Loading...
1// SPDX-License-Identifier: GPL-2.0
2#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
3
4#include <linux/errno.h>
5#include <linux/kernel.h>
6#include <linux/mm.h>
7#include <linux/smp.h>
8#include <linux/prctl.h>
9#include <linux/slab.h>
10#include <linux/sched.h>
11#include <linux/sched/idle.h>
12#include <linux/sched/debug.h>
13#include <linux/sched/task.h>
14#include <linux/sched/task_stack.h>
15#include <linux/init.h>
16#include <linux/export.h>
17#include <linux/pm.h>
18#include <linux/tick.h>
19#include <linux/random.h>
20#include <linux/user-return-notifier.h>
21#include <linux/dmi.h>
22#include <linux/utsname.h>
23#include <linux/stackprotector.h>
24#include <linux/cpuidle.h>
25#include <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/api.h>
34#include <asm/fpu/sched.h>
35#include <asm/fpu/xstate.h>
36#include <asm/debugreg.h>
37#include <asm/nmi.h>
38#include <asm/tlbflush.h>
39#include <asm/mce.h>
40#include <asm/vm86.h>
41#include <asm/switch_to.h>
42#include <asm/desc.h>
43#include <asm/prctl.h>
44#include <asm/spec-ctrl.h>
45#include <asm/io_bitmap.h>
46#include <asm/proto.h>
47#include <asm/frame.h>
48#include <asm/unwind.h>
49#include <asm/tdx.h>
50
51#include "process.h"
52
53/*
54 * per-CPU TSS segments. Threads are completely 'soft' on Linux,
55 * no more per-task TSS's. The TSS size is kept cacheline-aligned
56 * so they are allowed to end up in the .data..cacheline_aligned
57 * section. Since TSS's are completely CPU-local, we want them
58 * on exact cacheline boundaries, to eliminate cacheline ping-pong.
59 */
60__visible DEFINE_PER_CPU_PAGE_ALIGNED(struct tss_struct, cpu_tss_rw) = {
61 .x86_tss = {
62 /*
63 * .sp0 is only used when entering ring 0 from a lower
64 * privilege level. Since the init task never runs anything
65 * but ring 0 code, there is no need for a valid value here.
66 * Poison it.
67 */
68 .sp0 = (1UL << (BITS_PER_LONG-1)) + 1,
69
70#ifdef CONFIG_X86_32
71 .sp1 = TOP_OF_INIT_STACK,
72
73 .ss0 = __KERNEL_DS,
74 .ss1 = __KERNEL_CS,
75#endif
76 .io_bitmap_base = IO_BITMAP_OFFSET_INVALID,
77 },
78};
79EXPORT_PER_CPU_SYMBOL(cpu_tss_rw);
80
81DEFINE_PER_CPU(bool, __tss_limit_invalid);
82EXPORT_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 */
88int 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 /* Drop the copied pointer to current's fpstate */
95 dst->thread.fpu.fpstate = NULL;
96
97 return 0;
98}
99
100#ifdef CONFIG_X86_64
101void arch_release_task_struct(struct task_struct *tsk)
102{
103 if (fpu_state_size_dynamic())
104 fpstate_free(&tsk->thread.fpu);
105}
106#endif
107
108/*
109 * Free thread data structures etc..
110 */
111void exit_thread(struct task_struct *tsk)
112{
113 struct thread_struct *t = &tsk->thread;
114 struct fpu *fpu = &t->fpu;
115
116 if (test_thread_flag(TIF_IO_BITMAP))
117 io_bitmap_exit(tsk);
118
119 free_vm86(t);
120
121 fpu__drop(fpu);
122}
123
124static int set_new_tls(struct task_struct *p, unsigned long tls)
125{
126 struct user_desc __user *utls = (struct user_desc __user *)tls;
127
128 if (in_ia32_syscall())
129 return do_set_thread_area(p, -1, utls, 0);
130 else
131 return do_set_thread_area_64(p, ARCH_SET_FS, tls);
132}
133
134int copy_thread(struct task_struct *p, const struct kernel_clone_args *args)
135{
136 unsigned long clone_flags = args->flags;
137 unsigned long sp = args->stack;
138 unsigned long tls = args->tls;
139 struct inactive_task_frame *frame;
140 struct fork_frame *fork_frame;
141 struct pt_regs *childregs;
142 int ret = 0;
143
144 childregs = task_pt_regs(p);
145 fork_frame = container_of(childregs, struct fork_frame, regs);
146 frame = &fork_frame->frame;
147
148 frame->bp = encode_frame_pointer(childregs);
149 frame->ret_addr = (unsigned long) ret_from_fork;
150 p->thread.sp = (unsigned long) fork_frame;
151 p->thread.io_bitmap = NULL;
152 p->thread.iopl_warn = 0;
153 memset(p->thread.ptrace_bps, 0, sizeof(p->thread.ptrace_bps));
154
155#ifdef CONFIG_X86_64
156 current_save_fsgs();
157 p->thread.fsindex = current->thread.fsindex;
158 p->thread.fsbase = current->thread.fsbase;
159 p->thread.gsindex = current->thread.gsindex;
160 p->thread.gsbase = current->thread.gsbase;
161
162 savesegment(es, p->thread.es);
163 savesegment(ds, p->thread.ds);
164#else
165 p->thread.sp0 = (unsigned long) (childregs + 1);
166 savesegment(gs, p->thread.gs);
167 /*
168 * Clear all status flags including IF and set fixed bit. 64bit
169 * does not have this initialization as the frame does not contain
170 * flags. The flags consistency (especially vs. AC) is there
171 * ensured via objtool, which lacks 32bit support.
172 */
173 frame->flags = X86_EFLAGS_FIXED;
174#endif
175
176 fpu_clone(p, clone_flags, args->fn);
177
178 /* Kernel thread ? */
179 if (unlikely(p->flags & PF_KTHREAD)) {
180 p->thread.pkru = pkru_get_init_value();
181 memset(childregs, 0, sizeof(struct pt_regs));
182 kthread_frame_init(frame, args->fn, args->fn_arg);
183 return 0;
184 }
185
186 /*
187 * Clone current's PKRU value from hardware. tsk->thread.pkru
188 * is only valid when scheduled out.
189 */
190 p->thread.pkru = read_pkru();
191
192 frame->bx = 0;
193 *childregs = *current_pt_regs();
194 childregs->ax = 0;
195 if (sp)
196 childregs->sp = sp;
197
198 if (unlikely(args->fn)) {
199 /*
200 * A user space thread, but it doesn't return to
201 * ret_after_fork().
202 *
203 * In order to indicate that to tools like gdb,
204 * we reset the stack and instruction pointers.
205 *
206 * It does the same kernel frame setup to return to a kernel
207 * function that a kernel thread does.
208 */
209 childregs->sp = 0;
210 childregs->ip = 0;
211 kthread_frame_init(frame, args->fn, args->fn_arg);
212 return 0;
213 }
214
215 /* Set a new TLS for the child thread? */
216 if (clone_flags & CLONE_SETTLS)
217 ret = set_new_tls(p, tls);
218
219 if (!ret && unlikely(test_tsk_thread_flag(current, TIF_IO_BITMAP)))
220 io_bitmap_share(p);
221
222 return ret;
223}
224
225static void pkru_flush_thread(void)
226{
227 /*
228 * If PKRU is enabled the default PKRU value has to be loaded into
229 * the hardware right here (similar to context switch).
230 */
231 pkru_write_default();
232}
233
234void flush_thread(void)
235{
236 struct task_struct *tsk = current;
237
238 flush_ptrace_hw_breakpoint(tsk);
239 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
240
241 fpu_flush_thread();
242 pkru_flush_thread();
243}
244
245void disable_TSC(void)
246{
247 preempt_disable();
248 if (!test_and_set_thread_flag(TIF_NOTSC))
249 /*
250 * Must flip the CPU state synchronously with
251 * TIF_NOTSC in the current running context.
252 */
253 cr4_set_bits(X86_CR4_TSD);
254 preempt_enable();
255}
256
257static void enable_TSC(void)
258{
259 preempt_disable();
260 if (test_and_clear_thread_flag(TIF_NOTSC))
261 /*
262 * Must flip the CPU state synchronously with
263 * TIF_NOTSC in the current running context.
