Loading...
1// SPDX-License-Identifier: GPL-2.0
2#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
3
4#include <linux/errno.h>
5#include <linux/kernel.h>
6#include <linux/mm.h>
7#include <linux/smp.h>
8#include <linux/prctl.h>
9#include <linux/slab.h>
10#include <linux/sched.h>
11#include <linux/sched/idle.h>
12#include <linux/sched/debug.h>
13#include <linux/sched/task.h>
14#include <linux/sched/task_stack.h>
15#include <linux/init.h>
16#include <linux/export.h>
17#include <linux/pm.h>
18#include <linux/tick.h>
19#include <linux/random.h>
20#include <linux/user-return-notifier.h>
21#include <linux/dmi.h>
22#include <linux/utsname.h>
23#include <linux/stackprotector.h>
24#include <linux/cpuidle.h>
25#include <trace/events/power.h>
26#include <linux/hw_breakpoint.h>
27#include <asm/cpu.h>
28#include <asm/apic.h>
29#include <asm/syscalls.h>
30#include <linux/uaccess.h>
31#include <asm/mwait.h>
32#include <asm/fpu/internal.h>
33#include <asm/debugreg.h>
34#include <asm/nmi.h>
35#include <asm/tlbflush.h>
36#include <asm/mce.h>
37#include <asm/vm86.h>
38#include <asm/switch_to.h>
39#include <asm/desc.h>
40#include <asm/prctl.h>
41#include <asm/spec-ctrl.h>
42
43/*
44 * per-CPU TSS segments. Threads are completely 'soft' on Linux,
45 * no more per-task TSS's. The TSS size is kept cacheline-aligned
46 * so they are allowed to end up in the .data..cacheline_aligned
47 * section. Since TSS's are completely CPU-local, we want them
48 * on exact cacheline boundaries, to eliminate cacheline ping-pong.
49 */
50__visible DEFINE_PER_CPU_PAGE_ALIGNED(struct tss_struct, cpu_tss_rw) = {
51 .x86_tss = {
52 /*
53 * .sp0 is only used when entering ring 0 from a lower
54 * privilege level. Since the init task never runs anything
55 * but ring 0 code, there is no need for a valid value here.
56 * Poison it.
57 */
58 .sp0 = (1UL << (BITS_PER_LONG-1)) + 1,
59
60#ifdef CONFIG_X86_64
61 /*
62 * .sp1 is cpu_current_top_of_stack. The init task never
63 * runs user code, but cpu_current_top_of_stack should still
64 * be well defined before the first context switch.
65 */
66 .sp1 = TOP_OF_INIT_STACK,
67#endif
68
69#ifdef CONFIG_X86_32
70 .ss0 = __KERNEL_DS,
71 .ss1 = __KERNEL_CS,
72 .io_bitmap_base = INVALID_IO_BITMAP_OFFSET,
73#endif
74 },
75#ifdef CONFIG_X86_32
76 /*
77 * Note that the .io_bitmap member must be extra-big. This is because
78 * the CPU will access an additional byte beyond the end of the IO
79 * permission bitmap. The extra byte must be all 1 bits, and must
80 * be within the limit.
81 */
82 .io_bitmap = { [0 ... IO_BITMAP_LONGS] = ~0 },
83#endif
84};
85EXPORT_PER_CPU_SYMBOL(cpu_tss_rw);
86
87DEFINE_PER_CPU(bool, __tss_limit_invalid);
88EXPORT_PER_CPU_SYMBOL_GPL(__tss_limit_invalid);
89
90/*
91 * this gets called so that we can store lazy state into memory and copy the
92 * current task into the new thread.
93 */
94int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
95{
96 memcpy(dst, src, arch_task_struct_size);
97#ifdef CONFIG_VM86
98 dst->thread.vm86 = NULL;
99#endif
100
101 return fpu__copy(&dst->thread.fpu, &src->thread.fpu);
102}
103
104/*
105 * Free current thread data structures etc..
