Loading...
1#include <linux/errno.h>
2#include <linux/kernel.h>
3#include <linux/mm.h>
4#include <linux/smp.h>
5#include <linux/prctl.h>
6#include <linux/slab.h>
7#include <linux/sched.h>
8#include <linux/module.h>
9#include <linux/pm.h>
10#include <linux/clockchips.h>
11#include <linux/random.h>
12#include <linux/user-return-notifier.h>
13#include <linux/dmi.h>
14#include <linux/utsname.h>
15#include <trace/events/power.h>
16#include <linux/hw_breakpoint.h>
17#include <asm/cpu.h>
18#include <asm/system.h>
19#include <asm/apic.h>
20#include <asm/syscalls.h>
21#include <asm/idle.h>
22#include <asm/uaccess.h>
23#include <asm/i387.h>
24#include <asm/debugreg.h>
25
26struct kmem_cache *task_xstate_cachep;
27EXPORT_SYMBOL_GPL(task_xstate_cachep);
28
29int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
30{
31 int ret;
32
33 *dst = *src;
34 if (fpu_allocated(&src->thread.fpu)) {
35 memset(&dst->thread.fpu, 0, sizeof(dst->thread.fpu));
36 ret = fpu_alloc(&dst->thread.fpu);
37 if (ret)
38 return ret;
39 fpu_copy(&dst->thread.fpu, &src->thread.fpu);
40 }
41 return 0;
42}
43
44void free_thread_xstate(struct task_struct *tsk)
45{
46 fpu_free(&tsk->thread.fpu);
47}
48
49void free_thread_info(struct thread_info *ti)
50{
51 free_thread_xstate(ti->task);
52 free_pages((unsigned long)ti, get_order(THREAD_SIZE));
53}
54
55void arch_task_cache_init(void)
56{
57 task_xstate_cachep =
58 kmem_cache_create("task_xstate", xstate_size,
59 __alignof__(union thread_xstate),
60 SLAB_PANIC | SLAB_NOTRACK, NULL);
61}
62
63/*
64 * Free current thread data structures etc..
65 */
66void exit_thread(void)
67{
68 struct task_struct *me = current;
69 struct thread_struct *t = &me->thread;
70 unsigned long *bp = t->io_bitmap_ptr;
71
72 if (bp) {
73 struct tss_struct *tss = &per_cpu(init_tss, get_cpu());
74
75 t->io_bitmap_ptr = NULL;
76 clear_thread_flag(TIF_IO_BITMAP);
77 /*
78 * Careful, clear this in the TSS too:
79 */
80 memset(tss->io_bitmap, 0xff, t->io_bitmap_max);
81 t->io_bitmap_max = 0;
82 put_cpu();
83 kfree(bp);
84 }
85}
86
87void show_regs(struct pt_regs *regs)
88{
89 show_registers(regs);
90 show_trace(NULL, regs, (unsigned long *)kernel_stack_pointer(regs), 0);
91}
92
93void show_regs_common(void)
94{
95 const char *vendor, *product, *board;
96
97 vendor = dmi_get_system_info(DMI_SYS_VENDOR);
98 if (!vendor)
99 vendor = "";
100 product = dmi_get_system_info(DMI_PRODUCT_NAME);
101 if (!product)
102 product = "";
103
104 /* Board Name is optional */
105 board = dmi_get_system_info(DMI_BOARD_NAME);
106
107 printk(KERN_CONT "\n");
108 printk(KERN_DEFAULT "Pid: %d, comm: %.20s %s %s %.*s",
109 current->pid, current->comm, print_tainted(),
110 init_utsname()->release,
111 (int)strcspn(init_utsname()->version, " "),
112 init_utsname()->version);
113 printk(KERN_CONT " %s %s", vendor, product);
114 if (board)
115 printk(KERN_CONT "/%s", board);
116 printk(KERN_CONT "\n");
117}
118
119void flush_thread(void)
120{
121 struct task_struct *tsk = current;
122
123 flush_ptrace_hw_breakpoint(tsk);
124 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
125 /*
126 * Forget coprocessor state..
127 */
128 tsk->fpu_counter = 0;
129 clear_fpu(tsk);
130 clear_used_math();
131}
132
133static void hard_disable_TSC(void)
134{
135 write_cr4(read_cr4() | X86_CR4_TSD);
136}
137
138void disable_TSC(void)
139{
140 preempt_disable();
141 if (!test_and_set_thread_flag(TIF_NOTSC))
142 /*
143 * Must flip the CPU state synchronously with
144 * TIF_NOTSC in the current running context.
145 */
146 hard_disable_TSC();
147 preempt_enable();
148}
149
150static void hard_enable_TSC(void)
151{
152 write_cr4(read_cr4() & ~X86_CR4_TSD);
153}
154
155static void enable_TSC(void)
156{
157 preempt_disable();
158 if (test_and_clear_thread_flag(TIF_NOTSC))
159 /*
160 * Must flip the CPU state synchronously with
161 * TIF_NOTSC in the current running context.
