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/cpu.h>
9#include <linux/prctl.h>
10#include <linux/slab.h>
11#include <linux/sched.h>
12#include <linux/sched/idle.h>
13#include <linux/sched/debug.h>
14#include <linux/sched/task.h>
15#include <linux/sched/task_stack.h>
16#include <linux/init.h>
17#include <linux/export.h>
18#include <linux/pm.h>
19#include <linux/tick.h>
20#include <linux/random.h>
21#include <linux/user-return-notifier.h>
22#include <linux/dmi.h>
23#include <linux/utsname.h>
24#include <linux/stackprotector.h>
25#include <linux/cpuidle.h>
26#include <linux/acpi.h>
27#include <linux/elf-randomize.h>
28#include <linux/static_call.h>
29#include <trace/events/power.h>
30#include <linux/hw_breakpoint.h>
31#include <linux/entry-common.h>
32#include <asm/cpu.h>
33#include <asm/apic.h>
34#include <linux/uaccess.h>
35#include <asm/mwait.h>
36#include <asm/fpu/api.h>
37#include <asm/fpu/sched.h>
38#include <asm/fpu/xstate.h>
39#include <asm/debugreg.h>
40#include <asm/nmi.h>
41#include <asm/tlbflush.h>
42#include <asm/mce.h>
43#include <asm/vm86.h>
44#include <asm/switch_to.h>
45#include <asm/desc.h>
46#include <asm/prctl.h>
47#include <asm/spec-ctrl.h>
48#include <asm/io_bitmap.h>
49#include <asm/proto.h>
50#include <asm/frame.h>
51#include <asm/unwind.h>
52#include <asm/tdx.h>
53#include <asm/mmu_context.h>
54#include <asm/shstk.h>
55
56#include "process.h"
57
58/*
59 * per-CPU TSS segments. Threads are completely 'soft' on Linux,
60 * no more per-task TSS's. The TSS size is kept cacheline-aligned
61 * so they are allowed to end up in the .data..cacheline_aligned
62 * section. Since TSS's are completely CPU-local, we want them
63 * on exact cacheline boundaries, to eliminate cacheline ping-pong.
64 */
65__visible DEFINE_PER_CPU_PAGE_ALIGNED(struct tss_struct, cpu_tss_rw) = {
66 .x86_tss = {
67 /*
68 * .sp0 is only used when entering ring 0 from a lower
69 * privilege level. Since the init task never runs anything
70 * but ring 0 code, there is no need for a valid value here.
71 * Poison it.
72 */
73 .sp0 = (1UL << (BITS_PER_LONG-1)) + 1,
74
75#ifdef CONFIG_X86_32
76 .sp1 = TOP_OF_INIT_STACK,
77
78 .ss0 = __KERNEL_DS,
79 .ss1 = __KERNEL_CS,
80#endif
81 .io_bitmap_base = IO_BITMAP_OFFSET_INVALID,
82 },
83};
84EXPORT_PER_CPU_SYMBOL(cpu_tss_rw);
85
86DEFINE_PER_CPU(bool, __tss_limit_invalid);
87EXPORT_PER_CPU_SYMBOL_GPL(__tss_limit_invalid);
88
89/*
90 * this gets called so that we can store lazy state into memory and copy the
91 * current task into the new thread.
92 */
93int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
94{
95 memcpy(dst, src, arch_task_struct_size);
96#ifdef CONFIG_VM86
97 dst->thread.vm86 = NULL;
98#endif
99 /* Drop the copied pointer to current's fpstate */
100 dst->thread.fpu.fpstate = NULL;
101
102 return 0;
103}
104
105#ifdef CONFIG_X86_64
106void arch_release_task_struct(struct task_struct *tsk)
107{
108 if (fpu_state_size_dynamic())
109 fpstate_free(fpu: &tsk->thread.fpu);
110}
111#endif
112
113/*
114 * Free thread data structures etc..
115 */
116void exit_thread(struct task_struct *tsk)
117{
118 struct thread_struct *t = &tsk->thread;
119 struct fpu *fpu = &t->fpu;
120
121 if (test_thread_flag(TIF_IO_BITMAP))
122 io_bitmap_exit(tsk);
123
124 free_vm86(t);
125
126 shstk_free(p: tsk);
127 fpu__drop(fpu);
128}
129
130static int set_new_tls(struct task_struct *p, unsigned long tls)
131{
132 struct user_desc __user *utls = (struct user_desc __user *)tls;
133
134 if (in_ia32_syscall())
135 return do_set_thread_area(p, idx: -1, info: utls, can_allocate: 0);
136 else
137 return do_set_thread_area_64(p, ARCH_SET_FS, tls);
138}
139
140__visible void ret_from_fork(struct task_struct *prev, struct pt_regs *regs,
141 int (*fn)(void *), void *fn_arg)
142{
143 schedule_tail(prev);
144
145 /* Is this a kernel thread? */
146 if (unlikely(fn)) {
147 fn(fn_arg);
148 /*
149 * A kernel thread is allowed to return here after successfully
150 * calling kernel_execve(). Exit to userspace to complete the
151 * execve() syscall.
