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>
23 #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/interrupt.h>
30 #include <linux/utsname.h>
31 #include <linux/delay.h>
32 #include <linux/module.h>
33 #include <linux/ptrace.h>
34 #include <linux/random.h>
35 #include <linux/notifier.h>
36 #include <linux/kprobes.h>
37 #include <linux/kdebug.h>
38 #include <linux/tick.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>
50 #include <asm/proto.h>
54 asmlinkage extern void ret_from_fork(void);
56 unsigned long kernel_thread_flags = CLONE_VM | CLONE_UNTRACED;
58 unsigned long boot_option_idle_override = 0;
59 EXPORT_SYMBOL(boot_option_idle_override);
62 * Powermanagement idle function, if any..
64 void (*pm_idle)(void);
65 EXPORT_SYMBOL(pm_idle);
67 static ATOMIC_NOTIFIER_HEAD(idle_notifier);
69 void idle_notifier_register(struct notifier_block *n)
71 atomic_notifier_chain_register(&idle_notifier, n);
77 atomic_notifier_call_chain(&idle_notifier, IDLE_START, NULL);
80 static void __exit_idle(void)
82 if (test_and_clear_bit_pda(0, isidle) == 0)
84 atomic_notifier_call_chain(&idle_notifier, IDLE_END, NULL);
87 /* Called from interrupts to signify idle end */
90 /* idle loop has pid 0 */
97 * We use this if we don't have any better
100 void default_idle(void)
102 current_thread_info()->status &= ~TS_POLLING;
104 * TS_POLLING-cleared state must be visible before we
109 if (!need_resched()) {
110 safe_halt(); /* enables interrupts racelessly */
114 current_thread_info()->status |= TS_POLLING;
118 * On SMP it's slightly faster (but much more power-consuming!)
119 * to poll the ->need_resched flag instead of waiting for the
120 * cross-CPU IPI to arrive. Use this option with caution.
122 static void poll_idle(void)
128 #ifdef CONFIG_HOTPLUG_CPU
129 DECLARE_PER_CPU(int, cpu_state);
132 /* We halt the CPU with physical CPU hotplug */
133 static inline void play_dead(void)
139 __get_cpu_var(cpu_state) = CPU_DEAD;
146 static inline void play_dead(void)
150 #endif /* CONFIG_HOTPLUG_CPU */
153 * The idle thread. There's no useful work to be
154 * done, so just try to conserve power and have a
155 * low exit latency (ie sit in a loop waiting for
156 * somebody to say that they'd like to reschedule)
160 current_thread_info()->status |= TS_POLLING;
161 /* endless idle loop with no priority at all */
163 tick_nohz_stop_sched_tick();
164 while (!need_resched()) {
171 if (cpu_is_offline(smp_processor_id()))
174 * Idle routines should keep interrupts disabled
175 * from here on, until they go to idle.
176 * Otherwise, idle callbacks can misfire.
181 /* In many cases the interrupt that ended idle
182 has already called exit_idle. But some idle
183 loops can be woken up without interrupt. */
187 tick_nohz_restart_sched_tick();
188 preempt_enable_no_resched();
194 static void do_nothing(void *unused)
199 * cpu_idle_wait - Used to ensure that all the CPUs discard old value of
200 * pm_idle and update to new pm_idle value. Required while changing pm_idle
201 * handler on SMP systems.
203 * Caller must have changed pm_idle to the new value before the call. Old
204 * pm_idle value will not be used by any CPU after the return of this function.
206 void cpu_idle_wait(void)
209 /* kick all the CPUs so that they exit out of pm_idle */
210 smp_call_function(do_nothing, NULL, 0, 1);
212 EXPORT_SYMBOL_GPL(cpu_idle_wait);
215 * This uses new MONITOR/MWAIT instructions on P4 processors with PNI,
216 * which can obviate IPI to trigger checking of need_resched.
217 * We execute MONITOR against need_resched and enter optimized wait state
218 * through MWAIT. Whenever someone changes need_resched, we would be woken
219 * up from MWAIT (without an IPI).
