blob: 37af23052470eff99b2c3a3fdcf05898c7312602 [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>
Avi Kivity6aa8b732006-12-10 02:21:36 -080052
Avi Kivitye4956062007-06-28 14:15:57 -040053#include <asm/processor.h>
Avi Kivitye4956062007-06-28 14:15:57 -040054#include <asm/io.h>
David Matlack2ea75be2014-09-19 16:03:25 -070055#include <asm/ioctl.h>
Avi Kivitye4956062007-06-28 14:15:57 -040056#include <asm/uaccess.h>
Izik Eidus3e021bf2007-11-19 11:16:57 +020057#include <asm/pgtable.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080058
Laurent Vivier5f94c172008-05-30 16:05:54 +020059#include "coalesced_mmio.h"
Gleb Natapovaf585b92010-10-14 11:22:46 +020060#include "async_pf.h"
Paolo Bonzini3c3c29f2014-09-24 13:02:46 +020061#include "vfio.h"
Laurent Vivier5f94c172008-05-30 16:05:54 +020062
Marcelo Tosatti229456f2009-06-17 09:22:14 -030063#define CREATE_TRACE_POINTS
64#include <trace/events/kvm.h>
65
Janosch Frank536a6f82016-05-18 13:26:23 +020066/* Worst case buffer size needed for holding an integer. */
67#define ITOA_MAX_LEN 12
68
Avi Kivity6aa8b732006-12-10 02:21:36 -080069MODULE_AUTHOR("Qumranet");
70MODULE_LICENSE("GPL");
71
David Hildenbrand920552b2015-09-18 12:34:53 +020072/* Architectures should define their poll value according to the halt latency */
73static unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT;
Paolo Bonzinif7819512015-02-04 18:20:58 +010074module_param(halt_poll_ns, uint, S_IRUGO | S_IWUSR);
75
Wanpeng Liaca6ff22015-09-03 22:07:38 +080076/* Default doubles per-vcpu halt_poll_ns. */
77static unsigned int halt_poll_ns_grow = 2;
Christian Borntraeger6b6de682016-02-09 13:47:55 +010078module_param(halt_poll_ns_grow, uint, S_IRUGO | S_IWUSR);
Wanpeng Liaca6ff22015-09-03 22:07:38 +080079
80/* Default resets per-vcpu halt_poll_ns . */
81static unsigned int halt_poll_ns_shrink;
Christian Borntraeger6b6de682016-02-09 13:47:55 +010082module_param(halt_poll_ns_shrink, uint, S_IRUGO | S_IWUSR);
Wanpeng Liaca6ff22015-09-03 22:07:38 +080083
Marcelo Tosattifa40a822009-06-04 15:08:24 -030084/*
85 * Ordering of locks:
86 *
Xiubo Lib7d409d2015-02-26 14:58:24 +080087 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
Marcelo Tosattifa40a822009-06-04 15:08:24 -030088 */
89
Paolo Bonzini2f303b72013-09-25 13:53:07 +020090DEFINE_SPINLOCK(kvm_lock);
Paolo Bonzini4a937f92013-09-10 12:58:35 +020091static DEFINE_RAW_SPINLOCK(kvm_count_lock);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +080092LIST_HEAD(vm_list);
Avi Kivity133de902007-02-12 00:54:44 -080093
Rusty Russell7f59f492008-12-07 21:25:45 +103094static cpumask_var_t cpus_hardware_enabled;
Xiubo Lif4fee932015-02-26 14:58:21 +080095static int kvm_usage_count;
Alexander Graf10474ae2009-09-15 11:37:46 +020096static atomic_t hardware_enable_failed;
Avi Kivity1b6c0162007-05-24 13:03:52 +030097
Rusty Russellc16f8622007-07-30 21:12:19 +100098struct kmem_cache *kvm_vcpu_cache;
99EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
Avi Kivity1165f5f2007-04-19 17:27:43 +0300100
Avi Kivity15ad7142007-07-11 18:17:21 +0300101static __read_mostly struct preempt_ops kvm_preempt_ops;
102
Hollis Blanchard76f7c872008-04-15 16:05:42 -0500103struct dentry *kvm_debugfs_dir;
Paul Mackerrase23a8082015-03-28 14:21:01 +1100104EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800105
Janosch Frank536a6f82016-05-18 13:26:23 +0200106static int kvm_debugfs_num_entries;
107static const struct file_operations *stat_fops_per_vm[];
108
Avi Kivitybccf2152007-02-21 18:04:26 +0200109static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
110 unsigned long arg);
Christian Borntraegerde8e5d72015-02-03 09:35:15 +0100111#ifdef CONFIG_KVM_COMPAT
Alexander Graf1dda6062011-06-08 02:45:37 +0200112static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
113 unsigned long arg);
114#endif
Alexander Graf10474ae2009-09-15 11:37:46 +0200115static int hardware_enable_all(void);
116static void hardware_disable_all(void);
Avi Kivitybccf2152007-02-21 18:04:26 +0200117
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200118static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
Stephen Hemminger79408762013-12-29 12:12:29 -0800119
Dan Williamsba049e92016-01-15 16:56:11 -0800120static void kvm_release_pfn_dirty(kvm_pfn_t pfn);
Paolo Bonzinibc009e42015-05-26 12:43:41 +0200121static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, gfn_t gfn);
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200122
Andi Kleen52480132014-02-08 08:51:57 +0100123__visible bool kvm_rebooting;
Avi Kivityb7c41452010-12-02 17:52:50 +0200124EXPORT_SYMBOL_GPL(kvm_rebooting);
Avi Kivity4ecac3f2008-05-13 13:23:38 +0300125
Marcelo Tosatti54dee992009-06-11 12:07:44 -0300126static bool largepages_enabled = true;
127
Dan Williamsba049e92016-01-15 16:56:11 -0800128bool kvm_is_reserved_pfn(kvm_pfn_t pfn)
Ben-Ami Yassourcbff90a2008-07-28 19:26:24 +0300129{
Andrea Arcangeli11feeb42013-07-25 03:04:38 +0200130 if (pfn_valid(pfn))
Ard Biesheuvelbf4bea82014-11-10 08:33:56 +0000131 return PageReserved(pfn_to_page(pfn));
Ben-Ami Yassourcbff90a2008-07-28 19:26:24 +0300132
133 return true;
134}
135
Avi Kivity6aa8b732006-12-10 02:21:36 -0800136/*
137 * Switches to specified vcpu, until a matching vcpu_put()
138 */
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +0300139int vcpu_load(struct kvm_vcpu *vcpu)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800140{
Avi Kivity15ad7142007-07-11 18:17:21 +0300141 int cpu;
142
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +0300143 if (mutex_lock_killable(&vcpu->mutex))
144 return -EINTR;
Avi Kivity15ad7142007-07-11 18:17:21 +0300145 cpu = get_cpu();
146 preempt_notifier_register(&vcpu->preempt_notifier);
Carsten Otte313a3dc2007-10-11 19:16:52 +0200147 kvm_arch_vcpu_load(vcpu, cpu);
Avi Kivity15ad7142007-07-11 18:17:21 +0300148 put_cpu();
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +0300149 return 0;
Avi Kivitybccf2152007-02-21 18:04:26 +0200150}
151
Carsten Otte313a3dc2007-10-11 19:16:52 +0200152void vcpu_put(struct kvm_vcpu *vcpu)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800153{
Avi Kivity15ad7142007-07-11 18:17:21 +0300154 preempt_disable();
Carsten Otte313a3dc2007-10-11 19:16:52 +0200155 kvm_arch_vcpu_put(vcpu);
Avi Kivity15ad7142007-07-11 18:17:21 +0300156 preempt_notifier_unregister(&vcpu->preempt_notifier);
157 preempt_enable();
Avi Kivity6aa8b732006-12-10 02:21:36 -0800158 mutex_unlock(&vcpu->mutex);
159}
160
Avi Kivityd9e368d2007-06-07 19:18:30 +0300161static void ack_flush(void *_completed)
162{
Avi Kivityd9e368d2007-06-07 19:18:30 +0300163}
164
Tang Chen445b8232014-09-24 15:57:55 +0800165bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
Avi Kivityd9e368d2007-06-07 19:18:30 +0300166{
Avi Kivity597a5f52008-07-20 14:24:22 +0300167 int i, cpu, me;
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030168 cpumask_var_t cpus;
169 bool called = true;
Avi Kivityd9e368d2007-06-07 19:18:30 +0300170 struct kvm_vcpu *vcpu;
Avi Kivityd9e368d2007-06-07 19:18:30 +0300171
Li Zefan79f55992009-06-15 14:58:26 +0800172 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030173
Xiao Guangrong3cba4132011-01-12 15:41:22 +0800174 me = get_cpu();
Gleb Natapov988a2ca2009-06-09 15:56:29 +0300175 kvm_for_each_vcpu(i, vcpu, kvm) {
Xiao Guangrong3cba4132011-01-12 15:41:22 +0800176 kvm_make_request(req, vcpu);
Avi Kivityd9e368d2007-06-07 19:18:30 +0300177 cpu = vcpu->cpu;
Xiao Guangrong6b7e2d02011-01-12 15:40:31 +0800178
Lan Tianyua30a0502016-03-13 11:10:26 +0800179 /* Set ->requests bit before we read ->mode. */
180 smp_mb__after_atomic();
Xiao Guangrong6b7e2d02011-01-12 15:40:31 +0800181
182 if (cpus != NULL && cpu != -1 && cpu != me &&
183 kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030184 cpumask_set_cpu(cpu, cpus);
Avi Kivityd9e368d2007-06-07 19:18:30 +0300185 }
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030186 if (unlikely(cpus == NULL))
187 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
188 else if (!cpumask_empty(cpus))
189 smp_call_function_many(cpus, ack_flush, NULL, 1);
190 else
191 called = false;
Xiao Guangrong3cba4132011-01-12 15:41:22 +0800192 put_cpu();
Rusty Russell6ef7a1b2008-12-08 20:28:04 +1030193 free_cpumask_var(cpus);
Rusty Russell49846892008-12-08 20:26:24 +1030194 return called;
195}
196
Mario Smarducha6d51012015-01-15 15:58:52 -0800197#ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
Rusty Russell49846892008-12-08 20:26:24 +1030198void kvm_flush_remote_tlbs(struct kvm *kvm)
199{
Lan Tianyu4ae3cb32016-03-13 11:10:28 +0800200 /*
201 * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
202 * kvm_make_all_cpus_request.
203 */
204 long dirty_count = smp_load_acquire(&kvm->tlbs_dirty);
Xiao Guangronga086f6a2014-04-17 17:06:12 +0800205
Lan Tianyu4ae3cb32016-03-13 11:10:28 +0800206 /*
207 * We want to publish modifications to the page tables before reading
208 * mode. Pairs with a memory barrier in arch-specific code.
209 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
210 * and smp_mb in walk_shadow_page_lockless_begin/end.
211 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
212 *
213 * There is already an smp_mb__after_atomic() before
214 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
215 * barrier here.
216 */
Tang Chen445b8232014-09-24 15:57:55 +0800217 if (kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
Rusty Russell49846892008-12-08 20:26:24 +1030218 ++kvm->stat.remote_tlb_flush;
Xiao Guangronga086f6a2014-04-17 17:06:12 +0800219 cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
Avi Kivityd9e368d2007-06-07 19:18:30 +0300220}
Aneesh Kumar K.V2ba9f0d2013-10-07 22:17:59 +0530221EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
Mario Smarducha6d51012015-01-15 15:58:52 -0800222#endif
Avi Kivityd9e368d2007-06-07 19:18:30 +0300223
Marcelo Tosatti2e53d632008-02-20 14:47:24 -0500224void kvm_reload_remote_mmus(struct kvm *kvm)
225{
Tang Chen445b8232014-09-24 15:57:55 +0800226 kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
Marcelo Tosatti2e53d632008-02-20 14:47:24 -0500227}
228
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000229int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
230{
231 struct page *page;
232 int r;
233
234 mutex_init(&vcpu->mutex);
235 vcpu->cpu = -1;
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000236 vcpu->kvm = kvm;
237 vcpu->vcpu_id = id;
Rik van Riel34bb10b2011-02-01 09:52:41 -0500238 vcpu->pid = NULL;
Marcelo Tosatti85773702016-02-19 09:46:39 +0100239 init_swait_queue_head(&vcpu->wq);
Gleb Natapovaf585b92010-10-14 11:22:46 +0200240 kvm_async_pf_vcpu_init(vcpu);
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000241
Feng Wubf9f6ac2015-09-18 22:29:55 +0800242 vcpu->pre_pcpu = -1;
243 INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
244
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000245 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
246 if (!page) {
247 r = -ENOMEM;
248 goto fail;
249 }
250 vcpu->run = page_address(page);
251
Raghavendra K T4c088492012-07-18 19:07:46 +0530252 kvm_vcpu_set_in_spin_loop(vcpu, false);
253 kvm_vcpu_set_dy_eligible(vcpu, false);
Raghavendra K T3a08a8f2013-03-04 23:32:07 +0530254 vcpu->preempted = false;
Raghavendra K T4c088492012-07-18 19:07:46 +0530255
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +0800256 r = kvm_arch_vcpu_init(vcpu);
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000257 if (r < 0)
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +0800258 goto fail_free_run;
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000259 return 0;
260
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000261fail_free_run:
262 free_page((unsigned long)vcpu->run);
263fail:
Rusty Russell76fafa52007-10-08 10:50:48 +1000264 return r;
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000265}
266EXPORT_SYMBOL_GPL(kvm_vcpu_init);
267
268void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
269{
Rik van Riel34bb10b2011-02-01 09:52:41 -0500270 put_pid(vcpu->pid);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +0800271 kvm_arch_vcpu_uninit(vcpu);
Rusty Russellfb3f0f52007-07-27 17:16:56 +1000272 free_page((unsigned long)vcpu->run);
273}
274EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
275
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200276#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
277static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
278{
279 return container_of(mn, struct kvm, mmu_notifier);
280}
281
282static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
283 struct mm_struct *mm,
284 unsigned long address)
285{
286 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200287 int need_tlb_flush, idx;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200288
289 /*
290 * When ->invalidate_page runs, the linux pte has been zapped
291 * already but the page is still allocated until
292 * ->invalidate_page returns. So if we increase the sequence
293 * here the kvm page fault will notice if the spte can't be
294 * established because the page is going to be freed. If
295 * instead the kvm page fault establishes the spte before
296 * ->invalidate_page runs, kvm_unmap_hva will release it
297 * before returning.
298 *
299 * The sequence increase only need to be seen at spin_unlock
300 * time, and not at spin_lock time.
301 *
302 * Increasing the sequence after the spin_unlock would be
303 * unsafe because the kvm page fault could then establish the
304 * pte after kvm_unmap_hva returned, without noticing the page
305 * is going to be freed.
306 */
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200307 idx = srcu_read_lock(&kvm->srcu);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200308 spin_lock(&kvm->mmu_lock);
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900309
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200310 kvm->mmu_notifier_seq++;
Xiao Guangronga4ee1ca2010-11-23 11:13:00 +0800311 need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200312 /* we've to flush the tlb before the pages can be freed */
313 if (need_tlb_flush)
314 kvm_flush_remote_tlbs(kvm);
315
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900316 spin_unlock(&kvm->mmu_lock);
Tang Chenfe715572014-09-24 15:57:57 +0800317
318 kvm_arch_mmu_notifier_invalidate_page(kvm, address);
319
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900320 srcu_read_unlock(&kvm->srcu, idx);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200321}
322
Izik Eidus3da0dd42009-09-23 21:47:18 +0300323static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
324 struct mm_struct *mm,
325 unsigned long address,
326 pte_t pte)
327{
328 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200329 int idx;
Izik Eidus3da0dd42009-09-23 21:47:18 +0300330
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200331 idx = srcu_read_lock(&kvm->srcu);
Izik Eidus3da0dd42009-09-23 21:47:18 +0300332 spin_lock(&kvm->mmu_lock);
333 kvm->mmu_notifier_seq++;
334 kvm_set_spte_hva(kvm, address, pte);
335 spin_unlock(&kvm->mmu_lock);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200336 srcu_read_unlock(&kvm->srcu, idx);
Izik Eidus3da0dd42009-09-23 21:47:18 +0300337}
338
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200339static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
340 struct mm_struct *mm,
341 unsigned long start,
342 unsigned long end)
343{
344 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200345 int need_tlb_flush = 0, idx;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200346
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200347 idx = srcu_read_lock(&kvm->srcu);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200348 spin_lock(&kvm->mmu_lock);
349 /*
350 * The count increase must become visible at unlock time as no
351 * spte can be established without taking the mmu_lock and
352 * count is also read inside the mmu_lock critical section.
353 */
354 kvm->mmu_notifier_count++;
Takuya Yoshikawab3ae2092012-07-02 17:56:33 +0900355 need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
Xiao Guangronga4ee1ca2010-11-23 11:13:00 +0800356 need_tlb_flush |= kvm->tlbs_dirty;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200357 /* we've to flush the tlb before the pages can be freed */
358 if (need_tlb_flush)
359 kvm_flush_remote_tlbs(kvm);
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900360
361 spin_unlock(&kvm->mmu_lock);
362 srcu_read_unlock(&kvm->srcu, idx);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200363}
364
365static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
366 struct mm_struct *mm,
367 unsigned long start,
368 unsigned long end)
369{
370 struct kvm *kvm = mmu_notifier_to_kvm(mn);
371
372 spin_lock(&kvm->mmu_lock);
373 /*
374 * This sequence increase will notify the kvm page fault that
375 * the page that is going to be mapped in the spte could have
376 * been freed.
377 */
378 kvm->mmu_notifier_seq++;
Paul Mackerrasa355aa52011-12-12 12:37:21 +0000379 smp_wmb();
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200380 /*
381 * The above sequence increase must be visible before the
Paul Mackerrasa355aa52011-12-12 12:37:21 +0000382 * below count decrease, which is ensured by the smp_wmb above
383 * in conjunction with the smp_rmb in mmu_notifier_retry().
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200384 */
385 kvm->mmu_notifier_count--;
386 spin_unlock(&kvm->mmu_lock);
387
388 BUG_ON(kvm->mmu_notifier_count < 0);
389}
390
391static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
392 struct mm_struct *mm,
Andres Lagar-Cavilla57128462014-09-22 14:54:42 -0700393 unsigned long start,
394 unsigned long end)
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200395{
396 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200397 int young, idx;
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200398
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200399 idx = srcu_read_lock(&kvm->srcu);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200400 spin_lock(&kvm->mmu_lock);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200401
Andres Lagar-Cavilla57128462014-09-22 14:54:42 -0700402 young = kvm_age_hva(kvm, start, end);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200403 if (young)
404 kvm_flush_remote_tlbs(kvm);
405
Takuya Yoshikawa565f3be2012-02-10 15:28:31 +0900406 spin_unlock(&kvm->mmu_lock);
407 srcu_read_unlock(&kvm->srcu, idx);
408
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200409 return young;
410}
411
Vladimir Davydov1d7715c2015-09-09 15:35:41 -0700412static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
413 struct mm_struct *mm,
414 unsigned long start,
415 unsigned long end)
416{
417 struct kvm *kvm = mmu_notifier_to_kvm(mn);
418 int young, idx;
419
420 idx = srcu_read_lock(&kvm->srcu);
421 spin_lock(&kvm->mmu_lock);
422 /*
423 * Even though we do not flush TLB, this will still adversely
424 * affect performance on pre-Haswell Intel EPT, where there is
425 * no EPT Access Bit to clear so that we have to tear down EPT
426 * tables instead. If we find this unacceptable, we can always
427 * add a parameter to kvm_age_hva so that it effectively doesn't
428 * do anything on clear_young.
429 *
430 * Also note that currently we never issue secondary TLB flushes
431 * from clear_young, leaving this job up to the regular system
432 * cadence. If we find this inaccurate, we might come up with a
433 * more sophisticated heuristic later.
434 */
435 young = kvm_age_hva(kvm, start, end);
436 spin_unlock(&kvm->mmu_lock);
437 srcu_read_unlock(&kvm->srcu, idx);
438
439 return young;
440}
441
Andrea Arcangeli8ee53822011-01-13 15:47:10 -0800442static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
443 struct mm_struct *mm,
444 unsigned long address)
445{
446 struct kvm *kvm = mmu_notifier_to_kvm(mn);
447 int young, idx;
448
449 idx = srcu_read_lock(&kvm->srcu);
450 spin_lock(&kvm->mmu_lock);
451 young = kvm_test_age_hva(kvm, address);
452 spin_unlock(&kvm->mmu_lock);
453 srcu_read_unlock(&kvm->srcu, idx);
454
455 return young;
456}
457
Marcelo Tosatti85db06e2008-12-10 21:23:26 +0100458static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
459 struct mm_struct *mm)
460{
461 struct kvm *kvm = mmu_notifier_to_kvm(mn);
Lai Jiangshaneda2bed2010-04-20 14:29:29 +0800462 int idx;
463
464 idx = srcu_read_lock(&kvm->srcu);
Marcelo Tosatti2df72e92012-08-24 15:54:57 -0300465 kvm_arch_flush_shadow_all(kvm);
Lai Jiangshaneda2bed2010-04-20 14:29:29 +0800466 srcu_read_unlock(&kvm->srcu, idx);
Marcelo Tosatti85db06e2008-12-10 21:23:26 +0100467}
468
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200469static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
470 .invalidate_page = kvm_mmu_notifier_invalidate_page,
471 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
472 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
473 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
Vladimir Davydov1d7715c2015-09-09 15:35:41 -0700474 .clear_young = kvm_mmu_notifier_clear_young,
Andrea Arcangeli8ee53822011-01-13 15:47:10 -0800475 .test_young = kvm_mmu_notifier_test_young,
Izik Eidus3da0dd42009-09-23 21:47:18 +0300476 .change_pte = kvm_mmu_notifier_change_pte,
Marcelo Tosatti85db06e2008-12-10 21:23:26 +0100477 .release = kvm_mmu_notifier_release,
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200478};
Avi Kivity4c07b0a2009-12-20 14:54:04 +0200479
480static int kvm_init_mmu_notifier(struct kvm *kvm)
481{
482 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
483 return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
484}
485
486#else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
487
488static int kvm_init_mmu_notifier(struct kvm *kvm)
489{
490 return 0;
491}
492
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200493#endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
494
Paolo Bonzinia47d2b02015-05-17 11:41:37 +0200495static struct kvm_memslots *kvm_alloc_memslots(void)
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800496{
497 int i;
Paolo Bonzinia47d2b02015-05-17 11:41:37 +0200498 struct kvm_memslots *slots;
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800499
Paolo Bonzinia47d2b02015-05-17 11:41:37 +0200500 slots = kvm_kvzalloc(sizeof(struct kvm_memslots));
501 if (!slots)
502 return NULL;
503
504 /*
505 * Init kvm generation close to the maximum to easily test the
506 * code of handling generation number wrap-around.
507 */
508 slots->generation = -150;
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800509 for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
Xiao Guangrongf85e2cb2011-11-24 17:41:54 +0800510 slots->id_to_index[i] = slots->memslots[i].id = i;
Paolo Bonzinia47d2b02015-05-17 11:41:37 +0200511
512 return slots;
513}
514
515static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
516{
517 if (!memslot->dirty_bitmap)
518 return;
519
520 kvfree(memslot->dirty_bitmap);
521 memslot->dirty_bitmap = NULL;
522}
523
524/*
525 * Free any memory in @free but not in @dont.
