[linux-2.6-block.git] / virt / kvm / async_pf.c

// SPDX-License-Identifier: GPL-2.0-only
/*
 * kvm asynchronous fault support
 *
 * Copyright 2010 Red Hat, Inc.
 *
 * Author:
 *      Gleb Natapov <gleb@redhat.com>
 */

#include <linux/kvm_host.h>
#include <linux/slab.h>
#include <linux/module.h>
#include <linux/mmu_context.h>
#include <linux/sched/mm.h>

#include "async_pf.h"
#include <trace/events/kvm.h>

static struct kmem_cache *async_pf_cache;

int kvm_async_pf_init(void)
{
	async_pf_cache = KMEM_CACHE(kvm_async_pf, 0);

	if (!async_pf_cache)
		return -ENOMEM;

	return 0;
}

void kvm_async_pf_deinit(void)
{
	kmem_cache_destroy(async_pf_cache);
	async_pf_cache = NULL;
}

void kvm_async_pf_vcpu_init(struct kvm_vcpu *vcpu)
{
	INIT_LIST_HEAD(&vcpu->async_pf.done);
	INIT_LIST_HEAD(&vcpu->async_pf.queue);
	spin_lock_init(&vcpu->async_pf.lock);
}

static void async_pf_execute(struct work_struct *work)
{
	struct kvm_async_pf *apf =
		container_of(work, struct kvm_async_pf, work);
	struct mm_struct *mm = apf->mm;
	struct kvm_vcpu *vcpu = apf->vcpu;
	unsigned long addr = apf->addr;
	gpa_t cr2_or_gpa = apf->cr2_or_gpa;
	int locked = 1;
	bool first;

	might_sleep();

	/*
	 * This work is run asynchronously to the task which owns
	 * mm and might be done in another context, so we must
	 * access remotely.
	 */
	mmap_read_lock(mm);
	get_user_pages_remote(mm, addr, 1, FOLL_WRITE, NULL, &locked);
	if (locked)
		mmap_read_unlock(mm);
	mmput(mm);

	if (IS_ENABLED(CONFIG_KVM_ASYNC_PF_SYNC))
		kvm_arch_async_page_present(vcpu, apf);

	spin_lock(&vcpu->async_pf.lock);
	first = list_empty(&vcpu->async_pf.done);
	list_add_tail(&apf->link, &vcpu->async_pf.done);
	apf->vcpu = NULL;
	spin_unlock(&vcpu->async_pf.lock);

	if (!IS_ENABLED(CONFIG_KVM_ASYNC_PF_SYNC) && first)
		kvm_arch_async_page_present_queued(vcpu);

	/*
	 * apf may be freed by kvm_check_async_pf_completion() after
	 * this point
	 */

	trace_kvm_async_pf_completed(addr, cr2_or_gpa);

	__kvm_vcpu_wake_up(vcpu);
}

static void kvm_flush_and_free_async_pf_work(struct kvm_async_pf *work)
{
	/*
	 * The async #PF is "done", but KVM must wait for the work item itself,
	 * i.e. async_pf_execute(), to run to completion.  If KVM is a module,
	 * KVM must ensure *no* code owned by the KVM (the module) can be run
	 * after the last call to module_put().  Note, flushing the work item
	 * is always required when the item is taken off the completion queue.
	 * E.g. even if the vCPU handles the item in the "normal" path, the VM
	 * could be terminated before async_pf_execute() completes.
	 *
	 * Wake all events skip the queue and go straight done, i.e. don't
	 * need to be flushed (but sanity check that the work wasn't queued).
	 */
	if (work->wakeup_all)
		WARN_ON_ONCE(work->work.func);
	else
		flush_work(&work->work);
	kmem_cache_free(async_pf_cache, work);
}

void kvm_clear_async_pf_completion_queue(struct kvm_vcpu *vcpu)
{
	spin_lock(&vcpu->async_pf.lock);

