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;
112 smp_mb__after_clear_bit();
114 if (!need_resched()) {
115 /* Enables interrupts one instruction before HLT.
116 x86 special cases this so there is no race. */
120 current_thread_info()->status |= TS_POLLING;
124 * On SMP it's slightly faster (but much more power-consuming!)
125 * to poll the ->need_resched flag instead of waiting for the
126 * cross-CPU IPI to arrive. Use this option with caution.
128 static void poll_idle (void)
134 void cpu_idle_wait(void)
136 unsigned int cpu, this_cpu = get_cpu();
137 cpumask_t map, tmp = current->cpus_allowed;
139 set_cpus_allowed(current, cpumask_of_cpu(this_cpu));
143 for_each_online_cpu(cpu) {
144 per_cpu(cpu_idle_state, cpu) = 1;
148 __get_cpu_var(cpu_idle_state) = 0;
153 for_each_online_cpu(cpu) {
154 if (cpu_isset(cpu, map) &&
155 !per_cpu(cpu_idle_state, cpu))
158 cpus_and(map, map, cpu_online_map);
159 } while (!cpus_empty(map));
161 set_cpus_allowed(current, tmp);
163 EXPORT_SYMBOL_GPL(cpu_idle_wait);
165 #ifdef CONFIG_HOTPLUG_CPU
166 DECLARE_PER_CPU(int, cpu_state);
169 /* We halt the CPU with physical CPU hotplug */
170 static inline void play_dead(void)
176 __get_cpu_var(cpu_state) = CPU_DEAD;
183 static inline void play_dead(void)
187 #endif /* CONFIG_HOTPLUG_CPU */
190 * The idle thread. There's no useful work to be
191 * done, so just try to conserve power and have a
192 * low exit latency (ie sit in a loop waiting for
193 * somebody to say that they'd like to reschedule)
197 current_thread_info()->status |= TS_POLLING;
198 /* endless idle loop with no priority at all */
200 while (!need_resched()) {
203 if (__get_cpu_var(cpu_idle_state))
204 __get_cpu_var(cpu_idle_state) = 0;
210 if (cpu_is_offline(smp_processor_id()))
213 * Idle routines should keep interrupts disabled
214 * from here on, until they go to idle.
215 * Otherwise, idle callbacks can misfire.
220 /* In many cases the interrupt that ended idle
221 has already called exit_idle. But some idle
222 loops can be woken up without interrupt. */
226 preempt_enable_no_resched();
233 * This uses new MONITOR/MWAIT instructions on P4 processors with PNI,
234 * which can obviate IPI to trigger checking of need_resched.
235 * We execute MONITOR against need_resched and enter optimized wait state
236 * through MWAIT. Whenever someone changes need_resched, we would be woken
237 * up from MWAIT (without an IPI).
239 * New with Core Duo processors, MWAIT can take some hints based on CPU
242 void mwait_idle_with_hints(unsigned long eax, unsigned long ecx)
244 if (!need_resched()) {
245 __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)
255 if (!need_resched()) {
256 __monitor((void *)¤t_thread_info()->flags, 0, 0);
267 void __cpuinit select_idle_routine(const struct cpuinfo_x86 *c)
270 if (cpu_has(c, X86_FEATURE_MWAIT)) {
272 * Skip, if setup has overridden idle.
