blob: c0dff5337a506b2b01dff8486cf0265aad294a23 [file] [log] [blame]
Avi Kivity6aa8b732006-12-10 02:21:36 -08001/*
2 * Kernel-based Virtual Machine driver for Linux
3 *
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
6 *
7 * Copyright (C) 2006 Qumranet, Inc.
Nicolas Kaiser9611c182010-10-06 14:23:22 +02008 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
Avi Kivity6aa8b732006-12-10 02:21:36 -08009 *
10 * Authors:
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
13 *
14 * This work is licensed under the terms of the GNU GPL, version 2. See
15 * the COPYING file in the top-level directory.
16 *
17 */
18
Andre Przywaraaf669ac2015-03-26 14:39:29 +000019#include <kvm/iodev.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080020
Avi Kivityedf88412007-12-16 11:02:48 +020021#include <linux/kvm_host.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080022#include <linux/kvm.h>
23#include <linux/module.h>
24#include <linux/errno.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080025#include <linux/percpu.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080026#include <linux/mm.h>
27#include <linux/miscdevice.h>
28#include <linux/vmalloc.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080029#include <linux/reboot.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080030#include <linux/debugfs.h>
31#include <linux/highmem.h>
32#include <linux/file.h>
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +010033#include <linux/syscore_ops.h>
Avi Kivity774c47f2007-02-12 00:54:47 -080034#include <linux/cpu.h>
Alexey Dobriyane8edc6e2007-05-21 01:22:52 +040035#include <linux/sched.h>
Avi Kivityd9e368d2007-06-07 19:18:30 +030036#include <linux/cpumask.h>
37#include <linux/smp.h>
Avi Kivityd6d28162007-06-28 08:38:16 -040038#include <linux/anon_inodes.h>
Avi Kivity04d2cc72007-09-10 18:10:54 +030039#include <linux/profile.h>
Anthony Liguori7aa81cc2007-09-17 14:57:50 -050040#include <linux/kvm_para.h>
Izik Eidus6fc138d2007-10-09 19:20:39 +020041#include <linux/pagemap.h>
Anthony Liguori8d4e1282007-10-18 09:59:34 -050042#include <linux/mman.h>
Anthony Liguori35149e22008-04-02 14:46:56 -050043#include <linux/swap.h>
Sheng Yange56d5322009-03-12 21:45:39 +080044#include <linux/bitops.h>
Marcelo Tosatti547de292009-05-07 17:55:13 -030045#include <linux/spinlock.h>
Arnd Bergmann6ff58942009-10-22 14:19:27 +020046#include <linux/compat.h>
Marcelo Tosattibc6678a2009-12-23 14:35:21 -020047#include <linux/srcu.h>
Joerg Roedel8f0b1ab2010-01-28 12:37:56 +010048#include <linux/hugetlb.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090049#include <linux/slab.h>
Sasha Levin743eeb02011-07-27 16:00:48 +030050#include <linux/sort.h>
51#include <linux/bsearch.h>
Junaid Shahid61e191b2019-11-04 12:22:02 +010052#include <linux/kthread.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080053
Avi Kivitye4956062007-06-28 14:15:57 -040054#include <asm/processor.h>
Avi Kivitye4956062007-06-28 14:15:57 -040055#include <asm/io.h>
David Matlack2ea75be2014-09-19 16:03:25 -070056#include <asm/ioctl.h>
Avi Kivitye4956062007-06-28 14:15:57 -040057#include <asm/uaccess.h>
Izik Eidus3e021bf2007-11-19 11:16:57 +020058#include <asm/pgtable.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080059
Laurent Vivier5f94c172008-05-30 16:05:54 +020060#include "coalesced_mmio.h"
Gleb Natapovaf585b92010-10-14 11:22:46 +020061#include "async_pf.h"
Paolo Bonzini3c3c29f2014-09-24 13:02:46 +020062#include "vfio.h"
Laurent Vivier5f94c172008-05-30 16:05:54 +020063
Marcelo Tosatti229456f2009-06-17 09:22:14 -030064#define CREATE_TRACE_POINTS
65#include <trace/events/kvm.h>
66
Janosch Frank536a6f82016-05-18 13:26:23 +020067/* Worst case buffer size needed for holding an integer. */
68#define ITOA_MAX_LEN 12
69
Avi Kivity6aa8b732006-12-10 02:21:36 -080070MODULE_AUTHOR("Qumranet");
71MODULE_LICENSE("GPL");
72
David Hildenbrand920552b2015-09-18 12:34:53 +020073/* Architectures should define their poll value according to the halt latency */
74static unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT;
Paolo Bonzinif7819512015-02-04 18:20:58 +010075module_param(halt_poll_ns, uint, S_IRUGO | S_IWUSR);
76
Wanpeng Liaca6ff22015-09-03 22:07:38 +080077/* Default doubles per-vcpu halt_poll_ns. */
78static unsigned int halt_poll_ns_grow = 2;
Christian Borntraeger6b6de682016-02-09 13:47:55 +010079module_param(halt_poll_ns_grow, uint, S_IRUGO | S_IWUSR);
Wanpeng Liaca6ff22015-09-03 22:07:38 +080080
81/* Default resets per-vcpu halt_poll_ns . */
82static unsigned int halt_poll_ns_shrink;
Christian Borntraeger6b6de682016-02-09 13:47:55 +010083module_param(halt_poll_ns_shrink, uint, S_IRUGO | S_IWUSR);
Wanpeng Liaca6ff22015-09-03 22:07:38 +080084
Marcelo Tosattifa40a822009-06-04 15:08:24 -030085/*
86 * Ordering of locks:
87 *
Xiubo Lib7d409d2015-02-26 14:58:24 +080088 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
Marcelo Tosattifa40a822009-06-04 15:08:24 -030089 */
90
Junaid Shahidc6170b82019-01-03 17:14:28 -080091DEFINE_MUTEX(kvm_lock);
Paolo Bonzini4a937f92013-09-10 12:58:35 +020092static DEFINE_RAW_SPINLOCK(kvm_count_lock);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +080093LIST_HEAD(vm_list);
Avi Kivity133de902007-02-12 00:54:44 -080094
Rusty Russell7f59f492008-12-07 21:25:45 +103095static cpumask_var_t cpus_hardware_enabled;
Xiubo Lif4fee932015-02-26 14:58:21 +080096static int kvm_usage_count;
Alexander Graf10474ae2009-09-15 11:37:46 +020097static atomic_t hardware_enable_failed;
Avi Kivity1b6c0162007-05-24 13:03:52 +030098
Rusty Russellc16f8622007-07-30 21:12:19 +100099struct kmem_cache *kvm_vcpu_cache;
100EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
Avi Kivity1165f5f2007-04-19 17:27:43 +0300101
Avi Kivity15ad7142007-07-11 18:17:21 +0300102static __read_mostly struct preempt_ops kvm_preempt_ops;
103
Hollis Blanchard76f7c872008-04-15 16:05:42 -0500104struct dentry *kvm_debugfs_dir;
Paul Mackerrase23a8082015-03-28 14:21:01 +1100105EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800106
Janosch Frank536a6f82016-05-18 13:26:23 +0200107static int kvm_debugfs_num_entries;
108static const struct file_operations *stat_fops_per_vm[];
109
Avi Kivitybccf2152007-02-21 18:04:26 +0200110static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
111 unsigned long arg);
Christian Borntraegerde8e5d72015-02-03 09:35:15 +0100112#ifdef CONFIG_KVM_COMPAT
Alexander Graf1dda6062011-06-08 02:45:37 +0200113static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
114 unsigned long arg);
115#endif
Alexander Graf10474ae2009-09-15 11:37:46 +0200116static int hardware_enable_all(void);
117static void hardware_disable_all(void);
Avi Kivitybccf2152007-02-21 18:04:26 +0200118
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200119static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
Stephen Hemminger79408762013-12-29 12:12:29 -0800120
Dan Williamsba049e92016-01-15 16:56:11 -0800121static void kvm_release_pfn_dirty(kvm_pfn_t pfn);
Paolo Bonzinibc009e42015-05-26 12:43:41 +0200122static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, gfn_t gfn);
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200123
Andi Kleen52480132014-02-08 08:51:57 +0100124__visible bool kvm_rebooting;
Avi Kivityb7c41452010-12-02 17:52:50 +0200125EXPORT_SYMBOL_GPL(kvm_rebooting);
Avi Kivity4ecac3f2008-05-13 13:23:38 +0300126
Marcelo Tosatti54dee992009-06-11 12:07:44 -0300127static bool largepages_enabled = true;
128
Radim Krčmář9cf0eaf2017-11-30 19:05:45 +0100129__weak void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
130 unsigned long start, unsigned long end)
131{
132}
133
Sean Christophersone528acd2019-11-11 14:12:27 -0800134bool kvm_is_zone_device_pfn(kvm_pfn_t pfn)
135{
136 /*
137 * The metadata used by is_zone_device_page() to determine whether or
138 * not a page is ZONE_DEVICE is guaranteed to be valid if and only if
139 * the device has been pinned, e.g. by get_user_pages(). WARN if the
140 * page_count() is zero to help detect bad usage of this helper.
141 */
142 if (!pfn_valid(pfn) || WARN_ON_ONCE(!page_count(pfn_to_page(pfn))))
143 return false;
144
145 return is_zone_device_page(pfn_to_page(pfn));
146}
147
Dan Williamsba049e92016-01-15 16:56:11 -0800148bool kvm_is_reserved_pfn(kvm_pfn_t pfn)
Ben-Ami Yassourcbff90a2008-07-28 19:26:24 +0300149{
Sean Christophersone528acd2019-11-11 14:12:27 -0800150 /*
151 * ZONE_DEVICE pages currently set PG_reserved, but from a refcounting
152 * perspective they are "normal" pages, albeit with slightly different
153 * usage rules.
154 */
Andrea Arcangeli11feeb42013-07-25 03:04:38 +0200155 if (pfn_valid(pfn))
Sean Christophersone528acd2019-11-11 14:12:27 -0800156 return PageReserved(pfn_to_page(pfn)) &&
157 !kvm_is_zone_device_pfn(pfn);
Ben-Ami Yassourcbff90a2008-07-28 19:26:24 +0300158
159 return true;
160}
161
Avi Kivity6aa8b732006-12-10 02:21:36 -0800162/*
163 * Switches to specified vcpu, until a matching vcpu_put()
164 */
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +0300165int vcpu_load(struct kvm_vcpu *vcpu)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800166{
Avi Kivity15ad7142007-07-11 18:17:21 +0300167 int cpu;
168
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +0300169 if (mutex_lock_killable(&vcpu->mutex))
170 return -EINTR;
Avi Kivity15ad7142007-07-11 18:17:21 +0300171 cpu = get_cpu();
172 preempt_notifier_register(&vcpu->preempt_notifier);
Carsten Otte313a3dc2007-10-11 19:16:52 +0200173 kvm_arch_vcpu_load(vcpu, cpu);
Avi Kivity15ad7142007-07-11 18:17:21 +0300174 put_cpu();
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +0300175 return 0;
Avi Kivitybccf2152007-02-21 18:04:26 +0200176}
Jim Mattson2f1fe812016-07-08 15:36:06 -0700177EXPORT_SYMBOL_GPL(vcpu_load);
Avi Kivitybccf2152007-02-21 18:04:26 +0200178
Carsten Otte313a3dc2007-10-11 19:16:52 +0200179void vcpu_put(struct kvm_vcpu *vcpu)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800180{
Avi Kivity15ad7142007-07-11 18:17:21 +0300181 preempt_disable();
Carsten Otte313a3dc2007-10-11 19:16:52 +0200182 kvm_arch_vcpu_put(vcpu);
Avi Kivity15ad7142007-07-11 18:17:21 +0300183 preempt_notifier_unregister(&vcpu->preempt_notifier);
184 preempt_enable();
Avi Kivity6aa8b732006-12-10 02:21:36 -0800185 mutex_unlock(&vcpu->mutex);
186}
Jim Mattson2f1fe812016-07-08 15:36:06 -0700187EXPORT_SYMBOL_GPL(vcpu_put);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800188
Avi Kivityd9e368d2007-06-07 19:18:30 +0300189static void ack_flush(void *_completed)
190{
Avi Kivityd9e368d2007-06-07 19:18:30 +0300191}
192
Tang Chen445b8232014-09-24 15:57:55 +0800193bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
Avi Kivityd9e368d2007-06-07 19:18:30 +0300194{
Avi Kivity597a5f52008-07-20 14:24:22 +0300195 int i, cpu, me;
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030196 cpumask_var_t cpus;
197 bool called = true;
Avi Kivityd9e368d2007-06-07 19:18:30 +0300198 struct kvm_vcpu *vcpu;
Avi Kivityd9e368d2007-06-07 19:18:30 +0300199
Li Zefan79f55992009-06-15 14:58:26 +0800200 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030201
Xiao Guangrong3cba4132011-01-12 15:41:22 +0800202 me = get_cpu();
Gleb Natapov988a2ca2009-06-09 15:56:29 +0300203 kvm_for_each_vcpu(i, vcpu, kvm) {
Xiao Guangrong3cba4132011-01-12 15:41:22 +0800204 kvm_make_request(req, vcpu);
Avi Kivityd9e368d2007-06-07 19:18:30 +0300205 cpu = vcpu->cpu;
Xiao Guangrong6b7e2d02011-01-12 15:40:31 +0800206
Lan Tianyua30a0502016-03-13 11:10:26 +0800207 /* Set ->requests bit before we read ->mode. */
208 smp_mb__after_atomic();
Xiao Guangrong6b7e2d02011-01-12 15:40:31 +0800209
210 if (cpus != NULL && cpu != -1 && cpu != me &&
211 kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030212 cpumask_set_cpu(cpu, cpus);
Avi Kivityd9e368d2007-06-07 19:18:30 +0300213 }
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030214 if (unlikely(cpus == NULL))
215 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
216 else if (!cpumask_empty(cpus))
217 smp_call_function_many(cpus, ack_flush, NULL, 1);
218 else
219 called = false;
Xiao Guangrong3cba4132011-01-12 15:41:22 +0800220 put_cpu();
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030221 free_cpumask_var(cpus);
Rusty Russell49846892008-12-08 20:26:24 +1030222 return called;
223}
224
Mario Smarducha6d51012015-01-15 15:58:52 -0800225#ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
Rusty Russell49846892008-12-08 20:26:24 +1030226void kvm_flush_remote_tlbs(struct kvm *kvm)
227{
Lan Tianyu4ae3cb32016-03-13 11:10:28 +0800228 /*
229 * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
230 * kvm_make_all_cpus_request.
231 */
232 long dirty_count = smp_load_acquire(&kvm->tlbs_dirty);
Xiao Guangronga086f6a2014-04-17 17:06:12 +0800233
Lan Tianyu4ae3cb32016-03-13 11:10:28 +0800234 /*
235 * We want to publish modifications to the page tables before reading
236 * mode. Pairs with a memory barrier in arch-specific code.
237 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
238 * and smp_mb in walk_shadow_page_lockless_begin/end.
239 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
240 *
241 * There is already an smp_mb__after_atomic() before
242 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
243 * barrier here.
244 */
Tang Chen445b8232014-09-24 15:57:55 +0800245 if (kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
Rusty Russell49846892008-12-08 20:26:24 +1030246 ++kvm->stat.remote_tlb_flush;
Xiao Guangronga086f6a2014-04-17 17:06:12 +0800247 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
Avi Kivityd9e368d2007-06-07 19:18:30 +0300248}
Aneesh Kumar K.V2ba9f0d2013-10-07 22:17:59 +0530249EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
Mario Smarducha6d51012015-01-15 15:58:52 -0800250#endif
Avi Kivityd9e368d2007-06-07 19:18:30 +0300251
Marcelo Tosatti2e53d632008-02-20 14:47:24 -0500252void kvm_reload_remote_mmus(struct kvm *kvm)
253{
Tang Chen445b8232014-09-24 15:57:55 +0800254 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
Marcelo Tosatti2e53d632008-02-20 14:47:24 -0500255}
256
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000257int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
258{
259 struct page *page;
260 int r;
261
262 mutex_init(&vcpu->mutex);
263 vcpu->cpu = -1;
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000264 vcpu->kvm = kvm;
265 vcpu->vcpu_id = id;
Rik van Riel34bb10b2011-02-01 09:52:41 -0500266 vcpu->pid = NULL;
Marcelo Tosatti85773702016-02-19 09:46:39 +0100267 init_swait_queue_head(&vcpu->wq);
Gleb Natapovaf585b92010-10-14 11:22:46 +0200268 kvm_async_pf_vcpu_init(vcpu);
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000269
Feng Wubf9f6ac2015-09-18 22:29:55 +0800270 vcpu->pre_pcpu = -1;
271 INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
272
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000273 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
274 if (!page) {
275 r = -ENOMEM;
276 goto fail;
277 }
278 vcpu->run = page_address(page);
279
Raghavendra K T4c088492012-07-18 19:07:46 +0530280 kvm_vcpu_set_in_spin_loop(vcpu, false);
281 kvm_vcpu_set_dy_eligible(vcpu, false);
Raghavendra K T3a08a8f2013-03-04 23:32:07 +0530282 vcpu->preempted = false;
Raghavendra K T4c088492012-07-18 19:07:46 +0530283
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +0800284 r = kvm_arch_vcpu_init(vcpu);
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000285 if (r < 0)
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +0800286 goto fail_free_run;
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000287 return 0;
288
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000289fail_free_run:
290 free_page((unsigned long)vcpu->run);
291fail:
Rusty Russell76fafa52007-10-08 10:50:48 +1000292 return r;
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000293}
294EXPORT_SYMBOL_GPL(kvm_vcpu_init);
295
296void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
297{
Rik van Riel34bb10b2011-02-01 09:52:41 -0500298 put_pid(vcpu->pid);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +0800299 kvm_arch_vcpu_uninit(vcpu);
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000300 free_page((unsigned long)vcpu->run);
301}
302EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
303
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200304#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
305static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
306{
307 return container_of(mn, struct kvm, mmu_notifier);
308}
309
310static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
311 struct mm_struct *mm,
312 unsigned long address)
313{
314 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200315 int need_tlb_flush, idx;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200316
317 /*
318 * When ->invalidate_page runs, the linux pte has been zapped
319 * already but the page is still allocated until
320 * ->invalidate_page returns. So if we increase the sequence
321 * here the kvm page fault will notice if the spte can't be
322 * established because the page is going to be freed. If
323 * instead the kvm page fault establishes the spte before
324 * ->invalidate_page runs, kvm_unmap_hva will release it
325 * before returning.
326 *
327 * The sequence increase only need to be seen at spin_unlock
328 * time, and not at spin_lock time.
329 *
330 * Increasing the sequence after the spin_unlock would be
331 * unsafe because the kvm page fault could then establish the
332 * pte after kvm_unmap_hva returned, without noticing the page
333 * is going to be freed.
334 */
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200335 idx = srcu_read_lock(&kvm->srcu);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200336 spin_lock(&kvm->mmu_lock);
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900337
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200338 kvm->mmu_notifier_seq++;
Xiao Guangronga4ee1ca2010-11-23 11:13:00 +0800339 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200340 /* we've to flush the tlb before the pages can be freed */
341 if (need_tlb_flush)
342 kvm_flush_remote_tlbs(kvm);
343
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900344 spin_unlock(&kvm->mmu_lock);
Tang Chenfe715572014-09-24 15:57:57 +0800345
346 kvm_arch_mmu_notifier_invalidate_page(kvm, address);
347
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900348 srcu_read_unlock(&kvm->srcu, idx);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200349}
350
Izik Eidus3da0dd42009-09-23 21:47:18 +0300351static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
352 struct mm_struct *mm,
353 unsigned long address,
354 pte_t pte)
355{
356 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200357 int idx;
Izik Eidus3da0dd42009-09-23 21:47:18 +0300358
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200359 idx = srcu_read_lock(&kvm->srcu);
Izik Eidus3da0dd42009-09-23 21:47:18 +0300360 spin_lock(&kvm->mmu_lock);
361 kvm->mmu_notifier_seq++;
362 kvm_set_spte_hva(kvm, address, pte);
363 spin_unlock(&kvm->mmu_lock);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200364 srcu_read_unlock(&kvm->srcu, idx);
Izik Eidus3da0dd42009-09-23 21:47:18 +0300365}
366
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200367static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
368 struct mm_struct *mm,
369 unsigned long start,
370 unsigned long end)
371{
372 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200373 int need_tlb_flush = 0, idx;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200374
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200375 idx = srcu_read_lock(&kvm->srcu);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200376 spin_lock(&kvm->mmu_lock);
377 /*
378 * The count increase must become visible at unlock time as no
379 * spte can be established without taking the mmu_lock and
380 * count is also read inside the mmu_lock critical section.
381 */
382 kvm->mmu_notifier_count++;
Takuya Yoshikawab3ae2092012-07-02 17:56:33 +0900383 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
Xiao Guangronga4ee1ca2010-11-23 11:13:00 +0800384 need_tlb_flush |= kvm->tlbs_dirty;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200385 /* we've to flush the tlb before the pages can be freed */
386 if (need_tlb_flush)
387 kvm_flush_remote_tlbs(kvm);
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900388
389 spin_unlock(&kvm->mmu_lock);
Radim Krčmář9cf0eaf2017-11-30 19:05:45 +0100390
391 kvm_arch_mmu_notifier_invalidate_range(kvm, start, end);
392
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900393 srcu_read_unlock(&kvm->srcu, idx);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200394}
395
396static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
397 struct mm_struct *mm,
398 unsigned long start,
399 unsigned long end)
400{
401 struct kvm *kvm = mmu_notifier_to_kvm(mn);
402
403 spin_lock(&kvm->mmu_lock);
404 /*
405 * This sequence increase will notify the kvm page fault that
406 * the page that is going to be mapped in the spte could have
407 * been freed.
408 */
409 kvm->mmu_notifier_seq++;
Paul Mackerrasa355aa52011-12-12 12:37:21 +0000410 smp_wmb();
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200411 /*
412 * The above sequence increase must be visible before the
Paul Mackerrasa355aa52011-12-12 12:37:21 +0000413 * below count decrease, which is ensured by the smp_wmb above
414 * in conjunction with the smp_rmb in mmu_notifier_retry().
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200415 */
416 kvm->mmu_notifier_count--;
417 spin_unlock(&kvm->mmu_lock);
418
419 BUG_ON(kvm->mmu_notifier_count < 0);
420}
421
422static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
423 struct mm_struct *mm,
Andres Lagar-Cavilla57128462014-09-22 14:54:42 -0700424 unsigned long start,
425 unsigned long end)
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200426{
427 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200428 int young, idx;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200429
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200430 idx = srcu_read_lock(&kvm->srcu);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200431 spin_lock(&kvm->mmu_lock);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200432
Andres Lagar-Cavilla57128462014-09-22 14:54:42 -0700433 young = kvm_age_hva(kvm, start, end);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200434 if (young)
435 kvm_flush_remote_tlbs(kvm);
436
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900437 spin_unlock(&kvm->mmu_lock);
438 srcu_read_unlock(&kvm->srcu, idx);
439
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200440 return young;
441}
442
Vladimir Davydov1d7715c2015-09-09 15:35:41 -0700443static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
444 struct mm_struct *mm,
445 unsigned long start,
446 unsigned long end)
447{
448 struct kvm *kvm = mmu_notifier_to_kvm(mn);
449 int young, idx;
450
451 idx = srcu_read_lock(&kvm->srcu);
452 spin_lock(&kvm->mmu_lock);
453 /*
454 * Even though we do not flush TLB, this will still adversely
455 * affect performance on pre-Haswell Intel EPT, where there is
456 * no EPT Access Bit to clear so that we have to tear down EPT
457 * tables instead. If we find this unacceptable, we can always
458 * add a parameter to kvm_age_hva so that it effectively doesn't
459 * do anything on clear_young.
460 *
461 * Also note that currently we never issue secondary TLB flushes
462 * from clear_young, leaving this job up to the regular system
463 * cadence. If we find this inaccurate, we might come up with a
464 * more sophisticated heuristic later.