264 */
265 cr4_clear_bits(X86_CR4_TSD);
266 preempt_enable();
267}
268
269int get_tsc_mode(unsigned long adr)
270{
271 unsigned int val;
272
273 if (test_thread_flag(TIF_NOTSC))
274 val = PR_TSC_SIGSEGV;
275 else
276 val = PR_TSC_ENABLE;
277
278 return put_user(val, (unsigned int __user *)adr);
279}
280
281int set_tsc_mode(unsigned int val)
282{
283 if (val == PR_TSC_SIGSEGV)
284 disable_TSC();
285 else if (val == PR_TSC_ENABLE)
286 enable_TSC();
287 else
288 return -EINVAL;
289
290 return 0;
291}
292
293DEFINE_PER_CPU(u64, msr_misc_features_shadow);
294
295static void set_cpuid_faulting(bool on)
296{
297 u64 msrval;
298
299 msrval = this_cpu_read(msr_misc_features_shadow);
300 msrval &= ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT;
301 msrval |= (on << MSR_MISC_FEATURES_ENABLES_CPUID_FAULT_BIT);
302 this_cpu_write(msr_misc_features_shadow, msrval);
303 wrmsrl(MSR_MISC_FEATURES_ENABLES, msrval);
304}
305
306static void disable_cpuid(void)
307{
308 preempt_disable();
309 if (!test_and_set_thread_flag(TIF_NOCPUID)) {
310 /*
311 * Must flip the CPU state synchronously with
312 * TIF_NOCPUID in the current running context.
313 */
314 set_cpuid_faulting(true);
315 }
316 preempt_enable();
317}
318
319static void enable_cpuid(void)
320{
321 preempt_disable();
322 if (test_and_clear_thread_flag(TIF_NOCPUID)) {
323 /*
324 * Must flip the CPU state synchronously with
325 * TIF_NOCPUID in the current running context.
326 */
327 set_cpuid_faulting(false);
328 }
329 preempt_enable();
330}
331
332static int get_cpuid_mode(void)
333{
334 return !test_thread_flag(TIF_NOCPUID);
335}
336
337static int set_cpuid_mode(unsigned long cpuid_enabled)
338{
339 if (!boot_cpu_has(X86_FEATURE_CPUID_FAULT))
340 return -ENODEV;
341
342 if (cpuid_enabled)
343 enable_cpuid();
344 else
345 disable_cpuid();
346
347 return 0;
348}
349
350/*
351 * Called immediately after a successful exec.
352 */
353void arch_setup_new_exec(void)
354{
355 /* If cpuid was previously disabled for this task, re-enable it. */
356 if (test_thread_flag(TIF_NOCPUID))
357 enable_cpuid();
358
359 /*
360 * Don't inherit TIF_SSBD across exec boundary when
361 * PR_SPEC_DISABLE_NOEXEC is used.
362 */
363 if (test_thread_flag(TIF_SSBD) &&
364 task_spec_ssb_noexec(current)) {
365 clear_thread_flag(TIF_SSBD);
366 task_clear_spec_ssb_disable(current);
367 task_clear_spec_ssb_noexec(current);
368 speculation_ctrl_update(read_thread_flags());
369 }
370}
371
372#ifdef CONFIG_X86_IOPL_IOPERM
373static inline void switch_to_bitmap(unsigned long tifp)
374{
375 /*
376 * Invalidate I/O bitmap if the previous task used it. This prevents
377 * any possible leakage of an active I/O bitmap.
378 *
379 * If the next task has an I/O bitmap it will handle it on exit to
380 * user mode.
381 */
382 if (tifp & _TIF_IO_BITMAP)
383 tss_invalidate_io_bitmap();
384}
385
386static void tss_copy_io_bitmap(struct tss_struct *tss, struct io_bitmap *iobm)
387{
388 /*
389 * Copy at least the byte range of the incoming tasks bitmap which
390 * covers the permitted I/O ports.
391 *
392 * If the previous task which used an I/O bitmap had more bits
393 * permitted, then the copy needs to cover those as well so they
394 * get turned off.
395 */
396 memcpy(tss->io_bitmap.bitmap, iobm->bitmap,
397 max(tss->io_bitmap.prev_max, iobm->max));
398
399 /*
400 * Store the new max and the sequence number of this bitmap
401 * and a pointer to the bitmap itself.
402 */
403 tss->io_bitmap.prev_max = iobm->max;
404 tss->io_bitmap.prev_sequence = iobm->sequence;
405}
406
407/**
408 * native_tss_update_io_bitmap - Update I/O bitmap before exiting to user mode
409 */
410void native_tss_update_io_bitmap(void)
411{
412 struct tss_struct *tss = this_cpu_ptr(&cpu_tss_rw);
413 struct thread_struct *t = ¤t->thread;
414 u16 *base = &tss->x86_tss.io_bitmap_base;
415
416 if (!test_thread_flag(TIF_IO_BITMAP)) {
417 native_tss_invalidate_io_bitmap();
418 return;
419 }
420
421 if (IS_ENABLED(CONFIG_X86_IOPL_IOPERM) && t->iopl_emul == 3) {
422 *base = IO_BITMAP_OFFSET_VALID_ALL;
423 } else {
424 struct io_bitmap *iobm = t->io_bitmap;
425
426 /*
427 * Only copy bitmap data when the sequence number differs. The
428 * update time is accounted to the incoming task.
429 */
430 if (tss->io_bitmap.prev_sequence != iobm->sequence)
431 tss_copy_io_bitmap(tss, iobm);
432
433 /* Enable the bitmap */
434 *base = IO_BITMAP_OFFSET_VALID_MAP;
435 }
436
437 /*
438 * Make sure that the TSS limit is covering the IO bitmap. It might have
439 * been cut down by a VMEXIT to 0x67 which would cause a subsequent I/O
440 * access from user space to trigger a #GP because tbe bitmap is outside
441 * the TSS limit.
442 */
443 refresh_tss_limit();
444}
445#else /* CONFIG_X86_IOPL_IOPERM */
446static inline void switch_to_bitmap(unsigned long tifp) { }
447#endif
448
449#ifdef CONFIG_SMP
450
451struct ssb_state {
452 struct ssb_state *shared_state;
453 raw_spinlock_t lock;
454 unsigned int disable_state;
455 unsigned long local_state;
456};
457
458#define LSTATE_SSB 0
459
460static DEFINE_PER_CPU(struct ssb_state, ssb_state);
461
462void speculative_store_bypass_ht_init(void)
463{
464 struct ssb_state *st = this_cpu_ptr(&ssb_state);
465 unsigned int this_cpu = smp_processor_id();
466 unsigned int cpu;
467
468 st->local_state = 0;
469
470 /*
471 * Shared state setup happens once on the first bringup
472 * of the CPU. It's not destroyed on CPU hotunplug.
473 */
474 if (st->shared_state)
475 return;
476
477 raw_spin_lock_init(&st->lock);
478
479 /*
480 * Go over HT siblings and check whether one of them has set up the
481 * shared state pointer already.
482 */
483 for_each_cpu(cpu, topology_sibling_cpumask(this_cpu)) {
484 if (cpu == this_cpu)
485 continue;
486
487 if (!per_cpu(ssb_state, cpu).shared_state)
488 continue;
489
490 /* Link it to the state of the sibling: */
491 st->shared_state = per_cpu(ssb_state, cpu).shared_state;
492 return;
493 }
494
495 /*
496 * First HT sibling to come up on the core. Link shared state of
497 * the first HT sibling to itself. The siblings on the same core
498 * which come up later will see the shared state pointer and link
499 * themselves to the state of this CPU.
500 */
501 st->shared_state = st;
502}
503
504/*
505 * Logic is: First HT sibling enables SSBD for both siblings in the core
506 * and last sibling to disable it, disables it for the whole core. This how
507 * MSR_SPEC_CTRL works in "hardware":
508 *
509 * CORE_SPEC_CTRL = THREAD0_SPEC_CTRL | THREAD1_SPEC_CTRL
510 */
511static __always_inline void amd_set_core_ssb_state(unsigned long tifn)
512{
513 struct ssb_state *st = this_cpu_ptr(&ssb_state);
514 u64 msr = x86_amd_ls_cfg_base;
515
516 if (!static_cpu_has(X86_FEATURE_ZEN)) {
517 msr |= ssbd_tif_to_amd_ls_cfg(tifn);
518 wrmsrl(MSR_AMD64_LS_CFG, msr);
519 return;
520 }
521
522 if (tifn & _TIF_SSBD) {
523 /*
524 * Since this can race with prctl(), block reentry on the
525 * same CPU.
