xref: /openbmc/linux/arch/x86/kernel/process.c (revision 6c33a6f4)
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 <asm/syscalls.h>
32 #include <linux/uaccess.h>
33 #include <asm/mwait.h>
34 #include <asm/fpu/internal.h>
35 #include <asm/debugreg.h>
36 #include <asm/nmi.h>
37 #include <asm/tlbflush.h>
38 #include <asm/mce.h>
39 #include <asm/vm86.h>
40 #include <asm/switch_to.h>
41 #include <asm/desc.h>
42 #include <asm/prctl.h>
43 #include <asm/spec-ctrl.h>
44 #include <asm/io_bitmap.h>
45 #include <asm/proto.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 		/*
67 		 * .sp1 is cpu_current_top_of_stack.  The init task never
68 		 * runs user code, but cpu_current_top_of_stack should still
69 		 * be well defined before the first context switch.
70 		 */
71 		.sp1 = TOP_OF_INIT_STACK,
72 
73 #ifdef CONFIG_X86_32
74 		.ss0 = __KERNEL_DS,
75 		.ss1 = __KERNEL_CS,
76 #endif
77 		.io_bitmap_base	= IO_BITMAP_OFFSET_INVALID,
78 	 },
79 };
80 EXPORT_PER_CPU_SYMBOL(cpu_tss_rw);
81 
82 DEFINE_PER_CPU(bool, __tss_limit_invalid);
83 EXPORT_PER_CPU_SYMBOL_GPL(__tss_limit_invalid);
84 
85 /*
86  * this gets called so that we can store lazy state into memory and copy the
87  * current task into the new thread.
88  */
89 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
90 {
91 	memcpy(dst, src, arch_task_struct_size);
92 #ifdef CONFIG_VM86
93 	dst->thread.vm86 = NULL;
94 #endif
95 
96 	return fpu__copy(dst, src);
97 }
98 
99 /*
100  * Free current thread data structures etc..
101  */
102 void exit_thread(struct task_struct *tsk)
103 {
104 	struct thread_struct *t = &tsk->thread;
105 	struct fpu *fpu = &t->fpu;
106 
107 	if (test_thread_flag(TIF_IO_BITMAP))
108 		io_bitmap_exit();
109 
110 	free_vm86(t);
111 
112 	fpu__drop(fpu);
113 }
114 
115 static int set_new_tls(struct task_struct *p, unsigned long tls)
116 {
117 	struct user_desc __user *utls = (struct user_desc __user *)tls;
118 
119 	if (in_ia32_syscall())
120 		return do_set_thread_area(p, -1, utls, 0);
121 	else
122 		return do_set_thread_area_64(p, ARCH_SET_FS, tls);
123 }
124 
125 int copy_thread_tls(unsigned long clone_flags, unsigned long sp,
126 		    unsigned long arg, struct task_struct *p, unsigned long tls)
127 {
128 	struct inactive_task_frame *frame;
129 	struct fork_frame *fork_frame;
130 	struct pt_regs *childregs;
131 	int ret = 0;
132 
133 	childregs = task_pt_regs(p);
134 	fork_frame = container_of(childregs, struct fork_frame, regs);
135 	frame = &fork_frame->frame;
136 
137 	frame->bp = 0;
138 	frame->ret_addr = (unsigned long) ret_from_fork;
139 	p->thread.sp = (unsigned long) fork_frame;
140 	p->thread.io_bitmap = NULL;
141 	memset(p->thread.ptrace_bps, 0, sizeof(p->thread.ptrace_bps));
142 
143 #ifdef CONFIG_X86_64
144 	savesegment(gs, p->thread.gsindex);
145 	p->thread.gsbase = p->thread.gsindex ? 0 : current->thread.gsbase;
146 	savesegment(fs, p->thread.fsindex);
147 	p->thread.fsbase = p->thread.fsindex ? 0 : current->thread.fsbase;
148 	savesegment(es, p->thread.es);
149 	savesegment(ds, p->thread.ds);
150 #else
151 	p->thread.sp0 = (unsigned long) (childregs + 1);
152 	/*
153 	 * Clear all status flags including IF and set fixed bit. 64bit
154 	 * does not have this initialization as the frame does not contain
155 	 * flags. The flags consistency (especially vs. AC) is there
156 	 * ensured via objtool, which lacks 32bit support.
