Linux Audio

Check our new training course

Real-Time Linux with PREEMPT_RT training

Feb 18-20, 2025
Register
Loading...
v3.5.6
 
  1/*
  2 * KVM paravirt_ops implementation
  3 *
  4 * This program is free software; you can redistribute it and/or modify
  5 * it under the terms of the GNU General Public License as published by
  6 * the Free Software Foundation; either version 2 of the License, or
  7 * (at your option) any later version.
  8 *
  9 * This program is distributed in the hope that it will be useful,
 10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 12 * GNU General Public License for more details.
 13 *
 14 * You should have received a copy of the GNU General Public License
 15 * along with this program; if not, write to the Free Software
 16 * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301, USA.
 17 *
 18 * Copyright (C) 2007, Red Hat, Inc., Ingo Molnar <mingo@redhat.com>
 19 * Copyright IBM Corporation, 2007
 20 *   Authors: Anthony Liguori <aliguori@us.ibm.com>
 21 */
 22
 23#include <linux/module.h>
 
 
 
 
 24#include <linux/kernel.h>
 25#include <linux/kvm_para.h>
 26#include <linux/cpu.h>
 27#include <linux/mm.h>
 28#include <linux/highmem.h>
 29#include <linux/hardirq.h>
 30#include <linux/notifier.h>
 31#include <linux/reboot.h>
 32#include <linux/hash.h>
 33#include <linux/sched.h>
 34#include <linux/slab.h>
 35#include <linux/kprobes.h>
 
 
 
 
 
 36#include <asm/timer.h>
 37#include <asm/cpu.h>
 38#include <asm/traps.h>
 39#include <asm/desc.h>
 40#include <asm/tlbflush.h>
 41#include <asm/idle.h>
 
 
 
 
 
 
 
 
 
 
 42
 43static int kvmapf = 1;
 44
 45static int parse_no_kvmapf(char *arg)
 46{
 47        kvmapf = 0;
 48        return 0;
 49}
 50
 51early_param("no-kvmapf", parse_no_kvmapf);
 52
 53static int steal_acc = 1;
 54static int parse_no_stealacc(char *arg)
 55{
 56        steal_acc = 0;
 57        return 0;
 58}
 59
 60early_param("no-steal-acc", parse_no_stealacc);
 61
 62static DEFINE_PER_CPU(struct kvm_vcpu_pv_apf_data, apf_reason) __aligned(64);
 63static DEFINE_PER_CPU(struct kvm_steal_time, steal_time) __aligned(64);
 64static int has_steal_clock = 0;
 65
 
 66/*
 67 * No need for any "IO delay" on KVM
 68 */
 69static void kvm_io_delay(void)
 70{
 71}
 72
 73#define KVM_TASK_SLEEP_HASHBITS 8
 74#define KVM_TASK_SLEEP_HASHSIZE (1<<KVM_TASK_SLEEP_HASHBITS)
 75
 76struct kvm_task_sleep_node {
 77	struct hlist_node link;
 78	wait_queue_head_t wq;
 79	u32 token;
 80	int cpu;
 81	bool halted;
 82};
 83
 84static struct kvm_task_sleep_head {
 85	spinlock_t lock;
 86	struct hlist_head list;
 87} async_pf_sleepers[KVM_TASK_SLEEP_HASHSIZE];
 88
 89static struct kvm_task_sleep_node *_find_apf_task(struct kvm_task_sleep_head *b,
 90						  u32 token)
 91{
 92	struct hlist_node *p;
 93
 94	hlist_for_each(p, &b->list) {
 95		struct kvm_task_sleep_node *n =
 96			hlist_entry(p, typeof(*n), link);
 97		if (n->token == token)
 98			return n;
 99	}
100
101	return NULL;
102}
103
104void kvm_async_pf_task_wait(u32 token)
105{
106	u32 key = hash_32(token, KVM_TASK_SLEEP_HASHBITS);
107	struct kvm_task_sleep_head *b = &async_pf_sleepers[key];
108	struct kvm_task_sleep_node n, *e;
109	DEFINE_WAIT(wait);
110	int cpu, idle;
111
112	cpu = get_cpu();
113	idle = idle_cpu(cpu);
114	put_cpu();
115
116	spin_lock(&b->lock);
117	e = _find_apf_task(b, token);
118	if (e) {
119		/* dummy entry exist -> wake up was delivered ahead of PF */
120		hlist_del(&e->link);
 
121		kfree(e);
122		spin_unlock(&b->lock);
123		return;
124	}
125
126	n.token = token;
127	n.cpu = smp_processor_id();
128	n.halted = idle || preempt_count() > 1;
129	init_waitqueue_head(&n.wq);
130	hlist_add_head(&n.link, &b->list);
131	spin_unlock(&b->lock);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
132
133	for (;;) {
134		if (!n.halted)
135			prepare_to_wait(&n.wq, &wait, TASK_UNINTERRUPTIBLE);
136		if (hlist_unhashed(&n.link))
137			break;
138
139		if (!n.halted) {
140			local_irq_enable();
141			schedule();
142			local_irq_disable();
143		} else {
144			/*
145			 * We cannot reschedule. So halt.
146			 */
147			native_safe_halt();
148			local_irq_disable();
149		}
150	}
151	if (!n.halted)
152		finish_wait(&n.wq, &wait);
153
154	return;
155}
156EXPORT_SYMBOL_GPL(kvm_async_pf_task_wait);
157
158static void apf_task_wake_one(struct kvm_task_sleep_node *n)
159{
160	hlist_del_init(&n->link);
161	if (n->halted)
162		smp_send_reschedule(n->cpu);
163	else if (waitqueue_active(&n->wq))
164		wake_up(&n->wq);
165}
166
167static void apf_task_wake_all(void)
168{
169	int i;
170
171	for (i = 0; i < KVM_TASK_SLEEP_HASHSIZE; i++) {
172		struct hlist_node *p, *next;
173		struct kvm_task_sleep_head *b = &async_pf_sleepers[i];
174		spin_lock(&b->lock);
 
 
 
175		hlist_for_each_safe(p, next, &b->list) {
176			struct kvm_task_sleep_node *n =
177				hlist_entry(p, typeof(*n), link);
178			if (n->cpu == smp_processor_id())
179				apf_task_wake_one(n);
180		}
181		spin_unlock(&b->lock);
182	}
183}
184
185void kvm_async_pf_task_wake(u32 token)
186{
187	u32 key = hash_32(token, KVM_TASK_SLEEP_HASHBITS);
188	struct kvm_task_sleep_head *b = &async_pf_sleepers[key];
189	struct kvm_task_sleep_node *n;
190
191	if (token == ~0) {
192		apf_task_wake_all();
193		return;
194	}
195
196again:
197	spin_lock(&b->lock);
198	n = _find_apf_task(b, token);
199	if (!n) {
200		/*
201		 * async PF was not yet handled.
202		 * Add dummy entry for the token.
 