465 */
466 young = kvm_age_hva(kvm, start, end);
467 spin_unlock(&kvm->mmu_lock);
468 srcu_read_unlock(&kvm->srcu, idx);
469
470 return young;
471}
472
Andrea Arcangeli8ee53822011-01-13 15:47:10 -0800473static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
474 struct mm_struct *mm,
475 unsigned long address)
476{
477 struct kvm *kvm = mmu_notifier_to_kvm(mn);
478 int young, idx;
479
480 idx = srcu_read_lock(&kvm->srcu);
481 spin_lock(&kvm->mmu_lock);
482 young = kvm_test_age_hva(kvm, address);
483 spin_unlock(&kvm->mmu_lock);
484 srcu_read_unlock(&kvm->srcu, idx);
485
486 return young;
487}
488
Marcelo Tosatti85db06e2008-12-10 21:23:26 +0100489static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
490 struct mm_struct *mm)
491{
492 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Lai Jiangshaneda2bed2010-04-20 14:29:29 +0800493 int idx;
494
495 idx = srcu_read_lock(&kvm->srcu);
Marcelo Tosatti2df72e92012-08-24 15:54:57 -0300496 kvm_arch_flush_shadow_all(kvm);
Lai Jiangshaneda2bed2010-04-20 14:29:29 +0800497 srcu_read_unlock(&kvm->srcu, idx);
Marcelo Tosatti85db06e2008-12-10 21:23:26 +0100498}
499
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200500static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
501 .invalidate_page = kvm_mmu_notifier_invalidate_page,
502 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
503 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
504 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
Vladimir Davydov1d7715c2015-09-09 15:35:41 -0700505 .clear_young = kvm_mmu_notifier_clear_young,
Andrea Arcangeli8ee53822011-01-13 15:47:10 -0800506 .test_young = kvm_mmu_notifier_test_young,
Izik Eidus3da0dd42009-09-23 21:47:18 +0300507 .change_pte = kvm_mmu_notifier_change_pte,
Marcelo Tosatti85db06e2008-12-10 21:23:26 +0100508 .release = kvm_mmu_notifier_release,
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200509};
Avi Kivity4c07b0a2009-12-20 14:54:04 +0200510
511static int kvm_init_mmu_notifier(struct kvm *kvm)
512{
513 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
514 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
515}
516
517#else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
518
519static int kvm_init_mmu_notifier(struct kvm *kvm)
520{
521 return 0;
522}
523
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200524#endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
525
Paolo Bonzinia47d2b02015-05-17 11:41:37 +0200526static struct kvm_memslots *kvm_alloc_memslots(void)
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800527{
528 int i;
Paolo Bonzinia47d2b02015-05-17 11:41:37 +0200529 struct kvm_memslots *slots;
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800530
Paolo Bonzinia47d2b02015-05-17 11:41:37 +0200531 slots = kvm_kvzalloc(sizeof(struct kvm_memslots));
532 if (!slots)
533 return NULL;
534
535 /*
536 * Init kvm generation close to the maximum to easily test the
537 * code of handling generation number wrap-around.
538 */
539 slots->generation = -150;
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800540 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
Xiao Guangrongf85e2cb2011-11-24 17:41:54 +0800541 slots->id_to_index[i] = slots->memslots[i].id = i;
Paolo Bonzinia47d2b02015-05-17 11:41:37 +0200542
543 return slots;
544}
545
546static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
547{
548 if (!memslot->dirty_bitmap)
549 return;
550
551 kvfree(memslot->dirty_bitmap);
552 memslot->dirty_bitmap = NULL;
553}
554
555/*
556 * Free any memory in @free but not in @dont.
557 */
558static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
559 struct kvm_memory_slot *dont)
560{
561 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
562 kvm_destroy_dirty_bitmap(free);
563
564 kvm_arch_free_memslot(kvm, free, dont);
565
566 free->npages = 0;
567}
568
569static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
570{
571 struct kvm_memory_slot *memslot;
572
573 if (!slots)
574 return;
575
576 kvm_for_each_memslot(memslot, slots)
577 kvm_free_memslot(kvm, memslot, NULL);
578
579 kvfree(slots);
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800580}
581
Janosch Frank536a6f82016-05-18 13:26:23 +0200582static void kvm_destroy_vm_debugfs(struct kvm *kvm)
583{
584 int i;
585
586 if (!kvm->debugfs_dentry)
587 return;
588
589 debugfs_remove_recursive(kvm->debugfs_dentry);
590
Luiz Capitulino9d5a1dc2016-09-07 14:47:21 -0400591 if (kvm->debugfs_stat_data) {
592 for (i = 0; i < kvm_debugfs_num_entries; i++)
593 kfree(kvm->debugfs_stat_data[i]);
594 kfree(kvm->debugfs_stat_data);
595 }
Janosch Frank536a6f82016-05-18 13:26:23 +0200596}
597
598static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
599{
600 char dir_name[ITOA_MAX_LEN * 2];
601 struct kvm_stat_data *stat_data;
602 struct kvm_stats_debugfs_item *p;
603
604 if (!debugfs_initialized())
605 return 0;
606
607 snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
608 kvm->debugfs_dentry = debugfs_create_dir(dir_name,
609 kvm_debugfs_dir);
610 if (!kvm->debugfs_dentry)
611 return -ENOMEM;
612
613 kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
614 sizeof(*kvm->debugfs_stat_data),
615 GFP_KERNEL);
616 if (!kvm->debugfs_stat_data)
617 return -ENOMEM;
618
619 for (p = debugfs_entries; p->name; p++) {
620 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL);
621 if (!stat_data)
622 return -ENOMEM;
623
624 stat_data->kvm = kvm;
625 stat_data->offset = p->offset;
626 kvm->debugfs_stat_data[p - debugfs_entries] = stat_data;
627 if (!debugfs_create_file(p->name, 0444,
628 kvm->debugfs_dentry,
629 stat_data,
630 stat_fops_per_vm[p->kind]))
631 return -ENOMEM;
632 }
633 return 0;
634}
635
Junaid Shahidc6e94ac2019-11-04 12:22:03 +0100636/*
637 * Called after the VM is otherwise initialized, but just before adding it to
638 * the vm_list.
639 */
640int __weak kvm_arch_post_init_vm(struct kvm *kvm)
641{
642 return 0;
643}
644
645/*
646 * Called just after removing the VM from the vm_list, but before doing any
647 * other destruction.
648 */
649void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm)
650{
651}
652
Carsten Ottee08b9632012-01-04 10:25:20 +0100653static struct kvm *kvm_create_vm(unsigned long type)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800654{
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100655 int r, i;
656 struct kvm *kvm = kvm_arch_alloc_vm();
Avi Kivity6aa8b732006-12-10 02:21:36 -0800657
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100658 if (!kvm)
659 return ERR_PTR(-ENOMEM);
660
Paolo Bonzinie9ad4ec2016-03-21 10:15:25 +0100661 spin_lock_init(&kvm->mmu_lock);
662 atomic_inc(&current->mm->mm_count);
663 kvm->mm = current->mm;
664 kvm_eventfd_init(kvm);
665 mutex_init(&kvm->lock);
666 mutex_init(&kvm->irq_lock);
667 mutex_init(&kvm->slots_lock);
668 atomic_set(&kvm->users_count, 1);
669 INIT_LIST_HEAD(&kvm->devices);
670
Carsten Ottee08b9632012-01-04 10:25:20 +0100671 r = kvm_arch_init_vm(kvm, type);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100672 if (r)
Christian Borntraeger719d93c2014-01-16 13:44:20 +0100673 goto out_err_no_disable;
Alexander Graf10474ae2009-09-15 11:37:46 +0200674
675 r = hardware_enable_all();
676 if (r)
Christian Borntraeger719d93c2014-01-16 13:44:20 +0100677 goto out_err_no_disable;
Alexander Graf10474ae2009-09-15 11:37:46 +0200678
Paolo Bonzinic77dcac2014-08-06 14:24:45 +0200679#ifdef CONFIG_HAVE_KVM_IRQFD
Gleb Natapov136bdfe2009-08-24 11:54:23 +0300680 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
Avi Kivity75858a82009-01-04 17:10:50 +0200681#endif
Avi Kivity6aa8b732006-12-10 02:21:36 -0800682
Alex Williamson1e702d92012-12-10 10:33:32 -0700683 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
684
Marcelo Tosatti46a26bf2009-12-23 14:35:16 -0200685 r = -ENOMEM;
Paolo Bonzinif481b062015-05-17 17:30:37 +0200686 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
687 kvm->memslots[i] = kvm_alloc_memslots();
688 if (!kvm->memslots[i])
689 goto out_err_no_srcu;
690 }
Paolo Bonzini00f034a2014-08-20 14:29:21 +0200691
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200692 if (init_srcu_struct(&kvm->srcu))
Christian Borntraeger719d93c2014-01-16 13:44:20 +0100693 goto out_err_no_srcu;
694 if (init_srcu_struct(&kvm->irq_srcu))
695 goto out_err_no_irq_srcu;
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200696 for (i = 0; i < KVM_NR_BUSES; i++) {
697 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
698 GFP_KERNEL);
Jan Kiszka57e7fbe2010-11-09 12:42:12 +0100699 if (!kvm->buses[i])
Junaid Shahidc6e94ac2019-11-04 12:22:03 +0100700 goto out_err_no_mmu_notifier;
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200701 }
Mike Waychison74b5c5b2011-06-03 13:04:53 -0700702
Mike Waychison74b5c5b2011-06-03 13:04:53 -0700703 r = kvm_init_mmu_notifier(kvm);
704 if (r)
Junaid Shahidc6e94ac2019-11-04 12:22:03 +0100705 goto out_err_no_mmu_notifier;
706
707 r = kvm_arch_post_init_vm(kvm);
708 if (r)
Mike Waychison74b5c5b2011-06-03 13:04:53 -0700709 goto out_err;
710
Junaid Shahidc6170b82019-01-03 17:14:28 -0800711 mutex_lock(&kvm_lock);
Rusty Russell5e58cfe2007-07-23 17:08:21 +1000712 list_add(&kvm->vm_list, &vm_list);
Junaid Shahidc6170b82019-01-03 17:14:28 -0800713 mutex_unlock(&kvm_lock);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100714
Peter Zijlstra2ecd9d22015-07-03 18:53:58 +0200715 preempt_notifier_inc();
716
Avi Kivityf17abe92007-02-21 19:28:04 +0200717 return kvm;
Alexander Graf10474ae2009-09-15 11:37:46 +0200718
719out_err:
Junaid Shahidc6e94ac2019-11-04 12:22:03 +0100720#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
721 if (kvm->mmu_notifier.ops)
722 mmu_notifier_unregister(&kvm->mmu_notifier, current->mm);
723#endif
724out_err_no_mmu_notifier:
Christian Borntraeger719d93c2014-01-16 13:44:20 +0100725 cleanup_srcu_struct(&kvm->irq_srcu);
726out_err_no_irq_srcu:
Jan Kiszka57e7fbe2010-11-09 12:42:12 +0100727 cleanup_srcu_struct(&kvm->srcu);
Christian Borntraeger719d93c2014-01-16 13:44:20 +0100728out_err_no_srcu:
Alexander Graf10474ae2009-09-15 11:37:46 +0200729 hardware_disable_all();
Christian Borntraeger719d93c2014-01-16 13:44:20 +0100730out_err_no_disable:
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200731 for (i = 0; i < KVM_NR_BUSES; i++)
732 kfree(kvm->buses[i]);
Paolo Bonzinif481b062015-05-17 17:30:37 +0200733 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
734 kvm_free_memslots(kvm, kvm->memslots[i]);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100735 kvm_arch_free_vm(kvm);
Paolo Bonzinie9ad4ec2016-03-21 10:15:25 +0100736 mmdrop(current->mm);
Alexander Graf10474ae2009-09-15 11:37:46 +0200737 return ERR_PTR(r);
Avi Kivityf17abe92007-02-21 19:28:04 +0200738}
739
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900740/*
741 * Avoid using vmalloc for a small buffer.
742 * Should not be used when the size is statically known.
743 */
Takuya Yoshikawac1a7b322012-05-20 13:15:07 +0900744void *kvm_kvzalloc(unsigned long size)
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900745{
746 if (size > PAGE_SIZE)
747 return vzalloc(size);
748 else
749 return kzalloc(size, GFP_KERNEL);
750}
751
Scott Wood07f0a7b2013-04-25 14:11:23 +0000752static void kvm_destroy_devices(struct kvm *kvm)
753{
Geliang Tange6e3b5a2016-01-01 19:47:12 +0800754 struct kvm_device *dev, *tmp;
Scott Wood07f0a7b2013-04-25 14:11:23 +0000755
Christoffer Dalla28ebea2016-08-09 19:13:01 +0200756 /*
757 * We do not need to take the kvm->lock here, because nobody else
758 * has a reference to the struct kvm at this point and therefore
759 * cannot access the devices list anyhow.
760 */
Geliang Tange6e3b5a2016-01-01 19:47:12 +0800761 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
762 list_del(&dev->vm_node);
Scott Wood07f0a7b2013-04-25 14:11:23 +0000763 dev->ops->destroy(dev);
764 }
765}
766
Avi Kivityf17abe92007-02-21 19:28:04 +0200767static void kvm_destroy_vm(struct kvm *kvm)
768{
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200769 int i;
Avi Kivity6d4e4c42007-11-21 16:41:05 +0200770 struct mm_struct *mm = kvm->mm;
771
Janosch Frank536a6f82016-05-18 13:26:23 +0200772 kvm_destroy_vm_debugfs(kvm);
Sheng Yangad8ba2c2009-01-06 10:03:02 +0800773 kvm_arch_sync_events(kvm);
Junaid Shahidc6170b82019-01-03 17:14:28 -0800774 mutex_lock(&kvm_lock);
Avi Kivity133de902007-02-12 00:54:44 -0800775 list_del(&kvm->vm_list);
Junaid Shahidc6170b82019-01-03 17:14:28 -0800776 mutex_unlock(&kvm_lock);
Junaid Shahidc6e94ac2019-11-04 12:22:03 +0100777 kvm_arch_pre_destroy_vm(kvm);
778
Avi Kivity399ec802008-11-19 13:58:46 +0200779 kvm_free_irq_routing(kvm);
Peter Xuef46a132017-03-15 16:01:17 +0800780 for (i = 0; i < KVM_NR_BUSES; i++) {
David Hildenbrand15636252017-03-23 18:24:19 +0100781 if (kvm->buses[i])
782 kvm_io_bus_destroy(kvm->buses[i]);
Peter Xuef46a132017-03-15 16:01:17 +0800783 kvm->buses[i] = NULL;
784 }
Avi Kivity980da6c2009-12-20 15:13:43 +0200785 kvm_coalesced_mmio_free(kvm);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200786#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
787 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
Gleb Natapovf00be0c2009-03-19 12:20:36 +0200788#else
Marcelo Tosatti2df72e92012-08-24 15:54:57 -0300789 kvm_arch_flush_shadow_all(kvm);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200790#endif
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +0800791 kvm_arch_destroy_vm(kvm);
Scott Wood07f0a7b2013-04-25 14:11:23 +0000792 kvm_destroy_devices(kvm);
Paolo Bonzinif481b062015-05-17 17:30:37 +0200793 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
794 kvm_free_memslots(kvm, kvm->memslots[i]);
Paolo Bonzini820b3fc2014-06-03 13:44:17 +0200795 cleanup_srcu_struct(&kvm->irq_srcu);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100796 cleanup_srcu_struct(&kvm->srcu);
797 kvm_arch_free_vm(kvm);
Peter Zijlstra2ecd9d22015-07-03 18:53:58 +0200798 preempt_notifier_dec();
Alexander Graf10474ae2009-09-15 11:37:46 +0200799 hardware_disable_all();
Avi Kivity6d4e4c42007-11-21 16:41:05 +0200800 mmdrop(mm);
Avi Kivityf17abe92007-02-21 19:28:04 +0200801}
802
Izik Eidusd39f13b2008-03-30 16:01:25 +0300803void kvm_get_kvm(struct kvm *kvm)
804{
805 atomic_inc(&kvm->users_count);
806}
807EXPORT_SYMBOL_GPL(kvm_get_kvm);
808
809void kvm_put_kvm(struct kvm *kvm)
810{
811 if (atomic_dec_and_test(&kvm->users_count))
812 kvm_destroy_vm(kvm);
813}
814EXPORT_SYMBOL_GPL(kvm_put_kvm);
815
816
Avi Kivityf17abe92007-02-21 19:28:04 +0200817static int kvm_vm_release(struct inode *inode, struct file *filp)
818{
819 struct kvm *kvm = filp->private_data;
820
Gregory Haskins721eecb2009-05-20 10:30:49 -0400821 kvm_irqfd_release(kvm);
822
Izik Eidusd39f13b2008-03-30 16:01:25 +0300823 kvm_put_kvm(kvm);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800824 return 0;
825}
826
Takuya Yoshikawa515a0122010-10-27 18:23:54 +0900827/*
828 * Allocation size is twice as large as the actual dirty bitmap size.
Takuya Yoshikawa93474b22012-03-01 19:34:45 +0900829 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
Takuya Yoshikawa515a0122010-10-27 18:23:54 +0900830 */
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900831static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
832{
Takuya Yoshikawa515a0122010-10-27 18:23:54 +0900833 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900834
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900835 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900836 if (!memslot->dirty_bitmap)
837 return -ENOMEM;
838
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900839 return 0;
840}
841
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800842/*
Igor Mammedov0e60b072014-12-01 17:29:26 +0000843 * Insert memslot and re-sort memslots based on their GFN,
844 * so binary search could be used to lookup GFN.
845 * Sorting algorithm takes advantage of having initially
846 * sorted array and known changed memslot position.
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800847 */
Paolo Bonzini5cc15022014-11-14 10:55:31 +0100848static void update_memslots(struct kvm_memslots *slots,
849 struct kvm_memory_slot *new)
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800850{
Paolo Bonzini8593176c2014-11-14 10:22:07 +0100851 int id = new->id;
852 int i = slots->id_to_index[id];
Igor Mammedov063584d2014-11-13 23:00:13 +0000853 struct kvm_memory_slot *mslots = slots->memslots;
Xiao Guangrongf85e2cb2011-11-24 17:41:54 +0800854
Paolo Bonzini8593176c2014-11-14 10:22:07 +0100855 WARN_ON(mslots[i].id != id);
Igor Mammedov9c1a5d382014-12-01 17:29:27 +0000856 if (!new->npages) {
Paolo Bonzinidbaff302014-12-27 21:08:16 +0100857 WARN_ON(!mslots[i].npages);
Igor Mammedov9c1a5d382014-12-01 17:29:27 +0000858 if (mslots[i].npages)
859 slots->used_slots--;
860 } else {
861 if (!mslots[i].npages)
862 slots->used_slots++;
863 }
Igor Mammedov0e60b072014-12-01 17:29:26 +0000864
Igor Mammedov7f379cf2014-12-01 17:29:24 +0000865 while (i < KVM_MEM_SLOTS_NUM - 1 &&
Igor Mammedov0e60b072014-12-01 17:29:26 +0000866 new->base_gfn <= mslots[i + 1].base_gfn) {
867 if (!mslots[i + 1].npages)
868 break;
Igor Mammedov7f379cf2014-12-01 17:29:24 +0000869 mslots[i] = mslots[i + 1];
870 slots->id_to_index[mslots[i].id] = i;
871 i++;
872 }
Paolo Bonziniefbeec72014-12-27 18:01:00 +0100873
874 /*
875 * The ">=" is needed when creating a slot with base_gfn == 0,
876 * so that it moves before all those with base_gfn == npages == 0.
877 *
878 * On the other hand, if new->npages is zero, the above loop has
879 * already left i pointing to the beginning of the empty part of
880 * mslots, and the ">=" would move the hole backwards in this
881 * case---which is wrong. So skip the loop when deleting a slot.
882 */
883 if (new->npages) {
884 while (i > 0 &&
885 new->base_gfn >= mslots[i - 1].base_gfn) {
886 mslots[i] = mslots[i - 1];
887 slots->id_to_index[mslots[i].id] = i;
888 i--;
889 }
Paolo Bonzinidbaff302014-12-27 21:08:16 +0100890 } else
891 WARN_ON_ONCE(i != slots->used_slots);
Xiao Guangrongf85e2cb2011-11-24 17:41:54 +0800892
Paolo Bonzini8593176c2014-11-14 10:22:07 +0100893 mslots[i] = *new;
894 slots->id_to_index[mslots[i].id] = i;
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800895}
896
Paolo Bonzini09170a42015-05-18 13:59:39 +0200897static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
Xiao Guangronga50d64d2012-08-21 10:58:13 +0800898{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +0800899 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
900
Christoffer Dall0f8a4de2014-08-26 14:00:37 +0200901#ifdef __KVM_HAVE_READONLY_MEM
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +0800902 valid_flags |= KVM_MEM_READONLY;
903#endif
904
905 if (mem->flags & ~valid_flags)
Xiao Guangronga50d64d2012-08-21 10:58:13 +0800906 return -EINVAL;
907
908 return 0;
909}
910
Gleb Natapov7ec4fb42012-12-24 17:49:30 +0200911static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
Paolo Bonzinif481b062015-05-17 17:30:37 +0200912 int as_id, struct kvm_memslots *slots)
Gleb Natapov7ec4fb42012-12-24 17:49:30 +0200913{
Paolo Bonzinif481b062015-05-17 17:30:37 +0200914 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
Gleb Natapov7ec4fb42012-12-24 17:49:30 +0200915
David Matlackee3d1572014-08-18 15:46:06 -0700916 /*
917 * Set the low bit in the generation, which disables SPTE caching
918 * until the end of synchronize_srcu_expedited.
919 */
920 WARN_ON(old_memslots->generation & 1);
921 slots->generation = old_memslots->generation + 1;
922
Paolo Bonzinif481b062015-05-17 17:30:37 +0200923 rcu_assign_pointer(kvm->memslots[as_id], slots);
Gleb Natapov7ec4fb42012-12-24 17:49:30 +0200924 synchronize_srcu_expedited(&kvm->srcu);
Takuya Yoshikawae59dbe02013-07-04 13:40:29 +0900925
David Matlackee3d1572014-08-18 15:46:06 -0700926 /*
927 * Increment the new memslot generation a second time. This prevents
928 * vm exits that race with memslot updates from caching a memslot
929 * generation that will (potentially) be valid forever.
930 */
931 slots->generation++;
932
Paolo Bonzini15f46012015-05-17 21:26:08 +0200933 kvm_arch_memslots_updated(kvm, slots);
Takuya Yoshikawae59dbe02013-07-04 13:40:29 +0900934
935 return old_memslots;
Gleb Natapov7ec4fb42012-12-24 17:49:30 +0200936}
937
Avi Kivity6aa8b732006-12-10 02:21:36 -0800938/*
Avi Kivity6aa8b732006-12-10 02:21:36 -0800939 * Allocate some memory and give it an address in the guest physical address
940 * space.
941 *
942 * Discontiguous memory is allowed, mostly for framebuffers.
Sheng Yangf78e0e22007-10-29 09:40:42 +0800943 *
Dominik Dingel02d5d552014-10-27 16:22:56 +0100944 * Must be called holding kvm->slots_lock for write.
