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);
83 enum idle_state { CPU_IDLE, CPU_NOT_IDLE };
84 static DEFINE_PER_CPU(enum idle_state, idle_state) = CPU_NOT_IDLE;
88 __get_cpu_var(idle_state) = CPU_IDLE;
89 atomic_notifier_call_chain(&idle_notifier, IDLE_START, NULL);
92 static void __exit_idle(void)
94 __get_cpu_var(idle_state) = CPU_NOT_IDLE;
95 atomic_notifier_call_chain(&idle_notifier, IDLE_END, NULL);
98 /* Called from interrupts to signify idle end */
101 if (current->pid | read_pda(irqcount))
107 * We use this if we don't have any better
110 static void default_idle(void)
114 current_thread_info()->status &= ~TS_POLLING;
115 smp_mb__after_clear_bit();
116 while (!need_resched()) {
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)
141 "i" (_TIF_NEED_RESCHED),
142 "m" (current_thread_info()->flags));
145 void cpu_idle_wait(void)
147 unsigned int cpu, this_cpu = get_cpu();
150 set_cpus_allowed(current, cpumask_of_cpu(this_cpu));
154 for_each_online_cpu(cpu) {
155 per_cpu(cpu_idle_state, cpu) = 1;
159 __get_cpu_var(cpu_idle_state) = 0;
164 for_each_online_cpu(cpu) {
165 if (cpu_isset(cpu, map) &&
166 !per_cpu(cpu_idle_state, cpu))
169 cpus_and(map, map, cpu_online_map);
170 } while (!cpus_empty(map));
172 EXPORT_SYMBOL_GPL(cpu_idle_wait);
174 #ifdef CONFIG_HOTPLUG_CPU
175 DECLARE_PER_CPU(int, cpu_state);
178 /* We halt the CPU with physical CPU hotplug */
179 static inline void play_dead(void)
185 __get_cpu_var(cpu_state) = CPU_DEAD;
192 static inline void play_dead(void)
196 #endif /* CONFIG_HOTPLUG_CPU */
199 * The idle thread. There's no useful work to be
200 * done, so just try to conserve power and have a
201 * low exit latency (ie sit in a loop waiting for
202 * somebody to say that they'd like to reschedule)
206 current_thread_info()->status |= TS_POLLING;
207 /* endless idle loop with no priority at all */
209 while (!need_resched()) {
212 if (__get_cpu_var(cpu_idle_state))
213 __get_cpu_var(cpu_idle_state) = 0;
219 if (cpu_is_offline(smp_processor_id()))
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 static void mwait_idle(void)
243 while (!need_resched()) {
244 __monitor((void *)¤t_thread_info()->flags, 0, 0);
252 void __cpuinit select_idle_routine(const struct cpuinfo_x86 *c)
255 if (cpu_has(c, X86_FEATURE_MWAIT)) {
257 * Skip, if setup has overridden idle.
258 * One CPU supports mwait => All CPUs supports mwait
262 printk("using mwait in idle threads.\n");
265 pm_idle = mwait_idle;
270 static int __init idle_setup (char *str)
272 if (!strncmp(str, "poll", 4)) {
273 printk("using polling idle threads.\n");
277 boot_option_idle_override = 1;
281 __setup("idle=", idle_setup);
283 /* Prints also some state that isn't saved in the pt_regs */
284 void __show_regs(struct pt_regs * regs)
286 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L, fs, gs, shadowgs;
287 unsigned int fsindex,gsindex;
288 unsigned int ds,cs,es;
292 printk("Pid: %d, comm: %.20s %s %s %.*s\n",
293 current->pid, current->comm, print_tainted(),
294 system_utsname.release,
295 (int)strcspn(system_utsname.version, " "),
296 system_utsname.version);
297 printk("RIP: %04lx:[<%016lx>] ", regs->cs & 0xffff, regs->rip);
298 printk_address(regs->rip);
299 printk("RSP: %04lx:%016lx EFLAGS: %08lx\n", regs->ss, regs->rsp,
301 printk("RAX: %016lx RBX: %016lx RCX: %016lx\n",
302 regs->rax, regs->rbx, regs->rcx);
303 printk("RDX: %016lx RSI: %016lx RDI: %016lx\n",
304 regs->rdx, regs->rsi, regs->rdi);
305 printk("RBP: %016lx R08: %016lx R09: %016lx\n",
306 regs->rbp, regs->r8, regs->r9);
307 printk("R10: %016lx R11: %016lx R12: %016lx\n",
308 regs->r10, regs->r11, regs->r12);
309 printk("R13: %016lx R14: %016lx R15: %016lx\n",
310 regs->r13, regs->r14, regs->r15);
