2 * linux/arch/x86-64/kernel/process.c
4 * Copyright (C) 1995 Linus Torvalds
6 * Pentium III FXSR, SSE support
7 * Gareth Hughes <gareth@valinux.com>, May 2000
12 * CPU hotplug support - ashok.raj@intel.com
16 * This file handles the architecture-dependent parts of process handling..
21 #include <linux/cpu.h>
22 #include <linux/errno.h>
23 #include <linux/sched.h>
24 #include <linux/kernel.h>
26 #include <linux/elfcore.h>
27 #include <linux/smp.h>
28 #include <linux/slab.h>
29 #include <linux/user.h>
30 #include <linux/module.h>
31 #include <linux/a.out.h>
32 #include <linux/interrupt.h>
33 #include <linux/delay.h>
34 #include <linux/ptrace.h>
35 #include <linux/utsname.h>
36 #include <linux/random.h>
37 #include <linux/notifier.h>
38 #include <linux/kprobes.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>
49 #include <asm/kdebug.h>
51 #include <asm/proto.h>
55 asmlinkage extern void ret_from_fork(void);
57 unsigned long kernel_thread_flags =
58 CLONE_VM | CLONE_UNTRACED | CLONE_KTHREAD;
60 unsigned long boot_option_idle_override = 0;
61 EXPORT_SYMBOL(boot_option_idle_override);
64 * Powermanagement idle function, if any..
66 void (*pm_idle)(void);
67 EXPORT_SYMBOL(pm_idle);
68 static DEFINE_PER_CPU(unsigned int, cpu_idle_state);
70 static ATOMIC_NOTIFIER_HEAD(idle_notifier);
72 void idle_notifier_register(struct notifier_block *n)
74 atomic_notifier_chain_register(&idle_notifier, n);
76 EXPORT_SYMBOL_GPL(idle_notifier_register);
78 void idle_notifier_unregister(struct notifier_block *n)
80 atomic_notifier_chain_unregister(&idle_notifier, n);
82 EXPORT_SYMBOL(idle_notifier_unregister);
87 atomic_notifier_call_chain(&idle_notifier, IDLE_START, NULL);
90 static void __exit_idle(void)
92 if (test_and_clear_bit_pda(0, isidle) == 0)
94 atomic_notifier_call_chain(&idle_notifier, IDLE_END, NULL);
97 /* Called from interrupts to signify idle end */
100 /* idle loop has pid 0 */
107 * We use this if we don't have any better
110 static void default_idle(void)
112 current_thread_info()->status &= ~TS_POLLING;
114 * TS_POLLING-cleared state must be visible before we
119 if (!need_resched()) {
120 /* Enables interrupts one instruction before HLT.
121 x86 special cases this so there is no race. */
125 current_thread_info()->status |= TS_POLLING;
129 * On SMP it's slightly faster (but much more power-consuming!)
130 * to poll the ->need_resched flag instead of waiting for the
131 * cross-CPU IPI to arrive. Use this option with caution.
133 static void poll_idle (void)
139 void cpu_idle_wait(void)
141 unsigned int cpu, this_cpu = get_cpu();
142 cpumask_t map, tmp = current->cpus_allowed;
144 set_cpus_allowed(current, cpumask_of_cpu(this_cpu));
148 for_each_online_cpu(cpu) {
149 per_cpu(cpu_idle_state, cpu) = 1;
153 __get_cpu_var(cpu_idle_state) = 0;
158 for_each_online_cpu(cpu) {
159 if (cpu_isset(cpu, map) &&
160 !per_cpu(cpu_idle_state, cpu))
163 cpus_and(map, map, cpu_online_map);
164 } while (!cpus_empty(map));
166 set_cpus_allowed(current, tmp);
168 EXPORT_SYMBOL_GPL(cpu_idle_wait);
170 #ifdef CONFIG_HOTPLUG_CPU
171 DECLARE_PER_CPU(int, cpu_state);
174 /* We halt the CPU with physical CPU hotplug */
175 static inline void play_dead(void)
181 __get_cpu_var(cpu_state) = CPU_DEAD;
188 static inline void play_dead(void)
192 #endif /* CONFIG_HOTPLUG_CPU */
195 * The idle thread. There's no useful work to be
196 * done, so just try to conserve power and have a
197 * low exit latency (ie sit in a loop waiting for
198 * somebody to say that they'd like to reschedule)
202 current_thread_info()->status |= TS_POLLING;
203 /* endless idle loop with no priority at all */
205 while (!need_resched()) {
208 if (__get_cpu_var(cpu_idle_state))
209 __get_cpu_var(cpu_idle_state) = 0;
215 if (cpu_is_offline(smp_processor_id()))
218 * Idle routines should keep interrupts disabled
219 * from here on, until they go to idle.