157 	 */
158 	frame->flags = X86_EFLAGS_FIXED;
159 #endif
160 
161 	/* Kernel thread ? */
162 	if (unlikely(p->flags & PF_KTHREAD)) {
163 		memset(childregs, 0, sizeof(struct pt_regs));
164 		kthread_frame_init(frame, sp, arg);
165 		return 0;
166 	}
167 
168 	frame->bx = 0;
169 	*childregs = *current_pt_regs();
170 	childregs->ax = 0;
171 	if (sp)
172 		childregs->sp = sp;
173 
174 #ifdef CONFIG_X86_32
175 	task_user_gs(p) = get_user_gs(current_pt_regs());
176 #endif
177 
178 	/* Set a new TLS for the child thread? */
179 	if (clone_flags & CLONE_SETTLS)
180 		ret = set_new_tls(p, tls);
181 
182 	if (!ret && unlikely(test_tsk_thread_flag(current, TIF_IO_BITMAP)))
183 		io_bitmap_share(p);
184 
185 	return ret;
186 }
187 
188 void flush_thread(void)
189 {
190 	struct task_struct *tsk = current;
191 
192 	flush_ptrace_hw_breakpoint(tsk);
193 	memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
194 
195 	fpu__clear(&tsk->thread.fpu);
196 }
197 
198 void disable_TSC(void)
199 {
200 	preempt_disable();
201 	if (!test_and_set_thread_flag(TIF_NOTSC))
202 		/*
203 		 * Must flip the CPU state synchronously with
204 		 * TIF_NOTSC in the current running context.
205 		 */
206 		cr4_set_bits(X86_CR4_TSD);
207 	preempt_enable();
208 }
209 
210 static void enable_TSC(void)
211 {
212 	preempt_disable();
213 	if (test_and_clear_thread_flag(TIF_NOTSC))
214 		/*
215 		 * Must flip the CPU state synchronously with
216 		 * TIF_NOTSC in the current running context.
217 		 */
218 		cr4_clear_bits(X86_CR4_TSD);
219 	preempt_enable();
220 }
221 
222 int get_tsc_mode(unsigned long adr)
223 {
224 	unsigned int val;
225 
226 	if (test_thread_flag(TIF_NOTSC))
227 		val = PR_TSC_SIGSEGV;
228 	else
229 		val = PR_TSC_ENABLE;
230 
231 	return put_user(val, (unsigned int __user *)adr);
232 }
233 
234 int set_tsc_mode(unsigned int val)
235 {
236 	if (val == PR_TSC_SIGSEGV)
237 		disable_TSC();
238 	else if (val == PR_TSC_ENABLE)
239 		enable_TSC();
240 	else
241 		return -EINVAL;
242 
243 	return 0;
244 }
245 
246 DEFINE_PER_CPU(u64, msr_misc_features_shadow);
247 
248 static void set_cpuid_faulting(bool on)
249 {
250 	u64 msrval;
251 
252 	msrval = this_cpu_read(msr_misc_features_shadow);
253 	msrval &= ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT;
254 	msrval |= (on << MSR_MISC_FEATURES_ENABLES_CPUID_FAULT_BIT);
255 	this_cpu_write(msr_misc_features_shadow, msrval);
256 	wrmsrl(MSR_MISC_FEATURES_ENABLES, msrval);
257 }
258 
259 static void disable_cpuid(void)
260 {
261 	preempt_disable();
262 	if (!test_and_set_thread_flag(TIF_NOCPUID)) {
263 		/*
264 		 * Must flip the CPU state synchronously with
265 		 * TIF_NOCPUID in the current running context.
266 		 */
267 		set_cpuid_faulting(true);
268 	}
269 	preempt_enable();
270 }
271 
272 static void enable_cpuid(void)
273 {
274 	preempt_disable();
275 	if (test_and_clear_thread_flag(TIF_NOCPUID)) {
276 		/*
277 		 * Must flip the CPU state synchronously with
278 		 * TIF_NOCPUID in the current running context.
279 		 */
280 		set_cpuid_faulting(false);
281 	}
282 	preempt_enable();
283 }
284 
285 static int get_cpuid_mode(void)
286 {
287 	return !test_thread_flag(TIF_NOCPUID);
288 }
289 
290 static int set_cpuid_mode(struct task_struct *task, unsigned long cpuid_enabled)
291 {
292 	if (!boot_cpu_has(X86_FEATURE_CPUID_FAULT))
293 		return -ENODEV;
294 
295 	if (cpuid_enabled)
296 		enable_cpuid();
297 	else
298 		disable_cpuid();
299 
300 	return 0;
301 }
302 
303 /*
304  * Called immediately after a successful exec.
