blob: c8f5e87615d54db88d567596b70cc6d0a69b87df [file] [log] [blame]
Suravee Suthikulpanit44a95da2016-05-04 14:09:46 -05001#define pr_fmt(fmt) "SVM: " fmt
2
Avi Kivityedf88412007-12-16 11:02:48 +02003#include <linux/kvm_host.h>
4
Eddie Dong85f455f2007-07-06 12:20:49 +03005#include "irq.h"
Zhang Xiantao1d737c82007-12-14 09:35:10 +08006#include "mmu.h"
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03007#include "kvm_cache_regs.h"
Gleb Natapovfe4c7b12009-03-23 11:23:18 +02008#include "x86.h"
Julian Stecklina66f7b722012-12-05 15:26:19 +01009#include "cpuid.h"
Wei Huang25462f72015-06-19 15:45:05 +020010#include "pmu.h"
Avi Kivitye4956062007-06-28 14:15:57 -040011
Avi Kivity6aa8b732006-12-10 02:21:36 -080012#include <linux/module.h>
Josh Triplettae759542012-03-28 11:32:28 -070013#include <linux/mod_devicetable.h>
Ahmed S. Darwish9d8f5492007-02-19 14:37:46 +020014#include <linux/kernel.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080015#include <linux/vmalloc.h>
16#include <linux/highmem.h>
Joerg Roedelef0f6492020-03-31 12:17:38 -040017#include <linux/amd-iommu.h>
Alexey Dobriyane8edc6e2007-05-21 01:22:52 +040018#include <linux/sched.h>
Steven Rostedt (Red Hat)af658dc2015-04-29 14:36:05 -040019#include <linux/trace_events.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090020#include <linux/slab.h>
Suravee Suthikulpanit5881f732016-08-23 13:52:42 -050021#include <linux/hashtable.h>
Josh Poimboeufc207aee2017-06-28 10:11:06 -050022#include <linux/frame.h>
Brijesh Singhe9df0942017-12-04 10:57:33 -060023#include <linux/psp-sev.h>
Brijesh Singh1654efc2017-12-04 10:57:34 -060024#include <linux/file.h>
Brijesh Singh89c50582017-12-04 10:57:35 -060025#include <linux/pagemap.h>
26#include <linux/swap.h>
Tom Lendacky33af3a72019-10-03 21:17:48 +000027#include <linux/rwsem.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080028
Suravee Suthikulpanit8221c132016-05-04 14:09:52 -050029#include <asm/apic.h>
Joerg Roedel1018faa2012-02-29 14:57:32 +010030#include <asm/perf_event.h>
Joerg Roedel67ec6602010-05-17 14:43:35 +020031#include <asm/tlbflush.h>
Avi Kivitye4956062007-06-28 14:15:57 -040032#include <asm/desc.h>
Paolo Bonzinifacb0132014-02-21 10:32:27 +010033#include <asm/debugreg.h>
Gleb Natapov631bc482010-10-14 11:22:52 +020034#include <asm/kvm_para.h>
Suravee Suthikulpanit411b44b2016-08-23 13:52:43 -050035#include <asm/irq_remapping.h>
Uros Bizjak1c164cb2020-04-11 17:36:27 +020036#include <asm/mce.h>
Thomas Gleixner28a27752018-04-29 15:01:37 +020037#include <asm/spec-ctrl.h>
Thomas Gleixnerba5bade2020-03-20 14:13:46 +010038#include <asm/cpu_device_id.h>
Avi Kivity6aa8b732006-12-10 02:21:36 -080039
Eduardo Habkost63d11422008-11-17 19:03:20 -020040#include <asm/virtext.h>
Marcelo Tosatti229456f2009-06-17 09:22:14 -030041#include "trace.h"
Eduardo Habkost63d11422008-11-17 19:03:20 -020042
Joerg Roedel883b0a92020-03-24 10:41:52 +010043#include "svm.h"
44
Avi Kivity4ecac3f2008-05-13 13:23:38 +030045#define __ex(x) __kvm_handle_fault_on_reboot(x)
46
Avi Kivity6aa8b732006-12-10 02:21:36 -080047MODULE_AUTHOR("Qumranet");
48MODULE_LICENSE("GPL");
49
Valdis Klētnieks575b2552020-02-27 21:49:52 -050050#ifdef MODULE
Josh Triplettae759542012-03-28 11:32:28 -070051static const struct x86_cpu_id svm_cpu_id[] = {
Thomas Gleixner320debe2020-03-20 14:13:50 +010052 X86_MATCH_FEATURE(X86_FEATURE_SVM, NULL),
Josh Triplettae759542012-03-28 11:32:28 -070053 {}
54};
55MODULE_DEVICE_TABLE(x86cpu, svm_cpu_id);
Valdis Klētnieks575b2552020-02-27 21:49:52 -050056#endif
Josh Triplettae759542012-03-28 11:32:28 -070057
Avi Kivity6aa8b732006-12-10 02:21:36 -080058#define IOPM_ALLOC_ORDER 2
59#define MSRPM_ALLOC_ORDER 1
60
Avi Kivity6aa8b732006-12-10 02:21:36 -080061#define SEG_TYPE_LDT 2
62#define SEG_TYPE_BUSY_TSS16 3
63
Andre Przywara6bc31bd2010-04-11 23:07:28 +020064#define SVM_FEATURE_LBRV (1 << 1)
65#define SVM_FEATURE_SVML (1 << 2)
Andre Przywaraddce97a2010-12-21 11:12:03 +010066#define SVM_FEATURE_TSC_RATE (1 << 4)
67#define SVM_FEATURE_VMCB_CLEAN (1 << 5)
68#define SVM_FEATURE_FLUSH_ASID (1 << 6)
69#define SVM_FEATURE_DECODE_ASSIST (1 << 7)
Andre Przywara6bc31bd2010-04-11 23:07:28 +020070#define SVM_FEATURE_PAUSE_FILTER (1 << 10)
Joerg Roedel80b77062007-03-30 17:02:14 +030071
Joerg Roedel24e09cb2008-02-13 18:58:47 +010072#define DEBUGCTL_RESERVED_BITS (~(0x3fULL))
73
Joerg Roedelfbc0db72011-03-25 09:44:46 +010074#define TSC_RATIO_RSVD 0xffffff0000000000ULL
Joerg Roedel92a1f122011-03-25 09:44:51 +010075#define TSC_RATIO_MIN 0x0000000000000001ULL
76#define TSC_RATIO_MAX 0x000000ffffffffffULL
Joerg Roedelfbc0db72011-03-25 09:44:46 +010077
Joerg Roedel67ec6602010-05-17 14:43:35 +020078static bool erratum_383_found __read_mostly;
79
Joerg Roedel883b0a92020-03-24 10:41:52 +010080u32 msrpm_offsets[MSRPM_OFFSETS] __read_mostly;
Joerg Roedel323c3d82010-03-01 15:34:37 +010081
Boris Ostrovsky2b036c62012-01-09 14:00:35 -050082/*
83 * Set osvw_len to higher value when updated Revision Guides
84 * are published and we know what the new status bits are
85 */
86static uint64_t osvw_len = 4, osvw_status;
87
Joerg Roedelfbc0db72011-03-25 09:44:46 +010088static DEFINE_PER_CPU(u64, current_tsc_ratio);
89#define TSC_RATIO_DEFAULT 0x0100000000ULL
90
Mathias Krause09941fb2012-08-30 01:30:20 +020091static const struct svm_direct_access_msrs {
Joerg Roedelac72a9b2010-03-01 15:34:36 +010092 u32 index; /* Index of the MSR */
93 bool always; /* True if intercept is always on */
94} direct_access_msrs[] = {
Brian Gerst8c065852010-07-17 09:03:26 -040095 { .index = MSR_STAR, .always = true },
Joerg Roedelac72a9b2010-03-01 15:34:36 +010096 { .index = MSR_IA32_SYSENTER_CS, .always = true },
97#ifdef CONFIG_X86_64
98 { .index = MSR_GS_BASE, .always = true },
99 { .index = MSR_FS_BASE, .always = true },
100 { .index = MSR_KERNEL_GS_BASE, .always = true },
101 { .index = MSR_LSTAR, .always = true },
102 { .index = MSR_CSTAR, .always = true },
103 { .index = MSR_SYSCALL_MASK, .always = true },
104#endif
KarimAllah Ahmedb2ac58f2018-02-03 15:56:23 +0100105 { .index = MSR_IA32_SPEC_CTRL, .always = false },
Ashok Raj15d45072018-02-01 22:59:43 +0100106 { .index = MSR_IA32_PRED_CMD, .always = false },
Joerg Roedelac72a9b2010-03-01 15:34:36 +0100107 { .index = MSR_IA32_LASTBRANCHFROMIP, .always = false },
108 { .index = MSR_IA32_LASTBRANCHTOIP, .always = false },
109 { .index = MSR_IA32_LASTINTFROMIP, .always = false },
110 { .index = MSR_IA32_LASTINTTOIP, .always = false },
111 { .index = MSR_INVALID, .always = false },
Avi Kivity6c8166a2009-05-31 18:15:37 +0300112};
113
Joerg Roedel709ddeb2008-02-07 13:47:45 +0100114/* enable NPT for AMD64 and X86 with PAE */
115#if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
Joerg Roedel883b0a92020-03-24 10:41:52 +0100116bool npt_enabled = true;
Joerg Roedel709ddeb2008-02-07 13:47:45 +0100117#else
Joerg Roedel883b0a92020-03-24 10:41:52 +0100118bool npt_enabled;
Joerg Roedel709ddeb2008-02-07 13:47:45 +0100119#endif
Joerg Roedel6c7dac72008-02-07 13:47:40 +0100120
Babu Moger8566ac82018-03-16 16:37:26 -0400121/*
122 * These 2 parameters are used to config the controls for Pause-Loop Exiting:
123 * pause_filter_count: On processors that support Pause filtering(indicated
124 * by CPUID Fn8000_000A_EDX), the VMCB provides a 16 bit pause filter
125 * count value. On VMRUN this value is loaded into an internal counter.
126 * Each time a pause instruction is executed, this counter is decremented
127 * until it reaches zero at which time a #VMEXIT is generated if pause
128 * intercept is enabled. Refer to AMD APM Vol 2 Section 15.14.4 Pause
129 * Intercept Filtering for more details.
130 * This also indicate if ple logic enabled.
131 *
132 * pause_filter_thresh: In addition, some processor families support advanced
133 * pause filtering (indicated by CPUID Fn8000_000A_EDX) upper bound on
134 * the amount of time a guest is allowed to execute in a pause loop.
135 * In this mode, a 16-bit pause filter threshold field is added in the
136 * VMCB. The threshold value is a cycle count that is used to reset the
137 * pause counter. As with simple pause filtering, VMRUN loads the pause
138 * count value from VMCB into an internal counter. Then, on each pause
139 * instruction the hardware checks the elapsed number of cycles since
140 * the most recent pause instruction against the pause filter threshold.
141 * If the elapsed cycle count is greater than the pause filter threshold,
142 * then the internal pause count is reloaded from the VMCB and execution
143 * continues. If the elapsed cycle count is less than the pause filter
144 * threshold, then the internal pause count is decremented. If the count
145 * value is less than zero and PAUSE intercept is enabled, a #VMEXIT is
146 * triggered. If advanced pause filtering is supported and pause filter
147 * threshold field is set to zero, the filter will operate in the simpler,
148 * count only mode.
149 */
150
151static unsigned short pause_filter_thresh = KVM_DEFAULT_PLE_GAP;
152module_param(pause_filter_thresh, ushort, 0444);
153
154static unsigned short pause_filter_count = KVM_SVM_DEFAULT_PLE_WINDOW;
155module_param(pause_filter_count, ushort, 0444);
156
157/* Default doubles per-vcpu window every exit. */
158static unsigned short pause_filter_count_grow = KVM_DEFAULT_PLE_WINDOW_GROW;
159module_param(pause_filter_count_grow, ushort, 0444);
160
161/* Default resets per-vcpu window every exit to pause_filter_count. */
162static unsigned short pause_filter_count_shrink = KVM_DEFAULT_PLE_WINDOW_SHRINK;
163module_param(pause_filter_count_shrink, ushort, 0444);
164
165/* Default is to compute the maximum so we can never overflow. */
166static unsigned short pause_filter_count_max = KVM_SVM_DEFAULT_PLE_WINDOW_MAX;
167module_param(pause_filter_count_max, ushort, 0444);
168
Davidlohr Buesoe2358852012-01-17 14:09:50 +0100169/* allow nested paging (virtualized MMU) for all guests */
170static int npt = true;
Joerg Roedel6c7dac72008-02-07 13:47:40 +0100171module_param(npt, int, S_IRUGO);
Joerg Roedele3da3ac2008-02-07 13:47:39 +0100172
Davidlohr Buesoe2358852012-01-17 14:09:50 +0100173/* allow nested virtualization in KVM/SVM */
174static int nested = true;
Alexander Graf236de052008-11-25 20:17:10 +0100175module_param(nested, int, S_IRUGO);
176
Paolo Bonzinid647eb62019-06-20 14:13:33 +0200177/* enable/disable Next RIP Save */
178static int nrips = true;
179module_param(nrips, int, 0444);
180
Janakarajan Natarajan89c8a492017-07-06 15:50:47 -0500181/* enable/disable Virtual VMLOAD VMSAVE */
182static int vls = true;
183module_param(vls, int, 0444);
184
Janakarajan Natarajan640bd6e2017-08-23 09:57:19 -0500185/* enable/disable Virtual GIF */
186static int vgif = true;
187module_param(vgif, int, 0444);
Suravee Suthikulpanit5ea11f22016-08-23 13:52:41 -0500188
Brijesh Singhe9df0942017-12-04 10:57:33 -0600189/* enable/disable SEV support */
190static int sev = IS_ENABLED(CONFIG_AMD_MEM_ENCRYPT_ACTIVE_BY_DEFAULT);
191module_param(sev, int, 0444);
192
Paolo Bonzini6f2f8452019-05-20 15:34:35 +0200193static bool __read_mostly dump_invalid_vmcb = 0;
194module_param(dump_invalid_vmcb, bool, 0644);
195
Brijesh Singh7607b712018-02-19 10:14:44 -0600196static u8 rsm_ins_bytes[] = "\x0f\xaa";
197
Joerg Roedela5c38322009-08-07 11:49:32 +0200198static void svm_complete_interrupts(struct vcpu_svm *svm);
Suravee Suthikulpanit44a95da2016-05-04 14:09:46 -0500199
Harvey Harrison4866d5e2008-02-19 10:32:02 -0800200static unsigned long iopm_base;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800201
202struct kvm_ldttss_desc {
203 u16 limit0;
204 u16 base0;
Joerg Roedele0231712010-02-24 18:59:10 +0100205 unsigned base1:8, type:5, dpl:2, p:1;
206 unsigned limit1:4, zero0:3, g:1, base2:8;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800207 u32 base3;
208 u32 zero1;
209} __attribute__((packed));
210
Joerg Roedeleaf78262020-03-24 10:41:54 +0100211DEFINE_PER_CPU(struct svm_cpu_data *, svm_data);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800212
Mathias Krause09941fb2012-08-30 01:30:20 +0200213static const u32 msrpm_ranges[] = {0, 0xc0000000, 0xc0010000};
Avi Kivity6aa8b732006-12-10 02:21:36 -0800214
Ahmed S. Darwish9d8f5492007-02-19 14:37:46 +0200215#define NUM_MSR_MAPS ARRAY_SIZE(msrpm_ranges)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800216#define MSRS_RANGE_SIZE 2048
217#define MSRS_IN_RANGE (MSRS_RANGE_SIZE * 8 / 2)
218
Joerg Roedel883b0a92020-03-24 10:41:52 +0100219u32 svm_msrpm_offset(u32 msr)
Joerg Roedel455716f2010-03-01 15:34:35 +0100220{
221 u32 offset;
222 int i;
223
224 for (i = 0; i < NUM_MSR_MAPS; i++) {
225 if (msr < msrpm_ranges[i] ||
226 msr >= msrpm_ranges[i] + MSRS_IN_RANGE)
227 continue;
228
229 offset = (msr - msrpm_ranges[i]) / 4; /* 4 msrs per u8 */
230 offset += (i * MSRS_RANGE_SIZE); /* add range offset */
231
232 /* Now we have the u8 offset - but need the u32 offset */
233 return offset / 4;
234 }
235
236 /* MSR not in any range */
237 return MSR_INVALID;
238}
239
Avi Kivity6aa8b732006-12-10 02:21:36 -0800240#define MAX_INST_SIZE 15
241
Avi Kivity6aa8b732006-12-10 02:21:36 -0800242static inline void clgi(void)
243{
Uros Bizjakac5ffda22018-11-26 17:00:08 +0100244 asm volatile (__ex("clgi"));
Avi Kivity6aa8b732006-12-10 02:21:36 -0800245}
246
247static inline void stgi(void)
248{
Uros Bizjakac5ffda22018-11-26 17:00:08 +0100249 asm volatile (__ex("stgi"));
Avi Kivity6aa8b732006-12-10 02:21:36 -0800250}
251
252static inline void invlpga(unsigned long addr, u32 asid)
253{
Uros Bizjakac5ffda22018-11-26 17:00:08 +0100254 asm volatile (__ex("invlpga %1, %0") : : "c"(asid), "a"(addr));
Avi Kivity6aa8b732006-12-10 02:21:36 -0800255}
256
Yu Zhang855feb62017-08-24 20:27:55 +0800257static int get_npt_level(struct kvm_vcpu *vcpu)
Joerg Roedel4b161842010-09-10 17:31:03 +0200258{
259#ifdef CONFIG_X86_64
Yu Zhang2a7266a2017-08-24 20:27:54 +0800260 return PT64_ROOT_4LEVEL;
Joerg Roedel4b161842010-09-10 17:31:03 +0200261#else
262 return PT32E_ROOT_LEVEL;
263#endif
264}
265
Joerg Roedel883b0a92020-03-24 10:41:52 +0100266void svm_set_efer(struct kvm_vcpu *vcpu, u64 efer)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800267{
Paolo Bonzinic513f482020-05-18 13:08:37 -0400268 struct vcpu_svm *svm = to_svm(vcpu);
Zachary Amsden6dc696d2010-05-26 15:09:43 -1000269 vcpu->arch.efer = efer;
Paolo Bonzini9167ab72019-10-27 16:23:23 +0100270
271 if (!npt_enabled) {
272 /* Shadow paging assumes NX to be available. */
273 efer |= EFER_NX;
274
275 if (!(efer & EFER_LMA))
276 efer &= ~EFER_LME;
277 }
Avi Kivity6aa8b732006-12-10 02:21:36 -0800278
Paolo Bonzinic513f482020-05-18 13:08:37 -0400279 if (!(efer & EFER_SVME)) {
280 svm_leave_nested(svm);
281 svm_set_gif(svm, true);
282 }
283
284 svm->vmcb->save.efer = efer | EFER_SVME;
285 mark_dirty(svm->vmcb, VMCB_CR);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800286}
287
Avi Kivity6aa8b732006-12-10 02:21:36 -0800288static int is_external_interrupt(u32 info)
289{
290 info &= SVM_EVTINJ_TYPE_MASK | SVM_EVTINJ_VALID;
291 return info == (SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR);
292}
293
Paolo Bonzini37ccdcb2014-05-20 14:29:47 +0200294static u32 svm_get_interrupt_shadow(struct kvm_vcpu *vcpu)
Glauber Costa2809f5d2009-05-12 16:21:05 -0400295{
296 struct vcpu_svm *svm = to_svm(vcpu);
297 u32 ret = 0;
298
299 if (svm->vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK)
Paolo Bonzini37ccdcb2014-05-20 14:29:47 +0200300 ret = KVM_X86_SHADOW_INT_STI | KVM_X86_SHADOW_INT_MOV_SS;
301 return ret;
Glauber Costa2809f5d2009-05-12 16:21:05 -0400302}
303
304static void svm_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
305{
306 struct vcpu_svm *svm = to_svm(vcpu);
307
308 if (mask == 0)
309 svm->vmcb->control.int_state &= ~SVM_INTERRUPT_SHADOW_MASK;
310 else
311 svm->vmcb->control.int_state |= SVM_INTERRUPT_SHADOW_MASK;
312
313}
314
Vitaly Kuznetsovf8ea7c62019-08-13 15:53:30 +0200315static int skip_emulated_instruction(struct kvm_vcpu *vcpu)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800316{
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -0400317 struct vcpu_svm *svm = to_svm(vcpu);
318
Paolo Bonzinid647eb62019-06-20 14:13:33 +0200319 if (nrips && svm->vmcb->control.next_rip != 0) {
Dirk Müllerd2922422015-10-01 13:43:42 +0200320 WARN_ON_ONCE(!static_cpu_has(X86_FEATURE_NRIPS));
Andre Przywara6bc31bd2010-04-11 23:07:28 +0200321 svm->next_rip = svm->vmcb->control.next_rip;
Bandan Dasf1047652015-06-11 02:05:33 -0400322 }
Andre Przywara6bc31bd2010-04-11 23:07:28 +0200323
Sean Christopherson1957aa62019-08-27 14:40:39 -0700324 if (!svm->next_rip) {
325 if (!kvm_emulate_instruction(vcpu, EMULTYPE_SKIP))
326 return 0;
327 } else {
Sean Christopherson1957aa62019-08-27 14:40:39 -0700328 kvm_rip_write(vcpu, svm->next_rip);
329 }
Glauber Costa2809f5d2009-05-12 16:21:05 -0400330 svm_set_interrupt_shadow(vcpu, 0);
Vitaly Kuznetsovf8ea7c62019-08-13 15:53:30 +0200331
Sean Christopherson60fc3d02019-08-27 14:40:38 -0700332 return 1;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800333}
334
Wanpeng Licfcd20e2017-07-13 18:30:39 -0700335static void svm_queue_exception(struct kvm_vcpu *vcpu)
Jan Kiszka116a4752010-02-23 17:47:54 +0100336{
337 struct vcpu_svm *svm = to_svm(vcpu);
Wanpeng Licfcd20e2017-07-13 18:30:39 -0700338 unsigned nr = vcpu->arch.exception.nr;
339 bool has_error_code = vcpu->arch.exception.has_error_code;
Wanpeng Licfcd20e2017-07-13 18:30:39 -0700340 u32 error_code = vcpu->arch.exception.error_code;
Jan Kiszka116a4752010-02-23 17:47:54 +0100341
Jim Mattsonda998b42018-10-16 14:29:22 -0700342 kvm_deliver_exception_payload(&svm->vcpu);
343
Paolo Bonzinid647eb62019-06-20 14:13:33 +0200344 if (nr == BP_VECTOR && !nrips) {
Jan Kiszka66b71382010-02-23 17:47:56 +0100345 unsigned long rip, old_rip = kvm_rip_read(&svm->vcpu);
346
347 /*
348 * For guest debugging where we have to reinject #BP if some
349 * INT3 is guest-owned:
350 * Emulate nRIP by moving RIP forward. Will fail if injection
351 * raises a fault that is not intercepted. Still better than
352 * failing in all cases.
353 */
Vitaly Kuznetsovf8ea7c62019-08-13 15:53:30 +0200354 (void)skip_emulated_instruction(&svm->vcpu);
Jan Kiszka66b71382010-02-23 17:47:56 +0100355 rip = kvm_rip_read(&svm->vcpu);
356 svm->int3_rip = rip + svm->vmcb->save.cs.base;
357 svm->int3_injected = rip - old_rip;
358 }
359
Jan Kiszka116a4752010-02-23 17:47:54 +0100360 svm->vmcb->control.event_inj = nr
361 | SVM_EVTINJ_VALID
362 | (has_error_code ? SVM_EVTINJ_VALID_ERR : 0)
363 | SVM_EVTINJ_TYPE_EXEPT;
364 svm->vmcb->control.event_inj_err = error_code;
365}
366
Joerg Roedel67ec6602010-05-17 14:43:35 +0200367static void svm_init_erratum_383(void)
368{
369 u32 low, high;
370 int err;
371 u64 val;
372
Borislav Petkove6ee94d2013-03-20 15:07:27 +0100373 if (!static_cpu_has_bug(X86_BUG_AMD_TLB_MMATCH))
Joerg Roedel67ec6602010-05-17 14:43:35 +0200374 return;
375
376 /* Use _safe variants to not break nested virtualization */
377 val = native_read_msr_safe(MSR_AMD64_DC_CFG, &err);
378 if (err)
379 return;
380
381 val |= (1ULL << 47);
382
383 low = lower_32_bits(val);
384 high = upper_32_bits(val);
385
386 native_write_msr_safe(MSR_AMD64_DC_CFG, low, high);
387
388 erratum_383_found = true;
389}
390
Boris Ostrovsky2b036c62012-01-09 14:00:35 -0500391static void svm_init_osvw(struct kvm_vcpu *vcpu)
392{
393 /*
394 * Guests should see errata 400 and 415 as fixed (assuming that
395 * HLT and IO instructions are intercepted).
396 */
397 vcpu->arch.osvw.length = (osvw_len >= 3) ? (osvw_len) : 3;
398 vcpu->arch.osvw.status = osvw_status & ~(6ULL);
399
400 /*
401 * By increasing VCPU's osvw.length to 3 we are telling the guest that
402 * all osvw.status bits inside that length, including bit 0 (which is
403 * reserved for erratum 298), are valid. However, if host processor's
404 * osvw_len is 0 then osvw_status[0] carries no information. We need to
405 * be conservative here and therefore we tell the guest that erratum 298
406 * is present (because we really don't know).
407 */
408 if (osvw_len == 0 && boot_cpu_data.x86 == 0x10)
409 vcpu->arch.osvw.status |= 1;
410}
411
Avi Kivity6aa8b732006-12-10 02:21:36 -0800412static int has_svm(void)
413{
Eduardo Habkost63d11422008-11-17 19:03:20 -0200414 const char *msg;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800415
Eduardo Habkost63d11422008-11-17 19:03:20 -0200416 if (!cpu_has_svm(&msg)) {
Joe Perchesff81ff12009-01-08 11:05:17 -0800417 printk(KERN_INFO "has_svm: %s\n", msg);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800418 return 0;
419 }
420
Avi Kivity6aa8b732006-12-10 02:21:36 -0800421 return 1;
422}
423
Radim Krčmář13a34e02014-08-28 15:13:03 +0200424static void svm_hardware_disable(void)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800425{
Joerg Roedelfbc0db72011-03-25 09:44:46 +0100426 /* Make sure we clean up behind us */
427 if (static_cpu_has(X86_FEATURE_TSCRATEMSR))
428 wrmsrl(MSR_AMD64_TSC_RATIO, TSC_RATIO_DEFAULT);
429
Eduardo Habkost2c8dcee2008-11-17 19:03:21 -0200430 cpu_svm_disable();
Joerg Roedel1018faa2012-02-29 14:57:32 +0100431
432 amd_pmu_disable_virt();
Avi Kivity6aa8b732006-12-10 02:21:36 -0800433}
434
Radim Krčmář13a34e02014-08-28 15:13:03 +0200435static int svm_hardware_enable(void)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800436{
437
Tejun Heo0fe1e002009-10-29 22:34:14 +0900438 struct svm_cpu_data *sd;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800439 uint64_t efer;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800440 struct desc_struct *gdt;
441 int me = raw_smp_processor_id();
442
Alexander Graf10474ae2009-09-15 11:37:46 +0200443 rdmsrl(MSR_EFER, efer);
444 if (efer & EFER_SVME)
445 return -EBUSY;
446
Avi Kivity6aa8b732006-12-10 02:21:36 -0800447 if (!has_svm()) {
Borislav Petkov1f5b77f2012-10-20 20:20:04 +0200448 pr_err("%s: err EOPNOTSUPP on %d\n", __func__, me);
Alexander Graf10474ae2009-09-15 11:37:46 +0200449 return -EINVAL;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800450 }
Tejun Heo0fe1e002009-10-29 22:34:14 +0900451 sd = per_cpu(svm_data, me);
Tejun Heo0fe1e002009-10-29 22:34:14 +0900452 if (!sd) {
Borislav Petkov1f5b77f2012-10-20 20:20:04 +0200453 pr_err("%s: svm_data is NULL on %d\n", __func__, me);
Alexander Graf10474ae2009-09-15 11:37:46 +0200454 return -EINVAL;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800455 }
456
Tejun Heo0fe1e002009-10-29 22:34:14 +0900457 sd->asid_generation = 1;
458 sd->max_asid = cpuid_ebx(SVM_CPUID_FUNC) - 1;
459 sd->next_asid = sd->max_asid + 1;
Brijesh Singhed3cd232017-12-04 10:57:32 -0600460 sd->min_asid = max_sev_asid + 1;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800461
Thomas Garnier45fc8752017-03-14 10:05:08 -0700462 gdt = get_current_gdt_rw();
Tejun Heo0fe1e002009-10-29 22:34:14 +0900463 sd->tss_desc = (struct kvm_ldttss_desc *)(gdt + GDT_ENTRY_TSS);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800464
Alexander Graf9962d032008-11-25 20:17:02 +0100465 wrmsrl(MSR_EFER, efer | EFER_SVME);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800466
Linus Torvaldsd0316552009-12-14 09:58:24 -0800467 wrmsrl(MSR_VM_HSAVE_PA, page_to_pfn(sd->save_area) << PAGE_SHIFT);
Alexander Graf10474ae2009-09-15 11:37:46 +0200468
Joerg Roedelfbc0db72011-03-25 09:44:46 +0100469 if (static_cpu_has(X86_FEATURE_TSCRATEMSR)) {
470 wrmsrl(MSR_AMD64_TSC_RATIO, TSC_RATIO_DEFAULT);
Christoph Lameter89cbc762014-08-17 12:30:40 -0500471 __this_cpu_write(current_tsc_ratio, TSC_RATIO_DEFAULT);
Joerg Roedelfbc0db72011-03-25 09:44:46 +0100472 }
473
Boris Ostrovsky2b036c62012-01-09 14:00:35 -0500474
475 /*
476 * Get OSVW bits.
477 *
478 * Note that it is possible to have a system with mixed processor
479 * revisions and therefore different OSVW bits. If bits are not the same
480 * on different processors then choose the worst case (i.e. if erratum
481 * is present on one processor and not on another then assume that the
482 * erratum is present everywhere).
