xref: /openbmc/linux/arch/x86/kernel/process.c (revision 29c37341)
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 
46 #include "process.h"
47 
48 /*
49  * per-CPU TSS segments. Threads are completely 'soft' on Linux,
50  * no more per-task TSS's. The TSS size is kept cacheline-aligned
51  * so they are allowed to end up in the .data..cacheline_aligned
52  * section. Since TSS's are completely CPU-local, we want them
53  * on exact cacheline boundaries, to eliminate cacheline ping-pong.
54  */
55 __visible DEFINE_PER_CPU_PAGE_ALIGNED(struct tss_struct, cpu_tss_rw) = {
56 	.x86_tss = {
57 		/*
58 		 * .sp0 is only used when entering ring 0 from a lower
59 		 * privilege level.  Since the init task never runs anything
60 		 * but ring 0 code, there is no need for a valid value here.
61 		 * Poison it.
62 		 */
63 		.sp0 = (1UL << (BITS_PER_LONG-1)) + 1,
64 
65 		/*
66 		 * .sp1 is cpu_current_top_of_stack.  The init task never
67 		 * runs user code, but cpu_current_top_of_stack should still
68 		 * be well defined before the first context switch.
69 		 */
70 		.sp1 = TOP_OF_INIT_STACK,
71 
72 #ifdef CONFIG_X86_32
73 		.ss0 = __KERNEL_DS,
74 		.ss1 = __KERNEL_CS,
75 #endif
76 		.io_bitmap_base	= IO_BITMAP_OFFSET_INVALID,
77 	 },
78 };
79 EXPORT_PER_CPU_SYMBOL(cpu_tss_rw);
80 
81 DEFINE_PER_CPU(bool, __tss_limit_invalid);
82 EXPORT_PER_CPU_SYMBOL_GPL(__tss_limit_invalid);
83 
84 /*
85  * this gets called so that we can store lazy state into memory and copy the
86  * current task into the new thread.
87  */
88 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
89 {
90 	memcpy(dst, src, arch_task_struct_size);
91 #ifdef CONFIG_VM86
92 	dst->thread.vm86 = NULL;
93 #endif
94 
95 	return fpu__copy(dst, src);
96 }
97 
98 /*
99  * Free thread data structures etc..
100  */
101 void exit_thread(struct task_struct *tsk)
102 {
103 	struct thread_struct *t = &tsk->thread;
104 	struct fpu *fpu = &t->fpu;
105 
106 	if (test_thread_flag(TIF_IO_BITMAP))
107 		io_bitmap_exit(tsk);
108 
109 	free_vm86(t);
110 
111 	fpu__drop(fpu);
112 }
113 
114 static int set_new_tls(struct task_struct *p, unsigned long tls)
115 {
116 	struct user_desc __user *utls = (struct user_desc __user *)tls;
117 
118 	if (in_ia32_syscall())
119 		return do_set_thread_area(p, -1, utls, 0);
120 	else
121 		return do_set_thread_area_64(p, ARCH_SET_FS, tls);
122 }
123 
124 int copy_thread(unsigned long clone_flags, unsigned long sp, unsigned long arg,
125 		struct task_struct *p, unsigned long tls)
126 {
127 	struct inactive_task_frame *frame;
128 	struct fork_frame *fork_frame;
129 	struct pt_regs *childregs;
130 	int ret = 0;
131 
132 	childregs = task_pt_regs(p);
133 	fork_frame = container_of(childregs, struct fork_frame, regs);
134 	frame = &fork_frame->frame;
135 
136 	frame->bp = 0;
137 	frame->ret_addr = (unsigned long) ret_from_fork;
138 	p->thread.sp = (unsigned long) fork_frame;
139 	p->thread.io_bitmap = NULL;
140 	memset(p->thread.ptrace_bps, 0, sizeof(p->thread.ptrace_bps));
141 
142 #ifdef CONFIG_X86_64
143 	current_save_fsgs();
144 	p->thread.fsindex = current->thread.fsindex;
145 	p->thread.fsbase = current->thread.fsbase;
146 	p->thread.gsindex = current->thread.gsindex;
147 	p->thread.gsbase = current->thread.gsbase;
148 
149 	savesegment(es, p->thread.es);
150 	savesegment(ds, p->thread.ds);
151 #else
152 	p->thread.sp0 = (unsigned long) (childregs + 1);
153 	/*
154 	 * Clear all status flags including IF and set fixed bit. 64bit
155 	 * does not have this initialization as the frame does not contain
156 	 * flags. The flags consistency (especially vs. AC) is there
157 	 * ensured via objtool, which lacks 32bit support.