273 * One CPU supports mwait => All CPUs supports mwait
277 printk("using mwait in idle threads.\n");
280 pm_idle = mwait_idle;
285 static int __init idle_setup (char *str)
287 if (!strncmp(str, "poll", 4)) {
288 printk("using polling idle threads.\n");
292 boot_option_idle_override = 1;
296 __setup("idle=", idle_setup);
298 /* Prints also some state that isn't saved in the pt_regs */
299 void __show_regs(struct pt_regs * regs)
301 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L, fs, gs, shadowgs;
302 unsigned int fsindex,gsindex;
303 unsigned int ds,cs,es;
307 printk("Pid: %d, comm: %.20s %s %s %.*s\n",
308 current->pid, current->comm, print_tainted(),
309 init_utsname()->release,
310 (int)strcspn(init_utsname()->version, " "),
311 init_utsname()->version);
312 printk("RIP: %04lx:[<%016lx>] ", regs->cs & 0xffff, regs->rip);
313 printk_address(regs->rip);
314 printk("RSP: %04lx:%016lx EFLAGS: %08lx\n", regs->ss, regs->rsp,
316 printk("RAX: %016lx RBX: %016lx RCX: %016lx\n",
317 regs->rax, regs->rbx, regs->rcx);
318 printk("RDX: %016lx RSI: %016lx RDI: %016lx\n",
319 regs->rdx, regs->rsi, regs->rdi);
320 printk("RBP: %016lx R08: %016lx R09: %016lx\n",
321 regs->rbp, regs->r8, regs->r9);
322 printk("R10: %016lx R11: %016lx R12: %016lx\n",
323 regs->r10, regs->r11, regs->r12);
324 printk("R13: %016lx R14: %016lx R15: %016lx\n",
325 regs->r13, regs->r14, regs->r15);
327 asm("movl %%ds,%0" : "=r" (ds));
328 asm("movl %%cs,%0" : "=r" (cs));
329 asm("movl %%es,%0" : "=r" (es));
330 asm("movl %%fs,%0" : "=r" (fsindex));
331 asm("movl %%gs,%0" : "=r" (gsindex));
333 rdmsrl(MSR_FS_BASE, fs);
334 rdmsrl(MSR_GS_BASE, gs);
335 rdmsrl(MSR_KERNEL_GS_BASE, shadowgs);
337 asm("movq %%cr0, %0": "=r" (cr0));
338 asm("movq %%cr2, %0": "=r" (cr2));
339 asm("movq %%cr3, %0": "=r" (cr3));
340 asm("movq %%cr4, %0": "=r" (cr4));
342 printk("FS: %016lx(%04x) GS:%016lx(%04x) knlGS:%016lx\n",
343 fs,fsindex,gs,gsindex,shadowgs);
344 printk("CS: %04x DS: %04x ES: %04x CR0: %016lx\n", cs, ds, es, cr0);
345 printk("CR2: %016lx CR3: %016lx CR4: %016lx\n", cr2, cr3, cr4);
348 void show_regs(struct pt_regs *regs)
350 printk("CPU %d:", smp_processor_id());
352 show_trace(NULL, regs, (void *)(regs + 1));
356 * Free current thread data structures etc..
358 void exit_thread(void)
360 struct task_struct *me = current;
361 struct thread_struct *t = &me->thread;
363 if (me->thread.io_bitmap_ptr) {
364 struct tss_struct *tss = &per_cpu(init_tss, get_cpu());
366 kfree(t->io_bitmap_ptr);
367 t->io_bitmap_ptr = NULL;
368 clear_thread_flag(TIF_IO_BITMAP);
370 * Careful, clear this in the TSS too:
372 memset(tss->io_bitmap, 0xff, t->io_bitmap_max);
373 t->io_bitmap_max = 0;
378 void flush_thread(void)
380 struct task_struct *tsk = current;
381 struct thread_info *t = current_thread_info();
383 if (t->flags & _TIF_ABI_PENDING) {
384 t->flags ^= (_TIF_ABI_PENDING | _TIF_IA32);
385 if (t->flags & _TIF_IA32)
386 current_thread_info()->status |= TS_COMPAT;
388 t->flags &= ~_TIF_DEBUG;
390 tsk->thread.debugreg0 = 0;
391 tsk->thread.debugreg1 = 0;
392 tsk->thread.debugreg2 = 0;
393 tsk->thread.debugreg3 = 0;
394 tsk->thread.debugreg6 = 0;
395 tsk->thread.debugreg7 = 0;
396 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
398 * Forget coprocessor state..