220 * Otherwise, idle callbacks can misfire.
225 /* In many cases the interrupt that ended idle
226 has already called exit_idle. But some idle
227 loops can be woken up without interrupt. */
231 preempt_enable_no_resched();
238 * This uses new MONITOR/MWAIT instructions on P4 processors with PNI,
239 * which can obviate IPI to trigger checking of need_resched.
240 * We execute MONITOR against need_resched and enter optimized wait state
241 * through MWAIT. Whenever someone changes need_resched, we would be woken
242 * up from MWAIT (without an IPI).
244 * New with Core Duo processors, MWAIT can take some hints based on CPU
247 void mwait_idle_with_hints(unsigned long eax, unsigned long ecx)
249 if (!need_resched()) {
250 __monitor((void *)¤t_thread_info()->flags, 0, 0);
257 /* Default MONITOR/MWAIT with no hints, used for default C1 state */
258 static void mwait_idle(void)
260 if (!need_resched()) {
261 __monitor((void *)¤t_thread_info()->flags, 0, 0);
272 void __cpuinit select_idle_routine(const struct cpuinfo_x86 *c)
275 if (cpu_has(c, X86_FEATURE_MWAIT)) {
277 * Skip, if setup has overridden idle.
278 * One CPU supports mwait => All CPUs supports mwait
282 printk("using mwait in idle threads.\n");
285 pm_idle = mwait_idle;
290 static int __init idle_setup (char *str)
292 if (!strncmp(str, "poll", 4)) {
293 printk("using polling idle threads.\n");
297 boot_option_idle_override = 1;
301 __setup("idle=", idle_setup);
303 /* Prints also some state that isn't saved in the pt_regs */
304 void __show_regs(struct pt_regs * regs)
306 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L, fs, gs, shadowgs;
307 unsigned int fsindex,gsindex;
308 unsigned int ds,cs,es;
312 printk("Pid: %d:#%u, comm: %.20s %s %s %.*s\n",
313 current->pid, current->xid, current->comm, print_tainted(),
314 init_utsname()->release,
315 (int)strcspn(init_utsname()->version, " "),
316 init_utsname()->version);
317 printk("RIP: %04lx:[<%016lx>] ", regs->cs & 0xffff, regs->rip);
318 printk_address(regs->rip);
319 printk("RSP: %04lx:%016lx EFLAGS: %08lx\n", regs->ss, regs->rsp,
321 printk("RAX: %016lx RBX: %016lx RCX: %016lx\n",
322 regs->rax, regs->rbx, regs->rcx);
323 printk("RDX: %016lx RSI: %016lx RDI: %016lx\n",
324 regs->rdx, regs->rsi, regs->rdi);
325 printk("RBP: %016lx R08: %016lx R09: %016lx\n",
326 regs->rbp, regs->r8, regs->r9);
327 printk("R10: %016lx R11: %016lx R12: %016lx\n",
328 regs->r10, regs->r11, regs->r12);
329 printk("R13: %016lx R14: %016lx R15: %016lx\n",
330 regs->r13, regs->r14, regs->r15);
332 asm("movl %%ds,%0" : "=r" (ds));
333 asm("movl %%cs,%0" : "=r" (cs));
334 asm("movl %%es,%0" : "=r" (es));
335 asm("movl %%fs,%0" : "=r" (fsindex));
336 asm("movl %%gs,%0" : "=r" (gsindex));
338 rdmsrl(MSR_FS_BASE, fs);
339 rdmsrl(MSR_GS_BASE, gs);
340 rdmsrl(MSR_KERNEL_GS_BASE, shadowgs);
342 asm("movq %%cr0, %0": "=r" (cr0));
343 asm("movq %%cr2, %0": "=r" (cr2));
344 asm("movq %%cr3, %0": "=r" (cr3));
345 asm("movq %%cr4, %0": "=r" (cr4));
347 printk("FS: %016lx(%04x) GS:%016lx(%04x) knlGS:%016lx\n",
348 fs,fsindex,gs,gsindex,shadowgs);
349 printk("CS: %04x DS: %04x ES: %04x CR0: %016lx\n", cs, ds, es, cr0);
350 printk("CR2: %016lx CR3: %016lx CR4: %016lx\n", cr2, cr3, cr4);
353 void show_regs(struct pt_regs *regs)
355 printk("CPU %d:", smp_processor_id());
357 show_trace(NULL, regs, (void *)(regs + 1));
361 * Free current thread data structures etc..
