2 * Copyright (C) 1995 Linus Torvalds
4 * Pentium III FXSR, SSE support
5 * Gareth Hughes <gareth@valinux.com>, May 2000
10 * CPU hotplug support - ashok.raj@intel.com
14 * This file handles the architecture-dependent parts of process handling..
19 #include <linux/cpu.h>
20 #include <linux/errno.h>
21 #include <linux/sched.h>
22 #include <linux/kernel.h>
25 #include <linux/elfcore.h>
26 #include <linux/smp.h>
27 #include <linux/slab.h>
28 #include <linux/user.h>
29 #include <linux/module.h>
30 #include <linux/a.out.h>
31 #include <linux/interrupt.h>
32 #include <linux/delay.h>
33 #include <linux/ptrace.h>
34 #include <linux/utsname.h>
35 #include <linux/random.h>
36 #include <linux/notifier.h>
37 #include <linux/kprobes.h>
38 #include <linux/kdebug.h>
39 #include <linux/tick.h>
41 #include <asm/uaccess.h>
42 #include <asm/pgtable.h>
43 #include <asm/system.h>
45 #include <asm/processor.h>
47 #include <asm/mmu_context.h>
49 #include <asm/prctl.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;
210 tick_nohz_stop_sched_tick();
216 if (cpu_is_offline(smp_processor_id()))
219 * Idle routines should keep interrupts disabled
220 * from here on, until they go to idle.
221 * Otherwise, idle callbacks can misfire.
226 /* In many cases the interrupt that ended idle
227 has already called exit_idle. But some idle
228 loops can be woken up without interrupt. */
232 tick_nohz_restart_sched_tick();
233 preempt_enable_no_resched();
240 * This uses new MONITOR/MWAIT instructions on P4 processors with PNI,
241 * which can obviate IPI to trigger checking of need_resched.
242 * We execute MONITOR against need_resched and enter optimized wait state
243 * through MWAIT. Whenever someone changes need_resched, we would be woken
244 * up from MWAIT (without an IPI).
246 * New with Core Duo processors, MWAIT can take some hints based on CPU
249 void mwait_idle_with_hints(unsigned long eax, unsigned long ecx)
251 if (!need_resched()) {
252 __monitor((void *)¤t_thread_info()->flags, 0, 0);
259 /* Default MONITOR/MWAIT with no hints, used for default C1 state */
260 static void mwait_idle(void)
262 if (!need_resched()) {
263 __monitor((void *)¤t_thread_info()->flags, 0, 0);
274 void __cpuinit select_idle_routine(const struct cpuinfo_x86 *c)
277 if (cpu_has(c, X86_FEATURE_MWAIT)) {
279 * Skip, if setup has overridden idle.
280 * One CPU supports mwait => All CPUs supports mwait
284 printk(KERN_INFO "using mwait in idle threads.\n");
287 pm_idle = mwait_idle;
292 static int __init idle_setup (char *str)
294 if (!strcmp(str, "poll")) {
295 printk("using polling idle threads.\n");
297 } else if (!strcmp(str, "mwait"))
302 boot_option_idle_override = 1;
305 early_param("idle", idle_setup);
307 /* Prints also some state that isn't saved in the pt_regs */
308 void __show_regs(struct pt_regs * regs)
310 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L, fs, gs, shadowgs;