203		 */
204		n = kzalloc(sizeof(*n), GFP_ATOMIC);
205		if (!n) {
 
 
 
 
 
 
 
 
 
 
206			/*
207			 * Allocation failed! Busy wait while other cpu
208			 * handles async PF.
209			 */
210			spin_unlock(&b->lock);
211			cpu_relax();
212			goto again;
213		}
214		n->token = token;
215		n->cpu = smp_processor_id();
216		init_waitqueue_head(&n->wq);
217		hlist_add_head(&n->link, &b->list);
218	} else
 
219		apf_task_wake_one(n);
220	spin_unlock(&b->lock);
221	return;
 
 
 
222}
223EXPORT_SYMBOL_GPL(kvm_async_pf_task_wake);
224
225u32 kvm_read_and_reset_pf_reason(void)
226{
227	u32 reason = 0;
228
229	if (__get_cpu_var(apf_reason).enabled) {
230		reason = __get_cpu_var(apf_reason).reason;
231		__get_cpu_var(apf_reason).reason = 0;
232	}
233
234	return reason;
235}
236EXPORT_SYMBOL_GPL(kvm_read_and_reset_pf_reason);
237
238dotraplinkage void __kprobes
239do_async_page_fault(struct pt_regs *regs, unsigned long error_code)
240{
241	switch (kvm_read_and_reset_pf_reason()) {
242	default:
243		do_page_fault(regs, error_code);
244		break;
245	case KVM_PV_REASON_PAGE_NOT_PRESENT:
246		/* page is swapped out by the host. */
247		kvm_async_pf_task_wait((u32)read_cr2());
248		break;
249	case KVM_PV_REASON_PAGE_READY:
250		rcu_irq_enter();
251		exit_idle();
252		kvm_async_pf_task_wake((u32)read_cr2());
253		rcu_irq_exit();
254		break;
 
 
 
 
 
 
 
 
 
 
255	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
256}
257
258static void __init paravirt_ops_setup(void)
259{
260	pv_info.name = "KVM";
261	pv_info.paravirt_enabled = 1;
262
263	if (kvm_para_has_feature(KVM_FEATURE_NOP_IO_DELAY))
264		pv_cpu_ops.io_delay = kvm_io_delay;
265
266#ifdef CONFIG_X86_IO_APIC
267	no_timer_check = 1;
268#endif
269}
270
271static void kvm_register_steal_time(void)
272{
273	int cpu = smp_processor_id();
274	struct kvm_steal_time *st = &per_cpu(steal_time, cpu);
275
276	if (!has_steal_clock)
277		return;
278
279	memset(st, 0, sizeof(*st));
280
281	wrmsrl(MSR_KVM_STEAL_TIME, (__pa(st) | KVM_MSR_ENABLED));
282	printk(KERN_INFO "kvm-stealtime: cpu %d, msr %lx\n",
283		cpu, __pa(st));
 
 
 
 
 
 
 
 
 
 
 
 
 
 
284}
285
286void __cpuinit kvm_guest_cpu_init(void)
287{
288	if (!kvm_para_available())
289		return;
290
291	if (kvm_para_has_feature(KVM_FEATURE_ASYNC_PF) && kvmapf) {
292		u64 pa = __pa(&__get_cpu_var(apf_reason));
293
294#ifdef CONFIG_PREEMPT
295		pa |= KVM_ASYNC_PF_SEND_ALWAYS;
296#endif
297		wrmsrl(MSR_KVM_ASYNC_PF_EN, pa | KVM_ASYNC_PF_ENABLED);
298		__get_cpu_var(apf_reason).enabled = 1;
299		printk(KERN_INFO"KVM setup async PF for cpu %d\n",
300		       smp_processor_id());
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
301	}
302
303	if (has_steal_clock)
304		kvm_register_steal_time();
305}
306
307static void kvm_pv_disable_apf(void *unused)
308{
309	if (!__get_cpu_var(apf_reason).enabled)
310		return;
311
312	wrmsrl(MSR_KVM_ASYNC_PF_EN, 0);
313	__get_cpu_var(apf_reason).enabled = 0;
314
315	printk(KERN_INFO"Unregister pv shared memory for cpu %d\n",
316	       smp_processor_id());
317}
318
319static int kvm_pv_reboot_notify(struct notifier_block *nb,
320				unsigned long code, void *unused)
321{
322	if (code == SYS_RESTART)
323		on_each_cpu(kvm_pv_disable_apf, NULL, 1);
324	return NOTIFY_DONE;
325}
326
327static struct notifier_block kvm_pv_reboot_nb = {
328	.notifier_call = kvm_pv_reboot_notify,
329};
330
331static u64 kvm_steal_clock(int cpu)
332{
333	u64 steal;
334	struct kvm_steal_time *src;
335	int version;
336
337	src = &per_cpu(steal_time, cpu);
338	do {
339		version = src->version;
340		rmb();
341		steal = src->steal;
342		rmb();
343	} while ((version & 1) || (version != src->version));
344
345	return steal;
346}
347
348void kvm_disable_steal_time(void)
349{
350	if (!has_steal_clock)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
351		return;
352
353	wrmsr(MSR_KVM_STEAL_TIME, 0, 0);
 
 
 
 
354}
355
356#ifdef CONFIG_SMP
357static void __init kvm_smp_prepare_boot_cpu(void)
358{
359#ifdef CONFIG_KVM_CLOCK
360	WARN_ON(kvm_register_clock("primary cpu clock"));
361#endif
362	kvm_guest_cpu_init();
363	native_smp_prepare_boot_cpu();
 
 
 
 
364}
365
366static void __cpuinit kvm_guest_cpu_online(void *dummy)
367{
 
 
 
368	kvm_guest_cpu_init();
 
 
369}
370
371static void kvm_guest_cpu_offline(void *dummy)
 
 
 
 
372{
373	kvm_disable_steal_time();
374	kvm_pv_disable_apf(NULL);
375	apf_task_wake_all();
 
 
376}
377
378static int __cpuinit kvm_cpu_notify(struct notifier_block *self,
379				    unsigned long action, void *hcpu)
380{
381	int cpu = (unsigned long)hcpu;
382	switch (action) {
383	case CPU_ONLINE:
384	case CPU_DOWN_FAILED:
385	case CPU_ONLINE_FROZEN:
386		smp_call_function_single(cpu, kvm_guest_cpu_online, NULL, 0);
387		break;
388	case CPU_DOWN_PREPARE:
389	case CPU_DOWN_PREPARE_FROZEN:
390		smp_call_function_single(cpu, kvm_guest_cpu_offline, NULL, 1);
391		break;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
392	default:
 
393		break;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
394	}
395	return NOTIFY_OK;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
396}
 
397
398static struct notifier_block __cpuinitdata kvm_cpu_notifier = {
399        .notifier_call  = kvm_cpu_notify,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
400};
 
 
 
 
 
 
 
 
 
 
 
 
 
401#endif
402
403static void __init kvm_apf_trap_init(void)
 
 
 
404{
405	set_intr_gate(14, &async_page_fault);
 
 
406}
 
 
 
407
408void __init kvm_guest_init(void)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
409{
410	int i;
411
412	if (!kvm_para_available())
413		return;
414
415	paravirt_ops_setup();
416	register_reboot_notifier(&kvm_pv_reboot_nb);
417	for (i = 0; i < KVM_TASK_SLEEP_HASHSIZE; i++)
418		spin_lock_init(&async_pf_sleepers[i].lock);
419	if (kvm_para_has_feature(KVM_FEATURE_ASYNC_PF))
420		x86_init.irqs.trap_init = kvm_apf_trap_init;
421
422	if (kvm_para_has_feature(KVM_FEATURE_STEAL_TIME)) {
423		has_steal_clock = 1;
424		pv_time_ops.steal_clock = kvm_steal_clock;
 
 
 
 
 
 
 
 
 
 
 
425	}
426
427#ifdef CONFIG_SMP
 
 
 
 
 
 
428	smp_ops.smp_prepare_boot_cpu = kvm_smp_prepare_boot_cpu;
429	register_cpu_notifier(&kvm_cpu_notifier);
 
 
 
 
 