	/* cancel outstanding work queue item */
	while (!list_empty(&vcpu->async_pf.queue)) {
		struct kvm_async_pf *work =
			list_first_entry(&vcpu->async_pf.queue,
					 typeof(*work), queue);
		list_del(&work->queue);

		/*
		 * We know it's present in vcpu->async_pf.done, do
		 * nothing here.
		 */
		if (!work->vcpu)
			continue;

		spin_unlock(&vcpu->async_pf.lock);
#ifdef CONFIG_KVM_ASYNC_PF_SYNC
		flush_work(&work->work);
#else
		if (cancel_work_sync(&work->work)) {
			mmput(work->mm);
			kmem_cache_free(async_pf_cache, work);
		}
#endif
		spin_lock(&vcpu->async_pf.lock);
	}

	while (!list_empty(&vcpu->async_pf.done)) {
		struct kvm_async_pf *work =
			list_first_entry(&vcpu->async_pf.done,
					 typeof(*work), link);
		list_del(&work->link);

		spin_unlock(&vcpu->async_pf.lock);
		kvm_flush_and_free_async_pf_work(work);
		spin_lock(&vcpu->async_pf.lock);
	}
	spin_unlock(&vcpu->async_pf.lock);

	vcpu->async_pf.queued = 0;
}

void kvm_check_async_pf_completion(struct kvm_vcpu *vcpu)
{
	struct kvm_async_pf *work;

	while (!list_empty_careful(&vcpu->async_pf.done) &&
	      kvm_arch_can_dequeue_async_page_present(vcpu)) {
		spin_lock(&vcpu->async_pf.lock);
		work = list_first_entry(&vcpu->async_pf.done, typeof(*work),
					      link);
		list_del(&work->link);
		spin_unlock(&vcpu->async_pf.lock);

		kvm_arch_async_page_ready(vcpu, work);
		if (!IS_ENABLED(CONFIG_KVM_ASYNC_PF_SYNC))
			kvm_arch_async_page_present(vcpu, work);

		list_del(&work->queue);
		vcpu->async_pf.queued--;
		kvm_flush_and_free_async_pf_work(work);
	}
}

/*
 * Try to schedule a job to handle page fault asynchronously. Returns 'true' on
 * success, 'false' on failure (page fault has to be handled synchronously).
 */
bool kvm_setup_async_pf(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
			unsigned long hva, struct kvm_arch_async_pf *arch)
{
	struct kvm_async_pf *work;

	if (vcpu->async_pf.queued >= ASYNC_PF_PER_VCPU)
		return false;

	/* Arch specific code should not do async PF in this case */
	if (unlikely(kvm_is_error_hva(hva)))
		return false;

	/*
	 * do alloc nowait since if we are going to sleep anyway we
	 * may as well sleep faulting in page
	 */
	work = kmem_cache_zalloc(async_pf_cache, GFP_NOWAIT | __GFP_NOWARN);
	if (!work)
		return false;

	work->wakeup_all = false;
	work->vcpu = vcpu;
	work->cr2_or_gpa = cr2_or_gpa;
	work->addr = hva;
	work->arch = *arch;
	work->mm = current->mm;
	mmget(work->mm);

	INIT_WORK(&work->work, async_pf_execute);

	list_add_tail(&work->queue, &vcpu->async_pf.queue);
	vcpu->async_pf.queued++;
	work->notpresent_injected = kvm_arch_async_page_not_present(vcpu, work);

	schedule_work(&work->work);

	return true;
}

int kvm_async_pf_wakeup_all(struct kvm_vcpu *vcpu)
{
	struct kvm_async_pf *work;
	bool first;

	if (!list_empty_careful(&vcpu->async_pf.done))
		return 0;

	work = kmem_cache_zalloc(async_pf_cache, GFP_ATOMIC);
	if (!work)
		return -ENOMEM;

	work->wakeup_all = true;
	INIT_LIST_HEAD(&work->queue); /* for list_del to work */