311 unsigned long d0, d1, d2, d3, d6, d7;
312 unsigned int fsindex,gsindex;
313 unsigned int ds,cs,es;
317 printk("Pid: %d, comm: %.20s %s %s %.*s\n",
318 current->pid, current->comm, print_tainted(),
319 init_utsname()->release,
320 (int)strcspn(init_utsname()->version, " "),
321 init_utsname()->version);
322 printk("RIP: %04lx:[<%016lx>] ", regs->cs & 0xffff, regs->rip);
323 printk_address(regs->rip);
324 printk("RSP: %04lx:%016lx EFLAGS: %08lx\n", regs->ss, regs->rsp,
326 printk("RAX: %016lx RBX: %016lx RCX: %016lx\n",
327 regs->rax, regs->rbx, regs->rcx);
328 printk("RDX: %016lx RSI: %016lx RDI: %016lx\n",
329 regs->rdx, regs->rsi, regs->rdi);
330 printk("RBP: %016lx R08: %016lx R09: %016lx\n",
331 regs->rbp, regs->r8, regs->r9);
332 printk("R10: %016lx R11: %016lx R12: %016lx\n",
333 regs->r10, regs->r11, regs->r12);
334 printk("R13: %016lx R14: %016lx R15: %016lx\n",
335 regs->r13, regs->r14, regs->r15);
337 asm("movl %%ds,%0" : "=r" (ds));
338 asm("movl %%cs,%0" : "=r" (cs));
339 asm("movl %%es,%0" : "=r" (es));
340 asm("movl %%fs,%0" : "=r" (fsindex));
341 asm("movl %%gs,%0" : "=r" (gsindex));
343 rdmsrl(MSR_FS_BASE, fs);
344 rdmsrl(MSR_GS_BASE, gs);
345 rdmsrl(MSR_KERNEL_GS_BASE, shadowgs);
352 printk("FS: %016lx(%04x) GS:%016lx(%04x) knlGS:%016lx\n",
353 fs,fsindex,gs,gsindex,shadowgs);
354 printk("CS: %04x DS: %04x ES: %04x CR0: %016lx\n", cs, ds, es, cr0);
355 printk("CR2: %016lx CR3: %016lx CR4: %016lx\n", cr2, cr3, cr4);
360 printk("DR0: %016lx DR1: %016lx DR2: %016lx\n", d0, d1, d2);
364 printk("DR3: %016lx DR6: %016lx DR7: %016lx\n", d3, d6, d7);
367 void show_regs(struct pt_regs *regs)
369 printk("CPU %d:", smp_processor_id());
371 show_trace(NULL, regs, (void *)(regs + 1));
375 * Free current thread data structures etc..
377 void exit_thread(void)
379 struct task_struct *me = current;
380 struct thread_struct *t = &me->thread;
382 if (me->thread.io_bitmap_ptr) {
383 struct tss_struct *tss = &per_cpu(init_tss, get_cpu());
385 kfree(t->io_bitmap_ptr);
386 t->io_bitmap_ptr = NULL;
387 clear_thread_flag(TIF_IO_BITMAP);
389 * Careful, clear this in the TSS too:
391 memset(tss->io_bitmap, 0xff, t->io_bitmap_max);
392 t->io_bitmap_max = 0;
397 void flush_thread(void)
399 struct task_struct *tsk = current;
401 if (test_tsk_thread_flag(tsk, TIF_ABI_PENDING)) {
402 clear_tsk_thread_flag(tsk, TIF_ABI_PENDING);
403 if (test_tsk_thread_flag(tsk, TIF_IA32)) {
404 clear_tsk_thread_flag(tsk, TIF_IA32);
406 set_tsk_thread_flag(tsk, TIF_IA32);
407 current_thread_info()->status |= TS_COMPAT;
410 clear_tsk_thread_flag(tsk, TIF_DEBUG);
412 tsk->thread.debugreg0 = 0;
413 tsk->thread.debugreg1 = 0;
414 tsk->thread.debugreg2 = 0;
415 tsk->thread.debugreg3 = 0;
416 tsk->thread.debugreg6 = 0;
417 tsk->thread.debugreg7 = 0;
418 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
420 * Forget coprocessor state..
