1
2
3
4
5
6
7
8
9
10
11
12#include <stdarg.h>
13
14#include <linux/cpu.h>
15#include <linux/errno.h>
16#include <linux/sched.h>
17#include <linux/fs.h>
18#include <linux/kernel.h>
19#include <linux/mm.h>
20#include <linux/elfcore.h>
21#include <linux/smp.h>
22#include <linux/stddef.h>
23#include <linux/slab.h>
24#include <linux/vmalloc.h>
25#include <linux/user.h>
26#include <linux/interrupt.h>
27#include <linux/utsname.h>
28#include <linux/delay.h>
29#include <linux/reboot.h>
30#include <linux/init.h>
31#include <linux/mc146818rtc.h>
32#include <linux/module.h>
33#include <linux/kallsyms.h>
34#include <linux/ptrace.h>
35#include <linux/random.h>
36#include <linux/personality.h>
37#include <linux/tick.h>
38#include <linux/percpu.h>
39
40#include <asm/uaccess.h>
41#include <asm/pgtable.h>
42#include <asm/system.h>
43#include <asm/io.h>
44#include <asm/ldt.h>
45#include <asm/processor.h>
46#include <asm/i387.h>
47#include <asm/desc.h>
48#include <asm/vm86.h>
49#ifdef CONFIG_MATH_EMULATION
50#include <asm/math_emu.h>
51#endif
52
53#include <linux/err.h>
54
55#include <asm/tlbflush.h>
56#include <asm/cpu.h>
57#include <asm/kdebug.h>
58
59asmlinkage void ret_from_fork(void) __asm__("ret_from_fork");
60
61static int hlt_counter;
62
63unsigned long boot_option_idle_override = 0;
64EXPORT_SYMBOL(boot_option_idle_override);
65
66DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task;
67EXPORT_PER_CPU_SYMBOL(current_task);
68
69DEFINE_PER_CPU(int, cpu_number);
70EXPORT_PER_CPU_SYMBOL(cpu_number);
71
72
73
74
75unsigned long thread_saved_pc(struct task_struct *tsk)
76{
77 return ((unsigned long *)tsk->thread.sp)[3];
78}
79
80
81
82
83void (*pm_idle)(void);
84EXPORT_SYMBOL(pm_idle);
85
86void disable_hlt(void)
87{
88 hlt_counter++;
89}
90
91EXPORT_SYMBOL(disable_hlt);
92
93void enable_hlt(void)
94{
95 hlt_counter--;
96}
97
98EXPORT_SYMBOL(enable_hlt);
99
100
101
102
103
104void default_idle(void)
105{
106 if (!hlt_counter && boot_cpu_data.hlt_works_ok) {
107 current_thread_info()->status &= ~TS_POLLING;
108
109
110
111
112 smp_mb();
113
114 local_irq_disable();
115 if (!need_resched()) {
116 ktime_t t0, t1;
117 u64 t0n, t1n;
118
119 t0 = ktime_get();
120 t0n = ktime_to_ns(t0);
121 safe_halt();
122 local_irq_disable();
123 t1 = ktime_get();
124 t1n = ktime_to_ns(t1);
125 sched_clock_idle_wakeup_event(t1n - t0n);
126 }
127 local_irq_enable();
128 current_thread_info()->status |= TS_POLLING;
129 } else {
130
131 cpu_relax();
132 }
133}
134#ifdef CONFIG_APM_MODULE
135EXPORT_SYMBOL(default_idle);
136#endif
137
138
139
140
141
142
143static void poll_idle(void)
144{
145 cpu_relax();
146}
147
148#ifdef CONFIG_HOTPLUG_CPU
149#include <asm/nmi.h>
150
151static inline void play_dead(void)
152{
153
154 cpu_exit_clear();
155 wbinvd();
156 mb();
157
158 __get_cpu_var(cpu_state) = CPU_DEAD;
159
160
161
162
163 local_irq_disable();
164 while (1)
165 halt();
166}
167#else
168static inline void play_dead(void)
169{
170 BUG();
171}
172#endif
173
174
175
176
177
178
179
180void cpu_idle(void)
181{
182 int cpu = smp_processor_id();
183
184 current_thread_info()->status |= TS_POLLING;
185
186
187 while (1) {
188 tick_nohz_stop_sched_tick();
189 while (!need_resched()) {
