Thomas Gleixner | 20c8ccb | 2019-06-04 10:11:32 +0200 | [diff] [blame] | 1 | // SPDX-License-Identifier: GPL-2.0-only |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 2 | /* |
| 3 | * Kernel-based Virtual Machine driver for Linux |
| 4 | * |
| 5 | * This module enables machines with Intel VT-x extensions to run virtual |
| 6 | * machines without emulation or binary translation. |
| 7 | * |
| 8 | * Copyright (C) 2006 Qumranet, Inc. |
Nicolas Kaiser | 9611c18 | 2010-10-06 14:23:22 +0200 | [diff] [blame] | 9 | * Copyright 2010 Red Hat, Inc. and/or its affiliates. |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 10 | * |
| 11 | * Authors: |
| 12 | * Avi Kivity <avi@qumranet.com> |
| 13 | * Yaniv Kamay <yaniv@qumranet.com> |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 14 | */ |
| 15 | |
Andre Przywara | af669ac | 2015-03-26 14:39:29 +0000 | [diff] [blame] | 16 | #include <kvm/iodev.h> |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 17 | |
Avi Kivity | edf8841 | 2007-12-16 11:02:48 +0200 | [diff] [blame] | 18 | #include <linux/kvm_host.h> |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 19 | #include <linux/kvm.h> |
| 20 | #include <linux/module.h> |
| 21 | #include <linux/errno.h> |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 22 | #include <linux/percpu.h> |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 23 | #include <linux/mm.h> |
| 24 | #include <linux/miscdevice.h> |
| 25 | #include <linux/vmalloc.h> |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 26 | #include <linux/reboot.h> |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 27 | #include <linux/debugfs.h> |
| 28 | #include <linux/highmem.h> |
| 29 | #include <linux/file.h> |
Rafael J. Wysocki | fb3600c | 2011-03-23 22:16:23 +0100 | [diff] [blame] | 30 | #include <linux/syscore_ops.h> |
Avi Kivity | 774c47f | 2007-02-12 00:54:47 -0800 | [diff] [blame] | 31 | #include <linux/cpu.h> |
Ingo Molnar | 174cd4b | 2017-02-02 19:15:33 +0100 | [diff] [blame] | 32 | #include <linux/sched/signal.h> |
Ingo Molnar | 6e84f31 | 2017-02-08 18:51:29 +0100 | [diff] [blame] | 33 | #include <linux/sched/mm.h> |
Ingo Molnar | 03441a3 | 2017-02-08 18:51:35 +0100 | [diff] [blame] | 34 | #include <linux/sched/stat.h> |
Avi Kivity | d9e368d | 2007-06-07 19:18:30 +0300 | [diff] [blame] | 35 | #include <linux/cpumask.h> |
| 36 | #include <linux/smp.h> |
Avi Kivity | d6d2816 | 2007-06-28 08:38:16 -0400 | [diff] [blame] | 37 | #include <linux/anon_inodes.h> |
Avi Kivity | 04d2cc7 | 2007-09-10 18:10:54 +0300 | [diff] [blame] | 38 | #include <linux/profile.h> |
Anthony Liguori | 7aa81cc | 2007-09-17 14:57:50 -0500 | [diff] [blame] | 39 | #include <linux/kvm_para.h> |
Izik Eidus | 6fc138d | 2007-10-09 19:20:39 +0200 | [diff] [blame] | 40 | #include <linux/pagemap.h> |
Anthony Liguori | 8d4e128 | 2007-10-18 09:59:34 -0500 | [diff] [blame] | 41 | #include <linux/mman.h> |
Anthony Liguori | 35149e2 | 2008-04-02 14:46:56 -0500 | [diff] [blame] | 42 | #include <linux/swap.h> |
Sheng Yang | e56d532 | 2009-03-12 21:45:39 +0800 | [diff] [blame] | 43 | #include <linux/bitops.h> |
Marcelo Tosatti | 547de29 | 2009-05-07 17:55:13 -0300 | [diff] [blame] | 44 | #include <linux/spinlock.h> |
Arnd Bergmann | 6ff5894 | 2009-10-22 14:19:27 +0200 | [diff] [blame] | 45 | #include <linux/compat.h> |
Marcelo Tosatti | bc6678a | 2009-12-23 14:35:21 -0200 | [diff] [blame] | 46 | #include <linux/srcu.h> |
Joerg Roedel | 8f0b1ab | 2010-01-28 12:37:56 +0100 | [diff] [blame] | 47 | #include <linux/hugetlb.h> |
Tejun Heo | 5a0e3ad | 2010-03-24 17:04:11 +0900 | [diff] [blame] | 48 | #include <linux/slab.h> |
Sasha Levin | 743eeb0 | 2011-07-27 16:00:48 +0300 | [diff] [blame] | 49 | #include <linux/sort.h> |
| 50 | #include <linux/bsearch.h> |
Paolo Bonzini | c011d23 | 2019-05-17 14:08:53 +0200 | [diff] [blame] | 51 | #include <linux/io.h> |
Wanpeng Li | 2eb06c3 | 2019-05-17 16:49:49 +0800 | [diff] [blame] | 52 | #include <linux/lockdep.h> |
Junaid Shahid | c57c804 | 2019-11-04 12:22:02 +0100 | [diff] [blame] | 53 | #include <linux/kthread.h> |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 54 | |
Avi Kivity | e495606 | 2007-06-28 14:15:57 -0400 | [diff] [blame] | 55 | #include <asm/processor.h> |
David Matlack | 2ea75be | 2014-09-19 16:03:25 -0700 | [diff] [blame] | 56 | #include <asm/ioctl.h> |
Linus Torvalds | 7c0f6ba | 2016-12-24 11:46:01 -0800 | [diff] [blame] | 57 | #include <linux/uaccess.h> |
Izik Eidus | 3e021bf | 2007-11-19 11:16:57 +0200 | [diff] [blame] | 58 | #include <asm/pgtable.h> |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 59 | |
Laurent Vivier | 5f94c17 | 2008-05-30 16:05:54 +0200 | [diff] [blame] | 60 | #include "coalesced_mmio.h" |
Gleb Natapov | af585b9 | 2010-10-14 11:22:46 +0200 | [diff] [blame] | 61 | #include "async_pf.h" |
Paolo Bonzini | 3c3c29f | 2014-09-24 13:02:46 +0200 | [diff] [blame] | 62 | #include "vfio.h" |
Laurent Vivier | 5f94c17 | 2008-05-30 16:05:54 +0200 | [diff] [blame] | 63 | |
Marcelo Tosatti | 229456f | 2009-06-17 09:22:14 -0300 | [diff] [blame] | 64 | #define CREATE_TRACE_POINTS |
| 65 | #include <trace/events/kvm.h> |
| 66 | |
Janosch Frank | 536a6f8 | 2016-05-18 13:26:23 +0200 | [diff] [blame] | 67 | /* Worst case buffer size needed for holding an integer. */ |
| 68 | #define ITOA_MAX_LEN 12 |
| 69 | |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 70 | MODULE_AUTHOR("Qumranet"); |
| 71 | MODULE_LICENSE("GPL"); |
| 72 | |
David Hildenbrand | 920552b | 2015-09-18 12:34:53 +0200 | [diff] [blame] | 73 | /* Architectures should define their poll value according to the halt latency */ |
Suraj Jitindar Singh | ec76d81 | 2016-10-14 11:53:19 +1100 | [diff] [blame] | 74 | unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT; |
Roman Storozhenko | 039c5d1 | 2017-06-27 12:51:18 +0300 | [diff] [blame] | 75 | module_param(halt_poll_ns, uint, 0644); |
Suraj Jitindar Singh | ec76d81 | 2016-10-14 11:53:19 +1100 | [diff] [blame] | 76 | EXPORT_SYMBOL_GPL(halt_poll_ns); |
Paolo Bonzini | f781951 | 2015-02-04 18:20:58 +0100 | [diff] [blame] | 77 | |
Wanpeng Li | aca6ff2 | 2015-09-03 22:07:38 +0800 | [diff] [blame] | 78 | /* Default doubles per-vcpu halt_poll_ns. */ |
Suraj Jitindar Singh | ec76d81 | 2016-10-14 11:53:19 +1100 | [diff] [blame] | 79 | unsigned int halt_poll_ns_grow = 2; |
Roman Storozhenko | 039c5d1 | 2017-06-27 12:51:18 +0300 | [diff] [blame] | 80 | module_param(halt_poll_ns_grow, uint, 0644); |
Suraj Jitindar Singh | ec76d81 | 2016-10-14 11:53:19 +1100 | [diff] [blame] | 81 | EXPORT_SYMBOL_GPL(halt_poll_ns_grow); |
Wanpeng Li | aca6ff2 | 2015-09-03 22:07:38 +0800 | [diff] [blame] | 82 | |
Nir Weiner | 49113d3 | 2019-01-27 12:17:15 +0200 | [diff] [blame] | 83 | /* The start value to grow halt_poll_ns from */ |
| 84 | unsigned int halt_poll_ns_grow_start = 10000; /* 10us */ |
| 85 | module_param(halt_poll_ns_grow_start, uint, 0644); |
| 86 | EXPORT_SYMBOL_GPL(halt_poll_ns_grow_start); |
| 87 | |
Wanpeng Li | aca6ff2 | 2015-09-03 22:07:38 +0800 | [diff] [blame] | 88 | /* Default resets per-vcpu halt_poll_ns . */ |
Suraj Jitindar Singh | ec76d81 | 2016-10-14 11:53:19 +1100 | [diff] [blame] | 89 | unsigned int halt_poll_ns_shrink; |
Roman Storozhenko | 039c5d1 | 2017-06-27 12:51:18 +0300 | [diff] [blame] | 90 | module_param(halt_poll_ns_shrink, uint, 0644); |
Suraj Jitindar Singh | ec76d81 | 2016-10-14 11:53:19 +1100 | [diff] [blame] | 91 | EXPORT_SYMBOL_GPL(halt_poll_ns_shrink); |
Wanpeng Li | aca6ff2 | 2015-09-03 22:07:38 +0800 | [diff] [blame] | 92 | |
Marcelo Tosatti | fa40a82 | 2009-06-04 15:08:24 -0300 | [diff] [blame] | 93 | /* |
| 94 | * Ordering of locks: |
| 95 | * |
Xiubo Li | b7d409d | 2015-02-26 14:58:24 +0800 | [diff] [blame] | 96 | * kvm->lock --> kvm->slots_lock --> kvm->irq_lock |
Marcelo Tosatti | fa40a82 | 2009-06-04 15:08:24 -0300 | [diff] [blame] | 97 | */ |
| 98 | |
Junaid Shahid | 0d9ce16 | 2019-01-03 17:14:28 -0800 | [diff] [blame] | 99 | DEFINE_MUTEX(kvm_lock); |
Paolo Bonzini | 4a937f9 | 2013-09-10 12:58:35 +0200 | [diff] [blame] | 100 | static DEFINE_RAW_SPINLOCK(kvm_count_lock); |
Zhang Xiantao | e9b11c1 | 2007-11-14 20:38:21 +0800 | [diff] [blame] | 101 | LIST_HEAD(vm_list); |
Avi Kivity | 133de90 | 2007-02-12 00:54:44 -0800 | [diff] [blame] | 102 | |
Rusty Russell | 7f59f49 | 2008-12-07 21:25:45 +1030 | [diff] [blame] | 103 | static cpumask_var_t cpus_hardware_enabled; |
Xiubo Li | f4fee93 | 2015-02-26 14:58:21 +0800 | [diff] [blame] | 104 | static int kvm_usage_count; |
Alexander Graf | 10474ae | 2009-09-15 11:37:46 +0200 | [diff] [blame] | 105 | static atomic_t hardware_enable_failed; |
Avi Kivity | 1b6c016 | 2007-05-24 13:03:52 +0300 | [diff] [blame] | 106 | |
Rusty Russell | c16f862 | 2007-07-30 21:12:19 +1000 | [diff] [blame] | 107 | struct kmem_cache *kvm_vcpu_cache; |
| 108 | EXPORT_SYMBOL_GPL(kvm_vcpu_cache); |
Avi Kivity | 1165f5f | 2007-04-19 17:27:43 +0300 | [diff] [blame] | 109 | |
Avi Kivity | 15ad714 | 2007-07-11 18:17:21 +0300 | [diff] [blame] | 110 | static __read_mostly struct preempt_ops kvm_preempt_ops; |
| 111 | |
Hollis Blanchard | 76f7c87 | 2008-04-15 16:05:42 -0500 | [diff] [blame] | 112 | struct dentry *kvm_debugfs_dir; |
Paul Mackerras | e23a808 | 2015-03-28 14:21:01 +1100 | [diff] [blame] | 113 | EXPORT_SYMBOL_GPL(kvm_debugfs_dir); |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 114 | |
Janosch Frank | 536a6f8 | 2016-05-18 13:26:23 +0200 | [diff] [blame] | 115 | static int kvm_debugfs_num_entries; |
| 116 | static const struct file_operations *stat_fops_per_vm[]; |
| 117 | |
Avi Kivity | bccf215 | 2007-02-21 18:04:26 +0200 | [diff] [blame] | 118 | static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl, |
| 119 | unsigned long arg); |
Christian Borntraeger | de8e5d7 | 2015-02-03 09:35:15 +0100 | [diff] [blame] | 120 | #ifdef CONFIG_KVM_COMPAT |
Alexander Graf | 1dda606 | 2011-06-08 02:45:37 +0200 | [diff] [blame] | 121 | static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl, |
| 122 | unsigned long arg); |
Marc Zyngier | 7ddfd3e | 2018-06-17 10:16:21 +0100 | [diff] [blame] | 123 | #define KVM_COMPAT(c) .compat_ioctl = (c) |
| 124 | #else |
Marc Zyngier | 9cb09e7 | 2019-11-14 13:17:39 +0000 | [diff] [blame] | 125 | /* |
| 126 | * For architectures that don't implement a compat infrastructure, |
| 127 | * adopt a double line of defense: |
| 128 | * - Prevent a compat task from opening /dev/kvm |
| 129 | * - If the open has been done by a 64bit task, and the KVM fd |
| 130 | * passed to a compat task, let the ioctls fail. |
| 131 | */ |
Marc Zyngier | 7ddfd3e | 2018-06-17 10:16:21 +0100 | [diff] [blame] | 132 | static long kvm_no_compat_ioctl(struct file *file, unsigned int ioctl, |
| 133 | unsigned long arg) { return -EINVAL; } |
Marc Zyngier | b9876e6 | 2019-11-13 16:05:23 +0000 | [diff] [blame] | 134 | |
| 135 | static int kvm_no_compat_open(struct inode *inode, struct file *file) |
| 136 | { |
| 137 | return is_compat_task() ? -ENODEV : 0; |
| 138 | } |
| 139 | #define KVM_COMPAT(c) .compat_ioctl = kvm_no_compat_ioctl, \ |
| 140 | .open = kvm_no_compat_open |
Alexander Graf | 1dda606 | 2011-06-08 02:45:37 +0200 | [diff] [blame] | 141 | #endif |
Alexander Graf | 10474ae | 2009-09-15 11:37:46 +0200 | [diff] [blame] | 142 | static int hardware_enable_all(void); |
| 143 | static void hardware_disable_all(void); |
Avi Kivity | bccf215 | 2007-02-21 18:04:26 +0200 | [diff] [blame] | 144 | |
Marcelo Tosatti | e93f8a0 | 2009-12-23 14:35:24 -0200 | [diff] [blame] | 145 | static void kvm_io_bus_destroy(struct kvm_io_bus *bus); |
Stephen Hemminger | 7940876 | 2013-12-29 12:12:29 -0800 | [diff] [blame] | 146 | |
Paolo Bonzini | bc009e4 | 2015-05-26 12:43:41 +0200 | [diff] [blame] | 147 | static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, gfn_t gfn); |
Marcelo Tosatti | e93f8a0 | 2009-12-23 14:35:24 -0200 | [diff] [blame] | 148 | |
Andi Kleen | 5248013 | 2014-02-08 08:51:57 +0100 | [diff] [blame] | 149 | __visible bool kvm_rebooting; |
Avi Kivity | b7c4145 | 2010-12-02 17:52:50 +0200 | [diff] [blame] | 150 | EXPORT_SYMBOL_GPL(kvm_rebooting); |
Avi Kivity | 4ecac3f | 2008-05-13 13:23:38 +0300 | [diff] [blame] | 151 | |
Marcelo Tosatti | 54dee99 | 2009-06-11 12:07:44 -0300 | [diff] [blame] | 152 | static bool largepages_enabled = true; |
| 153 | |
Claudio Imbrenda | 286de8f | 2017-07-12 17:56:44 +0200 | [diff] [blame] | 154 | #define KVM_EVENT_CREATE_VM 0 |
| 155 | #define KVM_EVENT_DESTROY_VM 1 |
| 156 | static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm); |
| 157 | static unsigned long long kvm_createvm_count; |
| 158 | static unsigned long long kvm_active_vms; |
| 159 | |
Michal Hocko | 93065ac | 2018-08-21 21:52:33 -0700 | [diff] [blame] | 160 | __weak int kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm, |
| 161 | unsigned long start, unsigned long end, bool blockable) |
Radim Krčmář | b1394e7 | 2017-11-30 19:05:45 +0100 | [diff] [blame] | 162 | { |
Michal Hocko | 93065ac | 2018-08-21 21:52:33 -0700 | [diff] [blame] | 163 | return 0; |
Radim Krčmář | b1394e7 | 2017-11-30 19:05:45 +0100 | [diff] [blame] | 164 | } |
| 165 | |
Sean Christopherson | a78986a | 2019-11-11 14:12:27 -0800 | [diff] [blame] | 166 | bool kvm_is_zone_device_pfn(kvm_pfn_t pfn) |
| 167 | { |
| 168 | /* |
| 169 | * The metadata used by is_zone_device_page() to determine whether or |
| 170 | * not a page is ZONE_DEVICE is guaranteed to be valid if and only if |
| 171 | * the device has been pinned, e.g. by get_user_pages(). WARN if the |
| 172 | * page_count() is zero to help detect bad usage of this helper. |
| 173 | */ |
| 174 | if (!pfn_valid(pfn) || WARN_ON_ONCE(!page_count(pfn_to_page(pfn)))) |
| 175 | return false; |
| 176 | |
| 177 | return is_zone_device_page(pfn_to_page(pfn)); |
| 178 | } |
| 179 | |
Dan Williams | ba049e9 | 2016-01-15 16:56:11 -0800 | [diff] [blame] | 180 | bool kvm_is_reserved_pfn(kvm_pfn_t pfn) |
Ben-Ami Yassour | cbff90a | 2008-07-28 19:26:24 +0300 | [diff] [blame] | 181 | { |
Sean Christopherson | a78986a | 2019-11-11 14:12:27 -0800 | [diff] [blame] | 182 | /* |
| 183 | * ZONE_DEVICE pages currently set PG_reserved, but from a refcounting |
| 184 | * perspective they are "normal" pages, albeit with slightly different |
| 185 | * usage rules. |
| 186 | */ |
Andrea Arcangeli | 11feeb4 | 2013-07-25 03:04:38 +0200 | [diff] [blame] | 187 | if (pfn_valid(pfn)) |
Sean Christopherson | a78986a | 2019-11-11 14:12:27 -0800 | [diff] [blame] | 188 | return PageReserved(pfn_to_page(pfn)) && |
| 189 | !kvm_is_zone_device_pfn(pfn); |
Ben-Ami Yassour | cbff90a | 2008-07-28 19:26:24 +0300 | [diff] [blame] | 190 | |
| 191 | return true; |
| 192 | } |
| 193 | |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 194 | /* |
| 195 | * Switches to specified vcpu, until a matching vcpu_put() |
| 196 | */ |
Christoffer Dall | ec7660c | 2017-12-04 21:35:23 +0100 | [diff] [blame] | 197 | void vcpu_load(struct kvm_vcpu *vcpu) |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 198 | { |
Christoffer Dall | ec7660c | 2017-12-04 21:35:23 +0100 | [diff] [blame] | 199 | int cpu = get_cpu(); |
Avi Kivity | 15ad714 | 2007-07-11 18:17:21 +0300 | [diff] [blame] | 200 | preempt_notifier_register(&vcpu->preempt_notifier); |
Carsten Otte | 313a3dc | 2007-10-11 19:16:52 +0200 | [diff] [blame] | 201 | kvm_arch_vcpu_load(vcpu, cpu); |
Avi Kivity | 15ad714 | 2007-07-11 18:17:21 +0300 | [diff] [blame] | 202 | put_cpu(); |
Avi Kivity | bccf215 | 2007-02-21 18:04:26 +0200 | [diff] [blame] | 203 | } |
Jim Mattson | 2f1fe81 | 2016-07-08 15:36:06 -0700 | [diff] [blame] | 204 | EXPORT_SYMBOL_GPL(vcpu_load); |
Avi Kivity | bccf215 | 2007-02-21 18:04:26 +0200 | [diff] [blame] | 205 | |
Carsten Otte | 313a3dc | 2007-10-11 19:16:52 +0200 | [diff] [blame] | 206 | void vcpu_put(struct kvm_vcpu *vcpu) |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 207 | { |
Avi Kivity | 15ad714 | 2007-07-11 18:17:21 +0300 | [diff] [blame] | 208 | preempt_disable(); |
Carsten Otte | 313a3dc | 2007-10-11 19:16:52 +0200 | [diff] [blame] | 209 | kvm_arch_vcpu_put(vcpu); |
Avi Kivity | 15ad714 | 2007-07-11 18:17:21 +0300 | [diff] [blame] | 210 | preempt_notifier_unregister(&vcpu->preempt_notifier); |
| 211 | preempt_enable(); |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 212 | } |
Jim Mattson | 2f1fe81 | 2016-07-08 15:36:06 -0700 | [diff] [blame] | 213 | EXPORT_SYMBOL_GPL(vcpu_put); |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 214 | |
Paolo Bonzini | 7a97cec | 2017-04-27 14:33:43 +0200 | [diff] [blame] | 215 | /* TODO: merge with kvm_arch_vcpu_should_kick */ |
| 216 | static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req) |
| 217 | { |
| 218 | int mode = kvm_vcpu_exiting_guest_mode(vcpu); |
| 219 | |
| 220 | /* |
| 221 | * We need to wait for the VCPU to reenable interrupts and get out of |
| 222 | * READING_SHADOW_PAGE_TABLES mode. |
| 223 | */ |
| 224 | if (req & KVM_REQUEST_WAIT) |
| 225 | return mode != OUTSIDE_GUEST_MODE; |
| 226 | |
| 227 | /* |
| 228 | * Need to kick a running VCPU, but otherwise there is nothing to do. |
| 229 | */ |
| 230 | return mode == IN_GUEST_MODE; |
| 231 | } |
| 232 | |
Avi Kivity | d9e368d | 2007-06-07 19:18:30 +0300 | [diff] [blame] | 233 | static void ack_flush(void *_completed) |
| 234 | { |
Avi Kivity | d9e368d | 2007-06-07 19:18:30 +0300 | [diff] [blame] | 235 | } |
| 236 | |
Paolo Bonzini | b49defe | 2017-06-30 13:25:45 +0200 | [diff] [blame] | 237 | static inline bool kvm_kick_many_cpus(const struct cpumask *cpus, bool wait) |
| 238 | { |
| 239 | if (unlikely(!cpus)) |
| 240 | cpus = cpu_online_mask; |
| 241 | |
| 242 | if (cpumask_empty(cpus)) |
| 243 | return false; |
| 244 | |
| 245 | smp_call_function_many(cpus, ack_flush, NULL, wait); |
| 246 | return true; |
| 247 | } |
| 248 | |
Vitaly Kuznetsov | 7053df4 | 2018-05-16 17:21:28 +0200 | [diff] [blame] | 249 | bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req, |
| 250 | unsigned long *vcpu_bitmap, cpumask_var_t tmp) |
Avi Kivity | d9e368d | 2007-06-07 19:18:30 +0300 | [diff] [blame] | 251 | { |
Avi Kivity | 597a5f5 | 2008-07-20 14:24:22 +0300 | [diff] [blame] | 252 | int i, cpu, me; |
Avi Kivity | d9e368d | 2007-06-07 19:18:30 +0300 | [diff] [blame] | 253 | struct kvm_vcpu *vcpu; |
Vitaly Kuznetsov | 7053df4 | 2018-05-16 17:21:28 +0200 | [diff] [blame] | 254 | bool called; |
Rusty Russell | 6ef7a1b | 2008-12-08 20:28:04 +1030 | [diff] [blame] | 255 | |
Xiao Guangrong | 3cba413 | 2011-01-12 15:41:22 +0800 | [diff] [blame] | 256 | me = get_cpu(); |
Vitaly Kuznetsov | 7053df4 | 2018-05-16 17:21:28 +0200 | [diff] [blame] | 257 | |
Gleb Natapov | 988a2ca | 2009-06-09 15:56:29 +0300 | [diff] [blame] | 258 | kvm_for_each_vcpu(i, vcpu, kvm) { |
Vitaly Kuznetsov | a812297 | 2018-08-22 12:18:29 +0200 | [diff] [blame] | 259 | if (vcpu_bitmap && !test_bit(i, vcpu_bitmap)) |
Vitaly Kuznetsov | 7053df4 | 2018-05-16 17:21:28 +0200 | [diff] [blame] | 260 | continue; |
| 261 | |
Xiao Guangrong | 3cba413 | 2011-01-12 15:41:22 +0800 | [diff] [blame] | 262 | kvm_make_request(req, vcpu); |
Avi Kivity | d9e368d | 2007-06-07 19:18:30 +0300 | [diff] [blame] | 263 | cpu = vcpu->cpu; |
Xiao Guangrong | 6b7e2d0 | 2011-01-12 15:40:31 +0800 | [diff] [blame] | 264 | |
Radim Krčmář | 178f02f | 2017-04-26 22:32:26 +0200 | [diff] [blame] | 265 | if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu)) |
| 266 | continue; |
Radim Krčmář | 6c6e836 | 2017-04-26 22:32:23 +0200 | [diff] [blame] | 267 | |
Vitaly Kuznetsov | 7053df4 | 2018-05-16 17:21:28 +0200 | [diff] [blame] | 268 | if (tmp != NULL && cpu != -1 && cpu != me && |
Paolo Bonzini | 7a97cec | 2017-04-27 14:33:43 +0200 | [diff] [blame] | 269 | kvm_request_needs_ipi(vcpu, req)) |
Vitaly Kuznetsov | 7053df4 | 2018-05-16 17:21:28 +0200 | [diff] [blame] | 270 | __cpumask_set_cpu(cpu, tmp); |
Avi Kivity | d9e368d | 2007-06-07 19:18:30 +0300 | [diff] [blame] | 271 | } |
Vitaly Kuznetsov | 7053df4 | 2018-05-16 17:21:28 +0200 | [diff] [blame] | 272 | |
| 273 | called = kvm_kick_many_cpus(tmp, !!(req & KVM_REQUEST_WAIT)); |
Xiao Guangrong | 3cba413 | 2011-01-12 15:41:22 +0800 | [diff] [blame] | 274 | put_cpu(); |
Vitaly Kuznetsov | 7053df4 | 2018-05-16 17:21:28 +0200 | [diff] [blame] | 275 | |
| 276 | return called; |
| 277 | } |
| 278 | |
| 279 | bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req) |
| 280 | { |
| 281 | cpumask_var_t cpus; |
| 282 | bool called; |
Vitaly Kuznetsov | 7053df4 | 2018-05-16 17:21:28 +0200 | [diff] [blame] | 283 | |
| 284 | zalloc_cpumask_var(&cpus, GFP_ATOMIC); |
| 285 | |
Vitaly Kuznetsov | a812297 | 2018-08-22 12:18:29 +0200 | [diff] [blame] | 286 | called = kvm_make_vcpus_request_mask(kvm, req, NULL, cpus); |
Vitaly Kuznetsov | 7053df4 | 2018-05-16 17:21:28 +0200 | [diff] [blame] | 287 | |
Rusty Russell | 6ef7a1b | 2008-12-08 20:28:04 +1030 | [diff] [blame] | 288 | free_cpumask_var(cpus); |
Rusty Russell | 4984689 | 2008-12-08 20:26:24 +1030 | [diff] [blame] | 289 | return called; |
| 290 | } |
| 291 | |
Mario Smarduch | a6d5101 | 2015-01-15 15:58:52 -0800 | [diff] [blame] | 292 | #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL |
Rusty Russell | 4984689 | 2008-12-08 20:26:24 +1030 | [diff] [blame] | 293 | void kvm_flush_remote_tlbs(struct kvm *kvm) |
| 294 | { |
Lan Tianyu | 4ae3cb3 | 2016-03-13 11:10:28 +0800 | [diff] [blame] | 295 | /* |
| 296 | * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in |
| 297 | * kvm_make_all_cpus_request. |
| 298 | */ |
| 299 | long dirty_count = smp_load_acquire(&kvm->tlbs_dirty); |
Xiao Guangrong | a086f6a | 2014-04-17 17:06:12 +0800 | [diff] [blame] | 300 | |
Lan Tianyu | 4ae3cb3 | 2016-03-13 11:10:28 +0800 | [diff] [blame] | 301 | /* |
| 302 | * We want to publish modifications to the page tables before reading |
| 303 | * mode. Pairs with a memory barrier in arch-specific code. |
| 304 | * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest |
| 305 | * and smp_mb in walk_shadow_page_lockless_begin/end. |
| 306 | * - powerpc: smp_mb in kvmppc_prepare_to_enter. |
| 307 | * |
| 308 | * There is already an smp_mb__after_atomic() before |
| 309 | * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that |
| 310 | * barrier here. |
| 311 | */ |
Tianyu Lan | b08660e | 2018-07-19 08:40:17 +0000 | [diff] [blame] | 312 | if (!kvm_arch_flush_remote_tlb(kvm) |
| 313 | || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH)) |
Rusty Russell | 4984689 | 2008-12-08 20:26:24 +1030 | [diff] [blame] | 314 | ++kvm->stat.remote_tlb_flush; |
Xiao Guangrong | a086f6a | 2014-04-17 17:06:12 +0800 | [diff] [blame] | 315 | cmpxchg(&kvm->tlbs_dirty, dirty_count, 0); |
Avi Kivity | d9e368d | 2007-06-07 19:18:30 +0300 | [diff] [blame] | 316 | } |
Aneesh Kumar K.V | 2ba9f0d | 2013-10-07 22:17:59 +0530 | [diff] [blame] | 317 | EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs); |
Mario Smarduch | a6d5101 | 2015-01-15 15:58:52 -0800 | [diff] [blame] | 318 | #endif |
Avi Kivity | d9e368d | 2007-06-07 19:18:30 +0300 | [diff] [blame] | 319 | |
Marcelo Tosatti | 2e53d63 | 2008-02-20 14:47:24 -0500 | [diff] [blame] | 320 | void kvm_reload_remote_mmus(struct kvm *kvm) |
| 321 | { |
Tang Chen | 445b823 | 2014-09-24 15:57:55 +0800 | [diff] [blame] | 322 | kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD); |
Marcelo Tosatti | 2e53d63 | 2008-02-20 14:47:24 -0500 | [diff] [blame] | 323 | } |
| 324 | |
Rusty Russell | fb3f0f5 | 2007-07-27 17:16:56 +1000 | [diff] [blame] | 325 | int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id) |
| 326 | { |
| 327 | struct page *page; |
| 328 | int r; |
| 329 | |
| 330 | mutex_init(&vcpu->mutex); |
| 331 | vcpu->cpu = -1; |
Rusty Russell | fb3f0f5 | 2007-07-27 17:16:56 +1000 | [diff] [blame] | 332 | vcpu->kvm = kvm; |
| 333 | vcpu->vcpu_id = id; |
Rik van Riel | 34bb10b | 2011-02-01 09:52:41 -0500 | [diff] [blame] | 334 | vcpu->pid = NULL; |
Marcelo Tosatti | 8577370 | 2016-02-19 09:46:39 +0100 | [diff] [blame] | 335 | init_swait_queue_head(&vcpu->wq); |
Gleb Natapov | af585b9 | 2010-10-14 11:22:46 +0200 | [diff] [blame] | 336 | kvm_async_pf_vcpu_init(vcpu); |
Rusty Russell | fb3f0f5 | 2007-07-27 17:16:56 +1000 | [diff] [blame] | 337 | |
Feng Wu | bf9f6ac | 2015-09-18 22:29:55 +0800 | [diff] [blame] | 338 | vcpu->pre_pcpu = -1; |
| 339 | INIT_LIST_HEAD(&vcpu->blocked_vcpu_list); |
| 340 | |
Rusty Russell | fb3f0f5 | 2007-07-27 17:16:56 +1000 | [diff] [blame] | 341 | page = alloc_page(GFP_KERNEL | __GFP_ZERO); |
| 342 | if (!page) { |
| 343 | r = -ENOMEM; |
| 344 | goto fail; |
| 345 | } |
| 346 | vcpu->run = page_address(page); |
| 347 | |
Raghavendra K T | 4c08849 | 2012-07-18 19:07:46 +0530 | [diff] [blame] | 348 | kvm_vcpu_set_in_spin_loop(vcpu, false); |
| 349 | kvm_vcpu_set_dy_eligible(vcpu, false); |
Raghavendra K T | 3a08a8f | 2013-03-04 23:32:07 +0530 | [diff] [blame] | 350 | vcpu->preempted = false; |
Wanpeng Li | d73eb57 | 2019-07-18 19:39:06 +0800 | [diff] [blame] | 351 | vcpu->ready = false; |
Raghavendra K T | 4c08849 | 2012-07-18 19:07:46 +0530 | [diff] [blame] | 352 | |
Zhang Xiantao | e9b11c1 | 2007-11-14 20:38:21 +0800 | [diff] [blame] | 353 | r = kvm_arch_vcpu_init(vcpu); |
Rusty Russell | fb3f0f5 | 2007-07-27 17:16:56 +1000 | [diff] [blame] | 354 | if (r < 0) |
Zhang Xiantao | e9b11c1 | 2007-11-14 20:38:21 +0800 | [diff] [blame] | 355 | goto fail_free_run; |
Rusty Russell | fb3f0f5 | 2007-07-27 17:16:56 +1000 | [diff] [blame] | 356 | return 0; |
| 357 | |
Rusty Russell | fb3f0f5 | 2007-07-27 17:16:56 +1000 | [diff] [blame] | 358 | fail_free_run: |
| 359 | free_page((unsigned long)vcpu->run); |
| 360 | fail: |
Rusty Russell | 76fafa5 | 2007-10-08 10:50:48 +1000 | [diff] [blame] | 361 | return r; |
Rusty Russell | fb3f0f5 | 2007-07-27 17:16:56 +1000 | [diff] [blame] | 362 | } |
| 363 | EXPORT_SYMBOL_GPL(kvm_vcpu_init); |
| 364 | |
| 365 | void kvm_vcpu_uninit(struct kvm_vcpu *vcpu) |
| 366 | { |
Christian Borntraeger | 0e4524a | 2017-07-06 14:44:28 +0200 | [diff] [blame] | 367 | /* |
| 368 | * no need for rcu_read_lock as VCPU_RUN is the only place that |
| 369 | * will change the vcpu->pid pointer and on uninit all file |
| 370 | * descriptors are already gone. |
| 371 | */ |
| 372 | put_pid(rcu_dereference_protected(vcpu->pid, 1)); |
Zhang Xiantao | e9b11c1 | 2007-11-14 20:38:21 +0800 | [diff] [blame] | 373 | kvm_arch_vcpu_uninit(vcpu); |
Rusty Russell | fb3f0f5 | 2007-07-27 17:16:56 +1000 | [diff] [blame] | 374 | free_page((unsigned long)vcpu->run); |
| 375 | } |
| 376 | EXPORT_SYMBOL_GPL(kvm_vcpu_uninit); |
| 377 | |
Andrea Arcangeli | e930bff | 2008-07-25 16:24:52 +0200 | [diff] [blame] | 378 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
| 379 | static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn) |
| 380 | { |
| 381 | return container_of(mn, struct kvm, mmu_notifier); |
| 382 | } |
| 383 | |
Izik Eidus | 3da0dd4 | 2009-09-23 21:47:18 +0300 | [diff] [blame] | 384 | static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn, |
| 385 | struct mm_struct *mm, |
| 386 | unsigned long address, |
| 387 | pte_t pte) |
| 388 | { |
| 389 | struct kvm *kvm = mmu_notifier_to_kvm(mn); |
Marcelo Tosatti | bc6678a | 2009-12-23 14:35:21 -0200 | [diff] [blame] | 390 | int idx; |
Izik Eidus | 3da0dd4 | 2009-09-23 21:47:18 +0300 | [diff] [blame] | 391 | |
Marcelo Tosatti | bc6678a | 2009-12-23 14:35:21 -0200 | [diff] [blame] | 392 | idx = srcu_read_lock(&kvm->srcu); |
Izik Eidus | 3da0dd4 | 2009-09-23 21:47:18 +0300 | [diff] [blame] | 393 | spin_lock(&kvm->mmu_lock); |
| 394 | kvm->mmu_notifier_seq++; |
Lan Tianyu | 0cf853c | 2018-12-06 21:21:11 +0800 | [diff] [blame] | 395 | |
| 396 | if (kvm_set_spte_hva(kvm, address, pte)) |
| 397 | kvm_flush_remote_tlbs(kvm); |
| 398 | |
Izik Eidus | 3da0dd4 | 2009-09-23 21:47:18 +0300 | [diff] [blame] | 399 | spin_unlock(&kvm->mmu_lock); |
Marcelo Tosatti | bc6678a | 2009-12-23 14:35:21 -0200 | [diff] [blame] | 400 | srcu_read_unlock(&kvm->srcu, idx); |
Izik Eidus | 3da0dd4 | 2009-09-23 21:47:18 +0300 | [diff] [blame] | 401 | } |
| 402 | |
Michal Hocko | 93065ac | 2018-08-21 21:52:33 -0700 | [diff] [blame] | 403 | static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn, |
Jérôme Glisse | 5d6527a | 2018-12-28 00:38:05 -0800 | [diff] [blame] | 404 | const struct mmu_notifier_range *range) |
Andrea Arcangeli | e930bff | 2008-07-25 16:24:52 +0200 | [diff] [blame] | 405 | { |
| 406 | struct kvm *kvm = mmu_notifier_to_kvm(mn); |
Marcelo Tosatti | bc6678a | 2009-12-23 14:35:21 -0200 | [diff] [blame] | 407 | int need_tlb_flush = 0, idx; |
Michal Hocko | 93065ac | 2018-08-21 21:52:33 -0700 | [diff] [blame] | 408 | int ret; |
Andrea Arcangeli | e930bff | 2008-07-25 16:24:52 +0200 | [diff] [blame] | 409 | |
Marcelo Tosatti | bc6678a | 2009-12-23 14:35:21 -0200 | [diff] [blame] | 410 | idx = srcu_read_lock(&kvm->srcu); |
Andrea Arcangeli | e930bff | 2008-07-25 16:24:52 +0200 | [diff] [blame] | 411 | spin_lock(&kvm->mmu_lock); |
| 412 | /* |
| 413 | * The count increase must become visible at unlock time as no |
| 414 | * spte can be established without taking the mmu_lock and |
| 415 | * count is also read inside the mmu_lock critical section. |
| 416 | */ |
| 417 | kvm->mmu_notifier_count++; |
Jérôme Glisse | 5d6527a | 2018-12-28 00:38:05 -0800 | [diff] [blame] | 418 | need_tlb_flush = kvm_unmap_hva_range(kvm, range->start, range->end); |
Xiao Guangrong | a4ee1ca | 2010-11-23 11:13:00 +0800 | [diff] [blame] | 419 | need_tlb_flush |= kvm->tlbs_dirty; |
Andrea Arcangeli | e930bff | 2008-07-25 16:24:52 +0200 | [diff] [blame] | 420 | /* we've to flush the tlb before the pages can be freed */ |
| 421 | if (need_tlb_flush) |
| 422 | kvm_flush_remote_tlbs(kvm); |
Takuya Yoshikawa | 565f3be | 2012-02-10 15:28:31 +0900 | [diff] [blame] | 423 | |
| 424 | spin_unlock(&kvm->mmu_lock); |
Radim Krčmář | b1394e7 | 2017-11-30 19:05:45 +0100 | [diff] [blame] | 425 | |
Jérôme Glisse | 5d6527a | 2018-12-28 00:38:05 -0800 | [diff] [blame] | 426 | ret = kvm_arch_mmu_notifier_invalidate_range(kvm, range->start, |
Jérôme Glisse | dfcd666 | 2019-05-13 17:20:38 -0700 | [diff] [blame] | 427 | range->end, |
| 428 | mmu_notifier_range_blockable(range)); |
Radim Krčmář | b1394e7 | 2017-11-30 19:05:45 +0100 | [diff] [blame] | 429 | |
Takuya Yoshikawa | 565f3be | 2012-02-10 15:28:31 +0900 | [diff] [blame] | 430 | srcu_read_unlock(&kvm->srcu, idx); |
Michal Hocko | 93065ac | 2018-08-21 21:52:33 -0700 | [diff] [blame] | 431 | |
| 432 | return ret; |
Andrea Arcangeli | e930bff | 2008-07-25 16:24:52 +0200 | [diff] [blame] | 433 | } |
| 434 | |
| 435 | static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn, |
Jérôme Glisse | 5d6527a | 2018-12-28 00:38:05 -0800 | [diff] [blame] | 436 | const struct mmu_notifier_range *range) |
Andrea Arcangeli | e930bff | 2008-07-25 16:24:52 +0200 | [diff] [blame] | 437 | { |
| 438 | struct kvm *kvm = mmu_notifier_to_kvm(mn); |
| 439 | |
| 440 | spin_lock(&kvm->mmu_lock); |
| 441 | /* |
| 442 | * This sequence increase will notify the kvm page fault that |
| 443 | * the page that is going to be mapped in the spte could have |
| 444 | * been freed. |
| 445 | */ |
| 446 | kvm->mmu_notifier_seq++; |
Paul Mackerras | a355aa5 | 2011-12-12 12:37:21 +0000 | [diff] [blame] | 447 | smp_wmb(); |
Andrea Arcangeli | e930bff | 2008-07-25 16:24:52 +0200 | [diff] [blame] | 448 | /* |
| 449 | * The above sequence increase must be visible before the |
Paul Mackerras | a355aa5 | 2011-12-12 12:37:21 +0000 | [diff] [blame] | 450 | * below count decrease, which is ensured by the smp_wmb above |
| 451 | * in conjunction with the smp_rmb in mmu_notifier_retry(). |
Andrea Arcangeli | e930bff | 2008-07-25 16:24:52 +0200 | [diff] [blame] | 452 | */ |
| 453 | kvm->mmu_notifier_count--; |
| 454 | spin_unlock(&kvm->mmu_lock); |
| 455 | |
| 456 | BUG_ON(kvm->mmu_notifier_count < 0); |
| 457 | } |
| 458 | |
| 459 | static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn, |
| 460 | struct mm_struct *mm, |
Andres Lagar-Cavilla | 5712846 | 2014-09-22 14:54:42 -0700 | [diff] [blame] | 461 | unsigned long start, |
| 462 | unsigned long end) |
Andrea Arcangeli | e930bff | 2008-07-25 16:24:52 +0200 | [diff] [blame] | 463 | { |
| 464 | struct kvm *kvm = mmu_notifier_to_kvm(mn); |
Marcelo Tosatti | bc6678a | 2009-12-23 14:35:21 -0200 | [diff] [blame] | 465 | int young, idx; |
Andrea Arcangeli | e930bff | 2008-07-25 16:24:52 +0200 | [diff] [blame] | 466 | |
Marcelo Tosatti | bc6678a | 2009-12-23 14:35:21 -0200 | [diff] [blame] | 467 | idx = srcu_read_lock(&kvm->srcu); |
Andrea Arcangeli | e930bff | 2008-07-25 16:24:52 +0200 | [diff] [blame] | 468 | spin_lock(&kvm->mmu_lock); |
Andrea Arcangeli | e930bff | 2008-07-25 16:24:52 +0200 | [diff] [blame] | 469 | |
Andres Lagar-Cavilla | 5712846 | 2014-09-22 14:54:42 -0700 | [diff] [blame] | 470 | young = kvm_age_hva(kvm, start, end); |
Andrea Arcangeli | e930bff | 2008-07-25 16:24:52 +0200 | [diff] [blame] | 471 | if (young) |
| 472 | kvm_flush_remote_tlbs(kvm); |
| 473 | |
Takuya Yoshikawa | 565f3be | 2012-02-10 15:28:31 +0900 | [diff] [blame] | 474 | spin_unlock(&kvm->mmu_lock); |
| 475 | srcu_read_unlock(&kvm->srcu, idx); |
| 476 | |
Andrea Arcangeli | e930bff | 2008-07-25 16:24:52 +0200 | [diff] [blame] | 477 | return young; |
| 478 | } |
| 479 | |
Vladimir Davydov | 1d7715c | 2015-09-09 15:35:41 -0700 | [diff] [blame] | 480 | static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn, |
| 481 | struct mm_struct *mm, |
| 482 | unsigned long start, |
| 483 | unsigned long end) |
| 484 | { |
| 485 | struct kvm *kvm = mmu_notifier_to_kvm(mn); |
| 486 | int young, idx; |
| 487 | |
| 488 | idx = srcu_read_lock(&kvm->srcu); |
| 489 | spin_lock(&kvm->mmu_lock); |
| 490 | /* |
| 491 | * Even though we do not flush TLB, this will still adversely |
| 492 | * affect performance on pre-Haswell Intel EPT, where there is |
| 493 | * no EPT Access Bit to clear so that we have to tear down EPT |
| 494 | * tables instead. If we find this unacceptable, we can always |
| 495 | * add a parameter to kvm_age_hva so that it effectively doesn't |
| 496 | * do anything on clear_young. |
| 497 | * |
| 498 | * Also note that currently we never issue secondary TLB flushes |
| 499 | * from clear_young, leaving this job up to the regular system |
| 500 | * cadence. If we find this inaccurate, we might come up with a |
| 501 | * more sophisticated heuristic later. |
| 502 | */ |
| 503 | young = kvm_age_hva(kvm, start, end); |
| 504 | spin_unlock(&kvm->mmu_lock); |
| 505 | srcu_read_unlock(&kvm->srcu, idx); |
| 506 | |
| 507 | return young; |
| 508 | } |
| 509 | |
Andrea Arcangeli | 8ee5382 | 2011-01-13 15:47:10 -0800 | [diff] [blame] | 510 | static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn, |
| 511 | struct mm_struct *mm, |
| 512 | unsigned long address) |
| 513 | { |
| 514 | struct kvm *kvm = mmu_notifier_to_kvm(mn); |
| 515 | int young, idx; |
| 516 | |
| 517 | idx = srcu_read_lock(&kvm->srcu); |
| 518 | spin_lock(&kvm->mmu_lock); |
| 519 | young = kvm_test_age_hva(kvm, address); |
| 520 | spin_unlock(&kvm->mmu_lock); |
| 521 | srcu_read_unlock(&kvm->srcu, idx); |
| 522 | |
| 523 | return young; |
| 524 | } |
| 525 | |
Marcelo Tosatti | 85db06e | 2008-12-10 21:23:26 +0100 | [diff] [blame] | 526 | static void kvm_mmu_notifier_release(struct mmu_notifier *mn, |