106 */
107void exit_thread(struct task_struct *tsk)
108{
109 struct thread_struct *t = &tsk->thread;
110 unsigned long *bp = t->io_bitmap_ptr;
111 struct fpu *fpu = &t->fpu;
112
113 if (bp) {
114 struct tss_struct *tss = &per_cpu(cpu_tss_rw, get_cpu());
115
116 t->io_bitmap_ptr = NULL;
117 clear_thread_flag(TIF_IO_BITMAP);
118 /*
119 * Careful, clear this in the TSS too:
120 */
121 memset(tss->io_bitmap, 0xff, t->io_bitmap_max);
122 t->io_bitmap_max = 0;
123 put_cpu();
124 kfree(bp);
125 }
126
127 free_vm86(t);
128
129 fpu__drop(fpu);
130}
131
132void flush_thread(void)
133{
134 struct task_struct *tsk = current;
135
136 flush_ptrace_hw_breakpoint(tsk);
137 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
138
139 fpu__clear(&tsk->thread.fpu);
140}
141
142void disable_TSC(void)
143{
144 preempt_disable();
145 if (!test_and_set_thread_flag(TIF_NOTSC))
146 /*
147 * Must flip the CPU state synchronously with
148 * TIF_NOTSC in the current running context.
149 */
150 cr4_set_bits(X86_CR4_TSD);
151 preempt_enable();
152}
153
154static void enable_TSC(void)
155{
156 preempt_disable();
157 if (test_and_clear_thread_flag(TIF_NOTSC))
158 /*
159 * Must flip the CPU state synchronously with
160 * TIF_NOTSC in the current running context.
161 */
162 cr4_clear_bits(X86_CR4_TSD);
163 preempt_enable();
164}
165
166int get_tsc_mode(unsigned long adr)
167{
168 unsigned int val;
169
170 if (test_thread_flag(TIF_NOTSC))
171 val = PR_TSC_SIGSEGV;
172 else
173 val = PR_TSC_ENABLE;
174
175 return put_user(val, (unsigned int __user *)adr);
176}
177
178int set_tsc_mode(unsigned int val)
179{
180 if (val == PR_TSC_SIGSEGV)
181 disable_TSC();
182 else if (val == PR_TSC_ENABLE)
183 enable_TSC();
184 else
185 return -EINVAL;
186
187 return 0;
188}
189
190DEFINE_PER_CPU(u64, msr_misc_features_shadow);
191
192static void set_cpuid_faulting(bool on)
193{
194 u64 msrval;
195
196 msrval = this_cpu_read(msr_misc_features_shadow);
197 msrval &= ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT;
198 msrval |= (on << MSR_MISC_FEATURES_ENABLES_CPUID_FAULT_BIT);
199 this_cpu_write(msr_misc_features_shadow, msrval);
200 wrmsrl(MSR_MISC_FEATURES_ENABLES, msrval);
201}
202
203static void disable_cpuid(void)
204{
205 preempt_disable();
206 if (!test_and_set_thread_flag(TIF_NOCPUID)) {
207 /*
208 * Must flip the CPU state synchronously with
209 * TIF_NOCPUID in the current running context.
210 */
211 set_cpuid_faulting(true);
212 }
213 preempt_enable();
214}
215
216static void enable_cpuid(void)
217{
218 preempt_disable();
219 if (test_and_clear_thread_flag(TIF_NOCPUID)) {
220 /*
221 * Must flip the CPU state synchronously with
222 * TIF_NOCPUID in the current running context.
223 */
224 set_cpuid_faulting(false);
225 }
226 preempt_enable();
227}
228
229static int get_cpuid_mode(void)
230{
231 return !test_thread_flag(TIF_NOCPUID);
232}
233
234static int set_cpuid_mode(struct task_struct *task, unsigned long cpuid_enabled)
235{
236 if (!static_cpu_has(X86_FEATURE_CPUID_FAULT))
237 return -ENODEV;
238
239 if (cpuid_enabled)
240 enable_cpuid();
241 else
242 disable_cpuid();
243
244 return 0;
245}
246
247/*
248 * Called immediately after a successful exec.
249 */
250void arch_setup_new_exec(void)
251{
252 /* If cpuid was previously disabled for this task, re-enable it. */
253 if (test_thread_flag(TIF_NOCPUID))
254 enable_cpuid();
255}
256
257static inline void switch_to_bitmap(struct tss_struct *tss,
258 struct thread_struct *prev,
259 struct thread_struct *next,
260 unsigned long tifp, unsigned long tifn)
261{
262 if (tifn & _TIF_IO_BITMAP) {
263 /*
264 * Copy the relevant range of the IO bitmap.
265 * Normally this is 128 bytes or less:
266 */
267 memcpy(tss->io_bitmap, next->io_bitmap_ptr,
268 max(prev->io_bitmap_max, next->io_bitmap_max));
269 /*
270 * Make sure that the TSS limit is correct for the CPU
271 * to notice the IO bitmap.