305  */
306 void arch_setup_new_exec(void)
307 {
308 	/* If cpuid was previously disabled for this task, re-enable it. */
309 	if (test_thread_flag(TIF_NOCPUID))
310 		enable_cpuid();
311 
312 	/*
313 	 * Don't inherit TIF_SSBD across exec boundary when
314 	 * PR_SPEC_DISABLE_NOEXEC is used.
315 	 */
316 	if (test_thread_flag(TIF_SSBD) &&
317 	    task_spec_ssb_noexec(current)) {
318 		clear_thread_flag(TIF_SSBD);
319 		task_clear_spec_ssb_disable(current);
320 		task_clear_spec_ssb_noexec(current);
321 		speculation_ctrl_update(task_thread_info(current)->flags);
322 	}
323 }
324 
325 #ifdef CONFIG_X86_IOPL_IOPERM
326 static inline void tss_invalidate_io_bitmap(struct tss_struct *tss)
327 {
328 	/*
329 	 * Invalidate the I/O bitmap by moving io_bitmap_base outside the
330 	 * TSS limit so any subsequent I/O access from user space will
331 	 * trigger a #GP.
332 	 *
333 	 * This is correct even when VMEXIT rewrites the TSS limit
334 	 * to 0x67 as the only requirement is that the base points
335 	 * outside the limit.
336 	 */
337 	tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET_INVALID;
338 }
339 
340 static inline void switch_to_bitmap(unsigned long tifp)
341 {
342 	/*
343 	 * Invalidate I/O bitmap if the previous task used it. This prevents
344 	 * any possible leakage of an active I/O bitmap.
345 	 *
346 	 * If the next task has an I/O bitmap it will handle it on exit to
347 	 * user mode.
348 	 */
349 	if (tifp & _TIF_IO_BITMAP)
350 		tss_invalidate_io_bitmap(this_cpu_ptr(&cpu_tss_rw));
351 }
352 
353 static void tss_copy_io_bitmap(struct tss_struct *tss, struct io_bitmap *iobm)
354 {
355 	/*
356 	 * Copy at least the byte range of the incoming tasks bitmap which
357 	 * covers the permitted I/O ports.
358 	 *
359 	 * If the previous task which used an I/O bitmap had more bits
360 	 * permitted, then the copy needs to cover those as well so they
361 	 * get turned off.
362 	 */
363 	memcpy(tss->io_bitmap.bitmap, iobm->bitmap,
364 	       max(tss->io_bitmap.prev_max, iobm->max));
365 
366 	/*
367 	 * Store the new max and the sequence number of this bitmap
368 	 * and a pointer to the bitmap itself.
369 	 */
370 	tss->io_bitmap.prev_max = iobm->max;
371 	tss->io_bitmap.prev_sequence = iobm->sequence;
372 }
373 
374 /**
375  * tss_update_io_bitmap - Update I/O bitmap before exiting to usermode
376  */
377 void native_tss_update_io_bitmap(void)
378 {
379 	struct tss_struct *tss = this_cpu_ptr(&cpu_tss_rw);
380 	struct thread_struct *t = &current->thread;
381 	u16 *base = &tss->x86_tss.io_bitmap_base;
382 
383 	if (!test_thread_flag(TIF_IO_BITMAP)) {
384 		tss_invalidate_io_bitmap(tss);
385 		return;
386 	}
387 
388 	if (IS_ENABLED(CONFIG_X86_IOPL_IOPERM) && t->iopl_emul == 3) {
389 		*base = IO_BITMAP_OFFSET_VALID_ALL;
390 	} else {
391 		struct io_bitmap *iobm = t->io_bitmap;
392 
393 		/*
394 		 * Only copy bitmap data when the sequence number differs. The
395 		 * update time is accounted to the incoming task.
396 		 */
397 		if (tss->io_bitmap.prev_sequence != iobm->sequence)
398 			tss_copy_io_bitmap(tss, iobm);
399 
400 		/* Enable the bitmap */
401 		*base = IO_BITMAP_OFFSET_VALID_MAP;
402 	}
403 
404 	/*
405 	 * Make sure that the TSS limit is covering the IO bitmap. It might have
406 	 * been cut down by a VMEXIT to 0x67 which would cause a subsequent I/O
407 	 * access from user space to trigger a #GP because tbe bitmap is outside
408 	 * the TSS limit.
