xref: /linux/arch/x86/kernel/process.c (revision aaa44952bbd1d4db14a4d676bf9595bb5db7e7b0)
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 <linux/acpi.h>
26 #include <linux/elf-randomize.h>
27 #include <trace/events/power.h>
28 #include <linux/hw_breakpoint.h>
29 #include <asm/cpu.h>
30 #include <asm/apic.h>
31 #include <linux/uaccess.h>
32 #include <asm/mwait.h>
33 #include <asm/fpu/internal.h>
34 #include <asm/debugreg.h>
35 #include <asm/nmi.h>
36 #include <asm/tlbflush.h>
37 #include <asm/mce.h>
38 #include <asm/vm86.h>
39 #include <asm/switch_to.h>
40 #include <asm/desc.h>
41 #include <asm/prctl.h>
42 #include <asm/spec-ctrl.h>
43 #include <asm/io_bitmap.h>
44 #include <asm/proto.h>
45 #include <asm/frame.h>
46 
47 #include "process.h"
48 
49 /*
50  * per-CPU TSS segments. Threads are completely 'soft' on Linux,
51  * no more per-task TSS's. The TSS size is kept cacheline-aligned
52  * so they are allowed to end up in the .data..cacheline_aligned
53  * section. Since TSS's are completely CPU-local, we want them
54  * on exact cacheline boundaries, to eliminate cacheline ping-pong.
55  */
56 __visible DEFINE_PER_CPU_PAGE_ALIGNED(struct tss_struct, cpu_tss_rw) = {
57 	.x86_tss = {
58 		/*
59 		 * .sp0 is only used when entering ring 0 from a lower
60 		 * privilege level.  Since the init task never runs anything
61 		 * but ring 0 code, there is no need for a valid value here.
62 		 * Poison it.
63 		 */
64 		.sp0 = (1UL << (BITS_PER_LONG-1)) + 1,
65 
66 #ifdef CONFIG_X86_32
67 		.sp1 = TOP_OF_INIT_STACK,
68 
69 		.ss0 = __KERNEL_DS,
70 		.ss1 = __KERNEL_CS,
71 #endif
72 		.io_bitmap_base	= IO_BITMAP_OFFSET_INVALID,
73 	 },
74 };
75 EXPORT_PER_CPU_SYMBOL(cpu_tss_rw);
76 
77 DEFINE_PER_CPU(bool, __tss_limit_invalid);
78 EXPORT_PER_CPU_SYMBOL_GPL(__tss_limit_invalid);
79 
80 /*
81  * this gets called so that we can store lazy state into memory and copy the
82  * current task into the new thread.
83  */
84 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
85 {
86 	memcpy(dst, src, arch_task_struct_size);
87 #ifdef CONFIG_VM86
88 	dst->thread.vm86 = NULL;
89 #endif
90 
91 	return fpu__copy(dst, src);
92 }
93 
94 /*
95  * Free thread data structures etc..
96  */
97 void exit_thread(struct task_struct *tsk)
98 {
99 	struct thread_struct *t = &tsk->thread;
100 	struct fpu *fpu = &t->fpu;
101 
102 	if (test_thread_flag(TIF_IO_BITMAP))
103 		io_bitmap_exit(tsk);
104 
105 	free_vm86(t);
106 
107 	fpu__drop(fpu);
108 }
109 
110 static int set_new_tls(struct task_struct *p, unsigned long tls)
111 {
112 	struct user_desc __user *utls = (struct user_desc __user *)tls;
113 
114 	if (in_ia32_syscall())
115 		return do_set_thread_area(p, -1, utls, 0);
116 	else
117 		return do_set_thread_area_64(p, ARCH_SET_FS, tls);
118 }
119 
120 int copy_thread(unsigned long clone_flags, unsigned long sp, unsigned long arg,
121 		struct task_struct *p, unsigned long tls)
122 {
123 	struct inactive_task_frame *frame;
124 	struct fork_frame *fork_frame;
125 	struct pt_regs *childregs;
126 	int ret = 0;
127 
128 	childregs = task_pt_regs(p);
129 	fork_frame = container_of(childregs, struct fork_frame, regs);
130 	frame = &fork_frame->frame;
131 
132 	frame->bp = encode_frame_pointer(childregs);
133 	frame->ret_addr = (unsigned long) ret_from_fork;
134 	p->thread.sp = (unsigned long) fork_frame;
135 	p->thread.io_bitmap = NULL;
136 	memset(p->thread.ptrace_bps, 0, sizeof(p->thread.ptrace_bps));
137 
138 #ifdef CONFIG_X86_64
139 	current_save_fsgs();
140 	p->thread.fsindex = current->thread.fsindex;
141 	p->thread.fsbase = current->thread.fsbase;
142 	p->thread.gsindex = current->thread.gsindex;
143 	p->thread.gsbase = current->thread.gsbase;
144 
145 	savesegment(es, p->thread.es);
146 	savesegment(ds, p->thread.ds);
147 #else
148 	p->thread.sp0 = (unsigned long) (childregs + 1);
149 	/*
150 	 * Clear all status flags including IF and set fixed bit. 64bit
151 	 * does not have this initialization as the frame does not contain
152 	 * flags. The flags consistency (especially vs. AC) is there
153 	 * ensured via objtool, which lacks 32bit support.