404 void release_thread(struct task_struct *dead_task)
407 if (dead_task->mm->context.size) {
408 printk("WARNING: dead process %8s still has LDT? <%p/%d>\n",
410 dead_task->mm->context.ldt,
411 dead_task->mm->context.size);
417 static inline void set_32bit_tls(struct task_struct *t, int tls, u32 addr)
419 struct user_desc ud = {
426 struct n_desc_struct *desc = (void *)t->thread.tls_array;
428 desc->a = LDT_entry_a(&ud);
429 desc->b = LDT_entry_b(&ud);
432 static inline u32 read_32bit_tls(struct task_struct *t, int tls)
434 struct desc_struct *desc = (void *)t->thread.tls_array;
437 (((u32)desc->base1) << 16) |
438 (((u32)desc->base2) << 24);
442 * This gets called before we allocate a new thread and copy
443 * the current task into it.
445 void prepare_to_copy(struct task_struct *tsk)
450 int copy_thread(int nr, unsigned long clone_flags, unsigned long rsp,
451 unsigned long unused,
452 struct task_struct * p, struct pt_regs * regs)
455 struct pt_regs * childregs;
456 struct task_struct *me = current;
458 childregs = ((struct pt_regs *)
459 (THREAD_SIZE + task_stack_page(p))) - 1;
463 childregs->rsp = rsp;
465 childregs->rsp = (unsigned long)childregs;
467 p->thread.rsp = (unsigned long) childregs;
468 p->thread.rsp0 = (unsigned long) (childregs+1);
469 p->thread.userrsp = me->thread.userrsp;
471 set_tsk_thread_flag(p, TIF_FORK);
473 p->thread.fs = me->thread.fs;
474 p->thread.gs = me->thread.gs;
476 asm("mov %%gs,%0" : "=m" (p->thread.gsindex));
477 asm("mov %%fs,%0" : "=m" (p->thread.fsindex));
478 asm("mov %%es,%0" : "=m" (p->thread.es));
479 asm("mov %%ds,%0" : "=m" (p->thread.ds));
481 if (unlikely(test_tsk_thread_flag(me, TIF_IO_BITMAP))) {
482 p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL);
483 if (!p->thread.io_bitmap_ptr) {
484 p->thread.io_bitmap_max = 0;
487 memcpy(p->thread.io_bitmap_ptr, me->thread.io_bitmap_ptr,
489 set_tsk_thread_flag(p, TIF_IO_BITMAP);
493 * Set a new TLS for the child thread?
495 if (clone_flags & CLONE_SETTLS) {
496 #ifdef CONFIG_IA32_EMULATION
497 if (test_thread_flag(TIF_IA32))
498 err = ia32_child_tls(p, childregs);
501 err = do_arch_prctl(p, ARCH_SET_FS, childregs->r8);
507 if (err && p->thread.io_bitmap_ptr) {
508 kfree(p->thread.io_bitmap_ptr);
509 p->thread.io_bitmap_max = 0;
515 * This special macro can be used to load a debugging register
517 #define loaddebug(thread,r) set_debugreg(thread->debugreg ## r, r)
519 static inline void __switch_to_xtra(struct task_struct *prev_p,
520 struct task_struct *next_p,
521 struct tss_struct *tss)
523 struct thread_struct *prev, *next;
525 prev = &prev_p->thread,
526 next = &next_p->thread;
528 if (test_tsk_thread_flag(next_p, TIF_DEBUG)) {
538 if (test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
540 * Copy the relevant range of the IO bitmap.
541 * Normally this is 128 bytes or less:
543 memcpy(tss->io_bitmap, next->io_bitmap_ptr,
544 max(prev->io_bitmap_max, next->io_bitmap_max));
545 } else if (test_tsk_thread_flag(prev_p, TIF_IO_BITMAP)) {
547 * Clear any possible leftover bits:
549 memset(tss->io_bitmap, 0xff, prev->io_bitmap_max);
554 * switch_to(x,y) should switch tasks from x to y.
556 * This could still be optimized:
557 * - fold all the options into a flag word and test it with a single test.