526 */
527static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
528 struct kvm_memory_slot *dont)
529{
530 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
531 kvm_destroy_dirty_bitmap(free);
532
533 kvm_arch_free_memslot(kvm, free, dont);
534
535 free->npages = 0;
536}
537
538static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
539{
540 struct kvm_memory_slot *memslot;
541
542 if (!slots)
543 return;
544
545 kvm_for_each_memslot(memslot, slots)
546 kvm_free_memslot(kvm, memslot, NULL);
547
548 kvfree(slots);
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800549}
550
Janosch Frank536a6f82016-05-18 13:26:23 +0200551static void kvm_destroy_vm_debugfs(struct kvm *kvm)
552{
553 int i;
554
555 if (!kvm->debugfs_dentry)
556 return;
557
558 debugfs_remove_recursive(kvm->debugfs_dentry);
559
560 for (i = 0; i < kvm_debugfs_num_entries; i++)
561 kfree(kvm->debugfs_stat_data[i]);
562 kfree(kvm->debugfs_stat_data);
563}
564
565static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
566{
567 char dir_name[ITOA_MAX_LEN * 2];
568 struct kvm_stat_data *stat_data;
569 struct kvm_stats_debugfs_item *p;
570
571 if (!debugfs_initialized())
572 return 0;
573
574 snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
575 kvm->debugfs_dentry = debugfs_create_dir(dir_name,
576 kvm_debugfs_dir);
577 if (!kvm->debugfs_dentry)
578 return -ENOMEM;
579
580 kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
581 sizeof(*kvm->debugfs_stat_data),
582 GFP_KERNEL);
583 if (!kvm->debugfs_stat_data)
584 return -ENOMEM;
585
586 for (p = debugfs_entries; p->name; p++) {
587 stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL);
588 if (!stat_data)
589 return -ENOMEM;
590
591 stat_data->kvm = kvm;
592 stat_data->offset = p->offset;
593 kvm->debugfs_stat_data[p - debugfs_entries] = stat_data;
594 if (!debugfs_create_file(p->name, 0444,
595 kvm->debugfs_dentry,
596 stat_data,
597 stat_fops_per_vm[p->kind]))
598 return -ENOMEM;
599 }
600 return 0;
601}
602
Carsten Ottee08b9632012-01-04 10:25:20 +0100603static struct kvm *kvm_create_vm(unsigned long type)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800604{
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100605 int r, i;
606 struct kvm *kvm = kvm_arch_alloc_vm();
Avi Kivity6aa8b732006-12-10 02:21:36 -0800607
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100608 if (!kvm)
609 return ERR_PTR(-ENOMEM);
610
Paolo Bonzinie9ad4ec2016-03-21 10:15:25 +0100611 spin_lock_init(&kvm->mmu_lock);
612 atomic_inc(&current->mm->mm_count);
613 kvm->mm = current->mm;
614 kvm_eventfd_init(kvm);
615 mutex_init(&kvm->lock);
616 mutex_init(&kvm->irq_lock);
617 mutex_init(&kvm->slots_lock);
618 atomic_set(&kvm->users_count, 1);
619 INIT_LIST_HEAD(&kvm->devices);
620
Carsten Ottee08b9632012-01-04 10:25:20 +0100621 r = kvm_arch_init_vm(kvm, type);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100622 if (r)
Christian Borntraeger719d93c2014-01-16 13:44:20 +0100623 goto out_err_no_disable;
Alexander Graf10474ae2009-09-15 11:37:46 +0200624
625 r = hardware_enable_all();
626 if (r)
Christian Borntraeger719d93c2014-01-16 13:44:20 +0100627 goto out_err_no_disable;
Alexander Graf10474ae2009-09-15 11:37:46 +0200628
Paolo Bonzinic77dcac2014-08-06 14:24:45 +0200629#ifdef CONFIG_HAVE_KVM_IRQFD
Gleb Natapov136bdfe2009-08-24 11:54:23 +0300630 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
Avi Kivity75858a82009-01-04 17:10:50 +0200631#endif
Avi Kivity6aa8b732006-12-10 02:21:36 -0800632
Alex Williamson1e702d92012-12-10 10:33:32 -0700633 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
634
Marcelo Tosatti46a26bf2009-12-23 14:35:16 -0200635 r = -ENOMEM;
Paolo Bonzinif481b062015-05-17 17:30:37 +0200636 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
637 kvm->memslots[i] = kvm_alloc_memslots();
638 if (!kvm->memslots[i])
639 goto out_err_no_srcu;
640 }
Paolo Bonzini00f034a2014-08-20 14:29:21 +0200641
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200642 if (init_srcu_struct(&kvm->srcu))
Christian Borntraeger719d93c2014-01-16 13:44:20 +0100643 goto out_err_no_srcu;
644 if (init_srcu_struct(&kvm->irq_srcu))
645 goto out_err_no_irq_srcu;
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200646 for (i = 0; i < KVM_NR_BUSES; i++) {
647 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
648 GFP_KERNEL);
Jan Kiszka57e7fbe2010-11-09 12:42:12 +0100649 if (!kvm->buses[i])
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200650 goto out_err;
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200651 }
Mike Waychison74b5c5b2011-06-03 13:04:53 -0700652
Mike Waychison74b5c5b2011-06-03 13:04:53 -0700653 r = kvm_init_mmu_notifier(kvm);
654 if (r)
655 goto out_err;
656
Paolo Bonzini2f303b72013-09-25 13:53:07 +0200657 spin_lock(&kvm_lock);
Rusty Russell5e58cfe2007-07-23 17:08:21 +1000658 list_add(&kvm->vm_list, &vm_list);
Paolo Bonzini2f303b72013-09-25 13:53:07 +0200659 spin_unlock(&kvm_lock);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100660
Peter Zijlstra2ecd9d22015-07-03 18:53:58 +0200661 preempt_notifier_inc();
662
Avi Kivityf17abe92007-02-21 19:28:04 +0200663 return kvm;
Alexander Graf10474ae2009-09-15 11:37:46 +0200664
665out_err:
Christian Borntraeger719d93c2014-01-16 13:44:20 +0100666 cleanup_srcu_struct(&kvm->irq_srcu);
667out_err_no_irq_srcu:
Jan Kiszka57e7fbe2010-11-09 12:42:12 +0100668 cleanup_srcu_struct(&kvm->srcu);
Christian Borntraeger719d93c2014-01-16 13:44:20 +0100669out_err_no_srcu:
Alexander Graf10474ae2009-09-15 11:37:46 +0200670 hardware_disable_all();
Christian Borntraeger719d93c2014-01-16 13:44:20 +0100671out_err_no_disable:
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200672 for (i = 0; i < KVM_NR_BUSES; i++)
673 kfree(kvm->buses[i]);
Paolo Bonzinif481b062015-05-17 17:30:37 +0200674 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
675 kvm_free_memslots(kvm, kvm->memslots[i]);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100676 kvm_arch_free_vm(kvm);
Paolo Bonzinie9ad4ec2016-03-21 10:15:25 +0100677 mmdrop(current->mm);
Alexander Graf10474ae2009-09-15 11:37:46 +0200678 return ERR_PTR(r);
Avi Kivityf17abe92007-02-21 19:28:04 +0200679}
680
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900681/*
682 * Avoid using vmalloc for a small buffer.
683 * Should not be used when the size is statically known.
684 */
Takuya Yoshikawac1a7b322012-05-20 13:15:07 +0900685void *kvm_kvzalloc(unsigned long size)
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900686{
687 if (size > PAGE_SIZE)
688 return vzalloc(size);
689 else
690 return kzalloc(size, GFP_KERNEL);
691}
692
Scott Wood07f0a7b2013-04-25 14:11:23 +0000693static void kvm_destroy_devices(struct kvm *kvm)
694{
Geliang Tange6e3b5a2016-01-01 19:47:12 +0800695 struct kvm_device *dev, *tmp;
Scott Wood07f0a7b2013-04-25 14:11:23 +0000696
Geliang Tange6e3b5a2016-01-01 19:47:12 +0800697 list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
698 list_del(&dev->vm_node);
Scott Wood07f0a7b2013-04-25 14:11:23 +0000699 dev->ops->destroy(dev);
700 }
701}
702
Avi Kivityf17abe92007-02-21 19:28:04 +0200703static void kvm_destroy_vm(struct kvm *kvm)
704{
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200705 int i;
Avi Kivity6d4e4c42007-11-21 16:41:05 +0200706 struct mm_struct *mm = kvm->mm;
707
Janosch Frank536a6f82016-05-18 13:26:23 +0200708 kvm_destroy_vm_debugfs(kvm);
Sheng Yangad8ba2c2009-01-06 10:03:02 +0800709 kvm_arch_sync_events(kvm);
Paolo Bonzini2f303b72013-09-25 13:53:07 +0200710 spin_lock(&kvm_lock);
Avi Kivity133de902007-02-12 00:54:44 -0800711 list_del(&kvm->vm_list);
Paolo Bonzini2f303b72013-09-25 13:53:07 +0200712 spin_unlock(&kvm_lock);
Avi Kivity399ec802008-11-19 13:58:46 +0200713 kvm_free_irq_routing(kvm);
Marcelo Tosattie93f8a02009-12-23 14:35:24 -0200714 for (i = 0; i < KVM_NR_BUSES; i++)
715 kvm_io_bus_destroy(kvm->buses[i]);
Avi Kivity980da6c2009-12-20 15:13:43 +0200716 kvm_coalesced_mmio_free(kvm);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200717#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
718 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
Gleb Natapovf00be0c2009-03-19 12:20:36 +0200719#else
Marcelo Tosatti2df72e92012-08-24 15:54:57 -0300720 kvm_arch_flush_shadow_all(kvm);
Andrea Arcangelie930bff2008-07-25 16:24:52 +0200721#endif
Zhang Xiantaod19a9cd2007-11-18 18:43:45 +0800722 kvm_arch_destroy_vm(kvm);
Scott Wood07f0a7b2013-04-25 14:11:23 +0000723 kvm_destroy_devices(kvm);
Paolo Bonzinif481b062015-05-17 17:30:37 +0200724 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
725 kvm_free_memslots(kvm, kvm->memslots[i]);
Paolo Bonzini820b3fc2014-06-03 13:44:17 +0200726 cleanup_srcu_struct(&kvm->irq_srcu);
Jan Kiszkad89f5ef2010-11-09 17:02:49 +0100727 cleanup_srcu_struct(&kvm->srcu);
728 kvm_arch_free_vm(kvm);
Peter Zijlstra2ecd9d22015-07-03 18:53:58 +0200729 preempt_notifier_dec();
Alexander Graf10474ae2009-09-15 11:37:46 +0200730 hardware_disable_all();
Avi Kivity6d4e4c42007-11-21 16:41:05 +0200731 mmdrop(mm);
Avi Kivityf17abe92007-02-21 19:28:04 +0200732}
733
Izik Eidusd39f13b2008-03-30 16:01:25 +0300734void kvm_get_kvm(struct kvm *kvm)
735{
736 atomic_inc(&kvm->users_count);
737}
738EXPORT_SYMBOL_GPL(kvm_get_kvm);
739
740void kvm_put_kvm(struct kvm *kvm)
741{
742 if (atomic_dec_and_test(&kvm->users_count))
743 kvm_destroy_vm(kvm);
744}
745EXPORT_SYMBOL_GPL(kvm_put_kvm);
746
747
Avi Kivityf17abe92007-02-21 19:28:04 +0200748static int kvm_vm_release(struct inode *inode, struct file *filp)
749{
750 struct kvm *kvm = filp->private_data;
751
Gregory Haskins721eecb2009-05-20 10:30:49 -0400752 kvm_irqfd_release(kvm);
753
Izik Eidusd39f13b2008-03-30 16:01:25 +0300754 kvm_put_kvm(kvm);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800755 return 0;
756}
757
Takuya Yoshikawa515a0122010-10-27 18:23:54 +0900758/*
759 * Allocation size is twice as large as the actual dirty bitmap size.
Takuya Yoshikawa93474b22012-03-01 19:34:45 +0900760 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
Takuya Yoshikawa515a0122010-10-27 18:23:54 +0900761 */
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900762static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
763{
Takuya Yoshikawa515a0122010-10-27 18:23:54 +0900764 unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900765
Takuya Yoshikawa92eca8f2012-05-20 13:13:28 +0900766 memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900767 if (!memslot->dirty_bitmap)
768 return -ENOMEM;
769
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900770 return 0;
771}
772
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800773/*
Igor Mammedov0e60b072014-12-01 17:29:26 +0000774 * Insert memslot and re-sort memslots based on their GFN,
775 * so binary search could be used to lookup GFN.
776 * Sorting algorithm takes advantage of having initially
777 * sorted array and known changed memslot position.
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800778 */
Paolo Bonzini5cc15022014-11-14 10:55:31 +0100779static void update_memslots(struct kvm_memslots *slots,
780 struct kvm_memory_slot *new)
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800781{
Paolo Bonzini8593176c2014-11-14 10:22:07 +0100782 int id = new->id;
783 int i = slots->id_to_index[id];
Igor Mammedov063584d2014-11-13 23:00:13 +0000784 struct kvm_memory_slot *mslots = slots->memslots;
Xiao Guangrongf85e2cb2011-11-24 17:41:54 +0800785
Paolo Bonzini8593176c2014-11-14 10:22:07 +0100786 WARN_ON(mslots[i].id != id);
Igor Mammedov9c1a5d382014-12-01 17:29:27 +0000787 if (!new->npages) {
Paolo Bonzinidbaff302014-12-27 21:08:16 +0100788 WARN_ON(!mslots[i].npages);
Igor Mammedov9c1a5d382014-12-01 17:29:27 +0000789 if (mslots[i].npages)
790 slots->used_slots--;
791 } else {
792 if (!mslots[i].npages)
793 slots->used_slots++;
794 }
Igor Mammedov0e60b072014-12-01 17:29:26 +0000795
Igor Mammedov7f379cf2014-12-01 17:29:24 +0000796 while (i < KVM_MEM_SLOTS_NUM - 1 &&
Igor Mammedov0e60b072014-12-01 17:29:26 +0000797 new->base_gfn <= mslots[i + 1].base_gfn) {
798 if (!mslots[i + 1].npages)
799 break;
Igor Mammedov7f379cf2014-12-01 17:29:24 +0000800 mslots[i] = mslots[i + 1];
801 slots->id_to_index[mslots[i].id] = i;
802 i++;
803 }
Paolo Bonziniefbeec72014-12-27 18:01:00 +0100804
805 /*
806 * The ">=" is needed when creating a slot with base_gfn == 0,
807 * so that it moves before all those with base_gfn == npages == 0.
808 *
809 * On the other hand, if new->npages is zero, the above loop has
810 * already left i pointing to the beginning of the empty part of
811 * mslots, and the ">=" would move the hole backwards in this
812 * case---which is wrong. So skip the loop when deleting a slot.
813 */
814 if (new->npages) {
815 while (i > 0 &&
816 new->base_gfn >= mslots[i - 1].base_gfn) {
817 mslots[i] = mslots[i - 1];
818 slots->id_to_index[mslots[i].id] = i;
819 i--;
820 }
Paolo Bonzinidbaff302014-12-27 21:08:16 +0100821 } else
822 WARN_ON_ONCE(i != slots->used_slots);
Xiao Guangrongf85e2cb2011-11-24 17:41:54 +0800823
Paolo Bonzini8593176c2014-11-14 10:22:07 +0100824 mslots[i] = *new;
825 slots->id_to_index[mslots[i].id] = i;
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +0800826}
827
Paolo Bonzini09170a42015-05-18 13:59:39 +0200828static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
Xiao Guangronga50d64d2012-08-21 10:58:13 +0800829{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +0800830 u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
831
Christoffer Dall0f8a4de2014-08-26 14:00:37 +0200832#ifdef __KVM_HAVE_READONLY_MEM
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +0800833 valid_flags |= KVM_MEM_READONLY;
834#endif
835
836 if (mem->flags & ~valid_flags)
Xiao Guangronga50d64d2012-08-21 10:58:13 +0800837 return -EINVAL;
838
839 return 0;
840}
841
Gleb Natapov7ec4fb42012-12-24 17:49:30 +0200842static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
Paolo Bonzinif481b062015-05-17 17:30:37 +0200843 int as_id, struct kvm_memslots *slots)
Gleb Natapov7ec4fb42012-12-24 17:49:30 +0200844{
Paolo Bonzinif481b062015-05-17 17:30:37 +0200845 struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
Gleb Natapov7ec4fb42012-12-24 17:49:30 +0200846
David Matlackee3d1572014-08-18 15:46:06 -0700847 /*
848 * Set the low bit in the generation, which disables SPTE caching
849 * until the end of synchronize_srcu_expedited.
850 */
851 WARN_ON(old_memslots->generation & 1);
852 slots->generation = old_memslots->generation + 1;
853
Paolo Bonzinif481b062015-05-17 17:30:37 +0200854 rcu_assign_pointer(kvm->memslots[as_id], slots);
Gleb Natapov7ec4fb42012-12-24 17:49:30 +0200855 synchronize_srcu_expedited(&kvm->srcu);
Takuya Yoshikawae59dbe02013-07-04 13:40:29 +0900856
David Matlackee3d1572014-08-18 15:46:06 -0700857 /*
858 * Increment the new memslot generation a second time. This prevents
859 * vm exits that race with memslot updates from caching a memslot
860 * generation that will (potentially) be valid forever.
861 */
862 slots->generation++;
863
Paolo Bonzini15f46012015-05-17 21:26:08 +0200864 kvm_arch_memslots_updated(kvm, slots);
Takuya Yoshikawae59dbe02013-07-04 13:40:29 +0900865
866 return old_memslots;
Gleb Natapov7ec4fb42012-12-24 17:49:30 +0200867}
868
Avi Kivity6aa8b732006-12-10 02:21:36 -0800869/*
Avi Kivity6aa8b732006-12-10 02:21:36 -0800870 * Allocate some memory and give it an address in the guest physical address
871 * space.
872 *
873 * Discontiguous memory is allowed, mostly for framebuffers.
Sheng Yangf78e0e22007-10-29 09:40:42 +0800874 *
Dominik Dingel02d5d552014-10-27 16:22:56 +0100875 * Must be called holding kvm->slots_lock for write.
Avi Kivity6aa8b732006-12-10 02:21:36 -0800876 */
Sheng Yangf78e0e22007-10-29 09:40:42 +0800877int __kvm_set_memory_region(struct kvm *kvm,
Paolo Bonzini09170a42015-05-18 13:59:39 +0200878 const struct kvm_userspace_memory_region *mem)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800879{
Avi Kivity8234b222010-12-27 12:08:45 +0200880 int r;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800881 gfn_t base_gfn;
Heiko Carstens28bcb112009-09-03 17:35:35 +0200882 unsigned long npages;
Takuya Yoshikawaa843fac2013-01-11 18:27:43 +0900883 struct kvm_memory_slot *slot;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800884 struct kvm_memory_slot old, new;
Alex Williamsonb7f69c52012-12-10 10:33:03 -0700885 struct kvm_memslots *slots = NULL, *old_memslots;
Paolo Bonzinif481b062015-05-17 17:30:37 +0200886 int as_id, id;
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900887 enum kvm_mr_change change;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800888
Xiao Guangronga50d64d2012-08-21 10:58:13 +0800889 r = check_memory_region_flags(mem);
890 if (r)
891 goto out;
892
Avi Kivity6aa8b732006-12-10 02:21:36 -0800893 r = -EINVAL;
Paolo Bonzinif481b062015-05-17 17:30:37 +0200894 as_id = mem->slot >> 16;
895 id = (u16)mem->slot;
896
Avi Kivity6aa8b732006-12-10 02:21:36 -0800897 /* General sanity checks */
898 if (mem->memory_size & (PAGE_SIZE - 1))
899 goto out;
900 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
901 goto out;
Takuya Yoshikawafa3d3152011-05-07 16:35:38 +0900902 /* We can read the guest memory with __xxx_user() later on. */
Paolo Bonzinif481b062015-05-17 17:30:37 +0200903 if ((id < KVM_USER_MEM_SLOTS) &&
Takuya Yoshikawafa3d3152011-05-07 16:35:38 +0900904 ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
Heiko Carstens9e3bb6b2011-05-24 07:51:27 +0200905 !access_ok(VERIFY_WRITE,
906 (void __user *)(unsigned long)mem->userspace_addr,
907 mem->memory_size)))
Hollis Blanchard78749802008-11-07 13:32:12 -0600908 goto out;
Paolo Bonzinif481b062015-05-17 17:30:37 +0200909 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800910 goto out;
911 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
912 goto out;
913
Paolo Bonzinif481b062015-05-17 17:30:37 +0200914 slot = id_to_memslot(__kvm_memslots(kvm, as_id), id);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800915 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
916 npages = mem->memory_size >> PAGE_SHIFT;
917
Takuya Yoshikawa660c22c2010-04-13 22:47:24 +0900918 if (npages > KVM_MEM_MAX_NR_PAGES)
919 goto out;
920
Takuya Yoshikawaa843fac2013-01-11 18:27:43 +0900921 new = old = *slot;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800922
Paolo Bonzinif481b062015-05-17 17:30:37 +0200923 new.id = id;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800924 new.base_gfn = base_gfn;
925 new.npages = npages;
926 new.flags = mem->flags;
927
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900928 if (npages) {
929 if (!old.npages)
930 change = KVM_MR_CREATE;
931 else { /* Modify an existing slot. */
932 if ((mem->userspace_addr != old.userspace_addr) ||
Takuya Yoshikawa75d61fb2013-01-30 19:40:41 +0900933 (npages != old.npages) ||
934 ((new.flags ^ old.flags) & KVM_MEM_READONLY))
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900935 goto out;
936
937 if (base_gfn != old.base_gfn)
938 change = KVM_MR_MOVE;
939 else if (new.flags != old.flags)
940 change = KVM_MR_FLAGS_ONLY;
941 else { /* Nothing to change. */
942 r = 0;
943 goto out;
944 }
945 }
Paolo Bonzini09170a42015-05-18 13:59:39 +0200946 } else {
947 if (!old.npages)
948 goto out;
949
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900950 change = KVM_MR_DELETE;
Paolo Bonzini09170a42015-05-18 13:59:39 +0200951 new.base_gfn = 0;
952 new.flags = 0;
953 }
Avi Kivity6aa8b732006-12-10 02:21:36 -0800954
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900955 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
Takuya Yoshikawa0a706be2013-01-11 18:26:55 +0900956 /* Check for overlaps */
957 r = -EEXIST;
Paolo Bonzinif481b062015-05-17 17:30:37 +0200958 kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) {
Takuya Yoshikawaa843fac2013-01-11 18:27:43 +0900959 if ((slot->id >= KVM_USER_MEM_SLOTS) ||
Paolo Bonzinif481b062015-05-17 17:30:37 +0200960 (slot->id == id))
Takuya Yoshikawa0a706be2013-01-11 18:26:55 +0900961 continue;
962 if (!((base_gfn + npages <= slot->base_gfn) ||
963 (base_gfn >= slot->base_gfn + slot->npages)))
964 goto out;
965 }
Avi Kivity6aa8b732006-12-10 02:21:36 -0800966 }
Avi Kivity6aa8b732006-12-10 02:21:36 -0800967
Avi Kivity6aa8b732006-12-10 02:21:36 -0800968 /* Free page dirty bitmap if unneeded */
969 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
Al Viro8b6d44c2007-02-09 16:38:40 +0000970 new.dirty_bitmap = NULL;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800971
972 r = -ENOMEM;
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900973 if (change == KVM_MR_CREATE) {
Takuya Yoshikawa189a2f72012-02-08 13:01:09 +0900974 new.userspace_addr = mem->userspace_addr;
Takuya Yoshikawad89cc612012-08-01 18:03:28 +0900975
Aneesh Kumar K.V55870272013-10-07 22:18:00 +0530976 if (kvm_arch_create_memslot(kvm, &new, npages))
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +0900977 goto out_free;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800978 }
Joerg Roedelec04b262009-06-19 15:16:23 +0200979
Avi Kivity6aa8b732006-12-10 02:21:36 -0800980 /* Allocate page dirty bitmap if needed */
981 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
Takuya Yoshikawaa36a57b12010-10-27 18:22:19 +0900982 if (kvm_create_dirty_bitmap(&new) < 0)
Sheng Yangf78e0e22007-10-29 09:40:42 +0800983 goto out_free;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800984 }
985
Igor Mammedov74496132015-03-20 12:21:37 +0000986 slots = kvm_kvzalloc(sizeof(struct kvm_memslots));
Paolo Bonzinif2a81032014-11-14 10:46:45 +0100987 if (!slots)
988 goto out_free;
Paolo Bonzinif481b062015-05-17 17:30:37 +0200989 memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
Paolo Bonzinif2a81032014-11-14 10:46:45 +0100990
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +0900991 if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
Paolo Bonzinif481b062015-05-17 17:30:37 +0200992 slot = id_to_memslot(slots, id);
Xiao Guangrong28a37542011-11-24 19:04:35 +0800993 slot->flags |= KVM_MEMSLOT_INVALID;
994
Paolo Bonzinif481b062015-05-17 17:30:37 +0200995 old_memslots = install_new_memslots(kvm, as_id, slots);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -0200996
Alex Williamsone40f1932012-12-10 10:32:57 -0700997 /* slot was deleted or moved, clear iommu mapping */
998 kvm_iommu_unmap_pages(kvm, &old);
Marcelo Tosatti12d6e752012-08-24 15:54:58 -0300999 /* From this point no new shadow pages pointing to a deleted,
1000 * or moved, memslot will be created.