154 	 */
155 	frame->flags = X86_EFLAGS_FIXED;
156 #endif
157 
158 	/* Kernel thread ? */
159 	if (unlikely(p->flags & (PF_KTHREAD | PF_IO_WORKER))) {
160 		memset(childregs, 0, sizeof(struct pt_regs));
161 		kthread_frame_init(frame, sp, arg);
162 		return 0;
163 	}
164 
165 	frame->bx = 0;
166 	*childregs = *current_pt_regs();
167 	childregs->ax = 0;
168 	if (sp)
169 		childregs->sp = sp;
170 
171 #ifdef CONFIG_X86_32
172 	task_user_gs(p) = get_user_gs(current_pt_regs());
173 #endif
174 
175 	/* Set a new TLS for the child thread? */
176 	if (clone_flags & CLONE_SETTLS)
177 		ret = set_new_tls(p, tls);
178 
179 	if (!ret && unlikely(test_tsk_thread_flag(current, TIF_IO_BITMAP)))
180 		io_bitmap_share(p);
181 
182 	return ret;
183 }
184 
185 void flush_thread(void)
186 {
187 	struct task_struct *tsk = current;
188 
189 	flush_ptrace_hw_breakpoint(tsk);
190 	memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
191 
192 	fpu__clear_all(&tsk->thread.fpu);
193 }
194 
195 void disable_TSC(void)
196 {
197 	preempt_disable();
198 	if (!test_and_set_thread_flag(TIF_NOTSC))
199 		/*
200 		 * Must flip the CPU state synchronously with
201 		 * TIF_NOTSC in the current running context.
202 		 */
203 		cr4_set_bits(X86_CR4_TSD);
204 	preempt_enable();
205 }
206 
207 static void enable_TSC(void)
208 {
209 	preempt_disable();
210 	if (test_and_clear_thread_flag(TIF_NOTSC))
211 		/*
212 		 * Must flip the CPU state synchronously with
213 		 * TIF_NOTSC in the current running context.
214 		 */
215 		cr4_clear_bits(X86_CR4_TSD);
216 	preempt_enable();
217 }
218 
219 int get_tsc_mode(unsigned long adr)
220 {
221 	unsigned int val;
222 
223 	if (test_thread_flag(TIF_NOTSC))
224 		val = PR_TSC_SIGSEGV;
225 	else
226 		val = PR_TSC_ENABLE;
227 
228 	return put_user(val, (unsigned int __user *)adr);
229 }
230 
231 int set_tsc_mode(unsigned int val)
232 {
233 	if (val == PR_TSC_SIGSEGV)
234 		disable_TSC();
235 	else if (val == PR_TSC_ENABLE)
236 		enable_TSC();
237 	else
238 		return -EINVAL;
239 
240 	return 0;
241 }
242 
243 DEFINE_PER_CPU(u64, msr_misc_features_shadow);
244 
245 static void set_cpuid_faulting(bool on)
246 {
247 	u64 msrval;
248 
249 	msrval = this_cpu_read(msr_misc_features_shadow);
250 	msrval &= ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT;
251 	msrval |= (on << MSR_MISC_FEATURES_ENABLES_CPUID_FAULT_BIT);
252 	this_cpu_write(msr_misc_features_shadow, msrval);
253 	wrmsrl(MSR_MISC_FEATURES_ENABLES, msrval);
254 }
255 
256 static void disable_cpuid(void)
257 {
258 	preempt_disable();
259 	if (!test_and_set_thread_flag(TIF_NOCPUID)) {
260 		/*
261 		 * Must flip the CPU state synchronously with
262 		 * TIF_NOCPUID in the current running context.