409 	 */
410 	refresh_tss_limit();
411 }
412 #else /* CONFIG_X86_IOPL_IOPERM */
413 static inline void switch_to_bitmap(unsigned long tifp) { }
414 #endif
415 
416 #ifdef CONFIG_SMP
417 
418 struct ssb_state {
419 	struct ssb_state	*shared_state;
420 	raw_spinlock_t		lock;
421 	unsigned int		disable_state;
422 	unsigned long		local_state;
423 };
424 
425 #define LSTATE_SSB	0
426 
427 static DEFINE_PER_CPU(struct ssb_state, ssb_state);
428 
429 void speculative_store_bypass_ht_init(void)
430 {
431 	struct ssb_state *st = this_cpu_ptr(&ssb_state);
432 	unsigned int this_cpu = smp_processor_id();
433 	unsigned int cpu;
434 
435 	st->local_state = 0;
436 
437 	/*
438 	 * Shared state setup happens once on the first bringup
439 	 * of the CPU. It's not destroyed on CPU hotunplug.
440 	 */
441 	if (st->shared_state)
442 		return;
443 
444 	raw_spin_lock_init(&st->lock);
445 
446 	/*
447 	 * Go over HT siblings and check whether one of them has set up the
448 	 * shared state pointer already.
449 	 */
450 	for_each_cpu(cpu, topology_sibling_cpumask(this_cpu)) {
451 		if (cpu == this_cpu)
452 			continue;
453 
454 		if (!per_cpu(ssb_state, cpu).shared_state)
455 			continue;
456 
457 		/* Link it to the state of the sibling: */
458 		st->shared_state = per_cpu(ssb_state, cpu).shared_state;
459 		return;
460 	}
461 
462 	/*
463 	 * First HT sibling to come up on the core.  Link shared state of
464 	 * the first HT sibling to itself. The siblings on the same core
465 	 * which come up later will see the shared state pointer and link
466 	 * themself to the state of this CPU.
467 	 */
468 	st->shared_state = st;
469 }
470 
471 /*
472  * Logic is: First HT sibling enables SSBD for both siblings in the core
473  * and last sibling to disable it, disables it for the whole core. This how
474  * MSR_SPEC_CTRL works in "hardware":
475  *
476  *  CORE_SPEC_CTRL = THREAD0_SPEC_CTRL | THREAD1_SPEC_CTRL
477  */
478 static __always_inline void amd_set_core_ssb_state(unsigned long tifn)
479 {
480 	struct ssb_state *st = this_cpu_ptr(&ssb_state);
481 	u64 msr = x86_amd_ls_cfg_base;
482 
483 	if (!static_cpu_has(X86_FEATURE_ZEN)) {
484 		msr |= ssbd_tif_to_amd_ls_cfg(tifn);
485 		wrmsrl(MSR_AMD64_LS_CFG, msr);
486 		return;
487 	}
488 
489 	if (tifn & _TIF_SSBD) {
490 		/*
491 		 * Since this can race with prctl(), block reentry on the
492 		 * same CPU.
493 		 */
494 		if (__test_and_set_bit(LSTATE_SSB, &st->local_state))
495 			return;
496 
497 		msr |= x86_amd_ls_cfg_ssbd_mask;
498 
499 		raw_spin_lock(&st->shared_state->lock);
500 		/* First sibling enables SSBD: */
501 		if (!st->shared_state->disable_state)
502 			wrmsrl(MSR_AMD64_LS_CFG, msr);
503 		st->shared_state->disable_state++;
504 		raw_spin_unlock(&st->shared_state->lock);
505 	} else {
506 		if (!__test_and_clear_bit(LSTATE_SSB, &st->local_state))
507 			return;
508 
509 		raw_spin_lock(&st->shared_state->lock);
510 		st->shared_state->disable_state--;
511 		if (!st->shared_state->disable_state)
512 			wrmsrl(MSR_AMD64_LS_CFG, msr);
513 		raw_spin_unlock(&st->shared_state->lock);
514 	}
515 }
516 #else
517 static __always_inline void amd_set_core_ssb_state(unsigned long tifn)
518 {
519 	u64 msr = x86_amd_ls_cfg_base | ssbd_tif_to_amd_ls_cfg(tifn);
520 
521 	wrmsrl(MSR_AMD64_LS_CFG, msr);
522 }
523 #endif
524 
525 static __always_inline void amd_set_ssb_virt_state(unsigned long tifn)
526 {
527 	/*
528 	 * SSBD has the same definition in SPEC_CTRL and VIRT_SPEC_CTRL,
529 	 * so ssbd_tif_to_spec_ctrl() just works.