152 */
153 regs->ax = 0;
154 }
155
156 syscall_exit_to_user_mode(regs);
157}
158
159int copy_thread(struct task_struct *p, const struct kernel_clone_args *args)
160{
161 unsigned long clone_flags = args->flags;
162 unsigned long sp = args->stack;
163 unsigned long tls = args->tls;
164 struct inactive_task_frame *frame;
165 struct fork_frame *fork_frame;
166 struct pt_regs *childregs;
167 unsigned long new_ssp;
168 int ret = 0;
169
170 childregs = task_pt_regs(p);
171 fork_frame = container_of(childregs, struct fork_frame, regs);
172 frame = &fork_frame->frame;
173
174 frame->bp = encode_frame_pointer(regs: childregs);
175 frame->ret_addr = (unsigned long) ret_from_fork_asm;
176 p->thread.sp = (unsigned long) fork_frame;
177 p->thread.io_bitmap = NULL;
178 p->thread.iopl_warn = 0;
179 memset(p->thread.ptrace_bps, 0, sizeof(p->thread.ptrace_bps));
180
181#ifdef CONFIG_X86_64
182 current_save_fsgs();
183 p->thread.fsindex = current->thread.fsindex;
184 p->thread.fsbase = current->thread.fsbase;
185 p->thread.gsindex = current->thread.gsindex;
186 p->thread.gsbase = current->thread.gsbase;
187
188 savesegment(es, p->thread.es);
189 savesegment(ds, p->thread.ds);
190
191 if (p->mm && (clone_flags & (CLONE_VM | CLONE_VFORK)) == CLONE_VM)
192 set_bit(MM_CONTEXT_LOCK_LAM, addr: &p->mm->context.flags);
193#else
194 p->thread.sp0 = (unsigned long) (childregs + 1);
195 savesegment(gs, p->thread.gs);
196 /*
197 * Clear all status flags including IF and set fixed bit. 64bit
198 * does not have this initialization as the frame does not contain
199 * flags. The flags consistency (especially vs. AC) is there
200 * ensured via objtool, which lacks 32bit support.
201 */
202 frame->flags = X86_EFLAGS_FIXED;
203#endif
204
205 /*
206 * Allocate a new shadow stack for thread if needed. If shadow stack,
207 * is disabled, new_ssp will remain 0, and fpu_clone() will know not to
208 * update it.
209 */
210 new_ssp = shstk_alloc_thread_stack(p, clone_flags, stack_size: args->stack_size);
211 if (IS_ERR_VALUE(new_ssp))
212 return PTR_ERR(ptr: (void *)new_ssp);
213
214 fpu_clone(dst: p, clone_flags, minimal: args->fn, shstk_addr: new_ssp);
215
216 /* Kernel thread ? */
217 if (unlikely(p->flags & PF_KTHREAD)) {
218 p->thread.pkru = pkru_get_init_value();
219 memset(childregs, 0, sizeof(struct pt_regs));
220 kthread_frame_init(frame, fun: args->fn, arg: args->fn_arg);
221 return 0;
222 }
223
224 /*
225 * Clone current's PKRU value from hardware. tsk->thread.pkru
226 * is only valid when scheduled out.
227 */
228 p->thread.pkru = read_pkru();
229
230 frame->bx = 0;
231 *childregs = *current_pt_regs();
232 childregs->ax = 0;
233 if (sp)
234 childregs->sp = sp;
235
236 if (unlikely(args->fn)) {
237 /*
238 * A user space thread, but it doesn't return to
239 * ret_after_fork().
240 *
241 * In order to indicate that to tools like gdb,
242 * we reset the stack and instruction pointers.
243 *
244 * It does the same kernel frame setup to return to a kernel
245 * function that a kernel thread does.
246 */
247 childregs->sp = 0;
248 childregs->ip = 0;
249 kthread_frame_init(frame, fun: args->fn, arg: args->fn_arg);
250 return 0;
251 }
252
253 /* Set a new TLS for the child thread? */
254 if (clone_flags & CLONE_SETTLS)
255 ret = set_new_tls(p, tls);
256
257 if (!ret && unlikely(test_tsk_thread_flag(current, TIF_IO_BITMAP)))
258 io_bitmap_share(tsk: p);
259
260 return ret;
261}
262
263static void pkru_flush_thread(void)
264{
265 /*
266 * If PKRU is enabled the default PKRU value has to be loaded into
267 * the hardware right here (similar to context switch).