263 		 */
264 		set_cpuid_faulting(true);
265 	}
266 	preempt_enable();
267 }
268 
269 static void enable_cpuid(void)
270 {
271 	preempt_disable();
272 	if (test_and_clear_thread_flag(TIF_NOCPUID)) {
273 		/*
274 		 * Must flip the CPU state synchronously with
275 		 * TIF_NOCPUID in the current running context.
276 		 */
277 		set_cpuid_faulting(false);
278 	}
279 	preempt_enable();
280 }
281 
282 static int get_cpuid_mode(void)
283 {
284 	return !test_thread_flag(TIF_NOCPUID);
285 }
286 
287 static int set_cpuid_mode(struct task_struct *task, unsigned long cpuid_enabled)
288 {
289 	if (!boot_cpu_has(X86_FEATURE_CPUID_FAULT))
290 		return -ENODEV;
291 
292 	if (cpuid_enabled)
293 		enable_cpuid();
294 	else
295 		disable_cpuid();
296 
297 	return 0;
298 }
299 
300 /*
301  * Called immediately after a successful exec.
302  */
303 void arch_setup_new_exec(void)
304 {
305 	/* If cpuid was previously disabled for this task, re-enable it. */
306 	if (test_thread_flag(TIF_NOCPUID))
307 		enable_cpuid();
308 
309 	/*
310 	 * Don't inherit TIF_SSBD across exec boundary when
311 	 * PR_SPEC_DISABLE_NOEXEC is used.
312 	 */
313 	if (test_thread_flag(TIF_SSBD) &&
314 	    task_spec_ssb_noexec(current)) {
315 		clear_thread_flag(TIF_SSBD);
316 		task_clear_spec_ssb_disable(current);
317 		task_clear_spec_ssb_noexec(current);
318 		speculation_ctrl_update(task_thread_info(current)->flags);
319 	}
320 }
321 
322 #ifdef CONFIG_X86_IOPL_IOPERM
323 static inline void switch_to_bitmap(unsigned long tifp)
324 {
325 	/*
326 	 * Invalidate I/O bitmap if the previous task used it. This prevents
327 	 * any possible leakage of an active I/O bitmap.
328 	 *
329 	 * If the next task has an I/O bitmap it will handle it on exit to
330 	 * user mode.
331 	 */
332 	if (tifp & _TIF_IO_BITMAP)
333 		tss_invalidate_io_bitmap();
334 }
335 
336 static void tss_copy_io_bitmap(struct tss_struct *tss, struct io_bitmap *iobm)
337 {
338 	/*
339 	 * Copy at least the byte range of the incoming tasks bitmap which
340 	 * covers the permitted I/O ports.
341 	 *
342 	 * If the previous task which used an I/O bitmap had more bits
343 	 * permitted, then the copy needs to cover those as well so they
344 	 * get turned off.
345 	 */
346 	memcpy(tss->io_bitmap.bitmap, iobm->bitmap,
347 	       max(tss->io_bitmap.prev_max, iobm->max));
348 
349 	/*
350 	 * Store the new max and the sequence number of this bitmap
351 	 * and a pointer to the bitmap itself.
352 	 */
353 	tss->io_bitmap.prev_max = iobm->max;
354 	tss->io_bitmap.prev_sequence = iobm->sequence;
355 }
356 
357 /**
358  * tss_update_io_bitmap - Update I/O bitmap before exiting to usermode
359  */
360 void native_tss_update_io_bitmap(void)
361 {
362 	struct tss_struct *tss = this_cpu_ptr(&cpu_tss_rw);
363 	struct thread_struct *t = &current->thread;
364 	u16 *base = &tss->x86_tss.io_bitmap_base;
365 
366 	if (!test_thread_flag(TIF_IO_BITMAP)) {
367 		native_tss_invalidate_io_bitmap();
368 		return;
369 	}
370 
371 	if (IS_ENABLED(CONFIG_X86_IOPL_IOPERM) && t->iopl_emul == 3) {
372 		*base = IO_BITMAP_OFFSET_VALID_ALL;
373 	} else {
374 		struct io_bitmap *iobm = t->io_bitmap;
375 
376 		/*
377 		 * Only copy bitmap data when the sequence number differs. The
378 		 * update time is accounted to the incoming task.