 
430#else
 
431	kvm_guest_cpu_init();
432#endif
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
433}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
434
435static __init int activate_jump_labels(void)
436{
437	if (has_steal_clock) {
438		static_key_slow_inc(&paravirt_steal_enabled);
439		if (steal_acc)
440			static_key_slow_inc(&paravirt_steal_rq_enabled);
441	}
442
443	return 0;
444}
445arch_initcall(activate_jump_labels);
v6.2
   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   3 * KVM paravirt_ops implementation
   4 *
 
 
 
 
 
 
 
 
 
 
 
 
 
 
   5 * Copyright (C) 2007, Red Hat, Inc., Ingo Molnar <mingo@redhat.com>
   6 * Copyright IBM Corporation, 2007
   7 *   Authors: Anthony Liguori <aliguori@us.ibm.com>
   8 */
   9
  10#define pr_fmt(fmt) "kvm-guest: " fmt
  11
  12#include <linux/context_tracking.h>
  13#include <linux/init.h>
  14#include <linux/irq.h>
  15#include <linux/kernel.h>
  16#include <linux/kvm_para.h>
  17#include <linux/cpu.h>
  18#include <linux/mm.h>
  19#include <linux/highmem.h>
  20#include <linux/hardirq.h>
  21#include <linux/notifier.h>
  22#include <linux/reboot.h>
  23#include <linux/hash.h>
  24#include <linux/sched.h>
  25#include <linux/slab.h>
  26#include <linux/kprobes.h>
  27#include <linux/nmi.h>
  28#include <linux/swait.h>
  29#include <linux/syscore_ops.h>
  30#include <linux/cc_platform.h>
  31#include <linux/efi.h>
  32#include <asm/timer.h>
  33#include <asm/cpu.h>
  34#include <asm/traps.h>
  35#include <asm/desc.h>
  36#include <asm/tlbflush.h>
  37#include <asm/apic.h>
  38#include <asm/apicdef.h>
  39#include <asm/hypervisor.h>
  40#include <asm/tlb.h>
  41#include <asm/cpuidle_haltpoll.h>
  42#include <asm/ptrace.h>
  43#include <asm/reboot.h>
  44#include <asm/svm.h>
  45#include <asm/e820/api.h>
  46
  47DEFINE_STATIC_KEY_FALSE(kvm_async_pf_enabled);
  48
  49static int kvmapf = 1;
  50
  51static int __init parse_no_kvmapf(char *arg)
  52{
  53        kvmapf = 0;
  54        return 0;
  55}
  56
  57early_param("no-kvmapf", parse_no_kvmapf);
  58
  59static int steal_acc = 1;
  60static int __init parse_no_stealacc(char *arg)
  61{
  62        steal_acc = 0;
  63        return 0;
  64}
  65
  66early_param("no-steal-acc", parse_no_stealacc);
  67
  68static DEFINE_PER_CPU_DECRYPTED(struct kvm_vcpu_pv_apf_data, apf_reason) __aligned(64);
  69DEFINE_PER_CPU_DECRYPTED(struct kvm_steal_time, steal_time) __aligned(64) __visible;
  70static int has_steal_clock = 0;
  71
  72static int has_guest_poll = 0;
  73/*
  74 * No need for any "IO delay" on KVM
  75 */
  76static void kvm_io_delay(void)
  77{
  78}
  79
  80#define KVM_TASK_SLEEP_HASHBITS 8
  81#define KVM_TASK_SLEEP_HASHSIZE (1<<KVM_TASK_SLEEP_HASHBITS)
  82
  83struct kvm_task_sleep_node {
  84	struct hlist_node link;
  85	struct swait_queue_head wq;
  86	u32 token;
  87	int cpu;
 
  88};
  89
  90static struct kvm_task_sleep_head {
  91	raw_spinlock_t lock;
  92	struct hlist_head list;
  93} async_pf_sleepers[KVM_TASK_SLEEP_HASHSIZE];
  94
  95static struct kvm_task_sleep_node *_find_apf_task(struct kvm_task_sleep_head *b,
  96						  u32 token)
  97{
  98	struct hlist_node *p;
  99
 100	hlist_for_each(p, &b->list) {
 101		struct kvm_task_sleep_node *n =
 102			hlist_entry(p, typeof(*n), link);
 103		if (n->token == token)
 104			return n;
 105	}
 106
 107	return NULL;
 108}
 109
 110static bool kvm_async_pf_queue_task(u32 token, struct kvm_task_sleep_node *n)
 111{
 112	u32 key = hash_32(token, KVM_TASK_SLEEP_HASHBITS);
 113	struct kvm_task_sleep_head *b = &async_pf_sleepers[key];
 114	struct kvm_task_sleep_node *e;
 
 
 
 
 
 
 115
 116	raw_spin_lock(&b->lock);
 117	e = _find_apf_task(b, token);
 118	if (e) {
 119		/* dummy entry exist -> wake up was delivered ahead of PF */
 120		hlist_del(&e->link);
 121		raw_spin_unlock(&b->lock);
 122		kfree(e);
 123		return false;
 
 124	}
 125
 126	n->token = token;
 127	n->cpu = smp_processor_id();
 128	init_swait_queue_head(&n->wq);
 129	hlist_add_head(&n->link, &b->list);
 130	raw_spin_unlock(&b->lock);
 131	return true;
 132}
 133
 134/*
 135 * kvm_async_pf_task_wait_schedule - Wait for pagefault to be handled
 136 * @token:	Token to identify the sleep node entry
 137 *
 138 * Invoked from the async pagefault handling code or from the VM exit page
 139 * fault handler. In both cases RCU is watching.
 140 */
 141void kvm_async_pf_task_wait_schedule(u32 token)
 142{
 143	struct kvm_task_sleep_node n;
 144	DECLARE_SWAITQUEUE(wait);
 145
 146	lockdep_assert_irqs_disabled();
 147
 148	if (!kvm_async_pf_queue_task(token, &n))
 149		return;
 150
 151	for (;;) {
 152		prepare_to_swait_exclusive(&n.wq, &wait, TASK_UNINTERRUPTIBLE);
 
 153		if (hlist_unhashed(&n.link))
 154			break;
 155
 156		local_irq_enable();
 157		schedule();
 158		local_irq_disable();
 
 
 
 
 
 
 
 
 159	}
 160	finish_swait(&n.wq, &wait);
 
 
 
 161}
 162EXPORT_SYMBOL_GPL(kvm_async_pf_task_wait_schedule);
 163
 164static void apf_task_wake_one(struct kvm_task_sleep_node *n)
 165{
 166	hlist_del_init(&n->link);
 167	if (swq_has_sleeper(&n->wq))
 168		swake_up_one(&n->wq);
 
 
 169}
 170
 171static void apf_task_wake_all(void)
 172{
 173	int i;
 174
 175	for (i = 0; i < KVM_TASK_SLEEP_HASHSIZE; i++) {
 
 176		struct kvm_task_sleep_head *b = &async_pf_sleepers[i];
 177		struct kvm_task_sleep_node *n;
 178		struct hlist_node *p, *next;
 179
 180		raw_spin_lock(&b->lock);
 181		hlist_for_each_safe(p, next, &b->list) {
 182			n = hlist_entry(p, typeof(*n), link);
 