426 void release_thread(struct task_struct *dead_task)
429 if (dead_task->mm->context.size) {
430 printk("WARNING: dead process %8s still has LDT? <%p/%d>\n",
432 dead_task->mm->context.ldt,
433 dead_task->mm->context.size);
439 static inline void set_32bit_tls(struct task_struct *t, int tls, u32 addr)
441 struct user_desc ud = {
448 struct n_desc_struct *desc = (void *)t->thread.tls_array;
450 desc->a = LDT_entry_a(&ud);
451 desc->b = LDT_entry_b(&ud);
454 static inline u32 read_32bit_tls(struct task_struct *t, int tls)
456 struct desc_struct *desc = (void *)t->thread.tls_array;
459 (((u32)desc->base1) << 16) |
460 (((u32)desc->base2) << 24);
464 * This gets called before we allocate a new thread and copy
465 * the current task into it.
467 void prepare_to_copy(struct task_struct *tsk)
472 int copy_thread(int nr, unsigned long clone_flags, unsigned long rsp,
473 unsigned long unused,
474 struct task_struct * p, struct pt_regs * regs)
477 struct pt_regs * childregs;
478 struct task_struct *me = current;
480 childregs = ((struct pt_regs *)
481 (THREAD_SIZE + task_stack_page(p))) - 1;
485 childregs->rsp = rsp;
487 childregs->rsp = (unsigned long)childregs;
489 p->thread.rsp = (unsigned long) childregs;
490 p->thread.rsp0 = (unsigned long) (childregs+1);
491 p->thread.userrsp = me->thread.userrsp;
493 set_tsk_thread_flag(p, TIF_FORK);
495 p->thread.fs = me->thread.fs;
496 p->thread.gs = me->thread.gs;
498 asm("mov %%gs,%0" : "=m" (p->thread.gsindex));
499 asm("mov %%fs,%0" : "=m" (p->thread.fsindex));
500 asm("mov %%es,%0" : "=m" (p->thread.es));
501 asm("mov %%ds,%0" : "=m" (p->thread.ds));
503 if (unlikely(test_tsk_thread_flag(me, TIF_IO_BITMAP))) {
504 p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL);
505 if (!p->thread.io_bitmap_ptr) {
506 p->thread.io_bitmap_max = 0;
509 memcpy(p->thread.io_bitmap_ptr, me->thread.io_bitmap_ptr,
511 set_tsk_thread_flag(p, TIF_IO_BITMAP);
515 * Set a new TLS for the child thread?
517 if (clone_flags & CLONE_SETTLS) {
518 #ifdef CONFIG_IA32_EMULATION
519 if (test_thread_flag(TIF_IA32))
520 err = ia32_child_tls(p, childregs);
523 err = do_arch_prctl(p, ARCH_SET_FS, childregs->r8);
529 if (err && p->thread.io_bitmap_ptr) {
530 kfree(p->thread.io_bitmap_ptr);
531 p->thread.io_bitmap_max = 0;
537 * This special macro can be used to load a debugging register
539 #define loaddebug(thread,r) set_debugreg(thread->debugreg ## r, r)
541 static inline void __switch_to_xtra(struct task_struct *prev_p,
542 struct task_struct *next_p,
543 struct tss_struct *tss)
545 struct thread_struct *prev, *next;
547 prev = &prev_p->thread,
548 next = &next_p->thread;
550 if (test_tsk_thread_flag(next_p, TIF_DEBUG)) {
560 if (test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
562 * Copy the relevant range of the IO bitmap.
563 * Normally this is 128 bytes or less:
565 memcpy(tss->io_bitmap, next->io_bitmap_ptr,
566 max(prev->io_bitmap_max, next->io_bitmap_max));
567 } else if (test_tsk_thread_flag(prev_p, TIF_IO_BITMAP)) {
569 * Clear any possible leftover bits:
571 memset(tss->io_bitmap, 0xff, prev->io_bitmap_max);
576 * switch_to(x,y) should switch tasks from x to y.
578 * This could still be optimized:
579 * - fold all the options into a flag word and test it with a single test.
580 * - could test fs/gs bitsliced
582 * Kprobes not supported here. Set the probe on schedule instead.