| 527 | struct mm_struct *mm) |
| 528 | { |
| 529 | struct kvm *kvm = mmu_notifier_to_kvm(mn); |
Lai Jiangshan | eda2bed | 2010-04-20 14:29:29 +0800 | [diff] [blame] | 530 | int idx; |
| 531 | |
| 532 | idx = srcu_read_lock(&kvm->srcu); |
Marcelo Tosatti | 2df72e9 | 2012-08-24 15:54:57 -0300 | [diff] [blame] | 533 | kvm_arch_flush_shadow_all(kvm); |
Lai Jiangshan | eda2bed | 2010-04-20 14:29:29 +0800 | [diff] [blame] | 534 | srcu_read_unlock(&kvm->srcu, idx); |
Marcelo Tosatti | 85db06e | 2008-12-10 21:23:26 +0100 | [diff] [blame] | 535 | } |
| 536 | |
Andrea Arcangeli | e930bff | 2008-07-25 16:24:52 +0200 | [diff] [blame] | 537 | static const struct mmu_notifier_ops kvm_mmu_notifier_ops = { |
Andrea Arcangeli | e930bff | 2008-07-25 16:24:52 +0200 | [diff] [blame] | 538 | .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start, |
| 539 | .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end, |
| 540 | .clear_flush_young = kvm_mmu_notifier_clear_flush_young, |
Vladimir Davydov | 1d7715c | 2015-09-09 15:35:41 -0700 | [diff] [blame] | 541 | .clear_young = kvm_mmu_notifier_clear_young, |
Andrea Arcangeli | 8ee5382 | 2011-01-13 15:47:10 -0800 | [diff] [blame] | 542 | .test_young = kvm_mmu_notifier_test_young, |
Izik Eidus | 3da0dd4 | 2009-09-23 21:47:18 +0300 | [diff] [blame] | 543 | .change_pte = kvm_mmu_notifier_change_pte, |
Marcelo Tosatti | 85db06e | 2008-12-10 21:23:26 +0100 | [diff] [blame] | 544 | .release = kvm_mmu_notifier_release, |
Andrea Arcangeli | e930bff | 2008-07-25 16:24:52 +0200 | [diff] [blame] | 545 | }; |
Avi Kivity | 4c07b0a | 2009-12-20 14:54:04 +0200 | [diff] [blame] | 546 | |
| 547 | static int kvm_init_mmu_notifier(struct kvm *kvm) |
| 548 | { |
| 549 | kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops; |
| 550 | return mmu_notifier_register(&kvm->mmu_notifier, current->mm); |
| 551 | } |
| 552 | |
| 553 | #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */ |
| 554 | |
| 555 | static int kvm_init_mmu_notifier(struct kvm *kvm) |
| 556 | { |
| 557 | return 0; |
| 558 | } |
| 559 | |
Andrea Arcangeli | e930bff | 2008-07-25 16:24:52 +0200 | [diff] [blame] | 560 | #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */ |
| 561 | |
Paolo Bonzini | a47d2b0 | 2015-05-17 11:41:37 +0200 | [diff] [blame] | 562 | static struct kvm_memslots *kvm_alloc_memslots(void) |
Xiao Guangrong | bf3e05b | 2011-11-24 17:40:57 +0800 | [diff] [blame] | 563 | { |
| 564 | int i; |
Paolo Bonzini | a47d2b0 | 2015-05-17 11:41:37 +0200 | [diff] [blame] | 565 | struct kvm_memslots *slots; |
Xiao Guangrong | bf3e05b | 2011-11-24 17:40:57 +0800 | [diff] [blame] | 566 | |
Ben Gardon | b12ce36 | 2019-02-11 11:02:49 -0800 | [diff] [blame] | 567 | slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT); |
Paolo Bonzini | a47d2b0 | 2015-05-17 11:41:37 +0200 | [diff] [blame] | 568 | if (!slots) |
| 569 | return NULL; |
| 570 | |
Xiao Guangrong | bf3e05b | 2011-11-24 17:40:57 +0800 | [diff] [blame] | 571 | for (i = 0; i < KVM_MEM_SLOTS_NUM; i++) |
Xiao Guangrong | f85e2cb | 2011-11-24 17:41:54 +0800 | [diff] [blame] | 572 | slots->id_to_index[i] = slots->memslots[i].id = i; |
Paolo Bonzini | a47d2b0 | 2015-05-17 11:41:37 +0200 | [diff] [blame] | 573 | |
| 574 | return slots; |
| 575 | } |
| 576 | |
| 577 | static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot) |
| 578 | { |
| 579 | if (!memslot->dirty_bitmap) |
| 580 | return; |
| 581 | |
| 582 | kvfree(memslot->dirty_bitmap); |
| 583 | memslot->dirty_bitmap = NULL; |
| 584 | } |
| 585 | |
| 586 | /* |
| 587 | * Free any memory in @free but not in @dont. |
| 588 | */ |
| 589 | static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free, |
| 590 | struct kvm_memory_slot *dont) |
| 591 | { |
| 592 | if (!dont || free->dirty_bitmap != dont->dirty_bitmap) |
| 593 | kvm_destroy_dirty_bitmap(free); |
| 594 | |
| 595 | kvm_arch_free_memslot(kvm, free, dont); |
| 596 | |
| 597 | free->npages = 0; |
| 598 | } |
| 599 | |
| 600 | static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots) |
| 601 | { |
| 602 | struct kvm_memory_slot *memslot; |
| 603 | |
| 604 | if (!slots) |
| 605 | return; |
| 606 | |
| 607 | kvm_for_each_memslot(memslot, slots) |
| 608 | kvm_free_memslot(kvm, memslot, NULL); |
| 609 | |
| 610 | kvfree(slots); |
Xiao Guangrong | bf3e05b | 2011-11-24 17:40:57 +0800 | [diff] [blame] | 611 | } |
| 612 | |
Janosch Frank | 536a6f8 | 2016-05-18 13:26:23 +0200 | [diff] [blame] | 613 | static void kvm_destroy_vm_debugfs(struct kvm *kvm) |
| 614 | { |
| 615 | int i; |
| 616 | |
| 617 | if (!kvm->debugfs_dentry) |
| 618 | return; |
| 619 | |
| 620 | debugfs_remove_recursive(kvm->debugfs_dentry); |
| 621 | |
Luiz Capitulino | 9d5a1dc | 2016-09-07 14:47:21 -0400 | [diff] [blame] | 622 | if (kvm->debugfs_stat_data) { |
| 623 | for (i = 0; i < kvm_debugfs_num_entries; i++) |
| 624 | kfree(kvm->debugfs_stat_data[i]); |
| 625 | kfree(kvm->debugfs_stat_data); |
| 626 | } |
Janosch Frank | 536a6f8 | 2016-05-18 13:26:23 +0200 | [diff] [blame] | 627 | } |
| 628 | |
| 629 | static int kvm_create_vm_debugfs(struct kvm *kvm, int fd) |
| 630 | { |
| 631 | char dir_name[ITOA_MAX_LEN * 2]; |
| 632 | struct kvm_stat_data *stat_data; |
| 633 | struct kvm_stats_debugfs_item *p; |
| 634 | |
| 635 | if (!debugfs_initialized()) |
| 636 | return 0; |
| 637 | |
| 638 | snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd); |
Greg Kroah-Hartman | 929f45e | 2018-05-29 18:22:04 +0200 | [diff] [blame] | 639 | kvm->debugfs_dentry = debugfs_create_dir(dir_name, kvm_debugfs_dir); |
Janosch Frank | 536a6f8 | 2016-05-18 13:26:23 +0200 | [diff] [blame] | 640 | |
| 641 | kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries, |
| 642 | sizeof(*kvm->debugfs_stat_data), |
Ben Gardon | b12ce36 | 2019-02-11 11:02:49 -0800 | [diff] [blame] | 643 | GFP_KERNEL_ACCOUNT); |
Janosch Frank | 536a6f8 | 2016-05-18 13:26:23 +0200 | [diff] [blame] | 644 | if (!kvm->debugfs_stat_data) |
| 645 | return -ENOMEM; |
| 646 | |
| 647 | for (p = debugfs_entries; p->name; p++) { |
Ben Gardon | b12ce36 | 2019-02-11 11:02:49 -0800 | [diff] [blame] | 648 | stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT); |
Janosch Frank | 536a6f8 | 2016-05-18 13:26:23 +0200 | [diff] [blame] | 649 | if (!stat_data) |
| 650 | return -ENOMEM; |
| 651 | |
| 652 | stat_data->kvm = kvm; |
| 653 | stat_data->offset = p->offset; |
Paolo Bonzini | 833b45d | 2019-09-30 18:48:44 +0200 | [diff] [blame] | 654 | stat_data->mode = p->mode ? p->mode : 0644; |
Janosch Frank | 536a6f8 | 2016-05-18 13:26:23 +0200 | [diff] [blame] | 655 | kvm->debugfs_stat_data[p - debugfs_entries] = stat_data; |
Paolo Bonzini | 833b45d | 2019-09-30 18:48:44 +0200 | [diff] [blame] | 656 | debugfs_create_file(p->name, stat_data->mode, kvm->debugfs_dentry, |
Greg Kroah-Hartman | 929f45e | 2018-05-29 18:22:04 +0200 | [diff] [blame] | 657 | stat_data, stat_fops_per_vm[p->kind]); |
Janosch Frank | 536a6f8 | 2016-05-18 13:26:23 +0200 | [diff] [blame] | 658 | } |
| 659 | return 0; |
| 660 | } |
| 661 | |
Junaid Shahid | 1aa9b95 | 2019-11-04 20:26:00 +0100 | [diff] [blame] | 662 | /* |
| 663 | * Called after the VM is otherwise initialized, but just before adding it to |
| 664 | * the vm_list. |
| 665 | */ |
| 666 | int __weak kvm_arch_post_init_vm(struct kvm *kvm) |
| 667 | { |
| 668 | return 0; |
| 669 | } |
| 670 | |
| 671 | /* |
| 672 | * Called just after removing the VM from the vm_list, but before doing any |
| 673 | * other destruction. |
| 674 | */ |
| 675 | void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm) |
| 676 | { |
| 677 | } |
| 678 | |
Carsten Otte | e08b963 | 2012-01-04 10:25:20 +0100 | [diff] [blame] | 679 | static struct kvm *kvm_create_vm(unsigned long type) |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 680 | { |
Jan Kiszka | d89f5ef | 2010-11-09 17:02:49 +0100 | [diff] [blame] | 681 | struct kvm *kvm = kvm_arch_alloc_vm(); |
Jim Mattson | 9121923 | 2019-10-24 16:03:26 -0700 | [diff] [blame] | 682 | int r = -ENOMEM; |
| 683 | int i; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 684 | |
Jan Kiszka | d89f5ef | 2010-11-09 17:02:49 +0100 | [diff] [blame] | 685 | if (!kvm) |
| 686 | return ERR_PTR(-ENOMEM); |
| 687 | |
Paolo Bonzini | e9ad4ec | 2016-03-21 10:15:25 +0100 | [diff] [blame] | 688 | spin_lock_init(&kvm->mmu_lock); |
Vegard Nossum | f1f1007 | 2017-02-27 14:30:07 -0800 | [diff] [blame] | 689 | mmgrab(current->mm); |
Paolo Bonzini | e9ad4ec | 2016-03-21 10:15:25 +0100 | [diff] [blame] | 690 | kvm->mm = current->mm; |
| 691 | kvm_eventfd_init(kvm); |
| 692 | mutex_init(&kvm->lock); |
| 693 | mutex_init(&kvm->irq_lock); |
| 694 | mutex_init(&kvm->slots_lock); |
Paolo Bonzini | e9ad4ec | 2016-03-21 10:15:25 +0100 | [diff] [blame] | 695 | INIT_LIST_HEAD(&kvm->devices); |
| 696 | |
Jim Mattson | 9121923 | 2019-10-24 16:03:26 -0700 | [diff] [blame] | 697 | BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX); |
| 698 | |
Paolo Bonzini | 8a44119a | 2019-11-04 12:16:49 +0100 | [diff] [blame] | 699 | if (init_srcu_struct(&kvm->srcu)) |
| 700 | goto out_err_no_srcu; |
| 701 | if (init_srcu_struct(&kvm->irq_srcu)) |
| 702 | goto out_err_no_irq_srcu; |
| 703 | |
Paolo Bonzini | e2d3fca | 2019-11-04 13:23:53 +0100 | [diff] [blame] | 704 | refcount_set(&kvm->users_count, 1); |
Jim Mattson | 9121923 | 2019-10-24 16:03:26 -0700 | [diff] [blame] | 705 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) { |
| 706 | struct kvm_memslots *slots = kvm_alloc_memslots(); |
| 707 | |
| 708 | if (!slots) |
Jim Mattson | a97b0e7 | 2019-10-25 13:34:58 +0200 | [diff] [blame] | 709 | goto out_err_no_arch_destroy_vm; |
Jim Mattson | 9121923 | 2019-10-24 16:03:26 -0700 | [diff] [blame] | 710 | /* Generations must be different for each address space. */ |
| 711 | slots->generation = i; |
| 712 | rcu_assign_pointer(kvm->memslots[i], slots); |
| 713 | } |
| 714 | |
| 715 | for (i = 0; i < KVM_NR_BUSES; i++) { |
| 716 | rcu_assign_pointer(kvm->buses[i], |
| 717 | kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT)); |
| 718 | if (!kvm->buses[i]) |
Jim Mattson | a97b0e7 | 2019-10-25 13:34:58 +0200 | [diff] [blame] | 719 | goto out_err_no_arch_destroy_vm; |
Jim Mattson | 9121923 | 2019-10-24 16:03:26 -0700 | [diff] [blame] | 720 | } |
| 721 | |
Carsten Otte | e08b963 | 2012-01-04 10:25:20 +0100 | [diff] [blame] | 722 | r = kvm_arch_init_vm(kvm, type); |
Jan Kiszka | d89f5ef | 2010-11-09 17:02:49 +0100 | [diff] [blame] | 723 | if (r) |
Jim Mattson | a97b0e7 | 2019-10-25 13:34:58 +0200 | [diff] [blame] | 724 | goto out_err_no_arch_destroy_vm; |
Alexander Graf | 10474ae | 2009-09-15 11:37:46 +0200 | [diff] [blame] | 725 | |
| 726 | r = hardware_enable_all(); |
| 727 | if (r) |
Christian Borntraeger | 719d93c | 2014-01-16 13:44:20 +0100 | [diff] [blame] | 728 | goto out_err_no_disable; |
Alexander Graf | 10474ae | 2009-09-15 11:37:46 +0200 | [diff] [blame] | 729 | |
Paolo Bonzini | c77dcac | 2014-08-06 14:24:45 +0200 | [diff] [blame] | 730 | #ifdef CONFIG_HAVE_KVM_IRQFD |
Gleb Natapov | 136bdfe | 2009-08-24 11:54:23 +0300 | [diff] [blame] | 731 | INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list); |
Avi Kivity | 75858a8 | 2009-01-04 17:10:50 +0200 | [diff] [blame] | 732 | #endif |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 733 | |
Mike Waychison | 74b5c5b | 2011-06-03 13:04:53 -0700 | [diff] [blame] | 734 | r = kvm_init_mmu_notifier(kvm); |
| 735 | if (r) |
Junaid Shahid | 1aa9b95 | 2019-11-04 20:26:00 +0100 | [diff] [blame] | 736 | goto out_err_no_mmu_notifier; |
| 737 | |
| 738 | r = kvm_arch_post_init_vm(kvm); |
| 739 | if (r) |
Mike Waychison | 74b5c5b | 2011-06-03 13:04:53 -0700 | [diff] [blame] | 740 | goto out_err; |
| 741 | |
Junaid Shahid | 0d9ce16 | 2019-01-03 17:14:28 -0800 | [diff] [blame] | 742 | mutex_lock(&kvm_lock); |
Rusty Russell | 5e58cfe | 2007-07-23 17:08:21 +1000 | [diff] [blame] | 743 | list_add(&kvm->vm_list, &vm_list); |
Junaid Shahid | 0d9ce16 | 2019-01-03 17:14:28 -0800 | [diff] [blame] | 744 | mutex_unlock(&kvm_lock); |
Jan Kiszka | d89f5ef | 2010-11-09 17:02:49 +0100 | [diff] [blame] | 745 | |
Peter Zijlstra | 2ecd9d2 | 2015-07-03 18:53:58 +0200 | [diff] [blame] | 746 | preempt_notifier_inc(); |
| 747 | |
Avi Kivity | f17abe9 | 2007-02-21 19:28:04 +0200 | [diff] [blame] | 748 | return kvm; |
Alexander Graf | 10474ae | 2009-09-15 11:37:46 +0200 | [diff] [blame] | 749 | |
| 750 | out_err: |
Junaid Shahid | 1aa9b95 | 2019-11-04 20:26:00 +0100 | [diff] [blame] | 751 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
| 752 | if (kvm->mmu_notifier.ops) |
| 753 | mmu_notifier_unregister(&kvm->mmu_notifier, current->mm); |
| 754 | #endif |
| 755 | out_err_no_mmu_notifier: |
Alexander Graf | 10474ae | 2009-09-15 11:37:46 +0200 | [diff] [blame] | 756 | hardware_disable_all(); |
Christian Borntraeger | 719d93c | 2014-01-16 13:44:20 +0100 | [diff] [blame] | 757 | out_err_no_disable: |
Jim Mattson | a97b0e7 | 2019-10-25 13:34:58 +0200 | [diff] [blame] | 758 | kvm_arch_destroy_vm(kvm); |
Jim Mattson | a97b0e7 | 2019-10-25 13:34:58 +0200 | [diff] [blame] | 759 | out_err_no_arch_destroy_vm: |
Paolo Bonzini | e2d3fca | 2019-11-04 13:23:53 +0100 | [diff] [blame] | 760 | WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count)); |
Marcelo Tosatti | e93f8a0 | 2009-12-23 14:35:24 -0200 | [diff] [blame] | 761 | for (i = 0; i < KVM_NR_BUSES; i++) |
Paolo Bonzini | 3898da9 | 2017-08-02 17:55:54 +0200 | [diff] [blame] | 762 | kfree(kvm_get_bus(kvm, i)); |
Paolo Bonzini | f481b06 | 2015-05-17 17:30:37 +0200 | [diff] [blame] | 763 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) |
Paolo Bonzini | 3898da9 | 2017-08-02 17:55:54 +0200 | [diff] [blame] | 764 | kvm_free_memslots(kvm, __kvm_memslots(kvm, i)); |
Paolo Bonzini | 8a44119a | 2019-11-04 12:16:49 +0100 | [diff] [blame] | 765 | cleanup_srcu_struct(&kvm->irq_srcu); |
| 766 | out_err_no_irq_srcu: |
| 767 | cleanup_srcu_struct(&kvm->srcu); |
| 768 | out_err_no_srcu: |
Jan Kiszka | d89f5ef | 2010-11-09 17:02:49 +0100 | [diff] [blame] | 769 | kvm_arch_free_vm(kvm); |
Paolo Bonzini | e9ad4ec | 2016-03-21 10:15:25 +0100 | [diff] [blame] | 770 | mmdrop(current->mm); |
Alexander Graf | 10474ae | 2009-09-15 11:37:46 +0200 | [diff] [blame] | 771 | return ERR_PTR(r); |
Avi Kivity | f17abe9 | 2007-02-21 19:28:04 +0200 | [diff] [blame] | 772 | } |
| 773 | |
Scott Wood | 07f0a7b | 2013-04-25 14:11:23 +0000 | [diff] [blame] | 774 | static void kvm_destroy_devices(struct kvm *kvm) |
| 775 | { |
Geliang Tang | e6e3b5a | 2016-01-01 19:47:12 +0800 | [diff] [blame] | 776 | struct kvm_device *dev, *tmp; |
Scott Wood | 07f0a7b | 2013-04-25 14:11:23 +0000 | [diff] [blame] | 777 | |
Christoffer Dall | a28ebea | 2016-08-09 19:13:01 +0200 | [diff] [blame] | 778 | /* |
| 779 | * We do not need to take the kvm->lock here, because nobody else |
| 780 | * has a reference to the struct kvm at this point and therefore |
| 781 | * cannot access the devices list anyhow. |
| 782 | */ |
Geliang Tang | e6e3b5a | 2016-01-01 19:47:12 +0800 | [diff] [blame] | 783 | list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) { |
| 784 | list_del(&dev->vm_node); |
Scott Wood | 07f0a7b | 2013-04-25 14:11:23 +0000 | [diff] [blame] | 785 | dev->ops->destroy(dev); |
| 786 | } |
| 787 | } |
| 788 | |
Avi Kivity | f17abe9 | 2007-02-21 19:28:04 +0200 | [diff] [blame] | 789 | static void kvm_destroy_vm(struct kvm *kvm) |
| 790 | { |
Marcelo Tosatti | e93f8a0 | 2009-12-23 14:35:24 -0200 | [diff] [blame] | 791 | int i; |
Avi Kivity | 6d4e4c4 | 2007-11-21 16:41:05 +0200 | [diff] [blame] | 792 | struct mm_struct *mm = kvm->mm; |
| 793 | |
Claudio Imbrenda | 286de8f | 2017-07-12 17:56:44 +0200 | [diff] [blame] | 794 | kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm); |
Janosch Frank | 536a6f8 | 2016-05-18 13:26:23 +0200 | [diff] [blame] | 795 | kvm_destroy_vm_debugfs(kvm); |
Sheng Yang | ad8ba2c | 2009-01-06 10:03:02 +0800 | [diff] [blame] | 796 | kvm_arch_sync_events(kvm); |
Junaid Shahid | 0d9ce16 | 2019-01-03 17:14:28 -0800 | [diff] [blame] | 797 | mutex_lock(&kvm_lock); |
Avi Kivity | 133de90 | 2007-02-12 00:54:44 -0800 | [diff] [blame] | 798 | list_del(&kvm->vm_list); |
Junaid Shahid | 0d9ce16 | 2019-01-03 17:14:28 -0800 | [diff] [blame] | 799 | mutex_unlock(&kvm_lock); |
Junaid Shahid | 1aa9b95 | 2019-11-04 20:26:00 +0100 | [diff] [blame] | 800 | kvm_arch_pre_destroy_vm(kvm); |
| 801 | |
Avi Kivity | 399ec80 | 2008-11-19 13:58:46 +0200 | [diff] [blame] | 802 | kvm_free_irq_routing(kvm); |
Peter Xu | df630b8 | 2017-03-15 16:01:17 +0800 | [diff] [blame] | 803 | for (i = 0; i < KVM_NR_BUSES; i++) { |
Paolo Bonzini | 3898da9 | 2017-08-02 17:55:54 +0200 | [diff] [blame] | 804 | struct kvm_io_bus *bus = kvm_get_bus(kvm, i); |
Christian Borntraeger | 4a12f95 | 2017-07-07 10:51:38 +0200 | [diff] [blame] | 805 | |
Christian Borntraeger | 4a12f95 | 2017-07-07 10:51:38 +0200 | [diff] [blame] | 806 | if (bus) |
| 807 | kvm_io_bus_destroy(bus); |
Peter Xu | df630b8 | 2017-03-15 16:01:17 +0800 | [diff] [blame] | 808 | kvm->buses[i] = NULL; |
| 809 | } |
Avi Kivity | 980da6c | 2009-12-20 15:13:43 +0200 | [diff] [blame] | 810 | kvm_coalesced_mmio_free(kvm); |
Andrea Arcangeli | e930bff | 2008-07-25 16:24:52 +0200 | [diff] [blame] | 811 | #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) |
| 812 | mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm); |
Gleb Natapov | f00be0c | 2009-03-19 12:20:36 +0200 | [diff] [blame] | 813 | #else |
Marcelo Tosatti | 2df72e9 | 2012-08-24 15:54:57 -0300 | [diff] [blame] | 814 | kvm_arch_flush_shadow_all(kvm); |
Andrea Arcangeli | e930bff | 2008-07-25 16:24:52 +0200 | [diff] [blame] | 815 | #endif |
Zhang Xiantao | d19a9cd | 2007-11-18 18:43:45 +0800 | [diff] [blame] | 816 | kvm_arch_destroy_vm(kvm); |
Scott Wood | 07f0a7b | 2013-04-25 14:11:23 +0000 | [diff] [blame] | 817 | kvm_destroy_devices(kvm); |
Paolo Bonzini | f481b06 | 2015-05-17 17:30:37 +0200 | [diff] [blame] | 818 | for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) |
Paolo Bonzini | 3898da9 | 2017-08-02 17:55:54 +0200 | [diff] [blame] | 819 | kvm_free_memslots(kvm, __kvm_memslots(kvm, i)); |
Paolo Bonzini | 820b3fc | 2014-06-03 13:44:17 +0200 | [diff] [blame] | 820 | cleanup_srcu_struct(&kvm->irq_srcu); |
Jan Kiszka | d89f5ef | 2010-11-09 17:02:49 +0100 | [diff] [blame] | 821 | cleanup_srcu_struct(&kvm->srcu); |
| 822 | kvm_arch_free_vm(kvm); |
Peter Zijlstra | 2ecd9d2 | 2015-07-03 18:53:58 +0200 | [diff] [blame] | 823 | preempt_notifier_dec(); |
Alexander Graf | 10474ae | 2009-09-15 11:37:46 +0200 | [diff] [blame] | 824 | hardware_disable_all(); |
Avi Kivity | 6d4e4c4 | 2007-11-21 16:41:05 +0200 | [diff] [blame] | 825 | mmdrop(mm); |
Avi Kivity | f17abe9 | 2007-02-21 19:28:04 +0200 | [diff] [blame] | 826 | } |
| 827 | |
Izik Eidus | d39f13b | 2008-03-30 16:01:25 +0300 | [diff] [blame] | 828 | void kvm_get_kvm(struct kvm *kvm) |
| 829 | { |
Elena Reshetova | e3736c3 | 2017-02-20 13:06:21 +0200 | [diff] [blame] | 830 | refcount_inc(&kvm->users_count); |
Izik Eidus | d39f13b | 2008-03-30 16:01:25 +0300 | [diff] [blame] | 831 | } |
| 832 | EXPORT_SYMBOL_GPL(kvm_get_kvm); |
| 833 | |
| 834 | void kvm_put_kvm(struct kvm *kvm) |
| 835 | { |
Elena Reshetova | e3736c3 | 2017-02-20 13:06:21 +0200 | [diff] [blame] | 836 | if (refcount_dec_and_test(&kvm->users_count)) |
Izik Eidus | d39f13b | 2008-03-30 16:01:25 +0300 | [diff] [blame] | 837 | kvm_destroy_vm(kvm); |
| 838 | } |
| 839 | EXPORT_SYMBOL_GPL(kvm_put_kvm); |
| 840 | |
Sean Christopherson | 149487b | 2019-10-21 15:58:42 -0700 | [diff] [blame] | 841 | /* |
| 842 | * Used to put a reference that was taken on behalf of an object associated |
| 843 | * with a user-visible file descriptor, e.g. a vcpu or device, if installation |
| 844 | * of the new file descriptor fails and the reference cannot be transferred to |
| 845 | * its final owner. In such cases, the caller is still actively using @kvm and |
| 846 | * will fail miserably if the refcount unexpectedly hits zero. |
| 847 | */ |
| 848 | void kvm_put_kvm_no_destroy(struct kvm *kvm) |
| 849 | { |
| 850 | WARN_ON(refcount_dec_and_test(&kvm->users_count)); |
| 851 | } |
| 852 | EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy); |
Izik Eidus | d39f13b | 2008-03-30 16:01:25 +0300 | [diff] [blame] | 853 | |
Avi Kivity | f17abe9 | 2007-02-21 19:28:04 +0200 | [diff] [blame] | 854 | static int kvm_vm_release(struct inode *inode, struct file *filp) |
| 855 | { |
| 856 | struct kvm *kvm = filp->private_data; |
| 857 | |
Gregory Haskins | 721eecbf | 2009-05-20 10:30:49 -0400 | [diff] [blame] | 858 | kvm_irqfd_release(kvm); |
| 859 | |
Izik Eidus | d39f13b | 2008-03-30 16:01:25 +0300 | [diff] [blame] | 860 | kvm_put_kvm(kvm); |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 861 | return 0; |
| 862 | } |
| 863 | |
Takuya Yoshikawa | 515a012 | 2010-10-27 18:23:54 +0900 | [diff] [blame] | 864 | /* |
| 865 | * Allocation size is twice as large as the actual dirty bitmap size. |
Takuya Yoshikawa | 93474b2 | 2012-03-01 19:34:45 +0900 | [diff] [blame] | 866 | * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed. |
Takuya Yoshikawa | 515a012 | 2010-10-27 18:23:54 +0900 | [diff] [blame] | 867 | */ |
Takuya Yoshikawa | a36a57b1 | 2010-10-27 18:22:19 +0900 | [diff] [blame] | 868 | static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot) |
| 869 | { |
Takuya Yoshikawa | 515a012 | 2010-10-27 18:23:54 +0900 | [diff] [blame] | 870 | unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot); |
Takuya Yoshikawa | a36a57b1 | 2010-10-27 18:22:19 +0900 | [diff] [blame] | 871 | |
Ben Gardon | b12ce36 | 2019-02-11 11:02:49 -0800 | [diff] [blame] | 872 | memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL_ACCOUNT); |
Takuya Yoshikawa | a36a57b1 | 2010-10-27 18:22:19 +0900 | [diff] [blame] | 873 | if (!memslot->dirty_bitmap) |
| 874 | return -ENOMEM; |
| 875 | |
Takuya Yoshikawa | a36a57b1 | 2010-10-27 18:22:19 +0900 | [diff] [blame] | 876 | return 0; |
| 877 | } |
| 878 | |
Xiao Guangrong | bf3e05b | 2011-11-24 17:40:57 +0800 | [diff] [blame] | 879 | /* |
Igor Mammedov | 0e60b07 | 2014-12-01 17:29:26 +0000 | [diff] [blame] | 880 | * Insert memslot and re-sort memslots based on their GFN, |
| 881 | * so binary search could be used to lookup GFN. |
| 882 | * Sorting algorithm takes advantage of having initially |
| 883 | * sorted array and known changed memslot position. |
Xiao Guangrong | bf3e05b | 2011-11-24 17:40:57 +0800 | [diff] [blame] | 884 | */ |
Paolo Bonzini | 5cc1502 | 2014-11-14 10:55:31 +0100 | [diff] [blame] | 885 | static void update_memslots(struct kvm_memslots *slots, |
Wei Yang | 31fc4f9 | 2018-08-22 21:57:11 +0800 | [diff] [blame] | 886 | struct kvm_memory_slot *new, |
| 887 | enum kvm_mr_change change) |
Xiao Guangrong | bf3e05b | 2011-11-24 17:40:57 +0800 | [diff] [blame] | 888 | { |
Paolo Bonzini | 8593176 | 2014-11-14 10:22:07 +0100 | [diff] [blame] | 889 | int id = new->id; |
| 890 | int i = slots->id_to_index[id]; |
Igor Mammedov | 063584d | 2014-11-13 23:00:13 +0000 | [diff] [blame] | 891 | struct kvm_memory_slot *mslots = slots->memslots; |
Xiao Guangrong | f85e2cb | 2011-11-24 17:41:54 +0800 | [diff] [blame] | 892 | |
Paolo Bonzini | 8593176 | 2014-11-14 10:22:07 +0100 | [diff] [blame] | 893 | WARN_ON(mslots[i].id != id); |
Wei Yang | 31fc4f9 | 2018-08-22 21:57:11 +0800 | [diff] [blame] | 894 | switch (change) { |
| 895 | case KVM_MR_CREATE: |
| 896 | slots->used_slots++; |
| 897 | WARN_ON(mslots[i].npages || !new->npages); |
| 898 | break; |
| 899 | case KVM_MR_DELETE: |
| 900 | slots->used_slots--; |
| 901 | WARN_ON(new->npages || !mslots[i].npages); |
| 902 | break; |
| 903 | default: |
| 904 | break; |
Igor Mammedov | 9c1a5d38 | 2014-12-01 17:29:27 +0000 | [diff] [blame] | 905 | } |
Igor Mammedov | 0e60b07 | 2014-12-01 17:29:26 +0000 | [diff] [blame] | 906 | |
Igor Mammedov | 7f379cf | 2014-12-01 17:29:24 +0000 | [diff] [blame] | 907 | while (i < KVM_MEM_SLOTS_NUM - 1 && |
Igor Mammedov | 0e60b07 | 2014-12-01 17:29:26 +0000 | [diff] [blame] | 908 | new->base_gfn <= mslots[i + 1].base_gfn) { |
| 909 | if (!mslots[i + 1].npages) |
| 910 | break; |
Igor Mammedov | 7f379cf | 2014-12-01 17:29:24 +0000 | [diff] [blame] | 911 | mslots[i] = mslots[i + 1]; |
| 912 | slots->id_to_index[mslots[i].id] = i; |
| 913 | i++; |
| 914 | } |
Paolo Bonzini | efbeec7 | 2014-12-27 18:01:00 +0100 | [diff] [blame] | 915 | |
| 916 | /* |
| 917 | * The ">=" is needed when creating a slot with base_gfn == 0, |
| 918 | * so that it moves before all those with base_gfn == npages == 0. |
| 919 | * |
| 920 | * On the other hand, if new->npages is zero, the above loop has |
| 921 | * already left i pointing to the beginning of the empty part of |
| 922 | * mslots, and the ">=" would move the hole backwards in this |
| 923 | * case---which is wrong. So skip the loop when deleting a slot. |
| 924 | */ |
| 925 | if (new->npages) { |
| 926 | while (i > 0 && |
| 927 | new->base_gfn >= mslots[i - 1].base_gfn) { |
| 928 | mslots[i] = mslots[i - 1]; |
| 929 | slots->id_to_index[mslots[i].id] = i; |
| 930 | i--; |
| 931 | } |
Paolo Bonzini | dbaff30 | 2014-12-27 21:08:16 +0100 | [diff] [blame] | 932 | } else |
| 933 | WARN_ON_ONCE(i != slots->used_slots); |
Xiao Guangrong | f85e2cb | 2011-11-24 17:41:54 +0800 | [diff] [blame] | 934 | |
Paolo Bonzini | 8593176 | 2014-11-14 10:22:07 +0100 | [diff] [blame] | 935 | mslots[i] = *new; |
| 936 | slots->id_to_index[mslots[i].id] = i; |
Xiao Guangrong | bf3e05b | 2011-11-24 17:40:57 +0800 | [diff] [blame] | 937 | } |
| 938 | |
Paolo Bonzini | 09170a4 | 2015-05-18 13:59:39 +0200 | [diff] [blame] | 939 | static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem) |
Xiao Guangrong | a50d64d | 2012-08-21 10:58:13 +0800 | [diff] [blame] | 940 | { |
Xiao Guangrong | 4d8b81a | 2012-08-21 11:02:51 +0800 | [diff] [blame] | 941 | u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES; |
| 942 | |
Christoffer Dall | 0f8a4de | 2014-08-26 14:00:37 +0200 | [diff] [blame] | 943 | #ifdef __KVM_HAVE_READONLY_MEM |
Xiao Guangrong | 4d8b81a | 2012-08-21 11:02:51 +0800 | [diff] [blame] | 944 | valid_flags |= KVM_MEM_READONLY; |
| 945 | #endif |
| 946 | |
| 947 | if (mem->flags & ~valid_flags) |
Xiao Guangrong | a50d64d | 2012-08-21 10:58:13 +0800 | [diff] [blame] | 948 | return -EINVAL; |
| 949 | |
| 950 | return 0; |
| 951 | } |
| 952 | |
Gleb Natapov | 7ec4fb4 | 2012-12-24 17:49:30 +0200 | [diff] [blame] | 953 | static struct kvm_memslots *install_new_memslots(struct kvm *kvm, |
Paolo Bonzini | f481b06 | 2015-05-17 17:30:37 +0200 | [diff] [blame] | 954 | int as_id, struct kvm_memslots *slots) |
Gleb Natapov | 7ec4fb4 | 2012-12-24 17:49:30 +0200 | [diff] [blame] | 955 | { |
Paolo Bonzini | f481b06 | 2015-05-17 17:30:37 +0200 | [diff] [blame] | 956 | struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id); |
Sean Christopherson | 361209e | 2019-02-05 13:01:14 -0800 | [diff] [blame] | 957 | u64 gen = old_memslots->generation; |
Gleb Natapov | 7ec4fb4 | 2012-12-24 17:49:30 +0200 | [diff] [blame] | 958 | |
Sean Christopherson | 361209e | 2019-02-05 13:01:14 -0800 | [diff] [blame] | 959 | WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS); |
| 960 | slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS; |
David Matlack | ee3d157 | 2014-08-18 15:46:06 -0700 | [diff] [blame] | 961 | |
Paolo Bonzini | f481b06 | 2015-05-17 17:30:37 +0200 | [diff] [blame] | 962 | rcu_assign_pointer(kvm->memslots[as_id], slots); |
Gleb Natapov | 7ec4fb4 | 2012-12-24 17:49:30 +0200 | [diff] [blame] | 963 | synchronize_srcu_expedited(&kvm->srcu); |
Takuya Yoshikawa | e59dbe0 | 2013-07-04 13:40:29 +0900 | [diff] [blame] | 964 | |
David Matlack | ee3d157 | 2014-08-18 15:46:06 -0700 | [diff] [blame] | 965 | /* |
Sean Christopherson | 361209e | 2019-02-05 13:01:14 -0800 | [diff] [blame] | 966 | * Increment the new memslot generation a second time, dropping the |
| 967 | * update in-progress flag and incrementing then generation based on |
| 968 | * the number of address spaces. This provides a unique and easily |
| 969 | * identifiable generation number while the memslots are in flux. |
| 970 | */ |
| 971 | gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS; |
| 972 | |
| 973 | /* |
Paolo Bonzini | 4bd518f | 2017-02-03 20:44:51 -0800 | [diff] [blame] | 974 | * Generations must be unique even across address spaces. We do not need |
| 975 | * a global counter for that, instead the generation space is evenly split |
| 976 | * across address spaces. For example, with two address spaces, address |
Sean Christopherson | 164bf7e | 2019-02-05 13:01:18 -0800 | [diff] [blame] | 977 | * space 0 will use generations 0, 2, 4, ... while address space 1 will |
| 978 | * use generations 1, 3, 5, ... |
David Matlack | ee3d157 | 2014-08-18 15:46:06 -0700 | [diff] [blame] | 979 | */ |
Sean Christopherson | 164bf7e | 2019-02-05 13:01:18 -0800 | [diff] [blame] | 980 | gen += KVM_ADDRESS_SPACE_NUM; |
David Matlack | ee3d157 | 2014-08-18 15:46:06 -0700 | [diff] [blame] | 981 | |
Sean Christopherson | 1524825 | 2019-02-05 12:54:17 -0800 | [diff] [blame] | 982 | kvm_arch_memslots_updated(kvm, gen); |
| 983 | |
| 984 | slots->generation = gen; |
Takuya Yoshikawa | e59dbe0 | 2013-07-04 13:40:29 +0900 | [diff] [blame] | 985 | |
| 986 | return old_memslots; |
Gleb Natapov | 7ec4fb4 | 2012-12-24 17:49:30 +0200 | [diff] [blame] | 987 | } |
| 988 | |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 989 | /* |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 990 | * Allocate some memory and give it an address in the guest physical address |
| 991 | * space. |
| 992 | * |
| 993 | * Discontiguous memory is allowed, mostly for framebuffers. |
Sheng Yang | f78e0e2 | 2007-10-29 09:40:42 +0800 | [diff] [blame] | 994 | * |
Dominik Dingel | 02d5d55 | 2014-10-27 16:22:56 +0100 | [diff] [blame] | 995 | * Must be called holding kvm->slots_lock for write. |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 996 | */ |
Sheng Yang | f78e0e2 | 2007-10-29 09:40:42 +0800 | [diff] [blame] | 997 | int __kvm_set_memory_region(struct kvm *kvm, |
Paolo Bonzini | 09170a4 | 2015-05-18 13:59:39 +0200 | [diff] [blame] | 998 | const struct kvm_userspace_memory_region *mem) |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 999 | { |
Avi Kivity | 8234b22 | 2010-12-27 12:08:45 +0200 | [diff] [blame] | 1000 | int r; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1001 | gfn_t base_gfn; |
Heiko Carstens | 28bcb11 | 2009-09-03 17:35:35 +0200 | [diff] [blame] | 1002 | unsigned long npages; |
Takuya Yoshikawa | a843fac | 2013-01-11 18:27:43 +0900 | [diff] [blame] | 1003 | struct kvm_memory_slot *slot; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1004 | struct kvm_memory_slot old, new; |
Alex Williamson | b7f69c5 | 2012-12-10 10:33:03 -0700 | [diff] [blame] | 1005 | struct kvm_memslots *slots = NULL, *old_memslots; |
Paolo Bonzini | f481b06 | 2015-05-17 17:30:37 +0200 | [diff] [blame] | 1006 | int as_id, id; |
Takuya Yoshikawa | f64c039 | 2013-01-29 11:00:07 +0900 | [diff] [blame] | 1007 | enum kvm_mr_change change; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1008 | |
Xiao Guangrong | a50d64d | 2012-08-21 10:58:13 +0800 | [diff] [blame] | 1009 | r = check_memory_region_flags(mem); |
| 1010 | if (r) |
| 1011 | goto out; |
| 1012 | |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1013 | r = -EINVAL; |
Paolo Bonzini | f481b06 | 2015-05-17 17:30:37 +0200 | [diff] [blame] | 1014 | as_id = mem->slot >> 16; |
| 1015 | id = (u16)mem->slot; |
| 1016 | |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1017 | /* General sanity checks */ |
| 1018 | if (mem->memory_size & (PAGE_SIZE - 1)) |
| 1019 | goto out; |
| 1020 | if (mem->guest_phys_addr & (PAGE_SIZE - 1)) |
| 1021 | goto out; |
Takuya Yoshikawa | fa3d315 | 2011-05-07 16:35:38 +0900 | [diff] [blame] | 1022 | /* We can read the guest memory with __xxx_user() later on. */ |
Paolo Bonzini | f481b06 | 2015-05-17 17:30:37 +0200 | [diff] [blame] | 1023 | if ((id < KVM_USER_MEM_SLOTS) && |
Takuya Yoshikawa | fa3d315 | 2011-05-07 16:35:38 +0900 | [diff] [blame] | 1024 | ((mem->userspace_addr & (PAGE_SIZE - 1)) || |
Linus Torvalds | 96d4f26 | 2019-01-03 18:57:57 -0800 | [diff] [blame] | 1025 | !access_ok((void __user *)(unsigned long)mem->userspace_addr, |
Heiko Carstens | 9e3bb6b | 2011-05-24 07:51:27 +0200 | [diff] [blame] | 1026 | mem->memory_size))) |
Hollis Blanchard | 7874980 | 2008-11-07 13:32:12 -0600 | [diff] [blame] | 1027 | goto out; |
Paolo Bonzini | f481b06 | 2015-05-17 17:30:37 +0200 | [diff] [blame] | 1028 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM) |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1029 | goto out; |
| 1030 | if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr) |
| 1031 | goto out; |
| 1032 | |
Paolo Bonzini | f481b06 | 2015-05-17 17:30:37 +0200 | [diff] [blame] | 1033 | slot = id_to_memslot(__kvm_memslots(kvm, as_id), id); |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1034 | base_gfn = mem->guest_phys_addr >> PAGE_SHIFT; |
| 1035 | npages = mem->memory_size >> PAGE_SHIFT; |
| 1036 | |
Takuya Yoshikawa | 660c22c | 2010-04-13 22:47:24 +0900 | [diff] [blame] | 1037 | if (npages > KVM_MEM_MAX_NR_PAGES) |
| 1038 | goto out; |
| 1039 | |
Takuya Yoshikawa | a843fac | 2013-01-11 18:27:43 +0900 | [diff] [blame] | 1040 | new = old = *slot; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1041 | |
Paolo Bonzini | f481b06 | 2015-05-17 17:30:37 +0200 | [diff] [blame] | 1042 | new.id = id; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1043 | new.base_gfn = base_gfn; |
| 1044 | new.npages = npages; |
| 1045 | new.flags = mem->flags; |
| 1046 | |
Takuya Yoshikawa | f64c039 | 2013-01-29 11:00:07 +0900 | [diff] [blame] | 1047 | if (npages) { |
| 1048 | if (!old.npages) |
| 1049 | change = KVM_MR_CREATE; |
| 1050 | else { /* Modify an existing slot. */ |
| 1051 | if ((mem->userspace_addr != old.userspace_addr) || |
Takuya Yoshikawa | 75d61fb | 2013-01-30 19:40:41 +0900 | [diff] [blame] | 1052 | (npages != old.npages) || |
| 1053 | ((new.flags ^ old.flags) & KVM_MEM_READONLY)) |
Takuya Yoshikawa | f64c039 | 2013-01-29 11:00:07 +0900 | [diff] [blame] | 1054 | goto out; |
| 1055 | |
| 1056 | if (base_gfn != old.base_gfn) |
| 1057 | change = KVM_MR_MOVE; |
| 1058 | else if (new.flags != old.flags) |
| 1059 | change = KVM_MR_FLAGS_ONLY; |
| 1060 | else { /* Nothing to change. */ |
| 1061 | r = 0; |
| 1062 | goto out; |
| 1063 | } |
| 1064 | } |
Paolo Bonzini | 09170a4 | 2015-05-18 13:59:39 +0200 | [diff] [blame] | 1065 | } else { |
| 1066 | if (!old.npages) |
| 1067 | goto out; |
| 1068 | |
Takuya Yoshikawa | f64c039 | 2013-01-29 11:00:07 +0900 | [diff] [blame] | 1069 | change = KVM_MR_DELETE; |
Paolo Bonzini | 09170a4 | 2015-05-18 13:59:39 +0200 | [diff] [blame] | 1070 | new.base_gfn = 0; |
| 1071 | new.flags = 0; |
| 1072 | } |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1073 | |
Takuya Yoshikawa | f64c039 | 2013-01-29 11:00:07 +0900 | [diff] [blame] | 1074 | if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) { |
Takuya Yoshikawa | 0a706be | 2013-01-11 18:26:55 +0900 | [diff] [blame] | 1075 | /* Check for overlaps */ |