312 asm("movl %%ds,%0" : "=r" (ds));
313 asm("movl %%cs,%0" : "=r" (cs));
314 asm("movl %%es,%0" : "=r" (es));
315 asm("movl %%fs,%0" : "=r" (fsindex));
316 asm("movl %%gs,%0" : "=r" (gsindex));
318 rdmsrl(MSR_FS_BASE, fs);
319 rdmsrl(MSR_GS_BASE, gs);
320 rdmsrl(MSR_KERNEL_GS_BASE, shadowgs);
322 asm("movq %%cr0, %0": "=r" (cr0));
323 asm("movq %%cr2, %0": "=r" (cr2));
324 asm("movq %%cr3, %0": "=r" (cr3));
325 asm("movq %%cr4, %0": "=r" (cr4));
327 printk("FS: %016lx(%04x) GS:%016lx(%04x) knlGS:%016lx\n",
328 fs,fsindex,gs,gsindex,shadowgs);
329 printk("CS: %04x DS: %04x ES: %04x CR0: %016lx\n", cs, ds, es, cr0);
330 printk("CR2: %016lx CR3: %016lx CR4: %016lx\n", cr2, cr3, cr4);
333 void show_regs(struct pt_regs *regs)
335 printk("CPU %d:", smp_processor_id());
337 show_trace(NULL, regs, (void *)(regs + 1));
341 * Free current thread data structures etc..
343 void exit_thread(void)
345 struct task_struct *me = current;
346 struct thread_struct *t = &me->thread;
348 if (me->thread.io_bitmap_ptr) {
349 struct tss_struct *tss = &per_cpu(init_tss, get_cpu());
351 kfree(t->io_bitmap_ptr);
352 t->io_bitmap_ptr = NULL;
354 * Careful, clear this in the TSS too:
356 memset(tss->io_bitmap, 0xff, t->io_bitmap_max);
357 t->io_bitmap_max = 0;
362 void flush_thread(void)
364 struct task_struct *tsk = current;
365 struct thread_info *t = current_thread_info();
367 if (t->flags & _TIF_ABI_PENDING) {
368 t->flags ^= (_TIF_ABI_PENDING | _TIF_IA32);
369 if (t->flags & _TIF_IA32)
370 current_thread_info()->status |= TS_COMPAT;
373 tsk->thread.debugreg0 = 0;
374 tsk->thread.debugreg1 = 0;
375 tsk->thread.debugreg2 = 0;
376 tsk->thread.debugreg3 = 0;
377 tsk->thread.debugreg6 = 0;
378 tsk->thread.debugreg7 = 0;
379 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
381 * Forget coprocessor state..
387 void release_thread(struct task_struct *dead_task)
390 if (dead_task->mm->context.size) {
391 printk("WARNING: dead process %8s still has LDT? <%p/%d>\n",
393 dead_task->mm->context.ldt,
394 dead_task->mm->context.size);
400 static inline void set_32bit_tls(struct task_struct *t, int tls, u32 addr)
402 struct user_desc ud = {
409 struct n_desc_struct *desc = (void *)t->thread.tls_array;
411 desc->a = LDT_entry_a(&ud);
412 desc->b = LDT_entry_b(&ud);
415 static inline u32 read_32bit_tls(struct task_struct *t, int tls)
417 struct desc_struct *desc = (void *)t->thread.tls_array;
420 (((u32)desc->base1) << 16) |
421 (((u32)desc->base2) << 24);
425 * This gets called before we allocate a new thread and copy
426 * the current task into it.
428 void prepare_to_copy(struct task_struct *tsk)
433 int copy_thread(int nr, unsigned long clone_flags, unsigned long rsp,
434 unsigned long unused,
435 struct task_struct * p, struct pt_regs * regs)
438 struct pt_regs * childregs;
439 struct task_struct *me = current;
441 childregs = ((struct pt_regs *)
442 (THREAD_SIZE + task_stack_page(p))) - 1;
446 childregs->rsp = rsp;
448 childregs->rsp = (unsigned long)childregs;
450 p->thread.rsp = (unsigned long) childregs;
451 p->thread.rsp0 = (unsigned long) (childregs+1);
452 p->thread.userrsp = me->thread.userrsp;
454 set_tsk_thread_flag(p, TIF_FORK);
456 p->thread.fs = me->thread.fs;
457 p->thread.gs = me->thread.gs;
459 asm("mov %%gs,%0" : "=m" (p->thread.gsindex));
460 asm("mov %%fs,%0" : "=m" (p->thread.fsindex));
461 asm("mov %%es,%0" : "=m" (p->thread.es));
462 asm("mov %%ds,%0" : "=m" (p->thread.ds));
464 if (unlikely(me->thread.io_bitmap_ptr != NULL)) {
465 p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL);
466 if (!p->thread.io_bitmap_ptr) {
467 p->thread.io_bitmap_max = 0;
470 memcpy(p->thread.io_bitmap_ptr, me->thread.io_bitmap_ptr,
475 * Set a new TLS for the child thread?