584 __kprobes struct task_struct *
585 __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
587 struct thread_struct *prev = &prev_p->thread,
588 *next = &next_p->thread;
589 int cpu = smp_processor_id();
590 struct tss_struct *tss = &per_cpu(init_tss, cpu);
592 /* we're going to use this soon, after a few expensive things */
593 if (next_p->fpu_counter>5)
594 prefetch(&next->i387.fxsave);
597 * Reload esp0, LDT and the page table pointer:
599 tss->rsp0 = next->rsp0;
603 * This won't pick up thread selector changes, but I guess that is ok.
605 asm volatile("mov %%es,%0" : "=m" (prev->es));
606 if (unlikely(next->es | prev->es))
607 loadsegment(es, next->es);
609 asm volatile ("mov %%ds,%0" : "=m" (prev->ds));
610 if (unlikely(next->ds | prev->ds))
611 loadsegment(ds, next->ds);
620 asm volatile("movl %%fs,%0" : "=r" (fsindex));
621 /* segment register != 0 always requires a reload.
622 also reload when it has changed.
623 when prev process used 64bit base always reload
624 to avoid an information leak. */
625 if (unlikely(fsindex | next->fsindex | prev->fs)) {
626 loadsegment(fs, next->fsindex);
627 /* check if the user used a selector != 0
628 * if yes clear 64bit base, since overloaded base
629 * is always mapped to the Null selector
634 /* when next process has a 64bit base use it */
636 wrmsrl(MSR_FS_BASE, next->fs);
637 prev->fsindex = fsindex;
641 asm volatile("movl %%gs,%0" : "=r" (gsindex));
642 if (unlikely(gsindex | next->gsindex | prev->gs)) {
643 load_gs_index(next->gsindex);
648 wrmsrl(MSR_KERNEL_GS_BASE, next->gs);
649 prev->gsindex = gsindex;
652 /* Must be after DS reload */
656 * Switch the PDA and FPU contexts.
658 prev->userrsp = read_pda(oldrsp);
659 write_pda(oldrsp, next->userrsp);
660 write_pda(pcurrent, next_p);
662 write_pda(kernelstack,
663 (unsigned long)task_stack_page(next_p) + THREAD_SIZE - PDA_STACKOFFSET);
664 #ifdef CONFIG_CC_STACKPROTECTOR
665 write_pda(stack_canary, next_p->stack_canary);
667 * Build time only check to make sure the stack_canary is at
668 * offset 40 in the pda; this is a gcc ABI requirement
670 BUILD_BUG_ON(offsetof(struct x8664_pda, stack_canary) != 40);
674 * Now maybe reload the debug registers and handle I/O bitmaps
676 if (unlikely((task_thread_info(next_p)->flags & _TIF_WORK_CTXSW))
677 || test_tsk_thread_flag(prev_p, TIF_IO_BITMAP))
678 __switch_to_xtra(prev_p, next_p, tss);
680 /* If the task has used fpu the last 5 timeslices, just do a full
681 * restore of the math state immediately to avoid the trap; the
682 * chances of needing FPU soon are obviously high now
684 if (next_p->fpu_counter>5)
685 math_state_restore();
690 * sys_execve() executes a new program.
693 long sys_execve(char __user *name, char __user * __user *argv,
694 char __user * __user *envp, struct pt_regs regs)
699 filename = getname(name);
700 error = PTR_ERR(filename);
701 if (IS_ERR(filename))
703 error = do_execve(filename, argv, envp, ®s);
706 current->ptrace &= ~PT_DTRACE;
707 task_unlock(current);
713 void set_personality_64bit(void)
715 /* inherit personality from parent */
717 /* Make sure to be in 64bit mode */
718 clear_thread_flag(TIF_IA32);
720 /* TBD: overwrites user setup. Should have two bits.