162 */
163 hard_enable_TSC();
164 preempt_enable();
165}
166
167int get_tsc_mode(unsigned long adr)
168{
169 unsigned int val;
170
171 if (test_thread_flag(TIF_NOTSC))
172 val = PR_TSC_SIGSEGV;
173 else
174 val = PR_TSC_ENABLE;
175
176 return put_user(val, (unsigned int __user *)adr);
177}
178
179int set_tsc_mode(unsigned int val)
180{
181 if (val == PR_TSC_SIGSEGV)
182 disable_TSC();
183 else if (val == PR_TSC_ENABLE)
184 enable_TSC();
185 else
186 return -EINVAL;
187
188 return 0;
189}
190
191void __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p,
192 struct tss_struct *tss)
193{
194 struct thread_struct *prev, *next;
195
196 prev = &prev_p->thread;
197 next = &next_p->thread;
198
199 if (test_tsk_thread_flag(prev_p, TIF_BLOCKSTEP) ^
200 test_tsk_thread_flag(next_p, TIF_BLOCKSTEP)) {
201 unsigned long debugctl = get_debugctlmsr();
202
203 debugctl &= ~DEBUGCTLMSR_BTF;
204 if (test_tsk_thread_flag(next_p, TIF_BLOCKSTEP))
205 debugctl |= DEBUGCTLMSR_BTF;
206
207 update_debugctlmsr(debugctl);
208 }
209
210 if (test_tsk_thread_flag(prev_p, TIF_NOTSC) ^
211 test_tsk_thread_flag(next_p, TIF_NOTSC)) {
212 /* prev and next are different */
213 if (test_tsk_thread_flag(next_p, TIF_NOTSC))
214 hard_disable_TSC();
215 else
216 hard_enable_TSC();
217 }
218
219 if (test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
220 /*
221 * Copy the relevant range of the IO bitmap.
222 * Normally this is 128 bytes or less:
223 */
224 memcpy(tss->io_bitmap, next->io_bitmap_ptr,
225 max(prev->io_bitmap_max, next->io_bitmap_max));
226 } else if (test_tsk_thread_flag(prev_p, TIF_IO_BITMAP)) {
227 /*
228 * Clear any possible leftover bits:
229 */
230 memset(tss->io_bitmap, 0xff, prev->io_bitmap_max);
231 }
232 propagate_user_return_notify(prev_p, next_p);
233}
234
235int sys_fork(struct pt_regs *regs)
236{
237 return do_fork(SIGCHLD, regs->sp, regs, 0, NULL, NULL);
238}
239
240/*
241 * This is trivial, and on the face of it looks like it
242 * could equally well be done in user mode.
243 *
244 * Not so, for quite unobvious reasons - register pressure.
245 * In user mode vfork() cannot have a stack frame, and if
246 * done by calling the "clone()" system call directly, you
247 * do not have enough call-clobbered registers to hold all
248 * the information you need.
249 */
250int sys_vfork(struct pt_regs *regs)
251{
252 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->sp, regs, 0,
253 NULL, NULL);
254}
255
256long
257sys_clone(unsigned long clone_flags, unsigned long newsp,
258 void __user *parent_tid, void __user *child_tid, struct pt_regs *regs)
259{
260 if (!newsp)
261 newsp = regs->sp;
262 return do_fork(clone_flags, newsp, regs, 0, parent_tid, child_tid);
263}
264
265/*
266 * This gets run with %si containing the
267 * function to call, and %di containing
268 * the "args".
269 */
270extern void kernel_thread_helper(void);
271
272/*
273 * Create a kernel thread
274 */
275int kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
276{
277 struct pt_regs regs;
278
279 memset(®s, 0, sizeof(regs));
280
281 regs.si = (unsigned long) fn;
282 regs.di = (unsigned long) arg;
283
284#ifdef CONFIG_X86_32
285 regs.ds = __USER_DS;
286 regs.es = __USER_DS;
287 regs.fs = __KERNEL_PERCPU;
288 regs.gs = __KERNEL_STACK_CANARY;
289#else
290 regs.ss = __KERNEL_DS;
291#endif
292
293 regs.orig_ax = -1;
294 regs.ip = (unsigned long) kernel_thread_helper;
295 regs.cs = __KERNEL_CS | get_kernel_rpl();
296 regs.flags = X86_EFLAGS_IF | 0x2;
297
298 /* Ok, create the new process.. */
299 return do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0, ®s, 0, NULL, NULL);
300}
301EXPORT_SYMBOL(kernel_thread);
302
303/*
304 * sys_execve() executes a new program.