Marcelo Tosattibc6678a2009-12-23 14:35:21 -02001001 *
1002 * validation of sp->gfn happens in:
Xiubo Lib7d409d2015-02-26 14:58:24 +08001003 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1004 * - kvm_is_visible_gfn (mmu_check_roots)
Marcelo Tosattibc6678a2009-12-23 14:35:21 -02001005 */
Marcelo Tosatti2df72e92012-08-24 15:54:57 -03001006 kvm_arch_flush_shadow_memslot(kvm, slot);
Paolo Bonzinif2a81032014-11-14 10:46:45 +01001007
1008 /*
1009 * We can re-use the old_memslots from above, the only difference
1010 * from the currently installed memslots is the invalid flag. This
1011 * will get overwritten by update_memslots anyway.
1012 */
Alex Williamsonb7f69c52012-12-10 10:33:03 -07001013 slots = old_memslots;
Marcelo Tosattibc6678a2009-12-23 14:35:21 -02001014 }
Marcelo Tosatti34d4cb82008-07-10 20:49:31 -03001015
Takuya Yoshikawa7b6195a2013-02-27 19:44:34 +09001016 r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
Marcelo Tosattif7784b82009-12-23 14:35:18 -02001017 if (r)
Alex Williamsonb7f69c52012-12-10 10:33:03 -07001018 goto out_slots;
Marcelo Tosattif7784b82009-12-23 14:35:18 -02001019
Paolo Bonzinia47d2b02015-05-17 11:41:37 +02001020 /* actual memory is freed via old in kvm_free_memslot below */
Takuya Yoshikawaf64c0392013-01-29 11:00:07 +09001021 if (change == KVM_MR_DELETE) {
Marcelo Tosattibc6678a2009-12-23 14:35:21 -02001022 new.dirty_bitmap = NULL;
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +09001023 memset(&new.arch, 0, sizeof(new.arch));
Marcelo Tosattibc6678a2009-12-23 14:35:21 -02001024 }
1025
Paolo Bonzini5cc15022014-11-14 10:55:31 +01001026 update_memslots(slots, &new);
Paolo Bonzinif481b062015-05-17 17:30:37 +02001027 old_memslots = install_new_memslots(kvm, as_id, slots);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001028
Paolo Bonzinif36f3f22015-05-18 13:20:23 +02001029 kvm_arch_commit_memory_region(kvm, mem, &old, &new, change);
Zhang Xiantao3ad82a72007-11-20 13:11:38 +08001030
Paolo Bonzinia47d2b02015-05-17 11:41:37 +02001031 kvm_free_memslot(kvm, &old, &new);
Igor Mammedov74496132015-03-20 12:21:37 +00001032 kvfree(old_memslots);
Marcelo Tosattibc6678a2009-12-23 14:35:21 -02001033
Marcelo Tosattibc6678a2009-12-23 14:35:21 -02001034 /*
1035 * IOMMU mapping: New slots need to be mapped. Old slots need to be
Avi Kivity6aa8b732006-12-10 02:21:36 -08001036 * un-mapped and re-mapped if their base changes. Since base change
1037 * unmapping is handled above with slot deletion, mapping alone is
Sheng Yangf78e0e22007-10-29 09:40:42 +08001038 * needed here. Anything else the iommu might care about for existing
Avi Kivity6aa8b732006-12-10 02:21:36 -08001039 * slots (size changes, userspace addr changes and read-only flag
1040 * changes) is disallowed above, so any other attribute changes getting
1041 * here can be skipped.
Izik Eidus210c7c42007-10-24 23:52:57 +02001042 */
1043 if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
Sheng Yangf78e0e22007-10-29 09:40:42 +08001044 r = kvm_iommu_map_pages(kvm, &new);
Yang Zhange0230e12013-10-24 09:56:39 +08001045 return r;
Sheng Yangf78e0e22007-10-29 09:40:42 +08001046 }
1047
Avi Kivity6aa8b732006-12-10 02:21:36 -08001048 return 0;
1049
1050out_slots:
Igor Mammedov74496132015-03-20 12:21:37 +00001051 kvfree(slots);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001052out_free:
Paolo Bonzinia47d2b02015-05-17 11:41:37 +02001053 kvm_free_memslot(kvm, &new, &old);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001054out:
1055 return r;
Izik Eidus210c7c42007-10-24 23:52:57 +02001056}
1057EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
Sheng Yangf78e0e22007-10-29 09:40:42 +08001058
1059int kvm_set_memory_region(struct kvm *kvm,
Paolo Bonzini09170a42015-05-18 13:59:39 +02001060 const struct kvm_userspace_memory_region *mem)
Sheng Yangf78e0e22007-10-29 09:40:42 +08001061{
1062 int r;
1063
Marcelo Tosatti79fac952009-12-23 14:35:26 -02001064 mutex_lock(&kvm->slots_lock);
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +09001065 r = __kvm_set_memory_region(kvm, mem);
Marcelo Tosatti79fac952009-12-23 14:35:26 -02001066 mutex_unlock(&kvm->slots_lock);
Sheng Yangf78e0e22007-10-29 09:40:42 +08001067 return r;
1068}
Izik Eidus210c7c42007-10-24 23:52:57 +02001069EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1070
Stephen Hemminger79408762013-12-29 12:12:29 -08001071static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1072 struct kvm_userspace_memory_region *mem)
Izik Eidus210c7c42007-10-24 23:52:57 +02001073{
Paolo Bonzinif481b062015-05-17 17:30:37 +02001074 if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
Izik Eiduse0d62c72007-10-24 23:57:46 +02001075 return -EINVAL;
Paolo Bonzini09170a42015-05-18 13:59:39 +02001076
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +09001077 return kvm_set_memory_region(kvm, mem);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001078}
1079
Zhang Xiantao5bb064d2007-11-18 20:29:43 +08001080int kvm_get_dirty_log(struct kvm *kvm,
1081 struct kvm_dirty_log *log, int *is_dirty)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001082{
Paolo Bonzini9f6b8022015-05-17 16:20:07 +02001083 struct kvm_memslots *slots;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001084 struct kvm_memory_slot *memslot;
Paolo Bonzinif481b062015-05-17 17:30:37 +02001085 int r, i, as_id, id;
Takuya Yoshikawa87bf6e72010-04-12 19:35:35 +09001086 unsigned long n;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001087 unsigned long any = 0;
1088
Avi Kivity6aa8b732006-12-10 02:21:36 -08001089 r = -EINVAL;
Paolo Bonzinif481b062015-05-17 17:30:37 +02001090 as_id = log->slot >> 16;
1091 id = (u16)log->slot;
1092 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001093 goto out;
1094
Paolo Bonzinif481b062015-05-17 17:30:37 +02001095 slots = __kvm_memslots(kvm, as_id);
1096 memslot = id_to_memslot(slots, id);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001097 r = -ENOENT;
1098 if (!memslot->dirty_bitmap)
1099 goto out;
1100
Takuya Yoshikawa87bf6e72010-04-12 19:35:35 +09001101 n = kvm_dirty_bitmap_bytes(memslot);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001102
Uri Lublincd1a4a92007-02-22 16:43:09 +02001103 for (i = 0; !any && i < n/sizeof(long); ++i)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001104 any = memslot->dirty_bitmap[i];
1105
1106 r = -EFAULT;
1107 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1108 goto out;
1109
Zhang Xiantao5bb064d2007-11-18 20:29:43 +08001110 if (any)
1111 *is_dirty = 1;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001112
1113 r = 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001114out:
Avi Kivity6aa8b732006-12-10 02:21:36 -08001115 return r;
1116}
Aneesh Kumar K.V2ba9f0d2013-10-07 22:17:59 +05301117EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001118
Mario Smarduchba0513b2015-01-15 15:58:53 -08001119#ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1120/**
1121 * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1122 * are dirty write protect them for next write.
1123 * @kvm: pointer to kvm instance
1124 * @log: slot id and address to which we copy the log
1125 * @is_dirty: flag set if any page is dirty
1126 *
1127 * We need to keep it in mind that VCPU threads can write to the bitmap
1128 * concurrently. So, to avoid losing track of dirty pages we keep the
1129 * following order:
1130 *
1131 * 1. Take a snapshot of the bit and clear it if needed.
1132 * 2. Write protect the corresponding page.
1133 * 3. Copy the snapshot to the userspace.
1134 * 4. Upon return caller flushes TLB's if needed.
1135 *
1136 * Between 2 and 4, the guest may write to the page using the remaining TLB
1137 * entry. This is not a problem because the page is reported dirty using
1138 * the snapshot taken before and step 4 ensures that writes done after
1139 * exiting to userspace will be logged for the next call.
1140 *
1141 */
1142int kvm_get_dirty_log_protect(struct kvm *kvm,
1143 struct kvm_dirty_log *log, bool *is_dirty)
1144{
Paolo Bonzini9f6b8022015-05-17 16:20:07 +02001145 struct kvm_memslots *slots;
Mario Smarduchba0513b2015-01-15 15:58:53 -08001146 struct kvm_memory_slot *memslot;
Paolo Bonzinif481b062015-05-17 17:30:37 +02001147 int r, i, as_id, id;
Mario Smarduchba0513b2015-01-15 15:58:53 -08001148 unsigned long n;
1149 unsigned long *dirty_bitmap;
1150 unsigned long *dirty_bitmap_buffer;
1151
1152 r = -EINVAL;
Paolo Bonzinif481b062015-05-17 17:30:37 +02001153 as_id = log->slot >> 16;
1154 id = (u16)log->slot;
1155 if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
Mario Smarduchba0513b2015-01-15 15:58:53 -08001156 goto out;
1157
Paolo Bonzinif481b062015-05-17 17:30:37 +02001158 slots = __kvm_memslots(kvm, as_id);
1159 memslot = id_to_memslot(slots, id);
Mario Smarduchba0513b2015-01-15 15:58:53 -08001160
1161 dirty_bitmap = memslot->dirty_bitmap;
1162 r = -ENOENT;
1163 if (!dirty_bitmap)
1164 goto out;
1165
1166 n = kvm_dirty_bitmap_bytes(memslot);
1167
1168 dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
1169 memset(dirty_bitmap_buffer, 0, n);
1170
1171 spin_lock(&kvm->mmu_lock);
1172 *is_dirty = false;
1173 for (i = 0; i < n / sizeof(long); i++) {
1174 unsigned long mask;
1175 gfn_t offset;
1176
1177 if (!dirty_bitmap[i])
1178 continue;
1179
1180 *is_dirty = true;
1181
1182 mask = xchg(&dirty_bitmap[i], 0);
1183 dirty_bitmap_buffer[i] = mask;
1184
Takuya Yoshikawa58d29302015-03-17 16:19:58 +09001185 if (mask) {
1186 offset = i * BITS_PER_LONG;
1187 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1188 offset, mask);
1189 }
Mario Smarduchba0513b2015-01-15 15:58:53 -08001190 }
1191
1192 spin_unlock(&kvm->mmu_lock);
1193
1194 r = -EFAULT;
1195 if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1196 goto out;
1197
1198 r = 0;
1199out:
1200 return r;
1201}
1202EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1203#endif
1204
Takuya Yoshikawadb3fe4e2012-02-08 13:02:18 +09001205bool kvm_largepages_enabled(void)
1206{
1207 return largepages_enabled;
1208}
1209
Marcelo Tosatti54dee992009-06-11 12:07:44 -03001210void kvm_disable_largepages(void)
1211{
1212 largepages_enabled = false;
1213}
1214EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1215
Gleb Natapov49c77542010-10-18 15:22:23 +02001216struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1217{
1218 return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1219}
Avi Kivitya1f4d3952010-06-21 11:44:20 +03001220EXPORT_SYMBOL_GPL(gfn_to_memslot);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001221
Paolo Bonzini8e734852015-05-17 13:58:53 +02001222struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1223{
1224 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1225}
1226
Yaowei Bai33e94152015-11-14 11:21:06 +08001227bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
Izik Eiduse0d62c72007-10-24 23:57:46 +02001228{
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +08001229 struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
Izik Eiduse0d62c72007-10-24 23:57:46 +02001230
Alex Williamsonbbacc0c2012-12-10 10:33:09 -07001231 if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
Xiao Guangrongbf3e05b2011-11-24 17:40:57 +08001232 memslot->flags & KVM_MEMSLOT_INVALID)
Yaowei Bai33e94152015-11-14 11:21:06 +08001233 return false;
Izik Eiduse0d62c72007-10-24 23:57:46 +02001234
Yaowei Bai33e94152015-11-14 11:21:06 +08001235 return true;
Izik Eiduse0d62c72007-10-24 23:57:46 +02001236}
1237EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1238
Joerg Roedel8f0b1ab2010-01-28 12:37:56 +01001239unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1240{
1241 struct vm_area_struct *vma;
1242 unsigned long addr, size;
1243
1244 size = PAGE_SIZE;
1245
1246 addr = gfn_to_hva(kvm, gfn);
1247 if (kvm_is_error_hva(addr))
1248 return PAGE_SIZE;
1249
1250 down_read(&current->mm->mmap_sem);
1251 vma = find_vma(current->mm, addr);
1252 if (!vma)
1253 goto out;
1254
1255 size = vma_kernel_pagesize(vma);
1256
1257out:
1258 up_read(&current->mm->mmap_sem);
1259
1260 return size;
1261}
1262
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001263static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1264{
1265 return slot->flags & KVM_MEM_READONLY;
1266}
1267
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001268static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1269 gfn_t *nr_pages, bool write)
Izik Eidus539cb662007-11-11 22:05:04 +02001270{
Marcelo Tosattibc6678a2009-12-23 14:35:21 -02001271 if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
Xiao Guangrongca3a4902012-08-21 11:01:50 +08001272 return KVM_HVA_ERR_BAD;
Xiao Guangrong48987782010-08-22 19:11:43 +08001273
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001274 if (memslot_is_readonly(slot) && write)
1275 return KVM_HVA_ERR_RO_BAD;
Xiao Guangrong48987782010-08-22 19:11:43 +08001276
1277 if (nr_pages)
1278 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1279
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001280 return __gfn_to_hva_memslot(slot, gfn);
Izik Eidus539cb662007-11-11 22:05:04 +02001281}
Xiao Guangrong48987782010-08-22 19:11:43 +08001282
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001283static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1284 gfn_t *nr_pages)
1285{
1286 return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1287}
1288
1289unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
Stephen Hemminger79408762013-12-29 12:12:29 -08001290 gfn_t gfn)
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001291{
1292 return gfn_to_hva_many(slot, gfn, NULL);
1293}
1294EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1295
Xiao Guangrong48987782010-08-22 19:11:43 +08001296unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1297{
Gleb Natapov49c77542010-10-18 15:22:23 +02001298 return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
Xiao Guangrong48987782010-08-22 19:11:43 +08001299}
Sheng Yang0d150292008-04-25 21:44:50 +08001300EXPORT_SYMBOL_GPL(gfn_to_hva);
Izik Eidus539cb662007-11-11 22:05:04 +02001301
Paolo Bonzini8e734852015-05-17 13:58:53 +02001302unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1303{
1304 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1305}
1306EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1307
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001308/*
Paolo Bonziniba6a3542013-09-09 13:52:33 +02001309 * If writable is set to false, the hva returned by this function is only
1310 * allowed to be read.
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001311 */
Christoffer Dall64d83122014-08-19 12:15:00 +02001312unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1313 gfn_t gfn, bool *writable)
Gleb Natapov80300892010-10-19 18:13:41 +02001314{
Gleb Natapova2ac07f2013-10-01 19:58:36 +03001315 unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1316
1317 if (!kvm_is_error_hva(hva) && writable)
Paolo Bonziniba6a3542013-09-09 13:52:33 +02001318 *writable = !memslot_is_readonly(slot);
1319
Gleb Natapova2ac07f2013-10-01 19:58:36 +03001320 return hva;
Xiao Guangrong86ab8cf2012-08-21 10:59:53 +08001321}
1322
Christoffer Dall64d83122014-08-19 12:15:00 +02001323unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1324{
1325 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1326
1327 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1328}
1329
Paolo Bonzini8e734852015-05-17 13:58:53 +02001330unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1331{
1332 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1333
1334 return gfn_to_hva_memslot_prot(slot, gfn, writable);
1335}
1336
Dave Hansend4edcf02016-02-12 13:01:56 -08001337static int get_user_page_nowait(unsigned long start, int write,
1338 struct page **page)
Gleb Natapov0857b9e2011-02-01 13:21:47 +02001339{
1340 int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1341
1342 if (write)
1343 flags |= FOLL_WRITE;
1344
Dave Hansend4edcf02016-02-12 13:01:56 -08001345 return __get_user_pages(current, current->mm, start, 1, flags, page,
1346 NULL, NULL);
Gleb Natapov0857b9e2011-02-01 13:21:47 +02001347}
1348
Huang Yingfafc3db2011-01-30 11:15:49 +08001349static inline int check_user_page_hwpoison(unsigned long addr)
1350{
1351 int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1352
1353 rc = __get_user_pages(current, current->mm, addr, 1,
1354 flags, NULL, NULL, NULL);
1355 return rc == -EHWPOISON;
1356}
1357
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001358/*
1359 * The atomic path to get the writable pfn which will be stored in @pfn,
1360 * true indicates success, otherwise false is returned.
1361 */
1362static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
Dan Williamsba049e92016-01-15 16:56:11 -08001363 bool write_fault, bool *writable, kvm_pfn_t *pfn)
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001364{
1365 struct page *page[1];
1366 int npages;
1367
1368 if (!(async || atomic))
1369 return false;
1370
Xiao Guangrong12ce13f2012-08-21 11:00:49 +08001371 /*
1372 * Fast pin a writable pfn only if it is a write fault request
1373 * or the caller allows to map a writable pfn for a read fault
1374 * request.
1375 */
1376 if (!(write_fault || writable))
1377 return false;
1378
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001379 npages = __get_user_pages_fast(addr, 1, 1, page);
1380 if (npages == 1) {
1381 *pfn = page_to_pfn(page[0]);
1382
1383 if (writable)
1384 *writable = true;
1385 return true;
1386 }
1387
1388 return false;
1389}
1390
1391/*
1392 * The slow path to get the pfn of the specified host virtual address,
1393 * 1 indicates success, -errno is returned if error is detected.
1394 */
1395static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
Dan Williamsba049e92016-01-15 16:56:11 -08001396 bool *writable, kvm_pfn_t *pfn)
Avi Kivity954bbbc2007-03-30 14:02:32 +03001397{
Anthony Liguori8d4e1282007-10-18 09:59:34 -05001398 struct page *page[1];
Gleb Natapovaf585b92010-10-14 11:22:46 +02001399 int npages = 0;
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001400
1401 might_sleep();
1402
1403 if (writable)
1404 *writable = write_fault;
1405
1406 if (async) {
1407 down_read(&current->mm->mmap_sem);
Dave Hansend4edcf02016-02-12 13:01:56 -08001408 npages = get_user_page_nowait(addr, write_fault, page);
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001409 up_read(&current->mm->mmap_sem);
Andrea Arcangeli0664e572015-02-11 15:27:28 -08001410 } else
1411 npages = __get_user_pages_unlocked(current, current->mm, addr, 1,
1412 write_fault, 0, page,
1413 FOLL_TOUCH|FOLL_HWPOISON);
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001414 if (npages != 1)
1415 return npages;
1416
1417 /* map read fault as writable if possible */
Xiao Guangrong12ce13f2012-08-21 11:00:49 +08001418 if (unlikely(!write_fault) && writable) {
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001419 struct page *wpage[1];
1420
1421 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1422 if (npages == 1) {
1423 *writable = true;
1424 put_page(page[0]);
1425 page[0] = wpage[0];
1426 }
1427
1428 npages = 1;
1429 }
1430 *pfn = page_to_pfn(page[0]);
1431 return npages;
1432}
1433
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001434static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1435{
1436 if (unlikely(!(vma->vm_flags & VM_READ)))
1437 return false;
1438
1439 if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1440 return false;
1441
1442 return true;
1443}
1444
Xiao Guangrong12ce13f2012-08-21 11:00:49 +08001445/*
1446 * Pin guest page in memory and return its pfn.
1447 * @addr: host virtual address which maps memory to the guest
1448 * @atomic: whether this function can sleep
1449 * @async: whether this function need to wait IO complete if the
1450 * host page is not in the memory
1451 * @write_fault: whether we should get a writable host page
1452 * @writable: whether it allows to map a writable host page for !@write_fault
1453 *
1454 * The function will map a writable host page for these two cases:
1455 * 1): @write_fault = true
1456 * 2): @write_fault = false && @writable, @writable will tell the caller
1457 * whether the mapping is writable.