721 But 64bit processes have always behaved this way,
722 so it's not too bad. The main problem is just that
723 32bit childs are affected again. */
724 current->personality &= ~READ_IMPLIES_EXEC;
727 asmlinkage long sys_fork(struct pt_regs *regs)
729 return do_fork(SIGCHLD, regs->rsp, regs, 0, NULL, NULL);
733 sys_clone(unsigned long clone_flags, unsigned long newsp,
734 void __user *parent_tid, void __user *child_tid, struct pt_regs *regs)
738 return do_fork(clone_flags, newsp, regs, 0, parent_tid, child_tid);
742 * This is trivial, and on the face of it looks like it
743 * could equally well be done in user mode.
745 * Not so, for quite unobvious reasons - register pressure.
746 * In user mode vfork() cannot have a stack frame, and if
747 * done by calling the "clone()" system call directly, you
748 * do not have enough call-clobbered registers to hold all
749 * the information you need.
751 asmlinkage long sys_vfork(struct pt_regs *regs)
753 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->rsp, regs, 0,
757 unsigned long get_wchan(struct task_struct *p)
763 if (!p || p == current || p->state==TASK_RUNNING)
765 stack = (unsigned long)task_stack_page(p);
766 if (p->thread.rsp < stack || p->thread.rsp > stack+THREAD_SIZE)
768 fp = *(u64 *)(p->thread.rsp);
770 if (fp < (unsigned long)stack ||
771 fp > (unsigned long)stack+THREAD_SIZE)
773 rip = *(u64 *)(fp+8);
774 if (!in_sched_functions(rip))
777 } while (count++ < 16);
781 long do_arch_prctl(struct task_struct *task, int code, unsigned long addr)
784 int doit = task == current;
789 if (addr >= TASK_SIZE_OF(task))
792 /* handle small bases via the GDT because that's faster to
794 if (addr <= 0xffffffff) {
795 set_32bit_tls(task, GS_TLS, addr);
797 load_TLS(&task->thread, cpu);
798 load_gs_index(GS_TLS_SEL);
800 task->thread.gsindex = GS_TLS_SEL;
803 task->thread.gsindex = 0;
804 task->thread.gs = addr;
807 ret = checking_wrmsrl(MSR_KERNEL_GS_BASE, addr);
813 /* Not strictly needed for fs, but do it for symmetry
815 if (addr >= TASK_SIZE_OF(task))
818 /* handle small bases via the GDT because that's faster to
820 if (addr <= 0xffffffff) {
821 set_32bit_tls(task, FS_TLS, addr);
823 load_TLS(&task->thread, cpu);
824 asm volatile("movl %0,%%fs" :: "r"(FS_TLS_SEL));
826 task->thread.fsindex = FS_TLS_SEL;
829 task->thread.fsindex = 0;
830 task->thread.fs = addr;
832 /* set the selector to 0 to not confuse
834 asm volatile("movl %0,%%fs" :: "r" (0));
835 ret = checking_wrmsrl(MSR_FS_BASE, addr);
842 if (task->thread.fsindex == FS_TLS_SEL)
843 base = read_32bit_tls(task, FS_TLS);
845 rdmsrl(MSR_FS_BASE, base);
847 base = task->thread.fs;
848 ret = put_user(base, (unsigned long __user *)addr);
854 if (task->thread.gsindex == GS_TLS_SEL)
855 base = read_32bit_tls(task, GS_TLS);
857 asm("movl %%gs,%0" : "=r" (gsindex));
859 rdmsrl(MSR_KERNEL_GS_BASE, base);
861 base = task->thread.gs;
864 base = task->thread.gs;
865 ret = put_user(base, (unsigned long __user *)addr);
877 long sys_arch_prctl(int code, unsigned long addr)
879 return do_arch_prctl(current, code, addr);
883 * Capture the user space registers if the task is not running (in user space)
885 int dump_task_regs(struct task_struct *tsk, elf_gregset_t *regs)
887 struct pt_regs *pp, ptregs;
889 pp = task_pt_regs(tsk);
895 elf_core_copy_regs(regs, &ptregs);
900 unsigned long arch_align_stack(unsigned long sp)
902 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
903 sp -= get_random_int() % 8192;