Avi Kivity6aa8b732006-12-10 02:21:36 -0800945 */
Sheng Yangf78e0e22007-10-29 09:40:42 +0800946int __kvm_set_memory_region(struct kvm *kvm,
Paolo Bonzini09170a42015-05-18 13:59:39 +0200947 const struct kvm_userspace_memory_region *mem)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800948{
Avi Kivity8234b222010-12-27 12:08:45 +0200949 int r;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800950 gfn_t base_gfn;
Heiko Carstens28bcb112009-09-03 17:35:35 +0200951 unsigned long npages;
Takuya Yoshikawaa843fac2013-01-11 18:27:43 +0900952 struct kvm_memory_slot *slot;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800953 struct kvm_memory_slot old, new;
Alex Williamsonb7f69c52012-12-10 10:33:03 -0700954 struct kvm_memslots *slots = NULL, *old_memslots;
Paolo Bonzinif481b062015-05-17 17:30:37 +0200955 int as_id, id;
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900956 enum kvm_mr_change change;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800957
Xiao Guangronga50d64d2012-08-21 10:58:13 +0800958 r = check_memory_region_flags(mem);
959 if (r)
960 goto out;
961
Avi Kivity6aa8b732006-12-10 02:21:36 -0800962 r = -EINVAL;
Paolo Bonzinif481b062015-05-17 17:30:37 +0200963 as_id = mem->slot >> 16;
964 id = (u16)mem->slot;
965
Avi Kivity6aa8b732006-12-10 02:21:36 -0800966 /* General sanity checks */
967 if (mem->memory_size & (PAGE_SIZE - 1))
968 goto out;
969 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
970 goto out;
Takuya Yoshikawafa3d3152011-05-07 16:35:38 +0900971 /* We can read the guest memory with __xxx_user() later on. */
Paolo Bonzinif481b062015-05-17 17:30:37 +0200972 if ((id < KVM_USER_MEM_SLOTS) &&
Takuya Yoshikawafa3d3152011-05-07 16:35:38 +0900973 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
Heiko Carstens9e3bb6b2011-05-24 07:51:27 +0200974 !access_ok(VERIFY_WRITE,
975 (void __user *)(unsigned long)mem->userspace_addr,
976 mem->memory_size)))
Hollis Blanchard78749802008-11-07 13:32:12 -0600977 goto out;
Paolo Bonzinif481b062015-05-17 17:30:37 +0200978 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800979 goto out;
980 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
981 goto out;
982
Paolo Bonzinif481b062015-05-17 17:30:37 +0200983 slot = id_to_memslot(__kvm_memslots(kvm, as_id), id);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800984 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
985 npages = mem->memory_size >> PAGE_SHIFT;
986
Takuya Yoshikawa660c22c2010-04-13 22:47:24 +0900987 if (npages > KVM_MEM_MAX_NR_PAGES)
988 goto out;
989
Takuya Yoshikawaa843fac2013-01-11 18:27:43 +0900990 new = old = *slot;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800991
Paolo Bonzinif481b062015-05-17 17:30:37 +0200992 new.id = id;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800993 new.base_gfn = base_gfn;
994 new.npages = npages;
995 new.flags = mem->flags;
996
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900997 if (npages) {
998 if (!old.npages)
999 change = KVM_MR_CREATE;
1000 else { /* Modify an existing slot. */
1001 if ((mem->userspace_addr != old.userspace_addr) ||
Takuya Yoshikawa75d61fb2013-01-30 19:40:41 +09001002 (npages != old.npages) ||
1003 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +09001004 goto out;
1005
1006 if (base_gfn != old.base_gfn)
1007 change = KVM_MR_MOVE;
1008 else if (new.flags != old.flags)
1009 change = KVM_MR_FLAGS_ONLY;
1010 else { /* Nothing to change. */
1011 r = 0;
1012 goto out;
1013 }
1014 }
Paolo Bonzini09170a42015-05-18 13:59:39 +02001015 } else {
1016 if (!old.npages)
1017 goto out;
1018
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +09001019 change = KVM_MR_DELETE;
Paolo Bonzini09170a42015-05-18 13:59:39 +02001020 new.base_gfn = 0;
1021 new.flags = 0;
1022 }
Avi Kivity6aa8b732006-12-10 02:21:36 -08001023
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +09001024 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
Takuya Yoshikawa0a706be2013-01-11 18:26:55 +09001025 /* Check for overlaps */
1026 r = -EEXIST;
Paolo Bonzinif481b062015-05-17 17:30:37 +02001027 kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) {
Wanpeng Li2e112f32018-02-13 15:36:00 +01001028 if (slot->id == id)
Takuya Yoshikawa0a706be2013-01-11 18:26:55 +09001029 continue;
1030 if (!((base_gfn + npages <= slot->base_gfn) ||
1031 (base_gfn >= slot->base_gfn + slot->npages)))
1032 goto out;
1033 }
Avi Kivity6aa8b732006-12-10 02:21:36 -08001034 }
Avi Kivity6aa8b732006-12-10 02:21:36 -08001035
Avi Kivity6aa8b732006-12-10 02:21:36 -08001036 /* Free page dirty bitmap if unneeded */
1037 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
Al Viro8b6d44c2007-02-09 16:38:40 +00001038 new.dirty_bitmap = NULL;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001039
1040 r = -ENOMEM;
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +09001041 if (change == KVM_MR_CREATE) {
Takuya Yoshikawa189a2f72012-02-08 13:01:09 +09001042 new.userspace_addr = mem->userspace_addr;
Takuya Yoshikawad89cc612012-08-01 18:03:28 +09001043
Aneesh Kumar K.V55870272013-10-07 22:18:00 +05301044 if (kvm_arch_create_memslot(kvm, &new, npages))
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +09001045 goto out_free;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001046 }
Joerg Roedelec04b262009-06-19 15:16:23 +02001047
Avi Kivity6aa8b732006-12-10 02:21:36 -08001048 /* Allocate page dirty bitmap if needed */
1049 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +09001050 if (kvm_create_dirty_bitmap(&new) < 0)
Sheng Yangf78e0e22007-10-29 09:40:42 +08001051 goto out_free;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001052 }
1053
Igor Mammedov74496132015-03-20 12:21:37 +00001054 slots = kvm_kvzalloc(sizeof(struct kvm_memslots));
Paolo Bonzinif2a81032014-11-14 10:46:45 +01001055 if (!slots)
1056 goto out_free;
Paolo Bonzinif481b062015-05-17 17:30:37 +02001057 memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
Paolo Bonzinif2a81032014-11-14 10:46:45 +01001058
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +09001059 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
Paolo Bonzinif481b062015-05-17 17:30:37 +02001060 slot = id_to_memslot(slots, id);
Xiao Guangrong28a37542011-11-24 19:04:35 +08001061 slot->flags |= KVM_MEMSLOT_INVALID;
1062
Paolo Bonzinif481b062015-05-17 17:30:37 +02001063 old_memslots = install_new_memslots(kvm, as_id, slots);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -02001064
Alex Williamsone40f1932012-12-10 10:32:57 -07001065 /* slot was deleted or moved, clear iommu mapping */
1066 kvm_iommu_unmap_pages(kvm, &old);
Marcelo Tosatti12d6e752012-08-24 15:54:58 -03001067 /* From this point no new shadow pages pointing to a deleted,
1068 * or moved, memslot will be created.
Marcelo Tosattibc6678a2009-12-23 14:35:21 -02001069 *
1070 * validation of sp->gfn happens in:
Xiubo Lib7d409d2015-02-26 14:58:24 +08001071 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1072 * - kvm_is_visible_gfn (mmu_check_roots)
Marcelo Tosattibc6678a2009-12-23 14:35:21 -02001073 */
Marcelo Tosatti2df72e92012-08-24 15:54:57 -03001074 kvm_arch_flush_shadow_memslot(kvm, slot);
Paolo Bonzinif2a81032014-11-14 10:46:45 +01001075
1076 /*
1077 * We can re-use the old_memslots from above, the only difference
1078 * from the currently installed memslots is the invalid flag. This
1079 * will get overwritten by update_memslots anyway.
1080 */
Alex Williamsonb7f69c52012-12-10 10:33:03 -07001081 slots = old_memslots;
Marcelo Tosattibc6678a2009-12-23 14:35:21 -02001082 }
Marcelo Tosatti34d4cb82008-07-10 20:49:31 -03001083
Takuya Yoshikawa7b6195a2013-02-27 19:44:34 +09001084 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
Marcelo Tosattif7784b82009-12-23 14:35:18 -02001085 if (r)
Alex Williamsonb7f69c52012-12-10 10:33:03 -07001086 goto out_slots;
Marcelo Tosattif7784b82009-12-23 14:35:18 -02001087
Paolo Bonzinia47d2b02015-05-17 11:41:37 +02001088 /* actual memory is freed via old in kvm_free_memslot below */
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +09001089 if (change == KVM_MR_DELETE) {
Marcelo Tosattibc6678a2009-12-23 14:35:21 -02001090 new.dirty_bitmap = NULL;
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +09001091 memset(&new.arch, 0, sizeof(new.arch));
Marcelo Tosattibc6678a2009-12-23 14:35:21 -02001092 }
1093
Paolo Bonzini5cc15022014-11-14 10:55:31 +01001094 update_memslots(slots, &new);
Paolo Bonzinif481b062015-05-17 17:30:37 +02001095 old_memslots = install_new_memslots(kvm, as_id, slots);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001096
Paolo Bonzinif36f3f22015-05-18 13:20:23 +02001097 kvm_arch_commit_memory_region(kvm, mem, &old, &new, change);
Zhang Xiantao3ad82a72007-11-20 13:11:38 +08001098
Paolo Bonzinia47d2b02015-05-17 11:41:37 +02001099 kvm_free_memslot(kvm, &old, &new);
Igor Mammedov74496132015-03-20 12:21:37 +00001100 kvfree(old_memslots);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -02001101
Marcelo Tosattibc6678a2009-12-23 14:35:21 -02001102 /*
1103 * IOMMU mapping: New slots need to be mapped. Old slots need to be
Avi Kivity6aa8b732006-12-10 02:21:36 -08001104 * un-mapped and re-mapped if their base changes. Since base change
1105 * unmapping is handled above with slot deletion, mapping alone is
Sheng Yangf78e0e22007-10-29 09:40:42 +08001106 * needed here. Anything else the iommu might care about for existing
Avi Kivity6aa8b732006-12-10 02:21:36 -08001107 * slots (size changes, userspace addr changes and read-only flag
1108 * changes) is disallowed above, so any other attribute changes getting
1109 * here can be skipped.
Izik Eidus210c7c42007-10-24 23:52:57 +02001110 */
Herongguang (Stephen)808ed3b2017-03-27 15:21:17 +08001111 if (as_id == 0 && (change == KVM_MR_CREATE || change == KVM_MR_MOVE)) {
Avi Kivity6aa8b732006-12-10 02:21:36 -08001112 r = kvm_iommu_map_pages(kvm, &new);
Yang Zhange0230e12013-10-24 09:56:39 +08001113 return r;
Sheng Yangf78e0e22007-10-29 09:40:42 +08001114 }
1115
Avi Kivity6aa8b732006-12-10 02:21:36 -08001116 return 0;
1117
1118out_slots:
Igor Mammedov74496132015-03-20 12:21:37 +00001119 kvfree(slots);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001120out_free:
Paolo Bonzinia47d2b02015-05-17 11:41:37 +02001121 kvm_free_memslot(kvm, &new, &old);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001122out:
1123 return r;
Izik Eidus210c7c42007-10-24 23:52:57 +02001124}
1125EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
Sheng Yangf78e0e22007-10-29 09:40:42 +08001126
1127int kvm_set_memory_region(struct kvm *kvm,
Paolo Bonzini09170a42015-05-18 13:59:39 +02001128 const struct kvm_userspace_memory_region *mem)
Sheng Yangf78e0e22007-10-29 09:40:42 +08001129{
1130 int r;
1131
Marcelo Tosatti79fac952009-12-23 14:35:26 -02001132 mutex_lock(&kvm->slots_lock);
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +09001133 r = __kvm_set_memory_region(kvm, mem);
Marcelo Tosatti79fac952009-12-23 14:35:26 -02001134 mutex_unlock(&kvm->slots_lock);
Sheng Yangf78e0e22007-10-29 09:40:42 +08001135 return r;
1136}
Izik Eidus210c7c42007-10-24 23:52:57 +02001137EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1138
Stephen Hemminger79408762013-12-29 12:12:29 -08001139static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1140 struct kvm_userspace_memory_region *mem)
Izik Eidus210c7c42007-10-24 23:52:57 +02001141{
Paolo Bonzinif481b062015-05-17 17:30:37 +02001142 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
Izik Eiduse0d62c72007-10-24 23:57:46 +02001143 return -EINVAL;
Paolo Bonzini09170a42015-05-18 13:59:39 +02001144
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +09001145 return kvm_set_memory_region(kvm, mem);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001146}
1147
Zhang Xiantao5bb064d2007-11-18 20:29:43 +08001148int kvm_get_dirty_log(struct kvm *kvm,
1149 struct kvm_dirty_log *log, int *is_dirty)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001150{
Paolo Bonzini9f6b8022015-05-17 16:20:07 +02001151 struct kvm_memslots *slots;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001152 struct kvm_memory_slot *memslot;
Paolo Bonzinif481b062015-05-17 17:30:37 +02001153 int r, i, as_id, id;
Takuya Yoshikawa87bf6e72010-04-12 19:35:35 +09001154 unsigned long n;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001155 unsigned long any = 0;
1156
Avi Kivity6aa8b732006-12-10 02:21:36 -08001157 r = -EINVAL;
Paolo Bonzinif481b062015-05-17 17:30:37 +02001158 as_id = log->slot >> 16;
1159 id = (u16)log->slot;
1160 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001161 goto out;
1162
Paolo Bonzinif481b062015-05-17 17:30:37 +02001163 slots = __kvm_memslots(kvm, as_id);
1164 memslot = id_to_memslot(slots, id);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001165 r = -ENOENT;
1166 if (!memslot->dirty_bitmap)
1167 goto out;
1168
Takuya Yoshikawa87bf6e72010-04-12 19:35:35 +09001169 n = kvm_dirty_bitmap_bytes(memslot);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001170
Uri Lublincd1a4a92007-02-22 16:43:09 +02001171 for (i = 0; !any && i < n/sizeof(long); ++i)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001172 any = memslot->dirty_bitmap[i];
1173
1174 r = -EFAULT;
1175 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1176 goto out;
1177
Zhang Xiantao5bb064d2007-11-18 20:29:43 +08001178 if (any)
1179 *is_dirty = 1;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001180
1181 r = 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001182out:
Avi Kivity6aa8b732006-12-10 02:21:36 -08001183 return r;
1184}
Aneesh Kumar K.V2ba9f0d2013-10-07 22:17:59 +05301185EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001186
Mario Smarduchba0513b2015-01-15 15:58:53 -08001187#ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1188/**
1189 * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1190 * are dirty write protect them for next write.
1191 * @kvm: pointer to kvm instance
1192 * @log: slot id and address to which we copy the log
1193 * @is_dirty: flag set if any page is dirty
1194 *
1195 * We need to keep it in mind that VCPU threads can write to the bitmap
1196 * concurrently. So, to avoid losing track of dirty pages we keep the
1197 * following order:
1198 *
1199 * 1. Take a snapshot of the bit and clear it if needed.
1200 * 2. Write protect the corresponding page.
1201 * 3. Copy the snapshot to the userspace.
1202 * 4. Upon return caller flushes TLB's if needed.
1203 *
1204 * Between 2 and 4, the guest may write to the page using the remaining TLB
1205 * entry. This is not a problem because the page is reported dirty using
1206 * the snapshot taken before and step 4 ensures that writes done after
1207 * exiting to userspace will be logged for the next call.
1208 *
1209 */
1210int kvm_get_dirty_log_protect(struct kvm *kvm,
1211 struct kvm_dirty_log *log, bool *is_dirty)
1212{
Paolo Bonzini9f6b8022015-05-17 16:20:07 +02001213 struct kvm_memslots *slots;
Mario Smarduchba0513b2015-01-15 15:58:53 -08001214 struct kvm_memory_slot *memslot;
Paolo Bonzinif481b062015-05-17 17:30:37 +02001215 int r, i, as_id, id;
Mario Smarduchba0513b2015-01-15 15:58:53 -08001216 unsigned long n;
1217 unsigned long *dirty_bitmap;
1218 unsigned long *dirty_bitmap_buffer;
1219
1220 r = -EINVAL;
Paolo Bonzinif481b062015-05-17 17:30:37 +02001221 as_id = log->slot >> 16;
1222 id = (u16)log->slot;
1223 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
Mario Smarduchba0513b2015-01-15 15:58:53 -08001224 goto out;
1225
Paolo Bonzinif481b062015-05-17 17:30:37 +02001226 slots = __kvm_memslots(kvm, as_id);
1227 memslot = id_to_memslot(slots, id);
Mario Smarduchba0513b2015-01-15 15:58:53 -08001228
1229 dirty_bitmap = memslot->dirty_bitmap;
1230 r = -ENOENT;
1231 if (!dirty_bitmap)
1232 goto out;
1233
1234 n = kvm_dirty_bitmap_bytes(memslot);
1235
1236 dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
1237 memset(dirty_bitmap_buffer, 0, n);
1238
1239 spin_lock(&kvm->mmu_lock);
1240 *is_dirty = false;
1241 for (i = 0; i < n / sizeof(long); i++) {
1242 unsigned long mask;
1243 gfn_t offset;
1244
1245 if (!dirty_bitmap[i])
1246 continue;
1247
1248 *is_dirty = true;
1249
1250 mask = xchg(&dirty_bitmap[i], 0);
1251 dirty_bitmap_buffer[i] = mask;
1252
Takuya Yoshikawa58d29302015-03-17 16:19:58 +09001253 if (mask) {
1254 offset = i * BITS_PER_LONG;
1255 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1256 offset, mask);
1257 }
Mario Smarduchba0513b2015-01-15 15:58:53 -08001258 }
1259
1260 spin_unlock(&kvm->mmu_lock);
1261
1262 r = -EFAULT;
1263 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1264 goto out;
1265
1266 r = 0;
1267out:
1268 return r;
1269}
1270EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1271#endif
1272
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +09001273bool kvm_largepages_enabled(void)
1274{
1275 return largepages_enabled;
1276}
1277
Marcelo Tosatti54dee992009-06-11 12:07:44 -03001278void kvm_disable_largepages(void)
1279{
1280 largepages_enabled = false;
1281}
1282EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1283
Gleb Natapov49c77542010-10-18 15:22:23 +02001284struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1285{
1286 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1287}
Avi Kivitya1f4d3952010-06-21 11:44:20 +03001288EXPORT_SYMBOL_GPL(gfn_to_memslot);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001289
Paolo Bonzini8e734852015-05-17 13:58:53 +02001290struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1291{
1292 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1293}
1294
Yaowei Bai33e94152015-11-14 11:21:06 +08001295bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
Izik Eiduse0d62c72007-10-24 23:57:46 +02001296{
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +08001297 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
Izik Eiduse0d62c72007-10-24 23:57:46 +02001298
Alex Williamsonbbacc0c2012-12-10 10:33:09 -07001299 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +08001300 memslot->flags & KVM_MEMSLOT_INVALID)
Yaowei Bai33e94152015-11-14 11:21:06 +08001301 return false;
Izik Eiduse0d62c72007-10-24 23:57:46 +02001302
Yaowei Bai33e94152015-11-14 11:21:06 +08001303 return true;
Izik Eiduse0d62c72007-10-24 23:57:46 +02001304}
1305EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1306
Joerg Roedel8f0b1ab2010-01-28 12:37:56 +01001307unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1308{
1309 struct vm_area_struct *vma;
1310 unsigned long addr, size;
1311
1312 size = PAGE_SIZE;
1313
1314 addr = gfn_to_hva(kvm, gfn);
1315 if (kvm_is_error_hva(addr))
1316 return PAGE_SIZE;
1317
1318 down_read(&current->mm->mmap_sem);
1319 vma = find_vma(current->mm, addr);
1320 if (!vma)
1321 goto out;
1322
1323 size = vma_kernel_pagesize(vma);
1324
1325out:
1326 up_read(&current->mm->mmap_sem);
1327
1328 return size;
1329}
1330
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001331static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1332{
1333 return slot->flags & KVM_MEM_READONLY;
1334}
1335
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001336static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1337 gfn_t *nr_pages, bool write)
Izik Eidus539cb662007-11-11 22:05:04 +02001338{
Marcelo Tosattibc6678a2009-12-23 14:35:21 -02001339 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
Xiao Guangrongca3a4902012-08-21 11:01:50 +08001340 return KVM_HVA_ERR_BAD;
Xiao Guangrong48987782010-08-22 19:11:43 +08001341
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001342 if (memslot_is_readonly(slot) && write)
1343 return KVM_HVA_ERR_RO_BAD;
Xiao Guangrong48987782010-08-22 19:11:43 +08001344
1345 if (nr_pages)
1346 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1347
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001348 return __gfn_to_hva_memslot(slot, gfn);
Izik Eidus539cb662007-11-11 22:05:04 +02001349}
Xiao Guangrong48987782010-08-22 19:11:43 +08001350
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001351static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1352 gfn_t *nr_pages)
1353{
1354 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1355}
1356
1357unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
Stephen Hemminger79408762013-12-29 12:12:29 -08001358 gfn_t gfn)
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001359{
1360 return gfn_to_hva_many(slot, gfn, NULL);
1361}
1362EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1363
Xiao Guangrong48987782010-08-22 19:11:43 +08001364unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1365{
Gleb Natapov49c77542010-10-18 15:22:23 +02001366 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
Xiao Guangrong48987782010-08-22 19:11:43 +08001367}
Sheng Yang0d150292008-04-25 21:44:50 +08001368EXPORT_SYMBOL_GPL(gfn_to_hva);
Izik Eidus539cb662007-11-11 22:05:04 +02001369
Paolo Bonzini8e734852015-05-17 13:58:53 +02001370unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1371{
1372 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1373}
1374EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1375
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001376/*
Paolo Bonziniba6a3542013-09-09 13:52:33 +02001377 * If writable is set to false, the hva returned by this function is only
1378 * allowed to be read.
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001379 */
Christoffer Dall64d83122014-08-19 12:15:00 +02001380unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1381 gfn_t gfn, bool *writable)
Gleb Natapov80300892010-10-19 18:13:41 +02001382{
Gleb Natapova2ac07f2013-10-01 19:58:36 +03001383 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1384
1385 if (!kvm_is_error_hva(hva) && writable)
Paolo Bonziniba6a3542013-09-09 13:52:33 +02001386 *writable = !memslot_is_readonly(slot);
1387
Gleb Natapova2ac07f2013-10-01 19:58:36 +03001388 return hva;
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001389}
1390
Christoffer Dall64d83122014-08-19 12:15:00 +02001391unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1392{
1393 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1394
1395 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1396}
1397
Paolo Bonzini8e734852015-05-17 13:58:53 +02001398unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1399{
1400 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1401
1402 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1403}
1404
Dave Hansend4edcf02016-02-12 13:01:56 -08001405static int get_user_page_nowait(unsigned long start, int write,
1406 struct page **page)
Gleb Natapov0857b9e2011-02-01 13:21:47 +02001407{
Lorenzo Stoakes0d731752016-10-24 10:57:25 +01001408 int flags = FOLL_NOWAIT | FOLL_HWPOISON;
Gleb Natapov0857b9e2011-02-01 13:21:47 +02001409
1410 if (write)
1411 flags |= FOLL_WRITE;
1412
Lorenzo Stoakes0d731752016-10-24 10:57:25 +01001413 return get_user_pages(start, 1, flags, page, NULL);
Gleb Natapov0857b9e2011-02-01 13:21:47 +02001414}
1415
Huang Yingfafc3db2011-01-30 11:15:49 +08001416static inline int check_user_page_hwpoison(unsigned long addr)
1417{
Lorenzo Stoakes0d731752016-10-24 10:57:25 +01001418 int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
Huang Yingfafc3db2011-01-30 11:15:49 +08001419
Lorenzo Stoakes0d731752016-10-24 10:57:25 +01001420 rc = get_user_pages(addr, 1, flags, NULL, NULL);
Huang Yingfafc3db2011-01-30 11:15:49 +08001421 return rc == -EHWPOISON;
1422}
1423
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001424/*
1425 * The atomic path to get the writable pfn which will be stored in @pfn,
1426 * true indicates success, otherwise false is returned.
1427 */
1428static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
Dan Williamsba049e92016-01-15 16:56:11 -08001429 bool write_fault, bool *writable, kvm_pfn_t *pfn)
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001430{
1431 struct page *page[1];
1432 int npages;
1433
1434 if (!(async || atomic))
1435 return false;
1436
Xiao Guangrong12ce13f2012-08-21 11:00:49 +08001437 /*
1438 * Fast pin a writable pfn only if it is a write fault request
1439 * or the caller allows to map a writable pfn for a read fault
1440 * request.
1441 */
1442 if (!(write_fault || writable))
1443 return false;
1444
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001445 npages = __get_user_pages_fast(addr, 1, 1, page);
1446 if (npages == 1) {
1447 *pfn = page_to_pfn(page[0]);
1448
1449 if (writable)
1450 *writable = true;
1451 return true;
1452 }
1453
1454 return false;
1455}
1456
1457/*
1458 * The slow path to get the pfn of the specified host virtual address,
1459 * 1 indicates success, -errno is returned if error is detected.
1460 */
1461static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
Dan Williamsba049e92016-01-15 16:56:11 -08001462 bool *writable, kvm_pfn_t *pfn)
Avi Kivity954bbbc2007-03-30 14:02:32 +03001463{
Anthony Liguori8d4e1282007-10-18 09:59:34 -05001464 struct page *page[1];
Gleb Natapovaf585b92010-10-14 11:22:46 +02001465 int npages = 0;
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001466
1467 might_sleep();
1468
1469 if (writable)
1470 *writable = write_fault;
1471
1472 if (async) {
1473 down_read(&current->mm->mmap_sem);
Dave Hansend4edcf02016-02-12 13:01:56 -08001474 npages = get_user_page_nowait(addr, write_fault, page);
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001475 up_read(&current->mm->mmap_sem);
Lorenzo Stoakesd4944b02016-10-13 01:20:12 +01001476 } else {
1477 unsigned int flags = FOLL_TOUCH | FOLL_HWPOISON;
1478
1479 if (write_fault)
1480 flags |= FOLL_WRITE;
1481
Andrea Arcangeli0664e572015-02-11 15:27:28 -08001482 npages = __get_user_pages_unlocked(current, current->mm, addr, 1,
Lorenzo Stoakesd4944b02016-10-13 01:20:12 +01001483 page, flags);
1484 }
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001485 if (npages != 1)
1486 return npages;
1487
1488 /* map read fault as writable if possible */
Xiao Guangrong12ce13f2012-08-21 11:00:49 +08001489 if (unlikely(!write_fault) && writable) {
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001490 struct page *wpage[1];
1491
1492 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1493 if (npages == 1) {
1494 *writable = true;
1495 put_page(page[0]);
1496 page[0] = wpage[0];
1497 }
1498
1499 npages = 1;
1500 }
1501 *pfn = page_to_pfn(page[0]);
1502 return npages;
1503}
1504
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001505static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1506{
1507 if (unlikely(!(vma->vm_flags & VM_READ)))
1508 return false;
1509
1510 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1511 return false;
1512
1513 return true;
1514}
1515
Paolo Bonzini92176a82016-06-07 16:22:47 +02001516static int hva_to_pfn_remapped(struct vm_area_struct *vma,
1517 unsigned long addr, bool *async,
KarimAllah Ahmed9445fde2018-01-17 19:18:56 +01001518 bool write_fault, bool *writable,
1519 kvm_pfn_t *p_pfn)
Paolo Bonzini92176a82016-06-07 16:22:47 +02001520{
Paolo Bonziniadd6a0c2016-06-07 17:51:18 +02001521 unsigned long pfn;
1522 int r;
1523
1524 r = follow_pfn(vma, addr, &pfn);
1525 if (r) {
1526 /*
1527 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1528 * not call the fault handler, so do it here.