268 */
269 pkru_write_default();
270}
271
272void flush_thread(void)
273{
274 struct task_struct *tsk = current;
275
276 flush_ptrace_hw_breakpoint(tsk);
277 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
278
279 fpu_flush_thread();
280 pkru_flush_thread();
281}
282
283void disable_TSC(void)
284{
285 preempt_disable();
286 if (!test_and_set_thread_flag(TIF_NOTSC))
287 /*
288 * Must flip the CPU state synchronously with
289 * TIF_NOTSC in the current running context.
290 */
291 cr4_set_bits(X86_CR4_TSD);
292 preempt_enable();
293}
294
295static void enable_TSC(void)
296{
297 preempt_disable();
298 if (test_and_clear_thread_flag(TIF_NOTSC))
299 /*
300 * Must flip the CPU state synchronously with
301 * TIF_NOTSC in the current running context.
302 */
303 cr4_clear_bits(X86_CR4_TSD);
304 preempt_enable();
305}
306
307int get_tsc_mode(unsigned long adr)
308{
309 unsigned int val;
310
311 if (test_thread_flag(TIF_NOTSC))
312 val = PR_TSC_SIGSEGV;
313 else
314 val = PR_TSC_ENABLE;
315
316 return put_user(val, (unsigned int __user *)adr);
317}
318
319int set_tsc_mode(unsigned int val)
320{
321 if (val == PR_TSC_SIGSEGV)
322 disable_TSC();
323 else if (val == PR_TSC_ENABLE)
324 enable_TSC();
325 else
326 return -EINVAL;
327
328 return 0;
329}
330
331DEFINE_PER_CPU(u64, msr_misc_features_shadow);
332
333static void set_cpuid_faulting(bool on)
334{
335 u64 msrval;
336
337 msrval = this_cpu_read(msr_misc_features_shadow);
338 msrval &= ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT;
339 msrval |= (on << MSR_MISC_FEATURES_ENABLES_CPUID_FAULT_BIT);
340 this_cpu_write(msr_misc_features_shadow, msrval);
341 wrmsrl(MSR_MISC_FEATURES_ENABLES, val: msrval);
342}
343
344static void disable_cpuid(void)
345{
346 preempt_disable();
347 if (!test_and_set_thread_flag(TIF_NOCPUID)) {
348 /*
349 * Must flip the CPU state synchronously with
350 * TIF_NOCPUID in the current running context.
351 */
352 set_cpuid_faulting(true);
353 }
354 preempt_enable();
355}
356
357static void enable_cpuid(void)
358{
359 preempt_disable();
360 if (test_and_clear_thread_flag(TIF_NOCPUID)) {
361 /*
362 * Must flip the CPU state synchronously with
363 * TIF_NOCPUID in the current running context.
364 */
365 set_cpuid_faulting(false);
366 }
367 preempt_enable();
368}
369
370static int get_cpuid_mode(void)
371{
372 return !test_thread_flag(TIF_NOCPUID);
373}
374
375static int set_cpuid_mode(unsigned long cpuid_enabled)
376{
377 if (!boot_cpu_has(X86_FEATURE_CPUID_FAULT))
378 return -ENODEV;
379
380 if (cpuid_enabled)
381 enable_cpuid();
382 else
383 disable_cpuid();
384
385 return 0;
386}
387
388/*
389 * Called immediately after a successful exec.
390 */
391void arch_setup_new_exec(void)
392{
393 /* If cpuid was previously disabled for this task, re-enable it. */
394 if (test_thread_flag(TIF_NOCPUID))
395 enable_cpuid();
396
397 /*
398 * Don't inherit TIF_SSBD across exec boundary when
399 * PR_SPEC_DISABLE_NOEXEC is used.
400 */
401 if (test_thread_flag(TIF_SSBD) &&
402 task_spec_ssb_noexec(current)) {
403 clear_thread_flag(TIF_SSBD);
404 task_clear_spec_ssb_disable(current);
405 task_clear_spec_ssb_noexec(current);
406 speculation_ctrl_update(read_thread_flags());
407 }
408
409 mm_reset_untag_mask(current->mm);
410}
411
412#ifdef CONFIG_X86_IOPL_IOPERM
413static inline void switch_to_bitmap(unsigned long tifp)
414{
415 /*
416 * Invalidate I/O bitmap if the previous task used it. This prevents
417 * any possible leakage of an active I/O bitmap.
418 *
419 * If the next task has an I/O bitmap it will handle it on exit to
420 * user mode.
421 */
422 if (tifp & _TIF_IO_BITMAP)
423 tss_invalidate_io_bitmap();
424}
425
426static void tss_copy_io_bitmap(struct tss_struct *tss, struct io_bitmap *iobm)
427{
428 /*
429 * Copy at least the byte range of the incoming tasks bitmap which
430 * covers the permitted I/O ports.