 183			if (n->cpu == smp_processor_id())
 184				apf_task_wake_one(n);
 185		}
 186		raw_spin_unlock(&b->lock);
 187	}
 188}
 189
 190void kvm_async_pf_task_wake(u32 token)
 191{
 192	u32 key = hash_32(token, KVM_TASK_SLEEP_HASHBITS);
 193	struct kvm_task_sleep_head *b = &async_pf_sleepers[key];
 194	struct kvm_task_sleep_node *n, *dummy = NULL;
 195
 196	if (token == ~0) {
 197		apf_task_wake_all();
 198		return;
 199	}
 200
 201again:
 202	raw_spin_lock(&b->lock);
 203	n = _find_apf_task(b, token);
 204	if (!n) {
 205		/*
 206		 * Async #PF not yet handled, add a dummy entry for the token.
 207		 * Allocating the token must be down outside of the raw lock
 208		 * as the allocator is preemptible on PREEMPT_RT kernels.
 209		 */
 210		if (!dummy) {
 211			raw_spin_unlock(&b->lock);
 212			dummy = kzalloc(sizeof(*dummy), GFP_ATOMIC);
 213
 214			/*
 215			 * Continue looping on allocation failure, eventually
 216			 * the async #PF will be handled and allocating a new
 217			 * node will be unnecessary.
 218			 */
 219			if (!dummy)
 220				cpu_relax();
 221
 222			/*
 223			 * Recheck for async #PF completion before enqueueing
 224			 * the dummy token to avoid duplicate list entries.
 225			 */
 
 
 226			goto again;
 227		}
 228		dummy->token = token;
 229		dummy->cpu = smp_processor_id();
 230		init_swait_queue_head(&dummy->wq);
 231		hlist_add_head(&dummy->link, &b->list);
 232		dummy = NULL;
 233	} else {
 234		apf_task_wake_one(n);
 235	}
 236	raw_spin_unlock(&b->lock);
 237
 238	/* A dummy token might be allocated and ultimately not used.  */
 239	kfree(dummy);
 240}
 241EXPORT_SYMBOL_GPL(kvm_async_pf_task_wake);
 242
 243noinstr u32 kvm_read_and_reset_apf_flags(void)
 244{
 245	u32 flags = 0;
 246
 247	if (__this_cpu_read(apf_reason.enabled)) {
 248		flags = __this_cpu_read(apf_reason.flags);
 249		__this_cpu_write(apf_reason.flags, 0);
 250	}
 251
 252	return flags;
 253}
 254EXPORT_SYMBOL_GPL(kvm_read_and_reset_apf_flags);
 255
 256noinstr bool __kvm_handle_async_pf(struct pt_regs *regs, u32 token)
 
 257{
 258	u32 flags = kvm_read_and_reset_apf_flags();
 259	irqentry_state_t state;
 260
 261	if (!flags)
 262		return false;
 263
 264	state = irqentry_enter(regs);
 265	instrumentation_begin();
 266
 267	/*
 268	 * If the host managed to inject an async #PF into an interrupt
 269	 * disabled region, then die hard as this is not going to end well
 270	 * and the host side is seriously broken.
 271	 */
 272	if (unlikely(!(regs->flags & X86_EFLAGS_IF)))
 273		panic("Host injected async #PF in interrupt disabled region\n");
 274
 275	if (flags & KVM_PV_REASON_PAGE_NOT_PRESENT) {
 276		if (unlikely(!(user_mode(regs))))
 277			panic("Host injected async #PF in kernel mode\n");
 278		/* Page is swapped out by the host. */
 279		kvm_async_pf_task_wait_schedule(token);
 280	} else {
 281		WARN_ONCE(1, "Unexpected async PF flags: %x\n", flags);
 282	}
 283
 284	instrumentation_end();
 285	irqentry_exit(regs, state);
 286	return true;
 287}
 288
 289DEFINE_IDTENTRY_SYSVEC(sysvec_kvm_asyncpf_interrupt)
 290{
 291	struct pt_regs *old_regs = set_irq_regs(regs);
 292	u32 token;
 293
 294	ack_APIC_irq();
 295
 296	inc_irq_stat(irq_hv_callback_count);
 297
 298	if (__this_cpu_read(apf_reason.enabled)) {
 299		token = __this_cpu_read(apf_reason.token);
 300		kvm_async_pf_task_wake(token);
 301		__this_cpu_write(apf_reason.token, 0);
 302		wrmsrl(MSR_KVM_ASYNC_PF_ACK, 1);
 303	}
 304
 305	set_irq_regs(old_regs);
 306}
 307
 308static void __init paravirt_ops_setup(void)
 309{
 310	pv_info.name = "KVM";
 
 311
 312	if (kvm_para_has_feature(KVM_FEATURE_NOP_IO_DELAY))
 313		pv_ops.cpu.io_delay = kvm_io_delay;
 314
 315#ifdef CONFIG_X86_IO_APIC
 316	no_timer_check = 1;
 317#endif
 318}
 319
 320static void kvm_register_steal_time(void)
 321{
 322	int cpu = smp_processor_id();
 323	struct kvm_steal_time *st = &per_cpu(steal_time, cpu);
 324
 325	if (!has_steal_clock)
 326		return;
 327
 328	wrmsrl(MSR_KVM_STEAL_TIME, (slow_virt_to_phys(st) | KVM_MSR_ENABLED));
 329	pr_debug("stealtime: cpu %d, msr %llx\n", cpu,
 330		(unsigned long long) slow_virt_to_phys(st));
 331}
 332
 333static DEFINE_PER_CPU_DECRYPTED(unsigned long, kvm_apic_eoi) = KVM_PV_EOI_DISABLED;
 334
 335static notrace void kvm_guest_apic_eoi_write(u32 reg, u32 val)
 336{
 337	/**
 338	 * This relies on __test_and_clear_bit to modify the memory
 339	 * in a way that is atomic with respect to the local CPU.
 340	 * The hypervisor only accesses this memory from the local CPU so
 341	 * there's no need for lock or memory barriers.
 342	 * An optimization barrier is implied in apic write.
 343	 */
 344	if (__test_and_clear_bit(KVM_PV_EOI_BIT, this_cpu_ptr(&kvm_apic_eoi)))
 345		return;
 346	apic->native_eoi_write(APIC_EOI, APIC_EOI_ACK);
 347}
 348
 349static void kvm_guest_cpu_init(void)
 350{
 351	if (kvm_para_has_feature(KVM_FEATURE_ASYNC_PF_INT) && kvmapf) {
 352		u64 pa;
 353
 354		WARN_ON_ONCE(!static_branch_likely(&kvm_async_pf_enabled));
 
 355
 356		pa = slow_virt_to_phys(this_cpu_ptr(&apf_reason));
 357		pa |= KVM_ASYNC_PF_ENABLED | KVM_ASYNC_PF_DELIVERY_AS_INT;
 358
 359		if (kvm_para_has_feature(KVM_FEATURE_ASYNC_PF_VMEXIT))
 360			pa |= KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
 361
 362		wrmsrl(MSR_KVM_ASYNC_PF_INT, HYPERVISOR_CALLBACK_VECTOR);
 363
 364		wrmsrl(MSR_KVM_ASYNC_PF_EN, pa);
 365		__this_cpu_write(apf_reason.enabled, 1);
 366		pr_debug("setup async PF for cpu %d\n", smp_processor_id());
 367	}
 368
 369	if (kvm_para_has_feature(KVM_FEATURE_PV_EOI)) {
 370		unsigned long pa;
 371
 372		/* Size alignment is implied but just to make it explicit. */
 373		BUILD_BUG_ON(__alignof__(kvm_apic_eoi) < 4);
 374		__this_cpu_write(kvm_apic_eoi, 0);
 375		pa = slow_virt_to_phys(this_cpu_ptr(&kvm_apic_eoi))
 376			| KVM_MSR_ENABLED;
 377		wrmsrl(MSR_KVM_PV_EOI_EN, pa);
 378	}
 379
 380	if (has_steal_clock)
 381		kvm_register_steal_time();
 382}
 383
 384static void kvm_pv_disable_apf(void)
 385{
 386	if (!__this_cpu_read(apf_reason.enabled))
 387		return;
 388
 389	wrmsrl(MSR_KVM_ASYNC_PF_EN, 0);
 390	__this_cpu_write(apf_reason.enabled, 0);
 391
 392	pr_debug("disable async PF for cpu %d\n", smp_processor_id());
 