1529 */
1530 bool unlocked = false;
1531 r = fixup_user_fault(current, current->mm, addr,
1532 (write_fault ? FAULT_FLAG_WRITE : 0),
1533 &unlocked);
1534 if (unlocked)
1535 return -EAGAIN;
1536 if (r)
1537 return r;
1538
1539 r = follow_pfn(vma, addr, &pfn);
1540 if (r)
1541 return r;
1542
1543 }
1544
KarimAllah Ahmed9445fde2018-01-17 19:18:56 +01001545 if (writable)
1546 *writable = true;
Paolo Bonziniadd6a0c2016-06-07 17:51:18 +02001547
1548 /*
1549 * Get a reference here because callers of *hva_to_pfn* and
1550 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1551 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
1552 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1553 * simply do nothing for reserved pfns.
1554 *
1555 * Whoever called remap_pfn_range is also going to call e.g.
1556 * unmap_mapping_range before the underlying pages are freed,
1557 * causing a call to our MMU notifier.
1558 */
1559 kvm_get_pfn(pfn);
1560
1561 *p_pfn = pfn;
Paolo Bonzini92176a82016-06-07 16:22:47 +02001562 return 0;
1563}
1564
Xiao Guangrong12ce13f2012-08-21 11:00:49 +08001565/*
1566 * Pin guest page in memory and return its pfn.
1567 * @addr: host virtual address which maps memory to the guest
1568 * @atomic: whether this function can sleep
1569 * @async: whether this function need to wait IO complete if the
1570 * host page is not in the memory
1571 * @write_fault: whether we should get a writable host page
1572 * @writable: whether it allows to map a writable host page for !@write_fault
1573 *
1574 * The function will map a writable host page for these two cases:
1575 * 1): @write_fault = true
1576 * 2): @write_fault = false && @writable, @writable will tell the caller
1577 * whether the mapping is writable.
1578 */
Dan Williamsba049e92016-01-15 16:56:11 -08001579static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001580 bool write_fault, bool *writable)
1581{
1582 struct vm_area_struct *vma;
Dan Williamsba049e92016-01-15 16:56:11 -08001583 kvm_pfn_t pfn = 0;
Paolo Bonzini92176a82016-06-07 16:22:47 +02001584 int npages, r;
Avi Kivity954bbbc2007-03-30 14:02:32 +03001585
Gleb Natapovaf585b92010-10-14 11:22:46 +02001586 /* we can do it either atomically or asynchronously, not both */
1587 BUG_ON(atomic && async);
1588
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001589 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1590 return pfn;
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001591
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001592 if (atomic)
1593 return KVM_PFN_ERR_FAULT;
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001594
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001595 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1596 if (npages == 1)
1597 return pfn;
Gleb Natapovaf585b92010-10-14 11:22:46 +02001598
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001599 down_read(&current->mm->mmap_sem);
1600 if (npages == -EHWPOISON ||
1601 (!async && check_user_page_hwpoison(addr))) {
1602 pfn = KVM_PFN_ERR_HWPOISON;
1603 goto exit;
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001604 }
Izik Eidus539cb662007-11-11 22:05:04 +02001605
Paolo Bonziniadd6a0c2016-06-07 17:51:18 +02001606retry:
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001607 vma = find_vma_intersection(current->mm, addr, addr + 1);
Anthony Liguori8d4e1282007-10-18 09:59:34 -05001608
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001609 if (vma == NULL)
1610 pfn = KVM_PFN_ERR_FAULT;
Paolo Bonzini92176a82016-06-07 16:22:47 +02001611 else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
KarimAllah Ahmed9445fde2018-01-17 19:18:56 +01001612 r = hva_to_pfn_remapped(vma, addr, async, write_fault, writable, &pfn);
Paolo Bonziniadd6a0c2016-06-07 17:51:18 +02001613 if (r == -EAGAIN)
1614 goto retry;
Paolo Bonzini92176a82016-06-07 16:22:47 +02001615 if (r < 0)
1616 pfn = KVM_PFN_ERR_FAULT;
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001617 } else {
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001618 if (async && vma_is_valid(vma, write_fault))
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001619 *async = true;
1620 pfn = KVM_PFN_ERR_FAULT;
1621 }
1622exit:
1623 up_read(&current->mm->mmap_sem);
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001624 return pfn;
Anthony Liguori35149e22008-04-02 14:46:56 -05001625}
1626
Dan Williamsba049e92016-01-15 16:56:11 -08001627kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
1628 bool atomic, bool *async, bool write_fault,
1629 bool *writable)
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001630{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001631 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1632
Paolo Bonzinib2740d32016-02-23 15:36:01 +01001633 if (addr == KVM_HVA_ERR_RO_BAD) {
1634 if (writable)
1635 *writable = false;
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001636 return KVM_PFN_ERR_RO_FAULT;
Paolo Bonzinib2740d32016-02-23 15:36:01 +01001637 }
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001638
Paolo Bonzinib2740d32016-02-23 15:36:01 +01001639 if (kvm_is_error_hva(addr)) {
1640 if (writable)
1641 *writable = false;
Xiao Guangrong81c52c52012-10-16 20:10:59 +08001642 return KVM_PFN_NOSLOT;
Paolo Bonzinib2740d32016-02-23 15:36:01 +01001643 }
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001644
1645 /* Do not map writable pfn in the readonly memslot. */
1646 if (writable && memslot_is_readonly(slot)) {
1647 *writable = false;
1648 writable = NULL;
1649 }
1650
1651 return hva_to_pfn(addr, atomic, async, write_fault,
1652 writable);
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001653}
Paolo Bonzini35204692015-04-02 11:20:48 +02001654EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001655
Dan Williamsba049e92016-01-15 16:56:11 -08001656kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001657 bool *writable)
1658{
Paolo Bonzinie37afc62015-05-19 16:09:04 +02001659 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1660 write_fault, writable);
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001661}
1662EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1663
Dan Williamsba049e92016-01-15 16:56:11 -08001664kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001665{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001666 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001667}
Paolo Bonzinie37afc62015-05-19 16:09:04 +02001668EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001669
Dan Williamsba049e92016-01-15 16:56:11 -08001670kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
Xiao Guangrong037d92d2012-08-21 10:59:12 +08001671{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001672 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
Xiao Guangrong037d92d2012-08-21 10:59:12 +08001673}
1674EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1675
Dan Williamsba049e92016-01-15 16:56:11 -08001676kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
Paolo Bonzinie37afc62015-05-19 16:09:04 +02001677{
1678 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn);
1679}
1680EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1681
Dan Williamsba049e92016-01-15 16:56:11 -08001682kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
Paolo Bonzini8e734852015-05-17 13:58:53 +02001683{
1684 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1685}
1686EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1687
Dan Williamsba049e92016-01-15 16:56:11 -08001688kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
Paolo Bonzinie37afc62015-05-19 16:09:04 +02001689{
1690 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1691}
1692EXPORT_SYMBOL_GPL(gfn_to_pfn);
1693
Dan Williamsba049e92016-01-15 16:56:11 -08001694kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
Paolo Bonzini8e734852015-05-17 13:58:53 +02001695{
1696 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1697}
1698EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1699
Paolo Bonzinid9ef13c2015-05-19 16:01:50 +02001700int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1701 struct page **pages, int nr_pages)
Xiao Guangrong48987782010-08-22 19:11:43 +08001702{
1703 unsigned long addr;
1704 gfn_t entry;
1705
Paolo Bonzinid9ef13c2015-05-19 16:01:50 +02001706 addr = gfn_to_hva_many(slot, gfn, &entry);
Xiao Guangrong48987782010-08-22 19:11:43 +08001707 if (kvm_is_error_hva(addr))
1708 return -1;
1709
1710 if (entry < nr_pages)
1711 return 0;
1712
1713 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1714}
1715EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1716
Dan Williamsba049e92016-01-15 16:56:11 -08001717static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
Xiao Guangronga2766322012-07-26 11:58:59 +08001718{
Xiao Guangrong81c52c52012-10-16 20:10:59 +08001719 if (is_error_noslot_pfn(pfn))
Xiao Guangrong6cede2e2012-08-03 15:41:22 +08001720 return KVM_ERR_PTR_BAD_PAGE;
Xiao Guangronga2766322012-07-26 11:58:59 +08001721
Ard Biesheuvelbf4bea82014-11-10 08:33:56 +00001722 if (kvm_is_reserved_pfn(pfn)) {
Xiao Guangrongcb9aaa32012-08-03 15:42:10 +08001723 WARN_ON(1);
1724 return KVM_ERR_PTR_BAD_PAGE;
1725 }
1726
Xiao Guangronga2766322012-07-26 11:58:59 +08001727 return pfn_to_page(pfn);
1728}
1729
Anthony Liguori35149e22008-04-02 14:46:56 -05001730struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1731{
Dan Williamsba049e92016-01-15 16:56:11 -08001732 kvm_pfn_t pfn;
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001733
1734 pfn = gfn_to_pfn(kvm, gfn);
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001735
Xiao Guangronga2766322012-07-26 11:58:59 +08001736 return kvm_pfn_to_page(pfn);
Avi Kivity954bbbc2007-03-30 14:02:32 +03001737}
1738EXPORT_SYMBOL_GPL(gfn_to_page);
1739
Paolo Bonzini8e734852015-05-17 13:58:53 +02001740struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1741{
Dan Williamsba049e92016-01-15 16:56:11 -08001742 kvm_pfn_t pfn;
Paolo Bonzini8e734852015-05-17 13:58:53 +02001743
1744 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
1745
1746 return kvm_pfn_to_page(pfn);
1747}
1748EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
1749
Izik Eidusb4231d62007-11-20 11:49:33 +02001750void kvm_release_page_clean(struct page *page)
1751{
Xiao Guangrong32cad842012-08-03 15:42:52 +08001752 WARN_ON(is_error_page(page));
1753
Anthony Liguori35149e22008-04-02 14:46:56 -05001754 kvm_release_pfn_clean(page_to_pfn(page));
Izik Eidusb4231d62007-11-20 11:49:33 +02001755}
1756EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1757
Dan Williamsba049e92016-01-15 16:56:11 -08001758void kvm_release_pfn_clean(kvm_pfn_t pfn)
Anthony Liguori35149e22008-04-02 14:46:56 -05001759{
Ard Biesheuvelbf4bea82014-11-10 08:33:56 +00001760 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001761 put_page(pfn_to_page(pfn));
Anthony Liguori35149e22008-04-02 14:46:56 -05001762}
1763EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1764
Izik Eidusb4231d62007-11-20 11:49:33 +02001765void kvm_release_page_dirty(struct page *page)
Izik Eidus8a7ae052007-10-18 11:09:33 +02001766{
Xiao Guangronga2766322012-07-26 11:58:59 +08001767 WARN_ON(is_error_page(page));
1768
Anthony Liguori35149e22008-04-02 14:46:56 -05001769 kvm_release_pfn_dirty(page_to_pfn(page));
Izik Eidus8a7ae052007-10-18 11:09:33 +02001770}
Izik Eidusb4231d62007-11-20 11:49:33 +02001771EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
Izik Eidus8a7ae052007-10-18 11:09:33 +02001772
Dan Williamsba049e92016-01-15 16:56:11 -08001773static void kvm_release_pfn_dirty(kvm_pfn_t pfn)
Anthony Liguori35149e22008-04-02 14:46:56 -05001774{
1775 kvm_set_pfn_dirty(pfn);
1776 kvm_release_pfn_clean(pfn);
1777}
Anthony Liguori35149e22008-04-02 14:46:56 -05001778
Dan Williamsba049e92016-01-15 16:56:11 -08001779void kvm_set_pfn_dirty(kvm_pfn_t pfn)
Anthony Liguori35149e22008-04-02 14:46:56 -05001780{
Sean Christophersone528acd2019-11-11 14:12:27 -08001781 if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn)) {
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001782 struct page *page = pfn_to_page(pfn);
Xiubo Lif95ef0c2015-02-26 14:58:23 +08001783
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001784 if (!PageReserved(page))
1785 SetPageDirty(page);
1786 }
Anthony Liguori35149e22008-04-02 14:46:56 -05001787}
1788EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1789
Dan Williamsba049e92016-01-15 16:56:11 -08001790void kvm_set_pfn_accessed(kvm_pfn_t pfn)
Anthony Liguori35149e22008-04-02 14:46:56 -05001791{
Sean Christophersone528acd2019-11-11 14:12:27 -08001792 if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001793 mark_page_accessed(pfn_to_page(pfn));
Anthony Liguori35149e22008-04-02 14:46:56 -05001794}
1795EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1796
Dan Williamsba049e92016-01-15 16:56:11 -08001797void kvm_get_pfn(kvm_pfn_t pfn)
Anthony Liguori35149e22008-04-02 14:46:56 -05001798{
Ard Biesheuvelbf4bea82014-11-10 08:33:56 +00001799 if (!kvm_is_reserved_pfn(pfn))
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001800 get_page(pfn_to_page(pfn));
Anthony Liguori35149e22008-04-02 14:46:56 -05001801}
1802EXPORT_SYMBOL_GPL(kvm_get_pfn);
1803
Izik Eidus195aefd2007-10-01 22:14:18 +02001804static int next_segment(unsigned long len, int offset)
1805{
1806 if (len > PAGE_SIZE - offset)
1807 return PAGE_SIZE - offset;
1808 else
1809 return len;
1810}
1811
Paolo Bonzini8e734852015-05-17 13:58:53 +02001812static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
1813 void *data, int offset, int len)
Izik Eidus195aefd2007-10-01 22:14:18 +02001814{
Izik Eiduse0506bc2007-11-11 22:10:22 +02001815 int r;
1816 unsigned long addr;
Izik Eidus195aefd2007-10-01 22:14:18 +02001817
Paolo Bonzini8e734852015-05-17 13:58:53 +02001818 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
Izik Eiduse0506bc2007-11-11 22:10:22 +02001819 if (kvm_is_error_hva(addr))
Izik Eidus195aefd2007-10-01 22:14:18 +02001820 return -EFAULT;
Paolo Bonzini3180a7f2015-04-02 14:08:20 +02001821 r = __copy_from_user(data, (void __user *)addr + offset, len);
Izik Eiduse0506bc2007-11-11 22:10:22 +02001822 if (r)
1823 return -EFAULT;
Izik Eidus195aefd2007-10-01 22:14:18 +02001824 return 0;
1825}
Paolo Bonzini8e734852015-05-17 13:58:53 +02001826
1827int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1828 int len)
1829{
1830 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1831
1832 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1833}
Izik Eidus195aefd2007-10-01 22:14:18 +02001834EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1835
Paolo Bonzini8e734852015-05-17 13:58:53 +02001836int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
1837 int offset, int len)
1838{
1839 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1840
1841 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1842}
1843EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
1844
Izik Eidus195aefd2007-10-01 22:14:18 +02001845int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1846{
1847 gfn_t gfn = gpa >> PAGE_SHIFT;
1848 int seg;
1849 int offset = offset_in_page(gpa);
1850 int ret;
1851
1852 while ((seg = next_segment(len, offset)) != 0) {
1853 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1854 if (ret < 0)
1855 return ret;
1856 offset = 0;
1857 len -= seg;
1858 data += seg;
1859 ++gfn;
1860 }
1861 return 0;
1862}
1863EXPORT_SYMBOL_GPL(kvm_read_guest);
1864
Paolo Bonzini8e734852015-05-17 13:58:53 +02001865int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
1866{
1867 gfn_t gfn = gpa >> PAGE_SHIFT;
1868 int seg;
1869 int offset = offset_in_page(gpa);
1870 int ret;
1871
1872 while ((seg = next_segment(len, offset)) != 0) {
1873 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
1874 if (ret < 0)
1875 return ret;
1876 offset = 0;
1877 len -= seg;
1878 data += seg;
1879 ++gfn;
1880 }
1881 return 0;
1882}
1883EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
1884
1885static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1886 void *data, int offset, unsigned long len)
Marcelo Tosatti7ec54582007-12-20 19:18:23 -05001887{
1888 int r;
1889 unsigned long addr;
Marcelo Tosatti7ec54582007-12-20 19:18:23 -05001890
Paolo Bonzini8e734852015-05-17 13:58:53 +02001891 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
Marcelo Tosatti7ec54582007-12-20 19:18:23 -05001892 if (kvm_is_error_hva(addr))
1893 return -EFAULT;
Andrea Arcangeli0aac03f2008-01-30 19:57:35 +01001894 pagefault_disable();
Paolo Bonzini3180a7f2015-04-02 14:08:20 +02001895 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
Andrea Arcangeli0aac03f2008-01-30 19:57:35 +01001896 pagefault_enable();
Marcelo Tosatti7ec54582007-12-20 19:18:23 -05001897 if (r)
1898 return -EFAULT;
1899 return 0;
1900}
Marcelo Tosatti7ec54582007-12-20 19:18:23 -05001901
Paolo Bonzini8e734852015-05-17 13:58:53 +02001902int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1903 unsigned long len)
1904{
1905 gfn_t gfn = gpa >> PAGE_SHIFT;
1906 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1907 int offset = offset_in_page(gpa);
1908
1909 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1910}
1911EXPORT_SYMBOL_GPL(kvm_read_guest_atomic);
1912
1913int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
1914 void *data, unsigned long len)
1915{
1916 gfn_t gfn = gpa >> PAGE_SHIFT;
1917 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1918 int offset = offset_in_page(gpa);
1919
1920 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1921}
1922EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
1923
1924static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
1925 const void *data, int offset, int len)
Izik Eidus195aefd2007-10-01 22:14:18 +02001926{
Izik Eiduse0506bc2007-11-11 22:10:22 +02001927 int r;
1928 unsigned long addr;
Izik Eidus195aefd2007-10-01 22:14:18 +02001929
Radim Krčmář251eb842015-04-10 21:47:27 +02001930 addr = gfn_to_hva_memslot(memslot, gfn);
Izik Eiduse0506bc2007-11-11 22:10:22 +02001931 if (kvm_is_error_hva(addr))
Izik Eidus195aefd2007-10-01 22:14:18 +02001932 return -EFAULT;
Xiao Guangrong8b0cedf2011-05-15 23:22:04 +08001933 r = __copy_to_user((void __user *)addr + offset, data, len);
Izik Eiduse0506bc2007-11-11 22:10:22 +02001934 if (r)
1935 return -EFAULT;
Paolo Bonzinibc009e42015-05-26 12:43:41 +02001936 mark_page_dirty_in_slot(memslot, gfn);
Izik Eidus195aefd2007-10-01 22:14:18 +02001937 return 0;
1938}
Paolo Bonzini8e734852015-05-17 13:58:53 +02001939
1940int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
1941 const void *data, int offset, int len)
1942{
1943 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1944
1945 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1946}
Izik Eidus195aefd2007-10-01 22:14:18 +02001947EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1948
Paolo Bonzini8e734852015-05-17 13:58:53 +02001949int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
1950 const void *data, int offset, int len)
1951{
1952 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1953
1954 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1955}
1956EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
1957
Izik Eidus195aefd2007-10-01 22:14:18 +02001958int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1959 unsigned long len)
1960{
1961 gfn_t gfn = gpa >> PAGE_SHIFT;
1962 int seg;
1963 int offset = offset_in_page(gpa);
1964 int ret;
1965
1966 while ((seg = next_segment(len, offset)) != 0) {
1967 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1968 if (ret < 0)
1969 return ret;
1970 offset = 0;
1971 len -= seg;
1972 data += seg;
1973 ++gfn;
1974 }
1975 return 0;
1976}
Wincy Vanff651cb2014-12-11 08:52:58 +03001977EXPORT_SYMBOL_GPL(kvm_write_guest);
Izik Eidus195aefd2007-10-01 22:14:18 +02001978
Paolo Bonzini8e734852015-05-17 13:58:53 +02001979int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
1980 unsigned long len)
1981{
1982 gfn_t gfn = gpa >> PAGE_SHIFT;
1983 int seg;
1984 int offset = offset_in_page(gpa);
1985 int ret;
1986
1987 while ((seg = next_segment(len, offset)) != 0) {
1988 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
1989 if (ret < 0)
1990 return ret;
1991 offset = 0;
1992 len -= seg;
1993 data += seg;
1994 ++gfn;
1995 }
1996 return 0;
1997}
1998EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
1999
Gleb Natapov49c77542010-10-18 15:22:23 +02002000int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
Andrew Honig8f964522013-03-29 09:35:21 -07002001 gpa_t gpa, unsigned long len)
Gleb Natapov49c77542010-10-18 15:22:23 +02002002{
2003 struct kvm_memslots *slots = kvm_memslots(kvm);
2004 int offset = offset_in_page(gpa);
Andrew Honig8f964522013-03-29 09:35:21 -07002005 gfn_t start_gfn = gpa >> PAGE_SHIFT;
2006 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
2007 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
2008 gfn_t nr_pages_avail;
Gleb Natapov49c77542010-10-18 15:22:23 +02002009
2010 ghc->gpa = gpa;
2011 ghc->generation = slots->generation;
Andrew Honig8f964522013-03-29 09:35:21 -07002012 ghc->len = len;
2013 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
Radim Krčmářca3f0872015-04-08 14:16:48 +02002014 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, NULL);
2015 if (!kvm_is_error_hva(ghc->hva) && nr_pages_needed <= 1) {
Gleb Natapov49c77542010-10-18 15:22:23 +02002016 ghc->hva += offset;
Andrew Honig8f964522013-03-29 09:35:21 -07002017 } else {
2018 /*
2019 * If the requested region crosses two memslots, we still
2020 * verify that the entire region is valid here.