	spin_lock(&vcpu->async_pf.lock);
	first = list_empty(&vcpu->async_pf.done);
	list_add_tail(&work->link, &vcpu->async_pf.done);
	spin_unlock(&vcpu->async_pf.lock);

	if (!IS_ENABLED(CONFIG_KVM_ASYNC_PF_SYNC) && first)
		kvm_arch_async_page_present_queued(vcpu);

	vcpu->async_pf.queued++;
	return 0;
}
Commit	Line	Data
775c8a3d	1	// SPDX-License-Identifier: GPL-2.0-only
af585b92 GN	2	/*
	3	* kvm asynchronous fault support
	4	*
	5	* Copyright 2010 Red Hat, Inc.
	6	*
	7	* Author:
	8	* Gleb Natapov <gleb@redhat.com>
af585b92 GN	9	*/
	10
	11	#include <linux/kvm_host.h>
	12	#include <linux/slab.h>
	13	#include <linux/module.h>
	14	#include <linux/mmu_context.h>
6e84f315	15	#include <linux/sched/mm.h>
af585b92 GN	16
	17	#include "async_pf.h"
	18	#include <trace/events/kvm.h>
	19
	20	static struct kmem_cache *async_pf_cache;
	21
	22	int kvm_async_pf_init(void)
	23	{
	24	async_pf_cache = KMEM_CACHE(kvm_async_pf, 0);
	25
	26	if (!async_pf_cache)
	27	return -ENOMEM;
	28
	29	return 0;
	30	}
	31
	32	void kvm_async_pf_deinit(void)
	33	{
4f52696a	34	kmem_cache_destroy(async_pf_cache);
af585b92 GN	35	async_pf_cache = NULL;
	36	}
	37
	38	void kvm_async_pf_vcpu_init(struct kvm_vcpu *vcpu)
	39	{
	40	INIT_LIST_HEAD(&vcpu->async_pf.done);
	41	INIT_LIST_HEAD(&vcpu->async_pf.queue);
	42	spin_lock_init(&vcpu->async_pf.lock);
	43	}
	44
	45	static void async_pf_execute(struct work_struct *work)
	46	{
af585b92 GN	47	struct kvm_async_pf *apf =
	48	container_of(work, struct kvm_async_pf, work);
	49	struct mm_struct *mm = apf->mm;
	50	struct kvm_vcpu *vcpu = apf->vcpu;
	51	unsigned long addr = apf->addr;
736c291c	52	gpa_t cr2_or_gpa = apf->cr2_or_gpa;
8b7457ef	53	int locked = 1;
557a961a	54	bool first;
af585b92 GN	55
	56	might_sleep();
	57
1e987790	58	/*
bdd303cb	59	* This work is run asynchronously to the task which owns
1e987790	60	* mm and might be done in another context, so we must
8b7457ef	61	* access remotely.
1e987790	62	*/
d8ed45c5	63	mmap_read_lock(mm);
ca5e8632	64	get_user_pages_remote(mm, addr, 1, FOLL_WRITE, NULL, &locked);
8b7457ef	65	if (locked)
d8ed45c5	66	mmap_read_unlock(mm);
422eeb54	67	mmput(mm);
1e987790	68
4425f567 PB	69	if (IS_ENABLED(CONFIG_KVM_ASYNC_PF_SYNC))
4425f567 PB	70	kvm_arch_async_page_present(vcpu, apf);
af585b92 GN	71
af585b92 GN	72	spin_lock(&vcpu->async_pf.lock);
557a961a	73	first = list_empty(&vcpu->async_pf.done);
af585b92	74	list_add_tail(&apf->link, &vcpu->async_pf.done);
22583f0d	75	apf->vcpu = NULL;
af585b92 GN	76	spin_unlock(&vcpu->async_pf.lock);
af585b92 GN	77
557a961a VK	78	if (!IS_ENABLED(CONFIG_KVM_ASYNC_PF_SYNC) && first)
	79	kvm_arch_async_page_present_queued(vcpu);
	80
af585b92 GN	81	/*
	82	* apf may be freed by kvm_check_async_pf_completion() after
	83	* this point
	84	*/
	85
736c291c	86	trace_kvm_async_pf_completed(addr, cr2_or_gpa);