| 1076 | r = -EEXIST; |
Paolo Bonzini | f481b06 | 2015-05-17 17:30:37 +0200 | [diff] [blame] | 1077 | kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) { |
Wanpeng Li | b28676b | 2018-02-13 15:36:00 +0100 | [diff] [blame] | 1078 | if (slot->id == id) |
Takuya Yoshikawa | 0a706be | 2013-01-11 18:26:55 +0900 | [diff] [blame] | 1079 | continue; |
| 1080 | if (!((base_gfn + npages <= slot->base_gfn) || |
| 1081 | (base_gfn >= slot->base_gfn + slot->npages))) |
| 1082 | goto out; |
| 1083 | } |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1084 | } |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1085 | |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1086 | /* Free page dirty bitmap if unneeded */ |
| 1087 | if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES)) |
Al Viro | 8b6d44c | 2007-02-09 16:38:40 +0000 | [diff] [blame] | 1088 | new.dirty_bitmap = NULL; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1089 | |
| 1090 | r = -ENOMEM; |
Takuya Yoshikawa | f64c039 | 2013-01-29 11:00:07 +0900 | [diff] [blame] | 1091 | if (change == KVM_MR_CREATE) { |
Takuya Yoshikawa | 189a2f7 | 2012-02-08 13:01:09 +0900 | [diff] [blame] | 1092 | new.userspace_addr = mem->userspace_addr; |
Takuya Yoshikawa | d89cc61 | 2012-08-01 18:03:28 +0900 | [diff] [blame] | 1093 | |
Aneesh Kumar K.V | 5587027 | 2013-10-07 22:18:00 +0530 | [diff] [blame] | 1094 | if (kvm_arch_create_memslot(kvm, &new, npages)) |
Takuya Yoshikawa | db3fe4e | 2012-02-08 13:02:18 +0900 | [diff] [blame] | 1095 | goto out_free; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1096 | } |
Joerg Roedel | ec04b26 | 2009-06-19 15:16:23 +0200 | [diff] [blame] | 1097 | |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1098 | /* Allocate page dirty bitmap if needed */ |
| 1099 | if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) { |
Takuya Yoshikawa | a36a57b1 | 2010-10-27 18:22:19 +0900 | [diff] [blame] | 1100 | if (kvm_create_dirty_bitmap(&new) < 0) |
Sheng Yang | f78e0e2 | 2007-10-29 09:40:42 +0800 | [diff] [blame] | 1101 | goto out_free; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1102 | } |
| 1103 | |
Ben Gardon | b12ce36 | 2019-02-11 11:02:49 -0800 | [diff] [blame] | 1104 | slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT); |
Paolo Bonzini | f2a8103 | 2014-11-14 10:46:45 +0100 | [diff] [blame] | 1105 | if (!slots) |
| 1106 | goto out_free; |
Paolo Bonzini | f481b06 | 2015-05-17 17:30:37 +0200 | [diff] [blame] | 1107 | memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots)); |
Paolo Bonzini | f2a8103 | 2014-11-14 10:46:45 +0100 | [diff] [blame] | 1108 | |
Takuya Yoshikawa | f64c039 | 2013-01-29 11:00:07 +0900 | [diff] [blame] | 1109 | if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) { |
Paolo Bonzini | f481b06 | 2015-05-17 17:30:37 +0200 | [diff] [blame] | 1110 | slot = id_to_memslot(slots, id); |
Xiao Guangrong | 28a3754 | 2011-11-24 19:04:35 +0800 | [diff] [blame] | 1111 | slot->flags |= KVM_MEMSLOT_INVALID; |
| 1112 | |
Paolo Bonzini | f481b06 | 2015-05-17 17:30:37 +0200 | [diff] [blame] | 1113 | old_memslots = install_new_memslots(kvm, as_id, slots); |
Marcelo Tosatti | bc6678a | 2009-12-23 14:35:21 -0200 | [diff] [blame] | 1114 | |
Marcelo Tosatti | 12d6e75 | 2012-08-24 15:54:58 -0300 | [diff] [blame] | 1115 | /* From this point no new shadow pages pointing to a deleted, |
| 1116 | * or moved, memslot will be created. |
Marcelo Tosatti | bc6678a | 2009-12-23 14:35:21 -0200 | [diff] [blame] | 1117 | * |
| 1118 | * validation of sp->gfn happens in: |
Xiubo Li | b7d409d | 2015-02-26 14:58:24 +0800 | [diff] [blame] | 1119 | * - gfn_to_hva (kvm_read_guest, gfn_to_pfn) |
| 1120 | * - kvm_is_visible_gfn (mmu_check_roots) |
Marcelo Tosatti | bc6678a | 2009-12-23 14:35:21 -0200 | [diff] [blame] | 1121 | */ |
Marcelo Tosatti | 2df72e9 | 2012-08-24 15:54:57 -0300 | [diff] [blame] | 1122 | kvm_arch_flush_shadow_memslot(kvm, slot); |
Paolo Bonzini | f2a8103 | 2014-11-14 10:46:45 +0100 | [diff] [blame] | 1123 | |
| 1124 | /* |
| 1125 | * We can re-use the old_memslots from above, the only difference |
| 1126 | * from the currently installed memslots is the invalid flag. This |
| 1127 | * will get overwritten by update_memslots anyway. |
| 1128 | */ |
Alex Williamson | b7f69c5 | 2012-12-10 10:33:03 -0700 | [diff] [blame] | 1129 | slots = old_memslots; |
Marcelo Tosatti | bc6678a | 2009-12-23 14:35:21 -0200 | [diff] [blame] | 1130 | } |
Marcelo Tosatti | 34d4cb8 | 2008-07-10 20:49:31 -0300 | [diff] [blame] | 1131 | |
Takuya Yoshikawa | 7b6195a | 2013-02-27 19:44:34 +0900 | [diff] [blame] | 1132 | r = kvm_arch_prepare_memory_region(kvm, &new, mem, change); |
Marcelo Tosatti | f7784b8 | 2009-12-23 14:35:18 -0200 | [diff] [blame] | 1133 | if (r) |
Alex Williamson | b7f69c5 | 2012-12-10 10:33:03 -0700 | [diff] [blame] | 1134 | goto out_slots; |
Marcelo Tosatti | f7784b8 | 2009-12-23 14:35:18 -0200 | [diff] [blame] | 1135 | |
Paolo Bonzini | a47d2b0 | 2015-05-17 11:41:37 +0200 | [diff] [blame] | 1136 | /* actual memory is freed via old in kvm_free_memslot below */ |
Takuya Yoshikawa | f64c039 | 2013-01-29 11:00:07 +0900 | [diff] [blame] | 1137 | if (change == KVM_MR_DELETE) { |
Marcelo Tosatti | bc6678a | 2009-12-23 14:35:21 -0200 | [diff] [blame] | 1138 | new.dirty_bitmap = NULL; |
Takuya Yoshikawa | db3fe4e | 2012-02-08 13:02:18 +0900 | [diff] [blame] | 1139 | memset(&new.arch, 0, sizeof(new.arch)); |
Marcelo Tosatti | bc6678a | 2009-12-23 14:35:21 -0200 | [diff] [blame] | 1140 | } |
| 1141 | |
Wei Yang | 31fc4f9 | 2018-08-22 21:57:11 +0800 | [diff] [blame] | 1142 | update_memslots(slots, &new, change); |
Paolo Bonzini | f481b06 | 2015-05-17 17:30:37 +0200 | [diff] [blame] | 1143 | old_memslots = install_new_memslots(kvm, as_id, slots); |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1144 | |
Paolo Bonzini | f36f3f2 | 2015-05-18 13:20:23 +0200 | [diff] [blame] | 1145 | kvm_arch_commit_memory_region(kvm, mem, &old, &new, change); |
Zhang Xiantao | 3ad82a7 | 2007-11-20 13:11:38 +0800 | [diff] [blame] | 1146 | |
Paolo Bonzini | a47d2b0 | 2015-05-17 11:41:37 +0200 | [diff] [blame] | 1147 | kvm_free_memslot(kvm, &old, &new); |
Igor Mammedov | 7449613 | 2015-03-20 12:21:37 +0000 | [diff] [blame] | 1148 | kvfree(old_memslots); |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1149 | return 0; |
| 1150 | |
Alex Williamson | e40f193 | 2012-12-10 10:32:57 -0700 | [diff] [blame] | 1151 | out_slots: |
Igor Mammedov | 7449613 | 2015-03-20 12:21:37 +0000 | [diff] [blame] | 1152 | kvfree(slots); |
Sheng Yang | f78e0e2 | 2007-10-29 09:40:42 +0800 | [diff] [blame] | 1153 | out_free: |
Paolo Bonzini | a47d2b0 | 2015-05-17 11:41:37 +0200 | [diff] [blame] | 1154 | kvm_free_memslot(kvm, &new, &old); |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1155 | out: |
| 1156 | return r; |
Izik Eidus | 210c7c4 | 2007-10-24 23:52:57 +0200 | [diff] [blame] | 1157 | } |
Sheng Yang | f78e0e2 | 2007-10-29 09:40:42 +0800 | [diff] [blame] | 1158 | EXPORT_SYMBOL_GPL(__kvm_set_memory_region); |
| 1159 | |
| 1160 | int kvm_set_memory_region(struct kvm *kvm, |
Paolo Bonzini | 09170a4 | 2015-05-18 13:59:39 +0200 | [diff] [blame] | 1161 | const struct kvm_userspace_memory_region *mem) |
Sheng Yang | f78e0e2 | 2007-10-29 09:40:42 +0800 | [diff] [blame] | 1162 | { |
| 1163 | int r; |
| 1164 | |
Marcelo Tosatti | 79fac95 | 2009-12-23 14:35:26 -0200 | [diff] [blame] | 1165 | mutex_lock(&kvm->slots_lock); |
Takuya Yoshikawa | 47ae31e | 2013-02-27 19:43:00 +0900 | [diff] [blame] | 1166 | r = __kvm_set_memory_region(kvm, mem); |
Marcelo Tosatti | 79fac95 | 2009-12-23 14:35:26 -0200 | [diff] [blame] | 1167 | mutex_unlock(&kvm->slots_lock); |
Sheng Yang | f78e0e2 | 2007-10-29 09:40:42 +0800 | [diff] [blame] | 1168 | return r; |
| 1169 | } |
Izik Eidus | 210c7c4 | 2007-10-24 23:52:57 +0200 | [diff] [blame] | 1170 | EXPORT_SYMBOL_GPL(kvm_set_memory_region); |
| 1171 | |
Stephen Hemminger | 7940876 | 2013-12-29 12:12:29 -0800 | [diff] [blame] | 1172 | static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm, |
| 1173 | struct kvm_userspace_memory_region *mem) |
Izik Eidus | 210c7c4 | 2007-10-24 23:52:57 +0200 | [diff] [blame] | 1174 | { |
Paolo Bonzini | f481b06 | 2015-05-17 17:30:37 +0200 | [diff] [blame] | 1175 | if ((u16)mem->slot >= KVM_USER_MEM_SLOTS) |
Izik Eidus | e0d62c7 | 2007-10-24 23:57:46 +0200 | [diff] [blame] | 1176 | return -EINVAL; |
Paolo Bonzini | 09170a4 | 2015-05-18 13:59:39 +0200 | [diff] [blame] | 1177 | |
Takuya Yoshikawa | 47ae31e | 2013-02-27 19:43:00 +0900 | [diff] [blame] | 1178 | return kvm_set_memory_region(kvm, mem); |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1179 | } |
| 1180 | |
Zhang Xiantao | 5bb064d | 2007-11-18 20:29:43 +0800 | [diff] [blame] | 1181 | int kvm_get_dirty_log(struct kvm *kvm, |
| 1182 | struct kvm_dirty_log *log, int *is_dirty) |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1183 | { |
Paolo Bonzini | 9f6b802 | 2015-05-17 16:20:07 +0200 | [diff] [blame] | 1184 | struct kvm_memslots *slots; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1185 | struct kvm_memory_slot *memslot; |
Markus Elfring | 843574a | 2017-01-22 17:41:07 +0100 | [diff] [blame] | 1186 | int i, as_id, id; |
Takuya Yoshikawa | 87bf6e7 | 2010-04-12 19:35:35 +0900 | [diff] [blame] | 1187 | unsigned long n; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1188 | unsigned long any = 0; |
| 1189 | |
Paolo Bonzini | f481b06 | 2015-05-17 17:30:37 +0200 | [diff] [blame] | 1190 | as_id = log->slot >> 16; |
| 1191 | id = (u16)log->slot; |
| 1192 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS) |
Markus Elfring | 843574a | 2017-01-22 17:41:07 +0100 | [diff] [blame] | 1193 | return -EINVAL; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1194 | |
Paolo Bonzini | f481b06 | 2015-05-17 17:30:37 +0200 | [diff] [blame] | 1195 | slots = __kvm_memslots(kvm, as_id); |
| 1196 | memslot = id_to_memslot(slots, id); |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1197 | if (!memslot->dirty_bitmap) |
Markus Elfring | 843574a | 2017-01-22 17:41:07 +0100 | [diff] [blame] | 1198 | return -ENOENT; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1199 | |
Takuya Yoshikawa | 87bf6e7 | 2010-04-12 19:35:35 +0900 | [diff] [blame] | 1200 | n = kvm_dirty_bitmap_bytes(memslot); |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1201 | |
Uri Lublin | cd1a4a9 | 2007-02-22 16:43:09 +0200 | [diff] [blame] | 1202 | for (i = 0; !any && i < n/sizeof(long); ++i) |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1203 | any = memslot->dirty_bitmap[i]; |
| 1204 | |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1205 | if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n)) |
Markus Elfring | 843574a | 2017-01-22 17:41:07 +0100 | [diff] [blame] | 1206 | return -EFAULT; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1207 | |
Zhang Xiantao | 5bb064d | 2007-11-18 20:29:43 +0800 | [diff] [blame] | 1208 | if (any) |
| 1209 | *is_dirty = 1; |
Markus Elfring | 843574a | 2017-01-22 17:41:07 +0100 | [diff] [blame] | 1210 | return 0; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1211 | } |
Aneesh Kumar K.V | 2ba9f0d | 2013-10-07 22:17:59 +0530 | [diff] [blame] | 1212 | EXPORT_SYMBOL_GPL(kvm_get_dirty_log); |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1213 | |
Mario Smarduch | ba0513b | 2015-01-15 15:58:53 -0800 | [diff] [blame] | 1214 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
| 1215 | /** |
Jiang Biao | b8b0022 | 2019-04-23 19:40:30 +0800 | [diff] [blame] | 1216 | * kvm_get_dirty_log_protect - get a snapshot of dirty pages |
Paolo Bonzini | 2a31b9d | 2018-10-23 02:36:47 +0200 | [diff] [blame] | 1217 | * and reenable dirty page tracking for the corresponding pages. |
Mario Smarduch | ba0513b | 2015-01-15 15:58:53 -0800 | [diff] [blame] | 1218 | * @kvm: pointer to kvm instance |
| 1219 | * @log: slot id and address to which we copy the log |
Jiang Biao | b8b0022 | 2019-04-23 19:40:30 +0800 | [diff] [blame] | 1220 | * @flush: true if TLB flush is needed by caller |
Mario Smarduch | ba0513b | 2015-01-15 15:58:53 -0800 | [diff] [blame] | 1221 | * |
| 1222 | * We need to keep it in mind that VCPU threads can write to the bitmap |
| 1223 | * concurrently. So, to avoid losing track of dirty pages we keep the |
| 1224 | * following order: |
| 1225 | * |
| 1226 | * 1. Take a snapshot of the bit and clear it if needed. |
| 1227 | * 2. Write protect the corresponding page. |
| 1228 | * 3. Copy the snapshot to the userspace. |
| 1229 | * 4. Upon return caller flushes TLB's if needed. |
| 1230 | * |
| 1231 | * Between 2 and 4, the guest may write to the page using the remaining TLB |
| 1232 | * entry. This is not a problem because the page is reported dirty using |
| 1233 | * the snapshot taken before and step 4 ensures that writes done after |
| 1234 | * exiting to userspace will be logged for the next call. |
| 1235 | * |
| 1236 | */ |
| 1237 | int kvm_get_dirty_log_protect(struct kvm *kvm, |
Paolo Bonzini | 8fe65a8 | 2018-10-23 02:18:42 +0200 | [diff] [blame] | 1238 | struct kvm_dirty_log *log, bool *flush) |
Mario Smarduch | ba0513b | 2015-01-15 15:58:53 -0800 | [diff] [blame] | 1239 | { |
Paolo Bonzini | 9f6b802 | 2015-05-17 16:20:07 +0200 | [diff] [blame] | 1240 | struct kvm_memslots *slots; |
Mario Smarduch | ba0513b | 2015-01-15 15:58:53 -0800 | [diff] [blame] | 1241 | struct kvm_memory_slot *memslot; |
Markus Elfring | 58d6db3 | 2017-01-22 17:30:16 +0100 | [diff] [blame] | 1242 | int i, as_id, id; |
Mario Smarduch | ba0513b | 2015-01-15 15:58:53 -0800 | [diff] [blame] | 1243 | unsigned long n; |
| 1244 | unsigned long *dirty_bitmap; |
| 1245 | unsigned long *dirty_bitmap_buffer; |
| 1246 | |
Paolo Bonzini | f481b06 | 2015-05-17 17:30:37 +0200 | [diff] [blame] | 1247 | as_id = log->slot >> 16; |
| 1248 | id = (u16)log->slot; |
| 1249 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS) |
Markus Elfring | 58d6db3 | 2017-01-22 17:30:16 +0100 | [diff] [blame] | 1250 | return -EINVAL; |
Mario Smarduch | ba0513b | 2015-01-15 15:58:53 -0800 | [diff] [blame] | 1251 | |
Paolo Bonzini | f481b06 | 2015-05-17 17:30:37 +0200 | [diff] [blame] | 1252 | slots = __kvm_memslots(kvm, as_id); |
| 1253 | memslot = id_to_memslot(slots, id); |
Mario Smarduch | ba0513b | 2015-01-15 15:58:53 -0800 | [diff] [blame] | 1254 | |
| 1255 | dirty_bitmap = memslot->dirty_bitmap; |
Mario Smarduch | ba0513b | 2015-01-15 15:58:53 -0800 | [diff] [blame] | 1256 | if (!dirty_bitmap) |
Markus Elfring | 58d6db3 | 2017-01-22 17:30:16 +0100 | [diff] [blame] | 1257 | return -ENOENT; |
Mario Smarduch | ba0513b | 2015-01-15 15:58:53 -0800 | [diff] [blame] | 1258 | |
| 1259 | n = kvm_dirty_bitmap_bytes(memslot); |
Paolo Bonzini | 8fe65a8 | 2018-10-23 02:18:42 +0200 | [diff] [blame] | 1260 | *flush = false; |
Paolo Bonzini | 2a31b9d | 2018-10-23 02:36:47 +0200 | [diff] [blame] | 1261 | if (kvm->manual_dirty_log_protect) { |
| 1262 | /* |
| 1263 | * Unlike kvm_get_dirty_log, we always return false in *flush, |
| 1264 | * because no flush is needed until KVM_CLEAR_DIRTY_LOG. There |
| 1265 | * is some code duplication between this function and |
| 1266 | * kvm_get_dirty_log, but hopefully all architecture |
| 1267 | * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log |
| 1268 | * can be eliminated. |
| 1269 | */ |
| 1270 | dirty_bitmap_buffer = dirty_bitmap; |
| 1271 | } else { |
| 1272 | dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot); |
| 1273 | memset(dirty_bitmap_buffer, 0, n); |
Mario Smarduch | ba0513b | 2015-01-15 15:58:53 -0800 | [diff] [blame] | 1274 | |
Paolo Bonzini | 2a31b9d | 2018-10-23 02:36:47 +0200 | [diff] [blame] | 1275 | spin_lock(&kvm->mmu_lock); |
| 1276 | for (i = 0; i < n / sizeof(long); i++) { |
| 1277 | unsigned long mask; |
| 1278 | gfn_t offset; |
Mario Smarduch | ba0513b | 2015-01-15 15:58:53 -0800 | [diff] [blame] | 1279 | |
Paolo Bonzini | 2a31b9d | 2018-10-23 02:36:47 +0200 | [diff] [blame] | 1280 | if (!dirty_bitmap[i]) |
| 1281 | continue; |
Mario Smarduch | ba0513b | 2015-01-15 15:58:53 -0800 | [diff] [blame] | 1282 | |
Paolo Bonzini | 2a31b9d | 2018-10-23 02:36:47 +0200 | [diff] [blame] | 1283 | *flush = true; |
| 1284 | mask = xchg(&dirty_bitmap[i], 0); |
| 1285 | dirty_bitmap_buffer[i] = mask; |
Mario Smarduch | ba0513b | 2015-01-15 15:58:53 -0800 | [diff] [blame] | 1286 | |
Lan Tianyu | a67794c | 2019-02-02 17:20:27 +0800 | [diff] [blame] | 1287 | offset = i * BITS_PER_LONG; |
| 1288 | kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot, |
| 1289 | offset, mask); |
Takuya Yoshikawa | 58d2930 | 2015-03-17 16:19:58 +0900 | [diff] [blame] | 1290 | } |
Paolo Bonzini | 2a31b9d | 2018-10-23 02:36:47 +0200 | [diff] [blame] | 1291 | spin_unlock(&kvm->mmu_lock); |
Mario Smarduch | ba0513b | 2015-01-15 15:58:53 -0800 | [diff] [blame] | 1292 | } |
| 1293 | |
Mario Smarduch | ba0513b | 2015-01-15 15:58:53 -0800 | [diff] [blame] | 1294 | if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n)) |
Markus Elfring | 58d6db3 | 2017-01-22 17:30:16 +0100 | [diff] [blame] | 1295 | return -EFAULT; |
| 1296 | return 0; |
Mario Smarduch | ba0513b | 2015-01-15 15:58:53 -0800 | [diff] [blame] | 1297 | } |
| 1298 | EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect); |
Paolo Bonzini | 2a31b9d | 2018-10-23 02:36:47 +0200 | [diff] [blame] | 1299 | |
| 1300 | /** |
| 1301 | * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap |
| 1302 | * and reenable dirty page tracking for the corresponding pages. |
| 1303 | * @kvm: pointer to kvm instance |
| 1304 | * @log: slot id and address from which to fetch the bitmap of dirty pages |
Jiang Biao | b8b0022 | 2019-04-23 19:40:30 +0800 | [diff] [blame] | 1305 | * @flush: true if TLB flush is needed by caller |
Paolo Bonzini | 2a31b9d | 2018-10-23 02:36:47 +0200 | [diff] [blame] | 1306 | */ |
| 1307 | int kvm_clear_dirty_log_protect(struct kvm *kvm, |
| 1308 | struct kvm_clear_dirty_log *log, bool *flush) |
| 1309 | { |
| 1310 | struct kvm_memslots *slots; |
| 1311 | struct kvm_memory_slot *memslot; |
Tomas Bortoli | 98938aa | 2019-01-02 18:29:37 +0100 | [diff] [blame] | 1312 | int as_id, id; |
Paolo Bonzini | 2a31b9d | 2018-10-23 02:36:47 +0200 | [diff] [blame] | 1313 | gfn_t offset; |
Tomas Bortoli | 98938aa | 2019-01-02 18:29:37 +0100 | [diff] [blame] | 1314 | unsigned long i, n; |
Paolo Bonzini | 2a31b9d | 2018-10-23 02:36:47 +0200 | [diff] [blame] | 1315 | unsigned long *dirty_bitmap; |
| 1316 | unsigned long *dirty_bitmap_buffer; |
| 1317 | |
| 1318 | as_id = log->slot >> 16; |
| 1319 | id = (u16)log->slot; |
| 1320 | if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS) |
| 1321 | return -EINVAL; |
| 1322 | |
Paolo Bonzini | 76d58e0 | 2019-04-17 15:28:44 +0200 | [diff] [blame] | 1323 | if (log->first_page & 63) |
Paolo Bonzini | 2a31b9d | 2018-10-23 02:36:47 +0200 | [diff] [blame] | 1324 | return -EINVAL; |
| 1325 | |
| 1326 | slots = __kvm_memslots(kvm, as_id); |
| 1327 | memslot = id_to_memslot(slots, id); |
| 1328 | |
| 1329 | dirty_bitmap = memslot->dirty_bitmap; |
| 1330 | if (!dirty_bitmap) |
| 1331 | return -ENOENT; |
| 1332 | |
Peter Xu | 4ddc920 | 2019-05-08 17:15:45 +0800 | [diff] [blame] | 1333 | n = ALIGN(log->num_pages, BITS_PER_LONG) / 8; |
Tomas Bortoli | 98938aa | 2019-01-02 18:29:37 +0100 | [diff] [blame] | 1334 | |
| 1335 | if (log->first_page > memslot->npages || |
Paolo Bonzini | 76d58e0 | 2019-04-17 15:28:44 +0200 | [diff] [blame] | 1336 | log->num_pages > memslot->npages - log->first_page || |
| 1337 | (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63))) |
| 1338 | return -EINVAL; |
Tomas Bortoli | 98938aa | 2019-01-02 18:29:37 +0100 | [diff] [blame] | 1339 | |
Paolo Bonzini | 2a31b9d | 2018-10-23 02:36:47 +0200 | [diff] [blame] | 1340 | *flush = false; |
| 1341 | dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot); |
| 1342 | if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n)) |
| 1343 | return -EFAULT; |
| 1344 | |
| 1345 | spin_lock(&kvm->mmu_lock); |
Peter Xu | 53eac7a | 2019-05-08 17:15:46 +0800 | [diff] [blame] | 1346 | for (offset = log->first_page, i = offset / BITS_PER_LONG, |
| 1347 | n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--; |
Paolo Bonzini | 2a31b9d | 2018-10-23 02:36:47 +0200 | [diff] [blame] | 1348 | i++, offset += BITS_PER_LONG) { |
| 1349 | unsigned long mask = *dirty_bitmap_buffer++; |
| 1350 | atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i]; |
| 1351 | if (!mask) |
| 1352 | continue; |
| 1353 | |
| 1354 | mask &= atomic_long_fetch_andnot(mask, p); |
| 1355 | |
| 1356 | /* |
| 1357 | * mask contains the bits that really have been cleared. This |
| 1358 | * never includes any bits beyond the length of the memslot (if |
| 1359 | * the length is not aligned to 64 pages), therefore it is not |
| 1360 | * a problem if userspace sets them in log->dirty_bitmap. |
| 1361 | */ |
| 1362 | if (mask) { |
| 1363 | *flush = true; |
| 1364 | kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot, |
| 1365 | offset, mask); |
| 1366 | } |
| 1367 | } |
| 1368 | spin_unlock(&kvm->mmu_lock); |
| 1369 | |
| 1370 | return 0; |
| 1371 | } |
| 1372 | EXPORT_SYMBOL_GPL(kvm_clear_dirty_log_protect); |
Mario Smarduch | ba0513b | 2015-01-15 15:58:53 -0800 | [diff] [blame] | 1373 | #endif |
| 1374 | |
Takuya Yoshikawa | db3fe4e | 2012-02-08 13:02:18 +0900 | [diff] [blame] | 1375 | bool kvm_largepages_enabled(void) |
| 1376 | { |
| 1377 | return largepages_enabled; |
| 1378 | } |
| 1379 | |
Marcelo Tosatti | 54dee99 | 2009-06-11 12:07:44 -0300 | [diff] [blame] | 1380 | void kvm_disable_largepages(void) |
| 1381 | { |
| 1382 | largepages_enabled = false; |
| 1383 | } |
| 1384 | EXPORT_SYMBOL_GPL(kvm_disable_largepages); |
| 1385 | |
Gleb Natapov | 49c7754 | 2010-10-18 15:22:23 +0200 | [diff] [blame] | 1386 | struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn) |
| 1387 | { |
| 1388 | return __gfn_to_memslot(kvm_memslots(kvm), gfn); |
| 1389 | } |
Avi Kivity | a1f4d395 | 2010-06-21 11:44:20 +0300 | [diff] [blame] | 1390 | EXPORT_SYMBOL_GPL(gfn_to_memslot); |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 1391 | |
Paolo Bonzini | 8e73485 | 2015-05-17 13:58:53 +0200 | [diff] [blame] | 1392 | struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn) |
| 1393 | { |
| 1394 | return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn); |
| 1395 | } |
| 1396 | |
Yaowei Bai | 33e9415 | 2015-11-14 11:21:06 +0800 | [diff] [blame] | 1397 | bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn) |
Izik Eidus | e0d62c7 | 2007-10-24 23:57:46 +0200 | [diff] [blame] | 1398 | { |
Xiao Guangrong | bf3e05b | 2011-11-24 17:40:57 +0800 | [diff] [blame] | 1399 | struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn); |
Izik Eidus | e0d62c7 | 2007-10-24 23:57:46 +0200 | [diff] [blame] | 1400 | |
Alex Williamson | bbacc0c | 2012-12-10 10:33:09 -0700 | [diff] [blame] | 1401 | if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS || |
Xiao Guangrong | bf3e05b | 2011-11-24 17:40:57 +0800 | [diff] [blame] | 1402 | memslot->flags & KVM_MEMSLOT_INVALID) |
Yaowei Bai | 33e9415 | 2015-11-14 11:21:06 +0800 | [diff] [blame] | 1403 | return false; |
Izik Eidus | e0d62c7 | 2007-10-24 23:57:46 +0200 | [diff] [blame] | 1404 | |
Yaowei Bai | 33e9415 | 2015-11-14 11:21:06 +0800 | [diff] [blame] | 1405 | return true; |
Izik Eidus | e0d62c7 | 2007-10-24 23:57:46 +0200 | [diff] [blame] | 1406 | } |
| 1407 | EXPORT_SYMBOL_GPL(kvm_is_visible_gfn); |
| 1408 | |
Joerg Roedel | 8f0b1ab | 2010-01-28 12:37:56 +0100 | [diff] [blame] | 1409 | unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn) |
| 1410 | { |
| 1411 | struct vm_area_struct *vma; |
| 1412 | unsigned long addr, size; |
| 1413 | |
| 1414 | size = PAGE_SIZE; |
| 1415 | |
| 1416 | addr = gfn_to_hva(kvm, gfn); |
| 1417 | if (kvm_is_error_hva(addr)) |
| 1418 | return PAGE_SIZE; |
| 1419 | |
| 1420 | down_read(¤t->mm->mmap_sem); |
| 1421 | vma = find_vma(current->mm, addr); |
| 1422 | if (!vma) |
| 1423 | goto out; |
| 1424 | |
| 1425 | size = vma_kernel_pagesize(vma); |
| 1426 | |
| 1427 | out: |
| 1428 | up_read(¤t->mm->mmap_sem); |
| 1429 | |
| 1430 | return size; |
| 1431 | } |
| 1432 | |
Xiao Guangrong | 4d8b81a | 2012-08-21 11:02:51 +0800 | [diff] [blame] | 1433 | static bool memslot_is_readonly(struct kvm_memory_slot *slot) |
| 1434 | { |
| 1435 | return slot->flags & KVM_MEM_READONLY; |
| 1436 | } |
| 1437 | |
Xiao Guangrong | 4d8b81a | 2012-08-21 11:02:51 +0800 | [diff] [blame] | 1438 | static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn, |
| 1439 | gfn_t *nr_pages, bool write) |
Izik Eidus | 539cb66 | 2007-11-11 22:05:04 +0200 | [diff] [blame] | 1440 | { |
Marcelo Tosatti | bc6678a | 2009-12-23 14:35:21 -0200 | [diff] [blame] | 1441 | if (!slot || slot->flags & KVM_MEMSLOT_INVALID) |
Xiao Guangrong | ca3a490 | 2012-08-21 11:01:50 +0800 | [diff] [blame] | 1442 | return KVM_HVA_ERR_BAD; |
Xiao Guangrong | 4898778 | 2010-08-22 19:11:43 +0800 | [diff] [blame] | 1443 | |
Xiao Guangrong | 4d8b81a | 2012-08-21 11:02:51 +0800 | [diff] [blame] | 1444 | if (memslot_is_readonly(slot) && write) |
| 1445 | return KVM_HVA_ERR_RO_BAD; |
Xiao Guangrong | 4898778 | 2010-08-22 19:11:43 +0800 | [diff] [blame] | 1446 | |
| 1447 | if (nr_pages) |
| 1448 | *nr_pages = slot->npages - (gfn - slot->base_gfn); |
| 1449 | |
Xiao Guangrong | 4d8b81a | 2012-08-21 11:02:51 +0800 | [diff] [blame] | 1450 | return __gfn_to_hva_memslot(slot, gfn); |
Izik Eidus | 539cb66 | 2007-11-11 22:05:04 +0200 | [diff] [blame] | 1451 | } |
Xiao Guangrong | 4898778 | 2010-08-22 19:11:43 +0800 | [diff] [blame] | 1452 | |
Xiao Guangrong | 4d8b81a | 2012-08-21 11:02:51 +0800 | [diff] [blame] | 1453 | static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn, |
| 1454 | gfn_t *nr_pages) |
| 1455 | { |
| 1456 | return __gfn_to_hva_many(slot, gfn, nr_pages, true); |
| 1457 | } |
| 1458 | |
| 1459 | unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, |
Stephen Hemminger | 7940876 | 2013-12-29 12:12:29 -0800 | [diff] [blame] | 1460 | gfn_t gfn) |
Xiao Guangrong | 4d8b81a | 2012-08-21 11:02:51 +0800 | [diff] [blame] | 1461 | { |
| 1462 | return gfn_to_hva_many(slot, gfn, NULL); |
| 1463 | } |
| 1464 | EXPORT_SYMBOL_GPL(gfn_to_hva_memslot); |
| 1465 | |
Xiao Guangrong | 4898778 | 2010-08-22 19:11:43 +0800 | [diff] [blame] | 1466 | unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn) |
| 1467 | { |
Gleb Natapov | 49c7754 | 2010-10-18 15:22:23 +0200 | [diff] [blame] | 1468 | return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL); |
Xiao Guangrong | 4898778 | 2010-08-22 19:11:43 +0800 | [diff] [blame] | 1469 | } |
Sheng Yang | 0d15029 | 2008-04-25 21:44:50 +0800 | [diff] [blame] | 1470 | EXPORT_SYMBOL_GPL(gfn_to_hva); |
Izik Eidus | 539cb66 | 2007-11-11 22:05:04 +0200 | [diff] [blame] | 1471 | |
Paolo Bonzini | 8e73485 | 2015-05-17 13:58:53 +0200 | [diff] [blame] | 1472 | unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn) |
| 1473 | { |
| 1474 | return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL); |
| 1475 | } |
| 1476 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva); |
| 1477 | |
Xiao Guangrong | 86ab8cf | 2012-08-21 10:59:53 +0800 | [diff] [blame] | 1478 | /* |
Wei Yang | 970c0d4 | 2018-10-09 10:41:15 +0800 | [diff] [blame] | 1479 | * Return the hva of a @gfn and the R/W attribute if possible. |
| 1480 | * |
| 1481 | * @slot: the kvm_memory_slot which contains @gfn |
| 1482 | * @gfn: the gfn to be translated |
| 1483 | * @writable: used to return the read/write attribute of the @slot if the hva |
| 1484 | * is valid and @writable is not NULL |
Xiao Guangrong | 86ab8cf | 2012-08-21 10:59:53 +0800 | [diff] [blame] | 1485 | */ |
Christoffer Dall | 64d8312 | 2014-08-19 12:15:00 +0200 | [diff] [blame] | 1486 | unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot, |
| 1487 | gfn_t gfn, bool *writable) |
Gleb Natapov | 8030089 | 2010-10-19 18:13:41 +0200 | [diff] [blame] | 1488 | { |
Gleb Natapov | a2ac07f | 2013-10-01 19:58:36 +0300 | [diff] [blame] | 1489 | unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false); |
| 1490 | |
| 1491 | if (!kvm_is_error_hva(hva) && writable) |
Paolo Bonzini | ba6a354 | 2013-09-09 13:52:33 +0200 | [diff] [blame] | 1492 | *writable = !memslot_is_readonly(slot); |
| 1493 | |
Gleb Natapov | a2ac07f | 2013-10-01 19:58:36 +0300 | [diff] [blame] | 1494 | return hva; |
Xiao Guangrong | 86ab8cf | 2012-08-21 10:59:53 +0800 | [diff] [blame] | 1495 | } |
| 1496 | |
Christoffer Dall | 64d8312 | 2014-08-19 12:15:00 +0200 | [diff] [blame] | 1497 | unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable) |
| 1498 | { |
| 1499 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); |
| 1500 | |
| 1501 | return gfn_to_hva_memslot_prot(slot, gfn, writable); |
| 1502 | } |
| 1503 | |
Paolo Bonzini | 8e73485 | 2015-05-17 13:58:53 +0200 | [diff] [blame] | 1504 | unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable) |
| 1505 | { |
| 1506 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); |
| 1507 | |
| 1508 | return gfn_to_hva_memslot_prot(slot, gfn, writable); |
| 1509 | } |
| 1510 | |
Huang Ying | fafc3db | 2011-01-30 11:15:49 +0800 | [diff] [blame] | 1511 | static inline int check_user_page_hwpoison(unsigned long addr) |
| 1512 | { |
Lorenzo Stoakes | 0d73175 | 2016-10-24 10:57:25 +0100 | [diff] [blame] | 1513 | int rc, flags = FOLL_HWPOISON | FOLL_WRITE; |
Huang Ying | fafc3db | 2011-01-30 11:15:49 +0800 | [diff] [blame] | 1514 | |
Lorenzo Stoakes | 0d73175 | 2016-10-24 10:57:25 +0100 | [diff] [blame] | 1515 | rc = get_user_pages(addr, 1, flags, NULL, NULL); |
Huang Ying | fafc3db | 2011-01-30 11:15:49 +0800 | [diff] [blame] | 1516 | return rc == -EHWPOISON; |
| 1517 | } |
| 1518 | |
Xiao Guangrong | 2fc84311 | 2012-08-21 11:00:22 +0800 | [diff] [blame] | 1519 | /* |
Paolo Bonzini | b9b33da | 2018-07-27 17:44:41 +0200 | [diff] [blame] | 1520 | * The fast path to get the writable pfn which will be stored in @pfn, |
| 1521 | * true indicates success, otherwise false is returned. It's also the |
| 1522 | * only part that runs if we can are in atomic context. |
Xiao Guangrong | 2fc84311 | 2012-08-21 11:00:22 +0800 | [diff] [blame] | 1523 | */ |
Paolo Bonzini | b9b33da | 2018-07-27 17:44:41 +0200 | [diff] [blame] | 1524 | static bool hva_to_pfn_fast(unsigned long addr, bool write_fault, |
| 1525 | bool *writable, kvm_pfn_t *pfn) |
Xiao Guangrong | 2fc84311 | 2012-08-21 11:00:22 +0800 | [diff] [blame] | 1526 | { |
| 1527 | struct page *page[1]; |
| 1528 | int npages; |
| 1529 | |
Xiao Guangrong | 12ce13f | 2012-08-21 11:00:49 +0800 | [diff] [blame] | 1530 | /* |
| 1531 | * Fast pin a writable pfn only if it is a write fault request |
| 1532 | * or the caller allows to map a writable pfn for a read fault |
| 1533 | * request. |
| 1534 | */ |
| 1535 | if (!(write_fault || writable)) |
| 1536 | return false; |
| 1537 | |
Xiao Guangrong | 2fc84311 | 2012-08-21 11:00:22 +0800 | [diff] [blame] | 1538 | npages = __get_user_pages_fast(addr, 1, 1, page); |
| 1539 | if (npages == 1) { |
| 1540 | *pfn = page_to_pfn(page[0]); |
| 1541 | |
| 1542 | if (writable) |
| 1543 | *writable = true; |
| 1544 | return true; |
| 1545 | } |
| 1546 | |
| 1547 | return false; |
| 1548 | } |
| 1549 | |
| 1550 | /* |
| 1551 | * The slow path to get the pfn of the specified host virtual address, |
| 1552 | * 1 indicates success, -errno is returned if error is detected. |
| 1553 | */ |
| 1554 | static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault, |
Dan Williams | ba049e9 | 2016-01-15 16:56:11 -0800 | [diff] [blame] | 1555 | bool *writable, kvm_pfn_t *pfn) |
Avi Kivity | 954bbbc | 2007-03-30 14:02:32 +0300 | [diff] [blame] | 1556 | { |
Al Viro | ce53053 | 2017-11-19 17:47:33 -0500 | [diff] [blame] | 1557 | unsigned int flags = FOLL_HWPOISON; |
| 1558 | struct page *page; |
Gleb Natapov | af585b9 | 2010-10-14 11:22:46 +0200 | [diff] [blame] | 1559 | int npages = 0; |
Xiao Guangrong | 2fc84311 | 2012-08-21 11:00:22 +0800 | [diff] [blame] | 1560 | |
| 1561 | might_sleep(); |
| 1562 | |
| 1563 | if (writable) |
| 1564 | *writable = write_fault; |
| 1565 | |
Al Viro | ce53053 | 2017-11-19 17:47:33 -0500 | [diff] [blame] | 1566 | if (write_fault) |
| 1567 | flags |= FOLL_WRITE; |
| 1568 | if (async) |
| 1569 | flags |= FOLL_NOWAIT; |
Lorenzo Stoakes | d4944b0 | 2016-10-13 01:20:12 +0100 | [diff] [blame] | 1570 | |
Al Viro | ce53053 | 2017-11-19 17:47:33 -0500 | [diff] [blame] | 1571 | npages = get_user_pages_unlocked(addr, 1, &page, flags); |
Xiao Guangrong | 2fc84311 | 2012-08-21 11:00:22 +0800 | [diff] [blame] | 1572 | if (npages != 1) |
| 1573 | return npages; |
| 1574 | |
| 1575 | /* map read fault as writable if possible */ |
Xiao Guangrong | 12ce13f | 2012-08-21 11:00:49 +0800 | [diff] [blame] | 1576 | if (unlikely(!write_fault) && writable) { |
Al Viro | ce53053 | 2017-11-19 17:47:33 -0500 | [diff] [blame] | 1577 | struct page *wpage; |
Xiao Guangrong | 2fc84311 | 2012-08-21 11:00:22 +0800 | [diff] [blame] | 1578 | |
Al Viro | ce53053 | 2017-11-19 17:47:33 -0500 | [diff] [blame] | 1579 | if (__get_user_pages_fast(addr, 1, 1, &wpage) == 1) { |
Xiao Guangrong | 2fc84311 | 2012-08-21 11:00:22 +0800 | [diff] [blame] | 1580 | *writable = true; |
Al Viro | ce53053 | 2017-11-19 17:47:33 -0500 | [diff] [blame] | 1581 | put_page(page); |
| 1582 | page = wpage; |
Xiao Guangrong | 2fc84311 | 2012-08-21 11:00:22 +0800 | [diff] [blame] | 1583 | } |
Xiao Guangrong | 2fc84311 | 2012-08-21 11:00:22 +0800 | [diff] [blame] | 1584 | } |
Al Viro | ce53053 | 2017-11-19 17:47:33 -0500 | [diff] [blame] | 1585 | *pfn = page_to_pfn(page); |
Xiao Guangrong | 2fc84311 | 2012-08-21 11:00:22 +0800 | [diff] [blame] | 1586 | return npages; |
| 1587 | } |
| 1588 | |
Xiao Guangrong | 4d8b81a | 2012-08-21 11:02:51 +0800 | [diff] [blame] | 1589 | static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault) |
| 1590 | { |
| 1591 | if (unlikely(!(vma->vm_flags & VM_READ))) |
| 1592 | return false; |
| 1593 | |
| 1594 | if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE)))) |
| 1595 | return false; |
| 1596 | |
| 1597 | return true; |
| 1598 | } |
| 1599 | |
Paolo Bonzini | 92176a8 | 2016-06-07 16:22:47 +0200 | [diff] [blame] | 1600 | static int hva_to_pfn_remapped(struct vm_area_struct *vma, |
| 1601 | unsigned long addr, bool *async, |
KarimAllah Ahmed | a340b3e | 2018-01-17 19:18:56 +0100 | [diff] [blame] | 1602 | bool write_fault, bool *writable, |
| 1603 | kvm_pfn_t *p_pfn) |
Paolo Bonzini | 92176a8 | 2016-06-07 16:22:47 +0200 | [diff] [blame] | 1604 | { |
Paolo Bonzini | add6a0c | 2016-06-07 17:51:18 +0200 | [diff] [blame] | 1605 | unsigned long pfn; |
| 1606 | int r; |
| 1607 | |
| 1608 | r = follow_pfn(vma, addr, &pfn); |
| 1609 | if (r) { |
| 1610 | /* |
| 1611 | * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does |
| 1612 | * not call the fault handler, so do it here. |
| 1613 | */ |
| 1614 | bool unlocked = false; |
| 1615 | r = fixup_user_fault(current, current->mm, addr, |
| 1616 | (write_fault ? FAULT_FLAG_WRITE : 0), |
| 1617 | &unlocked); |
| 1618 | if (unlocked) |
| 1619 | return -EAGAIN; |
| 1620 | if (r) |
| 1621 | return r; |
| 1622 | |
| 1623 | r = follow_pfn(vma, addr, &pfn); |