431 *
432 * If the previous task which used an I/O bitmap had more bits
433 * permitted, then the copy needs to cover those as well so they
434 * get turned off.
435 */
436 memcpy(tss->io_bitmap.bitmap, iobm->bitmap,
437 max(tss->io_bitmap.prev_max, iobm->max));
438
439 /*
440 * Store the new max and the sequence number of this bitmap
441 * and a pointer to the bitmap itself.
442 */
443 tss->io_bitmap.prev_max = iobm->max;
444 tss->io_bitmap.prev_sequence = iobm->sequence;
445}
446
447/**
448 * native_tss_update_io_bitmap - Update I/O bitmap before exiting to user mode
449 */
450void native_tss_update_io_bitmap(void)
451{
452 struct tss_struct *tss = this_cpu_ptr(&cpu_tss_rw);
453 struct thread_struct *t = &current->thread;
454 u16 *base = &tss->x86_tss.io_bitmap_base;
455
456 if (!test_thread_flag(TIF_IO_BITMAP)) {
457 native_tss_invalidate_io_bitmap();
458 return;
459 }
460
461 if (IS_ENABLED(CONFIG_X86_IOPL_IOPERM) && t->iopl_emul == 3) {
462 *base = IO_BITMAP_OFFSET_VALID_ALL;
463 } else {
464 struct io_bitmap *iobm = t->io_bitmap;
465
466 /*
467 * Only copy bitmap data when the sequence number differs. The
468 * update time is accounted to the incoming task.
469 */
470 if (tss->io_bitmap.prev_sequence != iobm->sequence)
471 tss_copy_io_bitmap(tss, iobm);
472
473 /* Enable the bitmap */
474 *base = IO_BITMAP_OFFSET_VALID_MAP;
475 }
476
477 /*
478 * Make sure that the TSS limit is covering the IO bitmap. It might have
479 * been cut down by a VMEXIT to 0x67 which would cause a subsequent I/O
480 * access from user space to trigger a #GP because tbe bitmap is outside
481 * the TSS limit.
482 */
483 refresh_tss_limit();
484}
485#else /* CONFIG_X86_IOPL_IOPERM */
486static inline void switch_to_bitmap(unsigned long tifp) { }
487#endif
488
489#ifdef CONFIG_SMP
490
491struct ssb_state {
492 struct ssb_state *shared_state;
493 raw_spinlock_t lock;
494 unsigned int disable_state;
495 unsigned long local_state;
496};
497
498#define LSTATE_SSB 0
499
500static DEFINE_PER_CPU(struct ssb_state, ssb_state);
501
502void speculative_store_bypass_ht_init(void)
503{
504 struct ssb_state *st = this_cpu_ptr(&ssb_state);
505 unsigned int this_cpu = smp_processor_id();
506 unsigned int cpu;
507
508 st->local_state = 0;
509
510 /*
511 * Shared state setup happens once on the first bringup
512 * of the CPU. It's not destroyed on CPU hotunplug.
513 */
514 if (st->shared_state)
515 return;
516
517 raw_spin_lock_init(&st->lock);
518
519 /*
520 * Go over HT siblings and check whether one of them has set up the
521 * shared state pointer already.
522 */
523 for_each_cpu(cpu, topology_sibling_cpumask(this_cpu)) {
524 if (cpu == this_cpu)
525 continue;
526
527 if (!per_cpu(ssb_state, cpu).shared_state)
528 continue;
529
530 /* Link it to the state of the sibling: */
531 st->shared_state = per_cpu(ssb_state, cpu).shared_state;
532 return;
533 }
534
535 /*
536 * First HT sibling to come up on the core. Link shared state of
537 * the first HT sibling to itself. The siblings on the same core
538 * which come up later will see the shared state pointer and link
539 * themselves to the state of this CPU.
540 */
541 st->shared_state = st;
542}
543
544/*
545 * Logic is: First HT sibling enables SSBD for both siblings in the core
546 * and last sibling to disable it, disables it for the whole core. This how
547 * MSR_SPEC_CTRL works in "hardware":
548 *
549 * CORE_SPEC_CTRL = THREAD0_SPEC_CTRL | THREAD1_SPEC_CTRL
550 */
551static __always_inline void amd_set_core_ssb_state(unsigned long tifn)
552{
553 struct ssb_state *st = this_cpu_ptr(&ssb_state);
554 u64 msr = x86_amd_ls_cfg_base;
555
556 if (!static_cpu_has(X86_FEATURE_ZEN)) {
557 msr |= ssbd_tif_to_amd_ls_cfg(tifn);
558 wrmsrl(MSR_AMD64_LS_CFG, val: msr);
559 return;
560 }
561
562 if (tifn & _TIF_SSBD) {
563 /*
564 * Since this can race with prctl(), block reentry on the
565 * same CPU.