379 		 */
380 		if (tss->io_bitmap.prev_sequence != iobm->sequence)
381 			tss_copy_io_bitmap(tss, iobm);
382 
383 		/* Enable the bitmap */
384 		*base = IO_BITMAP_OFFSET_VALID_MAP;
385 	}
386 
387 	/*
388 	 * Make sure that the TSS limit is covering the IO bitmap. It might have
389 	 * been cut down by a VMEXIT to 0x67 which would cause a subsequent I/O
390 	 * access from user space to trigger a #GP because tbe bitmap is outside
391 	 * the TSS limit.
392 	 */
393 	refresh_tss_limit();
394 }
395 #else /* CONFIG_X86_IOPL_IOPERM */
396 static inline void switch_to_bitmap(unsigned long tifp) { }
397 #endif
398 
399 #ifdef CONFIG_SMP
400 
401 struct ssb_state {
402 	struct ssb_state	*shared_state;
403 	raw_spinlock_t		lock;
404 	unsigned int		disable_state;
405 	unsigned long		local_state;
406 };
407 
408 #define LSTATE_SSB	0
409 
410 static DEFINE_PER_CPU(struct ssb_state, ssb_state);
411 
412 void speculative_store_bypass_ht_init(void)
413 {
414 	struct ssb_state *st = this_cpu_ptr(&ssb_state);
415 	unsigned int this_cpu = smp_processor_id();
416 	unsigned int cpu;
417 
418 	st->local_state = 0;
419 
420 	/*
421 	 * Shared state setup happens once on the first bringup
422 	 * of the CPU. It's not destroyed on CPU hotunplug.
423 	 */
424 	if (st->shared_state)
425 		return;
426 
427 	raw_spin_lock_init(&st->lock);
428 
429 	/*
430 	 * Go over HT siblings and check whether one of them has set up the
431 	 * shared state pointer already.
432 	 */
433 	for_each_cpu(cpu, topology_sibling_cpumask(this_cpu)) {
434 		if (cpu == this_cpu)
435 			continue;
436 
437 		if (!per_cpu(ssb_state, cpu).shared_state)
438 			continue;
439 
440 		/* Link it to the state of the sibling: */
441 		st->shared_state = per_cpu(ssb_state, cpu).shared_state;
442 		return;
443 	}
444 
445 	/*
446 	 * First HT sibling to come up on the core.  Link shared state of
447 	 * the first HT sibling to itself. The siblings on the same core
448 	 * which come up later will see the shared state pointer and link
449 	 * themselves to the state of this CPU.
450 	 */
451 	st->shared_state = st;
452 }
453 
454 /*
455  * Logic is: First HT sibling enables SSBD for both siblings in the core
456  * and last sibling to disable it, disables it for the whole core. This how
457  * MSR_SPEC_CTRL works in "hardware":
458  *
459  *  CORE_SPEC_CTRL = THREAD0_SPEC_CTRL | THREAD1_SPEC_CTRL
460  */
461 static __always_inline void amd_set_core_ssb_state(unsigned long tifn)
462 {
463 	struct ssb_state *st = this_cpu_ptr(&ssb_state);
464 	u64 msr = x86_amd_ls_cfg_base;
465 
466 	if (!static_cpu_has(X86_FEATURE_ZEN)) {
467 		msr |= ssbd_tif_to_amd_ls_cfg(tifn);
468 		wrmsrl(MSR_AMD64_LS_CFG, msr);
469 		return;
470 	}
471 
472 	if (tifn & _TIF_SSBD) {
473 		/*
474 		 * Since this can race with prctl(), block reentry on the
475 		 * same CPU.
476 		 */
477 		if (__test_and_set_bit(LSTATE_SSB, &st->local_state))
478 			return;
479 
480 		msr |= x86_amd_ls_cfg_ssbd_mask;
481 
482 		raw_spin_lock(&st->shared_state->lock);
483 		/* First sibling enables SSBD: */
484 		if (!st->shared_state->disable_state)
485 			wrmsrl(MSR_AMD64_LS_CFG, msr);
486 		st->shared_state->disable_state++;
487 		raw_spin_unlock(&st->shared_state->lock);
488 	} else {
489 		if (!__test_and_clear_bit(LSTATE_SSB, &st->local_state))
490 			return;
491 
492 		raw_spin_lock(&st->shared_state->lock);
493 		st->shared_state->disable_state--;
494 		if (!st->shared_state->disable_state)
495 			wrmsrl(MSR_AMD64_LS_CFG, msr);
496 		raw_spin_unlock(&st->shared_state->lock);
497 	}
498 }
499 #else
500 static __always_inline void amd_set_core_ssb_state(unsigned long tifn)
501 {
502 	u64 msr = x86_amd_ls_cfg_base | ssbd_tif_to_amd_ls_cfg(tifn);
503 
504 	wrmsrl(MSR_AMD64_LS_CFG, msr);
505 }
506 #endif
507 
508 static __always_inline void amd_set_ssb_virt_state(unsigned long tifn)
509 {
510 	/*
511 	 * SSBD has the same definition in SPEC_CTRL and VIRT_SPEC_CTRL,
512 	 * so ssbd_tif_to_spec_ctrl() just works.