221 * New with Core Duo processors, MWAIT can take some hints based on CPU
224 void mwait_idle_with_hints(unsigned long ax, unsigned long cx)
226 if (!need_resched()) {
227 __monitor((void *)¤t_thread_info()->flags, 0, 0);
234 /* Default MONITOR/MWAIT with no hints, used for default C1 state */
235 static void mwait_idle(void)
237 if (!need_resched()) {
238 __monitor((void *)¤t_thread_info()->flags, 0, 0);
250 static int __cpuinit mwait_usable(const struct cpuinfo_x86 *c)
254 /* Any C1 states supported? */
255 return c->cpuid_level >= 5 && ((cpuid_edx(5) >> 4) & 0xf) > 0;
258 void __cpuinit select_idle_routine(const struct cpuinfo_x86 *c)
264 #ifdef CONFIG_X86_SMP
265 if (pm_idle == poll_idle && smp_num_siblings > 1) {
266 printk(KERN_WARNING "WARNING: polling idle and HT enabled,"
267 " performance may degrade.\n");
270 if (cpu_has(c, X86_FEATURE_MWAIT) && mwait_usable(c)) {
272 * Skip, if setup has overridden idle.
273 * One CPU supports mwait => All CPUs supports mwait
276 printk(KERN_INFO "using mwait in idle threads.\n");
277 pm_idle = mwait_idle;
283 static int __init idle_setup(char *str)
285 if (!strcmp(str, "poll")) {
286 printk("using polling idle threads.\n");
288 } else if (!strcmp(str, "mwait"))
293 boot_option_idle_override = 1;
296 early_param("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 long d0, d1, d2, d3, d6, d7;
303 unsigned int fsindex, gsindex;
304 unsigned int ds, cs, es;
308 printk("Pid: %d, comm: %.20s %s %s %.*s\n",
309 current->pid, current->comm, print_tainted(),
310 init_utsname()->release,
311 (int)strcspn(init_utsname()->version, " "),
312 init_utsname()->version);
313 printk("RIP: %04lx:[<%016lx>] ", regs->cs & 0xffff, regs->ip);
314 printk_address(regs->ip, 1);
315 printk("RSP: %04lx:%016lx EFLAGS: %08lx\n", regs->ss, regs->sp,
317 printk("RAX: %016lx RBX: %016lx RCX: %016lx\n",
318 regs->ax, regs->bx, regs->cx);
319 printk("RDX: %016lx RSI: %016lx RDI: %016lx\n",
320 regs->dx, regs->si, regs->di);
321 printk("RBP: %016lx R08: %016lx R09: %016lx\n",
322 regs->bp, regs->r8, regs->r9);
323 printk("R10: %016lx R11: %016lx R12: %016lx\n",
324 regs->r10, regs->r11, regs->r12);
325 printk("R13: %016lx R14: %016lx R15: %016lx\n",
326 regs->r13, regs->r14, regs->r15);
328 asm("movl %%ds,%0" : "=r" (ds));
329 asm("movl %%cs,%0" : "=r" (cs));
330 asm("movl %%es,%0" : "=r" (es));
331 asm("movl %%fs,%0" : "=r" (fsindex));
332 asm("movl %%gs,%0" : "=r" (gsindex));
334 rdmsrl(MSR_FS_BASE, fs);
335 rdmsrl(MSR_GS_BASE, gs);
336 rdmsrl(MSR_KERNEL_GS_BASE, shadowgs);
343 printk("FS: %016lx(%04x) GS:%016lx(%04x) knlGS:%016lx\n",
344 fs,fsindex,gs,gsindex,shadowgs);
345 printk("CS: %04x DS: %04x ES: %04x CR0: %016lx\n", cs, ds, es, cr0);
346 printk("CR2: %016lx CR3: %016lx CR4: %016lx\n", cr2, cr3, cr4);
351 printk("DR0: %016lx DR1: %016lx DR2: %016lx\n", d0, d1, d2);
355 printk("DR3: %016lx DR6: %016lx DR7: %016lx\n", d3, d6, d7);
358 void show_regs(struct pt_regs *regs)
360 printk("CPU %d:", smp_processor_id());
362 show_trace(NULL, regs, (void *)(regs + 1), regs->bp);
366 * Free current thread data structures etc..