566 */
567 if (__test_and_set_bit(LSTATE_SSB, &st->local_state))
568 return;
569
570 msr |= x86_amd_ls_cfg_ssbd_mask;
571
572 raw_spin_lock(&st->shared_state->lock);
573 /* First sibling enables SSBD: */
574 if (!st->shared_state->disable_state)
575 wrmsrl(MSR_AMD64_LS_CFG, val: msr);
576 st->shared_state->disable_state++;
577 raw_spin_unlock(&st->shared_state->lock);
578 } else {
579 if (!__test_and_clear_bit(LSTATE_SSB, &st->local_state))
580 return;
581
582 raw_spin_lock(&st->shared_state->lock);
583 st->shared_state->disable_state--;
584 if (!st->shared_state->disable_state)
585 wrmsrl(MSR_AMD64_LS_CFG, val: msr);
586 raw_spin_unlock(&st->shared_state->lock);
587 }
588}
589#else
590static __always_inline void amd_set_core_ssb_state(unsigned long tifn)
591{
592 u64 msr = x86_amd_ls_cfg_base | ssbd_tif_to_amd_ls_cfg(tifn);
593
594 wrmsrl(MSR_AMD64_LS_CFG, msr);
595}
596#endif
597
598static __always_inline void amd_set_ssb_virt_state(unsigned long tifn)
599{
600 /*
601 * SSBD has the same definition in SPEC_CTRL and VIRT_SPEC_CTRL,
602 * so ssbd_tif_to_spec_ctrl() just works.
603 */
604 wrmsrl(MSR_AMD64_VIRT_SPEC_CTRL, val: ssbd_tif_to_spec_ctrl(tifn));
605}
606
607/*
608 * Update the MSRs managing speculation control, during context switch.
609 *
610 * tifp: Previous task's thread flags
611 * tifn: Next task's thread flags
612 */
613static __always_inline void __speculation_ctrl_update(unsigned long tifp,
614 unsigned long tifn)
615{
616 unsigned long tif_diff = tifp ^ tifn;
617 u64 msr = x86_spec_ctrl_base;
618 bool updmsr = false;
619
620 lockdep_assert_irqs_disabled();
621
622 /* Handle change of TIF_SSBD depending on the mitigation method. */
623 if (static_cpu_has(X86_FEATURE_VIRT_SSBD)) {
624 if (tif_diff & _TIF_SSBD)
625 amd_set_ssb_virt_state(tifn);
626 } else if (static_cpu_has(X86_FEATURE_LS_CFG_SSBD)) {
627 if (tif_diff & _TIF_SSBD)
628 amd_set_core_ssb_state(tifn);
629 } else if (static_cpu_has(X86_FEATURE_SPEC_CTRL_SSBD) ||
630 static_cpu_has(X86_FEATURE_AMD_SSBD)) {
631 updmsr |= !!(tif_diff & _TIF_SSBD);
632 msr |= ssbd_tif_to_spec_ctrl(tifn);
633 }
634
635 /* Only evaluate TIF_SPEC_IB if conditional STIBP is enabled. */
636 if (IS_ENABLED(CONFIG_SMP) &&
637 static_branch_unlikely(&switch_to_cond_stibp)) {
638 updmsr |= !!(tif_diff & _TIF_SPEC_IB);
639 msr |= stibp_tif_to_spec_ctrl(tifn);
640 }
641
642 if (updmsr)
643 update_spec_ctrl_cond(val: msr);
644}
645
646static unsigned long speculation_ctrl_update_tif(struct task_struct *tsk)
647{
648 if (test_and_clear_tsk_thread_flag(tsk, TIF_SPEC_FORCE_UPDATE)) {
649 if (task_spec_ssb_disable(p: tsk))
650 set_tsk_thread_flag(tsk, TIF_SSBD);
651 else
652 clear_tsk_thread_flag(tsk, TIF_SSBD);
653
654 if (task_spec_ib_disable(p: tsk))
655 set_tsk_thread_flag(tsk, TIF_SPEC_IB);
656 else
657 clear_tsk_thread_flag(tsk, TIF_SPEC_IB);
658 }
659 /* Return the updated threadinfo flags*/
660 return read_task_thread_flags(tsk);
661}
662
663void speculation_ctrl_update(unsigned long tif)
664{
665 unsigned long flags;
666
667 /* Forced update. Make sure all relevant TIF flags are different */
668 local_irq_save(flags);
669 __speculation_ctrl_update(tifp: ~tif, tifn: tif);
670 local_irq_restore(flags);
671}
672
673/* Called from seccomp/prctl update */
674void speculation_ctrl_update_current(void)
675{
676 preempt_disable();
677 speculation_ctrl_update(tif: speculation_ctrl_update_tif(current));
678 preempt_enable();
679}
680
681static inline void cr4_toggle_bits_irqsoff(unsigned long mask)
682{
683 unsigned long newval, cr4 = this_cpu_read(cpu_tlbstate.cr4);
684
685 newval = cr4 ^ mask;
686 if (newval != cr4) {