513 	 */
514 	wrmsrl(MSR_AMD64_VIRT_SPEC_CTRL, ssbd_tif_to_spec_ctrl(tifn));
515 }
516 
517 /*
518  * Update the MSRs managing speculation control, during context switch.
519  *
520  * tifp: Previous task's thread flags
521  * tifn: Next task's thread flags
522  */
523 static __always_inline void __speculation_ctrl_update(unsigned long tifp,
524 						      unsigned long tifn)
525 {
526 	unsigned long tif_diff = tifp ^ tifn;
527 	u64 msr = x86_spec_ctrl_base;
528 	bool updmsr = false;
529 
530 	lockdep_assert_irqs_disabled();
531 
532 	/* Handle change of TIF_SSBD depending on the mitigation method. */
533 	if (static_cpu_has(X86_FEATURE_VIRT_SSBD)) {
534 		if (tif_diff & _TIF_SSBD)
535 			amd_set_ssb_virt_state(tifn);
536 	} else if (static_cpu_has(X86_FEATURE_LS_CFG_SSBD)) {
537 		if (tif_diff & _TIF_SSBD)
538 			amd_set_core_ssb_state(tifn);
539 	} else if (static_cpu_has(X86_FEATURE_SPEC_CTRL_SSBD) ||
540 		   static_cpu_has(X86_FEATURE_AMD_SSBD)) {
541 		updmsr |= !!(tif_diff & _TIF_SSBD);
542 		msr |= ssbd_tif_to_spec_ctrl(tifn);
543 	}
544 
545 	/* Only evaluate TIF_SPEC_IB if conditional STIBP is enabled. */
546 	if (IS_ENABLED(CONFIG_SMP) &&
547 	    static_branch_unlikely(&switch_to_cond_stibp)) {
548 		updmsr |= !!(tif_diff & _TIF_SPEC_IB);
549 		msr |= stibp_tif_to_spec_ctrl(tifn);
550 	}
551 
552 	if (updmsr)
553 		wrmsrl(MSR_IA32_SPEC_CTRL, msr);
554 }
555 
556 static unsigned long speculation_ctrl_update_tif(struct task_struct *tsk)
557 {
558 	if (test_and_clear_tsk_thread_flag(tsk, TIF_SPEC_FORCE_UPDATE)) {
559 		if (task_spec_ssb_disable(tsk))
560 			set_tsk_thread_flag(tsk, TIF_SSBD);
561 		else
562 			clear_tsk_thread_flag(tsk, TIF_SSBD);
563 
564 		if (task_spec_ib_disable(tsk))
565 			set_tsk_thread_flag(tsk, TIF_SPEC_IB);
566 		else
567 			clear_tsk_thread_flag(tsk, TIF_SPEC_IB);
568 	}
569 	/* Return the updated threadinfo flags*/
570 	return task_thread_info(tsk)->flags;
571 }
572 
573 void speculation_ctrl_update(unsigned long tif)
574 {
575 	unsigned long flags;
576 
577 	/* Forced update. Make sure all relevant TIF flags are different */
578 	local_irq_save(flags);
579 	__speculation_ctrl_update(~tif, tif);
580 	local_irq_restore(flags);
581 }
582 
583 /* Called from seccomp/prctl update */
584 void speculation_ctrl_update_current(void)
585 {
586 	preempt_disable();
587 	speculation_ctrl_update(speculation_ctrl_update_tif(current));
588 	preempt_enable();
589 }
590 
591 static inline void cr4_toggle_bits_irqsoff(unsigned long mask)
592 {
593 	unsigned long newval, cr4 = this_cpu_read(cpu_tlbstate.cr4);
594 
595 	newval = cr4 ^ mask;
596 	if (newval != cr4) {
597 		this_cpu_write(cpu_tlbstate.cr4, newval);
598 		__write_cr4(newval);
599 	}
600 }
601 
602 void __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p)
603 {
604 	unsigned long tifp, tifn;
605 
606 	tifn = READ_ONCE(task_thread_info(next_p)->flags);
607 	tifp = READ_ONCE(task_thread_info(prev_p)->flags);
608 
609 	switch_to_bitmap(tifp);
610 
611 	propagate_user_return_notify(prev_p, next_p);
612 
613 	if ((tifp & _TIF_BLOCKSTEP || tifn & _TIF_BLOCKSTEP) &&
614 	    arch_has_block_step()) {
615 		unsigned long debugctl, msk;
