]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - arch/x86/kvm/vmx.c
projects / mirror_ubuntu-artful-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
KVM: x86: lapic: Clean up find_highest_vector() and count_vectors()
[mirror_ubuntu-artful-kernel.git] / arch / x86 / kvm / vmx.c
CommitLineData
6aa8b732
AK		 1/*
2 * Kernel-based Virtual Machine driver for Linux
3 *
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
6 *
7 * Copyright (C) 2006 Qumranet, Inc.
9611c187 8 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
6aa8b732
AK		 9 *
10 * Authors:
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
13 *
14 * This work is licensed under the terms of the GNU GPL, version 2. See
15 * the COPYING file in the top-level directory.
16 *
17 */
18
85f455f7 19#include "irq.h"
1d737c8a 20#include "mmu.h"
00b27a3e 21#include "cpuid.h"
e495606d 22
edf88417 23#include <linux/kvm_host.h>
6aa8b732 24#include <linux/module.h>
9d8f549d 25#include <linux/kernel.h>
6aa8b732
AK		 26#include <linux/mm.h>
27#include <linux/highmem.h>
e8edc6e0 28#include <linux/sched.h>
c7addb90 29#include <linux/moduleparam.h>
e9bda3b3 30#include <linux/mod_devicetable.h>
229456fc 31#include <linux/ftrace_event.h>
5a0e3ad6 32#include <linux/slab.h>
cafd6659 33#include <linux/tboot.h>
5fdbf976 34#include "kvm_cache_regs.h"
35920a35 35#include "x86.h"
e495606d 36
6aa8b732 37#include <asm/io.h>
3b3be0d1 38#include <asm/desc.h>
13673a90 39#include <asm/vmx.h>
6210e37b 40#include <asm/virtext.h>
a0861c02 41#include <asm/mce.h>
2acf923e
DC		 42#include <asm/i387.h>
43#include <asm/xcr.h>
d7cd9796 44#include <asm/perf_event.h>
6aa8b732 45
229456fc
MT		 46#include "trace.h"
47
4ecac3fd 48#define __ex(x) __kvm_handle_fault_on_reboot(x)
5e520e62
AK		 49#define __ex_clear(x, reg) \
50 ____kvm_handle_fault_on_reboot(x, "xor " reg " , " reg)
4ecac3fd 51
6aa8b732
AK		 52MODULE_AUTHOR("Qumranet");
53MODULE_LICENSE("GPL");
54
e9bda3b3
JT		 55static const struct x86_cpu_id vmx_cpu_id[] = {
56 X86_FEATURE_MATCH(X86_FEATURE_VMX),
57 {}
58};
59MODULE_DEVICE_TABLE(x86cpu, vmx_cpu_id);
60
476bc001 61static bool __read_mostly enable_vpid = 1;
736caefe 62module_param_named(vpid, enable_vpid, bool, 0444);
2384d2b3 63
476bc001 64static bool __read_mostly flexpriority_enabled = 1;
736caefe 65module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
4c9fc8ef 66
476bc001 67static bool __read_mostly enable_ept = 1;
736caefe 68module_param_named(ept, enable_ept, bool, S_IRUGO);
d56f546d 69
476bc001 70static bool __read_mostly enable_unrestricted_guest = 1;
3a624e29
NK		 71module_param_named(unrestricted_guest,
72 enable_unrestricted_guest, bool, S_IRUGO);
73
83c3a331
XH		 74static bool __read_mostly enable_ept_ad_bits = 1;
75module_param_named(eptad, enable_ept_ad_bits, bool, S_IRUGO);
76
a27685c3 77static bool __read_mostly emulate_invalid_guest_state = true;
c1f8bc04 78module_param(emulate_invalid_guest_state, bool, S_IRUGO);
04fa4d32 79
476bc001 80static bool __read_mostly vmm_exclusive = 1;
b923e62e
DX		 81module_param(vmm_exclusive, bool, S_IRUGO);
82
476bc001 83static bool __read_mostly fasteoi = 1;
58fbbf26
KT		 84module_param(fasteoi, bool, S_IRUGO);
85
801d3424
NHE		 86/*
87 * If nested=1, nested virtualization is supported, i.e., guests may use
88 * VMX and be a hypervisor for its own guests. If nested=0, guests may not
89 * use VMX instructions.
90 */
476bc001 91static bool __read_mostly nested = 0;
801d3424
NHE		 92module_param(nested, bool, S_IRUGO);
93
cdc0e244
AK		 94#define KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST \
95 (X86_CR0_WP | X86_CR0_NE | X86_CR0_NW | X86_CR0_CD)
96#define KVM_GUEST_CR0_MASK \
97 (KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
98#define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST \
81231c69 99 (X86_CR0_WP | X86_CR0_NE)
cdc0e244
AK		 100#define KVM_VM_CR0_ALWAYS_ON \
101 (KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
4c38609a
AK		 102#define KVM_CR4_GUEST_OWNED_BITS \
103 (X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR \
104 | X86_CR4_OSXMMEXCPT)
105
cdc0e244
AK		 106#define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
107#define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
108
78ac8b47
AK		 109#define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
110
4b8d54f9
ZE		 111/*
112 * These 2 parameters are used to config the controls for Pause-Loop Exiting:
113 * ple_gap: upper bound on the amount of time between two successive
114 * executions of PAUSE in a loop. Also indicate if ple enabled.
00c25bce 115 * According to test, this time is usually smaller than 128 cycles.
4b8d54f9
ZE		 116 * ple_window: upper bound on the amount of time a guest is allowed to execute
117 * in a PAUSE loop. Tests indicate that most spinlocks are held for
118 * less than 2^12 cycles
119 * Time is measured based on a counter that runs at the same rate as the TSC,
120 * refer SDM volume 3b section 21.6.13 & 22.1.3.
121 */
00c25bce 122#define KVM_VMX_DEFAULT_PLE_GAP 128
4b8d54f9
ZE		 123#define KVM_VMX_DEFAULT_PLE_WINDOW 4096
124static int ple_gap = KVM_VMX_DEFAULT_PLE_GAP;
125module_param(ple_gap, int, S_IRUGO);
126
127static int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
128module_param(ple_window, int, S_IRUGO);
129
8bf00a52 130#define NR_AUTOLOAD_MSRS 8
ff2f6fe9 131#define VMCS02_POOL_SIZE 1
61d2ef2c 132
a2fa3e9f
GH		 133struct vmcs {
134 u32 revision_id;
135 u32 abort;
136 char data[0];
137};
138
d462b819
NHE		 139/*
140 * Track a VMCS that may be loaded on a certain CPU. If it is (cpu!=-1), also
141 * remember whether it was VMLAUNCHed, and maintain a linked list of all VMCSs
142 * loaded on this CPU (so we can clear them if the CPU goes down).
143 */
144struct loaded_vmcs {
145 struct vmcs *vmcs;
146 int cpu;
147 int launched;
148 struct list_head loaded_vmcss_on_cpu_link;
149};
150
26bb0981
AK		 151struct shared_msr_entry {
152 unsigned index;
153 u64 data;
d5696725 154 u64 mask;
26bb0981
AK		 155};
156
a9d30f33
NHE		 157/*
158 * struct vmcs12 describes the state that our guest hypervisor (L1) keeps for a
159 * single nested guest (L2), hence the name vmcs12. Any VMX implementation has
160 * a VMCS structure, and vmcs12 is our emulated VMX's VMCS. This structure is
161 * stored in guest memory specified by VMPTRLD, but is opaque to the guest,
162 * which must access it using VMREAD/VMWRITE/VMCLEAR instructions.
163 * More than one of these structures may exist, if L1 runs multiple L2 guests.
164 * nested_vmx_run() will use the data here to build a vmcs02: a VMCS for the
165 * underlying hardware which will be used to run L2.
166 * This structure is packed to ensure that its layout is identical across
167 * machines (necessary for live migration).
168 * If there are changes in this struct, VMCS12_REVISION must be changed.
169 */
22bd0358 170typedef u64 natural_width;
a9d30f33
NHE		 171struct __packed vmcs12 {
172 /* According to the Intel spec, a VMCS region must start with the
173 * following two fields. Then follow implementation-specific data.
174 */
175 u32 revision_id;
176 u32 abort;
22bd0358 177
27d6c865
NHE		 178 u32 launch_state; /* set to 0 by VMCLEAR, to 1 by VMLAUNCH */
179 u32 padding[7]; /* room for future expansion */
180
22bd0358
NHE		 181 u64 io_bitmap_a;
182 u64 io_bitmap_b;
183 u64 msr_bitmap;
184 u64 vm_exit_msr_store_addr;
185 u64 vm_exit_msr_load_addr;
186 u64 vm_entry_msr_load_addr;
187 u64 tsc_offset;
188 u64 virtual_apic_page_addr;
189 u64 apic_access_addr;
190 u64 ept_pointer;
191 u64 guest_physical_address;
192 u64 vmcs_link_pointer;
193 u64 guest_ia32_debugctl;
194 u64 guest_ia32_pat;
195 u64 guest_ia32_efer;
196 u64 guest_ia32_perf_global_ctrl;
197 u64 guest_pdptr0;
198 u64 guest_pdptr1;
199 u64 guest_pdptr2;
200 u64 guest_pdptr3;
201 u64 host_ia32_pat;
202 u64 host_ia32_efer;
203 u64 host_ia32_perf_global_ctrl;
204 u64 padding64[8]; /* room for future expansion */
205 /*
206 * To allow migration of L1 (complete with its L2 guests) between
207 * machines of different natural widths (32 or 64 bit), we cannot have
208 * unsigned long fields with no explict size. We use u64 (aliased
209 * natural_width) instead. Luckily, x86 is little-endian.
210 */
211 natural_width cr0_guest_host_mask;
212 natural_width cr4_guest_host_mask;
213 natural_width cr0_read_shadow;
214 natural_width cr4_read_shadow;
215 natural_width cr3_target_value0;
216 natural_width cr3_target_value1;
217 natural_width cr3_target_value2;
218 natural_width cr3_target_value3;
219 natural_width exit_qualification;
220 natural_width guest_linear_address;
221 natural_width guest_cr0;
222 natural_width guest_cr3;
223 natural_width guest_cr4;
224 natural_width guest_es_base;
225 natural_width guest_cs_base;
226 natural_width guest_ss_base;
227 natural_width guest_ds_base;
228 natural_width guest_fs_base;
229 natural_width guest_gs_base;
230 natural_width guest_ldtr_base;
231 natural_width guest_tr_base;
232 natural_width guest_gdtr_base;
233 natural_width guest_idtr_base;
234 natural_width guest_dr7;
235 natural_width guest_rsp;
236 natural_width guest_rip;
237 natural_width guest_rflags;
238 natural_width guest_pending_dbg_exceptions;
239 natural_width guest_sysenter_esp;
240 natural_width guest_sysenter_eip;
241 natural_width host_cr0;
242 natural_width host_cr3;
243 natural_width host_cr4;
244 natural_width host_fs_base;
245 natural_width host_gs_base;
246 natural_width host_tr_base;
247 natural_width host_gdtr_base;
248 natural_width host_idtr_base;
249 natural_width host_ia32_sysenter_esp;
250 natural_width host_ia32_sysenter_eip;
251 natural_width host_rsp;
252 natural_width host_rip;
253 natural_width paddingl[8]; /* room for future expansion */
254 u32 pin_based_vm_exec_control;
255 u32 cpu_based_vm_exec_control;
256 u32 exception_bitmap;
257 u32 page_fault_error_code_mask;
258 u32 page_fault_error_code_match;
259 u32 cr3_target_count;
260 u32 vm_exit_controls;
261 u32 vm_exit_msr_store_count;
262 u32 vm_exit_msr_load_count;
263 u32 vm_entry_controls;
264 u32 vm_entry_msr_load_count;
265 u32 vm_entry_intr_info_field;
266 u32 vm_entry_exception_error_code;
267 u32 vm_entry_instruction_len;
268 u32 tpr_threshold;
269 u32 secondary_vm_exec_control;
270 u32 vm_instruction_error;
271 u32 vm_exit_reason;
272 u32 vm_exit_intr_info;
273 u32 vm_exit_intr_error_code;
274 u32 idt_vectoring_info_field;
275 u32 idt_vectoring_error_code;
276 u32 vm_exit_instruction_len;
277 u32 vmx_instruction_info;
278 u32 guest_es_limit;
279 u32 guest_cs_limit;
280 u32 guest_ss_limit;
281 u32 guest_ds_limit;
282 u32 guest_fs_limit;
283 u32 guest_gs_limit;
284 u32 guest_ldtr_limit;
285 u32 guest_tr_limit;
286 u32 guest_gdtr_limit;
287 u32 guest_idtr_limit;
288 u32 guest_es_ar_bytes;
289 u32 guest_cs_ar_bytes;
290 u32 guest_ss_ar_bytes;
291 u32 guest_ds_ar_bytes;
292 u32 guest_fs_ar_bytes;
293 u32 guest_gs_ar_bytes;
294 u32 guest_ldtr_ar_bytes;
295 u32 guest_tr_ar_bytes;
296 u32 guest_interruptibility_info;
297 u32 guest_activity_state;
298 u32 guest_sysenter_cs;
299 u32 host_ia32_sysenter_cs;
300 u32 padding32[8]; /* room for future expansion */
301 u16 virtual_processor_id;
302 u16 guest_es_selector;
303 u16 guest_cs_selector;
304 u16 guest_ss_selector;
305 u16 guest_ds_selector;
306 u16 guest_fs_selector;
307 u16 guest_gs_selector;
308 u16 guest_ldtr_selector;
309 u16 guest_tr_selector;
310 u16 host_es_selector;
311 u16 host_cs_selector;
312 u16 host_ss_selector;
313 u16 host_ds_selector;
314 u16 host_fs_selector;
315 u16 host_gs_selector;
316 u16 host_tr_selector;
a9d30f33
NHE		 317};
318
319/*
320 * VMCS12_REVISION is an arbitrary id that should be changed if the content or
321 * layout of struct vmcs12 is changed. MSR_IA32_VMX_BASIC returns this id, and
322 * VMPTRLD verifies that the VMCS region that L1 is loading contains this id.
323 */
324#define VMCS12_REVISION 0x11e57ed0
325
326/*
327 * VMCS12_SIZE is the number of bytes L1 should allocate for the VMXON region
328 * and any VMCS region. Although only sizeof(struct vmcs12) are used by the
329 * current implementation, 4K are reserved to avoid future complications.
330 */
331#define VMCS12_SIZE 0x1000
332
ff2f6fe9
NHE		 333/* Used to remember the last vmcs02 used for some recently used vmcs12s */
334struct vmcs02_list {
335 struct list_head list;
336 gpa_t vmptr;
337 struct loaded_vmcs vmcs02;
338};
339
ec378aee
NHE		 340/*
341 * The nested_vmx structure is part of vcpu_vmx, and holds information we need
342 * for correct emulation of VMX (i.e., nested VMX) on this vcpu.
343 */
344struct nested_vmx {
345 /* Has the level1 guest done vmxon? */
346 bool vmxon;
a9d30f33
NHE		 347
348 /* The guest-physical address of the current VMCS L1 keeps for L2 */
349 gpa_t current_vmptr;
350 /* The host-usable pointer to the above */
351 struct page *current_vmcs12_page;
352 struct vmcs12 *current_vmcs12;
ff2f6fe9
NHE		 353
354 /* vmcs02_list cache of VMCSs recently used to run L2 guests */
355 struct list_head vmcs02_pool;
356 int vmcs02_num;
fe3ef05c 357 u64 vmcs01_tsc_offset;
644d711a
NHE		 358 /* L2 must run next, and mustn't decide to exit to L1. */
359 bool nested_run_pending;
fe3ef05c
NHE		 360 /*
361 * Guest pages referred to in vmcs02 with host-physical pointers, so
362 * we must keep them pinned while L2 runs.
363 */
364 struct page *apic_access_page;
ec378aee
NHE		 365};
366
a2fa3e9f 367struct vcpu_vmx {
fb3f0f51 368 struct kvm_vcpu vcpu;
313dbd49 369 unsigned long host_rsp;
29bd8a78 370 u8 fail;
69c73028 371 u8 cpl;
9d58b931 372 bool nmi_known_unmasked;
51aa01d1 373 u32 exit_intr_info;
1155f76a 374 u32 idt_vectoring_info;
6de12732 375 ulong rflags;
26bb0981 376 struct shared_msr_entry *guest_msrs;
a2fa3e9f
GH		 377 int nmsrs;
378 int save_nmsrs;
a2fa3e9f 379#ifdef CONFIG_X86_64
44ea2b17
AK		 380 u64 msr_host_kernel_gs_base;
381 u64 msr_guest_kernel_gs_base;
a2fa3e9f 382#endif
d462b819
NHE		 383 /*
384 * loaded_vmcs points to the VMCS currently used in this vcpu. For a
385 * non-nested (L1) guest, it always points to vmcs01. For a nested
386 * guest (L2), it points to a different VMCS.
387 */
388 struct loaded_vmcs vmcs01;
389 struct loaded_vmcs *loaded_vmcs;
390 bool __launched; /* temporary, used in vmx_vcpu_run */
61d2ef2c
AK		 391 struct msr_autoload {
392 unsigned nr;
393 struct vmx_msr_entry guest[NR_AUTOLOAD_MSRS];
394 struct vmx_msr_entry host[NR_AUTOLOAD_MSRS];
395 } msr_autoload;
a2fa3e9f
GH		 396 struct {
397 int loaded;
398 u16 fs_sel, gs_sel, ldt_sel;
b2da15ac
AK		 399#ifdef CONFIG_X86_64
400 u16 ds_sel, es_sel;
401#endif
152d3f2f
LV		 402 int gs_ldt_reload_needed;
403 int fs_reload_needed;
d77c26fc 404 } host_state;
9c8cba37 405 struct {
7ffd92c5 406 int vm86_active;
78ac8b47 407 ulong save_rflags;
f5f7b2fe
AK		 408 struct kvm_segment segs[8];
409 } rmode;
410 struct {
411 u32 bitmask; /* 4 bits per segment (1 bit per field) */
7ffd92c5
AK		 412 struct kvm_save_segment {
413 u16 selector;
414 unsigned long base;
415 u32 limit;
416 u32 ar;
f5f7b2fe 417 } seg[8];
2fb92db1 418 } segment_cache;
2384d2b3 419 int vpid;
04fa4d32 420 bool emulation_required;
3b86cd99
JK		 421
422 /* Support for vnmi-less CPUs */
423 int soft_vnmi_blocked;
424 ktime_t entry_time;
425 s64 vnmi_blocked_time;
a0861c02 426 u32 exit_reason;
4e47c7a6
SY		 427
428 bool rdtscp_enabled;
ec378aee
NHE		 429
430 /* Support for a guest hypervisor (nested VMX) */
431 struct nested_vmx nested;
a2fa3e9f
GH		 432};
433
2fb92db1
AK		 434enum segment_cache_field {
435 SEG_FIELD_SEL = 0,
436 SEG_FIELD_BASE = 1,
437 SEG_FIELD_LIMIT = 2,
438 SEG_FIELD_AR = 3,
439
440 SEG_FIELD_NR = 4
441};
442
a2fa3e9f
GH		 443static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
444{
fb3f0f51 445 return container_of(vcpu, struct vcpu_vmx, vcpu);
a2fa3e9f
GH		 446}
447
22bd0358
NHE		 448#define VMCS12_OFFSET(x) offsetof(struct vmcs12, x)
449#define FIELD(number, name) [number] = VMCS12_OFFSET(name)
450#define FIELD64(number, name) [number] = VMCS12_OFFSET(name), \
451 [number##_HIGH] = VMCS12_OFFSET(name)+4
452
772e0318 453static const unsigned short vmcs_field_to_offset_table[] = {
22bd0358
NHE		 454 FIELD(VIRTUAL_PROCESSOR_ID, virtual_processor_id),
455 FIELD(GUEST_ES_SELECTOR, guest_es_selector),
456 FIELD(GUEST_CS_SELECTOR, guest_cs_selector),
457 FIELD(GUEST_SS_SELECTOR, guest_ss_selector),
458 FIELD(GUEST_DS_SELECTOR, guest_ds_selector),
459 FIELD(GUEST_FS_SELECTOR, guest_fs_selector),
460 FIELD(GUEST_GS_SELECTOR, guest_gs_selector),
461 FIELD(GUEST_LDTR_SELECTOR, guest_ldtr_selector),
462 FIELD(GUEST_TR_SELECTOR, guest_tr_selector),
463 FIELD(HOST_ES_SELECTOR, host_es_selector),
464 FIELD(HOST_CS_SELECTOR, host_cs_selector),
465 FIELD(HOST_SS_SELECTOR, host_ss_selector),
466 FIELD(HOST_DS_SELECTOR, host_ds_selector),
467 FIELD(HOST_FS_SELECTOR, host_fs_selector),
468 FIELD(HOST_GS_SELECTOR, host_gs_selector),
469 FIELD(HOST_TR_SELECTOR, host_tr_selector),
470 FIELD64(IO_BITMAP_A, io_bitmap_a),
471 FIELD64(IO_BITMAP_B, io_bitmap_b),
472 FIELD64(MSR_BITMAP, msr_bitmap),
473 FIELD64(VM_EXIT_MSR_STORE_ADDR, vm_exit_msr_store_addr),
474 FIELD64(VM_EXIT_MSR_LOAD_ADDR, vm_exit_msr_load_addr),
475 FIELD64(VM_ENTRY_MSR_LOAD_ADDR, vm_entry_msr_load_addr),
476 FIELD64(TSC_OFFSET, tsc_offset),
477 FIELD64(VIRTUAL_APIC_PAGE_ADDR, virtual_apic_page_addr),
478 FIELD64(APIC_ACCESS_ADDR, apic_access_addr),
479 FIELD64(EPT_POINTER, ept_pointer),
480 FIELD64(GUEST_PHYSICAL_ADDRESS, guest_physical_address),
481 FIELD64(VMCS_LINK_POINTER, vmcs_link_pointer),
482 FIELD64(GUEST_IA32_DEBUGCTL, guest_ia32_debugctl),
483 FIELD64(GUEST_IA32_PAT, guest_ia32_pat),
484 FIELD64(GUEST_IA32_EFER, guest_ia32_efer),
485 FIELD64(GUEST_IA32_PERF_GLOBAL_CTRL, guest_ia32_perf_global_ctrl),
486 FIELD64(GUEST_PDPTR0, guest_pdptr0),
487 FIELD64(GUEST_PDPTR1, guest_pdptr1),
488 FIELD64(GUEST_PDPTR2, guest_pdptr2),
489 FIELD64(GUEST_PDPTR3, guest_pdptr3),
490 FIELD64(HOST_IA32_PAT, host_ia32_pat),
491 FIELD64(HOST_IA32_EFER, host_ia32_efer),
492 FIELD64(HOST_IA32_PERF_GLOBAL_CTRL, host_ia32_perf_global_ctrl),
493 FIELD(PIN_BASED_VM_EXEC_CONTROL, pin_based_vm_exec_control),
494 FIELD(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control),
495 FIELD(EXCEPTION_BITMAP, exception_bitmap),
496 FIELD(PAGE_FAULT_ERROR_CODE_MASK, page_fault_error_code_mask),
497 FIELD(PAGE_FAULT_ERROR_CODE_MATCH, page_fault_error_code_match),
498 FIELD(CR3_TARGET_COUNT, cr3_target_count),
499 FIELD(VM_EXIT_CONTROLS, vm_exit_controls),
500 FIELD(VM_EXIT_MSR_STORE_COUNT, vm_exit_msr_store_count),
501 FIELD(VM_EXIT_MSR_LOAD_COUNT, vm_exit_msr_load_count),
502 FIELD(VM_ENTRY_CONTROLS, vm_entry_controls),
503 FIELD(VM_ENTRY_MSR_LOAD_COUNT, vm_entry_msr_load_count),
504 FIELD(VM_ENTRY_INTR_INFO_FIELD, vm_entry_intr_info_field),
505 FIELD(VM_ENTRY_EXCEPTION_ERROR_CODE, vm_entry_exception_error_code),
506 FIELD(VM_ENTRY_INSTRUCTION_LEN, vm_entry_instruction_len),
507 FIELD(TPR_THRESHOLD, tpr_threshold),
508 FIELD(SECONDARY_VM_EXEC_CONTROL, secondary_vm_exec_control),
509 FIELD(VM_INSTRUCTION_ERROR, vm_instruction_error),
510 FIELD(VM_EXIT_REASON, vm_exit_reason),
511 FIELD(VM_EXIT_INTR_INFO, vm_exit_intr_info),
512 FIELD(VM_EXIT_INTR_ERROR_CODE, vm_exit_intr_error_code),
513 FIELD(IDT_VECTORING_INFO_FIELD, idt_vectoring_info_field),
514 FIELD(IDT_VECTORING_ERROR_CODE, idt_vectoring_error_code),
515 FIELD(VM_EXIT_INSTRUCTION_LEN, vm_exit_instruction_len),
516 FIELD(VMX_INSTRUCTION_INFO, vmx_instruction_info),
517 FIELD(GUEST_ES_LIMIT, guest_es_limit),
518 FIELD(GUEST_CS_LIMIT, guest_cs_limit),
519 FIELD(GUEST_SS_LIMIT, guest_ss_limit),
520 FIELD(GUEST_DS_LIMIT, guest_ds_limit),
521 FIELD(GUEST_FS_LIMIT, guest_fs_limit),
522 FIELD(GUEST_GS_LIMIT, guest_gs_limit),
523 FIELD(GUEST_LDTR_LIMIT, guest_ldtr_limit),
524 FIELD(GUEST_TR_LIMIT, guest_tr_limit),
525 FIELD(GUEST_GDTR_LIMIT, guest_gdtr_limit),
526 FIELD(GUEST_IDTR_LIMIT, guest_idtr_limit),
527 FIELD(GUEST_ES_AR_BYTES, guest_es_ar_bytes),
528 FIELD(GUEST_CS_AR_BYTES, guest_cs_ar_bytes),
529 FIELD(GUEST_SS_AR_BYTES, guest_ss_ar_bytes),
530 FIELD(GUEST_DS_AR_BYTES, guest_ds_ar_bytes),
531 FIELD(GUEST_FS_AR_BYTES, guest_fs_ar_bytes),
532 FIELD(GUEST_GS_AR_BYTES, guest_gs_ar_bytes),
533 FIELD(GUEST_LDTR_AR_BYTES, guest_ldtr_ar_bytes),
534 FIELD(GUEST_TR_AR_BYTES, guest_tr_ar_bytes),
535 FIELD(GUEST_INTERRUPTIBILITY_INFO, guest_interruptibility_info),
536 FIELD(GUEST_ACTIVITY_STATE, guest_activity_state),
537 FIELD(GUEST_SYSENTER_CS, guest_sysenter_cs),
538 FIELD(HOST_IA32_SYSENTER_CS, host_ia32_sysenter_cs),
539 FIELD(CR0_GUEST_HOST_MASK, cr0_guest_host_mask),
540 FIELD(CR4_GUEST_HOST_MASK, cr4_guest_host_mask),
541 FIELD(CR0_READ_SHADOW, cr0_read_shadow),
542 FIELD(CR4_READ_SHADOW, cr4_read_shadow),
543 FIELD(CR3_TARGET_VALUE0, cr3_target_value0),
544 FIELD(CR3_TARGET_VALUE1, cr3_target_value1),
545 FIELD(CR3_TARGET_VALUE2, cr3_target_value2),
546 FIELD(CR3_TARGET_VALUE3, cr3_target_value3),
547 FIELD(EXIT_QUALIFICATION, exit_qualification),
548 FIELD(GUEST_LINEAR_ADDRESS, guest_linear_address),
549 FIELD(GUEST_CR0, guest_cr0),
550 FIELD(GUEST_CR3, guest_cr3),
551 FIELD(GUEST_CR4, guest_cr4),
552 FIELD(GUEST_ES_BASE, guest_es_base),
553 FIELD(GUEST_CS_BASE, guest_cs_base),
554 FIELD(GUEST_SS_BASE, guest_ss_base),
555 FIELD(GUEST_DS_BASE, guest_ds_base),
556 FIELD(GUEST_FS_BASE, guest_fs_base),
557 FIELD(GUEST_GS_BASE, guest_gs_base),
558 FIELD(GUEST_LDTR_BASE, guest_ldtr_base),
559 FIELD(GUEST_TR_BASE, guest_tr_base),
560 FIELD(GUEST_GDTR_BASE, guest_gdtr_base),
561 FIELD(GUEST_IDTR_BASE, guest_idtr_base),
562 FIELD(GUEST_DR7, guest_dr7),
563 FIELD(GUEST_RSP, guest_rsp),
564 FIELD(GUEST_RIP, guest_rip),
565 FIELD(GUEST_RFLAGS, guest_rflags),
566 FIELD(GUEST_PENDING_DBG_EXCEPTIONS, guest_pending_dbg_exceptions),
567 FIELD(GUEST_SYSENTER_ESP, guest_sysenter_esp),
568 FIELD(GUEST_SYSENTER_EIP, guest_sysenter_eip),
569 FIELD(HOST_CR0, host_cr0),
570 FIELD(HOST_CR3, host_cr3),
571 FIELD(HOST_CR4, host_cr4),
572 FIELD(HOST_FS_BASE, host_fs_base),
573 FIELD(HOST_GS_BASE, host_gs_base),
574 FIELD(HOST_TR_BASE, host_tr_base),
575 FIELD(HOST_GDTR_BASE, host_gdtr_base),
576 FIELD(HOST_IDTR_BASE, host_idtr_base),
577 FIELD(HOST_IA32_SYSENTER_ESP, host_ia32_sysenter_esp),
578 FIELD(HOST_IA32_SYSENTER_EIP, host_ia32_sysenter_eip),
579 FIELD(HOST_RSP, host_rsp),
580 FIELD(HOST_RIP, host_rip),
581};
582static const int max_vmcs_field = ARRAY_SIZE(vmcs_field_to_offset_table);
583
584static inline short vmcs_field_to_offset(unsigned long field)
585{
586 if (field >= max_vmcs_field || vmcs_field_to_offset_table[field] == 0)
587 return -1;
588 return vmcs_field_to_offset_table[field];
589}
590
a9d30f33
NHE		 591static inline struct vmcs12 *get_vmcs12(struct kvm_vcpu *vcpu)
592{
593 return to_vmx(vcpu)->nested.current_vmcs12;
594}
595
596static struct page *nested_get_page(struct kvm_vcpu *vcpu, gpa_t addr)
597{
598 struct page *page = gfn_to_page(vcpu->kvm, addr >> PAGE_SHIFT);
32cad84f 599 if (is_error_page(page))
a9d30f33 600 return NULL;
32cad84f 601
a9d30f33
NHE		 602 return page;
603}
604
605static void nested_release_page(struct page *page)
606{
607 kvm_release_page_dirty(page);
608}
609
610static void nested_release_page_clean(struct page *page)
611{
612 kvm_release_page_clean(page);
613}
614
4e1096d2 615static u64 construct_eptp(unsigned long root_hpa);
4610c9cc
DX		 616static void kvm_cpu_vmxon(u64 addr);
617static void kvm_cpu_vmxoff(void);
aff48baa 618static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3);
776e58ea 619static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr);
b246dd5d
OW		 620static void vmx_set_segment(struct kvm_vcpu *vcpu,
621 struct kvm_segment *var, int seg);
622static void vmx_get_segment(struct kvm_vcpu *vcpu,
623 struct kvm_segment *var, int seg);
75880a01 624
6aa8b732
AK		 625static DEFINE_PER_CPU(struct vmcs *, vmxarea);
626static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
d462b819
NHE		 627/*
628 * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
629 * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
630 */
631static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
3444d7da 632static DEFINE_PER_CPU(struct desc_ptr, host_gdt);
6aa8b732 633
3e7c73e9
AK		 634static unsigned long *vmx_io_bitmap_a;
635static unsigned long *vmx_io_bitmap_b;
5897297b
AK		 636static unsigned long *vmx_msr_bitmap_legacy;
637static unsigned long *vmx_msr_bitmap_longmode;
fdef3ad1 638
110312c8 639static bool cpu_has_load_ia32_efer;
8bf00a52 640static bool cpu_has_load_perf_global_ctrl;
110312c8 641
2384d2b3
SY		 642static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
643static DEFINE_SPINLOCK(vmx_vpid_lock);
644
1c3d14fe 645static struct vmcs_config {
6aa8b732
AK		 646 int size;
647 int order;
648 u32 revision_id;
1c3d14fe
YS		 649 u32 pin_based_exec_ctrl;
650 u32 cpu_based_exec_ctrl;
f78e0e2e 651 u32 cpu_based_2nd_exec_ctrl;
1c3d14fe
YS		 652 u32 vmexit_ctrl;
653 u32 vmentry_ctrl;
654} vmcs_config;
6aa8b732 655
efff9e53 656static struct vmx_capability {
d56f546d
SY		 657 u32 ept;
658 u32 vpid;
659} vmx_capability;
660
6aa8b732
AK		 661#define VMX_SEGMENT_FIELD(seg) \
662 [VCPU_SREG_##seg] = { \
663 .selector = GUEST_##seg##_SELECTOR, \
664 .base = GUEST_##seg##_BASE, \
665 .limit = GUEST_##seg##_LIMIT, \
666 .ar_bytes = GUEST_##seg##_AR_BYTES, \
667 }
668
772e0318 669static const struct kvm_vmx_segment_field {
6aa8b732
AK		 670 unsigned selector;
671 unsigned base;
672 unsigned limit;
673 unsigned ar_bytes;
674} kvm_vmx_segment_fields[] = {
675 VMX_SEGMENT_FIELD(CS),
676 VMX_SEGMENT_FIELD(DS),
677 VMX_SEGMENT_FIELD(ES),
678 VMX_SEGMENT_FIELD(FS),
679 VMX_SEGMENT_FIELD(GS),
680 VMX_SEGMENT_FIELD(SS),
681 VMX_SEGMENT_FIELD(TR),
682 VMX_SEGMENT_FIELD(LDTR),
683};
684
26bb0981
AK		 685static u64 host_efer;
686
6de4f3ad
AK		 687static void ept_save_pdptrs(struct kvm_vcpu *vcpu);
688
4d56c8a7 689/*
8c06585d 690 * Keep MSR_STAR at the end, as setup_msrs() will try to optimize it
4d56c8a7
AK		 691 * away by decrementing the array size.