af585b92	87
d92a5d1c	88	__kvm_vcpu_wake_up(vcpu);
3d75b8aa SC	89	}
	90
	91	static void kvm_flush_and_free_async_pf_work(struct kvm_async_pf *work)
	92	{
	93	/*
	94	* The async #PF is "done", but KVM must wait for the work item itself,
	95	* i.e. async_pf_execute(), to run to completion. If KVM is a module,
	96	* KVM must ensure no code owned by the KVM (the module) can be run
	97	* after the last call to module_put(). Note, flushing the work item
	98	* is always required when the item is taken off the completion queue.
	99	* E.g. even if the vCPU handles the item in the "normal" path, the VM
	100	* could be terminated before async_pf_execute() completes.
	101	*
	102	* Wake all events skip the queue and go straight done, i.e. don't
	103	* need to be flushed (but sanity check that the work wasn't queued).
	104	*/
	105	if (work->wakeup_all)
	106	WARN_ON_ONCE(work->work.func);
	107	else
	108	flush_work(&work->work);
	109	kmem_cache_free(async_pf_cache, work);
af585b92 GN	110	}
	111
	112	void kvm_clear_async_pf_completion_queue(struct kvm_vcpu *vcpu)
	113	{
22583f0d PB	114	spin_lock(&vcpu->async_pf.lock);
22583f0d PB	115
af585b92 GN	116	/* cancel outstanding work queue item */
	117	while (!list_empty(&vcpu->async_pf.queue)) {
	118	struct kvm_async_pf *work =
433da860 GT	119	list_first_entry(&vcpu->async_pf.queue,
433da860 GT	120	typeof(*work), queue);
af585b92	121	list_del(&work->queue);
9f2ceda4	122
22583f0d PB	123	/*
	124	* We know it's present in vcpu->async_pf.done, do
	125	* nothing here.
	126	*/
	127	if (!work->vcpu)
	128	continue;
	129
	130	spin_unlock(&vcpu->async_pf.lock);
9f2ceda4 DD	131	#ifdef CONFIG_KVM_ASYNC_PF_SYNC
	132	flush_work(&work->work);
	133	#else
98fda169	134	if (cancel_work_sync(&work->work)) {
41c22f62	135	mmput(work->mm);
af585b92	136	kmem_cache_free(async_pf_cache, work);
28b441e2	137	}
9f2ceda4	138	#endif
22583f0d	139	spin_lock(&vcpu->async_pf.lock);
af585b92 GN	140	}
af585b92 GN	141
af585b92 GN	142	while (!list_empty(&vcpu->async_pf.done)) {
af585b92 GN	143	struct kvm_async_pf *work =
433da860 GT	144	list_first_entry(&vcpu->async_pf.done,
433da860 GT	145	typeof(*work), link);
af585b92	146	list_del(&work->link);
3d75b8aa SC	147
	148	spin_unlock(&vcpu->async_pf.lock);
	149	kvm_flush_and_free_async_pf_work(work);
	150	spin_lock(&vcpu->async_pf.lock);
af585b92 GN	151	}
	152	spin_unlock(&vcpu->async_pf.lock);
	153
	154	vcpu->async_pf.queued = 0;
	155	}
	156
	157	void kvm_check_async_pf_completion(struct kvm_vcpu *vcpu)
	158	{
	159	struct kvm_async_pf *work;
	160
15096ffc	161	while (!list_empty_careful(&vcpu->async_pf.done) &&
7c0ade6c	162	kvm_arch_can_dequeue_async_page_present(vcpu)) {
15096ffc XG	163	spin_lock(&vcpu->async_pf.lock);
	164	work = list_first_entry(&vcpu->async_pf.done, typeof(*work),
	165	link);
	166	list_del(&work->link);