526 */
527 if (__test_and_set_bit(LSTATE_SSB, &st->local_state))
528 return;
529
530 msr |= x86_amd_ls_cfg_ssbd_mask;
531
532 raw_spin_lock(&st->shared_state->lock);
533 /* First sibling enables SSBD: */
534 if (!st->shared_state->disable_state)
535 wrmsrl(MSR_AMD64_LS_CFG, msr);
536 st->shared_state->disable_state++;
537 raw_spin_unlock(&st->shared_state->lock);
538 } else {
539 if (!__test_and_clear_bit(LSTATE_SSB, &st->local_state))
540 return;
541
542 raw_spin_lock(&st->shared_state->lock);
543 st->shared_state->disable_state--;
544 if (!st->shared_state->disable_state)
545 wrmsrl(MSR_AMD64_LS_CFG, msr);
546 raw_spin_unlock(&st->shared_state->lock);
547 }
548}
549#else
550static __always_inline void amd_set_core_ssb_state(unsigned long tifn)
551{
552 u64 msr = x86_amd_ls_cfg_base | ssbd_tif_to_amd_ls_cfg(tifn);
553
554 wrmsrl(MSR_AMD64_LS_CFG, msr);
555}
556#endif
557
558static __always_inline void amd_set_ssb_virt_state(unsigned long tifn)
559{
560 /*
561 * SSBD has the same definition in SPEC_CTRL and VIRT_SPEC_CTRL,
562 * so ssbd_tif_to_spec_ctrl() just works.
563 */
564 wrmsrl(MSR_AMD64_VIRT_SPEC_CTRL, ssbd_tif_to_spec_ctrl(tifn));
565}
566
567/*
568 * Update the MSRs managing speculation control, during context switch.
569 *
570 * tifp: Previous task's thread flags
571 * tifn: Next task's thread flags
572 */
573static __always_inline void __speculation_ctrl_update(unsigned long tifp,
574 unsigned long tifn)
575{
576 unsigned long tif_diff = tifp ^ tifn;
577 u64 msr = x86_spec_ctrl_base;
578 bool updmsr = false;
579
580 lockdep_assert_irqs_disabled();
581
582 /* Handle change of TIF_SSBD depending on the mitigation method. */
583 if (static_cpu_has(X86_FEATURE_VIRT_SSBD)) {
584 if (tif_diff & _TIF_SSBD)
585 amd_set_ssb_virt_state(tifn);
586 } else if (static_cpu_has(X86_FEATURE_LS_CFG_SSBD)) {
587 if (tif_diff & _TIF_SSBD)
588 amd_set_core_ssb_state(tifn);
589 } else if (static_cpu_has(X86_FEATURE_SPEC_CTRL_SSBD) ||
590 static_cpu_has(X86_FEATURE_AMD_SSBD)) {
591 updmsr |= !!(tif_diff & _TIF_SSBD);
592 msr |= ssbd_tif_to_spec_ctrl(tifn);
593 }
594
595 /* Only evaluate TIF_SPEC_IB if conditional STIBP is enabled. */
596 if (IS_ENABLED(CONFIG_SMP) &&
597 static_branch_unlikely(&switch_to_cond_stibp)) {
598 updmsr |= !!(tif_diff & _TIF_SPEC_IB);
599 msr |= stibp_tif_to_spec_ctrl(tifn);
600 }
601
602 if (updmsr)
603 update_spec_ctrl_cond(msr);
604}
605
606static unsigned long speculation_ctrl_update_tif(struct task_struct *tsk)
607{
608 if (test_and_clear_tsk_thread_flag(tsk, TIF_SPEC_FORCE_UPDATE)) {
609 if (task_spec_ssb_disable(tsk))
610 set_tsk_thread_flag(tsk, TIF_SSBD);
611 else
612 clear_tsk_thread_flag(tsk, TIF_SSBD);
613
614 if (task_spec_ib_disable(tsk))
615 set_tsk_thread_flag(tsk, TIF_SPEC_IB);
616 else
617 clear_tsk_thread_flag(tsk, TIF_SPEC_IB);
618 }
619 /* Return the updated threadinfo flags*/
620 return read_task_thread_flags(tsk);
621}
622
623void speculation_ctrl_update(unsigned long tif)
624{
625 unsigned long flags;
626
627 /* Forced update. Make sure all relevant TIF flags are different */
628 local_irq_save(flags);
629 __speculation_ctrl_update(~tif, tif);
630 local_irq_restore(flags);
631}
632
633/* Called from seccomp/prctl update */
634void speculation_ctrl_update_current(void)
635{
636 preempt_disable();
637 speculation_ctrl_update(speculation_ctrl_update_tif(current));
638 preempt_enable();
639}
640
641static inline void cr4_toggle_bits_irqsoff(unsigned long mask)
642{
643 unsigned long newval, cr4 = this_cpu_read(cpu_tlbstate.cr4);
644
645 newval = cr4 ^ mask;
646 if (newval != cr4) {
647 this_cpu_write(cpu_tlbstate.cr4, newval);
648 __write_cr4(newval);
649 }
650}
651
652void __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p)
653{
654 unsigned long tifp, tifn;
655
656 tifn = read_task_thread_flags(next_p);
657 tifp = read_task_thread_flags(prev_p);
658
659 switch_to_bitmap(tifp);
660
661 propagate_user_return_notify(prev_p, next_p);
662
663 if ((tifp & _TIF_BLOCKSTEP || tifn & _TIF_BLOCKSTEP) &&
664 arch_has_block_step()) {
665 unsigned long debugctl, msk;
666
667 rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctl);
668 debugctl &= ~DEBUGCTLMSR_BTF;
669 msk = tifn & _TIF_BLOCKSTEP;
670 debugctl |= (msk >> TIF_BLOCKSTEP) << DEBUGCTLMSR_BTF_SHIFT;
671 wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctl);
672 }
673
674 if ((tifp ^ tifn) & _TIF_NOTSC)
675 cr4_toggle_bits_irqsoff(X86_CR4_TSD);
676
677 if ((tifp ^ tifn) & _TIF_NOCPUID)
678 set_cpuid_faulting(!!(tifn & _TIF_NOCPUID));
679
680 if (likely(!((tifp | tifn) & _TIF_SPEC_FORCE_UPDATE))) {
681 __speculation_ctrl_update(tifp, tifn);
682 } else {
683 speculation_ctrl_update_tif(prev_p);
684 tifn = speculation_ctrl_update_tif(next_p);
685
686 /* Enforce MSR update to ensure consistent state */
687 __speculation_ctrl_update(~tifn, tifn);
688 }
689}
690
691/*
692 * Idle related variables and functions
693 */
694unsigned long boot_option_idle_override = IDLE_NO_OVERRIDE;
695EXPORT_SYMBOL(boot_option_idle_override);
696
697static void (*x86_idle)(void);
698
699#ifndef CONFIG_SMP
700static inline void play_dead(void)
701{
702 BUG();
703}
704#endif
705
706void arch_cpu_idle_enter(void)
707{
708 tsc_verify_tsc_adjust(false);
709 local_touch_nmi();
710}
711
712void arch_cpu_idle_dead(void)
713{
714 play_dead();
715}
716
717/*
718 * Called from the generic idle code.
719 */
720void arch_cpu_idle(void)
721{
722 x86_idle();
723}
724
725/*
726 * We use this if we don't have any better idle routine..