477 if (clone_flags & CLONE_SETTLS) {
478 #ifdef CONFIG_IA32_EMULATION
479 if (test_thread_flag(TIF_IA32))
480 err = ia32_child_tls(p, childregs);
483 err = do_arch_prctl(p, ARCH_SET_FS, childregs->r8);
489 if (err && p->thread.io_bitmap_ptr) {
490 kfree(p->thread.io_bitmap_ptr);
491 p->thread.io_bitmap_max = 0;
497 * This special macro can be used to load a debugging register
499 #define loaddebug(thread,r) set_debugreg(thread->debugreg ## r, r)
502 * switch_to(x,y) should switch tasks from x to y.
504 * This could still be optimized:
505 * - fold all the options into a flag word and test it with a single test.
506 * - could test fs/gs bitsliced
508 * Kprobes not supported here. Set the probe on schedule instead.
510 __kprobes struct task_struct *
511 __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
513 struct thread_struct *prev = &prev_p->thread,
514 *next = &next_p->thread;
515 int cpu = smp_processor_id();
516 struct tss_struct *tss = &per_cpu(init_tss, cpu);
519 * Reload esp0, LDT and the page table pointer:
521 tss->rsp0 = next->rsp0;
525 * This won't pick up thread selector changes, but I guess that is ok.
527 asm volatile("mov %%es,%0" : "=m" (prev->es));
528 if (unlikely(next->es | prev->es))
529 loadsegment(es, next->es);
531 asm volatile ("mov %%ds,%0" : "=m" (prev->ds));
532 if (unlikely(next->ds | prev->ds))
533 loadsegment(ds, next->ds);
542 asm volatile("movl %%fs,%0" : "=r" (fsindex));
543 /* segment register != 0 always requires a reload.
544 also reload when it has changed.
545 when prev process used 64bit base always reload
546 to avoid an information leak. */
547 if (unlikely(fsindex | next->fsindex | prev->fs)) {
548 loadsegment(fs, next->fsindex);
549 /* check if the user used a selector != 0
550 * if yes clear 64bit base, since overloaded base
551 * is always mapped to the Null selector
556 /* when next process has a 64bit base use it */
558 wrmsrl(MSR_FS_BASE, next->fs);
559 prev->fsindex = fsindex;
563 asm volatile("movl %%gs,%0" : "=r" (gsindex));
564 if (unlikely(gsindex | next->gsindex | prev->gs)) {
565 load_gs_index(next->gsindex);
570 wrmsrl(MSR_KERNEL_GS_BASE, next->gs);
571 prev->gsindex = gsindex;
575 * Switch the PDA and FPU contexts.
577 prev->userrsp = read_pda(oldrsp);
578 write_pda(oldrsp, next->userrsp);
579 write_pda(pcurrent, next_p);
581 /* This must be here to ensure both math_state_restore() and
582 kernel_fpu_begin() work consistently.
583 And the AMD workaround requires it to be after DS reload. */
585 write_pda(kernelstack,
586 task_stack_page(next_p) + THREAD_SIZE - PDA_STACKOFFSET);
589 * Now maybe reload the debug registers
591 if (unlikely(next->debugreg7)) {
603 * Handle the IO bitmap
605 if (unlikely(prev->io_bitmap_ptr || next->io_bitmap_ptr)) {
606 if (next->io_bitmap_ptr)
608 * Copy the relevant range of the IO bitmap.
609 * Normally this is 128 bytes or less:
611 memcpy(tss->io_bitmap, next->io_bitmap_ptr,
612 max(prev->io_bitmap_max, next->io_bitmap_max));
615 * Clear any possible leftover bits:
617 memset(tss->io_bitmap, 0xff, prev->io_bitmap_max);
625 * sys_execve() executes a new program.