190 void (*idle)(void);
191
192 check_pgt_cache();
193 rmb();
194 idle = pm_idle;
195
196 if (rcu_pending(cpu))
197 rcu_check_callbacks(cpu, 0);
198
199 if (!idle)
200 idle = default_idle;
201
202 if (cpu_is_offline(cpu))
203 play_dead();
204
205 __get_cpu_var(irq_stat).idle_timestamp = jiffies;
206 idle();
207 }
208 tick_nohz_restart_sched_tick();
209 preempt_enable_no_resched();
210 schedule();
211 preempt_disable();
212 }
213}
214
215static void do_nothing(void *unused)
216{
217}
218
219
220
221
222
223
224
225
226
227void cpu_idle_wait(void)
228{
229 smp_mb();
230
231 smp_call_function(do_nothing, NULL, 0, 1);
232}
233EXPORT_SYMBOL_GPL(cpu_idle_wait);
234
235
236
237
238
239
240
241
242
243
244
245void mwait_idle_with_hints(unsigned long ax, unsigned long cx)
246{
247 if (!need_resched()) {
248 __monitor((void *)¤t_thread_info()->flags, 0, 0);
249 smp_mb();
250 if (!need_resched())
251 __mwait(ax, cx);
252 }
253}
254
255
256static void mwait_idle(void)
257{
258 local_irq_enable();
259 mwait_idle_with_hints(0, 0);
260}
261
262static int __cpuinit mwait_usable(const struct cpuinfo_x86 *c)
263{
264 if (force_mwait)
265 return 1;
266
267 return c->cpuid_level >= 5 && ((cpuid_edx(5) >> 4) & 0xf) > 0;
268}
269
270void __cpuinit select_idle_routine(const struct cpuinfo_x86 *c)
271{
272 static int selected;
273
274 if (selected)
275 return;
276#ifdef CONFIG_X86_SMP
277 if (pm_idle == poll_idle && smp_num_siblings > 1) {
278 printk(KERN_WARNING "WARNING: polling idle and HT enabled,"
279 " performance may degrade.\n");
280 }
281#endif
282 if (cpu_has(c, X86_FEATURE_MWAIT) && mwait_usable(c)) {
283
284
285
286
287 if (!pm_idle) {
288 printk(KERN_INFO "using mwait in idle threads.\n");
289 pm_idle = mwait_idle;
290 }
291 }
292 selected = 1;
293}
294
295static int __init idle_setup(char *str)
296{
297 if (!strcmp(str, "poll")) {
298 printk("using polling idle threads.\n");
299 pm_idle = poll_idle;
300 } else if (!strcmp(str, "mwait"))
301 force_mwait = 1;
302 else
303 return -1;
304
305 boot_option_idle_override = 1;
306 return 0;
307}
308early_param("idle", idle_setup);
309
310void __show_registers(struct pt_regs *regs, int all)
311{
312 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L;
313 unsigned long d0, d1, d2, d3, d6, d7;
314 unsigned long sp;
315 unsigned short ss, gs;
316
317 if (user_mode_vm(regs)) {
318 sp = regs->sp;
319 ss = regs->ss & 0xffff;
320 savesegment(gs, gs);
321 } else {
322 sp = (unsigned long) (®s->sp);
323 savesegment(ss, ss);
324 savesegment(gs, gs);
325 }
326
327 printk("\n");
328 printk("Pid: %d, comm: %s %s (%s %.*s)\n",
329 task_pid_nr(current), current->comm,
330 print_tainted(), init_utsname()->release,
331 (int)strcspn(init_utsname()->version, " "),
332 init_utsname()->version);
333
334 printk("EIP: %04x:[<%08lx>] EFLAGS: %08lx CPU: %d\n",
335 0xffff & regs->cs, regs->ip, regs->flags,
336 smp_processor_id());
337 print_symbol("EIP is at %s\n", regs->ip);
338
339 printk("EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
340 regs->ax, regs->bx, regs->cx, regs->dx);
341 printk("ESI: %08lx EDI: %08lx EBP: %08lx ESP: %08lx\n",
342 regs->si, regs->di, regs->bp, sp);
343 printk(" DS: %04x ES: %04x FS: %04x GS: %04x SS: %04x\n",
