {
392 struct tss_struct *tss = &per_cpu(init_tss, get_cpu());
394 kfree(t->io_bitmap_ptr);
395 t->io_bitmap_ptr = NULL;
396 clear_thread_flag(TIF_IO_BITMAP);
398 * Careful, clear this in the TSS too:
400 memset(tss->io_bitmap, 0xff, t->io_bitmap_max);
401 t->io_bitmap_max = 0;
406 void flush_thread(void)
408 struct task_struct *tsk = current;
410 if (test_tsk_thread_flag(tsk, TIF_ABI_PENDING)) {
411 clear_tsk_thread_flag(tsk, TIF_ABI_PENDING);
412 if (test_tsk_thread_flag(tsk, TIF_IA32)) {
413 clear_tsk_thread_flag(tsk, TIF_IA32);
415 set_tsk_thread_flag(tsk, TIF_IA32);
416 current_thread_info()->status |= TS_COMPAT;
419 clear_tsk_thread_flag(tsk, TIF_DEBUG);
421 tsk->thread.debugreg0 = 0;
422 tsk->thread.debugreg1 = 0;
423 tsk->thread.debugreg2 = 0;
424 tsk->thread.debugreg3 = 0;
425 tsk->thread.debugreg6 = 0;
426 tsk->thread.debugreg7 = 0;
427 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
429 * Forget coprocessor state..
435 void release_thread(struct task_struct *dead_task)
438 if (dead_task->mm->context.size) {
439 printk("WARNING: dead process %8s still has LDT? <%p/%d>\n",
441 dead_task->mm->context.ldt,
442 dead_task->mm->context.size);
448 static inline void set_32bit_tls(struct task_struct *t, int tls, u32 addr)
450 struct user_desc ud = {
457 struct desc_struct *desc = (void *)t->thread.tls_array;
462 static inline u32 read_32bit_tls(struct task_struct *t, int tls)
464 return get_desc_base(&t->thread.tls_array[tls]);
468 * This gets called before we allocate a new thread and copy
469 * the current task into it.
471 void prepare_to_copy(struct task_struct *tsk)
476 int copy_thread(int nr, unsigned long clone_flags, unsigned long sp,
477 unsigned long unused,
478 struct task_struct * p, struct pt_regs * regs)
481 struct pt_regs * childregs;
482 struct task_struct *me = current;
484 childregs = ((struct pt_regs *)
485 (THREAD_SIZE + task_stack_page(p))) - 1;
491 childregs->sp = (unsigned long)childregs;
493 p->thread.sp = (unsigned long) childregs;
494 p->thread.sp0 = (unsigned long) (childregs+1);
495 p->thread.usersp = me->thread.usersp;
497 set_tsk_thread_flag(p, TIF_FORK);
499 p->thread.fs = me->thread.fs;
500 p->thread.gs = me->thread.gs;
502 asm("mov %%gs,%0" : "=m" (p->thread.gsindex));
503 asm("mov %%fs,%0" : "=m" (p->thread.fsindex));
504 asm("mov %%es,%0" : "=m" (p->thread.es));
505 asm("mov %%ds,%0" : "=m" (p->thread.ds));
507 if (unlikely(test_tsk_thread_flag(me, TIF_IO_BITMAP))) {
508 p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL);
509 if (!p->thread.io_bitmap_ptr) {
510 p->thread.io_bitmap_max = 0;
513 memcpy(p->thread.io_bitmap_ptr, me->thread.io_bitmap_ptr,
515 set_tsk_thread_flag(p, TIF_IO_BITMAP);
519 * Set a new TLS for the child thread?
521 if (clone_flags & CLONE_SETTLS) {
522 #ifdef CONFIG_IA32_EMULATION
523 if (test_thread_flag(TIF_IA32))
524 err = do_set_thread_area(p, -1,
525 (struct user_desc __user *)childregs->si, 0);
528 err = do_arch_prctl(p, ARCH_SET_FS, childregs->r8);
534 if (err && p->thread.io_bitmap_ptr) {
535 kfree(p->thread.io_bitmap_ptr);
536 p->thread.io_bitmap_max = 0;
542 * This special macro can be used to load a debugging register
544 #define loaddebug(thread, r) set_debugreg(thread->debugreg ## r, r)
546 static inline void __switch_to_xtra(struct task_struct *prev_p,
547 struct task_struct *next_p,
548 struct tss_struct *tss)
550 struct thread_struct *prev, *next;
551 unsigned long debugctl;
553 prev = &prev_p->thread,
554 next = &next_p->thread;
556 debugctl = prev->debugctlmsr;
557 if (next->ds_area_msr != prev->ds_area_msr) {
558 /* we clear debugctl to make sure DS
559 * is not in use when we change it */
561 wrmsrl(MSR_IA32_DEBUGCTLMSR, 0);
562 wrmsrl(MSR_IA32_DS_AREA, next->ds_area_msr);
565 if (next->debugctlmsr != debugctl)
566 wrmsrl(MSR_IA32_DEBUGCTLMSR, next->debugctlmsr);
568 if (test_tsk_thread_flag(next_p, TIF_DEBUG)) {
578 if (test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
580 * Copy the relevant range of the IO bitmap.