2021 */
2022 while (start_gfn <= end_gfn) {
2023 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
2024 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
2025 &nr_pages_avail);
2026 if (kvm_is_error_hva(ghc->hva))
2027 return -EFAULT;
2028 start_gfn += nr_pages_avail;
2029 }
2030 /* Use the slow path for cross page reads and writes. */
2031 ghc->memslot = NULL;
2032 }
Gleb Natapov49c77542010-10-18 15:22:23 +02002033 return 0;
2034}
2035EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
2036
2037int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2038 void *data, unsigned long len)
2039{
2040 struct kvm_memslots *slots = kvm_memslots(kvm);
2041 int r;
2042
Andrew Honig8f964522013-03-29 09:35:21 -07002043 BUG_ON(len > ghc->len);
2044
Gleb Natapov49c77542010-10-18 15:22:23 +02002045 if (slots->generation != ghc->generation)
Andrew Honig8f964522013-03-29 09:35:21 -07002046 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
2047
2048 if (unlikely(!ghc->memslot))
2049 return kvm_write_guest(kvm, ghc->gpa, data, len);
Gleb Natapov49c77542010-10-18 15:22:23 +02002050
2051 if (kvm_is_error_hva(ghc->hva))
2052 return -EFAULT;
2053
Xiao Guangrong8b0cedf2011-05-15 23:22:04 +08002054 r = __copy_to_user((void __user *)ghc->hva, data, len);
Gleb Natapov49c77542010-10-18 15:22:23 +02002055 if (r)
2056 return -EFAULT;
Paolo Bonzinibc009e42015-05-26 12:43:41 +02002057 mark_page_dirty_in_slot(ghc->memslot, ghc->gpa >> PAGE_SHIFT);
Gleb Natapov49c77542010-10-18 15:22:23 +02002058
2059 return 0;
2060}
2061EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
2062
Gleb Natapove03b6442011-07-11 15:28:11 -04002063int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2064 void *data, unsigned long len)
2065{
2066 struct kvm_memslots *slots = kvm_memslots(kvm);
2067 int r;
2068
Andrew Honig8f964522013-03-29 09:35:21 -07002069 BUG_ON(len > ghc->len);
2070
Gleb Natapove03b6442011-07-11 15:28:11 -04002071 if (slots->generation != ghc->generation)
Andrew Honig8f964522013-03-29 09:35:21 -07002072 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
2073
2074 if (unlikely(!ghc->memslot))
2075 return kvm_read_guest(kvm, ghc->gpa, data, len);
Gleb Natapove03b6442011-07-11 15:28:11 -04002076
2077 if (kvm_is_error_hva(ghc->hva))
2078 return -EFAULT;
2079
2080 r = __copy_from_user(data, (void __user *)ghc->hva, len);
2081 if (r)
2082 return -EFAULT;
2083
2084 return 0;
2085}
2086EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
2087
Izik Eidus195aefd2007-10-01 22:14:18 +02002088int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
2089{
Heiko Carstens8a3caa62013-11-18 10:35:55 +01002090 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2091
2092 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
Izik Eidus195aefd2007-10-01 22:14:18 +02002093}
2094EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
2095
2096int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
2097{
2098 gfn_t gfn = gpa >> PAGE_SHIFT;
2099 int seg;
2100 int offset = offset_in_page(gpa);
2101 int ret;
2102
Kevin Mulveybfda0e82015-02-20 08:21:36 -05002103 while ((seg = next_segment(len, offset)) != 0) {
Izik Eidus195aefd2007-10-01 22:14:18 +02002104 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
2105 if (ret < 0)
2106 return ret;
2107 offset = 0;
2108 len -= seg;
2109 ++gfn;
2110 }
2111 return 0;
2112}
2113EXPORT_SYMBOL_GPL(kvm_clear_guest);
2114
Paolo Bonzinibc009e42015-05-26 12:43:41 +02002115static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
Stephen Hemminger79408762013-12-29 12:12:29 -08002116 gfn_t gfn)
Avi Kivity6aa8b732006-12-10 02:21:36 -08002117{
Rusty Russell7e9d6192007-07-31 20:41:14 +10002118 if (memslot && memslot->dirty_bitmap) {
2119 unsigned long rel_gfn = gfn - memslot->base_gfn;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002120
Takuya Yoshikawab74ca3b2012-10-04 17:13:12 -07002121 set_bit_le(rel_gfn, memslot->dirty_bitmap);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002122 }
2123}
2124
Gleb Natapov49c77542010-10-18 15:22:23 +02002125void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
2126{
2127 struct kvm_memory_slot *memslot;
2128
2129 memslot = gfn_to_memslot(kvm, gfn);
Paolo Bonzinibc009e42015-05-26 12:43:41 +02002130 mark_page_dirty_in_slot(memslot, gfn);
Gleb Natapov49c77542010-10-18 15:22:23 +02002131}
Aneesh Kumar K.V2ba9f0d2013-10-07 22:17:59 +05302132EXPORT_SYMBOL_GPL(mark_page_dirty);
Gleb Natapov49c77542010-10-18 15:22:23 +02002133
Paolo Bonzini8e734852015-05-17 13:58:53 +02002134void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
2135{
2136 struct kvm_memory_slot *memslot;
2137
2138 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2139 mark_page_dirty_in_slot(memslot, gfn);
2140}
2141EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
2142
Wanpeng Liaca6ff22015-09-03 22:07:38 +08002143static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
2144{
Christian Borntraeger6b6de682016-02-09 13:47:55 +01002145 unsigned int old, val, grow;
Wanpeng Liaca6ff22015-09-03 22:07:38 +08002146
Wanpeng Li2cbd7822015-09-03 22:07:39 +08002147 old = val = vcpu->halt_poll_ns;
Christian Borntraeger6b6de682016-02-09 13:47:55 +01002148 grow = READ_ONCE(halt_poll_ns_grow);
Wanpeng Liaca6ff22015-09-03 22:07:38 +08002149 /* 10us base */
Christian Borntraeger6b6de682016-02-09 13:47:55 +01002150 if (val == 0 && grow)
Wanpeng Liaca6ff22015-09-03 22:07:38 +08002151 val = 10000;
2152 else
Christian Borntraeger6b6de682016-02-09 13:47:55 +01002153 val *= grow;
Wanpeng Liaca6ff22015-09-03 22:07:38 +08002154
David Matlack313f6362016-03-08 16:19:44 -08002155 if (val > halt_poll_ns)
2156 val = halt_poll_ns;
2157
Wanpeng Liaca6ff22015-09-03 22:07:38 +08002158 vcpu->halt_poll_ns = val;
Wanpeng Li2cbd7822015-09-03 22:07:39 +08002159 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
Wanpeng Liaca6ff22015-09-03 22:07:38 +08002160}
2161
2162static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
2163{
Christian Borntraeger6b6de682016-02-09 13:47:55 +01002164 unsigned int old, val, shrink;
Wanpeng Liaca6ff22015-09-03 22:07:38 +08002165
Wanpeng Li2cbd7822015-09-03 22:07:39 +08002166 old = val = vcpu->halt_poll_ns;
Christian Borntraeger6b6de682016-02-09 13:47:55 +01002167 shrink = READ_ONCE(halt_poll_ns_shrink);
2168 if (shrink == 0)
Wanpeng Liaca6ff22015-09-03 22:07:38 +08002169 val = 0;
2170 else
Christian Borntraeger6b6de682016-02-09 13:47:55 +01002171 val /= shrink;
Wanpeng Liaca6ff22015-09-03 22:07:38 +08002172
2173 vcpu->halt_poll_ns = val;
Wanpeng Li2cbd7822015-09-03 22:07:39 +08002174 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
Wanpeng Liaca6ff22015-09-03 22:07:38 +08002175}
2176
Paolo Bonzinif7819512015-02-04 18:20:58 +01002177static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
2178{
2179 if (kvm_arch_vcpu_runnable(vcpu)) {
2180 kvm_make_request(KVM_REQ_UNHALT, vcpu);
2181 return -EINTR;
2182 }
2183 if (kvm_cpu_has_pending_timer(vcpu))
2184 return -EINTR;
2185 if (signal_pending(current))
2186 return -EINTR;
2187
2188 return 0;
2189}
2190
Eddie Dongb6958ce2007-07-18 12:15:21 +03002191/*
2192 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2193 */
Hollis Blanchard8776e512007-10-31 17:24:24 -05002194void kvm_vcpu_block(struct kvm_vcpu *vcpu)
Eddie Dongb6958ce2007-07-18 12:15:21 +03002195{
Paolo Bonzinif7819512015-02-04 18:20:58 +01002196 ktime_t start, cur;
Marcelo Tosatti85773702016-02-19 09:46:39 +01002197 DECLARE_SWAITQUEUE(wait);
Paolo Bonzinif7819512015-02-04 18:20:58 +01002198 bool waited = false;
Wanpeng Liaca6ff22015-09-03 22:07:38 +08002199 u64 block_ns;
Paolo Bonzinif7819512015-02-04 18:20:58 +01002200
2201 start = cur = ktime_get();
Wanpeng Li19020f82015-09-03 22:07:37 +08002202 if (vcpu->halt_poll_ns) {
2203 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
Xiubo Lif95ef0c2015-02-26 14:58:23 +08002204
Paolo Bonzini62bea5b2015-09-15 18:27:57 +02002205 ++vcpu->stat.halt_attempted_poll;
Paolo Bonzinif7819512015-02-04 18:20:58 +01002206 do {
2207 /*
2208 * This sets KVM_REQ_UNHALT if an interrupt
2209 * arrives.
2210 */
2211 if (kvm_vcpu_check_block(vcpu) < 0) {
2212 ++vcpu->stat.halt_successful_poll;
Christian Borntraeger3491caf2016-05-13 12:16:35 +02002213 if (!vcpu_valid_wakeup(vcpu))
2214 ++vcpu->stat.halt_poll_invalid;
Paolo Bonzinif7819512015-02-04 18:20:58 +01002215 goto out;
2216 }
2217 cur = ktime_get();
2218 } while (single_task_running() && ktime_before(cur, stop));
2219 }
Eddie Dongb6958ce2007-07-18 12:15:21 +03002220
Christoffer Dall3217f7c2015-08-27 16:41:15 +02002221 kvm_arch_vcpu_blocking(vcpu);
2222
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03002223 for (;;) {
Marcelo Tosatti85773702016-02-19 09:46:39 +01002224 prepare_to_swait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
Eddie Dongb6958ce2007-07-18 12:15:21 +03002225
Paolo Bonzinif7819512015-02-04 18:20:58 +01002226 if (kvm_vcpu_check_block(vcpu) < 0)
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03002227 break;
2228
Paolo Bonzinif7819512015-02-04 18:20:58 +01002229 waited = true;
Eddie Dongb6958ce2007-07-18 12:15:21 +03002230 schedule();
Eddie Dongb6958ce2007-07-18 12:15:21 +03002231 }
2232
Marcelo Tosatti85773702016-02-19 09:46:39 +01002233 finish_swait(&vcpu->wq, &wait);
Paolo Bonzinif7819512015-02-04 18:20:58 +01002234 cur = ktime_get();
2235
Christoffer Dall3217f7c2015-08-27 16:41:15 +02002236 kvm_arch_vcpu_unblocking(vcpu);
Paolo Bonzinif7819512015-02-04 18:20:58 +01002237out:
Wanpeng Liaca6ff22015-09-03 22:07:38 +08002238 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2239
Christian Borntraeger2086d322016-05-17 10:49:22 +02002240 if (!vcpu_valid_wakeup(vcpu))
2241 shrink_halt_poll_ns(vcpu);
2242 else if (halt_poll_ns) {
Wanpeng Liaca6ff22015-09-03 22:07:38 +08002243 if (block_ns <= vcpu->halt_poll_ns)
2244 ;
2245 /* we had a long block, shrink polling */
Christian Borntraeger2086d322016-05-17 10:49:22 +02002246 else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns)
Wanpeng Liaca6ff22015-09-03 22:07:38 +08002247 shrink_halt_poll_ns(vcpu);
2248 /* we had a short halt and our poll time is too small */
2249 else if (vcpu->halt_poll_ns < halt_poll_ns &&
2250 block_ns < halt_poll_ns)
2251 grow_halt_poll_ns(vcpu);
Wanpeng Liedb92722015-09-14 17:38:51 +08002252 } else
2253 vcpu->halt_poll_ns = 0;
Wanpeng Liaca6ff22015-09-03 22:07:38 +08002254
Christian Borntraeger3491caf2016-05-13 12:16:35 +02002255 trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
2256 kvm_arch_vcpu_block_finish(vcpu);
Eddie Dongb6958ce2007-07-18 12:15:21 +03002257}
Aneesh Kumar K.V2ba9f0d2013-10-07 22:17:59 +05302258EXPORT_SYMBOL_GPL(kvm_vcpu_block);
Eddie Dongb6958ce2007-07-18 12:15:21 +03002259
Marcelo Tosatti8c847802012-03-14 17:58:48 -03002260#ifndef CONFIG_S390
Radim Krčmářdd1a4cc2016-05-04 14:09:44 -05002261void kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
Christoffer Dallb6d33832012-03-08 16:44:24 -05002262{
Marcelo Tosatti85773702016-02-19 09:46:39 +01002263 struct swait_queue_head *wqp;
Christoffer Dallb6d33832012-03-08 16:44:24 -05002264
2265 wqp = kvm_arch_vcpu_wq(vcpu);
Marcelo Tosatti85773702016-02-19 09:46:39 +01002266 if (swait_active(wqp)) {
2267 swake_up(wqp);
Christoffer Dallb6d33832012-03-08 16:44:24 -05002268 ++vcpu->stat.halt_wakeup;
2269 }
2270
Radim Krčmářdd1a4cc2016-05-04 14:09:44 -05002271}
2272EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
2273
2274/*
2275 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2276 */
2277void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2278{
2279 int me;
2280 int cpu = vcpu->cpu;
2281
2282 kvm_vcpu_wake_up(vcpu);
Christoffer Dallb6d33832012-03-08 16:44:24 -05002283 me = get_cpu();
2284 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2285 if (kvm_arch_vcpu_should_kick(vcpu))
2286 smp_send_reschedule(cpu);
2287 put_cpu();
2288}
Yang Zhanga20ed542013-04-11 19:25:15 +08002289EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
Marcelo Tosatti8c847802012-03-14 17:58:48 -03002290#endif /* !CONFIG_S390 */
Christoffer Dallb6d33832012-03-08 16:44:24 -05002291
Dan Carpenterfa933842014-05-23 13:20:42 +03002292int kvm_vcpu_yield_to(struct kvm_vcpu *target)
Konstantin Weitz41628d32012-04-25 15:30:38 +02002293{
2294 struct pid *pid;
2295 struct task_struct *task = NULL;
Dan Carpenterfa933842014-05-23 13:20:42 +03002296 int ret = 0;
Konstantin Weitz41628d32012-04-25 15:30:38 +02002297
2298 rcu_read_lock();
2299 pid = rcu_dereference(target->pid);
2300 if (pid)
Sam Bobroff27fbe64b2014-09-19 09:40:41 +10002301 task = get_pid_task(pid, PIDTYPE_PID);
Konstantin Weitz41628d32012-04-25 15:30:38 +02002302 rcu_read_unlock();
2303 if (!task)
Raghavendra K Tc45c5282013-01-22 13:09:24 +05302304 return ret;
Raghavendra K Tc45c5282013-01-22 13:09:24 +05302305 ret = yield_to(task, 1);
Konstantin Weitz41628d32012-04-25 15:30:38 +02002306 put_task_struct(task);
Raghavendra K Tc45c5282013-01-22 13:09:24 +05302307
2308 return ret;
Konstantin Weitz41628d32012-04-25 15:30:38 +02002309}
2310EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2311
Raghavendra K T06e48c52012-07-19 15:17:52 +05302312/*
2313 * Helper that checks whether a VCPU is eligible for directed yield.
2314 * Most eligible candidate to yield is decided by following heuristics:
2315 *
2316 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2317 * (preempted lock holder), indicated by @in_spin_loop.
2318 * Set at the beiginning and cleared at the end of interception/PLE handler.
2319 *
2320 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2321 * chance last time (mostly it has become eligible now since we have probably
2322 * yielded to lockholder in last iteration. This is done by toggling
2323 * @dy_eligible each time a VCPU checked for eligibility.)
2324 *
2325 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2326 * to preempted lock-holder could result in wrong VCPU selection and CPU
2327 * burning. Giving priority for a potential lock-holder increases lock
2328 * progress.
2329 *
2330 * Since algorithm is based on heuristics, accessing another VCPU data without
2331 * locking does not harm. It may result in trying to yield to same VCPU, fail
2332 * and continue with next VCPU and so on.
2333 */
Stephen Hemminger79408762013-12-29 12:12:29 -08002334static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
Raghavendra K T06e48c52012-07-19 15:17:52 +05302335{
Scott Wood4a55dd72014-01-09 18:43:16 -06002336#ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
Raghavendra K T06e48c52012-07-19 15:17:52 +05302337 bool eligible;
2338
2339 eligible = !vcpu->spin_loop.in_spin_loop ||
Christian Borntraeger34656112014-09-04 21:13:31 +02002340 vcpu->spin_loop.dy_eligible;
Raghavendra K T06e48c52012-07-19 15:17:52 +05302341
2342 if (vcpu->spin_loop.in_spin_loop)
2343 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2344
2345 return eligible;
Scott Wood4a55dd72014-01-09 18:43:16 -06002346#else
2347 return true;
Raghavendra K T06e48c52012-07-19 15:17:52 +05302348#endif
Scott Wood4a55dd72014-01-09 18:43:16 -06002349}
Raghavendra K Tc45c5282013-01-22 13:09:24 +05302350
Rik van Riel217ece62011-02-01 09:53:28 -05002351void kvm_vcpu_on_spin(struct kvm_vcpu *me)
Zhai, Edwind255f4f2009-10-09 18:03:20 +08002352{
Rik van Riel217ece62011-02-01 09:53:28 -05002353 struct kvm *kvm = me->kvm;
2354 struct kvm_vcpu *vcpu;
2355 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2356 int yielded = 0;
Raghavendra K Tc45c5282013-01-22 13:09:24 +05302357 int try = 3;
Rik van Riel217ece62011-02-01 09:53:28 -05002358 int pass;
2359 int i;
Zhai, Edwind255f4f2009-10-09 18:03:20 +08002360
Raghavendra K T4c088492012-07-18 19:07:46 +05302361 kvm_vcpu_set_in_spin_loop(me, true);
Rik van Riel217ece62011-02-01 09:53:28 -05002362 /*
2363 * We boost the priority of a VCPU that is runnable but not
2364 * currently running, because it got preempted by something
2365 * else and called schedule in __vcpu_run. Hopefully that
2366 * VCPU is holding the lock that we need and will release it.
2367 * We approximate round-robin by starting at the last boosted VCPU.
2368 */
Raghavendra K Tc45c5282013-01-22 13:09:24 +05302369 for (pass = 0; pass < 2 && !yielded && try; pass++) {
Rik van Riel217ece62011-02-01 09:53:28 -05002370 kvm_for_each_vcpu(i, vcpu, kvm) {
Rik van Riel5cfc2aa2012-06-19 16:51:04 -04002371 if (!pass && i <= last_boosted_vcpu) {
Rik van Riel217ece62011-02-01 09:53:28 -05002372 i = last_boosted_vcpu;
2373 continue;
2374 } else if (pass && i > last_boosted_vcpu)
2375 break;
Raghavendra K T7bc7ae22013-03-04 23:32:27 +05302376 if (!ACCESS_ONCE(vcpu->preempted))
2377 continue;
Rik van Riel217ece62011-02-01 09:53:28 -05002378 if (vcpu == me)
2379 continue;
Marcelo Tosatti85773702016-02-19 09:46:39 +01002380 if (swait_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
Rik van Riel217ece62011-02-01 09:53:28 -05002381 continue;
Raghavendra K T06e48c52012-07-19 15:17:52 +05302382 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2383 continue;
Raghavendra K Tc45c5282013-01-22 13:09:24 +05302384
2385 yielded = kvm_vcpu_yield_to(vcpu);
2386 if (yielded > 0) {
Rik van Riel217ece62011-02-01 09:53:28 -05002387 kvm->last_boosted_vcpu = i;
Rik van Riel217ece62011-02-01 09:53:28 -05002388 break;
Raghavendra K Tc45c5282013-01-22 13:09:24 +05302389 } else if (yielded < 0) {
2390 try--;
2391 if (!try)
2392 break;
Rik van Riel217ece62011-02-01 09:53:28 -05002393 }
Rik van Riel217ece62011-02-01 09:53:28 -05002394 }
2395 }
Raghavendra K T4c088492012-07-18 19:07:46 +05302396 kvm_vcpu_set_in_spin_loop(me, false);
Raghavendra K T06e48c52012-07-19 15:17:52 +05302397
2398 /* Ensure vcpu is not eligible during next spinloop */
2399 kvm_vcpu_set_dy_eligible(me, false);
Zhai, Edwind255f4f2009-10-09 18:03:20 +08002400}
2401EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2402
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11002403static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02002404{
2405 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02002406 struct page *page;
2407
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11002408 if (vmf->pgoff == 0)
Avi Kivity039576c2007-03-20 12:46:50 +02002409 page = virt_to_page(vcpu->run);
Avi Kivity09566762008-01-23 18:14:23 +02002410#ifdef CONFIG_X86
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11002411 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
Zhang Xiantaoad312c72007-12-13 23:50:52 +08002412 page = virt_to_page(vcpu->arch.pio_data);
Avi Kivity09566762008-01-23 18:14:23 +02002413#endif
Laurent Vivier5f94c172008-05-30 16:05:54 +02002414#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2415 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2416 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2417#endif
Avi Kivity039576c2007-03-20 12:46:50 +02002418 else
Carsten Otte5b1c1492012-01-04 10:25:23 +01002419 return kvm_arch_vcpu_fault(vcpu, vmf);
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02002420 get_page(page);
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11002421 vmf->page = page;
2422 return 0;
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02002423}
2424
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04002425static const struct vm_operations_struct kvm_vcpu_vm_ops = {
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11002426 .fault = kvm_vcpu_fault,
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02002427};
2428
2429static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2430{
2431 vma->vm_ops = &kvm_vcpu_vm_ops;
2432 return 0;
2433}
2434
Avi Kivitybccf2152007-02-21 18:04:26 +02002435static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2436{
2437 struct kvm_vcpu *vcpu = filp->private_data;
2438
Luiz Capitulino45b59392016-09-16 10:27:35 -04002439 debugfs_remove_recursive(vcpu->debugfs_dentry);
Al Viro66c0b392008-04-19 20:33:56 +01002440 kvm_put_kvm(vcpu->kvm);
Avi Kivitybccf2152007-02-21 18:04:26 +02002441 return 0;
2442}
2443
Christian Borntraeger3d3aab12008-12-02 11:17:32 +01002444static struct file_operations kvm_vcpu_fops = {
Avi Kivitybccf2152007-02-21 18:04:26 +02002445 .release = kvm_vcpu_release,
2446 .unlocked_ioctl = kvm_vcpu_ioctl,
Christian Borntraegerde8e5d72015-02-03 09:35:15 +01002447#ifdef CONFIG_KVM_COMPAT
Alexander Graf1dda6062011-06-08 02:45:37 +02002448 .compat_ioctl = kvm_vcpu_compat_ioctl,
2449#endif
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02002450 .mmap = kvm_vcpu_mmap,
Arnd Bergmann6038f372010-08-15 18:52:59 +02002451 .llseek = noop_llseek,
Avi Kivitybccf2152007-02-21 18:04:26 +02002452};
2453
2454/*
2455 * Allocates an inode for the vcpu.
2456 */
2457static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2458{
Yann Droneaud24009b02013-08-24 22:14:07 +02002459 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
Avi Kivitybccf2152007-02-21 18:04:26 +02002460}
2461
Luiz Capitulino45b59392016-09-16 10:27:35 -04002462static int kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
2463{
2464 char dir_name[ITOA_MAX_LEN * 2];
2465 int ret;
2466
2467 if (!kvm_arch_has_vcpu_debugfs())
2468 return 0;
2469
2470 if (!debugfs_initialized())
2471 return 0;
2472
2473 snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
2474 vcpu->debugfs_dentry = debugfs_create_dir(dir_name,
2475 vcpu->kvm->debugfs_dentry);
2476 if (!vcpu->debugfs_dentry)
2477 return -ENOMEM;
2478
2479 ret = kvm_arch_create_vcpu_debugfs(vcpu);
2480 if (ret < 0) {
2481 debugfs_remove_recursive(vcpu->debugfs_dentry);
2482 return ret;
2483 }
2484
2485 return 0;
2486}
2487
Avi Kivityc5ea7662007-02-20 18:41:05 +02002488/*
2489 * Creates some virtual cpus. Good luck creating more than one.
2490 */
Gleb Natapov73880c82009-06-09 15:56:28 +03002491static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
Avi Kivityc5ea7662007-02-20 18:41:05 +02002492{
2493 int r;
David Hildenbrande09fefd2015-11-05 09:03:50 +01002494 struct kvm_vcpu *vcpu;
Avi Kivityc5ea7662007-02-20 18:41:05 +02002495
Greg Kurz0b1b1df2016-05-09 18:13:37 +02002496 if (id >= KVM_MAX_VCPU_ID)
Andy Honig338c7db2013-11-18 16:09:22 -08002497 return -EINVAL;
2498
Paolo Bonzini6c7caeb2016-06-13 14:48:25 +02002499 mutex_lock(&kvm->lock);
2500 if (kvm->created_vcpus == KVM_MAX_VCPUS) {
2501 mutex_unlock(&kvm->lock);
2502 return -EINVAL;
2503 }
2504
2505 kvm->created_vcpus++;
2506 mutex_unlock(&kvm->lock);
2507
Gleb Natapov73880c82009-06-09 15:56:28 +03002508 vcpu = kvm_arch_vcpu_create(kvm, id);
Paolo Bonzini6c7caeb2016-06-13 14:48:25 +02002509 if (IS_ERR(vcpu)) {
2510 r = PTR_ERR(vcpu);
2511 goto vcpu_decrement;
2512 }
Avi Kivityc5ea7662007-02-20 18:41:05 +02002513
Avi Kivity15ad7142007-07-11 18:17:21 +03002514 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2515
Avi Kivity26e52152007-11-20 15:30:24 +02002516 r = kvm_arch_vcpu_setup(vcpu);
2517 if (r)
Jan Kiszkad7805922011-05-23 10:33:05 +02002518 goto vcpu_destroy;
Avi Kivity26e52152007-11-20 15:30:24 +02002519
Luiz Capitulino45b59392016-09-16 10:27:35 -04002520 r = kvm_create_vcpu_debugfs(vcpu);
2521 if (r)
2522 goto vcpu_destroy;
2523
Shaohua Li11ec2802007-07-23 14:51:37 +08002524 mutex_lock(&kvm->lock);
David Hildenbrande09fefd2015-11-05 09:03:50 +01002525 if (kvm_get_vcpu_by_id(kvm, id)) {
2526 r = -EEXIST;
2527 goto unlock_vcpu_destroy;
2528 }
Gleb Natapov73880c82009-06-09 15:56:28 +03002529
2530 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
Rusty Russellfb3f0f52007-07-27 17:16:56 +10002531
2532 /* Now it's all set up, let userspace reach it */
Al Viro66c0b392008-04-19 20:33:56 +01002533 kvm_get_kvm(kvm);
Avi Kivitybccf2152007-02-21 18:04:26 +02002534 r = create_vcpu_fd(vcpu);
Gleb Natapov73880c82009-06-09 15:56:28 +03002535 if (r < 0) {
2536 kvm_put_kvm(kvm);
Jan Kiszkad7805922011-05-23 10:33:05 +02002537 goto unlock_vcpu_destroy;
Gleb Natapov73880c82009-06-09 15:56:28 +03002538 }
2539
2540 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
Paolo Bonzinidd489242015-07-29 11:32:20 +02002541
2542 /*
2543 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2544 * before kvm->online_vcpu's incremented value.