727 */
728void __cpuidle default_idle(void)
729{
730 raw_safe_halt();
731}
732#if defined(CONFIG_APM_MODULE) || defined(CONFIG_HALTPOLL_CPUIDLE_MODULE)
733EXPORT_SYMBOL(default_idle);
734#endif
735
736#ifdef CONFIG_XEN
737bool xen_set_default_idle(void)
738{
739 bool ret = !!x86_idle;
740
741 x86_idle = default_idle;
742
743 return ret;
744}
745#endif
746
747void __noreturn stop_this_cpu(void *dummy)
748{
749 local_irq_disable();
750 /*
751 * Remove this CPU:
752 */
753 set_cpu_online(smp_processor_id(), false);
754 disable_local_APIC();
755 mcheck_cpu_clear(this_cpu_ptr(&cpu_info));
756
757 /*
758 * Use wbinvd on processors that support SME. This provides support
759 * for performing a successful kexec when going from SME inactive
760 * to SME active (or vice-versa). The cache must be cleared so that
761 * if there are entries with the same physical address, both with and
762 * without the encryption bit, they don't race each other when flushed
763 * and potentially end up with the wrong entry being committed to
764 * memory.
765 *
766 * Test the CPUID bit directly because the machine might've cleared
767 * X86_FEATURE_SME due to cmdline options.
768 */
769 if (cpuid_eax(0x8000001f) & BIT(0))
770 native_wbinvd();
771 for (;;) {
772 /*
773 * Use native_halt() so that memory contents don't change
774 * (stack usage and variables) after possibly issuing the
775 * native_wbinvd() above.
776 */
777 native_halt();
778 }
779}
780
781/*
782 * AMD Erratum 400 aware idle routine. We handle it the same way as C3 power
783 * states (local apic timer and TSC stop).
784 *
785 * XXX this function is completely buggered vs RCU and tracing.
786 */
787static void amd_e400_idle(void)
788{
789 /*
790 * We cannot use static_cpu_has_bug() here because X86_BUG_AMD_APIC_C1E
791 * gets set after static_cpu_has() places have been converted via
792 * alternatives.
793 */
794 if (!boot_cpu_has_bug(X86_BUG_AMD_APIC_C1E)) {
795 default_idle();
796 return;
797 }
798
799 tick_broadcast_enter();
800
801 default_idle();
802
803 /*
804 * The switch back from broadcast mode needs to be called with
805 * interrupts disabled.
806 */
807 raw_local_irq_disable();
808 tick_broadcast_exit();
809 raw_local_irq_enable();
810}
811
812/*
813 * Prefer MWAIT over HALT if MWAIT is supported, MWAIT_CPUID leaf
814 * exists and whenever MONITOR/MWAIT extensions are present there is at
815 * least one C1 substate.
816 *
817 * Do not prefer MWAIT if MONITOR instruction has a bug or idle=nomwait
818 * is passed to kernel commandline parameter.
819 */
820static int prefer_mwait_c1_over_halt(const struct cpuinfo_x86 *c)
821{
822 u32 eax, ebx, ecx, edx;
823
824 /* User has disallowed the use of MWAIT. Fallback to HALT */
825 if (boot_option_idle_override == IDLE_NOMWAIT)
826 return 0;
827
828 /* MWAIT is not supported on this platform. Fallback to HALT */
829 if (!cpu_has(c, X86_FEATURE_MWAIT))
830 return 0;
831
832 /* Monitor has a bug. Fallback to HALT */
833 if (boot_cpu_has_bug(X86_BUG_MONITOR))
834 return 0;
835
836 cpuid(CPUID_MWAIT_LEAF, &eax, &ebx, &ecx, &edx);
837
838 /*
839 * If MWAIT extensions are not available, it is safe to use MWAIT
840 * with EAX=0, ECX=0.
841 */
842 if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED))
843 return 1;
844
845 /*
846 * If MWAIT extensions are available, there should be at least one
847 * MWAIT C1 substate present.
848 */
849 return (edx & MWAIT_C1_SUBSTATE_MASK);
850}
851
852/*
853 * MONITOR/MWAIT with no hints, used for default C1 state. This invokes MWAIT
854 * with interrupts enabled and no flags, which is backwards compatible with the
855 * original MWAIT implementation.
856 */
857static __cpuidle void mwait_idle(void)
858{
859 if (!current_set_polling_and_test()) {
860 if (this_cpu_has(X86_BUG_CLFLUSH_MONITOR)) {
861 mb(); /* quirk */
862 clflush((void *)¤t_thread_info()->flags);
863 mb(); /* quirk */
864 }
865
866 __monitor((void *)¤t_thread_info()->flags, 0, 0);
867 if (!need_resched())
868 __sti_mwait(0, 0);
869 else
870 raw_local_irq_enable();
871 } else {
872 raw_local_irq_enable();
873 }
874 __current_clr_polling();
875}
876
877void select_idle_routine(const struct cpuinfo_x86 *c)
878{
879#ifdef CONFIG_SMP
880 if (boot_option_idle_override == IDLE_POLL && smp_num_siblings > 1)
881 pr_warn_once("WARNING: polling idle and HT enabled, performance may degrade\n");
882#endif
883 if (x86_idle || boot_option_idle_override == IDLE_POLL)
884 return;
885
886 if (boot_cpu_has_bug(X86_BUG_AMD_E400)) {
887 pr_info("using AMD E400 aware idle routine\n");
888 x86_idle = amd_e400_idle;
889 } else if (prefer_mwait_c1_over_halt(c)) {
890 pr_info("using mwait in idle threads\n");
891 x86_idle = mwait_idle;
892 } else if (cpu_feature_enabled(X86_FEATURE_TDX_GUEST)) {
893 pr_info("using TDX aware idle routine\n");
894 x86_idle = tdx_safe_halt;
895 } else
896 x86_idle = default_idle;
897}
898
899void amd_e400_c1e_apic_setup(void)
900{
901 if (boot_cpu_has_bug(X86_BUG_AMD_APIC_C1E)) {
902 pr_info("Switch to broadcast mode on CPU%d\n", smp_processor_id());
903 local_irq_disable();
904 tick_broadcast_force();
905 local_irq_enable();
906 }
907}
908
909void __init arch_post_acpi_subsys_init(void)
910{
911 u32 lo, hi;
912
913 if (!boot_cpu_has_bug(X86_BUG_AMD_E400))
914 return;
915
916 /*
917 * AMD E400 detection needs to happen after ACPI has been enabled. If
918 * the machine is affected K8_INTP_C1E_ACTIVE_MASK bits are set in
919 * MSR_K8_INT_PENDING_MSG.
920 */
921 rdmsr(MSR_K8_INT_PENDING_MSG, lo, hi);
922 if (!(lo & K8_INTP_C1E_ACTIVE_MASK))
923 return;
924
925 boot_cpu_set_bug(X86_BUG_AMD_APIC_C1E);
926
927 if (!boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
928 mark_tsc_unstable("TSC halt in AMD C1E");
929 pr_info("System has AMD C1E enabled\n");
930}
931
932static int __init idle_setup(char *str)
933{
934 if (!str)
935 return -EINVAL;
936
937 if (!strcmp(str, "poll")) {
938 pr_info("using polling idle threads\n");
939 boot_option_idle_override = IDLE_POLL;
940 cpu_idle_poll_ctrl(true);
941 } else if (!strcmp(str, "halt")) {
942 /*
943 * When the boot option of idle=halt is added, halt is
944 * forced to be used for CPU idle. In such case CPU C2/C3
945 * won't be used again.
946 * To continue to load the CPU idle driver, don't touch
947 * the boot_option_idle_override.
948 */
949 x86_idle = default_idle;
950 boot_option_idle_override = IDLE_HALT;
951 } else if (!strcmp(str, "nomwait")) {
952 /*
953 * If the boot option of "idle=nomwait" is added,
954 * it means that mwait will be disabled for CPU C1/C2/C3
955 * states.
956 */
957 boot_option_idle_override = IDLE_NOMWAIT;
958 } else
959 return -1;
960
961 return 0;
962}
963early_param("idle", idle_setup);
964
965unsigned long arch_align_stack(unsigned long sp)
966{
967 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
968 sp -= get_random_u32_below(8192);
969 return sp & ~0xf;
970}
971
972unsigned long arch_randomize_brk(struct mm_struct *mm)
973{
974 return randomize_page(mm->brk, 0x02000000);
975}
976
977/*
978 * Called from fs/proc with a reference on @p to find the function
979 * which called into schedule(). This needs to be done carefully
980 * because the task might wake up and we might look at a stack
981 * changing under us.