305 */
306long sys_execve(const char __user *name,
307 const char __user *const __user *argv,
308 const char __user *const __user *envp, struct pt_regs *regs)
309{
310 long error;
311 char *filename;
312
313 filename = getname(name);
314 error = PTR_ERR(filename);
315 if (IS_ERR(filename))
316 return error;
317 error = do_execve(filename, argv, envp, regs);
318
319#ifdef CONFIG_X86_32
320 if (error == 0) {
321 /* Make sure we don't return using sysenter.. */
322 set_thread_flag(TIF_IRET);
323 }
324#endif
325
326 putname(filename);
327 return error;
328}
329
330/*
331 * Idle related variables and functions
332 */
333unsigned long boot_option_idle_override = IDLE_NO_OVERRIDE;
334EXPORT_SYMBOL(boot_option_idle_override);
335
336/*
337 * Powermanagement idle function, if any..
338 */
339void (*pm_idle)(void);
340#ifdef CONFIG_APM_MODULE
341EXPORT_SYMBOL(pm_idle);
342#endif
343
344#ifdef CONFIG_X86_32
345/*
346 * This halt magic was a workaround for ancient floppy DMA
347 * wreckage. It should be safe to remove.
348 */
349static int hlt_counter;
350void disable_hlt(void)
351{
352 hlt_counter++;
353}
354EXPORT_SYMBOL(disable_hlt);
355
356void enable_hlt(void)
357{
358 hlt_counter--;
359}
360EXPORT_SYMBOL(enable_hlt);
361
362static inline int hlt_use_halt(void)
363{
364 return (!hlt_counter && boot_cpu_data.hlt_works_ok);
365}
366#else
367static inline int hlt_use_halt(void)
368{
369 return 1;
370}
371#endif
372
373/*
374 * We use this if we don't have any better
375 * idle routine..
376 */
377void default_idle(void)
378{
379 if (hlt_use_halt()) {
380 trace_power_start(POWER_CSTATE, 1, smp_processor_id());
381 trace_cpu_idle(1, smp_processor_id());
382 current_thread_info()->status &= ~TS_POLLING;
383 /*
384 * TS_POLLING-cleared state must be visible before we
385 * test NEED_RESCHED:
386 */
387 smp_mb();
388
389 if (!need_resched())
390 safe_halt(); /* enables interrupts racelessly */
391 else
392 local_irq_enable();
393 current_thread_info()->status |= TS_POLLING;
394 trace_power_end(smp_processor_id());
395 trace_cpu_idle(PWR_EVENT_EXIT, smp_processor_id());
396 } else {
397 local_irq_enable();
398 /* loop is done by the caller */
399 cpu_relax();
400 }
401}
402#ifdef CONFIG_APM_MODULE
403EXPORT_SYMBOL(default_idle);
404#endif
405
406void stop_this_cpu(void *dummy)
407{
408 local_irq_disable();
409 /*
410 * Remove this CPU:
411 */
412 set_cpu_online(smp_processor_id(), false);
413 disable_local_APIC();
414
415 for (;;) {
416 if (hlt_works(smp_processor_id()))
417 halt();
418 }
419}
420
421static void do_nothing(void *unused)
422{
423}
424
425/*
426 * cpu_idle_wait - Used to ensure that all the CPUs discard old value of
427 * pm_idle and update to new pm_idle value. Required while changing pm_idle
428 * handler on SMP systems.
429 *
430 * Caller must have changed pm_idle to the new value before the call. Old
431 * pm_idle value will not be used by any CPU after the return of this function.
432 */
433void cpu_idle_wait(void)
434{
435 smp_mb();
436 /* kick all the CPUs so that they exit out of pm_idle */
437 smp_call_function(do_nothing, NULL, 1);
438}
439EXPORT_SYMBOL_GPL(cpu_idle_wait);
440
441/* Default MONITOR/MWAIT with no hints, used for default C1 state */
442static void mwait_idle(void)
443{
444 if (!need_resched()) {
445 trace_power_start(POWER_CSTATE, 1, smp_processor_id());
446 trace_cpu_idle(1, smp_processor_id());
447 if (this_cpu_has(X86_FEATURE_CLFLUSH_MONITOR))
448 clflush((void *)¤t_thread_info()->flags);
449
450 __monitor((void *)¤t_thread_info()->flags, 0, 0);
451 smp_mb();
452 if (!need_resched())
453 __sti_mwait(0, 0);
454 else
455 local_irq_enable();
456 trace_power_end(smp_processor_id());
457 trace_cpu_idle(PWR_EVENT_EXIT, smp_processor_id());
458 } else
459 local_irq_enable();
460}
461
462/*
463 * On SMP it's slightly faster (but much more power-consuming!)
464 * to poll the ->work.need_resched flag instead of waiting for the
465 * cross-CPU IPI to arrive. Use this option with caution.
