Linux Audio

Check our new training course

Loading...
v3.1
 
 
  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(&regs, 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, &regs, 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 *)&current_thread_info()->flags);
449
450		__monitor((void *)&current_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
v4.6
  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/module.h>
 11#include <linux/pm.h>
 12#include <linux/tick.h>
 13#include <linux/random.h>
 14#include <linux/user-return-notifier.h>
 15#include <linux/dmi.h>
 16#include <linux/utsname.h>
 17#include <linux/stackprotector.h>
 18#include <linux/tick.h>
 19#include <linux/cpuidle.h>
 20#include <trace/events/power.h>
 21#include <linux/hw_breakpoint.h>
 22#include <asm/cpu.h>
 
 23#include <asm/apic.h>
 24#include <asm/syscalls.h>
 25#include <asm/idle.h>
 26#include <asm/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
 35/*
 36 * per-CPU TSS segments. Threads are completely 'soft' on Linux,
 37 * no more per-task TSS's. The TSS size is kept cacheline-aligned
 38 * so they are allowed to end up in the .data..cacheline_aligned
 39 * section. Since TSS's are completely CPU-local, we want them
 40 * on exact cacheline boundaries, to eliminate cacheline ping-pong.
 41 */
 42__visible DEFINE_PER_CPU_SHARED_ALIGNED(struct tss_struct, cpu_tss) = {
 43	.x86_tss = {
 44		.sp0 = TOP_OF_INIT_STACK,
 45#ifdef CONFIG_X86_32
 46		.ss0 = __KERNEL_DS,
 47		.ss1 = __KERNEL_CS,
 48		.io_bitmap_base	= INVALID_IO_BITMAP_OFFSET,
 49#endif
 50	 },
 51#ifdef CONFIG_X86_32
 52	 /*
 53	  * Note that the .io_bitmap member must be extra-big. This is because
 54	  * the CPU will access an additional byte beyond the end of the IO
 55	  * permission bitmap. The extra byte must be all 1 bits, and must
 56	  * be within the limit.
 57	  */
 58	.io_bitmap		= { [0 ... IO_BITMAP_LONGS] = ~0 },
 59#endif
 60#ifdef CONFIG_X86_32
 61	.SYSENTER_stack_canary	= STACK_END_MAGIC,
 62#endif
 63};
 64EXPORT_PER_CPU_SYMBOL(cpu_tss);
 65
 66#ifdef CONFIG_X86_64
 67static DEFINE_PER_CPU(unsigned char, is_idle);
 68static ATOMIC_NOTIFIER_HEAD(idle_notifier);
 69
 70void idle_notifier_register(struct notifier_block *n)
 
 
 
 
 
 
 
 
 
 
 
 71{
 72	atomic_notifier_chain_register(&idle_notifier, n);
 73}
 74EXPORT_SYMBOL_GPL(idle_notifier_register);
 75
 76void idle_notifier_unregister(struct notifier_block *n)
 77{
 78	atomic_notifier_chain_unregister(&idle_notifier, n);
 
 79}
 80EXPORT_SYMBOL_GPL(idle_notifier_unregister);
 81#endif
 82
 83/*
 84 * this gets called so that we can store lazy state into memory and copy the
 85 * current task into the new thread.
 86 */
 87int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
 88{
 89	memcpy(dst, src, arch_task_struct_size);
 90#ifdef CONFIG_VM86
 91	dst->thread.vm86 = NULL;
 92#endif
 93
 94	return fpu__copy(&dst->thread.fpu, &src->thread.fpu);
 95}
 96
 97/*
 98 * Free current thread data structures etc..
 99 */
100void exit_thread(void)
101{
102	struct task_struct *me = current;
103	struct thread_struct *t = &me->thread;
104	unsigned long *bp = t->io_bitmap_ptr;
105	struct fpu *fpu = &t->fpu;
106
107	if (bp) {
108		struct tss_struct *tss = &per_cpu(cpu_tss, get_cpu());
109
110		t->io_bitmap_ptr = NULL;
111		clear_thread_flag(TIF_IO_BITMAP);
112		/*
113		 * Careful, clear this in the TSS too:
114		 */
115		memset(tss->io_bitmap, 0xff, t->io_bitmap_max);
116		t->io_bitmap_max = 0;
117		put_cpu();
118		kfree(bp);
119	}
 
120
121	free_vm86(t);
 
 
 
 
122
123	fpu__drop(fpu);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
124}
125
126void flush_thread(void)
127{
128	struct task_struct *tsk = current;
129
130	flush_ptrace_hw_breakpoint(tsk);
131	memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
132
133	fpu__clear(&tsk->thread.fpu);
 
 
 