| 1624 | if (r) |
| 1625 | return r; |
| 1626 | |
| 1627 | } |
| 1628 | |
KarimAllah Ahmed | a340b3e | 2018-01-17 19:18:56 +0100 | [diff] [blame] | 1629 | if (writable) |
| 1630 | *writable = true; |
Paolo Bonzini | add6a0c | 2016-06-07 17:51:18 +0200 | [diff] [blame] | 1631 | |
| 1632 | /* |
| 1633 | * Get a reference here because callers of *hva_to_pfn* and |
| 1634 | * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the |
| 1635 | * returned pfn. This is only needed if the VMA has VM_MIXEDMAP |
| 1636 | * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will |
| 1637 | * simply do nothing for reserved pfns. |
| 1638 | * |
| 1639 | * Whoever called remap_pfn_range is also going to call e.g. |
| 1640 | * unmap_mapping_range before the underlying pages are freed, |
| 1641 | * causing a call to our MMU notifier. |
| 1642 | */ |
| 1643 | kvm_get_pfn(pfn); |
| 1644 | |
| 1645 | *p_pfn = pfn; |
Paolo Bonzini | 92176a8 | 2016-06-07 16:22:47 +0200 | [diff] [blame] | 1646 | return 0; |
| 1647 | } |
| 1648 | |
Xiao Guangrong | 12ce13f | 2012-08-21 11:00:49 +0800 | [diff] [blame] | 1649 | /* |
| 1650 | * Pin guest page in memory and return its pfn. |
| 1651 | * @addr: host virtual address which maps memory to the guest |
| 1652 | * @atomic: whether this function can sleep |
| 1653 | * @async: whether this function need to wait IO complete if the |
| 1654 | * host page is not in the memory |
| 1655 | * @write_fault: whether we should get a writable host page |
| 1656 | * @writable: whether it allows to map a writable host page for !@write_fault |
| 1657 | * |
| 1658 | * The function will map a writable host page for these two cases: |
| 1659 | * 1): @write_fault = true |
| 1660 | * 2): @write_fault = false && @writable, @writable will tell the caller |
| 1661 | * whether the mapping is writable. |
| 1662 | */ |
Dan Williams | ba049e9 | 2016-01-15 16:56:11 -0800 | [diff] [blame] | 1663 | static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async, |
Xiao Guangrong | 2fc84311 | 2012-08-21 11:00:22 +0800 | [diff] [blame] | 1664 | bool write_fault, bool *writable) |
| 1665 | { |
| 1666 | struct vm_area_struct *vma; |
Dan Williams | ba049e9 | 2016-01-15 16:56:11 -0800 | [diff] [blame] | 1667 | kvm_pfn_t pfn = 0; |
Paolo Bonzini | 92176a8 | 2016-06-07 16:22:47 +0200 | [diff] [blame] | 1668 | int npages, r; |
Avi Kivity | 954bbbc | 2007-03-30 14:02:32 +0300 | [diff] [blame] | 1669 | |
Gleb Natapov | af585b9 | 2010-10-14 11:22:46 +0200 | [diff] [blame] | 1670 | /* we can do it either atomically or asynchronously, not both */ |
| 1671 | BUG_ON(atomic && async); |
| 1672 | |
Paolo Bonzini | b9b33da | 2018-07-27 17:44:41 +0200 | [diff] [blame] | 1673 | if (hva_to_pfn_fast(addr, write_fault, writable, &pfn)) |
Xiao Guangrong | 2fc84311 | 2012-08-21 11:00:22 +0800 | [diff] [blame] | 1674 | return pfn; |
Marcelo Tosatti | 612819c | 2010-10-22 14:18:18 -0200 | [diff] [blame] | 1675 | |
Xiao Guangrong | 2fc84311 | 2012-08-21 11:00:22 +0800 | [diff] [blame] | 1676 | if (atomic) |
| 1677 | return KVM_PFN_ERR_FAULT; |
Marcelo Tosatti | 612819c | 2010-10-22 14:18:18 -0200 | [diff] [blame] | 1678 | |
Xiao Guangrong | 2fc84311 | 2012-08-21 11:00:22 +0800 | [diff] [blame] | 1679 | npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn); |
| 1680 | if (npages == 1) |
| 1681 | return pfn; |
Gleb Natapov | af585b9 | 2010-10-14 11:22:46 +0200 | [diff] [blame] | 1682 | |
Xiao Guangrong | 2fc84311 | 2012-08-21 11:00:22 +0800 | [diff] [blame] | 1683 | down_read(¤t->mm->mmap_sem); |
| 1684 | if (npages == -EHWPOISON || |
| 1685 | (!async && check_user_page_hwpoison(addr))) { |
| 1686 | pfn = KVM_PFN_ERR_HWPOISON; |
| 1687 | goto exit; |
Xiao Guangrong | 887c08a | 2010-08-22 19:10:28 +0800 | [diff] [blame] | 1688 | } |
Izik Eidus | 539cb66 | 2007-11-11 22:05:04 +0200 | [diff] [blame] | 1689 | |
Paolo Bonzini | add6a0c | 2016-06-07 17:51:18 +0200 | [diff] [blame] | 1690 | retry: |
Xiao Guangrong | 2fc84311 | 2012-08-21 11:00:22 +0800 | [diff] [blame] | 1691 | vma = find_vma_intersection(current->mm, addr, addr + 1); |
Anthony Liguori | 8d4e128 | 2007-10-18 09:59:34 -0500 | [diff] [blame] | 1692 | |
Xiao Guangrong | 2fc84311 | 2012-08-21 11:00:22 +0800 | [diff] [blame] | 1693 | if (vma == NULL) |
| 1694 | pfn = KVM_PFN_ERR_FAULT; |
Paolo Bonzini | 92176a8 | 2016-06-07 16:22:47 +0200 | [diff] [blame] | 1695 | else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) { |
KarimAllah Ahmed | a340b3e | 2018-01-17 19:18:56 +0100 | [diff] [blame] | 1696 | r = hva_to_pfn_remapped(vma, addr, async, write_fault, writable, &pfn); |
Paolo Bonzini | add6a0c | 2016-06-07 17:51:18 +0200 | [diff] [blame] | 1697 | if (r == -EAGAIN) |
| 1698 | goto retry; |
Paolo Bonzini | 92176a8 | 2016-06-07 16:22:47 +0200 | [diff] [blame] | 1699 | if (r < 0) |
| 1700 | pfn = KVM_PFN_ERR_FAULT; |
Xiao Guangrong | 2fc84311 | 2012-08-21 11:00:22 +0800 | [diff] [blame] | 1701 | } else { |
Xiao Guangrong | 4d8b81a | 2012-08-21 11:02:51 +0800 | [diff] [blame] | 1702 | if (async && vma_is_valid(vma, write_fault)) |
Xiao Guangrong | 2fc84311 | 2012-08-21 11:00:22 +0800 | [diff] [blame] | 1703 | *async = true; |
| 1704 | pfn = KVM_PFN_ERR_FAULT; |
| 1705 | } |
| 1706 | exit: |
| 1707 | up_read(¤t->mm->mmap_sem); |
Anthony Liguori | 2e2e373 | 2008-04-30 15:37:07 -0500 | [diff] [blame] | 1708 | return pfn; |
Anthony Liguori | 35149e2 | 2008-04-02 14:46:56 -0500 | [diff] [blame] | 1709 | } |
| 1710 | |
Dan Williams | ba049e9 | 2016-01-15 16:56:11 -0800 | [diff] [blame] | 1711 | kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, |
| 1712 | bool atomic, bool *async, bool write_fault, |
| 1713 | bool *writable) |
Xiao Guangrong | 887c08a | 2010-08-22 19:10:28 +0800 | [diff] [blame] | 1714 | { |
Xiao Guangrong | 4d8b81a | 2012-08-21 11:02:51 +0800 | [diff] [blame] | 1715 | unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault); |
| 1716 | |
Paolo Bonzini | b2740d3 | 2016-02-23 15:36:01 +0100 | [diff] [blame] | 1717 | if (addr == KVM_HVA_ERR_RO_BAD) { |
| 1718 | if (writable) |
| 1719 | *writable = false; |
Xiao Guangrong | 4d8b81a | 2012-08-21 11:02:51 +0800 | [diff] [blame] | 1720 | return KVM_PFN_ERR_RO_FAULT; |
Paolo Bonzini | b2740d3 | 2016-02-23 15:36:01 +0100 | [diff] [blame] | 1721 | } |
Xiao Guangrong | 4d8b81a | 2012-08-21 11:02:51 +0800 | [diff] [blame] | 1722 | |
Paolo Bonzini | b2740d3 | 2016-02-23 15:36:01 +0100 | [diff] [blame] | 1723 | if (kvm_is_error_hva(addr)) { |
| 1724 | if (writable) |
| 1725 | *writable = false; |
Xiao Guangrong | 81c52c5 | 2012-10-16 20:10:59 +0800 | [diff] [blame] | 1726 | return KVM_PFN_NOSLOT; |
Paolo Bonzini | b2740d3 | 2016-02-23 15:36:01 +0100 | [diff] [blame] | 1727 | } |
Xiao Guangrong | 4d8b81a | 2012-08-21 11:02:51 +0800 | [diff] [blame] | 1728 | |
| 1729 | /* Do not map writable pfn in the readonly memslot. */ |
| 1730 | if (writable && memslot_is_readonly(slot)) { |
| 1731 | *writable = false; |
| 1732 | writable = NULL; |
| 1733 | } |
| 1734 | |
| 1735 | return hva_to_pfn(addr, atomic, async, write_fault, |
| 1736 | writable); |
Xiao Guangrong | 887c08a | 2010-08-22 19:10:28 +0800 | [diff] [blame] | 1737 | } |
Paolo Bonzini | 3520469 | 2015-04-02 11:20:48 +0200 | [diff] [blame] | 1738 | EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot); |
Xiao Guangrong | 887c08a | 2010-08-22 19:10:28 +0800 | [diff] [blame] | 1739 | |
Dan Williams | ba049e9 | 2016-01-15 16:56:11 -0800 | [diff] [blame] | 1740 | kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault, |
Marcelo Tosatti | 612819c | 2010-10-22 14:18:18 -0200 | [diff] [blame] | 1741 | bool *writable) |
| 1742 | { |
Paolo Bonzini | e37afc6 | 2015-05-19 16:09:04 +0200 | [diff] [blame] | 1743 | return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL, |
| 1744 | write_fault, writable); |
Marcelo Tosatti | 612819c | 2010-10-22 14:18:18 -0200 | [diff] [blame] | 1745 | } |
| 1746 | EXPORT_SYMBOL_GPL(gfn_to_pfn_prot); |
| 1747 | |
Dan Williams | ba049e9 | 2016-01-15 16:56:11 -0800 | [diff] [blame] | 1748 | kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn) |
Marcelo Tosatti | 506f0d6 | 2009-12-23 14:35:19 -0200 | [diff] [blame] | 1749 | { |
Xiao Guangrong | 4d8b81a | 2012-08-21 11:02:51 +0800 | [diff] [blame] | 1750 | return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL); |
Marcelo Tosatti | 506f0d6 | 2009-12-23 14:35:19 -0200 | [diff] [blame] | 1751 | } |
Paolo Bonzini | e37afc6 | 2015-05-19 16:09:04 +0200 | [diff] [blame] | 1752 | EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot); |
Marcelo Tosatti | 506f0d6 | 2009-12-23 14:35:19 -0200 | [diff] [blame] | 1753 | |
Dan Williams | ba049e9 | 2016-01-15 16:56:11 -0800 | [diff] [blame] | 1754 | kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn) |
Xiao Guangrong | 037d92d | 2012-08-21 10:59:12 +0800 | [diff] [blame] | 1755 | { |
Xiao Guangrong | 4d8b81a | 2012-08-21 11:02:51 +0800 | [diff] [blame] | 1756 | return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL); |
Xiao Guangrong | 037d92d | 2012-08-21 10:59:12 +0800 | [diff] [blame] | 1757 | } |
| 1758 | EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic); |
| 1759 | |
Dan Williams | ba049e9 | 2016-01-15 16:56:11 -0800 | [diff] [blame] | 1760 | kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn) |
Paolo Bonzini | e37afc6 | 2015-05-19 16:09:04 +0200 | [diff] [blame] | 1761 | { |
| 1762 | return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn); |
| 1763 | } |
| 1764 | EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic); |
| 1765 | |
Dan Williams | ba049e9 | 2016-01-15 16:56:11 -0800 | [diff] [blame] | 1766 | kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn) |
Paolo Bonzini | 8e73485 | 2015-05-17 13:58:53 +0200 | [diff] [blame] | 1767 | { |
| 1768 | return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn); |
| 1769 | } |
| 1770 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic); |
| 1771 | |
Dan Williams | ba049e9 | 2016-01-15 16:56:11 -0800 | [diff] [blame] | 1772 | kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn) |
Paolo Bonzini | e37afc6 | 2015-05-19 16:09:04 +0200 | [diff] [blame] | 1773 | { |
| 1774 | return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn); |
| 1775 | } |
| 1776 | EXPORT_SYMBOL_GPL(gfn_to_pfn); |
| 1777 | |
Dan Williams | ba049e9 | 2016-01-15 16:56:11 -0800 | [diff] [blame] | 1778 | kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn) |
Paolo Bonzini | 8e73485 | 2015-05-17 13:58:53 +0200 | [diff] [blame] | 1779 | { |
| 1780 | return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn); |
| 1781 | } |
| 1782 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn); |
| 1783 | |
Paolo Bonzini | d9ef13c | 2015-05-19 16:01:50 +0200 | [diff] [blame] | 1784 | int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn, |
| 1785 | struct page **pages, int nr_pages) |
Xiao Guangrong | 4898778 | 2010-08-22 19:11:43 +0800 | [diff] [blame] | 1786 | { |
| 1787 | unsigned long addr; |
Arnd Bergmann | 076b925 | 2017-08-10 14:14:39 +0200 | [diff] [blame] | 1788 | gfn_t entry = 0; |
Xiao Guangrong | 4898778 | 2010-08-22 19:11:43 +0800 | [diff] [blame] | 1789 | |
Paolo Bonzini | d9ef13c | 2015-05-19 16:01:50 +0200 | [diff] [blame] | 1790 | addr = gfn_to_hva_many(slot, gfn, &entry); |
Xiao Guangrong | 4898778 | 2010-08-22 19:11:43 +0800 | [diff] [blame] | 1791 | if (kvm_is_error_hva(addr)) |
| 1792 | return -1; |
| 1793 | |
| 1794 | if (entry < nr_pages) |
| 1795 | return 0; |
| 1796 | |
| 1797 | return __get_user_pages_fast(addr, nr_pages, 1, pages); |
| 1798 | } |
| 1799 | EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic); |
| 1800 | |
Dan Williams | ba049e9 | 2016-01-15 16:56:11 -0800 | [diff] [blame] | 1801 | static struct page *kvm_pfn_to_page(kvm_pfn_t pfn) |
Xiao Guangrong | a276632 | 2012-07-26 11:58:59 +0800 | [diff] [blame] | 1802 | { |
Xiao Guangrong | 81c52c5 | 2012-10-16 20:10:59 +0800 | [diff] [blame] | 1803 | if (is_error_noslot_pfn(pfn)) |
Xiao Guangrong | 6cede2e | 2012-08-03 15:41:22 +0800 | [diff] [blame] | 1804 | return KVM_ERR_PTR_BAD_PAGE; |
Xiao Guangrong | a276632 | 2012-07-26 11:58:59 +0800 | [diff] [blame] | 1805 | |
Ard Biesheuvel | bf4bea8 | 2014-11-10 08:33:56 +0000 | [diff] [blame] | 1806 | if (kvm_is_reserved_pfn(pfn)) { |
Xiao Guangrong | cb9aaa3 | 2012-08-03 15:42:10 +0800 | [diff] [blame] | 1807 | WARN_ON(1); |
| 1808 | return KVM_ERR_PTR_BAD_PAGE; |
| 1809 | } |
| 1810 | |
Xiao Guangrong | a276632 | 2012-07-26 11:58:59 +0800 | [diff] [blame] | 1811 | return pfn_to_page(pfn); |
| 1812 | } |
| 1813 | |
Anthony Liguori | 35149e2 | 2008-04-02 14:46:56 -0500 | [diff] [blame] | 1814 | struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn) |
| 1815 | { |
Dan Williams | ba049e9 | 2016-01-15 16:56:11 -0800 | [diff] [blame] | 1816 | kvm_pfn_t pfn; |
Anthony Liguori | 2e2e373 | 2008-04-30 15:37:07 -0500 | [diff] [blame] | 1817 | |
| 1818 | pfn = gfn_to_pfn(kvm, gfn); |
Anthony Liguori | 2e2e373 | 2008-04-30 15:37:07 -0500 | [diff] [blame] | 1819 | |
Xiao Guangrong | a276632 | 2012-07-26 11:58:59 +0800 | [diff] [blame] | 1820 | return kvm_pfn_to_page(pfn); |
Avi Kivity | 954bbbc | 2007-03-30 14:02:32 +0300 | [diff] [blame] | 1821 | } |
| 1822 | EXPORT_SYMBOL_GPL(gfn_to_page); |
| 1823 | |
KarimAllah Ahmed | e45adf6 | 2019-01-31 21:24:34 +0100 | [diff] [blame] | 1824 | static int __kvm_map_gfn(struct kvm_memory_slot *slot, gfn_t gfn, |
| 1825 | struct kvm_host_map *map) |
| 1826 | { |
| 1827 | kvm_pfn_t pfn; |
| 1828 | void *hva = NULL; |
| 1829 | struct page *page = KVM_UNMAPPED_PAGE; |
| 1830 | |
| 1831 | if (!map) |
| 1832 | return -EINVAL; |
| 1833 | |
| 1834 | pfn = gfn_to_pfn_memslot(slot, gfn); |
| 1835 | if (is_error_noslot_pfn(pfn)) |
| 1836 | return -EINVAL; |
| 1837 | |
| 1838 | if (pfn_valid(pfn)) { |
| 1839 | page = pfn_to_page(pfn); |
| 1840 | hva = kmap(page); |
Paolo Bonzini | d30b214 | 2019-05-20 12:06:36 +0200 | [diff] [blame] | 1841 | #ifdef CONFIG_HAS_IOMEM |
KarimAllah Ahmed | e45adf6 | 2019-01-31 21:24:34 +0100 | [diff] [blame] | 1842 | } else { |
| 1843 | hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB); |
Paolo Bonzini | d30b214 | 2019-05-20 12:06:36 +0200 | [diff] [blame] | 1844 | #endif |
KarimAllah Ahmed | e45adf6 | 2019-01-31 21:24:34 +0100 | [diff] [blame] | 1845 | } |
| 1846 | |
| 1847 | if (!hva) |
| 1848 | return -EFAULT; |
| 1849 | |
| 1850 | map->page = page; |
| 1851 | map->hva = hva; |
| 1852 | map->pfn = pfn; |
| 1853 | map->gfn = gfn; |
| 1854 | |
| 1855 | return 0; |
| 1856 | } |
| 1857 | |
| 1858 | int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map) |
| 1859 | { |
| 1860 | return __kvm_map_gfn(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, map); |
| 1861 | } |
| 1862 | EXPORT_SYMBOL_GPL(kvm_vcpu_map); |
| 1863 | |
| 1864 | void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, |
| 1865 | bool dirty) |
| 1866 | { |
| 1867 | if (!map) |
| 1868 | return; |
| 1869 | |
| 1870 | if (!map->hva) |
| 1871 | return; |
| 1872 | |
KarimAllah Ahmed | b614c602 | 2019-07-10 11:13:13 +0200 | [diff] [blame] | 1873 | if (map->page != KVM_UNMAPPED_PAGE) |
KarimAllah Ahmed | e45adf6 | 2019-01-31 21:24:34 +0100 | [diff] [blame] | 1874 | kunmap(map->page); |
Christian Borntraeger | eb1f2f3 | 2019-05-27 10:28:25 +0200 | [diff] [blame] | 1875 | #ifdef CONFIG_HAS_IOMEM |
KarimAllah Ahmed | e45adf6 | 2019-01-31 21:24:34 +0100 | [diff] [blame] | 1876 | else |
| 1877 | memunmap(map->hva); |
Christian Borntraeger | eb1f2f3 | 2019-05-27 10:28:25 +0200 | [diff] [blame] | 1878 | #endif |
KarimAllah Ahmed | e45adf6 | 2019-01-31 21:24:34 +0100 | [diff] [blame] | 1879 | |
| 1880 | if (dirty) { |
| 1881 | kvm_vcpu_mark_page_dirty(vcpu, map->gfn); |
| 1882 | kvm_release_pfn_dirty(map->pfn); |
| 1883 | } else { |
| 1884 | kvm_release_pfn_clean(map->pfn); |
| 1885 | } |
| 1886 | |
| 1887 | map->hva = NULL; |
| 1888 | map->page = NULL; |
| 1889 | } |
| 1890 | EXPORT_SYMBOL_GPL(kvm_vcpu_unmap); |
| 1891 | |
Paolo Bonzini | 8e73485 | 2015-05-17 13:58:53 +0200 | [diff] [blame] | 1892 | struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn) |
| 1893 | { |
Dan Williams | ba049e9 | 2016-01-15 16:56:11 -0800 | [diff] [blame] | 1894 | kvm_pfn_t pfn; |
Paolo Bonzini | 8e73485 | 2015-05-17 13:58:53 +0200 | [diff] [blame] | 1895 | |
| 1896 | pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn); |
| 1897 | |
| 1898 | return kvm_pfn_to_page(pfn); |
| 1899 | } |
| 1900 | EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page); |
| 1901 | |
Izik Eidus | b4231d61 | 2007-11-20 11:49:33 +0200 | [diff] [blame] | 1902 | void kvm_release_page_clean(struct page *page) |
| 1903 | { |
Xiao Guangrong | 32cad84 | 2012-08-03 15:42:52 +0800 | [diff] [blame] | 1904 | WARN_ON(is_error_page(page)); |
| 1905 | |
Anthony Liguori | 35149e2 | 2008-04-02 14:46:56 -0500 | [diff] [blame] | 1906 | kvm_release_pfn_clean(page_to_pfn(page)); |
Izik Eidus | b4231d61 | 2007-11-20 11:49:33 +0200 | [diff] [blame] | 1907 | } |
| 1908 | EXPORT_SYMBOL_GPL(kvm_release_page_clean); |
| 1909 | |
Dan Williams | ba049e9 | 2016-01-15 16:56:11 -0800 | [diff] [blame] | 1910 | void kvm_release_pfn_clean(kvm_pfn_t pfn) |
Anthony Liguori | 35149e2 | 2008-04-02 14:46:56 -0500 | [diff] [blame] | 1911 | { |
Ard Biesheuvel | bf4bea8 | 2014-11-10 08:33:56 +0000 | [diff] [blame] | 1912 | if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn)) |
Anthony Liguori | 2e2e373 | 2008-04-30 15:37:07 -0500 | [diff] [blame] | 1913 | put_page(pfn_to_page(pfn)); |
Anthony Liguori | 35149e2 | 2008-04-02 14:46:56 -0500 | [diff] [blame] | 1914 | } |
| 1915 | EXPORT_SYMBOL_GPL(kvm_release_pfn_clean); |
| 1916 | |
Izik Eidus | b4231d61 | 2007-11-20 11:49:33 +0200 | [diff] [blame] | 1917 | void kvm_release_page_dirty(struct page *page) |
Izik Eidus | 8a7ae05 | 2007-10-18 11:09:33 +0200 | [diff] [blame] | 1918 | { |
Xiao Guangrong | a276632 | 2012-07-26 11:58:59 +0800 | [diff] [blame] | 1919 | WARN_ON(is_error_page(page)); |
| 1920 | |
Anthony Liguori | 35149e2 | 2008-04-02 14:46:56 -0500 | [diff] [blame] | 1921 | kvm_release_pfn_dirty(page_to_pfn(page)); |
Izik Eidus | 8a7ae05 | 2007-10-18 11:09:33 +0200 | [diff] [blame] | 1922 | } |
Izik Eidus | b4231d61 | 2007-11-20 11:49:33 +0200 | [diff] [blame] | 1923 | EXPORT_SYMBOL_GPL(kvm_release_page_dirty); |
Izik Eidus | 8a7ae05 | 2007-10-18 11:09:33 +0200 | [diff] [blame] | 1924 | |
David Hildenbrand | f7a6509 | 2017-09-01 17:11:43 +0200 | [diff] [blame] | 1925 | void kvm_release_pfn_dirty(kvm_pfn_t pfn) |
Anthony Liguori | 35149e2 | 2008-04-02 14:46:56 -0500 | [diff] [blame] | 1926 | { |
| 1927 | kvm_set_pfn_dirty(pfn); |
| 1928 | kvm_release_pfn_clean(pfn); |
| 1929 | } |
David Hildenbrand | f7a6509 | 2017-09-01 17:11:43 +0200 | [diff] [blame] | 1930 | EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty); |
Anthony Liguori | 35149e2 | 2008-04-02 14:46:56 -0500 | [diff] [blame] | 1931 | |
Dan Williams | ba049e9 | 2016-01-15 16:56:11 -0800 | [diff] [blame] | 1932 | void kvm_set_pfn_dirty(kvm_pfn_t pfn) |
Anthony Liguori | 35149e2 | 2008-04-02 14:46:56 -0500 | [diff] [blame] | 1933 | { |
Sean Christopherson | a78986a | 2019-11-11 14:12:27 -0800 | [diff] [blame] | 1934 | if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn)) { |
Anthony Liguori | 2e2e373 | 2008-04-30 15:37:07 -0500 | [diff] [blame] | 1935 | struct page *page = pfn_to_page(pfn); |
Xiubo Li | f95ef0cd | 2015-02-26 14:58:23 +0800 | [diff] [blame] | 1936 | |
Paolo Bonzini | 8f946da | 2019-08-05 11:11:08 +0200 | [diff] [blame] | 1937 | SetPageDirty(page); |
Anthony Liguori | 2e2e373 | 2008-04-30 15:37:07 -0500 | [diff] [blame] | 1938 | } |
Anthony Liguori | 35149e2 | 2008-04-02 14:46:56 -0500 | [diff] [blame] | 1939 | } |
| 1940 | EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty); |
| 1941 | |
Dan Williams | ba049e9 | 2016-01-15 16:56:11 -0800 | [diff] [blame] | 1942 | void kvm_set_pfn_accessed(kvm_pfn_t pfn) |
Anthony Liguori | 35149e2 | 2008-04-02 14:46:56 -0500 | [diff] [blame] | 1943 | { |
Sean Christopherson | a78986a | 2019-11-11 14:12:27 -0800 | [diff] [blame] | 1944 | if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn)) |
Anthony Liguori | 2e2e373 | 2008-04-30 15:37:07 -0500 | [diff] [blame] | 1945 | mark_page_accessed(pfn_to_page(pfn)); |
Anthony Liguori | 35149e2 | 2008-04-02 14:46:56 -0500 | [diff] [blame] | 1946 | } |
| 1947 | EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed); |
| 1948 | |
Dan Williams | ba049e9 | 2016-01-15 16:56:11 -0800 | [diff] [blame] | 1949 | void kvm_get_pfn(kvm_pfn_t pfn) |
Anthony Liguori | 35149e2 | 2008-04-02 14:46:56 -0500 | [diff] [blame] | 1950 | { |
Ard Biesheuvel | bf4bea8 | 2014-11-10 08:33:56 +0000 | [diff] [blame] | 1951 | if (!kvm_is_reserved_pfn(pfn)) |
Anthony Liguori | 2e2e373 | 2008-04-30 15:37:07 -0500 | [diff] [blame] | 1952 | get_page(pfn_to_page(pfn)); |
Anthony Liguori | 35149e2 | 2008-04-02 14:46:56 -0500 | [diff] [blame] | 1953 | } |
| 1954 | EXPORT_SYMBOL_GPL(kvm_get_pfn); |
| 1955 | |
Izik Eidus | 195aefd | 2007-10-01 22:14:18 +0200 | [diff] [blame] | 1956 | static int next_segment(unsigned long len, int offset) |
| 1957 | { |
| 1958 | if (len > PAGE_SIZE - offset) |
| 1959 | return PAGE_SIZE - offset; |
| 1960 | else |
| 1961 | return len; |
| 1962 | } |
| 1963 | |
Paolo Bonzini | 8e73485 | 2015-05-17 13:58:53 +0200 | [diff] [blame] | 1964 | static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn, |
| 1965 | void *data, int offset, int len) |
Izik Eidus | 195aefd | 2007-10-01 22:14:18 +0200 | [diff] [blame] | 1966 | { |
Izik Eidus | e0506bc | 2007-11-11 22:10:22 +0200 | [diff] [blame] | 1967 | int r; |
| 1968 | unsigned long addr; |
Izik Eidus | 195aefd | 2007-10-01 22:14:18 +0200 | [diff] [blame] | 1969 | |
Paolo Bonzini | 8e73485 | 2015-05-17 13:58:53 +0200 | [diff] [blame] | 1970 | addr = gfn_to_hva_memslot_prot(slot, gfn, NULL); |
Izik Eidus | e0506bc | 2007-11-11 22:10:22 +0200 | [diff] [blame] | 1971 | if (kvm_is_error_hva(addr)) |
Izik Eidus | 195aefd | 2007-10-01 22:14:18 +0200 | [diff] [blame] | 1972 | return -EFAULT; |
Paolo Bonzini | 3180a7f | 2015-04-02 14:08:20 +0200 | [diff] [blame] | 1973 | r = __copy_from_user(data, (void __user *)addr + offset, len); |
Izik Eidus | e0506bc | 2007-11-11 22:10:22 +0200 | [diff] [blame] | 1974 | if (r) |
| 1975 | return -EFAULT; |
Izik Eidus | 195aefd | 2007-10-01 22:14:18 +0200 | [diff] [blame] | 1976 | return 0; |
| 1977 | } |
Paolo Bonzini | 8e73485 | 2015-05-17 13:58:53 +0200 | [diff] [blame] | 1978 | |
| 1979 | int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset, |
| 1980 | int len) |
| 1981 | { |
| 1982 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); |
| 1983 | |
| 1984 | return __kvm_read_guest_page(slot, gfn, data, offset, len); |
| 1985 | } |
Izik Eidus | 195aefd | 2007-10-01 22:14:18 +0200 | [diff] [blame] | 1986 | EXPORT_SYMBOL_GPL(kvm_read_guest_page); |
| 1987 | |
Paolo Bonzini | 8e73485 | 2015-05-17 13:58:53 +0200 | [diff] [blame] | 1988 | int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data, |
| 1989 | int offset, int len) |
| 1990 | { |
| 1991 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); |
| 1992 | |
| 1993 | return __kvm_read_guest_page(slot, gfn, data, offset, len); |
| 1994 | } |
| 1995 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page); |
| 1996 | |
Izik Eidus | 195aefd | 2007-10-01 22:14:18 +0200 | [diff] [blame] | 1997 | int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len) |
| 1998 | { |
| 1999 | gfn_t gfn = gpa >> PAGE_SHIFT; |
| 2000 | int seg; |
| 2001 | int offset = offset_in_page(gpa); |
| 2002 | int ret; |
| 2003 | |
| 2004 | while ((seg = next_segment(len, offset)) != 0) { |
| 2005 | ret = kvm_read_guest_page(kvm, gfn, data, offset, seg); |
| 2006 | if (ret < 0) |
| 2007 | return ret; |
| 2008 | offset = 0; |
| 2009 | len -= seg; |
| 2010 | data += seg; |
| 2011 | ++gfn; |
| 2012 | } |
| 2013 | return 0; |
| 2014 | } |
| 2015 | EXPORT_SYMBOL_GPL(kvm_read_guest); |
| 2016 | |
Paolo Bonzini | 8e73485 | 2015-05-17 13:58:53 +0200 | [diff] [blame] | 2017 | int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len) |
| 2018 | { |
| 2019 | gfn_t gfn = gpa >> PAGE_SHIFT; |
| 2020 | int seg; |
| 2021 | int offset = offset_in_page(gpa); |
| 2022 | int ret; |
| 2023 | |
| 2024 | while ((seg = next_segment(len, offset)) != 0) { |
| 2025 | ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg); |
| 2026 | if (ret < 0) |
| 2027 | return ret; |
| 2028 | offset = 0; |
| 2029 | len -= seg; |
| 2030 | data += seg; |
| 2031 | ++gfn; |
| 2032 | } |
| 2033 | return 0; |
| 2034 | } |
| 2035 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest); |
| 2036 | |
| 2037 | static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn, |
| 2038 | void *data, int offset, unsigned long len) |
Marcelo Tosatti | 7ec5458 | 2007-12-20 19:18:23 -0500 | [diff] [blame] | 2039 | { |
| 2040 | int r; |
| 2041 | unsigned long addr; |
Marcelo Tosatti | 7ec5458 | 2007-12-20 19:18:23 -0500 | [diff] [blame] | 2042 | |
Paolo Bonzini | 8e73485 | 2015-05-17 13:58:53 +0200 | [diff] [blame] | 2043 | addr = gfn_to_hva_memslot_prot(slot, gfn, NULL); |
Marcelo Tosatti | 7ec5458 | 2007-12-20 19:18:23 -0500 | [diff] [blame] | 2044 | if (kvm_is_error_hva(addr)) |
| 2045 | return -EFAULT; |
Andrea Arcangeli | 0aac03f | 2008-01-30 19:57:35 +0100 | [diff] [blame] | 2046 | pagefault_disable(); |
Paolo Bonzini | 3180a7f | 2015-04-02 14:08:20 +0200 | [diff] [blame] | 2047 | r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len); |
Andrea Arcangeli | 0aac03f | 2008-01-30 19:57:35 +0100 | [diff] [blame] | 2048 | pagefault_enable(); |
Marcelo Tosatti | 7ec5458 | 2007-12-20 19:18:23 -0500 | [diff] [blame] | 2049 | if (r) |
| 2050 | return -EFAULT; |
| 2051 | return 0; |
| 2052 | } |
Marcelo Tosatti | 7ec5458 | 2007-12-20 19:18:23 -0500 | [diff] [blame] | 2053 | |
Paolo Bonzini | 8e73485 | 2015-05-17 13:58:53 +0200 | [diff] [blame] | 2054 | int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data, |
| 2055 | unsigned long len) |
| 2056 | { |
| 2057 | gfn_t gfn = gpa >> PAGE_SHIFT; |
| 2058 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); |
| 2059 | int offset = offset_in_page(gpa); |
| 2060 | |
| 2061 | return __kvm_read_guest_atomic(slot, gfn, data, offset, len); |
| 2062 | } |
| 2063 | EXPORT_SYMBOL_GPL(kvm_read_guest_atomic); |
| 2064 | |
| 2065 | int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa, |
| 2066 | void *data, unsigned long len) |
| 2067 | { |
| 2068 | gfn_t gfn = gpa >> PAGE_SHIFT; |
| 2069 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); |
| 2070 | int offset = offset_in_page(gpa); |
| 2071 | |
| 2072 | return __kvm_read_guest_atomic(slot, gfn, data, offset, len); |
| 2073 | } |
| 2074 | EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic); |
| 2075 | |
| 2076 | static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn, |
| 2077 | const void *data, int offset, int len) |
Izik Eidus | 195aefd | 2007-10-01 22:14:18 +0200 | [diff] [blame] | 2078 | { |
Izik Eidus | e0506bc | 2007-11-11 22:10:22 +0200 | [diff] [blame] | 2079 | int r; |
| 2080 | unsigned long addr; |
Izik Eidus | 195aefd | 2007-10-01 22:14:18 +0200 | [diff] [blame] | 2081 | |
Radim Krčmář | 251eb84 | 2015-04-10 21:47:27 +0200 | [diff] [blame] | 2082 | addr = gfn_to_hva_memslot(memslot, gfn); |
Izik Eidus | e0506bc | 2007-11-11 22:10:22 +0200 | [diff] [blame] | 2083 | if (kvm_is_error_hva(addr)) |
Izik Eidus | 195aefd | 2007-10-01 22:14:18 +0200 | [diff] [blame] | 2084 | return -EFAULT; |
Xiao Guangrong | 8b0cedf | 2011-05-15 23:22:04 +0800 | [diff] [blame] | 2085 | r = __copy_to_user((void __user *)addr + offset, data, len); |
Izik Eidus | e0506bc | 2007-11-11 22:10:22 +0200 | [diff] [blame] | 2086 | if (r) |
| 2087 | return -EFAULT; |
Paolo Bonzini | bc009e4 | 2015-05-26 12:43:41 +0200 | [diff] [blame] | 2088 | mark_page_dirty_in_slot(memslot, gfn); |
Izik Eidus | 195aefd | 2007-10-01 22:14:18 +0200 | [diff] [blame] | 2089 | return 0; |
| 2090 | } |
Paolo Bonzini | 8e73485 | 2015-05-17 13:58:53 +0200 | [diff] [blame] | 2091 | |
| 2092 | int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, |
| 2093 | const void *data, int offset, int len) |
| 2094 | { |
| 2095 | struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn); |
| 2096 | |
| 2097 | return __kvm_write_guest_page(slot, gfn, data, offset, len); |
| 2098 | } |
Izik Eidus | 195aefd | 2007-10-01 22:14:18 +0200 | [diff] [blame] | 2099 | EXPORT_SYMBOL_GPL(kvm_write_guest_page); |
| 2100 | |
Paolo Bonzini | 8e73485 | 2015-05-17 13:58:53 +0200 | [diff] [blame] | 2101 | int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, |
| 2102 | const void *data, int offset, int len) |
| 2103 | { |
| 2104 | struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); |
| 2105 | |
| 2106 | return __kvm_write_guest_page(slot, gfn, data, offset, len); |
| 2107 | } |
| 2108 | EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page); |
| 2109 | |
Izik Eidus | 195aefd | 2007-10-01 22:14:18 +0200 | [diff] [blame] | 2110 | int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data, |
| 2111 | unsigned long len) |
| 2112 | { |
| 2113 | gfn_t gfn = gpa >> PAGE_SHIFT; |
| 2114 | int seg; |
| 2115 | int offset = offset_in_page(gpa); |
| 2116 | int ret; |
| 2117 | |
| 2118 | while ((seg = next_segment(len, offset)) != 0) { |
| 2119 | ret = kvm_write_guest_page(kvm, gfn, data, offset, seg); |
| 2120 | if (ret < 0) |
| 2121 | return ret; |
| 2122 | offset = 0; |
| 2123 | len -= seg; |
| 2124 | data += seg; |
| 2125 | ++gfn; |
| 2126 | } |
| 2127 | return 0; |
| 2128 | } |
Wincy Van | ff651cb | 2014-12-11 08:52:58 +0300 | [diff] [blame] | 2129 | EXPORT_SYMBOL_GPL(kvm_write_guest); |
Izik Eidus | 195aefd | 2007-10-01 22:14:18 +0200 | [diff] [blame] | 2130 | |
Paolo Bonzini | 8e73485 | 2015-05-17 13:58:53 +0200 | [diff] [blame] | 2131 | int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data, |
| 2132 | unsigned long len) |
| 2133 | { |
| 2134 | gfn_t gfn = gpa >> PAGE_SHIFT; |
| 2135 | int seg; |
| 2136 | int offset = offset_in_page(gpa); |
| 2137 | int ret; |
| 2138 | |
| 2139 | while ((seg = next_segment(len, offset)) != 0) { |
| 2140 | ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg); |
| 2141 | if (ret < 0) |
| 2142 | return ret; |
| 2143 | offset = 0; |
| 2144 | len -= seg; |
| 2145 | data += seg; |
| 2146 | ++gfn; |
| 2147 | } |
| 2148 | return 0; |
| 2149 | } |
| 2150 | EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest); |
| 2151 | |
Paolo Bonzini | 5a2d436 | 2017-02-03 20:32:28 -0800 | [diff] [blame] | 2152 | static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots, |
| 2153 | struct gfn_to_hva_cache *ghc, |
| 2154 | gpa_t gpa, unsigned long len) |
Gleb Natapov | 49c7754 | 2010-10-18 15:22:23 +0200 | [diff] [blame] | 2155 | { |
Gleb Natapov | 49c7754 | 2010-10-18 15:22:23 +0200 | [diff] [blame] | 2156 | int offset = offset_in_page(gpa); |
Andrew Honig | 8f96452 | 2013-03-29 09:35:21 -0700 | [diff] [blame] | 2157 | gfn_t start_gfn = gpa >> PAGE_SHIFT; |
| 2158 | gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT; |
| 2159 | gfn_t nr_pages_needed = end_gfn - start_gfn + 1; |
| 2160 | gfn_t nr_pages_avail; |
Jim Mattson | f1b9dd5 | 2018-12-17 13:53:33 -0800 | [diff] [blame] | 2161 | int r = start_gfn <= end_gfn ? 0 : -EINVAL; |
Gleb Natapov | 49c7754 | 2010-10-18 15:22:23 +0200 | [diff] [blame] | 2162 | |
| 2163 | ghc->gpa = gpa; |
| 2164 | ghc->generation = slots->generation; |
Andrew Honig | 8f96452 | 2013-03-29 09:35:21 -0700 | [diff] [blame] | 2165 | ghc->len = len; |
Jim Mattson | f1b9dd5 | 2018-12-17 13:53:33 -0800 | [diff] [blame] | 2166 | ghc->hva = KVM_HVA_ERR_BAD; |
| 2167 | |
| 2168 | /* |
| 2169 | * If the requested region crosses two memslots, we still |
| 2170 | * verify that the entire region is valid here. |
| 2171 | */ |
| 2172 | while (!r && start_gfn <= end_gfn) { |
| 2173 | ghc->memslot = __gfn_to_memslot(slots, start_gfn); |
| 2174 | ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, |
| 2175 | &nr_pages_avail); |
| 2176 | if (kvm_is_error_hva(ghc->hva)) |
| 2177 | r = -EFAULT; |
| 2178 | start_gfn += nr_pages_avail; |
Andrew Honig | 8f96452 | 2013-03-29 09:35:21 -0700 | [diff] [blame] | 2179 | } |
Jim Mattson | f1b9dd5 | 2018-12-17 13:53:33 -0800 | [diff] [blame] | 2180 | |
| 2181 | /* Use the slow path for cross page reads and writes. */ |
| 2182 | if (!r && nr_pages_needed == 1) |
| 2183 | ghc->hva += offset; |
| 2184 | else |
| 2185 | ghc->memslot = NULL; |
| 2186 | |
| 2187 | return r; |
Gleb Natapov | 49c7754 | 2010-10-18 15:22:23 +0200 | [diff] [blame] | 2188 | } |
Paolo Bonzini | 5a2d436 | 2017-02-03 20:32:28 -0800 | [diff] [blame] | 2189 | |
Paolo Bonzini | 4e335d9 | 2017-05-02 16:20:18 +0200 | [diff] [blame] | 2190 | int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
Paolo Bonzini | 5a2d436 | 2017-02-03 20:32:28 -0800 | [diff] [blame] | 2191 | gpa_t gpa, unsigned long len) |
| 2192 | { |
Paolo Bonzini | 4e335d9 | 2017-05-02 16:20:18 +0200 | [diff] [blame] | 2193 | struct kvm_memslots *slots = kvm_memslots(kvm); |
Paolo Bonzini | 5a2d436 | 2017-02-03 20:32:28 -0800 | [diff] [blame] | 2194 | return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len); |
| 2195 | } |
Paolo Bonzini | 4e335d9 | 2017-05-02 16:20:18 +0200 | [diff] [blame] | 2196 | EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init); |
Gleb Natapov | 49c7754 | 2010-10-18 15:22:23 +0200 | [diff] [blame] | 2197 | |
Paolo Bonzini | 4e335d9 | 2017-05-02 16:20:18 +0200 | [diff] [blame] | 2198 | int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
Jim Mattson | 7a86dab | 2018-12-14 14:34:43 -0800 | [diff] [blame] | 2199 | void *data, unsigned int offset, |
| 2200 | unsigned long len) |
Gleb Natapov | 49c7754 | 2010-10-18 15:22:23 +0200 | [diff] [blame] | 2201 | { |
Paolo Bonzini | 4e335d9 | 2017-05-02 16:20:18 +0200 | [diff] [blame] | 2202 | struct kvm_memslots *slots = kvm_memslots(kvm); |
Gleb Natapov | 49c7754 | 2010-10-18 15:22:23 +0200 | [diff] [blame] | 2203 | int r; |
Pan Xinhui | 4ec6e86 | 2016-11-02 05:08:34 -0400 | [diff] [blame] | 2204 | gpa_t gpa = ghc->gpa + offset; |
Gleb Natapov | 49c7754 | 2010-10-18 15:22:23 +0200 | [diff] [blame] | 2205 | |
Pan Xinhui | 4ec6e86 | 2016-11-02 05:08:34 -0400 | [diff] [blame] | 2206 | BUG_ON(len + offset > ghc->len); |
Andrew Honig | 8f96452 | 2013-03-29 09:35:21 -0700 | [diff] [blame] | 2207 | |
Gleb Natapov | 49c7754 | 2010-10-18 15:22:23 +0200 | [diff] [blame] | 2208 | if (slots->generation != ghc->generation) |
Paolo Bonzini | 5a2d436 | 2017-02-03 20:32:28 -0800 | [diff] [blame] | 2209 | __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len); |
Andrew Honig | 8f96452 | 2013-03-29 09:35:21 -0700 | [diff] [blame] | 2210 | |
| 2211 | if (unlikely(!ghc->memslot)) |
Paolo Bonzini | 4e335d9 | 2017-05-02 16:20:18 +0200 | [diff] [blame] | 2212 | return kvm_write_guest(kvm, gpa, data, len); |
Gleb Natapov | 49c7754 | 2010-10-18 15:22:23 +0200 | [diff] [blame] | 2213 | |
| 2214 | if (kvm_is_error_hva(ghc->hva)) |
| 2215 | return -EFAULT; |
| 2216 | |
Pan Xinhui | 4ec6e86 | 2016-11-02 05:08:34 -0400 | [diff] [blame] | 2217 | r = __copy_to_user((void __user *)ghc->hva + offset, data, len); |
Gleb Natapov | 49c7754 | 2010-10-18 15:22:23 +0200 | [diff] [blame] | 2218 | if (r) |