558 * - could test fs/gs bitsliced
560 * Kprobes not supported here. Set the probe on schedule instead.
562 __kprobes struct task_struct *
563 __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
565 struct thread_struct *prev = &prev_p->thread,
566 *next = &next_p->thread;
567 int cpu = smp_processor_id();
568 struct tss_struct *tss = &per_cpu(init_tss, cpu);
570 /* we're going to use this soon, after a few expensive things */
571 if (next_p->fpu_counter>5)
572 prefetch(&next->i387.fxsave);
575 * Reload esp0, LDT and the page table pointer:
577 tss->rsp0 = next->rsp0;
581 * This won't pick up thread selector changes, but I guess that is ok.
583 asm volatile("mov %%es,%0" : "=m" (prev->es));
584 if (unlikely(next->es | prev->es))
585 loadsegment(es, next->es);
587 asm volatile ("mov %%ds,%0" : "=m" (prev->ds));
588 if (unlikely(next->ds | prev->ds))
589 loadsegment(ds, next->ds);
598 asm volatile("movl %%fs,%0" : "=r" (fsindex));
599 /* segment register != 0 always requires a reload.
600 also reload when it has changed.
601 when prev process used 64bit base always reload
602 to avoid an information leak. */
603 if (unlikely(fsindex | next->fsindex | prev->fs)) {
604 loadsegment(fs, next->fsindex);
605 /* check if the user used a selector != 0
606 * if yes clear 64bit base, since overloaded base
607 * is always mapped to the Null selector
612 /* when next process has a 64bit base use it */
614 wrmsrl(MSR_FS_BASE, next->fs);
615 prev->fsindex = fsindex;
619 asm volatile("movl %%gs,%0" : "=r" (gsindex));
620 if (unlikely(gsindex | next->gsindex | prev->gs)) {
621 load_gs_index(next->gsindex);
626 wrmsrl(MSR_KERNEL_GS_BASE, next->gs);
627 prev->gsindex = gsindex;
630 /* Must be after DS reload */
634 * Switch the PDA and FPU contexts.
636 prev->userrsp = read_pda(oldrsp);
637 write_pda(oldrsp, next->userrsp);
638 write_pda(pcurrent, next_p);
640 write_pda(kernelstack,
641 (unsigned long)task_stack_page(next_p) + THREAD_SIZE - PDA_STACKOFFSET);
642 #ifdef CONFIG_CC_STACKPROTECTOR
643 write_pda(stack_canary, next_p->stack_canary);
645 * Build time only check to make sure the stack_canary is at
646 * offset 40 in the pda; this is a gcc ABI requirement
648 BUILD_BUG_ON(offsetof(struct x8664_pda, stack_canary) != 40);
652 * Now maybe reload the debug registers and handle I/O bitmaps
654 if (unlikely((task_thread_info(next_p)->flags & _TIF_WORK_CTXSW))
655 || test_tsk_thread_flag(prev_p, TIF_IO_BITMAP))
656 __switch_to_xtra(prev_p, next_p, tss);
658 /* If the task has used fpu the last 5 timeslices, just do a full
659 * restore of the math state immediately to avoid the trap; the
660 * chances of needing FPU soon are obviously high now
662 if (next_p->fpu_counter>5)
663 math_state_restore();
668 * sys_execve() executes a new program.
671 long sys_execve(char __user *name, char __user * __user *argv,
672 char __user * __user *envp, struct pt_regs regs)
677 filename = getname(name);
678 error = PTR_ERR(filename);
679 if (IS_ERR(filename))
681 error = do_execve(filename, argv, envp, ®s);
684 current->ptrace &= ~PT_DTRACE;
685 task_unlock(current);
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);
698 /* TBD: overwrites user setup. Should have two bits.