363 void exit_thread(void)
365 struct task_struct *me = current;
366 struct thread_struct *t = &me->thread;
368 if (me->thread.io_bitmap_ptr) {
369 struct tss_struct *tss = &per_cpu(init_tss, get_cpu());
371 kfree(t->io_bitmap_ptr);
372 t->io_bitmap_ptr = NULL;
373 clear_thread_flag(TIF_IO_BITMAP);
375 * Careful, clear this in the TSS too:
377 memset(tss->io_bitmap, 0xff, t->io_bitmap_max);
378 t->io_bitmap_max = 0;
383 void flush_thread(void)
385 struct task_struct *tsk = current;
386 struct thread_info *t = current_thread_info();
388 if (t->flags & _TIF_ABI_PENDING) {
389 t->flags ^= (_TIF_ABI_PENDING | _TIF_IA32);
390 if (t->flags & _TIF_IA32)
391 current_thread_info()->status |= TS_COMPAT;
393 t->flags &= ~_TIF_DEBUG;
395 tsk->thread.debugreg0 = 0;
396 tsk->thread.debugreg1 = 0;
397 tsk->thread.debugreg2 = 0;
398 tsk->thread.debugreg3 = 0;
399 tsk->thread.debugreg6 = 0;
400 tsk->thread.debugreg7 = 0;
401 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
403 * Forget coprocessor state..
409 void release_thread(struct task_struct *dead_task)
412 if (dead_task->mm->context.size) {
413 printk("WARNING: dead process %8s still has LDT? <%p/%d>\n",
415 dead_task->mm->context.ldt,
416 dead_task->mm->context.size);
422 static inline void set_32bit_tls(struct task_struct *t, int tls, u32 addr)
424 struct user_desc ud = {
431 struct n_desc_struct *desc = (void *)t->thread.tls_array;
433 desc->a = LDT_entry_a(&ud);
434 desc->b = LDT_entry_b(&ud);
437 static inline u32 read_32bit_tls(struct task_struct *t, int tls)
439 struct desc_struct *desc = (void *)t->thread.tls_array;
442 (((u32)desc->base1) << 16) |
443 (((u32)desc->base2) << 24);
447 * This gets called before we allocate a new thread and copy
448 * the current task into it.
450 void prepare_to_copy(struct task_struct *tsk)
455 int copy_thread(int nr, unsigned long clone_flags, unsigned long rsp,
456 unsigned long unused,
457 struct task_struct * p, struct pt_regs * regs)
460 struct pt_regs * childregs;
461 struct task_struct *me = current;
463 childregs = ((struct pt_regs *)
464 (THREAD_SIZE + task_stack_page(p))) - 1;
468 childregs->rsp = rsp;
470 childregs->rsp = (unsigned long)childregs;
472 p->thread.rsp = (unsigned long) childregs;
473 p->thread.rsp0 = (unsigned long) (childregs+1);
474 p->thread.userrsp = me->thread.userrsp;
476 set_tsk_thread_flag(p, TIF_FORK);
478 p->thread.fs = me->thread.fs;
479 p->thread.gs = me->thread.gs;
481 asm("mov %%gs,%0" : "=m" (p->thread.gsindex));
482 asm("mov %%fs,%0" : "=m" (p->thread.fsindex));
483 asm("mov %%es,%0" : "=m" (p->thread.es));
484 asm("mov %%ds,%0" : "=m" (p->thread.ds));
486 if (unlikely(test_tsk_thread_flag(me, TIF_IO_BITMAP))) {
487 p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL);
488 if (!p->thread.io_bitmap_ptr) {
489 p->thread.io_bitmap_max = 0;
492 memcpy(p->thread.io_bitmap_ptr, me->thread.io_bitmap_ptr,
494 set_tsk_thread_flag(p, TIF_IO_BITMAP);
498 * Set a new TLS for the child thread?