581 * Normally this is 128 bytes or less:
583 memcpy(tss->io_bitmap, next->io_bitmap_ptr,
584 max(prev->io_bitmap_max, next->io_bitmap_max));
585 } else if (test_tsk_thread_flag(prev_p, TIF_IO_BITMAP)) {
587 * Clear any possible leftover bits:
589 memset(tss->io_bitmap, 0xff, prev->io_bitmap_max);
592 if (test_tsk_thread_flag(prev_p, TIF_BTS_TRACE_TS))
593 ptrace_bts_take_timestamp(prev_p, BTS_TASK_DEPARTS);
595 if (test_tsk_thread_flag(next_p, TIF_BTS_TRACE_TS))
596 ptrace_bts_take_timestamp(next_p, BTS_TASK_ARRIVES);
600 * switch_to(x,y) should switch tasks from x to y.
602 * This could still be optimized:
603 * - fold all the options into a flag word and test it with a single test.
604 * - could test fs/gs bitsliced
606 * Kprobes not supported here. Set the probe on schedule instead.
609 __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
611 struct thread_struct *prev = &prev_p->thread,
612 *next = &next_p->thread;
613 int cpu = smp_processor_id();
614 struct tss_struct *tss = &per_cpu(init_tss, cpu);
616 /* we're going to use this soon, after a few expensive things */
617 if (next_p->fpu_counter>5)
618 prefetch(&next->i387.fxsave);
621 * Reload esp0, LDT and the page table pointer:
627 * This won't pick up thread selector changes, but I guess that is ok.
629 asm volatile("mov %%es,%0" : "=m" (prev->es));
630 if (unlikely(next->es | prev->es))
631 loadsegment(es, next->es);
633 asm volatile ("mov %%ds,%0" : "=m" (prev->ds));
634 if (unlikely(next->ds | prev->ds))
635 loadsegment(ds, next->ds);
644 asm volatile("movl %%fs,%0" : "=r" (fsindex));
645 /* segment register != 0 always requires a reload.
646 also reload when it has changed.
647 when prev process used 64bit base always reload
648 to avoid an information leak. */
649 if (unlikely(fsindex | next->fsindex | prev->fs)) {
650 loadsegment(fs, next->fsindex);
651 /* check if the user used a selector != 0
652 * if yes clear 64bit base, since overloaded base
653 * is always mapped to the Null selector
658 /* when next process has a 64bit base use it */
660 wrmsrl(MSR_FS_BASE, next->fs);
661 prev->fsindex = fsindex;
665 asm volatile("movl %%gs,%0" : "=r" (gsindex));
666 if (unlikely(gsindex | next->gsindex | prev->gs)) {
667 load_gs_index(next->gsindex);
672 wrmsrl(MSR_KERNEL_GS_BASE, next->gs);
673 prev->gsindex = gsindex;
676 /* Must be after DS reload */
680 * Switch the PDA and FPU contexts.
682 prev->usersp = read_pda(oldrsp);
683 write_pda(oldrsp, next->usersp);
684 write_pda(pcurrent, next_p);
686 write_pda(kernelstack,
687 (unsigned long)task_stack_page(next_p) + THREAD_SIZE - PDA_STACKOFFSET);
688 #ifdef CONFIG_CC_STACKPROTECTOR
689 write_pda(stack_canary, next_p->stack_canary);
691 * Build time only check to make sure the stack_canary is at
692 * offset 40 in the pda; this is a gcc ABI requirement
694 BUILD_BUG_ON(offsetof(struct x8664_pda, stack_canary) != 40);
698 * Now maybe reload the debug registers and handle I/O bitmaps
700 if (unlikely(task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT ||
701 task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV))
702 __switch_to_xtra(prev_p, next_p, tss);
704 /* If the task has used fpu the last 5 timeslices, just do a full
705 * restore of the math state immediately to avoid the trap; the
706 * chances of needing FPU soon are obviously high now
708 if (next_p->fpu_counter>5)
709 math_state_restore();
714 * sys_execve() executes a new program.