1458 */
Dan Williamsba049e92016-01-15 16:56:11 -08001459static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001460 bool write_fault, bool *writable)
1461{
1462 struct vm_area_struct *vma;
Dan Williamsba049e92016-01-15 16:56:11 -08001463 kvm_pfn_t pfn = 0;
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001464 int npages;
Avi Kivity954bbbc2007-03-30 14:02:32 +03001465
Gleb Natapovaf585b92010-10-14 11:22:46 +02001466 /* we can do it either atomically or asynchronously, not both */
1467 BUG_ON(atomic && async);
1468
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001469 if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1470 return pfn;
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001471
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001472 if (atomic)
1473 return KVM_PFN_ERR_FAULT;
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001474
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001475 npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1476 if (npages == 1)
1477 return pfn;
Gleb Natapovaf585b92010-10-14 11:22:46 +02001478
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001479 down_read(&current->mm->mmap_sem);
1480 if (npages == -EHWPOISON ||
1481 (!async && check_user_page_hwpoison(addr))) {
1482 pfn = KVM_PFN_ERR_HWPOISON;
1483 goto exit;
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001484 }
Izik Eidus539cb662007-11-11 22:05:04 +02001485
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001486 vma = find_vma_intersection(current->mm, addr, addr + 1);
Anthony Liguori8d4e1282007-10-18 09:59:34 -05001487
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001488 if (vma == NULL)
1489 pfn = KVM_PFN_ERR_FAULT;
1490 else if ((vma->vm_flags & VM_PFNMAP)) {
1491 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1492 vma->vm_pgoff;
Ard Biesheuvelbf4bea82014-11-10 08:33:56 +00001493 BUG_ON(!kvm_is_reserved_pfn(pfn));
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001494 } else {
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001495 if (async && vma_is_valid(vma, write_fault))
Xiao Guangrong2fc84312012-08-21 11:00:22 +08001496 *async = true;
1497 pfn = KVM_PFN_ERR_FAULT;
1498 }
1499exit:
1500 up_read(&current->mm->mmap_sem);
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001501 return pfn;
Anthony Liguori35149e22008-04-02 14:46:56 -05001502}
1503
Dan Williamsba049e92016-01-15 16:56:11 -08001504kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
1505 bool atomic, bool *async, bool write_fault,
1506 bool *writable)
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001507{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001508 unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1509
Paolo Bonzinib2740d32016-02-23 15:36:01 +01001510 if (addr == KVM_HVA_ERR_RO_BAD) {
1511 if (writable)
1512 *writable = false;
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001513 return KVM_PFN_ERR_RO_FAULT;
Paolo Bonzinib2740d32016-02-23 15:36:01 +01001514 }
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001515
Paolo Bonzinib2740d32016-02-23 15:36:01 +01001516 if (kvm_is_error_hva(addr)) {
1517 if (writable)
1518 *writable = false;
Xiao Guangrong81c52c52012-10-16 20:10:59 +08001519 return KVM_PFN_NOSLOT;
Paolo Bonzinib2740d32016-02-23 15:36:01 +01001520 }
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001521
1522 /* Do not map writable pfn in the readonly memslot. */
1523 if (writable && memslot_is_readonly(slot)) {
1524 *writable = false;
1525 writable = NULL;
1526 }
1527
1528 return hva_to_pfn(addr, atomic, async, write_fault,
1529 writable);
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001530}
Paolo Bonzini35204692015-04-02 11:20:48 +02001531EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
Xiao Guangrong887c08a2010-08-22 19:10:28 +08001532
Dan Williamsba049e92016-01-15 16:56:11 -08001533kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001534 bool *writable)
1535{
Paolo Bonzinie37afc62015-05-19 16:09:04 +02001536 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1537 write_fault, writable);
Marcelo Tosatti612819c2010-10-22 14:18:18 -02001538}
1539EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1540
Dan Williamsba049e92016-01-15 16:56:11 -08001541kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001542{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001543 return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001544}
Paolo Bonzinie37afc62015-05-19 16:09:04 +02001545EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
Marcelo Tosatti506f0d62009-12-23 14:35:19 -02001546
Dan Williamsba049e92016-01-15 16:56:11 -08001547kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
Xiao Guangrong037d92d2012-08-21 10:59:12 +08001548{
Xiao Guangrong4d8b81a2012-08-21 11:02:51 +08001549 return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
Xiao Guangrong037d92d2012-08-21 10:59:12 +08001550}
1551EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1552
Dan Williamsba049e92016-01-15 16:56:11 -08001553kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
Paolo Bonzinie37afc62015-05-19 16:09:04 +02001554{
1555 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn);
1556}
1557EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1558
Dan Williamsba049e92016-01-15 16:56:11 -08001559kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
Paolo Bonzini8e734852015-05-17 13:58:53 +02001560{
1561 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1562}
1563EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1564
Dan Williamsba049e92016-01-15 16:56:11 -08001565kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
Paolo Bonzinie37afc62015-05-19 16:09:04 +02001566{
1567 return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1568}
1569EXPORT_SYMBOL_GPL(gfn_to_pfn);
1570
Dan Williamsba049e92016-01-15 16:56:11 -08001571kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
Paolo Bonzini8e734852015-05-17 13:58:53 +02001572{
1573 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1574}
1575EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1576
Paolo Bonzinid9ef13c2015-05-19 16:01:50 +02001577int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1578 struct page **pages, int nr_pages)
Xiao Guangrong48987782010-08-22 19:11:43 +08001579{
1580 unsigned long addr;
1581 gfn_t entry;
1582
Paolo Bonzinid9ef13c2015-05-19 16:01:50 +02001583 addr = gfn_to_hva_many(slot, gfn, &entry);
Xiao Guangrong48987782010-08-22 19:11:43 +08001584 if (kvm_is_error_hva(addr))
1585 return -1;
1586
1587 if (entry < nr_pages)
1588 return 0;
1589
1590 return __get_user_pages_fast(addr, nr_pages, 1, pages);
1591}
1592EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1593
Dan Williamsba049e92016-01-15 16:56:11 -08001594static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
Xiao Guangronga2766322012-07-26 11:58:59 +08001595{
Xiao Guangrong81c52c52012-10-16 20:10:59 +08001596 if (is_error_noslot_pfn(pfn))
Xiao Guangrong6cede2e2012-08-03 15:41:22 +08001597 return KVM_ERR_PTR_BAD_PAGE;
Xiao Guangronga2766322012-07-26 11:58:59 +08001598
Ard Biesheuvelbf4bea82014-11-10 08:33:56 +00001599 if (kvm_is_reserved_pfn(pfn)) {
Xiao Guangrongcb9aaa32012-08-03 15:42:10 +08001600 WARN_ON(1);
1601 return KVM_ERR_PTR_BAD_PAGE;
1602 }
1603
Xiao Guangronga2766322012-07-26 11:58:59 +08001604 return pfn_to_page(pfn);
1605}
1606
Anthony Liguori35149e22008-04-02 14:46:56 -05001607struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1608{
Dan Williamsba049e92016-01-15 16:56:11 -08001609 kvm_pfn_t pfn;
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001610
1611 pfn = gfn_to_pfn(kvm, gfn);
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001612
Xiao Guangronga2766322012-07-26 11:58:59 +08001613 return kvm_pfn_to_page(pfn);
Avi Kivity954bbbc2007-03-30 14:02:32 +03001614}
1615EXPORT_SYMBOL_GPL(gfn_to_page);
1616
Paolo Bonzini8e734852015-05-17 13:58:53 +02001617struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1618{
Dan Williamsba049e92016-01-15 16:56:11 -08001619 kvm_pfn_t pfn;
Paolo Bonzini8e734852015-05-17 13:58:53 +02001620
1621 pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
1622
1623 return kvm_pfn_to_page(pfn);
1624}
1625EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
1626
Izik Eidusb4231d62007-11-20 11:49:33 +02001627void kvm_release_page_clean(struct page *page)
1628{
Xiao Guangrong32cad842012-08-03 15:42:52 +08001629 WARN_ON(is_error_page(page));
1630
Anthony Liguori35149e22008-04-02 14:46:56 -05001631 kvm_release_pfn_clean(page_to_pfn(page));
Izik Eidusb4231d62007-11-20 11:49:33 +02001632}
1633EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1634
Dan Williamsba049e92016-01-15 16:56:11 -08001635void kvm_release_pfn_clean(kvm_pfn_t pfn)
Anthony Liguori35149e22008-04-02 14:46:56 -05001636{
Ard Biesheuvelbf4bea82014-11-10 08:33:56 +00001637 if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001638 put_page(pfn_to_page(pfn));
Anthony Liguori35149e22008-04-02 14:46:56 -05001639}
1640EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1641
Izik Eidusb4231d62007-11-20 11:49:33 +02001642void kvm_release_page_dirty(struct page *page)
Izik Eidus8a7ae052007-10-18 11:09:33 +02001643{
Xiao Guangronga2766322012-07-26 11:58:59 +08001644 WARN_ON(is_error_page(page));
1645
Anthony Liguori35149e22008-04-02 14:46:56 -05001646 kvm_release_pfn_dirty(page_to_pfn(page));
Izik Eidus8a7ae052007-10-18 11:09:33 +02001647}
Izik Eidusb4231d62007-11-20 11:49:33 +02001648EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
Izik Eidus8a7ae052007-10-18 11:09:33 +02001649
Dan Williamsba049e92016-01-15 16:56:11 -08001650static void kvm_release_pfn_dirty(kvm_pfn_t pfn)
Anthony Liguori35149e22008-04-02 14:46:56 -05001651{
1652 kvm_set_pfn_dirty(pfn);
1653 kvm_release_pfn_clean(pfn);
1654}
Anthony Liguori35149e22008-04-02 14:46:56 -05001655
Dan Williamsba049e92016-01-15 16:56:11 -08001656void kvm_set_pfn_dirty(kvm_pfn_t pfn)
Anthony Liguori35149e22008-04-02 14:46:56 -05001657{
Ard Biesheuvelbf4bea82014-11-10 08:33:56 +00001658 if (!kvm_is_reserved_pfn(pfn)) {
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001659 struct page *page = pfn_to_page(pfn);
Xiubo Lif95ef0c2015-02-26 14:58:23 +08001660
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001661 if (!PageReserved(page))
1662 SetPageDirty(page);
1663 }
Anthony Liguori35149e22008-04-02 14:46:56 -05001664}
1665EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1666
Dan Williamsba049e92016-01-15 16:56:11 -08001667void kvm_set_pfn_accessed(kvm_pfn_t pfn)
Anthony Liguori35149e22008-04-02 14:46:56 -05001668{
Ard Biesheuvelbf4bea82014-11-10 08:33:56 +00001669 if (!kvm_is_reserved_pfn(pfn))
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001670 mark_page_accessed(pfn_to_page(pfn));
Anthony Liguori35149e22008-04-02 14:46:56 -05001671}
1672EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1673
Dan Williamsba049e92016-01-15 16:56:11 -08001674void kvm_get_pfn(kvm_pfn_t pfn)
Anthony Liguori35149e22008-04-02 14:46:56 -05001675{
Ard Biesheuvelbf4bea82014-11-10 08:33:56 +00001676 if (!kvm_is_reserved_pfn(pfn))
Anthony Liguori2e2e3732008-04-30 15:37:07 -05001677 get_page(pfn_to_page(pfn));
Anthony Liguori35149e22008-04-02 14:46:56 -05001678}
1679EXPORT_SYMBOL_GPL(kvm_get_pfn);
1680
Izik Eidus195aefd2007-10-01 22:14:18 +02001681static int next_segment(unsigned long len, int offset)
1682{
1683 if (len > PAGE_SIZE - offset)
1684 return PAGE_SIZE - offset;
1685 else
1686 return len;
1687}
1688
Paolo Bonzini8e734852015-05-17 13:58:53 +02001689static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
1690 void *data, int offset, int len)
Izik Eidus195aefd2007-10-01 22:14:18 +02001691{
Izik Eiduse0506bc2007-11-11 22:10:22 +02001692 int r;
1693 unsigned long addr;
Izik Eidus195aefd2007-10-01 22:14:18 +02001694
Paolo Bonzini8e734852015-05-17 13:58:53 +02001695 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
Izik Eiduse0506bc2007-11-11 22:10:22 +02001696 if (kvm_is_error_hva(addr))
Izik Eidus195aefd2007-10-01 22:14:18 +02001697 return -EFAULT;
Paolo Bonzini3180a7f2015-04-02 14:08:20 +02001698 r = __copy_from_user(data, (void __user *)addr + offset, len);
Izik Eiduse0506bc2007-11-11 22:10:22 +02001699 if (r)
1700 return -EFAULT;
Izik Eidus195aefd2007-10-01 22:14:18 +02001701 return 0;
1702}
Paolo Bonzini8e734852015-05-17 13:58:53 +02001703
1704int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1705 int len)
1706{
1707 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1708
1709 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1710}
Izik Eidus195aefd2007-10-01 22:14:18 +02001711EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1712
Paolo Bonzini8e734852015-05-17 13:58:53 +02001713int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
1714 int offset, int len)
1715{
1716 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1717
1718 return __kvm_read_guest_page(slot, gfn, data, offset, len);
1719}
1720EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
1721
Izik Eidus195aefd2007-10-01 22:14:18 +02001722int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1723{
1724 gfn_t gfn = gpa >> PAGE_SHIFT;
1725 int seg;
1726 int offset = offset_in_page(gpa);
1727 int ret;
1728
1729 while ((seg = next_segment(len, offset)) != 0) {
1730 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1731 if (ret < 0)
1732 return ret;
1733 offset = 0;
1734 len -= seg;
1735 data += seg;
1736 ++gfn;
1737 }
1738 return 0;
1739}
1740EXPORT_SYMBOL_GPL(kvm_read_guest);
1741
Paolo Bonzini8e734852015-05-17 13:58:53 +02001742int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
1743{
1744 gfn_t gfn = gpa >> PAGE_SHIFT;
1745 int seg;
1746 int offset = offset_in_page(gpa);
1747 int ret;
1748
1749 while ((seg = next_segment(len, offset)) != 0) {
1750 ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
1751 if (ret < 0)
1752 return ret;
1753 offset = 0;
1754 len -= seg;
1755 data += seg;
1756 ++gfn;
1757 }
1758 return 0;
1759}
1760EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
1761
1762static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1763 void *data, int offset, unsigned long len)
Marcelo Tosatti7ec54582007-12-20 19:18:23 -05001764{
1765 int r;
1766 unsigned long addr;
Marcelo Tosatti7ec54582007-12-20 19:18:23 -05001767
Paolo Bonzini8e734852015-05-17 13:58:53 +02001768 addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
Marcelo Tosatti7ec54582007-12-20 19:18:23 -05001769 if (kvm_is_error_hva(addr))
1770 return -EFAULT;
Andrea Arcangeli0aac03f2008-01-30 19:57:35 +01001771 pagefault_disable();
Paolo Bonzini3180a7f2015-04-02 14:08:20 +02001772 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
Andrea Arcangeli0aac03f2008-01-30 19:57:35 +01001773 pagefault_enable();
Marcelo Tosatti7ec54582007-12-20 19:18:23 -05001774 if (r)
1775 return -EFAULT;
1776 return 0;
1777}
Marcelo Tosatti7ec54582007-12-20 19:18:23 -05001778
Paolo Bonzini8e734852015-05-17 13:58:53 +02001779int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1780 unsigned long len)
1781{
1782 gfn_t gfn = gpa >> PAGE_SHIFT;
1783 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1784 int offset = offset_in_page(gpa);
1785
1786 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1787}
1788EXPORT_SYMBOL_GPL(kvm_read_guest_atomic);
1789
1790int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
1791 void *data, unsigned long len)
1792{
1793 gfn_t gfn = gpa >> PAGE_SHIFT;
1794 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1795 int offset = offset_in_page(gpa);
1796
1797 return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1798}
1799EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
1800
1801static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
1802 const void *data, int offset, int len)
Izik Eidus195aefd2007-10-01 22:14:18 +02001803{
Izik Eiduse0506bc2007-11-11 22:10:22 +02001804 int r;
1805 unsigned long addr;
Izik Eidus195aefd2007-10-01 22:14:18 +02001806
Radim Krčmář251eb842015-04-10 21:47:27 +02001807 addr = gfn_to_hva_memslot(memslot, gfn);
Izik Eiduse0506bc2007-11-11 22:10:22 +02001808 if (kvm_is_error_hva(addr))
Izik Eidus195aefd2007-10-01 22:14:18 +02001809 return -EFAULT;
Xiao Guangrong8b0cedf2011-05-15 23:22:04 +08001810 r = __copy_to_user((void __user *)addr + offset, data, len);
Izik Eiduse0506bc2007-11-11 22:10:22 +02001811 if (r)
1812 return -EFAULT;
Paolo Bonzinibc009e42015-05-26 12:43:41 +02001813 mark_page_dirty_in_slot(memslot, gfn);
Izik Eidus195aefd2007-10-01 22:14:18 +02001814 return 0;
1815}
Paolo Bonzini8e734852015-05-17 13:58:53 +02001816
1817int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
1818 const void *data, int offset, int len)
1819{
1820 struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1821
1822 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1823}
Izik Eidus195aefd2007-10-01 22:14:18 +02001824EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1825
Paolo Bonzini8e734852015-05-17 13:58:53 +02001826int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
1827 const void *data, int offset, int len)
1828{
1829 struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1830
1831 return __kvm_write_guest_page(slot, gfn, data, offset, len);
1832}
1833EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
1834
Izik Eidus195aefd2007-10-01 22:14:18 +02001835int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1836 unsigned long len)
1837{
1838 gfn_t gfn = gpa >> PAGE_SHIFT;
1839 int seg;
1840 int offset = offset_in_page(gpa);
1841 int ret;
1842
1843 while ((seg = next_segment(len, offset)) != 0) {
1844 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1845 if (ret < 0)
1846 return ret;
1847 offset = 0;
1848 len -= seg;
1849 data += seg;
1850 ++gfn;
1851 }
1852 return 0;
1853}
Wincy Vanff651cb2014-12-11 08:52:58 +03001854EXPORT_SYMBOL_GPL(kvm_write_guest);
Izik Eidus195aefd2007-10-01 22:14:18 +02001855
Paolo Bonzini8e734852015-05-17 13:58:53 +02001856int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
1857 unsigned long len)
1858{
1859 gfn_t gfn = gpa >> PAGE_SHIFT;
1860 int seg;
1861 int offset = offset_in_page(gpa);
1862 int ret;
1863
1864 while ((seg = next_segment(len, offset)) != 0) {
1865 ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
1866 if (ret < 0)
1867 return ret;
1868 offset = 0;
1869 len -= seg;
1870 data += seg;
1871 ++gfn;
1872 }
1873 return 0;
1874}
1875EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
1876
Gleb Natapov49c77542010-10-18 15:22:23 +02001877int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
Andrew Honig8f964522013-03-29 09:35:21 -07001878 gpa_t gpa, unsigned long len)
Gleb Natapov49c77542010-10-18 15:22:23 +02001879{
1880 struct kvm_memslots *slots = kvm_memslots(kvm);
1881 int offset = offset_in_page(gpa);
Andrew Honig8f964522013-03-29 09:35:21 -07001882 gfn_t start_gfn = gpa >> PAGE_SHIFT;
1883 gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1884 gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1885 gfn_t nr_pages_avail;
Gleb Natapov49c77542010-10-18 15:22:23 +02001886
1887 ghc->gpa = gpa;
1888 ghc->generation = slots->generation;
Andrew Honig8f964522013-03-29 09:35:21 -07001889 ghc->len = len;
1890 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
Radim Krčmářca3f0872015-04-08 14:16:48 +02001891 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, NULL);
1892 if (!kvm_is_error_hva(ghc->hva) && nr_pages_needed <= 1) {
Gleb Natapov49c77542010-10-18 15:22:23 +02001893 ghc->hva += offset;
Andrew Honig8f964522013-03-29 09:35:21 -07001894 } else {
1895 /*
1896 * If the requested region crosses two memslots, we still
1897 * verify that the entire region is valid here.