158 	 */
159 	frame->flags = X86_EFLAGS_FIXED;
160 #endif
161 
162 	/* Kernel thread ? */
163 	if (unlikely(p->flags & PF_KTHREAD)) {
164 		memset(childregs, 0, sizeof(struct pt_regs));
165 		kthread_frame_init(frame, sp, arg);
166 		return 0;
167 	}
168 
169 	frame->bx = 0;
170 	*childregs = *current_pt_regs();
171 	childregs->ax = 0;
172 	if (sp)
173 		childregs->sp = sp;
174 
175 #ifdef CONFIG_X86_32
176 	task_user_gs(p) = get_user_gs(current_pt_regs());
177 #endif
178 
179 	/* Set a new TLS for the child thread? */
180 	if (clone_flags & CLONE_SETTLS)
181 		ret = set_new_tls(p, tls);
182 
183 	if (!ret && unlikely(test_tsk_thread_flag(current, TIF_IO_BITMAP)))
184 		io_bitmap_share(p);
185 
186 	return ret;
187 }
188 
189 void flush_thread(void)
190 {
191 	struct task_struct *tsk = current;
192 
193 	flush_ptrace_hw_breakpoint(tsk);
194 	memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
195 
196 	fpu__clear_all(&tsk->thread.fpu);
197 }
198 
199 void disable_TSC(void)
200 {
201 	preempt_disable();
202 	if (!test_and_set_thread_flag(TIF_NOTSC))
203 		/*
204 		 * Must flip the CPU state synchronously with
205 		 * TIF_NOTSC in the current running context.
206 		 */
207 		cr4_set_bits(X86_CR4_TSD);
208 	preempt_enable();
209 }
210 
211 static void enable_TSC(void)
212 {
213 	preempt_disable();
214 	if (test_and_clear_thread_flag(TIF_NOTSC))
215 		/*
216 		 * Must flip the CPU state synchronously with
217 		 * TIF_NOTSC in the current running context.
218 		 */
219 		cr4_clear_bits(X86_CR4_TSD);
220 	preempt_enable();
221 }
222 
223 int get_tsc_mode(unsigned long adr)
224 {
225 	unsigned int val;
226 
227 	if (test_thread_flag(TIF_NOTSC))
228 		val = PR_TSC_SIGSEGV;
229 	else
230 		val = PR_TSC_ENABLE;
231 
232 	return put_user(val, (unsigned int __user *)adr);
233 }
234 
235 int set_tsc_mode(unsigned int val)
236 {
237 	if (val == PR_TSC_SIGSEGV)
238 		disable_TSC();
239 	else if (val == PR_TSC_ENABLE)
240 		enable_TSC();
241 	else
242 		return -EINVAL;
243 
244 	return 0;
245 }
246 
247 DEFINE_PER_CPU(u64, msr_misc_features_shadow);
248 
249 static void set_cpuid_faulting(bool on)
250 {
251 	u64 msrval;
252 
253 	msrval = this_cpu_read(msr_misc_features_shadow);
254 	msrval &= ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT;
255 	msrval |= (on << MSR_MISC_FEATURES_ENABLES_CPUID_FAULT_BIT);
256 	this_cpu_write(msr_misc_features_shadow, msrval);
257 	wrmsrl(MSR_MISC_FEATURES_ENABLES, msrval);
258 }
259 
260 static void disable_cpuid(void)
261 {
262 	preempt_disable();
263 	if (!test_and_set_thread_flag(TIF_NOCPUID)) {
264 		/*
265 		 * Must flip the CPU state synchronously with
266 		 * TIF_NOCPUID in the current running context.
267 		 */
268 		set_cpuid_faulting(true);
269 	}
270 	preempt_enable();
271 }
272 
273 static void enable_cpuid(void)
274 {
275 	preempt_disable();
276 	if (test_and_clear_thread_flag(TIF_NOCPUID)) {
277 		/*
278 		 * Must flip the CPU state synchronously with
279 		 * TIF_NOCPUID in the current running context.