 393}
 394
 395static void kvm_disable_steal_time(void)
 
 396{
 397	if (!has_steal_clock)
 398		return;
 
 
 399
 400	wrmsr(MSR_KVM_STEAL_TIME, 0, 0);
 401}
 
 402
 403static u64 kvm_steal_clock(int cpu)
 404{
 405	u64 steal;
 406	struct kvm_steal_time *src;
 407	int version;
 408
 409	src = &per_cpu(steal_time, cpu);
 410	do {
 411		version = src->version;
 412		virt_rmb();
 413		steal = src->steal;
 414		virt_rmb();
 415	} while ((version & 1) || (version != src->version));
 416
 417	return steal;
 418}
 419
 420static inline void __set_percpu_decrypted(void *ptr, unsigned long size)
 421{
 422	early_set_memory_decrypted((unsigned long) ptr, size);
 423}
 424
 425/*
 426 * Iterate through all possible CPUs and map the memory region pointed
 427 * by apf_reason, steal_time and kvm_apic_eoi as decrypted at once.
 428 *
 429 * Note: we iterate through all possible CPUs to ensure that CPUs
 430 * hotplugged will have their per-cpu variable already mapped as
 431 * decrypted.
 432 */
 433static void __init sev_map_percpu_data(void)
 434{
 435	int cpu;
 436
 437	if (!cc_platform_has(CC_ATTR_GUEST_MEM_ENCRYPT))
 438		return;
 439
 440	for_each_possible_cpu(cpu) {
 441		__set_percpu_decrypted(&per_cpu(apf_reason, cpu), sizeof(apf_reason));
 442		__set_percpu_decrypted(&per_cpu(steal_time, cpu), sizeof(steal_time));
 443		__set_percpu_decrypted(&per_cpu(kvm_apic_eoi, cpu), sizeof(kvm_apic_eoi));
 444	}
 445}
 446
 447static void kvm_guest_cpu_offline(bool shutdown)
 
 448{
 449	kvm_disable_steal_time();
 450	if (kvm_para_has_feature(KVM_FEATURE_PV_EOI))
 451		wrmsrl(MSR_KVM_PV_EOI_EN, 0);
 452	if (kvm_para_has_feature(KVM_FEATURE_MIGRATION_CONTROL))
 453		wrmsrl(MSR_KVM_MIGRATION_CONTROL, 0);
 454	kvm_pv_disable_apf();
 455	if (!shutdown)
 456		apf_task_wake_all();
 457	kvmclock_disable();
 458}
 459
 460static int kvm_cpu_online(unsigned int cpu)
 461{
 462	unsigned long flags;
 463
 464	local_irq_save(flags);
 465	kvm_guest_cpu_init();
 466	local_irq_restore(flags);
 467	return 0;
 468}
 469
 470#ifdef CONFIG_SMP
 471
 472static DEFINE_PER_CPU(cpumask_var_t, __pv_cpu_mask);
 473
 474static bool pv_tlb_flush_supported(void)
 475{
 476	return (kvm_para_has_feature(KVM_FEATURE_PV_TLB_FLUSH) &&
 477		!kvm_para_has_hint(KVM_HINTS_REALTIME) &&
 478		kvm_para_has_feature(KVM_FEATURE_STEAL_TIME) &&
 479		!boot_cpu_has(X86_FEATURE_MWAIT) &&
 480		(num_possible_cpus() != 1));
 481}
 482
 483static bool pv_ipi_supported(void)
 