272 */
273 refresh_tss_limit();
274 } else if (tifp & _TIF_IO_BITMAP) {
275 /*
276 * Clear any possible leftover bits:
277 */
278 memset(tss->io_bitmap, 0xff, prev->io_bitmap_max);
279 }
280}
281
282#ifdef CONFIG_SMP
283
284struct ssb_state {
285 struct ssb_state *shared_state;
286 raw_spinlock_t lock;
287 unsigned int disable_state;
288 unsigned long local_state;
289};
290
291#define LSTATE_SSB 0
292
293static DEFINE_PER_CPU(struct ssb_state, ssb_state);
294
295void speculative_store_bypass_ht_init(void)
296{
297 struct ssb_state *st = this_cpu_ptr(&ssb_state);
298 unsigned int this_cpu = smp_processor_id();
299 unsigned int cpu;
300
301 st->local_state = 0;
302
303 /*
304 * Shared state setup happens once on the first bringup
305 * of the CPU. It's not destroyed on CPU hotunplug.
306 */
307 if (st->shared_state)
308 return;
309
310 raw_spin_lock_init(&st->lock);
311
312 /*
313 * Go over HT siblings and check whether one of them has set up the
314 * shared state pointer already.
315 */
316 for_each_cpu(cpu, topology_sibling_cpumask(this_cpu)) {
317 if (cpu == this_cpu)
318 continue;
319
320 if (!per_cpu(ssb_state, cpu).shared_state)
321 continue;
322
323 /* Link it to the state of the sibling: */
324 st->shared_state = per_cpu(ssb_state, cpu).shared_state;
325 return;
326 }
327
328 /*
329 * First HT sibling to come up on the core. Link shared state of
330 * the first HT sibling to itself. The siblings on the same core
331 * which come up later will see the shared state pointer and link
332 * themself to the state of this CPU.
333 */
334 st->shared_state = st;
335}
336
337/*
338 * Logic is: First HT sibling enables SSBD for both siblings in the core
339 * and last sibling to disable it, disables it for the whole core. This how
340 * MSR_SPEC_CTRL works in "hardware":
341 *
342 * CORE_SPEC_CTRL = THREAD0_SPEC_CTRL | THREAD1_SPEC_CTRL
343 */
344static __always_inline void amd_set_core_ssb_state(unsigned long tifn)
345{
346 struct ssb_state *st = this_cpu_ptr(&ssb_state);
347 u64 msr = x86_amd_ls_cfg_base;
348
349 if (!static_cpu_has(X86_FEATURE_ZEN)) {
350 msr |= ssbd_tif_to_amd_ls_cfg(tifn);
351 wrmsrl(MSR_AMD64_LS_CFG, msr);
352 return;
353 }
354
355 if (tifn & _TIF_SSBD) {
356 /*
357 * Since this can race with prctl(), block reentry on the
358 * same CPU.
359 */
360 if (__test_and_set_bit(LSTATE_SSB, &st->local_state))
361 return;
362
363 msr |= x86_amd_ls_cfg_ssbd_mask;
364
365 raw_spin_lock(&st->shared_state->lock);
366 /* First sibling enables SSBD: */
367 if (!st->shared_state->disable_state)
368 wrmsrl(MSR_AMD64_LS_CFG, msr);
369 st->shared_state->disable_state++;
370 raw_spin_unlock(&st->shared_state->lock);
371 } else {
372 if (!__test_and_clear_bit(LSTATE_SSB, &st->local_state))
373 return;
374
375 raw_spin_lock(&st->shared_state->lock);
376 st->shared_state->disable_state--;
377 if (!st->shared_state->disable_state)
378 wrmsrl(MSR_AMD64_LS_CFG, msr);
379 raw_spin_unlock(&st->shared_state->lock);
380 }
381}
382#else
383static __always_inline void amd_set_core_ssb_state(unsigned long tifn)
384{
385 u64 msr = x86_amd_ls_cfg_base | ssbd_tif_to_amd_ls_cfg(tifn);
386
387 wrmsrl(MSR_AMD64_LS_CFG, msr);
388}
389#endif
390
391static __always_inline void amd_set_ssb_virt_state(unsigned long tifn)
392{
393 /*
394 * SSBD has the same definition in SPEC_CTRL and VIRT_SPEC_CTRL,
395 * so ssbd_tif_to_spec_ctrl() just works.