500 if (clone_flags & CLONE_SETTLS) {
501 #ifdef CONFIG_IA32_EMULATION
502 if (test_thread_flag(TIF_IA32))
503 err = ia32_child_tls(p, childregs);
506 err = do_arch_prctl(p, ARCH_SET_FS, childregs->r8);
512 if (err && p->thread.io_bitmap_ptr) {
513 kfree(p->thread.io_bitmap_ptr);
514 p->thread.io_bitmap_max = 0;
520 * This special macro can be used to load a debugging register
522 #define loaddebug(thread,r) set_debugreg(thread->debugreg ## r, r)
524 static inline void __switch_to_xtra(struct task_struct *prev_p,
525 struct task_struct *next_p,
526 struct tss_struct *tss)
528 struct thread_struct *prev, *next;
530 prev = &prev_p->thread,
531 next = &next_p->thread;
533 if (test_tsk_thread_flag(next_p, TIF_DEBUG)) {
543 if (test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
545 * Copy the relevant range of the IO bitmap.
546 * Normally this is 128 bytes or less:
548 memcpy(tss->io_bitmap, next->io_bitmap_ptr,
549 max(prev->io_bitmap_max, next->io_bitmap_max));
550 } else if (test_tsk_thread_flag(prev_p, TIF_IO_BITMAP)) {
552 * Clear any possible leftover bits:
554 memset(tss->io_bitmap, 0xff, prev->io_bitmap_max);
559 * switch_to(x,y) should switch tasks from x to y.
561 * This could still be optimized:
562 * - fold all the options into a flag word and test it with a single test.
563 * - could test fs/gs bitsliced
565 * Kprobes not supported here. Set the probe on schedule instead.
567 __kprobes struct task_struct *
568 __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
570 struct thread_struct *prev = &prev_p->thread,
571 *next = &next_p->thread;
572 int cpu = smp_processor_id();
573 struct tss_struct *tss = &per_cpu(init_tss, cpu);
575 /* we're going to use this soon, after a few expensive things */
576 if (next_p->fpu_counter>5)
577 prefetch(&next->i387.fxsave);
580 * Reload esp0, LDT and the page table pointer:
582 tss->rsp0 = next->rsp0;
586 * This won't pick up thread selector changes, but I guess that is ok.
588 asm volatile("mov %%es,%0" : "=m" (prev->es));
589 if (unlikely(next->es | prev->es))
590 loadsegment(es, next->es);
592 asm volatile ("mov %%ds,%0" : "=m" (prev->ds));
593 if (unlikely(next->ds | prev->ds))
594 loadsegment(ds, next->ds);
603 asm volatile("movl %%fs,%0" : "=r" (fsindex));
604 /* segment register != 0 always requires a reload.
605 also reload when it has changed.
606 when prev process used 64bit base always reload
607 to avoid an information leak. */
608 if (unlikely(fsindex | next->fsindex | prev->fs)) {
609 loadsegment(fs, next->fsindex);
610 /* check if the user used a selector != 0
611 * if yes clear 64bit base, since overloaded base
612 * is always mapped to the Null selector
617 /* when next process has a 64bit base use it */
619 wrmsrl(MSR_FS_BASE, next->fs);
620 prev->fsindex = fsindex;
624 asm volatile("movl %%gs,%0" : "=r" (gsindex));
625 if (unlikely(gsindex | next->gsindex | prev->gs)) {
626 load_gs_index(next->gsindex);
631 wrmsrl(MSR_KERNEL_GS_BASE, next->gs);
632 prev->gsindex = gsindex;
635 /* Must be after DS reload */
639 * Switch the PDA and FPU contexts.
641 prev->userrsp = read_pda(oldrsp);
642 write_pda(oldrsp, next->userrsp);
643 write_pda(pcurrent, next_p);
645 write_pda(kernelstack,
646 (unsigned long)task_stack_page(next_p) + THREAD_SIZE - PDA_STACKOFFSET);
647 #ifdef CONFIG_CC_STACKPROTECTOR
648 write_pda(stack_canary, next_p->stack_canary);
650 * Build time only check to make sure the stack_canary is at
651 * offset 40 in the pda; this is a gcc ABI requirement
653 BUILD_BUG_ON(offsetof(struct x8664_pda, stack_canary) != 40);
657 * Now maybe reload the debug registers and handle I/O bitmaps
659 if (unlikely((task_thread_info(next_p)->flags & _TIF_WORK_CTXSW))
660 || test_tsk_thread_flag(prev_p, TIF_IO_BITMAP))
661 __switch_to_xtra(prev_p, next_p, tss);
663 /* If the task has used fpu the last 5 timeslices, just do a full
664 * restore of the math state immediately to avoid the trap; the
665 * chances of needing FPU soon are obviously high now
667 if (next_p->fpu_counter>5)
668 math_state_restore();
673 * sys_execve() executes a new program.