368 void exit_thread(void)
370 struct task_struct *me = current;
371 struct thread_struct *t = &me->thread;
373 if (me->thread.io_bitmap_ptr) {
374 struct tss_struct *tss = &per_cpu(init_tss, get_cpu());
376 kfree(t->io_bitmap_ptr);
377 t->io_bitmap_ptr = NULL;
378 clear_thread_flag(TIF_IO_BITMAP);
380 * Careful, clear this in the TSS too:
382 memset(tss->io_bitmap, 0xff, t->io_bitmap_max);
383 t->io_bitmap_max = 0;
388 void flush_thread(void)
390 struct task_struct *tsk = current;
392 if (test_tsk_thread_flag(tsk, TIF_ABI_PENDING)) {
393 clear_tsk_thread_flag(tsk, TIF_ABI_PENDING);
394 if (test_tsk_thread_flag(tsk, TIF_IA32)) {
395 clear_tsk_thread_flag(tsk, TIF_IA32);
397 set_tsk_thread_flag(tsk, TIF_IA32);
398 current_thread_info()->status |= TS_COMPAT;
401 clear_tsk_thread_flag(tsk, TIF_DEBUG);
403 tsk->thread.debugreg0 = 0;
404 tsk->thread.debugreg1 = 0;
405 tsk->thread.debugreg2 = 0;
406 tsk->thread.debugreg3 = 0;
407 tsk->thread.debugreg6 = 0;
408 tsk->thread.debugreg7 = 0;
409 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
411 * Forget coprocessor state..
417 void release_thread(struct task_struct *dead_task)
420 if (dead_task->mm->context.size) {
421 printk("WARNING: dead process %8s still has LDT? <%p/%d>\n",
423 dead_task->mm->context.ldt,
424 dead_task->mm->context.size);
430 static inline void set_32bit_tls(struct task_struct *t, int tls, u32 addr)
432 struct user_desc ud = {
439 struct desc_struct *desc = t->thread.tls_array;
444 static inline u32 read_32bit_tls(struct task_struct *t, int tls)
446 return get_desc_base(&t->thread.tls_array[tls]);
450 * This gets called before we allocate a new thread and copy
451 * the current task into it.
453 void prepare_to_copy(struct task_struct *tsk)
458 int copy_thread(int nr, unsigned long clone_flags, unsigned long sp,
459 unsigned long unused,
460 struct task_struct * p, struct pt_regs * regs)
463 struct pt_regs * childregs;
464 struct task_struct *me = current;
466 childregs = ((struct pt_regs *)
467 (THREAD_SIZE + task_stack_page(p))) - 1;
473 childregs->sp = (unsigned long)childregs;
475 p->thread.sp = (unsigned long) childregs;
476 p->thread.sp0 = (unsigned long) (childregs+1);
477 p->thread.usersp = me->thread.usersp;
479 set_tsk_thread_flag(p, TIF_FORK);
481 p->thread.fs = me->thread.fs;
482 p->thread.gs = me->thread.gs;
484 asm("mov %%gs,%0" : "=m" (p->thread.gsindex));
485 asm("mov %%fs,%0" : "=m" (p->thread.fsindex));
486 asm("mov %%es,%0" : "=m" (p->thread.es));
487 asm("mov %%ds,%0" : "=m" (p->thread.ds));
489 if (unlikely(test_tsk_thread_flag(me, TIF_IO_BITMAP))) {
490 p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL);
491 if (!p->thread.io_bitmap_ptr) {
492 p->thread.io_bitmap_max = 0;
495 memcpy(p->thread.io_bitmap_ptr, me->thread.io_bitmap_ptr,
497 set_tsk_thread_flag(p, TIF_IO_BITMAP);
501 * Set a new TLS for the child thread?
503 if (clone_flags & CLONE_SETTLS) {
504 #ifdef CONFIG_IA32_EMULATION
505 if (test_thread_flag(TIF_IA32))
506 err = do_set_thread_area(p, -1,
507 (struct user_desc __user *)childregs->si, 0);
510 err = do_arch_prctl(p, ARCH_SET_FS, childregs->r8);
516 if (err && p->thread.io_bitmap_ptr) {
517 kfree(p->thread.io_bitmap_ptr);
518 p->thread.io_bitmap_max = 0;
524 start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp)
526 asm volatile("movl %0, %%fs; movl %0, %%es; movl %0, %%ds" :: "r"(0));
530 write_pda(oldrsp, new_sp);
531 regs->cs = __USER_CS;
532 regs->ss = __USER_DS;
536 EXPORT_SYMBOL_GPL(start_thread);
539 * This special macro can be used to load a debugging register
541 #define loaddebug(thread, r) set_debugreg(thread->debugreg ## r, r)
543 static inline void __switch_to_xtra(struct task_struct *prev_p,
544 struct task_struct *next_p,
545 struct tss_struct *tss)
547 struct thread_struct *prev, *next;
548 unsigned long debugctl;
550 prev = &prev_p->thread,
551 next = &next_p->thread;
553 debugctl = prev->debugctlmsr;
554 if (next->ds_area_msr != prev->ds_area_msr) {
555 /* we clear debugctl to make sure DS
556 * is not in use when we change it */
558 update_debugctlmsr(0);
559 wrmsrl(MSR_IA32_DS_AREA, next->ds_area_msr);
562 if (next->debugctlmsr != debugctl)
563 update_debugctlmsr(next->debugctlmsr);
565 if (test_tsk_thread_flag(next_p, TIF_DEBUG)) {
575 if (test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
577 * Copy the relevant range of the IO bitmap.