344 regs->ds & 0xffff, regs->es & 0xffff,
345 regs->fs & 0xffff, gs, ss);
346
347 if (!all)
348 return;
349
350 cr0 = read_cr0();
351 cr2 = read_cr2();
352 cr3 = read_cr3();
353 cr4 = read_cr4_safe();
354 printk("CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n",
355 cr0, cr2, cr3, cr4);
356
357 get_debugreg(d0, 0);
358 get_debugreg(d1, 1);
359 get_debugreg(d2, 2);
360 get_debugreg(d3, 3);
361 printk("DR0: %08lx DR1: %08lx DR2: %08lx DR3: %08lx\n",
362 d0, d1, d2, d3);
363
364 get_debugreg(d6, 6);
365 get_debugreg(d7, 7);
366 printk("DR6: %08lx DR7: %08lx\n",
367 d6, d7);
368}
369
370void show_regs(struct pt_regs *regs)
371{
372 __show_registers(regs, 1);
373 show_trace(NULL, regs, ®s->sp, regs->bp);
374}
375
376
377
378
379
380
381extern void kernel_thread_helper(void);
382
383
384
385
386int kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
387{
388 struct pt_regs regs;
389
390 memset(®s, 0, sizeof(regs));
391
392 regs.bx = (unsigned long) fn;
393 regs.dx = (unsigned long) arg;
394
395 regs.ds = __USER_DS;
396 regs.es = __USER_DS;
397 regs.fs = __KERNEL_PERCPU;
398 regs.orig_ax = -1;
399 regs.ip = (unsigned long) kernel_thread_helper;
400 regs.cs = __KERNEL_CS | get_kernel_rpl();
401 regs.flags = X86_EFLAGS_IF | X86_EFLAGS_SF | X86_EFLAGS_PF | 0x2;
402
403
404 return do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0, ®s, 0, NULL, NULL);
405}
406EXPORT_SYMBOL(kernel_thread);
407
408
409
410
411void exit_thread(void)
412{
413
414 if (unlikely(test_thread_flag(TIF_IO_BITMAP))) {
415 struct task_struct *tsk = current;
416 struct thread_struct *t = &tsk->thread;
417 int cpu = get_cpu();
418 struct tss_struct *tss = &per_cpu(init_tss, cpu);
419
420 kfree(t->io_bitmap_ptr);
421 t->io_bitmap_ptr = NULL;
422 clear_thread_flag(TIF_IO_BITMAP);
423
424
425
426 memset(tss->io_bitmap, 0xff, tss->io_bitmap_max);
427 t->io_bitmap_max = 0;
428 tss->io_bitmap_owner = NULL;
429 tss->io_bitmap_max = 0;
430 tss->x86_tss.io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
431 put_cpu();
432 }
433}
434
435void flush_thread(void)
436{
437 struct task_struct *tsk = current;
438
439 tsk->thread.debugreg0 = 0;
440 tsk->thread.debugreg1 = 0;
441 tsk->thread.debugreg2 = 0;
442 tsk->thread.debugreg3 = 0;
443 tsk->thread.debugreg6 = 0;
444 tsk->thread.debugreg7 = 0;
445 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
446 clear_tsk_thread_flag(tsk, TIF_DEBUG);
447
448
449
450 clear_fpu(tsk);
451 clear_used_math();
452}
453
454void release_thread(struct task_struct *dead_task)
455{
456 BUG_ON(dead_task->mm);
457 release_vm86_irqs(dead_task);
458}
459
460
461
462
463
464void prepare_to_copy(struct task_struct *tsk)
465{
466 unlazy_fpu(tsk);
467}
468
469int copy_thread(int nr, unsigned long clone_flags, unsigned long sp,
470 unsigned long unused,
471 struct task_struct * p, struct pt_regs * regs)
472{
473 struct pt_regs * childregs;
474 struct task_struct *tsk;
475 int err;
476
477 childregs = task_pt_regs(p);
478 *childregs = *regs;
479 childregs->ax = 0;
480 childregs->sp = sp;
481
482 p->thread.sp = (unsigned long) childregs;
483 p->thread.sp0 = (unsigned long) (childregs+1);
484
485 p->thread.ip = (unsigned long) ret_from_fork;