530 	 */
531 	wrmsrl(MSR_AMD64_VIRT_SPEC_CTRL, ssbd_tif_to_spec_ctrl(tifn));
532 }
533 
534 /*
535  * Update the MSRs managing speculation control, during context switch.
536  *
537  * tifp: Previous task's thread flags
538  * tifn: Next task's thread flags
539  */
540 static __always_inline void __speculation_ctrl_update(unsigned long tifp,
541 						      unsigned long tifn)
542 {
543 	unsigned long tif_diff = tifp ^ tifn;
544 	u64 msr = x86_spec_ctrl_base;
545 	bool updmsr = false;
546 
547 	lockdep_assert_irqs_disabled();
548 
549 	/*
550 	 * If TIF_SSBD is different, select the proper mitigation
551 	 * method. Note that if SSBD mitigation is disabled or permanentely
552 	 * enabled this branch can't be taken because nothing can set
553 	 * TIF_SSBD.
554 	 */
555 	if (tif_diff & _TIF_SSBD) {
556 		if (static_cpu_has(X86_FEATURE_VIRT_SSBD)) {
557 			amd_set_ssb_virt_state(tifn);
558 		} else if (static_cpu_has(X86_FEATURE_LS_CFG_SSBD)) {
559 			amd_set_core_ssb_state(tifn);
560 		} else if (static_cpu_has(X86_FEATURE_SPEC_CTRL_SSBD) ||
561 			   static_cpu_has(X86_FEATURE_AMD_SSBD)) {
562 			msr |= ssbd_tif_to_spec_ctrl(tifn);
563 			updmsr  = true;
564 		}
565 	}
566 
567 	/*
568 	 * Only evaluate TIF_SPEC_IB if conditional STIBP is enabled,
569 	 * otherwise avoid the MSR write.
570 	 */
571 	if (IS_ENABLED(CONFIG_SMP) &&
572 	    static_branch_unlikely(&switch_to_cond_stibp)) {
573 		updmsr |= !!(tif_diff & _TIF_SPEC_IB);
574 		msr |= stibp_tif_to_spec_ctrl(tifn);
575 	}
576 
577 	if (updmsr)
578 		wrmsrl(MSR_IA32_SPEC_CTRL, msr);
579 }
580 
581 static unsigned long speculation_ctrl_update_tif(struct task_struct *tsk)
582 {
583 	if (test_and_clear_tsk_thread_flag(tsk, TIF_SPEC_FORCE_UPDATE)) {
584 		if (task_spec_ssb_disable(tsk))
585 			set_tsk_thread_flag(tsk, TIF_SSBD);
586 		else
587 			clear_tsk_thread_flag(tsk, TIF_SSBD);
588 
589 		if (task_spec_ib_disable(tsk))
590 			set_tsk_thread_flag(tsk, TIF_SPEC_IB);
591 		else
592 			clear_tsk_thread_flag(tsk, TIF_SPEC_IB);
593 	}
594 	/* Return the updated threadinfo flags*/
595 	return task_thread_info(tsk)->flags;
596 }
597 
598 void speculation_ctrl_update(unsigned long tif)
599 {
600 	unsigned long flags;
601 
602 	/* Forced update. Make sure all relevant TIF flags are different */
603 	local_irq_save(flags);
604 	__speculation_ctrl_update(~tif, tif);
605 	local_irq_restore(flags);
606 }
607 
608 /* Called from seccomp/prctl update */
609 void speculation_ctrl_update_current(void)
610 {
611 	preempt_disable();
612 	speculation_ctrl_update(speculation_ctrl_update_tif(current));
613 	preempt_enable();
614 }
615 
616 void __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p)
617 {
618 	unsigned long tifp, tifn;
619 
620 	tifn = READ_ONCE(task_thread_info(next_p)->flags);
621 	tifp = READ_ONCE(task_thread_info(prev_p)->flags);
622 
623 	switch_to_bitmap(tifp);
624 
625 	propagate_user_return_notify(prev_p, next_p);
626 
627 	if ((tifp & _TIF_BLOCKSTEP || tifn & _TIF_BLOCKSTEP) &&
628 	    arch_has_block_step()) {
629 		unsigned long debugctl, msk;
630 
631 		rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctl);
632 		debugctl &= ~DEBUGCTLMSR_BTF;
633 		msk = tifn & _TIF_BLOCKSTEP;