1898 */
1899 while (start_gfn <= end_gfn) {
1900 ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1901 ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1902 &nr_pages_avail);
1903 if (kvm_is_error_hva(ghc->hva))
1904 return -EFAULT;
1905 start_gfn += nr_pages_avail;
1906 }
1907 /* Use the slow path for cross page reads and writes. */
1908 ghc->memslot = NULL;
1909 }
Gleb Natapov49c77542010-10-18 15:22:23 +02001910 return 0;
1911}
1912EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1913
1914int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1915 void *data, unsigned long len)
1916{
1917 struct kvm_memslots *slots = kvm_memslots(kvm);
1918 int r;
1919
Andrew Honig8f964522013-03-29 09:35:21 -07001920 BUG_ON(len > ghc->len);
1921
Gleb Natapov49c77542010-10-18 15:22:23 +02001922 if (slots->generation != ghc->generation)
Andrew Honig8f964522013-03-29 09:35:21 -07001923 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1924
1925 if (unlikely(!ghc->memslot))
1926 return kvm_write_guest(kvm, ghc->gpa, data, len);
Gleb Natapov49c77542010-10-18 15:22:23 +02001927
1928 if (kvm_is_error_hva(ghc->hva))
1929 return -EFAULT;
1930
Xiao Guangrong8b0cedf2011-05-15 23:22:04 +08001931 r = __copy_to_user((void __user *)ghc->hva, data, len);
Gleb Natapov49c77542010-10-18 15:22:23 +02001932 if (r)
1933 return -EFAULT;
Paolo Bonzinibc009e42015-05-26 12:43:41 +02001934 mark_page_dirty_in_slot(ghc->memslot, ghc->gpa >> PAGE_SHIFT);
Gleb Natapov49c77542010-10-18 15:22:23 +02001935
1936 return 0;
1937}
1938EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1939
Gleb Natapove03b6442011-07-11 15:28:11 -04001940int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1941 void *data, unsigned long len)
1942{
1943 struct kvm_memslots *slots = kvm_memslots(kvm);
1944 int r;
1945
Andrew Honig8f964522013-03-29 09:35:21 -07001946 BUG_ON(len > ghc->len);
1947
Gleb Natapove03b6442011-07-11 15:28:11 -04001948 if (slots->generation != ghc->generation)
Andrew Honig8f964522013-03-29 09:35:21 -07001949 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1950
1951 if (unlikely(!ghc->memslot))
1952 return kvm_read_guest(kvm, ghc->gpa, data, len);
Gleb Natapove03b6442011-07-11 15:28:11 -04001953
1954 if (kvm_is_error_hva(ghc->hva))
1955 return -EFAULT;
1956
1957 r = __copy_from_user(data, (void __user *)ghc->hva, len);
1958 if (r)
1959 return -EFAULT;
1960
1961 return 0;
1962}
1963EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1964
Izik Eidus195aefd2007-10-01 22:14:18 +02001965int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1966{
Heiko Carstens8a3caa62013-11-18 10:35:55 +01001967 const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1968
1969 return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
Izik Eidus195aefd2007-10-01 22:14:18 +02001970}
1971EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1972
1973int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1974{
1975 gfn_t gfn = gpa >> PAGE_SHIFT;
1976 int seg;
1977 int offset = offset_in_page(gpa);
1978 int ret;
1979
Kevin Mulveybfda0e82015-02-20 08:21:36 -05001980 while ((seg = next_segment(len, offset)) != 0) {
Izik Eidus195aefd2007-10-01 22:14:18 +02001981 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1982 if (ret < 0)
1983 return ret;
1984 offset = 0;
1985 len -= seg;
1986 ++gfn;
1987 }
1988 return 0;
1989}
1990EXPORT_SYMBOL_GPL(kvm_clear_guest);
1991
Paolo Bonzinibc009e42015-05-26 12:43:41 +02001992static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
Stephen Hemminger79408762013-12-29 12:12:29 -08001993 gfn_t gfn)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001994{
Rusty Russell7e9d6192007-07-31 20:41:14 +10001995 if (memslot && memslot->dirty_bitmap) {
1996 unsigned long rel_gfn = gfn - memslot->base_gfn;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001997
Takuya Yoshikawab74ca3b2012-10-04 17:13:12 -07001998 set_bit_le(rel_gfn, memslot->dirty_bitmap);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001999 }
2000}
2001
Gleb Natapov49c77542010-10-18 15:22:23 +02002002void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
2003{
2004 struct kvm_memory_slot *memslot;
2005
2006 memslot = gfn_to_memslot(kvm, gfn);
Paolo Bonzinibc009e42015-05-26 12:43:41 +02002007 mark_page_dirty_in_slot(memslot, gfn);
Gleb Natapov49c77542010-10-18 15:22:23 +02002008}
Aneesh Kumar K.V2ba9f0d2013-10-07 22:17:59 +05302009EXPORT_SYMBOL_GPL(mark_page_dirty);
Gleb Natapov49c77542010-10-18 15:22:23 +02002010
Paolo Bonzini8e734852015-05-17 13:58:53 +02002011void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
2012{
2013 struct kvm_memory_slot *memslot;
2014
2015 memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2016 mark_page_dirty_in_slot(memslot, gfn);
2017}
2018EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
2019
Wanpeng Liaca6ff22015-09-03 22:07:38 +08002020static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
2021{
Christian Borntraeger6b6de682016-02-09 13:47:55 +01002022 unsigned int old, val, grow;
Wanpeng Liaca6ff22015-09-03 22:07:38 +08002023
Wanpeng Li2cbd7822015-09-03 22:07:39 +08002024 old = val = vcpu->halt_poll_ns;
Christian Borntraeger6b6de682016-02-09 13:47:55 +01002025 grow = READ_ONCE(halt_poll_ns_grow);
Wanpeng Liaca6ff22015-09-03 22:07:38 +08002026 /* 10us base */
Christian Borntraeger6b6de682016-02-09 13:47:55 +01002027 if (val == 0 && grow)
Wanpeng Liaca6ff22015-09-03 22:07:38 +08002028 val = 10000;
2029 else
Christian Borntraeger6b6de682016-02-09 13:47:55 +01002030 val *= grow;
Wanpeng Liaca6ff22015-09-03 22:07:38 +08002031
David Matlack313f6362016-03-08 16:19:44 -08002032 if (val > halt_poll_ns)
2033 val = halt_poll_ns;
2034
Wanpeng Liaca6ff22015-09-03 22:07:38 +08002035 vcpu->halt_poll_ns = val;
Wanpeng Li2cbd7822015-09-03 22:07:39 +08002036 trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
Wanpeng Liaca6ff22015-09-03 22:07:38 +08002037}
2038
2039static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
2040{
Christian Borntraeger6b6de682016-02-09 13:47:55 +01002041 unsigned int old, val, shrink;
Wanpeng Liaca6ff22015-09-03 22:07:38 +08002042
Wanpeng Li2cbd7822015-09-03 22:07:39 +08002043 old = val = vcpu->halt_poll_ns;
Christian Borntraeger6b6de682016-02-09 13:47:55 +01002044 shrink = READ_ONCE(halt_poll_ns_shrink);
2045 if (shrink == 0)
Wanpeng Liaca6ff22015-09-03 22:07:38 +08002046 val = 0;
2047 else
Christian Borntraeger6b6de682016-02-09 13:47:55 +01002048 val /= shrink;
Wanpeng Liaca6ff22015-09-03 22:07:38 +08002049
2050 vcpu->halt_poll_ns = val;
Wanpeng Li2cbd7822015-09-03 22:07:39 +08002051 trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
Wanpeng Liaca6ff22015-09-03 22:07:38 +08002052}
2053
Paolo Bonzinif7819512015-02-04 18:20:58 +01002054static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
2055{
2056 if (kvm_arch_vcpu_runnable(vcpu)) {
2057 kvm_make_request(KVM_REQ_UNHALT, vcpu);
2058 return -EINTR;
2059 }
2060 if (kvm_cpu_has_pending_timer(vcpu))
2061 return -EINTR;
2062 if (signal_pending(current))
2063 return -EINTR;
2064
2065 return 0;
2066}
2067
Eddie Dongb6958ce2007-07-18 12:15:21 +03002068/*
2069 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2070 */
Hollis Blanchard8776e512007-10-31 17:24:24 -05002071void kvm_vcpu_block(struct kvm_vcpu *vcpu)
Eddie Dongb6958ce2007-07-18 12:15:21 +03002072{
Paolo Bonzinif7819512015-02-04 18:20:58 +01002073 ktime_t start, cur;
Marcelo Tosatti85773702016-02-19 09:46:39 +01002074 DECLARE_SWAITQUEUE(wait);
Paolo Bonzinif7819512015-02-04 18:20:58 +01002075 bool waited = false;
Wanpeng Liaca6ff22015-09-03 22:07:38 +08002076 u64 block_ns;
Paolo Bonzinif7819512015-02-04 18:20:58 +01002077
2078 start = cur = ktime_get();
Wanpeng Li19020f82015-09-03 22:07:37 +08002079 if (vcpu->halt_poll_ns) {
2080 ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
Xiubo Lif95ef0c2015-02-26 14:58:23 +08002081
Paolo Bonzini62bea5b2015-09-15 18:27:57 +02002082 ++vcpu->stat.halt_attempted_poll;
Paolo Bonzinif7819512015-02-04 18:20:58 +01002083 do {
2084 /*
2085 * This sets KVM_REQ_UNHALT if an interrupt
2086 * arrives.
2087 */
2088 if (kvm_vcpu_check_block(vcpu) < 0) {
2089 ++vcpu->stat.halt_successful_poll;
Christian Borntraeger3491caf2016-05-13 12:16:35 +02002090 if (!vcpu_valid_wakeup(vcpu))
2091 ++vcpu->stat.halt_poll_invalid;
Paolo Bonzinif7819512015-02-04 18:20:58 +01002092 goto out;
2093 }
2094 cur = ktime_get();
2095 } while (single_task_running() && ktime_before(cur, stop));
2096 }
Eddie Dongb6958ce2007-07-18 12:15:21 +03002097
Christoffer Dall3217f7c2015-08-27 16:41:15 +02002098 kvm_arch_vcpu_blocking(vcpu);
2099
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03002100 for (;;) {
Marcelo Tosatti85773702016-02-19 09:46:39 +01002101 prepare_to_swait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
Eddie Dongb6958ce2007-07-18 12:15:21 +03002102
Paolo Bonzinif7819512015-02-04 18:20:58 +01002103 if (kvm_vcpu_check_block(vcpu) < 0)
Marcelo Tosattie5c239c2008-05-08 19:47:01 -03002104 break;
2105
Paolo Bonzinif7819512015-02-04 18:20:58 +01002106 waited = true;
Eddie Dongb6958ce2007-07-18 12:15:21 +03002107 schedule();
Eddie Dongb6958ce2007-07-18 12:15:21 +03002108 }
2109
Marcelo Tosatti85773702016-02-19 09:46:39 +01002110 finish_swait(&vcpu->wq, &wait);
Paolo Bonzinif7819512015-02-04 18:20:58 +01002111 cur = ktime_get();
2112
Christoffer Dall3217f7c2015-08-27 16:41:15 +02002113 kvm_arch_vcpu_unblocking(vcpu);
Paolo Bonzinif7819512015-02-04 18:20:58 +01002114out:
Wanpeng Liaca6ff22015-09-03 22:07:38 +08002115 block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2116
Christian Borntraeger2086d322016-05-17 10:49:22 +02002117 if (!vcpu_valid_wakeup(vcpu))
2118 shrink_halt_poll_ns(vcpu);
2119 else if (halt_poll_ns) {
Wanpeng Liaca6ff22015-09-03 22:07:38 +08002120 if (block_ns <= vcpu->halt_poll_ns)
2121 ;
2122 /* we had a long block, shrink polling */
Christian Borntraeger2086d322016-05-17 10:49:22 +02002123 else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns)
Wanpeng Liaca6ff22015-09-03 22:07:38 +08002124 shrink_halt_poll_ns(vcpu);
2125 /* we had a short halt and our poll time is too small */
2126 else if (vcpu->halt_poll_ns < halt_poll_ns &&
2127 block_ns < halt_poll_ns)
2128 grow_halt_poll_ns(vcpu);
Wanpeng Liedb92722015-09-14 17:38:51 +08002129 } else
2130 vcpu->halt_poll_ns = 0;
Wanpeng Liaca6ff22015-09-03 22:07:38 +08002131
Christian Borntraeger3491caf2016-05-13 12:16:35 +02002132 trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
2133 kvm_arch_vcpu_block_finish(vcpu);
Eddie Dongb6958ce2007-07-18 12:15:21 +03002134}
Aneesh Kumar K.V2ba9f0d2013-10-07 22:17:59 +05302135EXPORT_SYMBOL_GPL(kvm_vcpu_block);
Eddie Dongb6958ce2007-07-18 12:15:21 +03002136
Marcelo Tosatti8c847802012-03-14 17:58:48 -03002137#ifndef CONFIG_S390
Radim Krčmářdd1a4cc2016-05-04 14:09:44 -05002138void kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
Christoffer Dallb6d33832012-03-08 16:44:24 -05002139{
Marcelo Tosatti85773702016-02-19 09:46:39 +01002140 struct swait_queue_head *wqp;
Christoffer Dallb6d33832012-03-08 16:44:24 -05002141
2142 wqp = kvm_arch_vcpu_wq(vcpu);
Marcelo Tosatti85773702016-02-19 09:46:39 +01002143 if (swait_active(wqp)) {
2144 swake_up(wqp);
Christoffer Dallb6d33832012-03-08 16:44:24 -05002145 ++vcpu->stat.halt_wakeup;
2146 }
2147
Radim Krčmářdd1a4cc2016-05-04 14:09:44 -05002148}
2149EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
2150
2151/*
2152 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2153 */
2154void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2155{
2156 int me;
2157 int cpu = vcpu->cpu;
2158
2159 kvm_vcpu_wake_up(vcpu);
Christoffer Dallb6d33832012-03-08 16:44:24 -05002160 me = get_cpu();
2161 if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2162 if (kvm_arch_vcpu_should_kick(vcpu))
2163 smp_send_reschedule(cpu);
2164 put_cpu();
2165}
Yang Zhanga20ed542013-04-11 19:25:15 +08002166EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
Marcelo Tosatti8c847802012-03-14 17:58:48 -03002167#endif /* !CONFIG_S390 */
Christoffer Dallb6d33832012-03-08 16:44:24 -05002168
Dan Carpenterfa933842014-05-23 13:20:42 +03002169int kvm_vcpu_yield_to(struct kvm_vcpu *target)
Konstantin Weitz41628d32012-04-25 15:30:38 +02002170{
2171 struct pid *pid;
2172 struct task_struct *task = NULL;
Dan Carpenterfa933842014-05-23 13:20:42 +03002173 int ret = 0;
Konstantin Weitz41628d32012-04-25 15:30:38 +02002174
2175 rcu_read_lock();
2176 pid = rcu_dereference(target->pid);
2177 if (pid)
Sam Bobroff27fbe64b2014-09-19 09:40:41 +10002178 task = get_pid_task(pid, PIDTYPE_PID);
Konstantin Weitz41628d32012-04-25 15:30:38 +02002179 rcu_read_unlock();
2180 if (!task)
Raghavendra K Tc45c5282013-01-22 13:09:24 +05302181 return ret;
Raghavendra K Tc45c5282013-01-22 13:09:24 +05302182 ret = yield_to(task, 1);
Konstantin Weitz41628d32012-04-25 15:30:38 +02002183 put_task_struct(task);
Raghavendra K Tc45c5282013-01-22 13:09:24 +05302184
2185 return ret;
Konstantin Weitz41628d32012-04-25 15:30:38 +02002186}
2187EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2188
Raghavendra K T06e48c52012-07-19 15:17:52 +05302189/*
2190 * Helper that checks whether a VCPU is eligible for directed yield.
2191 * Most eligible candidate to yield is decided by following heuristics:
2192 *
2193 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2194 * (preempted lock holder), indicated by @in_spin_loop.
2195 * Set at the beiginning and cleared at the end of interception/PLE handler.
2196 *
2197 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2198 * chance last time (mostly it has become eligible now since we have probably
2199 * yielded to lockholder in last iteration. This is done by toggling
2200 * @dy_eligible each time a VCPU checked for eligibility.)
2201 *
2202 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2203 * to preempted lock-holder could result in wrong VCPU selection and CPU
2204 * burning. Giving priority for a potential lock-holder increases lock
2205 * progress.
2206 *
2207 * Since algorithm is based on heuristics, accessing another VCPU data without
2208 * locking does not harm. It may result in trying to yield to same VCPU, fail
2209 * and continue with next VCPU and so on.
2210 */
Stephen Hemminger79408762013-12-29 12:12:29 -08002211static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
Raghavendra K T06e48c52012-07-19 15:17:52 +05302212{
Scott Wood4a55dd72014-01-09 18:43:16 -06002213#ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
Raghavendra K T06e48c52012-07-19 15:17:52 +05302214 bool eligible;
2215
2216 eligible = !vcpu->spin_loop.in_spin_loop ||
Christian Borntraeger34656112014-09-04 21:13:31 +02002217 vcpu->spin_loop.dy_eligible;
Raghavendra K T06e48c52012-07-19 15:17:52 +05302218
2219 if (vcpu->spin_loop.in_spin_loop)
2220 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2221
2222 return eligible;
Scott Wood4a55dd72014-01-09 18:43:16 -06002223#else
2224 return true;
Raghavendra K T06e48c52012-07-19 15:17:52 +05302225#endif
Scott Wood4a55dd72014-01-09 18:43:16 -06002226}
Raghavendra K Tc45c5282013-01-22 13:09:24 +05302227
Rik van Riel217ece62011-02-01 09:53:28 -05002228void kvm_vcpu_on_spin(struct kvm_vcpu *me)
Zhai, Edwind255f4f2009-10-09 18:03:20 +08002229{
Rik van Riel217ece62011-02-01 09:53:28 -05002230 struct kvm *kvm = me->kvm;
2231 struct kvm_vcpu *vcpu;
2232 int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2233 int yielded = 0;
Raghavendra K Tc45c5282013-01-22 13:09:24 +05302234 int try = 3;
Rik van Riel217ece62011-02-01 09:53:28 -05002235 int pass;
2236 int i;
Zhai, Edwind255f4f2009-10-09 18:03:20 +08002237
Raghavendra K T4c088492012-07-18 19:07:46 +05302238 kvm_vcpu_set_in_spin_loop(me, true);
Rik van Riel217ece62011-02-01 09:53:28 -05002239 /*
2240 * We boost the priority of a VCPU that is runnable but not
2241 * currently running, because it got preempted by something
2242 * else and called schedule in __vcpu_run. Hopefully that
2243 * VCPU is holding the lock that we need and will release it.
2244 * We approximate round-robin by starting at the last boosted VCPU.
2245 */
Raghavendra K Tc45c5282013-01-22 13:09:24 +05302246 for (pass = 0; pass < 2 && !yielded && try; pass++) {
Rik van Riel217ece62011-02-01 09:53:28 -05002247 kvm_for_each_vcpu(i, vcpu, kvm) {
Rik van Riel5cfc2aa2012-06-19 16:51:04 -04002248 if (!pass && i <= last_boosted_vcpu) {
Rik van Riel217ece62011-02-01 09:53:28 -05002249 i = last_boosted_vcpu;
2250 continue;
2251 } else if (pass && i > last_boosted_vcpu)
2252 break;
Raghavendra K T7bc7ae22013-03-04 23:32:27 +05302253 if (!ACCESS_ONCE(vcpu->preempted))
2254 continue;
Rik van Riel217ece62011-02-01 09:53:28 -05002255 if (vcpu == me)
2256 continue;
Marcelo Tosatti85773702016-02-19 09:46:39 +01002257 if (swait_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
Rik van Riel217ece62011-02-01 09:53:28 -05002258 continue;
Raghavendra K T06e48c52012-07-19 15:17:52 +05302259 if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2260 continue;
Raghavendra K Tc45c5282013-01-22 13:09:24 +05302261
2262 yielded = kvm_vcpu_yield_to(vcpu);
2263 if (yielded > 0) {
Rik van Riel217ece62011-02-01 09:53:28 -05002264 kvm->last_boosted_vcpu = i;
Rik van Riel217ece62011-02-01 09:53:28 -05002265 break;
Raghavendra K Tc45c5282013-01-22 13:09:24 +05302266 } else if (yielded < 0) {
2267 try--;
2268 if (!try)
2269 break;
Rik van Riel217ece62011-02-01 09:53:28 -05002270 }
Rik van Riel217ece62011-02-01 09:53:28 -05002271 }
2272 }
Raghavendra K T4c088492012-07-18 19:07:46 +05302273 kvm_vcpu_set_in_spin_loop(me, false);
Raghavendra K T06e48c52012-07-19 15:17:52 +05302274
2275 /* Ensure vcpu is not eligible during next spinloop */
2276 kvm_vcpu_set_dy_eligible(me, false);
Zhai, Edwind255f4f2009-10-09 18:03:20 +08002277}
2278EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2279
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11002280static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02002281{
2282 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02002283 struct page *page;
2284
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11002285 if (vmf->pgoff == 0)
Avi Kivity039576c2007-03-20 12:46:50 +02002286 page = virt_to_page(vcpu->run);
Avi Kivity09566762008-01-23 18:14:23 +02002287#ifdef CONFIG_X86
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11002288 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
Zhang Xiantaoad312c72007-12-13 23:50:52 +08002289 page = virt_to_page(vcpu->arch.pio_data);
Avi Kivity09566762008-01-23 18:14:23 +02002290#endif
Laurent Vivier5f94c172008-05-30 16:05:54 +02002291#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2292 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2293 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2294#endif
Avi Kivity039576c2007-03-20 12:46:50 +02002295 else
Carsten Otte5b1c1492012-01-04 10:25:23 +01002296 return kvm_arch_vcpu_fault(vcpu, vmf);
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02002297 get_page(page);
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11002298 vmf->page = page;
2299 return 0;
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02002300}
2301
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04002302static const struct vm_operations_struct kvm_vcpu_vm_ops = {
npiggin@suse.dee4a533a2007-12-05 18:15:52 +11002303 .fault = kvm_vcpu_fault,
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02002304};
2305
2306static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2307{
2308 vma->vm_ops = &kvm_vcpu_vm_ops;
2309 return 0;
2310}
2311
Avi Kivitybccf2152007-02-21 18:04:26 +02002312static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2313{
2314 struct kvm_vcpu *vcpu = filp->private_data;
2315
Al Viro66c0b392008-04-19 20:33:56 +01002316 kvm_put_kvm(vcpu->kvm);
Avi Kivitybccf2152007-02-21 18:04:26 +02002317 return 0;
2318}
2319
Christian Borntraeger3d3aab12008-12-02 11:17:32 +01002320static struct file_operations kvm_vcpu_fops = {
Avi Kivitybccf2152007-02-21 18:04:26 +02002321 .release = kvm_vcpu_release,
2322 .unlocked_ioctl = kvm_vcpu_ioctl,
Christian Borntraegerde8e5d72015-02-03 09:35:15 +01002323#ifdef CONFIG_KVM_COMPAT
Alexander Graf1dda6062011-06-08 02:45:37 +02002324 .compat_ioctl = kvm_vcpu_compat_ioctl,
2325#endif
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02002326 .mmap = kvm_vcpu_mmap,
Arnd Bergmann6038f372010-08-15 18:52:59 +02002327 .llseek = noop_llseek,
Avi Kivitybccf2152007-02-21 18:04:26 +02002328};
2329
2330/*
2331 * Allocates an inode for the vcpu.
2332 */
2333static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2334{
Yann Droneaud24009b02013-08-24 22:14:07 +02002335 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
Avi Kivitybccf2152007-02-21 18:04:26 +02002336}
2337
Avi Kivityc5ea7662007-02-20 18:41:05 +02002338/*
2339 * Creates some virtual cpus. Good luck creating more than one.
2340 */
Gleb Natapov73880c82009-06-09 15:56:28 +03002341static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
Avi Kivityc5ea7662007-02-20 18:41:05 +02002342{
2343 int r;
David Hildenbrande09fefd2015-11-05 09:03:50 +01002344 struct kvm_vcpu *vcpu;
Avi Kivityc5ea7662007-02-20 18:41:05 +02002345
Greg Kurz0b1b1df2016-05-09 18:13:37 +02002346 if (id >= KVM_MAX_VCPU_ID)
Andy Honig338c7db2013-11-18 16:09:22 -08002347 return -EINVAL;
2348
Gleb Natapov73880c82009-06-09 15:56:28 +03002349 vcpu = kvm_arch_vcpu_create(kvm, id);
Rusty Russellfb3f0f52007-07-27 17:16:56 +10002350 if (IS_ERR(vcpu))
2351 return PTR_ERR(vcpu);
Avi Kivityc5ea7662007-02-20 18:41:05 +02002352
Avi Kivity15ad7142007-07-11 18:17:21 +03002353 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2354
Avi Kivity26e52152007-11-20 15:30:24 +02002355 r = kvm_arch_vcpu_setup(vcpu);
2356 if (r)
Jan Kiszkad7805922011-05-23 10:33:05 +02002357 goto vcpu_destroy;
Avi Kivity26e52152007-11-20 15:30:24 +02002358
Shaohua Li11ec2802007-07-23 14:51:37 +08002359 mutex_lock(&kvm->lock);
Avi Kivity3e515702012-03-05 14:23:29 +02002360 if (!kvm_vcpu_compatible(vcpu)) {
2361 r = -EINVAL;
2362 goto unlock_vcpu_destroy;
2363 }
Gleb Natapov73880c82009-06-09 15:56:28 +03002364 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
2365 r = -EINVAL;
Jan Kiszkad7805922011-05-23 10:33:05 +02002366 goto unlock_vcpu_destroy;
Rusty Russellfb3f0f52007-07-27 17:16:56 +10002367 }
David Hildenbrande09fefd2015-11-05 09:03:50 +01002368 if (kvm_get_vcpu_by_id(kvm, id)) {
2369 r = -EEXIST;
2370 goto unlock_vcpu_destroy;
2371 }
Gleb Natapov73880c82009-06-09 15:56:28 +03002372
2373 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
Rusty Russellfb3f0f52007-07-27 17:16:56 +10002374
2375 /* Now it's all set up, let userspace reach it */
Al Viro66c0b392008-04-19 20:33:56 +01002376 kvm_get_kvm(kvm);
Avi Kivitybccf2152007-02-21 18:04:26 +02002377 r = create_vcpu_fd(vcpu);
Gleb Natapov73880c82009-06-09 15:56:28 +03002378 if (r < 0) {
2379 kvm_put_kvm(kvm);
Jan Kiszkad7805922011-05-23 10:33:05 +02002380 goto unlock_vcpu_destroy;
Gleb Natapov73880c82009-06-09 15:56:28 +03002381 }
2382
2383 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
Paolo Bonzinidd489242015-07-29 11:32:20 +02002384
2385 /*
2386 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2387 * before kvm->online_vcpu's incremented value.