466 */
467static void poll_idle(void)
468{
469 trace_power_start(POWER_CSTATE, 0, smp_processor_id());
470 trace_cpu_idle(0, smp_processor_id());
471 local_irq_enable();
472 while (!need_resched())
473 cpu_relax();
474 trace_power_end(smp_processor_id());
475 trace_cpu_idle(PWR_EVENT_EXIT, smp_processor_id());
476}
477
478/*
479 * mwait selection logic:
480 *
481 * It depends on the CPU. For AMD CPUs that support MWAIT this is
482 * wrong. Family 0x10 and 0x11 CPUs will enter C1 on HLT. Powersavings
483 * then depend on a clock divisor and current Pstate of the core. If
484 * all cores of a processor are in halt state (C1) the processor can
485 * enter the C1E (C1 enhanced) state. If mwait is used this will never
486 * happen.
487 *
488 * idle=mwait overrides this decision and forces the usage of mwait.
489 */
490
491#define MWAIT_INFO 0x05
492#define MWAIT_ECX_EXTENDED_INFO 0x01
493#define MWAIT_EDX_C1 0xf0
494
495int mwait_usable(const struct cpuinfo_x86 *c)
496{
497 u32 eax, ebx, ecx, edx;
498
499 if (boot_option_idle_override == IDLE_FORCE_MWAIT)
500 return 1;
501
502 if (c->cpuid_level < MWAIT_INFO)
503 return 0;
504
505 cpuid(MWAIT_INFO, &eax, &ebx, &ecx, &edx);
506 /* Check, whether EDX has extended info about MWAIT */
507 if (!(ecx & MWAIT_ECX_EXTENDED_INFO))
508 return 1;
509
510 /*
511 * edx enumeratios MONITOR/MWAIT extensions. Check, whether
512 * C1 supports MWAIT
513 */
514 return (edx & MWAIT_EDX_C1);
515}
516
517bool amd_e400_c1e_detected;
518EXPORT_SYMBOL(amd_e400_c1e_detected);
519
520static cpumask_var_t amd_e400_c1e_mask;
521
522void amd_e400_remove_cpu(int cpu)
523{
524 if (amd_e400_c1e_mask != NULL)
525 cpumask_clear_cpu(cpu, amd_e400_c1e_mask);
526}
527
528/*
529 * AMD Erratum 400 aware idle routine. We check for C1E active in the interrupt
530 * pending message MSR. If we detect C1E, then we handle it the same
531 * way as C3 power states (local apic timer and TSC stop)
532 */
533static void amd_e400_idle(void)
534{
535 if (need_resched())
536 return;
537
538 if (!amd_e400_c1e_detected) {
539 u32 lo, hi;
540
541 rdmsr(MSR_K8_INT_PENDING_MSG, lo, hi);
542
543 if (lo & K8_INTP_C1E_ACTIVE_MASK) {
544 amd_e400_c1e_detected = true;
545 if (!boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
546 mark_tsc_unstable("TSC halt in AMD C1E");
547 printk(KERN_INFO "System has AMD C1E enabled\n");
548 }
549 }
550
551 if (amd_e400_c1e_detected) {
552 int cpu = smp_processor_id();
553
554 if (!cpumask_test_cpu(cpu, amd_e400_c1e_mask)) {
555 cpumask_set_cpu(cpu, amd_e400_c1e_mask);
556 /*
557 * Force broadcast so ACPI can not interfere.
558 */
559 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_FORCE,
560 &cpu);
561 printk(KERN_INFO "Switch to broadcast mode on CPU%d\n",
562 cpu);
563 }
564 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_ENTER, &cpu);
565
566 default_idle();
567
568 /*
569 * The switch back from broadcast mode needs to be
570 * called with interrupts disabled.
571 */
572 local_irq_disable();
573 clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_EXIT, &cpu);
574 local_irq_enable();
575 } else
576 default_idle();
577}
578
579void __cpuinit select_idle_routine(const struct cpuinfo_x86 *c)
580{
581#ifdef CONFIG_SMP
582 if (pm_idle == poll_idle && smp_num_siblings > 1) {
583 printk_once(KERN_WARNING "WARNING: polling idle and HT enabled,"
584 " performance may degrade.\n");
585 }
586#endif
587 if (pm_idle)
588 return;
589
590 if (cpu_has(c, X86_FEATURE_MWAIT) && mwait_usable(c)) {
591 /*
592 * One CPU supports mwait => All CPUs supports mwait
593 */
594 printk(KERN_INFO "using mwait in idle threads.\n");
595 pm_idle = mwait_idle;
596 } else if (cpu_has_amd_erratum(amd_erratum_400)) {
597 /* E400: APIC timer interrupt does not wake up CPU from C1e */
598 printk(KERN_INFO "using AMD E400 aware idle routine\n");
599 pm_idle = amd_e400_idle;
600 } else
601 pm_idle = default_idle;
602}
603
604void __init init_amd_e400_c1e_mask(void)
605{
606 /* If we're using amd_e400_idle, we need to allocate amd_e400_c1e_mask. */
607 if (pm_idle == amd_e400_idle)
608 zalloc_cpumask_var(&amd_e400_c1e_mask, GFP_KERNEL);
609}
610
611static int __init idle_setup(char *str)
612{
613 if (!str)
614 return -EINVAL;
615
616 if (!strcmp(str, "poll")) {
617 printk("using polling idle threads.\n");
618 pm_idle = poll_idle;
619 boot_option_idle_override = IDLE_POLL;
620 } else if (!strcmp(str, "mwait")) {
621 boot_option_idle_override = IDLE_FORCE_MWAIT;
622 WARN_ONCE(1, "\"idle=mwait\" will be removed in 2012\n");
623 } else if (!strcmp(str, "halt")) {
624 /*
625 * When the boot option of idle=halt is added, halt is
626 * forced to be used for CPU idle. In such case CPU C2/C3
627 * won't be used again.
