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>
27 #include <linux/elfcore.h>
28 #include <linux/smp.h>
29 #include <linux/slab.h>
30 #include <linux/user.h>
31 #include <linux/module.h>
32 #include <linux/a.out.h>
33 #include <linux/interrupt.h>
34 #include <linux/delay.h>
35 #include <linux/ptrace.h>
36 #include <linux/utsname.h>
37 #include <linux/random.h>
38 #include <linux/notifier.h>
39 #include <linux/kprobes.h>
40 #include <linux/kdebug.h>
42 #include <asm/uaccess.h>
43 #include <asm/pgtable.h>
44 #include <asm/system.h>
46 #include <asm/processor.h>
48 #include <asm/mmu_context.h>
50 #include <asm/prctl.h>
52 #include <asm/proto.h>
56 asmlinkage extern void ret_from_fork(void);
58 unsigned long kernel_thread_flags = CLONE_VM | CLONE_UNTRACED;
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(KERN_INFO "using mwait in idle threads.\n");
285 pm_idle = mwait_idle;
290 static int __init idle_setup (char *str)
292 if (!strcmp(str, "poll")) {
293 printk("using polling idle threads.\n");
295 } else if (!strcmp(str, "mwait"))
300 boot_option_idle_override = 1;
303 early_param("idle", idle_setup);
305 /* Prints also some state that isn't saved in the pt_regs */
306 void __show_regs(struct pt_regs * regs)
308 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L, fs, gs, shadowgs;
309 unsigned long d0, d1, d2, d3, d6, d7;
310 unsigned int fsindex,gsindex;
311 unsigned int ds,cs,es;
315 printk("Pid: %d, comm: %.20s %s %s %.*s\n",
316 current->pid, current->comm, print_tainted(),
317 init_utsname()->release,
318 (int)strcspn(init_utsname()->version, " "),
319 init_utsname()->version);
320 printk("RIP: %04lx:[<%016lx>] ", regs->cs & 0xffff, regs->rip);
321 printk_address(regs->rip);
322 printk("RSP: %04lx:%016lx EFLAGS: %08lx\n", regs->ss, regs->rsp,
324 printk("RAX: %016lx RBX: %016lx RCX: %016lx\n",
325 regs->rax, regs->rbx, regs->rcx);
326 printk("RDX: %016lx RSI: %016lx RDI: %016lx\n",
327 regs->rdx, regs->rsi, regs->rdi);
328 printk("RBP: %016lx R08: %016lx R09: %016lx\n",
329 regs->rbp, regs->r8, regs->r9);
330 printk("R10: %016lx R11: %016lx R12: %016lx\n",
331 regs->r10, regs->r11, regs->r12);
332 printk("R13: %016lx R14: %016lx R15: %016lx\n",
333 regs->r13, regs->r14, regs->r15);
335 asm("movl %%ds,%0" : "=r" (ds));
336 asm("movl %%cs,%0" : "=r" (cs));
337 asm("movl %%es,%0" : "=r" (es));
338 asm("movl %%fs,%0" : "=r" (fsindex));
339 asm("movl %%gs,%0" : "=r" (gsindex));
341 rdmsrl(MSR_FS_BASE, fs);
342 rdmsrl(MSR_GS_BASE, gs);
343 rdmsrl(MSR_KERNEL_GS_BASE, shadowgs);
350 printk("FS: %016lx(%04x) GS:%016lx(%04x) knlGS:%016lx\n",
351 fs,fsindex,gs,gsindex,shadowgs);
352 printk("CS: %04x DS: %04x ES: %04x CR0: %016lx\n", cs, ds, es, cr0);
353 printk("CR2: %016lx CR3: %016lx CR4: %016lx\n", cr2, cr3, cr4);
358 printk("DR0: %016lx DR1: %016lx DR2: %016lx\n", d0, d1, d2);
362 printk("DR3: %016lx DR6: %016lx DR7: %016lx\n", d3, d6, d7);
365 void show_regs(struct pt_regs *regs)
367 printk("CPU %d:", smp_processor_id());
369 show_trace(NULL, regs, (void *)(regs + 1));
373 * Free current thread data structures etc..