692 */
6aa8b732 693static const u32 vmx_msr_index[] = {
05b3e0c2 694#ifdef CONFIG_X86_64
44ea2b17 695 MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
6aa8b732 696#endif
8c06585d 697 MSR_EFER, MSR_TSC_AUX, MSR_STAR,
6aa8b732 698};
9d8f549d 699#define NR_VMX_MSR ARRAY_SIZE(vmx_msr_index)
6aa8b732 700
31299944 701static inline bool is_page_fault(u32 intr_info)
6aa8b732
AK		 702{
703 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
704 INTR_INFO_VALID_MASK)) ==
8ab2d2e2 705 (INTR_TYPE_HARD_EXCEPTION | PF_VECTOR | INTR_INFO_VALID_MASK);
6aa8b732
AK		 706}
707
31299944 708static inline bool is_no_device(u32 intr_info)
2ab455cc
AL		 709{
710 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
711 INTR_INFO_VALID_MASK)) ==
8ab2d2e2 712 (INTR_TYPE_HARD_EXCEPTION | NM_VECTOR | INTR_INFO_VALID_MASK);
2ab455cc
AL		 713}
714
31299944 715static inline bool is_invalid_opcode(u32 intr_info)
7aa81cc0
AL		 716{
717 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
718 INTR_INFO_VALID_MASK)) ==
8ab2d2e2 719 (INTR_TYPE_HARD_EXCEPTION | UD_VECTOR | INTR_INFO_VALID_MASK);
7aa81cc0
AL		 720}
721
31299944 722static inline bool is_external_interrupt(u32 intr_info)
6aa8b732
AK		 723{
724 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
725 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
726}
727
31299944 728static inline bool is_machine_check(u32 intr_info)
a0861c02
AK		 729{
730 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
731 INTR_INFO_VALID_MASK)) ==
732 (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK);
733}
734
31299944 735static inline bool cpu_has_vmx_msr_bitmap(void)
25c5f225 736{
04547156 737 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS;
25c5f225
SY		 738}
739
31299944 740static inline bool cpu_has_vmx_tpr_shadow(void)
6e5d865c 741{
04547156 742 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW;
6e5d865c
YS		 743}
744
31299944 745static inline bool vm_need_tpr_shadow(struct kvm *kvm)
6e5d865c 746{
04547156 747 return (cpu_has_vmx_tpr_shadow()) && (irqchip_in_kernel(kvm));
6e5d865c
YS		 748}
749
31299944 750static inline bool cpu_has_secondary_exec_ctrls(void)
f78e0e2e 751{
04547156
SY		 752 return vmcs_config.cpu_based_exec_ctrl &
753 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
f78e0e2e
SY		 754}
755
774ead3a 756static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
f78e0e2e 757{
04547156
SY		 758 return vmcs_config.cpu_based_2nd_exec_ctrl &
759 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
760}
761
762static inline bool cpu_has_vmx_flexpriority(void)
763{
764 return cpu_has_vmx_tpr_shadow() &&
765 cpu_has_vmx_virtualize_apic_accesses();
f78e0e2e
SY		 766}
767
e799794e
MT		 768static inline bool cpu_has_vmx_ept_execute_only(void)
769{
31299944 770 return vmx_capability.ept & VMX_EPT_EXECUTE_ONLY_BIT;
e799794e
MT		 771}
772
773static inline bool cpu_has_vmx_eptp_uncacheable(void)
774{
31299944 775 return vmx_capability.ept & VMX_EPTP_UC_BIT;
e799794e
MT		 776}
777
778static inline bool cpu_has_vmx_eptp_writeback(void)
779{
31299944 780 return vmx_capability.ept & VMX_EPTP_WB_BIT;
e799794e
MT		 781}
782
783static inline bool cpu_has_vmx_ept_2m_page(void)
784{
31299944 785 return vmx_capability.ept & VMX_EPT_2MB_PAGE_BIT;
e799794e
MT		 786}
787
878403b7
SY		 788static inline bool cpu_has_vmx_ept_1g_page(void)
789{
31299944 790 return vmx_capability.ept & VMX_EPT_1GB_PAGE_BIT;
878403b7
SY		 791}
792
4bc9b982
SY		 793static inline bool cpu_has_vmx_ept_4levels(void)
794{
795 return vmx_capability.ept & VMX_EPT_PAGE_WALK_4_BIT;
796}
797
83c3a331
XH		 798static inline bool cpu_has_vmx_ept_ad_bits(void)
799{
800 return vmx_capability.ept & VMX_EPT_AD_BIT;
801}
802
31299944 803static inline bool cpu_has_vmx_invept_individual_addr(void)
d56f546d 804{
31299944 805 return vmx_capability.ept & VMX_EPT_EXTENT_INDIVIDUAL_BIT;
d56f546d
SY		 806}
807
31299944 808static inline bool cpu_has_vmx_invept_context(void)
d56f546d 809{
31299944 810 return vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT;
d56f546d
SY		 811}
812
31299944 813static inline bool cpu_has_vmx_invept_global(void)
d56f546d 814{
31299944 815 return vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT;
d56f546d
SY		 816}
817
518c8aee
GJ		 818static inline bool cpu_has_vmx_invvpid_single(void)
819{
820 return vmx_capability.vpid & VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT;
821}
822
b9d762fa
GJ		 823static inline bool cpu_has_vmx_invvpid_global(void)
824{
825 return vmx_capability.vpid & VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT;
826}
827
31299944 828static inline bool cpu_has_vmx_ept(void)
d56f546d 829{
04547156
SY		 830 return vmcs_config.cpu_based_2nd_exec_ctrl &
831 SECONDARY_EXEC_ENABLE_EPT;
d56f546d
SY		 832}
833
31299944 834static inline bool cpu_has_vmx_unrestricted_guest(void)
3a624e29
NK		 835{
836 return vmcs_config.cpu_based_2nd_exec_ctrl &
837 SECONDARY_EXEC_UNRESTRICTED_GUEST;
838}
839
31299944 840static inline bool cpu_has_vmx_ple(void)
4b8d54f9
ZE		 841{
842 return vmcs_config.cpu_based_2nd_exec_ctrl &
843 SECONDARY_EXEC_PAUSE_LOOP_EXITING;
844}
845
31299944 846static inline bool vm_need_virtualize_apic_accesses(struct kvm *kvm)
f78e0e2e 847{
6d3e435e 848 return flexpriority_enabled && irqchip_in_kernel(kvm);
f78e0e2e
SY		 849}
850
31299944 851static inline bool cpu_has_vmx_vpid(void)
2384d2b3 852{
04547156
SY		 853 return vmcs_config.cpu_based_2nd_exec_ctrl &
854 SECONDARY_EXEC_ENABLE_VPID;
2384d2b3
SY		 855}
856
31299944 857static inline bool cpu_has_vmx_rdtscp(void)
4e47c7a6
SY		 858{
859 return vmcs_config.cpu_based_2nd_exec_ctrl &
860 SECONDARY_EXEC_RDTSCP;
861}
862
ad756a16
MJ		 863static inline bool cpu_has_vmx_invpcid(void)
864{
865 return vmcs_config.cpu_based_2nd_exec_ctrl &
866 SECONDARY_EXEC_ENABLE_INVPCID;
867}
868
31299944 869static inline bool cpu_has_virtual_nmis(void)
f08864b4
SY		 870{
871 return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS;
872}
873
f5f48ee1
SY		 874static inline bool cpu_has_vmx_wbinvd_exit(void)
875{
876 return vmcs_config.cpu_based_2nd_exec_ctrl &
877 SECONDARY_EXEC_WBINVD_EXITING;
878}
879
04547156
SY		 880static inline bool report_flexpriority(void)
881{
882 return flexpriority_enabled;
883}
884
fe3ef05c
NHE		 885static inline bool nested_cpu_has(struct vmcs12 *vmcs12, u32 bit)
886{
887 return vmcs12->cpu_based_vm_exec_control & bit;
888}
889
890static inline bool nested_cpu_has2(struct vmcs12 *vmcs12, u32 bit)
891{
892 return (vmcs12->cpu_based_vm_exec_control &
893 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
894 (vmcs12->secondary_vm_exec_control & bit);
895}
896
644d711a
NHE		 897static inline bool nested_cpu_has_virtual_nmis(struct vmcs12 *vmcs12,
898 struct kvm_vcpu *vcpu)
899{
900 return vmcs12->pin_based_vm_exec_control & PIN_BASED_VIRTUAL_NMIS;
901}
902
903static inline bool is_exception(u32 intr_info)
904{
905 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
906 == (INTR_TYPE_HARD_EXCEPTION | INTR_INFO_VALID_MASK);
907}
908
909static void nested_vmx_vmexit(struct kvm_vcpu *vcpu);
7c177938
NHE		 910static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
911 struct vmcs12 *vmcs12,
912 u32 reason, unsigned long qualification);
913
8b9cf98c 914static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
7725f0ba
AK		 915{
916 int i;
917
a2fa3e9f 918 for (i = 0; i < vmx->nmsrs; ++i)
26bb0981 919 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
a75beee6
ED		 920 return i;
921 return -1;
922}
923
2384d2b3
SY		 924static inline void __invvpid(int ext, u16 vpid, gva_t gva)
925{
926 struct {
927 u64 vpid : 16;
928 u64 rsvd : 48;
929 u64 gva;
930 } operand = { vpid, 0, gva };
931
4ecac3fd 932 asm volatile (__ex(ASM_VMX_INVVPID)
2384d2b3
SY		 933 /* CF==1 or ZF==1 --> rc = -1 */
934 "; ja 1f ; ud2 ; 1:"
935 : : "a"(&operand), "c"(ext) : "cc", "memory");
936}
937
1439442c
SY		 938static inline void __invept(int ext, u64 eptp, gpa_t gpa)
939{
940 struct {
941 u64 eptp, gpa;
942 } operand = {eptp, gpa};
943
4ecac3fd 944 asm volatile (__ex(ASM_VMX_INVEPT)
1439442c
SY		 945 /* CF==1 or ZF==1 --> rc = -1 */
946 "; ja 1f ; ud2 ; 1:\n"
947 : : "a" (&operand), "c" (ext) : "cc", "memory");
948}
949
26bb0981 950static struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
a75beee6
ED		 951{
952 int i;
953
8b9cf98c 954 i = __find_msr_index(vmx, msr);
a75beee6 955 if (i >= 0)
a2fa3e9f 956 return &vmx->guest_msrs[i];
8b6d44c7 957 return NULL;
7725f0ba
AK		 958}
959
6aa8b732
AK		 960static void vmcs_clear(struct vmcs *vmcs)
961{
962 u64 phys_addr = __pa(vmcs);
963 u8 error;
964
4ecac3fd 965 asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
16d8f72f 966 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
6aa8b732
AK		 967 : "cc", "memory");
968 if (error)
969 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
970 vmcs, phys_addr);
971}
972
d462b819
NHE		 973static inline void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs)
974{
975 vmcs_clear(loaded_vmcs->vmcs);
976 loaded_vmcs->cpu = -1;
977 loaded_vmcs->launched = 0;
978}
979
7725b894
DX		 980static void vmcs_load(struct vmcs *vmcs)
981{
982 u64 phys_addr = __pa(vmcs);
983 u8 error;
984
985 asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
16d8f72f 986 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
7725b894
DX		 987 : "cc", "memory");
988 if (error)
2844d849 989 printk(KERN_ERR "kvm: vmptrld %p/%llx failed\n",
7725b894
DX		 990 vmcs, phys_addr);
991}
992
d462b819 993static void __loaded_vmcs_clear(void *arg)
6aa8b732 994{
d462b819 995 struct loaded_vmcs *loaded_vmcs = arg;
d3b2c338 996 int cpu = raw_smp_processor_id();
6aa8b732 997
d462b819
NHE		 998 if (loaded_vmcs->cpu != cpu)
999 return; /* vcpu migration can race with cpu offline */
1000 if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
6aa8b732 1001 per_cpu(current_vmcs, cpu) = NULL;
d462b819
NHE		 1002 list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
1003 loaded_vmcs_init(loaded_vmcs);
6aa8b732
AK		 1004}
1005
d462b819 1006static void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
8d0be2b3 1007{
d462b819
NHE		 1008 if (loaded_vmcs->cpu != -1)
1009 smp_call_function_single(
1010 loaded_vmcs->cpu, __loaded_vmcs_clear, loaded_vmcs, 1);
8d0be2b3
AK		 1011}
1012
1760dd49 1013static inline void vpid_sync_vcpu_single(struct vcpu_vmx *vmx)
2384d2b3
SY		 1014{
1015 if (vmx->vpid == 0)
1016 return;
1017
518c8aee
GJ		 1018 if (cpu_has_vmx_invvpid_single())
1019 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vmx->vpid, 0);
2384d2b3
SY		 1020}
1021
b9d762fa
GJ		 1022static inline void vpid_sync_vcpu_global(void)
1023{
1024 if (cpu_has_vmx_invvpid_global())
1025 __invvpid(VMX_VPID_EXTENT_ALL_CONTEXT, 0, 0);
1026}
1027
1028static inline void vpid_sync_context(struct vcpu_vmx *vmx)
1029{
1030 if (cpu_has_vmx_invvpid_single())
1760dd49 1031 vpid_sync_vcpu_single(vmx);
b9d762fa
GJ		 1032 else
1033 vpid_sync_vcpu_global();
1034}
1035
1439442c
SY		 1036static inline void ept_sync_global(void)
1037{
1038 if (cpu_has_vmx_invept_global())
1039 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
1040}
1041
1042static inline void ept_sync_context(u64 eptp)
1043{
089d034e 1044 if (enable_ept) {
1439442c
SY		 1045 if (cpu_has_vmx_invept_context())
1046 __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
1047 else
1048 ept_sync_global();
1049 }
1050}
1051
1052static inline void ept_sync_individual_addr(u64 eptp, gpa_t gpa)
1053{
089d034e 1054 if (enable_ept) {
1439442c
SY		 1055 if (cpu_has_vmx_invept_individual_addr())
1056 __invept(VMX_EPT_EXTENT_INDIVIDUAL_ADDR,
1057 eptp, gpa);
1058 else
1059 ept_sync_context(eptp);
1060 }
1061}
1062
96304217 1063static __always_inline unsigned long vmcs_readl(unsigned long field)
6aa8b732 1064{
5e520e62 1065 unsigned long value;
6aa8b732 1066
5e520e62
AK		 1067 asm volatile (__ex_clear(ASM_VMX_VMREAD_RDX_RAX, "%0")
1068 : "=a"(value) : "d"(field) : "cc");
6aa8b732
AK		 1069 return value;
1070}
1071
96304217 1072static __always_inline u16 vmcs_read16(unsigned long field)
6aa8b732
AK		 1073{
1074 return vmcs_readl(field);
1075}
1076
96304217 1077static __always_inline u32 vmcs_read32(unsigned long field)
6aa8b732
AK		 1078{
1079 return vmcs_readl(field);
1080}
1081
96304217 1082static __always_inline u64 vmcs_read64(unsigned long field)
6aa8b732 1083{
05b3e0c2 1084#ifdef CONFIG_X86_64
6aa8b732
AK		 1085 return vmcs_readl(field);
1086#else
1087 return vmcs_readl(field) | ((u64)vmcs_readl(field+1) << 32);
1088#endif
1089}
1090
e52de1b8
AK		 1091static noinline void vmwrite_error(unsigned long field, unsigned long value)
1092{
1093 printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
1094 field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
1095 dump_stack();
1096}
1097
6aa8b732
AK		 1098static void vmcs_writel(unsigned long field, unsigned long value)
1099{
1100 u8 error;
1101
4ecac3fd 1102 asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
d77c26fc 1103 : "=q"(error) : "a"(value), "d"(field) : "cc");
e52de1b8
AK		 1104 if (unlikely(error))
1105 vmwrite_error(field, value);
6aa8b732
AK		 1106}
1107
1108static void vmcs_write16(unsigned long field, u16 value)
1109{
1110 vmcs_writel(field, value);
1111}
1112
1113static void vmcs_write32(unsigned long field, u32 value)
1114{
1115 vmcs_writel(field, value);
1116}
1117
1118static void vmcs_write64(unsigned long field, u64 value)
1119{
6aa8b732 1120 vmcs_writel(field, value);
7682f2d0 1121#ifndef CONFIG_X86_64
6aa8b732
AK		 1122 asm volatile ("");
1123 vmcs_writel(field+1, value >> 32);
1124#endif
1125}
1126
2ab455cc
AL		 1127static void vmcs_clear_bits(unsigned long field, u32 mask)
1128{
1129 vmcs_writel(field, vmcs_readl(field) & ~mask);
1130}
1131
1132static void vmcs_set_bits(unsigned long field, u32 mask)
1133{
1134 vmcs_writel(field, vmcs_readl(field) | mask);
1135}
1136
2fb92db1
AK		 1137static void vmx_segment_cache_clear(struct vcpu_vmx *vmx)
1138{
1139 vmx->segment_cache.bitmask = 0;
1140}
1141
1142static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
1143 unsigned field)
1144{
1145 bool ret;
1146 u32 mask = 1 << (seg * SEG_FIELD_NR + field);
1147
1148 if (!(vmx->vcpu.arch.regs_avail & (1 << VCPU_EXREG_SEGMENTS))) {
1149 vmx->vcpu.arch.regs_avail |= (1 << VCPU_EXREG_SEGMENTS);
1150 vmx->segment_cache.bitmask = 0;
1151 }
1152 ret = vmx->segment_cache.bitmask & mask;
1153 vmx->segment_cache.bitmask |= mask;
1154 return ret;
1155}
1156
1157static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
1158{
1159 u16 *p = &vmx->segment_cache.seg[seg].selector;
1160
1161 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
1162 *p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
1163 return *p;
1164}
1165
1166static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
1167{
1168 ulong *p = &vmx->segment_cache.seg[seg].base;
1169
1170 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
1171 *p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
1172 return *p;
1173}
1174
1175static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
1176{
1177 u32 *p = &vmx->segment_cache.seg[seg].limit;
1178
1179 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
1180 *p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
1181 return *p;
1182}
1183
1184static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
1185{
1186 u32 *p = &vmx->segment_cache.seg[seg].ar;
1187
1188 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
1189 *p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
1190 return *p;
1191}
1192
abd3f2d6
AK		 1193static void update_exception_bitmap(struct kvm_vcpu *vcpu)
1194{
1195 u32 eb;
1196
fd7373cc
JK		 1197 eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
1198 (1u << NM_VECTOR) | (1u << DB_VECTOR);
1199 if ((vcpu->guest_debug &
1200 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
1201 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
1202 eb |= 1u << BP_VECTOR;
7ffd92c5 1203 if (to_vmx(vcpu)->rmode.vm86_active)
abd3f2d6 1204 eb = ~0;
089d034e 1205 if (enable_ept)
1439442c 1206 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
02daab21
AK		 1207 if (vcpu->fpu_active)
1208 eb &= ~(1u << NM_VECTOR);
36cf24e0
NHE		 1209
1210 /* When we are running a nested L2 guest and L1 specified for it a
1211 * certain exception bitmap, we must trap the same exceptions and pass
1212 * them to L1. When running L2, we will only handle the exceptions
1213 * specified above if L1 did not want them.
1214 */
1215 if (is_guest_mode(vcpu))
1216 eb |= get_vmcs12(vcpu)->exception_bitmap;
1217
abd3f2d6
AK		 1218 vmcs_write32(EXCEPTION_BITMAP, eb);
1219}
1220
8bf00a52
GN		 1221static void clear_atomic_switch_msr_special(unsigned long entry,
1222 unsigned long exit)
1223{
1224 vmcs_clear_bits(VM_ENTRY_CONTROLS, entry);
1225 vmcs_clear_bits(VM_EXIT_CONTROLS, exit);
1226}
1227
61d2ef2c
AK		 1228static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
1229{
1230 unsigned i;
1231 struct msr_autoload *m = &vmx->msr_autoload;
1232
8bf00a52
GN		 1233 switch (msr) {
1234 case MSR_EFER:
1235 if (cpu_has_load_ia32_efer) {
1236 clear_atomic_switch_msr_special(VM_ENTRY_LOAD_IA32_EFER,
1237 VM_EXIT_LOAD_IA32_EFER);
1238 return;
1239 }
1240 break;
1241 case MSR_CORE_PERF_GLOBAL_CTRL:
1242 if (cpu_has_load_perf_global_ctrl) {
1243 clear_atomic_switch_msr_special(
1244 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1245 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
1246 return;
1247 }
1248 break;
110312c8
AK		 1249 }
1250
61d2ef2c
AK		 1251 for (i = 0; i < m->nr; ++i)
1252 if (m->guest[i].index == msr)
1253 break;
1254
1255 if (i == m->nr)
1256 return;
1257 --m->nr;
1258 m->guest[i] = m->guest[m->nr];
1259 m->host[i] = m->host[m->nr];
1260 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1261 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1262}
1263
8bf00a52
GN		 1264static void add_atomic_switch_msr_special(unsigned long entry,
1265 unsigned long exit, unsigned long guest_val_vmcs,
1266 unsigned long host_val_vmcs, u64 guest_val, u64 host_val)
1267{
1268 vmcs_write64(guest_val_vmcs, guest_val);
1269 vmcs_write64(host_val_vmcs, host_val);
1270 vmcs_set_bits(VM_ENTRY_CONTROLS, entry);
1271 vmcs_set_bits(VM_EXIT_CONTROLS, exit);
1272}
1273
61d2ef2c
AK		 1274static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
1275 u64 guest_val, u64 host_val)
1276{
1277 unsigned i;
1278 struct msr_autoload *m = &vmx->msr_autoload;
1279
8bf00a52
GN		 1280 switch (msr) {
1281 case MSR_EFER:
1282 if (cpu_has_load_ia32_efer) {
1283 add_atomic_switch_msr_special(VM_ENTRY_LOAD_IA32_EFER,
1284 VM_EXIT_LOAD_IA32_EFER,
1285 GUEST_IA32_EFER,
1286 HOST_IA32_EFER,
1287 guest_val, host_val);
1288 return;
1289 }
1290 break;
1291 case MSR_CORE_PERF_GLOBAL_CTRL:
1292 if (cpu_has_load_perf_global_ctrl) {
1293 add_atomic_switch_msr_special(
1294 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1295 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
1296 GUEST_IA32_PERF_GLOBAL_CTRL,
1297 HOST_IA32_PERF_GLOBAL_CTRL,
1298 guest_val, host_val);
1299 return;
1300 }
1301 break;
110312c8
AK		 1302 }
1303
61d2ef2c
AK		 1304 for (i = 0; i < m->nr; ++i)
1305 if (m->guest[i].index == msr)
1306 break;
1307
e7fc6f93
GN		 1308 if (i == NR_AUTOLOAD_MSRS) {
1309 printk_once(KERN_WARNING"Not enough mst switch entries. "
1310 "Can't add msr %x\n", msr);
1311 return;
1312 } else if (i == m->nr) {
61d2ef2c
AK		 1313 ++m->nr;
1314 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1315 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1316 }
1317
1318 m->guest[i].index = msr;
1319 m->guest[i].value = guest_val;
1320 m->host[i].index = msr;
1321 m->host[i].value = host_val;
1322}
1323
33ed6329
AK		 1324static void reload_tss(void)
1325{
33ed6329
AK		 1326 /*
1327 * VT restores TR but not its size. Useless.
1328 */
d359192f 1329 struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
a5f61300 1330 struct desc_struct *descs;
33ed6329 1331
d359192f 1332 descs = (void *)gdt->address;
33ed6329
AK		 1333 descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
1334 load_TR_desc();
33ed6329
AK		 1335}
1336
92c0d900 1337static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
2cc51560 1338{
3a34a881 1339 u64 guest_efer;
51c6cf66
AK		 1340 u64 ignore_bits;
1341
f6801dff 1342 guest_efer = vmx->vcpu.arch.efer;
3a34a881 1343
51c6cf66 1344 /*
0fa06071 1345 * NX is emulated; LMA and LME handled by hardware; SCE meaningless
51c6cf66
AK		 1346 * outside long mode
1347 */
1348 ignore_bits = EFER_NX | EFER_SCE;
1349#ifdef CONFIG_X86_64
1350 ignore_bits |= EFER_LMA | EFER_LME;
1351 /* SCE is meaningful only in long mode on Intel */
1352 if (guest_efer & EFER_LMA)
1353 ignore_bits &= ~(u64)EFER_SCE;
1354#endif
51c6cf66
AK		 1355 guest_efer &= ~ignore_bits;
1356 guest_efer |= host_efer & ignore_bits;
26bb0981 1357 vmx->guest_msrs[efer_offset].data = guest_efer;
d5696725 1358 vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
84ad33ef
AK		 1359
1360 clear_atomic_switch_msr(vmx, MSR_EFER);
1361 /* On ept, can't emulate nx, and must switch nx atomically */
1362 if (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX)) {
1363 guest_efer = vmx->vcpu.arch.efer;
1364 if (!(guest_efer & EFER_LMA))
1365 guest_efer &= ~EFER_LME;
1366 add_atomic_switch_msr(vmx, MSR_EFER, guest_efer, host_efer);
1367 return false;
1368 }
1369
26bb0981 1370 return true;
51c6cf66
AK		 1371}
1372
2d49ec72
GN		 1373static unsigned long segment_base(u16 selector)
1374{
d359192f 1375 struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
2d49ec72
GN		 1376 struct desc_struct *d;
1377 unsigned long table_base;
1378 unsigned long v;
1379
1380 if (!(selector & ~3))
1381 return 0;
1382
d359192f 1383 table_base = gdt->address;
2d49ec72
GN		 1384
1385 if (selector & 4) { /* from ldt */
1386 u16 ldt_selector = kvm_read_ldt();
1387
1388 if (!(ldt_selector & ~3))
1389 return 0;
1390
1391 table_base = segment_base(ldt_selector);
1392 }
1393 d = (struct desc_struct *)(table_base + (selector & ~7));
1394 v = get_desc_base(d);
1395#ifdef CONFIG_X86_64
1396 if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
1397 v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
1398#endif
1399 return v;
1400}
1401
1402static inline unsigned long kvm_read_tr_base(void)
1403{
1404 u16 tr;
1405 asm("str %0" : "=g"(tr));
1406 return segment_base(tr);
1407}
1408
04d2cc77 1409static void vmx_save_host_state(struct kvm_vcpu *vcpu)
33ed6329 1410{
04d2cc77 1411 struct vcpu_vmx *vmx = to_vmx(vcpu);
26bb0981 1412 int i;
04d2cc77 1413
a2fa3e9f 1414 if (vmx->host_state.loaded)
33ed6329
AK		 1415 return;
1416
a2fa3e9f 1417 vmx->host_state.loaded = 1;
33ed6329
AK		 1418 /*
1419 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
1420 * allow segment selectors with cpl > 0 or ti == 1.
1421 */
d6e88aec 1422 vmx->host_state.ldt_sel = kvm_read_ldt();
152d3f2f 1423 vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
9581d442 1424 savesegment(fs, vmx->host_state.fs_sel);
152d3f2f 1425 if (!(vmx->host_state.fs_sel & 7)) {
a2fa3e9f 1426 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
152d3f2f
LV		 1427 vmx->host_state.fs_reload_needed = 0;
1428 } else {
33ed6329 1429 vmcs_write16(HOST_FS_SELECTOR, 0);
152d3f2f 1430 vmx->host_state.fs_reload_needed = 1;
33ed6329 1431 }
9581d442 1432 savesegment(gs, vmx->host_state.gs_sel);
a2fa3e9f
GH		 1433 if (!(vmx->host_state.gs_sel & 7))
1434 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
33ed6329
AK		 1435 else {
1436 vmcs_write16(HOST_GS_SELECTOR, 0);
152d3f2f 1437 vmx->host_state.gs_ldt_reload_needed = 1;
33ed6329
AK		 1438 }
1439
b2da15ac
AK		 1440#ifdef CONFIG_X86_64
1441 savesegment(ds, vmx->host_state.ds_sel);
1442 savesegment(es, vmx->host_state.es_sel);
1443#endif
1444
33ed6329
AK		 1445#ifdef CONFIG_X86_64
1446 vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
1447 vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
1448#else
a2fa3e9f
GH		 1449 vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
1450 vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
33ed6329 1451#endif
707c0874
AK		 1452
1453#ifdef CONFIG_X86_64
c8770e7b
AK		 1454 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1455 if (is_long_mode(&vmx->vcpu))
44ea2b17 1456 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
707c0874 1457#endif
26bb0981
AK		 1458 for (i = 0; i < vmx->save_nmsrs; ++i)
1459 kvm_set_shared_msr(vmx->guest_msrs[i].index,
d5696725
AK		 1460 vmx->guest_msrs[i].data,
1461 vmx->guest_msrs[i].mask);
33ed6329
AK		 1462}
1463
a9b21b62 1464static void __vmx_load_host_state(struct vcpu_vmx *vmx)
33ed6329 1465{
a2fa3e9f 1466 if (!vmx->host_state.loaded)
33ed6329
AK		 1467 return;
1468
e1beb1d3 1469 ++vmx->vcpu.stat.host_state_reload;
a2fa3e9f 1470 vmx->host_state.loaded = 0;
c8770e7b
AK		 1471#ifdef CONFIG_X86_64
1472 if (is_long_mode(&vmx->vcpu))
1473 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1474#endif
152d3f2f 1475 if (vmx->host_state.gs_ldt_reload_needed) {
d6e88aec 1476 kvm_load_ldt(vmx->host_state.ldt_sel);
33ed6329 1477#ifdef CONFIG_X86_64
9581d442 1478 load_gs_index(vmx->host_state.gs_sel);
9581d442
AK		 1479#else
1480 loadsegment(gs, vmx->host_state.gs_sel);
33ed6329 1481#endif
33ed6329 1482 }
0a77fe4c
AK		 1483 if (vmx->host_state.fs_reload_needed)
1484 loadsegment(fs, vmx->host_state.fs_sel);
b2da15ac
AK		 1485#ifdef CONFIG_X86_64
1486 if (unlikely(vmx->host_state.ds_sel | vmx->host_state.es_sel)) {
1487 loadsegment(ds, vmx->host_state.ds_sel);
1488 loadsegment(es, vmx->host_state.es_sel);
1489 }
b2da15ac 1490#endif
152d3f2f 1491 reload_tss();
44ea2b17 1492#ifdef CONFIG_X86_64
c8770e7b 1493 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
44ea2b17 1494#endif
1361b83a 1495 if (user_has_fpu())
1c11e713 1496 clts();
3444d7da 1497 load_gdt(&__get_cpu_var(host_gdt));
33ed6329
AK		 1498}
1499
a9b21b62
AK		 1500static void vmx_load_host_state(struct vcpu_vmx *vmx)
1501{
1502 preempt_disable();
1503 __vmx_load_host_state(vmx);
1504 preempt_enable();
1505}
1506
6aa8b732
AK		 1507/*
1508 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
1509 * vcpu mutex is already taken.
1510 */
15ad7146 1511static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
6aa8b732 1512{
a2fa3e9f 1513 struct vcpu_vmx *vmx = to_vmx(vcpu);
4610c9cc 1514 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
6aa8b732 1515
4610c9cc
DX		 1516 if (!vmm_exclusive)
1517 kvm_cpu_vmxon(phys_addr);
d462b819
NHE		 1518 else if (vmx->loaded_vmcs->cpu != cpu)
1519 loaded_vmcs_clear(vmx->loaded_vmcs);
6aa8b732 1520
d462b819
NHE		 1521 if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
1522 per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
1523 vmcs_load(vmx->loaded_vmcs->vmcs);
6aa8b732
AK		 1524 }
1525
d462b819 1526 if (vmx->loaded_vmcs->cpu != cpu) {
d359192f 1527 struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
6aa8b732
AK		 1528 unsigned long sysenter_esp;
1529
a8eeb04a 1530 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
92fe13be 1531 local_irq_disable();
d462b819
NHE		 1532 list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
1533 &per_cpu(loaded_vmcss_on_cpu, cpu));
92fe13be
DX		 1534 local_irq_enable();
1535
6aa8b732
AK		 1536 /*
1537 * Linux uses per-cpu TSS and GDT, so set these when switching
1538 * processors.
1539 */
d6e88aec 1540 vmcs_writel(HOST_TR_BASE, kvm_read_tr_base()); /* 22.2.4 */
d359192f 1541 vmcs_writel(HOST_GDTR_BASE, gdt->address); /* 22.2.4 */
6aa8b732
AK		 1542
1543 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
1544 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
d462b819 1545 vmx->loaded_vmcs->cpu = cpu;
6aa8b732 1546 }
6aa8b732
AK		 1547}
1548
1549static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
1550{
a9b21b62 1551 __vmx_load_host_state(to_vmx(vcpu));
4610c9cc 1552 if (!vmm_exclusive) {
d462b819
NHE		 1553 __loaded_vmcs_clear(to_vmx(vcpu)->loaded_vmcs);
1554 vcpu->cpu = -1;
4610c9cc
DX		 1555 kvm_cpu_vmxoff();
1556 }
6aa8b732
AK		 1557}
1558
5fd86fcf
AK		 1559static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
1560{
81231c69
AK		 1561 ulong cr0;
1562
5fd86fcf
AK		 1563 if (vcpu->fpu_active)
1564 return;
1565 vcpu->fpu_active = 1;
81231c69
AK		 1566 cr0 = vmcs_readl(GUEST_CR0);
1567 cr0 &= ~(X86_CR0_TS | X86_CR0_MP);
1568 cr0 |= kvm_read_cr0_bits(vcpu, X86_CR0_TS | X86_CR0_MP);
1569 vmcs_writel(GUEST_CR0, cr0);
5fd86fcf 1570 update_exception_bitmap(vcpu);
edcafe3c 1571 vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
36cf24e0
NHE		 1572 if (is_guest_mode(vcpu))
1573 vcpu->arch.cr0_guest_owned_bits &=
1574 ~get_vmcs12(vcpu)->cr0_guest_host_mask;
edcafe3c 1575 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
5fd86fcf
AK		 1576}
1577
edcafe3c
AK		 1578static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
1579
fe3ef05c
NHE		 1580/*
1581 * Return the cr0 value that a nested guest would read. This is a combination
1582 * of the real cr0 used to run the guest (guest_cr0), and the bits shadowed by
1583 * its hypervisor (cr0_read_shadow).
1584 */
1585static inline unsigned long nested_read_cr0(struct vmcs12 *fields)
1586{
1587 return (fields->guest_cr0 & ~fields->cr0_guest_host_mask) |
1588 (fields->cr0_read_shadow & fields->cr0_guest_host_mask);
1589}
1590static inline unsigned long nested_read_cr4(struct vmcs12 *fields)
1591{
1592 return (fields->guest_cr4 & ~fields->cr4_guest_host_mask) |
1593 (fields->cr4_read_shadow & fields->cr4_guest_host_mask);
1594}
1595
5fd86fcf
AK		 1596static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
1597{
36cf24e0
NHE		 1598 /* Note that there is no vcpu->fpu_active = 0 here. The caller must
1599 * set this *before* calling this function.
1600 */
edcafe3c 1601 vmx_decache_cr0_guest_bits(vcpu);
81231c69 1602 vmcs_set_bits(GUEST_CR0, X86_CR0_TS | X86_CR0_MP);
5fd86fcf 1603 update_exception_bitmap(vcpu);
edcafe3c
AK		 1604 vcpu->arch.cr0_guest_owned_bits = 0;
1605 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
36cf24e0
NHE		 1606 if (is_guest_mode(vcpu)) {
1607 /*
1608 * L1's specified read shadow might not contain the TS bit,
1609 * so now that we turned on shadowing of this bit, we need to
1610 * set this bit of the shadow. Like in nested_vmx_run we need
1611 * nested_read_cr0(vmcs12), but vmcs12->guest_cr0 is not yet
1612 * up-to-date here because we just decached cr0.TS (and we'll
1613 * only update vmcs12->guest_cr0 on nested exit).
1614 */
1615 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1616 vmcs12->guest_cr0 = (vmcs12->guest_cr0 & ~X86_CR0_TS) |
1617 (vcpu->arch.cr0 & X86_CR0_TS);
1618 vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
1619 } else
1620 vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0);
5fd86fcf
AK		 1621}
1622
6aa8b732
AK		 1623static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
1624{
78ac8b47 1625 unsigned long rflags, save_rflags;
345dcaa8 1626
6de12732
AK		 1627 if (!test_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail)) {
1628 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
1629 rflags = vmcs_readl(GUEST_RFLAGS);
1630 if (to_vmx(vcpu)->rmode.vm86_active) {
1631 rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
1632 save_rflags = to_vmx(vcpu)->rmode.save_rflags;
1633 rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
1634 }
1635 to_vmx(vcpu)->rflags = rflags;
78ac8b47 1636 }
6de12732 1637 return to_vmx(vcpu)->rflags;
6aa8b732
AK		 1638}
1639
1640static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1641{
6de12732 1642 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
69c73028 1643 __clear_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail);
6de12732 1644 to_vmx(vcpu)->rflags = rflags;
78ac8b47
AK		 1645 if (to_vmx(vcpu)->rmode.vm86_active) {
1646 to_vmx(vcpu)->rmode.save_rflags = rflags;
053de044 1647 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
78ac8b47 1648 }
6aa8b732
AK		 1649 vmcs_writel(GUEST_RFLAGS, rflags);
1650}
1651
2809f5d2
GC		 1652static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1653{
1654 u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1655 int ret = 0;
1656
1657 if (interruptibility & GUEST_INTR_STATE_STI)
48005f64 1658 ret |= KVM_X86_SHADOW_INT_STI;
2809f5d2 1659 if (interruptibility & GUEST_INTR_STATE_MOV_SS)
48005f64 1660 ret |= KVM_X86_SHADOW_INT_MOV_SS;
2809f5d2
GC		 1661
1662 return ret & mask;
1663}
1664
1665static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1666{
1667 u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1668 u32 interruptibility = interruptibility_old;
1669
1670 interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
1671
48005f64 1672 if (mask & KVM_X86_SHADOW_INT_MOV_SS)
2809f5d2 1673 interruptibility |= GUEST_INTR_STATE_MOV_SS;
48005f64 1674 else if (mask & KVM_X86_SHADOW_INT_STI)
2809f5d2
GC		 1675 interruptibility |= GUEST_INTR_STATE_STI;
1676
1677 if ((interruptibility != interruptibility_old))
1678 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
1679}
1680
6aa8b732
AK		 1681static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
1682{
1683 unsigned long rip;
6aa8b732 1684
5fdbf976 1685 rip = kvm_rip_read(vcpu);
6aa8b732 1686 rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
5fdbf976 1687 kvm_rip_write(vcpu, rip);
6aa8b732 1688
2809f5d2
GC		 1689 /* skipping an emulated instruction also counts */
1690 vmx_set_interrupt_shadow(vcpu, 0);
6aa8b732
AK		 1691}
1692
0b6ac343
NHE		 1693/*
1694 * KVM wants to inject page-faults which it got to the guest. This function
1695 * checks whether in a nested guest, we need to inject them to L1 or L2.
1696 * This function assumes it is called with the exit reason in vmcs02 being
1697 * a #PF exception (this is the only case in which KVM injects a #PF when L2
1698 * is running).
1699 */
1700static int nested_pf_handled(struct kvm_vcpu *vcpu)
1701{
1702 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1703
1704 /* TODO: also check PFEC_MATCH/MASK, not just EB.PF. */
95871901 1705 if (!(vmcs12->exception_bitmap & (1u << PF_VECTOR)))
0b6ac343
NHE		 1706 return 0;
1707
1708 nested_vmx_vmexit(vcpu);
1709 return 1;
1710}
1711
298101da 1712static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
ce7ddec4
JR		 1713 bool has_error_code, u32 error_code,
1714 bool reinject)
298101da 1715{
77ab6db0 1716 struct vcpu_vmx *vmx = to_vmx(vcpu);
8ab2d2e2 1717 u32 intr_info = nr | INTR_INFO_VALID_MASK;
77ab6db0 1718
0b6ac343
NHE		 1719 if (nr == PF_VECTOR && is_guest_mode(vcpu) &&
1720 nested_pf_handled(vcpu))
1721 return;
1722
8ab2d2e2 1723 if (has_error_code) {
77ab6db0 1724 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
8ab2d2e2
JK		 1725 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
1726 }
77ab6db0 1727
7ffd92c5 1728 if (vmx->rmode.vm86_active) {
71f9833b
SH		 1729 int inc_eip = 0;
1730 if (kvm_exception_is_soft(nr))
1731 inc_eip = vcpu->arch.event_exit_inst_len;
1732 if (kvm_inject_realmode_interrupt(vcpu, nr, inc_eip) != EMULATE_DONE)
a92601bb 1733 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
77ab6db0
JK		 1734 return;
1735 }
1736
66fd3f7f
GN		 1737 if (kvm_exception_is_soft(nr)) {
1738 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
1739 vmx->vcpu.arch.event_exit_inst_len);
8ab2d2e2
JK		 1740 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
1741 } else
1742 intr_info |= INTR_TYPE_HARD_EXCEPTION;
1743
1744 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
298101da
AK		 1745}
1746
4e47c7a6
SY		 1747static bool vmx_rdtscp_supported(void)
1748{
1749 return cpu_has_vmx_rdtscp();
1750}
1751
ad756a16
MJ		 1752static bool vmx_invpcid_supported(void)
1753{
1754 return cpu_has_vmx_invpcid() && enable_ept;
1755}
1756
a75beee6
ED		 1757/*
1758 * Swap MSR entry in host/guest MSR entry array.
1759 */
8b9cf98c 1760static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
a75beee6 1761{
26bb0981 1762 struct shared_msr_entry tmp;
a2fa3e9f
GH		 1763
1764 tmp = vmx->guest_msrs[to];
1765 vmx->guest_msrs[to] = vmx->guest_msrs[from];
1766 vmx->guest_msrs[from] = tmp;
a75beee6
ED		 1767}
1768
e38aea3e
AK		 1769/*
1770 * Set up the vmcs to automatically save and restore system
1771 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
1772 * mode, as fiddling with msrs is very expensive.
1773 */
8b9cf98c 1774static void setup_msrs(struct vcpu_vmx *vmx)
e38aea3e 1775{
26bb0981 1776 int save_nmsrs, index;
5897297b 1777 unsigned long *msr_bitmap;
e38aea3e 1778
a75beee6
ED		 1779 save_nmsrs = 0;
1780#ifdef CONFIG_X86_64
8b9cf98c 1781 if (is_long_mode(&vmx->vcpu)) {
8b9cf98c 1782 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
a75beee6 1783 if (index >= 0)
8b9cf98c
RR		 1784 move_msr_up(vmx, index, save_nmsrs++);
1785 index = __find_msr_index(vmx, MSR_LSTAR);
a75beee6 1786 if (index >= 0)
8b9cf98c
RR		 1787 move_msr_up(vmx, index, save_nmsrs++);
1788 index = __find_msr_index(vmx, MSR_CSTAR);
a75beee6 1789 if (index >= 0)
8b9cf98c 1790 move_msr_up(vmx, index, save_nmsrs++);
4e47c7a6
SY		 1791 index = __find_msr_index(vmx, MSR_TSC_AUX);
1792 if (index >= 0 && vmx->rdtscp_enabled)
1793 move_msr_up(vmx, index, save_nmsrs++);
a75beee6 1794 /*
8c06585d 1795 * MSR_STAR is only needed on long mode guests, and only
a75beee6
ED		 1796 * if efer.sce is enabled.