687 this_cpu_write(cpu_tlbstate.cr4, newval);
688 __write_cr4(x: newval);
689 }
690}
691
692void __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p)
693{
694 unsigned long tifp, tifn;
695
696 tifn = read_task_thread_flags(next_p);
697 tifp = read_task_thread_flags(prev_p);
698
699 switch_to_bitmap(tifp);
700
701 propagate_user_return_notify(prev: prev_p, next: next_p);
702
703 if ((tifp & _TIF_BLOCKSTEP || tifn & _TIF_BLOCKSTEP) &&
704 arch_has_block_step()) {
705 unsigned long debugctl, msk;
706
707 rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctl);
708 debugctl &= ~DEBUGCTLMSR_BTF;
709 msk = tifn & _TIF_BLOCKSTEP;
710 debugctl |= (msk >> TIF_BLOCKSTEP) << DEBUGCTLMSR_BTF_SHIFT;
711 wrmsrl(MSR_IA32_DEBUGCTLMSR, val: debugctl);
712 }
713
714 if ((tifp ^ tifn) & _TIF_NOTSC)
715 cr4_toggle_bits_irqsoff(X86_CR4_TSD);
716
717 if ((tifp ^ tifn) & _TIF_NOCPUID)
718 set_cpuid_faulting(!!(tifn & _TIF_NOCPUID));
719
720 if (likely(!((tifp | tifn) & _TIF_SPEC_FORCE_UPDATE))) {
721 __speculation_ctrl_update(tifp, tifn);
722 } else {
723 speculation_ctrl_update_tif(tsk: prev_p);
724 tifn = speculation_ctrl_update_tif(tsk: next_p);
725
726 /* Enforce MSR update to ensure consistent state */
727 __speculation_ctrl_update(tifp: ~tifn, tifn);
728 }
729}
730
731/*
732 * Idle related variables and functions
733 */
734unsigned long boot_option_idle_override = IDLE_NO_OVERRIDE;
735EXPORT_SYMBOL(boot_option_idle_override);
736
737/*
738 * We use this if we don't have any better idle routine..
739 */
740void __cpuidle default_idle(void)
741{
742 raw_safe_halt();
743 raw_local_irq_disable();
744}
745#if defined(CONFIG_APM_MODULE) || defined(CONFIG_HALTPOLL_CPUIDLE_MODULE)
746EXPORT_SYMBOL(default_idle);
747#endif
748
749DEFINE_STATIC_CALL_NULL(x86_idle, default_idle);
750
751static bool x86_idle_set(void)
752{
753 return !!static_call_query(x86_idle);
754}
755
756#ifndef CONFIG_SMP
757static inline void __noreturn play_dead(void)
758{
759 BUG();
760}
761#endif
762
763void arch_cpu_idle_enter(void)
764{
765 tsc_verify_tsc_adjust(resume: false);
766 local_touch_nmi();
767}
768
769void __noreturn arch_cpu_idle_dead(void)
770{
771 play_dead();
772}
773
774/*
775 * Called from the generic idle code.
776 */
777void __cpuidle arch_cpu_idle(void)
778{
779 static_call(x86_idle)();
780}
781EXPORT_SYMBOL_GPL(arch_cpu_idle);
782
783#ifdef CONFIG_XEN
784bool xen_set_default_idle(void)
785{
786 bool ret = x86_idle_set();
787
788 static_call_update(x86_idle, default_idle);
789
790 return ret;
791}
792#endif
793
794struct cpumask cpus_stop_mask;
795
796void __noreturn stop_this_cpu(void *dummy)
797{
798 struct cpuinfo_x86 *c = this_cpu_ptr(&cpu_info);
799 unsigned int cpu = smp_processor_id();
800
801 local_irq_disable();
802
803 /*
804 * Remove this CPU from the online mask and disable it
805 * unconditionally. This might be redundant in case that the reboot
806 * vector was handled late and stop_other_cpus() sent an NMI.
807 *
808 * According to SDM and APM NMIs can be accepted even after soft
809 * disabling the local APIC.
810 */
811 set_cpu_online(cpu, online: false);
812 disable_local_APIC();
813 mcheck_cpu_clear(c);
814
815 /*
816 * Use wbinvd on processors that support SME. This provides support
817 * for performing a successful kexec when going from SME inactive
818 * to SME active (or vice-versa). The cache must be cleared so that
819 * if there are entries with the same physical address, both with and
820 * without the encryption bit, they don't race each other when flushed
821 * and potentially end up with the wrong entry being committed to
822 * memory.