628 * To continue to load the CPU idle driver, don't touch
629 * the boot_option_idle_override.
630 */
631 pm_idle = default_idle;
632 boot_option_idle_override = IDLE_HALT;
633 } else if (!strcmp(str, "nomwait")) {
634 /*
635 * If the boot option of "idle=nomwait" is added,
636 * it means that mwait will be disabled for CPU C2/C3
637 * states. In such case it won't touch the variable
638 * of boot_option_idle_override.
639 */
640 boot_option_idle_override = IDLE_NOMWAIT;
641 } else
642 return -1;
643
644 return 0;
645}
646early_param("idle", idle_setup);
647
648unsigned long arch_align_stack(unsigned long sp)
649{
650 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
651 sp -= get_random_int() % 8192;
652 return sp & ~0xf;
653}
654
655unsigned long arch_randomize_brk(struct mm_struct *mm)
656{
657 unsigned long range_end = mm->brk + 0x02000000;
658 return randomize_range(mm->brk, range_end, 0) ? : mm->brk;
659}
660
1#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
2
3#include <linux/errno.h>
4#include <linux/kernel.h>
5#include <linux/mm.h>
6#include <linux/smp.h>
7#include <linux/prctl.h>
8#include <linux/slab.h>
9#include <linux/sched.h>
10#include <linux/init.h>
11#include <linux/export.h>
12#include <linux/pm.h>
13#include <linux/tick.h>
14#include <linux/random.h>
15#include <linux/user-return-notifier.h>
16#include <linux/dmi.h>
17#include <linux/utsname.h>
18#include <linux/stackprotector.h>
19#include <linux/tick.h>
20#include <linux/cpuidle.h>
21#include <trace/events/power.h>
22#include <linux/hw_breakpoint.h>
23#include <asm/cpu.h>
24#include <asm/apic.h>
25#include <asm/syscalls.h>
26#include <linux/uaccess.h>
27#include <asm/mwait.h>
28#include <asm/fpu/internal.h>
29#include <asm/debugreg.h>
30#include <asm/nmi.h>
31#include <asm/tlbflush.h>
32#include <asm/mce.h>
33#include <asm/vm86.h>
34#include <asm/switch_to.h>
35
36/*
37 * per-CPU TSS segments. Threads are completely 'soft' on Linux,
38 * no more per-task TSS's. The TSS size is kept cacheline-aligned
39 * so they are allowed to end up in the .data..cacheline_aligned
40 * section. Since TSS's are completely CPU-local, we want them
41 * on exact cacheline boundaries, to eliminate cacheline ping-pong.
42 */
43__visible DEFINE_PER_CPU_SHARED_ALIGNED(struct tss_struct, cpu_tss) = {
44 .x86_tss = {
45 .sp0 = TOP_OF_INIT_STACK,
46#ifdef CONFIG_X86_32
47 .ss0 = __KERNEL_DS,
48 .ss1 = __KERNEL_CS,
49 .io_bitmap_base = INVALID_IO_BITMAP_OFFSET,
50#endif
51 },
52#ifdef CONFIG_X86_32
53 /*
54 * Note that the .io_bitmap member must be extra-big. This is because
55 * the CPU will access an additional byte beyond the end of the IO
56 * permission bitmap. The extra byte must be all 1 bits, and must
57 * be within the limit.
58 */
59 .io_bitmap = { [0 ... IO_BITMAP_LONGS] = ~0 },
60#endif
61#ifdef CONFIG_X86_32
62 .SYSENTER_stack_canary = STACK_END_MAGIC,
63#endif
64};
65EXPORT_PER_CPU_SYMBOL(cpu_tss);
66
67/*
68 * this gets called so that we can store lazy state into memory and copy the
69 * current task into the new thread.
70 */
71int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
72{
73 memcpy(dst, src, arch_task_struct_size);
74#ifdef CONFIG_VM86
75 dst->thread.vm86 = NULL;
76#endif
77
78 return fpu__copy(&dst->thread.fpu, &src->thread.fpu);
79}
80
81/*
82 * Free current thread data structures etc..
