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 = 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);
75 EXPORT_SYMBOL_GPL(idle_notifier_register);
77 void idle_notifier_unregister(struct notifier_block *n)
79 atomic_notifier_chain_unregister(&idle_notifier, n);
81 EXPORT_SYMBOL(idle_notifier_unregister);
86 atomic_notifier_call_chain(&idle_notifier, IDLE_START, NULL);
89 static void __exit_idle(void)
91 if (test_and_clear_bit_pda(0, isidle) == 0)
93 atomic_notifier_call_chain(&idle_notifier, IDLE_END, NULL);
96 /* Called from interrupts to signify idle end */
99 /* idle loop has pid 0 */
106 * We use this if we don't have any better
109 static void default_idle(void)
111 current_thread_info()->status &= ~TS_POLLING;
113 * TS_POLLING-cleared state must be visible before we
118 if (!need_resched()) {
119 /* Enables interrupts one instruction before HLT.
120 x86 special cases this so there is no race. */
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 void cpu_idle_wait(void)
140 unsigned int cpu, this_cpu = get_cpu();
141 cpumask_t map, tmp = current->cpus_allowed;
143 set_cpus_allowed(current, cpumask_of_cpu(this_cpu));
147 for_each_online_cpu(cpu) {
148 per_cpu(cpu_idle_state, cpu) = 1;
152 __get_cpu_var(cpu_idle_state) = 0;
157 for_each_online_cpu(cpu) {
158 if (cpu_isset(cpu, map) &&
159 !per_cpu(cpu_idle_state, cpu))
162 cpus_and(map, map, cpu_online_map);
163 } while (!cpus_empty(map));
165 set_cpus_allowed(current, tmp);
167 EXPORT_SYMBOL_GPL(cpu_idle_wait);
169 #ifdef CONFIG_HOTPLUG_CPU
170 DECLARE_PER_CPU(int, cpu_state);
173 /* We halt the CPU with physical CPU hotplug */
174 static inline void play_dead(void)
180 __get_cpu_var(cpu_state) = CPU_DEAD;
187 static inline void play_dead(void)
191 #endif /* CONFIG_HOTPLUG_CPU */
194 * The idle thread. There's no useful work to be
195 * done, so just try to conserve power and have a
196 * low exit latency (ie sit in a loop waiting for
197 * somebody to say that they'd like to reschedule)
201 current_thread_info()->status |= TS_POLLING;
202 /* endless idle loop with no priority at all */
204 while (!need_resched()) {
207 if (__get_cpu_var(cpu_idle_state))
208 __get_cpu_var(cpu_idle_state) = 0;
214 if (cpu_is_offline(smp_processor_id()))
217 * Idle routines should keep interrupts disabled
218 * from here on, until they go to idle.
219 * Otherwise, idle callbacks can misfire.
224 /* In many cases the interrupt that ended idle
225 has already called exit_idle. But some idle
226 loops can be woken up without interrupt. */
230 preempt_enable_no_resched();
237 * This uses new MONITOR/MWAIT instructions on P4 processors with PNI,
238 * which can obviate IPI to trigger checking of need_resched.
239 * We execute MONITOR against need_resched and enter optimized wait state
240 * through MWAIT. Whenever someone changes need_resched, we would be woken
241 * up from MWAIT (without an IPI).
243 * New with Core Duo processors, MWAIT can take some hints based on CPU
246 void mwait_idle_with_hints(unsigned long eax, unsigned long ecx)
248 if (!need_resched()) {
249 __monitor((void *)¤t_thread_info()->flags, 0, 0);
256 /* Default MONITOR/MWAIT with no hints, used for default C1 state */
257 static void mwait_idle(void)
259 if (!need_resched()) {
260 __monitor((void *)¤t_thread_info()->flags, 0, 0);
271 void __cpuinit select_idle_routine(const struct cpuinfo_x86 *c)
274 if (cpu_has(c, X86_FEATURE_MWAIT)) {
276 * Skip, if setup has overridden idle.