717 long sys_execve(char __user *name, char __user * __user *argv,
718 char __user * __user *envp, struct pt_regs regs)
723 filename = getname(name);
724 error = PTR_ERR(filename);
725 if (IS_ERR(filename))
727 error = do_execve(filename, argv, envp, ®s);
732 void set_personality_64bit(void)
734 /* inherit personality from parent */
736 /* Make sure to be in 64bit mode */
737 clear_thread_flag(TIF_IA32);
739 /* TBD: overwrites user setup. Should have two bits.
740 But 64bit processes have always behaved this way,
741 so it's not too bad. The main problem is just that
742 32bit childs are affected again. */
743 current->personality &= ~READ_IMPLIES_EXEC;
746 asmlinkage long sys_fork(struct pt_regs *regs)
748 return do_fork(SIGCHLD, regs->sp, regs, 0, NULL, NULL);
752 sys_clone(unsigned long clone_flags, unsigned long newsp,
753 void __user *parent_tid, void __user *child_tid, struct pt_regs *regs)
757 return do_fork(clone_flags, newsp, regs, 0, parent_tid, child_tid);
761 * This is trivial, and on the face of it looks like it
762 * could equally well be done in user mode.
764 * Not so, for quite unobvious reasons - register pressure.
765 * In user mode vfork() cannot have a stack frame, and if
766 * done by calling the "clone()" system call directly, you
767 * do not have enough call-clobbered registers to hold all
768 * the information you need.
770 asmlinkage long sys_vfork(struct pt_regs *regs)
772 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->sp, regs, 0,
776 unsigned long get_wchan(struct task_struct *p)
782 if (!p || p == current || p->state==TASK_RUNNING)
784 stack = (unsigned long)task_stack_page(p);
785 if (p->thread.sp < stack || p->thread.sp > stack+THREAD_SIZE)
787 fp = *(u64 *)(p->thread.sp);
789 if (fp < (unsigned long)stack ||
790 fp > (unsigned long)stack+THREAD_SIZE)
793 if (!in_sched_functions(ip))
796 } while (count++ < 16);
800 long do_arch_prctl(struct task_struct *task, int code, unsigned long addr)
803 int doit = task == current;
808 if (addr >= TASK_SIZE_OF(task))
811 /* handle small bases via the GDT because that's faster to
813 if (addr <= 0xffffffff) {
814 set_32bit_tls(task, GS_TLS, addr);
816 load_TLS(&task->thread, cpu);
817 load_gs_index(GS_TLS_SEL);
819 task->thread.gsindex = GS_TLS_SEL;
822 task->thread.gsindex = 0;
823 task->thread.gs = addr;
826 ret = checking_wrmsrl(MSR_KERNEL_GS_BASE, addr);
832 /* Not strictly needed for fs, but do it for symmetry
834 if (addr >= TASK_SIZE_OF(task))
837 /* handle small bases via the GDT because that's faster to
839 if (addr <= 0xffffffff) {
840 set_32bit_tls(task, FS_TLS, addr);
842 load_TLS(&task->thread, cpu);
843 asm volatile("movl %0,%%fs" :: "r"(FS_TLS_SEL));
845 task->thread.fsindex = FS_TLS_SEL;
848 task->thread.fsindex = 0;
849 task->thread.fs = addr;
851 /* set the selector to 0 to not confuse
853 asm volatile("movl %0,%%fs" :: "r" (0));
854 ret = checking_wrmsrl(MSR_FS_BASE, addr);
861 if (task->thread.fsindex == FS_TLS_SEL)
862 base = read_32bit_tls(task, FS_TLS);
864 rdmsrl(MSR_FS_BASE, base);
866 base = task->thread.fs;
867 ret = put_user(base, (unsigned long __user *)addr);
873 if (task->thread.gsindex == GS_TLS_SEL)
874 base = read_32bit_tls(task, GS_TLS);
876 asm("movl %%gs,%0" : "=r" (gsindex));
878 rdmsrl(MSR_KERNEL_GS_BASE, base);
880 base = task->thread.gs;
883 base = task->thread.gs;
884 ret = put_user(base, (unsigned long __user *)addr);
896 long sys_arch_prctl(int code, unsigned long addr)
898 return do_arch_prctl(current, code, addr);
901 unsigned long arch_align_stack(unsigned long sp)
903 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
904 sp -= get_random_int() % 8192;
908 unsigned long arch_randomize_brk(struct mm_struct *mm)
910 unsigned long range_end = mm->brk + 0x02000000;
911 return randomize_range(mm->brk, range_end, 0) ? : mm->brk;