578 * Normally this is 128 bytes or less:
580 memcpy(tss->io_bitmap, next->io_bitmap_ptr,
581 max(prev->io_bitmap_max, next->io_bitmap_max));
582 } else if (test_tsk_thread_flag(prev_p, TIF_IO_BITMAP)) {
584 * Clear any possible leftover bits:
586 memset(tss->io_bitmap, 0xff, prev->io_bitmap_max);
590 if (test_tsk_thread_flag(prev_p, TIF_BTS_TRACE_TS))
591 ptrace_bts_take_timestamp(prev_p, BTS_TASK_DEPARTS);
593 if (test_tsk_thread_flag(next_p, TIF_BTS_TRACE_TS))
594 ptrace_bts_take_timestamp(next_p, BTS_TASK_ARRIVES);
599 * switch_to(x,y) should switch tasks from x to y.
601 * This could still be optimized:
602 * - fold all the options into a flag word and test it with a single test.
603 * - could test fs/gs bitsliced
605 * Kprobes not supported here. Set the probe on schedule instead.
608 __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
610 struct thread_struct *prev = &prev_p->thread,
611 *next = &next_p->thread;
612 int cpu = smp_processor_id();
613 struct tss_struct *tss = &per_cpu(init_tss, cpu);
615 /* we're going to use this soon, after a few expensive things */
616 if (next_p->fpu_counter>5)
617 prefetch(&next->i387.fxsave);
620 * Reload esp0, LDT and the page table pointer:
626 * This won't pick up thread selector changes, but I guess that is ok.
628 asm volatile("mov %%es,%0" : "=m" (prev->es));
629 if (unlikely(next->es | prev->es))
630 loadsegment(es, next->es);
632 asm volatile ("mov %%ds,%0" : "=m" (prev->ds));
633 if (unlikely(next->ds | prev->ds))
634 loadsegment(ds, next->ds);
643 asm volatile("movl %%fs,%0" : "=r" (fsindex));
644 /* segment register != 0 always requires a reload.
645 also reload when it has changed.
646 when prev process used 64bit base always reload
647 to avoid an information leak. */
648 if (unlikely(fsindex | next->fsindex | prev->fs)) {
649 loadsegment(fs, next->fsindex);
650 /* check if the user used a selector != 0
651 * if yes clear 64bit base, since overloaded base
652 * is always mapped to the Null selector
657 /* when next process has a 64bit base use it */
659 wrmsrl(MSR_FS_BASE, next->fs);
660 prev->fsindex = fsindex;
664 asm volatile("movl %%gs,%0" : "=r" (gsindex));
665 if (unlikely(gsindex | next->gsindex | prev->gs)) {
666 load_gs_index(next->gsindex);
671 wrmsrl(MSR_KERNEL_GS_BASE, next->gs);
672 prev->gsindex = gsindex;
675 /* Must be after DS reload */
679 * Switch the PDA and FPU contexts.
681 prev->usersp = read_pda(oldrsp);
682 write_pda(oldrsp, next->usersp);
683 write_pda(pcurrent, next_p);
685 write_pda(kernelstack,
686 (unsigned long)task_stack_page(next_p) + THREAD_SIZE - PDA_STACKOFFSET);
687 #ifdef CONFIG_CC_STACKPROTECTOR
688 write_pda(stack_canary, next_p->stack_canary);
690 * Build time only check to make sure the stack_canary is at
691 * offset 40 in the pda; this is a gcc ABI requirement
693 BUILD_BUG_ON(offsetof(struct x8664_pda, stack_canary) != 40);
697 * Now maybe reload the debug registers and handle I/O bitmaps
699 if (unlikely(task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT ||
700 task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV))
701 __switch_to_xtra(prev_p, next_p, tss);
703 /* If the task has used fpu the last 5 timeslices, just do a full
704 * restore of the math state immediately to avoid the trap; the
705 * chances of needing FPU soon are obviously high now
707 if (next_p->fpu_counter>5)
708 math_state_restore();
713 * sys_execve() executes a new program.