375 void exit_thread(void)
377 struct task_struct *me = current;
378 struct thread_struct *t = &me->thread;
380 if (me->thread.io_bitmap_ptr) {
381 struct tss_struct *tss = &per_cpu(init_tss, get_cpu());
383 kfree(t->io_bitmap_ptr);
384 t->io_bitmap_ptr = NULL;
385 clear_thread_flag(TIF_IO_BITMAP);
387 * Careful, clear this in the TSS too:
389 memset(tss->io_bitmap, 0xff, t->io_bitmap_max);
390 t->io_bitmap_max = 0;
395 void flush_thread(void)
397 struct task_struct *tsk = current;
399 if (test_tsk_thread_flag(tsk, TIF_ABI_PENDING)) {
400 clear_tsk_thread_flag(tsk, TIF_ABI_PENDING);
401 if (test_tsk_thread_flag(tsk, TIF_IA32)) {
402 clear_tsk_thread_flag(tsk, TIF_IA32);
404 set_tsk_thread_flag(tsk, TIF_IA32);
405 current_thread_info()->status |= TS_COMPAT;
408 clear_tsk_thread_flag(tsk, TIF_DEBUG);
410 tsk->thread.debugreg0 = 0;
411 tsk->thread.debugreg1 = 0;
412 tsk->thread.debugreg2 = 0;
413 tsk->thread.debugreg3 = 0;
414 tsk->thread.debugreg6 = 0;
415 tsk->thread.debugreg7 = 0;
416 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
418 * Forget coprocessor state..
424 void release_thread(struct task_struct *dead_task)
427 if (dead_task->mm->context.size) {
428 printk("WARNING: dead process %8s still has LDT? <%p/%d>\n",
430 dead_task->mm->context.ldt,
431 dead_task->mm->context.size);
437 static inline void set_32bit_tls(struct task_struct *t, int tls, u32 addr)
439 struct user_desc ud = {
446 struct n_desc_struct *desc = (void *)t->thread.tls_array;
448 desc->a = LDT_entry_a(&ud);
449 desc->b = LDT_entry_b(&ud);
452 static inline u32 read_32bit_tls(struct task_struct *t, int tls)
454 struct desc_struct *desc = (void *)t->thread.tls_array;
457 (((u32)desc->base1) << 16) |
458 (((u32)desc->base2) << 24);
462 * This gets called before we allocate a new thread and copy
463 * the current task into it.
465 void prepare_to_copy(struct task_struct *tsk)
470 int copy_thread(int nr, unsigned long clone_flags, unsigned long rsp,
471 unsigned long unused,
472 struct task_struct * p, struct pt_regs * regs)
475 struct pt_regs * childregs;
476 struct task_struct *me = current;
478 childregs = ((struct pt_regs *)
479 (THREAD_SIZE + task_stack_page(p))) - 1;
483 childregs->rsp = rsp;
485 childregs->rsp = (unsigned long)childregs;
487 p->thread.rsp = (unsigned long) childregs;
488 p->thread.rsp0 = (unsigned long) (childregs+1);
489 p->thread.userrsp = me->thread.userrsp;
491 set_tsk_thread_flag(p, TIF_FORK);
493 p->thread.fs = me->thread.fs;
494 p->thread.gs = me->thread.gs;
496 asm("mov %%gs,%0" : "=m" (p->thread.gsindex));
497 asm("mov %%fs,%0" : "=m" (p->thread.fsindex));
498 asm("mov %%es,%0" : "=m" (p->thread.es));
499 asm("mov %%ds,%0" : "=m" (p->thread.ds));
501 if (unlikely(test_tsk_thread_flag(me, TIF_IO_BITMAP))) {
502 p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL);
503 if (!p->thread.io_bitmap_ptr) {
504 p->thread.io_bitmap_max = 0;
507 memcpy(p->thread.io_bitmap_ptr, me->thread.io_bitmap_ptr,
509 set_tsk_thread_flag(p, TIF_IO_BITMAP);
513 * Set a new TLS for the child thread?
515 if (clone_flags & CLONE_SETTLS) {
516 #ifdef CONFIG_IA32_EMULATION
517 if (test_thread_flag(TIF_IA32))
518 err = ia32_child_tls(p, childregs);
521 err = do_arch_prctl(p, ARCH_SET_FS, childregs->r8);
527 if (err && p->thread.io_bitmap_ptr) {
528 kfree(p->thread.io_bitmap_ptr);
529 p->thread.io_bitmap_max = 0;
535 * This special macro can be used to load a debugging register
537 #define loaddebug(thread,r) set_debugreg(thread->debugreg ## r, r)
539 static inline void __switch_to_xtra(struct task_struct *prev_p,
540 struct task_struct *next_p,
541 struct tss_struct *tss)
543 struct thread_struct *prev, *next;
545 prev = &prev_p->thread,
546 next = &next_p->thread;
548 if (test_tsk_thread_flag(next_p, TIF_DEBUG)) {
558 if (test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
560 * Copy the relevant range of the IO bitmap.