823 *
824 * Test the CPUID bit directly because the machine might've cleared
825 * X86_FEATURE_SME due to cmdline options.
826 */
827 if (c->extended_cpuid_level >= 0x8000001f && (cpuid_eax(op: 0x8000001f) & BIT(0)))
828 native_wbinvd();
829
830 /*
831 * This brings a cache line back and dirties it, but
832 * native_stop_other_cpus() will overwrite cpus_stop_mask after it
833 * observed that all CPUs reported stop. This write will invalidate
834 * the related cache line on this CPU.
835 */
836 cpumask_clear_cpu(cpu, dstp: &cpus_stop_mask);
837
838 for (;;) {
839 /*
840 * Use native_halt() so that memory contents don't change
841 * (stack usage and variables) after possibly issuing the
842 * native_wbinvd() above.
843 */
844 native_halt();
845 }
846}
847
848/*
849 * AMD Erratum 400 aware idle routine. We handle it the same way as C3 power
850 * states (local apic timer and TSC stop).
851 *
852 * XXX this function is completely buggered vs RCU and tracing.
853 */
854static void amd_e400_idle(void)
855{
856 /*
857 * We cannot use static_cpu_has_bug() here because X86_BUG_AMD_APIC_C1E
858 * gets set after static_cpu_has() places have been converted via
859 * alternatives.
860 */
861 if (!boot_cpu_has_bug(X86_BUG_AMD_APIC_C1E)) {
862 default_idle();
863 return;
864 }
865
866 tick_broadcast_enter();
867
868 default_idle();
869
870 tick_broadcast_exit();
871}
872
873/*
874 * Prefer MWAIT over HALT if MWAIT is supported, MWAIT_CPUID leaf
875 * exists and whenever MONITOR/MWAIT extensions are present there is at
876 * least one C1 substate.
877 *
878 * Do not prefer MWAIT if MONITOR instruction has a bug or idle=nomwait
879 * is passed to kernel commandline parameter.
880 */
881static int prefer_mwait_c1_over_halt(const struct cpuinfo_x86 *c)
882{
883 u32 eax, ebx, ecx, edx;
884
885 /* User has disallowed the use of MWAIT. Fallback to HALT */
886 if (boot_option_idle_override == IDLE_NOMWAIT)
887 return 0;
888
889 /* MWAIT is not supported on this platform. Fallback to HALT */
890 if (!cpu_has(c, X86_FEATURE_MWAIT))
891 return 0;
892
893 /* Monitor has a bug. Fallback to HALT */
894 if (boot_cpu_has_bug(X86_BUG_MONITOR))
895 return 0;
896
897 cpuid(CPUID_MWAIT_LEAF, eax: &eax, ebx: &ebx, ecx: &ecx, edx: &edx);
898
899 /*
900 * If MWAIT extensions are not available, it is safe to use MWAIT
901 * with EAX=0, ECX=0.
902 */
903 if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED))
904 return 1;
905
906 /*
907 * If MWAIT extensions are available, there should be at least one
908 * MWAIT C1 substate present.
909 */
910 return (edx & MWAIT_C1_SUBSTATE_MASK);
911}
912
913/*
914 * MONITOR/MWAIT with no hints, used for default C1 state. This invokes MWAIT
915 * with interrupts enabled and no flags, which is backwards compatible with the
916 * original MWAIT implementation.
917 */
918static __cpuidle void mwait_idle(void)
919{
920 if (!current_set_polling_and_test()) {
921 if (this_cpu_has(X86_BUG_CLFLUSH_MONITOR)) {
922 mb(); /* quirk */
923 clflush(p: (void *)&current_thread_info()->flags);
924 mb(); /* quirk */
925 }
926
927 __monitor(eax: (void *)&current_thread_info()->flags, ecx: 0, edx: 0);
928 if (!need_resched()) {
929 __sti_mwait(eax: 0, ecx: 0);
930 raw_local_irq_disable();
931 }
932 }
933 __current_clr_polling();
934}
935
936void select_idle_routine(const struct cpuinfo_x86 *c)
937{
938#ifdef CONFIG_SMP
939 if (boot_option_idle_override == IDLE_POLL && smp_num_siblings > 1)
940 pr_warn_once("WARNING: polling idle and HT enabled, performance may degrade\n");
941#endif
942 if (x86_idle_set() || boot_option_idle_override == IDLE_POLL)
943 return;
944
945 if (boot_cpu_has_bug(X86_BUG_AMD_E400)) {
946 pr_info("using AMD E400 aware idle routine\n");
947 static_call_update(x86_idle, amd_e400_idle);
948 } else if (prefer_mwait_c1_over_halt(c)) {
949 pr_info("using mwait in idle threads\n");
950 static_call_update(x86_idle, mwait_idle);
951 } else if (cpu_feature_enabled(X86_FEATURE_TDX_GUEST)) {
952 pr_info("using TDX aware idle routine\n");
953 static_call_update(x86_idle, tdx_safe_halt);
954 } else
955 static_call_update(x86_idle, default_idle);
956}
957
958void amd_e400_c1e_apic_setup(void)
959{
960 if (boot_cpu_has_bug(X86_BUG_AMD_APIC_C1E)) {
961 pr_info("Switch to broadcast mode on CPU%d\n", smp_processor_id());
962 local_irq_disable();
963 tick_broadcast_force();
964 local_irq_enable();
965 }
966}
967
968void __init arch_post_acpi_subsys_init(void)
969{
970 u32 lo, hi;
971
972 if (!boot_cpu_has_bug(X86_BUG_AMD_E400))
973 return;
974
975 /*
976 * AMD E400 detection needs to happen after ACPI has been enabled. If
977 * the machine is affected K8_INTP_C1E_ACTIVE_MASK bits are set in
978 * MSR_K8_INT_PENDING_MSG.