634 		debugctl |= (msk >> TIF_BLOCKSTEP) << DEBUGCTLMSR_BTF_SHIFT;
635 		wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctl);
636 	}
637 
638 	if ((tifp ^ tifn) & _TIF_NOTSC)
639 		cr4_toggle_bits_irqsoff(X86_CR4_TSD);
640 
641 	if ((tifp ^ tifn) & _TIF_NOCPUID)
642 		set_cpuid_faulting(!!(tifn & _TIF_NOCPUID));
643 
644 	if (likely(!((tifp | tifn) & _TIF_SPEC_FORCE_UPDATE))) {
645 		__speculation_ctrl_update(tifp, tifn);
646 	} else {
647 		speculation_ctrl_update_tif(prev_p);
648 		tifn = speculation_ctrl_update_tif(next_p);
649 
650 		/* Enforce MSR update to ensure consistent state */
651 		__speculation_ctrl_update(~tifn, tifn);
652 	}
653 }
654 
655 /*
656  * Idle related variables and functions
657  */
658 unsigned long boot_option_idle_override = IDLE_NO_OVERRIDE;
659 EXPORT_SYMBOL(boot_option_idle_override);
660 
661 static void (*x86_idle)(void);
662 
663 #ifndef CONFIG_SMP
664 static inline void play_dead(void)
665 {
666 	BUG();
667 }
668 #endif
669 
670 void arch_cpu_idle_enter(void)
671 {
672 	tsc_verify_tsc_adjust(false);
673 	local_touch_nmi();
674 }
675 
676 void arch_cpu_idle_dead(void)
677 {
678 	play_dead();
679 }
680 
681 /*
682  * Called from the generic idle code.
683  */
684 void arch_cpu_idle(void)
685 {
686 	x86_idle();
687 }
688 
689 /*
690  * We use this if we don't have any better idle routine..
691  */
692 void __cpuidle default_idle(void)
693 {
694 	trace_cpu_idle_rcuidle(1, smp_processor_id());
695 	safe_halt();
696 	trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, smp_processor_id());
697 }
698 #if defined(CONFIG_APM_MODULE) || defined(CONFIG_HALTPOLL_CPUIDLE_MODULE)
699 EXPORT_SYMBOL(default_idle);
700 #endif
701 
702 #ifdef CONFIG_XEN
703 bool xen_set_default_idle(void)
704 {
705 	bool ret = !!x86_idle;
706 
707 	x86_idle = default_idle;
708 
709 	return ret;
710 }
711 #endif
712 
713 void stop_this_cpu(void *dummy)
714 {
715 	local_irq_disable();
716 	/*
717 	 * Remove this CPU:
718 	 */
719 	set_cpu_online(smp_processor_id(), false);
720 	disable_local_APIC();
721 	mcheck_cpu_clear(this_cpu_ptr(&cpu_info));
722 
723 	/*
724 	 * Use wbinvd on processors that support SME. This provides support
725 	 * for performing a successful kexec when going from SME inactive
726 	 * to SME active (or vice-versa). The cache must be cleared so that
727 	 * if there are entries with the same physical address, both with and
728 	 * without the encryption bit, they don't race each other when flushed
729 	 * and potentially end up with the wrong entry being committed to
730 	 * memory.
731 	 */
732 	if (boot_cpu_has(X86_FEATURE_SME))
733 		native_wbinvd();
734 	for (;;) {
735 		/*
736 		 * Use native_halt() so that memory contents don't change
737 		 * (stack usage and variables) after possibly issuing the
738 		 * native_wbinvd() above.
739 		 */
740 		native_halt();
741 	}
742 }
743 
744 /*
745  * AMD Erratum 400 aware idle routine. We handle it the same way as C3 power
746  * states (local apic timer and TSC stop).
747  */
748 static void amd_e400_idle(void)
749 {
750 	/*
751 	 * We cannot use static_cpu_has_bug() here because X86_BUG_AMD_APIC_C1E
752 	 * gets set after static_cpu_has() places have been converted via
753 	 * alternatives.