1797 */
8c06585d 1798 index = __find_msr_index(vmx, MSR_STAR);
f6801dff 1799 if ((index >= 0) && (vmx->vcpu.arch.efer & EFER_SCE))
8b9cf98c 1800 move_msr_up(vmx, index, save_nmsrs++);
a75beee6
ED		 1801 }
1802#endif
92c0d900
AK		 1803 index = __find_msr_index(vmx, MSR_EFER);
1804 if (index >= 0 && update_transition_efer(vmx, index))
26bb0981 1805 move_msr_up(vmx, index, save_nmsrs++);
e38aea3e 1806
26bb0981 1807 vmx->save_nmsrs = save_nmsrs;
5897297b
AK		 1808
1809 if (cpu_has_vmx_msr_bitmap()) {
1810 if (is_long_mode(&vmx->vcpu))
1811 msr_bitmap = vmx_msr_bitmap_longmode;
1812 else
1813 msr_bitmap = vmx_msr_bitmap_legacy;
1814
1815 vmcs_write64(MSR_BITMAP, __pa(msr_bitmap));
1816 }
e38aea3e
AK		 1817}
1818
6aa8b732
AK		 1819/*
1820 * reads and returns guest's timestamp counter "register"
1821 * guest_tsc = host_tsc + tsc_offset -- 21.3
1822 */
1823static u64 guest_read_tsc(void)
1824{
1825 u64 host_tsc, tsc_offset;
1826
1827 rdtscll(host_tsc);
1828 tsc_offset = vmcs_read64(TSC_OFFSET);
1829 return host_tsc + tsc_offset;
1830}
1831
d5c1785d
NHE		 1832/*
1833 * Like guest_read_tsc, but always returns L1's notion of the timestamp
1834 * counter, even if a nested guest (L2) is currently running.
1835 */
1836u64 vmx_read_l1_tsc(struct kvm_vcpu *vcpu)
1837{
1838 u64 host_tsc, tsc_offset;
1839
1840 rdtscll(host_tsc);
1841 tsc_offset = is_guest_mode(vcpu) ?
1842 to_vmx(vcpu)->nested.vmcs01_tsc_offset :
1843 vmcs_read64(TSC_OFFSET);
1844 return host_tsc + tsc_offset;
1845}
1846
4051b188 1847/*
cc578287
ZA		 1848 * Engage any workarounds for mis-matched TSC rates. Currently limited to
1849 * software catchup for faster rates on slower CPUs.
4051b188 1850 */
cc578287 1851static void vmx_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
4051b188 1852{
cc578287
ZA		 1853 if (!scale)
1854 return;
1855
1856 if (user_tsc_khz > tsc_khz) {
1857 vcpu->arch.tsc_catchup = 1;
1858 vcpu->arch.tsc_always_catchup = 1;
1859 } else
1860 WARN(1, "user requested TSC rate below hardware speed\n");
4051b188
JR		 1861}
1862
6aa8b732 1863/*
99e3e30a 1864 * writes 'offset' into guest's timestamp counter offset register
6aa8b732 1865 */
99e3e30a 1866static void vmx_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
6aa8b732 1867{
27fc51b2 1868 if (is_guest_mode(vcpu)) {
7991825b 1869 /*
27fc51b2
NHE		 1870 * We're here if L1 chose not to trap WRMSR to TSC. According
1871 * to the spec, this should set L1's TSC; The offset that L1
1872 * set for L2 remains unchanged, and still needs to be added
1873 * to the newly set TSC to get L2's TSC.
7991825b 1874 */
27fc51b2
NHE		 1875 struct vmcs12 *vmcs12;
1876 to_vmx(vcpu)->nested.vmcs01_tsc_offset = offset;
1877 /* recalculate vmcs02.TSC_OFFSET: */
1878 vmcs12 = get_vmcs12(vcpu);
1879 vmcs_write64(TSC_OFFSET, offset +
1880 (nested_cpu_has(vmcs12, CPU_BASED_USE_TSC_OFFSETING) ?
1881 vmcs12->tsc_offset : 0));
1882 } else {
1883 vmcs_write64(TSC_OFFSET, offset);
1884 }
6aa8b732
AK		 1885}
1886
f1e2b260 1887static void vmx_adjust_tsc_offset(struct kvm_vcpu *vcpu, s64 adjustment, bool host)
e48672fa
ZA		 1888{
1889 u64 offset = vmcs_read64(TSC_OFFSET);
1890 vmcs_write64(TSC_OFFSET, offset + adjustment);
7991825b
NHE		 1891 if (is_guest_mode(vcpu)) {
1892 /* Even when running L2, the adjustment needs to apply to L1 */
1893 to_vmx(vcpu)->nested.vmcs01_tsc_offset += adjustment;
1894 }
e48672fa
ZA		 1895}
1896
857e4099
JR		 1897static u64 vmx_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1898{
1899 return target_tsc - native_read_tsc();
1900}
1901
801d3424
NHE		 1902static bool guest_cpuid_has_vmx(struct kvm_vcpu *vcpu)
1903{
1904 struct kvm_cpuid_entry2 *best = kvm_find_cpuid_entry(vcpu, 1, 0);
1905 return best && (best->ecx & (1 << (X86_FEATURE_VMX & 31)));
1906}
1907
1908/*
1909 * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
1910 * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
1911 * all guests if the "nested" module option is off, and can also be disabled
1912 * for a single guest by disabling its VMX cpuid bit.
1913 */
1914static inline bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
1915{
1916 return nested && guest_cpuid_has_vmx(vcpu);
1917}
1918
b87a51ae
NHE		 1919/*
1920 * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
1921 * returned for the various VMX controls MSRs when nested VMX is enabled.
1922 * The same values should also be used to verify that vmcs12 control fields are
1923 * valid during nested entry from L1 to L2.
1924 * Each of these control msrs has a low and high 32-bit half: A low bit is on
1925 * if the corresponding bit in the (32-bit) control field *must* be on, and a
1926 * bit in the high half is on if the corresponding bit in the control field
1927 * may be on. See also vmx_control_verify().
1928 * TODO: allow these variables to be modified (downgraded) by module options
1929 * or other means.
1930 */
1931static u32 nested_vmx_procbased_ctls_low, nested_vmx_procbased_ctls_high;
1932static u32 nested_vmx_secondary_ctls_low, nested_vmx_secondary_ctls_high;
1933static u32 nested_vmx_pinbased_ctls_low, nested_vmx_pinbased_ctls_high;
1934static u32 nested_vmx_exit_ctls_low, nested_vmx_exit_ctls_high;
1935static u32 nested_vmx_entry_ctls_low, nested_vmx_entry_ctls_high;
1936static __init void nested_vmx_setup_ctls_msrs(void)
1937{
1938 /*
1939 * Note that as a general rule, the high half of the MSRs (bits in
1940 * the control fields which may be 1) should be initialized by the
1941 * intersection of the underlying hardware's MSR (i.e., features which
1942 * can be supported) and the list of features we want to expose -
1943 * because they are known to be properly supported in our code.
1944 * Also, usually, the low half of the MSRs (bits which must be 1) can
1945 * be set to 0, meaning that L1 may turn off any of these bits. The
1946 * reason is that if one of these bits is necessary, it will appear
1947 * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
1948 * fields of vmcs01 and vmcs02, will turn these bits off - and
1949 * nested_vmx_exit_handled() will not pass related exits to L1.
1950 * These rules have exceptions below.
1951 */
1952
1953 /* pin-based controls */
1954 /*
1955 * According to the Intel spec, if bit 55 of VMX_BASIC is off (as it is
1956 * in our case), bits 1, 2 and 4 (i.e., 0x16) must be 1 in this MSR.
1957 */
1958 nested_vmx_pinbased_ctls_low = 0x16 ;
1959 nested_vmx_pinbased_ctls_high = 0x16 |
1960 PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING |
1961 PIN_BASED_VIRTUAL_NMIS;
1962
1963 /* exit controls */
1964 nested_vmx_exit_ctls_low = 0;
b6f1250e 1965 /* Note that guest use of VM_EXIT_ACK_INTR_ON_EXIT is not supported. */
b87a51ae
NHE		 1966#ifdef CONFIG_X86_64
1967 nested_vmx_exit_ctls_high = VM_EXIT_HOST_ADDR_SPACE_SIZE;
1968#else
1969 nested_vmx_exit_ctls_high = 0;
1970#endif
1971
1972 /* entry controls */
1973 rdmsr(MSR_IA32_VMX_ENTRY_CTLS,
1974 nested_vmx_entry_ctls_low, nested_vmx_entry_ctls_high);
1975 nested_vmx_entry_ctls_low = 0;
1976 nested_vmx_entry_ctls_high &=
1977 VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_IA32E_MODE;
1978
1979 /* cpu-based controls */
1980 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS,
1981 nested_vmx_procbased_ctls_low, nested_vmx_procbased_ctls_high);
1982 nested_vmx_procbased_ctls_low = 0;
1983 nested_vmx_procbased_ctls_high &=
1984 CPU_BASED_VIRTUAL_INTR_PENDING | CPU_BASED_USE_TSC_OFFSETING |
1985 CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
1986 CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
1987 CPU_BASED_CR3_STORE_EXITING |
1988#ifdef CONFIG_X86_64
1989 CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
1990#endif
1991 CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
1992 CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_EXITING |
dbcb4e79 1993 CPU_BASED_RDPMC_EXITING | CPU_BASED_RDTSC_EXITING |
b87a51ae
NHE		 1994 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
1995 /*
1996 * We can allow some features even when not supported by the
1997 * hardware. For example, L1 can specify an MSR bitmap - and we
1998 * can use it to avoid exits to L1 - even when L0 runs L2
1999 * without MSR bitmaps.
2000 */
2001 nested_vmx_procbased_ctls_high |= CPU_BASED_USE_MSR_BITMAPS;
2002
2003 /* secondary cpu-based controls */
2004 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
2005 nested_vmx_secondary_ctls_low, nested_vmx_secondary_ctls_high);
2006 nested_vmx_secondary_ctls_low = 0;
2007 nested_vmx_secondary_ctls_high &=
2008 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
2009}
2010
2011static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
2012{
2013 /*
2014 * Bits 0 in high must be 0, and bits 1 in low must be 1.
2015 */
2016 return ((control & high) | low) == control;
2017}
2018
2019static inline u64 vmx_control_msr(u32 low, u32 high)
2020{
2021 return low | ((u64)high << 32);
2022}
2023
2024/*
2025 * If we allow our guest to use VMX instructions (i.e., nested VMX), we should
2026 * also let it use VMX-specific MSRs.
2027 * vmx_get_vmx_msr() and vmx_set_vmx_msr() return 1 when we handled a
2028 * VMX-specific MSR, or 0 when we haven't (and the caller should handle it
2029 * like all other MSRs).
2030 */
2031static int vmx_get_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2032{
2033 if (!nested_vmx_allowed(vcpu) && msr_index >= MSR_IA32_VMX_BASIC &&
2034 msr_index <= MSR_IA32_VMX_TRUE_ENTRY_CTLS) {
2035 /*
2036 * According to the spec, processors which do not support VMX
2037 * should throw a #GP(0) when VMX capability MSRs are read.
2038 */
2039 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
2040 return 1;
2041 }
2042
2043 switch (msr_index) {
2044 case MSR_IA32_FEATURE_CONTROL:
2045 *pdata = 0;
2046 break;
2047 case MSR_IA32_VMX_BASIC:
2048 /*
2049 * This MSR reports some information about VMX support. We
2050 * should return information about the VMX we emulate for the
2051 * guest, and the VMCS structure we give it - not about the
2052 * VMX support of the underlying hardware.
2053 */
2054 *pdata = VMCS12_REVISION |
2055 ((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
2056 (VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
2057 break;
2058 case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
2059 case MSR_IA32_VMX_PINBASED_CTLS:
2060 *pdata = vmx_control_msr(nested_vmx_pinbased_ctls_low,
2061 nested_vmx_pinbased_ctls_high);
2062 break;
2063 case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
2064 case MSR_IA32_VMX_PROCBASED_CTLS:
2065 *pdata = vmx_control_msr(nested_vmx_procbased_ctls_low,
2066 nested_vmx_procbased_ctls_high);
2067 break;
2068 case MSR_IA32_VMX_TRUE_EXIT_CTLS:
2069 case MSR_IA32_VMX_EXIT_CTLS:
2070 *pdata = vmx_control_msr(nested_vmx_exit_ctls_low,
2071 nested_vmx_exit_ctls_high);
2072 break;
2073 case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
2074 case MSR_IA32_VMX_ENTRY_CTLS:
2075 *pdata = vmx_control_msr(nested_vmx_entry_ctls_low,
2076 nested_vmx_entry_ctls_high);
2077 break;
2078 case MSR_IA32_VMX_MISC:
2079 *pdata = 0;
2080 break;
2081 /*
2082 * These MSRs specify bits which the guest must keep fixed (on or off)
2083 * while L1 is in VMXON mode (in L1's root mode, or running an L2).
2084 * We picked the standard core2 setting.
2085 */
2086#define VMXON_CR0_ALWAYSON (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
2087#define VMXON_CR4_ALWAYSON X86_CR4_VMXE
2088 case MSR_IA32_VMX_CR0_FIXED0:
2089 *pdata = VMXON_CR0_ALWAYSON;
2090 break;
2091 case MSR_IA32_VMX_CR0_FIXED1:
2092 *pdata = -1ULL;
2093 break;
2094 case MSR_IA32_VMX_CR4_FIXED0:
2095 *pdata = VMXON_CR4_ALWAYSON;
2096 break;
2097 case MSR_IA32_VMX_CR4_FIXED1:
2098 *pdata = -1ULL;
2099 break;
2100 case MSR_IA32_VMX_VMCS_ENUM:
2101 *pdata = 0x1f;
2102 break;
2103 case MSR_IA32_VMX_PROCBASED_CTLS2:
2104 *pdata = vmx_control_msr(nested_vmx_secondary_ctls_low,
2105 nested_vmx_secondary_ctls_high);
2106 break;
2107 case MSR_IA32_VMX_EPT_VPID_CAP:
2108 /* Currently, no nested ept or nested vpid */
2109 *pdata = 0;
2110 break;
2111 default:
2112 return 0;
2113 }
2114
2115 return 1;
2116}
2117
2118static int vmx_set_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
2119{
2120 if (!nested_vmx_allowed(vcpu))
2121 return 0;
2122
2123 if (msr_index == MSR_IA32_FEATURE_CONTROL)
2124 /* TODO: the right thing. */
2125 return 1;
2126 /*
2127 * No need to treat VMX capability MSRs specially: If we don't handle
2128 * them, handle_wrmsr will #GP(0), which is correct (they are readonly)
2129 */
2130 return 0;
2131}
2132
6aa8b732
AK		 2133/*
2134 * Reads an msr value (of 'msr_index') into 'pdata'.
2135 * Returns 0 on success, non-0 otherwise.
2136 * Assumes vcpu_load() was already called.
2137 */
2138static int vmx_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2139{
2140 u64 data;
26bb0981 2141 struct shared_msr_entry *msr;
6aa8b732
AK		 2142
2143 if (!pdata) {
2144 printk(KERN_ERR "BUG: get_msr called with NULL pdata\n");
2145 return -EINVAL;
2146 }
2147
2148 switch (msr_index) {
05b3e0c2 2149#ifdef CONFIG_X86_64
6aa8b732
AK		 2150 case MSR_FS_BASE:
2151 data = vmcs_readl(GUEST_FS_BASE);
2152 break;
2153 case MSR_GS_BASE:
2154 data = vmcs_readl(GUEST_GS_BASE);
2155 break;
44ea2b17
AK		 2156 case MSR_KERNEL_GS_BASE:
2157 vmx_load_host_state(to_vmx(vcpu));
2158 data = to_vmx(vcpu)->msr_guest_kernel_gs_base;
2159 break;
26bb0981 2160#endif
6aa8b732 2161 case MSR_EFER:
3bab1f5d 2162 return kvm_get_msr_common(vcpu, msr_index, pdata);
af24a4e4 2163 case MSR_IA32_TSC:
6aa8b732
AK		 2164 data = guest_read_tsc();
2165 break;
2166 case MSR_IA32_SYSENTER_CS:
2167 data = vmcs_read32(GUEST_SYSENTER_CS);
2168 break;
2169 case MSR_IA32_SYSENTER_EIP:
f5b42c33 2170 data = vmcs_readl(GUEST_SYSENTER_EIP);
6aa8b732
AK		 2171 break;
2172 case MSR_IA32_SYSENTER_ESP:
f5b42c33 2173 data = vmcs_readl(GUEST_SYSENTER_ESP);
6aa8b732 2174 break;
4e47c7a6
SY		 2175 case MSR_TSC_AUX:
2176 if (!to_vmx(vcpu)->rdtscp_enabled)
2177 return 1;
2178 /* Otherwise falls through */
6aa8b732 2179 default:
b87a51ae
NHE		 2180 if (vmx_get_vmx_msr(vcpu, msr_index, pdata))
2181 return 0;
8b9cf98c 2182 msr = find_msr_entry(to_vmx(vcpu), msr_index);
3bab1f5d
AK		 2183 if (msr) {
2184 data = msr->data;
2185 break;
6aa8b732 2186 }
3bab1f5d 2187 return kvm_get_msr_common(vcpu, msr_index, pdata);
6aa8b732
AK		 2188 }
2189
2190 *pdata = data;
2191 return 0;
2192}
2193
2194/*
2195 * Writes msr value into into the appropriate "register".
2196 * Returns 0 on success, non-0 otherwise.
2197 * Assumes vcpu_load() was already called.
2198 */
2199static int vmx_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
2200{
a2fa3e9f 2201 struct vcpu_vmx *vmx = to_vmx(vcpu);
26bb0981 2202 struct shared_msr_entry *msr;
2cc51560
ED		 2203 int ret = 0;
2204
6aa8b732 2205 switch (msr_index) {
3bab1f5d 2206 case MSR_EFER:
2cc51560 2207 ret = kvm_set_msr_common(vcpu, msr_index, data);
2cc51560 2208 break;
16175a79 2209#ifdef CONFIG_X86_64
6aa8b732 2210 case MSR_FS_BASE:
2fb92db1 2211 vmx_segment_cache_clear(vmx);
6aa8b732
AK		 2212 vmcs_writel(GUEST_FS_BASE, data);
2213 break;
2214 case MSR_GS_BASE:
2fb92db1 2215 vmx_segment_cache_clear(vmx);
6aa8b732
AK		 2216 vmcs_writel(GUEST_GS_BASE, data);
2217 break;
44ea2b17
AK		 2218 case MSR_KERNEL_GS_BASE:
2219 vmx_load_host_state(vmx);
2220 vmx->msr_guest_kernel_gs_base = data;
2221 break;
6aa8b732
AK		 2222#endif
2223 case MSR_IA32_SYSENTER_CS:
2224 vmcs_write32(GUEST_SYSENTER_CS, data);
2225 break;
2226 case MSR_IA32_SYSENTER_EIP:
f5b42c33 2227 vmcs_writel(GUEST_SYSENTER_EIP, data);
6aa8b732
AK		 2228 break;
2229 case MSR_IA32_SYSENTER_ESP:
f5b42c33 2230 vmcs_writel(GUEST_SYSENTER_ESP, data);
6aa8b732 2231 break;
af24a4e4 2232 case MSR_IA32_TSC:
99e3e30a 2233 kvm_write_tsc(vcpu, data);
6aa8b732 2234 break;
468d472f
SY		 2235 case MSR_IA32_CR_PAT:
2236 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2237 vmcs_write64(GUEST_IA32_PAT, data);
2238 vcpu->arch.pat = data;
2239 break;
2240 }
4e47c7a6
SY		 2241 ret = kvm_set_msr_common(vcpu, msr_index, data);
2242 break;
2243 case MSR_TSC_AUX:
2244 if (!vmx->rdtscp_enabled)
2245 return 1;
2246 /* Check reserved bit, higher 32 bits should be zero */
2247 if ((data >> 32) != 0)
2248 return 1;
2249 /* Otherwise falls through */
6aa8b732 2250 default:
b87a51ae
NHE		 2251 if (vmx_set_vmx_msr(vcpu, msr_index, data))
2252 break;
8b9cf98c 2253 msr = find_msr_entry(vmx, msr_index);
3bab1f5d
AK		 2254 if (msr) {
2255 msr->data = data;
2225fd56
AK		 2256 if (msr - vmx->guest_msrs < vmx->save_nmsrs) {
2257 preempt_disable();
9ee73970
AK		 2258 kvm_set_shared_msr(msr->index, msr->data,
2259 msr->mask);
2225fd56
AK		 2260 preempt_enable();
2261 }
3bab1f5d 2262 break;
6aa8b732 2263 }
2cc51560 2264 ret = kvm_set_msr_common(vcpu, msr_index, data);
6aa8b732
AK		 2265 }
2266
2cc51560 2267 return ret;
6aa8b732
AK		 2268}
2269
5fdbf976 2270static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
6aa8b732 2271{
5fdbf976
MT		 2272 __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
2273 switch (reg) {
2274 case VCPU_REGS_RSP:
2275 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
2276 break;
2277 case VCPU_REGS_RIP:
2278 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
2279 break;
6de4f3ad
AK		 2280 case VCPU_EXREG_PDPTR:
2281 if (enable_ept)
2282 ept_save_pdptrs(vcpu);
2283 break;
5fdbf976
MT		 2284 default:
2285 break;
2286 }
6aa8b732
AK		 2287}
2288
355be0b9 2289static void set_guest_debug(struct kvm_vcpu *vcpu, struct kvm_guest_debug *dbg)
6aa8b732 2290{
ae675ef0
JK		 2291 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
2292 vmcs_writel(GUEST_DR7, dbg->arch.debugreg[7]);
2293 else
2294 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
2295
abd3f2d6 2296 update_exception_bitmap(vcpu);
6aa8b732
AK		 2297}
2298
2299static __init int cpu_has_kvm_support(void)
2300{
6210e37b 2301 return cpu_has_vmx();
6aa8b732
AK		 2302}
2303
2304static __init int vmx_disabled_by_bios(void)
2305{
2306 u64 msr;
2307
2308 rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
cafd6659 2309 if (msr & FEATURE_CONTROL_LOCKED) {
23f3e991 2310 /* launched w/ TXT and VMX disabled */
cafd6659
SW		 2311 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2312 && tboot_enabled())
2313 return 1;
23f3e991 2314 /* launched w/o TXT and VMX only enabled w/ TXT */
cafd6659 2315 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
23f3e991 2316 && (msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
f9335afe
SW		 2317 && !tboot_enabled()) {
2318 printk(KERN_WARNING "kvm: disable TXT in the BIOS or "
23f3e991 2319 "activate TXT before enabling KVM\n");
cafd6659 2320 return 1;
f9335afe 2321 }
23f3e991
JC		 2322 /* launched w/o TXT and VMX disabled */
2323 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2324 && !tboot_enabled())
2325 return 1;
cafd6659
SW		 2326 }
2327
2328 return 0;
6aa8b732
AK		 2329}
2330
7725b894
DX		 2331static void kvm_cpu_vmxon(u64 addr)
2332{
2333 asm volatile (ASM_VMX_VMXON_RAX
2334 : : "a"(&addr), "m"(addr)
2335 : "memory", "cc");
2336}
2337
10474ae8 2338static int hardware_enable(void *garbage)
6aa8b732
AK		 2339{
2340 int cpu = raw_smp_processor_id();
2341 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
cafd6659 2342 u64 old, test_bits;
6aa8b732 2343
10474ae8
AG		 2344 if (read_cr4() & X86_CR4_VMXE)
2345 return -EBUSY;
2346
d462b819 2347 INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
6aa8b732 2348 rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
cafd6659
SW		 2349
2350 test_bits = FEATURE_CONTROL_LOCKED;
2351 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
2352 if (tboot_enabled())
2353 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX;
2354
2355 if ((old & test_bits) != test_bits) {
6aa8b732 2356 /* enable and lock */
cafd6659
SW		 2357 wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits);
2358 }
66aee91a 2359 write_cr4(read_cr4() | X86_CR4_VMXE); /* FIXME: not cpu hotplug safe */
10474ae8 2360
4610c9cc
DX		 2361 if (vmm_exclusive) {
2362 kvm_cpu_vmxon(phys_addr);
2363 ept_sync_global();
2364 }
10474ae8 2365
3444d7da
AK		 2366 store_gdt(&__get_cpu_var(host_gdt));
2367
10474ae8 2368 return 0;
6aa8b732
AK		 2369}
2370
d462b819 2371static void vmclear_local_loaded_vmcss(void)
543e4243
AK		 2372{
2373 int cpu = raw_smp_processor_id();
d462b819 2374 struct loaded_vmcs *v, *n;
543e4243 2375
d462b819
NHE		 2376 list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
2377 loaded_vmcss_on_cpu_link)
2378 __loaded_vmcs_clear(v);
543e4243
AK		 2379}
2380
710ff4a8
EH		 2381
2382/* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
2383 * tricks.
2384 */
2385static void kvm_cpu_vmxoff(void)
6aa8b732 2386{
4ecac3fd 2387 asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
6aa8b732
AK		 2388}
2389
710ff4a8
EH		 2390static void hardware_disable(void *garbage)
2391{
4610c9cc 2392 if (vmm_exclusive) {
d462b819 2393 vmclear_local_loaded_vmcss();
4610c9cc
DX		 2394 kvm_cpu_vmxoff();
2395 }
7725b894 2396 write_cr4(read_cr4() & ~X86_CR4_VMXE);
710ff4a8
EH		 2397}
2398
1c3d14fe 2399static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
d77c26fc 2400 u32 msr, u32 *result)
1c3d14fe
YS		 2401{
2402 u32 vmx_msr_low, vmx_msr_high;
2403 u32 ctl = ctl_min | ctl_opt;
2404
2405 rdmsr(msr, vmx_msr_low, vmx_msr_high);
2406
2407 ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
2408 ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */
2409
2410 /* Ensure minimum (required) set of control bits are supported. */
2411 if (ctl_min & ~ctl)
002c7f7c 2412 return -EIO;
1c3d14fe
YS		 2413
2414 *result = ctl;
2415 return 0;
2416}
2417
110312c8
AK		 2418static __init bool allow_1_setting(u32 msr, u32 ctl)
2419{
2420 u32 vmx_msr_low, vmx_msr_high;
2421
2422 rdmsr(msr, vmx_msr_low, vmx_msr_high);
2423 return vmx_msr_high & ctl;
2424}
2425
002c7f7c 2426static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
6aa8b732
AK		 2427{
2428 u32 vmx_msr_low, vmx_msr_high;
d56f546d 2429 u32 min, opt, min2, opt2;
1c3d14fe
YS		 2430 u32 _pin_based_exec_control = 0;
2431 u32 _cpu_based_exec_control = 0;
f78e0e2e 2432 u32 _cpu_based_2nd_exec_control = 0;
1c3d14fe
YS		 2433 u32 _vmexit_control = 0;
2434 u32 _vmentry_control = 0;
2435
2436 min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
f08864b4 2437 opt = PIN_BASED_VIRTUAL_NMIS;
1c3d14fe
YS		 2438 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
2439 &_pin_based_exec_control) < 0)
002c7f7c 2440 return -EIO;
1c3d14fe 2441
10166744 2442 min = CPU_BASED_HLT_EXITING |
1c3d14fe
YS		 2443#ifdef CONFIG_X86_64
2444 CPU_BASED_CR8_LOAD_EXITING |
2445 CPU_BASED_CR8_STORE_EXITING |
2446#endif
d56f546d
SY		 2447 CPU_BASED_CR3_LOAD_EXITING |
2448 CPU_BASED_CR3_STORE_EXITING |
1c3d14fe
YS		 2449 CPU_BASED_USE_IO_BITMAPS |
2450 CPU_BASED_MOV_DR_EXITING |
a7052897 2451 CPU_BASED_USE_TSC_OFFSETING |
59708670
SY		 2452 CPU_BASED_MWAIT_EXITING |
2453 CPU_BASED_MONITOR_EXITING |
fee84b07
AK		 2454 CPU_BASED_INVLPG_EXITING |
2455 CPU_BASED_RDPMC_EXITING;
443381a8 2456
f78e0e2e 2457 opt = CPU_BASED_TPR_SHADOW |
25c5f225 2458 CPU_BASED_USE_MSR_BITMAPS |
f78e0e2e 2459 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
1c3d14fe
YS		 2460 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
2461 &_cpu_based_exec_control) < 0)
002c7f7c 2462 return -EIO;
6e5d865c
YS		 2463#ifdef CONFIG_X86_64
2464 if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2465 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
2466 ~CPU_BASED_CR8_STORE_EXITING;
2467#endif
f78e0e2e 2468 if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
d56f546d
SY		 2469 min2 = 0;
2470 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2384d2b3 2471 SECONDARY_EXEC_WBINVD_EXITING |
d56f546d 2472 SECONDARY_EXEC_ENABLE_VPID |
3a624e29 2473 SECONDARY_EXEC_ENABLE_EPT |
4b8d54f9 2474 SECONDARY_EXEC_UNRESTRICTED_GUEST |
4e47c7a6 2475 SECONDARY_EXEC_PAUSE_LOOP_EXITING |
ad756a16
MJ		 2476 SECONDARY_EXEC_RDTSCP |
2477 SECONDARY_EXEC_ENABLE_INVPCID;
d56f546d
SY		 2478 if (adjust_vmx_controls(min2, opt2,
2479 MSR_IA32_VMX_PROCBASED_CTLS2,
f78e0e2e
SY		 2480 &_cpu_based_2nd_exec_control) < 0)
2481 return -EIO;
2482 }
2483#ifndef CONFIG_X86_64
2484 if (!(_cpu_based_2nd_exec_control &
2485 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
2486 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
2487#endif
d56f546d 2488 if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
a7052897
MT		 2489 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
2490 enabled */
5fff7d27
GN		 2491 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
2492 CPU_BASED_CR3_STORE_EXITING |
2493 CPU_BASED_INVLPG_EXITING);
d56f546d
SY		 2494 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
2495 vmx_capability.ept, vmx_capability.vpid);
2496 }
1c3d14fe
YS		 2497
2498 min = 0;
2499#ifdef CONFIG_X86_64
2500 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
2501#endif
468d472f 2502 opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT;
1c3d14fe
YS		 2503 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
2504 &_vmexit_control) < 0)
002c7f7c 2505 return -EIO;
1c3d14fe 2506
468d472f
SY		 2507 min = 0;
2508 opt = VM_ENTRY_LOAD_IA32_PAT;
1c3d14fe
YS		 2509 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
2510 &_vmentry_control) < 0)
002c7f7c 2511 return -EIO;
6aa8b732 2512
c68876fd 2513 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
1c3d14fe
YS		 2514
2515 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
2516 if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
002c7f7c 2517 return -EIO;
1c3d14fe
YS		 2518
2519#ifdef CONFIG_X86_64
2520 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
2521 if (vmx_msr_high & (1u<<16))
002c7f7c 2522 return -EIO;
1c3d14fe
YS		 2523#endif
2524
2525 /* Require Write-Back (WB) memory type for VMCS accesses. */
2526 if (((vmx_msr_high >> 18) & 15) != 6)
002c7f7c 2527 return -EIO;
1c3d14fe 2528
002c7f7c
YS		 2529 vmcs_conf->size = vmx_msr_high & 0x1fff;
2530 vmcs_conf->order = get_order(vmcs_config.size);
2531 vmcs_conf->revision_id = vmx_msr_low;
1c3d14fe 2532
002c7f7c
YS		 2533 vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
2534 vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
f78e0e2e 2535 vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
002c7f7c
YS		 2536 vmcs_conf->vmexit_ctrl = _vmexit_control;
2537 vmcs_conf->vmentry_ctrl = _vmentry_control;
1c3d14fe 2538
110312c8
AK		 2539 cpu_has_load_ia32_efer =
2540 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
2541 VM_ENTRY_LOAD_IA32_EFER)
2542 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
2543 VM_EXIT_LOAD_IA32_EFER);
2544
8bf00a52
GN		 2545 cpu_has_load_perf_global_ctrl =
2546 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
2547 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
2548 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
2549 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
2550
2551 /*
2552 * Some cpus support VM_ENTRY_(LOAD|SAVE)_IA32_PERF_GLOBAL_CTRL
2553 * but due to arrata below it can't be used. Workaround is to use
2554 * msr load mechanism to switch IA32_PERF_GLOBAL_CTRL.
2555 *
2556 * VM Exit May Incorrectly Clear IA32_PERF_GLOBAL_CTRL [34:32]
2557 *
2558 * AAK155 (model 26)
2559 * AAP115 (model 30)
2560 * AAT100 (model 37)
2561 * BC86,AAY89,BD102 (model 44)
2562 * BA97 (model 46)
2563 *
2564 */
2565 if (cpu_has_load_perf_global_ctrl && boot_cpu_data.x86 == 0x6) {
2566 switch (boot_cpu_data.x86_model) {
2567 case 26:
2568 case 30:
2569 case 37:
2570 case 44:
2571 case 46:
2572 cpu_has_load_perf_global_ctrl = false;
2573 printk_once(KERN_WARNING"kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
2574 "does not work properly. Using workaround\n");
2575 break;
2576 default:
2577 break;
2578 }
2579 }
2580
1c3d14fe 2581 return 0;
c68876fd 2582}
6aa8b732
AK		 2583
2584static struct vmcs *alloc_vmcs_cpu(int cpu)
2585{
2586 int node = cpu_to_node(cpu);
2587 struct page *pages;
2588 struct vmcs *vmcs;
2589
6484eb3e 2590 pages = alloc_pages_exact_node(node, GFP_KERNEL, vmcs_config.order);
6aa8b732
AK		 2591 if (!pages)
2592 return NULL;
2593 vmcs = page_address(pages);
1c3d14fe
YS		 2594 memset(vmcs, 0, vmcs_config.size);
2595 vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
6aa8b732
AK		 2596 return vmcs;
2597}
2598
2599static struct vmcs *alloc_vmcs(void)
2600{
d3b2c338 2601 return alloc_vmcs_cpu(raw_smp_processor_id());
6aa8b732
AK		 2602}
2603
2604static void free_vmcs(struct vmcs *vmcs)
2605{
1c3d14fe 2606 free_pages((unsigned long)vmcs, vmcs_config.order);
6aa8b732
AK		 2607}
2608
d462b819
NHE		 2609/*
2610 * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
2611 */
2612static void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2613{
2614 if (!loaded_vmcs->vmcs)
2615 return;
2616 loaded_vmcs_clear(loaded_vmcs);
2617 free_vmcs(loaded_vmcs->vmcs);
2618 loaded_vmcs->vmcs = NULL;
2619}
2620
39959588 2621static void free_kvm_area(void)
6aa8b732
AK		 2622{
2623 int cpu;
2624
3230bb47 2625 for_each_possible_cpu(cpu) {
6aa8b732 2626 free_vmcs(per_cpu(vmxarea, cpu));
3230bb47
ZA		 2627 per_cpu(vmxarea, cpu) = NULL;
2628 }
6aa8b732
AK		 2629}
2630
6aa8b732
AK		 2631static __init int alloc_kvm_area(void)
2632{
2633 int cpu;
2634
3230bb47 2635 for_each_possible_cpu(cpu) {
6aa8b732
AK		 2636 struct vmcs *vmcs;
2637
2638 vmcs = alloc_vmcs_cpu(cpu);
2639 if (!vmcs) {
2640 free_kvm_area();
2641 return -ENOMEM;
2642 }
2643
2644 per_cpu(vmxarea, cpu) = vmcs;
2645 }
2646 return 0;
2647}
2648
2649static __init int hardware_setup(void)
2650{
002c7f7c
YS		 2651 if (setup_vmcs_config(&vmcs_config) < 0)
2652 return -EIO;
50a37eb4
JR		 2653
2654 if (boot_cpu_has(X86_FEATURE_NX))
2655 kvm_enable_efer_bits(EFER_NX);
2656
93ba03c2
SY		 2657 if (!cpu_has_vmx_vpid())
2658 enable_vpid = 0;
2659
4bc9b982
SY		 2660 if (!cpu_has_vmx_ept() ||
2661 !cpu_has_vmx_ept_4levels()) {
93ba03c2 2662 enable_ept = 0;
3a624e29 2663 enable_unrestricted_guest = 0;
83c3a331 2664 enable_ept_ad_bits = 0;
3a624e29
NK		 2665 }
2666
83c3a331
XH		 2667 if (!cpu_has_vmx_ept_ad_bits())
2668 enable_ept_ad_bits = 0;
2669
3a624e29
NK		 2670 if (!cpu_has_vmx_unrestricted_guest())
2671 enable_unrestricted_guest = 0;
93ba03c2
SY		 2672
2673 if (!cpu_has_vmx_flexpriority())
2674 flexpriority_enabled = 0;
2675
95ba8273
GN		 2676 if (!cpu_has_vmx_tpr_shadow())
2677 kvm_x86_ops->update_cr8_intercept = NULL;
2678
54dee993
MT		 2679 if (enable_ept && !cpu_has_vmx_ept_2m_page())
2680 kvm_disable_largepages();
2681
4b8d54f9
ZE		 2682 if (!cpu_has_vmx_ple())
2683 ple_gap = 0;
2684
b87a51ae
NHE		 2685 if (nested)
2686 nested_vmx_setup_ctls_msrs();
2687
6aa8b732
AK		 2688 return alloc_kvm_area();
2689}
2690
2691static __exit void hardware_unsetup(void)
2692{
2693 free_kvm_area();
2694}
2695
f5f7b2fe 2696static void fix_pmode_dataseg(struct kvm_vcpu *vcpu, int seg, struct kvm_segment *save)
6aa8b732 2697{
772e0318 2698 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
c865c43d 2699 struct kvm_segment tmp = *save;
6aa8b732 2700
c865c43d
AK		 2701 if (!(vmcs_readl(sf->base) == tmp.base && tmp.s)) {
2702 tmp.base = vmcs_readl(sf->base);
2703 tmp.selector = vmcs_read16(sf->selector);
2704 tmp.s = 1;
6aa8b732 2705 }
c865c43d 2706 vmx_set_segment(vcpu, &tmp, seg);
6aa8b732
AK		 2707}
2708
2709static void enter_pmode(struct kvm_vcpu *vcpu)
2710{
2711 unsigned long flags;
a89a8fb9 2712 struct vcpu_vmx *vmx = to_vmx(vcpu);
6aa8b732 2713
a89a8fb9 2714 vmx->emulation_required = 1;
7ffd92c5 2715 vmx->rmode.vm86_active = 0;
6aa8b732 2716
2fb92db1
AK		 2717 vmx_segment_cache_clear(vmx);
2718
f5f7b2fe 2719 vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
6aa8b732
AK		 2720
2721 flags = vmcs_readl(GUEST_RFLAGS);
78ac8b47
AK		 2722 flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
2723 flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
6aa8b732
AK		 2724 vmcs_writel(GUEST_RFLAGS, flags);
2725
66aee91a
RR		 2726 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
2727 (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
6aa8b732
AK		 2728
2729 update_exception_bitmap(vcpu);
2730
a89a8fb9
MG		 2731 if (emulate_invalid_guest_state)
2732 return;
2733
f5f7b2fe
AK		 2734 fix_pmode_dataseg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
2735 fix_pmode_dataseg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
2736 fix_pmode_dataseg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
2737 fix_pmode_dataseg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
6aa8b732 2738
2fb92db1
AK		 2739 vmx_segment_cache_clear(vmx);
2740
6aa8b732
AK		 2741 vmcs_write16(GUEST_SS_SELECTOR, 0);
2742 vmcs_write32(GUEST_SS_AR_BYTES, 0x93);
2743
2744 vmcs_write16(GUEST_CS_SELECTOR,
2745 vmcs_read16(GUEST_CS_SELECTOR) & ~SELECTOR_RPL_MASK);
2746 vmcs_write32(GUEST_CS_AR_BYTES, 0x9b);
2747}
2748
d77c26fc 2749static gva_t rmode_tss_base(struct kvm *kvm)
6aa8b732 2750{
bfc6d222 2751 if (!kvm->arch.tss_addr) {
bc6678a3 2752 struct kvm_memslots *slots;
28a37544 2753 struct kvm_memory_slot *slot;
bc6678a3
MT		 2754 gfn_t base_gfn;
2755
90d83dc3 2756 slots = kvm_memslots(kvm);
28a37544
XG		 2757 slot = id_to_memslot(slots, 0);
2758 base_gfn = slot->base_gfn + slot->npages - 3;
2759
cbc94022
IE		 2760 return base_gfn << PAGE_SHIFT;
2761 }
bfc6d222 2762 return kvm->arch.tss_addr;
6aa8b732
AK		 2763}
2764
f5f7b2fe 2765static void fix_rmode_seg(int seg, struct kvm_segment *save)
baa7e81e 2766{
772e0318 2767 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
baa7e81e 2768
15b00f32 2769 vmcs_write16(sf->selector, save->base >> 4);
444e863d 2770 vmcs_write32(sf->base, save->base & 0xffff0);
6aa8b732
AK		 2771 vmcs_write32(sf->limit, 0xffff);
2772 vmcs_write32(sf->ar_bytes, 0xf3);
444e863d
GN		 2773 if (save->base & 0xf)
2774 printk_once(KERN_WARNING "kvm: segment base is not paragraph"
2775 " aligned when entering protected mode (seg=%d)",
2776 seg);
6aa8b732
AK		 2777}
2778
2779static void enter_rmode(struct kvm_vcpu *vcpu)
2780{
2781 unsigned long flags;
a89a8fb9 2782 struct vcpu_vmx *vmx = to_vmx(vcpu);
b246dd5d 2783 struct kvm_segment var;
6aa8b732 2784
3a624e29
NK		 2785 if (enable_unrestricted_guest)
2786 return;
2787
f5f7b2fe
AK		 2788 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
2789 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
2790 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
2791 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
2792 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
2793
a89a8fb9 2794 vmx->emulation_required = 1;
7ffd92c5 2795 vmx->rmode.vm86_active = 1;
6aa8b732 2796
baa7e81e 2797
776e58ea
GN		 2798 /*
2799 * Very old userspace does not call KVM_SET_TSS_ADDR before entering
2800 * vcpu. Call it here with phys address pointing 16M below 4G.