277 * One CPU supports mwait => All CPUs supports mwait
281 printk("using mwait in idle threads.\n");
284 pm_idle = mwait_idle;
289 static int __init idle_setup (char *str)
291 if (!strncmp(str, "poll", 4)) {
292 printk("using polling idle threads.\n");
296 boot_option_idle_override = 1;
300 __setup("idle=", idle_setup);
302 /* Prints also some state that isn't saved in the pt_regs */
303 void __show_regs(struct pt_regs * regs)
305 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L, fs, gs, shadowgs;
306 unsigned int fsindex,gsindex;
307 unsigned int ds,cs,es;
311 printk("Pid: %d, comm: %.20s %s %s %.*s\n",
312 current->pid, current->comm, print_tainted(),
313 init_utsname()->release,
314 (int)strcspn(init_utsname()->version, " "),
315 init_utsname()->version);
316 printk("RIP: %04lx:[<%016lx>] ", regs->cs & 0xffff, regs->rip);
317 printk_address(regs->rip);
318 printk("RSP: %04lx:%016lx EFLAGS: %08lx\n", regs->ss, regs->rsp,
320 printk("RAX: %016lx RBX: %016lx RCX: %016lx\n",
321 regs->rax, regs->rbx, regs->rcx);
322 printk("RDX: %016lx RSI: %016lx RDI: %016lx\n",
323 regs->rdx, regs->rsi, regs->rdi);
324 printk("RBP: %016lx R08: %016lx R09: %016lx\n",
325 regs->rbp, regs->r8, regs->r9);
326 printk("R10: %016lx R11: %016lx R12: %016lx\n",
327 regs->r10, regs->r11, regs->r12);
328 printk("R13: %016lx R14: %016lx R15: %016lx\n",
329 regs->r13, regs->r14, regs->r15);
331 asm("movl %%ds,%0" : "=r" (ds));
332 asm("movl %%cs,%0" : "=r" (cs));
333 asm("movl %%es,%0" : "=r" (es));
334 asm("movl %%fs,%0" : "=r" (fsindex));
335 asm("movl %%gs,%0" : "=r" (gsindex));
337 rdmsrl(MSR_FS_BASE, fs);
338 rdmsrl(MSR_GS_BASE, gs);
339 rdmsrl(MSR_KERNEL_GS_BASE, shadowgs);
341 asm("movq %%cr0, %0": "=r" (cr0));
342 asm("movq %%cr2, %0": "=r" (cr2));
343 asm("movq %%cr3, %0": "=r" (cr3));
344 asm("movq %%cr4, %0": "=r" (cr4));
346 printk("FS: %016lx(%04x) GS:%016lx(%04x) knlGS:%016lx\n",
347 fs,fsindex,gs,gsindex,shadowgs);
348 printk("CS: %04x DS: %04x ES: %04x CR0: %016lx\n", cs, ds, es, cr0);
349 printk("CR2: %016lx CR3: %016lx CR4: %016lx\n", cr2, cr3, cr4);
352 void show_regs(struct pt_regs *regs)
354 printk("CPU %d:", smp_processor_id());
356 show_trace(NULL, regs, (void *)(regs + 1));
360 * Free current thread data structures etc..
362 void exit_thread(void)
364 struct task_struct *me = current;
365 struct thread_struct *t = &me->thread;
367 if (me->thread.io_bitmap_ptr) {
368 struct tss_struct *tss = &per_cpu(init_tss, get_cpu());
370 kfree(t->io_bitmap_ptr);
371 t->io_bitmap_ptr = NULL;
372 clear_thread_flag(TIF_IO_BITMAP);
374 * Careful, clear this in the TSS too:
376 memset(tss->io_bitmap, 0xff, t->io_bitmap_max);
377 t->io_bitmap_max = 0;
382 void flush_thread(void)
384 struct task_struct *tsk = current;
386 if (test_tsk_thread_flag(tsk, TIF_ABI_PENDING)) {
387 clear_tsk_thread_flag(tsk, TIF_ABI_PENDING);
388 if (test_tsk_thread_flag(tsk, TIF_IA32)) {
389 clear_tsk_thread_flag(tsk, TIF_IA32);
391 set_tsk_thread_flag(tsk, TIF_IA32);
392 current_thread_info()->status |= TS_COMPAT;
395 clear_tsk_thread_flag(tsk, TIF_DEBUG);
397 tsk->thread.debugreg0 = 0;
398 tsk->thread.debugreg1 = 0;
399 tsk->thread.debugreg2 = 0;
400 tsk->thread.debugreg3 = 0;
401 tsk->thread.debugreg6 = 0;
402 tsk->thread.debugreg7 = 0;
403 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
405 * Forget coprocessor state..