754 	 */
755 	if (!boot_cpu_has_bug(X86_BUG_AMD_APIC_C1E)) {
756 		default_idle();
757 		return;
758 	}
759 
760 	tick_broadcast_enter();
761 
762 	default_idle();
763 
764 	/*
765 	 * The switch back from broadcast mode needs to be called with
766 	 * interrupts disabled.
767 	 */
768 	local_irq_disable();
769 	tick_broadcast_exit();
770 	local_irq_enable();
771 }
772 
773 /*
774  * Intel Core2 and older machines prefer MWAIT over HALT for C1.
775  * We can't rely on cpuidle installing MWAIT, because it will not load
776  * on systems that support only C1 -- so the boot default must be MWAIT.
777  *
778  * Some AMD machines are the opposite, they depend on using HALT.
779  *
780  * So for default C1, which is used during boot until cpuidle loads,
781  * use MWAIT-C1 on Intel HW that has it, else use HALT.
782  */
783 static int prefer_mwait_c1_over_halt(const struct cpuinfo_x86 *c)
784 {
785 	if (c->x86_vendor != X86_VENDOR_INTEL)
786 		return 0;
787 
788 	if (!cpu_has(c, X86_FEATURE_MWAIT) || boot_cpu_has_bug(X86_BUG_MONITOR))
789 		return 0;
790 
791 	return 1;
792 }
793 
794 /*
795  * MONITOR/MWAIT with no hints, used for default C1 state. This invokes MWAIT
796  * with interrupts enabled and no flags, which is backwards compatible with the
797  * original MWAIT implementation.
798  */
799 static __cpuidle void mwait_idle(void)
800 {
801 	if (!current_set_polling_and_test()) {
802 		trace_cpu_idle_rcuidle(1, smp_processor_id());
803 		if (this_cpu_has(X86_BUG_CLFLUSH_MONITOR)) {
804 			mb(); /* quirk */
805 			clflush((void *)&current_thread_info()->flags);
806 			mb(); /* quirk */
807 		}
808 
809 		__monitor((void *)&current_thread_info()->flags, 0, 0);
810 		if (!need_resched())
811 			__sti_mwait(0, 0);
812 		else
813 			local_irq_enable();
814 		trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, smp_processor_id());
815 	} else {
816 		local_irq_enable();
817 	}
818 	__current_clr_polling();
819 }
820 
821 void select_idle_routine(const struct cpuinfo_x86 *c)
822 {
823 #ifdef CONFIG_SMP
824 	if (boot_option_idle_override == IDLE_POLL && smp_num_siblings > 1)
825 		pr_warn_once("WARNING: polling idle and HT enabled, performance may degrade\n");
826 #endif
827 	if (x86_idle || boot_option_idle_override == IDLE_POLL)
828 		return;
829 
830 	if (boot_cpu_has_bug(X86_BUG_AMD_E400)) {
831 		pr_info("using AMD E400 aware idle routine\n");
832 		x86_idle = amd_e400_idle;
833 	} else if (prefer_mwait_c1_over_halt(c)) {
834 		pr_info("using mwait in idle threads\n");
835 		x86_idle = mwait_idle;
836 	} else
837 		x86_idle = default_idle;
838 }
839 
840 void amd_e400_c1e_apic_setup(void)
841 {
842 	if (boot_cpu_has_bug(X86_BUG_AMD_APIC_C1E)) {
843 		pr_info("Switch to broadcast mode on CPU%d\n", smp_processor_id());
844 		local_irq_disable();
845 		tick_broadcast_force();
846 		local_irq_enable();
847 	}
848 }
849 
850 void __init arch_post_acpi_subsys_init(void)
851 {
852 	u32 lo, hi;
853 
854 	if (!boot_cpu_has_bug(X86_BUG_AMD_E400))
855 		return;
856 
857 	/*
858 	 * AMD E400 detection needs to happen after ACPI has been enabled. If
859 	 * the machine is affected K8_INTP_C1E_ACTIVE_MASK bits are set in
860 	 * MSR_K8_INT_PENDING_MSG.