280 		 */
281 		set_cpuid_faulting(false);
282 	}
283 	preempt_enable();
284 }
285 
286 static int get_cpuid_mode(void)
287 {
288 	return !test_thread_flag(TIF_NOCPUID);
289 }
290 
291 static int set_cpuid_mode(struct task_struct *task, unsigned long cpuid_enabled)
292 {
293 	if (!boot_cpu_has(X86_FEATURE_CPUID_FAULT))
294 		return -ENODEV;
295 
296 	if (cpuid_enabled)
297 		enable_cpuid();
298 	else
299 		disable_cpuid();
300 
301 	return 0;
302 }
303 
304 /*
305  * Called immediately after a successful exec.
306  */
307 void arch_setup_new_exec(void)
308 {
309 	/* If cpuid was previously disabled for this task, re-enable it. */
310 	if (test_thread_flag(TIF_NOCPUID))
311 		enable_cpuid();
312 
313 	/*
314 	 * Don't inherit TIF_SSBD across exec boundary when
315 	 * PR_SPEC_DISABLE_NOEXEC is used.
316 	 */
317 	if (test_thread_flag(TIF_SSBD) &&
318 	    task_spec_ssb_noexec(current)) {
319 		clear_thread_flag(TIF_SSBD);
320 		task_clear_spec_ssb_disable(current);
321 		task_clear_spec_ssb_noexec(current);
322 		speculation_ctrl_update(task_thread_info(current)->flags);
323 	}
324 }
325 
326 #ifdef CONFIG_X86_IOPL_IOPERM
327 static inline void switch_to_bitmap(unsigned long tifp)
328 {
329 	/*
330 	 * Invalidate I/O bitmap if the previous task used it. This prevents
331 	 * any possible leakage of an active I/O bitmap.
332 	 *
333 	 * If the next task has an I/O bitmap it will handle it on exit to
334 	 * user mode.
335 	 */
336 	if (tifp & _TIF_IO_BITMAP)
337 		tss_invalidate_io_bitmap();
338 }
339 
340 static void tss_copy_io_bitmap(struct tss_struct *tss, struct io_bitmap *iobm)
341 {
342 	/*
343 	 * Copy at least the byte range of the incoming tasks bitmap which
344 	 * covers the permitted I/O ports.
345 	 *
346 	 * If the previous task which used an I/O bitmap had more bits
347 	 * permitted, then the copy needs to cover those as well so they
348 	 * get turned off.
349 	 */
350 	memcpy(tss->io_bitmap.bitmap, iobm->bitmap,
351 	       max(tss->io_bitmap.prev_max, iobm->max));
352 
353 	/*
354 	 * Store the new max and the sequence number of this bitmap
355 	 * and a pointer to the bitmap itself.
356 	 */
357 	tss->io_bitmap.prev_max = iobm->max;
358 	tss->io_bitmap.prev_sequence = iobm->sequence;
359 }
360 
361 /**
362  * tss_update_io_bitmap - Update I/O bitmap before exiting to usermode
363  */
364 void native_tss_update_io_bitmap(void)
365 {
366 	struct tss_struct *tss = this_cpu_ptr(&cpu_tss_rw);
367 	struct thread_struct *t = &current->thread;
368 	u16 *base = &tss->x86_tss.io_bitmap_base;
369 
370 	if (!test_thread_flag(TIF_IO_BITMAP)) {
371 		native_tss_invalidate_io_bitmap();
372 		return;
373 	}
374 
375 	if (IS_ENABLED(CONFIG_X86_IOPL_IOPERM) && t->iopl_emul == 3) {
376 		*base = IO_BITMAP_OFFSET_VALID_ALL;
377 	} else {
378 		struct io_bitmap *iobm = t->io_bitmap;
379 
380 		/*
381 		 * Only copy bitmap data when the sequence number differs. The
382 		 * update time is accounted to the incoming task.
383 		 */
384 		if (tss->io_bitmap.prev_sequence != iobm->sequence)
385 			tss_copy_io_bitmap(tss, iobm);
386 
387 		/* Enable the bitmap */
388 		*base = IO_BITMAP_OFFSET_VALID_MAP;
389 	}
390 
391 	/*
392 	 * Make sure that the TSS limit is covering the IO bitmap. It might have
393 	 * been cut down by a VMEXIT to 0x67 which would cause a subsequent I/O
394 	 * access from user space to trigger a #GP because tbe bitmap is outside
395 	 * the TSS limit.