616 
617 		rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctl);
618 		debugctl &= ~DEBUGCTLMSR_BTF;
619 		msk = tifn & _TIF_BLOCKSTEP;
620 		debugctl |= (msk >> TIF_BLOCKSTEP) << DEBUGCTLMSR_BTF_SHIFT;
621 		wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctl);
622 	}
623 
624 	if ((tifp ^ tifn) & _TIF_NOTSC)
625 		cr4_toggle_bits_irqsoff(X86_CR4_TSD);
626 
627 	if ((tifp ^ tifn) & _TIF_NOCPUID)
628 		set_cpuid_faulting(!!(tifn & _TIF_NOCPUID));
629 
630 	if (likely(!((tifp | tifn) & _TIF_SPEC_FORCE_UPDATE))) {
631 		__speculation_ctrl_update(tifp, tifn);
632 	} else {
633 		speculation_ctrl_update_tif(prev_p);
634 		tifn = speculation_ctrl_update_tif(next_p);
635 
636 		/* Enforce MSR update to ensure consistent state */
637 		__speculation_ctrl_update(~tifn, tifn);
638 	}
639 
640 	if ((tifp ^ tifn) & _TIF_SLD)
641 		switch_to_sld(tifn);
642 }
643 
644 /*
645  * Idle related variables and functions
646  */
647 unsigned long boot_option_idle_override = IDLE_NO_OVERRIDE;
648 EXPORT_SYMBOL(boot_option_idle_override);
649 
650 static void (*x86_idle)(void);
651 
652 #ifndef CONFIG_SMP
653 static inline void play_dead(void)
654 {
655 	BUG();
656 }
657 #endif
658 
659 void arch_cpu_idle_enter(void)
660 {
661 	tsc_verify_tsc_adjust(false);
662 	local_touch_nmi();
663 }
664 
665 void arch_cpu_idle_dead(void)
666 {
667 	play_dead();
668 }
669 
670 /*
671  * Called from the generic idle code.
672  */
673 void arch_cpu_idle(void)
674 {
675 	x86_idle();
676 }
677 
678 /*
679  * We use this if we don't have any better idle routine..
680  */
681 void __cpuidle default_idle(void)
682 {
683 	raw_safe_halt();
684 }
685 #if defined(CONFIG_APM_MODULE) || defined(CONFIG_HALTPOLL_CPUIDLE_MODULE)
686 EXPORT_SYMBOL(default_idle);
687 #endif
688 
689 #ifdef CONFIG_XEN
690 bool xen_set_default_idle(void)
691 {
692 	bool ret = !!x86_idle;
693 
694 	x86_idle = default_idle;
695 
696 	return ret;
697 }
698 #endif
699 
700 void stop_this_cpu(void *dummy)
701 {
702 	local_irq_disable();
703 	/*
704 	 * Remove this CPU:
705 	 */
706 	set_cpu_online(smp_processor_id(), false);
707 	disable_local_APIC();
708 	mcheck_cpu_clear(this_cpu_ptr(&cpu_info));
709 
710 	/*
711 	 * Use wbinvd on processors that support SME. This provides support
712 	 * for performing a successful kexec when going from SME inactive
713 	 * to SME active (or vice-versa). The cache must be cleared so that
714 	 * if there are entries with the same physical address, both with and
715 	 * without the encryption bit, they don't race each other when flushed
716 	 * and potentially end up with the wrong entry being committed to
717 	 * memory.
718 	 */
719 	if (boot_cpu_has(X86_FEATURE_SME))
720 		native_wbinvd();
721 	for (;;) {
722 		/*
723 		 * Use native_halt() so that memory contents don't change
724 		 * (stack usage and variables) after possibly issuing the
725 		 * native_wbinvd() above.
726 		 */
727 		native_halt();
728 	}
729 }
730 
731 /*
732  * AMD Erratum 400 aware idle routine. We handle it the same way as C3 power
733  * states (local apic timer and TSC stop).
734  *
735  * XXX this function is completely buggered vs RCU and tracing.