| 2219 | return -EFAULT; |
Pan Xinhui | 4ec6e86 | 2016-11-02 05:08:34 -0400 | [diff] [blame] | 2220 | mark_page_dirty_in_slot(ghc->memslot, gpa >> PAGE_SHIFT); |
Gleb Natapov | 49c7754 | 2010-10-18 15:22:23 +0200 | [diff] [blame] | 2221 | |
| 2222 | return 0; |
| 2223 | } |
Paolo Bonzini | 4e335d9 | 2017-05-02 16:20:18 +0200 | [diff] [blame] | 2224 | EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached); |
Pan Xinhui | 4ec6e86 | 2016-11-02 05:08:34 -0400 | [diff] [blame] | 2225 | |
Paolo Bonzini | 4e335d9 | 2017-05-02 16:20:18 +0200 | [diff] [blame] | 2226 | int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
| 2227 | void *data, unsigned long len) |
Pan Xinhui | 4ec6e86 | 2016-11-02 05:08:34 -0400 | [diff] [blame] | 2228 | { |
Paolo Bonzini | 4e335d9 | 2017-05-02 16:20:18 +0200 | [diff] [blame] | 2229 | return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len); |
Pan Xinhui | 4ec6e86 | 2016-11-02 05:08:34 -0400 | [diff] [blame] | 2230 | } |
Paolo Bonzini | 4e335d9 | 2017-05-02 16:20:18 +0200 | [diff] [blame] | 2231 | EXPORT_SYMBOL_GPL(kvm_write_guest_cached); |
Gleb Natapov | 49c7754 | 2010-10-18 15:22:23 +0200 | [diff] [blame] | 2232 | |
Paolo Bonzini | 4e335d9 | 2017-05-02 16:20:18 +0200 | [diff] [blame] | 2233 | int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc, |
| 2234 | void *data, unsigned long len) |
Gleb Natapov | e03b644 | 2011-07-11 15:28:11 -0400 | [diff] [blame] | 2235 | { |
Paolo Bonzini | 4e335d9 | 2017-05-02 16:20:18 +0200 | [diff] [blame] | 2236 | struct kvm_memslots *slots = kvm_memslots(kvm); |
Gleb Natapov | e03b644 | 2011-07-11 15:28:11 -0400 | [diff] [blame] | 2237 | int r; |
| 2238 | |
Andrew Honig | 8f96452 | 2013-03-29 09:35:21 -0700 | [diff] [blame] | 2239 | BUG_ON(len > ghc->len); |
| 2240 | |
Gleb Natapov | e03b644 | 2011-07-11 15:28:11 -0400 | [diff] [blame] | 2241 | if (slots->generation != ghc->generation) |
Paolo Bonzini | 5a2d436 | 2017-02-03 20:32:28 -0800 | [diff] [blame] | 2242 | __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len); |
Andrew Honig | 8f96452 | 2013-03-29 09:35:21 -0700 | [diff] [blame] | 2243 | |
| 2244 | if (unlikely(!ghc->memslot)) |
Paolo Bonzini | 4e335d9 | 2017-05-02 16:20:18 +0200 | [diff] [blame] | 2245 | return kvm_read_guest(kvm, ghc->gpa, data, len); |
Gleb Natapov | e03b644 | 2011-07-11 15:28:11 -0400 | [diff] [blame] | 2246 | |
| 2247 | if (kvm_is_error_hva(ghc->hva)) |
| 2248 | return -EFAULT; |
| 2249 | |
| 2250 | r = __copy_from_user(data, (void __user *)ghc->hva, len); |
| 2251 | if (r) |
| 2252 | return -EFAULT; |
| 2253 | |
| 2254 | return 0; |
| 2255 | } |
Paolo Bonzini | 4e335d9 | 2017-05-02 16:20:18 +0200 | [diff] [blame] | 2256 | EXPORT_SYMBOL_GPL(kvm_read_guest_cached); |
Gleb Natapov | e03b644 | 2011-07-11 15:28:11 -0400 | [diff] [blame] | 2257 | |
Izik Eidus | 195aefd | 2007-10-01 22:14:18 +0200 | [diff] [blame] | 2258 | int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len) |
| 2259 | { |
Heiko Carstens | 8a3caa6 | 2013-11-18 10:35:55 +0100 | [diff] [blame] | 2260 | const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0))); |
| 2261 | |
| 2262 | return kvm_write_guest_page(kvm, gfn, zero_page, offset, len); |
Izik Eidus | 195aefd | 2007-10-01 22:14:18 +0200 | [diff] [blame] | 2263 | } |
| 2264 | EXPORT_SYMBOL_GPL(kvm_clear_guest_page); |
| 2265 | |
| 2266 | int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len) |
| 2267 | { |
| 2268 | gfn_t gfn = gpa >> PAGE_SHIFT; |
| 2269 | int seg; |
| 2270 | int offset = offset_in_page(gpa); |
| 2271 | int ret; |
| 2272 | |
Kevin Mulvey | bfda0e8 | 2015-02-20 08:21:36 -0500 | [diff] [blame] | 2273 | while ((seg = next_segment(len, offset)) != 0) { |
Izik Eidus | 195aefd | 2007-10-01 22:14:18 +0200 | [diff] [blame] | 2274 | ret = kvm_clear_guest_page(kvm, gfn, offset, seg); |
| 2275 | if (ret < 0) |
| 2276 | return ret; |
| 2277 | offset = 0; |
| 2278 | len -= seg; |
| 2279 | ++gfn; |
| 2280 | } |
| 2281 | return 0; |
| 2282 | } |
| 2283 | EXPORT_SYMBOL_GPL(kvm_clear_guest); |
| 2284 | |
Paolo Bonzini | bc009e4 | 2015-05-26 12:43:41 +0200 | [diff] [blame] | 2285 | static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, |
Stephen Hemminger | 7940876 | 2013-12-29 12:12:29 -0800 | [diff] [blame] | 2286 | gfn_t gfn) |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 2287 | { |
Rusty Russell | 7e9d619 | 2007-07-31 20:41:14 +1000 | [diff] [blame] | 2288 | if (memslot && memslot->dirty_bitmap) { |
| 2289 | unsigned long rel_gfn = gfn - memslot->base_gfn; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 2290 | |
Takuya Yoshikawa | b74ca3b | 2012-10-04 17:13:12 -0700 | [diff] [blame] | 2291 | set_bit_le(rel_gfn, memslot->dirty_bitmap); |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 2292 | } |
| 2293 | } |
| 2294 | |
Gleb Natapov | 49c7754 | 2010-10-18 15:22:23 +0200 | [diff] [blame] | 2295 | void mark_page_dirty(struct kvm *kvm, gfn_t gfn) |
| 2296 | { |
| 2297 | struct kvm_memory_slot *memslot; |
| 2298 | |
| 2299 | memslot = gfn_to_memslot(kvm, gfn); |
Paolo Bonzini | bc009e4 | 2015-05-26 12:43:41 +0200 | [diff] [blame] | 2300 | mark_page_dirty_in_slot(memslot, gfn); |
Gleb Natapov | 49c7754 | 2010-10-18 15:22:23 +0200 | [diff] [blame] | 2301 | } |
Aneesh Kumar K.V | 2ba9f0d | 2013-10-07 22:17:59 +0530 | [diff] [blame] | 2302 | EXPORT_SYMBOL_GPL(mark_page_dirty); |
Gleb Natapov | 49c7754 | 2010-10-18 15:22:23 +0200 | [diff] [blame] | 2303 | |
Paolo Bonzini | 8e73485 | 2015-05-17 13:58:53 +0200 | [diff] [blame] | 2304 | void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn) |
| 2305 | { |
| 2306 | struct kvm_memory_slot *memslot; |
| 2307 | |
| 2308 | memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn); |
| 2309 | mark_page_dirty_in_slot(memslot, gfn); |
| 2310 | } |
| 2311 | EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty); |
| 2312 | |
Jan H. Schönherr | 20b7035 | 2017-11-24 22:39:01 +0100 | [diff] [blame] | 2313 | void kvm_sigset_activate(struct kvm_vcpu *vcpu) |
| 2314 | { |
| 2315 | if (!vcpu->sigset_active) |
| 2316 | return; |
| 2317 | |
| 2318 | /* |
| 2319 | * This does a lockless modification of ->real_blocked, which is fine |
| 2320 | * because, only current can change ->real_blocked and all readers of |
| 2321 | * ->real_blocked don't care as long ->real_blocked is always a subset |
| 2322 | * of ->blocked. |
| 2323 | */ |
| 2324 | sigprocmask(SIG_SETMASK, &vcpu->sigset, ¤t->real_blocked); |
| 2325 | } |
| 2326 | |
| 2327 | void kvm_sigset_deactivate(struct kvm_vcpu *vcpu) |
| 2328 | { |
| 2329 | if (!vcpu->sigset_active) |
| 2330 | return; |
| 2331 | |
| 2332 | sigprocmask(SIG_SETMASK, ¤t->real_blocked, NULL); |
| 2333 | sigemptyset(¤t->real_blocked); |
| 2334 | } |
| 2335 | |
Wanpeng Li | aca6ff2 | 2015-09-03 22:07:38 +0800 | [diff] [blame] | 2336 | static void grow_halt_poll_ns(struct kvm_vcpu *vcpu) |
| 2337 | { |
Nir Weiner | dee339b | 2019-01-27 12:17:16 +0200 | [diff] [blame] | 2338 | unsigned int old, val, grow, grow_start; |
Wanpeng Li | aca6ff2 | 2015-09-03 22:07:38 +0800 | [diff] [blame] | 2339 | |
Wanpeng Li | 2cbd782 | 2015-09-03 22:07:39 +0800 | [diff] [blame] | 2340 | old = val = vcpu->halt_poll_ns; |
Nir Weiner | dee339b | 2019-01-27 12:17:16 +0200 | [diff] [blame] | 2341 | grow_start = READ_ONCE(halt_poll_ns_grow_start); |
Christian Borntraeger | 6b6de68 | 2016-02-09 13:47:55 +0100 | [diff] [blame] | 2342 | grow = READ_ONCE(halt_poll_ns_grow); |
Nir Weiner | 7fa08e7 | 2019-01-27 12:17:14 +0200 | [diff] [blame] | 2343 | if (!grow) |
| 2344 | goto out; |
| 2345 | |
Nir Weiner | dee339b | 2019-01-27 12:17:16 +0200 | [diff] [blame] | 2346 | val *= grow; |
| 2347 | if (val < grow_start) |
| 2348 | val = grow_start; |
Wanpeng Li | aca6ff2 | 2015-09-03 22:07:38 +0800 | [diff] [blame] | 2349 | |
David Matlack | 313f636 | 2016-03-08 16:19:44 -0800 | [diff] [blame] | 2350 | if (val > halt_poll_ns) |
| 2351 | val = halt_poll_ns; |
| 2352 | |
Wanpeng Li | aca6ff2 | 2015-09-03 22:07:38 +0800 | [diff] [blame] | 2353 | vcpu->halt_poll_ns = val; |
Nir Weiner | 7fa08e7 | 2019-01-27 12:17:14 +0200 | [diff] [blame] | 2354 | out: |
Wanpeng Li | 2cbd782 | 2015-09-03 22:07:39 +0800 | [diff] [blame] | 2355 | trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old); |
Wanpeng Li | aca6ff2 | 2015-09-03 22:07:38 +0800 | [diff] [blame] | 2356 | } |
| 2357 | |
| 2358 | static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu) |
| 2359 | { |
Christian Borntraeger | 6b6de68 | 2016-02-09 13:47:55 +0100 | [diff] [blame] | 2360 | unsigned int old, val, shrink; |
Wanpeng Li | aca6ff2 | 2015-09-03 22:07:38 +0800 | [diff] [blame] | 2361 | |
Wanpeng Li | 2cbd782 | 2015-09-03 22:07:39 +0800 | [diff] [blame] | 2362 | old = val = vcpu->halt_poll_ns; |
Christian Borntraeger | 6b6de68 | 2016-02-09 13:47:55 +0100 | [diff] [blame] | 2363 | shrink = READ_ONCE(halt_poll_ns_shrink); |
| 2364 | if (shrink == 0) |
Wanpeng Li | aca6ff2 | 2015-09-03 22:07:38 +0800 | [diff] [blame] | 2365 | val = 0; |
| 2366 | else |
Christian Borntraeger | 6b6de68 | 2016-02-09 13:47:55 +0100 | [diff] [blame] | 2367 | val /= shrink; |
Wanpeng Li | aca6ff2 | 2015-09-03 22:07:38 +0800 | [diff] [blame] | 2368 | |
| 2369 | vcpu->halt_poll_ns = val; |
Wanpeng Li | 2cbd782 | 2015-09-03 22:07:39 +0800 | [diff] [blame] | 2370 | trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old); |
Wanpeng Li | aca6ff2 | 2015-09-03 22:07:38 +0800 | [diff] [blame] | 2371 | } |
| 2372 | |
Paolo Bonzini | f781951 | 2015-02-04 18:20:58 +0100 | [diff] [blame] | 2373 | static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu) |
| 2374 | { |
Junaid Shahid | 50c28f2 | 2018-06-27 14:59:11 -0700 | [diff] [blame] | 2375 | int ret = -EINTR; |
| 2376 | int idx = srcu_read_lock(&vcpu->kvm->srcu); |
| 2377 | |
Paolo Bonzini | f781951 | 2015-02-04 18:20:58 +0100 | [diff] [blame] | 2378 | if (kvm_arch_vcpu_runnable(vcpu)) { |
| 2379 | kvm_make_request(KVM_REQ_UNHALT, vcpu); |
Junaid Shahid | 50c28f2 | 2018-06-27 14:59:11 -0700 | [diff] [blame] | 2380 | goto out; |
Paolo Bonzini | f781951 | 2015-02-04 18:20:58 +0100 | [diff] [blame] | 2381 | } |
| 2382 | if (kvm_cpu_has_pending_timer(vcpu)) |
Junaid Shahid | 50c28f2 | 2018-06-27 14:59:11 -0700 | [diff] [blame] | 2383 | goto out; |
Paolo Bonzini | f781951 | 2015-02-04 18:20:58 +0100 | [diff] [blame] | 2384 | if (signal_pending(current)) |
Junaid Shahid | 50c28f2 | 2018-06-27 14:59:11 -0700 | [diff] [blame] | 2385 | goto out; |
Paolo Bonzini | f781951 | 2015-02-04 18:20:58 +0100 | [diff] [blame] | 2386 | |
Junaid Shahid | 50c28f2 | 2018-06-27 14:59:11 -0700 | [diff] [blame] | 2387 | ret = 0; |
| 2388 | out: |
| 2389 | srcu_read_unlock(&vcpu->kvm->srcu, idx); |
| 2390 | return ret; |
Paolo Bonzini | f781951 | 2015-02-04 18:20:58 +0100 | [diff] [blame] | 2391 | } |
| 2392 | |
Eddie Dong | b6958ce | 2007-07-18 12:15:21 +0300 | [diff] [blame] | 2393 | /* |
| 2394 | * The vCPU has executed a HLT instruction with in-kernel mode enabled. |
| 2395 | */ |
Hollis Blanchard | 8776e51 | 2007-10-31 17:24:24 -0500 | [diff] [blame] | 2396 | void kvm_vcpu_block(struct kvm_vcpu *vcpu) |
Eddie Dong | b6958ce | 2007-07-18 12:15:21 +0300 | [diff] [blame] | 2397 | { |
Paolo Bonzini | f781951 | 2015-02-04 18:20:58 +0100 | [diff] [blame] | 2398 | ktime_t start, cur; |
Marcelo Tosatti | 8577370 | 2016-02-19 09:46:39 +0100 | [diff] [blame] | 2399 | DECLARE_SWAITQUEUE(wait); |
Paolo Bonzini | f781951 | 2015-02-04 18:20:58 +0100 | [diff] [blame] | 2400 | bool waited = false; |
Wanpeng Li | aca6ff2 | 2015-09-03 22:07:38 +0800 | [diff] [blame] | 2401 | u64 block_ns; |
Paolo Bonzini | f781951 | 2015-02-04 18:20:58 +0100 | [diff] [blame] | 2402 | |
Marc Zyngier | 07ab0f8 | 2019-08-02 11:37:09 +0100 | [diff] [blame] | 2403 | kvm_arch_vcpu_blocking(vcpu); |
| 2404 | |
Paolo Bonzini | f781951 | 2015-02-04 18:20:58 +0100 | [diff] [blame] | 2405 | start = cur = ktime_get(); |
Christian Borntraeger | cdd6ad3 | 2019-03-05 05:30:01 -0500 | [diff] [blame] | 2406 | if (vcpu->halt_poll_ns && !kvm_arch_no_poll(vcpu)) { |
Wanpeng Li | 19020f8 | 2015-09-03 22:07:37 +0800 | [diff] [blame] | 2407 | ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns); |
Xiubo Li | f95ef0cd | 2015-02-26 14:58:23 +0800 | [diff] [blame] | 2408 | |
Paolo Bonzini | 62bea5b | 2015-09-15 18:27:57 +0200 | [diff] [blame] | 2409 | ++vcpu->stat.halt_attempted_poll; |
Paolo Bonzini | f781951 | 2015-02-04 18:20:58 +0100 | [diff] [blame] | 2410 | do { |
| 2411 | /* |
| 2412 | * This sets KVM_REQ_UNHALT if an interrupt |
| 2413 | * arrives. |
| 2414 | */ |
| 2415 | if (kvm_vcpu_check_block(vcpu) < 0) { |
| 2416 | ++vcpu->stat.halt_successful_poll; |
Christian Borntraeger | 3491caf | 2016-05-13 12:16:35 +0200 | [diff] [blame] | 2417 | if (!vcpu_valid_wakeup(vcpu)) |
| 2418 | ++vcpu->stat.halt_poll_invalid; |
Paolo Bonzini | f781951 | 2015-02-04 18:20:58 +0100 | [diff] [blame] | 2419 | goto out; |
| 2420 | } |
| 2421 | cur = ktime_get(); |
| 2422 | } while (single_task_running() && ktime_before(cur, stop)); |
| 2423 | } |
Eddie Dong | b6958ce | 2007-07-18 12:15:21 +0300 | [diff] [blame] | 2424 | |
Marcelo Tosatti | e5c239c | 2008-05-08 19:47:01 -0300 | [diff] [blame] | 2425 | for (;;) { |
Peter Zijlstra | b3dae10 | 2018-06-12 10:34:52 +0200 | [diff] [blame] | 2426 | prepare_to_swait_exclusive(&vcpu->wq, &wait, TASK_INTERRUPTIBLE); |
Eddie Dong | b6958ce | 2007-07-18 12:15:21 +0300 | [diff] [blame] | 2427 | |
Paolo Bonzini | f781951 | 2015-02-04 18:20:58 +0100 | [diff] [blame] | 2428 | if (kvm_vcpu_check_block(vcpu) < 0) |
Marcelo Tosatti | e5c239c | 2008-05-08 19:47:01 -0300 | [diff] [blame] | 2429 | break; |
| 2430 | |
Paolo Bonzini | f781951 | 2015-02-04 18:20:58 +0100 | [diff] [blame] | 2431 | waited = true; |
Eddie Dong | b6958ce | 2007-07-18 12:15:21 +0300 | [diff] [blame] | 2432 | schedule(); |
Eddie Dong | b6958ce | 2007-07-18 12:15:21 +0300 | [diff] [blame] | 2433 | } |
| 2434 | |
Marcelo Tosatti | 8577370 | 2016-02-19 09:46:39 +0100 | [diff] [blame] | 2435 | finish_swait(&vcpu->wq, &wait); |
Paolo Bonzini | f781951 | 2015-02-04 18:20:58 +0100 | [diff] [blame] | 2436 | cur = ktime_get(); |
Paolo Bonzini | f781951 | 2015-02-04 18:20:58 +0100 | [diff] [blame] | 2437 | out: |
Marc Zyngier | 07ab0f8 | 2019-08-02 11:37:09 +0100 | [diff] [blame] | 2438 | kvm_arch_vcpu_unblocking(vcpu); |
Wanpeng Li | aca6ff2 | 2015-09-03 22:07:38 +0800 | [diff] [blame] | 2439 | block_ns = ktime_to_ns(cur) - ktime_to_ns(start); |
| 2440 | |
Wanpeng Li | 44551b2 | 2019-09-29 09:06:56 +0800 | [diff] [blame] | 2441 | if (!kvm_arch_no_poll(vcpu)) { |
| 2442 | if (!vcpu_valid_wakeup(vcpu)) { |
Wanpeng Li | aca6ff2 | 2015-09-03 22:07:38 +0800 | [diff] [blame] | 2443 | shrink_halt_poll_ns(vcpu); |
Wanpeng Li | 44551b2 | 2019-09-29 09:06:56 +0800 | [diff] [blame] | 2444 | } else if (halt_poll_ns) { |
| 2445 | if (block_ns <= vcpu->halt_poll_ns) |
| 2446 | ; |
| 2447 | /* we had a long block, shrink polling */ |
| 2448 | else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns) |
| 2449 | shrink_halt_poll_ns(vcpu); |
| 2450 | /* we had a short halt and our poll time is too small */ |
| 2451 | else if (vcpu->halt_poll_ns < halt_poll_ns && |
| 2452 | block_ns < halt_poll_ns) |
| 2453 | grow_halt_poll_ns(vcpu); |
| 2454 | } else { |
| 2455 | vcpu->halt_poll_ns = 0; |
| 2456 | } |
| 2457 | } |
Wanpeng Li | aca6ff2 | 2015-09-03 22:07:38 +0800 | [diff] [blame] | 2458 | |
Christian Borntraeger | 3491caf | 2016-05-13 12:16:35 +0200 | [diff] [blame] | 2459 | trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu)); |
| 2460 | kvm_arch_vcpu_block_finish(vcpu); |
Eddie Dong | b6958ce | 2007-07-18 12:15:21 +0300 | [diff] [blame] | 2461 | } |
Aneesh Kumar K.V | 2ba9f0d | 2013-10-07 22:17:59 +0530 | [diff] [blame] | 2462 | EXPORT_SYMBOL_GPL(kvm_vcpu_block); |
Eddie Dong | b6958ce | 2007-07-18 12:15:21 +0300 | [diff] [blame] | 2463 | |
Radim Krčmář | 178f02f | 2017-04-26 22:32:26 +0200 | [diff] [blame] | 2464 | bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu) |
Christoffer Dall | b6d3383 | 2012-03-08 16:44:24 -0500 | [diff] [blame] | 2465 | { |
Marcelo Tosatti | 8577370 | 2016-02-19 09:46:39 +0100 | [diff] [blame] | 2466 | struct swait_queue_head *wqp; |
Christoffer Dall | b6d3383 | 2012-03-08 16:44:24 -0500 | [diff] [blame] | 2467 | |
| 2468 | wqp = kvm_arch_vcpu_wq(vcpu); |
Davidlohr Bueso | 5e0018b | 2017-09-13 13:08:22 -0700 | [diff] [blame] | 2469 | if (swq_has_sleeper(wqp)) { |
Peter Zijlstra | b3dae10 | 2018-06-12 10:34:52 +0200 | [diff] [blame] | 2470 | swake_up_one(wqp); |
Wanpeng Li | d73eb57 | 2019-07-18 19:39:06 +0800 | [diff] [blame] | 2471 | WRITE_ONCE(vcpu->ready, true); |
Christoffer Dall | b6d3383 | 2012-03-08 16:44:24 -0500 | [diff] [blame] | 2472 | ++vcpu->stat.halt_wakeup; |
Radim Krčmář | 178f02f | 2017-04-26 22:32:26 +0200 | [diff] [blame] | 2473 | return true; |
Christoffer Dall | b6d3383 | 2012-03-08 16:44:24 -0500 | [diff] [blame] | 2474 | } |
| 2475 | |
Radim Krčmář | 178f02f | 2017-04-26 22:32:26 +0200 | [diff] [blame] | 2476 | return false; |
Radim Krčmář | dd1a4cc | 2016-05-04 14:09:44 -0500 | [diff] [blame] | 2477 | } |
| 2478 | EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up); |
| 2479 | |
Paolo Bonzini | 0266c89 | 2017-05-04 15:14:13 +0200 | [diff] [blame] | 2480 | #ifndef CONFIG_S390 |
Radim Krčmář | dd1a4cc | 2016-05-04 14:09:44 -0500 | [diff] [blame] | 2481 | /* |
| 2482 | * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode. |
| 2483 | */ |
| 2484 | void kvm_vcpu_kick(struct kvm_vcpu *vcpu) |
| 2485 | { |
| 2486 | int me; |
| 2487 | int cpu = vcpu->cpu; |
| 2488 | |
Radim Krčmář | 178f02f | 2017-04-26 22:32:26 +0200 | [diff] [blame] | 2489 | if (kvm_vcpu_wake_up(vcpu)) |
| 2490 | return; |
| 2491 | |
Christoffer Dall | b6d3383 | 2012-03-08 16:44:24 -0500 | [diff] [blame] | 2492 | me = get_cpu(); |
| 2493 | if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu)) |
| 2494 | if (kvm_arch_vcpu_should_kick(vcpu)) |
| 2495 | smp_send_reschedule(cpu); |
| 2496 | put_cpu(); |
| 2497 | } |
Yang Zhang | a20ed54 | 2013-04-11 19:25:15 +0800 | [diff] [blame] | 2498 | EXPORT_SYMBOL_GPL(kvm_vcpu_kick); |
Paolo Bonzini | 0266c89 | 2017-05-04 15:14:13 +0200 | [diff] [blame] | 2499 | #endif /* !CONFIG_S390 */ |
Christoffer Dall | b6d3383 | 2012-03-08 16:44:24 -0500 | [diff] [blame] | 2500 | |
Dan Carpenter | fa93384 | 2014-05-23 13:20:42 +0300 | [diff] [blame] | 2501 | int kvm_vcpu_yield_to(struct kvm_vcpu *target) |
Konstantin Weitz | 41628d3 | 2012-04-25 15:30:38 +0200 | [diff] [blame] | 2502 | { |
| 2503 | struct pid *pid; |
| 2504 | struct task_struct *task = NULL; |
Dan Carpenter | fa93384 | 2014-05-23 13:20:42 +0300 | [diff] [blame] | 2505 | int ret = 0; |
Konstantin Weitz | 41628d3 | 2012-04-25 15:30:38 +0200 | [diff] [blame] | 2506 | |
| 2507 | rcu_read_lock(); |
| 2508 | pid = rcu_dereference(target->pid); |
| 2509 | if (pid) |
Sam Bobroff | 27fbe64b | 2014-09-19 09:40:41 +1000 | [diff] [blame] | 2510 | task = get_pid_task(pid, PIDTYPE_PID); |
Konstantin Weitz | 41628d3 | 2012-04-25 15:30:38 +0200 | [diff] [blame] | 2511 | rcu_read_unlock(); |
| 2512 | if (!task) |
Raghavendra K T | c45c528 | 2013-01-22 13:09:24 +0530 | [diff] [blame] | 2513 | return ret; |
Raghavendra K T | c45c528 | 2013-01-22 13:09:24 +0530 | [diff] [blame] | 2514 | ret = yield_to(task, 1); |
Konstantin Weitz | 41628d3 | 2012-04-25 15:30:38 +0200 | [diff] [blame] | 2515 | put_task_struct(task); |
Raghavendra K T | c45c528 | 2013-01-22 13:09:24 +0530 | [diff] [blame] | 2516 | |
| 2517 | return ret; |
Konstantin Weitz | 41628d3 | 2012-04-25 15:30:38 +0200 | [diff] [blame] | 2518 | } |
| 2519 | EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to); |
| 2520 | |
Raghavendra K T | 06e48c5 | 2012-07-19 15:17:52 +0530 | [diff] [blame] | 2521 | /* |
| 2522 | * Helper that checks whether a VCPU is eligible for directed yield. |
| 2523 | * Most eligible candidate to yield is decided by following heuristics: |
| 2524 | * |
| 2525 | * (a) VCPU which has not done pl-exit or cpu relax intercepted recently |
| 2526 | * (preempted lock holder), indicated by @in_spin_loop. |
| 2527 | * Set at the beiginning and cleared at the end of interception/PLE handler. |
| 2528 | * |
| 2529 | * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get |
| 2530 | * chance last time (mostly it has become eligible now since we have probably |
| 2531 | * yielded to lockholder in last iteration. This is done by toggling |
| 2532 | * @dy_eligible each time a VCPU checked for eligibility.) |
| 2533 | * |
| 2534 | * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding |
| 2535 | * to preempted lock-holder could result in wrong VCPU selection and CPU |
| 2536 | * burning. Giving priority for a potential lock-holder increases lock |
| 2537 | * progress. |
| 2538 | * |
| 2539 | * Since algorithm is based on heuristics, accessing another VCPU data without |
| 2540 | * locking does not harm. It may result in trying to yield to same VCPU, fail |
| 2541 | * and continue with next VCPU and so on. |
| 2542 | */ |
Stephen Hemminger | 7940876 | 2013-12-29 12:12:29 -0800 | [diff] [blame] | 2543 | static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu) |
Raghavendra K T | 06e48c5 | 2012-07-19 15:17:52 +0530 | [diff] [blame] | 2544 | { |
Scott Wood | 4a55dd7 | 2014-01-09 18:43:16 -0600 | [diff] [blame] | 2545 | #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT |
Raghavendra K T | 06e48c5 | 2012-07-19 15:17:52 +0530 | [diff] [blame] | 2546 | bool eligible; |
| 2547 | |
| 2548 | eligible = !vcpu->spin_loop.in_spin_loop || |
Christian Borntraeger | 3465611 | 2014-09-04 21:13:31 +0200 | [diff] [blame] | 2549 | vcpu->spin_loop.dy_eligible; |
Raghavendra K T | 06e48c5 | 2012-07-19 15:17:52 +0530 | [diff] [blame] | 2550 | |
| 2551 | if (vcpu->spin_loop.in_spin_loop) |
| 2552 | kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible); |
| 2553 | |
| 2554 | return eligible; |
Scott Wood | 4a55dd7 | 2014-01-09 18:43:16 -0600 | [diff] [blame] | 2555 | #else |
| 2556 | return true; |
Raghavendra K T | 06e48c5 | 2012-07-19 15:17:52 +0530 | [diff] [blame] | 2557 | #endif |
Scott Wood | 4a55dd7 | 2014-01-09 18:43:16 -0600 | [diff] [blame] | 2558 | } |
Raghavendra K T | c45c528 | 2013-01-22 13:09:24 +0530 | [diff] [blame] | 2559 | |
Wanpeng Li | 17e433b | 2019-08-05 10:03:19 +0800 | [diff] [blame] | 2560 | /* |
| 2561 | * Unlike kvm_arch_vcpu_runnable, this function is called outside |
| 2562 | * a vcpu_load/vcpu_put pair. However, for most architectures |
| 2563 | * kvm_arch_vcpu_runnable does not require vcpu_load. |
| 2564 | */ |
| 2565 | bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu) |
| 2566 | { |
| 2567 | return kvm_arch_vcpu_runnable(vcpu); |
| 2568 | } |
| 2569 | |
| 2570 | static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu) |
| 2571 | { |
| 2572 | if (kvm_arch_dy_runnable(vcpu)) |
| 2573 | return true; |
| 2574 | |
| 2575 | #ifdef CONFIG_KVM_ASYNC_PF |
| 2576 | if (!list_empty_careful(&vcpu->async_pf.done)) |
| 2577 | return true; |
| 2578 | #endif |
| 2579 | |
| 2580 | return false; |
| 2581 | } |
| 2582 | |
Longpeng(Mike) | 199b576 | 2017-08-08 12:05:32 +0800 | [diff] [blame] | 2583 | void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode) |
Zhai, Edwin | d255f4f | 2009-10-09 18:03:20 +0800 | [diff] [blame] | 2584 | { |
Rik van Riel | 217ece6 | 2011-02-01 09:53:28 -0500 | [diff] [blame] | 2585 | struct kvm *kvm = me->kvm; |
| 2586 | struct kvm_vcpu *vcpu; |
| 2587 | int last_boosted_vcpu = me->kvm->last_boosted_vcpu; |
| 2588 | int yielded = 0; |
Raghavendra K T | c45c528 | 2013-01-22 13:09:24 +0530 | [diff] [blame] | 2589 | int try = 3; |
Rik van Riel | 217ece6 | 2011-02-01 09:53:28 -0500 | [diff] [blame] | 2590 | int pass; |
| 2591 | int i; |
Zhai, Edwin | d255f4f | 2009-10-09 18:03:20 +0800 | [diff] [blame] | 2592 | |
Raghavendra K T | 4c08849 | 2012-07-18 19:07:46 +0530 | [diff] [blame] | 2593 | kvm_vcpu_set_in_spin_loop(me, true); |
Rik van Riel | 217ece6 | 2011-02-01 09:53:28 -0500 | [diff] [blame] | 2594 | /* |
| 2595 | * We boost the priority of a VCPU that is runnable but not |
| 2596 | * currently running, because it got preempted by something |
| 2597 | * else and called schedule in __vcpu_run. Hopefully that |
| 2598 | * VCPU is holding the lock that we need and will release it. |
| 2599 | * We approximate round-robin by starting at the last boosted VCPU. |
| 2600 | */ |
Raghavendra K T | c45c528 | 2013-01-22 13:09:24 +0530 | [diff] [blame] | 2601 | for (pass = 0; pass < 2 && !yielded && try; pass++) { |
Rik van Riel | 217ece6 | 2011-02-01 09:53:28 -0500 | [diff] [blame] | 2602 | kvm_for_each_vcpu(i, vcpu, kvm) { |
Rik van Riel | 5cfc2aa | 2012-06-19 16:51:04 -0400 | [diff] [blame] | 2603 | if (!pass && i <= last_boosted_vcpu) { |
Rik van Riel | 217ece6 | 2011-02-01 09:53:28 -0500 | [diff] [blame] | 2604 | i = last_boosted_vcpu; |
| 2605 | continue; |
| 2606 | } else if (pass && i > last_boosted_vcpu) |
| 2607 | break; |
Wanpeng Li | d73eb57 | 2019-07-18 19:39:06 +0800 | [diff] [blame] | 2608 | if (!READ_ONCE(vcpu->ready)) |
Raghavendra K T | 7bc7ae25 | 2013-03-04 23:32:27 +0530 | [diff] [blame] | 2609 | continue; |
Rik van Riel | 217ece6 | 2011-02-01 09:53:28 -0500 | [diff] [blame] | 2610 | if (vcpu == me) |
| 2611 | continue; |
Wanpeng Li | 17e433b | 2019-08-05 10:03:19 +0800 | [diff] [blame] | 2612 | if (swait_active(&vcpu->wq) && !vcpu_dy_runnable(vcpu)) |
Rik van Riel | 217ece6 | 2011-02-01 09:53:28 -0500 | [diff] [blame] | 2613 | continue; |
Wanpeng Li | 046ddee | 2019-08-01 11:30:14 +0800 | [diff] [blame] | 2614 | if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode && |
| 2615 | !kvm_arch_vcpu_in_kernel(vcpu)) |
Longpeng(Mike) | 199b576 | 2017-08-08 12:05:32 +0800 | [diff] [blame] | 2616 | continue; |
Raghavendra K T | 06e48c5 | 2012-07-19 15:17:52 +0530 | [diff] [blame] | 2617 | if (!kvm_vcpu_eligible_for_directed_yield(vcpu)) |
| 2618 | continue; |
Raghavendra K T | c45c528 | 2013-01-22 13:09:24 +0530 | [diff] [blame] | 2619 | |
| 2620 | yielded = kvm_vcpu_yield_to(vcpu); |
| 2621 | if (yielded > 0) { |
Rik van Riel | 217ece6 | 2011-02-01 09:53:28 -0500 | [diff] [blame] | 2622 | kvm->last_boosted_vcpu = i; |
Rik van Riel | 217ece6 | 2011-02-01 09:53:28 -0500 | [diff] [blame] | 2623 | break; |
Raghavendra K T | c45c528 | 2013-01-22 13:09:24 +0530 | [diff] [blame] | 2624 | } else if (yielded < 0) { |
| 2625 | try--; |
| 2626 | if (!try) |
| 2627 | break; |
Rik van Riel | 217ece6 | 2011-02-01 09:53:28 -0500 | [diff] [blame] | 2628 | } |
Rik van Riel | 217ece6 | 2011-02-01 09:53:28 -0500 | [diff] [blame] | 2629 | } |
| 2630 | } |
Raghavendra K T | 4c08849 | 2012-07-18 19:07:46 +0530 | [diff] [blame] | 2631 | kvm_vcpu_set_in_spin_loop(me, false); |
Raghavendra K T | 06e48c5 | 2012-07-19 15:17:52 +0530 | [diff] [blame] | 2632 | |
| 2633 | /* Ensure vcpu is not eligible during next spinloop */ |
| 2634 | kvm_vcpu_set_dy_eligible(me, false); |
Zhai, Edwin | d255f4f | 2009-10-09 18:03:20 +0800 | [diff] [blame] | 2635 | } |
| 2636 | EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin); |
| 2637 | |
Souptick Joarder | 1499fa8 | 2018-04-19 00:49:58 +0530 | [diff] [blame] | 2638 | static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf) |
Avi Kivity | 9a2bb7f | 2007-02-22 12:58:31 +0200 | [diff] [blame] | 2639 | { |
Dave Jiang | 11bac80 | 2017-02-24 14:56:41 -0800 | [diff] [blame] | 2640 | struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data; |
Avi Kivity | 9a2bb7f | 2007-02-22 12:58:31 +0200 | [diff] [blame] | 2641 | struct page *page; |
| 2642 | |
npiggin@suse.de | e4a533a | 2007-12-05 18:15:52 +1100 | [diff] [blame] | 2643 | if (vmf->pgoff == 0) |
Avi Kivity | 039576c | 2007-03-20 12:46:50 +0200 | [diff] [blame] | 2644 | page = virt_to_page(vcpu->run); |
Avi Kivity | 0956676 | 2008-01-23 18:14:23 +0200 | [diff] [blame] | 2645 | #ifdef CONFIG_X86 |
npiggin@suse.de | e4a533a | 2007-12-05 18:15:52 +1100 | [diff] [blame] | 2646 | else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET) |
Zhang Xiantao | ad312c7 | 2007-12-13 23:50:52 +0800 | [diff] [blame] | 2647 | page = virt_to_page(vcpu->arch.pio_data); |
Avi Kivity | 0956676 | 2008-01-23 18:14:23 +0200 | [diff] [blame] | 2648 | #endif |
Paolo Bonzini | 4b4357e | 2017-03-31 13:53:23 +0200 | [diff] [blame] | 2649 | #ifdef CONFIG_KVM_MMIO |
Laurent Vivier | 5f94c17 | 2008-05-30 16:05:54 +0200 | [diff] [blame] | 2650 | else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET) |
| 2651 | page = virt_to_page(vcpu->kvm->coalesced_mmio_ring); |
| 2652 | #endif |
Avi Kivity | 039576c | 2007-03-20 12:46:50 +0200 | [diff] [blame] | 2653 | else |
Carsten Otte | 5b1c149 | 2012-01-04 10:25:23 +0100 | [diff] [blame] | 2654 | return kvm_arch_vcpu_fault(vcpu, vmf); |
Avi Kivity | 9a2bb7f | 2007-02-22 12:58:31 +0200 | [diff] [blame] | 2655 | get_page(page); |
npiggin@suse.de | e4a533a | 2007-12-05 18:15:52 +1100 | [diff] [blame] | 2656 | vmf->page = page; |
| 2657 | return 0; |
Avi Kivity | 9a2bb7f | 2007-02-22 12:58:31 +0200 | [diff] [blame] | 2658 | } |
| 2659 | |
Alexey Dobriyan | f0f37e2f | 2009-09-27 22:29:37 +0400 | [diff] [blame] | 2660 | static const struct vm_operations_struct kvm_vcpu_vm_ops = { |
npiggin@suse.de | e4a533a | 2007-12-05 18:15:52 +1100 | [diff] [blame] | 2661 | .fault = kvm_vcpu_fault, |
Avi Kivity | 9a2bb7f | 2007-02-22 12:58:31 +0200 | [diff] [blame] | 2662 | }; |
| 2663 | |
| 2664 | static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma) |
| 2665 | { |
| 2666 | vma->vm_ops = &kvm_vcpu_vm_ops; |
| 2667 | return 0; |
| 2668 | } |
| 2669 | |
Avi Kivity | bccf215 | 2007-02-21 18:04:26 +0200 | [diff] [blame] | 2670 | static int kvm_vcpu_release(struct inode *inode, struct file *filp) |
| 2671 | { |
| 2672 | struct kvm_vcpu *vcpu = filp->private_data; |
| 2673 | |
Luiz Capitulino | 45b5939 | 2016-09-16 10:27:35 -0400 | [diff] [blame] | 2674 | debugfs_remove_recursive(vcpu->debugfs_dentry); |
Al Viro | 66c0b39 | 2008-04-19 20:33:56 +0100 | [diff] [blame] | 2675 | kvm_put_kvm(vcpu->kvm); |
Avi Kivity | bccf215 | 2007-02-21 18:04:26 +0200 | [diff] [blame] | 2676 | return 0; |
| 2677 | } |
| 2678 | |
Christian Borntraeger | 3d3aab1 | 2008-12-02 11:17:32 +0100 | [diff] [blame] | 2679 | static struct file_operations kvm_vcpu_fops = { |
Avi Kivity | bccf215 | 2007-02-21 18:04:26 +0200 | [diff] [blame] | 2680 | .release = kvm_vcpu_release, |
| 2681 | .unlocked_ioctl = kvm_vcpu_ioctl, |
Avi Kivity | 9a2bb7f | 2007-02-22 12:58:31 +0200 | [diff] [blame] | 2682 | .mmap = kvm_vcpu_mmap, |
Arnd Bergmann | 6038f37 | 2010-08-15 18:52:59 +0200 | [diff] [blame] | 2683 | .llseek = noop_llseek, |
Marc Zyngier | 7ddfd3e | 2018-06-17 10:16:21 +0100 | [diff] [blame] | 2684 | KVM_COMPAT(kvm_vcpu_compat_ioctl), |
Avi Kivity | bccf215 | 2007-02-21 18:04:26 +0200 | [diff] [blame] | 2685 | }; |
| 2686 | |
| 2687 | /* |
| 2688 | * Allocates an inode for the vcpu. |
| 2689 | */ |
| 2690 | static int create_vcpu_fd(struct kvm_vcpu *vcpu) |
| 2691 | { |
Masatake YAMATO | e46b469 | 2018-01-20 04:04:22 +0900 | [diff] [blame] | 2692 | char name[8 + 1 + ITOA_MAX_LEN + 1]; |
| 2693 | |
| 2694 | snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id); |
| 2695 | return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC); |
Avi Kivity | bccf215 | 2007-02-21 18:04:26 +0200 | [diff] [blame] | 2696 | } |
| 2697 | |
Greg KH | 3e7093d | 2019-07-31 20:56:20 +0200 | [diff] [blame] | 2698 | static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu) |
Luiz Capitulino | 45b5939 | 2016-09-16 10:27:35 -0400 | [diff] [blame] | 2699 | { |
Paolo Bonzini | 741cbba | 2019-08-03 08:14:25 +0200 | [diff] [blame] | 2700 | #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS |
Luiz Capitulino | 45b5939 | 2016-09-16 10:27:35 -0400 | [diff] [blame] | 2701 | char dir_name[ITOA_MAX_LEN * 2]; |
Luiz Capitulino | 45b5939 | 2016-09-16 10:27:35 -0400 | [diff] [blame] | 2702 | |
Luiz Capitulino | 45b5939 | 2016-09-16 10:27:35 -0400 | [diff] [blame] | 2703 | if (!debugfs_initialized()) |
Greg KH | 3e7093d | 2019-07-31 20:56:20 +0200 | [diff] [blame] | 2704 | return; |
Luiz Capitulino | 45b5939 | 2016-09-16 10:27:35 -0400 | [diff] [blame] | 2705 | |
| 2706 | snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id); |
| 2707 | vcpu->debugfs_dentry = debugfs_create_dir(dir_name, |
Greg KH | 3e7093d | 2019-07-31 20:56:20 +0200 | [diff] [blame] | 2708 | vcpu->kvm->debugfs_dentry); |
Luiz Capitulino | 45b5939 | 2016-09-16 10:27:35 -0400 | [diff] [blame] | 2709 | |
Greg KH | 3e7093d | 2019-07-31 20:56:20 +0200 | [diff] [blame] | 2710 | kvm_arch_create_vcpu_debugfs(vcpu); |
Paolo Bonzini | 741cbba | 2019-08-03 08:14:25 +0200 | [diff] [blame] | 2711 | #endif |
Luiz Capitulino | 45b5939 | 2016-09-16 10:27:35 -0400 | [diff] [blame] | 2712 | } |
| 2713 | |
Avi Kivity | c5ea766 | 2007-02-20 18:41:05 +0200 | [diff] [blame] | 2714 | /* |
| 2715 | * Creates some virtual cpus. Good luck creating more than one. |
| 2716 | */ |
Gleb Natapov | 73880c8 | 2009-06-09 15:56:28 +0300 | [diff] [blame] | 2717 | static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id) |
Avi Kivity | c5ea766 | 2007-02-20 18:41:05 +0200 | [diff] [blame] | 2718 | { |
| 2719 | int r; |
David Hildenbrand | e09fefd | 2015-11-05 09:03:50 +0100 | [diff] [blame] | 2720 | struct kvm_vcpu *vcpu; |
Avi Kivity | c5ea766 | 2007-02-20 18:41:05 +0200 | [diff] [blame] | 2721 | |
Greg Kurz | 0b1b1df | 2016-05-09 18:13:37 +0200 | [diff] [blame] | 2722 | if (id >= KVM_MAX_VCPU_ID) |
Andy Honig | 338c7db | 2013-11-18 16:09:22 -0800 | [diff] [blame] | 2723 | return -EINVAL; |
| 2724 | |
Paolo Bonzini | 6c7caeb | 2016-06-13 14:48:25 +0200 | [diff] [blame] | 2725 | mutex_lock(&kvm->lock); |
| 2726 | if (kvm->created_vcpus == KVM_MAX_VCPUS) { |
| 2727 | mutex_unlock(&kvm->lock); |
| 2728 | return -EINVAL; |
| 2729 | } |
| 2730 | |
| 2731 | kvm->created_vcpus++; |
| 2732 | mutex_unlock(&kvm->lock); |
| 2733 | |
Gleb Natapov | 73880c8 | 2009-06-09 15:56:28 +0300 | [diff] [blame] | 2734 | vcpu = kvm_arch_vcpu_create(kvm, id); |
Paolo Bonzini | 6c7caeb | 2016-06-13 14:48:25 +0200 | [diff] [blame] | 2735 | if (IS_ERR(vcpu)) { |
| 2736 | r = PTR_ERR(vcpu); |
| 2737 | goto vcpu_decrement; |
| 2738 | } |
Avi Kivity | c5ea766 | 2007-02-20 18:41:05 +0200 | [diff] [blame] | 2739 | |
Avi Kivity | 15ad714 | 2007-07-11 18:17:21 +0300 | [diff] [blame] | 2740 | preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops); |
| 2741 | |
Avi Kivity | 26e5215 | 2007-11-20 15:30:24 +0200 | [diff] [blame] | 2742 | r = kvm_arch_vcpu_setup(vcpu); |
| 2743 | if (r) |
Jan Kiszka | d780592 | 2011-05-23 10:33:05 +0200 | [diff] [blame] | 2744 | goto vcpu_destroy; |
Avi Kivity | 26e5215 | 2007-11-20 15:30:24 +0200 | [diff] [blame] | 2745 | |
Greg KH | 3e7093d | 2019-07-31 20:56:20 +0200 | [diff] [blame] | 2746 | kvm_create_vcpu_debugfs(vcpu); |
Luiz Capitulino | 45b5939 | 2016-09-16 10:27:35 -0400 | [diff] [blame] | 2747 | |
Shaohua Li | 11ec280 | 2007-07-23 14:51:37 +0800 | [diff] [blame] | 2748 | mutex_lock(&kvm->lock); |
David Hildenbrand | e09fefd | 2015-11-05 09:03:50 +0100 | [diff] [blame] | 2749 | if (kvm_get_vcpu_by_id(kvm, id)) { |
| 2750 | r = -EEXIST; |