 
134}
135
136static void hard_disable_TSC(void)
137{
138	cr4_set_bits(X86_CR4_TSD);
139}
140
141void disable_TSC(void)
142{
143	preempt_disable();
144	if (!test_and_set_thread_flag(TIF_NOTSC))
145		/*
146		 * Must flip the CPU state synchronously with
147		 * TIF_NOTSC in the current running context.
148		 */
149		hard_disable_TSC();
150	preempt_enable();
151}
152
153static void hard_enable_TSC(void)
154{
155	cr4_clear_bits(X86_CR4_TSD);
156}
157
158static void enable_TSC(void)
159{
160	preempt_disable();
161	if (test_and_clear_thread_flag(TIF_NOTSC))
162		/*
163		 * Must flip the CPU state synchronously with
164		 * TIF_NOTSC in the current running context.
165		 */
166		hard_enable_TSC();
167	preempt_enable();
168}
169
170int get_tsc_mode(unsigned long adr)
171{
172	unsigned int val;
173
174	if (test_thread_flag(TIF_NOTSC))
175		val = PR_TSC_SIGSEGV;
176	else
177		val = PR_TSC_ENABLE;
178
179	return put_user(val, (unsigned int __user *)adr);
180}
181
182int set_tsc_mode(unsigned int val)
183{
184	if (val == PR_TSC_SIGSEGV)
185		disable_TSC();
186	else if (val == PR_TSC_ENABLE)
187		enable_TSC();
188	else
189		return -EINVAL;
190
191	return 0;
192}
193
194void __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p,
195		      struct tss_struct *tss)
196{
197	struct thread_struct *prev, *next;
198
199	prev = &prev_p->thread;
200	next = &next_p->thread;
201
202	if (test_tsk_thread_flag(prev_p, TIF_BLOCKSTEP) ^
203	    test_tsk_thread_flag(next_p, TIF_BLOCKSTEP)) {
204		unsigned long debugctl = get_debugctlmsr();
205
206		debugctl &= ~DEBUGCTLMSR_BTF;
207		if (test_tsk_thread_flag(next_p, TIF_BLOCKSTEP))
208			debugctl |= DEBUGCTLMSR_BTF;
209
210		update_debugctlmsr(debugctl);
211	}
212
213	if (test_tsk_thread_flag(prev_p, TIF_NOTSC) ^
214	    test_tsk_thread_flag(next_p, TIF_NOTSC)) {
215		/* prev and next are different */
216		if (test_tsk_thread_flag(next_p, TIF_NOTSC))
217			hard_disable_TSC();
218		else
219			hard_enable_TSC();
220	}
221
222	if (test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
223		/*
224		 * Copy the relevant range of the IO bitmap.
225		 * Normally this is 128 bytes or less:
226		 */
227		memcpy(tss->io_bitmap, next->io_bitmap_ptr,
228		       max(prev->io_bitmap_max, next->io_bitmap_max));
229	} else if (test_tsk_thread_flag(prev_p, TIF_IO_BITMAP)) {
230		/*
231		 * Clear any possible leftover bits:
232		 */
233		memset(tss->io_bitmap, 0xff, prev->io_bitmap_max);
234	}
235	propagate_user_return_notify(prev_p, next_p);
236}
237
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
238/*
239 * Idle related variables and functions
240 */
241unsigned long boot_option_idle_override = IDLE_NO_OVERRIDE;
242EXPORT_SYMBOL(boot_option_idle_override);
243
244static void (*x86_idle)(void);
 
 
 
 
 
 
245
246#ifndef CONFIG_SMP
247static inline void play_dead(void)
 
 
 
 
 
248{
249	BUG();
250}
251#endif
252
253#ifdef CONFIG_X86_64
254void enter_idle(void)
255{
256	this_cpu_write(is_idle, 1);
257	atomic_notifier_call_chain(&idle_notifier, IDLE_START, NULL);
258}
 
259
260static void __exit_idle(void)
261{
262	if (x86_test_and_clear_bit_percpu(0, is_idle) == 0)
263		return;
264	atomic_notifier_call_chain(&idle_notifier, IDLE_END, NULL);
265}
266
267/* Called from interrupts to signify idle end */
268void exit_idle(void)
269{
270	/* idle loop has pid 0 */
271	if (current->pid)
272		return;
273	__exit_idle();
274}
275#endif
276
277void arch_cpu_idle_enter(void)
 
 
 
 
278{
279	local_touch_nmi();
280	enter_idle();
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
281}
 
 
 
282
283void arch_cpu_idle_exit(void)
284{
285	__exit_idle();
 
 
 
 
 
 
 
 
 
 
286}
287
288void arch_cpu_idle_dead(void)
289{
290	play_dead();
291}
292
293/*
294 * Called from the generic idle code.
 
 
 
 
 
295 */
296void arch_cpu_idle(void)
297{
298	x86_idle();
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
299}
300
301/*
302 * We use this if we don't have any better idle routine..
 