736  */
737 static void amd_e400_idle(void)
738 {
739 	/*
740 	 * We cannot use static_cpu_has_bug() here because X86_BUG_AMD_APIC_C1E
741 	 * gets set after static_cpu_has() places have been converted via
742 	 * alternatives.
743 	 */
744 	if (!boot_cpu_has_bug(X86_BUG_AMD_APIC_C1E)) {
745 		default_idle();
746 		return;
747 	}
748 
749 	tick_broadcast_enter();
750 
751 	default_idle();
752 
753 	/*
754 	 * The switch back from broadcast mode needs to be called with
755 	 * interrupts disabled.
756 	 */
757 	raw_local_irq_disable();
758 	tick_broadcast_exit();
759 	raw_local_irq_enable();
760 }
761 
762 /*
763  * Intel Core2 and older machines prefer MWAIT over HALT for C1.
764  * We can't rely on cpuidle installing MWAIT, because it will not load
765  * on systems that support only C1 -- so the boot default must be MWAIT.
766  *
767  * Some AMD machines are the opposite, they depend on using HALT.
768  *
769  * So for default C1, which is used during boot until cpuidle loads,
770  * use MWAIT-C1 on Intel HW that has it, else use HALT.
771  */
772 static int prefer_mwait_c1_over_halt(const struct cpuinfo_x86 *c)
773 {
774 	if (c->x86_vendor != X86_VENDOR_INTEL)
775 		return 0;
776 
777 	if (!cpu_has(c, X86_FEATURE_MWAIT) || boot_cpu_has_bug(X86_BUG_MONITOR))
778 		return 0;
779 
780 	return 1;
781 }
782 
783 /*
784  * MONITOR/MWAIT with no hints, used for default C1 state. This invokes MWAIT
785  * with interrupts enabled and no flags, which is backwards compatible with the
786  * original MWAIT implementation.
787  */
788 static __cpuidle void mwait_idle(void)
789 {
790 	if (!current_set_polling_and_test()) {
791 		if (this_cpu_has(X86_BUG_CLFLUSH_MONITOR)) {
792 			mb(); /* quirk */
793 			clflush((void *)&current_thread_info()->flags);
794 			mb(); /* quirk */
795 		}
796 
797 		__monitor((void *)&current_thread_info()->flags, 0, 0);
798 		if (!need_resched())
799 			__sti_mwait(0, 0);
800 		else
801 			raw_local_irq_enable();
802 	} else {
803 		raw_local_irq_enable();
804 	}
805 	__current_clr_polling();
806 }
807 
808 void select_idle_routine(const struct cpuinfo_x86 *c)
809 {
810 #ifdef CONFIG_SMP
811 	if (boot_option_idle_override == IDLE_POLL && smp_num_siblings > 1)
812 		pr_warn_once("WARNING: polling idle and HT enabled, performance may degrade\n");
813 #endif
814 	if (x86_idle || boot_option_idle_override == IDLE_POLL)
815 		return;
816 
817 	if (boot_cpu_has_bug(X86_BUG_AMD_E400)) {
818 		pr_info("using AMD E400 aware idle routine\n");
819 		x86_idle = amd_e400_idle;
820 	} else if (prefer_mwait_c1_over_halt(c)) {
821 		pr_info("using mwait in idle threads\n");
822 		x86_idle = mwait_idle;
823 	} else
824 		x86_idle = default_idle;
825 }
826 
827 void amd_e400_c1e_apic_setup(void)
828 {
829 	if (boot_cpu_has_bug(X86_BUG_AMD_APIC_C1E)) {
830 		pr_info("Switch to broadcast mode on CPU%d\n", smp_processor_id());
831 		local_irq_disable();
832 		tick_broadcast_force();
833 		local_irq_enable();
834 	}
835 }
836 
837 void __init arch_post_acpi_subsys_init(void)
838 {
839 	u32 lo, hi;
840 
841 	if (!boot_cpu_has_bug(X86_BUG_AMD_E400))
842 		return;
843 
844 	/*
845 	 * AMD E400 detection needs to happen after ACPI has been enabled. If
846 	 * the machine is affected K8_INTP_C1E_ACTIVE_MASK bits are set in
847 	 * MSR_K8_INT_PENDING_MSG.