 484{
 485	return (kvm_para_has_feature(KVM_FEATURE_PV_SEND_IPI) &&
 486	       (num_possible_cpus() != 1));
 487}
 488
 489static bool pv_sched_yield_supported(void)
 490{
 491	return (kvm_para_has_feature(KVM_FEATURE_PV_SCHED_YIELD) &&
 492		!kvm_para_has_hint(KVM_HINTS_REALTIME) &&
 493	    kvm_para_has_feature(KVM_FEATURE_STEAL_TIME) &&
 494	    !boot_cpu_has(X86_FEATURE_MWAIT) &&
 495	    (num_possible_cpus() != 1));
 496}
 497
 498#define KVM_IPI_CLUSTER_SIZE	(2 * BITS_PER_LONG)
 499
 500static void __send_ipi_mask(const struct cpumask *mask, int vector)
 501{
 502	unsigned long flags;
 503	int cpu, apic_id, icr;
 504	int min = 0, max = 0;
 505#ifdef CONFIG_X86_64
 506	__uint128_t ipi_bitmap = 0;
 507#else
 508	u64 ipi_bitmap = 0;
 509#endif
 510	long ret;
 511
 512	if (cpumask_empty(mask))
 513		return;
 514
 515	local_irq_save(flags);
 516
 517	switch (vector) {
 518	default:
 519		icr = APIC_DM_FIXED | vector;
 520		break;
 521	case NMI_VECTOR:
 522		icr = APIC_DM_NMI;
 523		break;
 524	}
 525
 526	for_each_cpu(cpu, mask) {
 527		apic_id = per_cpu(x86_cpu_to_apicid, cpu);
 528		if (!ipi_bitmap) {
 529			min = max = apic_id;
 530		} else if (apic_id < min && max - apic_id < KVM_IPI_CLUSTER_SIZE) {
 531			ipi_bitmap <<= min - apic_id;
 532			min = apic_id;
 533		} else if (apic_id > min && apic_id < min + KVM_IPI_CLUSTER_SIZE) {
 534			max = apic_id < max ? max : apic_id;
 535		} else {
 536			ret = kvm_hypercall4(KVM_HC_SEND_IPI, (unsigned long)ipi_bitmap,
 537				(unsigned long)(ipi_bitmap >> BITS_PER_LONG), min, icr);
 538			WARN_ONCE(ret < 0, "kvm-guest: failed to send PV IPI: %ld",
 539				  ret);
 540			min = max = apic_id;
 541			ipi_bitmap = 0;
 542		}
 543		__set_bit(apic_id - min, (unsigned long *)&ipi_bitmap);
 544	}
 545
 546	if (ipi_bitmap) {
 547		ret = kvm_hypercall4(KVM_HC_SEND_IPI, (unsigned long)ipi_bitmap,
 548			(unsigned long)(ipi_bitmap >> BITS_PER_LONG), min, icr);
 549		WARN_ONCE(ret < 0, "kvm-guest: failed to send PV IPI: %ld",
 550			  ret);
 551	}
 552
 553	local_irq_restore(flags);
 554}
 555
 556static void kvm_send_ipi_mask(const struct cpumask *mask, int vector)
 557{
 558	__send_ipi_mask(mask, vector);
 559}
 560
 561static void kvm_send_ipi_mask_allbutself(const struct cpumask *mask, int vector)
 562{
 563	unsigned int this_cpu = smp_processor_id();
 564	struct cpumask *new_mask = this_cpu_cpumask_var_ptr(__pv_cpu_mask);
 565	const struct cpumask *local_mask;
 566
 567	cpumask_copy(new_mask, mask);
 568	cpumask_clear_cpu(this_cpu, new_mask);
 569	local_mask = new_mask;
 570	__send_ipi_mask(local_mask, vector);
 571}
 572
 573static int __init setup_efi_kvm_sev_migration(void)
 574{
 575	efi_char16_t efi_sev_live_migration_enabled[] = L"SevLiveMigrationEnabled";
 576	efi_guid_t efi_variable_guid = AMD_SEV_MEM_ENCRYPT_GUID;
 577	efi_status_t status;
 578	unsigned long size;
 579	bool enabled;
 580
 581	if (!cc_platform_has(CC_ATTR_GUEST_MEM_ENCRYPT) ||
 582	    !kvm_para_has_feature(KVM_FEATURE_MIGRATION_CONTROL))
 583		return 0;
 584
 585	if (!efi_enabled(EFI_BOOT))
 586		return 0;
 587
 588	if (!efi_enabled(EFI_RUNTIME_SERVICES)) {
 589		pr_info("%s : EFI runtime services are not enabled\n", __func__);
 590		return 0;
 591	}
 592
 593	size = sizeof(enabled);
 594
 595	/* Get variable contents into buffer */
 596	status = efi.get_variable(efi_sev_live_migration_enabled,
 597				  &efi_variable_guid, NULL, &size, &enabled);
 598
 599	if (status == EFI_NOT_FOUND) {
 600		pr_info("%s : EFI live migration variable not found\n", __func__);
 601		return 0;
 602	}
 603
 604	if (status != EFI_SUCCESS) {
 605		pr_info("%s : EFI variable retrieval failed\n", __func__);
 606		return 0;
 607	}
 608
 609	if (enabled == 0) {
 610		pr_info("%s: live migration disabled in EFI\n", __func__);
 611		return 0;
 612	}
 613
 614	pr_info("%s : live migration enabled in EFI\n", __func__);
 615	wrmsrl(MSR_KVM_MIGRATION_CONTROL, KVM_MIGRATION_READY);
 616
 617	return 1;
 618}
 619
 620late_initcall(setup_efi_kvm_sev_migration);
 621
 622/*
 623 * Set the IPI entry points
 624 */
 625static void kvm_setup_pv_ipi(void)
 626{
 627	apic->send_IPI_mask = kvm_send_ipi_mask;
 628	apic->send_IPI_mask_allbutself = kvm_send_ipi_mask_allbutself;
 629	pr_info("setup PV IPIs\n");
 630}
 631
 632static void kvm_smp_send_call_func_ipi(const struct cpumask *mask)
 633{
 634	int cpu;
 635
 636	native_send_call_func_ipi(mask);
 637
 638	/* Make sure other vCPUs get a chance to run if they need to. */
 639	for_each_cpu(cpu, mask) {
 640		if (!idle_cpu(cpu) && vcpu_is_preempted(cpu)) {
 641			kvm_hypercall1(KVM_HC_SCHED_YIELD, per_cpu(x86_cpu_to_apicid, cpu));
 642			break;
 643		}
 644	}
 645}
 646
 647static void kvm_flush_tlb_multi(const struct cpumask *cpumask,
 648			const struct flush_tlb_info *info)
 649{
 650	u8 state;
 651	int cpu;
 652	struct kvm_steal_time *src;
 653	struct cpumask *flushmask = this_cpu_cpumask_var_ptr(__pv_cpu_mask);
 654
 655	cpumask_copy(flushmask, cpumask);
 656	/*
 657	 * We have to call flush only on online vCPUs. And
 658	 * queue flush_on_enter for pre-empted vCPUs
 659	 */
 660	for_each_cpu(cpu, flushmask) {
 661		/*
 662		 * The local vCPU is never preempted, so we do not explicitly
 663		 * skip check for local vCPU - it will never be cleared from
 664		 * flushmask.
 665		 */
 666		src = &per_cpu(steal_time, cpu);
 667		state = READ_ONCE(src->preempted);
 668		if ((state & KVM_VCPU_PREEMPTED)) {
 669			if (try_cmpxchg(&src->preempted, &state,
 670					state | KVM_VCPU_FLUSH_TLB))
 671				__cpumask_clear_cpu(cpu, flushmask);
 672		}
 673	}
 674
 675	native_flush_tlb_multi(flushmask, info);
 676}
 677
 678static __init int kvm_alloc_cpumask(void)
 679{
 680	int cpu;
 681
 682	if (!kvm_para_available() || nopv)
 683		return 0;
 684
 685	if (pv_tlb_flush_supported() || pv_ipi_supported())
 686		for_each_possible_cpu(cpu) {
 687			zalloc_cpumask_var_node(per_cpu_ptr(&__pv_cpu_mask, cpu),
 688				GFP_KERNEL, cpu_to_node(cpu));
 689		}
 690
 691	return 0;
 692}
 693arch_initcall(kvm_alloc_cpumask);
 694
 695static void __init kvm_smp_prepare_boot_cpu(void)
 696{
 697	/*
 698	 * Map the per-cpu variables as decrypted before kvm_guest_cpu_init()
 699	 * shares the guest physical address with the hypervisor.
 700	 */
 701	sev_map_percpu_data();
 702
 703	kvm_guest_cpu_init();
 704	native_smp_prepare_boot_cpu();
 705	kvm_spinlock_init();
 706}
 707
 708static int kvm_cpu_down_prepare(unsigned int cpu)
 709{
 710	unsigned long flags;
 711
 712	local_irq_save(flags);
 713	kvm_guest_cpu_offline(false);
 714	local_irq_restore(flags);
 715	return 0;
 716}
 717
 718#endif
 719
 720static int kvm_suspend(void)
 721{
 722	u64 val = 0;
 723
 724	kvm_guest_cpu_offline(false);
 725
 726#ifdef CONFIG_ARCH_CPUIDLE_HALTPOLL
 727	if (kvm_para_has_feature(KVM_FEATURE_POLL_CONTROL))
 728		rdmsrl(MSR_KVM_POLL_CONTROL, val);
 729	has_guest_poll = !(val & 1);
 730#endif
 731	return 0;
 732}
 733
 734static void kvm_resume(void)
 735{
 736	kvm_cpu_online(raw_smp_processor_id());
 737
 738#ifdef CONFIG_ARCH_CPUIDLE_HALTPOLL
 739	if (kvm_para_has_feature(KVM_FEATURE_POLL_CONTROL) && has_guest_poll)
 740		wrmsrl(MSR_KVM_POLL_CONTROL, 0);
 741#endif
 742}
 743
 744static struct syscore_ops kvm_syscore_ops = {
 745	.suspend	= kvm_suspend,
 746	.resume		= kvm_resume,
 747};
 748
 749static void kvm_pv_guest_cpu_reboot(void *unused)
 750{
 751	kvm_guest_cpu_offline(true);
 752}
 753
 754static int kvm_pv_reboot_notify(struct notifier_block *nb,
 755				unsigned long code, void *unused)
 756{
 757	if (code == SYS_RESTART)
 758		on_each_cpu(kvm_pv_guest_cpu_reboot, NULL, 1);
 759	return NOTIFY_DONE;
 760}
 761
 762static struct notifier_block kvm_pv_reboot_nb = {
 763	.notifier_call = kvm_pv_reboot_notify,
 764};
 765
 766/*
 767 * After a PV feature is registered, the host will keep writing to the
 768 * registered memory location. If the guest happens to shutdown, this memory
 769 * won't be valid. In cases like kexec, in which you install a new kernel, this
 770 * means a random memory location will be kept being written.
 771 */
 772#ifdef CONFIG_KEXEC_CORE
 773static void kvm_crash_shutdown(struct pt_regs *regs)
 774{
 775	kvm_guest_cpu_offline(true);
 776	native_machine_crash_shutdown(regs);
 777}
 778#endif
 779
 780#if defined(CONFIG_X86_32) || !defined(CONFIG_SMP)
 781bool __kvm_vcpu_is_preempted(long cpu);
 782
 783__visible bool __kvm_vcpu_is_preempted(long cpu)
 784{
 785	struct kvm_steal_time *src = &per_cpu(steal_time, cpu);
 786
 787	return !!(src->preempted & KVM_VCPU_PREEMPTED);
 788}
 789PV_CALLEE_SAVE_REGS_THUNK(__kvm_vcpu_is_preempted);
 790
 791#else
 792
 793#include <asm/asm-offsets.h>
 794
 795extern bool __raw_callee_save___kvm_vcpu_is_preempted(long);
 796
 797/*
 798 * Hand-optimize version for x86-64 to avoid 8 64-bit register saving and
 799 * restoring to/from the stack.
 800 */
 801#define PV_VCPU_PREEMPTED_ASM						     \
 802 "movq   __per_cpu_offset(,%rdi,8), %rax\n\t"				     \
 803 "cmpb   $0, " __stringify(KVM_STEAL_TIME_preempted) "+steal_time(%rax)\n\t" \
 804 "setne  %al\n\t"
 805
 806DEFINE_PARAVIRT_ASM(__raw_callee_save___kvm_vcpu_is_preempted,
 807		    PV_VCPU_PREEMPTED_ASM, .text);
 808#endif
 809
 810static void __init kvm_guest_init(void)
 811{
 812	int i;
 813
 