2801 */
2802 if (!vcpu->kvm->arch.tss_addr) {
2803 printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
2804 "called before entering vcpu\n");
2805 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
2806 vmx_set_tss_addr(vcpu->kvm, 0xfeffd000);
2807 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2808 }
2809
2fb92db1
AK		 2810 vmx_segment_cache_clear(vmx);
2811
6aa8b732 2812 vmcs_writel(GUEST_TR_BASE, rmode_tss_base(vcpu->kvm));
6aa8b732 2813 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
6aa8b732
AK		 2814 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
2815
2816 flags = vmcs_readl(GUEST_RFLAGS);
78ac8b47 2817 vmx->rmode.save_rflags = flags;
6aa8b732 2818
053de044 2819 flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
6aa8b732
AK		 2820
2821 vmcs_writel(GUEST_RFLAGS, flags);
66aee91a 2822 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
6aa8b732
AK		 2823 update_exception_bitmap(vcpu);
2824
a89a8fb9
MG		 2825 if (emulate_invalid_guest_state)
2826 goto continue_rmode;
2827
b246dd5d
OW		 2828 vmx_get_segment(vcpu, &var, VCPU_SREG_SS);
2829 vmx_set_segment(vcpu, &var, VCPU_SREG_SS);
2830
2831 vmx_get_segment(vcpu, &var, VCPU_SREG_CS);
2832 vmx_set_segment(vcpu, &var, VCPU_SREG_CS);
2833
2834 vmx_get_segment(vcpu, &var, VCPU_SREG_ES);
2835 vmx_set_segment(vcpu, &var, VCPU_SREG_ES);
2836
2837 vmx_get_segment(vcpu, &var, VCPU_SREG_DS);
2838 vmx_set_segment(vcpu, &var, VCPU_SREG_DS);
6aa8b732 2839
b246dd5d
OW		 2840 vmx_get_segment(vcpu, &var, VCPU_SREG_GS);
2841 vmx_set_segment(vcpu, &var, VCPU_SREG_GS);
6aa8b732 2842
b246dd5d
OW		 2843 vmx_get_segment(vcpu, &var, VCPU_SREG_FS);
2844 vmx_set_segment(vcpu, &var, VCPU_SREG_FS);
75880a01 2845
a89a8fb9 2846continue_rmode:
8668a3c4 2847 kvm_mmu_reset_context(vcpu);
6aa8b732
AK		 2848}
2849
401d10de
AS		 2850static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
2851{
2852 struct vcpu_vmx *vmx = to_vmx(vcpu);
26bb0981
AK		 2853 struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
2854
2855 if (!msr)
2856 return;
401d10de 2857
44ea2b17
AK		 2858 /*
2859 * Force kernel_gs_base reloading before EFER changes, as control
2860 * of this msr depends on is_long_mode().
2861 */
2862 vmx_load_host_state(to_vmx(vcpu));
f6801dff 2863 vcpu->arch.efer = efer;
401d10de
AS		 2864 if (efer & EFER_LMA) {
2865 vmcs_write32(VM_ENTRY_CONTROLS,
2866 vmcs_read32(VM_ENTRY_CONTROLS) |
2867 VM_ENTRY_IA32E_MODE);
2868 msr->data = efer;
2869 } else {
2870 vmcs_write32(VM_ENTRY_CONTROLS,
2871 vmcs_read32(VM_ENTRY_CONTROLS) &
2872 ~VM_ENTRY_IA32E_MODE);
2873
2874 msr->data = efer & ~EFER_LME;
2875 }
2876 setup_msrs(vmx);
2877}
2878
05b3e0c2 2879#ifdef CONFIG_X86_64
6aa8b732
AK		 2880
2881static void enter_lmode(struct kvm_vcpu *vcpu)
2882{
2883 u32 guest_tr_ar;
2884
2fb92db1
AK		 2885 vmx_segment_cache_clear(to_vmx(vcpu));
2886
6aa8b732
AK		 2887 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
2888 if ((guest_tr_ar & AR_TYPE_MASK) != AR_TYPE_BUSY_64_TSS) {
bd80158a
JK		 2889 pr_debug_ratelimited("%s: tss fixup for long mode. \n",
2890 __func__);
6aa8b732
AK		 2891 vmcs_write32(GUEST_TR_AR_BYTES,
2892 (guest_tr_ar & ~AR_TYPE_MASK)
2893 | AR_TYPE_BUSY_64_TSS);
2894 }
da38f438 2895 vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
6aa8b732
AK		 2896}
2897
2898static void exit_lmode(struct kvm_vcpu *vcpu)
2899{
6aa8b732
AK		 2900 vmcs_write32(VM_ENTRY_CONTROLS,
2901 vmcs_read32(VM_ENTRY_CONTROLS)
1e4e6e00 2902 & ~VM_ENTRY_IA32E_MODE);
da38f438 2903 vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
6aa8b732
AK		 2904}
2905
2906#endif
2907
2384d2b3
SY		 2908static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
2909{
b9d762fa 2910 vpid_sync_context(to_vmx(vcpu));
dd180b3e
XG		 2911 if (enable_ept) {
2912 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
2913 return;
4e1096d2 2914 ept_sync_context(construct_eptp(vcpu->arch.mmu.root_hpa));
dd180b3e 2915 }
2384d2b3
SY		 2916}
2917
e8467fda
AK		 2918static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
2919{
2920 ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
2921
2922 vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
2923 vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
2924}
2925
aff48baa
AK		 2926static void vmx_decache_cr3(struct kvm_vcpu *vcpu)
2927{
2928 if (enable_ept && is_paging(vcpu))
2929 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
2930 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
2931}
2932
25c4c276 2933static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
399badf3 2934{
fc78f519
AK		 2935 ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
2936
2937 vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
2938 vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
399badf3
AK		 2939}
2940
1439442c
SY		 2941static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
2942{
6de4f3ad
AK		 2943 if (!test_bit(VCPU_EXREG_PDPTR,
2944 (unsigned long *)&vcpu->arch.regs_dirty))
2945 return;
2946
1439442c 2947 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
ff03a073
JR		 2948 vmcs_write64(GUEST_PDPTR0, vcpu->arch.mmu.pdptrs[0]);
2949 vmcs_write64(GUEST_PDPTR1, vcpu->arch.mmu.pdptrs[1]);
2950 vmcs_write64(GUEST_PDPTR2, vcpu->arch.mmu.pdptrs[2]);
2951 vmcs_write64(GUEST_PDPTR3, vcpu->arch.mmu.pdptrs[3]);
1439442c
SY		 2952 }
2953}
2954
8f5d549f
AK		 2955static void ept_save_pdptrs(struct kvm_vcpu *vcpu)
2956{
2957 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
ff03a073
JR		 2958 vcpu->arch.mmu.pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
2959 vcpu->arch.mmu.pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
2960 vcpu->arch.mmu.pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
2961 vcpu->arch.mmu.pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
8f5d549f 2962 }
6de4f3ad
AK		 2963
2964 __set_bit(VCPU_EXREG_PDPTR,
2965 (unsigned long *)&vcpu->arch.regs_avail);
2966 __set_bit(VCPU_EXREG_PDPTR,
2967 (unsigned long *)&vcpu->arch.regs_dirty);
8f5d549f
AK		 2968}
2969
5e1746d6 2970static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
1439442c
SY		 2971
2972static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
2973 unsigned long cr0,
2974 struct kvm_vcpu *vcpu)
2975{
5233dd51
MT		 2976 if (!test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
2977 vmx_decache_cr3(vcpu);
1439442c
SY		 2978 if (!(cr0 & X86_CR0_PG)) {
2979 /* From paging/starting to nonpaging */
2980 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
65267ea1 2981 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
1439442c
SY		 2982 (CPU_BASED_CR3_LOAD_EXITING |
2983 CPU_BASED_CR3_STORE_EXITING));
2984 vcpu->arch.cr0 = cr0;
fc78f519 2985 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
1439442c
SY		 2986 } else if (!is_paging(vcpu)) {
2987 /* From nonpaging to paging */
2988 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
65267ea1 2989 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
1439442c
SY		 2990 ~(CPU_BASED_CR3_LOAD_EXITING |
2991 CPU_BASED_CR3_STORE_EXITING));
2992 vcpu->arch.cr0 = cr0;
fc78f519 2993 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
1439442c 2994 }
95eb84a7
SY		 2995
2996 if (!(cr0 & X86_CR0_WP))
2997 *hw_cr0 &= ~X86_CR0_WP;
1439442c
SY		 2998}
2999
6aa8b732
AK		 3000static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
3001{
7ffd92c5 3002 struct vcpu_vmx *vmx = to_vmx(vcpu);
3a624e29
NK		 3003 unsigned long hw_cr0;
3004
3005 if (enable_unrestricted_guest)
3006 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST)
3007 | KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
3008 else
3009 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK) | KVM_VM_CR0_ALWAYS_ON;
1439442c 3010
7ffd92c5 3011 if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
6aa8b732
AK		 3012 enter_pmode(vcpu);
3013
7ffd92c5 3014 if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
6aa8b732
AK		 3015 enter_rmode(vcpu);
3016
05b3e0c2 3017#ifdef CONFIG_X86_64
f6801dff 3018 if (vcpu->arch.efer & EFER_LME) {
707d92fa 3019 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
6aa8b732 3020 enter_lmode(vcpu);
707d92fa 3021 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
6aa8b732
AK		 3022 exit_lmode(vcpu);
3023 }
3024#endif
3025
089d034e 3026 if (enable_ept)
1439442c
SY		 3027 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
3028
02daab21 3029 if (!vcpu->fpu_active)
81231c69 3030 hw_cr0 |= X86_CR0_TS | X86_CR0_MP;
02daab21 3031
6aa8b732 3032 vmcs_writel(CR0_READ_SHADOW, cr0);
1439442c 3033 vmcs_writel(GUEST_CR0, hw_cr0);
ad312c7c 3034 vcpu->arch.cr0 = cr0;
69c73028 3035 __clear_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail);
6aa8b732
AK		 3036}
3037
1439442c
SY		 3038static u64 construct_eptp(unsigned long root_hpa)
3039{
3040 u64 eptp;
3041
3042 /* TODO write the value reading from MSR */
3043 eptp = VMX_EPT_DEFAULT_MT |
3044 VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT;
b38f9934
XH		 3045 if (enable_ept_ad_bits)
3046 eptp |= VMX_EPT_AD_ENABLE_BIT;
1439442c
SY		 3047 eptp |= (root_hpa & PAGE_MASK);
3048
3049 return eptp;
3050}
3051
6aa8b732
AK		 3052static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
3053{
1439442c
SY		 3054 unsigned long guest_cr3;
3055 u64 eptp;
3056
3057 guest_cr3 = cr3;
089d034e 3058 if (enable_ept) {
1439442c
SY		 3059 eptp = construct_eptp(cr3);
3060 vmcs_write64(EPT_POINTER, eptp);
9f8fe504 3061 guest_cr3 = is_paging(vcpu) ? kvm_read_cr3(vcpu) :
b927a3ce 3062 vcpu->kvm->arch.ept_identity_map_addr;
7c93be44 3063 ept_load_pdptrs(vcpu);
1439442c
SY		 3064 }
3065
2384d2b3 3066 vmx_flush_tlb(vcpu);
1439442c 3067 vmcs_writel(GUEST_CR3, guest_cr3);
6aa8b732
AK		 3068}
3069
5e1746d6 3070static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
6aa8b732 3071{
7ffd92c5 3072 unsigned long hw_cr4 = cr4 | (to_vmx(vcpu)->rmode.vm86_active ?
1439442c
SY		 3073 KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
3074
5e1746d6
NHE		 3075 if (cr4 & X86_CR4_VMXE) {
3076 /*
3077 * To use VMXON (and later other VMX instructions), a guest
3078 * must first be able to turn on cr4.VMXE (see handle_vmon()).
3079 * So basically the check on whether to allow nested VMX
3080 * is here.
3081 */
3082 if (!nested_vmx_allowed(vcpu))
3083 return 1;
3084 } else if (to_vmx(vcpu)->nested.vmxon)
3085 return 1;
3086
ad312c7c 3087 vcpu->arch.cr4 = cr4;
bc23008b
AK		 3088 if (enable_ept) {
3089 if (!is_paging(vcpu)) {
3090 hw_cr4 &= ~X86_CR4_PAE;
3091 hw_cr4 |= X86_CR4_PSE;
3092 } else if (!(cr4 & X86_CR4_PAE)) {
3093 hw_cr4 &= ~X86_CR4_PAE;
3094 }
3095 }
1439442c
SY		 3096
3097 vmcs_writel(CR4_READ_SHADOW, cr4);
3098 vmcs_writel(GUEST_CR4, hw_cr4);
5e1746d6 3099 return 0;
6aa8b732
AK		 3100}
3101
6aa8b732
AK		 3102static void vmx_get_segment(struct kvm_vcpu *vcpu,
3103 struct kvm_segment *var, int seg)
3104{
a9179499 3105 struct vcpu_vmx *vmx = to_vmx(vcpu);
6aa8b732
AK		 3106 u32 ar;
3107
a9179499
AK		 3108 if (vmx->rmode.vm86_active
3109 && (seg == VCPU_SREG_TR || seg == VCPU_SREG_ES
3110 || seg == VCPU_SREG_DS || seg == VCPU_SREG_FS
72636420 3111 || seg == VCPU_SREG_GS)) {
f5f7b2fe 3112 *var = vmx->rmode.segs[seg];
a9179499 3113 if (seg == VCPU_SREG_TR
2fb92db1 3114 || var->selector == vmx_read_guest_seg_selector(vmx, seg))
f5f7b2fe 3115 return;
1390a28b
AK		 3116 var->base = vmx_read_guest_seg_base(vmx, seg);
3117 var->selector = vmx_read_guest_seg_selector(vmx, seg);
3118 return;
a9179499 3119 }
2fb92db1
AK		 3120 var->base = vmx_read_guest_seg_base(vmx, seg);
3121 var->limit = vmx_read_guest_seg_limit(vmx, seg);
3122 var->selector = vmx_read_guest_seg_selector(vmx, seg);
3123 ar = vmx_read_guest_seg_ar(vmx, seg);
9fd4a3b7 3124 if ((ar & AR_UNUSABLE_MASK) && !emulate_invalid_guest_state)
6aa8b732
AK		 3125 ar = 0;
3126 var->type = ar & 15;
3127 var->s = (ar >> 4) & 1;
3128 var->dpl = (ar >> 5) & 3;
3129 var->present = (ar >> 7) & 1;
3130 var->avl = (ar >> 12) & 1;
3131 var->l = (ar >> 13) & 1;
3132 var->db = (ar >> 14) & 1;
3133 var->g = (ar >> 15) & 1;
3134 var->unusable = (ar >> 16) & 1;
3135}
3136
a9179499
AK		 3137static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
3138{
a9179499
AK		 3139 struct kvm_segment s;
3140
3141 if (to_vmx(vcpu)->rmode.vm86_active) {
3142 vmx_get_segment(vcpu, &s, seg);
3143 return s.base;
3144 }
2fb92db1 3145 return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
a9179499
AK		 3146}
3147
69c73028 3148static int __vmx_get_cpl(struct kvm_vcpu *vcpu)
2e4d2653 3149{
3eeb3288 3150 if (!is_protmode(vcpu))
2e4d2653
IE		 3151 return 0;
3152
f4c63e5d
AK		 3153 if (!is_long_mode(vcpu)
3154 && (kvm_get_rflags(vcpu) & X86_EFLAGS_VM)) /* if virtual 8086 */
2e4d2653
IE		 3155 return 3;
3156
2fb92db1 3157 return vmx_read_guest_seg_selector(to_vmx(vcpu), VCPU_SREG_CS) & 3;
2e4d2653
IE		 3158}
3159
69c73028
AK		 3160static int vmx_get_cpl(struct kvm_vcpu *vcpu)
3161{
d881e6f6
AK		 3162 struct vcpu_vmx *vmx = to_vmx(vcpu);
3163
3164 /*
3165 * If we enter real mode with cs.sel & 3 != 0, the normal CPL calculations
3166 * fail; use the cache instead.
3167 */
3168 if (unlikely(vmx->emulation_required && emulate_invalid_guest_state)) {
3169 return vmx->cpl;
3170 }
3171
69c73028
AK		 3172 if (!test_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail)) {
3173 __set_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail);
d881e6f6 3174 vmx->cpl = __vmx_get_cpl(vcpu);
69c73028 3175 }
d881e6f6
AK		 3176
3177 return vmx->cpl;
69c73028
AK		 3178}
3179
3180
653e3108 3181static u32 vmx_segment_access_rights(struct kvm_segment *var)
6aa8b732 3182{
6aa8b732
AK		 3183 u32 ar;
3184
f0495f9b 3185 if (var->unusable || !var->present)
6aa8b732
AK		 3186 ar = 1 << 16;
3187 else {
3188 ar = var->type & 15;
3189 ar |= (var->s & 1) << 4;
3190 ar |= (var->dpl & 3) << 5;
3191 ar |= (var->present & 1) << 7;
3192 ar |= (var->avl & 1) << 12;
3193 ar |= (var->l & 1) << 13;
3194 ar |= (var->db & 1) << 14;
3195 ar |= (var->g & 1) << 15;
3196 }
653e3108
AK		 3197
3198 return ar;
3199}
3200
3201static void vmx_set_segment(struct kvm_vcpu *vcpu,
3202 struct kvm_segment *var, int seg)
3203{
7ffd92c5 3204 struct vcpu_vmx *vmx = to_vmx(vcpu);
772e0318 3205 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
653e3108
AK		 3206 u32 ar;
3207
2fb92db1
AK		 3208 vmx_segment_cache_clear(vmx);
3209
7ffd92c5 3210 if (vmx->rmode.vm86_active && seg == VCPU_SREG_TR) {
a8ba6c26 3211 vmcs_write16(sf->selector, var->selector);
f5f7b2fe 3212 vmx->rmode.segs[VCPU_SREG_TR] = *var;
653e3108
AK		 3213 return;
3214 }
3215 vmcs_writel(sf->base, var->base);
3216 vmcs_write32(sf->limit, var->limit);
3217 vmcs_write16(sf->selector, var->selector);
7ffd92c5 3218 if (vmx->rmode.vm86_active && var->s) {
ce566803 3219 vmx->rmode.segs[seg] = *var;
653e3108
AK		 3220 /*
3221 * Hack real-mode segments into vm86 compatibility.
3222 */
3223 if (var->base == 0xffff0000 && var->selector == 0xf000)
3224 vmcs_writel(sf->base, 0xf0000);
3225 ar = 0xf3;
3226 } else
3227 ar = vmx_segment_access_rights(var);
3a624e29
NK		 3228
3229 /*
3230 * Fix the "Accessed" bit in AR field of segment registers for older
3231 * qemu binaries.
3232 * IA32 arch specifies that at the time of processor reset the
3233 * "Accessed" bit in the AR field of segment registers is 1. And qemu
0fa06071 3234 * is setting it to 0 in the userland code. This causes invalid guest
3a624e29
NK		 3235 * state vmexit when "unrestricted guest" mode is turned on.
3236 * Fix for this setup issue in cpu_reset is being pushed in the qemu
3237 * tree. Newer qemu binaries with that qemu fix would not need this
3238 * kvm hack.
3239 */
3240 if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
3241 ar |= 0x1; /* Accessed */
3242
6aa8b732 3243 vmcs_write32(sf->ar_bytes, ar);
69c73028 3244 __clear_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail);
b246dd5d
OW		 3245
3246 /*
3247 * Fix segments for real mode guest in hosts that don't have
3248 * "unrestricted_mode" or it was disabled.
3249 * This is done to allow migration of the guests from hosts with
3250 * unrestricted guest like Westmere to older host that don't have
3251 * unrestricted guest like Nehelem.
3252 */
3253 if (!enable_unrestricted_guest && vmx->rmode.vm86_active) {
3254 switch (seg) {
3255 case VCPU_SREG_CS:
3256 vmcs_write32(GUEST_CS_AR_BYTES, 0xf3);
3257 vmcs_write32(GUEST_CS_LIMIT, 0xffff);
3258 if (vmcs_readl(GUEST_CS_BASE) == 0xffff0000)
3259 vmcs_writel(GUEST_CS_BASE, 0xf0000);
3260 vmcs_write16(GUEST_CS_SELECTOR,
3261 vmcs_readl(GUEST_CS_BASE) >> 4);
3262 break;
3263 case VCPU_SREG_ES:
b246dd5d 3264 case VCPU_SREG_DS:
b246dd5d 3265 case VCPU_SREG_GS:
b246dd5d 3266 case VCPU_SREG_FS:
f5f7b2fe 3267 fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
b246dd5d
OW		 3268 break;
3269 case VCPU_SREG_SS:
3270 vmcs_write16(GUEST_SS_SELECTOR,
3271 vmcs_readl(GUEST_SS_BASE) >> 4);
3272 vmcs_write32(GUEST_SS_LIMIT, 0xffff);
3273 vmcs_write32(GUEST_SS_AR_BYTES, 0xf3);
3274 break;
3275 }
3276 }
6aa8b732
AK		 3277}
3278
6aa8b732
AK		 3279static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3280{
2fb92db1 3281 u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
6aa8b732
AK		 3282
3283 *db = (ar >> 14) & 1;
3284 *l = (ar >> 13) & 1;
3285}
3286
89a27f4d 3287static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
6aa8b732 3288{
89a27f4d
GN		 3289 dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
3290 dt->address = vmcs_readl(GUEST_IDTR_BASE);
6aa8b732
AK		 3291}
3292
89a27f4d 3293static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
6aa8b732 3294{
89a27f4d
GN		 3295 vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
3296 vmcs_writel(GUEST_IDTR_BASE, dt->address);
6aa8b732
AK		 3297}
3298
89a27f4d 3299static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
6aa8b732 3300{
89a27f4d
GN		 3301 dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
3302 dt->address = vmcs_readl(GUEST_GDTR_BASE);
6aa8b732
AK		 3303}
3304
89a27f4d 3305static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
6aa8b732 3306{
89a27f4d
GN		 3307 vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
3308 vmcs_writel(GUEST_GDTR_BASE, dt->address);
6aa8b732
AK		 3309}
3310
648dfaa7
MG		 3311static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
3312{
3313 struct kvm_segment var;
3314 u32 ar;
3315
3316 vmx_get_segment(vcpu, &var, seg);
3317 ar = vmx_segment_access_rights(&var);
3318
3319 if (var.base != (var.selector << 4))
3320 return false;
e2a610d7 3321 if (var.limit < 0xffff)
648dfaa7 3322 return false;
a81aba14 3323 if (((ar | (3 << AR_DPL_SHIFT)) & ~(AR_G_MASK | AR_DB_MASK)) != 0xf3)
648dfaa7
MG		 3324 return false;
3325
3326 return true;
3327}
3328
3329static bool code_segment_valid(struct kvm_vcpu *vcpu)
3330{
3331 struct kvm_segment cs;
3332 unsigned int cs_rpl;
3333
3334 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3335 cs_rpl = cs.selector & SELECTOR_RPL_MASK;
3336
1872a3f4
AK		 3337 if (cs.unusable)
3338 return false;
648dfaa7
MG		 3339 if (~cs.type & (AR_TYPE_CODE_MASK|AR_TYPE_ACCESSES_MASK))
3340 return false;
3341 if (!cs.s)
3342 return false;
1872a3f4 3343 if (cs.type & AR_TYPE_WRITEABLE_MASK) {
648dfaa7
MG		 3344 if (cs.dpl > cs_rpl)
3345 return false;
1872a3f4 3346 } else {
648dfaa7
MG		 3347 if (cs.dpl != cs_rpl)
3348 return false;
3349 }
3350 if (!cs.present)
3351 return false;
3352
3353 /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
3354 return true;
3355}
3356
3357static bool stack_segment_valid(struct kvm_vcpu *vcpu)
3358{
3359 struct kvm_segment ss;
3360 unsigned int ss_rpl;
3361
3362 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3363 ss_rpl = ss.selector & SELECTOR_RPL_MASK;
3364
1872a3f4
AK		 3365 if (ss.unusable)
3366 return true;
3367 if (ss.type != 3 && ss.type != 7)
648dfaa7
MG		 3368 return false;
3369 if (!ss.s)
3370 return false;
3371 if (ss.dpl != ss_rpl) /* DPL != RPL */
3372 return false;
3373 if (!ss.present)
3374 return false;
3375
3376 return true;
3377}
3378
3379static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
3380{
3381 struct kvm_segment var;
3382 unsigned int rpl;
3383
3384 vmx_get_segment(vcpu, &var, seg);
3385 rpl = var.selector & SELECTOR_RPL_MASK;
3386
1872a3f4
AK		 3387 if (var.unusable)
3388 return true;
648dfaa7
MG		 3389 if (!var.s)
3390 return false;
3391 if (!var.present)
3392 return false;
3393 if (~var.type & (AR_TYPE_CODE_MASK|AR_TYPE_WRITEABLE_MASK)) {
3394 if (var.dpl < rpl) /* DPL < RPL */
3395 return false;
3396 }
3397
3398 /* TODO: Add other members to kvm_segment_field to allow checking for other access
3399 * rights flags
3400 */
3401 return true;
3402}
3403
3404static bool tr_valid(struct kvm_vcpu *vcpu)
3405{
3406 struct kvm_segment tr;
3407
3408 vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
3409
1872a3f4
AK		 3410 if (tr.unusable)
3411 return false;
648dfaa7
MG		 3412 if (tr.selector & SELECTOR_TI_MASK) /* TI = 1 */
3413 return false;
1872a3f4 3414 if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
648dfaa7
MG		 3415 return false;
3416 if (!tr.present)
3417 return false;
3418
3419 return true;
3420}
3421
3422static bool ldtr_valid(struct kvm_vcpu *vcpu)
3423{
3424 struct kvm_segment ldtr;
3425
3426 vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
3427
1872a3f4
AK		 3428 if (ldtr.unusable)
3429 return true;
648dfaa7
MG		 3430 if (ldtr.selector & SELECTOR_TI_MASK) /* TI = 1 */
3431 return false;
3432 if (ldtr.type != 2)
3433 return false;
3434 if (!ldtr.present)
3435 return false;
3436
3437 return true;
3438}
3439
3440static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
3441{
3442 struct kvm_segment cs, ss;
3443
3444 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3445 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3446
3447 return ((cs.selector & SELECTOR_RPL_MASK) ==
3448 (ss.selector & SELECTOR_RPL_MASK));
3449}
3450
3451/*
3452 * Check if guest state is valid. Returns true if valid, false if
3453 * not.
3454 * We assume that registers are always usable
3455 */
3456static bool guest_state_valid(struct kvm_vcpu *vcpu)
3457{
3458 /* real mode guest state checks */
3eeb3288 3459 if (!is_protmode(vcpu)) {
648dfaa7
MG		 3460 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
3461 return false;
3462 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
3463 return false;
3464 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
3465 return false;
3466 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
3467 return false;
3468 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
3469 return false;
3470 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
3471 return false;
3472 } else {
3473 /* protected mode guest state checks */
3474 if (!cs_ss_rpl_check(vcpu))
3475 return false;
3476 if (!code_segment_valid(vcpu))
3477 return false;
3478 if (!stack_segment_valid(vcpu))
3479 return false;
3480 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
3481 return false;
3482 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
3483 return false;
3484 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
3485 return false;
3486 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
3487 return false;
3488 if (!tr_valid(vcpu))
3489 return false;
3490 if (!ldtr_valid(vcpu))
3491 return false;
3492 }
3493 /* TODO:
3494 * - Add checks on RIP
3495 * - Add checks on RFLAGS
3496 */
3497
3498 return true;
3499}
3500
d77c26fc 3501static int init_rmode_tss(struct kvm *kvm)
6aa8b732 3502{
40dcaa9f 3503 gfn_t fn;
195aefde 3504 u16 data = 0;
40dcaa9f 3505 int r, idx, ret = 0;
6aa8b732 3506
40dcaa9f
XG		 3507 idx = srcu_read_lock(&kvm->srcu);
3508 fn = rmode_tss_base(kvm) >> PAGE_SHIFT;
195aefde
IE		 3509 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3510 if (r < 0)
10589a46 3511 goto out;
195aefde 3512 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
464d17c8
SY		 3513 r = kvm_write_guest_page(kvm, fn++, &data,
3514 TSS_IOPB_BASE_OFFSET, sizeof(u16));
195aefde 3515 if (r < 0)
10589a46 3516 goto out;
195aefde
IE		 3517 r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
3518 if (r < 0)
10589a46 3519 goto out;
195aefde
IE		 3520 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3521 if (r < 0)
10589a46 3522 goto out;
195aefde 3523 data = ~0;
10589a46
MT		 3524 r = kvm_write_guest_page(kvm, fn, &data,
3525 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
3526 sizeof(u8));
195aefde 3527 if (r < 0)
10589a46
MT		 3528 goto out;
3529
3530 ret = 1;
3531out:
40dcaa9f 3532 srcu_read_unlock(&kvm->srcu, idx);
10589a46 3533 return ret;
6aa8b732
AK		 3534}
3535
b7ebfb05
SY		 3536static int init_rmode_identity_map(struct kvm *kvm)
3537{
40dcaa9f 3538 int i, idx, r, ret;
b7ebfb05
SY		 3539 pfn_t identity_map_pfn;
3540 u32 tmp;
3541
089d034e 3542 if (!enable_ept)
b7ebfb05
SY		 3543 return 1;
3544 if (unlikely(!kvm->arch.ept_identity_pagetable)) {
3545 printk(KERN_ERR "EPT: identity-mapping pagetable "
3546 "haven't been allocated!\n");
3547 return 0;
3548 }
3549 if (likely(kvm->arch.ept_identity_pagetable_done))
3550 return 1;
3551 ret = 0;
b927a3ce 3552 identity_map_pfn = kvm->arch.ept_identity_map_addr >> PAGE_SHIFT;
40dcaa9f 3553 idx = srcu_read_lock(&kvm->srcu);
b7ebfb05
SY		 3554 r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
3555 if (r < 0)
3556 goto out;
3557 /* Set up identity-mapping pagetable for EPT in real mode */
3558 for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
3559 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
3560 _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
3561 r = kvm_write_guest_page(kvm, identity_map_pfn,
3562 &tmp, i * sizeof(tmp), sizeof(tmp));
3563 if (r < 0)
3564 goto out;
3565 }
3566 kvm->arch.ept_identity_pagetable_done = true;
3567 ret = 1;
3568out:
40dcaa9f 3569 srcu_read_unlock(&kvm->srcu, idx);
b7ebfb05
SY		 3570 return ret;
3571}
3572
6aa8b732
AK		 3573static void seg_setup(int seg)
3574{
772e0318 3575 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3a624e29 3576 unsigned int ar;
6aa8b732
AK		 3577
3578 vmcs_write16(sf->selector, 0);
3579 vmcs_writel(sf->base, 0);
3580 vmcs_write32(sf->limit, 0xffff);
3a624e29
NK		 3581 if (enable_unrestricted_guest) {
3582 ar = 0x93;
3583 if (seg == VCPU_SREG_CS)
3584 ar |= 0x08; /* code segment */
3585 } else
3586 ar = 0xf3;
3587
3588 vmcs_write32(sf->ar_bytes, ar);
6aa8b732
AK		 3589}
3590
f78e0e2e
SY		 3591static int alloc_apic_access_page(struct kvm *kvm)
3592{
3593 struct kvm_userspace_memory_region kvm_userspace_mem;
3594 int r = 0;
3595
79fac95e 3596 mutex_lock(&kvm->slots_lock);
bfc6d222 3597 if (kvm->arch.apic_access_page)
f78e0e2e
SY		 3598 goto out;
3599 kvm_userspace_mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
3600 kvm_userspace_mem.flags = 0;
3601 kvm_userspace_mem.guest_phys_addr = 0xfee00000ULL;
3602 kvm_userspace_mem.memory_size = PAGE_SIZE;
3603 r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
3604 if (r)
3605 goto out;
72dc67a6 3606
bfc6d222 3607 kvm->arch.apic_access_page = gfn_to_page(kvm, 0xfee00);
f78e0e2e 3608out:
79fac95e 3609 mutex_unlock(&kvm->slots_lock);
f78e0e2e
SY		 3610 return r;
3611}
3612
b7ebfb05
SY		 3613static int alloc_identity_pagetable(struct kvm *kvm)
3614{
3615 struct kvm_userspace_memory_region kvm_userspace_mem;
3616 int r = 0;
3617
79fac95e 3618 mutex_lock(&kvm->slots_lock);
b7ebfb05
SY		 3619 if (kvm->arch.ept_identity_pagetable)
3620 goto out;
3621 kvm_userspace_mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
3622 kvm_userspace_mem.flags = 0;
b927a3ce
SY		 3623 kvm_userspace_mem.guest_phys_addr =
3624 kvm->arch.ept_identity_map_addr;
b7ebfb05
SY		 3625 kvm_userspace_mem.memory_size = PAGE_SIZE;
3626 r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
3627 if (r)
3628 goto out;
3629
b7ebfb05 3630 kvm->arch.ept_identity_pagetable = gfn_to_page(kvm,
b927a3ce 3631 kvm->arch.ept_identity_map_addr >> PAGE_SHIFT);
b7ebfb05 3632out:
79fac95e 3633 mutex_unlock(&kvm->slots_lock);
b7ebfb05
SY		 3634 return r;
3635}
3636
2384d2b3
SY		 3637static void allocate_vpid(struct vcpu_vmx *vmx)
3638{
3639 int vpid;
3640
3641 vmx->vpid = 0;
919818ab 3642 if (!enable_vpid)
2384d2b3
SY		 3643 return;
3644 spin_lock(&vmx_vpid_lock);
3645 vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
3646 if (vpid < VMX_NR_VPIDS) {
3647 vmx->vpid = vpid;
3648 __set_bit(vpid, vmx_vpid_bitmap);
3649 }
3650 spin_unlock(&vmx_vpid_lock);
3651}
3652
cdbecfc3
LJ		 3653static void free_vpid(struct vcpu_vmx *vmx)
3654{
3655 if (!enable_vpid)
3656 return;
3657 spin_lock(&vmx_vpid_lock);
3658 if (vmx->vpid != 0)
3659 __clear_bit(vmx->vpid, vmx_vpid_bitmap);
3660 spin_unlock(&vmx_vpid_lock);
3661}
3662
5897297b 3663static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap, u32 msr)
25c5f225 3664{
3e7c73e9 3665 int f = sizeof(unsigned long);
25c5f225
SY		 3666
3667 if (!cpu_has_vmx_msr_bitmap())
3668 return;
3669
3670 /*
3671 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3672 * have the write-low and read-high bitmap offsets the wrong way round.