861 	 */
862 	rdmsr(MSR_K8_INT_PENDING_MSG, lo, hi);
863 	if (!(lo & K8_INTP_C1E_ACTIVE_MASK))
864 		return;
865 
866 	boot_cpu_set_bug(X86_BUG_AMD_APIC_C1E);
867 
868 	if (!boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
869 		mark_tsc_unstable("TSC halt in AMD C1E");
870 	pr_info("System has AMD C1E enabled\n");
871 }
872 
873 static int __init idle_setup(char *str)
874 {
875 	if (!str)
876 		return -EINVAL;
877 
878 	if (!strcmp(str, "poll")) {
879 		pr_info("using polling idle threads\n");
880 		boot_option_idle_override = IDLE_POLL;
881 		cpu_idle_poll_ctrl(true);
882 	} else if (!strcmp(str, "halt")) {
883 		/*
884 		 * When the boot option of idle=halt is added, halt is
885 		 * forced to be used for CPU idle. In such case CPU C2/C3
886 		 * won't be used again.
887 		 * To continue to load the CPU idle driver, don't touch
888 		 * the boot_option_idle_override.
889 		 */
890 		x86_idle = default_idle;
891 		boot_option_idle_override = IDLE_HALT;
892 	} else if (!strcmp(str, "nomwait")) {
893 		/*
894 		 * If the boot option of "idle=nomwait" is added,
895 		 * it means that mwait will be disabled for CPU C2/C3
896 		 * states. In such case it won't touch the variable
897 		 * of boot_option_idle_override.
898 		 */
899 		boot_option_idle_override = IDLE_NOMWAIT;
900 	} else
901 		return -1;
902 
903 	return 0;
904 }
905 early_param("idle", idle_setup);
906 
907 unsigned long arch_align_stack(unsigned long sp)
908 {
909 	if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
910 		sp -= get_random_int() % 8192;
911 	return sp & ~0xf;
912 }
913 
914 unsigned long arch_randomize_brk(struct mm_struct *mm)
915 {
916 	return randomize_page(mm->brk, 0x02000000);
917 }
918 
919 /*
920  * Called from fs/proc with a reference on @p to find the function
921  * which called into schedule(). This needs to be done carefully
922  * because the task might wake up and we might look at a stack
923  * changing under us.
924  */
925 unsigned long get_wchan(struct task_struct *p)
926 {
927 	unsigned long start, bottom, top, sp, fp, ip, ret = 0;
928 	int count = 0;
929 
930 	if (p == current || p->state == TASK_RUNNING)
931 		return 0;
932 
933 	if (!try_get_task_stack(p))
934 		return 0;
935 
936 	start = (unsigned long)task_stack_page(p);
937 	if (!start)
938 		goto out;
939 
940 	/*
941 	 * Layout of the stack page:
942 	 *
943 	 * ----------- topmax = start + THREAD_SIZE - sizeof(unsigned long)
944 	 * PADDING
945 	 * ----------- top = topmax - TOP_OF_KERNEL_STACK_PADDING
946 	 * stack
947 	 * ----------- bottom = start
948 	 *
949 	 * The tasks stack pointer points at the location where the
950 	 * framepointer is stored. The data on the stack is:
951 	 * ... IP FP ... IP FP
952 	 *
953 	 * We need to read FP and IP, so we need to adjust the upper
954 	 * bound by another unsigned long.
955 	 */
956 	top = start + THREAD_SIZE - TOP_OF_KERNEL_STACK_PADDING;
957 	top -= 2 * sizeof(unsigned long);
958 	bottom = start;
959 
960 	sp = READ_ONCE(p->thread.sp);
961 	if (sp < bottom || sp > top)
962 		goto out;
963 
964 	fp = READ_ONCE_NOCHECK(((struct inactive_task_frame *)sp)->bp);
965 	do {
966 		if (fp < bottom || fp > top)
967 			goto out;
968 		ip = READ_ONCE_NOCHECK(*(unsigned long *)(fp + sizeof(unsigned long)));
969 		if (!in_sched_functions(ip)) {
970 			ret = ip;
971 			goto out;
972 		}
973 		fp = READ_ONCE_NOCHECK(*(unsigned long *)fp);
974 	} while (count++ < 16 && p->state != TASK_RUNNING);
975 
976 out:
977 	put_task_stack(p);
978 	return ret;
979 }
980 
981 long do_arch_prctl_common(struct task_struct *task, int option,
982 			  unsigned long cpuid_enabled)
983 {
984 	switch (option) {
985 	case ARCH_GET_CPUID:
986 		return get_cpuid_mode();
987 	case ARCH_SET_CPUID:
988 		return set_cpuid_mode(task, cpuid_enabled);
989 	}
990 
991 	return -EINVAL;
992 }
993