396 	 */
397 	refresh_tss_limit();
398 }
399 #else /* CONFIG_X86_IOPL_IOPERM */
400 static inline void switch_to_bitmap(unsigned long tifp) { }
401 #endif
402 
403 #ifdef CONFIG_SMP
404 
405 struct ssb_state {
406 	struct ssb_state	*shared_state;
407 	raw_spinlock_t		lock;
408 	unsigned int		disable_state;
409 	unsigned long		local_state;
410 };
411 
412 #define LSTATE_SSB	0
413 
414 static DEFINE_PER_CPU(struct ssb_state, ssb_state);
415 
416 void speculative_store_bypass_ht_init(void)
417 {
418 	struct ssb_state *st = this_cpu_ptr(&ssb_state);
419 	unsigned int this_cpu = smp_processor_id();
420 	unsigned int cpu;
421 
422 	st->local_state = 0;
423 
424 	/*
425 	 * Shared state setup happens once on the first bringup
426 	 * of the CPU. It's not destroyed on CPU hotunplug.
427 	 */
428 	if (st->shared_state)
429 		return;
430 
431 	raw_spin_lock_init(&st->lock);
432 
433 	/*
434 	 * Go over HT siblings and check whether one of them has set up the
435 	 * shared state pointer already.
436 	 */
437 	for_each_cpu(cpu, topology_sibling_cpumask(this_cpu)) {
438 		if (cpu == this_cpu)
439 			continue;
440 
441 		if (!per_cpu(ssb_state, cpu).shared_state)
442 			continue;
443 
444 		/* Link it to the state of the sibling: */
445 		st->shared_state = per_cpu(ssb_state, cpu).shared_state;
446 		return;
447 	}
448 
449 	/*
450 	 * First HT sibling to come up on the core.  Link shared state of
451 	 * the first HT sibling to itself. The siblings on the same core
452 	 * which come up later will see the shared state pointer and link
453 	 * themself to the state of this CPU.
454 	 */
455 	st->shared_state = st;
456 }
457 
458 /*
459  * Logic is: First HT sibling enables SSBD for both siblings in the core
460  * and last sibling to disable it, disables it for the whole core. This how
461  * MSR_SPEC_CTRL works in "hardware":
462  *
463  *  CORE_SPEC_CTRL = THREAD0_SPEC_CTRL | THREAD1_SPEC_CTRL
464  */
465 static __always_inline void amd_set_core_ssb_state(unsigned long tifn)
466 {
467 	struct ssb_state *st = this_cpu_ptr(&ssb_state);
468 	u64 msr = x86_amd_ls_cfg_base;
469 
470 	if (!static_cpu_has(X86_FEATURE_ZEN)) {
471 		msr |= ssbd_tif_to_amd_ls_cfg(tifn);
472 		wrmsrl(MSR_AMD64_LS_CFG, msr);
473 		return;
474 	}
475 
476 	if (tifn & _TIF_SSBD) {
477 		/*
478 		 * Since this can race with prctl(), block reentry on the
479 		 * same CPU.
480 		 */
481 		if (__test_and_set_bit(LSTATE_SSB, &st->local_state))
482 			return;
483 
484 		msr |= x86_amd_ls_cfg_ssbd_mask;
485 
486 		raw_spin_lock(&st->shared_state->lock);
487 		/* First sibling enables SSBD: */
488 		if (!st->shared_state->disable_state)
489 			wrmsrl(MSR_AMD64_LS_CFG, msr);
490 		st->shared_state->disable_state++;
491 		raw_spin_unlock(&st->shared_state->lock);
492 	} else {
493 		if (!__test_and_clear_bit(LSTATE_SSB, &st->local_state))
494 			return;
495 
496 		raw_spin_lock(&st->shared_state->lock);
497 		st->shared_state->disable_state--;
498 		if (!st->shared_state->disable_state)
499 			wrmsrl(MSR_AMD64_LS_CFG, msr);
500 		raw_spin_unlock(&st->shared_state->lock);
501 	}
502 }
503 #else
504 static __always_inline void amd_set_core_ssb_state(unsigned long tifn)
505 {
506 	u64 msr = x86_amd_ls_cfg_base | ssbd_tif_to_amd_ls_cfg(tifn);
507 
508 	wrmsrl(MSR_AMD64_LS_CFG, msr);
509 }
510 #endif
511 
512 static __always_inline void amd_set_ssb_virt_state(unsigned long tifn)
513 {
514 	/*
515 	 * SSBD has the same definition in SPEC_CTRL and VIRT_SPEC_CTRL,
516 	 * so ssbd_tif_to_spec_ctrl() just works.