3673 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3674 */
25c5f225 3675 if (msr <= 0x1fff) {
3e7c73e9
AK		 3676 __clear_bit(msr, msr_bitmap + 0x000 / f); /* read-low */
3677 __clear_bit(msr, msr_bitmap + 0x800 / f); /* write-low */
25c5f225
SY		 3678 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
3679 msr &= 0x1fff;
3e7c73e9
AK		 3680 __clear_bit(msr, msr_bitmap + 0x400 / f); /* read-high */
3681 __clear_bit(msr, msr_bitmap + 0xc00 / f); /* write-high */
25c5f225 3682 }
25c5f225
SY		 3683}
3684
5897297b
AK		 3685static void vmx_disable_intercept_for_msr(u32 msr, bool longmode_only)
3686{
3687 if (!longmode_only)
3688 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy, msr);
3689 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode, msr);
3690}
3691
a3a8ff8e
NHE		 3692/*
3693 * Set up the vmcs's constant host-state fields, i.e., host-state fields that
3694 * will not change in the lifetime of the guest.
3695 * Note that host-state that does change is set elsewhere. E.g., host-state
3696 * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
3697 */
3698static void vmx_set_constant_host_state(void)
3699{
3700 u32 low32, high32;
3701 unsigned long tmpl;
3702 struct desc_ptr dt;
3703
3704 vmcs_writel(HOST_CR0, read_cr0() | X86_CR0_TS); /* 22.2.3 */
3705 vmcs_writel(HOST_CR4, read_cr4()); /* 22.2.3, 22.2.5 */
3706 vmcs_writel(HOST_CR3, read_cr3()); /* 22.2.3 FIXME: shadow tables */
3707
3708 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
b2da15ac
AK		 3709#ifdef CONFIG_X86_64
3710 /*
3711 * Load null selectors, so we can avoid reloading them in
3712 * __vmx_load_host_state(), in case userspace uses the null selectors
3713 * too (the expected case).
3714 */
3715 vmcs_write16(HOST_DS_SELECTOR, 0);
3716 vmcs_write16(HOST_ES_SELECTOR, 0);
3717#else
a3a8ff8e
NHE		 3718 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
3719 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
b2da15ac 3720#endif
a3a8ff8e
NHE		 3721 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
3722 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
3723
3724 native_store_idt(&dt);
3725 vmcs_writel(HOST_IDTR_BASE, dt.address); /* 22.2.4 */
3726
3727 asm("mov $.Lkvm_vmx_return, %0" : "=r"(tmpl));
3728 vmcs_writel(HOST_RIP, tmpl); /* 22.2.5 */
3729
3730 rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
3731 vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
3732 rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
3733 vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl); /* 22.2.3 */
3734
3735 if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
3736 rdmsr(MSR_IA32_CR_PAT, low32, high32);
3737 vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
3738 }
3739}
3740
bf8179a0
NHE		 3741static void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
3742{
3743 vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
3744 if (enable_ept)
3745 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
fe3ef05c
NHE		 3746 if (is_guest_mode(&vmx->vcpu))
3747 vmx->vcpu.arch.cr4_guest_owned_bits &=
3748 ~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
bf8179a0
NHE		 3749 vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
3750}
3751
3752static u32 vmx_exec_control(struct vcpu_vmx *vmx)
3753{
3754 u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
3755 if (!vm_need_tpr_shadow(vmx->vcpu.kvm)) {
3756 exec_control &= ~CPU_BASED_TPR_SHADOW;
3757#ifdef CONFIG_X86_64
3758 exec_control |= CPU_BASED_CR8_STORE_EXITING |
3759 CPU_BASED_CR8_LOAD_EXITING;
3760#endif
3761 }
3762 if (!enable_ept)
3763 exec_control |= CPU_BASED_CR3_STORE_EXITING |
3764 CPU_BASED_CR3_LOAD_EXITING |
3765 CPU_BASED_INVLPG_EXITING;
3766 return exec_control;
3767}
3768
3769static u32 vmx_secondary_exec_control(struct vcpu_vmx *vmx)
3770{
3771 u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
3772 if (!vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
3773 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
3774 if (vmx->vpid == 0)
3775 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
3776 if (!enable_ept) {
3777 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
3778 enable_unrestricted_guest = 0;
ad756a16
MJ		 3779 /* Enable INVPCID for non-ept guests may cause performance regression. */
3780 exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
bf8179a0
NHE		 3781 }
3782 if (!enable_unrestricted_guest)
3783 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
3784 if (!ple_gap)
3785 exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
3786 return exec_control;
3787}
3788
ce88decf
XG		 3789static void ept_set_mmio_spte_mask(void)
3790{
3791 /*
3792 * EPT Misconfigurations can be generated if the value of bits 2:0
3793 * of an EPT paging-structure entry is 110b (write/execute).
3794 * Also, magic bits (0xffull << 49) is set to quickly identify mmio
3795 * spte.
3796 */
3797 kvm_mmu_set_mmio_spte_mask(0xffull << 49 | 0x6ull);
3798}
3799
6aa8b732
AK		 3800/*
3801 * Sets up the vmcs for emulated real mode.
3802 */
8b9cf98c 3803static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
6aa8b732 3804{
2e4ce7f5 3805#ifdef CONFIG_X86_64
6aa8b732 3806 unsigned long a;
2e4ce7f5 3807#endif
6aa8b732 3808 int i;
6aa8b732 3809
6aa8b732 3810 /* I/O */
3e7c73e9
AK		 3811 vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a));
3812 vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b));
6aa8b732 3813
25c5f225 3814 if (cpu_has_vmx_msr_bitmap())
5897297b 3815 vmcs_write64(MSR_BITMAP, __pa(vmx_msr_bitmap_legacy));
25c5f225 3816
6aa8b732
AK		 3817 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
3818
6aa8b732 3819 /* Control */
1c3d14fe
YS		 3820 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL,
3821 vmcs_config.pin_based_exec_ctrl);
6e5d865c 3822
bf8179a0 3823 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, vmx_exec_control(vmx));
6aa8b732 3824
83ff3b9d 3825 if (cpu_has_secondary_exec_ctrls()) {
bf8179a0
NHE		 3826 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
3827 vmx_secondary_exec_control(vmx));
83ff3b9d 3828 }
f78e0e2e 3829
4b8d54f9
ZE		 3830 if (ple_gap) {
3831 vmcs_write32(PLE_GAP, ple_gap);
3832 vmcs_write32(PLE_WINDOW, ple_window);
3833 }
3834
c3707958
XG		 3835 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
3836 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
6aa8b732
AK		 3837 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
3838
9581d442
AK		 3839 vmcs_write16(HOST_FS_SELECTOR, 0); /* 22.2.4 */
3840 vmcs_write16(HOST_GS_SELECTOR, 0); /* 22.2.4 */
a3a8ff8e 3841 vmx_set_constant_host_state();
05b3e0c2 3842#ifdef CONFIG_X86_64
6aa8b732
AK		 3843 rdmsrl(MSR_FS_BASE, a);
3844 vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
3845 rdmsrl(MSR_GS_BASE, a);
3846 vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
3847#else
3848 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
3849 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
3850#endif
3851
2cc51560
ED		 3852 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
3853 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
61d2ef2c 3854 vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
2cc51560 3855 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
61d2ef2c 3856 vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
6aa8b732 3857
468d472f 3858 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
a3a8ff8e
NHE		 3859 u32 msr_low, msr_high;
3860 u64 host_pat;
468d472f
SY		 3861 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
3862 host_pat = msr_low | ((u64) msr_high << 32);
3863 /* Write the default value follow host pat */
3864 vmcs_write64(GUEST_IA32_PAT, host_pat);
3865 /* Keep arch.pat sync with GUEST_IA32_PAT */
3866 vmx->vcpu.arch.pat = host_pat;
3867 }
3868
6aa8b732
AK		 3869 for (i = 0; i < NR_VMX_MSR; ++i) {
3870 u32 index = vmx_msr_index[i];
3871 u32 data_low, data_high;
a2fa3e9f 3872 int j = vmx->nmsrs;
6aa8b732
AK		 3873
3874 if (rdmsr_safe(index, &data_low, &data_high) < 0)
3875 continue;
432bd6cb
AK		 3876 if (wrmsr_safe(index, data_low, data_high) < 0)
3877 continue;
26bb0981
AK		 3878 vmx->guest_msrs[j].index = i;
3879 vmx->guest_msrs[j].data = 0;
d5696725 3880 vmx->guest_msrs[j].mask = -1ull;
a2fa3e9f 3881 ++vmx->nmsrs;
6aa8b732 3882 }
6aa8b732 3883
1c3d14fe 3884 vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
6aa8b732
AK		 3885
3886 /* 22.2.1, 20.8.1 */
1c3d14fe
YS		 3887 vmcs_write32(VM_ENTRY_CONTROLS, vmcs_config.vmentry_ctrl);
3888
e00c8cf2 3889 vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
bf8179a0 3890 set_cr4_guest_host_mask(vmx);
e00c8cf2 3891
99e3e30a 3892 kvm_write_tsc(&vmx->vcpu, 0);
f78e0e2e 3893
e00c8cf2
AK		 3894 return 0;
3895}
3896
3897static int vmx_vcpu_reset(struct kvm_vcpu *vcpu)
3898{
3899 struct vcpu_vmx *vmx = to_vmx(vcpu);
3900 u64 msr;
4b9d3a04 3901 int ret;
e00c8cf2 3902
5fdbf976 3903 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP));
e00c8cf2 3904
7ffd92c5 3905 vmx->rmode.vm86_active = 0;
e00c8cf2 3906
3b86cd99
JK		 3907 vmx->soft_vnmi_blocked = 0;
3908
ad312c7c 3909 vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
2d3ad1f4 3910 kvm_set_cr8(&vmx->vcpu, 0);
e00c8cf2 3911 msr = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
c5af89b6 3912 if (kvm_vcpu_is_bsp(&vmx->vcpu))
e00c8cf2
AK		 3913 msr |= MSR_IA32_APICBASE_BSP;
3914 kvm_set_apic_base(&vmx->vcpu, msr);
3915
10ab25cd
JK		 3916 ret = fx_init(&vmx->vcpu);
3917 if (ret != 0)
3918 goto out;
e00c8cf2 3919
2fb92db1
AK		 3920 vmx_segment_cache_clear(vmx);
3921
5706be0d 3922 seg_setup(VCPU_SREG_CS);
e00c8cf2
AK		 3923 /*
3924 * GUEST_CS_BASE should really be 0xffff0000, but VT vm86 mode
3925 * insists on having GUEST_CS_BASE == GUEST_CS_SELECTOR << 4. Sigh.
3926 */
c5af89b6 3927 if (kvm_vcpu_is_bsp(&vmx->vcpu)) {
e00c8cf2
AK		 3928 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
3929 vmcs_writel(GUEST_CS_BASE, 0x000f0000);
3930 } else {
ad312c7c
ZX		 3931 vmcs_write16(GUEST_CS_SELECTOR, vmx->vcpu.arch.sipi_vector << 8);
3932 vmcs_writel(GUEST_CS_BASE, vmx->vcpu.arch.sipi_vector << 12);
e00c8cf2 3933 }
e00c8cf2
AK		 3934
3935 seg_setup(VCPU_SREG_DS);
3936 seg_setup(VCPU_SREG_ES);
3937 seg_setup(VCPU_SREG_FS);
3938 seg_setup(VCPU_SREG_GS);
3939 seg_setup(VCPU_SREG_SS);
3940
3941 vmcs_write16(GUEST_TR_SELECTOR, 0);
3942 vmcs_writel(GUEST_TR_BASE, 0);
3943 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
3944 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
3945
3946 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
3947 vmcs_writel(GUEST_LDTR_BASE, 0);
3948 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
3949 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
3950
3951 vmcs_write32(GUEST_SYSENTER_CS, 0);
3952 vmcs_writel(GUEST_SYSENTER_ESP, 0);
3953 vmcs_writel(GUEST_SYSENTER_EIP, 0);
3954
3955 vmcs_writel(GUEST_RFLAGS, 0x02);
c5af89b6 3956 if (kvm_vcpu_is_bsp(&vmx->vcpu))
5fdbf976 3957 kvm_rip_write(vcpu, 0xfff0);
e00c8cf2 3958 else
5fdbf976
MT		 3959 kvm_rip_write(vcpu, 0);
3960 kvm_register_write(vcpu, VCPU_REGS_RSP, 0);
e00c8cf2 3961
e00c8cf2
AK		 3962 vmcs_writel(GUEST_DR7, 0x400);
3963
3964 vmcs_writel(GUEST_GDTR_BASE, 0);
3965 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
3966
3967 vmcs_writel(GUEST_IDTR_BASE, 0);
3968 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
3969
443381a8 3970 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
e00c8cf2
AK		 3971 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
3972 vmcs_write32(GUEST_PENDING_DBG_EXCEPTIONS, 0);
3973
e00c8cf2
AK		 3974 /* Special registers */
3975 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
3976
3977 setup_msrs(vmx);
3978
6aa8b732
AK		 3979 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
3980
f78e0e2e
SY		 3981 if (cpu_has_vmx_tpr_shadow()) {
3982 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
3983 if (vm_need_tpr_shadow(vmx->vcpu.kvm))
3984 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
afc20184 3985 __pa(vmx->vcpu.arch.apic->regs));
f78e0e2e
SY		 3986 vmcs_write32(TPR_THRESHOLD, 0);
3987 }
3988
3989 if (vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
3990 vmcs_write64(APIC_ACCESS_ADDR,
bfc6d222 3991 page_to_phys(vmx->vcpu.kvm->arch.apic_access_page));
6aa8b732 3992
2384d2b3
SY		 3993 if (vmx->vpid != 0)
3994 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
3995
fa40052c 3996 vmx->vcpu.arch.cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
7a4f5ad0 3997 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
4d4ec087 3998 vmx_set_cr0(&vmx->vcpu, kvm_read_cr0(vcpu)); /* enter rmode */
7a4f5ad0 3999 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8b9cf98c 4000 vmx_set_cr4(&vmx->vcpu, 0);
8b9cf98c 4001 vmx_set_efer(&vmx->vcpu, 0);
8b9cf98c
RR		 4002 vmx_fpu_activate(&vmx->vcpu);
4003 update_exception_bitmap(&vmx->vcpu);
6aa8b732 4004
b9d762fa 4005 vpid_sync_context(vmx);
2384d2b3 4006
3200f405 4007 ret = 0;
6aa8b732 4008
a89a8fb9
MG		 4009 /* HACK: Don't enable emulation on guest boot/reset */
4010 vmx->emulation_required = 0;
4011
6aa8b732
AK		 4012out:
4013 return ret;
4014}
4015
b6f1250e
NHE		 4016/*
4017 * In nested virtualization, check if L1 asked to exit on external interrupts.
4018 * For most existing hypervisors, this will always return true.
4019 */
4020static bool nested_exit_on_intr(struct kvm_vcpu *vcpu)
4021{
4022 return get_vmcs12(vcpu)->pin_based_vm_exec_control &
4023 PIN_BASED_EXT_INTR_MASK;
4024}
4025
3b86cd99
JK		 4026static void enable_irq_window(struct kvm_vcpu *vcpu)
4027{
4028 u32 cpu_based_vm_exec_control;
d6185f20
NHE		 4029 if (is_guest_mode(vcpu) && nested_exit_on_intr(vcpu)) {
4030 /*
4031 * We get here if vmx_interrupt_allowed() said we can't
4032 * inject to L1 now because L2 must run. Ask L2 to exit
4033 * right after entry, so we can inject to L1 more promptly.
b6f1250e 4034 */
d6185f20 4035 kvm_make_request(KVM_REQ_IMMEDIATE_EXIT, vcpu);
b6f1250e 4036 return;
d6185f20 4037 }
3b86cd99
JK		 4038
4039 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4040 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
4041 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
4042}
4043
4044static void enable_nmi_window(struct kvm_vcpu *vcpu)
4045{
4046 u32 cpu_based_vm_exec_control;
4047
4048 if (!cpu_has_virtual_nmis()) {
4049 enable_irq_window(vcpu);
4050 return;
4051 }
4052
30bd0c4c
AK		 4053 if (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
4054 enable_irq_window(vcpu);
4055 return;
4056 }
3b86cd99
JK		 4057 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4058 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_NMI_PENDING;
4059 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
4060}
4061
66fd3f7f 4062static void vmx_inject_irq(struct kvm_vcpu *vcpu)
85f455f7 4063{
9c8cba37 4064 struct vcpu_vmx *vmx = to_vmx(vcpu);
66fd3f7f
GN		 4065 uint32_t intr;
4066 int irq = vcpu->arch.interrupt.nr;
9c8cba37 4067
229456fc 4068 trace_kvm_inj_virq(irq);
2714d1d3 4069
fa89a817 4070 ++vcpu->stat.irq_injections;
7ffd92c5 4071 if (vmx->rmode.vm86_active) {
71f9833b
SH		 4072 int inc_eip = 0;
4073 if (vcpu->arch.interrupt.soft)
4074 inc_eip = vcpu->arch.event_exit_inst_len;
4075 if (kvm_inject_realmode_interrupt(vcpu, irq, inc_eip) != EMULATE_DONE)
a92601bb 4076 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
85f455f7
ED		 4077 return;
4078 }
66fd3f7f
GN		 4079 intr = irq | INTR_INFO_VALID_MASK;
4080 if (vcpu->arch.interrupt.soft) {
4081 intr |= INTR_TYPE_SOFT_INTR;
4082 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
4083 vmx->vcpu.arch.event_exit_inst_len);
4084 } else
4085 intr |= INTR_TYPE_EXT_INTR;
4086 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
85f455f7
ED		 4087}
4088
f08864b4
SY		 4089static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
4090{
66a5a347
JK		 4091 struct vcpu_vmx *vmx = to_vmx(vcpu);
4092
0b6ac343
NHE		 4093 if (is_guest_mode(vcpu))
4094 return;
4095
3b86cd99
JK		 4096 if (!cpu_has_virtual_nmis()) {
4097 /*
4098 * Tracking the NMI-blocked state in software is built upon
4099 * finding the next open IRQ window. This, in turn, depends on
4100 * well-behaving guests: They have to keep IRQs disabled at
4101 * least as long as the NMI handler runs. Otherwise we may
4102 * cause NMI nesting, maybe breaking the guest. But as this is
4103 * highly unlikely, we can live with the residual risk.
4104 */
4105 vmx->soft_vnmi_blocked = 1;
4106 vmx->vnmi_blocked_time = 0;
4107 }
4108
487b391d 4109 ++vcpu->stat.nmi_injections;
9d58b931 4110 vmx->nmi_known_unmasked = false;
7ffd92c5 4111 if (vmx->rmode.vm86_active) {
71f9833b 4112 if (kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0) != EMULATE_DONE)
a92601bb 4113 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
66a5a347
JK		 4114 return;
4115 }
f08864b4
SY		 4116 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
4117 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
f08864b4
SY		 4118}
4119
c4282df9 4120static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
33f089ca 4121{
3b86cd99 4122 if (!cpu_has_virtual_nmis() && to_vmx(vcpu)->soft_vnmi_blocked)
c4282df9 4123 return 0;
33f089ca 4124
c4282df9 4125 return !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
30bd0c4c
AK		 4126 (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
4127 | GUEST_INTR_STATE_NMI));
33f089ca
JK		 4128}
4129
3cfc3092
JK		 4130static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
4131{
4132 if (!cpu_has_virtual_nmis())
4133 return to_vmx(vcpu)->soft_vnmi_blocked;
9d58b931
AK		 4134 if (to_vmx(vcpu)->nmi_known_unmasked)
4135 return false;
c332c83a 4136 return vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
3cfc3092
JK		 4137}
4138
4139static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
4140{
4141 struct vcpu_vmx *vmx = to_vmx(vcpu);
4142
4143 if (!cpu_has_virtual_nmis()) {
4144 if (vmx->soft_vnmi_blocked != masked) {
4145 vmx->soft_vnmi_blocked = masked;
4146 vmx->vnmi_blocked_time = 0;
4147 }
4148 } else {
9d58b931 4149 vmx->nmi_known_unmasked = !masked;
3cfc3092
JK		 4150 if (masked)
4151 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
4152 GUEST_INTR_STATE_NMI);
4153 else
4154 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
4155 GUEST_INTR_STATE_NMI);
4156 }
4157}
4158
78646121
GN		 4159static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
4160{
b6f1250e 4161 if (is_guest_mode(vcpu) && nested_exit_on_intr(vcpu)) {
51cfe38e
NHE		 4162 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4163 if (to_vmx(vcpu)->nested.nested_run_pending ||
4164 (vmcs12->idt_vectoring_info_field &
4165 VECTORING_INFO_VALID_MASK))
b6f1250e
NHE		 4166 return 0;
4167 nested_vmx_vmexit(vcpu);
b6f1250e
NHE		 4168 vmcs12->vm_exit_reason = EXIT_REASON_EXTERNAL_INTERRUPT;
4169 vmcs12->vm_exit_intr_info = 0;
4170 /* fall through to normal code, but now in L1, not L2 */
4171 }
4172
c4282df9
GN		 4173 return (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
4174 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4175 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
78646121
GN		 4176}
4177
cbc94022
IE		 4178static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
4179{
4180 int ret;
4181 struct kvm_userspace_memory_region tss_mem = {
6fe63979 4182 .slot = TSS_PRIVATE_MEMSLOT,
cbc94022
IE		 4183 .guest_phys_addr = addr,
4184 .memory_size = PAGE_SIZE * 3,
4185 .flags = 0,
4186 };
4187
4188 ret = kvm_set_memory_region(kvm, &tss_mem, 0);
4189 if (ret)
4190 return ret;
bfc6d222 4191 kvm->arch.tss_addr = addr;
93ea5388
GN		 4192 if (!init_rmode_tss(kvm))
4193 return -ENOMEM;
4194
cbc94022
IE		 4195 return 0;
4196}
4197
6aa8b732
AK		 4198static int handle_rmode_exception(struct kvm_vcpu *vcpu,
4199 int vec, u32 err_code)
4200{
b3f37707
NK		 4201 /*
4202 * Instruction with address size override prefix opcode 0x67
4203 * Cause the #SS fault with 0 error code in VM86 mode.
4204 */
4205 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0)
51d8b661 4206 if (emulate_instruction(vcpu, 0) == EMULATE_DONE)
6aa8b732 4207 return 1;
77ab6db0
JK		 4208 /*
4209 * Forward all other exceptions that are valid in real mode.
4210 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
4211 * the required debugging infrastructure rework.
4212 */
4213 switch (vec) {
77ab6db0 4214 case DB_VECTOR:
d0bfb940
JK		 4215 if (vcpu->guest_debug &
4216 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
4217 return 0;
4218 kvm_queue_exception(vcpu, vec);
4219 return 1;
77ab6db0 4220 case BP_VECTOR:
c573cd22
JK		 4221 /*
4222 * Update instruction length as we may reinject the exception
4223 * from user space while in guest debugging mode.
4224 */
4225 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
4226 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
d0bfb940
JK		 4227 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
4228 return 0;
4229 /* fall through */
4230 case DE_VECTOR:
77ab6db0
JK		 4231 case OF_VECTOR:
4232 case BR_VECTOR:
4233 case UD_VECTOR:
4234 case DF_VECTOR:
4235 case SS_VECTOR:
4236 case GP_VECTOR:
4237 case MF_VECTOR:
4238 kvm_queue_exception(vcpu, vec);
4239 return 1;
4240 }
6aa8b732
AK		 4241 return 0;
4242}
4243
a0861c02
AK		 4244/*
4245 * Trigger machine check on the host. We assume all the MSRs are already set up
4246 * by the CPU and that we still run on the same CPU as the MCE occurred on.
4247 * We pass a fake environment to the machine check handler because we want
4248 * the guest to be always treated like user space, no matter what context
4249 * it used internally.
4250 */
4251static void kvm_machine_check(void)
4252{
4253#if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
4254 struct pt_regs regs = {
4255 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
4256 .flags = X86_EFLAGS_IF,
4257 };
4258
4259 do_machine_check(&regs, 0);
4260#endif
4261}
4262
851ba692 4263static int handle_machine_check(struct kvm_vcpu *vcpu)
a0861c02
AK		 4264{
4265 /* already handled by vcpu_run */
4266 return 1;
4267}
4268
851ba692 4269static int handle_exception(struct kvm_vcpu *vcpu)
6aa8b732 4270{
1155f76a 4271 struct vcpu_vmx *vmx = to_vmx(vcpu);
851ba692 4272 struct kvm_run *kvm_run = vcpu->run;
d0bfb940 4273 u32 intr_info, ex_no, error_code;
42dbaa5a 4274 unsigned long cr2, rip, dr6;
6aa8b732
AK		 4275 u32 vect_info;
4276 enum emulation_result er;
4277
1155f76a 4278 vect_info = vmx->idt_vectoring_info;
88786475 4279 intr_info = vmx->exit_intr_info;
6aa8b732 4280
a0861c02 4281 if (is_machine_check(intr_info))
851ba692 4282 return handle_machine_check(vcpu);
a0861c02 4283
6aa8b732 4284 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
65ac7264
AK		 4285 !is_page_fault(intr_info)) {
4286 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4287 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
4288 vcpu->run->internal.ndata = 2;
4289 vcpu->run->internal.data[0] = vect_info;
4290 vcpu->run->internal.data[1] = intr_info;
4291 return 0;
4292 }
6aa8b732 4293
e4a41889 4294 if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR)
1b6269db 4295 return 1; /* already handled by vmx_vcpu_run() */
2ab455cc
AL		 4296
4297 if (is_no_device(intr_info)) {
5fd86fcf 4298 vmx_fpu_activate(vcpu);
2ab455cc
AL		 4299 return 1;
4300 }
4301
7aa81cc0 4302 if (is_invalid_opcode(intr_info)) {
51d8b661 4303 er = emulate_instruction(vcpu, EMULTYPE_TRAP_UD);
7aa81cc0 4304 if (er != EMULATE_DONE)
7ee5d940 4305 kvm_queue_exception(vcpu, UD_VECTOR);
7aa81cc0
AL		 4306 return 1;
4307 }
4308
6aa8b732 4309 error_code = 0;
2e11384c 4310 if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
6aa8b732
AK		 4311 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
4312 if (is_page_fault(intr_info)) {
1439442c 4313 /* EPT won't cause page fault directly */
cf3ace79 4314 BUG_ON(enable_ept);
6aa8b732 4315 cr2 = vmcs_readl(EXIT_QUALIFICATION);
229456fc
MT		 4316 trace_kvm_page_fault(cr2, error_code);
4317
3298b75c 4318 if (kvm_event_needs_reinjection(vcpu))
577bdc49 4319 kvm_mmu_unprotect_page_virt(vcpu, cr2);
dc25e89e 4320 return kvm_mmu_page_fault(vcpu, cr2, error_code, NULL, 0);
6aa8b732
AK		 4321 }
4322
7ffd92c5 4323 if (vmx->rmode.vm86_active &&
6aa8b732 4324 handle_rmode_exception(vcpu, intr_info & INTR_INFO_VECTOR_MASK,
72d6e5a0 4325 error_code)) {
ad312c7c
ZX		 4326 if (vcpu->arch.halt_request) {
4327 vcpu->arch.halt_request = 0;
72d6e5a0
AK		 4328 return kvm_emulate_halt(vcpu);
4329 }
6aa8b732 4330 return 1;
72d6e5a0 4331 }
6aa8b732 4332
d0bfb940 4333 ex_no = intr_info & INTR_INFO_VECTOR_MASK;
42dbaa5a
JK		 4334 switch (ex_no) {
4335 case DB_VECTOR:
4336 dr6 = vmcs_readl(EXIT_QUALIFICATION);
4337 if (!(vcpu->guest_debug &
4338 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
4339 vcpu->arch.dr6 = dr6 | DR6_FIXED_1;
4340 kvm_queue_exception(vcpu, DB_VECTOR);
4341 return 1;
4342 }
4343 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
4344 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
4345 /* fall through */
4346 case BP_VECTOR:
c573cd22
JK		 4347 /*
4348 * Update instruction length as we may reinject #BP from
4349 * user space while in guest debugging mode. Reading it for
4350 * #DB as well causes no harm, it is not used in that case.
4351 */
4352 vmx->vcpu.arch.event_exit_inst_len =
4353 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
6aa8b732 4354 kvm_run->exit_reason = KVM_EXIT_DEBUG;
0a434bb2 4355 rip = kvm_rip_read(vcpu);
d0bfb940
JK		 4356 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
4357 kvm_run->debug.arch.exception = ex_no;
42dbaa5a
JK		 4358 break;
4359 default:
d0bfb940
JK		 4360 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
4361 kvm_run->ex.exception = ex_no;
4362 kvm_run->ex.error_code = error_code;
42dbaa5a 4363 break;
6aa8b732 4364 }
6aa8b732
AK		 4365 return 0;
4366}
4367
851ba692 4368static int handle_external_interrupt(struct kvm_vcpu *vcpu)
6aa8b732 4369{
1165f5fe 4370 ++vcpu->stat.irq_exits;
6aa8b732
AK		 4371 return 1;
4372}
4373
851ba692 4374static int handle_triple_fault(struct kvm_vcpu *vcpu)
988ad74f 4375{
851ba692 4376 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
988ad74f
AK		 4377 return 0;
4378}
6aa8b732 4379
851ba692 4380static int handle_io(struct kvm_vcpu *vcpu)
6aa8b732 4381{
bfdaab09 4382 unsigned long exit_qualification;
34c33d16 4383 int size, in, string;
039576c0 4384 unsigned port;
6aa8b732 4385
bfdaab09 4386 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
039576c0 4387 string = (exit_qualification & 16) != 0;
cf8f70bf 4388 in = (exit_qualification & 8) != 0;
e70669ab 4389
cf8f70bf 4390 ++vcpu->stat.io_exits;
e70669ab 4391
cf8f70bf 4392 if (string || in)
51d8b661 4393 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
e70669ab 4394
cf8f70bf
GN		 4395 port = exit_qualification >> 16;
4396 size = (exit_qualification & 7) + 1;
e93f36bc 4397 skip_emulated_instruction(vcpu);
cf8f70bf
GN		 4398
4399 return kvm_fast_pio_out(vcpu, size, port);
6aa8b732
AK		 4400}
4401
102d8325
IM		 4402static void
4403vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
4404{
4405 /*
4406 * Patch in the VMCALL instruction:
4407 */
4408 hypercall[0] = 0x0f;
4409 hypercall[1] = 0x01;
4410 hypercall[2] = 0xc1;
102d8325
IM		 4411}
4412
0fa06071 4413/* called to set cr0 as appropriate for a mov-to-cr0 exit. */
eeadf9e7
NHE		 4414static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
4415{
4416 if (to_vmx(vcpu)->nested.vmxon &&
4417 ((val & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON))
4418 return 1;
4419
4420 if (is_guest_mode(vcpu)) {
4421 /*
4422 * We get here when L2 changed cr0 in a way that did not change
4423 * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
4424 * but did change L0 shadowed bits. This can currently happen
4425 * with the TS bit: L0 may want to leave TS on (for lazy fpu
4426 * loading) while pretending to allow the guest to change it.
4427 */
4428 if (kvm_set_cr0(vcpu, (val & vcpu->arch.cr0_guest_owned_bits) |
4429 (vcpu->arch.cr0 & ~vcpu->arch.cr0_guest_owned_bits)))
4430 return 1;
4431 vmcs_writel(CR0_READ_SHADOW, val);
4432 return 0;
4433 } else
4434 return kvm_set_cr0(vcpu, val);
4435}
4436
4437static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
4438{
4439 if (is_guest_mode(vcpu)) {
4440 if (kvm_set_cr4(vcpu, (val & vcpu->arch.cr4_guest_owned_bits) |
4441 (vcpu->arch.cr4 & ~vcpu->arch.cr4_guest_owned_bits)))
4442 return 1;
4443 vmcs_writel(CR4_READ_SHADOW, val);
4444 return 0;
4445 } else
4446 return kvm_set_cr4(vcpu, val);
4447}
4448
4449/* called to set cr0 as approriate for clts instruction exit. */
4450static void handle_clts(struct kvm_vcpu *vcpu)
4451{
4452 if (is_guest_mode(vcpu)) {
4453 /*
4454 * We get here when L2 did CLTS, and L1 didn't shadow CR0.TS
4455 * but we did (!fpu_active). We need to keep GUEST_CR0.TS on,
4456 * just pretend it's off (also in arch.cr0 for fpu_activate).
4457 */
4458 vmcs_writel(CR0_READ_SHADOW,
4459 vmcs_readl(CR0_READ_SHADOW) & ~X86_CR0_TS);
4460 vcpu->arch.cr0 &= ~X86_CR0_TS;
4461 } else
4462 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
4463}
4464
851ba692 4465static int handle_cr(struct kvm_vcpu *vcpu)
6aa8b732 4466{
229456fc 4467 unsigned long exit_qualification, val;
6aa8b732
AK		 4468 int cr;
4469 int reg;
49a9b07e 4470 int err;
6aa8b732 4471
bfdaab09 4472 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6aa8b732
AK		 4473 cr = exit_qualification & 15;
4474 reg = (exit_qualification >> 8) & 15;
4475 switch ((exit_qualification >> 4) & 3) {
4476 case 0: /* mov to cr */
229456fc
MT		 4477 val = kvm_register_read(vcpu, reg);
4478 trace_kvm_cr_write(cr, val);
6aa8b732
AK		 4479 switch (cr) {
4480 case 0:
eeadf9e7 4481 err = handle_set_cr0(vcpu, val);
db8fcefa 4482 kvm_complete_insn_gp(vcpu, err);
6aa8b732
AK		 4483 return 1;
4484 case 3:
2390218b 4485 err = kvm_set_cr3(vcpu, val);
db8fcefa 4486 kvm_complete_insn_gp(vcpu, err);
6aa8b732
AK		 4487 return 1;
4488 case 4:
eeadf9e7 4489 err = handle_set_cr4(vcpu, val);
db8fcefa 4490 kvm_complete_insn_gp(vcpu, err);
6aa8b732 4491 return 1;
0a5fff19
GN		 4492 case 8: {
4493 u8 cr8_prev = kvm_get_cr8(vcpu);
4494 u8 cr8 = kvm_register_read(vcpu, reg);
eea1cff9 4495 err = kvm_set_cr8(vcpu, cr8);
db8fcefa 4496 kvm_complete_insn_gp(vcpu, err);
0a5fff19
GN		 4497 if (irqchip_in_kernel(vcpu->kvm))
4498 return 1;
4499 if (cr8_prev <= cr8)
4500 return 1;
851ba692 4501 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
0a5fff19
GN		 4502 return 0;
4503 }
6aa8b732
AK		 4504 };
4505 break;
25c4c276 4506 case 2: /* clts */
eeadf9e7 4507 handle_clts(vcpu);
4d4ec087 4508 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
25c4c276 4509 skip_emulated_instruction(vcpu);
6b52d186 4510 vmx_fpu_activate(vcpu);
25c4c276 4511 return 1;
6aa8b732
AK		 4512 case 1: /*mov from cr*/
4513 switch (cr) {
4514 case 3:
9f8fe504
AK		 4515 val = kvm_read_cr3(vcpu);
4516 kvm_register_write(vcpu, reg, val);
4517 trace_kvm_cr_read(cr, val);
6aa8b732
AK		 4518 skip_emulated_instruction(vcpu);
4519 return 1;
4520 case 8:
229456fc
MT		 4521 val = kvm_get_cr8(vcpu);
4522 kvm_register_write(vcpu, reg, val);
4523 trace_kvm_cr_read(cr, val);
6aa8b732
AK		 4524 skip_emulated_instruction(vcpu);
4525 return 1;
4526 }
4527 break;
4528 case 3: /* lmsw */
a1f83a74 4529 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
4d4ec087 4530 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
a1f83a74 4531 kvm_lmsw(vcpu, val);
6aa8b732
AK		 4532
4533 skip_emulated_instruction(vcpu);
4534 return 1;
4535 default:
4536 break;
4537 }
851ba692 4538 vcpu->run->exit_reason = 0;
a737f256 4539 vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
6aa8b732
AK		 4540 (int)(exit_qualification >> 4) & 3, cr);
4541 return 0;
4542}
4543
851ba692 4544static int handle_dr(struct kvm_vcpu *vcpu)
6aa8b732 4545{
bfdaab09 4546 unsigned long exit_qualification;
6aa8b732
AK		 4547 int dr, reg;
4548
f2483415 4549 /* Do not handle if the CPL > 0, will trigger GP on re-entry */
0a79b009
AK		 4550 if (!kvm_require_cpl(vcpu, 0))
4551 return 1;
42dbaa5a
JK		 4552 dr = vmcs_readl(GUEST_DR7);
4553 if (dr & DR7_GD) {
4554 /*
4555 * As the vm-exit takes precedence over the debug trap, we
4556 * need to emulate the latter, either for the host or the
4557 * guest debugging itself.