	167	spin_unlock(&vcpu->async_pf.lock);
af585b92	168
f2e10669	169	kvm_arch_async_page_ready(vcpu, work);
4425f567 PB	170	if (!IS_ENABLED(CONFIG_KVM_ASYNC_PF_SYNC))
4425f567 PB	171	kvm_arch_async_page_present(vcpu, work);
af585b92	172
15096ffc XG	173	list_del(&work->queue);
15096ffc XG	174	vcpu->async_pf.queued--;
3d75b8aa	175	kvm_flush_and_free_async_pf_work(work);
15096ffc	176	}
af585b92 GN	177	}
af585b92 GN	178
e8c22266 VK	179	/*
	180	* Try to schedule a job to handle page fault asynchronously. Returns 'true' on
	181	* success, 'false' on failure (page fault has to be handled synchronously).
	182	*/
	183	bool kvm_setup_async_pf(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
	184	unsigned long hva, struct kvm_arch_async_pf *arch)
af585b92 GN	185	{
	186	struct kvm_async_pf *work;
	187
	188	if (vcpu->async_pf.queued >= ASYNC_PF_PER_VCPU)
e8c22266	189	return false;
af585b92	190
7863e346 VK	191	/* Arch specific code should not do async PF in this case */
7863e346 VK	192	if (unlikely(kvm_is_error_hva(hva)))
e8c22266	193	return false;
af585b92 GN	194
	195	/*
	196	* do alloc nowait since if we are going to sleep anyway we
	197	* may as well sleep faulting in page
	198	*/
d7444794	199	work = kmem_cache_zalloc(async_pf_cache, GFP_NOWAIT \| __GFP_NOWARN);
af585b92	200	if (!work)
e8c22266	201	return false;
af585b92	202
f2e10669	203	work->wakeup_all = false;
af585b92	204	work->vcpu = vcpu;
736c291c	205	work->cr2_or_gpa = cr2_or_gpa;
e0ead41a	206	work->addr = hva;
af585b92 GN	207	work->arch = *arch;
af585b92 GN	208	work->mm = current->mm;
3fce371b	209	mmget(work->mm);
af585b92	210
af585b92	211	INIT_WORK(&work->work, async_pf_execute);
af585b92 GN	212
	213	list_add_tail(&work->queue, &vcpu->async_pf.queue);
	214	vcpu->async_pf.queued++;
2a18b7e7	215	work->notpresent_injected = kvm_arch_async_page_not_present(vcpu, work);
7863e346 VK	216
	217	schedule_work(&work->work);
	218
e8c22266	219	return true;
af585b92	220	}
344d9588 GN	221
	222	int kvm_async_pf_wakeup_all(struct kvm_vcpu *vcpu)
	223	{
	224	struct kvm_async_pf *work;
557a961a	225	bool first;
344d9588	226
64f638c7	227	if (!list_empty_careful(&vcpu->async_pf.done))
344d9588 GN	228	return 0;
	229
	230	work = kmem_cache_zalloc(async_pf_cache, GFP_ATOMIC);
	231	if (!work)
	232	return -ENOMEM;
	233
f2e10669	234	work->wakeup_all = true;
344d9588 GN	235	INIT_LIST_HEAD(&work->queue); /* for list_del to work */
344d9588 GN	236
64f638c7	237	spin_lock(&vcpu->async_pf.lock);
557a961a	238	first = list_empty(&vcpu->async_pf.done);
344d9588	239	list_add_tail(&work->link, &vcpu->async_pf.done);
64f638c7 XG	240	spin_unlock(&vcpu->async_pf.lock);
64f638c7 XG	241
557a961a VK	242	if (!IS_ENABLED(CONFIG_KVM_ASYNC_PF_SYNC) && first)
	243	kvm_arch_async_page_present_queued(vcpu);
	244
344d9588 GN	245	vcpu->async_pf.queued++;
	246	return 0;
	247	}