982 */
983unsigned long __get_wchan(struct task_struct *p)
984{
985 struct unwind_state state;
986 unsigned long addr = 0;
987
988 if (!try_get_task_stack(p))
989 return 0;
990
991 for (unwind_start(&state, p, NULL, NULL); !unwind_done(&state);
992 unwind_next_frame(&state)) {
993 addr = unwind_get_return_address(&state);
994 if (!addr)
995 break;
996 if (in_sched_functions(addr))
997 continue;
998 break;
999 }
1000
1001 put_task_stack(p);
1002
1003 return addr;
1004}
1005
1006long do_arch_prctl_common(int option, unsigned long arg2)
1007{
1008 switch (option) {
1009 case ARCH_GET_CPUID:
1010 return get_cpuid_mode();
1011 case ARCH_SET_CPUID:
1012 return set_cpuid_mode(arg2);
1013 case ARCH_GET_XCOMP_SUPP:
1014 case ARCH_GET_XCOMP_PERM:
1015 case ARCH_REQ_XCOMP_PERM:
1016 case ARCH_GET_XCOMP_GUEST_PERM:
1017 case ARCH_REQ_XCOMP_GUEST_PERM:
1018 return fpu_xstate_prctl(option, arg2);
1019 }
1020
1021 return -EINVAL;
1022}
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};
75EXPORT_PER_CPU_SYMBOL(cpu_tss_rw);
76
77DEFINE_PER_CPU(bool, __tss_limit_invalid);
78EXPORT_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 */
84int 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 return fpu_clone(dst);
91}
92
93/*
94 * Free thread data structures etc..
95 */
96void exit_thread(struct task_struct *tsk)
97{
98 struct thread_struct *t = &tsk->thread;
99 struct fpu *fpu = &t->fpu;
100
101 if (test_thread_flag(TIF_IO_BITMAP))
102 io_bitmap_exit(tsk);
103
104 free_vm86(t);
105
106 fpu__drop(fpu);
107}
108
109static int set_new_tls(struct task_struct *p, unsigned long tls)
110{
111 struct user_desc __user *utls = (struct user_desc __user *)tls;
112
113 if (in_ia32_syscall())
114 return do_set_thread_area(p, -1, utls, 0);
115 else
116 return do_set_thread_area_64(p, ARCH_SET_FS, tls);
117}
118
119int copy_thread(unsigned long clone_flags, unsigned long sp, unsigned long arg,
120 struct task_struct *p, unsigned long tls)
121{
122 struct inactive_task_frame *frame;
123 struct fork_frame *fork_frame;
124 struct pt_regs *childregs;
125 int ret = 0;
126
127 childregs = task_pt_regs(p);
128 fork_frame = container_of(childregs, struct fork_frame, regs);
129 frame = &fork_frame->frame;
130
131 frame->bp = encode_frame_pointer(childregs);
132 frame->ret_addr = (unsigned long) ret_from_fork;
133 p->thread.sp = (unsigned long) fork_frame;
134 p->thread.io_bitmap = NULL;
135 memset(p->thread.ptrace_bps, 0, sizeof(p->thread.ptrace_bps));
136
137#ifdef CONFIG_X86_64
138 current_save_fsgs();
139 p->thread.fsindex = current->thread.fsindex;
140 p->thread.fsbase = current->thread.fsbase;
141 p->thread.gsindex = current->thread.gsindex;
142 p->thread.gsbase = current->thread.gsbase;
143
144 savesegment(es, p->thread.es);
145 savesegment(ds, p->thread.ds);
146#else
147 p->thread.sp0 = (unsigned long) (childregs + 1);
148 /*
149 * Clear all status flags including IF and set fixed bit. 64bit
150 * does not have this initialization as the frame does not contain
151 * flags. The flags consistency (especially vs. AC) is there
152 * ensured via objtool, which lacks 32bit support.
153 */
154 frame->flags = X86_EFLAGS_FIXED;
155#endif
156
157 /* Kernel thread ? */
158 if (unlikely(p->flags & PF_KTHREAD)) {
159 p->thread.pkru = pkru_get_init_value();
160 memset(childregs, 0, sizeof(struct pt_regs));
161 kthread_frame_init(frame, sp, arg);
162 return 0;
163 }
164
165 /*
166 * Clone current's PKRU value from hardware. tsk->thread.pkru
167 * is only valid when scheduled out.
168 */
169 p->thread.pkru = read_pkru();
170
171 frame->bx = 0;
172 *childregs = *current_pt_regs();
173 childregs->ax = 0;
174 if (sp)
175 childregs->sp = sp;
176
177#ifdef CONFIG_X86_32
178 task_user_gs(p) = get_user_gs(current_pt_regs());
179#endif
180
181 if (unlikely(p->flags & PF_IO_WORKER)) {
182 /*
183 * An IO thread is a user space thread, but it doesn't
184 * return to ret_after_fork().
185 *
186 * In order to indicate that to tools like gdb,
187 * we reset the stack and instruction pointers.
188 *
189 * It does the same kernel frame setup to return to a kernel
190 * function that a kernel thread does.
191 */
192 childregs->sp = 0;
193 childregs->ip = 0;
194 kthread_frame_init(frame, sp, arg);
195 return 0;
196 }
197
198 /* Set a new TLS for the child thread? */
199 if (clone_flags & CLONE_SETTLS)
200 ret = set_new_tls(p, tls);
201
202 if (!ret && unlikely(test_tsk_thread_flag(current, TIF_IO_BITMAP)))
203 io_bitmap_share(p);
204
205 return ret;
206}
207
208static void pkru_flush_thread(void)
209{
210 /*
211 * If PKRU is enabled the default PKRU value has to be loaded into
212 * the hardware right here (similar to context switch).
213 */
214 pkru_write_default();
215}
216
217void flush_thread(void)
218{
219 struct task_struct *tsk = current;
220
221 flush_ptrace_hw_breakpoint(tsk);
222 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
223
224 fpu_flush_thread();
225 pkru_flush_thread();
226}
227
228void disable_TSC(void)
229{
230 preempt_disable();
231 if (!test_and_set_thread_flag(TIF_NOTSC))
232 /*
233 * Must flip the CPU state synchronously with
234 * TIF_NOTSC in the current running context.
235 */
236 cr4_set_bits(X86_CR4_TSD);
237 preempt_enable();
238}
239
240static void enable_TSC(void)
241{
242 preempt_disable();
243 if (test_and_clear_thread_flag(TIF_NOTSC))
244 /*
245 * Must flip the CPU state synchronously with
246 * TIF_NOTSC in the current running context.
247 */
248 cr4_clear_bits(X86_CR4_TSD);
249 preempt_enable();
250}
251
252int get_tsc_mode(unsigned long adr)
253{
254 unsigned int val;
255
256 if (test_thread_flag(TIF_NOTSC))
257 val = PR_TSC_SIGSEGV;
258 else
259 val = PR_TSC_ENABLE;
260
261 return put_user(val, (unsigned int __user *)adr);
262}
263
264int set_tsc_mode(unsigned int val)
265{
266 if (val == PR_TSC_SIGSEGV)
267 disable_TSC();
268 else if (val == PR_TSC_ENABLE)
269 enable_TSC();
270 else
271 return -EINVAL;
272
273 return 0;
274}
275
276DEFINE_PER_CPU(u64, msr_misc_features_shadow);
277
278static void set_cpuid_faulting(bool on)
279{
280 u64 msrval;
281
282 msrval = this_cpu_read(msr_misc_features_shadow);
283 msrval &= ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT;
284 msrval |= (on << MSR_MISC_FEATURES_ENABLES_CPUID_FAULT_BIT);
285 this_cpu_write(msr_misc_features_shadow, msrval);
286 wrmsrl(MSR_MISC_FEATURES_ENABLES, msrval);
287}
288
289static void disable_cpuid(void)
290{
291 preempt_disable();
292 if (!test_and_set_thread_flag(TIF_NOCPUID)) {
293 /*
294 * Must flip the CPU state synchronously with
295 * TIF_NOCPUID in the current running context.