628 long sys_execve(char __user *name, char __user * __user *argv,
629 char __user * __user *envp, struct pt_regs regs)
634 filename = getname(name);
635 error = PTR_ERR(filename);
636 if (IS_ERR(filename))
638 error = do_execve(filename, argv, envp, ®s);
643 void set_personality_64bit(void)
645 /* inherit personality from parent */
647 /* Make sure to be in 64bit mode */
648 clear_thread_flag(TIF_IA32);
651 asmlinkage long sys_fork(struct pt_regs *regs)
653 return do_fork(SIGCHLD, regs->rsp, regs, 0, NULL, NULL);
657 sys_clone(unsigned long clone_flags, unsigned long newsp,
658 void __user *parent_tid, void __user *child_tid, struct pt_regs *regs)
662 return do_fork(clone_flags, newsp, regs, 0, parent_tid, child_tid);
666 * This is trivial, and on the face of it looks like it
667 * could equally well be done in user mode.
669 * Not so, for quite unobvious reasons - register pressure.
670 * In user mode vfork() cannot have a stack frame, and if
671 * done by calling the "clone()" system call directly, you
672 * do not have enough call-clobbered registers to hold all
673 * the information you need.
675 asmlinkage long sys_vfork(struct pt_regs *regs)
677 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->rsp, regs, 0,
681 unsigned long get_wchan(struct task_struct *p)
687 if (!p || p == current || p->state==TASK_RUNNING)
689 stack = (unsigned long)task_stack_page(p);
690 if (p->thread.rsp < stack || p->thread.rsp > stack+THREAD_SIZE)
692 fp = *(u64 *)(p->thread.rsp);
694 if (fp < (unsigned long)stack ||
695 fp > (unsigned long)stack+THREAD_SIZE)
697 rip = *(u64 *)(fp+8);
698 if (!in_sched_functions(rip))
701 } while (count++ < 16);
705 long do_arch_prctl(struct task_struct *task, int code, unsigned long addr)
708 int doit = task == current;
713 if (addr >= TASK_SIZE_OF(task))
716 /* handle small bases via the GDT because that's faster to
718 if (addr <= 0xffffffff) {
719 set_32bit_tls(task, GS_TLS, addr);
721 load_TLS(&task->thread, cpu);
722 load_gs_index(GS_TLS_SEL);
724 task->thread.gsindex = GS_TLS_SEL;
727 task->thread.gsindex = 0;
728 task->thread.gs = addr;
731 ret = checking_wrmsrl(MSR_KERNEL_GS_BASE, addr);
737 /* Not strictly needed for fs, but do it for symmetry
739 if (addr >= TASK_SIZE_OF(task))
742 /* handle small bases via the GDT because that's faster to
744 if (addr <= 0xffffffff) {
745 set_32bit_tls(task, FS_TLS, addr);
747 load_TLS(&task->thread, cpu);
748 asm volatile("movl %0,%%fs" :: "r"(FS_TLS_SEL));
750 task->thread.fsindex = FS_TLS_SEL;
753 task->thread.fsindex = 0;
754 task->thread.fs = addr;
756 /* set the selector to 0 to not confuse
758 asm volatile("movl %0,%%fs" :: "r" (0));
759 ret = checking_wrmsrl(MSR_FS_BASE, addr);
766 if (task->thread.fsindex == FS_TLS_SEL)
767 base = read_32bit_tls(task, FS_TLS);
769 rdmsrl(MSR_FS_BASE, base);
771 base = task->thread.fs;
772 ret = put_user(base, (unsigned long __user *)addr);
778 if (task->thread.gsindex == GS_TLS_SEL)
779 base = read_32bit_tls(task, GS_TLS);
781 asm("movl %%gs,%0" : "=r" (gsindex));
783 rdmsrl(MSR_KERNEL_GS_BASE, base);
785 base = task->thread.gs;
788 base = task->thread.gs;
789 ret = put_user(base, (unsigned long __user *)addr);
801 long sys_arch_prctl(int code, unsigned long addr)
803 return do_arch_prctl(current, code, addr);
807 * Capture the user space registers if the task is not running (in user space)
809 int dump_task_regs(struct task_struct *tsk, elf_gregset_t *regs)
811 struct pt_regs *pp, ptregs;
813 pp = task_pt_regs(tsk);
819 elf_core_copy_regs(regs, &ptregs);
824 unsigned long arch_align_stack(unsigned long sp)
826 if (randomize_va_space)
827 sp -= get_random_int() % 8192;