2388 */
Gleb Natapov73880c82009-06-09 15:56:28 +03002389 smp_wmb();
2390 atomic_inc(&kvm->online_vcpus);
2391
Gleb Natapov73880c82009-06-09 15:56:28 +03002392 mutex_unlock(&kvm->lock);
Marcelo Tosatti42897d82012-11-27 23:29:02 -02002393 kvm_arch_vcpu_postcreate(vcpu);
Avi Kivitybccf2152007-02-21 18:04:26 +02002394 return r;
Avi Kivityc5ea7662007-02-20 18:41:05 +02002395
Jan Kiszkad7805922011-05-23 10:33:05 +02002396unlock_vcpu_destroy:
Glauber Costa7d8fece2008-09-17 23:16:59 -03002397 mutex_unlock(&kvm->lock);
Jan Kiszkad7805922011-05-23 10:33:05 +02002398vcpu_destroy:
Hollis Blanchardd40ccc62007-11-19 14:04:43 -06002399 kvm_arch_vcpu_destroy(vcpu);
Avi Kivityc5ea7662007-02-20 18:41:05 +02002400 return r;
2401}
2402
Avi Kivity1961d272007-03-05 19:46:05 +02002403static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2404{
2405 if (sigset) {
2406 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2407 vcpu->sigset_active = 1;
2408 vcpu->sigset = *sigset;
2409 } else
2410 vcpu->sigset_active = 0;
2411 return 0;
2412}
2413
Avi Kivitybccf2152007-02-21 18:04:26 +02002414static long kvm_vcpu_ioctl(struct file *filp,
2415 unsigned int ioctl, unsigned long arg)
Avi Kivity6aa8b732006-12-10 02:21:36 -08002416{
Avi Kivitybccf2152007-02-21 18:04:26 +02002417 struct kvm_vcpu *vcpu = filp->private_data;
Al Viro2f3669872007-02-09 16:38:35 +00002418 void __user *argp = (void __user *)arg;
Carsten Otte313a3dc2007-10-11 19:16:52 +02002419 int r;
Dave Hansenfa3795a2008-08-11 10:01:46 -07002420 struct kvm_fpu *fpu = NULL;
2421 struct kvm_sregs *kvm_sregs = NULL;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002422
Avi Kivity6d4e4c42007-11-21 16:41:05 +02002423 if (vcpu->kvm->mm != current->mm)
2424 return -EIO;
Avi Kivity2122ff52010-05-13 11:25:04 +03002425
David Matlack2ea75be2014-09-19 16:03:25 -07002426 if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2427 return -EINVAL;
2428
Sanjay Lal2f4d9b52012-11-21 18:34:15 -08002429#if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
Avi Kivity2122ff52010-05-13 11:25:04 +03002430 /*
2431 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2432 * so vcpu_load() would break it.
2433 */
Jens Freimann47b43c52014-11-11 20:57:06 +01002434 if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_S390_IRQ || ioctl == KVM_INTERRUPT)
Avi Kivity2122ff52010-05-13 11:25:04 +03002435 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2436#endif
2437
2438
Michael S. Tsirkin9fc77442012-09-16 11:50:30 +03002439 r = vcpu_load(vcpu);
2440 if (r)
2441 return r;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002442 switch (ioctl) {
Avi Kivity9a2bb7f2007-02-22 12:58:31 +02002443 case KVM_RUN:
Avi Kivityf0fe5102007-03-07 13:11:17 +02002444 r = -EINVAL;
2445 if (arg)
2446 goto out;
Christian Borntraeger7a72f7a2014-08-05 16:44:14 +02002447 if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
2448 /* The thread running this VCPU changed. */
2449 struct pid *oldpid = vcpu->pid;
2450 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
Xiubo Lif95ef0c2015-02-26 14:58:23 +08002451
Christian Borntraeger7a72f7a2014-08-05 16:44:14 +02002452 rcu_assign_pointer(vcpu->pid, newpid);
2453 if (oldpid)
2454 synchronize_rcu();
2455 put_pid(oldpid);
2456 }
Hollis Blanchardb6c7a5d2007-11-01 14:16:10 -05002457 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
Gleb Natapov64be5002010-10-24 16:49:08 +02002458 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002459 break;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002460 case KVM_GET_REGS: {
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002461 struct kvm_regs *kvm_regs;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002462
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002463 r = -ENOMEM;
2464 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2465 if (!kvm_regs)
2466 goto out;
2467 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002468 if (r)
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002469 goto out_free1;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002470 r = -EFAULT;
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002471 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2472 goto out_free1;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002473 r = 0;
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002474out_free1:
2475 kfree(kvm_regs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002476 break;
2477 }
2478 case KVM_SET_REGS: {
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002479 struct kvm_regs *kvm_regs;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002480
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002481 r = -ENOMEM;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002482 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2483 if (IS_ERR(kvm_regs)) {
2484 r = PTR_ERR(kvm_regs);
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002485 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002486 }
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002487 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
Xiantao Zhang3e4bb3a2008-02-25 18:52:20 +08002488 kfree(kvm_regs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002489 break;
2490 }
2491 case KVM_GET_SREGS: {
Dave Hansenfa3795a2008-08-11 10:01:46 -07002492 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2493 r = -ENOMEM;
2494 if (!kvm_sregs)
2495 goto out;
2496 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002497 if (r)
2498 goto out;
2499 r = -EFAULT;
Dave Hansenfa3795a2008-08-11 10:01:46 -07002500 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002501 goto out;
2502 r = 0;
2503 break;
2504 }
2505 case KVM_SET_SREGS: {
Sasha Levinff5c2c02011-12-04 19:36:29 +02002506 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2507 if (IS_ERR(kvm_sregs)) {
2508 r = PTR_ERR(kvm_sregs);
Guo Chao18595412012-11-02 18:33:21 +08002509 kvm_sregs = NULL;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002510 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002511 }
Dave Hansenfa3795a2008-08-11 10:01:46 -07002512 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002513 break;
2514 }
Marcelo Tosatti62d9f0d2008-04-11 13:24:45 -03002515 case KVM_GET_MP_STATE: {
2516 struct kvm_mp_state mp_state;
2517
2518 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2519 if (r)
2520 goto out;
2521 r = -EFAULT;
Xiubo Li893bdbf2015-02-26 14:58:19 +08002522 if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
Marcelo Tosatti62d9f0d2008-04-11 13:24:45 -03002523 goto out;
2524 r = 0;
2525 break;
2526 }
2527 case KVM_SET_MP_STATE: {
2528 struct kvm_mp_state mp_state;
2529
2530 r = -EFAULT;
Xiubo Li893bdbf2015-02-26 14:58:19 +08002531 if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
Marcelo Tosatti62d9f0d2008-04-11 13:24:45 -03002532 goto out;
2533 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
Marcelo Tosatti62d9f0d2008-04-11 13:24:45 -03002534 break;
2535 }
Avi Kivity6aa8b732006-12-10 02:21:36 -08002536 case KVM_TRANSLATE: {
2537 struct kvm_translation tr;
2538
2539 r = -EFAULT;
Xiubo Li893bdbf2015-02-26 14:58:19 +08002540 if (copy_from_user(&tr, argp, sizeof(tr)))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002541 goto out;
Zhang Xiantao8b006792007-11-16 13:05:55 +08002542 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002543 if (r)
2544 goto out;
2545 r = -EFAULT;
Xiubo Li893bdbf2015-02-26 14:58:19 +08002546 if (copy_to_user(argp, &tr, sizeof(tr)))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002547 goto out;
2548 r = 0;
2549 break;
2550 }
Jan Kiszkad0bfb942008-12-15 13:52:10 +01002551 case KVM_SET_GUEST_DEBUG: {
2552 struct kvm_guest_debug dbg;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002553
2554 r = -EFAULT;
Xiubo Li893bdbf2015-02-26 14:58:19 +08002555 if (copy_from_user(&dbg, argp, sizeof(dbg)))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002556 goto out;
Jan Kiszkad0bfb942008-12-15 13:52:10 +01002557 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002558 break;
2559 }
Avi Kivity1961d272007-03-05 19:46:05 +02002560 case KVM_SET_SIGNAL_MASK: {
2561 struct kvm_signal_mask __user *sigmask_arg = argp;
2562 struct kvm_signal_mask kvm_sigmask;
2563 sigset_t sigset, *p;
2564
2565 p = NULL;
2566 if (argp) {
2567 r = -EFAULT;
2568 if (copy_from_user(&kvm_sigmask, argp,
Xiubo Li893bdbf2015-02-26 14:58:19 +08002569 sizeof(kvm_sigmask)))
Avi Kivity1961d272007-03-05 19:46:05 +02002570 goto out;
2571 r = -EINVAL;
Xiubo Li893bdbf2015-02-26 14:58:19 +08002572 if (kvm_sigmask.len != sizeof(sigset))
Avi Kivity1961d272007-03-05 19:46:05 +02002573 goto out;
2574 r = -EFAULT;
2575 if (copy_from_user(&sigset, sigmask_arg->sigset,
Xiubo Li893bdbf2015-02-26 14:58:19 +08002576 sizeof(sigset)))
Avi Kivity1961d272007-03-05 19:46:05 +02002577 goto out;
2578 p = &sigset;
2579 }
Andi Kleen376d41f2010-06-10 13:10:47 +02002580 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
Avi Kivity1961d272007-03-05 19:46:05 +02002581 break;
2582 }
Avi Kivityb8836732007-04-01 16:34:31 +03002583 case KVM_GET_FPU: {
Dave Hansenfa3795a2008-08-11 10:01:46 -07002584 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2585 r = -ENOMEM;
2586 if (!fpu)
2587 goto out;
2588 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
Avi Kivityb8836732007-04-01 16:34:31 +03002589 if (r)
2590 goto out;
2591 r = -EFAULT;
Dave Hansenfa3795a2008-08-11 10:01:46 -07002592 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
Avi Kivityb8836732007-04-01 16:34:31 +03002593 goto out;
2594 r = 0;
2595 break;
2596 }
2597 case KVM_SET_FPU: {
Sasha Levinff5c2c02011-12-04 19:36:29 +02002598 fpu = memdup_user(argp, sizeof(*fpu));
2599 if (IS_ERR(fpu)) {
2600 r = PTR_ERR(fpu);
Guo Chao18595412012-11-02 18:33:21 +08002601 fpu = NULL;
Avi Kivityb8836732007-04-01 16:34:31 +03002602 goto out;
Sasha Levinff5c2c02011-12-04 19:36:29 +02002603 }
Dave Hansenfa3795a2008-08-11 10:01:46 -07002604 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
Avi Kivityb8836732007-04-01 16:34:31 +03002605 break;
2606 }
Avi Kivitybccf2152007-02-21 18:04:26 +02002607 default:
Carsten Otte313a3dc2007-10-11 19:16:52 +02002608 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
Avi Kivitybccf2152007-02-21 18:04:26 +02002609 }
2610out:
Avi Kivity2122ff52010-05-13 11:25:04 +03002611 vcpu_put(vcpu);
Dave Hansenfa3795a2008-08-11 10:01:46 -07002612 kfree(fpu);
2613 kfree(kvm_sregs);
Avi Kivitybccf2152007-02-21 18:04:26 +02002614 return r;
2615}
2616
Christian Borntraegerde8e5d72015-02-03 09:35:15 +01002617#ifdef CONFIG_KVM_COMPAT
Alexander Graf1dda6062011-06-08 02:45:37 +02002618static long kvm_vcpu_compat_ioctl(struct file *filp,
2619 unsigned int ioctl, unsigned long arg)
2620{
2621 struct kvm_vcpu *vcpu = filp->private_data;
2622 void __user *argp = compat_ptr(arg);
2623 int r;
2624
2625 if (vcpu->kvm->mm != current->mm)
2626 return -EIO;
2627
2628 switch (ioctl) {
2629 case KVM_SET_SIGNAL_MASK: {
2630 struct kvm_signal_mask __user *sigmask_arg = argp;
2631 struct kvm_signal_mask kvm_sigmask;
2632 compat_sigset_t csigset;
2633 sigset_t sigset;
2634
2635 if (argp) {
2636 r = -EFAULT;
2637 if (copy_from_user(&kvm_sigmask, argp,
Xiubo Li893bdbf2015-02-26 14:58:19 +08002638 sizeof(kvm_sigmask)))
Alexander Graf1dda6062011-06-08 02:45:37 +02002639 goto out;
2640 r = -EINVAL;
Xiubo Li893bdbf2015-02-26 14:58:19 +08002641 if (kvm_sigmask.len != sizeof(csigset))
Alexander Graf1dda6062011-06-08 02:45:37 +02002642 goto out;
2643 r = -EFAULT;
2644 if (copy_from_user(&csigset, sigmask_arg->sigset,
Xiubo Li893bdbf2015-02-26 14:58:19 +08002645 sizeof(csigset)))
Alexander Graf1dda6062011-06-08 02:45:37 +02002646 goto out;
Alan Cox760a9a32012-08-22 14:34:11 +01002647 sigset_from_compat(&sigset, &csigset);
2648 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2649 } else
2650 r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
Alexander Graf1dda6062011-06-08 02:45:37 +02002651 break;
2652 }
2653 default:
2654 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2655 }
2656
2657out:
2658 return r;
2659}
2660#endif
2661
Scott Wood852b6d52013-04-12 14:08:42 +00002662static int kvm_device_ioctl_attr(struct kvm_device *dev,
2663 int (*accessor)(struct kvm_device *dev,
2664 struct kvm_device_attr *attr),
2665 unsigned long arg)
2666{
2667 struct kvm_device_attr attr;
2668
2669 if (!accessor)
2670 return -EPERM;
2671
2672 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2673 return -EFAULT;
2674
2675 return accessor(dev, &attr);
2676}
2677
2678static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2679 unsigned long arg)
2680{
2681 struct kvm_device *dev = filp->private_data;
2682
2683 switch (ioctl) {
2684 case KVM_SET_DEVICE_ATTR:
2685 return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2686 case KVM_GET_DEVICE_ATTR:
2687 return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2688 case KVM_HAS_DEVICE_ATTR:
2689 return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2690 default:
2691 if (dev->ops->ioctl)
2692 return dev->ops->ioctl(dev, ioctl, arg);
2693
2694 return -ENOTTY;
2695 }
2696}
2697
Scott Wood852b6d52013-04-12 14:08:42 +00002698static int kvm_device_release(struct inode *inode, struct file *filp)
2699{
2700 struct kvm_device *dev = filp->private_data;
2701 struct kvm *kvm = dev->kvm;
2702
Scott Wood852b6d52013-04-12 14:08:42 +00002703 kvm_put_kvm(kvm);
2704 return 0;
2705}
2706
2707static const struct file_operations kvm_device_fops = {
2708 .unlocked_ioctl = kvm_device_ioctl,
Christian Borntraegerde8e5d72015-02-03 09:35:15 +01002709#ifdef CONFIG_KVM_COMPAT
Scott Wooddb6ae612013-04-30 20:00:45 -05002710 .compat_ioctl = kvm_device_ioctl,
2711#endif
Scott Wood852b6d52013-04-12 14:08:42 +00002712 .release = kvm_device_release,
2713};
2714
2715struct kvm_device *kvm_device_from_filp(struct file *filp)
2716{
2717 if (filp->f_op != &kvm_device_fops)
2718 return NULL;
2719
2720 return filp->private_data;
2721}
2722
Will Deacond60eacb2014-09-02 10:27:33 +01002723static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2724#ifdef CONFIG_KVM_MPIC
2725 [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops,
2726 [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops,
2727#endif
2728
2729#ifdef CONFIG_KVM_XICS
2730 [KVM_DEV_TYPE_XICS] = &kvm_xics_ops,
2731#endif
Will Deacond60eacb2014-09-02 10:27:33 +01002732};
2733
2734int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2735{
2736 if (type >= ARRAY_SIZE(kvm_device_ops_table))
2737 return -ENOSPC;
2738
2739 if (kvm_device_ops_table[type] != NULL)
2740 return -EEXIST;
2741
2742 kvm_device_ops_table[type] = ops;
2743 return 0;
2744}
2745
Wanpeng Li571ee1b2014-10-09 18:30:08 +08002746void kvm_unregister_device_ops(u32 type)
2747{
2748 if (kvm_device_ops_table[type] != NULL)
2749 kvm_device_ops_table[type] = NULL;
2750}
2751
Scott Wood852b6d52013-04-12 14:08:42 +00002752static int kvm_ioctl_create_device(struct kvm *kvm,
2753 struct kvm_create_device *cd)
2754{
2755 struct kvm_device_ops *ops = NULL;
2756 struct kvm_device *dev;
2757 bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2758 int ret;
2759
Will Deacond60eacb2014-09-02 10:27:33 +01002760 if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
Scott Wood852b6d52013-04-12 14:08:42 +00002761 return -ENODEV;
Will Deacond60eacb2014-09-02 10:27:33 +01002762
2763 ops = kvm_device_ops_table[cd->type];
2764 if (ops == NULL)
2765 return -ENODEV;
Scott Wood852b6d52013-04-12 14:08:42 +00002766
2767 if (test)
2768 return 0;
2769
2770 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2771 if (!dev)
2772 return -ENOMEM;
2773
2774 dev->ops = ops;
2775 dev->kvm = kvm;
Scott Wood852b6d52013-04-12 14:08:42 +00002776
2777 ret = ops->create(dev, cd->type);
2778 if (ret < 0) {
2779 kfree(dev);
2780 return ret;
2781 }
2782
Yann Droneaud24009b02013-08-24 22:14:07 +02002783 ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
Scott Wood852b6d52013-04-12 14:08:42 +00002784 if (ret < 0) {
2785 ops->destroy(dev);
2786 return ret;
2787 }
2788
Scott Wood07f0a7b2013-04-25 14:11:23 +00002789 list_add(&dev->vm_node, &kvm->devices);
Scott Wood852b6d52013-04-12 14:08:42 +00002790 kvm_get_kvm(kvm);
2791 cd->fd = ret;
2792 return 0;
2793}
2794
Alexander Graf92b591a2014-07-14 18:33:08 +02002795static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2796{
2797 switch (arg) {
2798 case KVM_CAP_USER_MEMORY:
2799 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2800 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
Alexander Graf92b591a2014-07-14 18:33:08 +02002801 case KVM_CAP_INTERNAL_ERROR_DATA:
2802#ifdef CONFIG_HAVE_KVM_MSI
2803 case KVM_CAP_SIGNAL_MSI:
2804#endif
Paul Mackerras297e2102014-06-30 20:51:13 +10002805#ifdef CONFIG_HAVE_KVM_IRQFD
Paolo Bonzinidc9be0f2015-03-05 11:54:46 +01002806 case KVM_CAP_IRQFD:
Alexander Graf92b591a2014-07-14 18:33:08 +02002807 case KVM_CAP_IRQFD_RESAMPLE:
2808#endif
Jason Wange9ea5062015-09-15 14:41:59 +08002809 case KVM_CAP_IOEVENTFD_ANY_LENGTH:
Alexander Graf92b591a2014-07-14 18:33:08 +02002810 case KVM_CAP_CHECK_EXTENSION_VM:
2811 return 1;
2812#ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2813 case KVM_CAP_IRQ_ROUTING:
2814 return KVM_MAX_IRQ_ROUTES;
2815#endif
Paolo Bonzinif481b062015-05-17 17:30:37 +02002816#if KVM_ADDRESS_SPACE_NUM > 1
2817 case KVM_CAP_MULTI_ADDRESS_SPACE:
2818 return KVM_ADDRESS_SPACE_NUM;
2819#endif
Greg Kurz0b1b1df2016-05-09 18:13:37 +02002820 case KVM_CAP_MAX_VCPU_ID:
2821 return KVM_MAX_VCPU_ID;
Alexander Graf92b591a2014-07-14 18:33:08 +02002822 default:
2823 break;
2824 }
2825 return kvm_vm_ioctl_check_extension(kvm, arg);
2826}
2827
Avi Kivitybccf2152007-02-21 18:04:26 +02002828static long kvm_vm_ioctl(struct file *filp,
2829 unsigned int ioctl, unsigned long arg)
2830{
2831 struct kvm *kvm = filp->private_data;
2832 void __user *argp = (void __user *)arg;
Carsten Otte1fe779f2007-10-29 16:08:35 +01002833 int r;
Avi Kivitybccf2152007-02-21 18:04:26 +02002834
Avi Kivity6d4e4c42007-11-21 16:41:05 +02002835 if (kvm->mm != current->mm)
2836 return -EIO;
Avi Kivitybccf2152007-02-21 18:04:26 +02002837 switch (ioctl) {
2838 case KVM_CREATE_VCPU:
2839 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
Avi Kivitybccf2152007-02-21 18:04:26 +02002840 break;
Izik Eidus6fc138d2007-10-09 19:20:39 +02002841 case KVM_SET_USER_MEMORY_REGION: {
2842 struct kvm_userspace_memory_region kvm_userspace_mem;
2843
2844 r = -EFAULT;
2845 if (copy_from_user(&kvm_userspace_mem, argp,
Xiubo Li893bdbf2015-02-26 14:58:19 +08002846 sizeof(kvm_userspace_mem)))
Izik Eidus6fc138d2007-10-09 19:20:39 +02002847 goto out;
2848
Takuya Yoshikawa47ae31e2013-02-27 19:43:00 +09002849 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002850 break;
2851 }
2852 case KVM_GET_DIRTY_LOG: {
2853 struct kvm_dirty_log log;
2854
2855 r = -EFAULT;
Xiubo Li893bdbf2015-02-26 14:58:19 +08002856 if (copy_from_user(&log, argp, sizeof(log)))
Avi Kivity6aa8b732006-12-10 02:21:36 -08002857 goto out;
Avi Kivity2c6f5df2007-02-20 18:27:58 +02002858 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002859 break;
2860 }
Laurent Vivier5f94c172008-05-30 16:05:54 +02002861#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2862 case KVM_REGISTER_COALESCED_MMIO: {
2863 struct kvm_coalesced_mmio_zone zone;
Xiubo Lif95ef0c2015-02-26 14:58:23 +08002864
Laurent Vivier5f94c172008-05-30 16:05:54 +02002865 r = -EFAULT;
Xiubo Li893bdbf2015-02-26 14:58:19 +08002866 if (copy_from_user(&zone, argp, sizeof(zone)))
Laurent Vivier5f94c172008-05-30 16:05:54 +02002867 goto out;
Laurent Vivier5f94c172008-05-30 16:05:54 +02002868 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
Laurent Vivier5f94c172008-05-30 16:05:54 +02002869 break;
2870 }
2871 case KVM_UNREGISTER_COALESCED_MMIO: {
2872 struct kvm_coalesced_mmio_zone zone;
Xiubo Lif95ef0c2015-02-26 14:58:23 +08002873
Laurent Vivier5f94c172008-05-30 16:05:54 +02002874 r = -EFAULT;
Xiubo Li893bdbf2015-02-26 14:58:19 +08002875 if (copy_from_user(&zone, argp, sizeof(zone)))
Laurent Vivier5f94c172008-05-30 16:05:54 +02002876 goto out;
Laurent Vivier5f94c172008-05-30 16:05:54 +02002877 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
Laurent Vivier5f94c172008-05-30 16:05:54 +02002878 break;
2879 }
2880#endif
Gregory Haskins721eecb2009-05-20 10:30:49 -04002881 case KVM_IRQFD: {
2882 struct kvm_irqfd data;
2883
2884 r = -EFAULT;
Xiubo Li893bdbf2015-02-26 14:58:19 +08002885 if (copy_from_user(&data, argp, sizeof(data)))