296 */
297 set_cpuid_faulting(true);
298 }
299 preempt_enable();
300}
301
302static void enable_cpuid(void)
303{
304 preempt_disable();
305 if (test_and_clear_thread_flag(TIF_NOCPUID)) {
306 /*
307 * Must flip the CPU state synchronously with
308 * TIF_NOCPUID in the current running context.
309 */
310 set_cpuid_faulting(false);
311 }
312 preempt_enable();
313}
314
315static int get_cpuid_mode(void)
316{
317 return !test_thread_flag(TIF_NOCPUID);
318}
319
320static int set_cpuid_mode(struct task_struct *task, unsigned long cpuid_enabled)
321{
322 if (!boot_cpu_has(X86_FEATURE_CPUID_FAULT))
323 return -ENODEV;
324
325 if (cpuid_enabled)
326 enable_cpuid();
327 else
328 disable_cpuid();
329
330 return 0;
331}
332
333/*
334 * Called immediately after a successful exec.
335 */
336void arch_setup_new_exec(void)
337{
338 /* If cpuid was previously disabled for this task, re-enable it. */
339 if (test_thread_flag(TIF_NOCPUID))
340 enable_cpuid();
341
342 /*
343 * Don't inherit TIF_SSBD across exec boundary when
344 * PR_SPEC_DISABLE_NOEXEC is used.
345 */
346 if (test_thread_flag(TIF_SSBD) &&
347 task_spec_ssb_noexec(current)) {
348 clear_thread_flag(TIF_SSBD);
349 task_clear_spec_ssb_disable(current);
350 task_clear_spec_ssb_noexec(current);
351 speculation_ctrl_update(task_thread_info(current)->flags);
352 }
353}
354
355#ifdef CONFIG_X86_IOPL_IOPERM
356static inline void switch_to_bitmap(unsigned long tifp)
357{
358 /*
359 * Invalidate I/O bitmap if the previous task used it. This prevents
360 * any possible leakage of an active I/O bitmap.
361 *
362 * If the next task has an I/O bitmap it will handle it on exit to
363 * user mode.
364 */
365 if (tifp & _TIF_IO_BITMAP)
366 tss_invalidate_io_bitmap();
367}
368
369static void tss_copy_io_bitmap(struct tss_struct *tss, struct io_bitmap *iobm)
370{
371 /*
372 * Copy at least the byte range of the incoming tasks bitmap which
373 * covers the permitted I/O ports.
374 *
375 * If the previous task which used an I/O bitmap had more bits
376 * permitted, then the copy needs to cover those as well so they
377 * get turned off.
378 */
379 memcpy(tss->io_bitmap.bitmap, iobm->bitmap,
380 max(tss->io_bitmap.prev_max, iobm->max));
381
382 /*
383 * Store the new max and the sequence number of this bitmap
384 * and a pointer to the bitmap itself.
385 */
386 tss->io_bitmap.prev_max = iobm->max;
387 tss->io_bitmap.prev_sequence = iobm->sequence;
388}
389
390/**
391 * tss_update_io_bitmap - Update I/O bitmap before exiting to usermode
392 */
393void native_tss_update_io_bitmap(void)
394{
395 struct tss_struct *tss = this_cpu_ptr(&cpu_tss_rw);
396 struct thread_struct *t = ¤t->thread;
397 u16 *base = &tss->x86_tss.io_bitmap_base;
398
399 if (!test_thread_flag(TIF_IO_BITMAP)) {
400 native_tss_invalidate_io_bitmap();
401 return;
402 }
403
404 if (IS_ENABLED(CONFIG_X86_IOPL_IOPERM) && t->iopl_emul == 3) {
405 *base = IO_BITMAP_OFFSET_VALID_ALL;
406 } else {
407 struct io_bitmap *iobm = t->io_bitmap;
408
409 /*
410 * Only copy bitmap data when the sequence number differs. The
411 * update time is accounted to the incoming task.
412 */
413 if (tss->io_bitmap.prev_sequence != iobm->sequence)
414 tss_copy_io_bitmap(tss, iobm);
415
416 /* Enable the bitmap */
417 *base = IO_BITMAP_OFFSET_VALID_MAP;
418 }
419
420 /*
421 * Make sure that the TSS limit is covering the IO bitmap. It might have
422 * been cut down by a VMEXIT to 0x67 which would cause a subsequent I/O
423 * access from user space to trigger a #GP because tbe bitmap is outside
424 * the TSS limit.
425 */
426 refresh_tss_limit();
427}
428#else /* CONFIG_X86_IOPL_IOPERM */
429static inline void switch_to_bitmap(unsigned long tifp) { }
430#endif
431
432#ifdef CONFIG_SMP
433
434struct ssb_state {
435 struct ssb_state *shared_state;
436 raw_spinlock_t lock;
437 unsigned int disable_state;
438 unsigned long local_state;
439};
440
441#define LSTATE_SSB 0
442
443static DEFINE_PER_CPU(struct ssb_state, ssb_state);
444
445void speculative_store_bypass_ht_init(void)
446{
447 struct ssb_state *st = this_cpu_ptr(&ssb_state);
448 unsigned int this_cpu = smp_processor_id();
449 unsigned int cpu;
450
451 st->local_state = 0;
452
453 /*
454 * Shared state setup happens once on the first bringup
455 * of the CPU. It's not destroyed on CPU hotunplug.
456 */
457 if (st->shared_state)
458 return;
459
460 raw_spin_lock_init(&st->lock);
461
462 /*
463 * Go over HT siblings and check whether one of them has set up the
464 * shared state pointer already.
465 */
466 for_each_cpu(cpu, topology_sibling_cpumask(this_cpu)) {
467 if (cpu == this_cpu)
468 continue;
469
470 if (!per_cpu(ssb_state, cpu).shared_state)
471 continue;
472
473 /* Link it to the state of the sibling: */
474 st->shared_state = per_cpu(ssb_state, cpu).shared_state;
475 return;
476 }
477
478 /*
479 * First HT sibling to come up on the core. Link shared state of
480 * the first HT sibling to itself. The siblings on the same core
481 * which come up later will see the shared state pointer and link
482 * themselves to the state of this CPU.
483 */
484 st->shared_state = st;
485}
486
487/*
488 * Logic is: First HT sibling enables SSBD for both siblings in the core
489 * and last sibling to disable it, disables it for the whole core. This how
490 * MSR_SPEC_CTRL works in "hardware":
491 *
492 * CORE_SPEC_CTRL = THREAD0_SPEC_CTRL | THREAD1_SPEC_CTRL
493 */
494static __always_inline void amd_set_core_ssb_state(unsigned long tifn)
495{
496 struct ssb_state *st = this_cpu_ptr(&ssb_state);
497 u64 msr = x86_amd_ls_cfg_base;
498
499 if (!static_cpu_has(X86_FEATURE_ZEN)) {
500 msr |= ssbd_tif_to_amd_ls_cfg(tifn);
501 wrmsrl(MSR_AMD64_LS_CFG, msr);
502 return;
503 }
504
505 if (tifn & _TIF_SSBD) {
506 /*
507 * Since this can race with prctl(), block reentry on the
508 * same CPU.
509 */
510 if (__test_and_set_bit(LSTATE_SSB, &st->local_state))
511 return;
512
513 msr |= x86_amd_ls_cfg_ssbd_mask;
514
515 raw_spin_lock(&st->shared_state->lock);
516 /* First sibling enables SSBD: */
517 if (!st->shared_state->disable_state)
518 wrmsrl(MSR_AMD64_LS_CFG, msr);
519 st->shared_state->disable_state++;
520 raw_spin_unlock(&st->shared_state->lock);
521 } else {
522 if (!__test_and_clear_bit(LSTATE_SSB, &st->local_state))
523 return;
524
525 raw_spin_lock(&st->shared_state->lock);
526 st->shared_state->disable_state--;
527 if (!st->shared_state->disable_state)
528 wrmsrl(MSR_AMD64_LS_CFG, msr);
529 raw_spin_unlock(&st->shared_state->lock);
530 }
531}
532#else
533static __always_inline void amd_set_core_ssb_state(unsigned long tifn)
534{
535 u64 msr = x86_amd_ls_cfg_base | ssbd_tif_to_amd_ls_cfg(tifn);
536
537 wrmsrl(MSR_AMD64_LS_CFG, msr);
538}
539#endif
540
541static __always_inline void amd_set_ssb_virt_state(unsigned long tifn)
542{
543 /*
544 * SSBD has the same definition in SPEC_CTRL and VIRT_SPEC_CTRL,
545 * so ssbd_tif_to_spec_ctrl() just works.
