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 static DEFINE_PER_CPU(unsigned int, cpu_idle_state);
68 static ATOMIC_NOTIFIER_HEAD(idle_notifier);
70 void idle_notifier_register(struct notifier_block *n)
72 atomic_notifier_chain_register(&idle_notifier, n);
74 EXPORT_SYMBOL_GPL(idle_notifier_register);
76 void idle_notifier_unregister(struct notifier_block *n)
78 atomic_notifier_chain_unregister(&idle_notifier, n);
80 EXPORT_SYMBOL(idle_notifier_unregister);
82 enum idle_state { CPU_IDLE, CPU_NOT_IDLE };
83 static DEFINE_PER_CPU(enum idle_state, idle_state) = CPU_NOT_IDLE;
87 __get_cpu_var(idle_state) = CPU_IDLE;
88 atomic_notifier_call_chain(&idle_notifier, IDLE_START, NULL);
91 static void __exit_idle(void)
93 __get_cpu_var(idle_state) = CPU_NOT_IDLE;
94 atomic_notifier_call_chain(&idle_notifier, IDLE_END, NULL);
97 /* Called from interrupts to signify idle end */
100 if (current->pid | read_pda(irqcount))
106 * We use this if we don't have any better
109 static void default_idle(void)
113 clear_thread_flag(TIF_POLLING_NRFLAG);
114 smp_mb__after_clear_bit();
115 while (!need_resched()) {
122 set_thread_flag(TIF_POLLING_NRFLAG);
126 * On SMP it's slightly faster (but much more power-consuming!)
127 * to poll the ->need_resched flag instead of waiting for the
128 * cross-CPU IPI to arrive. Use this option with caution.
130 static void poll_idle (void)
140 "i" (_TIF_NEED_RESCHED),
141 "m" (current_thread_info()->flags));
144 void cpu_idle_wait(void)
146 unsigned int cpu, this_cpu = get_cpu();
149 set_cpus_allowed(current, cpumask_of_cpu(this_cpu));
153 for_each_online_cpu(cpu) {
154 per_cpu(cpu_idle_state, cpu) = 1;
158 __get_cpu_var(cpu_idle_state) = 0;
163 for_each_online_cpu(cpu) {
164 if (cpu_isset(cpu, map) &&
165 !per_cpu(cpu_idle_state, cpu))
168 cpus_and(map, map, cpu_online_map);
169 } while (!cpus_empty(map));
171 EXPORT_SYMBOL_GPL(cpu_idle_wait);
173 #ifdef CONFIG_HOTPLUG_CPU
174 DECLARE_PER_CPU(int, cpu_state);
177 /* We halt the CPU with physical CPU hotplug */
178 static inline void play_dead(void)
184 __get_cpu_var(cpu_state) = CPU_DEAD;
191 static inline void play_dead(void)
195 #endif /* CONFIG_HOTPLUG_CPU */
198 * The idle thread. There's no useful work to be
199 * done, so just try to conserve power and have a
200 * low exit latency (ie sit in a loop waiting for
201 * somebody to say that they'd like to reschedule)
205 set_thread_flag(TIF_POLLING_NRFLAG);
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("\nRSP: %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(®s->rsp);
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);
370 tsk->thread.debugreg0 = 0;
371 tsk->thread.debugreg1 = 0;
372 tsk->thread.debugreg2 = 0;
373 tsk->thread.debugreg3 = 0;
374 tsk->thread.debugreg6 = 0;
375 tsk->thread.debugreg7 = 0;
376 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
378 * Forget coprocessor state..
384 void release_thread(struct task_struct *dead_task)
387 if (dead_task->mm->context.size) {
388 printk("WARNING: dead process %8s still has LDT? <%p/%d>\n",
390 dead_task->mm->context.ldt,
391 dead_task->mm->context.size);
397 static inline void set_32bit_tls(struct task_struct *t, int tls, u32 addr)
399 struct user_desc ud = {
406 struct n_desc_struct *desc = (void *)t->thread.tls_array;
408 desc->a = LDT_entry_a(&ud);
409 desc->b = LDT_entry_b(&ud);
412 static inline u32 read_32bit_tls(struct task_struct *t, int tls)
414 struct desc_struct *desc = (void *)t->thread.tls_array;
417 (((u32)desc->base1) << 16) |
418 (((u32)desc->base2) << 24);
422 * This gets called before we allocate a new thread and copy
423 * the current task into it.