83 */
84void exit_thread(struct task_struct *tsk)
85{
86 struct thread_struct *t = &tsk->thread;
87 unsigned long *bp = t->io_bitmap_ptr;
88 struct fpu *fpu = &t->fpu;
89
90 if (bp) {
91 struct tss_struct *tss = &per_cpu(cpu_tss, get_cpu());
92
93 t->io_bitmap_ptr = NULL;
94 clear_thread_flag(TIF_IO_BITMAP);
95 /*
96 * Careful, clear this in the TSS too:
97 */
98 memset(tss->io_bitmap, 0xff, t->io_bitmap_max);
99 t->io_bitmap_max = 0;
100 put_cpu();
101 kfree(bp);
102 }
103
104 free_vm86(t);
105
106 fpu__drop(fpu);
107}
108
109void flush_thread(void)
110{
111 struct task_struct *tsk = current;
112
113 flush_ptrace_hw_breakpoint(tsk);
114 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
115
116 fpu__clear(&tsk->thread.fpu);
117}
118
119static void hard_disable_TSC(void)
120{
121 cr4_set_bits(X86_CR4_TSD);
122}
123
124void disable_TSC(void)
125{
126 preempt_disable();
127 if (!test_and_set_thread_flag(TIF_NOTSC))
128 /*
129 * Must flip the CPU state synchronously with
130 * TIF_NOTSC in the current running context.
131 */
132 hard_disable_TSC();
133 preempt_enable();
134}
135
136static void hard_enable_TSC(void)
137{
138 cr4_clear_bits(X86_CR4_TSD);
139}
140
141static void enable_TSC(void)
142{
143 preempt_disable();
144 if (test_and_clear_thread_flag(TIF_NOTSC))
145 /*
146 * Must flip the CPU state synchronously with
147 * TIF_NOTSC in the current running context.
148 */
149 hard_enable_TSC();
150 preempt_enable();
151}
152
153int get_tsc_mode(unsigned long adr)
154{
155 unsigned int val;
156
157 if (test_thread_flag(TIF_NOTSC))
158 val = PR_TSC_SIGSEGV;
159 else
160 val = PR_TSC_ENABLE;
161
162 return put_user(val, (unsigned int __user *)adr);
163}
164
165int set_tsc_mode(unsigned int val)
166{
167 if (val == PR_TSC_SIGSEGV)
168 disable_TSC();
169 else if (val == PR_TSC_ENABLE)
170 enable_TSC();
171 else
172 return -EINVAL;
173
174 return 0;
175}
176
177void __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p,
178 struct tss_struct *tss)
179{
180 struct thread_struct *prev, *next;
181
182 prev = &prev_p->thread;
183 next = &next_p->thread;
184
185 if (test_tsk_thread_flag(prev_p, TIF_BLOCKSTEP) ^
186 test_tsk_thread_flag(next_p, TIF_BLOCKSTEP)) {
187 unsigned long debugctl = get_debugctlmsr();
188
189 debugctl &= ~DEBUGCTLMSR_BTF;
190 if (test_tsk_thread_flag(next_p, TIF_BLOCKSTEP))
191 debugctl |= DEBUGCTLMSR_BTF;
192
193 update_debugctlmsr(debugctl);
194 }
195
196 if (test_tsk_thread_flag(prev_p, TIF_NOTSC) ^
197 test_tsk_thread_flag(next_p, TIF_NOTSC)) {
198 /* prev and next are different */
199 if (test_tsk_thread_flag(next_p, TIF_NOTSC))
200 hard_disable_TSC();
201 else
202 hard_enable_TSC();
203 }
204
205 if (test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
206 /*
207 * Copy the relevant range of the IO bitmap.
208 * Normally this is 128 bytes or less:
209 */
210 memcpy(tss->io_bitmap, next->io_bitmap_ptr,
211 max(prev->io_bitmap_max, next->io_bitmap_max));
212 } else if (test_tsk_thread_flag(prev_p, TIF_IO_BITMAP)) {
213 /*
214 * Clear any possible leftover bits:
215 */
216 memset(tss->io_bitmap, 0xff, prev->io_bitmap_max);
217 }
218 propagate_user_return_notify(prev_p, next_p);
219}
220
221/*
222 * Idle related variables and functions
223 */
224unsigned long boot_option_idle_override = IDLE_NO_OVERRIDE;
225EXPORT_SYMBOL(boot_option_idle_override);
226
227static void (*x86_idle)(void);
228
229#ifndef CONFIG_SMP
230static inline void play_dead(void)
231{
232 BUG();
233}
234#endif
235
236void arch_cpu_idle_enter(void)
237{
238 tsc_verify_tsc_adjust(false);
239 local_touch_nmi();
240}
241
242void arch_cpu_idle_dead(void)
243{
244 play_dead();
245}
246
247/*
248 * Called from the generic idle code.