546 */
547 wrmsrl(MSR_AMD64_VIRT_SPEC_CTRL, ssbd_tif_to_spec_ctrl(tifn));
548}
549
550/*
551 * Update the MSRs managing speculation control, during context switch.
552 *
553 * tifp: Previous task's thread flags
554 * tifn: Next task's thread flags
555 */
556static __always_inline void __speculation_ctrl_update(unsigned long tifp,
557 unsigned long tifn)
558{
559 unsigned long tif_diff = tifp ^ tifn;
560 u64 msr = x86_spec_ctrl_base;
561 bool updmsr = false;
562
563 lockdep_assert_irqs_disabled();
564
565 /* Handle change of TIF_SSBD depending on the mitigation method. */
566 if (static_cpu_has(X86_FEATURE_VIRT_SSBD)) {
567 if (tif_diff & _TIF_SSBD)
568 amd_set_ssb_virt_state(tifn);
569 } else if (static_cpu_has(X86_FEATURE_LS_CFG_SSBD)) {
570 if (tif_diff & _TIF_SSBD)
571 amd_set_core_ssb_state(tifn);
572 } else if (static_cpu_has(X86_FEATURE_SPEC_CTRL_SSBD) ||
573 static_cpu_has(X86_FEATURE_AMD_SSBD)) {
574 updmsr |= !!(tif_diff & _TIF_SSBD);
575 msr |= ssbd_tif_to_spec_ctrl(tifn);
576 }
577
578 /* Only evaluate TIF_SPEC_IB if conditional STIBP is enabled. */
579 if (IS_ENABLED(CONFIG_SMP) &&
580 static_branch_unlikely(&switch_to_cond_stibp)) {
581 updmsr |= !!(tif_diff & _TIF_SPEC_IB);
582 msr |= stibp_tif_to_spec_ctrl(tifn);
583 }
584
585 if (updmsr)
586 wrmsrl(MSR_IA32_SPEC_CTRL, msr);
587}
588
589static unsigned long speculation_ctrl_update_tif(struct task_struct *tsk)
590{
591 if (test_and_clear_tsk_thread_flag(tsk, TIF_SPEC_FORCE_UPDATE)) {
592 if (task_spec_ssb_disable(tsk))
593 set_tsk_thread_flag(tsk, TIF_SSBD);
594 else
595 clear_tsk_thread_flag(tsk, TIF_SSBD);
596
597 if (task_spec_ib_disable(tsk))
598 set_tsk_thread_flag(tsk, TIF_SPEC_IB);
599 else
600 clear_tsk_thread_flag(tsk, TIF_SPEC_IB);
601 }
602 /* Return the updated threadinfo flags*/
603 return task_thread_info(tsk)->flags;
604}
605
606void speculation_ctrl_update(unsigned long tif)
607{
608 unsigned long flags;
609
610 /* Forced update. Make sure all relevant TIF flags are different */
611 local_irq_save(flags);
612 __speculation_ctrl_update(~tif, tif);
613 local_irq_restore(flags);
614}
615
616/* Called from seccomp/prctl update */
617void speculation_ctrl_update_current(void)
618{
619 preempt_disable();
620 speculation_ctrl_update(speculation_ctrl_update_tif(current));
621 preempt_enable();
622}
623
624static inline void cr4_toggle_bits_irqsoff(unsigned long mask)
625{
626 unsigned long newval, cr4 = this_cpu_read(cpu_tlbstate.cr4);
627
628 newval = cr4 ^ mask;
629 if (newval != cr4) {
630 this_cpu_write(cpu_tlbstate.cr4, newval);
631 __write_cr4(newval);
632 }
633}
634
635void __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p)
636{
637 unsigned long tifp, tifn;
638
639 tifn = READ_ONCE(task_thread_info(next_p)->flags);
640 tifp = READ_ONCE(task_thread_info(prev_p)->flags);
641
642 switch_to_bitmap(tifp);
643
644 propagate_user_return_notify(prev_p, next_p);
645
646 if ((tifp & _TIF_BLOCKSTEP || tifn & _TIF_BLOCKSTEP) &&
647 arch_has_block_step()) {
648 unsigned long debugctl, msk;
649
650 rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctl);
651 debugctl &= ~DEBUGCTLMSR_BTF;
652 msk = tifn & _TIF_BLOCKSTEP;
653 debugctl |= (msk >> TIF_BLOCKSTEP) << DEBUGCTLMSR_BTF_SHIFT;
654 wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctl);
655 }
656
657 if ((tifp ^ tifn) & _TIF_NOTSC)
658 cr4_toggle_bits_irqsoff(X86_CR4_TSD);
659
660 if ((tifp ^ tifn) & _TIF_NOCPUID)
661 set_cpuid_faulting(!!(tifn & _TIF_NOCPUID));
662
663 if (likely(!((tifp | tifn) & _TIF_SPEC_FORCE_UPDATE))) {
664 __speculation_ctrl_update(tifp, tifn);
665 } else {
666 speculation_ctrl_update_tif(prev_p);
667 tifn = speculation_ctrl_update_tif(next_p);
668
669 /* Enforce MSR update to ensure consistent state */
670 __speculation_ctrl_update(~tifn, tifn);
671 }
672
673 if ((tifp ^ tifn) & _TIF_SLD)
674 switch_to_sld(tifn);
675}
676
677/*
678 * Idle related variables and functions
679 */
680unsigned long boot_option_idle_override = IDLE_NO_OVERRIDE;
681EXPORT_SYMBOL(boot_option_idle_override);
682
683static void (*x86_idle)(void);
684
685#ifndef CONFIG_SMP
686static inline void play_dead(void)
687{
688 BUG();
689}
690#endif
691
692void arch_cpu_idle_enter(void)
693{
694 tsc_verify_tsc_adjust(false);
695 local_touch_nmi();
696}
697
698void arch_cpu_idle_dead(void)
699{
700 play_dead();
701}
702
703/*
704 * Called from the generic idle code.
705 */
706void arch_cpu_idle(void)
707{
708 x86_idle();
709}
710
711/*
712 * We use this if we don't have any better idle routine..
713 */
714void __cpuidle default_idle(void)
715{
716 raw_safe_halt();
717}
718#if defined(CONFIG_APM_MODULE) || defined(CONFIG_HALTPOLL_CPUIDLE_MODULE)
719EXPORT_SYMBOL(default_idle);
720#endif
721
722#ifdef CONFIG_XEN
723bool xen_set_default_idle(void)
724{
725 bool ret = !!x86_idle;
726
727 x86_idle = default_idle;
728
729 return ret;
730}
731#endif
732
733void stop_this_cpu(void *dummy)
734{
735 local_irq_disable();
736 /*
737 * Remove this CPU:
738 */
739 set_cpu_online(smp_processor_id(), false);
740 disable_local_APIC();
741 mcheck_cpu_clear(this_cpu_ptr(&cpu_info));
742
743 /*
744 * Use wbinvd on processors that support SME. This provides support
745 * for performing a successful kexec when going from SME inactive
746 * to SME active (or vice-versa). The cache must be cleared so that
747 * if there are entries with the same physical address, both with and
748 * without the encryption bit, they don't race each other when flushed
749 * and potentially end up with the wrong entry being committed to
750 * memory.
751 */
752 if (boot_cpu_has(X86_FEATURE_SME))
753 native_wbinvd();
754 for (;;) {
755 /*
756 * Use native_halt() so that memory contents don't change
757 * (stack usage and variables) after possibly issuing the
758 * native_wbinvd() above.
759 */
760 native_halt();
761 }
762}
763
764/*
765 * AMD Erratum 400 aware idle routine. We handle it the same way as C3 power
766 * states (local apic timer and TSC stop).
767 *
768 * XXX this function is completely buggered vs RCU and tracing.