4558 */
4559 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
851ba692
AK		 4560 vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
4561 vcpu->run->debug.arch.dr7 = dr;
4562 vcpu->run->debug.arch.pc =
42dbaa5a
JK		 4563 vmcs_readl(GUEST_CS_BASE) +
4564 vmcs_readl(GUEST_RIP);
851ba692
AK		 4565 vcpu->run->debug.arch.exception = DB_VECTOR;
4566 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
42dbaa5a
JK		 4567 return 0;
4568 } else {
4569 vcpu->arch.dr7 &= ~DR7_GD;
4570 vcpu->arch.dr6 |= DR6_BD;
4571 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
4572 kvm_queue_exception(vcpu, DB_VECTOR);
4573 return 1;
4574 }
4575 }
4576
bfdaab09 4577 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
42dbaa5a
JK		 4578 dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
4579 reg = DEBUG_REG_ACCESS_REG(exit_qualification);
4580 if (exit_qualification & TYPE_MOV_FROM_DR) {
020df079
GN		 4581 unsigned long val;
4582 if (!kvm_get_dr(vcpu, dr, &val))
4583 kvm_register_write(vcpu, reg, val);
4584 } else
4585 kvm_set_dr(vcpu, dr, vcpu->arch.regs[reg]);
6aa8b732
AK		 4586 skip_emulated_instruction(vcpu);
4587 return 1;
4588}
4589
020df079
GN		 4590static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
4591{
4592 vmcs_writel(GUEST_DR7, val);
4593}
4594
851ba692 4595static int handle_cpuid(struct kvm_vcpu *vcpu)
6aa8b732 4596{
06465c5a
AK		 4597 kvm_emulate_cpuid(vcpu);
4598 return 1;
6aa8b732
AK		 4599}
4600
851ba692 4601static int handle_rdmsr(struct kvm_vcpu *vcpu)
6aa8b732 4602{
ad312c7c 4603 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
6aa8b732
AK		 4604 u64 data;
4605
4606 if (vmx_get_msr(vcpu, ecx, &data)) {
59200273 4607 trace_kvm_msr_read_ex(ecx);
c1a5d4f9 4608 kvm_inject_gp(vcpu, 0);
6aa8b732
AK		 4609 return 1;
4610 }
4611
229456fc 4612 trace_kvm_msr_read(ecx, data);
2714d1d3 4613
6aa8b732 4614 /* FIXME: handling of bits 32:63 of rax, rdx */
ad312c7c
ZX		 4615 vcpu->arch.regs[VCPU_REGS_RAX] = data & -1u;
4616 vcpu->arch.regs[VCPU_REGS_RDX] = (data >> 32) & -1u;
6aa8b732
AK		 4617 skip_emulated_instruction(vcpu);
4618 return 1;
4619}
4620
851ba692 4621static int handle_wrmsr(struct kvm_vcpu *vcpu)
6aa8b732 4622{
ad312c7c
ZX		 4623 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
4624 u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
4625 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
6aa8b732
AK		 4626
4627 if (vmx_set_msr(vcpu, ecx, data) != 0) {
59200273 4628 trace_kvm_msr_write_ex(ecx, data);
c1a5d4f9 4629 kvm_inject_gp(vcpu, 0);
6aa8b732
AK		 4630 return 1;
4631 }
4632
59200273 4633 trace_kvm_msr_write(ecx, data);
6aa8b732
AK		 4634 skip_emulated_instruction(vcpu);
4635 return 1;
4636}
4637
851ba692 4638static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
6e5d865c 4639{
3842d135 4640 kvm_make_request(KVM_REQ_EVENT, vcpu);
6e5d865c
YS		 4641 return 1;
4642}
4643
851ba692 4644static int handle_interrupt_window(struct kvm_vcpu *vcpu)
6aa8b732 4645{
85f455f7
ED		 4646 u32 cpu_based_vm_exec_control;
4647
4648 /* clear pending irq */
4649 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4650 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
4651 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2714d1d3 4652
3842d135
AK		 4653 kvm_make_request(KVM_REQ_EVENT, vcpu);
4654
a26bf12a 4655 ++vcpu->stat.irq_window_exits;
2714d1d3 4656
c1150d8c
DL		 4657 /*
4658 * If the user space waits to inject interrupts, exit as soon as
4659 * possible
4660 */
8061823a 4661 if (!irqchip_in_kernel(vcpu->kvm) &&
851ba692 4662 vcpu->run->request_interrupt_window &&
8061823a 4663 !kvm_cpu_has_interrupt(vcpu)) {
851ba692 4664 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
c1150d8c
DL		 4665 return 0;
4666 }
6aa8b732
AK		 4667 return 1;
4668}
4669
851ba692 4670static int handle_halt(struct kvm_vcpu *vcpu)
6aa8b732
AK		 4671{
4672 skip_emulated_instruction(vcpu);
d3bef15f 4673 return kvm_emulate_halt(vcpu);
6aa8b732
AK		 4674}
4675
851ba692 4676static int handle_vmcall(struct kvm_vcpu *vcpu)
c21415e8 4677{
510043da 4678 skip_emulated_instruction(vcpu);
7aa81cc0
AL		 4679 kvm_emulate_hypercall(vcpu);
4680 return 1;
c21415e8
IM		 4681}
4682
ec25d5e6
GN		 4683static int handle_invd(struct kvm_vcpu *vcpu)
4684{
51d8b661 4685 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
ec25d5e6
GN		 4686}
4687
851ba692 4688static int handle_invlpg(struct kvm_vcpu *vcpu)
a7052897 4689{
f9c617f6 4690 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
a7052897
MT		 4691
4692 kvm_mmu_invlpg(vcpu, exit_qualification);
4693 skip_emulated_instruction(vcpu);
4694 return 1;
4695}
4696
fee84b07
AK		 4697static int handle_rdpmc(struct kvm_vcpu *vcpu)
4698{
4699 int err;
4700
4701 err = kvm_rdpmc(vcpu);
4702 kvm_complete_insn_gp(vcpu, err);
4703
4704 return 1;
4705}
4706
851ba692 4707static int handle_wbinvd(struct kvm_vcpu *vcpu)
e5edaa01
ED		 4708{
4709 skip_emulated_instruction(vcpu);
f5f48ee1 4710 kvm_emulate_wbinvd(vcpu);
e5edaa01
ED		 4711 return 1;
4712}
4713
2acf923e
DC		 4714static int handle_xsetbv(struct kvm_vcpu *vcpu)
4715{
4716 u64 new_bv = kvm_read_edx_eax(vcpu);
4717 u32 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
4718
4719 if (kvm_set_xcr(vcpu, index, new_bv) == 0)
4720 skip_emulated_instruction(vcpu);
4721 return 1;
4722}
4723
851ba692 4724static int handle_apic_access(struct kvm_vcpu *vcpu)
f78e0e2e 4725{
58fbbf26
KT		 4726 if (likely(fasteoi)) {
4727 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4728 int access_type, offset;
4729
4730 access_type = exit_qualification & APIC_ACCESS_TYPE;
4731 offset = exit_qualification & APIC_ACCESS_OFFSET;
4732 /*
4733 * Sane guest uses MOV to write EOI, with written value
4734 * not cared. So make a short-circuit here by avoiding
4735 * heavy instruction emulation.
4736 */
4737 if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
4738 (offset == APIC_EOI)) {
4739 kvm_lapic_set_eoi(vcpu);
4740 skip_emulated_instruction(vcpu);
4741 return 1;
4742 }
4743 }
51d8b661 4744 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
f78e0e2e
SY		 4745}
4746
851ba692 4747static int handle_task_switch(struct kvm_vcpu *vcpu)
37817f29 4748{
60637aac 4749 struct vcpu_vmx *vmx = to_vmx(vcpu);
37817f29 4750 unsigned long exit_qualification;
e269fb21
JK		 4751 bool has_error_code = false;
4752 u32 error_code = 0;
37817f29 4753 u16 tss_selector;
7f3d35fd 4754 int reason, type, idt_v, idt_index;
64a7ec06
GN		 4755
4756 idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
7f3d35fd 4757 idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
64a7ec06 4758 type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
37817f29
IE		 4759
4760 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4761
4762 reason = (u32)exit_qualification >> 30;
64a7ec06
GN		 4763 if (reason == TASK_SWITCH_GATE && idt_v) {
4764 switch (type) {
4765 case INTR_TYPE_NMI_INTR:
4766 vcpu->arch.nmi_injected = false;
654f06fc 4767 vmx_set_nmi_mask(vcpu, true);
64a7ec06
GN		 4768 break;
4769 case INTR_TYPE_EXT_INTR:
66fd3f7f 4770 case INTR_TYPE_SOFT_INTR:
64a7ec06
GN		 4771 kvm_clear_interrupt_queue(vcpu);
4772 break;
4773 case INTR_TYPE_HARD_EXCEPTION:
e269fb21
JK		 4774 if (vmx->idt_vectoring_info &
4775 VECTORING_INFO_DELIVER_CODE_MASK) {
4776 has_error_code = true;
4777 error_code =
4778 vmcs_read32(IDT_VECTORING_ERROR_CODE);
4779 }
4780 /* fall through */
64a7ec06
GN		 4781 case INTR_TYPE_SOFT_EXCEPTION:
4782 kvm_clear_exception_queue(vcpu);
4783 break;
4784 default:
4785 break;
4786 }
60637aac 4787 }
37817f29
IE		 4788 tss_selector = exit_qualification;
4789
64a7ec06
GN		 4790 if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
4791 type != INTR_TYPE_EXT_INTR &&
4792 type != INTR_TYPE_NMI_INTR))
4793 skip_emulated_instruction(vcpu);
4794
7f3d35fd
KW		 4795 if (kvm_task_switch(vcpu, tss_selector,
4796 type == INTR_TYPE_SOFT_INTR ? idt_index : -1, reason,
4797 has_error_code, error_code) == EMULATE_FAIL) {
acb54517
GN		 4798 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4799 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4800 vcpu->run->internal.ndata = 0;
42dbaa5a 4801 return 0;
acb54517 4802 }
42dbaa5a
JK		 4803
4804 /* clear all local breakpoint enable flags */
4805 vmcs_writel(GUEST_DR7, vmcs_readl(GUEST_DR7) & ~55);
4806
4807 /*
4808 * TODO: What about debug traps on tss switch?
4809 * Are we supposed to inject them and update dr6?
4810 */
4811
4812 return 1;
37817f29
IE		 4813}
4814
851ba692 4815static int handle_ept_violation(struct kvm_vcpu *vcpu)
1439442c 4816{
f9c617f6 4817 unsigned long exit_qualification;
1439442c 4818 gpa_t gpa;
4f5982a5 4819 u32 error_code;
1439442c 4820 int gla_validity;
1439442c 4821
f9c617f6 4822 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
1439442c
SY		 4823
4824 if (exit_qualification & (1 << 6)) {
4825 printk(KERN_ERR "EPT: GPA exceeds GAW!\n");
7f582ab6 4826 return -EINVAL;
1439442c
SY		 4827 }
4828
4829 gla_validity = (exit_qualification >> 7) & 0x3;
4830 if (gla_validity != 0x3 && gla_validity != 0x1 && gla_validity != 0) {
4831 printk(KERN_ERR "EPT: Handling EPT violation failed!\n");
4832 printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n",
4833 (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS),
f9c617f6 4834 vmcs_readl(GUEST_LINEAR_ADDRESS));
1439442c
SY		 4835 printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n",
4836 (long unsigned int)exit_qualification);
851ba692
AK		 4837 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
4838 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_VIOLATION;
596ae895 4839 return 0;
1439442c
SY		 4840 }
4841
4842 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
229456fc 4843 trace_kvm_page_fault(gpa, exit_qualification);
4f5982a5
XG		 4844
4845 /* It is a write fault? */
4846 error_code = exit_qualification & (1U << 1);
4847 /* ept page table is present? */
4848 error_code |= (exit_qualification >> 3) & 0x1;
4849
4850 return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
1439442c
SY		 4851}
4852
68f89400
MT		 4853static u64 ept_rsvd_mask(u64 spte, int level)
4854{
4855 int i;
4856 u64 mask = 0;
4857
4858 for (i = 51; i > boot_cpu_data.x86_phys_bits; i--)
4859 mask |= (1ULL << i);
4860
4861 if (level > 2)
4862 /* bits 7:3 reserved */
4863 mask |= 0xf8;
4864 else if (level == 2) {
4865 if (spte & (1ULL << 7))
4866 /* 2MB ref, bits 20:12 reserved */
4867 mask |= 0x1ff000;
4868 else
4869 /* bits 6:3 reserved */
4870 mask |= 0x78;
4871 }
4872
4873 return mask;
4874}
4875
4876static void ept_misconfig_inspect_spte(struct kvm_vcpu *vcpu, u64 spte,
4877 int level)
4878{
4879 printk(KERN_ERR "%s: spte 0x%llx level %d\n", __func__, spte, level);
4880
4881 /* 010b (write-only) */
4882 WARN_ON((spte & 0x7) == 0x2);
4883
4884 /* 110b (write/execute) */
4885 WARN_ON((spte & 0x7) == 0x6);
4886
4887 /* 100b (execute-only) and value not supported by logical processor */
4888 if (!cpu_has_vmx_ept_execute_only())
4889 WARN_ON((spte & 0x7) == 0x4);
4890
4891 /* not 000b */
4892 if ((spte & 0x7)) {
4893 u64 rsvd_bits = spte & ept_rsvd_mask(spte, level);
4894
4895 if (rsvd_bits != 0) {
4896 printk(KERN_ERR "%s: rsvd_bits = 0x%llx\n",
4897 __func__, rsvd_bits);
4898 WARN_ON(1);
4899 }
4900
4901 if (level == 1 || (level == 2 && (spte & (1ULL << 7)))) {
4902 u64 ept_mem_type = (spte & 0x38) >> 3;
4903
4904 if (ept_mem_type == 2 || ept_mem_type == 3 ||
4905 ept_mem_type == 7) {
4906 printk(KERN_ERR "%s: ept_mem_type=0x%llx\n",
4907 __func__, ept_mem_type);
4908 WARN_ON(1);
4909 }
4910 }
4911 }
4912}
4913
851ba692 4914static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
68f89400
MT		 4915{
4916 u64 sptes[4];
ce88decf 4917 int nr_sptes, i, ret;
68f89400
MT		 4918 gpa_t gpa;
4919
4920 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
4921
ce88decf
XG		 4922 ret = handle_mmio_page_fault_common(vcpu, gpa, true);
4923 if (likely(ret == 1))
4924 return x86_emulate_instruction(vcpu, gpa, 0, NULL, 0) ==
4925 EMULATE_DONE;
4926 if (unlikely(!ret))
4927 return 1;
4928
4929 /* It is the real ept misconfig */
68f89400
MT		 4930 printk(KERN_ERR "EPT: Misconfiguration.\n");
4931 printk(KERN_ERR "EPT: GPA: 0x%llx\n", gpa);
4932
4933 nr_sptes = kvm_mmu_get_spte_hierarchy(vcpu, gpa, sptes);
4934
4935 for (i = PT64_ROOT_LEVEL; i > PT64_ROOT_LEVEL - nr_sptes; --i)
4936 ept_misconfig_inspect_spte(vcpu, sptes[i-1], i);
4937
851ba692
AK		 4938 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
4939 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_MISCONFIG;
68f89400
MT		 4940
4941 return 0;
4942}
4943
851ba692 4944static int handle_nmi_window(struct kvm_vcpu *vcpu)
f08864b4
SY		 4945{
4946 u32 cpu_based_vm_exec_control;
4947
4948 /* clear pending NMI */
4949 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4950 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
4951 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
4952 ++vcpu->stat.nmi_window_exits;
3842d135 4953 kvm_make_request(KVM_REQ_EVENT, vcpu);
f08864b4
SY		 4954
4955 return 1;
4956}
4957
80ced186 4958static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
ea953ef0 4959{
8b3079a5
AK		 4960 struct vcpu_vmx *vmx = to_vmx(vcpu);
4961 enum emulation_result err = EMULATE_DONE;
80ced186 4962 int ret = 1;
49e9d557
AK		 4963 u32 cpu_exec_ctrl;
4964 bool intr_window_requested;
b8405c18 4965 unsigned count = 130;
49e9d557
AK		 4966
4967 cpu_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4968 intr_window_requested = cpu_exec_ctrl & CPU_BASED_VIRTUAL_INTR_PENDING;
ea953ef0 4969
b8405c18 4970 while (!guest_state_valid(vcpu) && count-- != 0) {
bdea48e3 4971 if (intr_window_requested && vmx_interrupt_allowed(vcpu))
49e9d557
AK		 4972 return handle_interrupt_window(&vmx->vcpu);
4973
de87dcdd
AK		 4974 if (test_bit(KVM_REQ_EVENT, &vcpu->requests))
4975 return 1;
4976
51d8b661 4977 err = emulate_instruction(vcpu, 0);
ea953ef0 4978
80ced186
MG		 4979 if (err == EMULATE_DO_MMIO) {
4980 ret = 0;
4981 goto out;
4982 }
1d5a4d9b 4983
de5f70e0
AK		 4984 if (err != EMULATE_DONE) {
4985 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4986 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4987 vcpu->run->internal.ndata = 0;
6d77dbfc 4988 return 0;
de5f70e0 4989 }
ea953ef0
MG		 4990
4991 if (signal_pending(current))
80ced186 4992 goto out;
ea953ef0
MG		 4993 if (need_resched())
4994 schedule();
4995 }
4996
7c068e45 4997 vmx->emulation_required = !guest_state_valid(vcpu);
80ced186
MG		 4998out:
4999 return ret;
ea953ef0
MG		 5000}
5001
4b8d54f9
ZE		 5002/*
5003 * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
5004 * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
5005 */
9fb41ba8 5006static int handle_pause(struct kvm_vcpu *vcpu)
4b8d54f9
ZE		 5007{
5008 skip_emulated_instruction(vcpu);
5009 kvm_vcpu_on_spin(vcpu);
5010
5011 return 1;
5012}
5013
59708670
SY		 5014static int handle_invalid_op(struct kvm_vcpu *vcpu)
5015{
5016 kvm_queue_exception(vcpu, UD_VECTOR);
5017 return 1;
5018}
5019
ff2f6fe9
NHE		 5020/*
5021 * To run an L2 guest, we need a vmcs02 based on the L1-specified vmcs12.
5022 * We could reuse a single VMCS for all the L2 guests, but we also want the
5023 * option to allocate a separate vmcs02 for each separate loaded vmcs12 - this
5024 * allows keeping them loaded on the processor, and in the future will allow
5025 * optimizations where prepare_vmcs02 doesn't need to set all the fields on
5026 * every entry if they never change.
5027 * So we keep, in vmx->nested.vmcs02_pool, a cache of size VMCS02_POOL_SIZE
5028 * (>=0) with a vmcs02 for each recently loaded vmcs12s, most recent first.
5029 *
5030 * The following functions allocate and free a vmcs02 in this pool.
5031 */
5032
5033/* Get a VMCS from the pool to use as vmcs02 for the current vmcs12. */
5034static struct loaded_vmcs *nested_get_current_vmcs02(struct vcpu_vmx *vmx)
5035{
5036 struct vmcs02_list *item;
5037 list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
5038 if (item->vmptr == vmx->nested.current_vmptr) {
5039 list_move(&item->list, &vmx->nested.vmcs02_pool);
5040 return &item->vmcs02;
5041 }
5042
5043 if (vmx->nested.vmcs02_num >= max(VMCS02_POOL_SIZE, 1)) {
5044 /* Recycle the least recently used VMCS. */
5045 item = list_entry(vmx->nested.vmcs02_pool.prev,
5046 struct vmcs02_list, list);
5047 item->vmptr = vmx->nested.current_vmptr;
5048 list_move(&item->list, &vmx->nested.vmcs02_pool);
5049 return &item->vmcs02;
5050 }
5051
5052 /* Create a new VMCS */
5053 item = (struct vmcs02_list *)
5054 kmalloc(sizeof(struct vmcs02_list), GFP_KERNEL);
5055 if (!item)
5056 return NULL;
5057 item->vmcs02.vmcs = alloc_vmcs();
5058 if (!item->vmcs02.vmcs) {
5059 kfree(item);
5060 return NULL;
5061 }
5062 loaded_vmcs_init(&item->vmcs02);
5063 item->vmptr = vmx->nested.current_vmptr;
5064 list_add(&(item->list), &(vmx->nested.vmcs02_pool));
5065 vmx->nested.vmcs02_num++;
5066 return &item->vmcs02;
5067}
5068
5069/* Free and remove from pool a vmcs02 saved for a vmcs12 (if there is one) */
5070static void nested_free_vmcs02(struct vcpu_vmx *vmx, gpa_t vmptr)
5071{
5072 struct vmcs02_list *item;
5073 list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
5074 if (item->vmptr == vmptr) {
5075 free_loaded_vmcs(&item->vmcs02);
5076 list_del(&item->list);
5077 kfree(item);
5078 vmx->nested.vmcs02_num--;
5079 return;
5080 }
5081}
5082
5083/*
5084 * Free all VMCSs saved for this vcpu, except the one pointed by
5085 * vmx->loaded_vmcs. These include the VMCSs in vmcs02_pool (except the one
5086 * currently used, if running L2), and vmcs01 when running L2.
5087 */
5088static void nested_free_all_saved_vmcss(struct vcpu_vmx *vmx)
5089{
5090 struct vmcs02_list *item, *n;
5091 list_for_each_entry_safe(item, n, &vmx->nested.vmcs02_pool, list) {
5092 if (vmx->loaded_vmcs != &item->vmcs02)
5093 free_loaded_vmcs(&item->vmcs02);
5094 list_del(&item->list);
5095 kfree(item);
5096 }
5097 vmx->nested.vmcs02_num = 0;
5098
5099 if (vmx->loaded_vmcs != &vmx->vmcs01)
5100 free_loaded_vmcs(&vmx->vmcs01);
5101}
5102
ec378aee
NHE		 5103/*
5104 * Emulate the VMXON instruction.
5105 * Currently, we just remember that VMX is active, and do not save or even
5106 * inspect the argument to VMXON (the so-called "VMXON pointer") because we
5107 * do not currently need to store anything in that guest-allocated memory
5108 * region. Consequently, VMCLEAR and VMPTRLD also do not verify that the their
5109 * argument is different from the VMXON pointer (which the spec says they do).
5110 */
5111static int handle_vmon(struct kvm_vcpu *vcpu)
5112{
5113 struct kvm_segment cs;
5114 struct vcpu_vmx *vmx = to_vmx(vcpu);
5115
5116 /* The Intel VMX Instruction Reference lists a bunch of bits that
5117 * are prerequisite to running VMXON, most notably cr4.VMXE must be
5118 * set to 1 (see vmx_set_cr4() for when we allow the guest to set this).
5119 * Otherwise, we should fail with #UD. We test these now:
5120 */
5121 if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE) ||
5122 !kvm_read_cr0_bits(vcpu, X86_CR0_PE) ||
5123 (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
5124 kvm_queue_exception(vcpu, UD_VECTOR);
5125 return 1;
5126 }
5127
5128 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
5129 if (is_long_mode(vcpu) && !cs.l) {
5130 kvm_queue_exception(vcpu, UD_VECTOR);
5131 return 1;
5132 }
5133
5134 if (vmx_get_cpl(vcpu)) {
5135 kvm_inject_gp(vcpu, 0);
5136 return 1;
5137 }
5138
ff2f6fe9
NHE		 5139 INIT_LIST_HEAD(&(vmx->nested.vmcs02_pool));
5140 vmx->nested.vmcs02_num = 0;
5141
ec378aee
NHE		 5142 vmx->nested.vmxon = true;
5143
5144 skip_emulated_instruction(vcpu);
5145 return 1;
5146}
5147
5148/*
5149 * Intel's VMX Instruction Reference specifies a common set of prerequisites
5150 * for running VMX instructions (except VMXON, whose prerequisites are
5151 * slightly different). It also specifies what exception to inject otherwise.
5152 */
5153static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
5154{
5155 struct kvm_segment cs;
5156 struct vcpu_vmx *vmx = to_vmx(vcpu);
5157
5158 if (!vmx->nested.vmxon) {
5159 kvm_queue_exception(vcpu, UD_VECTOR);
5160 return 0;
5161 }
5162
5163 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
5164 if ((vmx_get_rflags(vcpu) & X86_EFLAGS_VM) ||
5165 (is_long_mode(vcpu) && !cs.l)) {
5166 kvm_queue_exception(vcpu, UD_VECTOR);
5167 return 0;
5168 }
5169
5170 if (vmx_get_cpl(vcpu)) {
5171 kvm_inject_gp(vcpu, 0);
5172 return 0;
5173 }
5174
5175 return 1;
5176}
5177
5178/*
5179 * Free whatever needs to be freed from vmx->nested when L1 goes down, or
5180 * just stops using VMX.
5181 */
5182static void free_nested(struct vcpu_vmx *vmx)
5183{
5184 if (!vmx->nested.vmxon)
5185 return;
5186 vmx->nested.vmxon = false;
a9d30f33
NHE		 5187 if (vmx->nested.current_vmptr != -1ull) {
5188 kunmap(vmx->nested.current_vmcs12_page);
5189 nested_release_page(vmx->nested.current_vmcs12_page);
5190 vmx->nested.current_vmptr = -1ull;
5191 vmx->nested.current_vmcs12 = NULL;
5192 }
fe3ef05c
NHE		 5193 /* Unpin physical memory we referred to in current vmcs02 */
5194 if (vmx->nested.apic_access_page) {
5195 nested_release_page(vmx->nested.apic_access_page);
5196 vmx->nested.apic_access_page = 0;
5197 }
ff2f6fe9
NHE		 5198
5199 nested_free_all_saved_vmcss(vmx);
ec378aee
NHE		 5200}
5201
5202/* Emulate the VMXOFF instruction */
5203static int handle_vmoff(struct kvm_vcpu *vcpu)
5204{
5205 if (!nested_vmx_check_permission(vcpu))
5206 return 1;
5207 free_nested(to_vmx(vcpu));
5208 skip_emulated_instruction(vcpu);
5209 return 1;
5210}
5211
064aea77
NHE		 5212/*
5213 * Decode the memory-address operand of a vmx instruction, as recorded on an
5214 * exit caused by such an instruction (run by a guest hypervisor).
5215 * On success, returns 0. When the operand is invalid, returns 1 and throws
5216 * #UD or #GP.
5217 */
5218static int get_vmx_mem_address(struct kvm_vcpu *vcpu,
5219 unsigned long exit_qualification,
5220 u32 vmx_instruction_info, gva_t *ret)
5221{
5222 /*
5223 * According to Vol. 3B, "Information for VM Exits Due to Instruction
5224 * Execution", on an exit, vmx_instruction_info holds most of the
5225 * addressing components of the operand. Only the displacement part
5226 * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
5227 * For how an actual address is calculated from all these components,
5228 * refer to Vol. 1, "Operand Addressing".
5229 */
5230 int scaling = vmx_instruction_info & 3;
5231 int addr_size = (vmx_instruction_info >> 7) & 7;
5232 bool is_reg = vmx_instruction_info & (1u << 10);
5233 int seg_reg = (vmx_instruction_info >> 15) & 7;
5234 int index_reg = (vmx_instruction_info >> 18) & 0xf;
5235 bool index_is_valid = !(vmx_instruction_info & (1u << 22));
5236 int base_reg = (vmx_instruction_info >> 23) & 0xf;
5237 bool base_is_valid = !(vmx_instruction_info & (1u << 27));
5238
5239 if (is_reg) {
5240 kvm_queue_exception(vcpu, UD_VECTOR);
5241 return 1;
5242 }
5243
5244 /* Addr = segment_base + offset */
5245 /* offset = base + [index * scale] + displacement */
5246 *ret = vmx_get_segment_base(vcpu, seg_reg);
5247 if (base_is_valid)
5248 *ret += kvm_register_read(vcpu, base_reg);
5249 if (index_is_valid)
5250 *ret += kvm_register_read(vcpu, index_reg)<<scaling;
5251 *ret += exit_qualification; /* holds the displacement */
5252
5253 if (addr_size == 1) /* 32 bit */
5254 *ret &= 0xffffffff;
5255
5256 /*
5257 * TODO: throw #GP (and return 1) in various cases that the VM*
5258 * instructions require it - e.g., offset beyond segment limit,
5259 * unusable or unreadable/unwritable segment, non-canonical 64-bit
5260 * address, and so on. Currently these are not checked.
5261 */
5262 return 0;
5263}
5264
0140caea
NHE		 5265/*
5266 * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(),
5267 * set the success or error code of an emulated VMX instruction, as specified
5268 * by Vol 2B, VMX Instruction Reference, "Conventions".
5269 */
5270static void nested_vmx_succeed(struct kvm_vcpu *vcpu)
5271{
5272 vmx_set_rflags(vcpu, vmx_get_rflags(vcpu)
5273 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
5274 X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF));
5275}
5276
5277static void nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
5278{
5279 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
5280 & ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
5281 X86_EFLAGS_SF | X86_EFLAGS_OF))
5282 | X86_EFLAGS_CF);
5283}
5284
5285static void nested_vmx_failValid(struct kvm_vcpu *vcpu,
5286 u32 vm_instruction_error)
5287{
5288 if (to_vmx(vcpu)->nested.current_vmptr == -1ull) {
5289 /*
5290 * failValid writes the error number to the current VMCS, which
5291 * can't be done there isn't a current VMCS.
5292 */
5293 nested_vmx_failInvalid(vcpu);
5294 return;
5295 }
5296 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
5297 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
5298 X86_EFLAGS_SF | X86_EFLAGS_OF))
5299 | X86_EFLAGS_ZF);
5300 get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error;
5301}
5302
27d6c865
NHE		 5303/* Emulate the VMCLEAR instruction */
5304static int handle_vmclear(struct kvm_vcpu *vcpu)
5305{
5306 struct vcpu_vmx *vmx = to_vmx(vcpu);
5307 gva_t gva;
5308 gpa_t vmptr;
5309 struct vmcs12 *vmcs12;
5310 struct page *page;
5311 struct x86_exception e;
5312
5313 if (!nested_vmx_check_permission(vcpu))
5314 return 1;
5315
5316 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
5317 vmcs_read32(VMX_INSTRUCTION_INFO), &gva))
5318 return 1;
5319
5320 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vmptr,
5321 sizeof(vmptr), &e)) {
5322 kvm_inject_page_fault(vcpu, &e);
5323 return 1;
5324 }
5325
5326 if (!IS_ALIGNED(vmptr, PAGE_SIZE)) {
5327 nested_vmx_failValid(vcpu, VMXERR_VMCLEAR_INVALID_ADDRESS);
5328 skip_emulated_instruction(vcpu);
5329 return 1;
5330 }
5331
5332 if (vmptr == vmx->nested.current_vmptr) {
5333 kunmap(vmx->nested.current_vmcs12_page);
5334 nested_release_page(vmx->nested.current_vmcs12_page);
5335 vmx->nested.current_vmptr = -1ull;
5336 vmx->nested.current_vmcs12 = NULL;
5337 }
5338
5339 page = nested_get_page(vcpu, vmptr);
5340 if (page == NULL) {
5341 /*
5342 * For accurate processor emulation, VMCLEAR beyond available
5343 * physical memory should do nothing at all. However, it is
5344 * possible that a nested vmx bug, not a guest hypervisor bug,
5345 * resulted in this case, so let's shut down before doing any
5346 * more damage:
5347 */
5348 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5349 return 1;
5350 }
5351 vmcs12 = kmap(page);
5352 vmcs12->launch_state = 0;
5353 kunmap(page);
5354 nested_release_page(page);
5355
5356 nested_free_vmcs02(vmx, vmptr);
5357
5358 skip_emulated_instruction(vcpu);
5359 nested_vmx_succeed(vcpu);
5360 return 1;
5361}
5362
cd232ad0
NHE		 5363static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch);
5364
5365/* Emulate the VMLAUNCH instruction */
5366static int handle_vmlaunch(struct kvm_vcpu *vcpu)
5367{
5368 return nested_vmx_run(vcpu, true);
5369}
5370
5371/* Emulate the VMRESUME instruction */
5372static int handle_vmresume(struct kvm_vcpu *vcpu)
5373{
5374
5375 return nested_vmx_run(vcpu, false);
5376}
5377
49f705c5
NHE		 5378enum vmcs_field_type {
5379 VMCS_FIELD_TYPE_U16 = 0,
5380 VMCS_FIELD_TYPE_U64 = 1,
5381 VMCS_FIELD_TYPE_U32 = 2,
5382 VMCS_FIELD_TYPE_NATURAL_WIDTH = 3
5383};
5384
5385static inline int vmcs_field_type(unsigned long field)
5386{
5387 if (0x1 & field) /* the *_HIGH fields are all 32 bit */
5388 return VMCS_FIELD_TYPE_U32;
5389 return (field >> 13) & 0x3 ;
5390}
5391
5392static inline int vmcs_field_readonly(unsigned long field)
5393{
5394 return (((field >> 10) & 0x3) == 1);
5395}
5396
5397/*
5398 * Read a vmcs12 field. Since these can have varying lengths and we return
5399 * one type, we chose the biggest type (u64) and zero-extend the return value
5400 * to that size. Note that the caller, handle_vmread, might need to use only
5401 * some of the bits we return here (e.g., on 32-bit guests, only 32 bits of
5402 * 64-bit fields are to be returned).
5403 */
5404static inline bool vmcs12_read_any(struct kvm_vcpu *vcpu,
5405 unsigned long field, u64 *ret)
5406{
5407 short offset = vmcs_field_to_offset(field);
5408 char *p;
5409
5410 if (offset < 0)
5411 return 0;
5412
5413 p = ((char *)(get_vmcs12(vcpu))) + offset;
5414
5415 switch (vmcs_field_type(field)) {
5416 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
5417 *ret = *((natural_width *)p);
5418 return 1;
5419 case VMCS_FIELD_TYPE_U16:
5420 *ret = *((u16 *)p);
5421 return 1;
5422 case VMCS_FIELD_TYPE_U32:
5423 *ret = *((u32 *)p);
5424 return 1;
5425 case VMCS_FIELD_TYPE_U64:
5426 *ret = *((u64 *)p);
5427 return 1;
5428 default:
5429 return 0; /* can never happen. */
5430 }
5431}
5432
5433/*
5434 * VMX instructions which assume a current vmcs12 (i.e., that VMPTRLD was
5435 * used before) all generate the same failure when it is missing.
5436 */
5437static int nested_vmx_check_vmcs12(struct kvm_vcpu *vcpu)
5438{
5439 struct vcpu_vmx *vmx = to_vmx(vcpu);
5440 if (vmx->nested.current_vmptr == -1ull) {
5441 nested_vmx_failInvalid(vcpu);
5442 skip_emulated_instruction(vcpu);
5443 return 0;
5444 }
5445 return 1;
5446}
5447
5448static int handle_vmread(struct kvm_vcpu *vcpu)
5449{
5450 unsigned long field;
5451 u64 field_value;
5452 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5453 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5454 gva_t gva = 0;
5455
5456 if (!nested_vmx_check_permission(vcpu) ||
5457 !nested_vmx_check_vmcs12(vcpu))
5458 return 1;
5459
5460 /* Decode instruction info and find the field to read */
5461 field = kvm_register_read(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
5462 /* Read the field, zero-extended to a u64 field_value */
5463 if (!vmcs12_read_any(vcpu, field, &field_value)) {
5464 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5465 skip_emulated_instruction(vcpu);
5466 return 1;
5467 }
5468 /*
5469 * Now copy part of this value to register or memory, as requested.
5470 * Note that the number of bits actually copied is 32 or 64 depending
5471 * on the guest's mode (32 or 64 bit), not on the given field's length.
5472 */
5473 if (vmx_instruction_info & (1u << 10)) {
5474 kvm_register_write(vcpu, (((vmx_instruction_info) >> 3) & 0xf),
5475 field_value);
5476 } else {
5477 if (get_vmx_mem_address(vcpu, exit_qualification,
5478 vmx_instruction_info, &gva))
5479 return 1;
5480 /* _system ok, as nested_vmx_check_permission verified cpl=0 */
5481 kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, gva,
5482 &field_value, (is_long_mode(vcpu) ? 8 : 4), NULL);
5483 }
5484
5485 nested_vmx_succeed(vcpu);
5486 skip_emulated_instruction(vcpu);
5487 return 1;
5488}
5489
5490
5491static int handle_vmwrite(struct kvm_vcpu *vcpu)
5492{
5493 unsigned long field;
5494 gva_t gva;
5495 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5496 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5497 char *p;
5498 short offset;
5499 /* The value to write might be 32 or 64 bits, depending on L1's long
5500 * mode, and eventually we need to write that into a field of several
5501 * possible lengths. The code below first zero-extends the value to 64
5502 * bit (field_value), and then copies only the approriate number of
5503 * bits into the vmcs12 field.
5504 */
5505 u64 field_value = 0;
5506 struct x86_exception e;
5507
5508 if (!nested_vmx_check_permission(vcpu) ||
5509 !nested_vmx_check_vmcs12(vcpu))
5510 return 1;
5511
5512 if (vmx_instruction_info & (1u << 10))
5513 field_value = kvm_register_read(vcpu,
5514 (((vmx_instruction_info) >> 3) & 0xf));
5515 else {
5516 if (get_vmx_mem_address(vcpu, exit_qualification,
5517 vmx_instruction_info, &gva))
5518 return 1;
5519 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva,
5520 &field_value, (is_long_mode(vcpu) ? 8 : 4), &e)) {
5521 kvm_inject_page_fault(vcpu, &e);
5522 return 1;
5523 }
5524 }
5525
5526
5527 field = kvm_register_read(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
5528 if (vmcs_field_readonly(field)) {
5529 nested_vmx_failValid(vcpu,
5530 VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
5531 skip_emulated_instruction(vcpu);
5532 return 1;
5533 }
5534
5535 offset = vmcs_field_to_offset(field);
5536 if (offset < 0) {
5537 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5538 skip_emulated_instruction(vcpu);
5539 return 1;
5540 }
5541 p = ((char *) get_vmcs12(vcpu)) + offset;
5542
5543 switch (vmcs_field_type(field)) {
5544 case VMCS_FIELD_TYPE_U16:
5545 *(u16 *)p = field_value;
5546 break;
5547 case VMCS_FIELD_TYPE_U32:
5548 *(u32 *)p = field_value;
5549 break;
5550 case VMCS_FIELD_TYPE_U64:
5551 *(u64 *)p = field_value;
5552 break;
5553 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
5554 *(natural_width *)p = field_value;
5555 break;
5556 default:
5557 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5558 skip_emulated_instruction(vcpu);
5559 return 1;
5560 }
5561
5562 nested_vmx_succeed(vcpu);
5563 skip_emulated_instruction(vcpu);
5564 return 1;
5565}
5566
63846663
NHE		 5567/* Emulate the VMPTRLD instruction */
5568static int handle_vmptrld(struct kvm_vcpu *vcpu)
5569{
5570 struct vcpu_vmx *vmx = to_vmx(vcpu);
5571 gva_t gva;
5572 gpa_t vmptr;
5573 struct x86_exception e;
5574
5575 if (!nested_vmx_check_permission(vcpu))
5576 return 1;
5577
5578 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
5579 vmcs_read32(VMX_INSTRUCTION_INFO), &gva))
5580 return 1;
5581
5582 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vmptr,
5583 sizeof(vmptr), &e)) {
5584 kvm_inject_page_fault(vcpu, &e);
5585 return 1;
5586 }
5587
5588 if (!IS_ALIGNED(vmptr, PAGE_SIZE)) {
5589 nested_vmx_failValid(vcpu, VMXERR_VMPTRLD_INVALID_ADDRESS);
5590 skip_emulated_instruction(vcpu);
5591 return 1;
5592 }
5593
5594 if (vmx->nested.current_vmptr != vmptr) {
5595 struct vmcs12 *new_vmcs12;
5596 struct page *page;
5597 page = nested_get_page(vcpu, vmptr);
5598 if (page == NULL) {
5599 nested_vmx_failInvalid(vcpu);
5600 skip_emulated_instruction(vcpu);
5601 return 1;
5602 }
5603 new_vmcs12 = kmap(page);
5604 if (new_vmcs12->revision_id != VMCS12_REVISION) {
5605 kunmap(page);
5606 nested_release_page_clean(page);
5607 nested_vmx_failValid(vcpu,
5608 VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5609 skip_emulated_instruction(vcpu);
5610 return 1;
5611 }
5612 if (vmx->nested.current_vmptr != -1ull) {
5613 kunmap(vmx->nested.current_vmcs12_page);
5614 nested_release_page(vmx->nested.current_vmcs12_page);
5615 }
5616
5617 vmx->nested.current_vmptr = vmptr;
5618 vmx->nested.current_vmcs12 = new_vmcs12;
5619 vmx->nested.current_vmcs12_page = page;
5620 }
5621
5622 nested_vmx_succeed(vcpu);
5623 skip_emulated_instruction(vcpu);
5624 return 1;
5625}
5626
6a4d7550
NHE		 5627/* Emulate the VMPTRST instruction */
5628static int handle_vmptrst(struct kvm_vcpu *vcpu)
5629{
5630 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5631 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5632 gva_t vmcs_gva;
5633 struct x86_exception e;
5634
5635 if (!nested_vmx_check_permission(vcpu))
5636 return 1;
5637
5638 if (get_vmx_mem_address(vcpu, exit_qualification,
5639 vmx_instruction_info, &vmcs_gva))
5640 return 1;
5641 /* ok to use *_system, as nested_vmx_check_permission verified cpl=0 */
5642 if (kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, vmcs_gva,
5643 (void *)&to_vmx(vcpu)->nested.current_vmptr,
5644 sizeof(u64), &e)) {
5645 kvm_inject_page_fault(vcpu, &e);
5646 return 1;
5647 }
5648 nested_vmx_succeed(vcpu);
5649 skip_emulated_instruction(vcpu);
5650 return 1;
5651}
5652
6aa8b732
AK		 5653/*
5654 * The exit handlers return 1 if the exit was handled fully and guest execution
5655 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
5656 * to be done to userspace and return 0.