249 */
250void arch_cpu_idle(void)
251{
252 x86_idle();
253}
254
255/*
256 * We use this if we don't have any better idle routine..
257 */
258void __cpuidle default_idle(void)
259{
260 trace_cpu_idle_rcuidle(1, smp_processor_id());
261 safe_halt();
262 trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, smp_processor_id());
263}
264#ifdef CONFIG_APM_MODULE
265EXPORT_SYMBOL(default_idle);
266#endif
267
268#ifdef CONFIG_XEN
269bool xen_set_default_idle(void)
270{
271 bool ret = !!x86_idle;
272
273 x86_idle = default_idle;
274
275 return ret;
276}
277#endif
278void stop_this_cpu(void *dummy)
279{
280 local_irq_disable();
281 /*
282 * Remove this CPU:
283 */
284 set_cpu_online(smp_processor_id(), false);
285 disable_local_APIC();
286 mcheck_cpu_clear(this_cpu_ptr(&cpu_info));
287
288 for (;;)
289 halt();
290}
291
292/*
293 * AMD Erratum 400 aware idle routine. We handle it the same way as C3 power
294 * states (local apic timer and TSC stop).
295 */
296static void amd_e400_idle(void)
297{
298 /*
299 * We cannot use static_cpu_has_bug() here because X86_BUG_AMD_APIC_C1E
300 * gets set after static_cpu_has() places have been converted via
301 * alternatives.
302 */
303 if (!boot_cpu_has_bug(X86_BUG_AMD_APIC_C1E)) {
304 default_idle();
305 return;
306 }
307
308 tick_broadcast_enter();
309
310 default_idle();
311
312 /*
313 * The switch back from broadcast mode needs to be called with
314 * interrupts disabled.
315 */
316 local_irq_disable();
317 tick_broadcast_exit();
318 local_irq_enable();
319}
320
321/*
322 * Intel Core2 and older machines prefer MWAIT over HALT for C1.
323 * We can't rely on cpuidle installing MWAIT, because it will not load
324 * on systems that support only C1 -- so the boot default must be MWAIT.
325 *
326 * Some AMD machines are the opposite, they depend on using HALT.
327 *
328 * So for default C1, which is used during boot until cpuidle loads,
329 * use MWAIT-C1 on Intel HW that has it, else use HALT.
330 */
331static int prefer_mwait_c1_over_halt(const struct cpuinfo_x86 *c)
332{
333 if (c->x86_vendor != X86_VENDOR_INTEL)
334 return 0;
335
336 if (!cpu_has(c, X86_FEATURE_MWAIT) || static_cpu_has_bug(X86_BUG_MONITOR))
337 return 0;
338
339 return 1;
340}
341
342/*
343 * MONITOR/MWAIT with no hints, used for default C1 state. This invokes MWAIT
344 * with interrupts enabled and no flags, which is backwards compatible with the
345 * original MWAIT implementation.
346 */
347static __cpuidle void mwait_idle(void)
348{
349 if (!current_set_polling_and_test()) {
350 trace_cpu_idle_rcuidle(1, smp_processor_id());
351 if (this_cpu_has(X86_BUG_CLFLUSH_MONITOR)) {
352 mb(); /* quirk */
353 clflush((void *)¤t_thread_info()->flags);
354 mb(); /* quirk */
355 }
356
357 __monitor((void *)¤t_thread_info()->flags, 0, 0);
358 if (!need_resched())
359 __sti_mwait(0, 0);
360 else
361 local_irq_enable();
362 trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, smp_processor_id());
363 } else {
364 local_irq_enable();
365 }
366 __current_clr_polling();
367}
368
369void select_idle_routine(const struct cpuinfo_x86 *c)
370{
371#ifdef CONFIG_SMP
372 if (boot_option_idle_override == IDLE_POLL && smp_num_siblings > 1)
373 pr_warn_once("WARNING: polling idle and HT enabled, performance may degrade\n");
374#endif
375 if (x86_idle || boot_option_idle_override == IDLE_POLL)
376 return;
377
378 if (boot_cpu_has_bug(X86_BUG_AMD_E400)) {
379 pr_info("using AMD E400 aware idle routine\n");
380 x86_idle = amd_e400_idle;
381 } else if (prefer_mwait_c1_over_halt(c)) {
382 pr_info("using mwait in idle threads\n");
383 x86_idle = mwait_idle;
384 } else
385 x86_idle = default_idle;
386}
387
388void amd_e400_c1e_apic_setup(void)
389{
390 if (boot_cpu_has_bug(X86_BUG_AMD_APIC_C1E)) {
391 pr_info("Switch to broadcast mode on CPU%d\n", smp_processor_id());
392 local_irq_disable();
393 tick_broadcast_force();
394 local_irq_enable();
395 }
396}
397
398void __init arch_post_acpi_subsys_init(void)
399{
400 u32 lo, hi;
401
402 if (!boot_cpu_has_bug(X86_BUG_AMD_E400))
403 return;
404
405 /*
406 * AMD E400 detection needs to happen after ACPI has been enabled. If
407 * the machine is affected K8_INTP_C1E_ACTIVE_MASK bits are set in
408 * MSR_K8_INT_PENDING_MSG.