486
487 savesegment(gs, p->thread.gs);
488
489 tsk = current;
490 if (unlikely(test_tsk_thread_flag(tsk, TIF_IO_BITMAP))) {
491 p->thread.io_bitmap_ptr = kmemdup(tsk->thread.io_bitmap_ptr,
492 IO_BITMAP_BYTES, GFP_KERNEL);
493 if (!p->thread.io_bitmap_ptr) {
494 p->thread.io_bitmap_max = 0;
495 return -ENOMEM;
496 }
497 set_tsk_thread_flag(p, TIF_IO_BITMAP);
498 }
499
500 err = 0;
501
502
503
504
505 if (clone_flags & CLONE_SETTLS)
506 err = do_set_thread_area(p, -1,
507 (struct user_desc __user *)childregs->si, 0);
508
509 if (err && p->thread.io_bitmap_ptr) {
510 kfree(p->thread.io_bitmap_ptr);
511 p->thread.io_bitmap_max = 0;
512 }
513 return err;
514}
515
516#ifdef CONFIG_SECCOMP
517static void hard_disable_TSC(void)
518{
519 write_cr4(read_cr4() | X86_CR4_TSD);
520}
521void disable_TSC(void)
522{
523 preempt_disable();
524 if (!test_and_set_thread_flag(TIF_NOTSC))
525
526
527
528
529 hard_disable_TSC();
530 preempt_enable();
531}
532static void hard_enable_TSC(void)
533{
534 write_cr4(read_cr4() & ~X86_CR4_TSD);
535}
536#endif
537
538static noinline void
539__switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p,
540 struct tss_struct *tss)
541{
542 struct thread_struct *prev, *next;
543 unsigned long debugctl;
544
545 prev = &prev_p->thread;
546 next = &next_p->thread;
547
548 debugctl = prev->debugctlmsr;
549 if (next->ds_area_msr != prev->ds_area_msr) {
550
551
552 debugctl = 0;
553 wrmsrl(MSR_IA32_DEBUGCTLMSR, 0);
554 wrmsr(MSR_IA32_DS_AREA, next->ds_area_msr, 0);
555 }
556
557 if (next->debugctlmsr != debugctl)
558 wrmsr(MSR_IA32_DEBUGCTLMSR, next->debugctlmsr, 0);
559
560 if (test_tsk_thread_flag(next_p, TIF_DEBUG)) {
561 set_debugreg(next->debugreg0, 0);
562 set_debugreg(next->debugreg1, 1);
563 set_debugreg(next->debugreg2, 2);
564 set_debugreg(next->debugreg3, 3);
565
566 set_debugreg(next->debugreg6, 6);
567 set_debugreg(next->debugreg7, 7);
568 }
569
570#ifdef CONFIG_SECCOMP
571 if (test_tsk_thread_flag(prev_p, TIF_NOTSC) ^
572 test_tsk_thread_flag(next_p, TIF_NOTSC)) {
573
574 if (test_tsk_thread_flag(next_p, TIF_NOTSC))
575 hard_disable_TSC();
576 else
577 hard_enable_TSC();
578 }
579#endif
580
581#ifdef X86_BTS
582 if (test_tsk_thread_flag(prev_p, TIF_BTS_TRACE_TS))
583 ptrace_bts_take_timestamp(prev_p, BTS_TASK_DEPARTS);
584
585 if (test_tsk_thread_flag(next_p, TIF_BTS_TRACE_TS))
586 ptrace_bts_take_timestamp(next_p, BTS_TASK_ARRIVES);
587#endif
588
589
590 if (!test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
591
592
593
594
595 tss->x86_tss.io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
596 return;
597 }
598
599 if (likely(next == tss->io_bitmap_owner)) {
600
601
602
603
604
605 tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET;
606 return;
607 }
608
609
610
611
612
613
614
615
616
617 tss->x86_tss.io_bitmap_base = INVALID_IO_BITMAP_OFFSET_LAZY;
618}
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647struct task_struct * __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
648{
649 struct thread_struct *prev = &prev_p->thread,
650 *next = &next_p->thread;
651 int cpu = smp_processor_id();
652 struct tss_struct *tss = &per_cpu(init_tss, cpu);
653
654
655