5657 */
772e0318 5658static int (*const kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
6aa8b732
AK		 5659 [EXIT_REASON_EXCEPTION_NMI] = handle_exception,
5660 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
988ad74f 5661 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
f08864b4 5662 [EXIT_REASON_NMI_WINDOW] = handle_nmi_window,
6aa8b732 5663 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
6aa8b732
AK		 5664 [EXIT_REASON_CR_ACCESS] = handle_cr,
5665 [EXIT_REASON_DR_ACCESS] = handle_dr,
5666 [EXIT_REASON_CPUID] = handle_cpuid,
5667 [EXIT_REASON_MSR_READ] = handle_rdmsr,
5668 [EXIT_REASON_MSR_WRITE] = handle_wrmsr,
5669 [EXIT_REASON_PENDING_INTERRUPT] = handle_interrupt_window,
5670 [EXIT_REASON_HLT] = handle_halt,
ec25d5e6 5671 [EXIT_REASON_INVD] = handle_invd,
a7052897 5672 [EXIT_REASON_INVLPG] = handle_invlpg,
fee84b07 5673 [EXIT_REASON_RDPMC] = handle_rdpmc,
c21415e8 5674 [EXIT_REASON_VMCALL] = handle_vmcall,
27d6c865 5675 [EXIT_REASON_VMCLEAR] = handle_vmclear,
cd232ad0 5676 [EXIT_REASON_VMLAUNCH] = handle_vmlaunch,
63846663 5677 [EXIT_REASON_VMPTRLD] = handle_vmptrld,
6a4d7550 5678 [EXIT_REASON_VMPTRST] = handle_vmptrst,
49f705c5 5679 [EXIT_REASON_VMREAD] = handle_vmread,
cd232ad0 5680 [EXIT_REASON_VMRESUME] = handle_vmresume,
49f705c5 5681 [EXIT_REASON_VMWRITE] = handle_vmwrite,
ec378aee
NHE		 5682 [EXIT_REASON_VMOFF] = handle_vmoff,
5683 [EXIT_REASON_VMON] = handle_vmon,
f78e0e2e
SY		 5684 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold,
5685 [EXIT_REASON_APIC_ACCESS] = handle_apic_access,
e5edaa01 5686 [EXIT_REASON_WBINVD] = handle_wbinvd,
2acf923e 5687 [EXIT_REASON_XSETBV] = handle_xsetbv,
37817f29 5688 [EXIT_REASON_TASK_SWITCH] = handle_task_switch,
a0861c02 5689 [EXIT_REASON_MCE_DURING_VMENTRY] = handle_machine_check,
68f89400
MT		 5690 [EXIT_REASON_EPT_VIOLATION] = handle_ept_violation,
5691 [EXIT_REASON_EPT_MISCONFIG] = handle_ept_misconfig,
4b8d54f9 5692 [EXIT_REASON_PAUSE_INSTRUCTION] = handle_pause,
59708670
SY		 5693 [EXIT_REASON_MWAIT_INSTRUCTION] = handle_invalid_op,
5694 [EXIT_REASON_MONITOR_INSTRUCTION] = handle_invalid_op,
6aa8b732
AK		 5695};
5696
5697static const int kvm_vmx_max_exit_handlers =
50a3485c 5698 ARRAY_SIZE(kvm_vmx_exit_handlers);
6aa8b732 5699
644d711a
NHE		 5700/*
5701 * Return 1 if we should exit from L2 to L1 to handle an MSR access access,
5702 * rather than handle it ourselves in L0. I.e., check whether L1 expressed
5703 * disinterest in the current event (read or write a specific MSR) by using an
5704 * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
5705 */
5706static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
5707 struct vmcs12 *vmcs12, u32 exit_reason)
5708{
5709 u32 msr_index = vcpu->arch.regs[VCPU_REGS_RCX];
5710 gpa_t bitmap;
5711
5712 if (!nested_cpu_has(get_vmcs12(vcpu), CPU_BASED_USE_MSR_BITMAPS))
5713 return 1;
5714
5715 /*
5716 * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
5717 * for the four combinations of read/write and low/high MSR numbers.
5718 * First we need to figure out which of the four to use:
5719 */
5720 bitmap = vmcs12->msr_bitmap;
5721 if (exit_reason == EXIT_REASON_MSR_WRITE)
5722 bitmap += 2048;
5723 if (msr_index >= 0xc0000000) {
5724 msr_index -= 0xc0000000;
5725 bitmap += 1024;
5726 }
5727
5728 /* Then read the msr_index'th bit from this bitmap: */
5729 if (msr_index < 1024*8) {
5730 unsigned char b;
5731 kvm_read_guest(vcpu->kvm, bitmap + msr_index/8, &b, 1);
5732 return 1 & (b >> (msr_index & 7));
5733 } else
5734 return 1; /* let L1 handle the wrong parameter */
5735}
5736
5737/*
5738 * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
5739 * rather than handle it ourselves in L0. I.e., check if L1 wanted to
5740 * intercept (via guest_host_mask etc.) the current event.
5741 */
5742static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
5743 struct vmcs12 *vmcs12)
5744{
5745 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5746 int cr = exit_qualification & 15;
5747 int reg = (exit_qualification >> 8) & 15;
5748 unsigned long val = kvm_register_read(vcpu, reg);
5749
5750 switch ((exit_qualification >> 4) & 3) {
5751 case 0: /* mov to cr */
5752 switch (cr) {
5753 case 0:
5754 if (vmcs12->cr0_guest_host_mask &
5755 (val ^ vmcs12->cr0_read_shadow))
5756 return 1;
5757 break;
5758 case 3:
5759 if ((vmcs12->cr3_target_count >= 1 &&
5760 vmcs12->cr3_target_value0 == val) ||
5761 (vmcs12->cr3_target_count >= 2 &&
5762 vmcs12->cr3_target_value1 == val) ||
5763 (vmcs12->cr3_target_count >= 3 &&
5764 vmcs12->cr3_target_value2 == val) ||
5765 (vmcs12->cr3_target_count >= 4 &&
5766 vmcs12->cr3_target_value3 == val))
5767 return 0;
5768 if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
5769 return 1;
5770 break;
5771 case 4:
5772 if (vmcs12->cr4_guest_host_mask &
5773 (vmcs12->cr4_read_shadow ^ val))
5774 return 1;
5775 break;
5776 case 8:
5777 if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
5778 return 1;
5779 break;
5780 }
5781 break;
5782 case 2: /* clts */
5783 if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
5784 (vmcs12->cr0_read_shadow & X86_CR0_TS))
5785 return 1;
5786 break;
5787 case 1: /* mov from cr */
5788 switch (cr) {
5789 case 3:
5790 if (vmcs12->cpu_based_vm_exec_control &
5791 CPU_BASED_CR3_STORE_EXITING)
5792 return 1;
5793 break;
5794 case 8:
5795 if (vmcs12->cpu_based_vm_exec_control &
5796 CPU_BASED_CR8_STORE_EXITING)
5797 return 1;
5798 break;
5799 }
5800 break;
5801 case 3: /* lmsw */
5802 /*
5803 * lmsw can change bits 1..3 of cr0, and only set bit 0 of
5804 * cr0. Other attempted changes are ignored, with no exit.
5805 */
5806 if (vmcs12->cr0_guest_host_mask & 0xe &
5807 (val ^ vmcs12->cr0_read_shadow))
5808 return 1;
5809 if ((vmcs12->cr0_guest_host_mask & 0x1) &&
5810 !(vmcs12->cr0_read_shadow & 0x1) &&
5811 (val & 0x1))
5812 return 1;
5813 break;
5814 }
5815 return 0;
5816}
5817
5818/*
5819 * Return 1 if we should exit from L2 to L1 to handle an exit, or 0 if we
5820 * should handle it ourselves in L0 (and then continue L2). Only call this
5821 * when in is_guest_mode (L2).
5822 */
5823static bool nested_vmx_exit_handled(struct kvm_vcpu *vcpu)
5824{
5825 u32 exit_reason = vmcs_read32(VM_EXIT_REASON);
5826 u32 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
5827 struct vcpu_vmx *vmx = to_vmx(vcpu);
5828 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5829
5830 if (vmx->nested.nested_run_pending)
5831 return 0;
5832
5833 if (unlikely(vmx->fail)) {
bd80158a
JK		 5834 pr_info_ratelimited("%s failed vm entry %x\n", __func__,
5835 vmcs_read32(VM_INSTRUCTION_ERROR));
644d711a
NHE		 5836 return 1;
5837 }
5838
5839 switch (exit_reason) {
5840 case EXIT_REASON_EXCEPTION_NMI:
5841 if (!is_exception(intr_info))
5842 return 0;
5843 else if (is_page_fault(intr_info))
5844 return enable_ept;
5845 return vmcs12->exception_bitmap &
5846 (1u << (intr_info & INTR_INFO_VECTOR_MASK));
5847 case EXIT_REASON_EXTERNAL_INTERRUPT:
5848 return 0;
5849 case EXIT_REASON_TRIPLE_FAULT:
5850 return 1;
5851 case EXIT_REASON_PENDING_INTERRUPT:
5852 case EXIT_REASON_NMI_WINDOW:
5853 /*
5854 * prepare_vmcs02() set the CPU_BASED_VIRTUAL_INTR_PENDING bit
5855 * (aka Interrupt Window Exiting) only when L1 turned it on,
5856 * so if we got a PENDING_INTERRUPT exit, this must be for L1.
5857 * Same for NMI Window Exiting.
5858 */
5859 return 1;
5860 case EXIT_REASON_TASK_SWITCH:
5861 return 1;
5862 case EXIT_REASON_CPUID:
5863 return 1;
5864 case EXIT_REASON_HLT:
5865 return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
5866 case EXIT_REASON_INVD:
5867 return 1;
5868 case EXIT_REASON_INVLPG:
5869 return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
5870 case EXIT_REASON_RDPMC:
5871 return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
5872 case EXIT_REASON_RDTSC:
5873 return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
5874 case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
5875 case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
5876 case EXIT_REASON_VMPTRST: case EXIT_REASON_VMREAD:
5877 case EXIT_REASON_VMRESUME: case EXIT_REASON_VMWRITE:
5878 case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
5879 /*
5880 * VMX instructions trap unconditionally. This allows L1 to
5881 * emulate them for its L2 guest, i.e., allows 3-level nesting!
5882 */
5883 return 1;
5884 case EXIT_REASON_CR_ACCESS:
5885 return nested_vmx_exit_handled_cr(vcpu, vmcs12);
5886 case EXIT_REASON_DR_ACCESS:
5887 return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
5888 case EXIT_REASON_IO_INSTRUCTION:
5889 /* TODO: support IO bitmaps */
5890 return 1;
5891 case EXIT_REASON_MSR_READ:
5892 case EXIT_REASON_MSR_WRITE:
5893 return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
5894 case EXIT_REASON_INVALID_STATE:
5895 return 1;
5896 case EXIT_REASON_MWAIT_INSTRUCTION:
5897 return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
5898 case EXIT_REASON_MONITOR_INSTRUCTION:
5899 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
5900 case EXIT_REASON_PAUSE_INSTRUCTION:
5901 return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
5902 nested_cpu_has2(vmcs12,
5903 SECONDARY_EXEC_PAUSE_LOOP_EXITING);
5904 case EXIT_REASON_MCE_DURING_VMENTRY:
5905 return 0;
5906 case EXIT_REASON_TPR_BELOW_THRESHOLD:
5907 return 1;
5908 case EXIT_REASON_APIC_ACCESS:
5909 return nested_cpu_has2(vmcs12,
5910 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
5911 case EXIT_REASON_EPT_VIOLATION:
5912 case EXIT_REASON_EPT_MISCONFIG:
5913 return 0;
5914 case EXIT_REASON_WBINVD:
5915 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
5916 case EXIT_REASON_XSETBV:
5917 return 1;
5918 default:
5919 return 1;
5920 }
5921}
5922
586f9607
AK		 5923static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
5924{
5925 *info1 = vmcs_readl(EXIT_QUALIFICATION);
5926 *info2 = vmcs_read32(VM_EXIT_INTR_INFO);
5927}
5928
6aa8b732
AK		 5929/*
5930 * The guest has exited. See if we can fix it or if we need userspace
5931 * assistance.
5932 */
851ba692 5933static int vmx_handle_exit(struct kvm_vcpu *vcpu)
6aa8b732 5934{
29bd8a78 5935 struct vcpu_vmx *vmx = to_vmx(vcpu);
a0861c02 5936 u32 exit_reason = vmx->exit_reason;
1155f76a 5937 u32 vectoring_info = vmx->idt_vectoring_info;
29bd8a78 5938
80ced186
MG		 5939 /* If guest state is invalid, start emulating */
5940 if (vmx->emulation_required && emulate_invalid_guest_state)
5941 return handle_invalid_guest_state(vcpu);
1d5a4d9b 5942
b6f1250e
NHE		 5943 /*
5944 * the KVM_REQ_EVENT optimization bit is only on for one entry, and if
5945 * we did not inject a still-pending event to L1 now because of
5946 * nested_run_pending, we need to re-enable this bit.
5947 */
5948 if (vmx->nested.nested_run_pending)
5949 kvm_make_request(KVM_REQ_EVENT, vcpu);
5950
509c75ea
NHE		 5951 if (!is_guest_mode(vcpu) && (exit_reason == EXIT_REASON_VMLAUNCH ||
5952 exit_reason == EXIT_REASON_VMRESUME))
644d711a
NHE		 5953 vmx->nested.nested_run_pending = 1;
5954 else
5955 vmx->nested.nested_run_pending = 0;
5956
5957 if (is_guest_mode(vcpu) && nested_vmx_exit_handled(vcpu)) {
5958 nested_vmx_vmexit(vcpu);
5959 return 1;
5960 }
5961
5120702e
MG		 5962 if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
5963 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5964 vcpu->run->fail_entry.hardware_entry_failure_reason
5965 = exit_reason;
5966 return 0;
5967 }
5968
29bd8a78 5969 if (unlikely(vmx->fail)) {
851ba692
AK		 5970 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5971 vcpu->run->fail_entry.hardware_entry_failure_reason
29bd8a78
AK		 5972 = vmcs_read32(VM_INSTRUCTION_ERROR);
5973 return 0;
5974 }
6aa8b732 5975
d77c26fc 5976 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
1439442c 5977 (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
60637aac
JK		 5978 exit_reason != EXIT_REASON_EPT_VIOLATION &&
5979 exit_reason != EXIT_REASON_TASK_SWITCH))
5980 printk(KERN_WARNING "%s: unexpected, valid vectoring info "
5981 "(0x%x) and exit reason is 0x%x\n",
5982 __func__, vectoring_info, exit_reason);
3b86cd99 5983
644d711a
NHE		 5984 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked &&
5985 !(is_guest_mode(vcpu) && nested_cpu_has_virtual_nmis(
5986 get_vmcs12(vcpu), vcpu)))) {
c4282df9 5987 if (vmx_interrupt_allowed(vcpu)) {
3b86cd99 5988 vmx->soft_vnmi_blocked = 0;
3b86cd99 5989 } else if (vmx->vnmi_blocked_time > 1000000000LL &&
4531220b 5990 vcpu->arch.nmi_pending) {
3b86cd99
JK		 5991 /*
5992 * This CPU don't support us in finding the end of an
5993 * NMI-blocked window if the guest runs with IRQs
5994 * disabled. So we pull the trigger after 1 s of
5995 * futile waiting, but inform the user about this.
5996 */
5997 printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
5998 "state on VCPU %d after 1 s timeout\n",
5999 __func__, vcpu->vcpu_id);
6000 vmx->soft_vnmi_blocked = 0;
3b86cd99 6001 }
3b86cd99
JK		 6002 }
6003
6aa8b732
AK		 6004 if (exit_reason < kvm_vmx_max_exit_handlers
6005 && kvm_vmx_exit_handlers[exit_reason])
851ba692 6006 return kvm_vmx_exit_handlers[exit_reason](vcpu);
6aa8b732 6007 else {
851ba692
AK		 6008 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
6009 vcpu->run->hw.hardware_exit_reason = exit_reason;
6aa8b732
AK		 6010 }
6011 return 0;
6012}
6013
95ba8273 6014static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
6e5d865c 6015{
95ba8273 6016 if (irr == -1 || tpr < irr) {
6e5d865c
YS		 6017 vmcs_write32(TPR_THRESHOLD, 0);
6018 return;
6019 }
6020
95ba8273 6021 vmcs_write32(TPR_THRESHOLD, irr);
6e5d865c
YS		 6022}
6023
51aa01d1 6024static void vmx_complete_atomic_exit(struct vcpu_vmx *vmx)
cf393f75 6025{
00eba012
AK		 6026 u32 exit_intr_info;
6027
6028 if (!(vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY
6029 || vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI))
6030 return;
6031
c5ca8e57 6032 vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
00eba012 6033 exit_intr_info = vmx->exit_intr_info;
a0861c02
AK		 6034
6035 /* Handle machine checks before interrupts are enabled */
00eba012 6036 if (is_machine_check(exit_intr_info))
a0861c02
AK		 6037 kvm_machine_check();
6038
20f65983 6039 /* We need to handle NMIs before interrupts are enabled */
00eba012 6040 if ((exit_intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR &&
ff9d07a0
ZY		 6041 (exit_intr_info & INTR_INFO_VALID_MASK)) {
6042 kvm_before_handle_nmi(&vmx->vcpu);
20f65983 6043 asm("int $2");
ff9d07a0
ZY		 6044 kvm_after_handle_nmi(&vmx->vcpu);
6045 }
51aa01d1 6046}
20f65983 6047
51aa01d1
AK		 6048static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
6049{
c5ca8e57 6050 u32 exit_intr_info;
51aa01d1
AK		 6051 bool unblock_nmi;
6052 u8 vector;
6053 bool idtv_info_valid;
6054
6055 idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
20f65983 6056
cf393f75 6057 if (cpu_has_virtual_nmis()) {
9d58b931
AK		 6058 if (vmx->nmi_known_unmasked)
6059 return;
c5ca8e57
AK		 6060 /*
6061 * Can't use vmx->exit_intr_info since we're not sure what
6062 * the exit reason is.
6063 */
6064 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
cf393f75
AK		 6065 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
6066 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
6067 /*
7b4a25cb 6068 * SDM 3: 27.7.1.2 (September 2008)
cf393f75
AK		 6069 * Re-set bit "block by NMI" before VM entry if vmexit caused by
6070 * a guest IRET fault.
7b4a25cb
GN		 6071 * SDM 3: 23.2.2 (September 2008)
6072 * Bit 12 is undefined in any of the following cases:
6073 * If the VM exit sets the valid bit in the IDT-vectoring
6074 * information field.
6075 * If the VM exit is due to a double fault.
cf393f75 6076 */
7b4a25cb
GN		 6077 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
6078 vector != DF_VECTOR && !idtv_info_valid)
cf393f75
AK		 6079 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
6080 GUEST_INTR_STATE_NMI);
9d58b931
AK		 6081 else
6082 vmx->nmi_known_unmasked =
6083 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
6084 & GUEST_INTR_STATE_NMI);
3b86cd99
JK		 6085 } else if (unlikely(vmx->soft_vnmi_blocked))
6086 vmx->vnmi_blocked_time +=
6087 ktime_to_ns(ktime_sub(ktime_get(), vmx->entry_time));
51aa01d1
AK		 6088}
6089
83422e17
AK		 6090static void __vmx_complete_interrupts(struct vcpu_vmx *vmx,
6091 u32 idt_vectoring_info,
6092 int instr_len_field,
6093 int error_code_field)
51aa01d1 6094{
51aa01d1
AK		 6095 u8 vector;
6096 int type;
6097 bool idtv_info_valid;
6098
6099 idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
668f612f 6100
37b96e98
GN		 6101 vmx->vcpu.arch.nmi_injected = false;
6102 kvm_clear_exception_queue(&vmx->vcpu);
6103 kvm_clear_interrupt_queue(&vmx->vcpu);
6104
6105 if (!idtv_info_valid)
6106 return;
6107
3842d135
AK		 6108 kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
6109
668f612f
AK		 6110 vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
6111 type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
37b96e98 6112
64a7ec06 6113 switch (type) {
37b96e98
GN		 6114 case INTR_TYPE_NMI_INTR:
6115 vmx->vcpu.arch.nmi_injected = true;
668f612f 6116 /*
7b4a25cb 6117 * SDM 3: 27.7.1.2 (September 2008)
37b96e98
GN		 6118 * Clear bit "block by NMI" before VM entry if a NMI
6119 * delivery faulted.
668f612f 6120 */
654f06fc 6121 vmx_set_nmi_mask(&vmx->vcpu, false);
37b96e98 6122 break;
37b96e98 6123 case INTR_TYPE_SOFT_EXCEPTION:
66fd3f7f 6124 vmx->vcpu.arch.event_exit_inst_len =
83422e17 6125 vmcs_read32(instr_len_field);
66fd3f7f
GN		 6126 /* fall through */
6127 case INTR_TYPE_HARD_EXCEPTION:
35920a35 6128 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
83422e17 6129 u32 err = vmcs_read32(error_code_field);
37b96e98 6130 kvm_queue_exception_e(&vmx->vcpu, vector, err);
35920a35
AK		 6131 } else
6132 kvm_queue_exception(&vmx->vcpu, vector);
37b96e98 6133 break;
66fd3f7f
GN		 6134 case INTR_TYPE_SOFT_INTR:
6135 vmx->vcpu.arch.event_exit_inst_len =
83422e17 6136 vmcs_read32(instr_len_field);
66fd3f7f 6137 /* fall through */
37b96e98 6138 case INTR_TYPE_EXT_INTR:
66fd3f7f
GN		 6139 kvm_queue_interrupt(&vmx->vcpu, vector,
6140 type == INTR_TYPE_SOFT_INTR);
37b96e98
GN		 6141 break;
6142 default:
6143 break;
f7d9238f 6144 }
cf393f75
AK		 6145}
6146
83422e17
AK		 6147static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
6148{
66c78ae4
NHE		 6149 if (is_guest_mode(&vmx->vcpu))
6150 return;
83422e17
AK		 6151 __vmx_complete_interrupts(vmx, vmx->idt_vectoring_info,
6152 VM_EXIT_INSTRUCTION_LEN,
6153 IDT_VECTORING_ERROR_CODE);
6154}
6155
b463a6f7
AK		 6156static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
6157{
66c78ae4
NHE		 6158 if (is_guest_mode(vcpu))
6159 return;
b463a6f7
AK		 6160 __vmx_complete_interrupts(to_vmx(vcpu),
6161 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
6162 VM_ENTRY_INSTRUCTION_LEN,
6163 VM_ENTRY_EXCEPTION_ERROR_CODE);
6164
6165 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
6166}
6167
d7cd9796
GN		 6168static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
6169{
6170 int i, nr_msrs;
6171 struct perf_guest_switch_msr *msrs;
6172
6173 msrs = perf_guest_get_msrs(&nr_msrs);
6174
6175 if (!msrs)
6176 return;
6177
6178 for (i = 0; i < nr_msrs; i++)
6179 if (msrs[i].host == msrs[i].guest)
6180 clear_atomic_switch_msr(vmx, msrs[i].msr);
6181 else
6182 add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
6183 msrs[i].host);
6184}
6185
c801949d
AK		 6186#ifdef CONFIG_X86_64
6187#define R "r"
6188#define Q "q"
6189#else
6190#define R "e"
6191#define Q "l"
6192#endif
6193
a3b5ba49 6194static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
6aa8b732 6195{
a2fa3e9f 6196 struct vcpu_vmx *vmx = to_vmx(vcpu);
2a7921b7 6197 unsigned long debugctlmsr;
104f226b 6198
66c78ae4
NHE		 6199 if (is_guest_mode(vcpu) && !vmx->nested.nested_run_pending) {
6200 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6201 if (vmcs12->idt_vectoring_info_field &
6202 VECTORING_INFO_VALID_MASK) {
6203 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
6204 vmcs12->idt_vectoring_info_field);
6205 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
6206 vmcs12->vm_exit_instruction_len);
6207 if (vmcs12->idt_vectoring_info_field &
6208 VECTORING_INFO_DELIVER_CODE_MASK)
6209 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
6210 vmcs12->idt_vectoring_error_code);
6211 }
6212 }
6213
104f226b
AK		 6214 /* Record the guest's net vcpu time for enforced NMI injections. */
6215 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked))
6216 vmx->entry_time = ktime_get();
6217
6218 /* Don't enter VMX if guest state is invalid, let the exit handler
6219 start emulation until we arrive back to a valid state */
6220 if (vmx->emulation_required && emulate_invalid_guest_state)
6221 return;
6222
6223 if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
6224 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
6225 if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
6226 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
6227
6228 /* When single-stepping over STI and MOV SS, we must clear the
6229 * corresponding interruptibility bits in the guest state. Otherwise
6230 * vmentry fails as it then expects bit 14 (BS) in pending debug
6231 * exceptions being set, but that's not correct for the guest debugging
6232 * case. */
6233 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6234 vmx_set_interrupt_shadow(vcpu, 0);
6235
d7cd9796 6236 atomic_switch_perf_msrs(vmx);
2a7921b7 6237 debugctlmsr = get_debugctlmsr();
d7cd9796 6238
d462b819 6239 vmx->__launched = vmx->loaded_vmcs->launched;
104f226b 6240 asm(
6aa8b732 6241 /* Store host registers */
c801949d 6242 "push %%"R"dx; push %%"R"bp;"
40712fae 6243 "push %%"R"cx \n\t" /* placeholder for guest rcx */
c801949d 6244 "push %%"R"cx \n\t"
313dbd49
AK		 6245 "cmp %%"R"sp, %c[host_rsp](%0) \n\t"
6246 "je 1f \n\t"
6247 "mov %%"R"sp, %c[host_rsp](%0) \n\t"
4ecac3fd 6248 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
313dbd49 6249 "1: \n\t"
d3edefc0
AK		 6250 /* Reload cr2 if changed */
6251 "mov %c[cr2](%0), %%"R"ax \n\t"
6252 "mov %%cr2, %%"R"dx \n\t"
6253 "cmp %%"R"ax, %%"R"dx \n\t"
6254 "je 2f \n\t"
6255 "mov %%"R"ax, %%cr2 \n\t"
6256 "2: \n\t"
6aa8b732 6257 /* Check if vmlaunch of vmresume is needed */
e08aa78a 6258 "cmpl $0, %c[launched](%0) \n\t"
6aa8b732 6259 /* Load guest registers. Don't clobber flags. */
c801949d
AK		 6260 "mov %c[rax](%0), %%"R"ax \n\t"
6261 "mov %c[rbx](%0), %%"R"bx \n\t"
6262 "mov %c[rdx](%0), %%"R"dx \n\t"
6263 "mov %c[rsi](%0), %%"R"si \n\t"
6264 "mov %c[rdi](%0), %%"R"di \n\t"
6265 "mov %c[rbp](%0), %%"R"bp \n\t"
05b3e0c2 6266#ifdef CONFIG_X86_64
e08aa78a
AK		 6267 "mov %c[r8](%0), %%r8 \n\t"
6268 "mov %c[r9](%0), %%r9 \n\t"
6269 "mov %c[r10](%0), %%r10 \n\t"
6270 "mov %c[r11](%0), %%r11 \n\t"
6271 "mov %c[r12](%0), %%r12 \n\t"
6272 "mov %c[r13](%0), %%r13 \n\t"
6273 "mov %c[r14](%0), %%r14 \n\t"
6274 "mov %c[r15](%0), %%r15 \n\t"
6aa8b732 6275#endif
c801949d
AK		 6276 "mov %c[rcx](%0), %%"R"cx \n\t" /* kills %0 (ecx) */
6277
6aa8b732 6278 /* Enter guest mode */
cd2276a7 6279 "jne .Llaunched \n\t"
4ecac3fd 6280 __ex(ASM_VMX_VMLAUNCH) "\n\t"
cd2276a7 6281 "jmp .Lkvm_vmx_return \n\t"
4ecac3fd 6282 ".Llaunched: " __ex(ASM_VMX_VMRESUME) "\n\t"
cd2276a7 6283 ".Lkvm_vmx_return: "
6aa8b732 6284 /* Save guest registers, load host registers, keep flags */
40712fae
AK		 6285 "mov %0, %c[wordsize](%%"R"sp) \n\t"
6286 "pop %0 \n\t"
c801949d
AK		 6287 "mov %%"R"ax, %c[rax](%0) \n\t"
6288 "mov %%"R"bx, %c[rbx](%0) \n\t"
1c696d0e 6289 "pop"Q" %c[rcx](%0) \n\t"
c801949d
AK		 6290 "mov %%"R"dx, %c[rdx](%0) \n\t"
6291 "mov %%"R"si, %c[rsi](%0) \n\t"
6292 "mov %%"R"di, %c[rdi](%0) \n\t"
6293 "mov %%"R"bp, %c[rbp](%0) \n\t"
05b3e0c2 6294#ifdef CONFIG_X86_64
e08aa78a
AK		 6295 "mov %%r8, %c[r8](%0) \n\t"
6296 "mov %%r9, %c[r9](%0) \n\t"
6297 "mov %%r10, %c[r10](%0) \n\t"
6298 "mov %%r11, %c[r11](%0) \n\t"
6299 "mov %%r12, %c[r12](%0) \n\t"
6300 "mov %%r13, %c[r13](%0) \n\t"
6301 "mov %%r14, %c[r14](%0) \n\t"
6302 "mov %%r15, %c[r15](%0) \n\t"
6aa8b732 6303#endif
c801949d
AK		 6304 "mov %%cr2, %%"R"ax \n\t"
6305 "mov %%"R"ax, %c[cr2](%0) \n\t"
6306
1c696d0e 6307 "pop %%"R"bp; pop %%"R"dx \n\t"
e08aa78a
AK		 6308 "setbe %c[fail](%0) \n\t"
6309 : : "c"(vmx), "d"((unsigned long)HOST_RSP),
d462b819 6310 [launched]"i"(offsetof(struct vcpu_vmx, __launched)),
e08aa78a 6311 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
313dbd49 6312 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
ad312c7c
ZX		 6313 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
6314 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
6315 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
6316 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
6317 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
6318 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
6319 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
05b3e0c2 6320#ifdef CONFIG_X86_64
ad312c7c
ZX		 6321 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
6322 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
6323 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
6324 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
6325 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
6326 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
6327 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
6328 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
6aa8b732 6329#endif
40712fae
AK		 6330 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2)),
6331 [wordsize]"i"(sizeof(ulong))
c2036300 6332 : "cc", "memory"
07d6f555 6333 , R"ax", R"bx", R"di", R"si"
c2036300 6334#ifdef CONFIG_X86_64
c2036300
LV		 6335 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
6336#endif
6337 );
6aa8b732 6338
2a7921b7
GN		 6339 /* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
6340 if (debugctlmsr)
6341 update_debugctlmsr(debugctlmsr);
6342
aa67f609
AK		 6343#ifndef CONFIG_X86_64
6344 /*
6345 * The sysexit path does not restore ds/es, so we must set them to
6346 * a reasonable value ourselves.
6347 *
6348 * We can't defer this to vmx_load_host_state() since that function
6349 * may be executed in interrupt context, which saves and restore segments
6350 * around it, nullifying its effect.
6351 */
6352 loadsegment(ds, __USER_DS);
6353 loadsegment(es, __USER_DS);
6354#endif
6355
6de4f3ad 6356 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
6de12732 6357 | (1 << VCPU_EXREG_RFLAGS)
69c73028 6358 | (1 << VCPU_EXREG_CPL)
aff48baa 6359 | (1 << VCPU_EXREG_PDPTR)
2fb92db1 6360 | (1 << VCPU_EXREG_SEGMENTS)
aff48baa 6361 | (1 << VCPU_EXREG_CR3));
5fdbf976
MT		 6362 vcpu->arch.regs_dirty = 0;
6363
1155f76a
AK		 6364 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
6365
66c78ae4
NHE		 6366 if (is_guest_mode(vcpu)) {
6367 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6368 vmcs12->idt_vectoring_info_field = vmx->idt_vectoring_info;
6369 if (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK) {
6370 vmcs12->idt_vectoring_error_code =
6371 vmcs_read32(IDT_VECTORING_ERROR_CODE);
6372 vmcs12->vm_exit_instruction_len =
6373 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
6374 }
6375 }
6376
d462b819 6377 vmx->loaded_vmcs->launched = 1;
1b6269db 6378
51aa01d1 6379 vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);
1e2b1dd7 6380 trace_kvm_exit(vmx->exit_reason, vcpu, KVM_ISA_VMX);
51aa01d1
AK		 6381
6382 vmx_complete_atomic_exit(vmx);
6383 vmx_recover_nmi_blocking(vmx);
cf393f75 6384 vmx_complete_interrupts(vmx);
6aa8b732
AK		 6385}
6386
c801949d
AK		 6387#undef R
6388#undef Q
6389
6aa8b732
AK		 6390static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
6391{
fb3f0f51
RR		 6392 struct vcpu_vmx *vmx = to_vmx(vcpu);
6393
cdbecfc3 6394 free_vpid(vmx);
ec378aee 6395 free_nested(vmx);
d462b819 6396 free_loaded_vmcs(vmx->loaded_vmcs);
fb3f0f51
RR		 6397 kfree(vmx->guest_msrs);
6398 kvm_vcpu_uninit(vcpu);
a4770347 6399 kmem_cache_free(kvm_vcpu_cache, vmx);
6aa8b732
AK		 6400}
6401
fb3f0f51 6402static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
6aa8b732 6403{
fb3f0f51 6404 int err;
c16f862d 6405 struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
15ad7146 6406 int cpu;
6aa8b732 6407
a2fa3e9f 6408 if (!vmx)
fb3f0f51
RR		 6409 return ERR_PTR(-ENOMEM);
6410
2384d2b3
SY		 6411 allocate_vpid(vmx);
6412
fb3f0f51
RR		 6413 err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
6414 if (err)
6415 goto free_vcpu;
965b58a5 6416
a2fa3e9f 6417 vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
be6d05cf 6418 err = -ENOMEM;
fb3f0f51 6419 if (!vmx->guest_msrs) {
fb3f0f51
RR		 6420 goto uninit_vcpu;
6421 }
965b58a5 6422
d462b819
NHE		 6423 vmx->loaded_vmcs = &vmx->vmcs01;
6424 vmx->loaded_vmcs->vmcs = alloc_vmcs();
6425 if (!vmx->loaded_vmcs->vmcs)
fb3f0f51 6426 goto free_msrs;
d462b819
NHE		 6427 if (!vmm_exclusive)
6428 kvm_cpu_vmxon(__pa(per_cpu(vmxarea, raw_smp_processor_id())));
6429 loaded_vmcs_init(vmx->loaded_vmcs);
6430 if (!vmm_exclusive)
6431 kvm_cpu_vmxoff();
a2fa3e9f 6432
15ad7146
AK		 6433 cpu = get_cpu();
6434 vmx_vcpu_load(&vmx->vcpu, cpu);
e48672fa 6435 vmx->vcpu.cpu = cpu;
8b9cf98c 6436 err = vmx_vcpu_setup(vmx);
fb3f0f51 6437 vmx_vcpu_put(&vmx->vcpu);
15ad7146 6438 put_cpu();
fb3f0f51
RR		 6439 if (err)
6440 goto free_vmcs;
5e4a0b3c 6441 if (vm_need_virtualize_apic_accesses(kvm))
be6d05cf
JK		 6442 err = alloc_apic_access_page(kvm);
6443 if (err)
5e4a0b3c 6444 goto free_vmcs;
fb3f0f51 6445
b927a3ce
SY		 6446 if (enable_ept) {
6447 if (!kvm->arch.ept_identity_map_addr)
6448 kvm->arch.ept_identity_map_addr =
6449 VMX_EPT_IDENTITY_PAGETABLE_ADDR;
93ea5388 6450 err = -ENOMEM;
b7ebfb05
SY		 6451 if (alloc_identity_pagetable(kvm) != 0)
6452 goto free_vmcs;
93ea5388
GN		 6453 if (!init_rmode_identity_map(kvm))
6454 goto free_vmcs;
b927a3ce 6455 }
b7ebfb05 6456
a9d30f33
NHE		 6457 vmx->nested.current_vmptr = -1ull;
6458 vmx->nested.current_vmcs12 = NULL;
6459
fb3f0f51
RR		 6460 return &vmx->vcpu;
6461
6462free_vmcs:
5f3fbc34 6463 free_loaded_vmcs(vmx->loaded_vmcs);
fb3f0f51 6464free_msrs:
fb3f0f51
RR		 6465 kfree(vmx->guest_msrs);
6466uninit_vcpu:
6467 kvm_vcpu_uninit(&vmx->vcpu);
6468free_vcpu:
cdbecfc3 6469 free_vpid(vmx);
a4770347 6470 kmem_cache_free(kvm_vcpu_cache, vmx);
fb3f0f51 6471 return ERR_PTR(err);
6aa8b732
AK		 6472}
6473
002c7f7c
YS		 6474static void __init vmx_check_processor_compat(void *rtn)
6475{
6476 struct vmcs_config vmcs_conf;
6477
6478 *(int *)rtn = 0;
6479 if (setup_vmcs_config(&vmcs_conf) < 0)
6480 *(int *)rtn = -EIO;
6481 if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
6482 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
6483 smp_processor_id());
6484 *(int *)rtn = -EIO;
6485 }
6486}
6487
67253af5
SY		 6488static int get_ept_level(void)
6489{
6490 return VMX_EPT_DEFAULT_GAW + 1;
6491}
6492
4b12f0de 6493static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
64d4d521 6494{
4b12f0de
SY		 6495 u64 ret;
6496
522c68c4
SY		 6497 /* For VT-d and EPT combination
6498 * 1. MMIO: always map as UC
6499 * 2. EPT with VT-d:
6500 * a. VT-d without snooping control feature: can't guarantee the
6501 * result, try to trust guest.