Gregory Haskins721eecb2009-05-20 10:30:49 -04002886 goto out;
Alex Williamsond4db2932012-06-29 09:56:08 -06002887 r = kvm_irqfd(kvm, &data);
Gregory Haskins721eecb2009-05-20 10:30:49 -04002888 break;
2889 }
Gregory Haskinsd34e6b12009-07-07 17:08:49 -04002890 case KVM_IOEVENTFD: {
2891 struct kvm_ioeventfd data;
2892
2893 r = -EFAULT;
Xiubo Li893bdbf2015-02-26 14:58:19 +08002894 if (copy_from_user(&data, argp, sizeof(data)))
Gregory Haskinsd34e6b12009-07-07 17:08:49 -04002895 goto out;
2896 r = kvm_ioeventfd(kvm, &data);
2897 break;
2898 }
Jan Kiszka07975ad2012-03-29 21:14:12 +02002899#ifdef CONFIG_HAVE_KVM_MSI
2900 case KVM_SIGNAL_MSI: {
2901 struct kvm_msi msi;
2902
2903 r = -EFAULT;
Xiubo Li893bdbf2015-02-26 14:58:19 +08002904 if (copy_from_user(&msi, argp, sizeof(msi)))
Jan Kiszka07975ad2012-03-29 21:14:12 +02002905 goto out;
2906 r = kvm_send_userspace_msi(kvm, &msi);
2907 break;
2908 }
2909#endif
Christoffer Dall23d43cf2012-07-24 08:51:20 -04002910#ifdef __KVM_HAVE_IRQ_LINE
2911 case KVM_IRQ_LINE_STATUS:
2912 case KVM_IRQ_LINE: {
2913 struct kvm_irq_level irq_event;
2914
2915 r = -EFAULT;
Xiubo Li893bdbf2015-02-26 14:58:19 +08002916 if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
Christoffer Dall23d43cf2012-07-24 08:51:20 -04002917 goto out;
2918
Yang Zhangaa2fbe62013-04-11 19:21:40 +08002919 r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2920 ioctl == KVM_IRQ_LINE_STATUS);
Christoffer Dall23d43cf2012-07-24 08:51:20 -04002921 if (r)
2922 goto out;
2923
2924 r = -EFAULT;
2925 if (ioctl == KVM_IRQ_LINE_STATUS) {
Xiubo Li893bdbf2015-02-26 14:58:19 +08002926 if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
Christoffer Dall23d43cf2012-07-24 08:51:20 -04002927 goto out;
2928 }
2929
2930 r = 0;
2931 break;
2932 }
2933#endif
Alexander Grafaa8d5942013-04-15 21:12:53 +02002934#ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2935 case KVM_SET_GSI_ROUTING: {
2936 struct kvm_irq_routing routing;
2937 struct kvm_irq_routing __user *urouting;
2938 struct kvm_irq_routing_entry *entries;
2939
2940 r = -EFAULT;
2941 if (copy_from_user(&routing, argp, sizeof(routing)))
2942 goto out;
2943 r = -EINVAL;
2944 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2945 goto out;
2946 if (routing.flags)
2947 goto out;
2948 r = -ENOMEM;
2949 entries = vmalloc(routing.nr * sizeof(*entries));
2950 if (!entries)
2951 goto out;
2952 r = -EFAULT;
2953 urouting = argp;
2954 if (copy_from_user(entries, urouting->entries,
2955 routing.nr * sizeof(*entries)))
2956 goto out_free_irq_routing;
2957 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2958 routing.flags);
Xiubo Lia642a172015-02-26 14:58:20 +08002959out_free_irq_routing:
Alexander Grafaa8d5942013-04-15 21:12:53 +02002960 vfree(entries);
2961 break;
2962 }
2963#endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
Scott Wood852b6d52013-04-12 14:08:42 +00002964 case KVM_CREATE_DEVICE: {
2965 struct kvm_create_device cd;
2966
2967 r = -EFAULT;
2968 if (copy_from_user(&cd, argp, sizeof(cd)))
2969 goto out;
2970
2971 r = kvm_ioctl_create_device(kvm, &cd);
2972 if (r)
2973 goto out;
2974
2975 r = -EFAULT;
2976 if (copy_to_user(argp, &cd, sizeof(cd)))
2977 goto out;
2978
2979 r = 0;
2980 break;
2981 }
Alexander Graf92b591a2014-07-14 18:33:08 +02002982 case KVM_CHECK_EXTENSION:
2983 r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
2984 break;
Avi Kivityf17abe92007-02-21 19:28:04 +02002985 default:
Carsten Otte1fe779f2007-10-29 16:08:35 +01002986 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
Avi Kivityf17abe92007-02-21 19:28:04 +02002987 }
2988out:
2989 return r;
2990}
2991
Christian Borntraegerde8e5d72015-02-03 09:35:15 +01002992#ifdef CONFIG_KVM_COMPAT
Arnd Bergmann6ff58942009-10-22 14:19:27 +02002993struct compat_kvm_dirty_log {
2994 __u32 slot;
2995 __u32 padding1;
2996 union {
2997 compat_uptr_t dirty_bitmap; /* one bit per page */
2998 __u64 padding2;
2999 };
3000};
3001
3002static long kvm_vm_compat_ioctl(struct file *filp,
3003 unsigned int ioctl, unsigned long arg)
3004{
3005 struct kvm *kvm = filp->private_data;
3006 int r;
3007
3008 if (kvm->mm != current->mm)
3009 return -EIO;
3010 switch (ioctl) {
3011 case KVM_GET_DIRTY_LOG: {
3012 struct compat_kvm_dirty_log compat_log;
3013 struct kvm_dirty_log log;
3014
3015 r = -EFAULT;
3016 if (copy_from_user(&compat_log, (void __user *)arg,
3017 sizeof(compat_log)))
3018 goto out;
3019 log.slot = compat_log.slot;
3020 log.padding1 = compat_log.padding1;
3021 log.padding2 = compat_log.padding2;
3022 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
3023
3024 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
Arnd Bergmann6ff58942009-10-22 14:19:27 +02003025 break;
3026 }
3027 default:
3028 r = kvm_vm_ioctl(filp, ioctl, arg);
3029 }
3030
3031out:
3032 return r;
3033}
3034#endif
3035
Christian Borntraeger3d3aab12008-12-02 11:17:32 +01003036static struct file_operations kvm_vm_fops = {
Avi Kivityf17abe92007-02-21 19:28:04 +02003037 .release = kvm_vm_release,
3038 .unlocked_ioctl = kvm_vm_ioctl,
Christian Borntraegerde8e5d72015-02-03 09:35:15 +01003039#ifdef CONFIG_KVM_COMPAT
Arnd Bergmann6ff58942009-10-22 14:19:27 +02003040 .compat_ioctl = kvm_vm_compat_ioctl,
3041#endif
Arnd Bergmann6038f372010-08-15 18:52:59 +02003042 .llseek = noop_llseek,
Avi Kivityf17abe92007-02-21 19:28:04 +02003043};
3044
Carsten Ottee08b9632012-01-04 10:25:20 +01003045static int kvm_dev_ioctl_create_vm(unsigned long type)
Avi Kivityf17abe92007-02-21 19:28:04 +02003046{
Heiko Carstensaac87632010-10-27 17:22:10 +02003047 int r;
Avi Kivityf17abe92007-02-21 19:28:04 +02003048 struct kvm *kvm;
3049
Carsten Ottee08b9632012-01-04 10:25:20 +01003050 kvm = kvm_create_vm(type);
Avi Kivityd6d28162007-06-28 08:38:16 -04003051 if (IS_ERR(kvm))
3052 return PTR_ERR(kvm);
Takuya Yoshikawa6ce5a092010-03-15 22:13:30 +09003053#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
3054 r = kvm_coalesced_mmio_init(kvm);
3055 if (r < 0) {
3056 kvm_put_kvm(kvm);
3057 return r;
3058 }
3059#endif
Yann Droneaud24009b02013-08-24 22:14:07 +02003060 r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
Janosch Frank536a6f82016-05-18 13:26:23 +02003061 if (r < 0) {
Al Viro66c0b392008-04-19 20:33:56 +01003062 kvm_put_kvm(kvm);
Janosch Frank536a6f82016-05-18 13:26:23 +02003063 return r;
3064 }
3065
3066 if (kvm_create_vm_debugfs(kvm, r) < 0) {
3067 kvm_put_kvm(kvm);
3068 return -ENOMEM;
3069 }
Avi Kivityf17abe92007-02-21 19:28:04 +02003070
Heiko Carstensaac87632010-10-27 17:22:10 +02003071 return r;
Avi Kivityf17abe92007-02-21 19:28:04 +02003072}
3073
3074static long kvm_dev_ioctl(struct file *filp,
3075 unsigned int ioctl, unsigned long arg)
3076{
Avi Kivity07c45a32007-03-07 13:05:38 +02003077 long r = -EINVAL;
Avi Kivityf17abe92007-02-21 19:28:04 +02003078
3079 switch (ioctl) {
3080 case KVM_GET_API_VERSION:
Avi Kivityf0fe5102007-03-07 13:11:17 +02003081 if (arg)
3082 goto out;
Avi Kivityf17abe92007-02-21 19:28:04 +02003083 r = KVM_API_VERSION;
3084 break;
3085 case KVM_CREATE_VM:
Carsten Ottee08b9632012-01-04 10:25:20 +01003086 r = kvm_dev_ioctl_create_vm(arg);
Avi Kivityf17abe92007-02-21 19:28:04 +02003087 break;
Zhang Xiantao018d00d2007-11-15 23:07:47 +08003088 case KVM_CHECK_EXTENSION:
Alexander Graf784aa3d2014-07-14 18:27:35 +02003089 r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
Avi Kivity5d308f42007-03-01 17:56:20 +02003090 break;
Avi Kivity07c45a32007-03-07 13:05:38 +02003091 case KVM_GET_VCPU_MMAP_SIZE:
Avi Kivity07c45a32007-03-07 13:05:38 +02003092 if (arg)
3093 goto out;
Avi Kivityadb1ff42008-01-24 15:13:08 +02003094 r = PAGE_SIZE; /* struct kvm_run */
3095#ifdef CONFIG_X86
3096 r += PAGE_SIZE; /* pio data page */
3097#endif
Laurent Vivier5f94c172008-05-30 16:05:54 +02003098#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
3099 r += PAGE_SIZE; /* coalesced mmio ring page */
3100#endif
Avi Kivity07c45a32007-03-07 13:05:38 +02003101 break;
Feng(Eric) Liud4c9ff22008-04-10 08:47:53 -04003102 case KVM_TRACE_ENABLE:
3103 case KVM_TRACE_PAUSE:
3104 case KVM_TRACE_DISABLE:
Marcelo Tosatti2023a292009-06-18 11:47:28 -03003105 r = -EOPNOTSUPP;
Feng(Eric) Liud4c9ff22008-04-10 08:47:53 -04003106 break;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003107 default:
Carsten Otte043405e2007-10-10 17:16:19 +02003108 return kvm_arch_dev_ioctl(filp, ioctl, arg);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003109 }
3110out:
3111 return r;
3112}
3113
Avi Kivity6aa8b732006-12-10 02:21:36 -08003114static struct file_operations kvm_chardev_ops = {
Avi Kivity6aa8b732006-12-10 02:21:36 -08003115 .unlocked_ioctl = kvm_dev_ioctl,
3116 .compat_ioctl = kvm_dev_ioctl,
Arnd Bergmann6038f372010-08-15 18:52:59 +02003117 .llseek = noop_llseek,
Avi Kivity6aa8b732006-12-10 02:21:36 -08003118};
3119
3120static struct miscdevice kvm_dev = {
Avi Kivitybbe44322007-03-04 13:27:36 +02003121 KVM_MINOR,
Avi Kivity6aa8b732006-12-10 02:21:36 -08003122 "kvm",
3123 &kvm_chardev_ops,
3124};
3125
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09003126static void hardware_enable_nolock(void *junk)
Avi Kivity1b6c0162007-05-24 13:03:52 +03003127{
3128 int cpu = raw_smp_processor_id();
Alexander Graf10474ae2009-09-15 11:37:46 +02003129 int r;
Avi Kivity1b6c0162007-05-24 13:03:52 +03003130
Rusty Russell7f59f492008-12-07 21:25:45 +10303131 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
Avi Kivity1b6c0162007-05-24 13:03:52 +03003132 return;
Alexander Graf10474ae2009-09-15 11:37:46 +02003133
Rusty Russell7f59f492008-12-07 21:25:45 +10303134 cpumask_set_cpu(cpu, cpus_hardware_enabled);
Alexander Graf10474ae2009-09-15 11:37:46 +02003135
Radim Krčmář13a34e02014-08-28 15:13:03 +02003136 r = kvm_arch_hardware_enable();
Alexander Graf10474ae2009-09-15 11:37:46 +02003137
3138 if (r) {
3139 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3140 atomic_inc(&hardware_enable_failed);
Xiubo Li1170adc2015-02-26 14:58:26 +08003141 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
Alexander Graf10474ae2009-09-15 11:37:46 +02003142 }
Avi Kivity1b6c0162007-05-24 13:03:52 +03003143}
3144
Paolo Bonzini4fa92fb2013-09-10 12:57:17 +02003145static void hardware_enable(void)
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09003146{
Paolo Bonzini4a937f92013-09-10 12:58:35 +02003147 raw_spin_lock(&kvm_count_lock);
Paolo Bonzini4fa92fb2013-09-10 12:57:17 +02003148 if (kvm_usage_count)
3149 hardware_enable_nolock(NULL);
Paolo Bonzini4a937f92013-09-10 12:58:35 +02003150 raw_spin_unlock(&kvm_count_lock);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09003151}
3152
3153static void hardware_disable_nolock(void *junk)
Avi Kivity1b6c0162007-05-24 13:03:52 +03003154{
3155 int cpu = raw_smp_processor_id();
3156
Rusty Russell7f59f492008-12-07 21:25:45 +10303157 if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
Avi Kivity1b6c0162007-05-24 13:03:52 +03003158 return;
Rusty Russell7f59f492008-12-07 21:25:45 +10303159 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
Radim Krčmář13a34e02014-08-28 15:13:03 +02003160 kvm_arch_hardware_disable();
Avi Kivity1b6c0162007-05-24 13:03:52 +03003161}
3162
Paolo Bonzini4fa92fb2013-09-10 12:57:17 +02003163static void hardware_disable(void)
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09003164{
Paolo Bonzini4a937f92013-09-10 12:58:35 +02003165 raw_spin_lock(&kvm_count_lock);
Paolo Bonzini4fa92fb2013-09-10 12:57:17 +02003166 if (kvm_usage_count)
3167 hardware_disable_nolock(NULL);
Paolo Bonzini4a937f92013-09-10 12:58:35 +02003168 raw_spin_unlock(&kvm_count_lock);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09003169}
3170
Alexander Graf10474ae2009-09-15 11:37:46 +02003171static void hardware_disable_all_nolock(void)
3172{
3173 BUG_ON(!kvm_usage_count);
3174
3175 kvm_usage_count--;
3176 if (!kvm_usage_count)
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09003177 on_each_cpu(hardware_disable_nolock, NULL, 1);
Alexander Graf10474ae2009-09-15 11:37:46 +02003178}
3179
3180static void hardware_disable_all(void)
3181{
Paolo Bonzini4a937f92013-09-10 12:58:35 +02003182 raw_spin_lock(&kvm_count_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02003183 hardware_disable_all_nolock();
Paolo Bonzini4a937f92013-09-10 12:58:35 +02003184 raw_spin_unlock(&kvm_count_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02003185}
3186
3187static int hardware_enable_all(void)
3188{
3189 int r = 0;
3190
Paolo Bonzini4a937f92013-09-10 12:58:35 +02003191 raw_spin_lock(&kvm_count_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02003192
3193 kvm_usage_count++;
3194 if (kvm_usage_count == 1) {
3195 atomic_set(&hardware_enable_failed, 0);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09003196 on_each_cpu(hardware_enable_nolock, NULL, 1);
Alexander Graf10474ae2009-09-15 11:37:46 +02003197
3198 if (atomic_read(&hardware_enable_failed)) {
3199 hardware_disable_all_nolock();
3200 r = -EBUSY;
3201 }
3202 }
3203
Paolo Bonzini4a937f92013-09-10 12:58:35 +02003204 raw_spin_unlock(&kvm_count_lock);
Alexander Graf10474ae2009-09-15 11:37:46 +02003205
3206 return r;
3207}
3208
Avi Kivity774c47f2007-02-12 00:54:47 -08003209static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
3210 void *v)
3211{
Avi Kivity1a6f4d72007-11-11 18:37:32 +02003212 val &= ~CPU_TASKS_FROZEN;
Avi Kivity774c47f2007-02-12 00:54:47 -08003213 switch (val) {
Avi Kivitycec9ad22007-05-24 13:11:41 +03003214 case CPU_DYING:
Paolo Bonzini4fa92fb2013-09-10 12:57:17 +02003215 hardware_disable();
Avi Kivity6ec8a8562007-08-19 15:57:26 +03003216 break;
Zachary Amsdenda908f22010-08-19 22:07:27 -10003217 case CPU_STARTING:
Paolo Bonzini4fa92fb2013-09-10 12:57:17 +02003218 hardware_enable();
Avi Kivity774c47f2007-02-12 00:54:47 -08003219 break;
3220 }
3221 return NOTIFY_OK;
3222}
3223
Rusty Russell9a2b85c2007-07-17 23:17:55 +10003224static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
Mike Dayd77c26f2007-10-08 09:02:08 -04003225 void *v)
Rusty Russell9a2b85c2007-07-17 23:17:55 +10003226{
Sheng Yang8e1c1812009-04-29 11:09:04 +08003227 /*
3228 * Some (well, at least mine) BIOSes hang on reboot if
3229 * in vmx root mode.
3230 *
3231 * And Intel TXT required VMX off for all cpu when system shutdown.
3232 */
Xiubo Li1170adc2015-02-26 14:58:26 +08003233 pr_info("kvm: exiting hardware virtualization\n");
Sheng Yang8e1c1812009-04-29 11:09:04 +08003234 kvm_rebooting = true;
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09003235 on_each_cpu(hardware_disable_nolock, NULL, 1);
Rusty Russell9a2b85c2007-07-17 23:17:55 +10003236 return NOTIFY_OK;
3237}
3238
3239static struct notifier_block kvm_reboot_notifier = {
3240 .notifier_call = kvm_reboot,
3241 .priority = 0,
3242};
3243
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003244static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04003245{
3246 int i;
3247
3248 for (i = 0; i < bus->dev_count; i++) {
Sasha Levin743eeb02011-07-27 16:00:48 +03003249 struct kvm_io_device *pos = bus->range[i].dev;
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04003250
3251 kvm_iodevice_destructor(pos);
3252 }
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003253 kfree(bus);
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04003254}
3255
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02003256static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
Xiubo Li20e87b72015-02-26 14:58:25 +08003257 const struct kvm_io_range *r2)
Sasha Levin743eeb02011-07-27 16:00:48 +03003258{
Jason Wang8f4216c72015-09-15 14:41:57 +08003259 gpa_t addr1 = r1->addr;
3260 gpa_t addr2 = r2->addr;
3261
3262 if (addr1 < addr2)
Sasha Levin743eeb02011-07-27 16:00:48 +03003263 return -1;
Jason Wang8f4216c72015-09-15 14:41:57 +08003264
3265 /* If r2->len == 0, match the exact address. If r2->len != 0,
3266 * accept any overlapping write. Any order is acceptable for
3267 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3268 * we process all of them.
3269 */
3270 if (r2->len) {
3271 addr1 += r1->len;
3272 addr2 += r2->len;
3273 }
3274
3275 if (addr1 > addr2)
Sasha Levin743eeb02011-07-27 16:00:48 +03003276 return 1;
Jason Wang8f4216c72015-09-15 14:41:57 +08003277
Sasha Levin743eeb02011-07-27 16:00:48 +03003278 return 0;
3279}
3280
Paolo Bonzinia343c9b2013-07-16 13:03:29 +02003281static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
3282{
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02003283 return kvm_io_bus_cmp(p1, p2);
Paolo Bonzinia343c9b2013-07-16 13:03:29 +02003284}
3285
Geoff Levand39369f72013-04-05 19:20:30 +00003286static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
Sasha Levin743eeb02011-07-27 16:00:48 +03003287 gpa_t addr, int len)
3288{
Sasha Levin743eeb02011-07-27 16:00:48 +03003289 bus->range[bus->dev_count++] = (struct kvm_io_range) {
3290 .addr = addr,
3291 .len = len,
3292 .dev = dev,
3293 };
3294
3295 sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
3296 kvm_io_bus_sort_cmp, NULL);
3297
3298 return 0;
3299}
3300
Geoff Levand39369f72013-04-05 19:20:30 +00003301static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
Sasha Levin743eeb02011-07-27 16:00:48 +03003302 gpa_t addr, int len)
3303{
3304 struct kvm_io_range *range, key;
3305 int off;
3306
3307 key = (struct kvm_io_range) {
3308 .addr = addr,
3309 .len = len,
3310 };
3311
3312 range = bsearch(&key, bus->range, bus->dev_count,
3313 sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
3314 if (range == NULL)
3315 return -ENOENT;
3316
3317 off = range - bus->range;
3318
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02003319 while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
Sasha Levin743eeb02011-07-27 16:00:48 +03003320 off--;
3321
3322 return off;
3323}
3324
Nikolay Nikolaeve32edf42015-03-26 14:39:28 +00003325static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
Cornelia Huck126a5af2013-07-03 16:30:53 +02003326 struct kvm_io_range *range, const void *val)
3327{
3328 int idx;
3329
3330 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3331 if (idx < 0)
3332 return -EOPNOTSUPP;
3333
3334 while (idx < bus->dev_count &&
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02003335 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
Nikolay Nikolaeve32edf42015-03-26 14:39:28 +00003336 if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
Cornelia Huck126a5af2013-07-03 16:30:53 +02003337 range->len, val))
3338 return idx;
3339 idx++;
3340 }
3341
3342 return -EOPNOTSUPP;
3343}
3344
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03003345/* kvm_io_bus_write - called under kvm->slots_lock */
Nikolay Nikolaeve32edf42015-03-26 14:39:28 +00003346int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03003347 int len, const void *val)
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04003348{
Cornelia Huck126a5af2013-07-03 16:30:53 +02003349 struct kvm_io_bus *bus;
3350 struct kvm_io_range range;
3351 int r;
3352
3353 range = (struct kvm_io_range) {
3354 .addr = addr,
3355 .len = len,
3356 };
3357
Nikolay Nikolaeve32edf42015-03-26 14:39:28 +00003358 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3359 r = __kvm_io_bus_write(vcpu, bus, &range, val);
Cornelia Huck126a5af2013-07-03 16:30:53 +02003360 return r < 0 ? r : 0;
3361}
3362
3363/* kvm_io_bus_write_cookie - called under kvm->slots_lock */
Nikolay Nikolaeve32edf42015-03-26 14:39:28 +00003364int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3365 gpa_t addr, int len, const void *val, long cookie)
Cornelia Huck126a5af2013-07-03 16:30:53 +02003366{
Lai Jiangshan90d83dc2010-04-19 17:41:23 +08003367 struct kvm_io_bus *bus;
Sasha Levin743eeb02011-07-27 16:00:48 +03003368 struct kvm_io_range range;
3369
3370 range = (struct kvm_io_range) {
3371 .addr = addr,
3372 .len = len,
3373 };
Lai Jiangshan90d83dc2010-04-19 17:41:23 +08003374
Nikolay Nikolaeve32edf42015-03-26 14:39:28 +00003375 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
Cornelia Huck126a5af2013-07-03 16:30:53 +02003376
3377 /* First try the device referenced by cookie. */
3378 if ((cookie >= 0) && (cookie < bus->dev_count) &&
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02003379 (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
Nikolay Nikolaeve32edf42015-03-26 14:39:28 +00003380 if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
Cornelia Huck126a5af2013-07-03 16:30:53 +02003381 val))
3382 return cookie;
3383
3384 /*
3385 * cookie contained garbage; fall back to search and return the
3386 * correct cookie value.