676 long sys_execve(char __user *name, char __user * __user *argv,
677 char __user * __user *envp, struct pt_regs regs)
682 filename = getname(name);
683 error = PTR_ERR(filename);
684 if (IS_ERR(filename))
686 error = do_execve(filename, argv, envp, ®s);
691 void set_personality_64bit(void)
693 /* inherit personality from parent */
695 /* Make sure to be in 64bit mode */
696 clear_thread_flag(TIF_IA32);
699 asmlinkage long sys_fork(struct pt_regs *regs)
701 return do_fork(SIGCHLD, regs->rsp, regs, 0, NULL, NULL);
705 sys_clone(unsigned long clone_flags, unsigned long newsp,
706 void __user *parent_tid, void __user *child_tid, struct pt_regs *regs)
710 return do_fork(clone_flags, newsp, regs, 0, parent_tid, child_tid);
714 * This is trivial, and on the face of it looks like it
715 * could equally well be done in user mode.
717 * Not so, for quite unobvious reasons - register pressure.
718 * In user mode vfork() cannot have a stack frame, and if
719 * done by calling the "clone()" system call directly, you
720 * do not have enough call-clobbered registers to hold all
721 * the information you need.
723 asmlinkage long sys_vfork(struct pt_regs *regs)
725 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->rsp, regs, 0,
729 unsigned long get_wchan(struct task_struct *p)
735 if (!p || p == current || p->state==TASK_RUNNING)
737 stack = (unsigned long)task_stack_page(p);
738 if (p->thread.rsp < stack || p->thread.rsp > stack+THREAD_SIZE)
740 fp = *(u64 *)(p->thread.rsp);
742 if (fp < (unsigned long)stack ||
743 fp > (unsigned long)stack+THREAD_SIZE)
745 rip = *(u64 *)(fp+8);
746 if (!in_sched_functions(rip))
749 } while (count++ < 16);
753 long do_arch_prctl(struct task_struct *task, int code, unsigned long addr)
756 int doit = task == current;
761 if (addr >= TASK_SIZE_OF(task))
764 /* handle small bases via the GDT because that's faster to
766 if (addr <= 0xffffffff) {
767 set_32bit_tls(task, GS_TLS, addr);
769 load_TLS(&task->thread, cpu);
770 load_gs_index(GS_TLS_SEL);
772 task->thread.gsindex = GS_TLS_SEL;
775 task->thread.gsindex = 0;
776 task->thread.gs = addr;
779 ret = checking_wrmsrl(MSR_KERNEL_GS_BASE, addr);
785 /* Not strictly needed for fs, but do it for symmetry
787 if (addr >= TASK_SIZE_OF(task))
790 /* handle small bases via the GDT because that's faster to
792 if (addr <= 0xffffffff) {
793 set_32bit_tls(task, FS_TLS, addr);
795 load_TLS(&task->thread, cpu);
796 asm volatile("movl %0,%%fs" :: "r"(FS_TLS_SEL));
798 task->thread.fsindex = FS_TLS_SEL;
801 task->thread.fsindex = 0;
802 task->thread.fs = addr;
804 /* set the selector to 0 to not confuse
806 asm volatile("movl %0,%%fs" :: "r" (0));
807 ret = checking_wrmsrl(MSR_FS_BASE, addr);
814 if (task->thread.fsindex == FS_TLS_SEL)
815 base = read_32bit_tls(task, FS_TLS);
817 rdmsrl(MSR_FS_BASE, base);
819 base = task->thread.fs;
820 ret = put_user(base, (unsigned long __user *)addr);
826 if (task->thread.gsindex == GS_TLS_SEL)
827 base = read_32bit_tls(task, GS_TLS);
829 asm("movl %%gs,%0" : "=r" (gsindex));
831 rdmsrl(MSR_KERNEL_GS_BASE, base);
833 base = task->thread.gs;
836 base = task->thread.gs;
837 ret = put_user(base, (unsigned long __user *)addr);
849 long sys_arch_prctl(int code, unsigned long addr)
851 return do_arch_prctl(current, code, addr);
855 * Capture the user space registers if the task is not running (in user space)
857 int dump_task_regs(struct task_struct *tsk, elf_gregset_t *regs)
859 struct pt_regs *pp, ptregs;
861 pp = task_pt_regs(tsk);
867 elf_core_copy_regs(regs, &ptregs);
872 unsigned long arch_align_stack(unsigned long sp)
874 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
875 sp -= get_random_int() % 8192;