517 	 */
518 	wrmsrl(MSR_AMD64_VIRT_SPEC_CTRL, ssbd_tif_to_spec_ctrl(tifn));
519 }
520 
521 /*
522  * Update the MSRs managing speculation control, during context switch.
523  *
524  * tifp: Previous task's thread flags
525  * tifn: Next task's thread flags
526  */
527 static __always_inline void __speculation_ctrl_update(unsigned long tifp,
528 						      unsigned long tifn)
529 {
530 	unsigned long tif_diff = tifp ^ tifn;
531 	u64 msr = x86_spec_ctrl_base;
532 	bool updmsr = false;
533 
534 	lockdep_assert_irqs_disabled();
535 
536 	/* Handle change of TIF_SSBD depending on the mitigation method. */
537 	if (static_cpu_has(X86_FEATURE_VIRT_SSBD)) {
538 		if (tif_diff & _TIF_SSBD)
539 			amd_set_ssb_virt_state(tifn);
540 	} else if (static_cpu_has(X86_FEATURE_LS_CFG_SSBD)) {
541 		if (tif_diff & _TIF_SSBD)
542 			amd_set_core_ssb_state(tifn);
543 	} else if (static_cpu_has(X86_FEATURE_SPEC_CTRL_SSBD) ||
544 		   static_cpu_has(X86_FEATURE_AMD_SSBD)) {
545 		updmsr |= !!(tif_diff & _TIF_SSBD);
546 		msr |= ssbd_tif_to_spec_ctrl(tifn);
547 	}
548 
549 	/* Only evaluate TIF_SPEC_IB if conditional STIBP is enabled. */
550 	if (IS_ENABLED(CONFIG_SMP) &&
551 	    static_branch_unlikely(&switch_to_cond_stibp)) {
552 		updmsr |= !!(tif_diff & _TIF_SPEC_IB);
553 		msr |= stibp_tif_to_spec_ctrl(tifn);
554 	}
555 
556 	if (updmsr)
557 		wrmsrl(MSR_IA32_SPEC_CTRL, msr);
558 }
559 
560 static unsigned long speculation_ctrl_update_tif(struct task_struct *tsk)
561 {
562 	if (test_and_clear_tsk_thread_flag(tsk, TIF_SPEC_FORCE_UPDATE)) {
563 		if (task_spec_ssb_disable(tsk))
564 			set_tsk_thread_flag(tsk, TIF_SSBD);
565 		else
566 			clear_tsk_thread_flag(tsk, TIF_SSBD);
567 
568 		if (task_spec_ib_disable(tsk))
569 			set_tsk_thread_flag(tsk, TIF_SPEC_IB);
570 		else
571 			clear_tsk_thread_flag(tsk, TIF_SPEC_IB);
572 	}
573 	/* Return the updated threadinfo flags*/
574 	return task_thread_info(tsk)->flags;
575 }
576 
577 void speculation_ctrl_update(unsigned long tif)
578 {
579 	unsigned long flags;
580 
581 	/* Forced update. Make sure all relevant TIF flags are different */
582 	local_irq_save(flags);
583 	__speculation_ctrl_update(~tif, tif);
584 	local_irq_restore(flags);
585 }
586 
587 /* Called from seccomp/prctl update */
588 void speculation_ctrl_update_current(void)
589 {
590 	preempt_disable();
591 	speculation_ctrl_update(speculation_ctrl_update_tif(current));
592 	preempt_enable();
593 }
594 
595 static inline void cr4_toggle_bits_irqsoff(unsigned long mask)
596 {
597 	unsigned long newval, cr4 = this_cpu_read(cpu_tlbstate.cr4);
598 
599 	newval = cr4 ^ mask;
600 	if (newval != cr4) {
601 		this_cpu_write(cpu_tlbstate.cr4, newval);
602 		__write_cr4(newval);
603 	}
604 }
605 
606 void __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p)
607 {
608 	unsigned long tifp, tifn;
609 
610 	tifn = READ_ONCE(task_thread_info(next_p)->flags);
611 	tifp = READ_ONCE(task_thread_info(prev_p)->flags);
612 
613 	switch_to_bitmap(tifp);
614 
615 	propagate_user_return_notify(prev_p, next_p);
616 
617 	if ((tifp & _TIF_BLOCKSTEP || tifn & _TIF_BLOCKSTEP) &&
618 	    arch_has_block_step()) {
619 		unsigned long debugctl, msk;
620 
621 		rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctl);
622 		debugctl &= ~DEBUGCTLMSR_BTF;
623 		msk = tifn & _TIF_BLOCKSTEP;