396 */
397 wrmsrl(MSR_AMD64_VIRT_SPEC_CTRL, ssbd_tif_to_spec_ctrl(tifn));
398}
399
400static __always_inline void intel_set_ssb_state(unsigned long tifn)
401{
402 u64 msr = x86_spec_ctrl_base | ssbd_tif_to_spec_ctrl(tifn);
403
404 wrmsrl(MSR_IA32_SPEC_CTRL, msr);
405}
406
407static __always_inline void __speculative_store_bypass_update(unsigned long tifn)
408{
409 if (static_cpu_has(X86_FEATURE_VIRT_SSBD))
410 amd_set_ssb_virt_state(tifn);
411 else if (static_cpu_has(X86_FEATURE_LS_CFG_SSBD))
412 amd_set_core_ssb_state(tifn);
413 else
414 intel_set_ssb_state(tifn);
415}
416
417void speculative_store_bypass_update(unsigned long tif)
418{
419 preempt_disable();
420 __speculative_store_bypass_update(tif);
421 preempt_enable();
422}
423
424void __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p,
425 struct tss_struct *tss)
426{
427 struct thread_struct *prev, *next;
428 unsigned long tifp, tifn;
429
430 prev = &prev_p->thread;
431 next = &next_p->thread;
432
433 tifn = READ_ONCE(task_thread_info(next_p)->flags);
434 tifp = READ_ONCE(task_thread_info(prev_p)->flags);
435 switch_to_bitmap(tss, prev, next, tifp, tifn);
436
437 propagate_user_return_notify(prev_p, next_p);
438
439 if ((tifp & _TIF_BLOCKSTEP || tifn & _TIF_BLOCKSTEP) &&
440 arch_has_block_step()) {
441 unsigned long debugctl, msk;
442
443 rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctl);
444 debugctl &= ~DEBUGCTLMSR_BTF;
445 msk = tifn & _TIF_BLOCKSTEP;
446 debugctl |= (msk >> TIF_BLOCKSTEP) << DEBUGCTLMSR_BTF_SHIFT;
447 wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctl);
448 }
449
450 if ((tifp ^ tifn) & _TIF_NOTSC)
451 cr4_toggle_bits_irqsoff(X86_CR4_TSD);
452
453 if ((tifp ^ tifn) & _TIF_NOCPUID)
454 set_cpuid_faulting(!!(tifn & _TIF_NOCPUID));
455
456 if ((tifp ^ tifn) & _TIF_SSBD)
457 __speculative_store_bypass_update(tifn);
458}
459
460/*
461 * Idle related variables and functions
462 */
463unsigned long boot_option_idle_override = IDLE_NO_OVERRIDE;
464EXPORT_SYMBOL(boot_option_idle_override);
465
466static void (*x86_idle)(void);
467
468#ifndef CONFIG_SMP
469static inline void play_dead(void)
470{
471 BUG();
472}
473#endif
474
475void arch_cpu_idle_enter(void)
476{
477 tsc_verify_tsc_adjust(false);
478 local_touch_nmi();
479}
480
481void arch_cpu_idle_dead(void)
482{
483 play_dead();
484}
485
486/*
487 * Called from the generic idle code.
488 */
489void arch_cpu_idle(void)
490{
491 x86_idle();
492}
493
494/*
495 * We use this if we don't have any better idle routine..
496 */
497void __cpuidle default_idle(void)
498{
499 trace_cpu_idle_rcuidle(1, smp_processor_id());
500 safe_halt();
501 trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, smp_processor_id());
502}
503#ifdef CONFIG_APM_MODULE
504EXPORT_SYMBOL(default_idle);
505#endif
506
507#ifdef CONFIG_XEN
508bool xen_set_default_idle(void)
509{
510 bool ret = !!x86_idle;
511
512 x86_idle = default_idle;
513
514 return ret;
515}
516#endif
517
518void stop_this_cpu(void *dummy)
519{
520 local_irq_disable();
521 /*
522 * Remove this CPU:
523 */
524 set_cpu_online(smp_processor_id(), false);
525 disable_local_APIC();
526 mcheck_cpu_clear(this_cpu_ptr(&cpu_info));
527
528 /*
529 * Use wbinvd on processors that support SME. This provides support
530 * for performing a successful kexec when going from SME inactive
531 * to SME active (or vice-versa). The cache must be cleared so that
532 * if there are entries with the same physical address, both with and
533 * without the encryption bit, they don't race each other when flushed
534 * and potentially end up with the wrong entry being committed to
535 * memory.