6502 * b. VT-d with snooping control feature: snooping control feature of
6503 * VT-d engine can guarantee the cache correctness. Just set it
6504 * to WB to keep consistent with host. So the same as item 3.
a19a6d11 6505 * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
522c68c4
SY		 6506 * consistent with host MTRR
6507 */
4b12f0de
SY		 6508 if (is_mmio)
6509 ret = MTRR_TYPE_UNCACHABLE << VMX_EPT_MT_EPTE_SHIFT;
522c68c4
SY		 6510 else if (vcpu->kvm->arch.iommu_domain &&
6511 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY))
6512 ret = kvm_get_guest_memory_type(vcpu, gfn) <<
6513 VMX_EPT_MT_EPTE_SHIFT;
4b12f0de 6514 else
522c68c4 6515 ret = (MTRR_TYPE_WRBACK << VMX_EPT_MT_EPTE_SHIFT)
a19a6d11 6516 | VMX_EPT_IPAT_BIT;
4b12f0de
SY		 6517
6518 return ret;
64d4d521
SY		 6519}
6520
17cc3935 6521static int vmx_get_lpage_level(void)
344f414f 6522{
878403b7
SY		 6523 if (enable_ept && !cpu_has_vmx_ept_1g_page())
6524 return PT_DIRECTORY_LEVEL;
6525 else
6526 /* For shadow and EPT supported 1GB page */
6527 return PT_PDPE_LEVEL;
344f414f
JR		 6528}
6529
0e851880
SY		 6530static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
6531{
4e47c7a6
SY		 6532 struct kvm_cpuid_entry2 *best;
6533 struct vcpu_vmx *vmx = to_vmx(vcpu);
6534 u32 exec_control;
6535
6536 vmx->rdtscp_enabled = false;
6537 if (vmx_rdtscp_supported()) {
6538 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
6539 if (exec_control & SECONDARY_EXEC_RDTSCP) {
6540 best = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
6541 if (best && (best->edx & bit(X86_FEATURE_RDTSCP)))
6542 vmx->rdtscp_enabled = true;
6543 else {
6544 exec_control &= ~SECONDARY_EXEC_RDTSCP;
6545 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
6546 exec_control);
6547 }
6548 }
6549 }
ad756a16
MJ		 6550
6551 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
6552 /* Exposing INVPCID only when PCID is exposed */
6553 best = kvm_find_cpuid_entry(vcpu, 0x7, 0);
6554 if (vmx_invpcid_supported() &&
6555 best && (best->ecx & bit(X86_FEATURE_INVPCID)) &&
6556 guest_cpuid_has_pcid(vcpu)) {
6557 exec_control |= SECONDARY_EXEC_ENABLE_INVPCID;
6558 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
6559 exec_control);
6560 } else {
6561 exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
6562 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
6563 exec_control);
6564 if (best)
6565 best->ecx &= ~bit(X86_FEATURE_INVPCID);
6566 }
0e851880
SY		 6567}
6568
d4330ef2
JR		 6569static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
6570{
7b8050f5
NHE		 6571 if (func == 1 && nested)
6572 entry->ecx |= bit(X86_FEATURE_VMX);
d4330ef2
JR		 6573}
6574
fe3ef05c
NHE		 6575/*
6576 * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
6577 * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
6578 * with L0's requirements for its guest (a.k.a. vmsc01), so we can run the L2
6579 * guest in a way that will both be appropriate to L1's requests, and our
6580 * needs. In addition to modifying the active vmcs (which is vmcs02), this
6581 * function also has additional necessary side-effects, like setting various
6582 * vcpu->arch fields.
6583 */
6584static void prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
6585{
6586 struct vcpu_vmx *vmx = to_vmx(vcpu);
6587 u32 exec_control;
6588
6589 vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
6590 vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
6591 vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
6592 vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
6593 vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
6594 vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
6595 vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
6596 vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
6597 vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
6598 vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
6599 vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
6600 vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
6601 vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
6602 vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
6603 vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
6604 vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
6605 vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
6606 vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
6607 vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
6608 vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
6609 vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
6610 vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
6611 vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
6612 vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
6613 vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
6614 vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
6615 vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
6616 vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
6617 vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
6618 vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
6619 vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
6620 vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
6621 vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
6622 vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
6623 vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
6624 vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
6625
6626 vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl);
6627 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
6628 vmcs12->vm_entry_intr_info_field);
6629 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
6630 vmcs12->vm_entry_exception_error_code);
6631 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
6632 vmcs12->vm_entry_instruction_len);
6633 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
6634 vmcs12->guest_interruptibility_info);
6635 vmcs_write32(GUEST_ACTIVITY_STATE, vmcs12->guest_activity_state);
6636 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
6637 vmcs_writel(GUEST_DR7, vmcs12->guest_dr7);
6638 vmcs_writel(GUEST_RFLAGS, vmcs12->guest_rflags);
6639 vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
6640 vmcs12->guest_pending_dbg_exceptions);
6641 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
6642 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
6643
6644 vmcs_write64(VMCS_LINK_POINTER, -1ull);
6645
6646 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL,
6647 (vmcs_config.pin_based_exec_ctrl |
6648 vmcs12->pin_based_vm_exec_control));
6649
6650 /*
6651 * Whether page-faults are trapped is determined by a combination of
6652 * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.
6653 * If enable_ept, L0 doesn't care about page faults and we should
6654 * set all of these to L1's desires. However, if !enable_ept, L0 does
6655 * care about (at least some) page faults, and because it is not easy
6656 * (if at all possible?) to merge L0 and L1's desires, we simply ask
6657 * to exit on each and every L2 page fault. This is done by setting
6658 * MASK=MATCH=0 and (see below) EB.PF=1.
6659 * Note that below we don't need special code to set EB.PF beyond the
6660 * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
6661 * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
6662 * !enable_ept, EB.PF is 1, so the "or" will always be 1.
6663 *
6664 * A problem with this approach (when !enable_ept) is that L1 may be
6665 * injected with more page faults than it asked for. This could have
6666 * caused problems, but in practice existing hypervisors don't care.
6667 * To fix this, we will need to emulate the PFEC checking (on the L1
6668 * page tables), using walk_addr(), when injecting PFs to L1.
6669 */
6670 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK,
6671 enable_ept ? vmcs12->page_fault_error_code_mask : 0);
6672 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH,
6673 enable_ept ? vmcs12->page_fault_error_code_match : 0);
6674
6675 if (cpu_has_secondary_exec_ctrls()) {
6676 u32 exec_control = vmx_secondary_exec_control(vmx);
6677 if (!vmx->rdtscp_enabled)
6678 exec_control &= ~SECONDARY_EXEC_RDTSCP;
6679 /* Take the following fields only from vmcs12 */
6680 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
6681 if (nested_cpu_has(vmcs12,
6682 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS))
6683 exec_control |= vmcs12->secondary_vm_exec_control;
6684
6685 if (exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) {
6686 /*
6687 * Translate L1 physical address to host physical
6688 * address for vmcs02. Keep the page pinned, so this
6689 * physical address remains valid. We keep a reference
6690 * to it so we can release it later.
6691 */
6692 if (vmx->nested.apic_access_page) /* shouldn't happen */
6693 nested_release_page(vmx->nested.apic_access_page);
6694 vmx->nested.apic_access_page =
6695 nested_get_page(vcpu, vmcs12->apic_access_addr);
6696 /*
6697 * If translation failed, no matter: This feature asks
6698 * to exit when accessing the given address, and if it
6699 * can never be accessed, this feature won't do
6700 * anything anyway.
6701 */
6702 if (!vmx->nested.apic_access_page)
6703 exec_control &=
6704 ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
6705 else
6706 vmcs_write64(APIC_ACCESS_ADDR,
6707 page_to_phys(vmx->nested.apic_access_page));
6708 }
6709
6710 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
6711 }
6712
6713
6714 /*
6715 * Set host-state according to L0's settings (vmcs12 is irrelevant here)
6716 * Some constant fields are set here by vmx_set_constant_host_state().
6717 * Other fields are different per CPU, and will be set later when
6718 * vmx_vcpu_load() is called, and when vmx_save_host_state() is called.
6719 */
6720 vmx_set_constant_host_state();
6721
6722 /*
6723 * HOST_RSP is normally set correctly in vmx_vcpu_run() just before
6724 * entry, but only if the current (host) sp changed from the value
6725 * we wrote last (vmx->host_rsp). This cache is no longer relevant
6726 * if we switch vmcs, and rather than hold a separate cache per vmcs,
6727 * here we just force the write to happen on entry.
6728 */
6729 vmx->host_rsp = 0;
6730
6731 exec_control = vmx_exec_control(vmx); /* L0's desires */
6732 exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
6733 exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
6734 exec_control &= ~CPU_BASED_TPR_SHADOW;
6735 exec_control |= vmcs12->cpu_based_vm_exec_control;
6736 /*
6737 * Merging of IO and MSR bitmaps not currently supported.
6738 * Rather, exit every time.
6739 */
6740 exec_control &= ~CPU_BASED_USE_MSR_BITMAPS;
6741 exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
6742 exec_control |= CPU_BASED_UNCOND_IO_EXITING;
6743
6744 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
6745
6746 /* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
6747 * bitwise-or of what L1 wants to trap for L2, and what we want to
6748 * trap. Note that CR0.TS also needs updating - we do this later.
6749 */
6750 update_exception_bitmap(vcpu);
6751 vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
6752 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
6753
6754 /* Note: IA32_MODE, LOAD_IA32_EFER are modified by vmx_set_efer below */
6755 vmcs_write32(VM_EXIT_CONTROLS,
6756 vmcs12->vm_exit_controls | vmcs_config.vmexit_ctrl);
6757 vmcs_write32(VM_ENTRY_CONTROLS, vmcs12->vm_entry_controls |
6758 (vmcs_config.vmentry_ctrl & ~VM_ENTRY_IA32E_MODE));
6759
6760 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT)
6761 vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
6762 else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
6763 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
6764
6765
6766 set_cr4_guest_host_mask(vmx);
6767
27fc51b2
NHE		 6768 if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING)
6769 vmcs_write64(TSC_OFFSET,
6770 vmx->nested.vmcs01_tsc_offset + vmcs12->tsc_offset);
6771 else
6772 vmcs_write64(TSC_OFFSET, vmx->nested.vmcs01_tsc_offset);
fe3ef05c
NHE		 6773
6774 if (enable_vpid) {
6775 /*
6776 * Trivially support vpid by letting L2s share their parent
6777 * L1's vpid. TODO: move to a more elaborate solution, giving
6778 * each L2 its own vpid and exposing the vpid feature to L1.
6779 */
6780 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
6781 vmx_flush_tlb(vcpu);
6782 }
6783
6784 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)
6785 vcpu->arch.efer = vmcs12->guest_ia32_efer;
6786 if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
6787 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
6788 else
6789 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
6790 /* Note: modifies VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
6791 vmx_set_efer(vcpu, vcpu->arch.efer);
6792
6793 /*
6794 * This sets GUEST_CR0 to vmcs12->guest_cr0, with possibly a modified
6795 * TS bit (for lazy fpu) and bits which we consider mandatory enabled.
6796 * The CR0_READ_SHADOW is what L2 should have expected to read given
6797 * the specifications by L1; It's not enough to take
6798 * vmcs12->cr0_read_shadow because on our cr0_guest_host_mask we we
6799 * have more bits than L1 expected.
6800 */
6801 vmx_set_cr0(vcpu, vmcs12->guest_cr0);
6802 vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
6803
6804 vmx_set_cr4(vcpu, vmcs12->guest_cr4);
6805 vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
6806
6807 /* shadow page tables on either EPT or shadow page tables */
6808 kvm_set_cr3(vcpu, vmcs12->guest_cr3);
6809 kvm_mmu_reset_context(vcpu);
6810
6811 kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->guest_rsp);
6812 kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->guest_rip);
6813}
6814
cd232ad0
NHE		 6815/*
6816 * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
6817 * for running an L2 nested guest.
6818 */
6819static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
6820{
6821 struct vmcs12 *vmcs12;
6822 struct vcpu_vmx *vmx = to_vmx(vcpu);
6823 int cpu;
6824 struct loaded_vmcs *vmcs02;
6825
6826 if (!nested_vmx_check_permission(vcpu) ||
6827 !nested_vmx_check_vmcs12(vcpu))
6828 return 1;
6829
6830 skip_emulated_instruction(vcpu);
6831 vmcs12 = get_vmcs12(vcpu);
6832
7c177938
NHE		 6833 /*
6834 * The nested entry process starts with enforcing various prerequisites
6835 * on vmcs12 as required by the Intel SDM, and act appropriately when
6836 * they fail: As the SDM explains, some conditions should cause the
6837 * instruction to fail, while others will cause the instruction to seem
6838 * to succeed, but return an EXIT_REASON_INVALID_STATE.
6839 * To speed up the normal (success) code path, we should avoid checking
6840 * for misconfigurations which will anyway be caught by the processor
6841 * when using the merged vmcs02.
6842 */
6843 if (vmcs12->launch_state == launch) {
6844 nested_vmx_failValid(vcpu,
6845 launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
6846 : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
6847 return 1;
6848 }
6849
6850 if ((vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_MSR_BITMAPS) &&
6851 !IS_ALIGNED(vmcs12->msr_bitmap, PAGE_SIZE)) {
6852 /*TODO: Also verify bits beyond physical address width are 0*/
6853 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
6854 return 1;
6855 }
6856
6857 if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) &&
6858 !IS_ALIGNED(vmcs12->apic_access_addr, PAGE_SIZE)) {
6859 /*TODO: Also verify bits beyond physical address width are 0*/
6860 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
6861 return 1;
6862 }
6863
6864 if (vmcs12->vm_entry_msr_load_count > 0 ||
6865 vmcs12->vm_exit_msr_load_count > 0 ||
6866 vmcs12->vm_exit_msr_store_count > 0) {
bd80158a
JK		 6867 pr_warn_ratelimited("%s: VMCS MSR_{LOAD,STORE} unsupported\n",
6868 __func__);
7c177938
NHE		 6869 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
6870 return 1;
6871 }
6872
6873 if (!vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
6874 nested_vmx_procbased_ctls_low, nested_vmx_procbased_ctls_high) ||
6875 !vmx_control_verify(vmcs12->secondary_vm_exec_control,
6876 nested_vmx_secondary_ctls_low, nested_vmx_secondary_ctls_high) ||
6877 !vmx_control_verify(vmcs12->pin_based_vm_exec_control,
6878 nested_vmx_pinbased_ctls_low, nested_vmx_pinbased_ctls_high) ||
6879 !vmx_control_verify(vmcs12->vm_exit_controls,
6880 nested_vmx_exit_ctls_low, nested_vmx_exit_ctls_high) ||
6881 !vmx_control_verify(vmcs12->vm_entry_controls,
6882 nested_vmx_entry_ctls_low, nested_vmx_entry_ctls_high))
6883 {
6884 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
6885 return 1;
6886 }
6887
6888 if (((vmcs12->host_cr0 & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON) ||
6889 ((vmcs12->host_cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON)) {
6890 nested_vmx_failValid(vcpu,
6891 VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
6892 return 1;
6893 }
6894
6895 if (((vmcs12->guest_cr0 & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON) ||
6896 ((vmcs12->guest_cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON)) {
6897 nested_vmx_entry_failure(vcpu, vmcs12,
6898 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
6899 return 1;
6900 }
6901 if (vmcs12->vmcs_link_pointer != -1ull) {
6902 nested_vmx_entry_failure(vcpu, vmcs12,
6903 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_VMCS_LINK_PTR);
6904 return 1;
6905 }
6906
6907 /*
6908 * We're finally done with prerequisite checking, and can start with
6909 * the nested entry.
6910 */
6911
cd232ad0
NHE		 6912 vmcs02 = nested_get_current_vmcs02(vmx);
6913 if (!vmcs02)
6914 return -ENOMEM;
6915
6916 enter_guest_mode(vcpu);
6917
6918 vmx->nested.vmcs01_tsc_offset = vmcs_read64(TSC_OFFSET);
6919
6920 cpu = get_cpu();
6921 vmx->loaded_vmcs = vmcs02;
6922 vmx_vcpu_put(vcpu);
6923 vmx_vcpu_load(vcpu, cpu);
6924 vcpu->cpu = cpu;
6925 put_cpu();
6926
6927 vmcs12->launch_state = 1;
6928
6929 prepare_vmcs02(vcpu, vmcs12);
6930
6931 /*
6932 * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
6933 * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
6934 * returned as far as L1 is concerned. It will only return (and set
6935 * the success flag) when L2 exits (see nested_vmx_vmexit()).
6936 */
6937 return 1;
6938}
6939
4704d0be
NHE		 6940/*
6941 * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
6942 * because L2 may have changed some cr0 bits directly (CRO_GUEST_HOST_MASK).
6943 * This function returns the new value we should put in vmcs12.guest_cr0.
6944 * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
6945 * 1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
6946 * available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
6947 * didn't trap the bit, because if L1 did, so would L0).
6948 * 2. Bits that L1 asked to trap (and therefore L0 also did) could not have
6949 * been modified by L2, and L1 knows it. So just leave the old value of
6950 * the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
6951 * isn't relevant, because if L0 traps this bit it can set it to anything.
6952 * 3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
6953 * changed these bits, and therefore they need to be updated, but L0
6954 * didn't necessarily allow them to be changed in GUEST_CR0 - and rather
6955 * put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
6956 */
6957static inline unsigned long
6958vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
6959{
6960 return
6961 /*1*/ (vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
6962 /*2*/ (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
6963 /*3*/ (vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
6964 vcpu->arch.cr0_guest_owned_bits));
6965}
6966
6967static inline unsigned long
6968vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
6969{
6970 return
6971 /*1*/ (vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
6972 /*2*/ (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
6973 /*3*/ (vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
6974 vcpu->arch.cr4_guest_owned_bits));
6975}
6976
6977/*
6978 * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
6979 * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
6980 * and this function updates it to reflect the changes to the guest state while
6981 * L2 was running (and perhaps made some exits which were handled directly by L0
6982 * without going back to L1), and to reflect the exit reason.
6983 * Note that we do not have to copy here all VMCS fields, just those that
6984 * could have changed by the L2 guest or the exit - i.e., the guest-state and
6985 * exit-information fields only. Other fields are modified by L1 with VMWRITE,
6986 * which already writes to vmcs12 directly.
6987 */
6988void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
6989{
6990 /* update guest state fields: */
6991 vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
6992 vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
6993
6994 kvm_get_dr(vcpu, 7, (unsigned long *)&vmcs12->guest_dr7);
6995 vmcs12->guest_rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
6996 vmcs12->guest_rip = kvm_register_read(vcpu, VCPU_REGS_RIP);
6997 vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
6998
6999 vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
7000 vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
7001 vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
7002 vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
7003 vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
7004 vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
7005 vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
7006 vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
7007 vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
7008 vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
7009 vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
7010 vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
7011 vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
7012 vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
7013 vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
7014 vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
7015 vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
7016 vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
7017 vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
7018 vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
7019 vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
7020 vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
7021 vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
7022 vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
7023 vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
7024 vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
7025 vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
7026 vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
7027 vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
7028 vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
7029 vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
7030 vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
7031 vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
7032 vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
7033 vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
7034 vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
7035
7036 vmcs12->guest_activity_state = vmcs_read32(GUEST_ACTIVITY_STATE);
7037 vmcs12->guest_interruptibility_info =
7038 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
7039 vmcs12->guest_pending_dbg_exceptions =
7040 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
7041
7042 /* TODO: These cannot have changed unless we have MSR bitmaps and
7043 * the relevant bit asks not to trap the change */
7044 vmcs12->guest_ia32_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
7045 if (vmcs12->vm_entry_controls & VM_EXIT_SAVE_IA32_PAT)
7046 vmcs12->guest_ia32_pat = vmcs_read64(GUEST_IA32_PAT);
7047 vmcs12->guest_sysenter_cs = vmcs_read32(GUEST_SYSENTER_CS);
7048 vmcs12->guest_sysenter_esp = vmcs_readl(GUEST_SYSENTER_ESP);
7049 vmcs12->guest_sysenter_eip = vmcs_readl(GUEST_SYSENTER_EIP);
7050
7051 /* update exit information fields: */
7052
7053 vmcs12->vm_exit_reason = vmcs_read32(VM_EXIT_REASON);
7054 vmcs12->exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7055
7056 vmcs12->vm_exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
7057 vmcs12->vm_exit_intr_error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
7058 vmcs12->idt_vectoring_info_field =
7059 vmcs_read32(IDT_VECTORING_INFO_FIELD);
7060 vmcs12->idt_vectoring_error_code =
7061 vmcs_read32(IDT_VECTORING_ERROR_CODE);
7062 vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
7063 vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7064
7065 /* clear vm-entry fields which are to be cleared on exit */
7066 if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY))
7067 vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
7068}
7069
7070/*
7071 * A part of what we need to when the nested L2 guest exits and we want to
7072 * run its L1 parent, is to reset L1's guest state to the host state specified
7073 * in vmcs12.
7074 * This function is to be called not only on normal nested exit, but also on
7075 * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
7076 * Failures During or After Loading Guest State").
7077 * This function should be called when the active VMCS is L1's (vmcs01).
7078 */
7079void load_vmcs12_host_state(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
7080{
7081 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
7082 vcpu->arch.efer = vmcs12->host_ia32_efer;
7083 if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
7084 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
7085 else
7086 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
7087 vmx_set_efer(vcpu, vcpu->arch.efer);
7088
7089 kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->host_rsp);
7090 kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->host_rip);
7091 /*
7092 * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
7093 * actually changed, because it depends on the current state of
7094 * fpu_active (which may have changed).
7095 * Note that vmx_set_cr0 refers to efer set above.
7096 */
7097 kvm_set_cr0(vcpu, vmcs12->host_cr0);
7098 /*
7099 * If we did fpu_activate()/fpu_deactivate() during L2's run, we need
7100 * to apply the same changes to L1's vmcs. We just set cr0 correctly,
7101 * but we also need to update cr0_guest_host_mask and exception_bitmap.
7102 */
7103 update_exception_bitmap(vcpu);
7104 vcpu->arch.cr0_guest_owned_bits = (vcpu->fpu_active ? X86_CR0_TS : 0);
7105 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
7106
7107 /*
7108 * Note that CR4_GUEST_HOST_MASK is already set in the original vmcs01
7109 * (KVM doesn't change it)- no reason to call set_cr4_guest_host_mask();
7110 */
7111 vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
7112 kvm_set_cr4(vcpu, vmcs12->host_cr4);
7113
7114 /* shadow page tables on either EPT or shadow page tables */
7115 kvm_set_cr3(vcpu, vmcs12->host_cr3);
7116 kvm_mmu_reset_context(vcpu);
7117
7118 if (enable_vpid) {
7119 /*
7120 * Trivially support vpid by letting L2s share their parent
7121 * L1's vpid. TODO: move to a more elaborate solution, giving
7122 * each L2 its own vpid and exposing the vpid feature to L1.
7123 */
7124 vmx_flush_tlb(vcpu);
7125 }
7126
7127
7128 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
7129 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
7130 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
7131 vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
7132 vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
7133 vmcs_writel(GUEST_TR_BASE, vmcs12->host_tr_base);
7134 vmcs_writel(GUEST_GS_BASE, vmcs12->host_gs_base);
7135 vmcs_writel(GUEST_FS_BASE, vmcs12->host_fs_base);
7136 vmcs_write16(GUEST_ES_SELECTOR, vmcs12->host_es_selector);
7137 vmcs_write16(GUEST_CS_SELECTOR, vmcs12->host_cs_selector);
7138 vmcs_write16(GUEST_SS_SELECTOR, vmcs12->host_ss_selector);
7139 vmcs_write16(GUEST_DS_SELECTOR, vmcs12->host_ds_selector);
7140 vmcs_write16(GUEST_FS_SELECTOR, vmcs12->host_fs_selector);
7141 vmcs_write16(GUEST_GS_SELECTOR, vmcs12->host_gs_selector);
7142 vmcs_write16(GUEST_TR_SELECTOR, vmcs12->host_tr_selector);
7143
7144 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT)
7145 vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
7146 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
7147 vmcs_write64(GUEST_IA32_PERF_GLOBAL_CTRL,
7148 vmcs12->host_ia32_perf_global_ctrl);
7149}
7150
7151/*
7152 * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
7153 * and modify vmcs12 to make it see what it would expect to see there if
7154 * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
7155 */
7156static void nested_vmx_vmexit(struct kvm_vcpu *vcpu)
7157{
7158 struct vcpu_vmx *vmx = to_vmx(vcpu);
7159 int cpu;
7160 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7161
7162 leave_guest_mode(vcpu);
7163 prepare_vmcs12(vcpu, vmcs12);
7164
7165 cpu = get_cpu();
7166 vmx->loaded_vmcs = &vmx->vmcs01;
7167 vmx_vcpu_put(vcpu);
7168 vmx_vcpu_load(vcpu, cpu);
7169 vcpu->cpu = cpu;
7170 put_cpu();
7171
7172 /* if no vmcs02 cache requested, remove the one we used */
7173 if (VMCS02_POOL_SIZE == 0)
7174 nested_free_vmcs02(vmx, vmx->nested.current_vmptr);
7175
7176 load_vmcs12_host_state(vcpu, vmcs12);
7177
27fc51b2 7178 /* Update TSC_OFFSET if TSC was changed while L2 ran */
4704d0be
NHE		 7179 vmcs_write64(TSC_OFFSET, vmx->nested.vmcs01_tsc_offset);
7180
7181 /* This is needed for same reason as it was needed in prepare_vmcs02 */
7182 vmx->host_rsp = 0;
7183
7184 /* Unpin physical memory we referred to in vmcs02 */
7185 if (vmx->nested.apic_access_page) {
7186 nested_release_page(vmx->nested.apic_access_page);
7187 vmx->nested.apic_access_page = 0;
7188 }
7189
7190 /*
7191 * Exiting from L2 to L1, we're now back to L1 which thinks it just
7192 * finished a VMLAUNCH or VMRESUME instruction, so we need to set the
7193 * success or failure flag accordingly.
7194 */
7195 if (unlikely(vmx->fail)) {
7196 vmx->fail = 0;
7197 nested_vmx_failValid(vcpu, vmcs_read32(VM_INSTRUCTION_ERROR));
7198 } else
7199 nested_vmx_succeed(vcpu);
7200}
7201
7c177938
NHE		 7202/*
7203 * L1's failure to enter L2 is a subset of a normal exit, as explained in
7204 * 23.7 "VM-entry failures during or after loading guest state" (this also
7205 * lists the acceptable exit-reason and exit-qualification parameters).
7206 * It should only be called before L2 actually succeeded to run, and when
7207 * vmcs01 is current (it doesn't leave_guest_mode() or switch vmcss).
7208 */
7209static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
7210 struct vmcs12 *vmcs12,
7211 u32 reason, unsigned long qualification)
7212{
7213 load_vmcs12_host_state(vcpu, vmcs12);
7214 vmcs12->vm_exit_reason = reason | VMX_EXIT_REASONS_FAILED_VMENTRY;
7215 vmcs12->exit_qualification = qualification;
7216 nested_vmx_succeed(vcpu);
7217}
7218
8a76d7f2
JR		 7219static int vmx_check_intercept(struct kvm_vcpu *vcpu,
7220 struct x86_instruction_info *info,
7221 enum x86_intercept_stage stage)
7222{
7223 return X86EMUL_CONTINUE;
7224}
7225
cbdd1bea 7226static struct kvm_x86_ops vmx_x86_ops = {
6aa8b732
AK		 7227 .cpu_has_kvm_support = cpu_has_kvm_support,
7228 .disabled_by_bios = vmx_disabled_by_bios,
7229 .hardware_setup = hardware_setup,
7230 .hardware_unsetup = hardware_unsetup,
002c7f7c 7231 .check_processor_compatibility = vmx_check_processor_compat,
6aa8b732
AK		 7232 .hardware_enable = hardware_enable,
7233 .hardware_disable = hardware_disable,
04547156 7234 .cpu_has_accelerated_tpr = report_flexpriority,
6aa8b732
AK		 7235
7236 .vcpu_create = vmx_create_vcpu,
7237 .vcpu_free = vmx_free_vcpu,
04d2cc77 7238 .vcpu_reset = vmx_vcpu_reset,
6aa8b732 7239
04d2cc77 7240 .prepare_guest_switch = vmx_save_host_state,
6aa8b732
AK		 7241 .vcpu_load = vmx_vcpu_load,
7242 .vcpu_put = vmx_vcpu_put,
7243
7244 .set_guest_debug = set_guest_debug,
7245 .get_msr = vmx_get_msr,
7246 .set_msr = vmx_set_msr,
7247 .get_segment_base = vmx_get_segment_base,
7248 .get_segment = vmx_get_segment,
7249 .set_segment = vmx_set_segment,
2e4d2653 7250 .get_cpl = vmx_get_cpl,
6aa8b732 7251 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
e8467fda 7252 .decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
aff48baa 7253 .decache_cr3 = vmx_decache_cr3,
25c4c276 7254 .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
6aa8b732 7255 .set_cr0 = vmx_set_cr0,
6aa8b732
AK		 7256 .set_cr3 = vmx_set_cr3,
7257 .set_cr4 = vmx_set_cr4,
6aa8b732 7258 .set_efer = vmx_set_efer,
6aa8b732
AK		 7259 .get_idt = vmx_get_idt,
7260 .set_idt = vmx_set_idt,
7261 .get_gdt = vmx_get_gdt,
7262 .set_gdt = vmx_set_gdt,
020df079 7263 .set_dr7 = vmx_set_dr7,
5fdbf976 7264 .cache_reg = vmx_cache_reg,
6aa8b732
AK		 7265 .get_rflags = vmx_get_rflags,
7266 .set_rflags = vmx_set_rflags,
ebcbab4c 7267 .fpu_activate = vmx_fpu_activate,
02daab21 7268 .fpu_deactivate = vmx_fpu_deactivate,
6aa8b732
AK		 7269
7270 .tlb_flush = vmx_flush_tlb,
6aa8b732 7271
6aa8b732 7272 .run = vmx_vcpu_run,
6062d012 7273 .handle_exit = vmx_handle_exit,
6aa8b732 7274 .skip_emulated_instruction = skip_emulated_instruction,
2809f5d2
GC		 7275 .set_interrupt_shadow = vmx_set_interrupt_shadow,
7276 .get_interrupt_shadow = vmx_get_interrupt_shadow,
102d8325 7277 .patch_hypercall = vmx_patch_hypercall,
2a8067f1 7278 .set_irq = vmx_inject_irq,
95ba8273 7279 .set_nmi = vmx_inject_nmi,
298101da 7280 .queue_exception = vmx_queue_exception,
b463a6f7 7281 .cancel_injection = vmx_cancel_injection,
78646121 7282 .interrupt_allowed = vmx_interrupt_allowed,
95ba8273 7283 .nmi_allowed = vmx_nmi_allowed,
3cfc3092
JK		 7284 .get_nmi_mask = vmx_get_nmi_mask,
7285 .set_nmi_mask = vmx_set_nmi_mask,
95ba8273
GN		 7286 .enable_nmi_window = enable_nmi_window,
7287 .enable_irq_window = enable_irq_window,
7288 .update_cr8_intercept = update_cr8_intercept,
95ba8273 7289
cbc94022 7290 .set_tss_addr = vmx_set_tss_addr,
67253af5 7291 .get_tdp_level = get_ept_level,
4b12f0de 7292 .get_mt_mask = vmx_get_mt_mask,
229456fc 7293
586f9607 7294 .get_exit_info = vmx_get_exit_info,
586f9607 7295
17cc3935 7296 .get_lpage_level = vmx_get_lpage_level,
0e851880
SY		 7297
7298 .cpuid_update = vmx_cpuid_update,
4e47c7a6
SY		 7299
7300 .rdtscp_supported = vmx_rdtscp_supported,
ad756a16 7301 .invpcid_supported = vmx_invpcid_supported,
d4330ef2
JR		 7302
7303 .set_supported_cpuid = vmx_set_supported_cpuid,
f5f48ee1
SY		 7304
7305 .has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
99e3e30a 7306
4051b188 7307 .set_tsc_khz = vmx_set_tsc_khz,
99e3e30a 7308 .write_tsc_offset = vmx_write_tsc_offset,
e48672fa 7309 .adjust_tsc_offset = vmx_adjust_tsc_offset,
857e4099 7310 .compute_tsc_offset = vmx_compute_tsc_offset,
d5c1785d 7311 .read_l1_tsc = vmx_read_l1_tsc,
1c97f0a0
JR		 7312
7313 .set_tdp_cr3 = vmx_set_cr3,
8a76d7f2
JR		 7314
7315 .check_intercept = vmx_check_intercept,
6aa8b732
AK		 7316};
7317
7318static int __init vmx_init(void)
7319{
26bb0981
AK		 7320 int r, i;
7321
7322 rdmsrl_safe(MSR_EFER, &host_efer);
7323
7324 for (i = 0; i < NR_VMX_MSR; ++i)
7325 kvm_define_shared_msr(i, vmx_msr_index[i]);
fdef3ad1 7326
3e7c73e9 7327 vmx_io_bitmap_a = (unsigned long *)__get_free_page(GFP_KERNEL);
fdef3ad1
HQ		 7328 if (!vmx_io_bitmap_a)
7329 return -ENOMEM;
7330
2106a548
GC		 7331 r = -ENOMEM;
7332
3e7c73e9 7333 vmx_io_bitmap_b = (unsigned long *)__get_free_page(GFP_KERNEL);
2106a548 7334 if (!vmx_io_bitmap_b)
fdef3ad1 7335 goto out;
fdef3ad1 7336
5897297b 7337 vmx_msr_bitmap_legacy = (unsigned long *)__get_free_page(GFP_KERNEL);
2106a548 7338 if (!vmx_msr_bitmap_legacy)
25c5f225 7339 goto out1;
2106a548 7340
25c5f225 7341
5897297b 7342 vmx_msr_bitmap_longmode = (unsigned long *)__get_free_page(GFP_KERNEL);
2106a548 7343 if (!vmx_msr_bitmap_longmode)
5897297b 7344 goto out2;
2106a548 7345
5897297b 7346
fdef3ad1
HQ		 7347 /*
7348 * Allow direct access to the PC debug port (it is often used for I/O
7349 * delays, but the vmexits simply slow things down).
7350 */
3e7c73e9
AK		 7351 memset(vmx_io_bitmap_a, 0xff, PAGE_SIZE);
7352 clear_bit(0x80, vmx_io_bitmap_a);
fdef3ad1 7353
3e7c73e9 7354 memset(vmx_io_bitmap_b, 0xff, PAGE_SIZE);
fdef3ad1 7355
5897297b
AK		 7356 memset(vmx_msr_bitmap_legacy, 0xff, PAGE_SIZE);
7357 memset(vmx_msr_bitmap_longmode, 0xff, PAGE_SIZE);
25c5f225 7358
2384d2b3
SY		 7359 set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
7360
0ee75bea
AK		 7361 r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
7362 __alignof__(struct vcpu_vmx), THIS_MODULE);
fdef3ad1 7363 if (r)
5897297b 7364 goto out3;
25c5f225 7365
5897297b
AK		 7366 vmx_disable_intercept_for_msr(MSR_FS_BASE, false);
7367 vmx_disable_intercept_for_msr(MSR_GS_BASE, false);
7368 vmx_disable_intercept_for_msr(MSR_KERNEL_GS_BASE, true);
7369 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_CS, false);
7370 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_ESP, false);
7371 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_EIP, false);
fdef3ad1 7372
089d034e 7373 if (enable_ept) {
3f6d8c8a
XH		 7374 kvm_mmu_set_mask_ptes(0ull,
7375 (enable_ept_ad_bits) ? VMX_EPT_ACCESS_BIT : 0ull,
7376 (enable_ept_ad_bits) ? VMX_EPT_DIRTY_BIT : 0ull,
7377 0ull, VMX_EPT_EXECUTABLE_MASK);
ce88decf 7378 ept_set_mmio_spte_mask();
5fdbcb9d
SY		 7379 kvm_enable_tdp();
7380 } else
7381 kvm_disable_tdp();
1439442c 7382
fdef3ad1
HQ		 7383 return 0;
7384
5897297b
AK		 7385out3:
7386 free_page((unsigned long)vmx_msr_bitmap_longmode);
25c5f225 7387out2:
5897297b 7388 free_page((unsigned long)vmx_msr_bitmap_legacy);
fdef3ad1 7389out1:
3e7c73e9 7390 free_page((unsigned long)vmx_io_bitmap_b);
fdef3ad1 7391out:
3e7c73e9 7392 free_page((unsigned long)vmx_io_bitmap_a);
fdef3ad1 7393 return r;
6aa8b732
AK		 7394}
7395
7396static void __exit vmx_exit(void)
7397{
5897297b
AK		 7398 free_page((unsigned long)vmx_msr_bitmap_legacy);
7399 free_page((unsigned long)vmx_msr_bitmap_longmode);
3e7c73e9
AK		 7400 free_page((unsigned long)vmx_io_bitmap_b);
7401 free_page((unsigned long)vmx_io_bitmap_a);
fdef3ad1 7402
cb498ea2 7403 kvm_exit();
6aa8b732
AK		 7404}
7405
7406module_init(vmx_init)
7407module_exit(vmx_exit)
Ubuntu Artful kernel mirror