624 		debugctl |= (msk >> TIF_BLOCKSTEP) << DEBUGCTLMSR_BTF_SHIFT;
625 		wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctl);
626 	}
627 
628 	if ((tifp ^ tifn) & _TIF_NOTSC)
629 		cr4_toggle_bits_irqsoff(X86_CR4_TSD);
630 
631 	if ((tifp ^ tifn) & _TIF_NOCPUID)
632 		set_cpuid_faulting(!!(tifn & _TIF_NOCPUID));
633 
634 	if (likely(!((tifp | tifn) & _TIF_SPEC_FORCE_UPDATE))) {
635 		__speculation_ctrl_update(tifp, tifn);
636 	} else {
637 		speculation_ctrl_update_tif(prev_p);
638 		tifn = speculation_ctrl_update_tif(next_p);
639 
640 		/* Enforce MSR update to ensure consistent state */
641 		__speculation_ctrl_update(~tifn, tifn);
642 	}
643 
644 	if ((tifp ^ tifn) & _TIF_SLD)
645 		switch_to_sld(tifn);
646 }
647 
648 /*
649  * Idle related variables and functions
650  */
651 unsigned long boot_option_idle_override = IDLE_NO_OVERRIDE;
652 EXPORT_SYMBOL(boot_option_idle_override);
653 
654 static void (*x86_idle)(void);
655 
656 #ifndef CONFIG_SMP
657 static inline void play_dead(void)
658 {
659 	BUG();
660 }
661 #endif
662 
663 void arch_cpu_idle_enter(void)
664 {
665 	tsc_verify_tsc_adjust(false);
666 	local_touch_nmi();
667 }
668 
669 void arch_cpu_idle_dead(void)
670 {
671 	play_dead();
672 }
673 
674 /*
675  * Called from the generic idle code.
676  */
677 void arch_cpu_idle(void)
678 {
679 	x86_idle();
680 }
681 
682 /*
683  * We use this if we don't have any better idle routine..
684  */
685 void __cpuidle default_idle(void)
686 {
687 	trace_cpu_idle_rcuidle(1, smp_processor_id());
688 	safe_halt();
689 	trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, smp_processor_id());
690 }
691 #if defined(CONFIG_APM_MODULE) || defined(CONFIG_HALTPOLL_CPUIDLE_MODULE)
692 EXPORT_SYMBOL(default_idle);
693 #endif
694 
695 #ifdef CONFIG_XEN
696 bool xen_set_default_idle(void)
697 {
698 	bool ret = !!x86_idle;
699 
700 	x86_idle = default_idle;
701 
702 	return ret;
703 }
704 #endif
705 
706 void stop_this_cpu(void *dummy)
707 {
708 	local_irq_disable();
709 	/*
710 	 * Remove this CPU:
711 	 */
712 	set_cpu_online(smp_processor_id(), false);
713 	disable_local_APIC();
714 	mcheck_cpu_clear(this_cpu_ptr(&cpu_info));
715 
716 	/*
717 	 * Use wbinvd on processors that support SME. This provides support
718 	 * for performing a successful kexec when going from SME inactive
719 	 * to SME active (or vice-versa). The cache must be cleared so that
720 	 * if there are entries with the same physical address, both with and
721 	 * without the encryption bit, they don't race each other when flushed
722 	 * and potentially end up with the wrong entry being committed to
723 	 * memory.
724 	 */
725 	if (boot_cpu_has(X86_FEATURE_SME))
726 		native_wbinvd();
727 	for (;;) {
728 		/*
729 		 * Use native_halt() so that memory contents don't change
730 		 * (stack usage and variables) after possibly issuing the
731 		 * native_wbinvd() above.
732 		 */
733 		native_halt();
734 	}
735 }
736 
737 /*
738  * AMD Erratum 400 aware idle routine. We handle it the same way as C3 power
739  * states (local apic timer and TSC stop).
740  */
741 static void amd_e400_idle(void)
742 {
743 	/*
744 	 * We cannot use static_cpu_has_bug() here because X86_BUG_AMD_APIC_C1E
745 	 * gets set after static_cpu_has() places have been converted via
746 	 * alternatives.