536 */
537 if (boot_cpu_has(X86_FEATURE_SME))
538 native_wbinvd();
539 for (;;) {
540 /*
541 * Use native_halt() so that memory contents don't change
542 * (stack usage and variables) after possibly issuing the
543 * native_wbinvd() above.
544 */
545 native_halt();
546 }
547}
548
549/*
550 * AMD Erratum 400 aware idle routine. We handle it the same way as C3 power
551 * states (local apic timer and TSC stop).
552 */
553static void amd_e400_idle(void)
554{
555 /*
556 * We cannot use static_cpu_has_bug() here because X86_BUG_AMD_APIC_C1E
557 * gets set after static_cpu_has() places have been converted via
558 * alternatives.
559 */
560 if (!boot_cpu_has_bug(X86_BUG_AMD_APIC_C1E)) {
561 default_idle();
562 return;
563 }
564
565 tick_broadcast_enter();
566
567 default_idle();
568
569 /*
570 * The switch back from broadcast mode needs to be called with
571 * interrupts disabled.
572 */
573 local_irq_disable();
574 tick_broadcast_exit();
575 local_irq_enable();
576}
577
578/*
579 * Intel Core2 and older machines prefer MWAIT over HALT for C1.
580 * We can't rely on cpuidle installing MWAIT, because it will not load
581 * on systems that support only C1 -- so the boot default must be MWAIT.
582 *
583 * Some AMD machines are the opposite, they depend on using HALT.
584 *
585 * So for default C1, which is used during boot until cpuidle loads,
586 * use MWAIT-C1 on Intel HW that has it, else use HALT.
587 */
588static int prefer_mwait_c1_over_halt(const struct cpuinfo_x86 *c)
589{
590 if (c->x86_vendor != X86_VENDOR_INTEL)
591 return 0;
592
593 if (!cpu_has(c, X86_FEATURE_MWAIT) || static_cpu_has_bug(X86_BUG_MONITOR))
594 return 0;
595
596 return 1;
597}
598
599/*
600 * MONITOR/MWAIT with no hints, used for default C1 state. This invokes MWAIT
601 * with interrupts enabled and no flags, which is backwards compatible with the
602 * original MWAIT implementation.
603 */
604static __cpuidle void mwait_idle(void)
605{
606 if (!current_set_polling_and_test()) {
607 trace_cpu_idle_rcuidle(1, smp_processor_id());
608 if (this_cpu_has(X86_BUG_CLFLUSH_MONITOR)) {
609 mb(); /* quirk */
610 clflush((void *)¤t_thread_info()->flags);
611 mb(); /* quirk */
612 }
613
614 __monitor((void *)¤t_thread_info()->flags, 0, 0);
615 if (!need_resched())
616 __sti_mwait(0, 0);
617 else
618 local_irq_enable();
619 trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, smp_processor_id());
620 } else {
621 local_irq_enable();
622 }
623 __current_clr_polling();
624}
625
626void select_idle_routine(const struct cpuinfo_x86 *c)
627{
628#ifdef CONFIG_SMP
629 if (boot_option_idle_override == IDLE_POLL && smp_num_siblings > 1)
630 pr_warn_once("WARNING: polling idle and HT enabled, performance may degrade\n");
631#endif
632 if (x86_idle || boot_option_idle_override == IDLE_POLL)
633 return;
634
635 if (boot_cpu_has_bug(X86_BUG_AMD_E400)) {
636 pr_info("using AMD E400 aware idle routine\n");
637 x86_idle = amd_e400_idle;
638 } else if (prefer_mwait_c1_over_halt(c)) {
639 pr_info("using mwait in idle threads\n");
640 x86_idle = mwait_idle;
641 } else
642 x86_idle = default_idle;
643}
644
645void amd_e400_c1e_apic_setup(void)
646{
647 if (boot_cpu_has_bug(X86_BUG_AMD_APIC_C1E)) {
648 pr_info("Switch to broadcast mode on CPU%d\n", smp_processor_id());
649 local_irq_disable();
650 tick_broadcast_force();
651 local_irq_enable();
652 }
653}
654
655void __init arch_post_acpi_subsys_init(void)
656{
657 u32 lo, hi;
658
659 if (!boot_cpu_has_bug(X86_BUG_AMD_E400))
660 return;
661
662 /*
663 * AMD E400 detection needs to happen after ACPI has been enabled. If
664 * the machine is affected K8_INTP_C1E_ACTIVE_MASK bits are set in
665 * MSR_K8_INT_PENDING_MSG.