561 * Normally this is 128 bytes or less:
563 memcpy(tss->io_bitmap, next->io_bitmap_ptr,
564 max(prev->io_bitmap_max, next->io_bitmap_max));
565 } else if (test_tsk_thread_flag(prev_p, TIF_IO_BITMAP)) {
567 * Clear any possible leftover bits:
569 memset(tss->io_bitmap, 0xff, prev->io_bitmap_max);
574 * switch_to(x,y) should switch tasks from x to y.
576 * This could still be optimized:
577 * - fold all the options into a flag word and test it with a single test.
578 * - could test fs/gs bitsliced
580 * Kprobes not supported here. Set the probe on schedule instead.
582 __kprobes struct task_struct *
583 __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
585 struct thread_struct *prev = &prev_p->thread,
586 *next = &next_p->thread;
587 int cpu = smp_processor_id();
588 struct tss_struct *tss = &per_cpu(init_tss, cpu);
590 /* we're going to use this soon, after a few expensive things */
591 if (next_p->fpu_counter>5)
592 prefetch(&next->i387.fxsave);
595 * Reload esp0, LDT and the page table pointer:
597 tss->rsp0 = next->rsp0;
601 * This won't pick up thread selector changes, but I guess that is ok.
603 asm volatile("mov %%es,%0" : "=m" (prev->es));
604 if (unlikely(next->es | prev->es))
605 loadsegment(es, next->es);
607 asm volatile ("mov %%ds,%0" : "=m" (prev->ds));
608 if (unlikely(next->ds | prev->ds))
609 loadsegment(ds, next->ds);
618 asm volatile("movl %%fs,%0" : "=r" (fsindex));
619 /* segment register != 0 always requires a reload.
620 also reload when it has changed.
621 when prev process used 64bit base always reload
622 to avoid an information leak. */
623 if (unlikely(fsindex | next->fsindex | prev->fs)) {
624 loadsegment(fs, next->fsindex);
625 /* check if the user used a selector != 0
626 * if yes clear 64bit base, since overloaded base
627 * is always mapped to the Null selector
632 /* when next process has a 64bit base use it */
634 wrmsrl(MSR_FS_BASE, next->fs);
635 prev->fsindex = fsindex;
639 asm volatile("movl %%gs,%0" : "=r" (gsindex));
640 if (unlikely(gsindex | next->gsindex | prev->gs)) {
641 load_gs_index(next->gsindex);
646 wrmsrl(MSR_KERNEL_GS_BASE, next->gs);
647 prev->gsindex = gsindex;
650 /* Must be after DS reload */
654 * Switch the PDA and FPU contexts.
656 prev->userrsp = read_pda(oldrsp);
657 write_pda(oldrsp, next->userrsp);
658 write_pda(pcurrent, next_p);
660 write_pda(kernelstack,
661 (unsigned long)task_stack_page(next_p) + THREAD_SIZE - PDA_STACKOFFSET);
662 #ifdef CONFIG_CC_STACKPROTECTOR
663 write_pda(stack_canary, next_p->stack_canary);
665 * Build time only check to make sure the stack_canary is at
666 * offset 40 in the pda; this is a gcc ABI requirement
668 BUILD_BUG_ON(offsetof(struct x8664_pda, stack_canary) != 40);
672 * Now maybe reload the debug registers and handle I/O bitmaps
674 if (unlikely((task_thread_info(next_p)->flags & _TIF_WORK_CTXSW))
675 || test_tsk_thread_flag(prev_p, TIF_IO_BITMAP))
676 __switch_to_xtra(prev_p, next_p, tss);
678 /* If the task has used fpu the last 5 timeslices, just do a full
679 * restore of the math state immediately to avoid the trap; the
680 * chances of needing FPU soon are obviously high now
682 if (next_p->fpu_counter>5)
683 math_state_restore();
688 * sys_execve() executes a new program.