483 */
484 if (cpu_has(&boot_cpu_data, X86_FEATURE_OSVW)) {
485 uint64_t len, status = 0;
486 int err;
487
488 len = native_read_msr_safe(MSR_AMD64_OSVW_ID_LENGTH, &err);
489 if (!err)
490 status = native_read_msr_safe(MSR_AMD64_OSVW_STATUS,
491 &err);
492
493 if (err)
494 osvw_status = osvw_len = 0;
495 else {
496 if (len < osvw_len)
497 osvw_len = len;
498 osvw_status |= status;
499 osvw_status &= (1ULL << osvw_len) - 1;
500 }
501 } else
502 osvw_status = osvw_len = 0;
503
Joerg Roedel67ec6602010-05-17 14:43:35 +0200504 svm_init_erratum_383();
505
Joerg Roedel1018faa2012-02-29 14:57:32 +0100506 amd_pmu_enable_virt();
507
Alexander Graf10474ae2009-09-15 11:37:46 +0200508 return 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800509}
510
Joerg Roedel0da1db752008-07-02 16:02:11 +0200511static void svm_cpu_uninit(int cpu)
512{
Tejun Heo0fe1e002009-10-29 22:34:14 +0900513 struct svm_cpu_data *sd = per_cpu(svm_data, raw_smp_processor_id());
Joerg Roedel0da1db752008-07-02 16:02:11 +0200514
Tejun Heo0fe1e002009-10-29 22:34:14 +0900515 if (!sd)
Joerg Roedel0da1db752008-07-02 16:02:11 +0200516 return;
517
518 per_cpu(svm_data, raw_smp_processor_id()) = NULL;
Brijesh Singh70cd94e2017-12-04 10:57:34 -0600519 kfree(sd->sev_vmcbs);
Tejun Heo0fe1e002009-10-29 22:34:14 +0900520 __free_page(sd->save_area);
521 kfree(sd);
Joerg Roedel0da1db752008-07-02 16:02:11 +0200522}
523
Avi Kivity6aa8b732006-12-10 02:21:36 -0800524static int svm_cpu_init(int cpu)
525{
Tejun Heo0fe1e002009-10-29 22:34:14 +0900526 struct svm_cpu_data *sd;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800527
Tejun Heo0fe1e002009-10-29 22:34:14 +0900528 sd = kzalloc(sizeof(struct svm_cpu_data), GFP_KERNEL);
529 if (!sd)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800530 return -ENOMEM;
Tejun Heo0fe1e002009-10-29 22:34:14 +0900531 sd->cpu = cpu;
Brijesh Singh70cd94e2017-12-04 10:57:34 -0600532 sd->save_area = alloc_page(GFP_KERNEL);
Tejun Heo0fe1e002009-10-29 22:34:14 +0900533 if (!sd->save_area)
Miaohe Lind80b64f2020-01-04 16:56:49 +0800534 goto free_cpu_data;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800535
Brijesh Singh70cd94e2017-12-04 10:57:34 -0600536 if (svm_sev_enabled()) {
Kees Cook6da2ec52018-06-12 13:55:00 -0700537 sd->sev_vmcbs = kmalloc_array(max_sev_asid + 1,
538 sizeof(void *),
539 GFP_KERNEL);
Brijesh Singh70cd94e2017-12-04 10:57:34 -0600540 if (!sd->sev_vmcbs)
Miaohe Lind80b64f2020-01-04 16:56:49 +0800541 goto free_save_area;
Brijesh Singh70cd94e2017-12-04 10:57:34 -0600542 }
543
Tejun Heo0fe1e002009-10-29 22:34:14 +0900544 per_cpu(svm_data, cpu) = sd;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800545
546 return 0;
547
Miaohe Lind80b64f2020-01-04 16:56:49 +0800548free_save_area:
549 __free_page(sd->save_area);
550free_cpu_data:
Tejun Heo0fe1e002009-10-29 22:34:14 +0900551 kfree(sd);
Miaohe Lind80b64f2020-01-04 16:56:49 +0800552 return -ENOMEM;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800553
554}
555
Joerg Roedelac72a9b2010-03-01 15:34:36 +0100556static bool valid_msr_intercept(u32 index)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800557{
558 int i;
559
Joerg Roedelac72a9b2010-03-01 15:34:36 +0100560 for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++)
561 if (direct_access_msrs[i].index == index)
562 return true;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800563
Joerg Roedelac72a9b2010-03-01 15:34:36 +0100564 return false;
565}
566
KarimAllah Ahmedb2ac58f2018-02-03 15:56:23 +0100567static bool msr_write_intercepted(struct kvm_vcpu *vcpu, unsigned msr)
568{
569 u8 bit_write;
570 unsigned long tmp;
571 u32 offset;
572 u32 *msrpm;
573
574 msrpm = is_guest_mode(vcpu) ? to_svm(vcpu)->nested.msrpm:
575 to_svm(vcpu)->msrpm;
576
577 offset = svm_msrpm_offset(msr);
578 bit_write = 2 * (msr & 0x0f) + 1;
579 tmp = msrpm[offset];
580
581 BUG_ON(offset == MSR_INVALID);
582
583 return !!test_bit(bit_write, &tmp);
584}
585
Avi Kivity6aa8b732006-12-10 02:21:36 -0800586static void set_msr_interception(u32 *msrpm, unsigned msr,
587 int read, int write)
588{
Joerg Roedel455716f2010-03-01 15:34:35 +0100589 u8 bit_read, bit_write;
590 unsigned long tmp;
591 u32 offset;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800592
Joerg Roedelac72a9b2010-03-01 15:34:36 +0100593 /*
594 * If this warning triggers extend the direct_access_msrs list at the
595 * beginning of the file
596 */
597 WARN_ON(!valid_msr_intercept(msr));
598
Joerg Roedel455716f2010-03-01 15:34:35 +0100599 offset = svm_msrpm_offset(msr);
600 bit_read = 2 * (msr & 0x0f);
601 bit_write = 2 * (msr & 0x0f) + 1;
602 tmp = msrpm[offset];
Avi Kivity6aa8b732006-12-10 02:21:36 -0800603
Joerg Roedel455716f2010-03-01 15:34:35 +0100604 BUG_ON(offset == MSR_INVALID);
605
606 read ? clear_bit(bit_read, &tmp) : set_bit(bit_read, &tmp);
607 write ? clear_bit(bit_write, &tmp) : set_bit(bit_write, &tmp);
608
609 msrpm[offset] = tmp;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800610}
611
Joerg Roedelf65c2292008-02-13 18:58:46 +0100612static void svm_vcpu_init_msrpm(u32 *msrpm)
613{
Joerg Roedelac72a9b2010-03-01 15:34:36 +0100614 int i;
615
Joerg Roedelf65c2292008-02-13 18:58:46 +0100616 memset(msrpm, 0xff, PAGE_SIZE * (1 << MSRPM_ALLOC_ORDER));
617
Joerg Roedelac72a9b2010-03-01 15:34:36 +0100618 for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
619 if (!direct_access_msrs[i].always)
620 continue;
621
622 set_msr_interception(msrpm, direct_access_msrs[i].index, 1, 1);
623 }
Joerg Roedelf65c2292008-02-13 18:58:46 +0100624}
625
Joerg Roedel323c3d82010-03-01 15:34:37 +0100626static void add_msr_offset(u32 offset)
627{
628 int i;
629
630 for (i = 0; i < MSRPM_OFFSETS; ++i) {
631
632 /* Offset already in list? */
633 if (msrpm_offsets[i] == offset)
634 return;
635
636 /* Slot used by another offset? */
637 if (msrpm_offsets[i] != MSR_INVALID)
638 continue;
639
640 /* Add offset to list */
641 msrpm_offsets[i] = offset;
642
643 return;
644 }
645
646 /*
647 * If this BUG triggers the msrpm_offsets table has an overflow. Just
648 * increase MSRPM_OFFSETS in this case.
649 */
650 BUG();
651}
652
653static void init_msrpm_offsets(void)
654{
655 int i;
656
657 memset(msrpm_offsets, 0xff, sizeof(msrpm_offsets));
658
659 for (i = 0; direct_access_msrs[i].index != MSR_INVALID; i++) {
660 u32 offset;
661
662 offset = svm_msrpm_offset(direct_access_msrs[i].index);
663 BUG_ON(offset == MSR_INVALID);
664
665 add_msr_offset(offset);
666 }
Avi Kivity6aa8b732006-12-10 02:21:36 -0800667}
668
Joerg Roedel24e09cb2008-02-13 18:58:47 +0100669static void svm_enable_lbrv(struct vcpu_svm *svm)
670{
671 u32 *msrpm = svm->msrpm;
672
Janakarajan Natarajan0dc92112017-07-06 15:50:45 -0500673 svm->vmcb->control.virt_ext |= LBR_CTL_ENABLE_MASK;
Joerg Roedel24e09cb2008-02-13 18:58:47 +0100674 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 1, 1);
675 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 1, 1);
676 set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 1, 1);
677 set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 1, 1);
678}
679
680static void svm_disable_lbrv(struct vcpu_svm *svm)
681{
682 u32 *msrpm = svm->msrpm;
683
Janakarajan Natarajan0dc92112017-07-06 15:50:45 -0500684 svm->vmcb->control.virt_ext &= ~LBR_CTL_ENABLE_MASK;
Joerg Roedel24e09cb2008-02-13 18:58:47 +0100685 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHFROMIP, 0, 0);
686 set_msr_interception(msrpm, MSR_IA32_LASTBRANCHTOIP, 0, 0);
687 set_msr_interception(msrpm, MSR_IA32_LASTINTFROMIP, 0, 0);
688 set_msr_interception(msrpm, MSR_IA32_LASTINTTOIP, 0, 0);
689}
690
Joerg Roedel883b0a92020-03-24 10:41:52 +0100691void disable_nmi_singlestep(struct vcpu_svm *svm)
Ladi Prosek4aebd0e2017-06-21 09:06:57 +0200692{
693 svm->nmi_singlestep = false;
Janakarajan Natarajan640bd6e2017-08-23 09:57:19 -0500694
Ladi Prosekab2f4d732017-06-21 09:06:58 +0200695 if (!(svm->vcpu.guest_debug & KVM_GUESTDBG_SINGLESTEP)) {
696 /* Clear our flags if they were not set by the guest */
697 if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_TF))
698 svm->vmcb->save.rflags &= ~X86_EFLAGS_TF;
699 if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_RF))
700 svm->vmcb->save.rflags &= ~X86_EFLAGS_RF;
701 }
Ladi Prosek4aebd0e2017-06-21 09:06:57 +0200702}
703
Babu Moger8566ac82018-03-16 16:37:26 -0400704static void grow_ple_window(struct kvm_vcpu *vcpu)
705{
706 struct vcpu_svm *svm = to_svm(vcpu);
707 struct vmcb_control_area *control = &svm->vmcb->control;
708 int old = control->pause_filter_count;
709
710 control->pause_filter_count = __grow_ple_window(old,
711 pause_filter_count,
712 pause_filter_count_grow,
713 pause_filter_count_max);
714
Peter Xu4f75bcc2019-09-06 10:17:22 +0800715 if (control->pause_filter_count != old) {
Babu Moger8566ac82018-03-16 16:37:26 -0400716 mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
Peter Xu4f75bcc2019-09-06 10:17:22 +0800717 trace_kvm_ple_window_update(vcpu->vcpu_id,
718 control->pause_filter_count, old);
719 }
Babu Moger8566ac82018-03-16 16:37:26 -0400720}
721
722static void shrink_ple_window(struct kvm_vcpu *vcpu)
723{
724 struct vcpu_svm *svm = to_svm(vcpu);
725 struct vmcb_control_area *control = &svm->vmcb->control;
726 int old = control->pause_filter_count;
727
728 control->pause_filter_count =
729 __shrink_ple_window(old,
730 pause_filter_count,
731 pause_filter_count_shrink,
732 pause_filter_count);
Peter Xu4f75bcc2019-09-06 10:17:22 +0800733 if (control->pause_filter_count != old) {
Babu Moger8566ac82018-03-16 16:37:26 -0400734 mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
Peter Xu4f75bcc2019-09-06 10:17:22 +0800735 trace_kvm_ple_window_update(vcpu->vcpu_id,
736 control->pause_filter_count, old);
737 }
Babu Moger8566ac82018-03-16 16:37:26 -0400738}
739
Tom Lendacky52918ed2020-01-09 17:42:16 -0600740/*
741 * The default MMIO mask is a single bit (excluding the present bit),
742 * which could conflict with the memory encryption bit. Check for
743 * memory encryption support and override the default MMIO mask if
744 * memory encryption is enabled.
745 */
746static __init void svm_adjust_mmio_mask(void)
747{
748 unsigned int enc_bit, mask_bit;
749 u64 msr, mask;
750
751 /* If there is no memory encryption support, use existing mask */
752 if (cpuid_eax(0x80000000) < 0x8000001f)
753 return;
754
755 /* If memory encryption is not enabled, use existing mask */
756 rdmsrl(MSR_K8_SYSCFG, msr);
757 if (!(msr & MSR_K8_SYSCFG_MEM_ENCRYPT))
758 return;
759
760 enc_bit = cpuid_ebx(0x8000001f) & 0x3f;
761 mask_bit = boot_cpu_data.x86_phys_bits;
762
763 /* Increment the mask bit if it is the same as the encryption bit */
764 if (enc_bit == mask_bit)
765 mask_bit++;
766
767 /*
768 * If the mask bit location is below 52, then some bits above the
769 * physical addressing limit will always be reserved, so use the
770 * rsvd_bits() function to generate the mask. This mask, along with
771 * the present bit, will be used to generate a page fault with
772 * PFER.RSV = 1.
773 *
774 * If the mask bit location is 52 (or above), then clear the mask.
775 */
776 mask = (mask_bit < 52) ? rsvd_bits(mask_bit, 51) | PT_PRESENT_MASK : 0;
777
Paolo Bonzinie7581ca2020-05-19 05:04:49 -0400778 kvm_mmu_set_mmio_spte_mask(mask, PT_WRITABLE_MASK | PT_USER_MASK);
Tom Lendacky52918ed2020-01-09 17:42:16 -0600779}
780
Li RongQingdd58f3c2020-02-23 16:13:12 +0800781static void svm_hardware_teardown(void)
782{
783 int cpu;
784
Joerg Roedeleaf78262020-03-24 10:41:54 +0100785 if (svm_sev_enabled())
786 sev_hardware_teardown();
Li RongQingdd58f3c2020-02-23 16:13:12 +0800787
788 for_each_possible_cpu(cpu)
789 svm_cpu_uninit(cpu);
790
791 __free_pages(pfn_to_page(iopm_base >> PAGE_SHIFT), IOPM_ALLOC_ORDER);
792 iopm_base = 0;
793}
794
Sean Christopherson9b58b982020-03-02 15:56:42 -0800795static __init void svm_set_cpu_caps(void)
796{
797 kvm_set_cpu_caps();
798
Paolo Bonzini408e9a32020-03-05 16:11:56 +0100799 supported_xss = 0;
800
Sean Christophersona50718c2020-03-02 15:57:07 -0800801 /* CPUID 0x80000001 and 0x8000000A (SVM features) */
802 if (nested) {
Sean Christopherson9b58b982020-03-02 15:56:42 -0800803 kvm_cpu_cap_set(X86_FEATURE_SVM);
804
Sean Christopherson4eb87462020-03-02 15:57:08 -0800805 if (nrips)
Sean Christophersona50718c2020-03-02 15:57:07 -0800806 kvm_cpu_cap_set(X86_FEATURE_NRIPS);
807
808 if (npt_enabled)
809 kvm_cpu_cap_set(X86_FEATURE_NPT);
810 }
811
Sean Christopherson93c380e2020-03-02 15:56:54 -0800812 /* CPUID 0x80000008 */
813 if (boot_cpu_has(X86_FEATURE_LS_CFG_SSBD) ||
814 boot_cpu_has(X86_FEATURE_AMD_SSBD))
815 kvm_cpu_cap_set(X86_FEATURE_VIRT_SSBD);
Sean Christopherson9b58b982020-03-02 15:56:42 -0800816}
817
Avi Kivity6aa8b732006-12-10 02:21:36 -0800818static __init int svm_hardware_setup(void)
819{
820 int cpu;
821 struct page *iopm_pages;
Joerg Roedelf65c2292008-02-13 18:58:46 +0100822 void *iopm_va;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800823 int r;
824
Avi Kivity6aa8b732006-12-10 02:21:36 -0800825 iopm_pages = alloc_pages(GFP_KERNEL, IOPM_ALLOC_ORDER);
826
827 if (!iopm_pages)
828 return -ENOMEM;
Anthony Liguoric8681332007-04-30 09:48:11 +0300829
830 iopm_va = page_address(iopm_pages);
831 memset(iopm_va, 0xff, PAGE_SIZE * (1 << IOPM_ALLOC_ORDER));
Avi Kivity6aa8b732006-12-10 02:21:36 -0800832 iopm_base = page_to_pfn(iopm_pages) << PAGE_SHIFT;
833
Joerg Roedel323c3d82010-03-01 15:34:37 +0100834 init_msrpm_offsets();
835
Sean Christophersoncfc48182020-03-02 15:56:23 -0800836 supported_xcr0 &= ~(XFEATURE_MASK_BNDREGS | XFEATURE_MASK_BNDCSR);
837
Joerg Roedel50a37eb2008-01-31 14:57:38 +0100838 if (boot_cpu_has(X86_FEATURE_NX))
839 kvm_enable_efer_bits(EFER_NX);
840
Alexander Graf1b2fd702009-02-02 16:23:51 +0100841 if (boot_cpu_has(X86_FEATURE_FXSR_OPT))
842 kvm_enable_efer_bits(EFER_FFXSR);
843
Joerg Roedel92a1f122011-03-25 09:44:51 +0100844 if (boot_cpu_has(X86_FEATURE_TSCRATEMSR)) {
Joerg Roedel92a1f122011-03-25 09:44:51 +0100845 kvm_has_tsc_control = true;
Haozhong Zhangbc9b9612015-10-20 15:39:01 +0800846 kvm_max_tsc_scaling_ratio = TSC_RATIO_MAX;
847 kvm_tsc_scaling_ratio_frac_bits = 32;
Joerg Roedel92a1f122011-03-25 09:44:51 +0100848 }
849
Babu Moger8566ac82018-03-16 16:37:26 -0400850 /* Check for pause filtering support */
851 if (!boot_cpu_has(X86_FEATURE_PAUSEFILTER)) {
852 pause_filter_count = 0;
853 pause_filter_thresh = 0;
854 } else if (!boot_cpu_has(X86_FEATURE_PFTHRESHOLD)) {
855 pause_filter_thresh = 0;
856 }
857
Alexander Graf236de052008-11-25 20:17:10 +0100858 if (nested) {
859 printk(KERN_INFO "kvm: Nested Virtualization enabled\n");
Joerg Roedeleec4b142010-05-05 16:04:44 +0200860 kvm_enable_efer_bits(EFER_SVME | EFER_LMSLE);
Alexander Graf236de052008-11-25 20:17:10 +0100861 }
862
Brijesh Singhe9df0942017-12-04 10:57:33 -0600863 if (sev) {
864 if (boot_cpu_has(X86_FEATURE_SEV) &&
865 IS_ENABLED(CONFIG_KVM_AMD_SEV)) {
866 r = sev_hardware_setup();
867 if (r)
868 sev = false;
869 } else {
870 sev = false;
871 }
872 }
873
Tom Lendacky52918ed2020-01-09 17:42:16 -0600874 svm_adjust_mmio_mask();
875
Zachary Amsden3230bb42009-09-29 11:38:37 -1000876 for_each_possible_cpu(cpu) {
Avi Kivity6aa8b732006-12-10 02:21:36 -0800877 r = svm_cpu_init(cpu);
878 if (r)
Joerg Roedelf65c2292008-02-13 18:58:46 +0100879 goto err;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800880 }
Joerg Roedel33bd6a02008-02-07 13:47:38 +0100881
Avi Kivity2a6b20b2010-11-09 16:15:42 +0200882 if (!boot_cpu_has(X86_FEATURE_NPT))
Joerg Roedele3da3ac2008-02-07 13:47:39 +0100883 npt_enabled = false;
884
Sean Christopherson213e0e12020-03-02 15:57:01 -0800885 if (npt_enabled && !npt)
Joerg Roedel6c7dac72008-02-07 13:47:40 +0100886 npt_enabled = false;
Joerg Roedel6c7dac72008-02-07 13:47:40 +0100887
Sean Christopherson3bae0452020-04-27 17:54:22 -0700888 kvm_configure_mmu(npt_enabled, PG_LEVEL_1G);
Sean Christopherson213e0e12020-03-02 15:57:01 -0800889 pr_info("kvm: Nested Paging %sabled\n", npt_enabled ? "en" : "dis");
Joerg Roedele3da3ac2008-02-07 13:47:39 +0100890
Paolo Bonzinid647eb62019-06-20 14:13:33 +0200891 if (nrips) {
892 if (!boot_cpu_has(X86_FEATURE_NRIPS))
893 nrips = false;
894 }
895
Suravee Suthikulpanit5b8abf12016-06-15 17:24:36 -0500896 if (avic) {
897 if (!npt_enabled ||
898 !boot_cpu_has(X86_FEATURE_AVIC) ||
Suravee Suthikulpanit5881f732016-08-23 13:52:42 -0500899 !IS_ENABLED(CONFIG_X86_LOCAL_APIC)) {
Suravee Suthikulpanit5b8abf12016-06-15 17:24:36 -0500900 avic = false;
Suravee Suthikulpanit5881f732016-08-23 13:52:42 -0500901 } else {
Suravee Suthikulpanit5b8abf12016-06-15 17:24:36 -0500902 pr_info("AVIC enabled\n");
Suravee Suthikulpanit5881f732016-08-23 13:52:42 -0500903
Suravee Suthikulpanit5881f732016-08-23 13:52:42 -0500904 amd_iommu_register_ga_log_notifier(&avic_ga_log_notifier);
905 }
Suravee Suthikulpanit5b8abf12016-06-15 17:24:36 -0500906 }
Suravee Suthikulpanit44a95da2016-05-04 14:09:46 -0500907
Janakarajan Natarajan89c8a492017-07-06 15:50:47 -0500908 if (vls) {
909 if (!npt_enabled ||
Borislav Petkov5442c262017-08-01 20:55:52 +0200910 !boot_cpu_has(X86_FEATURE_V_VMSAVE_VMLOAD) ||
Janakarajan Natarajan89c8a492017-07-06 15:50:47 -0500911 !IS_ENABLED(CONFIG_X86_64)) {
912 vls = false;
913 } else {
914 pr_info("Virtual VMLOAD VMSAVE supported\n");
915 }
916 }
917
Janakarajan Natarajan640bd6e2017-08-23 09:57:19 -0500918 if (vgif) {
919 if (!boot_cpu_has(X86_FEATURE_VGIF))
920 vgif = false;
921 else
922 pr_info("Virtual GIF supported\n");
923 }
924
Sean Christopherson9b58b982020-03-02 15:56:42 -0800925 svm_set_cpu_caps();
Sean Christopherson66a69502020-03-02 15:56:41 -0800926
Avi Kivity6aa8b732006-12-10 02:21:36 -0800927 return 0;
928
Joerg Roedelf65c2292008-02-13 18:58:46 +0100929err:
Li RongQingdd58f3c2020-02-23 16:13:12 +0800930 svm_hardware_teardown();
Avi Kivity6aa8b732006-12-10 02:21:36 -0800931 return r;
932}
933
Avi Kivity6aa8b732006-12-10 02:21:36 -0800934static void init_seg(struct vmcb_seg *seg)
935{
936 seg->selector = 0;
937 seg->attrib = SVM_SELECTOR_P_MASK | SVM_SELECTOR_S_MASK |
Joerg Roedele0231712010-02-24 18:59:10 +0100938 SVM_SELECTOR_WRITE_MASK; /* Read/Write Data Segment */
Avi Kivity6aa8b732006-12-10 02:21:36 -0800939 seg->limit = 0xffff;
940 seg->base = 0;
941}
942
943static void init_sys_seg(struct vmcb_seg *seg, uint32_t type)
944{
945 seg->selector = 0;
946 seg->attrib = SVM_SELECTOR_P_MASK | type;
947 seg->limit = 0xffff;
948 seg->base = 0;
949}
950
Leonid Shatz326e7422018-11-06 12:14:25 +0200951static u64 svm_write_l1_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
Zachary Amsdenf4e1b3c2010-08-19 22:07:16 -1000952{
953 struct vcpu_svm *svm = to_svm(vcpu);
954 u64 g_tsc_offset = 0;
955
Joerg Roedel20307532010-11-29 17:51:48 +0100956 if (is_guest_mode(vcpu)) {
KarimAllah Ahmede79f2452018-04-14 05:10:52 +0200957 /* Write L1's TSC offset. */
Zachary Amsdenf4e1b3c2010-08-19 22:07:16 -1000958 g_tsc_offset = svm->vmcb->control.tsc_offset -
959 svm->nested.hsave->control.tsc_offset;
960 svm->nested.hsave->control.tsc_offset = offset;
Paolo Bonzini45c3af92018-11-25 18:45:35 +0100961 }
962
963 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
964 svm->vmcb->control.tsc_offset - g_tsc_offset,
965 offset);
Zachary Amsdenf4e1b3c2010-08-19 22:07:16 -1000966
967 svm->vmcb->control.tsc_offset = offset + g_tsc_offset;
Joerg Roedel116a0a22010-12-03 11:45:49 +0100968
969 mark_dirty(svm->vmcb, VMCB_INTERCEPTS);
Leonid Shatz326e7422018-11-06 12:14:25 +0200970 return svm->vmcb->control.tsc_offset;
Zachary Amsdenf4e1b3c2010-08-19 22:07:16 -1000971}
972
Paolo Bonzini56908912015-10-19 11:30:19 +0200973static void init_vmcb(struct vcpu_svm *svm)
Avi Kivity6aa8b732006-12-10 02:21:36 -0800974{
Joerg Roedele6101a92008-02-13 18:58:45 +0100975 struct vmcb_control_area *control = &svm->vmcb->control;
976 struct vmcb_save_area *save = &svm->vmcb->save;
Avi Kivity6aa8b732006-12-10 02:21:36 -0800977
Roedel, Joerg4ee546b2010-12-03 10:50:51 +0100978 svm->vcpu.arch.hflags = 0;
Avi Kivitybff78272010-01-07 13:16:08 +0200979
Roedel, Joerg4ee546b2010-12-03 10:50:51 +0100980 set_cr_intercept(svm, INTERCEPT_CR0_READ);
981 set_cr_intercept(svm, INTERCEPT_CR3_READ);
982 set_cr_intercept(svm, INTERCEPT_CR4_READ);
983 set_cr_intercept(svm, INTERCEPT_CR0_WRITE);
984 set_cr_intercept(svm, INTERCEPT_CR3_WRITE);
985 set_cr_intercept(svm, INTERCEPT_CR4_WRITE);
Suravee Suthikulpanit3bbf3562016-05-04 14:09:51 -0500986 if (!kvm_vcpu_apicv_active(&svm->vcpu))
987 set_cr_intercept(svm, INTERCEPT_CR8_WRITE);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800988
Paolo Bonzini5315c712014-03-03 13:08:29 +0100989 set_dr_intercepts(svm);
Avi Kivity6aa8b732006-12-10 02:21:36 -0800990
Joerg Roedel18c918c2010-11-30 18:03:59 +0100991 set_exception_intercept(svm, PF_VECTOR);
992 set_exception_intercept(svm, UD_VECTOR);
993 set_exception_intercept(svm, MC_VECTOR);
Eric Northup54a20552015-11-03 18:03:53 +0100994 set_exception_intercept(svm, AC_VECTOR);
Paolo Bonzinicbdb9672015-11-10 09:14:39 +0100995 set_exception_intercept(svm, DB_VECTOR);
Liran Alon97184202018-03-12 13:12:52 +0200996 /*
997 * Guest access to VMware backdoor ports could legitimately
998 * trigger #GP because of TSS I/O permission bitmap.
999 * We intercept those #GP and allow access to them anyway
1000 * as VMware does.
1001 */
1002 if (enable_vmware_backdoor)
1003 set_exception_intercept(svm, GP_VECTOR);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001004
Joerg Roedel8a05a1b82010-11-30 18:04:00 +01001005 set_intercept(svm, INTERCEPT_INTR);
1006 set_intercept(svm, INTERCEPT_NMI);
1007 set_intercept(svm, INTERCEPT_SMI);
1008 set_intercept(svm, INTERCEPT_SELECTIVE_CR0);
Avi Kivity332b56e2011-11-10 14:57:24 +02001009 set_intercept(svm, INTERCEPT_RDPMC);
Joerg Roedel8a05a1b82010-11-30 18:04:00 +01001010 set_intercept(svm, INTERCEPT_CPUID);
1011 set_intercept(svm, INTERCEPT_INVD);
Joerg Roedel8a05a1b82010-11-30 18:04:00 +01001012 set_intercept(svm, INTERCEPT_INVLPG);
1013 set_intercept(svm, INTERCEPT_INVLPGA);
1014 set_intercept(svm, INTERCEPT_IOIO_PROT);
1015 set_intercept(svm, INTERCEPT_MSR_PROT);
1016 set_intercept(svm, INTERCEPT_TASK_SWITCH);
1017 set_intercept(svm, INTERCEPT_SHUTDOWN);
1018 set_intercept(svm, INTERCEPT_VMRUN);
1019 set_intercept(svm, INTERCEPT_VMMCALL);
1020 set_intercept(svm, INTERCEPT_VMLOAD);
1021 set_intercept(svm, INTERCEPT_VMSAVE);
1022 set_intercept(svm, INTERCEPT_STGI);
1023 set_intercept(svm, INTERCEPT_CLGI);
1024 set_intercept(svm, INTERCEPT_SKINIT);
1025 set_intercept(svm, INTERCEPT_WBINVD);
Joerg Roedel81dd35d2010-12-07 17:15:06 +01001026 set_intercept(svm, INTERCEPT_XSETBV);
Jim Mattson0cb84102019-09-19 15:59:17 -07001027 set_intercept(svm, INTERCEPT_RDPRU);
Brijesh Singh7607b712018-02-19 10:14:44 -06001028 set_intercept(svm, INTERCEPT_RSM);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001029
Wanpeng Li4d5422c2018-03-12 04:53:02 -07001030 if (!kvm_mwait_in_guest(svm->vcpu.kvm)) {
Michael S. Tsirkin668fffa2017-04-21 12:27:17 +02001031 set_intercept(svm, INTERCEPT_MONITOR);
1032 set_intercept(svm, INTERCEPT_MWAIT);
1033 }
1034
Wanpeng Licaa057a2018-03-12 04:53:03 -07001035 if (!kvm_hlt_in_guest(svm->vcpu.kvm))
1036 set_intercept(svm, INTERCEPT_HLT);
1037
Tom Lendackyd0ec49d2017-07-17 16:10:27 -05001038 control->iopm_base_pa = __sme_set(iopm_base);
1039 control->msrpm_base_pa = __sme_set(__pa(svm->msrpm));
Avi Kivity6aa8b732006-12-10 02:21:36 -08001040 control->int_ctl = V_INTR_MASKING_MASK;
1041
1042 init_seg(&save->es);
1043 init_seg(&save->ss);
1044 init_seg(&save->ds);
1045 init_seg(&save->fs);
1046 init_seg(&save->gs);
1047
1048 save->cs.selector = 0xf000;
Paolo Bonzini04b66832013-03-19 16:30:26 +01001049 save->cs.base = 0xffff0000;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001050 /* Executable/Readable Code Segment */
1051 save->cs.attrib = SVM_SELECTOR_READ_MASK | SVM_SELECTOR_P_MASK |
1052 SVM_SELECTOR_S_MASK | SVM_SELECTOR_CODE_MASK;
1053 save->cs.limit = 0xffff;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001054
1055 save->gdtr.limit = 0xffff;
1056 save->idtr.limit = 0xffff;
1057
1058 init_sys_seg(&save->ldtr, SEG_TYPE_LDT);
1059 init_sys_seg(&save->tr, SEG_TYPE_BUSY_TSS16);
1060
Paolo Bonzini56908912015-10-19 11:30:19 +02001061 svm_set_efer(&svm->vcpu, 0);
Mike Dayd77c26f2007-10-08 09:02:08 -04001062 save->dr6 = 0xffff0ff0;
Avi Kivityf6e78472010-08-02 15:30:20 +03001063 kvm_set_rflags(&svm->vcpu, 2);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001064 save->rip = 0x0000fff0;
Marcelo Tosatti5fdbf972008-06-27 14:58:02 -03001065 svm->vcpu.arch.regs[VCPU_REGS_RIP] = save->rip;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001066
Joerg Roedele0231712010-02-24 18:59:10 +01001067 /*
Eduardo Habkost18fa0002009-10-24 02:49:59 -02001068 * svm_set_cr0() sets PG and WP and clears NW and CD on save->cr0.
Nadav Amitd28bc9d2015-04-13 14:34:08 +03001069 * It also updates the guest-visible cr0 value.
Avi Kivity6aa8b732006-12-10 02:21:36 -08001070 */
Paolo Bonzini79a80592015-09-21 07:46:55 +02001071 svm_set_cr0(&svm->vcpu, X86_CR0_NW | X86_CR0_CD | X86_CR0_ET);
Igor Mammedovebae8712015-09-18 15:39:05 +02001072 kvm_mmu_reset_context(&svm->vcpu);
Eduardo Habkost18fa0002009-10-24 02:49:59 -02001073
Rusty Russell66aee912007-07-17 23:34:16 +10001074 save->cr4 = X86_CR4_PAE;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001075 /* rdx = ?? */
Joerg Roedel709ddeb2008-02-07 13:47:45 +01001076
1077 if (npt_enabled) {
1078 /* Setup VMCB for Nested Paging */
Tom Lendackycea3a192017-12-04 10:57:24 -06001079 control->nested_ctl |= SVM_NESTED_CTL_NP_ENABLE;
Joerg Roedel8a05a1b82010-11-30 18:04:00 +01001080 clr_intercept(svm, INTERCEPT_INVLPG);
Joerg Roedel18c918c2010-11-30 18:03:59 +01001081 clr_exception_intercept(svm, PF_VECTOR);
Roedel, Joerg4ee546b2010-12-03 10:50:51 +01001082 clr_cr_intercept(svm, INTERCEPT_CR3_READ);
1083 clr_cr_intercept(svm, INTERCEPT_CR3_WRITE);
Radim Krčmář74545702015-04-27 15:11:25 +02001084 save->g_pat = svm->vcpu.arch.pat;
Joerg Roedel709ddeb2008-02-07 13:47:45 +01001085 save->cr3 = 0;
1086 save->cr4 = 0;
1087 }
Joerg Roedelf40f6a42010-12-03 15:25:15 +01001088 svm->asid_generation = 0;
Alexander Graf1371d902008-11-25 20:17:04 +01001089
Joerg Roedele6aa9ab2009-08-07 11:49:33 +02001090 svm->nested.vmcb = 0;
Joerg Roedel2af91942009-08-07 11:49:28 +02001091 svm->vcpu.arch.hflags = 0;
1092
Babu Moger8566ac82018-03-16 16:37:26 -04001093 if (pause_filter_count) {
1094 control->pause_filter_count = pause_filter_count;
1095 if (pause_filter_thresh)
1096 control->pause_filter_thresh = pause_filter_thresh;
Joerg Roedel8a05a1b82010-11-30 18:04:00 +01001097 set_intercept(svm, INTERCEPT_PAUSE);
Babu Moger8566ac82018-03-16 16:37:26 -04001098 } else {
1099 clr_intercept(svm, INTERCEPT_PAUSE);
Mark Langsdorf565d0992009-10-06 14:25:02 -05001100 }
1101
Suravee Suthikulpanit67034bb2017-09-12 10:42:42 -05001102 if (kvm_vcpu_apicv_active(&svm->vcpu))
Suravee Suthikulpanit44a95da2016-05-04 14:09:46 -05001103 avic_init_vmcb(svm);
1104
Janakarajan Natarajan89c8a492017-07-06 15:50:47 -05001105 /*
1106 * If hardware supports Virtual VMLOAD VMSAVE then enable it
1107 * in VMCB and clear intercepts to avoid #VMEXIT.