666 */
667 rdmsr(MSR_K8_INT_PENDING_MSG, lo, hi);
668 if (!(lo & K8_INTP_C1E_ACTIVE_MASK))
669 return;
670
671 boot_cpu_set_bug(X86_BUG_AMD_APIC_C1E);
672
673 if (!boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
674 mark_tsc_unstable("TSC halt in AMD C1E");
675 pr_info("System has AMD C1E enabled\n");
676}
677
678static int __init idle_setup(char *str)
679{
680 if (!str)
681 return -EINVAL;
682
683 if (!strcmp(str, "poll")) {
684 pr_info("using polling idle threads\n");
685 boot_option_idle_override = IDLE_POLL;
686 cpu_idle_poll_ctrl(true);
687 } else if (!strcmp(str, "halt")) {
688 /*
689 * When the boot option of idle=halt is added, halt is
690 * forced to be used for CPU idle. In such case CPU C2/C3
691 * won't be used again.
692 * To continue to load the CPU idle driver, don't touch
693 * the boot_option_idle_override.
694 */
695 x86_idle = default_idle;
696 boot_option_idle_override = IDLE_HALT;
697 } else if (!strcmp(str, "nomwait")) {
698 /*
699 * If the boot option of "idle=nomwait" is added,
700 * it means that mwait will be disabled for CPU C2/C3
701 * states. In such case it won't touch the variable
702 * of boot_option_idle_override.
703 */
704 boot_option_idle_override = IDLE_NOMWAIT;
705 } else
706 return -1;
707
708 return 0;
709}
710early_param("idle", idle_setup);
711
712unsigned long arch_align_stack(unsigned long sp)
713{
714 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
715 sp -= get_random_int() % 8192;
716 return sp & ~0xf;
717}
718
719unsigned long arch_randomize_brk(struct mm_struct *mm)
720{
721 return randomize_page(mm->brk, 0x02000000);
722}
723
724/*
725 * Called from fs/proc with a reference on @p to find the function
726 * which called into schedule(). This needs to be done carefully
727 * because the task might wake up and we might look at a stack
728 * changing under us.
729 */
730unsigned long get_wchan(struct task_struct *p)
731{
732 unsigned long start, bottom, top, sp, fp, ip, ret = 0;
733 int count = 0;
734
735 if (!p || p == current || p->state == TASK_RUNNING)
736 return 0;
737
738 if (!try_get_task_stack(p))
739 return 0;
740
741 start = (unsigned long)task_stack_page(p);
742 if (!start)
743 goto out;
744
745 /*
746 * Layout of the stack page:
747 *
748 * ----------- topmax = start + THREAD_SIZE - sizeof(unsigned long)
749 * PADDING
750 * ----------- top = topmax - TOP_OF_KERNEL_STACK_PADDING
751 * stack
752 * ----------- bottom = start
753 *
754 * The tasks stack pointer points at the location where the
755 * framepointer is stored. The data on the stack is:
756 * ... IP FP ... IP FP
757 *
758 * We need to read FP and IP, so we need to adjust the upper
759 * bound by another unsigned long.
760 */
761 top = start + THREAD_SIZE - TOP_OF_KERNEL_STACK_PADDING;
762 top -= 2 * sizeof(unsigned long);
763 bottom = start;
764
765 sp = READ_ONCE(p->thread.sp);
766 if (sp < bottom || sp > top)
767 goto out;
768
769 fp = READ_ONCE_NOCHECK(((struct inactive_task_frame *)sp)->bp);
770 do {
771 if (fp < bottom || fp > top)
772 goto out;
773 ip = READ_ONCE_NOCHECK(*(unsigned long *)(fp + sizeof(unsigned long)));
774 if (!in_sched_functions(ip)) {
775 ret = ip;
776 goto out;
777 }
778 fp = READ_ONCE_NOCHECK(*(unsigned long *)fp);
779 } while (count++ < 16 && p->state != TASK_RUNNING);
780
781out:
782 put_task_stack(p);
783 return ret;
784}
785
786long do_arch_prctl_common(struct task_struct *task, int option,
787 unsigned long cpuid_enabled)
788{
789 switch (option) {
790 case ARCH_GET_CPUID:
791 return get_cpuid_mode();
792 case ARCH_SET_CPUID:
793 return set_cpuid_mode(task, cpuid_enabled);
794 }
795
796 return -EINVAL;
797}
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