411 void release_thread(struct task_struct *dead_task)
414 if (dead_task->mm->context.size) {
415 printk("WARNING: dead process %8s still has LDT? <%p/%d>\n",
417 dead_task->mm->context.ldt,
418 dead_task->mm->context.size);
424 static inline void set_32bit_tls(struct task_struct *t, int tls, u32 addr)
426 struct user_desc ud = {
433 struct n_desc_struct *desc = (void *)t->thread.tls_array;
435 desc->a = LDT_entry_a(&ud);
436 desc->b = LDT_entry_b(&ud);
439 static inline u32 read_32bit_tls(struct task_struct *t, int tls)
441 struct desc_struct *desc = (void *)t->thread.tls_array;
444 (((u32)desc->base1) << 16) |
445 (((u32)desc->base2) << 24);
449 * This gets called before we allocate a new thread and copy
450 * the current task into it.
452 void prepare_to_copy(struct task_struct *tsk)
457 int copy_thread(int nr, unsigned long clone_flags, unsigned long rsp,
458 unsigned long unused,
459 struct task_struct * p, struct pt_regs * regs)
462 struct pt_regs * childregs;
463 struct task_struct *me = current;
465 childregs = ((struct pt_regs *)
466 (THREAD_SIZE + task_stack_page(p))) - 1;
470 childregs->rsp = rsp;
472 childregs->rsp = (unsigned long)childregs;
474 p->thread.rsp = (unsigned long) childregs;
475 p->thread.rsp0 = (unsigned long) (childregs+1);
476 p->thread.userrsp = me->thread.userrsp;
478 set_tsk_thread_flag(p, TIF_FORK);
480 p->thread.fs = me->thread.fs;
481 p->thread.gs = me->thread.gs;
483 asm("mov %%gs,%0" : "=m" (p->thread.gsindex));
484 asm("mov %%fs,%0" : "=m" (p->thread.fsindex));
485 asm("mov %%es,%0" : "=m" (p->thread.es));
486 asm("mov %%ds,%0" : "=m" (p->thread.ds));
488 if (unlikely(test_tsk_thread_flag(me, TIF_IO_BITMAP))) {
489 p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL);
490 if (!p->thread.io_bitmap_ptr) {
491 p->thread.io_bitmap_max = 0;
494 memcpy(p->thread.io_bitmap_ptr, me->thread.io_bitmap_ptr,
496 set_tsk_thread_flag(p, TIF_IO_BITMAP);
500 * Set a new TLS for the child thread?
502 if (clone_flags & CLONE_SETTLS) {
503 #ifdef CONFIG_IA32_EMULATION
504 if (test_thread_flag(TIF_IA32))
505 err = ia32_child_tls(p, childregs);
508 err = do_arch_prctl(p, ARCH_SET_FS, childregs->r8);
514 if (err && p->thread.io_bitmap_ptr) {
515 kfree(p->thread.io_bitmap_ptr);
516 p->thread.io_bitmap_max = 0;
522 * This special macro can be used to load a debugging register
524 #define loaddebug(thread,r) set_debugreg(thread->debugreg ## r, r)
526 static inline void __switch_to_xtra(struct task_struct *prev_p,
527 struct task_struct *next_p,
528 struct tss_struct *tss)
530 struct thread_struct *prev, *next;
532 prev = &prev_p->thread,
533 next = &next_p->thread;
535 if (test_tsk_thread_flag(next_p, TIF_DEBUG)) {
545 if (test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
547 * Copy the relevant range of the IO bitmap.
548 * Normally this is 128 bytes or less:
550 memcpy(tss->io_bitmap, next->io_bitmap_ptr,
551 max(prev->io_bitmap_max, next->io_bitmap_max));
552 } else if (test_tsk_thread_flag(prev_p, TIF_IO_BITMAP)) {
554 * Clear any possible leftover bits:
556 memset(tss->io_bitmap, 0xff, prev->io_bitmap_max);
561 * switch_to(x,y) should switch tasks from x to y.