1108 */
1109 if (vls) {
1110 clr_intercept(svm, INTERCEPT_VMLOAD);
1111 clr_intercept(svm, INTERCEPT_VMSAVE);
1112 svm->vmcb->control.virt_ext |= VIRTUAL_VMLOAD_VMSAVE_ENABLE_MASK;
1113 }
1114
Janakarajan Natarajan640bd6e2017-08-23 09:57:19 -05001115 if (vgif) {
1116 clr_intercept(svm, INTERCEPT_STGI);
1117 clr_intercept(svm, INTERCEPT_CLGI);
1118 svm->vmcb->control.int_ctl |= V_GIF_ENABLE_MASK;
1119 }
1120
Brijesh Singh35c6f6492017-12-04 10:57:39 -06001121 if (sev_guest(svm->vcpu.kvm)) {
Brijesh Singh1654efc2017-12-04 10:57:34 -06001122 svm->vmcb->control.nested_ctl |= SVM_NESTED_CTL_SEV_ENABLE;
Brijesh Singh35c6f6492017-12-04 10:57:39 -06001123 clr_exception_intercept(svm, UD_VECTOR);
1124 }
Brijesh Singh1654efc2017-12-04 10:57:34 -06001125
Roedel, Joerg8d28fec2010-12-03 13:15:21 +01001126 mark_all_dirty(svm->vmcb);
1127
Joerg Roedel2af91942009-08-07 11:49:28 +02001128 enable_gif(svm);
Suravee Suthikulpanit44a95da2016-05-04 14:09:46 -05001129
1130}
1131
Nadav Amitd28bc9d2015-04-13 14:34:08 +03001132static void svm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
Avi Kivity04d2cc72007-09-10 18:10:54 +03001133{
1134 struct vcpu_svm *svm = to_svm(vcpu);
Julian Stecklina66f7b722012-12-05 15:26:19 +01001135 u32 dummy;
1136 u32 eax = 1;
Avi Kivity04d2cc72007-09-10 18:10:54 +03001137
KarimAllah Ahmedb2ac58f2018-02-03 15:56:23 +01001138 svm->spec_ctrl = 0;
Thomas Gleixnerccbcd262018-05-09 23:01:01 +02001139 svm->virt_spec_ctrl = 0;
KarimAllah Ahmedb2ac58f2018-02-03 15:56:23 +01001140
Nadav Amitd28bc9d2015-04-13 14:34:08 +03001141 if (!init_event) {
1142 svm->vcpu.arch.apic_base = APIC_DEFAULT_PHYS_BASE |
1143 MSR_IA32_APICBASE_ENABLE;
1144 if (kvm_vcpu_is_reset_bsp(&svm->vcpu))
1145 svm->vcpu.arch.apic_base |= MSR_IA32_APICBASE_BSP;
1146 }
Paolo Bonzini56908912015-10-19 11:30:19 +02001147 init_vmcb(svm);
Avi Kivity70433382007-11-07 12:57:23 +02001148
Sean Christophersonf91af512020-03-04 17:34:37 -08001149 kvm_cpuid(vcpu, &eax, &dummy, &dummy, &dummy, false);
Sean Christophersonde3cd112019-04-30 10:36:17 -07001150 kvm_rdx_write(vcpu, eax);
Suravee Suthikulpanit44a95da2016-05-04 14:09:46 -05001151
1152 if (kvm_vcpu_apicv_active(vcpu) && !init_event)
1153 avic_update_vapic_bar(svm, APIC_DEFAULT_PHYS_BASE);
Avi Kivity04d2cc72007-09-10 18:10:54 +03001154}
1155
Sean Christopherson987b2592019-12-18 13:54:55 -08001156static int svm_create_vcpu(struct kvm_vcpu *vcpu)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001157{
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -04001158 struct vcpu_svm *svm;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001159 struct page *page;
Joerg Roedelf65c2292008-02-13 18:58:46 +01001160 struct page *msrpm_pages;
Alexander Grafb286d5d2008-11-25 20:17:05 +01001161 struct page *hsave_page;
Alexander Graf3d6368e2008-11-25 20:17:07 +01001162 struct page *nested_msrpm_pages;
Rusty Russellfb3f0f52007-07-27 17:16:56 +10001163 int err;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001164
Sean Christophersona9dd6f02019-12-18 13:54:52 -08001165 BUILD_BUG_ON(offsetof(struct vcpu_svm, vcpu) != 0);
1166 svm = to_svm(vcpu);
Rusty Russellfb3f0f52007-07-27 17:16:56 +10001167
Joerg Roedelf65c2292008-02-13 18:58:46 +01001168 err = -ENOMEM;
Ben Gardon1ec69642019-02-11 11:02:51 -08001169 page = alloc_page(GFP_KERNEL_ACCOUNT);
Takuya Yoshikawab7af4042010-03-09 14:55:19 +09001170 if (!page)
Sean Christopherson987b2592019-12-18 13:54:55 -08001171 goto out;
Takuya Yoshikawab7af4042010-03-09 14:55:19 +09001172
Ben Gardon1ec69642019-02-11 11:02:51 -08001173 msrpm_pages = alloc_pages(GFP_KERNEL_ACCOUNT, MSRPM_ALLOC_ORDER);
Joerg Roedelf65c2292008-02-13 18:58:46 +01001174 if (!msrpm_pages)
Takuya Yoshikawab7af4042010-03-09 14:55:19 +09001175 goto free_page1;
Alexander Graf3d6368e2008-11-25 20:17:07 +01001176
Ben Gardon1ec69642019-02-11 11:02:51 -08001177 nested_msrpm_pages = alloc_pages(GFP_KERNEL_ACCOUNT, MSRPM_ALLOC_ORDER);
Alexander Graf3d6368e2008-11-25 20:17:07 +01001178 if (!nested_msrpm_pages)
Takuya Yoshikawab7af4042010-03-09 14:55:19 +09001179 goto free_page2;
Joerg Roedelf65c2292008-02-13 18:58:46 +01001180
Ben Gardon1ec69642019-02-11 11:02:51 -08001181 hsave_page = alloc_page(GFP_KERNEL_ACCOUNT);
Alexander Grafb286d5d2008-11-25 20:17:05 +01001182 if (!hsave_page)
Takuya Yoshikawab7af4042010-03-09 14:55:19 +09001183 goto free_page3;
1184
Suravee Suthikulpanitdfa20092017-09-12 10:42:40 -05001185 err = avic_init_vcpu(svm);
1186 if (err)
1187 goto free_page4;
Suravee Suthikulpanit44a95da2016-05-04 14:09:46 -05001188
Suravee Suthikulpanit8221c132016-05-04 14:09:52 -05001189 /* We initialize this flag to true to make sure that the is_running
1190 * bit would be set the first time the vcpu is loaded.
1191 */
Suravee Suthikulpanit6c3e4422019-11-14 14:15:12 -06001192 if (irqchip_in_kernel(vcpu->kvm) && kvm_apicv_activated(vcpu->kvm))
1193 svm->avic_is_running = true;
Suravee Suthikulpanit8221c132016-05-04 14:09:52 -05001194
Joerg Roedele6aa9ab2009-08-07 11:49:33 +02001195 svm->nested.hsave = page_address(hsave_page);
Paolo Bonzinicc440cd2020-05-13 13:36:32 -04001196 clear_page(svm->nested.hsave);
Alexander Grafb286d5d2008-11-25 20:17:05 +01001197
Takuya Yoshikawab7af4042010-03-09 14:55:19 +09001198 svm->msrpm = page_address(msrpm_pages);
1199 svm_vcpu_init_msrpm(svm->msrpm);
1200
Joerg Roedele6aa9ab2009-08-07 11:49:33 +02001201 svm->nested.msrpm = page_address(nested_msrpm_pages);
Joerg Roedel323c3d82010-03-01 15:34:37 +01001202 svm_vcpu_init_msrpm(svm->nested.msrpm);
Alexander Graf3d6368e2008-11-25 20:17:07 +01001203
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -04001204 svm->vmcb = page_address(page);
1205 clear_page(svm->vmcb);
Tom Lendackyd0ec49d2017-07-17 16:10:27 -05001206 svm->vmcb_pa = __sme_set(page_to_pfn(page) << PAGE_SHIFT);
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -04001207 svm->asid_generation = 0;
Paolo Bonzini56908912015-10-19 11:30:19 +02001208 init_vmcb(svm);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001209
Sean Christopherson7f271792019-12-18 13:54:51 -08001210 svm_init_osvw(vcpu);
Paolo Bonzinibab0c312020-02-11 18:40:58 +01001211 vcpu->arch.microcode_version = 0x01000065;
Boris Ostrovsky2b036c62012-01-09 14:00:35 -05001212
Sean Christophersona9dd6f02019-12-18 13:54:52 -08001213 return 0;
Avi Kivity36241b82006-12-22 01:05:20 -08001214
Suravee Suthikulpanit44a95da2016-05-04 14:09:46 -05001215free_page4:
1216 __free_page(hsave_page);
Takuya Yoshikawab7af4042010-03-09 14:55:19 +09001217free_page3:
1218 __free_pages(nested_msrpm_pages, MSRPM_ALLOC_ORDER);
1219free_page2:
1220 __free_pages(msrpm_pages, MSRPM_ALLOC_ORDER);
1221free_page1:
1222 __free_page(page);
Sean Christopherson987b2592019-12-18 13:54:55 -08001223out:
Sean Christophersona9dd6f02019-12-18 13:54:52 -08001224 return err;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001225}
1226
Jim Mattsonfd65d312018-05-22 09:54:20 -07001227static void svm_clear_current_vmcb(struct vmcb *vmcb)
1228{
1229 int i;
1230
1231 for_each_online_cpu(i)
1232 cmpxchg(&per_cpu(svm_data, i)->current_vmcb, vmcb, NULL);
1233}
1234
Avi Kivity6aa8b732006-12-10 02:21:36 -08001235static void svm_free_vcpu(struct kvm_vcpu *vcpu)
1236{
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -04001237 struct vcpu_svm *svm = to_svm(vcpu);
1238
Jim Mattsonfd65d312018-05-22 09:54:20 -07001239 /*
1240 * The vmcb page can be recycled, causing a false negative in
1241 * svm_vcpu_load(). So, ensure that no logical CPU has this
1242 * vmcb page recorded as its current vmcb.
1243 */
1244 svm_clear_current_vmcb(svm->vmcb);
1245
Tom Lendackyd0ec49d2017-07-17 16:10:27 -05001246 __free_page(pfn_to_page(__sme_clr(svm->vmcb_pa) >> PAGE_SHIFT));
Joerg Roedelf65c2292008-02-13 18:58:46 +01001247 __free_pages(virt_to_page(svm->msrpm), MSRPM_ALLOC_ORDER);
Joerg Roedele6aa9ab2009-08-07 11:49:33 +02001248 __free_page(virt_to_page(svm->nested.hsave));
1249 __free_pages(virt_to_page(svm->nested.msrpm), MSRPM_ALLOC_ORDER);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001250}
1251
Avi Kivity15ad7142007-07-11 18:17:21 +03001252static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001253{
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -04001254 struct vcpu_svm *svm = to_svm(vcpu);
Ashok Raj15d45072018-02-01 22:59:43 +01001255 struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
Avi Kivity15ad7142007-07-11 18:17:21 +03001256 int i;
Avi Kivity0cc50642007-03-25 12:07:27 +02001257
Avi Kivity0cc50642007-03-25 12:07:27 +02001258 if (unlikely(cpu != vcpu->cpu)) {
Marcelo Tosatti4b656b12009-07-21 12:47:45 -03001259 svm->asid_generation = 0;
Roedel, Joerg8d28fec2010-12-03 13:15:21 +01001260 mark_all_dirty(svm->vmcb);
Avi Kivity0cc50642007-03-25 12:07:27 +02001261 }
Anthony Liguori94dfbdb2007-04-29 11:56:06 +03001262
Avi Kivity82ca2d12010-10-21 12:20:34 +02001263#ifdef CONFIG_X86_64
1264 rdmsrl(MSR_GS_BASE, to_svm(vcpu)->host.gs_base);
1265#endif
Avi Kivitydacccfd2010-10-21 12:20:33 +02001266 savesegment(fs, svm->host.fs);
1267 savesegment(gs, svm->host.gs);
1268 svm->host.ldt = kvm_read_ldt();
1269
Anthony Liguori94dfbdb2007-04-29 11:56:06 +03001270 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -04001271 rdmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
Joerg Roedelfbc0db72011-03-25 09:44:46 +01001272
Haozhong Zhangad7218832015-10-20 15:39:02 +08001273 if (static_cpu_has(X86_FEATURE_TSCRATEMSR)) {
1274 u64 tsc_ratio = vcpu->arch.tsc_scaling_ratio;
1275 if (tsc_ratio != __this_cpu_read(current_tsc_ratio)) {
1276 __this_cpu_write(current_tsc_ratio, tsc_ratio);
1277 wrmsrl(MSR_AMD64_TSC_RATIO, tsc_ratio);
1278 }
Joerg Roedelfbc0db72011-03-25 09:44:46 +01001279 }
Paolo Bonzini46896c72015-11-12 14:49:16 +01001280 /* This assumes that the kernel never uses MSR_TSC_AUX */
1281 if (static_cpu_has(X86_FEATURE_RDTSCP))
1282 wrmsrl(MSR_TSC_AUX, svm->tsc_aux);
Suravee Suthikulpanit8221c132016-05-04 14:09:52 -05001283
Ashok Raj15d45072018-02-01 22:59:43 +01001284 if (sd->current_vmcb != svm->vmcb) {
1285 sd->current_vmcb = svm->vmcb;
1286 indirect_branch_prediction_barrier();
1287 }
Suravee Suthikulpanit8221c132016-05-04 14:09:52 -05001288 avic_vcpu_load(vcpu, cpu);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001289}
1290
1291static void svm_vcpu_put(struct kvm_vcpu *vcpu)
1292{
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -04001293 struct vcpu_svm *svm = to_svm(vcpu);
Anthony Liguori94dfbdb2007-04-29 11:56:06 +03001294 int i;
1295
Suravee Suthikulpanit8221c132016-05-04 14:09:52 -05001296 avic_vcpu_put(vcpu);
1297
Avi Kivitye1beb1d2007-11-18 13:50:24 +02001298 ++vcpu->stat.host_state_reload;
Avi Kivitydacccfd2010-10-21 12:20:33 +02001299 kvm_load_ldt(svm->host.ldt);
1300#ifdef CONFIG_X86_64
1301 loadsegment(fs, svm->host.fs);
Andy Lutomirski296f7812016-04-26 12:23:29 -07001302 wrmsrl(MSR_KERNEL_GS_BASE, current->thread.gsbase);
Joerg Roedel893a5ab2011-01-14 16:45:01 +01001303 load_gs_index(svm->host.gs);
Avi Kivitydacccfd2010-10-21 12:20:33 +02001304#else
Avi Kivity831ca602011-03-08 16:09:51 +02001305#ifdef CONFIG_X86_32_LAZY_GS
Avi Kivitydacccfd2010-10-21 12:20:33 +02001306 loadsegment(gs, svm->host.gs);
1307#endif
Avi Kivity831ca602011-03-08 16:09:51 +02001308#endif
Anthony Liguori94dfbdb2007-04-29 11:56:06 +03001309 for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -04001310 wrmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001311}
1312
Avi Kivity6aa8b732006-12-10 02:21:36 -08001313static unsigned long svm_get_rflags(struct kvm_vcpu *vcpu)
1314{
Ladi Prosek9b611742017-06-21 09:06:59 +02001315 struct vcpu_svm *svm = to_svm(vcpu);
1316 unsigned long rflags = svm->vmcb->save.rflags;
1317
1318 if (svm->nmi_singlestep) {
1319 /* Hide our flags if they were not set by the guest */
1320 if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_TF))
1321 rflags &= ~X86_EFLAGS_TF;
1322 if (!(svm->nmi_singlestep_guest_rflags & X86_EFLAGS_RF))
1323 rflags &= ~X86_EFLAGS_RF;
1324 }
1325 return rflags;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001326}
1327
1328static void svm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1329{
Ladi Prosek9b611742017-06-21 09:06:59 +02001330 if (to_svm(vcpu)->nmi_singlestep)
1331 rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF);
1332
Paolo Bonziniae9fedc2014-05-14 09:39:49 +02001333 /*
Andrea Gelminibb3541f2016-05-21 14:14:44 +02001334 * Any change of EFLAGS.VM is accompanied by a reload of SS
Paolo Bonziniae9fedc2014-05-14 09:39:49 +02001335 * (caused by either a task switch or an inter-privilege IRET),
1336 * so we do not need to update the CPL here.
1337 */
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -04001338 to_svm(vcpu)->vmcb->save.rflags = rflags;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001339}
1340
Avi Kivity6de4f3a2009-05-31 22:58:47 +03001341static void svm_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
1342{
1343 switch (reg) {
1344 case VCPU_EXREG_PDPTR:
1345 BUG_ON(!npt_enabled);
Avi Kivity9f8fe502010-12-05 17:30:00 +02001346 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
Avi Kivity6de4f3a2009-05-31 22:58:47 +03001347 break;
1348 default:
Sean Christopherson34059c22019-09-27 14:45:23 -07001349 WARN_ON_ONCE(1);
Avi Kivity6de4f3a2009-05-31 22:58:47 +03001350 }
1351}
1352
Suravee Suthikulpanite14b7782020-05-06 08:17:55 -05001353static void svm_set_vintr(struct vcpu_svm *svm)
Paolo Bonzini64b5bd22020-03-04 13:12:35 -05001354{
1355 struct vmcb_control_area *control;
1356
1357 /* The following fields are ignored when AVIC is enabled */
1358 WARN_ON(kvm_vcpu_apicv_active(&svm->vcpu));
Suravee Suthikulpanite14b7782020-05-06 08:17:55 -05001359 set_intercept(svm, INTERCEPT_VINTR);
Paolo Bonzini64b5bd22020-03-04 13:12:35 -05001360
1361 /*
1362 * This is just a dummy VINTR to actually cause a vmexit to happen.
1363 * Actual injection of virtual interrupts happens through EVENTINJ.
1364 */
1365 control = &svm->vmcb->control;
1366 control->int_vector = 0x0;
1367 control->int_ctl &= ~V_INTR_PRIO_MASK;
1368 control->int_ctl |= V_IRQ_MASK |
1369 ((/*control->int_vector >> 4*/ 0xf) << V_INTR_PRIO_SHIFT);
1370 mark_dirty(svm->vmcb, VMCB_INTR);
1371}
1372
Alexander Graff0b85052008-11-25 20:17:01 +01001373static void svm_clear_vintr(struct vcpu_svm *svm)
1374{
Paolo Bonzinid8e4e582020-05-22 07:38:20 -04001375 const u32 mask = V_TPR_MASK | V_GIF_ENABLE_MASK | V_GIF_MASK | V_INTR_MASKING_MASK;
Joerg Roedel8a05a1b82010-11-30 18:04:00 +01001376 clr_intercept(svm, INTERCEPT_VINTR);
Paolo Bonzini64b5bd22020-03-04 13:12:35 -05001377
Paolo Bonzinid8e4e582020-05-22 07:38:20 -04001378 /* Drop int_ctl fields related to VINTR injection. */
1379 svm->vmcb->control.int_ctl &= mask;
1380 if (is_guest_mode(&svm->vcpu)) {
Paolo Bonzinifb7333d2020-06-08 07:11:47 -04001381 svm->nested.hsave->control.int_ctl &= mask;
1382
Paolo Bonzinid8e4e582020-05-22 07:38:20 -04001383 WARN_ON((svm->vmcb->control.int_ctl & V_TPR_MASK) !=
1384 (svm->nested.ctl.int_ctl & V_TPR_MASK));
1385 svm->vmcb->control.int_ctl |= svm->nested.ctl.int_ctl & ~mask;
1386 }
1387
Paolo Bonzini64b5bd22020-03-04 13:12:35 -05001388 mark_dirty(svm->vmcb, VMCB_INTR);
Alexander Graff0b85052008-11-25 20:17:01 +01001389}
1390
Avi Kivity6aa8b732006-12-10 02:21:36 -08001391static struct vmcb_seg *svm_seg(struct kvm_vcpu *vcpu, int seg)
1392{
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -04001393 struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001394
1395 switch (seg) {
1396 case VCPU_SREG_CS: return &save->cs;
1397 case VCPU_SREG_DS: return &save->ds;
1398 case VCPU_SREG_ES: return &save->es;
1399 case VCPU_SREG_FS: return &save->fs;
1400 case VCPU_SREG_GS: return &save->gs;
1401 case VCPU_SREG_SS: return &save->ss;
1402 case VCPU_SREG_TR: return &save->tr;
1403 case VCPU_SREG_LDTR: return &save->ldtr;
1404 }
1405 BUG();
Al Viro8b6d44c2007-02-09 16:38:40 +00001406 return NULL;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001407}
1408
1409static u64 svm_get_segment_base(struct kvm_vcpu *vcpu, int seg)
1410{
1411 struct vmcb_seg *s = svm_seg(vcpu, seg);
1412
1413 return s->base;
1414}
1415
1416static void svm_get_segment(struct kvm_vcpu *vcpu,
1417 struct kvm_segment *var, int seg)
1418{
1419 struct vmcb_seg *s = svm_seg(vcpu, seg);
1420
1421 var->base = s->base;
1422 var->limit = s->limit;
1423 var->selector = s->selector;
1424 var->type = s->attrib & SVM_SELECTOR_TYPE_MASK;
1425 var->s = (s->attrib >> SVM_SELECTOR_S_SHIFT) & 1;
1426 var->dpl = (s->attrib >> SVM_SELECTOR_DPL_SHIFT) & 3;
1427 var->present = (s->attrib >> SVM_SELECTOR_P_SHIFT) & 1;
1428 var->avl = (s->attrib >> SVM_SELECTOR_AVL_SHIFT) & 1;
1429 var->l = (s->attrib >> SVM_SELECTOR_L_SHIFT) & 1;
1430 var->db = (s->attrib >> SVM_SELECTOR_DB_SHIFT) & 1;
Jim Mattson80112c82014-07-08 09:47:41 +05301431
1432 /*
1433 * AMD CPUs circa 2014 track the G bit for all segments except CS.
1434 * However, the SVM spec states that the G bit is not observed by the
1435 * CPU, and some VMware virtual CPUs drop the G bit for all segments.
1436 * So let's synthesize a legal G bit for all segments, this helps
1437 * running KVM nested. It also helps cross-vendor migration, because
1438 * Intel's vmentry has a check on the 'G' bit.
1439 */
1440 var->g = s->limit > 0xfffff;
Amit Shah25022ac2008-10-27 09:04:17 +00001441
Joerg Roedele0231712010-02-24 18:59:10 +01001442 /*
1443 * AMD's VMCB does not have an explicit unusable field, so emulate it
Andre Przywara19bca6a2009-04-28 12:45:30 +02001444 * for cross vendor migration purposes by "not present"
1445 */
Gioh Kim8eae9572017-05-30 15:24:45 +02001446 var->unusable = !var->present;
Andre Przywara19bca6a2009-04-28 12:45:30 +02001447
Andre Przywara1fbdc7a2009-01-11 22:39:44 +01001448 switch (seg) {
Andre Przywara1fbdc7a2009-01-11 22:39:44 +01001449 case VCPU_SREG_TR:
1450 /*
1451 * Work around a bug where the busy flag in the tr selector
1452 * isn't exposed
1453 */
Amit Shahc0d09822008-10-27 09:04:18 +00001454 var->type |= 0x2;
Andre Przywara1fbdc7a2009-01-11 22:39:44 +01001455 break;
1456 case VCPU_SREG_DS:
1457 case VCPU_SREG_ES:
1458 case VCPU_SREG_FS:
1459 case VCPU_SREG_GS:
1460 /*
1461 * The accessed bit must always be set in the segment
1462 * descriptor cache, although it can be cleared in the
1463 * descriptor, the cached bit always remains at 1. Since
1464 * Intel has a check on this, set it here to support
1465 * cross-vendor migration.
1466 */
1467 if (!var->unusable)
1468 var->type |= 0x1;
1469 break;
Andre Przywarab586eb02009-04-28 12:45:43 +02001470 case VCPU_SREG_SS:
Joerg Roedele0231712010-02-24 18:59:10 +01001471 /*
1472 * On AMD CPUs sometimes the DB bit in the segment
Andre Przywarab586eb02009-04-28 12:45:43 +02001473 * descriptor is left as 1, although the whole segment has
1474 * been made unusable. Clear it here to pass an Intel VMX
1475 * entry check when cross vendor migrating.
1476 */
1477 if (var->unusable)
1478 var->db = 0;
Roman Pend9c1b542017-06-01 10:55:03 +02001479 /* This is symmetric with svm_set_segment() */
Jan Kiszka33b458d2014-06-29 17:12:43 +02001480 var->dpl = to_svm(vcpu)->vmcb->save.cpl;
Andre Przywarab586eb02009-04-28 12:45:43 +02001481 break;
Andre Przywara1fbdc7a2009-01-11 22:39:44 +01001482 }
Avi Kivity6aa8b732006-12-10 02:21:36 -08001483}
1484
Izik Eidus2e4d2652008-03-24 19:38:34 +02001485static int svm_get_cpl(struct kvm_vcpu *vcpu)
1486{
1487 struct vmcb_save_area *save = &to_svm(vcpu)->vmcb->save;
1488
1489 return save->cpl;
1490}
1491
Gleb Natapov89a27f42010-02-16 10:51:48 +02001492static void svm_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001493{
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -04001494 struct vcpu_svm *svm = to_svm(vcpu);
1495
Gleb Natapov89a27f42010-02-16 10:51:48 +02001496 dt->size = svm->vmcb->save.idtr.limit;
1497 dt->address = svm->vmcb->save.idtr.base;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001498}
1499
Gleb Natapov89a27f42010-02-16 10:51:48 +02001500static void svm_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001501{
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -04001502 struct vcpu_svm *svm = to_svm(vcpu);
1503
Gleb Natapov89a27f42010-02-16 10:51:48 +02001504 svm->vmcb->save.idtr.limit = dt->size;
1505 svm->vmcb->save.idtr.base = dt->address ;
Joerg Roedel17a703c2010-12-03 11:45:56 +01001506 mark_dirty(svm->vmcb, VMCB_DT);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001507}
1508
Gleb Natapov89a27f42010-02-16 10:51:48 +02001509static void svm_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001510{
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -04001511 struct vcpu_svm *svm = to_svm(vcpu);
1512
Gleb Natapov89a27f42010-02-16 10:51:48 +02001513 dt->size = svm->vmcb->save.gdtr.limit;
1514 dt->address = svm->vmcb->save.gdtr.base;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001515}
1516
Gleb Natapov89a27f42010-02-16 10:51:48 +02001517static void svm_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001518{
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -04001519 struct vcpu_svm *svm = to_svm(vcpu);
1520
Gleb Natapov89a27f42010-02-16 10:51:48 +02001521 svm->vmcb->save.gdtr.limit = dt->size;
1522 svm->vmcb->save.gdtr.base = dt->address ;
Joerg Roedel17a703c2010-12-03 11:45:56 +01001523 mark_dirty(svm->vmcb, VMCB_DT);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001524}
1525
Avi Kivityd2251572010-01-06 10:55:27 +02001526static void update_cr0_intercept(struct vcpu_svm *svm)
1527{
1528 ulong gcr0 = svm->vcpu.arch.cr0;
1529 u64 *hcr0 = &svm->vmcb->save.cr0;
1530
Paolo Bonzinibd7e5b02017-02-03 21:18:52 -08001531 *hcr0 = (*hcr0 & ~SVM_CR0_SELECTIVE_MASK)
1532 | (gcr0 & SVM_CR0_SELECTIVE_MASK);
Avi Kivityd2251572010-01-06 10:55:27 +02001533
Joerg Roedeldcca1a62010-12-03 11:45:54 +01001534 mark_dirty(svm->vmcb, VMCB_CR);
Avi Kivityd2251572010-01-06 10:55:27 +02001535
Paolo Bonzinibd7e5b02017-02-03 21:18:52 -08001536 if (gcr0 == *hcr0) {
Roedel, Joerg4ee546b2010-12-03 10:50:51 +01001537 clr_cr_intercept(svm, INTERCEPT_CR0_READ);
1538 clr_cr_intercept(svm, INTERCEPT_CR0_WRITE);
Avi Kivityd2251572010-01-06 10:55:27 +02001539 } else {
Roedel, Joerg4ee546b2010-12-03 10:50:51 +01001540 set_cr_intercept(svm, INTERCEPT_CR0_READ);
1541 set_cr_intercept(svm, INTERCEPT_CR0_WRITE);
Avi Kivityd2251572010-01-06 10:55:27 +02001542 }
1543}
1544
Joerg Roedel883b0a92020-03-24 10:41:52 +01001545void svm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001546{
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -04001547 struct vcpu_svm *svm = to_svm(vcpu);
1548
Avi Kivity05b3e0c2006-12-13 00:33:45 -08001549#ifdef CONFIG_X86_64
Avi Kivityf6801df2010-01-21 15:31:50 +02001550 if (vcpu->arch.efer & EFER_LME) {
Rusty Russell707d92fa2007-07-17 23:19:08 +10001551 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
Avi Kivityf6801df2010-01-21 15:31:50 +02001552 vcpu->arch.efer |= EFER_LMA;
Carlo Marcelo Arenas Belon2b5203e2007-12-01 06:17:11 -06001553 svm->vmcb->save.efer |= EFER_LMA | EFER_LME;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001554 }
1555
Mike Dayd77c26f2007-10-08 09:02:08 -04001556 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG)) {
Avi Kivityf6801df2010-01-21 15:31:50 +02001557 vcpu->arch.efer &= ~EFER_LMA;
Carlo Marcelo Arenas Belon2b5203e2007-12-01 06:17:11 -06001558 svm->vmcb->save.efer &= ~(EFER_LMA | EFER_LME);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001559 }
1560 }
1561#endif
Zhang Xiantaoad312c72007-12-13 23:50:52 +08001562 vcpu->arch.cr0 = cr0;
Avi Kivity888f9f32010-01-10 12:14:04 +02001563
1564 if (!npt_enabled)
1565 cr0 |= X86_CR0_PG | X86_CR0_WP;
Avi Kivity02daab22009-12-30 12:40:26 +02001566
Paolo Bonzinibcf166a2015-10-01 13:19:55 +02001567 /*
1568 * re-enable caching here because the QEMU bios
1569 * does not do it - this results in some delay at
1570 * reboot
1571 */
1572 if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
1573 cr0 &= ~(X86_CR0_CD | X86_CR0_NW);
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -04001574 svm->vmcb->save.cr0 = cr0;
Joerg Roedeldcca1a62010-12-03 11:45:54 +01001575 mark_dirty(svm->vmcb, VMCB_CR);
Avi Kivityd2251572010-01-06 10:55:27 +02001576 update_cr0_intercept(svm);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001577}
1578
Joerg Roedel883b0a92020-03-24 10:41:52 +01001579int svm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001580{
Andy Lutomirski1e02ce42014-10-24 15:58:08 -07001581 unsigned long host_cr4_mce = cr4_read_shadow() & X86_CR4_MCE;
Joerg Roedele5eab0c2008-09-09 19:11:51 +02001582 unsigned long old_cr4 = to_svm(vcpu)->vmcb->save.cr4;
1583
Nadav Har'El5e1746d2011-05-25 23:03:24 +03001584 if (cr4 & X86_CR4_VMXE)
1585 return 1;
1586
Joerg Roedele5eab0c2008-09-09 19:11:51 +02001587 if (npt_enabled && ((old_cr4 ^ cr4) & X86_CR4_PGE))
Sean Christophersonf55ac302020-03-20 14:28:12 -07001588 svm_flush_tlb(vcpu);
Joerg Roedel6394b642008-04-09 14:15:29 +02001589
Joerg Roedelec077262008-04-09 14:15:28 +02001590 vcpu->arch.cr4 = cr4;
1591 if (!npt_enabled)
1592 cr4 |= X86_CR4_PAE;
Joerg Roedel6394b642008-04-09 14:15:29 +02001593 cr4 |= host_cr4_mce;
Joerg Roedelec077262008-04-09 14:15:28 +02001594 to_svm(vcpu)->vmcb->save.cr4 = cr4;
Joerg Roedeldcca1a62010-12-03 11:45:54 +01001595 mark_dirty(to_svm(vcpu)->vmcb, VMCB_CR);
Nadav Har'El5e1746d2011-05-25 23:03:24 +03001596 return 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001597}
1598
1599static void svm_set_segment(struct kvm_vcpu *vcpu,
1600 struct kvm_segment *var, int seg)
1601{
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -04001602 struct vcpu_svm *svm = to_svm(vcpu);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001603 struct vmcb_seg *s = svm_seg(vcpu, seg);
1604
1605 s->base = var->base;
1606 s->limit = var->limit;
1607 s->selector = var->selector;
Roman Pend9c1b542017-06-01 10:55:03 +02001608 s->attrib = (var->type & SVM_SELECTOR_TYPE_MASK);
1609 s->attrib |= (var->s & 1) << SVM_SELECTOR_S_SHIFT;
1610 s->attrib |= (var->dpl & 3) << SVM_SELECTOR_DPL_SHIFT;
1611 s->attrib |= ((var->present & 1) && !var->unusable) << SVM_SELECTOR_P_SHIFT;
1612 s->attrib |= (var->avl & 1) << SVM_SELECTOR_AVL_SHIFT;
1613 s->attrib |= (var->l & 1) << SVM_SELECTOR_L_SHIFT;
1614 s->attrib |= (var->db & 1) << SVM_SELECTOR_DB_SHIFT;
1615 s->attrib |= (var->g & 1) << SVM_SELECTOR_G_SHIFT;
Paolo Bonziniae9fedc2014-05-14 09:39:49 +02001616
1617 /*
1618 * This is always accurate, except if SYSRET returned to a segment
1619 * with SS.DPL != 3. Intel does not have this quirk, and always
1620 * forces SS.DPL to 3 on sysret, so we ignore that case; fixing it
1621 * would entail passing the CPL to userspace and back.