 
303 */
304void default_idle(void)
305{
306	trace_cpu_idle_rcuidle(1, smp_processor_id());
307	safe_halt();
308	trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, smp_processor_id());
 
 
 
 
309}
310#ifdef CONFIG_APM_MODULE
311EXPORT_SYMBOL(default_idle);
312#endif
313
314#ifdef CONFIG_XEN
315bool xen_set_default_idle(void)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
316{
317	bool ret = !!x86_idle;
 
 
 
 
 
 
318
319	x86_idle = default_idle;
 
 
 
320
321	return ret;
322}
323#endif
324void stop_this_cpu(void *dummy)
325{
326	local_irq_disable();
327	/*
328	 * Remove this CPU:
 
329	 */
330	set_cpu_online(smp_processor_id(), false);
331	disable_local_APIC();
332	mcheck_cpu_clear(this_cpu_ptr(&cpu_info));
333
334	for (;;)
335		halt();
336}
337
338bool amd_e400_c1e_detected;
339EXPORT_SYMBOL(amd_e400_c1e_detected);
340
341static cpumask_var_t amd_e400_c1e_mask;
342
343void amd_e400_remove_cpu(int cpu)
344{
345	if (amd_e400_c1e_mask != NULL)
346		cpumask_clear_cpu(cpu, amd_e400_c1e_mask);
347}
348
349/*
350 * AMD Erratum 400 aware idle routine. We check for C1E active in the interrupt
351 * pending message MSR. If we detect C1E, then we handle it the same
352 * way as C3 power states (local apic timer and TSC stop)
353 */
354static void amd_e400_idle(void)
355{
 
 
 
356	if (!amd_e400_c1e_detected) {
357		u32 lo, hi;
358
359		rdmsr(MSR_K8_INT_PENDING_MSG, lo, hi);
360
361		if (lo & K8_INTP_C1E_ACTIVE_MASK) {
362			amd_e400_c1e_detected = true;
363			if (!boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
364				mark_tsc_unstable("TSC halt in AMD C1E");
365			pr_info("System has AMD C1E enabled\n");
366		}
367	}
368
369	if (amd_e400_c1e_detected) {
370		int cpu = smp_processor_id();
371
372		if (!cpumask_test_cpu(cpu, amd_e400_c1e_mask)) {
373			cpumask_set_cpu(cpu, amd_e400_c1e_mask);
374			/* Force broadcast so ACPI can not interfere. */
375			tick_broadcast_force();
376			pr_info("Switch to broadcast mode on CPU%d\n", cpu);
 
 
 
 
377		}
378		tick_broadcast_enter();
379
380		default_idle();
381
382		/*
383		 * The switch back from broadcast mode needs to be
384		 * called with interrupts disabled.
385		 */
386		local_irq_disable();
387		tick_broadcast_exit();
388		local_irq_enable();
389	} else
390		default_idle();
391}
392
393/*
394 * Intel Core2 and older machines prefer MWAIT over HALT for C1.
395 * We can't rely on cpuidle installing MWAIT, because it will not load
396 * on systems that support only C1 -- so the boot default must be MWAIT.
397 *
398 * Some AMD machines are the opposite, they depend on using HALT.
399 *
400 * So for default C1, which is used during boot until cpuidle loads,
401 * use MWAIT-C1 on Intel HW that has it, else use HALT.
402 */
403static int prefer_mwait_c1_over_halt(const struct cpuinfo_x86 *c)
404{
405	if (c->x86_vendor != X86_VENDOR_INTEL)
406		return 0;
407
408	if (!cpu_has(c, X86_FEATURE_MWAIT))
409		return 0;
410
411	return 1;
412}
413
414/*
415 * MONITOR/MWAIT with no hints, used for default C1 state. This invokes MWAIT
416 * with interrupts enabled and no flags, which is backwards compatible with the
417 * original MWAIT implementation.
418 */
419static void mwait_idle(void)
420{
421	if (!current_set_polling_and_test()) {
422		trace_cpu_idle_rcuidle(1, smp_processor_id());
423		if (this_cpu_has(X86_BUG_CLFLUSH_MONITOR)) {
424			mb(); /* quirk */
425			clflush((void *)&current_thread_info()->flags);
426			mb(); /* quirk */
427		}
428
429		__monitor((void *)&current_thread_info()->flags, 0, 0);
430		if (!need_resched())
431			__sti_mwait(0, 0);
432		else
433			local_irq_enable();
434		trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, smp_processor_id());
435	} else {
436		local_irq_enable();
437	}
438	__current_clr_polling();
439}
440
441void select_idle_routine(const struct cpuinfo_x86 *c)
442{
443#ifdef CONFIG_SMP
444	if (boot_option_idle_override == IDLE_POLL && smp_num_siblings > 1)
445		pr_warn_once("WARNING: polling idle and HT enabled, performance may degrade\n");
446#endif
447	if (x86_idle || boot_option_idle_override == IDLE_POLL)
448		return;
449
450	if (cpu_has_bug(c, X86_BUG_AMD_APIC_C1E)) {
 