769 */
770static void amd_e400_idle(void)
771{
772 /*
773 * We cannot use static_cpu_has_bug() here because X86_BUG_AMD_APIC_C1E
774 * gets set after static_cpu_has() places have been converted via
775 * alternatives.
776 */
777 if (!boot_cpu_has_bug(X86_BUG_AMD_APIC_C1E)) {
778 default_idle();
779 return;
780 }
781
782 tick_broadcast_enter();
783
784 default_idle();
785
786 /*
787 * The switch back from broadcast mode needs to be called with
788 * interrupts disabled.
789 */
790 raw_local_irq_disable();
791 tick_broadcast_exit();
792 raw_local_irq_enable();
793}
794
795/*
796 * Intel Core2 and older machines prefer MWAIT over HALT for C1.
797 * We can't rely on cpuidle installing MWAIT, because it will not load
798 * on systems that support only C1 -- so the boot default must be MWAIT.
799 *
800 * Some AMD machines are the opposite, they depend on using HALT.
801 *
802 * So for default C1, which is used during boot until cpuidle loads,
803 * use MWAIT-C1 on Intel HW that has it, else use HALT.
804 */
805static int prefer_mwait_c1_over_halt(const struct cpuinfo_x86 *c)
806{
807 if (c->x86_vendor != X86_VENDOR_INTEL)
808 return 0;
809
810 if (!cpu_has(c, X86_FEATURE_MWAIT) || boot_cpu_has_bug(X86_BUG_MONITOR))
811 return 0;
812
813 return 1;
814}
815
816/*
817 * MONITOR/MWAIT with no hints, used for default C1 state. This invokes MWAIT
818 * with interrupts enabled and no flags, which is backwards compatible with the
819 * original MWAIT implementation.
820 */
821static __cpuidle void mwait_idle(void)
822{
823 if (!current_set_polling_and_test()) {
824 if (this_cpu_has(X86_BUG_CLFLUSH_MONITOR)) {
825 mb(); /* quirk */
826 clflush((void *)¤t_thread_info()->flags);
827 mb(); /* quirk */
828 }
829
830 __monitor((void *)¤t_thread_info()->flags, 0, 0);
831 if (!need_resched())
832 __sti_mwait(0, 0);
833 else
834 raw_local_irq_enable();
835 } else {
836 raw_local_irq_enable();
837 }
838 __current_clr_polling();
839}
840
841void select_idle_routine(const struct cpuinfo_x86 *c)
842{
843#ifdef CONFIG_SMP
844 if (boot_option_idle_override == IDLE_POLL && smp_num_siblings > 1)
845 pr_warn_once("WARNING: polling idle and HT enabled, performance may degrade\n");
846#endif
847 if (x86_idle || boot_option_idle_override == IDLE_POLL)
848 return;
849
850 if (boot_cpu_has_bug(X86_BUG_AMD_E400)) {
851 pr_info("using AMD E400 aware idle routine\n");
852 x86_idle = amd_e400_idle;
853 } else if (prefer_mwait_c1_over_halt(c)) {
854 pr_info("using mwait in idle threads\n");
855 x86_idle = mwait_idle;
856 } else
857 x86_idle = default_idle;
858}
859
860void amd_e400_c1e_apic_setup(void)
861{
862 if (boot_cpu_has_bug(X86_BUG_AMD_APIC_C1E)) {
863 pr_info("Switch to broadcast mode on CPU%d\n", smp_processor_id());
864 local_irq_disable();
865 tick_broadcast_force();
866 local_irq_enable();
867 }
868}
869
870void __init arch_post_acpi_subsys_init(void)
871{
872 u32 lo, hi;
873
874 if (!boot_cpu_has_bug(X86_BUG_AMD_E400))
875 return;
876
877 /*
878 * AMD E400 detection needs to happen after ACPI has been enabled. If
879 * the machine is affected K8_INTP_C1E_ACTIVE_MASK bits are set in
880 * MSR_K8_INT_PENDING_MSG.
881 */
882 rdmsr(MSR_K8_INT_PENDING_MSG, lo, hi);
883 if (!(lo & K8_INTP_C1E_ACTIVE_MASK))
884 return;
885
886 boot_cpu_set_bug(X86_BUG_AMD_APIC_C1E);
887
888 if (!boot_cpu_has(X86_FEATURE_NONSTOP_TSC))
889 mark_tsc_unstable("TSC halt in AMD C1E");
890 pr_info("System has AMD C1E enabled\n");
891}
892
893static int __init idle_setup(char *str)
894{
895 if (!str)
896 return -EINVAL;
897
898 if (!strcmp(str, "poll")) {
899 pr_info("using polling idle threads\n");
900 boot_option_idle_override = IDLE_POLL;
901 cpu_idle_poll_ctrl(true);
902 } else if (!strcmp(str, "halt")) {
903 /*
904 * When the boot option of idle=halt is added, halt is
905 * forced to be used for CPU idle. In such case CPU C2/C3
906 * won't be used again.
907 * To continue to load the CPU idle driver, don't touch
908 * the boot_option_idle_override.
909 */
910 x86_idle = default_idle;
911 boot_option_idle_override = IDLE_HALT;
912 } else if (!strcmp(str, "nomwait")) {
913 /*
914 * If the boot option of "idle=nomwait" is added,
915 * it means that mwait will be disabled for CPU C2/C3
916 * states. In such case it won't touch the variable
917 * of boot_option_idle_override.
918 */
919 boot_option_idle_override = IDLE_NOMWAIT;
920 } else
921 return -1;
922
923 return 0;
924}
925early_param("idle", idle_setup);
926
927unsigned long arch_align_stack(unsigned long sp)
928{
929 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
930 sp -= get_random_int() % 8192;
931 return sp & ~0xf;
932}
933
934unsigned long arch_randomize_brk(struct mm_struct *mm)
935{
936 return randomize_page(mm->brk, 0x02000000);
937}
938
939/*
940 * Called from fs/proc with a reference on @p to find the function
941 * which called into schedule(). This needs to be done carefully
942 * because the task might wake up and we might look at a stack
943 * changing under us.
944 */
945unsigned long get_wchan(struct task_struct *p)
946{
947 unsigned long start, bottom, top, sp, fp, ip, ret = 0;
948 int count = 0;
949
950 if (p == current || task_is_running(p))
951 return 0;
952
953 if (!try_get_task_stack(p))
954 return 0;
955
956 start = (unsigned long)task_stack_page(p);
957 if (!start)
958 goto out;
959
960 /*
961 * Layout of the stack page:
962 *
963 * ----------- topmax = start + THREAD_SIZE - sizeof(unsigned long)
964 * PADDING
965 * ----------- top = topmax - TOP_OF_KERNEL_STACK_PADDING
966 * stack
967 * ----------- bottom = start
968 *
969 * The tasks stack pointer points at the location where the
970 * framepointer is stored. The data on the stack is:
971 * ... IP FP ... IP FP
972 *
973 * We need to read FP and IP, so we need to adjust the upper
974 * bound by another unsigned long.
975 */
976 top = start + THREAD_SIZE - TOP_OF_KERNEL_STACK_PADDING;
977 top -= 2 * sizeof(unsigned long);
978 bottom = start;
979
980 sp = READ_ONCE(p->thread.sp);
981 if (sp < bottom || sp > top)
982 goto out;
983
984 fp = READ_ONCE_NOCHECK(((struct inactive_task_frame *)sp)->bp);
985 do {
986 if (fp < bottom || fp > top)
987 goto out;
988 ip = READ_ONCE_NOCHECK(*(unsigned long *)(fp + sizeof(unsigned long)));
989 if (!in_sched_functions(ip)) {
990 ret = ip;
991 goto out;
992 }
993 fp = READ_ONCE_NOCHECK(*(unsigned long *)fp);
994 } while (count++ < 16 && !task_is_running(p));
995
996out:
997 put_task_stack(p);
998 return ret;
999}
1000
1001long do_arch_prctl_common(struct task_struct *task, int option,
1002 unsigned long cpuid_enabled)
1003{
1004 switch (option) {
1005 case ARCH_GET_CPUID:
1006 return get_cpuid_mode();
1007 case ARCH_SET_CPUID:
1008 return set_cpuid_mode(task, cpuid_enabled);
1009 }
1010
1011 return -EINVAL;
1012}