699 But 64bit processes have always behaved this way,
700 so it's not too bad. The main problem is just that
701 32bit childs are affected again. */
702 current->personality &= ~READ_IMPLIES_EXEC;
705 asmlinkage long sys_fork(struct pt_regs *regs)
707 return do_fork(SIGCHLD, regs->rsp, regs, 0, NULL, NULL);
711 sys_clone(unsigned long clone_flags, unsigned long newsp,
712 void __user *parent_tid, void __user *child_tid, struct pt_regs *regs)
716 return do_fork(clone_flags, newsp, regs, 0, parent_tid, child_tid);
720 * This is trivial, and on the face of it looks like it
721 * could equally well be done in user mode.
723 * Not so, for quite unobvious reasons - register pressure.
724 * In user mode vfork() cannot have a stack frame, and if
725 * done by calling the "clone()" system call directly, you
726 * do not have enough call-clobbered registers to hold all
727 * the information you need.
729 asmlinkage long sys_vfork(struct pt_regs *regs)
731 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->rsp, regs, 0,
735 unsigned long get_wchan(struct task_struct *p)
741 if (!p || p == current || p->state==TASK_RUNNING)
743 stack = (unsigned long)task_stack_page(p);
744 if (p->thread.rsp < stack || p->thread.rsp > stack+THREAD_SIZE)
746 fp = *(u64 *)(p->thread.rsp);
748 if (fp < (unsigned long)stack ||
749 fp > (unsigned long)stack+THREAD_SIZE)
751 rip = *(u64 *)(fp+8);
752 if (!in_sched_functions(rip))
755 } while (count++ < 16);
759 long do_arch_prctl(struct task_struct *task, int code, unsigned long addr)
762 int doit = task == current;
767 if (addr >= TASK_SIZE_OF(task))
770 /* handle small bases via the GDT because that's faster to
772 if (addr <= 0xffffffff) {
773 set_32bit_tls(task, GS_TLS, addr);
775 load_TLS(&task->thread, cpu);
776 load_gs_index(GS_TLS_SEL);
778 task->thread.gsindex = GS_TLS_SEL;
781 task->thread.gsindex = 0;
782 task->thread.gs = addr;
785 ret = checking_wrmsrl(MSR_KERNEL_GS_BASE, addr);
791 /* Not strictly needed for fs, but do it for symmetry
793 if (addr >= TASK_SIZE_OF(task))
796 /* handle small bases via the GDT because that's faster to
798 if (addr <= 0xffffffff) {
799 set_32bit_tls(task, FS_TLS, addr);
801 load_TLS(&task->thread, cpu);
802 asm volatile("movl %0,%%fs" :: "r"(FS_TLS_SEL));
804 task->thread.fsindex = FS_TLS_SEL;
807 task->thread.fsindex = 0;
808 task->thread.fs = addr;
810 /* set the selector to 0 to not confuse
812 asm volatile("movl %0,%%fs" :: "r" (0));
813 ret = checking_wrmsrl(MSR_FS_BASE, addr);
820 if (task->thread.fsindex == FS_TLS_SEL)
821 base = read_32bit_tls(task, FS_TLS);
823 rdmsrl(MSR_FS_BASE, base);
825 base = task->thread.fs;
826 ret = put_user(base, (unsigned long __user *)addr);
832 if (task->thread.gsindex == GS_TLS_SEL)
833 base = read_32bit_tls(task, GS_TLS);
835 asm("movl %%gs,%0" : "=r" (gsindex));
837 rdmsrl(MSR_KERNEL_GS_BASE, base);
839 base = task->thread.gs;
842 base = task->thread.gs;
843 ret = put_user(base, (unsigned long __user *)addr);
855 long sys_arch_prctl(int code, unsigned long addr)
857 return do_arch_prctl(current, code, addr);
861 * Capture the user space registers if the task is not running (in user space)
863 int dump_task_regs(struct task_struct *tsk, elf_gregset_t *regs)
865 struct pt_regs *pp, ptregs;
867 pp = task_pt_regs(tsk);
873 elf_core_copy_regs(regs, &ptregs);
878 unsigned long arch_align_stack(unsigned long sp)
880 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
881 sp -= get_random_int() % 8192;