425 void prepare_to_copy(struct task_struct *tsk)
430 int copy_thread(int nr, unsigned long clone_flags, unsigned long rsp,
431 unsigned long unused,
432 struct task_struct * p, struct pt_regs * regs)
435 struct pt_regs * childregs;
436 struct task_struct *me = current;
438 childregs = ((struct pt_regs *)
439 (THREAD_SIZE + task_stack_page(p))) - 1;
443 childregs->rsp = rsp;
445 childregs->rsp = (unsigned long)childregs;
447 p->thread.rsp = (unsigned long) childregs;
448 p->thread.rsp0 = (unsigned long) (childregs+1);
449 p->thread.userrsp = me->thread.userrsp;
451 set_tsk_thread_flag(p, TIF_FORK);
453 p->thread.fs = me->thread.fs;
454 p->thread.gs = me->thread.gs;
456 asm("mov %%gs,%0" : "=m" (p->thread.gsindex));
457 asm("mov %%fs,%0" : "=m" (p->thread.fsindex));
458 asm("mov %%es,%0" : "=m" (p->thread.es));
459 asm("mov %%ds,%0" : "=m" (p->thread.ds));
461 if (unlikely(me->thread.io_bitmap_ptr != NULL)) {
462 p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL);
463 if (!p->thread.io_bitmap_ptr) {
464 p->thread.io_bitmap_max = 0;
467 memcpy(p->thread.io_bitmap_ptr, me->thread.io_bitmap_ptr,
472 * Set a new TLS for the child thread?
474 if (clone_flags & CLONE_SETTLS) {
475 #ifdef CONFIG_IA32_EMULATION
476 if (test_thread_flag(TIF_IA32))
477 err = ia32_child_tls(p, childregs);
480 err = do_arch_prctl(p, ARCH_SET_FS, childregs->r8);
486 if (err && p->thread.io_bitmap_ptr) {
487 kfree(p->thread.io_bitmap_ptr);
488 p->thread.io_bitmap_max = 0;
494 * This special macro can be used to load a debugging register
496 #define loaddebug(thread,r) set_debugreg(thread->debugreg ## r, r)
499 * switch_to(x,y) should switch tasks from x to y.
501 * This could still be optimized:
502 * - fold all the options into a flag word and test it with a single test.
503 * - could test fs/gs bitsliced
505 * Kprobes not supported here. Set the probe on schedule instead.
507 __kprobes struct task_struct *
508 __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
510 struct thread_struct *prev = &prev_p->thread,
511 *next = &next_p->thread;
512 int cpu = smp_processor_id();
513 struct tss_struct *tss = &per_cpu(init_tss, cpu);
516 * Reload esp0, LDT and the page table pointer:
518 tss->rsp0 = next->rsp0;
522 * This won't pick up thread selector changes, but I guess that is ok.
524 asm volatile("mov %%es,%0" : "=m" (prev->es));
525 if (unlikely(next->es | prev->es))
526 loadsegment(es, next->es);
528 asm volatile ("mov %%ds,%0" : "=m" (prev->ds));
529 if (unlikely(next->ds | prev->ds))
530 loadsegment(ds, next->ds);
539 asm volatile("movl %%fs,%0" : "=r" (fsindex));
540 /* segment register != 0 always requires a reload.
541 also reload when it has changed.
542 when prev process used 64bit base always reload
543 to avoid an information leak. */
544 if (unlikely(fsindex | next->fsindex | prev->fs)) {
545 loadsegment(fs, next->fsindex);
546 /* check if the user used a selector != 0
547 * if yes clear 64bit base, since overloaded base
548 * is always mapped to the Null selector
553 /* when next process has a 64bit base use it */
555 wrmsrl(MSR_FS_BASE, next->fs);
556 prev->fsindex = fsindex;
560 asm volatile("movl %%gs,%0" : "=r" (gsindex));
561 if (unlikely(gsindex | next->gsindex | prev->gs)) {
562 load_gs_index(next->gsindex);
567 wrmsrl(MSR_KERNEL_GS_BASE, next->gs);
568 prev->gsindex = gsindex;
572 * Switch the PDA and FPU contexts.
574 prev->userrsp = read_pda(oldrsp);
575 write_pda(oldrsp, next->userrsp);
576 write_pda(pcurrent, next_p);
578 /* This must be here to ensure both math_state_restore() and
579 kernel_fpu_begin() work consistently.
580 And the AMD workaround requires it to be after DS reload. */
582 write_pda(kernelstack,
583 task_stack_page(next_p) + THREAD_SIZE - PDA_STACKOFFSET);
586 * Now maybe reload the debug registers
588 if (unlikely(next->debugreg7)) {
600 * Handle the IO bitmap
602 if (unlikely(prev->io_bitmap_ptr || next->io_bitmap_ptr)) {
603 if (next->io_bitmap_ptr)
605 * Copy the relevant range of the IO bitmap.
606 * Normally this is 128 bytes or less:
608 memcpy(tss->io_bitmap, next->io_bitmap_ptr,
609 max(prev->io_bitmap_max, next->io_bitmap_max));
612 * Clear any possible leftover bits:
614 memset(tss->io_bitmap, 0xff, prev->io_bitmap_max);
622 * sys_execve() executes a new program.