1622 */
1623 if (seg == VCPU_SREG_SS)
Roman Pend9c1b542017-06-01 10:55:03 +02001624 /* This is symmetric with svm_get_segment() */
1625 svm->vmcb->save.cpl = (var->dpl & 3);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001626
Joerg Roedel060d0c92010-12-03 11:45:57 +01001627 mark_dirty(svm->vmcb, VMCB_SEG);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001628}
1629
Paolo Bonzinicbdb9672015-11-10 09:14:39 +01001630static void update_bp_intercept(struct kvm_vcpu *vcpu)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001631{
Jan Kiszkad0bfb942008-12-15 13:52:10 +01001632 struct vcpu_svm *svm = to_svm(vcpu);
1633
Joerg Roedel18c918c2010-11-30 18:03:59 +01001634 clr_exception_intercept(svm, BP_VECTOR);
Gleb Natapov44c11432009-05-11 13:35:52 +03001635
Jan Kiszkad0bfb942008-12-15 13:52:10 +01001636 if (vcpu->guest_debug & KVM_GUESTDBG_ENABLE) {
Jan Kiszkad0bfb942008-12-15 13:52:10 +01001637 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
Joerg Roedel18c918c2010-11-30 18:03:59 +01001638 set_exception_intercept(svm, BP_VECTOR);
Jan Kiszkad0bfb942008-12-15 13:52:10 +01001639 } else
1640 vcpu->guest_debug = 0;
Gleb Natapov44c11432009-05-11 13:35:52 +03001641}
1642
Tejun Heo0fe1e002009-10-29 22:34:14 +09001643static void new_asid(struct vcpu_svm *svm, struct svm_cpu_data *sd)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001644{
Tejun Heo0fe1e002009-10-29 22:34:14 +09001645 if (sd->next_asid > sd->max_asid) {
1646 ++sd->asid_generation;
Brijesh Singh4faefff2017-12-04 10:57:25 -06001647 sd->next_asid = sd->min_asid;
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -04001648 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ALL_ASID;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001649 }
1650
Tejun Heo0fe1e002009-10-29 22:34:14 +09001651 svm->asid_generation = sd->asid_generation;
1652 svm->vmcb->control.asid = sd->next_asid++;
Joerg Roedeld48086d2010-12-03 11:45:51 +01001653
1654 mark_dirty(svm->vmcb, VMCB_ASID);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001655}
1656
Paolo Bonzinid67668e2020-05-06 06:40:04 -04001657static void svm_set_dr6(struct vcpu_svm *svm, unsigned long value)
Jan Kiszka73aaf249e2014-01-04 18:47:16 +01001658{
Paolo Bonzinid67668e2020-05-06 06:40:04 -04001659 struct vmcb *vmcb = svm->vmcb;
Jan Kiszka73aaf249e2014-01-04 18:47:16 +01001660
Paolo Bonzinid67668e2020-05-06 06:40:04 -04001661 if (unlikely(value != vmcb->save.dr6)) {
1662 vmcb->save.dr6 = value;
1663 mark_dirty(vmcb, VMCB_DR);
1664 }
Jan Kiszka73aaf249e2014-01-04 18:47:16 +01001665}
1666
Paolo Bonzinifacb0132014-02-21 10:32:27 +01001667static void svm_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
1668{
1669 struct vcpu_svm *svm = to_svm(vcpu);
1670
1671 get_debugreg(vcpu->arch.db[0], 0);
1672 get_debugreg(vcpu->arch.db[1], 1);
1673 get_debugreg(vcpu->arch.db[2], 2);
1674 get_debugreg(vcpu->arch.db[3], 3);
Paolo Bonzinid67668e2020-05-06 06:40:04 -04001675 /*
1676 * We cannot reset svm->vmcb->save.dr6 to DR6_FIXED_1|DR6_RTM here,
1677 * because db_interception might need it. We can do it before vmentry.
1678 */
Paolo Bonzini5679b802020-05-04 11:28:25 -04001679 vcpu->arch.dr6 = svm->vmcb->save.dr6;
Paolo Bonzinifacb0132014-02-21 10:32:27 +01001680 vcpu->arch.dr7 = svm->vmcb->save.dr7;
Paolo Bonzinifacb0132014-02-21 10:32:27 +01001681 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
1682 set_dr_intercepts(svm);
1683}
1684
Gleb Natapov020df072010-04-13 10:05:23 +03001685static void svm_set_dr7(struct kvm_vcpu *vcpu, unsigned long value)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001686{
Jan Kiszka42dbaa52008-12-15 13:52:10 +01001687 struct vcpu_svm *svm = to_svm(vcpu);
Jan Kiszka42dbaa52008-12-15 13:52:10 +01001688
Gleb Natapov020df072010-04-13 10:05:23 +03001689 svm->vmcb->save.dr7 = value;
Joerg Roedel72214b92010-12-03 11:45:55 +01001690 mark_dirty(svm->vmcb, VMCB_DR);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001691}
1692
Avi Kivity851ba692009-08-24 11:10:17 +03001693static int pf_interception(struct vcpu_svm *svm)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001694{
Brijesh Singh0ede79e2017-12-04 10:57:39 -06001695 u64 fault_address = __sme_clr(svm->vmcb->control.exit_info_2);
Wanpeng Li1261bfa2017-07-13 18:30:40 -07001696 u64 error_code = svm->vmcb->control.exit_info_1;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001697
Wanpeng Li1261bfa2017-07-13 18:30:40 -07001698 return kvm_handle_page_fault(&svm->vcpu, error_code, fault_address,
Brijesh Singh00b10fe2017-12-04 10:57:40 -06001699 static_cpu_has(X86_FEATURE_DECODEASSISTS) ?
1700 svm->vmcb->control.insn_bytes : NULL,
Paolo Bonzinid0006532017-08-11 18:36:43 +02001701 svm->vmcb->control.insn_len);
1702}
1703
1704static int npf_interception(struct vcpu_svm *svm)
1705{
Brijesh Singh0ede79e2017-12-04 10:57:39 -06001706 u64 fault_address = __sme_clr(svm->vmcb->control.exit_info_2);
Paolo Bonzinid0006532017-08-11 18:36:43 +02001707 u64 error_code = svm->vmcb->control.exit_info_1;
1708
1709 trace_kvm_page_fault(fault_address, error_code);
1710 return kvm_mmu_page_fault(&svm->vcpu, fault_address, error_code,
Brijesh Singh00b10fe2017-12-04 10:57:40 -06001711 static_cpu_has(X86_FEATURE_DECODEASSISTS) ?
1712 svm->vmcb->control.insn_bytes : NULL,
Paolo Bonzinid0006532017-08-11 18:36:43 +02001713 svm->vmcb->control.insn_len);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001714}
1715
Avi Kivity851ba692009-08-24 11:10:17 +03001716static int db_interception(struct vcpu_svm *svm)
Jan Kiszkad0bfb942008-12-15 13:52:10 +01001717{
Avi Kivity851ba692009-08-24 11:10:17 +03001718 struct kvm_run *kvm_run = svm->vcpu.run;
Vitaly Kuznetsov99c22172019-04-03 16:06:42 +02001719 struct kvm_vcpu *vcpu = &svm->vcpu;
Avi Kivity851ba692009-08-24 11:10:17 +03001720
Jan Kiszkad0bfb942008-12-15 13:52:10 +01001721 if (!(svm->vcpu.guest_debug &
Gleb Natapov44c11432009-05-11 13:35:52 +03001722 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) &&
Jan Kiszka6be7d302009-10-18 13:24:54 +02001723 !svm->nmi_singlestep) {
Paolo Bonzinid67668e2020-05-06 06:40:04 -04001724 u32 payload = (svm->vmcb->save.dr6 ^ DR6_RTM) & ~DR6_FIXED_1;
1725 kvm_queue_exception_p(&svm->vcpu, DB_VECTOR, payload);
Jan Kiszkad0bfb942008-12-15 13:52:10 +01001726 return 1;
1727 }
Gleb Natapov44c11432009-05-11 13:35:52 +03001728
Jan Kiszka6be7d302009-10-18 13:24:54 +02001729 if (svm->nmi_singlestep) {
Ladi Prosek4aebd0e2017-06-21 09:06:57 +02001730 disable_nmi_singlestep(svm);
Vitaly Kuznetsov99c22172019-04-03 16:06:42 +02001731 /* Make sure we check for pending NMIs upon entry */
1732 kvm_make_request(KVM_REQ_EVENT, vcpu);
Gleb Natapov44c11432009-05-11 13:35:52 +03001733 }
1734
1735 if (svm->vcpu.guest_debug &
Joerg Roedele0231712010-02-24 18:59:10 +01001736 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP)) {
Gleb Natapov44c11432009-05-11 13:35:52 +03001737 kvm_run->exit_reason = KVM_EXIT_DEBUG;
Paolo Bonzinidee919d2020-05-04 09:34:10 -04001738 kvm_run->debug.arch.dr6 = svm->vmcb->save.dr6;
1739 kvm_run->debug.arch.dr7 = svm->vmcb->save.dr7;
Gleb Natapov44c11432009-05-11 13:35:52 +03001740 kvm_run->debug.arch.pc =
1741 svm->vmcb->save.cs.base + svm->vmcb->save.rip;
1742 kvm_run->debug.arch.exception = DB_VECTOR;
1743 return 0;
1744 }
1745
1746 return 1;
Jan Kiszkad0bfb942008-12-15 13:52:10 +01001747}
1748
Avi Kivity851ba692009-08-24 11:10:17 +03001749static int bp_interception(struct vcpu_svm *svm)
Jan Kiszkad0bfb942008-12-15 13:52:10 +01001750{
Avi Kivity851ba692009-08-24 11:10:17 +03001751 struct kvm_run *kvm_run = svm->vcpu.run;
1752
Jan Kiszkad0bfb942008-12-15 13:52:10 +01001753 kvm_run->exit_reason = KVM_EXIT_DEBUG;
1754 kvm_run->debug.arch.pc = svm->vmcb->save.cs.base + svm->vmcb->save.rip;
1755 kvm_run->debug.arch.exception = BP_VECTOR;
1756 return 0;
1757}
1758
Avi Kivity851ba692009-08-24 11:10:17 +03001759static int ud_interception(struct vcpu_svm *svm)
Anthony Liguori7aa81cc2007-09-17 14:57:50 -05001760{
Wanpeng Li082d06e2018-04-03 16:28:48 -07001761 return handle_ud(&svm->vcpu);
Anthony Liguori7aa81cc2007-09-17 14:57:50 -05001762}
1763
Eric Northup54a20552015-11-03 18:03:53 +01001764static int ac_interception(struct vcpu_svm *svm)
1765{
1766 kvm_queue_exception_e(&svm->vcpu, AC_VECTOR, 0);
1767 return 1;
1768}
1769
Liran Alon97184202018-03-12 13:12:52 +02001770static int gp_interception(struct vcpu_svm *svm)
1771{
1772 struct kvm_vcpu *vcpu = &svm->vcpu;
1773 u32 error_code = svm->vmcb->control.exit_info_1;
Liran Alon97184202018-03-12 13:12:52 +02001774
1775 WARN_ON_ONCE(!enable_vmware_backdoor);
1776
Sean Christophersona6c6ed12019-08-27 14:40:30 -07001777 /*
1778 * VMware backdoor emulation on #GP interception only handles IN{S},
1779 * OUT{S}, and RDPMC, none of which generate a non-zero error code.
1780 */
1781 if (error_code) {
1782 kvm_queue_exception_e(vcpu, GP_VECTOR, error_code);
1783 return 1;
1784 }
Sean Christopherson60fc3d02019-08-27 14:40:38 -07001785 return kvm_emulate_instruction(vcpu, EMULTYPE_VMWARE_GP);
Liran Alon97184202018-03-12 13:12:52 +02001786}
1787
Joerg Roedel67ec6602010-05-17 14:43:35 +02001788static bool is_erratum_383(void)
1789{
1790 int err, i;
1791 u64 value;
1792
1793 if (!erratum_383_found)
1794 return false;
1795
1796 value = native_read_msr_safe(MSR_IA32_MC0_STATUS, &err);
1797 if (err)
1798 return false;
1799
1800 /* Bit 62 may or may not be set for this mce */
1801 value &= ~(1ULL << 62);
1802
1803 if (value != 0xb600000000010015ULL)
1804 return false;
1805
1806 /* Clear MCi_STATUS registers */
1807 for (i = 0; i < 6; ++i)
1808 native_write_msr_safe(MSR_IA32_MCx_STATUS(i), 0, 0);
1809
1810 value = native_read_msr_safe(MSR_IA32_MCG_STATUS, &err);
1811 if (!err) {
1812 u32 low, high;
1813
1814 value &= ~(1ULL << 2);
1815 low = lower_32_bits(value);
1816 high = upper_32_bits(value);
1817
1818 native_write_msr_safe(MSR_IA32_MCG_STATUS, low, high);
1819 }
1820
1821 /* Flush tlb to evict multi-match entries */
1822 __flush_tlb_all();
1823
1824 return true;
1825}
1826
Uros Bizjak1c164cb2020-04-11 17:36:27 +02001827/*
1828 * Trigger machine check on the host. We assume all the MSRs are already set up
1829 * by the CPU and that we still run on the same CPU as the MCE occurred on.
1830 * We pass a fake environment to the machine check handler because we want
1831 * the guest to be always treated like user space, no matter what context
1832 * it used internally.
1833 */
1834static void kvm_machine_check(void)
1835{
1836#if defined(CONFIG_X86_MCE)
1837 struct pt_regs regs = {
1838 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
1839 .flags = X86_EFLAGS_IF,
1840 };
1841
1842 do_machine_check(&regs, 0);
1843#endif
1844}
1845
Joerg Roedelfe5913e2010-05-17 14:43:34 +02001846static void svm_handle_mce(struct vcpu_svm *svm)
Joerg Roedel53371b52008-04-09 14:15:30 +02001847{
Joerg Roedel67ec6602010-05-17 14:43:35 +02001848 if (is_erratum_383()) {
1849 /*
1850 * Erratum 383 triggered. Guest state is corrupt so kill the
1851 * guest.
1852 */
1853 pr_err("KVM: Guest triggered AMD Erratum 383\n");
1854
Avi Kivitya8eeb042010-05-10 12:34:53 +03001855 kvm_make_request(KVM_REQ_TRIPLE_FAULT, &svm->vcpu);
Joerg Roedel67ec6602010-05-17 14:43:35 +02001856
1857 return;
1858 }
1859
Joerg Roedel53371b52008-04-09 14:15:30 +02001860 /*
1861 * On an #MC intercept the MCE handler is not called automatically in
1862 * the host. So do it by hand here.
1863 */
Uros Bizjak1c164cb2020-04-11 17:36:27 +02001864 kvm_machine_check();
Joerg Roedelfe5913e2010-05-17 14:43:34 +02001865}
1866
1867static int mc_interception(struct vcpu_svm *svm)
1868{
Joerg Roedel53371b52008-04-09 14:15:30 +02001869 return 1;
1870}
1871
Avi Kivity851ba692009-08-24 11:10:17 +03001872static int shutdown_interception(struct vcpu_svm *svm)
Joerg Roedel46fe4dd2007-01-26 00:56:42 -08001873{
Avi Kivity851ba692009-08-24 11:10:17 +03001874 struct kvm_run *kvm_run = svm->vcpu.run;
1875
Joerg Roedel46fe4dd2007-01-26 00:56:42 -08001876 /*
1877 * VMCB is undefined after a SHUTDOWN intercept
1878 * so reinitialize it.
1879 */
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -04001880 clear_page(svm->vmcb);
Paolo Bonzini56908912015-10-19 11:30:19 +02001881 init_vmcb(svm);
Joerg Roedel46fe4dd2007-01-26 00:56:42 -08001882
1883 kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
1884 return 0;
1885}
1886
Avi Kivity851ba692009-08-24 11:10:17 +03001887static int io_interception(struct vcpu_svm *svm)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001888{
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02001889 struct kvm_vcpu *vcpu = &svm->vcpu;
Mike Dayd77c26f2007-10-08 09:02:08 -04001890 u32 io_info = svm->vmcb->control.exit_info_1; /* address size bug? */
Sean Christophersondca7f122018-03-08 08:57:27 -08001891 int size, in, string;
Avi Kivity039576c2007-03-20 12:46:50 +02001892 unsigned port;
Avi Kivity6aa8b732006-12-10 02:21:36 -08001893
Rusty Russelle756fc62007-07-30 20:07:08 +10001894 ++svm->vcpu.stat.io_exits;
Laurent Viviere70669a2007-08-05 10:36:40 +03001895 string = (io_info & SVM_IOIO_STR_MASK) != 0;
Avi Kivity039576c2007-03-20 12:46:50 +02001896 in = (io_info & SVM_IOIO_TYPE_MASK) != 0;
Tom Lendacky8370c3d2016-11-23 12:01:50 -05001897 if (string)
Sean Christopherson60fc3d02019-08-27 14:40:38 -07001898 return kvm_emulate_instruction(vcpu, 0);
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02001899
Avi Kivity039576c2007-03-20 12:46:50 +02001900 port = io_info >> 16;
1901 size = (io_info & SVM_IOIO_SIZE_MASK) >> SVM_IOIO_SIZE_SHIFT;
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02001902 svm->next_rip = svm->vmcb->control.exit_info_2;
Gleb Natapovcf8f70b2010-03-18 15:20:23 +02001903
Sean Christophersondca7f122018-03-08 08:57:27 -08001904 return kvm_fast_pio(&svm->vcpu, size, port, in);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001905}
1906
Avi Kivity851ba692009-08-24 11:10:17 +03001907static int nmi_interception(struct vcpu_svm *svm)
Joerg Roedelc47f0982008-04-30 17:56:00 +02001908{
1909 return 1;
1910}
1911
Avi Kivity851ba692009-08-24 11:10:17 +03001912static int intr_interception(struct vcpu_svm *svm)
Joerg Roedela0698052008-04-30 17:56:01 +02001913{
1914 ++svm->vcpu.stat.irq_exits;
1915 return 1;
1916}
1917
Avi Kivity851ba692009-08-24 11:10:17 +03001918static int nop_on_interception(struct vcpu_svm *svm)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001919{
1920 return 1;
1921}
1922
Avi Kivity851ba692009-08-24 11:10:17 +03001923static int halt_interception(struct vcpu_svm *svm)
Avi Kivity6aa8b732006-12-10 02:21:36 -08001924{
Rusty Russelle756fc62007-07-30 20:07:08 +10001925 return kvm_emulate_halt(&svm->vcpu);
Avi Kivity6aa8b732006-12-10 02:21:36 -08001926}
1927
Avi Kivity851ba692009-08-24 11:10:17 +03001928static int vmmcall_interception(struct vcpu_svm *svm)
Avi Kivity02e235b2007-02-19 14:37:47 +02001929{
Andrey Smetanin0d9c0552016-02-11 16:44:59 +03001930 return kvm_emulate_hypercall(&svm->vcpu);
Avi Kivity02e235b2007-02-19 14:37:47 +02001931}
1932
Avi Kivity851ba692009-08-24 11:10:17 +03001933static int vmload_interception(struct vcpu_svm *svm)
Alexander Graf55426752008-11-25 20:17:06 +01001934{
Joerg Roedel9966bf62009-08-07 11:49:40 +02001935 struct vmcb *nested_vmcb;
KarimAllah Ahmed8c5fbf12019-01-31 21:24:40 +01001936 struct kvm_host_map map;
Ladi Prosekb742c1e2017-06-22 09:05:26 +02001937 int ret;
Joerg Roedel9966bf62009-08-07 11:49:40 +02001938
Alexander Graf55426752008-11-25 20:17:06 +01001939 if (nested_svm_check_permissions(svm))
1940 return 1;
1941
KarimAllah Ahmed8c5fbf12019-01-31 21:24:40 +01001942 ret = kvm_vcpu_map(&svm->vcpu, gpa_to_gfn(svm->vmcb->save.rax), &map);
1943 if (ret) {
1944 if (ret == -EINVAL)
1945 kvm_inject_gp(&svm->vcpu, 0);
Joerg Roedel9966bf62009-08-07 11:49:40 +02001946 return 1;
KarimAllah Ahmed8c5fbf12019-01-31 21:24:40 +01001947 }
1948
1949 nested_vmcb = map.hva;
Joerg Roedel9966bf62009-08-07 11:49:40 +02001950
Ladi Prosekb742c1e2017-06-22 09:05:26 +02001951 ret = kvm_skip_emulated_instruction(&svm->vcpu);
Joerg Roedele3e9ed32011-04-06 12:30:03 +02001952
Joerg Roedel9966bf62009-08-07 11:49:40 +02001953 nested_svm_vmloadsave(nested_vmcb, svm->vmcb);
KarimAllah Ahmed8c5fbf12019-01-31 21:24:40 +01001954 kvm_vcpu_unmap(&svm->vcpu, &map, true);
Alexander Graf55426752008-11-25 20:17:06 +01001955
Ladi Prosekb742c1e2017-06-22 09:05:26 +02001956 return ret;
Alexander Graf55426752008-11-25 20:17:06 +01001957}
1958
Avi Kivity851ba692009-08-24 11:10:17 +03001959static int vmsave_interception(struct vcpu_svm *svm)
Alexander Graf55426752008-11-25 20:17:06 +01001960{
Joerg Roedel9966bf62009-08-07 11:49:40 +02001961 struct vmcb *nested_vmcb;
KarimAllah Ahmed8c5fbf12019-01-31 21:24:40 +01001962 struct kvm_host_map map;
Ladi Prosekb742c1e2017-06-22 09:05:26 +02001963 int ret;
Joerg Roedel9966bf62009-08-07 11:49:40 +02001964
Alexander Graf55426752008-11-25 20:17:06 +01001965 if (nested_svm_check_permissions(svm))
1966 return 1;
1967
KarimAllah Ahmed8c5fbf12019-01-31 21:24:40 +01001968 ret = kvm_vcpu_map(&svm->vcpu, gpa_to_gfn(svm->vmcb->save.rax), &map);
1969 if (ret) {
1970 if (ret == -EINVAL)
1971 kvm_inject_gp(&svm->vcpu, 0);
Joerg Roedel9966bf62009-08-07 11:49:40 +02001972 return 1;
KarimAllah Ahmed8c5fbf12019-01-31 21:24:40 +01001973 }
1974
1975 nested_vmcb = map.hva;
Joerg Roedel9966bf62009-08-07 11:49:40 +02001976
Ladi Prosekb742c1e2017-06-22 09:05:26 +02001977 ret = kvm_skip_emulated_instruction(&svm->vcpu);
Joerg Roedele3e9ed32011-04-06 12:30:03 +02001978
Joerg Roedel9966bf62009-08-07 11:49:40 +02001979 nested_svm_vmloadsave(svm->vmcb, nested_vmcb);
KarimAllah Ahmed8c5fbf12019-01-31 21:24:40 +01001980 kvm_vcpu_unmap(&svm->vcpu, &map, true);
Alexander Graf55426752008-11-25 20:17:06 +01001981
Ladi Prosekb742c1e2017-06-22 09:05:26 +02001982 return ret;
Alexander Graf55426752008-11-25 20:17:06 +01001983}
1984
Avi Kivity851ba692009-08-24 11:10:17 +03001985static int vmrun_interception(struct vcpu_svm *svm)
Alexander Graf3d6368e2008-11-25 20:17:07 +01001986{
Alexander Graf3d6368e2008-11-25 20:17:07 +01001987 if (nested_svm_check_permissions(svm))
1988 return 1;
1989
Vitaly Kuznetsove7134c12019-08-13 15:53:34 +02001990 return nested_svm_vmrun(svm);
Alexander Graf3d6368e2008-11-25 20:17:07 +01001991}
1992
Paolo Bonziniffdf7f92020-05-22 12:18:27 -04001993void svm_set_gif(struct vcpu_svm *svm, bool value)
1994{
1995 if (value) {
1996 /*
1997 * If VGIF is enabled, the STGI intercept is only added to
1998 * detect the opening of the SMI/NMI window; remove it now.
1999 * Likewise, clear the VINTR intercept, we will set it
2000 * again while processing KVM_REQ_EVENT if needed.
2001 */
2002 if (vgif_enabled(svm))
2003 clr_intercept(svm, INTERCEPT_STGI);
Paolo Bonzinifb7333d2020-06-08 07:11:47 -04002004 if (is_intercept(svm, INTERCEPT_VINTR))
Paolo Bonziniffdf7f92020-05-22 12:18:27 -04002005 svm_clear_vintr(svm);
2006
2007 enable_gif(svm);
2008 if (svm->vcpu.arch.smi_pending ||
2009 svm->vcpu.arch.nmi_pending ||
2010 kvm_cpu_has_injectable_intr(&svm->vcpu))
2011 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
2012 } else {
2013 disable_gif(svm);
2014
2015 /*
2016 * After a CLGI no interrupts should come. But if vGIF is
2017 * in use, we still rely on the VINTR intercept (rather than
2018 * STGI) to detect an open interrupt window.