747 	 */
748 	if (!boot_cpu_has_bug(X86_BUG_AMD_APIC_C1E)) {
749 		default_idle();
750 		return;
751 	}
752 
753 	tick_broadcast_enter();
754 
755 	default_idle();
756 
757 	/*
758 	 * The switch back from broadcast mode needs to be called with
759 	 * interrupts disabled.
760 	 */
761 	local_irq_disable();
762 	tick_broadcast_exit();
763 	local_irq_enable();
764 }
765 
766 /*
767  * Intel Core2 and older machines prefer MWAIT over HALT for C1.
768  * We can't rely on cpuidle installing MWAIT, because it will not load
769  * on systems that support only C1 -- so the boot default must be MWAIT.
770  *
771  * Some AMD machines are the opposite, they depend on using HALT.
772  *
773  * So for default C1, which is used during boot until cpuidle loads,
774  * use MWAIT-C1 on Intel HW that has it, else use HALT.
775  */
776 static int prefer_mwait_c1_over_halt(const struct cpuinfo_x86 *c)
777 {
778 	if (c->x86_vendor != X86_VENDOR_INTEL)
779 		return 0;
780 
781 	if (!cpu_has(c, X86_FEATURE_MWAIT) || boot_cpu_has_bug(X86_BUG_MONITOR))
782 		return 0;
783 
784 	return 1;
785 }
786 
787 /*
788  * MONITOR/MWAIT with no hints, used for default C1 state. This invokes MWAIT
789  * with interrupts enabled and no flags, which is backwards compatible with the
790  * original MWAIT implementation.
791  */
792 static __cpuidle void mwait_idle(void)
793 {
794 	if (!current_set_polling_and_test()) {
795 		trace_cpu_idle_rcuidle(1, smp_processor_id());
796 		if (this_cpu_has(X86_BUG_CLFLUSH_MONITOR)) {
797 			mb(); /* quirk */
798 			clflush((void *)&current_thread_info()->flags);
799 			mb(); /* quirk */
800 		}
801 
802 		__monitor((void *)&current_thread_info()->flags, 0, 0);
803 		if (!need_resched())
804 			__sti_mwait(0, 0);
805 		else
806 			local_irq_enable();
807 		trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, smp_processor_id());
808 	} else {
809 		local_irq_enable();
810 	}
811 	__current_clr_polling();
812 }
813 
814 void select_idle_routine(const struct cpuinfo_x86 *c)
815 {
816 #ifdef CONFIG_SMP
817 	if (boot_option_idle_override == IDLE_POLL && smp_num_siblings > 1)
818 		pr_warn_once("WARNING: polling idle and HT enabled, performance may degrade\n");
819 #endif
820 	if (x86_idle || boot_option_idle_override == IDLE_POLL)
821 		return;
822 
823 	if (boot_cpu_has_bug(X86_BUG_AMD_E400)) {
824 		pr_info("using AMD E400 aware idle routine\n");
825 		x86_idle = amd_e400_idle;
826 	} else if (prefer_mwait_c1_over_halt(c)) {
827 		pr_info("using mwait in idle threads\n");
828 		x86_idle = mwait_idle;
829 	} else
830 		x86_idle = default_idle;
831 }
832 
833 void amd_e400_c1e_apic_setup(void)
834 {
835 	if (boot_cpu_has_bug(X86_BUG_AMD_APIC_C1E)) {
836 		pr_info("Switch to broadcast mode on CPU%d\n", smp_processor_id());
837 		local_irq_disable();
838 		tick_broadcast_force();
839 		local_irq_enable();
840 	}
841 }
842 
843 void __init arch_post_acpi_subsys_init(void)
844 {
845 	u32 lo, hi;
846 
847 	if (!boot_cpu_has_bug(X86_BUG_AMD_E400))
848 		return;
849 
850 	/*
851 	 * AMD E400 detection needs to happen after ACPI has been enabled. If
852 	 * the machine is affected K8_INTP_C1E_ACTIVE_MASK bits are set in
853 	 * MSR_K8_INT_PENDING_MSG.