| 2751 | goto unlock_vcpu_destroy; |
| 2752 | } |
Gleb Natapov | 73880c8 | 2009-06-09 15:56:28 +0300 | [diff] [blame] | 2753 | |
Radim Krčmář | 8750e72 | 2019-11-07 07:53:42 -0500 | [diff] [blame] | 2754 | vcpu->vcpu_idx = atomic_read(&kvm->online_vcpus); |
| 2755 | BUG_ON(kvm->vcpus[vcpu->vcpu_idx]); |
Rusty Russell | fb3f0f5 | 2007-07-27 17:16:56 +1000 | [diff] [blame] | 2756 | |
| 2757 | /* Now it's all set up, let userspace reach it */ |
Al Viro | 66c0b39 | 2008-04-19 20:33:56 +0100 | [diff] [blame] | 2758 | kvm_get_kvm(kvm); |
Avi Kivity | bccf215 | 2007-02-21 18:04:26 +0200 | [diff] [blame] | 2759 | r = create_vcpu_fd(vcpu); |
Gleb Natapov | 73880c8 | 2009-06-09 15:56:28 +0300 | [diff] [blame] | 2760 | if (r < 0) { |
Sean Christopherson | 149487b | 2019-10-21 15:58:42 -0700 | [diff] [blame] | 2761 | kvm_put_kvm_no_destroy(kvm); |
Jan Kiszka | d780592 | 2011-05-23 10:33:05 +0200 | [diff] [blame] | 2762 | goto unlock_vcpu_destroy; |
Gleb Natapov | 73880c8 | 2009-06-09 15:56:28 +0300 | [diff] [blame] | 2763 | } |
| 2764 | |
Radim Krčmář | 8750e72 | 2019-11-07 07:53:42 -0500 | [diff] [blame] | 2765 | kvm->vcpus[vcpu->vcpu_idx] = vcpu; |
Paolo Bonzini | dd48924 | 2015-07-29 11:32:20 +0200 | [diff] [blame] | 2766 | |
| 2767 | /* |
| 2768 | * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus |
| 2769 | * before kvm->online_vcpu's incremented value. |
| 2770 | */ |
Gleb Natapov | 73880c8 | 2009-06-09 15:56:28 +0300 | [diff] [blame] | 2771 | smp_wmb(); |
| 2772 | atomic_inc(&kvm->online_vcpus); |
| 2773 | |
Gleb Natapov | 73880c8 | 2009-06-09 15:56:28 +0300 | [diff] [blame] | 2774 | mutex_unlock(&kvm->lock); |
Marcelo Tosatti | 42897d8 | 2012-11-27 23:29:02 -0200 | [diff] [blame] | 2775 | kvm_arch_vcpu_postcreate(vcpu); |
Avi Kivity | bccf215 | 2007-02-21 18:04:26 +0200 | [diff] [blame] | 2776 | return r; |
Avi Kivity | c5ea766 | 2007-02-20 18:41:05 +0200 | [diff] [blame] | 2777 | |
Jan Kiszka | d780592 | 2011-05-23 10:33:05 +0200 | [diff] [blame] | 2778 | unlock_vcpu_destroy: |
Glauber Costa | 7d8fece | 2008-09-17 23:16:59 -0300 | [diff] [blame] | 2779 | mutex_unlock(&kvm->lock); |
Luiz Capitulino | 45b5939 | 2016-09-16 10:27:35 -0400 | [diff] [blame] | 2780 | debugfs_remove_recursive(vcpu->debugfs_dentry); |
Jan Kiszka | d780592 | 2011-05-23 10:33:05 +0200 | [diff] [blame] | 2781 | vcpu_destroy: |
Hollis Blanchard | d40ccc6 | 2007-11-19 14:04:43 -0600 | [diff] [blame] | 2782 | kvm_arch_vcpu_destroy(vcpu); |
Paolo Bonzini | 6c7caeb | 2016-06-13 14:48:25 +0200 | [diff] [blame] | 2783 | vcpu_decrement: |
| 2784 | mutex_lock(&kvm->lock); |
| 2785 | kvm->created_vcpus--; |
| 2786 | mutex_unlock(&kvm->lock); |
Avi Kivity | c5ea766 | 2007-02-20 18:41:05 +0200 | [diff] [blame] | 2787 | return r; |
| 2788 | } |
| 2789 | |
Avi Kivity | 1961d27 | 2007-03-05 19:46:05 +0200 | [diff] [blame] | 2790 | static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset) |
| 2791 | { |
| 2792 | if (sigset) { |
| 2793 | sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP)); |
| 2794 | vcpu->sigset_active = 1; |
| 2795 | vcpu->sigset = *sigset; |
| 2796 | } else |
| 2797 | vcpu->sigset_active = 0; |
| 2798 | return 0; |
| 2799 | } |
| 2800 | |
Avi Kivity | bccf215 | 2007-02-21 18:04:26 +0200 | [diff] [blame] | 2801 | static long kvm_vcpu_ioctl(struct file *filp, |
| 2802 | unsigned int ioctl, unsigned long arg) |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 2803 | { |
Avi Kivity | bccf215 | 2007-02-21 18:04:26 +0200 | [diff] [blame] | 2804 | struct kvm_vcpu *vcpu = filp->private_data; |
Al Viro | 2f366987 | 2007-02-09 16:38:35 +0000 | [diff] [blame] | 2805 | void __user *argp = (void __user *)arg; |
Carsten Otte | 313a3dc | 2007-10-11 19:16:52 +0200 | [diff] [blame] | 2806 | int r; |
Dave Hansen | fa3795a | 2008-08-11 10:01:46 -0700 | [diff] [blame] | 2807 | struct kvm_fpu *fpu = NULL; |
| 2808 | struct kvm_sregs *kvm_sregs = NULL; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 2809 | |
Avi Kivity | 6d4e4c4 | 2007-11-21 16:41:05 +0200 | [diff] [blame] | 2810 | if (vcpu->kvm->mm != current->mm) |
| 2811 | return -EIO; |
Avi Kivity | 2122ff5 | 2010-05-13 11:25:04 +0300 | [diff] [blame] | 2812 | |
David Matlack | 2ea75be | 2014-09-19 16:03:25 -0700 | [diff] [blame] | 2813 | if (unlikely(_IOC_TYPE(ioctl) != KVMIO)) |
| 2814 | return -EINVAL; |
| 2815 | |
Avi Kivity | 2122ff5 | 2010-05-13 11:25:04 +0300 | [diff] [blame] | 2816 | /* |
Paolo Bonzini | 5cb0944 | 2017-12-12 17:41:34 +0100 | [diff] [blame] | 2817 | * Some architectures have vcpu ioctls that are asynchronous to vcpu |
| 2818 | * execution; mutex_lock() would break them. |
Avi Kivity | 2122ff5 | 2010-05-13 11:25:04 +0300 | [diff] [blame] | 2819 | */ |
Paolo Bonzini | 5cb0944 | 2017-12-12 17:41:34 +0100 | [diff] [blame] | 2820 | r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg); |
| 2821 | if (r != -ENOIOCTLCMD) |
Michael S. Tsirkin | 9fc7744 | 2012-09-16 11:50:30 +0300 | [diff] [blame] | 2822 | return r; |
Avi Kivity | 2122ff5 | 2010-05-13 11:25:04 +0300 | [diff] [blame] | 2823 | |
Christoffer Dall | ec7660c | 2017-12-04 21:35:23 +0100 | [diff] [blame] | 2824 | if (mutex_lock_killable(&vcpu->mutex)) |
| 2825 | return -EINTR; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 2826 | switch (ioctl) { |
Christian Borntraeger | 0e4524a | 2017-07-06 14:44:28 +0200 | [diff] [blame] | 2827 | case KVM_RUN: { |
| 2828 | struct pid *oldpid; |
Avi Kivity | f0fe510 | 2007-03-07 13:11:17 +0200 | [diff] [blame] | 2829 | r = -EINVAL; |
| 2830 | if (arg) |
| 2831 | goto out; |
Christian Borntraeger | 0e4524a | 2017-07-06 14:44:28 +0200 | [diff] [blame] | 2832 | oldpid = rcu_access_pointer(vcpu->pid); |
Eric W. Biederman | 71dbc8a | 2017-07-16 21:39:32 -0500 | [diff] [blame] | 2833 | if (unlikely(oldpid != task_pid(current))) { |
Christian Borntraeger | 7a72f7a | 2014-08-05 16:44:14 +0200 | [diff] [blame] | 2834 | /* The thread running this VCPU changed. */ |
Christoffer Dall | bd2a639 | 2018-02-23 17:23:57 +0100 | [diff] [blame] | 2835 | struct pid *newpid; |
Xiubo Li | f95ef0cd | 2015-02-26 14:58:23 +0800 | [diff] [blame] | 2836 | |
Christoffer Dall | bd2a639 | 2018-02-23 17:23:57 +0100 | [diff] [blame] | 2837 | r = kvm_arch_vcpu_run_pid_change(vcpu); |
| 2838 | if (r) |
| 2839 | break; |
| 2840 | |
| 2841 | newpid = get_task_pid(current, PIDTYPE_PID); |
Christian Borntraeger | 7a72f7a | 2014-08-05 16:44:14 +0200 | [diff] [blame] | 2842 | rcu_assign_pointer(vcpu->pid, newpid); |
| 2843 | if (oldpid) |
| 2844 | synchronize_rcu(); |
| 2845 | put_pid(oldpid); |
| 2846 | } |
Hollis Blanchard | b6c7a5d | 2007-11-01 14:16:10 -0500 | [diff] [blame] | 2847 | r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run); |
Gleb Natapov | 64be500 | 2010-10-24 16:49:08 +0200 | [diff] [blame] | 2848 | trace_kvm_userspace_exit(vcpu->run->exit_reason, r); |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 2849 | break; |
Christian Borntraeger | 0e4524a | 2017-07-06 14:44:28 +0200 | [diff] [blame] | 2850 | } |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 2851 | case KVM_GET_REGS: { |
Xiantao Zhang | 3e4bb3a | 2008-02-25 18:52:20 +0800 | [diff] [blame] | 2852 | struct kvm_regs *kvm_regs; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 2853 | |
Xiantao Zhang | 3e4bb3a | 2008-02-25 18:52:20 +0800 | [diff] [blame] | 2854 | r = -ENOMEM; |
Ben Gardon | b12ce36 | 2019-02-11 11:02:49 -0800 | [diff] [blame] | 2855 | kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT); |
Xiantao Zhang | 3e4bb3a | 2008-02-25 18:52:20 +0800 | [diff] [blame] | 2856 | if (!kvm_regs) |
| 2857 | goto out; |
| 2858 | r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs); |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 2859 | if (r) |
Xiantao Zhang | 3e4bb3a | 2008-02-25 18:52:20 +0800 | [diff] [blame] | 2860 | goto out_free1; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 2861 | r = -EFAULT; |
Xiantao Zhang | 3e4bb3a | 2008-02-25 18:52:20 +0800 | [diff] [blame] | 2862 | if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs))) |
| 2863 | goto out_free1; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 2864 | r = 0; |
Xiantao Zhang | 3e4bb3a | 2008-02-25 18:52:20 +0800 | [diff] [blame] | 2865 | out_free1: |
| 2866 | kfree(kvm_regs); |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 2867 | break; |
| 2868 | } |
| 2869 | case KVM_SET_REGS: { |
Xiantao Zhang | 3e4bb3a | 2008-02-25 18:52:20 +0800 | [diff] [blame] | 2870 | struct kvm_regs *kvm_regs; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 2871 | |
Xiantao Zhang | 3e4bb3a | 2008-02-25 18:52:20 +0800 | [diff] [blame] | 2872 | r = -ENOMEM; |
Sasha Levin | ff5c2c0 | 2011-12-04 19:36:29 +0200 | [diff] [blame] | 2873 | kvm_regs = memdup_user(argp, sizeof(*kvm_regs)); |
| 2874 | if (IS_ERR(kvm_regs)) { |
| 2875 | r = PTR_ERR(kvm_regs); |
Xiantao Zhang | 3e4bb3a | 2008-02-25 18:52:20 +0800 | [diff] [blame] | 2876 | goto out; |
Sasha Levin | ff5c2c0 | 2011-12-04 19:36:29 +0200 | [diff] [blame] | 2877 | } |
Xiantao Zhang | 3e4bb3a | 2008-02-25 18:52:20 +0800 | [diff] [blame] | 2878 | r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs); |
Xiantao Zhang | 3e4bb3a | 2008-02-25 18:52:20 +0800 | [diff] [blame] | 2879 | kfree(kvm_regs); |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 2880 | break; |
| 2881 | } |
| 2882 | case KVM_GET_SREGS: { |
Ben Gardon | b12ce36 | 2019-02-11 11:02:49 -0800 | [diff] [blame] | 2883 | kvm_sregs = kzalloc(sizeof(struct kvm_sregs), |
| 2884 | GFP_KERNEL_ACCOUNT); |
Dave Hansen | fa3795a | 2008-08-11 10:01:46 -0700 | [diff] [blame] | 2885 | r = -ENOMEM; |
| 2886 | if (!kvm_sregs) |
| 2887 | goto out; |
| 2888 | r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs); |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 2889 | if (r) |
| 2890 | goto out; |
| 2891 | r = -EFAULT; |
Dave Hansen | fa3795a | 2008-08-11 10:01:46 -0700 | [diff] [blame] | 2892 | if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs))) |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 2893 | goto out; |
| 2894 | r = 0; |
| 2895 | break; |
| 2896 | } |
| 2897 | case KVM_SET_SREGS: { |
Sasha Levin | ff5c2c0 | 2011-12-04 19:36:29 +0200 | [diff] [blame] | 2898 | kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs)); |
| 2899 | if (IS_ERR(kvm_sregs)) { |
| 2900 | r = PTR_ERR(kvm_sregs); |
Guo Chao | 1859541 | 2012-11-02 18:33:21 +0800 | [diff] [blame] | 2901 | kvm_sregs = NULL; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 2902 | goto out; |
Sasha Levin | ff5c2c0 | 2011-12-04 19:36:29 +0200 | [diff] [blame] | 2903 | } |
Dave Hansen | fa3795a | 2008-08-11 10:01:46 -0700 | [diff] [blame] | 2904 | r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs); |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 2905 | break; |
| 2906 | } |
Marcelo Tosatti | 62d9f0d | 2008-04-11 13:24:45 -0300 | [diff] [blame] | 2907 | case KVM_GET_MP_STATE: { |
| 2908 | struct kvm_mp_state mp_state; |
| 2909 | |
| 2910 | r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state); |
| 2911 | if (r) |
| 2912 | goto out; |
| 2913 | r = -EFAULT; |
Xiubo Li | 893bdbf | 2015-02-26 14:58:19 +0800 | [diff] [blame] | 2914 | if (copy_to_user(argp, &mp_state, sizeof(mp_state))) |
Marcelo Tosatti | 62d9f0d | 2008-04-11 13:24:45 -0300 | [diff] [blame] | 2915 | goto out; |
| 2916 | r = 0; |
| 2917 | break; |
| 2918 | } |
| 2919 | case KVM_SET_MP_STATE: { |
| 2920 | struct kvm_mp_state mp_state; |
| 2921 | |
| 2922 | r = -EFAULT; |
Xiubo Li | 893bdbf | 2015-02-26 14:58:19 +0800 | [diff] [blame] | 2923 | if (copy_from_user(&mp_state, argp, sizeof(mp_state))) |
Marcelo Tosatti | 62d9f0d | 2008-04-11 13:24:45 -0300 | [diff] [blame] | 2924 | goto out; |
| 2925 | r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state); |
Marcelo Tosatti | 62d9f0d | 2008-04-11 13:24:45 -0300 | [diff] [blame] | 2926 | break; |
| 2927 | } |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 2928 | case KVM_TRANSLATE: { |
| 2929 | struct kvm_translation tr; |
| 2930 | |
| 2931 | r = -EFAULT; |
Xiubo Li | 893bdbf | 2015-02-26 14:58:19 +0800 | [diff] [blame] | 2932 | if (copy_from_user(&tr, argp, sizeof(tr))) |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 2933 | goto out; |
Zhang Xiantao | 8b0067913 | 2007-11-16 13:05:55 +0800 | [diff] [blame] | 2934 | r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr); |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 2935 | if (r) |
| 2936 | goto out; |
| 2937 | r = -EFAULT; |
Xiubo Li | 893bdbf | 2015-02-26 14:58:19 +0800 | [diff] [blame] | 2938 | if (copy_to_user(argp, &tr, sizeof(tr))) |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 2939 | goto out; |
| 2940 | r = 0; |
| 2941 | break; |
| 2942 | } |
Jan Kiszka | d0bfb94 | 2008-12-15 13:52:10 +0100 | [diff] [blame] | 2943 | case KVM_SET_GUEST_DEBUG: { |
| 2944 | struct kvm_guest_debug dbg; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 2945 | |
| 2946 | r = -EFAULT; |
Xiubo Li | 893bdbf | 2015-02-26 14:58:19 +0800 | [diff] [blame] | 2947 | if (copy_from_user(&dbg, argp, sizeof(dbg))) |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 2948 | goto out; |
Jan Kiszka | d0bfb94 | 2008-12-15 13:52:10 +0100 | [diff] [blame] | 2949 | r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg); |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 2950 | break; |
| 2951 | } |
Avi Kivity | 1961d27 | 2007-03-05 19:46:05 +0200 | [diff] [blame] | 2952 | case KVM_SET_SIGNAL_MASK: { |
| 2953 | struct kvm_signal_mask __user *sigmask_arg = argp; |
| 2954 | struct kvm_signal_mask kvm_sigmask; |
| 2955 | sigset_t sigset, *p; |
| 2956 | |
| 2957 | p = NULL; |
| 2958 | if (argp) { |
| 2959 | r = -EFAULT; |
| 2960 | if (copy_from_user(&kvm_sigmask, argp, |
Xiubo Li | 893bdbf | 2015-02-26 14:58:19 +0800 | [diff] [blame] | 2961 | sizeof(kvm_sigmask))) |
Avi Kivity | 1961d27 | 2007-03-05 19:46:05 +0200 | [diff] [blame] | 2962 | goto out; |
| 2963 | r = -EINVAL; |
Xiubo Li | 893bdbf | 2015-02-26 14:58:19 +0800 | [diff] [blame] | 2964 | if (kvm_sigmask.len != sizeof(sigset)) |
Avi Kivity | 1961d27 | 2007-03-05 19:46:05 +0200 | [diff] [blame] | 2965 | goto out; |
| 2966 | r = -EFAULT; |
| 2967 | if (copy_from_user(&sigset, sigmask_arg->sigset, |
Xiubo Li | 893bdbf | 2015-02-26 14:58:19 +0800 | [diff] [blame] | 2968 | sizeof(sigset))) |
Avi Kivity | 1961d27 | 2007-03-05 19:46:05 +0200 | [diff] [blame] | 2969 | goto out; |
| 2970 | p = &sigset; |
| 2971 | } |
Andi Kleen | 376d41f | 2010-06-10 13:10:47 +0200 | [diff] [blame] | 2972 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, p); |
Avi Kivity | 1961d27 | 2007-03-05 19:46:05 +0200 | [diff] [blame] | 2973 | break; |
| 2974 | } |
Avi Kivity | b883673 | 2007-04-01 16:34:31 +0300 | [diff] [blame] | 2975 | case KVM_GET_FPU: { |
Ben Gardon | b12ce36 | 2019-02-11 11:02:49 -0800 | [diff] [blame] | 2976 | fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT); |
Dave Hansen | fa3795a | 2008-08-11 10:01:46 -0700 | [diff] [blame] | 2977 | r = -ENOMEM; |
| 2978 | if (!fpu) |
| 2979 | goto out; |
| 2980 | r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu); |
Avi Kivity | b883673 | 2007-04-01 16:34:31 +0300 | [diff] [blame] | 2981 | if (r) |
| 2982 | goto out; |
| 2983 | r = -EFAULT; |
Dave Hansen | fa3795a | 2008-08-11 10:01:46 -0700 | [diff] [blame] | 2984 | if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu))) |
Avi Kivity | b883673 | 2007-04-01 16:34:31 +0300 | [diff] [blame] | 2985 | goto out; |
| 2986 | r = 0; |
| 2987 | break; |
| 2988 | } |
| 2989 | case KVM_SET_FPU: { |
Sasha Levin | ff5c2c0 | 2011-12-04 19:36:29 +0200 | [diff] [blame] | 2990 | fpu = memdup_user(argp, sizeof(*fpu)); |
| 2991 | if (IS_ERR(fpu)) { |
| 2992 | r = PTR_ERR(fpu); |
Guo Chao | 1859541 | 2012-11-02 18:33:21 +0800 | [diff] [blame] | 2993 | fpu = NULL; |
Avi Kivity | b883673 | 2007-04-01 16:34:31 +0300 | [diff] [blame] | 2994 | goto out; |
Sasha Levin | ff5c2c0 | 2011-12-04 19:36:29 +0200 | [diff] [blame] | 2995 | } |
Dave Hansen | fa3795a | 2008-08-11 10:01:46 -0700 | [diff] [blame] | 2996 | r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu); |
Avi Kivity | b883673 | 2007-04-01 16:34:31 +0300 | [diff] [blame] | 2997 | break; |
| 2998 | } |
Avi Kivity | bccf215 | 2007-02-21 18:04:26 +0200 | [diff] [blame] | 2999 | default: |
Carsten Otte | 313a3dc | 2007-10-11 19:16:52 +0200 | [diff] [blame] | 3000 | r = kvm_arch_vcpu_ioctl(filp, ioctl, arg); |
Avi Kivity | bccf215 | 2007-02-21 18:04:26 +0200 | [diff] [blame] | 3001 | } |
| 3002 | out: |
Christoffer Dall | ec7660c | 2017-12-04 21:35:23 +0100 | [diff] [blame] | 3003 | mutex_unlock(&vcpu->mutex); |
Dave Hansen | fa3795a | 2008-08-11 10:01:46 -0700 | [diff] [blame] | 3004 | kfree(fpu); |
| 3005 | kfree(kvm_sregs); |
Avi Kivity | bccf215 | 2007-02-21 18:04:26 +0200 | [diff] [blame] | 3006 | return r; |
| 3007 | } |
| 3008 | |
Christian Borntraeger | de8e5d7 | 2015-02-03 09:35:15 +0100 | [diff] [blame] | 3009 | #ifdef CONFIG_KVM_COMPAT |
Alexander Graf | 1dda606 | 2011-06-08 02:45:37 +0200 | [diff] [blame] | 3010 | static long kvm_vcpu_compat_ioctl(struct file *filp, |
| 3011 | unsigned int ioctl, unsigned long arg) |
| 3012 | { |
| 3013 | struct kvm_vcpu *vcpu = filp->private_data; |
| 3014 | void __user *argp = compat_ptr(arg); |
| 3015 | int r; |
| 3016 | |
| 3017 | if (vcpu->kvm->mm != current->mm) |
| 3018 | return -EIO; |
| 3019 | |
| 3020 | switch (ioctl) { |
| 3021 | case KVM_SET_SIGNAL_MASK: { |
| 3022 | struct kvm_signal_mask __user *sigmask_arg = argp; |
| 3023 | struct kvm_signal_mask kvm_sigmask; |
Alexander Graf | 1dda606 | 2011-06-08 02:45:37 +0200 | [diff] [blame] | 3024 | sigset_t sigset; |
| 3025 | |
| 3026 | if (argp) { |
| 3027 | r = -EFAULT; |
| 3028 | if (copy_from_user(&kvm_sigmask, argp, |
Xiubo Li | 893bdbf | 2015-02-26 14:58:19 +0800 | [diff] [blame] | 3029 | sizeof(kvm_sigmask))) |
Alexander Graf | 1dda606 | 2011-06-08 02:45:37 +0200 | [diff] [blame] | 3030 | goto out; |
| 3031 | r = -EINVAL; |
Al Viro | 3968cf6 | 2017-09-03 21:45:17 -0400 | [diff] [blame] | 3032 | if (kvm_sigmask.len != sizeof(compat_sigset_t)) |
Alexander Graf | 1dda606 | 2011-06-08 02:45:37 +0200 | [diff] [blame] | 3033 | goto out; |
| 3034 | r = -EFAULT; |
Al Viro | 3968cf6 | 2017-09-03 21:45:17 -0400 | [diff] [blame] | 3035 | if (get_compat_sigset(&sigset, (void *)sigmask_arg->sigset)) |
Alexander Graf | 1dda606 | 2011-06-08 02:45:37 +0200 | [diff] [blame] | 3036 | goto out; |
Alan Cox | 760a9a3 | 2012-08-22 14:34:11 +0100 | [diff] [blame] | 3037 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset); |
| 3038 | } else |
| 3039 | r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL); |
Alexander Graf | 1dda606 | 2011-06-08 02:45:37 +0200 | [diff] [blame] | 3040 | break; |
| 3041 | } |
| 3042 | default: |
| 3043 | r = kvm_vcpu_ioctl(filp, ioctl, arg); |
| 3044 | } |
| 3045 | |
| 3046 | out: |
| 3047 | return r; |
| 3048 | } |
| 3049 | #endif |
| 3050 | |
Cédric Le Goater | a1cd3f0 | 2019-04-18 12:39:36 +0200 | [diff] [blame] | 3051 | static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma) |
| 3052 | { |
| 3053 | struct kvm_device *dev = filp->private_data; |
| 3054 | |
| 3055 | if (dev->ops->mmap) |
| 3056 | return dev->ops->mmap(dev, vma); |
| 3057 | |
| 3058 | return -ENODEV; |
| 3059 | } |
| 3060 | |
Scott Wood | 852b6d5 | 2013-04-12 14:08:42 +0000 | [diff] [blame] | 3061 | static int kvm_device_ioctl_attr(struct kvm_device *dev, |
| 3062 | int (*accessor)(struct kvm_device *dev, |
| 3063 | struct kvm_device_attr *attr), |
| 3064 | unsigned long arg) |
| 3065 | { |
| 3066 | struct kvm_device_attr attr; |
| 3067 | |
| 3068 | if (!accessor) |
| 3069 | return -EPERM; |
| 3070 | |
| 3071 | if (copy_from_user(&attr, (void __user *)arg, sizeof(attr))) |
| 3072 | return -EFAULT; |
| 3073 | |
| 3074 | return accessor(dev, &attr); |
| 3075 | } |
| 3076 | |
| 3077 | static long kvm_device_ioctl(struct file *filp, unsigned int ioctl, |
| 3078 | unsigned long arg) |
| 3079 | { |
| 3080 | struct kvm_device *dev = filp->private_data; |
| 3081 | |
Sean Christopherson | ddba918 | 2019-02-15 12:48:39 -0800 | [diff] [blame] | 3082 | if (dev->kvm->mm != current->mm) |
| 3083 | return -EIO; |
| 3084 | |
Scott Wood | 852b6d5 | 2013-04-12 14:08:42 +0000 | [diff] [blame] | 3085 | switch (ioctl) { |
| 3086 | case KVM_SET_DEVICE_ATTR: |
| 3087 | return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg); |
| 3088 | case KVM_GET_DEVICE_ATTR: |
| 3089 | return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg); |
| 3090 | case KVM_HAS_DEVICE_ATTR: |
| 3091 | return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg); |
| 3092 | default: |
| 3093 | if (dev->ops->ioctl) |
| 3094 | return dev->ops->ioctl(dev, ioctl, arg); |
| 3095 | |
| 3096 | return -ENOTTY; |
| 3097 | } |
| 3098 | } |
| 3099 | |
Scott Wood | 852b6d5 | 2013-04-12 14:08:42 +0000 | [diff] [blame] | 3100 | static int kvm_device_release(struct inode *inode, struct file *filp) |
| 3101 | { |
| 3102 | struct kvm_device *dev = filp->private_data; |
| 3103 | struct kvm *kvm = dev->kvm; |
| 3104 | |
Cédric Le Goater | 2bde9b3 | 2019-04-18 12:39:41 +0200 | [diff] [blame] | 3105 | if (dev->ops->release) { |
| 3106 | mutex_lock(&kvm->lock); |
| 3107 | list_del(&dev->vm_node); |
| 3108 | dev->ops->release(dev); |
| 3109 | mutex_unlock(&kvm->lock); |
| 3110 | } |
| 3111 | |
Scott Wood | 852b6d5 | 2013-04-12 14:08:42 +0000 | [diff] [blame] | 3112 | kvm_put_kvm(kvm); |
| 3113 | return 0; |
| 3114 | } |
| 3115 | |
| 3116 | static const struct file_operations kvm_device_fops = { |
| 3117 | .unlocked_ioctl = kvm_device_ioctl, |
| 3118 | .release = kvm_device_release, |
Marc Zyngier | 7ddfd3e | 2018-06-17 10:16:21 +0100 | [diff] [blame] | 3119 | KVM_COMPAT(kvm_device_ioctl), |
Cédric Le Goater | a1cd3f0 | 2019-04-18 12:39:36 +0200 | [diff] [blame] | 3120 | .mmap = kvm_device_mmap, |
Scott Wood | 852b6d5 | 2013-04-12 14:08:42 +0000 | [diff] [blame] | 3121 | }; |
| 3122 | |
| 3123 | struct kvm_device *kvm_device_from_filp(struct file *filp) |
| 3124 | { |
| 3125 | if (filp->f_op != &kvm_device_fops) |
| 3126 | return NULL; |
| 3127 | |
| 3128 | return filp->private_data; |
| 3129 | } |
| 3130 | |
Steven Price | 8538cb2 | 2019-10-21 16:28:19 +0100 | [diff] [blame] | 3131 | static const struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = { |
Will Deacon | d60eacb | 2014-09-02 10:27:33 +0100 | [diff] [blame] | 3132 | #ifdef CONFIG_KVM_MPIC |
| 3133 | [KVM_DEV_TYPE_FSL_MPIC_20] = &kvm_mpic_ops, |
| 3134 | [KVM_DEV_TYPE_FSL_MPIC_42] = &kvm_mpic_ops, |
| 3135 | #endif |
Will Deacon | d60eacb | 2014-09-02 10:27:33 +0100 | [diff] [blame] | 3136 | }; |
| 3137 | |
Steven Price | 8538cb2 | 2019-10-21 16:28:19 +0100 | [diff] [blame] | 3138 | int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type) |
Will Deacon | d60eacb | 2014-09-02 10:27:33 +0100 | [diff] [blame] | 3139 | { |
| 3140 | if (type >= ARRAY_SIZE(kvm_device_ops_table)) |
| 3141 | return -ENOSPC; |
| 3142 | |
| 3143 | if (kvm_device_ops_table[type] != NULL) |
| 3144 | return -EEXIST; |
| 3145 | |
| 3146 | kvm_device_ops_table[type] = ops; |
| 3147 | return 0; |
| 3148 | } |
| 3149 | |
Wanpeng Li | 571ee1b | 2014-10-09 18:30:08 +0800 | [diff] [blame] | 3150 | void kvm_unregister_device_ops(u32 type) |
| 3151 | { |
| 3152 | if (kvm_device_ops_table[type] != NULL) |
| 3153 | kvm_device_ops_table[type] = NULL; |
| 3154 | } |
| 3155 | |
Scott Wood | 852b6d5 | 2013-04-12 14:08:42 +0000 | [diff] [blame] | 3156 | static int kvm_ioctl_create_device(struct kvm *kvm, |
| 3157 | struct kvm_create_device *cd) |
| 3158 | { |
Steven Price | 8538cb2 | 2019-10-21 16:28:19 +0100 | [diff] [blame] | 3159 | const struct kvm_device_ops *ops = NULL; |
Scott Wood | 852b6d5 | 2013-04-12 14:08:42 +0000 | [diff] [blame] | 3160 | struct kvm_device *dev; |
| 3161 | bool test = cd->flags & KVM_CREATE_DEVICE_TEST; |
Paolo Bonzini | 1d487e9 | 2019-04-11 11:16:47 +0200 | [diff] [blame] | 3162 | int type; |
Scott Wood | 852b6d5 | 2013-04-12 14:08:42 +0000 | [diff] [blame] | 3163 | int ret; |
| 3164 | |
Will Deacon | d60eacb | 2014-09-02 10:27:33 +0100 | [diff] [blame] | 3165 | if (cd->type >= ARRAY_SIZE(kvm_device_ops_table)) |
Scott Wood | 852b6d5 | 2013-04-12 14:08:42 +0000 | [diff] [blame] | 3166 | return -ENODEV; |
Will Deacon | d60eacb | 2014-09-02 10:27:33 +0100 | [diff] [blame] | 3167 | |
Paolo Bonzini | 1d487e9 | 2019-04-11 11:16:47 +0200 | [diff] [blame] | 3168 | type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table)); |
| 3169 | ops = kvm_device_ops_table[type]; |
Will Deacon | d60eacb | 2014-09-02 10:27:33 +0100 | [diff] [blame] | 3170 | if (ops == NULL) |
| 3171 | return -ENODEV; |
Scott Wood | 852b6d5 | 2013-04-12 14:08:42 +0000 | [diff] [blame] | 3172 | |
| 3173 | if (test) |
| 3174 | return 0; |
| 3175 | |
Ben Gardon | b12ce36 | 2019-02-11 11:02:49 -0800 | [diff] [blame] | 3176 | dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT); |
Scott Wood | 852b6d5 | 2013-04-12 14:08:42 +0000 | [diff] [blame] | 3177 | if (!dev) |
| 3178 | return -ENOMEM; |
| 3179 | |
| 3180 | dev->ops = ops; |
| 3181 | dev->kvm = kvm; |
Scott Wood | 852b6d5 | 2013-04-12 14:08:42 +0000 | [diff] [blame] | 3182 | |
Christoffer Dall | a28ebea | 2016-08-09 19:13:01 +0200 | [diff] [blame] | 3183 | mutex_lock(&kvm->lock); |
Paolo Bonzini | 1d487e9 | 2019-04-11 11:16:47 +0200 | [diff] [blame] | 3184 | ret = ops->create(dev, type); |
Scott Wood | 852b6d5 | 2013-04-12 14:08:42 +0000 | [diff] [blame] | 3185 | if (ret < 0) { |
Christoffer Dall | a28ebea | 2016-08-09 19:13:01 +0200 | [diff] [blame] | 3186 | mutex_unlock(&kvm->lock); |
Scott Wood | 852b6d5 | 2013-04-12 14:08:42 +0000 | [diff] [blame] | 3187 | kfree(dev); |
| 3188 | return ret; |
| 3189 | } |
Christoffer Dall | a28ebea | 2016-08-09 19:13:01 +0200 | [diff] [blame] | 3190 | list_add(&dev->vm_node, &kvm->devices); |
| 3191 | mutex_unlock(&kvm->lock); |
Scott Wood | 852b6d5 | 2013-04-12 14:08:42 +0000 | [diff] [blame] | 3192 | |
Christoffer Dall | 023e9fd | 2016-08-09 19:13:00 +0200 | [diff] [blame] | 3193 | if (ops->init) |
| 3194 | ops->init(dev); |
| 3195 | |
Jann Horn | cfa3938 | 2019-01-26 01:54:33 +0100 | [diff] [blame] | 3196 | kvm_get_kvm(kvm); |
Yann Droneaud | 24009b0 | 2013-08-24 22:14:07 +0200 | [diff] [blame] | 3197 | ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC); |
Scott Wood | 852b6d5 | 2013-04-12 14:08:42 +0000 | [diff] [blame] | 3198 | if (ret < 0) { |
Sean Christopherson | 149487b | 2019-10-21 15:58:42 -0700 | [diff] [blame] | 3199 | kvm_put_kvm_no_destroy(kvm); |
Christoffer Dall | a28ebea | 2016-08-09 19:13:01 +0200 | [diff] [blame] | 3200 | mutex_lock(&kvm->lock); |
| 3201 | list_del(&dev->vm_node); |
| 3202 | mutex_unlock(&kvm->lock); |
Dan Carpenter | a0f1d21 | 2016-11-30 22:21:05 +0300 | [diff] [blame] | 3203 | ops->destroy(dev); |
Scott Wood | 852b6d5 | 2013-04-12 14:08:42 +0000 | [diff] [blame] | 3204 | return ret; |
| 3205 | } |
| 3206 | |
Scott Wood | 852b6d5 | 2013-04-12 14:08:42 +0000 | [diff] [blame] | 3207 | cd->fd = ret; |
| 3208 | return 0; |
| 3209 | } |
| 3210 | |
Alexander Graf | 92b591a | 2014-07-14 18:33:08 +0200 | [diff] [blame] | 3211 | static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg) |
| 3212 | { |
| 3213 | switch (arg) { |
| 3214 | case KVM_CAP_USER_MEMORY: |
| 3215 | case KVM_CAP_DESTROY_MEMORY_REGION_WORKS: |
| 3216 | case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS: |
Alexander Graf | 92b591a | 2014-07-14 18:33:08 +0200 | [diff] [blame] | 3217 | case KVM_CAP_INTERNAL_ERROR_DATA: |
| 3218 | #ifdef CONFIG_HAVE_KVM_MSI |
| 3219 | case KVM_CAP_SIGNAL_MSI: |
| 3220 | #endif |
Paul Mackerras | 297e210 | 2014-06-30 20:51:13 +1000 | [diff] [blame] | 3221 | #ifdef CONFIG_HAVE_KVM_IRQFD |
Paolo Bonzini | dc9be0f | 2015-03-05 11:54:46 +0100 | [diff] [blame] | 3222 | case KVM_CAP_IRQFD: |
Alexander Graf | 92b591a | 2014-07-14 18:33:08 +0200 | [diff] [blame] | 3223 | case KVM_CAP_IRQFD_RESAMPLE: |
| 3224 | #endif |
Jason Wang | e9ea506 | 2015-09-15 14:41:59 +0800 | [diff] [blame] | 3225 | case KVM_CAP_IOEVENTFD_ANY_LENGTH: |
Alexander Graf | 92b591a | 2014-07-14 18:33:08 +0200 | [diff] [blame] | 3226 | case KVM_CAP_CHECK_EXTENSION_VM: |
Paolo Bonzini | e5d83c7 | 2017-02-16 10:40:56 +0100 | [diff] [blame] | 3227 | case KVM_CAP_ENABLE_CAP_VM: |
Paolo Bonzini | 2a31b9d | 2018-10-23 02:36:47 +0200 | [diff] [blame] | 3228 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
Peter Xu | d7547c5 | 2019-05-08 17:15:47 +0800 | [diff] [blame] | 3229 | case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: |
Paolo Bonzini | 2a31b9d | 2018-10-23 02:36:47 +0200 | [diff] [blame] | 3230 | #endif |
Alexander Graf | 92b591a | 2014-07-14 18:33:08 +0200 | [diff] [blame] | 3231 | return 1; |
Paolo Bonzini | 4b4357e | 2017-03-31 13:53:23 +0200 | [diff] [blame] | 3232 | #ifdef CONFIG_KVM_MMIO |
Paolo Bonzini | 3042255 | 2017-03-31 13:53:22 +0200 | [diff] [blame] | 3233 | case KVM_CAP_COALESCED_MMIO: |
| 3234 | return KVM_COALESCED_MMIO_PAGE_OFFSET; |
Peng Hao | 0804c84 | 2018-10-14 07:09:55 +0800 | [diff] [blame] | 3235 | case KVM_CAP_COALESCED_PIO: |
| 3236 | return 1; |
Paolo Bonzini | 3042255 | 2017-03-31 13:53:22 +0200 | [diff] [blame] | 3237 | #endif |
Alexander Graf | 92b591a | 2014-07-14 18:33:08 +0200 | [diff] [blame] | 3238 | #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING |
| 3239 | case KVM_CAP_IRQ_ROUTING: |
| 3240 | return KVM_MAX_IRQ_ROUTES; |
| 3241 | #endif |
Paolo Bonzini | f481b06 | 2015-05-17 17:30:37 +0200 | [diff] [blame] | 3242 | #if KVM_ADDRESS_SPACE_NUM > 1 |
| 3243 | case KVM_CAP_MULTI_ADDRESS_SPACE: |
| 3244 | return KVM_ADDRESS_SPACE_NUM; |
| 3245 | #endif |
Paolo Bonzini | c110ae5 | 2019-03-28 17:24:03 +0100 | [diff] [blame] | 3246 | case KVM_CAP_NR_MEMSLOTS: |
| 3247 | return KVM_USER_MEM_SLOTS; |
Alexander Graf | 92b591a | 2014-07-14 18:33:08 +0200 | [diff] [blame] | 3248 | default: |
| 3249 | break; |
| 3250 | } |
| 3251 | return kvm_vm_ioctl_check_extension(kvm, arg); |
| 3252 | } |
| 3253 | |
Paolo Bonzini | e5d83c7 | 2017-02-16 10:40:56 +0100 | [diff] [blame] | 3254 | int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm, |
| 3255 | struct kvm_enable_cap *cap) |
| 3256 | { |
| 3257 | return -EINVAL; |
| 3258 | } |
| 3259 | |
| 3260 | static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm, |
| 3261 | struct kvm_enable_cap *cap) |
| 3262 | { |
| 3263 | switch (cap->cap) { |
Paolo Bonzini | 2a31b9d | 2018-10-23 02:36:47 +0200 | [diff] [blame] | 3264 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
Peter Xu | d7547c5 | 2019-05-08 17:15:47 +0800 | [diff] [blame] | 3265 | case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: |
Paolo Bonzini | 2a31b9d | 2018-10-23 02:36:47 +0200 | [diff] [blame] | 3266 | if (cap->flags || (cap->args[0] & ~1)) |
| 3267 | return -EINVAL; |
| 3268 | kvm->manual_dirty_log_protect = cap->args[0]; |
| 3269 | return 0; |
| 3270 | #endif |
Paolo Bonzini | e5d83c7 | 2017-02-16 10:40:56 +0100 | [diff] [blame] | 3271 | default: |
| 3272 | return kvm_vm_ioctl_enable_cap(kvm, cap); |
| 3273 | } |
| 3274 | } |
| 3275 | |
Avi Kivity | bccf215 | 2007-02-21 18:04:26 +0200 | [diff] [blame] | 3276 | static long kvm_vm_ioctl(struct file *filp, |
| 3277 | unsigned int ioctl, unsigned long arg) |
| 3278 | { |
| 3279 | struct kvm *kvm = filp->private_data; |
| 3280 | void __user *argp = (void __user *)arg; |
Carsten Otte | 1fe779f | 2007-10-29 16:08:35 +0100 | [diff] [blame] | 3281 | int r; |
Avi Kivity | bccf215 | 2007-02-21 18:04:26 +0200 | [diff] [blame] | 3282 | |
Avi Kivity | 6d4e4c4 | 2007-11-21 16:41:05 +0200 | [diff] [blame] | 3283 | if (kvm->mm != current->mm) |
| 3284 | return -EIO; |
Avi Kivity | bccf215 | 2007-02-21 18:04:26 +0200 | [diff] [blame] | 3285 | switch (ioctl) { |
| 3286 | case KVM_CREATE_VCPU: |
| 3287 | r = kvm_vm_ioctl_create_vcpu(kvm, arg); |
Avi Kivity | bccf215 | 2007-02-21 18:04:26 +0200 | [diff] [blame] | 3288 | break; |
Paolo Bonzini | e5d83c7 | 2017-02-16 10:40:56 +0100 | [diff] [blame] | 3289 | case KVM_ENABLE_CAP: { |
| 3290 | struct kvm_enable_cap cap; |
| 3291 | |
| 3292 | r = -EFAULT; |
| 3293 | if (copy_from_user(&cap, argp, sizeof(cap))) |
| 3294 | goto out; |
| 3295 | r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap); |
| 3296 | break; |
| 3297 | } |
Izik Eidus | 6fc138d | 2007-10-09 19:20:39 +0200 | [diff] [blame] | 3298 | case KVM_SET_USER_MEMORY_REGION: { |
| 3299 | struct kvm_userspace_memory_region kvm_userspace_mem; |
| 3300 | |
| 3301 | r = -EFAULT; |
| 3302 | if (copy_from_user(&kvm_userspace_mem, argp, |
Xiubo Li | 893bdbf | 2015-02-26 14:58:19 +0800 | [diff] [blame] | 3303 | sizeof(kvm_userspace_mem))) |
Izik Eidus | 6fc138d | 2007-10-09 19:20:39 +0200 | [diff] [blame] | 3304 | goto out; |
| 3305 | |
Takuya Yoshikawa | 47ae31e | 2013-02-27 19:43:00 +0900 | [diff] [blame] | 3306 | r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem); |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 3307 | break; |
| 3308 | } |
| 3309 | case KVM_GET_DIRTY_LOG: { |
| 3310 | struct kvm_dirty_log log; |
| 3311 | |
| 3312 | r = -EFAULT; |
Xiubo Li | 893bdbf | 2015-02-26 14:58:19 +0800 | [diff] [blame] | 3313 | if (copy_from_user(&log, argp, sizeof(log))) |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 3314 | goto out; |
Avi Kivity | 2c6f5df | 2007-02-20 18:27:58 +0200 | [diff] [blame] | 3315 | r = kvm_vm_ioctl_get_dirty_log(kvm, &log); |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 3316 | break; |
| 3317 | } |
Paolo Bonzini | 2a31b9d | 2018-10-23 02:36:47 +0200 | [diff] [blame] | 3318 | #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT |
| 3319 | case KVM_CLEAR_DIRTY_LOG: { |
| 3320 | struct kvm_clear_dirty_log log; |
| 3321 | |
| 3322 | r = -EFAULT; |
| 3323 | if (copy_from_user(&log, argp, sizeof(log))) |
| 3324 | goto out; |
| 3325 | r = kvm_vm_ioctl_clear_dirty_log(kvm, &log); |
| 3326 | break; |
| 3327 | } |
| 3328 | #endif |
Paolo Bonzini | 4b4357e | 2017-03-31 13:53:23 +0200 | [diff] [blame] | 3329 | #ifdef CONFIG_KVM_MMIO |
Laurent Vivier | 5f94c17 | 2008-05-30 16:05:54 +0200 | [diff] [blame] | 3330 | case KVM_REGISTER_COALESCED_MMIO: { |
| 3331 | struct kvm_coalesced_mmio_zone zone; |
Xiubo Li | f95ef0cd | 2015-02-26 14:58:23 +0800 | [diff] [blame] | 3332 | |
Laurent Vivier | 5f94c17 | 2008-05-30 16:05:54 +0200 | [diff] [blame] | 3333 | r = -EFAULT; |
Xiubo Li | 893bdbf | 2015-02-26 14:58:19 +0800 | [diff] [blame] | 3334 | if (copy_from_user(&zone, argp, sizeof(zone))) |
Laurent Vivier | 5f94c17 | 2008-05-30 16:05:54 +0200 | [diff] [blame] | 3335 | goto out; |
Laurent Vivier | 5f94c17 | 2008-05-30 16:05:54 +0200 | [diff] [blame] | 3336 | r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone); |
Laurent Vivier | 5f94c17 | 2008-05-30 16:05:54 +0200 | [diff] [blame] | 3337 | break; |
| 3338 | } |
| 3339 | case KVM_UNREGISTER_COALESCED_MMIO: { |
| 3340 | struct kvm_coalesced_mmio_zone zone; |
Xiubo Li | f95ef0cd | 2015-02-26 14:58:23 +0800 | [diff] [blame] | 3341 | |
Laurent Vivier | 5f94c17 | 2008-05-30 16:05:54 +0200 | [diff] [blame] | 3342 | r = -EFAULT; |
Xiubo Li | 893bdbf | 2015-02-26 14:58:19 +0800 | [diff] [blame] | 3343 | if (copy_from_user(&zone, argp, sizeof(zone))) |