2019 */
2020 if (!vgif_enabled(svm))
2021 svm_clear_vintr(svm);
2022 }
2023}
2024
Avi Kivity851ba692009-08-24 11:10:17 +03002025static int stgi_interception(struct vcpu_svm *svm)
Alexander Graf1371d902008-11-25 20:17:04 +01002026{
Ladi Prosekb742c1e2017-06-22 09:05:26 +02002027 int ret;
2028
Alexander Graf1371d902008-11-25 20:17:04 +01002029 if (nested_svm_check_permissions(svm))
2030 return 1;
2031
Ladi Prosekb742c1e2017-06-22 09:05:26 +02002032 ret = kvm_skip_emulated_instruction(&svm->vcpu);
Paolo Bonziniffdf7f92020-05-22 12:18:27 -04002033 svm_set_gif(svm, true);
Ladi Prosekb742c1e2017-06-22 09:05:26 +02002034 return ret;
Alexander Graf1371d902008-11-25 20:17:04 +01002035}
2036
Avi Kivity851ba692009-08-24 11:10:17 +03002037static int clgi_interception(struct vcpu_svm *svm)
Alexander Graf1371d902008-11-25 20:17:04 +01002038{
Ladi Prosekb742c1e2017-06-22 09:05:26 +02002039 int ret;
2040
Alexander Graf1371d902008-11-25 20:17:04 +01002041 if (nested_svm_check_permissions(svm))
2042 return 1;
2043
Ladi Prosekb742c1e2017-06-22 09:05:26 +02002044 ret = kvm_skip_emulated_instruction(&svm->vcpu);
Paolo Bonziniffdf7f92020-05-22 12:18:27 -04002045 svm_set_gif(svm, false);
Ladi Prosekb742c1e2017-06-22 09:05:26 +02002046 return ret;
Alexander Graf1371d902008-11-25 20:17:04 +01002047}
2048
Avi Kivity851ba692009-08-24 11:10:17 +03002049static int invlpga_interception(struct vcpu_svm *svm)
Alexander Grafff092382009-06-15 15:21:24 +02002050{
2051 struct kvm_vcpu *vcpu = &svm->vcpu;
Alexander Grafff092382009-06-15 15:21:24 +02002052
Sean Christophersonde3cd112019-04-30 10:36:17 -07002053 trace_kvm_invlpga(svm->vmcb->save.rip, kvm_rcx_read(&svm->vcpu),
2054 kvm_rax_read(&svm->vcpu));
Joerg Roedelec1ff792009-10-09 16:08:31 +02002055
Alexander Grafff092382009-06-15 15:21:24 +02002056 /* Let's treat INVLPGA the same as INVLPG (can be optimized!) */
Sean Christophersonde3cd112019-04-30 10:36:17 -07002057 kvm_mmu_invlpg(vcpu, kvm_rax_read(&svm->vcpu));
Alexander Grafff092382009-06-15 15:21:24 +02002058
Ladi Prosekb742c1e2017-06-22 09:05:26 +02002059 return kvm_skip_emulated_instruction(&svm->vcpu);
Alexander Grafff092382009-06-15 15:21:24 +02002060}
2061
Joerg Roedel532a46b2009-10-09 16:08:32 +02002062static int skinit_interception(struct vcpu_svm *svm)
2063{
Sean Christophersonde3cd112019-04-30 10:36:17 -07002064 trace_kvm_skinit(svm->vmcb->save.rip, kvm_rax_read(&svm->vcpu));
Joerg Roedel532a46b2009-10-09 16:08:32 +02002065
2066 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2067 return 1;
2068}
2069
David Kaplandab429a2015-03-02 13:43:37 -06002070static int wbinvd_interception(struct vcpu_svm *svm)
2071{
Kyle Huey6affcbe2016-11-29 12:40:40 -08002072 return kvm_emulate_wbinvd(&svm->vcpu);
David Kaplandab429a2015-03-02 13:43:37 -06002073}
2074
Joerg Roedel81dd35d2010-12-07 17:15:06 +01002075static int xsetbv_interception(struct vcpu_svm *svm)
2076{
2077 u64 new_bv = kvm_read_edx_eax(&svm->vcpu);
Sean Christophersonde3cd112019-04-30 10:36:17 -07002078 u32 index = kvm_rcx_read(&svm->vcpu);
Joerg Roedel81dd35d2010-12-07 17:15:06 +01002079
2080 if (kvm_set_xcr(&svm->vcpu, index, new_bv) == 0) {
Ladi Prosekb742c1e2017-06-22 09:05:26 +02002081 return kvm_skip_emulated_instruction(&svm->vcpu);
Joerg Roedel81dd35d2010-12-07 17:15:06 +01002082 }
2083
2084 return 1;
2085}
2086
Jim Mattson0cb84102019-09-19 15:59:17 -07002087static int rdpru_interception(struct vcpu_svm *svm)
2088{
2089 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2090 return 1;
2091}
2092
Avi Kivity851ba692009-08-24 11:10:17 +03002093static int task_switch_interception(struct vcpu_svm *svm)
Avi Kivity6aa8b732006-12-10 02:21:36 -08002094{
Izik Eidus37817f22008-03-24 23:14:53 +02002095 u16 tss_selector;
Gleb Natapov64a7ec02009-03-30 16:03:29 +03002096 int reason;
2097 int int_type = svm->vmcb->control.exit_int_info &
2098 SVM_EXITINTINFO_TYPE_MASK;
Gleb Natapov8317c292009-04-12 13:37:02 +03002099 int int_vec = svm->vmcb->control.exit_int_info & SVM_EVTINJ_VEC_MASK;
Gleb Natapovfe8e7f82009-04-23 17:03:48 +03002100 uint32_t type =
2101 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_TYPE_MASK;
2102 uint32_t idt_v =
2103 svm->vmcb->control.exit_int_info & SVM_EXITINTINFO_VALID;
Jan Kiszkae269fb22010-04-14 15:51:09 +02002104 bool has_error_code = false;
2105 u32 error_code = 0;
Izik Eidus37817f22008-03-24 23:14:53 +02002106
2107 tss_selector = (u16)svm->vmcb->control.exit_info_1;
Gleb Natapov64a7ec02009-03-30 16:03:29 +03002108
Izik Eidus37817f22008-03-24 23:14:53 +02002109 if (svm->vmcb->control.exit_info_2 &
2110 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_IRET))
Gleb Natapov64a7ec02009-03-30 16:03:29 +03002111 reason = TASK_SWITCH_IRET;
2112 else if (svm->vmcb->control.exit_info_2 &
2113 (1ULL << SVM_EXITINFOSHIFT_TS_REASON_JMP))
2114 reason = TASK_SWITCH_JMP;
Gleb Natapovfe8e7f82009-04-23 17:03:48 +03002115 else if (idt_v)
Gleb Natapov64a7ec02009-03-30 16:03:29 +03002116 reason = TASK_SWITCH_GATE;
2117 else
2118 reason = TASK_SWITCH_CALL;
2119
Gleb Natapovfe8e7f82009-04-23 17:03:48 +03002120 if (reason == TASK_SWITCH_GATE) {
2121 switch (type) {
2122 case SVM_EXITINTINFO_TYPE_NMI:
2123 svm->vcpu.arch.nmi_injected = false;
2124 break;
2125 case SVM_EXITINTINFO_TYPE_EXEPT:
Jan Kiszkae269fb22010-04-14 15:51:09 +02002126 if (svm->vmcb->control.exit_info_2 &
2127 (1ULL << SVM_EXITINFOSHIFT_TS_HAS_ERROR_CODE)) {
2128 has_error_code = true;
2129 error_code =
2130 (u32)svm->vmcb->control.exit_info_2;
2131 }
Gleb Natapovfe8e7f82009-04-23 17:03:48 +03002132 kvm_clear_exception_queue(&svm->vcpu);
2133 break;
2134 case SVM_EXITINTINFO_TYPE_INTR:
2135 kvm_clear_interrupt_queue(&svm->vcpu);
2136 break;
2137 default:
2138 break;
2139 }
2140 }
Gleb Natapov64a7ec02009-03-30 16:03:29 +03002141
Gleb Natapov8317c292009-04-12 13:37:02 +03002142 if (reason != TASK_SWITCH_GATE ||
2143 int_type == SVM_EXITINTINFO_TYPE_SOFT ||
2144 (int_type == SVM_EXITINTINFO_TYPE_EXEPT &&
Vitaly Kuznetsovf8ea7c62019-08-13 15:53:30 +02002145 (int_vec == OF_VECTOR || int_vec == BP_VECTOR))) {
Sean Christopherson60fc3d02019-08-27 14:40:38 -07002146 if (!skip_emulated_instruction(&svm->vcpu))
Sean Christopherson738fece2019-08-27 14:40:34 -07002147 return 0;
Vitaly Kuznetsovf8ea7c62019-08-13 15:53:30 +02002148 }
Gleb Natapov64a7ec02009-03-30 16:03:29 +03002149
Kevin Wolf7f3d35f2012-02-08 14:34:38 +01002150 if (int_type != SVM_EXITINTINFO_TYPE_SOFT)
2151 int_vec = -1;
2152
Sean Christopherson10517782019-08-27 14:40:35 -07002153 return kvm_task_switch(&svm->vcpu, tss_selector, int_vec, reason,
Sean Christopherson60fc3d02019-08-27 14:40:38 -07002154 has_error_code, error_code);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002155}
2156
Avi Kivity851ba692009-08-24 11:10:17 +03002157static int cpuid_interception(struct vcpu_svm *svm)
Avi Kivity6aa8b732006-12-10 02:21:36 -08002158{
Kyle Huey6a908b62016-11-29 12:40:37 -08002159 return kvm_emulate_cpuid(&svm->vcpu);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002160}
2161
Avi Kivity851ba692009-08-24 11:10:17 +03002162static int iret_interception(struct vcpu_svm *svm)
Gleb Natapov95ba8273132009-04-21 17:45:08 +03002163{
2164 ++svm->vcpu.stat.nmi_window_exits;
Joerg Roedel8a05a1b82010-11-30 18:04:00 +01002165 clr_intercept(svm, INTERCEPT_IRET);
Gleb Natapov44c11432009-05-11 13:35:52 +03002166 svm->vcpu.arch.hflags |= HF_IRET_MASK;
Avi Kivitybd3d1ec2011-02-03 15:29:52 +02002167 svm->nmi_iret_rip = kvm_rip_read(&svm->vcpu);
Radim Krčmářf303b4c2014-01-17 20:52:42 +01002168 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
Gleb Natapov95ba8273132009-04-21 17:45:08 +03002169 return 1;
2170}
2171
Avi Kivity851ba692009-08-24 11:10:17 +03002172static int invlpg_interception(struct vcpu_svm *svm)
Marcelo Tosattia7052892008-09-23 13:18:35 -03002173{
Andre Przywaradf4f31082010-12-21 11:12:06 +01002174 if (!static_cpu_has(X86_FEATURE_DECODEASSISTS))
Sean Christopherson60fc3d02019-08-27 14:40:38 -07002175 return kvm_emulate_instruction(&svm->vcpu, 0);
Andre Przywaradf4f31082010-12-21 11:12:06 +01002176
2177 kvm_mmu_invlpg(&svm->vcpu, svm->vmcb->control.exit_info_1);
Ladi Prosekb742c1e2017-06-22 09:05:26 +02002178 return kvm_skip_emulated_instruction(&svm->vcpu);
Marcelo Tosattia7052892008-09-23 13:18:35 -03002179}
2180
Avi Kivity851ba692009-08-24 11:10:17 +03002181static int emulate_on_interception(struct vcpu_svm *svm)
Avi Kivity6aa8b732006-12-10 02:21:36 -08002182{
Sean Christopherson60fc3d02019-08-27 14:40:38 -07002183 return kvm_emulate_instruction(&svm->vcpu, 0);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002184}
2185
Brijesh Singh7607b712018-02-19 10:14:44 -06002186static int rsm_interception(struct vcpu_svm *svm)
2187{
Sean Christopherson60fc3d02019-08-27 14:40:38 -07002188 return kvm_emulate_instruction_from_buffer(&svm->vcpu, rsm_ins_bytes, 2);
Brijesh Singh7607b712018-02-19 10:14:44 -06002189}
2190
Avi Kivity332b56e2011-11-10 14:57:24 +02002191static int rdpmc_interception(struct vcpu_svm *svm)
2192{
2193 int err;
2194
Paolo Bonzinid647eb62019-06-20 14:13:33 +02002195 if (!nrips)
Avi Kivity332b56e2011-11-10 14:57:24 +02002196 return emulate_on_interception(svm);
2197
2198 err = kvm_rdpmc(&svm->vcpu);
Kyle Huey6affcbe2016-11-29 12:40:40 -08002199 return kvm_complete_insn_gp(&svm->vcpu, err);
Avi Kivity332b56e2011-11-10 14:57:24 +02002200}
2201
Xiubo Li52eb5a62015-03-13 17:39:45 +08002202static bool check_selective_cr0_intercepted(struct vcpu_svm *svm,
2203 unsigned long val)
Joerg Roedel628afd22011-04-04 12:39:36 +02002204{
2205 unsigned long cr0 = svm->vcpu.arch.cr0;
2206 bool ret = false;
2207 u64 intercept;
2208
Paolo Bonzinie670bf62020-05-13 13:16:12 -04002209 intercept = svm->nested.ctl.intercept;
Joerg Roedel628afd22011-04-04 12:39:36 +02002210
2211 if (!is_guest_mode(&svm->vcpu) ||
2212 (!(intercept & (1ULL << INTERCEPT_SELECTIVE_CR0))))
2213 return false;
2214
2215 cr0 &= ~SVM_CR0_SELECTIVE_MASK;
2216 val &= ~SVM_CR0_SELECTIVE_MASK;
2217
2218 if (cr0 ^ val) {
2219 svm->vmcb->control.exit_code = SVM_EXIT_CR0_SEL_WRITE;
2220 ret = (nested_svm_exit_handled(svm) == NESTED_EXIT_DONE);
2221 }
2222
2223 return ret;
2224}
2225
Andre Przywara7ff76d52010-12-21 11:12:04 +01002226#define CR_VALID (1ULL << 63)
2227
2228static int cr_interception(struct vcpu_svm *svm)
2229{
2230 int reg, cr;
2231 unsigned long val;
2232 int err;
2233
2234 if (!static_cpu_has(X86_FEATURE_DECODEASSISTS))
2235 return emulate_on_interception(svm);
2236
2237 if (unlikely((svm->vmcb->control.exit_info_1 & CR_VALID) == 0))
2238 return emulate_on_interception(svm);
2239
2240 reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK;
David Kaplan5e575182015-03-06 14:44:35 -06002241 if (svm->vmcb->control.exit_code == SVM_EXIT_CR0_SEL_WRITE)
2242 cr = SVM_EXIT_WRITE_CR0 - SVM_EXIT_READ_CR0;
2243 else
2244 cr = svm->vmcb->control.exit_code - SVM_EXIT_READ_CR0;
Andre Przywara7ff76d52010-12-21 11:12:04 +01002245
2246 err = 0;
2247 if (cr >= 16) { /* mov to cr */
2248 cr -= 16;
2249 val = kvm_register_read(&svm->vcpu, reg);
2250 switch (cr) {
2251 case 0:
Joerg Roedel628afd22011-04-04 12:39:36 +02002252 if (!check_selective_cr0_intercepted(svm, val))
2253 err = kvm_set_cr0(&svm->vcpu, val);
Joerg Roedel977b2d02011-04-18 11:42:52 +02002254 else
2255 return 1;
2256
Andre Przywara7ff76d52010-12-21 11:12:04 +01002257 break;
2258 case 3:
2259 err = kvm_set_cr3(&svm->vcpu, val);
2260 break;
2261 case 4:
2262 err = kvm_set_cr4(&svm->vcpu, val);
2263 break;
2264 case 8:
2265 err = kvm_set_cr8(&svm->vcpu, val);
2266 break;
2267 default:
2268 WARN(1, "unhandled write to CR%d", cr);
2269 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2270 return 1;
2271 }
2272 } else { /* mov from cr */
2273 switch (cr) {
2274 case 0:
2275 val = kvm_read_cr0(&svm->vcpu);
2276 break;
2277 case 2:
2278 val = svm->vcpu.arch.cr2;
2279 break;
2280 case 3:
Avi Kivity9f8fe502010-12-05 17:30:00 +02002281 val = kvm_read_cr3(&svm->vcpu);
Andre Przywara7ff76d52010-12-21 11:12:04 +01002282 break;
2283 case 4:
2284 val = kvm_read_cr4(&svm->vcpu);
2285 break;
2286 case 8:
2287 val = kvm_get_cr8(&svm->vcpu);
2288 break;
2289 default:
2290 WARN(1, "unhandled read from CR%d", cr);
2291 kvm_queue_exception(&svm->vcpu, UD_VECTOR);
2292 return 1;
2293 }
2294 kvm_register_write(&svm->vcpu, reg, val);
2295 }
Kyle Huey6affcbe2016-11-29 12:40:40 -08002296 return kvm_complete_insn_gp(&svm->vcpu, err);
Andre Przywara7ff76d52010-12-21 11:12:04 +01002297}
2298
Andre Przywaracae37972010-12-21 11:12:05 +01002299static int dr_interception(struct vcpu_svm *svm)
2300{
2301 int reg, dr;
2302 unsigned long val;
Andre Przywaracae37972010-12-21 11:12:05 +01002303
Paolo Bonzinifacb0132014-02-21 10:32:27 +01002304 if (svm->vcpu.guest_debug == 0) {
2305 /*
2306 * No more DR vmexits; force a reload of the debug registers
2307 * and reenter on this instruction. The next vmexit will
2308 * retrieve the full state of the debug registers.
2309 */
2310 clr_dr_intercepts(svm);
2311 svm->vcpu.arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
2312 return 1;
2313 }
2314
Andre Przywaracae37972010-12-21 11:12:05 +01002315 if (!boot_cpu_has(X86_FEATURE_DECODEASSISTS))
2316 return emulate_on_interception(svm);
2317
2318 reg = svm->vmcb->control.exit_info_1 & SVM_EXITINFO_REG_MASK;
2319 dr = svm->vmcb->control.exit_code - SVM_EXIT_READ_DR0;
2320
2321 if (dr >= 16) { /* mov to DRn */
Nadav Amit16f8a6f2014-10-03 01:10:05 +03002322 if (!kvm_require_dr(&svm->vcpu, dr - 16))
2323 return 1;
Andre Przywaracae37972010-12-21 11:12:05 +01002324 val = kvm_register_read(&svm->vcpu, reg);
2325 kvm_set_dr(&svm->vcpu, dr - 16, val);
2326 } else {
Nadav Amit16f8a6f2014-10-03 01:10:05 +03002327 if (!kvm_require_dr(&svm->vcpu, dr))
2328 return 1;
2329 kvm_get_dr(&svm->vcpu, dr, &val);
2330 kvm_register_write(&svm->vcpu, reg, val);
Andre Przywaracae37972010-12-21 11:12:05 +01002331 }
2332
Ladi Prosekb742c1e2017-06-22 09:05:26 +02002333 return kvm_skip_emulated_instruction(&svm->vcpu);
Andre Przywaracae37972010-12-21 11:12:05 +01002334}
2335
Avi Kivity851ba692009-08-24 11:10:17 +03002336static int cr8_write_interception(struct vcpu_svm *svm)
Joerg Roedel1d075432007-12-06 21:02:25 +01002337{
Avi Kivity851ba692009-08-24 11:10:17 +03002338 struct kvm_run *kvm_run = svm->vcpu.run;
Andre Przywaraeea1cff2010-12-21 11:12:00 +01002339 int r;
Avi Kivity851ba692009-08-24 11:10:17 +03002340
Gleb Natapov0a5fff192009-04-21 17:45:06 +03002341 u8 cr8_prev = kvm_get_cr8(&svm->vcpu);
2342 /* instruction emulation calls kvm_set_cr8() */
Andre Przywara7ff76d52010-12-21 11:12:04 +01002343 r = cr_interception(svm);
Paolo Bonzini35754c92015-07-29 12:05:37 +02002344 if (lapic_in_kernel(&svm->vcpu))
Andre Przywara7ff76d52010-12-21 11:12:04 +01002345 return r;
Gleb Natapov0a5fff192009-04-21 17:45:06 +03002346 if (cr8_prev <= kvm_get_cr8(&svm->vcpu))
Andre Przywara7ff76d52010-12-21 11:12:04 +01002347 return r;
Joerg Roedel1d075432007-12-06 21:02:25 +01002348 kvm_run->exit_reason = KVM_EXIT_SET_TPR;
2349 return 0;
2350}
2351
Tom Lendacky801e4592018-02-21 13:39:51 -06002352static int svm_get_msr_feature(struct kvm_msr_entry *msr)
2353{
Tom Lendackyd1d93fa2018-02-24 00:18:20 +01002354 msr->data = 0;
2355
2356 switch (msr->index) {
2357 case MSR_F10H_DECFG:
2358 if (boot_cpu_has(X86_FEATURE_LFENCE_RDTSC))
2359 msr->data |= MSR_F10H_DECFG_LFENCE_SERIALIZE;
2360 break;
2361 default:
2362 return 1;
2363 }
2364
2365 return 0;
Tom Lendacky801e4592018-02-21 13:39:51 -06002366}
2367
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002368static int svm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
Avi Kivity6aa8b732006-12-10 02:21:36 -08002369{
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -04002370 struct vcpu_svm *svm = to_svm(vcpu);
2371
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002372 switch (msr_info->index) {
Brian Gerst8c065852010-07-17 09:03:26 -04002373 case MSR_STAR:
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002374 msr_info->data = svm->vmcb->save.star;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002375 break;
Avi Kivity0e859ca2006-12-22 01:05:08 -08002376#ifdef CONFIG_X86_64
Avi Kivity6aa8b732006-12-10 02:21:36 -08002377 case MSR_LSTAR:
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002378 msr_info->data = svm->vmcb->save.lstar;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002379 break;
2380 case MSR_CSTAR:
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002381 msr_info->data = svm->vmcb->save.cstar;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002382 break;
2383 case MSR_KERNEL_GS_BASE:
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002384 msr_info->data = svm->vmcb->save.kernel_gs_base;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002385 break;
2386 case MSR_SYSCALL_MASK:
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002387 msr_info->data = svm->vmcb->save.sfmask;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002388 break;
2389#endif
2390 case MSR_IA32_SYSENTER_CS:
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002391 msr_info->data = svm->vmcb->save.sysenter_cs;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002392 break;
2393 case MSR_IA32_SYSENTER_EIP:
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002394 msr_info->data = svm->sysenter_eip;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002395 break;
2396 case MSR_IA32_SYSENTER_ESP:
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002397 msr_info->data = svm->sysenter_esp;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002398 break;
Paolo Bonzini46896c72015-11-12 14:49:16 +01002399 case MSR_TSC_AUX:
2400 if (!boot_cpu_has(X86_FEATURE_RDTSCP))
2401 return 1;
2402 msr_info->data = svm->tsc_aux;
2403 break;
Joerg Roedele0231712010-02-24 18:59:10 +01002404 /*
2405 * Nobody will change the following 5 values in the VMCB so we can
2406 * safely return them on rdmsr. They will always be 0 until LBRV is
2407 * implemented.
2408 */
Joerg Roedela2938c82008-02-13 16:30:28 +01002409 case MSR_IA32_DEBUGCTLMSR:
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002410 msr_info->data = svm->vmcb->save.dbgctl;
Joerg Roedela2938c82008-02-13 16:30:28 +01002411 break;
2412 case MSR_IA32_LASTBRANCHFROMIP:
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002413 msr_info->data = svm->vmcb->save.br_from;
Joerg Roedela2938c82008-02-13 16:30:28 +01002414 break;
2415 case MSR_IA32_LASTBRANCHTOIP:
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002416 msr_info->data = svm->vmcb->save.br_to;
Joerg Roedela2938c82008-02-13 16:30:28 +01002417 break;
2418 case MSR_IA32_LASTINTFROMIP:
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002419 msr_info->data = svm->vmcb->save.last_excp_from;
Joerg Roedela2938c82008-02-13 16:30:28 +01002420 break;
2421 case MSR_IA32_LASTINTTOIP:
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002422 msr_info->data = svm->vmcb->save.last_excp_to;
Joerg Roedela2938c82008-02-13 16:30:28 +01002423 break;
Alexander Grafb286d5d2008-11-25 20:17:05 +01002424 case MSR_VM_HSAVE_PA:
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002425 msr_info->data = svm->nested.hsave_msr;
Alexander Grafb286d5d2008-11-25 20:17:05 +01002426 break;
Joerg Roedeleb6f3022008-11-25 20:17:09 +01002427 case MSR_VM_CR:
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002428 msr_info->data = svm->nested.vm_cr_msr;
Joerg Roedeleb6f3022008-11-25 20:17:09 +01002429 break;
KarimAllah Ahmedb2ac58f2018-02-03 15:56:23 +01002430 case MSR_IA32_SPEC_CTRL:
2431 if (!msr_info->host_initiated &&
Paolo Bonzinidf7e8812020-02-05 16:10:52 +01002432 !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL) &&
2433 !guest_cpuid_has(vcpu, X86_FEATURE_AMD_STIBP) &&
Konrad Rzeszutek Wilk6ac2f492018-06-01 10:59:20 -04002434 !guest_cpuid_has(vcpu, X86_FEATURE_AMD_IBRS) &&
2435 !guest_cpuid_has(vcpu, X86_FEATURE_AMD_SSBD))
KarimAllah Ahmedb2ac58f2018-02-03 15:56:23 +01002436 return 1;
2437
2438 msr_info->data = svm->spec_ctrl;
2439 break;
Tom Lendackybc226f02018-05-10 22:06:39 +02002440 case MSR_AMD64_VIRT_SPEC_CTRL:
2441 if (!msr_info->host_initiated &&
2442 !guest_cpuid_has(vcpu, X86_FEATURE_VIRT_SSBD))
2443 return 1;
2444
2445 msr_info->data = svm->virt_spec_ctrl;
2446 break;
Borislav Petkovae8b7872015-11-23 11:12:23 +01002447 case MSR_F15H_IC_CFG: {
2448
2449 int family, model;
2450
2451 family = guest_cpuid_family(vcpu);
2452 model = guest_cpuid_model(vcpu);
2453
2454 if (family < 0 || model < 0)
2455 return kvm_get_msr_common(vcpu, msr_info);
2456
2457 msr_info->data = 0;
2458
2459 if (family == 0x15 &&
2460 (model >= 0x2 && model < 0x20))
2461 msr_info->data = 0x1E;
2462 }
2463 break;
Tom Lendackyd1d93fa2018-02-24 00:18:20 +01002464 case MSR_F10H_DECFG:
2465 msr_info->data = svm->msr_decfg;
2466 break;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002467 default:
Paolo Bonzini609e36d2015-04-08 15:30:38 +02002468 return kvm_get_msr_common(vcpu, msr_info);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002469 }
2470 return 0;
2471}
2472
Avi Kivity851ba692009-08-24 11:10:17 +03002473static int rdmsr_interception(struct vcpu_svm *svm)
Avi Kivity6aa8b732006-12-10 02:21:36 -08002474{
Sean Christopherson1edce0a2019-09-05 14:22:55 -07002475 return kvm_emulate_rdmsr(&svm->vcpu);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002476}
2477
Joerg Roedel4a810182010-02-24 18:59:15 +01002478static int svm_set_vm_cr(struct kvm_vcpu *vcpu, u64 data)
2479{
2480 struct vcpu_svm *svm = to_svm(vcpu);
2481 int svm_dis, chg_mask;
2482
2483 if (data & ~SVM_VM_CR_VALID_MASK)
2484 return 1;
2485
2486 chg_mask = SVM_VM_CR_VALID_MASK;
2487
2488 if (svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK)
2489 chg_mask &= ~(SVM_VM_CR_SVM_LOCK_MASK | SVM_VM_CR_SVM_DIS_MASK);
2490
2491 svm->nested.vm_cr_msr &= ~chg_mask;
2492 svm->nested.vm_cr_msr |= (data & chg_mask);
2493
2494 svm_dis = svm->nested.vm_cr_msr & SVM_VM_CR_SVM_DIS_MASK;
2495
2496 /* check for svm_disable while efer.svme is set */
2497 if (svm_dis && (vcpu->arch.efer & EFER_SVME))
2498 return 1;
2499
2500 return 0;
2501}
2502
Will Auld8fe8ab42012-11-29 12:42:12 -08002503static int svm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
Avi Kivity6aa8b732006-12-10 02:21:36 -08002504{
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -04002505 struct vcpu_svm *svm = to_svm(vcpu);
2506
Will Auld8fe8ab42012-11-29 12:42:12 -08002507 u32 ecx = msr->index;
2508 u64 data = msr->data;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002509 switch (ecx) {
Paolo Bonzini15038e12017-10-26 09:13:27 +02002510 case MSR_IA32_CR_PAT:
2511 if (!kvm_mtrr_valid(vcpu, MSR_IA32_CR_PAT, data))
2512 return 1;
2513 vcpu->arch.pat = data;
2514 svm->vmcb->save.g_pat = data;
2515 mark_dirty(svm->vmcb, VMCB_NPT);
2516 break;
KarimAllah Ahmedb2ac58f2018-02-03 15:56:23 +01002517 case MSR_IA32_SPEC_CTRL:
2518 if (!msr->host_initiated &&
Paolo Bonzinidf7e8812020-02-05 16:10:52 +01002519 !guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL) &&
2520 !guest_cpuid_has(vcpu, X86_FEATURE_AMD_STIBP) &&
Konrad Rzeszutek Wilk6ac2f492018-06-01 10:59:20 -04002521 !guest_cpuid_has(vcpu, X86_FEATURE_AMD_IBRS) &&
2522 !guest_cpuid_has(vcpu, X86_FEATURE_AMD_SSBD))
KarimAllah Ahmedb2ac58f2018-02-03 15:56:23 +01002523 return 1;
2524
Paolo Bonzini6441fa62020-01-20 16:33:06 +01002525 if (data & ~kvm_spec_ctrl_valid_bits(vcpu))
KarimAllah Ahmedb2ac58f2018-02-03 15:56:23 +01002526 return 1;
2527
2528 svm->spec_ctrl = data;
KarimAllah Ahmedb2ac58f2018-02-03 15:56:23 +01002529 if (!data)
2530 break;
2531
2532 /*
2533 * For non-nested:
2534 * When it's written (to non-zero) for the first time, pass
2535 * it through.
2536 *
2537 * For nested:
2538 * The handling of the MSR bitmap for L2 guests is done in
2539 * nested_svm_vmrun_msrpm.
2540 * We update the L1 MSR bit as well since it will end up
2541 * touching the MSR anyway now.
2542 */
2543 set_msr_interception(svm->msrpm, MSR_IA32_SPEC_CTRL, 1, 1);
2544 break;
Ashok Raj15d45072018-02-01 22:59:43 +01002545 case MSR_IA32_PRED_CMD:
2546 if (!msr->host_initiated &&
Borislav Petkove7c587d2018-05-02 18:15:14 +02002547 !guest_cpuid_has(vcpu, X86_FEATURE_AMD_IBPB))
Ashok Raj15d45072018-02-01 22:59:43 +01002548 return 1;
2549
2550 if (data & ~PRED_CMD_IBPB)
2551 return 1;
Paolo Bonzini6441fa62020-01-20 16:33:06 +01002552 if (!boot_cpu_has(X86_FEATURE_AMD_IBPB))
2553 return 1;
Ashok Raj15d45072018-02-01 22:59:43 +01002554 if (!data)
2555 break;
2556
2557 wrmsrl(MSR_IA32_PRED_CMD, PRED_CMD_IBPB);
Ashok Raj15d45072018-02-01 22:59:43 +01002558 set_msr_interception(svm->msrpm, MSR_IA32_PRED_CMD, 0, 1);
2559 break;
Tom Lendackybc226f02018-05-10 22:06:39 +02002560 case MSR_AMD64_VIRT_SPEC_CTRL:
2561 if (!msr->host_initiated &&
2562 !guest_cpuid_has(vcpu, X86_FEATURE_VIRT_SSBD))
2563 return 1;
2564
2565 if (data & ~SPEC_CTRL_SSBD)
2566 return 1;
2567
2568 svm->virt_spec_ctrl = data;
2569 break;
Brian Gerst8c065852010-07-17 09:03:26 -04002570 case MSR_STAR:
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -04002571 svm->vmcb->save.star = data;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002572 break;
Robert P. J. Day49b14f22007-01-29 13:19:50 -08002573#ifdef CONFIG_X86_64
Avi Kivity6aa8b732006-12-10 02:21:36 -08002574 case MSR_LSTAR:
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -04002575 svm->vmcb->save.lstar = data;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002576 break;
2577 case MSR_CSTAR:
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -04002578 svm->vmcb->save.cstar = data;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002579 break;
2580 case MSR_KERNEL_GS_BASE:
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -04002581 svm->vmcb->save.kernel_gs_base = data;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002582 break;
2583 case MSR_SYSCALL_MASK:
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -04002584 svm->vmcb->save.sfmask = data;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002585 break;
2586#endif
2587 case MSR_IA32_SYSENTER_CS:
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -04002588 svm->vmcb->save.sysenter_cs = data;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002589 break;
2590 case MSR_IA32_SYSENTER_EIP:
Andre Przywara017cb992009-05-28 11:56:31 +02002591 svm->sysenter_eip = data;
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -04002592 svm->vmcb->save.sysenter_eip = data;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002593 break;
2594 case MSR_IA32_SYSENTER_ESP:
Andre Przywara017cb992009-05-28 11:56:31 +02002595 svm->sysenter_esp = data;
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -04002596 svm->vmcb->save.sysenter_esp = data;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002597 break;
Paolo Bonzini46896c72015-11-12 14:49:16 +01002598 case MSR_TSC_AUX:
2599 if (!boot_cpu_has(X86_FEATURE_RDTSCP))
2600 return 1;
2601
2602 /*
2603 * This is rare, so we update the MSR here instead of using
2604 * direct_access_msrs. Doing that would require a rdmsr in
2605 * svm_vcpu_put.
2606 */
2607 svm->tsc_aux = data;
2608 wrmsrl(MSR_TSC_AUX, svm->tsc_aux);
2609 break;
Joerg Roedela2938c82008-02-13 16:30:28 +01002610 case MSR_IA32_DEBUGCTLMSR:
Avi Kivity2a6b20b2010-11-09 16:15:42 +02002611 if (!boot_cpu_has(X86_FEATURE_LBRV)) {
Christoffer Dalla737f252012-06-03 21:17:48 +03002612 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTL 0x%llx, nop\n",
2613 __func__, data);
Joerg Roedel24e09cb2008-02-13 18:58:47 +01002614 break;
2615 }
2616 if (data & DEBUGCTL_RESERVED_BITS)
2617 return 1;
2618
2619 svm->vmcb->save.dbgctl = data;
Joerg Roedelb53ba3f2010-12-03 11:45:59 +01002620 mark_dirty(svm->vmcb, VMCB_LBR);
Joerg Roedel24e09cb2008-02-13 18:58:47 +01002621 if (data & (1ULL<<0))
2622 svm_enable_lbrv(svm);
2623 else
2624 svm_disable_lbrv(svm);
Joerg Roedela2938c82008-02-13 16:30:28 +01002625 break;
Alexander Grafb286d5d2008-11-25 20:17:05 +01002626 case MSR_VM_HSAVE_PA:
Joerg Roedele6aa9ab2009-08-07 11:49:33 +02002627 svm->nested.hsave_msr = data;
Alexander Grafb286d5d2008-11-25 20:17:05 +01002628 break;
Alexander Graf3c5d0a42009-06-15 15:21:23 +02002629 case MSR_VM_CR:
Joerg Roedel4a810182010-02-24 18:59:15 +01002630 return svm_set_vm_cr(vcpu, data);
Alexander Graf3c5d0a42009-06-15 15:21:23 +02002631 case MSR_VM_IGNNE:
Christoffer Dalla737f252012-06-03 21:17:48 +03002632 vcpu_unimpl(vcpu, "unimplemented wrmsr: 0x%x data 0x%llx\n", ecx, data);
Alexander Graf3c5d0a42009-06-15 15:21:23 +02002633 break;
Tom Lendackyd1d93fa2018-02-24 00:18:20 +01002634 case MSR_F10H_DECFG: {
2635 struct kvm_msr_entry msr_entry;
2636
2637 msr_entry.index = msr->index;
2638 if (svm_get_msr_feature(&msr_entry))
2639 return 1;
2640
2641 /* Check the supported bits */
2642 if (data & ~msr_entry.data)
2643 return 1;
2644
2645 /* Don't allow the guest to change a bit, #GP */
2646 if (!msr->host_initiated && (data ^ msr_entry.data))
2647 return 1;
2648
2649 svm->msr_decfg = data;
2650 break;
2651 }
Suravee Suthikulpanit44a95da2016-05-04 14:09:46 -05002652 case MSR_IA32_APICBASE:
2653 if (kvm_vcpu_apicv_active(vcpu))
2654 avic_update_vapic_bar(to_svm(vcpu), data);
Gustavo A. R. Silvab2869f22019-01-25 12:23:17 -06002655 /* Fall through */
Avi Kivity6aa8b732006-12-10 02:21:36 -08002656 default:
Will Auld8fe8ab42012-11-29 12:42:12 -08002657 return kvm_set_msr_common(vcpu, msr);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002658 }
2659 return 0;
2660}
2661
Avi Kivity851ba692009-08-24 11:10:17 +03002662static int wrmsr_interception(struct vcpu_svm *svm)
Avi Kivity6aa8b732006-12-10 02:21:36 -08002663{
Sean Christopherson1edce0a2019-09-05 14:22:55 -07002664 return kvm_emulate_wrmsr(&svm->vcpu);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002665}
2666
Avi Kivity851ba692009-08-24 11:10:17 +03002667static int msr_interception(struct vcpu_svm *svm)
Avi Kivity6aa8b732006-12-10 02:21:36 -08002668{
Rusty Russelle756fc62007-07-30 20:07:08 +10002669 if (svm->vmcb->control.exit_info_1)
Avi Kivity851ba692009-08-24 11:10:17 +03002670 return wrmsr_interception(svm);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002671 else
Avi Kivity851ba692009-08-24 11:10:17 +03002672 return rdmsr_interception(svm);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002673}
2674
Avi Kivity851ba692009-08-24 11:10:17 +03002675static int interrupt_window_interception(struct vcpu_svm *svm)
Dor Laorc1150d82007-01-05 16:36:24 -08002676{
Avi Kivity3842d132010-07-27 12:30:24 +03002677 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
Alexander Graff0b85052008-11-25 20:17:01 +01002678 svm_clear_vintr(svm);
Suravee Suthikulpanitf3515dc2019-11-14 14:15:15 -06002679
2680 /*
2681 * For AVIC, the only reason to end up here is ExtINTs.
2682 * In this case AVIC was temporarily disabled for
2683 * requesting the IRQ window and we have to re-enable it.