 
 
 
 
 
451		/* E400: APIC timer interrupt does not wake up CPU from C1e */
452		pr_info("using AMD E400 aware idle routine\n");
453		x86_idle = amd_e400_idle;
454	} else if (prefer_mwait_c1_over_halt(c)) {
455		pr_info("using mwait in idle threads\n");
456		x86_idle = mwait_idle;
457	} else
458		x86_idle = default_idle;
459}
460
461void __init init_amd_e400_c1e_mask(void)
462{
463	/* If we're using amd_e400_idle, we need to allocate amd_e400_c1e_mask. */
464	if (x86_idle == amd_e400_idle)
465		zalloc_cpumask_var(&amd_e400_c1e_mask, GFP_KERNEL);
466}
467
468static int __init idle_setup(char *str)
469{
470	if (!str)
471		return -EINVAL;
472
473	if (!strcmp(str, "poll")) {
474		pr_info("using polling idle threads\n");
 
475		boot_option_idle_override = IDLE_POLL;
476		cpu_idle_poll_ctrl(true);
 
 
477	} else if (!strcmp(str, "halt")) {
478		/*
479		 * When the boot option of idle=halt is added, halt is
480		 * forced to be used for CPU idle. In such case CPU C2/C3
481		 * won't be used again.
482		 * To continue to load the CPU idle driver, don't touch
483		 * the boot_option_idle_override.
484		 */
485		x86_idle = default_idle;
486		boot_option_idle_override = IDLE_HALT;
487	} else if (!strcmp(str, "nomwait")) {
488		/*
489		 * If the boot option of "idle=nomwait" is added,
490		 * it means that mwait will be disabled for CPU C2/C3
491		 * states. In such case it won't touch the variable
492		 * of boot_option_idle_override.
493		 */
494		boot_option_idle_override = IDLE_NOMWAIT;
495	} else
496		return -1;
497
498	return 0;
499}
500early_param("idle", idle_setup);
501
502unsigned long arch_align_stack(unsigned long sp)
503{
504	if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
505		sp -= get_random_int() % 8192;
506	return sp & ~0xf;
507}
508
509unsigned long arch_randomize_brk(struct mm_struct *mm)
510{
511	unsigned long range_end = mm->brk + 0x02000000;
512	return randomize_range(mm->brk, range_end, 0) ? : mm->brk;
513}
514
515/*
516 * Called from fs/proc with a reference on @p to find the function
517 * which called into schedule(). This needs to be done carefully
518 * because the task might wake up and we might look at a stack
519 * changing under us.
520 */
521unsigned long get_wchan(struct task_struct *p)
522{
523	unsigned long start, bottom, top, sp, fp, ip;
524	int count = 0;
525
526	if (!p || p == current || p->state == TASK_RUNNING)
527		return 0;
528
529	start = (unsigned long)task_stack_page(p);
530	if (!start)
531		return 0;
532
533	/*
534	 * Layout of the stack page:
535	 *
536	 * ----------- topmax = start + THREAD_SIZE - sizeof(unsigned long)
537	 * PADDING
538	 * ----------- top = topmax - TOP_OF_KERNEL_STACK_PADDING
539	 * stack
540	 * ----------- bottom = start + sizeof(thread_info)
541	 * thread_info
542	 * ----------- start
543	 *
544	 * The tasks stack pointer points at the location where the
545	 * framepointer is stored. The data on the stack is:
546	 * ... IP FP ... IP FP
547	 *
548	 * We need to read FP and IP, so we need to adjust the upper
549	 * bound by another unsigned long.
550	 */
551	top = start + THREAD_SIZE - TOP_OF_KERNEL_STACK_PADDING;
552	top -= 2 * sizeof(unsigned long);
553	bottom = start + sizeof(struct thread_info);
554
555	sp = READ_ONCE(p->thread.sp);
556	if (sp < bottom || sp > top)
557		return 0;
558
559	fp = READ_ONCE_NOCHECK(*(unsigned long *)sp);
560	do {
561		if (fp < bottom || fp > top)
562			return 0;
563		ip = READ_ONCE_NOCHECK(*(unsigned long *)(fp + sizeof(unsigned long)));
564		if (!in_sched_functions(ip))
565			return ip;
566		fp = READ_ONCE_NOCHECK(*(unsigned long *)fp);
567	} while (count++ < 16 && p->state != TASK_RUNNING);
568	return 0;
569}