848 	 */
849 	rdmsr(MSR_K8_INT_PENDING_MSG, lo, hi);
850 	if (!(lo & K8_INTP_C1E_ACTIVE_MASK))
851 		return;
852 
853 	boot_cpu_set_bug(X86_BUG_AMD_APIC_C1E);
854 
855 	if (!boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
856 		mark_tsc_unstable("TSC halt in AMD C1E");
857 	pr_info("System has AMD C1E enabled\n");
858 }
859 
860 static int __init idle_setup(char *str)
861 {
862 	if (!str)
863 		return -EINVAL;
864 
865 	if (!strcmp(str, "poll")) {
866 		pr_info("using polling idle threads\n");
867 		boot_option_idle_override = IDLE_POLL;
868 		cpu_idle_poll_ctrl(true);
869 	} else if (!strcmp(str, "halt")) {
870 		/*
871 		 * When the boot option of idle=halt is added, halt is
872 		 * forced to be used for CPU idle. In such case CPU C2/C3
873 		 * won't be used again.
874 		 * To continue to load the CPU idle driver, don't touch
875 		 * the boot_option_idle_override.
876 		 */
877 		x86_idle = default_idle;
878 		boot_option_idle_override = IDLE_HALT;
879 	} else if (!strcmp(str, "nomwait")) {
880 		/*
881 		 * If the boot option of "idle=nomwait" is added,
882 		 * it means that mwait will be disabled for CPU C2/C3
883 		 * states. In such case it won't touch the variable
884 		 * of boot_option_idle_override.
885 		 */
886 		boot_option_idle_override = IDLE_NOMWAIT;
887 	} else
888 		return -1;
889 
890 	return 0;
891 }
892 early_param("idle", idle_setup);
893 
894 unsigned long arch_align_stack(unsigned long sp)
895 {
896 	if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
897 		sp -= get_random_int() % 8192;
898 	return sp & ~0xf;
899 }
900 
901 unsigned long arch_randomize_brk(struct mm_struct *mm)
902 {
903 	return randomize_page(mm->brk, 0x02000000);
904 }
905 
906 /*
907  * Called from fs/proc with a reference on @p to find the function
908  * which called into schedule(). This needs to be done carefully
909  * because the task might wake up and we might look at a stack
910  * changing under us.
911  */
912 unsigned long get_wchan(struct task_struct *p)
913 {
914 	unsigned long start, bottom, top, sp, fp, ip, ret = 0;
915 	int count = 0;
916 
917 	if (p == current || p->state == TASK_RUNNING)
918 		return 0;
919 
920 	if (!try_get_task_stack(p))
921 		return 0;
922 
923 	start = (unsigned long)task_stack_page(p);
924 	if (!start)
925 		goto out;
926 
927 	/*
928 	 * Layout of the stack page:
929 	 *
930 	 * ----------- topmax = start + THREAD_SIZE - sizeof(unsigned long)
931 	 * PADDING
932 	 * ----------- top = topmax - TOP_OF_KERNEL_STACK_PADDING
933 	 * stack
934 	 * ----------- bottom = start
935 	 *
936 	 * The tasks stack pointer points at the location where the
937 	 * framepointer is stored. The data on the stack is:
938 	 * ... IP FP ... IP FP
939 	 *
940 	 * We need to read FP and IP, so we need to adjust the upper
941 	 * bound by another unsigned long.
942 	 */
943 	top = start + THREAD_SIZE - TOP_OF_KERNEL_STACK_PADDING;
944 	top -= 2 * sizeof(unsigned long);
945 	bottom = start;
946 
947 	sp = READ_ONCE(p->thread.sp);
948 	if (sp < bottom || sp > top)
949 		goto out;
950 
951 	fp = READ_ONCE_NOCHECK(((struct inactive_task_frame *)sp)->bp);
952 	do {
953 		if (fp < bottom || fp > top)
954 			goto out;
955 		ip = READ_ONCE_NOCHECK(*(unsigned long *)(fp + sizeof(unsigned long)));
956 		if (!in_sched_functions(ip)) {
957 			ret = ip;
958 			goto out;
959 		}
960 		fp = READ_ONCE_NOCHECK(*(unsigned long *)fp);
961 	} while (count++ < 16 && p->state != TASK_RUNNING);
962 
963 out:
964 	put_task_stack(p);
965 	return ret;
966 }
967 
968 long do_arch_prctl_common(struct task_struct *task, int option,
969 			  unsigned long cpuid_enabled)
970 {
971 	switch (option) {
972 	case ARCH_GET_CPUID:
973 		return get_cpuid_mode();
974 	case ARCH_SET_CPUID:
975 		return set_cpuid_mode(task, cpuid_enabled);
976 	}
977 
978 	return -EINVAL;
979 }
980