563 * This could still be optimized:
564 * - fold all the options into a flag word and test it with a single test.
565 * - could test fs/gs bitsliced
567 * Kprobes not supported here. Set the probe on schedule instead.
569 __kprobes struct task_struct *
570 __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
572 struct thread_struct *prev = &prev_p->thread,
573 *next = &next_p->thread;
574 int cpu = smp_processor_id();
575 struct tss_struct *tss = &per_cpu(init_tss, cpu);
577 /* we're going to use this soon, after a few expensive things */
578 if (next_p->fpu_counter>5)
579 prefetch(&next->i387.fxsave);
582 * Reload esp0, LDT and the page table pointer:
584 tss->rsp0 = next->rsp0;
588 * This won't pick up thread selector changes, but I guess that is ok.
590 asm volatile("mov %%es,%0" : "=m" (prev->es));
591 if (unlikely(next->es | prev->es))
592 loadsegment(es, next->es);
594 asm volatile ("mov %%ds,%0" : "=m" (prev->ds));
595 if (unlikely(next->ds | prev->ds))
596 loadsegment(ds, next->ds);
605 asm volatile("movl %%fs,%0" : "=r" (fsindex));
606 /* segment register != 0 always requires a reload.
607 also reload when it has changed.
608 when prev process used 64bit base always reload
609 to avoid an information leak. */
610 if (unlikely(fsindex | next->fsindex | prev->fs)) {
611 loadsegment(fs, next->fsindex);
612 /* check if the user used a selector != 0
613 * if yes clear 64bit base, since overloaded base
614 * is always mapped to the Null selector
619 /* when next process has a 64bit base use it */
621 wrmsrl(MSR_FS_BASE, next->fs);
622 prev->fsindex = fsindex;
626 asm volatile("movl %%gs,%0" : "=r" (gsindex));
627 if (unlikely(gsindex | next->gsindex | prev->gs)) {
628 load_gs_index(next->gsindex);
633 wrmsrl(MSR_KERNEL_GS_BASE, next->gs);
634 prev->gsindex = gsindex;
637 /* Must be after DS reload */
641 * Switch the PDA and FPU contexts.
643 prev->userrsp = read_pda(oldrsp);
644 write_pda(oldrsp, next->userrsp);
645 write_pda(pcurrent, next_p);
647 write_pda(kernelstack,
648 (unsigned long)task_stack_page(next_p) + THREAD_SIZE - PDA_STACKOFFSET);
649 #ifdef CONFIG_CC_STACKPROTECTOR
650 write_pda(stack_canary, next_p->stack_canary);
652 * Build time only check to make sure the stack_canary is at
653 * offset 40 in the pda; this is a gcc ABI requirement
655 BUILD_BUG_ON(offsetof(struct x8664_pda, stack_canary) != 40);
659 * Now maybe reload the debug registers and handle I/O bitmaps
661 if (unlikely((task_thread_info(next_p)->flags & _TIF_WORK_CTXSW))
662 || test_tsk_thread_flag(prev_p, TIF_IO_BITMAP))
663 __switch_to_xtra(prev_p, next_p, tss);
665 /* If the task has used fpu the last 5 timeslices, just do a full
666 * restore of the math state immediately to avoid the trap; the
667 * chances of needing FPU soon are obviously high now
669 if (next_p->fpu_counter>5)
670 math_state_restore();
675 * sys_execve() executes a new program.
678 long sys_execve(char __user *name, char __user * __user *argv,
679 char __user * __user *envp, struct pt_regs regs)
684 filename = getname(name);
685 error = PTR_ERR(filename);
686 if (IS_ERR(filename))
688 error = do_execve(filename, argv, envp, ®s);
691 current->ptrace &= ~PT_DTRACE;
692 task_unlock(current);
698 void set_personality_64bit(void)
700 /* inherit personality from parent */
702 /* Make sure to be in 64bit mode */
703 clear_thread_flag(TIF_IA32);
705 /* TBD: overwrites user setup. Should have two bits.