854 	 */
855 	rdmsr(MSR_K8_INT_PENDING_MSG, lo, hi);
856 	if (!(lo & K8_INTP_C1E_ACTIVE_MASK))
857 		return;
858 
859 	boot_cpu_set_bug(X86_BUG_AMD_APIC_C1E);
860 
861 	if (!boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
862 		mark_tsc_unstable("TSC halt in AMD C1E");
863 	pr_info("System has AMD C1E enabled\n");
864 }
865 
866 static int __init idle_setup(char *str)
867 {
868 	if (!str)
869 		return -EINVAL;
870 
871 	if (!strcmp(str, "poll")) {
872 		pr_info("using polling idle threads\n");
873 		boot_option_idle_override = IDLE_POLL;
874 		cpu_idle_poll_ctrl(true);
875 	} else if (!strcmp(str, "halt")) {
876 		/*
877 		 * When the boot option of idle=halt is added, halt is
878 		 * forced to be used for CPU idle. In such case CPU C2/C3
879 		 * won't be used again.
880 		 * To continue to load the CPU idle driver, don't touch
881 		 * the boot_option_idle_override.
882 		 */
883 		x86_idle = default_idle;
884 		boot_option_idle_override = IDLE_HALT;
885 	} else if (!strcmp(str, "nomwait")) {
886 		/*
887 		 * If the boot option of "idle=nomwait" is added,
888 		 * it means that mwait will be disabled for CPU C2/C3
889 		 * states. In such case it won't touch the variable
890 		 * of boot_option_idle_override.
891 		 */
892 		boot_option_idle_override = IDLE_NOMWAIT;
893 	} else
894 		return -1;
895 
896 	return 0;
897 }
898 early_param("idle", idle_setup);
899 
900 unsigned long arch_align_stack(unsigned long sp)
901 {
902 	if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
903 		sp -= get_random_int() % 8192;
904 	return sp & ~0xf;
905 }
906 
907 unsigned long arch_randomize_brk(struct mm_struct *mm)
908 {
909 	return randomize_page(mm->brk, 0x02000000);
910 }
911 
912 /*
913  * Called from fs/proc with a reference on @p to find the function
914  * which called into schedule(). This needs to be done carefully
915  * because the task might wake up and we might look at a stack
916  * changing under us.
917  */
918 unsigned long get_wchan(struct task_struct *p)
919 {
920 	unsigned long start, bottom, top, sp, fp, ip, ret = 0;
921 	int count = 0;
922 
923 	if (p == current || p->state == TASK_RUNNING)
924 		return 0;
925 
926 	if (!try_get_task_stack(p))
927 		return 0;
928 
929 	start = (unsigned long)task_stack_page(p);
930 	if (!start)
931 		goto out;
932 
933 	/*
934 	 * Layout of the stack page:
935 	 *
936 	 * ----------- topmax = start + THREAD_SIZE - sizeof(unsigned long)
937 	 * PADDING
938 	 * ----------- top = topmax - TOP_OF_KERNEL_STACK_PADDING
939 	 * stack
940 	 * ----------- bottom = start
941 	 *
942 	 * The tasks stack pointer points at the location where the
943 	 * framepointer is stored. The data on the stack is:
944 	 * ... IP FP ... IP FP
945 	 *
946 	 * We need to read FP and IP, so we need to adjust the upper
947 	 * bound by another unsigned long.
948 	 */
949 	top = start + THREAD_SIZE - TOP_OF_KERNEL_STACK_PADDING;
950 	top -= 2 * sizeof(unsigned long);
951 	bottom = start;
952 
953 	sp = READ_ONCE(p->thread.sp);
954 	if (sp < bottom || sp > top)
955 		goto out;
956 
957 	fp = READ_ONCE_NOCHECK(((struct inactive_task_frame *)sp)->bp);
958 	do {
959 		if (fp < bottom || fp > top)
960 			goto out;
961 		ip = READ_ONCE_NOCHECK(*(unsigned long *)(fp + sizeof(unsigned long)));
962 		if (!in_sched_functions(ip)) {
963 			ret = ip;
964 			goto out;
965 		}
966 		fp = READ_ONCE_NOCHECK(*(unsigned long *)fp);
967 	} while (count++ < 16 && p->state != TASK_RUNNING);
968 
969 out:
970 	put_task_stack(p);
971 	return ret;
972 }
973 
974 long do_arch_prctl_common(struct task_struct *task, int option,
975 			  unsigned long cpuid_enabled)
976 {
977 	switch (option) {
978 	case ARCH_GET_CPUID:
979 		return get_cpuid_mode();
980 	case ARCH_SET_CPUID:
981 		return set_cpuid_mode(task, cpuid_enabled);
982 	}
983 
984 	return -EINVAL;
985 }
986