625 long sys_execve(char __user *name, char __user * __user *argv,
626 char __user * __user *envp, struct pt_regs regs)
631 filename = getname(name);
632 error = PTR_ERR(filename);
633 if (IS_ERR(filename))
635 error = do_execve(filename, argv, envp, ®s);
638 current->ptrace &= ~PT_DTRACE;
639 task_unlock(current);
645 void set_personality_64bit(void)
647 /* inherit personality from parent */
649 /* Make sure to be in 64bit mode */
650 clear_thread_flag(TIF_IA32);
652 /* TBD: overwrites user setup. Should have two bits.
653 But 64bit processes have always behaved this way,
654 so it's not too bad. The main problem is just that
655 32bit childs are affected again. */
656 current->personality &= ~READ_IMPLIES_EXEC;
659 asmlinkage long sys_fork(struct pt_regs *regs)
661 return do_fork(SIGCHLD, regs->rsp, regs, 0, NULL, NULL);
665 sys_clone(unsigned long clone_flags, unsigned long newsp,
666 void __user *parent_tid, void __user *child_tid, struct pt_regs *regs)
670 return do_fork(clone_flags, newsp, regs, 0, parent_tid, child_tid);
674 * This is trivial, and on the face of it looks like it
675 * could equally well be done in user mode.
677 * Not so, for quite unobvious reasons - register pressure.
678 * In user mode vfork() cannot have a stack frame, and if
679 * done by calling the "clone()" system call directly, you
680 * do not have enough call-clobbered registers to hold all
681 * the information you need.
683 asmlinkage long sys_vfork(struct pt_regs *regs)
685 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->rsp, regs, 0,
689 unsigned long get_wchan(struct task_struct *p)
695 if (!p || p == current || p->state==TASK_RUNNING)
697 stack = (unsigned long)task_stack_page(p);
698 if (p->thread.rsp < stack || p->thread.rsp > stack+THREAD_SIZE)
700 fp = *(u64 *)(p->thread.rsp);
702 if (fp < (unsigned long)stack ||
703 fp > (unsigned long)stack+THREAD_SIZE)
705 rip = *(u64 *)(fp+8);
706 if (!in_sched_functions(rip))
709 } while (count++ < 16);
713 long do_arch_prctl(struct task_struct *task, int code, unsigned long addr)
716 int doit = task == current;
721 if (addr >= TASK_SIZE_OF(task))
724 /* handle small bases via the GDT because that's faster to
726 if (addr <= 0xffffffff) {
727 set_32bit_tls(task, GS_TLS, addr);
729 load_TLS(&task->thread, cpu);
730 load_gs_index(GS_TLS_SEL);
732 task->thread.gsindex = GS_TLS_SEL;
735 task->thread.gsindex = 0;
736 task->thread.gs = addr;
739 ret = checking_wrmsrl(MSR_KERNEL_GS_BASE, addr);
745 /* Not strictly needed for fs, but do it for symmetry
747 if (addr >= TASK_SIZE_OF(task))
750 /* handle small bases via the GDT because that's faster to
752 if (addr <= 0xffffffff) {
753 set_32bit_tls(task, FS_TLS, addr);
755 load_TLS(&task->thread, cpu);
756 asm volatile("movl %0,%%fs" :: "r"(FS_TLS_SEL));
758 task->thread.fsindex = FS_TLS_SEL;
761 task->thread.fsindex = 0;
762 task->thread.fs = addr;
764 /* set the selector to 0 to not confuse
766 asm volatile("movl %0,%%fs" :: "r" (0));
767 ret = checking_wrmsrl(MSR_FS_BASE, addr);
774 if (task->thread.fsindex == FS_TLS_SEL)
775 base = read_32bit_tls(task, FS_TLS);
777 rdmsrl(MSR_FS_BASE, base);
779 base = task->thread.fs;
780 ret = put_user(base, (unsigned long __user *)addr);
786 if (task->thread.gsindex == GS_TLS_SEL)
787 base = read_32bit_tls(task, GS_TLS);
789 asm("movl %%gs,%0" : "=r" (gsindex));
791 rdmsrl(MSR_KERNEL_GS_BASE, base);
793 base = task->thread.gs;
796 base = task->thread.gs;
797 ret = put_user(base, (unsigned long __user *)addr);
809 long sys_arch_prctl(int code, unsigned long addr)
811 return do_arch_prctl(current, code, addr);
815 * Capture the user space registers if the task is not running (in user space)
817 int dump_task_regs(struct task_struct *tsk, elf_gregset_t *regs)
819 struct pt_regs *pp, ptregs;
821 pp = task_pt_regs(tsk);
827 elf_core_copy_regs(regs, &ptregs);
832 unsigned long arch_align_stack(unsigned long sp)
834 if (randomize_va_space)
835 sp -= get_random_int() % 8192;