Laurent Vivier | 5f94c17 | 2008-05-30 16:05:54 +0200 | [diff] [blame] | 3344 | goto out; |
Laurent Vivier | 5f94c17 | 2008-05-30 16:05:54 +0200 | [diff] [blame] | 3345 | r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone); |
Laurent Vivier | 5f94c17 | 2008-05-30 16:05:54 +0200 | [diff] [blame] | 3346 | break; |
| 3347 | } |
| 3348 | #endif |
Gregory Haskins | 721eecbf | 2009-05-20 10:30:49 -0400 | [diff] [blame] | 3349 | case KVM_IRQFD: { |
| 3350 | struct kvm_irqfd data; |
| 3351 | |
| 3352 | r = -EFAULT; |
Xiubo Li | 893bdbf | 2015-02-26 14:58:19 +0800 | [diff] [blame] | 3353 | if (copy_from_user(&data, argp, sizeof(data))) |
Gregory Haskins | 721eecbf | 2009-05-20 10:30:49 -0400 | [diff] [blame] | 3354 | goto out; |
Alex Williamson | d4db293 | 2012-06-29 09:56:08 -0600 | [diff] [blame] | 3355 | r = kvm_irqfd(kvm, &data); |
Gregory Haskins | 721eecbf | 2009-05-20 10:30:49 -0400 | [diff] [blame] | 3356 | break; |
| 3357 | } |
Gregory Haskins | d34e6b1 | 2009-07-07 17:08:49 -0400 | [diff] [blame] | 3358 | case KVM_IOEVENTFD: { |
| 3359 | struct kvm_ioeventfd data; |
| 3360 | |
| 3361 | r = -EFAULT; |
Xiubo Li | 893bdbf | 2015-02-26 14:58:19 +0800 | [diff] [blame] | 3362 | if (copy_from_user(&data, argp, sizeof(data))) |
Gregory Haskins | d34e6b1 | 2009-07-07 17:08:49 -0400 | [diff] [blame] | 3363 | goto out; |
| 3364 | r = kvm_ioeventfd(kvm, &data); |
| 3365 | break; |
| 3366 | } |
Jan Kiszka | 07975ad | 2012-03-29 21:14:12 +0200 | [diff] [blame] | 3367 | #ifdef CONFIG_HAVE_KVM_MSI |
| 3368 | case KVM_SIGNAL_MSI: { |
| 3369 | struct kvm_msi msi; |
| 3370 | |
| 3371 | r = -EFAULT; |
Xiubo Li | 893bdbf | 2015-02-26 14:58:19 +0800 | [diff] [blame] | 3372 | if (copy_from_user(&msi, argp, sizeof(msi))) |
Jan Kiszka | 07975ad | 2012-03-29 21:14:12 +0200 | [diff] [blame] | 3373 | goto out; |
| 3374 | r = kvm_send_userspace_msi(kvm, &msi); |
| 3375 | break; |
| 3376 | } |
| 3377 | #endif |
Christoffer Dall | 23d43cf | 2012-07-24 08:51:20 -0400 | [diff] [blame] | 3378 | #ifdef __KVM_HAVE_IRQ_LINE |
| 3379 | case KVM_IRQ_LINE_STATUS: |
| 3380 | case KVM_IRQ_LINE: { |
| 3381 | struct kvm_irq_level irq_event; |
| 3382 | |
| 3383 | r = -EFAULT; |
Xiubo Li | 893bdbf | 2015-02-26 14:58:19 +0800 | [diff] [blame] | 3384 | if (copy_from_user(&irq_event, argp, sizeof(irq_event))) |
Christoffer Dall | 23d43cf | 2012-07-24 08:51:20 -0400 | [diff] [blame] | 3385 | goto out; |
| 3386 | |
Yang Zhang | aa2fbe6 | 2013-04-11 19:21:40 +0800 | [diff] [blame] | 3387 | r = kvm_vm_ioctl_irq_line(kvm, &irq_event, |
| 3388 | ioctl == KVM_IRQ_LINE_STATUS); |
Christoffer Dall | 23d43cf | 2012-07-24 08:51:20 -0400 | [diff] [blame] | 3389 | if (r) |
| 3390 | goto out; |
| 3391 | |
| 3392 | r = -EFAULT; |
| 3393 | if (ioctl == KVM_IRQ_LINE_STATUS) { |
Xiubo Li | 893bdbf | 2015-02-26 14:58:19 +0800 | [diff] [blame] | 3394 | if (copy_to_user(argp, &irq_event, sizeof(irq_event))) |
Christoffer Dall | 23d43cf | 2012-07-24 08:51:20 -0400 | [diff] [blame] | 3395 | goto out; |
| 3396 | } |
| 3397 | |
| 3398 | r = 0; |
| 3399 | break; |
| 3400 | } |
| 3401 | #endif |
Alexander Graf | aa8d594 | 2013-04-15 21:12:53 +0200 | [diff] [blame] | 3402 | #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING |
| 3403 | case KVM_SET_GSI_ROUTING: { |
| 3404 | struct kvm_irq_routing routing; |
| 3405 | struct kvm_irq_routing __user *urouting; |
Paolo Bonzini | f8c1b85 | 2016-06-01 14:09:22 +0200 | [diff] [blame] | 3406 | struct kvm_irq_routing_entry *entries = NULL; |
Alexander Graf | aa8d594 | 2013-04-15 21:12:53 +0200 | [diff] [blame] | 3407 | |
| 3408 | r = -EFAULT; |
| 3409 | if (copy_from_user(&routing, argp, sizeof(routing))) |
| 3410 | goto out; |
| 3411 | r = -EINVAL; |
David Hildenbrand | 5c0aea0 | 2017-04-28 17:06:20 +0200 | [diff] [blame] | 3412 | if (!kvm_arch_can_set_irq_routing(kvm)) |
| 3413 | goto out; |
Xiubo Li | caf1ff2 | 2016-06-15 18:00:33 +0800 | [diff] [blame] | 3414 | if (routing.nr > KVM_MAX_IRQ_ROUTES) |
Alexander Graf | aa8d594 | 2013-04-15 21:12:53 +0200 | [diff] [blame] | 3415 | goto out; |
| 3416 | if (routing.flags) |
| 3417 | goto out; |
Paolo Bonzini | f8c1b85 | 2016-06-01 14:09:22 +0200 | [diff] [blame] | 3418 | if (routing.nr) { |
| 3419 | r = -ENOMEM; |
Kees Cook | 42bc47b | 2018-06-12 14:27:11 -0700 | [diff] [blame] | 3420 | entries = vmalloc(array_size(sizeof(*entries), |
| 3421 | routing.nr)); |
Paolo Bonzini | f8c1b85 | 2016-06-01 14:09:22 +0200 | [diff] [blame] | 3422 | if (!entries) |
| 3423 | goto out; |
| 3424 | r = -EFAULT; |
| 3425 | urouting = argp; |
| 3426 | if (copy_from_user(entries, urouting->entries, |
| 3427 | routing.nr * sizeof(*entries))) |
| 3428 | goto out_free_irq_routing; |
| 3429 | } |
Alexander Graf | aa8d594 | 2013-04-15 21:12:53 +0200 | [diff] [blame] | 3430 | r = kvm_set_irq_routing(kvm, entries, routing.nr, |
| 3431 | routing.flags); |
Xiubo Li | a642a17 | 2015-02-26 14:58:20 +0800 | [diff] [blame] | 3432 | out_free_irq_routing: |
Alexander Graf | aa8d594 | 2013-04-15 21:12:53 +0200 | [diff] [blame] | 3433 | vfree(entries); |
| 3434 | break; |
| 3435 | } |
| 3436 | #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */ |
Scott Wood | 852b6d5 | 2013-04-12 14:08:42 +0000 | [diff] [blame] | 3437 | case KVM_CREATE_DEVICE: { |
| 3438 | struct kvm_create_device cd; |
| 3439 | |
| 3440 | r = -EFAULT; |
| 3441 | if (copy_from_user(&cd, argp, sizeof(cd))) |
| 3442 | goto out; |
| 3443 | |
| 3444 | r = kvm_ioctl_create_device(kvm, &cd); |
| 3445 | if (r) |
| 3446 | goto out; |
| 3447 | |
| 3448 | r = -EFAULT; |
| 3449 | if (copy_to_user(argp, &cd, sizeof(cd))) |
| 3450 | goto out; |
| 3451 | |
| 3452 | r = 0; |
| 3453 | break; |
| 3454 | } |
Alexander Graf | 92b591a | 2014-07-14 18:33:08 +0200 | [diff] [blame] | 3455 | case KVM_CHECK_EXTENSION: |
| 3456 | r = kvm_vm_ioctl_check_extension_generic(kvm, arg); |
| 3457 | break; |
Avi Kivity | f17abe9 | 2007-02-21 19:28:04 +0200 | [diff] [blame] | 3458 | default: |
Carsten Otte | 1fe779f | 2007-10-29 16:08:35 +0100 | [diff] [blame] | 3459 | r = kvm_arch_vm_ioctl(filp, ioctl, arg); |
Avi Kivity | f17abe9 | 2007-02-21 19:28:04 +0200 | [diff] [blame] | 3460 | } |
| 3461 | out: |
| 3462 | return r; |
| 3463 | } |
| 3464 | |
Christian Borntraeger | de8e5d7 | 2015-02-03 09:35:15 +0100 | [diff] [blame] | 3465 | #ifdef CONFIG_KVM_COMPAT |
Arnd Bergmann | 6ff5894 | 2009-10-22 14:19:27 +0200 | [diff] [blame] | 3466 | struct compat_kvm_dirty_log { |
| 3467 | __u32 slot; |
| 3468 | __u32 padding1; |
| 3469 | union { |
| 3470 | compat_uptr_t dirty_bitmap; /* one bit per page */ |
| 3471 | __u64 padding2; |
| 3472 | }; |
| 3473 | }; |
| 3474 | |
| 3475 | static long kvm_vm_compat_ioctl(struct file *filp, |
| 3476 | unsigned int ioctl, unsigned long arg) |
| 3477 | { |
| 3478 | struct kvm *kvm = filp->private_data; |
| 3479 | int r; |
| 3480 | |
| 3481 | if (kvm->mm != current->mm) |
| 3482 | return -EIO; |
| 3483 | switch (ioctl) { |
| 3484 | case KVM_GET_DIRTY_LOG: { |
| 3485 | struct compat_kvm_dirty_log compat_log; |
| 3486 | struct kvm_dirty_log log; |
| 3487 | |
Arnd Bergmann | 6ff5894 | 2009-10-22 14:19:27 +0200 | [diff] [blame] | 3488 | if (copy_from_user(&compat_log, (void __user *)arg, |
| 3489 | sizeof(compat_log))) |
Markus Elfring | f6a3b16 | 2017-01-22 11:30:21 +0100 | [diff] [blame] | 3490 | return -EFAULT; |
Arnd Bergmann | 6ff5894 | 2009-10-22 14:19:27 +0200 | [diff] [blame] | 3491 | log.slot = compat_log.slot; |
| 3492 | log.padding1 = compat_log.padding1; |
| 3493 | log.padding2 = compat_log.padding2; |
| 3494 | log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap); |
| 3495 | |
| 3496 | r = kvm_vm_ioctl_get_dirty_log(kvm, &log); |
Arnd Bergmann | 6ff5894 | 2009-10-22 14:19:27 +0200 | [diff] [blame] | 3497 | break; |
| 3498 | } |
| 3499 | default: |
| 3500 | r = kvm_vm_ioctl(filp, ioctl, arg); |
| 3501 | } |
Arnd Bergmann | 6ff5894 | 2009-10-22 14:19:27 +0200 | [diff] [blame] | 3502 | return r; |
| 3503 | } |
| 3504 | #endif |
| 3505 | |
Christian Borntraeger | 3d3aab1 | 2008-12-02 11:17:32 +0100 | [diff] [blame] | 3506 | static struct file_operations kvm_vm_fops = { |
Avi Kivity | f17abe9 | 2007-02-21 19:28:04 +0200 | [diff] [blame] | 3507 | .release = kvm_vm_release, |
| 3508 | .unlocked_ioctl = kvm_vm_ioctl, |
Arnd Bergmann | 6038f37 | 2010-08-15 18:52:59 +0200 | [diff] [blame] | 3509 | .llseek = noop_llseek, |
Marc Zyngier | 7ddfd3e | 2018-06-17 10:16:21 +0100 | [diff] [blame] | 3510 | KVM_COMPAT(kvm_vm_compat_ioctl), |
Avi Kivity | f17abe9 | 2007-02-21 19:28:04 +0200 | [diff] [blame] | 3511 | }; |
| 3512 | |
Carsten Otte | e08b963 | 2012-01-04 10:25:20 +0100 | [diff] [blame] | 3513 | static int kvm_dev_ioctl_create_vm(unsigned long type) |
Avi Kivity | f17abe9 | 2007-02-21 19:28:04 +0200 | [diff] [blame] | 3514 | { |
Heiko Carstens | aac8763 | 2010-10-27 17:22:10 +0200 | [diff] [blame] | 3515 | int r; |
Avi Kivity | f17abe9 | 2007-02-21 19:28:04 +0200 | [diff] [blame] | 3516 | struct kvm *kvm; |
Al Viro | 506cfba | 2016-07-14 18:54:17 +0200 | [diff] [blame] | 3517 | struct file *file; |
Avi Kivity | f17abe9 | 2007-02-21 19:28:04 +0200 | [diff] [blame] | 3518 | |
Carsten Otte | e08b963 | 2012-01-04 10:25:20 +0100 | [diff] [blame] | 3519 | kvm = kvm_create_vm(type); |
Avi Kivity | d6d2816 | 2007-06-28 08:38:16 -0400 | [diff] [blame] | 3520 | if (IS_ERR(kvm)) |
| 3521 | return PTR_ERR(kvm); |
Paolo Bonzini | 4b4357e | 2017-03-31 13:53:23 +0200 | [diff] [blame] | 3522 | #ifdef CONFIG_KVM_MMIO |
Takuya Yoshikawa | 6ce5a09 | 2010-03-15 22:13:30 +0900 | [diff] [blame] | 3523 | r = kvm_coalesced_mmio_init(kvm); |
Markus Elfring | 78588335 | 2017-11-21 13:40:17 +0100 | [diff] [blame] | 3524 | if (r < 0) |
| 3525 | goto put_kvm; |
Takuya Yoshikawa | 6ce5a09 | 2010-03-15 22:13:30 +0900 | [diff] [blame] | 3526 | #endif |
Al Viro | 506cfba | 2016-07-14 18:54:17 +0200 | [diff] [blame] | 3527 | r = get_unused_fd_flags(O_CLOEXEC); |
Markus Elfring | 78588335 | 2017-11-21 13:40:17 +0100 | [diff] [blame] | 3528 | if (r < 0) |
| 3529 | goto put_kvm; |
| 3530 | |
Al Viro | 506cfba | 2016-07-14 18:54:17 +0200 | [diff] [blame] | 3531 | file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR); |
| 3532 | if (IS_ERR(file)) { |
| 3533 | put_unused_fd(r); |
Markus Elfring | 78588335 | 2017-11-21 13:40:17 +0100 | [diff] [blame] | 3534 | r = PTR_ERR(file); |
| 3535 | goto put_kvm; |
Al Viro | 506cfba | 2016-07-14 18:54:17 +0200 | [diff] [blame] | 3536 | } |
Janosch Frank | 536a6f8 | 2016-05-18 13:26:23 +0200 | [diff] [blame] | 3537 | |
Paolo Bonzini | 525df86 | 2017-06-27 15:45:09 +0200 | [diff] [blame] | 3538 | /* |
| 3539 | * Don't call kvm_put_kvm anymore at this point; file->f_op is |
| 3540 | * already set, with ->release() being kvm_vm_release(). In error |
| 3541 | * cases it will be called by the final fput(file) and will take |
| 3542 | * care of doing kvm_put_kvm(kvm). |
| 3543 | */ |
Janosch Frank | 536a6f8 | 2016-05-18 13:26:23 +0200 | [diff] [blame] | 3544 | if (kvm_create_vm_debugfs(kvm, r) < 0) { |
Al Viro | 506cfba | 2016-07-14 18:54:17 +0200 | [diff] [blame] | 3545 | put_unused_fd(r); |
| 3546 | fput(file); |
Janosch Frank | 536a6f8 | 2016-05-18 13:26:23 +0200 | [diff] [blame] | 3547 | return -ENOMEM; |
| 3548 | } |
Claudio Imbrenda | 286de8f | 2017-07-12 17:56:44 +0200 | [diff] [blame] | 3549 | kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm); |
Avi Kivity | f17abe9 | 2007-02-21 19:28:04 +0200 | [diff] [blame] | 3550 | |
Al Viro | 506cfba | 2016-07-14 18:54:17 +0200 | [diff] [blame] | 3551 | fd_install(r, file); |
Heiko Carstens | aac8763 | 2010-10-27 17:22:10 +0200 | [diff] [blame] | 3552 | return r; |
Markus Elfring | 78588335 | 2017-11-21 13:40:17 +0100 | [diff] [blame] | 3553 | |
| 3554 | put_kvm: |
| 3555 | kvm_put_kvm(kvm); |
| 3556 | return r; |
Avi Kivity | f17abe9 | 2007-02-21 19:28:04 +0200 | [diff] [blame] | 3557 | } |
| 3558 | |
| 3559 | static long kvm_dev_ioctl(struct file *filp, |
| 3560 | unsigned int ioctl, unsigned long arg) |
| 3561 | { |
Avi Kivity | 07c45a3 | 2007-03-07 13:05:38 +0200 | [diff] [blame] | 3562 | long r = -EINVAL; |
Avi Kivity | f17abe9 | 2007-02-21 19:28:04 +0200 | [diff] [blame] | 3563 | |
| 3564 | switch (ioctl) { |
| 3565 | case KVM_GET_API_VERSION: |
Avi Kivity | f0fe510 | 2007-03-07 13:11:17 +0200 | [diff] [blame] | 3566 | if (arg) |
| 3567 | goto out; |
Avi Kivity | f17abe9 | 2007-02-21 19:28:04 +0200 | [diff] [blame] | 3568 | r = KVM_API_VERSION; |
| 3569 | break; |
| 3570 | case KVM_CREATE_VM: |
Carsten Otte | e08b963 | 2012-01-04 10:25:20 +0100 | [diff] [blame] | 3571 | r = kvm_dev_ioctl_create_vm(arg); |
Avi Kivity | f17abe9 | 2007-02-21 19:28:04 +0200 | [diff] [blame] | 3572 | break; |
Zhang Xiantao | 018d00d | 2007-11-15 23:07:47 +0800 | [diff] [blame] | 3573 | case KVM_CHECK_EXTENSION: |
Alexander Graf | 784aa3d | 2014-07-14 18:27:35 +0200 | [diff] [blame] | 3574 | r = kvm_vm_ioctl_check_extension_generic(NULL, arg); |
Avi Kivity | 5d308f4 | 2007-03-01 17:56:20 +0200 | [diff] [blame] | 3575 | break; |
Avi Kivity | 07c45a3 | 2007-03-07 13:05:38 +0200 | [diff] [blame] | 3576 | case KVM_GET_VCPU_MMAP_SIZE: |
Avi Kivity | 07c45a3 | 2007-03-07 13:05:38 +0200 | [diff] [blame] | 3577 | if (arg) |
| 3578 | goto out; |
Avi Kivity | adb1ff4 | 2008-01-24 15:13:08 +0200 | [diff] [blame] | 3579 | r = PAGE_SIZE; /* struct kvm_run */ |
| 3580 | #ifdef CONFIG_X86 |
| 3581 | r += PAGE_SIZE; /* pio data page */ |
| 3582 | #endif |
Paolo Bonzini | 4b4357e | 2017-03-31 13:53:23 +0200 | [diff] [blame] | 3583 | #ifdef CONFIG_KVM_MMIO |
Laurent Vivier | 5f94c17 | 2008-05-30 16:05:54 +0200 | [diff] [blame] | 3584 | r += PAGE_SIZE; /* coalesced mmio ring page */ |
| 3585 | #endif |
Avi Kivity | 07c45a3 | 2007-03-07 13:05:38 +0200 | [diff] [blame] | 3586 | break; |
Feng(Eric) Liu | d4c9ff2 | 2008-04-10 08:47:53 -0400 | [diff] [blame] | 3587 | case KVM_TRACE_ENABLE: |
| 3588 | case KVM_TRACE_PAUSE: |
| 3589 | case KVM_TRACE_DISABLE: |
Marcelo Tosatti | 2023a29 | 2009-06-18 11:47:28 -0300 | [diff] [blame] | 3590 | r = -EOPNOTSUPP; |
Feng(Eric) Liu | d4c9ff2 | 2008-04-10 08:47:53 -0400 | [diff] [blame] | 3591 | break; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 3592 | default: |
Carsten Otte | 043405e | 2007-10-10 17:16:19 +0200 | [diff] [blame] | 3593 | return kvm_arch_dev_ioctl(filp, ioctl, arg); |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 3594 | } |
| 3595 | out: |
| 3596 | return r; |
| 3597 | } |
| 3598 | |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 3599 | static struct file_operations kvm_chardev_ops = { |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 3600 | .unlocked_ioctl = kvm_dev_ioctl, |
Arnd Bergmann | 6038f37 | 2010-08-15 18:52:59 +0200 | [diff] [blame] | 3601 | .llseek = noop_llseek, |
Marc Zyngier | 7ddfd3e | 2018-06-17 10:16:21 +0100 | [diff] [blame] | 3602 | KVM_COMPAT(kvm_dev_ioctl), |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 3603 | }; |
| 3604 | |
| 3605 | static struct miscdevice kvm_dev = { |
Avi Kivity | bbe4432 | 2007-03-04 13:27:36 +0200 | [diff] [blame] | 3606 | KVM_MINOR, |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 3607 | "kvm", |
| 3608 | &kvm_chardev_ops, |
| 3609 | }; |
| 3610 | |
Takuya Yoshikawa | 75b7127 | 2010-11-16 17:37:41 +0900 | [diff] [blame] | 3611 | static void hardware_enable_nolock(void *junk) |
Avi Kivity | 1b6c016 | 2007-05-24 13:03:52 +0300 | [diff] [blame] | 3612 | { |
| 3613 | int cpu = raw_smp_processor_id(); |
Alexander Graf | 10474ae | 2009-09-15 11:37:46 +0200 | [diff] [blame] | 3614 | int r; |
Avi Kivity | 1b6c016 | 2007-05-24 13:03:52 +0300 | [diff] [blame] | 3615 | |
Rusty Russell | 7f59f49 | 2008-12-07 21:25:45 +1030 | [diff] [blame] | 3616 | if (cpumask_test_cpu(cpu, cpus_hardware_enabled)) |
Avi Kivity | 1b6c016 | 2007-05-24 13:03:52 +0300 | [diff] [blame] | 3617 | return; |
Alexander Graf | 10474ae | 2009-09-15 11:37:46 +0200 | [diff] [blame] | 3618 | |
Rusty Russell | 7f59f49 | 2008-12-07 21:25:45 +1030 | [diff] [blame] | 3619 | cpumask_set_cpu(cpu, cpus_hardware_enabled); |
Alexander Graf | 10474ae | 2009-09-15 11:37:46 +0200 | [diff] [blame] | 3620 | |
Radim Krčmář | 13a34e0 | 2014-08-28 15:13:03 +0200 | [diff] [blame] | 3621 | r = kvm_arch_hardware_enable(); |
Alexander Graf | 10474ae | 2009-09-15 11:37:46 +0200 | [diff] [blame] | 3622 | |
| 3623 | if (r) { |
| 3624 | cpumask_clear_cpu(cpu, cpus_hardware_enabled); |
| 3625 | atomic_inc(&hardware_enable_failed); |
Xiubo Li | 1170adc | 2015-02-26 14:58:26 +0800 | [diff] [blame] | 3626 | pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu); |
Alexander Graf | 10474ae | 2009-09-15 11:37:46 +0200 | [diff] [blame] | 3627 | } |
Avi Kivity | 1b6c016 | 2007-05-24 13:03:52 +0300 | [diff] [blame] | 3628 | } |
| 3629 | |
Thomas Gleixner | 8c18b2d | 2016-07-13 17:16:37 +0000 | [diff] [blame] | 3630 | static int kvm_starting_cpu(unsigned int cpu) |
Takuya Yoshikawa | 75b7127 | 2010-11-16 17:37:41 +0900 | [diff] [blame] | 3631 | { |
Paolo Bonzini | 4a937f9 | 2013-09-10 12:58:35 +0200 | [diff] [blame] | 3632 | raw_spin_lock(&kvm_count_lock); |
Paolo Bonzini | 4fa92fb | 2013-09-10 12:57:17 +0200 | [diff] [blame] | 3633 | if (kvm_usage_count) |
| 3634 | hardware_enable_nolock(NULL); |
Paolo Bonzini | 4a937f9 | 2013-09-10 12:58:35 +0200 | [diff] [blame] | 3635 | raw_spin_unlock(&kvm_count_lock); |
Thomas Gleixner | 8c18b2d | 2016-07-13 17:16:37 +0000 | [diff] [blame] | 3636 | return 0; |
Takuya Yoshikawa | 75b7127 | 2010-11-16 17:37:41 +0900 | [diff] [blame] | 3637 | } |
| 3638 | |
| 3639 | static void hardware_disable_nolock(void *junk) |
Avi Kivity | 1b6c016 | 2007-05-24 13:03:52 +0300 | [diff] [blame] | 3640 | { |
| 3641 | int cpu = raw_smp_processor_id(); |
| 3642 | |
Rusty Russell | 7f59f49 | 2008-12-07 21:25:45 +1030 | [diff] [blame] | 3643 | if (!cpumask_test_cpu(cpu, cpus_hardware_enabled)) |
Avi Kivity | 1b6c016 | 2007-05-24 13:03:52 +0300 | [diff] [blame] | 3644 | return; |
Rusty Russell | 7f59f49 | 2008-12-07 21:25:45 +1030 | [diff] [blame] | 3645 | cpumask_clear_cpu(cpu, cpus_hardware_enabled); |
Radim Krčmář | 13a34e0 | 2014-08-28 15:13:03 +0200 | [diff] [blame] | 3646 | kvm_arch_hardware_disable(); |
Avi Kivity | 1b6c016 | 2007-05-24 13:03:52 +0300 | [diff] [blame] | 3647 | } |
| 3648 | |
Thomas Gleixner | 8c18b2d | 2016-07-13 17:16:37 +0000 | [diff] [blame] | 3649 | static int kvm_dying_cpu(unsigned int cpu) |
Takuya Yoshikawa | 75b7127 | 2010-11-16 17:37:41 +0900 | [diff] [blame] | 3650 | { |
Paolo Bonzini | 4a937f9 | 2013-09-10 12:58:35 +0200 | [diff] [blame] | 3651 | raw_spin_lock(&kvm_count_lock); |
Paolo Bonzini | 4fa92fb | 2013-09-10 12:57:17 +0200 | [diff] [blame] | 3652 | if (kvm_usage_count) |
| 3653 | hardware_disable_nolock(NULL); |
Paolo Bonzini | 4a937f9 | 2013-09-10 12:58:35 +0200 | [diff] [blame] | 3654 | raw_spin_unlock(&kvm_count_lock); |
Thomas Gleixner | 8c18b2d | 2016-07-13 17:16:37 +0000 | [diff] [blame] | 3655 | return 0; |
Takuya Yoshikawa | 75b7127 | 2010-11-16 17:37:41 +0900 | [diff] [blame] | 3656 | } |
| 3657 | |
Alexander Graf | 10474ae | 2009-09-15 11:37:46 +0200 | [diff] [blame] | 3658 | static void hardware_disable_all_nolock(void) |
| 3659 | { |
| 3660 | BUG_ON(!kvm_usage_count); |
| 3661 | |
| 3662 | kvm_usage_count--; |
| 3663 | if (!kvm_usage_count) |
Takuya Yoshikawa | 75b7127 | 2010-11-16 17:37:41 +0900 | [diff] [blame] | 3664 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
Alexander Graf | 10474ae | 2009-09-15 11:37:46 +0200 | [diff] [blame] | 3665 | } |
| 3666 | |
| 3667 | static void hardware_disable_all(void) |
| 3668 | { |
Paolo Bonzini | 4a937f9 | 2013-09-10 12:58:35 +0200 | [diff] [blame] | 3669 | raw_spin_lock(&kvm_count_lock); |
Alexander Graf | 10474ae | 2009-09-15 11:37:46 +0200 | [diff] [blame] | 3670 | hardware_disable_all_nolock(); |
Paolo Bonzini | 4a937f9 | 2013-09-10 12:58:35 +0200 | [diff] [blame] | 3671 | raw_spin_unlock(&kvm_count_lock); |
Alexander Graf | 10474ae | 2009-09-15 11:37:46 +0200 | [diff] [blame] | 3672 | } |
| 3673 | |
| 3674 | static int hardware_enable_all(void) |
| 3675 | { |
| 3676 | int r = 0; |
| 3677 | |
Paolo Bonzini | 4a937f9 | 2013-09-10 12:58:35 +0200 | [diff] [blame] | 3678 | raw_spin_lock(&kvm_count_lock); |
Alexander Graf | 10474ae | 2009-09-15 11:37:46 +0200 | [diff] [blame] | 3679 | |
| 3680 | kvm_usage_count++; |
| 3681 | if (kvm_usage_count == 1) { |
| 3682 | atomic_set(&hardware_enable_failed, 0); |
Takuya Yoshikawa | 75b7127 | 2010-11-16 17:37:41 +0900 | [diff] [blame] | 3683 | on_each_cpu(hardware_enable_nolock, NULL, 1); |
Alexander Graf | 10474ae | 2009-09-15 11:37:46 +0200 | [diff] [blame] | 3684 | |
| 3685 | if (atomic_read(&hardware_enable_failed)) { |
| 3686 | hardware_disable_all_nolock(); |
| 3687 | r = -EBUSY; |
| 3688 | } |
| 3689 | } |
| 3690 | |
Paolo Bonzini | 4a937f9 | 2013-09-10 12:58:35 +0200 | [diff] [blame] | 3691 | raw_spin_unlock(&kvm_count_lock); |
Alexander Graf | 10474ae | 2009-09-15 11:37:46 +0200 | [diff] [blame] | 3692 | |
| 3693 | return r; |
| 3694 | } |
| 3695 | |
Rusty Russell | 9a2b85c | 2007-07-17 23:17:55 +1000 | [diff] [blame] | 3696 | static int kvm_reboot(struct notifier_block *notifier, unsigned long val, |
Mike Day | d77c26f | 2007-10-08 09:02:08 -0400 | [diff] [blame] | 3697 | void *v) |
Rusty Russell | 9a2b85c | 2007-07-17 23:17:55 +1000 | [diff] [blame] | 3698 | { |
Sheng Yang | 8e1c181 | 2009-04-29 11:09:04 +0800 | [diff] [blame] | 3699 | /* |
| 3700 | * Some (well, at least mine) BIOSes hang on reboot if |
| 3701 | * in vmx root mode. |
| 3702 | * |
| 3703 | * And Intel TXT required VMX off for all cpu when system shutdown. |
| 3704 | */ |
Xiubo Li | 1170adc | 2015-02-26 14:58:26 +0800 | [diff] [blame] | 3705 | pr_info("kvm: exiting hardware virtualization\n"); |
Sheng Yang | 8e1c181 | 2009-04-29 11:09:04 +0800 | [diff] [blame] | 3706 | kvm_rebooting = true; |
Takuya Yoshikawa | 75b7127 | 2010-11-16 17:37:41 +0900 | [diff] [blame] | 3707 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
Rusty Russell | 9a2b85c | 2007-07-17 23:17:55 +1000 | [diff] [blame] | 3708 | return NOTIFY_OK; |
| 3709 | } |
| 3710 | |
| 3711 | static struct notifier_block kvm_reboot_notifier = { |
| 3712 | .notifier_call = kvm_reboot, |
| 3713 | .priority = 0, |
| 3714 | }; |
| 3715 | |
Marcelo Tosatti | e93f8a0 | 2009-12-23 14:35:24 -0200 | [diff] [blame] | 3716 | static void kvm_io_bus_destroy(struct kvm_io_bus *bus) |
Gregory Haskins | 2eeb2e9 | 2007-05-31 14:08:53 -0400 | [diff] [blame] | 3717 | { |
| 3718 | int i; |
| 3719 | |
| 3720 | for (i = 0; i < bus->dev_count; i++) { |
Sasha Levin | 743eeb0 | 2011-07-27 16:00:48 +0300 | [diff] [blame] | 3721 | struct kvm_io_device *pos = bus->range[i].dev; |
Gregory Haskins | 2eeb2e9 | 2007-05-31 14:08:53 -0400 | [diff] [blame] | 3722 | |
| 3723 | kvm_iodevice_destructor(pos); |
| 3724 | } |
Marcelo Tosatti | e93f8a0 | 2009-12-23 14:35:24 -0200 | [diff] [blame] | 3725 | kfree(bus); |
Gregory Haskins | 2eeb2e9 | 2007-05-31 14:08:53 -0400 | [diff] [blame] | 3726 | } |
| 3727 | |
Paolo Bonzini | c21fbff | 2013-08-27 15:41:41 +0200 | [diff] [blame] | 3728 | static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1, |
Xiubo Li | 20e87b7 | 2015-02-26 14:58:25 +0800 | [diff] [blame] | 3729 | const struct kvm_io_range *r2) |
Sasha Levin | 743eeb0 | 2011-07-27 16:00:48 +0300 | [diff] [blame] | 3730 | { |
Jason Wang | 8f4216c7 | 2015-09-15 14:41:57 +0800 | [diff] [blame] | 3731 | gpa_t addr1 = r1->addr; |
| 3732 | gpa_t addr2 = r2->addr; |
| 3733 | |
| 3734 | if (addr1 < addr2) |
Sasha Levin | 743eeb0 | 2011-07-27 16:00:48 +0300 | [diff] [blame] | 3735 | return -1; |
Jason Wang | 8f4216c7 | 2015-09-15 14:41:57 +0800 | [diff] [blame] | 3736 | |
| 3737 | /* If r2->len == 0, match the exact address. If r2->len != 0, |
| 3738 | * accept any overlapping write. Any order is acceptable for |
| 3739 | * overlapping ranges, because kvm_io_bus_get_first_dev ensures |
| 3740 | * we process all of them. |
| 3741 | */ |
| 3742 | if (r2->len) { |
| 3743 | addr1 += r1->len; |
| 3744 | addr2 += r2->len; |
| 3745 | } |
| 3746 | |
| 3747 | if (addr1 > addr2) |
Sasha Levin | 743eeb0 | 2011-07-27 16:00:48 +0300 | [diff] [blame] | 3748 | return 1; |
Jason Wang | 8f4216c7 | 2015-09-15 14:41:57 +0800 | [diff] [blame] | 3749 | |
Sasha Levin | 743eeb0 | 2011-07-27 16:00:48 +0300 | [diff] [blame] | 3750 | return 0; |
| 3751 | } |
| 3752 | |
Paolo Bonzini | a343c9b | 2013-07-16 13:03:29 +0200 | [diff] [blame] | 3753 | static int kvm_io_bus_sort_cmp(const void *p1, const void *p2) |
| 3754 | { |
Paolo Bonzini | c21fbff | 2013-08-27 15:41:41 +0200 | [diff] [blame] | 3755 | return kvm_io_bus_cmp(p1, p2); |
Paolo Bonzini | a343c9b | 2013-07-16 13:03:29 +0200 | [diff] [blame] | 3756 | } |
| 3757 | |
Geoff Levand | 39369f7 | 2013-04-05 19:20:30 +0000 | [diff] [blame] | 3758 | static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus, |
Sasha Levin | 743eeb0 | 2011-07-27 16:00:48 +0300 | [diff] [blame] | 3759 | gpa_t addr, int len) |
| 3760 | { |
| 3761 | struct kvm_io_range *range, key; |
| 3762 | int off; |
| 3763 | |
| 3764 | key = (struct kvm_io_range) { |
| 3765 | .addr = addr, |
| 3766 | .len = len, |
| 3767 | }; |
| 3768 | |
| 3769 | range = bsearch(&key, bus->range, bus->dev_count, |
| 3770 | sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp); |
| 3771 | if (range == NULL) |
| 3772 | return -ENOENT; |
| 3773 | |
| 3774 | off = range - bus->range; |
| 3775 | |
Paolo Bonzini | c21fbff | 2013-08-27 15:41:41 +0200 | [diff] [blame] | 3776 | while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0) |
Sasha Levin | 743eeb0 | 2011-07-27 16:00:48 +0300 | [diff] [blame] | 3777 | off--; |
| 3778 | |
| 3779 | return off; |
| 3780 | } |
| 3781 | |
Nikolay Nikolaev | e32edf4 | 2015-03-26 14:39:28 +0000 | [diff] [blame] | 3782 | static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus, |
Cornelia Huck | 126a5af | 2013-07-03 16:30:53 +0200 | [diff] [blame] | 3783 | struct kvm_io_range *range, const void *val) |
| 3784 | { |
| 3785 | int idx; |
| 3786 | |
| 3787 | idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len); |
| 3788 | if (idx < 0) |
| 3789 | return -EOPNOTSUPP; |
| 3790 | |
| 3791 | while (idx < bus->dev_count && |
Paolo Bonzini | c21fbff | 2013-08-27 15:41:41 +0200 | [diff] [blame] | 3792 | kvm_io_bus_cmp(range, &bus->range[idx]) == 0) { |
Nikolay Nikolaev | e32edf4 | 2015-03-26 14:39:28 +0000 | [diff] [blame] | 3793 | if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr, |
Cornelia Huck | 126a5af | 2013-07-03 16:30:53 +0200 | [diff] [blame] | 3794 | range->len, val)) |
| 3795 | return idx; |
| 3796 | idx++; |
| 3797 | } |
| 3798 | |
| 3799 | return -EOPNOTSUPP; |
| 3800 | } |
| 3801 | |
Michael S. Tsirkin | bda9020 | 2009-06-29 22:24:32 +0300 | [diff] [blame] | 3802 | /* kvm_io_bus_write - called under kvm->slots_lock */ |
Nikolay Nikolaev | e32edf4 | 2015-03-26 14:39:28 +0000 | [diff] [blame] | 3803 | int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr, |
Michael S. Tsirkin | bda9020 | 2009-06-29 22:24:32 +0300 | [diff] [blame] | 3804 | int len, const void *val) |
Gregory Haskins | 2eeb2e9 | 2007-05-31 14:08:53 -0400 | [diff] [blame] | 3805 | { |
Cornelia Huck | 126a5af | 2013-07-03 16:30:53 +0200 | [diff] [blame] | 3806 | struct kvm_io_bus *bus; |
| 3807 | struct kvm_io_range range; |
| 3808 | int r; |
| 3809 | |
| 3810 | range = (struct kvm_io_range) { |
| 3811 | .addr = addr, |
| 3812 | .len = len, |
| 3813 | }; |
| 3814 | |
Nikolay Nikolaev | e32edf4 | 2015-03-26 14:39:28 +0000 | [diff] [blame] | 3815 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
David Hildenbrand | 90db104 | 2017-03-23 18:24:19 +0100 | [diff] [blame] | 3816 | if (!bus) |
| 3817 | return -ENOMEM; |
Nikolay Nikolaev | e32edf4 | 2015-03-26 14:39:28 +0000 | [diff] [blame] | 3818 | r = __kvm_io_bus_write(vcpu, bus, &range, val); |
Cornelia Huck | 126a5af | 2013-07-03 16:30:53 +0200 | [diff] [blame] | 3819 | return r < 0 ? r : 0; |
| 3820 | } |
Leo Yan | a242010 | 2019-02-22 16:10:09 +0800 | [diff] [blame] | 3821 | EXPORT_SYMBOL_GPL(kvm_io_bus_write); |
Cornelia Huck | 126a5af | 2013-07-03 16:30:53 +0200 | [diff] [blame] | 3822 | |
| 3823 | /* kvm_io_bus_write_cookie - called under kvm->slots_lock */ |
Nikolay Nikolaev | e32edf4 | 2015-03-26 14:39:28 +0000 | [diff] [blame] | 3824 | int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, |
| 3825 | gpa_t addr, int len, const void *val, long cookie) |
Cornelia Huck | 126a5af | 2013-07-03 16:30:53 +0200 | [diff] [blame] | 3826 | { |
Lai Jiangshan | 90d83dc | 2010-04-19 17:41:23 +0800 | [diff] [blame] | 3827 | struct kvm_io_bus *bus; |
Sasha Levin | 743eeb0 | 2011-07-27 16:00:48 +0300 | [diff] [blame] | 3828 | struct kvm_io_range range; |
| 3829 | |
| 3830 | range = (struct kvm_io_range) { |
| 3831 | .addr = addr, |
| 3832 | .len = len, |
| 3833 | }; |
Lai Jiangshan | 90d83dc | 2010-04-19 17:41:23 +0800 | [diff] [blame] | 3834 | |
Nikolay Nikolaev | e32edf4 | 2015-03-26 14:39:28 +0000 | [diff] [blame] | 3835 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
David Hildenbrand | 90db104 | 2017-03-23 18:24:19 +0100 | [diff] [blame] | 3836 | if (!bus) |
| 3837 | return -ENOMEM; |
Cornelia Huck | 126a5af | 2013-07-03 16:30:53 +0200 | [diff] [blame] | 3838 | |
| 3839 | /* First try the device referenced by cookie. */ |
| 3840 | if ((cookie >= 0) && (cookie < bus->dev_count) && |
Paolo Bonzini | c21fbff | 2013-08-27 15:41:41 +0200 | [diff] [blame] | 3841 | (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0)) |
Nikolay Nikolaev | e32edf4 | 2015-03-26 14:39:28 +0000 | [diff] [blame] | 3842 | if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len, |
Cornelia Huck | 126a5af | 2013-07-03 16:30:53 +0200 | [diff] [blame] | 3843 | val)) |
| 3844 | return cookie; |
| 3845 | |
| 3846 | /* |
| 3847 | * cookie contained garbage; fall back to search and return the |
| 3848 | * correct cookie value. |
| 3849 | */ |
Nikolay Nikolaev | e32edf4 | 2015-03-26 14:39:28 +0000 | [diff] [blame] | 3850 | return __kvm_io_bus_write(vcpu, bus, &range, val); |
Cornelia Huck | 126a5af | 2013-07-03 16:30:53 +0200 | [diff] [blame] | 3851 | } |
| 3852 | |
Nikolay Nikolaev | e32edf4 | 2015-03-26 14:39:28 +0000 | [diff] [blame] | 3853 | static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus, |
| 3854 | struct kvm_io_range *range, void *val) |
Cornelia Huck | 126a5af | 2013-07-03 16:30:53 +0200 | [diff] [blame] | 3855 | { |
| 3856 | int idx; |
| 3857 | |
| 3858 | idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len); |
Sasha Levin | 743eeb0 | 2011-07-27 16:00:48 +0300 | [diff] [blame] | 3859 | if (idx < 0) |
| 3860 | return -EOPNOTSUPP; |
| 3861 | |
| 3862 | while (idx < bus->dev_count && |
Paolo Bonzini | c21fbff | 2013-08-27 15:41:41 +0200 | [diff] [blame] | 3863 | kvm_io_bus_cmp(range, &bus->range[idx]) == 0) { |
Nikolay Nikolaev | e32edf4 | 2015-03-26 14:39:28 +0000 | [diff] [blame] | 3864 | if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr, |
Cornelia Huck | 126a5af | 2013-07-03 16:30:53 +0200 | [diff] [blame] | 3865 | range->len, val)) |
| 3866 | return idx; |
Sasha Levin | 743eeb0 | 2011-07-27 16:00:48 +0300 | [diff] [blame] | 3867 | idx++; |
| 3868 | } |
| 3869 | |
Michael S. Tsirkin | bda9020 | 2009-06-29 22:24:32 +0300 | [diff] [blame] | 3870 | return -EOPNOTSUPP; |
| 3871 | } |
Gregory Haskins | 2eeb2e9 | 2007-05-31 14:08:53 -0400 | [diff] [blame] | 3872 | |
Michael S. Tsirkin | bda9020 | 2009-06-29 22:24:32 +0300 | [diff] [blame] | 3873 | /* kvm_io_bus_read - called under kvm->slots_lock */ |
Nikolay Nikolaev | e32edf4 | 2015-03-26 14:39:28 +0000 | [diff] [blame] | 3874 | int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr, |
Marcelo Tosatti | e93f8a0 | 2009-12-23 14:35:24 -0200 | [diff] [blame] | 3875 | int len, void *val) |
Michael S. Tsirkin | bda9020 | 2009-06-29 22:24:32 +0300 | [diff] [blame] | 3876 | { |
Cornelia Huck | 126a5af | 2013-07-03 16:30:53 +0200 | [diff] [blame] | 3877 | struct kvm_io_bus *bus; |
| 3878 | struct kvm_io_range range; |
| 3879 | int r; |
| 3880 | |
| 3881 | range = (struct kvm_io_range) { |
| 3882 | .addr = addr, |
| 3883 | .len = len, |
| 3884 | }; |
| 3885 | |
Nikolay Nikolaev | e32edf4 | 2015-03-26 14:39:28 +0000 | [diff] [blame] | 3886 | bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu); |
David Hildenbrand | 90db104 | 2017-03-23 18:24:19 +0100 | [diff] [blame] | 3887 | if (!bus) |
| 3888 | return -ENOMEM; |
Nikolay Nikolaev | e32edf4 | 2015-03-26 14:39:28 +0000 | [diff] [blame] | 3889 | r = __kvm_io_bus_read(vcpu, bus, &range, val); |
Cornelia Huck | 126a5af | 2013-07-03 16:30:53 +0200 | [diff] [blame] | 3890 | return r < 0 ? r : 0; |
| 3891 | } |
| 3892 | |
Marcelo Tosatti | 79fac95 | 2009-12-23 14:35:26 -0200 | [diff] [blame] | 3893 | /* Caller must hold slots_lock. */ |
Sasha Levin | 743eeb0 | 2011-07-27 16:00:48 +0300 | [diff] [blame] | 3894 | int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr, |
| 3895 | int len, struct kvm_io_device *dev) |
Michael S. Tsirkin | 6c47469 | 2009-06-29 22:24:26 +0300 | [diff] [blame] | 3896 | { |
Gal Hammer | d4c67a7 | 2018-01-16 15:34:41 +0200 | [diff] [blame] | 3897 | int i; |
Marcelo Tosatti | e93f8a0 | 2009-12-23 14:35:24 -0200 | [diff] [blame] | 3898 | struct kvm_io_bus *new_bus, *bus; |
Gal Hammer | d4c67a7 | 2018-01-16 15:34:41 +0200 | [diff] [blame] | 3899 | struct kvm_io_range range; |
Gregory Haskins | 090b7af | 2009-07-07 17:08:44 -0400 | [diff] [blame] | 3900 | |
Christian Borntraeger | 4a12f95 | 2017-07-07 10:51:38 +0200 | [diff] [blame] | 3901 | bus = kvm_get_bus(kvm, bus_idx); |
David Hildenbrand | 90db104 | 2017-03-23 18:24:19 +0100 | [diff] [blame] | 3902 | if (!bus) |
| 3903 | return -ENOMEM; |
| 3904 | |
Amos Kong | 6ea34c9 | 2013-05-25 06:44:15 +0800 | [diff] [blame] | 3905 | /* exclude ioeventfd which is limited by maximum fd */ |
| 3906 | if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1) |
Gregory Haskins | 090b7af | 2009-07-07 17:08:44 -0400 | [diff] [blame] | 3907 | return -ENOSPC; |
| 3908 | |
Gustavo A. R. Silva | 90952cd | 2019-01-30 17:07:47 +0100 | [diff] [blame] | 3909 | new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1), |
Ben Gardon | b12ce36 | 2019-02-11 11:02:49 -0800 | [diff] [blame] | 3910 | GFP_KERNEL_ACCOUNT); |
Marcelo Tosatti | e93f8a0 | 2009-12-23 14:35:24 -0200 | [diff] [blame] | 3911 | if (!new_bus) |
| 3912 | return -ENOMEM; |
Gal Hammer | d4c67a7 | 2018-01-16 15:34:41 +0200 | [diff] [blame] | 3913 | |
| 3914 | range = (struct kvm_io_range) { |
| 3915 | .addr = addr, |
| 3916 | .len = len, |
| 3917 | .dev = dev, |
| 3918 | }; |
| 3919 | |
| 3920 | for (i = 0; i < bus->dev_count; i++) |
| 3921 | if (kvm_io_bus_cmp(&bus->range[i], &range) > 0) |
| 3922 | break; |
| 3923 | |
| 3924 | memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range)); |
| 3925 | new_bus->dev_count++; |
| 3926 | new_bus->range[i] = range; |
| 3927 | memcpy(new_bus->range + i + 1, bus->range + i, |
| 3928 | (bus->dev_count - i) * sizeof(struct kvm_io_range)); |
Marcelo Tosatti | e93f8a0 | 2009-12-23 14:35:24 -0200 | [diff] [blame] | 3929 | rcu_assign_pointer(kvm->buses[bus_idx], new_bus); |
| 3930 | synchronize_srcu_expedited(&kvm->srcu); |
| 3931 | kfree(bus); |
Gregory Haskins | 090b7af | 2009-07-07 17:08:44 -0400 | [diff] [blame] | 3932 | |