2545 */
Gleb Natapov73880c82009-06-09 15:56:28 +03002546 smp_wmb();
2547 atomic_inc(&kvm->online_vcpus);
2548
Gleb Natapov73880c82009-06-09 15:56:28 +03002549 mutex_unlock(&kvm->lock);
Marcelo Tosatti42897d82012-11-27 23:29:02 -02002550 kvm_arch_vcpu_postcreate(vcpu);
Avi Kivitybccf2152007-02-21 18:04:26 +02002551 return r;
Avi Kivityc5ea7662007-02-20 18:41:05 +02002552
Jan Kiszkad7805922011-05-23 10:33:05 +02002553unlock_vcpu_destroy:
Glauber Costa7d8fece2008-09-17 23:16:59 -03002554 mutex_unlock(&kvm->lock);
Luiz Capitulino45b59392016-09-16 10:27:35 -04002555 debugfs_remove_recursive(vcpu->debugfs_dentry);
Jan Kiszkad7805922011-05-23 10:33:05 +02002556vcpu_destroy:
Hollis Blanchardd40ccc62007-11-19 14:04:43 -06002557 kvm_arch_vcpu_destroy(vcpu);
Paolo Bonzini6c7caeb2016-06-13 14:48:25 +02002558vcpu_decrement:
2559 mutex_lock(&kvm->lock);
2560 kvm->created_vcpus--;
2561 mutex_unlock(&kvm->lock);
Avi Kivityc5ea7662007-02-20 18:41:05 +02002562 return r;
2563}
2564
Avi Kivity1961d272007-03-05 19:46:05 +02002565static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2566{
2567 if (sigset) {
2568 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2569 vcpu->sigset_active = 1;
2570 vcpu->sigset = *sigset;
2571 } else
2572 vcpu->sigset_active = 0;
2573 return 0;
2574}
2575
Avi Kivitybccf2152007-02-21 18:04:26 +02002576static long kvm_vcpu_ioctl(struct file *filp,
2577 unsigned int ioctl, unsigned long arg)
Avi Kivity6aa8b732006-12-10 02:21:36 -08002578{
Avi Kivitybccf2152007-02-21 18:04:26 +02002579 struct kvm_vcpu *vcpu = filp->private_data;
Al Viro2f3669872007-02-09 16:38:35 +00002580 void __user *argp = (void __user *)arg;
Carsten Otte313a3dc2007-10-11 19:16:52 +02002581 int r;
Dave Hansenfa3795a2008-08-11 10:01:46 -07002582 struct kvm_fpu *fpu = NULL;
2583 struct kvm_sregs *kvm_sregs = NULL;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002584
Avi Kivity6d4e4c42007-11-21 16:41:05 +02002585 if (vcpu->kvm->mm != current->mm)
2586 return -EIO;
Avi Kivity2122ff52010-05-13 11:25:04 +03002587
David Matlack2ea75be2014-09-19 16:03:25 -07002588 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2589 return -EINVAL;
2590
Sanjay Lal2f4d9b52012-11-21 18:34:15 -08002591#if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
Avi Kivity2122ff52010-05-13 11:25:04 +03002592 /*
2593 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2594 * so vcpu_load() would break it.
2595 */
Jens Freimann47b43c52014-11-11 20:57:06 +01002596 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_S390_IRQ || ioctl == KVM_INTERRUPT)
Avi Kivity2122ff52010-05-13 11:25:04 +03002597 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2598#endif
2599
2600
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +03002601 r = vcpu_load(vcpu);
2602 if (r)
2603 return r;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002604 switch (ioctl) {
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02002605 case KVM_RUN:
Avi Kivityf0fe5102007-03-07 13:11:17 +02002606 r = -EINVAL;
2607 if (arg)
2608 goto out;
Christian Borntraeger7a72f7a2014-08-05 16:44:14 +02002609 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
2610 /* The thread running this VCPU changed. */
2611 struct pid *oldpid = vcpu->pid;
2612 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
Xiubo Lif95ef0c2015-02-26 14:58:23 +08002613
Christian Borntraeger7a72f7a2014-08-05 16:44:14 +02002614 rcu_assign_pointer(vcpu->pid, newpid);
2615 if (oldpid)
2616 synchronize_rcu();
2617 put_pid(oldpid);
2618 }
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05002619 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
Gleb Natapov64be5002010-10-24 16:49:08 +02002620 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002621 break;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002622 case KVM_GET_REGS: {
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002623 struct kvm_regs *kvm_regs;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002624
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002625 r = -ENOMEM;
2626 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2627 if (!kvm_regs)
2628 goto out;
2629 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002630 if (r)
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002631 goto out_free1;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002632 r = -EFAULT;
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002633 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2634 goto out_free1;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002635 r = 0;
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002636out_free1:
2637 kfree(kvm_regs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002638 break;
2639 }
2640 case KVM_SET_REGS: {
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002641 struct kvm_regs *kvm_regs;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002642
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002643 r = -ENOMEM;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002644 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2645 if (IS_ERR(kvm_regs)) {
2646 r = PTR_ERR(kvm_regs);
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002647 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002648 }
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002649 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002650 kfree(kvm_regs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002651 break;
2652 }
2653 case KVM_GET_SREGS: {
Dave Hansenfa3795a2008-08-11 10:01:46 -07002654 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2655 r = -ENOMEM;
2656 if (!kvm_sregs)
2657 goto out;
2658 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002659 if (r)
2660 goto out;
2661 r = -EFAULT;
Dave Hansenfa3795a2008-08-11 10:01:46 -07002662 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002663 goto out;
2664 r = 0;
2665 break;
2666 }
2667 case KVM_SET_SREGS: {
Sasha Levinff5c2c02011-12-04 19:36:29 +02002668 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2669 if (IS_ERR(kvm_sregs)) {
2670 r = PTR_ERR(kvm_sregs);
Guo Chao18595412012-11-02 18:33:21 +08002671 kvm_sregs = NULL;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002672 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002673 }
Dave Hansenfa3795a2008-08-11 10:01:46 -07002674 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002675 break;
2676 }
Marcelo Tosatti62d9f0d2008-04-11 13:24:45 -03002677 case KVM_GET_MP_STATE: {
2678 struct kvm_mp_state mp_state;
2679
2680 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2681 if (r)
2682 goto out;
2683 r = -EFAULT;
Xiubo Li893bdbf2015-02-26 14:58:19 +08002684 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
Marcelo Tosatti62d9f0d2008-04-11 13:24:45 -03002685 goto out;
2686 r = 0;
2687 break;
2688 }
2689 case KVM_SET_MP_STATE: {
2690 struct kvm_mp_state mp_state;
2691
2692 r = -EFAULT;
Xiubo Li893bdbf2015-02-26 14:58:19 +08002693 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
Marcelo Tosatti62d9f0d2008-04-11 13:24:45 -03002694 goto out;
2695 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
Marcelo Tosatti62d9f0d2008-04-11 13:24:45 -03002696 break;
2697 }
Avi Kivity6aa8b732006-12-10 02:21:36 -08002698 case KVM_TRANSLATE: {
2699 struct kvm_translation tr;
2700
2701 r = -EFAULT;
Xiubo Li893bdbf2015-02-26 14:58:19 +08002702 if (copy_from_user(&tr, argp, sizeof(tr)))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002703 goto out;
Zhang Xiantao8b006792007-11-16 13:05:55 +08002704 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002705 if (r)
2706 goto out;
2707 r = -EFAULT;
Xiubo Li893bdbf2015-02-26 14:58:19 +08002708 if (copy_to_user(argp, &tr, sizeof(tr)))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002709 goto out;
2710 r = 0;
2711 break;
2712 }
Jan Kiszkad0bfb942008-12-15 13:52:10 +01002713 case KVM_SET_GUEST_DEBUG: {
2714 struct kvm_guest_debug dbg;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002715
2716 r = -EFAULT;
Xiubo Li893bdbf2015-02-26 14:58:19 +08002717 if (copy_from_user(&dbg, argp, sizeof(dbg)))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002718 goto out;
Jan Kiszkad0bfb942008-12-15 13:52:10 +01002719 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002720 break;
2721 }
Avi Kivity1961d272007-03-05 19:46:05 +02002722 case KVM_SET_SIGNAL_MASK: {
2723 struct kvm_signal_mask __user *sigmask_arg = argp;
2724 struct kvm_signal_mask kvm_sigmask;
2725 sigset_t sigset, *p;
2726
2727 p = NULL;
2728 if (argp) {
2729 r = -EFAULT;
2730 if (copy_from_user(&kvm_sigmask, argp,
Xiubo Li893bdbf2015-02-26 14:58:19 +08002731 sizeof(kvm_sigmask)))
Avi Kivity1961d272007-03-05 19:46:05 +02002732 goto out;
2733 r = -EINVAL;
Xiubo Li893bdbf2015-02-26 14:58:19 +08002734 if (kvm_sigmask.len != sizeof(sigset))
Avi Kivity1961d272007-03-05 19:46:05 +02002735 goto out;
2736 r = -EFAULT;
2737 if (copy_from_user(&sigset, sigmask_arg->sigset,
Xiubo Li893bdbf2015-02-26 14:58:19 +08002738 sizeof(sigset)))
Avi Kivity1961d272007-03-05 19:46:05 +02002739 goto out;
2740 p = &sigset;
2741 }
Andi Kleen376d41f2010-06-10 13:10:47 +02002742 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
Avi Kivity1961d272007-03-05 19:46:05 +02002743 break;
2744 }
Avi Kivityb8836732007-04-01 16:34:31 +03002745 case KVM_GET_FPU: {
Dave Hansenfa3795a2008-08-11 10:01:46 -07002746 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2747 r = -ENOMEM;
2748 if (!fpu)
2749 goto out;
2750 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
Avi Kivityb8836732007-04-01 16:34:31 +03002751 if (r)
2752 goto out;
2753 r = -EFAULT;
Dave Hansenfa3795a2008-08-11 10:01:46 -07002754 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
Avi Kivityb8836732007-04-01 16:34:31 +03002755 goto out;
2756 r = 0;
2757 break;
2758 }
2759 case KVM_SET_FPU: {
Sasha Levinff5c2c02011-12-04 19:36:29 +02002760 fpu = memdup_user(argp, sizeof(*fpu));
2761 if (IS_ERR(fpu)) {
2762 r = PTR_ERR(fpu);
Guo Chao18595412012-11-02 18:33:21 +08002763 fpu = NULL;
Avi Kivityb8836732007-04-01 16:34:31 +03002764 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002765 }
Dave Hansenfa3795a2008-08-11 10:01:46 -07002766 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
Avi Kivityb8836732007-04-01 16:34:31 +03002767 break;
2768 }
Avi Kivitybccf2152007-02-21 18:04:26 +02002769 default:
Carsten Otte313a3dc2007-10-11 19:16:52 +02002770 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
Avi Kivitybccf2152007-02-21 18:04:26 +02002771 }
2772out:
Avi Kivity2122ff52010-05-13 11:25:04 +03002773 vcpu_put(vcpu);
Dave Hansenfa3795a2008-08-11 10:01:46 -07002774 kfree(fpu);
2775 kfree(kvm_sregs);
Avi Kivitybccf2152007-02-21 18:04:26 +02002776 return r;
2777}
2778
Christian Borntraegerde8e5d72015-02-03 09:35:15 +01002779#ifdef CONFIG_KVM_COMPAT
Alexander Graf1dda6062011-06-08 02:45:37 +02002780static long kvm_vcpu_compat_ioctl(struct file *filp,
2781 unsigned int ioctl, unsigned long arg)
2782{
2783 struct kvm_vcpu *vcpu = filp->private_data;
2784 void __user *argp = compat_ptr(arg);
2785 int r;
2786
2787 if (vcpu->kvm->mm != current->mm)
2788 return -EIO;
2789
2790 switch (ioctl) {
2791 case KVM_SET_SIGNAL_MASK: {
2792 struct kvm_signal_mask __user *sigmask_arg = argp;
2793 struct kvm_signal_mask kvm_sigmask;
2794 compat_sigset_t csigset;
2795 sigset_t sigset;
2796
2797 if (argp) {
2798 r = -EFAULT;
2799 if (copy_from_user(&kvm_sigmask, argp,
Xiubo Li893bdbf2015-02-26 14:58:19 +08002800 sizeof(kvm_sigmask)))
Alexander Graf1dda6062011-06-08 02:45:37 +02002801 goto out;
2802 r = -EINVAL;
Xiubo Li893bdbf2015-02-26 14:58:19 +08002803 if (kvm_sigmask.len != sizeof(csigset))
Alexander Graf1dda6062011-06-08 02:45:37 +02002804 goto out;
2805 r = -EFAULT;
2806 if (copy_from_user(&csigset, sigmask_arg->sigset,
Xiubo Li893bdbf2015-02-26 14:58:19 +08002807 sizeof(csigset)))
Alexander Graf1dda6062011-06-08 02:45:37 +02002808 goto out;
Alan Cox760a9a32012-08-22 14:34:11 +01002809 sigset_from_compat(&sigset, &csigset);
2810 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2811 } else
2812 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
Alexander Graf1dda6062011-06-08 02:45:37 +02002813 break;
2814 }
2815 default:
2816 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2817 }
2818
2819out:
2820 return r;
2821}
2822#endif
2823
Scott Wood852b6d52013-04-12 14:08:42 +00002824static int kvm_device_ioctl_attr(struct kvm_device *dev,
2825 int (*accessor)(struct kvm_device *dev,
2826 struct kvm_device_attr *attr),
2827 unsigned long arg)
2828{
2829 struct kvm_device_attr attr;
2830
2831 if (!accessor)
2832 return -EPERM;
2833
2834 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2835 return -EFAULT;
2836
2837 return accessor(dev, &attr);
2838}
2839
2840static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2841 unsigned long arg)
2842{
2843 struct kvm_device *dev = filp->private_data;
2844
Sean Christophersonc217b3c2019-02-15 12:48:39 -08002845 if (dev->kvm->mm != current->mm)
2846 return -EIO;
2847
Scott Wood852b6d52013-04-12 14:08:42 +00002848 switch (ioctl) {
2849 case KVM_SET_DEVICE_ATTR:
2850 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2851 case KVM_GET_DEVICE_ATTR:
2852 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2853 case KVM_HAS_DEVICE_ATTR:
2854 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2855 default:
2856 if (dev->ops->ioctl)
2857 return dev->ops->ioctl(dev, ioctl, arg);
2858
2859 return -ENOTTY;
2860 }
2861}
2862
Scott Wood852b6d52013-04-12 14:08:42 +00002863static int kvm_device_release(struct inode *inode, struct file *filp)
2864{
2865 struct kvm_device *dev = filp->private_data;
2866 struct kvm *kvm = dev->kvm;
2867
Scott Wood852b6d52013-04-12 14:08:42 +00002868 kvm_put_kvm(kvm);
2869 return 0;
2870}
2871
2872static const struct file_operations kvm_device_fops = {
2873 .unlocked_ioctl = kvm_device_ioctl,
Christian Borntraegerde8e5d72015-02-03 09:35:15 +01002874#ifdef CONFIG_KVM_COMPAT
Scott Wooddb6ae612013-04-30 20:00:45 -05002875 .compat_ioctl = kvm_device_ioctl,
2876#endif
Scott Wood852b6d52013-04-12 14:08:42 +00002877 .release = kvm_device_release,
2878};
2879
2880struct kvm_device *kvm_device_from_filp(struct file *filp)
2881{
2882 if (filp->f_op != &kvm_device_fops)
2883 return NULL;
2884
2885 return filp->private_data;
2886}
2887
Will Deacond60eacb2014-09-02 10:27:33 +01002888static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2889#ifdef CONFIG_KVM_MPIC
2890 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
2891 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
2892#endif
2893
2894#ifdef CONFIG_KVM_XICS
2895 [KVM_DEV_TYPE_XICS] = &kvm_xics_ops,
2896#endif
Will Deacond60eacb2014-09-02 10:27:33 +01002897};
2898
2899int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2900{
2901 if (type >= ARRAY_SIZE(kvm_device_ops_table))
2902 return -ENOSPC;
2903
2904 if (kvm_device_ops_table[type] != NULL)
2905 return -EEXIST;
2906
2907 kvm_device_ops_table[type] = ops;
2908 return 0;
2909}
2910
Wanpeng Li571ee1b2014-10-09 18:30:08 +08002911void kvm_unregister_device_ops(u32 type)
2912{
2913 if (kvm_device_ops_table[type] != NULL)
2914 kvm_device_ops_table[type] = NULL;
2915}
2916
Scott Wood852b6d52013-04-12 14:08:42 +00002917static int kvm_ioctl_create_device(struct kvm *kvm,
2918 struct kvm_create_device *cd)
2919{
2920 struct kvm_device_ops *ops = NULL;
2921 struct kvm_device *dev;
2922 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2923 int ret;
2924
Will Deacond60eacb2014-09-02 10:27:33 +01002925 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
Scott Wood852b6d52013-04-12 14:08:42 +00002926 return -ENODEV;
Will Deacond60eacb2014-09-02 10:27:33 +01002927
2928 ops = kvm_device_ops_table[cd->type];
2929 if (ops == NULL)
2930 return -ENODEV;
Scott Wood852b6d52013-04-12 14:08:42 +00002931
2932 if (test)
2933 return 0;
2934
2935 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2936 if (!dev)
2937 return -ENOMEM;
2938
2939 dev->ops = ops;
2940 dev->kvm = kvm;
Scott Wood852b6d52013-04-12 14:08:42 +00002941
Christoffer Dalla28ebea2016-08-09 19:13:01 +02002942 mutex_lock(&kvm->lock);
Scott Wood852b6d52013-04-12 14:08:42 +00002943 ret = ops->create(dev, cd->type);
2944 if (ret < 0) {
Christoffer Dalla28ebea2016-08-09 19:13:01 +02002945 mutex_unlock(&kvm->lock);
Scott Wood852b6d52013-04-12 14:08:42 +00002946 kfree(dev);
2947 return ret;
2948 }
Christoffer Dalla28ebea2016-08-09 19:13:01 +02002949 list_add(&dev->vm_node, &kvm->devices);
2950 mutex_unlock(&kvm->lock);
Scott Wood852b6d52013-04-12 14:08:42 +00002951
Christoffer Dall023e9fd2016-08-09 19:13:00 +02002952 if (ops->init)
2953 ops->init(dev);
2954
Jann Horn0c42df12019-01-26 01:54:33 +01002955 kvm_get_kvm(kvm);
Yann Droneaud24009b02013-08-24 22:14:07 +02002956 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
Scott Wood852b6d52013-04-12 14:08:42 +00002957 if (ret < 0) {
Jann Horn0c42df12019-01-26 01:54:33 +01002958 kvm_put_kvm(kvm);
Christoffer Dalla28ebea2016-08-09 19:13:01 +02002959 mutex_lock(&kvm->lock);
2960 list_del(&dev->vm_node);
2961 mutex_unlock(&kvm->lock);
Dan Carpentera0f1d212016-11-30 22:21:05 +03002962 ops->destroy(dev);
Scott Wood852b6d52013-04-12 14:08:42 +00002963 return ret;
2964 }
2965
Scott Wood852b6d52013-04-12 14:08:42 +00002966 cd->fd = ret;
2967 return 0;
2968}
2969
Alexander Graf92b591a2014-07-14 18:33:08 +02002970static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2971{
2972 switch (arg) {
2973 case KVM_CAP_USER_MEMORY:
2974 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2975 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
Alexander Graf92b591a2014-07-14 18:33:08 +02002976 case KVM_CAP_INTERNAL_ERROR_DATA:
2977#ifdef CONFIG_HAVE_KVM_MSI
2978 case KVM_CAP_SIGNAL_MSI:
2979#endif
Paul Mackerras297e2102014-06-30 20:51:13 +10002980#ifdef CONFIG_HAVE_KVM_IRQFD
Paolo Bonzinidc9be0f2015-03-05 11:54:46 +01002981 case KVM_CAP_IRQFD:
Alexander Graf92b591a2014-07-14 18:33:08 +02002982 case KVM_CAP_IRQFD_RESAMPLE:
2983#endif
Jason Wange9ea5062015-09-15 14:41:59 +08002984 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
Alexander Graf92b591a2014-07-14 18:33:08 +02002985 case KVM_CAP_CHECK_EXTENSION_VM:
2986 return 1;
2987#ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2988 case KVM_CAP_IRQ_ROUTING:
2989 return KVM_MAX_IRQ_ROUTES;
2990#endif
Paolo Bonzinif481b062015-05-17 17:30:37 +02002991#if KVM_ADDRESS_SPACE_NUM > 1
2992 case KVM_CAP_MULTI_ADDRESS_SPACE:
2993 return KVM_ADDRESS_SPACE_NUM;
2994#endif
Greg Kurz0b1b1df2016-05-09 18:13:37 +02002995 case KVM_CAP_MAX_VCPU_ID:
2996 return KVM_MAX_VCPU_ID;
Alexander Graf92b591a2014-07-14 18:33:08 +02002997 default:
2998 break;
2999 }
3000 return kvm_vm_ioctl_check_extension(kvm, arg);
3001}
3002
Avi Kivitybccf2152007-02-21 18:04:26 +02003003static long kvm_vm_ioctl(struct file *filp,
3004 unsigned int ioctl, unsigned long arg)
3005{
3006 struct kvm *kvm = filp->private_data;
3007 void __user *argp = (void __user *)arg;
Carsten Otte1fe779f2007-10-29 16:08:35 +01003008 int r;
Avi Kivitybccf2152007-02-21 18:04:26 +02003009
Avi Kivity6d4e4c42007-11-21 16:41:05 +02003010 if (kvm->mm != current->mm)
3011 return -EIO;
Avi Kivitybccf2152007-02-21 18:04:26 +02003012 switch (ioctl) {
3013 case KVM_CREATE_VCPU:
3014 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
Avi Kivitybccf2152007-02-21 18:04:26 +02003015 break;
Izik Eidus6fc138d2007-10-09 19:20:39 +02003016 case KVM_SET_USER_MEMORY_REGION: {
3017 struct kvm_userspace_memory_region kvm_userspace_mem;
3018
3019 r = -EFAULT;
3020 if (copy_from_user(&kvm_userspace_mem, argp,
Xiubo Li893bdbf2015-02-26 14:58:19 +08003021 sizeof(kvm_userspace_mem)))
Izik Eidus6fc138d2007-10-09 19:20:39 +02003022 goto out;
3023
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +09003024 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003025 break;
3026 }
3027 case KVM_GET_DIRTY_LOG: {
3028 struct kvm_dirty_log log;
3029
3030 r = -EFAULT;
Xiubo Li893bdbf2015-02-26 14:58:19 +08003031 if (copy_from_user(&log, argp, sizeof(log)))
Avi Kivity6aa8b732006-12-10 02:21:36 -08003032 goto out;
Avi Kivity2c6f5df2007-02-20 18:27:58 +02003033 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003034 break;
3035 }
Laurent Vivier5f94c172008-05-30 16:05:54 +02003036#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
3037 case KVM_REGISTER_COALESCED_MMIO: {
3038 struct kvm_coalesced_mmio_zone zone;
Xiubo Lif95ef0c2015-02-26 14:58:23 +08003039
Laurent Vivier5f94c172008-05-30 16:05:54 +02003040 r = -EFAULT;
Xiubo Li893bdbf2015-02-26 14:58:19 +08003041 if (copy_from_user(&zone, argp, sizeof(zone)))
Laurent Vivier5f94c172008-05-30 16:05:54 +02003042 goto out;
Laurent Vivier5f94c172008-05-30 16:05:54 +02003043 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
Laurent Vivier5f94c172008-05-30 16:05:54 +02003044 break;
3045 }
3046 case KVM_UNREGISTER_COALESCED_MMIO: {
3047 struct kvm_coalesced_mmio_zone zone;
Xiubo Lif95ef0c2015-02-26 14:58:23 +08003048
Laurent Vivier5f94c172008-05-30 16:05:54 +02003049 r = -EFAULT;
Xiubo Li893bdbf2015-02-26 14:58:19 +08003050 if (copy_from_user(&zone, argp, sizeof(zone)))
Laurent Vivier5f94c172008-05-30 16:05:54 +02003051 goto out;