2684 */
2685 svm_toggle_avic_for_irq_window(&svm->vcpu, true);
2686
Jason Wang675acb72012-03-08 18:07:56 +08002687 ++svm->vcpu.stat.irq_window_exits;
Dor Laorc1150d82007-01-05 16:36:24 -08002688 return 1;
2689}
2690
Mark Langsdorf565d0992009-10-06 14:25:02 -05002691static int pause_interception(struct vcpu_svm *svm)
2692{
Longpeng(Mike)de63ad42017-08-08 12:05:33 +08002693 struct kvm_vcpu *vcpu = &svm->vcpu;
2694 bool in_kernel = (svm_get_cpl(vcpu) == 0);
2695
Babu Moger8566ac82018-03-16 16:37:26 -04002696 if (pause_filter_thresh)
2697 grow_ple_window(vcpu);
2698
Longpeng(Mike)de63ad42017-08-08 12:05:33 +08002699 kvm_vcpu_on_spin(vcpu, in_kernel);
Mark Langsdorf565d0992009-10-06 14:25:02 -05002700 return 1;
2701}
2702
Gabriel L. Somlo87c00572014-05-07 16:52:13 -04002703static int nop_interception(struct vcpu_svm *svm)
2704{
Ladi Prosekb742c1e2017-06-22 09:05:26 +02002705 return kvm_skip_emulated_instruction(&(svm->vcpu));
Gabriel L. Somlo87c00572014-05-07 16:52:13 -04002706}
2707
2708static int monitor_interception(struct vcpu_svm *svm)
2709{
2710 printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
2711 return nop_interception(svm);
2712}
2713
2714static int mwait_interception(struct vcpu_svm *svm)
2715{
2716 printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
2717 return nop_interception(svm);
2718}
2719
Mathias Krause09941fb2012-08-30 01:30:20 +02002720static int (*const svm_exit_handlers[])(struct vcpu_svm *svm) = {
Andre Przywara7ff76d52010-12-21 11:12:04 +01002721 [SVM_EXIT_READ_CR0] = cr_interception,
2722 [SVM_EXIT_READ_CR3] = cr_interception,
2723 [SVM_EXIT_READ_CR4] = cr_interception,
2724 [SVM_EXIT_READ_CR8] = cr_interception,
David Kaplan5e575182015-03-06 14:44:35 -06002725 [SVM_EXIT_CR0_SEL_WRITE] = cr_interception,
Joerg Roedel628afd22011-04-04 12:39:36 +02002726 [SVM_EXIT_WRITE_CR0] = cr_interception,
Andre Przywara7ff76d52010-12-21 11:12:04 +01002727 [SVM_EXIT_WRITE_CR3] = cr_interception,
2728 [SVM_EXIT_WRITE_CR4] = cr_interception,
Joerg Roedele0231712010-02-24 18:59:10 +01002729 [SVM_EXIT_WRITE_CR8] = cr8_write_interception,
Andre Przywaracae37972010-12-21 11:12:05 +01002730 [SVM_EXIT_READ_DR0] = dr_interception,
2731 [SVM_EXIT_READ_DR1] = dr_interception,
2732 [SVM_EXIT_READ_DR2] = dr_interception,
2733 [SVM_EXIT_READ_DR3] = dr_interception,
2734 [SVM_EXIT_READ_DR4] = dr_interception,
2735 [SVM_EXIT_READ_DR5] = dr_interception,
2736 [SVM_EXIT_READ_DR6] = dr_interception,
2737 [SVM_EXIT_READ_DR7] = dr_interception,
2738 [SVM_EXIT_WRITE_DR0] = dr_interception,
2739 [SVM_EXIT_WRITE_DR1] = dr_interception,
2740 [SVM_EXIT_WRITE_DR2] = dr_interception,
2741 [SVM_EXIT_WRITE_DR3] = dr_interception,
2742 [SVM_EXIT_WRITE_DR4] = dr_interception,
2743 [SVM_EXIT_WRITE_DR5] = dr_interception,
2744 [SVM_EXIT_WRITE_DR6] = dr_interception,
2745 [SVM_EXIT_WRITE_DR7] = dr_interception,
Jan Kiszkad0bfb942008-12-15 13:52:10 +01002746 [SVM_EXIT_EXCP_BASE + DB_VECTOR] = db_interception,
2747 [SVM_EXIT_EXCP_BASE + BP_VECTOR] = bp_interception,
Anthony Liguori7aa81cc2007-09-17 14:57:50 -05002748 [SVM_EXIT_EXCP_BASE + UD_VECTOR] = ud_interception,
Joerg Roedele0231712010-02-24 18:59:10 +01002749 [SVM_EXIT_EXCP_BASE + PF_VECTOR] = pf_interception,
Joerg Roedele0231712010-02-24 18:59:10 +01002750 [SVM_EXIT_EXCP_BASE + MC_VECTOR] = mc_interception,
Eric Northup54a20552015-11-03 18:03:53 +01002751 [SVM_EXIT_EXCP_BASE + AC_VECTOR] = ac_interception,
Liran Alon97184202018-03-12 13:12:52 +02002752 [SVM_EXIT_EXCP_BASE + GP_VECTOR] = gp_interception,
Joerg Roedele0231712010-02-24 18:59:10 +01002753 [SVM_EXIT_INTR] = intr_interception,
Joerg Roedelc47f0982008-04-30 17:56:00 +02002754 [SVM_EXIT_NMI] = nmi_interception,
Avi Kivity6aa8b732006-12-10 02:21:36 -08002755 [SVM_EXIT_SMI] = nop_on_interception,
2756 [SVM_EXIT_INIT] = nop_on_interception,
Dor Laorc1150d82007-01-05 16:36:24 -08002757 [SVM_EXIT_VINTR] = interrupt_window_interception,
Avi Kivity332b56e2011-11-10 14:57:24 +02002758 [SVM_EXIT_RDPMC] = rdpmc_interception,
Avi Kivity6aa8b732006-12-10 02:21:36 -08002759 [SVM_EXIT_CPUID] = cpuid_interception,
Gleb Natapov95ba8273132009-04-21 17:45:08 +03002760 [SVM_EXIT_IRET] = iret_interception,
Avi Kivitycf5a94d2007-10-28 16:11:58 +02002761 [SVM_EXIT_INVD] = emulate_on_interception,
Mark Langsdorf565d0992009-10-06 14:25:02 -05002762 [SVM_EXIT_PAUSE] = pause_interception,
Avi Kivity6aa8b732006-12-10 02:21:36 -08002763 [SVM_EXIT_HLT] = halt_interception,
Marcelo Tosattia7052892008-09-23 13:18:35 -03002764 [SVM_EXIT_INVLPG] = invlpg_interception,
Alexander Grafff092382009-06-15 15:21:24 +02002765 [SVM_EXIT_INVLPGA] = invlpga_interception,
Joerg Roedele0231712010-02-24 18:59:10 +01002766 [SVM_EXIT_IOIO] = io_interception,
Avi Kivity6aa8b732006-12-10 02:21:36 -08002767 [SVM_EXIT_MSR] = msr_interception,
2768 [SVM_EXIT_TASK_SWITCH] = task_switch_interception,
Joerg Roedel46fe4dd2007-01-26 00:56:42 -08002769 [SVM_EXIT_SHUTDOWN] = shutdown_interception,
Alexander Graf3d6368e2008-11-25 20:17:07 +01002770 [SVM_EXIT_VMRUN] = vmrun_interception,
Avi Kivity02e235b2007-02-19 14:37:47 +02002771 [SVM_EXIT_VMMCALL] = vmmcall_interception,
Alexander Graf55426752008-11-25 20:17:06 +01002772 [SVM_EXIT_VMLOAD] = vmload_interception,
2773 [SVM_EXIT_VMSAVE] = vmsave_interception,
Alexander Graf1371d902008-11-25 20:17:04 +01002774 [SVM_EXIT_STGI] = stgi_interception,
2775 [SVM_EXIT_CLGI] = clgi_interception,
Joerg Roedel532a46b2009-10-09 16:08:32 +02002776 [SVM_EXIT_SKINIT] = skinit_interception,
David Kaplandab429a2015-03-02 13:43:37 -06002777 [SVM_EXIT_WBINVD] = wbinvd_interception,
Gabriel L. Somlo87c00572014-05-07 16:52:13 -04002778 [SVM_EXIT_MONITOR] = monitor_interception,
2779 [SVM_EXIT_MWAIT] = mwait_interception,
Joerg Roedel81dd35d2010-12-07 17:15:06 +01002780 [SVM_EXIT_XSETBV] = xsetbv_interception,
Jim Mattson0cb84102019-09-19 15:59:17 -07002781 [SVM_EXIT_RDPRU] = rdpru_interception,
Paolo Bonzinid0006532017-08-11 18:36:43 +02002782 [SVM_EXIT_NPF] = npf_interception,
Brijesh Singh7607b712018-02-19 10:14:44 -06002783 [SVM_EXIT_RSM] = rsm_interception,
Suravee Suthikulpanit18f40c52016-05-04 14:09:48 -05002784 [SVM_EXIT_AVIC_INCOMPLETE_IPI] = avic_incomplete_ipi_interception,
2785 [SVM_EXIT_AVIC_UNACCELERATED_ACCESS] = avic_unaccelerated_access_interception,
Avi Kivity6aa8b732006-12-10 02:21:36 -08002786};
2787
Joe Perchesae8cc052011-04-24 22:00:50 -07002788static void dump_vmcb(struct kvm_vcpu *vcpu)
Joerg Roedel3f10c842010-05-05 16:04:42 +02002789{
2790 struct vcpu_svm *svm = to_svm(vcpu);
2791 struct vmcb_control_area *control = &svm->vmcb->control;
2792 struct vmcb_save_area *save = &svm->vmcb->save;
2793
Paolo Bonzini6f2f8452019-05-20 15:34:35 +02002794 if (!dump_invalid_vmcb) {
2795 pr_warn_ratelimited("set kvm_amd.dump_invalid_vmcb=1 to dump internal KVM state.\n");
2796 return;
2797 }
2798
Joerg Roedel3f10c842010-05-05 16:04:42 +02002799 pr_err("VMCB Control Area:\n");
Joe Perchesae8cc052011-04-24 22:00:50 -07002800 pr_err("%-20s%04x\n", "cr_read:", control->intercept_cr & 0xffff);
2801 pr_err("%-20s%04x\n", "cr_write:", control->intercept_cr >> 16);
2802 pr_err("%-20s%04x\n", "dr_read:", control->intercept_dr & 0xffff);
2803 pr_err("%-20s%04x\n", "dr_write:", control->intercept_dr >> 16);
2804 pr_err("%-20s%08x\n", "exceptions:", control->intercept_exceptions);
2805 pr_err("%-20s%016llx\n", "intercepts:", control->intercept);
2806 pr_err("%-20s%d\n", "pause filter count:", control->pause_filter_count);
Babu Moger1d8fb442018-03-16 16:37:25 -04002807 pr_err("%-20s%d\n", "pause filter threshold:",
2808 control->pause_filter_thresh);
Joe Perchesae8cc052011-04-24 22:00:50 -07002809 pr_err("%-20s%016llx\n", "iopm_base_pa:", control->iopm_base_pa);
2810 pr_err("%-20s%016llx\n", "msrpm_base_pa:", control->msrpm_base_pa);
2811 pr_err("%-20s%016llx\n", "tsc_offset:", control->tsc_offset);
2812 pr_err("%-20s%d\n", "asid:", control->asid);
2813 pr_err("%-20s%d\n", "tlb_ctl:", control->tlb_ctl);
2814 pr_err("%-20s%08x\n", "int_ctl:", control->int_ctl);
2815 pr_err("%-20s%08x\n", "int_vector:", control->int_vector);
2816 pr_err("%-20s%08x\n", "int_state:", control->int_state);
2817 pr_err("%-20s%08x\n", "exit_code:", control->exit_code);
2818 pr_err("%-20s%016llx\n", "exit_info1:", control->exit_info_1);
2819 pr_err("%-20s%016llx\n", "exit_info2:", control->exit_info_2);
2820 pr_err("%-20s%08x\n", "exit_int_info:", control->exit_int_info);
2821 pr_err("%-20s%08x\n", "exit_int_info_err:", control->exit_int_info_err);
2822 pr_err("%-20s%lld\n", "nested_ctl:", control->nested_ctl);
2823 pr_err("%-20s%016llx\n", "nested_cr3:", control->nested_cr3);
Suravee Suthikulpanit44a95da2016-05-04 14:09:46 -05002824 pr_err("%-20s%016llx\n", "avic_vapic_bar:", control->avic_vapic_bar);
Joe Perchesae8cc052011-04-24 22:00:50 -07002825 pr_err("%-20s%08x\n", "event_inj:", control->event_inj);
2826 pr_err("%-20s%08x\n", "event_inj_err:", control->event_inj_err);
Janakarajan Natarajan0dc92112017-07-06 15:50:45 -05002827 pr_err("%-20s%lld\n", "virt_ext:", control->virt_ext);
Joe Perchesae8cc052011-04-24 22:00:50 -07002828 pr_err("%-20s%016llx\n", "next_rip:", control->next_rip);
Suravee Suthikulpanit44a95da2016-05-04 14:09:46 -05002829 pr_err("%-20s%016llx\n", "avic_backing_page:", control->avic_backing_page);
2830 pr_err("%-20s%016llx\n", "avic_logical_id:", control->avic_logical_id);
2831 pr_err("%-20s%016llx\n", "avic_physical_id:", control->avic_physical_id);
Joerg Roedel3f10c842010-05-05 16:04:42 +02002832 pr_err("VMCB State Save Area:\n");
Joe Perchesae8cc052011-04-24 22:00:50 -07002833 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
2834 "es:",
2835 save->es.selector, save->es.attrib,
2836 save->es.limit, save->es.base);
2837 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
2838 "cs:",
2839 save->cs.selector, save->cs.attrib,
2840 save->cs.limit, save->cs.base);
2841 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
2842 "ss:",
2843 save->ss.selector, save->ss.attrib,
2844 save->ss.limit, save->ss.base);
2845 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
2846 "ds:",
2847 save->ds.selector, save->ds.attrib,
2848 save->ds.limit, save->ds.base);
2849 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
2850 "fs:",
2851 save->fs.selector, save->fs.attrib,
2852 save->fs.limit, save->fs.base);
2853 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
2854 "gs:",
2855 save->gs.selector, save->gs.attrib,
2856 save->gs.limit, save->gs.base);
2857 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
2858 "gdtr:",
2859 save->gdtr.selector, save->gdtr.attrib,
2860 save->gdtr.limit, save->gdtr.base);
2861 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
2862 "ldtr:",
2863 save->ldtr.selector, save->ldtr.attrib,
2864 save->ldtr.limit, save->ldtr.base);
2865 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
2866 "idtr:",
2867 save->idtr.selector, save->idtr.attrib,
2868 save->idtr.limit, save->idtr.base);
2869 pr_err("%-5s s: %04x a: %04x l: %08x b: %016llx\n",
2870 "tr:",
2871 save->tr.selector, save->tr.attrib,
2872 save->tr.limit, save->tr.base);
Joerg Roedel3f10c842010-05-05 16:04:42 +02002873 pr_err("cpl: %d efer: %016llx\n",
2874 save->cpl, save->efer);
Joe Perchesae8cc052011-04-24 22:00:50 -07002875 pr_err("%-15s %016llx %-13s %016llx\n",
2876 "cr0:", save->cr0, "cr2:", save->cr2);
2877 pr_err("%-15s %016llx %-13s %016llx\n",
2878 "cr3:", save->cr3, "cr4:", save->cr4);
2879 pr_err("%-15s %016llx %-13s %016llx\n",
2880 "dr6:", save->dr6, "dr7:", save->dr7);
2881 pr_err("%-15s %016llx %-13s %016llx\n",
2882 "rip:", save->rip, "rflags:", save->rflags);
2883 pr_err("%-15s %016llx %-13s %016llx\n",
2884 "rsp:", save->rsp, "rax:", save->rax);
2885 pr_err("%-15s %016llx %-13s %016llx\n",
2886 "star:", save->star, "lstar:", save->lstar);
2887 pr_err("%-15s %016llx %-13s %016llx\n",
2888 "cstar:", save->cstar, "sfmask:", save->sfmask);
2889 pr_err("%-15s %016llx %-13s %016llx\n",
2890 "kernel_gs_base:", save->kernel_gs_base,
2891 "sysenter_cs:", save->sysenter_cs);
2892 pr_err("%-15s %016llx %-13s %016llx\n",
2893 "sysenter_esp:", save->sysenter_esp,
2894 "sysenter_eip:", save->sysenter_eip);
2895 pr_err("%-15s %016llx %-13s %016llx\n",
2896 "gpat:", save->g_pat, "dbgctl:", save->dbgctl);
2897 pr_err("%-15s %016llx %-13s %016llx\n",
2898 "br_from:", save->br_from, "br_to:", save->br_to);
2899 pr_err("%-15s %016llx %-13s %016llx\n",
2900 "excp_from:", save->last_excp_from,
2901 "excp_to:", save->last_excp_to);
Joerg Roedel3f10c842010-05-05 16:04:42 +02002902}
2903
Avi Kivity586f9602010-11-18 13:09:54 +02002904static void svm_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
2905{
2906 struct vmcb_control_area *control = &to_svm(vcpu)->vmcb->control;
2907
2908 *info1 = control->exit_info_1;
2909 *info2 = control->exit_info_2;
2910}
2911
Wanpeng Li404d5d72020-04-28 14:23:25 +08002912static int handle_exit(struct kvm_vcpu *vcpu, fastpath_t exit_fastpath)
Avi Kivity6aa8b732006-12-10 02:21:36 -08002913{
Avi Kivity04d2cc72007-09-10 18:10:54 +03002914 struct vcpu_svm *svm = to_svm(vcpu);
Avi Kivity851ba692009-08-24 11:10:17 +03002915 struct kvm_run *kvm_run = vcpu->run;
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -04002916 u32 exit_code = svm->vmcb->control.exit_code;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002917
Paolo Bonzini8b89fe12015-12-10 18:37:32 +01002918 trace_kvm_exit(exit_code, vcpu, KVM_ISA_SVM);
2919
Roedel, Joerg4ee546b2010-12-03 10:50:51 +01002920 if (!is_cr_intercept(svm, INTERCEPT_CR0_WRITE))
Joerg Roedel2be4fc72010-04-22 12:33:09 +02002921 vcpu->arch.cr0 = svm->vmcb->save.cr0;
2922 if (npt_enabled)
2923 vcpu->arch.cr3 = svm->vmcb->save.cr3;
Joerg Roedelaf9ca2d2008-04-30 17:56:03 +02002924
Paolo Bonzini36e2e982020-05-22 06:04:57 -04002925 svm_complete_interrupts(svm);
2926
Joerg Roedel20307532010-11-29 17:51:48 +01002927 if (is_guest_mode(vcpu)) {
Joerg Roedel410e4d52009-08-07 11:49:44 +02002928 int vmexit;
2929
Joerg Roedeld8cabdd2009-10-09 16:08:28 +02002930 trace_kvm_nested_vmexit(svm->vmcb->save.rip, exit_code,
2931 svm->vmcb->control.exit_info_1,
2932 svm->vmcb->control.exit_info_2,
2933 svm->vmcb->control.exit_int_info,
Stefan Hajnoczie097e5f2011-07-22 12:46:52 +01002934 svm->vmcb->control.exit_int_info_err,
2935 KVM_ISA_SVM);
Joerg Roedeld8cabdd2009-10-09 16:08:28 +02002936
Joerg Roedel410e4d52009-08-07 11:49:44 +02002937 vmexit = nested_svm_exit_special(svm);
2938
2939 if (vmexit == NESTED_EXIT_CONTINUE)
2940 vmexit = nested_svm_exit_handled(svm);
2941
2942 if (vmexit == NESTED_EXIT_DONE)
Alexander Grafcf74a782008-11-25 20:17:08 +01002943 return 1;
Alexander Grafcf74a782008-11-25 20:17:08 +01002944 }
2945
Avi Kivity04d2cc72007-09-10 18:10:54 +03002946 if (svm->vmcb->control.exit_code == SVM_EXIT_ERR) {
2947 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
2948 kvm_run->fail_entry.hardware_entry_failure_reason
2949 = svm->vmcb->control.exit_code;
Joerg Roedel3f10c842010-05-05 16:04:42 +02002950 dump_vmcb(vcpu);
Avi Kivity04d2cc72007-09-10 18:10:54 +03002951 return 0;
2952 }
2953
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -04002954 if (is_external_interrupt(svm->vmcb->control.exit_int_info) &&
Joerg Roedel709ddeb2008-02-07 13:47:45 +01002955 exit_code != SVM_EXIT_EXCP_BASE + PF_VECTOR &&
Joerg Roedel55c5e462010-09-10 17:31:04 +02002956 exit_code != SVM_EXIT_NPF && exit_code != SVM_EXIT_TASK_SWITCH &&
2957 exit_code != SVM_EXIT_INTR && exit_code != SVM_EXIT_NMI)
Borislav Petkov6614c7d2013-04-26 00:22:01 +02002958 printk(KERN_ERR "%s: unexpected exit_int_info 0x%x "
Avi Kivity6aa8b732006-12-10 02:21:36 -08002959 "exit_code 0x%x\n",
Harvey Harrisonb8688d52008-03-03 12:59:56 -08002960 __func__, svm->vmcb->control.exit_int_info,
Avi Kivity6aa8b732006-12-10 02:21:36 -08002961 exit_code);
2962
Wanpeng Li404d5d72020-04-28 14:23:25 +08002963 if (exit_fastpath != EXIT_FASTPATH_NONE)
Wanpeng Li1e9e2622019-11-21 11:17:11 +08002964 return 1;
Wanpeng Li404d5d72020-04-28 14:23:25 +08002965
2966 if (exit_code >= ARRAY_SIZE(svm_exit_handlers)
Joe Perches56919c52007-11-12 20:06:51 -08002967 || !svm_exit_handlers[exit_code]) {
Liran Alon7396d332019-08-26 13:16:43 +03002968 vcpu_unimpl(vcpu, "svm: unexpected exit reason 0x%x\n", exit_code);
2969 dump_vmcb(vcpu);
2970 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
2971 vcpu->run->internal.suberror =
2972 KVM_INTERNAL_ERROR_UNEXPECTED_EXIT_REASON;
2973 vcpu->run->internal.ndata = 1;
2974 vcpu->run->internal.data[0] = exit_code;
2975 return 0;
Avi Kivity6aa8b732006-12-10 02:21:36 -08002976 }
2977
Andrea Arcangeli3dcb2a32019-11-04 18:00:00 -05002978#ifdef CONFIG_RETPOLINE
2979 if (exit_code == SVM_EXIT_MSR)
2980 return msr_interception(svm);
2981 else if (exit_code == SVM_EXIT_VINTR)
2982 return interrupt_window_interception(svm);
2983 else if (exit_code == SVM_EXIT_INTR)
2984 return intr_interception(svm);
2985 else if (exit_code == SVM_EXIT_HLT)
2986 return halt_interception(svm);
2987 else if (exit_code == SVM_EXIT_NPF)
2988 return npf_interception(svm);
2989#endif
Avi Kivity851ba692009-08-24 11:10:17 +03002990 return svm_exit_handlers[exit_code](svm);
Avi Kivity6aa8b732006-12-10 02:21:36 -08002991}
2992
2993static void reload_tss(struct kvm_vcpu *vcpu)
2994{
2995 int cpu = raw_smp_processor_id();
2996
Tejun Heo0fe1e002009-10-29 22:34:14 +09002997 struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
2998 sd->tss_desc->type = 9; /* available 32/64-bit TSS */
Avi Kivity6aa8b732006-12-10 02:21:36 -08002999 load_TR_desc();
3000}
3001
Rusty Russelle756fc62007-07-30 20:07:08 +10003002static void pre_svm_run(struct vcpu_svm *svm)
Avi Kivity6aa8b732006-12-10 02:21:36 -08003003{
3004 int cpu = raw_smp_processor_id();
3005
Tejun Heo0fe1e002009-10-29 22:34:14 +09003006 struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003007
Brijesh Singh70cd94e2017-12-04 10:57:34 -06003008 if (sev_guest(svm->vcpu.kvm))
3009 return pre_sev_run(svm, cpu);
3010
Marcelo Tosatti4b656b12009-07-21 12:47:45 -03003011 /* FIXME: handle wraparound of asid_generation */
Tejun Heo0fe1e002009-10-29 22:34:14 +09003012 if (svm->asid_generation != sd->asid_generation)
3013 new_asid(svm, sd);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003014}
3015
Gleb Natapov95ba8273132009-04-21 17:45:08 +03003016static void svm_inject_nmi(struct kvm_vcpu *vcpu)
3017{
3018 struct vcpu_svm *svm = to_svm(vcpu);
3019
3020 svm->vmcb->control.event_inj = SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_NMI;
3021 vcpu->arch.hflags |= HF_NMI_MASK;
Joerg Roedel8a05a1b82010-11-30 18:04:00 +01003022 set_intercept(svm, INTERCEPT_IRET);
Gleb Natapov95ba8273132009-04-21 17:45:08 +03003023 ++vcpu->stat.nmi_injections;
3024}
Avi Kivity6aa8b732006-12-10 02:21:36 -08003025
Gleb Natapov66fd3f72009-05-11 13:35:50 +03003026static void svm_set_irq(struct kvm_vcpu *vcpu)
Eddie Dong2a8067f2007-08-06 16:29:07 +03003027{
3028 struct vcpu_svm *svm = to_svm(vcpu);
3029
Joerg Roedel2af91942009-08-07 11:49:28 +02003030 BUG_ON(!(gif_set(svm)));
Alexander Grafcf74a782008-11-25 20:17:08 +01003031
Gleb Natapov9fb2d2b2010-05-23 14:28:26 +03003032 trace_kvm_inj_virq(vcpu->arch.interrupt.nr);
3033 ++vcpu->stat.irq_injections;
3034
Alexander Graf219b65d2009-06-15 15:21:25 +02003035 svm->vmcb->control.event_inj = vcpu->arch.interrupt.nr |
3036 SVM_EVTINJ_VALID | SVM_EVTINJ_TYPE_INTR;
Eddie Dong2a8067f2007-08-06 16:29:07 +03003037}
3038
Gleb Natapov95ba8273132009-04-21 17:45:08 +03003039static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
3040{
3041 struct vcpu_svm *svm = to_svm(vcpu);
3042
Liran Alon49d654d2019-11-11 14:26:21 +02003043 if (svm_nested_virtualize_tpr(vcpu))
Joerg Roedel88ab24a2010-02-19 16:23:06 +01003044 return;
3045
Radim Krčmář596f3142014-03-11 19:11:18 +01003046 clr_cr_intercept(svm, INTERCEPT_CR8_WRITE);
3047
Gleb Natapov95ba8273132009-04-21 17:45:08 +03003048 if (irr == -1)
3049 return;
3050
3051 if (tpr >= irr)
Roedel, Joerg4ee546b2010-12-03 10:50:51 +01003052 set_cr_intercept(svm, INTERCEPT_CR8_WRITE);
Gleb Natapov95ba8273132009-04-21 17:45:08 +03003053}
3054
Paolo Bonzinicae96af2020-04-23 14:19:26 -04003055bool svm_nmi_blocked(struct kvm_vcpu *vcpu)
Joerg Roedelaaacfc92008-04-16 16:51:18 +02003056{
3057 struct vcpu_svm *svm = to_svm(vcpu);
3058 struct vmcb *vmcb = svm->vmcb;
Sean Christopherson88c604b2020-04-22 19:25:41 -07003059 bool ret;
Cathy Avery9c3d3702020-04-14 16:11:06 -04003060
Paolo Bonzinicae96af2020-04-23 14:19:26 -04003061 if (!gif_set(svm))
Paolo Bonzinibbdad0b2020-04-23 08:06:43 -04003062 return true;
3063
Paolo Bonzinicae96af2020-04-23 14:19:26 -04003064 if (is_guest_mode(vcpu) && nested_exit_on_nmi(svm))
3065 return false;
3066
3067 ret = (vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK) ||
3068 (svm->vcpu.arch.hflags & HF_NMI_MASK);
Joerg Roedel924584c2010-04-22 12:33:07 +02003069
3070 return ret;
Joerg Roedelaaacfc92008-04-16 16:51:18 +02003071}
3072
Paolo Bonzinic9d40912020-05-22 11:21:49 -04003073static int svm_nmi_allowed(struct kvm_vcpu *vcpu, bool for_injection)
Paolo Bonzinicae96af2020-04-23 14:19:26 -04003074{
3075 struct vcpu_svm *svm = to_svm(vcpu);
3076 if (svm->nested.nested_run_pending)
Paolo Bonzinic9d40912020-05-22 11:21:49 -04003077 return -EBUSY;
Paolo Bonzinicae96af2020-04-23 14:19:26 -04003078
Paolo Bonzinic300ab92020-04-23 14:08:58 -04003079 /* An NMI must not be injected into L2 if it's supposed to VM-Exit. */
3080 if (for_injection && is_guest_mode(vcpu) && nested_exit_on_nmi(svm))
Paolo Bonzinic9d40912020-05-22 11:21:49 -04003081 return -EBUSY;
Paolo Bonzinic300ab92020-04-23 14:08:58 -04003082
3083 return !svm_nmi_blocked(vcpu);
Paolo Bonzinicae96af2020-04-23 14:19:26 -04003084}
3085
Jan Kiszka3cfc3092009-11-12 01:04:25 +01003086static bool svm_get_nmi_mask(struct kvm_vcpu *vcpu)
3087{
3088 struct vcpu_svm *svm = to_svm(vcpu);
3089
3090 return !!(svm->vcpu.arch.hflags & HF_NMI_MASK);
3091}
3092
3093static void svm_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
3094{
3095 struct vcpu_svm *svm = to_svm(vcpu);
3096
3097 if (masked) {
3098 svm->vcpu.arch.hflags |= HF_NMI_MASK;
Joerg Roedel8a05a1b82010-11-30 18:04:00 +01003099 set_intercept(svm, INTERCEPT_IRET);
Jan Kiszka3cfc3092009-11-12 01:04:25 +01003100 } else {
3101 svm->vcpu.arch.hflags &= ~HF_NMI_MASK;
Joerg Roedel8a05a1b82010-11-30 18:04:00 +01003102 clr_intercept(svm, INTERCEPT_IRET);
Jan Kiszka3cfc3092009-11-12 01:04:25 +01003103 }
3104}
3105
Paolo Bonzinicae96af2020-04-23 14:19:26 -04003106bool svm_interrupt_blocked(struct kvm_vcpu *vcpu)
Gleb Natapov78646122009-03-23 12:12:11 +02003107{
3108 struct vcpu_svm *svm = to_svm(vcpu);
3109 struct vmcb *vmcb = svm->vmcb;
Joerg Roedel7fcdb512009-09-16 15:24:15 +02003110
Paolo Bonzinifc6f7c02020-04-23 18:02:45 -04003111 if (!gif_set(svm))
Paolo Bonzinicae96af2020-04-23 14:19:26 -04003112 return true;
Joerg Roedel7fcdb512009-09-16 15:24:15 +02003113
Paolo Bonzinifc6f7c02020-04-23 18:02:45 -04003114 if (is_guest_mode(vcpu)) {
3115 /* As long as interrupts are being delivered... */
Paolo Bonzinie9fd7612020-05-13 13:28:23 -04003116 if ((svm->nested.ctl.int_ctl & V_INTR_MASKING_MASK)
Paolo Bonzini08245e62020-05-19 09:21:04 -04003117 ? !(svm->nested.hsave->save.rflags & X86_EFLAGS_IF)
Paolo Bonzinifc6f7c02020-04-23 18:02:45 -04003118 : !(kvm_get_rflags(vcpu) & X86_EFLAGS_IF))
3119 return true;
3120
3121 /* ... vmexits aren't blocked by the interrupt shadow */
3122 if (nested_exit_on_intr(svm))
3123 return false;
3124 } else {
3125 if (!(kvm_get_rflags(vcpu) & X86_EFLAGS_IF))
3126 return true;
3127 }
3128
3129 return (vmcb->control.int_state & SVM_INTERRUPT_SHADOW_MASK);
Paolo Bonzinicae96af2020-04-23 14:19:26 -04003130}
3131
Paolo Bonzinic9d40912020-05-22 11:21:49 -04003132static int svm_interrupt_allowed(struct kvm_vcpu *vcpu, bool for_injection)
Paolo Bonzinicae96af2020-04-23 14:19:26 -04003133{
3134 struct vcpu_svm *svm = to_svm(vcpu);
3135 if (svm->nested.nested_run_pending)
Paolo Bonzinic9d40912020-05-22 11:21:49 -04003136 return -EBUSY;
Paolo Bonzinicae96af2020-04-23 14:19:26 -04003137
Paolo Bonzinic300ab92020-04-23 14:08:58 -04003138 /*
3139 * An IRQ must not be injected into L2 if it's supposed to VM-Exit,
3140 * e.g. if the IRQ arrived asynchronously after checking nested events.