| 3933 | return 0; |
| 3934 | } |
| 3935 | |
Marcelo Tosatti | 79fac95 | 2009-12-23 14:35:26 -0200 | [diff] [blame] | 3936 | /* Caller must hold slots_lock. */ |
David Hildenbrand | 90db104 | 2017-03-23 18:24:19 +0100 | [diff] [blame] | 3937 | void kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx, |
| 3938 | struct kvm_io_device *dev) |
Gregory Haskins | 090b7af | 2009-07-07 17:08:44 -0400 | [diff] [blame] | 3939 | { |
David Hildenbrand | 90db104 | 2017-03-23 18:24:19 +0100 | [diff] [blame] | 3940 | int i; |
Marcelo Tosatti | e93f8a0 | 2009-12-23 14:35:24 -0200 | [diff] [blame] | 3941 | struct kvm_io_bus *new_bus, *bus; |
Michael S. Tsirkin | 6c47469 | 2009-06-29 22:24:26 +0300 | [diff] [blame] | 3942 | |
Christian Borntraeger | 4a12f95 | 2017-07-07 10:51:38 +0200 | [diff] [blame] | 3943 | bus = kvm_get_bus(kvm, bus_idx); |
Peter Xu | df630b8 | 2017-03-15 16:01:17 +0800 | [diff] [blame] | 3944 | if (!bus) |
David Hildenbrand | 90db104 | 2017-03-23 18:24:19 +0100 | [diff] [blame] | 3945 | return; |
Peter Xu | df630b8 | 2017-03-15 16:01:17 +0800 | [diff] [blame] | 3946 | |
Amos Kong | a1300716 | 2012-03-09 12:17:32 +0800 | [diff] [blame] | 3947 | for (i = 0; i < bus->dev_count; i++) |
| 3948 | if (bus->range[i].dev == dev) { |
Gregory Haskins | 090b7af | 2009-07-07 17:08:44 -0400 | [diff] [blame] | 3949 | break; |
| 3950 | } |
Marcelo Tosatti | e93f8a0 | 2009-12-23 14:35:24 -0200 | [diff] [blame] | 3951 | |
David Hildenbrand | 90db104 | 2017-03-23 18:24:19 +0100 | [diff] [blame] | 3952 | if (i == bus->dev_count) |
| 3953 | return; |
Amos Kong | a1300716 | 2012-03-09 12:17:32 +0800 | [diff] [blame] | 3954 | |
Gustavo A. R. Silva | 90952cd | 2019-01-30 17:07:47 +0100 | [diff] [blame] | 3955 | new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1), |
Ben Gardon | b12ce36 | 2019-02-11 11:02:49 -0800 | [diff] [blame] | 3956 | GFP_KERNEL_ACCOUNT); |
David Hildenbrand | 90db104 | 2017-03-23 18:24:19 +0100 | [diff] [blame] | 3957 | if (!new_bus) { |
| 3958 | pr_err("kvm: failed to shrink bus, removing it completely\n"); |
| 3959 | goto broken; |
| 3960 | } |
Amos Kong | a1300716 | 2012-03-09 12:17:32 +0800 | [diff] [blame] | 3961 | |
| 3962 | memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range)); |
| 3963 | new_bus->dev_count--; |
| 3964 | memcpy(new_bus->range + i, bus->range + i + 1, |
| 3965 | (new_bus->dev_count - i) * sizeof(struct kvm_io_range)); |
Marcelo Tosatti | e93f8a0 | 2009-12-23 14:35:24 -0200 | [diff] [blame] | 3966 | |
David Hildenbrand | 90db104 | 2017-03-23 18:24:19 +0100 | [diff] [blame] | 3967 | broken: |
Marcelo Tosatti | e93f8a0 | 2009-12-23 14:35:24 -0200 | [diff] [blame] | 3968 | rcu_assign_pointer(kvm->buses[bus_idx], new_bus); |
| 3969 | synchronize_srcu_expedited(&kvm->srcu); |
| 3970 | kfree(bus); |
David Hildenbrand | 90db104 | 2017-03-23 18:24:19 +0100 | [diff] [blame] | 3971 | return; |
Gregory Haskins | 2eeb2e9 | 2007-05-31 14:08:53 -0400 | [diff] [blame] | 3972 | } |
| 3973 | |
Andre Przywara | 8a39d00 | 2016-07-15 12:43:26 +0100 | [diff] [blame] | 3974 | struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx, |
| 3975 | gpa_t addr) |
| 3976 | { |
| 3977 | struct kvm_io_bus *bus; |
| 3978 | int dev_idx, srcu_idx; |
| 3979 | struct kvm_io_device *iodev = NULL; |
| 3980 | |
| 3981 | srcu_idx = srcu_read_lock(&kvm->srcu); |
| 3982 | |
| 3983 | bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu); |
David Hildenbrand | 90db104 | 2017-03-23 18:24:19 +0100 | [diff] [blame] | 3984 | if (!bus) |
| 3985 | goto out_unlock; |
Andre Przywara | 8a39d00 | 2016-07-15 12:43:26 +0100 | [diff] [blame] | 3986 | |
| 3987 | dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1); |
| 3988 | if (dev_idx < 0) |
| 3989 | goto out_unlock; |
| 3990 | |
| 3991 | iodev = bus->range[dev_idx].dev; |
| 3992 | |
| 3993 | out_unlock: |
| 3994 | srcu_read_unlock(&kvm->srcu, srcu_idx); |
| 3995 | |
| 3996 | return iodev; |
| 3997 | } |
| 3998 | EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev); |
| 3999 | |
Janosch Frank | 536a6f8 | 2016-05-18 13:26:23 +0200 | [diff] [blame] | 4000 | static int kvm_debugfs_open(struct inode *inode, struct file *file, |
| 4001 | int (*get)(void *, u64 *), int (*set)(void *, u64), |
| 4002 | const char *fmt) |
| 4003 | { |
| 4004 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *) |
| 4005 | inode->i_private; |
| 4006 | |
| 4007 | /* The debugfs files are a reference to the kvm struct which |
| 4008 | * is still valid when kvm_destroy_vm is called. |
| 4009 | * To avoid the race between open and the removal of the debugfs |
| 4010 | * directory we test against the users count. |
| 4011 | */ |
Elena Reshetova | e3736c3 | 2017-02-20 13:06:21 +0200 | [diff] [blame] | 4012 | if (!refcount_inc_not_zero(&stat_data->kvm->users_count)) |
Janosch Frank | 536a6f8 | 2016-05-18 13:26:23 +0200 | [diff] [blame] | 4013 | return -ENOENT; |
| 4014 | |
Paolo Bonzini | 833b45d | 2019-09-30 18:48:44 +0200 | [diff] [blame] | 4015 | if (simple_attr_open(inode, file, get, |
| 4016 | stat_data->mode & S_IWUGO ? set : NULL, |
| 4017 | fmt)) { |
Janosch Frank | 536a6f8 | 2016-05-18 13:26:23 +0200 | [diff] [blame] | 4018 | kvm_put_kvm(stat_data->kvm); |
| 4019 | return -ENOMEM; |
| 4020 | } |
| 4021 | |
| 4022 | return 0; |
| 4023 | } |
| 4024 | |
| 4025 | static int kvm_debugfs_release(struct inode *inode, struct file *file) |
| 4026 | { |
| 4027 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *) |
| 4028 | inode->i_private; |
| 4029 | |
| 4030 | simple_attr_release(inode, file); |
| 4031 | kvm_put_kvm(stat_data->kvm); |
| 4032 | |
| 4033 | return 0; |
| 4034 | } |
| 4035 | |
| 4036 | static int vm_stat_get_per_vm(void *data, u64 *val) |
| 4037 | { |
| 4038 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data; |
| 4039 | |
Suraj Jitindar Singh | 8a7e75d | 2016-08-02 14:03:22 +1000 | [diff] [blame] | 4040 | *val = *(ulong *)((void *)stat_data->kvm + stat_data->offset); |
Janosch Frank | 536a6f8 | 2016-05-18 13:26:23 +0200 | [diff] [blame] | 4041 | |
| 4042 | return 0; |
| 4043 | } |
| 4044 | |
Suraj Jitindar Singh | ce35ef2 | 2016-10-19 13:49:47 +1100 | [diff] [blame] | 4045 | static int vm_stat_clear_per_vm(void *data, u64 val) |
| 4046 | { |
| 4047 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data; |
| 4048 | |
| 4049 | if (val) |
| 4050 | return -EINVAL; |
| 4051 | |
| 4052 | *(ulong *)((void *)stat_data->kvm + stat_data->offset) = 0; |
| 4053 | |
| 4054 | return 0; |
| 4055 | } |
| 4056 | |
Janosch Frank | 536a6f8 | 2016-05-18 13:26:23 +0200 | [diff] [blame] | 4057 | static int vm_stat_get_per_vm_open(struct inode *inode, struct file *file) |
| 4058 | { |
| 4059 | __simple_attr_check_format("%llu\n", 0ull); |
| 4060 | return kvm_debugfs_open(inode, file, vm_stat_get_per_vm, |
Suraj Jitindar Singh | ce35ef2 | 2016-10-19 13:49:47 +1100 | [diff] [blame] | 4061 | vm_stat_clear_per_vm, "%llu\n"); |
Janosch Frank | 536a6f8 | 2016-05-18 13:26:23 +0200 | [diff] [blame] | 4062 | } |
| 4063 | |
| 4064 | static const struct file_operations vm_stat_get_per_vm_fops = { |
| 4065 | .owner = THIS_MODULE, |
| 4066 | .open = vm_stat_get_per_vm_open, |
| 4067 | .release = kvm_debugfs_release, |
| 4068 | .read = simple_attr_read, |
| 4069 | .write = simple_attr_write, |
Geliang Tang | 3bed888 | 2017-05-06 23:37:19 +0800 | [diff] [blame] | 4070 | .llseek = no_llseek, |
Janosch Frank | 536a6f8 | 2016-05-18 13:26:23 +0200 | [diff] [blame] | 4071 | }; |
| 4072 | |
| 4073 | static int vcpu_stat_get_per_vm(void *data, u64 *val) |
| 4074 | { |
| 4075 | int i; |
| 4076 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data; |
| 4077 | struct kvm_vcpu *vcpu; |
| 4078 | |
| 4079 | *val = 0; |
| 4080 | |
| 4081 | kvm_for_each_vcpu(i, vcpu, stat_data->kvm) |
Suraj Jitindar Singh | 8a7e75d | 2016-08-02 14:03:22 +1000 | [diff] [blame] | 4082 | *val += *(u64 *)((void *)vcpu + stat_data->offset); |
Janosch Frank | 536a6f8 | 2016-05-18 13:26:23 +0200 | [diff] [blame] | 4083 | |
| 4084 | return 0; |
| 4085 | } |
| 4086 | |
Suraj Jitindar Singh | ce35ef2 | 2016-10-19 13:49:47 +1100 | [diff] [blame] | 4087 | static int vcpu_stat_clear_per_vm(void *data, u64 val) |
| 4088 | { |
| 4089 | int i; |
| 4090 | struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data; |
| 4091 | struct kvm_vcpu *vcpu; |
| 4092 | |
| 4093 | if (val) |
| 4094 | return -EINVAL; |
| 4095 | |
| 4096 | kvm_for_each_vcpu(i, vcpu, stat_data->kvm) |
| 4097 | *(u64 *)((void *)vcpu + stat_data->offset) = 0; |
| 4098 | |
| 4099 | return 0; |
| 4100 | } |
| 4101 | |
Janosch Frank | 536a6f8 | 2016-05-18 13:26:23 +0200 | [diff] [blame] | 4102 | static int vcpu_stat_get_per_vm_open(struct inode *inode, struct file *file) |
| 4103 | { |
| 4104 | __simple_attr_check_format("%llu\n", 0ull); |
| 4105 | return kvm_debugfs_open(inode, file, vcpu_stat_get_per_vm, |
Suraj Jitindar Singh | ce35ef2 | 2016-10-19 13:49:47 +1100 | [diff] [blame] | 4106 | vcpu_stat_clear_per_vm, "%llu\n"); |
Janosch Frank | 536a6f8 | 2016-05-18 13:26:23 +0200 | [diff] [blame] | 4107 | } |
| 4108 | |
| 4109 | static const struct file_operations vcpu_stat_get_per_vm_fops = { |
| 4110 | .owner = THIS_MODULE, |
| 4111 | .open = vcpu_stat_get_per_vm_open, |
| 4112 | .release = kvm_debugfs_release, |
| 4113 | .read = simple_attr_read, |
| 4114 | .write = simple_attr_write, |
Geliang Tang | 3bed888 | 2017-05-06 23:37:19 +0800 | [diff] [blame] | 4115 | .llseek = no_llseek, |
Janosch Frank | 536a6f8 | 2016-05-18 13:26:23 +0200 | [diff] [blame] | 4116 | }; |
| 4117 | |
| 4118 | static const struct file_operations *stat_fops_per_vm[] = { |
| 4119 | [KVM_STAT_VCPU] = &vcpu_stat_get_per_vm_fops, |
| 4120 | [KVM_STAT_VM] = &vm_stat_get_per_vm_fops, |
| 4121 | }; |
| 4122 | |
Christoph Hellwig | 8b88b09 | 2008-02-08 04:20:26 -0800 | [diff] [blame] | 4123 | static int vm_stat_get(void *_offset, u64 *val) |
Avi Kivity | ba1389b | 2007-11-18 16:24:12 +0200 | [diff] [blame] | 4124 | { |
| 4125 | unsigned offset = (long)_offset; |
Avi Kivity | ba1389b | 2007-11-18 16:24:12 +0200 | [diff] [blame] | 4126 | struct kvm *kvm; |
Janosch Frank | 536a6f8 | 2016-05-18 13:26:23 +0200 | [diff] [blame] | 4127 | struct kvm_stat_data stat_tmp = {.offset = offset}; |
| 4128 | u64 tmp_val; |
Avi Kivity | ba1389b | 2007-11-18 16:24:12 +0200 | [diff] [blame] | 4129 | |
Christoph Hellwig | 8b88b09 | 2008-02-08 04:20:26 -0800 | [diff] [blame] | 4130 | *val = 0; |
Junaid Shahid | 0d9ce16 | 2019-01-03 17:14:28 -0800 | [diff] [blame] | 4131 | mutex_lock(&kvm_lock); |
Janosch Frank | 536a6f8 | 2016-05-18 13:26:23 +0200 | [diff] [blame] | 4132 | list_for_each_entry(kvm, &vm_list, vm_list) { |
| 4133 | stat_tmp.kvm = kvm; |
| 4134 | vm_stat_get_per_vm((void *)&stat_tmp, &tmp_val); |
| 4135 | *val += tmp_val; |
| 4136 | } |
Junaid Shahid | 0d9ce16 | 2019-01-03 17:14:28 -0800 | [diff] [blame] | 4137 | mutex_unlock(&kvm_lock); |
Christoph Hellwig | 8b88b09 | 2008-02-08 04:20:26 -0800 | [diff] [blame] | 4138 | return 0; |
Avi Kivity | ba1389b | 2007-11-18 16:24:12 +0200 | [diff] [blame] | 4139 | } |
| 4140 | |
Suraj Jitindar Singh | ce35ef2 | 2016-10-19 13:49:47 +1100 | [diff] [blame] | 4141 | static int vm_stat_clear(void *_offset, u64 val) |
| 4142 | { |
| 4143 | unsigned offset = (long)_offset; |
| 4144 | struct kvm *kvm; |
| 4145 | struct kvm_stat_data stat_tmp = {.offset = offset}; |
| 4146 | |
| 4147 | if (val) |
| 4148 | return -EINVAL; |
| 4149 | |
Junaid Shahid | 0d9ce16 | 2019-01-03 17:14:28 -0800 | [diff] [blame] | 4150 | mutex_lock(&kvm_lock); |
Suraj Jitindar Singh | ce35ef2 | 2016-10-19 13:49:47 +1100 | [diff] [blame] | 4151 | list_for_each_entry(kvm, &vm_list, vm_list) { |
| 4152 | stat_tmp.kvm = kvm; |
| 4153 | vm_stat_clear_per_vm((void *)&stat_tmp, 0); |
| 4154 | } |
Junaid Shahid | 0d9ce16 | 2019-01-03 17:14:28 -0800 | [diff] [blame] | 4155 | mutex_unlock(&kvm_lock); |
Suraj Jitindar Singh | ce35ef2 | 2016-10-19 13:49:47 +1100 | [diff] [blame] | 4156 | |
| 4157 | return 0; |
| 4158 | } |
| 4159 | |
| 4160 | DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n"); |
Avi Kivity | ba1389b | 2007-11-18 16:24:12 +0200 | [diff] [blame] | 4161 | |
Christoph Hellwig | 8b88b09 | 2008-02-08 04:20:26 -0800 | [diff] [blame] | 4162 | static int vcpu_stat_get(void *_offset, u64 *val) |
Avi Kivity | 1165f5f | 2007-04-19 17:27:43 +0300 | [diff] [blame] | 4163 | { |
| 4164 | unsigned offset = (long)_offset; |
Avi Kivity | 1165f5f | 2007-04-19 17:27:43 +0300 | [diff] [blame] | 4165 | struct kvm *kvm; |
Janosch Frank | 536a6f8 | 2016-05-18 13:26:23 +0200 | [diff] [blame] | 4166 | struct kvm_stat_data stat_tmp = {.offset = offset}; |
| 4167 | u64 tmp_val; |
Avi Kivity | 1165f5f | 2007-04-19 17:27:43 +0300 | [diff] [blame] | 4168 | |
Christoph Hellwig | 8b88b09 | 2008-02-08 04:20:26 -0800 | [diff] [blame] | 4169 | *val = 0; |
Junaid Shahid | 0d9ce16 | 2019-01-03 17:14:28 -0800 | [diff] [blame] | 4170 | mutex_lock(&kvm_lock); |
Janosch Frank | 536a6f8 | 2016-05-18 13:26:23 +0200 | [diff] [blame] | 4171 | list_for_each_entry(kvm, &vm_list, vm_list) { |
| 4172 | stat_tmp.kvm = kvm; |
| 4173 | vcpu_stat_get_per_vm((void *)&stat_tmp, &tmp_val); |
| 4174 | *val += tmp_val; |
| 4175 | } |
Junaid Shahid | 0d9ce16 | 2019-01-03 17:14:28 -0800 | [diff] [blame] | 4176 | mutex_unlock(&kvm_lock); |
Christoph Hellwig | 8b88b09 | 2008-02-08 04:20:26 -0800 | [diff] [blame] | 4177 | return 0; |
Avi Kivity | 1165f5f | 2007-04-19 17:27:43 +0300 | [diff] [blame] | 4178 | } |
| 4179 | |
Suraj Jitindar Singh | ce35ef2 | 2016-10-19 13:49:47 +1100 | [diff] [blame] | 4180 | static int vcpu_stat_clear(void *_offset, u64 val) |
| 4181 | { |
| 4182 | unsigned offset = (long)_offset; |
| 4183 | struct kvm *kvm; |
| 4184 | struct kvm_stat_data stat_tmp = {.offset = offset}; |
| 4185 | |
| 4186 | if (val) |
| 4187 | return -EINVAL; |
| 4188 | |
Junaid Shahid | 0d9ce16 | 2019-01-03 17:14:28 -0800 | [diff] [blame] | 4189 | mutex_lock(&kvm_lock); |
Suraj Jitindar Singh | ce35ef2 | 2016-10-19 13:49:47 +1100 | [diff] [blame] | 4190 | list_for_each_entry(kvm, &vm_list, vm_list) { |
| 4191 | stat_tmp.kvm = kvm; |
| 4192 | vcpu_stat_clear_per_vm((void *)&stat_tmp, 0); |
| 4193 | } |
Junaid Shahid | 0d9ce16 | 2019-01-03 17:14:28 -0800 | [diff] [blame] | 4194 | mutex_unlock(&kvm_lock); |
Suraj Jitindar Singh | ce35ef2 | 2016-10-19 13:49:47 +1100 | [diff] [blame] | 4195 | |
| 4196 | return 0; |
| 4197 | } |
| 4198 | |
| 4199 | DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear, |
| 4200 | "%llu\n"); |
Avi Kivity | ba1389b | 2007-11-18 16:24:12 +0200 | [diff] [blame] | 4201 | |
Alexey Dobriyan | 828c095 | 2009-10-01 15:43:56 -0700 | [diff] [blame] | 4202 | static const struct file_operations *stat_fops[] = { |
Avi Kivity | ba1389b | 2007-11-18 16:24:12 +0200 | [diff] [blame] | 4203 | [KVM_STAT_VCPU] = &vcpu_stat_fops, |
| 4204 | [KVM_STAT_VM] = &vm_stat_fops, |
| 4205 | }; |
Avi Kivity | 1165f5f | 2007-04-19 17:27:43 +0300 | [diff] [blame] | 4206 | |
Claudio Imbrenda | 286de8f | 2017-07-12 17:56:44 +0200 | [diff] [blame] | 4207 | static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm) |
| 4208 | { |
| 4209 | struct kobj_uevent_env *env; |
Claudio Imbrenda | 286de8f | 2017-07-12 17:56:44 +0200 | [diff] [blame] | 4210 | unsigned long long created, active; |
| 4211 | |
| 4212 | if (!kvm_dev.this_device || !kvm) |
| 4213 | return; |
| 4214 | |
Junaid Shahid | 0d9ce16 | 2019-01-03 17:14:28 -0800 | [diff] [blame] | 4215 | mutex_lock(&kvm_lock); |
Claudio Imbrenda | 286de8f | 2017-07-12 17:56:44 +0200 | [diff] [blame] | 4216 | if (type == KVM_EVENT_CREATE_VM) { |
| 4217 | kvm_createvm_count++; |
| 4218 | kvm_active_vms++; |
| 4219 | } else if (type == KVM_EVENT_DESTROY_VM) { |
| 4220 | kvm_active_vms--; |
| 4221 | } |
| 4222 | created = kvm_createvm_count; |
| 4223 | active = kvm_active_vms; |
Junaid Shahid | 0d9ce16 | 2019-01-03 17:14:28 -0800 | [diff] [blame] | 4224 | mutex_unlock(&kvm_lock); |
Claudio Imbrenda | 286de8f | 2017-07-12 17:56:44 +0200 | [diff] [blame] | 4225 | |
Ben Gardon | b12ce36 | 2019-02-11 11:02:49 -0800 | [diff] [blame] | 4226 | env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT); |
Claudio Imbrenda | 286de8f | 2017-07-12 17:56:44 +0200 | [diff] [blame] | 4227 | if (!env) |
| 4228 | return; |
| 4229 | |
| 4230 | add_uevent_var(env, "CREATED=%llu", created); |
| 4231 | add_uevent_var(env, "COUNT=%llu", active); |
| 4232 | |
Claudio Imbrenda | fdeaf7e | 2017-07-24 13:40:03 +0200 | [diff] [blame] | 4233 | if (type == KVM_EVENT_CREATE_VM) { |
Claudio Imbrenda | 286de8f | 2017-07-12 17:56:44 +0200 | [diff] [blame] | 4234 | add_uevent_var(env, "EVENT=create"); |
Claudio Imbrenda | fdeaf7e | 2017-07-24 13:40:03 +0200 | [diff] [blame] | 4235 | kvm->userspace_pid = task_pid_nr(current); |
| 4236 | } else if (type == KVM_EVENT_DESTROY_VM) { |
Claudio Imbrenda | 286de8f | 2017-07-12 17:56:44 +0200 | [diff] [blame] | 4237 | add_uevent_var(env, "EVENT=destroy"); |
Claudio Imbrenda | fdeaf7e | 2017-07-24 13:40:03 +0200 | [diff] [blame] | 4238 | } |
| 4239 | add_uevent_var(env, "PID=%d", kvm->userspace_pid); |
Claudio Imbrenda | 286de8f | 2017-07-12 17:56:44 +0200 | [diff] [blame] | 4240 | |
Greg Kroah-Hartman | 8ed0579 | 2019-02-28 16:34:37 +0100 | [diff] [blame] | 4241 | if (!IS_ERR_OR_NULL(kvm->debugfs_dentry)) { |
Ben Gardon | b12ce36 | 2019-02-11 11:02:49 -0800 | [diff] [blame] | 4242 | char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT); |
Claudio Imbrenda | 286de8f | 2017-07-12 17:56:44 +0200 | [diff] [blame] | 4243 | |
Claudio Imbrenda | fdeaf7e | 2017-07-24 13:40:03 +0200 | [diff] [blame] | 4244 | if (p) { |
| 4245 | tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX); |
| 4246 | if (!IS_ERR(tmp)) |
| 4247 | add_uevent_var(env, "STATS_PATH=%s", tmp); |
| 4248 | kfree(p); |
Claudio Imbrenda | 286de8f | 2017-07-12 17:56:44 +0200 | [diff] [blame] | 4249 | } |
| 4250 | } |
| 4251 | /* no need for checks, since we are adding at most only 5 keys */ |
| 4252 | env->envp[env->envp_idx++] = NULL; |
| 4253 | kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp); |
| 4254 | kfree(env); |
Claudio Imbrenda | 286de8f | 2017-07-12 17:56:44 +0200 | [diff] [blame] | 4255 | } |
| 4256 | |
Greg Kroah-Hartman | 929f45e | 2018-05-29 18:22:04 +0200 | [diff] [blame] | 4257 | static void kvm_init_debug(void) |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 4258 | { |
| 4259 | struct kvm_stats_debugfs_item *p; |
| 4260 | |
Hollis Blanchard | 76f7c87 | 2008-04-15 16:05:42 -0500 | [diff] [blame] | 4261 | kvm_debugfs_dir = debugfs_create_dir("kvm", NULL); |
Hamo | 4f69b68 | 2011-12-15 14:23:16 +0800 | [diff] [blame] | 4262 | |
Janosch Frank | 536a6f8 | 2016-05-18 13:26:23 +0200 | [diff] [blame] | 4263 | kvm_debugfs_num_entries = 0; |
| 4264 | for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) { |
Paolo Bonzini | 833b45d | 2019-09-30 18:48:44 +0200 | [diff] [blame] | 4265 | int mode = p->mode ? p->mode : 0644; |
| 4266 | debugfs_create_file(p->name, mode, kvm_debugfs_dir, |
Greg Kroah-Hartman | 929f45e | 2018-05-29 18:22:04 +0200 | [diff] [blame] | 4267 | (void *)(long)p->offset, |
| 4268 | stat_fops[p->kind]); |
Hamo | 4f69b68 | 2011-12-15 14:23:16 +0800 | [diff] [blame] | 4269 | } |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 4270 | } |
| 4271 | |
Rafael J. Wysocki | fb3600c | 2011-03-23 22:16:23 +0100 | [diff] [blame] | 4272 | static int kvm_suspend(void) |
Avi Kivity | 59ae6c6 | 2007-02-12 00:54:48 -0800 | [diff] [blame] | 4273 | { |
Alexander Graf | 10474ae | 2009-09-15 11:37:46 +0200 | [diff] [blame] | 4274 | if (kvm_usage_count) |
Takuya Yoshikawa | 75b7127 | 2010-11-16 17:37:41 +0900 | [diff] [blame] | 4275 | hardware_disable_nolock(NULL); |
Avi Kivity | 59ae6c6 | 2007-02-12 00:54:48 -0800 | [diff] [blame] | 4276 | return 0; |
| 4277 | } |
| 4278 | |
Rafael J. Wysocki | fb3600c | 2011-03-23 22:16:23 +0100 | [diff] [blame] | 4279 | static void kvm_resume(void) |
Avi Kivity | 59ae6c6 | 2007-02-12 00:54:48 -0800 | [diff] [blame] | 4280 | { |
Zachary Amsden | ca84d1a | 2010-08-19 22:07:28 -1000 | [diff] [blame] | 4281 | if (kvm_usage_count) { |
Wanpeng Li | 2eb06c3 | 2019-05-17 16:49:49 +0800 | [diff] [blame] | 4282 | #ifdef CONFIG_LOCKDEP |
| 4283 | WARN_ON(lockdep_is_held(&kvm_count_lock)); |
| 4284 | #endif |
Takuya Yoshikawa | 75b7127 | 2010-11-16 17:37:41 +0900 | [diff] [blame] | 4285 | hardware_enable_nolock(NULL); |
Zachary Amsden | ca84d1a | 2010-08-19 22:07:28 -1000 | [diff] [blame] | 4286 | } |
Avi Kivity | 59ae6c6 | 2007-02-12 00:54:48 -0800 | [diff] [blame] | 4287 | } |
| 4288 | |
Rafael J. Wysocki | fb3600c | 2011-03-23 22:16:23 +0100 | [diff] [blame] | 4289 | static struct syscore_ops kvm_syscore_ops = { |
Avi Kivity | 59ae6c6 | 2007-02-12 00:54:48 -0800 | [diff] [blame] | 4290 | .suspend = kvm_suspend, |
| 4291 | .resume = kvm_resume, |
| 4292 | }; |
| 4293 | |
Avi Kivity | 15ad714 | 2007-07-11 18:17:21 +0300 | [diff] [blame] | 4294 | static inline |
| 4295 | struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn) |
| 4296 | { |
| 4297 | return container_of(pn, struct kvm_vcpu, preempt_notifier); |
| 4298 | } |
| 4299 | |
| 4300 | static void kvm_sched_in(struct preempt_notifier *pn, int cpu) |
| 4301 | { |
| 4302 | struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn); |
Xiubo Li | f95ef0cd | 2015-02-26 14:58:23 +0800 | [diff] [blame] | 4303 | |
Wanpeng Li | 046ddee | 2019-08-01 11:30:14 +0800 | [diff] [blame] | 4304 | WRITE_ONCE(vcpu->preempted, false); |
Wanpeng Li | d73eb57 | 2019-07-18 19:39:06 +0800 | [diff] [blame] | 4305 | WRITE_ONCE(vcpu->ready, false); |
Avi Kivity | 15ad714 | 2007-07-11 18:17:21 +0300 | [diff] [blame] | 4306 | |
Radim Krčmář | e790d9e | 2014-08-21 18:08:05 +0200 | [diff] [blame] | 4307 | kvm_arch_sched_in(vcpu, cpu); |
| 4308 | |
Zhang Xiantao | e9b11c1 | 2007-11-14 20:38:21 +0800 | [diff] [blame] | 4309 | kvm_arch_vcpu_load(vcpu, cpu); |
Avi Kivity | 15ad714 | 2007-07-11 18:17:21 +0300 | [diff] [blame] | 4310 | } |
| 4311 | |
| 4312 | static void kvm_sched_out(struct preempt_notifier *pn, |
| 4313 | struct task_struct *next) |
| 4314 | { |
| 4315 | struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn); |
| 4316 | |
Wanpeng Li | d73eb57 | 2019-07-18 19:39:06 +0800 | [diff] [blame] | 4317 | if (current->state == TASK_RUNNING) { |
Wanpeng Li | 046ddee | 2019-08-01 11:30:14 +0800 | [diff] [blame] | 4318 | WRITE_ONCE(vcpu->preempted, true); |
Wanpeng Li | d73eb57 | 2019-07-18 19:39:06 +0800 | [diff] [blame] | 4319 | WRITE_ONCE(vcpu->ready, true); |
| 4320 | } |
Zhang Xiantao | e9b11c1 | 2007-11-14 20:38:21 +0800 | [diff] [blame] | 4321 | kvm_arch_vcpu_put(vcpu); |
Avi Kivity | 15ad714 | 2007-07-11 18:17:21 +0300 | [diff] [blame] | 4322 | } |
| 4323 | |
Sean Christopherson | f257d6d | 2019-04-19 22:18:17 -0700 | [diff] [blame] | 4324 | static void check_processor_compat(void *rtn) |
| 4325 | { |
| 4326 | *(int *)rtn = kvm_arch_check_processor_compat(); |
| 4327 | } |
| 4328 | |
Avi Kivity | 0ee75be | 2010-04-28 15:39:01 +0300 | [diff] [blame] | 4329 | int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align, |
Rusty Russell | c16f862 | 2007-07-30 21:12:19 +1000 | [diff] [blame] | 4330 | struct module *module) |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 4331 | { |
| 4332 | int r; |
Yang, Sheng | 002c7f7 | 2007-07-31 14:23:01 +0300 | [diff] [blame] | 4333 | int cpu; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 4334 | |
Zhang Xiantao | f8c16bb | 2007-11-14 20:40:21 +0800 | [diff] [blame] | 4335 | r = kvm_arch_init(opaque); |
| 4336 | if (r) |
Zhang Xiantao | d2308784 | 2007-11-29 15:35:39 +0800 | [diff] [blame] | 4337 | goto out_fail; |
Zhang Xiantao | cb498ea | 2007-11-14 20:39:31 +0800 | [diff] [blame] | 4338 | |
Asias He | 7dac16c | 2013-05-08 10:57:29 +0800 | [diff] [blame] | 4339 | /* |
| 4340 | * kvm_arch_init makes sure there's at most one caller |
| 4341 | * for architectures that support multiple implementations, |
| 4342 | * like intel and amd on x86. |
Paolo Bonzini | 36343f6 | 2016-10-26 13:35:56 +0200 | [diff] [blame] | 4343 | * kvm_arch_init must be called before kvm_irqfd_init to avoid creating |
| 4344 | * conflicts in case kvm is already setup for another implementation. |
Asias He | 7dac16c | 2013-05-08 10:57:29 +0800 | [diff] [blame] | 4345 | */ |
Paolo Bonzini | 36343f6 | 2016-10-26 13:35:56 +0200 | [diff] [blame] | 4346 | r = kvm_irqfd_init(); |
| 4347 | if (r) |
| 4348 | goto out_irqfd; |
Asias He | 7dac16c | 2013-05-08 10:57:29 +0800 | [diff] [blame] | 4349 | |
Avi Kivity | 8437a617 | 2009-06-06 14:52:35 -0700 | [diff] [blame] | 4350 | if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) { |
Rusty Russell | 7f59f49 | 2008-12-07 21:25:45 +1030 | [diff] [blame] | 4351 | r = -ENOMEM; |
| 4352 | goto out_free_0; |
| 4353 | } |
| 4354 | |
Zhang Xiantao | e9b11c1 | 2007-11-14 20:38:21 +0800 | [diff] [blame] | 4355 | r = kvm_arch_hardware_setup(); |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 4356 | if (r < 0) |
Miaohe Lin | faf0be2 | 2019-11-23 10:45:50 +0800 | [diff] [blame] | 4357 | goto out_free_1; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 4358 | |
Yang, Sheng | 002c7f7 | 2007-07-31 14:23:01 +0300 | [diff] [blame] | 4359 | for_each_online_cpu(cpu) { |
Sean Christopherson | f257d6d | 2019-04-19 22:18:17 -0700 | [diff] [blame] | 4360 | smp_call_function_single(cpu, check_processor_compat, &r, 1); |
Yang, Sheng | 002c7f7 | 2007-07-31 14:23:01 +0300 | [diff] [blame] | 4361 | if (r < 0) |
Miaohe Lin | faf0be2 | 2019-11-23 10:45:50 +0800 | [diff] [blame] | 4362 | goto out_free_2; |
Yang, Sheng | 002c7f7 | 2007-07-31 14:23:01 +0300 | [diff] [blame] | 4363 | } |
| 4364 | |
Thomas Gleixner | 73c1b41 | 2016-12-21 20:19:54 +0100 | [diff] [blame] | 4365 | r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting", |
Thomas Gleixner | 8c18b2d | 2016-07-13 17:16:37 +0000 | [diff] [blame] | 4366 | kvm_starting_cpu, kvm_dying_cpu); |
Avi Kivity | 774c47f | 2007-02-12 00:54:47 -0800 | [diff] [blame] | 4367 | if (r) |
Zhang Xiantao | d2308784 | 2007-11-29 15:35:39 +0800 | [diff] [blame] | 4368 | goto out_free_2; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 4369 | register_reboot_notifier(&kvm_reboot_notifier); |
| 4370 | |
Rusty Russell | c16f862 | 2007-07-30 21:12:19 +1000 | [diff] [blame] | 4371 | /* A kmem cache lets us meet the alignment requirements of fx_save. */ |
Avi Kivity | 0ee75be | 2010-04-28 15:39:01 +0300 | [diff] [blame] | 4372 | if (!vcpu_align) |
| 4373 | vcpu_align = __alignof__(struct kvm_vcpu); |
Paolo Bonzini | 4651573 | 2017-10-26 15:45:46 +0200 | [diff] [blame] | 4374 | kvm_vcpu_cache = |
| 4375 | kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align, |
| 4376 | SLAB_ACCOUNT, |
| 4377 | offsetof(struct kvm_vcpu, arch), |
| 4378 | sizeof_field(struct kvm_vcpu, arch), |
| 4379 | NULL); |
Rusty Russell | c16f862 | 2007-07-30 21:12:19 +1000 | [diff] [blame] | 4380 | if (!kvm_vcpu_cache) { |
| 4381 | r = -ENOMEM; |
Rafael J. Wysocki | fb3600c | 2011-03-23 22:16:23 +0100 | [diff] [blame] | 4382 | goto out_free_3; |
Rusty Russell | c16f862 | 2007-07-30 21:12:19 +1000 | [diff] [blame] | 4383 | } |
| 4384 | |
Gleb Natapov | af585b9 | 2010-10-14 11:22:46 +0200 | [diff] [blame] | 4385 | r = kvm_async_pf_init(); |
| 4386 | if (r) |
| 4387 | goto out_free; |
| 4388 | |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 4389 | kvm_chardev_ops.owner = module; |
Christian Borntraeger | 3d3aab1 | 2008-12-02 11:17:32 +0100 | [diff] [blame] | 4390 | kvm_vm_fops.owner = module; |
| 4391 | kvm_vcpu_fops.owner = module; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 4392 | |
| 4393 | r = misc_register(&kvm_dev); |
| 4394 | if (r) { |
Xiubo Li | 1170adc | 2015-02-26 14:58:26 +0800 | [diff] [blame] | 4395 | pr_err("kvm: misc device register failed\n"); |
Gleb Natapov | af585b9 | 2010-10-14 11:22:46 +0200 | [diff] [blame] | 4396 | goto out_unreg; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 4397 | } |
| 4398 | |
Rafael J. Wysocki | fb3600c | 2011-03-23 22:16:23 +0100 | [diff] [blame] | 4399 | register_syscore_ops(&kvm_syscore_ops); |
| 4400 | |
Avi Kivity | 15ad714 | 2007-07-11 18:17:21 +0300 | [diff] [blame] | 4401 | kvm_preempt_ops.sched_in = kvm_sched_in; |
| 4402 | kvm_preempt_ops.sched_out = kvm_sched_out; |
| 4403 | |
Greg Kroah-Hartman | 929f45e | 2018-05-29 18:22:04 +0200 | [diff] [blame] | 4404 | kvm_init_debug(); |
Darrick J. Wong | 0ea4ed8 | 2009-10-14 16:21:00 -0700 | [diff] [blame] | 4405 | |
Paolo Bonzini | 3c3c29f | 2014-09-24 13:02:46 +0200 | [diff] [blame] | 4406 | r = kvm_vfio_ops_init(); |
| 4407 | WARN_ON(r); |
| 4408 | |
Avi Kivity | c7addb9 | 2007-09-16 18:58:32 +0200 | [diff] [blame] | 4409 | return 0; |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 4410 | |
Gleb Natapov | af585b9 | 2010-10-14 11:22:46 +0200 | [diff] [blame] | 4411 | out_unreg: |
| 4412 | kvm_async_pf_deinit(); |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 4413 | out_free: |
Rusty Russell | c16f862 | 2007-07-30 21:12:19 +1000 | [diff] [blame] | 4414 | kmem_cache_destroy(kvm_vcpu_cache); |
Zhang Xiantao | d2308784 | 2007-11-29 15:35:39 +0800 | [diff] [blame] | 4415 | out_free_3: |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 4416 | unregister_reboot_notifier(&kvm_reboot_notifier); |
Thomas Gleixner | 8c18b2d | 2016-07-13 17:16:37 +0000 | [diff] [blame] | 4417 | cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING); |
Zhang Xiantao | d2308784 | 2007-11-29 15:35:39 +0800 | [diff] [blame] | 4418 | out_free_2: |
Zhang Xiantao | e9b11c1 | 2007-11-14 20:38:21 +0800 | [diff] [blame] | 4419 | kvm_arch_hardware_unsetup(); |
Miaohe Lin | faf0be2 | 2019-11-23 10:45:50 +0800 | [diff] [blame] | 4420 | out_free_1: |
Rusty Russell | 7f59f49 | 2008-12-07 21:25:45 +1030 | [diff] [blame] | 4421 | free_cpumask_var(cpus_hardware_enabled); |
Zhang Xiantao | d2308784 | 2007-11-29 15:35:39 +0800 | [diff] [blame] | 4422 | out_free_0: |
Cornelia Huck | a0f155e | 2013-02-28 12:33:18 +0100 | [diff] [blame] | 4423 | kvm_irqfd_exit(); |
Paolo Bonzini | 36343f6 | 2016-10-26 13:35:56 +0200 | [diff] [blame] | 4424 | out_irqfd: |
Asias He | 7dac16c | 2013-05-08 10:57:29 +0800 | [diff] [blame] | 4425 | kvm_arch_exit(); |
| 4426 | out_fail: |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 4427 | return r; |
| 4428 | } |
Zhang Xiantao | cb498ea | 2007-11-14 20:39:31 +0800 | [diff] [blame] | 4429 | EXPORT_SYMBOL_GPL(kvm_init); |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 4430 | |
Zhang Xiantao | cb498ea | 2007-11-14 20:39:31 +0800 | [diff] [blame] | 4431 | void kvm_exit(void) |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 4432 | { |
Janosch Frank | 4bd33b5 | 2015-10-14 12:37:35 +0200 | [diff] [blame] | 4433 | debugfs_remove_recursive(kvm_debugfs_dir); |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 4434 | misc_deregister(&kvm_dev); |
Rusty Russell | c16f862 | 2007-07-30 21:12:19 +1000 | [diff] [blame] | 4435 | kmem_cache_destroy(kvm_vcpu_cache); |
Gleb Natapov | af585b9 | 2010-10-14 11:22:46 +0200 | [diff] [blame] | 4436 | kvm_async_pf_deinit(); |
Rafael J. Wysocki | fb3600c | 2011-03-23 22:16:23 +0100 | [diff] [blame] | 4437 | unregister_syscore_ops(&kvm_syscore_ops); |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 4438 | unregister_reboot_notifier(&kvm_reboot_notifier); |
Thomas Gleixner | 8c18b2d | 2016-07-13 17:16:37 +0000 | [diff] [blame] | 4439 | cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING); |
Takuya Yoshikawa | 75b7127 | 2010-11-16 17:37:41 +0900 | [diff] [blame] | 4440 | on_each_cpu(hardware_disable_nolock, NULL, 1); |
Zhang Xiantao | e9b11c1 | 2007-11-14 20:38:21 +0800 | [diff] [blame] | 4441 | kvm_arch_hardware_unsetup(); |
Zhang Xiantao | f8c16bb | 2007-11-14 20:40:21 +0800 | [diff] [blame] | 4442 | kvm_arch_exit(); |
Cornelia Huck | a0f155e | 2013-02-28 12:33:18 +0100 | [diff] [blame] | 4443 | kvm_irqfd_exit(); |
Rusty Russell | 7f59f49 | 2008-12-07 21:25:45 +1030 | [diff] [blame] | 4444 | free_cpumask_var(cpus_hardware_enabled); |
Wanpeng Li | 571ee1b | 2014-10-09 18:30:08 +0800 | [diff] [blame] | 4445 | kvm_vfio_ops_exit(); |
Avi Kivity | 6aa8b73 | 2006-12-10 02:21:36 -0800 | [diff] [blame] | 4446 | } |
Zhang Xiantao | cb498ea | 2007-11-14 20:39:31 +0800 | [diff] [blame] | 4447 | EXPORT_SYMBOL_GPL(kvm_exit); |
Junaid Shahid | c57c804 | 2019-11-04 12:22:02 +0100 | [diff] [blame] | 4448 | |
| 4449 | struct kvm_vm_worker_thread_context { |
| 4450 | struct kvm *kvm; |
| 4451 | struct task_struct *parent; |
| 4452 | struct completion init_done; |
| 4453 | kvm_vm_thread_fn_t thread_fn; |
| 4454 | uintptr_t data; |
| 4455 | int err; |
| 4456 | }; |
| 4457 | |
| 4458 | static int kvm_vm_worker_thread(void *context) |
| 4459 | { |
| 4460 | /* |
| 4461 | * The init_context is allocated on the stack of the parent thread, so |
| 4462 | * we have to locally copy anything that is needed beyond initialization |
| 4463 | */ |
| 4464 | struct kvm_vm_worker_thread_context *init_context = context; |
| 4465 | struct kvm *kvm = init_context->kvm; |
| 4466 | kvm_vm_thread_fn_t thread_fn = init_context->thread_fn; |
| 4467 | uintptr_t data = init_context->data; |
| 4468 | int err; |
| 4469 | |
| 4470 | err = kthread_park(current); |
| 4471 | /* kthread_park(current) is never supposed to return an error */ |
| 4472 | WARN_ON(err != 0); |
| 4473 | if (err) |
| 4474 | goto init_complete; |
| 4475 | |
| 4476 | err = cgroup_attach_task_all(init_context->parent, current); |
| 4477 | if (err) { |
| 4478 | kvm_err("%s: cgroup_attach_task_all failed with err %d\n", |
| 4479 | __func__, err); |
| 4480 | goto init_complete; |
| 4481 | } |
| 4482 | |
| 4483 | set_user_nice(current, task_nice(init_context->parent)); |
| 4484 | |
| 4485 | init_complete: |
| 4486 | init_context->err = err; |
| 4487 | complete(&init_context->init_done); |
| 4488 | init_context = NULL; |
| 4489 | |
| 4490 | if (err) |
| 4491 | return err; |
| 4492 | |
| 4493 | /* Wait to be woken up by the spawner before proceeding. */ |
| 4494 | kthread_parkme(); |
| 4495 | |
| 4496 | if (!kthread_should_stop()) |
| 4497 | err = thread_fn(kvm, data); |
| 4498 | |
| 4499 | return err; |
| 4500 | } |
| 4501 | |
| 4502 | int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn, |
| 4503 | uintptr_t data, const char *name, |
| 4504 | struct task_struct **thread_ptr) |
| 4505 | { |
| 4506 | struct kvm_vm_worker_thread_context init_context = {}; |
| 4507 | struct task_struct *thread; |
| 4508 | |
| 4509 | *thread_ptr = NULL; |
| 4510 | init_context.kvm = kvm; |
| 4511 | init_context.parent = current; |
| 4512 | init_context.thread_fn = thread_fn; |
| 4513 | init_context.data = data; |
| 4514 | init_completion(&init_context.init_done); |
| 4515 | |
| 4516 | thread = kthread_run(kvm_vm_worker_thread, &init_context, |
| 4517 | "%s-%d", name, task_pid_nr(current)); |
| 4518 | if (IS_ERR(thread)) |
| 4519 | return PTR_ERR(thread); |
| 4520 | |
| 4521 | /* kthread_run is never supposed to return NULL */ |
| 4522 | WARN_ON(thread == NULL); |
| 4523 | |
| 4524 | wait_for_completion(&init_context.init_done); |
| 4525 | |
| 4526 | if (!init_context.err) |
| 4527 | *thread_ptr = thread; |
| 4528 | |
| 4529 | return init_context.err; |
| 4530 | } |