979 */
980 rdmsr(MSR_K8_INT_PENDING_MSG, lo, hi);
981 if (!(lo & K8_INTP_C1E_ACTIVE_MASK))
982 return;
983
984 boot_cpu_set_bug(X86_BUG_AMD_APIC_C1E);
985
986 if (!boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
987 mark_tsc_unstable(reason: "TSC halt in AMD C1E");
988 pr_info("System has AMD C1E enabled\n");
989}
990
991static int __init idle_setup(char *str)
992{
993 if (!str)
994 return -EINVAL;
995
996 if (!strcmp(str, "poll")) {
997 pr_info("using polling idle threads\n");
998 boot_option_idle_override = IDLE_POLL;
999 cpu_idle_poll_ctrl(enable: true);
1000 } else if (!strcmp(str, "halt")) {
1001 /*
1002 * When the boot option of idle=halt is added, halt is
1003 * forced to be used for CPU idle. In such case CPU C2/C3
1004 * won't be used again.
1005 * To continue to load the CPU idle driver, don't touch
1006 * the boot_option_idle_override.
1007 */
1008 static_call_update(x86_idle, default_idle);
1009 boot_option_idle_override = IDLE_HALT;
1010 } else if (!strcmp(str, "nomwait")) {
1011 /*
1012 * If the boot option of "idle=nomwait" is added,
1013 * it means that mwait will be disabled for CPU C1/C2/C3
1014 * states.
1015 */
1016 boot_option_idle_override = IDLE_NOMWAIT;
1017 } else
1018 return -1;
1019
1020 return 0;
1021}
1022early_param("idle", idle_setup);
1023
1024unsigned long arch_align_stack(unsigned long sp)
1025{
1026 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
1027 sp -= get_random_u32_below(ceil: 8192);
1028 return sp & ~0xf;
1029}
1030
1031unsigned long arch_randomize_brk(struct mm_struct *mm)
1032{
1033 return randomize_page(start: mm->brk, range: 0x02000000);
1034}
1035
1036/*
1037 * Called from fs/proc with a reference on @p to find the function
1038 * which called into schedule(). This needs to be done carefully
1039 * because the task might wake up and we might look at a stack
1040 * changing under us.
1041 */
1042unsigned long __get_wchan(struct task_struct *p)
1043{
1044 struct unwind_state state;
1045 unsigned long addr = 0;
1046
1047 if (!try_get_task_stack(tsk: p))
1048 return 0;
1049
1050 for (unwind_start(state: &state, task: p, NULL, NULL); !unwind_done(state: &state);
1051 unwind_next_frame(state: &state)) {
1052 addr = unwind_get_return_address(state: &state);
1053 if (!addr)
1054 break;
1055 if (in_sched_functions(addr))
1056 continue;
1057 break;
1058 }
1059
1060 put_task_stack(tsk: p);
1061
1062 return addr;
1063}
1064
1065long do_arch_prctl_common(int option, unsigned long arg2)
1066{
1067 switch (option) {
1068 case ARCH_GET_CPUID:
1069 return get_cpuid_mode();
1070 case ARCH_SET_CPUID:
1071 return set_cpuid_mode(arg2);
1072 case ARCH_GET_XCOMP_SUPP:
1073 case ARCH_GET_XCOMP_PERM:
1074 case ARCH_REQ_XCOMP_PERM:
1075 case ARCH_GET_XCOMP_GUEST_PERM:
1076 case ARCH_REQ_XCOMP_GUEST_PERM:
1077 return fpu_xstate_prctl(option, arg2);
1078 }
1079
1080 return -EINVAL;
1081}
1082

source code of linux/arch/x86/kernel/process.c