3141 */
3142 if (for_injection && is_guest_mode(vcpu) && nested_exit_on_intr(svm))
Paolo Bonzinic9d40912020-05-22 11:21:49 -04003143 return -EBUSY;
Paolo Bonzinic300ab92020-04-23 14:08:58 -04003144
3145 return !svm_interrupt_blocked(vcpu);
Gleb Natapov78646122009-03-23 12:12:11 +02003146}
3147
Jan Kiszkac9a79532014-03-07 20:03:15 +01003148static void enable_irq_window(struct kvm_vcpu *vcpu)
Gleb Natapov9222be12009-04-23 17:14:37 +03003149{
Alexander Graf219b65d2009-06-15 15:21:25 +02003150 struct vcpu_svm *svm = to_svm(vcpu);
Alexander Graf219b65d2009-06-15 15:21:25 +02003151
Joerg Roedele0231712010-02-24 18:59:10 +01003152 /*
3153 * In case GIF=0 we can't rely on the CPU to tell us when GIF becomes
3154 * 1, because that's a separate STGI/VMRUN intercept. The next time we
3155 * get that intercept, this function will be called again though and
Janakarajan Natarajan640bd6e2017-08-23 09:57:19 -05003156 * we'll get the vintr intercept. However, if the vGIF feature is
3157 * enabled, the STGI interception will not occur. Enable the irq
3158 * window under the assumption that the hardware will set the GIF.
Joerg Roedele0231712010-02-24 18:59:10 +01003159 */
Paolo Bonzinib518ba92020-03-04 16:46:47 -05003160 if (vgif_enabled(svm) || gif_set(svm)) {
Suravee Suthikulpanitf3515dc2019-11-14 14:15:15 -06003161 /*
3162 * IRQ window is not needed when AVIC is enabled,
3163 * unless we have pending ExtINT since it cannot be injected
3164 * via AVIC. In such case, we need to temporarily disable AVIC,
3165 * and fallback to injecting IRQ via V_IRQ.
3166 */
3167 svm_toggle_avic_for_irq_window(vcpu, false);
Alexander Graf219b65d2009-06-15 15:21:25 +02003168 svm_set_vintr(svm);
Alexander Graf219b65d2009-06-15 15:21:25 +02003169 }
Gleb Natapov9222be12009-04-23 17:14:37 +03003170}
3171
Jan Kiszkac9a79532014-03-07 20:03:15 +01003172static void enable_nmi_window(struct kvm_vcpu *vcpu)
Avi Kivity6aa8b732006-12-10 02:21:36 -08003173{
Avi Kivity04d2cc72007-09-10 18:10:54 +03003174 struct vcpu_svm *svm = to_svm(vcpu);
Eddie Dong85f455f2007-07-06 12:20:49 +03003175
Gleb Natapov44c11432009-05-11 13:35:52 +03003176 if ((svm->vcpu.arch.hflags & (HF_NMI_MASK | HF_IRET_MASK))
3177 == HF_NMI_MASK)
Jan Kiszkac9a79532014-03-07 20:03:15 +01003178 return; /* IRET will cause a vm exit */
Gleb Natapov44c11432009-05-11 13:35:52 +03003179
Janakarajan Natarajan640bd6e2017-08-23 09:57:19 -05003180 if (!gif_set(svm)) {
3181 if (vgif_enabled(svm))
3182 set_intercept(svm, INTERCEPT_STGI);
Ladi Prosek1a5e1852017-06-21 09:07:01 +02003183 return; /* STGI will cause a vm exit */
Janakarajan Natarajan640bd6e2017-08-23 09:57:19 -05003184 }
Ladi Prosek1a5e1852017-06-21 09:07:01 +02003185
Joerg Roedele0231712010-02-24 18:59:10 +01003186 /*
3187 * Something prevents NMI from been injected. Single step over possible
3188 * problem (IRET or exception injection or interrupt shadow)
3189 */
Ladi Prosekab2f4d732017-06-21 09:06:58 +02003190 svm->nmi_singlestep_guest_rflags = svm_get_rflags(vcpu);
Jan Kiszka6be7d302009-10-18 13:24:54 +02003191 svm->nmi_singlestep = true;
Gleb Natapov44c11432009-05-11 13:35:52 +03003192 svm->vmcb->save.rflags |= (X86_EFLAGS_TF | X86_EFLAGS_RF);
Eddie Dong85f455f2007-07-06 12:20:49 +03003193}
3194
Izik Eiduscbc94022007-10-25 00:29:55 +02003195static int svm_set_tss_addr(struct kvm *kvm, unsigned int addr)
3196{
3197 return 0;
3198}
3199
Sean Christopherson2ac52ab2018-03-20 12:17:19 -07003200static int svm_set_identity_map_addr(struct kvm *kvm, u64 ident_addr)
3201{
3202 return 0;
3203}
3204
Sean Christophersonf55ac302020-03-20 14:28:12 -07003205void svm_flush_tlb(struct kvm_vcpu *vcpu)
Avi Kivityd9e368d2007-06-07 19:18:30 +03003206{
Joerg Roedel38e5e922010-12-03 15:25:16 +01003207 struct vcpu_svm *svm = to_svm(vcpu);
3208
Sean Christopherson4a41e432020-03-20 14:28:17 -07003209 /*
3210 * Flush only the current ASID even if the TLB flush was invoked via
3211 * kvm_flush_remote_tlbs(). Although flushing remote TLBs requires all
3212 * ASIDs to be flushed, KVM uses a single ASID for L1 and L2, and
3213 * unconditionally does a TLB flush on both nested VM-Enter and nested
3214 * VM-Exit (via kvm_mmu_reset_context()).
3215 */
Joerg Roedel38e5e922010-12-03 15:25:16 +01003216 if (static_cpu_has(X86_FEATURE_FLUSHBYASID))
3217 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ASID;
3218 else
3219 svm->asid_generation--;
Avi Kivityd9e368d2007-06-07 19:18:30 +03003220}
3221
Junaid Shahidfaff8752018-06-29 13:10:05 -07003222static void svm_flush_tlb_gva(struct kvm_vcpu *vcpu, gva_t gva)
3223{
3224 struct vcpu_svm *svm = to_svm(vcpu);
3225
3226 invlpga(gva, svm->vmcb->control.asid);
3227}
3228
Avi Kivity04d2cc72007-09-10 18:10:54 +03003229static void svm_prepare_guest_switch(struct kvm_vcpu *vcpu)
3230{
3231}
3232
Joerg Roedeld7bf8222008-04-16 16:51:17 +02003233static inline void sync_cr8_to_lapic(struct kvm_vcpu *vcpu)
3234{
3235 struct vcpu_svm *svm = to_svm(vcpu);
3236
Suravee Suthikulpanit3bbf3562016-05-04 14:09:51 -05003237 if (svm_nested_virtualize_tpr(vcpu))
Joerg Roedel88ab24a2010-02-19 16:23:06 +01003238 return;
3239
Roedel, Joerg4ee546b2010-12-03 10:50:51 +01003240 if (!is_cr_intercept(svm, INTERCEPT_CR8_WRITE)) {
Joerg Roedeld7bf8222008-04-16 16:51:17 +02003241 int cr8 = svm->vmcb->control.int_ctl & V_TPR_MASK;
Gleb Natapov615d5192009-04-21 17:45:05 +03003242 kvm_set_cr8(vcpu, cr8);
Joerg Roedeld7bf8222008-04-16 16:51:17 +02003243 }
3244}
3245
Joerg Roedel649d6862008-04-16 16:51:15 +02003246static inline void sync_lapic_to_cr8(struct kvm_vcpu *vcpu)
3247{
3248 struct vcpu_svm *svm = to_svm(vcpu);
3249 u64 cr8;
3250
Suravee Suthikulpanit3bbf3562016-05-04 14:09:51 -05003251 if (svm_nested_virtualize_tpr(vcpu) ||
3252 kvm_vcpu_apicv_active(vcpu))
Joerg Roedel88ab24a2010-02-19 16:23:06 +01003253 return;
3254
Joerg Roedel649d6862008-04-16 16:51:15 +02003255 cr8 = kvm_get_cr8(vcpu);
3256 svm->vmcb->control.int_ctl &= ~V_TPR_MASK;
3257 svm->vmcb->control.int_ctl |= cr8 & V_TPR_MASK;
3258}
3259
Gleb Natapov9222be12009-04-23 17:14:37 +03003260static void svm_complete_interrupts(struct vcpu_svm *svm)
3261{
3262 u8 vector;
3263 int type;
3264 u32 exitintinfo = svm->vmcb->control.exit_int_info;
Jan Kiszka66b71382010-02-23 17:47:56 +01003265 unsigned int3_injected = svm->int3_injected;
3266
3267 svm->int3_injected = 0;
Gleb Natapov9222be12009-04-23 17:14:37 +03003268
Avi Kivitybd3d1ec2011-02-03 15:29:52 +02003269 /*
3270 * If we've made progress since setting HF_IRET_MASK, we've
3271 * executed an IRET and can allow NMI injection.
3272 */
3273 if ((svm->vcpu.arch.hflags & HF_IRET_MASK)
3274 && kvm_rip_read(&svm->vcpu) != svm->nmi_iret_rip) {
Gleb Natapov44c11432009-05-11 13:35:52 +03003275 svm->vcpu.arch.hflags &= ~(HF_NMI_MASK | HF_IRET_MASK);
Avi Kivity3842d132010-07-27 12:30:24 +03003276 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3277 }
Gleb Natapov44c11432009-05-11 13:35:52 +03003278
Gleb Natapov9222be12009-04-23 17:14:37 +03003279 svm->vcpu.arch.nmi_injected = false;
3280 kvm_clear_exception_queue(&svm->vcpu);
3281 kvm_clear_interrupt_queue(&svm->vcpu);
3282
3283 if (!(exitintinfo & SVM_EXITINTINFO_VALID))
3284 return;
3285
Avi Kivity3842d132010-07-27 12:30:24 +03003286 kvm_make_request(KVM_REQ_EVENT, &svm->vcpu);
3287
Gleb Natapov9222be12009-04-23 17:14:37 +03003288 vector = exitintinfo & SVM_EXITINTINFO_VEC_MASK;
3289 type = exitintinfo & SVM_EXITINTINFO_TYPE_MASK;
3290
3291 switch (type) {
3292 case SVM_EXITINTINFO_TYPE_NMI:
3293 svm->vcpu.arch.nmi_injected = true;
3294 break;
3295 case SVM_EXITINTINFO_TYPE_EXEPT:
Jan Kiszka66b71382010-02-23 17:47:56 +01003296 /*
3297 * In case of software exceptions, do not reinject the vector,
3298 * but re-execute the instruction instead. Rewind RIP first
3299 * if we emulated INT3 before.
3300 */
3301 if (kvm_exception_is_soft(vector)) {
3302 if (vector == BP_VECTOR && int3_injected &&
3303 kvm_is_linear_rip(&svm->vcpu, svm->int3_rip))
3304 kvm_rip_write(&svm->vcpu,
3305 kvm_rip_read(&svm->vcpu) -
3306 int3_injected);
Alexander Graf219b65d2009-06-15 15:21:25 +02003307 break;
Jan Kiszka66b71382010-02-23 17:47:56 +01003308 }
Gleb Natapov9222be12009-04-23 17:14:37 +03003309 if (exitintinfo & SVM_EXITINTINFO_VALID_ERR) {
3310 u32 err = svm->vmcb->control.exit_int_info_err;
Joerg Roedelce7ddec2010-04-22 12:33:13 +02003311 kvm_requeue_exception_e(&svm->vcpu, vector, err);
Gleb Natapov9222be12009-04-23 17:14:37 +03003312
3313 } else
Joerg Roedelce7ddec2010-04-22 12:33:13 +02003314 kvm_requeue_exception(&svm->vcpu, vector);
Gleb Natapov9222be12009-04-23 17:14:37 +03003315 break;
3316 case SVM_EXITINTINFO_TYPE_INTR:
Gleb Natapov66fd3f72009-05-11 13:35:50 +03003317 kvm_queue_interrupt(&svm->vcpu, vector, false);
Gleb Natapov9222be12009-04-23 17:14:37 +03003318 break;
3319 default:
3320 break;
3321 }
3322}
3323
Avi Kivityb463a6f2010-07-20 15:06:17 +03003324static void svm_cancel_injection(struct kvm_vcpu *vcpu)
3325{
3326 struct vcpu_svm *svm = to_svm(vcpu);
3327 struct vmcb_control_area *control = &svm->vmcb->control;
3328
3329 control->exit_int_info = control->event_inj;
3330 control->exit_int_info_err = control->event_inj_err;
3331 control->event_inj = 0;
3332 svm_complete_interrupts(svm);
3333}
3334
Wanpeng Li404d5d72020-04-28 14:23:25 +08003335static fastpath_t svm_exit_handlers_fastpath(struct kvm_vcpu *vcpu)
Wanpeng Lia9ab13f2020-04-10 10:47:03 -07003336{
3337 if (!is_guest_mode(vcpu) &&
3338 to_svm(vcpu)->vmcb->control.exit_code == SVM_EXIT_MSR &&
3339 to_svm(vcpu)->vmcb->control.exit_info_1)
3340 return handle_fastpath_set_msr_irqoff(vcpu);
3341
3342 return EXIT_FASTPATH_NONE;
3343}
3344
Uros Bizjak56a87e52020-04-09 13:49:26 +02003345void __svm_vcpu_run(unsigned long vmcb_pa, unsigned long *regs);
Uros Bizjak199cd1d2020-03-30 15:02:13 +02003346
Wanpeng Li404d5d72020-04-28 14:23:25 +08003347static fastpath_t svm_vcpu_run(struct kvm_vcpu *vcpu)
Avi Kivity6aa8b732006-12-10 02:21:36 -08003348{
Wanpeng Li404d5d72020-04-28 14:23:25 +08003349 fastpath_t exit_fastpath;
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -04003350 struct vcpu_svm *svm = to_svm(vcpu);
Avi Kivityd9e368d2007-06-07 19:18:30 +03003351
Joerg Roedel2041a062010-04-22 12:33:08 +02003352 svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
3353 svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
3354 svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP];
3355
Joerg Roedelcd3ff652009-10-09 16:08:26 +02003356 /*
Ladi Proseka12713c2017-06-21 09:07:00 +02003357 * Disable singlestep if we're injecting an interrupt/exception.
3358 * We don't want our modified rflags to be pushed on the stack where
3359 * we might not be able to easily reset them if we disabled NMI
3360 * singlestep later.
3361 */
3362 if (svm->nmi_singlestep && svm->vmcb->control.event_inj) {
3363 /*
3364 * Event injection happens before external interrupts cause a
3365 * vmexit and interrupts are disabled here, so smp_send_reschedule
3366 * is enough to force an immediate vmexit.
3367 */
3368 disable_nmi_singlestep(svm);
3369 smp_send_reschedule(vcpu->cpu);
3370 }
3371
Rusty Russelle756fc62007-07-30 20:07:08 +10003372 pre_svm_run(svm);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003373
Joerg Roedel649d6862008-04-16 16:51:15 +02003374 sync_lapic_to_cr8(vcpu);
3375
Joerg Roedelcda0ffd2009-08-07 11:49:45 +02003376 svm->vmcb->save.cr2 = vcpu->arch.cr2;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003377
Paolo Bonzinid67668e2020-05-06 06:40:04 -04003378 /*
3379 * Run with all-zero DR6 unless needed, so that we can get the exact cause
3380 * of a #DB.
3381 */
3382 if (unlikely(svm->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT))
3383 svm_set_dr6(svm, vcpu->arch.dr6);
3384 else
3385 svm_set_dr6(svm, DR6_FIXED_1 | DR6_RTM);
3386
Avi Kivity04d2cc72007-09-10 18:10:54 +03003387 clgi();
Aaron Lewis139a12c2019-10-21 16:30:25 -07003388 kvm_load_guest_xsave_state(vcpu);
Avi Kivity04d2cc72007-09-10 18:10:54 +03003389
Wanpeng Lib6c4bc62019-05-20 16:18:09 +08003390 if (lapic_in_kernel(vcpu) &&
3391 vcpu->arch.apic->lapic_timer.timer_advance_ns)
3392 kvm_wait_lapic_expire(vcpu);
3393
KarimAllah Ahmedb2ac58f2018-02-03 15:56:23 +01003394 /*
3395 * If this vCPU has touched SPEC_CTRL, restore the guest's value if
3396 * it's non-zero. Since vmentry is serialising on affected CPUs, there
3397 * is no need to worry about the conditional branch over the wrmsr
3398 * being speculatively taken.
3399 */
Thomas Gleixnerccbcd262018-05-09 23:01:01 +02003400 x86_spec_ctrl_set_guest(svm->spec_ctrl, svm->virt_spec_ctrl);
KarimAllah Ahmedb2ac58f2018-02-03 15:56:23 +01003401
Uros Bizjak199cd1d2020-03-30 15:02:13 +02003402 __svm_vcpu_run(svm->vmcb_pa, (unsigned long *)&svm->vcpu.arch.regs);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003403
Thomas Gleixner15e6c222018-05-11 15:21:01 +02003404#ifdef CONFIG_X86_64
3405 wrmsrl(MSR_GS_BASE, svm->host.gs_base);
3406#else
3407 loadsegment(fs, svm->host.fs);
3408#ifndef CONFIG_X86_32_LAZY_GS
3409 loadsegment(gs, svm->host.gs);
3410#endif
3411#endif
3412
KarimAllah Ahmedb2ac58f2018-02-03 15:56:23 +01003413 /*
3414 * We do not use IBRS in the kernel. If this vCPU has used the
3415 * SPEC_CTRL MSR it may have left it on; save the value and
3416 * turn it off. This is much more efficient than blindly adding
3417 * it to the atomic save/restore list. Especially as the former
3418 * (Saving guest MSRs on vmexit) doesn't even exist in KVM.
3419 *
3420 * For non-nested case:
3421 * If the L01 MSR bitmap does not intercept the MSR, then we need to
3422 * save it.
3423 *
3424 * For nested case:
3425 * If the L02 MSR bitmap does not intercept the MSR, then we need to
3426 * save it.
3427 */
Paolo Bonzini946fbbc2018-02-22 16:43:18 +01003428 if (unlikely(!msr_write_intercepted(vcpu, MSR_IA32_SPEC_CTRL)))
Paolo Bonziniecb586b2018-02-22 16:43:17 +01003429 svm->spec_ctrl = native_read_msr(MSR_IA32_SPEC_CTRL);
KarimAllah Ahmedb2ac58f2018-02-03 15:56:23 +01003430
Avi Kivity6aa8b732006-12-10 02:21:36 -08003431 reload_tss(vcpu);
3432
Thomas Gleixner024d83c2018-08-12 20:41:45 +02003433 x86_spec_ctrl_restore_host(svm->spec_ctrl, svm->virt_spec_ctrl);
3434
Avi Kivity13c34e02010-10-21 12:20:31 +02003435 vcpu->arch.cr2 = svm->vmcb->save.cr2;
3436 vcpu->arch.regs[VCPU_REGS_RAX] = svm->vmcb->save.rax;
3437 vcpu->arch.regs[VCPU_REGS_RSP] = svm->vmcb->save.rsp;
3438 vcpu->arch.regs[VCPU_REGS_RIP] = svm->vmcb->save.rip;
3439
Joerg Roedel3781c012011-01-14 16:45:02 +01003440 if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI))
Andi Kleendd60d212017-07-25 17:20:32 -07003441 kvm_before_interrupt(&svm->vcpu);
Joerg Roedel3781c012011-01-14 16:45:02 +01003442
Aaron Lewis139a12c2019-10-21 16:30:25 -07003443 kvm_load_host_xsave_state(vcpu);
Joerg Roedel3781c012011-01-14 16:45:02 +01003444 stgi();
3445
3446 /* Any pending NMI will happen here */
Wanpeng Lia9ab13f2020-04-10 10:47:03 -07003447 exit_fastpath = svm_exit_handlers_fastpath(vcpu);
Joerg Roedel3781c012011-01-14 16:45:02 +01003448
3449 if (unlikely(svm->vmcb->control.exit_code == SVM_EXIT_NMI))
Andi Kleendd60d212017-07-25 17:20:32 -07003450 kvm_after_interrupt(&svm->vcpu);
Joerg Roedel3781c012011-01-14 16:45:02 +01003451
Joerg Roedeld7bf8222008-04-16 16:51:17 +02003452 sync_cr8_to_lapic(vcpu);
3453
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -04003454 svm->next_rip = 0;
Paolo Bonzini2d8a42b2020-05-22 03:50:14 -04003455 if (is_guest_mode(&svm->vcpu)) {
3456 sync_nested_vmcb_control(svm);
3457 svm->nested.nested_run_pending = 0;
3458 }
Gleb Natapov9222be12009-04-23 17:14:37 +03003459
Joerg Roedel38e5e922010-12-03 15:25:16 +01003460 svm->vmcb->control.tlb_ctl = TLB_CONTROL_DO_NOTHING;
3461
Gleb Natapov631bc482010-10-14 11:22:52 +02003462 /* if exit due to PF check for async PF */
3463 if (svm->vmcb->control.exit_code == SVM_EXIT_EXCP_BASE + PF_VECTOR)
Vitaly Kuznetsov68fd66f2020-05-25 16:41:17 +02003464 svm->vcpu.arch.apf.host_apf_flags =
3465 kvm_read_and_reset_apf_flags();
Gleb Natapov631bc482010-10-14 11:22:52 +02003466
Avi Kivity6de4f3a2009-05-31 22:58:47 +03003467 if (npt_enabled) {
3468 vcpu->arch.regs_avail &= ~(1 << VCPU_EXREG_PDPTR);
3469 vcpu->arch.regs_dirty &= ~(1 << VCPU_EXREG_PDPTR);
3470 }
Joerg Roedelfe5913e2010-05-17 14:43:34 +02003471
3472 /*
3473 * We need to handle MC intercepts here before the vcpu has a chance to
3474 * change the physical cpu
3475 */
3476 if (unlikely(svm->vmcb->control.exit_code ==
3477 SVM_EXIT_EXCP_BASE + MC_VECTOR))
3478 svm_handle_mce(svm);
Roedel, Joerg8d28fec2010-12-03 13:15:21 +01003479
3480 mark_all_clean(svm->vmcb);
Wanpeng Lia9ab13f2020-04-10 10:47:03 -07003481 return exit_fastpath;
Avi Kivity6aa8b732006-12-10 02:21:36 -08003482}
3483
Paolo Bonzini727a7e22020-03-05 03:52:50 -05003484static void svm_load_mmu_pgd(struct kvm_vcpu *vcpu, unsigned long root)
Avi Kivity6aa8b732006-12-10 02:21:36 -08003485{
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -04003486 struct vcpu_svm *svm = to_svm(vcpu);
Paolo Bonzini689f3bf2020-03-03 10:11:10 +01003487 unsigned long cr3;
Gregory Haskinsa2fa3e92007-07-27 08:13:10 -04003488
Paolo Bonzini689f3bf2020-03-03 10:11:10 +01003489 cr3 = __sme_set(root);
3490 if (npt_enabled) {
3491 svm->vmcb->control.nested_cr3 = cr3;
3492 mark_dirty(svm->vmcb, VMCB_NPT);
Avi Kivity6aa8b732006-12-10 02:21:36 -08003493
Paolo Bonzini689f3bf2020-03-03 10:11:10 +01003494 /* Loading L2's CR3 is handled by enter_svm_guest_mode. */
Paolo Bonzini978ce582020-05-20 08:37:37 -04003495 if (!test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
3496 return;
3497 cr3 = vcpu->arch.cr3;
Paolo Bonzini689f3bf2020-03-03 10:11:10 +01003498 }
Joerg Roedel1c97f0a2010-09-10 17:30:41 +02003499
Paolo Bonzini978ce582020-05-20 08:37:37 -04003500 svm->vmcb->save.cr3 = cr3;
3501 mark_dirty(svm->vmcb, VMCB_CR);
Joerg Roedel1c97f0a2010-09-10 17:30:41 +02003502}
3503
Avi Kivity6aa8b732006-12-10 02:21:36 -08003504static int is_disabled(void)
3505{
Joerg Roedel6031a612007-06-22 12:29:50 +03003506 u64 vm_cr;
3507
3508 rdmsrl(MSR_VM_CR, vm_cr);
3509 if (vm_cr & (1 << SVM_VM_CR_SVM_DISABLE))
3510 return 1;
3511
Avi Kivity6aa8b732006-12-10 02:21:36 -08003512 return 0;
3513}
3514
Ingo Molnar102d8322007-02-19 14:37:47 +02003515static void
3516svm_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
3517{
3518 /*
3519 * Patch in the VMMCALL instruction:
3520 */
3521 hypercall[0] = 0x0f;
3522 hypercall[1] = 0x01;
3523 hypercall[2] = 0xd9;
Ingo Molnar102d8322007-02-19 14:37:47 +02003524}
3525
Sean Christophersonf257d6d2019-04-19 22:18:17 -07003526static int __init svm_check_processor_compat(void)
Yang, Sheng002c7f72007-07-31 14:23:01 +03003527{
Sean Christophersonf257d6d2019-04-19 22:18:17 -07003528 return 0;
Yang, Sheng002c7f72007-07-31 14:23:01 +03003529}
3530
Avi Kivity774ead32007-12-26 13:57:04 +02003531static bool svm_cpu_has_accelerated_tpr(void)
3532{
3533 return false;
3534}
3535
Sean Christophersoncb97c2d2020-02-18 15:40:11 -08003536static bool svm_has_emulated_msr(u32 index)
Paolo Bonzini6d396b52015-04-01 14:25:33 +02003537{
Vitaly Kuznetsove87555e2018-12-19 12:06:13 +01003538 switch (index) {
3539 case MSR_IA32_MCG_EXT_CTL:
Paolo Bonzini95c5c7c2019-07-02 14:45:24 +02003540 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
Vitaly Kuznetsove87555e2018-12-19 12:06:13 +01003541 return false;
3542 default:
3543 break;
3544 }
3545
Paolo Bonzini6d396b52015-04-01 14:25:33 +02003546 return true;
3547}
3548
Paolo Bonzinifc07e762015-10-01 13:20:22 +02003549static u64 svm_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
3550{
3551 return 0;
3552}
3553
Sheng Yang0e851882009-12-18 16:48:46 +08003554static void svm_cpuid_update(struct kvm_vcpu *vcpu)
3555{
Joerg Roedel6092d3d2015-10-14 15:10:54 +02003556 struct vcpu_svm *svm = to_svm(vcpu);
3557
Aaron Lewis72041602019-10-21 16:30:20 -07003558 vcpu->arch.xsaves_enabled = guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
Sean Christopherson96be4e02019-12-10 14:44:15 -08003559 boot_cpu_has(X86_FEATURE_XSAVE) &&
Aaron Lewis72041602019-10-21 16:30:20 -07003560 boot_cpu_has(X86_FEATURE_XSAVES);
3561
Joerg Roedel6092d3d2015-10-14 15:10:54 +02003562 /* Update nrips enabled cache */
Sean Christopherson4eb87462020-03-02 15:57:08 -08003563 svm->nrips_enabled = kvm_cpu_cap_has(X86_FEATURE_NRIPS) &&
3564 guest_cpuid_has(&svm->vcpu, X86_FEATURE_NRIPS);
Suravee Suthikulpanit46781ea2016-05-04 14:09:50 -05003565
3566 if (!kvm_vcpu_apicv_active(vcpu))
3567 return;
3568
Oliver Uptoncc7f5572020-02-28 00:59:04 -08003569 /*
3570 * AVIC does not work with an x2APIC mode guest. If the X2APIC feature
3571 * is exposed to the guest, disable AVIC.
3572 */
3573 if (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC))
3574 kvm_request_apicv_update(vcpu->kvm, false,
3575 APICV_INHIBIT_REASON_X2APIC);
Suravee Suthikulpanit9a0bf052019-11-14 14:15:14 -06003576
3577 /*
3578 * Currently, AVIC does not work with nested virtualization.
3579 * So, we disable AVIC when cpuid for SVM is set in the L1 guest.