 
 
 814	paravirt_ops_setup();
 815	register_reboot_notifier(&kvm_pv_reboot_nb);
 816	for (i = 0; i < KVM_TASK_SLEEP_HASHSIZE; i++)
 817		raw_spin_lock_init(&async_pf_sleepers[i].lock);
 
 
 818
 819	if (kvm_para_has_feature(KVM_FEATURE_STEAL_TIME)) {
 820		has_steal_clock = 1;
 821		static_call_update(pv_steal_clock, kvm_steal_clock);
 822
 823		pv_ops.lock.vcpu_is_preempted =
 824			PV_CALLEE_SAVE(__kvm_vcpu_is_preempted);
 825	}
 826
 827	if (kvm_para_has_feature(KVM_FEATURE_PV_EOI))
 828		apic_set_eoi_write(kvm_guest_apic_eoi_write);
 829
 830	if (kvm_para_has_feature(KVM_FEATURE_ASYNC_PF_INT) && kvmapf) {
 831		static_branch_enable(&kvm_async_pf_enabled);
 832		alloc_intr_gate(HYPERVISOR_CALLBACK_VECTOR, asm_sysvec_kvm_asyncpf_interrupt);
 833	}
 834
 835#ifdef CONFIG_SMP
 836	if (pv_tlb_flush_supported()) {
 837		pv_ops.mmu.flush_tlb_multi = kvm_flush_tlb_multi;
 838		pv_ops.mmu.tlb_remove_table = tlb_remove_table;
 839		pr_info("KVM setup pv remote TLB flush\n");
 840	}
 841
 842	smp_ops.smp_prepare_boot_cpu = kvm_smp_prepare_boot_cpu;
 843	if (pv_sched_yield_supported()) {
 844		smp_ops.send_call_func_ipi = kvm_smp_send_call_func_ipi;
 845		pr_info("setup PV sched yield\n");
 846	}
 847	if (cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN, "x86/kvm:online",
 848				      kvm_cpu_online, kvm_cpu_down_prepare) < 0)
 849		pr_err("failed to install cpu hotplug callbacks\n");
 850#else
 851	sev_map_percpu_data();
 852	kvm_guest_cpu_init();
 853#endif
 854
 855#ifdef CONFIG_KEXEC_CORE
 856	machine_ops.crash_shutdown = kvm_crash_shutdown;
 857#endif
 858
 859	register_syscore_ops(&kvm_syscore_ops);
 860
 861	/*
 862	 * Hard lockup detection is enabled by default. Disable it, as guests
 863	 * can get false positives too easily, for example if the host is
 864	 * overcommitted.
 865	 */
 866	hardlockup_detector_disable();
 867}
 868
 869static noinline uint32_t __kvm_cpuid_base(void)
 870{
 871	if (boot_cpu_data.cpuid_level < 0)
 872		return 0;	/* So we don't blow up on old processors */
 873
 874	if (boot_cpu_has(X86_FEATURE_HYPERVISOR))
 875		return hypervisor_cpuid_base(KVM_SIGNATURE, 0);
 876
 877	return 0;
 878}
 879
 880static inline uint32_t kvm_cpuid_base(void)
 881{
 882	static int kvm_cpuid_base = -1;
 883
 884	if (kvm_cpuid_base == -1)
 885		kvm_cpuid_base = __kvm_cpuid_base();
 886
 887	return kvm_cpuid_base;
 888}
 889
 890bool kvm_para_available(void)
 891{
 892	return kvm_cpuid_base() != 0;
 893}
 894EXPORT_SYMBOL_GPL(kvm_para_available);
 895
 896unsigned int kvm_arch_para_features(void)
 897{
 898	return cpuid_eax(kvm_cpuid_base() | KVM_CPUID_FEATURES);
 899}
 900
 901unsigned int kvm_arch_para_hints(void)
 902{
 903	return cpuid_edx(kvm_cpuid_base() | KVM_CPUID_FEATURES);
 904}
 905EXPORT_SYMBOL_GPL(kvm_arch_para_hints);
 906
 907static uint32_t __init kvm_detect(void)
 908{
 909	return kvm_cpuid_base();
 910}
 911
 912static void __init kvm_apic_init(void)
 913{
 914#ifdef CONFIG_SMP
 915	if (pv_ipi_supported())
 916		kvm_setup_pv_ipi();
 917#endif
 918}
 919
 920static bool __init kvm_msi_ext_dest_id(void)
 921{
 922	return kvm_para_has_feature(KVM_FEATURE_MSI_EXT_DEST_ID);
 923}
 924
 925static void kvm_sev_hc_page_enc_status(unsigned long pfn, int npages, bool enc)
 926{
 927	kvm_sev_hypercall3(KVM_HC_MAP_GPA_RANGE, pfn << PAGE_SHIFT, npages,
 928			   KVM_MAP_GPA_RANGE_ENC_STAT(enc) | KVM_MAP_GPA_RANGE_PAGE_SZ_4K);
 929}
 930
 931static void __init kvm_init_platform(void)
 932{
 933	if (cc_platform_has(CC_ATTR_GUEST_MEM_ENCRYPT) &&
 934	    kvm_para_has_feature(KVM_FEATURE_MIGRATION_CONTROL)) {
 935		unsigned long nr_pages;
 936		int i;
 937
 938		pv_ops.mmu.notify_page_enc_status_changed =
 939			kvm_sev_hc_page_enc_status;
 940
 941		/*
 942		 * Reset the host's shared pages list related to kernel
 943		 * specific page encryption status settings before we load a
 944		 * new kernel by kexec. Reset the page encryption status
 945		 * during early boot intead of just before kexec to avoid SMP
 946		 * races during kvm_pv_guest_cpu_reboot().
 947		 * NOTE: We cannot reset the complete shared pages list
 948		 * here as we need to retain the UEFI/OVMF firmware
 949		 * specific settings.
 950		 */
 951
 952		for (i = 0; i < e820_table->nr_entries; i++) {
 953			struct e820_entry *entry = &e820_table->entries[i];
 954
 955			if (entry->type != E820_TYPE_RAM)
 956				continue;
 957
 958			nr_pages = DIV_ROUND_UP(entry->size, PAGE_SIZE);
 959
 960			kvm_sev_hypercall3(KVM_HC_MAP_GPA_RANGE, entry->addr,
 961				       nr_pages,
 962				       KVM_MAP_GPA_RANGE_ENCRYPTED | KVM_MAP_GPA_RANGE_PAGE_SZ_4K);
 963		}
 964
 965		/*
 966		 * Ensure that _bss_decrypted section is marked as decrypted in the
 967		 * shared pages list.
 968		 */
 969		nr_pages = DIV_ROUND_UP(__end_bss_decrypted - __start_bss_decrypted,
 970					PAGE_SIZE);
 971		early_set_mem_enc_dec_hypercall((unsigned long)__start_bss_decrypted,
 972						nr_pages, 0);
 973
 974		/*
 975		 * If not booted using EFI, enable Live migration support.
 976		 */
 977		if (!efi_enabled(EFI_BOOT))