656 __unlazy_fpu(prev_p);
657
658
659
660 if (next_p->fpu_counter > 5)
661 prefetch(&next->i387.fxsave);
662
663
664
665
666 load_sp0(tss, next);
667
668
669
670
671
672
673
674
675
676
677
678 savesegment(gs, prev->gs);
679
680
681
682
683 load_TLS(next, cpu);
684
685
686
687
688
689
690
691 if (get_kernel_rpl() && unlikely(prev->iopl != next->iopl))
692 set_iopl_mask(next->iopl);
693
694
695
696
697 if (unlikely(task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV ||
698 task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT))
699 __switch_to_xtra(prev_p, next_p, tss);
700
701
702
703
704
705
706
707
708 arch_leave_lazy_cpu_mode();
709
710
711
712
713
714 if (next_p->fpu_counter > 5)
715 math_state_restore();
716
717
718
719
720 if (prev->gs | next->gs)
721 loadsegment(gs, next->gs);
722
723 x86_write_percpu(current_task, next_p);
724
725 return prev_p;
726}
727
728asmlinkage int sys_fork(struct pt_regs regs)
729{
730 return do_fork(SIGCHLD, regs.sp, ®s, 0, NULL, NULL);
731}
732
733asmlinkage int sys_clone(struct pt_regs regs)
734{
735 unsigned long clone_flags;
736 unsigned long newsp;
737 int __user *parent_tidptr, *child_tidptr;
738
739 clone_flags = regs.bx;
740 newsp = regs.cx;
741 parent_tidptr = (int __user *)regs.dx;
742 child_tidptr = (int __user *)regs.di;
743 if (!newsp)
744 newsp = regs.sp;
745 return do_fork(clone_flags, newsp, ®s, 0, parent_tidptr, child_tidptr);
746}
747
748
749
750
751
752
753
754
755
756
757
758asmlinkage int sys_vfork(struct pt_regs regs)
759{
760 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs.sp, ®s, 0, NULL, NULL);
761}
762
763
764
765
766asmlinkage int sys_execve(struct pt_regs regs)
767{
768 int error;
769 char * filename;
770
771 filename = getname((char __user *) regs.bx);
772 error = PTR_ERR(filename);
773 if (IS_ERR(filename))
774 goto out;
775 error = do_execve(filename,
776 (char __user * __user *) regs.cx,
777 (char __user * __user *) regs.dx,
778 ®s);
779 if (error == 0) {
780
781 set_thread_flag(TIF_IRET);
782 }
783 putname(filename);
784out:
785 return error;
786}
787
788#define top_esp (THREAD_SIZE - sizeof(unsigned long))
789#define top_ebp (THREAD_SIZE - 2*sizeof(unsigned long))
790
791unsigned long get_wchan(struct task_struct *p)
792{
793 unsigned long bp, sp, ip;
794 unsigned long stack_page;
795 int count = 0;
796 if (!p || p == current || p->state == TASK_RUNNING)
797 return 0;
798 stack_page = (unsigned long)task_stack_page(p);
799 sp = p->thread.sp;
800 if (!stack_page || sp < stack_page || sp > top_esp+stack_page)
801 return 0;
802
803 bp = *(unsigned long *) sp;
804 do {
805 if (bp < stack_page || bp > top_ebp+stack_page)
806 return 0;
807 ip = *(unsigned long *) (bp+4);
808 if (!in_sched_functions(ip))
809 return ip;
810 bp = *(unsigned long *) bp;
811 } while (count++ < 16);
812 return 0;
813}
814
815unsigned long arch_align_stack(unsigned long sp)
816{
817 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
818 sp -= get_random_int() % 8192;
819 return sp & ~0xf;
820}
821
822unsigned long arch_randomize_brk(struct mm_struct *mm)
823{
824 unsigned long range_end = mm->brk + 0x02000000;
825 return randomize_range(mm->brk, range_end, 0) ? : mm->brk;
826}
827