Laurent Vivier5f94c172008-05-30 16:05:54 +02003052 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
Laurent Vivier5f94c172008-05-30 16:05:54 +02003053 break;
3054 }
3055#endif
Gregory Haskins721eecb2009-05-20 10:30:49 -04003056 case KVM_IRQFD: {
3057 struct kvm_irqfd data;
3058
3059 r = -EFAULT;
Xiubo Li893bdbf2015-02-26 14:58:19 +08003060 if (copy_from_user(&data, argp, sizeof(data)))
Gregory Haskins721eecb2009-05-20 10:30:49 -04003061 goto out;
Alex Williamsond4db2932012-06-29 09:56:08 -06003062 r = kvm_irqfd(kvm, &data);
Gregory Haskins721eecb2009-05-20 10:30:49 -04003063 break;
3064 }
Gregory Haskinsd34e6b12009-07-07 17:08:49 -04003065 case KVM_IOEVENTFD: {
3066 struct kvm_ioeventfd data;
3067
3068 r = -EFAULT;
Xiubo Li893bdbf2015-02-26 14:58:19 +08003069 if (copy_from_user(&data, argp, sizeof(data)))
Gregory Haskinsd34e6b12009-07-07 17:08:49 -04003070 goto out;
3071 r = kvm_ioeventfd(kvm, &data);
3072 break;
3073 }
Jan Kiszka07975ad2012-03-29 21:14:12 +02003074#ifdef CONFIG_HAVE_KVM_MSI
3075 case KVM_SIGNAL_MSI: {
3076 struct kvm_msi msi;
3077
3078 r = -EFAULT;
Xiubo Li893bdbf2015-02-26 14:58:19 +08003079 if (copy_from_user(&msi, argp, sizeof(msi)))
Jan Kiszka07975ad2012-03-29 21:14:12 +02003080 goto out;
3081 r = kvm_send_userspace_msi(kvm, &msi);
3082 break;
3083 }
3084#endif
Christoffer Dall23d43cf2012-07-24 08:51:20 -04003085#ifdef __KVM_HAVE_IRQ_LINE
3086 case KVM_IRQ_LINE_STATUS:
3087 case KVM_IRQ_LINE: {
3088 struct kvm_irq_level irq_event;
3089
3090 r = -EFAULT;
Xiubo Li893bdbf2015-02-26 14:58:19 +08003091 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
Christoffer Dall23d43cf2012-07-24 08:51:20 -04003092 goto out;
3093
Yang Zhangaa2fbe62013-04-11 19:21:40 +08003094 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
3095 ioctl == KVM_IRQ_LINE_STATUS);
Christoffer Dall23d43cf2012-07-24 08:51:20 -04003096 if (r)
3097 goto out;
3098
3099 r = -EFAULT;
3100 if (ioctl == KVM_IRQ_LINE_STATUS) {
Xiubo Li893bdbf2015-02-26 14:58:19 +08003101 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
Christoffer Dall23d43cf2012-07-24 08:51:20 -04003102 goto out;
3103 }
3104
3105 r = 0;
3106 break;
3107 }
3108#endif
Alexander Grafaa8d5942013-04-15 21:12:53 +02003109#ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3110 case KVM_SET_GSI_ROUTING: {
3111 struct kvm_irq_routing routing;
3112 struct kvm_irq_routing __user *urouting;
Paolo Bonzinif8c1b852016-06-01 14:09:22 +02003113 struct kvm_irq_routing_entry *entries = NULL;
Alexander Grafaa8d5942013-04-15 21:12:53 +02003114
3115 r = -EFAULT;
3116 if (copy_from_user(&routing, argp, sizeof(routing)))
3117 goto out;
3118 r = -EINVAL;
Xiubo Licaf1ff22016-06-15 18:00:33 +08003119 if (routing.nr > KVM_MAX_IRQ_ROUTES)
Alexander Grafaa8d5942013-04-15 21:12:53 +02003120 goto out;
3121 if (routing.flags)
3122 goto out;
Paolo Bonzinif8c1b852016-06-01 14:09:22 +02003123 if (routing.nr) {
3124 r = -ENOMEM;
3125 entries = vmalloc(routing.nr * sizeof(*entries));
3126 if (!entries)
3127 goto out;
3128 r = -EFAULT;
3129 urouting = argp;
3130 if (copy_from_user(entries, urouting->entries,
3131 routing.nr * sizeof(*entries)))
3132 goto out_free_irq_routing;
3133 }
Alexander Grafaa8d5942013-04-15 21:12:53 +02003134 r = kvm_set_irq_routing(kvm, entries, routing.nr,
3135 routing.flags);
Xiubo Lia642a172015-02-26 14:58:20 +08003136out_free_irq_routing:
Alexander Grafaa8d5942013-04-15 21:12:53 +02003137 vfree(entries);
3138 break;
3139 }
3140#endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
Scott Wood852b6d52013-04-12 14:08:42 +00003141 case KVM_CREATE_DEVICE: {
3142 struct kvm_create_device cd;
3143
3144 r = -EFAULT;
3145 if (copy_from_user(&cd, argp, sizeof(cd)))
3146 goto out;
3147
3148 r = kvm_ioctl_create_device(kvm, &cd);
3149 if (r)
3150 goto out;
3151
3152 r = -EFAULT;
3153 if (copy_to_user(argp, &cd, sizeof(cd)))
3154 goto out;
3155
3156 r = 0;
3157 break;
3158 }
Alexander Graf92b591a2014-07-14 18:33:08 +02003159 case KVM_CHECK_EXTENSION:
3160 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
3161 break;
Avi Kivityf17abe92007-02-21 19:28:04 +02003162 default:
Carsten Otte1fe779f2007-10-29 16:08:35 +01003163 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
Avi Kivityf17abe92007-02-21 19:28:04 +02003164 }
3165out:
3166 return r;
3167}
3168
Christian Borntraegerde8e5d72015-02-03 09:35:15 +01003169#ifdef CONFIG_KVM_COMPAT
Arnd Bergmann6ff58942009-10-22 14:19:27 +02003170struct compat_kvm_dirty_log {
3171 __u32 slot;
3172 __u32 padding1;
3173 union {
3174 compat_uptr_t dirty_bitmap; /* one bit per page */
3175 __u64 padding2;
3176 };
3177};
3178
3179static long kvm_vm_compat_ioctl(struct file *filp,
3180 unsigned int ioctl, unsigned long arg)
3181{
3182 struct kvm *kvm = filp->private_data;
3183 int r;
3184
3185 if (kvm->mm != current->mm)
3186 return -EIO;
3187 switch (ioctl) {
3188 case KVM_GET_DIRTY_LOG: {
3189 struct compat_kvm_dirty_log compat_log;
3190 struct kvm_dirty_log log;
3191
3192 r = -EFAULT;
3193 if (copy_from_user(&compat_log, (void __user *)arg,
3194 sizeof(compat_log)))
3195 goto out;
3196 log.slot = compat_log.slot;
3197 log.padding1 = compat_log.padding1;
3198 log.padding2 = compat_log.padding2;
3199 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
3200
3201 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
Arnd Bergmann6ff58942009-10-22 14:19:27 +02003202 break;
3203 }
3204 default:
3205 r = kvm_vm_ioctl(filp, ioctl, arg);
3206 }
3207
3208out:
3209 return r;
3210}
3211#endif
3212
Christian Borntraeger3d3aab12008-12-02 11:17:32 +01003213static struct file_operations kvm_vm_fops = {
Avi Kivityf17abe92007-02-21 19:28:04 +02003214 .release = kvm_vm_release,
3215 .unlocked_ioctl = kvm_vm_ioctl,
Christian Borntraegerde8e5d72015-02-03 09:35:15 +01003216#ifdef CONFIG_KVM_COMPAT
Arnd Bergmann6ff58942009-10-22 14:19:27 +02003217 .compat_ioctl = kvm_vm_compat_ioctl,
3218#endif
Arnd Bergmann6038f372010-08-15 18:52:59 +02003219 .llseek = noop_llseek,
Avi Kivityf17abe92007-02-21 19:28:04 +02003220};
3221
Carsten Ottee08b9632012-01-04 10:25:20 +01003222static int kvm_dev_ioctl_create_vm(unsigned long type)
Avi Kivityf17abe92007-02-21 19:28:04 +02003223{
Heiko Carstensaac87632010-10-27 17:22:10 +02003224 int r;
Avi Kivityf17abe92007-02-21 19:28:04 +02003225 struct kvm *kvm;
Al Viro506cfba2016-07-14 18:54:17 +02003226 struct file *file;
Avi Kivityf17abe92007-02-21 19:28:04 +02003227
Carsten Ottee08b9632012-01-04 10:25:20 +01003228 kvm = kvm_create_vm(type);
Avi Kivityd6d28162007-06-28 08:38:16 -04003229 if (IS_ERR(kvm))
3230 return PTR_ERR(kvm);
Takuya Yoshikawa6ce5a092010-03-15 22:13:30 +09003231#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
3232 r = kvm_coalesced_mmio_init(kvm);
3233 if (r < 0) {
3234 kvm_put_kvm(kvm);
3235 return r;
3236 }
3237#endif
Al Viro506cfba2016-07-14 18:54:17 +02003238 r = get_unused_fd_flags(O_CLOEXEC);
Janosch Frank536a6f82016-05-18 13:26:23 +02003239 if (r < 0) {
Al Viro66c0b392008-04-19 20:33:56 +01003240 kvm_put_kvm(kvm);
Janosch Frank536a6f82016-05-18 13:26:23 +02003241 return r;
3242 }
Al Viro506cfba2016-07-14 18:54:17 +02003243 file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
3244 if (IS_ERR(file)) {
3245 put_unused_fd(r);
3246 kvm_put_kvm(kvm);
3247 return PTR_ERR(file);
3248 }
Janosch Frank536a6f82016-05-18 13:26:23 +02003249
3250 if (kvm_create_vm_debugfs(kvm, r) < 0) {
Al Viro506cfba2016-07-14 18:54:17 +02003251 put_unused_fd(r);
3252 fput(file);
Janosch Frank536a6f82016-05-18 13:26:23 +02003253 return -ENOMEM;
3254 }
Avi Kivityf17abe92007-02-21 19:28:04 +02003255
Al Viro506cfba2016-07-14 18:54:17 +02003256 fd_install(r, file);
Heiko Carstensaac87632010-10-27 17:22:10 +02003257 return r;
Avi Kivityf17abe92007-02-21 19:28:04 +02003258}
3259
3260static long kvm_dev_ioctl(struct file *filp,
3261 unsigned int ioctl, unsigned long arg)
3262{
Avi Kivity07c45a32007-03-07 13:05:38 +02003263 long r = -EINVAL;
Avi Kivityf17abe92007-02-21 19:28:04 +02003264
3265 switch (ioctl) {
3266 case KVM_GET_API_VERSION:
Avi Kivityf0fe5102007-03-07 13:11:17 +02003267 if (arg)
3268 goto out;
Avi Kivityf17abe92007-02-21 19:28:04 +02003269 r = KVM_API_VERSION;
3270 break;
3271 case KVM_CREATE_VM:
Carsten Ottee08b9632012-01-04 10:25:20 +01003272 r = kvm_dev_ioctl_create_vm(arg);
Avi Kivityf17abe92007-02-21 19:28:04 +02003273 break;
Zhang Xiantao018d00d2007-11-15 23:07:47 +08003274 case KVM_CHECK_EXTENSION:
Alexander Graf784aa3d2014-07-14 18:27:35 +02003275 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
Avi Kivity5d308f42007-03-01 17:56:20 +02003276 break;
Avi Kivity07c45a32007-03-07 13:05:38 +02003277 case KVM_GET_VCPU_MMAP_SIZE:
Avi Kivity07c45a32007-03-07 13:05:38 +02003278 if (arg)
3279 goto out;
Avi Kivityadb1ff42008-01-24 15:13:08 +02003280 r = PAGE_SIZE; /* struct kvm_run */
3281#ifdef CONFIG_X86
3282 r += PAGE_SIZE; /* pio data page */
3283#endif
Laurent Vivier5f94c172008-05-30 16:05:54 +02003284#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
3285 r += PAGE_SIZE; /* coalesced mmio ring page */
3286#endif
Avi Kivity07c45a32007-03-07 13:05:38 +02003287 break;
Feng(Eric) Liud4c9ff22008-04-10 08:47:53 -04003288 case KVM_TRACE_ENABLE:
3289 case KVM_TRACE_PAUSE:
3290 case KVM_TRACE_DISABLE:
Marcelo Tosatti2023a292009-06-18 11:47:28 -03003291 r = -EOPNOTSUPP;
Feng(Eric) Liud4c9ff22008-04-10 08:47:53 -04003292 break;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003293 default:
Carsten Otte043405e2007-10-10 17:16:19 +02003294 return kvm_arch_dev_ioctl(filp, ioctl, arg);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003295 }
3296out:
3297 return r;
3298}
3299
Avi Kivity6aa8b732006-12-10 02:21:36 -08003300static struct file_operations kvm_chardev_ops = {
Avi Kivity6aa8b732006-12-10 02:21:36 -08003301 .unlocked_ioctl = kvm_dev_ioctl,
3302 .compat_ioctl = kvm_dev_ioctl,
Arnd Bergmann6038f372010-08-15 18:52:59 +02003303 .llseek = noop_llseek,
Avi Kivity6aa8b732006-12-10 02:21:36 -08003304};
3305
3306static struct miscdevice kvm_dev = {
Avi Kivitybbe44322007-03-04 13:27:36 +02003307 KVM_MINOR,
Avi Kivity6aa8b732006-12-10 02:21:36 -08003308 "kvm",
3309 &kvm_chardev_ops,
3310};
3311
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09003312static void hardware_enable_nolock(void *junk)
Avi Kivity1b6c0162007-05-24 13:03:52 +03003313{
3314 int cpu = raw_smp_processor_id();
Alexander Graf10474ae2009-09-15 11:37:46 +02003315 int r;
Avi Kivity1b6c0162007-05-24 13:03:52 +03003316
Rusty Russell7f59f492008-12-07 21:25:45 +10303317 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
Avi Kivity1b6c0162007-05-24 13:03:52 +03003318 return;
Alexander Graf10474ae2009-09-15 11:37:46 +02003319
Rusty Russell7f59f492008-12-07 21:25:45 +10303320 cpumask_set_cpu(cpu, cpus_hardware_enabled);
Alexander Graf10474ae2009-09-15 11:37:46 +02003321
Radim Krčmář13a34e02014-08-28 15:13:03 +02003322 r = kvm_arch_hardware_enable();
Alexander Graf10474ae2009-09-15 11:37:46 +02003323
3324 if (r) {
3325 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3326 atomic_inc(&hardware_enable_failed);
Xiubo Li1170adc2015-02-26 14:58:26 +08003327 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
Alexander Graf10474ae2009-09-15 11:37:46 +02003328 }
Avi Kivity1b6c0162007-05-24 13:03:52 +03003329}
3330
Thomas Gleixner8c18b2d2016-07-13 17:16:37 +00003331static int kvm_starting_cpu(unsigned int cpu)
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09003332{
Paolo Bonzini4a937f92013-09-10 12:58:35 +02003333 raw_spin_lock(&kvm_count_lock);
Paolo Bonzini4fa92fb2013-09-10 12:57:17 +02003334 if (kvm_usage_count)
3335 hardware_enable_nolock(NULL);
Paolo Bonzini4a937f92013-09-10 12:58:35 +02003336 raw_spin_unlock(&kvm_count_lock);
Thomas Gleixner8c18b2d2016-07-13 17:16:37 +00003337 return 0;
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09003338}
3339
3340static void hardware_disable_nolock(void *junk)
Avi Kivity1b6c0162007-05-24 13:03:52 +03003341{
3342 int cpu = raw_smp_processor_id();
3343
Rusty Russell7f59f492008-12-07 21:25:45 +10303344 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
Avi Kivity1b6c0162007-05-24 13:03:52 +03003345 return;
Rusty Russell7f59f492008-12-07 21:25:45 +10303346 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
Radim Krčmář13a34e02014-08-28 15:13:03 +02003347 kvm_arch_hardware_disable();
Avi Kivity1b6c0162007-05-24 13:03:52 +03003348}
3349
Thomas Gleixner8c18b2d2016-07-13 17:16:37 +00003350static int kvm_dying_cpu(unsigned int cpu)
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09003351{
Paolo Bonzini4a937f92013-09-10 12:58:35 +02003352 raw_spin_lock(&kvm_count_lock);
Paolo Bonzini4fa92fb2013-09-10 12:57:17 +02003353 if (kvm_usage_count)
3354 hardware_disable_nolock(NULL);
Paolo Bonzini4a937f92013-09-10 12:58:35 +02003355 raw_spin_unlock(&kvm_count_lock);
Thomas Gleixner8c18b2d2016-07-13 17:16:37 +00003356 return 0;
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09003357}
3358
Alexander Graf10474ae2009-09-15 11:37:46 +02003359static void hardware_disable_all_nolock(void)
3360{
3361 BUG_ON(!kvm_usage_count);
3362
3363 kvm_usage_count--;
3364 if (!kvm_usage_count)
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09003365 on_each_cpu(hardware_disable_nolock, NULL, 1);
Alexander Graf10474ae2009-09-15 11:37:46 +02003366}
3367
3368static void hardware_disable_all(void)
3369{
Paolo Bonzini4a937f92013-09-10 12:58:35 +02003370 raw_spin_lock(&kvm_count_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02003371 hardware_disable_all_nolock();
Paolo Bonzini4a937f92013-09-10 12:58:35 +02003372 raw_spin_unlock(&kvm_count_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02003373}
3374
3375static int hardware_enable_all(void)
3376{
3377 int r = 0;
3378
Paolo Bonzini4a937f92013-09-10 12:58:35 +02003379 raw_spin_lock(&kvm_count_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02003380
3381 kvm_usage_count++;
3382 if (kvm_usage_count == 1) {
3383 atomic_set(&hardware_enable_failed, 0);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09003384 on_each_cpu(hardware_enable_nolock, NULL, 1);
Alexander Graf10474ae2009-09-15 11:37:46 +02003385
3386 if (atomic_read(&hardware_enable_failed)) {
3387 hardware_disable_all_nolock();
3388 r = -EBUSY;
3389 }
3390 }
3391
Paolo Bonzini4a937f92013-09-10 12:58:35 +02003392 raw_spin_unlock(&kvm_count_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02003393
3394 return r;
3395}
3396
Rusty Russell9a2b85c2007-07-17 23:17:55 +10003397static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
Mike Dayd77c26f2007-10-08 09:02:08 -04003398 void *v)
Rusty Russell9a2b85c2007-07-17 23:17:55 +10003399{
Sheng Yang8e1c1812009-04-29 11:09:04 +08003400 /*
3401 * Some (well, at least mine) BIOSes hang on reboot if
3402 * in vmx root mode.
3403 *
3404 * And Intel TXT required VMX off for all cpu when system shutdown.
3405 */
Xiubo Li1170adc2015-02-26 14:58:26 +08003406 pr_info("kvm: exiting hardware virtualization\n");
Sheng Yang8e1c1812009-04-29 11:09:04 +08003407 kvm_rebooting = true;
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09003408 on_each_cpu(hardware_disable_nolock, NULL, 1);
Rusty Russell9a2b85c2007-07-17 23:17:55 +10003409 return NOTIFY_OK;
3410}
3411
3412static struct notifier_block kvm_reboot_notifier = {
3413 .notifier_call = kvm_reboot,
3414 .priority = 0,
3415};
3416
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003417static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04003418{
3419 int i;
3420
3421 for (i = 0; i < bus->dev_count; i++) {
Sasha Levin743eeb02011-07-27 16:00:48 +03003422 struct kvm_io_device *pos = bus->range[i].dev;
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04003423
3424 kvm_iodevice_destructor(pos);
3425 }
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003426 kfree(bus);
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04003427}
3428
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02003429static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
Xiubo Li20e87b72015-02-26 14:58:25 +08003430 const struct kvm_io_range *r2)
Sasha Levin743eeb02011-07-27 16:00:48 +03003431{
Jason Wang8f4216c72015-09-15 14:41:57 +08003432 gpa_t addr1 = r1->addr;
3433 gpa_t addr2 = r2->addr;
3434
3435 if (addr1 < addr2)
Sasha Levin743eeb02011-07-27 16:00:48 +03003436 return -1;
Jason Wang8f4216c72015-09-15 14:41:57 +08003437
3438 /* If r2->len == 0, match the exact address. If r2->len != 0,
3439 * accept any overlapping write. Any order is acceptable for
3440 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3441 * we process all of them.
3442 */
3443 if (r2->len) {
3444 addr1 += r1->len;
3445 addr2 += r2->len;
3446 }
3447
3448 if (addr1 > addr2)
Sasha Levin743eeb02011-07-27 16:00:48 +03003449 return 1;
Jason Wang8f4216c72015-09-15 14:41:57 +08003450
Sasha Levin743eeb02011-07-27 16:00:48 +03003451 return 0;
3452}
3453
Paolo Bonzinia343c9b2013-07-16 13:03:29 +02003454static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
3455{
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02003456 return kvm_io_bus_cmp(p1, p2);
Paolo Bonzinia343c9b2013-07-16 13:03:29 +02003457}
3458
Geoff Levand39369f72013-04-05 19:20:30 +00003459static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
Sasha Levin743eeb02011-07-27 16:00:48 +03003460 gpa_t addr, int len)
3461{
Sasha Levin743eeb02011-07-27 16:00:48 +03003462 bus->range[bus->dev_count++] = (struct kvm_io_range) {
3463 .addr = addr,
3464 .len = len,
3465 .dev = dev,
3466 };
3467
3468 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
3469 kvm_io_bus_sort_cmp, NULL);
3470
3471 return 0;
3472}
3473
Geoff Levand39369f72013-04-05 19:20:30 +00003474static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
Sasha Levin743eeb02011-07-27 16:00:48 +03003475 gpa_t addr, int len)
3476{
3477 struct kvm_io_range *range, key;
3478 int off;
3479
3480 key = (struct kvm_io_range) {
3481 .addr = addr,
3482 .len = len,
3483 };
3484
3485 range = bsearch(&key, bus->range, bus->dev_count,
3486 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
3487 if (range == NULL)
3488 return -ENOENT;
3489
3490 off = range - bus->range;
3491
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02003492 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
Sasha Levin743eeb02011-07-27 16:00:48 +03003493 off--;
3494
3495 return off;
3496}
3497
Nikolay Nikolaeve32edf42015-03-26 14:39:28 +00003498static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
Cornelia Huck126a5af2013-07-03 16:30:53 +02003499 struct kvm_io_range *range, const void *val)
3500{
3501 int idx;
3502
3503 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3504 if (idx < 0)
3505 return -EOPNOTSUPP;
3506
3507 while (idx < bus->dev_count &&
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02003508 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
Nikolay Nikolaeve32edf42015-03-26 14:39:28 +00003509 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
Cornelia Huck126a5af2013-07-03 16:30:53 +02003510 range->len, val))
3511 return idx;
3512 idx++;
3513 }
3514
3515 return -EOPNOTSUPP;
3516}
3517
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03003518/* kvm_io_bus_write - called under kvm->slots_lock */
Nikolay Nikolaeve32edf42015-03-26 14:39:28 +00003519int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03003520 int len, const void *val)
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04003521{
Cornelia Huck126a5af2013-07-03 16:30:53 +02003522 struct kvm_io_bus *bus;
3523 struct kvm_io_range range;
3524 int r;
3525
3526 range = (struct kvm_io_range) {
3527 .addr = addr,
3528 .len = len,
3529 };
3530
Nikolay Nikolaeve32edf42015-03-26 14:39:28 +00003531 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
David Hildenbrand15636252017-03-23 18:24:19 +01003532 if (!bus)
3533 return -ENOMEM;
Nikolay Nikolaeve32edf42015-03-26 14:39:28 +00003534 r = __kvm_io_bus_write(vcpu, bus, &range, val);
Cornelia Huck126a5af2013-07-03 16:30:53 +02003535 return r < 0 ? r : 0;
3536}
3537
3538/* kvm_io_bus_write_cookie - called under kvm->slots_lock */
Nikolay Nikolaeve32edf42015-03-26 14:39:28 +00003539int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3540 gpa_t addr, int len, const void *val, long cookie)
Cornelia Huck126a5af2013-07-03 16:30:53 +02003541{
Lai Jiangshan90d83dc2010-04-19 17:41:23 +08003542 struct kvm_io_bus *bus;
Sasha Levin743eeb02011-07-27 16:00:48 +03003543 struct kvm_io_range range;
3544
3545 range = (struct kvm_io_range) {
3546 .addr = addr,
3547 .len = len,
3548 };
Lai Jiangshan90d83dc2010-04-19 17:41:23 +08003549
Nikolay Nikolaeve32edf42015-03-26 14:39:28 +00003550 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
David Hildenbrand15636252017-03-23 18:24:19 +01003551 if (!bus)
3552 return -ENOMEM;
Cornelia Huck126a5af2013-07-03 16:30:53 +02003553
3554 /* First try the device referenced by cookie. */
3555 if ((cookie >= 0) && (cookie < bus->dev_count) &&
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02003556 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
Nikolay Nikolaeve32edf42015-03-26 14:39:28 +00003557 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
Cornelia Huck126a5af2013-07-03 16:30:53 +02003558 val))
3559 return cookie;
3560
3561 /*
3562 * cookie contained garbage; fall back to search and return the
3563 * correct cookie value.