409 */
410 rdmsr(MSR_K8_INT_PENDING_MSG, lo, hi);
411 if (!(lo & K8_INTP_C1E_ACTIVE_MASK))
412 return;
413
414 boot_cpu_set_bug(X86_BUG_AMD_APIC_C1E);
415
416 if (!boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
417 mark_tsc_unstable("TSC halt in AMD C1E");
418 pr_info("System has AMD C1E enabled\n");
419}
420
421static int __init idle_setup(char *str)
422{
423 if (!str)
424 return -EINVAL;
425
426 if (!strcmp(str, "poll")) {
427 pr_info("using polling idle threads\n");
428 boot_option_idle_override = IDLE_POLL;
429 cpu_idle_poll_ctrl(true);
430 } else if (!strcmp(str, "halt")) {
431 /*
432 * When the boot option of idle=halt is added, halt is
433 * forced to be used for CPU idle. In such case CPU C2/C3
434 * won't be used again.
435 * To continue to load the CPU idle driver, don't touch
436 * the boot_option_idle_override.
437 */
438 x86_idle = default_idle;
439 boot_option_idle_override = IDLE_HALT;
440 } else if (!strcmp(str, "nomwait")) {
441 /*
442 * If the boot option of "idle=nomwait" is added,
443 * it means that mwait will be disabled for CPU C2/C3
444 * states. In such case it won't touch the variable
445 * of boot_option_idle_override.
446 */
447 boot_option_idle_override = IDLE_NOMWAIT;
448 } else
449 return -1;
450
451 return 0;
452}
453early_param("idle", idle_setup);
454
455unsigned long arch_align_stack(unsigned long sp)
456{
457 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
458 sp -= get_random_int() % 8192;
459 return sp & ~0xf;
460}
461
462unsigned long arch_randomize_brk(struct mm_struct *mm)
463{
464 return randomize_page(mm->brk, 0x02000000);
465}
466
467/*
468 * Return saved PC of a blocked thread.
469 * What is this good for? it will be always the scheduler or ret_from_fork.
470 */
471unsigned long thread_saved_pc(struct task_struct *tsk)
472{
473 struct inactive_task_frame *frame =
474 (struct inactive_task_frame *) READ_ONCE(tsk->thread.sp);
475 return READ_ONCE_NOCHECK(frame->ret_addr);
476}
477
478/*
479 * Called from fs/proc with a reference on @p to find the function
480 * which called into schedule(). This needs to be done carefully
481 * because the task might wake up and we might look at a stack
482 * changing under us.
483 */
484unsigned long get_wchan(struct task_struct *p)
485{
486 unsigned long start, bottom, top, sp, fp, ip, ret = 0;
487 int count = 0;
488
489 if (!p || p == current || p->state == TASK_RUNNING)
490 return 0;
491
492 if (!try_get_task_stack(p))
493 return 0;
494
495 start = (unsigned long)task_stack_page(p);
496 if (!start)
497 goto out;
498
499 /*
500 * Layout of the stack page:
501 *
502 * ----------- topmax = start + THREAD_SIZE - sizeof(unsigned long)
503 * PADDING
504 * ----------- top = topmax - TOP_OF_KERNEL_STACK_PADDING
505 * stack
506 * ----------- bottom = start
507 *
508 * The tasks stack pointer points at the location where the
509 * framepointer is stored. The data on the stack is:
510 * ... IP FP ... IP FP
511 *
512 * We need to read FP and IP, so we need to adjust the upper
513 * bound by another unsigned long.
514 */
515 top = start + THREAD_SIZE - TOP_OF_KERNEL_STACK_PADDING;
516 top -= 2 * sizeof(unsigned long);
517 bottom = start;
518
519 sp = READ_ONCE(p->thread.sp);
520 if (sp < bottom || sp > top)
521 goto out;
522
523 fp = READ_ONCE_NOCHECK(((struct inactive_task_frame *)sp)->bp);
524 do {
525 if (fp < bottom || fp > top)
526 goto out;
527 ip = READ_ONCE_NOCHECK(*(unsigned long *)(fp + sizeof(unsigned long)));
528 if (!in_sched_functions(ip)) {
529 ret = ip;
530 goto out;
531 }
532 fp = READ_ONCE_NOCHECK(*(unsigned long *)fp);
533 } while (count++ < 16 && p->state != TASK_RUNNING);
534
535out:
536 put_task_stack(p);
537 return ret;
538}