3387 */
Nikolay Nikolaeve32edf42015-03-26 14:39:28 +00003388 return __kvm_io_bus_write(vcpu, bus, &range, val);
Cornelia Huck126a5af2013-07-03 16:30:53 +02003389}
3390
Nikolay Nikolaeve32edf42015-03-26 14:39:28 +00003391static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3392 struct kvm_io_range *range, void *val)
Cornelia Huck126a5af2013-07-03 16:30:53 +02003393{
3394 int idx;
3395
3396 idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
Sasha Levin743eeb02011-07-27 16:00:48 +03003397 if (idx < 0)
3398 return -EOPNOTSUPP;
3399
3400 while (idx < bus->dev_count &&
Paolo Bonzinic21fbff2013-08-27 15:41:41 +02003401 kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
Nikolay Nikolaeve32edf42015-03-26 14:39:28 +00003402 if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
Cornelia Huck126a5af2013-07-03 16:30:53 +02003403 range->len, val))
3404 return idx;
Sasha Levin743eeb02011-07-27 16:00:48 +03003405 idx++;
3406 }
3407
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03003408 return -EOPNOTSUPP;
3409}
Michael S. Tsirkin68c3b4d2014-03-31 21:50:44 +03003410EXPORT_SYMBOL_GPL(kvm_io_bus_write);
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04003411
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03003412/* kvm_io_bus_read - called under kvm->slots_lock */
Nikolay Nikolaeve32edf42015-03-26 14:39:28 +00003413int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003414 int len, void *val)
Michael S. Tsirkinbda90202009-06-29 22:24:32 +03003415{
Cornelia Huck126a5af2013-07-03 16:30:53 +02003416 struct kvm_io_bus *bus;
3417 struct kvm_io_range range;
3418 int r;
3419
3420 range = (struct kvm_io_range) {
3421 .addr = addr,
3422 .len = len,
3423 };
3424
Nikolay Nikolaeve32edf42015-03-26 14:39:28 +00003425 bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3426 r = __kvm_io_bus_read(vcpu, bus, &range, val);
Cornelia Huck126a5af2013-07-03 16:30:53 +02003427 return r < 0 ? r : 0;
3428}
3429
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04003430
Marcelo Tosatti79fac952009-12-23 14:35:26 -02003431/* Caller must hold slots_lock. */
Sasha Levin743eeb02011-07-27 16:00:48 +03003432int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3433 int len, struct kvm_io_device *dev)
Michael S. Tsirkin6c474692009-06-29 22:24:26 +03003434{
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003435 struct kvm_io_bus *new_bus, *bus;
Gregory Haskins090b7af2009-07-07 17:08:44 -04003436
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003437 bus = kvm->buses[bus_idx];
Amos Kong6ea34c92013-05-25 06:44:15 +08003438 /* exclude ioeventfd which is limited by maximum fd */
3439 if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
Gregory Haskins090b7af2009-07-07 17:08:44 -04003440 return -ENOSPC;
3441
Jason Wangd3febdd2015-08-25 17:05:46 +08003442 new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count + 1) *
Amos Konga13007162012-03-09 12:17:32 +08003443 sizeof(struct kvm_io_range)), GFP_KERNEL);
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003444 if (!new_bus)
3445 return -ENOMEM;
Amos Konga13007162012-03-09 12:17:32 +08003446 memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3447 sizeof(struct kvm_io_range)));
Sasha Levin743eeb02011-07-27 16:00:48 +03003448 kvm_io_bus_insert_dev(new_bus, dev, addr, len);
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003449 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3450 synchronize_srcu_expedited(&kvm->srcu);
3451 kfree(bus);
Gregory Haskins090b7af2009-07-07 17:08:44 -04003452
3453 return 0;
3454}
3455
Marcelo Tosatti79fac952009-12-23 14:35:26 -02003456/* Caller must hold slots_lock. */
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003457int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3458 struct kvm_io_device *dev)
Gregory Haskins090b7af2009-07-07 17:08:44 -04003459{
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003460 int i, r;
3461 struct kvm_io_bus *new_bus, *bus;
Michael S. Tsirkin6c474692009-06-29 22:24:26 +03003462
Sasha Levincdfca7b2011-12-04 19:36:28 +02003463 bus = kvm->buses[bus_idx];
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003464 r = -ENOENT;
Amos Konga13007162012-03-09 12:17:32 +08003465 for (i = 0; i < bus->dev_count; i++)
3466 if (bus->range[i].dev == dev) {
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003467 r = 0;
Gregory Haskins090b7af2009-07-07 17:08:44 -04003468 break;
3469 }
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003470
Amos Konga13007162012-03-09 12:17:32 +08003471 if (r)
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003472 return r;
Amos Konga13007162012-03-09 12:17:32 +08003473
Jason Wangd3febdd2015-08-25 17:05:46 +08003474 new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count - 1) *
Amos Konga13007162012-03-09 12:17:32 +08003475 sizeof(struct kvm_io_range)), GFP_KERNEL);
3476 if (!new_bus)
3477 return -ENOMEM;
3478
3479 memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3480 new_bus->dev_count--;
3481 memcpy(new_bus->range + i, bus->range + i + 1,
3482 (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
Marcelo Tosattie93f8a02009-12-23 14:35:24 -02003483
3484 rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3485 synchronize_srcu_expedited(&kvm->srcu);
3486 kfree(bus);
3487 return r;
Gregory Haskins2eeb2e92007-05-31 14:08:53 -04003488}
3489
Avi Kivity774c47f2007-02-12 00:54:47 -08003490static struct notifier_block kvm_cpu_notifier = {
3491 .notifier_call = kvm_cpu_hotplug,
Avi Kivity774c47f2007-02-12 00:54:47 -08003492};
3493
Janosch Frank536a6f82016-05-18 13:26:23 +02003494static int kvm_debugfs_open(struct inode *inode, struct file *file,
3495 int (*get)(void *, u64 *), int (*set)(void *, u64),
3496 const char *fmt)
3497{
3498 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3499 inode->i_private;
3500
3501 /* The debugfs files are a reference to the kvm struct which
3502 * is still valid when kvm_destroy_vm is called.
3503 * To avoid the race between open and the removal of the debugfs
3504 * directory we test against the users count.
3505 */
3506 if (!atomic_add_unless(&stat_data->kvm->users_count, 1, 0))
3507 return -ENOENT;
3508
3509 if (simple_attr_open(inode, file, get, set, fmt)) {
3510 kvm_put_kvm(stat_data->kvm);
3511 return -ENOMEM;
3512 }
3513
3514 return 0;
3515}
3516
3517static int kvm_debugfs_release(struct inode *inode, struct file *file)
3518{
3519 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3520 inode->i_private;
3521
3522 simple_attr_release(inode, file);
3523 kvm_put_kvm(stat_data->kvm);
3524
3525 return 0;
3526}
3527
3528static int vm_stat_get_per_vm(void *data, u64 *val)
3529{
3530 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3531
3532 *val = *(u32 *)((void *)stat_data->kvm + stat_data->offset);
3533
3534 return 0;
3535}
3536
3537static int vm_stat_get_per_vm_open(struct inode *inode, struct file *file)
3538{
3539 __simple_attr_check_format("%llu\n", 0ull);
3540 return kvm_debugfs_open(inode, file, vm_stat_get_per_vm,
3541 NULL, "%llu\n");
3542}
3543
3544static const struct file_operations vm_stat_get_per_vm_fops = {
3545 .owner = THIS_MODULE,
3546 .open = vm_stat_get_per_vm_open,
3547 .release = kvm_debugfs_release,
3548 .read = simple_attr_read,
3549 .write = simple_attr_write,
3550 .llseek = generic_file_llseek,
3551};
3552
3553static int vcpu_stat_get_per_vm(void *data, u64 *val)
3554{
3555 int i;
3556 struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3557 struct kvm_vcpu *vcpu;
3558
3559 *val = 0;
3560
3561 kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
3562 *val += *(u32 *)((void *)vcpu + stat_data->offset);
3563
3564 return 0;
3565}
3566
3567static int vcpu_stat_get_per_vm_open(struct inode *inode, struct file *file)
3568{
3569 __simple_attr_check_format("%llu\n", 0ull);
3570 return kvm_debugfs_open(inode, file, vcpu_stat_get_per_vm,
3571 NULL, "%llu\n");
3572}
3573
3574static const struct file_operations vcpu_stat_get_per_vm_fops = {
3575 .owner = THIS_MODULE,
3576 .open = vcpu_stat_get_per_vm_open,
3577 .release = kvm_debugfs_release,
3578 .read = simple_attr_read,
3579 .write = simple_attr_write,
3580 .llseek = generic_file_llseek,
3581};
3582
3583static const struct file_operations *stat_fops_per_vm[] = {
3584 [KVM_STAT_VCPU] = &vcpu_stat_get_per_vm_fops,
3585 [KVM_STAT_VM] = &vm_stat_get_per_vm_fops,
3586};
3587
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003588static int vm_stat_get(void *_offset, u64 *val)
Avi Kivityba1389b2007-11-18 16:24:12 +02003589{
3590 unsigned offset = (long)_offset;
Avi Kivityba1389b2007-11-18 16:24:12 +02003591 struct kvm *kvm;
Janosch Frank536a6f82016-05-18 13:26:23 +02003592 struct kvm_stat_data stat_tmp = {.offset = offset};
3593 u64 tmp_val;
Avi Kivityba1389b2007-11-18 16:24:12 +02003594
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003595 *val = 0;
Paolo Bonzini2f303b72013-09-25 13:53:07 +02003596 spin_lock(&kvm_lock);
Janosch Frank536a6f82016-05-18 13:26:23 +02003597 list_for_each_entry(kvm, &vm_list, vm_list) {
3598 stat_tmp.kvm = kvm;
3599 vm_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
3600 *val += tmp_val;
3601 }
Paolo Bonzini2f303b72013-09-25 13:53:07 +02003602 spin_unlock(&kvm_lock);
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003603 return 0;
Avi Kivityba1389b2007-11-18 16:24:12 +02003604}
3605
3606DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3607
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003608static int vcpu_stat_get(void *_offset, u64 *val)
Avi Kivity1165f5f2007-04-19 17:27:43 +03003609{
3610 unsigned offset = (long)_offset;
Avi Kivity1165f5f2007-04-19 17:27:43 +03003611 struct kvm *kvm;
Janosch Frank536a6f82016-05-18 13:26:23 +02003612 struct kvm_stat_data stat_tmp = {.offset = offset};
3613 u64 tmp_val;
Avi Kivity1165f5f2007-04-19 17:27:43 +03003614
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003615 *val = 0;
Paolo Bonzini2f303b72013-09-25 13:53:07 +02003616 spin_lock(&kvm_lock);
Janosch Frank536a6f82016-05-18 13:26:23 +02003617 list_for_each_entry(kvm, &vm_list, vm_list) {
3618 stat_tmp.kvm = kvm;
3619 vcpu_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
3620 *val += tmp_val;
3621 }
Paolo Bonzini2f303b72013-09-25 13:53:07 +02003622 spin_unlock(&kvm_lock);
Christoph Hellwig8b88b092008-02-08 04:20:26 -08003623 return 0;
Avi Kivity1165f5f2007-04-19 17:27:43 +03003624}
3625
Avi Kivityba1389b2007-11-18 16:24:12 +02003626DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3627
Alexey Dobriyan828c0952009-10-01 15:43:56 -07003628static const struct file_operations *stat_fops[] = {
Avi Kivityba1389b2007-11-18 16:24:12 +02003629 [KVM_STAT_VCPU] = &vcpu_stat_fops,
3630 [KVM_STAT_VM] = &vm_stat_fops,
3631};
Avi Kivity1165f5f2007-04-19 17:27:43 +03003632
Hamo4f69b682011-12-15 14:23:16 +08003633static int kvm_init_debug(void)
Avi Kivity6aa8b732006-12-10 02:21:36 -08003634{
Paolo Bonzini0c8eb042013-10-30 12:12:13 +01003635 int r = -EEXIST;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003636 struct kvm_stats_debugfs_item *p;
3637
Hollis Blanchard76f7c872008-04-15 16:05:42 -05003638 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
Hamo4f69b682011-12-15 14:23:16 +08003639 if (kvm_debugfs_dir == NULL)
3640 goto out;
3641
Janosch Frank536a6f82016-05-18 13:26:23 +02003642 kvm_debugfs_num_entries = 0;
3643 for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) {
Janosch Frank4bd33b52015-10-14 12:37:35 +02003644 if (!debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3645 (void *)(long)p->offset,
3646 stat_fops[p->kind]))
Hamo4f69b682011-12-15 14:23:16 +08003647 goto out_dir;
3648 }
3649
3650 return 0;
3651
3652out_dir:
3653 debugfs_remove_recursive(kvm_debugfs_dir);
3654out:
3655 return r;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003656}
3657
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003658static int kvm_suspend(void)
Avi Kivity59ae6c62007-02-12 00:54:48 -08003659{
Alexander Graf10474ae2009-09-15 11:37:46 +02003660 if (kvm_usage_count)
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09003661 hardware_disable_nolock(NULL);
Avi Kivity59ae6c62007-02-12 00:54:48 -08003662 return 0;
3663}
3664
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003665static void kvm_resume(void)
Avi Kivity59ae6c62007-02-12 00:54:48 -08003666{
Zachary Amsdenca84d1a2010-08-19 22:07:28 -10003667 if (kvm_usage_count) {
Paolo Bonzini4a937f92013-09-10 12:58:35 +02003668 WARN_ON(raw_spin_is_locked(&kvm_count_lock));
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09003669 hardware_enable_nolock(NULL);
Zachary Amsdenca84d1a2010-08-19 22:07:28 -10003670 }
Avi Kivity59ae6c62007-02-12 00:54:48 -08003671}
3672
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003673static struct syscore_ops kvm_syscore_ops = {
Avi Kivity59ae6c62007-02-12 00:54:48 -08003674 .suspend = kvm_suspend,
3675 .resume = kvm_resume,
3676};
3677
Avi Kivity15ad7142007-07-11 18:17:21 +03003678static inline
3679struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3680{
3681 return container_of(pn, struct kvm_vcpu, preempt_notifier);
3682}
3683
3684static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3685{
3686 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
Xiubo Lif95ef0c2015-02-26 14:58:23 +08003687
Raghavendra K T3a08a8f2013-03-04 23:32:07 +05303688 if (vcpu->preempted)
3689 vcpu->preempted = false;
Avi Kivity15ad7142007-07-11 18:17:21 +03003690
Radim Krčmáře790d9e2014-08-21 18:08:05 +02003691 kvm_arch_sched_in(vcpu, cpu);
3692
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003693 kvm_arch_vcpu_load(vcpu, cpu);
Avi Kivity15ad7142007-07-11 18:17:21 +03003694}
3695
3696static void kvm_sched_out(struct preempt_notifier *pn,
3697 struct task_struct *next)
3698{
3699 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3700
Raghavendra K T3a08a8f2013-03-04 23:32:07 +05303701 if (current->state == TASK_RUNNING)
3702 vcpu->preempted = true;
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003703 kvm_arch_vcpu_put(vcpu);
Avi Kivity15ad7142007-07-11 18:17:21 +03003704}
3705
Avi Kivity0ee75be2010-04-28 15:39:01 +03003706int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
Rusty Russellc16f8622007-07-30 21:12:19 +10003707 struct module *module)
Avi Kivity6aa8b732006-12-10 02:21:36 -08003708{
3709 int r;
Yang, Sheng002c7f72007-07-31 14:23:01 +03003710 int cpu;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003711
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08003712 r = kvm_arch_init(opaque);
3713 if (r)
Zhang Xiantaod23087842007-11-29 15:35:39 +08003714 goto out_fail;
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08003715
Asias He7dac16c2013-05-08 10:57:29 +08003716 /*
3717 * kvm_arch_init makes sure there's at most one caller
3718 * for architectures that support multiple implementations,
3719 * like intel and amd on x86.
3720 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3721 * conflicts in case kvm is already setup for another implementation.
3722 */
3723 r = kvm_irqfd_init();
3724 if (r)
3725 goto out_irqfd;
3726
Avi Kivity8437a612009-06-06 14:52:35 -07003727 if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
Rusty Russell7f59f492008-12-07 21:25:45 +10303728 r = -ENOMEM;
3729 goto out_free_0;
3730 }
3731
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003732 r = kvm_arch_hardware_setup();
Avi Kivity6aa8b732006-12-10 02:21:36 -08003733 if (r < 0)
Rusty Russell7f59f492008-12-07 21:25:45 +10303734 goto out_free_0a;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003735
Yang, Sheng002c7f72007-07-31 14:23:01 +03003736 for_each_online_cpu(cpu) {
3737 smp_call_function_single(cpu,
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003738 kvm_arch_check_processor_compat,
Jens Axboe8691e5a2008-06-06 11:18:06 +02003739 &r, 1);
Yang, Sheng002c7f72007-07-31 14:23:01 +03003740 if (r < 0)
Zhang Xiantaod23087842007-11-29 15:35:39 +08003741 goto out_free_1;
Yang, Sheng002c7f72007-07-31 14:23:01 +03003742 }
3743
Avi Kivity774c47f2007-02-12 00:54:47 -08003744 r = register_cpu_notifier(&kvm_cpu_notifier);
3745 if (r)
Zhang Xiantaod23087842007-11-29 15:35:39 +08003746 goto out_free_2;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003747 register_reboot_notifier(&kvm_reboot_notifier);
3748
Rusty Russellc16f8622007-07-30 21:12:19 +10003749 /* A kmem cache lets us meet the alignment requirements of fx_save. */
Avi Kivity0ee75be2010-04-28 15:39:01 +03003750 if (!vcpu_align)
3751 vcpu_align = __alignof__(struct kvm_vcpu);
3752 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
Joe Perches56919c52007-11-12 20:06:51 -08003753 0, NULL);
Rusty Russellc16f8622007-07-30 21:12:19 +10003754 if (!kvm_vcpu_cache) {
3755 r = -ENOMEM;
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003756 goto out_free_3;
Rusty Russellc16f8622007-07-30 21:12:19 +10003757 }
3758
Gleb Natapovaf585b92010-10-14 11:22:46 +02003759 r = kvm_async_pf_init();
3760 if (r)
3761 goto out_free;
3762
Avi Kivity6aa8b732006-12-10 02:21:36 -08003763 kvm_chardev_ops.owner = module;
Christian Borntraeger3d3aab12008-12-02 11:17:32 +01003764 kvm_vm_fops.owner = module;
3765 kvm_vcpu_fops.owner = module;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003766
3767 r = misc_register(&kvm_dev);
3768 if (r) {
Xiubo Li1170adc2015-02-26 14:58:26 +08003769 pr_err("kvm: misc device register failed\n");
Gleb Natapovaf585b92010-10-14 11:22:46 +02003770 goto out_unreg;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003771 }
3772
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003773 register_syscore_ops(&kvm_syscore_ops);
3774
Avi Kivity15ad7142007-07-11 18:17:21 +03003775 kvm_preempt_ops.sched_in = kvm_sched_in;
3776 kvm_preempt_ops.sched_out = kvm_sched_out;
3777
Hamo4f69b682011-12-15 14:23:16 +08003778 r = kvm_init_debug();
3779 if (r) {
Xiubo Li1170adc2015-02-26 14:58:26 +08003780 pr_err("kvm: create debugfs files failed\n");
Hamo4f69b682011-12-15 14:23:16 +08003781 goto out_undebugfs;
3782 }
Darrick J. Wong0ea4ed82009-10-14 16:21:00 -07003783
Paolo Bonzini3c3c29f2014-09-24 13:02:46 +02003784 r = kvm_vfio_ops_init();
3785 WARN_ON(r);
3786
Avi Kivityc7addb92007-09-16 18:58:32 +02003787 return 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003788
Hamo4f69b682011-12-15 14:23:16 +08003789out_undebugfs:
3790 unregister_syscore_ops(&kvm_syscore_ops);
Wei Yongjunafc2f792013-05-05 20:03:40 +08003791 misc_deregister(&kvm_dev);
Gleb Natapovaf585b92010-10-14 11:22:46 +02003792out_unreg:
3793 kvm_async_pf_deinit();
Avi Kivity6aa8b732006-12-10 02:21:36 -08003794out_free:
Rusty Russellc16f8622007-07-30 21:12:19 +10003795 kmem_cache_destroy(kvm_vcpu_cache);
Zhang Xiantaod23087842007-11-29 15:35:39 +08003796out_free_3:
Avi Kivity6aa8b732006-12-10 02:21:36 -08003797 unregister_reboot_notifier(&kvm_reboot_notifier);
Avi Kivity774c47f2007-02-12 00:54:47 -08003798 unregister_cpu_notifier(&kvm_cpu_notifier);
Zhang Xiantaod23087842007-11-29 15:35:39 +08003799out_free_2:
Zhang Xiantaod23087842007-11-29 15:35:39 +08003800out_free_1:
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003801 kvm_arch_hardware_unsetup();
Rusty Russell7f59f492008-12-07 21:25:45 +10303802out_free_0a:
3803 free_cpumask_var(cpus_hardware_enabled);
Zhang Xiantaod23087842007-11-29 15:35:39 +08003804out_free_0:
Cornelia Hucka0f155e2013-02-28 12:33:18 +01003805 kvm_irqfd_exit();
3806out_irqfd:
Asias He7dac16c2013-05-08 10:57:29 +08003807 kvm_arch_exit();
3808out_fail:
Avi Kivity6aa8b732006-12-10 02:21:36 -08003809 return r;
3810}
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08003811EXPORT_SYMBOL_GPL(kvm_init);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003812
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08003813void kvm_exit(void)
Avi Kivity6aa8b732006-12-10 02:21:36 -08003814{
Janosch Frank4bd33b52015-10-14 12:37:35 +02003815 debugfs_remove_recursive(kvm_debugfs_dir);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003816 misc_deregister(&kvm_dev);
Rusty Russellc16f8622007-07-30 21:12:19 +10003817 kmem_cache_destroy(kvm_vcpu_cache);
Gleb Natapovaf585b92010-10-14 11:22:46 +02003818 kvm_async_pf_deinit();
Rafael J. Wysockifb3600c2011-03-23 22:16:23 +01003819 unregister_syscore_ops(&kvm_syscore_ops);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003820 unregister_reboot_notifier(&kvm_reboot_notifier);
Avi Kivity59ae6c62007-02-12 00:54:48 -08003821 unregister_cpu_notifier(&kvm_cpu_notifier);
Takuya Yoshikawa75b71272010-11-16 17:37:41 +09003822 on_each_cpu(hardware_disable_nolock, NULL, 1);
Zhang Xiantaoe9b11c12007-11-14 20:38:21 +08003823 kvm_arch_hardware_unsetup();
Zhang Xiantaof8c16bb2007-11-14 20:40:21 +08003824 kvm_arch_exit();
Cornelia Hucka0f155e2013-02-28 12:33:18 +01003825 kvm_irqfd_exit();
Rusty Russell7f59f492008-12-07 21:25:45 +10303826 free_cpumask_var(cpus_hardware_enabled);
Wanpeng Li571ee1b2014-10-09 18:30:08 +08003827 kvm_vfio_ops_exit();
Avi Kivity6aa8b732006-12-10 02:21:36 -08003828}
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08003829EXPORT_SYMBOL_GPL(kvm_exit);