 978			wrmsrl(MSR_KVM_MIGRATION_CONTROL,
 979			       KVM_MIGRATION_READY);
 980	}
 981	kvmclock_init();
 982	x86_platform.apic_post_init = kvm_apic_init;
 983}
 984
 985#if defined(CONFIG_AMD_MEM_ENCRYPT)
 986static void kvm_sev_es_hcall_prepare(struct ghcb *ghcb, struct pt_regs *regs)
 987{
 988	/* RAX and CPL are already in the GHCB */
 989	ghcb_set_rbx(ghcb, regs->bx);
 990	ghcb_set_rcx(ghcb, regs->cx);
 991	ghcb_set_rdx(ghcb, regs->dx);
 992	ghcb_set_rsi(ghcb, regs->si);
 993}
 994
 995static bool kvm_sev_es_hcall_finish(struct ghcb *ghcb, struct pt_regs *regs)
 996{
 997	/* No checking of the return state needed */
 998	return true;
 999}
1000#endif
1001
1002const __initconst struct hypervisor_x86 x86_hyper_kvm = {
1003	.name				= "KVM",
1004	.detect				= kvm_detect,
1005	.type				= X86_HYPER_KVM,
1006	.init.guest_late_init		= kvm_guest_init,
1007	.init.x2apic_available		= kvm_para_available,
1008	.init.msi_ext_dest_id		= kvm_msi_ext_dest_id,
1009	.init.init_platform		= kvm_init_platform,
1010#if defined(CONFIG_AMD_MEM_ENCRYPT)
1011	.runtime.sev_es_hcall_prepare	= kvm_sev_es_hcall_prepare,
1012	.runtime.sev_es_hcall_finish	= kvm_sev_es_hcall_finish,
1013#endif
1014};
1015
1016static __init int activate_jump_labels(void)
1017{
1018	if (has_steal_clock) {
1019		static_key_slow_inc(&paravirt_steal_enabled);
1020		if (steal_acc)
1021			static_key_slow_inc(&paravirt_steal_rq_enabled);
1022	}
1023
1024	return 0;
1025}
1026arch_initcall(activate_jump_labels);
1027
1028#ifdef CONFIG_PARAVIRT_SPINLOCKS
1029
1030/* Kick a cpu by its apicid. Used to wake up a halted vcpu */
1031static void kvm_kick_cpu(int cpu)
1032{
1033	int apicid;
1034	unsigned long flags = 0;
1035
1036	apicid = per_cpu(x86_cpu_to_apicid, cpu);
1037	kvm_hypercall2(KVM_HC_KICK_CPU, flags, apicid);
1038}
1039
1040#include <asm/qspinlock.h>
1041
1042static void kvm_wait(u8 *ptr, u8 val)
1043{
1044	if (in_nmi())
1045		return;
1046
1047	/*
1048	 * halt until it's our turn and kicked. Note that we do safe halt
1049	 * for irq enabled case to avoid hang when lock info is overwritten
1050	 * in irq spinlock slowpath and no spurious interrupt occur to save us.
1051	 */
1052	if (irqs_disabled()) {
1053		if (READ_ONCE(*ptr) == val)
1054			halt();
1055	} else {
1056		local_irq_disable();
1057
1058		/* safe_halt() will enable IRQ */
1059		if (READ_ONCE(*ptr) == val)
1060			safe_halt();
1061		else
1062			local_irq_enable();
1063	}
1064}
1065
1066/*
1067 * Setup pv_lock_ops to exploit KVM_FEATURE_PV_UNHALT if present.
1068 */
1069void __init kvm_spinlock_init(void)
1070{
1071	/*
1072	 * In case host doesn't support KVM_FEATURE_PV_UNHALT there is still an
1073	 * advantage of keeping virt_spin_lock_key enabled: virt_spin_lock() is
1074	 * preferred over native qspinlock when vCPU is preempted.
1075	 */
1076	if (!kvm_para_has_feature(KVM_FEATURE_PV_UNHALT)) {
1077		pr_info("PV spinlocks disabled, no host support\n");
1078		return;
1079	}
1080
1081	/*
1082	 * Disable PV spinlocks and use native qspinlock when dedicated pCPUs
1083	 * are available.
1084	 */
1085	if (kvm_para_has_hint(KVM_HINTS_REALTIME)) {
1086		pr_info("PV spinlocks disabled with KVM_HINTS_REALTIME hints\n");
1087		goto out;
1088	}
1089
1090	if (num_possible_cpus() == 1) {
1091		pr_info("PV spinlocks disabled, single CPU\n");
1092		goto out;
1093	}
1094
1095	if (nopvspin) {
1096		pr_info("PV spinlocks disabled, forced by \"nopvspin\" parameter\n");
1097		goto out;
1098	}
1099
1100	pr_info("PV spinlocks enabled\n");
1101
1102	__pv_init_lock_hash();
1103	pv_ops.lock.queued_spin_lock_slowpath = __pv_queued_spin_lock_slowpath;
1104	pv_ops.lock.queued_spin_unlock =
1105		PV_CALLEE_SAVE(__pv_queued_spin_unlock);
1106	pv_ops.lock.wait = kvm_wait;
1107	pv_ops.lock.kick = kvm_kick_cpu;
1108
1109	/*
1110	 * When PV spinlock is enabled which is preferred over
1111	 * virt_spin_lock(), virt_spin_lock_key's value is meaningless.
1112	 * Just disable it anyway.
1113	 */
1114out:
1115	static_branch_disable(&virt_spin_lock_key);
1116}
1117
1118#endif	/* CONFIG_PARAVIRT_SPINLOCKS */
1119
1120#ifdef CONFIG_ARCH_CPUIDLE_HALTPOLL
1121
1122static void kvm_disable_host_haltpoll(void *i)
1123{
1124	wrmsrl(MSR_KVM_POLL_CONTROL, 0);
1125}
1126
1127static void kvm_enable_host_haltpoll(void *i)
1128{
1129	wrmsrl(MSR_KVM_POLL_CONTROL, 1);
1130}
1131
1132void arch_haltpoll_enable(unsigned int cpu)
1133{
1134	if (!kvm_para_has_feature(KVM_FEATURE_POLL_CONTROL)) {
1135		pr_err_once("host does not support poll control\n");
1136		pr_err_once("host upgrade recommended\n");
1137		return;
1138	}
1139
1140	/* Enable guest halt poll disables host halt poll */
1141	smp_call_function_single(cpu, kvm_disable_host_haltpoll, NULL, 1);
1142}
1143EXPORT_SYMBOL_GPL(arch_haltpoll_enable);
1144
1145void arch_haltpoll_disable(unsigned int cpu)
1146{
1147	if (!kvm_para_has_feature(KVM_FEATURE_POLL_CONTROL))
1148		return;
1149
1150	/* Disable guest halt poll enables host halt poll */
1151	smp_call_function_single(cpu, kvm_enable_host_haltpoll, NULL, 1);
1152}
1153EXPORT_SYMBOL_GPL(arch_haltpoll_disable);
1154#endif