691 long sys_execve(char __user *name, char __user * __user *argv,
692 char __user * __user *envp, struct pt_regs regs)
697 filename = getname(name);
698 error = PTR_ERR(filename);
699 if (IS_ERR(filename))
701 error = do_execve(filename, argv, envp, ®s);
704 current->ptrace &= ~PT_DTRACE;
705 task_unlock(current);
711 void set_personality_64bit(void)
713 /* inherit personality from parent */
715 /* Make sure to be in 64bit mode */
716 clear_thread_flag(TIF_IA32);
718 /* TBD: overwrites user setup. Should have two bits.
719 But 64bit processes have always behaved this way,
720 so it's not too bad. The main problem is just that
721 32bit childs are affected again. */
722 current->personality &= ~READ_IMPLIES_EXEC;
725 asmlinkage long sys_fork(struct pt_regs *regs)
727 return do_fork(SIGCHLD, regs->rsp, regs, 0, NULL, NULL);
731 sys_clone(unsigned long clone_flags, unsigned long newsp,
732 void __user *parent_tid, void __user *child_tid, struct pt_regs *regs)
736 return do_fork(clone_flags, newsp, regs, 0, parent_tid, child_tid);
740 * This is trivial, and on the face of it looks like it
741 * could equally well be done in user mode.
743 * Not so, for quite unobvious reasons - register pressure.
744 * In user mode vfork() cannot have a stack frame, and if
745 * done by calling the "clone()" system call directly, you
746 * do not have enough call-clobbered registers to hold all
747 * the information you need.
749 asmlinkage long sys_vfork(struct pt_regs *regs)
751 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->rsp, regs, 0,
755 unsigned long get_wchan(struct task_struct *p)
761 if (!p || p == current || p->state==TASK_RUNNING)
763 stack = (unsigned long)task_stack_page(p);
764 if (p->thread.rsp < stack || p->thread.rsp > stack+THREAD_SIZE)
766 fp = *(u64 *)(p->thread.rsp);
768 if (fp < (unsigned long)stack ||
769 fp > (unsigned long)stack+THREAD_SIZE)
771 rip = *(u64 *)(fp+8);
772 if (!in_sched_functions(rip))
775 } while (count++ < 16);
779 long do_arch_prctl(struct task_struct *task, int code, unsigned long addr)
782 int doit = task == current;
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, GS_TLS, addr);
795 load_TLS(&task->thread, cpu);
796 load_gs_index(GS_TLS_SEL);
798 task->thread.gsindex = GS_TLS_SEL;
801 task->thread.gsindex = 0;
802 task->thread.gs = addr;
805 ret = checking_wrmsrl(MSR_KERNEL_GS_BASE, addr);
811 /* Not strictly needed for fs, but do it for symmetry
813 if (addr >= TASK_SIZE_OF(task))
816 /* handle small bases via the GDT because that's faster to
818 if (addr <= 0xffffffff) {
819 set_32bit_tls(task, FS_TLS, addr);
821 load_TLS(&task->thread, cpu);
822 asm volatile("movl %0,%%fs" :: "r"(FS_TLS_SEL));
824 task->thread.fsindex = FS_TLS_SEL;
827 task->thread.fsindex = 0;
828 task->thread.fs = addr;
830 /* set the selector to 0 to not confuse
832 asm volatile("movl %0,%%fs" :: "r" (0));
833 ret = checking_wrmsrl(MSR_FS_BASE, addr);
840 if (task->thread.fsindex == FS_TLS_SEL)
841 base = read_32bit_tls(task, FS_TLS);
843 rdmsrl(MSR_FS_BASE, base);
845 base = task->thread.fs;
846 ret = put_user(base, (unsigned long __user *)addr);
852 if (task->thread.gsindex == GS_TLS_SEL)
853 base = read_32bit_tls(task, GS_TLS);
855 asm("movl %%gs,%0" : "=r" (gsindex));
857 rdmsrl(MSR_KERNEL_GS_BASE, base);
859 base = task->thread.gs;
862 base = task->thread.gs;
863 ret = put_user(base, (unsigned long __user *)addr);
875 long sys_arch_prctl(int code, unsigned long addr)
877 return do_arch_prctl(current, code, addr);
881 * Capture the user space registers if the task is not running (in user space)
883 int dump_task_regs(struct task_struct *tsk, elf_gregset_t *regs)
885 struct pt_regs *pp, ptregs;
887 pp = task_pt_regs(tsk);
893 elf_core_copy_regs(regs, &ptregs);
898 unsigned long arch_align_stack(unsigned long sp)
900 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
901 sp -= get_random_int() % 8192;