3580 */
3581 if (nested && guest_cpuid_has(vcpu, X86_FEATURE_SVM))
3582 kvm_request_apicv_update(vcpu->kvm, false,
3583 APICV_INHIBIT_REASON_NESTED);
Sheng Yang0e851882009-12-18 16:48:46 +08003584}
3585
Sheng Yangf5f48ee2010-06-30 12:25:15 +08003586static bool svm_has_wbinvd_exit(void)
3587{
3588 return true;
3589}
3590
Joerg Roedel80612522011-04-04 12:39:33 +02003591#define PRE_EX(exit) { .exit_code = (exit), \
Avi Kivity40e19b52011-04-21 12:35:41 +03003592 .stage = X86_ICPT_PRE_EXCEPT, }
Joerg Roedelcfec82c2011-04-04 12:39:28 +02003593#define POST_EX(exit) { .exit_code = (exit), \
Avi Kivity40e19b52011-04-21 12:35:41 +03003594 .stage = X86_ICPT_POST_EXCEPT, }
Joerg Roedeld7eb8202011-04-04 12:39:32 +02003595#define POST_MEM(exit) { .exit_code = (exit), \
Avi Kivity40e19b52011-04-21 12:35:41 +03003596 .stage = X86_ICPT_POST_MEMACCESS, }
Joerg Roedelcfec82c2011-04-04 12:39:28 +02003597
Mathias Krause09941fb2012-08-30 01:30:20 +02003598static const struct __x86_intercept {
Joerg Roedelcfec82c2011-04-04 12:39:28 +02003599 u32 exit_code;
3600 enum x86_intercept_stage stage;
Joerg Roedelcfec82c2011-04-04 12:39:28 +02003601} x86_intercept_map[] = {
3602 [x86_intercept_cr_read] = POST_EX(SVM_EXIT_READ_CR0),
3603 [x86_intercept_cr_write] = POST_EX(SVM_EXIT_WRITE_CR0),
3604 [x86_intercept_clts] = POST_EX(SVM_EXIT_WRITE_CR0),
3605 [x86_intercept_lmsw] = POST_EX(SVM_EXIT_WRITE_CR0),
3606 [x86_intercept_smsw] = POST_EX(SVM_EXIT_READ_CR0),
Joerg Roedel3b88e412011-04-04 12:39:29 +02003607 [x86_intercept_dr_read] = POST_EX(SVM_EXIT_READ_DR0),
3608 [x86_intercept_dr_write] = POST_EX(SVM_EXIT_WRITE_DR0),
Joerg Roedeldee6bb72011-04-04 12:39:30 +02003609 [x86_intercept_sldt] = POST_EX(SVM_EXIT_LDTR_READ),
3610 [x86_intercept_str] = POST_EX(SVM_EXIT_TR_READ),
3611 [x86_intercept_lldt] = POST_EX(SVM_EXIT_LDTR_WRITE),
3612 [x86_intercept_ltr] = POST_EX(SVM_EXIT_TR_WRITE),
3613 [x86_intercept_sgdt] = POST_EX(SVM_EXIT_GDTR_READ),
3614 [x86_intercept_sidt] = POST_EX(SVM_EXIT_IDTR_READ),
3615 [x86_intercept_lgdt] = POST_EX(SVM_EXIT_GDTR_WRITE),
3616 [x86_intercept_lidt] = POST_EX(SVM_EXIT_IDTR_WRITE),
Joerg Roedel01de8b02011-04-04 12:39:31 +02003617 [x86_intercept_vmrun] = POST_EX(SVM_EXIT_VMRUN),
3618 [x86_intercept_vmmcall] = POST_EX(SVM_EXIT_VMMCALL),
3619 [x86_intercept_vmload] = POST_EX(SVM_EXIT_VMLOAD),
3620 [x86_intercept_vmsave] = POST_EX(SVM_EXIT_VMSAVE),
3621 [x86_intercept_stgi] = POST_EX(SVM_EXIT_STGI),
3622 [x86_intercept_clgi] = POST_EX(SVM_EXIT_CLGI),
3623 [x86_intercept_skinit] = POST_EX(SVM_EXIT_SKINIT),
3624 [x86_intercept_invlpga] = POST_EX(SVM_EXIT_INVLPGA),
Joerg Roedeld7eb8202011-04-04 12:39:32 +02003625 [x86_intercept_rdtscp] = POST_EX(SVM_EXIT_RDTSCP),
3626 [x86_intercept_monitor] = POST_MEM(SVM_EXIT_MONITOR),
3627 [x86_intercept_mwait] = POST_EX(SVM_EXIT_MWAIT),
Joerg Roedel80612522011-04-04 12:39:33 +02003628 [x86_intercept_invlpg] = POST_EX(SVM_EXIT_INVLPG),
3629 [x86_intercept_invd] = POST_EX(SVM_EXIT_INVD),
3630 [x86_intercept_wbinvd] = POST_EX(SVM_EXIT_WBINVD),
3631 [x86_intercept_wrmsr] = POST_EX(SVM_EXIT_MSR),
3632 [x86_intercept_rdtsc] = POST_EX(SVM_EXIT_RDTSC),
3633 [x86_intercept_rdmsr] = POST_EX(SVM_EXIT_MSR),
3634 [x86_intercept_rdpmc] = POST_EX(SVM_EXIT_RDPMC),
3635 [x86_intercept_cpuid] = PRE_EX(SVM_EXIT_CPUID),
3636 [x86_intercept_rsm] = PRE_EX(SVM_EXIT_RSM),
Joerg Roedelbf608f82011-04-04 12:39:34 +02003637 [x86_intercept_pause] = PRE_EX(SVM_EXIT_PAUSE),
3638 [x86_intercept_pushf] = PRE_EX(SVM_EXIT_PUSHF),
3639 [x86_intercept_popf] = PRE_EX(SVM_EXIT_POPF),
3640 [x86_intercept_intn] = PRE_EX(SVM_EXIT_SWINT),
3641 [x86_intercept_iret] = PRE_EX(SVM_EXIT_IRET),
3642 [x86_intercept_icebp] = PRE_EX(SVM_EXIT_ICEBP),
3643 [x86_intercept_hlt] = POST_EX(SVM_EXIT_HLT),
Joerg Roedelf6511932011-04-04 12:39:35 +02003644 [x86_intercept_in] = POST_EX(SVM_EXIT_IOIO),
3645 [x86_intercept_ins] = POST_EX(SVM_EXIT_IOIO),
3646 [x86_intercept_out] = POST_EX(SVM_EXIT_IOIO),
3647 [x86_intercept_outs] = POST_EX(SVM_EXIT_IOIO),
Vitaly Kuznetsov02d41602019-08-13 15:53:32 +02003648 [x86_intercept_xsetbv] = PRE_EX(SVM_EXIT_XSETBV),
Joerg Roedelcfec82c2011-04-04 12:39:28 +02003649};
3650
Joerg Roedel80612522011-04-04 12:39:33 +02003651#undef PRE_EX
Joerg Roedelcfec82c2011-04-04 12:39:28 +02003652#undef POST_EX
Joerg Roedeld7eb8202011-04-04 12:39:32 +02003653#undef POST_MEM
Joerg Roedelcfec82c2011-04-04 12:39:28 +02003654
Joerg Roedel8a76d7f2011-04-04 12:39:27 +02003655static int svm_check_intercept(struct kvm_vcpu *vcpu,
3656 struct x86_instruction_info *info,
Sean Christopherson21f1b8f2020-02-18 15:29:42 -08003657 enum x86_intercept_stage stage,
3658 struct x86_exception *exception)
Joerg Roedel8a76d7f2011-04-04 12:39:27 +02003659{
Joerg Roedelcfec82c2011-04-04 12:39:28 +02003660 struct vcpu_svm *svm = to_svm(vcpu);
3661 int vmexit, ret = X86EMUL_CONTINUE;
3662 struct __x86_intercept icpt_info;
3663 struct vmcb *vmcb = svm->vmcb;
3664
3665 if (info->intercept >= ARRAY_SIZE(x86_intercept_map))
3666 goto out;
3667
3668 icpt_info = x86_intercept_map[info->intercept];
3669
Avi Kivity40e19b52011-04-21 12:35:41 +03003670 if (stage != icpt_info.stage)
Joerg Roedelcfec82c2011-04-04 12:39:28 +02003671 goto out;
3672
3673 switch (icpt_info.exit_code) {
3674 case SVM_EXIT_READ_CR0:
3675 if (info->intercept == x86_intercept_cr_read)
3676 icpt_info.exit_code += info->modrm_reg;
3677 break;
3678 case SVM_EXIT_WRITE_CR0: {
3679 unsigned long cr0, val;
3680 u64 intercept;
3681
3682 if (info->intercept == x86_intercept_cr_write)
3683 icpt_info.exit_code += info->modrm_reg;
3684
Jan Kiszka62baf442014-06-29 21:55:53 +02003685 if (icpt_info.exit_code != SVM_EXIT_WRITE_CR0 ||
3686 info->intercept == x86_intercept_clts)
Joerg Roedelcfec82c2011-04-04 12:39:28 +02003687 break;
3688
Paolo Bonzinie670bf62020-05-13 13:16:12 -04003689 intercept = svm->nested.ctl.intercept;
Joerg Roedelcfec82c2011-04-04 12:39:28 +02003690
3691 if (!(intercept & (1ULL << INTERCEPT_SELECTIVE_CR0)))
3692 break;
3693
3694 cr0 = vcpu->arch.cr0 & ~SVM_CR0_SELECTIVE_MASK;
3695 val = info->src_val & ~SVM_CR0_SELECTIVE_MASK;
3696
3697 if (info->intercept == x86_intercept_lmsw) {
3698 cr0 &= 0xfUL;
3699 val &= 0xfUL;
3700 /* lmsw can't clear PE - catch this here */
3701 if (cr0 & X86_CR0_PE)
3702 val |= X86_CR0_PE;
3703 }
3704
3705 if (cr0 ^ val)
3706 icpt_info.exit_code = SVM_EXIT_CR0_SEL_WRITE;
3707
3708 break;
3709 }
Joerg Roedel3b88e412011-04-04 12:39:29 +02003710 case SVM_EXIT_READ_DR0:
3711 case SVM_EXIT_WRITE_DR0:
3712 icpt_info.exit_code += info->modrm_reg;
3713 break;
Joerg Roedel80612522011-04-04 12:39:33 +02003714 case SVM_EXIT_MSR:
3715 if (info->intercept == x86_intercept_wrmsr)
3716 vmcb->control.exit_info_1 = 1;
3717 else
3718 vmcb->control.exit_info_1 = 0;
3719 break;
Joerg Roedelbf608f82011-04-04 12:39:34 +02003720 case SVM_EXIT_PAUSE:
3721 /*
3722 * We get this for NOP only, but pause
3723 * is rep not, check this here
3724 */
3725 if (info->rep_prefix != REPE_PREFIX)
3726 goto out;
Jan H. Schönherr49a8afc2017-09-05 23:58:44 +02003727 break;
Joerg Roedelf6511932011-04-04 12:39:35 +02003728 case SVM_EXIT_IOIO: {
3729 u64 exit_info;
3730 u32 bytes;
3731
Joerg Roedelf6511932011-04-04 12:39:35 +02003732 if (info->intercept == x86_intercept_in ||
3733 info->intercept == x86_intercept_ins) {
Jan Kiszka6cbc5f52014-06-30 12:52:55 +02003734 exit_info = ((info->src_val & 0xffff) << 16) |
3735 SVM_IOIO_TYPE_MASK;
Joerg Roedelf6511932011-04-04 12:39:35 +02003736 bytes = info->dst_bytes;
Jan Kiszka6493f152014-06-30 11:07:05 +02003737 } else {
Jan Kiszka6cbc5f52014-06-30 12:52:55 +02003738 exit_info = (info->dst_val & 0xffff) << 16;
Jan Kiszka6493f152014-06-30 11:07:05 +02003739 bytes = info->src_bytes;
Joerg Roedelf6511932011-04-04 12:39:35 +02003740 }
3741
3742 if (info->intercept == x86_intercept_outs ||
3743 info->intercept == x86_intercept_ins)
3744 exit_info |= SVM_IOIO_STR_MASK;
3745
3746 if (info->rep_prefix)
3747 exit_info |= SVM_IOIO_REP_MASK;
3748
3749 bytes = min(bytes, 4u);
3750
3751 exit_info |= bytes << SVM_IOIO_SIZE_SHIFT;
3752
3753 exit_info |= (u32)info->ad_bytes << (SVM_IOIO_ASIZE_SHIFT - 1);
3754
3755 vmcb->control.exit_info_1 = exit_info;
3756 vmcb->control.exit_info_2 = info->next_rip;
3757
3758 break;
3759 }
Joerg Roedelcfec82c2011-04-04 12:39:28 +02003760 default:
3761 break;
3762 }
3763
Bandan Dasf1047652015-06-11 02:05:33 -04003764 /* TODO: Advertise NRIPS to guest hypervisor unconditionally */
3765 if (static_cpu_has(X86_FEATURE_NRIPS))
3766 vmcb->control.next_rip = info->next_rip;
Joerg Roedelcfec82c2011-04-04 12:39:28 +02003767 vmcb->control.exit_code = icpt_info.exit_code;
3768 vmexit = nested_svm_exit_handled(svm);
3769
3770 ret = (vmexit == NESTED_EXIT_DONE) ? X86EMUL_INTERCEPTED
3771 : X86EMUL_CONTINUE;
3772
3773out:
3774 return ret;
Joerg Roedel8a76d7f2011-04-04 12:39:27 +02003775}
3776
Wanpeng Lia9ab13f2020-04-10 10:47:03 -07003777static void svm_handle_exit_irqoff(struct kvm_vcpu *vcpu)
Yang Zhanga547c6d2013-04-11 19:25:10 +08003778{
Yang Zhanga547c6d2013-04-11 19:25:10 +08003779}
3780
Radim Krčmářae97a3b2014-08-21 18:08:06 +02003781static void svm_sched_in(struct kvm_vcpu *vcpu, int cpu)
3782{
Babu Moger8566ac82018-03-16 16:37:26 -04003783 if (pause_filter_thresh)
3784 shrink_ple_window(vcpu);
Radim Krčmářae97a3b2014-08-21 18:08:06 +02003785}
3786
Borislav Petkov74f16902017-03-26 23:51:24 +02003787static void svm_setup_mce(struct kvm_vcpu *vcpu)
3788{
3789 /* [63:9] are reserved. */
3790 vcpu->arch.mcg_cap &= 0x1ff;
3791}
3792
Paolo Bonzinicae96af2020-04-23 14:19:26 -04003793bool svm_smi_blocked(struct kvm_vcpu *vcpu)
Ladi Prosek72d7b372017-10-11 16:54:41 +02003794{
Ladi Prosek05cade72017-10-11 16:54:45 +02003795 struct vcpu_svm *svm = to_svm(vcpu);
3796
3797 /* Per APM Vol.2 15.22.2 "Response to SMI" */
3798 if (!gif_set(svm))
Paolo Bonzinicae96af2020-04-23 14:19:26 -04003799 return true;
3800
3801 return is_smm(vcpu);
3802}
3803
Paolo Bonzinic9d40912020-05-22 11:21:49 -04003804static int svm_smi_allowed(struct kvm_vcpu *vcpu, bool for_injection)
Paolo Bonzinicae96af2020-04-23 14:19:26 -04003805{
3806 struct vcpu_svm *svm = to_svm(vcpu);
3807 if (svm->nested.nested_run_pending)
Paolo Bonzinic9d40912020-05-22 11:21:49 -04003808 return -EBUSY;
Ladi Prosek05cade72017-10-11 16:54:45 +02003809
Paolo Bonzinic300ab92020-04-23 14:08:58 -04003810 /* An SMI must not be injected into L2 if it's supposed to VM-Exit. */
3811 if (for_injection && is_guest_mode(vcpu) && nested_exit_on_smi(svm))
Paolo Bonzinic9d40912020-05-22 11:21:49 -04003812 return -EBUSY;
Paolo Bonzinic300ab92020-04-23 14:08:58 -04003813
Paolo Bonzinicae96af2020-04-23 14:19:26 -04003814 return !svm_smi_blocked(vcpu);
Ladi Prosek72d7b372017-10-11 16:54:41 +02003815}
3816
Ladi Prosek0234bf82017-10-11 16:54:40 +02003817static int svm_pre_enter_smm(struct kvm_vcpu *vcpu, char *smstate)
3818{
Ladi Prosek05cade72017-10-11 16:54:45 +02003819 struct vcpu_svm *svm = to_svm(vcpu);
3820 int ret;
3821
3822 if (is_guest_mode(vcpu)) {
3823 /* FED8h - SVM Guest */
3824 put_smstate(u64, smstate, 0x7ed8, 1);
3825 /* FEE0h - SVM Guest VMCB Physical Address */
3826 put_smstate(u64, smstate, 0x7ee0, svm->nested.vmcb);
3827
3828 svm->vmcb->save.rax = vcpu->arch.regs[VCPU_REGS_RAX];
3829 svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
3830 svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP];
3831
3832 ret = nested_svm_vmexit(svm);
3833 if (ret)
3834 return ret;
3835 }
Ladi Prosek0234bf82017-10-11 16:54:40 +02003836 return 0;
3837}
3838
Sean Christophersoned193212019-04-02 08:03:09 -07003839static int svm_pre_leave_smm(struct kvm_vcpu *vcpu, const char *smstate)
Ladi Prosek0234bf82017-10-11 16:54:40 +02003840{
Ladi Prosek05cade72017-10-11 16:54:45 +02003841 struct vcpu_svm *svm = to_svm(vcpu);
3842 struct vmcb *nested_vmcb;
KarimAllah Ahmed8c5fbf12019-01-31 21:24:40 +01003843 struct kvm_host_map map;
Sean Christophersoned193212019-04-02 08:03:09 -07003844 u64 guest;
3845 u64 vmcb;
Ladi Prosek05cade72017-10-11 16:54:45 +02003846
Sean Christophersoned193212019-04-02 08:03:09 -07003847 guest = GET_SMSTATE(u64, smstate, 0x7ed8);
3848 vmcb = GET_SMSTATE(u64, smstate, 0x7ee0);
Ladi Prosek05cade72017-10-11 16:54:45 +02003849
Sean Christophersoned193212019-04-02 08:03:09 -07003850 if (guest) {
KarimAllah Ahmed8c5fbf12019-01-31 21:24:40 +01003851 if (kvm_vcpu_map(&svm->vcpu, gpa_to_gfn(vmcb), &map) == -EINVAL)
Sean Christopherson9ec19492019-04-02 08:03:11 -07003852 return 1;
KarimAllah Ahmed8c5fbf12019-01-31 21:24:40 +01003853 nested_vmcb = map.hva;
Paolo Bonzini69c9dfa2020-05-13 12:57:26 -04003854 enter_svm_guest_mode(svm, vmcb, nested_vmcb);
3855 kvm_vcpu_unmap(&svm->vcpu, &map, true);
Ladi Prosek05cade72017-10-11 16:54:45 +02003856 }
Sean Christopherson9ec19492019-04-02 08:03:11 -07003857 return 0;
Ladi Prosek0234bf82017-10-11 16:54:40 +02003858}
3859
Paolo Bonzinic9d40912020-05-22 11:21:49 -04003860static void enable_smi_window(struct kvm_vcpu *vcpu)
Ladi Prosekcc3d9672017-10-17 16:02:39 +02003861{
3862 struct vcpu_svm *svm = to_svm(vcpu);
3863
3864 if (!gif_set(svm)) {
3865 if (vgif_enabled(svm))
3866 set_intercept(svm, INTERCEPT_STGI);
3867 /* STGI will cause a vm exit */
Paolo Bonzinic9d40912020-05-22 11:21:49 -04003868 } else {
3869 /* We must be in SMM; RSM will cause a vmexit anyway. */
Ladi Prosekcc3d9672017-10-17 16:02:39 +02003870 }
Ladi Prosekcc3d9672017-10-17 16:02:39 +02003871}
3872
Singh, Brijesh05d5a482019-02-15 17:24:12 +00003873static bool svm_need_emulation_on_page_fault(struct kvm_vcpu *vcpu)
3874{
Liran Alon118154b2019-07-17 02:56:58 +03003875 unsigned long cr4 = kvm_read_cr4(vcpu);
3876 bool smep = cr4 & X86_CR4_SMEP;
3877 bool smap = cr4 & X86_CR4_SMAP;
3878 bool is_user = svm_get_cpl(vcpu) == 3;
Singh, Brijesh05d5a482019-02-15 17:24:12 +00003879
3880 /*
Paolo Bonzinie72436b2020-04-17 12:21:06 -04003881 * If RIP is invalid, go ahead with emulation which will cause an
3882 * internal error exit.
3883 */
3884 if (!kvm_vcpu_gfn_to_memslot(vcpu, kvm_rip_read(vcpu) >> PAGE_SHIFT))
3885 return true;
3886
3887 /*
Liran Alon118154b2019-07-17 02:56:58 +03003888 * Detect and workaround Errata 1096 Fam_17h_00_0Fh.
3889 *
3890 * Errata:
3891 * When CPU raise #NPF on guest data access and vCPU CR4.SMAP=1, it is
3892 * possible that CPU microcode implementing DecodeAssist will fail
3893 * to read bytes of instruction which caused #NPF. In this case,
3894 * GuestIntrBytes field of the VMCB on a VMEXIT will incorrectly
3895 * return 0 instead of the correct guest instruction bytes.
3896 *
3897 * This happens because CPU microcode reading instruction bytes
3898 * uses a special opcode which attempts to read data using CPL=0
3899 * priviledges. The microcode reads CS:RIP and if it hits a SMAP
3900 * fault, it gives up and returns no instruction bytes.
3901 *
3902 * Detection:
3903 * We reach here in case CPU supports DecodeAssist, raised #NPF and
3904 * returned 0 in GuestIntrBytes field of the VMCB.
3905 * First, errata can only be triggered in case vCPU CR4.SMAP=1.
3906 * Second, if vCPU CR4.SMEP=1, errata could only be triggered
3907 * in case vCPU CPL==3 (Because otherwise guest would have triggered
3908 * a SMEP fault instead of #NPF).
3909 * Otherwise, vCPU CR4.SMEP=0, errata could be triggered by any vCPU CPL.
3910 * As most guests enable SMAP if they have also enabled SMEP, use above
3911 * logic in order to attempt minimize false-positive of detecting errata
3912 * while still preserving all cases semantic correctness.
3913 *
3914 * Workaround:
3915 * To determine what instruction the guest was executing, the hypervisor
3916 * will have to decode the instruction at the instruction pointer.
Singh, Brijesh05d5a482019-02-15 17:24:12 +00003917 *
3918 * In non SEV guest, hypervisor will be able to read the guest
3919 * memory to decode the instruction pointer when insn_len is zero
3920 * so we return true to indicate that decoding is possible.
3921 *
3922 * But in the SEV guest, the guest memory is encrypted with the
3923 * guest specific key and hypervisor will not be able to decode the
3924 * instruction pointer so we will not able to workaround it. Lets
3925 * print the error and request to kill the guest.
3926 */
Liran Alon118154b2019-07-17 02:56:58 +03003927 if (smap && (!smep || is_user)) {
Singh, Brijesh05d5a482019-02-15 17:24:12 +00003928 if (!sev_guest(vcpu->kvm))
3929 return true;
3930
Liran Alon118154b2019-07-17 02:56:58 +03003931 pr_err_ratelimited("KVM: SEV Guest triggered AMD Erratum 1096\n");
Singh, Brijesh05d5a482019-02-15 17:24:12 +00003932 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3933 }
3934
3935 return false;
3936}
3937
Liran Alon4b9852f2019-08-26 13:24:49 +03003938static bool svm_apic_init_signal_blocked(struct kvm_vcpu *vcpu)
3939{
3940 struct vcpu_svm *svm = to_svm(vcpu);
3941
3942 /*
3943 * TODO: Last condition latch INIT signals on vCPU when
3944 * vCPU is in guest-mode and vmcb12 defines intercept on INIT.
Paolo Bonzini33b22172020-04-17 10:24:18 -04003945 * To properly emulate the INIT intercept,
3946 * svm_check_nested_events() should call nested_svm_vmexit()
3947 * if an INIT signal is pending.
Liran Alon4b9852f2019-08-26 13:24:49 +03003948 */
3949 return !gif_set(svm) ||
3950 (svm->vmcb->control.intercept & (1ULL << INTERCEPT_INIT));
3951}
3952
Joerg Roedeleaf78262020-03-24 10:41:54 +01003953static void svm_vm_destroy(struct kvm *kvm)
3954{
3955 avic_vm_destroy(kvm);
3956 sev_vm_destroy(kvm);
3957}
3958
3959static int svm_vm_init(struct kvm *kvm)
3960{
3961 if (avic) {
3962 int ret = avic_vm_init(kvm);
3963 if (ret)
3964 return ret;
3965 }
3966
3967 kvm_apicv_init(kvm, avic);
3968 return 0;
3969}
3970
Sean Christopherson9c14ee22020-03-21 13:26:03 -07003971static struct kvm_x86_ops svm_x86_ops __initdata = {
Li RongQingdd58f3c2020-02-23 16:13:12 +08003972 .hardware_unsetup = svm_hardware_teardown,
Avi Kivity6aa8b732006-12-10 02:21:36 -08003973 .hardware_enable = svm_hardware_enable,
3974 .hardware_disable = svm_hardware_disable,
Avi Kivity774ead32007-12-26 13:57:04 +02003975 .cpu_has_accelerated_tpr = svm_cpu_has_accelerated_tpr,
Tom Lendackybc226f02018-05-10 22:06:39 +02003976 .has_emulated_msr = svm_has_emulated_msr,
Avi Kivity6aa8b732006-12-10 02:21:36 -08003977
3978 .vcpu_create = svm_create_vcpu,
3979 .vcpu_free = svm_free_vcpu,
Avi Kivity04d2cc72007-09-10 18:10:54 +03003980 .vcpu_reset = svm_vcpu_reset,
Avi Kivity6aa8b732006-12-10 02:21:36 -08003981
Sean Christopherson562b6b02020-01-26 16:41:13 -08003982 .vm_size = sizeof(struct kvm_svm),
Suravee Suthikulpanit4e19c362019-11-14 14:15:05 -06003983 .vm_init = svm_vm_init,
Brijesh Singh1654efc2017-12-04 10:57:34 -06003984 .vm_destroy = svm_vm_destroy,
Suravee Suthikulpanit44a95da2016-05-04 14:09:46 -05003985
Avi Kivity04d2cc72007-09-10 18:10:54 +03003986 .prepare_guest_switch = svm_prepare_guest_switch,
Avi Kivity6aa8b732006-12-10 02:21:36 -08003987 .vcpu_load = svm_vcpu_load,
3988 .vcpu_put = svm_vcpu_put,
Suravee Suthikulpanit8221c132016-05-04 14:09:52 -05003989 .vcpu_blocking = svm_vcpu_blocking,
3990 .vcpu_unblocking = svm_vcpu_unblocking,
Avi Kivity6aa8b732006-12-10 02:21:36 -08003991
Paolo Bonzinia96036b2015-11-10 11:55:36 +01003992 .update_bp_intercept = update_bp_intercept,
Tom Lendacky801e4592018-02-21 13:39:51 -06003993 .get_msr_feature = svm_get_msr_feature,
Avi Kivity6aa8b732006-12-10 02:21:36 -08003994 .get_msr = svm_get_msr,
3995 .set_msr = svm_set_msr,
3996 .get_segment_base = svm_get_segment_base,
3997 .get_segment = svm_get_segment,
3998 .set_segment = svm_set_segment,
Izik Eidus2e4d2652008-03-24 19:38:34 +02003999 .get_cpl = svm_get_cpl,
Rusty Russell1747fb72007-09-06 01:21:32 +10004000 .get_cs_db_l_bits = kvm_get_cs_db_l_bits,
Avi Kivity6aa8b732006-12-10 02:21:36 -08004001 .set_cr0 = svm_set_cr0,
Avi Kivity6aa8b732006-12-10 02:21:36 -08004002 .set_cr4 = svm_set_cr4,
4003 .set_efer = svm_set_efer,
4004 .get_idt = svm_get_idt,
4005 .set_idt = svm_set_idt,
4006 .get_gdt = svm_get_gdt,
4007 .set_gdt = svm_set_gdt,
Gleb Natapov020df072010-04-13 10:05:23 +03004008 .set_dr7 = svm_set_dr7,
Paolo Bonzinifacb0132014-02-21 10:32:27 +01004009 .sync_dirty_debug_regs = svm_sync_dirty_debug_regs,
Avi Kivity6de4f3a2009-05-31 22:58:47 +03004010 .cache_reg = svm_cache_reg,
Avi Kivity6aa8b732006-12-10 02:21:36 -08004011 .get_rflags = svm_get_rflags,
4012 .set_rflags = svm_set_rflags,
Huaitong Hanbe94f6b2016-03-22 16:51:20 +08004013
Sean Christopherson77809382020-03-20 14:28:18 -07004014 .tlb_flush_all = svm_flush_tlb,
Sean Christophersoneeeb4f62020-03-20 14:28:20 -07004015 .tlb_flush_current = svm_flush_tlb,
Junaid Shahidfaff8752018-06-29 13:10:05 -07004016 .tlb_flush_gva = svm_flush_tlb_gva,
Sean Christopherson72b38322020-03-20 14:28:13 -07004017 .tlb_flush_guest = svm_flush_tlb,
Avi Kivity6aa8b732006-12-10 02:21:36 -08004018
Avi Kivity6aa8b732006-12-10 02:21:36 -08004019 .run = svm_vcpu_run,
Avi Kivity04d2cc72007-09-10 18:10:54 +03004020 .handle_exit = handle_exit,
Avi Kivity6aa8b732006-12-10 02:21:36 -08004021 .skip_emulated_instruction = skip_emulated_instruction,
Oliver Upton5ef8acb2020-02-07 02:36:07 -08004022 .update_emulated_instruction = NULL,
Glauber Costa2809f5d2009-05-12 16:21:05 -04004023 .set_interrupt_shadow = svm_set_interrupt_shadow,
4024 .get_interrupt_shadow = svm_get_interrupt_shadow,
Ingo Molnar102d8322007-02-19 14:37:47 +02004025 .patch_hypercall = svm_patch_hypercall,
Eddie Dong2a8067f2007-08-06 16:29:07 +03004026 .set_irq = svm_set_irq,
Gleb Natapov95ba8273132009-04-21 17:45:08 +03004027 .set_nmi = svm_inject_nmi,
Avi Kivity298101d2007-11-25 13:41:11 +02004028 .queue_exception = svm_queue_exception,
Avi Kivityb463a6f2010-07-20 15:06:17 +03004029 .cancel_injection = svm_cancel_injection,
Gleb Natapov78646122009-03-23 12:12:11 +02004030 .interrupt_allowed = svm_interrupt_allowed,
Gleb Natapov95ba8273132009-04-21 17:45:08 +03004031 .nmi_allowed = svm_nmi_allowed,
Jan Kiszka3cfc3092009-11-12 01:04:25 +01004032 .get_nmi_mask = svm_get_nmi_mask,
4033 .set_nmi_mask = svm_set_nmi_mask,
Gleb Natapov95ba8273132009-04-21 17:45:08 +03004034 .enable_nmi_window = enable_nmi_window,
4035 .enable_irq_window = enable_irq_window,
4036 .update_cr8_intercept = update_cr8_intercept,
Jim Mattson8d860bb2018-05-09 16:56:05 -04004037 .set_virtual_apic_mode = svm_set_virtual_apic_mode,
Andrey Smetanind62caab2015-11-10 15:36:33 +03004038 .refresh_apicv_exec_ctrl = svm_refresh_apicv_exec_ctrl,
Suravee Suthikulpanitef8efd72019-11-14 14:15:10 -06004039 .check_apicv_inhibit_reasons = svm_check_apicv_inhibit_reasons,
Suravee Suthikulpanit2de9d0c2019-11-14 14:15:11 -06004040 .pre_update_apicv_exec_ctrl = svm_pre_update_apicv_exec_ctrl,
Yang Zhangc7c9c562013-01-25 10:18:51 +08004041 .load_eoi_exitmap = svm_load_eoi_exitmap,
Suravee Suthikulpanit44a95da2016-05-04 14:09:46 -05004042 .hwapic_irr_update = svm_hwapic_irr_update,
4043 .hwapic_isr_update = svm_hwapic_isr_update,
Liran Alonfa59cc02017-12-24 18:12:53 +02004044 .sync_pir_to_irr = kvm_lapic_find_highest_irr,
Suravee Suthikulpanitbe8ca172016-05-04 14:09:49 -05004045 .apicv_post_state_restore = avic_post_state_restore,
Izik Eiduscbc94022007-10-25 00:29:55 +02004046
4047 .set_tss_addr = svm_set_tss_addr,
Sean Christopherson2ac52ab2018-03-20 12:17:19 -07004048 .set_identity_map_addr = svm_set_identity_map_addr,
Sheng Yang67253af2008-04-25 10:20:22 +08004049 .get_tdp_level = get_npt_level,
Sheng Yang4b12f0d2009-04-27 20:35:42 +08004050 .get_mt_mask = svm_get_mt_mask,
Marcelo Tosatti229456f2009-06-17 09:22:14 -03004051
Avi Kivity586f9602010-11-18 13:09:54 +02004052 .get_exit_info = svm_get_exit_info,
Avi Kivity586f9602010-11-18 13:09:54 +02004053
Sheng Yang0e851882009-12-18 16:48:46 +08004054 .cpuid_update = svm_cpuid_update,
Sheng Yang4e47c7a2009-12-18 16:48:47 +08004055
Sheng Yangf5f48ee2010-06-30 12:25:15 +08004056 .has_wbinvd_exit = svm_has_wbinvd_exit,
Zachary Amsden99e3e302010-08-19 22:07:17 -10004057
Leonid Shatz326e7422018-11-06 12:14:25 +02004058 .write_l1_tsc_offset = svm_write_l1_tsc_offset,
Joerg Roedel1c97f0a2010-09-10 17:30:41 +02004059
Paolo Bonzini727a7e22020-03-05 03:52:50 -05004060 .load_mmu_pgd = svm_load_mmu_pgd,
Joerg Roedel8a76d7f2011-04-04 12:39:27 +02004061
4062 .check_intercept = svm_check_intercept,
Sean Christopherson95b5a482019-04-19 22:50:59 -07004063 .handle_exit_irqoff = svm_handle_exit_irqoff,
Radim Krčmářae97a3b2014-08-21 18:08:06 +02004064
Sean Christophersond264ee02018-08-27 15:21:12 -07004065 .request_immediate_exit = __kvm_request_immediate_exit,
4066
Radim Krčmářae97a3b2014-08-21 18:08:06 +02004067 .sched_in = svm_sched_in,
Wei Huang25462f72015-06-19 15:45:05 +02004068
4069 .pmu_ops = &amd_pmu_ops,
Paolo Bonzini33b22172020-04-17 10:24:18 -04004070 .nested_ops = &svm_nested_ops,
4071
Suravee Suthikulpanit340d3bc2016-05-04 14:09:47 -05004072 .deliver_posted_interrupt = svm_deliver_avic_intr,
Wanpeng Li17e433b2019-08-05 10:03:19 +08004073 .dy_apicv_has_pending_interrupt = svm_dy_apicv_has_pending_interrupt,
Suravee Suthikulpanit411b44b2016-08-23 13:52:43 -05004074 .update_pi_irte = svm_update_pi_irte,
Borislav Petkov74f16902017-03-26 23:51:24 +02004075 .setup_mce = svm_setup_mce,
Ladi Prosek0234bf82017-10-11 16:54:40 +02004076
Ladi Prosek72d7b372017-10-11 16:54:41 +02004077 .smi_allowed = svm_smi_allowed,
Ladi Prosek0234bf82017-10-11 16:54:40 +02004078 .pre_enter_smm = svm_pre_enter_smm,
4079 .pre_leave_smm = svm_pre_leave_smm,
Ladi Prosekcc3d9672017-10-17 16:02:39 +02004080 .enable_smi_window = enable_smi_window,
Brijesh Singh1654efc2017-12-04 10:57:34 -06004081
4082 .mem_enc_op = svm_mem_enc_op,
Brijesh Singh1e80fdc2017-12-04 10:57:38 -06004083 .mem_enc_reg_region = svm_register_enc_region,
4084 .mem_enc_unreg_region = svm_unregister_enc_region,
Vitaly Kuznetsov57b119d2018-10-16 18:50:01 +02004085
Singh, Brijesh05d5a482019-02-15 17:24:12 +00004086 .need_emulation_on_page_fault = svm_need_emulation_on_page_fault,
Liran Alon4b9852f2019-08-26 13:24:49 +03004087
4088 .apic_init_signal_blocked = svm_apic_init_signal_blocked,
Avi Kivity6aa8b732006-12-10 02:21:36 -08004089};
4090
Sean Christophersond008dfd2020-03-21 13:25:56 -07004091static struct kvm_x86_init_ops svm_init_ops __initdata = {
4092 .cpu_has_kvm_support = has_svm,
4093 .disabled_by_bios = is_disabled,
4094 .hardware_setup = svm_hardware_setup,
4095 .check_processor_compatibility = svm_check_processor_compat,
4096
4097 .runtime_ops = &svm_x86_ops,
Avi Kivity6aa8b732006-12-10 02:21:36 -08004098};
4099
4100static int __init svm_init(void)
4101{
Sean Christophersond008dfd2020-03-21 13:25:56 -07004102 return kvm_init(&svm_init_ops, sizeof(struct vcpu_svm),
Avi Kivity0ee75be2010-04-28 15:39:01 +03004103 __alignof__(struct vcpu_svm), THIS_MODULE);
Avi Kivity6aa8b732006-12-10 02:21:36 -08004104}
4105
4106static void __exit svm_exit(void)
4107{
Zhang Xiantaocb498ea2007-11-14 20:39:31 +08004108 kvm_exit();
Avi Kivity6aa8b732006-12-10 02:21:36 -08004109}
4110
4111module_init(svm_init)
4112module_exit(svm_exit)