706 But 64bit processes have always behaved this way,
707 so it's not too bad. The main problem is just that
708 32bit childs are affected again. */
709 current->personality &= ~READ_IMPLIES_EXEC;
712 asmlinkage long sys_fork(struct pt_regs *regs)
714 return do_fork(SIGCHLD, regs->rsp, regs, 0, NULL, NULL);
718 sys_clone(unsigned long clone_flags, unsigned long newsp,
719 void __user *parent_tid, void __user *child_tid, struct pt_regs *regs)
723 return do_fork(clone_flags, newsp, regs, 0, parent_tid, child_tid);
727 * This is trivial, and on the face of it looks like it
728 * could equally well be done in user mode.
730 * Not so, for quite unobvious reasons - register pressure.
731 * In user mode vfork() cannot have a stack frame, and if
732 * done by calling the "clone()" system call directly, you
733 * do not have enough call-clobbered registers to hold all
734 * the information you need.
736 asmlinkage long sys_vfork(struct pt_regs *regs)
738 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->rsp, regs, 0,
742 unsigned long get_wchan(struct task_struct *p)
748 if (!p || p == current || p->state==TASK_RUNNING)
750 stack = (unsigned long)task_stack_page(p);
751 if (p->thread.rsp < stack || p->thread.rsp > stack+THREAD_SIZE)
753 fp = *(u64 *)(p->thread.rsp);
755 if (fp < (unsigned long)stack ||
756 fp > (unsigned long)stack+THREAD_SIZE)
758 rip = *(u64 *)(fp+8);
759 if (!in_sched_functions(rip))
762 } while (count++ < 16);
766 long do_arch_prctl(struct task_struct *task, int code, unsigned long addr)
769 int doit = task == current;
774 if (addr >= TASK_SIZE_OF(task))
777 /* handle small bases via the GDT because that's faster to
779 if (addr <= 0xffffffff) {
780 set_32bit_tls(task, GS_TLS, addr);
782 load_TLS(&task->thread, cpu);
783 load_gs_index(GS_TLS_SEL);
785 task->thread.gsindex = GS_TLS_SEL;
788 task->thread.gsindex = 0;
789 task->thread.gs = addr;
792 ret = checking_wrmsrl(MSR_KERNEL_GS_BASE, addr);
798 /* Not strictly needed for fs, but do it for symmetry
800 if (addr >= TASK_SIZE_OF(task))
803 /* handle small bases via the GDT because that's faster to
805 if (addr <= 0xffffffff) {
806 set_32bit_tls(task, FS_TLS, addr);
808 load_TLS(&task->thread, cpu);
809 asm volatile("movl %0,%%fs" :: "r"(FS_TLS_SEL));
811 task->thread.fsindex = FS_TLS_SEL;
814 task->thread.fsindex = 0;
815 task->thread.fs = addr;
817 /* set the selector to 0 to not confuse
819 asm volatile("movl %0,%%fs" :: "r" (0));
820 ret = checking_wrmsrl(MSR_FS_BASE, addr);
827 if (task->thread.fsindex == FS_TLS_SEL)
828 base = read_32bit_tls(task, FS_TLS);
830 rdmsrl(MSR_FS_BASE, base);
832 base = task->thread.fs;
833 ret = put_user(base, (unsigned long __user *)addr);
839 if (task->thread.gsindex == GS_TLS_SEL)
840 base = read_32bit_tls(task, GS_TLS);
842 asm("movl %%gs,%0" : "=r" (gsindex));
844 rdmsrl(MSR_KERNEL_GS_BASE, base);
846 base = task->thread.gs;
849 base = task->thread.gs;
850 ret = put_user(base, (unsigned long __user *)addr);
862 long sys_arch_prctl(int code, unsigned long addr)
864 return do_arch_prctl(current, code, addr);
868 * Capture the user space registers if the task is not running (in user space)
870 int dump_task_regs(struct task_struct *tsk, elf_gregset_t *regs)
872 struct pt_regs *pp, ptregs;
874 pp = task_pt_regs(tsk);
880 elf_core_copy_regs(regs, &ptregs);
885 unsigned long arch_align_stack(unsigned long sp)
887 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
888 sp -= get_random_int() % 8192;