3564 */
Nikolay Nikolaeve32edf42015-03-26 14:39:28 +00003565 return __kvm_io_bus_write(vcpu, bus, &range, val);
Cornelia Huck126a5af2013-07-03 16:30:53 +02003566}
3567
Nikolay Nikolaeve32edf42015-03-26 14:39:28 +00003568static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3569 struct kvm_io_range *range, void *val)
Cornelia Huck126a5af2013-07-03 16:30:53 +02003570{
3571 int idx;
3572
3573 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
Sasha Levin743eeb02011-07-27 16:00:48 +03003574 if (idx < 0)
3575 return -EOPNOTSUPP;
3576
3577 while (idx < bus->dev_count &&
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02003578 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
Nikolay Nikolaeve32edf42015-03-26 14:39:28 +00003579 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
Cornelia Huck126a5af2013-07-03 16:30:53 +02003580 range->len, val))
3581 return idx;
Sasha Levin743eeb02011-07-27 16:00:48 +03003582 idx++;
3583 }
3584
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03003585 return -EOPNOTSUPP;
3586}
Michael S. Tsirkin68c3b4d2014-03-31 21:50:44 +03003587EXPORT_SYMBOL_GPL(kvm_io_bus_write);
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04003588
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03003589/* kvm_io_bus_read - called under kvm->slots_lock */
Nikolay Nikolaeve32edf42015-03-26 14:39:28 +00003590int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003591 int len, void *val)
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03003592{
Cornelia Huck126a5af2013-07-03 16:30:53 +02003593 struct kvm_io_bus *bus;
3594 struct kvm_io_range range;
3595 int r;
3596
3597 range = (struct kvm_io_range) {
3598 .addr = addr,
3599 .len = len,
3600 };
3601
Nikolay Nikolaeve32edf42015-03-26 14:39:28 +00003602 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
David Hildenbrand15636252017-03-23 18:24:19 +01003603 if (!bus)
3604 return -ENOMEM;
Nikolay Nikolaeve32edf42015-03-26 14:39:28 +00003605 r = __kvm_io_bus_read(vcpu, bus, &range, val);
Cornelia Huck126a5af2013-07-03 16:30:53 +02003606 return r < 0 ? r : 0;
3607}
3608
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04003609
Marcelo Tosatti79fac952009-12-23 14:35:26 -02003610/* Caller must hold slots_lock. */
Sasha Levin743eeb02011-07-27 16:00:48 +03003611int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3612 int len, struct kvm_io_device *dev)
Michael S. Tsirkin6c474692009-06-29 22:24:26 +03003613{
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003614 struct kvm_io_bus *new_bus, *bus;
Gregory Haskins090b7af2009-07-07 17:08:44 -04003615
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003616 bus = kvm->buses[bus_idx];
David Hildenbrand15636252017-03-23 18:24:19 +01003617 if (!bus)
3618 return -ENOMEM;
3619
Amos Kong6ea34c92013-05-25 06:44:15 +08003620 /* exclude ioeventfd which is limited by maximum fd */
3621 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
Gregory Haskins090b7af2009-07-07 17:08:44 -04003622 return -ENOSPC;
3623
Jason Wangd3febdd2015-08-25 17:05:46 +08003624 new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count + 1) *
Amos Konga13007162012-03-09 12:17:32 +08003625 sizeof(struct kvm_io_range)), GFP_KERNEL);
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003626 if (!new_bus)
3627 return -ENOMEM;
Amos Konga13007162012-03-09 12:17:32 +08003628 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3629 sizeof(struct kvm_io_range)));
Sasha Levin743eeb02011-07-27 16:00:48 +03003630 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003631 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3632 synchronize_srcu_expedited(&kvm->srcu);
3633 kfree(bus);
Gregory Haskins090b7af2009-07-07 17:08:44 -04003634
3635 return 0;
3636}
3637
Marcelo Tosatti79fac952009-12-23 14:35:26 -02003638/* Caller must hold slots_lock. */
David Hildenbrand15636252017-03-23 18:24:19 +01003639void kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3640 struct kvm_io_device *dev)
Gregory Haskins090b7af2009-07-07 17:08:44 -04003641{
David Hildenbrand15636252017-03-23 18:24:19 +01003642 int i;
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003643 struct kvm_io_bus *new_bus, *bus;
Michael S. Tsirkin6c474692009-06-29 22:24:26 +03003644
Sasha Levincdfca7b2011-12-04 19:36:28 +02003645 bus = kvm->buses[bus_idx];
Peter Xuef46a132017-03-15 16:01:17 +08003646 if (!bus)
David Hildenbrand15636252017-03-23 18:24:19 +01003647 return;
Peter Xuef46a132017-03-15 16:01:17 +08003648
Amos Konga13007162012-03-09 12:17:32 +08003649 for (i = 0; i < bus->dev_count; i++)
3650 if (bus->range[i].dev == dev) {
Gregory Haskins090b7af2009-07-07 17:08:44 -04003651 break;
3652 }
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003653
David Hildenbrand15636252017-03-23 18:24:19 +01003654 if (i == bus->dev_count)
3655 return;
Amos Konga13007162012-03-09 12:17:32 +08003656
Jason Wangd3febdd2015-08-25 17:05:46 +08003657 new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count - 1) *
Amos Konga13007162012-03-09 12:17:32 +08003658 sizeof(struct kvm_io_range)), GFP_KERNEL);
David Hildenbrand15636252017-03-23 18:24:19 +01003659 if (!new_bus) {
3660 pr_err("kvm: failed to shrink bus, removing it completely\n");
3661 goto broken;
3662 }
Amos Konga13007162012-03-09 12:17:32 +08003663
3664 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3665 new_bus->dev_count--;
3666 memcpy(new_bus->range + i, bus->range + i + 1,
3667 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003668
David Hildenbrand15636252017-03-23 18:24:19 +01003669broken:
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003670 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3671 synchronize_srcu_expedited(&kvm->srcu);
3672 kfree(bus);
David Hildenbrand15636252017-03-23 18:24:19 +01003673 return;
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04003674}
3675
Andre Przywara8a39d002016-07-15 12:43:26 +01003676struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3677 gpa_t addr)
3678{
3679 struct kvm_io_bus *bus;
3680 int dev_idx, srcu_idx;
3681 struct kvm_io_device *iodev = NULL;
3682
3683 srcu_idx = srcu_read_lock(&kvm->srcu);
3684
3685 bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
David Hildenbrand15636252017-03-23 18:24:19 +01003686 if (!bus)
3687 goto out_unlock;
Andre Przywara8a39d002016-07-15 12:43:26 +01003688
3689 dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
3690 if (dev_idx < 0)
3691 goto out_unlock;
3692
3693 iodev = bus->range[dev_idx].dev;
3694
3695out_unlock:
3696 srcu_read_unlock(&kvm->srcu, srcu_idx);
3697
3698 return iodev;
3699}
3700EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
3701
Janosch Frank536a6f82016-05-18 13:26:23 +02003702static int kvm_debugfs_open(struct inode *inode, struct file *file,
3703 int (*get)(void *, u64 *), int (*set)(void *, u64),
3704 const char *fmt)
3705{
3706 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3707 inode->i_private;
3708
3709 /* The debugfs files are a reference to the kvm struct which
3710 * is still valid when kvm_destroy_vm is called.
3711 * To avoid the race between open and the removal of the debugfs
3712 * directory we test against the users count.
3713 */
3714 if (!atomic_add_unless(&stat_data->kvm->users_count, 1, 0))
3715 return -ENOENT;
3716
3717 if (simple_attr_open(inode, file, get, set, fmt)) {
3718 kvm_put_kvm(stat_data->kvm);
3719 return -ENOMEM;
3720 }
3721
3722 return 0;
3723}
3724
3725static int kvm_debugfs_release(struct inode *inode, struct file *file)
3726{
3727 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3728 inode->i_private;
3729
3730 simple_attr_release(inode, file);
3731 kvm_put_kvm(stat_data->kvm);
3732
3733 return 0;
3734}
3735
3736static int vm_stat_get_per_vm(void *data, u64 *val)
3737{
3738 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3739
Suraj Jitindar Singh8a7e75d2016-08-02 14:03:22 +10003740 *val = *(ulong *)((void *)stat_data->kvm + stat_data->offset);
Janosch Frank536a6f82016-05-18 13:26:23 +02003741
3742 return 0;
3743}
3744
3745static int vm_stat_get_per_vm_open(struct inode *inode, struct file *file)
3746{
3747 __simple_attr_check_format("%llu\n", 0ull);
3748 return kvm_debugfs_open(inode, file, vm_stat_get_per_vm,
3749 NULL, "%llu\n");
3750}
3751
3752static const struct file_operations vm_stat_get_per_vm_fops = {
3753 .owner = THIS_MODULE,
3754 .open = vm_stat_get_per_vm_open,
3755 .release = kvm_debugfs_release,
3756 .read = simple_attr_read,
3757 .write = simple_attr_write,
3758 .llseek = generic_file_llseek,
3759};
3760
3761static int vcpu_stat_get_per_vm(void *data, u64 *val)
3762{
3763 int i;
3764 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3765 struct kvm_vcpu *vcpu;
3766
3767 *val = 0;
3768
3769 kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
Suraj Jitindar Singh8a7e75d2016-08-02 14:03:22 +10003770 *val += *(u64 *)((void *)vcpu + stat_data->offset);
Janosch Frank536a6f82016-05-18 13:26:23 +02003771
3772 return 0;
3773}
3774
3775static int vcpu_stat_get_per_vm_open(struct inode *inode, struct file *file)
3776{
3777 __simple_attr_check_format("%llu\n", 0ull);
3778 return kvm_debugfs_open(inode, file, vcpu_stat_get_per_vm,
3779 NULL, "%llu\n");
3780}
3781
3782static const struct file_operations vcpu_stat_get_per_vm_fops = {
3783 .owner = THIS_MODULE,
3784 .open = vcpu_stat_get_per_vm_open,
3785 .release = kvm_debugfs_release,
3786 .read = simple_attr_read,
3787 .write = simple_attr_write,
3788 .llseek = generic_file_llseek,
3789};
3790
3791static const struct file_operations *stat_fops_per_vm[] = {
3792 [KVM_STAT_VCPU] = &vcpu_stat_get_per_vm_fops,
3793 [KVM_STAT_VM] = &vm_stat_get_per_vm_fops,
3794};
3795
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003796static int vm_stat_get(void *_offset, u64 *val)
Avi Kivityba1389b2007-11-18 16:24:12 +02003797{
3798 unsigned offset = (long)_offset;
Avi Kivityba1389b2007-11-18 16:24:12 +02003799 struct kvm *kvm;
Janosch Frank536a6f82016-05-18 13:26:23 +02003800 struct kvm_stat_data stat_tmp = {.offset = offset};
3801 u64 tmp_val;
Avi Kivityba1389b2007-11-18 16:24:12 +02003802
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003803 *val = 0;
Junaid Shahidc6170b82019-01-03 17:14:28 -08003804 mutex_lock(&kvm_lock);
Janosch Frank536a6f82016-05-18 13:26:23 +02003805 list_for_each_entry(kvm, &vm_list, vm_list) {
3806 stat_tmp.kvm = kvm;
3807 vm_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
3808 *val += tmp_val;
3809 }
Junaid Shahidc6170b82019-01-03 17:14:28 -08003810 mutex_unlock(&kvm_lock);
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003811 return 0;
Avi Kivityba1389b2007-11-18 16:24:12 +02003812}
3813
3814DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3815
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003816static int vcpu_stat_get(void *_offset, u64 *val)
Avi Kivity1165f5f2007-04-19 17:27:43 +03003817{
3818 unsigned offset = (long)_offset;
Avi Kivity1165f5f2007-04-19 17:27:43 +03003819 struct kvm *kvm;
Janosch Frank536a6f82016-05-18 13:26:23 +02003820 struct kvm_stat_data stat_tmp = {.offset = offset};
3821 u64 tmp_val;
Avi Kivity1165f5f2007-04-19 17:27:43 +03003822
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003823 *val = 0;
Junaid Shahidc6170b82019-01-03 17:14:28 -08003824 mutex_lock(&kvm_lock);
Janosch Frank536a6f82016-05-18 13:26:23 +02003825 list_for_each_entry(kvm, &vm_list, vm_list) {
3826 stat_tmp.kvm = kvm;
3827 vcpu_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
3828 *val += tmp_val;
3829 }
Junaid Shahidc6170b82019-01-03 17:14:28 -08003830 mutex_unlock(&kvm_lock);
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003831 return 0;
Avi Kivity1165f5f2007-04-19 17:27:43 +03003832}
3833
Avi Kivityba1389b2007-11-18 16:24:12 +02003834DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3835
Alexey Dobriyan828c0952009-10-01 15:43:56 -07003836static const struct file_operations *stat_fops[] = {
Avi Kivityba1389b2007-11-18 16:24:12 +02003837 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3838 [KVM_STAT_VM] = &vm_stat_fops,
3839};
Avi Kivity1165f5f2007-04-19 17:27:43 +03003840
Hamo4f69b682011-12-15 14:23:16 +08003841static int kvm_init_debug(void)
Avi Kivity6aa8b732006-12-10 02:21:36 -08003842{
Paolo Bonzini0c8eb042013-10-30 12:12:13 +01003843 int r = -EEXIST;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003844 struct kvm_stats_debugfs_item *p;
3845
Hollis Blanchard76f7c872008-04-15 16:05:42 -05003846 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
Hamo4f69b682011-12-15 14:23:16 +08003847 if (kvm_debugfs_dir == NULL)
3848 goto out;
3849
Janosch Frank536a6f82016-05-18 13:26:23 +02003850 kvm_debugfs_num_entries = 0;
3851 for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) {
Janosch Frank4bd33b52015-10-14 12:37:35 +02003852 if (!debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3853 (void *)(long)p->offset,
3854 stat_fops[p->kind]))
Hamo4f69b682011-12-15 14:23:16 +08003855 goto out_dir;
3856 }
3857
3858 return 0;
3859
3860out_dir:
3861 debugfs_remove_recursive(kvm_debugfs_dir);
3862out:
3863 return r;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003864}
3865
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003866static int kvm_suspend(void)
Avi Kivity59ae6c62007-02-12 00:54:48 -08003867{
Alexander Graf10474ae2009-09-15 11:37:46 +02003868 if (kvm_usage_count)
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09003869 hardware_disable_nolock(NULL);
Avi Kivity59ae6c62007-02-12 00:54:48 -08003870 return 0;
3871}
3872
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003873static void kvm_resume(void)
Avi Kivity59ae6c62007-02-12 00:54:48 -08003874{
Zachary Amsdenca84d1a2010-08-19 22:07:28 -10003875 if (kvm_usage_count) {
Paolo Bonzini4a937f92013-09-10 12:58:35 +02003876 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09003877 hardware_enable_nolock(NULL);
Zachary Amsdenca84d1a2010-08-19 22:07:28 -10003878 }
Avi Kivity59ae6c62007-02-12 00:54:48 -08003879}
3880
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003881static struct syscore_ops kvm_syscore_ops = {
Avi Kivity59ae6c62007-02-12 00:54:48 -08003882 .suspend = kvm_suspend,
3883 .resume = kvm_resume,
3884};
3885
Avi Kivity15ad7142007-07-11 18:17:21 +03003886static inline
3887struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3888{
3889 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3890}
3891
3892static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3893{
3894 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
Xiubo Lif95ef0c2015-02-26 14:58:23 +08003895
Raghavendra K T3a08a8f2013-03-04 23:32:07 +05303896 if (vcpu->preempted)
3897 vcpu->preempted = false;
Avi Kivity15ad7142007-07-11 18:17:21 +03003898
Radim Krčmáře790d9e2014-08-21 18:08:05 +02003899 kvm_arch_sched_in(vcpu, cpu);
3900
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003901 kvm_arch_vcpu_load(vcpu, cpu);
Avi Kivity15ad7142007-07-11 18:17:21 +03003902}
3903
3904static void kvm_sched_out(struct preempt_notifier *pn,
3905 struct task_struct *next)
3906{
3907 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3908
Raghavendra K T3a08a8f2013-03-04 23:32:07 +05303909 if (current->state == TASK_RUNNING)
3910 vcpu->preempted = true;
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003911 kvm_arch_vcpu_put(vcpu);
Avi Kivity15ad7142007-07-11 18:17:21 +03003912}
3913
Avi Kivity0ee75be2010-04-28 15:39:01 +03003914int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
Rusty Russellc16f8622007-07-30 21:12:19 +10003915 struct module *module)
Avi Kivity6aa8b732006-12-10 02:21:36 -08003916{
3917 int r;
Yang, Sheng002c7f72007-07-31 14:23:01 +03003918 int cpu;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003919
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08003920 r = kvm_arch_init(opaque);
3921 if (r)
Zhang Xiantaod23087842007-11-29 15:35:39 +08003922 goto out_fail;
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08003923
Asias He7dac16c2013-05-08 10:57:29 +08003924 /*
3925 * kvm_arch_init makes sure there's at most one caller
3926 * for architectures that support multiple implementations,
3927 * like intel and amd on x86.
Paolo Bonzini36343f62016-10-26 13:35:56 +02003928 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3929 * conflicts in case kvm is already setup for another implementation.
Asias He7dac16c2013-05-08 10:57:29 +08003930 */
Paolo Bonzini36343f62016-10-26 13:35:56 +02003931 r = kvm_irqfd_init();
3932 if (r)
3933 goto out_irqfd;
Asias He7dac16c2013-05-08 10:57:29 +08003934
Avi Kivity8437a612009-06-06 14:52:35 -07003935 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
Rusty Russell7f59f492008-12-07 21:25:45 +10303936 r = -ENOMEM;
3937 goto out_free_0;
3938 }
3939
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003940 r = kvm_arch_hardware_setup();
Avi Kivity6aa8b732006-12-10 02:21:36 -08003941 if (r < 0)
Rusty Russell7f59f492008-12-07 21:25:45 +10303942 goto out_free_0a;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003943
Yang, Sheng002c7f72007-07-31 14:23:01 +03003944 for_each_online_cpu(cpu) {
3945 smp_call_function_single(cpu,
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003946 kvm_arch_check_processor_compat,
Jens Axboe8691e5a2008-06-06 11:18:06 +02003947 &r, 1);
Yang, Sheng002c7f72007-07-31 14:23:01 +03003948 if (r < 0)
Zhang Xiantaod23087842007-11-29 15:35:39 +08003949 goto out_free_1;
Yang, Sheng002c7f72007-07-31 14:23:01 +03003950 }
3951
Thomas Gleixner8c18b2d2016-07-13 17:16:37 +00003952 r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "AP_KVM_STARTING",
3953 kvm_starting_cpu, kvm_dying_cpu);
Avi Kivity774c47f2007-02-12 00:54:47 -08003954 if (r)
Zhang Xiantaod23087842007-11-29 15:35:39 +08003955 goto out_free_2;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003956 register_reboot_notifier(&kvm_reboot_notifier);
3957
Rusty Russellc16f8622007-07-30 21:12:19 +10003958 /* A kmem cache lets us meet the alignment requirements of fx_save. */
Avi Kivity0ee75be2010-04-28 15:39:01 +03003959 if (!vcpu_align)
3960 vcpu_align = __alignof__(struct kvm_vcpu);
3961 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
Shakeel Butt206e1622017-10-05 18:07:24 -07003962 SLAB_ACCOUNT, NULL);
Rusty Russellc16f8622007-07-30 21:12:19 +10003963 if (!kvm_vcpu_cache) {
3964 r = -ENOMEM;
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003965 goto out_free_3;
Rusty Russellc16f8622007-07-30 21:12:19 +10003966 }
3967
Gleb Natapovaf585b92010-10-14 11:22:46 +02003968 r = kvm_async_pf_init();
3969 if (r)
3970 goto out_free;
3971
Avi Kivity6aa8b732006-12-10 02:21:36 -08003972 kvm_chardev_ops.owner = module;
Christian Borntraeger3d3aab12008-12-02 11:17:32 +01003973 kvm_vm_fops.owner = module;
3974 kvm_vcpu_fops.owner = module;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003975
3976 r = misc_register(&kvm_dev);
3977 if (r) {
Xiubo Li1170adc2015-02-26 14:58:26 +08003978 pr_err("kvm: misc device register failed\n");
Gleb Natapovaf585b92010-10-14 11:22:46 +02003979 goto out_unreg;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003980 }
3981
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003982 register_syscore_ops(&kvm_syscore_ops);
3983
Avi Kivity15ad7142007-07-11 18:17:21 +03003984 kvm_preempt_ops.sched_in = kvm_sched_in;
3985 kvm_preempt_ops.sched_out = kvm_sched_out;
3986
Hamo4f69b682011-12-15 14:23:16 +08003987 r = kvm_init_debug();
3988 if (r) {
Xiubo Li1170adc2015-02-26 14:58:26 +08003989 pr_err("kvm: create debugfs files failed\n");
Hamo4f69b682011-12-15 14:23:16 +08003990 goto out_undebugfs;
3991 }
Darrick J. Wong0ea4ed82009-10-14 16:21:00 -07003992
Paolo Bonzini3c3c29f2014-09-24 13:02:46 +02003993 r = kvm_vfio_ops_init();
3994 WARN_ON(r);
3995
Avi Kivityc7addb92007-09-16 18:58:32 +02003996 return 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003997
Hamo4f69b682011-12-15 14:23:16 +08003998out_undebugfs:
3999 unregister_syscore_ops(&kvm_syscore_ops);
Wei Yongjunafc2f792013-05-05 20:03:40 +08004000 misc_deregister(&kvm_dev);
Gleb Natapovaf585b92010-10-14 11:22:46 +02004001out_unreg:
4002 kvm_async_pf_deinit();
Avi Kivity6aa8b732006-12-10 02:21:36 -08004003out_free:
Rusty Russellc16f8622007-07-30 21:12:19 +10004004 kmem_cache_destroy(kvm_vcpu_cache);
Zhang Xiantaod23087842007-11-29 15:35:39 +08004005out_free_3:
Avi Kivity6aa8b732006-12-10 02:21:36 -08004006 unregister_reboot_notifier(&kvm_reboot_notifier);
Thomas Gleixner8c18b2d2016-07-13 17:16:37 +00004007 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
Zhang Xiantaod23087842007-11-29 15:35:39 +08004008out_free_2:
Zhang Xiantaod23087842007-11-29 15:35:39 +08004009out_free_1:
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08004010 kvm_arch_hardware_unsetup();
Rusty Russell7f59f492008-12-07 21:25:45 +10304011out_free_0a:
4012 free_cpumask_var(cpus_hardware_enabled);
Zhang Xiantaod23087842007-11-29 15:35:39 +08004013out_free_0:
Cornelia Hucka0f155e2013-02-28 12:33:18 +01004014 kvm_irqfd_exit();
Paolo Bonzini36343f62016-10-26 13:35:56 +02004015out_irqfd:
Asias He7dac16c2013-05-08 10:57:29 +08004016 kvm_arch_exit();
4017out_fail:
Avi Kivity6aa8b732006-12-10 02:21:36 -08004018 return r;
4019}
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08004020EXPORT_SYMBOL_GPL(kvm_init);
Avi Kivity6aa8b732006-12-10 02:21:36 -08004021
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08004022void kvm_exit(void)
Avi Kivity6aa8b732006-12-10 02:21:36 -08004023{
Janosch Frank4bd33b52015-10-14 12:37:35 +02004024 debugfs_remove_recursive(kvm_debugfs_dir);
Avi Kivity6aa8b732006-12-10 02:21:36 -08004025 misc_deregister(&kvm_dev);
Rusty Russellc16f8622007-07-30 21:12:19 +10004026 kmem_cache_destroy(kvm_vcpu_cache);
Gleb Natapovaf585b92010-10-14 11:22:46 +02004027 kvm_async_pf_deinit();
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01004028 unregister_syscore_ops(&kvm_syscore_ops);
Avi Kivity6aa8b732006-12-10 02:21:36 -08004029 unregister_reboot_notifier(&kvm_reboot_notifier);
Thomas Gleixner8c18b2d2016-07-13 17:16:37 +00004030 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09004031 on_each_cpu(hardware_disable_nolock, NULL, 1);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08004032 kvm_arch_hardware_unsetup();
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08004033 kvm_arch_exit();
Cornelia Hucka0f155e2013-02-28 12:33:18 +01004034 kvm_irqfd_exit();
Rusty Russell7f59f492008-12-07 21:25:45 +10304035 free_cpumask_var(cpus_hardware_enabled);
Wanpeng Li571ee1b2014-10-09 18:30:08 +08004036 kvm_vfio_ops_exit();
Avi Kivity6aa8b732006-12-10 02:21:36 -08004037}
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08004038EXPORT_SYMBOL_GPL(kvm_exit);
Junaid Shahid61e191b2019-11-04 12:22:02 +01004039
4040struct kvm_vm_worker_thread_context {
4041 struct kvm *kvm;
4042 struct task_struct *parent;
4043 struct completion init_done;
4044 kvm_vm_thread_fn_t thread_fn;
4045 uintptr_t data;
4046 int err;
4047};
4048
4049static int kvm_vm_worker_thread(void *context)
4050{
4051 /*
4052 * The init_context is allocated on the stack of the parent thread, so
4053 * we have to locally copy anything that is needed beyond initialization
4054 */
4055 struct kvm_vm_worker_thread_context *init_context = context;
4056 struct kvm *kvm = init_context->kvm;
4057 kvm_vm_thread_fn_t thread_fn = init_context->thread_fn;
4058 uintptr_t data = init_context->data;
4059 int err;
4060
4061 err = kthread_park(current);
4062 /* kthread_park(current) is never supposed to return an error */
4063 WARN_ON(err != 0);
4064 if (err)
4065 goto init_complete;
4066
4067 err = cgroup_attach_task_all(init_context->parent, current);
4068 if (err) {
4069 kvm_err("%s: cgroup_attach_task_all failed with err %d\n",
4070 __func__, err);
4071 goto init_complete;
4072 }
4073
4074 set_user_nice(current, task_nice(init_context->parent));
4075
4076init_complete:
4077 init_context->err = err;
4078 complete(&init_context->init_done);
4079 init_context = NULL;
4080
4081 if (err)
4082 return err;
4083
4084 /* Wait to be woken up by the spawner before proceeding. */
4085 kthread_parkme();
4086
4087 if (!kthread_should_stop())
4088 err = thread_fn(kvm, data);
4089
4090 return err;
4091}
4092
4093int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn,
4094 uintptr_t data, const char *name,
4095 struct task_struct **thread_ptr)
4096{
4097 struct kvm_vm_worker_thread_context init_context = {};
4098 struct task_struct *thread;
4099
4100 *thread_ptr = NULL;
4101 init_context.kvm = kvm;
4102 init_context.parent = current;
4103 init_context.thread_fn = thread_fn;
4104 init_context.data = data;
4105 init_completion(&init_context.init_done);
4106
4107 thread = kthread_run(kvm_vm_worker_thread, &init_context,
4108 "%s-%d", name, task_pid_nr(current));
4109 if (IS_ERR(thread))
4110 return PTR_ERR(thread);
4111
4112 /* kthread_run is never supposed to return NULL */
4113 WARN_ON(thread == NULL);
4114
4115 wait_for_completion(&init_context.init_done);
4116
4117 if (!init_context.err)
4118 *thread_ptr = thread;
4119
4120 return init_context.err;
4121}