716 long sys_execve(char __user *name, char __user * __user *argv,
717 char __user * __user *envp, struct pt_regs *regs)
722 filename = getname(name);
723 error = PTR_ERR(filename);
724 if (IS_ERR(filename))
726 error = do_execve(filename, argv, envp, regs);
731 void set_personality_64bit(void)
733 /* inherit personality from parent */
735 /* Make sure to be in 64bit mode */
736 clear_thread_flag(TIF_IA32);
738 /* TBD: overwrites user setup. Should have two bits.
739 But 64bit processes have always behaved this way,
740 so it's not too bad. The main problem is just that
741 32bit childs are affected again. */
742 current->personality &= ~READ_IMPLIES_EXEC;
745 asmlinkage long sys_fork(struct pt_regs *regs)
747 return do_fork(SIGCHLD, regs->sp, regs, 0, NULL, NULL);
751 sys_clone(unsigned long clone_flags, unsigned long newsp,
752 void __user *parent_tid, void __user *child_tid, struct pt_regs *regs)
756 return do_fork(clone_flags, newsp, regs, 0, parent_tid, child_tid);
760 * This is trivial, and on the face of it looks like it
761 * could equally well be done in user mode.
763 * Not so, for quite unobvious reasons - register pressure.
764 * In user mode vfork() cannot have a stack frame, and if
765 * done by calling the "clone()" system call directly, you
766 * do not have enough call-clobbered registers to hold all
767 * the information you need.
769 asmlinkage long sys_vfork(struct pt_regs *regs)
771 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->sp, regs, 0,
775 unsigned long get_wchan(struct task_struct *p)
781 if (!p || p == current || p->state==TASK_RUNNING)
783 stack = (unsigned long)task_stack_page(p);
784 if (p->thread.sp < stack || p->thread.sp > stack+THREAD_SIZE)
786 fp = *(u64 *)(p->thread.sp);
788 if (fp < (unsigned long)stack ||
789 fp > (unsigned long)stack+THREAD_SIZE)
792 if (!in_sched_functions(ip))
795 } while (count++ < 16);
799 long do_arch_prctl(struct task_struct *task, int code, unsigned long addr)
802 int doit = task == current;
807 if (addr >= TASK_SIZE_OF(task))
810 /* handle small bases via the GDT because that's faster to
812 if (addr <= 0xffffffff) {
813 set_32bit_tls(task, GS_TLS, addr);
815 load_TLS(&task->thread, cpu);
816 load_gs_index(GS_TLS_SEL);
818 task->thread.gsindex = GS_TLS_SEL;
821 task->thread.gsindex = 0;
822 task->thread.gs = addr;
825 ret = checking_wrmsrl(MSR_KERNEL_GS_BASE, addr);
831 /* Not strictly needed for fs, but do it for symmetry
833 if (addr >= TASK_SIZE_OF(task))
836 /* handle small bases via the GDT because that's faster to
838 if (addr <= 0xffffffff) {
839 set_32bit_tls(task, FS_TLS, addr);
841 load_TLS(&task->thread, cpu);
842 asm volatile("movl %0,%%fs" :: "r"(FS_TLS_SEL));
844 task->thread.fsindex = FS_TLS_SEL;
847 task->thread.fsindex = 0;
848 task->thread.fs = addr;
850 /* set the selector to 0 to not confuse
852 asm volatile("movl %0,%%fs" :: "r" (0));
853 ret = checking_wrmsrl(MSR_FS_BASE, addr);
860 if (task->thread.fsindex == FS_TLS_SEL)
861 base = read_32bit_tls(task, FS_TLS);
863 rdmsrl(MSR_FS_BASE, base);
865 base = task->thread.fs;
866 ret = put_user(base, (unsigned long __user *)addr);
872 if (task->thread.gsindex == GS_TLS_SEL)
873 base = read_32bit_tls(task, GS_TLS);
875 asm("movl %%gs,%0" : "=r" (gsindex));
877 rdmsrl(MSR_KERNEL_GS_BASE, base);
879 base = task->thread.gs;
882 base = task->thread.gs;
883 ret = put_user(base, (unsigned long __user *)addr);
895 long sys_arch_prctl(int code, unsigned long addr)
897 return do_arch_prctl(current, code, addr);
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;
907 unsigned long arch_randomize_brk(struct mm_struct *mm)
909 unsigned long range_end = mm->brk + 0x02000000;
910 return randomize_range(mm->brk, range_end, 0) ? : mm->brk;